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
|
|---|---|---|---|---|---|---|---|---|---|
1305.1391 | 1 | 1305 | 2013-05-07T03:19:45 | A polynomial identity for the bilinear operation in Lie-Yamaguti algebras | [
"math.RA",
"math.DG"
] | We use computer algebra to demonstrate the existence of a multilinear polynomial identity of degree 8 satisfied by the bilinear operation in every Lie-Yamaguti algebra. This identity is a consequence of the defining identities for Lie-Yamaguti algebras, but is not a consequence of anticommutativity. We give an explicit form of this identity as an alternating sum over all permutations of the variables in a polynomial with 8 terms. Our computations also show that such identities do not exist in degrees less than 8. | math.RA | math |
A POLYNOMIAL IDENTITY FOR THE BILINEAR OPERATION
IN LIE-YAMAGUTI ALGEBRAS
MURRAY R. BREMNER
Abstract. We use computer algebra to demonstrate the existence of a mul-
tilinear polynomial identity of degree 8 satisfied by the bilinear operation in
every Lie-Yamaguti algebra. This identity is a consequence of the defining
identities for Lie-Yamaguti algebras, but is not a consequence of anticommu-
tativity. We give an explicit form of this identity as an alternating sum over all
permutations of the variables in a polynomial with 8 terms. Our computations
also show that such identities do not exist in degrees less than 8.
1. Introduction
The binary-ternary structures now called Lie-Yamaguti algebras were introduced
by Yamaguti [12] in his study of torsion and curvature on symmetric spaces.
Definition 1.1. A vector space L with a bilinear operation [−, −] : L × L → L and
a trilinear operation h−, −, −i : L × L × L → L is a Lie-Yamaguti algebra if the
operations satisfy the following polynomial identities for all a, b, c, d, e ∈ L:
(1)
(2)
(3)
(4)
(5)
(6)
[a, a] ≡ 0,
ha, a, bi ≡ 0,
Xa,b,c(cid:0) [[a, b], c] + ha, b, ci(cid:1) ≡ 0 (cyclic sum),
Xa,b,c
h [a, b], c, d i ≡ 0 (cyclic sum),
h a, b, [c, d] i ≡ [ ha, b, ci, d ] + [ c, ha, b, di ],
h a, b, hc, d, ei i ≡ h ha, b, ci, d, e i + h c, ha, b, di, e i + h c, d, ha, b, ei i.
From this it is clear that Lie-Yamaguti algebras are a generalization of Lie al-
gebras and Lie triple systems: if the binary (resp. ternary) operation is identically
zero, we obtain a Lie triple system (resp. Lie algebra). Yamaguti called these struc-
tures general Lie triple systems; the modern name was introduced by Kinyon and
Weinstein [9].
2010 Mathematics Subject Classification. Primary 17A30. Secondary 17-04, 17A32, 17A40,
17A50, 17B01.
Key words and phrases. Lie-Yamaguti algebras, polynomial identities, computer algebra, rep-
resentation theory of the symmetric group.
This research was supported by a Discovery Grant from NSERC, the Natural Sciences and
Engineering Research Council of Canada. The author thanks Pilar Benito and the Department of
Mathematics and Computer Science at the University of La Rioja in Logrono for their hospitality
in April and May 2013.
1
2
MURRAY R. BREMNER
Given two elements a, b ∈ L we define a linear operator Da,b ∈ End(L) by
Da,b(c) = ha, b, ci; this is a derivation of both the bilinear and trilinear operations.
We denote by D ⊆ End(L) the subspace spanned by all Da,b (a, b ∈ L), and define
an anticommutative product · on G = D ⊕ L as follows:
Da,b · Dc,d = Dha,b,ci,d + Dc,ha,b,di, Da,b · c = ha, b, ci,
a · b = Da,b + [a, b].
Yamaguti showed that G is a Lie algebra, the standard enveloping algebra of L.
Furthermore, D is a Lie subalgebra of G, and D · L ⊆ L, giving a reductive de-
composition of G. Finally, we recover the bilinear and trilinear operations on L
from the product in G by the formulas [a, b] = pL(a · b) and ha, b, ci = pD(a · b) · c
where pD and pL are the projections onto D and L. The structure theory of finite
dimensional Lie-Yamaguti algebras has been developed during the last few years
by Benito, Elduque, and Mart´ın-Herce [2, 3].
Loday [10] introduced the notion of a Leibniz algebra, which is a nonassociative
algebra A with product {a, b} satisfying the derivation identity,
{{a, b}, c} ≡ {{a, c}, b} + {a, {b, c}}.
Kinyon and Weinstein showed that the skew-symmetrization [a, b] = {a, b} − {b, a}
of the product on any Leibniz algebra can be realized as the bilinear operation on
a Lie-Yamaguti algebra, where the trilinear operation is ha, b, ci = − 1
4 {{a, b}, c}.
In spite of all this, the bilinear operation in a Lie-Yamaguti algebra remains
somewhat mysterious; the only identity in Definition 1.1 which involves the bilinear
operation by itself is anticommutativity (1). This raises the following question:
Let LY (X) be the free Lie-Yamaguti algebra on the set X of gener-
ators, and let BLY (X) be the subalgebra generated by X using only
the bilinear operation. Is BLY (X) a free anticommutative algebra?
(For free anticommutative algebras, see Shirshov [11].) This question can be refor-
mulated in terms of polynomial identities as follows:
Do there exist polynomial identities satisfied by the operation [−, −]
in every Lie-Yamaguti algebra which follow from the defining iden-
tities (1) -- (6) but do not follow from anticommutativity (1) alone?
From the calculations of the author and Hentzel [5], it is known that if such an iden-
tity exists, it must have degree at least 7. That paper studied the anticommutative
product on the sl2(C)-module V (10) obtained from the decomposition
Λ2V (10) ≈ V (18) ⊕ V (14) ⊕ V (10) ⊕ V (6) ⊕ V (2).
In this isomorphism, the direct sum of V (10) with the adjoint module V (2) ∼= sl2(C)
can be given the structure of a Lie algebra of type G2, and hence V (10) becomes a
Lie-Yamaguti algebra where the operations [a, b] = β(a, b) and ha, b, ci = τ (a, b) · c
are determined by the projections β : Λ2V (10) → V (10) and τ : Λ2V (10) → V (2).
The simplest identities satisfied by the bilinear operation β(a, b) which do not follow
from anticommutativity have degree 7. Similar calculations for the Lie-Yamaguti
structure on V (6) were done by the author and Douglas [4]. (For a detailed analysis
of Lie-Yamaguti algebras related to G2, see Benito, Draper and Elduque [1].)
In this note, we describe calculations with the computer algebra system Maple
that demonstrate the existence of a polynomial identity in degree 8 which is sat-
isfied by the bilinear operation in every Lie-Yamaguti algebra but which is not a
consequence of anticommutativity. We give an explicit statement of this identity,
and show that such identities do not exist in degrees less than 8.
A POLYNOMIAL IDENTITY IN LIE-YAMAGUTI ALGEBRAS
3
2. Polynomial identities in Lie-Yamaguti algebras
We consider binary-ternary algebras, or BT-algebras for short, with a bilinear
operation [−, −] and a trilinear operation h−, −, −i satisfying identities (1) and (2).
Definition 2.1. We write BT (n) for the multilinear subspace of degree n in the
free BT-algebra on n generators. This space is a module over the symmetric group
Sn acting by permutations of the subscripts: σ ·xi = xσ(i) for σ ∈ Sn and 1 ≤ i ≤ n.
We need to determine the number btn of inequivalent placements of the binary
and ternary operation symbols in a monomial of degree n in a free BT-algebra.
Lemma 2.2. The number btn of binary-ternary association types in degree n is
given by bt1 = 1 and the following recurrence relation:
⌊(n−1)/2⌋
btn =
btn−ibti +(cid:18)btn/2+1
2 (cid:19)
Xi=1
Xi=1" ⌊(n−i−1)/2⌋
Xj=1
n−2
+
btn−i−jbtjbti +(cid:18)bt(n−i)/2+1
2
(cid:19)bti#.
The terms with binomial coefficients occur only when defined: the first for n even
and the second for n−i even.
In this recurrence relation, the quadratic (resp. cubic) terms correspond to ap-
plications of the bilinear (resp. trilinear) product. The btn association types are
partitioned into three subtypes: bn binary types involve only the bilinear product,
tn ternary types involve only the trilinear product, and the remaining mn mixed
types involve both. We have btn = bn + tn + mn for n ≥ 2.
n
btn
bn
tn
mn
1 2
1 1
1 1
1 0
0 0
3 4
2 5
1 2
1 0
0 3
6
5
12
13 38 113 354 1128 3688 12229 41161
11
10
7
8
9
3
2
8
6
0
32
11
6
96
46
19
23
0
451
0
331 1063 3590 11955 40710
207
67
98
0
We will order the association types in each degree with the ternary and mixed
types first, and the binary types last. Tables 1 and 2 display the BT types in
degrees 5 and 6, ordered according to the following definition.
Definition 2.3. If T is a binary-ternary association type then we set α(T ) = {2}
(resp. α(T ) = 3) if T involves only the binary (resp. ternary) operation, and α(T ) =
{2, 3} if T involves both operations. We impose the order {3} ≺ {2, 3} ≺ {2}. If
T = [U1, U2] (resp. T = hU1, U2, U3i) then we set ω(T ) = 2 (resp. ω(T ) = 3). If T
and T ′ are binary-ternary association types of degrees n and n′ then we say that
T ≺ T ′ in deglex order (degree-lexicographical order) if:
• n < n′, or
• n = n′ and α(T ) ≺ α(T ′), or
• n = n′, α(T ) = α(T ′) and ω(T ) < ω(T ′), or
• n = n′, α(T ) = α(T ′), ω(T ) = ω(T ′) and Ui ≺ U ′
2, U ′
2] (or T = hU1, U2, U3i, T ′ = hU ′
1, U ′
T ′ = [U1, U ′
index for which Ui 6= U ′
i .
i , where T = [U1, U2],
3i) and i is the least
4
MURRAY R. BREMNER
h− − h− − −ii
hh− − −i − −i
[h− − −i[−−]]
h− − [[−−]−]i
[[h− − −i−]−]
h[−−] − [−−]i
[h− − [−−]i−]
h[−−][−−]−i
[h[−−] − −i−]
h[[−−]−] − −i
[[[−−]−][−−]]
[[[−−][−−]]−]
[[[[−−]−]−]−]
Table 1. Binary-ternary association types in degree 5
[h− − −ih− − −i]
[h[−−] − −i[−−]]
[[[h− − −i−]−]−]
[h[−−] − [−−]i−]
h− − h− − [−−]ii
h[−−] − h− − −ii
hh− − −i[−−]−i
hh− − [−−]i − −i
[h− − −i[[−−]−]]
[h− − h− − −ii−]
[[h− − [−−]i−]−]
[h[−−][−−]−i−]
h− − h[−−] − −ii
h[−−] − [[−−]−]i
h[[−−]−] − [−−]i
hh[−−] − −i − −i
[[h− − −i−][−−]]
[hh− − −i − −i−]
[[h[−−] − −i−]−]
[h[[−−]−] − −i−]
h− − [[−−][−−]]i
h[−−][−−][−−]i
h[[−−]−][−−]−i
h[[−−][−−]] − −i
[h− − [−−]i[−−]]
[[h− − −i[−−]]−]
[h− − [[−−]−]i−]
h− − [h− − −i−]i
h− − [[[−−]−]−]i
hh− − −i − [−−]i
h[h− − −i−] − −i
h[[[−−]−]−] − −i
[[[−−]−][[−−]−]]
[[[[−−][−−]]−]−]
[[[−−][−−]][−−]]
[[[[[−−]−]−]−]−]
[[[[−−]−]−][−−]]
[[[[−−]−][−−]]−]
Table 2. Binary-ternary association types in degree 6
Definition 2.4. Let T be a binary-ternary association type in degree n. Let
ι ∈ Sn be the identity permutation, and consider σ ∈ Sn with σ 6= ι but σ2 = ι.
Let [ι]T +[σ]T be the multilinear polynomial in BT (n) whose two terms have type T
with variables x1 · · · xn and xσ(1) · · · xσ(n) respectively. If [ι]T + [σ]T ≡ 0 in BT (n)
then we call this identity a skew-symmetry of association type T . We write s(T )
for the number of skew-symmetrices of association type T .
We will only require the skew-symmetries of the binary association types in each
degree. In degree 8 the 23 binary types are displayed in Table 3 with their index
numbers as subscripts. The corresponding 74 skew-symmetries are displayed in
Table 4; we give only the index number t, together with σ expressed as a product
of disjoint transpositions and its sign.
Definition 2.5. Let f = f (a1, . . . , an) be an element of BT (n): a multilinear poly-
nomial of degree n in the free BT-algebra on n generators. The bilinear operation
produces n+1 consequences of f in degree n+1, called the binary liftings of f ,
using n substitutions and one multiplication:
f ([a1, an+1], a2, . . . , an),
[f (a1, . . . , an), an+1].
. . . ,
f (a1, . . . , an−1, [an, an+1]),
The trilinear operation produces n+2 consequences of f in degree n+2, called the
ternary liftings of f , using n substitutions and two multiplications:
f (ha1, an+1, an+2i, a2, . . . , an),
hf (a1, . . . , an), an+1, an+2i,
. . . ,
f (a1, . . . , an−1, han, an+1, an+2i),
han+1, an+2, f (a1, . . . , an)i.
A POLYNOMIAL IDENTITY IN LIE-YAMAGUTI ALGEBRAS
5
[[[−−][−−]][[−−][−−]]]1
[[[[−−]−][−−]][[−−]−]]4
[[[[−−]−][[−−]−]][−−]]7
[[[[[−−]−][−−]]−][−−]]10
[[[[−−][−−]][[−−]−]]−]13
[[[[[−−][−−]]−][−−]]−]16
[[[[[−−][−−]][−−]]−]−]19
[[[[[[−−][−−]]−]−]−]−]22
[[[−−][−−]][[[−−]−]−]]2
[[[[−−][−−]]−][[−−]−]]5
[[[[−−][−−]][−−]][−−]]8
[[[[[−−][−−]]−]−][−−]]11
[[[[[−−]−]−][[−−]−]]−]14
[[[[[[−−]−]−]−][−−]]−]17
[[[[[[−−]−]−][−−]]−]−]20
[[[[[[[−−]−]−]−]−]−]−]23
[[[[−−]−]−][[[−−]−]−]]3
[[[[[−−]−]−]−][[−−]−]]6
[[[[[−−]−]−][−−]][−−]]9
[[[[[[−−]−]−]−]−][−−]]12
[[[[[−−]−][−−]][−−]]−]15
[[[[[−−]−][[−−]−]]−]−]18
[[[[[[−−]−][−−]]−]−]−]21
Table 3. Binary association types in degree 8
t σ
ε(σ)
t σ
ε(σ)
t σ
ε(σ)
1 (12)
1 (56)
1 (15)(26)(37)(48)
2 (13)(24)
3 (56)
4 (45)
5 (34)
6 (12)
7 (45)
8 (12)
8 (56)
9 (56)
10 (45)
11 (34)
12 (12)
13 (34)
14 (12)
15 (45)
16 (34)
17 (12)
18 (45)
19 (34)
20 (12)
21 (45)
22 (13)(24)
−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 (34)
1 (78)
2 (12)
2 (56)
3 (15)(26)(37)(48)
4 (67)
5 (13)(24)
6 (67)
7 (14)(25)(36)
8 (34)
8 (78)
9 (78)
10 (78)
11 (13)(24)
12 (78)
13 (13)(24)
14 (56)
15 (67)
16 (13)(24)
17 (67)
18 (14)(25)(36)
19 (13)(24)
20 (56)
22 (12)
23 (12)
(13)(24)
1
(57)(68)
1
(34)
2
(12)
3
(12)
4
(12)
5
(67)
5
(12)
7
(78)
7
(13)(24)
8
(12)
9
10 (12)
11 (12)
11 (78)
13 (12)
13 (56)
15 (12)
16 (12)
16 (67)
18 (12)
19 (12)
19 (56)
21 (12)
22 (34)
−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
Table 4. Skew-symmetries of binary types in degree 8
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 apply this lifting process to identities (3)-(6), which we write as follows:
f (a, b, c) = [[a, b], c] − [[a, c], b] + [[b, c], a] + ha, b, ci − ha, c, bi + hb, c, ai,
g1(a, b, c, d) = h[a, b], c, di − h[a, c], b, di + h[b, c], a, di,
g2(a, b, c, d) = ha, b, [c, d]i − [ha, b, ci, d] + [ha, b, di, c],
h(a, b, c, d, e) = ha, b, hc, d, eii − hha, b, ci, d, ei + hha, b, di, c, ei − hc, d, ha, b, eii.
6
MURRAY R. BREMNER
h(a, b, c, d, e),
g1(a, b, c, [d, e]),
g2(a, b, [c, e], d),
f ([a, [d, e]], b, c),
f ([a, e], [b, d], c),
[f (a, [b, d], c), e],
f (a, b, [c, [d, e]]),
[f (a, b, [c, e]), d],
f (a, hb, d, ei, c),
g1([a, e], b, c, d),
[g1(a, b, c, d), e],
g2(a, b, c, [d, e]),
f ([a, d], [b, e], c),
f (a, [[b, e], d], c),
f ([a, e], b, [c, d]),
[f (a, b, [c, d]), e],
[f (a, b, c), [d, e]],
f (a, b, hc, d, ei),
g1(a, [b, e], c, d),
g2([a, e], b, c, d),
[g2(a, b, c, d), e],
f ([a, d], b, [c, e]),
f (a, [b, [d, e]], c),
f (a, [b, e], [c, d]),
[f ([a, e], b, c), d],
[[f (a, b, c), d], e],
hf (a, b, c), d, ei,
g1(a, b, [c, e], d),
g2(a, [b, e], c, d),
f ([[a, e], d], b, c),
[f ([a, d], b, c), e],
f (a, [b, d], [c, e]),
f (a, b, [[c, e], d]),
[f (a, [b, e], c), d],
f (ha, d, ei, b, c),
hd, e, f (a, b, c)i.
Table 5. Lie-Yamaguti identities in degree 5
In degree 3, we have only one identity:
f (a, b, c) ≡ 0.
In degree 4, we have g1, g2 and four consequences of f using [−, −]:
g1(a, b, c, d) ≡ 0,
f (a, [b, d], c) ≡ 0,
g2(a, b, c, d) ≡ 0,
f (a, b, [c, d]) ≡ 0,
f ([a, d], b, c) ≡ 0,
[f (a, b, c), d] ≡ 0.
In degree 5, we have h, five consequences of each identity in degree 4 using [−, −],
and five consequences of f using h−, −, −i; see Table 5, where we omit "≡ 0" to save
space. These 36 identities generate the S5-module of identities in degree 5 satisfied
by Lie-Yamaguti algebras. Some of these identities are redundant, but this issue
will not concern us until degree 6. For example, we do not need to include both
f ([[a, e], d], b, c),
f ([a, [d, e]], b, c) = −f ([[d, e], a], b, c),
in our set of generators, since they differ only by the transposition (ad) and a sign.
Definition 2.6. We write LY (n) ⊂ BT (n) for the Sn-submodule of all multilinear
polynomial identities in degree n satisfied by Lie-Yamaguti algebras.
Each of the 36 generators of LY (5) produces 6 identities in degree 6 using [−, −],
and each of the six generators of LY (4) produces 6 identities in degree 6 using
h−, −, −i, giving a total of 252 generators for the S6-module LY (6). Continuing
the lifting process, we obtain (252 + 36) · 7 = 2016 generators for the S7-module
LY (7), and (2016 + 252) · 8 = 18144 generators for the S8-module LY (8).
Lemma 2.7. The number λ(n) of multilinear Lie-Yamaguti identities in degree n
obtained by the lifting procedure described above equals
λ(n) =
1
20
(n+1)!
(n ≥ 4).
Proof. The sequence λ(n) satisfies λ(4) = 6, λ(5) = 36, and the recurrence relation
(cid:3)
λ(n) = n(cid:0)λ(n−1) + λ(n−2(cid:1), which has the indicated solution.
We will discuss in the next section how to eliminate redundancies in these sets
of Lie-Yamaguti identities in order to reduce the size of the computations.
A POLYNOMIAL IDENTITY IN LIE-YAMAGUTI ALGEBRAS
7
Lemma 2.8. The number of multilinear monomials forming a basis of BT (n) is
where T1, . . . , Tm is a complete list of the association types in degree n.
µ(n) =
n!
2s(Ti) ,
m
Xi=1
Proof. If T is an association type in degree n with s(T ) skew-symmetries, then the
number of distinct multilinear monomials with type T is n!/2s(T ).
(cid:3)
The matrix of Lie-Yamaguti identities. In principle, for each n we construct
a matrix of size λ(n) × µ(n) in which the (i, j) entry is the coefficient of the j-
th multilinear monomial in the i-th Lie-Yamaguti identity. We then compute the
row canonical form (RCF) of this matrix, and identify those rows whose leading
1s occur in a column labelled by a monomial in a binary association type. (Recall
that we have sorted the association types so that the binary types correspond to
the rightmost columns of the matrix. This idea of using a convenient ordering
of the association types was introduced by Correa, Hentzel and Peresi [8].) Let
b(n) be the number of columns corresponding to the binary types, and let a(n) be
the number of rows of the RCF whose leading 1s occur in the last b(n) columns.
Since we have already eliminated all skew-symmetries of the association types in
our enumeration of the multilinear monomials, it follows that if a(n) > 0 then the
corresponding rows are polynomial identities satisfied by the bilinear operation in
every Lie-Yamaguti algebra which are not consequences of identities (1) and (2).
Representation theory of the symmetric group. We have
µ(5) = 300,
µ(6) = 5310,
µ(7) = 109620,
µ(8) = 2751840.
In order to reduce the size of the matrices, we use the representation theory of the
symmetric group. A theoretical and algorithmic exposition of this method has been
given by Bremner and Peresi [6]; here we summarize the basic ideas.
The irreducible representations of Sn are in one-to-one correspondence with the
partitions π of n. For π = (n1, . . . , nk), n1 ≥ · · · ≥ nk ≥ 1, n1 + · · · + nk = n, we
write dπ for the dimension of the corresponding representation. Over any field F of
characteristic 0 or p > n, the group algebra F Sn is semisimple and has the following
decomposition into the direct sum of simple two-sided ideals, where Md(F ) denotes
the d × d matrices over F :
F Sn ∼=Mπ
Mdπ (F ).
We write Rπ : F Sn → Mdπ (F ) for the projection corresponding to π. The matrices
Rπ(σ) for σ ∈ Sn can be computed using the method of Clifton [7]. Let T1, . . . , Tm
be the BT association types in degree n, and let I1, . . . , Iℓ be the Lie-Yamaguti
identities in degree n. We collect the terms of each identity by association type:
Ii =
Iij ,
m
Xj=1
where Iij ∈ F Sn contains the terms of the i-th identity with the j-th type. Thus
each identity can be written as an element of the direct sum of m copies of F Sn.
For each partition π, we write Lπ for the ℓdπ × mdπ matrix consisting of dπ × dπ
blocks in which the (i, j) block contains the representation matrix Rπ(Iij ). We
compute RCF(Lπ) and extract the aπ × bndπ submatrix Aπ containing the nonzero
8
MURRAY R. BREMNER
rows whose leading 1s occur in one of the dπ × dπ blocks corresponding to the
bn binary types. This submatrix represents the consequences of the Lie-Yamaguti
identities for partition π which involve only the bilinear operation:
ternary and mixed BT types
binary types
RCF(Lπ) =
z
"
∗
0
}
· · · · · ·
· · · · · ·
{
∗
0
∗ · · · ∗
z
}
Aπ
{
#
We need to exclude the identities which are consequences of the skew-symmetries
of the binary types. We construct the sndπ × bndπ matrix Bπ in which the (i, j)
block contains the matrix Rπ([ι]j + [σi]j) representing the terms of the i-th skew-
symmetry in the j-th binary type. We compute RCF(Bπ) and write Bπ for the
cπ × bndπ submatrix containing the nonzero rows.
If the row space of Aπ is a subspace of the row space of Bπ, then for partition π
we see that every identity for the bilinear operation is a consequence of the skew-
symmetries of the binary types. If the row space of Aπ is not a subspace of the row
space of Bπ, then each row of Aπ which does not belong to the row space of Bπ
represents an identity for the bilinear operation which is not a consequence of the
skew-symmetries of the binary types.
If computer memory permits, we do these calculations using rational arithmetic.
However, this becomes impractical except in low degrees. Therefore, we use mod-
ular arithmetic with a prime p greater than the degree n of the identities. By the
structure theory of the group algebra F Sn, the ranks of the matrices will be the
same as they would have been if we had used rational arithmetic.
3. Main Theorem
We first show that there are no identities of the required form in degrees 6 and 7,
and we then study degree 8, where we find an identity for the bilinear operation.
In degree 6, there are 38 binary-ternary association types, and 252 Lie-Yamaguti
identities. For each partition π, we construct the 252dπ × 38dπ matrix in which the
(i, j) block contains the representation matrix for the terms with the j-th association
type in the i-th Lie-Yamaguti identity. We compute the RCF of this matrix and
extract the lower right block containing the binary identities.
There are 6 binary types with 15 skew-symmetries. We construct the 15dπ × 6dπ
matrix in which the (i, j) block contains the representation matrix for the terms
with the j-th binary type in the i-th skew-symmetry, and compute its RCF.
For every partition π, we find that the row space of the lower right block of
the first matrix is a subspace of the row space of the second matrix. Thus every
identity in degree 6 satisfied by the bilinear operation is a consequence of the skew-
symmetries of the binary types.
We repeat the same computation, but we reduce the matrix after each Lie-
Yamaguti identity in order to keep track of those identities which cause the rank
to increase for at least one partition. We find that the 252 identities in degree 6
are consequences of a subset of 48 identities: more than 80% are redundant.
In degree 7, there are 113 binary-ternary association types, and 2016 Lie-Yamaguti
identities. There are 11 binary types with 30 skew-symmetries. Following the same
procedure as in degree 6, we find that every identity in degree 7 satisfied by the
bilinear operation is a consequence of the skew-symmetries of the binary types.
A POLYNOMIAL IDENTITY IN LIE-YAMAGUTI ALGEBRAS
9
·
1
· 1
·
·
·
·
·
·
·
·
·
·
·
· 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1
·
· 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· − 3
· −1 1
2
·
·
·
·
· 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1 ·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 2 ·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 3
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1 ·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 2 −2
·
·
·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
Table 6. RCF of binary Lie-Yamaguti identities for partition 18
·
·
·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
·
1
·
· 1
·
·
·
·
·
·
·
·
·
·
· 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
Table 7. RCF of binary skew-symmetries for partition 18
·
·
·
·
·
·
·
·
· 1 ·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1 ·
·
·
·
·
·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1
·
·
·
·
·
·
·
·
·
·
Furthermore, the 2016 identities in degree 7 are consequences of a subset of 154
identities: more than 92% are redundant.
In degree 8, we first compute the binary liftings of the 154 identities from degree
7 and the ternary liftings of the 48 identities from degree 6. We obtain altogether
(154 + 48) · 8 = 1616 identities, less than 9% of the original set of 18144 identities.
Of the 22 irreducible representations of S8, the first 21 contain only identities for
the bilinear operation which are consequences of the skew-symmetries of the binary
types. The last representation (the sign representation of S8) produces an identity
which is not a consequence of the skew-symmetries. For this representation, the
corresponding elements of the group algebra F S8 are the alternating sums over all
permutations of the variables in each association type. Since this representation
is 1-dimensional, the matrices are relatively small, and it is possible to do these
calculations using rational arithmetic.
The lower right block of the RCF of the matrix representing the Lie-Yamaguti
identities consists of those rows whose leading 1s occur in a column corresponding
to a binary association type; this matrix has rank 11 and is displayed in Table 6.
In the RCF of the matrix representing the skew-symmetries of the binary asso-
ciation types, the columns which consist entirely of 0 correspond to those types T
whose skew-symmetries all have the form [ι]T + [σ]T where σ is an odd permutation
(see Table 4); this matrix has rank 10 and is displayed in Table 7.
10
MURRAY R. BREMNER
Row 4 of the matrix in Table 6 represents an identity satisfied by the bilinear
operation in every Lie-Yamaguti algebra which is not a consequence of anticommu-
tativity. This identity is stated explicitly in the next theorem.
Theorem. The following polynomial identity is satisfied by the bilinear operation
in every Lie-Yamaguti algebra over a field of characteristic 0 or p > 2, but is not a
consequence of anticommutativity:
Xσ∈S8
ε(σ)(cid:16) [[[[a, b], c], [d, e]], [[f, g], h]] − 3
− [[[[[a, b], c], d], [e, f ]], [g, h]] + [[[[[a, b], c], [d, e]], f ], [g, h]]
2 [[[[a, b], c], [[d, e], f ]], [g, h]]
+ 2 [[[[[a, b], c], d], [[e, f ], g]], h] + 3 [[[[[a, b], c], [[d, e], f ]], g], h]
+ 2 [[[[[[a, b], c], d], [e, f ]], g], h] − 2 [[[[[[a, b], c], [d, e]], f ], g], h](cid:17) ≡ 0,
where ε(σ) is the sign of σ ∈ S8 which is applied to a, b, c, d, e, f, g, h.
Corollary. Let LY (X) be the free Lie-Yamaguti algebra on the set X of generators
over a field of characteristic 0 or p > 2. Let BLY (X) the subalgebra of LY (X)
generated by X using only the bilinear operation. If X ≥ 8 then BLY (X) is not
a free anticommutative algebra.
References
[1] P. Benito, C. Draper, A. Elduque: Lie-Yamaguti algebras related to g2. J. Pure Appl.
Algebra 202 (2005), no. 1-3, 22 -- 54.
[2] P. Benito, A. Elduque, F. Mart´ın-Herce: Irreducible Lie-Yamaguti algebras. J. Pure
Appl. Algebra 213 (2009), no. 5, 795 -- 808.
[3] 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.
[4] M. R. Bremner, A. F. Douglas: The simple non-Lie Malcev algebra as a Lie-Yamaguti
algebra. J. Algebra 358 (2012) 269 -- 291.
[5] M. R. Bremner, I. R. Hentzel: Invariant nonassociative algebra structures on irreducible
representations of simple Lie algebras. Experiment. Math. 13 (2004), no. 2, 231 -- 256.
[6] M. R. Bremner, L. A. Peresi: Special identities for quasi-Jordan algebras. Comm. Algebra
39 (2011) no. 7, 2313 -- 2337.
[7] J. M. Clifton: A simplification of the computation of the natural representation of the
symmetric group Sn. Proc. Amer. Math. Soc. 83 (1981) no. 2, 248 -- 250.
[8] I. Correa, I. R. Hentzel, L. A. Peresi: Minimal identities of Bernstein algebras. Algebras
Groups Geom. 11 (1994), no. 2, 181 -- 199.
[9] M. K. Kinyon, A. Weinstein: Leibniz algebras, Courant algebroids, and multiplications on
reductive homogeneous spaces. Amer. J. Math. 123 (2001), no. 3, 525 -- 550.
[10] J.-L. Loday: Une version non commutative des alg`ebres de Lie:
les alg`ebres de Leibniz.
Enseign. Math. (2) 39 (1993), no. 3-4, 269 -- 293.
[11] A. I. Shirshov: Subalgebras of free commutative and free anticommutative algebras. Mat.
Sbornik N.S. 34 (1954) 81 -- 88.
[12] K. Yamaguti: On the Lie triple system and its generalization. J. Sci. Hiroshima Univ. Ser.
A 21 (1957/1958) 155 -- 160.
Department of Mathematics and Statistics, University of Saskatchewan, Canada
E-mail address: [email protected]
|
1812.08209 | 1 | 1812 | 2018-12-19T19:32:18 | On Generalized $(m, n)$-Jordan Derivations and Centralizers of Semiprime Rings | [
"math.RA"
] | In this paper we give an affirmative answer to two conjectures on generalized $(m,n)$-Jordan derivations and generalized $(m,n)$-Jordan centralizers raised in [S. Ali and A. Fo\v{s}ner, \textit{On Generalized $(m, n)$-Derivations and Generalized $(m, n)$-Jordan Derivations in Rings,} Algebra Colloq. \textbf{21} (2014), 411--420] and [A. Fo\v{s}ner, \textit{A note on generalized $(m,n)$-Jordan centralizers,} Demonstratio Math. \textbf{46} (2013), 254--262]. Precisely, when $R$ is a semiprime ring, we prove, under some suitable torsion restrictions, that every nonzero generalized $(m,n)$-Jordan derivation (resp., a generalized $(m,n)$-Jordan centralizer) is a derivation (resp., a two-sided centralizer). | math.RA | math |
On Generalized (m, n)-Jordan Derivations and
Centralizers of Semiprime Rings
Driss Bennis1,a, Basudeb Dhara2 and Brahim Fahid1,b
1: Centre de Recherche de Math´ematiques et Applications de Rabat
(CeReMAR), Faculty of Sciences, Mohammed V University in Rabat, Rabat,
Morocco
a: [email protected]; driss [email protected]
b: [email protected]
2: Department of Mathematics, Belda College, Belda,
Paschim Medinipur, 721424, W.B., India.
basu [email protected]
Abstract. In this paper we give an affirmative answer to two conjectures
on generalized (m, n)-Jordan derivations and generalized (m, n)-Jordan cen-
tralizers raised in [S. Ali and A. Fosner, On Generalized (m, n)-Derivations
and Generalized (m, n)-Jordan Derivations in Rings, Algebra Colloq. 21
(2014), 411 -- 420] and [A. Fosner, A note on generalized (m, n)-Jordan cen-
tralizers, Demonstratio Math. 46 (2013), 254 -- 262]. Precisely, when R is a
semiprime ring, we prove, under some suitable torsion restrictions, that ev-
ery nonzero generalized (m, n)-Jordan derivation (resp., a generalized (m, n)-
Jordan centralizer) is a derivation (resp., a two-sided centralizer).
2010 Mathematics Subject Classification. 16N60, 16W25
Key Words. semiprime ring, generalized (m, n)-derivation, generalized (m, n)-
Jordan derivation, (m, n)-Jordan centralizer, generalized (m, n)-Jordan centralizer
1
Introduction
Throughout this paper, R will represent an associative ring with center Z(R). We
denote by char(R) the characteristic of a prime ring R. Let n ≥ 2 be an integer.
2
D. Bennis, B. Dhara and B. Fahid
A ring R is said to be n-torsion free if, for all x ∈ R, nx = 0 implies x = 0. Recall
that a ring R is prime if, for any a, b ∈ R, aRb = {0} implies a = 0 or b = 0. A
ring R is called semiprime if, for any a ∈ R, aRa = {0} implies a = 0.
An additive mapping d : R −→ R is called a derivation, if d(xy) = d(x)y+xd(y)
holds for all x, y ∈ R, and it is called a Jordan derivation, if d(x2) = d(x)x + xd(x)
holds for all x ∈ R. An additive mapping T : R −→ R is called a left (resp., right)
centralizer if T (xy) = T (x)y (resp., T (xy) = xT (y)) is fulfilled for all x, y ∈ R,
and it is called a left (resp., right) Jordan centralizer if T (x2) = T (x)x (resp.,
T (x2) = xT (x)) is fulfilled for all x ∈ R. We call an additive mapping T : R −→ R
a two-sided centralizer (resp., a two-sided Jordan centralizer) if T is both a left as
well as a right centralizer (resp., a left and a right Jordan centralizer).
An additive mapping F : R −→ R is called a generalized derivation if F (xy) =
F (x)y + xd(y) holds for all x, y ∈ R, where d : R −→ R is a derivation. The
concept of generalized derivations was introduced by Bresar in [3] and covers both
the concepts of derivations and left centralizers. It is easy to see that generalized
derivations are exactly those additive mappings F which can be written in the
form F = d + T , where d is a derivation and T is a left centralizer.
The Jordan counterpart of the notion of generalized derivation was introduced
by Jing and Lu in [10] as follows: An additive mapping F : R −→ R is called a
generalized Jordan derivation if F (x2) = F (x)x + xd(x) is fulfilled for all x ∈ R,
where d : R −→ R is a Jordan derivation.
The study of relations between various sorts of derivations goes back to Her-
stein's classical result [9] which shows that any Jordan derivation on a 2-torsion
free prime ring is a derivation (see also [5] for a brief proof of Herstein's result).
In [7], Cusack generalized Herstein's result to 2-torsion free semiprime rings (see
also [2] for an alternative proof). Motivated by these classical results, Vukman
[17] proved that any generalized Jordan derivation on a 2-torsion free semiprime
ring is a generalized derivation.
In the last few years several authors have introduced and studied various sorts
of parameterized derivations. In [1], Ali and Fosner defined the notion of (m, n)-
derivations as follows: Let m, n ≥ 0 be two fixed integers with m + n 6= 0. An
additive mapping d : R −→ R is called an (m, n)-derivation if
(m + n)d(xy) = 2md(x)y + 2nxd(y)
(1.1)
holds for all x, y ∈ R.
Obviously, a (1, 1)-derivation on a 2-torsion free ring is a derivation.
In the same paper [1], a generalized (m, n)-derivation was defined as follows:
Let m, n ≥ 0 be two fixed integers with m + n 6= 0. An additive mapping D :
On Generalized (m, n)-Jordan Derivations and Centralizers
3
R −→ R is called a generalized (m, n)-derivation if there exists an (m, n)-derivation
d : R −→ R such that
(m + n)D(xy) = 2mD(x)y + 2nxd(y)
(1.2)
holds for all x, y ∈ R.
Obviously, every generalized (1, 1)-derivation on a 2-torsion free ring is a gen-
eralized derivation.
In [18], Vukman defined an (m, n)-Jordan derivation as follows: Let m, n ≥ 0
be two fixed integers with m + n 6= 0. An additive mapping d : R −→ R is called
an (m, n)-Jordan derivation if
(m + n)d(x2) = 2md(x)x + 2nxd(x)
(1.3)
holds for all x, y ∈ R.
Clearly, every (1, 1)-Jordan derivation on a 2-torsion free ring is a Jordan
derivation.
Recently, in [11], Kosi-Ulbl and Vukman proved the following result.
Theorem 1.1 ([11], Theorem 1.5) Let m, n ≥ 1 be distinct integers, R a mn(m+
n)m − n-torsion free semiprime ring an d : R −→ R an (m, n)-Jordan derivation.
Then d is a derivation which maps R into Z(R).
The (m, n)-generalized counterpart of the notion of an (m, n)-Jordan derivation
is introduced by Ali and Fosner in [1] as follows: Let m, n ≥ 0 be two fixed integers
with m + n 6= 0. An additive mapping F : R −→ R is called a generalized (m, n)-
Jordan derivation if there exists an (m, n)-Jordan derivation d : R −→ R such
that
(m + n)F (x2) = 2mF (x)x + 2nxd(x)
(1.4)
holds for all x, y ∈ R.
Based on some observations and inspired by the classical results, Ali and Fosner
in [1] made the following conjecture.
Conjecture 1.2 ([1], Conjecture 1) Let m, n ≥ 1 be two fixed integers, let R
be a semiprime ring with suitable torsion restrictions, and let F : R −→ R be a
nonzero generalized (m, n)-Jordan derivation. Then F is a derivation which maps
R into Z(R).
The first aim of this paper is to give an affirmative answer to this conjecture.
Namely, our first main result is the following theorem.
4
D. Bennis, B. Dhara and B. Fahid
Theorem 1.3 Let m, n ≥ 1 be distinct integers, let R be a k-torsion free semiprime
ring, where k = 6mn(m + n)m − n, and let F : R −→ R be a nonzero generalized
(m, n)-Jordan derivation. Then F is a derivation which maps R into Z(R).
On the other hand and in parallel, there are similar works which study relations
between various sorts of Jordan centralizers and centralizers. Namely, in [20], Zalar
proved that any left (resp., right) Jordan centralizer on a 2-torsion free semiprime
ring is a left (resp., right) centralizer. In [15], Vukman proved that, for a 2-torsion
free semiprime ring R, every additive mapping T : R −→ R satisfying the relation
"2T (x2) = T (x)x + xT (x) for all x ∈ R" is a two-sided centralizer. Motivated by
these results and inspired by his work [15], Vukman in [19] introduced the notion
of an (m, n)-Jordan centralizer as follows: Let m, n ≥ 0 be two fixed integers with
m+n 6= 0. An additive mapping T : R −→ R is called an (m, n)-Jordan centralizer
if
(m + n)T (x2) = mT (x)x + nxT (x)
(1.5)
holds for all x, y ∈ R.
Obviously, a (1, 0)-Jordan centralizer (resp., (0, 1)-Jordan centralizer) is a left
(resp., a right) Jordan centralizer. When n = m = 1, we recover the maps studied
in [15].
Based on some observations and results, Vukman conjectured that, on semiprime
rings with suitable torsion restrictions, every (m, n)-Jordan centralizer is a two-
sided centralizer (see [19]). Recently, this conjecture was solved affirmatively by
Kosi-Ulbl and Vukman in [12]. Namely, they proved the following result.
Theorem 1.4 ([12], Theorem 1.5) Let m, n ≥ 1 be distinct integers, let R be
an mn(m + n)-torsion free semiprime ring, and let T : R −→ R be an (m, n)-
Jordan centralizer. Then T is a two-sided centralizer.
Inspired by the work of Vukman [15, 19], Fosner [8] introduced more generalized
version of (m, n)-Jordan centralizers as follows: Let m, n ≥ 0 be two fixed integers
with m + n 6= 0. An additive mapping T : R −→ R is called a generalized (m, n)-
Jordan centralizer if there exists an (m, n)-Jordan centralizer T0 : R −→ R such
that
(m + n)T (x2) = mT (x)x + nxT0(x)
(1.6)
holds for all x ∈ R.
Thus, a generalized (1, 0)-Jordan centralizer is a left Jordan centralizer.
In [8], Fosner showed that, on a prime ring with a specific torsion condition,
every generalized (m, n)-Jordan centralizer is a two-sided centralizer. This led
Fosner to make the following conjecture.
On Generalized (m, n)-Jordan Derivations and Centralizers
5
Conjecture 1.5 ([8], Conjecture 1) Let m, n ≥ 1 be two fixed integers, let R
be a semiprime ring with suitable torsion restrictions, and let T : R −→ R be a
generalized (m, n)-Jordan centralizer. Then T is a two-sided centralizer.
The second aim of this paper is to give an affirmative answer to Fosner's
conjecture. Namely, our second main result is the following theorem.
Theorem 1.6 Let m, n ≥ 1 be two fixed integers, let R be an 6mn(m+n)(2n+m)-
torsion free semiprime ring, and let T : R −→ R be a nonzero generalized (m, n)-
Jordan centralizer. Then T is a two-sided centralizer.
2 Proof of the main theorems
In the proof of our main results, Theorems 1.3 and 1.6, we shall use the following
results.
Lemma 2.1 ([1], Lemma 1) Let m, n ≥ 0 be distinct integers with m + n 6= 0,
let R be a 2-torsion free ring, and let F : R −→ R be a nonzero generalized
(m, n)-Jordan derivation with an associated (m, n)-Jordan derivation d. Then,
(m + n)2F (xyx) = m(n − m)F (x)xy + m(3m + n)F (x)yx + m(m − n)F (y)x2 +
4mnxd(y)x+ n(n − m)x2d(y)+ n(m + 3n)xyd(x)+ n(m − n)yxd(x) for all x, y ∈ R.
Lemma 2.2 ([6], Theorem 3.3) Let n ≥ 2 be a fixed integer and let R be a
prime ring with char(R) = 0 or char(R) ≥ n. If T : R −→ R is an additive
mapping satisfying the relation T (xn) = T (x)xn−1 for all x ∈ R, then T (xy) =
T (x)y for all x, y ∈ R.
Lemma 2.3 ([8], Lemma 1) Let m, n ≥ 0 be distinct integers with m + n 6= 0,
let R be a ring, and let T : R −→ R be a nonzero generalized (m, n)-Jordan cen-
tralizer with an associated (m, n)-Jordan centralizer T0. Then, 2(m+n)2T (xyx) =
mnT (x)xy + m(2m + n)T (x)yx − mnT (y)x2 + 2mnxT0(y)x − mnx2T0(y) + n(m +
2n)xyT0(x) + mnyxT0(x) for all x, y ∈ R.
Lemma 2.4 ([16], Lemma 3) Let R be a semiprime ring and let T : R −→ R
be an additive mapping. If either T (x)x = 0 or xT (x) = 0 holds for all x ∈ R,
then T = 0.
We shall use the relation between semiprime rings and prime ideals. Namely,
it is well-known that a ring R is semiprime if and only if the intersection of all
prime ideals of R is zero if and only if R has no nonzero nilpotent (left, right)
ideals (see for instance Lam's book [13] or the recent book of Bresar [4]). Due to
6
D. Bennis, B. Dhara and B. Fahid
the classical Levitzki's paper [14], several authors prefer to refer to a such result
by Levitzki's lemma.
Let I be an ideal of R. For an element x ∈ R, we use x to denote the equivalence
class of x modulo I.
Lemma 2.5 Let R be both a 2-torsion free and a 3-torsion free semiprime ring
and let T : R −→ R be an additive map such that T (x)x3 = 0 and T (x4) = 0 for
all x ∈ R. Then T (xy) = T (x)y for all x, y ∈ R.
Proof. Let x, y ∈ R. We prove that T (xy) = T (x)y. We may assume that x and
y are not 0. Let P be a prime ideal of R and set R = R/P . Consider an element
p ∈ P . By hypothesis, 0 = T (x+ p)(x+ p)3 = (T (x)+ T (p))(x3 + xpx + px2 + p2x+
x2p + xp2 + pxp + p3). Thus, 0 = T (x)(xpx + px2 + p2x + x2p + xp2 + pxp + p3) +
T (p)(x3 + xpx + px2 + p2x + x2p + xp2 + pxp). Hence, T (p)x3 ∈ P , equivalently
T (p)x3 = 0. By Levitzki's lemma, T (p)x = 0, and then T (p) = 0 (since R is a
prime ring). Thus, T (P ) ⊆ P , which implies that T (x + P ) = T (x) + P . Then,
the induced map T : R/P → R/P such that T (x) = T (x) for every x ∈ R, is well
defined. Now, since T (x)x3 = 0 and T (x4) = 0, T (x4) = T (x)x3. This shows,
using Lemma 2.2, that T (xy) = T (x)y. Therefore, T (xy) − T (x)y ∈ P . Finally,
by the semiprimeness of R, we get the desired result.
Now we are ready to prove the first main result.
Proof of Theorem 1.3. Let d be the associated (m, n)-Jordan derivation of
F . Since R is a semiprime ring, d is a derivation which maps R into Z(R) (by
Theorem 1.1). Let us denote F − d by D. Then, we have (m + n)D(x2) = (m +
n)F (x2) − (m + n)d(x2) = 2mF (x)x + 2nxd(x) − 2md(x)x − 2nxd(x) = 2mD(x)x
for all x ∈ R. Thus
(m + n)D(x2) = 2mD(x)x, x ∈ R.
Replacing x with x2 in (2.1), we get
(m + n)D(x4) = 2mD(x2)x2, x ∈ R.
Multiplying by m + n and then using (2.1), we get
(m + n)2D(x4) = 4m2D(x)x3, x ∈ R.
(2.1)
(2.2)
(2.3)
On the other hand, putting x2 for y in the relation of Lemma 2.1 and using the
fact that D is a generalized (m, n)-Jordan derivation associated with the zero map
as an (m, n)-Jordan derivation, we get
(m+n)2D(x4) = m(n−m)D(x)x3 +m(3m+n)D(x)x3 +m(m−n)D(x2)x2, x ∈ R.
(2.4)
On Generalized (m, n)-Jordan Derivations and Centralizers
7
Multiplying both sides in (2.4) by 2 we get
2(m+n)2D(x4) = 2m(n−m)D(x)x3+2m(3m+n)D(x)x3+2m(m−n)D(x2)x2, x ∈ R.
(2.5)
Combining (2.2) and (2.5), we get
2(m+n)2D(x4) = 2m(n−m)D(x)x3+2m(3m+n)D(x)x3+(m+n)(m−n)D(x4), x ∈ R,
which gives
(m + n)(m + 3n)D(x4) = 4m(m + n)D(x)x3, x ∈ R.
Multiplying both sides in (2.7) by m + n, we get
(2.6)
(2.7)
(m + n)2(m + 3n)D(x4) = 4m(m + n)2D(x)x3, x ∈ R.
(2.8)
Multiplying by m + 3n in (2.3), we get
(m + n)2(m + 3n)D(x4) = 4m2(m + 3n)D(x)x3, x ∈ R.
(2.9)
By comparing (2.8) and (2.9), we get
4mn(m − n)D(x)x3 = 0, x ∈ R.
(2.10)
Since R is a 2mnn − m-torsion free ring, D(x)x3 = 0 for all x ∈ R. Applying
D(x)x3 = 0 in equation (2.3), we get (m + n)2D(x4) = 0 for all x ∈ R. By
using the torsion free restriction, we have D(x4) = 0 for all x ∈ R. Hence,
D(xy) = D(x)y for all x, y ∈ R (by Lemma 2.5). Applying this in (2.1), yields
(m + n)D(x)x = 2mD(x)x for all x ∈ R, equivalently (m − n)D(x)x = 0. Since
R is an m − n-torsion free ring, D(x)x = 0 for all x ∈ R. Therefore, by Lemma
2.4, D = 0. This completes the proof.
The second main result is proved similarly. Nevertheless, we include a proof
for completeness.
Proof of Theorem 1.6. Let T0 be the associated (m, n)-Jordan centralizer of T .
Since R is a semiprime ring, T0 is a two-sided centralizer (by Theorem 1.4). Let us
denote T −T0 by D. Then, we have (m+n)D(x2) = (m+n)T (x2)−(m+n)T0(x2) =
mT (x)x + nxT0(x) − mT0(x)x − nxT0(x) = mD(x)x for all x ∈ R. Thus
(m + n)D(x2) = mD(x)x, x ∈ R.
Replacing x with x2 in (2.11), we get
(m + n)D(x4) = mD(x2)x2, x ∈ R.
(2.11)
(2.12)
8
D. Bennis, B. Dhara and B. Fahid
Multiplying by m + n and then using (2.11), we get
(m + n)2D(x4) = m2D(x)x3, x ∈ R.
(2.13)
On the other hand, if we put y = x2 in the relation of Lemma 2.3, we get
2(m + n)2D(x4) = mnD(x)x3 + m(2m + n)D(x)x3 − mnD(x2)x2, x ∈ R. (2.14)
Multiplying both sides in (2.13) by 2 we get
2(m + n)2D(x4) = 2m2D(x)x3, x ∈ R.
(2.15)
Combining (2.12) and (2.14), we get
2(m+n)2D(x4) = mnD(x)x3+m(2m+n)D(x)x3−n(m+n)D(x4), x ∈ R, (2.16)
which implies
(m + n)(2m + 3n)D(x4) = 2m(m + n)D(x)x3, x ∈ R.
(2.17)
Multiplying both sides of above relation by m + n, we have
(m + n)2(2m + 3n)D(x4) = 2m(m + n)2D(x)x3, x ∈ R.
(2.18)
Multiplying by (2m + 3n) in (2.13), we get
(m + n)2(2m + 3n)D(x4) = m2(2m + 3n)D(x)x3, x ∈ R.
(2.19)
By comparing (2.18) and (2.19), we get
mn(2n + m)D(x)x3 = 0, x ∈ R.
(2.20)
Since R is a mn(2n + m)-torsion free ring, D(x)x3 = 0 for all x ∈ R. Applying
D(x)x3 = 0 in equation (2.13) and then using (m + n)-torsion freeness of R, we
get D(x4) = 0 for all x ∈ R. Moreover, since R is a 2 and a 3-torsion free ring,
by Lemma 2.5, we get D(xy) = D(x)y for all x, y ∈ R. Applying this in (2.11),
yields (m + n)D(x)x = mD(x)x for all x ∈ R. So nD(x)x = 0, which implies that
D(x)x = 0 for all x ∈ R. Therefore, by Lemma 2.4, D = 0. This completes the
proof.
Acknowledgements. The authors would like to thank Professor Abdellah Mamouni
for useful discussions.
On Generalized (m, n)-Jordan Derivations and Centralizers
9
References
[1] S. Ali and A. Fosner, On Generalized (m, n)-Derivations and Generalized
(m, n)-Jordan Derivations in Rings, Algebra Colloq. 21 (2014), 411 -- 420.
[2] M. Bresar, Jordan derivations on semiprime rings, Proc. Amer. Math. Soc.
104 (1988), 1003 -- 1006.
[3] M. Bresar, On the distance of the composition of two derivations to the
generalized derivations, Glasg. Math. J. 33 (1991), 89 -- 93.
[4] M. Bresar, Introduction to noncommutative algebra, Universitext, Springer,
2014.
[5] M. Bresar and J. Vukman, Jordan derivations on prime rings, Bull. Aust.
Math. Soc. 37 (1988), 321 -- 322.
[6] D. Benkovic and D. Eremita, Characterizing left centralizers by their action
on a polynomial, Publ. Math. Debercen 64 (2004), 343 -- 351.
[7] J. Cusak, Jordan derivations on rings, Proc. Amer. Math. Soc. 53 (1975),
321 -- 324.
[8] A. Fosner, A note on generalized (m, n)-Jordan centralizers, Demonstratio
Math. 46 (2013), 254 -- 262.
[9] I. N. Herstein, Jordan derivations of prime rings, Proc. Amer. Math. Soc. 8
(1957), 1104 -- 1119.
[10] W. Jing and S. Lu, Generalized Jordan derivations on prime rings and stan-
dard operator algebras, Taiwanese J. Math. 7 (2003), 605 -- 613.
[11] I. Kosi-Ulbl and J. Vukman, A note on (m, n)-Jordan derivation of rings
and banach algebras, Bull. Aust. Math. Soc. 93 (2016), 231 -- 237.
[12] I. Kosi-Ulbl and J.Vukman, On (m, n)-Jordan centralizers of semiprime
rings, Publ. Math. Debrecen 7490 (2016), 1 -- 9.
[13] T. Y. Lam, A first course in noncommutative rings, Graduate Texts in Math-
ematics, 123, Springer-Verlag, New York, 1991.
[14] J. Levitzki, Prime ideals and the lower radical, Amer. J. Math. 73 (1951),
25 -- 29.
[15] J. Vukman, An identity related to centralizers in semiprime rings, Comment.
Math. Univ. Carolin. 40 (1999), 447 -- 456.
10
D. Bennis, B. Dhara and B. Fahid
[16] J. Vukman, Identities with derivations and automorphisms on semiprime
rings, Int. J. Math. Math. Sci. 7 (2005), 1031 -- 1038.
[17] J. Vukman, A note on generalized derivations of semiprime rings, Taiwanese
J. Math. 11 (2007), 367 -- 370.
[18] J. Vukman, On (m, n)-Jordan derivations and commutativity of prime rings,
Demonstratio Math. 41 (2008), 773 -- 778.
[19] J. Vukman, On (m, n)-Jordan centralizers in rings and algebras, Glasg.
Math. J. 45 (2010), 43 -- 53.
[20] B. Zalar, On centralizers of semiprime rings, Comment. Math. Univ. Car-
olin. 32 (1991), 609 -- 614.
|
1209.4597 | 1 | 1209 | 2012-09-19T11:28:21 | A negative answer to the question of the linearity of Tate's Trace for the sum of two endomorphisms | [
"math.RA"
] | The aim of this note is to solve a problem proposed by J. Tate in 1968 by offering a counter-example of the linearity of the trace for the sum of two finite potent operators on an infinite-dimensional vector space. | math.RA | math |
A NEGATIVE ANSWER TO THE QUESTION
OF THE LINEARITY OF TATE'S TRACE
FOR THE SUM OF TWO ENDOMORPHISMS
JULIA RAMOS GONZ ´ALEZ (*)
FERNANDO PABLOS ROMO (**)
Abstract. The aim of this note is to solve a problem proposed by J. Tate in
1968 by offering a counter-example of the linearity of the trace for the sum of
two finite potent operators on an infinite-dimensional vector space.
Contents
Introduction
1.
2. Counter-example of the linearity property of the Tate's trace
References
1
2
4
1. Introduction
Let k be a fixed ground field and V a vector space over k. If we consider an
endomorphism ϕ of V , according to [3] we say that ϕ is "finite-potent" if ϕnV is
finite dimensional for some n, and a trace TrV (ϕ) ∈ k may be defined, having the
following properties:
(1) If V is finite dimensional, then TrV (ϕ) is the ordinary trace.
(2) If W is a subspace of V such that ϕW ⊂ W , then
TrV (ϕ) = TrW (ϕ) + TrV /W (ϕ) .
(3) If ϕ is nilpotent, then TrV (ϕ) = 0.
(4) If F is a "finite-potent" subspace of End(V ) (i.e., if there exists an n such
that for any family of n elements ϕ1, . . . , ϕn ∈ F , the space ϕ1 . . . ϕnV is
finite dimensional), then TrV : F −→ k is k-linear.
(5) If f : V ′ → V and g : V → V ′ are k-linear and f ◦ g is finite potent, then
g ◦ f is finite potent, and
TrV (f ◦ g) = TrV ′ (g ◦ f ) .
Remark 1.1. Properties (1), (2) and (3) characterize traces, because if W is a finite
dimensional subspace of V such that ϕW ⊆ W and ϕnV ⊆ W , for some n, then
TrV (ϕ) = TrW (ϕ). And, since ϕ is finite potent, we may take W = ϕnV .
Key words and phrases. vector space, trace, linearity.
2010 Mathematics Subject Classification: 15A03, 15A04
This work is partially supported by a collaboration-fellowship of the Spanish Government (*), and
by the DGESYC research contract no. MTM2009-11393 (**).
1
2
JULIA RAMOS GONZ ´ALEZ (*) FERNANDO PABLOS ROMO (**)
This trace is the main tool used by J. Tate in his elegant definition of the residue
offered in [3].
An open problem has been to determine whether this trace satisfies the linearity
property. In fact, in the article mentioned J. Tate wrote: "I doubt whether the rule
TrV θ1 + TrV θ2 = TrV (θ1 + θ2)
holds in general, i.e., whenever all three endomorphisms θ1, θ2 and θ1 +θ2 are finite-
potent, although I do not know a counter-example. (If a counter-example exists
at all, then there will be one with θ1 and θ2 nilpotent, because every finite-potent
endomorphism is the sum of a nilpotent one and one with finite range.)"
Lately, the second author of this note has studied this problem, solving the
question of the linearity with a negative answer for the case of the sum of three
finite potent endomorphisms.
Indeed, in [2] F. Pablos Romo offered three nilpotent endomorphisms θ1, θ2
and θ3 of an infinite-dimensional Z/2Z-vector space V , such that θ1 + θ2 + θ3 is a
finite-potent endomorphism of V and
TrV (θ1 + θ2 + θ3) = 1 .
Accordingly, in general Tate's trace does not satisfy the linearity property.
Moreover, M. Argerami, F. Szechtman and R. Tifenbach have given an alter-
native characterization of finite potent endomorphisms that can be used to reduce
the question to a special case. They have shown in [1] that an endomorphism ϕ
is finite potent if and only if V admits a ϕ-invariant decomposition V = Wϕ ⊕ Uϕ
such that ϕUϕ
: Wϕ −→ Wϕ is an
isomorphism. This decomposition is unique and one has that TrV (ϕ) = TrW (ϕWϕ
).
From this characterization, the question of the linearity of Tate's trace can be
reduced to a particular case of finite potent linear operators in the vector space
V = k < s, t > /I, where I is a left ideal of k < s, t > such that:
is nilpotent, Wϕ is finite dimensional, and ϕWϕ
• I contains the two-sided ideal generated by sn, tm, f (s+t) for some n, m ≥ 2
and f (x) ∈ k[x], where f (x) = xlg(x), l ≥ 2 and g(x) 6= 1 is monic and
relatively prime to x.
• If J is the right ideal of k < s, t > generated by (s+ t)l, then V1 = (J + I)/I
is finite dimensional and non-trivial.
• V is infinite dimensional.
The aim of this note is to give a negative answer to the question of the linearity
of Tate's trace for the sum of two endomorphisms by offering a counter-example
of this property. For this counter-example we consider a nilpotent endomorphism
over a space V of countable dimension with diagonal entries 1,0,0,0,.... -exhibited
in [1]-.
2. Counter-example of the linearity property of the Tate's trace
Let V be a vector space of countable dimension over an arbitrary ground field
k. Let {e1, e2, e3, . . . } be a basis of V indexed by the natural numbers.
Let us consider the linear operator θ1 of V defined in [1] by:
θ1(ei) = (cid:26) 0
ei−1
if i is odd
if i is even
.
If we denote by {vj}j∈N the basis constructed according to the following scheme:
v1 = e2, v2 = e2+e4, v3 = e1+e4, v2i = e2i+e2i+2, v2i+1 = e2i−1+e2i+2 for all i ≥ 1 ,
it is easy to check that the matrix associated with θ1 in the basis {vj}j∈N is:
LINEARITY OF TATE'S TRACE
3
1
1
0
θ1 ≡
· · ·
0 −1 0
0 −1
1
0
1
0 −1 · · ·
0
1
0
−1
0 −1
· · ·
0
0
1
1
0
0
0
0 −1 0
· · ·
0 −1 −1
1
0
0
0
· · ·
0
1
1
0
0 −1 · · ·
0
0 −1 −1
0
· · ·
1
0
0
1
0
· · ·
0 −1 −1 0
0
0
· · ·
1
0
0
0
...
...
...
...
. . .
Note that the explicit expression of θ1 in that basis is:
1
0
0
0
0
...
0
1
0
...
1
0
0
...
0
1
0
0
1
0
1
...
1
...
.
v1 − v2 + v3
v3 − v4 + v5
−v1 + v2 − v4 + v5
v2k+1 − v2k+2 + v2k+3
if i = 1
if i = 2
if i = 3
if i = 2k for k ≥ 2
(−1)kv1 + (−1)k+1v2 + [
i−1
X2j=4
j≥2
(−1)j+kv2j ] − vi+1 + vi+2
if i = 2k + 1 for k ≥ 2
θ1(vi) =
If {vj}j∈N is again the basis of V described above, let us now consider the linear
operator θ2 of V defined by:
θ2(vi) =
v1 − v3
v1 − v2 − v3 + v4 − v5
v1 − v2 + v4
if i = 1
if i = 2
if i = 3
v4k+2
i+1
if i = 4k for k ≥ 1
−v1 + v2 +
X2j=4
j≥2
0
(−1)j−1v2j
if i = 4k + 1 for k ≥ 1
if i = 4k + 2 for k ≥ 1
v1 − v2 + [
i+1
X2j=4
j≥2
(−1)jv2j] − vi+2
if i = 4k + 3 for k ≥ 1
A computation shows that θ2 is nilpotent of order 6 and its matrix with respect
to the basis {vj}j∈N is:
θ2 ≡
1
1
1
0
1
0
0
1
0
0 −1 0
0 −1 · · ·
1
· · ·
0 −1 0
0 −1 −1 0
0
0
−1 −1
· · ·
0
0 −1 · · ·
0 −1 0
0
1
0 −1
0
0
0
· · ·
· · ·
0 −1 0
1
0
0
0
0
0
0
0
· · ·
0 −1 · · ·
0
0
0
0
· · ·
0 −1 0
0
0
0
...
...
...
...
...
. . .
0
1
0
0
0
0
0
...
0
1
0
0
0
...
0
1
0
0
...
0
1
1
...
.
4
JULIA RAMOS GONZ ´ALEZ (*) FERNANDO PABLOS ROMO (**)
Thus, if ϕ = θ1 + θ2, one has that ϕ is a finite potent endomorphism of V ,
and the ϕ-invariant decomposition of V referred to above is V = Wϕ ⊕ Uϕ, with
Wϕ =< v1, v2 > and Uϕ =< vr >r≥3.
Regarding the basis {v1, v2}, it is clear that the isomorphism ϕWϕ
is
ϕWϕ
≡ (cid:18) 2
−1 −1(cid:19) .
1
Accordingly, the explicit expression of the linear operator ϕUϕ
in the basis
{vr}r≥3 is:
ϕUϕ
(vr) =
v5
if r = 3
v4k+1 + v4k+3
if r = 4k for k ≥ 1
v4k+3
if r = 4k + 1 for k ≥ 1
v4k+3 − v4k+4 + v4k+5
if r = 4k + 2 for k ≥ 1
0
if r = 4k + 3 for k ≥ 1
which is a nilpotent endomorphism of order 4, whose matrix in this basis is:
θ1 + θ2 ≡
.
0
0
1
0
0
0
0
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
1 1
0 0 −1 0 0
0 1
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
1 1
...
...
0
0
0
0
0
0
0
0
0
...
· · ·
· · ·
· · ·
· · ·
· · ·
· · ·
· · ·
· · ·
· · ·
. . .
Hence, θ1 and θ2 are nilpotent endomorphisms of V , and θ1 + θ2 is a finite potent
endomorphism with TrV (θ1 + θ2) = 1. Thus, we obtain a counter-example of the
linearity property of Tate's trace for finite potent endomorphisms and we solve the
above referred problem proposed by J. Tate in [3].
ACKNOWLEDGMENT
The first author wishes to thank the "Instituto Universitario de F´ısica Funda-
mental y Matem´aticas (IUFFyM)" of the University of Salamanca for a Research
Fellowship in the summer of 2011 that allowed her to explore this problem and
begin an approach to its solution.
References
[1] Argerami, M.; Szechtman, F.; Tifenbach, R. On Tate's trace, Linear Multilinear Alge-
bra 55(6), (2007) 515-520.
[2] Pablos Romo, F. On the linearity property of Tates trace, Linear Multilinear Algebra
55(4), (2007) 523-526.
[3] Tate, J. Residues of Differentials on Curves, Ann. Scient. ´Ec. Norm. Sup. 1, 4a s´erie,
(1968) 149-159.
Departamento de Matem´aticas, Universidad de Salamanca, Plaza de la Merced 1-4,
37008 Salamanca, Espana
E-mail address:
E-mail address:
(*) [email protected]
(**) [email protected]
|
0911.2903 | 2 | 0911 | 2010-02-16T18:30:46 | Alg\`ebres amass\'ees et applications | [
"math.RA",
"math.RT"
] | Sergey Fomin and Andrei Zelevinsky have invented cluster algebras at the beginning of this decade with the aim of creating an algebraic framework for the study of canonical bases in quantum groups and total positivity in algebraic groups. It soon turned out that the combinatorics of cluster algebras also appear in many other subjects and notably in the representation theory of quivers and finite-dimensional algebras. In this talk, we give a concise introduction to cluster algebras and sketch two significant applications concerning the study of certain discrete dynamical systems and the construction of dual semicanonical bases.
-----
Sergey Fomin et Andrei Zelevinsky ont invent\'e les alg\`ebres amass\'ees (cluster algebras) au d\'ebut des ann\'ees 2000 dans le but de fournir un cadre alg\'ebrique \`a l'\'etude des bases canoniques dans les groupes quantiques et de la positivit\'e totale dans les groupes alg\'ebriques. Il s'est av\'er\'e rapidement que la combinatoire des alg\`ebres amass\'ees intervenait \'egalement dans de nombreux autres sujets et notamment en th\'eorie des repr\'esentations des carquois et des alg\`ebres de dimension finie. Dans cet expos\'e, nous donnons une introduction concise aux alg\`ebres amass\'ees et esquissons deux applications significatives portant sur l'\'etude de certains syst\`emes dynamiques discrets et la construction de bases duales semi-canoniques. | math.RA | math |
Séminaire BOURBAKI
62ème année, 2009-2010, no 1014
Novembre 2009
ALGÈBRES AMASSÉES ET APPLICATIONS
[d'après Fomin-Zelevinsky, . . . ]
par Bernhard KELLER
INTRODUCTION
Les algèbres amassées (cluster algebras), inventées [38] par Sergey Fomin et Andrei
Zelevinsky au début des années 2000, sont des algèbres commutatives, dont les généra-
teurs et les relations sont construits de façon récursive. Parmi ces algèbres se trouvent
les algèbres de coordonnées homogènes sur les grassmanniennes, les variétés de drapeaux
et beaucoup d'autres variétés qui jouent un rôle important en géométrie et théorie des
représentations. La motivation principale de Fomin et Zelevinsky était de trouver un
cadre combinatoire pour l'étude des bases canoniques dont on dispose [68] [81] dans
ces algèbres et qui sont étroitement liées à la notion de positivité totale [82] dans les
variétés associées. Il s'est avéré rapidement que la combinatoire des algèbres amassées
intervenait également dans de nombreux autres sujets, par exemple dans
-- la géométrie de Poisson [52] [53] [54] [9] . . . ;
-- les systèmes dynamiques discrets [41] [69] [24] [61] . . . ;
-- les espaces de Teichmüller supérieurs [30] [31] [32] [33] . . . ;
-- la combinatoire et en particulier l'étude de polyèdres tels les associaèdres de Sta-
sheff [19] [18] [60] [76] [36] [37] [85] [86] . . . ;
-- la géométrie algébrique (commutative ou non commutative) et en particulier
l'étude des conditions de stabilité de Bridgeland [10], les algèbres Calabi-Yau [64]
[55], les invariants de Donaldson-Thomas [66] [75] [91] [46] . . . ;
-- et la théorie des représentations des carquois et des algèbres de dimension finie,
voir par exemple les articles de synthèse [3] [92] [93] [51] [70].
Nous renvoyons aux articles d'initiation [40] [104] [105] [106] [107] et au portail des
algèbres amassées [35] pour plus d'informations sur les algèbres amassées et leurs liens
avec d'autres sujets mathématiques (et physiques).
Dans cet exposé, nous donnons une introduction concise aux algèbres amassées (sec-
tion 1) et présentons deux applications :
-- la démonstration de la périodicité de certains systèmes dynamiques discrets,
d'après Fomin-Zelevinsky [41] et l'auteur [70] [73] (section 2.3) ;
1014 -- 02
-- la construction de bases duales semi-canoniques, d'après Geiss-Leclerc-Schröer [49]
(section 3.4).
Ces applications sont fondées sur la catégorification additive des algèbres amassées à
l'aide de catégories de représentations de carquois (avec relations). Nous en décrivons
les idées principales à la section 4. Nous y esquissons également des développements
récents importants liés à la catégorification monoïdale d'algèbres amassées [58] [87] et
à leur étude via les carquois à potentiel [23] [22].
1. DESCRIPTION ET PREMIERS EXEMPLES
1.1. Description
Une algèbre amassée est une Q-algèbre commutative munie d'un ensemble de généra-
teurs distingués (les variables d'amas) regroupés dans des parties (les amas) de cardinal
constant (le rang) qui sont construites récursivement par mutation à partir d'un amas
initial. L'ensemble des variables d'amas peut être fini ou infini.
Théorème 1.1 ([39]). -- Les algèbres amassées n'ayant qu'un nombre fini de variables
d'amas sont paramétrées par les systèmes de racines finis.
La classification est donc analogue à celle des algèbres de Lie semi-simples complexes.
Nous allons préciser le théorème (dans le cas simplement lacé) à la section 2.
1.2. Premier exemple
Pour illustrer la description et le théorème, présentons [107] l'algèbre amassée AA2
associée au système de racines A2. Par définition, elle est engendrée sur Q par les
variables d'amas xm, m ∈ Z, soumises aux relations d'échange
xm−1xm+1 = 1 + xm , m ∈ Z.
Ses amas sont par définition les paires de variables consécutives {xm, xm+1}, m ∈ Z.
L'amas initial est {x1, x2} et deux amas sont reliés par une mutation si et seulement si
ils ont exactement une variable d'amas en commun.
Les relations d'échange permettent d'exprimer toute variable d'amas comme fonction
rationnelle des variables initiales x1, x2 et donc d'identifier l'algèbre AA2 à une sous-
algèbre du corps Q(x1, x2). Afin d'expliciter cette sous-algèbre, calculons les xm pour
m ≥ 3. Nous avons :
(1.2.1)
(1.2.2)
(1.2.3)
x3 =
x4 =
x5 =
1 + x2
x1
1 + x3
x2
1 + x4
x3
=
=
x1 + 1 + x2
x1x2
x1x2 + x1 + 1 + x2
x1x2
÷
1 + x2
x1
=
1 + x1
x2
.
1014 -- 03
Notons que, contrairement à ce qu'on pourrait attendre, le dénominateur dans 1.2.3
reste un monôme ! En fait, toute variable d'amas dans une algèbre amassée quelconque
est un polynôme de Laurent, voir le théorème 2.1. Continuons le calcul :
(1.2.4)
(1.2.5)
x6 =
1 + x5
x4
=
x2 + 1 + x1
÷
x1 + 1 + x2
x1x2
= x1
x7 = (1 + x1) ÷
= x2.
x2
1 + x1
x2
Il est alors clair que la suite des xm, m ∈ Z, est 5-périodique et que le nombre de
variables d'amas est effectivement fini et égal à cinq. Outre les deux variables initiales
x1 et x2 nous avons trois variables non initiales x3, x4 et x5. En examinant leurs
dénominateurs, nous voyons qu'elles sont en bijection naturelle avec les racines positives
α1, α1 + α2, α2 du système de racines de type A2. Ceci se généralise à tout diagramme
de Dynkin, voir le théorème 2.1.
1.3. Algèbres amassées de rang 2
À tout couple d'entiers positifs (b, c) est associée une algèbre amassée A(b,c). On la
définit de la même manière que AA2, mais en remplaçant les relations d'échange par
xm−1xm+1 =(cid:26) xb
m + 1 si m est impair,
m + 1 si m est pair.
xc
L'algèbre A(b,c) n'a qu'un nombre fini de variables d'amas si et seulement si bc ≤ 3,
autrement dit si la matrice
(cid:20) 2 −b
2 (cid:21)
−c
est la matrice de Cartan d'un système de racines Φ de rang 2. Le lecteur pourra s'amuser
à vérifier que, dans ce cas, les variables d'amas non initiales sont toujours paramétrées
par les racines positives de Φ.
2. ALGÈBRES AMASSÉES ASSOCIÉES AUX CARQUOIS
2.1. Mutation des carquois
Un carquois est un graphe orienté, c'est-à-dire un quadruplet Q = (Q0, Q1, s, t) formé
d'un ensemble de sommets Q0, d'un ensemble de flèches Q1 et de deux applications s
et t de Q1 dans Q0 qui, à une flèche α, associent respectivement sa source et son but.
En pratique, on représente un carquois par un dessin comme dans l'exemple qui suit :
Q :
λ
1
3
ν
µ
^>>>>>>>>
/ 2
α
5
β
γ
/ 6
4.
&
&
/
/
/
/
/
/
/
/
^
o
o
1014 -- 04
Une flèche α dont la source et le but coïncident est une boucle ; un 2-cycle est un
couple de flèches distinctes β et γ telles que s(β) = t(γ) et t(β) = s(γ). De même,
on définit les n-cycles pour tout entier positif n. Un sommet i d'un carquois est une
source (respectivement un puits) s'il n'existe aucune flèche de but i (respectivement de
source i).
Appelons bon carquois un carquois fini sans boucles ni 2-cycles dont l'ensemble des
sommets est l'ensemble des entiers 1, . . . , n pour un entier positif n. À un isomorphisme
fixant les sommets près, un tel carquois Q est donné par la matrice antisymétrique
B = BQ dont le coefficient bij est la différence entre le nombre de flèches de i à j et
le nombre de flèches de j à i pour tous 1 ≤ i, j ≤ n. Réciproquement, toute matrice
antisymétrique B à coefficients entiers provient d'un bon carquois QB.
Soient Q un bon carquois et k un sommet de Q. La carquois muté µk(Q) est le
carquois obtenu à partir de Q comme suit :
(1) pour tout sous-carquois i
β
/ k
α
/ j , on rajoute une nouvelle flèche [αβ] :
i → j ;
(2) on renverse toutes les flèches de source ou de but k ;
(3) on supprime les flèches d'un ensemble maximal de 2-cycles disjoints deux à deux.
Si B est la matrice antisymétrique associée à Q et B ′ celle associée à µk(Q), on a
ij =(cid:26) −bij
b′
bij + sgn(bik) max(0, bikbkj)
si i = k ou j = k ;
sinon.
C'est la règle de mutation des matrices antisymétriques (plus généralement : anti-
symétrisables) introduite par Fomin-Zelevinsky dans [38], voir aussi [42].
On vérifie sans peine que µk est une involution. Par exemple, les carquois
(2.1.1)
1
E
22222222
2
et
3
2
1
Y33333333
3
sont reliés par la mutation par rapport au sommet 1. Notons que, du point de la théorie
des représentations, ces carquois sont très différents. Deux carquois sont équivalents par
mutation s'ils sont reliés par une suite finie de mutations. On vérifie facilement, par
exemple à l'aide de [71], que les trois carquois suivants sont équivalents par mutation
(2.1.2)
10
1
1
5
E
333
E
222
F
222
3
9
222
F
F
222
2
8
4
F
7
6
222
10
-\\
5
!!
7
8
(RR
m\\\
9
!!
2
5
jUU
~}}
6
4
66
xx
10
*UU
9
6
Z55
5ll
4
X11
3
7
$$
8
.
3
Q$$
2
1
/
/
E
o
o
Y
E
o
o
F
F
o
o
F
o
o
E
F
o
o
o
o
o
o
X
Z
5
-
m
(
"
"
Q
<
<
j
~
*
>
>
1014 -- 05
La classe de mutation commune de ces carquois comporte 5739 carquois (à isomor-
phisme près). La classe de mutation de la «plupart» des carquois est infinie. La classi-
fication des carquois ayant une classe de mutation finie est un problème difficile, résolu
récemment par Felikson-Shapiro-Tumarkin [28] : outre les carquois à deux sommets et
les carquois associés à des surfaces à bord marquées [37], la liste contient 11 carquois
exceptionnels, dont le plus grand est dans la classe de mutation des carquois 2.1.2.
2.2. Mutation des graines, algèbres amassées
Soient n ≥ 1 un entier et F le corps Q(x1, . . . , xn) engendré par n indéterminées
x1, . . . , xn. Une graine (appelée aussi X-graine) est un couple (R, u), où R est un bon
carquois et u une suite u1, . . . , un qui engendre librement le corps F. Si (R, u) est une
graine et k un sommet de R, la mutation µk(R, u) est la graine (R′, u′), où R′ = µk(R)
et u′ est obtenu à partir de u en remplaçant l'élément uk par l'élément u′
k défini par la
relation d'échange
(2.2.1)
u′
kuk = Ys(α)=k
ut(α) + Yt(α)=k
us(α).
On vérifie que µ2
k(R, u) = (R, u). Par exemple, les mutations de la graine
1
/ 2
/ 3 , {x1, x2, x3}
par rapport aux sommets 1 et 2 sont les graines
(2.2.2)
1
2
/ 3 , {
, x2, x3}
et
1
2
3
, {x1,
1 + x2
x1
x1 + x3
x2
, x3}.
Fixons maintenant un bon carquois Q. La graine initiale est (Q, {x1, . . . , xn}). Un
amas associé à Q est une suite u qui apparaît dans une graine (R, u) obtenue à partir de
la graine initiale par mutation itérée. Les variables d'amas sont les éléments des amas.
L'algèbre amassée AQ est la sous-algèbre de F engendrée par les variables d'amas.
Clairement, si (R, u) est une graine associée à Q, l'isomorphisme naturel
Q(u1, . . . , un) ∼→ Q(x1, . . . , xn)
induit un isomorphisme de AR sur AQ qui préserve les variables d'amas et les amas.
L'algèbre amassée AQ est donc un invariant de la classe de mutation de Q.
Il est
utile d'introduire un objet combinatoire qui code la construction récursive des graines :
le graphe d'échange. Par définition, ses sommets sont les classes d'isomorphisme de
graines (les isomorphismes renumérotent les sommets et les variables) et ses arêtes
correspondent aux mutations. Par exemple, le graphe d'échange obtenu à partir du
/
/
o
o
/
'
'
o
o
o
o
1014 -- 06
carquois Q : 1
/ 2
/ 3 est le 1-squelette de l'associaèdre de Stasheff [97] [19]:
◦
◦
----
◦
7654
01232
PPPPPPPPPPPPP
$$$$$$$$$$$
111
uuuu
◦
7654
01231
??
???
7654
01233
◦
tttttttttttttttt
◦
,,,
◦
◦
qqqqqqqqqq
*****
◦
MMMMM
::::::::::::
◦
◦
Le sommet 1 correspond à la graine initiale et les sommets 2 et 3 aux graines 2.2.2.
Fixons un bon carquois connexe Q. Si son graphe sous-jacent est un diagramme de
Dynkin simplement lacé de type ∆, nous disons que Q est un carquois de Dynkin de
type ∆.
Théorème 2.1 ([39]). -- (a) Toute variable d'amas de AQ est un polynôme de Lau-
rent à coefficients entiers [38].
(b) L'algèbre amassée AQ n'a qu'un nombre fini de variables d'amas si et seulement si
Q est équivalent par mutation à un carquois de Dynkin. Dans ce cas, le graphe ∆
sous-jacent à ce carquois est unique et s'appelle le type amassé de Q.
(c) Si Q est un carquois de Dynkin de type ∆, alors les variables d'amas non initiales
de AQ sont en bijection avec les racines positives du système de racines Φ de ∆ ;
plus précisément, si α1, . . . , αn sont les racines simples, alors pour toute racine
positive α = d1α1 + · · · + dnαn, il existe une unique variable d'amas non initiale
Xα de dénominateur xd1
1 · · · xdn
n .
Un monôme d'amas est un produit de puissances positives de variables d'amas qui ap-
partiennent toutes au même amas. La construction d'une «base canonique» de l'algèbre
amassée AQ est un problème important et encore très largement ouvert, voir par ex-
emple [96] [26] [17]. On s'attend à ce qu'une telle base contienne tous les monômes
d'amas, d'où la conjecture :
Conjecture 2.2 ([39]). -- Les monômes d'amas sont linéairement indépendants sur
le corps Q.
Si Q est un carquois de Dynkin, on sait [15] que les monômes d'amas forment une
base de AQ. Si Q est acyclique, c'est-à-dire n'admet aucun cycle orienté, la conjec-
ture résulte d'un théorème de Geiss-Leclerc-Schröer [47], qui montrent l'existence d'une
«base générique» contenant les monômes d'amas. La conjecture a aussi été démontrée
pour des classes d'algèbres amassées à coefficients (voir la section 3), par exemple dans
les travaux [44] [47] [21].
/
/
1014 -- 07
Conjecture 2.3 ([39]). -- Les variables d'amas s'écrivent comme des polynômes de
Laurent à coefficients entiers positifs en les variables de tout amas.
La catégorification monoïdale développée par Leclerc [80] et Hernandez-Leclerc [58]
(voir la section 4.3) a permis récemment de montrer cette conjecture d'abord pour les
carquois de type An et D4, voir [58], puis pour tout carquois admettant une orientation
bipartite [87], c'est-à-dire une orientation où tout sommet est une source ou un puits.
Elle est démontrée de façon combinatoire par Musiker-Schiffler-Williams [86] pour tous
les carquois associés à des surfaces à bord marquées [37] et par Di Francesco-Kedem
[25] pour les carquois associés au T -système de type A.
Nous renvoyons à [40] et [42] pour de nombreuses autres conjectures sur les algèbres
amassées et à [22] pour la solution d'une bonne partie de ces conjectures grâce à la
catégorification (voir la section 4.4).
2.3. Y -graines, application à la conjecture de périodicité
Soient n ≥ 1 un entier et G le corps Q(y1, . . . , yn) engendré par des indéterminées yi.
Une Y -graine est un couple (R, v), où R est un bon carquois et v une suite v1, . . . , vn
qui engendre librement le corps G (nous nous écartons quelque peu de la définition dans
[42]). Si (R, v) est une Y -graine et k un sommet de R, la mutation µk(R, v) est la
Y -graine (R′, v′), où R′ = µk(R) et
v′
i =
v−1
i
vi(1 + vk)m
vi(1 + v−1
vi
si i = k,
si le nombre de flèches i → k est m ≥ 1,
k )−m si le nombre de flèches k → i est m ≥ 1,
sinon.
Par exemple, les Y -graines obtenues à partir de y1 → y2 par mutation sont, en écrivant
les variables vi à la place des sommets i :
y2
1 + y1 + y1y2
1 + y1
y2 (cid:19) µ17→(cid:18)1 + y1 + y1y2
y2
→
1
y1(1 + y2)(cid:19)
←
(cid:18) 1
y1
←
y1y2
1 + y1(cid:19) µ27→(cid:18)
µ27→(cid:18) 1
y2
→ y1(1 + y2)(cid:19) µ17→ (y2 ← y1) .
Les Y -graines jouent un rôle important dans la théorie de Teichmüller supérieure de
Fock-Goncharov [30] [34] et dans l'étude par Kontsevich-Soibelman [75] des invariants
de Donaldson-Thomas des carquois à potentiel [23]. Elles sont liées aux X-graines par
des conjectures de dualité [30] étudiées systématiquement par Fomin-Zelevinsky dans
[42]. En particulier, dans [42], les auteurs montrent que les deux types de graines se
déterminent mutuellement si, en même temps que AQ, on considère aussi A eQ, où eQ est
l'extension principale de Q obtenue à partir de Q en rajoutant de nouveaux sommets
n + 1, . . . , 2n et une nouvelle flèche (n + i) → i pour tout 1 ≤ i ≤ n. Ces liens
combinés avec la catégorification additive (voir section 4.1) ont permis récemment une
application des algèbres amassées à l'étude de systèmes dynamiques discrets issus de la
physique mathématique.
1014 -- 08
Soient ∆ et ∆′ deux diagrammes de Dynkin simplement lacés. Notons 1, . . . , n et
1, . . . , n′ leurs sommets et A et A′ leurs matrices d'incidence, le coefficient en position
(i, j) valant 1 s'il existe une arête entre i et j et 0 sinon. Notons h et h′ les nombres de
Coxeter de ∆ et ∆′. Le Y -système associé à ∆ et ∆′ est un système infini d'équations
de récurrence en des variables Yi,j,t associées aux sommets (i, j) du produit ∆ × ∆′ et
dépendant d'un paramètre de temps discret t ∈ Z. Les équations du Y -système sont
Yi,i′,t−1Yi,i′,t+1 = Qn
Qn′
j=1(1 + Yj,i′,t)aij
i,j ′,t)a′
j ′=1(1 + Y −1
i′ j′
(2.3.1)
,
pour tout sommet (i, i′) du produit et tout entier t.
Théorème 2.4. -- Toutes les solutions du Y -système sont périodiques par rapport au
paramètre t de période divisant 2(h + h′).
Ce théorème vient confirmer la «conjecture de périodicité» formulée par Al. B. Za-
molodchikov [103, (12)] pour ∆′ = A1, par Kuniba-Nakanishi [77, (2a)] pour ∆′ = Am
et par Ravanini-Valleriani-Tateo [90, (6.2)] dans le cas général. Le théorème a été
démontré
-- pour (An, A1) par Frenkel-Szenes [43] (qui donnent des solutions explicites) et par
Gliozzi-Tateo [56] (à l'aide de calculs de volumes de 3-variétés) ;
-- par Fomin-Zelevinsky [41] pour (∆, A1), où ∆ n'est pas nécessairement simplement
lacé (ils utilisent les méthodes de leur théorie des algèbres amassées et un calcul
sur ordinateur pour les types exceptionnels; ce calcul peut être évité maintenant
grâce à [102]) ;
-- pour (An, Am) par Volkov [100], qui construit des solutions explicites grâce à des
considérations de géométrie projective élémentaire, et par Szenes [98], qui inter-
prète le système comme un système de connexions plates sur un graphe; la démon-
stration d'un énoncé équivalent est due à Henriques [57] ;
-- pour (∆, ∆′) quelconques dans [70] [73] à l'aide de la catégorification additive, voir
la section 4.1.
Pour des diagrammes non simplement lacés ∆ et ∆′, deux variantes généralisées de
la conjecture existent :
la première se ramène au théorème 2.4 par la technique du
«pliage», voir [41] ; la deuxième, formulée par Kuniba-Nakanishi [77, (2a)] et Kuniba-
Nakanishi-Suzuki [78, B.6], fait intervenir le double de la somme des nombres de Coxeter
duaux (voir par exemple le chapitre 6 de [67]) ; elle a été démontrée dans [62] [63].
3. ALGÈBRES AMASSÉES À COEFFICIENTS
Nous allons généraliser légèrement la définition donnée à la section 2 pour obtenir
la classe des «algèbres amassées antisymétriques de type géométrique». Cette classe
1014 -- 09
contient de nombreuses algèbres d'origine géométrique munies de «bases duales semi-
canoniques». La construction d'une grande partie d'une telle base est l'une des appli-
cations les plus remarquables des algèbres amassées.
Nous renvoyons à [42] pour la définition des «algèbres amassées antisymétrisables à
coefficients dans un semi-corps», qui constituent la classe la plus générale considérée
jusqu'à maintenant.
3.1. Définition
bon carquois à m sommets et qui ne comporte aucune flèche entre sommets i, j tous les
Q dont les sommets sont 1, . . . , n (un sous-carquois est plein si, avec deux sommets, il
contient toutes les flèches qui les relient). Les sommets n + 1, . . . , m sont les sommets
Soient 1 ≤ n ≤ m des entiers. Soit eQ un carquois glacé de type (n, m), c'est-à-dire un
deux strictement plus grands que n. La partie principale de eQ est le sous-carquois plein
gelés. L'algèbre amassée associée au carquois glacé eQ
A eQ ⊂ Q(x1, . . . , xm)
est définie de la même façon que l'algèbre amassée associée à un carquois (section 2)
sauf que
-- seules les mutations par rapport à des sommets non gelés sont admises et aucune
flèche entre sommets gelés n'est introduite lors des mutations ;
-- les variables xn+1, . . . , xm, qui font partie de tous les amas, sont appelées coeffi-
cients plutôt que variables d'amas ;
-- le type amassé du carquois glacé est celui de sa partie principale (s'il est défini).
Souvent, on considère des localisations de A eQ obtenues en inversant certains des coef-
ficients. Si K est une extension de Q et A une K-algèbre (associative avec 1), une
structure d'algèbre amassée à coefficients de type eQ sur A est la donnée d'un isomor-
phisme ϕ de A eQ ⊗Q K sur A. Un tel isomorphisme est déterminé par les images des
coefficients et des variables de la graine initiale ϕ(xi), 1 ≤ i ≤ m. Nous appellerons
graine initiale de A la donnée du carquois eQ et des ϕ(xi).
3.2. Exemple : le cône sur la grassmannienne des plans d'un espace vectoriel
Soit n ≥ 1 un entier. Soit A l'algèbre des fonctions polynomiales sur le cône au-
dessus de la grassmannienne des plans de Cn+3. Cette algèbre est engendrée par les
coordonnées de Plücker xij, 1 ≤ i < j ≤ n + 3, assujetties aux relations de Plücker :
pour tout quadruplet d'entiers i < j < k < l compris entre 1 et n + 3, nous avons
(3.2.1)
xikxjl = xijxkl + xjkxil.
1014 -- 10
Notons que les monômes dans cette relation sont naturellement associés aux diagonales
et aux côtés du carré
i
l
<<<<<<<<
j
k
L'idée est d'interpréter cette relation comme une relation d'échange dans une algèbre
amassée (à coefficients). Pour décrire cette algèbre, considérons, dans le plan affine
euclidien, un polygone régulier P dont les sommets sont numérotés de 1 à n + 3. Con-
sidérons la variable xij comme associée au segment [ij] joignant les sommets i et j.
Proposition 3.1 ([39, Example 12.6]). -- L'algèbre A a une structure d'algèbre
amassée à coefficients telle que
- les coefficients soient les variables xij associées aux côtés de P ;
- les variables d'amas soient les variables xij associées aux diagonales de P ;
- les amas soient les n-uplets de variables d'amas correspondant à des diagonales
qui forment une triangulation de P .
En outre, les relations d'échange sont exactement les relations de Plücker et le type
amassé est An.
Une triangulation de P détermine une graine initiale pour l'algèbre amassée et les
relations d'échange vérifiées par les variables d'amas initiales déterminent le carquois
glacé eQ. Par exemple, on vérifie que, dans le dessin suivant, la triangulation et le
carquois glacé (dont les sommets gelés sont entourés de boîtes) se correspondent
5
4
oooooo
NNNNNN
0
3
OOOOOO
4444444444
oooooo
1
2
45
05
U,,,
04
34
01
02
23
03
gOO
?
woo
??
12
De nombreuses autres algèbres de coordonnées (homogènes) de variétés algébriques
classiques admettent également des structures d'algèbres amassées (supérieures, voir la
section 3.3), notamment les grassmanniennes [95] et les doubles cellules de Bruhat [7].
Certaines de ces algèbres n'ont qu'un nombre fini de variables d'amas et donc un type
amassé bien défini. Voici quelques exemples extraits de [40], où N est un sous-groupe
unipotent maximal :
Gr2,n+3 Gr3,6 Gr3,7 Gr3,8 SL3/N SL4/N SL5/N Sp4/N SL2 SL3
A1 D4
D4
D6
An
B2
E6
E8
A1
A3
Un analogue de la proposition 3.1 pour les doubles cellules de Bruhat réduites [8] est
dû à Yang et Zelevinsky [102]. Ils obtiennent ainsi une algèbre amassée (à coefficients
O
O
O
O
O
O
U
g
/
/
w
?
O
O
/
/
1014 -- 11
principaux) avec une description explicite des variables d'amas pour tout diagramme
de Dynkin.
3.3. Exemple : le sous-groupe unipotent maximal de SL(n + 1, C)
Soient n un entier positif et N le sous-groupe de SL(n + 1, C) formé des matrices
triangulaires supérieures dont les coefficients diagonaux sont tous égaux à 1. Pour
1 ≤ i, j ≤ n + 1 et g ∈ N, soit Fij(g) la sous-matrice carrée de taille maximale de g qui
comporte le coefficient gij dans son coin inférieur gauche. Soit fij(g) le déterminant de
Fij(g). Nous considérons les fonctions polynomiales fij : N → C pour 1 ≤ i ≤ n et
est la sous-algèbre de Q(x1, . . . , xm) formée des éléments qui s'expriment comme des
i + j ≤ n + 2. L'algèbre amassée supérieure associée à un carquois glacé eQ à m sommets
polynômes de Laurent en les variables de tout amas associé à eQ.
Théorème 3.1 ([7]). -- L'algèbre des fonctions polynomiales C[N] a une structure
d'algèbre amassée supérieure dont la graine initiale est donnée par
f12
/ f13
f14
/ . . .
/ f1,n+1
~~~~~~~~~~
/ f23
/ . . .
f2,n
/ . . .
f22
...
fn,2
Il n'est pas difficile de vérifier que cette structure est de type amassé A3 pour n = 3,
D6 pour n = 4 et qu'elle a une infinité de variables d'amas pour n ≥ 5.
Un théorème de Fekete [27], généralisé dans [6], affirme qu'une matrice carrée à n + 1
lignes et à coefficients réels est totalement positive (i.e. tous ses mineurs sont > 0)
si les (n + 1)2 mineurs suivants sont strictement positifs : tous les mineurs formés des
k premières lignes et de k colonnes consécutives pour 1 ≤ k ≤ n+1. Une matrice g ∈ N
à coefficients réels est totalement positive si tous les mineurs non identiquement nuls
sur N sont strictement positifs en g. Cette condition est équivalente à ce que l'on ait
fij(g) > 0 pour les fij de la graine initiale du théorème. Comme les relations d'échange
ne font pas intervenir de soustraction, tout amas non initial C donne également un
critère de positivité : la matrice g ∈ N est totalement positive si et seulement si l'on a
uij(g) > 0 pour toute variable d'amas uij dans C.
/
/
/
/
/
O
O
/
O
O
/
O
O
/
~
~
O
O
1014 -- 12
3.4. Exemple d'application aux bases duales semi-canoniques
L'exemple 3.3 se généralise. Soit, en effet, G un groupe algébrique semi-simple com-
plexe et N un sous-groupe unipotent maximal de G. Alors d'après [7], l'algèbre C[N]
est munie d'une structure canonique d'algèbre amassée supérieure. Soit n l'algèbre
de Lie du groupe algébrique N. Dans [83], Lusztig construit une base distinguée, la
base semi-canonique, de (l'espace vectoriel complexe sous-jacent à) l'algèbre envelop-
pante U(n). Le dual restreint de la cogèbre U(n) est canoniquement isomorphe à C[N],
qui est donc muni de la base duale de celle construite par Lusztig, appelée base duale
semi-canonique.
Théorème 3.2 (Geiss-Leclerc-Schröer [49]). -- Tout monôme d'amas de C[N] (munie
de la structure d'algèbre amassée du théorème 3.1) fait partie de la base duale semi-
canonique.
Notons que ce théorème implique la conjecture 2.2 sur l'indépendance des monômes
d'amas pour cette classe d'algèbres amassées. L'algèbre U(n) est également munie de
la base canonique obtenue par spécialisation à partir de la base canonique [68] [81] du
groupe quantique Uq(n).
Théorème 3.3 (Geiss-Leclerc-Schröer [48]). -- Pour G = SL(n + 1, C), la base cano-
nique coïncide avec la base semi-canonique de U(n) si et seulement si n ≤ 4.
Néanmoins, Geiss-Leclerc-Schröer conjecturent qu'au moins les «parties rigides» des
bases duales canonique et semi-canonique coïncident. Plus précisément, un cas parti-
culier de la conjecture 23.2 de [47] nous donne la conjecture qui suit.
Conjecture 3.4 (Geiss-Leclerc-Schröer [47]). -- Tout monôme d'amas de C[N] ap-
partient aussi à la base duale canonique.
Dans un travail de longue haleine qui a abouti à [47], Geiss-Leclerc-Schröer ont
généralisé les théorèmes 3.1 et 3.2 de l'algèbre C[N] aux algèbres de coordonnées de
cellules unipotentes de groupes de Kac-Moody simplement lacés. Nous renvoyons à [51]
pour une introduction et une synthèse des résultats dans le cas fini, et à [21] pour une
extension (partielle) au cas non simplement lacé.
4. CATÉGORIFICATIONS
Les démonstrations du théorème de périodicité (Théorème 2.4) et du théorème sur la
base duale semi-canonique (Théorème 3.2) s'appuient sur la «catégorification additive»
des algèbres amassées ; celle des cas connus de la conjecture de positivité 2.3 sur la «caté-
gorification monoïdale». Nous allons esquisser les idées principales de ces méthodes.
La section 4.4 est consacrée à la méthode de la catégorification à l'aide des «carquois
à potentiel». À ce jour, c'est la seule méthode qui permette de traiter des algèbres
amassées associées à des carquois finis arbitraires (sans boucles ni 2-cycles).
1014 -- 13
4.1. Catégorification additive : la catégorie amassée
Soit Q un carquois fini d'ensemble de sommets {1, . . . , n}. Un chemin de Q est une
composition formelle (jαs . . . α1i) d'un nombre s positif ou nul de flèches αi telle
que s(αi) = t(αi−1) pour 1 ≤ i ≤ s. En particulier, pour tout sommet i, nous avons
le chemin paresseux ei = (ii) de longueur nulle, neutre pour la composition naturelle
des chemins. Une représentation (complexe) de Q est la donnée V d'espaces vectoriels
complexes de dimension finie Vi, i ∈ Q0, et d'applications linéaires Vα : Vi → Vj
pour toute flèche α : i → j de Q. Une représentation est donc un diagramme d'espaces
vectoriels de la forme donnée par Q. Un morphisme de représentations est un morphisme
de diagrammes. On obtient ainsi la catégorie rep(Q) des représentations de Q. C'est
une catégorie abélienne équivalente à la catégorie des modules de C-dimension finie sur
une algèbre, à savoir l'algèbre des chemins CQ (une base de cette algèbre est formée
des chemins de Q ; le produit de deux chemins composables est leur composition, le
produit de chemins non composables est nul). En particulier, nous avons des notions
naturelles de sous-représentation, de représentation simple, de somme directe et de
représentation indécomposable (= représentation non nulle qui n'est pas somme directe
de deux sous-représentations non nulles).
tation indécomposable V , cette bijection associe la racine Pn
Supposons que Q est un carquois de Dynkin de type ∆. Alors d'après le théorème de
Gabriel [45], on a une bijection de l'ensemble des classes d'isomorphisme de représenta-
tions indécomposables de Q sur l'ensemble des racines positives de ∆ ; à une représen-
i=1(dim Vi)αi, où les αi
sont les racines simples. En composant cette bijection avec celle de la partie c) du
théorème 2.1 de Fomin-Zelevinsky, nous obtenons une bijection de l'ensemble des classes
d'isomorphisme de représentations indécomposables sur l'ensemble des variables d'amas
non initiales de l'algèbre amassée AQ : à une représentation indécomposable V , cette
bijection associe l'unique variable d'amas non initiale XV dont le dénominateur est
xd1
n , où di = dim Vi. Il est remarquable que le numérateur de XV admette aussi
1 · · · xdn
une interprétation naturelle en termes de la représentation V . Pour expliciter cette
interprétation, nous avons besoin de quelques notations supplémentaires : soient V une
représentation quelconque de Q et di = dim Vi, i ∈ Q0. Pour un élément e ∈ Nn,
notons Gre(V ) l'ensemble des sous-representations U de V telles que dim Ui = ei. La
donnée d'un point de Gre(V ) est donc la donnée d'une famille de sous-espaces vecto-
riels Ui ⊂ Vi telle que Ui soit de dimension ei et que Vα(Ui) ⊂ Uj pour toute flèche
α : i → j de Q. Cette description montre que Gre(V ) est une sous-variété fermée du
produit des grassmanniennes Grei(Vi). En particulier, c'est une variété projective (sin-
gulière en général). Elle est appelée grassmannienne des sous-représentations (quiver
Grassmannian) et étudiée dans [16], par exemple. On note χ(Gre(Vi)) la caractéristique
d'Euler-Poincaré de son espace topologique sous-jacent. Posons
CC(V ) =
1
xd1
1 · · · xdn
n Xe
χ(Gre(V ))
nYi=1
Pj→i ej+Pi→j (dj −ej)
x
i
,
1014 -- 14
où les sommes dans l'exposant portent sur les flèches de but i respectivement de source i.
Théorème 4.1 (Caldero-Chapoton [12]). -- Si V est indécomposable, nous avons
XV = CC(V ).
Notons que la formule pour CC(V ) a un sens pour toute représentation de tout
carquois fini Q. Supposons maintenant que Q est un carquois sans cycles orientés
quelconque. Une représentation V de Q est rigide si son groupe d'auto-extensions
Ext1(V, V ) dans la catégorie rep(Q) s'annule. La partie c) du théorème 2.1 ne s'applique
plus, mais nous avons néanmoins une paramétrisation des variables d'amas non initiales
en termes de représentations de Q.
Théorème 4.2 ([14]). -- L'application V 7→ CC(V ) induit une bijection de l'ensemble
des classes d'isomorphisme de représentations rigides indécomposables de Q sur
l'ensemble des variables d'amas non initiales de AQ.
Ce théorème fournit une interprétation catégorique de la quasi-totalité des variables
d'amas de l'algèbre amassée. Il se pose la question d'étendre cette interprétation aux
variables initiales, aux relations d'échange et aux amas. Pour ce faire, on agrandit
la catégorie des représentations : soit DQ la catégorie dérivée bornée de la catégorie
abélienne rep(Q). Les objets de DQ sont donc les complexes bornés de représentations
et ses morphismes sont obtenus à partir des morphismes de complexes en inversant
formellement les quasi-isomorphismes. La catégorie DQ est une catégorie triangulée ;
on note Σ son foncteur suspension (qui n'est autre que le foncteur de décalage des
complexes X 7→ X[1]). Les ensembles de morphismes de DQ sont des espaces vectoriels
de dimension finie et DQ admet un foncteur de Serre, c'est-à-dire une auto-équivalence
S : DQ → DQ telle qu'on ait des isomorphismes bifonctoriels
D Hom(X, Y ) = Hom(Y, SX) ,
où D = HomC(?, C) est la dualité des espaces vectoriels complexes. La catégorie amassée
est la catégorie d'orbites
CQ = DQ/(S −1 ◦ Σ2)Z
de DQ sous l'action du groupe cyclique engendré par l'automorphisme S −1 ◦ Σ2. Elle est
due à Buan-Marsh-Reineke-Reiten-Todorov [4] et, de façon indépendante et sous une
forme très différente, à Caldero-Chapoton-Schiffler [13] pour les carquois de Dynkin de
type A. La catégorie CQ est canoniquement triangulée [72]. Ses espaces de morphismes
sont de dimension finie et son foncteur de Serre (induit par S) est isomorphe au carré
de son foncteur suspension (induit par Σ). Cela signifie que CQ est Calabi-Yau de
dimension 2. Notons π : DQ → CQ le foncteur de projection canonique et
Ext1(L, M) = Hom(L, ΣM)
pour des objets L et M de CQ. Grâce à la propriété de Calabi-Yau, on a D Ext1(L, M) =
Ext1(M, L). On appelle rigide un objet L tel que Ext1(L, L) s'annule. Notons Pi la
représentation qui correspond au CQ-module CQei, i ∈ Q0. On peut montrer [4] que
tout objet L de CQ se décompose de façon unique (à isomorphisme près) sous la forme
1014 -- 15
Σπ(Pi)mi ,
L = π(M) ⊕Mi∈Q0
CC(L) = CC(M) · Yi∈Q0
pour une représentation M et des multiplicités mi. On pose
xmi
i
.
Théorème 4.3 ([14]). -- a) On a CC(L ⊕ M) = CC(L) · CC(M) pour tous L et
M dans CQ,
b) si L et M sont des objets de CQ tels que Ext1(L, M) soit de dimension 1 et
L → E → M → ΣL et M → E ′ → L → ΣM
soient deux triangles non scindés, on a
(4.1.1)
CC(L) · CC(M) = CC(E) + CC(E ′).
c) L'application CC induit une bijection de l'ensemble des objets rigides de CQ sur
l'ensemble des monômes d'amas de AQ.
d) Par cette bijection, les objets rigides indécomposables correspondent aux variables
d'amas, et un ensemble d'indécomposables rigides T1, . . . , Tn correspond à un
amas si et seulement si Ext1(Ti, Tj) = 0 pour tous i, j.
Les propriétés a) et b) fournissent une interprétation des relations d'échange. La
démonstration du théorème est fondée sur le travail de plusieurs groupes d'auteurs :
Buan-Marsh-Reiten-Todorov [5], Buan-Marsh-Reiten [11], Buan-Marsh-Reineke-
Reiten-Todorov [4], Marsh-Reineke-Zelevinsky [84], . . . et surtout Caldero-Chapoton
[12]. Une autre démonstration de la formule de multiplication 4.1.1 est due à Hubery
[59] pour des carquois dont le graphe sous-jacent est un diagramme de Dynkin étendu,
et à Xiao-Xu [101] dans le cas général.
La construction de la catégorie amassée a été généralisée des algèbres CQ à une
classe d'algèbres de dimension globale 2 par Amiot [1]. Une version généralisée de
l'application de Caldero-Chapoton et de la formule de multiplication 4.1.1 est due à
Palu [89]. L'extension des résultats de Palu au cas de certaines algèbres amassées à
coefficients est obtenue dans [44]. La mutation dans une catégorie 2-Calabi-Yau générale
est construite par Iyama-Yoshino [65]. La démonstration du théorème de périodicité 2.4
dans [70] [73] est fondée sur ces travaux.
4.2. Catégorification additive : modules sur les algèbres préprojectives
Nous allons décrire l'idée de base de la démonstration du théorème 3.2. Soient ∆
un diagramme de Dynkin, g l'algèbre de Lie simple complexe qui lui correspond et
n une sous-algèbre nilpotente maximale de g. Soit N le groupe algébrique unipotent
associé à n. L'algèbre de coordonnées C[N] est le dual restreint de la cogèbre U(n).
La base duale semi-canonique de l'algèbre C[N] est duale de la base semi-canonique de
1014 -- 16
U(n) construite par Lusztig [83]. Dans un premier temps, nous allons décrire (suivant
[49]) la base duale semi-canonique en termes de modules sur l'algèbre préprojective :
soit Q un carquois de Dynkin de type ∆. Soit Q le double carquois, obtenu à partir
de Q en rajoutant une flèche α∗ : j → i pour chaque flèche α : i → j. Soit Λ
l'algèbre préprojective de Q, c'est-à-dire le quotient de l'algèbre des chemins CQ (voir
la section 4.1) par l'idéal bilatère engendré par la somme P[α, α∗] prise sur l'ensemble
des flèches de Q. C'est une algèbre de dimension finie sur C qui est auto-injective
(c'est-à-dire injective comme module sur elle-même). Appelons Λ-module un Λ-module
à gauche de dimension finie sur C. Pour un tel module M, son vecteur dimension est la
suite des entiers dim eiM, i ∈ Q0. Soit d une famille d'entiers positifs indexés par Q0.
On note rep(Λ, d) la variété formée des familles de matrices
Mα : Cds(α) → Cdt(α) , α ∈ Q1 ,
qui vérifient les relations de Λ, c'est-à-dire définissent une structure de Λ-module sur
la somme directe des Cdi, i ∈ Q0. La variété rep(Λ, d) porte une action naturelle par
sont en bijection avec les classes d'isomorphisme de Λ-modules de vecteur dimension d.
Notons Md l'espace vectoriel des fonctions constructibles et Gd-invariantes sur la variété
«changement de base» du groupe Gd = Q GL(di, C) et les orbites de cette action
rep(Λ, d). Notons U(n)d la composante graduée de U(n) associée au vecteur P diαi, où
les αi sont les racines simples. Lusztig [83] a défini une injection linéaire λd de U(n)d
dans Md. Chaque Λ-module M définit une forme linéaire sur Md, à savoir la forme qui,
à une fonction f, associe sa valeur f (M) en l'orbite déterminée par M. Par composition,
le module M nous donne une forme linéaire
δM : U(n)d
λd→ Md → C.
d → C[N]d. Comme dans [49], nous
Or nous avons l'isomorphisme canonique ι : U(n)∗
posons ϕM = ι(δM ). Nous obtenons ainsi une application M 7→ ϕM de la classe
des Λ-modules dans l'algèbre C[N]. On peut expliciter cette application en termes
de caractéristiques d'Euler-Poincaré de variétés de drapeaux de sous-représentations
de M, voir [49]. Cette description montre que l'application M 7→ ϕM est constructible
sur la variété algébrique rep(Λ, d). Donc chaque composante irréductible de cette variété
contient un ouvert dense où la fonction M 7→ ϕM est constante. On appelle génériques
les modules M appartenant à de tels ouverts. Alors la base duale semi-canonique n'est
autre que
{ϕM M est générique}.
Un module M est rigide si l'espace Ext1(M, M) s'annule. De façon équivalente [49],
l'orbite de M dans rep(Λ, d), où d est le vecteur dimension de M, est ouverte. En
particulier, si M est rigide, alors il est générique et la fonction ϕM appartient à la
base duale semi-canonique. Pour montrer le théorème 3.2, il suffit donc de montrer que
chaque monôme d'amas est de la forme ϕM pour un module rigide M. Pour cela, on
procède par récurrence : on montre [50] que les éléments de la graine initiale sont des
images de modules rigides indécomposables canoniques T (0)
m . Puis on
, . . . , T (0)
1
, T (0)
2
1014 -- 17
relève l'opération de mutation des amas à une classe convenable de suites T1, . . . , Tm
de modules rigides indécomposables. Appelons accessibles les modules rigides dont les
facteurs directs indécomposables sont obtenus par mutation itérée à partir de la suite
T (0)
m . Le théorème suivant est l'analogue précis du théorème 4.3. Ses
1
parties a) et b) permettent de relier la mutation des amas à la mutation des modules
rigides et donc d'effectuer la récurrence qui termine la démonstration du théorème 3.2.
, . . . , T (0)
, T (0)
2
Théorème 4.4 (Geiss-Leclerc-Schröer [49]). -- On a
a) ϕL⊕M = ϕLϕM ,
b) Si Ext1(L, M) est de dimension 1 et que l'on a les suites exactes non scindées
0 → L → E → M → 0 et 0 → M → E ′ → L → 0 ,
alors on a ϕLϕM = ϕE + ϕE ′.
c) L'application M 7→ ϕM induit une bijection de l'ensemble des classes d'isomorphisme
de modules rigides accessibles sur l'ensemble des monômes d'amas.
d) Les rigides indécomposables accessibles correspondent aux variables d'amas et une
suite T1, . . . , Tm de tels modules correspond à un amas si et seulement si l'on a
Ext1(Ti, Tj) = 0 pour tous i et j.
4.3. Catégorification monoïdale
Les catégorifications additives décrites ci-dessus se sont avérées très utiles et on sait
les construire pour de grandes classes d'algèbres amassées. De l'autre côté, elles sem-
blent difficiles à exploiter pour démontrer la conjecture de positivité 2.3, et la notion
même de catégorification additive semble peu naturelle. La catégorification monoïdale,
introduite par Leclerc [80] et Hernandez-Leclerc [58], consiste à réaliser une algèbre
amassée comme l'anneau de Grothendieck d'une catégorie abélienne monoïdale. Elle
est donc très naturelle. En outre, comme nous allons le voir, son existence donne im-
médiatement la conjecture de positivité 2.3 (et la conjecture d'indépendance 2.2). De
l'autre côté, les catégorifications monoïdales semblent très difficiles à construire.
Les notions suivantes [80] sont fondamentales pour la suite : un objet simple S d'une
catégorie abélienne monoïdale est premier s'il n'admet pas de factorisation tensorielle
non triviale ; il est réel si son carré tensoriel est encore simple.
Soient A une algèbre amassée à coefficients et AZ son sous-anneau engendré par les
variables d'amas et les coefficients. Suivant [58], une catégorification monoïdale de A
est la donnée d'une catégorie abélienne monoïdale M et d'un isomorphisme d'anneaux
tel que ϕ induise
ϕ : AZ
∼→ K0(M)
a) une bijection de l'ensemble des monômes d'amas sur l'ensemble des classes d'objets
simples réels de M et
b) une bijection de l'ensemble des variables d'amas et des coefficients sur l'ensemble
des classes d'objets simples, réels et premiers de M.
1014 -- 18
Le tableau suivant, extrait de [80], résume les correspondances entre les structures
associées à une algèbre amassée et leurs relèvements dans une catégorification additive
respectivement monoïdale.
algèbre amassée A catégorification additive C catégorification monoïdale M
+
×
?
⊕
⊕
⊗
monôme d'amas
variable d'amas
objet rigide
indécomposable rigide
objet simple réel
simple premier réel
L'existence d'une catégorification monoïdale M d'une algèbre amassée A a des con-
séquences très fortes pour A : en effet,
l'algèbre A est alors munie d'une «base
canonique», à savoir la base fournie par les objets simples de M et cette base con-
tient les monômes d'amas car ceux-ci correspondent bijectivement aux classes dans
K0(M) de certains objets simples. En particulier, la conjecture d'indépendance est
vérifiée pour A. De même, la conjecture de positivité est vérifiée pour A. En effet, si
on exprime une variable d'amas x comme polynôme de Laurent
x =
P (u1, . . . , um)
ud1
1 · · · udm
m
en les variables d'un amas u1, . . . , um, alors les coefficients de P sont les multiplicités de
certains objets simples dans la classe du produit tensoriel ϕ(xud1
m ) et sont donc
des entiers positifs.
1 · · · udm
L'existence d'une catégorification monoïdale ϕ : AZ → K0(M) donne également des
renseignements précieux sur la structure monoïdale de M : en effet, elle montre que
le comportement des objets simples réels de M est gouverné par la combinatoire des
amas de A.
Dans [58], Hernandez-Leclerc exhibent des catégorifications monoïdales conjecturales
un diagramme de Dynkin simplement lacé et l ∈ N un «niveau». Voici l'exemple du
Ml pour les algèbres amassées Al associées à certains carquois glacés eQ(∆, l), où ∆ est
carquois eQ(D5, 3), où les sommets gelés sont marqués par des •.
/ ◦
/ •
◦
J
◦
◦
Y2222
/ ◦
◦
2222
◦
Y2222
/ ◦
◦
J
/ •
•
/ •
•
◦
◦
/ ◦
◦
◦
/ ◦
Hernandez-Leclerc construisent les catégories Ml comme des sous-catégories monoïdales
de la catégorie des représentations de dimension finie de l'algèbre affine quantique Uq(bg)
associée à ∆. Ils construisent un morphisme d'anneaux ϕ : (Al)Z → K0(Ml) qui envoie
/
o
o
/
/
o
o
/
Y
J
o
o
/
J
Y
o
o
/
O
O
o
o
/
O
O
o
o
/
O
O
o
o
1014 -- 19
les variables de l'amas initial sur les classes de certains modules de Kirillov-Reshetikhin
et conjecturent que ϕ est une catégorification monoïdale de Al (Conjecture 13.2 de
[58]). Ils démontrent leur conjecture pour l ≤ 1 et ∆ de type An, n ≥ 1, ou D4 ainsi
que pour ∆ de type A2 et l = 2 (et observent que pour ∆ = A1 et tout l ∈ N, la
conjecture résulte du travail de Chari-Pressley [20]).
Dans [87], Nakajima construit des catégorifications monoïdales conjecturales Nl pour
les carquois eQ(R, l) associés à un carquois R bipartite et un niveau l ∈ N. Les catégories
Nl sont réalisées comme des catégories de faisceaux pervers sur des variétés de carquois
gradués [88] munies du produit tensoriel construit géométriquement dans [99]. Si R
est un carquois de Dynkin et l = 1, la catégorie Nl est équivalente à Ml et Nakajima
démontre qu'elle est une catégorification monoïdale de l'algèbre Al confirmant ainsi la
conjecture de Hernandez-Leclerc. Pour un carquois bipartite R quelconque et l = 1,
il montre que l'anneau (Al)Z se plonge dans K0(Nl) de telle façon que les monômes
d'amas sont envoyés sur des objets simples. Ceci entraîne la conjecture de positivité
pour Al.
4.4. Catégorification via les carquois à potentiel
Inspirés par des travaux de physiciens (voir par exemple la section 6 dans [29])
Derksen-Weyman-Zelevinsky ont étendu [23] l'opération de mutation des carquois aux
carquois à potentiel et leurs représentations décorées. Décrivons brièvement ces notions
plétée, c'est-à-dire la complétion de CQ par rapport à l'idéal bilatère engendré par les
flèches de Q. L'espace CQ admet donc une base topologique formée de tous les chemins
en suivant [23]. Soit en effet Q un carquois fini. Notons dCQ l'algèbre des chemins com-
de Q. L'homologie de Hochschild continue HH0(dCQ) est le complété de l'espace quo-
tient de CQ par le sous-espace [CQ, CQ] engendré par tous les commutateurs. Il admet
une base topologique formée de tous les cycles de Q, c'est-à-dire les orbites sous l'action
du groupe cyclique Z/tZ de chemins cycliques de longueur t ≥ 0. Pour chaque flèche α
de Q, la dérivée cyclique [94] est l'unique application linéaire continue
∂α : HH0(dCQ) →dCQ
p = uαv en des chemins u et v de longueur supérieure ou égale à zéro. Soit W un
qui envoie la classe d'un chemin p sur la sommeP vu prise sur toutes les décompositions
potentiel sur Q, c'est-à-dire un élément de HH0(dCQ). L'algèbre de Jacobi P(Q, W ) est
la complétion du quotient de dCQ par l'idéal bilatère engendré par les dérivées cycliques
∂αW , où α parcourt les flèches de Q. Une représentation décorée (M, V ) de (Q, W ) est
formée d'un module M sur P(Q, W ) et d'une famille d'espaces vectoriels Vi, i ∈ Q0,
où M et les Vi sont supposés de dimension finie sur C. Par exemple, si Q n'a pas de
cycles orientés (et donc W = 0 et P(Q, W ) = CQ), toute représentation décorée (M, V )
fournit un objet
π(M) ⊕Mi∈Q0
Σπ(Vi ⊗C Pi)
de la catégorie amassée (voir la section 4.1).
1014 -- 20
Dans [23] et [22], Derksen-Weyman-Zelevinsky construisent et étudient l'opération
de mutation pour les carquois à potentiel et leurs représentations décorées. Des diffi-
cultés techniques nombreuses et subtiles sont dues au fait que cette opération n'est ni
fonctorielle ni définie partout. Ceci est aussi la raison pour laquelle les représentations
décorées ne forment pas, en général, une catégorie. Néanmoins, la théorie développée
par Derksen-Weyman-Zelevinsky est assez proche de la catégorification additive. Elle en
diffère par le fait que les objets combinatoires centraux ne sont plus les variables d'amas
mais les F -polynômes et g-vecteurs introduits dans [42] et qui sont peut-être encore plus
fondamentaux que les variables d'amas. Dans [22], Derksen-Weyman-Zelevinsky ap-
pliquent leur théorie en démontrant de nombreuses conjectures formulées dans [42]. Ils
y parviennent sous la seule hypothèse que les algèbres amassées considérées proviennent
de bons carquois (non glacés), c'est-à-dire de matrices antisymétriques quelconques, ce
qui représente un progrès remarquable par rapport aux approches précédentes. La
construction de bases et la conjecture de positivité restent néanmoins des problèmes
complètement ouverts dans cette généralité. Les idées de Derksen-Weyman-Zelevinsky
ont été liées à la catégorification additive au sens des sections 4.2 et 4.1 dans [2] [74] [1].
Un lien important avec les surfaces à bord marquées est établi dans [79].
Remerciements
Je remercie Caroline Gruson, Bernard Leclerc et Rached Mneimné pour leurs conseils
avisés sur une version antérieure de ce texte.
RÉFÉRENCES
[1] Claire Amiot -- Cluster categories for algebras of global dimension 2 and quivers
with potential, arXiv:0805.1035, à paraître dans Ann. Inst. Fourier.
[2] Aslak Bakke Buan, Osamu Iyama, Idun Reiten & David Smith -- Mutation
of cluster-tilting objects and potentials, arXiv:0804.3813.
[3] Aslak Bakke Buan & Robert Marsh -- Cluster-tilting theory, Trends in rep-
resentation theory of algebras and related topics, Contemp. Math., vol. 406, Amer.
Math. Soc., Providence, RI, 2006, pp. 1 -- 30.
[4] Aslak Bakke Buan, Robert J. Marsh, Markus Reineke, Idun Reiten
& Gordana Todorov -- Tilting theory and cluster combinatorics, Adv. Math. 204
(2006), 572 -- 618.
1014 -- 21
, Quantum cluster algebras, Adv. Math. 195 (2005), 405 -- 455.
[5] Aslak Bakke Buan, Robert J. Marsh, Idun Reiten & Gordana Todorov
-- Clusters and seeds in acyclic cluster algebras, Proc. Amer. Math. Soc. 135
(2007),3049 -- 3060 (electronic), With an appendix coauthored in addition by P. Caldero
and B. Keller.
[6] Arkady Berenstein, Sergey Fomin & Andrei Zelevinsky -- Parametriza-
tions of canonical bases and totally positive matrices, Adv. Math. 122 (1996), 49 -- 149.
, Cluster algebras. III. Upper bounds and double Bruhat cells, Duke Math. J.
[7]
126 (2005), 1 -- 52.
[8] Arkady Berenstein & Andrei Zelevinsky -- Tensor product multiplicities,
canonical bases and totally positive varieties, Invent. Math. 143 (2001), 77 -- 128.
[9]
[10] Tom Bridgeland -- Stability conditions on triangulated categories, Ann. of Math.
(2) 166 (2007), 317 -- 345.
[11] Aslak Bakke Buan, Robert J. Marsh & Idun Reiten -- Cluster mutation
via quiver representations, Comment. Math. Helv. 83 (2008), 143 -- 177.
[12] Philippe Caldero & Frédéric Chapoton -- Cluster algebras as Hall algebras
of quiver representations, Comment. Math. Helv. 81 (2006), 595 -- 616.
[13] Philippe Caldero, Frédéric Chapoton & Ralf Schiffler -- Quivers with
relations arising from clusters (An case), Trans. Amer. Math. Soc. 358 (2006), 1347 --
1364.
[14] Philippe Caldero & Bernhard Keller -- From triangulated categories to
cluster algebras. II, Ann. Sci. École Norm. Sup. (4) 39 (2006), 983 -- 1009.
[15] Philippe Caldero & Bernhard Keller -- From triangulated categories to
cluster algebras, Inv. Math. 172 (2008), 169 -- 211.
[16] Philippe Caldero and Markus Reineke -- On the quiver Grassmannian in
the acyclic case, J. Pure Appl. Algebra 212 (2008), 2369 -- 2380.
[17] Giovanni Cerulli Irelli -- Canonically positive basis of cluster algebras of type
A(1)
[18] Frédéric Chapoton -- Enumerative properties of generalized associahedra, Sém.
Lothar. Combin. 51 (2004/05), Art. B51b, 16 pp. (electronic).
[19] Frédéric Chapoton, Sergey Fomin & Andrei Zelevinsky -- Polytopal real-
izations of generalized associahedra, Canad. Math. Bull. 45 (2002), 537 -- 566, Dedicated
to Robert V. Moody.
[20] Vyjayanthi Chari & Andrew Pressley -- Quantum affine algebras, Comm.
Math. Phys. 142 (1991), 261 -- 283.
[21] Laurent Demonet -- Categorification of skew-symmetrizable cluster algebras,
arXiv:0909.1633.
[22] Harm Derksen, Jerzy Weyman & Andrei Zelevinsky -- Quivers with poten-
tials and their representations II: Applications to cluster algebras, arXiv:0904.0676v1.
, Quivers with potentials and their representations I: Mutations, Selecta
[23]
Mathematica 14 (2008), 59 -- 119.
2 , arXiv:0904.2543.
1014 -- 22
, Positivity of the T -system cluster algebra, arXiv:0908.3122.
variables
in acyclic
cluster
algebras,
, Cluster algebras: notes for the CDM-03 conference, Current developments
, Cluster ensembles, quantization and the dilogarithm II: The intertwiner,
[24] Philippe Di Francesco & Rinat Kedem -- Q-systems as cluster algebras II:
Cartan matrix of finite type and the polynomial property, arXiv:0803.0362.
[25]
[26] Grégoire Dupont -- Generic
arXiv:0811.2909v1.
[27] M. Fekete -- Ueber ein Problem von Laguerre, Rend. Circ. Mat. Palermo 34
(1912), 89 -- 100, 110 -- 120.
[28] Anna Felikson, Michael Shapiro & Pavel Tumarkin -- Skew-symmetric
cluster algebras of finite mutation type, arXiv:0811.1703.
[29] Bo Feng, Amihay Hanany, Yang-Hui He & Angel M. Uranga -- Toric
duality as Seiberg duality and brane diamonds, J. High Energy Phys. (2001), Paper 35,
29.
[30] V. V. Fock & A. B. Goncharov -- Cluster ensembles, quantization and the
dilogarithm, arXiv:math.AG/0311245. À paraître dans le fasc. 6 des Ann. Scient. Éc.
Norm. Sup. 42 (2009).
[31]
arXiv:math/0702398.
, Cluster X -varieties, amalgamation & Poisson-Lie groups, Algebraic ge-
[32]
ometry and number theory, Progr. Math., vol. 253, Birkhäuser Boston, Boston, MA,
2006, pp. 27 -- 68.
[33]
Invent. Math. 175 (2009), 223 -- 286.
[34]
Inst. Hautes Études Sci. (2006), 1 -- 211.
[35] Sergey Fomin -- Cluster algebras portal, www.math.lsa.umich.edu/~fomin/
cluster.html.
[36] Sergey Fomin & Nathan Reading -- Generalized cluster complexes and
Coxeter combinatorics, Int. Math. Res. Not. (2005), 2709 -- 2757.
[37] Sergey Fomin, Michael Shapiro & Dylan Thurston -- Cluster algebras
and triangulated surfaces. I. Cluster complexes, Acta Math. 201 (2008), 83 -- 146.
[38] Sergey Fomin & Andrei Zelevinsky -- Cluster algebras. I. Foundations,
J. Amer. Math. Soc. 15 (2002), 497 -- 529 (electronic).
[39]
63 -- 121.
[40]
in mathematics, 2003, Int. Press, Somerville, MA, 2003, pp. 1 -- 34.
[41]
977 -- 1018.
[42]
112 -- 164.
, Cluster algebras. II. Finite type classification, Invent. Math. 154 (2003),
, Y -systems and generalized associahedra, Ann. of Math. (2) 158 (2003),
, Cluster algebras IV: Coefficients, Compositio Mathematica 143 (2007),
, Moduli spaces of local systems and higher Teichmüller theory, Publ. Math.
, The quantum dilogarithm and representations of quantum cluster varieties,
1014 -- 23
, Semicanonical bases and preprojective algebras, Ann. Sci. École Norm.
, Rigid modules over preprojective algebras, Invent. Math. 165 (2006), 589 --
, On the properties of the exchange graph of a cluster algebra, Math. Res.
, Partial flag varieties and preprojective algebras, Ann. Inst. Fourier (Greno-
[43] Edward Frenkel & András Szenes -- Thermodynamic Bethe ansatz and dilog-
arithm identities. I, Math. Res. Lett. 2 (1995), 677 -- 693.
[44] Changjian Fu & Bernhard Keller -- On cluster algebras with coefficients and
2-Calabi-Yau categories, arXiv:0710.3152.
[45] P. Gabriel -- Représentations indécomposables, Sém. Bourbaki (1973/74),
Exp. n◦ 444, Lecture Notes in Math. 431 (1975), 143 -- 169.
[46] Davide Gaiotto, Gregory W. Moore & Andrew Neitzke -- Wall-crossing,
Hitchin systems & the WKB approximation, arXiv:0907.3987.
[47] Christof Geiss, Bernard Leclerc & Jan Schröer -- Cluster algebra struc-
tures and semicanonical bases for unipotent groups, arXiv:math/0703039.
[48]
Sup. 38 (2005), 193 -- 253.
[49]
632.
[50]
ble) 58 (2008), 825 -- 876.
, Preprojective algebras and cluster algebras, Trends in representation theory
[51]
of algebras and related topics, EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich, 2008,
pp. 253 -- 283.
[52] Michael Gekhtman, Michael Shapiro & Alek Vainshtein -- Cluster al-
gebras and Poisson geometry, Mosc. Math. J. 3 (2003), 899 -- 934, 1199 (Dedicated to
Vladimir Igorevich Arnold on the occasion of his 65th birthday).
[53]
291 -- 311.
[54]
Lett. 15 (2008), 321 -- 330.
[55] Victor Ginzburg -- Calabi-Yau algebras, arXiv:math/0612139v3.
[56] F. Gliozzi & R. Tateo -- Thermodynamic Bethe ansatz and three-fold triangu-
lations, Internat. J. Modern Phys. A 11 (1996), 4051 -- 4064.
[57] André Henriques -- A periodicity theorem for the octahedron recurrence, J. Al-
gebraic Combin. 26 (2007), 1 -- 26.
[58] David Hernandez & Bernard Leclerc -- Cluster algebras and quantum affine
algebras, arXiv:0903.1452v1.
[59] Andrew Hubery -- Acyclic cluster algebras via Ringel-Hall algebras, Preprint
available at the author's home page.
[60] Colin Ingalls & Hugh Thomas -- Noncrossing partitions and representations
of quivers, arXiv:math/0612219.
[61] Rei Inoue, Osamu Iyama, Atsuo Kuniba, Tomoki Nakanishi & Junji
Suzuki -- Periodicities of T -systems and Y -systems, arXiv:0812.0667.
, Cluster algebras and Weil-Petersson forms, Duke Math. J. 127 (2005),
1014 -- 24
, Quiver mutation in Java, applet Java disponible sur la page personnelle
, Periodicities of T and Y -systems, dilogarithm identities, and cluster alge-
[62] Rei Inoue, Osamu Iyama, Bernhard Keller, Atsuo Kuniba & Tomoki
Nakanishi -- Periodicities of T and Y -systems, dilogarithm identities, and cluster
algebras I: type Br, arXiv:1001.1880.
[63]
bras II: types Cr, F4, and G2, arXiv:1001.1881.
[64] Osamu Iyama & Idun Reiten -- Fomin-Zelevinsky mutation and tilting modules
over Calabi-Yau algebras, Amer. J. Math. 130 (2008), 1087 -- 1149.
[65] Osamu Iyama & Yuji Yoshino -- Mutations in triangulated categories and rigid
Cohen-Macaulay modules, Inv. Math. 172 (2008), 117 -- 168.
[66] Dominic Joyce & Yinan Song -- A theory of generalized Donaldson-Thomas
invariants. II. Multiplicative identities for Behrend functions, arXiv:0901.2872.
[67] Victor G. Kac -- Infinite-dimensional Lie algebras, third ed., Cambridge Univ.
Press, Cambridge, 1990.
[68] Masaki Kashiwara -- Bases cristallines, C. R. Acad. Sci. Paris Sér. I Math. 311
(1990), 277 -- 280.
[69] Rinat Kedem -- Q-systems as cluster algebras, J. Phys. A 41 (2008), 194011, 14.
[70] Bernhard Keller -- Cluster algebras, quiver representations and triangulated
categories, arXiv:0807.1960.
[71]
de l'auteur.
, On triangulated orbit categories, Doc. Math. 10 (2005), 551 -- 581.
[72]
[73]
, The periodicity conjecture for pairs of Dynkin diagrams, arXiv:1001.1531.
[74] Bernhard Keller & Dong Yang -- Derived equivalences from mutations of
quivers with potential, arXiv:0906.0761.
[75] Maxim Kontsevich & Yan Soibelman -- Stability structures, Donaldson-
Thomas invariants and cluster transformations, arXiv:0811.2435.
[76] Christian Krattenthaler -- The F -triangle of the generalised cluster complex,
Topics in discrete mathematics, Algorithms Combin., vol. 26, Springer, Berlin, 2006,
pp. 93 -- 126.
[77] A. Kuniba & T. Nakanishi -- Spectra in conformal field theories from the Rogers
dilogarithm, Modern Phys. Lett. A 7 (1992), 3487 -- 3494.
[78] Atsuo Kuniba, Tomoki Nakanishi & Junji Suzuki -- Functional relations in
solvable lattice models. I. Functional relations and representation theory, Internat. J.
Modern Phys. A 9 (1994), 5215 -- 5266.
[79] Daniel Labardini-Fragoso -- Quivers with potentials associated to triangulated
surfaces, Proc. Lond. Math. Soc. (3) 98 (2009), 797 -- 839.
[80] Bernard Leclerc -- Algèbres affines quantiques et algèbres amassées, Notes d'un
exposé au séminaire d'algèbre à l'institut Henri Poincaré le 14 janvier 2008.
[81] G. Lusztig -- Canonical bases arising from quantized enveloping algebras, J. Amer.
Math. Soc. 3 (1990), 447 -- 498.
1014 -- 25
, Total positivity in reductive groups, Lie theory and geometry, Progr. Math.,
, Semicanonical bases arising from enveloping algebras, Adv. Math. 151
[82]
vol. 123, Birkhäuser Boston, Boston, MA, 1994, pp. 531 -- 568.
[83]
(2000), 129 -- 139.
[84] Robert Marsh, Markus Reineke & Andrei Zelevinsky -- Generalized as-
sociahedra via quiver representations, Trans. Amer. Math. Soc. 355 (2003), 4171 -- 4186
(electronic).
[85] Gregg Musiker -- A graph theoretic expansion formula for cluster algebras of
type Bn and Dn, arXiv:0710.3574v1.
[86] Gregg Musiker, Ralf Schiffler & Lauren Williams -- Positivity for clus-
ter algebras from surfaces, arXiv:0906.0748.
[87] Hiraku Nakajima -- Quiver varieties and cluster algebras, arXiv:0905.0002v3.
[88]
algebras, J. Amer. Math. Soc. 14 (2001), 145 -- 238 (electronic).
[89] Yann Palu -- Cluster characters for 2-Calabi-Yau triangulated categories, Ann.
Inst. Fourier (Grenoble) 58 (2008), 2221 -- 2248.
[90] F. Ravanini, A. Valleriani & R. Tateo -- Dynkin TBAs, Internat. J. Modern
Phys. A 8 (1993), 1707 -- 1727.
[91] Markus Reineke -- Cohomology of quiver moduli, functional equations & inte-
grality of Donaldson-Thomas type invariants, arXiv:0903.0261.
[92] Idun Reiten -- Tilting theory and cluster algebras, prépublication disponible sur
, Quiver varieties and finite-dimensional representations of quantum affine
www.institut.math.jussieu.fr/e keller/ictp2006/lecturenotes/reiten.pdf.
[93] Claus Michael Ringel -- Some remarks concerning tilting modules and tilted
algebras. Origin. Relevance. Future., Handbook of Tilting Theory, LMS Lecture Note
Series, vol. 332, Cambridge Univ. Press, Cambridge, 2007, pp. 49 -- 104.
[94] Gian-Carlo Rota, Bruce Sagan & Paul R. Stein -- A cyclic derivative in
noncommutative algebra, J. Algebra 64 (1980), 54 -- 75.
[95] Joshua S. Scott -- Grassmannians and cluster algebras, Proc. London Math.
Soc. (3) 92 (2006), 345 -- 380.
[96] Paul Sherman and Andrei Zelevinsky -- Positivity and canonical bases in
rank 2 cluster algebras of finite and affine types, Mosc. Math. J. 4 (2004), no. 4,
947 -- 974, 982.
[97] James Dillon Stasheff -- Homotopy associativity of H-spaces. I, II, Trans.
Amer. Math. Soc. 108 (1963), 275-292; ibid. 108 (1963), 293 -- 312.
[98] András
arXiv:math.RT/0606377.
[99] M. Varagnolo & E. Vasserot -- Perverse sheaves and quantum Grothendieck
rings, Studies in memory of Issai Schur (Chevaleret/Rehovot, 2000), Progr. Math.,
vol. 210, Birkhäuser Boston, Boston, MA, 2003, pp. 345 -- 365.
-- Periodicity
connections,
Szenes
flat
of
Y -systems
and
1014 -- 26
[100] Alexandre Yu. Volkov -- On the periodicity conjecture for Y -systems, Comm.
Math. Phys. 276 (2007), 509 -- 517.
[101] Jie Xiao & Fan Xu -- Green's formula with C∗-action and Caldero-Keller's
formula for cluster algebras, arXiv:0707.1175.
[102] Shih-Wei Yang & Andrei Zelevinsky -- Cluster algebras of finite type via
Coxeter elements and principal minors, arXiv:0804.3303.
[103] Al. B. Zamolodchikov -- On the thermodynamic Bethe ansatz equations for
reflectionless ADE scattering theories, Phys. Lett. B 253 (1991), 391 -- 394.
[104] Andrei Zelevinsky -- Cluster algebras: notes for 2004 IMCC (Chonju, Korea,
August 2004), arXiv:math.RT/0407414.
, From Littlewood-Richardson coefficients to cluster algebras in three lec-
[105]
tures, Symmetric functions 2001: surveys of developments and perspectives, NATO
Sci. Ser. II Math. Phys. Chem., vol. 74, Kluwer Acad. Publ., Dordrecht, 2002, pp. 253 --
273.
[106]
towards millennium problems, SAS Int. Publ., Delhi, 2005, pp. 85 -- 105.
[107]
, What is a cluster algebra?, Notices of the A.M.S. 54 (2007), 1494 -- 1495.
, Cluster algebras: origins, results and conjectures, Advances in algebra
Bernhard KELLER
Université Paris Diderot -- Paris 7
Institut de Mathématiques de Jussieu
U.M.R. 7586 du CNRS
U.F.R. de Mathématiques
Case 7012
Bâtiment Chevaleret
F -- 75205 Paris Cedex 13
E-mail : [email protected]
|
1901.01655 | 1 | 1901 | 2019-01-07T03:29:12 | Orthogonal decompositions of classical Lie algebras over finite commutative rings | [
"math.RA"
] | Let $R$ be a finite commutative ring with identity. In this paper, we give a necessary condition for the existence of an orthogonal decomposition of the special linear Lie algebra over $R$. Additionally, we study orthogonal decompositions of the symplectic Lie algebra and the special orthogonal Lie algebra over $R$. | math.RA | math |
ORTHOGONAL DECOMPOSITIONS OF CLASSICAL LIE ALGEBRAS
OVER FINITE COMMUTATIVE RINGS
SONGPON SRIWONGSA
Abstract. Let R be a finite commutative ring with identity. In this paper, we give a
necessary condition for the existence of an orthogonal decomposition of the special linear
Lie algebra over R. Additionally, we study orthogonal decompositions of the symplectic
Lie algebra and the special orthogonal Lie algebra over R.
1. Introduction
An orthogonal decomposition (OD) of a finite dimensional Lie algebra L over the field
of complex numbers C is a decomposition of L into a direct sum of its Cartan subalgebras
which are pairwise orthogonal with respect to the Killing form. The earliest recorded
mention for orthogonal decompositions of Lie algebras was by Thompson who used an
OD of the Lie algebra E8 for the construction of a finite simple group of a special order,
also known as the Thompson group [13, 14]. In the 1980s, Kostrikin et al. developed the
theory of orthogonal decompositions of simple Lie algebras of types A, B, C and D over
C [7, 8]. During the past four decades, the OD problem of Lie algebras has attracted
greater attentions due to its applications in other fields. For instance, an OD of the special
linear Lie algebra sln(C) is related to mutually unbiased bases (MUBs) in Cn which play an
important role in quantum information theory [2, 11]. A connection between the problem
of constructing maximal collections of MUBs and the existence problem of an OD of sln(C)
was found by Boykin et al. [2].
An OD of sln(C) has been constructed for all n which are a power of a prime integer
[7]. In the latter cases, the existence of an OD is still an open question even when n = 6
(the first positive integer that is not a power of a prime). We refer the reader to [1] for
a recent development of the OD problem of sl6(C). Note that the symplectic Lie algebra
sp6(C) is a subalgebra of sl6(C). The OD problem of sp6(C) was studied in [15].
It is
natural to ask when an orthogonal decomposition exists for Lie algebras over other fields
or, more generally, over other rings. Recently, the problem in this direction for the case of
Lie algebra sln over a finite commutative ring with identity was studied in [12].
2010 Mathematics Subject Classification. Primary: 17B50; Secondary: 13M05.
Key words and phrases. Cartan subalgebras; Local rings; Orthogonal decomposition.
1
2
SONGPON SRIWONGSA
Throughout this paper we assume that all modules are unitary and L denotes a Lie
algebra over a finite commutative ring R with identity that is free of rank n as an R-module.
Recall that a Cartan subalgebra of L is a nilpotent subalgebra which equals its normalizer
in L. Every Cartan subalgebra of a finite dimensional semisimple Lie algebra over C is
abelian. However, for a Lie algebra over a general finite commutative ring with identity,
a Cartan subalgebra is not necessary abelian. For example, there is a finite dimensional
semisimple Lie algebra over a finite field which has a non-abelian Cartan subalgebra [4].
Here, we only consider an orthogonal decomposition of L that is formed by abelian Cartan
subalgebras and use the abbreviation ODAC (AC for "abelian Cartan"). The orthogonality
is defined via the Killing form
K(A, B) := T r(ad A · ad B).
This form is well-defined because all considered Lie algebras are free modules of finite rank.
Therefore, an ODAC of L is a decomposition
L = H0 ⊕ H1 ⊕ . . . ⊕ Hk, k ∈ N0,
where Hi's are pairwise orthogonal abelian Cartan subalgebras of L with respect to the
Killing form.
It was proved in [12] that an ODAC of sln(R) can be constructed under some sufficient
conditions on the ring R and n. In this paper, we continue to study the existence problem
of an ODAC of sln(R) by considering a necessary condition on R and n. One of our
motivations is to illuminate the result for the orthogonal decomposition problem of sln
when n is a non prime power integer especially n = 6. Moreover, we consider the problem
of ODAC of the symplectic Lie algebra spn and the special orthogonal Lie algebra son
over R with odd characteristic by using the techniques motivated by the previous works of
Kostrikin et al. on these Lie algebras over C [8, 9].
In Section 2, we first relate the problem of ODAC of L with the ring decomposition of R.
We use the result to show that if sln(R) has an ODAC, then char(R) is relatively prime to
n. In Section 3, we construct an ODAC of sp2m+1(R) when char(R) is odd, by restricting
the ODAC of sl2m+1(R) constructed in [12]. The restriction was used for sp2m+1(C) (see [9,
Lemma 2.1.4]) and one can use Lie's theorem to verify that the restricted decomposition
is an OD of sp2m+1(C). However, Lie's theorem does not exist in a general commutative
ring case. Thus, some arguments need to be modified in our setting here. In the complex
number case, an OD of son(C) was constructed by using its standard basis elements [8]. For
n = 2k, the authors verified this OD by relating the construction to 1-factorization of the
complete graph with 2k vertices and used the result to form an OD of this Lie algebra when
n = 2k − 1. One can realize that the similar technique also works for any commutative ring
ORTHOGONAL DECOMPOSITIONS
3
with identity case. For the sake of completeness, we describe in detail about this approach
for son(R) when char(R) is odd, in Section 4.
2. A Necessary condition for sln
We begin with some assertions for the Lie algebra L over R. Note that R can be
decomposed into a finite direct product of finite local rings, i.e., R = R1 × R2 × · · · × Rt,
where Ri is a finite local ring [10, Theorem VI. 2]. We use this fact to observe that if L
can be decomposed into a direct sum of Lie algebras over R1, R2, . . . , Rt, respectively, then
L has an ODAC if and only if each component of L has an ODAC. In particular, sln(R)
has an ODAC if and only if sln(Ri) has an ODAC for all i = 1, 2, . . . , t. Moreover, we show
that if sln(R) has an ODAC, then char(R) must be relatively prime to n.
For each i = 1, 2, . . . , t, let Li be a free Ri-module of finite rank. Suppose that all Li's
have the same rank n. Then L1 ⊕ L2 ⊕ · · · ⊕ Lt is a free R-module of rank n by defining
the scalar multiplication as follows: for all r ∈ R, r = (r1, r2, . . . , rt) for some ri ∈ Ri,
r.(x1, x2, . . . , xt) := (r1x1, r2x2, . . . , rtxt)
where (x1, x2, . . . , xt) ∈ L1 ⊕ L2 ⊕ · · · ⊕ Lt. Assume further that each Li is a Lie algebra
over Ri, then we can naturally define the Lie bracket on L1 ⊕ L2 ⊕ · · · ⊕ Lt by taking the
componentwise bracket. More precisely, for (x1, x2, . . . , xt), (y1, y2, . . . , yt) ∈ L1 ⊕ L2 ⊕ · · · ⊕
Lt,
[(x1, x2, . . . , xt), (y1, y2, . . . , yt)] := ([x1, y1], [x2, y2], . . . , [xt, yt]).
Then L1 ⊕ L2 ⊕ · · · ⊕ Lt is a Lie algebra over R.
From now on we fix R, R1, R2, . . . , Rt and L1, L2, . . . , Lt as above and for 1 ≤ i ≤ t, let
Proji denote a projection from L1 ⊕ L2 ⊕ · · · ⊕ Lt onto Li.
Lemma 2.1. Under the above setting, suppose that there is a Lie algebra isomorphism
φ : L → L1 ⊕ L2 ⊕ · · · ⊕ Lt.
If H is an abelian Cartan subalgebra of L, then Proji(φ(H)) is an abelian Cartan subalgebra
of Li for all i = 1, 2, . . . , t.
Proof. We first note that for a subalgebra A of φ(L),
A = Proj1(A) ⊕ Proj2(A) ⊕ · · · Projt(A)
and if x ∈ Proji(A), then (0, . . . , 0, x
, 0, . . . , 0) ∈ φ(L).
To prove the lemma, it suffices to assume that t = 2 and i = 1 since the similar arguments
It is clear that Proj1(φ(H)) is an R1-submodule of L1. Let
hold for the other cases.
{z}i-th
4
SONGPON SRIWONGSA
x, y ∈ Proj1(φ(H)). Using the scalar (1, 0), we observe that (x, 0) and (y, 0) are in φ(H).
Since H is abelian, so is φ(H) and
[x, y] = Proj1([x, y], [0, 0]) = Proj1[(x, 0), (y, 0)] = Proj1(0, 0) = 0.
Thus, Proj1(φ(H)) is an abelian subalgebra of L1 and so it is nilpotent.
We show that Proj1(φ(H)) is a self-normalizer in L1. For convenience, we denote Hi :=
Proji(φ(H)) for all i = 1, 2. Let x ∈ NL1(H1). Then [x, H1] ⊆ H1. For any h ∈ H,
[(x, 0), φ(h)] = [(x, 0), (Proj1(φ(h)), Proj2(φ(h)))]
= ([x, Proj1(φ(h))], [0, Proj2(φ(h))])
= ([x, Proj1(φ(h))], 0) ∈ (H1, H2) = φ(H).
Then, [(x, 0), φ(H)] ⊆ φ(H). Since H is a self-normalizer in L, so is φ(H) in L1 ⊕ L2. Thus,
(x, 0) ∈ φ(H). Therefore, NL1(H1) = H1.
(cid:3)
From the above lemma, we can derive the following criteria for the Lie algebra L over R
to admit an ODAC.
Theorem 2.2. Under the assumption in Lemma 2.1, L has an ODAC if and only if each
Li has an ODAC.
Proof. Assume that L = H1 ⊕ H2 ⊕ · · · ⊕ Hk is an ODAC of L. We only prove that L1 has
an ODAC
L1 = Proj1(φ(H1)) ⊕ Proj1(φ(H2)) ⊕ · · · ⊕ Proj1(φ(Hk))
and the similar arguments work for the other Li's by using suitable projection maps. By
Lemma 2.1, each Proj1(φ(Hj)) is an abelian Cartan subalgebra of L1. Let x1 ∈ L1 and
x0 ∈ L such that φ(x0) = (x1, 0, . . . , 0). Due to the ODAC of L, we have
for some x0,j ∈ Hj, and
x0 = x0,1 + x0,2 + · · · + x0,k,
x1 = Proj1(φ(x0)) = Proj1(φ(x0,1)) + · · · + Proj1(φ(x0,k)).
So, L1 ⊆Pk
j=1 Proj1(φ(Hj)). On the other hand, it is clear that
L1 ⊇
Proj1(φ(Hj)).
k
Xj=1
Next, let j0 ∈ {1, 2, . . . , k} and x1 ∈ Proj1(φ(Hj0)) ∩Pj6=j0 Proj1(φ(Hj)). Then there exist
i=2 Li such that
(h2, . . . , ht), (h′
2, . . . , h′
(x1, h2, . . . , ht) ∈ φ(Hj0) and (x1, h′
2, . . . , h′
φ(Hj).
t) ∈Pt
t) ∈Xj6=j0
Let r = (1, 0, . . . , 0) ∈ R1 × R2 × . . . × Rt. So,
ORTHOGONAL DECOMPOSITIONS
5
(x1, 0, . . . , 0) = r(x1, h2, . . . , ht) = r(x1, h′
2, . . . , h′
φ(Hj) = {0}
t) ∈ φ(Hj0) ∩Xj6=j0
and hence x1 = 0. So, the sum is direct. Let Ki : Li × Li → Ri be the Killing form. We
show that the Killing form Kφ(L) of φ(L) is equal to
Kφ(L)(φ(x), φ(y)) = (K1(x1, y1), K2(x2, y2), · · · , Kt(xt, yt))
where φ(x) = (x1, x2, . . . , xt) and φ(y) = (y1, y2, . . . , yt) for all x, y ∈ L. Fix a basis
{v1, v2, . . . , vn} for L and a basis {φ(v1), φ(v2), . . . , φ(vn)} for φ(L). For each i = 1, 2, . . . , t,
let v(i)
2 , . . . , v(i)
1 = Proji(φ(v1)), v(i)
n }
is a basis for Li. We have
n = Proji(φ(vn)). Then {v(i)
2 = Proji(φ(v2)), . . . , v(i)
1 , v(i)
[φ(x), φ(vi)] =
aijφ(vj)
n
Xj=1
for some aij ∈ R and aij = (a(1)
ij , a(2)
ij , . . . , a(t)
ij ) ∈ R1 × R2 × · · · × Rt. Moreover,
([x1, v(1)
i
], [x2, v(2)
i
], . . . , [xt, v(t)
i
]) =(cid:16) n
Xj=1
a(1)
ij v(1)
j
,
n
Xj=1
a(2)
ij v(2)
j
, . . . ,
n
Xj=1
a(t)
ij v(t)
j (cid:17).
Then ad φ(x) = (aij) and ad xl = (a(l)
replace a(ij) with b(ij), we have ad φ(y) = (bij) and ad yl = (b(l)
we find that
ij ) for all l = 1, 2, . . . , t. By the similar arguments and
ij ) for all l = 1, 2, . . . , t. So,
Kφ(L)(φ(x), φ(y)) = Tr(ad φ(x) ad φ(y))
= Tr((aij)(bij))
aisbsi
n
n
n
n
=
=
Xs=1
Xi=1
Xs=1
Xi=1
Xs=1
Xi=1
=(cid:16) n
Xs=1
Xi=1
=
n
n
n
(a(1)
is , a(2)
is , . . . , a(t)
is )(b(1)
si , b(2)
si , . . . , b(t)
si )
(a(1)
is b(1)
si , a(2)
is b(2)
si , . . . , a(t)
is b(t)
si )
is b(1)
a(1)
si ,
a(2)
is b(2)
si , . . . ,
n
Xi=1
n
Xs=1
n
Xi=1
n
Xs=1
a(t)
is b(t)
si(cid:17)
ij )(b(1)
= (Tr((a(1)
ij )))
= (Tr(ad x1 ad y1), Tr(ad x2 ad y2), . . . , Tr(ad xt ad yt))
ij )), . . . , Tr((a(t)
ij )), Tr((a(2)
ij )(b(2)
ij )(b(t)
= (K1(x1, y1), K2(x2, y2), · · · , Kt(xt, yt))
6
SONGPON SRIWONGSA
Next, we prove that Proj1(φ(Hj1)) is orthogonal to Proj1(φ(Hj2)) with respect to the Killing
from K1 if j1 6= j2. Let x1 ∈ Proj1(φ(Hj1)) and y1 ∈ Proj1(φ(Hj2)). Then (x1, 0, . . . , 0) ∈
φ(Hj1) and (y1, 0, . . . , 0) ∈ φ(Hj2) are orthogonal to each other. Moreover,
(K1(x1, y1), K2(0, 0), . . . , Kt(0, 0)) = Kφ(L)((x1, 0, . . . , 0), (y1, 0, . . . , 0)) = 0.
Therefore, K1(x1, y1) = 0.
Conversely, we suppose that each Li, i = 1, 2, . . . , t, has an ODAC with ki components.
Let k = max{ki : i = 1, 2, . . . , t}. For each i = 1, 2, . . . , t and j = 1, 2, . . . , k, let Hij be the
jth component of ODAC of Li if j ≤ ki and a zero submodule if otherwise. Then L has an
ODAC
L = H1 ⊕ H2 ⊕ · · · ⊕ Hk,
where Hj = φ−1(H1j, H2j, . . . , Htj).
(cid:3)
We relate the decomposition of a finite commutative ring to an ODAC of a linear Lie
algebra. A Lie algebra isomorphism from gln(R) to gln(R1) ⊕ gln(R2) ⊕ · · · ⊕ gln(Rt) can
be defined as follows. Note that the multiplication
(A(1), A(2), . . . , A(t)) · (B(1), B(2), . . . , B(t)) = (A(1)B(1), A(2)B(2), . . . , A(t)B(t)),
defines an R-algebra structure on gln(R1) ⊕ gln(R2) ⊕ · · · ⊕ gln(Rt). Consequently, we
can define a bracket [·, ·] on it to be the componentwise commutator. It follows that this
R-algebra is a Lie algebra over R. For each a ∈ R, we can write a = (a(1), a(2), . . . , a(t))
uniquely. Define
φ :gln(R) −→ gln(R1) ⊕ gln(R2) ⊕ · · · ⊕ gln(Rt)
(aij) 7−→ ((a(1)
ij ), (a(2)
ij ), . . . , (a(t)
ij ))
for all (aij) ∈ gln(R). Clearly, φ is an R-module isomorphism.
Let A = (aij), B = (bij) ∈ gln(R). Then
φ(AB) = φ((aij)(bij))
il b(1)
a(1)
lj ), (Xl
il b(2)
a(2)
lj ), . . . , (Xl
il b(t)
a(t)
lj ))
ailblj))
= φ((Xl
= ((Xl
= ((a(1)
= ((a(1)
ij )(b(1)
ij ), (a(2)
ij )(b(2)
ij ), (a(2)
ij ), . . . , (a(t)
ij ), . . . , (a(t)
ij )) · ((b(1)
= φ((aij))φ((bij)) = φ(A)φ(B).
ij )(b(t)
ij ), (b(2)
ij ))
ij ), . . . , (b(t)
ij ))
So, φ([A, B]) = [φ(A), φ(B)]. Thus, φ is a Lie algebra isomorphism and by Theorem 2.2,
we have the following theorem.
ORTHOGONAL DECOMPOSITIONS
7
Theorem 2.3. Under the above setting, we have the following:
(i) There is a Lie algebra (over R) isomorphism
φ : gln(R) −→ gln(R1) ⊕ gln(R2) ⊕ · · · ⊕ gln(Rt).
(ii) If g is a Lie subalgebra of gln(R), then
g ∼= Proj1(φ(g)) ⊕ Proj2(φ(g)) ⊕ · · · ⊕ Projt(φ(g)).
Moreover, g has an ODAC if and only if Proji(φ(g)) has an ODAC for all i =
1, 2, . . . , t.
We now consider the special linear Lie algebra over R. By Theorem 2.3,
sln(R) ∼= Proj1(φ(sln(R))) ⊕ Proj2(φ(sln(R))) ⊕ · · · ⊕ Projt(φ(sln(R))).
It is straightforward to verify that Proji(φ(sln(R))) = sln(Ri) for all i = 1, 2, . . . , t. There-
fore, we have:
Theorem 2.4. sln(R) has an ODAC if and only if sln(Ri) has an ODAC for all i =
1, 2, . . . , t.
Using the above theorem, we obtain a necessary condition on the ring R and n for the
existence of an ODAC of sln(R).
Theorem 2.5. If sln(R) admits an ODAC, then char(R) is relatively prime to n.
Proof. Suppose that char(R) is not relatively prime to n. Then
and
char(R) = paps1
1 ps2
2 · · · psl
l
n = pbpt1
1 pt2
2 · · · ptl
l
where p and pi's are all distinct prime integers and a, s1, . . . , sl, b, t1, . . . , tl are non negative
integers. Since R = R1 × R2 × · · · × Rt is a finite product of finite local rings and each
Ri has characteristic a power of a prime integer, there exists i0 ∈ {1, 2, . . . , t} such that
char(Ri0) = pa. Consider sln(Ri0); we have two distinct cases.
Case 1: b ≥ a. Then n is divisible by pa and so the trace of the identity matrix In is 0.
Thus, sln(Ri0) contains In and so does every abelian Cartan subalgebra. Thus, any two
abelian Cartan subalgebras have a nontrivial intersection. Since sln(Ri0) is not abelian, it
does not have an ODAC.
Case 2: b < a. Then pa−bIn is an element of sln(Ri0). By the similar reason to the case 1,
sln(Ri0) does not admit an ODAC.
Hence, by Theorem 2.4, sln(R) does not have an ODAC.
(cid:3)
By the above theorem, we have the following example.
8
SONGPON SRIWONGSA
Example 1. sl6(R) does not have an ODAC if R has one of the following rings as its
summand: F2m, F3m, Z2m and Z3m.
3. ODAC of sp2m+1
In the complex number case, the OD problem of Lie algebra of type C has the same
difficulty as type A. However, in the special case of the Lie algebra of type C2m, it is
manageable because sp2m+1(C) is a subalgebra of sl2m+1(C) and an OD of this Lie algebra
is constructible [9, Chapter 1]. Here, we consider the ODAC problem of sp2m+1(R) when
the characteristic of R is odd. Note that −1 ∈ R is the primitive square root of unity and
−2 is a unit in R. By Theorem 3.1 in [12], an ODAC of sl2m+1(R) exists. Restricting this
ODAC of sl2m+1(R), we can show that sp2m+1(R) also has an ODAC. Note that the Killing
form for spn(R) is equal to
K(A, B) = (4n + 2)Tr(AB)
for all A, B ∈ sp2n(R).
We recall that
sp2m+1(R) = {X ∈ M2m+1(R) : XK + KX T = 0},
where K = 0
−I2m
I2m
0 !. Let
D = 1
0 −1! and P = 0 1
1 0! .
0
Let W = F2m+1 ⊕ F2m+1 be a 2(m + 1)-dimensional vector space over F2 equipped with a
symplectic form h·, ·i : W × W → F2 defined by the field trace1 as follows: for any elements
~w = (α; β), ~w′ = (α′; β′) ∈ W ,
h ~w, ~w′i = T rF
2m+1 /F2(αβ′ − α′β).
Then, by Corollary 3.3 of [16], W possesses a symplectic basis B = {~e1, . . . , ~em+1, ~f1, . . . , ~fm+1}
where {~e1, . . . , ~em+1} and { ~f1, . . . , ~fm+1} span the first and the second factor, respectively,
such that
h ~w, ~w′i =
(aib′
i − a′
ibi),
m+1
Xi=1
1The field trace of α ∈ F2m+1 is defined to be the sum of all Galois conjugates of α, i.e.
TrF
2m+1 /F2(α) = α + α2 + · · · + α2m
.
ORTHOGONAL DECOMPOSITIONS
9
where ~w = Pm+1
vector ~w ∈ W as
i=1 (ai~ei + bi
~fi) and ~w′ = Pm+1
i=1 (a′
i~ei + b′
i
~fi). With the basis B, write each
~w = (a1, . . . , am+1; b1, . . . , bm+1)
and associate it with a matrix
J ~w = J(a1,b1) ⊗ J(a2,b2) ⊗ · · · ⊗ J(am+1,bm+1),
where J(a,b) = DaP b and ⊗ is a Kronecker product 2. Moreover, we define
q( ~w) :=
m+1
Xi=1
aibi + (a1 + b1).
Then the above symplectic form is equal to
h ~w, ~w′i = q( ~w) + q( ~w′) + q( ~w + ~w′)
for all ~w, ~w′ ∈ W and (W, q) is a nondegenerate quadratic space with Witt index m (Propo-
sition 1.5.42 in [3]). We note that (W, h·, ·i) is a symplectic space with maximum totally
isotropic subspaces of dimension m + 1.
Let Q = { ~w ∈ W : q( ~w) = 1}. We will describe a special basis of sp2m+1(R) by using Q
in the next theorem. This special basis will be used for the construction of an ODAC of
sp2m+1(R).
Theorem 3.1. The Lie algebra sp2m+1(R) has {J ~w : ~w ∈ Q} as a basis.
2The Kronecker product of an m × n matrix A = (aij ) and a p × q matrix B is defined to be the mp × nq
a11B · · ·
. . .
am1B · · ·
...
a1nB
...
amnB
block matrix: A ⊗ B =
.
10
SONGPON SRIWONGSA
Proof. Write J ~w = J(a1,b1) ⊗ J~v, where ~v = (a2, . . . , am+1; b2, . . . , bm+1). Note that K =
DP ⊗ I2m. We show that if ~w ∈ Q, then J ~w ∈ sp2m+1(R). Set S =Pm+1
KJ T
~w = (−1)SKJ ~w
i=1 aibi. Consider
= (−1)S(DP ⊗ I2m)(J(a1,b1) ⊗ J~v)
= (−1)S(DP J(a1,b1)) ⊗ J~v
= (−1)S(DP Da1P b1) ⊗ J~v
= (−1)S+a1+b1(Da1P b1DP ) ⊗ J~v
= (−1)S+a1+b1(J(a1,b1)DP ) ⊗ J~v
= (−1)S+a1+b1(J(a1,b1)DP ) ⊗ J~vI2m
= (−1)S+a1+b1(J(a1,b1) ⊗ J~v)(DP ⊗ I2m)
= (−1)S+a1+b1J ~wK
= (−1)q( ~w)J ~wK.
Since ~w ∈ Q, KJ T
~w = −J ~wK, i.e. J ~w ∈ sp2m+1(R).
Note that the set {J(0,0), J(0,1), J(1,0), J(1,1)} is linearly independent. It follows from basic
properties about Kronecker (tensor) products that the set {J ~w : 0 6= ~w ∈ W } is linearly
independent and so is the set {J ~w : ~w ∈ Q}. To complete the proof, we show that
Q = 2m(2m+1 + 1) which is the rank of sp2m+1(R) as a free R-module. Then SpanR({J ~w :
~w ∈ Q}) = sp2m+1(R) since R is finite. Let ~w = (a1, . . . , am+1; b1, . . . , bm+1) ∈ Q. Then
a1b1 + a1 + b1 = 1 +
aibi.
m+1
Xi=2
Case 1: a1 = 0. Then b1 = 1 +Pm+1
i=2 aibi. Hence,
Ω0 = { ~w ∈ Q : ~w = (0, a2, . . . , am+1; b1, . . . , bm+1)}
has 22m elements.
Case 2: a1 = 1. Then Pm+1
Ωj =({ ~w ∈ Q : ~w = (1, 0, . . . , 0; b1, . . . , bm+1)}
i=2 aibi = 0 and b1 is 0 or 1. Let
{ ~w ∈ Q : ~w = (1, 0, . . . , 0, 1, aj+1, . . . , am+1; b1, . . . , bm+1)}
if j = 1,
if 2 ≤ j ≤ m + 1.
Then Ω1 = 2m+1. For 2 ≤ j ≤ m + 1, if a2 = . . . = aj−1 = 0 and aj = 1, then b2, . . . , bj−1
i=j+1 aibi. Thus, Ωj = 22m−j+1. If a2 = . . . = am = 0 and am+1 = 1,
then b2, . . . , bm are 0 or 1 and am+1 = bm+1 = 1. Thus, Ωm+1 = 2m.
are 0 or 1 and bj =Pm+1
ORTHOGONAL DECOMPOSITIONS
Note that {Ω0, Ω1, . . . , Ωm+1} is a partition of Q. Therefore,
Q =
m+1
Xj=0
Ωj = 22m + 2m+1 +
m+1
Xj=2
22m−j+1 = 2m(2m+1 + 1)
as desired.
11
(cid:3)
Next we construct an ODAC of sp2m+1(R) by using the basis in the above theorem.
Note that sp2m+1(R) is a subalgebra of sl2m+1(R) and by Theorem 3.1 in [12], an ODAC of
sl2m+1(R) is
sl2m+1(R) = H∞ ⊕ (⊕α∈F
2m+1 Hα),
and Hα = hJ(α;λα)λ ∈ F×
2m+1iF
where H∞ = hJ(0;λ)λ ∈ F×
for all α ∈ F2m+1.
The basis in Theorem 3.1 is the union of some subsets of these Hj's. We show that the
components of an ODAC of sp2m+1(R) can be obtained from the Hi's by picking up the
elements whose index belongs to Q. We use the following lemma to verify the constructed
decomposition is an ODAC.
2m+1iF
2m+1
2m+1
Lemma 3.2. For each α ∈ F2m+1, let Wα = {(λ; αλ) ∈ W : λ ∈ F×
{(0; λ) ∈ W : λ ∈ F×
2m+1}. Then
(1) W = (cid:16)Sα∈F
2m+1
subspaces of W .
Wα(cid:17) ∪ W∞ where Wα = Wα ∪ {(0; 0)},
2m+1}, and let W∞ =
W∞ = W∞ ∪ {(0; 0)} are
(2) For α ∈ F2m+1 ∪ {∞}, if Qα = Wα ∩ Q, then Wα = hQαiF2.
Proof. It is clear that the Wα's are subspaces of W and (1) holds. To prove (2), we first
note that Qc = W \ Q = { ~w ∈ W : q( ~w) = 0} and
Qc \ {(0; 0)} = W \ {(0; 0)} − Q
= (22(m+1) − 1) − 2m(2m+1 + 1)
= (2m − 1)(2m+1 + 1).
(3.1)
We show that for all α ∈ F2m+1 ∪ {∞}, Wα ∩ (Qc \ {(0; 0)}) ≥ 2m − 1. Suppose, to
the contrary, that there exists an α such that Wα ∩ (Qc \ {(0; 0)}) < 2m − 1. Then by
(3.1), there exists an α′ such that Wα′ ∩ (Qc \ {(0; 0)}) ≥ 2m. So Wα′ ∩ Qc ≥ 2m + 1.
Wα′ ∩ Qc is a totally isotopic subspace of (W, q). Indeed, if ~w1, ~w2 ∈ Wα′ ∩ Qc, then
But
q( ~w1 + ~w2) = q( ~w1) + q( ~w2) + h ~w1, ~w2i = 0. Thus, dim( Wα′ ∩ Qc) ≤ m and as a subspace
over F2, Wα′ ∩ Qc ≤ 2m. This is a contradiction.
Now, for each α ∈ F2m+1 ∪ {∞}, by (3.1), Wα ∩ (Qc \ {(0; 0)}) = 2m − 1, and hence,
Wα ∩ Q = (2m+1 − 1) − (2m − 1) = 2m.
Let Qα = Wα ∩ Q. Then hQαiF2 is a totally isotopic subspace of (W, h·, ·i) and Wα ⊇
hQαiF2. We have dim(hQαiF2) ≤ m + 1. But since hQαiF2 ≥ Qα + 1 = 2m + 1,
12
SONGPON SRIWONGSA
dim(hQαiF2) ≥ m + 1 which forces dim(hQαiF2) = m + 1. Thus, hQαiF2 = 2m+1 = Wα,
and so Wα = hQαiF2.
(cid:3)
Theorem 3.3. For a positive integer m, sp2m+1(R) has an ODAC obtained by restricting
an ODAC of sl2m+1(R) constructed in Theorem 3.1 in [12].
Proof. For each α ∈ F2m+1, let
H ′
α = hJ(λ;αλ)λ ∈ F×
2m+1 and (λ; αλ) ∈ QiR
,
and let
H ′
∞ = hJ(0;λ)λ ∈ F×
2m+1 and (0; λ) ∈ QiR
.
It follows from the proof of Theorem 3.1 in [12] and Theorem 3.1 that all H ′
F2m+1 ∪ {∞} are orthogonal abelian subalgebras of sp2m+1(R) and the sum of all these H ′
is direct. Thus,
α's, α ∈
α's
sp2m+1(R) = H ′
∞ ⊕ (⊕α∈F
2m+1 H ′
α).
To show that each H ′
Then
α is a self-normalizer in sp2m+1(R), let α ∈ F2m+1 and A ∈ Nsp2m+1 (R)(H ′
α).
bλ′J(0;λ′)
bλ′[J(0;λ′), J(λ;αλ)] ∈ H ′
α.
A = Xβ ′∈Fq Xλ′∈F×
q
(λ′;β ′λ′)∈Q
a(λ′,β ′)J(λ′;β ′λ′)! + Xλ′∈F×
q
(0;λ′)∈Q
where a(λ′,β ′) and bλ′ are elements in R. For any J(λ;αλ) ∈ H ′
α,
[A, J(λ;αλ)] = Xβ ′∈Fq Xλ′∈F×
q
(λ′;β ′λ′)∈Q
a(λ′,β ′)[J(λ′;β ′λ′), J(λ;αλ)]! + Xλ′∈F×
q
(0;λ′)∈Q
This implies
Xβ ′∈Fq
β ′6=α
Xλ′∈F×
q
(λ′;β ′λ′)∈Q
a(λ′,β ′)[J(λ′;β ′λ′), J(λ;αλ)]! + Xλ′∈F×
q
(0;λ′)∈Q
bλ′[J(0;λ′), J(λ;αλ)] ∈ H ′
α.
For each (λ′, β′), if for all (λ; αλ) ∈ Q, h(λ; αλ), (λ′; β′λ′)i = 0, then by Lemma 3.2,
J(λ′;β ′λ′) would be in Nsl2m+1 (R)(Hα) = Hα which is a contradiction. So, we may assume
that we can choose (λ; αλ) ∈ Q such that h(λ; αλ), (λ′; β′λ′)i = 1. Argue as in the proof
of Theorem 3.1 in [12], we obtain a(λ′,β ′) = 0. Similarly, bλ′ = 0. Thus, A ∈ H ′
α, and so
Nsp2m+1 (R)(H ′
α. By an analogous argument, we also have Nsp2m+1 (R)(H ′
∞) = H ′
∞.
Hence, sp2m+1(R) has an ODAC.
(cid:3)
α) = H ′
ORTHOGONAL DECOMPOSITIONS
13
4. ODAC of son
We again assume that R has odd characteristic. Recall that
so2n(R) = hX(i,j)1 ≤ i 6= j ≤ 2niR ,
where X(i,j) = eij − eji and eij is the matrix having 1 in the (i, j) position and 0 elsewhere.
We utilize these basis elements to construct an ODAC of this Lie algebra. This technique
was also used for an OD of so2n(C) [8, 9]. Note that the Killing form is equal to
K(A, B) = (2n − 2)Tr(AB)
for all A, B ∈ so2n(R).
The matrices X(i,j)'s satisfy the following properties:
Lemma 4.1. Keep the above notations and denoted by {·, ·} an unordered pair, we have
(1) X(i,j) = −X(j,i).
(2) If {i, j} 6= {k, l}, then Tr(X(i,j)X(k,l)) = 0.
if j = k,
if {i, j} ∩ {k, l} = Ø.
(3) [X(i,j), X(k,l)] =(X(i,l)
0
Proof. The first property is clear from the definition. To prove (2), we first compute
X(i,j)X(k,l) = (eij − eji)(ekl − elk)
= eijekl − eijelk − ejiekl + ejielk.
Assume that {i, j} 6= {k, l}. We consider two distinct cases.
Case 1: i 6= k and l. We have X(i,j)X(k,l) = eijekl − eijelk. Then Tr(X(i,j)X(k,l)) = 0.
Case 2: j 6= k and l. We have X(i,j)X(k,l) = −ejiekl + ejielk. Then Tr(X(i,j)X(k,l)) = 0.
Finally,
[X(i,j), X(k,l)] = X(i,j)X(k,l) − X(k,l)X(i,j)
= (eijekl − eijelk − ejiekl + ejielk) − (ekleij − ekleji − elkeij + elkeji)
=(X(i,l)
0
if j = k,
if {i, j} ∩ {k, l} = Ø,
as claimed.
(cid:3)
We use the relations in the above lemma to construct an ODAC of so2n(R). To do that,
we introduce the following set of unordered pairs and its partition. Let
and let
X = {{i, j} : 1 ≤ i 6= j ≤ 2n}
P = {Mk : 1 ≤ k ≤ 2n − 1}
14
SONGPON SRIWONGSA
be a partition of X, where Mk = n and α ∩ β = Ø for any α, β ∈ Mk such that α 6= β.
This P can be viewed as a partition of the complete graph with vertex set {1, 2, . . . , 2n}
and edge set X, it is also called 1-factorization of the graph which is constructible [6,
Theorem 9.1]. Note that X = n(2n − 1) which is equal to the rank of so2n(R) as an
R-module.
Theorem 4.2. For a positive integer n, so2n(R) has an ODAC
so2n(R) = H1 ⊕ H2 ⊕ · · · ⊕ H2n−1,
where Hk = hX(i,j){i, j} ∈ MkiR.
Proof. By Lemma 4.1 (2) and (3), we have the orthogonality and the commutativity of
Hk's. Next, we show that Nso2n(R)(Hk) = Hk. Let A ∈ Nso2n(R)(Hk) and write it as a linear
combination of the elements X(i,j)
For any X(s,t) ∈ Hk,
and so
αijX(i,j).
A =Xi6=j
[A, X(s,t)] =Xi6=j
αij[X(i,j), X(s,t)] ∈ Hk,
αij[X(i,j), X(s,t)] ∈ Hk.
Xi6=j
{i,j} /∈Mk
For each pair (i, j), since the Mk's form a partition of X, there exists X(j,t) ∈ Hk such that
t 6= i and [X(i,j), X(j,t)] = X(i,t) 6= 0 by Lemma 4.1. Therefore, αij = 0, and so A ∈ Hk. (cid:3)
Finally, we present the existence of an ODAC of the Lie algebra
so2n−1(R) = hX(i,j)1 ≤ i 6= j ≤ 2n − 1iR .
Note that the Killing form is equal to
K(A, B) = (2n − 3)Tr(AB)
for all A, B ∈ so2n−1(R). Similarly, we let
X ′ = {{i, j} : 1 ≤ i 6= j ≤ 2n − 1}.
In the next step, we construct a partition of this set into subsets M ′
k satisfying
M ′
k = n − 1 and α ∩ β = Ø for all α, β ∈ M ′
k, α 6= β.
The construction can be obtained from all Mk's of the construction of an ODAC of so2n(R)
in the above discussion. Without loss of generality, we assume that each Mk contains the
pair {k, 2n}. Let M ′
k = Mk \ {k, 2n}.
Theorem 4.3. For a positive integer n ≥ 2, so2n−1(R) has an ODAC
ORTHOGONAL DECOMPOSITIONS
15
so2n−1(R) = H ′
1 ⊕ H ′
2 ⊕ · · · ⊕ H ′
2n−1,
where H ′
k = hX(i,j){i, j} ∈ M ′
kiR.
Proof. We only need to show that each H ′
k is a self-normalizer because analogous arguments
from the proof of Theorem 4.2 can be used to prove the rest. Let A ∈ Nso2n−1(R)(Hk) and
write it as a linear combination of the elements X(i,j)
For any X(s,t) ∈ Hk,
and so
αijX(i,j).
A =Xi6=j
[A, X(s,t)] =Xi6=j
αij[X(i,j), X(s,t)] ∈ Hk,
αij[X(i,j), X(s,t)] ∈ Hk.
Xi6=j
{i,j} /∈M ′
k
For each pair (i, j), if j 6= k, we can use the argument provided in Theorem 4.2 to show
αij = 0.
If j = k, we use the relation (1) of Lemma 4.1 to interchange i and j. This
completes the proof.
(cid:3)
References
[1] A. Bondal and I. Zhdanovskiy, Orthogonal pairs and mutually unbiased bases, J. Math. Sci. (N.Y.),
216 (2016), no. 1, 23 -- 40.
[2] P. O. Boykin, M. Sitharam, P. H. Tiep and P. Wocjan, Mutually unbiased bases and orthogonal
decompositions of Lie algebras, Quantum Inf. Comput., 7 (2007) 371 -- 382.
[3] J. N. Bray, D. F. Holt and C. Roney-Dougal, The maximal subgroups of the low-dimensional finite
classical groups, LMS Lecture Notes Ser. 407, Cambridge UP, 2013.
[4] S. P. Demushkin, Cartan subalgebras of simple nonclassical Lie p-algebras, Izv. Akad. Nauk SSSR Ser.
Mat., 36:5 (1972) 915 -- 932.
[5] T. Durt, B. G. Englert, I. Bengtsson and K. Zyczkowski, On mutually unbiased bases, Int. J. Quantum
Inform., 8 (2010) 535 -- 640.
[6] F. Harary, Graph Theory, Addison-Wesley, Reading, Mass., 1969.
[7] A. I. Kostrikin, I. A. Kostrikin and V. A. Ufnarovskii, Orthogonal decompositions of simple Lie
algebras (type An), Trudy Mat. Inst. Steklov., 158 (1981) 105 -- 120.
[8] A. I. Kostrikin, I. A. Kostrikin and V. A. Ufnarovskii, On decompositions of classical Lie algebras,
Trudy Mat. Inst. Steklov., 166 (1984) 117 -- 134.
[9] A. I. Kostrikin and P. H. Tiep, Orthogonal Decompositions and Integral Lattices, Walter de Gruyter,
1994.
[10] B. R. McDonald, Finite Rings with Identity, Marcel Dekker, New York, 1974.
[11] M. B. Ruskai, Some connections between frames, mutually unbiased bases, and POVM's in quantum
information theory, Acta Appl. Math., 108 (2009), no. 3, 709 -- 719.
16
SONGPON SRIWONGSA
[12] S. Sriwongsa, Y. M. Zou, Orthogonal Cartan subalgebra decomposition of sln over a finite commutative
ring, Linear Multilinear Algebra, 2018 In press, arXiv:1802.02275 [math.RA].
[13] J. G. Thompson, A conjugacy theorem for E8, J. Algebra, 38 (1976), no. 2, 525 -- 530.
[14] J. G. Thompson, A simple subgroup of E8(3). In Finite Groups Symposium, N. Iwahori ed., Japan
Soc. for Promotion of Science, pages 113 -- 116, 1976.
[15] A. Torstensson, On the existence of orthogonal decompositions of the simple Lie algebra of type C3,
Comput. Sci. J. Moldova, 8 (2000), no. 1(22), 16 -- 41.
[16] Z. Wan, Geometry of Classical Groups over Finite Fields, 2nd Edition. Beijing, New York: Science
Press; 2002.
Songpon Sriwongsa, Department of Mathematical Sciences, University of Wisconsin-
Milwaukee, USA
E-mail address: [email protected]
|
1110.3007 | 1 | 1110 | 2011-10-13T17:42:49 | Cohomology of Restricted Lie-Rinehart Algebras and the Brauer Group | [
"math.RA",
"math.KT"
] | We give an interpretation of the Brauer group of a purely inseparable extension of exponent 1, in terms of restricted Lie-Rinehart cohomology. In particular, we define and study the category $p$-$\rm{LR}(A)$ of restricted Lie-Rinehart algebras over a commutative algebra $A$. We define cotriple cohomology groups $H_{p-LR}(L,M)$ for $L\in p$-$\rm{LR}(A)$ and $M$ a Beck $L$-module. We classify restricted Lie-Rinehart extensions. Thus, we obtain a classification theorem for regular extensions considered by Hoshschild. | math.RA | math |
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS
AND THE BRAUER GROUP
IOANNIS DOKAS
Abstract. We give an interpretation of the Brauer group of a purely insep-
arable extension of exponent 1, in terms of restricted Lie-Rinehart cohomol-
ogy.
In particular, we define and study the category p-LR(A) of restricted
Lie-Rinehart algebras over a commutative algebra A. We define cotriple co-
homology groups Hp−LR(L, M ) for L ∈ p-LR(A) and M a Beck L-module.
We classify restricted Lie-Rinehart extensions. Thus, we obtain a classification
theorem for regular extensions considered by Hoshschild.
Introduction
In classical theory of simple algebras it is known that if E/F is a Galois extension
of fields, then the Brauer group BE
F of this extension is isomorphic with the group
of equivalence classes of group extensions of the Galois group Gal(E/F ) by the
multiplicative group F ∗ of F . Moreover, we have an isomorphism of groups relating
the Galois cohomology and the Brauer group:
(0.1)
BE
F ≃ H 2(Gal(E/F ), F ∗)
In the context of purely inseparable extensions of exponent 1 there is a Galois
theory due to Jacobson (see [18]). The role of Galois group is now played by the
group of derivations. Purely inseparable extensions occur naturally in algebraic
geometry.
In particular, such extensions appear in the theory of elliptic curves
in prime characteristic. If k is a field such that char k = p and V an algebraic
variety over k of dimension greater than 0, then the function field k(V ) is a purely
inseparable extension over the subfield k(V )p of pth powers.
Hochschild in [14] proves that the Brauer group BK
k of purely inseparable exten-
sion K/k of exponent 1, is isomorphic with the set of equivalence classes of regular
extensions of restricted Lie algebras of Derk(K) by K. We remark that the second
Hochschild cohomology group H 2
Hoch(L, M ) where L is a restricted Lie algebra and
M a restricted Lie -module classifies abelian extensions such that M [p] = 0. This
implies that, there is not an isomorphism between the Brauer group BK
k and a
subgroup of the cohomology group H 2
Hoch(Derk(K), K). This remark has been the
motivation to undertake this research. In order to classify regular extensions and
obtain the analogue to (0.1) cohomological interpretation of the Brauer group, we
are led to define and study the category of restricted Lie-Rinehart algebras. The
concept of Lie-Rinehart algebra is the algebraic counterpart of the notion of Lie
algebroid (see [23]). It seems that the notion of Lie-Rinehart algebra appears first
under the name pseudo-alg`ebre de Lie in the paper [12] of Herz. Also, the notion
1991 Mathematics Subject Classification. 17B63, 17A32, 18G60, 18G15.
Key words and phrases. Restricted Lie algebra, Lie-Rinehart algebra, Brauer group, Quillen-
Barr-Beck cohomology.
1
2
IOANNIS DOKAS
has been examined by Palais under the name d-ring. The first thorough study of
the notion has been done by Rinehart in [28]. Rinehart is the first who defined
cohomology groups for the category of Lie-Rinehart algebras with coefficients in a
Lie-Rinehart module and further developments has been done by Huebschmann in
[15]. Moreover, cotriple cohomology for the category of Lie-Rinehart algebras has
been defined in [9] by Casas- Lada- Pirashvili. Besides, the notion of Lie-Rinehart
algebra is closely related to the notion of Poisson algebra. Loday-Vallette in [21] re-
marked that a Lie-Rinehart algebra is a Poisson algebra. Thus, all the constructions
and properties of Poisson algebras apply to Lie-Rinehart algebras.
In Section 1 we define the category p-LR(A) of restricted Lie-Rinehart alge-
bras over a commutative algebra A. We give examples which occur naturally. In
Section 2 we introduce the notion of restricted enveloping algebra and restricted
Lie-Rinehart module. We prove a Poincar´e-Birkhoff-Witt type theorem for the cat-
egory of restricted Lie-Rinehart algebras. There is a notion of free Lie-Rinehart
algebra made explicit by Casas-Pirashvili (see [9]) and Kapranov (see [20]). We ex-
tend this notion for the category p-LR(A) and we construct free functor left adjoint
to the forgetful functor. In Section 3 following the general scheme of Quillen-Barr-
Beck cohomology theory for universal algebras, we determine the Beck modules and
Beck derivations. In Section 4 we define cohomology groups H ∗
p−LR(A)(L, M ) for
L ∈ p-LR(A) and M a Beck L-module. We prove that Quillen-Barr-Beck cohomol-
ogy for p-LR(A) classifies extensions of restricted Lie-Rinehart algebras. Regular
extensions considered by Hochschild are restricted Lie-Rinehart extensions. As a
consequence in Section 5, we prove that if K/k is purely inseparable extension of
exponent 1, there is an isomorphism of groups
BK
k ≃ H 1
p−LR(Derk(K), K)
1. Restricted Lie-Rinehart algebras
In many cases when we study Lie algebras over a field of prime characteristic
we are led to consider a richer structure than an ordinary Lie algebra. Indeed the
notion of a Lie algebra has to be replaced by the notion of restricted Lie algebra
introduced by N. Jacobson in [17]. Let us recall the definition.
Definition 1.1. A restricted Lie algebra (L, (−)[pL]) over a field k of characteristic
p 6= 0 is a Lie algebra L over k together with a map (−)[pL] : L → L called the
p-map such that the following relations hold:
(1.1)
(1.2)
(1.3)
(αx)[pL] = αp x[pL]
[x, y[pL]] = [x, y], y], · · · , y
]
p
{z
}
(x + y)[pL] = x[pL] + y[pL] +
p−1
Xi=1
si(x, y)
where isi(x, y) is the coefficient of λi−1 in adp−1
λx+y(x), where adx : L → L denotes
the adjoint representation given by adx(y) := [y, x] and x, y ∈ L, α ∈ k. We denote
by RLie the category of restricted Lie algebras over k.
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP3
A Lie-module A over a restricted Lie algebra (L, (−)[p]) is called restricted if
x[p]m = (x · · · (x(x
m) · · · )
p
{z
}
Example 1.2. Let R be any associative algebra over a field k with characteristic
p 6= 0. We denote by RLie the induced Lie algebra with the bracket given by
[x, y] := xy − yx, for all x, y ∈ R. Then (R, (−)p) is a restricted Lie algebra
where (−)p is the Frobenious map given by x 7→ xp. Thus, there is a functor
(−)RLie : As → RLie from the category of associative algebras to the category of
restricted Lie algebras.
Proposition 1.3. Let L be a restricted Lie algebra and A a restricted L-module.
If we denote by A ⊕ L the direct sum of the underlying vectors spaces of A and L
then A ⊕ L is endowed with the structure of restricted Lie-algebra.
Proof. It is well known that A ⊕ L is endowed with the structure of a Lie algebra
with bracket given by:
[a + X, b + Y ] = (X(b) − Y (a)) + [X, Y ]
for any a, b ∈ A and X, Y ∈ L.
Moreover, we have:
[a + X, Y, ] · · · , Y
] = −(Y (Y (. . . (Y
(a)))) + [X, Y ], · · · , Y
]
p
{z
}
p
(a)))) + [X, Y [p]]
{z
}
p
{z
{z
p
}
}
= −(Y (Y (. . . (Y
= [a + X, Y [p]]
Besides,
[a + X, b] · · · , b
] = 0
Therefore, from Jacobson's theorem there exists a unique p-map
p
{z }
(−)[p] : A ⊕ L → A ⊕ L
which extents the p-map on L and such that a[p] = 0 for all a ∈ A. In particular
we see that the p-map on A ⊕ L is given by:
(a + X)[p] = (X(X(. . . (X
(a)))) + X [p]
We denote this restricted Lie algebra structure on A ⊕ L by A ⋊ L.
p−1
{z
}
(cid:3)
Remark 1.4. Let A be a restricted L-module. We consider the invariants for the
Lie action AL = {a ∈ A : la = 0,
for all l ∈ L, a ∈ A}. If f : A → AL is a p-semi
linear map, then it is easily seen that A ⊕ L is a restricted Lie algebra with p-map
given by (a + X)[p] := (X(X(. . . (X
(a)))) + X [p] + f (a). We denote this restricted
Lie algebra by A ⋊f L.
p−1
{z
}
Let A be a commutative algebra over a field k. A k-linear map D : A → A is
called a k-derivation if
D(ab) = aD(b) + D(a)b
4
IOANNIS DOKAS
It is well known that if D, D′ ∈
Let Derk(A) be the set of k-derivations of A.
Der(A), then [D, D′] := DD′ − D′D is a derivation. Thus, (Der(A), [−, −]) is a
Lie algebra. Moreover, if D ∈ Derk(A) and a, x ∈ A then aD : A → A, given
by: (aD)(x) := aD(x) is a derivation. Therefore, Derk(A) has the structure of an
A-module. Besides the following relation holds:
[D, aD′] = a[D, D′] + D(a)D′
The structure on Derk(A) described above is the prototype example of the notion
of Lie-Rinehart algebra. Let us recall the definition.
Definition 1.5. A Lie-Rinehart algebra over A, or (k − A)-Lie algebra, is a pair
(A, L) where, A is a commutative algebra over k, L is a Lie algebra over k equipped
with the structure of an A-module together with a map called anchor α : L →
Derk(A) which is an A-module and a Lie algebra morphism such that:
for all a ∈ A and X, Y ∈ L.
[X, aY ] = a[X, Y ] + α(X)(a)Y
In order to simplify the notation we denote α(X)(a) by X(a). Moreover, we
denote by LR(A) the category of Lie-Rinehart algebras over A.
Example 1.6. We easily see that (A, Derk(A)) is a Lie-Rinehart algebra with
anchor map id : Derk(A) → Derk(A).
Suppose now that the ground field k is a field of characteristic p 6= 0. Let
D ∈ Derk(A), from Leibniz rule for all a, b ∈ A we have
Since char k = p we get:
Dp(ab) =
i=p
Xi=0
(cid:18)p
i(cid:19)Di(a)Dp−i(b)
Dp(ab) = aDp(b) + Dp(a)b
Therefore, Dp is a derivation. In other words Derk(A) is equipped with the structure
of restricted Lie algebra. Moreover, by Hochschild's Lemma 1 in [14] we get the
relation:
(1.4)
(aD)p = apDp + (aD)p−1(a)D
Therefore, we see in prime characteristic that the set of derivations Derk(A) has a
richer structure than just a Lie-Rinehart algebra structure. We are naturally led
to the following definition of restricted Lie-Rinehart algebra.
From now on we fix a field k of characteristic p 6= 0.
Definition 1.7. A restricted Lie-Rinehart algebra (A, L, (−)[p]) over a commu-
tative k-algebra A, is a Lie-Rinehart algebra over A such that: (L, (−)[p]) is a
restricted Lie algebra over k, the anchor map is a restricted Lie homomorphism,
and the following relation holds:
(1.5)
(aX)[p] = apX [p] + (aX)p−1(a)X
for all a ∈ A and X ∈ L.
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP5
Let (A, L, (−)[p]) and (A′, L′, (−)[p]) be restricted Lie-Rinehart algebras. Then
a Lie-Rinehart morphism (ξ, f ) : (A, L, (−)[p]) → (A′, L′, (−)[p]) is called restricted
Lie-Rinehart morphism if f is a restricted Lie morphism, namely: f (x[p]) = f (x)[p]
for all x ∈ L. We will denote by p − LR the category of restricted Lie-Rinehart
algebras.
Example 1.8. As we have seen if A is a commutative algebra over k, then Derk(A)
is a restricted Lie-Rinehart algebra.
Example 1.9. Any restricted Lie algebra over k is a restricted Lie-Rinehart algebra
(A, L, (−)[p]), where A = k.
The structure of restricted Lie Rinehart algebra appears in Jacobson-Galois the-
ory of purely inseparable extensions of exponent 1.
Example 1.10. Let K/k be a purely inseparable field extension of exponent 1.
Then, there is a one-to-one correspondence between intermediate fields and re-
stricted Lie Rinehart sub-algebras of the restricted Lie Rinehart algebra Derk(K)
over K (see [18]).
The Jacobson-Galois correspondence for purely inseparable extensions of expo-
nent 1 has been used by several authors in order to study isogenies of algebraic
groups especially of abelian varietes. The notion of restricted Lie-Rinehart appears
in this study.
Example 1.11. Let G be an algebraic group over k, and K be the field of rationales
functions of G. Then, the K-space of derivations Dk(K) is a restricted Lie-Rinehart
algebra over K (see [31]).
The structure of restricted Lie Rinehart algebras emerges also in connection with
theory of characteristic classes.
Example 1.12. Let k ⊂ K be fields of characteristic p 6= 0. Then, Maakestad
considers the category LieK/k (see [22]) whose objects are restricted Lie Rinehart
sub-algebras of the restricted Lie Rinehart algebra Derk(K) over K. Moreover,
Maakestad construct a contravariant functor which associates g ∈ LieK/k to the
Grothendiek ring K0(g) of the category of g-connections. (for details see Theorem
3.2 in [22]).
Example 1.13. Let P be a Poisson algebra over k. If we denote by
C := {c ∈ P [c, −] = 0}
then we see that C is closed under the commutative and Lie bracket. Thus, C is a
Poisson subalgebra of P . We call a Poisson derivation a k-linear map D : P → P
which is at the same time a derivation with respect to commutative and Lie product.
We can easily see that, Derk(P) has the structure of Lie k-algebra. Moreover, by
Leibniz rule we can see that Dp is a derivation with respect to commutative and the
Lie product. Thus, Derk(P) has the structure of restricted Lie k-algebra. Besides,
if c ∈ C and D ∈ Derk(P) then cD ∈ Derk(P). Moreover, by Hochchild's Lemma 1
(see [14]) we have the relation:
(cD)p = cpDp + (cD)p−1(c)D
for all c ∈ C Therefore, Derk(P) is a restricted Lie-Rinehart algebra over C.
6
IOANNIS DOKAS
2. Restricted enveloping algebras and restricted modules
Let (A, L) be a Lie-Rinehart algebra. There is a notion of univeral enveloping
algebra U (A, L) of (A, L) defined by Rinehart in [28]. The universal enveloping
algebra U (A, L) is an associative A-algebra which verifies the appropriate universal
property (see [15]). We recall the definition of the enveloping associative algebra
U (A, L).
The direct sum A ⊕ L of the underlying vector spaces has the structure of k-Lie
algebra given in the Proposition 1.3. Let (U(A ⊕ L), ι) be the enveloping algebra
where ι : A⊕L → U(A⊕L) is the canonical embedding. We consider the subalgebra
U +(A ⊕ L) generated by A ⊕ L. Moreover, A ⊕ L has the structure of an A-module
via
a(a′ + X) := aa′ + aX
for all a, a′ ∈ A and X ∈ L. Then, the enveloping algebra U (A, L) is defined as the
quotient:
U (A, L) := U +(A ⊕ L)/ < ι(a)ι(a′ + X) − ι(a(a′ + X)) >
The canonical map ιA is an A-algebra homomorphism. The canonical represen-
tation A → Endk(A) given by the multiplication is faithful. Thus, by the universal
property of U (A, L) we obtain that the A-algebra homomorphism ιA : A → U (A, L)
is injective. The canonical map ιL is a Lie algebra homomorphism. Moreover, in
U (A, L) the following relations hold:
ιA(a)ιL(X) = ιL(aX), and [ιL(X), ιA(a)] = ιA(X(a))
for all a ∈ A and X ∈ L.
The enveloping algebra U (A, L) has a canonical filtration:
A = U0(A, L) ⊂ U1(A, L) ⊂ U2(A, L) · · ·
where Un(A, L) is spanned by A and the powers ιL(L)n. Therefore, we can construct
the associated graded algebra given by
gr(U (A, L)) = ⊕∞
n=0Un(A, L)/Un−1(A, L)
where we set U−1(A, L) = 0. We note that gr(U (A, L)) is a commutative A-
algebra. There is a theorem of Poincare-Birkhoff- Witt type due to Rinehart. In
particular, it is proved (see [28], Theorem 3.1) that if L is a projective A-module
and SA(L) denotes the symmetric A-algebra on L then the canonical epimorphism
θ : SA(L) → gr(U (A, L)) is an isomorphism of A-algebras. Moreover, we obtain
that ιL : L → U (A, L) is injective.
Let (A, L) be a restricted Lie-Rinehart algebra over A and suppose that L is free
as an A-module. Let {ui, i ∈ I} be an ordered A-basis of L. Let C(U (A, L)) denote
the center of U (A, L). Since L is a restricted Lie algebra we obtain: for all ui there
is a zi ∈ C(U (A, L)) such that up
i − u[p]
i = zi.
Theorem 2.1. Let (A, L) be a restricted Lie-Rinehart algebra such that L is free
as an A-module. Then the set,
B := {zh1
i1 zh2
i2 · · · zhr
ir
uk1
i1 uk2
i2 · · · ukr
ir
}
where i1 < i2 < · · · < ir, hi ≥ 0 and 0 ≤ ki < p is an A-basis of U (A, L).
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP7
Proof. It is proved in Theorem 3.1 in [28] that the standard monomials of the form
us1
i1 us2
i2 · · · usr
where i1 < i2 < · · · < ir and si ≥ 0 form an A-basis of U (A, L). Let
ir
us1
us2
· · · usr
be a standard monomial and s = s1 + · · · + sr. By induction on s
i1
i2
ir
we prove that the set B generates U (A, L). If all si are such that si < p then it is
clear. Suppose that there is sij > p then we have
us1
i1
us2
i2
· · · usr
ir
= zij us1
i1
us2
i2
· · · u
−p
sij
ij
· · · usr
ir
+ us1
i1
us2
i2
· · · u
−p
sij
ij
u[p]
ij
· · · usr
ir
We notice that the terms us1
i1
· · · usr
ir
belong in Us−1(A, L), thus by induction can be written as linear combination of
elements of B. Next we prove that the elements of B are linearly independent. Let
zh1
i1 zh2
be an element of B. Since zi = up
i2 · · · zhr
i2 · · · ukr
i we get
and us1
i1
i − u[p]
uk1
i1 uk2
· · · usr
ir
· · · u
us2
i2
us2
i2
· · · u
ir
ir
u[p]
ij
−p
sij
ij
−p
sij
ij
(2.1)
zh1
i1
zh2
i2
· · · zhr
ir
uk1
i1
uk2
i2
· · · ukr
ir
= (up
1 )h1 (up
1−u[p]
≡ uh1p+k1
2−u[p]
uh2p+k2
i2
2 )h2 · · · (up
· · · uhrp+kr
ir
i1
r−u[p]
uk2
i2
r )hr uk1
i1
mod Us−1(A, L)
· · · ukr
ir
where s = Pi=r
i=1 hip + ki and i1 < i2 < · · · < ir.
i < p it is obliged to have hi = h′
Let (hi1 , hi2, · · · , hir ), (ki1 , ki2 , · · · , kir ) and (h′
ir )
be sequences such that hip+ ki = h′
i for all i = 1, 2, · · · r. Since 0 < ki < p and
0 < k′
i for all i = 1, 2, · · · r. Moreover,
by Poincare-Birkhoff-Witt theorem we have an isomorphism of A-algebras SA(L) ≃
gr(U (A, L)). Thus, by the relation (2.1) follows that the elements of B are linearly
independent.
(cid:3)
i and ki = k′
i2 , · · · , h′
i2 , · · · , k′
ir ), (k′
ip+ k′
i1 , h′
i1 , k′
Proposition 2.2. Let (A, L) be a Lie-Rinehart algebra such that L is free as an
A-module. Let {ui, I} be an ordered A-basis of L. If there is a map ui → u[p]
such
i
that adp
for all i ∈ I then (A, L) can be equipped with the structure of
restricted Lie-Rinehart algebra with a p-map which extends the map (−)[p].
ui = adu[p]
i
Proof. Let J be the ideal of U (A, L) generated by the zi := up
. We consider
the associative algebra Up(A, L) := U (A, L)/J. By the previous theorem we get
that the elements of the form
i − u[p]
i
uk1
i1
uk2
i2
· · · ukr
ir
where i1 < i2 < · · · < ir, and 0 ≤ ki < p constitute an A-basis for Up(A, L).
The canonical A-algebra homomorphism ιA : A → U (A, L) induce an A-algebra
homomorphism iA : A → Up(A, L). Moreover, the Lie homomorphism ιL : L →
U (A, L)Lie induce a Lie algebra homomorphism iL : L → Up(A, L)Lie which is
injective. Moreover, we have
(iL(ui))p = up
i + J = u[p]
i + J
and (iL(ui))p ∈ iL(L). Besides,
(iL(aui))p = (aui)p + J
8
IOANNIS DOKAS
By Hochschild's relation Lemma 1 in [14] we get in U (A, L) the relation
(aui)p = apup
i + [aui, [aui, · · · [aui
, a], · · · ]ui
}
p−1
{z
= apup
i + (aui)p−1(a)ui
Therefore,
(iL(aui))p = apup
= apu[p]
i + (aui)p−1(a)ui + J
i + (aui)p−1(a)ui + J
and (iL(aui))p ∈ iL(L). Therefore, by relation (1.3) we get that for all x ∈ L we
have xp ∈ iL(L) and L is a restricted Lie k-subalgebra of Up(A, L)RLie. Obviously
the relation (1.5) of the definition holds and (A, L) is equipped with the structure
of a restricted Lie-Rinehart algebra.
(cid:3)
Remark 2.3. Let (A, L) be a Lie-Rinehart algebra such that L is free as an A-
module. Let {ui, I} be an ordered A-basis of L. We easily see using the relations
1.3 and 1.4, that the ideal of U (A, L) generated by the elements {X p−X [p], X ∈ L}
is equal to J.
Next, we introduce the notion of restricted enveloping algebra of a restricted
Lie-Rinehart algebra. Let (A, L, (−)[p]) be a restricted Lie-Rinehart algebra. We
define as restricted enveloping algebra Up(A, L) of a restricted Lie-Rinehart algebra
(A, L, (−)[p]), the quotient:
Up(A, L) := U (A, L)/ < X [p] − X p >
where, a ∈ A and X ∈ L.
Remark 2.4. We note that in Proposition 2.2 is proved that the map iL : L →
Up(A, L)RLie is a restricted Lie monomorphism when L is a free as an A-module.
The following proposition gives us the universal property of the restricted en-
veloping algebra.
Proposition 2.5. Let B be an A-algebra such that there is an A-algebra homo-
morphism φA : A → B and φL : L → BRLie a restricted Lie k-homomorphism. If
we have:
φA(a)φL(X) = φL(aX), and [φL(X), φA(a)] = φA(X(a))
for all a ∈ A and X ∈ L then there exists a unique homomorphism of associative
algebras Φp : Up(A, L) → B such that ΦpiA = φA and ΦpiL = φL.
Proof. We easily see that the map f : A ⊕ L → BLie given by
f (a + X) = φA(a) + φL(X),
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP9
for all a ∈ A and X ∈ L is a Lie morphism. Therefore, there is a an algebra
morphism f ′ : U +(A ⊕ L) → B. Moreover,
f ′(ι(a(a′ + X))) = f (aa′ + aX)
= φA(aa′) + φL(aX)
= φA(a)φA(a′) + φA(a)φL(X)
= f (a)f (a′ + X)
= f ′(ι(a))f ′(ι(a′ + X))
= f ′(ι(a)(ι(a′ + X))
Thus, f ′ induces an algebra morphism Φ : U (A, L) → B. Since, φL is a restricted
Lie homomorphism we have Φ(X [p]) = Φ(X p). Therefore, Φ induces an algebra
morphism Φp : Up(A, L) → B.
(cid:3)
Remark 2.6. The canonical representation A → Endk(A) given by the multiplica-
tion, is faithful. By the universal property of Up(A, L) we get that the A-algebra
homomorphism iA : A → Up(A, L) is injective.
Definition 2.7. Let (A, L, (−)[p]) be a restricted Lie-Rinehart algebra. A restricted
Lie-Rinehart module, is a Lie-Rinehart (A − L)-module M which additionally is a
restricted Lie L-module. In other words, a restricted Lie-Rinehart (A − L)-module
is a k-module M equipped with the structures of an A-module and a Lie L-module
such that:
(aX)m = a(Xm)
X(am) = aX(m) + X(a)m
X [p]m = (X(X(. . . (X
m))))
for all a ∈ A, X ∈ L and m ∈ M .
p
{z
}
Let (A, L, (−)[p]) be a restricted Lie-Rinehart algebra. The category of restricted
(A, L)-modules is equivalent to the category of Up(A, L)-modules.
Example 2.8. The notion of a restricted Lie-Rinehart module recovers in a par-
ticular case the notion of regular module defined in [6]. Namely, if K/k is a purely
inseparable extension of exponent 1 then a regular module is just a restricted Lie-
Rinehart module over the restricted Lie Rinehart algebra Derk(K).
Example 2.9. Let g ∈ LieK/k be a restricted Lie-Rinehart algebra (see 1.12 and
[22]) then a p-flat connection is a restricted Lie-Rinehart g-module.
An important example of Lie algebroid is the transformation Lie algebroid for a
differential manifold. There is an algebraic generalazation of this notion called the
transformation Lie-Rihart algebra.
Proposition 2.10. Let g ∈ RLie be a restricted Lie k-algebra and A a commutative
k-algebra. If there is a restricted Lie homomorphism δ : g → Der(A), then the
transformation Lie-Rinehart algebra (A, A ⊗k g) can be endowed with the structure
of a restricted Lie -Rinehart algebra.
10
IOANNIS DOKAS
Proof. The Lie-Rinehart algebra of transformation is a Lie k-algebra with Lie
bracket given by:
[a ⊗ g, a′ ⊗ g′] := aa′ ⊗ [g, g′] + aδ(g)(a′) ⊗ g′ − a′δ(g′)(a) ⊗ g
for all a, a′ ∈ A and g, g′ ∈ g, and with anchor map α : A ⊗k g → Der(A) given
by α(a ⊗ g)(a′) = aδ(g)(a′). Let {gi, i ∈ I} be a k-basis of g, then the elements
{1 ⊗ gi, i ∈ I} form an A-basis of A ⊗ g. Let τ1⊗gi , ρ1⊗gi : A ⊗k g → A ⊗ g be the
k-linear maps given by τ1⊗gi (a ⊗ g) := a ⊗ [g, gi] and ρ1⊗gi (a ⊗ g) := δ(gi)(a) ⊗ g
respectively. We note that:
Since char k = p we have:
τ1⊗gi ρ1⊗gi = ρ1⊗gi τ1⊗gi
adp
1⊗gi = (τ1⊗gi − ρ1⊗gi )p
1⊗gi ρp−j
1⊗gi
=
j=p
(cid:18)p
j(cid:19)τ j
Xj=0
1⊗gi − ρp
= τ p
= ad1⊗g[p]
i
1⊗gi
Therefore, by Proposition 2.2 above we get that A ⊗k g can be equipped with
the structure of a restricted Lie-Rinehart algebra and the p-map is given by:
(a ⊗ g)[p] = ap ⊗ g[p] − (aδ(g))p−1(a) ⊗ g
By equation (1.4), we see that the anchor map α is actually a restricted Lie homo-
morphism.
(cid:3)
In the next subsection we extend for the category p − LR(A) the notion of free
Lie-Rinehart algebra defined by Casas-Pirashvili in [9] and Kapranov in [20].
2.1. Free restricted Lie-Rinehart algebra. Let A be a commutative k-algebra.
We denote by Vect/Der(A) the category whose objects are k-linear morphisms
ψ : V → Der(A) where V ∈ Vect and morphisms f : ψ → ψ′ are k-linear morphisms
f : V → V ′ such that ψ′f = ψ. We denote by V : p − LR(A) → Vect/Der(A)
the forgetful functor from the category of restricted Lie-Rinehart A-algebras to the
category Vect/Der(A) which assigns a restricted Lie-Rinehart algebra L over A to
α : L → Der(A) the anchor map of L.
Proposition 2.11. There is a left adjoint functor F : Vect/Der(A) → p − LR(A)
to the functor V : p − LR(A) → Vect/Der(A)
Homp−LR(A)(F (ψ), L) ≃ HomVect/Der(A)(ψ, V(L))
Proof. Let ψ : V → Der(A). Then, by the universal property of the free re-
stricted Lie algebra Lp(V ) generated by V there is a restricted Lie homomorphism
Φ : Lp(V ) → Der(A) such that ΦiV = ψ, where iV : V → Lp(V ) denotes the
canonical map . By Proposition 2.2, we get that A ⊗ Lp(V ) is equipped with the
structure of a restricted Lie-Rinehart algebra. Therefore, we construct a functor
F : Vect/Der(A) → p − LR which assigns ψ ∈ Vect/Der(A) to A ⊗k Lp(V ).
Let f ∈ HomVect/Der(A)(ψ, V(L)). Then, we have that αf = ψ. Moreover, by
the universal property of the free restricted Lie algebra Lp(V ) generated by V , there
is a restricted Lie homomorphism φ : Lp(V ) → L such that φiV = f and Φ = αφ.
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP11
Let fp : A ⊗k Lp(V ) → L be the homomorphism of restricted Lie-Rinehart algebras
given by fp(a ⊗ x) := aφ(x), where a ∈ A and x ∈ Lp(V ). Thus, we construct a
map f 7→ fp. Conversely, for fp ∈ Homp−LR(A)(F (ψ), L) we consider the k-linear
map f : V → L given by f := fp¯i where ¯i : V → A ⊗k Lp(V ) given by v 7→ (1 ⊗ v),
and v ∈ V . We easily see that the maps f 7→ fp and fp 7→ f are inverse to each
other.
(cid:3)
3. Beck modules and Beck derivations
Beck in his desertion (see [5]) gave an answer of what should be the right notion
of coefficient module for cohomology. The notion of Beck-module encompasses for
various categories, the usual known notions of coefficient module for cohomology
(see [3]). In this section we determine the category of Beck modules and the group
of Beck derivations (see [5], [4]) for the category of restricted Lie-Rinehart algebras
p-LR(A) over a commutative algebra A.
Definition 3.1. Let L be an object in a category p-LR(A). We denote by (p-
LR(A)/L)ab the category of abelian group objects of the comma category p-LR(A)/L
and by IL : (p-LR(A)/L)ab → p-LR(A)/L the forgetful functor. An object M ∈ (p-
LR(A)/L)ab is called a Beck L-module. Let g ∈ p-LR(A)/L and M a Beck-L-
module. The group Homp−LR(A)/L(g, IL(M )) is called the group of Beck deriva-
tions of g by M .
Notation 3.2. Let M
by p ¯M the restriction of the p-map pM of M to ¯M .
µ
−→ L be a Beck L-module. We denote by ¯M := ker µ and
−→ L be Beck L-module. Then, ¯M ⋊p ¯M
Theorem 3.3. Let L ∈ p-LR(A) and M
L is defined and endowed with the structure of a restricted Lie-Rinehart algebra.
Moreover, there is an isomorphism of restricted Lie-Rinehart algebras:
µ
¯M ⋊p ¯M L ≃ M
µ
Proof. Since M
−→ L is an abelian group object in (p-LR(A)/L)ab a fortiori is
an abelian group object in (Lie/L)ab, where Lie/L denotes the slice category of
Lie algebras over L. It is well known that the category of abelian group objects
(Lie/L)ab is equivalent to the category of Lie L-modules. In particular if z : L → M
denotes the zero map for the structure of group object we have a split extension in
the category of Lie algebras:
0 → ¯M → M
µ
−−→
z←−
L → 0
Moreover, there is an isomorphism of Lie algebras ψ : ¯M ⊕ L ≃ M given by
ψ( ¯m + X) := ¯m + z(X), for all ¯m ∈ ¯M and X ∈ L. Since z is a restricted Lie
homomorphism we have:
X [pL] ¯m = [z(X [pL]), ¯m]
= [z(X)[pM ], ¯m]
= [z(X), · · · [z(X), [z(X)
, ¯m]] · · · ]
p
{z
}
12
IOANNIS DOKAS
Therefore, ¯M is endowed with the structure of restricted L-module. Besides ¯M is
abelian thus,
[z(X), ¯m[p ¯M ]] = [z(X), ¯m]..., ¯m
]..] = 0
Therefore, p ¯M : ¯M → ¯M L and ¯M ⋊p ¯M L is defined. Since ¯M is abelian
p
{z }
(ψ(X + ¯m))pM = z(X)pM + ¯mpM +
i=p−1
Xi=1
si(z(X), ¯m)
= z(X)pM + ¯mpM + adp−1
z(X)( ¯m)
Therefore, the isomorphism ψ is an isomorphism of a restricted Lie algebras
ψ : ¯M ⋊p ¯M L ≃ M
The k-module ¯M ⋊p ¯M L has the structure of an A-module given by the formula
a( ¯m + X) := a ¯m + aX, a ∈ A
and ψ is an A-module homomorphism. We easily see that, ¯M ⋊p ¯M L is endowed with
the structure of a Lie-Rinehart algebra with anchor map α : ¯M ⋊p ¯M L → Derk(A)
given by
In this way, ψ becomes a Lie-Rinehart isomorphism. Besides, we have
α( ¯m + X)(a) := X(a) = z(X)(a)
ψ((a(X + ¯m))[p]) = (ψ(a(X + ¯m)))[p]
= (a(ψ(X + ¯m)))[p]
= ap(ψ(X + ¯m))[p] + (aψ(X + ¯m))p−1(a)ψ(X + ¯m)
= ap(ψ(X + ¯m)[p])) + (ψ(a(X + ¯m))p−1(a)(ψ(X + ¯m)))
= ψ(ap(X + ¯m)[p]) + (a(z(X) + ¯m))p−1(a)ψ(X + ¯m))
= ψ(ap(X + ¯m)[p]) + ψ((a(z(X) + ¯m))p−1(a)(X + ¯m))
= ψ(ap(X + ¯m)[p] + (a(X + ¯m))p−1(a)(X + ¯m))
Therefore, M ⋊p ¯M L is a restricted Lie-Rinehart algebra and ψ : ¯M ⋊p ¯M L ≃ M is
a restricted Lie-Rinehart isomorphism.
(cid:3)
Lemma 3.4. Let L be a Lie-Rinehart algebra over A. Then, the following relation
holds
(aX)p−1(ab) = apX p−1(b) + (aX)p−1(a)b
for all X ∈ p-LR(A) and a, b ∈ A.
Proof. We consider the restricted enveloping algebra U (A, L) of the Lie-Rinehart
algebra L over A. By Hochschild's Lemma 1, we get in U (A, L) that:
(a(b + iL(X)))p = ap(b + iL(X))p + (aX))p−1(a)(b + iL(X))
= ap(bp + iL(X)p + X p−1(b)) + (aX)p−1(a)b + (aX)p−1(a)iL(X)
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP13
Besides,
(a(b + iL(X)))p = (ab + aiL(X))p
= ((ab)p + aiL(X))p + (aX)p−1(ab)
= apbp + apiL(X)p + (aX)p−1(a)iL(X) + (aX)p−1(ab)
Therefore, we get:
(aX)p−1(ab) = apX p−1(b) + (aX)p−1(a)b
for all X ∈ L and a, b ∈ A.
(cid:3)
Let L be a Lie-Rinehart algebra over A. Let M be a Lie-Rinehart (A − L)-
module. We consider the commutative algebra semi-direct product A ⊕ M with
product given by:
(a ⊕ m)(a′ ⊕ m′) := aa′ ⊕ (am′ + a′m)
There is an anchor map α : L → Der(A ⊕ M ) given by
α(X)(a ⊕ m) := X(a) ⊕ X(m)
for all X ∈ L, a ∈ A and m ∈ M . Moreover, L becomes an A ⊕ M -module via
the action (a ⊕ m)X := aX. Then, we easily see that, L becomes a Lie-Rinehart
algebra over A ⊕ M . Therefore, by Lemma 3.4 above we get:
(3.1)
(aX)p−1(am) = apX p−1(m) + (aX)p−1(a)m
where X ∈ L and a ∈ A, m ∈ M .
Proposition 3.5. Let L be a restricted Lie-Rinehart algebra over A and ¯M a
restricted Lie-Rinehart module. Then, ¯M ⋊ L is endowed with the structure of
restricted Lie-Rinehart algebra.
Proof. We observe that ¯M ⋊ L is an A-module via the action a( ¯m + X) := a ¯m + aX
where a ∈ A, ¯m ∈ ¯M and X ∈ L. Moreover, there is anchor map α : ¯M ⋊ L →
Der(A) given by α(( ¯m + X))(a) := X(a). We easily see that ¯M ⋊ L is endowed
with the structure of Lie-Rinehart algebra over A. Besides, by the relation (3.1)
above we get
(a( ¯m + X))[p] = (aX)[p] + (aX)p−1(a ¯m)
= apX [p] + (aX)p−1(a)X + apX p−1( ¯m) + (aX)p−1(a) ¯m
= ap( ¯m + X)[p] + (a( ¯m + X))p−1(a)( ¯m + X)
Thus, ¯M ⋊ L is a restricted Lie-Rinehart algebra.
(cid:3)
Let A[P ] be the polynomial ring given by
A[P ] := {
i=n
Xi=0
aiP i, : P a = apP for all ai, a ∈ A}
We consider the ring W (A, L) which as an A-module is given by
W (A, L) := A[P ] ⊗A Up(A, L)
14
IOANNIS DOKAS
and such that A[P ] → W (A, L) and Up(A, L) → W (A, L) are A-algebra homomor-
phisms and the multiplication is such that
(3.2)
(3.3)
(P ⊗ 1)(1 ⊗ iL(X)) : = P ⊗ iL(X)
(1 ⊗ iL(X))(P ⊗ 1) : = 0
Proposition 3.6. Let L be a restricted Lie-Rinehart algebra over A. Then, the
category of Beck L-modules is equivalent to the category of W (A, L)-modules.
Proof. Let M be a W (A, L)-module. Using the homomorphism Up(A, L) → W (A, L)
we can see M as a Up(A, L)-module which we denote by ¯M . Besides, the A[P ] action
endows ¯M with a p-semi-linear map p ¯M : ¯M → ¯M L given by
p ¯M ( ¯m) := P ¯m, ¯m ∈ ¯M
From Proposition 3.5 and Remark 1.4 we see that ¯M ⋊p ¯M L is a restricted Lie-
Rinehart algebra. Thus, by Theorem 3.3, any W (A, L)-module M is associated to
the Beck L-module ¯M ⋊p ¯M L. Conversely, let M be a Beck L-module, then by
Theorem 3.3 we have that M ≃ ¯M ⋊p ¯M L. Moreover, we observe that ¯M is a
W (A, L)-module. Thus, we have an equivalence of categories.
(cid:3)
3.1. Beck derivations. Let g be a restricted Lie-Rinehart algebra and M =
¯M ⋊p ¯M L be a Beck g-module. Then, ¯M is a Lie-Rinehart g-module. We de-
note by Der(g, ¯M ) the group of Lie-Rinehart derivations. We recall that a k-Lie
algebra derivation d : g → ¯M is called Lie Rinehart if d is A-linear. The group of
Beck derivations is defined as follows:
Derp(g, M ) := {d ∈ DerA(g, ¯M ) : d(X [p]) = X · · · X
d(X) + (d(X))[p ¯M ], X ∈ g}
p−1
{z }
We note that Derp(g, M ) is a group under the addition since p ¯M is a p-semi-linear
map.
Proposition 3.7. Let L be a restricted Lie-Rinehart algebra over A and g ∈ p-
LR(A)/L. If M is a Beck L-module, then we have the following isomorphism
Homp−LR(A)/L(g, ¯M ⋊p ¯M L) ≃ Derp(g, M )
Proof. Let f : g → ¯M ⋊p ¯M L and π : ¯M ⋊p ¯M L → ¯M be the canonical projection.
We easily see that, df := πf is a Beck derivation and therefore, is defined a map
Φ : f 7→ df . Moreover, let d : g → ¯M be a Beck derivation we consider the map
fd : g → ¯M ⋊p ¯M L given by fd(X) := d(X) + γ(X), where X ∈ g and γ : g → L
is the structural map. The maps Ψ : d 7→ fd and Φ : f 7→ df are inverse to each
other.
(cid:3)
4. Quillen-Barr-Beck cohomology for restricted Lie-Rinehart
algebras
There is a general theory of cohomology for univeral algebras due to Quillen-Barr-
Beck (see [25], [3], [4]). Moreover, Quillen in [26] proves that the cohomology theory
for universal algebras defined in [25], is a special case of the general definition of
sheave cohomology due to Grothendieck. Following the general scheme of Quillen-
Barr-Beck cohomology theory we define cohomology groups for the category of
restricted Lie-Rinehart algebras. By Proposition 2.8 there is a functor
F : Vect/Der(A) → p − LR(A)
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP15
left adjoint to the functor
V : p − LR(A) → Vect/Der(A)
The adjoint pair (F, V) induce a cotriple G = (G∗, ǫ, δ) such that G∗ := F V.
Thus, for all L ∈ p-LR(A) we obtain a simplicial resolution G∗L → L known as
cotriple resolution or Godement resolution (see [10], [4]).
Definition 4.1. Let L be a restricted Lie-Rinehart algebra and M a Beck L-
module. Then, for n ∈ N∗, Quillen-Barr-Beck cohomology groups are defined by
the following formula
H n
p−LR(L, M ) := H n(Homp−LR(A)/L(G∗L, M ))
where in the right hand side of the formula H denotes the cohomology of a cosim-
plicial object.
Remark 4.2. We observe that in degree 0 we have H 0
p−LR(L, M ) = Derp(L, M ).
4.1. Cohomology in degree 1 and extensions. A principal bundle gives rise
to Atiyah sequence introduced by Atiyah in [2] (see for details [16]). The algebraic
analogue of "Atiyah sequence" is an extension of Lie-Rinehart algebras.
In this
subsection we consider extensions of restricted Lie-Rinehart algebras. We prove that
the Quillen-Barr-Beck cohomology in degree one classifies restricted Lie-Rinehart
extensions.
Definition 4.3. An extension (e) of restricted Lie-Rinehart algebras (of L by M )
is a short exact sequence of restricted Lie-Rinehart algebras
such that [M, M ] = 0.
0 → M → E → L → 0
(e)
Remark 4.4. Let L be a restricted Lie-Rinehart algebra over A and (e) an extension
of L by M . Since M is abelian a section s : L → E defines a restricted Lie L-action
on M . Moreover, we easily see that M becomes a Up(A, L)-module. Besides, there
is an A[P ] action on M such that P m := m[pM ], m ∈ M . Thus, we obtain that M
is endowed with the structure of a W (A, L)-module.
Two restricted Lie-Rinehart extensions (e), (e′) of L by M are called equivalent
if there is a restricted Lie-Rinehart isomorphism f : E → E ′ such that the following
diagram commutes
0 −−−−→ M −−−−→ E −−−−→ L −−−−→ 0
(cid:13)(cid:13)(cid:13)
fy
(cid:13)(cid:13)(cid:13)
0 −−−−→ M −−−−→ E ′ −−−−→ L −−−−→ 0
We denote the set of equivalent classes by Extp(L, M ).
Baer sum of restricted Lie-Rinehart algebras. Let (e), (e′) be two Lie
algebra extensions of L by M
and
0 → M → E
f
−→ L → 0
0 → M → E ′ f ′
−→ L → 0
16
IOANNIS DOKAS
Let E ×L E ′ = {(e, e′), : f (e) = f ′(e′)} be the pullback in Lie. We denote by I the
ideal
I :=< {(m, 0) − (0, m′), m, m′ ∈ M } >
and we consider the Lie algebra
Y := E ×L E ′/I
The Baer sum of (e) and (e′) is the extension of Lie algebras of L by M
0 → M ι−→ Y
ψ
−→ L → 0
where ι(m) := (m, 0) and the last one ψ(cid:0)(e, e′)(cid:1) := f (e) = f (e′). The set ExtLie(L, M )
of equivalence classes of extensions of Lie algebras of L by M endowed with the
operation of Baer sum is a group. Moreover, let (e), (e′) be restricted Lie-Rinehart
extensions of L by M . Then, E ×L E ′ is a restricted Lie algebra with p-map given
by (e, e′)[p] = (e[p], e′[p]). Let a ∈ A and (X, Y ) ∈ E ×L E ′. Then, there is an
action of A on E ×L E ′ given by a(X, Y ) := (aX, aY ). Besides, there is an action
E ×L E ′ → Der(A) given by (X, Y )(a) := X(a) = Y (a). The above actions endow
E ×L E ′ with the structure of a restricted Lie-Rinehart algebra. Moreover, the ideal
I is a restricted Lie-Rinehart ideal. Thus, Y is a restricted Lie-Rinehart algebra.
The Baer sum endows Extp(L, M ), with the structure of a group and Extp(L, M )
is a subgroup of ExtLie(L, M ).
Theorem 4.5. Let L be a restricted Lie-Rinehart algebra over A and M a Beck
L-module. There is a bijection
(4.1)
H 1
p−LR(L, M ) ≃ Extp(L, M )
Proof. Duskin in [7] develops the theory of torsors and gives an interpretation
of cotriple cohomology. There is a bijection between the first cohomology group
H 1
G(L, M ) and the set T orsp−LR(L, M ) of the isomorphism classes of objects E ∈ p-
LR(A)/L which are torsors for the abelian group object M . An object E ∈ p-
LR(A)/L is M torsor, if E → L is an epimorphism and there is a restricted Lie-
Rinehart morphism
such that the map
ω : ( ¯M ⋊p ¯M L) ×L E → E
(ω, π) : ( ¯M ⋊p ¯M L) ×L E → E ×L E
where π denotes projection, is a restricted Lie-Rinehart isomorphism.
Let E ∈ Extp(L, M ), then we have a short exact sequence in p-LR(A)
0 → M → E
f
−→ L → 0
such that [M, M ] = 0 and the induced W (A, L)-action on M , recovers the given
W (A, L)-action on M . We define a map
ω : ( ¯M ⋊p ¯M L) ×L E → E
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP17
given by ω(m + f (e), e) := m + e, for all m ∈ ¯M and e ∈ E. We easily see that, ω
is a Lie-Rinehart homomorphism. Moreover, we have
ω((m + f (e), e)[p]) = ω((m + f (e))[p], e[p])
ω(f (e)p−1m + m[p ¯M ] + (f (e))[p], e[p])
= ω(ep−1m + m[p ¯M ] + f (e[p]), e[p])
= ep−1m + m[p ¯M ] + e[p]
= (m + e)[p]
= (ω(m + f (e), e))[p]
Thus, we get that ω is a restricted Lie-Rinehart homomorphism. It is easy to check
that
(ω, π) : ( ¯M ⋊p ¯M L) ×L E → E ×L E
is a restricted Lie-Rinehart isomorphism. Therefore, E → L is torsor for M .
Conversely, let E ∈ p-LR(A)/L be an object torsor for ¯M ⋊p ¯M L → L. Then, the
structural map f : E → L, is a restricted Lie-Rinehart epimorphism. If K := ker f ,
then there is an injection i : K ֒→ E ×L E with i(k) := (k, 0). Besides, there is an
injection j : ¯M ֒→ ( ¯M ⋊p ¯M L) ×L E given by j( ¯m) := (( ¯m, 0), 0). The restriction
of the restricted Lie-Rinehaert isomorphism (ω, π) on ¯M implies an isomorphism
of restricted Lie-Rinehart algebras (A, ¯M , (−)[p ¯M ]) ≃ (A, K, (−)[pk]). Moreover, let
X ∈ L, m ∈ ¯M and e ∈ E such that f (e) = X. Then, we have
Besides, we have
ω(Xm, 0) = ω([(X, e), (m, 0)])
= [ω(X, e), ω(m, 0)]
(ω, π)((X, e), e) = (ω(X, e), e)
it follows that f (ω(X, e)) = X. Therefore, ¯M and K are isomorphic as restricted
Lie L-modules. Since ω is a restricted Lie-Rinehart homomorphism, it follows
that, ¯M and K are isomorphic as Up(A, L)-modules. Thus, we get an extension of
restricted Lie-Rinehart algebras
0 → M → E → L → 0
Therefore, there is a bijection of sets H 1
p−LR(L, M ) ≃ Extp(L, M ).
(cid:3)
Remark 4.6. We note that, in terms of Quillen-Barr-Beck cohomology, there is a
shift in the dimension, in compare to the classical notation, concerning classification
theorems of extensions.
If L is a Lie algebra and M a U(L)-module, then cotriple cohomology groups
Lie(L, M ) := H ∗(HomLie(GLie(L), M )) are defined (see [3]). The cotriple which
H ∗
is used is given by GLie := LF where L : V ect → Lie is the free Lie algebra functor
left adjoint to the forgetful functor F : Lie → V ect.
Corollary 4.7. There is an isomorphism of groups
H 1
p−LR(L, M ) ≃ Extp(L, M )
18
IOANNIS DOKAS
Proof. Duskin in [7],[8] and Glen in [11], develop the theory of torsors. In particular,
is proved (see Section 4 in [8] and [11]) that the set of torsors can be endowed with
the structure of a group. It follows from the general theory (see Section 5 in [8]
and [11]) that, there is a group isomorphism
H 1
p−LR(L, M ) ≃ T orsp−LR(L, M )
and respectively for the category Lie algebras
H 1
Lie(L, M ) ≃ T orsLie(L, M )
Besides, the category of Lie algebras is a category of interest in sense of Orzech
(see [24]). Therefore by a general result of Vale in [32], we obtain a group isomor-
phism ExtLie(L, M ) ≃ T orsLie(L, M ) . If L ∈ p − LR(A) and M a W (L)-module,
then there is a natural embedding H 1
p−LR(L, M ) ֒→ H 1
Lie(L, M ).
Thus, we have the following commutative diagram
p−LR(L, M )
/ T orsp−LR(L, M )
/ Extp(L, M )
H 1
∽ /
H 1
Lie(L, M )
Therefore, the bijection (4.1)
/ T orsLie(L, M )
∽ /
∽ /
/ ExtLie(L, M )
H 1
p−LR(L, M ) ≃ Extp(L, M )
is an isomorphism of groups.
(cid:3)
5. Brauer group and cohomology
The problem of classification of finite-dimensional central simple algebras is re-
lated to the notion of Brauer group. The theory of Brauer groups has strong
ties with number theory, algebraic geometry (see [30], Local fields) and algebraic
k-theory [19]. In connection with Galois theory we have that if E/F is a Galois
extension then the relative Brauer group BE
F is isomorphic with the group of equiva-
lence classes of group extensions of the Galois group Gal(E/F ) by the multiplicative
group F ∗ of F . Moreover, we have an isomorphism of groups relating the Galois
cohomology and the Brauer group:
(5.1)
BE
F ≃ H 2(Gal(E/F ), F ∗)
In the context of purely inseparable extension of exponent 1 there is a Galois
theory due to Jacobson (see [18]). The role of Galois group is now played by the
group of derivations. In particular, let K/k be a finite purely inseparable extension
of exponent 1 and A a finite dimensional algebra with center k which contains
K as a maximal commutative subring. Let S := {s ∈ A, Ds(K) ⊂ K}, where
Ds denotes the derivation Ds(a) := sa − as. Then, Hochschild in [14] considers
particular class of restricted Lie algebra extensions of Derk(K) by K
0 → K → S → Derk(K) → 0
called regular extensions. A regular extension is nothing more than a restricted
Lie-Rinehart extension of the restricted Lie-Rinehart algebra Derk(K) by K. We
note that K is an abelian restricted Lie algebra over k with its natural p-map (see
[14] for details). Moreover, Hochschild in [14] proves that there is an isomorphism
between the Brauer group BK
k and the set of equivalence classes of regular restricted
Lie algebra extensions of Derk(K) by K.
_
_
/
_
COHOMOLOGY OF RESTRICTED LIE-RINEHART ALGEBRAS AND THE BRAUER GROUP19
As motivation to carry out this research was to establish an isomorphism in terms
of cohomology of restricted Lie-Rinehart algebras analogue to the isomorphism
(5.1) in terms of Galois cohomology. Hochschild in [13] defined cohomology groups
Hoch := H ∗(L, M ) where L ∈ RLie and M a restricted Lie module. Since the
H ∗
second cohomology group H 2
Hoch(L, M ) classifies extensions where M is strongly
abelian, i.e such that M [p] = 0 we do not have an isomorphism of the Brauer group
BK
k with a sub group of H 2
Hoch(Derk(K), K). Nevertheless, using the Quillen-
Barr-Beck cohomology for p-LR(A) we can establish the analogue isomorphism to
(5.1).
Theorem 5.1. Let K/k be a finite purely inseparable extension of exponent 1.
Then, we have the following isomorphism of groups
BK
k ≃ H 1
p−LR(Derk(K), K)
Proof. The proof follows from Corollary (4.7) and the isomorphism proved by
Hochschild in [14] between the Brauer group BK
k and the group of equivalence
classes of regular extensions of Derk(K) by K.
(cid:3)
Remark 5.2. Given a commutative ring C and a commutative C-algebra A, Amut-
sur in [1] defined cohomology groups H ∗
Am(A). Let K be a purely inseparable
extension field of k of exponent 1. Then, Rosenberg and Zelinsky in Section 4 in
[29] exhibit an isomorphism between H 2
k . There-
fore, by the above theorem we get an isomorphism
Am(K) and the Brauer group BK
H 2
Am(K) ≃ H 1
p−LR(Derk(K), K)
References
[1] S. A. Amitsur, Simple algebras and cohomology groups of arbitrary fields, Trans. Amer. Math.
Soc. 90, (1959), 73-112.
[2] M. F. Atiyah, Complex analytic connections in fibre bundles, Trans. Amer. Math. Soc. 85,
(1957), 181-207.
[3] M. Barr, Cartan-Eilenberg cohomology and triples. J. Pure Appl, Algebra 112 (1996), no. 3,
219-238.
[4] M. Barr ; J. Beck, Acyclic models and triples, 1966 Proc. Conf. Categorical Algebra (La Jolla,
Calif., (1965), 336343, Springer, New York.
[5] J. Beck, Triples, algebras and cohomology, Ph.D. Thesis, Columbia University, (1967).
[6] A. J. Berkson, On Amitsur's complex and restricted Lie algebras, Trans. Amer. Math. Soc.
109, (1963), 430-443.
[7] J.W. Duskin, K(π, n)K(,n)-torsors and the interpretation of triple cohomology, Proc. Nat.
Acad. Sci. U.S.A. 71 (1974), 2554-2557.
[8] J.W. Duskin, Higher-dimensional torsors and the cohomology of topoi: the abelian theory,
Applications of sheaves Proc. Res. Sympos. Appl. Sheaf Theory to Logic, Algebra and Anal.,
Univ. Durham, Durham, 1977), 255-279, Lecture Notes in Math., 753, Springer, Berlin,
(1979).
[9] J.M. Casas; M. Ladra; T. Pirashvili, Triple cohomology of Lie-Rinehart algebras and the
canonical class of associative algebras, J. Algebra 291 (2005), no. 1, 144-163.
[10] R. Godement, Topologie Alg`ebrique et Th`eorie des Faisceaux, Herman, Paris, (1958).
[11] P. Glenn, Realization of cohomology classes in arbitrary exact categories, J. Pure Appl.
Algebra 25, (1982), no. 1, 33-105.
[12] J.C. Herz, Pseudo-algebres de Lie, C. R. Acad. Sci. Paris 236, (1953), 1935-1937.
[13] G. Hochschild, Cohomology of restricted Lie algebras. Amer. J. Math, 76, (1954), 555-580.
[14] G. Hochschild, Simple algebras with purely inseparable splitting fields of exponent 1, Trans.
Amer. Math. Soc. 79, (1955), 477-489.
[15] J. Huebschmann, Poisson cohomology and quantization, J. Reine Angew. Math, 408 (1990),
57-113.
20
IOANNIS DOKAS
[16] J. Huebschmann, Extensions of Lie-Rinehart algebras and the Chern-Weil construction,
Higher homotopy structures in topology and mathematical physics (Poughkeepsie, NY, 1996),
145-176, Contemp. Math., 227, Amer. Math. Soc., Providence, RI, (1999).
[17] N. Jacobson, Restricted Lie algebras of characteristic p, Trans. Amer. Math. Soc. 50, (1941),
15-25.
[18] N. Jacobson, An extension of Galois theory to non-normal and non-separable fields, Amer.
J. Math. 66, (1944), 1-29.
[19] I. Kersten, Brauergruppen von Korpern, Aspects of Mathematics, D6. Friedr. Vieweg and
Sohn, Braunschweig, (1990).
[20] M. Kapranov, Free Lie Algebroids and Space of Paths, (English summary), Selecta Math.
(N.S.) 13 (2007), no. 2, 277-319.
[21] J.-L. Loday; B. Vallette, Algebraic Operads, http://www-irma.u-strasbg.fr/∼loday/PAPERS/LodayVallette.pdf
[22] H. Maakestad, Chern character for Lie-Rinehart algebras, Ann. Inst. Fourier (Grenoble) 55
(2005), no. 7, 2551-2574
[23] K. Mackenzie, Lie groupoids and Lie algebroids in differential geometry, London Mathemat-
ical Society Lecture Note Series, 124. Cambridge University Press, Cambridge, 1987.
[24] G. Orzech, Obstruction theory in algebraic categories I, II, J. Pure Appl. Algebra 2 (1972),
287-314, 315-340.
[25] D. Quillen, On the (co-) homology of commutative rings, (1970) Applications of Categorical
Algebra (Proc. Sympos. Pure Math., Vol. XVII, New York, (1968), 65-87. Amer. Math. Soc.,
Providence, R.I.
[26] D. Quillen, Homotopical algebra, Lecture Notes in Mathematics, No. 43 Springer-Verlag,
Berlin-New York (1967).
[27] R. S. Palais, The cohomology of Lie rings, (1961) Proc. Sympos. Pure Math., Vol. III, 130-137
American Mathematical Society, Providence.
[28] G. S. Rinehart, Differential forms on general commutative algebras, Trans. Amer. Math. Soc.
108, (1963), 195-222.
[29] A. Rosenberg ; D. Zelinsky, On Amitsur's complex, Trans. Amer. Math, Soc. 97, (1960),
327-356.
[30] J. P. Serre, Local fields, Translated from the French by Marvin Jay Greenberg, Graduate
Texts in Mathematics, 67, Springer-Verlag, New York-Berlin, (1979).
[31] J.P. Serre, Quelques proprietes des varietes abeliennes en caracteristique p, (French) Amer.
J. Math. 80, (1958), 715-739.
[32] M. Vale, Torsors and special extensions, Cahiers Topologie Geom. Differentielle Categ. 26,
(1985), no. 1, 63-90.
E-mail address: [email protected]
|
1605.00803 | 1 | 1605 | 2016-05-03T09:24:50 | B-lattice of nil-extensions of rectangular skew-rings | [
"math.RA"
] | Every quasi completely regular semiring is a b-lattice of completely Archimedean semirings, i.e., a b-lattice of nil-extensions of completely simple semirings. In this paper we consider the semiring which is a b-lattice of nil-extensions of orthodox completely simple semirings. | math.RA | math |
b-lattice of nil-extensions of rectangular
skew-rings
S. K. Maity and R. Chatterjee
Department of Mathematics, University of Burdwan,
Golapbag, Burdwan-713104, India.
e-mail: [email protected]
Abstract
Every quasi completely regular semiring is a b-lattice of completely Archimedean semir-
ings, i.e., a b-lattice of nil-extensions of completely simple semirings. In this paper we
consider the semiring which is a b-lattice of nil-extensions of orthodox completely simple
semirings.
AMS Mathematics Subject Classification (2010): 16A78, 20M10, 20M07.
Keywords : Completely simple semiring, completely Archimedean semiring, rectan-
gular skew-ring, left skew-ring, orthodox, b-lattice, nil-extension.
1
Introduction
Semiring is one of the many concepts of universal algebra which is an established and
recognized area of study. The structure of semirings has been studied by many authors,
for example, by F. Pastijn, Y. Q. Guo, M. K. Sen, K. P. Shum and others ([7], [9]). In
[4] the authors studied properties of quasi completely regular semirings. They proved
that a semiring is a quasi completely regular semiring if and only if it is a b-lattice
of completely Archimedean semirings.
In [5], the authors proved that a semiring is
a completely Archimedean semiring if and only if it is nil-extension of a completely
1
simple semiring. In this paper we intend to study the semirings which are b-lattice of
orthodox completely Archimedean semirings. The preliminaries and prerequisites we
need for this article are discussed in section 2. In section 3, we discuss our main result.
2 Preliminaries
A semiring (S, +, ·) is a type (2, 2)-algebra whose semigroup reducts (S, +) and (S, ·)
are connected by ring like distributivity, that is, a(b+c) = ab+ac and (b+c)a = ba+ca
for all a, b, c ∈ S. An element a in a semiring (S, +, ·) is said to be additively regular
if there exists an element x ∈ S such that a = a + x + a. An element a of a semiring
(S, +, ·) is called completely regular [9] if there exists an element x ∈ S such that
a = a + x + a, a + x = x + a and a(a + x) = a + x. We call a semiring (S, +, ·)
completely regular if every element of S is completely regular. A semiring (S, +, ·) is
called a skew-ring if its additive reduct (S, +) is an arbitrary group. A semiring (S, +, ·)
is called a b-lattice if (S, ·) is a band and (S, +) is a semilattice. If both the reducts
(S, +) and (S, ·) are bands, then the semiring (S, +, ·) is said to be an idempotent
semiring. An element a in a semiring (S, +, ·) is said to be additively quasi regular if
there exists a positive integer n such that na is additively regular. An element a in
a semiring (S, +, ·) is said to be quasi completely regular [4] if there exists a positive
integer n such that na is completely regular. Naturally, a semiring (S, +, ·) is said to
be quasi completely regular if all the elements of S are quasi completely regular.
A semigroup S is a rectangular band if for every a, x ∈ S, a = axa. An element
x of a semigroup S is a left zero of S if xa = x for all a ∈ S. A semigroup all of
whose elements are left zeros is said to be a left zero semigroup. A rectangular group is
isomorphic to a direct product of a rectangular band and a group. Again a left group is
isomorphic to a direct product of a left zero semigroup and a group. A semigroup S in
which the idempotents form a subsemigroup is orthodox. A completely simple orthodox
semigroup is a rectangular group, and conversely. We naturally aim at extending the
concepts of rectangular group and left group to that of semirings.
Throughout this paper, we always let E +(S) and Reg+S respectively be the set of
all additive idempotents and the set of all additively regular elements of the semiring
2
S. Also we denote the set of all additive inverse elements of an additively regular
element a in a semiring (S, +, ·) by V +(a). As usual, we denote the Green's relations
on the semiring (S, +, ·) by L , R, D, J and H and correspondingly, the L -relation,
R-relation, D-relation, J -relation and H -relation on (S, +) are denoted by L +, R +,
D +, J + and H +, respectively. In fact, the relations L +, R +, D +, J + and H + are
all congruence relations on the multiplicative reduct (S, ·). Thus if any one of these
happens to be a congruence on (S, +), it will be a congruence on the semiring (S, +, ·).
We further denote the Green's relations on a quasi completely regular semigroup as
L ∗,R ∗, D ∗,J ∗ and H ∗. For other notations and terminologies not given in this
paper, the reader is referred to the texts of Howie [3] & Golan [2]. Now we state some
important definitions and results on quasi completely regular semirings and completely
simple semirings.
Definition 2.1. [9] A completely regular semiring (S, +, ·) is called a completely simple
semiring if J + = S × S.
Theorem 2.2. [10] Let R be a skew-ring, (I, ·) and (Λ, ·) are bands such that I ∩ Λ =
{o}. Let P = (pλ,i) be a matrix over R, i ∈ I, λ ∈ Λ and assume
1. pλ,o = po,i = 0,
2. pλµ,kj = pλµ,ij − pνµ,ij + pνµ,kj,
3. pµλ,jk = pµλ,ji − pµν,ji + pµν,jk,
4. apλ,i = pλ,ia = 0,
5. ab + poµ,io = poµ,io + ab,
6. ab + pλo,oj = pλo,oj + ab,
for all i, j, k ∈ I, λ, µ, ν ∈ Λ and a, b ∈ R.
Let M consist of all the elements of I × R × Λ and define operations '+' and '·' on M
by
and
(i, a, λ) + (j, b, µ) = (i, a + pλ,j + b, µ)
(i, a, λ) · (j, b, µ) = (ij, −pλµ,ij + ab, λµ).
Then (M , +, ·) is a completely simple semiring. Conversely, every completely simple
semiring is isomorphic to such a semiring.
The semiring constructed in the above theorem is denoted by M (I, R, Λ; P ) and is
called a Rees matrix semiring.
3
Corollary 2.3.
[10] Let M (I, R, Λ; P ) be a Rees matrix semiring. Then pλµ,ij =
pλo,oj + poµ,io holds for all i, j ∈ I; λ, µ ∈ Λ. This yields pλ,i = pλo,oi + poλ,io and hence
by assumption (5) and (6) stated in Theorem 2.2, ab + pλ,i = pλ,i + ab for all i ∈ I;
λ ∈ Λ and a, b ∈ R.
Definition 2.4. Let (S, +, ·) be an additively quasi regular semiring. We consider the
relations L ∗
+, R ∗
+, J ∗
+, H ∗
+ and D ∗
+ on S defined by: for a, b ∈ S;
aL ∗
aR ∗
+b if and only if maL +nb,
+b if and only if maR +nb,
+b if and only if maJ +nb,
+ ∩ R ∗
+ and D ∗
+ = L ∗
aJ ∗
+ = L ∗
H ∗
+o R ∗
+
where m and n are the smallest positive integers such that ma and nb are additively
regular.
Definition 2.5. [4] A quasi completely regular semiring (S, +, ·) is called completely
Archimedean if J ∗
+ = S × S.
Definition 2.6. [4] Let R be subskew-ring of a semiring S. If for every a ∈ S there
exists a positive integer n such that na ∈ R, then S is said to be a quasi skew-ring.
Definition 2.7. A congruence ρ on a semiring S is called a b-lattice congruence (idem-
potent semiring congruence) if S/ρ is a b-lattice (respectively, an idempotent semiring).
A semiring S is called a b-lattice (idempotent semiring) Y of semirings Sα(α ∈ Y ) if S
admits a b-lattice congruence (respectively, an idempotent semiring congruence) ρ on
S such that Y = S/ρ and each Sα is a ρ-class.
Theorem 2.8.
[4] The following conditions on a semiring (S, +, ·) are equivalent.
1. S is a quasi completely regular semiring.
2. Every H ∗
+-class is a quasi skew-ring.
3. S is (disjoint) union of quasi skew-rings.
4. S is a b-lattice of completely Archimedean semirings.
5. S is an idempotent semiring of quasi skew-rings.
4
Definition 2.9. [5] Let (S, +, ·) be a semiring. A nonempty subset I of S is said to
be a bi-ideal of S if a ∈ I and x ∈ S imply that a + x, x + a, ax, xa ∈ I.
Definition 2.10. [5] Let I be a bi-ideal of a semiring S. We define a relation ρI on
S by aρI b if and only if either a, b ∈ I or a = b where a, b ∈ S. It is easy to verify that
ρI is a congruence on S. This congruence is said to be Rees congruence on S and the
quotient semiring S/ρI contains a zero, namely I. This quotient semiring S/ρI is said
to be the Rees quotient semiring and is denoted by S/I. In this case the semiring S is
said to be an ideal extension or simply an extension of I by the semiring S/I. An ideal
extension S of a semiring I is said to be a nil-extension of I if for any a ∈ S there
exists a positive integer n such that na ∈ I.
Theorem 2.11. [5] A semiring S is a quasi skew-ring if and only if S is a nil-extension
of a skew-ring.
Theorem 2.12.
[5] The following conditions on a semiring are equivalent:
1. S is a completely Archimedean semiring.
2. S is a nil-extension of a completely simple semiring.
3. S is Archimedean and quasi completely regular.
3 Main Results
In this section we characterize b-lattice of nil-extensions of rectangular skew-rings.
Definition 3.1. An idempotent semiring (S, +, ·) is said to be a rectangular band
semiring if (S, +) is a rectangular band.
Definition 3.2. A semiring (S, +, ·) is said to be a rectangular skew-ring if it is a
completely simple semiring and (S, +) is an orthodox semigroup, i.e., E +(S) is a sub-
semigroup of the semigroup (S, +) .
Theorem 3.3. A semiring is a rectangular skew-ring if and only if it is isomorphic to
direct product of a rectangular band semiring and a skew-ring.
5
Proof. Let (S, +, ·) be a rectangular skew-ring. Then by definition (S, +, ·) is a com-
pletely simple semiring and E +(S) is a subsemigroup of the semigroup (S, +). Let
S = M (I, R, Λ; P ). As E +(S) is a subsemigroup of the semigroup (S, +) then we have
(i, −pλ,i, λ) + (j, −pµ,j, µ) = (i, −pµ,i, µ). This implies −pλ,i + pλ,j − pµ,j = −pµ,i, i.e.,
pλ,i − pµ,i + pµ,j = pλ,j. Putting j = o, we get pλ,i = pµ,i. Similarly, putting i = o we
get pλ,j = pµ,j.
Obviously, E +(S) is a rectangular band semiring.
Now we define a mapping φ : S → E +(S) × R by
φ(i, a, λ) = (cid:16)(i, −pλ,i, λ), pλ,i + a(cid:17) for all (i, a, λ) ∈ S.
Now,
φ(i, a, λ) + φ(j, b, µ) = (cid:16)(i, −pλ,i, λ), pλ,i + a(cid:17) + (cid:16)(j, −pµ,j, µ), pµ,j + b(cid:17)
= (cid:16)(i, −pλ,i, λ) + (j, −pµ,j, µ), pλ,i + a + pµ,j + b(cid:17)
= (cid:16)(i, −pµ,i, µ), pµ,i + a + pλ,j + b(cid:17)
= φ(i, a + pλ,j + b, µ)
= φ(cid:16)(i, a, λ) + (j, b, µ)(cid:17).
Again,
Also
φ(i, a, λ) · φ(j, b, µ) = (cid:16)(i, −pλ,i, λ), pλ,i + a(cid:17) · (cid:16)(j, −pµ,j, µ), pµ,j + b(cid:17)
= (cid:16)(i, −pλ,i, λ)(j, −pµ,j, µ), (pλ,i + a)(pµ,j + b(cid:17)
= (cid:16)(ij, −pλµ,ij, λµ), ab(cid:17).
φ(cid:16)(i, a, λ) · (j, b, µ)(cid:17) = φ(ij, −pλµ,ij + ab, λµ)
= (cid:16)(ij, −pλµ,ij, λµ), pλµ,ij − pλµ,ij + ab(cid:17)
= (cid:16)(ij, −pλµ,ij, λµ), ab(cid:17).
So φ(i, a, λ) · φ(j, b, µ) = φ(cid:16)(i, a, λ) · (j, b, µ)(cid:17).
Hence φ is a homomorphism. It is easy to see that φ is one-one and onto. Hence φ is
an isomorphism from a rectangular skew-ring S onto a direct product of a rectangular
band semiring E +(S) and a skew-ring R.
Converse part is obvious.
6
Theorem 3.4. The following conditions on a semiring S are equivalent:
(i) S is a b-lattice of nil-extensions of rectangular skew-rings,
(ii) S is a quasi completely regular semiring and for every e, f ∈ E +(S), there exists
n ∈ N such that n(e + f ) = (n + 1)(e + f ),
(iii) S is additively quasi regular, b2 H ∗
+ b for all b ∈ S and a = a + x + a implies
a = a + 2x + 2a.
Proof. (i) =⇒ (ii): Let S be a b-lattice of nil-extensions of rectangular skew-rings.
Then S is a b-lattice of nil-extensions of orthodox completely simple semirings. So by
Theorem 2.12, it follows that S is a b-lattice of completely Archimedean semirings and
hence by Theorem 2.8, we have S is a quasi completely regular semiring. Again, (S, +)
is a semilattice of nil-extensions of orthodox completely simple semigroups, i.e., (S, +)
is a semilattice of nil-extensions of rectangular groups. Then by Theorem X.2.1 [1],
for every e, f ∈ E +(S), there exists n ∈ Z + such that n(e + f ) = (n + 1)(e + f ).
(ii) =⇒ (iii):
Let (S, +, ·) be a quasi completely regular semiring and for every
e, f ∈ E +(S), there exists n ∈ Z + such that n(e + f ) = (n + 1)(e + f ). Hence S is an
additively quasi regular semiring and b2 H ∗
+ b for all b ∈ S. Let a = a + x + a. Then a
is additively regular and hence by Theorem X.2.1 [1], it follows that a = a + 2x + 2a.
(iii) =⇒ (i):
Let S is additively quasi regular, b2 H ∗
+ b for all b ∈ S and a =
a + x + a implies a = a + 2x + 2a. Then by Theorem X.2.1 [1], (S, +) is a semilattice
of nil-extensions of rectangular groups. This implies (S, +) is a GV-semigroup. To
complete the proof, it suffices to show that every additively regular element of S is
completely regular. For this let a = a + x + a. Then a is regular and hence a is
completely regular in the semigroup (S, +). So there exists an element y ∈ S such
that a = a + y + a and a + y = y + a. Now since H ∗
+ is a congruence on (S, ·)
and (a + y) H ∗
+ (a + y). Since each
+ a, it follows that a(a + y) H ∗
+ a2 H ∗
+ a H ∗
H ∗
+-class contains a unique additive idempotent and a(a + y) H ∗
+ (a + y), it follows
that a(a + y) = a + y. Hence S is a quasi completely regular semiring. Consequently,
S is a b-lattice of nil-extensions of rectangular skew-rings.
Theorem 3.5. A semiring S is nil-extension of a rectangular skew-ring if and only if
S is a completely Archimedean semiring and E +(S) is a subsemigroup of (S, +).
7
Proof. Let (S, +, ·) be a nil-extension of a rectangular skew-ring (K, +, ·). Then (K, +, ·)
is a completely simple semiring and E +(K) is a subsemigroup of (K, +). Hence S is
a completely Archimedean semiring. Clearly, E +(S) = E +(K) and thus E +(S) is a
subsemigroup of (S, +).
Conversely, let (S, +, ·) be a completely Archimedean semiring and E +(S) be a
subsemigroup of (S, +). Then (S, +, ·) is a nil-extension of a completely simple semiring
(K, +, ·). Also, E +(S) = E +(K) implies (K, +, ·) is a completely simple semiring such
that (K, +) is orthodox. Thus K is a rectangular skew-ring and hence S is nil-extension
of a rectangular skew-ring K.
Theorem 3.6. The following conditions are equivalent on a semiring (S, +, ·) :
(i) (S, +, ·) is a quasi completely regular semiring and E +(S) is a subsemigroup of
(S, +).
(ii) S is additively quasi regular, b2 H ∗
+ b for all b ∈ S, a = a + x + a implies
a = a + 2x + 2a and Reg+S is a subsemigroup of (S, +).
(iii) S is a b-lattice of nil-extensions of rectangular skew-rings and E +(S) is a
subsemigroup of (S, +).
Proof. (i) =⇒ (ii): Let S is a quasi completely regular semiring and E +(S) is a subsemi-
group of (S, +). Then clearly S is a b-lattice of nil-extensions of rectangular skew-rings.
Hence by Theorem 3.4, it follows that S is additively quasi regular, b2 H ∗
+ b for all
b ∈ S and a = a+x+a implies a = a+2x+2a. Also E +(S) is a subsemigroup of (S, +).
Then by proposition X.2.1 [1], for any a, b, x, y ∈ S, a = a + x + a and b = b + y + b
implies a + b = a + b + y + x + a + b, i.e., a, b ∈ Reg+S implies a + b ∈ Reg+S. Hence
Reg+S is a subsemigroup of (S, +).
(ii) =⇒ (iii): Let S be additively quasi regular, b2 H ∗
+ b for all b ∈ S, a = a + x + a
implies a = a + 2x + 2a and Reg+S is a subsemigroup of (S, +). Then by Theorem
3.4, S is a b-lattice of nil-extensions of rectangular skew-rings. Let e, f ∈ E +(S) ⊂
Reg+(S). Then e + f ∈ Reg+(S) and let x ∈ S be an additive inverse of e + f . Then
f + x + e = f + (x + e + f + x) + e = 2(f + x + e). Now (e + f ) + (f + x + e) + (e + f ) =
e + f + x + e + f = e + f implies e + f = (e + f ) + 2(f + x + e) + 2(e + f ) =
(e + f ) + (f + x + e) + (e + f ) + (e + f ) = (e + f ) + (e + f ) = 2(e + f ). Hence we have
8
E +(S) is a subsemigroup of (S, +).
(iii) =⇒ (i): It is obvious by Theorem 3.4.
Theorem 3.7. The following conditions are equivalent on a semiring (S, +, ·) :
(i) S is a quasi completely regular semiring and for every e, f ∈ E +(S), e+f = f +e,
(ii) S is b-lattice of quasi skew-rings and for every e, f ∈ E +(S), e + f = f + e,
(iii) S is additively quasi regular and Reg+S is a subsemigroup of (S, +) which is
a b-lattice of skew-rings.
Proof. (i) =⇒ (ii):
Let S be a quasi completely regular semiring and for every
e, f ∈ E +(S), e + f = f + e. Then H ∗+ is a b-lattice congruence on S. Hence S is
b-lattice of quasi skew-rings and for every e, f ∈ E +(S), e + f = f + e.
(ii) =⇒ (i): This part is obvious.
(i) =⇒ (iii):
Let S be a quasi completely regular semiring such that for every
e, f ∈ E +(S), e + f = f + e. Then obviously S is additively quasi regular. Now let
a, b ∈ Reg+S. Then there exist x, y ∈ S such that a = a + x + a and b = b + y + b. So
a + x, x + a, b + y, y + b ∈ E +(S). Now, a + b = a + x + a + b + y + b = a + b + y + x + a + b
implies a + b ∈ Reg+S. Hence Reg+S is a subsemigroup of (S, +). Since in a quasi
completely regular semiring, every additively regular element is completely regular, it
follows that Reg+S is a completely regular semiring. Thus by Theorem 3.6 [9], Reg+S
can be regarded as a b-lattice Y of completely simple semirings Rα (α ∈ Y ), where
Y = S/J + and each Rα is a J +-class in S. Let e and f be two additive idempotents
in Rα. Then e + f = f + e and e J + f . Since Rα is a completely simple semiring then
(Rα, +) is a completely simple semigroup and so e D + f and thus we have e = f . This
shows that each Rα contains a single additive idempotent, so that (Rα, +) is a group
and hence (Rα, +, ·) is a skew-ring. In other words, we have shown that Reg+S is a
b-lattice Y of skew-rings Rα.
(iii) =⇒ (i):
Let S be an additively quasi regular semiring and Reg+S be
a subsemigroup of (S, +) which is a b-lattice of skew-rings. Then clearly S is a
quasi completely regular semiring. Moreover, (Reg+(S), +) is a Clifford semigroup.
Let e, f ∈ E +(S). Then e and f are two idempotents in the Clifford semigroup
(Reg+(S), +). Hence e + f = f + e.
9
Definition 3.8. An idempotent semiring (S, +, ·) is said to be a left zero semiring if
(S, +) is a left zero band.
Definition 3.9. A semiring S is said to be a left skew-ring if it is isomorphic to a
direct product of a left zero semiring and a skew-ring.
We recall that a left group is isomorphic to a direct product of a left zero semigroup
and a group. Thus, if a semiring (S, +, ·) is a left skew-ring then the semiring (S, +, ·)
is a rectangular skew-ring and the semigroup (S, +) is a left group and conversely.
Theorem 3.10. The following conditions are equivalent on a semiring (S, +, ·) :
(i) S is a b-lattice of nil-extensions of left skew-rings.
(ii) S is a quasi completely regular semiring and for every e, f ∈ E +(S), there exists
n ∈ N such that n(e + f ) = n(e + f + e).
(iii) S is additively quasi regular, b2 H ∗
+ b for all b ∈ S and a = a + x + a implies
a + x = a + 2x + a.
Proof. (i) =⇒ (ii): Let S be a b-lattice Y of nil-extensions of left skew-rings Sα (α ∈ Y ).
Then S is a b-lattice of nil-extensions of rectangular skew-rings. Hence by Theorem 3.4,
S is a quasi completely regular semiring. Again (S, +) is a semilattice of nil-extensions
of left groups (Sα, +) (α ∈ Y ). Hence by Theorem X.2.2 [1], it follows that for every
e, f ∈ E +(S), there exists a positive integer n such that n(e + f ) = n(e + f + e).
(ii) =⇒ (iii): Let S be a quasi completely regular semiring and for every e, f ∈
E +(S), there exists n ∈ N such that n(e + f ) = n(e + f + e). Then clearly S is
additively quasi regular and b2 H ∗
+ b for all b ∈ S. Again (S, +) is a GV-semigroup
such that for every e, f ∈ E +(S), there exists n ∈ N such that n(e + f ) = n(e + f + e).
Hence by Theorem X.2.2 [1], follows that a = a + x + a implies a + x = a + 2x + a.
(iii) =⇒ (i): Let S be an additively quasi regular such that b2 H ∗
+ b for all b ∈ S
and a = a + x + a implies a + x = a + 2x + a. Then a = a + x + a = (a + x) + a =
(a + 2x + a) + a = a + 2x + 2a. So by Theorem 3.4, it follows that S is a b-lattice Y of
nil-extensions of rectangular skew-rings Sα (α ∈ Y ). By Theorem X.2.2 [1], it follows
that (Sα, +) is a left group. Hence Sα is a left skew-ring. Consequently, S is a b-lattice
of nil-extensions of left skew-rings.
10
Theorem 3.11. The following conditions are equivalent on a semiring (S, +, ·) :
(i) S is a b-lattice of nil-extensions of left skew-rings and E +(S) is a subsemigroup
of (S, +).
(ii) S is a quasi completely regular semiring and for every e, f ∈ E +(S), e + f =
e + f + e.
Proof. (i) =⇒ (ii): Let S be a b-lattice of nil-extensions of left skew-rings and E +(S)
is a subsemigroup of (S, +). Then by Theorem 3.10, it follows that S is a quasi
completely regular semiring and for every e, f ∈ E +(S), there exists a positive integer
n such that n(e + f ) = n(e + f + e). Now since E +(S) is a subsemigroup of (S, +), we
have e + f, e + f + e ∈ E +(S). Hence e + f = n(e + f ) = n(e + f + e) = e + f + e.
(ii) =⇒ (i): Let S be a quasi completely regular semiring and for every e, f ∈
E +(S), e + f = e + f + e. Then by Theorem 3.10, we have S is a b-lattice of nil-
extensions of left skew-rings. Also for every e, f ∈ E +(S), 2(e + f ) = e + f + e + f =
e + f + f = e + f which implies that e + f ∈ E +(S), i.e., E +(S) is a subsemigroup of
(S, +).
References
[1] Bogdanovi´c, S., Semigroups with a system of subsemigroups, Novi Sad, 1985.
[2] Golan, J. S., The Theory of Semirings with Applications in Mathematics and The-
oretical Computer Science. Pitman Monographs and Surveys in Pure and Applied
Mathematics 54, Longman Scientific, 1992.
[3] Howie, J. M., Introduction to the theory of semigroups. Academic Press, 1976.
[4] Maity, S. K. and Ghosh, R., On Quasi Completely Regular Semirings, Semigroup
Forum 89, no. 1 (2014), 422-430.
[5] Maity, S. K. and Ghosh, R., Nil-extension of a Completely Simple Semiring, Dis-
cussiones Mathematicae-General Algebra and Applications 33 (2013), 201-209.
[6] Maity, S. K., Ghosh, R. and Chatterjee, R., On Quasi Completely Inverse Semir-
ings, Accepted for publication in Southeast Asian Bulletin of Mathematics.
[7] Pastijn, F. and Guo. Y. Q., The lattice of idempotent distributive semiring vari-
eties., Science in China (Series A) 42 (8) (1999), 785-804.
11
[8] Petrich, M. and Reilly, N. R., Completely regular semigroups, Wiley, New York,
1999.
[9] Sen, M. K., Maity, S. K. and Shum, K. P., On Completely Regular Semirings,
Bull. Cal. Math. Soc. 98 (2006), 319-328.
[10] Sen, M. K., Maity, S. K. and Weinert, H. J., Completely Simple Semirings, Bull.
Cal. Math. Soc. 97 (2005), 163-172.
12
|
1901.06362 | 1 | 1901 | 2019-01-18T18:09:53 | Ideals of direct products of rings | [
"math.RA"
] | It is known that an ideal of a direct product of commutative unitary rings is directly decomposable into ideals of the corresponding factors. We show that this does not hold in general for commutative rings and we find necessary and sufficient conditions for direct decomposability of ideals. For varieties of commutative rings we derive a Mal'cev type condition characterizing direct decomposability of ideals and we determine explicitly all varieties satisfying this condition. | math.RA | math |
Ideals of direct products of rings
Ivan Chajda, Günther Eigenthaler and Helmut Länger
Abstract
It is known that an ideal of a direct product of commutative unitary rings is
directly decomposable into ideals of the corresponding factors. We show that this
does not hold in general for commutative rings and we find necessary and sufficient
conditions for direct decomposability of ideals. For varieties of commutative rings
we derive a Mal'cev type condition characterizing direct decomposability of ideals
and we determine explicitly all varieties satisfying this condition.
AMS Subject Classification: 13A15, 16R40, 08B05
Keywords: Commutative ring, ring ideal, direct product, directly decomposable ideal,
Mal'cev condition, variety of commutative rings
Recall that an ideal of a ring R = (R, +, ·) is a non-empty subset I of R such that if
a, b ∈ I then a − b ∈ I and ar, ra ∈ I for every r ∈ R. For other concepts used here the
reader is referred to any monograph on rings, see e.g. [1] and [4]. Ideals play a crucial
role in the theory of rings since the kernels of homomorphisms are ideals and rings can
be factorized by means of ideals.
It is an elementary fact that in rings, ideals and congruences are in a one-to-one corre-
spondence. Hence, if R = (R, +, ·) is a ring then the ideal lattice Id R := (Id R, ⊆) and
the congruence lattice Con R := (Con R, ⊆) are isomorphic. Hence Id R is a modular
bounded lattice with the least element {0} and the greatest element R where the supre-
mum and infimum of ideals I1, I2 are given by I1 ∨ I2 = I1 + I2 and I1 ∧ I2 = I1 ∩ I2,
respectively.
Having two rings R1 = (R1, +, ·) and R2 = (R2, +, ·), it is elementary that for I1 ∈ Id R1
and I2 ∈ Id R2 we have I1 × I2 ∈ Id(R1 × R2). On the other hand, if I ∈ Id(R1 × R2)
then there need not exist I1 ∈ Id R1 and I2 ∈ Id R2 with I1 × I2 = I. If such ideals I1, I2
do not exist, then I is called skew. Otherwise, I will be called directly decomposable. For
commutative rings R1, R2 we will derive conditions under which R1 × R2 has no skew
ideals.
We say that a direct product of finitely many rings has directly decomposable ideals if
every ideal of this product is a direct product of ideals of the corresponding factors.
For sets M1, M2 and i ∈ {1, 2} let πi denote the i-th projection from M1 × M2 onto Mi.
1Support of the research of all three authors by ÖAD, project CZ 04/2017, support of the first and
the third author by IGA, project PřF 2018 012, as well as support of the second and the third author
by the Austrian Science Fund (FWF), project I 1923-N25, is gratefully acknowledged.
2Electronic version of an article published in Asian-European Journal of Mathematics, Vol. 11, No.
4, 2018, pages 1850094-1 -- 1850094-6, DOI: 10.1142/S1793557118500948 c(cid:13) World Scientific Publishing
Company, https://www.worldscientific.com/worldscinet/aejm
1
It is easy to see that an ideal I of R1 × R2 is directly decomposable if and only if
π1(I) × π2(I) = I which is equivalent to π1(I) × π2(I) ⊆ I.
Direct decomposability of ideals in commutative rings was used by the first and the third
author in their study of complementation in ideal lattices, see [2].
Let K4 = ({0, a, b, c}, +) denote the four-element Kleinian group whose operation table
for + looks as follows:
+ 0 a b
0
0 a b
a a 0 c
b
c
c
c
b
c 0 a
b a 0
b
c
We show an example of rings whose direct product contains a skew ideal.
Example 1. Consider the zero-ring K = (K, +, ·) whose additive group is K4, i.e.,
xy = 0 for all x, y ∈ K. Then the principal ideal I(a, c) = {(0, 0), (a, c)} of K × K
generated by (a, c) is skew since I 6= π1(I) × π2(I).
In what follows, we derive necessary and sufficient conditions under which an ideal of the
direct product of rings is directly decomposable. For this purpose, we borrow the method
developed by Fraser and Horn ([3]) for congruences.
Theorem 2. Let R1 = (R1, +, ·) and R2 = (R2, +, ·) be rings and I ∈ Id(R1 × R2).
Then the following are equivalent:
(i) I is directly decomposable.
(ii) (R1 × {0}) ∩ (({0} × R2) + I) ⊆ I and
((R1 × {0}) + I) ∩ ({0} × R2) ⊆ I.
(iii) If (a, b) ∈ I then (a, 0), (0, b) ∈ I.
(iv) ((R1 × {0}) + I) ∩ (({0} × R2) + I) = I.
Proof.
(i) ⇒ (ii):
If I = I1 × I2 then
(R1 × {0}) ∩ (({0} × R2) + I) = (R1 × {0}) ∩ (({0} × R2) + (I1 × I2)) =
= (R1 × {0}) ∩ (I1 × R2) = I1 × {0} ⊆ I,
((R1 × {0}) + I) ∩ ({0} × R2) = ((R1 × {0}) + (I1 × I2)) ∩ ({0} × R2) =
= (R1 × I2) ∩ ({0} × R2) = {0} × I2 ⊆ I.
(ii) ⇒ (iii):
If (a, b) ∈ I then
(a, 0) ∈ (R1 × {0}) ∩ (({0} × R2) + I) ⊆ I,
(0, b) ∈ ((R1 × {0}) + I) ∩ ({0} × R2) ⊆ I.
(iii) ⇒ (i):
If (a, b) ∈ π1(I) × π2(I) then there exists some (c, d) ∈ R1 × R2 with (a, d), (c, b) ∈ I,
2
hence (a, 0), (0, b) ∈ I which shows (a, b) = (a, 0) + (0, b) ∈ I.
(i) ⇒ (iv):
If I = I1 × I2 then
((R1 × {0}) + I) ∩ (({0} × R2) + I) =
= ((R1 × {0}) + (I1 × I2)) ∩ (({0} × R2) + (I1 × I2)) = (R1 × I2) ∩ (I1 × R2) =
= I1 × I2 = I.
(iv) ⇒ (ii):
This follows immediately.
The following result is already known but it follows easily from the equivalence of (i) and
(iii) in Theorem 2.
Corollary 3. If R1 and R2 are unitary rings then I is directly decomposable since (a, b) ∈
I implies (a, 0) = (a, b)(1, 0) ∈ I and (0, b) = (a, b)(0, 1) ∈ I and hence (iii) holds.
Corollary 4. Let R1, R2 be rings. Then Id(R1 × R2) is distributive if and only if every
ideal of R1 × R2 is directly decomposable and Id R1, Id R2 are distributive.
Proof. If Id(R1 × R2) is distributive, then (ii) of Theorem 2 is satisfied for every I ∈
Id(R1 × R2). Thus every ideal of R1 × R2 is directly decomposable. Moreover, for
i = 1, 2 we have that Id Ri is isomorphic to the principal filter of Id(R1 × R2) generated
by the kernel of πi. Therefore, Id R1 and Id R2 are distributive. Conversely, suppose
that Id R1, Id R2 are distributive and every ideal of R1 × R2 is directly decomposable.
Let I, J ∈ Id(R1 ×R2). Then there exist I1, J1 ∈ Id R1 and I2, J2 ∈ Id R2 with I1 ×I2 = I
and J1 × J2 = J. Now we have
I ∨ J = (I1 × I2) + (J1 × J2) = (I1 + J1) × (I2 + J2),
I ∧ J = (I1 × I2) ∩ (J1 × J2) = (I1 ∩ J1) × (I2 ∩ J2).
Hence, join and meet in Id(R1 × R2) are computed "component-wise" showing distribu-
tivity of Id(R1 × R2).
Another application of Theorem 2 is the following example:
Example 5. If R1 = (R1, +, ·) is a Boolean ring (i.e., xx ≈ x) and R2 = (R2, +, ·) a
unitary ring then R1 × R2 has no skew ideals. This follows directly by (iii) of Theorem 2
since if (a, b) ∈ I then (a, 0) = (a, b)(a, 0) ∈ I and (0, b) = (a, b)(0, 1) ∈ I.
Denote by Z the ring of integers. As usually, for a ∈ Z put aZ := {ax x ∈ Z}. Of course,
for each a ∈ Z, aZ is a commutative ring which is unitary only in case a ∈ {−1, 0, 1}.
The next theorem shows that, in general, the rings aZ × bZ contain skew ideals. Hence,
the rather exotic ring K × K from Example 1 is not the only commutative ring possessing
skew ideals.
Theorem 6. Let a, b, c, d ∈ Z with ca and db and consider the ideal I(a, b) of cZ × dZ
generated by (a, b). Then I(a, b) is directly decomposable if and only if either a = 0 or
b = 0 or gcd(c, d) = 1.
3
Proof. Put (a, b)Z := {(ax, bx) x ∈ Z}. Obviously, I(a, b) = (a, b)Z+(acZ×bdZ). Since
π1(I(a, b)) = aZ and π2(I(a, b)) = bZ, direct decomposability of I(a, b) is equivalent to
aZ × bZ ⊆ I(a, b). If a = 0 or b = 0 then obviously aZ × bZ ⊆ I(a, b). Now assume
a, b 6= 0. If gcd(c, d) = 1 then there exist e, f ∈ Z with ce + df = 1 and hence
(ax, by) = (a, b)(df x + cey) + ((ac)(e(x − y)), (bd)(−f (x − y))) ∈ I(a, b)
for all x, y ∈ Z proving direct decomposability of I(a, b). Finally, assume gcd(c, d) 6= 1.
Suppose I(a, b) to be directly decomposable. Then (a, 0) ∈ I(a, b) and hence there exist
g, h, i ∈ Z with (a, 0) = (ag + ach, bg + bdi). From this we conclude g = −di and
hence ch − di = 1 whence gcd(c, d) = 1, a contradiction. Hence I(a, b) is not directly
decomposable in this case.
Although the rings of the form cZ are rings of integers, Theorem 6 yields the following
result.
Corollary 7. If c, d ∈ Z and gcd(c, d) 6= 1 then cZ × dZ has skew ideals.
Example 8. The principal ideal
I(2, 2) = (2, 2)Z + (4Z × 4Z) = (4Z × 4Z) ∪ ((2 + 4Z) × (2 + 4Z))
of 2Z × 2Z generated by (2, 2) is skew since (0, 0), (2, 2) ∈ I(2, 2), but (0, 2) /∈ I(2, 2).
This is in accordance with Theorem 6.
Using Theorem 2 we can generalize the situation described in Example 5. Namely, we
can derive a Mal'cev type condition characterizing varieties of commutative rings whose
ideals are directly decomposable.
Theorem 9. Let V be a variety of commutative rings. Then V has directly decomposable
ideals if and only if there exists a unary term t satisfying the identity xt(x) ≈ x.
Proof. First assume V to have directly decomposable ideals. Consider the free commu-
tative ring F(x) = (F (x), +, ·) with one free generator x and the principal ideal
I(x, x) = {n(x, x) + (x, x)(r(x), s(x)) n ∈ Z and (r(x), s(x)) ∈ F (x) × F (x)} =
= {(nx + xr(x), nx + xs(x)) n ∈ Z and r(x), s(x) ∈ F (x)}
of F(x) × F(x) generated by (x, x). Since I(x, x) is directly decomposable and (0, 0),
(x, x) ∈ I(x, x), we have (x, 0) ∈ I(x, x). Hence there exist some n ∈ Z and r(x), s(x) ∈
F (x) with (nx + xr(x), nx + xs(x)) = (x, 0) which implies xt(x) = x with t(x) :=
r(x) − s(x). Conversely, assume there exists a unary term t satisfying xt(x) ≈ x. Let
R1 = (R1, +, ·), R2 = (R2, +, ·) ∈ V and I ∈ Id(R1 × R2) and assume (a, b) ∈ I. Then
(a, 0) = (a, b)(t(a), 0) ∈ I and (0, b) = (a, b)(0, t(b)) ∈ I. According to (iii) of Theorem 2,
I is directly decomposable.
Using Theorem 9, we can explicitly describe all varieties of commutative rings having
directly decomposable ideals.
Corollary 10. A variety of commutative rings has directly decomposable ideals if and
only if it satisfies an identity of the form
aixi ≈ x with n ≥ 2 and a2, . . . , an ∈ Z.
n
P
i=2
4
Proof. This follows immediately from Theorem 9 since the unary terms in a variety of
n
commutative rings are exactly the terms of the form
aixi with n ≥ 1 and a1, . . . , an ∈
P
i=1
Z.
Corollary 11. The variety of Boolean rings has directly decomposable ideals since it
satisfies the identity x2 ≈ x.
However, we can also consider classes of commutative rings which do not form a variety.
In this case we cannot apply Theorem 9 in order to prove that rings belonging to such a
class have directly decomposable ideals. A typical example is the following:
Example 12. Let R = (R, +, ·) be a commutative ring, F = (F, +, ·) a field, I ∈
Id(R × F) and J := π1(I). If I ⊆ R × {0} then I = J × {0}. Now assume I 6⊆ R × {0}.
Then there exists some (a, b) ∈ I with b 6= 0. If (c, d) ∈ J × F then there exists some
e ∈ F with (c, e) ∈ I and hence
(c, d) = (c, e) + (a, b)(0, b−1(d − e)) ∈ I
showing I = J × F . Hence, R × F has directly decomposable ideals. By induction we
obtain that R × F1 × · · · × Fn has directly decomposable ideals if n ≥ 1 and F1, . . . , Fn
are fields.
In particular, we can consider the case where R denotes the commutative ring K from
Example 1. Then K × F has directly decomposable ideals despite the fact that K × K
does not have this property. Similarly, if c and d are integers satisfying gcd(c, d) 6= 1 then
(cZ × dZ) × F, cZ × F and dZ × F have directly decomposable ideals though cZ × dZ
does not have this property.
Remark 13. Note that Example 12 remains valid in case that R is not commutative.
Acknowledgement. The authors thank the referee for his valuable suggestions which
increased the quality of the paper.
References
[1] I. T. Adamson, Rings, modules and algebras. Oliver and Boyd, Edinburgh 1971.
[2] I. Chajda and H. Länger, Commutative rings whose ideal lattices are complemented.
Asian-European J. Math. (to appear).
[3] G. A. Fraser and A. Horn, Congruence relations in direct products. Proc. Amer. Math.
Soc. 26 (1970), 390 -- 394.
[4] J. Lambek, Lectures on rings and modules. Chelsea Publ. Co., New York 1976.
Authors' addresses:
Ivan Chajda
Palacký University Olomouc
5
Faculty of Science
Department of Algebra and Geometry
17. listopadu 12
771 46 Olomouc
Czech Republic
[email protected]
Günther Eigenthaler
TU Wien
Faculty of Mathematics and Geoinformation
Institute of Discrete Mathematics and Geometry
Wiedner Hauptstrasse 8-10
1040 Vienna
Austria
[email protected]
Helmut Länger
TU Wien
Faculty of Mathematics and Geoinformation
Institute of Discrete Mathematics and Geometry
Wiedner Hauptstrasse 8-10
1040 Vienna
Austria, and
Palacký University Olomouc
Faculty of Science
Department of Algebra and Geometry
17. listopadu 12
771 46 Olomouc
Czech Republic
[email protected]
6
|
1709.01759 | 1 | 1709 | 2017-09-06T10:52:46 | Complexity of term representations of finitary functions | [
"math.RA",
"math.LO"
] | The clone of term operations of an algebraic structure consists of all operations that can be expressed by a term in the language of the structure. We consider bounds for the length and the height of the terms expressing these functions, and we show that these bounds are often robust against the change of the basic operations of the structure. | math.RA | math |
COMPLEXITY OF TERM REPRESENTATIONS OF FINITARY
FUNCTIONS
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
Abstract. The clone of term operations of an algebraic structure consists of
all operations that can be expressed by a term in the language of the structure.
We consider bounds for the length and the height of the terms expressing these
functions, and we show that these bounds are often robust against the change
of the basic operations of the structure.
1. Motivation
Following [BS81], an algebraic structure, or an algebra, is a pair A = (A, F )
where A is a non-empty set called the universe of A, and F is a set of operations on
A. The operations in F are called the basic operations of A. A language (sometimes
called signature, or type) of an algebra A is a set of symbols, one for each basic
operation, together with their arities. An n-ary term over the variables x1, . . . , xn
is any properly formed expression using these operation symbols and variables. A
natural measure of the complexity of a term is its length as a string, i.e., the total
number of both operation and variable symbols it contains. We will denote this
quantity by len(t) and call it the length of the term t. Formally, len(x) = 1 if x is
a variable or a nullary operation symbol, and len(f (t1, . . . , tk)) = 1 + Pk
i=1 len(ti)
if f is a k-ary operation symbol and t1, . . . , tk are terms. A term t over x1, . . . , xn
naturally induces a function tA from An to A, which maps a tuple (a1, . . . , an) to
the value obtained by substituting ai for each xi and interpreting the operation
symbols by the corresponding operations. Next, we measure the complexity of
a term function of a finite algebra A: the length of a term function tA is the length
of the shortest term that induces this function; formally,
lenA(tA) = min{len(s) : s is a term in the language of A with sA = tA}.
From this we define a sequence lA(n) whose n-th element expresses the worst-case
complexity of an n-ary term function of A:
lA(n) = max{lenA(tA) : t is an n-ary term in the language of A}.
In other words, lA(n) is the smallest number of symbols that is sufficient to write
down any n-ary term operation of A.
In the present paper we study how the
asymptotics of the sequence lA(n) depends on the properties of the algebra A. The
algebras A and B are term equivalent if they are defined on the same universe
and have the same term operations. Unlike other similar sequences studied in the
literature, e.g., the free spectrum, the sequences lA and lB may differ even when A
and B are term equivalent. Nevertheless, in some cases we are able to prove that
the asymptotics of these sequences does not change. This motivates the following
definition.
Date: October 10, 2018.
Supported by the Austrian Science Fund (FWF):P29931, Research Grant 174018 of the Min-
istry of Science and Education of the Republic of Serbia, and European Research Council (Grant
Agreement no. 681988, CSP-Infinity).
1
2
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
Definition 1.1. Let A be a finite algebra of finite type. Then A is resilient against
change of signature, or simply resilient if for every algebra B of finite type that is
term equivalent to A, there exist polynomials p, q such that lA(n) ≤ p(lB(n)) and
lB(n) ≤ q(lA(n)) for each n ∈ N.
It is not known to the authors whether there is a finite algebra of finite type which
is not resilient against change of signature. We show that two classes of algebras are
resilient. In both cases, the sequences lA(n) are close to the theoretical lower bound
given by a simple counting argument described in the following section. These
two cases are primal algebras and supernilpotent algebras in congruence modular
varieties. We recall that an algebra A with a universe A is called primal if every
operation on A is a term operation of A.
Theorem 1.2. Let A be a finite primal algebra of finite type with at least 2 elements.
Then there exist positive real numbers c1, c2 such that for all n ∈ N, 2c1n ≤ lA(n) ≤
2c2n.
Corollary 1.3. Every finite primal algebra of finite type is resilient.
The proof of these results is given in Section 3.
The other property, supernilpotency, can be defined by a condition on higher
commutators of the algebra. Higher commutators, introduced in [Bul01], generalize
binary commutators, whose theory has been described in [FM87]. They have been
studied in [AM10] and recently in [Opr16, Moo16, Wir17]. By a result of Kearnes
[Kea99], the supernilpotent algebras distinguish themselves among other algebras
in a congruence modular variety by having a small number of term operations. For
a precise statement of this fact and a self-contained definition of supernilpotency,
we refer the reader to Section 4.
Theorem 1.4. Let A be a finite supernilpotent algebra in a congruence modular
variety with at least 2 elements. Then there exists a polynomial q such that for all
n ∈ N, n − 1 ≤ lA(n) ≤ q(n).
Corollary 1.5. Every finite supernilpotent algebra of finite type in a congruence
modular variety is resilient.
These results are proved in Section 4. For groups, Horv´ath and Nehaniv have
proved that for each nilpotent group G, lG(n) is bounded from above by a poly-
nomial in n, and for simple groups, one has lG(n) ≤ C(G) · n8 · Gn where C(G)
depends only on the group G [HN15].
2. Notation and general bounds
We will write N for the set of positive integers, N0 = N ∪ {0}, and log(n) is the
logarithm of n in base e. For an algebra A, the set of all term operations is denoted
by Clo A and called the clone of term operations of A. Its n-ary part is denoted by
Clon A. Clon A can be endowed with a structure of an algebra, denoted by FA(n),
in the same language as A by setting:
f FA(n)(tA
1 , . . . , tA
k ) = (f (t1, . . . , tk))A,
or equivalently defining the operations coordinatewise, seeing Clon A as a subuni-
verse of AAn
. It is well-known that this algebra is freely generated (in HSP A) by
the projections (the set {xA
n }). The sequence defined by
1 , . . . , xA
SpecA(n) = FA(n)
for n ∈ N is called the free spectrum of A.
We will now introduce another measure for the complexity of a term. Every term
can be expressed as a rooted tree whose vertices correspond to the function and
variable symbols appearing in it. The variables correspond to leaves, each function
COMPLEXITY OF TERMS
3
q
q
p
x1
x2
x2 x4
q
x1
x1
Figure 1. Tree representation of term q(q(x1, x2), p(x2, x4, q(x1, x1))).
symbol has exactly as many children as is the arity of the symbol, and the children
have a (usually implicitly) given order. We call this tree the tree representation of
a term t. The number of vertices of the tree representing t is exactly len(t). The
height of the term t, denoted by ht(t), represents the height of the tree representation
of t, with an adjustment if t contains a nullary operation symbol: Precisely, we define
ht(t) inductively by setting: ht(x) = 0 if x is a variable; ht(f ) = 1 if f is a nullary
operation symbol; and if f is a k-ary operation symbol where k ∈ N, and t1, . . . , tk
are terms, we set ht(f (t1, . . . , tk)) = 1 + max{ht(t1), . . . , ht(tk)}. Through the rest
of the present paper, we set max ∅ = 0. This will simplify some proofs since it saves
a case distinction between nullary and non-nullary operation symbols.
The height of a term function tA is defined by
htA(tA) = min{ht(s) : s is a term in the language of A with sA = tA}.
This allows us to define a second sequence measuring the complexity of terms of
an algebra A. For n ∈ N, let
hA(n) = max{htA(tA) : t is an n-ary term in the language of A}.
Lemma 2.1. Let A be a finite algebra of finite type, let r be the number of operation
symbols of A and m be their maximal arity. Then:
(1) SpecA(n) < (n + r + 1)lA(n) for all n ∈ N;
(2) there exists a positive real number c such that for all n ≥ 2 we have
SpecA(n) ≤ 2c log(n) lA(n);
(3) hA(n) ≤ SpecA(n) for all n ∈ N;
(4) hA(n) ≤ lA(n) for all n ∈ N;
(5) lA(n) ≤ 1 + m + · · · + mhA(n) ≤ (m + 1)hA(n) for all n ∈ N.
Proof. (1) By the definition of the sequence lA(n), every n-ary term operation
is given by a term whose representation in prefix notation is a string of length
at most lA(n) using n symbols for variables and r symbols for basic operations.
Therefore, there are at most PlA(n)
l=1 (n + r)l < (n + r + 1)lA(n) term operations of
A of arity n. Now (2) follows because there is a positive real number c such that
log2(n + r + 1) ≤ c log n for all n ≥ 2.
(3) Let F ≤ AAn
be the universe of the free algebra generated by the projections
π1, . . . , πn, i.e., the elements of F are precisely the n-ary term functions of A. We
consider the following alternative construction of the set F : We start by taking
H0 = {π1, . . . , πn}, and then inductively define sets Hk by
Hk+1 = {f AAn
(t1, . . . , tm) : t1, . . . , tm ∈ Hk, f is a basic operation of A} ∪ Hk
for k ∈ N0. If for some k, Hk+1 = Hk, then Hk is closed under the operations of
AAn
, therefore a subuniverse, and hence F = Hk. Let k0 be the smallest k such
that Hk = Hk+1. Then F = Hk0 . Now, since Hk+1 ≥ Hk + 1 for all k < k0, we
obtain F ≥ k0. Further, observe that for any k, Hk is the set of all n-ary term
4
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
operations of A which can be expressed by a term of depth at most k, in other
words, Hk = {f A ∈ Clon(A) : ht(f A) ≤ k}. Therefore, since Hk0 = F , we obtain
that ht(f A) ≤ k0 ≤ F for any n-ary term operation f A of A, which immediately
yields the required inequality.
(4) Let t be an n-ary term. Then there is a term s with len(s) ≤ lA(n) such that
tA = sA. For every term, we have len(s) ≥ ht(s), and therefore ht(s) ≤ lA(n). This
implies hA(n) ≤ lA(n).
(5) We first observe that for every term t,
len(t) ≤ 1 + m + · · · + mht(t)
(2.1)
because a rooted tree in which every vertex has at most m children has at most md
vertices of depth d. In order to obtain an upper bound for lA(n), we let t be an
n-ary term. Hence ht(tA) ≤ hA(n), and hence there is a term s with ht(s) ≤ hA(n)
and sA = tA. Now (2.1) implies that len(s), and therefore also lenA(tA), is at most
PhA(n)
i=0 mi which completes the proof of the first inequality. The second inequality
is immediate.
(cid:3)
For an n-ary term t and terms t1, . . . , tn, we write t(t1, . . . , tn) for the term
obtained by substituting every occurrence of xi in t by ti.
Lemma 2.2. Let n ∈ N, let L be a language, and let t, t1, . . . , tn be terms of type
L. If t is n-ary, then ht(t(t1, . . . , tn)) ≤ ht(t) + max{ht(t1), . . . , ht(tn)}.
Proof. We proceed by induction on ht(t): If t is a variable xi, then ht(t) = 0, and
the statement reduces to ht(ti) ≤ max{ht(t1), . . . , ht(tn)} which is trivial. For the
induction step, we assume that t = f (s1, . . . , sk) for some operation symbol f and n-
ary terms s1, . . . , sk. To simplify the formulae below, let m = max{ht(t1), . . . , ht(tn)}.
Observe that ht(si) ≤ ht(t) − 1. From the induction hypothesis we obtain
ht(si(t1, . . . , tn)) ≤ ht(si) + m ≤ ht(t) + m − 1
for every i ∈ {1, . . . , k}. Therefore,
ht(t(t1, . . . , tn)) = ht(f (s1(t1, . . . , tn), . . . , sk(t1, . . . , tn)))
= 1 + max{ht(s1(t1, . . . , tn)), . . . , ht(sk(t1, . . . , tn))} ≤ ht(t) + m.
(cid:3)
Lemma 2.3. Let A and B be two term equivalent finite algebras of finite type.
Then there exist positive real numbers c1, c2 such that for all n ∈ N,
c1 hB(n) ≤ hA(n) ≤ c2 hB(n).
Proof. We first consider the second inequality. We choose c2 such that every basic
operation f B of B can be expressed as a term operation of A of height at most c2;
such a c2 exists because B is of finite type and the two algebras are term equivalent.
Now, given that tA is a term operation of A of arity n, and A and B are term
equivalent, then tA is also a term operation of B. Therefore, there exists a term s
in the language of B of height at most hB(n) such that sB = tA. We will turn this
term into a term r in the language of A with rA = sB by substituting every basic
operation symbol by its definition as a term of A (again, we use that A and B are
term equivalent). The term r which is obtained this way will be of height at most
c2 ht(s) ≤ c2 hB(n). Formally, we prove the existence of a term r with rA = sB and
ht(r) ≤ c2 ht(s) by induction on the height of s. If s is a variable, we set r := s and
obtain the required inequality from the fact that ht(r) = ht(s) = 0. Now assume
that s = f (s1, . . . , sk) for some k ∈ N0, some terms s1, . . . , sk and a basic operation
f of B. Then ht(si) ≤ ht(s) − 1 for all i. By the induction hypothesis, we know
that there are terms r1, . . . , rk such that
ht(ri) ≤ c2 ht(si) ≤ c2(ht(s) − 1) and rA
i = sB
i .
COMPLEXITY OF TERMS
5
There is also a term g in the language of A of height at most c2 such that gA = f B.
Let r = g(r1, . . . , rn). Then rA = gA(rA
k ) = sB. More-
over, from Lemma 2.2 we obtain that ht(r) ≤ ht(g) + max{ht(ri) : i ∈ {1, . . . , k}} ≤
c2 + c2(ht(s) − 1) = c2 ht(s) ≤ c2 hB(n). This completes the induction step. There-
fore hA(n) ≤ c2 hB(n).
k ) = f B(sB
1 , . . . , rA
1 , . . . , sB
For proving the first inequality, we interchange the roles of A and B to obtain a
c′ > 0 such that for all n ∈ N, hB(n) ≤ c′ hA(n). Then the inequality is satisfied
with c1 := 1/c′.
(cid:3)
Corollary 2.4. Let A and B be two term equivalent finite algebras of finite type.
Then there exist positive real numbers d1, d2 such that for all n ∈ N,
d1 log2(lB(n)) ≤ lA(n) ≤ 2d2 lB(n).
Proof. Let c2 be the constant from the previous lemma, i.e., we have hA(n) ≤
c2 hB(n) for all n ∈ N. Let m be the maximal arity of operation symbols of A, and
let d2 := c2 log2(m + 1). By Lemma 2.1 (5), we have lA(n) ≤ (m + 1)hA(n). Using
the inequality on the height, we obtain
(m + 1)hA(n) ≤ (m + 1)c2 hB(n) = 2d2 hB(n).
Now by Lemma 2.1 (4), we have 2d2 hB(n) ≤ 2d2 lB(n), and therefore the second
inequality holds. For proving the first inequality, we interchange the roles of A and
B to obtain a d′ > 0 such that for all n ∈ N, lB(n) ≤ 2d′ hA(n). Hence log2(lB(n)) ≤
d′ hA(n). Hence the required inequality is satisfied with d1 := 1/d′.
(cid:3)
We are not aware of finite term equivalent algebras A and B of finite type such
that there is a positive real number c with lA(n) ≥ 2c lB(n) for all n ∈ N. A natural
example that the change of signatures can make it significantly easier to write down
certain functions is given by the commutator operation in a group. Let A = A4 to
be the alternating group on four elements, and let B be the algebra obtained from
A by adding a single binary operation [·, ·] expressing the commutator of the two
elements, i.e., [x, y] = x−1y−1xy. Observe that while the n-ary iterated commutator
tn = [. . .[[x1, x2], x3], . . . xn] has linear length in n, the corresponding term of A (the
one obtained by simply substituting the definition of the commutator for each of its
appearance) has length more than 2n. Unfortunately, we are unable to prove that
the term functions tB
n cannot be represented by terms in the language of A whose
length would be bounded by a polynomial in n. Nevertheless, using [HS12] and
under the assumption that P 6= NP, such terms cannot be produced in polynomial
time:
Proposition 2.5. Let A := (A4, ·, −1, 1) be the alternating group on four letters,
let B := (A4, ·, −1, 1, [., .]) be its expansion with the commutator operation [x, y] :=
x−1y−1xy, let t1 = x1, and for n ∈ N, let tn := [tn−1, xn]. If P 6= NP, then there
is no algorithm which, given n, produces a term sn in the language of A such that
sA
n = tB
n and which runs in polynomial time in n.
Proof. We derive this result from results of [HS12]: the equation solvability problem
for A is in P [HS12, Theorem 6]. Furthermore, we will use their reduction of 3-
colorability to the equation solvability problem for B [HS12, Theorem 13].
Suppose that there is a polynomial time algorithm producing sn. We will use
this algorithm to reduce 3-colorability to equation solvability in A. To this end, let
Γ = (V, E) be a graph with n vertices and k edges, and let V = {v1, . . . , vn} and
E = {e1, . . . , ek}. In [HS12], the authors produce a term tΓ in the language of B of
polynomial length in n and a ∈ A4 such that tΓ ≈ a has a solution for a 6= 1 if and
only if the graph Γ is 3-colorable. This term is defined by
tΓ(x1, x2, y1, . . . , yn) = [. . .[[x1, x2], ge1 ], . . . . . . , gek ],
6
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
j
where g(vi,vj ) = yiy−1
. The term tΓ is of a polynomial length in n as a term of B.
Using this term, Horv´ath and Szab´o reduce 3-colorability to equation solvability in
B. We have to do one more step to reduce it to equation solvability in A, that is
find a polynomial length term in the language of groups. For that we run the given
algorithm to produce a term sk+2 in the language of A such that sA
k+2.
Note that this term has to be of polynomial length in k since it is produced by
a polynomial time algorithm. Finally, let
k+2 = tB
sΓ(x1, x2, y1, . . . , yn) = sk+2(x1, x2, ge1 , . . . . . . , gek ).
This term can be computed from sk+2 in linear time, and since sA
Γ , the
equation tΓ ≈ a has a solution in A4 if and only if sΓ ≈ a does. Hence Γ is 3-colorable
if and only if sΓ ≈ a is solvable. This completes the reduction from 3-colorability to
equation solvability in A. By [HS12, Theorem 6], equational solvability in A is in P.
Altogether, we have produced a polynomial time algorithm for 3-colorability. Since
3-colorability is NP-complete [Kar72], this contradicts the assumption P 6= NP.
Therefore, assuming P 6= NP, such an algorithm producing sn does not exist. (cid:3)
Γ = tB
We will also use the following elementary lemma.
Lemma 2.6. Let f : N → R be a non-decreasing function, let n0 ∈ N, 0 < c < 1,
and d > 0 such that cn0 ≥ 1 and for all n ≥ n0 we have f (n) ≤ f (⌊cn⌋) + d. Then
for all n ∈ N, we have f (n) ≤ d log1/c n + f (n0).
Proof. We will prove the claim by induction on n. For n ≤ n0, the statement
follows from monotonicity of f . For the induction step, suppose that n > n0 and
the statement is true for all m < n. We obtain
f (n) ≤ f (⌊cn⌋) + d ≤ d log1/c(⌊cn⌋) + f (n0) + d
from the induction hypothesis and the monotonicity of the logarithm.
(cid:3)
≤ d(1 + log1/c(cn)) + f (n0) = d log1/c n + f (n0)
3. Primal algebras
In this section, we will consider sums x1 + x2 + · · · + xn, where + is some binary
operation. These sums can be parenthesized in various ways. We will put the
parentheses in a way that yield a balanced binary tree.
Definition 3.1. Let t be a binary symbol. We define the sequence (σt
in the language {t} by σt
1(x1) := x1 and
n)n∈N of terms
σt
n(x1, . . . , xn) := t(σt
⌈n/2⌉(x1, . . . , x⌈n/2⌉), σt
⌊n/2⌋(x⌈n/2⌉+1, . . . , xn))
for n ≥ 2.
Lemma 3.2. Let t be a binary symbol in the language of an algebra A and 0 ∈ A
such that tA(x, 0) = tA(0, x) = x for all x ∈ A. Then for all n ∈ N and x ∈ A, we
have (σt
n)A(0, . . . , 0, x, 0, . . . , 0) = x.
Proof. By induction on n. For n = 1 the statement is obviously true by the def-
inition of (σt
n)n∈N. Let n ≥ 2, and let x ∈ A. We first consider the case that
x appears at place i with i > ⌈n/2⌉. Then the equalities σt
(0, . . . , 0) = 0
and σt
the assumption tA(0, x) = x, we obtain (σt
i ≤ ⌈n/2⌉ is done similarly.
(0, . . . , 0, x, 0, . . . , 0) = x follow from the induction hypothesis. Using
n)A(0, . . . , 0, x, 0, . . . , 0) = x. The case
(cid:3)
⌈n/2⌉
⌊n/2⌋
A
A
The height of a binary balanced tree can be determined from its number of leaves.
In our setting, this means:
Lemma 3.3. Let t be a binary symbol. Then ht(σt
n) = ⌈log2 n⌉.
COMPLEXITY OF TERMS
7
Proof. We prove the statement
∀k ∈ N0 ∀n ∈ N : 2k−1 < n ≤ 2k ⇒ ht(σt
n) = k
by induction on k. It is clearly true for k ∈ {0, 1}. Now assume that k > 1. Since
2k−1 < n ≤ 2k, have 2k−2 < n/2 ≤ 2k−1, and therefore, 2k−2 < ⌈n/2⌉ ≤ 2k−1
and ⌊n/2⌋ ≤ 2k−1. From these inequalities and the induction hypothesis, we obtain
that ht(σt
⌊n/2⌋) ≤ k − 1. Hence, from the definition of σt
n,
we get ht(σt
n) = k. This completes the induction proof. Now we notice that if
2k−1 < n ≤ 2k, then k = ⌈log2 n⌉, which implies the result.
(cid:3)
⌈n/2⌉) = k − 1 and ht(σt
Lemma 3.4. Let A be a finite algebra that has binary operations + and · as fun-
damental operations as well as characteristic functions χa for all a ∈ A such that
x + 0 = 0 + x = x, x · 1 = x and x · 0 = 0 for all x ∈ A. If all unary constant
operations of A are term functions then there is a positive real number d such that
hA(n) ≤ dn for all n ∈ N.
Proof. Let n ∈ N, let A = m, m ∈ N, let A = {a1, . . . , am} and let us denote
Πn
i=1χαi (xi) = (. . . (χα1 (x1) · χα2(x2)) · . . .) · χαn (xn) by χα1(x1) · · · χαn (xn) for all
α1, . . . , αn ∈ A and x1, . . . , xn ∈ A. By Lemma 3.2 we have that arbitrary term
function f can be represented as
σ+
mn (f (a1, . . . , a1) · χa1 (x1) · · · χa1(xn), . . . , f (am, . . . , am) · χam (x1) · · · χam (xn)).
(3.1)
We note that ht(x + y) = ht(x · y) = 1 and ht(χai (x)) = 1 for all i ∈ {1, . . . , m}.
Let s be the maximal height of the terms that represent constant functions. Now we
can calculate the height of (3.1). Using Lemma 3.3, we have ht(σ+
mn ) ≤ ⌈log2 An⌉.
Using the definition of the height we obtain
ht(f (α1, . . . , αn) · Πn
i=1χαi (xi)) = 1 + max{ht(f (α1, . . . , αn)), ht(Πn
i=1χαi(xi))}
and ht(Πn
i=1χαi(xi)) = n for all α1, . . . , αn ∈ A. Hence, for all n ∈ N, we have
ht(f (α1, . . . , αn) · Πn
i=1χαi (xi)) ≤ 1 + max{s, n}.
Hence there is c ∈ N such that ht(f (α1, . . . , αn) · Πn
i=1χαi(xi)) ≤ cn for all n ∈ N.
Therefore, for each n ∈ N, by Lemma 2.2 the height of (3.1) is at most ⌈log2 An⌉ +
cn. There is a positive real d such that for all n ∈ N, ⌈log2 An⌉ + cn ≤ dn. Hence
hA(n) ≤ dn.
(cid:3)
Proposition 3.5. Let A = (A, F ) be a finite primal algebra of finite type, and let
n ∈ N. Then there is a positive real number c such that lA(n) ≤ 2cn for all n ∈ N.
Proof. Let A be a primal algebra and let u be the maximal arity of its fundamen-
tal operations. By [KP01, Theorem 3.1.5] every primal algebra A = (A, F ) such
that A = {a1, . . . , am}, is term equivalent to B = (A, +, ·, χa1 , . . . , χam, 0, 1) where
χa1 , . . . , χam are characteristic functions, 0 and 1 are elements from A and + and
· are binary operations such that x + 0 = 0 + x = x, x · 1 = x, x · 0 = 0 for all
x ∈ A and every constant function is a term function of B. By Lemma 2.3, there
is a positive real number a such that hA(n) ≤ a hB(n) for all n ∈ N. Using Lemma
3.4, we obtain a positive real d such that hA(n) ≤ dn for all n ∈ N. Using Lemma
2.1 we have lA(n) ≤ (u + 1)dn, and thus we can find a positive real c such that
lA(n) ≤ 2cn for all n ∈ N.
(cid:3)
Proof of Theorem 1.2. For every primal algebra A and n ∈ N we have SpecA(n) =
AAn
. Now the first inequality follows from Lemma 2.1 (2). The second inequality
is given in Proposition 3.5.
(cid:3)
8
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
4. Supernilpotent algebras
The notion of supernilpotency was introduced in [AE06, Definition 4.1] for ex-
pansions of groups, and in its general form in [AM10, Definition 7.1]. It is closely
related to Bulatov's higher commutators introduced in [Bul01].
Definition 4.1. An algebra A is said to be supernilpotent of degree n if it satisfies
the commutator identity [1A, . . . , 1A] = 0A where the commutator has arity n + 1.
The commutator identity in the above definition is described by the following
term condition: for all i = 1, . . . , n + 1, ki ∈ N, ai, bi ∈ Aki with ai 6= bi, and all
terms t of arity P ki that satisfy
t(x1, . . . , xn, an+1) = t(x1, . . . , xn, bn+1)
for all choices of xi's between ai and bi except the case where all are bi, we have
t(b1, . . . , bn, an+1) = t(b1, . . . , bn, bn+1).
Recently, Moorhead [Moo16] provided a condition using two terms that is equiva-
lent to the 'one term' condition given by Bulatov in the case that the algebra lies in
a congruence modular variety. This gives another simple description of supernilpo-
tency: loosely speaking, for a fixed ai and bi, the value of any term s on the tuple
(b1, . . . , bn+1) is determined by its values on all other tuples consisting of ai and
bi in the right order. The next theorem, which is based on the results in [Opr16],
shows that this unique value can be obtained as a result of a certain (2n − 1)-ary
term (a strong cube term [Opr16, p. 375]) applied to the values of all the other
tuples. We recall that every algebra A with a Mal'cev term q has a strong n-cube
term qn for every n > 1, and moreover such a term can be obtained recursively
from the Mal'cev term by q2(x, y, z) = q(y, x, z) and
qn+1(x0, . . . , x2n+1−1) = q2(qn(x0, . . . , x2n−2), x2n−1, qn(x2n , . . . , x2n+1−2)).
Also, by [Opr16, Lemma 4.1], for every n ≥ 2, every algebra with a strong n-cube
term has a Mal'cev term. By a polynomial term of the algebra A, we understand
a term of the algebra A∗, which is the expansion of A with one nullary constant
operation for every element of A.
Theorem 4.2. Let n ≥ 2 and let A be an algebra with a strong n-cube term qn.
Then the following are equivalent
(1) A is supernilpotent of degree n − 1;
(2) for all m1, . . . , mn ∈ N, for all terms t of arity m = m1 + · · · + mn, and for
all a1, b1 ∈ Am1 , . . . , an, bn ∈ Amn , we have
qA
n (cid:0)tA(a1, . . . , an), tA(b1, a2, . . . , an), . . . , tA(a1, b2, . . . , bn)(cid:1) = tA(b1, b2, . . . , bn);
(3) for all n-ary polynomial terms t and all a1, b1, . . . , an, bn ∈ A, we have
n (cid:0)tA(a1, . . . , an), tA(b1, a2, . . . , an), . . . , tA(a1, b2, . . . , bn)(cid:1) = tA(b1, b2, . . . , bn).
qA
The proof heavily relies on the properties of the relation ∆(α1, . . . , αn).
In
the case A is a Mal'cev algebra, this relation of arity 2n is described in [Opr16,
Lemma 3.3] by
∆(α1, . . . , αn) = (cid:8)(cid:0)t(a1, . . . , an), t(b1, a2, . . . , an), . . . , t(b1, b2, . . . , bn)(cid:1)
for all i ∈ {1, 2, . . . , n} : mi ∈ N0, ai, bi ∈ Ami, (ai, bi) ∈ αmi
and t ∈ CloPn
i
,
i=1 mi A(cid:9).
Proof. To simplify the notation, let ∆n denote the relation ∆(1A, . . . , 1A), where
1A appears n times, and let [1]n denote the n-ary commutator [1A, . . . , 1A].
(1) ⇒ (2): This implication is a consequence of [Opr16, Lemma 4.2]. First,
observe that (cid:0)tA(a1, . . . , an), tA(b1, a2 . . . , an), . . . , tA(b1, b2 . . . , bn)(cid:1) ∈ ∆n. From
COMPLEXITY OF TERMS
9
the mentioned lemma, we get that the last element of this tuple is [1]n-related to
the result of qn applied to all the previous elements. But since A is supernilpotent
of degree n, and therefore [1]n = 0A, this gives the desired identity.
(2) ⇒ (3): For the n-ary polynomial term t(x1, . . . , xn), there is a term s in the
language of A and there are c1, . . . , cm ∈ A such that
t(x1, . . . , xn) = s(x1, . . . , xn, c1, . . . , cm).
Now we apply (2) for the term s and for a1 := a1, b1 := b1, . . . , an−1 := an−1,
bn−1 := bn−1, an := (an, c1, . . . , cm), bn := (bn, c1, . . . , cm).
(3) ⇒ (1): We will prove that the condition (3) implies that for any n-tuple of
principal congruences θ1, . . . , θn, we have [θ1, . . . , θn] = 0A. The claim then follows
from join distributivity of the higher commutator. Suppose that θi = Cg (ai, bi)
for all i. We know that the relation ∆(θ1, . . . , θn) [Opr16, Lemma 3.3] consists of
tuples of the form
(cid:0)t(a1, . . . , an), t(b1, a2, . . . , an), . . . , t(b1, b2, . . . , bn)(cid:1)
where ai ≡θi bi and t is a term operation of A. Since A is a Mal'cev algebra, and
therefore any reflexive binary compatible relation is a congruence, and since δi is
generated by (ai, bi), every pair (c, d) ∈ θi is of the form (pA(ai), pA(bi)), where p
is a polynomial term of A. Hence, ai and bi are of the form (ti1(ai), . . . , timi (ai))
and (ti1(bi), . . . , timi (bi)) for some unary polynomial operations tij. By composing
these polynomials with t, we obtain
∆(θ1, . . . , θn) = {(cid:0)p(a1, . . . , an), p(b1, a2, . . . , an), . . . , p(b1, b2, . . . , bn)(cid:1)
p is an an n-ary polynomial operation of A}.
By combining this observation with (3), we obtain that the last coordinate of a tu-
ple in ∆(θ1, . . . , θn) is determined by the other coordinates, therefore by [Opr16,
Theorem 1.2], we get that [θ1, . . . , θn] = 0A, as required.
(cid:3)
It is a consequence of [BM14, Lemma 2.7] (which builds upon Lemma 14.6 of
[FM87]) that a finite supernilpotent algebra in a congruence permutable variety has
a finitely generated clone of term operations. Theorem 4.2 provides another way of
establishing this fact.
Corollary 4.3 (cf. [BM14, Lemma 2.7]). Let n ∈ N, and let A be a supernilpotent
Mal'cev algebra of degree n. Then the clone of term operations is generated by the
Mal'cev term operation together with all term operations of arity at most n + 1.
Proof. By induction on k, we show that every k-ary term operation of A can be
generated. We use [Opr16, Lemma 4.1] to produce a strong cube term qn+1 of arity
2n+1 − 1 for A. Let k ≥ n + 2, and let f (x1, x2, . . . , xk) be a k-ary term operation.
We set b1 = x1, . . . , bn = xn, bn+1 = (xn+1, xn+2, . . . , xk), a1 = · · · = an = xn+1,
and an+1 to the (k − n)-tuple (xn+1, . . . , xn+1). Theorem 4.2 (2) implies
qA
n+1(f (xn+1, xn+1, . . . , xn+1, . . . , xn+1), f (x1, xn+1, . . . , xn+1, . . . , xn+1), . . . ,
f (xn+1, x2, . . . , xn+1, . . . , xk)) = f (x1, . . . , xk).
Each of the 2n+1 − 1 arguments of qA
n+1 contains at least two occurrences of xn+1
and is therefore of essential arity at most k − 1. By the induction hypothesis,
each of these arguments describes a function that lies in the clone generated by the
Mal'cev operation and the (n + 1)-ary functions. Since qn+1 is composed from the
Mal'cev term, f can be generated by the Mal'cev term and functions of arity at
most n + 1.
(cid:3)
This generalizes [AM10, Proposition 6.18] to clones that do not contain all con-
In contrast to the constantive case, term functions of arity n
stant operations.
10
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
may not suffice: as an example consider the clone C on the set M2×2(Z2) of
2 × 2 matrices over Z2 that contains all functions (X1, . . . , Xk) 7→ Pk
i=1 AiXi with
A1, . . . , Ak ∈ M2×2(Z2) and Pk
i=1 Ai = 1. Then the algebra A = (M2×2(Z2), C) is
1-supernilpotent, but C is not generated by the identity mapping and the unique
Mal'cev operation in the clone.
The condition (2) in Theorem 4.2 also provides an explicit, though infinite, set
of identities that defines supernilpotency in a Mal'cev variety:
Corollary 4.4. Let V be a variety with a strong cube term qn. Then the class of
all supernilpotent algebras of degree n forms a subvariety of V. This subvariety is
defined by the collection of identities of the form
qn(cid:0)t(x1, . . . , xn), t(y1, x2, . . . , xn), . . . , t(x1, y2, . . . , yn)(cid:1) ≈ t(y1, y2, . . . , yn),
where t is a term of arity k ≥ n and k1, . . . , kn ∈ N are such that k1 + · · · + kn = k.
Here, xi and yi denote the tuples of variables (xi1, . . . , xiki ) and (yi1, . . . , yiki ),
respectively.
We will now use these identities to express terms of higher arity using the strong
cube term qk and terms of smaller arity. This method allows us to prove that there
is a logarithmic bound on the sequence hA(n) for every supernilpotent finite algebra
with a Mal'cev term, and as a consequence, we obtain a polynomial bound on lA(n).
Theorem 4.5. Let A be a finite supernilpotent Mal'cev algebra. Then there exist
positive real numbers c1, c2 such that for all n ∈ N, hA(n) ≤ c1 log n + c2.
Proof. Let k be 1 plus the degree of supernilpotency of A, and let qk be a strong
cube term of A. Further assume that ht(qk) = d. We will prove that there is
a constant c < 1 such that for any large enough n, we have hA(n) ≤ d + hA(⌊cn⌋).
To do that, we start with a term f in the language of A of high-enough arity n, and
we will group its variables into k pieces of almost the same length, and then use the
identity from item (2) in Theorem 4.2 to replace f by a composition of the strong
cube term qk with terms of arity lower than n. More precisely, let n = qk + r where
r < k, q > 1. We group the variables of f into r many (q + 1)-tuples x1, . . . , xr and
(k − r) many q-tuples xr+1, . . . , xk so that x1 = (x1, . . . , xi1 ), x2 = (xi1+1, . . . , xi2 ),
etc. We take a new variable y, and for i ∈ {1, . . . , k}, we let yi denote the tuple
(y, . . . , y) of the same length as xi (i.e., yi is a (q + 1)-tuple for i ≤ r and a q-tuple
for i > r). Now applying the condition (2) of Theorem 4.2, we get that
f (x1, . . . , xk) ≈ qk(f (y1, x2, . . . , xk), . . . , f (x1, y2, . . . , yk))
(4.1)
is satisfied in A. The right hand side is an application of qk on terms obtained from
f by substituting one or more of xi's by yi. The maximal arity of these 2k − 1
terms is obtained, e.g., when only xk is substituted by yk. In this case, omitting xk
reduces the arity of f by q = ⌊ n
k ⌋ and adds 1 for the new variable y. Hence, each
of the 2k − 1 arguments of qk in (4.1) contains at most n + 1 − ⌊ n
k ⌋ many different
variables. For each of these 2k − 1 arguments, we pick a term ui of height at most
hA(n + 1 − ⌊ n
k ⌋) representing the same function on A. From Lemma 2.2, we obtain
that qk(u1, . . . , u2k−1) is a term of height at most d + hA(n + 1 − ⌊ n
k ⌋) that induces
the same function on A as f . Therefore,
hA(n) ≤ d + hA(n + 1 − ⌊
n
k
⌋)
for every n ∈ N. We choose ǫ ∈ R such that 0 < ǫ < 1/k, we set c = 1 − 1/k + ǫ,
and let n0 > k be big enough so that ǫn0 ≥ 2 and cn0 ≥ 1. Then for any n ≥ n0,
we have
n + 1 − ⌊
⌋ < n −
+ 2 ≤ (1 −
+ ǫ)n = cn.
n
k
n
k
1
k
COMPLEXITY OF TERMS
11
Therefore, hA(n) ≤ d + hA(⌊cn⌋) for any n ≥ n0. From Lemma 2.6, we obtain
that hA(n) ≤ d log1/c n + hA(n0) for all n ∈ N. Choosing c1 := d/ log(1/c) and
c2 = hA(n0) we obtain the required result.
(cid:3)
Corollary 4.6. Let A be a finite supernilpotent Mal'cev algebra, then there exist
an integer k > 0 and a positive real c such that for all n ∈ N, lA(n) ≤ cnk.
Proof. The algebra A need not be of finite type. However, by Corollary 4.3, its
clone of term operations is finitely generated. Therefore, we can choose a finite
subset of the fundamental operations of A that generates all other fundamental
operations, and we let A′ be the reduct of A with only these finitely many fun-
damental operations;
let m be their maximal arity. By Theorem 4.5, we have
hA′ (n) ≤ c1 log(n) + c2. Now Lemma 2.1 yields lA′ (n) ≤ (m + 1)hA′ (n), and thus
there is a positive real c and k ∈ N such that for all n ∈ N, lA′ (n) ≤ cnk. We clearly
have lA(n) ≤ lA′ (n) for all n ∈ N which implies the result.
(cid:3)
Proof of Theorem 1.4. Since the variety generated by A is locally finite and has
a weak difference term, using [Wir17, Theorem 4.8] and its proof, we get that A
has a Mal'cev term. Corollary 4.6 now yields the second inequality. For the first
equality, we show that for every k ∈ N0, there is an (2k + 1)-ary term tk such
that tA depends on all of its arguments. To this end, let m be a Mal'cev term, let
t0(x1) = x1, and tk(x1, . . . , x2k+1) = m(tk−1(x1, . . . , x2k−1), x2k, x2k+1) for k ∈ N.
Let a, b be different elements of A. We consider y := tA
k (a, . . . , a, b, . . . , b) where
the first 2k + 1 − l arguments are a and the remaining l arguments are set to b.
Then y = a if l is even, and y = b otherwise. This proves that tk depends on
all of its 2k + 1 arguments. We will now show lA(n) ≥ n − 1:
if n is even, then
tA
(n−2)/2(x1, . . . , xn−1) depends on n − 1 arguments. Hence every term representing
tA
(n−2)/2 must contain at least n − 1 variables, and is thus of length at least n − 1.
If n is odd, then tA
(n−1)/2(x1, . . . , xn) depends on all of its n arguments, and thus
lenA(tA
(n−1)/2) ≥ n.
(cid:3)
In the rest of this section, we give an argument that out of finite algebras in
congruence modular varieties, only supernilpotent ones have a polynomial bound
on the length of term functions. This argument is based on a description of the
sequence SpecA(n). A rough asymptotic behavior of this sequence for congruence
modular algebras have been first described by Kearnes in [Kea99]. He proved that
an algebra of finite type in a congruence modular variety has a doubly exponential
lower bound if and only if it is not a product of prime-power order nilpotent algebras
which is now known to be equivalent to being supernilpotent [AM10]. We present
a refinement of this result which is given by a combination of several different
sources.
Proposition 4.7. Let A be a finite algebra in a congruence modular variety, and
let k ∈ N. Then A is k-supernilpotent if and only if there is a polynomial p of degree
k such that for all n ∈ N, SpecA(n) ≤ 2p(n).
Proof. For the "if"-part, first observe that 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 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 (in the sense
of [Kea99, p. 179]) of A∗ are of rank at most k. Hence all commutator polynomials
12
ERHARD AICHINGER, NEBOJSA MUDRINSKI, AND JAKUB OPRSAL
(in the sense of [AM10, Definition 7.2]) of A are of rank at most k, and then [AM10,
Lemma 7.5] yields that A is k-supernilpotent.
For the "only if"-part, we assume that A is k-supernilpotent. Then from the
proof of [Wir17, Theorem 4.8], it follows that A has a Mal'cev term, and thus by
Lemma 7.5 of [AM10] each commutator term of A is of rank at most k. Now from
the proof of Theorem 1 in [BB87], we obtain a polynomial p of degree at most k
such that for all n ∈ N, FA(n) has exactly 2p(n) elements.
(cid:3)
Section 4 of [Aic14] contains a self-contained version of Proposition 4.7 for the
case that A is an expanded group.
Corollary 4.8. Let A be a finite algebra of finite type in a congruence modular
variety. Then the following are equivalent.
(1) A is supernilpotent;
(2) there exists constants c1, c2 > 0 such that hA(n) ≤ c1 log n + c2 for all
n > 0;
(3) there exists a polynomial p1 such that lA(n) ≤ p1(n) for all n > 0;
(4) there exists a polynomial p2 such that SpecA(n) ≤ 2p2(n) for all n > 0.
Proof. (1) ⇒ (2): as noted in the proof of Proposition 4.7, a finite supernilpotent
algebra has a Mal'cev term, therefore Theorem 4.5 applies in this case.
The implications (2) ⇒ (3) and (3) ⇒ (4) are given by Lemma 2.1, and (4) ⇒ (1)
(cid:3)
is implied by Proposition 4.7.
Acknowledgements. The authors would like to thank ´Agnes Szendrei for supply-
ing the argument leading to Lemma 2.1 (3).
References
[AE06] Erhard Aichinger and Jurgen Ecker. Every (k +1)-affine complete nilpotent group of class
k is affine complete. Internat. J. Algebra Comput., 16(2):259 -- 274, 2006.
[Aic14] Erhard Aichinger. On the direct decomposition of nilpotent expanded groups. Comm.
Algebra, 42(6):2651 -- 2662, 2014.
[AM10] Erhard Aichinger and Nebojsa Mudrinski. Some applications of higher commutators in
[BB87]
Mal'cev algebras. Algebra Universalis, 63(4):367 -- 403, 2010.
Joel Berman and W. J. Blok. Free spectra of nilpotent varieties. Algebra Universalis,
24(3):279 -- 282, 1987.
[BM14] Wolfram Bentz and Peter Mayr. Supernilpotence prevents dualizability. J. Aust. Math.
[BS81]
Soc., 96(1):1 -- 24, 2014.
Stanley Burris and Hanamantagouda P. Sankappanavar. A course in universal algebra,
volume 78 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1981.
[Bul01] Andrei Bulatov. On the number of finite Mal'tsev algebras. In Contributions to general
algebra, 13 (Velk´e Karlovice, 1999/Dresden, 2000), pages 41 -- 54. Heyn, Klagenfurt, 2001.
[FM87] Ralph Freese and Ralph McKenzie. Commutator Theory for Congruence Modular vari-
eties, volume 125 of London Math. Soc. Lecture Note Ser. Cambridge University Press,
1987.
[HM88] David Hobby and Ralph McKenzie. The structure of finite algebras, volume 76 of Con-
temporary mathematics. American Mathematical Society, 1988.
[HN15] G´abor Horv´ath and C. L. Nehaniv. Length of polynomials over finite groups. J. Comput.
System Sci., 81(8):1614 -- 1622, 2015.
[HS12] G´abor Horv´ath and Csaba Szab´o. Equivalence and equation solvability problems for the
alternating group a4. Journal of Pure and Applied Algebra, 216(10):2170 -- 2176, 2012.
[Kar72] Richard M. Karp. Reducibility among combinatorial problems. In Complexity of Com-
puter Computations (Proc. Sympos., IBM Thomas J. Watson Res. Center, Yorktown
Heights, N.Y., pages 85 -- 103. Plenum, New York, 1972.
[Kea99] Keith A. Kearnes. Congruence modular varieties with small free spectra. Algebra Uni-
versalis, 42(3):165 -- 181, 1999.
[KP01] Kalle Kaarli and Alden F. Pixley. Polynomial completeness in algebraic systems. Chap-
man & Hall / CRC, Boca Raton, Florida, 2001.
[Moo16] Andrew Moorhead. Higher commutator theory for congruence modular varieties. ArXiv
e-prints, 2016. https://arxiv.org/abs/1610.07087.
COMPLEXITY OF TERMS
13
[Opr16] Jakub Oprsal. A relational description of higher commutators in Mal'cev varieties. Algebra
universalis, 76(3):367 -- 383, 2016.
[Wir17] Alexander Wires. On
supernilpotent
algebras.
ArXiv
e-prints,
2017.
https://arxiv.org/abs/1701.08949.
Institut fur Algebra, Johannes Kepler Universitat Linz, 4040 Linz, Austria
E-mail address: [email protected]
Department of Mathematics and Informatics, Faculty of Sciences, University of Novi
Sad, Trg Dositeja Obradovi´ca 4, 21000 Novi Sad, Serbia
E-mail address: [email protected]
Institut fur Algebra, Technische Universitat Dresden, 01062 Dresden, Germany
E-mail address: [email protected]
|
1506.04475 | 1 | 1506 | 2015-06-15T04:21:57 | Fuzzy derivations KU-ideals on KU-algebras | [
"math.RA"
] | In this manuscript, we introduce a new concept, which is called fuzzy left (right) derivations KU- ideals in KU-algebras. We state and prove some theorems about fundamental properties of it. Moreover, we give the concepts of the image and the pre-image of fuzzy left (right) derivations KU-ideals under homomorphism of KU- algebras and investigated some its properties. Further, we have proved that every the image and the pre-image of fuzzy left (right) derivations KU-ideals under homomorphism of KU- algebras are fuzzy left (right) derivations KU-ideals | math.RA | math | Fuzzy derivations KU-ideals on KU-algebras
BY
Samy M.Mostafa 1 , Ahmed Abd-eldayem2
[email protected] [email protected]
1,2Department of mathematics -Faculty of Education -Ain Shams University Roxy, Cairo, Egypt
Abstract. In this manuscript, we introduce a new concept, which is called fuzzy
left (right) derivations KU- ideals in KU-algebras. We state and prove some
theorems about fundamental properties of it. Moreover, we give the concepts of the
image and the pre-image of fuzzy left (right) derivations KU-ideals under
homomorphism of KU- algebras and investigated some its properties. Further, we
have proved that every the image and the pre-image of fuzzy left (right) derivations
KU-ideals under homomorphism of KU- algebras are fuzzy left (right) derivations
KU-ideals. Furthermore, we give the concept of the Cartesian product of fuzzy left
(right) derivations KU - ideals in Cartesian product of KU – algebras.
AMS Subject Classification: 03G25, 06F35
Keywords. KU-algebras ,fuzzy left (right) derivations of KU-ideals , the image
and the per- image of fuzzy left (right) derivations KU – ideals, the Cartesian
product of fuzzy left (right) derivations KU – ideals.
Corresponding Author : Samy M. Mostafa ( [email protected] )
1. Introduction
As it is well known, BCK and BCI-algebras are two classes of algebras of logic.
They were introduced by Imai and Iseki [10,11,12] and have been extensively
investigated by many researchers. It is known that the class of BCK-algebras is a
proper sub class of the BCI-algebras.The class of all BCK-algebras is a
quasivariety. Is´eki posed an interesting problem (solved by Wro´nski [24])
whether the class of BCK-algebras is a variety. In connection with this problem,
Komori [15] introduced a notion of BCC-algebras, and Dudek [7] redefined the
notion of BCC-algebras by using a dual form of the ordinary definition in the
sense of Komori. Dudek and Zhang [8] introduced a new notion of ideals in BCC-
algebras and described connections between such ideals and congruences .
1
C.Prabpayak and U.Leerawat ( [22 ], [23 ]) introduced a new algebraic structure
which is called KU - algebra . They gave the concept of homomorphisms of KU-
algebras and investigated some related properties. Several authors [2,3 ,5 ,6,9,14]
have studied derivations in rings and near rings. Jun and Xin [13] applied the
notion of derivations in ring and near-ring theory to BCI-algebras, and they also
introduced a new concept called a regular derivation in BCI -algebras. They
investigated some of its properties, defined a d -derivation ideal and gave
conditions for an ideal to be d-derivation. Later, Hamza and Al-Shehri [1],
defined a left derivation in BCI-algebras and investigated a regular left derivation.
Zhan and Liu [27 ] studied f-derivations in BCI-algebras and proved some results.
( -derivation in a BCI-
G. Muhiuddin etl [20,21] introduced the notion of
algebra and investigated related properties. They provided a condition for a
( - derivation to be regular. They also introduced the concepts of a
-
)
)
,
,
( d
)
,
,
,
)
)
( -derivation and α-ideal, and then they investigated their relations.
invariant
( - derivations.
Furthermore, they obtained some results on regular
Moreover, they studied the notion of t-derivations on BCI-algebras and obtained
some of its related properties. Further, they characterized the notion of p-semi-
simple BCI-algebra X by using the notion of t-derivation. Later, Mostafa et al
[18,19], introduced the notions of
) -derivation of a KU-algebra and
some related properties are explored. The concept of fuzzy sets was introduced by
Zadeh [26]. In 1991, Xi [25] applied the concept of fuzzy sets to BCI, BCK, MV-
algebras .Since its inception, the theory of fuzzy sets , ideal theory and its
fuzzification has developed in many directions and is finding applications in a
wide variety of fields. Mostafa et al, in 2011[17] introduced the notion of fuzzy
KU-ideals of KU-algebras and then they investigated several basic properties
which are related to fuzzy KU-ideals.
r
),(
),( r
-(
In this paper, we introduce the notion of fuzzy left (right) derivations KU- ideals
in KU - algebras. The homomorphic image ( preimage) of fuzzy left (right) -
derivations KU- ideals in KU - algebras under homomorprhism of a KU-algebras
are discussed . Many related results have been derived.
2
2. Preliminaries
In this section, we recall some basic definitions and results that are needed for
our work.
Definition 2.1 [22,23 ] Let X be a set with a binary operation and a constant
:
0. (
) is called KU-algebra if the following axioms hold :
Xzyx
,
,X
0,
,
(
KU
)
1
(
x
y
)
[(
y
z
)
(
zx
)]
0
(
KU
)
2
x
0
0
(
KU
3
0)
x
x
(
KU
)
4
xif
xy
0
y
implies
x
y
Define a binary relation by :
ais
partially
ordered
set
.
x
0
x
y
y
, we can prove that
(
X
,
)
Throughout this article, X will denote a KU-algebra unless otherwise mentioned
Corollary 2.2 [17,22] In KU-algebra the following identities are true for
all
Xzyx
,
,
:
(i)
z
0 z
(ii)
z
(
x
z
)
0
(iii) If
x
y
implies
that
y
(v)
z
(
y
x
)
y
(
z
x
)
(vi)
y
[(
y
x
)
x
0]
x
z
z
Definition 2.3 [22,23] A subset S of KU-algebra X is called sub algebra of X if
x
, whenever
Syx ,
S
y
Definition 2.4 [22,23 ] Anon empty subset A of KU-algebra X is called ideal of
X if it is satisfied the following conditions:
(ii)
y
AyAz
,
(i)
A0
implies
Az
3
Xzy
,
.
Definition 2.5 [17] A non - empty subset A of a KU-algebra X is called an KU ideal of X
if it satisfies the following conditions :
(1) 0 A ,
(2) x * (y * z)A , y A implies x * z A , for all x , y , zX
Definition 2.6[17] Let X be a KU - algebra, a fuzzy set µ in X is called fuzzy sub-algebra
if it satisfies:
(S1) µ (0) ≥ µ (x) ,
(S2) µ (x) ≥ {µ (x * y) , µ (y)} for all x , y X .
Definition 2.7 [17] Let X be a KU-algebra , a fuzzy set µ in X is called a fuzzy KU-ideal
of X if it satisfies the following conditions:
(F1) µ (0) ≥ µ (x) ,
(F2) µ (x * z) ≥ min {µ (x * (y * z)), µ (y)}.
Definition 2.8 For elements x and y of KU-algebra (
x
.
y
x
y
)
y
(
,X
0,
), we denote
Definition 2.9[18] Let X be a KU-algebra. A self map
derivation (briefly,
r
),(
-derivation) of X if it satisfies the identity
Xd :
X
is a left –right
xd
(
y
)
xd
)((
y
)
(
ydx
(
))
,
Xyx
If d satisfies the identity
xd
(
y
)
(
ydx
(
))
xd
)((
y
)
,
Xyx
d is called right-left derivation (briefly,
r
),(
derivation
r
),(
then
and
d
),( r
-derivation) of X . Moreover, if d is both
is called a derivation of X .
Definition 2.10[18] A derivation of KU-algebra is said to be regular if
d
0)0(
.
4
4.3,2,1,0
Lemma 2.11[18] A derivation d of KU-algebra X is regular.
Example 2.12 [18] Let X= {
follows:
4
4
4
4
4
0
} be a set in which the operation is defined as
0
1
0
1
0
0
0
0
0
0
0
1
2
2
2
0
0
1
0
1
2
3
4
3
3
2
1
0
1
Using the algorithms in Appendix A, we can prove that (X, *, 0) is a KU-algebra. Define
a map
Xd :
by
X
xd
)(
0
4
xif
xif
3,2,1,0
4
Then it is easy to show that d is both a
r
),(
and
),( r
-derivation of X .
Example 2.12. Let
0
be the set of all positive integers and 0 . The operation (* )
on
0
is defined as follows: x*y=y–x ,where "–" the minus operation .Define a
y
x
by :
x
. Then
0
binary relation on
0
KU-algebra. We define a map
,
d(x*y)=d(y−x)=y−x−1……………………………………………………….(I),
d(x)*y =y − d(x)=y− (x−1) =1+ y – x and x*d(y) =d(y) –x =y–1–x=y–x–1, but
by d (x ) =x −1 for all
0
.Then
,we have
(
Xd :
Xyx
x
y
X
X
)0,*,
is a
d(x)*y x*d(y) =((1+y –x )* (y–x–1))* (y–x–1)= (y–x–1) –[(y–x–1) –(1+y–x )]=
= y+1– x…………………………… (II)
From (I) and (II), d is not
On other hand
r
),(
derivation of X.
5
x*d(y)= d(y) −x=y−x−1 , d(x)*y= y− d(x) =y−(x−1)=y+1−x , but
x*d(y) d(x)*y = [ (x*d(y))* (d(x)*y)]* (d(x)*y)= ( d(x) −y) −[(y −d(x)) – (d(y) –x)]
= y−x−1…………………………. (III)
From (I) and (III), d is
r
),(
derivation are not coincide .
),( r
derivation of X . Hence
),( r
-derivation and
Proposition 2.13[18] Let X be a KU-algebra with partial order , and let d be a
derivation of X . Then the following hold
Xyx
,
:
(i)
xd
)(
x
.
(ii)
xd
(
y
)
xd
)(
y
.
(iii)
xd
(
y
)
ydx
)(
.
(v)
xdxd
(
(
))
0
.
(vi)
d
{)0(1
xdXx
)(
}0
is a sub algebra of X .
Definition 2.14 [18] Let X be a KU-algebra and d be a derivation of X .
Denote
Fixd
(
X
{)
xdXx
)(
:
x
}
.
Proposition 2.15[18] Let X be a KU-algebra and d be a derivation of X .Then
Fixd
(X
)
is a sub algebra of X .
6
3. Fuzzy derivations KU- ideals of KU-algebras
In this section, we will discuss and investigate a new notion called fuzzy- left derivations
KU - ideals of KU - algebras and study several basic properties which are related to fuzzy
left derivations KU - ideals.
Definition 3.1 Let X be a KU-algebra and
Xd :
X
be self map .A non - empty subset
A of a KU-algebra X is called left derivations KU ideal of X if it satisfies the following
conditions:
(1) 0 A,
(2) d(x) * (y * z)A , d(y) A implies d(x * z ) A , for all x , y , zX
Definition 3.2 Let X be a KU-algebra and
Xd :
X
be self map .A non - empty subset
A of a KU-algebra X is called right derivations KU ideal of X if it satisfies the following
conditions:
(1) 0 A,
(2) x * d(y * z)A , d(y) A implies d(x * z ) A , for all x , y , zX.
Definition 3.3 Let X be a KU-algebra and
be self map .A non - empty subset
Xd :
X
A of a KU-algebra X is called derivations KU -ideal of X if it satisfies the following
conditions:
(1) 0 A,
(2) d(x * (y * z))A , d(y) A implies d(x * z ) A , for all x , y , zX
Definition 3.4 Let X be a KU-algebra and
be self map. A fuzzy set
in X is called a fuzzy left derivations KU-ideal(briefly,
-derivation) of X if it satisfies the following conditions:
: X
F
),
(
Xd :
X
]1,0[
(F1) µ (0) ≥ µ (x),
(FL2) µ (d(x * z)) ≥ min{ µ(d(x)*(y*z)), µ (d(y))} .
7
Definition 3.5 Let X be a KU-algebra and
in X is called a fuzzy right derivations KU-ideal(briefly,
be self map. A fuzzy set
-derivation)
Xd :
rF
),
X
(
]1,0[
: X
of X if it satisfies the following conditions:
(F1) µ (0) ≥ µ (x).
(FR2) µ (d(x * z)) ≥ min { µ(x*d(y*z)), µ (d(y))}.
Definition 3.6 Let X be a KU-algebra and
be self map. A fuzzy set
Xd :
X
]1,0[
in X is called a fuzzy derivations KU-ideal ,if it satisfies the following
: X
conditions
(F1) µ (0) ≥ µ (x).
(F2) µ (d(x * z)) ≥ min{ µ(d(x* (y*z)), µ (d(y))}.
Remark3.7 (I) If d is fixed , the definitions (3.1, 3.2 ,3.3) gives the definition KU-ideal.
(II) If d is fixed , the definitions (3.4,3.5, 3.6) gives the definition fuzzy
KU-ideal.
Example 3.8 Let X = {
Using the algorithms in Appendix A, we can prove that (X, *, 0) is a KU-algebra.
} be a set in which the operation is defined as follows:
4.3,2,1,0
Define a self map
*
0
1
2
3
4
0
0
0
0
0
0
1
1
0
1
0
0
2
2
2
0
0
0
3
3
3
1
0
0
4
4
3
4
3
0
Xd :
X
by
8
xd
)(
0
4
xif
xif
3,2,1,0
4
.
Define a fuzzy set µ : d(X) → [0,1] ,by d(µ(0)) = t0 , µ (d(1)) =µ (d(2)) = t1 , µ (d(3)) =
µ (d (4)) = t2 , where t0 , t1 , t2 [0,1] with t0 > t1 > t2 .Routine calculations give that µ is
a fuzzy left (right)- derivations KU- ideal of KU- algebra X.
Lemma 3.9 Let µ be a fuzzy left derivations KU - ideal of KU - algebra X , if the
inequality , x*y≤ d( z) holds in X, then µ (d(y)) ≥ min {µ (d(x)) , µ (z) } .
Proof. Assume that the inequality x*y≤ d( z) holds in X , then
d(z )* (x * y) = 0 , (z )* (x * y) = 0 , since d(z) ≤z from (Proposition 2.13(i)) and by(FL2),
we have µ (d(z * y)) ≥ min{ µ(d(z)*(x*y)),µ (d(x))}= min{ µ(0), µ (d(x))}= µ (d(x))
Put z=0 , we have µ (d(0 * y)) = µ (d(y)) ≥ min{ µ(x*y),µ (d(x))}………… (i),
but µ (x * y) ≥ min {µ (x * (z * y) , µ (z)} = min {µ (z * (x * y)) , µ (z)}
=min {µ (0) , µ (z)} = µ (z) ………(ii)
From (i) , (ii) , we get µ (d(y)) ≥ min {µ (z) , µ (d(x))}, this completes the proof.
Lemma 3.10 If µ is a fuzzy left derivations KU - ideal of KU - algebra X and if x ≤ d(y) ,
then μ(d(x)) ≥ μ (d(y)).
Proof . If x ≤ d(y),then d(y) * x = 0 , y*x=0 since d(y) ≤ y (from Proposition 2.13(i)) this
together with 0 * x = x and μ(0) ≥ μ (y),we get
µ (d(0 * x)) = µ (d(x)) ≥ min {µ (d(0) * (y * x)) , µ (d(y))} = min {µ (0 * 0) , µ (d(y))} =
= min {µ (0),µ (d(y))} = µ (d(y)).
Proposition 3.11 The intersection of any set of fuzzy left derivations KU - ideals of KU –
algebra X is also fuzzy left derivations KU - ideal.
Proof. let i be a family of fuzzy left derivations KU - ideals of KU- algebra X , then
for any x , y , z ∈ X ,
(
))0(
xd
((
and
)0)(
)(
xd
(
))
inf(
inf(
i
i
i
)))
(
i
9
(
)(
i
min
inf(
inf(
zxd
))*(
i
xd
zy
*(*)(
((
i
zxd
))*((
inf(min
yd
((
inf(
))),
))
i
yd
xd
zy
((
))
*(*)(
((
i
i
xd
zy
(min
*(*)(
))),
(
i
i
)((
)),
)(
yd
(
.))
This completes the proof .
Lemma 3.12 The intersection of any set of fuzzy right derivations KU - ideals of KU –
algebra X is also fuzzy right derivations KU - ideal.
proof. Clear
Theorem3.13 Let µ be a fuzzy set in X then µ is a fuzzy left derivations KU- ideal of X if
and only if it satisfies :
For all α∈[0,1]),U (μ , α) ≠ φ implies U(μ ,α) is KU- ideal of X……… (A),
where U (μ , α) = {x ∈ X / μ (d(x)) ≥ α} .
Proof . Assume that µ is a fuzzy left derivations KU- ideal of X , let α ∈ [0 , 1] be such
that U (μ , α) ≠ φ , and x , y ∈ X such that x ∈ U (μ , α) , then µ (d(x)) ≥ α and so by
(FL2 ) , µ (d( y * 0)) = µ (d(0)) ≥ min { µ ( d(y )* (x * 0) ) , µ (d(x))}=
min{µ (d(y) * 0), µ (d(x))} = min {µ (0) , µ (d(x))} = α , hence 0 ∈ U (μ , α) .
Let d(x) * (y * z) ∈ U (μ, α ) , d(y) ∈ U (μ, α), It follows from(FL2) that
µ (d(x * z)) ≥ min {µ (d(x) * (y * z)) , µ (d(y))} = α , so that x * z ∈ U (µ, α) .
Hence U (μ, α ) is KU - ideal of X .
Conversely, suppose that µ satisfies (A) , let x , y , z ∈ X be such that
µ (d(x * z)) < min {µ (d(x) * (y * z)) , µ (d(y))},taking
β0 = 1/2 {µ (d(x * z)) + min {µ (d(x) * (y * z)) , µ(d(y)) } , we have
β0 ∈ [0,1] and µ ( d(x * z)) < β0 < min {μ (d(x) * (y * z)) , µ(d(y)) } ,it follows that
d(x) * (y * z) ∈ U (μ, β0) and d(x * z) ∈ U (μ, β0) , this is a contradiction and therefore µ
is a fuzzy left derivations KU - ideal of X .
Theorem3.14 Let µ be a fuzzy set in X then µ is a fuzzy right derivations KU- ideal of X
if and only if it satisfies : For all α∈[0,1]),U (μ , α) ≠ φ implies U(μ ,α) is KU- ideal of X.
Proposition 3.15 If µ is a fuzzy left derivations KU - ideal of X , then
µ (d(x) * (x * y)) ≥ µ (d(y))
10
proof . Taking z = x * y in (FL2) and using (ku2) and (F1) , we get
µ(d(x) * (x * y)) ≥ min { µ (d(x) * (y * (x * y)) , µ(d(y)) } = min {µ(d(x) * (x * (y * y)) ,
µ(d(y)) } = min {µ( d(x) * (x * 0)) , µ(d(y)) }= min {µ (0) , µ (d(y)) } = µ (d(y)).
Definition3.16 Let µ be a fuzzy left derivations KU - ideal of KU - algebra X ,.the KU -
ideals
t , t∈ [0,1] are called level KU - ideal of µ .
Corollary3.17 Let I be an KU - ideal of KU - algebra X , then for any fixed number t in
an open interval (0,1) , there exist a fuzzy left derivations KU – ideal µ of X such that
t = I .
proof. The proof is similar the corollary 4.4 [17] .
4 Image (Pre-image) of fuzzy derivations KU-ideals under
homomorphism
In this section, we introduce the concepts of the image and the pre-image of fuzzy left
derivations KU-ideals in KU-algebras under homomorphism.
Definition 4.1 Let f be a mapping from the set X to a set Y. If is a fuzzy subset of X,
then the fuzzy subset β of Y defined by
f
(
)y)(
)y(
sup
1
f
y
(
)
x
0
if ),x(
f
1
f
, Xx{)y(
)x(
}y
otherwise
is said to be the image of under f.
Similarly if β is a fuzzy subset of Y , then the fuzzy subset µ = β f in X ( i.e the fuzzy
subset defined by µ (x) = β (f (x)) for all x X ) is called the primage of β under f .
11
Theorem 4.2 An onto homomorphic preimage of a fuzzy left derivations KU - ideal is
also a fuzzy left derivations KU - ideal .
Proof.Let f : X → X` be an onto homomorphism of KU - algebras , β a fuzzy left
derivations KU - ideal of X` and µ the preimage of β under f , then β (f (d(x)) = µ
(d(x)) , for all x X . Let x X , then µ (d(0)) = β (f(d(0))) ≥ β (f (d(x))) = µ (d(x)).
Now let x , y , z X , then µ (d(x * z)) = β (f (d(x * z))) ≥
min {β(f (d(x)) *` (f (y) *` f(z)), β(f (d(y)))} =
min { β (f(d(x)*(y * z)),β (f (d(y)) }=
min {µ(d(x) * (y * z))) , µ(d (y))} .
The proof is completed.
Definition 4.3 [4 ] A fuzzy subset µ of X has sup property if for any subset T of X , there
exist t0 T such that ,
t
( 0
)
t
)(
.
SUP
Tt
Theorem 4.4 Let f : X → Y be a homomorphism between KU - algebras X and Y .
For every fuzzy left derivations KU - ideal µ in X , f (µ) is a fuzzy left derivations KU -
ideal of Y .
Proof. By definition
yd
((
))
f
)(
(
yd
(
))
sup
1
f
((
xd
)(
yd
(
))
xd
((
))
for all
y
Y
and
sup .We have to prove that
0
xd
((
))
z
min{
xd
((
z
)
),
y
(
yd
((
))},
x`, y`, z`Y. Let f : X Y be an onto a homomorphism of KU - algebras , µ a
fuzzy left derivations KU - ideal of X with sup property and β the image of μ under f ,
since µ is a fuzzy left derivations KU - ideal of X , we have µ(d(0)) ≥ µ(d(x)) for all
xX . Note that 0 f 1 (0`) , where 0 , 0` are the zero of X and Y respectively
12
Thus,
d
))0((
sup
1
f
d
(
td
)(
))0(
td
))((
d
))0((
)0(
xd
((
)),
for all
Xx ,
which implies that
d
))0((
sup
1
f
(
xd
(
td
)(
td
((
))
xd
((
)),
for any
x
Y
. For
))
any
zyx
,
,
Y
,let
xd
(
0
)
f
1
xd
((
))
,
yd
(
0
)
f
1
yd
((
))
,
zd
(
0
)
f
1
zd
((
))
be Such that
xd
((
0
z
0
))
td
)(
f
sup
(
(
1
zxd
td
((
))
,
(
y
0
)
\
))
sup
f
td
)(
(
1
yd
(
td
((
))
))
and
xd
((
0
)
(
(
xd
((
0
()
y
0
z
Then
y
0
sup
f
))
0
z
0
))
xdf
)
(({
0
xd
((
0
)
(
y
(
0
y
0
z
))}
0
z
))}
0
xd
((
1
(
xd
(
()
y
z
))
y
)
(
sup
xd
()
(
(
1
z
))
y
z
))
td
((
.
))
td
)(
f
xd
((
))
z
sup
f
(
(
1
zxd
td
)(
td
((
))
xd
((
0
z
0
))
min{
xd
((
0
)
(
y
0
z
0
)),
yd
((
0
))}
=
))
min
sup
xd
()
(
(
y
z
)))
(
td
, ))(
sup
1
f
(
yd
(
\
))
td
)(
(
td
(
))
td
)(
1
f
=
Hence β is a fuzzy left derivations KU-ideal of Y.
min{
xd
((
)
(
y
z
,))
yd
((
))}
.
Theorem 4.5 An onto homomorphic preimage of a fuzzy right derivations KU - ideal is
also a fuzzy right derivations KU - ideal
Theorem 4.6 Let f : X → Y be a homomorphism between KU - algebras X and Y .
For every fuzzy right derivations KU - ideal µ in X , f (µ) is a fuzzy right derivations
KU - ideal of Y .
proof. Clear
13
5. Cartesian product of fuzzy left derivations KU-ideals
Definition 5.1[4] A fuzzy µ is called a fuzzy relation on any set S , if µ is a fuzzy
subset µ : S × S → [0,1] .
Definition 5.2 [4] If µ is a fuzzy relation on a set S and β is a fuzzy subset of S ,
then μ is a fuzzy relation on β if μ (x , y) ≤ min {β (x) , β (y)}, ∀ x , y ∈ S .
of μ and β is define by (μ × β) (x , y) = min {μ (x) , β (y)} , ∀ x , y ∈ S .
Definition 5.3 [4] Let µ and β be fuzzy subset of a set S, the Cartesian product
Lemma 5.4[4] Let μ and β be fuzzy subset of a set S ,then
(i) is a fuzzy relation on S .
(ii)
t)
(
=
t ×
t for all t ∈ [0,1].
Definition 5.5 If µ is a fuzzy left derivations relation on a set S and β is a fuzzy left
derivations subset of S , then µ is a fuzzy left derivations relation on β if
µ(d (x , y)) ≤ min {β(d (x)) , β(d (y))}, x , y S .
Definition 5.6 [4] Let µ and β be fuzzy left derivations subset of a set S , the Cartesian
product of µ and β is define by (µ × β)(d (x , y)) = min {µ(d (x)) , β(d (y))} , x , y S
Lemma 5.7[4] Let µ and β be fuzzy subset of a set S ,then
(i)
(ii)
is a fuzzy relation on S ,
(
t)
=
t ×
t for all t [0,1].
14
Definition 5.8 If β is a fuzzy left derivations subset of a set S , the strongest fuzzy
relation on S , that is a fuzzy derivations relation on β is µβ given by
µβ(d (x , y)) = min {β(d (x)) , β(d (y))}, x , y S .
Lemma 5.9 For a given fuzzy left derivations subset S , let µβ be the strongest fuzzy left
derivations relation on S ,then for t [0,1] , we have (µβ)t= βt × βt .
Proposition 5.10 For a given fuzzy subset β of KU - algebra X , let µβ be the strongest
left fuzzy derivations relation on X . If µβ is a fuzzy left derivations KU - ideal of X × X ,
then β(d (x)) ≤ β(d (0))= β(0) for all x X .
Proof . Since µβ is a fuzzy left derivations KU- ideal of X × X , it follows from (F1) that
µβ (x , x) = min {β(d (x)) , β(d (x))} ≤ β (d(0 , 0)) = min {β(d (0)) , β(d (0))} ,
where (0 , 0) X × X then β (d(x)) ≤ β(d (0)) = β(0) .
Remark5.11 Let X and Y be KU- algebras , we define * on X × Y by :
For every (x , y), (u , v)X x Y , (x , y ) * (u , v) = ( x * u , y * v) , then clearly
(X × Y, * , (0 , 0) ) is a KU- algebra .
Theorem 5.12 Let µ and β be a fuzzy left derivations KU- ideals of KU - algebra X ,then
µ × β is a fuzzy left derivations KU-ideal of X × X .
Proof : for any (x , y) X × X ,we have ,
(µ × β) (d((0, 0)) = min {µ(d (0)) , β(d (0))}= min {µ(0) , β(0)} ≥
min {µ(d (x)) , β(d (x))} = (µ x β)(d (x , y)) .
Now let (x1 , x2) , (y1 , y2) , (z1 , z2) X × X , then ,
(µ x β)(d (x1 * z1 , x2 * z2)) = min {µ(d (x1,z1)) , β(d (x2 , z2))}
≥ min{min {µ(d (x1 )*(y1 * z1))), µ(d(y1))}} , min {β(d (x2) * (y2 * z2))) , β(d(y2))}}
15
= min{min{µ(d (x1) * (y1 * z1))) , µ(d (x2) * (y2 * z2)))} , min{µ(d(y1)), β(d(y2))}}
= min{(µ × β) ( (d(x1) * (y1 * z1) , d(x2) * (y2 * z2))) ,( µ × β) (d(y1), d(y2))} .
Hence µ × β is a fuzzy left derivations KU- ideal of X × X .
Analogous to theorem 3.2 [ 16] , we have a similar results for fuzzy left derivations KU-
ideal , which can be proved in similar manner , we state the results without proof .
Theorem 5.13 Let µ and β be a fuzzy left derivations subset of KU-algebra X ,such that
µ × β is a fuzzy left derivations KU-ideal of X × X , then
(i) Either µ(d (x)) ≤ µ(d (0)) or β(d (x)) ≤ β(d (0)) for all x X ,
(ii) If µ(d (x)) ≤ µ(d (0)) for all x X , then either µ(d (x)) ≤ β(d (0)) or
β(d (x)) ≤ β(d(0)) ,
(iii) If β(d (x)) ≤ β(d (0)) for all x X , then either µ(d (x)) ≤ µ(d (0)) or
β(d (x)) ≤ µ(d(0)),
(iv) Either µ or β is a fuzzy left derivations KU- ideal of X .
Theorem 5.14 Let β be a fuzzy subset of KU- algebra X and let µβ be the strongest fuzzy
left derivations relation on X , then β is a fuzzy left derivations KU - ideal of X if and
only if µβ is a fuzzy left derivations KU- ideal of X × X .
proof : Assume that β is a fuzzy left derivations KU- ideal X , we note from (F1) that :
µβ (0, 0) = min {β (d(0)) , β (d(0))} =min {β (0) , β (0)} ≥ min {β (d(x)) , β (d(y)) }
= µβ (d(x) , d(y) ).
Now, for any (x1,x2) , (y1,y2) ,(z1,z2) X x X , we have from (F2) :
µβ (d(x1 * z1) , d(x2 * z2)) = min {β (d(x1 * z1)) , β (d(x2 * z2))}
≥ min {min{β (d(x1) * (y1 * z1)) , β (d(y1))} , min {β (d(x2) * (y2 * z2)) , β(y2)}}
= min{min{β(d(x1) * (y1 * z1)) , β (d(x2) * (y2 * z2))} , min {β(d(y1)), β(d(y2))}}
= min {µβ (d(x1) * (y1 * z1) , d(x2) * (y2 * z2)) , µβ (d(y1) , d(y2))} .
16
Hence µβ is a fuzzy left derivations KU - ideal of X × X .
Conversely . For all (x , y) X × X , we have
Min {β (0) , β (0) } = µβ (x , y) = min {β (x) , β (y)}. It follows that
β (0) ≥ β (x) for all x X , which prove (F1).
Now, let (x1 , x2) , (y1 , y2), (z1 , z2) X × X , then
min {β (d(x1 * z1)) , β d(x2 * z2))} = µβ (d(x1 * z1) , d( x2 * z2))
≥ min {µβ (d(x1 , x2) * ((y1 , y2) * (z1, z2) ) , µβ (d(y1),d(y2))}
= min {µβ (d(x1) * (y1 * z1) , d(x2) * (y2 * z2)) , µβ (d(y1) , d(y2))}
= min {min {β (d(x1)* (y1 * z1)) , β (d(x2) * (y2 * z2))} , min {β (d(y1)) , β (d(y2))}}
= min {min {β (d(x1) * (y1 * z1)) , β (d(y1))} , min {β((dx2) * (y2 * z2)) , β (d(y2))}}
In particular, if we take x2 = y2 = z2 =0 , then ,
β (d(x1 * z1)) ≥ min { β (d(x1) * (y1 * z1)), β (d(y1))} This prove (FL2) and completes the
proof.
Conclusion
Derivation is a very interesting and important area of research in the theory of algebraic
structures in mathematics. In the present paper, the notion of fuzzy left derivations KU -
ideal in KU-algebra are introduced and investigated the useful properties of fuzzy left
derivations KU - ideals in KU-algebras.
In our opinion, these definitions and main results can be similarly extended to some other
algebraic systems such as BCI-algebra, BCH-algebra ,Hilbert algebra ,BF-algebra -J-
algebra ,WS-algebra ,CI-algebra, SU-algebra ,BCL-algebra ,BP-algebra ,Coxeter algebra
,BO-algebra ,PU- algebras and so forth.
The main purpose of our future work is to investigate:
(1) The interval value, bipolar and intuitionistic fuzzy left derivations KU - ideal in KU-
algebra.
(2) To consider the cubic structure of left derivations KU - ideal in KU-algebra.
We hope the fuzzy left derivations KU - ideals in KU-algebras, have applications in
different branches of theoretical physics and computer science.
17
Algorithm for KU-algebras
Input (
:X set,
: binary operation)
Output (“ X is a KU-algebra or not”)
Begin
If
X
then go to (1.);
End If
If
X0
then go to (1.);
End If
Stop: =false;
1:i
;
While
i
X
and not (Stop) do
If
x
i
x
i
0
then
Stop: = true;
End If
1:j
While
j
X
and not (Stop) do
If
((
y
x
i
)
x
i
)
j
0
then
Stop: = true;
End If
End If
1:k
While
k
X
and not (Stop) do
If
(
x
i
y
j
)
((
y
z
)
(
x
i
k
z
k
j
))
0
then
18
Stop: = true;
End If
End While
End While
End While
If Stop then
(1.) Output (“ X is not a KU-algebra”)
Else
Output (“ X is a KU-algebra”)
End If
End.
References
[1] H. A. S. Abujabal and N. O. Al-Shehri, “On left derivations of BCI-algebras,”
Soochow Journal of Mathematics, vol. 33, no. 3, pp. 435–444, 2007.
[2] H. E. Bell and L.-C. Kappe, “Rings in which derivations satisfy certain algebraic
conditions,” Acta Mathematica Hungarica, vol. 53, no. 3-4, pp. 339–346, 1989.
[3] H. E. Bell and G. Mason, “On derivations in near-rings, near-rings and Near-fields,”
North-Holland Mathematics Studies, vol. 137, pp. 31–35, 1987.
[4] P.Bhattacharye and N.P.Mukheriee , Fuzzy relations and fuzzy group inform ,
sci ,36(1985), 267-282
[5] M. Bresar and J. Vukman, “On left derivations and related mappings,” Proceedings
of the American Mathematical Society, vol. 110, no. 1, pp. 7–16, 1990.
[6] M. Bresar, “On the distance of the composition of two derivations to the generalized
derivations,” Glasgow Mathematical Journal, vol. 33, no. 1, pp. 89–93, 1991.
[7] W. A. Dudek, The number of subalgebras of finite BCC-algebras, Bull.
Inst. Math. Acad. Sinica, 20 (1992), 129-136.
[8] W. A. Dudek and X. H. Zhang, On ideals and congruences in BCCalgebras
, Czechoslovak Math. J., 48(123) (1998), 21-29.
[9] B. Hvala, “Generalized derivations in rings,” Communications in Algebra, vol. 26,
no. 4, pp. 1147–1166, 1998
19
[10] Y.Imai and Iseki K: On axiom systems of Propositional calculi, XIV, Proc. Japan
Acad. Ser A, Math Sci., 42(1966),19-22.
[11] k.Iseki: An algebra related with a propositional calculi, Proc. Japan Acad. Ser A
Math. Sci., 42 (1966), 26-29.
[12] K Iseki and Tanaka S: An introduction to theory of BCK-algebras, Math. Japo., 23
(1978) 1-26.
[13] Y. B. Jun and X. L. Xin, “On derivations of BCI-algebras,” Information Sciences,
vol. 159, no. 3-4, pp. 167–176, 2004.
[14]A. A. M. Kamal, “σ-derivations on prime near-rings,” Tamkang Journal of
Mathematics, vol. 32, no. 2, pp. 89–93, 2001
[15]Y. Komori, The class of BCC-algebras is not a variety, Math. Japonica,
29 (1984), 391-394.
sets and systems , 43(1991) ,117-123 .
[16 ] D.S.Malik and J.N.Mordeson , Fuzzy relation on rings and groups , fuzzy
[ 17] S. M. Mostafa – M. A . Abd-Elnaby- M.M. Yousef,“ Fuzzy ideals of KU –
algebra”International Mathematical Forum , Vol. 6 , 2011 .no. 63 ,3139 – 3149.
[18] S. M.Mostafa , R. A. K. Omar , A. Abd-eldayem, Properties of derivations on KU-
algebras ,gournal of advances in mathematics Vol .9, No 10.
[19]S. M. Mostafa, F. F.Kareem , Left fixed maps and α-derivations of KU-algebra ,
gournal of advances in mathematics Vol .9, No 7
)
[20] G. Muhiuddin and Abdullah M. Al-Roqi On
.Discrete Dynamics in Nature and Society Volume 2012.
[21] G. Muhiuddin and Abdullah M. Al-roqi ,On t-Derivations of BCI-Algebras
( -Derivations in BCI-Algebras
,
Abstract and Applied Analysis Volume 2012, Article ID 872784, 12 pages
[22] C.Prabpayak and U.Leerawat, On ideals and congruence in KU-algebras, scientia
Magna in- ternational book series, Vol. 5(2009), No.1, 54-57.
[23] C.Prabpayak and U.Leerawat, On isomorphisms of KU-algebras, scientia Magna
international book series, 5(2009), no .3, 25-31.
[24] A. Wronski, BCK-algebras do not form a variety, Math. Japonica, 28
(1983), 211-213.
[25] O.G.Xi , fuzzy BCK-algebras , Math . Japan ,36(1991).935-942 .
20
[26] L.A.Zadeh , Fuzzy sets , inform . and control ,8(1965) , 338-353 .
[27] J. Zhan and Y. L. Liu, “On f-derivations of BCI-algebras,” International Journal of
Mathematics and Mathematical Sciences, no. 11, pp. 1675–1684, 2005.
Samy M. Mostafa [email protected]
Department of Mathematics, Faculty of Education, Ain Shams University, Roxy, Cairo,
Egypt.
Ahmed Abd-eldayem [email protected]
Egypt.
Department of Mathematics, Faculty of Education, Ain Shams University, Roxy, Cairo,
21
|
1903.03584 | 1 | 1903 | 2019-03-08T18:02:10 | On Lenagan's Theorem for finite length bimodules | [
"math.RA",
"math.RT"
] | We offer a self-contained proof of Lenagan's Theorem which does not rely on Goldie's Theorem | math.RA | math |
ON LENAGAN'S THEOREM FOR FINITE LENGTH
BIMODULES
ANDREW HUBERY
Abstract. We offer a self-contained proof of Lenagan's Theorem which does
not rely on Goldie's Theorem.
1. Introduction
Lenagan's Theorem states that for a bimodule ΓMΛ, if ΓM has finite length and
MΛ is Noetherian, then MΛ also has finite length. This theorem first appeared in
[7], and as Lam writes
Indeed, although the argument above is quite short, it seemed to
have used the full force of Goldie's First Theorem, and it is not
clear at all how one could have proved [it] otherwise. [6, p. 333]
In fact, Lenagan's Theorem was generalised by Crawley-Boevey [5]:
One important consequence is that the left and right Artin radicals of a Noether-
ian ring agree. We can also regard Lenagan's Theorem as a generalisation of the
classical result that a left Artinian ring is right Artinian if and only if it is right
Noetherian. This latter is a direct consequence of the Hopkins-Levitzki Theorem.
if ΓM is
Artinian and MΛ is Noetherian, then ΓM and MΛ both have finite length. We give
a new proof of this result without recourse to Goldie's Theorem. Instead it follows
from a strengthening of the Hopkins -- Levitzki Theorem. In a similar way we also
obtain a result of Bjork on subrings of semiprimary rings. The starting point for
our proof is a result of Camps and Dicks on semilocal rings.
Let Λ be a ring, with Jacobson radical rad(Λ). We call Λ semilocal if ¯Λ :=
Λ/ rad(Λ) is semisimple, and semiprimary if moreover rad(Λ) is nilpotent.
It is
well-known that the endomorphism ring of a finite length object in an abelian
category is always semiprimary.
2. Finite length modules
We begin with a beautiful proof, due to Camps and Dicks [2]. We follow the
proof in the second edition [3], which incorporates ideas from Camillo and Nielsen
[4].
Theorem 1. Let M be an Artinian object in some abelian category, and set E :=
End(M ). If Λ ⊂ E is a subring such that Λ× = Λ ∩ E×, then Λ is semilocal.
Proof. Observe first that for all x, y ∈ E we have
Ker(x − xyx) = Ker(x) ⊕ Ker(1 − xy),
2010 Mathematics Subject Classification. Primary 16P20; Secondary 16D20.
1
2
ANDREW HUBERY
induced by the idempotent endomorphism xy on Ker(x − xyx). Also, M satisfies
the ascending chain condition with respect to direct summands: given Ui ≤ M with
Ui = Ui−1 ⊕ Vi, the Vi are eventually all zero. For, given such a sequence, we can
construct the descending chain of submodules Li>n Vi.
Now let I ≤ Λ be a maximal right ideal. Take x 6∈ I such that Ker(x) is
maximal, with respect to direct summands, in {Ker(y) : y 6∈ I}. We claim that
xΛ ∩ I ⊂ rad(Λ), so suppose xy ∈ I. Then for all z ∈ Λ we have Ker(x − xyzx) =
Ker(x) ⊕ Ker(1 − xyz). Since x − xyz 6∈ I, we see that 1 − xyz is injective. By
Fitting's Lemma it is invertible in E, so is invertible in Λ by assumption.
It follows that ¯Λ equals ¯I ⊕ x¯Λ. In particular, x¯Λ is simple, so the socle of ¯Λ is
(cid:3)
not contained in any maximal right ideal, and hence ¯Λ is semisimple.
The following proposition is a strengthening of the Hopkins -- Levitzki Theorem,
which deals with the case Λ = E.
Proposition 2. Let Λ subring of semiprimary ring E, and M a right E-module.
Then MΛ is Noetherian if and only if it is Artinian.
Proof. Since E is semiprimary, ME has finite Loewy length, and hence we may
assume that ME is semisimple. Then MΛ Artinian or Noetherian implies the same
for ME, and hence we may assume that ME is simple. Note that this already proves
the result when Λ = E; that is, ME is Noetherian if and only if it is Artinian.
After taking the quotient by the annihilator of M , we may assume that E =
Mn(∆) for a division ring ∆, and hence regard M as a ∆-E-bimodule. We claim
that, in this situation, Λ is semiprimary, so MΛ has finite length as above.
Since EE ∼= M n
E, we know that EΛ, and hence also ΛΛ, is Artinian or Noetherian.
If Λ is right Artinian, then it is semiprimary, and we are done. Assume therefore
that EΛ is right Noetherian.
Suppose x ∈ Λ has inverse y ∈ E. For some d we have yd ∈ Pi<d yiΛ, and thus
y ∈ Pi<d xd−i−1Λ ⊂ Λ. Since ∆M has finite length, it follows from the theorem
that Λ is semilocal. Also, E · radn(Λ) form a descending chain of left ∆-modules,
so stabilises. Nakayama's Lemma then gives E · radn(Λ) = 0 for some n. Thus
radn(Λ) = 0, and Λ is semiprimary as claimed.
(cid:3)
The next corollary appears as Theorem 3.11 in [1].
Corollary 3. Let E be semiprimary and Λ ⊂ E a subring. If EΛ is Noetherian,
then Λ is right Artinian.
Proof. Apply the proposition to M = E.
(cid:3)
Corollary 4 (Lenagan,Crawley-Boevey). Let ΓMΛ be a bimodule such that ΓM is
Artinian and MΛ is Noetherian. Then ΓM and MΛ both have finite length.
Proof. It is enough to prove that E := EndΓ(M ) is semiprimary, since then MΛ
has finite length by the proposition, so EndΛ(M ) is also semiprimary, and hence
ΓM has finite length by the proposition once more.
Now, the theorem tells us that that E is semilocal. Also, the Γ-submodules
M · radn(E) form a descending chain, so must stabilise. As MΛ is Noetherian, so
too is ME. Thus Nakayama's Lemma gives M · radn(E) = 0 for some n, and hence
that radn(E) = 0.
(cid:3)
ON LENAGAN'S THEOREM FOR FINITE LENGTH BIMODULES
3
[1] Bjork, J.-E.: Conditions which imply that subrings of semiprimary rings are semiprimary. J.
Algebra 19, 384 -- 395 (1971)
References
[2] Camps, R., Dicks, W.: On semilocal rings. Israel J. Math. 81, 203 -- 211 (1993)
[3] Camps,
2nd
http://mat.uab.cat/~dicks/SemilocalNew.pdf (2010) Accessed 8 March 2019
Dicks,
W.:
R.,
On
semilocal
rings.
edition.
[4] Camillo, V.P., Nielsen, P.P.: On a theorem of Camps and Dicks. In: Van Huynh, D., L´opez-
Permouth, S.R. (eds) Advances in ring theory. Trends in Mathematics, pp. 83 -- 84. Birkh
auser, Basel (2010)
[5] Crawley-Boevey, W.W.: Modules of finite length over their endomorphism rings. In:
Tachikawa, H., Brenner, S. (eds) Representations of algebras and related topics (Kyoto, 1990).
London Math. Soc. Lecture Note Ser. 168, pp. 127 -- 184. Cambridge Univ. Press, Cambridge
(1992)
[6] Lam, T.Y.: Lectures on modules and rings. Graduate Texts in Math. 189. Springer-Verlag,
New York (1999)
[7] Lenagan, T.H.: Artinian ideals in Noetherian rings. Proc. Amer. Math. Soc. 51, 499 -- 500
(1975)
Bielefeld University, 33501 Bielefeld, Germany
E-mail address: [email protected]
|
1903.08210 | 1 | 1903 | 2019-03-19T18:44:36 | Determinants for integral forms in lattice type vertex operator algebras | [
"math.RA",
"math.QA"
] | We prove a determinant formula for the standard integral form of a lattice vertex operator algebra. | math.RA | math |
Determinants for integral forms in lattice
type vertex operator algebras
3 April, 2018
Chongying Dong
Department of Mathematics,
University of California,
Santa Cruz, CA 95064 USA
&
School of Mathematics and Statistics,
Qingdao University,
Qingdao 266071 CHINA
[email protected]
and
Robert L. Griess Jr.
Department of Mathematics,
University of Michigan,
Ann Arbor, MI 48109-1043 USA
[email protected]
Abstract
We prove a determinant formula for the standard integral form of
a lattice vertex operator algebra.
1
Introduction
We have studied group-invariant integral forms in vertex operator algebras
[2, 3].
In this article, we study standard integral forms in lattice vertex
operator algebras and give the determinant of each homogeneous piece as a
particular integral power of determinant of the input positive definite even
integral lattice. When the lattice is unimodular, all these homogeneous pieces
have determinant 1, already proved in [2]. For lattices of other determinants,
there did not seem to be an obvious answer. Borcherds stated without proof
1
in [1] that the determinant of a homogeneous piece was some (unspecified)
integral power of the input lattice.
The standard integral form VL.Z for a lattice vertex operator algebra VL
is reviewed in Section 4, Definition (4.1). Lemma (3.3) shows that our main
theorem (5.1) is reduced to a study of determinants for integral forms within
a certain symmetric algebra. The latter determinant is therefore our main
object of study in this article.
2 Background
Lemma 2.1. If x1, . . . , xk are variables, then the number of monomials
1 · · · xak
xa1
k , ai ∈ Z≥0, of total degree n is (cid:0)n+k−1
k−1 (cid:1).
Proof. This is essentially a counting result, called Balls in Urns. Monomials
correspond to the set of k−1 marker balls to be chosen among a set of n+k−1
balls arranged in a straight line. One adds marker ball 0 at the very beginning
and marker ball k at the very end. The sequence a1, . . . , ak gives the lengths
of the gaps between successive marker balls. (cid:3)
Lemma 2.2. If J ≤ K are finite rank lattices and the index J : K is finite,
then det(K) = J : K2det(J).
3 Symmetric algebas
Notation 3.1. Let H be a k-dimensional vector space over C and let t be
a variable. For r ≥ 1, let Hr = H ⊗ Ct−r be a copy of H, defined to have
degree r, and set h(−r) = h ⊗ t−r for h ∈ H. We shall work in the symmetric
algebra M(1) = S[H[t−1]t−1] = C[h(−r)h ∈ H] where
Then
H[t−1]t−1 = ⊕r≥1H ⊗ Ct−r.
M(1) = ⊕n≥0M(1)n
is graded such that M(1)n is spanned by h1(−r) · · · hp(−rp) for h1, ..., hp ∈ H
and r1, ..., rp ∈ N with Pi ri = n. For a sequence a = (a1, . . . , an, . . . ) of
nonnegative integers which is almost all zero, define wt(a) :=Pj≥1 jaj. For
n ≥ 0, define A(n) to be the set of such a of weight n. Note that A(0) = ∅.
2
Suppose that x1, . . . , xk is a basis of H. Then L = Zx1 +· · ·+Zxk is a free
abelian group of rank r in H and M(1) = C[x1(−r), · · · xk(−r)r ≥ 1]. Define
B(a) := B(x1, . . . xk; a) to be the Z-span of all words w1 · · · wn in M(1) where
wi is a product of length ai in the variables xj(−i), for j ∈ {1, 2, . . . , k}. Fi-
nally, for an integer n ≥ 0, define BL(n) := B(x1, . . . xk; n) := ⊕a∈A(n)B(a)
and BL := ⊕n≥0BL(n) = Z[x1(−r), · · · xk(−r)r ≥ 1]. Then BL is a subring
of R. Note that these objects are unchanged if x1, . . . , xk is replaced by any
basis of the Z span of x1, . . . , xk.
Notation 3.2. Let AL be the Z-submodule of M(1) generated by sα,n for
αi ∈ {x1, . . . xk} and n ≥ 0 where
E−(−α, z) := exp Xn>0
α(−n)
n
zn! =Xn≥0
sα,nzn.
Although we do not use the vertex operator algebra structure on M(1), we
use the notations E−(−α, z) and sα,n from [4] and [2] here. Then AL has a
Z-base B1B2 · · · Bk where
Bi = {sxi,n1 · · · sxi,nqn1 ≥ · · · ≥ nq ≥ 0}
and B1 · · · Bk = {u1u2 · · · ukui ∈ Bi}.
We also set AL(n) = AL ∩ M(1)n for all n. The following result will be
useful in computing the determinants for the lattice vertex operator algebras.
Lemma 3.3. BL is a subring of AL and the index [AL(n) : BL(n)] is inde-
pendent of the base {x1, ..., xk} for any n ≥ 0.
Proof. We first prove that BL is a subring of AL. It is good enough to show
that α(−n) ∈ BL for α ∈ {x1, ..., xk} and n ≥ 0. Note that
E−(−nα, z) = E−(−α, z)n = (Xm≥0
sα,mzm)n.
coefficient of zn in (Pn
So the coefficient cn of zn in E−(−nα, z) lies in AL. Clearly, cn is also the
m=0 sα,mzm)n. A straightforward computation shows
that an = α(−n)+u where u is a Z-linear combination of elements of the form
sα,m1sα,m2 · · · withmi < n and m1 + m2 + · · · = n. As a result, α(−n) ∈ AL.
To show that the index [AL(n) : BL(n)] is independent of the base
{x1, ..., xk} for any n ≥ 0, we let {y1, ..., yk} be another basis of H and
3
K = Zy1 + · · · + Zyk. Then a group isomorphism f from L to K by sending
xi to yi induces a ring isomorphism f from R to itself such that f (AL) = AK
and f (BL) = BK. It is evident that f is a degree preserving map. As a re-
sult, f (AL(n)) = AK(n) and f (BL) = BK. Thus, [AL(n) : BL(n)] = [AK(n) :
BK(n)]. (cid:3)
Note that both AL and BL are Z-forms of M(1).
Lemma 3.4. (i) rank(Sm(H)) =(cid:0)m+k−1
k−1 (cid:1);
(ii) rank(B(a)) =Qn
(iii) wt(B(a)) =Pj≥1 jaj;
(iv) rank(B(n)) =Pa:wt(a)=n rank(B(a)).
j=1(cid:0)aj +k−1
k−1 (cid:1);
Proof. Straightforward, with Lemma (2.1). (cid:3)
So far, there is no bilinear form in this discussion. We shall introduce
forms later, after Corollary 3.8.
We now compare what happens to the B(a) when x1, . . . , xn is replaced
by another basis. We already noted in Notation 3.1 that B(x1, . . . xk; a) =
B(y1, . . . yk; a) if spanZ(x1, . . . xk) = spanZ(y1, . . . yk).
Using the proof of Lemma 3.3 we can easily have:
Lemma 3.5. If x1, . . . xk and y1, . . . yk are bases and if spanZ(x1, . . . xk) con-
tains spanZ(y1, . . . yk), then for any invertible linear transformation T on H,
B(x1, . . . xk; a)/B(y1, . . . yk; a) ∼= B(T x1, . . . T xk; a)/B(T y1, . . . T yk; a).
In
particular, we have equality of indices B(x1, . . . xk; a) : B(y1, . . . yk; a) =
B(T x1, . . . T xk; a) : B(T y1, . . . T yk; a).
Lemma 3.6. Suppose that p > 0 is an integer. Then B(x1, x2, . . . , xk; a)
contains B(px1, px2, . . . , pxk; a) and the index is pN (k,a) where N(k, a) :=
Proof. The free abelian group B(x1, x2, . . . , xk; a) has basis consisting of
monomials in the xt(−j). Such a monomial has a unique expression w1 · · · wwt(a),
Qn
j=1 aj(cid:0)aj +k−1
k−1 (cid:1).
where wj is a monomial in the xt(−j). There are (cid:0)aj +k−1
mula for N(k, a) is now clear. (cid:3)
k−1 (cid:1) such wj. The for-
Lemma 3.7. Suppose that p > 0 is an integer. Then B(x1, x2, . . . , xk; a)
k N (k,a) where N(k, a) :=
contains B(px1, x2, . . . , xk; a) and the index is p
1
Qn
j=1 aj(cid:0)aj +k−1
k−1 (cid:1).
4
Proof. Observe that we have a chain
spanZ(x1, x2, x3, . . . xk) > spanZ(px1, x2, x3, . . . xk) >
spanZ(px1, px2, x3, . . . xk) > · · · > spanZ(px1, px2, px3, . . . pxk).
By Lemma (3.5), the indices for each containment
B(x1, x2, x3, . . . xk; a) > B(px1, x2, x3, . . . xk; a) >
B(px1, px2, x3, . . . xk; a) > · · · > B(px1, px2, px3, . . . pxk; a)
are equal. We then deduce the result from Lemma (3.6). (cid:3)
Corollary 3.8. In the notation of Lemma (3.7), the index
B(x1, x2, . . . , xk; a) : B(px1, x2, . . . , xk; a)
is p
1
k Pa∈A(n) N (k,a) = p
1
k P(aj )=a∈A(n) Qn
j=1 aj(aj +k−1
k−1 ).
Now assume that H has a nondegenerate symmetric bilinear form h··i.
Then we can make M(1) an irreducible module for the affine algebra
H = H ⊗ C[t, t−1] ⊕ CK
such that H ⊗ C[t] annihilates 1 and the central element K acts as 1. We
abbrevuate h ⊗ tm by writing h(m) for h ∈ H and m ∈ Z.
Notation 3.9. There is a unique nondegenerate symmetric bilinear form h··i
on M(1) such that h11i = 1 and hh(m)uvi = −huh(−m)vi for u, v ∈ M(1)
and h ∈ H (see [5], [2]).
Furthermore, BL(n) = B(x1, ..., xk; n) is a Z-form of M(1)n. Note that if
L = Zx1 + · · · + Zxk is rational lattice of H in the sense that for any α, β ∈ L
hαβi ∈ Q, then B(x1, ..., xk; n) is also a rational lattice, due to the form. In
the notation of Corollary (3.8), we have
Corollary 3.10. Assume existence of the form as in Notation 3.9. For k ≥ 1
and n ≥ 0, define
S(k, n) :=
1
k Xa∈A(n)
N(k, a) =
1
k Xa=(aj )∈A(n)
n
Yj=1
k − 1 (cid:19).
aj(cid:18)aj + k − 1
Then S(k, 0) = 0 and
det(B(px1, x2, . . . , xk; n)) = det(B(x1, x2, . . . , xk; n))p2S(k,n)
for all k ≥ 1 and n ≥ 0.
5
Remark 3.11. This presentation helps us understand the "homogeneous
part" of the standard integral form in the symmetric algebra spanned over
Z by all monomials made from a basis. The integral form involves expres-
sions like Schur functions which have fractional coefficients so are not in the
homogeneous part. We shall study the quotient of that integral form by its
homogeneous part.
4
Integral forms of M (1)
Let L be a positive definite integral lattice with basis x1, . . . , xk and we denote
the form on L by h· ·i. We recall the standard integral form for the lattice
vertex operator algebra based on L.
Note from [4] that M(1) := C[xi(−n)i = 1, ..., k; n > 0] is the Heisen-
berg vertex operator algebra and VL = M(1) ⊗ Cǫ[L] is the correspond-
ing lattice vertex operator algebra where ǫ is a bimultiplicative map from
L × L → h±1i such that ǫ(α, β)ǫ(β, α) = (−1)hαβi and ǫ(α, α) = (−1)hααi/2
and where Cǫ[L] = ⊕α∈LCeα is the twisted group algebra. There is a unique
nondegenerate symmetric invariant bilinear form h· ·i on VL such that
and
heα eβi = δα+β,0
hα(n)u vi = −hu α(−n)vi
for all u, v ∈ VL α ∈ L and n ∈ Z (see [1], [2]).
Recall the subring AL from Section 3. Then AL is a Z-form of M(1) in
the sense that AL is a vertex algebra over Z, hu vi ∈ Z for u, v ∈ AL and
M(1) = C ⊗Z AL [2].
We also set (VL)Z = ⊕α∈LAL ⊗eα. Then (VL)Z is a vertex operator algebra
over Z generated by e±xi for i = 1, ..., k and is a free Z-module such that
VL = C ⊗Z (VL)Z.
Definition 4.1. (VL)Z = ⊕α∈LAL ⊗ eα. is the standard integral form in the
lattice vertex operator algebra VL.
Let (VL)Z,n consists of vectors of weight n in (VL)Z. To get det((VL)Z,n),
we first understand det(AL(n)) in terms of det(L). Recall BL = Z[xi(−n)
i = 1, ..., k; n > 0] and BL(n) = BL ∩ M(1)n for n ≥ 0. By Lemma 3.3,
[AL(n) : BL(n)] only depends on the rank of L and integer n.
6
Using Lemma 3.3 we can give an explicit expression of [AL(n) : BL(n)].
We define numbers b0 := 1 and for n > 0,
bn :=
Ya=(a1,a2,··· )∈A(n)Yi≥1
iai · ai!.
Lemma 4.2. The index [AL(n) : BL(n)] is the square root of
Yn1,...,nk≥0, P ni=n
bn1 · · · bnk
for n ≥ 0.
Proof. By Lemma 3.3, [AL(n) : BL(n)] is independent of lattice L. So we
can choose L = Zx1 + · · ·+ Zxk such that {x1, ..., xk} is an orthonormal basis
of H for convenience of computation. Then AL(n) is a unimodular lattice by
Proposition 3.6 of [2]. It is easy to show that
hxi(−p)s xi(−p)si = (−1)sh1 xi(p)sxi(−p)1i
= (−1)ss!psh1 1i
= (−1)ss!ps
for any i, s. This shows that det(BL(n)) equalsQn1,...,nk≥0, P ni=n bn1 · · · bnk .
Since det(BL(n)) = [AL(n) : BL,n]2, the result follows immediately. (cid:3)
Lemma 4.3. Let A1, A2, C1, C2 be lattices with the same rank such that Ci ⊂
Ai for i = 1, 2, C2 ⊂ C1. Then det(A2) = [C1:C2]2[A1:C1]2 det(A1)
. In particular,
if det(A1) = 1 and [A1 : C1] = [A2 : C2] then det(A2) = [C1 : C2]2.
[A2:C2]2
Proof. The result follows from the following relations
det(Ci) = det(Ai)[Ai : Ci]2, det(C2) = det(C1)[C1 : C2]2
for i = 1, 2. (cid:3)
Theorem 4.4. Let L be an positive definite integral
lattice with a base
{x1, ..., xk} as before. Then for each n ≥ 0, det(AL(n)) is an integer power
of det(L). In fact, det(AL(n)) = det(L)2S(k,n), where S(k, n) is given by
Lemma (3.10).
7
Proof. We prove the theorem in several steps. If K contains the sublattice
J with finite index, then one may deduce the results for K from those for J,
and conversely, from the results of Section 3.
The result det(AL(n)) = 1 when L is unimodular was proved in [2]. Let
p be a positive integer.
Case (a): Let 0 6= p ∈ Z and L = Zpe1 ⊕ Ze2 ⊕ · · · ⊕ Zek be a sublattice of
Zk = Ze1 ⊕ · · · ⊕ Zek where {e1, ..., ek} is the standard orthonormal basis of
Rk. Using Lemmas 3.3, 4.2 with A1 = AZk (n), A2 = AL(n), C1 = BZk (n) and
C2 = BL(n) gives det(AL(n)) = [BZk (n) : BL(n)]2. Note that det(L) = p2.
By Corollary 3.10, det(AL(n)) = p2S(k,n) = det(L)2S(k,n).
Case (b): Let L = Zp1e1 ⊕ Zp2e2 ⊕ · · · ⊕ Zpkek for any positive integers
k and det(AL(n)) = det(L)2S(k,n) by Case
1 · · · p2
p1, ..., pk. Then det(L) = p2
(a).
Case (c): Let T be a positive integer such that L is a rank k sublattice
of K = 1
Zk, i.e., L ⊂ Qk. Then det(L) = [K : L]2T −2k. There exist a base
T
{u1, ..., uk} of K and positive integers p1, ..., pk such that {p1u1, ..., pkuk} is a
base of L. This implies that [K : L] = p1 · · · pk. From Lemma 4.3 and discus-
sion in Case (b), we see that det(AL(n)) = [BK(n) : BL(n)]2 det(AK(n)) =
(p1 · · · pk)2S(k,n) det(AK(n)). On the other hand,
1 = det(AZk(n)) = [BK(n) : BZk (n)]2 det(AK(n)) = T 2kS(k,n) det(AK(n)).
Thus
det(AL(n)) = (p1 · · · pk)2S(k,n)T −2kS(k,n) = det(L)2S(k,n).
Case (d): Let L be an arbitrary integral lattice in Euclidean space Rk.
The problem with applying (c) is that L is not necessarily a sublattice of
Qk. However, we can use a sequence of rational lattices {Lii ∈ N} such
that "limi→∞ Li = L". We fix a base {v1, ..., vk} of L and take linearly
independent vectors vi
i for all i, j. It is
clear that limi→∞ det(Li) = det(L) and limi→∞ det(ALi(n)) = det(AL(n))
for any n ≥ 0. It follows from Case (c) that det(AL(n)) = det(L)2S(k,n), as
desired. (cid:3)
k ∈ Qk such that vi
j − vj < 1
1, ..., vi
5
Integral forms of VL
We now assume that L is a positive definite even lattice. Recall that (VL)Z =
⊕α∈LAL ⊗ eα. Also recall (VL)Z,n from Section 4. We determine det((VL)Z,n)
in this section.
8
For m ≥ 0 we set L2m = {α ∈ Lhα αi = 2m}. Define Y0 := Lo = {0}.
For m ≥ 1, let Y2m be a subset of L2m such that 2Y2m = L2m and L2m =
Y2m ∪ (−Y2m). For α ∈ L we set W α = M(1)L ⊗ eα + M(1)L ⊗ e−α ⊂ (VL)Z.
Let W α
n 6= 0 if and only if α ∈ L2m and m ≤ n. In
this case, W α
n = AL(n − m) ⊗ eα + AL(n − m) ⊗ e−α. Observe that
n = W α ∩ (V )L,n. Then W α
and hW α W βi = 0 if α 6= β. So
(VL)Z,n = ⊕n
m=0 ⊕α∈Y2m W α
n
det((VL)Z,n) =
n
Ym=0 Yα∈Y2m
det(W α
m).
From the definition of the bilinear form, we know that for α ∈ Y2m with
m 6= 0
det(W α
n ) = det(AL(n − m))2.
Also, det(W 0
0 ) = det(Z1) = 1.
Here is our main theorem, an immediate consequence of Theorem 4.4.
Theorem 5.1. For all n ≥ 0, we have
det((VL)Z,n) =
n
Ym=0
det(L)L2m2S(k,n−m).
6 Acknowledgements
C. Dong was supported by China NSF grant 11871351. R. Griess was sup-
ported by funds from his Collegiate Professorship and Distinguished Univer-
sity Professorship at the University of Michigan.
References
[1] R. Borcherds, Vertex algebras, Kac-Moody algebras, and the Mon-
ster, Proc. Nat. Acad. Sci. U.S.A. 83 (1986), 3068-3071.
[2] C. Dong and R. L. Griess Jr, Integral forms in vertex operator
algebras which are invariant under finite groups, J. Algebra 365
(2012), 184-198.
9
[3] C. Dong and R. L. Griess Jr, Lattice-integrality of certain group-
invariant integral forms in vertex operator algebras, J. Algebra 474
(2017), 505-516.
[4] I. B. Frenkel, J. Lepowsky and A. Meurman, Vertex Operator Alge-
bras and the Monster, Pure and Applied Math., Vol. 134, Academic
Press, 1988.
[5] H. Li, Symmetric invariant bilinear forms on vertex operator alge-
bras, Pure and Appl. Math. 96 (1994), 279-297.
10
|
1302.5476 | 1 | 1302 | 2013-02-22T03:57:15 | On the definition of nucleus for dialgebras | [
"math.RA"
] | Malcev dialgebras have been introduced recently by Bremner, Peresi and S\'anchez-Ortega. In the present paper, we continue their study by introducing the notion of the generalized alternative di-nucleus of a 0-dialgebra. A general conjecture about the speciality of Malcev dialgebras in terms of this di-nucleus is formulated. In the last section, we introduce the appropriate generalization of the associative nucleus for dialgebras, and prove an analogue of Kleinfeld's theorem for the setting of dialgebras. | math.RA | math |
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
JUANA S ´ANCHEZ-ORTEGA
Abstract. Malcev dialgebras were introduced recently by Bremner, Peresi
and S´anchez-Ortega. In the present paper, we continue their study by intro-
ducing the notion of the generalized alternative di-nucleus of a 0-dialgebra. A
general conjecture about the speciality of Malcev dialgebras in terms of this
di-nucleus is formulated.
In the last section, we introduce the appropriate
generalization of the associative nucleus for dialgebras, and prove an analogue
of Kleinfeld's theorem for the setting of dialgebras.
1. Introduction
This paper is devoted to a better understanding of Malcev dialgebras, recently
introduced by Bremner, Peresi and S´anchez-Ortega [6].
It is well known that every associative algebra A gives rise to a Lie algebra A−,
when the associative product xy is replaced by the commutator [x, y] = xy − yx.
And conversely, the famous Poincar´e-Birkhoff-Witt Theorem [13] states that any
Lie algebra is isomorphic to a subalgebra of A− for some associative algebra A.
Loday and Pirashvili [22] proved that the Poincar´e-Birkhoff-Witt Theorem remains
true for Leibniz algebras, a "noncommutative" version of Lie algebras; see also
Aymon and Grivel [1], Insua and Ladra [12] for other approaches. The role played
by associative algebras is taken now by associative dialgebras, introduced by Loday
[20, 21] in the last decade of the 20 th century.
For nonassociative algebras, Malcev [23] showed that the commutator in an al-
ternative algebra satisfies the defining identities for Malcev algebras. This result
has been extended to the setting of dialgebras (see [6, Section 4] for details). How-
ever, nowadays it still remains an open problem whether any Malcev algebra is
isomorphic to a subalgebra of A− for some alternative algebra A. Some partial
results on this problem were obtained in [11, 32]. Therefore, trying to solve the
analogous problem in the dialgebra setting seems to be a very ambitious task.
The speciality of Malcev algebras has been also treated from another perspec-
tive: a nice result of P´erez-Izquierdo and Shestakov [28] establishes that any Malcev
algebra can be naturally embedded into a subalgebra of the generalized alternative
nucleus for some nonassociative algebra. The generalized alternative nucleus, previ-
ously introduced by Morandi and P´erez-Izquierdo [25] in the context of composition
algebras, turns out to be a Malcev algebra with the commutator.
The purpose of the present paper is to develop the necessary machinery to ap-
proach similar problems for Malcev dialgebras in future projects. The paper is
2010 Mathematics Subject Classification. Primary 17D10. Secondary 17A20, 17A30, 17A32,
17D05.
Key words and phrases. nonassociative algebras and dialgebras, nucleus, ternary derivations,
computer algebra.
1
2
S ´ANCHEZ-ORTEGA
organized as follows: in section 2 we gather together basic definitions from the the-
ory of associative and nonassociative dialgebras, and present a simplified statement
of the general Kolesnikov-Pozhidaev (KP) algorithm for converting an arbitrary
variety of multioperator algebras into a variety of dialgebras. The relationship be-
tween the KP algorithm and the Bremner-S´anchez-Ortega (BSO) algorithm (for
extending multilinear operations in an associative algebra to an associative dialge-
bra) is also described. In section 3 we apply the KP algorithm to the definining
identities for the generalized alternative nucleus; as a result, we obtain a system
of polynomial identities which will define the so-called generalized alternative di-
nucleus. The rest of the section focusses on the construction of a Malcev dialgebra
from the generalized alternative di-nucleus. Section 4 begins with some observations
about the necessity to have a nonlinear di-Malcev identity similar to the identity
for Malcev algebras expressed in terms of the Jacobian. After applying the BSO
to the Jacobian we find the natural candidate to be called the nonlinear di-Malcev
identity, but unfortunately it turns out not to be equivalent to the di-Malcev iden-
tity. Therefore, it is natural to ask about the existence of a nonlinear di-Malcev
identity; we close section 4 with this question. In Section 5 we formulate a general
conjecture about the speciality of Malcev dialgebras in terms of the generalized
alternative di-nucleus. Finally in the last section, inspired by a classical result of
Kleinfeld [14] which measures the associativeness of a semiprime algebra by impos-
ing some conditions on the associators and the associative nucleus; we introduce
the proper generalization of the associative nucleus to the setting of dialgebras and
study whether Kleinfeld's theorem remains true.
2. Preliminaries
2.1. Associative and alternative dialgebras. Leibniz algebras. Associative
dialgebras were introduced by Loday [20, 21] to provide a natural setting for Leibniz
algebras.
The concept of Leibniz algebra was originally introduced in the mid-1960's by
Bloh under the name "D-algebra".
Definition 2.1. (Bloh [2, 3], Loday [19], Cuvier [10].) A (right) Leibniz algebra
is a vector space L, with a bilinear map L × L → L, denoted (x, y) 7→ hx, yi,
satisfying the (right) Leibniz identity, which states that right multiplications
are derivations:
(L)
hhx, yi, zi ≡ hhx, zi, yi + hx, hy, zii.
If hx, xi ≡ 0 then the Leibniz identity is the Jacobi identity and L is a Lie algebra.
An associative algebra becomes a Lie algebra if the product xy is replaced by the
Lie bracket xy −yx. The notion of dialgebra gives, by a similar procedure, a Leibniz
algebra. Loday's idea was to replace the associative products xy and yx by two
distinct operations, so that the resulting bracket is not necessarily skew-symmetric.
Definition 2.2. (Loday [20]) A dialgebra is a vector space D with two bilinear
operations ⊣ : D × D → D and ⊢ : D × D → D, called the left and right products.
Definition 2.3. (Kolesnikov [16]) A 0-dialgebra is a dialgebra satisfying the left
and right bar identities:
(x ⊣ y) ⊢ z ≡ (x ⊢ y) ⊢ z,
x ⊣ (y ⊣ z) ≡ x ⊣ (y ⊢ z).
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
3
Bar identities say that on the bar side of the products, the operation symbols are
interchangeable.
Definition 2.4. (Loday [20]) An associative dialgebra is a 0-dialgebra satisfying
left, right and inner associativity:
(x ⊣ y) ⊣ z ≡ x ⊣ (y ⊣ z),
(x ⊢ y) ⊢ z ≡ x ⊢ (y ⊢ z),
(x ⊢ y) ⊣ z ≡ x ⊢ (y ⊣ z).
Definition 2.5. In any dialgebra D the dicommutator is the bilinear operation
hx, yi = x ⊣ y − y ⊢ x.
In what follows, we will write D− to denote (D, h−, −i), i.e., the underlying vector
space of D with the dicommutator.
It is easy to check that the dicommutator in any associative dialgebra D satisfies
the Leibniz identity, and hence D gives rise to a Leibniz algebra D−. Conversely,
to motivate the definition of an associative dialgebra, suppose we are given a vector
space D with bilinear maps ⊣ and ⊢, and we want to determine the identities
that must be satisfied so that the dicommutator satisfies the Leibniz identity. We
calculate as follows:
hhx, yi, zi − hhx, zi, yi − hx, hy, zii =
(cid:0)(x ⊣ y) ⊣ z − x ⊣ (y ⊣ z)(cid:1) −(cid:0)(x ⊣ z) ⊣ y − x ⊣ (z ⊢ y)(cid:1) −(cid:0)(y ⊢ x) ⊣ z − y ⊢ (x ⊣ z)(cid:1)
−(cid:0)y ⊢ (z ⊢ x) − (y ⊣ z) ⊢ x(cid:1) −(cid:0)z ⊢ (x ⊣ y) − (z ⊢ x) ⊣ y(cid:1) +(cid:0)z ⊢ (y ⊢ x) − (z ⊢ y) ⊢ x(cid:1).
If we set the differences within each pair of large parentheses to zero, we obtain
identities equivalent to the defining identities for associative dialgebras.
Ten years after Loday's definition of associative dialgebras, Liu [18] introduced
alternative dialgebras, the natural analogue of alternative algebras in the setting of
structures with two operations
Definition 2.6. (Liu [18]) An alternative dialgebra is a 0-dialgebra satisfying
(x, y, z)⊣ + (z, y, x)⊢ ≡ 0,
(x, y, z)⊣ − (y, z, x)⊢ ≡ 0,
(x, y, z)× + (x, z, y)⊢ ≡ 0,
where
(x, y, z)⊣ = (x ⊣ y) ⊣ z − x ⊣ (y ⊣ z),
(x, y, z)× = (x ⊢ y) ⊣ z − x ⊢ (y ⊣ z),
(x, y, z)⊢ = (x ⊢ y) ⊢ z − x ⊢ (y ⊢ z),
are the left, inner and right associators, respectively. We will refer to them as
the dialgebra associators. It is straightforward to check that every associative
dialgebra is an alternative dialgebra.
2.2. KP algorithm. Kolesnikov [16] introduced a general algorithm for trans-
forming the defining polynomial identities for a variety of binary algebras into the
defining identities for the corresponding variety of dialgebras. This procedure was
extended by Pozhidaev [29] to varieties of arbitrary n-ary algebras. In this subsec-
tion, we recall a simplified statement of the Kolesnikov-Pozhidaev (KP) algorithm
given in [5]. See Chapoton [9], Vallette [33], Kolesnikov and Voronin [17] for the
underlying construction of the KP algorithm in the theory of operads.
Consider a multilinear n-ary operation {−, . . . , −}, and introduce n new n-ary
operations {−, . . . , −}j distinguished by subscripts j = 1, . . . , n.
4
S ´ANCHEZ-ORTEGA
First, we introduce the following 0-identities for i, j = 1, . . . , n with i 6= j and
k, ℓ = 1, . . . , n; these identities say that the new operations are interchangeable in
argument i of operation j when i 6= j:
{a1, . . . , ai−1, {b1, · · · , bn}k, ai+1, . . . , an}j ≡
{a1, . . . , ai−1, {b1, · · · , bn}ℓ, ai+1, . . . , an}j.
Note that the 0-identities are generalizations of the bar identities for associative
dialgebras.
Second, we consider a multilinear identity I(a1, . . . , ad) of degree d in the n-ary
operation {−, . . . , −}. We apply the following rule to each monomial of the identity;
let aπ(1)aπ(2) . . . aπ(d) be such a monomial with some placement of operation sym-
bols where π is a permutation of 1, . . . , d. For i = 1, . . . , d we convert this monomial
into a new monomial of the same degree in the n new operations according to the
position of the variable ai, called the central argument of the monomial. For each
occurrence of the original operation, we have the following cases:
• If ai occurs in argument j then {. . . } becomes {. . . }j.
• If ai does not occur in any argument then
-- if ai occurs to the left of the original operation, {. . . } becomes {. . . }1,
-- if ai occurs to the right of the original operation, {. . . } becomes {. . . }n.
The resulting new identity is called the KP identity corresponding to I(a1, . . . , ad).
The choice of new operations, in the two subcases under the second bullet above,
gives a convenient normal form for the monomial: by the 0-identities, the subscripts
1 and n can be replaced by any other subscripts. Suppose that ai is the central
argument and that the identity I(a1, . . . , ad) contains a monomial of this form:
{. . . ,
{−, . . . , −}, . . . ,
{. . . , ai, . . . }, . . . ,
{−, . . . , −}, . . . }.
z
}
{
z
argument i
argument j
argument k
}
{
z
}
{
Since ai occurs in argument j, the outermost operation must receive subscript j:
{. . . , {−, . . . , −}, . . . , {. . . , ai, . . . }, . . . , {−, . . . , −}, . . . }j.
Our convention above attaches subscripts n and 1 to arguments i and k respectively:
{. . . , {−, . . . , −}n, . . . , {. . . , ai, . . . }, . . . , {−, . . . , −}1, . . . }j.
Since these subscripts occur in arguments i 6= j and k 6= j of operation j, the
0-identities imply that any other subscripts would give an equivalent identity.
Remark 2.7. Applying the KP algorithm to the associativity law {{x, y}, z} ≡
{x, {y, z}} gives the defining identities for an associative dialgebra. The defining
identities for an alternative dialgebra can be obtained by an application to the
KP algorithm to the linearization of right and left alternativity: (x, x, y) ≡ 0 and
(x, y, y) ≡ 0. (See [6, Examples 7 and 8])
2.3. BSO algorithm. Bremner and the author [4] have introduced an algorithm
(BSO) for extending multilinear operations in an associative algebra to correspond-
ing operations in an associative dialgebra. The BSO algorithm is based on the
following notion.
Definition 2.8. (Loday [21]) A dialgebra monomial in the free 0-dialgebra on
a set X of generators is a product x = x1x2 · · · xn where x1, . . . , xn ∈ X and the
bar indicates some placement of parentheses and some choice of operations. The
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
5
center of x is defined inductively: if n = 1 (x ∈ X) then c(x) = x; if n ≥ 2 then
x = y ⊣ z or x = y ⊢ z and c(y ⊣ z) = c(y) or c(y ⊢ z) = c(z) following the
direction of the product symbols. Using other words, the center of a monomial is
the element which has all the product symbols pointing inwards to it.
See [5, Definition 5.1] for a generalized statement of the BSO algorithm. The
input is an n-multilinear operation ω(x1, . . . , xn) in an associative algebra, and
the output are n operations in an associative dialgebra obtained by making xi the
center of each monomial of ω. For example, the Lie bracket [x, y] = xy − yx gives
rise to the operations:
[x, y]1 = x ⊣ y − y ⊢ x,
[x, y]2 = x ⊢ y − y ⊣ x.
Note that [x, y]2 = −[y, x]1, and moreover [x, y]1 = hx, yi, the dicommutator.
The BSO algorithm also works for nonassociative algebras and dialgebras. We
have already presented an example: the left, inner and right associators (see Defi-
nition 2.6) can be obtained by applying the BSO algorithm to the associator in a
nonassociative algebra.
2.4. Relation between the KP and BSO algorithms. In [5, Section 6] a gen-
eral conjecture was stated in terms of a commutative diagram relating the output
of the KP and BSO algorithms. Given ω, a multilinear n-ary operation in an asso-
ciative algebra, this conjecture established that under a mild technical condition,
the following two processes produce the same results:
• Find the identities satisfied by ω, and apply the KP algorithm.
• Apply the BSO algorithm to ω, and find the identities satisfied by ω1, . . . , ωn.
This conjecture has been recently proved by Kolesnikov and Voronin [17] using
operads.
2.5. Malcev dialgebras. Malcev dialgebras, the appropriate generalization of
Malcev algebras to the setting of dialgebras, have been recently introduced by
Bremner, Peresi and the author [6]. Malcev dialgebras are related to alternative
dialgebras in the same way that Malcev algebras are related to alternative algebras.
Before stating their definition, let us first recall the definition of a Malcev algebra.
Definition 2.9. (Malcev [23]) A Malcev algebra is a vector space with a bilinear
operation xy satisfying anticommutativity and the Malcev identity:
x2 ≡ 0,
(xy)(xz) ≡ ((xy)z)x + ((yz)x)x + ((zx)x)y.
Lemma 2.10. (Sagle [30]) If the characteristic is not 2, then an algebra is Malcev
if and only if it satisfies the following multilinear identities:
xy + yx ≡ 0,
(xz)(yt) ≡ ((xy)z)t + ((yz)t)x + ((zt)x)y + ((tx)y)z.
The defining identities for Malcev dialgebras were obtained by applying the KP
algorithm to the multilinear identities displayed in the previous lemma.
Definition 2.11. (Bremner, Peresi, JSO [6]) Over a field of characteristic not 2, a
(right) Malcev dialgebra is a vector space with a bilinear operation xy satisfying
right anticommutativity and the di-Malcev identity:
x(yz) + x(zy) ≡ 0,
((xy)z)t − ((xt)y)z − (x(zt))y − (xz)(yt) − x((yz)t) ≡ 0.
6
S ´ANCHEZ-ORTEGA
3. The generalized alternative di-nucleus
Malcev [23] showed that an alternative algebra becomes a Malcev algebra by
considering the same underlying vector space under the commutator. Bremner,
Peresi and the author [6] have used computer algebra to show that any subspace of
an alternative dialgebra closed under the dicommutator is a Malcev dialgebra.
A few years ago, in 2004, P´erez-Izquierdo and Shestakov [28] established a more
general way of constructing Malcev algebras. Given an algebra A, the generalized
alternative nucleus Nalt(A) of A, introduced by Morandi and P´erez-Izquierdo
[25], is defined as
Nalt(A) = {a ∈ A (a, x, y) = −(x, a, y) = (x, y, a) for all x, y ∈ A},
where (x, y, z) = (xy)z − x(yz) denotes the associator of A. As was pointed out in
[25], Nalt(A) may not be a subalgebra of A but it is closed under the commutator,
so it is a subalgebra of A−, that is, (A, [−, −]). Moreover, Nalt(A)− is a Malcev
algebra. (See [25, Proposition 4.3])
Remark 3.1. It is easy to see that the elements of Nalt(A) satisfy right and left
alternativity, i.e., the defining identities for an alternative algebra. In fact, if A is
an alternative algebra then Nalt(A) = A, and the construction of Malcev algebras
from alternative algebras is recovered.
In this section we introduce the analogue of the generalized alternative nucleus
for dialgebras. We prove that it is closed under the dicommutator and satisfies the
defining identities for Malcev dialgebras.
3.1. Definition of the generalized alternative di-nucleus of a 0-dialgebra.
In this subsection we apply the KP algorithm to the defining identities for the
generalized alternative nucleus Nalt(A) of an algebra A. Expanding the associators
and using the operation symbol {−, −} we get the following identities:
{{a, x}, y} + {{x, a}, y} − {a, {x, y}} − {x, {a, y}} ≡ 0,
{{x, y}, a} + {{x, a}, y} − {x, {y, a}} − {x, {a, y}} ≡ 0.
The KP identities are obtained by making a, x, y in turn the central argument:
{{a, x}1, y}1 + {{x, a}2, y}1 − {a, {x, y}1}1 − {x, {a, y}1}2 ≡ 0,
{{a, x}2, y}1 + {{x, a}1, y}1 − {a, {x, y}1}2 − {x, {a, y}1}1 ≡ 0,
{{a, x}2, y}2 + {{x, a}2, y}2 − {a, {x, y}2}2 − {x, {a, y}2}2 ≡ 0,
{{x, y}2, a}2 + {{x, a}2, y}1 − {x, {y, a}2}2 − {x, {a, y}1}2 ≡ 0,
{{x, y}1, a}1 + {{x, a}1, y}1 − {x, {y, a}1}1 − {x, {a, y}1}1 ≡ 0,
{{x, y}2, a}1 + {{x, a}2, y}2 − {x, {y, a}1}2 − {x, {a, y}2}2 ≡ 0,
Using the notation ⋆ ⊣ • = {⋆, •}1 and ⋆ ⊢ • = {⋆, •}2, the identities above become
(1)
(2)
(3)
(4)
(5)
(6)
(a ⊣ x) ⊣ y + (x ⊢ a) ⊣ y − a ⊣ (x ⊣ y) − x ⊢ (a ⊣ y) ≡ 0,
(a ⊢ x) ⊣ y + (x ⊣ a) ⊣ y − a ⊢ (x ⊣ y) − x ⊣ (a ⊣ y) ≡ 0,
(a ⊢ x) ⊢ y + (x ⊢ a) ⊢ y − a ⊢ (x ⊢ y) − x ⊢ (a ⊢ y) ≡ 0,
(x ⊢ y) ⊢ a + (x ⊢ a) ⊣ y − x ⊢ (y ⊢ a) − x ⊢ (a ⊣ y) ≡ 0,
(x ⊣ y) ⊣ a + (x ⊣ a) ⊣ y − x ⊣ (y ⊣ a) − x ⊣ (a ⊣ y) ≡ 0,
(x ⊢ y) ⊣ a + (x ⊢ a) ⊢ y − x ⊢ (y ⊣ a) − x ⊢ (a ⊢ y) ≡ 0.
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
7
Using the dialgebra associators, identities (1)-(6) can be rewritten as follows:
(a, x, y)⊣ + (x, a, y)× ≡ 0,
(a, x, y)× + (x, a, y)⊣ ≡ 0,
(a, x, y)⊢ + (x, a, y)⊢ ≡ 0,
(x, y, a)⊢ + (x, a, y)× ≡ 0,
(x, y, a)⊣ + (x, a, y)⊣ ≡ 0,
(x, y, a)× + (x, a, y)⊢ ≡ 0.
To finish, note that the 0-identities
{a, {b, c}1}1 ≡ {a, {b, c}2}1,
{{a, b}1, c}2 ≡ {{a, b}2, c}2,
become the bar identities by replacing the symbols {−, −}1, {−, −}2 by ⊣ and ⊢,
respectively. More precisely,
a ⊣ (b ⊣ c) ≡ a ⊣ (b ⊢ c),
(a ⊣ b) ⊢ c ≡ (a ⊢ b) ⊢ c.
The above calculations make possible to introduce the generalized alternative nu-
cleus for the setting of dialgebras.
Definition 3.2. The generalized alternative di-nucleus Nalt(D) of a 0-dialgebra
D is the set of elements a ∈ D which satisfies the bar identities and the following:
(GAN1)
(GAN2)
(GAN3)
(a, x, y)⊣ ≡ −(x, a, y)× ≡ (x, y, a)⊢,
(a, x, y)× ≡ −(x, a, y)⊣ ≡ (x, y, a)⊣,
(a, x, y)⊢ ≡ −(x, a, y)⊢ ≡ (x, y, a)×.
Proposition 3.3. Let D be a 0-dialgebra. Then the elements of Nalt(D) satisfy
the defining identities for an alternative dialgebra. Moreover, if D is alternative
then Nalt(D) = D.
Proof. Straightforward.
(cid:3)
Remark 3.4. In the setting of algebras, the generalized alternative nucleus does
not have a subalgebra structure. As a consequence, we can affirm that Nalt(D) is
not in general a sub-dialgebra of D.
Our next goal will be to show that Nalt(D) is closed under the dicommutator.
Our argument shares some of the ideas developed in [25, Section 4]. We start by
introducing some definitions.
Definition 3.5. Let D be a dialgebra and a ∈ D. The multiplication operators
L⊢
a : D → D are given by
a , L⊣
Definition 3.6. A ternary derivation of a 0-dialgebra D is a triple (δ1, δ2, δ3) ∈
EndF (D) × EndF (D) × EndF (D) such that
(7)
δ1(x ⊣ y) = δ2(x) ⊣ y + x ⊣ δ3(y),
δ1(x ⊢ y) = δ2(x) ⊢ y + x ⊢ δ3(y),
for all x, y ∈ D.
a , R⊢
L⊢
a , R⊣
a (x) = a ⊢ x, L⊣
for any x ∈ D. We also introduce T ×
a(x) = a ⊣ x, R⊢
a := L⊢
a (x) = x ⊢ a, R⊣
eT ×
a := R⊢
a + R⊣
a ,
a (x) = x ⊣ a,
a + L⊣
a .
8
S ´ANCHEZ-ORTEGA
The set of all ternary derivations Tder(D) of D has a Lie algebra structure with
the Lie bracket defined to be
(8)
[(δ1, δ2, δ3), (µ1, µ2, µ3)] := ([δ1, µ1], [δ2, µ2], [δ3, µ3]),
for all (δ1, δ2, δ3), (µ1, µ2, µ3) ∈ Tder(D). In case δ1 = δ2 = δ3, equation (7) says
that δ1 is a derivation of D.
Remark 3.7. Using the terminology of ternary derivations, it is easy to see that
an element a ∈ D satisfies (GAN2) and (GAN3) if and only if (L⊢
a),
(R⊣
a , −R⊣
At this point, one may ask about the remaining condition (GAN1). It is not
a ) ∈ Tder(D).
a , −L⊢
a , T ×
a , T ×
surprising that (GAN1) will be related to the operators L⊣
a , R⊢
a ) are no longer ternary
a , eT ×
a .
derivations of D. To be more precise, given a ∈ D, it follows that
a , −L⊣
a), (R⊢
a , −R⊢
a , eT ×
Unfortunately, the triples (L⊣
a , eT ×
(x, a, y)× ≡ −(a, x, y)⊢ ⇔ L⊣
(x, a, y)× ≡ −(x, y, a)⊢ ⇔ R⊢
a (x ⊣ y) = eT ×
a (x ⊢ y) = −R⊢
a (y),
a (x) ⊣ y − x ⊢ L⊣
a (x) ⊣ y + x ⊢ eT ×
a (y),
for any x, y ∈ D. Note that it is natural to obtain the above expressions since the
bar identities can be applied to get
L⊣
a (x ⊣ y) = L⊣
a (x ⊢ y),
R⊢
a (x ⊢ y) = R⊢
a (x ⊣ y),
for all x, y ∈ D.
Definition 3.8. A triple (δ1, δ2, δ3) ∈ EndF (D) × EndF (D) × EndF (D) of a 0-
dialgebra D, is called a quasi-ternary derivation if it satisfies
δ1(x ⊣ y) = δ2(x) ⊣ y + x ⊢ δ3(y),
δ1(x ⊢ y) = δ2(x) ⊣ y + x ⊢ δ3(y),
for all x, y ∈ D. Denote by QTder(D) the set of all quasi-ternary derivations of D.
Note that an element a ∈ D satisfies (GAN1) if and only if (L⊣
a , −R⊢
a),
a ) are quasi-ternary derivations. We have just proved the following
a , −L⊣
a , eT ×
(R⊢
result.
a , eT ×
a , T ×
a , eT ×
a , T ×
a , eT ×
Lemma 3.9. Let D be 0-dialgebra and a ∈ D. Then a ∈ Nalt(D) if and only if a
satisfies the following conditions:
a , −R⊣
a , −R⊢
a), (R⊣
a), (R⊢
a , −L⊢
a , −L⊣
a ) ∈ QTder(D).
a ) ∈ Tder(D).
(ii) (L⊣
(i) (L⊢
The following lemma relates the product of a ternary derivation with a quasi-
ternary derivation.
Lemma 3.10. Let D be a 0-dialgebra. If (δ1, δ2, δ3) ∈ QTder(D) and (µ1, µ2, µ3) ∈
Tder(D), then ([δ1, µ1], [δ2, µ2], [δ3, µ3]) ∈ QTder(D).
Proof. Given x, y ∈ D, we have
[δ1, µ1](x ⊣ y) = δ1(µ1(x ⊣ y)) − µ1(δ1(x ⊣ y)) = δ1(µ2(x) ⊣ y) + δ1(x ⊣ µ3(y))
− µ1(δ2(x) ⊣ y) − µ1(x ⊢ δ3(y)) = δ2(µ2(x)) ⊣ y + µ2(x) ⊢ δ3(y)
+ δ2(x) ⊣ µ3(y) + x ⊢ δ3(µ3(y)) − µ2(δ2(x)) ⊣ y − δ2(x) ⊣ µ3(y)
− µ2(x) ⊢ δ3(y) − x ⊢ µ3(δ3(y)) = [δ2, µ2](x) ⊣ y + x ⊣ [δ3, µ3](y),
as desired.
(cid:3)
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
9
Next, we derive some properties of the multiplication operators that will be
useful for our purposes.
Lemma 3.11. Let D be a 0-dialgebra, a, b ∈ Nalt(D) and x ∈ D. Then
(i) L⊢
(ii) L⊣
(iii) R⊢
(iv) R⊣
(v) [L⊢
(vi) L⊢
a , L⊢
x],
a , L⊢
x],
x , L⊣
a ],
x , L⊢
a ],
a⊣x = L⊢
a⊣x = L⊣
a⊣x = R⊣
a⊣x = R⊣
a , R⊣
ha,bi = [L⊢
a L⊢
a L⊣
x R⊢
x R⊣
b ] = [R⊣
a , L⊢
x + [R⊣
x + [R⊢
a + [R⊣
a + [R⊣
a , L⊢
b ],
b ] + 2[R⊣
[L⊣
a , R⊣
a , L⊢
b ],
a + [L⊢
a + [L⊢
x + [L⊣
x + [L⊢
x, R⊣
a ].
x, R⊢
a ].
a , R⊣
x ].
a , R⊣
x ].
L⊢
L⊣
xL⊢
xL⊣
a R⊢
a R⊣
x⊢a = L⊢
x⊢a = L⊢
x⊢a = R⊢
R⊢
x⊢a = R⊣
a , L⊢
b ].
ha,bi = [L⊣
L⊣
b ] = [R⊢
R⊣
(vii) −L⊢
(viii) R⊣
ha,bi = [L⊢
ha,bi = −[R⊣
a , L⊢
b ] − 2[T ×
a , L⊢
b ], −L⊣
a , R⊣
b ] − 2[L⊢
a , R⊣
b ],
a , L⊢
b ] + 2[R⊢
a , L⊢
b ].
b ] − 2[eT ×
a , R⊣
a , L⊢
b ].
a , L⊢
ha,bi = [L⊣
ha,bi = −[R⊢
R⊢
b ] − 2[L⊣
a , R⊣
b ].
b ] + 2[eT ×
a , R⊣
b ].
(x) T ×
(xi) eT ×
ha,bi = [T ×
ha,bi = [eT ×
b ] = −[T ×
b ] = −[eT ×
(ix) −R⊣
ha,bi = −[R⊣
a , T ×
a , R⊣
b ] + 2[T ×
a , T ×
b ] − 2[R⊣
a , R⊣
b ], −R⊢
a , T ×
ha,bi = −[R⊢
b ] + 2[L⊢
a , T ×
b ].
a , R⊣
a , T ×
b ] − 2[R⊢
a , T ×
a , T ×
b ] + 2[L⊣
a , T ×
b ].
Proof. Let a, b ∈ Nalt(D) and x, y ∈ D.
(i). Applying the left bar identity we get
L⊢
a⊣x(y) = (a ⊣ x) ⊢ y = (a ⊢ x) ⊢ y.
On the other hand, we have
a , L⊢
x(y) + [R⊣
L⊢
a⊣x(y) = L⊢
x(y) + [R⊣
a L⊢
a L⊢
x](y) = a ⊢ (x ⊢ y) + (x ⊢ y) ⊣ a − x ⊢ (y ⊣ a).
Thus, L⊢
holds since a ∈ Nalt(D). Analogously, one can show that L⊢
[L⊢
x](y) if and only if (a, x, y)⊢ = −(x, y, a)× which
a (y) +
x⊢a(y) = L⊢
a , L⊢
xL⊢
a ](y).
x, R⊣
(ii), (iii) and (iv) can be proved analogously.
(v). By definition we have
[L⊢
[R⊣
a , R⊣
a , L⊢
b ](y) = a ⊢ (y ⊣ b) − (a ⊢ y) ⊣ b = −(a, y, b)×,
b ](y) = b ⊢ (y ⊣ a) − (b ⊢ y) ⊣ a = (b, y, a)×.
Applying that b ∈ Nalt(D) from (GAN2) we obtain (b, y, a)× = (y, a, b)⊣. Since
a ∈ Nalt(D) a second use of (GAN2) gives (y, a, b)⊣ = −(a, y, b)×. The proof of
the second equality is similar: apply (GAN3) with b ∈ Nalt(D), and (GAN1) with
a ∈ Nalt(D).
(vi) follows from (i), (ii) and (v).
(vii) follows from (vi) and the definitions of T ×
(viii) is a consequence of (iii), (iv) and (v).
(ix) is obtained by an application of (viii), taking into account the definitions of
a , eT ×
a .
T ×
a .
a , eT ×
(x). Applying (vi), (viii) and (v) we get
T ×
ha,bi = L⊢
ha,bi + R⊣
ha,bi = [L⊢
a , L⊢
b ] − [R⊣
a , R⊣
b ].
10
S ´ANCHEZ-ORTEGA
On the other hand
(9)
[T ×
a , T ×
b ] = [L⊢
a + R⊣
a , L⊢
b + R⊣
b ] = [L⊢
a , L⊢
b ] + 2[R⊣
a , L⊢
b ] + [R⊣
a , R⊣
b ],
which implies
T ×
ha,bi = [T ×
a , T ×
b ] − 2[R⊣
a , L⊢
b ] − 2[R⊣
a , R⊣
b ] = [T ×
a , T ×
b ] − 2[R⊣
a , T ×
b ]
= −[T ×
a , T ×
b ] + 2[L⊢
a , T ×
b ].
(xi) can be shown similarly.
(cid:3)
Proposition 3.12. The generalized alternative di-nucleus of a 0-dialgebra is closed
under the dicommutator.
Proof. Let D be a 0-dialgebra and Nalt(D) its generalized alternative di-nucleus.
Given a, b ∈ Nalt(D); in order to show that ha, bi ∈ Nalt(D) we are going to use
the characterization in terms of ternary and quasi-ternary derivations described
in Lemma 3.9. Let us start by proving the claim that (L⊢
ha,bi) ∈
Tder(D). Since a, b ∈ Nalt(D), Lemma 3.9 (i) allows us to conclude that
ha,bi, −L⊢
ha,bi, T ×
a , −L⊢
a ),
(L⊢
b , T ×
b , −L⊢
b ),
(R⊣
a , −R⊣
a , T ×
a ) ∈ Tder(D),
(L⊢
a , T ×
which implies that
a , L⊢
(cid:0)[L⊢
(cid:0)[L⊢
a , L⊢
= (cid:0)[L⊢
= (L⊢
b ], [T ×
a , L⊢
ha,bi, T ×
b ], [T ×
a , T ×
b ], [L⊢
a , L⊢
a , L⊢
b ], [−R⊣
a , T ×
b ], [T ×
a , −L⊢
On the other hand, applying Lemma 3.11 (vi), (vii), (x) we get
b ], [T ×
a , L⊢
a , T ×
b ] + 2[R⊣
b ], [−R⊣
a , T ×
a , T ×
b ], [L⊢
b ] + 2[−R⊣
b ], [L⊢
a , L⊢
a , L⊢
a , L⊢
b ](cid:1) + 2(cid:0)[R⊣
b ](cid:1) + 2(cid:0)[R⊣
a , T ×
b ], [T ×
b ](cid:1) ∈ Tder(D).
b ](cid:1)
a , −L⊢
b ](cid:1)
b ] + 2[T ×
a , −L⊢
a , T ×
b ](cid:17) − 2(cid:16)[L⊣
a , R⊣
b ], −[R⊢
b ], [eT ×
a , R⊣
a , R⊣
b ] + 2[eT ×
a , −R⊣
b ], −[eT ×
b ](cid:17)
a , T ×
b ] + 2[L⊣
b ], [−L⊣
a , T ×
b ](cid:17)
a , T ×
(cid:3)
Theorem 3.13. Let D be a 0-dialgebra over a field of characteristic not 2 or 3.
Then (Nalt(D), h−, −i) is a Malcev dialgebra.
Proof. Taking into account Propositions 3.3 and 3.12, the result follows from the
fact that every subspace of an alternative dialgebra, which is closed under the
dicommutator, is a Malcev dialgebra. (See [6, Section 4] for details.)
(cid:3)
ha,bi, −R⊣
ha,bi, T ×
ha,bi, −R⊢
It remains
ha,bi) ∈ QTder(D). Let
ha,bi) ∈ QTder(D). Lemma 3.9 (ii) yields
a ) ∈ QTder(D), while from Lemma 3.9 (i) we get
ha,bi) ∈ Tder(D).
ha,bi, eT ×
ha,bi), (R⊢
ha,bi, eT ×
b ) ∈ Tder(D). Now apply Lemma 3.10 to conclude that
b ](cid:17)−2(cid:16)[L⊣
a , R⊣
b ], [eT ×
a , −R⊣
b ], [−L⊣
a , T ×
b ](cid:17) ∈ QTder(D).
On the other hand by Lemma 3.11 (viii), (ix), (xi), we obtain
ha,bi, −L⊢
ha,bi),
which concludes the proof of the claim.
Similarly, one can show that (R⊣
to check that (L⊣
us now prove that (R⊢
(L⊣
that (R⊣
ha,bi, eT ×
a , −R⊢
a , −L⊣
b , −R⊢
a), (R⊢
b , T ×
a , eT ×
−(cid:16)[R⊢
a , R⊣
b ], [R⊢
a , R⊣
b ], [eT ×
ha,bi, −L⊣
ha,bi, −R⊢
a , eT ×
a , T ×
−(cid:16)[R⊢
a , R⊣
= (cid:16)−[R⊢
b ], [R⊢
a , R⊣
= (R⊢
ha,bi, −R⊢
a , R⊣
b ], [eT ×
b ] − 2[L⊣
a , R⊣
ha,bi, eT ×
ha,bi),
which finishes the proof.
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
11
4. On the search for a nonlinear di-Malcev identity
In the previous section, we have shown that the generalized alternative di-nucleus
of a 0-dialgebra, endowed with the dicommutator, is a Malcev dialgebra (see Theo-
rem 3.13). To prove Theorem 3.13, we have used that any subspace of an alternative
dialgebra, which is closed under the dicommutator, is a Malcev dialgebra. A more
interesting fact would be to prove Theorem 3.13 independently; since it would give
us a more general construction of Malcev dialgebras. For this task, based on the
proof of the corresponding result for algebras (see [25, Proposition 4.3] and [26,
p. 9]), one can expect that it will be crucial to have a nonlinear version of the
di-Malcev identity.
Motivated by the following result, due to Myung, our first step will be to intro-
duce the analogous operation to the Jacobian for dialgebras.
Proposition 4.1. [26, Proposition 1.1] In a Malcev algebra, the Malcev identity is
equivalent to the identity
(10)
J(x, y, xz) ≡ J(x, y, z)x,
where J(x, y, z) = (xy)z + (yz)x + (zx)y is the Jacobian.
Inspired by the fact that the Jacobian vanishes in any Lie algebra, we introduce
the following trilinear operation on any Malcev dialgebra:
L(x, y, z) = (xy)z − x(yz) − (xz)y.
Note that L equals to zero in every (right) Leibniz algebra. We will refer to L as
the di-Jacobian. The following two remarks justify our terminology.
Remark 4.2. In a Malcev algebra, an application of the anticommutativity tells
us that the di-Jacobian coincides with the Jacobian.
Remark 4.3. The di-Jacobian could also be obtained by applying the BSO al-
gorithm to the Jacobian. In fact, making x, y and z the center in the Jacobian
gives
J1(x, y, z) = (x ⊣ y) ⊣ z + (y ⊢ z) ⊢ x + (z ⊢ x) ⊣ y,
J2(x, y, z) = (x ⊢ y) ⊣ z + (y ⊣ z) ⊣ x + (z ⊢ x) ⊢ y,
J3(x, y, z) = (x ⊢ y) ⊢ z + (y ⊢ z) ⊣ x + (z ⊣ x) ⊢ y.
Since J2(x, y, z) = J1(y, z, x) and J3(x, y, z) = J1(z, x, y), we discard J2 and J3,
and we retain J1. At this point, note that in a Malcev dialgebra with product xy,
the right product is superfluous, since x ⊣ y = −y ⊢ x = xy (see [6, Section 3] for
more details). Thus, rewriting J1(x, y, z) in terms of the operation xy and applying
right anticommutativity, we obtain
J1(x, y, z) = (x ⊣ y) ⊣ z + (y ⊢ z) ⊢ x + (z ⊢ x) ⊣ y
= (xy)z + x(zy) − (xz)y = (xy)z − x(yz) − (xz)y = L(x, y, z),
as claimed.
In what follows, we will show that the di-Jacobian satisfies some properties
analogous to those of the Jacobian. We start by recalling some basic notions and
facts about the so-called Malcev admissible algebras.
12
S ´ANCHEZ-ORTEGA
An algebra A is called Malcev admissible if A− = (A, [−, −]) is a Malcev
algebra. In the theory of Malcev admissible algebras, the trilinear operation
(11)
S(x, y, z) := (x, y, z) + (y, z, x) + (z, x, y),
plays an important role. More precisely, given an algebra A, over a field F of
arbitrary characteristic, expanding the associators, we get
(12)
S(x, y, z) − S(x, z, y) = JA−(x, y, z),
where JA− stands for the Jacobian of A−.
If A is flexible, then the function S
alternates in its second and third arguments, i.e., A satisfies S(x, y, z) ≡ −S(x, z, y),
and so (12) applies to get
(13)
2S(x, y, z) ≡ JA−(x, y, z).
From (13) and Proposition 4.1 (see also [26, Lemma 1.2 (ii)]) it follows that a
flexible algebra is Malcev admissible if and only if the following identity is satisfied.
(14)
2S(x, y, [x, z]) ≡ 2[S(x, y, z), x].
Coming back to the dialgebra setting, our first task will be to introduce the
analogue of the operation S. To this end, we first expand the associators in (11):
(15)
S(x, y, z) = (xy)z − x(yz) + (yz)x − y(zx) + (zx)y − z(xy).
Next, applying the BSO algorithm (by making x the center in each monomial)
produces the following trilinear operation in a nonassociative dialgebra:
eS(x, y, z) = (x ⊣ y) ⊣ z − x ⊣ (y ⊣ z) + (y ⊢ z) ⊢ x − y ⊢ (z ⊢ x) + (z ⊢ x) ⊣ y
− z ⊢ (x ⊣ y) = (x, y, z)⊣ + (y, z, x)⊢ + (z, x, y)×.
Note that making y or z the center in (15) does not give anything new: if Si(x, y, z)
is the operation obtained from S(x, y, z) by making the i-th argument the center,
then eS(x, y, z) = S1(x, y, z) = S2(z, x, y) = S3(y, z, x).
Definition 4.4. A 0-dialgebra D is called Malcev admissible if D− = (D, h−, −i)
is a Malcev dialgebra.
We have already seen an example of a Malcev admissible dialgebra, that is, the
generalized alternative di-nucleus Nalt(D) of a 0-dialgebra D.
Recall (see [7, Section 7] for details) that a flexible dialgebra is a 0-dialgebra
which satisfies the identities:
(16)
(x, y, z)⊣ + (z, y, x)⊢ ≡ 0,
(x, y, z)× + (z, y, x)× ≡ 0.
Note that the first identity in (16) coincides with the first identity in the definition
of an alternative dialgebra. Moreover, every alternative dialgebra is flexible.
The following result collects together some properties of the operation eS.
Lemma 4.5. Let D be a flexible dialgebra over a field of characteristic different
from 2. Then
(i) eS(x, y, z) = −eS(x, z, y),
(ii) D− is a Leibniz algebra if and only if eS(x, y, z) ≡ 0.
2eS(x, y, z) = LD−(x, y, z) for all x, y, z ∈ D.
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
13
Proof. (i). For x, y, z ∈ D, it follows
eS(x, y, z) = (x, y, z)⊣ + (y, z, x)⊢ + (z, x, y)×
(16)
Moreover, applying the bar identities we get
≡ −(z, y, x)⊢ − (x, z, y)⊣ − (y, x, z)× = −eS(x, z, y).
LD−(x, y, z) = hhx, yi, zi − hx, hy, zii − hhx, zi, yi = (x ⊣ y − y ⊢ x) ⊣ z
− z ⊢ (x ⊣ y − y ⊢ x) − x ⊣ (y ⊣ z − z ⊢ y) + (y ⊣ z − z ⊢ y) ⊢ x
− (x ⊣ z − z ⊢ x) ⊣ y + y ⊢ (x ⊣ z − z ⊢ x) ≡ (x, y, z)⊣ + (y, z, x)⊢
+ (z, x, y)× − (x, z, y)⊣ − (z, y, x)⊢ − (y, x, z)× = eS(x, y, z) − eS(x, z, y)
= 2eS(x, y, z),
as desired.
(ii) is a consequence of (i).
(cid:3)
Inspired by (10), we introduce the following nonlinear identity in a Malcev dial-
gebra
(17)
L(y, x, zx) ≡ L(y, z, x)x.
Remark 4.6. Note that in a Malcev algebra identities (10) and (17) turn to be
equal. This follows from the skew symmetries of the Jacobian and the anticommu-
tativity law.
Remark 4.7. In a Malcev admissible dialgebra, identity (17) can be rewritten, in
terms of the operation eS, as follows:
(18)
eS(x, y, hz, yi) ≡ heS(x, z, y), yi.
At this point, it is natural ask whether identity (17) will be the nonlinear ana-
logue of the di-Malcev identity. Unfortunately, against what seems to be natural
identity (17) and the di-Malcev identity turn out to be non-equivalent. We will use
computer algebra to prove this claim. To this end, we will regard the subspace of all
identities of degree n for a certain algebra A as a module over the symmetric group
Sn acting by permutations of the variables. Given identities f, f1, . . . , fk of degree
n, we say that f is a consequence of f1, . . . , fk if f belongs to the Sn-submodule
generated by f1, . . . , fk.
Theorem 4.8. The di-Malcev identity is not equivalent to identity (17) in the free
right anticommutative algebra.
Proof. A binary operation has five association types in degree 4, namely:
((ab)c)d,
(a(bc))d,
(ab)(cd),
a(b(cd)),
a((bc)d).
An application of the right anticommutativity law eliminates type 4, since a(b(cd)) =
−a((cd)b). Moreover, types 2, 3 and 5 have the following skew-symmetries:
(19)
(a(cb))d = −(a(bc))d,
(ab)(dc) = −(ab)(cd),
a((cb)d) = −a((bc)d).
Each skew-symmetry halves the number of multilinear monomials, giving the 60
monomials of Table 1 which form an ordered basis of the multilinear subspace of
degree 4 in the free right anticommutative algebra on four generators.
14
S ´ANCHEZ-ORTEGA
((ab)c)d,
((ba)c)d,
((ca)b)d,
((da)b)c,
(a(bc))d,
(c(ab))d,
(ab)(cd),
(ca)(bd),
a((bc)d),
c((ab)d),
((ab)d)c,
((ba)d)c,
((ca)d)b,
((da)c)b,
(a(bd))c,
(c(ad))b,
(ac)(bd),
(cb)(ad),
a((bd)c),
c((ad)b),
((ac)b)d,
((bc)a)d,
((cb)a)d,
((db)a)c,
(a(cd))b,
(c(bd))a,
(ad)(bc),
(cd)(ab),
a((cd)b),
c((bd)a),
((ac)d)b,
((bc)d)a,
((cb)d)a,
((db)c)a,
(b(ac))d,
(d(ab))c,
(ba)(cd),
(da)(bc),
b((ac)d),
d((ab)c),
((ad)b)c,
((bd)a)c,
((cd)a)b,
((dc)a)b,
(b(ad))c,
(d(ac))b,
(bc)(ad),
(db)(ac),
b((ad)c),
d((ac)b),
((ad)c)b,
((bd)c)a,
((cd)b)a,
((dc)b)a,
(b(cd))a,
(d(bc))a,
(bd)(ac),
(dc)(ab),
b((cd)a),
d((bc)a).
Table 1. Right anticommutative monomials in degree 4
We first process identity (17). We create a 48 × 60 matrix M , initialized to zero.
We fill the first 24 rows with the coefficient vectors obtained by applying all 24
permutations of the variables a, b, c, d to identity (17) and straightening the terms
by using right anticommutativity. The rank of the resulting matrix is 8. Next, we
perform the same calculations with the di-Malcev identity and store the resulting
vectors in rows 25−48 of M ; the rank is now 20. We then reverse this procedure, first
processing the di-Malcev identity, obtaining rank 20 and then processing identity
(17), which does not increase the rank.
We conclude that identity (17) is a consequence of the di-Malcev identity but the
converse is not true: the di-Malcev identity can not be obtained from identity (17).
These calculations show that identity (17) and the di-Malcev identity generate a
20-dimensional subspace in the 60-dimensional space spanned by the right anticom-
mutative monomials. Moreover, identity (17) generates a 8-dimensional subspace
while the di-Malcev identity generates the entire 20-dimensional subspace.
These calculations were performed by using the Maple 16 package LinearAlgebra.
(cid:3)
Question 4.9. The results of the present section make us to ask whether there
exists a nonlinear identity, which has an expression in terms of the di-Jacobian,
equivalent to the di-Malcev identity.
5. Conjecture: speciality on Malcev dialgebras
P´erez-Izquierdo and Shestakov [28] proved that any Malcev algebra is isomorphic
to a subalgebra of the generalized alternative nucleus Nalt(A) of a certain algebra A.
More precisely, given a Malcev algebra M they constructed an algebra U(M ), and
a monomorphism ι : M → U(M )− such that the image of M lies in the generalized
alternative nucleus of U(M ), and U(M ) is a universal object with respect to such
homomorphisms. They showed that U(M ) has a basis of Poincar´e-Birkhoff-Witt
type over M , and inherits some good properties of universal enveloping algebras of
Lie algebras.
Motivated by this result and based on the results of the previous sections, it
seems natural to ask whether any Malcev dialgebra arises from a subalgebra of the
generalized alternative di-nucleus of a certain 0-dialgebra. We leave it as an open
problem.
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
15
6. The associative di-nucleus
The associators vanish in any associative algebra. Concerning to nonassociative
algebras, several authors have analyzed what happens if we impose that the asso-
ciators satisfy certain polynomial identities. For example, Thedy [34] studied the
case in which all associators commute with all the elements. Later on, Kleinfeld and
Widmer [15] considered rings which associators satisfy (x, y, z) = (y, z, x); previ-
ously studied by Outcalt [27] and Sterling [31]. In the present section, we focus our
attention on a result due to Kleinfeld [14], which stated that a semiprime algebra
with all its associators in the associative nucleus is associative.
Let us recall that the associative nucleus N(A) of an algebra A is defined by
N(A) = {a ∈ A (a, A, A) = (A, a, A) = (A, A, a) = 0}.
As we have seen, the theory of dialgebras is not entirely analogous to the theory
of algebras; in the sense that we can not translate directly an arbitrary result
from algebras to dialgebras, and hope that the resulting result will also hold in
the dialgebra setting. Likely, in this section, we will show that the analogue to
Kleinfeld's theorem holds for a 0-dialgebra.
The definition of the associative di-nucleus of a 0-dialgebra can be obtained by
applying the KP algorithm to the defining identities for the associative nucleus
N(A) of an algebra A. Proceeding as in subsection 3.1 (we omit here the details)
we will obtain the following definition.
Definition 6.1. Let D be a 0-dialgebra. The associative di-nucleus N(D) of
D is the set of elements a ∈ D, which satisfies the bar identities jointly with the
following identities.
(AN1)
(AN2)
(AN3)
(a, D, D)⊣ ≡ (D, a, D)⊣ ≡ (D, D, a)⊣ ≡ 0,
(a, D, D)× ≡ (D, a, D)×≡ (D, D, a)× ≡ 0,
(a, D, D)⊢ ≡ (D, a, D)⊢ ≡ (D, D, a)⊢ ≡ 0.
One of the differences between the generalized alternative di-nucleus Nalt(D) and
the associative nucleus N(D) is that N(D) is a subdialgebra of D.
In order to prove this important property of N(D), we need to introduce some
notions. The following identity, so-called the Teichmuller identity
(T)
(wx, y, z) − (w, xy, z) + (w, x, yz) ≡ w(x, y, z) + (w, x, y)z,
holds in any algebra. Due to the relation between the BSO and the KP algo-
rithms the following identities hold in any 0-dialgebra. We will refer to them as the
Teichmuller di-identities.
(T1)
(w ⊣ x, y, z)⊣ − (w, x ⊣ y, z)⊣ + (w, x, y ⊣ z)⊣ ≡ w ⊣ (x, y, z)⊣ + (w, x, y)⊣ ⊣ z,
(T2)
(w ⊢ x, y, z)⊣ − (w, x ⊣ y, z)× + (w, x, y ⊣ z)× ≡ w ⊢ (x, y, z)⊣ + (w, x, y)× ⊣ z,
(T3)
(w ⊢ x, y, z)× − (w, x ⊢ y, z)× + (w, x, y ⊣ z)⊢ ≡ w ⊢ (x, y, z)× + (w, x, y)⊢ ⊣ z,
(T4)
(w ⊢ x, y, z)⊢ − (w, x ⊢ y, z)⊢ + (w, x, y ⊢ z)⊢ ≡ w ⊢ (x, y, z)⊢ + (w, x, y)⊢ ⊢ z.
16
S ´ANCHEZ-ORTEGA
Note that (T1)-(T4) are obtained by expanding the associators in (T), and making
w, x, y and z, respectively, the center of each monomial.
Lemma 6.2. The associative di-nucleus N(D) of a 0-dialgebra D is a subdialgebra.
Proof. We will show that N(D) is closed under the left product. Similarly, one can
prove that it is also closed under the right product.
Given a, b ∈ N(D) applying (T1), by taking into account that (AN1) holds for a
and b, we get that a ⊣ b satisfies (AN1). Applications of the left bar identity and
(T4) give that (AN3) is satisfied by a ⊣ b. To finish, in order to show that a ⊣ b
also verifies (AN2) apply the left bar identity jointly with (T2) and (T3).
(cid:3)
Any nonassociative algebra has a particular ideal, called the associator ideal
defined to be the smallest ideal which contains all associators. Kleinfeld [14] noticed
that its elements are either finite sums of associators or right multiples of associ-
ators. In what follows, we will develop the necessary machinery to find a similar
notion in the dialgebra setting.
Definition 6.3. A subspace I of a dialgebra D is called a di-ideal if it satisfies
that I ⊣ D, I ⊢ D, D ⊣ I, D ⊢ I ⊆ I.
Let D be a 0-dialgebra, let us denote by Assoc(D) the set consisting of all finite
sums of dialgebra associators of D jointly with all its right multiples of dialgebra
associators of D. To be more precise, an arbitrary element of Assoc(D) is of one of
the following types:
• A finite sum of dialgebra associators:
(x, y, z)⊣,
(x, y, z)×,
(x, y, z)⊢
• A right multiple of a dialgebra associator:
(x, y, z)⊣ ⊣ t,
(x, y, z)× ⊣ t,
(x, y, z)⊢ ⊣ t,
(x, y, z)⊢ ⊢ t
where x, y, z, t ∈ D.
Remark 6.4. Notice that the bar identities apply to get
(x, y, z)⊣ ⊢ t ≡ (x, y, z)× ⊢ t ≡ (x, y, z)⊢ ⊢ t.
Lemma 6.5. Let D be a 0-dialgebra. Then Assoc(D) is a di-ideal of D. Moreover,
Assoc(D) is the smallest di-ideal of D containing all the dialgebra associators.
Proof. Due to the bar identities, the result follows by noticing the following:
((x, y, z)⋆ ⊣ t) ⊣ u = ((x, y, z)⋆, t, u)⊣ − (x, y, z)⋆ ⊣ (t ⊣ u),
((x, y, z)⊢ ⊢ t) ⊢ u = ((x, y, z)⊢, t, u)⊢ − (x, y, z)⊢ ⊢ (t ⊢ u),
u ⊣ (x, y, z)⊣
(T1)
≡ (u ⊣ x, y, z)⊣ − (u, x ⊣ y, z)⊣ + (u, x, y ⊣ z)⊣ − (u, x, y)⊣ ⊣ z,
u ⊢ (x, y, z)⊣
(T2)
≡ (u ⊢ x, y, z)⊣ − (u, x ⊣ y, z)× + (u, x, y ⊣ z)× − (u, x, y)× ⊣ z,
u ⊢ (x, y, z)×
(T3)
≡ (u ⊢ x, y, z)× − (u, x ⊢ y, z)× + (u, x, y ⊣ z)⊢ − (u, x, y)⊢ ⊣ z,
(T4)
≡ (u ⊢ x, y, z)⊢ − (u, x ⊢ y, z)⊢ + (u, x, y ⊢ z)⊢ − (u, x, y)⊢ ⊢ z,
u ⊢ (x, y, z)⊢
u ⊢ ((x, y, z)⋆ ⊣ t) = (u, (x, y, z)⋆, t)× − (u ⊢ (x, y, z)⋆) ⊣ t,
u ⊢ ((x, y, z)⊢ ⊢ t) = (u, (x, y, z)⊢, t)⊢ − (u ⊢ (x, y, z)⊢) ⊢ t,
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
u ⊣ ((x, y, z)⊣ ⊣ t) = (u, (x, y, z)⊣, t)⊣ − (u ⊣ (x, y, z)⊣) ⊣ t,
where ⋆ ∈ {⊣, ×, ⊢}.
17
(cid:3)
Definition 6.6. The associator di-ideal of a 0-dialgebra D is the di-ideal Assoc(D).
Definition 6.7. We say that a dialgebra is semiprime if it does not contain di-
ideals which left and right square products are zero; that is, I ⊣ I = I ⊣ I = 0
implies I = 0 for every di-ideal I.
Theorem 6.8. Let D be a 0-dialgebra over a field of characteristic not 2. Assume
that D is semiprime and has all its associators in N(D). Then D is an associative
dialgebra.
Lemma 6.9. Let D a 0-dialgebra. Then for a ∈ N(D), x, y, z ∈ D and ⋆ ∈ {⊣
, ×, ⊢}, the following identities hold:
(20)
(21)
(22)
(23)
(24)
(25)
(26)
(27)
(a ⊣ x, y, z)⊣ ≡ a ⊣ (x, y, z)⊣,
(a ⊢ x, y, z)⋆ ≡ a ⊢ (x, y, z)⋆,
(x ⊣ a, y, z)⋆ ≡ (x, a ⊣ y, z)⋆,
(x ⊢ a, y, z)⊣ ≡ (x, a ⊣ y, z)×,
(x, y ⊣ a, z)⋆ ≡ (x, y, a ⊢ z)⋆,
(x, y ⊢ a, z)× ≡ (x, y, a ⊣ z)⊢,
(x, y, z ⊣ a)⋆ ≡ (x, y, z)⋆ ⊣ a,
(x, y, z ⊢ a)⊢ ≡ (x, y, z)⊢ ⊢ a.
Proof. It follows by applying the Teichmuller di-identities and the bar identities. (cid:3)
Proof of Theorem 6.8. Write I to denote the associative di-nucleus Assoc(D) of D.
We are going to show that I ⊣ I = I ⊢ I = 0 which, by the semiprimeness of D,
will allow us to conclude that I = 0.
Given x, y, z, t, u, v ∈ D and ⋆ ∈ {⊣, ×, ⊢} we claim that
(28)
(29)
(x, y, z)⋆ ⊣ (t, u, v)⊣ ≡ (x, y, z)⋆ ⊣ (t, u, v)×≡ (x, y, z)⋆ ⊣ (t, u, v)⊢ = 0,
(x, y, z)⊣ ⊢ (t, u, v)⋆ ≡ (x, y, z)× ⊢ (t, u, v)⋆≡ (x, y, z)⊢ ⊢ (t, u, v)⋆ = 0,
Set p := (x, y, z)⊣ ⊣ (t, u, v)⊣. Since (x, y, z)⊣ ∈ N(D), an application of (20)
gives p ≡ ((x, y, z)⊣ ⊣ t, u, v)⊣. On the other hand from (T1) we obtain
(x, y, z)⊣ ⊣ t ≡ (x ⊣ y, z, t)⊣ − (x, y ⊣ z, t)⊣ + (x, y, z ⊣ t)⊣ − x ⊣ (y, z, t)⊣,
which by the hypothesis yields p ≡ −(x ⊣ (y, z, t)⊣, u, v)⊣. Next (22) and (T1)
apply to get p ≡ −(x, (y, z, t)⊣ ⊣ u, v)⊣ ≡ (x, y ⊣ (z, t, u)⊣v)⊣. Using the right bar
identity, (24) and (T1) we have p ≡ (x, y, (z, t, u)⊣ ⊣ v)⊣ ≡ −(x, y, z ⊣ (t, u, v)⊣)⊣.
To finish apply (26) to get p ≡ −(x, y, z)⊣ ⊣ (t, u, v)⊣ = −p. Thus 2p ≡ 0 and
therefore p ≡ 0. Reasoning in a similar way, one can complete the proof of (28)
and show (29).
Next, for x, y, z, s, t, u, v, w ∈ D and ⋆ ∈ {⊣, ×, ⊢} we claim that
(30)
(31)
((x, y, z)⋆ ⊣ t) ⊣ ((u, v, w)⊣ ⊣ s) ≡ ((x, y, z)⋆ ⊣ t)⊣ ((u, v, w)× ⊣ s) ≡
((x, y, z)⋆ ⊣ t) ⊣ ((u, v, w)⊢ ⊣ s) ≡ ((x, y, z)⋆ ⊣ t) ⊣ ((u, v, w)⊢ ⊢ s) = 0,
((x, y, z)⊣ ⊣ t) ⊢ ((u, v, w)⋆ ⊣ s) ≡ ((x, y, z)× ⊣ t) ⊢ ((u, v, w)⋆ ⊣ s) ≡
18
(32)
(33)
S ´ANCHEZ-ORTEGA
((x, y, z)⊢ ⊣ t) ⊢ ((u, v, w)⋆ ⊣ s) ≡ ((x, y, z)⊢ ⊢ t) ⊢ ((u, v, w)⋆ ⊣ s) ≡ 0,
((x, y, z)⊢ ⊢ t) ⊣ ((u, v, w)⊣ ⊣ s) ≡ ((x, y, z)⊢ ⊢ t)⊣ ((u, v, w)× ⊣ s) ≡
((x, y, z)⊢ ⊢ t) ⊣ ((u, v, w)⊢ ⊣ s) ≡ ((x, y, z)⊢ ⊢ t) ⊣ ((u, v, w)⊢ ⊢ s) ≡ 0,
((x, y, z)⊢ ⊢ t) ⊢ ((u, v, w)⊢ ⊢ s) ≡ ((x, y, z)⊣ ⊣ t) ⊢ ((u, v, w)⊢ ⊢ s) ≡
((x, y, z)×⊣ t) ⊢ ((u, v, w)⊢ ⊢ s) ≡ ((x, y, z)⊢ ⊣ t) ⊢ ((u, v, w)⊢ ⊢ s) ≡ 0.
Let us check that ((x, y, z)⊢ ⊢ t) ⊢ ((u, v, w)⊢ ⊢ s) ≡ 0. Similarly, one can show
that all the elements above equal zero.
Notice that ((x, y, z)⊢ ⊢ t) ⊢ (u, v, w)⊢ ≡ (x, y, z)⊢ ⊢ (t ⊢ (u, v, w)⊢), since
((x, y, z)⊢, t, (u, v, w)⊢)⊢ ≡ 0 by the hypothesis and (AN3). Thus, it makes sense
to write (x, y, z)⊢ ⊢ t ⊢ (u, v, w)⊢. From (T4) we get
(x, y, z)⊢ ⊢ t ≡ (x ⊢ y, z, t)⊢ − (x, y ⊢ z, t)⊢ + (x, y, z ⊢ t)⊢ − x ⊢ (y, z, t)⊢,
which yields
(x, y, z)⊢ ⊢ t ⊢ (u, v, w)⊢ ≡ (x ⊢ y, z, t)⊢ ⊢ (u, v, w)⊢ − (x, y ⊢ z, t)⊢ ⊢ (u, v, w)⊢+
(x, y, z ⊢ t)⊢ ⊢ (u, v, w)⊢ − (x ⊢ (y, z, t)⊢) ⊢ (u, v, w)⊢
by (AN3) and (29). Then applying (AN3) we get
≡ −x ⊢ (y, z, t)⊢ ⊢ (u, v, w)⊢ ≡ 0,
((x, y, z)⊢ ⊢ t) ⊢ ((u, v, w)⊢ ⊢ s) = ((x, y, z)⊢ ⊢ t ⊢ (u, v, w)⊢) ⊢ s ≡ 0,
as desired. To finish, notice that (28)-(33) yield that I ⊣ I = I ⊢ I = 0, which
concludes the proof.
Acknowledgements
The author thanks Joe Repka for his carefully reading of this manuscript. She
also thanks Sara Madariaga for her help with the Maple calculations. The au-
thor was supported by the Spanish MEC and Fondos FEDER jointly through
project MTM2010-15223, and by the Junta de Andaluc´ıa (projects FQM-336 and
FQM2467).
References
[1] M. Aymon, P. P. Grivel: Un Theoreme de Poincar´e-Birkhoff-Witt pour les algebres de
Leibniz. Comm. Algebra 31 (2003), no. 3, 527 -- 544.
[2] A. Bloh: On a generalization of the concept of Lie algebra. Dokl. Akad. Nauk SSSR 165
(1965) 471 -- 473.
[3] A. Bloh: Cartan-Eilenberg homology theory for a generalized class of Lie algebras. Dokl.
Akad. Nauk SSSR 175 (1967) 266 -- 268.
[4] M. R. Bremner, J. S´anchez-Ortega: The partially alternating ternary sum in an associa-
tive dialgebra. J. Physics A 43 (2010) 455215.
[5] M. R. Bremner, R. Felipe, J. S´anchez-Ortega: Jordan triple disystems. Comput. Math.
Appl. 63 (2012) 1039 -- 1055.
[6] M. R. Bremner, L. A. Peresi, J. S´anchez-Ortega: Malcev dialgebras. Linear Multilinear
Algebra 60 (2012), no. 10, 1125 -- 1141.
[7] M. R. Bremner, R. Felipe, R. Felipe-Sosa, M. K. Kinyon, J. S´anchez-Ortega: The
Cayley-Dickson process for dialgebras. arXiv:1209.2645.
[8] R. H. Bruck, E. Kleinfeld: The structure of alternative division rings. Proc. Amer. Math.
Soc. 2 (1951) 878 -- 890.
[9] F. Chapoton: Un endofoncteur de la cat´egorie des op´erades. Dialgebras and Related Operads,
105 -- 110. Lecture Notes in Mathematics, 1763. Springer, Berlin, 2001.
ON THE DEFINITIONS OF NUCLEUS FOR DIALGEBRAS
19
[10] C. Cuvier: Alg`ebres de Leibnitz: d´efinitions, propri´et´es. Ann. Sci. ´Ecole Norm. Sup. (4) 27
(1994) 1 -- 45.
[11] V. T. Filippov: The measure of non-Lieness for Malcev algebras. Algebra Logic 31 (1992),
no. 2, 126 -- 140.
[12] M. A. Insua, M. Ladra: Gr´'obner bases in universal enveloping algebras of Leibniz algebras.
J. Symbolic Comput. 44 (2009) 517 -- 526.
[13] N. Jacobson: Lie algebras. Donver, New York, 1979.
[14] E. Kleinfeld: A class of rings which are very nearly associative. Amer. Math. Monthly 93
(1986), no. 9, 720 -- 722.
[15] E. Kleinfeld, L. Widmer: Rings satisfying (x, y, z) = (y, z, x). Comm. Algebra 17: 11
(1989) 2683 -- 2687.
[16] P. S. Kolesnikov: Varieties of dialgebras and conformal algebras. Sib. Math. J. 93 (1986),
no. 9, 720 -- 722. 49 (2008) 257 -- 272.
[17] P. S. Kolesnikov, V. Y. Voronin: On special identities for dialgebras. Linear Multilinear
Algebra 61 (2013), no. 3, 377 -- 391.
[18] D. Liu: Steinberg-Leibniz algebras and superalgebras. J. Algebra 283 (2005) 199 -- 221.
[19] J.-L. Loday: Une version non commutative des alg`ebres de Lie:
les alg`ebres de Leibniz.
Enseign. Math. 39 (1993) 269 -- 293.
[20] J.-L. Loday: Alg`ebres ayant deux op´erations associatives (dig`ebres). C. R. Math. Acad.
Sci. Paris 321 (1995) 141 -- 146.
[21] J.-L. Loday: Dialgebras. In: Dialgebras and Related Operads, 7 -- 66. Lectures Notes in Math-
ematics, 1763. Springer, 2001.
[22] J.-L. Loday, T. Pirashvili: Universal enveloping algebras of Leibniz algebras and
(co)homology. Math. Ann. 296 (1993) 139 -- 158.
[23] A. I. Malcev: Analytic loops. Matematicheskiı Sbornik N. S. 36/78 (1955) 569 -- 576.
[24] C. Mart´ın: Associative dialgebras from a structural viewpoint. Comm. Algebra (to appear)
arXiv:1012.4984.
[25] P. J. Morandi, J. M. P´erez-Izquierdo: On the tensor product of composition algebras. J.
Algebra 243 (2001) 41 -- 68.
[26] H. C. Myung: Malcev-admissible Algebras, Progress in Mathematics, Vol. 64, Birkhauser,
Basel, 1986.
[27] D. L. Outcalt: An extension of the class of alternative rings. Can. J. Math. 17 (1965)
130 -- 141.
[28] J. M. P´erez-Izquierdo, I. P. Shestakov: An envelope for Malcev algebras. J. Algebra 272
(2004) 379 -- 393.
[29] A. P. Pozhidaev: Algebraic Systems of Lie Type. Doctor of Science Thesis. Novosibirsk,
2010, 230 pages (in Russian).
[30] A. A. Sagle: Malcev algebras. Trans. Amer. Math. Soc. 101 (1961) 426 -- 458.
[31] J. Sterling: Rings satisfying (x, y, z) = (y, z, x). Can. J. Math. (1968) 913 -- 918.
[32] S. R. Sverchkov: Varieties of special algebras. Comm. Algebra 16 (1988) 1877 -- 1919.
[33] B. Vallette: Manin products, Koszul duality, Loday algebras and Deligne conjecture. J.
Reine Angew. Math. 620 (2008) 105 -- 164.
[34] A. Thedy: On rings satisfying [a, (b, c, d)] = 0. Proc. Amer. Math. Soc. 29 (1971) 250 -- 254.
Department of Algebra, Geometry and Topology, University of M´alaga, Spain
E-mail address: [email protected]
|
1106.4525 | 3 | 1106 | 2016-10-03T09:48:53 | Pure Dimension and Projectivity of Tropical Polytopes | [
"math.RA",
"math.AG"
] | We study how geometric properties of tropical convex sets and polytopes, which are of interest in many application areas, manifest themselves in their algebraic structure as modules over the tropical semiring. Our main results establish a close connection between pure dimension of tropical convex sets, and projectivity (in the sense of ring theory). These results lead to a geometric understanding of idempotency for tropical matrices. As well as their direct interest, our results suggest that there is substantial scope to apply ideas and techniques from abstract algebra (in particular, ring theory) in tropical geometry. | math.RA | math |
This is the author accepted manuscript of an article published in
Advances in Mathematics, Vol. 303 (2016), pp. 1236 -- 1263.
http://dx.doi.org/10.1016/j.aim.2016.08.033.
c(cid:13) 2016. This manuscript version is made available under the CC-BY-NC-
ND 4.0 licence http://creativecommons.org/licences/by-nc-nd/4.0/
PURE DIMENSION AND PROJECTIVITY OF
TROPICAL POLYTOPES
ZUR IZHAKIAN1, MARIANNE JOHNSON2 and MARK KAMBITES3
Abstract. We study how geometric properties of tropical convex sets
and polytopes, which are of interest in many application areas, mani-
fest themselves in their algebraic structure as modules over the tropical
semiring. Our main results establish a close connection between pure
dimension of tropical convex sets, and projectivity (in the sense of ring
theory). These results lead to a geometric understanding of idempo-
tency for tropical matrices. As well as their direct interest, our results
suggest that there is substantial scope to apply ideas and techniques
from abstract algebra (in particular, ring theory) in tropical geometry.
1. Introduction
Tropical mathematics can be loosely defined as the study of the real num-
bers (sometimes augmented with −∞) under the operations of addition and
maximum (or equivalently, minimum). It has been an active area of study in
its own right since the 1970's [13] and also has well-documented applications
in diverse areas such as analysis of discrete event systems, control theory,
combinatorial optimisation and scheduling problems [27], formal languages
and automata [29], phylogenetics [16], statistical inference [28], combinato-
rial/geometric group theory [5] and most recently in algebraic geometry (see
for example [21]). A key role in most of these areas is played by tropically
convex sets and tropical polytopes. These subsets of tropical space are natu-
rally endowed not only with a geometric structure (as Euclidean polyhedral
complexes), but also with a purely algebraic structure (as modules over the
tropical semiring).
1 Institute of Mathematics, University of Aberdeen, AB24 3UE, UK. Email
[email protected]. Research supported by Israel Science Foundation grant number 448/09
and by a Leibniz Fellowship at the Mathematisches Forschungsinstut Oberwolfach. Zur
Izhakian also gratefully acknowledges the support of EPSRC grant EP/H000801/1 and
the hospitality of the University of Manchester during a visit to Manchester.
2School of Mathematics, University of Manchester, Manchester M13 9PL, UK. Email
[email protected]. Research supported by EPSRC Grant
EP/H000801/1.
3School of Mathematics, University of Manchester, Manchester M13 9PL, England.
Email [email protected]. Research supported by an RCUK Academic
Fellowship and EPSRC Grant EP/H000801/1. Mark Kambites gratefully acknowledges
the hospitality of the Mathematisches Forschungsinstut Oberwolfach during a visit to
Oberwolfach.
1
2
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
In this paper we consider the way in which geometric properties of convex
sets and polytopes, of interest in many application areas, manifest them-
selves in their algebraic structure. Our main results establish a close con-
nection between pure dimension (a geometric property of interest for appli-
cations of tropical methods) and projectivity (in the sense of ring theory and
category theory). These results lead to a geometric understanding of idem-
potency for tropical matrices. As a corollary, we obtain the fact that all of
the widely studied notions of rank for tropical matrices coincide where the
matrices are idempotent or, more generally, von Neumann regular. (This
fact was mentioned without proof in [1, Fact 4, Section 35.7], with reference
given to a preprint of Cohen, Gaubert and Quadrat which at the time of
writing is still not available to the public.) As well as their direct inter-
est, these results suggest that there is substantial scope to apply ideas and
techniques from abstract algebra (in particular, ring theory) to understand
problems in tropical geometry.
We denote by FT the (finitary) tropical semiring, which consists of the
real numbers under the operations of addition and maximum. We write a⊕b
to denote the maximum of a and b, and a ⊗ b or just ab to denote the sum
of a and b. It is readily checked that both operations are associative and
commutative, ⊗ has a neutral element (0), admits inverses and distributes
over ⊕, while ⊕ is idempotent (a ⊕ a = a for all a). These properties mean
that FT has the structure of an idempotent semifield (without zero).
The space FTn of tropical n-vectors admits natural operations of com-
ponentwise maximum and the obvious scaling by FT, which makes it into
an FT-module. It also has a natural partial order. For detailed definitions
see Section 2 below. Submodules of FTn play a vital role in tropical mathe-
matics; as well as their obvious algebraic importance, they have a geometric
structure in view of which they are usually called (tropical) convex sets or
sometimes convex cones. Particularly important are the finitely generated
convex sets, which are called (tropical) polytopes.
There are several important notions of dimension for convex sets. The
(affine) tropical dimension is the topological dimension of the set, viewed as
a subset of Rn with the usual topology. The projective tropical dimension
(sometimes just called dimension in the algebraic geometry literature) is
one less than the affine tropical dimension. Note that, in contrast to the
classical (Euclidean) case, tropical convex sets may have regions of different
topological dimension. We say that a set X has pure (affine) dimension k if
every open (within X with the induced topology) subset of X has topological
dimension k. The generator dimension (sometimes also called the weak
dimension) of a convex set is the minimal cardinality of a generating subset,
under the linear operations of scaling and addition. The dual dimension is
the minimal cardinality of a generating set under scaling and the induced
operation of greatest lower bound within the convex set. We shall see later
(Section 3) that if a convex set X is the column space of a matrix, then its
dual dimension is the generator dimension of the row space, and also that
the dual dimension of X is the minimum k such that X embeds linearly in
FTk.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
3
Since tropical convex sets also have the aspect of FT-modules, it is natural
to ask about their algebraic structure. In particular, one might ask whether
important geometric and order-theoretic properties manifest themselves in
a natural way in their algebraic structure as modules, and vice versa.
If
they do, this raises the twin possibilities of addressing geometric problems
involving polytopes by the use of (tropically linear) algebraic methods and,
conversely, using geometric intuition to provide insight into problems in
tropical linear algebra.
One of the most important properties in the study of modules is projec-
tivity; recall that a module P is called projective if every morphism from
P to another module M factors through every surjective module morphism
onto M . The main results of this paper characterise projectivity for tropical
polytopes, in terms of the geometric and order-theoretic structure on these
sets. Our most striking result gives a direct connection between projective
modules and the notions of dimension and pure dimension discussed above:
Theorem 1.1. Let X ⊆ FTn be a tropical polytope. Then X is a projec-
tive FT-module if and only if it has pure dimension equal to its generator
dimension and its dual dimension.
Theorem 1.1 will be proved at the end of Section 6 below. Recall that
a square matrix A is called von Neumann regular if there is a matrix B
such that ABA = A. In ring theory there is a well-established three-way
correspondence between von Neumann regularity, idempotency and projec-
tivity:
for a finitely generated module, being projective is equivalent to
being isomorphic to the row space of an idempotent matrix, which in turn
is equivalent to being isomorphic to the row space of a von Neumann reg-
ular matrix. The corresponding result in the case of the tropical semiring
with zero can be found in the work of Gaubert; indeed one can obtain an
equivalent statement by combining Theorem 100 and Theorem 104 of [17],
the latter being based on joint work of Cohen, Gaubert and Quadrat [9,
Theorem 15] (which applies to a more general class of semifields). The same
correspondence even applies when working over FT, as a consequence of
which Theorem 1.1 immediately yields a new geometric characterisation of
von Neumann regularity:
Corollary 1.2. A square matrix over FT is von Neumann regular if and
only if its row space and column space have the same pure dimension equal
to their generator dimension.
Recall that a matrix A is called idempotent if A2 = A. Idempotent trop-
ical matrices are of particular significance for metric geometry, because of a
natural relationship between the tropical idempotency condition on a matrix
and the triangle inequality for an associated distance function (see for exam-
ple [15] for more details). A matrix is von Neumann regular if and only if it
shares its column space (or equivalently, its row space) with an idempotent
matrix (see, for example, [17, Theorem 104]), so Corollary 1.2 also exactly
characterises the row and column spaces of idempotent matrices. In fact,
many of our results below are proved by working with idempotents, and we
believe the technical understanding of tropical idempotency developed may
prove to be of independent interest.
4
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
Numerous definitions of rank have been introduced and studied for trop-
ical matrices, mostly corresponding to different notions of dimension of the
row or column space. For example, the tropical rank of a matrix is the
tropical dimension of its row space, which by [15, Theorem 23] for example,
coincides with that of its column space (variations of this result can also be
found in [22] and [3]). The row rank or row generator rank is the generator
dimension of its row space, which we shall see below (Section 3) coincides
with the dual dimension of its column space. The column rank or column
generator rank is defined dually. Other important notions of rank include
factor rank (also known as Barvinok rank or Schein rank ), Gondran-Minoux
row rank, Gondran-Minoux column rank, determinantal rank and Kapranov
rank ; since these do not play a key role in the present paper we refer the
interested reader to [1, 2, 14] for definitions. Of these different ranks, none
are ever lower than the tropical rank, and none are ever higher than the
greater of row rank and column rank; the non-obvious parts of this claim
are given in [2, Remark 7.8 and Theorem 8.4] and [14, Theorem 1.4]. Thus,
as a corollary we obtain a proof of the following result (which was mentioned
without proof in [1, Fact 4, Section 35.7]):
Corollary 1.3. Let M be a square von Neumann regular matrix (for ex-
ample, an idempotent matrix) over FT. Then the row generator rank, col-
umn generator rank, tropical rank, factor/Barvinok/Schein rank, Gondran-
Minoux row rank, Gondran-Minoux column rank, determinantal rank and
Kapranov rank of M are all equal.
We also obtain an order-theoretic description of projectivity. For convex
sets whose generator dimension and dual dimension coincide with the di-
mension of the ambient space (essentially, a non-singularity condition), this
has a particularly appealing form:
Theorem 1.4. A tropical polytope in FTn of generator dimension n and
dual dimension n is a projective FT-module if and only if it is min-plus (as
well as max-plus) convex.
Theorem 1.4 can also be deduced from results mentioned in the abstract
of the talk [12] but for which a proof has not yet been published. In greater
generality the formulation is slightly more technical, but still quite straight-
forward:
Theorem 1.5. A tropical polytope is projective if and only if it has generator
dimension equal to its dual dimension (equal to k, say), and is linearly
isomorphic to a submodule of FTk that is min-plus convex (as well as max-
plus convex).
Theorems 1.4 and 1.5 are established in Section 5 below. Polytopes that
are min-plus (as well as max-plus) convex have been studied in detail by
Joswig and Kulas [26], who term them polytropes.
In the terminology of
[26], a consequence of Theorem 1.4 is that a tropical n-polytope in FTn is
a polytrope if and only if it is a projective FT-module. Theorem 1.5 says
that a general tropical polytope is a projective FT-module if and only if it
is linearly isomorphic to a polytrope in some dimension.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
5
As well as connecting algebraic and geometric aspects of polytopes, our
approach also yields further insight into the abstract algebraic structure of
the semigroup of all n×n tropical matrices, and in particular the idempotent
elements. For example, we prove that any matrix of full column rank or row
rank is R-related (or L-related) to at most one idempotent (see Section 2
below for definitions and Theorem 5.7 for the formal statement and proof).
There are a number of different variants on the tropical semiring, which
arise naturally in different areas (for example algebraic geometry, traditional
max-plus algebra, and idempotent analysis). As well as the (theoretically
trivial, but potentially confusing) issue of whether to use maximum or min-
imum, one may choose to augment FT with an additive zero element (see
Section 2 below) or a "top" element (see for example [10]).
In bridging
different areas, and drawing on ideas and results from all of them, we face
the question of exactly which semiring to work in. For simplicity, in this
paper we have chosen to establish most of our main results only for FT
as defined above; this choice seems to makes the ideas behind the proofs
clearest and also minimises the extent to which we must modify and re-
prove existing geometric results to make them suitable for our needs.
In
places we are nevertheless forced to reprove some foundational results which
are known for T-modules in the setting of FT-modules; in other places it
is more convenient to use known results over T directly. We have tried to
give detailed references for any results which are known in similar settings,
but have included the proofs for the reader less familiar with the semiring
literature. It is likely that our main geometric results can be extended to
the augmented semirings themselves; the main modifications required would
be the replacement of subtraction in the proofs by an appropriate notion of
residuation (see [6]) and the extension of certain existing results which we
rely upon (for example, those of [15]) to the new setting.
In addition to this introduction, this article comprises six sections. Sec-
tion 2 briefly revises some necessary definitions, while Section 3 introduces
the key concept of the dual dimension of a polytope, and proves several
equivalent formulations. Section 4 proves some foundational results con-
necting projective modules, free modules and idempotent matrices over FT
(some of which are already known in the case of modules over a semiring
with 0 element, or specifically over T). Section 5 and Section 6 prove our
main results, giving order-theoretic and geometric characterisations respec-
tively of projective polytopes. Finally, Section 7 presents some examples of
how our concepts and results apply to tropical polytopes in low dimension;
while these are collected in one place for ease of discussion, the reader may
wish to refer to them at various times throughout the paper.
2. Preliminaries
Recall that we denote by FT the set R equipped with the operations of
maximum (denoted by ⊕) and addition (denoted by ⊗, or where more con-
venient by + or simply by juxtaposition). Thus, we write a ⊕ b = max(a, b)
and a ⊗ b = ab = a + b. Note that 0 acts as a multiplicative identity.
We denote by T the set FT∪{−∞} with the operations ⊕ and ⊗ extended
from the above so as to make −∞ an additive identity and a multiplicative
6
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
zero, that is
(−∞) ⊕ x = x ⊕ (−∞) = x and (−∞)x = x(−∞) = −∞
for all x ∈ T. We also extend the usual order on R to T in the obvious way,
namely by −∞ ≤ x for all x ∈ T.
By an FT-module 4 we mean a commutative semigroup M (with operation
⊕) equipped with a left action of FT such that λ(µm) = (λµ)m, (λ ⊕
µ)m = λm ⊕ µm, λ(m ⊕ n) = λm ⊕ λn and 0m = m for all λ, µ ∈ FT
and m, n ∈ M . A T-module is a commutative monoid M (with operation ⊕
and neutral element 0M ) satisfying the above conditions with the additional
requirement that λ0M = 0M = (−∞)m for all λ ∈ T and m ∈ M . Note
that the idempotency of addition in FT and T forces the addition in any
FT-module or T-module also to be idempotent. There is an obvious notion
of isomorphism between modules; we write X ∼= Y to indicate that two
modules are isomorphic.
Let R ∈ {FT, T}. We consider the space Rn of n-tuples of R; if x ∈ Rn
then we write xi for the ith component of x. Then Rn admits an addition
and a scaling action of R defined respectively by (x ⊕ y)i = xi ⊕ yi and
(λx)i = λ(xi) = λ + xi. It is readily verified that these operations make
Rn into an R-module. Rn also admits a partial order, given by x ≤ y if
xi ≤ yi for all i, and a corresponding componentwise minimum operation,
the minimum of two elements being greatest lower bound with respect to
the partial order.
In the case R = T the vector (−∞, . . . , −∞) ∈ Tn is an additive neutral
element for Tn. The vector (−∞, . . . , −∞, 0, −∞, . . . , −∞) with the 0 in
component i is called the ith standard basis vector for Tn, and denoted ei.
We write Mn(R) for the set of all n × n square matrices over R. Since ⊕
distributes over ⊗, these operations induce an associative multiplication for
matrices in the usual way, namely:
(AB)ij =
n
M
k=1
Aik ⊗ Bkj
giving Mn(R) the structure of a semigroup. (Of course one may equip it with
entrywise addition to form a (non-commutative) semiring, but we shall not
be concerned with this extra structure here.) The semigroup Mn(R) acts
on the left and right of the space Rn in the obvious way, by viewing vectors
as n × 1 or 1 × n matrices respectively.
A subset X ⊆ Rn is called (max-plus) convex if it is closed under ⊕ and
the action of R, that is, if it is an R-submodule of Rn. It is called min-plus
convex if it is closed under componentwise minimum and the action of R.
A non-empty and finitely generated (under the linear operations of scaling
and ⊕) submodule of FTn is called a (tropical) polytope. Tropical polytopes
in FTn projectivize to compact subsets of Rn−1 with the usual topology
[25, Proposition 2.6]. Some examples of tropical polytopes are collected in
Section 7 at the end of this paper.
4Some authors prefer the term semimodule, to emphasise the non-invertibility of addi-
tion, but since no other kind of module is possible over FT we have preferred the more
concise term.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
7
If M is a matrix over R then its column space CR(M ) and row space
RR(M ) are the polytopes generated by its columns and its rows respectively.
A non-zero element x in a convex set X ⊆ Tn or X ⊆ FTn is called
extremal if for every expression
x =
k
M
i=1
yi
with each yi ∈ X we have that yi = x for some i. Note that if x is extremal
then λx is also extremal for all λ ∈ FT. It is immediate from the definition
that if X ⊆ FTn then its extremal points do not depend upon whether it is
considered as a subset of FTn or Tn. It is known (see for example [8, 30])
that if X is finitely generated then it is generated by its extremal points,
and every generating set for X contains a scaling of every extremal point.
We shall also make use of some of Green's relations, which are a tool used
in semigroup theory to describe the principal ideal structure of a semigroup
or monoid. Let S be any semigroup. If S is a monoid, we set S1 = S, and
otherwise we denote by S1 the monoid obtained by adjoining a new identity
element 1 to S. We define binary relations L and R on S by aLb if and
only if S1a = S1b, and aRb if and only if aS1 = bS1. We define the binary
relation H on S by aHb if and only if aLb and aRb. Finally, the binary
relation D is defined by aDb if and only if there exists an element c ∈ S such
that aRc and cLa. Each of L, R, H and D is an equivalence relation on S
[20].
The following theorem, which first appeared in [17] over T (see also [19,
23] for proofs over both T and FT) summarises some results characterising
Green's relations in the semigroups Mn(FT) and Mn(T).
Theorem 2.1. Let A, B ∈ Mn(R) for R ∈ {FT, T}.
(i) ALB if and only if RR(A) = RR(B);
(ii) ARB if and only if CR(A) = CR(B);
(iii) ADB if and only if CR(A) and CR(B) are linearly isomorphic;
(iv) ADB if and only if RR(A) and RR(B) are linearly isomorphic.
We also need the following version of [17, Theorem 104, part 3] over FT,
which can be seen as a consequence of tropical duality (see for example
[10, 15, 19]). This result follows immediately from Theorem 2.1(iii) and (iv)
in the case of square matrices of the same size, but we shall also make use of
it in the non-square, non-uniform case. We give a brief proof for expository
purposes.
Theorem 2.2. Let M and N be matrices over FT (not necessarily square
or of the same size). Then CFT(M ) ∼= CFT(N ) if and only if RFT(M ) ∼=
RFT(N ).
Proof. Suppose f : CFT(M ) → CFT(N ) is an isomorphism of FT-modules.
By [19, Theorem 2.4] (see also [10, 15]) there are anti-isomorphisms (bi-
jections which invert scaling and reverse the partial order) g : RFT(M ) →
CFT(M ) and h : CFT(N ) → RFT(N ). Then the composite f ◦ g : RFT(M ) →
CFT(N ) is clearly also an anti-isomorphism, so by [19, Lemma 2.3], the map
h ◦ f ◦ g : RFT(M ) → RFT(N ) is an isomorphism of FT-modules.
The converse is dual.
(cid:3)
8
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
3. Dual Dimension
Let X ⊆ FTn be a convex set. We define the dual dimension of X to be
the minimum cardinality of a generating set for X under the operations of
scaling and greatest lower bound within X. Beware that the greatest lower
bound operation within X can differ from the componentwise minimum
operation, that is, the greatest lower bound operation in the ambient space
FTn. Indeed, they will coincide exactly if X is min-plus as well as max-plus
convex.
The concept of dual dimension is in some sense implicit in the theory
of duality for tropical modules (see for example [10]), but to the authors'
knowledge it was first explicitly mentioned in [24], and has yet to be ex-
tensively studied. Some examples of the dual dimension of polytopes are
presented in Section 7 below. We next prove some alternative character-
isations of dual dimension, which we hope will convince the reader of its
significance.
Proposition 3.1. Let M be a (not necessarily square) matrix over FT.
Then the dual dimension of CFT(M ) is the generator dimension of RFT(M )
(that is, the row generator rank of the matrix M ).
Proof. It is known [19, Theorem 2.4] (see also [10, 15]) that there is an
anti-isomorphism (a bijection which inverts scaling and reverses the order)
from RFT(M ) to CFT(M ). This map takes scalings to scalings, and maps
the ⊕ operation in RFT(M ) to greatest lower bound within CFT(M ). Thus,
the generator dimension of RFT(M ) (the minimum number of generators for
RFT(M ) under ⊕ and scaling) is equal to the dual dimension of CFT(M ) (the
minimum number of generators for CFT(M ) under greatest lower bound and
scaling).
(cid:3)
Theorem 3.2. Let X ⊆ FTn be a tropical polytope. Then the dual dimen-
sion of X is the smallest k such that X embeds linearly in FTk. In particular,
the dual dimension of X cannot exceed n.
Proof. Suppose X has generator dimension q and dual dimension k. Now X
is the column space of an n × q matrix M , and it follows by Proposition 3.1
that RFT(M ) has generator dimension k; in particular, k is finite. Thus,
RFT(M ) has k distinct (up to scaling) extremal points and these must all
occur as rows of M . Choose k rows to represent the extremal points, and
discard the others to obtain a k × q matrix N . Then RFT(N ) = RFT(M ), so
by Theorem 2.2, X = CFT(M ) is isomorphic to CFT(N ) ⊆ FTk.
Now suppose X embeds linearly into FTp; we need to show that k ≤ p.
The image of this embedding is a convex set of generator dimension q in
FTp, and so can be expressed as the column space of a p × q matrix N say.
By Proposition 3.1, the row rank of N is the dual dimension k of X. But the
size of N means that this cannot exceed p, so we have k ≤ p as required. (cid:3)
4. Projectivity, Free Modules, Idempotents and Regularity
In this section, we briefly discuss some properties of finitely generated
projective FT-modules, and their relationship to idempotency and von Neu-
mann regularity of matrices. The corresponding relationship over rings is
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
9
well known, and was extended to cover distributive idempotent semifields
(such as T) in [9, 11, 17]. Some of the results given there carry over to the
case of FT-modules by making suitable modifications; where this is the case
we give references to the original result.
To begin with we shall need a simple description of free FT-modules of
finite rank. For any positive integer k, define
Fk = Tk \ {(−∞, . . . , −∞)}.
Then Fk is closed under addition and scaling by reals, and hence has the
structure of an FT-module.
Proposition 4.1. Fk is a free FT-module on the subset {e1, . . . , ek} of stan-
dard basis vectors.
Proof. It follows from general results about semirings with zero (see for
example [18, Proposition 17.12]) that Tk is a free T-module of rank k, with
free basis {e1, . . . , ek}. We claim that Fk is a free FT-module with the same
basis. Suppose M is an FT-module and f : {e1, . . . , ek} → M is a function.
We may obtain from M a T-module M 0 by adjoining a new element 0M ,
and defining 0M ⊕ m = m ⊕ 0M = m, (−∞)m = 0M and λ0M = 0M for
all m ∈ M 0 and λ ∈ T. Now by freeness of Tk, there is a unique T-module
morphism g : Tk → M 0 extending f . It follows from the definition of M 0
that g maps elements of Fk to elements of M , so it restricts to an FT-module
morphism h : Fk → M extending f . Moreover, if h′ : Fk → M were another
such map, then it would extend to a distinct morphism from Tk to M 0
extending f , contradicting the uniqueness of g.
(cid:3)
An abstract characterisation of the finitely generated projective modules
over a distributive semifield (such as T), given in terms of direct factors5,
can be found in [11, Theorem 15]. Since our focus is on the connection
between projectivity and dimension, we consider the following formulation,
parts of which are well known for semirings with zero (see for example [18,
Proposition 17.16]) but which needs slightly more work for FT. Recall that
a retraction of an algebraic structure is an idempotent endomorphism; the
image of a retraction is called a retract.
Theorem 4.2. Let X be a polytope of generator dimension k. Then the
following are equivalent:
(i) X is projective;
(ii) X is isomorphic to a retract of the free FT-module Fk;
(iii) X is isomorphic to the column space of a k × k idempotent matrix
over FT;
(iv) X is isomorphic to the column space of an idempotent square matrix
over FT.
Proof. Suppose (i) holds. Since X is k-generated, Proposition 4.1 means
that there is a surjective morphism π : Fk → X. We also have the identity
5Specifically, it is shown that a homomorphism B between free modules U and X is (von
Neumann) regular if and only of there exists a homomorphism between free modules X and
Y whose kernel congruence induces a well-defined projection that maps each equivalence
class onto a unique representative lying in the image of B.
10
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
morphism ιM : X → X. By projectivity, there is a map ψ : X → Fk such
that π ◦ ψ = ιX . But now ψ ◦ π : Fk → Fk is a retraction with image
isomorphic to X, so (ii) holds.
Now suppose (ii) holds, and let π : Fk → Fk be a retraction with image
isomorphic to X. For each standard basis vector ei, define
xi = π(ei) ∈ Fk
and let E be the matrix whose ith column is xi. Now viewing E as a matrix
over T, we see that
Eei = xi
for each ei. So the action of E agrees with the action of π on the standard
basis vectors, and hence by linearity on the whole of Fk. Since E clearly
also fixes the zero vector in Tk, this means that E represents an idempotent
map on Tk, and hence is an idempotent matrix.
It remains to show that E ∈ Mk(FT), that is, that E has no −∞ entries.
Suppose for a contradiction that Eij = −∞. We claim that Eim = −∞ for
all m. Indeed, if we had Eim 6= −∞ then there would be no λ ∈ FT such
that λxm ≤ xi, which clearly cannot happen since xm and xi lie in CFT(E)
which is isomorphic to X, a subset of FTn. Since E is idempotent we have
xi =
k
M
p=1
Epixp.
But since Eii = −∞ this writes xi as a linear combination of the other
columns. This means that CFT(E) is generated by k − 1 vectors, which
contradicts the assumption that X, which is isomorphic to CFT(E), has
generator dimension k.
That (iii) implies (iv) is obvious.
Finally, suppose (iv) holds, and let E ∈ Mm(FT) be an idempotent matrix
with column space isomorphic to X. Then E viewed as a matrix over T
acts on Tm by left multiplication. Since E does not contain −∞ it clearly
cannot map a non-zero vector to zero, so it induces an idempotent map
π : Fm → Fm with image CFT(E). Now suppose A and B are FT-modules,
g : CFT(E) → B is a morphism and f : A → B is a surjective morphism.
By the surjectivity of f , for each standard basis vector ei of Fm we may
choose an element ai ∈ A such that f (ai) = g(π(ei)). Now since Fm is free
by Proposition 4.1, there is a (unique) morphism q : Fm → A taking each ei
to ai. Now for each i we have
f (q(ei)) = f (ai) = g(π(ei)),
so by linearity, f (q(x)) = g(π(x)) for all x ∈ Fm. But then by idempotency
of π,
f (q(π(x))) = g(π(π(x))) = g(π(x))
for all x ∈ Fm. Thus, if we let p be the restriction of q to π(Fm) then we
have f ◦ p = g, as required to show that π(Fm) = CFT(E) is projective and
hence X is projective.
(cid:3)
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
11
Recall that an element x of a semigroup (or semiring) is called von Neu-
mann regular 6 if there exists an element y satisfying xyx = x; thus a matrix
is von Neumann regular as defined above exactly if it is von Neumann reg-
ular in the containing full matrix semigroup.
It is a standard fact from
semigroup theory (see for example [20]) that an element is von Neumann
regular exactly if it is D-related (or equivalently, L-related or R-related) to
an idempotent.
The following result extends to FT a fact which is known for a class of
semirings including T ([9, Theorem 15], [17, Theorem 104] and [11, Propo-
sition 5]).
Proposition 4.3. Let A ∈ Mn(FT). Then the following are equivalent:
(i) A is a von Neumann regular element of Mn(T);
(ii) A is a von Neumann regular of Mn(FT);
(iii) CFT(A) is a projective FT-module;
(iv) RFT(A) is a projective FT-module.
Proof. We prove the equivalence of (i), (ii) and (iii), the equivalence of (i),
(ii) and (iv) being dual.
If (i) holds, then A = ABA for some B ∈ Mn(T). An easy calculation
shows that replacing any −∞ entries in B with sufficiently small finite values
yields a matrix B ′ ∈ Mn(FT) satisfying A = AB ′A, so (ii) holds.
If (ii) holds then A is von Neumann regular, so it is R-related to an
idempotent matrix in E ∈ Mn(FT). Now by Theorem 2.1 we have CFT(A) =
CFT(E). But CFT(E) is projective by Theorem 4.2, so (iii) holds.
Finally, suppose (iii) holds, so CFT(A) is projective, and let k be the
generator dimension of CFT(A). Note that k ≤ n, since CFT(A) is generated
by the n columns of A. By Theorem 4.2 there is an idempotent matrix
E ∈ Mk(FT) such that CFT(E) is isomorphic to CFT(A). By adding n − k
rows and columns of −∞ entries, we obtain from E an idempotent matrix
F ∈ Mn(T) satisfying CT(F ) ∼= CT(E) = CT(A). But now Theorem 2.1
gives F DA, which suffices to establish (i).
(cid:3)
Theorem 4.4. Every projective tropical polytope has generator dimension
equal to its dual dimension.
Proof. We show that the dual dimension cannot exceed the generator dimen-
sion, the reverse inequality being dual by Propositions 3.1 and 4.3. Suppose
then for a contradiction that X is projective with dual dimension k strictly
greater than its generator dimension m. Then by Theorem 3.2, k is mini-
mal such that X embeds in FTk. But by Theorem 4.2, X is isomorphic to
the column space of an m × m idempotent matrix E, say, which means X
embeds in FTm. Since m < k this is a contradiction.
(cid:3)
Note that Theorem 4.2 says that every projective polytope is abstractly
isomorphic to the column space of an idempotent matrix (of size its gen-
erator dimension). Since a polytope is itself a submodule of some FTn, we
might ask whether every projective polytope is itself the column space of
6In the literature of semigroup theory such elements are usually just called "regular";
we use the longer term "von Neumann regular" for disambiguation from other concepts
of regularity for tropical matrices (see for example [7]).
12
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
an idempotent (of size the dimension of the containing space). We conclude
this section by noting that [17, Theorem 104] and [11, Proposition 5] extend
to FT, or in other words, projective polytopes are exactly the column spaces
of idempotents:
Theorem 4.5. Let X ⊆ FTn be a tropical polytope. Then X is projective if
and only if X is the column space of an idempotent matrix in Mn(FT).
Proof. If X is the column space of an idempotent over FT then Theorem 4.2
tells us that X is projective.
Conversely, suppose X is projective. By Theorem 4.4, the generator di-
mension of X is equal to dual dimension of X, which by Theorem 3.2 cannot
exceed n. Thus, we may write X as the column space of an n × n matrix
A. By Proposition 4.3 this matrix is von Neumann regular as an element
of Mn(FT), so it is R-related to an idempotent in Mn(FT), which by Theo-
rem 2.1 also has column space X.
(cid:3)
5. Order-Theoretic Properties of Projective Polytopes
In this section we establish our order-theoretic characterisation of pro-
jective tropical polytopes. For this, we first need some elementary order-
theoretic properties of idempotent matrices over the tropical semiring. These
will be familiar to experts but to aid the non-specialist reader we include
some short direct proofs.
Proposition 5.1. For any matrix A ∈ Mn(T) and vectors x, y ∈ Tn, if
x ≥ y then Ax ≥ Ay and xA ≥ yA.
Proof. If x ≥ y then x ⊕ y = x, so by linearity Ax ⊕ Ay = A(x ⊕ y) = Ax
which means that Ax ≥ Ay. The other claim is dual.
(cid:3)
The following result (and the corollaries that follow) can be seen as a
special case of the well-developed spectral theory for tropical matrices (see
for example [4, Theorem 3.101]).
Lemma 5.2. Let E ∈ Mn(T) be an idempotent matrix, and let x be an
extremal point of the column space CT(E). Then there exists a λ ∈ FT such
that λx occurs as a column of E with 0 in the diagonal position.
Proof. Clearly every extremal point of CT(E) occurs (up to scaling) as a
column of E, since they are by definition needed to generate the column
space. Let c1, . . . , cn be the columns of E, and suppose ci is an extremal
point.
Considering the equation E2 = E, we have
ci =
n
M
j=1
Ejicj =
n
(ci)jcj.
M
j=1
Since ci is extremal, it must in fact be equal to one of the terms in this sum,
say
ci = Ejicj = (ci)jcj,
giving that cj is a multiple of ci. Moreover, since ci is extremal, and hence
not the zero vector, it follows that (ci)j 6= −∞. But now
(ci)j = ((ci)jcj)j = (ci)j(cj)j
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
13
which since (ci)j 6= −∞ means that (cj)j = Ejj = 0. Since cj is a multiple
of ci, this completes the proof.
(cid:3)
Corollary 5.3. If E ∈ Mn(T) is an idempotent matrix of column generator
rank n (or row generator rank n) then every diagonal entry of E is 0.
Corollary 5.4. If E ∈ Mn(T) is an idempotent matrix of column generator
rank n (or row generator rank n) then Ex ≥ x and xE ≥ x for all x ∈ Tn.
Proposition 5.5. If E ∈ Mn(T) [respectively, E ∈ Mn(FT)] is an idem-
potent matrix of column generator rank n (or row generator rank n) then
CT(E) and RT(E) [respectively, CFT(E) and RFT(E)] are min-plus convex.
Proof. We prove the claim for CT(E), that for the row space being dual
and the FT cases very similar. Suppose a, b ∈ CT(E), and let c be the
componentwise minimum of a and b. It will suffice to show that c ∈ CT(E).
Since a, b ∈ CT(E) and E is idempotent, we have a = Ea and b = Eb. Since
c ≤ a and c ≤ b, by Proposition 5.1 we have Ec ≤ Ea = a and Ec ≤ Eb = b.
This means that Ec ≤ min{a, b} = c. But by Corollary 5.4 we have Ec ≥ c,
so it must be that Ec = c and c ∈ CT(E), as required.
(cid:3)
Proposition 5.6. Let E ∈ Mn(T) be an idempotent of column generator
rank n. Then for any vector x ∈ Tn, the vector Ex is the minimum (with
respect to the partial order ≤) of all elements y ∈ CT(E) such that y ≥ x.
Proof. By definition we have Ex ∈ CT(E) and by Corollary 5.4 we have
Ex ≥ x, so Ex is itself an element of CT(E) which lies above x. Thus, it
will suffice to show that every other such element lies above Ex. Suppose,
then that z ∈ CT(E) and z ≥ x. Since z ∈ CT(E) and E is idempotent
we have Ez = z. But since z ≥ x, Proposition 5.1 gives z = Ez ≥ Ex, as
required.
(cid:3)
We note that Proposition 5.5 can also be deduced as a consequence of
Proposition 5.6. Proposition 5.6 has the following interesting semigroup-
theoretic corollary:
Theorem 5.7. Any R-class [L-class] in Mn(T) consisting of matrices of
column generator rank n [or row generator rank n] contains at most one
idempotent.
Proof. We prove the claim for R-classes, that for L-classes being dual. Let E
be an idempotent such that CT(E) has generator rank n. For i ∈ {1, . . . , n},
applying Proposition 5.6 with x = ei the ith standard basis vector shows
that the ith column Eei of E is the minimum element of CT(E) greater than
or equal to ei. Thus, E is completely determined by its column space and
the fact that it is idempotent. Now if F were another idempotent R-related
to E then by Theorem 2.1(ii) we would have CT(E) = CT(F ), which by the
preceding argument would mean that E = F .
(cid:3)
Note the row or column generator rank hypothesis in Theorem 5.7 cannot
be removed. Indeed, in [23] it was shown that every H-class corresponding
to a 1-generated column space in M2(T) contains an idempotent, so the
corresponding R− and L-classes each contain a continuum of idempotents.
We are now ready to prove our first main result, namely Theorem 1.4
from the introduction.
14
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
Theorem 1.4. A tropical polytope in FTn of generator dimension n and
dual dimension n is a projective FT-module if and only if it is min-plus (as
well as max-plus) convex.
Proof. The direct implication is immediate from Theorems 4.5 and 5.5, so
we need only prove the converse.
Suppose, then, that X ⊆ FTn is min-plus and max-plus convex, and let
M be the matrix whose ith column is the infimum (in FTn) of all elements
y ∈ X such that y ≥ ei, where ei is the ith standard basis vector (that
is, such that y has non-negative ith coordinate). Such an infimum exists.
Indeed, if n = 1 take y = 0. Otherwise, for each coordinate j 6= i, consider
the set
{uj u ∈ X with ui ≥ 0}.
It is easy to see that this set is non-empty and, since X is finitely generated,
it has a lower bound and hence an infimum. It follows from the fact X is
closed that this infimum will be attained; choose a vector wj ∈ X such that
wjj attains it at wji ≥ 0. In fact, by the minimality of wjj and the fact that
X is closed under scaling, we will have wji = 0. Now let v be the minimum
of all the wj's. Then vi = 0 and v is clearly less than or equal to all vectors
u ∈ X with ui ≥ 0. Moreover, by min-plus convexity, it lies in X, which
means it must be the desired minimum.
Notice that since X is closed under scaling, it will have elements in which
the ith coordinate is 0.
It follows that the ith column of M will in fact
have ith coordinate 0, that is, that every diagonal entry of M is 0. We have
shown that every column of M lies in X, so CFT(M ) ⊆ X. We aim to show
that M is idempotent with column space X.
First, we claim that each column of M is an extremal point of X. Indeed,
suppose for a contradiction that the pth column, call it y, is not an extremal
point of X. Then by definition we may write y as a finite sum of elements
in X which are not multiples of y, say y = z1 ⊕ · · · ⊕ zk. Let j be such
that zj agrees with y in the pth coordinate. Then zj < y (since zj forms
part of a linear combination for y, and was chosen not to be a multiple of y)
but zj ≥ ep and zj ∈ X, which contradicts the choice of y as the minimum
element in X above ep.
Next, we claim that no two columns of M are scalings of one another.
Indeed, suppose the ith column vi and jth column vj are scalings of one
another. For any x ∈ X we have (−xi)x ≥ ei and (−xi)x ∈ X, so by the
definition of vi we have vi ≤ (−xi)x. Thus, using the fact that vii = 0,
vij − vii ≤ (−xi)xj = xj − xi.
By applying the same argument with i and j exchanged we also obtain
vji − vjj ≤ (−xj)xi = xi − xj.
But since vi is a multiple of vj, we have vji − vjj = vii − vij so negating we
get
vij − vii ≥ xj − xi.
Thus we have shown that xj − xi = vij − vii for every x ∈ X.
In other
words, the jth entry of every vector in X is determined by the ith entry.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
15
This implies that X embeds linearly into FTn−1, which by Theorem 3.2
contradicts the fact that X has dual dimension n.
We have shown that the n columns of M are extremal points of X, and
that no two are scalings of each other. Since X has generator dimension n
it has precisely n extremal points up to scaling, so we conclude that every
extremal point of X must occur (up to scaling) as a column of M . Thus,
X ⊆ CFT(M ), and so CFT(M ) = X.
Finally, we need to show that M is idempotent. We have already observed
that every diagonal entry of M is 0. It follows from the definition of matrix
multiplication that for all i and j,
(M 2)ij ≥ MijMjj = Mij0 = Mij.
Now let i, j, k ∈ {1, . . . , n}. To complete the proof, it will suffice to show
that Mij ≥ MikMkj. Let vj and vk be the jth and kth columns of M ,
and consider the vector w = (−Mkj)vj = (−vjk)vj. Then w lies in X and
has kth component 0, so by the definition of M , w is greater than vk. In
particular, comparing the ith entries of these vectors, we have
(−Mkj)Mij = wi ≥ vki = Mik
Mij ≥ MikMkj
and so
as required.
(cid:3)
Combining Theorem 1.4 with Proposition 4.3 yields an order-theoretic
characterisation of von Neumann regularity for matrices of full column and
row generator rank over FT.
Theorem 5.8. Let M ∈ Mn(FT) be a matrix of column generator rank n
and row generator rank n. Then the following are equivalent:
(i) M is von Neumann regular;
(ii) CFT(M ) is min-plus convex;
(iii) RFT(M ) is min-plus convex.
Proof. By Proposition 4.3, M is von Neumann regular if and only if CFT(M )
is projective, which by Theorem 1.4 is true exactly if CFT(M ) is min-plus
convex. A similar argument applies to RFT(M ).
(cid:3)
Next we prove Theorem 1.5 from the introduction.
Theorem 1.5. A tropical polytope is projective if and only if it has generator
dimension equal to its dual dimension (equal to k, say), and is linearly
isomorphic to a submodule of FTk that is min-plus convex (as well as max-
plus convex).
Proof. Suppose X ⊆ FTn is finitely generated and projective. By Theo-
rem 4.4 it has generator dimension equal to its dual dimension; let k be this
value. By Theorem 4.2, X is isomorphic to the column space of an idempo-
tent matrix in Mk(FT). This column space is projective and has generator
dimension k and dual dimension k, and so by Theorem 1.4 is min-plus convex
as well as max-plus convex.
16
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
Conversely, if X has dual dimension and generator dimension k and is
isomorphic to a convex set in FTk which is min-plus as well as max-plus
convex, then X is projective by Theorem 1.4.
(cid:3)
6. Geometric characterisation of projective polytopes
Our aim in this section is to prove Theorem 1.1, which gives a geometric
characterisation of projective tropical polytopes in terms of pure dimension,
generator dimension and dual dimension.
We require some preliminary terminology, notation and results from [15].
Let v1, . . . , vk ∈ FTn and let X ⊆ FTn be the polytope they generate. Let
x ∈ FTn. The type of x (with respect to the vectors v1, . . . , vk) is an n-
tuple of sets, the pth component of which consists of the indices of those
generators which can contribute in the pth position to a linear combination
for x. Formally,
type(x)p = {i ∈ {1, . . . , k} ∃λ ∈ FT such that λvi ≤ x and λvip = xp}
= {i ∈ {1, . . . , k} xp − vip ≤ xq − viq for all q ∈ {1, . . . , n}}.
It is easily seen that X itself consists of those vectors whose types have every
component non-empty.
For a given type S we write Sp for the pth component of S. We denote by
GS the undirected graph having vertices {1, . . . , n}, and an edge between p
and q if and only if Sp ∩ Sq 6= ∅. We define union and inclusion for types
if S and T are types then S ∪ T is the type given by
in the obvious way:
(S ∪ T )i = Si ∪ Ti, and S ⊆ T if Si ⊆ Ti for all i, that is, if S ∪ T = T . We
write XS for the set of all points having type containing S; it is readily
verified that XS is a closed set of pure dimension. A face of X is a set XS
such that S is the type of a point of X.
We require the following result of Develin and Sturmfels [15], which we
rephrase slightly for compatibility with the terminology and conventions of
the present paper ([15] instead using the min-plus convention and using the
term "dimension" to mean projective dimension).
Lemma 6.1 ([15, Proposition 17]). With notation as above, the tropical
dimension of XS is the number of connected components in GS.
It is easily seen that a polytope has pure dimension k if and only if every
point lies inside a closed face of dimension k.
Let E ∈ Mn(FT) be an idempotent matrix with columns v1, . . . , vn (so
that vij = Eji for all i, j). We shall show that the column space CFT(E) has
pure dimension. To do this we need some lemmas.
Lemma 6.2. Let E, v1, . . . , vn be as above, and let i, j ∈ {1, . . . , n} be such
that
• vi and vj are extremal points in CFT(E);
• vj is not a multiple of vi; and
• vii = vjj = 0.
Then for every k ∈ {1, . . . n} we have
and in the case k = j this inequality is strict.
vji − vii = vji ≤ vjk − vik,
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
17
Proof. For any k, computing the (k, j) entry of E2, we see that
vjk = Ekj = (E2)kj ≥ EkiEij = vik + vji
which, since vii = 0, yields
vji − vii = vji ≤ vjk − vik
as required.
Now let k = j, and suppose for a contradiction that the inequality is not
strict, that is, that
vji = vjk − vik = vjj − vij.
Since vjj = 0, the above equation yields vji = −vij. Now for any index
p ∈ {1, . . . , n} by the above we have
vjp − vip ≥ vji.
By symmetry of assumption, we may also apply a corresponding inequality
with i and j exchanged, which yields
vip − vjp ≥ vij = −vji
and hence by negating both sides
vjp − vip ≤ vji.
So vji = vjp − vip for all p, which means that vj = vjivi. But this contradicts
the hypothesis that vj is not a scalar multiple of vi.
(cid:3)
Lemma 6.3. Let E, v1, . . . , vn be as above, and let J ⊆ {1, . . . , n} be such
that the corresponding columns form a set of unique representatives for the
extremal points of CFT(E), and vjj = 0 for every j ∈ J. Let x ∈ CFT(E),
and j ∈ J. Then j ∈ type(x)j.
Proof. Suppose for a contradiction that j /∈ type(x)j. Write
x = M
i∈J
λivi
with the λi maximal. The fact that j /∈ type(x)j means precisely that
λjvjj < xj.
By definition of the sum, there is a k ∈ J such that
and by the above k 6= j. Rearranging, we obtain
λkvkj = xj > λjvjj
vkj − vjj > λj − λk.
On the other hand, by maximality of the λi's, there is a p such that λjvjp =
xp. Then certainly we have
which combining with the above yields
λkvkp ≤ xp = λjvjp
vkp − vjp ≤ λj − λk < vkj − vjj,
contradicting Lemma 6.2 applied to columns vk and vj.
(cid:3)
18
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
Lemma 6.4. Let E, v1, . . . , vn be as above and let J ⊆ {1, . . . , n} be such
that the corresponding columns form a set of unique representatives for the
extremal points of CFT(E), and vjj = 0 for all j ∈ J. Let x ∈ CFT(E). Then
there is an element y ∈ CFT(E) such that type(y) ⊆ type(x) and type(y) is
a vector of singletons.
Proof. For every vector y ∈ FTn, define
Ty = {(i, j, p) ∈ J × J × {1, . . . , n} i 6= j and i, j ∈ type(y)p}.
If y ∈ CFT(E) then, as discussed above, the components of type(y) are non-
empty, so type(y) is a vector of singletons exactly if Ty is empty. Thus,
the claim to be proven is that there is a vector y ∈ CFT(E) with type(y) ⊆
type(x) and Ty empty. Suppose false for a contradiction, and choose z ∈
CFT(E) such that type(z) ⊆ type(x), and the cardinality of Tz is minimal
amongst vectors having this property.
Write
z = M
i∈J
λivi
with the λi's maximal. By the supposition, Tz is non-empty, so we may
choose some (i, j, p) ∈ Tz. Then by the definition of types we have
Notice that we cannot have both
λivip = zp = λjvjp.
λivij = zj and λjvji = zi.
Indeed, by Lemma 6.3 we have λivii = zi and λjvjj = zj, so we would have
vjj − vij = λi − λj = vji − vii
which contradicts the strict inequality guaranteed by Lemma 6.2. Thus, by
exchanging i and j if necessary, we may assume without loss of generality
that λivij < zj.
Now choose an ε > 0 such that
ε < zq − λkvkq
for all k ∈ J and q ∈ {1, . . . , n} such that λkvkq 6= zq. (Notice that by
the definition of the λk we can never have zq < λkvkq, so the condition
λkvkq 6= zq is sufficient to make zq − λkvkq positive.)
Define
y = z ⊕ (λi + ε)vi.
Since z ∈ CFT(E) and vi is a column of E, we have y ∈ CFT(E). Write
y = M
k∈J
µkvk
with the µk's maximal. Notice that since λivi ≤ z, we have
y = z ⊕ (λi + ε)vi = z ⊕ ε(λivi) ≤ εz.
In other words, no coordinate of y can exceed the corresponding coordinate
in z by more than ε. It follows immediately that
for all k ∈ J. It is also clear that µi = λi + ε.
µk ≤ λk + ε
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
19
We claim that type(y) ⊆ type(z).
Indeed, suppose k /∈ type(z)p, that
is, λkvkp < zp. Then by the choice of ε, we have ε < zp − λkvkp, that is,
ελkvkp < zp. So using the previous paragraph we have
µkvkp ≤ ελkvkp < zp ≤ yp,
which means that k /∈ type(y)p.
It follows immediately that Ty ⊆ Tz; we claim that the containment is
strict. Indeed, we know that (i, j, p) ∈ Tz; suppose for a contradiction that
it lies also in Ty, that is, that i, j ∈ type(y)p. Then by definition we have
µjvjp = yp = µivip = ελivip = ελjvjp,
from which we deduce that µj = λj + ε. Thus, using Lemma 6.3, we have
yj = µjvjj = ελjvjj = εzj > zj.
Since y = z ⊕ ελivi, the only way this can happen is if
yj = (ελivi)j = ελivij.
But then ελivij = yj = ελjvjj, so
λivij = λjvjj = zj.
This contradicts our assumption that λivij < zj, and so proves the claim
that (i, j, p) /∈ Ty. Thus, Ty is a strict subset of Tz, which contradicts the
minimality assumption on Tz, and completes the proof of the lemma.
(cid:3)
Theorem 6.5. Let E ∈ Mn(FT) be an idempotent matrix of column gener-
ator rank r. Then CFT(E) has pure dimension r.
Proof. Let x ∈ CFT(E). It will suffice to show that x lies in a face of tropical
dimension r.
Let J ⊆ {1, . . . , n} be such that the corresponding columns form a set
of unique representatives for the extremal points of CFT(E). Thus, J has
cardinality r. By Lemma 5.2, we may choose J so that vjj = 0 for all j ∈ J.
Now consider types with respect to the generating set of CFT(E) formed by
the columns corresponding to indices in J.
By Lemma 6.4 there is a point y ∈ CFT(E) such that type(y) ⊆ type(x)
and type(y) is a vector of singletons. By Lemma 6.3, we have j ∈ type(y)j
for every j ∈ J. It follows that the graph Gtype(y) has exactly r connected
components (one corresponding to each generator vi).
Hence, by Lemma 6.1, the face Xtype(y) has tropical dimension r. More-
over, since type(y) ⊆ type(x), it follows from the definition of Xtype(y) that
x lies in a face of tropical dimension r, as required.
(cid:3)
Lemma 6.6. Let X be a tropical polytope in FTn of generator dimension n
or less. Then X contains at most one face of dimension n.
Proof. Choose generators v1, . . . , vn for X, and suppose for a contradiction
that XS and XT are distinct faces of dimension n. By Lemma 6.1, both S
and T are n-tuples of singleton sets containing all the numbers from 1 to n.
By reordering our chosen generating set if necessary we may thus assume
that
S = ({1}, {2}, . . . , {n})
20
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
while
T = ({σ(1)}, {σ(2)}, . . . , {σ(n)})
for some permutation σ of {1, . . . , n}. Since S and T are distinct, σ must
be non-trivial.
Now choose points a, b ∈ X with type(a) = S and type(b) = T . Write
a =
n
M
i=1
λivi and b =
n
M
i=1
µivi
with the λi's and µi's all maximal. By the definition of types, for all i we
have
ai = λivii ≥ λσ(i)vσ(i)i and µivii ≤ µσ(i)vσ(i)i = bi
and these inequalities are strict provided i 6= σ(i). Rearranging these, we
get
λi − λσ(i) ≥ vσ(i)i − vii ≥ µi − µσ(i)
and again, the inequalities are strict provided i 6= σ(i).
Now since σ is a non-trivial permutation of a finite set, it contains a non-
trivial cycle. In other words, there is a p ∈ {1, . . . , n} and an integer k ≥ 2
such that p 6= σ(p), but p = σk(p). Note that, σi(p) 6= σi+1(p) for any i, so
using our strict inequalities above we have
k
k
0 =
(λσi(p) − λσi+1(p)) >
X
i=1
(µσi(p) − µσi+1(p)) = 0
X
i=1
giving the required contradiction.
(cid:3)
Theorem 6.7. Suppose X ⊆ FTn is a tropical polytope of generator dimen-
sion n and pure dimension n. Then X is projective.
Proof. Let u1, . . . , un be a minimal generating set for X (so that the elements
ui are unique representatives of the extremal points of X). Consider the
types of points in X with respect to this generating set. Since X has pure
dimension, for each i, ui is contained in a closed face of dimension n. It
follows by Lemma 6.6 that all of the ui's are contained in the same face
of dimension n, say XS for some type S. Since XS is a face of X, the
components of S are non-empty, so it follows by Lemma 6.1 that S consists
of singletons and contains every i. By reordering the ui's, we may assume
without loss of generality that Sk = {k} for all k ∈ {1, . . . , n}.
Moreover, by scaling the ui's if necessary, we may assume that uii = 0
for each i. Let E ∈ Mn(FT) be the matrix whose ith column is ui. It is
immediate that CFT(E) = X, and from our rescaling of the ui's that the
diagonal entries of E are 0. We claim that E is idempotent, that is,
(E2)ij = Eij
for all i, j ∈ {1, . . . , n}.
Fix i, j ∈ {1, . . . , n}. From the definition of matrix multiplication we have
(E2)ij =
n
M
k=1
EikEkj.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
21
Since the diagonal entries of E are 0, we have
(E2)ij ≥ EiiEij = 0Eij = Eij.
On the other hand, let k ∈ {1, . . . , n}. Recall that Sk = {k}. Since
uj appears in the face XS, by definition we have S ⊆ type(uj), and so
k ∈ type(uj)k. It follows from the definition of types that
ujk − ukk ≤ uji − uki.
But ukk = 0 so rearranging yields uki + ujk ≤ uji for all k. Thus we have
(E2)ij =
n
M
k=1
EikEkj =
n
M
k=1
uki + ujk ≤
n
M
k=1
uji = uji = Eij.
as required to complete the proof of the claim that E is idempotent.
Thus, X is the column space of an idempotent matrix, so by Theorem 4.5
(cid:3)
we deduce that X is projective.
We are finally ready to prove Theorem 1.1 from the introduction.
Theorem 1.1. Let X ⊆ FTn be a tropical polytope. Then X is a projec-
tive FT-module if and only if it has pure dimension equal to its generator
dimension and its dual dimension.
Proof. Suppose X ⊆ FTn is a projective polytope. Then by Theorem 4.4,
there is a k ≤ n such that X has generator dimension k and dual dimension
k. Now by Theorem 4.2, X is isomorphic to the column space CFT(E) of
an idempotent matrix E ∈ Mk(FT). It follows by Theorem 6.5 that CFT(E)
has pure dimension k. Moreover, it is easy to see that a linear isomorphism
of convex sets is a homeomorphism with respect to the standard product
topology inherited from the real numbers.
Indeed, an isomorphism is a
bijection, and both it and its inverse are continuous because addition and
multiplication in FT are continuous. Since pure dimension is an abstract
topological property it follows that X has pure dimension k.
Conversely, suppose X has pure dimension, generator dimension and dual
dimension all equal to k. Then by Theorem 3.2, X is isomorphic to a convex
set Y ⊆ FTk. Now Y also has generator dimension k and, by the same
argument as above, pure dimension k so by Theorem 6.7, Y is projective,
and so X is projective.
(cid:3)
7. Examples
In this section we collect together some examples of tropical polytopes in
low dimension, and show how the concepts and results of this paper apply
to them.
We consider first the (somewhat degenerate) 2-dimensional case.
It is
well known and easily seen that every polytope in FT2 is either (a) a line
of gradient 1, or (b) the closed region between two such lines. Figure 1
illustrates these possibilities.
It is readily verified that polytopes of type
(a) have pure dimension, generator dimension and dual dimension all equal
to 1, while those of type (b) have pure dimension, generator dimension
and dual dimension all equal to 2. We deduce by Theorem 1.1 that every
tropical polytope in FT2 is projective. By Corollary 1.2, we recover the fact
22
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
(a)
(b)
Figure 1. Polytopes in FT2.
(proved by explicit computation in [23]) that every 2 × 2 tropical matrix is
von Neumann regular, that is, that the semigroup of all such matrices is a
regular semigroup. It also follows by Corollary 1.3 that the various notions
of rank discussed in the introduction all coincide for 2 × 2 matrices.
Recall that from affine tropical n-space we obtain projective tropical (n −
1)-space, denoted PFT(n−1), by identifying two vectors if one is a tropical
multiple of the other by an element of FT. Thus we may identify PFTn−1
with Rn−1 via the map
(x1, . . . , xn) 7→ (x1 − xn, x2 − xn, . . . , xn−1 − xn).
(7.1)
Each convex set X ⊆ FTn induces a subset of the corresponding projective
space, termed the projectivisation of X.
(c)
(d)
(3,3)
(e)
(0,0)
(3,0)
Figure 2. Some projective tropical polytopes in PFT2.
All three polytopes shown in Figure 2 have pure dimension. Polytopes (c)
and (e) have tropical dimension 3, generator dimension 3 and dual dimen-
sion 3, while polytope (d) has tropical dimension 2, generator dimension 2
and dual dimension 2, and so by Theorem 1.1 they are all projective. By
Theorem 1.4 polytopes (c) and (e) must be min-plus (as well as max-plus)
convex, and indeed this can be verified by inspection. Polytope (d) is not
b
b
b
b
b
b
b
b
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
23
min-plus convex, but by Theorem 1.5 it must be isomorphic to a polytope
in FT2 which is min-plus convex; in fact it will be isomorphic to something
of the form shown in Figure 1(b).
By Corollary 1.2 we deduce that every matrix whose row space is one
of these polytopes must be von Neumann regular, and so there is at least
one idempotent matrix with each of these row spaces. In cases (c) and (e),
Theorem 5.7 tells us that there is a unique such idempotent. In case (d)
Theorem 5.7 does not apply (since the dimension is not maximal) and in
fact there are continuum-many such idempotents. The unique idempotent
in case (c) is
0 −3 −3
0 −3
0
0
0
0
.
(f)
(g)
(h)
Figure 3. Some non-projective tropical polytopes in PFT2.
Figure 3 illustrates three polytopes in FT3 which fail to be projective for
different reasons. Polytope (f) does not have pure dimension, and so by
Theorem 1.1 cannot be projective. Since the generator dimension and dual
dimension are both equal to the dimension of the ambient space, we may
also deduce this from Theorem 1.4 and the fact it is not min-plus convex.
Polytope (g), which is equivalent to [11, Example 18], does have pure
dimension, but its tropical dimension (2) differs from its generator dimension
and dual dimension (both 3), and hence by Theorem 1.1 is not projective.
Again, since the generator dimension and dual dimension are both equal
to the dimension of the ambient space, non-projectivity also follows from
Theorem 1.4 and the lack of min-plus convexity.
Polytope (h) has pure dimension, but this time its tropical and dual
dimension (3) fail to agree with its generator dimension (4), so again by
Theorem 1.1 it is not projective. In this case Theorem 1.4 does not apply.
Note that if we choose a 4 × 3 matrix with row space polytope (h), the
column space of this matrix will (by Proposition 3.1) yield an example of a
polytope in FT4 with pure tropical dimension 3, generator dimension 3 and
dual dimension 4.
b
b
b
b
b
b
b
b
b
b
24
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
Acknowledgement
The authors thank the anonymous referee for their many useful comments,
and in particular for drawing our attention to several related results in the
literature.
References
[1] M. Akian, R. Bapat, and S. Gaubert. Max-plus algebra. In Handbook
of Linear Algebra. Chapman and Hall, 2006.
[2] M. Akian, S. Gaubert, and A. Guterman. Linear independence over
tropical semirings and beyond. In Tropical and idempotent mathemat-
ics, volume 495 of Contemp. Math., pages 1 -- 38. Amer. Math. Soc.,
Providence, RI, 2009.
[3] M. Akian, S. Gaubert, and A. Guterman. Tropical polyhedra are equiv-
alent to mean payoff games. Internat. J. Algebra Comput., 22:1250001,
2012.
[4] F. L. Baccelli, G. Cohen, G. J. Olsder, and J.-P. Quadrat. Synchro-
nization and linearity. Wiley Series in Probability and Mathematical
Statistics. John Wiley & Sons Ltd., Chichester, 1992.
[5] R. Bieri and J. R. J. Groves. The geometry of the set of characters
induced by valuations. J. Reine Angew. Math., 347:168 -- 195, 1984.
[6] T. S. Blyth and M. F. Janowitz. Residuation theory. Pergamon Press,
Oxford, 1972. International Series of Monographs in Pure and Applied
Mathematics, Vol. 102.
[7] P. Butkovic. Max-linear systems:
theory and algorithms. Springer,
2010.
[8] P. Butkovic, H. Schneider, and S. Sergeev. Generators, extremals and
bases of max cones. Linear Algebra Appl., 421(2-3):394 -- 406, 2007.
[9] G. Cohen, S. Gaubert, and J.-P. Quadrat. Linear projectors in the max-
plus algebra. In Proceedings of the 5th IEEE Mediterranean Conference
on Control and Systems, Paphos, Cyprus, 1997.
[10] G. Cohen, S. Gaubert, and J.-P. Quadrat. Duality and separation the-
orems in idempotent semimodules. Linear Algebra Appl., 379:395 -- 422,
2004.
[11] G. Cohen, S. Gaubert, and J.-P. Quadrat. Projection and aggregation
in max-plus semimodules. In Current Trends in Nonlinear Systems and
Control, Systems and Control: Foundations and Applications, pages
443 -- 454. Birkhauser, Boston, 2006.
[12] G. Cohen, S. Gaubert, and J.-P. Quadrat. Linear projectors on tropical
spaces. Abstract of a talk given at the 2nd International Conference on
Matrix Methods and Operator Equations, Moscow, 2007.
[13] R. Cuninghame-Green. Minimax algebra. Springer-Verlag, Berlin, 1979.
[14] M. Develin, F. Santos, and B. Sturmfels. On the rank of a tropical
matrix. In Combinatorial and computational geometry, volume 52 of
Math. Sci. Res. Inst. Publ., pages 213 -- 242. Cambridge Univ. Press,
Cambridge, 2005.
[15] M. Develin and B. Sturmfels. Tropical convexity. Doc. Math., 9:1 -- 27
(electronic), 2004.
PURE DIMENSION AND PROJECTIVITY OF TROPICAL POLYTOPES
25
[16] N. Eriksson, K. Ranestad, B. Sturmfels, and S. Sullivant. Phylogenetic
algebraic geometry. In Projective varieties with unexpected properties,
pages 237 -- 255. Walter de Gruyter GmbH & Co. KG, Berlin, 2005.
[17] S. Gaubert. Two lectures on max-plus algebra. In Proceedings of the
26th Spring School of Theoretical Computer Science, 1998.
[18] J. S. Golan. Semirings and their applications. Kluwer Academic Pub-
lishers, Dordrecht, 2010.
[19] C. Hollings and M. Kambites. Tropical matrix duality and Green's D
relation. J. London Math. Soc., 86:520 -- 538, 2012.
[20] J. M. Howie. Fundamentals of Semigroup Theory. Clarendon Press,
1995.
[21] I. Itenberg, G. Mikhalkin, and E. Shustin. Tropical algebraic geometry,
volume 35 of Oberwolfach Seminars. Birkhauser Verlag, Basel, second
edition, 2009.
[22] Z. Izhakian and L. Rowen. The tropical rank of a tropical matrix.
Comm. Algebra, 37(11):3912 -- 3927, 2009.
[23] M. Johnson and M. Kambites. Multiplicative structure of 2 × 2 tropical
matrices. Linear Algebra Appl., 435:1612 -- 1625, 2011.
[24] M. Johnson and M. Kambites. Green's J -order and the rank of tropical
matrices. J. Pure Appl. Algebra, 217:280 -- 292, 2013.
[25] M. Joswig. Tropical halfspaces. In Combinatorial and computational
geometry, volume 52 of Math. Sci. Res. Inst. Publ., pages 409 -- 431.
Cambridge Univ. Press, Cambridge, 2005.
[26] M. Joswig and K. Kulas. Tropical and ordinary convexity combined.
Adv. Geom., 10(2):333 -- 352, 2010.
[27] Max Plus Working Group. Max-plus algebra and applications to system
theory and optimal control. In Proceedings of the International Con-
gress of Mathematicians, Vol. 1, 2 (Zurich, 1994), pages 1511 -- 1522,
1995.
[28] L. Pachter and B. Sturmfels. Tropical geometry of statistical models.
Proc. Natl. Acad. Sci. USA, 101(46):16132 -- 16137 (electronic), 2004.
[29] J.-E. Pin. Tropical semirings. In Idempotency (Bristol, 1994), volume 11
of Publ. Newton Inst., pages 50 -- 69. Cambridge Univ. Press, Cambridge,
1998.
[30] E. Wagneur. Moduloıds and pseudomodules I. Dimension theory. Dis-
crete Math., 98(1):57 -- 73, 1991.
|
1711.03708 | 2 | 1711 | 2018-11-02T15:59:41 | Extensions of Enveloping Algebras by Anti-Cocommutative Elements | [
"math.RA"
] | Anti-cocommutative elements were introduced by Wang, Zhang, Zhuang (2013) in their paper Coassociative Lie Algebras. We use this notion to extend universal enveloping algebras of Lie algebras with regards to their Hopf structure, and see if these connected Hopf algebras are enveloping algebras. Furthermore, we apply these results to compare global dimension of connected Hopf algebras and the dimension of their corresponding Lie algebras of primitive elements. Title has been renamed to "GK-Dimension of some Connected Hopf Algebras". | math.RA | math |
GK-Dimension of some Connected Hopf Algebras
Daniel Yee
Bradley University, Peoria, IL 61606
Abstract
While it was identified that the growth of any connected Hopf algebras is
either a positive integer or infinite (see [15]), we have yet to determine the
GK-dimension of a given connected Hopf algebra. We use the notion of
anti-cocommutative elements introduced in [13] to analyze the structure of
connected Hopf algebras generated by anti-cocommutative elements and com-
pute the Gelfand-Kirillov dimension of said algebras. Additionally, we apply
these results to compare global dimension of connected Hopf algebras and
the dimension of their corresponding Lie algebras of primitive elements.
Keywords: Hopf Algebras, Gelfand-Kirillov Dimension
2010 MSC: 16T05
1. Introduction
Connected Hopf algebras are generalizations of universal enveloping alge-
bras U(g) with respect to the Hopf structure, so it is natural to ask if any of
the ring-theoretic properties of enveloping algebras hold for connected Hopf
algebras. Throughout the article, we focus on the Gelfand-Kirillov dimen-
sion, denoted GK.dim, and the global dimension, denoted gl.dim of some
connected Hopf algebras.
We assume that k is an algebraically closed field k of characteristic zero.
All vector spaces, algebras, tensor products, affine-ness and linear maps are
over k. We let τ : V ⊗W → W ⊗V represent the twist map τ (v ⊗w) = w ⊗v,
and let τ ◦ δ be the composition between two linear maps τ and δ.
Given a Hopf algebra H, we denote ∆ : H → H ⊗ H as comultiplication
on H with the Sweedler notation
∆(h) =Xh
h1 ⊗ h2.
Preprint submitted to Elsevier
November 5, 2018
Additionally, let S : H → H be the antipode of H, P (H) be the Lie subal-
gebra of primitive elements of H, i.e.
P (H) = {x ∈ H : ∆(x) = x ⊗ 1 + 1 ⊗ x},
let {Hn : n ≥ 0} be the coradical filtration on H, and gr H be the associated
graded algebra with respect to the coradical filtration. Because P (H) is a
Lie algebra, there exists a Hopf algebra isomorphism between the enveloping
algebra U(P (H)) and the Hopf subalgebra of H generated by P (H), which
we will denote this Hopf subalgebra as UH.
Recall that a Hopf algebra H is connected if H0 = F , and a connected
Hopf algebra is locally finite if every coradical filter is finite dimensional,
or equivalently the Lie algebra of primitive elements is finite dimensional.
Furthermore, given a connected Hopf algebra H, for any n ≥ 1 and for every
h ∈ Hn, we have ∆(h) = h ⊗ 1 + 1 ⊗ h + w, where w ∈ Hn−1 ⊗ Hn−1 (see [9,
Lemma 5.3.2]), and let h be the element in the subspace Hn/Hn−1.
Given a connected Hopf algebra H, we take into consideration a particular
vector space
P2(H) := {c ∈ H : τ ◦ δ(c) = −δ(c), and δ(c) ∈ P (H) ⊗ P (H)},
where the linear map δ : H → H ⊗ H is defined by
δ(h) = ∆(h) − (h ⊗ 1 + 1 ⊗ h),
for all h ∈ H. We call elements belonging to P2(H) anti-cocommutative; it is
clear that δ(P (H)) = 0, so primitive elements are anti-cocommutative. The
reader can find more about anti-cocommutativity and P2(H) in [14]. Finally,
we say that H is primitively thick if GK.dim(H) = dim P (H) + 1 < ∞, or
equivalently, if H is generated by P2(H) as a Hopf algebra, and dim P2(H) =
dim P (H) + 1 is finite (see [14, Lemma 2.6] and [3, Theorem 6.2.12]).
In section 1, we determine the GK-dimension of some connected Hopf
algebras H. In particular, our focus is on Hopf algebras generated by P2(H),
so we will denote A(g) as the class of locally finite connected Hopf algebras
H generated by P2(H) given a finite dimensional Lie algebra g = P (H)
∼= U(g). It immediately follows that every algebra in A(g) is
and H 6= UH
affine. To calculate GK-dimension of these algebras, we need the normality
condition. Recall that given a Hopf algebra H, a Hopf subalgebra A of H is
normal if both
adl[h](a) :=Xh
h1aS(h2) ∈ A, and adr[h](a) :=Xh
S(h1)ah2,
2
for every h ∈ H and every a ∈ A.
Theorem 1.1. If H ∈ A(g), and if UH is a normal Hopf subalgebra of H,
then GK.dim(H) = dim P2(H).
Theorem 1.1 is analogous to the fact that GK.dim(U(g)) = dim g, and g
is the generating subspace of U(g). With finite GK-dimension, algebras in
A(g) have nice ring-theoretic properties thanks to [15, Corollary 6.10].
Corollary 1.2. If H ∈ A(g), and UH is a normal Hopf subalgebra of H,
then H is a Noetherian, Cohen-Macaulay, Auslander-regular domain with
gl.dim(H) = GK.dim(H).
In section 2, we apply the tools in proving Theorem 1.1 into the fol-
lowing result on global dimension of connected Hopf algebras. Note that
gl.dim(H) = l.gl.dim(H) = r.gl.dim(H) for any Hopf algebra H due to [12,
Proposition A.1].
Theorem 1.3. Suppose His any locally finite connected Hopf algebra with
dim P (H) = gl.dim(H) < ∞.
If one of the following conditions holds
1. P (H) is (completely) solvable, or
2. UH is a normal Hopf subalgebra of H,
then H = UH .
The motivation behind Theorem 1.3 is to mimic [15, Lemma 7.2] but we
replace GK-dimension for global dimension.
2. Gelfand-Kirillov Dimension
To compute GK-dimension, we need to connect normal Hopf algebras
with a ring theoretic notion: almost centralizing extensions.
Given algebras R ⊆ S, where S is generated by {x1, ..., xd} over R, we
say that S is an almost centralizing extension of R if the following hold:
1. [r, xi] := rxi − xir ∈ R for all r ∈ R,
3
2. [xi, xj] := xixj − xjxi ∈Pd
for all i, j ≤ d.
m=1 xmR + R,
Lemma 2.1. Let A ⊆ H be connected Hopf algebras, and let {h1, ..., hd}
generate H over A such that δ(hi) ∈ A ⊗ A for all i ≤ d. Then H is an
almost centralizing extension of A if and only if both conditions are satisfied:
1. A is a normal Hopf subalgebra of H, and
2. δ([hi, hj]) ∈Pd
m=1 δ(hmA) + δ(A) for all i, j ≤ d.
Proof. First, assume that H is an almost centralizing extension. It is clear
that condition 2 of the definition implies condition 2, since δ is a linear map.
To show normality, consider any a ∈ A. Since the adjoint map adl[h] is linear
and adl[bc] = adl[c] ◦ adl[b], for all b, c ∈ H, then without loss of generality
let h ∈ {h1, ..., hd} and ∆(h) = h ⊗ 1 + 1 ⊗ h +Ph h1 ⊗ h2. It follows that
S(h) = −h −Ph h1S(h2), and
h1aS(h2)
adl[h](a) = ha + aS(h) +Xh
= ha − ah −Xh
= [h, a] +Xh
ah1S(h2) +Xh
[h1, a]S(h2).
h1aS(h2)
Our assumption δ(hi) ∈ A ⊗ A and A is a Hopf subalgebra implies that
Ph[h1, a]S(h2) ∈ A. Furthermore, almost centralizing extension implies that
[h, a] ∈ A, hence adl[h](A) ⊆ A. Similarly, adr[h](A) ⊆ A, therefore A is a
normal Hopf subalgebra of H.
Conversely, assume that both 1 and 2 hold. Normality combined with
the adjoint computation from the previous paragraph shows that [h, a] =
adl[h](a) −Ph[h1, a]S(h2) ∈ A, for all a ∈ A and all h ∈ {h1, ..., hd}.
With 2, write δ([hi, hj]) = Pm=1 δ(hmam) + δ(a0), where a0, ..., am ∈ A.
Since we have
∆([hi, hj]) = [hi, hj] ⊗ 1 + 1 ⊗ [hi, hj] + δ([hi, hj]),
∆(hmam) = hmam ⊗ 1 + 1 ⊗ hmam + δ(hmam),
∆(a0) = a0 ⊗ 1 + 1 ⊗ a0 + δ(a0),
4
for all m ≤ d, it follows that
∆ [hi, hj] −
d
Xm=1
hmam − a0! = [hi, hj] −
hmam − a0! ⊗ 1
d
Xm=1
+ 1 ⊗ [hi, hj] −
hmam − a0! ,
d
Xm=1
m=1 hmam − a0 ∈ P (H). By our assumption, the generat-
ing set {h1, ..., hd} contains the basis elements of P (H) that are not in P (A).
m=1 hmam − a0 = 0. Since i, j ≤ d are
whence [hi, hj] −Pd
Thus, we may assume that [hi, hj] −Pd
arbitrary, then [hi, hj] ∈ Pm=1 hmA + A, hence H is an almost centralizing
extension of A.
Though the next corollary is not exactly Theorem 1.1, there are cases
where it may be used to compute the GK-dimension of connected Hopf al-
gebras.
Corollary 2.2. Suppose A ⊆ H are affine locally finite connected Hopf al-
gebras such that {h1, ..., hd} ⊆ H generate H over A. If the following hold:
1. δ(hi) ∈ A ⊗ A for all i ≤ d,
2. δ([hi, hj]) ∈Pd
m=1 δ(hmA) + δ(A) for all i, j ≤ d,
3. A is a normal Hopf subalgebra of H with finite GK-dimension,
then GK.dim(H) = GK.dim(A) + d < ∞.
Proof. Lemma 2.1 states that H must be an almost centralizing extension of
A. By [15, Proposition 6.4] the associated graded algebra with respect to the
coradical filtration is gr H = gr A[h1, ..., hd], the commutative polyonomial
ring over gr A. Thus, by [7, Proposition 8.6.7] we have that GK.dim(gr H) =
GK.dim(gr A) + d. Assuming A has finite GK-dimension implies that gr A
is affine and gr A has finite GK-dimension by [15, Theorem 6.9], and hence
gr H has finite GK-dimension. Therefore,
GK.dim(H) = GK.dim(gr H) = GK.dim(A) + d < ∞,
as desired.
5
Example 2.3. Let's consider the connected Hopf algebra H generated by
kx1 + kx2 + kx3 + kz with the relations
[x1, x2] = x2
[z, x1] = z
[x1, x3] = [x2, x3] = 0
[z, x2] = 0
[z, x3] = x2,
and the Hopf structure is given by
ε(kx1 + kx2 + kx3 + kz) = 0, where ε is the counit,
∆(xi) = xi ⊗ 1 + 1 ⊗ xi, for i = 1, 2, 3,
∆(z) = z ⊗ 1 + 1 ⊗ z + x1 ⊗ x2 − x2 ⊗ x1.
(This algebra H is [14, Theorem 3.5(a)] with a = c = 0 and b = 1.) Clearly
every xi ∈ P (H) and z ∈ P2(H). Because of the relation [z, x1] = z, it
follows that UH is not a normal Hopf subalgebra of H, thus we cannot apply
Theorem 1.1. But we can apply Corollary 2.2 to compute GK.dim(H).
Let A be the Hopf subalgebra generated by kx1 + kx2 + kz. It follows
that A is algebra-isomorphic to the enveloping algebra U(g), where the Lie
algebra g = kX1 + kX2 + kZ has the relations [X1, X2] = [Z, X2] = 0,
[Z, X1] = Z. Hence GK.dim(A) = 3. Obviously {x3} generates H over
A and δ(x3) ∈ A ⊗ A, thus satisfying condition 1. Condition 2 is trivial,
and normality immediately follows from adl[x3](a) = −adr[x3](a) = x3a −
ax3 ∈ A for any a ∈ kx1 + kx2 + kz. Therefore, Corollary 2.2 implies that
GK.dim(H) = GK.dim(A) + 1 = 4.
We focus our attention to proving Theorem 1.1. We first decompose the
linear map δ : H → H ⊗ H to two distinct parts.
Definition 2.4. Given a connected Hopf algebra H, we define
δcc = 1
2(δ + τ ◦ δ), and δac = 1
2(δ − τ ◦ δ)
Note that δ = δcc + δac.
Lemma 2.5. Suppose H is a connected Hopf algebra with P = P2(H) and
U = UH. Then
1. δcc([s, t]) = [δ(s), δ(t)] in H ⊗ H, for any s, t ∈ P2(H).
2. δacU = 0 while δccU = δU .
6
3. δacP = δP while δccP = 0.
Proof. In H ⊗ H, notice that for any non-primitive s, t ∈ P , we have
δ([s, t]) = [(s ⊗ 1 + 1 ⊗ s), δ(t)] + [δ(s), (t ⊗ 1 + 1 ⊗ t)] + [δ(s), δ(t)].
Applying the twist map yields
τ ◦ δ([s, t]) = −[(s ⊗ 1 + 1 ⊗ s), δ(t)] − [δ(s), (t ⊗ 1 + 1 ⊗ t)] + [δ(s), δ(t)].
Therefore, (δ + τ ◦ δ)([s, t]) = 2[δ(s), δ(t)], and (δ − τ ◦ δ)([s, t]) = 2[(s ⊗ 1 +
1 ⊗ s), δ(t)] + 2[δ(s), (t ⊗ 1 + 1 ⊗ t)].
The rest is straightforward.
In short, δcc preserves the cocommutative part of δ, while δac preserves
the anti-cocommutative part.
Given a connected Hopf algebra H, since P2(H) is the largest subcoalge-
bra of H consisting of anti-cocommutative elements ([14, Lemma 2.5]), then
one would expect that the preimage of δac belongs to P2(A), and similarly
the preimage of δcc belongs to U(P (H)).
Lemma 2.6. Suppose H ∈ A(g). Let s, t ∈ P2(H) be non-primitive, and let
Un denote the coradical filtration of UH , i.e. Un = UH ∩ Hn for all n ≥ 0.
Then δcc([s, t]) ∈ δ(U3) if and only if δac([s, t]) ∈ δ(P2(H)).
Proof. Assume δcc([s, t]) = δ(w) for some w ∈ U3. By Lemma 2.5, we have
∆([s, t] − w) = ([s, t] − w) ⊗ 1 + 1 ⊗ ([s, t] − w) + δac([s, t]),
thus δ([s, t] − w) = δac([s, t]). Since τ ◦ δac = −δac, then [s, t] − w ∈ P2(H),
whence δ([s, t] − w) = δac([s, t]) ∈ δ(P2(H)).
Now let δac([s, t]) = δ(v) for some v ∈ P2(A). By Lemma 2.5, we have
∆([s, t] − v) = ([s, t] − v) ⊗ 1 + 1 ⊗ ([s, t] − v) + δcc([s, t]),
which implies that [s, t] − v is cocommutative. Since UH is the largest
cocommutative subcoalgebra of H, then [s, t] − v ∈ UH.
It follows that
st ∈ H4, whence [s, t] − v ∈ H4 ∩ UH = U4. Since δcc(v) = 0, we have
7
δcc([s, t]) = δ([s, t] − v) ∈ δ(U4). To show that δcc([s, t]) ∈ δ(U3), consider for
any x1, x2, y1, y2 ∈ P (H),
[(x1 ⊗ y1 − y1 ⊗ x1), (x2 ⊗ y2 − y2 ⊗ x2)]
= x1x2 ⊗ y1y2 − x1y2 ⊗ y1x2 − y1x2 ⊗ x1y2 + y1y2 ⊗ x1x2 − x2x1 ⊗ y2y1
+ y2y1 ⊗ x2x1 + x2x1 ⊗ y2y1 − y2y1 ⊗ x2x1
= x2x1 ⊗ [y1, y2] + [y1, y2] ⊗ x2x1 + y2y1 ⊗ [x1, x2] + [x1, x2] ⊗ y2y1 + [x1, x2] ⊗ [y1, y2]
+ [y1, y2] ⊗ [x1, x2] − (x1x2 ⊗ [y1, x2] + [y1, x2] ⊗ x1y2 + y1x2 ⊗ [x1, y2] + [x1, y2] ⊗ y1x2)
+ [x1, y2] ⊗ [y1, x2] + [y1, x2] ⊗ [x1, y2]
∈ (U2/U1) ⊗ U1 + U1 ⊗ (U2/U1).
Because δ(s), δ(t) ∈ P (H) ⊗ P (H), then δcc([s, t]) ∈ (U2/U1) ⊗ U1 + U1 ⊗
(U2/U1). If [s, t] − v /∈ U3, then δcc([s, t]) = v1 + v2 + u, where v1 ∈ (U3/U2) ⊗
U1, v2 ∈ U1 ⊗ (U3/U2) are both nonzero, and u ∈ U2 ⊗ U2. But this is absurd,
therefore [s, t] − v ∈ U3, hence δcc([s, t]) ∈ δ(U3).
We need an equivalent condition of normality for the Hopf subalgebra
generated by primitive elements, namely one that is analogous to the state-
ment: given a Lie algebra g, a subspace j of g is an ideal if and only if U(j)
is a normal Hopf subalgebra of U(g) (see [8]).
Lemma 2.7. Suppose H ∈ A(g). Then UH is a normal Hopf subalgebra of
H if and only if [t, g] ⊆ g for all t ∈ P2(H).
Proof. Assume t ∈ P2(H) is non-primitive, as t ∈ g is trivial. Since t is
a linear combination of the basis of P2(H), then without loss of generality,
assume that t ∈ P2(H) is a basis element with δ(t) = x ⊗ y − y ⊗ x. It follows
that S(t) = −t + [x, y], so for any g ∈ g,
adr[t](g) = S(t)g + gt − xgy + ygx
= −tg + xyg − yxg + gt + ygx − xgy
= −[t, g] + y[g, x] + x[y, g],
adl[t](g) = [t, g] + [g, x]y + [y, g]x.
If we assume that UH is a normal Hopf subalgebra, then [t, g] ∈ UH , and
since [t, g] ∈ P2(H), we have that [t, g] ∈ UH ∩ P2(H) = P2(UH) = g.
Conversely, given [t, g] ⊆ g, we have adr[t](g) ⊆ UH, for all t ∈ P2(H).
Since adr[ba] = adr[a]◦adr[b] for all a, b ∈ H, then it follows that adr[H](UH) ⊆
UH, and similarly adl[H](UH) ⊆ UH, therefore UH is a normal Hopf subalge-
bra of H.
8
We can now prove Theorem 1.1.
Proof of Theorem 1.1. Since P2(H) is finite dimensional by [14, Lemma 2.5],
is non-primitive. By
definition, H is generated by {t1, ..., tn} over UH . Because δ(P2(H)) ⊆
P (H) ⊗ P (H), showing condition 2 of Corollary 2.2 proves our result.
write P2(H) = (Pn
i=1 kti) ⊕ P (H), where each ti
Without loss of generality, assume that δ(t1) = x1 ⊗ x2 − x2 ⊗ x1, and
δ(t2) = y1 ⊗ y2 − y2 ⊗ y1, where x1, x2, y1, y2 ∈ P (H). Then in H ⊗ H we
have
δac([t1, t2]) = δ([t1, t2]) − [δ(t1), δ(t2)]
= ([t1, y1] ⊗ y2 − y2 ⊗ [t1, y1]) + (y1 ⊗ [t1, y2] − [t1, y2] ⊗ y1)
+ ([x1, t2] ⊗ x2 − x2 ⊗ [x1, t2]) + (x1 ⊗ [x2, t2] − [x2, t2] ⊗ x1).
Lemma 2.7 implies that [ti, P (H)] ⊆ P (H) for all i ≤ n, whence δac([t1, t2]) ∈
δ(P2(H)). Furthermore, by Lemma 2.6, δcc([t1, t2]) ∈ δ(UH). Therefore,
δ([t1, t2]) = δac([t1, t2]) + δcc([t1, t2])
∈ δ(P2(H)) + δ(UH )
⊆
n
Xi=1
δ(tiUH) + δ(UH ).
Since t1, t2 ∈ {t1, ..., tn} are arbitrary, then condition 2 holds, and hence H
is an almost centralizing extension of UH. Therefore,
GK.dim(H) = dim P (H) + n = dim P2(H),
by Corollary 2.2.
Example 2.8. There are many examples that satisfy Theorem 1.1. For ex-
ample, the connected Hopfa algebras generated by the anti-cocommutative
coassociative Lie algebra L in [14, Theorem 3.5(b), (c), (d), (g)], the con-
nected Hopf algebra A in [14, Example 4.1], and the construction of the
connected Hopf algebra L in [3, Section 7.2.3].
Unfortunately not all H ∈ A(g) satisfies Theorem 1.1, especially when g
is a semisimple Lie algebra.
Theorem 2.9. If H is a locally finite connected Hopf algebra such that P (H)
is a semsimple Lie algebra, and if UH is a normal Hopf subalgebra, then we
have H = UH.
9
Proof. If H 6= UH, then by [14, Lemma 2.6], P2(H) 6= P (H). Since UH is a
normal Hopf subalgebra of H, for any non-primitive t ∈ P2(H), [P (H), t] ⊆
P (H) by Lemma 2.5, hence if A is the subalgebra of H generated by the
subspace P (H) ⊕ kt, then A is a Hopf sublagebra since δ(t) ∈ P (H) ⊗ P (H).
Due to the relation, if we view the subspace h = P (H) ⊕ kt as a Lie algebra,
it follows that A is isomorphic to the enveloping algebra U(h) as algebras,
hence GK.dim(A) = dim P (H) + 1, hence A is primitively thick. But this
contradicts [3, Proposition 6.4.5], therefore we must have H = UH .
Transitioning into our next section on global dimension, we first pose the
following question.
Question 2.10. If H is a locally finite connected Hopf algebra, is the GK-
dimension of H finite if and only if the global dimension of H is finite? And
does GK.dim(H) = gl.dim(H) when either is finite?
The result [15, Corollary 6.10] showed that whenever a connected Hopf
algebra H has finite GK-dimension, then the global dimension H is exactly
the GK-dimension of H, which proves only one direction.
3. Application: Global Dimension
In this section we ask if the dimension of P (H) is exactly the global di-
mension of the connected Hopf algebra H, is H isomorphic to the enveloping
algebra U(P (H))?
Lemma 3.1. Suppose that H is any connected Hopf algebra and A is a Hopf
subalgebra of H. Then gl.dim(A) ≤ gl.dim(H) when A is left Noetherian with
finite left global dimension.
Proof. By [11], H is a faithfully flat left A-module. As A is left Noetherian
with finite global dimension, [7, Theorem 7.2.6] implies that gl.dim(A) ≤
gl.dim(H).
Corollary 3.2. If H is a locally finite connected Hopf algebra, then dim P (H) ≤
gl.dim(H).
Proof. Apply Lemma 3.1 with A = UH.
Before proving Theorem 1.3, we need to see how the Lie algebra P (H)
effects the structure of H.
10
Lemma 3.3. Suppose H ∈ A(g) where g is a (completely) solvable Lie alge-
bra. Then there exists a primitively thick Hopf subalgebra A of H such that
UH ⊆ A.
Proof. Clearly the vector space P2(H)/g is a g-module. Applying [4, Corol-
lary 2.4.3], there exists v ∈ P2(H)/g such that x(v) = λ(x)v for all x ∈ g,
where λ : g → k is a linear map. Since x(v) = [x, v] in H, we let A be the
subalgebra of H generated by g ⊕ kv. It is clear that UH ⊆ A, and A is a
Hopf algebra since ∆(v) ∈ g ⊗ g. Additonally, we may view h = g ⊕ kv as a
Lie algebra since [g, v] = kv. It follows that A is isomorphic to the envelop-
ing algebra U(h) as algebras, and thus GK.dim(A) = dim g + 1, hence A is
primitively thick.
We can now prove Theorem 1.3.
Proof of Theorem 1.3. We start by assuming the contrary, H 6= UH . As a
consequence, P2(H) 6= P (H) by [14, Lemma 2.6].
1. First we assume that P (H) is (completely) solvable. By Lemma
3.3, there exists a primitively thick Hopf subalgebra A of H such that A
is algebra-isomorphic to an enveloping algebra U(h), where h is a Lie al-
gebra of dimension dim P (H) + 1. This implies that A is Noetherian and
gl.dim(A) = dim P (H) + 1, and thus Lemma 3.1 forces
dim P (H) < gl.dim(A) ≤ gl.dim(H),
which is a contradiction to our assumption. Therefore H = UH.
2.
If we assume that UH is normal, then consider any non-primitive
t ∈ P2(H). By Lemma 2.7, [t, P (H)] ⊆ P (H). Let A be the Hopf subalgebra
of H generated by P (H)⊕kt. It follows that A is primitively thick containing
UH, thus it must be algebra-isomorphic to an enveloping algebra U(L), where
L is a Lie algebra of dimension dim P (H) + 1. Hence A is a Noetherian Hopf
subalgebra with gl.dim(A) = dim P (H) + 1. Applying Lemma 3.1 yields
dim P (H) < gl.dim(A) ≤ gl.dim(H),
a contradiction. Therefore H = UH.
Theorem 1.3 also provides us a strict lower bound of Corollary 3.2, namely
dim P (H) < gl.dim(H), whenever condition 1 or 2 hold. Relating to GK-
dimension, a consequence of Theorem 1.3 shows that gl.dim(H) > dim P (H)
if and only if GK.dim(H) > dim P (H).
11
Futhermore, we see that Theorem 1.3 is a specific setting for a more
general question.
Question 3.4. If A ⊆ H are (affine) locally finite connected Hopf algebras
with gl.dim(H) = gl.dim(A), does H = A?
Acknowledgments
The author would like to thank his adviser Allen Bell, and the Mathe-
matics Department at both University of Wisconsin-Milwaukee and Bradley
University for their guidance and support of the author's work. The author
is also grateful to the referees for their constructive feedback and advice.
References
References
[1] K.A. Brown, S. P. Gilmartin, J.J. Zhang, Connected (graded) Hopf Al-
gebras, preprint, arXiv:1601.06687, 2016.
[2] Enveloping Algebras, Graduate Studies in Matheamtics #11, American
Mathematical Society, Providence, 1996.
[3] P. Gilmartin, Connected Hopf Algebras of Finite Gelfand-Kirillov Di-
mension, Ph.D. Thesis, University of Glasgow, 2016.
[4] W.A. de Graff, Lie Algebras: Theory and Algorithms, Vol. 56, North-
Holland Mathematical Library, Elsevier Science B.V., 2000.
[5] K.R. Goodearl, R. Warfield, Intorduction to Noncommutative Noethe-
rian Rings, London Mathematical Society Student Texts #61, Cam-
bridge University Press, Cambridge, 2004.
[6] N. Jacobson, Lie Algebras, Dover Publications Inc., Mineola, 1962.
[7] J.C. McConnell, J.C. Robson, Noncommutative Noetherian Rings,
Graduate Studies in matheamtics #30, American Mathematical Soci-
ety, Providence, 2001.
[8] J.W. Milnor, J.C. Moore, On the Structure of Hopf Algebras, The An-
nals of Mathematics, Second Series 81 (1965) 211-264.
12
[9] S. Montgomery, Hopf Algebras and Their Actions on Rings, Confer-
ence Board of the Mathematical Sciences #82, American Mathematical
Society, Providence, 1994.
[10] D. Radford, Hopf Algebras, Series on Knots and Everything Vol. #49,
World Scientific, 2011.
[11] M. Takeuchi, A Correspondence between Hopf ideals and sub-Hopf al-
gebras, Manuscripta Mathematica 7 (1972), 251-270.
[12] X. Wang, X. Yu, Y. Zhang, Calabi-Yau property under monoidal Morita-
Takeuchi equivalence, Pacific J. Math., 290, no. 2 (2017), 481-510.
[13] D.G. Wang, J.J. Zhang, G. Zhuang, Coassociative Lie Algebras, Glasgow
Mathematics Journal 55A (2013) 195-215.
[14] D.G. Wang, J.J. Zhang, G. Zhuang, Connected Hopf Algebras of
Gelfand-Kirillov Dimension Four, Transactions of the American Math-
ematical Society 367 (2015) 5597-5632.
[15] G. Zhuang, Properties of Pointed and Connected Hopf Algebras of Finite
Gelfand-Kirillov Dimension, Journal of London Mathematical Society 87
(2013) 877-898.
Department of Mathematics, Bradley University, Peoria,
Illinois 61606
E-mail addresss: [email protected]
13
|
1107.5369 | 1 | 1107 | 2011-07-27T02:25:38 | Thin Hessenberg Pairs and Double Vandermonde Matrices | [
"math.RA"
] | A square matrix is called {\it Hessenberg} whenever each entry below the subdiagonal is zero and each entry on the subdiagonal is nonzero. Let $V$ denote a nonzero finite-dimensional vector space over a field $\fld$. We consider an ordered pair of linear transformations $A: V \rightarrow V$ and $A^*: V \rightarrow V$ which satisfy both (i), (ii) below.
(i) There exists a basis for $V$ with respect to which the matrix representing $A$ is Hessenberg and the matrix representing $A^*$ is diagonal.
(ii) There exists a basis for $V$ with respect to which the matrix representing $A$ is diagonal and the matrix representing $A^*$ is Hessenberg.
\noindent We call such a pair a {\it thin Hessenberg pair} (or {\it TH pair}). By the {\it diameter} of the pair we mean the dimension of $V$ minus one. There is an "oriented" version of a TH pair called a TH system. In this paper we investigate a connection between TH systems and double Vandermonde matrices. We give a bijection between any two of the following three sets:
\cdot The set of isomorphism classes of TH systems over $\K$ of diameter $d$.
\cdot The set of normalized west-south Vandermonde systems in $\Mdf$.
\cdot The set of parameter arrays over $\K$ of diameter $d$.
We give a bijection between any two of the following five sets:
\cdot The set of affine isomorphism classes of TH systems over $\K$ of diameter $d$.
\cdot The set of isomorphism classes of RTH systems over $\K$ of diameter $d$.
\cdot The set of affine classes of normalized west-south Vandermonde systems in $\Mdf$.
\cdot The set of normalized west-south Vandermonde matrices in $\Mdf$.
\cdot The set of reduced parameter arrays over $\K$ of diameter $d$. | math.RA | math |
Thin Hessenberg Pairs and Double
Vandermonde Matrices
Ali Godjali
July 26, 2011
Abstract
A square matrix is called Hessenberg whenever each entry below the subdiagonal is
zero and each entry on the subdiagonal is nonzero. Let V denote a nonzero finite-
dimensional vector space over a field K. We consider an ordered pair of linear trans-
formations A : V → V and A∗ : V → V which satisfy both (i), (ii) below.
(i) There exists a basis for V with respect to which the matrix representing A is
Hessenberg and the matrix representing A∗ is diagonal.
(ii) There exists a basis for V with respect to which the matrix representing A is
diagonal and the matrix representing A∗ is Hessenberg.
We call such a pair a thin Hessenberg pair (or TH pair). By the diameter of the pair
we mean the dimension of V minus one. There is an "oriented" version of a TH pair
called a TH system. In this paper we investigate a connection between TH systems
and double Vandermonde matrices. We have two main results. For the first result we
give a bijection between any two of the following three sets:
• The set of isomorphism classes of TH systems over K of diameter d.
• The set of normalized west-south Vandermonde systems in Matd+1(K).
• The set of parameter arrays over K of diameter d.
For the second result we give a bijection between any two of the following five sets:
• The set of affine isomorphism classes of TH systems over K of diameter d.
• The set of isomorphism classes of RTH systems over K of diameter d.
• The set of affine classes of normalized west-south Vandermonde systems in Matd+1(K).
• The set of normalized west-south Vandermonde matrices in Matd+1(K).
• The set of reduced parameter arrays over K of diameter d.
Keywords: Leonard pair, Hessenberg pair, Vandermonde matrix.
2010 Mathematics Subject Classification: 15A04.
1
1
Introduction
This paper is about a linear algebraic object called a thin Hessenberg pair [1]. To recall its
definition, we will use the following term. A square matrix is called Hessenberg whenever
each entry below the subdiagonal is zero and each entry on the subdiagonal is nonzero.
Throughout the paper, K will denote a field.
Definition 1.1. [1, Definition 1.1] Let V denote a nonzero finite-dimensional vector space
over K. By a thin Hessenberg pair (or TH pair) on V , we mean an ordered pair of linear
transformations A : V → V and A∗ : V → V which satisfy both (i), (ii) below.
(i) There exists a basis for V with respect to which the matrix representing A is Hessenberg
and the matrix representing A∗ is diagonal.
(ii) There exists a basis for V with respect to which the matrix representing A is diagonal
and the matrix representing A∗ is Hessenberg.
We call V the underlying vector space and say that A, A∗ is over K. By the diameter of
A, A∗ we mean the dimension of V minus one.
Note 1.2. It is a common notational convention to use A∗ to represent the conjugate-
transpose of A. We are not using this convention. In a TH pair A, A∗ the linear transforma-
tions A and A∗ are arbitrary subject to (i), (ii) above.
A TH pair is a generalization of a Leonard pair [4]. Roughly speaking, a Leonard pair is a
pair of linear transformations as in Definition 1.1, with the Hessenberg requirement replaced
by an irreducible tridiagonal requirement. Leonard pairs have been extensively studied; for
more information see [5] and the references therein.
In [1] we introduced the concept of a TH pair and began a systematic study of these
objects. We now summarize the content of [1].
In [1, Definition 2.2] we introduced an
"oriented" version of a TH pair called a TH system. A TH system is described as follows.
Let A, A∗ denote a TH pair on V of diameter d. By definition A is diagonalizable. It turns
out that each eigenspace of A has dimension one [1, Lemma 2.1]. Therefore a basis from
Definition 1.1(ii) induces an ordering {Vi}d
i=0 of the eigenspaces of A. For 0 ≤ i ≤ d, let Ei
denote the primitive idempotent of A that corresponds to Vi. We call {Ei}d
i=0 a standard
ordering of the primitive idempotents of A. A standard ordering of the primitive idempotents
of A∗ is defined similarly. A TH system is a TH pair A, A∗ together with a standard ordering
of the primitive idempotents of A and a standard ordering of the primitive idempotents of
A∗. Let (A; {Ei}d
i=0) denote a TH system on V . In [1] we investigated six bases
for V with respect to which the matrices representing A and A∗ are attractive. We displayed
these matrices along with the transition matrices relating the bases. We classified the TH
systems up to isomorphism.
i=0; A∗; {E∗
i }d
In the present paper, we continue our study of TH pairs and TH systems. Our focus is
on a connection between TH systems and double Vandermonde matrices. We establish two
main results. These results have the following form. In the first result we display three sets
and show any two are in bijection. In the second result we display five sets and show any
two are in bijection. We now describe the first result. To do this we display the three sets
and then discuss the meaning. The three sets are:
2
• The set of isomorphism classes of TH systems over K of diameter d.
• The set of normalized west-south Vandermonde systems in Matd+1(K).
• The set of parameter arrays over K of diameter d.
i }d
i }d
i=0, {φi}d
i=0, {θ∗
i }d
d−i(θ∗
i=0, {θ∗
i }d
i=0 are mutually distinct; (ii) {θ∗
i=0) such that: (i) X is a matrix in Matd+1(K); (ii) {θi}d
We now describe the above three sets in more detail. The first set is clear, so consider
the second set. For an indeterminate λ let K[λ] denote the K-algebra consisting of the
polynomials in λ that have all coefficients in K. Let {fi}d
i=0 denote a sequence of polynomials
in K[λ]. We say that {fi}d
i=0 is graded whenever f0 = 1 and fi has degree i for 0 ≤ i ≤
d. By a normalized west-south Vandermonde system in Matd+1(K) we mean a sequence
(X, {θi}d
i=0 is a sequence of
mutually distinct scalars in K; (iii) {θ∗
i=0 is a sequence of mutually distinct scalars in K;
(iv) there exists a graded sequence of polynomials {fi}d
i=0 in K[λ] such that Xij = fj(θi)
for 0 ≤ i, j ≤ d; (v) there exists a graded sequence of polynomials {f ∗
i=0 in K[λ] such
that Xij = f ∗
j ) for 0 ≤ i, j ≤ d. We now describe the third set. By a parameter array
over K of diameter d we mean a sequence ({θi}d
i=1) of scalars taken from K
such that: (i) {θi}d
i=1
are all nonzero. We have now described the three sets. We now describe the bijections
between these sets. We start by describing the bijection from the first set to the second
set. Let Φ = (A; {Ei}d
i=0) denote a TH system on V . Associated with Φ is
a certain matrix P ∈ Matd+1(K). This is the transition matrix from a basis in Definition
1.1(ii) to a basis in Definition 1.1(i), where the bases are normalized so that each entry in
the leftmost column and the bottom row of P is 1. For 0 ≤ i ≤ d let θi (resp. θ∗
i ) denote
the eigenvalue of A (resp. A∗) that corresponds to Ei (resp. E∗
i ). Our bijection sends
the isomorphism class of Φ to (P, {θi}d
i=0). We now describe the bijection from the
third set to the first set. Let ({θi}d
i=1) denote a parameter array over K of
diameter d. Let A denote the lower bidiagonal matrix in Matd+1(K) with entries Aii = θd−i
for 0 ≤ i ≤ d and Ai,i−1 = φi for 1 ≤ i ≤ d. Let A∗ denote the upper bidiagonal matrix
in Matd+1(K) with entries A∗
i−1,i = 1 for 1 ≤ i ≤ d. Observe
i=0) is an ordering of the eigenvalues of A (resp. A∗). For 0 ≤ i ≤ d
that {θi}d
i ) denote the primitive idempotent of A (resp. A∗) that corresponds to θi
let Ei (resp. E∗
(resp. θ∗
i=0) is a TH system. Our bijection sends
({θi}d
i }d
i=0 are mutually distinct; (iii) {φi}d
i ). We show that Φ = (A; {Ei}d
ii = θ∗
i for 0 ≤ i ≤ d and A∗
i=1) to the isomorphism class of Φ.
i }d
i=0, {φi}d
i=0 (resp. {θ∗
i }d
i=0, {θ∗
i }d
i=0, {φi}d
i=0, {θ∗
i=0, {θ∗
i }d
i=0; A∗; {E∗
i }d
i=0; A∗; {E∗
i }d
We now describe our second result, which is a variation on the first result. We mentioned
above that the second result involves five sets. The five sets are:
• The set of affine isomorphism classes of TH systems over K of diameter d.
• The set of isomorphism classes of RTH systems over K of diameter d.
• The set of affine classes of normalized west-south Vandermonde systems in Matd+1(K).
• The set of normalized west-south Vandermonde matrices in Matd+1(K).
• The set of reduced parameter arrays over K of diameter d.
3
i }d
i }d
i + β∗}d
i=0; {E∗
i }d
i=0) induced by a TH system (A; {Ei}d
i=0; A∗; {E∗
i=0; α∗A∗ + β∗I; {E∗
We now describe the above five sets in more detail. Throughout the description let α, β, α∗, β∗
denote scalars in K with α, α∗ nonzero. We now describe the first set. Let
Φ = (A; {Ei}d
i=0) denote a TH system over K. Observe that the sequence
(αA + βI; {Ei}d
i=0) is a TH system over K, said to be an affine trans-
formation of Φ. We now describe the second set. By an RTH system over K we mean
the sequence ({Ei}d
i=0) over K.
We now describe the third set. Let (X, {θi}d
i=0) denote a normalized west-south
Vandermonde system in Matd+1(K). One checks that (X, {αθi + β}d
i=0)
is a normalized west-south Vandermonde system in Matd+1(K). These two systems are
said to be affine related. We now describe the fourth set. By a normalized west-south
Vandermonde matrix in Matd+1(K) we mean the matrix X induced by a normalized west-
south Vandermonde system (X, {θi}d
i=0) in Matd+1(K). We now describe the fifth
set. Let ({θi}d
i=1) denote a parameter array over K. Observe that ({αθi +
β}d
i=1) is a parameter array over K. These two parameter arrays
are said to be affine related. This affine relation is an equivalence relation; the equivalence
classes are called reduced parameter arrays. We have now described the five sets. We omit
the description of the bijections between these sets as they are not hard to guess.
i }d
i=0, {φi}d
i=0, {αα∗φi}d
i=0; A∗; {E∗
i }d
i=0, {α∗θ∗
i=0, {θ∗
i }d
i=0, {θ∗
i }d
i=0, {θ∗
i=0, {α∗θ∗
i + β∗}d
This paper is organized as follows.
In Section 5 we discuss affine transformations of a TH system.
In Sections 2, 3 we review some basic concepts
regarding TH pairs and TH systems. In Section 4 we summarize the classification of TH
systems given in [1].
In
Sections 6, 7 we discuss how a given TH system yields three more TH systems called the
relatives. In Sections 8, 9 we discuss some scalars that are helpful in describing a given TH
system. In Section 10 we use these scalars to describe the relatives of a given TH system.
In Sections 11, 12 we discuss the transition matrix P and a related matrix P . In Section 13
we define the notion of a Vandermonde system. In Sections 14 -- 16 we discuss the connection
between Vandermonde systems, graded sequences of polynomials, and Hessenberg matrices.
In Section 17 we discuss the double Vandermonde structure of the transition matrices P and
P . Sections 18, 19 contain the main results of the paper.
2 Thin Hessenberg systems
In our study of a TH pair, it is often helpful to consider a closely related object called a
TH system. Before defining this notion, we make some definitions and observations. For the
rest of the paper, fix an integer d ≥ 0. Let Matd+1(K) denote the K-algebra consisting of
the (d + 1) × (d + 1) matrices that have all entries in K. We index the rows and columns by
0, 1, . . . , d. Let Kd+1 denote the K-vector space consisting of the (d + 1) × 1 matrices that
have all entries in K. We index the columns by 0, 1, . . . , d. Observe that Matd+1(K) acts
on Kd+1 by left multiplication. For the rest of the paper, fix a vector space V over K with
dimension d + 1. Let End(V ) denote the K-algebra consisting of the linear transformations
from V to V . Suppose that {vi}d
i=0 is a basis for V . For X ∈ Matd+1(K) and Y ∈ End(V ),
we say that X represents Y with respect to {vi}d
i=0 Xijvi for 0 ≤ j ≤ d.
For A ∈ End(V ) and W ⊆ V , we call W an eigenspace of A whenever W 6= 0 and there
exists θ ∈ K such that W = {v ∈ V Av = θv}. In this case θ is called the eigenvalue
of A corresponding to W . We say that A is diagonalizable whenever V is spanned by the
i=0 whenever Y vj = Pd
4
eigenspaces of A. We say that A is multiplicity-free whenever A is diagonalizable and each
eigenspace of A has dimension one.
Lemma 2.1. [1, Lemma 2.1] Let A, A∗ denote a TH pair on V . Then each of A, A∗ is
multiplicity-free.
i=0 denote an ordering of the eigenspaces of A and let {θi}d
We recall a few more concepts from linear algebra. Let A denote a multiplicity-free element
of End(V ). Let {Vi}d
i=0 denote
the corresponding ordering of the eigenvalues of A. For 0 ≤ i ≤ d, define Ei ∈ End(V )
such that (Ei − I)Vi = 0 and EiVj = 0 for j 6= i (0 ≤ j ≤ d). Here I denotes the identity
of End(V ). We call Ei the primitive idempotent of A corresponding to Vi (or θi). Observe
i=0 Ei; (ii) EiEj = δi,jEi (0 ≤ i, j ≤ d); (iii) Vi = EiV (0 ≤ i ≤ d); (iv)
that (i) I = Pd
A = Pd
i=0 θiEi. Moreover
Ei = Y0≤j≤d
j6=i
A − θjI
θi − θj
(0 ≤ i ≤ d).
(1)
Note that each of {Ai}d
i=0 is a basis for the K-subalgebra of End(V ) generated by
i=0, {Ei}d
i=0(A − θiI) = 0.
A. Moreover Qd
We now define a TH system.
Definition 2.2. By a thin Hessenberg system (or TH system) on V we mean a sequence
Φ = (A; {Ei}d
i=0; A∗; {E∗
i }d
i=0)
which satisfies (i) -- (v) below.
(i) Each of A, A∗ is a multiplicity-free element of End(V ).
(ii) {Ei}d
i=0 is an ordering of the primitive idempotents of A.
(iii) {E∗
i }d
i=0 is an ordering of the primitive idempotents of A∗.
(iv) EiA∗Ej = (cid:26) 0,
j = (cid:26) 0,
(v) E∗
i AE∗
6= 0,
6= 0,
if i − j > 1
if i − j = 1
if i − j > 1
if i − j = 1
(0 ≤ i, j ≤ d).
(0 ≤ i, j ≤ d).
We refer to d as the diameter of Φ. We call V the underlying vector space and say that Φ is
over K.
i V ). Then the sequence {vi}d
We comment on how TH pairs and TH systems are related. Let (A; {Ei}d
i=0)
denote a TH system on V . For 0 ≤ i ≤ d, let vi (resp. v∗
i ) denote a nonzero vector
in EiV (resp. E∗
i }d
i=0) is a basis for V which
satisfies Definition 1.1(ii) (resp. Definition 1.1(i)). Therefore the pair A, A∗ is a TH pair
on V . Conversely, let A, A∗ denote a TH pair on V . Then each of A, A∗ is multiplicity-free
by Lemma 2.1. Let {vi}d
i=0) denote a basis for V which satisfies Definition
1.1(ii) (resp. Definition 1.1(i)). For 0 ≤ i ≤ d, the vector vi (resp. v∗
i ) is an eigenvector
for A (resp. A∗); let Ei (resp. E∗
i ) denote the corresponding primitive idempotent. Then
(A; {Ei}d
i=0) is a TH system on V .
i=0 (resp. {v∗
i=0 (resp. {v∗
i=0; A∗; {E∗
i=0; A∗; {E∗
i }d
i }d
i }d
5
Definition 2.3. Let Φ = (A; {Ei}d
A, A∗ is a TH pair on V . We say that this pair is associated with Φ.
i=0; A∗; {E∗
i }d
i=0) denote a TH system on V . Observe that
Remark 2.4. With reference to Definition 2.3, conceivably a given TH pair is associated
with many TH systems.
We now recall several definitions and results on TH systems.
Definition 2.5. Let Φ = (A; {Ei}d
let θi (resp. θ∗
refer to {θi}d
sequence of Φ. We observe that {θi}d
{θ∗
i }d
i ) denote the eigenvalue of A (resp. A∗) corresponding to Ei (resp. E∗
i=0; A∗; {E∗
i }d
i=0) denote a TH system on V . For 0 ≤ i ≤ d,
i ). We
i=0 as the dual eigenvalue
i=0 are mutually distinct and contained in K. Similarly
i }d
i=0 as the eigenvalue sequence of Φ. We refer to {θ∗
i=0 are mutually distinct and contained in K.
Definition 2.6. Let A, A∗ denote a TH pair. By an eigenvalue sequence (resp. dual eigen-
value sequence) of A, A∗, we mean the eigenvalue sequence (resp. dual eigenvalue sequence)
of an associated TH system. We emphasize that a given TH pair could have many eigenvalue
and dual eigenvalue sequences.
Let K[λ] denote the K-algebra consisting of the polynomials in λ that have all coefficients
in K.
Notation 2.7. Let {θi}d
0 ≤ i ≤ d + 1, let τi, τ ∗
i , ηi, η∗
i=0, {θ∗
i }d
i=0 denote two sequences of scalars taken from K. For
i denote the following polynomials in K[λ].
i−1
i−1
(λ − θh),
(λ − θd−h),
(λ − θ∗
h),
(λ − θ∗
d−h).
ηi =
η∗
i =
Yh=0
Yh=0
i−1
τi =
τ ∗
i =
i−1
Yh=0
Yh=0
i , η∗
We observe that each of τi, ηi, τ ∗
i is monic with degree i.
By (1), for 0 ≤ i ≤ d
Ei =
τi(A)ηd−i(A)
τi(θi)ηd−i(θi)
,
E∗
i =
τ ∗
i (A∗)η∗
i )η∗
τ ∗
i (θ∗
d−i(A∗)
d−i(θ∗
i )
.
(2)
By a decomposition of V we mean a sequence {Ui}d
i=0 consisting of one-dimensional
subspaces of V such that
V = U0 + U1 + · · · + Ud
(direct sum).
For notational convenience, set U−1 = 0 and Ud+1 = 0.
i=0; A∗; {E∗
Let Φ = (A; {Ei}d
i V }d
i=0 is a de-
composition of V , said to be Φ-standard. Let 0 6= ξ0 ∈ E0V . The sequence {E∗
i ξ0}d
i=0 is a
basis for V [1, Lemma 8.1], said to be Φ-standard. We recall another decomposition of V
associated with Φ. For 0 ≤ i ≤ d, let
i=0) denote a TH system on V . Then {E∗
i }d
6
Ui = (E∗
0V + E∗
1 V + · · · + E∗
i V ) ∩ (E0V + E1V + · · · + Ed−iV ).
(3)
The sequence {Ui}d
for 0 ≤ i ≤ d, both
i=0 is a decomposition of V [1, Section 4], said to be Φ-split. Moreover
(A − θd−iI)Ui = Ui+1,
(A∗ − θ∗
i I)Ui = Ui−1.
Setting i = d in (3) we find Ud = E0V . Combining this with (5) we find
Ui = η∗
d−i(A∗)E0V
(0 ≤ i ≤ d).
(4)
(5)
(6)
Recall 0 6= ξ0 ∈ E0V . From (6) we find that for 0 ≤ i ≤ d, the vector η∗
for Ui. By this and since {Ui}d
i=0 is a decomposition of V , the sequence
η∗
d−i(A∗)ξ0
(0 ≤ i ≤ d)
d−i(A∗)ξ0 is a basis
(7)
is a basis for V , said to be Φ-split. Let 1 ≤ i ≤ d. By (5) we have (A∗ −θ∗
i I)Ui = Ui−1, and by
(4) we have (A − θd−i+1I)Ui−1 = Ui. Therefore Ui is invariant under (A − θd−i+1I)(A∗ − θ∗
i I)
and the corresponding eigenvalue is a nonzero element of K. We denote this eigenvalue by
φi. We call the sequence {φi}d
i=1 the split sequence of Φ. For notational convenience, set
φ0 = 0 and φd+1 = 0.
Proposition 2.8. [1, Proposition 4.4] Let Φ = (A; {Ei}d
on V with eigenvalue sequence {θi}d
{φi}d
i=0) denote a TH system
i }d
i=0, and split sequence
i=1. Then the matrices representing A and A∗ with respect to a Φ-split basis for V are
i }d
i=0, dual eigenvalue sequence {θ∗
i=0; A∗; {E∗
θd
φ1 θd−1
φ2
0
θd−2
·
·
·
0
·
φd θ0
,
θ∗
0
1
θ∗
1
1
θ∗
2
0
·
·
·
·
0
1
θ∗
d
(8)
respectively.
Next we describe the matrices representing the primitive idempotents of A, A∗ with respect
to a Φ-split basis for V .
Proposition 2.9. Let Φ = (A; {Ei}d
i=0) denote a TH system on V with eigen-
value sequence {θi}d
i=1. For
0 ≤ r ≤ d, consider the matrices in Matd+1(K) that represent Er and E∗
r with respect to a
Φ-split basis. For 0 ≤ i, j ≤ d, their (i, j)-entry is described as follows. For Er this entry is
i=0, dual eigenvalue sequence {θ∗
i=0, and split sequence {φi}d
i=0; A∗; {E∗
i }d
i }d
φ1φ2 · · · φi
φ1φ2 · · · φj
τd−i(θr)ηj(θr)
τr(θr)ηd−r(θr)
,
and for E∗
r this entry is
τ ∗
i (θ∗
r (θ∗
τ ∗
r )η∗
r )η∗
d−j(θ∗
r )
d−r(θ∗
r )
.
7
(9)
(10)
Proof: Fix a Φ-split basis for V . For notational convenience, identify each element of End(V )
with the matrix in Matd+1(K) that represents it with respect to this basis. We first show
that the (i, j)-entry of E∗
r is given by (10). Computing the (i, j)-entry of A∗E∗
r using
matrix multiplication, and taking into account the form of A∗ in (8), we find
r = θ∗
r E∗
(E∗
r )i+1,j = (θ∗
r − θ∗
i )(E∗
r )ij
if i ≤ d − 1. Replacing i by i − 1 in the above line, we find
(E∗
r )ij = (θ∗
r − θ∗
i−1)(E∗
r )i−1,j
if i ≥ 1. Using the recursion (11), we routinely find
(E∗
r )ij = (θ∗
= τ ∗
r − θ∗
i (θ∗
i−1)(θ∗
r )0j.
r )(E∗
r − θ∗
i−2) · · · (θ∗
r − θ∗
0)(E∗
r )0j
(11)
(12)
Computing the (0, j)-entry of E∗
account the form of A∗, we find
r A∗ = θ∗
r E∗
r using matrix multiplication, and taking into
(E∗
r )0,j−1 = (θ∗
r − θ∗
j )(E∗
r )0j
if j ≥ 1. Replacing j by j + 1 in the above line we find
(E∗
r )0j = (θ∗
r − θ∗
j+1)(E∗
r )0,j+1
if j ≤ d − 1. Using the recursion (13), we routinely find
(E∗
r )0j = (θ∗
= η∗
r − θ∗
d−j(θ∗
j+1)(θ∗
r )(E∗
r )0d.
r − θ∗
j+2) · · · (θ∗
r − θ∗
d)(E∗
r )0d
Combining (12), (14), we find
(E∗
r )ij = τ ∗
i (θ∗
r )η∗
d−j(θ∗
r )c,
(13)
(14)
(15)
where we abbreviate c = (E∗
is a polynomial in A∗, we see E∗
of (E∗
r )rr equals 0 or 1. We show (E∗
r )0d. We now find c. Since A∗ is upper triangular, and since E∗
r
r , so the diagonal entry
r is upper triangular. Recall E∗2
r = E∗
r )rr = 1. Setting i = r, j = r in (15),
(E∗
r )rr = τ ∗
r (θ∗
r )η∗
d−r(θ∗
r )c.
(16)
r (θ∗
d−r(θ∗
r ) 6= 0 and η∗
r ) 6= 0 by Notation 2.7, and since {θ∗
Observe τ ∗
c 6= 0; otherwise E∗
(E∗
evaluating (15) using the result, we find the (i, j)-entry of E∗
We now show that the (i, j)-entry of Er is given by (9). Let G ∈ Matd+1(K) denote the
i=0 are distinct. Observe
r = 0 in view of (15). Apparently the right side of (16) is not 0, so
r )rr = 1 in (16), solving for c, and
r )rr 6= 0, and we conclude (E∗
r )rr = 1. Setting (E∗
r is given by (10).
i }d
8
diagonal matrix with (i, i)-entry φ1φ2 · · · φi for 0 ≤ i ≤ d and set A′ := GAtG−1, where A is
the matrix on the left of (8). The matrix A′ is equal to
θd
1
θd−1
1
θd−2
0
·
·
·
·
0
1
θ0
(17)
r denote the primitive idempotent of A′ associated with the eigenvalue θr. We find E′
Let E′
in two ways. On one hand, applying (10) to A′, we find E′
r has (i, j)-entry
r
for 0 ≤ i, j ≤ d. On the other hand, by elementary linear algebra
τd−j(θr)ηi(θr)
τr(θr)ηd−r(θr)
E′
r = GEt
rG−1,
so E′
r has (i, j)-entry
Gii(Er)jiG−1
jj =
φ1φ2 · · · φi
φ1φ2 · · · φj
(Er)ji
(18)
(19)
for 0 ≤ i, j ≤ d. Equating (18) and the right side of (19), and solving for (Er)ji, we routinely
obtain the result.
✷
Example 2.10. Referring to Proposition 2.9, assume d = 2. With respect to a Φ-split basis,
the matrices representing E0, E1, E2 are
0
0
φ1φ2
0
0
φ2
(θ0−θ2)(θ0−θ1)
θ0−θ1
0
0
1
,
0
φ1
θ1−θ2
φ1φ2
0
1
φ2
(θ1−θ0)(θ1−θ2)
θ1−θ0
0
0
0
,
1
φ1
0 0
0 0
(θ2−θ0)(θ2−θ1) 0 0
θ2−θ1
φ1φ2
respectively. Moreover the matrices representing E∗
0, E∗
1 , E∗
2 are
1
1
1
0
0
θ∗
0 −θ∗
0
0
respectively.
(θ∗
0 −θ∗
0 −θ∗
2 )
1
1)(θ∗
0
0
,
0
0
0
1
0
1 −θ∗
θ∗
1
0
(θ∗
1 −θ∗
0)(θ∗
1 −θ∗
2 )
1
1
2
1 −θ∗
θ∗
0
,
0 0
0 0
0 0
(θ∗
2 −θ∗
1 )(θ∗
2 −θ∗
0 )
1
1
1
2 −θ∗
θ∗
1
,
,
We now give some characterizations of the split sequence.
Lemma 2.11. Let (A; {Ei}d
{θi}d
i=0, dual eigenvalue sequence {θ∗
i=0; A∗; {E∗
i }d
i }d
i=0, and split sequence {φi}d
i=0) denote a TH system with eigenvalue sequence
i=1. Then
E∗
0 ηi(A)E∗
0 =
(θ∗
0 − θ∗
1)(θ∗
2) · · · (θ∗
0 − θ∗
i )
φ1φ2 · · · φi
0 − θ∗
E∗
0
(0 ≤ i ≤ d).
(20)
9
Proof: Let Φ denote the TH system in question and assume V is the underlying vector space.
Let {Ui}d
0V .
By this and (4), (5) we obtain
i=0 denote the Φ-split decomposition of V . Setting i = 0 in (3) we find U0 = E∗
(A∗ − θ∗
1I)(A∗ − θ∗
2I) · · · (A∗ − θ∗
i I)ηi(A) = φ1φ2 · · · φiI
(21)
0 V . To obtain (20), multiply both sides of (21) on the left by E∗
0 and use E∗
0A∗ = θ∗
0E∗
0 .
on E∗
✷
Corollary 2.12. Let (A; {Ei}d
{θi}d
i=0, dual eigenvalue sequence {θ∗
i=0; A∗; {E∗
i }d
i }d
i=0) denote a TH system with eigenvalue sequence
i=0, and split sequence {φi}d
i=1. Then for 0 ≤ i ≤ d,
φ1φ2 · · · φi = (θ∗
0 − θ∗
1)(θ∗
0 − θ∗
2) · · · (θ∗
0 − θ∗
i )trace(ηi(A)E∗
0 ).
(22)
Moreover ηi(A)E∗
0 has nonzero trace.
Proof: To obtain (22), in (20) take the trace of each side and simplify the result using the
fact that trace(E∗
0 ) = trace(ηi(A)E∗
0). This
gives (22). The last assertion follows since φi 6= 0 for 1 ≤ i ≤ d.
✷
0) = 1 and trace(E∗
0 ) = trace(ηi(A)E∗
0ηi(A)E∗
0E∗
Corollary 2.13. Let (A; {Ei}d
{θi}d
i=0, dual eigenvalue sequence {θ∗
i=0; A∗; {E∗
i }d
i }d
i=0) denote a TH system with eigenvalue sequence
i=0, and split sequence {φi}d
i=1. Then
φi = (θ∗
0 − θ∗
i )trace(ηi(A)E∗
0)/trace(ηi−1(A)E∗
0)
Proof: Routine by Corollary 2.12.
(1 ≤ i ≤ d).
(23)
✷
In Section 7 we give some more characterizations of the split sequence.
3
Isomorphisms for TH pairs and TH systems
In this section we discuss the notion of isomorphism for TH pairs and TH systems.
Lemma 3.1. For X ∈ Matd+1(K) the following (i) -- (iii) are equivalent.
(i) X is diagonal.
(ii) DX = XD for all diagonal D ∈ Matd+1(K).
(iii) There exists a diagonal D ∈ Matd+1(K) that has mutually distinct diagonal entries and
DX = XD.
Proof: (i) ⇒ (ii) Clear.
(ii) ⇒ (iii) Clear.
(iii) ⇒ (i) For 0 ≤ i, j ≤ d with i 6= j, we show Xij = 0. Comparing the (i, j)-entry of DX
and XD, we find DiiXij = XijDjj. By assumption Dii 6= Djj, so Xij = 0.
✷
Let A, A∗ denote a TH pair on V . In general, End(V ) may not be generated by A, A∗.
Moreover there may exist a subspace W of V such that AW ⊆ W, A∗W ⊆ W, W 6= 0, W 6= V .
However we do have the following result.
10
Lemma 3.2. Let A, A∗ denote a TH pair on V . Let ∆ denote an element of End(V ) such
that ∆A = A∆ and ∆A∗ = A∗∆. Then ∆ ∈ KI.
Proof: Pick a basis for V from Definition 1.1(i). For notational convenience, identify each
element of End(V ) with the matrix that represents it with respect to this basis. Thus the
matrix A is Hessenberg and the matrix A∗ is diagonal. Moreover the diagonal entries of A∗
are mutually distinct by Lemma 2.1. Applying Lemma 3.1 with D = A∗ and X = ∆, we
find ∆ is diagonal. For 1 ≤ i ≤ d, comparing the (i, i − 1)-entry of ∆A and A∆, we find
∆iiAi,i−1 = Ai,i−1∆i−1,i−1. Observe that Ai,i−1 6= 0 since A is Hessenberg, so ∆ii = ∆i−1,i−1.
Therefore ∆ii is independent of i for 0 ≤ i ≤ d. Consequently ∆ ∈ KI.
✷
For the rest of this section, let W denote a vector space over K with dimension d + 1.
Let Γ : V → W denote a K-vector space isomorphism. Then there exists a unique K-
algebra isomorphism γ : End(V ) → End(W ) such that Sγ = ΓSΓ−1 for all S ∈ End(V ).
Conversely let γ : End(V ) → End(W ) denote a K-algebra isomorphism. By the Skolem-
Noether theorem [2, Corollary 9.122] there exists a K-vector space isomorphism Γ : V → W
such that Sγ = ΓSΓ−1 for all S ∈ End(V ). Moreover Γ is unique up to multiplication by a
nonzero scalar in K.
Definition 3.3. Let A, A∗ denote a TH pair on V and let B, B∗ denote a TH pair on W .
By an isomorphism of TH pairs from A, A∗ to B, B∗ we mean a K-algebra isomorphism
γ : End(V ) → End(W ) such that B = Aγ and B∗ = A∗γ. We say that the TH pairs A, A∗
and B, B∗ are isomorphic whenever there exists an isomorphism of TH pairs from A, A∗ to
B, B∗.
Lemma 3.4. Let A, A∗ and B, B∗ denote isomorphic TH pairs over K. Then the isomor-
phism of TH pairs from A, A∗ to B, B∗ is unique.
Proof: Let γ and γ′ denote isomorphisms of TH pairs from A, A∗ to B, B∗. We show that
γ = γ′. By the comments above Definition 3.3, there exists a K-vector space isomorphism
Γ : V → W (resp. Γ′ : V → W ) such that Sγ = ΓSΓ−1 (resp. Sγ′ = Γ′SΓ′−1) for all
S ∈ End(V ). Consider the composition ∆ = Γ−1Γ′. Observe that ∆ is an invertible element
of End(V ). By construction, ∆A = A∆ and ∆A∗ = A∗∆. Therefore ∆ ∈ KI by Lemma 3.2.
By these comments, there exists 0 6= α ∈ K such that ∆ = αI. Hence Γ′ = αΓ, so γ = γ′. ✷
Definition 3.5. Let Φ = (A; {Ei}d
(B; {Fi}d
from Φ to Ψ we mean a K-algebra isomorphism γ : End(V ) → End(W ) such that
i=0) denote a TH system on V and let Ψ =
i=0) denote a TH system on W . By an isomorphism of TH systems
i=0; A∗; {E∗
i=0; B∗; {F ∗
i }d
i }d
B = Aγ,
B∗ = A∗γ,
Fi = Eγ
i ,
i = E∗γ
F ∗
i
(0 ≤ i ≤ d).
We say that the TH systems Φ and Ψ are isomorphic whenever there exists an isomorphism
of TH systems from Φ to Ψ.
Lemma 3.6. Let Φ and Ψ denote isomorphic TH systems over K. Then the isomorphism
of TH systems from Φ to Ψ is unique.
Proof: Similar to the proof of Lemma 3.4.
✷
We give another interpretation of isomorphism for TH pairs and TH systems.
11
Lemma 3.7. Let A, A∗ denote a TH pair on V and let B, B∗ denote a TH pair on W . Then
the following (i), (ii) are equivalent.
(i) The TH pairs A, A∗ and B, B∗ are isomorphic.
(ii) There exists a K-vector space isomorphism Γ : V → W such that BΓ = ΓA and
B∗Γ = ΓA∗.
Moreover assume (i), (ii) hold. Then Γ is unique up to a multiplication by a nonzero scalar
in K.
Lemma 3.8. Let Φ = (A; {Ei}d
(B; {Fi}d
lent.
i=0; B∗; {F ∗
i }d
i=0) denote a TH system on V and let Ψ =
i=0) denote a TH system on W . Then the following (i), (ii) are equiva-
i=0; A∗; {E∗
i }d
(i) The TH systems Φ and Ψ are isomorphic.
(ii) There exists a K-vector space isomorphism Γ : V → W such that
BΓ = ΓA,
B∗Γ = ΓA∗,
FiΓ = ΓEi,
F ∗
i Γ = ΓE∗
i
(0 ≤ i ≤ d).
Moreover assume (i), (ii) hold. Then Γ is unique up to a multiplication by a nonzero scalar
in K.
4 The classification of TH systems
In [1] we classified the TH systems up to isomorphism. We recall the result in this section.
Definition 4.1. Let Φ denote a TH system. By the parameter array of Φ we mean the
sequence ({θi}d
i=0) is the eigenvalue (resp.
dual eigenvalue) sequence of Φ and {φi}d
i=1 is the split sequence of Φ.
i=1), where {θi}d
i=0 (resp. {θ∗
i=0, {φi}d
i=0, {θ∗
i }d
i }d
Theorem 4.2. [1, Theorem 6.3] Let
({θi}d
i=0, {θ∗
i }d
i=0, {φi}d
i=1)
(24)
denote a sequence of scalars taken from K. Then there exists a TH system Φ over K with
parameter array (24) if and only if (i) -- (iii) hold below.
(i) θi 6= θj
if
i 6= j
(ii) θ∗
i 6= θ∗
j
if
i 6= j
(iii) φi 6= 0
(0 ≤ i, j ≤ d).
(0 ≤ i, j ≤ d).
(1 ≤ i ≤ d).
Moreover assume (i) -- (iii) hold. Then Φ is unique up to isomorphism of TH systems.
Definition 4.3. By a parameter array over K of diameter d we mean a sequence of scalars
({θi}d
i=1) taken from K that satisfies conditions (i) -- (iii) of Theorem 4.2.
i=0, {φi}d
i=0, {θ∗
i }d
12
Corollary 4.4. The map which sends a given TH system to its parameter array induces a
bijection from the set of isomorphism classes of TH systems over K of diameter d, to the set
of parameter arrays over K of diameter d.
Proof: Immediate from Theorem 4.2.
✷
To illuminate the bijection in Corollary 4.4 we now describe its inverse in concrete terms.
Let π denote the bijection in Corollary 4.4.
i }d
i=0, {θ∗
i=0, {φi}d
Proposition 4.5. Let ({θi}d
i=1) denote a parameter array over K of diam-
eter d. Let A (resp. A∗) denote the matrix on the left (resp. right) in (8). Observe that
A (resp. A∗) is multiplicity-free with eigenvalues {θi}d
i=0). For 0 ≤ i ≤ d
i ) denote the primitive idempotent of A (resp. A∗) that corresponds to θi
let Ei (resp. E∗
i=0) is a TH system over K. Moreover π−1 sends
(resp. θ∗
({θi}d
i=1) to the isomorphism class of Φ.
i ). Then Φ = (A; {Ei}d
i=0 (resp. {θ∗
i=0; A∗; {E∗
i=0, {φi}d
i=0, {θ∗
i }d
i }d
i }d
Proof: This is proven as part of the proof of [1, Theorem 6.3].
✷
5 The affine transformations of a TH system
A given TH system can be modified in several ways to get a new TH system. In this section
we describe one way. In the next section we describe another way.
Lemma 5.1. Let Φ = (A; {Ei}d
denote scalars in K with α, α∗ nonzero. Then the sequence
i=0; A∗; {E∗
i }d
i=0) denote a TH system on V . Let α, β, α∗, β∗
(αA + βI; {Ei}d
i=0; α∗A∗ + β∗I; {E∗
i }d
i=0)
is a TH system on V .
Proof: Routine.
(25)
✷
Definition 5.2. Referring to Lemma 5.1, we call the TH system (25) the affine transforma-
tion of Φ associated with α, β, α∗, β∗.
Definition 5.3. Let Φ and Φ′ denote TH systems over K. We say that Φ and Φ′ are affine
isomorphic whenever Φ is isomorphic to an affine transformation of Φ′. Observe that affine
isomorphism is an equivalence relation.
Lemma 5.4. With reference to Lemma 5.1, let ({θi}d
eter array of Φ. Then the parameter array of the TH system (25) is ({αθi + β}d
β∗}d
i=1) denote the param-
i +
i=0, {α∗θ∗
i=0, {φi}d
i=0, {θ∗
i }d
i=0, {αα∗φi}d
i=1).
Proof: Let Φ′ denote the TH system (25). By Definition 2.5, for 0 ≤ i ≤ d the scalar θi is the
eigenvalue of A associated with Ei, so αθi + β is the eigenvalue of αA + βI associated with
Ei. Thus {αθi + β}d
i=0 is the dual
eigenvalue sequence of Φ′. In (23), if we replace A by αA + βI and replace θj (resp. θ∗
j ) by
j + β∗) for 0 ≤ j ≤ d, then the left-hand side becomes αα∗φi. Therefore
αθj + β (resp. α∗θ∗
{αα∗φi}d
✷
i=0 is the eigenvalue sequence of Φ′. Similarly {α∗θ∗
i=1 is the split sequence of Φ′.
i + β∗}d
13
6 The relatives of a TH system
Let Φ denote a TH system. In the previous section we modified Φ in a certain way to get
another TH system. In this section we modify Φ in a different way to obtain two more TH
systems. These TH systems are called Φ∗ and Φ. We start with Φ∗.
Definition 6.1. Let Φ = (A; {Ei}d
(A∗; {E∗
i=0; A; {Ei}d
i }d
i=0) is a TH system on V , which we denote by Φ∗.
i=0; A∗; {E∗
i }d
i=0) denote a TH system on V . Observe that
Lemma 6.2. [1, Lemma 6.4] Let Φ denote a TH system with parameter array
({θi}d
i=1). Then the TH system Φ∗ has parameter array ({θ∗
i=0, {φi}d
i=0, {θ∗
i }d
i }d
i=0, {θi}d
i=0, {φd−i+1}d
i=1).
We now consider Φ. For the rest of this section, let W denote a vector space over K
with dimension d + 1. For K-algebras A and A′, by a K-algebra anti-isomorphism from
A to A′ we mean a K-vector space isomorphism † : A → A′ such that (RS)† = S†R†
for all R, S ∈ A. By a K-algebra anti-automorphism of A we mean a K-algebra anti-
isomorphism from A to A. The anti-automorphisms of Matd+1(K) are described as follows.
Let R denote an invertible element of Matd+1(K). Then there exists a unique K-algebra anti-
automorphism † of Matd+1(K) such that S† = RStR−1 for all S ∈ Matd+1(K). Conversely,
let † denote a K-algebra anti-automorphism of Matd+1(K). By the Skolem-Noether theorem
[2, Corollary 9.122], there exists an invertible R ∈ Matd+1(K) such that S† = RStR−1 for all
S ∈ Matd+1(K). Moreover R is unique up to a multiplication by a nonzero scalar in K.
Define Z ∈ Matd+1(K) such that Zij = δi+j,d for 0 ≤ i, j ≤ d. Observe that Z −1 = Z.
Define ς to be the K-algebra anti-automorphism of Matd+1(K) such that Sς = ZStZ for
all S ∈ Matd+1(K). For S ∈ Matd+1(K), Sς is obtained from S by reflecting about the
diagonal connecting the top right corner of S and the bottom left corner of S.
In other
words, (Sς)ij = Sd−j,d−i for 0 ≤ i, j ≤ d. For example,
S =
1 2 3
4 5 6
7 8 9
,
Sς =
9 6 3
8 5 2
7 4 1
.
Observe that (Sς)ς = S for all S ∈ Matd+1(K). Note that if H ∈ Matd+1(K) is Hessenberg
then H ς is Hessenberg.
Lemma 6.3. Let A denote a multiplicity-free element of End(V ) with eigenvalues {θi}d
i=0.
For 0 ≤ i ≤ d, let Ei ∈ End(V ) denote the primitive idempotent of A corresponding to θi.
For any anti-isomorphism † : End(V ) → End(W ), the following (i), (ii) hold.
(i) A† is a multiplicity-free element of End(W ) with eigenvalues {θi}d
i=0.
(ii) For 0 ≤ i ≤ d, E†
i is the primitive idempotent of A† corresponding to θi.
Proof: (i) For f ∈ K[λ] we have f (A) = 0 if and only if f (A†) = 0. Therefore A and A†
i=0(λ − θi) so the
i=0 are mutually distinct,
i=0. Recall dim W = d + 1 so A† is multiplicity-free.
✷
have the same minimal polynomial. The minimal polynomial of A is Qd
minimal polynomial of A† is Qd
A† is diagonalizable with eigenvalues {θi}d
(ii) Apply † to (1).
i=0(λ − θi). By this and since {θi}d
14
Proposition 6.4. Let (A; {Ei}d
anti-isomorphism from End(V ) to End(W ). Then (A†; {E†
system on W .
i=0; A∗; {E∗
i=0) denote a TH system on V . Let † denote an
i=0) is a TH
i=0; A∗†; {E∗†
d−i}d
d−i}d
i }d
Proof: Define Ψ = (A†; {E†
In order to show that Ψ is a TH sys-
tem on W , we show that Ψ satisfies conditions (i) -- (v) of Definition 2.2. By Lemma
6.3, Ψ satisfies conditions (i) -- (iii). We now show that Ψ satisfies condition (iv). Since
(A; {Ei}d
i=0) is a TH system, we have
i=0; A∗†; {E∗†
i=0; A∗; {E∗
d−i}d
d−i}d
i=0).
(0 ≤ i, j ≤ d).
(26)
(0 ≤ i, j ≤ d).
(27)
(0 ≤ i, j ≤ d).
i }d
EiA∗Ej = (cid:26) 0, if i − j > 1
6= 0, if i − j = 1
Applying † to (26), we find
E†
j A∗†E†
i = (cid:26) 0, if i − j > 1
6= 0, if i − j = 1
Relabelling the indices in (27), we find
E†
d−iA∗†E†
d−j = (cid:26) 0, if i − j > 1
6= 0, if i − j = 1
Therefore Ψ satisfies condition (iv). Similarly Ψ satisfies condition (v). Therefore Ψ is a TH
system on W .
✷
Definition 6.5. Let Φ = (A; {Ei}d
(B; {Fi}d
from Φ to Ψ we mean a K-algebra anti-isomorphism † : End(V ) → End(W ) such that
i=0) denote a TH system on V and let Ψ =
i=0) denote a TH system on W . By an anti-isomorphism of TH systems
i=0; B∗; {F ∗
i=0; A∗; {E∗
i }d
i }d
B = A†,
B∗ = A∗†,
Fi = E†
d−i,
i = E∗†
F ∗
d−i
(0 ≤ i ≤ d).
Observe that if † is an anti-isomorphism of TH systems from Φ to Ψ, then †−1 is an anti-
isomorphism of TH systems from Ψ to Φ. We say that the TH systems Φ and Ψ are
anti-isomorphic whenever there exists an anti-isomorphism of TH systems from Φ to Ψ.
Lemma 6.6. Let Φ denote a TH system over K. Then there exists a TH system Ψ over
K such that Φ and Ψ are anti-isomorphic. Moreover Ψ is unique up to isomorphism of TH
systems.
i=0; A∗; {E∗
Proof: We first show that Ψ exists. Write Φ = (A; {Ei}d
i=0) and assume V is
the vector space underlying Φ. By elementary linear algebra, there exists a K-algebra anti-
automorphism † of End(V ). Define Ψ = (A†; {E†
i=0). By Proposition 6.4,
Ψ is a TH system on V . By Definition 6.5, Φ and Ψ are anti-isomorphic. We have shown
that Ψ exists. Next we show that Ψ is unique. Suppose that Ψ′ is a TH system on W
such that Φ and Ψ′ are anti-isomorphic. We show that Ψ and Ψ′ are isomorphic. Let †′
denote an anti-isomorphism of TH systems from Φ to Ψ′. Then the composition †′†−1 is an
isomorphism of TH systems from Ψ to Ψ′. Therefore Ψ and Ψ′ are isomorphic.
✷
i=0; A∗†; {E∗†
d−i}d
d−i}d
i }d
15
Lemma 6.7. Let Φ and Ψ denote anti-isomorphic TH systems over K. Then the anti-
isomorphism of TH systems from Φ to Ψ is unique.
Proof: Let † and †′ denote anti-isomorphisms of TH systems from Φ to Ψ. We show that
† = †′. Observe that the composition †−1†′ is an isomorphism of TH systems from Φ to Φ.
The map †−1†′ is the identity by Lemma 3.6, so † = †′.
✷
Proposition 6.8. Let Φ denote a TH system over K with parameter array
({θi}d
i }d
are equivalent.
i=1). Let Ψ denote a TH system over K. Then the following (i), (ii)
i=0, {φi}d
i=0, {θ∗
(i) Φ and Ψ are anti-isomorphic.
(ii) The parameter array of Ψ is ({θd−i}d
i=0, {θ∗
i }d
i=0, {φd−i+1}d
d−i}d
i=0) and Ψ = (B; {Fi}d
i=1).
Proof: (ii)⇒(i) Write Φ = (A; {Ei}d
i=0). As-
sume V (resp. W ) is the vector space underlying Φ (resp. Ψ). For notational convenience,
fix a Φ-split basis for V (resp. Ψ-split basis for W ) and identify each element of End(V )
(resp. End(W )) with the matrix in Matd+1(K) that represents it with respect to this basis.
By Proposition 2.8,
i=0; B∗; {F ∗
i=0; A∗; {E∗
i }d
A =
θd
φ1 θd−1
φ2
0
θd−2
·
·
·
0
·
φd θ0
,
Moreover
B =
θ0
φd
0
θ1
φd−1 θ2
·
0
·
·
·
φ1 θd
,
θ∗
d
0
A∗ =
B∗ =
θ∗
0
1
θ∗
1
1
θ∗
2
0
1
θ∗
d−1
·
·
·
·
0
1
θ∗
d
.
1
θ∗
d−2
·
·
·
·
0
1
θ∗
0
.
Recall the K-algebra anti-automorphism ς of Matd+1(K) from above Lemma 6.3. Observe
that B = Aς and B∗ = A∗ς. By this and (1) we find Fi = Eς
d−i for
0 ≤ i ≤ d. Therefore ς is an anti-isomorphism of TH systems from Φ to Ψ, so Φ and Ψ are
anti-isomorphic.
(i)⇒(ii) Routine by Theorem 4.2, Lemma 6.6, and the previous part.
d−i and F ∗
i = E∗ς
✷
We now discuss the notion of anti-isomorphism for TH pairs.
Definition 6.9. Let A, A∗ denote a TH pair on V and let B, B∗ denote a TH pair on
W . By an anti-isomorphism of TH pairs from A, A∗ to B, B∗ we mean a K-algebra anti-
isomorphism † : End(V ) → End(W ) such that B = A† and B∗ = A∗†. Observe that if † is an
anti-isomorphism of TH pairs from A, A∗ to B, B∗, then †−1 is an anti-isomorphism of TH
pairs from B, B∗ to A, A∗. We say that the TH pairs A, A∗ and B, B∗ are anti-isomorphic
whenever there exists an anti-isomorphism of TH pairs from A, A∗ to B, B∗.
16
Lemma 6.10. Let A, A∗ denote a TH pair over K. Then there exists a TH pair B, B∗
over K such that A, A∗ and B, B∗ are anti-isomorphic. Moreover B, B∗ is unique up to
isomorphism of TH pairs.
Proof: Similar to the proof of Lemma 6.6.
✷
Lemma 6.11. Let A, A∗ and B, B∗ denote anti-isomorphic TH pairs over K. Then the
anti-isomorphism of TH pairs from A, A∗ to B, B∗ is unique.
Proof: Similar to the proof of Lemma 6.7.
✷
We recall some more terms and facts from elementary linear algebra. A map h , i :
V × W → K is called a bilinear form whenever the following conditions hold for all v, v′ ∈ V ,
w, w′ ∈ W , and α ∈ K: (i) hv + v′, wi = hv, wi + hv′, wi; (ii) hαv, wi = αhv, wi; (iii)
hv, w + w′i = hv, wi + hv, w′i; (iv) hv, αwi = αhv, wi. We observe that a scalar multiple of a
bilinear form is a bilinear form.
Let h , i : V × W → K denote a bilinear form. Then the following are equivalent: (i)
there exists a nonzero v ∈ V such that hv, wi = 0 for all w ∈ W ; (ii) there exists a nonzero
w ∈ W such that hv, wi = 0 for all v ∈ V . The form h , i is said to be degenerate whenever
(i), (ii) hold and nondegenerate otherwise.
Bilinear forms are related to anti-isomorphisms as follows. Let h , i : V × W → K denote
a nondegenerate bilinear form. Then there exists a unique anti-isomorphism † : End(V ) →
End(W ) such that hSv, wi = hv, S†wi for all v ∈ V , w ∈ W , and S ∈ End(V ). Conversely,
given an anti-isomorphism † : End(V ) → End(W ) there exists a nonzero bilinear form
h , i : V × W → K such that hSv, wi = hv, S†wi for all v ∈ V , w ∈ W , and S ∈ End(V ).
This bilinear form is nondegenerate, and uniquely determined by † up to multiplication by
a nonzero scalar in K. We say that the form h , i is associated with †.
Define V to be the dual space of V , consisting of all K-linear transformations from V to
K. By elementary linear algebra, V is a vector space over K and dim V = dim V . Define
a bilinear form h , i : V × V → K such that hv, f i = f (v) for all v ∈ V and f ∈ V .
The form h , i is nondegenerate. We call h , i the canonical bilinear form between V and
V . Let σ : End(V ) → End( V ) denote the anti-isomorphism associated with h , i. Thus
hSv, f i = hv, Sσf i for all v ∈ V , f ∈ V , and S ∈ End(V ). We call σ the canonical anti-
isomorphism from End(V ) to End( V ).
Definition 6.12. Let Φ = (A; {Ei}d
i=0) denote a TH system on V with pa-
rameter array ({θi}d
i=0, {φi}d
i=0), where
σ : End(V ) → End( V ) is the canonical anti-isomorphism. By Proposition 6.4, Φ is a TH
system on V . By Definition 6.5, Φ and Φ are anti-isomorphic. By Proposition 6.8, Φ has
parameter array ({θd−i}d
i=0; A∗; {E∗
i=1). Define Φ = (Aσ; {Eσ
i=0; A∗σ; {E∗σ
i=0, {θ∗
i=0, {φd−i+1}d
i=0, {θ∗
d−i}d
d−i}d
i }d
d−i}d
i=1).
i }d
7 The Z2 × Z2 action
Let Φ = (A; {Ei}d
each of the following is a TH system:
i=0; A∗; {E∗
i }d
i=0) denote a TH system. We saw in the previous section that
Φ∗ = (A∗; {E∗
Φ = (Aσ; {Eσ
i }d
d−i}d
i=0; A; {Ei}d
i=0),
i=0; A∗σ; {E∗σ
d−i}d
i=0).
17
Viewing ∗, ∼ as permutations on the set of all TH systems,
∗2 = ∼2 = 1,
∗ ∼ = ∼ ∗.
(28)
The group generated by symbols ∗, ∼ subject to the relations (28) is the group Z2 × Z2.
Thus ∗, ∼ induce an action of Z2 × Z2 on the set of all TH systems. Two TH systems will
be called relatives whenever they are in the same orbit of this Z2 × Z2 action. The relatives
of Φ are as follows:
name
Φ
Φ∗
Φ
Φ∗
relative
i=0; A∗; {E∗
(A; {Ei}d
i }d
(A∗; {E∗
i }d
i=0; A; {Ei}d
i=0; A∗σ; {E∗σ
d−i}d
i=0; Aσ; {Eσ
d−i}d
(Aσ; {Eσ
(A∗σ; {E∗σ
i=0)
i=0)
d−i}d
d−i}d
i=0)
i=0)
Corollary 7.1. Let Φ denote a TH system with parameter array ({θi}d
Then the parameter arrays of its relatives are as follows:
i=0, {θ∗
i }d
i=0, {φi}d
i=1).
name
Φ
Φ∗
Φ
Φ∗
parameter array
i=0, {θ∗
i=0, {θi}d
i=0, {θ∗
i }d
i=1)
i=0, {φd−i+1}d
i=0, {φi}d
i=1)
i=0, {φd−i+1}d
i=1)
i=0, {φi}d
i=0, {θd−i}d
d−i}d
i=1)
({θi}d
i }d
({θd−i}d
({θ∗
({θ∗
d−i}d
Proof: Immediate from Lemma 6.2 and Proposition 6.8.
✷
We will use the following notational convention.
Definition 7.2. Let Φ denote a TH system. For g ∈ Z2 × Z2 and for an object f associated
with Φ, let f g denote the corresponding object associated with Φg.
We end this section by giving some more characterizations of the split sequence, as promised
at the end of Section 2.
Lemma 7.3. Let (A; {Ei}d
({θi}d
i=0, {φi}d
i=0, {θ∗
i }d
i=0; A∗; {E∗
i }d
i=0) denote a TH system with parameter array
i=1). Then the following (i) -- (iii) hold for 0 ≤ i ≤ d.
(i) E0η∗
i (A∗)E0 =
φdφd−1 · · · φd−i+1
(θ0 − θ1)(θ0 − θ2) · · · (θ0 − θi)
E0.
(ii) E∗
dτi(A)E∗
d =
φdφd−1 · · · φd−i+1
(θ∗
d − θ∗
d−1)(θ∗
d − θ∗
d−2) · · · (θ∗
d − θ∗
d−i)
E∗
d.
(iii) Edτ ∗
i (A∗)Ed =
φ1φ2 · · · φi
(θd − θd−1)(θd − θd−2) · · · (θd − θd−i)
Ed.
Proof: Let Φ denote the TH system in question.
(i) Apply Lemma 2.11 to Φ∗.
(ii) Apply Lemma 2.11 to Φ and then apply σ−1 to each side of the resulting equations.
(iii) Apply (ii) to Φ∗.
✷
18
Corollary 7.4. Let (A; {Ei}d
({θi}d
i=0, {φi}d
i=0, {θ∗
i }d
i=0; A∗; {E∗
i }d
i=0) denote a TH system with parameter array
i=1). Then the following (i) -- (iii) hold for 0 ≤ i ≤ d.
(i) φdφd−1 · · · φd−i+1 = (θ0 − θ1)(θ0 − θ2) · · · (θ0 − θi)trace(η∗
i (A∗)E0).
(ii) φdφd−1 · · · φd−i+1 = (θ∗
d − θ∗
d−1)(θ∗
d − θ∗
d−2) · · · (θ∗
d − θ∗
d−i)trace(τi(A)E∗
d).
(iii) φ1φ2 · · · φi = (θd − θd−1)(θd − θd−2) · · · (θd − θd−i)trace(τ ∗
i (A∗)Ed).
Moreover each of η∗
i (A∗)E0, τi(A)E∗
d, τ ∗
i (A∗)Ed has nonzero trace.
Proof: Let Φ denote the TH system in question.
(i) Apply Corollary 2.12 to Φ∗.
(ii) In the equation of Lemma 7.3(ii), take the trace of each side and simplify the result using
the fact that trace(E∗
(iii) Apply (ii) to Φ∗.
The last assertion follows since φi 6= 0 for 1 ≤ i ≤ d.
d) = 1 and trace(E∗
d) = trace(τi(A)E∗
d) = trace(τi(A)E∗
dτi(A)E∗
dE∗
d).
✷
Corollary 7.5. Let (A; {Ei}d
({θi}d
i=0, {φi}d
i=0, {θ∗
i }d
i=0; A∗; {E∗
i }d
i=0) denote a TH system with parameter array
i=1). Then the following (i) -- (iii) hold for 1 ≤ i ≤ d.
(i) φi = (θ0 − θd−i+1)trace(η∗
d−i+1(A∗)E0)/trace(η∗
d−i(A∗)E0).
(ii) φi = (θ∗
d − θ∗
i−1)trace(τd−i+1(A)E∗
d)/trace(τd−i(A)E∗
d).
(iii) φi = (θd − θd−i)trace(τ ∗
i (A∗)Ed)/trace(τ ∗
i−1(A∗)Ed).
Proof: Routine by Corollary 7.4.
✷
8 The scalars {ℓi}d
i=0
Let Φ denote a TH system. In this section we associate with Φ a sequence of scalars {ℓi}d
that will help us describe Φ.
i=0
Definition 8.1. Let Φ denote a TH system with dual eigenvalue sequence {θ∗
0 ≤ i ≤ d, define
i }d
i=0. For
ℓi =
=
η∗
d(θ∗
0)
d−i(θ∗
i )η∗
i (θ∗
τ ∗
i )
(θ∗
i − θ∗
0)(θ∗
i − θ∗
1) · · · (θ∗
(θ∗
0 − θ∗
1)(θ∗
i − θ∗
0 − θ∗
i−1)(θ∗
2) · · · (θ∗
i − θ∗
0 − θ∗
d)
i+1) · · · (θ∗
i − θ∗
d−1)(θ∗
i − θ∗
d)
.
Observe that ℓ0 = 1.
19
Lemma 8.2. Let Φ denote a TH system with eigenvalue sequence {θi}d
sequence {θ∗
i=0. Then for 0 ≤ i ≤ d,
i }d
i=0 and dual eigenvalue
ℓ∗
i =
ηd(θ0)
τi(θi)ηd−i(θi)
=
ℓi =
=
ℓ∗
i =
=
(θ0 − θ1)(θ0 − θ2) · · · (θ0 − θd)
(θi − θ0)(θi − θ1) · · · (θi − θi−1)(θi − θi+1) · · · (θi − θd−1)(θi − θd)
,
d (θ∗
τ ∗
d)
d−i(θ∗
d−i)τ ∗
d−i)
η∗
i (θ∗
(θ∗
d−i − θ∗
d)(θ∗
d−i − θ∗
(θ∗
d − θ∗
d−1) · · · (θ∗
d−1)(θ∗
d−i − θ∗
d − θ∗
d−i+1)(θ∗
d−2) · · · (θ∗
d−i − θ∗
d − θ∗
0)
d−i−1) · · · (θ∗
d−i − θ∗
1)(θ∗
d−i − θ∗
0)
τd(θd)
ηi(θd−i)τd−i(θd−i)
(θd−i − θd)(θd−i − θd−1) · · · (θd−i − θd−i+1)(θd−i − θd−i−1) · · · (θd−i − θ1)(θd−i − θ0)
(θd − θd−1)(θd − θd−2) · · · (θd − θ0)
Moreover ℓi =
d (θ∗
τ ∗
d)
η∗
d(θ∗
0)
ℓd−i and ℓ∗
i =
τd(θd)
ηd(θ0)
ℓ∗
d−i.
Proof: Combine Corollary 7.1 and Definition 8.1.
We give one significance of the sequence {ℓi}d
i=0.
,
.
✷
Lemma 8.3. Let (A; {Ei}d
for 0 ≤ i ≤ d.
i=0; A∗; {E∗
i }d
i=0) denote a TH system. Then EdE∗
i E0 = ℓiEdE∗
0E0
Proof: Let Φ denote the TH system in question and assume V is the underlying vector space.
For notational convenience, fix a Φ-split basis for V and identify each element of End(V )
with the matrix in Matd+1(K) that represents it with respect to this basis. We show that
Ed(E∗
0)E0 = 0. By (9) the entries of all but the first column of Ed are zero and the
entries of all but the last row of E0 are zero. Therefore for 0 ≤ m, n ≤ d, the (m, n)-entry of
Ed(E∗
i − ℓiE∗
i − ℓiE∗
0)E0 is
(Ed)m0(E∗
i − ℓiE∗
0)0d(E0)dn.
(29)
By (10) the middle factor in (29) is 0, so (29) is 0. Therefore Ed(E∗
result follows.
i − ℓiE∗
0 )E0 = 0 and the
✷
Corollary 8.4. Let (A; {Ei}d
(iii) hold for 0 ≤ i ≤ d.
i=0; A∗; {E∗
i }d
i=0) denote a TH system. Then the following (i) --
(i) E∗
dEiE∗
i E∗
0 = ℓ∗
dE0E∗
0.
i E0 = ℓd−iEdE∗
dE0.
(ii) EdE∗
(iii) E∗
dEiE∗
0 = ℓ∗
d−iE∗
dEdE∗
0.
20
Proof: Let Φ denote the TH system in question.
(i) Apply Lemma 8.3 to Φ∗.
(ii) Apply Lemma 8.3 to Φ and then apply σ−1 to each side of the resulting equations.
(iii) Apply (ii) to Φ∗.
✷
Definition 8.5. Let Φ denote a TH system of diameter d. We associate with Φ a diagonal
matrix L ∈ Matd+1(K) with (i, i)-entry ℓi for 0 ≤ i ≤ d.
9 The scalar ν
Let Φ denote a TH system. In this section we associate with Φ a scalar ν that will help us
describe Φ.
Definition 9.1. Let (A; {Ei}d
trace(E0E∗
0) is nonzero. Let ν denote the reciprocal of trace(E0E∗
0 ).
i=0; A∗; {E∗
i }d
i=0) denote a TH system. By [1, Lemma 7.5],
We give one significance of the scalar ν.
Lemma 9.2. [1, Lemma 7.4] Let (A; {Ei}d
i=0; A∗; {E∗
i }d
i=0) denote a TH system. Then both
νE0E∗
0 E0 = E0,
νE∗
0 E0E∗
0 = E∗
0 .
Lemma 9.3. Let Φ denote a TH system with parameter array ({θi}d
let ν denote the scalar from Definition 9.1. Then
i=0, {θ∗
i }d
i=0, {φi}d
i=1) and
ν =
ν =
(θ0 − θ1)(θ0 − θ2) · · · (θ0 − θd)(θ∗
0 − θ∗
1)(θ∗
0 − θ∗
2) · · · (θ∗
0 − θ∗
d)
φ1φ2 · · · φd
,
(30)
(θd − θd−1)(θd − θd−2) · · · (θd − θ0)(θ∗
d − θ∗
d−1)(θ∗
d − θ∗
d−2) · · · (θ∗
d − θ∗
0)
φ1φ2 · · · φd
.
Moreover ν∗ = ν and ν∗ = ν.
Proof: Line (30) holds by [1, Lemma 7.6]. The remaining assertions follow from Corollary
7.1.
✷
Lemma 9.4. Let (A; {Ei}d
i=0; A∗; {E∗
i }d
i=0) denote a TH system. Then both
νEdE∗
dEd = Ed,
νE∗
dEdE∗
d = E∗
d.
Proof: Let Φ denote the TH system in question. Apply Lemma 9.2 to Φ and then apply σ−1
to each side of the resulting equations.
✷
21
10 Anti-isomorphic TH systems and the bilinear form
Let Φ denote a TH system on V and let Φ denote the relative of Φ from Definition 6.12.
Recall that Φ and Φ are anti-isomorphic. In this section, we discuss further the relationship
between Φ and Φ. Recall the canonical bilinear form h , i : V × V → K from above
Definition 6.12. Let U (resp. U) denote a subspace of V (resp. V ). We say that U and U
are orthogonal whenever hx, yi = 0 for all x ∈ U and y ∈ U . Let {Vi}d
i=0)
denote a decomposition of V (resp. V ). We say that {Vi}d
i=0 are dual whenever
Vi and Vj are orthogonal for 0 ≤ i, j ≤ d, i 6= j. By elementary linear algebra, for any
i=0) of V (resp. V ) there exists a unique decomposition
decomposition {Vi}d
{ Vi}d
i=0 (resp. {Vi}d
(resp. {vi}d
whenever hvi, vji = δij for 0 ≤ i, j ≤ d. By elementary linear algebra, for any basis {vi}d
(resp. {vi}d
V ) such that {vi}d
{αi}d
i=0 and { Vi}d
i=0) denote a basis for V (resp. V ). We say that {vi}d
i=0 (resp. { Vi}d
i=0) of V (resp. V ) such that {Vi}d
i=0) for V (resp. V ) there exists a unique basis {vi}d
i=0) for V (resp.
i=0, by the inversion of
i=0 are dual. Let {vi}d
i=0 and {vi}d
i=0 are dual. Given any sequence {αi}d
i=0
i=0 are dual
i=0
i=0 (resp. { Vi}d
i=0 (resp. {vi}d
i=0 and { Vi}d
i=0 and {vi}d
i=0 we mean the sequence {αd−i}d
Recall the Φ-standard decomposition {E∗
i=0.
i=0 from above (3). Observe by Definition
i=0 is the Φ-standard decomposition. We now compare these two decom-
i V }d
6.12 that {E∗σ
d−i
positions.
V }d
Lemma 10.1. With reference to Definition 6.12, the following (i), (ii) are inverted dual.
(i) The Φ-standard decomposition of V .
(ii) The Φ-standard decomposition of V .
V . Simplify the equation hA∗u, vi = hu, A∗σvi using A∗u = θ∗
j . Therefore E∗
i − θ∗
j )hu, vi = 0. Now hu, vi = 0 since θ∗
i 6= θ∗
i V and E∗σ
j
V are orthogonal. Let
i u and
i V and
✷
i V and v ∈ E∗σ
j
Proof: For distinct i, j (0 ≤ i, j ≤ d) we show that E∗
u ∈ E∗
A∗σv = θ∗
E∗σ
j
V are orthogonal and the result follows.
Let 0 6= ξ0 ∈ E0V and recall the Φ-standard basis {E∗
j v to obtain (θ∗
0 6= ξd ∈ Eσ
These two bases are related as follows.
V and observe by Definition 6.12 that {E∗σ
d−i
d
i ξ0}d
ξd}d
i=0 for V from above (3). Let
i=0 is a Φ-standard basis for V .
Proposition 10.2. With reference to Definition 6.12, let 0 6= ξ0 ∈ E0V and 0 6= ξd ∈ Eσ
Then for 0 ≤ i, j ≤ d,
d
V .
hE∗
i ξ0, E∗σ
j
ξdi = δijℓihE∗
0ξ0, ξdi,
where ℓi is from Definition 8.1.
Proof: Using the definition of σ from above Definition 6.12 along with Lemma 8.3, we find
hE∗
i ξ0, E∗σ
j
ξdi = hE∗
i E0ξ0, E∗σ
j E∗
= hE∗
= δijhE∗
j Eσ
i E0ξ0, Eσ
i E0ξ0, Eσ
ξdi
ξdi
ξdi
d
d
d
22
= δijhEdE∗
= δijℓihEdE∗
= δijℓihE∗
= δijℓihE∗
i E0ξ0, ξdi
0E0ξ0, ξdi
ξdi
0E0ξ0, Eσ
0ξ0, ξdi.
d
Corollary 10.3. With reference to Definition 6.12, let {vi}d
for V (resp. V ). Suppose that {vi}d
(ii) are equivalent.
i=0 and {wi}d
i=0) denote a basis
i=0 are inverted dual. Then the following (i),
i=0 (resp. {wi}d
✷
(i) {ℓivi}d
(ii) {wi}d
i=0 is a Φ-standard basis for V .
i=0 is a Φ-standard basis for V .
Proof: Use Proposition 10.2.
✷
By Definition 6.1, the sequence {EiV }d
i=0 is the Φ∗-standard decomposition. By Defi-
i=0 is the Φ∗-standard decomposition. We now compare
V }d
nition 6.12, the sequence {Eσ
these two decompositions.
d−i
Lemma 10.4. With reference to Definition 6.12, the following (i), (ii) are inverted dual.
(i) The Φ∗-standard decomposition of V .
(ii) The Φ∗-standard decomposition of V .
Proof: Apply Lemma 10.1 to Φ∗.
✷
Let 0 6= ξ∗
0 ∈ E∗
for V . Let 0 6= ξ∗
basis for V . These two bases are related as follows.
d ∈ E∗σ
d
0 V and observe by Definition 6.1 that {Eiξ∗
0}d
V and observe by Definition 6.12 that {Eσ
i=0 is a Φ∗-standard basis
i=0 is a Φ∗-standard
ξ∗
d}d
d−i
Proposition 10.5. With reference to Definition 6.12, let 0 6= ξ∗
Then for 0 ≤ i, j ≤ d,
0 ∈ E∗
0V and 0 6= ξ∗
d ∈ E∗σ
d
hEiξ∗
0, Eσ
j
where ℓ∗
i is from Lemma 8.2.
Proof: Apply Proposition 10.2 to Φ∗.
ξ∗
di = δijℓ∗
i hE0ξ∗
0, ξ∗
di,
V .
✷
Corollary 10.6. With reference to Definition 6.12, let {vi}d
for V (resp. V ). Suppose that {vi}d
(ii) are equivalent.
i=0 and {wi}d
i=0) denote a basis
i=0 are inverted dual. Then the following (i),
i=0 (resp. {wi}d
(i) {ℓ∗
i vi}d
(ii) {wi}d
i=0 is a Φ∗-standard basis for V .
i=0 is a Φ∗-standard basis for V .
23
Proof: Apply Corollary 10.3 to Φ∗.
Let {Ui}d
i=0 denote the Φ-split decomposition of V . Recall from (3) that for 0 ≤ i ≤ d,
✷
Ui = (E∗
(31)
i=0 denote the Φ-split decomposition of V . Combining Definition 6.12 and (31), we
i V ) ∩ (E0V + E1V + · · · + Ed−iV ).
1 V + · · · + E∗
0V + E∗
Let { Ui}d
find that for 0 ≤ i ≤ d,
Ui = (E∗σ
d
V + E∗σ
d−1
V + · · · + E∗σ
d−i
V ) ∩ (Eσ
d
V + Eσ
d−1
V + · · · + Eσ
i
V ).
(32)
We now compare these two decompositions.
Lemma 10.7. With reference to Definition 6.12, the following (i), (ii) are inverted dual.
(i) The Φ-split decomposition of V .
(ii) The Φ-split decomposition of V .
0 V +E∗
Proof: We use the notation from above this lemma. For 0 ≤ i, j ≤ d with i + j 6= d, we show
that Ui and Uj are orthogonal. We consider two cases: i + j < d and i + j > d. First suppose
V .
that i+j < d. Abbreviate M = E∗
Observe that Ui ⊆ M by (31) and Uj ⊆ N by (32). Moreover M and N are orthogonal by
our assumption and Lemma 10.1. Therefore Ui and Uj are orthogonal. Next suppose that
V .
i + j > d. Abbreviate S = E0V + E1V + · · · + Ed−iV and T = Eσ
d
Observe Ui ⊆ S by (31) and Uj ⊆ T by (32). Moreover S and T are orthogonal by our
assumption and Lemma 10.4. Therefore Ui and Uj are orthogonal and the result follows. ✷
Let 0 6= ξ0 ∈ E0V and recall from (7) that the sequence
i V and N = E∗σ
V + · · · + Eσ
1V +· · ·+E∗
V +· · ·+E∗σ
V +E∗σ
d−1
V + Eσ
d−1
d−j
d
j
η∗
d−i(A∗)ξ0
is a Φ-split basis for V . Let 0 6= ξd ∈ Eσ
d−i(A∗σ) ξd
τ ∗
d
(0 ≤ i ≤ d)
(33)
V and observe by Definition 6.12 that the sequence
(0 ≤ i ≤ d)
(34)
is a Φ-split basis for V . The bases (33), (34) are related as follows.
Proposition 10.8. With reference to Definition 6.12, let 0 6= ξ0 ∈ E0V and 0 6= ξd ∈ Eσ
Then for 0 ≤ i, j ≤ d,
d
V .
hη∗
i (A∗)ξ0, τ ∗
j (A∗σ) ξdi = δi+j,d η∗
d(θ∗
0)hE∗
0ξ0, ξdi.
Proof: First suppose i + j 6= d. Then the result holds by Lemma 10.7 and the comments
above this proposition. Next suppose that i + j = d. Using (2), Proposition 10.2, and the
definition of σ from above Definition 6.12, we find
hη∗
i (A∗)ξ0, τ ∗
j (A∗σ) ξdi = hη∗
= hτ ∗
= τ ∗
= τ ∗
= τ ∗
= τ ∗
= η∗
d−j(A∗)ξ0, τ ∗
j (A∗)η∗
j (θ∗
j )η∗
j (θ∗
j )η∗
j )η∗
j (θ∗
j (θ∗
j )η∗
0)hE∗
d(θ∗
j (A∗σ) ξdi
d−j(A∗)ξ0, ξdi
j ξ0, ξdi
d−j(θ∗
d−j(θ∗
j E∗
j ξ0, E∗σ
d−j(θ∗
0ξ0, ξdi
d−j(θ∗
0ξ0, ξdi.
j )hE∗
j )hE∗
j )hE∗
j )ℓjhE∗
j ξ0, ξdi
ξdi
j
24
Corollary 10.9. With reference to Definition 6.12, let {vi}d
for V (resp. V ). Suppose that {vi}d
(ii) are equivalent.
i=0 and {wi}d
i=0) denote a basis
i=0 are inverted dual. Then the following (i),
i=0 (resp. {wi}d
✷
(i) {vi}d
(ii) {wi}d
i=0 is a Φ-split basis for V .
i=0 is a Φ-split basis for V .
Proof: Use Proposition 10.8.
✷
At the end of of Section 17, we give the relationship between a Φ-standard (resp. Φ∗-
standard) basis for V and a Φ∗-standard (resp. Φ-standard) basis for V . This relationship
is of a different type than the ones in the present section.
11 The transition matrices for a TH system
vj = Pd
Let Φ denote a TH system. In this section we consider several transition matrices associated
with Φ. First we clarify our terms. Let {ui}d
i=0 denote bases for V . By the
transition matrix from {ui}d
i=0, we mean the matrix T ∈ Matd+1(K) such that
i=0 Tijui for 0 ≤ j ≤ d.
i=0 and {vi}d
i=0 to {vi}d
Definition 11.1. [1, Definition 10.6] Let Φ = (A; {Ei}d
on V . Let 0 6= ξ0 ∈ E0V and 0 6= ξ∗
and the Φ∗-standard basis {Eiξ∗
from {Eiξ∗
i=0) denote a TH system
i=0 for V
i=0 for V . Let P ∈ Matd+1(K) denote the transition matrix
0V . Recall the Φ-standard basis {E∗
i=0; A∗; {E∗
0 chosen so that ξ∗
i=0, with ξ0, ξ∗
0 ∈ E∗
0 = E∗
0}d
i=0 to {E∗
i ξ0}d
i ξ0}d
0 ξ0.
0}d
i }d
Theorem 11.2. [1, Theorem 10.8] Let Φ denote a TH system with parameter array
({θi}d
the (i, j)-entry of P is equal to ℓj times
i=1) and let P denote the matrix from Definition 11.1. For 0 ≤ i, j ≤ d,
i=0, {φi}d
i=0, {θ∗
i }d
d
Xh=0
(θi − θd)(θi − θd−1) · · · (θi − θd−h+1)(θ∗
φ1φ2 · · · φh
j − θ∗
0)(θ∗
j − θ∗
1) · · · (θ∗
j − θ∗
h−1)
where ℓj is from Definition 8.1.
Corollary 11.3. With reference to Definition 11.1, for 0 ≤ i ≤ d both
Pi0 = 1,
Pdi = ℓi,
where ℓi is from Definition 8.1.
Proof: Use Theorem 11.2.
,
(35)
✷
The following definition is motivated by Definition 11.1 and Corollary 11.3.
Definition 11.4. We call the matrix P from Definition 11.1 the west normalized transition
matrix of Φ.
25
Motivated by the sum (35), we make a definition. Let λ, µ denote commuting indeterminates.
Let K[λ, µ] denote the K-algebra consisting of the polynomials in λ, µ that have all coefficients
in K.
Definition 11.5. Let Φ denote a TH system with parameter array ({θi}d
Define p ∈ K[λ, µ] by
i=0, {θ∗
i }d
i=0, {φi}d
i=1).
p =
d
Xh=0
ηh(λ)τ ∗
h (µ)
φ1φ2 · · · φh
,
(36)
where {τ ∗
Φ.
i }d
i=0 and {ηi}d
i=0 are from Notation 2.7. We call p the two-variable polynomial of
Example 11.6. With reference to Definition 11.5, assume d = 2. Then
p = 1 +
(λ − θ2)(µ − θ∗
0)
φ1
+
(λ − θ2)(λ − θ1)(µ − θ∗
0)(µ − θ∗
1)
φ1φ2
.
Remark 11.7. With reference to Definition 11.5, for 0 ≤ i, j ≤ d the scalar p(θi, θ∗
sum (35).
j ) is the
Definition 11.8. Let Φ denote a TH system with parameter array ({θi}d
i=1).
Let P ∈ Matd+1(K) denote the matrix with (i, j)-entry p(θi, θ∗
j ) for 0 ≤ i, j ≤ d. Observe
that by Theorem 11.2 and Remark 11.7, the matrix P from Definition 11.1 is equal to PL,
where L is the matrix from Definition 8.5.
i=0, {φi}d
i=0, {θ∗
i }d
Example 11.9. With reference to Definition 11.8, assume d = 2. Then
P =
1 1 + (θ0−θ2)(θ∗
1 1 + (θ1−θ2)(θ∗
1
1
φ1
φ1
1 −θ∗
0 )
1 −θ∗
0 )
1 + (θ0−θ2)(θ∗
2 −θ∗
0)
+ (θ0−θ2)(θ0−θ1)(θ∗
2 −θ∗
0)(θ∗
2 −θ∗
1 )
φ1
φ1φ2
1 + (θ1−θ2)(θ∗
2 −θ∗
0 )
φ1
1
.
Corollary 11.10. With reference to Definition 11.8, for 0 ≤ i ≤ d both
Pi0 = 1,
Pdi = 1.
Proof: Routine.
✷
We now interpret the matrix P as a transition matrix. Let {ui}d
i=0 denote the inverted dual
of a Φ-standard basis for V . By Corollary 10.3, {ℓiui}d
i=0 is a Φ-standard basis for V , where
ℓi is from Definition 8.1. Recall the canonical bilinear form h , i : V × V → K from above
Definition 6.12.
d
i=0; A∗; {E∗
i }d
0 V and 0 6= ξd ∈ Eσ
V . Note by Lemma 10.1 that hξ∗
Corollary 11.11. Let Φ = (A; {Ei}d
ξ∗
0 ∈ E∗
standard basis {Eiξ∗
the matrix from Definition 11.8. Then αP is the transition matrix from {Eiξ∗
inverted dual of {E∗σ
d−i
0, ξd so that hξ∗
ξ∗
of {E∗σ
i=0.
d−i
i=0) denote a TH system on V . Let 0 6=
0, ξdi 6= 0. Recall the Φ∗-
ξd}d
i=0 for V . Let P denote
i=0 to the
0, ξdi. In particular if we choose
i=0 to the inverted dual
0, ξdi = 1, then P is the transition matrix from {Eiξ∗
i=0 for V and the Φ-standard basis {E∗σ
i=0, where α is the reciprocal of hξ∗
ξd}d
ξd}d
0}d
0}d
0}d
d−i
26
0 = E∗
Proof: Choose ξ0 ∈ E0V so that ξ∗
matrix P from Definition 11.1 is the transition matrix from {Eiξ∗
by Definition 11.8, αP is the transition matrix from {Eiξ∗
by Proposition 10.2, {αℓ−1
i E∗
transition matrix from {Eiξ∗
0 ξ0, and recall the Φ-standard basis {E∗
i=0 to {E∗
i E∗
i ξ0}d
i ξ0}d
i=0. The
i ξ0}d
i=0. Now
i=0. Moreover
i=0. Therefore αP is the
✷
i=0 is the inverted dual of {E∗σ
i=0 to the inverted dual of {E∗σ
i=0 to {αℓ−1
i ξ0}d
0}d
ξd}d
i=0.
ξd}d
0}d
0}d
d−i
d−i
The following definition is motivated by Corollary 11.10 and Corollary 11.11.
Definition 11.12. We call the matrix P from Definition 11.8 the west-south normalized
transition matrix of Φ.
12 The transition matrices P, P and their relatives
Let Φ denote a TH system. In the previous section we discussed two closely related transition
matrices P, P associated with Φ. In this section we find the relationship between P, P and
their relatives. There are two types of relations; one type is best expressed in terms of P
and its relatives, while the other is best expressed in terms of P and its relatives.
Proposition 12.1. [1, Proposition 10.9] With reference to Definition 11.1, both
P P ∗ = P ∗P = νI,
P P ∗ = P ∗ P = νI,
where ν, ν are from Definition 9.1.
Let {ui}d
i=0 and {vi}d
matrix from {ui}d
from the dual of {vi}d
the inverted dual of {vi}d
i=0 to {vi}d
i=0 denote bases for V . Let T ∈ Matd+1(K) denote the transition
i=0. By elementary linear algebra, T t is the transition matrix
i=0. Therefore T ς is the transition matrix from
i=0 to the dual of {ui}d
i=0 to the inverted dual of {ui}d
i=0.
Proposition 12.2. With reference to Definition 11.8, both
P ς = P ∗,
(P ∗)ς = P.
(37)
0}d
0 ∈ E∗
0V and 0 6= ξd ∈ Eσ
i=0 for V and the Φ-standard basis {E∗σ
Proof: Choose 0 6= ξ∗
basis {Eiξ∗
is the transition matrix from {Eiξ∗
i=0 to the inverted dual of {E∗σ
Corollary 11.11 applied to Φ∗, P ∗ is the transition matrix from {E∗σ
dual of {Eiξ∗
equation on the left in (37). The equation on the right in (37) is similarly obtained.
0, ξdi = 1. Recall the Φ∗-standard
i=0 for V . By Corollary 11.11, P
i=0. Moreover by
i=0 to the inverted
i=0. By these comments and the ones above this proposition we obtain the
✷
V so that hξ∗
ξd}d
ξd}d
ξd}d
0}d
0}d
d−i
d−i
d−i
d
The following lemma could be used to give another proof of Proposition 12.2.
Lemma 12.3. With reference to Definition 11.5, both
p(µ, λ) = p∗(λ, µ),
p∗(µ, λ) = p(λ, µ).
27
Proof: Combining Corollary 7.1 and Definition 11.5, we find
p∗(λ, µ) =
d
Xh=0
τ ∗
h (λ)ηh(µ)
φ1φ2 · · · φh
.
Therefore p(µ, λ) = p∗(λ, µ). The proof for the other claim is similar.
✷
We will continue our discussion of the transition matrices P, P in Section 17. In Sections
13 -- 16 we recall some linear algebra that will be needed in the discussion.
13 Vandermonde matrices and systems
Let Φ denote a TH system. In Section 17 we will show that each of the transition matrices
P, P of Φ has a certain structure said to be double Vandermonde. To prepare for that, over
the next few sections we discuss some linear algebra related to Vandermonde matrices.
Definition 13.1. Let n denote a nonnegative integer. Let {fi}n
polynomials in K[λ]. We say that {fi}n
i=0 is graded whenever
i=0 denote a sequence of
(i) f0 = 1;
(ii) the degree of fi is equal to i for 0 ≤ i ≤ n.
Definition 13.2. A matrix X ∈ Matd+1(K) is called west Vandermonde whenever the
following (i), (ii) hold.
(i) There exists a sequence of mutually distinct scalars {θi}d
i=0 taken from K and a graded
sequence of polynomials {fi}d
i=0 in K[λ] such that
Xij = Xi0fj(θi)
(0 ≤ i, j ≤ d).
(38)
(ii) Xi0 6= 0 for 0 ≤ i ≤ d.
With reference to Definition 13.2, assume X is west Vandermonde. As we will see,
i=0 but the
i=0 is not unique. To facilitate our discussion of this issue, we introduce the
i=0 are uniquely determined by the sequence of scalars {θi}d
the polynomials {fi}d
sequence {θi}d
following term.
Definition 13.3. Let X ∈ Matd+1(K) denote a west Vandermonde matrix. Let {θi}d
denote a sequence of scalars taken from K. We say that X and {θi}d
whenever
i=0
i=0 are compatible
(i) θi 6= θj if i 6= j
(0 ≤ i, j ≤ d);
(ii) there exists a graded sequence of polynomials {fi}d
i=0 in K[λ] that satisfies (38).
Observe that if X and {θi}d
any α, β ∈ K with α 6= 0.
i=0 are compatible, then X and {αθi + β}d
i=0 are compatible for
28
Lemma 13.4. Let X ∈ Matd+1(K) denote a west Vandermonde matrix. Let {θi}d
i=0 denote
a sequence of scalars taken from K. Then the following (i), (ii) are equivalent provided d ≥ 1.
(i) X and {θi}d
i=0 are compatible.
(ii) There exists a, b ∈ K with a 6= 0 such that θi = aXi1/Xi0 + b for 0 ≤ i ≤ d.
Proof: (i) ⇒ (ii) By Definition 13.3, there exists a polynomial f1 ∈ K[λ] of degree 1 such
that Xi1 = Xi0f1(θi) for 0 ≤ i ≤ d. Write f1 = αλ + β for some α, β ∈ K with α 6= 0. Thus
Xi1 = Xi0(αθi + β) for 0 ≤ i ≤ d. Rearranging terms, we find that there exists a, b ∈ K with
a 6= 0 such that θi = aXi1/Xi0 + b for 0 ≤ i ≤ d.
(ii) ⇒ (i) By Definition 13.2, there exists a sequence of scalars {θ′
i=0 taken from K that
is compatible with X. By the previous part, there exists a′, b′ ∈ K with a′
6= 0 such that
i = a′Xi1/Xi0 + b′ for 0 ≤ i ≤ d. Thus there exists α, β ∈ K with α 6= 0 such that
θ′
θi = αθ′
i=0 are compatible by the observation at the end
of Definition 13.3.
✷
i + β for 0 ≤ i ≤ d. Now X and {θi}d
i}d
Lemma 13.5. Let X ∈ Matd+1(K) denote a west Vandermonde matrix. Let {θi}d
a sequence of scalars taken from K that is compatible with X. Let {θ′
of scalars taken from K. Then the following (i), (ii) are equivalent.
i=0 denote
i=0 denote a sequence
i}d
(i) X and {θ′
i}d
i=0 are compatible.
(ii) There exists α, β ∈ K with α 6= 0 such that θ′
i = αθi + β for 0 ≤ i ≤ d.
Proof: (i) ⇒ (ii) Routine by Lemma 13.4.
(ii) ⇒ (i) This is the observation at the end of Definition 13.3.
✷
Lemma 13.6. Let {θi}d
X ∈ Matd+1(K) denote a west Vandermonde matrix that is compatible with {θi}d
there exists a unique graded sequence of polynomials {fi}d
i=0 denote a sequence of mutually distinct scalars taken from K. Let
i=0. Then
i=0 in K[λ] that satisfies (38).
i=0 in K[λ]
i=0 is a graded
i = fi for 0 ≤ i ≤ d.
i − fi. Using (38), we find gi(θj) = 0 for 0 ≤ j ≤ d. Since
i=0 are mutually distinct and gi has degree at most i, it follows that gi = 0. Therefore
✷
Proof: By Definition 13.3 there exists a graded sequence of polynomials {fi}d
that satisfies (38). We show that this sequence is unique. Suppose that {f ′
sequences of polynomials in K[λ] that satisfies (38). We show that f ′
Let i be given and define gi = f ′
{θi}d
f ′
i = fi. We have shown that the sequence {fi}d
Lemma 13.7. Let {θi}d
i=0 denote a sequence of mutually distinct scalars taken from K.
Let {fi}d
i=0 denote a sequence
of nonzero scalars taken from K. Define X ∈ Matd+1(K) such that Xij = cifj(θi) for
0 ≤ i, j ≤ d. Then X is west Vandermonde and compatible with {θi}d
are the corresponding polynomials from Lemma 13.6.
i=0 denote a graded sequence of polynomials in K[λ]. Let {ci}d
i=0. Moreover {fi}d
i=0 is unique.
i }d
i=0
Proof: Routine.
✷
i=0 denote a sequence of mutually distinct scalars taken from K. Let
i=0 and let
i=0 denote the corresponding polynomials from Lemma 13.6. Let X ′ ∈ Matd+1(K). Then
Lemma 13.8. Let {θi}d
X ∈ Matd+1(K) denote a west Vandermonde matrix that is compatible with {θi}d
{fi}d
the following (i), (ii) are equivalent.
29
(i) X ′ is a west Vandermonde matrix that is compatible with {θi}d
i=0 and {fi}d
i=0 are the
corresponding polynomials from Lemma 13.6.
(ii) There exists an invertible diagonal matrix D ∈ Matd+1(K) such that X ′ = DX.
Proof: (i) ⇒ (ii) Using (38), we find that for 0 ≤ i, j ≤ d,
X ′
ij
X ′
i0
=
Xij
Xi0
= fj(θi).
(39)
Define a diagonal matrix D ∈ Matd+1(K) with (i, i)-entry X ′
that D is invertible. Moreover X ′ = DX by (39).
(ii) ⇒ (i) Since X ′ = DX we find that for 0 ≤ i, j ≤ d,
i0/Xi0 for 0 ≤ i ≤ d. Observe
X ′
ij = X ′
i0
Xij
Xi0
.
ij = X ′
i0fj(θi) and (i) follows.
By this and (38), we find that X ′
Lemma 13.9. Let {θi}d
i=0 denote a sequence of mutually distinct scalars taken from K. Let
X ∈ Matd+1(K) denote a west Vandermonde matrix that is compatible with {θi}d
i=0 and let
{fi}d
i=0 denote the corresponding polynomials from Lemma 13.6. Let D ∈ Matd+1(K) denote
an invertible diagonal matrix. Then XD is a west Vandermonde matrix that is compatible
with {θi}d
Proof: Routine using (38).
i=0 are the corresponding polynomials from Lemma 13.6.
i=0 and {Diifi/D00}d
✷
✷
Definition 13.10. By a west Vandermonde system in Matd+1(K) we mean a sequence
(X, {θi}d
i=0) such that
(i) X is a west Vandermonde matrix in Matd+1(K);
(ii) {θi}d
i=0 is a sequence of mutually distinct scalars taken from K that is compatible with
X.
fd+1 = Qd
Definition 13.11. Let (X, {θi}d
Lemma 13.6 we associated (X, {θi}d
i=0(λ − θi). We call {fi}d+1
i=0) denote a west Vandermonde system in Matd+1(K). In
i=0. For convenience, let
i=0) with some polynomials {fi}d
i=0 the polynomials of (X, {θi}d
i=0).
Definition 13.12. Let X ∈ Matd+1(K). Let X ′ ∈ Matd+1(K) denote a matrix that is
obtained by rotating X clockwise 90 degrees. We call X south Vandermonde whenever X ′
is west Vandermonde.
The above notions regarding west Vandermonde matrices carry over to south Vandermonde
matrices.
We end this section with a comment.
Lemma 13.13. Let X ∈ Matd+1(K) denote a west or south Vandermonde matrix. Then X
is invertible.
Proof: First assume that X is west Vandermonde. Perform invertible row and column
operations on X so that the resulting matrix X ′ has (i, j)-entry θj
i for 0 ≤ i, j ≤ d. The
i=0 are mutually distinct so this
determinant is nonzero. Therefore X ′ is invertible so X is invertible. The case of south
Vandermonde is similar.
✷
determinant of X ′ is equal to Q0≤i<j≤d(θj − θi). The {θi}d
30
14 Hessenberg matrices and graded sequences of poly-
nomials
Recall the notion of a Hessenberg matrix from Section 1. In the next section we discuss the
role Vandermonde matrices play in the diagonalization of Hessenberg matrices. To prepare
for that, in this section we discuss the relationship between Hessenberg matrices and graded
sequences of polynomials.
Lemma 14.1. Let H ∈ Matd+1(K) denote a Hessenberg matrix. Then the minimal polyno-
mial of H is equal to the characteristic polynomial of H.
Proof: Using the Hessenberg shape of H, we find I, H, H 2, . . . , H d are linearly independent.
Therefore the minimal polynomial of H has degree d + 1. The result follows.
✷
Given a Hessenberg matrix H, we are interested in finding the polynomial in Lemma
14.1.
Notation 14.2. Let H ∈ Matd+1(K) denote a Hessenberg matrix. We denote by cH the
product Qd
i=1 Hi,i−1. Observe that cH is nonzero.
Definition 14.3. Let H ∈ Matd+1(K) denote a Hessenberg matrix. Define a sequence of
polynomials {fi}d+1
i=0 in K[λ] such that
(i) f0 = 1;
(ii) λfj = Pj+1
(iii) λfd = c−1
i=0 Hijfi for 0 ≤ j ≤ d − 1;
H fd+1 +Pd
We call {fi}d+1
i=0 the polynomials of H.
i=0 Hidfi, where cH is from Notation 14.2.
Definition 14.4. A graded sequence of polynomials {fi}d+1
ever fd+1 is monic.
i=0 in K[λ] is called standard when-
Lemma 14.5. Let H ∈ Matd+1(K) denote a Hessenberg matrix with polynomials {fi}d+1
i=0 .
Then the following (i) -- (iii) hold.
(i) For 0 ≤ i ≤ d, fi has degree i with λi coefficient (Qi
(ii) fd+1 is monic with degree d + 1.
j=1 Hj,j−1)−1.
(iii) The sequence {fi}d+1
i=0 is graded and standard.
Proof: Routine.
✷
Let I denote the identity matrix in Matd+1(K). For 0 ≤ i ≤ d, let ǫi denote the ith
column of I. Observe that {ǫi}d
i=0 is a basis for the vector space Kd+1.
Lemma 14.6. Let H ∈ Matd+1(K) denote a Hessenberg matrix. Then there exists a unique
standard graded sequence of polynomials {fi}d+1
i=0 in K[λ] such that fi(H)ǫ0 = ǫi for 0 ≤ i ≤ d
and fd+1(H)ǫ0 = 0. The {fi}d+1
i=0 are the polynomials of H from Definition 14.3.
31
Proof: Concerning existence, let {fi}d+1
i=0 denote the polynomials of H. By Lemma 14.5(iii)
the sequence {fi}d+1
i=0 is graded and standard. We show that fi(H)ǫ0 = ǫi for 0 ≤ i ≤ d and
fd+1(H)ǫ0 = 0. Abbreviate vi = fi(H)ǫ0 for 0 ≤ i ≤ d + 1 and note that v0 = ǫ0. From
Definition 14.3, we have
Hvj =
j+1
Xi=0
Hijvi
(0 ≤ j ≤ d),
where Hd+1,d = c−1
H . By the definition of {ǫi}d
i=0, we have
Hǫj =
j+1
Xi=0
Hijǫi
(0 ≤ j ≤ d),
(40)
(41)
where ǫd+1 = 0. Comparing (40), (41) and using v0 = ǫ0, we find vi = ǫi for 0 ≤ i ≤ d + 1.
i }d+1
Therefore fi(H)ǫ0 = ǫi for 0 ≤ i ≤ d and fd+1(H)ǫ0 = 0. Concerning uniqueness, let {f ′
i=0
denote a standard graded sequence of polynomials in K[λ] such that f ′
i (H)ǫ0 = ǫi for 0 ≤
i ≤ d and f ′
i = fi for 0 ≤ i ≤ d + 1. Let i be given and define
gi = f ′
i − fi. Observe that gi(H)ǫ0 = 0. Thus gi(H)ǫj = gi(H)fj(H)ǫ0 = fj(H)gi(H)ǫ0 = 0
for 0 ≤ j ≤ d, so gi(H) = 0. Therefore the minimal polynomial of H divides gi. The
polynomial gi has degree at most d, and the minimal polynomial of H has degree d + 1 by
Lemma 14.1. Therefore gi = 0 so f ′
✷
d+1(H)ǫ0 = 0. We show that f ′
i = fi.
Corollary 14.7. Let H ∈ Matd+1(K) denote a Hessenberg matrix with polynomials {fi}d+1
i=0 .
Then fd+1 is both the minimal polynomial and the characteristic polynomial of H.
Proof: Using Lemma 14.6 we find that fd+1(H)ǫi = fd+1(H)fi(H)ǫ0 = fi(H)fd+1(H)ǫ0 = 0
for 0 ≤ i ≤ d. Therefore fd+1(H) = 0. The result follows by Lemma 14.1 and Lemma
14.5(ii).
✷
So far, given a Hessenberg matrix we obtain a graded sequence of polynomials. Now
turning things around, given a graded sequence of polynomials we obtain a Hessenberg
matrix.
Definition 14.8. Let {fi}d+1
that for 0 ≤ j ≤ d, λfj is in the span of {fi}j+1
sequence of scalars {cij}j+1
the (i, j)-connection coefficient for the given graded sequence of polynomials.
i=0 denote a graded sequence of polynomials in K[λ]. Observe
i=0 . So for 0 ≤ j ≤ d, there exists a unique
i=0 cijfi. We call the scalar cij
i=0 taken from K such that λfj = Pj+1
Definition 14.9. Let {fi}d+1
i=0 denote a graded sequence of polynomials in K[λ]. By the
connection coefficient matrix of {fi}d+1
i=0 , we mean the Hessenberg matrix H ∈ Matd+1(K)
such that Hij = cij for 0 ≤ i, j ≤ d, i − j ≤ 1. The scalars cij are from Definition 14.8.
Observe that the scalar cd+1,d plays no role in the definition of H.
Lemma 14.10. Let {fi}d+1
i=0 ) denote a graded sequence of polynomials in K[λ]
with connection coefficient matrix H (resp. H ′). Then the following (i), (ii) are equivalent.
i=0 (resp. {f ′
i }d+1
(i) H = H ′.
32
(ii) fi = f ′
i for 0 ≤ i ≤ d and there exists 0 6= c ∈ K such that fd+1 = cf ′
d+1.
Proof: Routine by Definition 14.9.
✷
Lemma 14.11. Let {fi}d+1
i=0 denote a standard graded sequence of polynomials in K[λ] and
let H ∈ Matd+1(K) denote a Hessenberg matrix. Then the following (i), (ii) are equivalent.
(i) {fi}d+1
i=0 are the polynomials of H.
(ii) H is the connection coefficient matrix of {fi}d+1
i=0 .
Proof: Routine.
✷
15 A Vandermonde matrix as a transition matrix
In this section we discuss the role that Vandermonde matrices play in the diagonalization of
a Hessenberg matrix.
Lemma 15.1. Let H ∈ Matd+1(K) denote a Hessenberg matrix with polynomials {fi}d+1
i=0 .
Then the following (i) -- (iii) are equivalent.
(i) H is diagonalizable.
(ii) H is multiplicity-free.
(iii) fd+1 has d + 1 distinct roots in K.
Proof: Recall that the polynomial fd+1 has degree d + 1 and it is the minimal polynomial of
H by Corollary 14.7. By elementary linear algebra a matrix in Matd+1(K) is diagonalizable
if and only if its minimal polynomial has distinct roots in K. The result follows.
✷
Let H ∈ Matd+1(K) denote a multiplicity-free Hessenberg matrix and let {θi}d
i=0 denote
an ordering of the eigenvalues of H. Let D ∈ Matd+1(K) denote the diagonal matrix with
(i, i)-entry θi for 0 ≤ i ≤ d. By elementary linear algebra, there exists an invertible X ∈
Matd+1(K) such that H = X −1DX. We comment on the uniqueness of X. Suppose that Y ∈
Matd+1(K) is invertible and H = Y −1DY . Then X −1DX = Y −1DY so DY X −1 = Y X −1D.
Therefore Y X −1 is diagonal by Lemma 3.1. By construction Y X −1 is invertible. By these
comments there exists an invertible diagonal matrix ∆ ∈ Matd+1(K) such that Y = ∆X.
Lemma 15.2. Let H ∈ Matd+1(K) denote a multiplicity-free Hessenberg matrix with polyno-
mials {fi}d+1
i=0 denote an ordering of the eigenvalues of H and let D ∈ Matd+1(K)
denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤ d. For X ∈ Matd+1(K), the fol-
lowing (i), (ii) are equivalent.
i=0 . Let {θi}d
(i) X is invertible and H = X −1DX.
(ii) (X, {θi}d
i=0) is a west Vandermonde system with polynomials {fi}d+1
i=0 .
33
Proof: (i) ⇒ (ii) Observe that {θi}d
show that
i=0 are mutually distinct since H is multiplicity-free. We
Xij = Xi0fj(θi)
(0 ≤ i, j ≤ d).
(42)
Since H = X −1DX we have fj(H) = X −1fj(D)X so Xfj(H) = fj(D)X. Hence ǫt
iXfj(H)ǫ0 =
ǫt
ifj(D)Xǫ0. Simplify this equation using fj(H)ǫ0 = ǫj from Lemma 14.6 together with ma-
trix multiplication to obtain (42). By (42) and since X is invertible we find Xi0 6= 0 for
i=0)
i=0(λ − θi). By these comments (X, {θi}d
0 ≤ i ≤ d. By Corollary 14.7 we have fd+1 = Qd
is a west Vandermonde system with polynomials {fi}d+1
i=0 .
(ii) ⇒ (i) By Lemma 13.13 X is invertible. We now show that H = X −1DX. For 0 ≤ i ≤ d,
evaluate the equations in Definition 14.3(ii),(iii) at λ = θi. In the resulting equations multiply
n=0 HnjXin
✷
each side by Xi0 and simplify using (38) and Corollary 14.7 to obtain θiXij = Pd
for 0 ≤ j ≤ d. Therefore DX = XH so H = X −1DX.
i=0) denote a west Vandermonde system with polynomials
i=0 . Let D ∈ Matd+1(K) denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤
i=0 . Moreover X −1DX is
Corollary 15.3. Let (X, {θi}d
{fi}d+1
d. Then X −1DX is the connection coefficient matrix of {fi}d+1
multiplicity-free and Hessenberg.
Proof: Let H ∈ Matd+1(K) denote the connection coefficient matrix of {fi}d+1
i=0 . Then H is
Hessenberg by Definition 14.9. We show that H = X −1DX and that H is multiplicity-free.
By Lemma 14.11 the {fi}d+1
i=0(λ − θi) and
{θi}d
i=0 are mutually distinct, we find using Corollary 14.7 that H is multiplicity-free. Now
by Lemma 15.2 we find H = X −1DX. The result follows.
✷
i=0 are the polynomials of H. Since fd+1 = Qd
We have been discussing west Vandermonde systems. We now obtain analogous results for
south Vandermonde systems.
Definition 15.4. Let {fi}d+1
i=0 denote a standard graded sequence of polynomials in K[λ]. Let
H ∈ Matd+1(K) denote the connection coefficient matrix of {fi}d+1
i=0 . Recall from Definition
14.9 that H is Hessenberg, so H ς is Hessenberg. The polynomials of H ς will be denoted by
i }d+1
{f ς
i=0 are said to be associated. Note
that f ς
i=0 . The two polynomial sequences {fi}d+1
d+1 = fd+1 by Corollary 14.7.
i=0 and {f ς
i }d+1
Lemma 15.5. Let H ∈ Matd+1(K) denote a multiplicity-free Hessenberg matrix with polyno-
mials {fi}d+1
i=0 denote an ordering of the eigenvalues of H and let D ∈ Matd+1(K)
denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤ d. For X ∈ Matd+1(K), the fol-
lowing (i), (ii) are equivalent.
i=0 . Let {θi}d
(i) X is invertible and H = XDX −1.
(ii) (X, {θi}d
i=0) is a south Vandermonde system with polynomials {f ς
i }d+1
i=0 .
Proof: (i) ⇒ (ii) In the equation H = XDX −1, apply ς to each side to obtain H ς =
(X ς)−1DςX ς. By this and Lemma 15.2 the sequence (X ς, {θd−i}d
i=0) is a west Vandermonde
system with polynomials {f ς
i=0) is a south Vandermonde system
with polynomials {f ς
i=0 . Therefore (X, {θi}d
i }d+1
i }d+1
i=0 .
34
i }d+1
i=0 . Therefore (X ς, {θd−i}d
(ii) ⇒ (i) By assumption (X, {θi}d
i=0) is a south Vandermonde system with polynomials
i }d+1
{f ς
i=0 .
Applying Lemma 15.2 we find that X ς is invertible and H ς = (X ς)−1DςX ς. Applying ς we
find that X is invertible and H = XDX −1.
✷
i=0) is a west Vandermonde system with polynomials {f ς
i=0) denote a south Vandermonde system with polynomials
i=0 . Let D ∈ Matd+1(K) denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤
i=0 . Moreover XDX −1 is
Corollary 15.6. Let (X, {θi}d
{fi}d+1
d. Then XDX −1 is the connection coefficient matrix of {f ς
multiplicity-free and Hessenberg.
i }d+1
Proof: Similar to the proof of Corollary 15.3.
✷
16 The inverse of a Vandermonde matrix
Let X ∈ Matd+1(K) denote a west or south Vandermonde matrix.
showed that X is invertible. In this section we discuss the matrix X −1.
In Lemma 13.13 we
Proposition 16.1. Let (X, {θi}d
{fi}d+1
i=0 . Then the following (i), (ii) hold.
i=0) denote a west Vandermonde system with polynomials
(i) (X −1, {θi}d
i=0) is a south Vandermonde system with polynomials {f ς
i }d+1
i=0 , where {f ς
i }d+1
i=0
are the associated polynomials of {fi}d+1
i=0 .
(ii) (X −1)dj =
of {fi}d+1
i=0 and cH is from Notation 14.2.
cH
τj (θj )ηd−j (θj )Xj0
for 0 ≤ j ≤ d, where H is the connection coefficient matrix
i=0(λ − θi) and {θi}d
i=0) is a south Vandermonde system with polynomials {f ς
H by Lemma 14.11. Since fd+1 = Qd
Proof: (i) Let D ∈ Matd+1(K) denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤ d.
Observe that H is Hessenberg by Definition 14.9 and that {fi}d+1
i=0 are the polynomials of
i=0 are mutually distinct, we find
using Corollary 14.7 that H is multiplicity-free and {θi}d
i=0 is an ordering of the eigenvalues
of H. Therefore H = X −1DX by Lemma 15.2. Applying Lemma 15.5 to X −1, we find
(X −1, {θi}d
(ii) First assume that Xi0 = 1 for 0 ≤ i ≤ d. Let h ∈ K[λ] denote the polynomial
Pd
i=0(X −1)ijfi. In the equation XX −1 = I, evaluate the jth-column using matrix multi-
plication to find that h(θi) = δij for 0 ≤ i ≤ d. Let ej ∈ K[λ] denote the polynomial
τj (θj )ηd−j (θj ). Observe that ej(θi) = δij for 0 ≤ i ≤ d. Thus h(θi) = ej(θi) for 0 ≤ i ≤ d. It
follows that h = ej since both h and ej have degree d. In particular, the leading coefficient of
h is equal to the leading coefficient of ej. By Lemma 14.5(i) the leading coefficient of fd is c−1
H ,
so the leading coefficient of h is (X −1)djc−1
H . The leading coefficient of ej is (τj(θj)ηd−j(θj))−1.
By these comments (X −1)djc−1
τj (θj )ηd−j (θj ). The result is
now proven for the special case in which Xi0 = 1 for 0 ≤ i ≤ d. For the general case, apply
the special case to the west Vandermonde system (∆−1X, {θi}d
i=0), where ∆ ∈ Matd+1(K) is
the diagonal matrix with (i, i)-entry Xi0 for 0 ≤ i ≤ d.
✷
H = (τj(θj)ηd−j(θj))−1 so (X −1)dj =
i }d+1
i=0 .
τj ηd−j
cH
Proposition 16.2. Let (X, {θi}d
{fi}d+1
i=0 . Then the following (i), (ii) hold.
i=0) denote a south Vandermonde system with polynomials
35
(i) (X −1, {θi}d
i=0) is a west Vandermonde system with polynomials {f ς
are the associated polynomials of {fi}d+1
i=0 .
i }d+1
i=0 , where {f ς
i }d+1
i=0
(ii) (X −1)i0 =
cH
τi(θi)ηd−i(θi)Xdi
for 0 ≤ i ≤ d, where H is the connection coefficient matrix of
{fi}d+1
i=0 and cH is from Notation 14.2.
Proof: Similar to the proof of Proposition 16.1.
✷
In the next section we return to our discussion of TH systems.
17 The transition matrices P, P and their Vandermonde
structures
We return our attention to TH systems. Let Φ denote a TH system. Recall the transition
matrices P and P of Φ from Definition 11.1 and Definition 11.8. We will show that each of
P, P has a west Vandermonde structure and a south Vandermonde structure. We start by
associating with Φ a graded sequence of polynomials.
Definition 17.1. Let Φ = (A; {Ei}d
i=0
denote a Φ-standard basis for V and let H ∈ Matd+1(K) denote the matrix representing A
with respect to {vi}d
i=0 denote the polynomials
of H from Definition 14.3, so that si(A)v0 = vi for 0 ≤ i ≤ d by Lemma 14.6 and sd+1 is the
minimal polynomial of A by Corollary 14.7.
i=0. Observe that H is Hessenberg. Let {si}d+1
i=0) denote a TH system on V . Let {vi}d
i=0; A∗; {E∗
i }d
The following normalization of the {si}d+1
i=0 will be useful.
Definition 17.2. With reference to Definition 17.1, let {ti}d+1
nomials in K[λ] that satisfies (i), (ii) below.
i=0 denote the sequence of poly-
(i) For 0 ≤ i ≤ d, ti = si/ℓi where ℓi is from Definition 8.1.
(ii) td+1 = sd+1.
We will show in Corollary 17.10 that ti(θd) = 1 for 0 ≤ i ≤ d.
In Definition 17.1 we saw how the polynomials {si}d+1
i=0 arise naturally from the action
of A on a Φ-standard basis for V . We now discuss the meaning of the polynomials {ti}d+1
i=0
i=0 denote the inverted dual of a Φ-standard basis for V .
from this point of view. Let {ui}d
By Corollary 10.3, {ℓiui}d
i=0 is a Φ-standard basis for V , where ℓi is from Definition 8.1.
Therefore by Definition 17.2, ti(A)u0 = ui for 0 ≤ i ≤ d and td+1 is the minimal polynomial
of A.
Our next goal is to show that the polynomials {si}d+1
i=0 are associated in the
sense of Definition 15.4. We will use the following fact. Let {vi}d
i=0 denote a basis for V
and let R ∈ End(V ). Let S ∈ Matd+1(K) denote the matrix representing R with respect to
i=0. By elementary linear algebra, St is the matrix representing Rσ with respect to the
{vi}d
i=0, where σ : End(V ) → End( V ) is the canonical anti-isomorphism from above
dual of {vi}d
Definition 6.12. Therefore the matrix Sς represents Rσ with respect to the inverted dual of
{vi}d
i=0 and {ti}d+1
i=0.
36
Lemma 17.3. With reference to Definition 17.1 and Definition 17.2, for each column of the
table below, the two graded sequences of polynomials are associated in the sense of Definition
15.4.
{si}d+1
i=0
{ti}d+1
i=0
{s∗
{t∗
i }d+1
i=0
i }d+1
i=0
{si}d+1
i=0
{ti}d+1
i=0
{s∗
{t∗
i }d+1
i=0
i }d+1
i=0
i=0. By Definition 17.1, {si}d+1
i=0 denote the inverted dual of {vi}d
Proof: Let {vi}d
i=0 denote a Φ-standard basis for V and let H ∈ Matd+1(K) denote the matrix
representing A with respect to {vi}d
i=0 are the polynomials of H.
Let {wi}d
i=0. Applying the comments below Definition
17.2 to Φ, we find ti(Aσ)w0 = wi for 0 ≤ i ≤ d and td+1 is the minimal polynomial of Aσ.
Moreover by the comments above the present lemma, the matrix H ς represents Aσ with
i=0. Now by Lemma 14.6 the {ti}d+1
respect to {wi}d
i=0 are the polynomials of H ς. Therefore
{si}d+1
i=0 are associated by Definition 15.4. We have verified our assertions about
the first column of the above table. Our assertions about the remaining columns follow from
Definition 7.2.
✷
i=0 and {ti}d+1
We recall some elementary linear algebra. Let {ui}d
i=0 and {vi}d
Let T ∈ Matd+1(K) denote the transition matrix from {ui}d
and let S ∈ Matd+1(K) denote the matrix that represents A with respect to {ui}d
the matrix T −1ST represents A with respect to {vi}d
i=0 to {vi}d
i=0.
i=0 denote bases for V .
i=0. Pick A ∈ End(V )
i=0. Then
We now display a west Vandermonde structure for P .
i=0 and dual
i=0. Let P denote the transition matrix of Φ from Definition 11.1.
i=0) is a west Vandermonde system, and the corresponding polynomials are the
i=0 from Definition 17.1. For each relative of P we display a west Vandermonde system
Proposition 17.4. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
Then (P, {θi}d
{si}d+1
along with the corresponding polynomials.
i }d
west Vandermonde system corresponding polynomials
(P, {θi}d
i=0)
(P ∗, {θ∗
i }d
i=0)
( P , {θd−i}d
i=0)
( P ∗, {θ∗
d−i}d
i=0)
{si}d+1
i=0
i }d+1
{s∗
i=0
{si}d+1
i=0
i }d+1
{s∗
i=0
i }d
i=0; A∗; {E∗
Proof: Write Φ = (A; {Ei}d
i=0) and assume V is the vector space underlying Φ.
Let H ∈ Matd+1(K) (resp. D ∈ Matd+1(K)) denote the matrix representing A with respect
to a Φ-standard (resp. Φ∗-standard) basis for V . By construction H is Hessenberg and
multiplicity-free with an ordering of the eigenvalues {θi}d
i=0. By construction D is diagonal
with (i, i)-entry θi for 0 ≤ i ≤ d. By Definition 17.1 {si}d+1
i=0 are the polynomials of H. By
Definition 11.1 and the comments above this proposition, we have H = P −1DP . Therefore
by Lemma 15.2 (P, {θi}d
i=0 . We
have verified our assertions about the first row of the above table. Our assertions about the
remaining rows follow from Corollary 7.1.
✷
i=0) is a west Vandermonde system with polynomials {si}d+1
We now display a south Vandermonde structure for P .
37
Proposition 17.5. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
Then (P, {θ∗
the {t∗
system along with the corresponding polynomials.
i=0 and dual
i=0. Let P denote the transition matrix of Φ from Definition 11.1.
i=0) is a south Vandermonde system, and the corresponding polynomials are
i=0 from Definition 17.2. For each relative of P we display a south Vandermonde
i }d+1
i }d
i }d
south Vandermonde system corresponding polynomials
(P, {θ∗
i }d
(P ∗, {θi}d
( P , {θ∗
d−i}d
( P ∗, {θd−i}d
i=0)
i=0)
i=0)
i=0)
i }d
{t∗
i }d+1
i=0
{ti}d+1
i=0
i }d+1
{t∗
i=0
{ti}d+1
i=0
i }d
i }d+1
i=0 . Thus by Proposition 16.1 and Lemma 17.3, ((P ∗)−1, {θ∗
Proof: By Proposition 17.4 (P ∗, {θ∗
i=0) is a west Vandermonde system with polynomials
{s∗
i=0) is a south Vander-
monde system with polynomials {t∗
i=0) is a
south Vandermonde system with polynomials {t∗
i }d+1
i=0)
is a south Vandermonde system with polynomials {t∗
i=0 . We have verified our assertions
about the first row of the above table. Our assertions about the remaining rows follow from
Corollary 7.1.
✷
i=0 . By this and since P P ∗ = νI, (ν−1P, {θ∗
i=0 . Therefore by Lemma 13.8 (P, {θ∗
i }d+1
i }d+1
i }d
i }d
We now turn to the matrix P. Below we display a west Vandermonde structure and a
south Vandermonde structure for P. We begin with the west Vandermonde structure.
Corollary 17.6. Let Φ denote a TH system with eigenvalue sequence {θi}d
i=0 and dual eigen-
value sequence {θ∗
i=0. Let P denote the transition matrix of Φ from Definition 11.8. Then
i=0) is a west Vandermonde system, and the corresponding polynomials are the {ti}d+1
(P, {θi}d
i=0
from Definition 17.2. For each relative of P we display a west Vandermonde system along
with the corresponding polynomials.
i }d
west Vandermonde system corresponding polynomials
(P, {θi}d
(P ∗, {θ∗
i }d
( P, {θd−i}d
( P ∗, {θ∗
d−i}d
i=0)
i=0)
i=0)
i=0)
{ti}d+1
i=0
i }d+1
{t∗
i=0
{ti}d+1
i=0
{t∗
i }d+1
i=0
Proof: Routine by Lemma 13.9 and Proposition 17.4.
We now display a south Vandermonde structure for P.
✷
i=0 and dual eigen-
i=0. Let P denote the transition matrix of Φ from Definition 11.8. Then
i=0) is a south Vandermonde system, and the corresponding polynomials are the
i=0 from Definition 17.2. For each relative of P we display a south Vandermonde sys-
Corollary 17.7. Let Φ denote a TH system with eigenvalue sequence {θi}d
value sequence {θ∗
(P, {θ∗
{t∗
i }d+1
tem along with the corresponding polynomials.
i }d
i }d
south Vandermonde system corresponding polynomials
(P, {θ∗
i }d
(P ∗, {θi}d
( P, {θ∗
d−i}d
( P ∗, {θd−i}d
i=0)
i=0)
i=0)
i=0)
38
{t∗
i }d+1
i=0
{ti}d+1
i=0
i }d+1
{t∗
i=0
{ti}d+1
i=0
Proof: Similar to the proof of Corollary 17.6 using Proposition 17.5.
✷
We have now displayed the west Vandermonde and south Vandermonde structures for P
and P. As corollaries to these results, we now obtain some facts involving the polynomials
{si}d+1
i=0 and {ti}d+1
i=0 .
Corollary 17.8. Let Φ denote a TH system with eigenvalue sequence {θi}d
i=0 and dual eigen-
value sequence {θ∗
i=0. Let P denote the transition matrix of Φ from Definition 11.1 and let
P denote the corresponding matrix from Definition 11.8. Let {si}d+1
i=0 ) denote
the polynomials of Φ from Definition 17.1 (resp. Definition 17.2). Then the following (i),
(ii) hold for 0 ≤ i, j ≤ d.
i=0 (resp. {ti}d+1
i }d
(i) Pij = ℓjtj(θi) = ℓjt∗
d−i(θ∗
j ) = sj(θi) = ℓj s∗
d−i(θ∗
j )/ℓ∗
d−i.
(ii) Pij = tj(θi) = t∗
d−i(θ∗
j ) = sj(θi)/ℓj = s∗
d−i(θ∗
j )/ℓ∗
d−i.
Here ℓj, ℓ∗
d−i are from Definition 8.1 and Lemma 8.2 respectively.
Proof: (i) Using Corollary 11.3, (38), and Proposition 17.4, we find Pij = sj(θi). Similarly
using Proposition 17.5 in place of Proposition 17.4 we find Pij = ℓjt∗
j ). The remaining
assertions follow using Definition 17.2.
(ii) Use (i) and the fact that Pij = Pijℓj for 0 ≤ i, j ≤ d.
d−i(θ∗
✷
We emphasize one aspect of Corollary 17.8 which is telling us that the {si}d+1
i=0 each satisfy a variation on the Askey-Wilson duality [3, Theorems 14.7 -- 14.9].
{ti}d+1
i=0 and the
Corollary 17.9. Let Φ denote a TH system with eigenvalue sequence {θi}d
i=0 and dual eigen-
value sequence {θ∗
i=0 ) denote the corresponding polynomials
from Definition 17.1 (resp. Definition 17.2). Then the following (i), (ii) hold for 0 ≤ i, j ≤ d.
i=0 (resp. {ti}d+1
i=0. Let {si}d+1
i }d
(i) tj(θi) = t∗
d−i(θ∗
j ).
(ii) sj(θi)/ℓj = s∗
d−i(θ∗
j )/ℓ∗
d−i, where ℓj, ℓ∗
d−i are from Definition 8.1 and Lemma 8.2 respec-
tively.
We have a comment on how the polynomials {si}d+1
i=0 and {ti}d+1
i=0 are normalized.
Corollary 17.10. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
als from Definition 17.1 (resp. Definition 17.2). Then the following (i), (ii) hold.
i=0 and dual
i=0 ) denote the corresponding polynomi-
i=0 (resp. {ti}d+1
i=0. Let {si}d+1
i }d
(i) si(θd) = ℓi for 0 ≤ i ≤ d, where ℓi is from Definition 8.1.
(ii) ti(θd) = 1 for 0 ≤ i ≤ d.
Proof: Use Corollary 11.3 and Corollary 17.8(i).
The polynomials {ti}d+1
i=0 are not orthogonal in general; however we do have the following.
✷
Corollary 17.11. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
tion 17.2. Then the following (i), (ii) hold.
i=0 and dual
i=0 denote the corresponding polynomials from Defini-
i=0. Let {ti}d+1
i }d
39
(i)
(ii)
d
Xn=0
d
Xi=0
ti(θn)tj(θn)ℓ∗
n = δi+j,d νℓ−1
i
(0 ≤ i, j ≤ d).
ti(θm)td−i(θn)ℓi = δmn ν(ℓ∗
m)−1
(0 ≤ m, n ≤ d).
Here {ℓi}d
Definition 9.1.
i=0, {ℓ∗
i }d
i=0 are from Definition 8.1 and Lemma 8.2 respectively, and ν is from
n=0 P ∗
d−j,n using Corollary 17.8 to obtain Pd
Proof: (i) Let P denote the transition matrix of Φ from Definition 11.1. In the equation
d−j,nPni = δi+j,dν. In
tj(θn)ℓiti(θn) =
n=0 ℓ∗
P ∗P = νI, compare the (d − j, i)-entry of each side to obtain Pd
this equation, evaluate Pni and P ∗
δi+j,dν. The result follows.
(ii) Let P denote the transition matrix of Φ from Definition 11.1. In the equation P P ∗ = νI,
In this equation,
td−i(θn) = δmnν. The
✷
evaluate Pmi and P ∗
result follows.
We now give an analogue of Corollary 17.11 that applies to the polynomials {si}d+1
i=0 .
compare the (m, n)-entry of each side to obtain Pd
in using Corollary 17.8 to obtain Pd
i=0 ℓiti(θm)ℓ∗
in = δmnν.
i=0 PmiP ∗
n
n
Corollary 17.12. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
tion 17.1. Then the following (i), (ii) hold.
i=0 and dual
i=0 denote the corresponding polynomials from Defini-
i=0. Let {si}d+1
i }d
si(θn)sj(θn)ℓ∗
n = δi+j,d ν ℓj
(0 ≤ i, j ≤ d).
(i)
d
Xn=0
d
si(θm)sd−i(θn)(ℓd−i)−1 = δmn ν(ℓ∗
m)−1
(0 ≤ m, n ≤ d).
(ii)
Xi=0
Here {ℓi}d
i=0, {ℓ∗
i }d
i=0 are from Lemma 8.2 and ν is from Definition 9.1.
Proof: Use Definition 17.2(i) and Corollary 17.11.
We now express the polynomials {ti}d+1
To do this we will use the two-variable polynomial p of Φ from Definition 11.5.
i=0 and {si}d+1
i=0 in terms of the parameter array of Φ.
✷
Corollary 17.13. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
tion 17.2. Then the following (i), (ii) hold.
i=0 and dual
i=0 denote the corresponding polynomials from Defini-
i=0. Let {ti}d+1
i }d
(i) For 0 ≤ i ≤ d, ti = p(λ, θ∗
i ) where p is from Definition 11.5.
(ii) td+1 = Qd
i=0(λ − θi).
Proof: Let P denote the transition matrix of Φ from Definition 11.8. Since Pij = p(θi, θ∗
j )
for 0 ≤ i, j ≤ d, we find that (P, {θi}d
i=0) is a west Vandermonde system with polynomials
{fi}d+1
i=0(λ − θi). The result follows by
Corollary 17.6.
✷
i ) for 0 ≤ i ≤ d and fd+1 = Qd
i=0 where fi = p(λ, θ∗
40
Example 17.14. With reference to Definition 17.2, assume d = 2. Then
t0 = 1,
t1 = 1 +
t2 = 1 +
(λ − θ2)(θ∗
1 − θ∗
0)
φ1
(λ − θ2)(θ∗
2 − θ∗
0)
φ1
,
+
(λ − θ2)(λ − θ1)(θ∗
φ1φ2
2 − θ∗
0)(θ∗
2 − θ∗
1)
.
Corollary 17.15. Let Φ denote a TH system with eigenvalue sequence {θi}d
eigenvalue sequence {θ∗
tion 17.1. Then the following (i), (ii) hold.
i=0 and dual
i=0 denote the corresponding polynomials from Defini-
i=0. Let {si}d+1
i }d
(i) For 0 ≤ i ≤ d, si = ℓip(λ, θ∗
i ) where ℓi is from Definition 8.1 and p is from Definition
11.5.
(ii) sd+1 = Qd
i=0(λ − θi).
Proof: Use Definition 17.2 and Corollary 17.13.
✷
Remark 17.16. In view of Corollary 17.13, one may wonder about the polynomial p(θi, λ).
By Lemma 12.3 and Corollary 17.13, p(θi, λ) = p∗(λ, θi) = t∗
d−i for 0 ≤ i ≤ d.
i=0; A∗; {E∗
i }d
We now give the results promised at the end of Section 10. Let Φ = (A; {Ei}d
i=0)
denote a TH system on V . Let 0 6= ξ0 ∈ E0V and recall the Φ-standard basis {E∗
i ξ0}d
i=0 for
ξ∗
V from above (3). Let 0 6= ξ∗
i=0 for V
d}d
from above Proposition 10.5. These two bases are related as follows.
V and recall the Φ∗-standard basis {Eσ
d ∈ E∗σ
d−i
d
Proposition 17.17. With reference to the TH system Φ in Definition 6.12, let 0 6= ξ0 ∈ E0V
and 0 6= ξ∗
V . Then for 0 ≤ i, j ≤ d,
d ∈ E∗σ
d
hE∗
i ξ0, Eσ
j
ξ∗
di = ν−1ℓiℓ∗
j ti(θj)hξ0, ξ∗
di.
Here ℓi, ℓ∗
9.1 and Definition 17.2 respectively.
j are from Definition 8.1 and Lemma 8.2 respectively, and ν, ti are from Definition
Proof: Let P denote the transition matrix of Φ from Definition 11.1. Let ξ∗
observe by Lemma 9.2 that
0 = E∗
0ξ0 and
E0ξ∗
0 = E0E∗
0 ξ0 = E0E∗
0E0ξ0 = ν−1E0ξ0 = ν−1ξ0.
(43)
We may now argue
d
hE∗
i ξ0, Eσ
j
ξ∗
di =
ξ∗
di
j
0, Eσ
PnihEnξ∗
Xn=0
0, ξ∗
j PjihE0ξ∗
di
0, ξ∗
j si(θj)hE0ξ∗
di
j si(θj)hξ0, ξ∗
di
j ti(θj)hξ0, ξ∗
di
= ℓ∗
= ℓ∗
= ν−1ℓ∗
= ν−1ℓiℓ∗
41
(by Definition 11.1)
(by Proposition 10.5)
(by Corollary 17.8)
(by (43))
(by Definition 17.2).
Let Φ = (A; {Ei}d
i=0) denote a TH system on V . Let 0 6= ξ∗
i=0; A∗; {E∗
i }d
i=0 for V from above Proposition 10.5. Let 0 6= ξd ∈ Eσ
recall the Φ∗-standard basis {Eiξ∗
0}d
and recall the Φ-standard basis {E∗σ
bases are related as follows.
✷
0V and
V
i=0 for V from above Proposition 10.2. These two
0 ∈ E∗
ξd}d
d−i
d
Proposition 17.18. With reference to the TH system Φ in Definition 6.12, let 0 6= ξ∗
and 0 6= ξd ∈ Eσ
V . Then for 0 ≤ i, j ≤ d,
0 ∈ E∗
0V
d
hEiξ∗
0, E∗σ
j
ξdi = ν−1ℓ∗
i ℓjt∗
i (θ∗
j )hξ∗
0, ξdi.
i , ℓj are from Lemma 8.2 and Definition 8.1 respectively, and ν, t∗
i are from Definition
Here ℓ∗
9.1 and Definition 17.2 respectively.
Proof: Apply Proposition 17.17 to Φ∗.
✷
18 TH systems and Vandermonde systems
In the previous sections we discussed TH systems and Vandermonde systems. In this section
we give a natural correspondence between these two objects.
Definition 18.1. A matrix X ∈ Matd+1(K) is called west-south (or double) Vandermonde
whenever X is both west Vandermonde and south Vandermonde. Assume X is west-south
Vandermonde. We say that X is west (resp. south) normalized whenever Xi0 = 1 (resp.
Xdi = 1) for 0 ≤ i ≤ d. We say that X is normalized whenever it is both west normalized
and south normalized.
Definition 18.2. By a west-south (or double) Vandermonde system in Matd+1(K), we
i }d
mean a sequence (X, {θi}d
i=0) is a west Vandermonde sys-
tem in Matd+1(K) and (X, {θ∗
i=0) is a south Vandermonde system in Matd+1(K). Let
(X, {θi}d
i=0) denote a west-south Vandermonde system. Observe that X is west-
south Vandermonde. We say that (X, {θi}d
i=0) is west normalized (resp. south nor-
malized) (resp. normalized) whenever X is west normalized (resp. south normalized) (resp.
normalized) in the sense of Definition 18.1.
i=0) such that (X, {θi}d
i=0, {θ∗
i }d
i=0, {θ∗
i=0, {θ∗
i }d
i }d
Our main goal in this section is to establish a bijection between the following two sets:
the set of isomorphism classes of TH systems over K of diameter d,
the set of normalized west-south Vandermonde systems in Matd+1(K).
(44)
(45)
i }d
To do this we define a map ρ from (44) to (45) and a map χ from (45) to (44), and show
that they are inverses of each other. We start with an observation. Let Φ denote a TH
system over K of diameter d, with eigenvalue sequence {θi}d
i=0 and dual eigenvalue sequence
{θ∗
i=0. Let P denote the transition matrix of Φ from Definition 11.8. By Corollary 17.6
the sequence (P, {θi}d
i=0) is a west Vandermonde system in Matd+1(K), and by Corollary
17.7 the sequence (P, {θ∗
i=0) is a south Vandermonde system in Matd+1(K). By Corollary
11.10, P is both west normalized and south normalized. Therefore (P, {θi}d
i=0) is a
normalized west-south Vandermonde system in Matd+1(K).
i=0, {θ∗
i }d
i }d
42
Definition 18.3. We define a map ρ from (44) to (45). We do this as follows. Let Φ denote
a TH system over K of diameter d, with eigenvalue sequence {θi}d
i=0 and dual eigenvalue
sequence {θ∗
i=0. Let P denote the transition matrix of Φ from Definition 11.8. By the above
comment (P, {θi}d
i=0) is a normalized west-south Vandermonde system in Matd+1(K).
The map ρ sends the isomorphism class of Φ to (P, {θi}d
i=0, {θ∗
i }d
i }d
i=0, {θ∗
i }d
i=0).
i=0)
i=0, {θ∗
i=0, {vi}d
i=0 to {vi}d
i ). Now define Φ = (A; {Ei}d
i vi for 0 ≤ i ≤ d. For 0 ≤ i ≤ d let Ei (resp. E∗
i=0 for V such that X is the transition matrix from {ui}d
We now define the map χ. We start with the following construction. Let (X, {θi}d
i }d
denote a normalized west-south Vandermonde system in Matd+1(K). We construct a TH sys-
tem Φ on V as follows. Recall that X is invertible by Lemma 13.13. Therefore there exist
bases {ui}d
i=0.
Define A ∈ End(V ) such that Aui = θiui for 0 ≤ i ≤ d. Define A∗ ∈ End(V ) such that
A∗vi = θ∗
i ) denote the primitive idempotent of
A (resp. A∗) corresponding to θi (resp. θ∗
i=0). We
claim that Φ is a TH system on V . To prove the claim we show that Φ satisfies conditions
(i) -- (v) in Definition 2.2. Conditions (i) -- (iii) hold by construction. Let D∗ ∈ Matd+1(K)
denote the diagonal matrix with (i, i)-entry θ∗
i for 0 ≤ i ≤ d. Observe that D∗ represents A∗
i=0. Hence by the comment above Proposition 17.4, the matrix XD∗X −1
with respect to {vi}d
represents A∗ with respect to {ui}d
i=0) is a south Vandermonde
system, XD∗X −1 is Hessenberg by Corollary 15.6. By these comments, condition (iv) holds.
Let D ∈ Matd+1(K) denote the diagonal matrix with (i, i)-entry θi for 0 ≤ i ≤ d. Observe
that D represents A with respect to {ui}d
i=0. Hence by the comment above Proposition 17.4,
the matrix X −1DX represents A with respect to {vi}d
i=0) is a
west Vandermonde system, X −1DX is Hessenberg by Corollary 15.3. By these comments,
condition (v) holds. Therefore Φ is a TH system on V . By construction Φ has eigenvalue
sequence {θi}d
i=0. Moreover since (X, {θi}d
i=0 and dual eigenvalue sequence {θ∗
i=0. Moreover since (X, {θ∗
i=0; A∗; {E∗
i=0.
i }d
i }d
i }d
i=0, {θ∗
Definition 18.4. We define a map χ from (45) to (44). We do this as follows. Let
(X, {θi}d
i=0) denote a normalized west-south Vandermonde system in Matd+1(K). Let
Φ denote the corresponding TH system constructed above. The map χ sends (X, {θi}d
i }d
to the isomorphism class of Φ.
i=0, {θ∗
i }d
i=0)
Our next goal is to show that the maps ρ and χ are inverses of each other. We first recall
some elementary linear algebra. Let {ui}d
i=0 denote bases for V . Let
T ∈ Matd+1(K) denote the transition matrix from {ui}d
i=0 and let S ∈ Matd+1(K)
denote the transition matrix from {vi}d
i=0. Then T S is the transition matrix from
{ui}d
i=0 to {wi}d
i=0 to {vi}d
i=0, {wi}d
i=0, {vi}d
i=0 to {wi}d
i=0.
Lemma 18.5. Let (X, {θi}d
i=0) denote a normalized west-south Vandermonde system
in Matd+1(K) and let Φ denote the TH system constructed above Definition 18.4. Then X
is the transition matrix of Φ from Definition 11.8.
i=0, {θ∗
i }d
Proof: Let P denote the transition matrix of Φ from Definition 11.8. We show that P = X.
In what follows we refer to the construction of Φ above Definition 18.4. By the construction
of A (resp. A∗) we find that ui ∈ EiV (resp. vi ∈ E∗
i V ) for 0 ≤ i ≤ d. Recall that X is
the transition matrix from {ui}d
i=0. By these comments, Definition 11.8, and the
comment above this lemma, there exist invertible diagonal matrices D1, D2 ∈ Matd+1(K)
such that P = D1XD2. The matrices P and X are west normalized, meaning that Pi0 =
i=0 to {vi}d
43
Xi0 = 1 for 0 ≤ i ≤ d. The matrices P and X are south normalized, meaning that
Pdi = Xdi = 1 for 0 ≤ i ≤ d. Evaluating the equation P = D1XD2 using these comments,
we find that D1 is a nonzero scalar multiple of the identity and D2 is the inverse of D1.
Therefore P = X.
✷
Theorem 18.6. The map ρ from Definition 18.3 and the map χ from Definition 18.4 are
inverses of each other. Moreover each of ρ, χ is bijective.
i }d
i }d
i }d
i }d
i=0, {θ∗
i=0, {θ∗
i=0, {θ∗
i=0 and dual eigenvalue sequence {θ∗
Proof: We show that ρ ◦ χ is the identity map on (45) and χ ◦ ρ is the identity map on
(44). We first show that ρ ◦ χ is the identity map on (45). Let (X, {θi}d
i=0) denote a
normalized west-south Vandermonde system in Matd+1(K). Let Φ denote the corresponding
TH system constructed above Definition 18.4. The map χ sends (X, {θi}d
i=0) to the
isomorphism class of Φ. Recall from above Definition 18.4 that Φ has eigenvalue sequence
{θi}d
i=0. By Lemma 18.5 X is the transition matrix
of Φ from Definition 11.8. By these comments and Definition 18.3 the map ρ sends the
isomorphism class of Φ to (X, {θi}d
i=0). Therefore ρ ◦ χ is the identity map on (45).
Next we show that χ ◦ ρ is the identity map on (44). Let Φ denote a TH system over K
of diameter d, with eigenvalue sequence {θi}d
i=0. Let P
denote the transition matrix of Φ from Definition 11.8. The map ρ sends the isomorphism
class of Φ to (P, {θi}d
i=0) to the isomorphism
class of Φ′, where Φ′ is the corresponding TH system constructed above Definition 18.4.
Recall from above Definition 18.4 that Φ′ has eigenvalue sequence {θi}d
i=0 and dual eigenvalue
sequence {θ∗
i=0. We show that Φ and Φ′ are isomorphic. To do this we will invoke Lemma
3.8. Write Φ = (A; {Ei}d
i=0). Let V (resp.
V ′) denote the vector space underlying Φ (resp. Φ′). Let 0 6= ξ∗
0 V (resp. 0 6=
0 V ′) and recall the Φ∗-standard basis (resp. Φ∗′-standard basis) {Eiξ∗
0 ∈ E∗′
ξ∗′
i=0 (resp.
{E′
i=0) for V (resp. V ′) from above Proposition 10.5. Let Γ : V → V ′ denote the
0 }d
K-vector space isomorphism which sends Eiξ∗
i=0 and dual eigenvalue sequence {θ∗
i=0). The map χ sends (P, {θi}d
i=0) and Φ′ = (A′; {E′
0 for 0 ≤ i ≤ d. We show that
i=0; A∗′; {E∗′
i=0; A∗; {E∗
0 ∈ E∗
i=0, {θ∗
i=0, {θ∗
0 to E′
i }d
iξ∗′
iξ∗′
0}d
i }d
i }d
i }d
i }d
i }d
i}d
A′Γ = ΓA,
A∗′Γ = ΓA∗,
E′
iΓ = ΓEi,
E∗′
i Γ = ΓE∗
i
(0 ≤ i ≤ d).
(46)
iE′
iξ∗′
jξ∗′
iΓEjξ∗
0 = E′
i Γ = ΓE∗
0 = δijE′
0 and ΓEiEjξ∗
A′ = Pd
0}d
0 = δijΓEiξ∗
i since Φ′ has eigenvalue sequence {θi}d
iΓ and ΓEi agree at each vector in the Φ∗-standard basis {Ejξ∗
iΓ = ΓEi for 0 ≤ i ≤ d. Let i be given. In order to show that E′
We first show that E′
we show that E′
that for 0 ≤ j ≤ d, E′
Thus E′
iΓ = ΓEi. Next we show that A′Γ = ΓA. Recall A = Pd
i=0 θiE′
iΓ = ΓEi,
j=0. Observe
0 = δijE′
iξ∗′
0 .
i=0 θiEi. Observe that
i=0. By these comments A′Γ = ΓA.
Next we show that E∗′
i for 0 ≤ i ≤ d. Let P (resp. P ′) denote the transition matrix
of Φ (resp. Φ′) from Definition 11.1, and let L (resp. L′) denote the matrix associated with
Φ (resp. Φ′) from Definition 8.5. Observe that L = L′ by Definition 8.1, since Φ and Φ′
have the same dual eigenvalue sequence {θ∗
i=0. By Lemma 18.5 P is the transition matrix
of Φ′ from Definition 11.8. By these comments and Definition 11.8, we have P = P ′. Let
0 6= ξ0 ∈ E0V (resp. 0 6= ξ′
0V ′) such that ξ∗
0). Recall
the Φ-standard basis (resp. Φ′-standard basis) {E∗
i=0) for V (resp.
V ′) from above (3). By Definition 11.1 and since P = P ′, P is the transition matrix from
{Eiξ∗
i=0). We can now easily show that
0ξ0 (resp. ξ∗′
i ξ′
0 = E∗′
0}d
0 = E∗
i ξ0}d
i=0 (resp. {E∗′
i=0 (resp. {E∗′
i=0 (resp. {E′
i=0) to {E∗
0 ∈ E′
i ξ0}d
0 }d
0 ξ′
i ξ′
0}d
iξ∗′
i }d
0}d
44
E∗′
E∗′
i Γ = ΓE∗
i Γ and ΓE∗
i for 0 ≤ i ≤ d. Let i be given. In order to show that E∗′
j ξ0}d
i agree at each vector in the Φ-standard basis {E∗
i Γ = ΓE∗
j=0. For 0 ≤ j ≤ d,
i we show that
E∗′
i ΓE∗
ΓE∗
i E∗
d
i Γ
j ξ0 = E∗′
Xh=0
j ξ0 = δijΓE∗
PhjEhξ∗
0 = E∗′
i
d
j ξ0 = δijΓ
Xh=0
i Γ and ΓE∗
d
Xh=0
PhjE′
hξ∗′
0 = E∗′
i E∗′
j ξ′
0 = δijE∗′
j ξ′
0,
PhjEhξ∗
0 = δij
d
Xh=0
PhjE′
hξ∗′
0 = δijE∗′
j ξ′
0.
j ξ0}d
j=0. Therefore E∗′
We have now shown that E∗′
{E∗
i agree at each vector in the Φ-standard basis
i E∗
i .
i=0. By these
comments A∗′Γ = ΓA∗. We have now shown (46). Now Φ and Φ′ are isomorphic in view of
Lemma 3.8. Therefore χ ◦ ρ is the identity map on (44). The result follows.
✷
i . Next we show that A∗′Γ = ΓA∗. Recall A∗ = Pd
Observe that A∗′ = Pd
since Φ′ has dual eigenvalue sequence {θ∗
i Γ = ΓE∗
i=0 θ∗
i=0 θ∗
i E∗′
i
i }d
Combining Corollary 4.4 and Theorem 18.6, we get a bijection between any two of the
following three sets:
• The set of isomorphism classes of TH systems over K of diameter d.
• The set of normalized west-south Vandermonde systems in Matd+1(K).
• The set of parameter arrays over K of diameter d.
19 Reduced TH systems and Vandermonde matrices
In the previous section we explained how double Vandermonde systems correspond with TH
systems. In this section we turn our attention to double Vandermonde matrices and explain
how these correspond with objects called reduced TH systems.
Definition 19.1. A sequence ({Ei}d
system) on V whenever there exist A, A∗ ∈ End(V ) such that (A; {Ei}d
a TH system on V . Let Φ = (A; {Ei}d
({Ei}d
i=0) is called a reduced TH system (or RTH
i=0) is
i=0) denote a TH system on V . Then
i=0) is an RTH system on V , called the reduction of Φ.
i=0; A∗; {E∗
i=0; A∗; {E∗
i=0; {E∗
i=0; {E∗
i }d
i }d
i }d
i }d
i }d
i=0; {E∗
Definition 19.2. Let Λ = ({Ei}d
i=0) denote an RTH system on V . Let W denote
a vector space over K with dimension d + 1, and let Ω = ({Fi}d
i=0) denote an RTH
system on W . By an isomorphism of RTH systems from Λ to Ω we mean a K-algebra
isomorphism γ : End(V ) → End(W ) such that Fi = Eγ
for 0 ≤ i ≤ d. We
say that the RTH systems Λ and Ω are isomorphic whenever there exists an isomorphism of
RTH systems from Λ to Ω.
i and F ∗
i = E∗γ
i=0; {F ∗
i }d
i
Proposition 19.3. Let Φ and Φ′ denote TH systems over K. Then the following (i), (ii)
are equivalent.
(i) The reduction of Φ is isomorphic to the reduction of Φ′.
45
(ii) Φ is affine isomorphic to Φ′.
Proof: (i) ⇒ (ii) Let P (resp. P ′) denote the transition matrix of Φ (resp. Φ′) from Definition
11.1. From its definition, we see that P (resp. P ′) is determined by the primitive idempotents
of Φ (resp. Φ′). Hence by our assumption P = P ′. Let {θi}d
i=0) denote the
eigenvalue sequence of Φ (resp. Φ′). By Proposition 17.4 P and {θi}d
i=0 are compatible.
i=0 are compatible. Since P = P ′, we conclude that P is compatible
Similarly P ′ and {θ′
with each of {θi}d
i=0. Hence by Lemma 13.5 there exist α, β ∈ K with α 6= 0
such that θ′
i=0) denote the dual eigenvalue
sequence of Φ (resp. Φ′). By a similar argument, there exist α∗, β∗ ∈ K with α∗ 6= 0 such
that θ∗′
i = α∗θ∗
(ii) ⇒ (i) Clear.
i + β∗ for 0 ≤ i ≤ d. It follows that Φ is affine isomorphic to Φ′.
i = αθi + β for 0 ≤ i ≤ d. Let {θ∗
i=0 (resp. {θ∗′
i=0 (resp. {θ′
i }d
i }d
i}d
i=0 and {θ′
i}d
i}d
✷
Corollary 19.4. Let Λ and Ω denote isomorphic RTH systems over K. Then the isomor-
phism of RTH systems from Λ to Ω is unique.
Proof: Let γ and γ′ denote isomorphisms of RTH systems from Λ to Ω. We show that
γ = γ′. Let Φ (resp. Ψ) denote a TH system over K whose reduction is Λ (resp. Ω). By
Proposition 19.3 Φ is affine isomorphic to Ψ. In other words, Φ is isomorphic to an affine
transformation Ψ′ of Ψ. By construction Φ and Ψ′ have the same eigenvalue sequence and
dual eigenvalue sequence. By this and the comment (iv) above (1), we find that each of γ
and γ′ is an isomorphism of TH systems from Φ to Ψ′. Now γ = γ′ in view of Lemma 3.6.
The result follows.
✷
We now give a correspondence between TH systems and reduced TH systems.
Corollary 19.5. The map which sends a TH system to its reduction induces a bijection
from the set of affine isomorphism classes of TH systems over K to the set of isomorphism
classes of RTH systems over K.
Proof: Immediate from Proposition 19.3.
✷
We now turn our attention to double Vandermonde systems and double Vandermonde ma-
trices.
Lemma 19.6. Let Ω = (X, {θi}d
i=0) denote a normalized west-south Vandermonde
system in Matd+1(K). Let α, β, α∗, β∗ denote scalars in K with α, α∗ nonzero. Then the
sequence
i=0, {θ∗
i }d
i=0)
is a normalized west-south Vandermonde system in Matd+1(K).
(X, {αθi + β}d
i=0; {α∗θ∗
i + β∗}d
Proof: Routine by Lemma 13.5.
(47)
✷
Definition 19.7. Referring to Lemma 19.6, we call (47) the affine transformation of Ω
associated with α, β, α∗, β∗.
Definition 19.8. Let Ω and Ω′ denote normalized west-south Vandermonde systems in
Matd+1(K). We say that Ω and Ω′ are affine related whenever Ω is an affine transformation
of Ω′. Observe that the affine relation is an equivalence relation.
46
Lemma 19.9. Let Ω = (X, {θi}d
i=0) denote nor-
malized west-south Vandermonde systems in Matd+1(K). Then the following (i), (ii) are
equivalent.
i=0) and Ω′ = (X ′, {θ′
i=0, {θ∗′
i=0, {θ∗
i }d
i }d
i}d
(i) X = X ′.
(ii) Ω is affine related to Ω′.
Proof: (i) ⇒ (ii) Immediate from Lemma 13.5.
(ii) ⇒ (i) Clear.
✷
We now give a correspondence between normalized double Vandermonde systems and nor-
malized double Vandermonde matrices.
i=0, {θ∗
Corollary 19.10. The map which sends a normalized west-south Vandermonde system
(X, {θi}d
i=0) to the matrix X induces a bijection from the set of affine classes of
normalized west-south Vandermonde systems in Matd+1(K) to the set of normalized west-
south Vandermonde matrices in Matd+1(K).
i }d
Proof: Immediate from Lemma 19.9.
✷
Next we give a correspondence between affine isomorphism classes of TH systems and affine
classes of normalized double Vandermonde systems. Recall the map ρ from Definition 18.3.
Lemma 19.11. Let Φ = (A; {Ei}d
i=0) denote the image under ρ of the isomorphism class of Φ. Let α, β, α∗, β∗
(X, {θi}d
denote scalars in K with α, α∗ nonzero and consider the TH system (25). Then ρ sends the
isomorphism class of (25) to (X, {αθi + β}d
i=0) denote a TH system over K. Let
i=0; A∗; {E∗
i=0, {θ∗
i }d
i }d
i + β∗}d
i=0, {α∗θ∗
i=0).
Proof: Immediate from Lemma 5.4.
✷
Corollary 19.12. The bijection ρ from Definition 18.3 induces a bijection from the set of
affine isomorphism classes of TH systems over K of diameter d, to the set of affine classes
of normalized west-south Vandermonde systems in Matd+1(K).
Proof: Immediate from Lemma 19.11.
✷
We now bring the parameter arrays into the discussion.
Definition 19.13. We define a binary relation on the set of parameter arrays over K of diam-
i=0, {φ′
eter d. We do this as follows. Let p = ({θi}d
denote parameter arrays over K of diameter d. We say that p and p′ are affine related when-
ever there exist scalars α, β, α∗, β∗ in K with α, α∗ nonzero such that the following (i) -- (iii)
hold.
i=1) and p′ = ({θ′
i=0, {φi}d
i=0, {θ∗′
i=0, {θ∗
i }d
i }d
i}d
i}d
i=1)
(i) θ′
i = αθi + β
(0 ≤ i ≤ d).
(ii) θ∗′
i = α∗θ∗
i + β∗
(0 ≤ i ≤ d).
(iii) φ′
i = αα∗φi
(1 ≤ i ≤ d).
47
Observe that the affine relation is an equivalence relation. By a reduced parameter array we
mean an equivalence class of this relation.
Corollary 19.14. The bijection from Corollary 4.4 induces a bijection from the set of affine
isomorphism classes of TH systems over K of diameter d, to the set of reduced parameter
arrays over K of diameter d.
Proof: Immediate from Lemma 5.4.
✷
Combining Corollaries 19.5, 19.10, 19.12, 19.14, we get a bijection between any two of
the following five sets:
• The set of affine isomorphism classes of TH systems over K of diameter d.
• The set of isomorphism classes of RTH systems over K of diameter d.
• The set of affine classes of normalized west-south Vandermonde systems in Matd+1(K).
• The set of normalized west-south Vandermonde matrices in Matd+1(K).
• The set of reduced parameter arrays over K of diameter d.
20 Acknowledgement
This paper was written while the author was a graduate student at the University of
Wisconsin-Madison. The author would like to thank his advisor Paul Terwilliger for his
many valuable ideas and suggestions.
References
[1] A. Godjali. Thin Hessenberg pairs. Linear Algebra Appl. 432 (2010) 3231 -- 3249;
arXiv:math.RA/0911.4118v1.
[2] J. J. Rotman. Advanced modern algebra. Prentice Hall, Saddle River NJ 2002.
[3] P. Terwilliger. Leonard pairs and the q-Racah polynomials. Linear Algebra Appl. 387
(2004) 235 -- 276; arXiv:math.QA/0306301.
[4] P. Terwilliger. Two linear transformations each tridiagonal with respect to an eigenbasis
of the other. Linear Algebra Appl. 330 (2001) 149 -- 203; arXiv:math.RA/0406555.
[5] P. Terwilliger. An algebraic approach to the Askey scheme of orthogonal polynomials.
Orthogonal polynomials and special functions, 255 -- 330, Lecture Notes in Math., 1883,
Springer, Berlin, 2006; arXiv:math.QA/0408390.
48
Ali Godjali
Department of Mathematics
University of Wisconsin
Van Vleck Hall
480 Lincoln Drive
Madison, WI 53706-1388 USA
email: [email protected]
49
|
0912.3159 | 2 | 0912 | 2010-09-09T21:19:25 | Universal deformation formulas and braided module algebras | [
"math.RA",
"math.KT"
] | We study formal deformations of a crossed product $S(V)#_f G$, of a polynomial algebra with a group, induced from a universal deformation formula introduced by Witherspoon. These deformations arise from braided actions of Hopf algebras generated by automorphisms and skew derivations. We show that they are nontrivial in the characteristic free context, even if $G$ is infinite, by showing that their infinitesimals are not coboundaries. For this we construct a new complex which computes the Hochschild cohomology of $S(V)#_f G$. | math.RA | math |
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED
MODULE ALGEBRAS
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
Abstract. We study formal deformations of a crossed product S(V )#f G,
of a polynomial algebra with a group, induced from a universal deformation
formula introduced by Witherspoon. These deformations arise from braided
actions of Hopf algebras generated by automorphisms and skew derivations.
We show that they are nontrivial in the characteristic free context, even if G
is infinite, by showing that their infinitesimals are not coboundaries. For this
we construct a new complex which computes the Hochschild cohomology of
S(V )#f G.
Introduction
In [G-Z] Giaquinto and Zhang develop the notion of a universal deformation formula
based on a bialgebra H, extending earlier formulas based on universal enveloping
algebras of Lie algebras. Each one of these formulas is called universal because
it provides a formal deformation for any H-module algebra.
In the same paper
the authors construct the first family of such formulas based on noncommutative
bialgebras, namely the enveloping algebras of central extensions of a Heisenberg
Lie algebra L. Another of these formulas, based on a Hopf algebra Hq over C,
where q ∈ C× is a parameter, generated by group like elements σ±1 and two skew
primitive elements D1, D2, were obtained in the generic case by the same authors,
but were not published. In [W] the author generalizes this formula to include the
case where q is a root of unity, and she uses it to construct formal deformations of a
crossed product S(V )#f G, where S(V ) is the polynomial algebra and the group G
acts linearly on V . More precisely, she deals with deformations whose infinitesimal
sends V ⊗ V to S(V )wg, where g is a central element of G.
In this paper we prove that some results established in [W] under the hypothesis
that G is a finite group, remain valid for arbitrary groups, and with C replaced by
an arbitrary field. For instance we show that the determinant of the action of g
on V is always 1. Moreover, we do not only consider standard Hq-module algebra
structures on S(V )#f G, but also the more general ones introduced in [G-G1], and
we work with actions which depend on two central elements g1 and g2 of G and
two polynomials P1 and P2. When the actions are the standard ones, g1 = 1 and
P1 = 1, we obtain the case considered in [W]. Finally, in Subsection 3.2 we show
how to extend the explicit formulas obtained previously, to non central g1 and g2.
2000 Mathematics Subject Classification. Primary 16S80; Secondary 16S35.
Key words and phrases. Crossed product; Deformation; Hochschild cohomology.
Supported by UBACYT 095, PIP 112-200801-00900 (CONICET) and PUCP-DAI-2009-0042.
Supported by UBACYT 095 and PIP 112-200801-00900 (CONICET).
Supported by PUCP-DAI-2009-0042, Lucet 90-DAI-L005, SFB 478 U. Munster, Konrad Ade-
nauer Stiftung.
The second author thanks the appointment as a visiting professor "C´atedra Jos´e Tola Pasquel"
and the hospitality during his stay at the PUCP.
1
2
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
As was noted by Witherspoon, these formulas necessarily involve all components
of S(V )#f G corresponding to the elements of a union of conjugacy classes of G.
The paper is organized as follows: in the first section we review the concept of
braided module algebra introduced in [G-G1], we adapt the notion of Universal
Deformation Formula (UDF) to the braided context, and we show that each one
of these formulas produces a deformation on any braided H-module algebra whose
transposition (see Definition 1.6) satisfy a suitable hypothesis. We remark that,
when the bialgebra H is standard, the use of braided module algebra gives rise to
more deformations than the ones obtained using only module algebras, because the
transposition can be different from the flip. With this in mind, although we are
going to work with the standard Hopf algebra Hq, we establish the basic properties
of UDF's in the braided case, because it is the most appropriate setting to deal
with arbitrary transpositions. In the second section we recall the definitions of the
Hopf algebra Hq and of the UDF expq considered in [W, Section 3], which we are
going to study. We also introduce the concept of a good transposition of Hq on an
algebra A, and we study some of its properties. Perhaps the most important result
in this section is Theorem 2.4, in which we obtain a description of all the Hq-module
algebras (A, s), with s a good transposition. This is the first of several results in
which we give a systematic account of the necessary and sufficient conditions that
an algebra (in general a crossed product S(V )#f G) must satisfy in order to support
a braided Hq-module algebra structure satisfying suitable hypothesis. In Section 4
of [W], using the UDF expq the author constructs a large family of deformations
whose infinitesimal sends V ⊗ V to S(V )wg, where g is a central element of G.
Using cohomological methods she proves that if G is finite, these deformations are
non trivial, that the action of g on V has determinant 1 and that the codimension of
gV is 0 or 2. In the first part of Section 3 we study a larger family of deformations
and we prove that the last two results hold for this family even if G is infinite and
the characteristic of k is non zero. Finally, in Section 4 we show that, under very
general hypothesis, the deformations constructed in the previous section are non
trivial. Once again, we do not assume characteristic zero, nor that the group G
is finite. One of the interesting points in this paper is the method developed to
deal with the cohomology of S(V )#f G when k[G] is non semisimple. As far as we
know it is the first time that this type of cochain complexes is used to prove the
non triviality of a Hochschild cocycle.
1 Preliminaries
After introducing some basic notations we recall briefly the concepts of braided
bialgebra and braided Hopf algebra following the presentation given in [T1] (see
also [T2], [L1], [F-M-S], [A-S], [D], [So] and [B-K-L-T]). Then we review the notion
of braided module algebra introduced in [G-G1], we recall the concept of Universal
Deformation Formula based on a bialgebra H, due to Giaquinto an Zhang, and
we show that such a UDF produces a formal deformation when it is applied to an
H-braided module algebra, satisfying suitable hypothesis, generalizing slightly a
result in [G-Z].
In this paper k is a field, k× = k \ {0}, all the vector spaces are over k, and
⊗ = ⊗k. Moreover we will use the usual notation (i)q = 1 + q + · · · + qi−1 and
(i)!q = (1)q · · · (i)q, for q ∈ k× and i ∈ N.
Let V , W be vector spaces and let c : V ⊗ W → W ⊗ V be a k-linear map. Recall
that:
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
3
(η ⊗ W ) = W ⊗ η and c
- If V is an algebra, then c is compatible with the algebra structure of V
if c
(V ⊗ c), where
η : k → V and µ : V ⊗ V → V denotes the unit and the multiplication map
of V , respectively.
(µ ⊗ W ) = (W ⊗ µ)
(c ⊗ V )
- If V is a coalgebra, then c is compatible with the coalgebra structure of V
(∆ ⊗ W ), where
if (W ⊗ ǫ)
ǫ : V → k and ∆ : V → V ⊗ V denotes the counit and the comultiplication
map of V , respectively.
c = ǫ ⊗ W and (W ⊗ ∆)
c = (c ⊗ V )
(V ⊗ c)
Of course, there are similar compatibilities when W is an algebra or a coalgebra.
1.1 Braided bialgebras and braided Hopf algebras
Definition 1.1. A braided bialgebra is a vector space H endowed with an algebra
structure, a coalgebra structure and a braiding operator c ∈ Autk(H ⊗2) (called the
braid of H), such that c is compatible with the algebra and coalgebra structures of
H, ∆
(∆ ⊗ ∆), η is a coalgebra morphism and ǫ is an
algebra morphism. Furthermore, if there exists a k-linear map S : H → H, which
is the inverse of the identity map for the convolution product, then we say that H
is a braided Hopf algebra and we call S the antipode of H.
(H ⊗ c ⊗ H)
µ = (µ ⊗ µ)
Usually H denotes a braided bialgebra, understanding the structure maps, and
c denotes its braid. If necessary, we will use notations as cH , µH , etcetera.
Remark 1.2. Assume that H is an algebra and a coalgebra and c ∈ Autk(H ⊗2)
is a solution of the braiding equation, which is compatible with the algebra and
coalgebra structures of H. Let H ⊗c H be the algebra with underlying vector space
H ⊗2 and multiplication map given by µH⊗cH := (µ ⊗ µ)
(H ⊗ c ⊗ H). It is easy to
see that H is a braided bialgebra with braid c if and only if ∆ : H → H ⊗c H and
ǫ : H → k are morphisms of algebras.
Definition 1.3. Let H and L be braided bialgebras. A map g : H → L is a
morphism of braided bialgebras if it is an algebra homomorphism, a coalgebra ho-
momorphism and c
(g ⊗ g) = (g ⊗ g)
c.
Let H and L be braided Hopf algebras. It is well known that if g : H → L is a
morphism of braided bialgebras, then g
S = S
g.
1.2 Braided module algebras
Definition 1.4. Let H be a braided bialgebra. A left H-braided space (V, s) is a
vector space V , endowed with a bijective k-linear map s : H ⊗ V → V ⊗ H, which
is compatible with the bialgebra structure of H and satisfies
(s ⊗ H)
(H ⊗ s)
(c ⊗ V ) = (V ⊗ c)
(s ⊗ H)
(H ⊗ s)
(compatibility of s with the braid). Let (V ′, s′) be another left H-braided space.
A k-linear map f : V → V ′ is said to be a morphism of left H-braided spaces, from
(V, s) to (V ′, s′), if (f ⊗ H)
(H ⊗ f ).
s = s′
We let LBH denote the category of all left H-braided spaces. It is easy to check
that this is a monoidal category with:
- unit (k, τ ), where τ : H ⊗ k → k ⊗ H is the flip,
- tensor product (V, sV ) ⊗ (U, sU ) := (V ⊗ U, sV ⊗U ), where sV ⊗U is the map
sV ⊗U := (V ⊗ sU )
(sV ⊗ U ),
- the usual associativity and unit constraints.
4
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
Definition 1.5. We will say that (A, s) is a left H-braided algebra or simply a left
H-algebra if it is an algebra in LBH .
We let ALBH denote the category of left H-braided algebras.
Definition 1.6. Let A be an algebra. A left transposition of H on A is a bijective
map s : H ⊗ A → A ⊗ H, satisfying:
(1) (A, s) is a left H-braided space,
(2) s is compatible with the algebra structure of A.
Remark 1.7. A left H-braided algebra is a pair (A, s) consisting of an algebra A
and a left transposition s of H on A. Let (A′, s′) be another left H-braided algebra.
A map f : A → A′ is a morphism of left H-braided algebras, from (A, s) to (A′, s′),
if and only if it is a morphism of standard algebras and (f ⊗ H)
(H ⊗ f ).
s = s′
Note that (H, c) is an algebra in LBH . Hence, one can consider left and right
(H, c)-modules in this monoidal category.
Definition 1.8. We will say that (V, s) is a left H-braided module or simply a left
H-module to mean that it is a left (H, c)-module in LBH .
We let H (LBH ) denote the category of left H-braided modules.
Remark 1.9. A left H-braided space (V, s) is a left H-module if and only if V is a
standard left H-module and
s
(H ⊗ ρ) = (ρ ⊗ H)
(H ⊗ s)
(c ⊗ V ),
where ρ denotes the action of H on V . Furthermore, a map f : V → V ′ is a
morphism of left H-modules, from (V, s) to (V ′, s′), if and only if it is H-linear and
(f ⊗ H)
(H ⊗ f ).
s = s′
Given left H-modules (V, sV ) and (U, sU ), with actions ρV and ρU respectively,
we let ρV ⊗U denote the diagonal action
ρV ⊗U := (ρV ⊗ ρU )
(H ⊗ sV ⊗ U )
(∆H ⊗ V ⊗ U ).
The following proposition says in particular that (k, τ ) is a left H-module via the
trivial action and that (V, sV ) ⊗ (U, sU ) is a left H-module via ρV ⊗U .
Proposition 1.10 (see [G-G1]). The category H (LBH ), of left H-braided modules,
endowed with the usual associativity and unit constraints, is monoidal.
Definition 1.11. We say that (A, s) is a left H-braided module algebra or simply
a left H-module algebra if it is an algebra in H (LBH ).
We let H (ALBH ) denote the category of left H-braided module algebras.
Remark 1.12. (A, s) is a left H-module algebra if and only if the following facts
hold:
(1) A is an algebra,
(2) s is a left transposition of H on A,
(3) A is a standard left H-module,
(4) s
(5) µA
(6) h · 1 = ǫ(h)1 for all h ∈ H,
(H ⊗ ρ) = (ρ ⊗ H)
(H ⊗ s ⊗ A)
(H ⊗ s)
(ρ ⊗ ρ)
(c ⊗ A),
(∆H ⊗ A ⊗ A) = ρ
(H ⊗ µA),
where ρ denotes the action of H on A. So, (A, s) is a left H-module algebra if and
only if it is a left H-algebra, a left H-module and satisfies conditions (5) and (6).
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
5
In the sequel, given a map ρ : H ⊗ A → A, sometimes we will write h · a to
denotes ρ(h ⊗ a).
Remark 1.13. If X generates H as a k-algebra, then conditions (4), (5) and (6) of
the above remark are satisfied if and only if
s(h ⊗ l·a) = (ρ ⊗ H)
(H ⊗ s)
(c ⊗ A)(h ⊗ l ⊗ a),
h · (ab) = µA
h·1 = ǫ(h),
(ρ ⊗ ρ)
(H ⊗ s ⊗ A)(∆(h) ⊗ a ⊗ b),
for all a, b ∈ A and h, l ∈ X.
Let (A′, s′) be another left H-module algebra. A map f : A → A′ is a morphism
of left H-module algebras, from (A, s) to (A′, s′), if and only if it is an H-linear
morphism of standard algebras that satisfies (f ⊗ H)
(H ⊗ f ).
s = s′
1.3 Bialgebra actions and universal deformation formulas
Most of the results of [G-Z, Section 1] remain valid in our more general context,
with the same arguments and minimal changes. In particular Theorem 1.15 below
holds.
Let H be a braided bialgebra. Given a left H-module algebra (A, s) and an
element F ∈ H ⊗ H, we let Fl : A ⊗ A → A ⊗ A denote the map defined by
Fl(a ⊗ b) := (ρ ⊗ ρ)
(H ⊗ s ⊗ A)(F ⊗ a ⊗ b),
where ρ : H ⊗ A → A is the action of H on A. We let AF denote A endowed with
the multiplication map µA
Fl.
Definition 1.14. We say that F ∈ H ⊗ H is a twisting element (based on H) if
(1) (ǫ ⊗ id)(F ) = (id ⊗ǫ)(F ) = 1,
(2) [(∆ ⊗ id)(F )](F ⊗ 1) = [(id ⊗∆)(F )](1 ⊗ F ) in H ⊗c H ⊗c H,
(3) (c ⊗ H)
(H ⊗ c)(F ⊗ h) = h ⊗ F , for all h ∈ H.
Theorem 1.15. Let (A, s) be a left H-module algebra. If F ∈ H ⊗ H is a twisting
element such that (s ⊗ H)
(H ⊗ s)(F ⊗ a) = a ⊗ F , for all a ∈ A, then AF is an
associative algebra with unit 1A.
The notions of braided bialgebra, left H-braided module algebra and twisting
element make sense in arbitrary monoidal categories. Here we consider the monoi-
dal category M[[t]] defined as follows:
- the objects are the k[[t]]-modules of the form M [[t]] where M is a k-vector
space,
- the arrows are the k[[t]]-linear maps,
- the tensor product is the completation
of the algebraic tensor product M [[t]] ⊗k[[t]] N [[t]] with respect to the t-adic
topology,
M [[t]]b⊗k[[t]]N [[t]]
- the unit and the associativity constrains are the evident ones.
We identify M [[t]]b⊗k[[t]]N [[t]] with (M ⊗ N )[[t]] by the map
Θ : M [[t]]b⊗k[[t]]N [[t]] → (M ⊗ N )[[t]]
given by Θ(mti ⊗ ntj) := (m ⊗ n)ti+j .
6
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
If A is a k-algebra, then A[[t]] is an algebra in M[[t]] via the multiplication map
(A ⊗ A)[[t]]
P(ai ⊗ bi)ti
µ
,
/ A[[t]]
/P aibiti
where aibi = µA(ai ⊗ bi). The unit map is the canonical inclusion k[[t]] ֒→ A[[t]].
If H is a braided bialgebra over k, then H[[t]] is a braided bialgebra in M[[t]].
The multiplication and unit maps are as above. The comultiplication and counits
are the maps
H[[t]] ∆ /
P hiti
/ (H ⊗ H)[[t]]
/P ∆H (hi)ti
and the braid operator is the map
H[[t]]
P hiti
,
/ k[[t]]
/P ǫH(hi)ti
ǫ
.
and
c[[t]]
s[[t]]
ρ
(H ⊗ H)[[t]]
P(hi ⊗ li)ti
/ (H ⊗ H)[[t]]
/P cH (hi ⊗ li)ti
If (A, s) is a H-module algebra, then (A[[t]], s[[t]]), where s[[t]] is the map
(H ⊗ A)[[t]]
P(hi ⊗ ai)ti
is an H[[t]]-module algebra, via
,
/ (A ⊗ H)[[t]]
/P s(hi ⊗ ai)ti
(H ⊗ A)[[t]]
P(hi ⊗ ai)ti
.
/ A[[t]]
/P ρA(hi ⊗ ai)ti
satisfying conditions (1) -- (3) of Definition 1.14. It is easy to check that a power
A twisting element based on H[[t]] in M[[t]] is an element F ∈ H[[t]]b⊗k[[t]]H[[t]]
series F =P Fi ti ∈ (H ⊗ H)[[t]] corresponds via Θ−1 to a twisting element if and
only if
(1) (ǫ ⊗ id)(F0) = (id ⊗ǫ)(F0) = 1 and (ǫ ⊗ id)(Fi) = (id ⊗ǫ)(Fi) = 0 for i ≥ 1,
(2) For all n ≥ 0,
Xi+j=n
(∆ ⊗ id)(Fi)(Fj ⊗ 1) = Xi+j=n
(id ⊗∆)(Fi)(1 ⊗ Fj )
in H ⊗c H ⊗c H,
(3) (c ⊗ H)
(H ⊗ c)(Fn ⊗ h) = h ⊗ Fn, for all h ∈ H and n ≥ 0.
We will say that F is an universal deformation formula (UDF) based on H if,
moreover, F0 = 1 ⊗ 1.
Theorem 1.16. Let (A, s) be a left H-module algebra. If F =P Fi ti is an UDF
based on H, such that
(s ⊗ H)
(H ⊗ s)(Fi ⊗ a) = a ⊗ Fi
for all i ≥ 0 and a ∈ A,
then, the construction considered in Theorem 1.15, applied to the left H[[t]]-module
algebra (A[[t]], s[[t]]) introduced above, produces a formal deformation of A.
Proof. It is immediate.
(cid:3)
/
/
/
/
/
/
/
/
/
/
/
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
7
2 Hq-module algebra structures and deformations
In this section, we briefly review the construction of the Hopf algebra Hq and
the UDF expq based on Hq considered in [W], we introduce the notion of a good
transposition of Hq on an algebra A, and we describe all the braided Hq-module
algebras whose transposition is good.
Let q ∈ k× and let H be the algebra generated by D1, D2, σ±1, subject to the
relations
D1D2 = D2D1,
σσ−1 = σ−1σ = 1
and qσDi = Diσ for i = 1, 2,
It is easy to check that H is a Hopf algebra with
∆(D1) := D1 ⊗ σ + 1 ⊗ D1,
ǫ(D1) := 0,
∆(D2) := D2 ⊗ 1 + σ ⊗ D2,
ǫ(D2) := 0,
S(D1) := −D1σ−1,
S(D2) := −σ−1D2,
∆(σ) := σ ⊗ σ,
ǫ(σ) := 1,
S(σ) := σ−1.
If q is a primitive l-root of unity with l ≥ 2, then the ideal I of H generated by Dl
1
and Dl
2 is a Hopf ideal. So, the quotient H/I is also a Hopf algebra. Let
Hq :=(H/I
H
if q is a primitive l-root of unity with l ≥ 2,
if q = 1 or it is not a root of unity.
The Hopf algebra Hq was considered in the paper [W], where it was proved that
1
(i)!q
1
(i)q!
(tD1 ⊗ D2)i
if q is a primitive l-root of unity (l ≥ 2),
(tD1⊗D2)i
if q = 1 or it is not a root of unity,
l−1Xi=0
∞Xi=0
expq(tD1⊗D2) :=
is an UDF based on Hq.
2.1 Good transpositions of Hq on an algebra
One of our main purposes in this paper is to construct formal deformation of alge-
bras by using the UDF expq(tD1 ⊗D2). By Theorem 1.16, it will be sufficient to
obtain examples of Hq-module algebras (A, s), whose underlying transpositions s
satisfy
(s ⊗ Hq)
(Hq ⊗ s)(D1 ⊗ D2 ⊗ a) = a ⊗ D1 ⊗ D2
for all a ∈ A.
(2.1)
Definition 2.1. A k-linear map s : Hq ⊗ A → A ⊗ Hq is good if condition (2.1) is
fulfilled.
It is evident that s : Hq ⊗ A → A ⊗ Hq is good if and only if there exists a
bijective k-linear map α : A → A such that
s(D1 ⊗ a) = α(a) ⊗ D1
and s(D2 ⊗ a) = α−1(a) ⊗ D2
for all a ∈ A.
Lemma 2.2. Let k[σ±1] denote the subHopfalgebra of Hq generated by σ. Each
transposition s : Hq ⊗ A → A ⊗ Hq takes k[σ±1] ⊗ A onto A ⊗ k[σ±1].
Proof. Let τ be the flip. Since τ
s(σ±1 ⊗ a) ∈ A ⊗ k[σ±1] for all a ∈ A. Write
s−1
τ is a transposition, it suffices to prove that
s(σ ⊗ a) =Xijk
γijk(a) ⊗ σiDj
1Dk
2 .
8
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
Since S2(D1) = q−1D1, S2(D2) = qD2 and S2(σ±1) = σ±1, we have
Xijk
γijk(a) ⊗ σiDj
1Dk
2 = s(σ ⊗ a)
(S2 ⊗ A)(σ ⊗ a)
= s
= (A ⊗ S2)
s(σ ⊗ a)
qk−j γijk(a) ⊗ σiDj
1Dk
2 ,
=Xijk
and so γijk = 0 for j 6= k. Using now that
Xij
γijj (a) ⊗ ∆(σ)i∆(D1)j∆(D2)j = (A ⊗ ∆)
s(σ ⊗ a)
= (s ⊗ Hq)
(Hq ⊗ s)
(∆ ⊗ A)(σ ⊗ a)
γi′j ′j ′ (γijj (a)) ⊗ σi′
Dj ′
1 Dj ′
2 ⊗ σiDj
1Dj
2,
= Xiji′ j ′
it is easy to check that γijj = 0 if j > 0 (use that in each term of the right side the
exponent of D1 equals the exponent of D2). For σ−1 the same argument carries
over. This finishes the proof.
(cid:3)
In the following result we obtain a characterization of the good transpositions of
Hq on an algebra A.
Theorem 2.3. The following facts hold:
(1) If s : Hq ⊗ A → A ⊗ Hq is a good transposition, then s(σ±1 ⊗ a) = a ⊗ σ±1
for all a ∈ A and the map α : A → A, defined by s(D1 ⊗ a) = α(a) ⊗ D1, is
an algebra homomorphism.
(2) Given an algebra automorphism α : A → A, there exists only one good
transposition s : Hq ⊗ A → A ⊗ Hq such that s(D1 ⊗ a) = α(a) ⊗ D1 for all
a ∈ A.
Proof. (1) By Lemma 2.2, we know that s induces by restriction a transposition
of k[σ±1] on A. Hence, by [G-G1, Theorem 4.14], there is a superalgebra structure
A = A+ ⊕ A− such that
s(σi ⊗ a) =(a ⊗ σi
a ⊗ σ−i
if a ∈ A+,
if a ∈ A−.
Let α : A → A be as in the statement. Since σ is a transposition, if a ∈ A−, then
α(a) ⊗ D1 ⊗ σ + α(a) ⊗ 1 ⊗ D1 = (A ⊗ ∆)
= (s ⊗ Hq)
(∆ ⊗ A)(D1 ⊗ a)
= α(a) ⊗ D1 ⊗ σ−1 + α(a) ⊗ 1 ⊗ D1.
s(D1 ⊗ a)
(Hq ⊗ s)
So, A− = 0. Finally, α is an algebra homomorphism, because
s(h ⊗ 1) = 1 ⊗ h for each h ∈ Hq and s
(Hq ⊗ µA) = (µA ⊗ Hq)
(A ⊗ s)
(s ⊗ A).
(2) By item (1) and the comment preceding Lemma 2.2, it must be
s(σ±1 ⊗ a) = a ⊗ σ±1,
s(D1 ⊗ a) = α(a) ⊗ D1 and s(D2 ⊗ a) = α−1(a) ⊗ D2.
So, necessarily
s(σiDj
1Dk
2 ⊗ a) = αj−k(a) ⊗ σiDj
1Dk
2 .
We leave to the reader the task to prove that s is a good transposition.
(cid:3)
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
9
2.2 Some Hq-module algebra structures
Let A be an algebra. Let us consider k-linear maps ς, δ1, δ2 : A → A. It is evident
that there is a (necessarily unique) action ρ : Hq ⊗ A → A such that
ρ(σ ⊗ a) = ς(a),
ρ(D1 ⊗ a) = δ1(a) and ρ(D2 ⊗ a) = δ2(a)
(2.2)
for all a ∈ A, if and only if the maps ς, δ1 and δ2 satisfy the following conditions:
(1) ς is a bijective map,
(2) δ1
δ2 = δ2
δ1,
δi = δi
(3) qς
(4) If q 6= 1 and ql = 1, then δl
ς for i = 1, 2.
1 = δl
2 = 0.
Let s : Hq ⊗ A → A ⊗ Hq be a good transposition and let α be the associated
automorphism. Let ς, δ1 and δ2 be k-linear endomorphisms of A satisfying (1) -- (4).
Next, we determine the conditions that ς, δ1 and δ2 must satisfy in order that (A, s)
becomes an Hq-module algebra via the action ρ defined by (2.2).
Theorem 2.4. (A, s) is an Hq-module algebra via ρ if and only if:
(5) ς is an algebra automorphism,
(6) α
δi = δi
α for i = 1, 2,
(7) α
ς = ς
α,
(8) δi(1) = 0 for i = 1, 2,
(9) δ1(ab) = δ1(a)ς(b) + α(a)δ1(b) for all a, b ∈ A,
(10) δ2(ab) = δ2(a)b + ς(cid:0)α−1(a)(cid:1)δ2(b) for all a, b ∈ A.
Proof. Assume that (A, s) is an Hq-module algebra and let τ : Hq ⊗ Hq → Hq ⊗ Hq
be the flip. Evaluating the equality
s
(Hq ⊗ ρ) = (ρ ⊗ Hq)
(Hq ⊗ s)
(τ ⊗ A)
successively on D1 ⊗ Di ⊗ a and D1 ⊗ σ ⊗ a with i ∈ {1, 2} and a ∈ A arbitrary,
we verify that items (6) and (7) are satisfied. Item (8) follows from the fact that
D1 ·1 = D2 ·1 = 0. Finally, using that σ·1 = 1 and evaluating the equality
ρ
(Hq ⊗ µA) = µA
(∆ ⊗ A ⊗ A)
(ρ ⊗ ρ)
(Hq ⊗ s ⊗ A)
on σ ⊗a⊗b and Di ⊗a⊗b, with i = 1, 2 and a, b ∈ A arbitrary, we see that items (5),
(9) and (10) hold. So, conditions (5) -- (10) are necessary. By Remark 1.13, in order
to verify that they are also sufficient, it is enough to check that they imply that
h·1 = ǫ(h),
s(h ⊗ l·a) = (ρ ⊗ Hq)
(ρ ⊗ ρ)
h · (ab) = µA
(Hq ⊗ s)(l ⊗ h ⊗ a),
(Hq ⊗ s ⊗ A)(∆(h) ⊗ a ⊗ b),
for all a, b ∈ A and h, l ∈ {D1, D2, σ±1}. We leave this task to the reader.
(cid:3)
Note that condition (8) in Theorem 2.4 is redundant since it can be obtained by
applying condition (9) and (10) with a = b = 1.
10
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
3 Hq-module algebra structures on crossed products
Let G be a group endowed with a representation on a k-vector space V of dimen-
sion n. Consider the symmetric k-algebra S(V ) equipped with the unique action of
G by automorphisms that extends the action of G on V and take A = S(V )#f G,
where f : G × G → k× is a normal cocycle. By definition the k-algebra A is a free
left S(V )-module with basis {wg : g ∈ G}. Its product is given by
(P wg)(Qwh) := P gQf (g, h)wgh,
where gQ denotes the action of g on Q. This section is devoted to the study of the
Hq-module algebras (A, s), with s good, that satisfy:
s(Hq ⊗ V ) ⊆ V ⊗ Hq,
s(Hq ⊗ kwg) ⊆ kwg ⊗ Hq, σ·v ∈ V and σ·wg ∈ kwg,
for all v ∈ V and g ∈ G. In Theorem 3.5 we give a general characterization of these
module algebras, and in Subsection 3.1 we consider a specific case which is more
suitable for finding concrete examples, and we study it in detail. Finally in Sub-
section 3.2 we consider the case where the cocycle involves several non necessarily
central elements of G.
In the following proposition we characterize the good transpositions s of Hq on
A satisfying the hypothesis mentioned above. By theorem 2.3 this is equivalent to
require that the k-linear map α : A → A associated with α, takes V to V and kwg
to kwg for all g ∈ G.
Proposition 3.1. Let α : V → V be a k-linear map and χα : G → k× a map.
There is a good transposition s : Hq ⊗ A → A ⊗ Hq, such that
s(D1 ⊗ v) = α(v) ⊗ D1
and
s(D1 ⊗ wg) = χα(g)wg ⊗ D1
for all v ∈ V and g ∈ G, if and only if α is a bijective k[G]-linear map and χα is a
group homomorphism.
Proof. By Theorem 2.3 we know that s exists if an only if the k-linear map α : A → A
defined by
α(v1 · · · vmwg) := α(v1) · · · α(vm)χα(g)wg,
is an automorphism. But, if this happens, then
a) χα is a morphism since
χα(g)χα(h)f (g, h)wgh = α(wg)α(wh) = α(wgwh) = χα(gh)f (g, h)wgh
for all g, h ∈ G,
b) α is a bijective k[G]-linear map, since it is the restriction and correstriction of
α to V , and
α(gv) = α(wg)α(v)α(w−1
g ) = χα(g)wg α(v)(χα(g)wg)−1 = wg α(v)w−1
g = gα(v).
Conversely, if α is a bijective map then α is also, and if α is a k[G]-linear map and
χα is a morphism, then
α(wg)α(v) = χα(g)wg α(v) = gα(v)χα(g)wg = α(gv)α(wg)
and
α(wg)α(wh) = χα(g)wgχα(h)wh = f (g, h)χα(gh)wgh = α(f (g, h)wgh),
for all v ∈ V and g, h ∈ G, from which it follows easily that α is a morphism. (cid:3)
Let A = S(V )#f G be as above. Throughout this section we fix a morphism
χα : G → k× and a bijective k[G]-linear map α : V → V , and we let α : A → A
denote the automorphism determined by α and χα. Moreover we will call
s : Hq ⊗ A → A ⊗ Hq
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
11
the good transposition associated with α. Our purpose is to obtain all the Hq-mo-
dule algebra structures on (A, s) such that
σ·v ∈ V
and σ·wg ∈ kwg
for all v ∈ V and g ∈ G.
(3.3)
Under these restrictions we obtain conditions which allow us to construct all Hq-
module structures in concrete examples. Thanks to Theorem 1.16 and the fact
that expq(tD1 ⊗ D2) is an UDF based on Hq, each one of these examples produces
automatically a formal deformation of A. First note that given an Hq-module
algebra structure on (A, s) satisfying (3.3), we can define k-linear maps
δ1 : V → A,
δ2 : V → A and ς : V → V
δ1 : G → A,
δ2 : G → A and χς : G → k×,
and maps
by
δi(v) := Di ·v,
ς(v) := σ·v,
δi(g) := Di ·wg
and σ·wg := χς (g)wg.
Lemma 3.2. Let ς : V → V be a k-linear map and χς : G → k× be a map. Then,
the map ς : A → A defined by
ς(v1mwg) := ς(v1) · · · ς(vm)χς (g)wg,
is a k-algebra automorphism if and only if ς is a bijective k[G]-linear map and χς
is a group homomorphism.
Proof. This was checked in the proof of Proposition 3.1.
(cid:3)
Lemma 3.3. Let δ1 : V → A and δ2 : V → A be k-linear maps and let δ1 : G → A
and δ2 : G → A be maps.
(1) The k-linear map δ1 : A → A given by
δ1(v1mwg) :=
mXj=1
α(v1,j−1)δ1(vj)ς(vj+1,mwg) + α(v1m)δ1(g),
where vhl = vh · · · vl, is well defined if and only if
δ1(v)ς(w) + α(v)δ1(w) = δ1(w)ς(v) + α(w)δ1(v)
for all v, w ∈ V . (3.4)
(2) The map δ2 : A → A given by
δ2(v1mwg) :=
mXj=1
ς(cid:0)α−1(v1,j−1)(cid:1)δ2(vj )vj+1,mwg + ς(cid:0)α−1(cid:1)(v1m)δ2(g)
is well defined if and only if
δ2(v)w+ς(cid:0)α−1(v)(cid:1)δ2(w) = δ2(w)v+ς(cid:0)α−1(w)(cid:1)δ2(v) for all v, w ∈ V . (3.5)
Proof. We prove the first assertion and leave the second one, which is similar, to
the reader. The only if part follows immediately by noting that
δ1(v)ς(w) + α(v)δ1(w) = δ1(vw) = δ1(wv) = δ1(w)ς(v) + α(w)δ1(v).
In order to prove the if part it suffices to check that
δ1(v1 · · · vi−1vi+1vivi+2 · · · vmwg) = δ1(v1mwg)
for all i < m,
which follows easily from the hypothesis.
(cid:3)
Lemma 3.4. Assume that ς is an algebra automorphism and δ1, δ2 are well defined.
The following facts hold:
12
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
(1) The map δ1 satisfies
δ1(x1 · · · xm) =
α(x1 . . . xj−1)δ1(xj )ς(xj+1 · · · xm)
for all x1, . . . , xm ∈ k#f G ∪ V , if and only if
(a) δ1(gv)χς (g)wg + α(gv)δ1(g) = δ1(g)ς(v) + χα(g)wg δ1(v),
(b) f (g, h)δ1(gh) = δ1(g)χς (h)wh + χα(g)wgδ1(h),
for all v ∈ V and g, h ∈ G.
(2) The map δ2 satisfies
δ2(x1 · · · xm) =
α−1(x1 . . . xj−1)δ1(xj )xj+1 · · · xm
ς
mXj=1
mXj=1
mXj=1
mXj=1
for all x1, . . . , xm ∈ k#f G ∪ V , if and only if
(a) δ2(gv)wg + ς(cid:0)α−1(gv)(cid:1)δ2(g) = δ2(g)v + χς (g)χ−1
(b) f (g, h)δ2(gh) = δ2(g)wh + χς (g)χ−1
for all v ∈ V and g, h ∈ G.
α (g)wgδ2(h),
α (g)wg δ2(v),
Proof. We prove the first assertion and leave the second one to the reader. For the
only if part it suffices to note that
δ1(gv)ς(wg) + α(gv)δ1(g) = δ1(gvwg) = δ1(wgv) = δ1(g)ς(v) + α(wg)δ1(v),
f (g, h)δ1(gh) = δ1(wgwh) = δ1(g)ς(wh) + α(wg)δ1(h),
and to use the definitions of ς(wg) and α(wg). We prove the sufficient part by induc-
tion on r = m + 1 − i, where i is the first index with xi ∈ k#f G (if x1, . . . , xm ∈ V
we set r := 0). For r ∈ {0, 1} the result follows immediately from the definition of
δ1. Assume that it is true when r < r0 and that m + 1 − i = r0. If xi = wg and
xi+1 = v ∈ V , then
δ1(x1 · · · xm) = δ1(y1 · · · ym) where yj =
and hence, by the inductive hypothesis and item (a),
xj
gv
wg
if j /∈ {i, i + 1},
if j = i,
if j = i + 1,
δ1(x1 · · · xm) =
=
α(y1 . . . yj−1)δ1(yj)ς(yj+1 · · · ym)
α(x1 . . . xj−1)δ1(xj)ς(xj+1 · · · xm).
If xi = wg and xi+1 = wh, then
δ1(x1 · · · xm) = f (g, h)δ1(y1 · · · ym−1) where yj =
and hence, by the inductive hypothesis and item (b),
xj
wgh
xj+1
if j < i,
if j = i,
if j > i,
δ1(x1 · · · xm) =
=
m−1Xj=1
mXj=1
f (g, h)α(y1 . . . yj−1)δ1(yj)ς(yj+1 · · · ym−1)
α(x1 . . . xj−1)δ1(xj )ς(xj+1 · · · xm),
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
13
as we want.
(cid:3)
Theorem 3.5. Let δ1 : V → A, δ2 : V → A and ς : V → V be k-linear maps and let
δ1 : G → A, δ2 : G → A and χς : G → k× be maps. There is an Hq-module algebra
structure on (A, s), such that
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v)
and Di ·wg = δi(g)
for all v ∈ V , g ∈ G and i ∈ {1, 2}, if and only if
(1) ς : V → V is a bijective k[G]-linear map and χς is a group homomorphism,
(2) Conditions (3.4) and (3.5) in Lemma 3.3 and items (1)(a), (1)(b), (2)(a)
and (2)(b) in Lemma 3.4 are satisfied,
(3) δi
α = α
δi,
(4) χα(g)δi(g) = α(cid:0)δi(g)(cid:1) for all g ∈ G,
α = α
(5) ς
(6) The maps ς : A → A, δ1 : A → A and δ2 : A → A, introduced in Lemmas 3.2
ς
and 3.3, satisfy the following properties:
δ2
δ1 = δ1
δ2,
χς (g)δi(g) = qς(cid:0)δi(g)(cid:1),
δi,
δi
ς = qς
δl
1 = δl
2 = 0 if q 6= 1 and ql = 1.
δ2
δ1 = δ1
δ2,
Proof. By Theorem 2.4 and the discussion above it, we know that to have an
Hq-module algebra structure on (A, s) satisfying the requirements in the statement
is equivalent to have maps ς, δ1, δ2 : A → A satisfying conditions (1) -- (10) in Sub-
section 2.2 and such that
ς(v) = ς(v),
ς(wg) = χς (g)wg,
δi(v) = δi(v)
and δi(wg) = δi(g)
for all v ∈ V , g ∈ G and i ∈ {1, 2}. Now, it is easy to see that
a) If ς, δ1 and δ2 satisfy conditions (5), (9) and (10) in Subsection 2.2, then
ς(v1mwg) = ς(v1) · · · ς(vm)χς (g)wg,
α(v1,j−1)δ1(vj )ς(vj+1,mwg) + α(v1m)δ1(g),
ς(cid:0)α−1(v1,j−1)(cid:1)δ2(vj )vj+1,mwg + ς(cid:0)α−1(v1m)(cid:1)δ2(g),
δ1(v1mwg) =
δ2(v1mwg) =
mXj=1
mXj=1
where vhl = vh · · · vl.
b) By Lemmas 3.2, 3.3 and 3.4, the maps defined in a) satisfy conditions (1), (5),
(8), (9) and (10) in Subsection 2.2 if and only if items (1) and (2) of the present
theorem are fulfilled.
So, in order to finish the proof it suffices to check that:
c) Conditions (6) and (7) in Subsection 2.2 are satisfied if and only if items (3) -- (5)
of the present theorem are fulfilled,
d) Conditions (2), (3) and (4) in Subsection 2.2 are satisfied if and only if item (6)
of the present theorem is fulfilled.
We leave this task to the reader.
(cid:3)
We are going now to consider several particular cases, with the purpose of ob-
taining more precise results. This will allow us to give some specific examples of
formal deformations of associative algebras.
14
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
3.1 First case
Let α, χα, α and s be as in the discussion following Proposition 3.1. Let δ1 : V → A,
δ2 : V → A and ς : V → V be k-linear maps and let χς : G → k× be a map. Assume
that the kernels of δ1 and δ2 have codimension 1, ker δ1 6= ker δ2 and there exist
xi ∈ V \ ker δi, such that δi(xi) = Piwgi with Pi ∈ S(V ) and gi ∈ G. Without loss
of generality we can assume that x1 ∈ ker δ2 and x2 ∈ ker δ1 (and we do it). For
g ∈ G and i ∈ {1, 2}, let λig, ωi, νi ∈ k be the elements defined by the following
conditions:
gxi − λigxi ∈ ker δi,
ς(xi) − ωixi ∈ ker δi
and α(xi) − νixi ∈ ker δi.
Theorem 3.6. There is an Hq-module algebra structure on (A, s), satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0
for all v ∈ V , g ∈ G and i ∈ {1, 2}, if and only if
1 α(v) for all v ∈ ker δ1 and ς(v) = g2α(v) for all v ∈ ker δ2,
(1) ς is a bijective k[G]-linear map and χς is a group homomorphism,
(2) ς(v) = g−1
(3) g1 and g2 belong to the center of G,
(4) ker δ1 and ker δ2 are G-submodules of V ,
(5) gP1 = λ1gχ−1
(6) gP2 = λ2gχα(g)χ−1
(7) α(ker δi) = ker δi for i ∈ {1, 2},
(8) P1 ∈ ker δ2 and P2 ∈ ker δ1, where δ1 and δ2 are the maps defined by
α (g)χς (g)f −1(g, g1)f (g1, g)P1 for all g ∈ G,
ς (g)f −1(g, g2)f (g2, g)P2 for all g ∈ G,
δ1(v1mwg) :=
δ2(v1mwg) :=
mXj=1
mXj=1
α(v1,j−1)δ1(vj)ς(vj+1,mwg),
ς(cid:0)α−1(v1,j−1)(cid:1)δ2(vj)vj+1,mwg,
in which vhl = vh · · · vl,
(9) ς(Pi) = q−1ωiχ−1
the map given by
ς (gi)Pi and α(Pi) = νiχ−1
α (gi)Pi for i ∈ {1, 2}, where ς is
ς(v1mwg) = ς(v1) · · · ς(vm)χς (g)wg,
(10) If q 6= 1 and ql = 1, then δl
1 = δl
2 = 0.
In order to prove this result we first need to establish some auxiliary results.
Lemma 3.7. The following facts hold:
(1) Condition (3.4) of Lemma 3.3 is satisfied if and only if g1ς(v) = α(v) for
all v ∈ ker δ1.
(2) Condition (3.5) of Lemma 3.3 is satisfied if and only if g2v = ς(cid:0)α−1(v)(cid:1) for
all v ∈ ker δ2.
Proof. We prove item 1) and we leave item 2), which is similar, to the reader. We
must check that
δ1(v)ς(w) + α(v)δ1(w) = δ1(w)ς(v) + α(w)δ1(v)
for all v, w ∈ V
(3.6)
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
15
if and only if ς1(v) = g−1
1 α(v) for all v ∈ ker δ1. It is clear that we can suppose that
v, w ∈ {x1} ∪ ker δ1. When v, w ∈ ker δ1 or v = w = x1 the equality (3.6) is trivial.
Assume v = x1 and w ∈ ker δ1. Then,
δ1(v)ς(w) + α(v)δ1(w) = P1wg1 ς(w) = P1
g1ς(w)wg1
and
δ1(w)ς(v) + α(w)δ1(v) = α(w)P1wg1 = P1 α(w)wg1
So, in this case, the result is true. Case v ∈ ker δ1 and w = x1 can be treated in a
similar way.
(cid:3)
Lemma 3.8. The following facts hold:
(1) Items (1)(a) and (1)(b) of Lemma 3.4 are satisfied if and only if
(a) ker δ1 is a G-submodule of V ,
(b) g1 belongs to the center of G,
(c) gP1 = λ1gχ−1
α (g)χς (g)f −1(g, g1)f (g1, g)P1, for all g ∈ G.
(2) Items (2)(a) and (2)(b) of Lemma 3.4 are satisfied if and only if
(a) ker δ2 is a G-submodule of V ,
(b) g2 belongs to the center of G,
(c) gP2 = λ2gχα(g)χ−1
ς
(g)f −1(g, g2)f (g2, g)P2, for all g ∈ G.
Proof. We prove item 1) and we leave item 2) to the reader. Since δ1 = 0, it is
sufficient to prove that
δ1(gv)χς (g)wg = χα(g)wg δ1(v)
for all v ∈ V and g ∈ G,
(3.7)
if and only if conditions (1)(a), (1)(b) and (1)(c) are satisfied. We can assume
that v ∈ {x1} ∪ ker δ1. When v ∈ ker δ1, then equality (3.7) is true if and only if
gv ∈ ker δ1. Now, since
δ1(gx1)χς (g)wg = λ1gP1wg1 χς (g)wg = λ1gP1χς (g)f (g1, g)wg1g
and
χα(g)wg δ1(x1) = χα(g)wgP1wg1 = χα(g)gP1f (g, g1)wgg1 ,
equality (3.7) is true for v = x1 and g ∈ G if and only if conditions (1)(b) and (1)(c)
are satisfied.
(cid:3)
Proof of Theorem 3.6. First note that item (1) coincide with item (1) of Theo-
rem 3.5 and that, by Lemmas 3.7 and 3.8, item (2) of Theorem 3.5 is equivalent to
items (2) -- (6). Item (4) of Theorem 3.5 and two of the equalities in item (6) of the
same theorem, are trivially satisfied because δ1 = δ2 = 0. Since
item (3) of Theorem 3.5 is true if and only if item (7) and the second equality in
item (9) hold. Since α is k[G]-linear, item (5) of Theorem 3.5 is an immediate con-
sequence of item (2) of Theorem 3.6. Finally we consider the non-trivial equalities
and α(cid:0)δi(xi)(cid:1) = α(Piwgi ) = α(Pi)χα(gi)wgi ,
δi(cid:0) α(xi)(cid:1) = νi δi(xi) = νPiwgi
in item (6) of Theorem 3.5. It is easy to see that δi(cid:0)ς(xi)(cid:1) = qς(cid:0)δi(xi)(cid:1) if and only
if the first equality in item (9) holds. On the other hand δi(cid:0)ς(v)(cid:1) = qς(cid:0)δi(v)(cid:1) for all
The equality δ2(cid:0)δ1(v)(cid:1) = δ1(cid:0)δ2(v)(cid:1) is trivially satisfied for v ∈ ker δ1 ∩ ker δ2, and
for v ∈ {x1, x2} it is equivalent to item (8). Lastly, the remaining equality coincides
with item (10).
(cid:3)
v ∈ ker δi if and only if ς(ker δi) ⊆ ker δi, which follows from items (2), (4) and (7).
Remark 3.9. The following facts hold:
16
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
- Since α and ς are bijective k[G]-linear maps, from item (2) Theorem 3.6 it
follows that
g−1
1 v = g2v
for all v ∈ ker δ1 ∩ ker δ2.
(3.8)
- Since x1 ∈ ker δ2 and ker δ2 is G-stable, gx1 − λ1gx1 ∈ ker δ1 ∩ ker δ2.
Similarly gx2 − λ1gx2 ∈ ker δ1 ∩ ker δ2.
- Since ker δi is a G-submodule of V and the k-linear map
V
v
/ V
/ gv
is an isomorphism for each g ∈ G, it is impossible that gxi ∈ ker δi. Con-
sequently, λig ∈ k× for each g ∈ G. Moreover, using again that ker δi is
a G-submodule of V , it is easy to see that the map g 7→ λig is a group
homomorphism.
Items (1), (2), (4), (7) and the fact that α is bijective
imply that also ω1, ω2, ν1, ν2 ∈ k×.
- Since
ς(x1) = α(g2x1) ≡ λ1g2 α(x1) ≡ λ1g2 ν1x1
(mod ker δ1),
we have ω1 = λ1g2 ν1. A similar argument shows that ν2 = λ2g1 ω2.
Corollary 3.10. Assume that the conditions above Theorem 3.6 are fulfilled and
that there exists an Hq-module algebra structure on (A, s) satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0
for all v ∈ V , g ∈ G and i ∈ {1, 2}. If P1 ∈ S(ker δ1) and P2 ∈ S(ker δ2), then
λ1g1 λ1g2 = q
and λ2g1 λ2g2 = q−1.
Moreover g0 := g1g2 has determinant 1 as an operator on V .
Proof. By items (9), (2) and (5) of Theorem 3.6,
q−1ω1χ−1
ς (g1)P1 = ς(P1) = g−1
1 P1 = ν1λ−1
1g1 χ−1
ς (g1)P1.
1 α(P1) = ν1χ−1
α (g1)g−1
Hence λ1g1 λ1g2 = q as we want, since ω1 = ν1λ1g2 . The proof that λ2g1 λ2g2 = q−1
It remains to check that det(g0) = 1. Since ker δ1 and ker δ2 are G-
is similar.
invariant, we have
gx1 ∈ ker δ2
and gx2 ∈ ker δ1
for all g ∈ G,
and so
g0x1 ∈ λ1g1 λ1g2 x1 + W and g0x2 ∈ λ2g1 λ2g2 x1 + W,
where W = ker δ1 ∩ ker δ2. Moreover, by Remark 3.9 we know that g0 acts as the
identity map on W and hence det(g0) = λ1g1 λ1g2 λ2g1 λ2g2 = 1.
(cid:3)
Remark 3.11. A particular case is the Hq-module algebra A considered in [W,
Section 4], in which P1 = 1, g1 = 1 and α is the identity map. Our P2, g2 and
f correspond in [W] to s, g and α, respectively. Our computations show that the
condition that h(s) = x1(h)x2(h)α(g, h)α−1(h, g)s, which appears as informed by
the cohomology of finite groups in [W], is in fact necessary for the existence of
the Hq-module algebra structure of A, and it does not depend on cohomological
considerations. In particular we need this condition for any group G, finite or not.
Similarly the conditions that g is central and det(g) = 1 are necessary even for
infinite groups.
/
/
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
17
Let G, V , f : G × G → k× and A be as at the beginning of this section. Let
α : V → V be a bijective k[G]-linear map, χα : G → k× a group homomorphism,
α : A → A the algebra automorphism induced by α and χα, and s the good trans-
position associated with α. Let
a) V1 6= V2 subspaces of codimension 1 of V such that V1 and V2 are α-stable
G-submodules of V ,
b) g1 and g2 central elements of G such that g−1
1 v = g2v for all v ∈ V1 ∩ V2,
c) χς : G → k× a group homomorphism and ς : V → V the map defined by
ς(v) :=(α(g−1
1 v)
α(g2v)
if v ∈ V1,
if v ∈ V2,
d) x1 ∈ V2 \ V1, x2 ∈ V1 \ V2, P1 ∈ S(V1), P2 ∈ S(V2) and δ1, δ2 : V → A the
maps defined by
ker δi := Vi
and δi(xi) := Piwgi .
For g ∈ G and i ∈ {1, 2}, let λig, νi, ωi ∈ k× be the elements defined by the
conditions gxi − λigxi ∈ Vi, α(xi) − νixi ∈ Vi and ς(xi) − ωixi ∈ Vi.
The following result is a sort of a reformulation of Theorem 3.6, more appropriate
to construct explicit examples. The only new hypothesis that we need is that
Pi ∈ S(Vi).
Corollary 3.12. There is an Hq-module algebra structure on (A, s), satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0
for all v ∈ V , g ∈ G and i ∈ {1, 2}, if and only if
(1) q = λ1g1 λ1g2 and q−1 = λ2g1 λ2g2 ,
(2) gP1 = λ1gχ−1
(3) gP2 = λ2gχα(g)χ−1
(4) α(Pi) = νiχ−1
(5) P1 ∈ ker δ2 and P2 ∈ ker δ1, where δ1, δ2 : A → A are the maps defined in
α (g)χς (g)f −1(g, g1)f (g1, g)P1,
ς (g)f −1(g, g2)f (g2, g)P2,
α (gi)Pi,
item (8) of Theorem 3.6,
(6) If q 6= 1 and ql = 1, then δl
1 = δl
2 = 0.
Proof. ⇐) By a), b), c) and d), it is obvious that items (1), (2), (3), (4) and (7) of
Theorem 3.6 are satisfied. Moreover items (2), (3), (5) and (6) are items (5), (6), (8)
and (10) of Theorem 3.6. So, we only must to check that item (9) of Theorem 3.6 is
satisfied. But the second equality in this item is exactly the one required in item (4)
of the present corollary, and we are going to check that the first one is true with
q = λ1g1 λ1g2 . Arguing as in Remark 3.9, and using item (2) with g = g1, item (1)
and item (4), we obtain
q−1ω1χ−1
ς (g1)P1 = q−1λ1g2 ν1χ−1
ς
(g1)P1
α (g1) g−1
1 P1
α (g1) g−1
= q−1λ1g1 λ1g2 ν1χ−1
= ν1χ−1
= g−1
= ς(P1),
1 α(P1)
1 P1
18
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
where the last equality is true since P1 ∈ S(V1). Again arguing as in Remark 3.9,
and using item (3) with g = g2, item (1) and item (4), we obtain
q−1ω2χ−1
ς (g2)P2
2g2 ν2χ−1
α (g2) g2P2
2g1 ν2χ−1
2g1 λ−1
α (g2) g2P2
ς (g2)P2 = q−1λ−1
= q−1λ−1
= ν2χ−1
= g2α(P2)
= ς(P2),
where the last equality is true since P2 ∈ S(V2).
⇒) Items (2), (3), (5) and (6) are items (5), (6), (8) and (1) of Theorem 3.6, and
item (4) is the first equality in item (9) of that theorem. Finally item (1) follows
from Corollary 3.10.
(cid:3)
The following result shows that if x1 and x2 are eigenvectors of the maps v 7→ g1v
and v 7→ g2v, then item (5) in the statement of Corollary 3.12 can be easily tested
and item (6) can be removed from the hypothesis.
Proposition 3.13. Assume that conditions a), b), c) and d) above Corollary 3.12
are fulfilled. Let δ1 and δ2 be the maps introduced in item (8) of Theorem 3.6. If
λ1g1 λ1g2 = q,
λ2g1 λ2g2 = q−1
and
gixj = λjgi xj
for 1 ≤ i, j ≤ 2,
then:
1 = δl
(1) δl
(2) If q = 1 or it is not a root of unity, then P1 ∈ ker δ2 and P2 ∈ ker δ1 if and
2 = 0, whenever q 6= 1 and ql = 1.
only if P1, P2 ∈ S(V1 ∩ V2).
(3) If q 6= 1 is a primitive l-root of unity, then P1 ∈ ker δ2 and P2 ∈ ker δ1 if
and only if P1 ∈ S(cid:0)k xl
2 ⊕ (V1 ∩ V2)(cid:1) and P2 ∈ S(cid:0)k xl
1 ⊕ (V1 ∩ V2)(cid:1).
Proof. The proposition is a direct consequence of the following formulas:
and
1(xr1
δs
1 · · · xrn
2(xr1
δs
1 · · · xrn
c = χs
ς (g)χs(s−1)/2
ς
d = λsr2−s(s+1)/2
2g2
1
2
0
0
gs
1 g
1 xr3
3 · · · xrn
xr2
2 · · · xrn
n (cid:1)wgs
n wg) =(cαs(cid:0)xr1−s
n wg) =(dxr2−s
n (cid:1)wgs
2(cid:0)xr1
(r1 − k)q! s−1Yk=0
(g1) s−1Yk=0
(r2 − k)q! s−1Yk=0
2 g)! s−1Yk=0
f (g2, gk
(g1)χs(s−1)/2
α
s−1Yk=0
for s ≤ r1,
otherwise,
2 g
for s ≤ r2,
otherwise,
1 g)!αs−1(P s
1 ),
f (g1, gk
gk
2P2!.
where αs denotes the s-fold composition of α,
We will prove the formula for δs
1 and we will leave the other one to the reader. We
begin with the case s = 1. Since x2, . . . , xn ∈ ker δ1 and δ1(x1) = P1wg1 , from the
definition of δ1 it follows that
δ1(xr1
1 · · · xrn
n wg) =
r1−1Xj=0
α(xj
1)P1wg1 ς(xr1−j−1
1
xr2
2 · · · xrn
n wg).
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
19
Thus, using the definition of ς, item c) above Corollary 3.12, the fact that α is
G-linear and the hypothesis, we obtain
r1−1Xj=0
r1−1Xj=0
δ1(xr1
1 · · · xrn
n wg) =
=
α(xj
1)P1wg1 α(g2xr1−j−1
1
)g−1
1 α(xr2
2 · · · xrn
n )χς (g)wg
α(xj
1)P1α(g1g2xr1−j−1
1
)α(xr2
2 · · · xrn
n )χς (g)f (g1, g)wg1g
= χς (g)(r1)qf (g1, g)P1α(xr1 −1
xr2
2 · · · xrn
n )wg1g.
1
Assume that s ≤ r1 and that the formula for δs
1 holds. Since c depends on s, r1 and
g, it will be convenient for us to use the more precise notation cs,r1(g) for c. From
items (3) and (5) of Theorem 3.5 and item (9) of Theorem 2.4, It follows easily that
α on S(V ). Using this fact, item (9) of Theorem 2.4 and the inductive
α
hypothesis, we obtain
δ1 = δ1
xr2
2 · · · xrn
1
1
(xr1
(xr1
δs+1
1
1 · · · xrn
xr2
2 · · · xrn
n ) = 0. Otherwise,
1 · · · xrn
If s = r1, then δ1(xr1−s
δs+1
1
n wg) = α(cid:0)csr1 (g)(cid:1)αs(cid:0)δ1(xr1−s
n )wg1(cid:1)ς(wgs
where c = α(cid:0)cs,r1(g)(cid:1)αs(cid:0)c1,r1−s(1)(cid:1). The formula for δs+1
xr2
2 · · · xrn
xr2
2 · · · xrn
n wg) = cαs(cid:0)α(xr1−s−1
= cαs+1(xr1−s−1
from this fact.
n )χs
α(g1)χs
1
1
1
1 g)
ς (g1)χς (g)f (g1, gs
1g)wgs+1
g,
1
follows immediately
(cid:3)
n )(cid:1)ς(wgs
1 g).
Example 3.14. Let G = hgi be an order r cyclic group, ξ an element of k× and
fξ : G ⊗ G → k the cocycle defined by
fξ(gu, gv) :=(1 if u + v < r,
otherwise.
ξ
Of course, if r = ∞, then for any ξ this is the trivial cocycle. Let V be a vector space
endowed with an action of G and let A be the crossed product A = S(V )#fξ G. Let
{x1, . . . , xn} be a basis of V . Let us V1 and V2 denote the subspaces of V generated
by {x2, . . . , xn} and {x1, x3, . . . , xn} respectively. Let α : V → V be a bijective
k[G]-linear map. Assume that V1 and V2 are α-stable G-submodules of V and that
there exist λ1, λ2 ∈ k× such that gx1 = λ1x1 and gx2 = λ2x2. Let m1, m2 ∈ Z.
Assume that gm1 +m2v = v for all v ∈ V1 ∩V2 (if r < ∞ we can take 0 ≤ m1, m2 < r).
Let ς : V → V be the map defined by
ς(v) :=(α(g−m1v)
α(gm2v)
if v ∈ V1,
if v ∈ V2,
and let χα, χς : G → k× be two morphisms. Consider the automorphism of algebras
α : A → A given by α(v) := α(v) for v ∈ V and α(wg) = χα(g)wg, and define
δ1, δ2 : V → A by
δ1(x2) = · · · = δ1(xn) := 0,
δ2(x1) = δ2(x3) = · · · = δ1(xn) := 0,
δ1(x1) := P1wgm1 ,
δ2(x2) := P2wgm2 ,
where P1 ∈ S(V1) \ {0} and P2 ∈ S(V2) \ {0}. Let s be the transposition of Hq with
A associated with α. There is an Hq-module algebra structure over (A, s) satisfying
σ·v = ς(v), σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0 for all v ∈ V ,
if and only if
20
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
1
2
,
and q−1 = λm1+m2
α (g)χς (g)P1 and gP2 = λ2χα(g)χ−1
(1) q = λm1+m2
(2) gP1 = λ1χ−1
(3) α(P1) = ν1χ−m1
(4) If q = 1 or q is not a root of unity, then P1, P2 ∈ k[x3, . . . , xn],
(5) If q 6= 1 is a primitive l-root of unity, then
(g)P1 and α(P2) = ν2χ−m2
ς
(g)P2,
(g)P2,
α
α
P1 ∈ k[xl
2, x3, . . . , xn] and P2 ∈ k[xl
1, x3, . . . , xn].
Consequently, in order to obtain explicit examples of braided Hq-module algebra
structures on an algebra A of the shape S(V )#fξ G, where V is a k-vector space
with basis {x1, . . . , xn} and G = hgi is a cyclic group of order r ≤ ∞, we proceed
as follows:
First: We define an action of G on V . For this we choose
- a k-linear automorphism γ of V12 := hx3, . . . , xni, whose order divides
r if r < ∞,
- λ1, λ2 ∈ k× such that λr
1 = λr
2 = 1 if r < ∞,
and we set
gxi :=
λ1x1
λ2x2
γ(xi)
if i = 1,
if i = 2,
if i ≥ 3.
Second: We construct the algebra A. For this we choose ξ ∈ k× and we
define A = S(V )#fξ G, where fξ is the cocycle associate with ξ.
Third: We endow A with a k-algebra automorphism α. For this we take
ν1, ν2, η ∈ k× such that ηr = 1 if r < ∞, a k-linear automorphism α′ of V12
and v1, v2 ∈ V12, and we define
Fourth: We choose m1, m2 ∈ Z and ζ ∈ k× such that
(λ1λ2)m1+m2 = 1 and ζr = 1 if r < ∞,
α(wg) := ηwg
γm1+m2 = id,
and we define
ς(wg) := ζwg
and α(xi) :=
and ς(xi) :=
ν1x1 + v1
ν2x2 + v2
α′(xi)
if i = 1,
if i = 2,
if i ≥ 3.
λm2
1 (ν1x1 + v1)
λ−m1
2
(ν2x2 + v2)
α′(cid:0)γm2(xi)(cid:1)
if i = 1,
if i = 2,
if i ≥ 3.
Fifth: we set q := λm1+m2
1
and we choose P1, P2 ∈ S(V ) \ {0} such that
- if q is not a root of unity, then P1, P2 ∈ k[x3, . . . , xn],
- if q is a primitive l-root of unity, then
P1 ∈ k[xl
2, x3, . . . , xn] and P2 ∈ k[xl
1, x3, . . . , xn],
- gP1 = λ1η−1ζP1 and gP2 = λ2ηζ −1P2,
- α(P1) = ν1η−m1 P1 and α(P2) = ν2η−m2 P2.
Now, by the discussion at the beginning of this example, there is an Hq-module
algebra structure on (A, s), where s : Hq ⊗ A → A ⊗ Hq is the good transposition
associated with α, such that
σ·xj = ς(xj ), σ·wg = ζwg, Di ·wg = 0
and Di(xj ) =(0
Piwgmi
if i 6= j,
if i = j,
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
21
where i ∈ {1, 2} and j ∈ {1, . . . , n}.
Remark 3.15. If P1(0) 6= 0 and P2(0) 6= 0, then the conditions in the first step
are fulfilled if and only if λ1λ2 = 1, η = λ1ζ, ν1 = ηm1 , ν2 = ηm2 , P1 and P2 are
G-invariants, α(P1) = P1 and α(P2) = P2.
3.2 Second case.
Let α, χα, α and s be as in the discussion following Proposition 3.1, let χς : G → k×
be a map and let δ1 : V → A, δ2 : V → A and ς : V → V be k-linear maps such that
ker δ1 6= ker δ2 are subspaces of codimension 1 of V . Here we are going to consider
a more general situation that the one studied in the previous subsection. Assume
that for each i ∈ {1, 2} there exist
- an element xi ∈ V \ ker(δi),
- different elements gi1, . . . , gini of G,
- polynomials P (i)
gi1 , . . . , P (i)
gini ∈ S(V ) \ {0},
such that
δi(xi) =
P (i)
gij wgij .
niXj=1
(The reason for the notation P (i)
gij instead of the more simple Pij will became clear
in items (5) and (6) of the following theorem). Without loss of generality we can
assume that x1 ∈ ker δ2 and x2 ∈ ker δ1 (and we do it). For g ∈ G and i ∈ {1, 2},
let λig, ωi, νi ∈ k be the elements defined by the following conditions:
gxi − λigxi ∈ ker δi,
ς(xi) − ωixi ∈ ker δi
and α(xi) − νixi ∈ ker δi.
Lemma 3.16. The following facts hold:
(1) Condition (3.4) of Lemma 3.3 is satisfied if and only if g1jς(v) = α(v) for
all j ≤ n1 and v ∈ ker δ1.
(2) Condition (3.5) of Lemma 3.3 is satisfied if and only if g2jv = ς(cid:0) α−1(v)(cid:1)
for all j ≤ n2 and v ∈ ker δ2.
Proof. Mimic the proof of Lemma 3.7.
Lemma 3.17. The following facts hold:
(1) Items (1)(a) and (1)(b) of Lemma 3.4 are satisfied if and only if
(a) ker δ1 is a G-submodule of V ,
(b) {g1j : 1 ≤ j ≤ n1} is a union of conjugacy classes of G,
(c) gP (1)
α (g)χς (g)f −1(g, g1j)f (gg1jg−1, g)P (1)
g1j = λ1gχ−1
gg1j g−1 for j ≤ n1.
(2) Items (2)(a) and (2)(b) of Lemma 3.4 are satisfied if and only if
(a) ker δ2 is a G-submodule of V ,
(b) {g2j : 1 ≤ j ≤ n2} is a union of conjugacy classes of G,
(c) gP (2)
ς (g)f −1(g, g2j)f (gg2jg−1, g)P (2)
g2j = λ2gχα(g)χ−1
gg2j g−1 for j ≤ n2.
Proof. Mimic the proof of Lemma 3.8.
Theorem 3.18. There is an Hq-module algebra structure on (A, s), satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0
for all v ∈ V , g ∈ G and i ∈ {1, 2}, if and only if
(cid:3)
(cid:3)
22
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
(1) ς is a bijective k[G]-linear map and χς is a group homomorphism,
(2) ς(v) = g−1
1j α(v) for j ≤ n1 and all v ∈ ker δ1, and ς(v) = g2jα(v) for j ≤ n2
and all v ∈ ker δ2,
(3) {gij : 1 ≤ j ≤ ni} is a union of conjugacy classes of G for i ∈ {1, 2},
(4) ker δ1 and ker δ2 are G-submodules of V ,
(5) gP (1)
g1j = λ1gχ−1
gg1j g−1 for j ≤ n1,
α (g)χς (g)f −1(g, g1j)f (gg1jg−1, g)P (1)
ς (g)f −1(g, g2j)f (gg2jg−1, g)P (2)
(6) gP (2)
g2j = λ2gχα(g)χ−1
gg2j g−1 for j ≤ n2,
(7) α(ker δi) = ker δi for i ∈ {1, 2},
(8) Pn1
j=1 P (1)
g1j wg1j ∈ ker δ2 and Pn2
the maps defined by
j=1 P (2)
g2j wg2j ∈ ker δ1, where δ1 and δ2 are
δ1(v1mwg) :=
δ2(v1mwg) :=
in which vhl = vh · · · vl,
mXj=1
mXj=1
α(v1,j−1)δ1(vj)ς(vj+1,mwg),
ς(cid:0)α−1(v1,j−1)(cid:1)δ2(vj)vj+1,mwg,
(9) ς(P (i)
gij ) = q−1ωiχ−1
gij and α(P (i)
and j ≤ ni, where ς is the map given by
ς (gij)P (i)
gij ) = νiχ−1
α (gij )P (i)
gij for i ∈ {1, 2}
ς(v1mwg) := ς(v1) · · · ς(vm)χς (g)wg.
(10) If q 6= 1 and ql = 1, then δl
1 = δl
2 = 0.
Proof. Mimic the proof of Theorem 3.6, but using Lemmas 3.16 and 3.17 instead
of Lemmas 3.7 and 3.8, respectively.
(cid:3)
Remark 3.19. Since α and ς are bijective k[G]-linear maps, from item (2) it follows
that
g1jv = g1hv
g2jv = g2hv
g−1
1j v = g2hv
for 1 ≤ j, h ≤ n1 and all v ∈ ker δ1,
for 1 ≤ j, h ≤ n2 and all v ∈ ker δ2,
for 1 ≤ j ≤ n1, 1 ≤ h ≤ n2 and all v ∈ ker δ1 ∩ ker δ2.
(3.9)
(3.10)
(3.11)
On the other hand, arguing as in Remark 3.9 we can check that
- gxi − λ1gxi ∈ ker δ1 ∩ ker δ2 for all g ∈ G,
- λig ∈ k× for all g ∈ G,
- the maps g 7→ λig are morphisms,
- ω1, ω2, ν1, ν2 ∈ k×.
Finally, since
ς(x1) = α(g2jx1) ≡ λ1g2j α(x1)
(mod ker δ1),
we have ω1 = λ1g2j ν1 for j ≤ n2. Similarly, ν2 = λ2g1j ω2 for j ≤ n1. Consequently,
λ1g21 = · · · = λ1g2n2
and λ2g11 = · · · = λ2g1n1
,
which also follows from (3.9) and (3.10).
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
23
Corollary 3.20. Assume that the conditions at the beginning of the present sub-
section are fulfilled and that there exists an Hq-module algebra structure on (A, s),
satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Di ·v = δi(v) and Di ·wg = 0
for all v ∈ V , g ∈ G and i ∈ {1, 2}. If P (1)
j ≤ n1 and h ≤ n2, then
g1j ∈ S(ker δ1) and P (2)
g2h ∈ S(ker δ2) for all
λ1g1j λ1g2h = q
and λ2g1j λ2g2h = q−1.
Moreover g1jg2h has determinant 1 as an operator on V .
Proof. This result generalizes Corollary 3.10, and its proof is similar.
(cid:3)
Let G, V , f : G × G → k×, A, α : V → V , χα : G → k×, α : A → A and s be as
below of Remark 3.11. Assume we have
a) subspaces V1 6= V2 of codimension 1 of V such that V1 and V2 are α-stable
G-submodules of V , and vectors x1 ∈ V2 \ V1 and x2 ∈ V1 \ V2,
b) different elements gi1, . . . , gini of G, where i ∈ {1, 2}, such that:
-- {g11, . . . , g1n1} and {g21, . . . g2n2} are unions of conjugacy classes of G,
-- g1jv = g1hv for 1 ≤ j, h ≤ n1 and all v ∈ V1,
-- g2jv = g2hv for 1 ≤ j, h ≤ n2 and all v ∈ V2,
-- g−1
1j v = g2hv for 1 ≤ j ≤ n1, 1 ≤ h ≤ n2 and all v ∈ V1 ∩ V2.
c) a morphism χς : G → k×
d) Nonzero polynomials P (1)
g1j ∈ S(V1) and P (2)
g2h ∈ S(V2), where 1 ≤ j ≤ n1
and 1 ≤ h ≤ n2.
Let ς : V → V and δ1, δ2 : V → A be the maps defined by
For g ∈ G and i ∈ {1, 2}, let λig, νi ∈ k× be the elements defined by the following
conditions: gxi − λigxi ∈ Vi and α(xi) − νixi ∈ Vi. Note that, by item b),
λ2g11 = · · · = λ2g1n1
and λ1g21 = · · · = λ1g2n2
.
Corollary 3.21. There is an Hq-module algebra structure on (A, s), satisfying
σ·v = ς(v),
σ·wg = χς (g)wg, Dh·v = δh(v) and Dh·wg = 0,
for all v ∈ V , g ∈ G and i ∈ {1, 2}, if and only if for all j ≤ n1 and h ≤ n2 the
following facts hold:
g1j = λ1gχ−1
(1) q = λ1g1j λ1g21 and q−1 = λ2g11 λ2g2h ,
(2) gP (1)
(3) gP (2)
(4) α(P (1)
α (g)χς (g)f −1(g, g1j)f (gg1jg−1, g)P (1)
ς (g)f −1(g, g2h)f (gg2hg−1, g)P (2)
g2h = λ2gχα(g)χ−1
gg1j g−1 ,
gg2hg−1 ,
α (g2h)P (2)
g2h ,
α (g1j)P (1)
g2h ) = ν2χ−1
g1j ) = ν1χ−1
j=1 P (1)
g1j wg1j ∈ ker δ2 and Pn2
g1j and α(P (2)
h=1 P (2)
(5) Pn1
are the maps defined in item (8) of Theorem 3.18
(6) If q 6= 1 and ql = 1, then δl
1 = δl
2 = 0.
g2h wg2h ∈ ker δ1, where δ1, δ2 : A → A
Proof. It is similar to the proof of Corollary 3.12, using Theorem 3.18 instead
of Theorem 3.6. The proof that ς is G-linear requires additionally the fact that
ggij g−1
(cid:3)
v = gijv for 1 ≤ i ≤ 2 and 1 ≤ j ≤ ni, which is true by b).
ς(v) :=(α(g−1
11 v)
α(g21v)
if v ∈ V1,
if v ∈ V2,
ker δi := Vi
and δi(xi) :=
P (i)
gij wgij .
niXj=1
24
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
Remark 3.22. Assume that the hypothesis of Corollary 3.21 are fulfilled. Then, as
it was note above this corollary,
λ2g11 = · · · = λ2g1n1
Moreover, by item (1) it is clear that
and λ1g21 = · · · = λ1g2n2
.
λ1g11 = · · · = λ1g1n1
and λ2g21 = · · · = λ2g2n2
.
Proposition 3.23. Let G, V , f , A, α, V1, V2, g11, . . . , g1n1, g21, . . . , g2n2, ς, χς ,
δ1, δ2, x1, x2, ν1, ν2, λ1g and λ2g, where g ∈ G, be as in the discussion above
Corollary 3.21. Assume that
λ2g11 = · · · = λ2g1n1
λ1g11 = · · · = λ1g1n1
,
,
λ1g21 = · · · = λ1g2n2
λ2g21 = · · · = λ2g2n2
,
,
and that conditions a), b), c) and d) above that corollary are fulfilled. If
λ1g11 λ1g21 = q,
λ2g11 λ2g21 = q−1
and gihxj = λjgih xj ,
for 1 ≤ i, j ≤ 2 and 1 ≤ h ≤ ni, then:
1 = δl
2 = 0, whenever q 6= 1 and ql = 1.
(1) δl
(2) If q = 1 or q is not a root of unity, then P (1)
g1j ∈ ker δ2 and P (2)
g2h ∈ ker δ1 if
and only if P (1)
g1j , P (2)
g2h ∈ S(V1 ∩ V2).
(3) If q 6= 1 is a primitive l-root of unity, then P (1)
g1j ∈ ker δ2 and P (2)
g2h ∈ ker δ1
g1j ∈ S(cid:0)k xl
2 ⊕ (V1 ∩ V2)(cid:1) and P (2)
g2h ∈ S(cid:0)k xl
1 ⊕ (V1 ∩ V2)(cid:1).
n . Using the hypothesis it is easy to check by induction
if and only if P (1)
Proof. Let xr = xr1
on s that
1 · · · xrn
chc′
hαs(xr1−s
1
xr2
2 · · · xrn
n )wg1hs g1hs−1
···g1h1 g
for s ≤ r1,
dhd′
hxr2−s
2
Xh∈Is
n2
0
gs
21(cid:0)xr1
1 xr3
3 · · · xrn
n (cid:1)wg2hs g2hs−1
otherwise,
···g2h1 g
for s ≤ r2,
otherwise,
Xh∈Is
n1
0
δs
δs
1(xrwg) =
2(xrwg) =
and
where
Is
ni = Ini × · · · × Ini
αs denotes the s-fold composition of α,
, with Ini = {1, . . . , ni},
ς
s times
}
χs−k
χk−1
(g1hk )
α (g1hk ),
{z
s−1Yk=1
sYk=2
(r1 − k)q! sYk=1
f (g1hk, g1hk−1 · · · g1h1g)! sYk=1
(r2 − k)q! sYk=1
f (g2hk , g2hk−1 · · · g2h1 g)! s−1Yk=0
,
ch = χs
ς (g)
c′
d′
h = s−1Yk=0
h = s−1Yk=0
2g21
dh = λsr2−s(s+1)/2
The result follows easily from these formulas.
αs−1(P (1)
g1hk
)!,
g2hs−k!.
gk
21P (2)
(cid:3)
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
25
Example 3.24. Let Du be the Dihedral group Du := hs, t s2, tu, ststi. Then
Du acts on k[X1, X2] via
sX1 = −X1,
and tX2 = X2.
sX2 = −X2,
tX1 = X1
Let A = k[X1, X2]#Du. We have
- Assume u is even. Then, there is an H1-module algebra structure on A,
such that
σ·X1 = X1,
σ·X2 = X2,
D1 ·X1 = wt + wt−1 , D1·X2 = 0,
D2 ·X1 = 0,
σ·wti = wti , σ·wtis = −wtis,
D1·wti = 0, D1 ·wtis = 0,
D2·X2 = wtu/2 , D2·wti = 0, D2 ·wtis = 0.
- There is an H−1-module algebra structure on A, such that
σ·X1 = X1,
σ·X2 = −X2,
σ·wti = wti , σ·wtis = −wtis,
u−1Xi=0
D1 ·X1 =
D2 ·X1 = 0,
wtis, D1·X2 = 0,
D1 ·wti = 0, D1·wtis = 0,
D2·X2 = wt + wt−1 , D2 ·wti = 0, D2·wtis = 0.
- Assume u is even. Let α : A → A be the k-algebra map defined by
α(Qwti ) := Qwti
and α(Qwtis) := −Qwtis,
and let s : H1 ⊗ A → A ⊗ H1 be the transposition associated with α. There
is an H1-module algebra structure on A, such that
σ·X1 = X1,
σ·X2 = X2,
D1·X1 = wt + wt−1 , D1 ·X2 = 0,
D2·X1 = 0,
σ·wti = wti, σ·wtis = wtis,
D1 ·wti = 0, D1·wtis = 0,
D2 ·X2 = wtu/2 , D2 ·wti = 0, D2·wtis = 0.
4 Non triviality of the deformations
Let A = S(V )#f G be as in Section 3. By Theorem 1.16 we know that each
Hq-module algebra (A, s), with s a good transposition, produces to a formal defor-
mation AF of A, which is constructed using the UDF F = expq(tD1⊗D2). The aim
of this section is to prove that if (A, s) satisfies the conditions required in Corol-
lary 3.21 and P (1)
g2h ∈ S(V1 ∩ V2) for 1 ≤ j ≤ n1 and 1 ≤ h ≤ n2, then AF is
non trivial. We will prove this showing that its infinitesimal
g1j , P (2)
Φ(a ⊗ b) = δ1(cid:0)α−1(a)(cid:1)δ2(b),
∗
is not a coboundary. For this we use a complex X
cohomology of A, which is simpler than the canonical one.
(A), giving the Hochschild
4.1 A simple resolution
Given a symmetric k-algebra S := S(V ), we consider the differential graded algebra
(Y∗, ∂∗) generated by elements yv and zv, of zero degree, and v, of degree one, where
v ∈ V , subject to the relations
zλv+w = λzv + zw,
yvyw = ywyv,
yλv+w = λyv + yw,
yvzw = zwyv,
vyw = ywv,
vzw = zwv,
v + w = λv + w,
zvzw = zwzv,
v2 = 0,
26
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
where λ ∈ k and v, w ∈ V , and with differential ∂ defined by ∂(v) := ρv, where
ρv = zv − yv.
Note that S is a subalgebra of Y∗ via the embedding that takes v to yv for all
v ∈ V . This produces an structure of left S-module on Y∗. Similarly we consider Y∗
as a right S-module via the embedding of S in Y∗ that takes v to zv for all v ∈ V .
for all v ∈ V . The S-bimodule complex
Proposition 4.1. Leteµ : Y0 → S be the algebra map defined byeµ(yv) =eµ(zv) := v
∂1
∂2
∂3
∂4
eµ
Y4
∂5
Y5
∂6
. . .
(4.12)
S
Y0
Y1
Y2
Y3
is contractible as a left S-module complex.
Proof. Let {x1, . . . , xn} be a basis of V . We will write yi, zi, ρi and vi instead of
yxi, zxi, ρxi and vxi, respectively. A contracting homotopy
ς0 : S → Y0
and ςr+1 : Yr → Yr+1
(r ≥ 0),
of (4.12) is given by
ς(1) := 1,
i1 vδ1
i1 · · · ρml
il
vδl
ς(cid:0)ρm1
il(cid:1) :=((−1)sρm1
0
i1
vδ1
i1
· · · ρml−1
il−1
vδl−1
il−1
ρml−1
il
vil
if δl = 0,
ifδl = 1,
where we assume that i1 < · · · < il, δ1 + · · · + δl = s and ml + δl > 0. In fact, a
direct computation shows that
ς(1) = ∂(0) = 0,
, where ml > 0 and x′ = ρm1
i1
· · · ρml−1
il−1
with i1 < · · · < il, then
and ∂
ς(x) = ∂(x′ρml−1
vil) = x,
il
- Let x = x′ρml
il
, where ml + δl > 0 and x′ = ρm1
i1 < · · · < il and δ1 + · · · + δl = s > 0. If δl = 0, then
vδl
il
i1 vδ1
i1 · · · ρml−1
il−1
vδl−1
il−1
with
ς
- ς
- If x = x′ρml
il
σ−1(1) =eµ(1) = 1,
- eµ
eµ(1) = ς(1) = 1 and ∂
eµ(x) = ς(0) = 0
∂(x) = ς(cid:0)∂(x′)ρml
ς(x) = ∂(cid:0)(−1)sx′ρml−1
∂(x) = ς(cid:0)∂(x′)ρml
il
ς(x) = ∂(0) = 0.
and if δl = 1, then
∂
∂
ς
ς
il
il (cid:1) = (−1)s−1∂(x′)ρml−1
il
vil,
il
vil(cid:1) = (−1)s∂(x′)ρml−1
(cid:1) = x,
il
vil + (−1)s−1x′ρml+1
vil + x,
The result follows immediately from all these facts.
(cid:3)
Let G be a group acting on V . We consider S as a k[G]-module algebra via the
action induced by the one of G on V . Let f : k[G] × k[G] → k× be a normal cocycle
and let A = S#f k[G] be the associated crossed product. In the sequel we will use
the following
Notation 4.2. We let k[G] denote k[G]/k. Moreover:
- Given g1, . . . , gs ∈ k[G] and 1 ≤ i < j ≤ s, we set gij := gi ⊗ · · · ⊗ gj.
- Given v1, . . . , vr ∈ V and 1 ≤ i < j ≤ r, we set vij := vi · · · vj .
o
o
o
o
o
o
o
o
o
o
o
o
o
o
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
27
For all r, s ≥ 0, let
Zs = (A ⊗ k[G]
⊗s
) ⊗S A and Xrs = (A ⊗ k[G]
⊗s
) ⊗S Yr ⊗S A,
where we consider A ⊗ k[G]
⊗s
as a right S-module via
(a0wg0 ⊗ g1s)·a = a0
g0···gsawg0 ⊗ g1s.
The Xrs's and the Zs's are A-bimodules in a canonical way. Note that
Zs ≃ A ⊗ k[G]
⊗s
⊗ k[G]
and Xrs ≃ A ⊗ k[G]
⊗s
⊗ ΛrV ⊗ A.
In particular, Xrs is a free A-bimodule. Consider the diagram of A-bimodules and
A-bimodule maps
...
−δ2
µ2
Z2
X02
−δ2
µ1
Z1
X01
−δ1
µ0
Z0
X00
d0
12
d0
11
d0
10
d0
22
. . .
X12
d0
21
. . .
X11
d0
20
. . . ,
X10
where
- each δs is defined by
δ(1 ⊗ g1s ⊗S 1) := wg1 ⊗ g2s ⊗S 1
+
(−1)if (gi, gi+1) ⊗ g1,i−1 ⊗ gigi+1 ⊗ gi+2,s ⊗S 1
s−1Xi+1
+ (−1)s1 ⊗ g1,s−1 ⊗S wgs,
- for each s ≥ 0, the complex (X∗s, d∗s) is (−1)s times (Y∗, ∂∗), tensored over
S, on the right with A and on the left with A ⊗ k[G]
- for each s ≥ 0, the map µs is defined by
µ(1 ⊗ g1s ⊗ 1) := 1 ⊗ g1s ⊗S 1.
⊗s
,
Each row in this diagram is contractible as a left A-module. A contracting homo-
topy
ς 0
0s : Zs → X0s
and ς 0
r+1,s : Xrs → Xr+1,s
(r ≥ 0),
is given by
ς 0(1 ⊗ g1s ⊗S 1) := 1 ⊗ g1s ⊗ 1,
ς 0(1 ⊗ g1s ⊗S P ⊗S 1) := (−1)s1 ⊗ g1s ⊗S ς(P) ⊗S 1.
For r ≥ 0 and 1 ≤ l ≤ s, we define A-bimodule maps dl
rs : Xrs → Xr+l−1,s−l,
recursively on l and r, by:
dl(x) :=
ς 0
δ
−ς 0
µ(x)
d1
j=1 ς 0
j=0 ς 0
−Pl−1
−Pl−1
d0(x)
dl−j
dl−j
dj(x)
dj(x)
if l = 1 and r = 0,
if l = 1 and r > 0,
if 1 < l and r = 0,
if 1 < l and r > 0,
for x = 1 ⊗ g1s ⊗ v1r ⊗ 1.
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
28
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
Theorem 4.3. There is a resolution of A as an A-bimodule
A
−µ
X0
d1
X1
d2
X2
d3
X3
d4
X4
d5
. . . ,
where µ : X00 → A is the multiplication map,
Xn = Mr+s=n
Xrs
and dn =
dl
0n +
nXl=1
nXr=1
n−rXl=0
dl
r,n−r.
Proof. See [G-G2, Appendix A].
(cid:3)
Proposition 4.4. The maps dl vanish for all l ≥ 2. Moreover
d1(1 ⊗ g1s ⊗ v1r ⊗ 1) = wg1 ⊗ g2s ⊗ v1r ⊗ 1
+
(−1)if (gi, gi+1) ⊗ g1,i−1 ⊗ gigi+1 ⊗ gi+2,s ⊗ v1r ⊗ 1
s−1Xi=1
+ (−1)s1 ⊗ g1,s−1 ⊗ gsv1 · · · gsvr ⊗ wgs .
In particular, (X∗, d∗) is the total complex of the double complex
...
...
...
d1
03
d1
13
d1
23
d0
12
d0
11
d0
10
X02
d1
02
X01
d1
01
X00
d0
22
d0
21
d0
20
X12
d1
12
X11
d1
11
X10
d0
32
. . .
d0
31
. . .
d0
30
. . . ,
X22
d1
22
X21
d1
21
X20
Proof. The computation of d1
rs can be obtained easily by induction on r, using that
d1(x) = ς 0
δ
µ(x)
for x = 1 ⊗ g1s ⊗ 1,
and
d1(x) = −ς 0
d1
d0(x)
for r ≥ 1 and x = 1 ⊗ g1s ⊗ v1r ⊗ 1.
The assertion for dl
definition of dl
rs.
rs, with l ≥ 2, follows by induction on l and r, using the recursive
(cid:3)
4.2 A comparison map
Let A = A/k. In this subsection we introduce and study a comparison map from
∗
(X∗, d∗) to the canonical normalized Hochschild resolution (A ⊗ A
∗). It is
well known that there is an A-bimodule homotopy equivalence
⊗ A, b′
∗
θ∗ : (X∗, d∗) → (A ⊗ A
⊗ A, b′
∗)
such that θ0 = idA⊗A. It can be recursively defined by θ0 := idA⊗A and
θ(x) := (−1)r+sθ
d(x) ⊗ 1 for x = 1 ⊗ g1s ⊗ v1r ⊗ 1 with r + s ≥ 1.
Next we give a closed formula for θ∗. In order to establish this result we need to
introduce a new notation. We recursively define (wg1 ⊗ · · · ⊗ wgs ) ∗ (P1 ⊗ · · · ⊗ Pr)
by
- (wg1 ⊗ · · · ⊗ wgs ) ∗ (Q1 ⊗ · · · ⊗ Qr) := (Q1 ⊗ · · · ⊗ Qr) if s = 0,
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
29
- (wg1 ⊗ · · · ⊗ wgs ) ∗ (Q1 ⊗ · · · ⊗ Qr) := (wg1 ⊗ · · · ⊗ wgs ) if r = 0,
- If r, s ≥ 1, then (wg1 ⊗ · · · ⊗ wgs ) ∗ (Q1 ⊗ · · · ⊗ Qr) equals
(−1)i(wg1 ⊗ · · · ⊗ wgs−1 ) ∗ (gsQ1 ⊗ · · · ⊗ gsQi) ⊗ wgs ⊗ Qi+1 ⊗ · · · ⊗ Qr.
rXi=0
θ(cid:0)1 ⊗ g1s ⊗ v1r ⊗ 1(cid:1) = (−1)r Xτ ∈Sr
Proposition 4.5. We have:
sg(τ ) ⊗(cid:0)wg1 ⊗ · · · ⊗ wgs(cid:1) ∗ vτ (1r) ⊗ 1,
where Sr is the symmetric group in r elements and vτ (1r) = vτ (1) ⊗ · · · ⊗ vτ (r).
Proof. We proceed by induction on n = r + s. The case n = 0 is obvious. Suppose
that r + s = n and the result is valid for θn−1. By the recursive definition of θ and
Theorem 4.3,
θ(1 ⊗ g1s ⊗ v1r ⊗ 1) = (−1)nθ
= (−1)nθ
d(1 ⊗ g1s ⊗ v1r ⊗ 1) ⊗ 1
(d0 + d1)(1 ⊗ g1s ⊗ v1r ⊗ 1) ⊗ 1
(−1)i+rθ(g1···gsvi ⊗ g1s ⊗ v1,i−1vi+1,r ⊗ 1) ⊗ 1
=
−
rXi=1
rXi=1
s−1Xi=1
(−1)i+rθ(1 ⊗ g1s ⊗ v1,i−1vi+1,r ⊗ vi) ⊗ 1
+ (−1)nθ(wg1 ⊗ g2s ⊗ v1r ⊗ 1) ⊗ 1
+
(−1)n+iθ(1 ⊗ g1,i−1 ⊗ gigi+1 ⊗ gi+1,s ⊗ v1r ⊗ 1) ⊗ 1
+ (−1)rθ(1 ⊗ g1,s−1 ⊗ g1,s−1 ⊗ gsv1 · · · gsvr ⊗ wgs ) ⊗ 1.
The desired result follows now from the inductive hypothesis.
(cid:3)
4.3 The Hochschild cohomology
Let M be an A-bimodule and Ae the enveloping algebra of A. Applying the functor
HomAe(−, M ) to (X∗∗, d0
∗∗) and using the identifications
∗∗, d1
HomAe(Xrs, M ) ≃ Homk(k[G]
⊗s
⊗ ΛrV, M )
we obtain the double complex
...
...
...
d
03
1
13
d
1
23
d
1
d
12
0
d
11
0
d
10
0
02
X
d
02
1
01
X
d
01
1
00
X
22
d
0
21
d
0
20
d
0
12
/ X
12
d
1
11
/ X
11
d
1
10
/ X
32
d
0
. . .
31
d
0
. . .
22
/ X
22
d
1
21
/ X
21
d
1
20
/ X
d
30
0
/ . . . ,
where
rs
X
= Homk(k[G]
⊗s
⊗ ΛrV, M ),
O
O
/
O
O
/
/
/
O
O
O
O
/
O
O
/
/
/
O
O
O
O
/
O
O
/
O
O
/
30
JORGE A. GUCCIONE, JUAN J. GUCCIONE, AND CHRISTIAN VALQUI
d0(ϕ)(g1s ⊗ v1,r+1) =
+
(−1)s+i+1ϕ(g1s ⊗ v1,i−1vi+1,r+1)vi
(−1)s+i g1···gsviϕ(g1s ⊗ v1,i−1vi+1,r+1),
d1(ϕ)(g1,s+1 ⊗ v1r) = wg1 ϕ(g2,s+1 ⊗ v1r)
r+1Xi=1
r+1Xi=1
sXi=1
+
(−1)if (gi, gi+1)ϕ(g1,i−1 ⊗ gigi+1 ⊗ gi+1,s+1 ⊗ v1r)
+ (−1)s+1ϕ(g1s ⊗ gs+1v1 · · · gs+1vr)wgs+1 ,
whose total complex X
coefficients in M . The comparison map θ∗ induces a quasi-isomorphism
(M ) gives the Hochschild cohomology H∗(A, M ) of A with
∗
It is immediate that
∗
θ
:(cid:0)Homk(A
θ(ϕ)(g1s ⊗ v1r) = (−1)r Xτ ∈Sr
∗
, M ), b
∗(cid:1) → X
∗
(M ).
sg(τ )ϕ(cid:0)(wg1 ⊗ · · · ⊗ wgs ) ∗ vτ (1r)(cid:1).
where Sr is the symmetric group in r elements and vτ (1r) = vτ (1) ⊗ · · · ⊗ vτ (r).
From now on we take M = A and we write HH∗(A) instead of H∗(A, A).
4.4 Proof of the main result
We are ready to prove that the cocycle Φ is non trivial. For this it is sufficient to
show that θ(Φ) is not a coboundary. Let x1, . . . , xn, P (1)
g2n2 ,
g11, . . . g1n1 and g21, . . . g2n2 be as in Corollary 3.21. A direct computation, using
the formulas for δ1 and δ2 obtained in the proof of Proposition 3.23, shows that
g11 , . . . , P (1)
g21 , . . . , P (2)
g1n1 , P (2)
θ(Φ)(g ⊗ v) = 0
and θ(Φ)(g ⊗ h) = 0.
for g, h ∈ G and v ∈ V , and that
θ(Φ)(x1 x2) =
n1Xj=1
n2Xh=1
and
χ−1
α (g1j)f (g1j, g2h)α−1(P (1)
g1j )g1jP (2)
g2h wg1j g2h
θ(Φ)(xi xj) = 0
for 1 ≤ i < j ≤ n with (i, j) 6= (1, 2).
We next prove that θ(Φ) is not a coboundary. Let ϕ0 ∈ X 01 and ϕ1 ∈ X 10. By
definition
d1(ϕ0)(g ⊗ h) = wgϕ0(h) − f (g, h)ϕ0(gh) + ϕ0(g)wh,
d0(ϕ0)(g ⊗ v) = gvϕ0(g) − ϕ0(g)v,
d1(ϕ1)(g ⊗ v) = wgϕ1(v) − ϕ1(gv)wg,
d0(ϕ1)(v1 v2) = ϕ1(v2)v1 − v1ϕ1(v2) + v2ϕ1(v1) − ϕ1(v1)v2,
and so θ(Φ) is a coboundary if and only if there exist ϕ0 and ϕ1 such that
wgϕ0(h) − f (g, h)ϕ0(gh) + ϕ0(g)wh = 0
gvϕ0(g) − ϕ0(g)v + wgϕ1(v) − ϕ1(gv)wg = 0
[ϕ1(xj), xi] + [xj , ϕ1(xi)] = 0
for all g, h ∈ G,
for all g ∈ G and v ∈ V ,
for all i < j with (i, j) 6= (1, 2),
UNIVERSAL DEFORMATION FORMULAS AND BRAIDED MODULE ALGEBRAS
31
where, as usual, [a, b] = ab − ba, and
[ϕ1(x2), x1] + [x2, ϕ1(x1)] =
But, since wgxj = gxj wg,
n1Xj=1
n2Xh=1
χ−1
α (g1j)f (g1j, g2h)α−1(P (1)
g1j )g1jP (2)
g2h wg1j g2h .
wg1j g2h x1 = f (g1j, g2h)−1wg1j wg2h x1 = qx1
and wg1j g2h x2 = q−1x2,
if
then necessarily
ϕ1(x1) =Xg∈G
Q(1)
g wg
Xg∈Υ
where
(q − 1)(cid:0)x1Q(2)
Q(2)
g wg,
and ϕ1(x2) =Xg∈G
n1Xj=1
n2Xh=1
g (cid:1)wg =
g + q−1x2Q(1)
Djhα−1(P (1)
g1j )g1jP (2)
g2h wg1j g2h ,
Djh = χ−1
α (g1j)f (g1j, g2h) and Υ = {g1jg2h : 1 ≤ j ≤ n1 and 1 ≤ h ≤ n2},
which is impossible because α−1(P (1)
g1j )g1jP (2)
g2h ∈ k[x3, . . . , xn] \ {0}.
References
[A-S]
N. Andruskiewitsch and H. J. Schneider, Hopf algebras of order p2 and braided Hopf
algebras of order p, Journal of Algebra, vol 199 (1998) 430 -- 454.
[B-K-L-T] Y. Bespalov, T. Kerler, V. Lyubashenko and V. Turaev, Integrals for braided Hopf
[D]
algebras, Journal of Pure and Applied Algebra, vol 148 (2000) 113 -- 164.
Y. Doi, Hopf modules in Yetter Drinfeld categories, Communications in Algebra, vol 26
(1998) 3057 -- 3070.
[G-Z]
[G-G1]
[G-G2]
[F-M-S] D. Fishman, S. Montgomery and H. J. Schneider, Frobenius extensions of subalgebras
of Hopf algebras, Transactions of the American Mathematical Society, vol 349 (1997)
4857 -- 4895.
Jorge A. Guccione and Juan J. Guccione, Theory of braided Hopf Crossed Products,
Journal of Algebra, vol 261 (2003) 54 -- 101.
Jorge A. Guccione and Juan J. Guccione, Hochschild (co)homology of Hopf crossed
products, K-theory, Vol 25, (2002) 138 -- 169.
Giaquinto and James J. Zang, Bialgebra actions, twists, and universal deformation
formulas, Journal of Pure and Applied Algebra, vol 128 (1998) 133 -- 151.
V. Lyubashenko Modular tranformations for tensor categories, Journal of Pure and
Applied Algebra, vol 98 (1995) 279 -- 327.
Y. Sommerhauser, Integrals for braided Hopf algebras, preprint.
M. Takeuchi, Survey of braided Hopf algebras, Contemporary Mathematics, vol 267
(2000) 301 -- 323.
M. Takeuchi, Finite Hopf algebras in braided tensor categories, Journal of Pure and
Applied Algebra, vol 138 (1999) 59 -- 82.
S. Withersponn, Skew derivations and deformations of a family of group crossed prod-
ucts, Communications in algebra, vol 34 (2006) 4187 -- 4206.
[So]
[T1]
[T2]
[L1]
[W]
Departamento de Matem´atica, Facultad de Ciencias Exactas y Naturales, Pabell´on 1,
Ciudad Universitaria, (1428) Buenos Aires, Argentina.
E-mail address: [email protected]
Departamento de Matem´atica, Facultad de Ciencias Exactas y Naturales, Pabell´on 1,
Ciudad Universitaria, (1428) Buenos Aires, Argentina.
E-mail address: [email protected]
Pontificia Universidad Cat´olica del Per´u - Instituto de Matem´atica y Ciencias Afi-
nes, Secci´on Matem´aticas, PUCP, Av. Universitaria 1801, San Miguel, Lima 32, Per´u.
E-mail address: [email protected]
|
1601.04269 | 2 | 1601 | 2017-04-06T13:02:08 | Co-Poisson structures on polynomial Hopf algebras | [
"math.RA"
] | The Hopf dual $H^\circ$ of any Poisson Hopf algebra $H$ is proved to be a co-Poisson Hopf algebra provided $H$ is noetherian. Without noetherian assumption, it is not true in general. There is no nontrivial Poisson Hopf structure on the universal enveloping algebra of a non-abelian Lie algebra. The Poisson Hopf structures on $A=k[x_1, x_2, \cdots, x_d]$, viewed as the universal enveloping algebra of a finite-dimensional abelian Lie algebra, are exactly linear Poisson structures on $A$. The co-Poisson structures on polynomial Hopf algebra $A$ are characterized. Some correspondences between co-Poisson and Poisson structures are also established. | math.RA | math | CO-POISSON STRUCTURES ON POLYNOMIAL HOPF
ALGEBRAS
QI LOU AND QUANSHUI WU
Abstract. The Hopf dual H ◦ of any Poisson Hopf algebra H is proved to
be a co-Poisson Hopf algebra provided H is noetherian. Without noetherian
assumption, unlike it is claimed in literature, the statement does not hold. It
is proved that there is no nontrivial Poisson Hopf structure on the universal
enveloping algebra of a non-abelian Lie algebra. So the polynomial Hopf alge-
bra, viewed as the universal enveloping algebra of a finite-dimensional abelian
Lie algebra, is considered. The Poisson Hopf structures on polynomial Hopf
algebras are exactly linear Poisson structures. The co-Poisson structures on
polynomial Hopf algebras are characterized. Some correspondences between
co-Poisson and Poisson structures are also established.
7
1
0
2
r
p
A
6
]
.
A
R
h
t
a
m
[
2
v
9
6
2
4
0
.
1
0
6
1
:
v
i
X
r
a
1. Introduction
Poisson structure naturally appears in classical/quantum mechanics, in mathe-
matical physics, and in deformation theory. It is an important algebra structure in
Poisson geometry, algebraic geometry and non-commutative geometry. There are
lots of research in the related subjects.
Co-Poisson structure is a dual concept of Poisson structure in categorial point of
view. It arises also in mathematics and mathematical physics naturally as explained
in the next two paragraphs.
Let G be a Lie group and O(G) be its algebra of functions. A Lie group G is
said to be a Poisson Lie group if O(G) is a Poisson Hopf algebra. It is well-know
that the category of connected and simply-connected Lie groups is equivalent to
the category of finite-dimensional Lie algebras.
In this case, O(G) is identified
with the Hopf dual U (g)◦ of the universal enveloping algebra U (g), where g is
the corresponding Lie algebra of G. The Poisson counterpart of this fact holds
also, namely, the category of connected and simply-connected Poisson Lie groups is
equivalent to the category of finite-dimensional Lie bialgebras ([KS, Theorem 3.3.1]
or [Dr, Theorem 1]). However, the Lie bialgebra structures on any Lie algebra g
is in one-to-one correspondence with the co-Poisson Hopf structures on U (g) ([CP,
Proposition 6.2.3]).
On the other hand, to quantize a Lie group or Lie algebra one should equip it
with an extra structure, namely, a Poisson Lie group structure or Lie bialgebra
structure, respectively. Therefore co-Poisson structure naturally appears in the
theory of quantum groups and in mathematical physics.
If G is a connected and simply-connected Poisson Lie group and g is the cor-
responding Lie bialgebra, then the Poisson Hopf structure on U (g)◦ ∼= O(G) is
the dual of the co-Poisson Hopf structures on U (g). In [OP], the authors proved
that the dual Hopf algebra U (g)◦ of U (g) is a Poisson Hopf algebra for any finite-
dimensional Lie bialgebra g. In fact, in general, as stated in [KS, Proposition 3.1.5]
earlier, the dual Hopf algebra of any co-Poisson Hopf algebra is a Poisson Hopf
2010 Mathematics Subject Classification. Primary 17B63, 16W10, 16S30.
Key words and phrases. Poisson algebra, co-Poisson coalgebra, Poisson Hopf algebra, co-
Poisson Hopf algebra.
1
2
QI LOU AND QUANSHUI WU
algebra. A complete proof is given in a recent paper by Oh [Oh, Theorem 2.2].
The dual proposition that the dual Hopf algebra of any Poisson Hopf algebra is a
co-Poisson Hopf algebra is also stated in [KS, Proposition 3.1.5]. Unfortunately,
this claim is not true in general as showed in our Example 3.6. Under an addi-
tional assumption that the algebra is noetherian, we prove the statement is true in
Proposition 3.5.
We prove that there is no nontrivial Poisson Hopf structure on the universal
enveloping algebra of a non-abelian Lie algebra in Proposition 2.5. So, we turn
to consider in latter sections the abelian case, i.e., the polynomial Hopf algebra
A = k[x1, x2, · · · , xd], viewed as the universal enveloping algebra of an abelian
Lie algebra of dimension d. The Poisson Hopf structures on A = k[x1, x2, · · · , xd]
are exactly linear Poisson structures on A (see Proposition 5.3). By establishing a
reciprocity law between two linear maps of A → A⊗A for A = k[x1, x2, · · · , xd] (see
Proposition 4.3), all co-Poisson coalgebra and co-Poisson Hopf algebra structures on
A = k[x1, x2, · · · , xd] are described in Theorem 4.7 and Proposition 5.4 respectively.
In particular, the co-Poisson coalgebra structures on A = k[x, y] are given by the
linear maps I : A → k(x ⊗ y − y ⊗ x) (see Proposition 4.8). By using the algebra of
divided power series, the co-Poisson coalgebra structures on A = k[x1, x2, · · · , xd]
are showed to be in one-to-one correspondence with the Poisson algebra structures
on A = k[[x1, x2, · · · , xd]], the algebra of formal power series in Theorem 5.8.
The paper is organized as follows. The definitions of (co-)Poisson (co)algebras
are recalled in Section 2. Some preliminary results and examples are also given
in Section 2. In Section 3, we establish some dual properties between co-Poisson
structures and Poisson structures. In Section 4, we characterize co-Poisson coalge-
bra structures on polynomial Hopf algebra. In Section 5, we characterize co-Poisson
Hopf structures on polynomial Hopf algebras.
Convention. Let k be a base field. All vector spaces, algebras, coalgebras, and
Hopf algebras are over k. All linear maps mean k-linear. Unadorned ⊗ means ⊗k.
Let V be a vector space. Let tn : V ⊗n → V ⊗n (n ∈ N+) be the linear map given
by v1 ⊗ · · · ⊗ vn 7→ vn ⊗ v1 ⊗ · · · ⊗ vn−1. For convenience, let (cid:9)= 1 + t3 + t2
3.
Suppose (C, ∆, ε) is a coalgebra where ∆ is the comultiplication and ε is the
counit. We frequently use the sigma notation
∆(c) =X c1 ⊗ c2 and (∆ ⊗ 1)∆(c) =X c1 ⊗ c2 ⊗ c3,
Let ∆(2) = (∆ ⊗ 1) ◦ ∆ = (1 ⊗ ∆) ◦ ∆ : C → C ⊗ C ⊗ C, and ∆′ = ∆ − t2 ◦ ∆
where P is often omitted in the computations.
be the cocommutator.
Suppose (A, µ, η) is an algebra where µ is the multiplication and η is the unit.
For any a, b ∈ A, [a, b] = ab − ba is the commutator.
2. Poisson structures and co-Poisson structures
2.1. Poisson algebras and Poisson Hopf algebras.
Definition 2.1. [Li, Wei] An algebra A equipped with a linear map {−, −} :
A ⊗ A → A is called a Poisson algebra if
(1) A with {−, −} : A ⊗ A → A is a Lie algebra.
(2) {−, c} : A → A is a derivation with respect to the multiplication of A for
all c ∈ A, that is, {ab, c} = a{b, c} + {a, c}b for all a, b ∈ A.
It should be noted that we don't assume that A is commutative in general. As
showed in [FL, Theorem 1.2], if A is prime and not commutative, then any nontrivial
Poisson structure {−, −} on A is the commutator bracket [−, −] up to some scalar
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
3
α in the center of the Martindale right quotient ring of A, where α is essentially
determined by a bimodule morphism from some nonzero ideal to A.
Proposition 2.2. [FL, Theorem 1.2] Let A be a Poisson algebra with {A, A} 6= 0
and [A, A] 6= 0. If A is prime, then
(1) for any a, b ∈ A, there is an isomorphism of A-bimodules
f(a,b) : A[a, b]A → A{a, b}A, [a, b] 7→ {a, b};
(2) in the Martindale right quotient ring of A, {−, −} = α[−, −], where α is
represented by f(a,b) for any a, b ∈ A such that [a, b] 6= 0.
Proof. (1) By using Leibniz rule and calculating {ac, bd} in both variables (see [Vo,
Theorem 1] or [FL, Lemma 1.1]), it follows that, for all a, b, c, d ∈ A,
[a, b]{c, d} = {a, b}[c, d].
Then, for all a, b, c, d, e ∈ A, [a, b]{ec, d} = {a, b}[ec, d], which implies that
[a, b]e{c, d} = {a, b}e[c, d].
(2.1)
(2.2)
Since [A, A] 6= 0, {A, A} 6= 0 and A is prime, then [a, b] = 0 if and only if
{a, b} = 0 for any a, b ∈ A.
Suppose [a, b] 6= 0. If Pn
Xi=1
0 =
n
Then Pn
i=1 xi[a, b]yi = 0, then for any c ∈ A,
({a, b}c)xi[a, b]yi =
[a, b]c(xi{a, b}yi).
n
Xi=1
i=1 xi{a, b}yi = 0 as A is prime and [a, b] 6= 0. This shows that the map
f(a,b) : A[a, b]A → A{a, b}A,
xi[a, b]yi 7→
xi{a, b}yi
n
n
Xi=1
Xi=1
is well-defined. So, f(a,b) is an A-bimodule morphism, and is in fact an isomorphism.
(2) If [a, b] 6= 0, then the bimodule morphism f(a,b) : A[a, b]A → A, [a, b] 7→ {a, b}
represents an element in the center of the Martindale right quotient ring of A ([MR,
10.3.5] or [He, Chapter 1, §3]). It follows from (2.2) that f(a,b) and f(c,d) represent
the same element, say α, in the Martindale right quotient ring of A if both [a, b]
and [c, d] are nonzero. Then {−, −} = α[−, −] in the Matindale right quotient ring
of A.
(cid:3)
By Proposition 2.2, any Poisson algebra structure {−, −} on the Weyl algebra
An(k) or the matrix algebra Mn(k) is α[−, −] for some α ∈ k.
Commutative Poisson algebras appear naturally in geometry and algebra. For
more examples of commutative Poisson algebras, see [LPV], [KS] and [LWW].
Definition 2.3. ([CP, Definition 6.2.1]) Let (H, µ, η, ∆, ε, S) be a Hopf algebra
with a linear map {−, −} : H ⊗ H → H. Then H is called a Poisson Hopf algebra
if
(1) (H, {−, −}) is a Poisson algebra.
(2) The structures are compatible in the sense that, for all a, b ∈ H,
∆ ({a, b}) =X{a1, b1} ⊗ a2b2 +X a1b1 ⊗ {a2, b2}.
Let A and B be two Poisson algebras. An algebra morphism f : A → B is called
a Poisson algebra morphism if f ({a, b}A) = {f (a), f (b)}B for all a, b ∈ A.
Equation (2.3) means that ∆ : H → H ⊗ H is a Poisson algebra morphism when
H is commutative by the following lemma.
(2.3)
4
QI LOU AND QUANSHUI WU
Lemma 2.4. Let A and B be two commutative Poisson algebras. There is naturally
a Poisson structure {−, −}A⊗B on the tensor product algebra A ⊗ B given by
{a ⊗ b, a′ ⊗ b′}A⊗B = {a, a′}A ⊗ bb′ + aa′ ⊗ {b, b′}B.
It is easy to see that there is no nontrivial Poisson Hopf algebra structure on
any group algebra k[G]. There is also no nontrivial Poisson Hopf structure on the
universal enveloping algebra U (g) if g is a non-abelian Lie algebra.
Proposition 2.5. Let g be a non-abelian Lie algebra over a field of characteristic
6= 2. Then there is no nontrivial Poisson Hopf structure on U (g).
Proof. Suppose {−, −} : U (g)⊗U (g) → U (g) is a non-trivial Poisson Hopf structure
on U (g). Since g is non-abelian and U (g) is prime, then by Proposition 2.2, for any
a, b ∈ g, [a, b] = 0 if and only if {a, b} = 0. Recall the compatible condition (2.3),
i.e., for all x, y ∈ U (g),
∆{x, y} = x1y1 ⊗ {x2, y2} + {x1, y1} ⊗ x2y2.
This implies that if both x and y are primitive then so is {x, y}. If we take x = ab
and y = cd for a, b, c, d ∈ g, the compatible condition (2.3) will imply that
[b, c] ⊗ {a, d} + [b, d] ⊗ {a, c}
is skew-symmetric for all a, b, c, d ∈ g. So, [b, c] ⊗ {a, b} is skew-symmetric by taking
d = b. Suppose [b, c] 6= 0 as [g, g] 6= 0. Since both [b, c] and {a, b} are primitive, then
{a, b} = 0 for all a ∈ g as char k 6= 2. In particular, {c, b} = 0, and so, [c, b] = 0,
which is a contradiction.
(cid:3)
As pointed out in [KS, Remark 3.1.4], if H is a commutative Poisson Hopf
algebra, then the counit ε : H → k is a Poisson algebra morphism, and the antipode
S : H → H is a Poisson algebra anti-morphism. Here is a proof of the facts (see
also [Oh, Lemma 4.2]).
Lemma 2.6. Let H be a Poisson Hopf algebra. Then the counit ε : H → k is a
Poisson algebra morphism. The antipode S : H → H is a Poisson algebra anti-
morphism provided H is commutative.
Proof. We need to show ε ({a, b}) = 0 and S ({a, b}) = {S(b), S(a)} for all a, b ∈ H.
Since ε(h) = ε(h1)ε(h2) for all h ∈ H, by (2.3),
ε ({a, b}) = ε ({a1, b1}) ε(a2b2) + ε(a1b1)ε ({a2, b2}) = 2ε ({a, b}) .
Thus ε ({a, b}) = 0. For the second assertion,
{S(b), S(a)} ={S(b1), S(a)}b2S(b3)
= − S(b1){b2, S(a)}S(b3)
= − S(b1){b2, S(a1)}a2S(a3)S(b3)
=S(b1)S(a1){b2, a2}S(a3)S(b3)
= − S(b1)S(a1)b2a2S ({b3, a3})
=S ({a, b}) ,
where the second last " = " follows from the fact {a1, b1}S(a2b2)+a1b1S ({a2, b2}) =
0 for all a, b ∈ H, and the last " = " follows from the commutativity of H.
(cid:3)
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
5
2.2. Co-Poisson coalgebras and co-Poisson Hopf algebras. The concepts of
co-Poisson coalgebra and co-Poisson Hopf algebra are introduced in [CP, Chapter
6] to study quantizations of Lie bialgebras. They are dual to Poisson algebra and
Poisson Hopf algebra.
Definition 2.7. ([CP, Definition 6.2.2]) A coalgebra (C, ∆, ε) equipped with a
linear map q : C → C ⊗ C is called a co-Poisson coalgebra if
(1) C with q : C → C ⊗ C is a Lie coalgebra, that is,
(1 + t2) ◦ q = 0,
(1 + t3 + t2
3) ◦ (q ⊗ 1) ◦ q = 0.
(2) (∆ ⊗ 1)q = (1 ⊗ q)∆ − t2
3(q ⊗ 1)∆.
(skew-symmetric)
(co-Jacobi identity)
(co-Leibniz rule)
In co-Poisson coalgebra C, we use the sigma notations
∆(c) =X c1 ⊗ c2, q(c) =X c(1) ⊗ c(2) and (q ⊗ 1)q(c) =X c(1) ⊗ c(2) ⊗ c(3)
where P is often omitted in the computations. Then,
(1 ⊗ q)q(c) = (1 ⊗ q)(−c(2) ⊗ c(1)) = −c(3) ⊗ c(1) ⊗ c(2) = c(3) ⊗ c(2) ⊗ c(1).
Remark 2.8. By using the sigma notation, the co-Leibniz rule reads as
X c(1)1 ⊗ c(1)2 ⊗ c(2) =X c1 ⊗ c2(1) ⊗ c2(2) −X c1(2) ⊗ c2 ⊗ c1(1)
for all c ∈ C. It is equivalent to
If the coalgebra C is cocommutative, then the co-Leibniz rule is also equivalent to
(1 ⊗ ∆)q = (q ⊗ 1)∆ − t3(1 ⊗ q)∆.
(2.4)
(∆ ⊗ 1)q = (1 − t3)(1 ⊗ q)∆.
(2.5)
The cocommutator ∆′ gives a co-Poisson structure on any coalgebra (C, ∆, ε).
It follows from the co-Leibniz rule that there is no nontrivial co-Poisson coalgebra
structure on any group coalgebra k[G] .
Definition 2.9. Let (C, ∆, ε) be a coalgebra. A linear map d : C → C is called a
coderivation if, for all c ∈ C,
∆(d(c)) = d(c1) ⊗ c2 + c1 ⊗ d(c2).
Example 2.10. Let C be a cocommutative coalgebra, d1 and d2 be two coderiva-
tions of C such that d1d2 = d2d1. Let q : C → C ⊗ C be the map c 7→ q(c),
where
q(c) = d1(c1) ⊗ d2(c2) − d2(c1) ⊗ d1(c2).
Then (C, q) is a co-Poisson coalgebra.
Dual to the fact {a, −} : A → A is a derivation in any Poisson algebra A is the
following.
Lemma 2.11. Let (C, ∆, ε, q) be a co-Poisson coalgebra. Then for any f ∈ C∗,
(f ⊗ 1)q : C → C and (1 ⊗ f )q : C → C are coderivations of C.
Dual to the fact {1A, a} = {a, 1A} = 0 in any Poisson algebra A is the following.
Lemma 2.12. Let (C, ∆, ε, q) be a co-Poisson coalgebra. Then
(ε ⊗ 1)q = (1 ⊗ ε)q = 0, that is, ε(c(1))c(2) = c(1)ε(c(2)) = 0 for all c ∈ C.
Definition 2.13. [CP, Definition 6.2.2] A Hopf algebra (H, µ, η, ∆, ε, S) equipped
with a linear map q : H → H ⊗ H is called a co-Poisson Hopf algebra if
(1) H with q : H → H ⊗ H is a co-Poisson coalgebra.
6
QI LOU AND QUANSHUI WU
(2) q is a ∆-derivation, that is, for all a, b ∈ H,
q(ab) = q(a)∆(b) + ∆(a)q(b).
(2.6)
Co-Poisson Hopf structures on U (g) are known.
Proposition 2.14. [CP, Proposition 6.2.3] Let g be a Lie algebra over a field
k of characteristic zero. Then the co-Poisson Hopf structures on the universal
enveloping algebra U (g) are determined uniquely by the Lie bialgebra structures on
g.
Let C and D be two co-Poisson coalgebras. A coalgebra morphism g : C → D is
called a co-Poisson coalgebra morphism if (g ⊗ g)(qC (c)) = qD (g(c)) for all c ∈ C.
Equation (2.6) means that µ : H ⊗ H → H is a co-Poisson coalgebra morphism
provided H is cocommutative by Lemma 2.15.
If C and D are two coalgebras, then C ⊗ D is a coalgebra with
∆C⊗D = (1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D),
i.e., ∆C⊗D(a ⊗ b) = a1 ⊗ b1 ⊗ a2 ⊗ b2.
Lemma 2.15. Let (C, qC ) and (D, qD) be two cocommutative co-Poisson coalge-
bras. Then (C ⊗ D, qC⊗D) is a co-Poisson coalgebra, with qC⊗D being the compo-
sition
C ⊗ D
qC ⊗∆D+∆C ⊗qD
−−−−−−−−−−−→ C ⊗ C ⊗ D ⊗ D
1⊗t2⊗1
−−−−−→ C ⊗ D ⊗ C ⊗ D,
that is,
qC⊗D(a ⊗ b) = a(1) ⊗ b1 ⊗ a(2) ⊗ b2 + a1 ⊗ b(1) ⊗ a2 ⊗ b(2).
The dual form of Lemma 2.6 is the following.
Lemma 2.16. Let (H, µ, η, ∆, ε, S, q) be a co-Poisson Hopf algebra.
(1) η is a co-Poisson coalgebra morphism.
(2) If H is cocommutative, then S is a co-Poisson coalgebra anti-morphism.
Proof. (1) This is trivial as q(1H ) = 0 by (2.6).
(2) We need to show q (S(h)) = S(h(2)) ⊗ S(h(1)) for all h ∈ H. It follows from
the co-Leibniz rule that for any h ∈ H,
h1(1)1 ⊗ h1(1)2 ⊗ h2 ⊗ h1(2) = h1 ⊗ h2(1) ⊗ h3 ⊗ h2(2) − h1(2) ⊗ h2 ⊗ h3 ⊗ h1(1).
Since S(h1(1)1)h1(1)2S(h2) ⊗ S(h1(2)) = 0 by Lemma 2.12, then
S(h1)h2(1)S(h3) ⊗ S(h2(2)) = S(h1(2))ε(h2) ⊗ S(h1(1)) = S(h(2)) ⊗ S(h(1)),
where the last " = " is derived from the co-Leibniz rule (2.4) and Lemma 2.12.
On the other hand, by the co-Leibniz rule,
h1 ⊗ h2 ⊗ h3(1)1 ⊗ h3(1)2 ⊗ h3(2) ⊗ h4
=h1 ⊗ h2 ⊗ h3 ⊗ h4(1) ⊗ h4(2) ⊗ h5 − h1 ⊗ h2 ⊗ h3(2) ⊗ h4 ⊗ h3(1) ⊗ h5.
By Lemma 2.12, S(h1)h3(2)S(h4) ⊗ S(h2)h3(1)1S(h3(1)2) = 0. Thus
S(h1)h3(1)S(h5) ⊗ S(h2)h3(2)S(h4)
=S(h1)h4(2)S(h5) ⊗ S(h2)h3S(h4(1))
=S(h1)h2(2)S(h3) ⊗ S(h2(1))
=S(h(1)) ⊗ S(h(2)).
It remains to show that S(h1)h3(2)S(h5) ⊗ S(h2)h3(1)S(h4) = q (S(h)). Since
0 = q(h1S(h2)) = q(h1)∆ (S(h2)) + ∆(h1)q (S(h2)) for any h ∈ H, then
h1(1)S(h3) ⊗ h1(2)S(h2) + h1S(h3)(1) ⊗ h2S(h3)(2) = 0.
Hence
h1 ⊗ h3S(h5)(1) ⊗ h2 ⊗ h4S(h5)(2) + h1 ⊗ h3(1)S(h5) ⊗ h2 ⊗ h3(2)S(h4) = 0.
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
7
Thus
S(h1)h3(2)S(h5) ⊗ S(h2)h3(1)S(h4)
=S(h1)h3S(h5)(1) ⊗ S(h2)h4S(h5)(2)
=S(h1)h2S(h5)(1) ⊗ S(h3)h4S(h5)(2)
=S(h)(1) ⊗ S(h)(2)
=q (S(h)) ,
where the cocommutativity is used for the second " = ".
(cid:3)
The dual form of (2.1) for co-Poisson coalgebras is the following proposition.
Proposition 2.17. Let (C, ∆, ε) be a co-Poisson coalgebra with co-Poisson struc-
ture q. Then
(q ⊗ ∆′) ◦ ∆ = (∆′ ⊗ q) ◦ ∆.
Proof. It follows by using the co-Leibniz rule to calculate
(∆ ⊗ ∆)q(a) = (∆ ⊗ 1 ⊗ 1)(1 ⊗ ∆)q(a) = (1 ⊗ 1 ⊗ ∆)(∆ ⊗ 1)q(a).
(cid:3)
The dual form of (2.2) for co-Poisson coalgebras is the following.
Corollary 2.18. Let (C, ∆, ε) be a co-Poisson coalgebra with co-Poisson structure
q. Then
(∆′ ⊗ (1 ⊗ q)∆) ◦ ∆ = (q ⊗ (1 ⊗ ∆′)∆) ◦ ∆,
i.e, for any a ∈ C,
a1 ⊗ a2 ⊗ a3 ⊗ q(a4) − a2 ⊗ a1 ⊗ a3 ⊗ q(a4)
= q(a1) ⊗ a2 ⊗ a3 ⊗ a4 − q(a1) ⊗ a2 ⊗ a4 ⊗ a3.
Poisson and co-Poisson structures on 4-dimensional Sweedler Hopf algebra are
clear.
Example 2.19. Let H4 = khx, g x2 = 0, g2 = 1, xg = −gxi be the 4-dimensional
Sweedler Hopf algebra, where the coalgebra structure is given by ∆(g) = g ⊗ g,
∆(x) = x⊗1+g⊗x, ε(g) = 1, ε(x) = 0; and the antipode is given by S(g) = g = g−1,
S(x) = −gx. Assume char k 6= 2.
(1) Any Poisson structure on H4 is given by {g, x} = λx+µgx for some λ, µ ∈ k.
(2) There is no nontrivial Poisson Hopf structure on H4. In fact, if the Poisson
structure given by {g, x} = λx + µgx is a Poisson Hopf structure, then by
applying (2.3) with a = g, b = x and a = x, b = gx we get λ = µ = 0.
(3) Any co-Poisson structure q on H4 is given by
q(1) = q(g) = 0,
q(x) = α(1 ⊗ x − x ⊗ 1 + x ⊗ g − g ⊗ x),
q(gx) = β(1 ⊗ gx − gx ⊗ 1 + gx ⊗ g − g ⊗ gx)
for some α, β ∈ k. In fact, for any h ∈ H4, we may assume
q(h) = α1(1 ⊗ x − x ⊗ 1) + α2(1 ⊗ g − g ⊗ 1) + α3(1 ⊗ gx − gx ⊗ 1)
+ α4(x ⊗ g − g ⊗ x) + α5(x ⊗ gx − gx ⊗ x) + α6(g ⊗ gx − gx ⊗ g)
as q(h) is skew-symmetric. It follows from (ε ⊗ 1)q = 0 (Lemma 2.12) that
α1 = α4, α2 = 0, α3 = −α6.
Then, the co-Leibniz rule implies that q(1) = q(g) = 0,
q(x) = α(1 ⊗ x − x ⊗ 1 + x ⊗ g − g ⊗ x),
8
QI LOU AND QUANSHUI WU
q(gx) = β(1 ⊗ gx − gx ⊗ 1 + gx ⊗ g − g ⊗ gx)
for some α, β ∈ k.
(4) There is no nontrivial co-Poisson Hopf structure on H4. In fact, suppose a
co-Poisson structure q as given in (3) is a co-Poisson Hopf structure on H4,
then α = β = 0 by the equations
q(gx) = q(g)∆(x) + ∆(g)q(x) and 0 = q(x2) = q(x)∆(x) + ∆(x)q(x).
3. Dual properties between Poisson and co-Poisson Hopf algebras
As the vector space dual of any coalgebra is an algebra, the dual of any co-Poisson
coalgebra is a Poisson algebra.
Proposition 3.1.
(1) Suppose (C, ∆, ε) is a coalgebra, q : C → C ⊗ C is a linear map. Then
(C, q) is a co-Poisson coalgebra if and only if (C∗, q∗) is a Poisson algebra.
(2) Suppose σ : C → D is a linear map between co-Poisson coalgebras. Then
σ : C → D is a co-Poisson coalgebra morphism if and only if σ∗ : D∗ → C∗
is a Poisson algebra morphism.
Proof. (1) Note that q∗ is the map q∗ : C∗ ⊗ C∗ ⊆ (C ⊗ C)∗ → C∗, f ⊗ g 7→ {f, g},
where {f, g} ∈ C∗ is the map C → k, c 7→ (f ⊗ g)q(c) = f (c(1))g(c(2)).
Note that ({f, g} + {g, f })(c) = (f ⊗ g)(1 + t2)q(c).
It is easy to see that
{−, −}C ∗ : C∗ × C∗ → C∗ is skew-symmetric if and only if q is skew-symmetric.
Since {{f, g}, h}(c) = (f ⊗ g ⊗ h)(q ⊗ 1)q(c), then
(cid:9) {{f, g}, h}(c) = (f ⊗ g ⊗ h)(1 + t3 + t2
3)(q ⊗ 1)q(c).
It follows that {−, −}C ∗ satisfies the Jacobi identity if and only if q satisfies the
co-Jacobi identity.
Similarly, since {f ∗ g, h}(c) = (f ⊗ g ⊗ h) ((∆ ⊗ 1)q(c)) and
((f ∗ {g, h}) − ({h, f } ∗ g))(c) = (f ⊗ g ⊗ h)((1 ⊗ q)∆ − t2
3(q ⊗ 1)∆)(c),
then {f ∗ g, h} = f ∗ {g, h} + {f, h} ∗ g if and only if (∆⊗ 1)q = (1 ⊗ q)∆− t2
3(q ⊗ 1)∆.
(2) It is well known that σ : C → D is a coalgebra morphism if and only if
σ∗ : D∗ → C∗ is an algebra morphism. Since
(σ∗{f, g}D∗ − {σ∗f, σ∗g}C ∗)(c) = (f ⊗ g)(∆Dσ − (σ ⊗ σ)∆C )(c),
then σ∗{−, −}D∗ = {−, −}C ∗(σ∗ ⊗ σ∗) if and only if ∆Dσ = (σ ⊗ σ)∆C . The proof
is finished.
(cid:3)
If A is an algebra, then A◦ = {f ∈ A∗ ker f contains a cofinite ideal of A} is a
coalgebra, which is called the finite dual of A. Suppose (A, p = {−, −}) is a Poisson
algebra. Example 3.2 shows that (A◦, q = p∗) may not be a co-Poisson coalgebra
because p∗ : A∗ → (A ⊗ A)∗ may not be restricted to a map p∗
A◦ : A◦ → A◦ ⊗ A◦.
Example 3.2. Let A = k[x1, x2, · · · , xn, · · · ] be a polynomial algebra with infin-
itely many variables {xi i ≥ 1}. Let p(xi ⊗ xj) = {xi, xj } = 1 for all i < j.
Then p gives a Poisson algebra structure on A. Suppose ε : A → k, xi 7→ 0, is the
augmentation map. Then ε ∈ A◦, but p∗(ε) = εp /∈ (A ⊗ A)◦.
In fact, if εp ∈ (A ⊗ A)◦, then there is a cofinite ideal I of A such that I ⊗ A +
A ⊗ I ⊆ ker(εp). Then, {I, A} ⊆ ker ε. Since A/I is finite-dimensional, there exists
a nonzero linear polynomial in I, which will imply that 1 ∈ {I, A}. It contradicts
to {I, A} ⊆ ker ε.
But if A is left or right noetherian, then we have the following positive conclusion.
Proposition 3.3. Let (A, p = {−, −}) be a Poisson algebra, and f : A → B be a
Poisson algebra morphism.
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
9
(1) If A is a (left or right) noetherian algebra, then (A◦, q = p∗) is a co-Poisson
coalgebra.
(2) If both A and B are (left or right) noetherian, then f ∗ : B◦ → A◦ is a
co-Poisson coalgebra morphism.
Proof. It suffices to show that p∗ : A∗ → (A ⊗ A)∗ restricts to a map
p∗A◦ : A◦ → A◦ ⊗ A◦ ∼= (A ⊗ A)◦.
Suppose f ∈ A◦ and I ⊆ ker f is a cofinite ideal of A. Since A is left or
right noetherian, then I/I 2 is a finitely generated left or right A/I-module. Hence
dimk I/I 2 < ∞ as dimk A/I < ∞. So, I 2 is also a cofinite ideal of A. Let J =
I 2 ⊗ A + A ⊗ I 2. Then J is a cofinite ideal of A ⊗ A, and
p∗(f )(J) = f p(J) ⊆ f (I) = {0}.
It follows that p∗(f ) ∈ (A ⊗ A)◦.
(cid:3)
Note that A◦ = 0 for the Weyl algebra A = An(k). So, even in noetherian case
(A, p) may not be a Poisson algebra when (A◦, q = p∗) is a co-Poisson coalgebra.
In [OP], the authors prove that the Hopf dual H ◦ of a co-Poisson Hopf algebra
H is a Poisson Hopf algebra when H is an almost normalizing extension over k. In
fact, this is true in general as stated in [KS, Proposition 3.1.5]. We give a proof
here. Oh gives a proof also in a recent paper ([Oh, Theorem 2.2]).
Proposition 3.4. Let (H, µ, η, ∆, ε, S, q) be a co-Poisson Hopf algebra. Then the
Hopf dual H ◦ is a Poisson Hopf algebra.
Proof. It is well-know that H ◦ is a Hopf algebra ([Sw, Section 6.2], [Mo, Theorem
9.1.3]). By Proposition 3.1, to show H ◦ is a Poisson algebra it suffices to show that
q∗ : (H ⊗ H)∗ → H ∗ restricts to a map q◦ : H ◦ ⊗ H ◦ ∼= (H ⊗ H)◦ → H ◦, that is,
q∗ ((H ⊗ H)◦) ⊆ H ◦.
Suppose F ∈ (H ⊗ H)◦ and J ⊆ ker F is a cofinite ideal in H ⊗ H. Since
∆ is an algebra morphism, ∆−1(J) is an ideal of H.
It follows from q(ab) =
q(a)∆(b) + ∆(a)q(b) that I = q−1(J) ∩ ∆−1(J) is an ideal of H. Since the linear
map
H ⊗ H
H ⊗ H
,
J
h 7→(cid:16)q(h), ∆(h)(cid:17)
H/I →
⊕
J
is injective, I is a cofinite ideal of H. Note that q∗(F )(I) = F (q(I)) ⊆ F (J) = 0,
i.e., I ⊆ ker q∗(F ). It follows that q∗(F ) ∈ H ◦.
To finish the proof, we need to show that µ∗ : H ◦ → H ◦ ⊗ H ◦ satisfies the
compatible condition (2.3), that is, for all f, g ∈ H ◦,
µ∗({f, g}H ◦ ) = {f1, g1}H ◦ ⊗ (f2 ∗ g2) + (f1 ∗ g1) ⊗ {f2, g2}H ◦ .
This is true, because, for any x, y ∈ H, on one hand, by (2.6),
µ∗({f, g}H ◦ )(x ⊗ y)
={f, g}H ◦(xy) = (f ⊗ g)q(xy) = f ((xy)(1))g((xy)(2))
=f ((x(1)y1)g((x(2)y2) + f ((x1y(1))g((x2y(2)),
on the other hand,
({f1, g1}H ◦ ⊗ (f2 ∗ g2) + (f1 ∗ g1) ⊗ {f2, g2}H ◦ )(x ⊗ y)
=({f1, g1}H ◦ (x) (f2 ∗ g2)(y) + (f1 ∗ g1)(x) {f2, g2}H ◦ (y)
=f1(x(1))g1(x(2))f2(y1)g2(y2) + f1(x1)g1(x2)f2(y(1))g2(y(2))
=f ((x(1)y1)g((x(2)y2) + f ((x1y(1))g((x2y(2)).
(cid:3)
10
QI LOU AND QUANSHUI WU
The following is also stated in [KS, Proposition 3.1.5] without noetherian hy-
pothesis. Without this hypothesis, it is not true as showed in Example 3.6.
Proposition 3.5. Let (H, µ, η, ∆, ε, S, p) be a left or right noetherian Poisson Hopf
algebra. Then the Hopf dual H ◦ is a co-Poisson Hopf algebra.
Proof. By Proposition 3.3, H ◦ is a co-Poisson coalgebra. We only need to check
the compatible condition (2.6), that is, for all f, g ∈ H ◦,
p∗(f ∗ g) = p∗(f )µ∗(g) + µ∗(f )p∗(g).
This is true, because for any x, y ∈ H,
p∗(f ∗ g)(x ⊗ y) = (f ∗ g)({x, y})
=f ({x1, y1}) g(x2y2) + f (x1y1) g({x2, y2}),
and
(p∗(f )µ∗(g) + µ∗(f )p∗(g))(x ⊗ y)
=(f(1) ∗ g1 ⊗ f(2) ∗ g2)(x ⊗ y) + (f1 ∗ g(1) ⊗ f2 ∗ g(2))(x ⊗ y)
=f(1)(x1)g1(x2)f(2)(y1)g2(y2) + f1(x1)g(1)(x2)f2(y1)g(2)(y2)
=f ({x1, y1}) g(x2y2) + f (x1y1) g({x2, y2}).
Example 3.6. Let A = k[x1, · · · , xd, · · · ] be a polynomial algebra with infinitely
many variables {xi i ≥ 1}, which is a Hopf algebra viewed as the enveloping
algebra of the abelian Lie algebra g = kx1 ⊕ kx2 ⊕ · · · ⊕ kxd ⊕ · · · . Let
(cid:3)
{x1, xi} = 0 for all i ∈ N,
{xi, xj} =(x1,
0,
j = i + 1,
otherwise
for all 1 < i < j ∈ N.
Then {{xi, xj}, xk} = 0 for all i, j, k. As in [LPV, Proposition 1.8], A is endowed
with a Poisson algebra structure. It is easy to check by induction on the degree of
homogeneous elements that (2.3) holds. So, A is a Poisson Hopf algebra.
We assert that A◦ is not a co-Poisson Hopf algebra by showing that {−, −}∗(A◦) *
A◦ ⊗ A◦. Let f : A → k be the linear map given by
(f (1A) = f (x1) = 1k,
f (a) = 0 for all other monic monomials a ∈ A.
1, x2, x3, · · · }i ⊆ ker f . Note that I is a cofinite ideal of A, and so,
Then I = h{x2
f ∈ A◦. Suppose {−.−}∗(f ) ∈ A◦ ⊗ A◦ and
n
with Ii ∈ ker gi and Ji ∈ ker hi are all cofinite ideals of A. Then
{−.−}∗(f ) =
gi ⊗ hi,
Xi=1
J =
(Ii ∩ Ji).
n
\i=1
i=1(gi ⊗ hi)(J ⊗ A) = 0. It
is a cofinite ideal of A, and {−, −}∗(f )(J ⊗ A) = Pn
follows that {J, A} ⊆ ker f .
Since J is cofinite in A, {x2 + J, x3 + J, · · · } is linearly dependent in A/J. Then
there exists i ≥ 2 and λ2, · · · , λi−1, λi ∈ K such that λi 6= 0 and
λ2x2 + λ3x3 + · · · + λixi ∈ J.
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
11
Now
{λ2x2 + λ3x3 + · · · + λixi, xi+1} = λix1 /∈ ker f,
which contradicts to {J, A} ⊆ ker f .
4. Co-Poisson coalgebra structures on k[x1, x2, · · · , xd]
Suppose g = kx1 ⊕ kx2 ⊕ · · · ⊕ kxd is a d-dimensional Lie algebra. The co-
Poisson Hopf structures on U (g) are in one-to-one correspondence with the Lie
bialgebra structures on g. If g is non-abelian, then there is no nontrivial Poisson
Hopf structure on U (g) (see Proposition 2.5). So, we turn to consider the case when
g is abelian from now on. Then A = U (g) = k[x1, · · · , xd]. We first characterize
co-Poisson coalgebra structures on k[x1, x2, · · · , xd] in this section.
Suppose (C, ∆, ε) is a coalgebra. Let P (C) be the subspace of C consisting of
all primitive elements of C. Assume P (C) = ⊕i∈I kei as a vector space, with (I, <)
being well ordered by using Well Ordering Principle. Let
I = Mi,j∈I,i<j
k(ei ⊗ ej − ej ⊗ ei).
Lemma 4.1. Retain the notations above. Suppose X ∈ C ⊗ C. Then X ∈ I if and
only if
(1 + t2)(X) = 0 and (∆ ⊗ 1)(X) = (1 − t3)(1 ⊗ X).
Proof. "⇒" Trivial.
"⇐" Assume 0 6= X = Pn
{b1, · · · , bn} are both linearly independent. Since (1+t2)(X) = 0, i.e.,Pn
−Pn
i=1 ai ⊗ bi with n minimal. Then {a1, · · · , an} and
i=1 ai⊗bi =
i=1 bi ⊗ ai, then
n
n
(1 ⊗ ai ⊗ bi − bi ⊗ 1 ⊗ ai) =
(1 ⊗ ai + ai ⊗ 1) ⊗ bi.
(1 − t3)(1 ⊗ X) =
Thus ai ∈ P (C) by the independence of the bi's.
Since (∆⊗1)(X) = (1−t3)(1⊗X), thenPn
By using Pn
i=1 ai ⊗ bi = −Pn
P (C) as well. The assertion follows.
i=1 ∆(ai)⊗bi =Pn
i=1 bi ⊗ ai and a similar discussion, we have bi ∈
(cid:3)
i=1(1⊗ai +ai ⊗1)⊗bi.
The following facts are obvious.
Lemma 4.2. Let B be a bialgebra. Then
(1) If X ∈ B ⊗ B is skew-symmetric, then so is X∆(x) for any x ∈ P (B).
(2) For any a ∈ B and X ∈ B ⊗ B,
(∆ ⊗ 1)(X∆(a)) = (∆ ⊗ 1)(X) · ∆(2)(a).
In the following, A = U (g) = k[x1, x2, · · · , xd]. Let H(A) be the set of all monic
monomials of A. For any a ∈ H(A), ∆(a) = P a1 ⊗ a2 is always assumed to be
the expression by the standard k-basis of k[x1, x2, · · · , xd]. For any a ∈ H(A), a
is the degree of a.
First, we establish a reciprocity law for two linear maps from A to A ⊗ A, which
is a key step to characterize the co-Poisson structures.
Proposition 4.3. Let q : A → A ⊗ A and I : A → A ⊗ A be two linear maps.
Then I(a) = (−1)a2q(a1)∆(a2) for all a ∈ A if and only if q(a) = I(a1)∆(a2) for
all a ∈ A.
Proof. First note that in our case (the algebra is generated by primitive elements),
for any 1 6= a ∈ H(A),
(−1)a2∆(a1a2) = 0.
(4.1)
Xi=1
Xi=1
12
QI LOU AND QUANSHUI WU
Since a ⊗ 1 is one of the terms of ∆(a) for any a ∈ H(A), then
(−1)a2+1∆(a1a2) = ∆(a).
(4.2)
Xa26=1
"⇒" Obviously, I(1) = q(1), and so q(a) = I(a1)∆(a2) for a = 1. We prove
q(a) = I(a1)∆(a2) holds for any a ∈ H(A) by induction on the degree of a. Suppose
q(a) = I(a1)∆(a2) holds for all a ∈ H(A) of degree no more than n. To finish the
proof, it suffices to show that q(ax) = I(a1x)∆(a2)+I(a1)∆(a2x) for any x ∈ P (A).
Since, by assumption,
I(ax) = (−1)a2q(a1x)∆(a2) + (−1)a2+1q(a1)∆(a2x)
(−1)a3(I(a1x)∆(a2) + I(a1)∆(a2x))∆(a3)
=q(ax) + Xa36=1
+ (−1)a3+1I(a1)∆(a2)∆(a3x),
then
I(ax) + I(a1)∆(a2x) = q(ax) + Xa36=1
Thus q(ax) = I(a1x)∆(a2) + I(a1)∆(a2x) by (4.2).
(−1)a3I(a1x)∆(a2a3).
"⇐" Obviously, I(a) = (−1)a2q(a1)∆(a2) holds for a = 1 as q(1) = I(1) by
q(a) = I(a1)∆(a2). It suffices to show that if I(a) = (−1)a2q(a1)∆(a2) then for
any x ∈ P (A),
I(ax) = (−1)a2q(a1x)∆(a2) + (−1)a2+1q(a1)∆(a2x).
This is equivalent to that
I(ax) = (−1)a3(I(a1x)∆(a2)+ I(a1)∆(a2x))∆(a3)+ (−1)a3+1I(a1)∆(a2)∆(a3x),
i.e.,
I(ax) = (−1)a3I(a1x)∆(a2a3),
which is always true by (4.1).
(cid:3)
To prove Proposition 4.5, we need the following lemma.
Lemma 4.4. Let A = U (g) = k[x1, · · · , xd]. Then for any linear map q : A →
A ⊗ A and a ∈ H(A),
(−1)a2(1 ⊗ q)∆(a1) · ∆(2)(a2) = (−1)a21 ⊗ q(a1)∆(a2).
Proof. We claim first for any a ∈ H(A),
X(−1)a1+a2a1a3 ⊗ a2 ⊗ a4 =X(−1)a11 ⊗ a1 ⊗ a2.
It is obviously true for a = 1. Now assume (4.3) holds for a. We show (4.3) holds
for ax for any x ∈ P (A).
(4.3)
Since ∆(3)(x) = 1 ⊗ 1 ⊗ 1 ⊗ x + 1 ⊗ 1 ⊗ x ⊗ 1 + 1 ⊗ x ⊗ 1 ⊗ 1 + x ⊗ 1 ⊗ 1 ⊗ 1, then
(−1)(ax)1+(ax)2(ax)1(ax)3 ⊗ (ax)2 ⊗ (ax)4
=(−1)a1+a2a1a3 ⊗ a2 ⊗ a4x + (−1)a1+a2a1a3x ⊗ a2 ⊗ a4
− (−1)a1+a2a1a3 ⊗ a2x ⊗ a4 − (−1)a1+a2a1xa3 ⊗ a2 ⊗ a4
=(−1)a1+a2a1a3 ⊗ a2 ⊗ a4x − (−1)a1+a2a1a3 ⊗ a2x ⊗ a4
=(−1)a11 ⊗ a1 ⊗ a2x − (−1)a11 ⊗ a1x ⊗ a2
=(−1)(ax)11 ⊗ (ax)1 ⊗ (ax)2.
Hence (4.3) holds for all a ∈ H(A). By applying 1 ⊗ q ⊗ ∆ to (4.3), then
(−1)a1+a2a1a3 ⊗ q(a2) ⊗ ∆(a4) = (−1)a11 ⊗ q(a1) ⊗ ∆(a2).
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
13
Thus
(−1)a1(1 ⊗ q)∆(a1) · ∆(2)(a2) = (−1)a11 ⊗ q(a1)∆(a2),
and the proof is finished.
(cid:3)
Proposition 4.5. Let A = U (g) = k[x1, · · · , xd] and I = ⊕i<jk(xi ⊗ xj − xj ⊗ xi).
Then, for any linear map q : A → A ⊗ A, the following are equivalent.
(1) q is skew-symmetric and satisfies the co-Leibniz rule.
(2) I(a) = (−1)a2q(a1)∆(a2) ∈ I for all a ∈ A.
Proof. "(1) ⇒ (2)" By Lemma 4.1, it suffice to show (1+t2)(cid:0)(−1)a2q(a1)∆(a2)(cid:1) =
0 and (∆ ⊗ 1)((−1)a2q(a1)∆(a2)) = (1 − t3)(cid:0)1 ⊗ (−1)a2q(a1)∆(a2)(cid:1) .
Note that A is generated by primitive elements. It follows from Lemma 4.2 that
(1 + t2)((−1)a2q(a1)∆(a2)) = 0 for all a ∈ H(A).
By Lemmas 4.2, 4.4 and the co-Leibniz rule (2.5),
(∆ ⊗ 1)((−1)a2q(a1)∆(a2))
=(−1)a2(∆ ⊗ 1)q(a1) · ∆(2)(a2)
=(1 − t3)(−1)a2(1 ⊗ q)∆(a1) · ∆(2)(a2)
=(1 − t3)(cid:16)(−1)a2(1 ⊗ q)∆(a1) · ∆(2)(a2)(cid:17)
=(1 − t3)(cid:16)1 ⊗ (−1)a2q(a1)∆(a2)(cid:17) .
"(2) ⇒ (1)" Suppose (−1)a2q(a1)∆(a2) ∈ I for all a ∈ H(A). Then q(1) ∈ I.
We check the skew symmetric property of q and the co-Leibniz rule
(∆ ⊗ 1)q(a) = (1 − t3)(1 ⊗ q)∆(a)
by induction on the degree of a. They are true for a = 1 by Lemma 4.1. Assume
they are true for a of degree no more than n. Now for any a with deg a = n + 1.
Since I(a) = (−1)a2q(a1)∆(a2) ∈ I,
q(a) = Xa26=1
(−1)a2+1q(a1)∆(a2) + I(a).
Since q(a1) is skew symmetric for a2 6= 1 by induction hypothesis and I(a) ∈ I,
then q(a) is skew symmetric by Lemma 4.2. By Lemma 4.4,
(∆ ⊗ 1)(q(a))
(∆ ⊗ 1)q(a1) · ∆(2)(a2) + (∆ ⊗ 1)(I(a))
(−1)a2+1q(a1)∆(a2) + I(a)
=(∆ ⊗ 1)
Xa26=1
=(−1)a2+1 Xa26=1
=(1 − t3)
Xa26=1
=(1 − t3)
Xa26=1
+ (1 − t3)(cid:16)X(−1)a2(1 ⊗ q(a1)∆(a2))(cid:17)
(−1)a2+1(1 ⊗ q)∆(a1)
(−1)a2+1(1 ⊗ q)∆(a1) · ∆(2)(a2)
· ∆(2)(a2) + (1 − t3)(1 ⊗ I(a))
=(1 − t3) ((1 ⊗ q)∆(a)) .
The proof is finished.
(cid:3)
14
QI LOU AND QUANSHUI WU
Proposition 4.6. For any a ∈ H(A), let I(a) = P1≤i,j≤d λij
a xi ⊗ xj ∈ I with
(λij
a )d×d ∈ Md(k) skew-symmetric. Then the linear map q : A → A ⊗ A, a 7→
I(a1)∆(a2) defines a co-Poisson structure on the coalgebra A if and only if for all
1 ≤ i < j < k ≤ d and a ∈ A,
d
a1 λij
xsa2 + λsi
a1 λjk
xsa2 + λsj
a1 λki
(4.4)
Xs=1(cid:0)λsk
xsa2(cid:1) = 0.
Proof. By Proposition 4.5, we need only to care for the co-Jacobi identity. For any
a ∈ H(A),
(q ⊗ 1)q(a) = (q ⊗ 1)(I(a1)∆(a2))
λst
a1 (xs ⊗ xt)∆(a2)) =Xs,t
λst
a1 q(xsa2) ⊗ xta3
λst
a1 (I(xsa2)∆(a3) + I(a2)∆(xsa3)) ⊗ xta4
xsa2 (xi ⊗ xj ⊗ xt)(a3 ⊗ a4 ⊗ a5)
a1 λij
λst
a2 (xixs ⊗ xj ⊗ xt + xi ⊗ xjxs ⊗ xt)(a3 ⊗ a4 ⊗ a5)
=(q ⊗ 1)(Xs,t
=Xs,t
= Xs,t,i,j
+ Xs,t,i,j
a1 λij
λst
=U + V,
V = Xs,t,i,j
(cid:9) V = Xs,t,i,j
+ Xs,t,i,j
+ Xs,t,i,j
=0.
where U is the first term, i.e.,
U = Xs,t,i,j
and V is the second term, i.e.,
a1 λij
λst
xsa2(xi ⊗ xj ⊗ xt)(a3 ⊗ a4 ⊗ a5),
a1 λij
λst
a2 (xixs ⊗ xj ⊗ xt + xi ⊗ xj xs ⊗ xt)(a3 ⊗ a4 ⊗ a5).
Then, by the skew-symmetric property of {λij
a }d×d and the cocommutativity of A,
(λst
a1 λij
a2 + λij
a1 λts
a2 )(xixs ⊗ xj ⊗ xt)(a3 ⊗ a4 ⊗ a5)
(λst
a1 λij
a2 + λji
a1 λst
a2 )(xi ⊗ xj xs ⊗ xt)(a3 ⊗ a4 ⊗ a5)
(λst
a1 λij
a2 + λji
a1 λst
a2 )(xt ⊗ xi ⊗ xj xs)(a3 ⊗ a4 ⊗ a5)
(cid:9) U = Xs,t,i,j
=Xi,j,kXs (cid:0)λsk
So, the co-Jacobi identity holds if and only if (cid:9) U = 0. Note that
a1 λij
λst
xsa2(xi ⊗ xj ⊗ xt + xt ⊗ xi ⊗ xj + xj ⊗ xt ⊗ xi)(a3 ⊗ a4 ⊗ a5)
a1 λij
xsa2 + λsi
a1 λjk
xsa2 + λsj
a1 λki
xsa2(cid:1) (xi ⊗ xj ⊗ xk)(a3 ⊗ a4 ⊗ a5).
If (cid:9) U = 0, then, by considering the coefficients of elements in degree 3 in (cid:9) U
(i.e., when a3 ⊗ a4 ⊗ a5 = 1 ⊗ 1 ⊗ 1),
cijk =Xs (cid:0)λsk
a1 λij
xsa2 + λsi
a1 λjk
xsa2 + λsj
a1 λki
xsa2(cid:1) = 0
for all 1 ≤ i, j, k ≤ d and a ∈ H(A).
Conversely, if cijk = 0 for all 1 ≤ i, j, k ≤ d and a ∈ H(A), then (cid:9) U = 0.
(cid:3)
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
15
In summary, we have the following result.
Theorem 4.7. Let A = U (g) = k[x1, · · · , xd]. Then a linear map q : A → A ⊗ A
gives a co-Poisson coalgebra structure on A if and only if there is a linear map
I : A → A ⊗ A such that for all a ∈ H(A),
(1) q(a) = I(a1)∆(a2).
(2) I(a) =P1≤i,j≤d λij
(3) For all 1 ≤ i < j < k ≤ d, Pd
a xi ⊗ xj ∈ I with (λij
a1 λij
s=1(cid:0)λsk
a )d×d ∈ Md(k) skew-symmetric.
xsa2 + λsi
xsa2 + λsj
a1 λjk
a1 λki
xsa2(cid:1) = 0.
The following proposition shows that the co-Jacobi identity holds trivially in two
variables case. So, the co-Poisson coalgebra structures on k[x, y] are given by the
linear maps I : k[x, y] → k(x ⊗ y − y ⊗ x).
Proposition 4.8. Let A = k[x, y]. Then there is an one-to-one correspondence
between the co-Poisson coalgebra structures on A and the linear maps A → I =
k(x ⊗ y − y ⊗ x), given by
(q : A → A ⊗ A) 7→ (I : A → I, a 7→ (−1)a2q(a1)∆(a2),
with the inverse map (I : A → I) 7→ (q : A → A ⊗ A, a 7→ I(a1)∆(a2)).
Proof. We only need to show the co-Jacobi identity always holds in this case. As-
sume I(a) = la(x ⊗ y − y ⊗ x) with la ∈ k. For any a ∈ A,
q(a) = I(a1)∆(a2) = la1 (x ⊗ y − y ⊗ x)(a2 ⊗ a3) = la1(xa2 ⊗ ya3 − ya2 ⊗ xa3).
Then (cid:9) (q ⊗ 1)q(a) =(cid:9) la1 (q(xa2) ⊗ ya3 − q(ya2) ⊗ xa3) . Now
(cid:9) la1 q(xa2) ⊗ ya3
= (cid:9) la1 (I(xa2)∆(a3) + I(a2)∆(xa3)) ⊗ ya4
= (cid:9) la1 lxa2(x ⊗ y − y ⊗ x)(a3 ⊗ a4) ⊗ ya5
+ (cid:9) la1la2 (x ⊗ y − y ⊗ x)(xa3 ⊗ a4 + a3 ⊗ xa4) ⊗ ya5
= (cid:9) la1 lxa2(xa3 ⊗ ya4 ⊗ ya5 − ya3 ⊗ xa4 ⊗ ya5)
+ (cid:9) la1la2 (x2a3 ⊗ ya4 ⊗ ya5 + xa3 ⊗ xya4 ⊗ ya5
− xya3 ⊗ xa4 ⊗ ya5 − ya3 ⊗ x2a4 ⊗ ya5)
= (cid:9) la1 la2(xa3 ⊗ xya4 ⊗ ya5 − xya3 ⊗ xa4 ⊗ ya5).
Similarly,
(cid:9) la1q(ya2) ⊗ xa3 =(cid:9) la1la2(ya3 ⊗ xya4 ⊗ xa5 − xya3 ⊗ ya4 ⊗ xa5).
Thus (cid:9) (q ⊗ 1)q(a) = 0 for all a ∈ A = k[x, y].
(cid:3)
5. (Co-)Poisson Hopf structures on k[x1, x2, · · · , xd]
5.1. Poisson Hopf structures. As proved in [LPV, Proposition 1.8], any Poisson
structure on the polynomial algebra A = k[x1, x2, · · · , xd] is given by {xi, xj} = fij
where {fij}d×d is a skew-symmetric matrix over A such that for all 1 ≤ i < j <
k ≤ d,
d
Xl=1(cid:18)flk
∂fij
∂xl
+ fli
∂fjk
∂xl
+ flj
∂fki
∂xl (cid:19) = 0.
(5.1)
If A = U (g) = k[x1, x2, · · · , xd] is viewed as a Hopf algebra, then the Poisson
algebra structures on A can be described in a form dual to Theorem 4.7. The
following is a reciprocity law for linear maps A ⊗ A → A.
Lemma 5.1. Let p : A ⊗ A → A and J : A ⊗ A → A be two linear maps. Then
p(a⊗ b) = J(a1 ⊗ b1)a2b2 for all a, b ∈ A if and only if J(a⊗ b) = (−1)a2+b2p(a1 ⊗
b1)a2b2 for all a, b ∈ A.
16
QI LOU AND QUANSHUI WU
Proof. Similar to that of Proposition 4.3.
(cid:3)
Proposition 5.2. Let A = U (g) = k[x1, · · · , xd]. Then a linear map p : A⊗A → A
gives a Poisson structure on A if and only if there is a linear map J : A ⊗ A → A
such that for all a, b ∈ A,
(1) p(a ⊗ b) = J(a1 ⊗ b1)a2b2.
(2) J is skew-symmetric, and J(a ⊗ b) = 0 except both a and b are of degree 1.
(3) The Jacobi identity holds for J.
Actually, Proposition 5.2 is exactly [LPV, Proposition 1.8]. If p : A ⊗ A → A is
a Poisson algebra structure on A, and J(a ⊗ b) = (−1)a2+b2p(a1 ⊗ b1)a2b2, then
J satisfies conditions (2) and (3) in Proposition 5.2. In this case, condition (1) is
the same as
{f, g} = X1≤i,j≤d
∂f
∂xi
∂g
∂xj
{xi, xj}.
The Poisson Hopf structures on A are classified in the following proposition.
They are exactly linear Poisson structures on A.
Proposition 5.3. Any Poisson Hopf structure on A = k[x1, x2, · · · , xd] is given
by
d
{xi, xj} =
λij
l xl
(1 ≤ i, j ≤ d),
Xl=1
l , subject to the relations, for all 1 ≤ i < j < k ≤ d and all
where λij
1 ≤ s ≤ d,
l = −λji
d
Xl=1(cid:16)λij
l λlk
s + λjk
l λli
s + λki
l λlj
s(cid:17) = 0.
Proof. Suppose {−, −} is a Poisson Hopf structure on A. Then, by (2.3),
∆ ({xi, xj}) = 1 ⊗ {xi, xj} + {xi, xj} ⊗ 1,
that is, {xi, xj} is a primitive element of A. Hence {xi, xj} = Pd
l xl for
for all l. Then (5.1) is equivalent to, for all
l=1 λij
some λij
l = −λji
l
1 ≤ i < j < k ≤ d and all 1 ≤ s ≤ d,
l ∈ k such that λij
d
Xl=1(cid:16)λij
l λlk
s + λjk
l λli
s + λki
l λlj
s(cid:17) = 0.
Conversely, any such a Poisson algebra structure is in fact a Poisson Hopf struc-
ture on A. We need to check that ∆ ({a, b}) = a1b1 ⊗ {a2, b2} + {a1, b1} ⊗ a2b2 for
all a, b ∈ H(A), which can be done by induction.
(cid:3)
5.2. Co-Poisson Hopf structures. Next we discuss co-Poisson Hopf structures
2 · · · xnd
on A. We fix some notations here. For any a = xn1
d ∈ H(A), we denote
1 xm2
by a! = n1! · · · nd! and a(i) = ni for 1 ≤ i ≤ d. For any b = xm1
d ∈ H(A),
2
if b a, we denote by
· · · xmd
1 xn2
(cid:18)a
b(cid:19) =(cid:18) n1
m1(cid:19)(cid:18) n2
m2(cid:19) · · ·(cid:18) nd
md(cid:19);
if b ∤ a, we set (cid:0)a
b(cid:1) = 0.
(1) q(a) = I(a1)∆(a2).
Proposition 5.4. Let A = U (g) = k[x1, x2, · · · , xd]. Then a linear map q : A →
A ⊗ A gives a co-Poisson Hopf algebra structure on A if and only if there is a linear
map I : A → A ⊗ A such that for all a ∈ H(A),
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
17
s xi ⊗ xj ∈ I
with (λij
s )d×d ∈ Md(k) skew-symmetric.
(2) For all 1 ≤ s ≤ d, I(a) = 0 if a 6= xs and I(xs) =P1≤i,j≤d λij
(3) For all 0 ≤ i < j < k ≤ d, 1 ≤ s ≤ d, Pd
l=1(cid:16)λlk
l + λli
s λjk
s λij
l + λlj
s λki
l (cid:17) = 0.
Proof. "⇒" Let I : A → A ⊗ A be the map I(a) = (−1)a2q(a1)∆(a2) for all a ∈ A.
Then, by Proposition 4.3, q(a) = I(a1)∆(a2), i.e., (1) holds.
Since q(ab) = q(a)∆(b) + ∆(a)q(b) for all a, b ∈ A, I(1) = q(1) = 0. Thus
q(xi) = I(xi) for all 1 ≤ i ≤ d. Then, by induction and (2.6), for any a ∈ H(A),
q(a) =
d
Xl=1(cid:18) a
xl(cid:19)I(xl)∆(cid:18) a
xl(cid:19) .
If a ∈ H(A) with deg a = 2, then I(a) = 0 by q(a) = I(a1)∆(a2). By induction, we
see I(a) = 0 for any a ∈ H(A) of degree ≥ 3. By Proposition 4.5, (2) holds.
l + λlj
In this case, the equation (4.4) is equivalent toPd
l=1(cid:16)λlk
0, i.e., (3) holds.
l (cid:17) =
"⇐" If I(a) = 0 for all xi 6= a ∈ H(A), then q(a) = I(a1)∆(a2) becomes
l + λli
s λjk
s λij
s λki
q(a) =
d
Xl=1(cid:18) a
xl(cid:19) .
xl(cid:19)I(xl)∆(cid:18) a
Now it is easy to check q(ab) = q(a)∆(b) + ∆(a)q(b) for all a, b ∈ H(A).
(cid:3)
Example 5.5. Let A = k[x, y]. Then there is an one-to-one correspondence be-
tween the co-Poisson Hopf structures on A and the set {(I(x), I(y)) I(x), I(y) ∈
k(x ⊗ y − y ⊗ x)}, given by
q 7→ (q(x), q(y)), and
(I(x), I(y)) 7→ q : xnym 7→ nI(x)∆(xn−1ym) + mI(y)∆(xnym−1).
Example 5.6. Let A = k[x1, x2, x3]. Then any Poisson Hopf structure on A is
given by
{x1, x2} = λ1
12x1 + λ2
12x2 + λ3
12x3,
{x2, x3} = λ1
23x1 + λ2
23x2 + λ3
23x3,
{x3, x1} = λ1
31x1 + λ2
31x2 + λ3
31x3,
subject to the relations,
λ1
12λl
31 + λ2
23λl
12 + λ3
31λl
23 = λ2
12λl
23 + λ3
23λl
31 + λ1
31λl
12, for all 1 ≤ l ≤ 3.
Any co-Poisson Hopf structure on A is given by
q(x1) = λ12
1 (x1 ⊗ x2 − x2 ⊗ x1) + λ23
1 (x2 ⊗ x3 − x3 ⊗ x2) + λ31
1 (x3 ⊗ x1 − x1 ⊗ x3),
q(x2) = λ12
2 (x1 ⊗ x2 − x2 ⊗ x1) + λ23
2 (x2 ⊗ x3 − x3 ⊗ x2) + λ31
2 (x3 ⊗ x1 − x1 ⊗ x3),
q(x3) = λ12
3 (x1 ⊗ x2 − x2 ⊗ x1) + λ23
3 (x2 ⊗ x3 − x3 ⊗ x2) + λ31
3 (x3 ⊗ x1 − x1 ⊗ x3),
subject to the relations,
k λ31
λ12
1 + λ23
k λ12
2 + λ31
k λ23
3 = λ31
k λ12
1 + λ12
k λ23
2 + λ23
k λ31
3 , for any 1 ≤ k ≤ 3.
18
QI LOU AND QUANSHUI WU
5.3. Dual (co-)Poisson structures. It is well-known that the dual algebra A∗ of
the coalgebra A = k[x1, x2, · · · , xd] is the algebra of formal divided power series (it is
called Hurwitz series in [Ke, Proposition 2.4] in one variable), which is isomorphic to
the algebra of formal power series k[[x1, x2, · · · , xd]]. By Cartier-Gabriel-Kostant-
Milnor-Moore Theorem ([Sw, Theorem 8.1.5] and [MM, § 6]), the finite dual A◦ of
the polynomial Hopf algebra A = k[x1, x2, · · · , xd] is isomorphic to A ⋉ kn as Hopf
algebras, where kn carries the additive group structure (the group-like elements in
the Hopf dual). Similar to [LPV, Proposition 1.8], the following lemma holds.
Lemma 5.7. Let A = k[[x1, · · · , xd]] be the algebra of formal power series. Any
Poisson structure on A is given by {xi, xj} = fij, where (fij )d×d is a skew-
symmetric matrix over A such that for all 1 ≤ i < j < k ≤ d,
{{xi, xj}, xk} + {{xj, xk}, xi} + {{xk, xi}, xj} = 0.
In this case, for all f, g ∈ A, {f, g} =Pd
i,j=1
∂f
∂xi
∂g
∂xj
fij.
Theorem 5.8. Suppose char k = 0. Let A = k[[x1, x2, · · · , xd]] be the algebra
of formal power series and A = k[x1, x2, · · · , xd]. Then there is an one-to-one
correspondence between the Poisson algebra structures on A and the co-Poisson
coalgebra structures on A.
Proof. Suppose {xi, xj} = fij =Pa∈H(A) λij
structure on A. Let αij
a = a!λij
a and
a a (1 ≤ i, j ≤ d) gives a Poisson algebra
I(a) = X1≤i,j≤d
αij
a xi ⊗ xj ,
q(a) = I(a1)∆(a2) for all a ∈ H(A).
Then q : A → A ⊗ A is a co-Poisson coalgebra structure on A.
1 ≤ i < j < k ≤ d,
In fact, for all
d
{{xi, xj}, xk} = Xa,b∈H(A)
Xl=1
λij
a λlk
b
∂a
∂xl
ab=c
d
b = Xc∈H(A)
Xa,b∈H(A)
Xl=1
(a(l) + 1)λij
axlλlk
b c.
Then (cid:9) {{xi, xj }, xk} = 0 if and only if for all c ∈ H(A) and all 1 ≤ i < j < k ≤ d,
d
(a(l) + 1)(cid:16)λij
Xl=1
axlλlk
b + λjk
axlλli
b + λki
axl λlj
b (cid:17) = 0.
Note that αij
a . It is equivalent to
ab=c
Xa,b∈H(A)
a = a!λij
ab=c
d
Xa,b∈H(A)
Xl=1
1
a!b!(cid:16)αij
axlαlk
b + αjk
axlαli
b + αki
axlαlj
b (cid:17) = 0.
By multiplying it with c!, it is easy to see that it is equivalent to
d
Xl=1(cid:0)αij
c1xlαlk
c2 + αjk
c1xlαli
c2 + αki
c1xlαlj
c2(cid:1) = 0
for all c ∈ H(A) and 1 ≤ i < j < k ≤ d.
On the other hand, suppose q : A → A ⊗ A is a co-Poisson coalgebra structure
on A. Let I(a) = (−1)a2q(a1)∆(a2) := αij
a xi ⊗ xj, and λij
a = 1
a! αij
a . Then
{xi, xj} = fij := Xa∈H(A)
λij
a a.
gives a Poisson algebra structure on A.
(cid:3)
CO-POISSON STRUCTURES ON POLYNOMIAL HOPF ALGEBRAS
19
Remark 5.9. A co-Poisson coalgebra structure q on A given by I : A → A ⊗ A
is called rational if there is an integer n such that I(a) = 0 for all a ∈ H(A) with
deg a ≥ n. Then there is an one-to-one correspondence between Poisson algebra
structures on A and rational co-Poisson coalgebra structures on A.
Combining Propositions 5.3 and 5.4, we have the following.
Theorem 5.10. There is an one-to-one correspondence between Poisson Hopf
structures on A and co-Poisson Hopf structures on A. More precisely, assume
is a Poisson Hopf structure on A. Let
{xi, xj} = λij
1 x1 + · · · + λij
d xd
I(xs) = X1≤i,j≤d
λij
s xi ⊗ xj
for all 1 ≤ s ≤ d and I(a) = 0 for all other a ∈ H(A). Then q(a) = I(a1)∆(a2)
defines a co-Poisson Hopf structure on A.
Acknowledgments
The authors thank Ruipeng Zhu for useful discussions. This research is supported
by NSFC key project 11331006 and NSFC project 11171067.
References
[CP] V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge University Press, Provi-
dence, 1994.
[Dr] V. G. Drinfeld, Quantum groups, Proc. Internat. Congr. Math. (Berkeley, 1986), Amer.
Math. Soc., Providence, RI, 1987, 798 -- 820.
[FL] D. R. Farkas, G. Letzter, Ring theory from symplectic geometry, J. Pure Appl. Algebra
[He]
125 (1998), 155 -- 190.
I. N. Herstein, Rings with Involution, Chicago Lecture in Math., Univ. of Chicago Press,
Chicago, 1976.
[Ke] W. F. Keigher, On the ring of Hurwitz series, Comm. Algebra 25 (1997), 1845 -- 1859.
[KS]
L. I. Korogodski, Y. S. Soibelman, Algebras of Functions on Quantum Groups, Part I,
Mathematical surveys and monographs, V. 56, Amer. Math. Soc., Providence, 1998.
A. Lichnerowicz, Les varieties de Poisson et leurs algebras de Lie associees (French), J.
Differential Geometry 12 (1977), 253 -- 300.
[Li]
[LPV] C. Laurent-Gengoux, A. Pichereau and P. Vanhaecke, Poisson Structures, Grundlehren
der Mathematischen Wissenschaften 347, Springer, Heidelberg, 2013.
[LWW] J. Luo, S.-Q. Wang, Q.-S. Wu, Twisted Poincar´e duality between Poisson homology and
cohomology, J. Algebra 442 (2015), 484 -- 505.
[MM] J. W. Milnor, J. C. Moore, On the structure of Hopf algebras, Ann. of Math., 81 (1965),
[Mo]
211 -- 264.
S. Montgomery, Hopf Algebras and Their Actions on Rings, CBMS Reg. Conf. Ser. Math.
82, Amer. Math. Soc., Providence, RI, 1993.
[MR] J. C. McConnell and J. C . Robson, Noncommutative Noetherian Rings, Wiley, Chichester,
[Oh]
[OP]
1987.
S.-Q. Oh, A Poisson Hopf algebra related to a twisted quantum group, Comm. Algebra 45
(2017), 76 -- 104.
S.-Q. Oh, H.-M. Park, Duality of co-Poisson Hopf algebras, Bull. Korean Math. Soc. 48
(2011), 17 -- 21.
[Sw] M. E. Sweedler, Hopf Algebras, Benjamin, New York, 1969.
[Vo]
T. Voronov, On the Poisson envelope of a Lie algebra, "Noncommutative" moment space,
Funct. Anal. Appl. 29 (1995), 196 -- 199.
[Wei] A. Weinstein, Lecture on Symplectic Manifolds, CBMS Conference series in Math. 29,
1977.
School of Mathematical Sciences, Fudan University, Shanghai 200433, China
E-mail address: [email protected]
School of Mathematical Sciences, Fudan University, Shanghai 200433, China
E-mail address: [email protected]
|
1804.03870 | 1 | 1804 | 2018-04-11T08:38:36 | Leibniz algebras constructed by Witt algebras | [
"math.RA"
] | We describe infinite-dimensional Leibniz algebras whose associated Lie algebra is the Witt algebra and we prove the triviality of low-dimensional Leibniz cohomology groups of the Witt algebra with the coefficients in itself. | math.RA | math | LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
L.M. CAMACHO1, B.A. OMIROV2, T.K. KURBANBAEV3
1 Dpto. Matemática Aplicada I. Universidad de Sevilla. Avda. Reina Mercedes, 41012 Sevilla,
Spain, [email protected]
8
1
0
2
2 National University of Uzbekistan, 100174, Tashkent, Uzbekistan, [email protected]
3 Institute of Mathematics of Uzbek Academy of Sciences, 100125, Tashkent, Uzbekistan,
[email protected]
r
p
A
1
1
]
.
A
R
h
t
a
m
[
1
v
0
7
8
3
0
.
4
0
8
1
:
v
i
X
r
a
Abstract. We describe infinite-dimensional Leibniz algebras whose associated Lie algebra is
the Witt algebra and we prove the triviality of low-dimensional Leibniz cohomology groups of
the Witt algebra with the coefficients in itself.
1. Introduction
Mathematical structures are important mostly in mathematics and its applications. Leibniz
algebras were introduced and developed by J-L. Loday [20]. They are a generalization of Lie
algebras, removing the restriction that the product is anti-commutative or that the square of an
element is zero. Many results of Lie algebra theory are extended to the case of Leibniz algebras,
while "pure" Leibniz results (which are true for non Lie Leibiz algebras) are obtained as well
[1–11, 23, 24].
For Lie algebras it is known that an arbitrary finite-dimensional Lie algebra over a field of char-
acteristic zero decomposes into the semidirect sum of the maximal solvable ideal and its semisimple
subalgebra (Levi's Theorem, [17]). Similar result is also true for Leibiz algebras, namely, a finite-
dimensional Leibniz algebra decomposes into the semidirect sum of a maximal solvable ideal and
a semisimple Lie subalgebra (Levi's Theorem, [8]). Complete description of semisimple finite-
dimensional Lie algebras over a field of characteristic zero is known [16, 17]. Therefore, the study
of finite-dimensional Leibniz algebras is reduced to the study of solvable ones.
Infinite-dimensional case is more complicated even in Lie algebra structures. The most simple
infinite-dimensional structure are Witt and Virasoro algebras. Complex Witt algebra was first
considered by E. Cartan [12] in 1909. This algebra is an example of an infinite-dimensional simple
Lie algebra. Recall that Virasoro algebra was introduced due to Witt algebra. In fact, the Vi-
rasoro algebra is the one-dimensional central extension of the Witt algebra. In the work [18] the
infinite-dimensional Lie algebras and their representations are studied. We also mention that the
investigations of representations of Witt algebra are found in [14], [22], [25].
A non-Lie Leibniz algebra L contains the non-trivial ideal, generated by the squares of elements
of the algebra L (denoted by I and usually called the Leibniz kernel), i.e., I = h[x, x] x ∈ Li. The
ideal I plays an important role in the theory of Leibniz algebras since it determines the (possible)
non-Lie property of a Leibniz algebra. Moreover, this ideal belongs to the right annihilator of L
and it is the minimal ideal with the property that the quotient algebra L/I is a Lie algebra.
The usual notion of simplicity for non Lie Leibniz algebras has no sense (because of non-triviality
of the ideal I). Therefore, it is proposed to use the adapted version of simplicity for Leibniz
algebras. Namely, Leibniz algebra L is called simple, if [L, L] 6= I and its only ideals are {0},
2010 Mathematics Subject Classification. 17A32, 17B30, 17B10.
Key words and phrases. Leibniz algebra, Witt Lie algebra, Leibniz representation, right Lie module, classification.
The work was partially supported was supported by Agencia Estatal de Investigación (Spain), grant MTM2016-
79661-P (European FEDER support included, UE). The third named author was supported by VI PPIT-US and
by Instituto de Matemáticas de la Universidad de Sevilla.
1
2
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
I and L. Clearly, if a Leibniz algebra is simple, then its corresponding Lie algebra is simple as
well. However, the converse is not true, in general. The analogue of Levi's theorem for Leibniz
algebras imply that any finite-dimensional simple Leibniz algebra decomposes into the semidirect
sum of simple Lie algebra and the ideal I, where I can be considered as an irreducible right
module over simple Lie algebra. The approach in the investigation of Leibniz algebras with a given
corresponding solvable Lie algebra was applied in [10, 11, 24].
In finite-dimensional case we know that for a given simple Lie algebra and its irreducible right
module one can construct a unique simple Leibniz algebra such that its corresponding Lie algebra
is the simple Lie algebra and the ideal I is the given right module. In this paper we describe the
Leibniz algebras whose corresponding Lie algebra is the Witt algebra and the ideal I is its right
module (see Theorems 2 and 3). Moreover, we prove the triviality of low-dimensional cohomology
groups of these Leibniz algebras (see Theorem 4).
Throughout the paper algebras are considered to be over the field of the complex numbers.
Moreover, in the table of multiplications of algebras the omitted products are assumed to be zero.
2. Prelimiaries
In this section we give preliminary definitions and results on Leibniz algebras and modules over
the Witt algebra.
An algebra L with multiplication [·, ·] over a field F is called Leibniz algebra if for any x, y, z ∈ L
the so-called Leibniz identity
[x, [y, z]] = [[x, y], z] − [[x, z], y]
holds.
Further we will use the notation
L(x, y, z) = [x, [y, z]] − [[x, y], z] + [[x, z], y].
It is obvious that Leibniz algebras are determined by the identity L(x, y, z) = 0.
For a Leibniz algebra L we consider the natural homomorphism onto the quotient Lie algebra
L/I, which is called the corresponding Lie algebra to the Leibniz algebra L (in some references
this algebra is called the liezation of L).
Now we present a construction of Leibniz algebras by a given Lie algebra and its right module.
Let (G, [−, −]) be a Lie algebra and let V be a right G-module. We equip the vector space
Q(G, V ) = G ⊕ V with the multiplication (−, −) in the following way:
(x + v, y + w) := [x, y] + v ⋆ y,
x, y ∈ G, v, w ∈ V.
(1)
Then Q(G, V ) is a Leibniz algebra and from (1) we get (G, V ) = (V, V ) = 0.
Note that if V ⋆ G = V , then V is nothing else but the ideal I of the Leibniz algebra Q(G, V ).
The map I × L/I −→ I defined as (v, x) −→ [v, x], v ∈ I, x ∈ L endows I with a structure of
a right L/I-module (it is well-defined due to I being in the right annihilator). Thus, for a given
Leibniz algebra L we have a Lie algebra L/I and its right module I.
The main goal of this paper is to describe Leibniz algebras such that its corresponding Lie
algebra is a given Lie algebra G and the ideal I is a given right G-module V .
Let A = C[z, z−1] be the algebra of Laurent polynomials in one variable. The Lie algebra of
derivations
Der(A) = span{f (z)
d
dz
: f ∈ C[z, z−1]}
is called Witt algebra and denoted by W. Obviously, the basis of W can be chosen as {dj j ∈ Z},
where dj = −zj+1 d
dz .
Then the table of multiplications of W in this basis have the following form:
On a vector space V (α, β) = {v(n) n ∈ Z} we define W-module structure [14] as follows:
[dm, dn] = (m − n)dm+n, m, n ∈ Z.
(2)
v(n) ⋆ dm = (α + n + βm)v(n + m), n ∈ Z, α, β ∈ C, α 6= 0.
(3)
dz on the elements v(n) = zn+α(dz)β.
This bracket would be discovered by the action dm = −zm+1 d
Below we present a result of [18].
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
3
Theorem 1. The representation V (α, β) is reducible if (i) α ∈ Z and β = 0, or (ii) α ∈ Z and
β = 1; otherwise it is irreducible.
Let L be Leibniz algebra such that its corresponding Lie algebra is W and the right W-module
V (α, β) is the ideal I. We equip the direct sum of the vector spaces L = W ⊕ I with a product
[−, −] satisfying the conditions:
[V (α, β), W] = V (α, β) ⋆ W,
[W, V (α, β)] = [V (α, β), V (α, β)] = 0,
(4)
where the product V (α, β) ⋆ W follows from (3).
Thus, in order to completely describe the Leibniz algebra L we have to clarify the product
[W, W].
For getting acquainted with cohomology of Leibniz algebras we refer the reader to works [20], [21]
and references therein. Here we just give the definition of the second group of cohomology for
Leibniz algebras with coefficient in itself. In fact, the second cohomology group of a Leibniz algebra
L with coefficient itself is the quotient space
HL2(L, L) := ZL2(L, L)/BL2(L, L),
where elements f ∈ BL2(L, L) and ϕ ∈ ZL2(L, L) are defined by conditions:
f (x, y) = [d(x), y] + [x, d(y)] − d([x, y]), for some linear map d ∈ Hom(L, L),
[x, ϕ(y, z)] − [ϕ(x, y), z] + [ϕ(x, z), y] + ϕ(x, [y, z]) − ϕ([x, y], z) + ϕ([x, z], y) = 0,
(5)
respectively.
It is obvious that a Leibniz 2-cocycle ϕ of a Leibniz algebra L is determined by the identity
Φ(ϕ)(x, y, z) = 0, x, y, z ∈ L where
Φ(ϕ)(x, y, z) = [x, ϕ(y, z)] − [ϕ(x, y), z] + [ϕ(x, z), y] + ϕ(x, [y, z]) − ϕ([x, y], z) + ϕ([x, z], y).
3. Some Leibniz algebras with corresponding Lie algebra W .
In this section we describe Leibniz algebras whose corresponding Lie algebra is the Witt algebra
and the ideal I is V (α, β) which satisfy (4). First, let us consider the case α /∈ Z.
Theorem 2. Let L be an arbitrary Leibniz algebra whose corresponding Lie algebra is the Witt
algebra W and the ideal I of L is considered as V (α, β) which satisfy the condition (4). If α /∈ Z,
then there exists a basis {di, v(i) i ∈ Z} of L such that its table of multiplications have the form:
( [v(k), di] = (k + α + βi)v(i + k),
[di, dj] = (i − j)di+j .
Proof. Let {di i ∈ Z} be the basis of W and let {v(i) i ∈ Z} be the basis of the space V (α, β).
Let us introduce notations
[v(k), di] = (k + α + βi)v(i + k),
with α + j + βj /∈ Z, j ∈ Z and α, β ∈ C.
Taking into account α /∈ Z by taking the change of basis elements di in the following way:
d′
0 = d0 −Xk
σ0,k
α + k
v(k),
d′
s = ds +Xk
γs,0,k
s − α − k
v(k), s ∈ Z∗,
one can assume that
[d0, d0] = 0, [dj, d0] = jdj,
j ∈ Z.
γijkv(k),
δjkv(k),
[di, dj] = (i − j)di+j +Pk
[dj, d−j] = 2jd0 +Pk
[dj, dj] =Pk
σjkv(k),
k + α + βi /∈ Z
i /∈ {j, −j},
j 6= 0,
4
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
By considering Leibniz identity for triples of elements given below we have the following con-
straints:
Leibniz identity
Constraints
L(d0, dj, d0) = 0
L(dj, dj, d0) = 0
L(dj, d0, d−j) = 0 =⇒ δj,k = 0, k 6= −α
L(dj, d0, d−j) = 0 =⇒ γijk = 0, k 6= i + j − α =⇒ [di, dj] = (i − j)di+j .
=⇒ [d0, dj] = −jdj,
=⇒ [dj, dj] = 0,
=⇒ [dj, d−j] = 2jd0,
=⇒ γ0,j,k = 0, α + k 6= j
=⇒ σj,k = 0, k 6= 2j − α
j ∈ Z,
Thus, we have proved [di, dj] = (i−j)di+j, that is, [W, W] = W, which completes the description
(cid:3)
of the structure of L.
Now we consider the case α ∈ Z.
Theorem 3. Let L be an arbitrary Leibniz algebra with corresponding Lie algebra W and the ideal
I of L is considered as V (α, β) which satisfies the conditions (4). If α ∈ Z, then there exists a
basis {di, v(i) i ∈ Z} of L such that the table of multiplications of L have one of the following
form:
(I) :( [v(k), di] = (k + α + βi)v(i + k), β /∈ {−1, 0, 1, 2, 3},
[di, dj] = (i − j)di+j ,
[v(k), di] = (k + α + 3i)v(i + k),
[dj , dj] = bj,jv(2j − α),
[dj , d−j] = 2jd0 − 1
[di, dj] = (i − j)di+j + bi,jv(i + j − 1),
3 bj,jv(−α),
j /∈ {−1, 0, 1},
j /∈ {−1, 0, 1},
i /∈ {j, −j},
1
(2i+1)(i−2) ((i + 1)(2i − 3)bi−1,i−1 − (2i − 1)),
b2,2 = 2, b−2,−2 = −2,
bi,i =
bi,i+1 =
bi,j = j
i+1
(2i+1)(i−1) ((2i − 1)bi,i − 1),
i(j−2) ((i − 1)bi+1,j−1 + (j − i − 1)),
i /∈ {−2, −1, 0, 1, 2},
i /∈ {j, −j, }, i, j /∈ {−1, 0, 1}, j 6= i + 1.
b2,2 = 9, b−2,−2 = −9,
bi,i =
(i + 1)(2i + 1)bi−1,i−1 − (i + 1)(4i − 3)ai−1 − (4i − 1)(i − 2)a−i
(i − 2)(2i − 3)
bi,i+1 =
(i + 1)((2i + 3)bi,i + (4i + 1)ai)
(i − 1)(2i + 1)
,
, i /∈ {−2, −1, 0, 1, 2},
i /∈ {−1, 0, 1},
bi,j =
j((i − 1)bi+1,j−1 + (3i + j)(j − 1)aj−1 − (j − i − 1)(j + i − 1)aj+i−1 − i(i + 3j − 2)ai)
i(j − 2)
j 6= i + 1, i /∈ {j, −j}, i, j /∈ {−1, 0, 1},
,
(i − 1)(i + 2)(i + 3)
(i + 1)(i − 2)(i − 3)
, a−i = −
,
i /∈ {−1, 0, 1},
ai =
20
20
[v(k), di] = (k + α + i)v(i + k),
[d0, di] = −idi,
[di, d0] = idi,
[di, di] = (i3 − i)v(2i − α),
[di, d−i] = 2id0 + (i3 − i)v(−α),
[di, dj] = (i − j)di+j + j(ij − 1)v(i + j − α),
i 6= 0,
i 6= 0,
i 6= 0,
i 6= 0,
i /∈ {j, −j, 0},
j 6= 0.
[v(k), di] = (α + k − i)v(k + i),
[dj, dj] = bj,jv(2j − α),
[dj, d−j] = 2jd0 − 1
[di, dj] = (i − j)di+j + bi,jv(i + j − α),
3 bj,jv(−α),
j 6= 0,
j 6= 0,
i /∈ {j, −j},
(II) :
where
(III) :
(IV ) :
where
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
5
Proof. Let us chose a basis {di, v(i) i ∈ Z} such that {di, i ∈ Z} and {v(n) n ∈ Z} are bases
of the spaces W and V (α, β), respectively. From the assumption of theorem we have the products
[v(n), dm] = (α + n + βm)v(n + m), n ∈ Z, α, β ∈ C, α 6= 0,
[W, V (α, β)] = [V (α, β), V (α, β)] = 0.
In order to complete the proof of theorem, that is, to clarify the structure of L we have to
describe the products [W, W].
We introduce notations:
γi,j,kv(k),
i 6= −j,
i 6= j,
[di, dj] = (i − j)di+j +Pk
[dj , d−j] = 2jd0 +Pk
[dj , dj] =Pk
γj,j,kv(k).
γj,−j,kv(k),
j 6= 0,
Taking the change of basis elements di as follows:
d′
0 = d0 − Xk,k6=−α
γ0,0,k
α + k
v(k), d′
s = ds + Xk,k6=s−α
γs,0,k
s − α − k
v(k)
s 6= 0.
one can assume that
[d0, d0] = γ0,0,−αv(−α),
[dj , d0] = jdj + γj,0,j−αv(j − α),
j 6= 0.
Considering the Leibniz identity for the triples of elements given below we have the following
constraints:
Leibniz identity
Constraints
L(dj, dj, d0) = 0, j 6= 0
L(d0, d0, dj) = 0, j 6= 0
=⇒
=⇒
L(dj, d0, d−j) = 0, j 6= 0 =⇒
For the sake of convenience we denote
γj,0,j−α = −βγ0,0,−α,
γ0,j,k = 0,
[d0, dj] = −jdj + γ0,j,j−αv(j − α),
j 6= 0,
k 6= j − α,
j 6= 0.
(1 + β)γj,0,j−α = 0,
γj,j,k = 0,
[dj, dj] = γj,j,2j−αv(2j − α),
j 6= 0,
k 6= 2j − α,
j ∈ Z,
γj,−j,k = 0,
k 6= −α,
(1 − β)γj,0,j−α − 2γ0,0,−α = 0,
[dj, d−j] = 2jd0 + γj,−j,−αv(−α).
γ0,0 = γ0,0,−α,
γ0,j = γ0,j,j−α,
γj,j = γj,j,2j−α,
γj,−j = γj,−j,−α.
Making the change
we derive γ0,j = 0.
d′
j = dj −
γ0,j
j
v(j − α),
j 6= 0,
The equalities L(di, d0, dj) = 0 with i 6= {−j, 0, j} imply γi,j,k = 0, k 6= i + j − α, i /∈ {−j, 0, j}.
Therefore, we apply new notations γi,j = γi,j,i+j−α.
6
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
Considering the Leibniz identity for the triples elements given below we obtain the constraints:
Identity
Constraints
L(d0, dj, d−j) = 0 =⇒ γj,−j + γ−j,j = −2γ0,0, j 6= 0
L(d0, di, dj) = 0
=⇒ iγi,j = jγj,i,
i 6= {0, j, −j}, j 6= 0
L(dj , d−j, dj) = 0 =⇒ βγj,−j + (β − 2)γj,j = 2βγ0,0, j 6= 0,
L(di, dj, dk) = 0
=⇒ (j − k)γi,j+k = (i − j)γi+j,k + (i + j + βk)γi,j − (i − k)γi+k,j −
In new notations the table of multiplication of L has the form
(i + k + βj)γi,k.
[v(k), di] = (k + α + βi)v(i + k),
[d0, d0] = γ0,0v(−α),
[d0, dj] = −jdj,
[dj, d0] = jdj + γj,0v(j − α),
[dj, dj ] = γj,jv(2j − α),
[dj, d−j] = 2jd0 + γj,−jv(−α),
[di, dj] = (i − j)di+j + γi,jv(i + j − α),
j 6= 0,
j 6= 0,
j 6= 0,
j 6= 0,
i /∈ {−j, 0, j}.
with the following restrictions:
γj,0 + βγ0,0 = (1 + β)γj,0 = (1 − β)γj,0 − 2γ0,0 = 0, j 6= 0
γj,−j + γ−j,j = −2γ0,0,
j 6= 0,
iγi,j = jγj,i,
i 6= {0, j, −j}, j 6= 0,
βγj,−j + (β − 2)γj,j = 2βγ0,0,
j 6= 0,
(j − k)γi,j+k = (i − j)γi+j,k + (i + j + βk)γi,j − (i − k)γi+k,j − (i + k + βj)γi,k.
Let us consider possible cases.
• Case 1. Let β 6= −1. Then γ0,0 = γj,0 = 0.
Case 1.1. Let β 6= 0. Taking the following basis transformation
d′
0 = d0 +
γ1,1
β(β + 1)
v(−α), d′
i = di −
γ1,1
1 + β
v(i − α),
i 6= 0, v′(i) = v(i),
we may assume γ1,1 = 0. From the above restrictions we derive γ1,−1 = γ−1,1 = 0.
Equality (9) with the values i = k = 1 imply
From this equality and (7) we obtain
(j − 1)(γ1,j+1 + γj+1,1) = (j + 1 + β)γ1,j.
(6)
(7)
(8)
(9)
(i − 1)(i + 2)γi+1,1 = (i + 1 + β)iγi,1, i /∈ {−2, −1, 0, 1}.
(10)
Using the induction it is easy to proof the equality
γi,1 =
(i − 1)(3 + β)(4 + β) · · · (i + β)
4 · 5 · · · (i + 1)
γ2,1,
i ∈ Z+ \ {1, 2}
Equality (9) with the values i = k = −1 imply
(j + 1)(γ−1,j−1 + γj−1,−1) = (j − 1 − β)γ−1,j − (βj − 2)γ−1,−1.
From this equality together with (7) we get
(i + 1)(i − 2)γi−1,−1 = (i − 1 − β)iγi,−1 − (βi − 2)γ−1,−1, i /∈ {−1, 0, 1, 2}.
(11)
Case 1.1.1. Let β 6= 1.
Case 1.1.1.1. Let β 6= 2. Then equality (8) with j = −1 imply γ−1,−1 = 0.
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
7
Substituting instead of parameters {i, j, k} the following values
(1, −2, 1), (1, −2, −1), (−1, 2, −1), (−1, 2, 1), (1, 2, −1), (−1, −2, 1)
in equality (9) we obtain
γ1,−2 = γ1,−3 = γ−1,2 = γ−1,3 = (3 − β)γ1,2 = (β − 3)γ−1,−2 = 0.
From these equalities and (7) we get γ−2,1 = γ−3,1 = γ2,−1 = γ3,−1 = 0.
We distinguish the possible subcases.
(a) Let β 6= 3. Then γ1,2 = γ−1,−2 = γ−2,−1 = γ2,1 = 0. From the equalities (10), (11) and (7)
we derive
Equality (9) with i = ±1, j := ±i, k = ∓1 together with (7) imply
γi,1 = γ1,i = γ−i,−1 = γ−1,−i = 0, i ∈ Z+ \ {1, 2}.
γi,−1 = γ−1,i = γ−i,1 = γ1,−i = 0,
i ∈ Z+ \ {1, 2}.
Let us write the combining of the above restriction
γ1,i = γi,1 = γ−1,i = γi,−1 = 0,
i ∈ Z∗.
(12)
Substituting in (9) i = 1, j := i, k := j we get
(1 − i)γ1+i,j = (1 − j)γ1+j,i,
i, j ∈ Z \ {−1, 1}.
In this equality putting j = −2 we obtain (1 − i)γ1+i,−2 = 3γ−1,i = 0.
Therefore, γi,−2 = γ−2,i = 0 for i ∈ Z \ {−2, 0, 2}.
Similarly, for j ≤ −3 we deduce γi,j = γj,i = 0 with i ∈ Z∗. So, we have γi,j = γj,i = 0 with
i ∈ Z∗ and j ∈ Z−.
From equality (9) we have
(13)
If i + (1 − β)j 6= 0, then from (13) with i = 2, j ≥ 3 we obtain γ2,j = γj,2 = 0, j ∈ Z+ \ {1, 2}
(i + (1 − β)j)γi,j = (i + j)γi−j,j, i 6= j, i, j ∈ Z+.
and repeating this consideration for i ≥ 3, we deduce
γi,j = γj,i = 0, i /∈ {j, −j}, i, j ∈ Z∗.
If i + (1 − β)j = 0, then equality (13) implies
γi,j = γj,i = 0, i /∈ {j, −j}, i, j ∈ Z∗.
Taking into account the above equation and equality (9) with j := i, k := j and i /∈ {−j} we
get γi,i = 0. Finally, from (8) we have γi,−i = 0, i ∈ Z∗ and the first algebra of the list of theorem
is obtained.
(b) Let β = 3. Then we have
γ1,−2 = γ−2,1 = γ−1,2 = γ2,−1 = γ1,−3 = γ−3,1 = γ−1,3 = γ3,−1 = γ1,1 = γ1,−1 = γ−1,1 = 0.
Note that from (8) we get γj,j = −3γj,−j.
Substituting β = 3 in equalities (10) and (11) we derive
(i − 1)(i + 2)γi+1,1 = (i + 4)iγi,1,
i ∈ Z \ {−1, 0, 1, 2},
(1 + i)(i − 2)γi−1,−1 = (i − 4)iγi,−1,
i ∈ Z \ {−1, 0, 1, 2}.
Applying induction in equality (14) one can prove
20
Similarly, using (15) it is easy to prove that
(i − 1)(i + 2)(i + 3)
γi,1 =
γ2,1,
i ≥ 3.
γi,−1 = −
(i + 1)(i − 2)(i − 3)
20
γ−2,−1,
i ≤ −3.
(14)
(15)
(16)
(17)
Substituting instead of parameters {i, j, k} the following values (1, i, −1), (−1, −i, 1) in equality
(9) we obtain
(1 − i)γi+1,−1 = (1 + i)γ1,i−1 − (i − 2)γ1,i, i ∈ Z \ {1, 2},
(i − 1)γ−i−1,1 = −(1 + i)γ−1,−i+1 + (i − 2)γ−1,−i, i ∈ Z \ {1, 2}.
8
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
Now substitute (16) in the first of the above equations to deduce
γi,−1 =
(i + 1)(i − 2)(i − 3)
20
γ2,1,
i ∈ Z+ \ {1, 2}.
Similarly, applying (17) in the second of the above equalities we get
γ−i,1 =
(i + 1)(i − 2)(i − 3)
20
γ−2,−1,
i ∈ Z+ \ {1, 2}.
Taking into account that γ2,−2 = −γ−2,2 in the equalities γ2,2 = −3γ2,−2
γ−2,−2 = −3γ−2,2
we obtain γ2,2 + γ−2,−2 = 0. Considering equality (9) with values (i, j, k) equal to (2, 2 − 1) and
(−2, −2, 1) lead to γ2,2 = 9γ2,1, γ−2,−2 = 9γ−2,−1.
Consequently, γ2,1 = −γ−2,−1 and
where
γi,1 = aiγ2,1,
γi,−1 = −a−iγ2,1,
i ∈ Z \ {−1, 0, 1},
ai =
(i − 1)(i + 2)(i + 3)
20
, a−i = −
(i + 1)(i − 2)(i − 3)
20
.
If in equality (9) we take (i, j, k) = (i, i, ∓1), then we obtain
(i + 1)(γi,i−1 + γi−1,i) = (2i − 3)γi,i − (4i − 1)γi,−1, i /∈ {−1, 0, 1},
(i − 1)(γi,i+1 + γi+1,i) = (2i + 3)γi,i − (4i + 1)γi,1, i /∈ {−1, 0, 1}
(18)
(19)
Putting i = 2 and i = 3 in (19) and (18), respectively, we derive
γ2,3 + γ3,2 = 54γ2,1,
4(γ2,3 + γ3,2) = 3γ3,3 =⇒ γ3,3 = 72γ2,1.
For i /∈ {−2, −1, 0, 1, 2} we set γi,i = bi,iγ2,1. Clearly, b2,2 = 9 and b3,3 = 72.
If we replace the parameter i in (19) to i − 1 and multiply both sides of equality (18) by i − 2,
then we derive
(cid:26) (i − 2)(γi−1,i + γi,i−1) = (2i + 1)γi−1,i−1 − (4i − 3)γi−1,1,
(i + 1)(i − 2)(γi−1,i + γi,i−1) = (i − 2)(2i − 3)γi,i − (i − 2)(4i − 1)γi,−1.
From which we have
(i + 1)((2i + 1)γi−1,i−1 − (4i − 3)γi−1,1) = (i − 2)(2i − 3)γi,i − (i − 2)(4i − 1)γi,−1.
Taking into account that γi−1,i−1 = bi−1,i−1γ2,1, γi−1,1 = ai−1γ2,1, γi,−1 = −a−iγ2,1, we get
γi,i =
1
(i − 2)(2i − 3)
((i + 1)(2i + 1)bi−1,i−1 − (i + 1)(4i − 3)ai−1 − (4i − 1)(i − 2)a−i)γ2,1.
Thus, we obtain the following recursive relations:
b2,2 = 9, b−2,−2 = −9,
bi,i =
(i + 1)(2i + 1)bi−1,i−1 − (i + 1)(4i − 3)ai−1 − (4i − 1)(i − 2)a−i
(i − 2)(2i − 3)
,
i 6= {−2, −1, 0, 1, 2}.
Since iγi,j = jγj,i, in order to clarify the parameters γi,j it is enough to discover γi,i+j. First,
we find parameters γi,i+1. The analysis of (7) and (19) lead to
γi,i+1 = bi,i+1γ2,1,
i 6= {−1, 0, 1},
where bi,i+1 =
i + 1
(i − 1)(2i + 1)
((2i + 3)bi,i − (4i + 1)ai).
Considering (9) for i = 1, j := i and k := i + 1 with i 6= {−1, 0, 1} we have
γi,i+2 = bi,i+2γ2,1,
i 6= {−1, 0, 1},
where bi,i+2 =
i + 2
i2 ((i − 1)bi+1,i+1 + 2(2i + 1)(i + 1)ai+1 − (2i + 1)a2i+1 − i(4i + 4)ai).
Applying similar arguments we obtain
where
γi,j = bi,jγ2,1,
bi,j =
j
i(j − 2)
((i − 1)bi+1,j−1 + (3i + j)(j − 1)aj−1−
− (j − i − 1)(j + i − 1)aj+i−1 − i(i + 3j − 2)ai).
Finally, from (8) we get γi,−i = − 1
3 bi,iγ2,1,
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
9
If γ2,1 = 0, then we have the algebra (I) with β = 3.
If γ2,1 6= 0, using the scale of basis we may assume γ2,1 = 1. Thus, the second algebra of the
list of theorem is obtained.
Case 1.1.1.2. Let β = 2. Then, due to (8) we get γ−j,j = γj,−j = 0. Note that γ1,1 = γ−1,1 =
γ1,−1 = 0. Consequently, the table of multiplications of the algebra has the following form:
[v(k), di] = (k + α + βi)v(i + k),
[d0, dj] = −jdj,
[dj , d0] = jdj,
[dj , dj] = γj,jv(2j − α),
[dj , d−j] = 2jd0,
[di, dj] = (i − j)di+j + γi,jv(i + j − α),
j 6= 0,
j 6= 0,
j /∈ {0, 1},
j 6= 0,
i 6= j,
i 6= −j.
Substituting instead of parameters {i, j, k} the following values
(i, i, −2i), (1, 2, 1), (1, −2, 1), (1, −2, −1), (−1, 2, 1), (1, 2, −1)
in equality (9) we obtain restriction on structure constants of the algebra
2γi,i = −γi,−2i,
3γ−1,−1 = −γ1,−3,
γ−1,3 = 3γ−1,2,
γ3,−1 = γ1,2.
i 6= 0,
γ1,3 = −γ3,1 + 5γ1,2,
γ1,−2 = 0,
(20)
Taking into account (20) and equation (9) with j := i, k := j we get
−(i + j)γi,−2i = (i − j)(γi+j,i + γi,i+j ) + (j + 3i)γi,j.
This equality with i = 1 has the form
−(1 + j)γ1,−2 = (1 − j)(γ1+j,1 + γ1,1+j) + (j + 3)γ1,j.
Due to (7) the above equality can be written as follows
(1 − j)(2 + j)γ1+j,1 + j(j + 3)γj,1 = −(1 + j)γ1,−2.
(21)
Equality (21) imply
Considering (9) with (i, j, k) = (−1, 3, −1) and (7) we conclude
4γ3,1 = 5γ1,2, γ−2,1 = 0, γ−1,3 = −3γ1,2
γ−1,1 = −γ−1,2.
Thanks to (20) we have γ−1,−1 = γ−1,2 = 0.
From equality (9) with j := i, k := j and i /∈ {−j} we obtain
(i − j)(γi,i+j + γi+j,i) = 2(i + j)γi,i − (3i + j)γi,j.
Setting i = −1 in the above equality and using (7), we obtain
(j + 1)(j − 2)γj−1,−1 = (j − 3)γ−1,j,
from which we derive γi,−1 = 0 with i ∈ Z+ \ {1, 2}.
Equality (21) for j ≤ −4 deduces γi,1 = γ1,i = 0, i ∈ Z−.
Substituting in equality (9) the values of parameters (i, j, k) as follows:
(1, i, −1), (−1, i, 1), (1, i, j),
we derive relations:
γ1,i = (1 + i)γ1,i−1, i ∈ Z+ \ {1, 2},
(i + 1)γ−1,i = (i − 1)γ−1,i+1, i ∈ Z− \ {−1},
(1 − i)γ1+i,j = (1 − j)γ1+j,i, i ∈ Z− \ {−1, 1}.
(22)
(23)
(24)
Taking i ≥ 3 in (22), i ≤ −2 in (23) and j = −2 in (24), we obtain
γ1,i = γi,1 = 0, i ∈ Z,
γ−1,i = γi,−1 = 0, i ∈ Z,
γi,−2 = γ−2,i = 0, i ∈ Z \ {0, 2}.
Applying similar arguments for j = −3, −4, . . . we get
γi,j = γj,i = 0, i 6= j, i ∈ Z∗, j ∈ Z−.
Thanks to the first equality of (20) we have γi,i = 0, i ∈ Z.
10
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
From equality (9) with k := −j and i 6= j we derive
(i − j)γi,j = (i + j)γi−j,j,
i 6= j,
i, j ∈ Z+.
Considering in this equality various values of parameters i and j we deduce
γi,j = γj,i = 0,
i 6= j, −j,
i, j ∈ Z∗.
Thus, we obtain the algebra (I) with β = 2.
Case 1.1.1. Let β = 1. Then from (6) and (8) we have γ1,1 = γ−1,1 = γ1,−1 = γ−1,−1 = 0 and
the following family of algebras:
[v(k), di] = (k + α + i)v(i + k),
[d0, dj] = −jdj,
[dj , d0] = jdj,
[dj , dj] = γj,jv(2j − α),
[dj , d−j] = 2jd0 + γj,jv(−α),
[di, dj] = (i − j)di+j + γi,jv(i + j − α),
j 6= 0,
j 6= 0,
j 6= 0,
j 6= 0,
i 6= j,
i 6= −j.
Substituting in equality (9) the values of parameters (i, j, k) as follows:
(−1, 2, 1), (1, −2, −1), (1, 2, −1), (−1, −2, 1), (2, 2, −1),
(−2, −2, 1), (1, 2, −2), (2, −2, 1), (−1, −2, 2), (2, −2, −1),
we get the following relations between structure constants:
γ−3,1 = 4γ−2,−1,
γ−1,3 = −12γ2,1,
γ−1,2 = −6γ2,1,
γ−2,−2 = 6γ−2,−1,
γ2,−2 = 12γ2,1 + 6γ−2,−1,
γ2,−3 = −30γ−2,−1 − 9γ2,1, γ2,1 = −γ−2,−1.
γ1,−3 = −12γ−2,−1,
γ1,−2 = −6γ−2,−1,
γ2,−1 = 3γ2,1,
γ3,−2 = 20γ2,1 + 6γ−2,−1,
γ−2,2 = −12γ2,1 − 6γ−2,−1, γ−3,2 = 20γ−2,−1 + 6γ2,1,
γ3,−1 = 4γ2,1,
γ−2,1 = 3γ−2,−1,
γ2,2 = 6γ2,1,
γ−2,3 = −30γ2,1 − 9γ−2,−1,
Note that from (8) with j = 2 we have γ2,−2 = γ2,2.
From the identity (10) we obtain
γi,1 = (i − 1)γ2,1,
γ−i,−1 = (i − 1)γ−2,−1 = −(i − 1)γ2,1,
i ∈ Z+ \ {1, 2},
i ∈ Z+ \ {1, 2}.
Considering (9) for i = ±1, j := ±i, k := ∓1, j and applying (7) we conclude
γi,−1 = (i + 1)γ2,1, i ∈ Z+ \ {1},
γ−i,1 = −(i + 1)γ2,1, i ∈ Z+ \ {1},
(1 − i)γ1+i,j = (1 − j)γ1+j,i − (i − j)(i + j − 1)γ2,1.
Analyzing these obtained equalities for j ≤ −2, we derive
γi,j = j(ij − 1)γ2,1, i 6= ±j, i, j ∈ Z \ {−1, 0, 1}
Now equality (9) with j := i and k := j implies
γi,i = (i3 − i)γ2,1,
i ∈ Z \ {−1, 0, 1}.
In addition, from (8) we get γi,−i = (i3 − i)γ2,1, i ∈ Z \ {−1, 0, 1}.
If γ2,1 6= 0, then by rescaling the basis we can assume γ2,1 = 1 and the third algebra of the list
of theorem is obtained.
If γ2,1 = 0, we have the algebra (I) with β = 1.
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
11
Case 1.2. Let β = 0 be. Then (8) has the form γj,j = 0, j 6= 0 and the have the algebra has
the following table of multiplications:
[v(k), di] = (k + α)v(i + k),
[d0, dj] = −jdj,
[dj , d0] = jdj,
[dj , dj] = 0,
[dj , d−j] = 2jd0 + γj,−jv(−α),
[di, dj] = (i − j)di+j + γi,jv(i + j − α),
j 6= 0,
j 6= 0,
j 6= 0,
i 6= j,
i 6= −j.
Taking the change of d0 as follows d′
0 = d0 + 1
2 γ1,−1v(−α) we may suppose γ1,−1 = 0. Then
from (6) we get γ−1,1 = 0.
Substituting in equality (9) the values of parameters (i, j, k) as (±1, ±2, ±1) we get the following
relations between structure constants:
γ3,−1 = 3γ1,2,
γ−3,1 = 3γ−1,−2,
γ−1,2 = γ−1,3,
γ1,−2 = 2γ1,−3,
γ1,3 = −γ3,1,
Thanks to (7) the above relations can be written as
γ1,−2 = γ−1,2 = 0,
γ−1,−3 = −γ−3,−1.
γ±1,±2 = γ±2,±1 = γ±1,±3 = γ±3,±1 = 0.
Equality (9) with i = ∓1, j := i, k = ∓1 with i /∈ {0, 1, −1} together with (7) lead to
(1 + i)(i − 2)γi−1,−1 = (i − 1)iγi,−1,
(i + 2)(i − 1)γi+1,1 = (i + 1)iγi,1,
from which we obtain
γ−1,i = γi,−1 = 0,
γ1,i = γi,1 = 0,
i ∈ Z \ {−1, 0, 1}.
Again we consider equality (9) with i = 1, j := i, k := j, i 6= ±j, i, j ∈ Z \ {−1, 0, 1} and with
j := 1 − i, k := −1, i ∈ Z+ \ {1, 2}, we derive
(1 − i)γ1+i,j = (1 − j)γ1+j,i, i /∈ {j, −j}, i, j ∈ Z \ {−1, 0, 1},
(2 − i)γi,−i = −(i + 1)γi−1,1−i, i ∈ Z+ \ {1, 2}.
The simple analysis of these relations bring us to
γi,j = 0, i, j ∈ Z
and hence, we obtain the algebra (I) with β = 0.
Case 2. Let β = −1. This case is carried out in a similar way as Case 1 and we obtain the
(cid:3)
rest pf the algebras from the list of the theorem.
Remark 1. From Theorem 3 we conclude that in the infinite-dimensional case, unlike to the
finite-dimensional case there are many simple Leibniz algebras whose corresponding Lie algebra is
a simple Lie algebra and the ideal I is an irreducible right module over the simple Lie algebra.
4. Annihilation of low-dimensional Leibniz cohomology groups of the Witt
algebra.
In this section we prove the coincidence of the second Lie and Leibniz cohomologies for Witt
algebra. We start with the table of multiplication of the Witt algebra W. Let us consider a basis
{di, i ∈ Z} of W in which the table of multiplications of W has the form (2).
The following theorem is true.
Theorem 4. HLi(W, W) = 0,
i = 1, 2.
Proof. Taking into account that for any Lie algebra G we have HL1(G, G) = H 1(G, G) (see [21])
and the result H 1(W, W) = 0 of paper [13], we conclude HL1(W, W) = 0.
Let ϕ(di, dj ) =Pk
From the equalities
αk
i,jdk be a Leibniz 2-cocycle.
Φ(ϕ)(d0, d0, d0) = Φ(ϕ)(di, d0, d0) = Φ(ϕ)(d0, di, di) = Φ(ϕ)(di, di, d0) = Φ(ϕ)(di, di, dj) = 0
12
LEIBNIZ ALGEBRAS CONSTRUCTED BY WITT ALGEBRAS
we obtain
ϕ(di, di) = 0, i ∈ Z, ϕ(di, dj ) = −ϕ(dj, di), i 6= j.
Hence, ϕ is a Lie 2-cocycle, that is, ZL2(W, W) = Z 2(W, W).
For an arbitrary Leibniz 2-coboundary f we have have the existence of a linear map g. Since
f (di, dj) = [g(di), dj] + [di, g(dj)] − g([di, dj]) = −[dj, f (di)] − [g(dj), di] + g([dj, di]) = −f (dj, di),
we conclude BL2(W, W) = B2(W, W). Therefore,
HL2(W, W) = H 2(W, W).
Due to [15] we have H 2(W, W). Hence, HL2(W, W) = 0.
(cid:3)
References
[1] Adashev J.Q., Omirov B.A., Uguz S., Leibniz algebras associated with representations of Euclidean Lie algebra,
arxiv:1607.04949. (2016)
[2] Albeverio S.A., Ayupov Sh.A., Omirov B.A., On nilpotent and simple Leibniz algebras, Comm. Algebra. 33(1)
(2005), 159-172.
[3] Albeverio S.A., Ayupov Sh.A., Omirov B.A., Cartan subalgebras, weight spaces, and criterion of solvability of
finite dimensional Leibniz algebras, Rev.Mat.Complut., 19(1)(2006)183-195.
[4] Ayupov Sh.A., Camacho L.M., Khudoyberdiyev A.K., Omirov B.A., Leibniz algebras associated with represen-
tations of filiform Lie algebras, J.Geom.Phys. 98(2015)181-195.
[5] Ayupov Sh.A., Kudaybergenov K.K., Omirov B.A., Zhao K., Semisimple Leibniz algebras and their derivations
and automorphisms, arxiv., 1708.08082v1.
[6] Ayupov Sh.A., Omirov B.A., On Leibniz algebras, Algebra and operator theory (Tashkent, 1997), Kluwer Acad.
Publ., Dordrecht, 1998, pp. 1-12.
[7] Barnes, D. On Engel's Theorem for Leibniz Algebras, Comm. Algebra. 40 (4) (2012), 1388-1389.
[8] Barnes D., On Levi's Theorem for Leibniz algebra, Bull. of the Australian Math. Soc., 86 (2012), 184-185.
[9] Bosko, L., 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.
[10] Calderón A.J., Camacho L.M., Omirov B.A., Leibniz algebras of Heisenberg type, J.Algebra., 452 (2016) 427-
447.
[11] Camacho L.M., Karimjanov I.A., Ladra M., Omirov B.A., Leibniz algebras constructed by representations of
General Diamond Lie algebras, Bull. Malays. Math. Sci. Soc. (2017). https://doi.org/10.1007/s40840-017-0541-5
[12] Cartan E., Les groupes de transformations continus, infinis, simples. Ann. Sci. Ecole Norm. Sup., 26 (1909)
93-161.
[13] Ecker J., Schlichenmaier M., The vanishing of low-dimensional cohomology groups of the Witt and the Virasoro
algebra, arxiv., 1707.06106v1.
[14] Eswara Rao S., Representations of Witt algebras, Publ. Res. Inst. Math. Sci. 30 (1994) 191 201.
[15] Fialowski A., Schlichenmaier M. Global deformations of the Witt algebra of Krichever-Noviker type, Commu-
nications in Contemparary Mathematics., 5(6) (2003) 921-945.
[16] Humphreys J.E., Introduction to Lie algebras and Representation theory, Springer-Verlag New York, 1972.
[17] Jacobson N., Lie algebras, 340p. Interscience Publishers, Wiley, New York, (1962).
[18] Kac V., Raina A., Bombay lectures on highest weight representations of infinite-dimensional Lie algebras, World
Sci.Singapore, 1987.
[19] Kinyon M.K., Weinstein A., Leibniz algebras, Courant algebroids, and multiplications on reductive homoge-
neous spaces, Amer. J. Math., 123 (3) (2001) 525-550.
[20] Loday J.-L. Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, Enseign.Math., (2) 39
(3-4) (1993) 269-293.
[21] Loday J.-L., Pirashvili T., Universal envoloping algebras of Leibniz algebras and (co)homology, Math.Ann 296,
(1993), 139-158-748.
[22] L u R., Guo X., Zhao K., Irreducible modules over the Virasoro algebra, Doc.Math., 16 (2011), 709-721.
[23] Omirov, B., Conjugacy of Cartan subalgebras of complex finite dimensional Leibniz algebras, J. Algebra. 302(2)
(2006), 887-896.
[24] Uguz S., Karimjanov I.A., Omirov B.A., Leibniz algebras associated with representations of the Diamond Lie
algebra, Algebr Represent Theor, 20 (2017) 175-195. https://doi.org/10.1007/s10468-016-9636-1
[25] Zhao K., Weight modules over generalized Witt algebras with 1-dimensional weight spaces, Forum Math. Vol.16
(2004), 725-748.
|
1106.0393 | 1 | 1106 | 2011-06-02T09:40:16 | A kind of infinite-dimensional Novikov algebras and its realization | [
"math.RA"
] | In this paper, we construct a kind of infinite-dimensional Novikov algebras and give its realization by hyperbolic sine functions and hyperbolic cosine functions. | math.RA | math |
A kind of infinite-dimensional Novikov
algebras and its realization ∗
Liangyun Chen, Yao Ma, Haijun Yu
Department of Mathematics, Northeast Normal University,
Changchun 130024, CHINA
Abstract
In this paper, we construct a kind of infinite-dimensional Novikov algebras and
give its realization by hyperbolic sine functions and hyperbolic cosine functions.
Key words: Novikov algebras, left symmetric algebras, adjoining Lie algebras.
AMS Subject Classification: 17A30.
§1
Introduction
The Hamilton operator is an important operator of the calculus of variations. When
I. M. Gel'fand and I. Ya. Dorfman [5-7] studied the following operator:
Hij = Xk
cijku(1)
k + dijku(0)
k
d
dx
,
cijk ∈ C,
dijk = cijk + cjik,
they gave the definition of Novikov algebras. Concretely, let cijk be the structural
coefficients, a product of L = L(e0, e1, · · ·) be ◦ such that
ei ◦ ej = X cijkek.
Then the product is Hamilton operator if and only if ◦ satisfies
(a ◦ b) ◦ c = (a ◦ c) ◦ b
∗Supported by NNSF of China (No.10871057), Natural Science Foundation of Jilin
province (No.201115006), Scientific Research Foundation for Returned Scholars Ministry of
Education of China and the Fundamental Research Funds for the Central Universities.
Corresponding author(L. Chen): [email protected]
1
(a ◦ b) ◦ c + c ◦ (a ◦ b) = (c ◦ b) ◦ a + a ◦ (c ◦ b).
In 1987, E. I. Zel'manov[15] began to study Novikov algebras and proved that
the dimension of finite simple Novikov algebras over a field of characteristic zero is
one. In algebras, what are paid attention to by mathematician are classifications
and structures, but so far we haven't got the systematic theory for general Novikov
algebras. In 1992, J. M. Osborn [9-10] had finished the classification of infinite simple
Novikov algebras with nilpotent elements over a field of characteristic zero and finite
simple Novikov algebras with nilpotent elements over a field of characteristic p > 0.
In 1995, X. P. Xu [11-14] developed his theory and got the classification of simple
Novikov algebras over an algebraically closed field of characteristic zero. C. M. Bai
and D. J. Meng [1-3] has serial work on low dimensional Novikov algebras, such as
the structure and classification. We construct two kinds of Novikov algebras [4].
Recently, people obtain some properties in Novikov superalgebras [8, 16].
In this paper, we construct a new infinite-dimensional Novikov algebras and give
its realization by hyperbolic sine functions and hyperbolic cosine functions.
Definition 1.1
Let (A, ◦) be an algebra over F such that:
(a, b, c) = (b, a, c),
(a ◦ b) ◦ c = (a ◦ c) ◦ b,
∀a, b, c ∈ A,
(1.1)
(1.2)
then A is called a Novikov algebra over F.
Remark 1.2
(1) Condition (1.1) is usually written by
a ◦ (b ◦ c) − (a ◦ b) ◦ c = b ◦ (a ◦ c) − (b ◦ a) ◦ c.
(1.3)
(2) An algebra A is called a left symmetric algebra if it only satisfies (1.1). It is
clear that left symmetric algebras contain Novikov algebras.
Remark 1.3
(1) If (A, ◦) is a left symmetric algebra satisfying
[a, b] = a ◦ b − b ◦ a,
∀a, b ∈ A,
(1.4)
then (A, [, ]) is a Lie algebra. Usually, it is called an adjoining Lie algebra.
(2) Let (A, ·) be a commutative algebra, then (A, d0, ◦) is a Novikov algebra if
d0 is a derivation of A with a bilinear operator ◦ such that
a ◦ b = a · d0(b),
∀a, b ∈ A.
(1.5)
2
§2 Main results
Lemma 2.1 Let { b0, a1, b1, a2, b2, · · · an, bn, · · · } be a basis of the linear space
A over a field F of characteristic p 6= 2 satisfying
aman =
1
2
(bm+n − bm−n),
bmbn =
1
2
(bm+n + bm−n),
ambn = bnam =
1
2
(am+n + am−n),
(2.1)
where b−m = bm, a−m = −am. Then A is a commutative and associative algebra.
Proof. It is clear that A is a commutative algebra over F.
=
=
ak(bm+n − bn−m) −
(bk+n − bk−n)am
(ak, an, am) = ak(anam) − (akan)am
1
2
1
4
−am+k+n − am−k−n + am+k−n + am−k+n)
(ak+m+n + ak−m−n − ak+n−m − ak−n+m
1
2
= 0.
Similarly, we have that (bk, bn, bm) = (ak, an, bm) = (ak, bn, am) = (bk, an, am) =
(bk, bn, am) = (bk, an, bm) = (ak, bn, bm) = 0. Then (a, b, c) = 0, ∀a, b, c ∈ A. The
result follows.
(cid:3)
Corollary 2.2 b0 of Lemma 2.1 is a unity of A.
Lemma 2.3 Let A be a commutative and associative algebra satisfying Lemma
2.1. Then the following statements hold:
1)
If D0 is a linear transformation of A such that
(cid:26) D0(an) = nbn, n = 1, 2, 3, · · · ,
D0(bn) = nan, n = 0, 1, 2, · · · ,
(2.2)
then D0 is a derivation of A.
2)
If aD0 is a linear transformation of A such that
(aD0)(b) = aD0(b), ∀a, b ∈ A,
(2.3)
then aD0 is a derivation of A.
3
3) D1 = {aD0a ∈ A} is a subalgebra of Lie algebra DerA.
Proof.
(1) We have
D0(anam) = D0(cid:18)1
2
(bn+m − bn−m)(cid:19) =
1
2
((m + n)an+m − (n − m)an−m)
and
D0(an)am + anD0(am) = nbnam + manbm
=
=
n
2
1
2
(an+m − an−m) +
m
2
(an+m − am−n)
((m + n)am+n − (n − m)an−m).
So D0 is a derivation of A.
2) For ∀a, b, c ∈ A, we have
(aD0)(bc) = aD0(bc) = aD0(b)c + abD0(c) = (aD0)(b)c + b(aD0)(c),
so aD0 is a derivation of A.
3) For ∀a, b, c ∈ A, we have
[aD0, bD0](c) = (aD0)(bD0)(c) − (bD0)(aD0)(c)
= aD0(b)D0(c) − bD0(a)D0(c)
= (aD0(b) − bD0(a))D0(c).
Then [aD0, bD0] = (aD0(b) − bD0(a))D0 ∈ D1, and so 3) holds.
(cid:3)
Theorem 2.4 Let A be a commutative and associative algebra satisfying Lemma
2.1, and let a be an element of A. If D0 satisfies Lemma 2.3 and ◦ satisfies
b ◦ c = baD0(c), ∀b, c ∈ A,
(2.4)
then the following statements hold:
1)
2)
(A, aD0, ◦) is a Novikov algebra.
(A, aD0, [, ]) is an adjoining Lie algebra of (A, aD0, ◦) and [ , ] such that
[b, c] = a(bD0(c) − cD0(b)),
∀b, c ∈ A.
(2.5)
Proof. 1) By Lemma 2.3, aD0 is a derivation of the commutative algebra A. So
(A, aD0, ◦) is a Novikov algebra by Remark 1.3 (2).
4
2)
(A, aD0, [, ]) is an adjoining Lie algebra of (A, aD0, ◦) by Remark 1.3 (1).
For ∀b, c ∈ A, ∃a ∈ A, we have
[b, c] = b ◦ c − c ◦ b = baD0(c) − caD0(b) = a(bD0(c) − cD0(b))
since A is commutative. Hence we obtain the desired result.
(cid:3)
Let b0 be a unity of A.
If we set a = b0 in Theorem 2.4, then an ◦ am =
anb0D0(am) = an(mbm) = m
2 (am+n + an−m). Similarly, we obtain the following
corollary:
Corollary 2.5 Let A be a commutative and associative algebra satisfying Lemma
2.1. Then the following statements hold:
an ◦ am = m
bn ◦ bm = m
an ◦ bm = m
bn ◦ am = m
2 (an+m + an−m)
2 (an+m + am−n)
2 (bn+m − bn−m)
2 (bn+m + bn−m)
[an, am] = 1
[bn, bm] = 1
[an, bm] = 1
[bn, am] = 1
2(m − n)an+m + 1
2(m − n)an+m − 1
2 (m − n)bn+m − 1
2 (m − n)bn+m + 1
2(m + n)an−m
2 (m + n)an−m
2 (n + m)bn−m
2(m + n)bn−m.
and
(2.6)
(2.7)
The following, let sinh x = ex
−e−x
2
, cosh x = ex+e−x
2
the field F be assumed R or
C. We will construct Novikov algebras over the linear space which is generated by
sinh x and cosh x.
First, let T be a linear space generated by {sinh mx, cosh nxm, n ∈ N} over F.
Lemma 2.6 T satisfying the above product is a commutative associative algebra.
Proof. Since the above product is commutative and associative, we only need
that T is closed for the product. In fact,
sinh mx sinh nx =
1
2
[cosh(m + n)x − cosh(m − n)x]
cosh mx cosh nx =
1
2
[cosh(m + n)x + cosh(m − n)x]
(2.8)
sinh mx cosh nx =
1
2
[sinh(m + n)x + sinh(m − n)x].
So T is a commutative and associative algebra.
(cid:3)
5
Lemma 2.7 Let T be a linear space generated by {sinh mx, cosh nxm, n ∈ N}
over F, then {1, sinh mx, cosh nxm, n ∈ N0} is a basis of T .
Proof. For ∀n ∈ N0, suppose that there are c0, ai, bj ∈ F, i, j ∈ N0 such that
c0 + a1 sinh x + b1 cosh x + · · · + an sinh nx + bn cosh nx = 0.
(2.9)
We take derivative for (2.9) such that its derivative order is 2k − 1 (k ∈ N0), and
put x = 0. Then we have
a1 + 22k−1a2 + · · · + n2k−1an = 0.
Let k = 1, 2, · · · , n, then we obtain the following system of n linear equations:
a1 + 2a2 + · · · + nan = 0
a1 + 23a2 + · · · + n3an = 0
· · · · · · · · · · · · · · · · · · · · · · · · · · ·
a1 + 22n−1a2 + · · · + n2n−1an = 0.
(2.10)
If a1, · · · , an are seen to be unknown, then the coefficient matrix of (2.10) is the
Vandermonde matrix whose determinant is not 0, so ai = 0, i = 1, · · · , n.
We take derivative for (2.9) such that its derivative order is 2k (k ∈ N0), and
put x = 0. Then we have
b1 + 22kb2 + · · · + n2kbn = 0.
Let k = 1, 2, · · · , n, then we obtain the following system of n linear equations:
b1 + 22b2 + · · · + n2bn = 0
b1 + 24b2 + · · · + n4bn = 0
· · · · · · · · · · · · · · · · · · · · · · · · · · ·
b1 + 22nb2 + · · · + n2nbn = 0.
(2.11)
If b1, · · · , bn are seen to be unknown, then the coefficient matrix of (2.11) is the
Vandermonde matrix whose determinant is not 0, so bi = 0, i = 1, · · · , n. Since
for ∀i ∈ N0, ai = 0 and bi = 0 satisfy (2.9), we have c0 = 0. Hence {1,
sinh x, cosh x, · · ·, sinh nx, cosh nx} are linearly independent for ∀n ∈ N0, then
{1, sinh nx, cosh mxn, m ∈ N0} are linearly independent and so they form a basis
of T as desired.
(cid:3)
Theorem 2.8 Let A1, A2 be commutative and associative algebras over F. If ϕ:
A1 −→ A2 is an isomorphism and D1 ∈ DerA1, then the following statements hold:
6
1) D2 := ϕD1ϕ−1 ∈ DerA2.
2) ϕ: (A1, D1, ◦) −→ (A2, D2, ◦) is also an isomorphism of Novikov algebras.
Proof. 1) For ∀a, b ∈ A1, we have
(ϕD1ϕ−1)(ϕ(a)ϕ(b)) = (ϕD1ϕ−1)(ϕ(ab))
= ϕD1(ab) = ϕ(D1(a)b + aD1(b)) = ϕ(D1(a))ϕ(b) + ϕ(a)ϕ(D1(b))
= (ϕD1ϕ−1)(ϕ(a))ϕ(b) + ϕ(a)(ϕD1ϕ−1)(ϕ(b)).
So 1) holds.
2) For ∀a, b ∈ A1, we have
ϕ(a ◦ b) = ϕ(aD1(b)) = ϕ(a)ϕ(D1(b))
= ϕ(a)(ϕD1ϕ−1)(ϕ(b)) = ϕ(a)D2(ϕ(b))
= ϕ(a) ◦ ϕ(b).
So 2) holds.
(cid:3)
Theorem 2.9 Let A be a commutative and associative algebra over F satisfy-
ing Lemma 2.1, D0 be its derivation satisfying (2.2) and T be a commutative and
associative algebra over F satisfying Lemmas 2.6 and 2.7. If ϕ : A −→ T satisfies
ϕ(bm) = cosh mx, m = 0, 1, 2, · · · , ϕ(an) = sinh nx, n = 1, 2, · · · ,
(2.12)
then the following statements hold:
1) ϕ is an isomorphism of commutative and associative algebras.
2) ϕD0ϕ−1 = d
dx .
3) ϕ : (A, aD0, ◦) −→ (T , ϕ(a) d
dx , ◦) is an isomorphism of Novikov algebras.
Proof. It is clear by Lemma 2.7, (2.1) and (2.8).
2) By (2.2) and (2.12), we have
ϕD0ϕ−1(sinh nx) = ϕD0(an)
= ϕ(nbn) = n cosh nx
=
d sinh nx
dx
,
ϕD0ϕ−1(cosh nx) = ϕD0(bn)
= ϕ(nan) = n sinh nx
=
d cosh nx
dx
.
7
So 2) holds.
3)
It is clear that ϕ(aD0)ϕ−1 = ϕ(a)d/dx. By (2.12) and (2.2), we have
ϕ(aD0)ϕ−1(sinh nx) = ϕ(aD0)(an)
= ϕ(aD0(an)) = ϕ(anbn)
= ϕ(a)ϕ(nbn) = ϕ(a)n cosh nx
= ϕ(a)d(sinh nx)/dx.
Similarly, we have ϕ(aD0)ϕ−1(cosh nx) = ϕ(a)d(cosh nx)/dx. So ϕ(aD0)ϕ−1 =
ϕ(a)d/dx.
By Theorems 2.4, 2.8 and Remark 1.3 (2), we have
ϕ(b ◦ c) = ϕ(baD0(c))
= ϕ(b)ϕ(aD0(c))
= ϕ(b)[ϕ(aD0)ϕ−1(ϕ(c))]
= ϕ(b)ϕ(a)d/dx(ϕ(c))
= ϕ(b) ◦ ϕ(c), ∀b, c ∈ A.
So ϕ : (A0, aD0, ◦) −→ (T , ϕ(a) d
dx, ◦) is an isomorphism of Novikov algebras.
(cid:3)
References
[1] Bai C M and Meng D J 2000 The structure of bi-symmetric algebras and their
subadjacent Lie algebras Commun. Algebra 28 2717-2734
[2] Bai C M and Meng D J 2001 The classification of Novikov algebras in low
dimensions J. Phys. A: Math. Gen. 34 1581-1594
[3] Bai C M and Meng D J 2001 The realizations of non-transitive Novikov algebrss
J. Phys. A: Math. Gen. 34 6435-6442
[4] Chen L Y, Niu Y J and Meng D J 2008 Two kinds of Novikov algebras and
their realizations J. Pure Appl. Algebra 212 902-909
8
[5] Gel'fand I M and Diki L A 1975 Asymptotic behavior of the resolvent of Sturm-
Liou-ville equations and the Lie algebra of the Korteweg-de Vries equations
Russian Math. Surveys 30 77-113
[6] Gel'fand I M and Diki L A 1976 A Lie algebra structure in a formal variational
cal-culation Functional Anal. Appl. 10 16-22
[7] Gel'fand I M and Dorfman I Y 1979 Hamiltonian operators and algebraic struc-
tures related to them Functional Anal. Appl. 13 248-262
[8] Kang Y F and Chen Z Q 2009 Novikov superalgebras in low dimensions J.
Nonlinear math.Phy. 16(3) 251-257
[9] Osborn J M 1992 simple Novikov algebras with an idempotent Commun. Alge-
bra 20 2729-2753
[10] Osborn J M 1994 Infinite dimentional Novikov algebras of characteristic 0 J.
Algebra 16 146-167
[11] Xu X P 1995 Hamiltonian operators and associative algebras whih a derivation
Lett. Math. Phys. 33 1-6
[12] Xu X P 1996 On simple Novikov algebra and their irreducible modules J. Al-
gebra 185 905-934
[13] Xu X P 1997 Novikov-Poisson algebras J. Algebra 190 253-279
[14] Xu X P 2000 Variational calculus of supervarables and related algebraic struc-
tures J. Algebra 223 396-437
[15] Zel'manov E I 1987 On a class of local translation invariant Lie algebras Soviet
Math. Dokl. 35(1) 216-218
[16] Zhu F H and Chen Z Q 2010 Novikov superalgebras with A0 = A1A1 Czechoslo-
vak Mathematical Journal 60(135) 903-907
9
|
1004.4761 | 1 | 1004 | 2010-04-27T10:58:56 | Analogues of the adjoint matrix for generalized inverses and corresponding Cramer rules | [
"math.RA"
] | In this article, we introduce determinantal representations of the Moore - Penrose inverse and the Drazin inverse which are based on analogues of the classical adjoint matrix. Using the obtained analogues of the adjoint matrix, we get Cramer rules for the least squares solution and for the Drazin inverse solution of singular linear systems. Finally, determinantal expressions for ${\rm {\bf A}}^{+} {\rm {\bf A}}$, ${\rm {\bf A}} {\rm {\bf A}}^{+}$, and ${\rm {\bf A}}^{D} {\rm {\bf A}}$ are presented. | math.RA | math |
Analogues of the adjoint matrix for generalized
inverses and corresponding Cramer rules.
Kyrchei I. I.∗
Abstract
In this article, we introduce determinantal representations of the
Moore - Penrose inverse and the Drazin inverse which are based on
analogues of the classical adjoint matrix. Using the obtained analogues
of the adjoint matrix, we get Cramer rules for the least squares solution
and for the Drazin inverse solution of singular linear systems. Finally,
determinantal expressions for A+A, AA+, and ADA are presented.
Keywords: Moore-Penrose inverse; Drazin inverse; system of linear equations;
least squares solution; Cramer rule
AMS classification: 15A09, 15A57
1
Introduction
Determinantal representation of the Moore - Penrose inverse was studied in
[1, 2, 8, 11, 12]. The main result consists in the following theorem.
Theorem 1.1 The Moore - Penrose inverse A+ = (a+
has the following determinantal representation
ij)n×m of A ∈ Cm×n
r
a+
ij =
α(cid:12)(cid:12)(cid:12) ∂
P(α, β)∈Nr{j, i}(cid:12)(cid:12)(cid:12)(A∗)β
∂aj i(cid:12)(cid:12)Aα
β(cid:12)(cid:12)
γ(cid:12)(cid:12)(cid:12) Aγ
P(γ, δ)∈Nr(cid:12)(cid:12)(cid:12)(A∗)δ
δ
, 1 ≤ i, j ≤ n.
∗Pidstrygach Institute for Applied Problems of Mechanics and Mathematics,
str.Naukova 3b, Lviv, Ukraine, 79005, [email protected]
1
Stanimirovic' [13] introduced a determinantal representation of the Drazin
inverse by the following theorem.
Cn×n with IndA = k possesses the following determinantal representation
Theorem 1.2 The Drazin inverse AD = (cid:0)aD
ij(cid:1) of an arbitrary matrix A ∈
α(cid:12)(cid:12)(cid:12) ∂
P(α, β)∈Nrk {j, i}(cid:12)(cid:12)(cid:12)(As)β
∂aj i(cid:12)(cid:12)Aα
β(cid:12)(cid:12)
γ(cid:12)(cid:12)(cid:12) Aγ
P(γ, δ)∈Nrk(cid:12)(cid:12)(cid:12)(As)δ
where s ≥ k and rk = rankAs.
These determinantal representations of generalized inverses are based on cor-
responding full-rank representations.
, 1 ≤ i, j ≤ n;
aD
ij =
δ
r
Then (cid:12)(cid:12)Aα
We use the following notations from [1, 12]. Let Cm×n be the set of m by
n matrices with complex entries, Cm×n
be the subset of Cm×n in which every
matrix has rank r. Im denotes the identity matrix of order m, and k.k = k.k2
is the Euclidean vector norm. Let α := {α1, . . . , αk} ⊆ {1, . . . , m} and
β := {β1, . . . , βk} ⊆ {1, . . . , n} be subsets of the order 1 ≤ k ≤ min {m, n}.
β(cid:12)(cid:12) denotes the minor of A ∈ Cm×n determined by the rows in-
dexed by α and the columns indexed by β. Clearly, Aα
α denotes a prin-
cipal minor determined by the rows and columns indexed by α. The co-
factor of aij in A ∈ Cn×n is denoted by
A. For 1 ≤ k ≤ n, de-
note by Lk,n := { α : α = (α1, . . . , αk) , 1 ≤ α1 ≤ . . . ≤ αk ≤ n} the collec-
tion of strictly increasing sequences of k integers chosen from {1, . . . , n}. Let
Nk := Lk,m × Lk,n. For fixed α ∈ Lp,m, β ∈ Lp,n, 1 ≤ p ≤ k, let
∂aij
∂
Ik, m (α) := {I : I ∈ Lk, m, I ⊇ α},
Jk, n (β) := {J : J ∈ Lk, n, J ⊇ β},
Nk (α, β) := Ik, m (α) × Jk, n (β)
For case i ∈ α and j ∈ β, we denote
Ik,m{i} := {α : α ∈ Lk,m, i ∈ α}, Jk, n{j} := {β : β ∈ Lk,n, j ∈ β},
Nk{i, j} := Ik, m{i} × Jk, n{j}.
In this paper we introduce determinantal representations of the Moore -
Penrose inverse and of the Drazin inverse based on corresponding limit rep-
resentations. The obtained determinantal representations can be considered
as founded on some analogues of the classical adjoint matrix. The corre-
sponding Cramer rules for the complex system of linear equations with a
rectangular or singular coefficient matrix follow from these analogues.
2
2 Analogues of the classical adjoint matrix
for the Moore - Penrose inverse
We shall use the following well-known facts (see, for example, [9]).
Definition 2.1 The matrix A+ ∈ Cn×m is called the Moore - Penrose in-
verse of an arbitrary A ∈ Cm×n if it satisfies the equations
AA+A = A; A+AA+ = A+; (AA+)∗ = AA+; (A+A)∗ = A+A.
The superscript ∗ denotes conjugate transpose matrix.
Lemma 2.1 [9] There exists a unique Moore - Penrose inverse A+ of A ∈
Cm×n.
Lemma 2.2 [9] If A ∈ Cm×n, then
A+ = lim
λ→0
A∗ (AA∗ + λI)−1 = lim
λ→0
(A∗A + λI)−1 A∗,
where λ ∈ R+, and R+ is the set of positive real numbers.
Lemma 2.3 [9] If A ∈ Cm×n, then the following statements are true.
i) If rank A = n, then A+ = (A∗A)−1 A∗ .
ii) If rank A = m, then A+ = A∗ (AA∗)−1 .
iii) If rank A = n = m, then A+ = A−1 .
Theorem 2.1 [9] Let dr be the sum of principal minors of order r of A ∈
Cn×n. Then its characteristic polynomial pA (t) can be expressed as pA (t) =
det (tI − A) = tn − d1tn−1 + d2tn−2 − . . . + (−1)n dn.
Denote by a.j and ai. the jth column and the ith row of A ∈ Cm×n respec-
tively. In the same way, denote by a∗
i. the jth column and the ith row
of Hermitian adjoint matrix A∗. Let A.j (b) denote the matrix obtained from
A by replacing its jth column with some vector b, and let Ai. (b) denote the
matrix obtained from A by replacing its ith row with b.
.j and a∗
Lemma 2.4 If A ∈ Cm×n
r
, then rank (A∗A). i(cid:0)a∗
.j(cid:1) ≤ r.
3
Proof . Let Pi k (−aj k) ∈ Cn×n, (k 6= i), be the matrix with −aj k in the
(i, k) entry, 1 in all diagonal entries, and 0 in others. It is the matrix of an
elementary transformation. It follows that
(A∗A). i(cid:0)a∗
. j(cid:1) ·Yk6=i
Pi k (−aj k) =
Pk6=j
Pk6=j
. . .
a∗
1kak1
. . . a∗
1j
a∗
nkak1
. . .
. . .
. . . a∗
nj
i−th
. . . Pk6=j
. . . Pk6=j
. . .
. . .
a∗
1kakn
a∗
nkakn
.
The obtained above matrix has the following factorization.
Pk6=j
Pk6=j
. . .
a∗
1kak1
. . . a∗
1j
a∗
nkak1
. . .
. . .
. . . a∗
nj
i−th
. . . Pk6=j
. . . Pk6=j
. . .
. . .
a∗
1kakn
a∗
nkakn
=
=
a∗
11 a∗
12
21 a∗
a∗
22
. . .
. . .
a∗
n1 a∗
n2
1m
. . . a∗
. . . a∗
2m
. . .
. . .
. . . a∗
nm
a11
. . .
0
. . .
am1
. . . 0
. . .
. . . 1
. . .
. . . 0
. . .
. . .
. . . an1
. . .
. . .
. . . 0
. . .
. . .
. . . amn
j − th.
Denote by A :=
a11
. . .
0
. . .
am1
. . . 0
. . .
. . . 1
. . .
. . . 0
. . .
. . .
i−th
j − th. The matrix A is ob-
i−th
. . . a1n
. . .
. . .
. . . 0
. . .
. . .
. . . amn
It follows that rank (A∗A). i(cid:0)a∗
tained from A by replacing all entries of the jth row and of the ith column
with zeroes except that the (j, i) entry equals 1. Elementary transformations
of a matrix do not change its rank.
.j(cid:1) ≤
minnrank A∗, rank Ao. Since rank A ≥ rank A = rank A∗ and rank A∗A =
rank A the proof is completed.(cid:4)
The following lemma is proved in the same way.
Lemma 2.5 If A ∈ Cm×n
r
, then rank (AA∗)i .(cid:0)a∗
j .(cid:1) ≤ r.
4
Theorem 2.2 The Moore-Penrose inverse A+ of A ∈ Cm×n
sented as follows
r
can be repre-
or
dr(A∗A)(cid:17)n×m
, where
A+ =(cid:16) lij
β(cid:12)(cid:12)(cid:12),
lij = Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)(cid:0)(A∗A). i(a∗
β(cid:12)(cid:12)(cid:12), dr(A∗A) = Pβ∈Jr, n(cid:12)(cid:12)(cid:12)(A∗A)β
.j)(cid:1)β
A+ =(cid:16) rij
rij = Pα∈Ir, m{j}(cid:12)(cid:12)((AA∗)j .(a∗
α(cid:12)(cid:12), dr(AA∗) = Pα∈Ir, m
dr(AA∗)(cid:17)n×m
, where
i .))α
(1)
(2)
(AA∗)α
α.
Proof . At first we shall obtain the representation (1). If λ ∈ R+, then the
matrix (λI + A∗A) ∈ Cn×n is Hermitian and rank (λI + A∗A) = n. Hence,
there exists its inverse
(λI + A∗A)−1 =
1
det (λI + A∗A)
L11 L21
L12 L22
. . .
. . .
L1 n L2 n
. . . Ln 1
. . . Ln 2
. . .
. . .
. . . Ln n
,
where Lij (∀i, j = 1, n) is a cofactor in λI + A∗A. By Lemma 2.2, A+ =
lim
λ→0
(λI + A∗A)−1 A∗, so that
A+ = lim
λ→0
det(λI+A∗A).1(a∗
det(λI+A∗A)
. 1)
. . .
det(λI+A∗A). n(a∗
. 1)
det(λI+A∗A)
. . .
. . .
. . .
det(λI+A∗A). 1(a∗
det(λI+A∗A)
. m)
. . .
det(λI+A∗A). n(a∗
. m)
det(λI+A∗A)
.
(3)
From Theorem 2.1 we get
det (λI + A∗A) = λn + d1λn−1 + d2λn−2 + . . . + dn,
where dr (∀r = 1, n − 1) is a sum of principal minors of A∗A of order r and
dn = det A∗A. Since rank A∗A = rank A = r, then dn = dn−1 = . . . =
dr+1 = 0 and
det (λI + A∗A) = λn + d1λn−1 + d2λn−2 + . . . + drλn−r.
(4)
5
In the same way, we have for arbitrary 1 ≤ i ≤ n and 1 ≤ j ≤ m from
Theorem 2.1
det (λI + A∗A). i(cid:0)a∗
.j(cid:1) = l(ij)
1 λn−1 + l(ij)
2 λn−2 + . . . + l(ij)
n ,
l(i j)
where for an arbitrary 1 ≤ k ≤ n − 1, l(ij)
k = Pβ∈Jk, n{i}(cid:12)(cid:12)(cid:12)(cid:0)(A∗A). i(a∗
β(cid:12)(cid:12)(cid:12), and
.j)(cid:1)β
n = det (A∗A). i(cid:0)a∗
. j(cid:1). By Lemma 2.4, rank (A∗A). i(cid:0)a∗
. j(cid:1) ≤ r so that if
k > r, then (cid:12)(cid:12)(cid:12)(cid:0)(A∗A) . i(a∗
β(cid:12)(cid:12)(cid:12) = 0, (∀β ∈ Jk, n{i}, ∀i = 1, n, ∀j = 1, m).
.j)(cid:1)β
k = Pβ∈Jk, n{i}(cid:12)(cid:12)(cid:12)(cid:0)(A∗A) . i(a∗
β(cid:12)(cid:12)(cid:12) = 0 and
.j)(cid:1)β
n = det (A∗A). i(cid:0)a∗
. j(cid:1) = 0, (cid:0)∀i = 1, n, ∀j = 1, m(cid:1). Finally we obtain
Therefore if r + 1 ≤ k < n, then l(ij)
l(i j)
1 λn−1 + l(i j)
2 λn−2 + . . . + l(ij)
r λn−r.
(5)
det (λI + A∗A). i(cid:0)a∗
. j(cid:1) = l(i j)
By replacing the denominators and the numerators of the fractions in
entries of matrix (3) with the expressions (4) and (5) respectively, we get
A+ = lim
λ→0
λn−1+...+l(11)
l(11)
λn−r
1
λn+d1λn−1+...+drλn−r
r
. . .
(n1)
λn−1+...+l
r
(n1)
λn−r
l
1
λn+d1λn−1+...+drλn−r
. . .
. . .
. . .
λn−1+...+l(1m)
l(1m)
λn−r
1
λn+d1λn−1+...+drλn−r
r
. . .
(nm)
λn−1+...+l
r
(nm)
λn−r
l
1
λn+d1λn−1+...+drλn−r
=
=
l(11)
r
dr
. . .
l(n1)
r
dr
. . .
. . .
. . .
l(1m)
r
dr
. . .
l(nm)
r
dr
.
From here the representation (1) of A+ follows by denoting l(ij)
r = lij.
We obtain the representation (2) in the same way. (cid:4)
Remark 2.1 If rank A = n, then from Lemma 2.3 we get A+ = (A∗A)−1 A∗.
Representing (A∗A)−1 by the classical adjoint matrix, we have
A+ =
1
det(A∗A)
det(A∗A).1 (a∗
.1)
. . .
det(A∗A). n (a∗
. 1)
If n < m, then (2) is valid.
6
. . . det(A∗A).1 (a∗
. . .
. . . det(A∗A). n (a∗
. . .
. m)
. m)
.
(6)
Remark 2.2 As above, if rank A = m, then
A+ =
1
det(AA∗)
det(AA∗)1 . (a∗
1 .)
. . .
det(AA∗)1 . (a∗
n .)
. . . det(AA∗)m . (a∗
. . .
. . . det(AA∗)m . (a∗
. . .
1 .)
n .)
.
(7)
If n > m, then (1) is valid as well.
Remark 2.3 The representation (1) can be obtained from Theorem 1.1 by
using the Binet-Cauchy formula. We use another method, which is the same
for the determinantal representations of the Moore-Penrose inverse by (1)
and (2), and of the Drazin inverse by (11).
n−1C r−1
n + C r−1
Remark 2.4 To obtain an entry of A+ by Theorem 1.1 one calculates (C r
nC r
C r−1
m−1) determinants of order r. Whereas by (1) we calculate as much as
m + C r−1
n−1) determinants of order r or we calculate the total of (C r
(C r
m−1)
determinants by (2). Therefore the calculation of entries of A+ by Theorem
2.2 is easier than by Theorem 1.1.
m+
Corollary 2.1 If A ∈ Cm×n
projection matrix P = A+A can be represented as
r
and r < min {m, n} or r = m < n, then the
where d. j denotes the jth column of (A∗A) and, for arbitrary 1 ≤ i, j ≤ n,
pij
dr (A∗A)(cid:19)n×n
,
P =(cid:18)
pij = Pβ∈Jr,n{i}(cid:12)(cid:12)(cid:12)((A∗A) . i(d.j))β
β(cid:12)(cid:12)(cid:12).
Proof . Representing the Moore - Penrose inverse A+ by (1), we obtain
Therefore, for arbitrary 1 ≤ i, j ≤ n we get
P =
dr (A∗A)
1
. . .
. . .
. . .
. . .
l12
l22
. . .
ln2
l11
l21
. . .
ln1
l1m
l2m
. . .
lnm
pi j =Pk Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)((A∗A). i(a∗
. k · ak j))β
= Pβ∈Jr, n{i}Pk (cid:12)(cid:12)(cid:12)((A∗A). i(a∗
corollary can be proved in the same way.
7
a12
a11
a22
a21
. . .
. . .
am 1 am 2
. . . a1n
. . . a2n
. . .
. . .
. . . am n
.
. k))β
β(cid:12)(cid:12)(cid:12) · ak j =
β(cid:12)(cid:12)(cid:12) = Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)(cid:0)(A∗A). i(d∗
β(cid:12)(cid:12)(cid:12).(cid:4)
.j)(cid:1)β
Using the representation (2) of the Moore - Penrose inverse the following
Corollary 2.2 If A ∈ Cm×n
projection matrix Q = AA+ can be represented as
r
, where r < min {m, n} or r = n < m, then a
qij
dr (AA∗)(cid:19)m×m
,
Q =(cid:18)
qi j = Pα∈Ir,m{j}(cid:12)(cid:12)((AA∗)j. (gi. ))α
α(cid:12)(cid:12).
where gi. denotes the ith row of (AA∗) and, for arbitrary 1 ≤ i, j ≤ m,
Remark 2.5 By definition of the classical adjoint Adj(A) for an arbitrary
invertible matrix A ∈ Cn×n one may put Adj(A)·A = det A·In. If A ∈ Cm×n
and rank A = n, then by Lemma 2.3, A+A = In. Representing the matrix
A+ by (6) as A+ =
det(A∗A) , we obtain LA = det (A∗A)·In. This means that
the matrix L is a left analogue of Adj(A) , where A ∈ Cm×n
. If rank A = m,
then by Lemma 2.3, AA+ = Im. Representing the matrix A+ by (7) as
det(AA∗), we obtain AR = Im · det (AA∗). This means that the matrix
A+ =
R is a right analogue of Adj(A), where A ∈ Cm×n
m .
R
n
L
r
L
and r < min{m, n}, then by (1) we have A+ =
Remark 2.6 If A ∈ Cm×n
dr(A∗A), where L = (lij) ∈ Cn×m. From Corollary 2.1 we get LA = dr (A∗A)·
P. The matrix P is idempotent. All eigenvalues of an idempotent matrix
chose from 1 or 0 only. Thus, there exists an unitary matrix U such that
LA = dr (A∗A) Udiag (1, . . . , 1, 0, . . . , 0)U∗, where diag (1, . . . , 1, 0, . . . , 0) ∈
Cn×n is a diagonal matrix. Therefore, the matrix L can be considered as a
left analogue of Adj(A), where A ∈ Cm×n
.
In the same way, if A ∈ Cm×n
r
and r < min{m, n}, then by (2) we
have A+ =
dr(AA∗), where R = (rij) ∈ Cn×m. From Corollary 2.2 we get
AR = dr (AA∗) · Q. The matrix Q is idempotent. There exists an uni-
tary matrix V such that AR = dr (AA∗) Vdiag (1, . . . , 1, 0, . . . , 0)V∗, where
diag (1, . . . , 1, 0, . . . , 0) ∈ Cm×m. Therefore, the matrix R can be considered
as a right analogue of Adj(A) in this case.
R
r
3 An analogue of the classical adjoint matrix
for the Drazin inverse
Definition 3.1 [4, 7] Let A ∈ Cn×n with IndA = k, where a nonnega-
tive integer IndA := min
k∈N ∪{0}(cid:8)rank Ak+1 = rank Ak(cid:9). Then the matrix X
8
satisfying
Ak+1X = Ak; XAX = X; AX = XA
(8)
is called the Drazin inverse of A and is denoted by X = AD. In particular,
if IndA = 1, then the matrix X in (8) is called the group inverse and is
denoted by X = A#.
Remark 3.1 If IndA = 0, then A is nonsingular, and AD ≡ A−1.
The Drazin inverse can be represented explicitly by the Jordan canonical
form as follows.
Theorem 3.1 [4] If A ∈ Cn×n with IndA = k and
A = P(cid:18)C 0
0 N(cid:19) P−1,
(9)
where C is nonsingular, rank C = rank Ak, and N is nilpotent of order k,
then
AD = P(cid:18)C−1 0
0(cid:19) P−1.
0
We use the following theorem about the limit representation of the Drazin
inverse.
Theorem 3.2 [4] If A ∈ Cn×n, then
AD = lim
λ→0(cid:0)λIn + Ak+1(cid:1)−1
Ak,
where k = IndA and λ ∈ R+.
Denote by a(k)
.j and a(k)
i.
the jth column and the ith row of Ak respectively.
Lemma 3.1 If A ∈ Cn×n with IndA = k, then
rank Ak+1
. i (cid:16)a(k)
.j (cid:17) ≤ rank Ak+1,
∀i, j = 1, n.
(10)
Proof . The proof of this lemma is similar to that of Lemma 2.4.
In the following theorem we introduce a determinantal representation of
the Drazin inverse.
9
Theorem 3.3 If IndA = k and rank Ak+1 = rank Ak = r ≤ n for an
arbitrary matrix A ∈ Cn×n, then
,
where
dr(Ak+1)(cid:19)n×n
AD =(cid:18) dij
dr(cid:0)Ak+1(cid:1) = Pβ∈Jr,n(cid:12)(cid:12)(cid:12)(cid:0)Ak+1(cid:1)β
β(cid:12)(cid:12)(cid:12),
β(cid:12)(cid:12)(cid:12)(cid:12), (cid:0)∀i, j = 1, n(cid:1) .
dij = Pβ∈Jr,n{i}(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)Ak+1
.j (cid:17)(cid:17)β
. i (cid:16)a(k)
(11)
Proof . The proof of this theorem is analogous to that of Theorem 2.2 by
using Theorem 2.1, Lemma 3.1, and Theorem 3.2.
In the following corollaries we introduce determinantal representations of
the group inverse A# and the matrix ADA respectively.
Corollary 3.1 If IndA = 1 and rank A2 = rank A = r ≤ n for an arbitrary
matrix A ∈ Cn×n, then
where dr (A2) = Pβ∈Jr,n(cid:12)(cid:12)(cid:12)(A2)β
,
A# =(cid:18) gij
dr (A2)(cid:19)n×n
β(cid:12)(cid:12)(cid:12), gij = Pβ∈Jr,n{i}(cid:12)(cid:12)(cid:12)(A2
. i (a.j))β
β(cid:12)(cid:12)(cid:12), (cid:0)∀i, j = 1, n(cid:1) .
Proof . The proof follows from Theorem 3.3 in view of k = 1.
Corollary 3.2 If IndA = k and rank Ak+1 = rank Ak = r ≤ n for an
arbitrary matrix A ∈ Cn×n, then
where vij = Pβ∈Jr,n{i}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1
. i
Proof . Representing the Drazin inverse AD by (11) we obtain
ADA =(cid:18)
dij
dr (Ak+1)(cid:19)n×n
vij
dr (Ak+1)(cid:19)n×n
.
ADA =(cid:18)
(cid:0)a.j
(k+1)(cid:1)(cid:1)β
,
vij
dr (Ak+1)(cid:19)n×n
β(cid:12)(cid:12)(cid:12), (cid:0)∀i, j = 1, n(cid:1) .
· (aij)n×n =(cid:18)
10
Here for arbitrary 1 ≤ i, j ≤ n we have
β(cid:12)(cid:12)(cid:12)(cid:12)as j =
vi j =Ps Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)(Ak+1). i(a(k)
. s )(cid:17)β
. s · as j)(cid:17)β
= Pβ∈Jr, n{i}Ps (cid:12)(cid:12)(cid:12)(cid:12)(cid:16)(Ak+1). i(a(k)
classical adjoint matrix, where D = dr(cid:0)Ak+1(cid:1)AD.
β(cid:12)(cid:12)(cid:12)(cid:12) = Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)(Ak+1). i(a(k+1)
.j
β(cid:12)(cid:12)(cid:12)(cid:12).(cid:4)
)(cid:17)β
Remark 3.2 The matrix (ADA) is idempotent. Similarly to the case of
Remark 2.6, the matrix D = (dij)n×n can be considered as an analogue of the
4 Cramer rules for generalized inverse solu-
tions
Definition 4.1 Suppose in a complex system of linear equations:
A · x = y
(12)
the coefficient matrix A ∈ Cm×n
Cm. The least squares solution of the system (12) is the vector x0 ∈ Cn sat-
isfying
and a column of constants y = (y1, . . . , ym)T ∈
r
x∈Cn(cid:26)kxk kA · x − yk = min
x∈Cn
kA · x − yk(cid:27),
(cid:13)(cid:13)x0(cid:13)(cid:13) = min
where Cn is an n-dimension complex vector space.
Theorem 4.1 [9] The vector x = A+y is the least squares solution of the
system (12).
Theorem 4.2 The following statements are true for the system of linear
equations (12).
i) If rank A = n, then the components of the least squares solution x =
(x0
1, . . . , x0
n)T are obtained by the formula
x0
j =
det(A∗A). j (f)
det A∗A
,
(cid:0)∀j = 1, n(cid:1) ,
(13)
where f = A∗y.
11
ii) If rank A = r ≤ m < n, then
j = Pβ∈Jr,n{j}(cid:12)(cid:12)(cid:12)((A∗A). j(f))β
β(cid:12)(cid:12)(cid:12)
dr (A∗A)
x0
,
(cid:0)∀j = 1, n(cid:1) .
(14)
Proof. i) If rank A = n, then we can represent A+ by (7). By multiplying
A+ into y we get (13).
ii) If rank A = k ≤ m < n, then A+ can be represented by (1). By
multiplying A+ into y the least squares solution of the linear system (12) is
given by components as in (14). (cid:4)
Using (3) and (8), we can obtain another representation of the Cramer
rule for the least squares solution of a linear system.
Theorem 4.3 The following statements are true for a system of linear equa-
tions written in the form x · A = y.
i) If rank A = m, then the components of the least squares solution x0 =
yA+ are obtained by the formula
x0
i =
det(AA∗)i . (g)
det AA∗
,
(cid:0)∀i = 1, m(cid:1) ,
where g = yA∗.
ii) If rank A = r ≤ n < m, then
i = Pα∈Ir,m{i}
x0
((AA∗) i .(g))α
α
dr (AA∗)
,
(cid:0)∀i = 1, m(cid:1) .
Proof . The proof of this theorem is analogous to that of Theorem 4.2.
Remark 4.1 The obtained formulas of the Cramer rule for the least squares
solution differ from similar formulas in [3, 6, 10, 14, 15, 16]. They give a
closer approximation to the Cramer rule for consistent nonsingular systems
of linear equations.
In some situations, however, people pay more attention to the Drazin
inverse solution of singular linear systems [6, 16]. Consider a general system
of linear equations (12), where A ∈ Cn×n and x, y are vectors in Cn. R(A)
denotes the range of A and N(A) denotes the null space of A. The charac-
teristic of the Drazin inverse solution ADy is given in [16] by the following
theorem.
12
Theorem 4.4 Let A ∈ Cn×n with Ind(A) = k. Then ADy is both the
unique solution in R(Ak) of
Ak+1x = Aky,
(15)
and the unique minimal P-norm least squares solution of (12).
Remark 4.2 The P-norm is defined as kxkP = kP−1xk for x ∈ Cn, where
P is a nonsingular matrix that transforms A into its Jordan canonical form
(9).
Remark 4.3 Since (15) is analogous to the normal system A∗Ax = A∗y,
the system (15) is called the generalized normal equations of (12), (see [16]).
We obtain the Cramer rule for the P-norm least squares solution of (12) in
the following theorem.
Theorem 4.5 Let A ∈ Cn×n with Ind(A) = k. Then the unique minimal
P-norm least squares solution bx = (bx1, . . . ,bxn)T of the system (12) is given
by
∀i = 1, n,
(16)
. i
β(cid:12)(cid:12)(cid:12)
bxi = Pβ∈Jr, n{i}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1
(g)(cid:1)β
β(cid:12)(cid:12)(cid:12)
Pβ∈Jr,n(cid:12)(cid:12)(cid:12)(Ak+1)β
where g = Aky.
Proof . Representing the Drazin inverse by (11) and by virtue of Theorem
4.4, we have
Therefore,
= ADy =
. . .
bx =
bx1
bxn
1
dr (Ak+1)
1
dr (Ak+1)
nXs=1 Xβ∈ Jr,n{i}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1
. i
bxi =
13
. . .
.
d1sys
dnsys
nPs=1
nPs=1
β(cid:12)(cid:12)(cid:12) · ys =
(cid:0)a(k)
. s (cid:1)(cid:1)β
=
=
1
dr (Ak+1) Xβ∈Jr, n{i}
dr (Ak+1) Xβ∈Jr, n{i}
1
. i
nXs=1 (cid:12)(cid:12)(cid:12)(cid:0)Ak+1
β(cid:12)(cid:12)(cid:12) · ys =
(cid:0)a(k)
. s (cid:1)(cid:1)β
nXs=1 (cid:12)(cid:12)(cid:12)(cid:0)Ak+1
β(cid:12)(cid:12)(cid:12).
. s · ys(cid:1)(cid:1)β
(cid:0)a(k)
. i
From this (16) follows immediately. (cid:4)
5 Examples
1. Let us consider the system of linear equations.
2x1 − 5x3 + 4x4 = 1,
7x1 − 4x2 − 9x3 + 1.5x4 = 2,
3x1 − 4x2 + 7x3 − 6.5x4 = 3,
x1 − 4x2 + 12x3 − 10.5x4 = 1.
(17)
The coefficient matrix of the system is the matrix A =
We calculate the rank of A which is equal to 3, and we have
4
2
0 −5
1.5
7 −4 −9
3 −4
7 −6.5
1 −4 12 −10.5
63
−44
−40 −40
−11.5
−44 −40 −11.5
48 −40
62
299 −205
62 −205 170.75
63
−44
−11.5
62
−44 −11.5
48
62
62
170.75
+
(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)
1.5 −6.5 −10.5
At first we obtain entries of A+ by (1):
1
−4
12
, A∗A =
+(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)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
63 −44 −40
−44
48 −40
−40 −40 299
A∗ =
7
2
3
0 −4 −4
−5 −9
7
4
d3(A∗A) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
l11 =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
2 −44 −40
0
48 −40
−5 −40 299
= 25779,
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
−40 −11.5
63
−40
299 −205
−11.5 −205 170.75
48 −40
−40
299 −205
62 −205 170.75
= 102060,
2 −44 −11.5
0
4
170.75
48
62
62
2 −40 −11.5
−5
299 −205
4 −205 170.75
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
14
and so forth. Continuing in the same way, we get
A+ =
1
102060
25779 −4905
20742 −5037
−3840 −2880 −4800 −960
28350 −17010 22680 −5670
39558 −18810 26484 −13074
.
Now we obtain the least squares solution of the system (17) by the matrix
method.
x0 =
x0
11
x0
21
x0
31
x0
41
=
1
102060
25779 −4905
20742 −5037
−3840 −2880 −4800 −960
28350 −17010 22680 −5670
39558 −18810 26484 −13074
·
1
2
3
1
=
=
1
102060
73158
−24960
56700
68316
=
12193
17010
− 416
1071
5
9
5693
8505
.
Next we get the least squares solution of the system (17) by the Cramer rule
(14), where
f =
7
2
3
0 −4 −4
−5 −9
7
4
1.5 −6.5 −10.5
1
2
3
1
=
26
−24
10
−23
.
1
−4
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)
=
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
15
Thus we have
x0
1 =
1
102060
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
1
102060
x0
2 =
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
26 −44 −40
−24
48 −40
10 −40 299
26 −44 −11.5
−24
−23
170.75
48
62
62
26 −40 −11.5
10
299 −205
23 −205 170.75
73158
102060
=
12193
17010
;
63
26 −40
−44 −24 −40
−40
299
10
26 −11.5
63
−44 −24
−11.5 −23 170.75
62
+
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
x0
3 =
x0
4 =
+(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
1
102060
1
102060
(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)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
62
−24 −40
10
299 −205
−23 −205 170.75
−24960
102060
= −
416
1071
;
63 −44
−44
−40 −40
26
48 −24
10
26 −11.5
63
−40
10 −205
−11.5 −23 170.75
+
62
48 −24
−40
10 −205
62 −23 170.75
63
−44
−11.5
26
−44
48 −24
62 −23
48 −40 −24
−40
10
62 −205 −23
299
56700
102060
=
5
9
;
−40
299
26
63
−40
10
−11.5 −205 −23
68316
102060
=
5693
8505
.
(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)(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)(cid:12)(cid:12)(cid:12)(cid:12)
=
2. Let us consider the following system of linear equations.
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+
(18)
x1 − x2 + x3 + x4 = 1,
x2 − x3 + x4 = 2,
x1 − x2 + x3 + 2x4 = 3,
x1 − x2 + x3 + x4 = 1.
1
1 −1
0
1 −1
1 −1
1
1 −1 1
2
1
1
1
.
The coefficient matrix of the system is the matrix A =
It is easy to verify the following:
and rank A = 3, rank A2 = rank A3 = 2. This implies k = Ind(A) = 2. We
obtain entries of AD by (11).
A2 =
4
3
1 −1 0
4
3
5
4
3 −4
0
4 −5
3 −4
d2(A3) =(cid:12)(cid:12)(cid:12)(cid:12)
+(cid:12)(cid:12)(cid:12)(cid:12)
−18
2 −2
10 −14
−1
2
18 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
14
10
2 −2 −1
13
10
18
14
,
10 −14
−1
13 −18
10 −14
, A3 =
13 18 (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)
10 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
2 −1
10 14
−14
16
10 10
10 10 (cid:12)(cid:12)(cid:12)(cid:12)
14 10 (cid:12)(cid:12)(cid:12)(cid:12) = 8,
18 13
and so forth. Continuing in the same way, we get AD =
Now we obtain the Drazin inverse solutionbx of the system (18) by the Cramer
rule (16), where
0.5 −0.5
0.5
0.5
1.75 2.5 −2.5 1.75
1.25 1.5 −1.5 1.25
0.5
0.5
0.5 −0.5
.
g = A2y =
Thus we have
d11 =(cid:12)(cid:12)(cid:12)(cid:12)
4
5
4
1
2
3
1
2
2
3 14
3 10
3 −14
0
(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
3 −4
0
4 −5
3 −4
3
1 −1 0
4
3
4 18 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
3 10 (cid:12)(cid:12)(cid:12)(cid:12) = 4,
.
=
·
10 10 (cid:12)(cid:12)(cid:12)(cid:12)
13 18 (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:19) =
10 (cid:12)(cid:12)(cid:12)(cid:12)
18 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
−1 −1 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
(cid:19) = 1,
10 10 (cid:12)(cid:12)(cid:12)(cid:12)
−18 13 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
13 13 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
(cid:19) = 1,
14 10 (cid:12)(cid:12)(cid:12)(cid:12)
−14 10 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
10 10 (cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)
(cid:19) =
10 −14
−1
−1 −1
10
−1 −2
13
10
−1
13
10
2 −1
2 −1
10
10
10 14
10 10
10 10
13 13
18 13
1
2
,
1
2
.
10 10
1
1
8(cid:18)(cid:12)(cid:12)(cid:12)(cid:12)
8(cid:18)(cid:12)(cid:12)(cid:12)(cid:12)
8(cid:18)(cid:12)(cid:12)(cid:12)(cid:12)
8(cid:18)(cid:12)(cid:12)(cid:12)(cid:12)
1
1
bx1 =
bx2 =
bx3 =
bx4 =
Acknowledgment. The author would like to thank Professor Chi-
Kwong Li and the referee for their useful suggestions.
References
[1] R. B. Bapat, K. P. S. Bhaskara, K. Manjunatha Prasad, Generalized
inverses over integral domains, Linear Algebra Appl. 140 (1990), 181 --
196.
[2] A. Ben-Israel, Generalized inverses of marices: a perspective of the work
of Penrose, Math. Proc. Camb. Phil. Soc. 100 (1986), 401-425.
17
[3] A. Ben-Israel, A Cramer rule for least-norm solutions of consistent linear
equations, Linear Algebra Appl. 43 (1982), 223-226.
[4] S. L. Campbell and C. D. Meyer Jr., Generalized Inverse of Linear
Transformations, Pitman, London, (1979).
[5] D. Carl, C. D. Meyer Jr., Limits and the index of a square matrix, SIAM
J. Appl. Math. 26, no. 3 (1974), 506-515.
[6] Y. Chen, A Cramer rule for solution of the general restricted linear
equation, Linear and Multilinear Algebra 34 (1993), 177-186
[7] M.P. Drazin, Pseudoinverses in associative rings and semigroups, Amer.
Math. Monthly 65 (1958), 506-515.
[8] R. Gabriel, Das verallgemeinerte inverse eineer matrix, deren elemente
einem beliebigen Korper angehoren, J.Reine angew math. 234 (1967),
107-122.
[9] R. A. Horn, C. R. Johnson, The Matrix analysis, Cambridge University
Press, (1986).
[10] J. Ji, Explicit expressions of the generalized inverses and condensed
Cramer rules, Linear Algebra Appl. 404 (2005), 183-192.
[11] E. H. Moore, On reciprocal of the general algebraic matrix. Bull. Amer.
Math. Soc. 26 (1920), 394-395.
[12] P. S. Stanimirovic', General determinantal representation of pseudoin-
verses of matrices, Mat. Vesnik 48 (1996), 1-9.
[13] P.S. Stanimirovic', D.S. Djordjevic', Full-rank and determinantal repre-
sentation of the Drazin inverse, Linear Algebra Appl. 311 (2000), 131-
151.
[14] G. Wang, A Cramer rule for minimum-norm(T) least-square(S) solution
of inconsistent linear equations, Linear Algebra Appl. 74 (1986), 213-
218.
[15] G. Wang, A Cramer rule for finding the solution of a class of singular
equations, Linear Algebra Appl. 116 (1989), 27-34.
18
[16] Y. M. Wei, H. B. Wu, Additional results on index splittings for Drazin
inverse solutions of singular linear systems, The Electronic Journal of
Linear Algebra 8 (2001), 83-93.
[17] H. J. Werner, On extension of Cramer's rule for solutions of restricted
linear systems, Linear and Multilinear Algebra 15 (1984), 319-330.
19
|
0907.4780 | 7 | 0907 | 2018-05-01T10:48:24 | Commutative 2-cocycles on Lie algebras | [
"math.RA"
] | On Lie algebras, we study commutative 2-cocycles, i.e., symmetric bilinear forms satisfying the usual cocycle equation. We note their relationship with antiderivations and compute them for some classes of Lie algebras, including finite-dimensional semisimple, current and Kac-Moody algebras. | math.RA | math |
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
ASKAR DZHUMADIL'DAEV AND PASHA ZUSMANOVICH
ABSTRACT. On Lie algebras, we study commutative 2-cocycles, i.e., symmetric bilinear forms satisfying
the usual cocycle equation. We note their relationship with antiderivations and compute them for some
classes of Lie algebras, including finite-dimensional semisimple, current and Kac-Moody algebras.
INTRODUCTION
A commutative 2-cocycle on a Lie algebra L over a field K is a symmetric bilinear form ϕ : L × L → K
satisfying the cocycle equation
(1)
ϕ([x, y], z) + ϕ([z, x], y) + ϕ([y, z], x) = 0
for any x, y, z ∈ L. Commutative 2-cocycles appear at least in two different contexts. First, they appear
in the study of nonassociative algebras satisfying certain skew-symmetric identities. It turns out that all
skew-symmetric identities of degree 3 reduce to a number of identities, among which
and
[a, b]c + [b, c]a + [c, a]b = 0
a[b, c] + b[c, a] + c[a, b] = 0
play a prominent role (here multiplication in an algebra is denoted by juxtaposition, and [a, b] = ab−ba).
Algebras satisfying both these identities are dubbed two-sided Alia algebras in [D3] and [DB]. Note
that the class of two-sided Alia algebras contains both Lie algebras and Novikov algebras (for the latter,
see the end of §2), so it appears to be a natural and interesting class of algebras to study.
Moreover, it is easy to see that an algebra A is two-sided Alia if and only if the associated "minus"
algebra A(−) with multiplication defined by the bracket [ · , · ], is a Lie algebra (in other words, A is
Lie-admissible), and multiplication in A can be written as
(2)
ab = [a, b] + ϕ(a, b), †
where ϕ is an A(−)-valued commutative 2-cocycle on A(−).
Note, however, that also any commutative (nonassociative) algebra is two-sided Alia, so the question
of description of simple algebras in this class does not make much sense without imposing additional
conditions. One such natural condition is, in a sense, opposite to the condition of commutativity of A
-- namely, that the Lie algebra A(−) is simple. In this way we arrive to the problem of description of
commutative 2-cocycles on simple Lie algebras.
Second, commutative 2-cocycles appear naturally in the description of the second cohomology of
current Lie algebras ([Zu3, Theorem 1]). All this, as well as a sheer curiosity in what happens with
the usual second Lie algebra cohomology when we replace the condition of the skew-symmetricity of
cochains by its opposite -- symmetricity -- makes them worth to study.
It is worth to note that this situation is similar (and somewhat dual) to the question which goes back to
A.A. Albert and was a subject of an intensive study later, namely, determination of Lie-admissible third
power-associative algebras A whose "minus" algebra A(−) belongs to some distinguished class of Lie
algebras, such as finite-dimensional simple Lie algebras, or Kac-Moody algebras. It is easy to see that
Date: last minor revision September 9, 2017.
J. Algebra 324 (2010), N4, 732 -- 748; arXiv:0907.4780.
† Added May 5, 2017: The correct formula is: ab = 1
2 [a, b] + ϕ(a, b).
1
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
2
in terms of decomposition (2), for a given Lie algebra L = A(−) the third power-associativity implies the
condition
[ϕ(x, y), z] + [ϕ(z, x), y] + [ϕ(y, z), x] = 0
for any x, y, z ∈ L (see, for example, [Be, p. 39]). The latter condition, together with (1), could be viewed
as two parts of the usual 2-cocycle equation on a Lie algebra with coefficients in the adjoint module.
Of course, the very definition of commutative 2-cocycles begs for a proper generalization -- to define
higher cohomology groups such that commutative 2-cocycles constitute cohomology of low degree.
However, all "naive" attempts to construct such higher cohomology seemingly fail. Maybe it could be
developed in the framework of operadic cohomology of two-sided Alia algebras.
The contents of the paper are as follows. In §1 we exhibit an exact sequence relating commutative
2-cocycles and so-called antiderivations, similar to the well known relationship between the second
cohomology and derivations with values in the coadjoint module. In §2 we prove that in the class of
simple Lie algebras the property to possess nonzero commutative 2-cocycles is equivalent to the property
to satisfy the standard identity of degree 5. In the class of finite-dimensional simple Lie algebras, this
provides yet another characterization of sl(2) and the modular Zassenhaus algebra. We also compute
the space of commutative 2-cocycles on the Zassenhaus algebra, and note how this may be utilized in
classification of simple Novikov algebras. §3 is devoted to some speculations about the characteristic 3
case. In §4 we establish formula expressing the space of commutative 2-cocycles on current Lie algebras
in terms of its tensor factors, similar to the known formula for the second cohomology. In the next two
sections that formula is applied to compute the space of commutative 2-cocycles for various classes of
Lie algebras: Kac-Moody algebras in §5, and modular semisimple Lie algebras in §6.
Not surprisingly, the study of commutative 2-cocycles turns out to be similar in some aspects to the
study of the second cohomology of Lie algebras, with some results having direct counterparts. There
are, however, also significant differences: for example, in the class of simple Lie algebras and close to
them, non-trivial commutative 2-cocycles turns out to be a very rare phenomenon.
Throughout the paper, all algebras and vector spaces are defined over a ground field K of characteristic
6= 2, 3, except in §3.
NOTATION AND CONVENTIONS
For a given Lie algebra L, the space of all commutative 2-cocycles on it will be denoted as Z2
comm(L).
Occasionally, we will consider the M-valued commutative 2-cocycles, that is, bilinear forms L × L → M,
where M is an L-module (or, rather, just a vector space), satisfying the cocycle equation (1).
It is
immediate that if either L or M is finite-dimensional, then the space of all such cocycles is isomorphic
to Z2
comm(L) ⊗ M.
Recall that a Lie algebra L is called perfect if [L, L] = L. Obviously, a symmetric bilinear form on L
that vanishes whenever one of the arguments belongs to [L, L], is a commutative 2-cocycle. Such cocy-
cles will be called trivial and they exist on any non-perfect Lie algebra. The space of trivial commutative
2-cocycles is isomorphic to S2(L/[L, L])∗.
The rest of our notation and definitions is mostly standard.
H n(L, M) and Zn(L, M) denote the usual Chevalley -- Eilenberg nth order cohomology and the space of
nth order cocycles, respectively, of a Lie algebra L with coefficients in a module M.
When being considered as an L-module, the ground field K is always understood as the one-dimensi-
onal trivial module.
HC1(A) denotes the first order cyclic cohomology of an associative commutative algebra A. Recall
that it is nothing but the space of all skew symmetric bilinear forms α : A × A → K such that
α(ab, c) + α(ca, b) + α(bc, a) = 0
for any a, b, c ∈ A. Note an obvious but useful fact: if A contains a unit 1, then α(1, A) = 0 for any
α ∈ HC1(A).
Der(A) denotes the Lie algebra of derivations of A.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
3
For an associative algebra A, we may consider associated Lie [ · , · ] and Jordan ◦ products on A:
[a, b] = ab − ba and a ◦ b = ab + ba for a, b ∈ A. The vector space A with the bracket [ · , · ] forms a Lie
algebra which is denoted as A(−).
For a vector space V , S2(V ) and ∧2(V ) denotes the symmetric and skew-symmetric product respec-
tively (so, the space of symmetric, respectively skew-symmetric bilinear forms on V is isomorphic to
S2(V )∗, respectively to ∧2(V )∗).
Other nonstandard notions (antiderivations, cyclic forms) are introduced in §1.
1. AN EXACT SEQUENCE CONNECTING COMMUTATIVE 2-COCYCLES AND ANTIDERIVATIONS
A relationship between the second cohomology with coefficients in the trivial module H 2(L, K), and
the first cohomology with coefficients in the coadjoint module H 1(L, L∗) was noted many times in
slightly different forms in the literature, and goes back to the classical works of Koszul and Hochschild --
Serre. Namely, there is an exact sequence
(3)
0 → H 2(L, K) u→ H 1(L, L∗) v→ B(L) w→ H 3(L, K).
Here B(L) denotes the space of symmetric bilinear invariant forms on L. The maps are defined as
follows: for the representative ϕ ∈ Z2(L, K) of a given cohomology class, we have to take the class of
u(ϕ), the latter being given by
(4)
(u(ϕ)(x))(y) = ϕ(x, y)
for any x, y ∈ L, v is sending the class of a given cocycle d ∈ Z1(L, L∗) to the bilinear form v(d) : L × L →
K defined by the formula
v(d)(x, y) = d(x)(y) + d(y)(x),
and w is sending a given symmetric bilinear invariant form ϕ : L × L → K to the class of the cocycle
ω ∈ Z3(L, K) defined by
(5)
ω(x, y, z) = ϕ([x, y], z)
(see, for example, [D2], where a certain long exact sequence is obtained, of which this one is the
beginning, and references therein for many earlier particular variations; this exact sequence was also
established in [NW, Proposition 7.2] with two additional terms on the right).
The following is a "commutative" version of this exact sequence, connecting commutative 2-cocycles
and antiderivations.
Definition. An antiderivation of a Lie algebra L to an L-module M is a linear map d : L → M such that
d([x, y]) = y • d(x) − x • d(y)
for any x, y ∈ L, where • denotes the L-action on M. The set of all such maps will be denoted as
ADer(L, M).
When M = L, the adjoint module, we get the notion of an antiderivation of a Lie algebra (to itself)
with the defining condition
d([x, y]) = −[d(x), y] − [x, d(y)],
what was the subject of study in a number of papers, including [F].
The third ingredient in our exact sequence, a counterpart of symmetric bilinear invariant forms, is
defined as follows.
Definition. A bilinear form ϕ : L × L → K is said to be cyclic if
(6)
ϕ([x, y], z) = ϕ([z, x], y)
for any x, y, z ∈ L. The space of all cyclic skew-symmetric bilinear forms on a Lie algebra L will be
denoted as C(L).
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
4
Proposition 1.1. For any Lie algebra L, there is an exact sequence
0 → Z2
comm(L) u→ ADer(L, L∗) v→ C(L)
where the map u is defined by formula (4), and v is sending a given antiderivation d ∈ ADer(L, L∗) to
the bilinear form v(d) : L × L → K defined by the formula
Proof. If ϕ ∈ Z2
comm(L), then the cocycle equation yields
v(d)(x, y) = d(x)(y) − d(y)(x).
(u(ϕ)([x, y]))(z) = ϕ([x, y], z) = −ϕ(x, [y, z]) + ϕ(y, [x, z]) = (y • u(ϕ)(x))(z) − (x • u(ϕ)(y))(z)
for any x, y, z ∈ L, where • denotes the standard L-action on L∗, i.e. u(ϕ) ∈ ADer(L, L∗).
If d ∈ ADer(L, L∗), then v(d) is obviously skew-symmetric, and
v(d)([x, y], z) = d([x, y])(z) − d(z)([x, y])
= −d(x)([y, z]) + d(y)([x, z]) − d(z)([x, y]) = d([z, x])(y) − d(y)([z, x])
= v(d)([z, x], y)
for any x, y, z ∈ L, i.e. v(d) ∈ C(L).
Now let us check exactness. Obviously, u is injective, so the sequence is exact at Z2
comm(L). Next,
Im u consists of all antiderivations d ∈ ADer(L, L∗) such that the bilinear form (x, y) 7→ d(x)(y) satisfies
the cocycle equation, what is equivalent to d being antiderivation, and is symmetric, what is equivalent
to d(x)(y) = d(y)(x) for any x, y ∈ L. But the latter condition is equivalent to d belonging to Ker v, hence
Im u = Ker v and the sequence is exact at ADer(L, L∗).
(cid:3)
Unfortunately, we do not know how to extend this exact sequence further. The map defined by the
formula (5) maps C(L) to the space of skew-symmetric trilinear forms L × L × L → K, but the resulting
images are neither Chevalley -- Eilenberg (or Leibniz) cocycles, nor do they satisfy any other natural
condition. Obviously, this is related to the difficulty to define higher analogs of commutative 2-cocycles.
Another difficulty is related to the fact that C(L) turns out to be not a very fascinating invariant of a
Lie algebra: it vanishes in the most interesting cases.
Lemma 1.2 (Referee). If L is a perfect Lie algebra, then C(L) = 0.
Proof. For any ϕ ∈ C(L), and any x, y, z, t ∈ L, we have
ϕ([[x, y], z], t)
(Jacobi identity) = −ϕ([[z, x], y], t) − ϕ([[y, z], x], t)
(cyclicity of ϕ) = −ϕ([y, t], [z, x]) − ϕ([x, t], [y, z])
(skew-symmetry of ϕ) = −ϕ([x, z], [y, t]) + ϕ([y, z], [x, t])
(cyclicity of ϕ) = −ϕ([[y, t], x], z) + ϕ([[x, t], y], z)
(Jacobi identity) = ϕ([[x, y], t], z)
(cyclicity of ϕ) = ϕ([z, [x, y]], t)
= −ϕ([[x, y], z], t).
Therefore, ϕ([[L, L], L], L) = 0, and the asserted statement follows.
(cid:3)
The exact sequence (3) was utilized many times in the literature to evaluate H 2(L, K) basing on
H 1(L, L∗) (see, for example, references in [D2]). We will utilize Proposition 1.1 in a similar way. For
this, we shall need to establish some facts about antiderivations of Lie algebras, what is done in the next
section.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
5
2. SIMPLE LIE ALGEBRAS
In [F], Filippov obtained many results about Lie algebras possessing a nonzero antiderivation. A slight
modification of his reasonings allows to extend some of these results to antiderivations with values in a
Lie algebra module.
Recall the standard identity of degree 5 in the class of Lie algebras:
(7)
(−1)σ[[[[y, xσ(1)], xσ(2)], xσ(3)], xσ(4)] = 0,
∑
σ∈S4
where the summation is performed over all elements of the symmetric group S4. The word at the left
side of (7) will be denoted as s4(x1, x2, x3, x4, y). For a Lie algebra L, let s4(L) denote the linear span
of values of this word for any x1, x2, x3, x4, y ∈ L. By the general result about verbal ideals (see, for
example, [AS, Chapter 14], Theorem 2.8 and remark after it), s4(L) is an ideal of L.
Lemma 2.1. If a Lie algebra possesses nonzero commutative 2-cocycles, then it has a nonzero homo-
morphic image satisfying the standard identity of degree 5.
Proof. If a Lie algebra L possesses nonzero commutative 2-cocycles, then by Proposition 1.1, L pos-
sesses a nonzero antiderivation to its coadjoint module. By [Zu4, Lemma 4.4] (which is just a slight
generalization of [F, Theorem 4]), s4(L) lies in the kernel of this antiderivation, and hence is a proper
ideal in L. The required homomorphic image is L/s4(L).
(cid:3)
Lemma 2.2. A Lie algebra with a subalgebra of codimension 1 which is not an ideal, possesses nonzero
commutative 2-cocycles.
Proof. Let S be a subalgebra of codimension 1 in a Lie algebra L. Pick x ∈ L\S and let f : S → K be
a linear map such that [x, y] ∈ f (y)x + S for any y ∈ S. The Jacobi identity implies f ([S, S]) = 0. Since
S is not an ideal in L, f is nonzero. It is easy to see that any bilinear map ϕ : L × L → K satisfying the
conditions
ϕ(S, S) = 0, ϕ(x, y) = ϕ(y, x) = f (y)
for y ∈ S, is a commutative 2-cocycle (in fact, the cocycle equation in that case is equivalent to the Jacobi
identity; ϕ(x, x) can take any value).
(cid:3)
Theorem 2.3. A simple Lie algebra possesses nonzero commutative 2-cocycles if and only if it satisfies
the standard identity of degree 5.
Proof. The "only if" part follows from Lemma 2.1.
The "if" part. Suppose a simple Lie algebra L satisfies the identity of degree 5. By Razmyslov's
characterization of such algebras ([R, Proposition 46.1]), there exists a field extension F of the centroid
C of L such that the Lie F-algebra L ⊗C F contains a subalgebra of codimension 1, and, by Lemma
2.2, possesses nonzero commutative 2-cocycles. As the space of commutative 2-cocycles is obviously
preserved under the ground field extension, the Lie C-algebra L ⊗K C possesses nonzero commutative
2-cocycles. Each such cocycle, being restricted to L, gives rise to a nonzero bilinear K-map L × L → C,
which is a commutative C-valued 2-cocycle on L.
(cid:3)
Using other characterizations of simple Lie algebras satisfying the standard identity of degree 5, it
is possible to give an alternative proof of the "if" part of Theorem 2.3 in the case where the ground
field has positive characteristic. Namely, according to [R, Theorem 46.2], each simple Lie algebra of
dimension > 3 over the field of positive characteristic, satisfying the standard identity of degree 5, can be
represented as a derivation algebra A∂ = {a∂ a ∈ A} of an associative commutative algebra A with unit,
generated as an A-module by a single derivation ∂ ∈ Der(A). In addition, A does not have ∂-invariant
ideals. For such algebras, we have
Lemma 2.4. A∗ is embedded into Z2
comm(A∂).
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
6
Proof. Consider the map F : A → A∂ defined by a 7→ a∂ for a ∈ A. The kernel of this map is a ∂-invariant
ideal of A, and hence this map is injective.
For any χ ∈ A∗, define the map Φ(χ) : A∂ × A∂ → K as Φ(χ)(a∂, b∂) = χ(ab) for a, b ∈ A. This
map is well defined due to injectivity of F and is a commutative 2-cocycle:
Φ(χ)([a∂, b∂], c∂) + Φ(χ)([c∂, a∂], b∂) + Φ(χ)([b∂, c∂], a∂)
= Φ(χ)((a∂(b) − b∂(a))∂, c∂)
+ Φ(χ)((c∂(a) − a∂(c))∂, b∂)
+ Φ(χ)((b∂(c) − c∂(b))∂, a∂)
= χ(ac∂(b)) − χ(bc∂(a)) + χ(bc∂(a)) − χ(ab∂(c)) + χ(ab∂(c)) − χ(ac∂(b)) = 0
for any a, b, c ∈ A. It is obvious that Φ is injective.
(cid:3)
Corollary 2.5. A finite-dimensional central simple Lie algebra possesses nonzero commutative 2-co-
cycles if and only if it is isomorphic either to a form of sl(2) or, in the case where the characteristic of
the ground field is positive, to the Zassenhaus algebra W1(n).
Proof. Obviously, we may pass to the algebraic closure of the ground field. Finite-dimensional simple
Lie algebras over an algebraically closed field satisfying the standard identity of degree 5, are precisely
sl(2) and the Zassenhaus algebra. This well known fact can be derived in several ways, perhaps the
easiest one is to invoke once again [R, Proposition 46.1] to establish that such Lie algebras have a sub-
algebra of codimension 1, and then refer to the known classification of such algebras (see, for example,
[E] and references therein).
(cid:3)
This generalizes [DB, Theorem 1.1], where the same result is proved for Lie algebras of classical type
by performing computations with the corresponding root system. Another proof for such Lie algebras
could be derived by combining results of [L] and [Zu3]. In a sense, root space computations in [DB] are
equivalent to the appropriate part of computations in [L].
Note that the same reasoning (Theorem 2.3 coupled with Razmyslov's results) shows that among
infinite-dimensional Lie algebras of Cartan type, only the Witt algebra possesses nonzero commuta-
tive 2-cocycles. The latters are described in [D3, Theorem 6.7]. Another important class of infinite-
dimensional Lie algebras -- Kac-Moody algebras -- is treated in §5.
The three-dimensional algebra sl(2) and the Zassenhaus algebra are characterized among simple
finite-dimensional Lie algebras in various interesting ways: these are algebras having a subalgebra of
codimension 1, algebras having a solvable maximal subalgebra (see [S, Corollary 9.1.0]), algebras with
certain properties of the lattice of subalgebras (see, for example, [BTW] and references therein), algebras
having non-trivial δ-derivations ([Zu4, §2]), etc. Corollary 2.5 adds another characterization to this list.
Of course, it is interesting to compute exactly the space of commutative 2-cocycles on these algebras.
comm(sl(2)) forms a 5-dimensional hyperplane in the 6-dimensional
As noted in [D3, Theorem 6.5], Z2
space of symmetric bilinear maps ϕ : sl(2) × sl(2) → K, determined by the equation
in the sl(2)-basis {e−, e+, h} with multiplication table [h, e±] = ±e±, [e−, e+] = h.
ϕ(h, h) = 2ϕ(e−, e+)
The famous Zassenhaus algebra W1(n) can be realized in two different ways. One realization is
the algebra of derivations of the divided powers algebra O1(n) = {xi 0 ≤ i ≤ pn−1}, where p is the
characteristic of the ground field, with multiplication given by
xix j =(cid:18)i + j
j (cid:19)xi+ j,
of the form O1(n)∂, where the derivation ∂ ∈ Der(O1(n)) acts as follows: ∂(xi) = xi−1. Another
realization is the algebra with the basis {eα α ∈ G}, where G is an additive subgroup of order pn of the
ground field K, with multiplication
(8)
[eα, eβ] = (β − α)eα+β
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
7
for α,β ∈ G. We formulate the result in terms of the first realization, but perform actual computations
with the second one.
Proposition 2.6. Z2
comm(W1(n)) ≃ O1(n)∗, each cocycle being of the form
(a∂, b∂) 7→ f (ab)
where a, b ∈ O1(n), for some f ∈ O1(n)∗.
Proof. The proof consists of straightforward computations reminiscent both the computation of the
second cohomology of W1(n) in [Bl, §5] and computation of the space of commutative 2-cocycles on
the infinite-dimensional Witt algebra in [D3, Theorem 6.7].
Let ϕ ∈ Z2
comm(W1(n)). Writing the cocycle equation in terms of the basis with multiplication (8) for
e0, eα, eβ, where α,β ∈ G such that α 6= β, we get
(9)
ϕ(eα, eβ) = ϕ(e0, eα+β).
Now, taking this into account, and writing the cocycle equation for eα, eβ, eα+β, where α,β ∈ G such
that α,β 6= 0, we get
ϕ(eα+β, eα+β) = ϕ(e0, e2α+2β).
Consequently, (9) holds for any α,β ∈ G. Conversely, each symmetric bilinear map satisfying the
condition (9) is easily seen to satisfy the cocycle equation. Each such map can be decomposed into the
sum of the maps of the form
(eα, eβ) 7→(1, α+ β = γ
0, otherwise
for each γ ∈ G. Thus we get G = pn linearly independent cocycles, so the whole space Z2
comm(W1(n)) is
pn-dimensional. Now, switching to the first realization as derivation algebra of O1(n), applying Lemma
2.4 and comparing dimensions, we see that the embedding of Lemma 2.4 is an isomorphism in this
particular case.
(cid:3)
Note that the results of this section allow to provide a somewhat alternative way for classification of
finite-dimensional simple Novikov algebras. Recall that an algebra is called Novikov if it satisfies the
identities
and
x(yz) − (xy)z = y(xz) − (yx)z
(xy)z = (xz)y.
Novikov algebras are ubiquitous in various branches of mathematics and physics (see [Bu], [O] and [Ze]
with a transitive closure of references therein)†.
A well known important fact is that each Novikov algebra A is Lie-admissible, i.e., A(−) is a Lie
algebra.
In [Ze], Zelmanov proved that finite-dimensional simple Novikov algebras over a field of
characteristic zero are 1-dimensional (i.e., coincide with the ground field), and in [O], Osborn proved
that if A is a finite-dimensional simple Novikov algebra over a field of positive characteristic, then A(−) is
either 1-dimensional, or isomorphic to the Zassenhaus algebra. The latter was the key result in obtaining
later a complete classification of simple Novikov algebras.
Another important observation -- which is a matter of simple calculations (see, for example, [Bu,
Lemma 2.3]) -- is that Novikov algebras are two-sided Alia. Further, it follows from the proof of [O,
Theorem 3.5], that if A is a finite-dimensional simple Novikov algebra, then the Lie algebra A(−) is also
simple. Thus, as noted in the Introduction, the multiplication in a finite-dimensional simple Novikov
algebra A could be written in the form (2), where [ · , · ] defines a simple Lie algebra structure on A, and
† Actually, what we define here are left Novikov algebras. Right Novikov algebras are defined by the opposite identities,
interchanging left and right multiplications. Often in the literature, left Novikov and right Novikov are confused, even if
authors make an explicit attempt not to do so (for example, [Ze] treats right Novikov algebras, while [O] and [Bu] -- left
Novikov ones). However, almost every result about left Novikov algebras is easily transformed into one about right Novikov
algebras, and vice versa.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
8
ϕ defines a commutative 2-cocycle on that Lie algebra. Note further that ϕ is a nonzero cocycle, as
otherwise A is a Lie algebra, what quickly leads to contradiction.
Then, instead of appealing to some structural results about graded and filtered Lie algebras, like in
[O], we may invoke Corollary 2.5 to see that A(−) is isomorphic either to 1-dimensional abelian algebra,
or to sl(2), or to the Zassenhaus algebra. Straightforward calculations based on [D3, Theorem 6.5]
exclude the case of sl(2), and in this way we get the main results of [Ze] and [O]. Proposition 2.6 may
be utilized in the subsequent classification.
3. CHARACTERISTIC 3
The case of characteristic 3 differs drastically: as noted in [D3, §6.2], in this case any symmetric
bilinear invariant form on a Lie algebra is a commutative 2-cocycle. Let us rephrase this observation in
another trivial, yet not without interest, form.
To start with, let us make a useful observation valid in any characteristic. For a Lie algebra L, consider
the standard Der(L)-action on the space of bilinear forms on L: for ϕ being such a form, and D ∈ Der(L),
(10)
(D • ϕ)(x, y) = −ϕ(D(x), y) − ϕ(x, D(y)),
x, y ∈ L. It is well known (and easy to verify) that the space of symmetric bilinear invariant forms on L
is closed under this action. It turns out that the same is true for the space of commutative 2-cocycles:
Lemma 3.1. For any Lie algebra L, Z2
comm(L) is closed under the action (10).
Proof. This follows from the following equality, valid for any bilinear form ϕ : L × L → K, any D ∈
Der(L), and any x, y, z ∈ L:
d(D • ϕ)(x, y, z) = − dϕ(x, y, D(z)) − dϕ(z, x, D(y)) − dϕ(y, z, D(x)).
where dϕ(x, y, z) denotes the left-hand side of equality (1).
(cid:3)
In particular, Z2
comm(L) is closed under the action of L (via inner derivations). This is, essentially, the
same observation which is used in deriving a very useful fact about triviality of the Lie algebra action
on its cohomology. Note, however, that, unlike for cohomology, in the case of commutative 2-cocycles
we do not have coboundaries, so we cannot normalize the cocycle appropriately to derive the triviality
of this action. Moreover, for invariants of this action we have the following obvious dichotomy:
comm(L)L =(B(L),
Z2
{ϕ : L × L → K ϕ is symmetric, ϕ([L, L], L) = 0},
p = 3
p 6= 3.
Still, we can make use of Lemma 3.1 in the same way as in cohomological considerations: when
T is a torus in a Lie algebra L such that L decomposes into the direct sum of the root spaces with
respect to the action of T (what always takes places if L is finite-dimensional and the ground field is
algebraically closed), then Z2
comm(L) decomposes into the direct sum of root spaces with respect to the
induced T -action.
Naturally, in order to compute the space of commutative 2-cocycles on any class of Lie algebras of
characteristic 3, it would be beneficial to elucidate first what the space of symmetric bilinear invariant
forms on these algebras looks like.
Conjecture 3.2. The space of symmetric bilinear invariant forms on any central Lie algebra of classical
type over a field of arbitrary characteristic is 1-dimensional.
Note that in small characteristics (including characteristic 3) not all Lie algebras obtained via the usual
Chevalley basis construction, are simple (see, for example, [S, §4.4]). In such cases, under Lie algebras
of classical type we mean both these non-simple Lie algebras, and their central simple quotients.
Of course, for p = 0 and p > 5 this conjecture trivially follows from the well known statement about
the Killing form. However, in small characteristics the Killing form, and, more general, any trace form,
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
9
may vanish (see [GP] and references therein), so this conjecture, perhaps, comes as a bit of surprise. It
is, however, supported by computer calculations†.
Conjecture 3.3. The space of commutative 2-cocycles on a finite-dimensional Lie algebra of classical
type different from sl(2), over the field of characteristic 3, coincides with the space of symmetric bilinear
invariant forms (and, hence, by Conjecture 3.2, is 1-dimensional).
Note that for Lie algebras not of classical type, it is no longer true that every commutative 2-cocycle
is a symmetric bilinear invariant form: for example, for the Zassenhaus algebra W1(n), Proposition 2.6
is valid also in characteristic 3 (with, essentially, the same proof). On the other hand, since W1(n) is
simple, the space of symmetric bilinear invariant forms on it is at most 1-dimensional (it is, in fact,
1-dimensional in characteristic 3, as shown in [D1, §2, Corollary] or [SF, §4.6, Theorem 6.3]).
A straightforward way to prove Conjectures 3.2 and 3.3 seems to employ the strategy of [DB, Theo-
rem 1.1]: first to prove the statements for algebras of rank 2, and then derive the general case.
Lie algebras of classical type of rank 2 are A2, B2 and G2. For these algebras, one may verify on a
computer that both the spaces of commutative 2-cocycles (for p = 3) and of symmetric bilinear invariant
forms (for any p), are 1-dimensional. The general case should follow then from considerations of 2-
sections in a Cartan decomposition, with the help of Lemma 3.1.
However, there are lot of subtleties when dealing with structure constants of classical Lie algebras
in small characteristics, as demonstrated by a noticeable amount of errors in works devoted to such
algebras. We postpone this laborious task to the future.
In this section we consider the current Lie algebras, i.e. Lie algebras of the form L ⊗ A where L is a
Lie algebra and A is an associative commutative algebra, with multiplication defined by
4. CURRENT LIE ALGEBRAS
for x, y ∈ L, a, b ∈ A.
[x ⊗ a, y ⊗ b] = [x, y] ⊗ ab
Theorem 4.1. Let L be a Lie algebra, A an associative commutative algebra, and at least one of L,
A is finite-dimensional. Then each commutative 2-cocycle on L ⊗ A can be represented as a sum of
decomposable cocycles ϕ ⊗ α, ϕ : L × L → K, α : A × A → K of one of the 8 following types:
(i) ϕ([x, y], z) + ϕ([z, x], y) + ϕ([y, z], x) = 0 and α(ab, c) = α(ca, b),
(ii) ϕ([x, y], z) = ϕ([z, x], y) and α(ab, c) + α(ca, b) + α(bc, a) = 0,
(iii) ϕ([L, L], L) = 0,
(iv) α(AA, A) = 0,
and each of these four types splits into two subtypes: with both ϕ and α symmetric, and with both ϕ
and α skew-symmetric.
Proof. The proof goes almost verbatim to the proof of Theorem 1 in [Zu3].
(cid:3)
Corollary 4.2. Suppose all assumptions of Theorem 4.1 hold, and, additionally, A contains a unit. Then
Z2
comm(L ⊗ A) ≃ Z2
comm(L) ⊗ A∗
⊕C(L) ⊗ HC1(A)
⊕ (S2(L/[L, L]))∗ ⊗ Ker(S2(A) → A)∗
⊕ (∧2(L/[L, L]))∗ ⊗ {ab ∧ c + ca ∧ b + bc ∧ a a, b, c ∈ A}∗,
where the map S2(A) → A is induced by multiplication in A.
† A simple-minded GAP code for calculation of the spaces of commutative 2-cocycles and of symmetric bilinear invariant
forms on a Lie algebra, is available at http://justpasha.org/math/comm2.gap . [Added January 4, 2016: currently
available as an ancillary file accompanying the arXiv version of this paper]. The code works by writing the corresponding
conditions in terms of a certain basis of a Lie algebra, and solving the arising linear homogeneous system.
Proof. Consider the cases of Theorem 4.1.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
10
Case (i). Substituting c = 1 in the condition α(ab, c) = α(ca, b), we get α(a, b) = β(ab) for a certain
comm(L) and
linear form β : A → K. Consequently, both ϕ and α are necessarily symmetric, hence ϕ ∈ Z2
the linear span of cocycles of this type is isomorphic to Z2
comm(L) ⊗ A∗.
Case (ii). Similarly, substituting b = c = 1 in the condition
α(ab, c) + α(ca, b) + α(bc, a) = 0,
and assuming that α is symmetric, we get α(a, 1) = 0. Now substituting just c = 1 in the same condition,
we get that α vanishes. Consequently, for cocycles of this type both ϕ and α are necessarily skew-
symmetric, hence ϕ ∈ C(L) and α ∈ HC1(A).
Case (iv). If A contains a unit, then AA = A, so cocycles of this type vanish.
Singling out from the linear span of the remaining cocycles of type (iii) the direct sum complement to
the linear span of cocycles of type (i) and (ii), and rearranging it in an obvious way, we get the desired
isomorphism.
(cid:3)
Evidently, the last two direct summands at the right-hand side of the isomorphism of Corollary 4.2
constitute trivial cocycles.
It is possible to extend Theorem 4.1 and Corollary 4.2 to various generalizations of current Lie alge-
bras, such as twisted algebras, extended affine Lie algebras, toroidal Lie algebras, Lie algebras graded
by root systems, etc. Some of computations could be quite cumbersome, but all of them seem to be
amenable to the technique used in [Zu3].
It is possible also to consider a sort of noncommutative version of Corollary 4.2, namely, commuta-
tive 2-cocycles on the Lie algebra sl(n, A) for an associative (and not necessarily commutative) algebra
A with unit. It is possible to show that any homomorphic image of such an algebra is closely related to
the algebra of the form sl(n, B), where B is a homomorphic image of A. If sl(n, A) possesses nonzero
2-commutative cocycles, then by Lemma 2.1, at least one of these homomorphic images satisfies the
standard identity of degree 5. As sl(n) is a subalgebra of sl(n, B), this implies that there are no com-
mutative 2-cocycles if n > 2. The algebra sl(2, A), on the contrary, possesses nonzero 2-commutative
cocycles, but in general there seems no nice expression for them in terms of A. As the final answer turns
out not to be very interesting in either case, we are not going into details.
5. KAC-MOODY ALGEBRAS
If we want to apply the results of the preceding section to Kac-Moody algebras, we should deal not
with the current Lie algebras and their twisted analogs, but their extensions by means of central elements
and derivations. To this end, we make the following elementary observations.
Lemma 5.1. Let L be a Lie algebra and I an ideal of L. Then Z2
comm(L/I) is embedded into Z2
comm(L).
Proof. There is an obvious bijection between Z2
ϕ ∈ Z2
comm(L) such that ϕ(L, I) = 0.
comm(L/I) and the set of cocycles
(cid:3)
Of course, the similar embedding exists for ordinary (skew-symmetric) cocycles, but this embedding
is usually not preserved on the level of cohomology. For example, free Lie algebras possess, in a sense,
"the most" of 2-cocycles, accumulating all cocycles from their homomorphic images, but in the skew-
symmetric case all cocycles are killed off by coboundaries. This has no parallel in the commutative case
due to absence of "commutative 2-coboundaries".
The following is a very particular complement, in a sense, to Lemma 5.1.
Lemma 5.2. Let L be a Lie algebra and I a perfect ideal of codimension 1 of L. Then Z2
embedded into Z2
comm(I) ⊕ K, the second direct summand being represented by a trivial cocycle.
comm(L) is
Proof. Write L = I ⊕ Kx for some x ∈ L\I. Let ϕ ∈ Z2
comm(L). The cocycle equation on L is equivalent
to the cocycle equation on I, plus the cocycle equation for a, b ∈ I and x, the latter could be written in
the form
ϕ([a, b], x) = ϕ([a, x], b) − ϕ([b, x], a).
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
11
As [I, I] = I, this formula provides a well defined extension of ϕ from I × I to I × Kx.
There are no further restrictions on the value of ϕ(x, x), so defining a bilinear form ϕ on L by ϕ(x, x) =
1, ϕ(I, I) = ϕ(I, x) = 0, we get a trivial cocycle which corresponds to the second direct summand. (cid:3)
current Lie algebras (see [K, Chapters 7 and 8]). Let g =L j∈Zn
We use the realization of affine Kac-Moody Lie algebras as extensions of non-twisted and twisted
g j be a simple finite-dimensional Zn-
graded Lie algebra (it is possible that g = g0, i.e., n = 1 and the grading is trivial, what corresponds to
the non-twisted case). Consider the following subalgebra of the current Lie algebra g ⊗ K[t, t−1]:
L(g, n) =Mi∈Z
gi(mod n) ⊗ t i.
represented in the form
This algebra has a non-split perfect central extensioneL(g, n). Each affine Kac-Moody algebra can be
for a suitable g, where multiplication between elements ofeL(g, n) and t d
We will consider first the case of non-twisted affine Kac-Moody algebras in a bit more general situ-
ation. Let L be a Lie algebra, h·, ·i a nonzero symmetric bilinear invariant form on L, A a commutative
associative algebra with unit, D a Lie subalgebra of Der(A), and ξ a nonzero D-invariant element of
HC1(A). Consider a Lie algebra defined as the vector space (L ⊗ A) ⊕ Kz ⊕ D with the following multi-
plication:
bL(g, n) =eL(g, n) ⊕ Kt
latter as derivation on the algebra of Laurent polynomials K[t, t−1].
dt is defined by the action of the
d
dt
[x ⊗ a, y ⊗ b] = [x, y] ⊗ ab + hx, yiξ(a, b)z
[x ⊗ a, d] = x ⊗ d(a)
for x, y ∈ L, a, b ∈ A, d ∈ D, and z belongs to the center. Note that the semidirect sum (L ⊗ A) A D
is a quotient by the 1-dimensional central ideal Kz, and (L ⊗ A) ⊕ Kz is a subalgebra, the latter being
central extension of the current Lie algebra L ⊗ A (for generalities about central extensions of current
Lie algebras, see [Zu2, Zu3, NW]).
Specializing this construction to the case where K is an algebraically closed field of characteristic
zero, L = g, a simple finite-dimensional Lie algebra, h·, ·i is the Killing form on g, A = K[t, t−1], D =
Kt d
dt ) for f , g ∈ K[t, t,−1 ], we get non-twisted affine Kac-Moody algebras.
dt , and ξ( f , g) = Res(g d f
Lemma 5.3. Let L be perfect, and one of L, A is finite-dimensional. Then†
Z2
comm((L ⊗ A) A D)
comm(L) ⊗ {χ ∈ A∗ χ(d(a)b − ad(b)) = 0 for any a, b ∈ A, d ∈ D}
≃ Z2
⊕ C(L) ⊗ {β ∈ HC1(A) β(d(a), b) − β(a, d(b)) = 0 for any a, b ∈ A, d ∈ D}
⊕ Z2
comm(D).
Proof. Let Φ ∈ Z2
on L ⊗ A. By Corollary 4.2, there are ϕ ∈ Z2
comm((L ⊗ A) A D). A restriction of Φ to (L ⊗ A) × (L ⊗ A) is a commutative 2-cocycle
comm(L), χ ∈ A∗, α ∈ C(L) and β ∈ HC1(A) such that
(11)
Φ(x ⊗ a, y ⊗ b) = ϕ(x, y)χ(ab) + α(x, y)β(a, b)
for any x, y ∈ L, a, b ∈ A. Writing the cocycle equation for x ⊗ a, y ⊗ b, d, we get
(12)
Φ([x, y] ⊗ ab, d) = ϕ(x, y)χ(d(a)b − ad(b)) + α(x, y)(β(d(a), b) − β(a, d(b)))
† Added May 24, 2011: By Lemma 1.2, the term containing C(L) is redundant here. The proofs of Lemmata 5.3 and 5.4
can be simplified accordingly.
for any x, y ∈ L, a, b ∈ A, and d ∈ D. Substituting here a = 1 and b = 1, we get respectively:
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
12
Φ([x, y] ⊗ b, d) = −ϕ(x, y)χ(d(b))
Φ([x, y] ⊗ a, d) = ϕ(x, y)χ(d(a)),
what implies Φ(L ⊗ A, D) = 0. Hence the right-hand side of (12) vanishes, and symmetrizing it fur-
ther with respect to x, y, we get that both summands vanish separately. Thus we see that χ ∈ A∗
and β ∈ HC1(A) in formula (11) satisfy additional conditions χ(d(a)b − ad(b)) = 0 and β(d(a), b) −
β(a, d(b)) = 0 for any a, b ∈ A.
The cocycle equation for one element from L ⊗ A and two elements from D is satisfied trivially, and
(cid:3)
restriction of Φ to D × D gives rise to an element from Z2
comm(D).
Lemma 5.4. Let L be perfect, and one of L, A is finite-dimensional. Then
comm((L ⊗ A) ⊕ Kz A D) ≃ Z2
Z2
comm((L ⊗ A) A D).
Proof. By Lemma 5.1, we have an embedding of Z2
Z2
comm((L ⊗ A) ⊕ Kz A D), and this embedding is an isomorphism if and only if
comm((L ⊗ A) A D) into
(13)
Φ((L ⊗ A) ⊕ Kz A D, z) = 0
for any Φ ∈ Z2
comm((L ⊗ A) ⊕ Kz A D).
Writing the cocycle equation for x ⊗ a, y ⊗ b, z, we get
(14)
Φ([x, y] ⊗ ab + hx, yiξ(a, b)z, z) = 0
for any x, y ∈ L, a, b ∈ A. Substituting here b = 1, we get
(15)
Φ(L ⊗ A, z) = 0.
Substituting the latter equality back to (14), we get Φ(z, z) = 0.
Being restricted to ((L ⊗ A) ⊕ Kz) × ((L ⊗ A) ⊕ Kz), Φ gives rise to a commutative 2-cocycle on
(L ⊗ A) ⊕ Kz, and due to (15), to a commutative 2-cocycle on L ⊗ A. Now we proceed as in the proof
of Lemma 5.3: by Corollary 4.2, the equality (11) holds for any x, y ∈ L, a, b ∈ A and appropriate
comm(L), χ ∈ A∗, α ∈ C(L), and β ∈ HC1(A). Then the cocycle equation for x ⊗ a, y ⊗ b, d, gives
ϕ ∈ Z2
(16) Φ([x, y] ⊗ ab, d) + hx, yiξ(a, b)Φ(z, d)
for any x, y ∈ L, a, b ∈ A, d ∈ D, and substitution of units in this equality yields Φ(L ⊗ A, D) = 0.
Substituting this back to (16) and symmetrizing with respect to x, y, gives
= ϕ(x, y)χ(d(a)b − ad(b)) + α(x, y)(β(d(a), b) − β(a, d(b)))
hx, yiξ(a, b)Φ(z, d) = ϕ(x, y)χ(d(a)b − ad(b))
for any x, y ∈ L, a, b ∈ A, d ∈ D. Either both sides of this equality vanishes, in which case Φ(z, d) = 0
for any d ∈ D, or ϕ(x, y) = λhx, yi for some nonzero λ ∈ K. But the latter is obviously impossible (note
that the condition that the characteristic of the ground field is different from 3 is crucial here: as noted
in [D3, §6.2], in characteristic 3 every symmetric bilinear invariant form is a commutative 2-cocycle).
Therefore, (13) holds and the desired isomorphism follows.
(cid:3)
Affine Kac-Moody algebras, being non-perfect, possess trivial nonzero commutative 2-cocycles. As
the commutant is always of codimension 1, the space of such cocycles is 1-dimensional. There are no
other cocycles, as the following result shows.
Theorem 5.5. Affine Kac-Moody algebras do not possess non-trivial commutative 2-cocycles.
Proof. First consider the case of non-twisted algebras. By Lemmata 5.3 and 5.4, the space of commuta-
tive 2-cocycles on a non-twisted affine Kac-Moody algebra
(g ⊗ K[t, t−1]) ⊕ Kz A Kt
d
dt
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
13
is isomorphic to
comm(g) ⊗ {χ ∈ K[t, t−1]∗ χ(t
Z2
⊕C(g) ⊗ {β ∈ HC1(K[t, t−1) β(t
⊕Z2
comm(Kt
d
dt
).
d f
dt
d f
dt
g − f t
dg
dt
) = 0 for any f , g ∈ K[t, t−1]}
, g) − β( f , t
dg
dt
) = 0 for any f , g ∈ K[t, t−1]}
Let us look at the second tensor factor in the first direct summand. Substituting f = t i, g = t j in the
defining condition
for χ ∈ K[t, t−1]∗, we get (i − j)χ(t i+ j) = 0 for any i, j ∈ Z such that i 6= j. Hence χ = 0, and the first
direct summand vanishes.
χ(t
d f
dt
g − f t
dg
dt
) = 0
By Lemma 1.2, C(g) = 0, hence the second direct summand vanishes too.
Finally, Z2
dt ), being the space of commutative cocycles on the 1-dimensional Lie algebra, is
comm(Kt d
1-dimensional and constitute trivial cocycles.
Now consider the general (twisted) case. Suppose that bL(g, n) possesses non-trivial commutative
2-cocycles. Then by Lemma 5.2,eL(g, n) possesses nonzero commutative 2-cocycles, and by Lemma
2.1, eL(g, n) has a nonzero homomorphic image satisfying the standard identity of degree 5. Every
homomorphic image ofeL(g, n) is either a homomorphic image of L(g, n), or is a central extension of
such. In the latter case, factoring by the central element, we will get again a homomorphic image of
L(g, n) satisfying the standard identity of degree 5. But g is a homomorphic image of L(g, n) under the
evaluation homomorphism
∑
i∈Z
xi ⊗ t i 7→ ∑
i∈Z
xi,
where xi ∈ g, and all but a finite number of summands vanish. Thus g satisfies the standard identity
(1)
of degree 5 too, hence g ≃ sl(2), and the corresponding Kac-Moody algebra is of type A
1 , i.e. is a
non-twisted algebra of the form
(sl(2) ⊗ K[t, t−1]) ⊕ Kz A Kt
d
dt
.
But the case of non-twisted algebras was already covered.
(cid:3)
Note that there is some ambiguity in definition of affine Kac-Moody algebras, and sometimes they
are defined without employing the derivation extension, i.e. merely as central extensions of non-twisted
or twisted current algebras (though some people argue that these are not "real" Kac-Moody algebras,
as they are not Lie algebras corresponding to Cartan matrices). Such algebras are perfect, and possess
(1)
1 . The proof is absolutely
nonzero commutative 2-cocycles only in the case of non-twisted type A
similar to those presented above.
According to Lemma 5.4 (with D = 0) and Corollary 4.2, the space of commutative 2-cocycles on
is infinite-dimensional and is isomorphic to
(1)
the Kac-Moody algebra (sl(2) ⊗ K[t, t−1]) ⊕ Kz of type A
1
comm(sl(2)) ⊗ K[t, t−1]∗, each cocycle being of the form
Z2
(x ⊗ f , y ⊗ g) 7→ ϕ(x, y) ⊗ χ( f g)
(x ⊗ f , z) 7→ 0
(z, z) 7→ 0
where x, y ∈ sl(2), f , g ∈ K[t, t−1], for some ϕ ∈ Z2
comm(sl(2)), χ ∈ K[t, t−1]∗.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
14
6. MODULAR SEMISIMPLE LIE ALGEBRAS
Essentially the same approach as in the previous section, allows to compute the space of commutative
2-cocycles on finite-dimensional semisimple Lie algebras over the field of positive characteristic p.
According to Block's classical theorem (see, for example, [S, Corollary 3.3.6]), the typical examples
of such algebras are Lie algebras of the form (S ⊗ On) A D where S is a simple Lie algebra, On =
K[x1, . . . , xn]/(xp
n) is the reduced polynomial algebra in n variables, and D is a Lie subalgebra
of Wn = Der(On), the simple Lie algebra of the general Cartan type. To ensure semisimplicity, it is
assumed that On does not contain proper D-invariant ideals.
comm((S ⊗ On) A D) ≃ Z2
Theorem 6.1. Let n ≥ 1. Then Z2
1, . . . , xp
comm(D).
Proof. According to Lemma 5.3, Z2
comm((S ⊗ On) A D) is isomorphic to
comm(S) ⊗ {χ ∈ O∗
Z2
n χ(d( f )g − f d(g)) = 0 for any f , g ∈ On, d ∈ D}
⊕C(S) ⊗ {β ∈ HC1(On) β(d( f ), g) − β( f , d(g)) = 0 for any f , g ∈ On, d ∈ D}
⊕Z2
comm(D).
Here again, the second tensor factor in the first direct summand vanishes. Indeed, substitution of
g = 1 in the defining condition for χ ∈ O∗
n, gives χ(d( f )) = 0 for any f ∈ On and d ∈ D. But then
χ(d( f )g) =
1
2
χ(d( f )g − f d(g)) +
1
2
χ(d( f )g + f (d(g))) = χ(d( f g)) = 0
for any f , g ∈ On, d ∈ D. The space D(On)On is evidently a D-invariant ideal of On, hence it coincides
with the whole On, and χ = 0.
And again, C(S) = 0 by Lemma 1.2, so the second direct summand vanishes too, and the desired
(cid:3)
isomorphism follows.
This is similar in spirit to the computation of the second cohomology of some modular semisimple
Lie algebras in [Zu1, §3 and Erratum and addendum].
ACKNOWLEDGEMENTS
Thanks are due to the anonymous referee for pointing few inaccuracies in and improvements to the
previous version of the manuscript, to Dimitry Leites for helpful comments, and to J. Marshall Osborn
who kindly supplied a reprint of his (hard-to-find) paper [O].
REFERENCES
R.K. Amayo and I. Stewart, Infinite-dimensional Lie Algebras, Noordhoff, Leyden, 1974.
G. Benkart, Power-associative Lie-admissible algebras, J. Algebra 90 (1984), 37 -- 58.
R.E. Block, On the extensions of Lie algebras, Canad. J. Math. 20 (1968), 1439 -- 1450.
D. Burde, Classical r-matrices and Novikov algebras, Geom. Dedicata 122 (2006), 145 -- 157.
[AS]
[Be]
[Bl]
[Bu]
[BTW] K. Bowman, D.A. Towers, and V.R. Varea, On flags and maximal chains of lower modular subalgebras of Lie
[D1]
[D2]
[D3]
[DB]
[E]
[F]
[GP]
algebras, J. Lie Theory 17 (2007), 605 -- 616.
A.S. Dzhumadil'daev, Generalized Casimir elements, Math. USSR Izv. 27 (1986), 391 -- 400.
, Symmetric (co)homologies of Lie algebras, Comptes Rendus Acad. Sci. Paris 324 (1997), 497 -- 502.
, Algebras with skew-symmetric identity of degree 3, J. Math. Sci. 161 (2009), 11 -- 30.
and A.B. Bakirova, Simple two-sided anti-Lie-admissible algebras, J. Math. Sci. 161 (2009), 31 -- 36.
A. Elduque, On Lie algebras with a subalgebra of codimension one, Lie Algebras, Madison 1987 (ed. G. Benkart
and J.M. Osborn), Lect. Notes Math. 1373 (1989), 58 -- 66.
V.T. Filippov, Lie algebras satisfying identities of degree 5, Algebra and Logic 34 (1996), 379 -- 394.
S. Garibaldi (with an appendix by A. Premet), Vanishing of trace forms in low characteristics, Algebra and Number
Theory 3 (2009), 543 -- 566; arXiv:0712.3764.
V. Kac, Infinite Dimensional Lie Algebras, 3rd ed., Cambridge Univ. Press, 1995.
A. Larsson, A periodisation of semisimple Lie algebras, Homology, Homotopy and Appl. 4 (2002), 337 -- 355.
[K]
[L]
[NW] K.-H. Neeb and F. Wagemann, The second cohomology of current algebras of general Lie algebras, Canad. J. Math.
60 (2008), 892 -- 922; arXiv:math/0511260.
COMMUTATIVE 2-COCYCLES ON LIE ALGEBRAS
15
J.M. Osborn, Novikov algebras, Nova J. Algebra Geom. 1 (1992), 1 -- 13.
Yu.P. Razmyslov, Identities of Algebras and Their Representations, AMS, Providence, R.I., 1994.
H. Strade, Simple Lie Algebras over Fields of Positive Characteristic. I. Structure Theory, de Gruyter, 2004.
[O]
[R]
[S]
[SF]
[Ze]
[Zu1] P. Zusmanovich, Central extensions of current algebras, Trans. Amer. Math. Soc. 334 (1992), 143 -- 152; Erratum
and R. Farnsteiner, Modular Lie Algebras and Their Representations, Marcel Dekker, 1988.
E.I. Zelmanov, On a class of local translation invariant Lie algebras, Soviet Math. Dokl. 35 (1987), 216 -- 218.
and addendum: to appear; arXiv:0812.2625v2.
[Zu2]
[Zu3]
[Zu4]
, The second homology group of current Lie algebras, Ast´erisque 226 (1994), 435 -- 452; arXiv:0808.0217.
, Invariants of Lie algebras extended over commutative algebras without unit, J. Nonlin. Math. Phys., to
appear; arXiv:0901.1395v1.
, On δ-derivations of Lie algebras and superalgebras, J. Algebra, to appear; arXiv:0907.2034v2.
KAZAKH-BRITISH TECHNICAL UNIVERSITY, TOLE BI 59, ALMATY 050000, KAZAKHSTAN
E-mail address: [email protected]
REYKJAV´IK ACADEMY, ICELAND
E-mail address: [email protected]
|
1011.4112 | 2 | 1011 | 2012-07-04T17:29:49 | The local integration of Leibniz algebras | [
"math.RA",
"math.DG"
] | This article gives a local answer to the coquecigrue problem. Hereby we mean the problem, formulated by J-L. Loday in \cite{LodayEns}, is that of finding a generalization of the Lie's third theorem for Leibniz algebra. That is, we search a manifold provided with an algebraic structure which generalizes the structure of a (local) Lie group, and such that the tangent space at a distinguished point is a Leibniz algebra structure. Moreover, when the Leibniz algebra is a Lie algebra, we want that the integrating manifold is a Lie group. In his article \cite{Kinyon}, M.K. Kinyon solves the particular case of split Leibniz algebras. He shows, in particular, that the tangent space at the neutral element of a Lie rack is provided with a Leibniz algebra structure. Hence it seemed reasonable to think that Lie racks give a solution to the coquecigrue problem, but M.K. Kinyon also showed that a Lie algebra can be integrated into a Lie rack which is not a Lie group. Therefore, we have to specify inside the category of Lie racks, which objects are the coquecigrues. In this article we give a local solution to this problem. We show that every Leibniz algebra becomes integrated into a \textit{local augmented Lie rack}. The proof is inspired by E. Cartan's proof of Lie's third theorem, and, viewing a Leibniz algebra as a central extension by some center, proceeds by integrating explicitely the corresponding Leibniz 2-cocycle into a rack 2-cocycle. This proof gives us a way to construct local augmented Lie racks which integrate Leibniz algebras, and this article ends with examples of the integration of non split Leibniz algebras in dimension 4 and 5. | math.RA | math |
The local integration of Leibniz algebras
Simon Covez
Abstract
This article gives a local answer to the coquecigrue problem. Hereby we mean the
problem, formulated by J-L. Loday in [Lod93], is that of finding a generalization of the
Lie's third theorem for Leibniz algebra. That is, we search a manifold provided with an
algebraic structure which generalizes the structure of a (local) Lie group, and such that the
tangent space at a distinguished point is a Leibniz algebra structure. Moreover, when the
Leibniz algebra is a Lie algebra, we want that the integrating manifold is a Lie group. In
his article [Kin07], M.K. Kinyon solves the particular case of split Leibniz algebras. He
shows, in particular, that the tangent space at the neutral element of a Lie rack is provided
with a Leibniz algebra structure. Hence it seemed reasonable to think that Lie racks give a
solution to the coquecigrue problem, but M.K. Kinyon also showed that a Lie algebra can
be integrated into a Lie rack which is not a Lie group. Therefore, we have to specify inside
the category of Lie racks, which objects are the coquecigrues. In this article we give a local
solution to this problem. We show that every Leibniz algebra becomes integrated into a local
augmented Lie rack. The proof is inspired by E. Cartan's proof of Lie's third theorem, and,
viewing a Leibniz algebra as a central extension by some center, proceeds by integrating
explicitely the corresponding Leibniz 2-cocycle into a rack 2-cocycle. This proof gives us a
way to construct local augmented Lie racks which integrate Leibniz algebras, and this article
ends with an example of the integration of a non split Leibniz algebra in dimension 5.
Introduction
The main result of this article is a local answer to the coquecigrue problem. By coquecigrue
problem, we mean the problem of integrating Leibniz algebras. This question was formulated
by J.-L. Loday in [Lod93] and consists in finding a generalisation of the Lie's third theorem for
Leibniz algebras. This theorem establishes that for every Lie algebra g, there exists a Lie group
G such that its tangent space at 1 is provided with a structure of Lie algebra isomorphic to g.
Leibniz algebras are generalisations of Lie algebras, they are their non-commutative analogues.
Precisely, a (left) Leibniz algebra (over R) is an R-vector space g provided with a bilinear map
[−,−] : g× g → g called the bracket and satisfying the (left) Leibniz identity for all x, y and z in
g
[x, [y, z]] = [[x, y], z] + [y, [x, z]]
Hence, a natural question is to know if, for every Leibniz algebra, there exists a manifold provided
with an algebraic structure generalizing the group structure, and such that the tangent space at
a distinguished point, called 1, can be provided with a Leibniz algebra structure isomorphic to
the given Leibniz algebra. As we want this integration to be the generalization of the Lie algebra
case, we also require that, when the Leibniz algebra is a Lie algebra, the integrating manifold is
a Lie group.
One result about this question was given by M.K. Kinyon in [Kin07]. In his article he solves
the particular case of split Leibniz algebras. A Leibniz algebras is split when it is isomorphic
1
to the demisemidirect product of a Lie algebra and a module over this Lie algebra, that is
isomorphic to g⊕ a as vector space and where the bracket is given by [(x, a), (y, b)] = ([x, y], x.a).
In this case he shows that the algebraic structure which answers the problem is the structure
of a digroup. A digroup is a set with two binary operation (cid:96) and (cid:97), a neutral element 1 and
some compatibility conditions. More precisely, Kinyon shows that a digroup structure induces a
pointed rack structure (pointed in 1), and it is this algebraic structure which gives the tangent
space at 1 a Leibniz algebra structure. Of course, not every Leibniz algebra is isomorphic to
a demisemidirect product, so we have to find a more general structure to solve the problem.
One should think that the right structure is that of a pointed rack, but M.K. Kinyon showed
in [Kin07] that the second condition (Lie algebra becomes integrated into a Lie group) is not
always fulfilled. Thus we have to specify the structure inside the category of pointed racks.
In this article we don't give a complete answer to the coquecigrue problem in the sense that
we only construct a local algebraic structure and not a global one. Indeed, to define an algebraic
structure on a tangent space at a given point on a manifold, we just need an algebraic structure
in a neighborhood of this point. We will show in chapter 3 that a local answer to the problem
is given by the pointed augmented local racks which are abelian extensions of a Lie group by an
anti-symmetric module.
Our approach to the problem is similar to the one given by E. Cartan in [Car30]. The main
idea comes from the fact that we know the Lie's first and second theorem on a class of Lie
algebras. For example, every abelian Lie algebra or every Lie subalgebra of the Lie algebra
End(V ) is integrable (using the Lie's first theorem). More precisely, let g be a Lie algebra, Z(g)
its center and g0 the quotient of g by Z(g). The Lie algebra Z(g) is abelian and g0 is a Lie
subalgebra of End(g), thus there exist Lie groups, respectively Z(g) and G0, which integrate
these Lie algebras. As a vector space, g is isomorphic to the direct sum g0 ⊕ Z(g), thus the
tangent space at (1, 0) of the manifold G0 × Z(g) is isomorphic to g. As a Lie algebra, g is
isomorphic to the central extension g0 ⊕ω Z(g) where ω is a Lie 2-cocycle on g0 with coefficients
in Z(g). That is, the bracket on g0 ⊕ω Z(g) is defined by
[(x, a), (y, b)] =(cid:0)[x, y], ω(x, y)(cid:1)
where ω is an anti-symmetric bilinear form on g0 with value on Z(g) which satisfies the Lie
algebra cocycle identity
ω([x, y], z) − ω(x, [y, z]) + ω(y, [x, z]) = 0
Hence we have to find a group structure on G0×Z(g) which gives this Lie algebra structure on the
tangent space at (1, 0). It is clear that the bracket (1) is completely determined by the bracket
on g0 and the cocycle ω. Hence, the only thing we have to understand is ω. The Lie algebra g is
a central extension of g0 by Z(g), thus we can hope that the Lie group which integrates g should
be a central extension of G0 by Z(g). To follow this idea, we have to find a group 2-cocycle on
G0 with coefficients in Z(g). In this case, the group structure on G0 × Z(g) is given by
(g, a).(h, b) =(cid:0)gh, a + b + f (g, h)(cid:1)
(2)
where f is a map from G × G → Z(g) vanishing on (1, g) and (g, 1) and satisfying the group
cocycle identity
f (h, k) − f (gh, k) + f (g, hk) − f (g, h) = 0
(1)
(3)
With such a cocycle, the conjugation in the group is given by the formula
(g, a).(h, b).(g, a)−1 =(cid:0)ghg−1, a + f (g, h) − f (ghg−1, g)(cid:1)
2
and by imposing a smoothness condition on f in a neighborhood of 1, we can differentiate this
formula twice, and obtain a bracket on g0 ⊕ Z(g) defined by
[(x, a), (y, b)] =(cid:0)[x, y], D2f (x, y)(cid:1)
where D2f (x, y) = d2f (1, 1)((x, 0), (0, y)) − d2f (1, 1)((y, 0), (0, x)). Thus, if D2f (x, y) equals
ω(x, y), we recover the bracket (1). Hence, if we associate to ω a group cocyle f satisfying some
smoothness conditions and such that D2f = ω, then our integration problem is solved. This can
be done in two steps. The first one consists in finding a local Lie group cocycle defined around
1. Precisely, we want a map f defined on a subset of G0 × G0 containing (1, 1) with values in
Z(g) which satisfies the local group cocycle identity (cf. [vE58] for a definition of local group).
We can construct explicitely such a local group cocycle. This construction is the following one
(cf. Lemma 5.2 in [Nee04]) :
Let V be an open convex 0-neighborhood in g0 and φ : V → G0 a chart of G0 with φ(0) = 1
and dφ(0) = idg0. For all (g, h) ∈ φ(V ) × φ(V ) such that gh ∈ φ(V ) let us define f (g, h) ∈ Z(g)
by the formula
(cid:90)
f (g, h) =
ωinv
γg,h
where ωinv ∈ Ω2(G0, Z(g)) is the invariant differential form on G0 associated to ω and γg,h is
the smooth singular 2-chain defined by
(cid:32)
(cid:18)
φ−1(cid:16)
gφ(cid:0)sφ−1(h)(cid:1)(cid:17)(cid:19)
(cid:18)
φ−1(cid:16)
gφ(cid:0)(1 − t)φ−1(h)(cid:1)(cid:17)(cid:19)(cid:33)
+ s
γg,h(t, s) = φ
t
The formula for f defines a smooth function such that D2f (x, y) = ω(x, y). We now only have to
check whether f satisfies the local group cocycle identity. Let (g, h, k) ∈ φ(V )3 such that gh, hk
and ghk are in φ(V ). We have
f (h, k) − f (gh, k) + f (g, hk) − f (g, h) =
=
(cid:90)
(cid:90)
ωinv +
ωinv −
ωinv
γh,k
γgh,k
γg,hk
γg,h
(cid:90)
(cid:90)
(cid:90)
ωinv −
ωinv
∂γg,h,k
(cid:90)
(cid:90)
where γg,h,k is a smooth singular 3-chain in φ(V ) such that ∂γg,h,k = gγh,k − γgh,k + γg,hk − γg,h
(such a chain exists because φ(V ) is homeomorphic to the convex open subset V of g0). Thus
f (h, k) − f (gh, k) + f (g, hk) − f (g, h) =
ωinv =
ddRωinv = 0
∂γg,h,k
γg,h,k
because ωinv is a closed 2-form. Hence, we have associated to ω a local group 2-cocycle, smooth
in a neighborhood in 1, and such that D2f (x, y) = ω(x, y). Thus we can define a local Lie group
structure on G0 × Z(g) by setting
(g, a)(h, b) =(cid:0)gh, a + g.b + f (g, h)(cid:1),
and the tangent space at (1, 0) of this local Lie group is isomorphic to g. If we want a global
structure, we have to extend this local cocycle to the whole group G0. First P.A. Smith ([Smi52,
Smi51]), then W.T. Van Est ([vE62a, vE62b]) have shown that it is precisely this enlargement
which may meet an obstruction coming from both π2(G0) and π1(G0).
In finite dimension,
π2(G0) = 0, thus there is no obstruction to integrate Lie algebras. This equality is no longer
3
true in infinite dimension, hence this obstruction prevents the integration of infinite dimensional
Lie algebras into global Lie groups (cf. [Nee02, Nee04]).
To integrate Leibniz algebras into pointed racks, we follow a similar approach. In this context,
we use the fact that we know how to integrate any (finite dimensional) Lie subalgebra of End(V )
for V a vector space.
In a similar way as the Lie algebra case, we associate to any Leibniz
algebra an abelian extension of a Lie algebra g0 by an anti-symmetric representation ZL(g). As
we have the theorem for Lie algebras, we can integrate g0 and ZL(g) into the Lie groups G0 and
ZL(g), and, using the Lie's second theorem, ZL(g) is a G0-module. Then, the main difficulty
becomes the integration of the Leibniz cocycle into a local Lie rack cocycle. In chapter 3 we
explain how to solve this problem. We make a similar construction as in the Lie algebra case,
but in this context, there are several difficulties which appear. One of them is that our cocycle
is not anti-symmetric, so we can't consider the equivariant form associated to it and integrate
this form. To solve this problem, we will use Proposition 1.1 which, in particular, establishes an
isomorphism from the 2-nd cohomology group of a Leibniz algebra g with coefficients in an anti-
symmetric representation aa to the 1-st cohomology group of g with coefficients in the symmetric
representation Hom(g, a)s. In this way, we get a 1-form that we can now integrate. Another
difficulty is to specify on which domain this 1-form should be integrated. In the Lie algebra case,
we integrate over a 2-simplex and the cocycle identity is verified by integrating over a 3-simplex,
whereas in our context we will replace the 2-simplex by the 2-cube and the 3-simplex by a 3-cube.
Let us describe the content of the article section-wise.
Section 1: Leibniz algebras
This whole section, except Proposition 1.1, is based on [Lod93, LP93, Lod98]. We first give
the basic definitions we need about Leibniz algebras. Unlike J.-L. Loday and T. Pirashvili, who
work with right Leibniz algebras, we study left Leibniz algebras. Hence, we have to translate
all the definitions needed into our context. As we have seen above, we translate our integration
problem into a cohomological problem, thus we need a cohomology theory for Leibniz algebras
and, a fortiori, a notion of representation. We take the definition of a representation of a Leibniz
algebra given by J.-L. Loday and T. Pirashvili in [LP93]. We end this section with a fondamental
result (Proposition 1.1). This proposition establishes an isomorphism of cochain complexes from
CLn(g, aa) to CLn−1(g, Hom(g, a)s). The important fact in this result is the transfer from an
anti-symmetric representation to a symmetric one. This will be useful when we will have to
associate a local Lie rack 2-cocycle to a Leibniz 2-cocycle.
Section 2: Lie racks
The notion of rack comes from topology, in particular, the theory of invariants of knots and links
(cf. for example [FR92]). It is M.K. Kinyon in [Kin07] who was the first to link racks to Leibniz
algebras. The idea of linking these two structures comes from the case of Lie groups and Lie
algebras and in particular from the construction of the bracket using the conjugation. Indeed, a
way to define a bracket on the tangent space at 1 of a Lie group is to differentiate the conjugation
morphism twice. Let G a Lie group, the conjugation is the group morphism c : G → Aut(G)
defined by cg(h) = ghg−1. If we differentiate this expression with respect to the variable h at 1,
we obtain a Lie group morphism Ad : G → Aut(g). We can still derive this morphism at 1 to
obtain a linear map ad : g → End(g). Then, we are allowed to define a bracket [−,−] on g by
setting [x, y] = ad(x)(y). We can show that this bracket satisfies the left Leibniz identity, and
that this identity is induced by the equality cg(ch(k)) = ccg(h)(cg(k)). Thus, if we denote cg(h)
by g(cid:66)h, the only properties we use to define a Lie bracket on g are
4
1. g(cid:66) : G → G is a bijection for all g ∈ G.
2. g(cid:66)(h(cid:66)k) = (g(cid:66)h)(cid:66)(g(cid:66)k) for all g, h, k ∈ G
3. g(cid:66)1 = 1 and 1(cid:66)g = g for all g ∈ G.
Hence, we call (left) rack, a set provided with a binary operation (cid:66) satisfying the first and the
second condition. A rack is called pointed if there exists an element 1 which satisfies the third
condition. We begin this chapter by giving definitions and examples, for this we follow [FR92].
They work with right racks, hence, as in the Leibniz algebra case, we translate the definitions to
left racks. In particular, we give the most important example called (pointed) augmented rack.
This example presents similarities with crossed modules of groups, and in this case, the rack
structure is induced by a group action.
As in the group case, we want to construct a pointed rack associated to a Leibniz algebra
using an abelian extension. Hence, we need a cohomology theory where the second cohomology
group corresponds to the extension classes of a rack by a module. We take the definitions given
by N. Andruskiewitsch and M. Graña in [AG03].
At the end of this section, we give the definitions of local rack cohomology and (local) Lie
rack cohomology.
Section 3: Lie racks and Leibniz algebras
This section is the heart of this article. It gives the local solution for the coquecigrue problem. To
our knowledge, all the results in this chapter are new, except Proposition 3.1 due to M.K. Kinyon
([Kin07]). First, we recall the link between (local) Lie racks and Leibniz algebras explained by
M.K. Kinyon in [Kin07] (Proposition 3.1). Then, we study the passage from smooth As(X)-
modules to Leibniz representations (Proposition 3.6) and (local) Lie rack cohomology to Leibniz
cohomology. We define a morphism from the (local) Lie rack cohomology of a rack X with
coefficients in a As(X)-module As (resp. Aa) to the Leibniz cohomology of the Leibniz algebra
associated to X with coefficients in as = T0A (resp. aa) (Proposition 3.7).
The end of this section (section 3.4 to 3.7) is on the integration of Leibniz algebras into
local Lie racks. We use the same approach as E. Cartan for the Lie groups case. That is, for
every Leibniz algebra, we consider the abelian extension by the left center and integrate it. This
extension is caracterized by a 2-cocycle, and we construct (Proposition 3.15) a local Lie rack
2-cocycle integrating it by an explicit construction similar to the one explained in the Lie group
case. This construction is summarized in our main theorem (Theorem 3.21). We remark that the
constructed 2-cocycle has more structure (Proposition 3.19). That is, the rack cocycle identity
is induced by another one. This other identity permits us to provide our constructed local Lie
rack with a structure of augmented local Lie rack (Proposition 3.24). We end this section with
an example of the integration of a non split Leibniz algebra in dimension 5.
1 Leibniz algebras
As it is written in the introduction, we work with left Leibniz algebras instead of right Leibniz
algebras. The main reason comes from the fact that M.K. Kinyon works in this context in his
article [Kin07]. Indeed, this article is our starting point of the integration problem for Leibniz
algebras. Thus, we have chosen to work in this context.
A (left) Leibniz algebra (over R) is a vector space g (over R) provided with a bracket [−,−] :
g ⊗ g → g, which satisfies the left Leibniz identity
[x, [y, z]] = [[x, y], z] + [y, [x, z]].
5
Remark that an equivalent way to define a left Leibniz algebra is to say that, for all x ∈ g,
[x,−] is a derivation for the bracket [−,−]. The first example of a Leibniz algebra is a Lie
algebra. Indeed, if the bracket is anti-symmetric, then the Leibniz identity is equivalent to the
Jacobi identity. Hence, we have a functor inc : Lie → Leib. This functor has a left adjoint
(−)Lie : Leib → Lie which is defined on the objects by gLie = g/gann, where gann is the two-
sided ideal of g generated by the set {[x, x] ∈ g x ∈ g}. We can remark that there are other
ways to construct a Lie algebra from a Leibniz algebra. One is to quotient g by the left center
ZL(g) = {x ∈ g [x,−] = 0}, but this construction is not functorial.
To define a cohomology theory for Leibniz algebras, we need a notion of representation of such
algebraic structure. As we work with left Leibniz algebra, we have to translate the definition given
by J.L. Loday and T. Pirashvili in their article [LP93]. In our context, a representation over a
Leibniz algebra g, becomes a vector space M provided with two linear maps [−,−]L : g⊗M → M
and [−,−]R : M ⊗ g → g, satisfying the following three axioms
[x, [y, m]L]L = [[x, y], m]L + [y, [x, m]L]L
[x, [m, y]R]L = [[x, m]L, y]R + [m, [x, y]]R
[m, [x, y]]R = [[m, x]R, y]R + [x, [m, y]R]L
(LLM)
(LM L)
(M LL)
Recall that, for a Lie algebra g, a representation of g is a vector space M provided with a linear
map [−,−] : g⊗ M → M satisfying [[x, y], m] = [x, [y, m]]− [y, [x, m]]. A Lie algebra is a Leibniz
algebra, hence we want that a Lie representation M of a Lie algebra g, is a Leibniz representation
of g. We have two canonical choices for putting a Leibniz representation structure on M. One
possibility is by setting [−,−]L = [−,−] and [−,−]R = −[−,−], and a second one is by setting
[−,−]L = [−,−] and [−,−]R = 0. These Leibniz representations are examples of particular
Leibniz representations. The first one is an example of a symmetric representation, and the
second one is an example of an anti-symmetric representation. A symmetric representation is
a Leibniz representation where [−,−]L = −[−,−]R and an anti-symmetric representation is a
Leibniz representation where [−,−]R = 0. A Leibniz representation which is symmetric and
anti-symmetric is called trivial.
Now, we are ready to define a cohomology theory for Leibniz algebras. The existence of a
cohomology (and homology) theory for these algebras is one of the main motivation for studying
them because, restricted to Lie algebras, this theory gives us new invariants (cf. [Lod93]). For g a
Leibniz algebra and M a representation of g, we define a cochain complex {CLn(g, M ), dLn}n∈N
by setting
CLn(g, M ) = Hom(g⊗n, M )
and
dLnω(x0, . . . , xn) =
n−1(cid:88)
(cid:88)
i=0
+
(−1)i[xi, ω(x0, . . . , xi, . . . , xn)]L + (−1)n−1[ω(x0, . . . , xn−1), xn]R
(−1)i+1ω(x0, . . . , xj−1, [xi, xj], xj+1, . . . , xn)
0≤i<j≤n
To prove that dLn+1 ◦ dLn = 0, we use Cartan's formulas. These formulas are described in
[LP93] in the right Leibniz algebra context, but we can adapt them easily in our context.
Like for many algebraic structures, the second cohomology group of a Leibniz algebra g
with coefficients in a representation M is in bijection with the set of equivalence classes of
abelian extensions of g by M (cf.
[LP93]). An abelian extension of a Leibniz algebra g by M
p(cid:16) g is a short exact sequence of Leibniz algebra
is a Leibniz algebra g such that, M
(cid:44)→ g
i
6
(where M is considered as an abelian Leibniz algebra) and the representation structure of M
is compatible with the representation structure induced by this short exact sequence. That is,
[m, x]R = i−1[i(m), s(x)] and [x, m]L = i−1[s(x), i(m)] where s is a section of p and the bracket
is that of g (of course, we have to justify that this representation structure of g on M induced
by the short exact sequence doesn't depend on s, but we deduce it easily from the fact that the
difference of two sections of p is in i(M )).
There are canonical abelian extensions associated to a Leibniz algebra. The one we will use
to integrate Leibniz algebras is the abelian extension by the left center
ZL(g)
i
(cid:44)→ g
p(cid:16) g0
where g0 := g/ZL(g). This is an extension of a Lie algebra by an anti-symmetric representation.
In a sense, a symmetric representation is closer to a Lie representation than to an anti-symmetric
representation. Hence, it is convenient to pass from a anti-symmetric representation to a sym-
metric representation. Let g be a Lie algebra and M a Lie representation of g, then we define a
Lie representation structure on Hom(g, M ) by setting
(x.α)(y) := x.(α(y)) − α([x, y])
for all x, y ∈ g and α ∈ Hom(g, M ). The following proposition establishes an isomorphism from
HLn(g, M a) to HLn−1(g, Hom(g, M )s), where M a (resp. Hom(g, M )s) means that M (resp.
Hom(g, M )) is provided with a anti-symmetric (resp. symmetric) g-representation structure.
Proposition 1.1. Let g be a Lie algebra and M a Lie representation of g. We have an isomor-
phism of cochain complexes
CLn(g, M a) τ n→ CLn−1(g, Hom(g, M )s)
given by ω (cid:55)→ τ n(ω) where τ n(ω)(x1, . . . , xn−1)(xn) = ω(x1, . . . , xn).
Proof. This morphism is clearly an isomorphism ∀n ≥ 0. Moreover, we have
dLτ n(ω)(x0, . . . , xn−1)(xn) =
i=0
0≤i<j≤n−1
+ (−1)n−1[xn−1, τ n(ω)(x0, . . . , xn−2)](xn)
+
(−1)i+1τ n(ω)(x0, . . . , xj−1, [xi, xj], xj+1, . . . , xn−1)(xn)
n−2(cid:88)
(−1)i[xi, τ n(ω)(x0, . . . ,(cid:98)xi, . . . , xn−1)](xn)
(cid:88)
n−1(cid:88)
(−1)i([xi, ω(x0, . . . , xi, . . . , xn−1, xn)] − ω(x0, . . . ,(cid:98)xi, . . . , xn−1, [xi, xn]))
(cid:88)
n−1(cid:88)
(−1)i[xi, ω(x0, . . . ,(cid:98)xi, . . . , xn−1, xn)]
(cid:88)
(−1)i+1ω(x0, . . . , xj−1, [xi, xj], xj+1, . . . , xn−1, xn)
(−1)i+1ω(x0, . . . , xj−1, [xi, xj], xj+1, . . . , xn−1, xn)
=
+
i=0
0≤i<j≤n−1
i=0
+
0≤i<j≤n
dLτ n(ω)(x0, . . . , xn−1)(xn) =
Hence {τ n}n≥0 is a morphism of cochain complexes.
= dLω(x0, . . . , xn−1, xn)
= τ n+1(dLω)(x0, . . . , xn−1)(xn)
7
2 Lie racks
2.1 Definitions and examples
Like in the Leibniz algebra case, we can define left racks and right racks. Because we have made
the choice to work with left Leibniz algebras, we take the definition of left racks. A (left) rack is
a set X provided with a product (cid:66) : X × X → X, which satisfies the left rack identity, that is
for all x, y, z ∈ X :
x(cid:66)(y(cid:66)z) = (x(cid:66)y)(cid:66)(x(cid:66)z),
and such that x(cid:66)− : X → X is a bijection for all x ∈ X. A rack is said to be pointed if there
exists an element 1 ∈ X, called the neutral element, which satisfies 1(cid:66)x = x and x(cid:66)1 = 1 for
all x ∈ X. A rack morphism is a map f : X → Y satisfying f (x(cid:66)y) = f (x)(cid:66)f (y), and a pointed
rack morphism is a rack morphism f such that f (1) = 1.
The first example of a rack is a group provided with the conjugation. Indeed, let G be a
group, we define a rack product (cid:66) on G by setting g(cid:66)h = ghg−1 for all g, h ∈ G. Clearly,
g(cid:66)− is a bijection with inverse g−1(cid:66)− and, an easy computation shows that the rack identity
is satisfied. Hence, we have a functor Conj : Group → Rack. This functor has a left adjoint
As : Rack → Group defined on the objects by As(X) = F (X)/ < {xyx−1(x(cid:66)y−1) x, y ∈ X} >
where F (X) is the free group generated by X, and < {xyx−1(x(cid:66)y−1) x, y ∈ X} > is the normal
subgroup generated by {xyx−1(x(cid:66)y−1) x, y ∈ X}. We can remark that Conj(G) is a pointed
rack. Indeed, we have 1(cid:66)g = g and g(cid:66)1 = 1 for all g ∈ G, where 1 is the neutral element for the
group product. Hence, Conj is a functor from Group to P ointedRack. This functor has a left
adjoint Asp : P ointedRack → Group, defined on the objects by Asp(X) = As(X)/ < {[1]} >,
where < {[1]} > is the subgroup of As(X) generated by the class [1] ∈ As(X).
A second example, and maybe the most important, is the example of augmented racks.
p→ G satisfying the
An augmented rack is the data of a group G, a G-set X, and a map X
augmentation identity, that is for all g ∈ G and x ∈ X
p(g.x) = gp(x)g−1.
Then, we define a rack structure on X by setting x(cid:66)y = p(x).y. If there exists an element 1 ∈ X
such that p(1) = 1 and g.1 = 1 for all g ∈ G, then the augmented rack X
p→ G is said to be
pointed, and the associated rack (X, (cid:66)) is pointed. We can remark that crossed modules and
precrossed modules of groups are examples of augmented racks.
2.2 Pointed rack cohomology
To define a pointed rack cohomology theory, we need a good notion of pointed rack module.
In this article, we take the definition given by N. Andruskiewitsch and M. Graña in [AG03].
Let X be a pointed rack, an X-module is an abelian group A, provided with two families of
homomorphisms of the abelian group A, (φx,y)x,y∈X and (ψx,y)x,y∈X, satisfying the following
axioms
(M0) φx,y is an isomorphism.
(M1) φx,y(cid:66)z ◦ φy,z = φx(cid:66)y,x(cid:66)z ◦ φx,z
(M2) φx,y(cid:66)z ◦ ψy,z = ψx(cid:66)y,x(cid:66)z ◦ φx,y
(M3) ψx,y(cid:66)z = φx(cid:66)y,x(cid:66)z ◦ ψx,z + ψx(cid:66)y,x(cid:66)z ◦ ψx,y
(M4) φ1,y = idA ∀y ∈ X and ψx,1 = 0 ∀x ∈ X
8
Remark 2.1. There is a more general definition of (pointed) rack module given by N. Jackson
in [Jac05], but we don't need this degree of generality. This definition of pointed rack module
coincides with the definition of homogeneous pointed rack module given in [Jac05].
For example, there are two canonical X-module structures on an Asp(X)-module. Indeed, let A
be a Asp(X)-module, that is an abelian group provided with a group morphism ρ : Asp(X) →
Aut(A), the first X-module structure, called symmetric, that we can define on A is given for all
x, y ∈ X by
φx,y(a) = ρx(a) and ψx,y(a) = a − ρx(cid:66)y(a).
The second, called anti-symmetric, is given for all x, y ∈ X by
φx,y(a) = ρx(a) and ψx,y(a) = 0.
With this definition of module, N. Andruskiewitsch and M. Graña define a cohomology theory
for pointed racks. For X a pointed rack and A a X-module, they define a cochain complex
{CRn(X, A), dn
R}n∈N by setting
CRn(X, A) = {f : X n → A f (x1, . . . , 1, . . . , xn) = 0}
and
n(cid:88)
i=1
dn
Rf (x1, . . . , xn+1) =
(−1)i−1(cid:0)φx1(cid:66)...(cid:66)xi,x1(cid:66)...(cid:66)(cid:98)xi(cid:66)...(cid:66)xn+1(f (x1, . . . ,(cid:98)xi, . . . , xn+1)) − f (x1, . . . , xi(cid:66)xi+1, . . . , xi(cid:66)xn+1)(cid:1)
+(−1)nψx1(cid:66)...(cid:66)xn,x1(cid:66)...(cid:66)xn−1(cid:66)xn+1(f (x1, . . . , xn))
This complex is the same as the one defined in [Jac05], but in the left rack context. Adapting
the proof given by N. Jackson in [Jac05], one easily sees that the second cohomology group
HR2(X, A) is in bijection with the set of equivalence classes of abelian extensions of a pointed
rack X by a X-module A. An abelian extension of a pointed rack X by a X-module A is a
surjective pointed rack homomorphism E
(E0) for all x ∈ X, there is a simply transitively right action of A on p−1(x).
(E1) for all u ∈ p−1(x), v ∈ p−1(y), a ∈ A, we have (u.a)(cid:66)v = (u(cid:66)v).ψx,y(a).
(E2) for all u ∈ p−1(x), v ∈ p−1(y), a ∈ A, we have u(cid:66)(v.a) = (u(cid:66)v).φx,y(a).
p(cid:16) X which satisfies the following axioms
p2(cid:16) X are called equivalent, if there exists a pointed rack
p1(cid:16) X , E2
θ→ E2 which satisfies the following axioms
and two extensions E1
isomorphism E1
1. p2 ◦ θ = p1.
2. for all x ∈ X, u ∈ p−1(x), a ∈ A, we have θ(u.a) = θ(u).a.
2.3 Lie racks
To generalize Lie groups, we need a pointed rack provided with a differentiable structure com-
patible with the algebraic structure. This is the notion of Lie racks. A Lie rack is a smooth
manifold X provided with a pointed rack structure such that the product (cid:66) is smooth, and such
9
that for all x ∈ X cx is a diffeomorphism. We will see in section 3 that the tangent space at the
neutral element of a Lie rack is provided with a Leibniz algebra structure.
s (X, A), dn
Let X be a Lie rack, a X-module A is said smooth if A is a abelian Lie group, and if
φ : X × X × A → A and ψ : X × X × A → A are smooth. Then we can define a cohomology
theory for Lie racks with values in a smooth module. For this we define a cochain complex
{CRn
s (X, A) is the set of functions f : X n → A which are smooth in
a neighborhood of (1, . . . , 1) ∈ X n and such that f (x1, . . . , 1, . . . , xn) = 0 for all x1, . . . , xn ∈ X.
The formula for the differential dR is the same as the one defined previously. We will see that a Lie
rack cocycle (respectively a coboundary) derives itself in a Leibniz algebra cocycle (respectively
coboundary).
R}n∈N where CRn
2.4 Local racks
To define a Lie algebra structure on the tangent space at the neutral element of a Lie group, we
can remark that we only use the local Lie group structure in the neighborhood of 1. We will see
that this remark remains true for Lie racks and Leibniz algebras.
A local rack is a set X provided with a product (cid:66) defined on a subset Ω of X × X with values
in X, and such that the following axioms are satisfied:
1. If (x, y), (x, z), (y, z), (x, y(cid:66)z) and (x(cid:66)y, x(cid:66)z) ∈ Ω, then x(cid:66)(y(cid:66)z) = (x(cid:66)y)(cid:66)(x(cid:66)z).
2. If (x, y), (x, z) ∈ Ω and x(cid:66)y = x(cid:66)z, then y = z.
A local rack is said to be pointed if there is a element 1 ∈ X such that 1(cid:66)x and x(cid:66)1 are defined
for all x ∈ X and respectively equal to x and 1. We called this element the neutral element.
Then a local Lie rack is a pointed local rack (X, Ω, 1) where X is a smooth manifold, Ω is a open
subset of X, and (cid:66) : Ω → X is smooth. For example, every Lie rack open subset containing
the neutral element is a local Lie rack. Given such a local Lie rack, we can define a associated
cohomology theory.
Let X be a Lie rack, U a subset of X containing the neutral element 1 and A a smooth
s (U, A) as the set of maps f : Un−loc → A, smooth
X-module. We define for all n ∈ N, CRn
If A is
in a neighborhood of the neutral element, and such that f (x1, . . . , 1, . . . , xn) = 0.
not anti-symmetric, then Un−loc is the subset of elements (x1, . . . , xn) of X × U n−1 satisfy-
ing xi1
If A is anti-symmetric, Un−loc
is the subset of elements (x1, . . . , xn) of X n−1 × U satisfying xi1
(cid:66)xn ∈ U, for all
i1 < ··· < ij < n, 1 ≤ j ≤ n− 1. One easily checks that the formula for the differential dR allows
R}n∈N. Then we define U-local Lie rack cohomology
us to define a cochain complex {CRn
of X with coefficients in A as the cohomology of the cochain complex {CRn
(cid:66) . . . (cid:66)xij ∈ U, for all i1 < ··· < ij, 2 ≤ j ≤ n.
(cid:66) . . . (cid:66)xij
s (U, A), dn
s (U, A), dn
R}n∈N.
3 Lie racks and Leibniz algebras
3.1 From Lie racks to Leibniz algebras
In this section we recall how a Leibniz algebra is canonically associated to a Lie rack.
Proposition 3.1 ([Kin07]). Let X be a Lie rack, then T1X is provided with a Leibniz algebra
structure.
10
Let X be a Lie rack and denote by x the tangent space to X at 1. The Leibniz algebra structure
on T1X is constructed as follow. The conjugation (cid:66) induces for all x ∈ X an automorphism of
Lie racks cx : X → X defined by cx(y) = x(cid:66)y. Define for all x ∈ X the map
Adx = T1cx ∈ GL(x).
x and c1 = id, hence Ad : X →
The pointed rack structure on X implies that cx(cid:66)y = cx ◦ cy ◦ c−1
GL(x) is a morphism of Lie racks. Let ad = D1Ad : x → gl(x) the differential of Ad at 1. Define
a bracket [−,−] on x = T1X by setting
[u, v] = ad(u)(v).
Differentiate the rack identity x(cid:66)(y(cid:66)z) = (x(cid:66)y)(cid:66)(x(cid:66)z) with respect to each variables involves
the Leibniz identity for the bracket [−,−] (cf. [Kin07]).
Example 3.2 (Group). Let G be a Lie group. We get in this way the canonical Lie algebra
structure on T1G.
p→ G be an augmented Lie rack. The linear map
Example 3.3 (Augmented rack). Let X
T1p→ T1G is a Lie algebra in the category of linear maps (see [LP98]). This structure induces
T1X
a Leibniz algebra structure on T1X which is isomorphic to the one induces by the Lie rack
structure on X.
We remark that a local smooth structure around 1 is sufficient to provide T1X with a Leibniz
algebra structure.
Proposition 3.4. Let X be a local Lie rack, then T1X is a Leibniz algebra.
3.2 From Asp(X)-modules to Leibniz representations
Let X be a rack. An Asp(X)-module is an abelian group A provided with a morphism of groups
φ : Asp(X) → Aut(A). By adjointness, this is the same thing as a morphism of pointed racks
φ : X → Conj(Aut(A)).
Definition 3.5. Let X be a Lie rack, a smooth As(X)-module is an Asp(X)module A where
A is an abelian Lie group and φ : X × A → A is smooth.
Recall that, given a Leibniz algebra g, a g-representation a is a vector space provided with
two linear maps [−,−]L : g⊗a → a and [−,−]R : a⊗g → a satisfying the axioms (LLM ), (LM L)
and (M LL) given in section 1.
There are two particular classes of modules. The first, called symmetric, are the modules
where [−,−]L = −[−,−]R. The second, called anti-symmetric, are the modules where [−,−]R =
0. Given a Leibniz algebra g and a a vector space equipped with a morphism of Leibniz algebra
φ : g → End(a), we can put two structures of g-representation on a. One is symmetric and
defined by
[x, a]L = φx(a) and [a, x]R = −φx(a), ∀x ∈ g, a ∈ a.
The other is anti-symmetric and defined by
[x, a]L = φx(a) and [a, x]R = 0, ∀x ∈ g, a ∈ a.
Moreover, given a rack X and A a (smooth) As(X)-module, we can put two structures of
(smooth) X-module on A. One is called symmetric and defined by
φx,y(a) = φx(a) and ψx,y(a) = a − φx(cid:66)y(a), ∀x, y ∈ X, a ∈ A.
11
The other is called anti-symmetric and defined by
φx,y(a) = φx(a) and ψx,y(a) = 0, ∀x, y ∈ X, a ∈ A.
These constructions are related to each other because one is the infinitesimal version of the
other. Indeed, let (A, φ, ψ) be a smooth symmetric X-module. We have by definition two smooth
maps
φ : X × X × A → A and ψ : X × X × A → A
with φ1,1 = id, ψ1,1 = 0. Thus the differentials of these maps at (1, 1) give us two maps
: X × X → Aut(a); (x, y) = T1φx,y and χ : X × X → End(a); χ(x, y) = T1ψx,y.
These maps are smooth, so we can differentiate them at (1, 1) to obtain
T(1,1) : x ⊕ x → End(a) and T(1,1)χ : x ⊕ x → End(a).
Then we define two linear maps [−,−]L : x ⊗ a → a and [−,−]R : a ⊗ x → a by
[u, m]L = T(1,1)(u, 0)(m) and [m, u]R = T(1,1)χ(0, u)(m).
Proposition 3.6. Let X be a Lie rack, x be its Leibniz algebra, A be an abelian Lie group and
a be its Lie algebra. If (A, φ, ψ) is a smooth symmetric (resp. anti-symmetric) X-module, then
(a, [−,−]L, [−,−]R) is a symmetric (resp. anti-symmetric) x-module.
Proof. It is clear that if (A, φ, ψ) is symmetric then [−,−]L = −[−,−]R, and if (A, φ, ψ) is
anti-symmetric then [−,−]R = 0.
Now let us prove that [−,−]L satisfies the axiom (LLM ). By hypothesis on φ, the relation
φx,y(cid:66)z ◦ φy,z = φx(cid:66)y,x(cid:66)z ◦ φx,z is true for all x, y, z ∈ X. Taking z = 1 we obtain φx,1 ◦ φy,1 =
φx(cid:66)y,1 ◦ φx,1. By differentiating this equality with respect to each variables we find that [−,−]L
satisfies the axiom (LLM ).
3.3 From Lie rack cohomology to Leibniz cohomology
Proposition 3.7. Let X be a Lie rack and let A be a smooth As(X)-module. We have morphisms
of cochains complexes
given by δn(f )(a1, . . . , an) = dnf (1, . . . , 1)(cid:0)(a1, 0, . . . , 0), . . . , (0, . . . , 0, an)(cid:1) (where dnf is the n-
δn→ CLn(x, as) and CRn
δn→ CLn(x, aa),
p (X, Aa)s
CRn
p (X, As)s
th differential of f).
Proof. Let f ∈ CRn
p (X, As) and (x0, . . . , xn) ∈ X n+1, we want to prove that
Let (γ0(t0), . . . , γn(tn)) be a family of paths γi :]−i, +i[→ V such that γi(0) = 1 and ∂
xi. Because f (x0, . . . , 1, . . . , xn) = 0, for all i ∈ {1, . . . , n} we have
∂s
δn+1(dn
Rf ) = dn
L(δn(f )).
(cid:12)(cid:12)s=0 γi(s) =
(cid:12)(cid:12)(cid:12)(cid:12)ti=0
∂n+1
∂t0 . . . ∂tn
φγ0(t0)(cid:66)...(cid:66)γi(ti)(f (γ0(t0), . . . , γi−1(ti−1), γi+1(ti+1), . . . , γn(tn)))
= ai.dn(f )(a0, . . . ,(cid:98)ai, . . . , an)
12
(cid:12)(cid:12)(cid:12)(cid:12)tl=0
Moreover for all i ∈ {1, . . . , n},
is equal to
∂n+1
∂t0 . . . ∂tn
f (γ0(t0), . . . , γi(ti)(cid:66)γi+1(ti+1), . . . , γi(ti)(cid:66)γn(tn))
(cid:12)(cid:12)(cid:12)(cid:12)ti=0
∂
∂ti
which is equal to
dnf (1, . . . , 1)(cid:0)(a0, 0, . . . , 0), . . . , (0, . . . , Adγi(ti)(ai+1), . . . , 0), . . . , (0, . . . , 0, Adγi(ti)(an)(cid:1)
n(cid:88)
dnf (1, . . . , 1)(cid:0)(a0, 0, . . . , 0), . . . , (0, . . . , [ai, ak], . . . , 0), . . . , (0, . . . , 0, an)(cid:1)
k=i+1
Hence
δn+1(dn
Rf )(a0, . . . , an) =
=
+
that is
(−1)i(cid:16)
n(cid:88)
n(cid:88)
(cid:88)
i=0
i=0
0≤i<k≤n
ai.δn(f )(a0, . . . , ai, . . . , an) − n(cid:88)
δnf (a0, . . . , [ai, ak], . . . , an)
(cid:17)
k=i+1
(−1)iai.δn(f )(a0, . . . , ai, . . . , an)
(−1)i+1δnf (a0, . . . , [ai, ak], . . . , an)
δn+1(dn
Rf ) = dn
L(δn(f ))
This is exactly the same proof as for the case where A is anti-symmetric.
We remark that we only need a local cocyle identity around 1. Thus we have
Proposition 3.8. Let X be a Lie rack, let U be a 1-neighborhood in X and let A be a smooth
As(X)-module. We have morphisms of cochain complexes
CRn
p (U, As) δn→ CLn(x, as) and CRn
p (U, Aa) δn→ CLn(x, aa),
given by δn(f )(a0, . . . , an) = dnf (1, . . . , 1)((a1, 0, . . . , 0), . . . , (0, . . . , 0, an)).
3.4 From Leibniz cohomology to local Lie rack cohomology
In this section, we study two cases of Leibniz cocyles integration. This section will be used in
the following section to integrate a Leibniz algebra into a local augmented Lie rack.
First, we study the integration of a 1-cocycle in ZL1(g, as) into a Lie rack 1-cocycle in
p(G, as)s, where G is a simply connected Lie group with Lie algebra g and a a representation
ZR1
of G.
Secondly, we use the result of the first part to study the integration of a 2-cocycle in ZL2(g, aa)
into a local Lie rack 2-cocycle in ZR2
p(U, aa)s, where U is a 1-neighborhood in a simply connected
Lie group G with Lie algebra g, and a a representation of G. It is this second part that we will
use to integrate Leibniz algebras.
13
3.4.1 From Leibniz 1-cocycles to Lie rack 1-cocycles
Let G be a simply connected Lie group and a a representation of G. We want to define a
morphism I 1 from ZL1(g, as) to ZR1
p(G, as)s. For this,
we put
p(G, as)s which sends BL1(g, as) into BR1
(cid:90)
I 1(ω)(g) =
ωeq,
where ω ∈ ZL1(g, as), γ : G × [0, 1] → G is a smooth map such that γg is a path from 1 to g, γ1
is the constant path equal to 1, and ωeq is the closed left equivariant differential form in Ω1(G, a)
defined by
γg
ωeq(g)(m) = g.(ω(TgLg−1(m))).
By definition, it is clear that I 1(ω)(1) = 0.
For the moment, I 1(ω) depends on γ, but because ω is a cocycle and G is simply connected,
the dependence with respect to γ disappears.
Proposition 3.9. I 1 does not depend on γ.
Proof. Let γ, γ(cid:48) : G × [0, 1] → G such that γg(0) = γ(cid:48)
H1(G) = 0, the cycle γg − γ(cid:48)
ωeq −
g is a boundary ∂σg. So
ωeq =
ωeq =
(cid:90)
(cid:90)
(cid:90)
(cid:90)
γg
γ(cid:48)
g
γg−γ(cid:48)
g
and I 1 does not depend on γ.
g(0) = 1 and γg(1) = γ(cid:48)
g(1) = g. As
ωeq =
ddRωeq = 0,
∂σg
σg
Proposition 3.10. I 1 sends cocycles to cocycles and coboundaries to coboundaries.
Proof. First, let ω ∈ ZL1(g, as), we have
dRI(ω)(g, h) = g.I(ω)(h) − I(ω)(g(cid:66)h) − (g(cid:66)h).I(ω)(g) + I(ω)(g)
ωeq
ωeq − (g(cid:66)h).
ωeq −
ωeq +
= g.
γg(cid:66)h
γh
g.ωeq −
γg
γg
ωeq −
(g(cid:66)h).ωeq +
ωeq
γg(cid:66)h
γg
ωeq −
ωeq −
γg(cid:66)h
ωeq +
(g(cid:66)h)γg
(cid:90)
(cid:90)
(cid:90)
γg
ωeq
γg
(cid:90)
γh
gγh
(cid:90)
(cid:90)
(cid:90)
=
=
=
As H 1(G) = 0 and ∂(gγh − γg(cid:66)h − (g(cid:66)h)γg + γg) = 0, there exists γg,h : [0, 1]2 → G such that
∂γg,h = gγh − γg(cid:66)h − (g(cid:66)h)γg + γg. Hence, we have
gγh−γg(cid:66)h−(g(cid:66)h)γg+γg
dRI(ω)(g, h) =
ωeq =
ddRωeq = 0.
∂γg,h
γg,h
Hence ZL1(g, as) is sent to ZR1
p(G, as)s.
Secondly, let ω ∈ BL1(g, as). There exists β ∈ a such that ω(m) = m.β. We have
(cid:90)
(cid:90)
ddRβeq = βeq(g) − βeq(1) = g.β − β = dRβ(g).
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
ωeq.
(cid:90)
I(ω)(g) =
ωeq =
(dLβ)eq =
γg
γg
γg
Hence BL1(g, as) is sent to BR1
p(G, as)s.
14
Proposition 3.11. I 1 is a left inverse for δ1.
Proof. Let ω ∈ ZL1(g, as). Let ϕ : U → g be a local chart around 1 such that ϕ(1) = 0 and
dϕ−1(0) = id. We define for x ∈ g the smooth map αx :] − , +[→ U by setting αx(s) =
ϕ−1(sx), and we define for all s ∈] − , +[ the smooth map γαx(s) : [0, 1] → U by setting
γαx(s)(t) = ϕ−1(tsx). We have
δ1(I 1(ω))(x) =
∂
∂s
I 1(ω)(αx(s)) =
γ∗
αx(s)ωeq
[0,1]
(ϕ−1)∗ωeq(tsx)(sx)dt
(cid:12)(cid:12)(cid:12)(cid:12)s=0
(cid:12)(cid:12)(cid:12)(cid:12)s=0
[0,1]
(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)s=0
(cid:12)(cid:12)(cid:12)(cid:12)s=0
∂
∂s
∂
∂s
∂
∂s
(cid:90)
(cid:90)
=
=
=
[0,1]
[0,1]
= ω(x).
(cid:12)(cid:12)(cid:12)(cid:12)s=0
∂
∂s
(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)t=0
ωeq =
γαx (s)
γαx(s)(t))dt
(cid:90)
ωeq(γαx(s)(t))(
∂
∂t
ωeq(ϕ−1(tsx))(sx)dt =
s(ϕ−1)∗ωeq(tsx)(x)dt = ω(x)
[0,1]
∂
∂s
(cid:90)
(cid:90)
∂
∂s
(cid:12)(cid:12)(cid:12)(cid:12)s=0
(cid:12)(cid:12)(cid:12)(cid:12)s=0
dt
[0,1]
Hence δ1 ◦ I 1 = id.
Remark 3.12. In fact, I 1(ω) is also a Lie group 1-cocycle. Indeed, the formula to define I 1(ω)
is the same as the one defined by K.H. Neeb in Section 3 of [Nee04], and in this article he shows
that I 1(ω) is a group cocycle. The following calculation shows that this group cocycle identity
satisfied by I 1(ω) implies the rack cocycle identity satisfied by I 1(ω). Indeed, I 1(ω) is a group
cocycle, thus we have
dGpI 1(ω)(g, h) − dGpI 1(ω)(g(cid:66)h, g) = 0.
Moreover dGpI 1(ω)(g, h) − dGpI 1(ω)(g(cid:66)h, g) = dRI 1(ω)(g, h), thus dRI 1(ω)(g, h) = 0, and we
see clearly that the rack cocycle identity is implied by the group cocycle identity. We will use
this remark in Proposition 3.19.
3.4.2 From Leibniz 2-cocycles to Lie local rack 2-cocycles
Let G be a simply connected Lie group, let U be a 1-neigbourhood in G such that log is defined
on U and let a be a representation of G. In Proposition 3.8 we have defined for all n ∈ N the
maps
HRn
s (U, aa)
[δn]−→ HLn(g, aa).
In the next section, we will see that a Leibniz algebra can be integrated into a local Lie rack
since the morphism [δ2] is surjective. More precisely, if we can construct a left inverse for [δ2],
then it gives us an explicit method to construct the local Lie rack which integrates the Leibniz
algebra.
In this section, we define a morphism [I 2] from HL2(g, aa) to HR2
s(U, aa), and we show that
it is a left inverse for [δ2]. To construct the map [I 2], we adapt an integration method of Lie
algebra cocycles into Lie group cocycles by integration over simplex. This method is due to W.T.
Van Est ([vE58]) and used by K.H. Neeb ([Nee02, Nee04]) for the infinite dimensional case.
Definition of I 2
We want to define a map from ZL2(g, aa) to ZR2
p(U, aa)s.
BR2
bra g with coefficients in a symmetric module as.
p(U, aa)s such that BL2(g, aa) is sent to
In the previous section, we have integrated a Leibniz 1-cocycle on a Lie alge-
In Proposition 1.1, we have shown that
15
there is an isomorphism between CL2(g, aa) and CL1(g, Hom(g, a)s), which sends ZL2(g, aa) to
ZL1(g, Hom(g, a)s) and BL2(g, aa) to BL1(g, Hom(g, a)s). Hence, we can define a map
I : ZL2(g, aa) → ZR1
p(G, Hom(g, a)s)s,
which sends BL2(g, aa) into BR1
p(G, Hom(g, a)s)s. This is the composition
ZL2(g, aa) τ 2→ ZL1(g, Hom(g, a)s) I 1→ ZR1
p(G, Hom(g, a)s)s.
Now, we want to define a map from ZR1
β has values in the representation Hom(g, a), so for all g ∈ G, we can consider the equivariant
differential form β(g)eq ∈ Ω1(G, a) defined by
p(G, Hom(g, a)s)s to ZR2
p(G, Hom(g, a)s)s,
p(U, aa). Let β ∈ CR1
β(g)eq(h)(m) := h.(β(g)(ThLh−1 (m))).
Then we define an element in CR2
p(U, aa) by setting
f (g, h) =
(β(g))eq,
γg(cid:66)h
where γ : G × [0, 1] → G is a smooth map such that for all g ∈ G, γg is a path from 1 to g in G
and γ1 = 1.
For the moment, an element of ZR1
ZR2
p(U, aa)s. To reach our goal, we have to specify the map γ, and we define it by setting
p(G, Hom(g, a)s)s is not necessarily sent to an element of
Then, we define I 2 : ZL2(g, aa) → CR2
Section 2.4)
γg(s) = exp(s log(g)).
p(U, aa)s by setting for all (g, h) ∈ U2−loc (cf. notation in
I 2(ω)(g, h) =
(I(ω)(g))eq.
By definition, it is clear that I 2(ω)(g, 1) = I 2(ω)(1, g) = 0.
Properties of I 2
Proposition 3.13. I 2 sends ZL2(g, aa) into ZR2
p(U, aa)s.
Proof. Let ω ∈ ZL2(g, aa) and (g, h, k) ∈ U3−loc. We have
dR(I 2(ω))(g, h, k) = g.I 2(ω)(h, k) − I 2(ω)(g(cid:66)h, g(cid:66)k) − (g(cid:66)h).I 2(ω)(g, k) + I 2(ω)(g, h(cid:66)k)
(cid:90)
g.((I(ω)(h))eq) −
(I(ω)(g(cid:66)h))eq −
(g(cid:66)h).((I(ω)(g))eq)
γg(cid:66)(h(cid:66)k)
γg(cid:66)k
(cid:90)
(cid:90)
(cid:90)
(cid:90)
=
+
γh(cid:66)k
γg(cid:66)(h(cid:66)k)
(I(ω)(g))eq.
For all g ∈ G we have g.(ωeq) = c∗
g((g.ω)eq), thus
dR(I 2(ω))(g, h, k) =
+
cg◦γh(cid:66)k
γg(cid:66)(h(cid:66)k)
(g.I(ω)(h)eq −
I(ω)(g)eq.
(cid:90)
(cid:90)
γg(cid:66)h
(cid:90)
(cid:90)
γg(cid:66)(h(cid:66)k)
16
I(ω)(g(cid:66)h)eq −
(cid:90)
cg(cid:66)h◦γg(cid:66)k
((g(cid:66)h).I(ω)(g))eq
(cid:90)
(cid:90)
(cid:90)
(cid:90)
(cid:90)
βeq −
γh
γg(cid:66)h
By naturality of the exponantial and the logarithm, we have for all (g, h) ∈ U2−loc γg(cid:66)h = g (cid:66) γh,
thus
dR(I 2(ω))(g, h, k) =
Hence ZL2(g, aa) is sent to ZR2
p(U, aa)s.
dR(I(ω))(g, h) = 0.
γg(cid:66)(h(cid:66)k)
Proposition 3.14. I 2 sends BL2(g, aa) into BR2
Proof. Let ω ∈ BL2(g, aa), there exists an element β ∈ CL1(g, aa) such that ω = dLβ. By
(τ 2(dLβ))eq, and because {τ n}n∈N is a morphism of cochain complexes
definition I(ω)(g) =
p(U, aa)s.
(cid:90)
γg
I(ω)(g) = I 1(τ 2(ω))(g) =
(τ 2(ω))eq =
(τ 2(dLβ))eq.
γg
γg
Let (g, h) ∈ U2−loc. Using the same kind of computation as in the proof of Proposition 3.13 we
find
I2(ω)(g, h) =
(I(ω)(g))eq = g.
βeq = dR(I 1(β))(g, h)
(cid:90)
γg(cid:66)h
Hence BL2(g, aa) is sent to BR2
p(U, a)s.
Proposition 3.15. I 2 is a left inverse for δ2.
Proof. Let x, y ∈ g, and Ix (resp Iy) be an interval in R such that x(s) = exp(sx) (resp
y(s) = exp(sy)) be defined for all s ∈ Ix (resp for all s ∈ Iy). The map x(cid:66)y : Ix × Iy → G
is continuous, thus there exists W an open subset of Ix × Iy such that (x(cid:66)y)(W ) ⊆ U. Hence
there exists an interval J ⊆ Ix ∩ Iy such that x(s)(cid:66)y(t) ∈ U for all (s, t) ∈ J × J.
We have to prove
δ2 ◦ I 2 = id.
x(s)(I(ω)(x(s)))eq
δ2(I 2(ω))(x, y) =
I 2(ω)(x(s), y(s)) =
Let ω ∈ ZL2(g, aa). By definition
(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)t=0
∂2
∂s∂t
=
∂
∂s
(cid:12)(cid:12)t=0
(
(cid:12)(cid:12)(cid:12)(cid:12)s,t=0
(cid:12)(cid:12)(cid:12)(cid:12)s=0
(cid:82)
(cid:12)(cid:12)(cid:12)(cid:12)t=0
γy (t)
First, we compute ∂
∂t
and βt = γy(t). We have
(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)t=0
α =
∂
∂t
∂
∂t
where ft(r) = α(βt(r))(β(cid:48)
We have
β
t(r)).
(cid:12)(cid:12)(cid:12)(cid:12)t=0
∂
∂t
ft(r) = (
[0,1]
(cid:12)(cid:12)(cid:12)(cid:12)t=0
∂
∂t
(cid:12)(cid:12)(cid:12)(cid:12)s,t=0
∂2
∂s∂t
(cid:90)
γx (s)(cid:66)y (t)
(I(ω)(x(s)))eq
γy (t)
c∗
x(s)(I(ω)(x(s)))eq).
∂
∂t
x(s)(I(ω)(x(s)))eq. For the sake of clarity, we put α = c∗
c∗
(cid:90)
(cid:90)
(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)t=0
β∗α =
∂
∂t
ft(r)dr =
[0,1]
[0,1]
ft(r)dr,
(cid:12)(cid:12)(cid:12)(cid:12)t=0
∂
∂t
(cid:12)(cid:12)(cid:12)(cid:12)t=0
∂
∂t
β(cid:48)
t(r)).
α(βt(r)))β(cid:48)
0(r) + (α(β0(r)))(
17
(cid:12)(cid:12)t=0 β(cid:48)
(cid:90)
γx(s)
(cid:12)(cid:12)(cid:12)(cid:12)s=0
∂
∂s
Moreover, α(β0(r)) = α(1), β(cid:48)
α(1)(y) and δ2(I 2(ω))(x, y) = ∂
∂s
0(r) = 0, and ∂
∂t
t(r) = y. So ∂
∂t
[0,1] α(1)(x)dr =
x(s)(I(ω)(x(s)))eq)(1)(y). Furthermore we have
(cid:12)(cid:12)s=0 (c∗
c∗
x(s)(I(ω)(x(s))eq)(1)(y) = (I(ω)(x(s)))eq(cx(s)(1))(Adx(s)(y))
(cid:12)(cid:12)t=0
(cid:82)
β α =(cid:82)
= I(ω)(x(s))(Adx(s)(y))
= (
τ 2(ω)eq)(Adx(s)(y)).
τ 2(ω)eq = σ(s) and Adx(s)(y) = λ(s), we have
If we put(cid:82)
(cid:12)(cid:12)(cid:12)(cid:12)s=0
∂
∂s
γx (s)
(cid:90)
((
τ 2(ω)eq)(Adx(s)(y))) =
γx (s)
We have σ(0) = 0, λ(0) = y, and σ(cid:48)(0) = τ 2(ω)(x). Thus
(cid:16)(cid:0)(cid:90)
(cid:12)(cid:12)(cid:12)(cid:12)s=0
∂
∂s
γx(s)
τ 2(ω)eq(cid:1)(cid:0)Adx(s)(y)(cid:1)(cid:17)
= τ 2(ω)(x)(y).
σ(s)(λ(s)) = σ(cid:48)(0)(λ(0)) + σ(0)(λ(cid:48)(0)).
Hence δ2(I 2(ω))(x, y) = ω(x, y).
Remark 3.16. Suppose that we have a Leibniz 2-cocycle ω which is also a Lie 2-cocycle. In
this case, we can integrate ω into a local Lie rack cocycle, but also into a local Lie group cocycle
(cf. Introduction). Then it is natural to ask if the two constructions are related to each other.
Proposition 3.17. Let G be a Lie group, g be its Lie algebra, a be a representation of G,
ω ∈ Λ2(g, a) and γ1, γ2 smooth paths in G pointed in 1. Then
(cid:90)
(cid:0)(cid:90)
(τ 2(ω))eq(cid:1)eq
(cid:90)
=
ωeq
γ1
γ2
γ1γ2
ωeq(γ1γ2)(
∂
∂s
γ1(t)γ2(s),
∂
∂t
[0,1]2
∂
∂s
γ2(s)), Adγ2(s)−1 (dγ1(t)Lγ1(t)−1(
γ1(t)γ2(s))dsdt,
γ1(t)))(cid:1)dsdt.
∂
∂t
(4)
where γ1γ2 : [0, 1]2 → G; (s, t) (cid:55)→ γ1(t)γ2(s).
Proof. On the one hand, we have
(cid:90)
(cid:90)
(cid:90)
and this expression is equal to
(cid:90)
(cid:0)(cid:90)
[0,1]2
ωeq =
(γ1γ2)∗ωeq =
γ1γ2
[0,1]2
γ1(t)γ2(s).ω(cid:0)dγ2(s)Lγ2(s)−1(
(cid:90)
(τ 2(ω))eq(cid:1)eq
=
γ∗
1 (
[0,1]
γ2
On the other hand, we have
γ1
γ2
=
(cid:90)
(cid:90)
(cid:90)
[0,1]
and this expression is equal to
(cid:90)
(cid:90)
(cid:90)
(τ 2(ω)eq)eq)
γ1(t).(
[0,1]
(τ 2(ω))eq(γ2(s))(
∂
∂s
γ2(s)))(dγ1(t)Lγ1(t)−1(
∂
∂t
γ1(t)))dt.
γ1(t).(
[0,1]
[0,1]
γ2(s).ω(dγ2(s)Lγ2(s)−1(
γ2(s)), Adγ2(s)−1(·))ds)(dγ1(t)Lγ1(t)−1(
∂
∂s
∂
∂t
γ1(t)))dt.
Using the Fubini theorem we see that this expression is equal to (4).
18
If we apply this result to the case where γ1(s) = γg(cid:66)h(s) = exp(s log(g(cid:66)h)) and γ2(s) = γg(s) =
exp(s log(g)) for (g, h) ∈ U2−loc, then we obtain the folowing corollary.
Corollary 3.18. If ω ∈ ZL2(g, aa) ∩ Z 2(g, a), then for all g, h ∈ U2−loc such that gh ∈ U2−loc
we have
I 2(ω)(g, h) = ι2(ω)(g, h) − ι2(ω)(g(cid:66)h, g),
(5)
with
ι2(ω)(g, h) =
ωeq,
(cid:90)
and where γg,h is a smooth singular 2-chain in G such that ∂γg,h = γg − γgh + gγh.
γg,h
We can remark that I 2 is more than a local Lie rack cocycle. Precisely, if ω is in ZL2(g, aa)
then the local rack cocycle identity satisfied by I 2(ω), comes from another identity satisfied by
I 2(ω). Indeed, I 2 is defined using I, and to verify that I 2 sends Leibniz cocycles into local rack
cocycles, we have used Proposition 3.10. This proposition establishes that I 1 sends Lie cocycles
into rack cocycle. But, we have remarked (Remark 3.12) that the rack cocycle identity satisfied
by I 1(ω), comes from the group cocycle identity. Hence, we can think that we forgot structure
on I 2(ω). The following proposition points out the identity satisfied by I 2(ω) which induced the
local rack identity.
Proposition 3.19. If ω ∈ ZL2(g, aa), then I 2(ω) satisfies the identity
g.I 2(ω)(h, k) − I 2(ω)(gh, k) + I 2(ω)(g, h(cid:66)k) = 0, ∀(g, h, k) ∈ U3−loc.
Moreover, this identity induces the local rack cocycle identity.
Proof. Let ω ∈ ZL2(g, aa) and (g, h, k) ∈ U3−loc. Using the same kind of computation as in the
proof of Proposition 3.13 we find :
g.I 2(ω)(h, k) − I 2(ω)(gh, k) + I 2(ω)(g, h(cid:66)k) =
d(I(ω))(g, h) = 0.
γg(cid:66)(h(cid:66)k)
Let (g, h, k) ∈ U3−loc. Denote the expression g.I 2(ω)(h, k) − I 2(ω)(gh, k) + I 2(ω)(g, h(cid:66)k) by
b(I 2(ω))(g, h, k)). Inserting −I 2(ω)(gh, k) + I 2(ω)(gh, k) in the formula for dR(I 2(ω))(g, h, k),
we find
dR(I 2(ω))(g, h, k) = b(I 2(ω))(g, h, k)) − b(I 2(ω)(g (cid:66) h, g, k)).
We will see in the next section that this identity makes it possible to integrate a Leibniz
algebra into a local augmented Lie rack.
3.5 From Leibniz algebras to local Lie racks
In this section, we present the main theorem of this article. In Proposition 3.1 we have seen
that the tangent space at 1 of a (local) Lie rack is provided with a Leibniz algebra structure.
Conversely, we now show that every Leibniz algebra can be integrated into an augmented local
Lie rack. Our construction is explicit, and by this construction, a Lie algebra is integrated into
a Lie group. Conversely, we show that an augmented local Lie rack whose tangent space at 1 is
a Lie algebra is necessarily a (local) Lie group. That is, there is a structure of Lie group on this
19
(cid:90)
augmented local Lie rack, and the conjugation on the augmented local Lie rack is the conjugation
in the group.
The idea of the proof is simple and uses the knowledge of the Lie's first theorem and Lie's
second theorem. Let g be a Leibniz algebra. First, we decompose the vector space g into a direct
sum of Leibniz algebras g0 and a that we know how to integrate. As we know the theorem for Lie
subalgebras of endomorphisms of a finite dimensional vector space V , the factors are integrable
if g is isomorphic (as a vector space) to an abelian extension of a Lie subalgebra g0 of End(V )
by a g0-representation a. Hence g is isomorphic to a ⊕ω g0, the Leibniz algebra a is abelian so
becomes integrated into a, and g0 is a Lie subalgebra of End(V ) so becomes integrated into a
simply connected Lie subgroup G0 of Aut(V ). Now, we have to understand how to patch a and
G0. That is, we have to understand how the gluing data ω becomes integrated into a gluing data
f between a and G0. It is the local Lie rack cocycle I 2(ω), constructed in the preceding section,
which answers this question. Hence, we showed that a Leibniz algebra g becomes integrated into
a local Lie rack of the form a ×f G0.
Let g be a Leibniz algebra, there are several ways to see g as an abelian extension of a Lie
subalgebra g0 of End(V ) by a g0-representation a. Here, we take the abelian extension associated
to the (left) center of g. By definition the left center is
ZL(g) = {x ∈ g [x, y] = 0 ∀y ∈ g}.
The left center ZL(g) is an ideal in g and we can consider the quotient of g by ZL(g). By
definition, ZL(g) is the kernel of the adjoint representation adL : g → End(g), x (cid:55)→ [x,−]. Thus
this quotient is isomorphic to a Lie subalgebra of End(g). We denote this quotient by g0. Hence,
to a Leibniz algebra g there is a canonical abelian extension given by
ZL(g)
i
(cid:44)→ g
p(cid:16) g0.
This extension gives a structure of g0-representation to ZL(g), and by definition of ZL(g), this
representation is anti-symmetric. The equivalence class of this extension is characterised by a
cohomology class in HL2(g0, ZL(g)). Hence there is ω ∈ ZL2(g0, ZL(g)) such that the abelian
extension ZL(g)
p(cid:16) g0 is equivalent to
i
(cid:44)→ g
ZL(g)
i
(cid:44)→ g0 ⊕ω ZL(g)
π(cid:16) g0.
Here g0 is a Lie subalgebra of End(g), so becomes integrated into a simply connected Lie subgroup
G0 of Aut(g), and ZL(g) is an abelian Lie algebra, so becomes integrated into itself. ZL(g) is
a g0-representation (in the sense of Lie algebra) and G0 is simply connected, thus by the Lie's
second theorem, ZL(g) is a smooth G0-module (in the Lie group sense) and we can provide
ZL(g) with an anti-symmetric smooth G0-module structure. The cocycle ω ∈ ZL2(g, ZL(g))
becomes integrated into the local Lie rack cocycle I 2(ω) ∈ ZR2
p(G0, ZL(g))s, and we can put on
the cartesian product G0 × ZL(g) a structure of local Lie rack by setting
(g, a)(cid:66)(h, b) = (g(cid:66)h, φg,h(b) + ψg,h(a) + I 2(ω)(g, h)),
where φg,h(b) = g.b and ψg,h(a) = 0. That is we have
(g, a)(cid:66)(h, b) = (g(cid:66)h, g.b + I 2(ω)(g, h))).
It is clear by construction that this local Lie rack has its tangent space at 1 provided with a
Leibniz algebra structure isomorphic to g. Finally, we have shown the following theorem.
20
Theorem 3.20. Every Leibniz algebra g can be integrated into a local Lie rack of the form
G0 ×I 2(ω) aa,
with conjugation
(g, a)(cid:66)(h, b) = (g(cid:66)h, g.b + I 2(ω)(g, h)),
(6)
and neutral element (1, 0), where G0 is a Lie group, a a G0-module and ω ∈ ZL2(g0, aa). Con-
versely, every local Lie rack of this form has its tangent space at 1 provides with a Leibniz algebra
structure.
We ask more in our original problem.
Indeed, we ask that, using the same procedure, a
Lie algebra becomes integrated into a Lie group. That is, we have to show that when g is a
Lie algebra, then G0 × ZL(g) is provided with a Lie group structure, and the conjugation on
G0 ×I 2(ω) ZL(g) is induced by the rack product in Conj(G0 × ZL(g)).
i
(cid:44)→ g
Let g be a Lie algebra, the left center ZL(g) is equal to the center Z(g). The abelian
p(cid:16) g0 provides ZL(g) with an anti-symmetric structure but also a symmetric
extension ZL(g)
structure, so a trivial structure. This extension becomes a central extension and the cocycle
ω ∈ ZL2(g0, Z(g)) is also in Z 2(g0, Z(g)). On the hand, with ω we can construct a local Lie rack
cocycle I 2(ω), and on the other hand, we can construct a Lie group cocycle ι2(ω). Hence, using
the formula (5) relating I 2(ω) and ι2(ω), the conjugation in G0 ×I 2(ω) Z(g) can be written
(g, a)(cid:66)(h, b) = (g(cid:66)h, I 2(ω)(g, h)) = (g(cid:66)h, ι2(ω)(g, h) − ι2(ω)(g(cid:66)h, g)),
and a easy calculation shows that this is the formula for the conjugation in the group G0 ×ι2(ω)
Z(g), where the product is defined by
(g, a)(h, b) = (gh, ι2(g, h)).
Conversely, suppose that a local Lie rack of the form G0 ×I 2(ω) aa has its tangent space at
1, g0 ⊕ω aa, provided with a Lie algebra structure. Necessarily, a is a trivial g0-representation
and ω ∈ Z 2(g0, a). Hence, as before we have the formula (5) relating I 2(ω) and ι2(ω) and the
conjugation defined by the formula (6) is induced by the conjugation coming from the group
structure on G0 ×ι2(ω) a. Finally, we have the following refinement of Theorem 3.20.
Theorem 3.21. Every Leibniz algebra g can be integrated into a local Lie rack of the form
G0 ×I 2(ω) aa,
with conjugation
(g, a)(cid:66)(h, b) = (g(cid:66)h, g.b + I 2(ω)(g, h)),
(7)
and neutral element (1, 0), where G0 is a Lie group, a a representation of G0 and ω ∈ ZL2(g0, aa).
Conversely, every local Lie rack of this form has its tangent space at 1 provided with a Leibniz
algebra structure.
Moreover, in the special case where g is a Lie algebra, the above construction provides
G0 ×I 2(ω) aa with a rack product coming from the conjugation in a Lie group. Conversely, if
the tangent space at 1 of G0 ×I 2(ω) aa is a Lie algebra, then G0 ×I 2(ω) aa can be provided with a
Lie group structure, and the conjugation induced by the Lie group structure is the one defined by
(7).
21
3.6 From Leibniz algebras to local augmented Lie racks
Let g0 be a Lie algebra, a a g-representation and ω ∈ ZL2(g0, aa).
showed that I 2(ω) is a local Lie rack cocycle. We showed also that it satisfies the identity
In Proposition 3.13, we
g.I 2(ω)(h, k) − I 2(ω)(gh, k) + I 2(ω)(g, h(cid:66)k) = 0
(8)
for all (g, h, k) ∈ U3−loc. The natural question is to know which algebraic structure on G0×I 2(ω)aa
is encoded by this identity. We will see that the answer is the structure of a local augmented Lie
rack.
Definition 3.22. Let G be a group. A local G-set is a set X provides with a map ρ defined on
a subset Ω of G × X with values in X such that the followings axioms are satisfied
1. If (h, x), (gh, x), (g, ρ(h, x)) ∈ Ω, then ρ(g, ρ(h, x)) = ρ(gh, x).
2. For all x ∈ X, we have (1, x) ∈ Ω and ρ(1, x) = x.
A local topological (resp.(smooth)) G-set is a topological set (resp. smooth manifold) X with
a structure of a local G-set where Ω is an open subset of X and ρ : Ω → X is continuous
(resp. smooth). A fixed point is an element x0 ∈ X such that for all g ∈ G, (g, x0) ∈ Ω and
ρ(g, x0) = x0.
In the following proposition, we show that the identity (8) provides G0 ×I 2(ω) aa with a
structure of a local G0-set.
Proposition 3.23. G0 ×I 2(ω) aa is a local smooth G0-set, and (1, 0) is a fixed point.
Proof. We define an open subset Ω and a smooth map ρ by
1. Ω = {(g, (h, b)) ∈ G0 × (G0 ×I 2(ω) aa)(g, h) ∈ U2−loc}.
2. ρ(g, (h, b)) = (g(cid:66)h, g.b + I 2(ω)(g, h)).
Let (h, (k, z)), (gh, (k, z)), (g, ρ(h, (k, z))) ∈ Ω. This is equivalent to the condition (h, k), (gh, k), (g, h(cid:66)k) ∈
U2−loc, that is (g, h, k) ∈ U3−loc. We have
ρ(g, ρ(h, (k, z))) = (g(cid:66)(h(cid:66)k), g.(h.z) + g.I 2(ω)(h, k) + I 2(ω)(g, h(cid:66)k)).
Using the identities (8) and (gh)(cid:66)k = g(cid:66)(h(cid:66)k), we have
ρ(g, ρ(h, (k, z))) = ((gh)(cid:66)k, (gh).z + I 2(ω)(gh, k)) = ρ(gh, ρ(k, z)).
Moreover, ρ(1, (k, z)) = (1(cid:66)k, 1.z+I 2(ω)(1, k)) = (k, z) and ρ(g, (1, 0)) = (g(cid:66)1, g.0+I 2(ω)(g, 1)) =
(1, 0). Hence G0 ×I 2(ω) aa is a local smooth G0-set and (1, 0) is a fixed point for this local ac-
tion.
We remark that we can reconstruct the rack product in G0 ×I 2(ω) aa from the formula of the
G0-action. Indeed, we have (g, a)(cid:66)(h, b) = g.(h, b) = p(g, a).(h, b) where p is the projection on
the first factor G0 ×I 2(ω) aa
p(cid:16) G0. Because p(1, 0) = 1 and p is equivariant we have shown the
following proposition.
Proposition 3.24. G0 ×I 2(ω) aa
p(cid:16) G0 is a local augmented Lie rack.
Hence we can rewrite our main theorem
22
Theorem 3.25. Every Leibniz algebra g can be integrated into a local augmented Lie rack of the
form
G0 ×I 2(ω) aa
p(cid:16) G0,
with local action
g.(h, b) = (g(cid:66)h, g.b + I 2(ω)(g, h)),
and neutral element (1, 0), where G0 is a Lie group, a a representation of G0 and ω ∈ ZL2(g0, aa).
Conversely, every local augmented Lie rack of this form has its tangent space at 1 provided with
a Leibniz algebra structure.
Moreover, in the special case where g is a Lie algebra, the above construction provides
G0 ×I 2(ω) aa with a rack product coming from the conjugation in a Lie group. Conversely, if
the tangent space at 1 of G0 ×I 2(ω) aa is a Lie algebra, then G0 ×I 2(ω) aa can be provided with a
Lie group structure, and the conjugation induced by the Lie group structure is the one defined by
(7).
3.7 Example of a non-split Leibniz algebra integration
In this section we construct the Lie rack associated to a Leibniz algebra of dimension 5 by
following the method explained above. Other examples of integration in dimension 4 can be
found in [Cov10].
Let g = R5. We define a bilinear map on g by
[e1, e1] = [e1, e2] = e3
[e2, e1] = [e2, e2] = [e1, e3] = e4
[e1, e4] = [e2, e3] = e5
We have [(x1, x2, x3, x4), (y1, y2, y3, y4)] = (0, 0, x1(y1 + y2), x2(y1 + y2) + x1y3, x1y4 + x2y3) and
(g, [−,−]) is a Leibniz algebra.
To follow the integration method explained above, we have to determine the left center ZL(g),
the quotient of g by ZL(g) denoted g0, the action of g0 on ZL(g) and the Leibniz 2-cocycle
describing the abelian extension ZL(g) (cid:44)→ g (cid:16) g0.
Let x ∈ ZL(g), for y = (0, 0, 1, 0, 0) in g, we have [x, y] = 0. This implies that x1 = x2 = 0.
Conversely, every element in g with the first two coordinates equal to 0 is in ZL(g). Hence
ZL(g) =< e3, e4, e5 > and g0 (cid:39)< e1, e2 >. The bracket on g0 is equal to zero, hence g0 is an
abelian Lie algebra. The action of g0 on ZL(g) is given by
ρx(y) = [(x1, x2, 0, 0, 0), (0, 0, y3, y4, y5)] = (0, 0, 0, x1y3, x1y4 + x2y3),
and the Leibniz 2-cocycle is given by
ω(x, y) = [(x1, x2, 0, 0, 0), (y1, y2, 0, 0, 0)] = (0, 0, x1(y1 + y2), x2(y1 + y2), 0).
Moreover, we have [x, x] = (0, 0, x1(x1 + x2), x2(x1 + x2) + x1x3, x1x4 + x2x3), hence taking
x = (1, 0, 0, 0, 0), (0, 1, 0, 0, 0) and (0, 1, 1, 0, 0), we see easily that gann = ZL(g). This Leibniz
algebra is not split because for α ∈ Hom(g, ZL(g)) and x, y ∈ g0, we have dLα(x, y) = ρx(α(y)) =
(0, 0, 0, x1α(y)3, x1α(y)4 + x2α(y)3).
Now, we have to determine the Lie group G0 associated to g0, the action of G0 on ZL(g)
integrating ρ : g0 → End(ZL(g)) (the action of g0 on ZL(g)), and the Lie rack cocycle integrating
ω.
23
.
0
0
0
0
x1
0
x2 x1
0
1
(cid:90)
(cid:90)
γb
γa
ρx =
(cid:90)
(cid:90)
[0,1]
[0,1]
The Lie algebra g0 is abelian, thus a Lie group integrating g0 is G0 = g0. To integrate the
action ρ, we use the exponential exp : End(ZL(g)) → Aut(ZL(g)). Indeed, for all x ∈ g0, we
have
Hence, we define a Lie group morphism φ : G0 → Aut(ZL(g)) by setting
φx = exp(ρx) =
0
1
1 x1
x1
x2 + 1
2 x2
.
0
0
0
It is easy to see that d1φ = ρ. What remains to be done is the integration of the cocycle ω. A
formula for f, a Lie rack cocycle integrating ω, is
where γa(s) = sa and γb(t) = tb. Let a ∈ G0 and x, y ∈ g0. We have
f (a, b) =
(
τ 2(ω)eq)eq,
(cid:12)(cid:12)(cid:12)(cid:12)s=0
γa(s))ds
(cid:90)
γa
Thus
We have
φγb(t)(
Thus
γa
τ 2(ω)eq =
=
τ 2(ω)eq(γa(s))(
∂
∂s
φγa(s) ◦ τ 2(ω)(a)ds
τ 2(ω)eq =
a1
1
2 a2
1 + a2
a1a2 + 1
6 a3
a1
1 + a2
1 a1a2 + 1
6 a3
2 a2
1
1
f (a, b) =
(
τ 2(ω)eq)eq =
γb
γa
and
φγb(t)(
a1
1
2 a2
1 + a2
a1a2 + 1
6 a3
a1
1 + a2
1 a1a2 + 1
6 a3
2 a2
1
1
1
tb1
tb2 + 1
2 (tb1)2
0
1
tb1
(cid:90)
(b)) =
[0,1]
=
(b))dt.
a1
1
2 a2
1 + a2
a1a2 + 1
6 a3
a1
1 + a2
1 a1a2 + 1
6 a3
2 a2
1
1
0
0
0
1
a1
1 + a2
1 a1a2 + 1
6 a3
2 a2
1
a1
1
2 a2
1 + a2
a1a2 + 1
6 a3
a1(b1 + b2)
2 a2
(cid:19)
(cid:18)b1
b2
.
(a1a2 + 1
6 a3
1 + 1
(tb1a1 + a2 + 1
2 tb1a2
1 + tb2a1 + tb1a2 + 1
1)(b1 + b2)
2 (tb1)2a1)(b1 + b2)
a1(b1 + b2)
2 a2
2 b2a1 + 1
1)(b1 + b2)
2 b1a2 + 1
6 (b1)2a1)(b1 + b2)
.
(cid:90)
(cid:90)
(cid:90)
6 a3
1 + 1
f (a, b) =
(a1a2 + 1
2 b1a1 + a2 + 1
( 1
1 + 1
4 b1a2
and the conjugation in G0 ×f ZL(g) = R5 is given by
a1
a2
a3
a4
a5
=
(cid:66)
a1b3 + b4 + ( 1
1)b3 + a1b4 + b5 + (a1a2 + 1
b1
b2
b3
b4
b5
(a2 + 1
2 a2
b1
b2
24
b3 + a1(b1 + b2)
2 b1a1 + a2 + 1
2 a2
1)(b1 + b2)
1 + 1
2 b2a1 + 1
6 a3
4 b1a2
1 + 1
2 b1a2 + 1
6 (b1)2a1)(b1 + b2)
.
References
[AG03] N. Andruskiewitsch and M. Graña. From racks to pointed Hopf algebras. Adv. in Math.,
178 (2):177 -- 243, 2003.
[Car30] E. Cartan. Le troisième théorème fondamental de Lie. C.R. Acad. Sc. T., 190:914 -- 1005,
1930.
[Cov10] S. Covez. L'intégration locale des algèbres de Leibniz. 2010. PhD thesis. Available at
http://tel.archives-ouvertes.fr/tel-00495469/.
[FR92] Roger Fenn and Colin Rourke. Racks and links in codimension two. J. Knot Theory
Ramifications, 1(4):343 -- 406, 1992.
[Jac05] Nicholas Jackson. Extensions of racks and quandles. Homology Homotopy Appl.,
7(1):151 -- 167, 2005.
[Kin07] Michael K. Kinyon. Leibniz algebras, Lie racks, and digroups. J. Lie Theory, 17(1):99 --
114, 2007.
[Lod93] Jean-Louis Loday. Une version non commutative des algèbres de Lie:
les algèbres de
Leibniz. In R.C.P. 25, Vol. 44 (French) (Strasbourg, 1992), volume 1993/41 of Prépubl.
Inst. Rech. Math. Av., pages 127 -- 151. Univ. Louis Pasteur, Strasbourg, 1993.
[Lod98] Jean-Louis Loday. Cyclic homology, volume 301 of Grundlehren der Mathematischen
Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag,
Berlin, second edition, 1998. Appendix E by María O. Ronco, Chapter 13 by the author
in collaboration with Teimuraz Pirashvili.
[LP93] Jean-Louis Loday and Teimuraz Pirashvili. Universal enveloping algebras of Leibniz
algebras and (co)homology. Math. Ann., 296(1):139 -- 158, 1993.
[LP98] J. L. Loday and T. Pirashvili. The tensor category of linear maps and Leibniz algebras.
Georgian Math. J., 5(3):263 -- 276, 1998.
[Nee02] Karl-Hermann Neeb. Central extensions of infinite-dimensional Lie groups. Ann. Inst.
Fourier (Grenoble), 52(5):1365 -- 1442, 2002.
[Nee04] Karl-Hermann Neeb. Abelian extensions of infinite-dimensional Lie groups. In Travaux
mathématiques. Fasc. XV, Trav. Math., XV, pages 69 -- 194. Univ. Luxemb., Luxem-
bourg, 2004.
[Smi51] P. A. Smith. The complex of a group relative to a set of generators. II. Ann. of Math.
(2), 54:403 -- 424, 1951.
[Smi52] P. A. Smith. Some topological notions connected with a set of generators. In Proceedings
of the International Congress of Mathematicians, Cambridge, Mass., 1950, vol. 2, pages
436 -- 441, Providence, R. I., 1952. Amer. Math. Soc.
[vE58] W. T. van Est. A group theoretic interpretation of area in the elementary geometries.
Simon Stevin, 32:29 -- 38, 1958.
[vE62a] W. T. van Est. Local and global groups. I. Nederl. Akad. Wetensch. Proc. Ser. A 65 =
Indag. Math., 24:391 -- 408, 1962.
[vE62b] W. T. van Est. Local and global groups. II. Nederl. Akad. Wetensch. Proc. Ser. A 65
= Indag. Math., 24:409 -- 425, 1962.
25
|
1108.0953 | 1 | 1108 | 2011-08-03T22:24:41 | The Clifford Twist | [
"math.RA"
] | Gives an elementary exposition of the twisted group algebra rep- resentation of simple Clifford algebras | math.RA | math | THE CLIFFORD TWIST
JOHN W. BALES
Abstract. This is an elementary exposition of the twisted group algebra rep-
resentation of simple Clifford algebras.
g
u
A
3
]
.
A
R
h
t
a
m
[
1
v
3
5
9
0
.
8
0
1
1
:
v
i
X
r
a
1. Clifford Algebra
Clifford Algebra is an algebra defined on a potentially infinite set e1, e2, e3, · · ·
of linearly independent unit vectors, their finite products (called multi-vectors) and
the unit scalar 1 (denoted e0 ). Every element of the algebra is a linear combination
of these basis elements over some ring, usually the real numbers.
The vectors e1, e2, e3, · · · are referred to as '1-blades.' A product of two vectors
is called a '2-blade.' three vectors a '3-blade' and so forth. The scalar e0 is a
'0-blade.' An n -blade multi-vector is said to be of grade n.
There are four fundamental multiplication properties of 1-blades.
(1) The square of 1-blades is µ (where µ2 = 1 ).
(2) The product of 1-blades is anti-commutative.
(3) The product of 1-blades is associative.
(4) Every n -blade can be factored into the product of n distinct 1-blades.
The product of ei and ej is denoted eij if i < j and by −eij if i > j. Likewise
for higher order blades. For example, if i < j < k then eiejek = eijk.
Any two n -blades may be multiplied by first factoring them into 1-blades. For
example, the product of e134 and e23, is computed as follows:
e134e23 = e1e3e4e2e3
= −e1e4e3e2e3
= e1e4e2e3e3
= µe1e4e2
= −µe1e2e4
= −µe124
2. Representing Clifford Algebra as a twisted group algebra
Each of the basis elements of Clifford algebra 1, e1, e2, e12, e3, · · · can be associ-
ated with an element of the set G of non-negative integers.
Each vector ek is associated with the integer 2k−1 and the scalar e0 is associated
with 0. A multi-vector is associated with the sum of the integers associated with
its vector factors. Thus, for example, the multi-vector e134 is associated with the
sum 20 + 22 + 23 = 13. Notice that the binary representation of 13 is 1101 with
2000 Mathematics Subject Classification. 16S99,16W99.
Key words and phrases. twisted group algebra, twist, Clifford algebra, geometric algebra.
1
2
JOHN W. BALES
bits 1, 3 and 4 set. We will represent the sequence 1, e1, e2, e12, e3, · · · by the
sequence i0, i1, i2, i3, i4, · · · where the subscript of i is the number associated with
the corresponding vector or multi-vector.
Notice that, since the square of a vector ek is µ which is either 1 or −1 , the
product of two basis elements ip and iq will always be either ir or −ir where r
is the XOR (exclusive or) of the binary representations of integers p and q . The
set G of non-negative integers is a group under XOR. For brevity, we will denote
the operation p XOR q by simple concatenation pq. Thus there is a function φ
mapping G × G into {−1, 1} such that if p, q ∈ G then
(2.1)
ipiq = φ(p, q)ipq
thereby representing Clifford algebra as a twisted group algebra.
Let 2p denote the double of p. Notice that the vector factors of i2p are the
successors of the vector factors of ip in the sense that ek is a vector factor of i2p
if and only if ek−1 is a vector factor of ip. For example, i13 = e134 and i26 = e245.
This is more intuitive if the subscripts are represented in binary. 13 = 1101B with
bits 1,3 and 4 set, and 2(13) = 26 = 11010B with bits 2,4 and 5 set. Multiplying
by 2 in binary shifts bits to the left and appends a 0 on the right.
The next two lemmas are then immediately obvious.
Lemma 2.1.
e1i2p = i2p+1
Lemma 2.2.
e1i2p+1 = µi2p
Let β(p) denote the sum of the bits of p. Then β(p) is the grade of ip. The
remaining lemmas follow from the fact that i2p contains exactly β(p) vector factors
and e1 must be 'commuted' with each of them to 'find its place' so to speak.
Lemma 2.3.
i2pe1 = (−1)β(p)i2p+1
Lemma 2.4.
i2p+1e1 = (−1)β(p)µi2p
Theorem 2.5. There is a twist φ(p, q) mapping G × G into {−1, 1} such that if
p, q ∈ G, then ipiq = φ(p, q)ipq.
Proof. Let Gn = {p 0 ≤ p < 2n} with group operation "bit-wise exclusive or."
To begin with, i0i0 = φ(0, 0)i0 = 1 provided φ(0, 0) = 1.
This defines the twist for G0.
If p and q are in Gn+1, then there are elements u and v in Gn such that one
of the following is true:
(1) p = 2u and q = 2v
(2) p = 2u and q = 2v + 1
(3) p = 2u + 1 and q = 2v
(4) p = 2u + 1 and q = 2v + 1
Assume φ is defined for u, v ∈ Gn, then consider these four cases in order.
THE CLIFFORD TWIST
3
(1) p = 2u and q = 2v
ipiq = i2ui2v
= φ(u, v)i2uv
= φ(2u, 2v)i(2u)(2v)
= φ(p, q)ipq
provided φ(2u, 2v) = φ(u, v).
(2) p = 2u and q = 2v + 1
ipiq = i2ui2v+1
= i2ue1i2v
= (−1)β(u)e1i2ui2v
= (−1)β(u)e1 φ(2u, 2v)i2uv
= (−1)β(u) φ(u, v)i2uv+1
= φ(2u, 2v + 1)i2uv+1
= φ(p, q)ipq
provided φ(2u, 2v + 1) = (−1)β(u) φ(u, v).
(3) p = 2u + 1 and q = 2v
ipiq = i2u+1i2v
= e1i2ui2v
= e1 φ(u, v)i2uv
= φ(u, v)i2uv+1
= φ(2u + 1, v)i2uv+1
= φ(p, q)ipq
provided φ(2u + 1, 2v) = φ(u, v).
(4) p = 2u + 1 and q = 2v + 1
ipiq = i2u+1i2v+1
= e1i2ue1i2v
= (−1)β(u)e1e1i2ui2v
= (−1)β(u)µ φ(u, v)i2uv
= φ(2u + 1, 2v + 1)i2uv
= φ(p, q)ipq
provided φ(2u + 1, 2v + 1) = (−1)β(u)µ φ(u, v).
(cid:3)
Corollary 2.6. Assume p, q ∈ Gn. The Clifford algebra twist can be defined re-
cursively as follows:
(1) φ(0, 0) = 1
(2) φ(2p, 2q) = φ(2p + 1, 2q) = φ(p, q)
(3) φ(2p, 2q + 1) = (−1)β(p) φ(p, q)
4
JOHN W. BALES
(4) φ(2p + 1, 2q + 1) = (−1)β(p)µ φ(p, q)
Stated another way
(2.2)
(cid:20) φ(2p, 2q)
φ(2p + 1, 2q) φ(2p + 1, 2q + 1)(cid:21) = φ(p, q)(cid:20)1
φ(2p, 2q + 1)
1
(−1)β(p)
(−1)β(p)µ(cid:21)
3. Recursive generation of twist matrices for higher dimensions
The twist matrix for dimension one is found when p = q = 0
1
(cid:20) 1
1 µ (cid:21)
For two dimensions, 0 ≤ p ≤ 1, 0 ≤ q ≤ 1, the twist matrix is
1
1
µ 1
1
1
1
µ
1 −1 µ −µ
1 −µ µ −1
For µ = −1 these coincide with the twist tables for complex numbers and
quaternions. For dimension 3, however, we do not get the twist table for the
octonions, rather
1
1
1
µ
1 −1
1 −µ
1 −1 −1
1 −µ −1
1
1
1 −µ −µ µ
µ −µ −1 µ
1
1
µ 1
1
1
1
µ
µ −µ 1 −1
µ −1
1 −µ
1 µ −µ −µ
µ µ −1 −µ
1
1
1
µ
µ −µ
µ −1
µ
1
µ −1 −1
1 −1 −µ
For dimension four the twist matrix is too large to represent in this form, so we
make the following substitutions:
(3.1)
(3.2)
A = (cid:20) 1
1 µ (cid:21)
1
B = (cid:20) 1 −1
1 −µ (cid:21)
The matrices A and B are simply the values of M (p) = (cid:20)1
1
(−1)β(p)
(−1)β(p)µ(cid:21) when
(−1)β(p) is positive and negative, respectively.
Then the dimension 4 twist table can be represented as follows.
THE CLIFFORD TWIST
5
A
A
µB
B
−B
B
A −µA
−B
B
A −µA
A
B
A
B
A
µB
µB −µB
−A
µA
−B
−A
A
B
A
A
µB
B
−B
B
A −µA
A
µB
µB −µB
−A
µA
B µB −µB −µB
µB
−A −µA
A
−A
−A
µA
−B −µB
B
µA µA
A −µA −µA µA
−B µB
µB −µB
The twist tables for the various dimensions can be generated recursively begin-
ning with A for dimension 1, then making the following replacements to generate
the twist table for each successively higher dimension:
(3.3)
(3.4)
A =⇒ (cid:20) A
B µB (cid:21)
A
B =⇒ (cid:20) B
A −µA (cid:21)
−B
4. A tree for computing the Clifford twist
In [3] a tree for computing the Cayley-Dickson twist is described. The same
procedure applies to the Clifford twist.
The tree consists of only four components which repeat indefinitely, beginning
at node A . There are two versions, one for each value of µ.
A
-- A
A A B B
B
-- A -- A -- B -- B
-- B
B -- B A -- A
-- B B -- A A
Figure 1. Clifford twist tree for µ = 1.
A
-- A
A A B -- B
B
-- A -- A -- B B
-- B
B -- B A A
-- B B -- A -- A
Figure 2. Clifford twist tree for µ = −1.
Let us illustrate the use of the tree to compute the product i2636i1143 given
µ = −1.
6
JOHN W. BALES
(1) Convert the subscripts to binary notation. 2636 = 101001001100B and
1143 = 10001110111B.
(2) Pair the bits of the first subscript with the bits of the second by placing
one over the other. Pad the smaller with zero bits if necessary.
1−→
,
0
0−→
,
1
0−→
,
1
1−→
,
0
1−→
,
1
0−→
,
1
0−→
1
1−→
,
1
1−→
,
0
0−→
,
0
0−→
,
0
0−→
,
1
(3) Each binary pair is an instruction for traversing one of the four tree com-
ponents. A zero is an instruction to move down a left branch and a one
is an instruction to move down a right branch. The result is the following
path.
A
1−→
0
0
−→
1
1−→
0
0−→
0
0−→
0
1−→
1
0−→
1
0−→
1
1−→
0
1−→
1
0−→
1
0−→
1
B
−B
−A
−A
−A
B
−B
B
A
−B
B
−B
Since the result is −B , φ(2636, 1143) = −1. Whenever the result is
−A or −B , φ = −1 and whenever the result is A or B , φ = +1. Since
101001001100 XOR 010001110111 = 111000111011 = 3643 the result is
or
i2636 · i1143 = −i3643
e347ac · e123567b = −e12456abc
References
[1] R. Busby, H. Smith Representations of twisted group algebras Trans. Am. Math. Soc.
Vol. 149, No. 2, (1970) 503-537
[2] J. Bales Properly twisted groups and their algebras (2006) arXiv:1107.1297
[3] J. Bales A tree for computing the Cayley-Dickson twist Missouri J. of Math. Sci. Vol 21
No. 2 (2009)
THE CLIFFORD TWIST
7
Department of Mathematics, Tuskegee University, Tuskegee, AL 36088, USA
E-mail address: [email protected]
|
1106.3818 | 3 | 1106 | 2011-08-11T18:53:47 | Non-reproductive and reproductive solutions of some matrix equations | [
"math.RA"
] | In this paper we analyzed solutions of some complex matrix equations related to pseudoinverses using the concept of reproductivity. Especially for matrix equation AXB=C it is shown that Penrose's general solution is actually the case of the reproductive solution. | math.RA | math | NON-REPRODUCTIVE AND REPRODUCTIVE SOLUTIONS OF SOME
MATRIX EQUATIONS
B.J. Malešević*, B.M. Radičić
* Faculty of Electrical Engineering, University of Belgrade, Serbia
In this paper we analyzed solutions of some complex matrix equations related to pseudoinverses using the
concept of reproductivity . Especially for matrix equation AXB = C it is shown that Penrose's general solution
is actually the case of the reproductive solution.
Reproductive equations
A concept of reproductive equations was introduced by S.B. Prešić
.4 in 1968. The
implementation of the concept of reproducibility was also considered by J.D. Kečkić
6. , .7 ) and S.B. Prešić and J.D. Kečkić
.12 .
(
Definition. The reproductive equations are the equations of the following form:
x
)( xf
where x is a unknown, S is a given set and
f
Sf :
S
is a given function which satisfies
f
f
. (1)
The condition (1) is called the condition of reproductivity because the function f which
satisfies the condition (1), after iteration, reproduces itself.
In literature, functions which satisfy the condition (1) are also called projectors or
idempotent maps.
The following two statements (S.B. Prešić
.4 ) give the fundamental properties of the
being in the same time reproductive as well. ◊
reproductive equations.
Theorem 1. For any of the consistent equations
there is an equation of the
)( xJ
, for any of the values y S. ◊
form, which is equivalent to
)( xJ
Theorem 2. If a certain equation
is equivalent to the reproductive one
x
)( xf
)( xJ
the general solution is given by the formula
x
)( yf
The first implementation of reproducibility on some matrix equations was considered by S.B.
Prešić ([2.], [4.]).
Solutions of the matrix equation AXB=C
nm
and C is the field of complex number. The set of all matrices of order
Let
Nnm ,
nm over
nmC . By
aC we denote the set of all matrices of order
nm
over C we denote by
1mC we use the denotement
nC 1
C which are of rank
.a Instead of
mC while instead of
we use the denotement
nC .
, then
)( xf
x
nmCA
. The solution of matrix equation
A
AXA (2)
Let
)1(A . The set of all {1}-inverses of A is denoted
is called {1}-inverse of A and it is denoted by
by 1A . More information about {1}-inverses and other types of inverses can be found in A.
Ben-Israel and T.N.E. Greville
.13 and S.L. Campbell and C.D. Meyer [16].
. The following statement (R. Penrose .1 ) gives the
nmCA
,
and
Let
gpCB
qmCC
necessary and sufficient condition for a matrix equation
C
AXB (3)
is consistent. The formula of general solution of the matrix equation (3) is also given in the
following statement.
)1(B of
)1(A and
Theorem 3. The matrix equation (3) is consistent iff for some of {1}-inverses
the matrices A and B the following condition is true
AA CB B C (4)
(1)
(1)
The general solution of the matrix equation (3) is given by the formula
where Y is an arbitrary matrix corresponding dimensions. ◊
AY
X
A
CB
Yf
)(
)1(
)1(
)1(B of the matrices
(1) ,A
If the condition (4) is true for at least one choice of {1}-inverses
)1(B . The
(1) ,A
,A B respectively, then the condition (4) is true for any choice of {1}-inverses
following statement is an extension of Theorem 3.
(5)
AYBB
)1(
)1(
0
X
AYBB
)1(
AY
)1(
YgX
)(
pnC
X
Theorem 3’. If the matrix
is any of the particular solutions of the matrix equation
0
(3), then the general solution of the matrix equation (3) is given by the formula
)1(B of the matrices
(1) ,A
,A B respectively and where Y is an
for any choice of {1}-inverses
arbitrary matrix corresponding dimensions. The general solution (6) of the matrix equation (3)
X
CB
A
.)1(
)1(
is reproductive one iff
0
Proof. It is easily to show that the solution of the matrix equation is given by (6). On the
contrary, let X is any of the solutions of the matrix equation 3 . Based on Theorem 3, there is
X
Yf
CB
A
AYBB
AY
)(
Y so that
XY
Z
A
CB
)1(
)1(
)1(
)1(
. Then, for
it is true
)1(
)1(
XZf
X
A
CB
Yf
)(
(
)
)1(
)1(
i.e.
(6)
0
0
XZf
AZ
X
A
CB
X
AZBB
Zg
(
(
)
)
0
0
From that we conclude that 6 is the general solution of the matrix equation 3 .
)1(
)1(
)1(
)1(
.
)1(
)1(
. ◊
X
A
CB
Yg
Yg
)(
)(
(
)
2
Considering that the equality
is valid, the formula of general
0
X
A
CB
)1(
)1(
solution (6) of the matrix equation (3) is reproductive iff
0
From the previous theorem we can see that Penrose's general solution (5) is the reproductive
0X which can not be
solution. In this paper we examine whether there are particular solutions
X
CB
A
)1(
)1(
represented in the form
(see the example at the end of this paper). The first
0
appearances of these solutions were in
.2 . In that paper S.B. Prešić analyzed the matrix
equation (2) in the case when A is a square matrix and proved the following statement.
1AB
Theorem 4. For the square matrix
the following is true:
and matrix
nnCA
. ◊
CY
BAY
AX
CY
YX
A
AX
YI
X
BAY
nn
nn
1)
,
2)
,
0
YX
XA
CY
A
YI
XA
CY
X
YAB
YAB
nn
nn
0
,
4)
3)
,
YB
X
CY
AXA
A
BAYAB
nn
5)
gives both Prešić’s solutions of the given equations and Penrose’s solutions of the given
equations. In the paper .8 M. Haverić showed that we can get Penrose’s solution from
The general solutions 2), 4) and 5) of the previous theorem are non-reproductive and as such
Prešić’s solution. Namely, she proved the following statement
they do not directly appear according to Penrose’s theorem. Let us notice that the formula (6)
1AB
Theorem 5. For the square matrix
the following is true:
and matrix
nnCA
. ◊
A
Y
CY
AX
X
BA
BAY
CY
YX
AX
BAY
nn
nn
1)
,
,
0
2)
YX
XA
CY
YAB
YAB
Y
AB
X
CY
A
XA
nn
nn
0
,
4)
3)
,
BAB
X
CY
AXA
A
BAYAB
Y
nn
5)
The matrix equation (3) is the subject of the contemporary research (see [10.],[11.],[14.],[15.]).
Solutions of non-homogeneous linear systems
In this part of the text we are going to show that the general solution of the consistent non-
homogeneous linear system can be obtain by general {1}-inverse. Namely, for each matrix
nmCA
nnCP
mmCQ
there are regular matrices
and
such that
QAP
E
a
I
a
0
0
0
, (7)
A
P
)1(
iw , ] are matrices with total
iv ,
iu , ],
W [
V [
mutually
] and
U [
where
2anmk
j
j
j
independent variables. In the paper
5. V. Perić mentioned that the representation (8) of the
general {1}-inverse was given by C. Rohde in his doctoral thesis .3 .
(8)
UI
a
WV
Q
where
( A
rank
a
).
following form
Then, each {1}-inverse
)1(A of the matrix A can be represented in the
j
j
mCc
A C
Theorem 6. Let
,m n
and
a
Ax (9)
c
)1(A of the matrix
is a consistent non-homogeneous linear system. Then, there is {1}-inverse
an arbitrary variables such that a vector
A in the form (8) with
cAx
)1(
(10)
determines the general solution of linear system (9).
Proof. Let for the matrix A the regular matrices Q and P be determined so that (7) is true.
Let us also notice that the vector (10) always determines one solution of the linear system (9)
)1(A . We conclude
iv , and
,
in the general {1}-inverse
for any choice of the variables
iu ,
iw ,
j
that
cQAQ
cA ])
rank
a
rank
]) (11)
([
([
and
AQ
a
rank
A
rank
(
(
)
)
using the consistency of the linear system (9). From (11) we can conclude that the vector
jc (
cQc
thj
'
has zeroes on the last
)
coordinate
coordinates and there is
'
j 1
a
tham
' jc
0
such that
because the linear system (9) is non-homogeneous. Therefore, in the case of
the general inverse (8) the following is true
1
0
vc
'
i
a
i
1
vc
'
0
1
0
iani
,
. By choosing the submatrix V in the form
0
1
v
v
2,1
,1
v
aan
,
ian
,
v
c
0
0
'
j
j
,1
(12)
V
c
v
0
0
'
jan
j
,
arbitrary variables, determines the general solution of linear system (9). ◊
v
jv ,1
. From this we conclude that the directrix of the previously
, … ,
we get
1
jan
an
,
an
an dimension. The vector
cAx
cAx
)1(
)1(
, with
is of
determined affine space
1,2
a
, …,
an
i
1
1
a
for
i
1
0
0
c
'
a
vc
'
i
,1
I a
U
WV
u
ama
,
w
,1
am
w
aman
,
0
v
1,1
1,
u
a
w
1,1
c
'
1
c
'
2
c
'
a
0
c
'
1
c
'
2
c
'
a
0
0
0
0
1
cQ
=
1
u
1,1
u
u
,1
u
,2
am
am
w
an
.
1,
an
,
P
v
an
1,
P
P
a
i
1
0
0
a
P
0
0
v
an
2,
c
'
1
c
'
2
0
1
i
i vc ,1
'
i
0
Px
P
and
,
c C
n
B C
m n
b
)1(B of the matrix
c
xB (13)
Theorem 7. Let
is a consistent non-homogeneous linear system. Then, there is {1}-inverse
B in the form (8) with m b arbitrary variables such that a vector
Proof. The proof is analogous to the previous proof. ◊
(14)
x
)1(cB
determines the general solution of linear system (13).
Using the Kronecker (tensor) product of matrices we can rewrite the matrix equation
.13 , A.K. Jain .9 ) as
(A. Ben-Israel and T.N.E. Greville
X
denotes the vector of the matrix X which is formed by writing the rows of the
vec
where
matrix X
into a single column vector. Namely, vec determines
the operator
vec
x
x
C
vec
C
. The inverse operator
that is defined as follows
nm
mn
:
jn
i
i
nm
nm
,
)1(
,
,
mat
x
i n
C
C
mat
x
p
q
i
n
is defined as follows
for
and
.
nm
mn
, ,
/
:
1
mod
i
p q
Example. Let be given the matrix equation:
(15)
AXB
C
BA
T
(Cvec
),
vec
X
nm
,
m n
,
j
AXB
C
121
010
111
X
11
11
22
3
1
2
3
1
2
. (16)
Using the Kronecker product the matrix equation (16) may be considered in the form of
equivalent linear system:
211422211
211422211
000211000
000211000
211211211
211211211
BA
T
vec
X
Cvec
(
)
x
1,1
x
2,1
x
3,1
x
1,2
x
2,2
x
3,2
x
1,3
x
2,3
x
3,3
3
3
1
1
2
2
. (17)
Based on (8) the general {1}-inverses of
A
121
010
111
and
B
11
11
22
are given by the
1
e
a
cb
2
b
c
e
a
db
2
2
b
1
d
e
ae
eb
2
b
e
,
Cedcba
,
,
,
,
following matrices:
)1(
A
and
The matrix
X
0
A
)1(
CB
)1(
B
)1(
1
g
h
qp
2
qp
r
2
2
r
g
q
rhq
r
,
Crqphg
,
,
,
,
.
31
cg
h
g
dc
3
2
h
g
2
1
cg
ch
d
6
3
c
3
ch
6
dg
2
dh
g
dg
h
cg
3
g
cg
3
ch
3
h
ch
3
dh
dh
dg
2
dh
dg
is one solution of the matrix equation (16) but we are going to show that it is not the general
solution,
Chgdc
.
,
,
,
Now, we will solve the linear system (17) using Theorem 6. Applying the elementary
66 CQ
such that
and
operations we get the regular matrices
99 CP
BAQ
T
EP
r
I
r
0
0
0
,
r
rank
TBA
rank A rank B
. Namely, we get that
2r
and
where
Q
0
1
0
0
11
0
0
01
01
000
0
1
000
0
000
0011
1
010
1
100
,
P
1
0
0
0
0
0
0
0
0
2
0
0
1
0
0
0
0
0
1
1
0
0
0
0
0
0
0
2
0
1
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
2
0
1
0
0
0
1
0
0
0
0
0
1
0
0
1
0
0
0
0
0
0
1
0
2
0
0
0
0
0
0
0
1
.
C Q vec C
'
(
)
3
1 0 0 0 0 T
Then,
and
vec
(
X
)
BA
(
T
)
)1(
Cvec
(
)
P
1
0
v
1,1
3
v
1,2
3
v
1,7
3
0
1
0
0
0
u
1,1
u
1,2
w
1,1
w
1,2
w
1,7
u
2,1
u
2,2
w
2,1
w
2,2
w
2,7
u
3,1
u
3,2
w
3,1
w
3,2
w
3,7
u
4,1
u
4,2
w
4,1
w
4,2
w
4,7
'
C
From (18) we get the matrix of the general solution
1
v
1,1
v
2
1,2
v
1,5
v
1,6
v
2
1,7
1
v
2
1,4
v
1,1
v
1,2
v
1,3
v
1,3
v
1,4
v
1,5
v
1,6
v
1,7
. (18)
X mat vec X
(
(
))
=
1
v
1,1
v
v
1,5
v
2
1,4
v
2
1,2
v
1
1,3
v
1,5
v
2
1,7
1,6
v
1,1
v
1,3
v
1,6
v
1,2
v
1,4
v
1,7
.
For
v
1,1
v
1,2
v
1,6
v
11,7
and
v
1,3
v
1,4
v
1,5
0
, the matrix
1X
117
001
0
11
)1(A ,
)1(B because
). Finally, we
1
(1) .B ◊
A
CB
X
for any choice of {1}-inverses
is the solution of (16), but
)1(
)1(
1
and
we obtain the contradiction (
X
X
0g
dg
cg
from
0
3
0
1
(1) ,A
X
CB
A
proved that
for any choice of {1}-inverses
)1(
)1(
1
Acknowledgements
The research is partially supported by the Ministry of Education and Science, Serbia, Grant No.
174032.
References
1. R. Penrose, A generalized inverses for matrices, Math. Proc. Cambridge Philos. Soc. 51
(1955).
2. S.B. Prešić, Certaines equations matricielles, Publ. Elektrotehn. Fak. Ser. Mat.-Fiz. 121,
Beograd, 1963.
f
,
f 2
3. C.A. Rohde, Contibution to the theory, computation and application of generalized inverses,
doctoral dissertation, University of North Carolina at Releigh, May, 1964.
4. S.B. Prešić, Une classe d’equations matricielles et l’´equation fonctionnelle
Publications de l’institut mathematique, Nouvelle serie, tome 8 (29), Beograd, 1968.
5. V. Perić, Generalizirana reciproka matrice, Stručno-metodički časopis Matematika, Zagreb
1982.
6. J.D. Kečkić, Reproductivity of some equations of analysis I, Publ. Inst. Math, 31(45), 73-81,
1982.
7. J.D. Kečkić, Reproductivity of some equations of analysis II, Publ. Inst. Math, 33(47), 109-
118, 1983.
8. M. Haverić, Formulae for general reproductive solutions of certain matrix equations,
Publications de l’institut mathematique, Nouvelle serie, tome 34 (48), Beograd, 1983.
9. A.K. Jain, Fundamentals of Digital Image Processing, Prentice Hall, 1989.
10. Lj. Kočinac, P. Stanimirović and S. Đorđević, Representation of {1}–inverses and the
group inverse by means of rational canonical form, Scientific Review 21–22 (1996), 47–55.
11. Lj. Kočinac, P. Stanimirović and S. Đorđević, {1}-inverses of square matrices and rational
canonical form, Math. Moravica 1 (1997), 41-49.
12. J.D. Kečkić, S.B. Prešić, Reproductivity – A general approach to equations, Facta
Universitatis (Niš), Ser. Math. Inform. 12 (1997), 157-184.
13. A. Ben-Israel and T.N.E. Greville, Generalized Inverses: Theory and Applications,
Springer, 2003.
14. D.S. Cvetković, The reflexive solutions of the matrix equation AXB = C, Comp. Math.
Appl. , 51 (2006), 897-902.
15. D.S. Cvetković-Ilić, Re-nnd solutions of the matrix equation AXB = C, Journal of the
Australian Mathematical Society, 84 (2008), 63 –72.
16. S.L. Campbell and C.D. Meyer, Generalized Inverses of Linear Transformations, Society
for Industrial and Applied Mathematics, 2009.
|
1102.1567 | 1 | 1102 | 2011-02-08T11:04:43 | Minimal clones generated by majority operations | [
"math.RA"
] | We determine all majority operations on a four-element set that generate a minimal clone. | math.RA | math |
MINIMAL CLONES GENERATED BY MAJORITY OPERATIONS
TAM ´AS WALDHAUSER
Abstract. The minimal majority functions of the four-element set are deter-
mined.
1. Introduction
A set C of finitary operations on a set A is a clone if it is closed under composition
of functions and contains all projections. In this paper we shall be concerned only
with clones on a finite set.
The set of all finitary operations on A is a clone as well as the set of all projections.
These are the largest and the smallest clones on A, the latter is often called the
trivial clone.
The clone generated by a set F of finitary functions on A is the intersection of
all clones containing F , i.e. the smallest clone containing F . This clone is denoted
by [F ]. If F = {f } then we simply write [f ]. Clearly [F ] consists of those functions
which can be obtained from the elements of F and from the projections by a finite
number of compositions. In other words, [F ] is the set of term functions of the
algebra hA, F i.
We say that the n-ary function f preserves the relation ρ ⊆ Ak if for all ai,j ∈ A
(i = 1, . . . , k, j = 1, . . . , n)
(a11, a21, . . . , ak1) , (a12, a22, . . . , ak2) , . . . , (a1n, a2n, . . . , akn) ∈ ρ
implies
(f (a11, a12, . . . , a1n) , f (a21, a22, . . . , a2n) , . . . , f (ak1, ak2, . . . , akn)) ∈ ρ.
Preserving a relation is inherited when composing functions:
(1.1)
If f preserves a relation ρ and g ∈ [f ], then g also preserves ρ.
An important special case is that of unary relations: f preserves B ⊆ A iff B is
closed under f . If f preserves all subsets of A then we say that f is conservative
(cf. [5]).
A clone is minimal if it has no proper subclones except for the trivial one. On
finite sets every clone contains a minimal one (cf. [6]). Obviously a clone is minimal
iff it is generated by every nontrivial member of it. (By a nontrivial function we
mean a function which is not a projection.)
If C is a minimal clone and f is
nontrivial, and of minimum arity in C then we say that f is a minimal function
(cf. [7] p.408).
By a theorem of I. G. Rosenberg [7], every minimal clone (on a finite set) is
generated by a nontrivial minimal function f , for which one of the following holds:
1) f is unary and f 2(x) ≈ f (x) or f p(x) ≈ x for some prime p.
2) f is binary idempotent, i.e. f (x, x) ≈ x.
3) f is ternary majority, i.e. f (x, x, y) ≈ f (x, y, x) ≈ f (y, x, x) ≈ x.
2000 Mathematics Subject Classification. 08A40.
Key words and phrases. Clones, majority operations.
This research was supported by Ministry of Culture and Education of Hungary, grant No.
FKFP 0877/1997.
1
2
T. WALDHAUSER
4) f is a semiprojection, i.e. there exists an i such that f (x1, x2, . . . , xn) = xi
whenever the arguments are not pairwise distinct.
5) f (x, y, z) = x + y + z where hA, +i is a Boolean group.
Note that in each case f cannot generate a nontrivial function which is of lesser
arity than f . This means that f is a minimal function iff [f ] is a minimal clone. In
cases 1) and 5) the conditions ensure the minimality of f , while in the other cases
they do not.
In [4] Post described all clones on a two-element set, in [1] B. Cs´ak´any determined
the minimal clones of a three-element set. For the four-element case binary minimal
clones were described by B. Szczepara in [8].
Conservative minimal majority and binary functions were determined on any
finite set by B. Cs´ak´any in [2].
In this paper we prove the following description of all the minimal majority
functions of a four-element set.
Theorem 1.1. If C is a minimal clone on a four-element set A and it contains
a majority function, then C = [f ] where f is either conservative or hA, f i ∼=
h{1, 2, 3, 4}, Mii for some i ∈ {1, 2, 3} (see the table below).
(1, 2, 3)
(2, 3, 1)
(3, 1, 2)
(2, 1, 3)
(1, 3, 2)
(3, 2, 1)
{1, 2, 4}
{1, 3, 4}
(4, 2, 3)
(2, 3, 4)
(3, 4, 2)
(2, 4, 3)
(4, 3, 2)
(3, 2, 4)
M1 M2 M3
3
4
3
4
4
3
4
4
4
4
4
4
4
4
4
4
3
4
3
4
4
3
4
4
4
4
4
4
4
2
3
2
4
3
4
4
4
2
3
2
4
3
The middle two rows mean that if
{a, b, c} = {1, 2, 4} or {1, 3, 4}, then
Mi(a, b, c) = 4 for i = 1, 2, 3. For
the triplets not listed in the table the
majority rule defines the value of the
functions.
2. Majority functions on finite sets
If C is a clone which is generated by a majority function then we shall briefly
say that C is a majority clone.
Let A be a finite set and f be a majority function on A. We define the range of
f in the following way:
range(f ) = { f (a, b, c) a, b, c ∈ A are pairwise distinct } .
A simple induction argument shows that if g is a nontrivial function in a majority
clone, then g is a so-called near-unanimity function, i.e.
g(y, x, x, . . . , x, x) ≈ g(x, y, x, . . . , x, x) ≈ . . . ≈ g(x, x, x, . . . , x, y) ≈ x
(cf. (7) of [2]).
In Rosenberg's theorem f cannot be a near-unanimity function except for the
majority case, so any minimal subclone of a majority clone is again a majority
clone. This means that in order to prove the minimality of a majority clone C, it
suffices to show that any two majority functions in C generate each other.
To show the nonminimality of a clone [f ] we will make use of the following facts.
Lemma 2.1. Let f be a majority function on A.
MINIMAL CLONES GENERATED BY MAJORITY OPERATIONS
3
(2.1) If f is a minimal function and it preserves B ⊆ A then f B must be a
minimal function on B.
(2.2) If a nontrivial g ∈ [f ] preserves some B ⊆ A but f does not, then [f ] is not
minimal.
(2.3) If the range of some nontrivial g ∈ [f ] does not contain an element which
belongs to the range of f , then [f ] is not minimal. (Cf. Corollary 1.25 of
[9].)
Proof.
(2.1) Composing functions and restricting functions commute.
(2.2) This follows from (1.1).
(2.3) This also follows from (1.1) since a 6∈ range(g) iff g preserves the equivalence
(cid:3)
relation whose blocks are {a} and A\{a}.
We can use (2.1) for three-element set B because we know the minimal majority
clones for such set. These are described in [1] as follows. If C is a minimal majority
clone on a three-element set A, then there exists an f ∈ C such that
hA, f i ∼= h{1, 2, 3}, mii for some i = 1, 2, 3,
where m1, m2, m3 are the following majority functions.
For {x0, x1, x2} = {1, 2, 3} we have
m1(x0, x1, x2) = 1
m2(x0, x1, x2) = x0
m3(x0, x1, x2) = xi+1 if xi = 1 (subscripts taken modulo 3).
The clones generated by m1, m2, m3 contain 1,3, and 8 majority functions re-
spectively; these are shown in the following table.
(1, 2, 3)
(2, 3, 1)
(3, 1, 2)
(2, 1, 3)
(1, 3, 2)
(3, 2, 1)
m1 m2
1
1
2
1
3
1
1
2
1
1
1
3
m3
2
2
2
3
3
3
2
3
1
1
3
2
3
1
2
3
2
1
3
2
2
3
2
3
2 2
2 3
3 2
2 3
3 3
3 2
3 2
3 3
3 3
2 2
2 3
2 2
3 3
3 2
2 3
3 2
2 2
2 3
Conservative minimal majority clones are described in [2] as follows:
If C is a conservative minimal majority clone then there exists an f ∈ C such
that for every three-element B ⊆ A there is an iB ∈ {1, 2, 3} such that
hB, f Bi ∼= h{1, 2, 3}, miBi.
Now we formulate a theorem which helps us reducing the number of functions
to be checked, when searching for minimal clones.
Theorem 2.2. Let f be a majority function on a finite set A. Then there exists a
majority function g ∈ [f ] which satisfies the following identity:
g(cid:0) g(x, y, z) , g(y, z, x) , g(z, x, y)(cid:1) ≈ g(x, y, z).
(∗)
Proof. We define functions f (k) (k ≥ 1) in the following way:
f (1)(x, y, z) = f (x, y, z),
f (k+1)(x, y, z) = f(cid:0) f (k)(x, y, z) , f (k)(y, z, x) , f (k)(z, x, y)(cid:1).
We assert that
f (k+l)(x, y, z) ≈ f (k)(cid:0) f (l)(x, y, z) , f (l)(y, z, x) , f (l)(z, x, y)(cid:1)
for k, l ≥ 1.
4
T. WALDHAUSER
This can be proved by induction on k; the proof is left to the reader. Let us
define a binary operation ∗ on the set D = (cid:8)f (k) : k ∈ N(cid:9) as follows:
(cid:0)f (k) ∗ f (l)(cid:1)(x, y, z) = f (k)(cid:0) f (l)(x, y, z) , f (l)(y, z, x) , f (l)(z, x, y)(cid:1).
The above assertion means that the map k 7→ f (k) is a homomorphism from
hN, +i to hD, ∗i. So the latter is a finite semigroup, hence it has an idempotent
element, say f (k) ∗ f (k) = f (k). And this is just the desired identity for g = f (k). (cid:3)
Now we introduce some more notation. The function g = f (k) which corresponds
to f in the theorem will be denoted by bf . We put habci = {(a, b, c), (b, c, a), (c, a, b)},
and we will use the symbol f habci≡ u to mean that f (a, b, c) = f (b, c, a) =
f (c, a, b) = u, and f habci= p to mean that f (a, b, c) = a, f (b, c, a) = b, f (c, a, b) = c.
The following lemma tells us what identity (∗) means for a majority function.
Lemma 2.3. Let f be a majority function satisfying (∗) and let a, b, c be pairwise
distinct elements of A. Let u = f (a, b, c), v = f (b, c, a), w = f (c, a, b). Then
{u, v, w} 6= 2 and if u, v, w are pairwise different, then f huvwi= p.
Proof. To prove the first statement, let us suppose (without loss of generality)
that u = v 6= w. Then (∗) for x = c, y = a, z = b yields that f (w, u, v) = w,
contradicting the majority property of f .
The second statement of the lemma follows similarly from (∗).
(cid:3)
We can say a bit more then Lemma 2.3 when f is a minimal function.
Theorem 2.4. If f is a minimal majority function satisfying (∗) and u = f (a, b, c),
v = f (b, c, a), w = f (c, a, b) are pairwise different then f huvwi= p and also
f hvuwi= p.
Proof. By the previous lemma we have f huvwi= p. Now the nontrivial superposi-
tion g(x, y, z) = f (f (x, y, z), f (x, z, y), x) preserves {u, v, w} hence f does too, and
then from the description of the minimal majority functions on the three-element
set we get the conclusion of the theorem.
(cid:3)
3. The four-element case
We have seen that every conservative minimal majority clone is generated by a
function f having the following property:
(3.1)
f habci≡ u or f habci= p
for every a, b, c ∈ A with a suitable u (depending of course, on a, b, c).
One would hope that it holds for nonconservative clones too. In the first part of
this section we are going to try to prove this for a four-element A. It will turn out,
that the conjecture is not true, but (in the four-element case) there is essentially
only one exception. In the second part we determine the minimal ones among the
functions satisfying property (3.1), and in the third part we prove the minimality
of the clones we have found.
3.1.
Let S denote the set of those majority functions on the set A = {1, 2, 3, 4} for
which (3.1) holds for any a, b, c ∈ A.
In this section we will show that a minimal majority function which satisfies (∗)
must belong to the set S, or it is isomorphic to M2. Since we will consider the
values of the functions on the set {1, 2, 3}, we introduce one more notation. Let
[p, q, r; s, t, u] denote the set of majority functions f on A for which f (1, 2, 3) = p,
If
f (2, 3, 1) = q, f (3, 1, 2) = r, f (2, 1, 3) = s, f (1, 3, 2) = t, f (3, 2, 1) = u.
MINIMAL CLONES GENERATED BY MAJORITY OPERATIONS
5
we do not want to specify all these six values of f , than we will use ∗ to indi-
cate an arbitrary element of A. For example f ∈ [4, ∗, ∗; ∗, ∗, ∗] means just that
f (1, 2, 3) = 4. The letters a, b, c, d will always denote arbitrary distinct elements of
A, i.e. {1, 2, 3, 4} = A = {a, b, c, d}.
First we define and examine a superposition which we will use frequently later
on. For a ternary function f let fx, fy, fz stand for the composite functions where
the first, second resp. third variable of f is replaced by f itself.
fx(x, y, z) = f (f (x, y, z), y, z)
fy(x, y, z) = f (x, f (x, y, z), z)
fz(x, y, z) = f (x, y, f (x, y, z))
We will briefly write fzy instead of (fz)y. We will also use the convention that
zy means (fzy)(2) and not (f (2))zy,
lower indices have priority to upper ones. So f (2)
and also bfzy stands for [(fzy).
The proof of the following lemma is just a straightforward calculation, so we
omit it.
If this is not the
Lemma 3.1. If f (a, b, c) 6= d then fzy(a, b, c) = f (a, b, c).
case, then fzy(a, b, c) = f (a, b, d) if the latter does not equal d. If it does, then
fzy(a, b, c) = f (a, d, c) if it is not b. If f (a, d, c) = b then fzy(a, b, c) = f (a, d, b).
From now on f will always denote an arbitrary majority function on A, satisfying
(∗). In the following lemma we prove a nice property of f , then through five claims
we reach the main result of this section, which is stated in Theorem 3.8. Let us
recall that habci is just the set {(a, b, c), (b, c, a), (c, a, b)}, hence f (habci) denotes
{f (a, b, c), f (b, c, a), f (c, a, b)}.
Lemma 3.2. If f is minimal and f (habci) ⊆ {a, b, c} then either f habci= p and
f hbaci= p or f habci≡ u and f hbaci≡ v for some u, v ∈ A.
Proof. The set f (habci) has three or one elements by Lemma 2.3. If it has three
elements then it is {a, b, c}, and then by Theorem 2.4 we have f habci= p and
f hbaci= p. In the latter case we may suppose f habci≡ a. If d /∈ f (hbaci) then f
preserves {a, b, c} and then the description of the minimal majority functions on the
three-element set yields f hbaci≡ v. If a ∈ f (hbaci) then we permute cyclically the
variables to have f (b, a, c) = a, and then g(2) preserves {a, b, c} for the superposition
g of Theorem 2.4, contradicting the minimality of f . Finally, if a /∈ f (hbaci)
but d ∈ f (hbaci) then f (hbaci) = {b, c, d}. Now we may suppose f (b, a, c) = c,
f (a, c, b) = d, f (c, b, a) = b or f (b, a, c) = b, f (a, c, b) = d, f (c, b, a) = c after a
cyclic permutation of variables. In the first case g, in the second case g(2) shows
that f is not minimal, since they preserve {a, b, c}.
(cid:3)
Claim 3.3. In either of the following four cases f is not minimal.
(1) f ∈ [4, 2, 1; ∗, ∗, ∗]
(2) f ∈ [4, 1, 2; ∗, ∗, ∗]
(3) f ∈ [4, 1, 3; ∗, ∗, ∗]
(4) f ∈ [4, 3, 1; ∗, ∗, ∗]
Proof.
(1) Lemma 3.1 shows that fzy preserves {1, 2, 3} (and hence f is not min-
imal) except when f (3, 2, 1) = 4, f (3, 2, 4) = 4 and f (3, 4, 1) = 4 or
f (3, 4, 1) = 2, f (3, 4, 2) = 4. Let us examine the set f (h213i).
It has
one or three elements by Lemma 2.3. If it is {1}, {2} or {3}, then Lemma
3.2 shows that f cannot be minimal. If we have f h213i≡ 4, then we can
compute that fzy ∈ [1, 2, 1; 2, u, 4], where u 6= 4. Depending on whether
6
T. WALDHAUSER
u = 2, 3, 1 resp. it can be shown that bfzy or fzy(y, z, fzy(x, y, z)) preserves
{1, 2, 3} or bfzy is not minimal by Lemma 3.2. Now let us suppose that
f (h213i) is a three-element set. If it is {1, 3, 4}, then Theorem 2.4 implies
f h341i= p, hence f (3, 4, 1) = 3, but we have seen that it is 4 or 2. Sim-
ilarly f (h213i) = {1, 2, 3}, {2, 3, 4} is also impossible. So f (h213i) can be
nothing else but {1, 2, 4}. Since f (3, 2, 1) = 4 there are only two possibili-
ties: f ∈ [4, 2, 1; 1, 2, 4] or f ∈ [4, 2, 1; 2, 1, 4], and then fzy ∈ [1, 2, 1; 1, 2, 4]
In both cases Lemma 3.2 yields that bfzy is not
or fzy ∈ [1, 2, 1; 2, 1, 4].
minimal, hence neither is f , and we have finished the proof.
(2) Here we can use the same argument, the only difference is that in this case
fzy ∈ [1, 1, 2; ∗, ∗, ∗].
(3) The function f (x, z, y) is isomorphic to a function which is not minimal by
case (1). (We shall note here that changing the second and third variable
does not influence the identity (∗).)
(4) Now f (x, z, y) falls under case (2) after renaming the elements of the base
(cid:3)
set.
Claim 3.4. If f ∈ [4, 3, 2; ∗, ∗, ∗] then f is not minimal.
Proof. Just as in the previous claim, we examine f (h213i). If it is {1}, {2}, {3}
or {1, 2, 3} then Lemma 3.2 shows that f is not minimal.
If f h213i≡ 4 then
g(x, y, z) = f (z, y, f (x, y, z)) ∈ [3, 3, 2; u, 2, v]. If none of u and v equals 4, then g
preserves {1, 2, 3}. If u 6= 3 then bh does, for h(x, y, z) = g(g(x, y, z), z, x). Only
u = 3, v = 4 remains, but in this case bg ∈ [3, 3, 3; 2, 4, 3], and it is not a mini-
mal function by Lemma 3.2. Now let us suppose that f (h213i) is a three-element
set containing 4.
If it is {1, 2, 4}, then according to Claim 3.3, we must have
f ∈ [4, 3, 2; 2, 1, 4] or f ∈ [4, 3, 2; 1, 2, 4], and in both cases fzy preserves {1, 2, 3}.
Similarly f (h213i) = {1, 3, 4} implies f ∈ [4, 3, 2; 4, 1, 3] or f ∈ [4, 3, 2; 4, 3, 1], and
again fzy preserves {1, 2, 3}. Finally, if f (h213i) = {2, 3, 4} then f ∈ [4, 3, 2; 3, 4, 2]
In the first case g(x, y, z) = f (z, y, f (x, y, z)) preserves
or f ∈ [4, 3, 2; 2, 4, 3].
{1, 2, 3}, in the second case bg does.
(cid:3)
Claim 3.5. If f ∈ [4, 2, 3; 2, 1, 4] or f ∈ [4, 2, 3; 4, 1, 3] then f is not minimal.
Proof. In the first case fz preserves {1, 2, 3}, in the second case fy does.
(cid:3)
Claim 3.6. If f ∈ [4, 2, 3; 2, 4, 3] then f = M2.
Proof. Lemma 2.3 yields f h234i= p and f h324i= p. Let us put g(x, y, z) =
f (x, y, f (x, y, z)). Then g ∈ [f (1, 2, 4), 2, 3; 2, f (1, 3, 4), 3].
If none of f (1, 2, 4),
f (1, 3, 4) equals 4 then g preserves {1, 2, 3}. If one of them equals 4, the other not,
then bg is not minimal by Lemma 3.2. So we must have f (1, 2, 4) = f (1, 3, 4) = 4.
In the same way we get f (2, 1, 4) = f (3, 1, 4) = 4, f (1, 4, 2) = f (1, 4, 3) = 4, etc.
by using g(x, y, z) = f (y, x, f (x, y, z)), f (x, f (x, y, z), y) etc.
(cid:3)
Claim 3.7. If f ∈ [4, 2, 3; 4, 4, 4] then f is not minimal.
Proof. If fzy(2, 1, 3) = 1 then for h(x, y, z) = fzy(z, x, fzy(x, y, z)) either bh preserves
{1, 2, 3} or fails to be minimal by Lemma 3.2. If fzy(2, 1, 3) 6= 1 then the same
holds for fzy itself, except when fzy ∈ [4, 2, 3; 2, 4, 3]. In this case Claim 3.6 yields
fzy = M2. We will see later that the clone generated by M2 contains only three
majority functions, and none of them equals f .
(cid:3)
Theorem 3.8. Any minimal nonconservative majority function on A which satis-
fies (∗) is isomorphic to M2 or it belongs to the set S.
MINIMAL CLONES GENERATED BY MAJORITY OPERATIONS
7
Proof. Let f be a function as stated in the theorem. According to Claim 3.3 and
Claim 3.4, for every a, b, c if neither f habci= p nor f habci≡ u holds, then we must
have that on two of the three triplets of habci the value of f equals the first variable,
while on the third one f equals d. If f /∈ S then this case really appears, so we
can suppose (after an isomorphism if necessary) that f (1, 2, 3) = 4, f (2, 3, 1) = 2,
f (3, 1, 2) = 3. Now if 4 /∈ f (h213i) then we get a contradiction by Lemma 3.2. If
f h213i≡ 4 than Claim 3.7 implies that f is not minimal. So f (h213i) must be a
three-element set containing 4, and then again by Claims 3.3 and 3.4 we must have
f ∈ [4, 2, 3; 2, 1, 4], f ∈ [4, 2, 3; 4, 1, 3], or f ∈ [4, 2, 3; 2, 4, 3]. The first two of these
is impossible by Claim 3.5, and in the third case Claim 3.6 shows that f equals
M2.
(cid:3)
3.2.
In this section we are going to search for the minimal functions of the set S.
The conservative ones are already described, so we deal only with nonconservative
functions. We assume f to be such a function and we will prove several properties of
f , until we find that only a few functions (essentially two) possess these properties,
and these happen to be minimal.
Definition 3.9. A ternary function g is said to be cyclically commutative if it is
invariant under the cyclic permutation of variables, i.e.
Claim 3.10. The function f is cyclically commutative.
g(x, y, z) ≈ g(y, z, x) ≈ g(z, x, y).
Proof. For contradiction, let us suppose f h124i= p, and then by Lemma 3.2 we
have also f h214i= p. Since f is not conservative, we may also suppose (with-
out loss of generality) that f h123i≡ 4, and again by Lemma 3.2 we must have
f h213i≡ u. First let us suppose u 6= 4. Then fzy preserves {1, 2, 3} except when
f (3, 1, 2) = f (3, 1, 4) = f (3, 4, 2) = 4 or f (3, 1, 2) = f (3, 1, 4) = f (3, 4, 1) = 4 and
f (3, 4, 2) = 1. In the first case bfzy ∈ [1, 2, 4; u, u, u] so it is not a minimal function
by Lemma 3.2, while in the second case fzy ∈ [1, 1, 4; u, u, u], hence bfzy preserves
{1, 2, 3}. If u = 4 then we have also fzy ∈ [1, 2, 4; ∗, ∗, ∗] or fzy ∈ [1, 1, 4; ∗, ∗, ∗] or
fzy(h123i) ⊆ {1, 2, 3}. The first case is impossible, since then Theorem 3.8 implies
that bfzy is isomorphic to M2, but then f /∈ [ bfzy] shows that f is not minimal.
(In fact, the clone generated by M2 contains no function from S except for the
first projection.) For h213i we have also three possibilities: fzy ∈ [∗, ∗, ∗; 2, 1, 4],
fzy ∈ [∗, ∗, ∗; 2, 2, 4] and fzy(h213i) ⊆ {1, 2, 3}. The first one of these is impossible
for the same reason as above. In the remaining cases bfzy preserves {1, 2, 3}.
(cid:3)
In the following we suppose f to be a nonconservative cyclically commutative
minimal majority function on A. In [3] these are determined by computer, here we
give a straightforward description. We again suppose f h123i≡ 4, and f h213i≡ u.
First we show that f preserves {1, 2, 4}, {1, 3, 4} and {2, 3, 4}.
Claim 3.11. The only subset of A not preserved by f is {1, 2, 3}.
Proof. We separate two cases upon u.
Case 1. f h213i≡ u 6= 4. For contradiction let us suppose that f does not preserve
In the first case f (y, x, f (x, y, z)(2)
{1, 2, 4}. Then f h214i≡ 3 or f h124i≡ 3.
preserves {1, 2, 3} or {1, 2, 4}, in the second case f (x, fz(x, y, z), z)(2) or fz preserves
{1, 2, 3} depending on whether 4 ∈ {f (2, 3, 4), f (3, 1, 4)} or not.
Case 2. f h213i≡ 4. What we have already proved of this claim means that if
f habci≡ d and f hbaci≡ a then f preserves the other three subsets of A, namely
{a, b, d}, {a, c, d}, {b, c, d}. So if we again suppose that f does not preserve {1, 2, 4}
8
T. WALDHAUSER
then we must have f h124i≡ 3 and f h214i≡ 3. Similarly, f h234i≡ 1 if and
only if f h324i≡ 1 and f h134i≡ 2 iff f h314i≡ 2. One can check that fz or bfz
preserves {1, 2, 3} or {1, 2, 4} except for only two functions (up to isomorphism and
permutation of variables). For both of them f h134i≡ 3 and f h314i≡ 4, and for one
we have f h234i≡ 1 and f h324i≡ 1, for the other one f h234i≡ 4 and f h324i≡ 3. In
both cases bfz ∈ [4, 4, 4; 3, 3, 3], hence by Case 1 it preserves {1, 2, 4}. We supposed
that f does not preserve this set, so f /∈ [bfz] and this contradicts the minimality of
f .
(cid:3)
We have proved that if f ∈ S is a minimal function, then f is cyclically commu-
tative and preserves all but one three-element subsets of A. In the following two
claims -- as usually -- we suppose that f preserves {1, 2, 4}, {1, 3, 4}, {2, 3, 4} and
f h123i≡ 4, f h213i≡ u. Depending on whether u = 4 or not, we will finally reach
M1 and M3.
Claim 3.12. If f h213i≡ u 6= 4 then hA; f i ∼= hA; M3i.
Proof. We can suppose f h213i≡ 3 without loss of generality. We also suppose
f h124i≡ 4, f h314i≡ 4, f h234i≡ 4, since otherwise bfzy preserves {1, 2, 3}. For
g(x, y, z) = f (y, x, f (x, y, z)) we have g ∈ [f (1, 2, 4), f (3, 2, 4), f (1, 3, 4); 4, 3, 3]. If
none of f (1, 2, 4), f (3, 2, 4), f (1, 3, 4) equals 4, then g(3) preserves {1, 2, 3}.
If
there is a 4 amongst them, but 3 does not appear, then we put h(x, y, z) =
f (g(x, y, z), g(z, y, x), g(x, z, y)) and one calculate that the range of h(2) does not
contain 4, hence f is not minimal by (2.3). Only nine functions remain;
for
two of them g is isomorphic to M3, hence f is also.
(The clone generated by
M3 contains only two functions from S, and only one of them can be equal to
f .)
if
g ∈ [1, 3, 4; 4, 4, 3] then g(g(x, y, z), y, g(y, z, x))(2) does so. In the remaining five
cases {1, 2, 3} is preserved by f (g(x, y, z), g(z, x, y), g(y, z, x)).
(cid:3)
Claim 3.13. If f h213i≡ 4 then hA; f i ∼= hA; M1i.
If g ∈ [2, 4, 3; 4, 4, 3] then g(y, g(y, z, x), g(x, y, z))(2) preserves {1, 2, 3},
Proof. Let U = {f (1, 2, 4), f (3, 1, 4), f (2, 3, 4)} and V = {f (2, 1, 4), f (1, 3, 4),
f (3, 2, 4)}. If U 6= {4} and V 6= {4}, then bfzy preserves {1, 2, 3}. If U = V = {4}
then f = M1. Now let us suppose U = {4} 6= V . If 4 /∈ V then bfzy is not minimal
by Lemma 3.2. If this is not the case then by the previous claim bfzy is isomorphic
to M3, but the clone [M3] contains no function which isomorphic to f . The case
V = {4} 6= U is similar.
(cid:3)
3.3.
We have now -- up to isomorphism -- only three functions: M1, M2, M3, and these
generate minimal clones.
Theorem 3.14. M1, M2, M3 are minimal functions on {1, 2, 3, 4}.
Proof. The proof is the same for all the three functions, so let f be any of them. This
function preserves the equivalence relation whose blocks are {1, 4}, {2}, {3}, and its
range does not contain the element 1. According to (1.1) and (2.3), the same is valid
for an arbitrary majority function g in [f ]. These properties determine g {1,2,3}
provided g {2,3,4} is given. Since f preserves {2, 3, 4} and f {2,3,4} is minimal,
there exists an h ∈ [g] such that h {2,3,4}=f {2,3,4}. Now h has also the above
mentioned two properties, so h {1,2,3} is uniquely determined:
it can be nothing
else than f {1,2,3}. On {1, 2, 4} and on {1, 3, 4} f is constant 4, consequently so
are g and h, hence h = f . Thus, for arbitrary g ∈ [f ], f ∈ [g] also holds, proving
that [f ] is a minimal clone.
(cid:3)
MINIMAL CLONES GENERATED BY MAJORITY OPERATIONS
9
Remark 3.15. From the proof it is clear, that restriction to {2, 3, 4} gives a one-to-
one correspondence between the majority functions in [Mi] and [mi]. Hence these
clones contain also 1,3 or 8 majority functions. They can be seen in the following
table.
(1, 2, 3)
(2, 3, 1)
(3, 1, 2)
(2, 1, 3)
(1, 3, 2)
(3, 2, 1)
{1, 2, 4}
{1, 3, 4}
(4, 2, 3)
(2, 3, 4)
(3, 4, 2)
(2, 4, 3)
(4, 3, 2)
(3, 2, 4)
M1 M2
4
4
2
4
3
4
4
2
4
4
3
4
4
4
4
4
4
4
2
4
3
4
4
2
4
4
4
3
M3
3
3
3
4
4
4
4
4
3
3
3
4
4
4
2
3
4
4
3
2
4
4
2
3
4
4
3
2
3
4
2
3
2
4
4
4
3
4
2
3
2
4
3 4
4 3
3 3
3 4
4 4
4 3
4 4
4 4
3 4
4 3
3 3
3 4
4 4
4 3
3 4
3 4
4 4
4 3
3 3
4 3
4 4
4 4
3 4
3 4
4 4
4 3
3 3
4 3
4
3
4
4
3
3
4
4
4
3
4
4
3
3
3 4
4 4
4 3
3 3
3 4
4 3
4 4
4 4
3 4
4 4
4 3
3 3
3 4
4 3
References
[1] B. Cs´ak´any, All minimal clones on the three-element set, Acta Cybernet. 6 (1983), no. 3,
227 -- 238.
[2] B. Cs´ak´any, On conservative minimal operations, Lectures in Universal Algebra (Szeged,
1983), Colloq. Math. Soc. J´anos Bolyai, 43, North-Holland, Amsterdam, 1986, 49 -- 60.
[3] Z. Csibor: Majority functions, Masters Thesis (Szeged, 1988) (In Hungarian).
[4] E. Post, The two-valued iterative systems of mathematical logic, Annals of Mathematics Stud-
ies, no. 5, Princeton University Press, Princeton, 1941.
[5] R. W. Quackenbush: Some remarks on categorical algebras, Algebra Universalis 2 (1972), 246.
[6] R. W. Quackenbush: A survey of minimal clones, Aequationes Mathematicae 50 (1995), 3-16.
[7] I. G. Rosenberg, Minimal clones I. The five types, Lectures in Universal Algebra (Szeged,
1983), Colloq. Math. Soc. J´anos Bolyai, 43, North-Holland, Amsterdam, 1986, 405 -- 427.
[8] B. Szczepara, Minimal clones generated by groupoids, Ph.D. Thesis, Universit´e de Montr´eal,
1995.
[9] ´A. Szendrei, Clones in Universal Algebra, S´eminaire de Math´ematiques Sup´erieures, 99,
Presses de L'Universit´e de Montr´eal, 1986.
Bolyai Institute, University of Szeged, Aradi v´ertan´uk tere 1, H6720, Szeged, Hun-
gary
E-mail address: [email protected]
|
0904.0676 | 3 | 0904 | 2010-03-23T14:29:01 | Quivers with potentials and their representations II: Applications to cluster algebras | [
"math.RA",
"math.RT"
] | We continue the study of quivers with potentials and their representations initiated in the first paper of the series. Here we develop some applications of this theory to cluster algebras. As shown in the "Cluster algebras IV" paper, the cluster algebra structure is to a large extent controlled by a family of integer vectors called g-vectors, and a family of integer polynomials called F-polynomials. In the case of skew-symmetric exchange matrices we find an interpretation of these g-vectors and F-polynomials in terms of (decorated) representations of quivers with potentials. Using this interpretation, we prove most of the conjectures about g-vectors and F-polynomials made in loc. cit. | math.RA | math |
QUIVERS WITH POTENTIALS AND THEIR REPRESENTATIONS II:
APPLICATIONS TO CLUSTER ALGEBRAS
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Abstract. We continue the study of quivers with potentials and their representations
initiated in the first paper of the series. Here we develop some applications of this theory
to cluster algebras. As shown in the "Cluster algebras IV" paper, the cluster algebra
structure is to a large extent controlled by a family of integer vectors called g-vectors,
and a family of integer polynomials called F -polynomials. In the case of skew-symmetric
exchange matrices we find an interpretation of these g-vectors and F -polynomials in terms
of (decorated) representations of quivers with potentials. Using this interpretation, we
prove most of the conjectures about g-vectors and F -polynomials made in loc. cit.
Contents
Introduction
1.
2. Background on g-vectors and F -polynomials
3. F -polynomials of quiver representations
4. Background on quivers with potentials and their representations
5. QP-interpretation of g-vectors and F -polynomials
6. Mutations preserve homomorphisms modulo confined ones
7. The E-invariant
8. Lower bounds for the E-invariant
9. Applications to cluster algebras
10. Homological interpretation of the E-invariant
References
1
6
10
13
18
25
27
30
34
35
43
1. Introduction
This paper continues our study of quivers with potentials and their representations ini-
tiated in [9]. Here we develop some applications of this theory to the theory of cluster
algebras. As shown in [12], the structure of cluster algebras is to a large extent controlled
by a family of integer vectors called g-vectors, and a family of integer polynomials called
F -polynomials. In the case of skew-symmetric exchange matrices (the terminology will be
Date: April 3, 2009; revised November 13, 2009; final version March 23, 2010.
2000 Mathematics Subject Classification. Primary 16G10, Secondary 16G20, 16S38, 16D90.
Research of H. D. supported by the NSF grants DMS-0349019 and DMS-0901298. Research of J. W.
supported by the NSF grant DMS-0600229. Research of A. Z. supported by the NSF grants DMS-0500534
and DMS-0801187, and by a Humboldt Research Award.
1
2
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
recalled later), we find an interpretation of g-vectors and F -polynomials in terms of rep-
resentations of quivers with potentials. Using this interpretation, we prove most of the
conjectures about g-vectors and F -polynomials made in [12].
k
Now we describe the main results of the paper in more detail. Fix a positive integer n. As
in [11] and [12, Definition 2.8], we work with the n-regular tree Tn whose edges are labeled
by the numbers 1, . . . , n, so that the n edges emanating from each vertex receive different
t′ to indicate that vertices t, t′ ∈ Tn are joined by an edge labeled
labels. We write t
by k. We also fix a vertex t0 ∈ Tn and a skew-symmetrizable integer n × n matrix B = (bi,j)
(recall that this means that dibi,j = −djbj,i for some positive integers d1, . . . , dn). We refer
to B as the exchange matrix at t0. To t0 and B we associate a family of integer vectors
gℓ;t = gB;t0
ℓ;t ∈ Z[u1, . . . , un]
(F -polynomials) in n independent variables u1, . . . , un; here ℓ = 1, . . . , n, and t ∈ Tn. Both
families can be defined via the recurrence relations on the tree Tn given by (2.1) -- (2.3) and
(2.4) -- (2.6) below.
ℓ;t ∈ Zn (g-vectors) and a family of integer polynomials Fℓ;t = F B;t0
Now we state some conjectures from [12].
Conjecture 1.1. ([12, Conjecture 5.4]) Each polynomial F B;t0
ℓ;t
has constant term 1.
In view of [12, Proposition 5.3], Conjecture 1.1 is equivalent to the following.
ℓ;t
Conjecture 1.2. ([12, Conjecture 5.5]) Each polynomial F B;t0
has a unique monomial of
maximal degree. Furthermore, this monomial has coefficient 1, and it is divisible by all the
other occurring monomials.
Conjecture 1.3. ([12, Conjecture 6.13]) For every t ∈ Tn, the vectors gB;t0
n;t are
1;t
sign-coherent, i.e., for any i = 1, . . . , n, the i-th components of all these vectors are either
all nonnegative, or all nonpositive.
Conjecture 1.4. ([12, Conjecture 7.10(2)]) For every t ∈ Tn, the vectors gB;t0
1;t
gB;t0
n;t
form a Z-basis of the lattice Zn.
, . . . , gB;t0
, . . . ,
Conjecture 1.5. ([12, Conjecture 7.10(1)]) Suppose we have
Xi∈I
aigB;t0
i;t =Xi∈I ′
igB;t0
a′
i;t′
for some t, t′ ∈ Tn, some nonempty subsets I, I ′ ⊆ {1, . . . , n} and some positive integers ai
and a′
i. Then there is a bijection σ : I → I ′ such that, for every i ∈ I, we have
ai = a′
σ(i),
gB;t0
i;t = gB;t0
σ(i);t′, F B;t0
i;t = F B;t0
σ(i);t′.
In particular, for given B and t0, each polynomial F B;t0
i;t
is determined by the vector gB;t0
i;t
.
To state our last conjecture, we need to recall the matrix mutation introduced in [11].
For any k = 1, . . . , n, we define an integer n × n matrix µk(B) = (b′
i,j) by setting
bi,j + [bi,k]+ [bk,j]+ − [−bi,k]+ [−bk,j]+ otherwise,
if i = k or j = k;
(1.1)
i,j =(−bi,j
b′
where we use the notation
(1.2)
[b]+ = max(b, 0).
QUIVERS WITH POTENTIALS II
3
Conjecture 1.6. ([12, Conjecture 7.12]) Let t0
let B′ = µk(B). Then, for any t ∈ Tn and ℓ = 1, . . . , n, the g-vectors gB;t0
and gB ′;t1
t1 be two adjacent vertices in Tn, and
ℓ;t = (g1, . . . , gn)
n) are related as follows:
ℓ;t = (g′
1, . . . , g′
k
(1.3)
j =(−gk
g′
gj + [bj,k]+gk − bj,k min(gk, 0)
if j = k;
if j 6= k.
We can now state one of our main results.
Theorem 1.7. The conjectures 1.1 -- 1.6 hold under the assumption that the exchange
matrix B is skew-symmetric.
Remark 1.8. As explained in [12, Remark 7.11], Conjectures 1.1 and 1.5 imply the lin-
ear independence of cluster monomials in any cluster algebra satisfying a mild additional
condition [12, (7.10)].
Remark 1.9. The above conjectures were established in [13] under some additional condi-
tions (that the cluster algebras in question admit a certain categorification). Our method
described below has an advantage that the only condition we need is that the matrix B is
skew-symmetric.
As mentioned already, our proof of Theorem 1.7 is based on interpreting g-vectors and
F -polynomials in terms of representations of quivers with potentials. First of all, a skew-
symmetric integer n × n matrix B can be encoded by a quiver Q(B) without loops and
oriented 2-cycles on the set of vertices [1, n] = {1, . . . , n}. This is done as follows:
(1.4)
for any two vertices i 6= j there are [bi,j]+ arrows from j to i in Q(B).
As is customary these days, we represent a quiver by a quadruple (Q0, Q1, h, t) consisting of a
pair of finite sets Q0 (vertices) and Q1 (arrows) supplied with two maps h : Q1 → Q0 (head )
and t : Q1 → Q0 (tail ); every arrow a ∈ Q1 is viewed as a directed edge a : t(a) → h(a).
For the quiver Q(B), the vertex set Q0 is identified with [1, n].
Recall that a representation M of a quiver Q is specified by a family of finite-dimensional
vector spaces (M(i))i∈Q0 (for simplicity we work over C) and a family of linear maps a =
aM : M(t(a)) → M(h(a)) for a ∈ Q1. The dimension vector dM of M is given by
(1.5)
dM = (dim M(1), . . . , dim M(n)).
For every integer vector e = (e1, . . . , en), we denote by Gre(M) the quiver Grassmannian
of subrepresentations N ⊆ M with dN = e.
In simple terms, an element of Gre(M) is
an n-tuple (N(1), . . . , N(n)), where each N(i) is a subspace of dimension ei in M(i), and
aM (N(j)) ⊆ N(i) for any arrow a : j → i. Thus, Gre(M) is a closed subvariety of the
i=1 Grei(M(i)), hence a projective algebraic variety.
Let χ(Gre(M)) denote the Euler-Poincar´e characteristic of Gre(M) (see e.g., [14, Sec-
tion 4.5]). We associate to a quiver representation M the polynomial FM ∈ Z[u1, . . . , un]
given by
product of ordinary GrassmanniansQn
(1.6)
FM (u1, . . . , un) =Xe
χ(Gre(M))
uei
i .
nYi=1
4
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
We refer to FM as the F -polynomial of M.
It is immediate from (1.6) that every polynomial FM satisfies properties in Conjectures 1.1
and 1.2. Thus, to prove these conjectures for any skew-symmetric matrix B, it suffices to
construct, for t0, ℓ and t as above, a representation M = M B;t0
of Q(B) such that
ℓ;t
(1.7)
F B;t0
ℓ;t = FM .
We do this in Theorem 5.1 using mutations of quivers with potentials and their representa-
tions introduced and studied in [9].
To prove the conjectures involving g-vectors, we need to consider quiver representations
equipped with some extra structure. First, following [16], we work with decorated represen-
tations M = (M, V ), where M is a representation of Q(B), and V = (V (i))i∈Q0 is a family
of finite-dimensional C-vector spaces, with no maps attached. Second, M must be nilpotent,
that is, annihilated by all sufficiently long paths in Q(B). Finally and most importantly,
the action of arrows in M must satisfy the relations from the Jacobian ideal of a generic
potential on Q(B). The corresponding setup developed in [9] will be recalled in Section 4,
here we just describe a general form of the relations. For every two arrows a, b ∈ Q1 with
h(a) = t(b), a generic potential S on Q(B) gives rise to an element ∂ba(S) of the complete
path algebra of Q(B): this is a (possibly infinite) linear combination of paths from h(b) to
t(a). For every k ∈ Q0, these elements give rise to the triangle of linear maps
(1.8)
M(k)
αk
:uuuuuuuuu
βk
$JJJJJJJJJ
Min(k)
γk
Mout(k)
Here the spaces Min(k) and Mout(k) are given by
Min(k) = Mh(a)=k
M(t(a)), Mout(k) = Mt(b)=k
M(h(b)),
(1.9)
(1.10)
the maps αk and βk are given by
αk = Xh(a)=k
aM ,
βk = Xt(b)=k
bM ,
and, for each a, b ∈ Q1 with h(a) = t(b) = k, the component γa,b : M(h(b)) → M(t(a)) of
γk is given by
(1.11)
γa,b = (∂baS)M .
In these terms, the relations on M imposed by the choice of S are just the following:
(1.12)
αk ◦ γk = 0,
γk ◦ βk = 0.
We refer to a decorated representation with these properties as a QP-representation (for
"quivers with potentials").
Now we define the g-vector gM = (g1, . . . , gn) ∈ Zn of a QP-representation M = (M, V )
by setting
(1.13)
gk = dim ker γk − dim M(k) + dim V (k) .
$
:
o
o
QUIVERS WITH POTENTIALS II
5
As a first step towards proving Conjectures 1.3 - 1.6 for B skew-symmetric, in Theorem 5.1
ℓ;t of
we construct, for t0, ℓ and t as above, an indecomposable QP-representation M = MB;t0
Q(B) such that
(1.14)
gB;t0
ℓ;t = gM
(note that M = (M, V ), where the quiver representation M = M B;t0
satisfies (1.7)).
ℓ;t
Our main tool in working with QP-representations is the mutation operation M 7→
µk(M) (for each k ∈ Q0) sending QP-representations of the quiver Q(B) to those of
Q(µk(B)). This operation was introduced and studied in [9], where it was shown in partic-
ular that µk sends indecomposable QP-representations into indecomposable ones. In terms
of the mutations, the family of QP-representations MB;t0
is determined by the following
ℓ;t
two properties:
• For t = t0, we have
(1.15)
MB;t0
ℓ;t0 = S −
ℓ ,
the negative simple QP-representation such that the only nonzero space among the
M(i) and V (i) is V (ℓ) = C.
• If t0
k
t1 in Tn, and B′ = µk(B) then
(1.16)
MB ′;t1
ℓ;t = µk(MB;t0
ℓ;t ).
In contrast with the situation for F -polynomials, where the interpretation (1.7) imme-
diately implies Conjectures 1.1 and 1.2, deducing Conjectures 1.3 -- 1.6 from (1.14) re-
quires further work. The main new ingredient is the following integer-valued function on
for a QP-representation M = (M, V ) of a quiver Q, we define the
QP-representations:
E-invariant by
(1.17)
E(M) = dim HomQ(M, M) +
gk dim M(k),
nXk=1
Note that in view of (1.15) and (1.16), the QP-representations MB;t0
ℓ;t
As a consequence, for each M of the form MB;t0
, the right hand side of (1.18) is equal to 0,
ℓ;t
and this information turns out to be exactly what we need for proving Conjectures 1.3 - 1.6.
can be characterized
as those obtained by a sequence of mutations from a negative simple representation. We
conjecture that this family coincides with the family of indecomposable QP-representations
M such that E(M) = 0. As a possible step towards proving this conjecture, in Section 10
where gk is given by (1.13), and HomQ stands for the space of homomorphisms of quiver
In Theorem 7.1 we prove that E(M) is invariant under mutations, i.e.,
representations.
for every k we have E(µk(M)) = E(M). Then it follows from (1.15) and (1.16) that
E(MB;t0
ℓ;t ) = 0 for all ℓ and t.
Since the numbers gk may be negative, it is not a priori clear that E(M) takes nonnegative
values. We prove this property in Theorem 8.1, establishing the following much sharper
lower bound:
(1.18)
(dim ker βk · dim(ker γk/ im βk) + dim M(k) · dim V (k)).
E(M) ≥ Xk∈Q0
6
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
we develop a homological interpretation of E(M) in the case where the potential is finite
and the Jacobian algebra is finite dimensional. This interpretation is based on constructing
a projective presentation for QP-representations, see Proposition 10.4.
The paper is organized as follows. Sections 2 -- 4 are devoted to preliminaries. The
necessary background on cluster algebras is recalled in Section 2. In Section 3 we collect
some general properties of F -polynomials of quiver representations to be used later. We
conclude this section with two examples, showing that a quiver Grassmannian may be
singular, and that it may have negative Euler characteristic. The necessary background
from [9] on quivers with potentials (QP's) and their representations is collected in Section 4.
Section 5 contains the first important new result of the paper -- Theorem 5.1. It asserts
that the family of QP-representations recursively defined by conditions (1.15) and (1.16)
provides a representation-theoretic interpretation given by (1.7) and (1.14) of F -polynomials
and g-vectors arising in the theory of cluster algebras. As a consequence, we obtain in
Corollary 5.3 a formula for cluster variables in the coefficient-free cluster algebra, which
generalizes the Caldero-Chapoton formula in [7, Theorem 3].
In Section 6 we prove Proposition 6.1, a technical result preparing the ground for the later
proof of the invariance under mutations of the function E(M) given by (1.17). Roughly
speaking, Proposition 6.1 says that the mutation at a vertex k preserves the space of ho-
momorphisms between any two QP-representations modulo the homomorphisms "confined"
to k. This result of independent interest was already established in [3, Theorem 7.1] but
the present proof seems to be much simpler.
In the rest of Section 6 we show that the
isomorphism in Proposition 6.1 can be stated in a functorial way.
The main result in Section 7 is Theorem 7.1 establishing in particular the invariance of
E(M) under mutations. Another useful result there is Proposition 7.3 saying that E(M)
is invariant under passing to the dual QP-representation of the opposite QP.
In Section 8 we prove the bound (1.18) (Theorem 8.1). The proof of Theorem 1.7 is
obtained by combining this result with the results in the preceding sections; this is done
in Section 9. The concluding Section 10 is devoted to the above-mentioned homological
interpretation of the E-invariant of QP-representations.
Acknowledgement. The authors are grateful to Grzegorz Bobi´nski for providing useful
references, and to an anonymous referee for several helpful suggestions.
2. Background on g-vectors and F -polynomials
First of all, we recall that the same rule as in (1.1) defines the matrix mutation µk for
any integer m × n matrix B = (bi,j) with m ≥ n, and any k = 1, . . . , n. This is an
involution on the set of integer m × n matrices. We call the top n × n submatrix B of B
the principal part of B; then µk(B) is the principal part of µk( B). Note also that, if B is
skew-symmetrizable, that is, dibi,j = −djbj,i for some positive integers d1, . . . , dn, then the
same choice of d1, . . . , dn makes µk(B) skew-symmetrizable as well. In particular, if B is
skew-symmetric then µk(B) is also skew-symmetric.
We say that a family of m × n integer matrices ( B(t)t∈Tn) is a skew-symmetrizable (resp.
skew-symmetric) matrix pattern of format m × n on Tn if the principal part B(t) of each
QUIVERS WITH POTENTIALS II
7
k
B(t) is skew-symmetrizable (resp. skew-symmetric), and we have B(t′) = µk( B(t)) when-
t′. Clearly, such a pattern is uniquely determined by each of its matrices B(t0),
ever t
which can be chosen arbitrarily with the only condition that its principal part is skew-
symmetrizable (resp. skew-symmetric).
Now choose any skew-symmetrizable n × n integer matrix B and any vertex t0 ∈ Tn. We
associate to B and t0 the skew-symmetrizable matrix pattern of format 2n × n such that
B(t0) = (bi,j) has principal part B, and its bottom part is the n × n identity matrix, that is,
bn+i,j = δi,j for i, j = 1, . . . , n; we refer to this pattern as the principal coefficients pattern
associated to B and t0. Let us denote this pattern simply as ( B(t) = (bi,j(t)))t∈Tn (with the
understanding that B and t0 are fixed).
Now, according to [12, Proposition 6.6], the vectors gℓ;t = gB;t0
ℓ;t
can be defined by the
initial conditions
(2.1)
gℓ;t0 = eℓ
(ℓ = 1, . . . , n)
together with the recurrence relations
gℓ;t′ = gℓ;t
for ℓ 6= k;
gk;t′ = −gk;t +
[bi,k(t)]+gi;t −
[bn+i,k(t)]+bi
nXi=1
nXi=1
t′ in Tn . Here e1, . . . , en are the unit vectors in Zn, and b1, . . . , bn are
for every edge t
the columns of B.
k
Similarly, by [12, Proposition 5.1], the polynomials Fℓ;t = F B;t0
(u1, . . . , un) can be defined
ℓ;t
by the initial conditions
(2.4)
Fℓ;t0 = 1 (ℓ = 1, . . . , n) ,
together with the recurrence relations
(2.2)
(2.3)
(2.5)
(2.6)
6.7]).
for ℓ 6= k;
i=1 u[bn+i,k(t)]+
i
Fℓ;t′ = Fℓ;t
Fk;t′ = Qn
for every edge t
k
t′ in Tn .
Fk;t
F [bi,k(t)]+
i;t
+Qn
k;t1 = −ek +Pn
For instance, if t1
Here is a specific example for the cluster algebra of type A2 (cf. [12, Examples 2.10, 3.4,
k;t1 = uk + 1.
i=1[−bi,k]+ei, and F B;t0
t0 , then gB;t0
k
i=1 u[−bn+i,k(t)]+
i
F [−bi,k(t)]+
i;t
,
Example 2.1. Let n = 2. The tree T2 is an infinite chain. We denote its vertices by
. . . , t−1 , t0 , t1 , t2 , . . . , and label its edges as follows:
(2.7)
· · ·
2
t−1
1
t0
2
Let B = [ 0 1
Table 1 (the last column will be explained later).
−1 0 ]. The g-vectors gℓ;t = gB;t0
ℓ;t
2
1
1
t3
t2
t1
and F -polynomials Fℓ;t = F B;t0
· · · .
ℓ;t
are shown in
Observing that B(t5) is obtained from B(t0) by interchanging the two columns, and
comparing g-vectors and F -polynomials at t0 and t5, we obtain the following periodicity
8
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
t
B(t)
g1;t
g2;t
F1;t
F2;t
h1;t
h2;t
t1
t0
1 0
1 0
−1 0
1 0
1
−1 0
−1 0
(cid:20) 0 1
0 1(cid:21)
0 −1(cid:21)
(cid:20) 0 −1
t2 (cid:20) 0
0 −1(cid:21)
t3 (cid:20) 0 −1
−1 1 (cid:21)
0 (cid:21)
t4 (cid:20) 0
1 0 (cid:21)
(cid:20) 0 −1
1
−1 0
1 −1
1
1
0
−1 0
1 0
0 1
t5
(cid:20)1
0(cid:21)
(cid:20)1
0(cid:21)
(cid:20)−1
0 (cid:21)
(cid:20)−1
0 (cid:21)
(cid:20)0
1(cid:21)
(cid:20)0
1(cid:21)
1
1
1
u2 + 1
(cid:20)0
0(cid:21)
(cid:20)0
0(cid:21)
(cid:20)0
1(cid:21)
(cid:20)0
(cid:20) 0
0(cid:21)
(cid:20) 0
−1(cid:21)
−1(cid:21)
−1(cid:21) u1u2 + u1 + 1 u2 + 1 (cid:20)−1
(cid:20) 0
0 (cid:21) (cid:20) 0
−1(cid:21)
1 (cid:21) u1u2 + u1 + 1 u1 + 1 (cid:20)−1
(cid:20)−1
0 (cid:21) (cid:20)−1
0 (cid:21)
(cid:20)0
(cid:20)−1
(cid:20)−1
1 (cid:21)
0(cid:21)
0 (cid:21)
(cid:20)0
(cid:20)0
(cid:20)1
0(cid:21)
0(cid:21)
0(cid:21)
u1 + 1
1
1
1
Table 1. g-vectors, F -polynomials, and h-vectors in type A2
property:
gℓ;tm+5 = g3−ℓ;tm, Fℓ;tm+5(u1, u2) = F3−ℓ;tm(u2, u1)
(m ∈ Z) .
Returning to the general situation, we note that the definition makes it clear that all
Fℓ;t(u1, . . . , un) are rational functions with coefficients in Q. The following stronger state-
ment was proven in [12, Propositions 3.6, 5.2].
Proposition 2.2. Each of the rational functions Fℓ;t(u1, . . . , un) is a polynomial with integer
coefficients, which is not divisible by any ui.
k
ℓ;t
ℓ;t
and F B;t0
We now fix ℓ and t, and discuss the dependency of gB;t0
ℓ;t
on the initial vertex t0
and the initial exchange matrix B. More precisely, choose some k ∈ [1, n], and suppose that
t0
, and the polynomials
F B;t0
t1 and B1 = µk(B). We will relate the vectors gB;t0
ℓ;t
and F B1;t1
. This requires some preparation.
ℓ;t
Recall that a semifield (P, ·, +) is an abelian multiplicative group (P, ·) endowed with a
binary operation of addition which is commutative, associative, and distributive with respect
to the multiplication in P. With every finite family of indeterminates u1, . . . , uℓ one can
associate two semifields: the universal semifield Q sf(u1, . . . , uℓ), and the tropical semifield
Trop(u1, . . . , uℓ) (cf. [12, Definitions 2.1, 2.2]). Recall that Q sf(u1, . . . , uℓ) is the set of all
rational functions in u1, . . . , uℓ which can be written as subtraction-free rational expressions,
while Trop(u1, . . . , uℓ) is the multiplicative group of Laurent monomials ua1
ℓ with the
1 · · · uaℓ
and gB1;t1
ℓ;t
QUIVERS WITH POTENTIALS II
9
addition ⊕ given by
(2.8)
Yj
uaj
j ⊕Yj
ubj
j =Yj
umin(aj ,bj)
j
.
Since (2.6) does not involve subtraction, every F -polynomial F B;t0
(u1, . . . , un) belongs to
Q sf(u1, . . . , un) (although it is still not known in general whether all these polynomials have
positive coefficients). Note that every subtraction-free rational expression F (u1, . . . , un) (in
particular, every F B;t0
) can be evaluated at any n-tuple of elements y1, . . . , yn of an arbitrary
semifield P. We denote the result of this evaluation by F P(yi ← ui). Using this notation,
we denote by hB;t0
ℓ;t = (h1, . . . , hn) the integer vector given by
ℓ;t
ℓ;t
(2.9)
xh1
1 · · · xhn
n = F B;t0
ℓ;t
Trop(x1,...,xn)(x−1
x[−bj,i]+
j
← ui).
i Yj6=i
Example 2.3. In the situation of Example 2.1, the vectors hℓ;t = hB;t0
ℓ;t are given in the last
column of Table 1. In this case, the formula (2.9) for the vector hℓ;t = (h1, h2) takes the
form
xh1
1 xh2
2 = Fℓ;tTrop(x1,x2)(x−1
1 x2, x−1
2 ).
For example, since F1;t2 = u1u2 + u1 + 1, we obtain
2 ) = x−1
F1;t2Trop(x1,x2)(x−1
1 x2, x−1
1 ⊕ x−1
1 x2 ⊕ 1 = x−1
1 ,
hence h1;t2 =(cid:20)−1
0 (cid:21).
Next we recall the Y -seeds and their mutations (see [12, Definitions 2.3, 2.4]). A (labeled)
Y -seed in a semifield P is a pair (y, B), where
• y = (y1, . . . , yn) is an n-tuple of elements of P, and
• B = (bi,j) is an n×n skew-symmetrizable integer matrix.
The Y -seed mutation at k ∈ [1, n] transforms (y, B) into a Y -seed µk(y, B) = (y′, B′),
where B′ = µk(B) is given by (1.1), and the n-tuple y′ = (y′
n) is given by
1, . . . , y′
(2.10)
i =(y−1
k
y′
yiy[bk,i]+
k
if i = k;
(yk + 1)−bk,i
if i 6= k.
The following result is immediate from [12, Proposition 6.8, formulas (6.26),(6.28)].
Proposition 2.4. Suppose t0
obtained from (y, B) by the mutation at k. Let hk (resp. h′
vector hB;t0
ℓ;t
are related by
t1 in Tn, and the Y -seed (y′, B1) in Q sf(y1, . . . , yn) is
k) be the k-th component of the
1, . . . , g′
n)
ℓ;t = (g1, . . . , gn) and gB1;t1
). Then the g-vectors gB;t0
(resp. hB1;t1
ℓ;t = (g′
ℓ;t
k
(2.11)
We also have
(2.12)
j =(−gk
g′
gj + [bj,k]+gk − bj,khk
if j = k;
if j 6= k.
gk = hk − h′
k,
10
and
(2.13)
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
(yk + 1)hkF B;t0
ℓ;t
(y1, . . . , yn) = (y′
k + 1)h′
kF B1;t1
ℓ;t
(y′
1, . . . , y′
n) .
We conclude this section by recalling [12, Corollary 6.3] that explains why the g-vectors
and F -polynomials play a crucial role in the theory of cluster algebras. Recall that a cluster
algebra A is specified by a choice of a Y -seed (y, B) in a semifield P. Let F = QP(x1, . . . , xn)
be the field of rational functions in commuting independent variables x1, . . . , xn over the
quotient field QP of the integer group ring ZP of the multiplicative group P. Then each ℓ
and t as above gives rise to a cluster variable xℓ;t ∈ F given by
(2.14)
xℓ;t =
ℓ;t
F B;t0
F B;t0
ℓ;t
F (y1, . . . , yn)
P(y1, . . . , yn)
xg1
1 · · · xgn
n ,
where (g1, . . . , gn) = gB;t0
ℓ;t
(2.15)
, and the elements y1, . . . , yn ∈ F are given by
yj = yjYi
xbi,j
i
.
Furthermore, all cluster variables are of this form, and A is the ZP-subalgebra of F gener-
ated by all the xℓ;t.
3. F -polynomials of quiver representations
In this section we use the terminology on quiver representations from the introduction.
We work with a quiver Q = Q(B) (see (1.4)). Our goal is to develop some basic properties
of the F -polynomial FM (u1, . . . , un) associated to any representation M of Q in accordance
with (1.6).
Proposition 3.1. Each polynomial FM (u1, . . . , un) has constant term 1. Furthermore,
with coefficient 1, and it is divisible by
i=1 udim M (i)
i
FM (u1, . . . , un) contains the monomialQn
all the other occurring monomials.
Proof. It is enough to notice that, for e = (0, . . . , 0) or e = dM (see (1.5)), the quiver
Grassmannian Gre(M) consists of one point.
(cid:3)
Proposition 3.2. For all representations M ′ and M ′′ of Q, we have
(3.1)
FM ′⊕M ′′ = FM ′FM ′′.
Proof. We use the following well-known property of the Euler-Poincar´e characteristic: if a
complex torus T acts algebraically on a variety X, then χ(X) = χ(X T ), where X T is the set
of T -fixed points (see for example [2]). Take X = Gre(M ′ ⊕ M ′′), and consider the action
of T = C∗ on X induced by the T -action on M ′ ⊕ M ′′ given by
t · (m′, m′′) = (tm′, m′′)
(m′ ∈ M ′, m′′ ∈ M ′′).
Then a point N ∈ X is T -fixed if and only if the submodule N ⊆ M ′ ⊕ M ′′ splits into
N = N ′ ⊕ N ′′ for some N ′ ⊆ M ′ and N ′′ ⊆ M ′′. Thus, we have
X T = Ge′+e′′=e
(Gre′(M ′) × Gre′′(M ′′)),
QUIVERS WITH POTENTIALS II
and so
implying (3.1).
χ(Gre(M ′ ⊕ M ′′)) = Xe′+e′′=e
χ(Gre′(M ′))χ(Gre′′(M ′′)),
11
(cid:3)
To state our next result, we recall the maps αk and βk in (1.10) (the map γk is undefined
for arbitrary quiver representations). We will denote these maps αk;M and βk;M if necessary
to stress the dependency of a representation M. We denote by hM = (h1, . . . , hn) the integer
vector given by
(3.2)
hk = hk(M) = − dim ker βk;M .
Now assume that FM belongs to Q sf(u1, . . . , un) (the semifield of subtraction-free rational
expressions), hence can be evaluated in an arbitrary semifield 1(see the discussion after
Proposition 2.2). The definition (3.2) is then justified by the following analog of (2.9).
Proposition 3.3. Under the assumption that FM ∈ Q sf(u1, . . . , un), the components of the
vector hM appear as the exponents in the tropical evaluation
(3.3)
xh1
1 · · · xhn
n = FM Trop(x1,...,xn)(x−1
x[−bj,i]+
j
← ui).
i Yj6=i
Proof. First a general lemma following easily from the definition of a tropical semifield (see
(2.8)).
Lemma 3.4. If F (u1, . . . , un) is a Laurent polynomial belonging to Q sf(u1, . . . , un) then the
result of any evaluation of F in a tropical semifield does not change if we replace F with
the sum of the terms (taken with coefficient 1) corresponding to the vertices of its Newton
polytope.
Now suppose that N ∈ Gre(M), i.e., N is a subrepresentation of M with dN = e. Then
the exponent of xk in the tropical evaluation
(ue1
1 · · · uen
n )Trop(x1,...,xn)(x−1
x[−bj,i]+
j
← ui)
i Yj6=i
can be rewritten as
−ek +Xi6=k
[−bk,i]+ei = −ek +Xi6=k
(we used the fact that B is skew-symmetric). Note that
[bi,k]+ei = dim Nout(k) − dim N(k)
dim Nout(k) − dim N(k) ≥ dim β(N(k)) − dim N(k)
= − dim(N(k) ∩ ker βk)
≥ − dim ker βk = hk.
In view of Lemma 3.4, this implies that hk does not exceed the exponent of xk in the right
hand side of (3.3).
Now take e = −hkek (recall that ek stands for the k-th unit vector in Zn), and notice that
Gre(M) consists of one point N (with N(i) = {0} for i 6= k, and N(k) = ker βk), and that e
1It is conceivable that this condition holds for arbitrary quiver representations.
12
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
is obviously a vertex of the Newton polytope of FM . This implies that the exponent of xk in
the right hand side of (3.3) does not exceed hk, completing the proof of Proposition 3.3. (cid:3)
Proposition 3.5. Suppose that Q is a quiver with no oriented cycles, and M is a general
representation of dimension d = (d1, . . . , dn). Then every quiver Grassmannian Gre(M) is
smooth. In particular, this is the case if M is rigid, that is, Ext1(M, M) = 0.
This proposition follows from the results of Schofield ([18, §3]). For the convenience of
the reader we give an outline of the proof.
Proof. If M is a representation with dimension vector d, then we may identify M(i) with
Cdi by choosing a basis in M(i) for all i. Then M is represented as an element
(aM )a∈Q1 ∈ Repd(Q) := Ya∈Q1
K dha×dta.
The group GLd =Qn
i=1 GLdi(C) acts on Repd(Q) by base change. This way, isomorphism
classes of d-dimensional representations correspond to GLd-orbits in Repd(Q). For a di-
mension vector e = (e1, . . . , en), let
Ze,d ⊆ Repd(Q) ×
Grei(Cdi)
nYi=1
have natural projections p : Ze,d → Repd(Q) and q : Ze,d →Qn
be defined as the set of all (M, (N1, N2, . . . , Nn)) for which aM (Nta) ⊆ Nha for all a ∈ Q1. We
i=1 Grei(Cdi). One can show
that the projection q makes Ze,d into a vector bundle over the product of Grassmannians,
hence Ze,d is smooth. Now the quiver Grassmannian Gre(M) is equal to the fiber p−1(M).
If M is a general representation of dimension d, then the fiber p−1(M) is smooth by the
second Bertini Theorem ([20, Chapter II, §6.2, Theorem 2]).
(cid:3)
If Q is a quiver without oriented cycles, and M is indecomposable and rigid, then all the
quiver Grassmannians are smooth by the proposition above. It was shown in [7, 8]2, that
the F -polynomial of M has nonnegative coefficients in this case. The next two examples
show that in general, the coefficients can be negative, and the quiver Grassmannian may be
singular.
Example 3.6. Consider the quiver Q given by
a1,a2,a3,a4
1
/ 2
and let M be a general representation of Q of dimension d = (3, 4). The arrows a1, . . . , a4
act in M as four linear maps C3 → C4 in general position. Choose e = (1, 3). Since M
is in general position, Gre(M) is smooth by the discussion above. Now the first projection
Gre(M) → Gr1(C3) = P2 identifies Gre(M) with the projective curve C given by the
equation
det(a1(m), a2(m), a3(m), a4(m)) = 0
(m ∈ C3).
2It was pointed out in [17] that the proof in [8] contains a gap.
/
/
/
/
/
/
/
QUIVERS WITH POTENTIALS II
13
Since C is a smooth curve of degree 4, it has genus g = (4 − 1)(4 − 2)/2 = 3 and Euler
characteristic 2 − 2g = 2 − 2 · 3 = −4 (see [20, Chapter IV, 2.3]). So we have
Example 3.7. Consider the quiver Q given by
χ(Gre(M)) = −4.
c
@
3
1
b
a
=======
2
Let M1, M2, M3 be the indecomposable representations of Q of dimensions (0, 1, 1), (1, 0, 1),
and (1, 1, 0), respectively, and M = M1 ⊕ M2 ⊕ M3. It is immediate from the definition
(1.6) that
FM1(u1, u2, u3) = 1 + u3 + u2u3,
FM2(u1, u2, u3) = 1 + u1 + u1u3,
FM3(u1, u2, u3) = 1 + u2 + u1u2.
By Proposition 3.2, we have FM = FM1FM2FM3. In particular, the coefficient of u1u2u3 in
FM is 4. Thus, χ(Gr(1,1,1)(M)) = 4.
Geometrically this result can be seen as follows. The variety Gr(1,1,1)(M) is a subvariety
in Gr1(M(1)) × Gr1(M(2)) × Gr1(M(3)) = P1 × P1 × P1. Let P = (P (1), P (2), P (3)) ∈
P1 × P1 × P1 be given by
P (1) = ker aM = im cM , P (2) = ker bM = im aM , P (3) = ker cM = im bM ;
then Gr(1,1,1)(M) consists of all points N ∈ P1 × P1 × P1 such that N and P have at least
two common components. Thus, Gr(1,1,1)(M) is the union of three copies of P1 meeting at
a single point P . In other words, Gr(1,1,1)(M) is the disjoint union of three copies of A1 and
the single point {P }, so
χ(Gre(N)) = 3χ(A1) + χ({P }) = 3 · 1 + 1 = 4.
Note that Gr(1,1,1)(M) is singular at P .
4. Background on quivers with potentials and their representations
Let Q = (Q0, Q1, h, t) be a quiver (see Introduction). We denote by R the vertex span of Q,
that is, the commutative algebra over C with the basis {ei : i ∈ Q0} and the multiplication
given by eiej = δi,jei. The arrow span of Q is the finite-dimensional R-bimodule A with the
C-basis identified with Q1, and the R-bimodule structure given by
Ai,j = eiAej = Ma:j→i
∞Yd=0
RhhAii =
Ca.
A⊗Rd.
(4.1)
The complete path algebra of Q is defined as
@
o
o
14
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Thus, the elements of RhhAii are (possibly infinite) C-linear combinations of paths in Q;
note that by the convention (4.1) all the paths are traced in the right-to-left order. We view
RhhAii as a topological algebra with respect to the m-adic topology, where the (two-sided)
ideal m ⊂ RhhAii is given by
(4.2)
m =
A⊗Rd.
∞Yd=1
A potential on Q is an element S ∈ mcyc = Li∈Q0
mi,i, i.e., a possibly infinite linear
combination of cyclic paths in RhhAii. We view potentials up to cyclical equivalence defined
as follows: two potentials S and S′ are cyclically equivalent if S − S′ lies in the closure of
the span of all elements of the form a1 · · · ad − a2 · · · ada1, where a1 · · · ad is a cyclic path.
For any arrow a ∈ Q1, the cyclic derivative ∂a is the continuous linear map mcyc →
RhhAiit(a),h(a) acting on cyclic paths by
(4.3)
ap+1 · · · ada1 · · · ap−1.
∂a(a1 · · · ad) = Xp:ap=a
The Jacobian ideal J(S) of a potential S is the closure of the (two-sided) ideal in RhhAii
generated by the elements ∂a(S) for all a ∈ Q1. We call the quotient RhhAii/J(S) the
Jacobian algebra of S, and denote it by P(Q, S) or P(A, S).
The cyclic derivatives of a potential S can be expressed in terms of another important
family of elements ∂ba(S) ∈ RhhAii associated with pairs of arrows a, b ∈ Q1 such that
h(a) = t(b). Namely, the definition of a continuous linear map
∂ba : mcyc → RhhAiit(a),h(b).
is similar to (4.3): replacing if necessary a potential S with a cyclically equivalent one, we
can assume that no cyclic path occurring in S starts with an arrow a; for every such cyclic
path a1 · · · ad, we set
(4.4)
(4.5)
An easy check shows that, for any b ∈ Q1, we have
∂ba(a1 · · · ad) = Xν:aν−1=b,aν =a
a · ∂ba(S) = Xc:t(c)=h(b)
Xa:h(a)=t(b)
aν+1 · · · ada1 · · · aν−2.
∂cb(S) · c = ∂b(S).
A (decorated) representation of a quiver with potential (Q, S) (QP for short) is a pair
M = (M, V ), where M is a finite-dimensional P(Q, S)-module, and V is a finite-dimensional
R-module. A more concrete description was given in the introduction (see [9, Section 10]):
V is simply a collection (V (i))i∈Q0 of finite-dimensional vector spaces, while M = (M(i))i∈Q0
is a representation of Q annihilated by mN for N ≫ 0, and by all cyclic derivatives of S.
In view of (4.5), the latter relations are equivalent to (1.12), where the map γk in the
triangle (1.8) is defined as follows: for each a, b ∈ Q1 with h(a) = t(b) = k, the component
γa,b : M(h(b)) → M(t(a)) of γk is given by (1.11). We can also express γk in matrix form:
set
(4.6)
{a1, . . . , ar} = {a ∈ Q1 : h(a) = k},
{b1, . . . , bs} = {b ∈ Q1 : t(b) = k},
QUIVERS WITH POTENTIALS II
15
and let Hk(S) be the r × s matrix whose (p, q) entry is ∂bqapS; then the action of γk in M
is given by the matrix
(4.7)
γk = (Hk(S))M .
In what follows, we refer to a decorated representation M = (M, V ) of a QP (Q, S) as a
QP-representation. The direct sums and indecomposable QP-representations are defined in
a natural way. We say that M is positive if V = {0}, and negative if M = {0}. Thus, inde-
composable positive QP-representations are just indecomposable P(Q, S)-modules, while
indecomposable negative QP-representations are negative simple representations S −
for
k
k ∈ Q0 defined as follows:
(4.8)
S −
k (Q, S) = ({0}, V ), dim V (i) = δi,k.
As in [9, Definitions 4.2, 10.2], we view QP's and their representations up to right-
equivalence. Recall that QP's (Q, S) and (Q, S′) on the same underlying quiver Q are
right-equivalent if there is an automorphism ϕ of RhhAii (as an algebra and R-bimodule)
such that ϕ(S) is cyclically equivalent to S′. In view of [9, Proposition 3.7], we then have
ϕ(J(S)) = J(S′); therefore, every P(Q, S)-module M carries a structure of a P(Q, S′)-
module (which we denote ϕM) with the "twisted" action of RhhAii given by
ϕ(u) ⋆ m = um (u ∈ RhhAii, m ∈ M).
Now a QP-representation M′ = (M ′, V ′) of (Q, S′) is right-equivalent to a QP-representation
M = (M, V ) of (Q, S) if M ′ is isomorphic to ϕM as a P(Q, S′)-module, and V ′ is isomorphic
to V as an R-module.
Let ϕ be an automorphism of RhhAii as above. Fix a vertex k ∈ Q0, and use the notation
in (4.6). We would like to express the matrix Hk(ϕ(S)) in terms of Hk(S). As shown in the
proof of Lemma 5.3 in [9], we have
(4.9)
where:
(cid:0)ϕ(a1) ϕ(a2)
· · · ϕ(ar)(cid:1) =(cid:0)a1 a2
· · · ar(cid:1) (C0 + C1),
• C0 is an invertible r × r matrix with entries in C such that its (p, q)-entry is 0 unless
t(ap) = t(aq);
• C1 is a r × r matrix whose (p, q)-entry belongs to mt(ap),t(aq ).
Similarly, we have
(4.10)
where:
ϕ(b1)
ϕ(b2)
...
ϕ(bs)
= (D0 + D1)
b1
b2
...
bs
,
• D0 is an invertible s × s matrix with entries in C such that its (p, q)-entry is 0 unless
h(bp) = h(bq);
• D1 is a s × s matrix whose (p, q)-entry belongs to mh(bp),h(bq).
Note that both matrices C0 + C1 and D0 + D1 are invertible, and their inverses are of the
same form.
16
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
In the above notation, we claim that
(4.11)
all entries of the matrix
Hk(ϕ(S)) − (C0 + C1) ϕ(Hk(S)) (D0 + D1) belong to J(ϕ(S))
(here the matrix ϕ(Hk(S)) is obtained by applying ϕ to each entry of Hk(S)). As a con-
sequence, for the representation M ′ = ϕM as above, the corresponding map γ′
k is given
by
(4.12)
γ′
k = (C0 + C1)M ′ ◦ γk ◦ (D0 + D1)M ′,
where (C0 + C1)M ′ (resp. (D0 + D1)M ′) is an R-bimodule automorphism of Min(k) (resp. of
Mout(k)). Note that (4.12) is the equality (10.16) in [9], while (4.11) is implicit in the proof
of this equality.
We now recall one of the main technical results of [9], the Splitting Theorem ([9, Theo-
rem 4.6]). Let Q be a quiver without loops (but possibly having oriented 2-cycles). We say
that a QP (Q, S) is trivial if S is a linear combination of cyclic 2-paths, and J(S) = m; in
other words (see [9, Proposition 4.4]), the set of arrows Q1 consists of 2N distinct arrows
a1, b1, . . . , aN , bN such that each aνbν is a cyclic 2-path, and there is an R-bimodule auto-
morphism ϕ of the arrow span A such that ϕ(S) is cyclically equivalent to a1b1 + · · ·+ aN bN .
We say that a QP (Q, S) is reduced if S ∈ m3 (note that Q is still allowed to have oriented
2-cycles). Now the Splitting Theorem asserts that
(4.13)
any QP (Q, S) is right-equivalent to the direct sum of a reduced QP
(Q, S)red and a trivial QP (Q, S)triv, each of which is determined by
(Q, S) up to right-equivalence.
We refer to (Q, S)red as the reduced part of (Q, S). The operation of taking the re-
duced part naturally extends to representations. Namely, if M = (M, V ) is a representa-
tion of (Q, S), then Mred is obtained by transforming M into a representation (ϕM, V ) of
(Q, S)red ⊕ (Q, S)triv with the help of a right-equivalence in (4.13), and then restricting the
resulting representation to (Q, S)red (see [9, Definition 10.4] for more details). By [9, Propo-
sition 10.5], the reduction of representations is well-defined on the level of right-equivalence
classes.
Now everything is in place for introducing our main tool -- mutations of reduced QP's
and their representations. Let (Q, S) be a reduced QP, and k ∈ Q0 a vertex such that Q
has no oriented 2-cycles through k. Following [9], we define the mutation (Q, S) = µk(Q, S)
at k as the reduced part (eQ,eS)red, where the "premutation" (eQ,eS) =eµk(Q, S) is defined as
follows. First, the quiver eQ is obtained from Q by the following two-step procedure:
Step 1. For every pair of arrows a, b ∈ Q1 with h(a) = k = t(b), create a "composite" arrow
[ba] with h([ba]) = h(b) and t([ba]) = t(a).
Step 2. Reverse all arrows at k; that is, replace each arrow a with h(a) = k (resp. each
arrow b with t(b) = k) by an arrow a⋆ with t(a⋆) = k and h(a⋆) = t(a) (resp. b⋆ with
h(b⋆) = k and t(b⋆) = h(b)).
Second, the potential eS on eQ is obtained from S as follows: replacing S if necessary with a
cyclically equivalent potential, we can assume that no cyclic path occurring in S starts and
QUIVERS WITH POTENTIALS II
17
ends at k; then we set
(4.14)
where
(4.15)
eS = [S] + ∆,
Xa,b∈Q1: h(a)=t(b)=k
[ba]a⋆b⋆,
∆ =
and [S] is obtained by substituting [aνaν+1] for each factor aνaν+1 with t(aν) = h(aν+1) = k
of any cyclic path a1 · · · ad occurring in the expansion of S. As shown in [9, Theorem 5.2],
k(Q, S) is right-equivalent to (Q, S).
Now let M = (M, V ) be a QP-representation of a reduced QP (Q, S). Fix a vertex k
orem 5.7], the mutation µk acts as an involution on the set of right-equivalence classes of
reduced QPs, that is, µ2
the right-equivalence class ofeµk(Q, S) is determined by the right-equivalence class of (Q, S);
hence by (4.13), the same is true for µk(Q, S) = (eµk(Q, S))red. Furthermore, by [9, The-
and let (eQ,eS) = eµk(Q, S), and (Q, S) = µk(Q, S) = (eQ,eS)red. We define the mutated
fM =eµk(M) = (M , V ) of (eQ,eS) given by the following construction (see [9, Section 10]).
QP-representation M = µk(M) of (Q, S) as the reduced part of the QP-representation
First, we set
M (i) = M(i),
V (i) = V (i)
(i 6= k),
(4.16)
and define the spaces M(k) and V (k) by
(4.17)
M (k) =
ker γk
im βk
⊕ im γk ⊕
ker αk
im γk
⊕ V (k),
V (k) =
ker βk
ker βk ∩ im αk
(see (1.8)). For every arrow c of eQ, the corresponding linear map cM : M (t(c)) → M (h(c))
We set cM = cM for every arrow c not incident to k, and [ba]M = bM aM for all arrows a
is defined as follows.
and b in Q with h(a) = k = t(b). It remains to define the linear maps
αk : M in(k) = Mout(k) → M(k),
βk : M (k) → M out(k) = Min(k)
in the counterpart of the triangle (1.8) for the representation fM. We use the following
notational convention: whenever we have a pair U1 ⊆ U2 of vector spaces, denote by ι :
U1 → U2 the inclusion map, and by π : U2 → U2/U1 the natural projection. We now
introduce the following splitting data:
(4.18)
Choose a linear map ρ : Mout(k) → ker γk such that ρι = idker γk .
(4.19)
Choose a linear map σ : ker αk/im γk → ker αk such that πσ =
idker αk/imγk .
Then we define:
(4.20)
αk =
−πρ
−γk
0
0
,
βk =(cid:0)0 ι
ισ 0(cid:1) .
18
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
As shown in [9, Propositions 10.7, 10.9, 10.10], the above construction makes eµk(M) =
(M , V ) a QP-representation of (eQ,eS), whose isomorphism class does not depend on the
choice of the splitting data (4.18) -- (4.19), and whose right-equivalence class is determined
by the right-equivalence class of M. Furthermore, we have
(4.21)
and
(4.22)
γk = βkαk,
ker αk = im βk,
im αk =
ker γk
im βk
⊕ im γk ⊕ {0} ⊕ {0},
ker βk =
ker γk
im βk
⊕ {0} ⊕ {0} ⊕ V (k),
im βk = ker αk.
(see [9, (10.25), (10.26)]).
Since by the definition, the representation µk(M) of (Q, S) is the reduced part of fM =
eµk(M) = (M , V ), the right-equivalence class of µk(M) is determined by the right-equiva-
lence class of M. Furthermore, in view of [9, Theorem 10.13], the mutation µk of QP-
representations is an involution:
(4.23)
for every QP-representation M of a reduced QP (Q, S), the QP-repre-
sentation µ2
k(M) is right-equivalent to M.
Since by construction, the mutations send direct sums of QP-representations to the direct
sums, (4.23) implies that (cf. [9, Corollary 10.14])
(4.24)
any mutation µk sends indecomposable QP-representations of reduced
QPs to indecomposable ones.
Now suppose that the quiver Q has no oriented 2-cycles, i.e., it is of the form Q(B)
for some skew-symmetric integer matrix B (see (1.4)). Then the mutated QP µk(Q, S) =
(Q, S) is well-defined for any vertex k and any potential S on Q. However, the quiver
Q may acquire some oriented 2-cycle, say involving vertices i and j, which would make
mutations µi and µj undefined for the QP (Q, S). Following [9, Definition 7.2], we say that
a QP (Q, S) is nondegenerate if this does not happen, and moreover if any finite sequence
of mutations µkℓ · · · µk1 can be applied to (Q, S) without creating oriented 2-cycles along
the way. According to this definition, the class of nondegenerate QPs is stable under all
mutations. Furthermore, according to [9, Proposition 7.1], mutations of nondegenerate QPs
are compatible with matrix mutations:
if µk(Q(B), S) = (Q, S) then Q = Q(µk(B)) with
µk(B) given by (1.1).
Finally we note that every quiver Q(B) has a potential S such that (Q(B), S) is a nonde-
generate QP. More precisely, in view of [9, Corollary 7.4], the non-degeneracy of (Q(B), S)
is guaranteed by non-vanishing at S of countably many nonzero polynomial functions on
the space of potentials on Q(B) (taken up to cyclical equivalence).
5. QP-interpretation of g-vectors and F -polynomials
We retain all the notation and conventions of the preceding sections. To a QP-representa-
tion M = (M, V ) we associate the g-vector gM = (g1, . . . , gn) ∈ Zn given by (1.13), and the
F -polynomial FM = FM given by (1.6) (in particular, if M is negative then FM = 1). Note
that gM = gM′ and FM = FM′ if M and M′ are right-equivalent (for the F -polynomial,
QUIVERS WITH POTENTIALS II
19
this is immediate from (1.6); for the g-vector, this is a consequence of (4.12)). Note also
that
gM⊕M′ = gM + gM′
(5.1)
for any QP-representations M and M′ of the same QP.
Let B be a skew-symmetric integer n × n matrix, t0, t ∈ Tn, and ℓ ∈ {1, . . . , n}. Let
Q = Q(B) and let S be a potential on Q such that (Q, S) is a nondegenerate QP. The main
result of this section is a construction of a QP-representation M = MB;t0
of (Q, S) such
ℓ;t
that gM = gB;t0
ℓ;t were
ℓ;t
introduced in Section 2.
, where the g-vectors gB;t0
ℓ;t
and F -polynomials F B;t0
and FM = F B;t0
ℓ;t
The family of QP-representations MB;t0
ℓ;t
is uniquely determined by the properties (1.15)
and (1.16). More explicitly, let
t0
k1
t1
k2
· · ·
kp
tp = t
be the (unique) path joining t0 and t in Tn. We set
(Q(t), S(t)) = µkp · · · µk1(Q, S),
which is well-defined because (Q, S) is nondegenerate. Let S −
simple representation of (Q(t), S(t)) at a vertex ℓ (see (4.8)). Then we have
ℓ (Q(t), S(t)) be the negative
MB;t0
(5.2)
in view of (4.23), replacing MB;t0
ℓ;t
assume that it is a QP-representation of (Q, S).
ℓ;t = µk1 · · · µkp(S −
ℓ (Q(t), S(t)));
if necessary by a right-equivalent representation, we can
Theorem 5.1. We have
(5.3)
where M = MB;t0
ℓ;t
.
gB;t0
ℓ;t = gM, F B;t0
ℓ;t = FM,
Proof. We deduce Theorem 5.1 from the following key lemma.
Lemma 5.2. Let M = (M, V ) be an arbitrary QP-representation of a nondegenerate QP
(Q(B), S), let M = (M , V ) = µk(M) for some k ∈ Q(B)0, and suppose that the Y -seed
(y′, B1) in Q sf(y1, . . . , yn) is obtained from (y, B) by the mutation at k. Let hk (resp. h′
k)
be the k-th component of the vector hM (resp. hM ) given by (3.2). Then the g-vector
gM = (g1, . . . , gn) satisfies (2.12), and is related to the g-vector gM = (g′
n) via
(2.11). Furthermore, the F -polynomials FM and FM are related by
(5.4)
(yk + 1)hkFM(y1, . . . , yn) = (y′
1, . . . , g′
1, . . . , y′
n) .
k + 1)h′
kFM(y′
Before proving Lemma 5.2, we first show how it implies Theorem 5.1. Let M = (M, V ) =
MB;t0
ℓ;t
(5.5)
. We prove (5.3) together with the equality
hM = hB;t0
ℓ;t
(see (2.9)) by induction on the distance between t0 and t in the tree Tn. The basis of
induction is the case t = t0. By (1.15), we have MB;t0
ℓ (Q(B), S). The fact that the g-
vector and F -polynomial of this QP-representation agree with (2.1) and (2.4), is immediate
from the definitions, while both sides of (5.5) are equal to 0.
ℓ;t0 = S −
20
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Now assume that (5.3) and (5.5) are satisfied for some ℓ and t, and that t0
t1 in Tn.
In view of (1.16), the QP-representation M = (M , V ) in Lemma 5.2 is equal to MB ′;t1
,
where B′ = µk(B). To finish the proof, it suffices to show that
ℓ;t
k
ℓ;t
ℓ;t
ℓ;t
(1) gM = gB1;t1
;
(2) FM = F B1;t1
;
(3) hM = hB1;t1
.
To prove (1), it suffices to observe that, by Lemma 5.2, the vector gM is obtained from
gM by the same rule (2.11) that expresses gB1;t1
. Then, since by Lemma 5.2
the numbers h′
k is the k-th component
of the vector hB1;t1
. Next, using the latter claim, and comparing (5.4) with the relation
(2.13) in Proposition 2.4, we obtain the proof of (2). Finally, to prove (3) it is enough to
apply Proposition 3.3 to the representation M (note that in view of (2), the polynomial FM
is a subtraction-free rational expression, which makes Proposition 3.3 applicable).
k, hk and gk are related by (2.12), we conclude that h′
ℓ;t
in terms of gB;t0
ℓ;t
ℓ;t
It remains to prove Lemma 5.2, which we accomplish in several steps.
Step 1. We start by proving that the numbers h′
(2.12), which we rewrite as
k, hk and gk in Lemma 5.2 are related by
−h′
k = gk − hk.
Remembering (3.2) and (1.13), we can rewrite this equality as
dim ker βk = dim ker γk − dim M(k) + dim V (k) + dim ker βk = dim(cid:16) ker γk
Qn
which is immediate from (4.22).
Step 2. Our next target is the identity (5.4). Suppose that N = (N(1), . . . , N(n)) ∈
i=1 Grei(M(i)), and let Nin(k) and Nout(k) be the corresponding subspaces of Min(k) and
Mout(k), respectively. The condition that N ∈ Gre(M) can be stated as the combination of
the following two conditions:
⊕ V (k)(cid:17),
im βk
(5.6)
cM (N(j)) ⊆ N(i) for any arrow c : j → i not incident to k in Q(B).
αk(Nin(k)) ⊆ N(k) ⊆ β−1
(5.7)
Now let e′ = (ei)i6=k denote the integer vector obtained from e by forgetting the com-
ponent ek. For every such vector e′ and every pair of nonnegative integers r ≤ s, we
denote by Ze′;r,s(M) the variety of tuples (N(i))i6=k satisfying the inclusions (5.6) and
αk(Nin(k)) ⊆ β−1
k (Nout(k)), and such that dim N(i) = ei for i 6= k, and
k (Nout(k)).
dim αk(Nin(k)) = r,
dim β−1
k (Nout(k)) = s.
Let Ze;r,s(M) denote the subset of Gre(M) consisting of all N = (N(1), . . . , N(n)) such that
the tuple obtained from N by forgetting N(k) belongs to Ze′;r,s(M). Then Gre(M) is the
disjoint union of the subsets Ze;r,s(M) over all pairs (r, s); and in view of (5.7), each Ze;r,s(M)
is the fiber bundle over Ze′;r,s(M) with the fiber Grek−r(Cs−r). Since χ(Grek−r(Cs−r)) =
(cid:0) s−r
ek−r(cid:1), it follows that
χ(Gre(M)) =Xr,s (cid:18) s − r
ek − r(cid:19)χ(Ze′;r,s(M)).
QUIVERS WITH POTENTIALS II
21
Substituting this expression into (1.6) and performing the summation with respect to ek,
we obtain
(5.8)
χ(Ze′;r,s(M))yr
k(yk + 1)s−r
FM(y1, . . . , yn) = Xe′,r,s
yei
i .
nYi6=k
The proof of (5.4) is based on the following observation:
(5.9)
Ze′;r,s(M) = Ze′;r,s(M ),
where r and s are given by
(5.10)
[bi,k]+ei − hk − s,
[−bi,k]+ei − h′
k − r.
r =Xi
s =Xi
In view of the symmetry between M and M, to prove (5.9), it is enough to show that every
(N(i))i6=k ∈ Ze′;r,s(M) belongs to Ze′;r,s(M ).
First of all, we need to show that βkαk(Nout(k)) ⊆ Nin(k), that is, the counterpart for
M of the inclusion αk(Nin(k)) ⊆ β−1
k (Nout(k)). As an immediate consequence of (4.20),
we get βkαk = −γk. In view of (1.11), each of the components of the map γk is a linear
combination of compositions of maps of the kind cM or bM aM (where a, b, c ∈ Q1 are such
that h(a) = t(b) = k, and c is not incident to k); thus, the defining conditions (5.6) and
(5.7) imply the desired inclusion γk(Nout(k)) ⊆ Nin(k).
To conclude the proof of (5.9), it remains to show that
(5.11)
dim αk(Nout(k)) = r,
dim β
−1
k (Nin(k)) = s.
To show the first equality, recall from (4.22) that ker αk = im βk, implying that
dim αk(Nout(k)) = dim Nout(k)/(Nout(k) ∩ im βk) =Xi
Using the exact sequence
[bi,k]+ei − dim(Nout(k) ∩ im βk).
0 → ker βk → β−1
k (Nout(k)) → Nout(k) ∩ im βk → 0,
we conclude that
dim(Nout(k) ∩ im βk) = dim β−1
k (Nout(k)) − dim ker βk = s + hk,
implying the first equality in (5.11). The second equality can be shown by similar arguments
but also follows from the first one applied to M instead of M.
The rest of the proof of (5.4) is straightforward: use (5.8) and (5.9) for rewriting its
right-hand side in the form
(y′
k + 1)h′
kFM(y′
1, . . . , y′
n) = (y′
k + 1)h′
χ(Ze′;r,s(M))(y′
k)r(y′
kXe′,r,s
k + 1)s−rYi6=k
(y′
i)ei,
1, . . . , y′
then substitute for y′
n (resp. for r and s) the expressions given by (2.10) (resp. by
(5.10)), simplify the resulting expression, and use (5.8) again to see that it is equal to the
left-hand-side of (5.4).
Step 3. To finish the proof of Lemma 5.2, it remains to show that the vectors gM and
gM′ are related by (2.11). As shown in Step 1, we have gk = hk − h′
k, implying the equality
g′
k = −gk.
22
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Now let i 6= k. Using (1.13), (3.2), and the fact that the matrix B is skew-symmetric, we
can rewrite the desired second equality in (2.11) as
dim ker γi − [bk,i]+ dim ker βk = dim ker γi − [bi,k]+ dim ker βk.
Interchanging M and M if necessary, we see that it suffices to prove the following:
(5.12)
if bk,i ≥ 0 then dim ker γi = dim ker γi − bk,i dim ker βk.
We first show that (5.12) holds if we replace the map γi : M out(i) → M in(i) with its
space Mout(i) as
counterparteγi : fMout(i) → fMin(i) for the representation fM =eµk(M). We decompose the
Mout(i) = M(k)bk,i ⊕ M ′
out(i),
where the first summand corresponds to the bk,i arrows from i to k. Accordingly, we have
Tracing the definitions, we see that the maps γi : M(k)bk,i ⊕ M ′
Mout(k)bk,i ⊕ M ′
out(i) → M (k)bk,i ⊕ Min(i) can be written in the block-matrix form as
out(i) → Min(i) and eγi :
fMout(i) = Mout(k)bk,i ⊕ M ′
γi =(cid:16)ψ ◦ βbk,i
k
out(i), fMin(i) = M (k)bk,i ⊕ Min(i).
η(cid:17) ,
eγi =(cid:18)αbk,i
η(cid:19)
k
ψ
0
k
for some linear maps ψ and η, where βbk,i
stand for the direct (diagonal) sums of
bk,i copies of the maps βk : M(k) → Mout(k) and αk : Mout(k) → M (k). Using the equality
ker αk = im βk ((4.22)), it is easy to see that there is an exact sequence
and αbk,i
k
We conclude that
where the map ker γi → kereγi sends a pair (u, v) ∈ ker γi ⊆ M(k)bk,i ⊕ M ′
0 → (ker βk)bk,i ⊕ {0} → ker γi → kereγi → 0,
dim kereγi = dim ker γi − bk,i dim ker βk.
To complete the proof of (5.12), it remains to show that
(5.13)
out(i) to (βbk,i
k u, v).
dim kereγi = dim ker γi.
In view of (4.12), dim kereγi does not change if we replace (eQ,eS) with a right-equivalent QP.
Thus, in proving (5.13), we can assume that (eQ,eS) = (Q, S) ⊕ (Q′, S′), where (Q′, S′) is a
trivial QP. In accordance with this decomposition, we can decompose the spaces fMin(i) and
fMout(i) as
where the spaces fM ′
out(i) correspond to the arrows from Q′. Thus, eγi has the
where ι is a vector space isomorphism fM ′
in(i), fMout(i) = M out(i) ⊕fM ′
eγi =(cid:18)γi 0
ι(cid:19) ,
out(i) → fM ′
fMin(i) = M in(i) ⊕fM ′
in(i) and fM ′
in(i). This implies (5.13), which com-
(cid:3)
Theorem 5.1 yields a formula for cluster variables in the coefficient-free cluster algebra
pletes the proofs of Lemma 5.2 and Theorem 5.1.
following block-matrix form:
out(i),
0
(that is, the one with the coefficient semifield P = {1}).
QUIVERS WITH POTENTIALS II
23
Corollary 5.3. Suppose that F B;t0
is positive
(that is, M = (M, 0)). Let xℓ;t be the corresponding cluster variable in the coefficient-free
cluster algebra . Then xℓ;t is given by the formula
6= 1, hence the QP-representation M = MB;t0
ℓ;t
ℓ;t
(5.14)
xℓ;t =
nYi=1
x−di
i Xe
χ(Gre(M))
nYi=1
where di = dim M(i).
Proof. It suffices to rewrite (1.13) as
−rkγi+Pj ([bi,j]+ej+[−bi,j]+(dj −ej))
x
i
,
gi = dim Mout(i) − rkγi − dim M(i) =Xj
[−bi,j]+dj − rkγi − di ,
and apply (2.14) and (2.15).
(cid:3)
Remark 5.4. If the quiver Q(B) has no oriented cycles then S = 0, hence γi = 0 for all i.
In this case (5.14) specializes to the Caldero-Chapoton formula for cluster variables (see [6])
obtained in this generality in [7, Theorem 3].
Recall that the denominator vector of a cluster variable z with respect to the initial
cluster (x1, . . . , xn) is the integer vector (d1(z), . . . , dn(z)) such that
z =
P (x1, . . . , xn)
xd1(z)
· · · xdn(z)
1
n
,
where P is a polynomial not divisible by any xi. Conjecture 7.17 in [12] claims that if z does
not belong to the initial cluster then the denominator vector of z is equal to the multidegree
of the corresponding F -polynomial. By Proposition 3.1 and Theorem 5.1, this conjecture is
equivalent to the equality
(5.15)
di(xℓ;t) = di = dim M(i)
(in the notation of Corollary 5.3). It was shown in [7] in the case where Q(B) has no oriented
cycles, that (5.14) implies (5.15). A direct proof of this was given in [15, Theorem 10]. In
full generality, (5.15) was disproved by a counterexample in [13] (based on the ideas in [4]).
Using Theorem 5.1, we obtain the following partial result.
Corollary 5.5. In the notation of Corollary 5.3, we have the inequality
(5.16)
di(xℓ;t) ≤ di .
Furthermore, a necessary condition for the equality in (5.16) is the existence of a quiver
subrepresentation N of M such that
(5.17)
ker γi ⊆ Nout(i),
γi(Nout(i)) = Nin(i) .
Proof. In view of (5.14), we have
(5.18)
di − di(xℓ;t) = min
e
(−rkγi +Xj
([bi,j]+ej + [−bi,j]+(dj − ej))) ,
24
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
where the minimum is over all dimension vectors e such that χ(Gre(M)) 6= 0. In particular,
Gre(M) must be nonempty, i.e., M must have a subrepresentation N with ei = dim N(i)
for all i. In terms of N, we have
Xj
[bi,j]+ej = dim Nin(i), Xj
Therefore,
[−bi,j]+(dj − ej) = dim Mout(i) − dim Nout(i) .
− rkγi +Xj
([bi,j]+ej + [−bi,j]+(dj − ej))
= −rkγi + dim Mout(i) + dim Nin(i) − dim Nout(i)
= dim ker γi + dim Nin(i) − dim Nout(i)
= dim
ker γi
ker γi ∩ Nout(i)
+ dim
Nin(i)
γi(Nout(i))
,
making clear both assertions in question.
(cid:3)
Remark 5.6. A counterpart of Corollary 5.5 in the context of 2-Calabi-Yau categories was
obtained in [13, Proposition 5.8].
We conclude this section by applying the above results for an explicit construction of a
special class of QP-representations corresponding to cluster variables. Let T be a subset
of vertices of Q = Q(B) such that the induced subgraph on T is a tree; in particular,
bi,j ∈ {0, ±1} for i, j ∈ T , so inside T there are no multiple arrows. Without loss of
generality, we can assume that T = [1, ℓ] ⊆ [1, n] = Q0, and that each i ∈ T is a leaf of
the subtree of T on vertices [i, ℓ]; in other words, for each i ∈ [1, ℓ − 1] there is a a unique
j ∈ [i + 1, ℓ] connected by an edge with i. Let M = MT be a Q-representation such that
M(i) = C for i ∈ T , M(i) = 0 for i /∈ T , and aM : M(t(a)) → M(h(a)) is an isomorphism
whenever h(a) and t(a) belong to T . The condition that T is a tree implies that M is a
P(Q, S)-module for any potential S (since every cyclic derivative ∂aS is a linear combination
of paths from h(a) to t(a), and every such path acts as 0 in M).
Proposition 5.7. Let t0
1
t1
2
· · ·
ℓ
tℓ = t be a path in Tn. Then
(5.19)
MB;t0
ℓ;t = (MT , 0) .
Proof. We need to show that the sequence of mutations µℓ ◦ · · · ◦ µ1 takes the QP-repre-
sentation (MT , 0) to the negative simple representation S −
(see (1.15)). For ℓ = 1, the
ℓ
representation MT is just the (positive) simple module Sℓ; using (4.17), we see that the
mutation µℓ turns it into S −
ℓ . For ℓ > 1, again using (4.17), we see that the mutation µ1
turns MT into MT ′, where the tree T ′ is obtained from T by removing the leaf 1. The proof
is finished by induction on ℓ.
(cid:3)
Corollary 5.8. In the situation of Proposition 5.7, the F -polynomial F B;t0
follows:
ℓ;t
is given as
(5.20)
F B;t0
ℓ;t
(u1, . . . , un) =XZ Yi∈Z
ui ,
QUIVERS WITH POTENTIALS II
25
where Z runs over all subsets of T = [1, ℓ] with the property that if j ∈ Z then i ∈ Z
for every arrow j → i in T . Furthermore, the denominator vector of the cluster variable
xℓ;t is the indicator vector of [1, ℓ] (that is, di(xℓ;t) = 1 for i ∈ [1, ℓ], and di(xℓ;t) = 0 for
i ∈ [ℓ + 1, n]).
Proof. By Theorem 5.1 and Proposition 5.7, we have F B;t0
ℓ;t = FMT . The equality (5.20) is
then immediate from the definition (1.6): clearly, the quiver Grassmannian Gre(MT ) consists
of one point if e is the indicator vector of a subset Z as in (5.20), otherwise Gre(MT ) = ∅.
Turning to the denominator vector, in view of Corollary 5.5, it is enough to show that
di(xℓ;t) = 1 for i ∈ T . Fix a vertex i ∈ T , and let Z be the subset of all vertices j ∈ T
that can be reached from i by a directed path in T . Let N = ⊕j∈ZMT (j). Then N is a
quiver subrepresentation of MT . The fact that T is a tree implies easily that N satisfies
(5.17) (indeed, we have γi = 0, Nout(i) = Mout(i), and Nin(i) = 0). Furthermore, N is the
only element in its quiver Grassmannian, which makes (5.17) not only necessary but also a
sufficient condition for the equality di(xℓ;t) = di(M) = 1.
(cid:3)
Remark 5.9. The computation of the g-vector of MT is more involved, since the map γi
is not necessarily 0 if i /∈ T . However, γi = 0 for i ∈ T , hence for i ∈ T the component gi
of gB;t0
ℓ;t
is equal to {j ∈ T : i → j} − 1.
6. Mutations preserve homomorphisms modulo confined ones
Let M = (M, V ) and N = (N, W ) be QP-representations of a reduced QP (Q, S). We fix
a vertex k ∈ Q0 and assume that Q has no oriented 2-cycles through k. Thus, the mutated
QP (Q, S) = µk(Q, S) is well-defined, as well as its QP-representations M = (M, V ) =
µk(M) and N = (N, W ) = µk(N ).
We abbreviate
M(bk) =Mi6=k
M(i),
and say that a homomorphism ϕ ∈ HomQ(M, N) is confined to k if ϕ(m) = 0 for m ∈ M(bk).
Denote the space of such homomorphisms by Hom[k]
a vector space isomorphism
Q (M, N). Restricting ϕ to M(k) yields
(6.1)
Hom[k]
Q (M, N) = HomC(coker αk;M , ker βk;N ).
The goal of this section is to prove the following proposition. It was already established
in [3, Theorem 7.1] but the present proof seems to be much simpler.
Proposition 6.1. The mutation µk induces an isomorphism
HomQ(M, N)/ Hom[k]
Q (M, N) = HomQ(M , N)/ Hom[k]
Q
(M , N).
Proof. We can view a P(Q, S)-module M as a module over the subalgebra
P(Q, S)bk,bk = Mi,j6=k
P(Q, S)i,j.
26
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Clearly, M(bk) is a P(Q, S)bk,bk-submodule of M, and so we have the restriction map ρ :
HomQ(M, N) → HomP(Q,S)k,k
HomQ(bk)(M, N) ={ϕ ∈ HomP(Q,S)k,k
(M(bk), N(bk)). Denote
ϕ(ker αk;M ) ⊆ ker αk;N , ϕ(im βk;M ) ⊆ im βk;N}.
(M(bk), N(bk)) :
As an easy consequence of the definitions, we have
ker ρ = Hom[k]
Q (M, N),
im ρ = HomQ(bk)(M, N).
Thus, ρ induces an isomorphism
(6.2)
HomQ(M, N)/ Hom[k]
Q (M, N) = HomQ(bk)(M, N).
Now recall from [9, Proposition 6.1, Corollary 6.6] that the mutation µk induces an
isomorphism between P(Q, S)bk,bk and P(Q, S)bk,bk. This isomorphism is explicitly described
in the proof of Proposition 6.1 in [9]:
it preserves all arrows not incident to k, and sends
each product ba (for a an incoming, and b an outgoing arrow at k) to the "composite
arrow" [ba]. Identifying P(Q, S)bk,bk and P(Q, S)bk,bk with the help of this isomorphism, and
recalling the definition of M in Section 4, we see that the P(Q, S)bk,bk-module structure on
M (bk) becomes identical to the P(Q, S)bk,bk-module structure on M(bk). Furthermore, by
(M(bk), N(bk)) gets identified with HomQ(bk)(M, N ). This completes the proof of
(4.22) we have ker αk = im βk,
HomP(Q,S)k,k
Proposition 6.1.
im βk = ker αk. Therefore, the subspace HomQ(bk)(M, N) ⊆
The isomorphism in Proposition 6.1 can be viewed as functorial in the following way.
Let C(Q, S) be the category whose objects are QP-representations of a QP (Q, S), and the
morphisms are given by
(cid:3)
HomC(Q,S)((M, V ), (N, W )) = HomQ(M, N) ⊕ HomR(V, W ).
For a vertex k ∈ Q0, let C[bk](Q, S) be the quotient category of C(Q, S) with the same objects,
and the morphisms given by
HomC[k](Q,S)((M, V ), (N, W )) =
HomC(Q,S)((M, V ), (N, W ))
Hom[k]
Q (M, N) ⊕ HomR(V, W )
.
Proposition 6.2. The mutation µk induces an equivalence of categories
µk : C[bk](Q, S) → C[bk](Q, S).
Proof. In view of (6.2), we have
HomC[k](Q,S)((M, V ), (N, W )) = HomQ(bk)(M, N).
It follows from the proof of Proposition 6.1 that the mutation at k gives rise to a functor
from C[bk](Q, S) to C[bk](Q, S). The fact that this functor is an equivalence of categories is a
consequence of the following basic result in the category theory (see [19, Proposition 16.3.2]).
Proposition 6.3. Let C and C be categories, and suppose F : C → C is a functor with the
following properties:
QUIVERS WITH POTENTIALS II
27
(1) For every object M of C there is an object M of C such that F (M) is isomorphic
to M;
(2) For any pair of objects M, N of C, the functor F induces a bijection
HomC(M, N ) ∼= HomC(FM, F N ).
Then F is an equivalence of categories, i.e., there exists a functor G : C → C such that the
the composition functors G ◦ F and F ◦ G are naturally equivalent to the identity functors
of C and C, respectively.
(cid:3)
Remark 6.4. The proof of Proposition 6.3 is based on a strong version of the axiom of
choice (see [19, §3.1, Remark 16.3.3]): for any class of sets and any equivalence relation on
this class we can choose a representative in every class.
Let M = (M, V ) and N = (N, W ) be QP-representations of the same nondegenerate QP
7. The E-invariant
(Q, S). We abbreviate
(7.1)
and
hM, Ni = dim HomQ(M, N),
(7.2)
d−
i (M) = dim V (i),
so that the components of the g-vector gM = (g1, . . . , gn) are given by
di(M) = di(M) = dim M(i),
(7.3)
gi = gi(M) = dim ker γi;M − di(M) + d−
i (M).
We now define the integer function
(7.4)
Einj(M, N ) = hM, Ni +
and its symmetrized version
di(M)gi(N ),
nXi=1
(7.5)
Esym(M, N ) = Einj(M, N ) + Einj(N , M).
In view of (1.17), the E-invariant of a QP-representation is given by
(7.6)
E(M) = Einj(M, M) =
Esym(M, M)
2
.
Now let µk(M) = M = (M , V ) and µk(N ) = N = (N, W ) be QP-representations (of
the QP (Q, S) = µk(Q, S)) obtained from M and N by the mutation at a vertex k.
Theorem 7.1. We have
(7.7)
In particular, Esym(M, N ) and E(M) are invariant under QP-mutations, i.e.,
Einj(M, N ) − Einj(M, N ) = hk(M )hk(N) − hk(M)hk(N ).
Esym(µk(M), µk(N )) = Esym(M, N ), E(µk(M)) = E(M)
for any vertex k.
28
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Proof. Our starting point is the equality
(7.8)
hM, Ni + dim(coker αk;M ) · hk(N) = hM, Ni + dim(coker αk;M ) · hK(N) ,
obtained by combining Proposition 6.1 with (6.1) (and recalling the notation (3.2)). We
claim that dim(coker αk;M ) and dim(coker αk;M ) are given by
(7.9)
and
(7.10)
dim(coker αk;M ) = hk(M ) + dk(M) + dk(M) −Xi
dim(coker αk;M ) = hk(M) + dk(M) + dk(M ) −Xi
[−bi,k]+di(M) ,
[bi,k]+di(M) .
Note that (7.10) follows from (7.9) by interchanging M with M , so it is enough to prove
(7.9). Using the equality ker αk;M = im βk;M in (4.22), we obtain
dim(coker αk;M ) = dk(M) − rk αk;M
= dk(M) − dim Min(k) + dim(ker αk;M )
= dk(M) −Xi
= dk(M) −Xi
[−bi,k]+di(M) + rk βk;M
[−bi,k]+di(M) + dk(M) − dim(ker βk;M ) ,
which implies (7.9) in view of (3.2).
Using (7.9) and (7.10), we can rewrite (7.8) as follows:
hM, Ni − hM, Ni = dim(coker αk;M ) · hk(N) − dim(coker αk;M ) · hK(N )
= (hk(M ) + dk(M) + dk(M ) −Xi
− (hk(M) + dk(M) + dk(M ) −Xi
[−bi,k]+di(M)) · hK(N)
[bi,k]+di(M)) · hK(N) .
In view of Lemma 5.2, we have hk(N) = hk(N) − gk(N ), which allows us to rewrite (7.8)
further as
(7.11)
hM, N i − hM, Ni − (hk(M)hk(N) − hk(M)hk(N))
= (Xi
bi,kdi(M)) · hk(N) + (dk(M) + dk(M ) −Xi
[bi,k]+di(M)) · gk(N ) .
Comparing (7.11) with the desired equality (7.7), we see that it remains to show that the
right hand side of (7.11) is equal to
(di(M)gi(N ) − di(M )gi(N )).
Xi
QUIVERS WITH POTENTIALS II
29
Using the equality di(M ) = di(M) for i 6= k, and the assertion (proved in Lemma 5.2) that
the transformation gN 7→ gN is given by (2.11), we obtain
Xi
(di(M)gi(N ) − di(M )gi(N ))
= (dk(M) + dk(M )) · gk(N ) +Xi6=k
= (Xi
bi,kdi(M)) · hk(N) + (dk(M) + dk(M) −Xi
di(M)(bi,khk(N) − [bi,k]+gk(N ))
[bi,k]+di(M)) · gk(N ) ,
finishing the proof of Theorem 7.1.
(cid:3)
Corollary 7.2. If M is obtained by a sequence of mutations from a negative QP-repre-
sentation ({0}, V ), then E(M) = 0. In particular, this is the case for any representation
MB;t0
ℓ;t
given by (5.2).
Proof. By the definition (1.17), we have E(({0}, V )) = 0, hence E(M) = 0 as well.
(cid:3)
We conclude this section by one more invariance property of E(M). For a quiver
Q = (Q0, Q1, h, t), we denote by Qop the opposite quiver (Q0, Q1, t, h) obtained from Q
by reversing all arrows. To distinguish the arrows of Qop from those of Q, we denote by aop
the arrow of Qop corresponding to an arrow a of Q. The correspondence a 7→ aop extends to
an anti-isomorphism u 7→ uop of completed path algebras RhhAii → RhhAopii (identical on
the vertex span R). In particular, every QP (Q, S) gives rise to the opposite QP (Qop, Sop).
By the definition (4.3) of a cyclic derivative, we have ∂aopSop = (∂aS)op for any arrow a
of Q. Thus, J(Sop) = (J(S))op. This implies that every QP-representation M = (M, V ) of
(Q, S) gives rise to a QP-representation M⋆ = (M ⋆, V ) of (Qop, Sop) obtained from M by
replacing each space M(k) with its dual M(k)⋆, and setting (aop)M ⋆ = (aM )⋆ for any arrow
a of Q.
Proposition 7.3. We have E(M⋆) = E(M) for any QP-representation M.
Proof. Using the notation in (7.2) and (7.3), we can express gi(M) as
(7.12)
gi(M) = dim Mout(i) − rk γi;M − di(M) + d−
i (M),
and E(M) as
(7.13)
E(M) = hM, Mi +
nXi=1
+ Xa∈Q1
dh(a)(M)dt(a)(M) .
di(M)(d−
i (M) − rk γi;M − di(M))
It remains to observe that passing from M to M⋆ does not change any of the terms in
(7.13) (since HomQop(M ⋆, M ⋆) is isomorphic to HomQ(M, M), and γi;M ⋆ = (γi;M )⋆).
(cid:3)
30
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
8. Lower bounds for the E-invariant
Fix a QP-representation M = (M, V ) of a reduced QP (Q, S). The goal of this section
is to prove the lower bound (1.18) for E(M). Using the notation in (7.2), we can state the
result as follows (since M is fixed, we allow ourselves to skip references to it in most of the
formulas below).
Theorem 8.1. The E-invariant of a QP-representation satisfies
(dim(ker βi) · dim(ker γi/ im βi) + di(M) · d−
i (M)).
(8.1)
(8.2)
E(M) ≥Xi∈Q0
E(M) ≥Xi∈Q0
Proof. The desired lower bound for E(M) follows from another one:
(dim(coker αi) · dim(ker αi/ im γi) + di(M) · d−
i (M));
indeed, to deduce (8.1) from (8.2) it suffices to apply the latter bound to the dual QP-
representation M⋆ and use Proposition 7.3.
Substituting into (8.2) the expression (7.13) for E(M), regrouping the terms and simpli-
fying, we can rewrite it as follows:
hM, Mi + Xa∈Q1
−Xi∈Q0
(8.3)
dh(a)(M) · dt(a)(M)
(di(M)2 + dim(coker αi) · dim(ker αi)) ≥Xi∈Q0
rkγi · rkαi.
We abbreviate
U =Mi∈Q0
HomC(Min(i), M(i)),
and define the subspaces U1 ⊆ U2 in U by
(8.4)
U1 = {(ψi : Min(i) → M(i))i∈Q0 :
U2 = {(ψi : Min(i) → M(i))i∈Q0 : ψi(ker αi) ⊆ im αi for all i}.
im ψi ⊆ im αi for all i},
Now we can state the key lemma.
Lemma 8.2. There exist two linear maps
HomR(M, M)
Φ
/ U
Ψ
satisfying the following conditions:
HomC(M(h(b)), M(t(b)))
/Lb∈Q1
(1) ker Φ = HomQ(M, M);
(2) im Φ ⊆ U2;
(3) im Φ ⊆ ker Ψ;
(4) dim Ψ(U1) ≥Pi∈Q0
rk γi · rk αi.
/
/
Before proving Lemma 8.2, we show that it implies (8.3). By the definition of U2, we have
QUIVERS WITH POTENTIALS II
31
dim U2 = dim U −Xi∈Q0
dim(coker αi) · dim(ker αi)
= Xa∈Q1
dh(a)(M) · dt(a)(M) −Xi∈Q0
HomR(M, M) =Mi∈Q0
EndC(M(i)).
Note also that
By (1), we have
dim(coker αi) · dim(ker αi).
rk Φ = dim(HomR(M, M)) − dim(ker Φ) =Xi∈Q0
di(M)2 − hM, Mi.
In view of (2), the left hand side of (8.3) is equal to dim(U2/im Φ). Now we use (3) and (4)
to conclude that
dim(U2/im Φ) ≥ dim(U2/(U2 ∩ ker Ψ)) = dim(Ψ(U2))
≥ dim(Ψ(U1)) ≥Xi∈Q0
rk γi · rk αi,
finishing the proof of (8.3).
To complete the proof of Theorem 8.1 it remains to prove Lemma 8.2. We define the map
Φ by setting, for ξ ∈ HomR(M, M),
(8.5)
Φ(ξ) = (ηi : Min(i) → M(i))i∈Q0 ∈ U,
ηi = ξαi − αiξ.
Properties (1) and (2) from Lemma 8.2 are immediate from this definition.
The definition of Ψ requires some preparation. First of all, we identify the space U with
HomC(M(t(a)), M(h(a))), so view Ψ as a linear map
Ψ : Ma∈Q1
HomC(M(t(a)), M(h(a))) →Mb∈Q1
HomC(M(h(b)), M(t(b))) .
Now recall from [9, (3.2)] that each arrow a ∈ Q1 gives rise to a continuous linear map
La∈Q1
(8.6)
such that for every path a1 · · · ad we have
here we use the notation
a1 · · · ap−1 ⊗ ap+1 · · · ad ;
∆a : RhhAii → RhhAii b⊗ RhhAii ,
∆a(a1 · · · ad) = Xp:ap=a
RhhAii b⊗ RhhAii = Yi,j≥0
∆a(∂bS) ∈ RhhAiit(b),h(a) b⊗ RhhAiit(a),h(b) ;
(A⊗Ri ⊗ A⊗Rj) ,
and the convention that if a1 = a (resp. ad = a) then the corresponding term in (8.6) is
eh(a) ⊗ a2 · · · ad (resp. a1 · · · ad−1 ⊗ et(a)). In particular, for every a, b ∈ Q1, we have
32
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
accordingly, we express ∆a(∂bS) as
u(ν)
b,a ⊗ v(ν)
(8.7)
a,b
∆a(∂bS) =Xν
(u(ν)
b,a ∈ RhhAiit(b),h(a), v(ν)
a,b ∈ RhhAiit(a),h(b)) .
Now we define the component Ψb,a : HomC(M(t(a)), M(h(a))) → HomC(M(h(b)), M(t(b)))
of Ψ by setting
(8.8)
Ψb,a(ηa) =Xν
(u(ν)
b,a )M ◦ ηa ◦ (v(ν)
a,b )M .
We postpone the proof of property (3) in Lemma 8.2, that is, of the equality Ψ ◦ Φ = 0,
until Section 10, see Corollary 10.2 and Remark 10.3 below.
It remains to check property (4). We start with the following observation which is a direct
consequence of the definitions: for every pair of arrows a and b, we have
(8.9)
∆a∂bS ≡ et(b) ⊗ δh(a),t(b)∂baS mod mb⊗RhhAii .
In view of (1.11) and (8.8), it follows that, for every ηa ∈ HomC(M(t(a)), M(h(a))), the
morphism Ψb,a(ηa) − δh(a),t(b)ηa ◦ γa,b ∈ HomC(M(h(b)), M(t(b))) is a linear combination of
the morphisms of the form uM ◦ ηa ◦ vM with u ∈ m, v ∈ RhhAii.
Since m acts nilpotently on M, we have a descending filtration of R-modules
For p = 0, . . . , ℓ − 1, choose a R-submodule M (p) in M such that
M ⊃ mM ⊃ · · · ⊃ m
ℓM = 0 .
m
pM = M (p) ⊕ m
p+1M .
For s ∈ C⋆, define λ(s) ∈ EndR(M) as the R-module automorphism of M acting on each
M (p) as multiplication by sp. This definition makes it clear that
(8.10)
for each u ∈ m.
λ(s) ◦ uM ◦ λ(s)−1 = 0
lim
s→0
Now for each s ∈ C⋆, define the linear map
Ψ(s) = (Ψ(s)
b,a) : U =Ma
HomC(M(t(a)), M(h(a))) →Mb
by setting
HomC(M(h(b)), M(t(b)))
Ψ(s)
b,a(ηa) = λ(s) ◦ Ψb,a(λ(s)−1 ◦ ηa).
Since Ψ(s) is obtained from Ψ by composing it with invertible linear maps on both sides, we
have rank Ψ(s) = rank Ψ for all s ∈ C⋆, and more generally, dim Ψ(s)(U ′) = dim Ψ(U ′) for
any subspace U ′ ⊆ U invariant under the automorphism (ηa) 7→ (λ(s)−1 ◦ ηa). Note that
the subspace U1 ⊆ U given by (8.4) satisfies this condition; indeed, under the identification
of U withLa HomC(M(t(a)), M(h(a))), U1 identifies withLa HomC(M(t(a)), mM(h(a))).
Now consider the linear map Ψ(0) = lims→0 Ψ(s). Since under the continuous deformation
the rank of a linear map depends semi-continuously on the deformation parameter, we
conclude that
dim Ψ(U1) ≥ dim Ψ(0)(U1);
QUIVERS WITH POTENTIALS II
33
to finish the proof of property (4) in Lemma 8.2, it suffices to show that
(8.11)
In view of (8.10), each component Ψ(0)
b,a of Ψ(0) acts by
dim Ψ(0)(U1) =Xi
rkγi · rkαi .
Ψ(0)
b,a(ηa) = δh(a),t(b)ηa ◦ γa,b.
Using natural identifications
Ma
Mb
HomC(M(t(a)), M(h(a))) =Mi
HomC(M(h(b)), M(t(b))) =Mi
HomC(Min(i), M(i)) ,
HomC(Mout(i), M(i)) ,
the operator Ψ(0) translates into the direct sum of operators
Ψ(0)
i
: HomC(Min(i), M(i)) → HomC(Mout(i), M(i))
acting by
Ψ(0)
i (ηi) = ηi ◦ γi .
This description makes (8.11) clear, finishing the proofs of Lemma 8.2 and Theorem 8.1. (cid:3)
The following corollary is immediate from (8.1).
Corollary 8.3. Suppose M = (M, V ) is a QP-representation such that E(M) = 0. Then
for every vertex k we have:
(1) Either M(k) = {0} or V (k) = {0};
(2) Either ker βk = {0} or im βk = ker γk.
Since E(M) is invariant under mutations, Corollary 8.3 implies that if E(M) = 0 then
every QP-representation obtained from M by a sequence of mutations satisfies the properties
(1) and (2). The following example shows that the converse is not true.
Let Q be the Kronecker quiver
a,b
1
/ 2 .
For every positive integer n, let Mn = (Mn, {0}) be the indecomposable positive QP-
representation of (Q, 0) such that Mn(1) = Mn(2) = Cn, and the linear maps aMn and
bMn from Mn(1) to Mn(2) are as follows: aMn = I is the identity map, while bMn = J is
the nilpotent Jordan n-block. Recalling (1.13), we see that the g-vector of Mn is equal to
(n, −n). Also HomQ(Mn, Mn) is naturally isomorphic to the centralizer of J in End(Cn),
hence we have hMn, Mni = n. Recalling (7.4), we get
E(Mn) = hMn, Mni + d1(Mn)g1(Mn) + d2(Mn)g2(Mn) = n + n2 − n2 = n .
On the other hand, it is easy to see that Mn as well as all representations obtained from it
by mutations, satisfy properties (1) and (2) in Corollary 8.3 (in fact, every representation
obtained from Mn by mutations is either right-equivalent to Mn, or differs from it just by
interchanging vertices 1 and 2).
/
/
/
34
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
9. Applications to cluster algebras
Proof of Theorem 1.7. We fix t0, t ∈ Tn, a skew-symmetric integer n × n matrix B, and a
nondegenerate potential S on the quiver Q = Q(B). Recall that in Theorem 5.1 the g-vector
gB;t0
from the theory of cluster algebras are interpreted as the
ℓ;t
g-vector gM and the F -polynomial FM associated with the QP-representation M = MB;t0
ℓ;t
of (Q, S) given by (5.2).
and the F -polynomial F B;t0
ℓ;t
Conjectures 1.1 and 1.2 are immediate from this interpretation, see Proposition 3.1.
Our next target is Conjecture 1.6. Comparing the desired formula (1.3) with (2.11), we
see that it is enough to prove the equality
(9.1)
min(0, gk) = hk ,
where gk and hk are given by (7.3) and (3.2), respectively. Substituting these expressions
into (9.1) and adding dim ker βk to both sides, we arrive at the equality
k (M)) = 0 .
min(dim(ker βk), dim(ker γk/im βk) + d−
(9.2)
To finish the proof, it remains to observe that, in view of Corollary 7.2, the QP-representation
M satisfies properties (1) and (2) in Corollary 8.3, and that (9.2) clearly holds for any rep-
resentation with these properties.
Now we are ready to prove Conjecture 1.3. The key observation is that the above argument
ℓ;t but also for the direct sum
proves the equality (9.1) not only for each representation MB;t0
M = MB;t0
1;t ⊕ · · · ⊕ MB;t0
n;t
(since M satisfies the assumption in Corollary 7.2). In view of Lemma 5.2, we have gk =
hk − h′
k = − dim ker βk is the k-th component of the vector hM for the QP-
representation M = µk(M). Thus, (9.1) is equivalent to
k, where h′
(9.3)
max(hk, h′
k) = 0 .
Suppose that hk = 0, that is, ker βk = 0. Then the same property holds for each direct
summand Mℓ = MB;t0
ℓ;t
, implying that
gk(Mℓ) = −hk(Mℓ) ≥ 0
for all ℓ = 1, . . . , n. This shows that the k-th coordinates of the vectors gB;t0
1;t
nonnegative. If h′
vectors are nonpositive, finishing the proof of Conjecture 1.3.
n;t are
k = 0, then the same argument shows that the k-th coordinates of all these
, . . . , gB;t0
As for Conjecture 1.4, it is an easy consequence of already proven Conjectures 1.6 and
1.3 combined with the following observation made already in [12, Remark 7.14]:
if g1, . . . , gn are sign-coherent vectors forming a Z-basis in Zn, then the
transformation (1.3) sends them to a Z-basis in Zn.
(9.4)
Indeed, to show that the vectors gB;t0
form a Z-basis of Zn, proceed by induction
1;t
on the distance between t0 and t in Tn. The basic step t = t0 is clear from (2.1), and the
inductive step follows from (9.4).
, . . . , gB;t0
n;t
To finish the proof of Theorem 1.7, it remains to prove Conjecture 1.5.
Lemma 9.1. For a QP-representation M = (M, V ), the following conditions are equivalent:
QUIVERS WITH POTENTIALS II
35
(1) M is negative, i.e., M = {0}.
(2) E(M) = 0, and the g-vector gM = (g1, . . . , gn) is nonnegative.
Under these conditions, we have dim V (i) = gi for all i, so M is uniquely determined by its
g-vector.
Proof. The only non-trivial statement is the implication (2) =⇒ (1). We have already
established that the equality E(M) = 0 implies (9.1), so if gM is nonnegative then hk = 0
for all k. Thus we have ker βk = 0 for all k. It remains to observe that the latter condition
cannot hold for a nonzero nilpotent quiver representation M. Indeed, if mℓ−1M 6= 0, and
mℓM = 0 for some ℓ ≥ 1, then 0 6= mℓ−1M ⊆ ⊕k ker βk, finishing the proof.
(cid:3)
Lemma 9.2. Let M and M′ be QP-representations of the same nondegenerate QP, and
suppose that M′ is mutation-equivalent to a negative representation. The following condi-
tions are equivalent:
(1) M is right-equivalent to M′.
(2) E(M) = 0, and gM = gM′.
Proof. Again only the implication (2) =⇒ (1) needs a proof. Since E(M) = E(M′) =
0, the already established formula (1.3) shows that the g-vectors remain the same under
applying to M and M′ the same sequence of mutations. Since mutations also preserve
right-equivalence, in proving that (2) =⇒ (1) we may assume that M′ is negative, in which
case the statement follows from Lemma 9.1.
(cid:3)
Under the assumptions of Conjecture 1.5, consider the QP-representations
Also we have E(M) = E(M′) = 0 since both M and M′ are mutation-equivalent to
negative QP-representations. By Lemma 9.2, M is right-equivalent to M′. Because of the
uniqueness of the decomposition into indecomposables, there exists a bijection σ : I → I ′
such that MB;t0
is right-equivalent to MB;t0
σ(i) for i ∈ I. Thus we have
i;t
σ(i);t′, and ai = a′
i;t = F B;t0
σ(i);t′, F B;t0
σ(i);t′
gB;t0
i;t = gB;t0
for all i ∈ I, finishing the proofs of Conjecture 1.5 and of Theorem 1.7.
(cid:3)
10. Homological interpretation of the E-invariant
Throughout this section we fix a quiver Q without oriented 2-cycles, and a QP (Q, S). Let
M = (M, V ) and N = (N, W ) be two QP-representations of (Q, S). Our aim is to associate
to M and N a vector space E inj(M, N ) such that dim E inj(M, N ) = Einj(M, N ), the integer
function defined in (7.4).
It will be more convenient for us to work with the "twisted"
function Eproj(M, N ) = Einj(N ⋆, M⋆), where M⋆ and N ⋆ are QP-representations of the
In view of (5.1), we have
(MB;t0
M =Mi∈I
gM =Xi∈I
i;t )ai, M′ =Mi∈I ′
i;t =Xi∈I ′
aigB;t0
aigB;t0
(MB;t0
i;t′ )a′
i .
i;t′ = gM′ .
36
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
opposite QP (Qop, Sop) constructed before Proposition 7.3. Clearly, we have hN ⋆, M ⋆i =
hM, Ni, implying
(10.1)
Eproj(M, N ) = hM, Ni + Xk∈Q0
gk(M⋆)dk(N).
We turn to the construction of a vector space E proj(M, N ) such that
(10.2)
dim E proj(M, N ) = Eproj(M, N ).
Let P(Q, S) be the Jacobian algebra of (Q, S) (see Section 2). In the rest of the section we
assume that
(10.3)
the potential S belongs to the path algebra RhAi, and the two-sided
ideal J0 in RhAi generated by all cyclic derivatives ∂aS contains some
power mN .
(Recall that in our general setup, S belongs to the completed path algebra RhhAii, and the
Jacobian ideal J of S is the closure of J0 in RhhAii.) Under this assumption, the Jacobian
algebra P(Q, S) = RhhAii/J is identified with RhAi/J0, and it is finite-dimensional. In this
situation, all the P(Q, S)-modules considered below will be finite-dimensional as well.
For every vertex k ∈ Q0 let Pk denote the indecomposable projective P(Q, S)-module
corresponding to k. Recall that Pk is given by
(10.4)
Pk =Mi∈Q0
P(Q, S)i,k,
where the double Q0-grading on P(Q, S) comes from the R-bimodule structure (see Sec-
tion 4). In particular, each Pk is finite-dimensional in view of (10.3).
To every (finite-dimensional) P(Q, S)-module M, we associate the sequence of P(Q, S)-
module homomorphisms
(10.5) Mb∈Q1
ψ→ Ma∈Q1
ϕ→ Mk∈Q0
(Pt(b) ⊗ M(h(b)))
(Ph(a) ⊗ M(t(a)))
(Pk ⊗ M(k)) ev→ M → 0
defined as follows. The P(Q, S)-module homomorphisms ev and ϕ are given by
(10.6)
and
(10.7)
ev(p ⊗ m) = pm (p ∈ Pk, m ∈ M(k)),
ϕ(p ⊗ m) = pa ⊗ m − p ⊗ aM (m)
(p ∈ Ph(a), m ∈ M(t(a))),
while the component ψa,b : Pt(b) ⊗ M(h(b)) → Ph(a) ⊗ M(t(a)) of ψ is given by (in the
notation of (8.7) and (8.8))
(10.8)
ψa,b(p ⊗ m) =Xν
pu(ν)
b,a ⊗ (v(ν)
a,b )M · m .
Proposition 10.1. The sequence (10.5) is exact.
Proof. As pointed out by the referee, this proposition follows from the results of [5]. For the
convenience of the reader we present some details (also kindly provided by the referee). To
make our notation closer to that of [5], in the following argument we denote the Jacobian
algebra RhAi/J0 by Λ, and rename J0 into I.
QUIVERS WITH POTENTIALS II
37
The ring Λ is a bimodule over itself. If we splice the exact sequence [5, (1.4)] for n = 0
and n = 1 together, we get a bimodule resolution of Λ as follows
(10.9)
Λ ⊗
I
I m + mI
d2
⊗ Λ
/ Λ ⊗ A ⊗ Λ
d1
d0
/ Λ ⊗ Λ
/ Λ
/ 0.
Here the tensor products are over R, and we have identified A with m/m2. Note that
I/(I m + mI) is spanned by all partial derivatives of the potential. The differential d2 and
d1 are given after (1.3) in [5]. Define
µ : RhAi → RhAi ⊗ A ⊗ RhAi
by µ(a1a2 · · · as) =Ps
i=1 a1 · · · ai−1 ⊗ ai ⊗ ai+1 · · · as (this map is denoted by ∆ in [5]). Then
d2 sends the residue class of an element 1 ⊗ u ⊗ 1 to ∆(u). The partial derivative ∂ξ was
defined for ξ ∈ A⋆ in [9, (3.1)]. By identifying A with A⋆ using a basis of arrows, we have
defined ∂b for an arrow b ∈ A. We have a surjection
(10.10)
A⋆ →
I
I m + mI
defined by ξ 7→ ∂ξS +I m+ mI. We can replace the module on the left in (10.9) by Λ⊗A⋆ ⊗Λ
using (10.10). Therefore, we have an exact sequence:
(10.11)
Λ ⊗ A⋆ ⊗ Λ
/ Λ ⊗ A ⊗ Λ
/ Λ ⊗ Λ
/ Λ
/ 0.
If we apply the functor • ⊗Λ M to (10.11), we obtain (10.5). Note that the sequence remains
exact after applying the functor, because (10.11) splits as a sequence of right Λ-modules.
(cid:3)
Corollary 10.2. The maps Φ and Ψ given by (8.5) and (8.8) satisfy the condition Ψ◦Φ = 0.
Proof. Note that, for every P(Q, S)-module M, a vector space U, and a vertex k ∈ Q0,
there is a natural isomorphism
(10.12)
HomC(U, M(k)) → HomP(Q,S)(Pk ⊗ U, M)
sending σ ∈ HomC(U, M(k)) to the composed morphism
Pk ⊗ U id⊗σ−→ Pk ⊗ M(k) ev→ M .
An easy check shows that the maps Φ and Ψ are obtained from the maps ϕ and ψ given
by (10.7) and (10.8) by applying the contravariant functor HomP(Q,S)(−, M) and using
the isomorphism in (10.12). So the statement in question follows from the exactness of
(10.5).
(cid:3)
Remark 10.3. Corollary 10.2 is still true without the assumption (10.3).
In the more
general case, the modules Pi may be infinite dimensional, but the composition ϕ ◦ ψ is still
equal to 0 in (10.5).
The sequence (10.5) produces a presentation of the P(Q, S)-module M, which can be
rewritten as
(10.13)
(Pk ⊗ Min(k))
Mk∈Q0
ϕ
→ Mk∈Q0
(Pk ⊗ M(k)) ev→ M → 0,
/
/
/
/
/
/
/
/
38
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
where with some abuse of notation we use the same symbol ϕ for the leftmost map: this
map is now given by
(10.14)
(pa ⊗ prt(a)m) − p ⊗ αk(m)
(p ∈ Pk, m ∈ Min(k))
ϕ(p ⊗ m) = Xh(a)=k
(here prt(a) stands for the projection Min(k) =Lh(a)=k M(t(a)) → M(t(a))).
We claim that the presentation (10.13) can be truncated as follows. For every k ∈ Q0,
choose subspaces U ′
k, U ′′
k ⊆ Min(k) and M (0)(k) ⊆ M(k) such that
(10.15)
ker(αk) = im(γk) ⊕ U ′
k, Min(k) = ker(αk) ⊕ U ′′
k , M(k) = M (0)(k) ⊕ im(αk),
and consider the projective P(Q, S)-modules
(10.16)
Proposition 10.4.
P ′ = Mk∈Q0
P (1) = Mk∈Q0
(Pk ⊗ ker(αk)), P ′′ = Mk∈Q0
k), P (0) = Mk∈Q0
(Pk ⊗ U ′
(Pk ⊗ U ′′
k ),
(Pk ⊗ M (0)(k)).
(1) For every p′ ∈ P ′, there exists a unique p′′ ∈ P ′′ such that ϕ(p′ − p′′) ∈ P (0). The
map ϕ : P ′ → P (0) given by ϕ(p′) = ϕ(p′ − p′′) is a P(Q, S)-module homomorphism.
(2) The restrictions of ϕ to P (1) and of ev to P (0) make the sequence
(10.17)
P (1) ϕ
→ P (0) ev→ M → 0
exact, thus giving a presentation of M.
(3) The presentation (10.17) is minimal, that is, the map ϕ : P (1) → P (0) induces
an isomorphism P (1)/mP (1) → im(ϕ)/m im(ϕ), where m is the maximal ideal in
P(Q, S).
Before proving Proposition 10.4, we use it to construct the space E proj(M, N ) (for any
QP-representations M = (M, V ) and N = (N, W ) of (Q, S)) satisfying (10.2). Note that
the P(Q, S)-module homomorphism ϕ : P (1) → P (0) in Proposition 10.4 induces a C-linear
map
ϕ⋆ : HomP(Q,S)(P (0), N) → HomP(Q,S)(P (1), N).
We now define the space E proj(M, N) as the cokernel of ϕ⋆, that is, from an exact sequence
(10.18)
HomP(Q,S)(P (0), N)
ϕ⋆
→ HomP(Q,S)(P (1), N) → E proj(M, N) → 0 .
And finally we set
(10.19)
E proj(M, N ) = E proj(M, N) ⊕ HomR(V, N) .
Theorem 10.5. The space E proj(M, N ) satisfies (10.2), i.e., its dimension is given by
(10.1).
Proof. Using the presentation (10.17), we include (10.18) into a longer exact sequence
(10.20)
0 → HomP(Q,S)(M, N) → HomP(Q,S)(P (0)(M), N)
→ HomP(Q,S)(P (1)(M), N) → E proj(M, N) → 0 .
QUIVERS WITH POTENTIALS II
39
Computing the dimensions of the terms in (10.20), we get
dim E proj(M, N) =
hM, Ni − dim HomP(Q,S)(P (0)(M), N) + dim HomP(Q,S)(P (1)(M), N) =
hM, Ni − Xk∈Q0
hM, Ni + Xk∈Q0
hM, Ni + Xk∈Q0
dim coker(αk;M ) · dk(N) + Xk∈Q0
dim
ker(αk;M )
im(γk;M )
· dk(N) =
(dim Min(k) − dk(M) + rk(γk;M)) · dk(N) =
(gk(M⋆) − d−
k (M)) · dk(N)
(for the last equality see (7.12)); note that in view of (10.12), HomP(Q,S)(Pk, N) is naturally
isomorphic to N(k), hence dim HomP(Q,S)(Pk, N) = dk(N).
To finish the proof of (10.2), it remains to note that
HomR(V, N) = Mk∈Q0
dim HomR(V, N) = Xk∈Q0
HomC(V (k), N(k)),
d−
k (M)dk(N).
dim E proj(M, N ) = dim E proj(M, N) + Xk∈Q0
= hM, Ni + Xk∈Q0
gk(M⋆)dk(N) = Eproj(M, N )
d−
k (M)dk(N)
(cid:3)
implying
Thus,
by (10.1).
Proof of Proposition 10.4. We start by showing that the map ev : P (0) → M is surjective.
This is a special case of the following lemma.
Lemma 10.6. Suppose η : K → L is a surjection of finite-dimensional P(Q, S)-modules.
Suppose that K = K ′ ⊕ K ′′ is the direct sum of two submodules, and that η(K ′′) ⊆ mL.
Then η(K ′) = L.
Proof. Choose the direct complement L(0) to mL in L. Then L(0) generates L as a P(Q, S)-
module. Indeed, we have
L = mL + L(0) = m
2L + mL(0) + L(0) = · · · = m
N +1L + m
N L(0) + · · · + L(0)
for each N ≥ 0; choosing N big enough so that mN +1L = {0}, we see that L = P(Q, S)L(0).
This argument also shows that mL = mL(0).
Since η is a homomorphism of P(Q, S)-modules, to prove that η(K ′) = L, it suffices to
show that L(0) ⊆ η(K ′). Using the surjectivity of η : K → L and the inclusion η(K ′′) ⊆
40
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
mL(0), we get
L(0) ⊆ η(K ′) + mL(0) ⊆ η(K ′) + mη(K ′) + m
N η(K ′) + m
⊆ η(K ′) + mη(K ′) + · · · + m
2L(0) ⊆ · · ·
N +1L(0) = η(K ′) + m
N +1L(0)
for each N ≥ 0, implying as above that L(0) ⊆ η(K ′).
(cid:3)
Now the fact that ev(P (0)) = M follows by applying Lemma 10.6 to the map ev :
(Pk ⊗ M(k)) → M in place of η : K → L, and to the submodules K ′ and K ′′
Lk∈Q0
given by
(10.21)
K ′ = P (0), K ′′ = Mk∈Q0
(Pk ⊗ im(αk)).
Continuing the proof of Proposition 10.4, we adopt the notation in (10.16) and (10.21),
thus viewing ϕ as a homomorphism of P(Q, S)-modules P ′ ⊕ P ′′ → K ′ ⊕ K ′′. We write ϕ
as ϕ(1) + ϕ(0) in accordance with the decomposition in (10.14). The following properties are
immediate from (10.14):
(10.22)
(10.23)
ϕ(0)P ′ = 0, while the restriction of ϕ(0) to P ′′ is an isomorphism between
P ′′ and K ′′;
im(ϕ(1)) ⊆ m(K ′ ⊕ K ′′).
We claim that these properties imply the following:
(10.24)
(10.25)
(10.26)
K ′′ ⊆ mK ′ + ϕ(P ′′);
K ′ ∩ ϕ(P ′′) = {0};
the restriction of ϕ to P ′′ is injective.
Note that these facts imply Part (1) of Proposition 10.4. Indeed, by (10.24) and (10.25),
we have
(10.27)
K ′ ⊕ K ′′ = K ′ ⊕ ϕ(P ′′).
This allows us to define the map ϕ : P ′ → K ′ = P (0) as the composition pr1 ◦ ϕ, where pr1 is
the projection of K ′ ⊕ K ′′ onto K ′ along ϕ(P ′′). Using (10.26), we see that ϕ is exactly the
map in Part (1) of Proposition 10.4 (the fact that ϕ is a P(Q, S)-module homomorphism is
obvious since so are ϕ and pr1).
To prove (10.24), we use (10.22) and (10.23) to get
K ′′ = ϕ(0)(P ′′) ⊆ ϕ(1)(P ′′) + ϕ(P ′′) ⊆ mK ′ + ϕ(P ′′) + mK ′′
⊆ mK ′ + ϕ(P ′′) + m(mK ′ + ϕ(P ′′) + mK ′′) ⊆ mK ′ + ϕ(P ′′) + m
2K ′′.
Iterating, we see that K ′′ ⊆ mK ′ + ϕ(P ′′) + mN K ′′ for all N ≥ 1, implying the desired
inclusion (10.24).
To prove (10.25) and (10.26), suppose that ϕ(p′′) = k′ for some k′ ∈ K ′ and p′′ ∈ P ′′.
Using (10.22) and (10.23), we see that
k′ − ϕ(0)(p′′) = ϕ(1)(p′′) ∈ m(K ′ ⊕ K ′′),
QUIVERS WITH POTENTIALS II
41
implying that k′ ∈ mK ′ and ϕ(0)(p′′) ∈ mK ′′. Once again applying (10.22), we conclude
that p′′ ∈ mP ′′. Iterating this argument, we conclude that k′ ∈ mN K ′ and p′′ ∈ mN P ′′ for
all N ≥ 1, hence k′ = p′′ = 0, implying both (10.25) and (10.26).
Turning to the proof of Part 2 of Proposition 10.4, we first show that the sequence
(10.28)
P ′ ϕ→ K ′ ev→ M → 0
is exact. The surjectivity of ev : K ′ → M is already proved above, so (using the exactness
of (10.13)) it remains to show that ϕ(P ′) = ϕ(P ′ ⊕ P ′′) ∩ K ′; but this is immediate from
the definition of ϕ.
To prove Part 2, it remains to show that the restriction of the map ϕ : P ′ → K ′ to the
submodule P (1) ⊆ P ′ has the same image as ϕ. Note that P ′ = P (1) ⊕ P ′
1, where
P ′
1 = Mk∈Q0
(Pk ⊗ im(γk)).
By Lemma 10.6, it is enough to show that
ϕ(P ′
(10.29)
1) ⊆ mϕ(P ′).
It follows easily from (10.27) that
mϕ(P ′) = m(ϕ(P ′ ⊕ P ′′) ∩ K ′) = mϕ(P ′ ⊕ P ′′) ∩ K ′,
and also that
ϕ(P ′) = ϕ(1)(P ′) ⊆ m(K ′ ⊕ K ′′) = mK ′ ⊕ mϕ(P ′′),
implying the inclusion ϕ(P ′
a consequence of the following:
1) ⊆ ϕ(P ′
1 ⊕ mP ′′). We see that the desired inclusion (10.29) is
(10.30)
ϕ(P ′
1) ⊆ mϕ(P ′ ⊕ P ′′).
To prove (10.30), it suffices to show that ϕ(ek ⊗ γk(m)) ∈ mϕ(P ′ ⊕ P ′′) for every m ∈
M(h(b)), where b is an arrow with t(b) = k. But this follows from the exactness of the
sequence (10.5) (more precisely, from the fact that im(ψ) ⊆ ker(ϕ)), since in view of (8.9)
we have
(10.31)
ek ⊗ γk(m) ≡ ψ(ek ⊗ m) mod m(P ′ ⊕ P ′′) .
This concludes the proof of Part 2 of Proposition 10.4.
To prove Part 3, note that (10.31) implies the inclusion
ker(ϕ) = im(ψ) ⊆ P ′
1 + m(P ′ ⊕ P ′′).
Now suppose that p ∈ P (1) is such that ϕ(p) ∈ ϕ(mP (1)). Remembering the definition of ϕ,
we conclude that
p ∈ mP (1) + P ′′ + ker(ϕ) ⊆ mP (1) ⊕ P ′
Therefore, p ∈ mP (1), finishing the proof of Proposition 10.4.
1 ⊕ P ′′.
(cid:3)
Remark 10.7. The presentation (10.17) is minimal by part (3) of Proposition 10.4. Minimal
presentations are unique up to isomorphism. One can show that, up to an isomorphism,
the presentation (10.17) does not depend on the choice of splitting subspaces in (10.15).
42
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Remark 10.8. To emphasize the dependence of indecomposable projective modules Pk
(for k ∈ Q0) on the underlying QP (Q, S), we will denote them Pk = Pk(Q, S). The
indecomposable injective P(Q, S)-modules Ik = Ik(Q, S) can be defined by going to the
opposite QP:
(10.32)
Ik(Q, S) = (Pk(Qop, Sop))⋆.
By this definition, there is a duality between projective and injective P(Q, S)-modules:
every exact sequence involving the modules Pk gives rise to the exact sequence (with the
arrows reversed) involving the Ik.
In particular, the presentation (10.17) gives rise to a
"co-presentation"
0 → M → Mk∈Q0
(Ik ⊗ ker(βk)⋆) → Mk∈Q0
(Ik ⊗ U ⋆
k ),
where Uk is a direct complement of im(βk) in ker(γk).
Recall that E proj(M, N) is defined in (10.18). We also define E inj(M, N) = E proj(N ⋆, M ⋆).
Corollary 10.9. We have the following isomorphisms
(10.33)
(10.34)
E proj(M, N) = HomP(Q,S)(N, τ (M))⋆
E inj(M, N) = HomP(Q,S)(τ −1(N), M)⋆
where τ is the Auslander-Reiten translation functor (see e.g., [1, Section IV.2]).
Proof. For this proof we will rely on the book [1]. We should point out that the authors
of [1] use the convention that all modules are right-modules unless stated otherwise, where
we assume modules to be left modules by default. Let ν be the Nakayama functor (see [1,
Section III, Definition 2.8]) from P(Q, S)-modules to P(Q, S)-modules defined by
ν(M) = HomP(Q,S)(M, P(Q, S))⋆.
This functor has the property that
ν(Pk) = Ik
for every vertex k. In particular, we have an isomorphism
(10.35)
HomP(Q,S)(P, M) = HomP(Q,S)(M, ν(P ))⋆
for every projective module P (see [1, Lemma 2.1]). Consider the minimal presentation
(10.17). It follows from [1, Section IV, Proposition 2.4] that the sequence
(10.36)
is exact. If we apply HomP(Q,S)(N, ·)⋆ to (10.36), then it follows from (10.35) that
0 → τ (M) → ν(P (1)) → ν(P (0))
(10.37)
is exact. It follows from (10.20) that E proj(M, N) = HomP(Q,S)(N, τ (M))⋆.
HomP(Q,S)(P (0), N) → HomP(Q,S)(P (1), N) → HomP(Q,S)(N, τ (M))⋆ → 0
We have τ (N ⋆) = τ −1(N)⋆. So it follows that
E inj(M, N) = E proj(N ⋆, M ⋆) = HomP(Q,S)op(M ⋆, τ (N ⋆))⋆ =
= HomP(Q,S)op(M ⋆, τ −1(N)⋆)⋆ = HomP(Q,S)(τ −1(N), M)⋆.
(cid:3)
QUIVERS WITH POTENTIALS II
43
Remark 10.10. Auslander-Reiten duality states that Ext1
P(Q,S)(M, N) is isomorphic to
HomP(Q,S)(M, N)⋆, where HomP(Q,S)(M, N) is equal to HomP(Q,S)(M, N) modulo the mor-
phisms that factor through injective modules (see [1, IV.2, Theorem 2.13]). We may view
Ext1
P(Q,S)(M, N) as a subspace of E proj(M, N). If M has projective dimension ≤ 1 then we
have equality by [1, IV.2, Corollary 2.14]. Similarly, Ext1
P(Q,S)(M, N) can be viewed as a
subspace of E inj(M, N), with equality when N has injective dimension ≤ 1.
References
[1] I. Assem, D. Simson, A. Skowro´nski, Elements of the Representation Theory of Associative Algebras,
London Mathematical Society Student Texts 65, Cambridge University Press, 2006.
[2] A. Bia´lynicki-Birula, On fixed point schemes of actions of multiplicative and additive groups, Topology
12 (1973), 99-103.
[3] A. Buan, O. Iyama, I. Reiten, D. Smith, Mutation of cluster-tilting objects and potentials,
arXiv:0804.3813, to appear in American J. Math.
[4] A. Buan, R. Marsh, I. Reiten, Denominators of cluster variables, J. Lond. Math. Soc. (2) 79 (2009),
no. 3, 589 -- 611.
[5] M. Butler, A. King, Minimal resolutions of algebras, J. Algebra 212 (1999), no.1, 323 -- 362.
[6] P. Caldero, F. Chapoton, Cluster algebras as Hall algebras of quiver representations, Comment. Math.
Helv. 81 (2006), no. 3, 595 -- 616.
[7] P. Caldero, B. Keller, From triangulated categories to cluster algebras II, Annales Sci. de l' ´Ecole
Norm. Sup. (4) 39 (2006), 983-1009.
[8] P. Caldero, M. Reineke, On the quiver Grassmannian in the acyclic case, J. Pure Appl. Algebra 212
(2008), no. 11, 2369 -- 2380.
[9] H. Derksen, J. Weyman, A. Zelevinsky, Quivers with potentials and their representations I: Mutations,
Selecta Math. 14 (2008), no. 1, 59 -- 119..
[10] V. V. Fock, A. B. Goncharov, Cluster ensembles, quantization and the dilogarithm, Annales Sci. de
l' ´Ecole Norm. Sup. (4) 42 (2009), 865 -- 930.
[11] S. Fomin, A. Zelevinsky, Cluster algebras I: Foundations, J. Amer. Math. Soc. 15 (2002), 497 -- 529.
[12] S. Fomin, A. Zelevinsky, Cluster algebras IV: Coefficients, Comp. Math. 143 (2007), 112 -- 164.
[13] C. Fu, B. Keller, On cluster algebras with coefficients and 2-Calabi-Yau categories, Trans. Amer.
Math. Soc. 362 (2010), no. 2, 859 -- 895.
[14] W. Fulton, Introduction to Toric Varieties, Ann. of Math. Stud., vol. 131, Princeton University Press,
1993.
[15] A. Hubery, Acyclic cluster algebras via Ringel-Hall algebras, preprint.
[16] R. Marsh, M. Reineke, A. Zelevinsky, Generalized associahedra via quiver representations. Trans.
Amer. Math. Soc. 355 (2003), no. 10, 4171 -- 4186.
[17] H. Nakajima, Quiver varieties and cluster algebras, arXiv:0905.0002.
[18] A. Schofield, General representations of quivers, Proc. London. Math. Soc. (3) 65 (1992), 46 -- 64.
[19] H. Schubert, Categories, Springer, New York Heidelberg, Berlin, 1972. Revised and enlarged transla-
tion of Kategorien I and Kategorien II, Heidelberger Taschenbucher, Volume 65, 66, 1970.
[20] I. R. Shafarevich, Basic Algebraic Geometry 1 -- Varieties in Projective Space, second ed., Springer,
1994.
44
HARM DERKSEN, JERZY WEYMAN AND ANDREI ZELEVINSKY
Harm Derksen
Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, USA
E-mail address: [email protected]
Jerzy Weyman
Department of Mathematics, Northeastern University, Boston, MA 02115, USA
E-mail address: [email protected]
Andrei Zelevinsky
Department of Mathematics, Northeastern University, Boston, MA 02115, USA
E-mail address: [email protected]
|
1205.3673 | 1 | 1205 | 2012-05-12T03:27:16 | Special bases for the vector space of square matrices | [
"math.RA"
] | We describe families of complete orthogonal bases of full rank matrices which span the vector spaces of square matrices. The proposed bases generalise non-trivially the Pauli matrice while shedding light on their algebraic properties. Finally we introduce the notion k -pseudo-closure for orthogonal bases spanning vector subspaces of square matrices and discuss their connections with hadamard matrices. | math.RA | math |
SPECIAL BASES FOR THE VECTOR SPACE OF SQUARE
MATRICES
EDINAH K. GNANG
Abstract. We describe families of complete orthogonal bases of full rank
matrices which span the vector spaces of square matrices. The proposed bases
generalise non-trivially the Pauli matrice while shedding light on their alge-
braic properties. Finally we introduce the notion k-pseudo-closure for orthog-
onal bases spanning vector subspaces of square matrices and discuss their
connections with hadamard matrices.
1. Introduction
It is a common practice when manipulating n × n matrices to express them as
linear combinations of basis elements from the complete orthonormal basis
Bn =(cid:8)eiet
j(cid:9)0≤i,j<n
where the set {ei}0≤i<n denotes the canonical euclidean basis of n-dimensional
column vectors. Although a prevalent choice as canonical basis for expressing square
matrices, we observe that the basis Bn is not as well behaved with respect to the
matrix product operation, more precisely
(cid:0)ei1 et
j1(cid:1)(cid:0)ei2 et
j2(cid:1) =( ei1 et
0
j2
if j1 = i2
otherwise
,
in addition elements of Bn are rank 1 matrices. An important insight confered by
linear algebra is the fact that the choice of a basis for a vector spaces is crucial
to the investigation of properties of a particular set of vectors.
Incidentally we
discuss here alternative complete orthogonal bases of full rank matrices which in
some instances are also pseudo-closed under matrix product as defined bellow.
Definition 1. [Pseudo-closure] Let k ≥ 1 and G denote an orthogonal basis for a
subspace of square matrices. G is said to be pseudo-closed of order k with respect
to the matrix product operation or simply a k-pseudo-closed basis if
(1.1)
∀A, B ∈ G, ( A⋆k+1
B⋆k+1
= A
= B
,
∃ D a diagonal matrix s.t. D (A B) ∈ G and Dk = I.
1991 Mathematics Subject Classification. 15A52.
Edinah K. Gnang is supported by NSF grant DGE-0549115.
1
2
EDINAH K. GNANG
where for A = (ai,j)0≤i,j<n we define A⋆k+1
:=(cid:0)ak+1
i,j (cid:1)0≤i,j<n
.
2. Orthonormal basis Induced by the Unitary group
Let {ei}0≤i<n denote the canonical euclidean basis of column vectors. We define
the fourier complete orthogonal basis to be the set Fn,
Fn :=
B (k, l) = X0≤j<n(cid:18)ej et
{j+k mod n}(cid:19) exp(cid:26)i
2π
n
.
j l(cid:27)
0≤k,l<n
We think of the set of matrices as an inner-product space with the inner-product
being defined as follows
(2.1)
hence
(2.2)
hA, Mi := Tr(cid:8)A M†(cid:9)
∀ A ∈ Cn×n, A = X0≤k,l<n
n−1 hA, B (k, l)i B (k, l) .
The proposed fourier basis generalizes the Pauli Matrices. In contrast to the con-
ventional canonical orthonormal matrix basis Bn, the basis elements of the fourier
basis are each full rank and it follows from their definition that they consititute
a n-pseudo-closed complete orthogonal basis spanning the vector space of n × n
matrices. It also follows from the defintion of Fn that it's elements generate a finite
mulitplicative group of matrices of order bounded by nn+1. Furthermore the group
generated by Fn is isomorphic to a subgroup of Sn2 which we represente here using
matrices in {0, 1}n2×n2
. Let T denote the group Isomorphism
T : Cn×n → Cn2×n2
(2.3)
T (B (k, l)) = X0≤u<n(cid:18)euet
{u+k mod n}(cid:19)⊗
X0≤v<n(cid:18)evet
{v+j×l mod n}(cid:19)
= F (k, l) ,
the isomorphism is naturally extended to the set of all n × n matrice as follows
(2.4)
∀A ∈ Cn×n,
T (A) = X0≤k,l<n
n−1 hA, B (k, l)i F (k, l) .
so that
(2.5)
( T (A1 + A2) = T (A1) × T (A2)
T (A1 × A2) = T (A1) × T (A2)
∀A1, A2 ∈ Cn×n .
We also note that the matrix P0≤k<n B (k, k) is unitary and correponds to the
Discrete Fourier Transform (DFT) matrix. from which it follows that the matrix
SPECIAL BASES FOR THE VECTOR SPACE OF SQUARE MATRICES
3
product of the DFT matrix with an arbitrary matrix A with entries in Qhei 2π
3 i
can be recovered without using complex numbers at all.
Incidentaly there are families of complete orthogonal matrix basis analogous to
the fourier basis associated with arbitrary unitary matrices and expressed by
(2.6)
where
QU (k, l) = X0≤j<n(cid:18)ej et
{j+k mod n}(cid:19) uj l
U U† = I
0≤k,l<n
0≤k,l<n
It also follows from this observation that for a hadamard matrix H, the orthonormal
basis
(2.7)
QH (k, l) = X0≤j<n(cid:18)ej et
{j+k mod n}(cid:19) hj l
is a 2-pseudo-closed complete orthogonal basis spanning the set of n × n matrices.
Theorem 2. Let Mn denote the vector space of n×n matrices with complex entries
there exist a set of 2-pseudo-close complete orthgonal basis of full rank matrices B
which spans Mn if and only if
(2.8)
∃H ∈ Mn such that HT H = I and H ⋆ H = 1n×n.
Similarly to the fourier basis case, the complete orthogonal matrix basis associated
with a hadamard matrix H generate a finite matrix group H of order bouded by
n2n which we call the Hadamard group. The Hadamard group is isomorphic to a
subgroup of S2n and the group isomorphism R is described bellow
Corollary 3. It follows that hadamard matrix H induces a non trivial injective
map
R : Mn → M2n
∀A ∈ Cn×n,
(2.9)
R (A) = X0≤k,l<n
n−1 hA, QH (k, l)i X0≤j<n(cid:18)ej et
{j+k mod n}(cid:19)⊗
so that
(2.10)
( R (A + B) = R (A) + R (B)
R (A × B) = R (A) × R (B)
.
0
1
0
1
0
1 ! if hj l = 1
0 ! if hj l = −1
1
.
4
EDINAH K. GNANG
Proof. The fact that Hadamard matrices can be used to construct 2-pseudo-close
complete orthogonal basis is immediate from the discussion in section[1], conse-
quently our proof shall focus on showing that the existence of a 2-pseudo-close
complete orthogonal basis of full rank matrices B = {N (k, l)}0≤k,l<n, implies the
existence of an n × n hadamard matrix.
If B is a complete basis then it must also express diagonal matrices, hence for a
diagonal matrix D we have
(2.11)
D = X0≤k,l<n
1
kN (k, l)k2
ℓ2
hD, N (k, l)i N (k, l) .
We recall that the defining property for diagonal matrices is the fact that
(2.12)
∀m > 2 Dm = D⋆m
from which it follows that the elements of B for which hD, N (k, l)i 6= 0 must also
be diagonal matrices. Futhermore since the basis element have entries belong to
the set {0, ±1} and most importantly are full rank and it follows that the diagonal
entries should be non zero and the diagonal elements of B should span a vector
space of dimension n , which completes the proof (cid:3).
(cid:3)
3. Conclusion
We have discussed here a variety of matrix bases and illustrated how the notion of
k-pseudo closure for complete orthogonal matrix bases ties together closure proper-
ties with respect to the matrix product operation so often associated with groups on
one hand and complete orthogonal basis commonly associated with vector spaces on
the other hand. We point out that the k-pseudo-close complete matrix basis gener-
alize Pauli matrices and simultaneously provide us with an alternative approach to
generalizing the algebra of quaternions and octonions. Furthermore by analogy to
discrete fourier analysis we argue that these basis suggest a natural framework for
matrix fourier transform providing us with a choice of basis from which on might
select the one which is best suited to some particular application. Finally from an
algorithmic point of view, recalling the fact that the fast fourier transform plays
a crucial role for fast integer mulitplication algorithm, It might be of interest to
investigate whether matrix fourier transform and their corresponding convolution
products also suggest efficient algorithms for matrix multiplication.
4. acknowledgement
The author is indebted to Prof. Doron Zeilberger, Prof. Vladimir Retakh, Prof.
Ahmed Elgammal and Prof. Henry Cohn for insightful discussions and precious
advice.
SPECIAL BASES FOR THE VECTOR SPACE OF SQUARE MATRICES
5
References
[1] J. Patera and H. Zassenhaus, The Pauli Matrices in n dimensions grading of simple
Lie Algebras of type An−1, AIP Journal of Mathematical Physics, 1987
[2] Antonio
Lao,
Hadmard
vs.
Pauli-Dirac
Matrices
www.toequest.com/forum/attachment.php?attachmentid=1589.
[3] Kazuyuki Fujii, Quantum Optical Construction of Generalized Pauli and Walsh-
Hadamard Matrices in Three Level Systems, arXiv:quant-ph/0309132v1
[4] Maurice R Kibler, Variations on a theme of Heisenberg, Pauli and Weyl (2008) J.
Phys. A: Math. Theor. 41 375302.
Department of Computer Science, Rutgers, New Brunswick NJ 08854-8019
E-mail address: [email protected]
|
1405.1575 | 3 | 1405 | 2015-08-30T14:22:41 | Primitive spaces of matrices with upper rank two over the field with two elements | [
"math.RA"
] | For fields with more than $2$ elements, the classification of the vector spaces of matrices with rank at most $2$ is already known. In this work, we complete that classification for the field $\mathbb{F}_2$. We apply the results to obtain the classification of triples of locally linearly dependent operators over $\mathbb{F}_2$, the classification of the $3$-dimensional subspaces of $\text{M}_3(\mathbb{F}_2)$ in which no matrix has a non-zero eigenvalue, and the classification of the $3$-dimensional affine spaces that are included in the general linear group $\text{GL}_3(\mathbb{F}_2)$. | math.RA | math | Primitive spaces of matrices with upper rank two over
the field with two elements
Cl´ement de Seguins Pazzis∗†
September 24, 2018
Abstract
For fields with more than 2 elements, the classification of the vector
spaces of matrices with rank at most 2 is already known. In this work, we
complete that classification for the field F2. We apply the results to obtain
the classification of triples of locally linearly dependent operators over F2,
the classification of the 3-dimensional subspaces of M3(F2) in which no
matrix has a non-zero eigenvalue, and the classification of the 3-dimensional
affine spaces that are included in the general linear group GL3(F2).
AMS Classification: 15A04, 15A30, 15A03
Keywords: spaces of bounded rank matrices, field with two elements, trivial
spectrum spaces.
1
Introduction
Let n and p be non-negative integers and K be an arbitrary field. Given integers
i and j such that i ≤ j, we denote by [[i, j]] the set of all integers k such that
i ≤ k ≤ j.
Given vector spaces U and V over K, one denotes by L(U, V ) the space of
all linear maps from U to V .
We denote by Mn,p(K) the space of all n × p matrices with entries in K, by
Mn(K) the space of all n × n matrices with entries in K, by Sn(K) (respectively,
∗Universit´e de Versailles Saint-Quentin-en-Yvelines, Laboratoire de Math´ematiques de Ver-
sailles, 45 avenue des Etats-Unis, 78035 Versailles cedex, France
†e-mail address: [email protected]
1
5
1
0
2
g
u
A
0
3
]
.
A
R
h
t
a
m
[
3
v
5
7
5
1
.
5
0
4
1
:
v
i
X
r
a
by An(K), by T +
n (K), by NTn(K)) the space of all n × n symmetric (respectively,
alternating, upper-triangular, and strictly upper-triangular) matrices with en-
tries in K. Recall that an alternating matrix is a skew-symmetric matrix in
which the diagonal entries equal zero. The group of all invertible matrices of
Mn(K) is denoted by GLn(K). Given a matrix M ∈ Mn,p(K), the entry of M at
the (i, j)-spot will be denoted by mi,j or, alternatively, by Mi,j. Given a matrix
M ∈ Mn,p(K), a scalar λ ∈ K and distinct integers i and j in [[1, n]] (respectively,
in [[1, p]]), the row operation Li ← Li + λLj (respectively, the column opera-
tion Ci ← Ci + λCj), takes M to the matrix with the same rows (respectively,
columns) except the i-th one, which equals the sum of the i-th row (respectively,
column) of M with the product of the j-th row (respectively, column) of M by
λ. One defines the row swap Li ↔ Lj (respectively, the column swap Ci ↔ Cj)
likewise.
The upper rank of a linear subspace V of Mn,p(K) is the maximal rank for a
matrix in V: We denote it by urk(V). Two linear subspaces V1 and V2 of Mn,p(K)
are called equivalent, and we write V1 ∼ V2, when there are non-singular matrices
P ∈ GLn(K) and Q ∈ GLp(K) such that V2 = P V1 Q, meaning that V1 and V2
represent the same space of linear operators between finite-dimensional vector
spaces in different choices of bases of the source and target spaces. If n = p we
say that V1 and V2 are similar, and we write V1 ≃ V2, when the above condition
holds with Q = P −1.
A linear subspace V of Mn,p(K) with upper rank r is called primitive when
it satisfies the following conditions:
(i) No non-zero vector belongs to the kernel of every matrix of V.
(ii) The span of all the ranges of the matrices of V is Kn.
where urk H(T ) ≤ r − 1.
[?]n×1(cid:3)
(iii) V is not equivalent to a space T of matrices of the form M =(cid:2)H(M )
(iv) V is not equivalent to a space T of matrices of the form M = (cid:20)H(M )
[?]1×p(cid:21)
where urk H(T ) ≤ r − 1.
We say that V is reduced whenever it satisfies conditions (i) and (ii), and
semi-primitive whenever it satisfies conditions (i), (ii) and (iii). Note that
those definitions are invariant under replacing V with an equivalent subspace.
Thus, we can define primitive/semi-primitive/reduced operator spaces between
finite-dimensional vector spaces.
2
2(cid:19) or
In particular, for
Primitive spaces of bounded rank matrices were initially introduced by Atkin-
son and Lloyd [1, 2] and later rediscovered by Eisenbud and Harris [4].
In
particular, Atkinson proved a general classification theorem [1] for all primi-
tive subspaces of Mn,p(K) with upper rank r and for which n > 1 +(cid:18)r
p > 1 +(cid:18)r
2(cid:19), provided that K has more than r elements.
r = 2, his theorem yields that, up to equivalence, the space A3(K) of all 3 × 3 al-
ternating matrices is the sole primitive matrix space with upper rank 2 provided
that the underlying field has more than 2 elements.
Recent new insights have put the theory of primitive spaces back into the
spotlight. First of all, Atkinson's classification theorem for primitive spaces
(and, more precisely, its generalization to semi-primitive spaces as given in [7])
has been shown to yield a generalization of Gerstenhaber's theorem for fields
with large cardinality, and we believe that this new insight should help one have
a better grasp of the structure of large spaces of nilpotent matrices [10]. On
the other hand, semi-primitive matrix spaces are deeply connected to minimal
locally linearly dependent spaces of operators, and classification theorems for
the former have been recently used to expand our understanding of the latter
[7].
Considering the above, there is a renewed motivation for finding classification
theorems for primitive spaces over small fields. An earlier article of Beasley [3]
contained some information on spaces of rank 2 matrices over F2 but fell short
of giving a complete classification. It is the main purpose of the present work to
achieve that classification.
For (s, t) ∈ [[0, n]] × [[0, p]], we denote by R(s, t) the space of all n × p matrices
of the form
(cid:20) [?]s×t
[?](n−s)×t
[?]s×(p−t)
[0](n−s)×(p−t)(cid:21) .
When we use this notation, the number of rows and columns will always be
obvious from the context. If s + t ≤ min(n, p), then R(s, t) has upper rank s + t.
In particular, if n ≥ 2 and p ≥ 2 the space R(1, 1) has upper rank 2.
It is known that a space with upper rank 1 is either equivalent to a subspace
of R(1, 0) or to a subspace of R(0, 1) (this classical result dates back to Issai
Schur). From there, one can determine the non-primitive reduced subspaces
with upper rank 2. Indeed, let V be such a space. If V is not semi-primitive,
3
then it is equivalent to a subspace V ′ of Mn,p(K) in which every matrix splits
up as M =(cid:2)H(M ) ?(cid:3) and H(V ′) ⊂ Mn,p−1(K) has upper rank 1; Then, V is
equivalent to a subspace of R(1, 1) or to a subspace of R(0, 2), whether H(V ′) is
equivalent to a subspace of R(1, 0) or to one of R(0, 1); In the second case p ≤ 2
as V is reduced, and hence p = 2. If the transpose of V is not semi-primitive,
then either n = 2 or V is equivalent to a subspace of R(1, 1). Conversely, if
p = 2 then V cannot be semi-primitive (just delete the second column), and the
same holds if V is equivalent to a subspace of R(1, 1) (delete the first column
from the matrices of R(1, 1)).
Thus:
Proposition 1.1. Let V be a reduced linear subspace of Mn,p(K) with upper rank
2. Then, V is semi-primitive if and only if p > 2 and V is not equivalent to a
subspace of R(1, 1).
Moreover, V is primitive if and only if n > 2, p > 2 and V is not equivalent to
a subspace of R(1, 1).
Thus, V is semi-primitive if and only if n = 2 or V is primitive.
The classification, up to equivalence, of the reduced linear subspaces of
R(1, 1) is an easy exercise:
Proposition 1.2. Assume that n ≥ 2 and p ≥ 2. For every reduced linear
subspace V of R(1, 1) ⊂ Mn,p(K), there is a unique integer r ∈ [[0, min(n − 1, p −
1)]] such that V is equivalent to one (and only one) of the following spaces:
and
(
(
(a, X, C, L) ∈ K×Kr×Kn−r−1×M1,p−r−1(K))
(X, C, L) ∈ Kr×Kn−r−1×M1,p−r−1(K)).
X T
[0]r×r
a
X
C [0](n−r−1)×r
L
[0]r×(p−r−1)
[0](n−r−1)×(p−r−1)
X T
[0]r×r
0
X
C [0](n−r−1)×r
L
[0]r×(p−r−1)
[0](n−r−1)×(p−r−1)
From that point on, we shall focus on classifying primitive matrix spaces
with upper rank 2 over F2.
In the rest of the article, we consider only the situation of a field K with two
elements, denoted by F2.
4
It is known that for every field K the space A3(K) is primitive with upper
rank 2 (see [2]), and this holds in particular for K = F2. Now, we introduce three
additional examples of primitive spaces with upper rank 2 over F2. To simplify
the discourse, it is convenient to describe such matrix spaces by generic matrices:
Recall that a generic matrix of a linear subspace V of Mn,p(K) is a matrix of the
form x1A1 + · · · + xmAm, where x1, . . . , xm are independent indeterminates and
(A1, . . . , Am) is a basis of V.
Notation 1. We define three linear subspaces of M3(F2) by generic matrices in
the following array:
Space
Generic matrix
J3(F2)
c
a
d
0 a + b e
0
b
0
U3(F2)
0
a
a a + c
0
a + b c
b
0
V3(F2)
0
c
a c + d
0
a + b d
b
0
Alternatively, J3(F2) can be seen as the space of all upper-triangular 3 × 3
matrices with trace zero, whereas V3(F2) can be seen as the space of all matrices
M = (mi,j) ∈ M3(F2) with diagonal zero and m1,2 + m2,3 + m3,1 = m3,2 + m2,1 +
m1,3 = 0. Note that U3(F2) is a linear subspace of V3(F2).
Given three scalars a, b, c in F2 with a + b + c = 0, one of them must be
zero whence abc = 0. Computing the determinant, it is then obvious that every
matrix in J3(F2) is singular, and so is every matrix in V3(F2) (or in U3(F2)).
Now, we state our three main results:
Proposition 1.3. The spaces U3(F2) and V3(F2) are primitive subspaces of
M3(F2) with upper rank 2. Moreover, every non-zero matrix of one of those
spaces has rank 2.
Proposition 1.4. The space J3(F2) is a primitive subspace of M3(F2) with
upper rank 2.
A linear subspace of J3(F2) is primitive with upper rank 2 if and only if, for all
(a, b) ∈ (F2)2, it contains at least one matrix of the form
?
a
?
0 a + b ?
b
0
0
.
Theorem 1.5 (Classification of primitive spaces with upper rank 2 over F2).
Let V be a primitive subspace of Mn,p(F2) with upper rank 2. Then, n = p = 3
and exactly one of the following four conditions holds:
5
(i) V is equivalent to a linear subspace of J3(F2);
(ii) V is equivalent to A3(F2);
(iii) V is equivalent to U3(F2);
(iv) V is equivalent to V3(F2).
In Section 4, we shall also describe, up to equivalence, all the primitive spaces
that are equivalent to a linear subspace of J3(F2).
Remark 1. Note that if V is a primitive subspace of Mn,p(K), then its transpose
is also primitive with the same upper rank. It is obvious that each one of the
spaces A3(F2), U3(F2) and V3(F2) is equal to its transpose. On the other hand,
one sees that J3(F2) is equivalent (and even similar) to its transpose by noting
that J3(F2)T is the space of all lower-triangular matrices of M3(F2) with trace
zero, and hence it equals KJ3(F2)K −1 for the matrix K :=
Let us immediately discuss some corollaries of the above results:
0 0 1
0 1 0
1 0 0
.
Corollary 1.6. Let V be a primitive 4-dimensional subspace of Mn,p(F2). As-
sume that V is a rank -- 2 space, i.e. all its non-zero matrices have rank 2. Then,
V is equivalent to V3(F2).
a rank 1 matrix in V.
To see this, it suffices to show that V cannot be equivalent to a subspace
of J3(F2). This is easily obtained by noting that every 4-dimensional subspace
of J3(F2) is a hyperplane of it, whence it must have a non-zero common vector
with the 2-dimensional subspace of all matrices of the form
(
(a, b, c, d) ∈ F4
2) and (
a
c
d a + b 0
0 c + d b
0 ? ?
0 0 0
0 0 0
, yielding
(a, b, c, d) ∈ F4
2)
In [3], Beasley stated without proof that the two 4-dimensional subspaces
c
a
d a + b
d
d
c
c
b
0
are inequivalent rank-2 spaces. However, although it is true that both are rank-2
spaces, the above corollary shows that they are equivalent as one easily checks
that both are primitive.
6
Note finally that Theorem 1.5 yields a quick proof of a result of [9] on the
classification of subspaces of singular matrices of M3(F2) with dimension at least
5. Indeed, given such a subspace V:
• Either V is non-reduced, and hence it is equivalent to a subspace of R(2, 0)
or to a subspace of R(0, 2).
• Or V is reduced and non-primitive, and hence it is equivalent to a subspace
of R(1, 1); in that case, as dim R(1, 1) = 5, we see that V is equivalent to
R(1, 1) itself.
• Or V is primitive, and hence, as dim V ≥ 5, Theorem 1.5 yields that V
is equivalent to a linear subspace of J3(F2), and hence it is equivalent to
J3(F2) because dim J3(F2) = 5.
The article is laid out as follows: In Section 2, we prove Proposition 1.3. Sec-
tion 3 is devoted to the proof of Theorem 1.5. In Section 4, we prove Proposition
1.4 and we classify all the primitive subspaces of J3(F2). In the last section, we
use our results to classify triples of locally linearly dependent operators over F2
(Section 5.1), to classify the 3-dimensional linear subspaces of M3(F2) in which
no matrix has 1 as eigenvalue (Section 5.2), and to classify the 3-dimensional
affine subspaces of M3(F2) that are included in GL3(F2) (Section 5.3).
2 The structure of U3(F2) and V3(F2)
Remember that V3(F2) is the space of all matrices M ∈ M3(F2) with diagonal
zero and m1,2 + m2,3 + m3,1 = 0 = m3,2 + m2,1 + m1,3, and that U3(F2) is a
hyperplane of V3(F2).
Lemma 2.1. Every non-zero matrix of V3(F2) has rank 2.
Proof. We have already shown that every matrix of V3(F2) is singular. Let
M ∈ V3(F2) be with rk M ≤ 1. As tr M = 0, we deduce that M 2 = 0. As
the diagonal of M is zero, this yields mi,jmj,k = 0 for all distinct i, j, k in
[[1, 3]]. In particular, among m1,2, m2,3, m3,1, at most one entry equals 1, and as
their sum equals zero, we deduce that they are all zero. Similarly, we obtain
m2,1 = m1,3 = m3,2 = 0, whence M = 0.
Lemma 2.2. For all x ∈ (F2)3 r {0}, one has dim U3(F2)x = 2.
7
Proof. Denote by \U3(F2) the space of all linear operators M ∈ U3(F2) 7→ M X ∈
x
y
z
(F2)3, with X ∈ (F2)3. For X =
7→
M =
a a + c
0
c
a + b
0
a
b
0
in (F2)3, the operator M 7→ M X reads
.
×
y + z 0 z
z
0
x y
=
ax + bz
a
b
c
x
x
ax + bx + cy
a(y + z) + cz
Thus, in a well chosen basis of U3(F2) and in the canonical basis of K3, the space
\U3(F2) is represented by the matrix space
(
y + z 0 z
z
0
x y
x
x
(x, y, z) ∈ (F2)3).
By successively applying the row operation L3 ← L3 + L2 and the column
operations C2 ↔ C1 and C3 ↔ C2, this space is seen to be equivalent to U3(F2).
As this space has dimension 3 and every non-zero matrix of U3(F2) has rank 2,
we deduce that dim U3(F2)x = 2 for all non-zero vectors x ∈ (F2)3.
As U3(F2)T = U3(F2), it follows that dim U3(F2)T x = 2 for all non-zero
vectors x ∈ (F2)3, whence U3(F2) is reduced. It also follows from Lemma 2.2
that U3(F2) is not equivalent to a linear subspace of R(1, 1). Therefore, U3(F2)
is primitive and it ensues that V3(F2) is also primitive since it contains U3(F2)
and shares the same upper rank. Thus, Proposition 1.3 is established.
3 Proof of the main classification theorem
This section is devoted to the proof of our main classification theorem, that is
Theorem 1.5. First of all, we shall prove that cases (i) to (iv) are pairwise incom-
patible. Then, we will examine two special cases with n = p = 3. Afterwards,
we will prove that Mn,p(F2) has a primitive subspace with upper rank 2 only
if n = p = 3. Finally, we will classify the primitive subspaces of M3(F2) with
upper rank 2.
8
3.1
Incompatibility between Cases (i) to (iv)
To see that no two cases of Cases (i) to (iv) in Theorem 1.5 can occur simul-
taneously, note that, whenever V falls into one of Cases (ii) to (iv), we have
dim Vx ≥ 2 for every non-zero vector x ∈ (F2)3, which rules Case (i) out.
Case (iv) is incompatible with Cases (ii) and (iii) because dim V3(F2) = 4,
whereas dim U3(F2) = dim A3(F2) = 3.
Finally, Case (iii) is incompatible with Case (ii) because if Case (ii) holds,
for every M ∈ V r {0}, the non-zero vector x of Ker M satisfies Vx = Im M
(indeed, in the special case when V = A3(F2), we have Im M = {x}⊥ = Vx where
⊥ refers to the canonical bilinear form (X, Y ) 7→ X T Y on (F2)3), whereas this
is not always the case for the space U3(F2). Indeed, the matrix M =
belongs to U3(F2), the non-zero vector x =(cid:2)0 1 1(cid:3)T
we have U3(F2)x =(cid:8)(cid:2)a b a(cid:3)T (a, b) ∈ (F2)2(cid:9), which is obviously unequal to
0 1 1
1 0 0
1 0 0
belongs to its kernel, but
Im M .
3.2 Two basic lemmas
Lemma 3.1. Let V be a primitive subspace of M3(F2) with upper rank 2. Assume
that there is a non-zero vector x ∈ (F2)3 such that dim Vx ≤ 1. Then V is
equivalent to a subspace of J3(F2).
Proof. Without loss of generality, we may assume that every matrix M of V
splits up as
M =(cid:20)a(M )
[0]2×1 K(M )(cid:21) with a(M ) ∈ F2 and K(M ) ∈ M2(F2).
[?]1×2
As every matrix of V is singular, a(M ) = 0 whenever K(M ) is non-singular.
Assume that K(V) is inequivalent to a subspace of T +
2 (F2). In particular,
K(V) must contain a non-singular matrix (by the classification of spaces with
upper rank 1). Then, we have some M0 ∈ V such that K(M0) is non-singular and
hence a(M0) = 0. For all M ∈ V, if K(M ) = 0 then K(M + M0) is non-singular
and hence a(M + M0) = 0, which yields a(M ) = 0. It follows that there is a
linear form ϕ : K(V) → F2 such that a(M ) = ϕ(K(M )) for all M ∈ V. As a 6= 0
(because V is reduced), we see that K(V) cannot be spanned by its non-singular
If K(V) were a hyperplane of M2(F2), then it would be equivalent
matrices.
9
to T +
2 (F2) or to S2(F2), whether its orthogonal subspace for (A, B) 7→ tr(AB)
contained a rank 1 matrix or not. However, S2(F2) is spanned by its non-
singular elements, and so does M2(F2), whence dim K(V) ≤ 2. Moreover, as
K(V) is inequivalent to a subspace of T +
2 (F2), we have K(V)y = (F2)2 for all
non-zero vectors y ∈ (F2)2; It ensues that dim K(V) = 2 and that all the non-
zero matrices of K(V) are non-singular, contradicting the fact that K(V) is not
spanned by its non-singular matrices.
Thus, K(V) is actually equivalent to a subspace of T +
generality is lost in assuming that V is actually a linear subspace of T +
2 (F2). Therefore, no
3 (F2).
For M ∈ V, denote by δ(M ) = (cid:2)m1,1 m2,2 m3,3(cid:3)T ∈ (F2)3 its diagonal vec-
(cid:2)1 1 1(cid:3)T
(cid:2)1 1 1(cid:3)T
tor. Then, δ(V) is a linear subspace of (F2)3 that does not contain the vector
(since no matrix of V is invertible). Thus, δ(V) is included in a hy-
perplane with the same property. Moreover, since V is primitive, δ(V) is included
in none of the three canonical hyperplanes (with equations x1 = 0, x2 = 0 and
x3 = 0, respectively). The only remaining hyperplane which does not contain
is the one defined by the equation x1 + x2 + x3 = 0, whence every
matrix of V has trace 0. We conclude that V is a linear subspace of J3(F2).
Lemma 3.2. Let V be a primitive subspace of M3(F2) with upper rank 2. Assume
that V contains a rank 1 matrix. Then, V is equivalent to a subspace of J3(F2).
Proof. Without loss of generality, we may assume that V contains the elementary
matrix E1,1 (with entry 1 at the (1, 1)-spot, and zero entries everywhere else).
Then, we split every M ∈ V as
M =(cid:20)a(M ) R(M )
S(M ) K(M )(cid:21) with a(M ) ∈ F2 and K(M ) ∈ M2(F2).
We contend that every matrix of K(V) is singular. Indeed, if we let M ∈ V,
then both matrices M and M + E1,1 belong to V, and therefore
0 = det(M + E1,1) − det M = det K(M ).
It follows that urk K(V) ≤ 1. Then, there are two cases to consider:
• Either there is a non-zero vector of (F2)2 on which all the matrices of
K(V) vanish; in this case we find a non-zero vector x of (F2)3 for which
dim Vx ≤ 1, and Lemma 3.1 shows that V is equivalent to a linear subspace
of J3(F2).
10
• Or the non-zero matrices of K(V) have the same range, whence there is a
non-zero vector x ∈ (F2)3 for which dim V T x ≤ 1. As V T is primitive with
upper rank 2, we deduce from Lemma 3.1 that it is equivalent to a linear
subspace of J3(F2). However, we have seen in Remark 1 that J3(F2)T is
equivalent to J3(F2), whence V is equivalent to a linear subspace of J3(F2).
3.3 Basic identities
In the rest of the proof, we let V be a primitive subspace of Mn,p(F2) with upper
rank 2. Note that n ≥ 3 and p ≥ 3.
As V contains a rank 2 matrix and as such a matrix is equivalent to
J2 :=(cid:20)
I2
[0](n−2)×2
[0]2×(p−2)
[0](n−2)×(p−2)(cid:21) ,
we lose no generality in assuming that V contains J2.
We split every matrix M of V up as
M =(cid:20)A(M ) C(M )
B(M ) D(M )(cid:21)
along the same pattern as J2.
Let i ∈ [[3, n]] and j ∈ [[3, p]]. The 3 by 3 sub-matrix of M obtained by
selecting row indices in {1, 2, i} and column indices in {1, 2, j} is singular since
rk M ≤ 2, and on the other hand its determinant reads
(cid:0)det A(M )(cid:1) D(M )i−2,j−2 − B(M )i−2
^A(M ) C(M )j−2,
where eN denotes the transpose of the comatrix of the square matrix N , and
B(M )i−2 and C(M )j−2 respectively denote the (i − 2)-th row of B(M ) and the
(j − 2)-th column of C(M ). Varying i and j then yields the matrix identity
(cid:0)det A(M )(cid:1)D(M ) = B(M ) ^A(M ) C(M ),
Note that N 7→ eN is linear on M2(F2). Moreover
∀M ∈ V, D(M ) = 0 ⇒ B(M ) C(M ) = 0.
(1)
(2)
To see this, it suffices to apply identity (1) to both matrices M and M + J2.
11
3.4 The proof that n = p = 3
Now, we prove the following result:
Proposition 3.3. Let V be a primitive subspace of Mn,p(F2) with upper rank 2.
Then, n = p = 3.
The proof has several steps. First of all, we lose no generality in assuming
that V contains J2, as in the preceding section. Our first step establishes an
important relationship between the matrices B(M ) and D(M ), for M in V:
Step 1. Let M ∈ V. Denote by B1(M ) and B2(M ) the columns of B(M ).
Then,
rk(cid:2)B1(M ) D(M )(cid:3) ≤ 1 and
rk(cid:2)B2(M ) D(M )(cid:3) ≤ 1.
Proof. Take two distinct indices i1 and i2 in [[3, n]], two distinct indices j1 and j2
in [[2, p]], and denote by ∆(M ) the 3 × 3 sub-matrix of M obtained by selecting
the row indices in {1, i1, i2} and the column indices in {1, j1, j2}. Then, we see
that det ∆(M + J2) − det ∆(M ) is the determinant of the 2 × 2 submatrix of
M obtained by selecting row indices in {i1, i2} and column indices in {j1, j2}.
As det ∆(M + J2) = 0 = det ∆(M ), we deduce that rk(cid:2)B2(M ) D(M )(cid:3) ≤ 1 by
varying i1, i2, j1, j2. The first inequality is proved in a similar fashion.
As an immediate corollary, we deduce:
Step 2. The upper rank of D(V) is less than or equal to 1.
It follows that either all the non-zero matrices of D(V) have the same kernel,
or all of them have the same range.
Step 3. Assume that D(V) 6= {0}. If all the non-zero matrices of D(V) have
the same range (respectively, the same kernel), then n = 3 (respectively, p = 3).
the same kernel, then this kernel cannot be {0} × (F2)p−2 as D(V) 6= {0};
then, as this kernel must have dimension p − 2, it must contain a vector of
Proof. Note that if all the non-zero matrices of the form(cid:2)B1(M ) D(M )(cid:3) have
(F2)p−1 r(cid:0){0} × (F2)p−2(cid:1), which yields a column matrix X ∈ Kp−2 such that
by D. By Step 1, if all the non-zero matrices of the form(cid:2)B1(M ) D(M )(cid:3) did
not have the same range, then they would all have the same kernel -- owing the
Assume that all the non-zero matrices of D(V) have the same range, denoted
B1(M ) = D(M )X for all M ∈ V.
12
classification of matrix spaces with upper rank at most 1 -- and hence the above
remark shows that B1(M ) ∈ Im D(M ) ⊂ D for all M ∈ V. If all those matrices
have the same range, it must be D because D(V) 6= {0}. In any case, we obtain
B1(M ) ∈ D for all M ∈ V. Similarly, one obtains B2(M ) ∈ D for all M ∈ V. As
V is reduced, we deduce that n = 3.
Using V T instead of V, we deduce that p = 3 if all the non-zero matrices of
D(V) have the same kernel.
Step 4. One has dim D(V) ≤ 1.
Proof. Assume that dim D(V) > 1. Assume also that all the non-zero matrices
of D(V) have the same kernel. Then, p = 3. Moreover, by Step 3, all the non-
zero matrices of the form(cid:2)B2(M ) D(M )(cid:3) cannot have the same range, which,
by Step 1, yields a scalar µ such that B2(M ) = µ D(M ) for all M ∈ V. Similarly,
one finds λ ∈ F2 for which B1(M ) = λ D(M ) for all M ∈ V. Performing the
column operations C1 ← C1 − λC3 and C2 ← C2 − µC3 changes none of the
above assumptions and reduces the situation to the one where B(M ) = 0 for all
M ∈ V. Note that every matrix M of V then splits up as
M =(cid:20) A(M )
[0](n−2)×2 D(M )(cid:21) ,
[?]2×1
whence rk A(M ) = 2 ⇒ D(M ) = 0.
Let M ∈ V be such that A(M ) = 0. Then, A(M + J2) = I2, whence
0 = D(M + J2) = D(M ). This yields a linear map ϕ : A(V) → (F2)n−2 such
that D(M ) = ϕ(A(M )) for all M ∈ V, and ϕ vanishes at every rank 2 matrix
of A(V). Note that dim Ker ϕ ≥ 1 and rk ϕ ≥ 2, whence dim A(V) ≥ 3.
If
dim A(V) = 4, then A(V) = M2(F2) is spanned by its rank 2 elements, which
leads to ϕ = 0. Thus, dim A(V) = 3, rk ϕ = 2 and dim Ker ϕ = 1. But again, we
find a contradiction by noting that every linear hyperplane of M2(F2) contains
several rank 2 matrices (this is obvious as such a hyperplane must be equivalent
to S2(F2) or to T +
2 (F2), as we have already explained in the course of the proof
of Lemma 3.1). Therefore, the non-zero matrices of D(V) cannot share the same
kernel.
Similarly, by working with V T , we see that the non-zero matrices of D(V)
cannot share the same range. Therefore, we have contradicted the fact that
D(V) has upper rank 1.
From there, we can complete our proof of Proposition 3.3:
13
Step 5. One has n = p = 3.
Proof. If D(V) 6= {0}, then dim D(V) = 1, whence all the non-zero matrices of
D(V) have the same range and the same kernel (there is only one such matrix!),
and Step 3 yields n = p = 3.
In the rest of the proof, we assume that D(V) = {0}. As V is reduced, we
have B(V) 6= {0} and C(V) 6= {0}. For all M ∈ V, we know from identity (2)
that B(M )C(M ) = 0. In particular B(M ) = 0 whenever rk C(M ) = 2. Assume
that some matrix M0 is such that rk C(M0) = 2. Then, B(M0) = 0. For every
M ∈ V, we find B(M )C(M + M0) = B(M + M0)C(M + M0) = 0, whence
B(M )C(M0) = B(M )C(M + M0) − B(M )C(M ) = 0, which leads to B(M ) = 0,
contradicting our assumptions. Thus, urk C(V) = 1, and similarly urk B(V) = 1.
If all the matrices of C(V) have the same kernel, we obtain that p = 3
since V is reduced. Assume now that all the non-zero matrices of C(V) have
the same range. Without loss of generality, we can assume that this range is
F2 × {0}. Note that dim C(V) = p − 2 because of condition (i) in the definition
of a primitive space. Then, for all M ∈ V, we write C(M ) =(cid:20) L(M )
[0]1×(p−2)(cid:21). If we
let M ∈ V, then identity (2) yields B1(M )L(M ) = 0, whence either B1(M ) = 0
or L(M ) = 0. As V is not the union of two of its proper linear subspaces and
as L(V) 6= 0, we deduce that B1(V) = {0}. As B(V) 6= {0} and V is reduced,
we deduce that B2(V) = (F2)n−2. Now, denote by α(M ) the entry of M ∈ V
at the (2, 1)-spot. If α = 0, then we contradict condition (iii) in the definition
of a primitive space (by deleting the second column). Thus, α 6= 0, B2 6= 0 and
L 6= 0. Fix M ∈ V, and note that
M =
?
α(M )
?
?
[0](n−2)×1 B2(M )
L(M )
[0]1×(p−2)
[0](n−2)×(p−2)
.
As rk M ≤ 2, one of the matrices B2(M ), L(M ) or α(M ) must be zero. However,
the linear maps B2, L and α on V are all non-zero, and we have just shown that
V is the union of their respective kernels. If p > 3, then Ker L has codimension
at least 2 in V, whence Lemma 2.5 of [8] yields a contradiction. Therefore, p = 3.
By applying the above line of reasoning to V T , we obtain n = 3.
This completes the proof of Proposition 3.3.
14
3.5 Completing the classification
Let V be a primitive subspace of Mn,p(F2) with upper rank 2. By Proposition
3.3, we know that V is actually a linear subspace of M3(F2). Moreover, we can
assume that V contains J2, and we keep the notation from Section 3.3. We also
make the following additional assumption:
(H1) V is inequivalent to a linear subspace of J3(F2).
From there, our aim is to prove that V is equivalent to A3(F2), U3(F2) or V3(F2).
Using (H1), we see from Lemmas 3.1 and 3.2 that every non-zero matrix
of V has rank 2, and dim Vx ≥ 2 for all x ∈ (F2)3 r {0}. By Remark 1, V T
is also inequivalent to a subspace of J3(F2), and hence dim V T x ≥ 2 for all
x ∈ (F2)3 r {0}.
Claim 1. (a) If D(V) = {0}, then dim V = 3.
(b) If, for every non-zero matrix M ∈ V, we have V Ker M = Im M , then V is
equivalent to A3(F2).
Proof. (a) Assume that D(V) = {0}. Denoting by e3 the third vector of the
standard basis of (F2)3, we deduce that Ve3 ⊂ (F2)2 × {0}, whence Ve3 =
(F2)2 × {0} as dim Ve3 ≥ 2. Thus, C(V) = (F2)2. By (2), we have
∀M ∈ V, B(M )C(M ) = 0.
Polarizing this quadratic identity yields B(M )C(N )+B(N )C(M ) = 0 for all
(M, N ) ∈ V 2. It follows that for every M ∈ V such that C(M ) = 0, we have
B(M )C(N ) = 0 for all N ∈ V, which yields B(M ) = 0 since C(V) = (F2)2.
This yields a (non-zero) matrix K ∈ M2(F2) such that B(M ) = C(M )T K
for all M ∈ V. Then, C(M )T KC(M ) = 0 for all M ∈ V, which shows
that K is alternating. Therefore, K =(cid:20)0 1
1 0(cid:21) (the sole non-zero matrix in
A2(F2)).
Let M0 ∈ V be such that C(M0) = 0. Then, B(M0) = 0 and, for all M ∈ V,
we find, by identity (1),
C(M )T K ^A(M0)C(M ) = B(M ) ^A(M0)C(M )
= B(M + M0)
^(cid:0)A(M ) + A(M0)(cid:1)C(M + M0) − B(M )^A(M )C(M ) = 0,
15
whence K ^A(M0) is alternating. Thus, K ^A(M0) ∈ F2 K, and hence A(M0) ∈
{0, I2}. Noting that C(J2) = 0, we deduce that Ker C = F2J2, whence
dim V = 3.
(b) Assume that for every non-zero matrix M ∈ V, we have V Ker M = Im M .
In particular, the case M = J2 yields D(V) = {0}, whence the above proof
shows that C(V) = (F2)2 and B(M ) = C(M )T K for all M ∈ V. We choose
M1 ∈ V with C(M1) =(cid:20)1
0(cid:21), so that
M1 =
?
? 1
?
? 0
0 1 0
Replacing M1 with M1 + J2 if necessary, we can assume that the entry of
M1 at the (1, 1)-spot is 0. As M1 is singular, its entry at the (2, 1)-spot is
0. Using row operations of the form L1 ← L1 − λL3 and L2 ← L2 − µL3,
we see that no generality is lost in assuming that
.
.
.
M1 =
M2 =
0 0 1
0 0 0
0 1 0
0 0 0
a b 1
1 0 0
With a similar line of reasoning, we find scalars a and b such that V contains
a matrix of the form
As rk(M1 + M2) ≤ 2, one finds a = b by computing the determinant. As
e1 ∈ Ker M1, we must have M2e1 ∈ Im M1, whence a = 0. We conclude that
a = b = 0, and hence, as dim V = 3, we have V = span(J2, M1, M2), i.e. V is
associated with the generic matrix
a 0 b
0 a c
c b 0
.
Swapping the first two rows finally shows that V is equivalent to A3(F2).
16
Claim 2. One has 3 ≤ dim V ≤ 4 and there is at least one non-zero vector
x ∈ (F2)3 for which dim Vx = 2. Moreover, if dim V = 3, then dim Vx =
dim V T x = 2 for all non-zero vectors x ∈ (F2)3.
Proof. Set d := dim V. We use a counting argument: Denote by N the set of all
pairs (M, x) ∈(cid:0)V r {0}(cid:1) ×(cid:0)(F2)3 r {0}(cid:1) for which M x = 0. Remember that
dim Vx ∈ {2, 3} for all non-zero vectors x ∈ (F2)3. For i ∈ {2, 3}, denote by
ni the number of non-zero vectors x ∈ (F2)3 for which dim Vx = i. For every
non-zero vector x ∈ (F2)3, the set of all matrices M ∈ V for which M x = 0 is
the kernel of M 7→ M x, and hence it has dimension d − dim Vx. Thus,
#N = (2d−2 − 1) n2 + (2d−3 − 1) n3.
On the other hand, every non-zero matrix of V has rank 2 and hence it annihilates
exactly one non-zero vector of (F2)3. Therefore,
#N = 2d − 1.
As n3 = 7 − n2, we deduce that 2d−3n2 = 2d − 7 × 2d−3 + 6, which leads to
n2 = 1 + 3 × 24−d.
In particular, we deduce that n2 > 0. As n2 must be an integer, we find 4−d ≥ 0.
As n2 ≤ 7, we also find d ≥ 3. Thus, d ∈ {3, 4}. Finally, if d = 3, then n2 = 7
whence dim Vx = 2 for every non-zero vector x ∈ (F2)3; V T must satisfy the
same conclusion as it has dimension 3.
Now, we make an additional assumption:
(H2) V is inequivalent to A3(F2).
We shall conclude by distinguishing between two cases, whether V has di-
mension 3 or 4.
Claim 3. Assume that dim V = 3. Then, V is equivalent to U3(F2).
Proof. If there are two distinct matrices of V with the same (two-dimensional)
range, then, by choosing a non-zero vector x in the orthogonal complement
of this range, we would find dim V T x ≤ 1, contradicting Claim 2. Thus, two
distinct matrices of V cannot have the same range.
17
Combining point (b) of Claim 1 with assumption (H2), we find a matrix
M ∈ V such that V Ker M 6= Im M . We lose no generality in assuming that
M = J2. As dim Ve3 = 2, no further generality is lost in assuming that the
space of all third columns of the matrices of V is F2 × {0} × F2. This yields two
matrices in V of the following forms
M1 =
? ? 1
? 0 0
? ? 0
and M2 =
0 ? 0
? ? 0
? ? 1
,
as we may add J2 if necessary. Note that J2, M1, M2 are obviously linearly
independent whence V = span(J2, M1, M2). By identity (2), the entry of M1
at the (3, 1)-spot must be zero. As M1 6= J2 and rk M1 = rk J2 = 2, we must
have Im M1 6⊂ Im J2, and hence the entry of M1 at the (3, 2)-spot is non-zero.
It follows that
As rk M1 = 2, we deduce that
Performing column operations of the forms C1 ← C1 − λC3 and C2 ← C2 − µC3,
we see that no generality is lost in assuming that
As dim Ve2 = 2 and V contains J2 and M1, the entry of M2 at the (1, 2)-spot is
zero, whence
? ? 1
? 0 0
0 1 0
? ? 1
0 0 0
0 1 0
.
.
.
.
M1 =
M1 =
M1 =
M2 =
for some (a, b, c) ∈ (F2)3.
0 0 0
? ? 0
? ? 1
0 0 1
0 0 0
0 1 0
18
As J2 + M2 is singular, we deduce that
M2 =
0 0 0
a 1 0
c 1
b
As M1 + M2 is singular, we find a(c + 1) = b by computing the determinant.
Using the singularity of M1 + M2 + J2, we obtain a(c + 1) = 0. Thus, b = 0.
However, as dim(Ve1) = 2 and V = span(J2, M1, M2), we cannot have a = 0,
whence a = 1 and c = 1. Thus,
We deduce that V is associated with the generic matrix
0 0 0
1 1 0
0 1 1
M2 =
0
b
a
c a + c
0
0 b + c c
.
.
Using the operations L1 ↔ L3 and C2 ↔ C3, we obtain that V is equivalent to
U3(F2).
Claim 4. Assume that dim V = 4. Then, V is equivalent to V3(F2).
Proof. By Claim 2, we can choose a vector x ∈ (F2)3 for which dim Vx = 2.
Then, there is a non-zero matrix of V which annihilates x, and this matrix has
rank 2. Without loss of generality, we may assume that this matrix is J2, in
which case x is the third vector of the standard basis. Point (a) of Claim 1
yields that Vx 6= Im J2; Thus, we can choose a basis (y, y′) of Vx such that
y ∈ Im J2 and y′ 6∈ Im J2, then we choose x1 ∈ (F2)3 such that J2x1 = y, and
we extend (x1, x) into a basis (x1, x2, x) of (F2)3. Then, by replacing V with
an equivalent subspace -- so that our new source basis is (x1, x2, x) and our new
target basis is (y, J2x2, y′) -- we see that no further generality is lost in assuming
that Vx = F2 × {0} × F2. Since dim V = 4, we may extend J2 into a basis
(J2, M1, M2, M3) of V with
Adding J2 to M1 and M2 if necessary, we may assume that
M1 =
? ? 0
? ? 0
? ? 0
, M2 =
M1 =
0 ? 0
? ? 0
? ? 0
? ? 1
? ? 0
? ? 0
and M3 =
.
? ? 1
? 0 0
? ? 0
and M2 =
? ? 0
? ? 0
? ? 1
.
19
Applying identity (2) to M2 and M1 + M2, we find that
M1 =
0 ? 0
? ? 0
0 ? 0
, M2 =
? ? 1
? 0 0
0 ? 0
.
Assume first that B(M1) = 0. As dim(V T e3) ≥ 2, we must have B(M2) 6= 0,
as M1 + M2 is singular, we also obtain (M1)2,1 = 0. Thus, the first and third
columns of M1 equal zero, whence M1 has rank 1, which is absurd.
whence B(M2) =(cid:2)0 1(cid:3). Then, as M2 is singular, we find that (M2)2,1 = 0 and,
We deduce that B(M1) =(cid:2)0 1(cid:3). Then, as we lose no generality in replacing
M2 with a matrix of the form M2 + a M1 + b J2, we can assume that B(M2) = 0.
From there, using column operations of the form C1 ← C1 + λC3 and C2 ←
C2 + µC3, we see that no generality is lost in assuming that
As rk M2 > 1, we deduce that
M2 =
M2 =
0 0 1
? 0 0
0 0 0
0 0 1
1 0 0
0 0 0
.
.
If the entry of M1 at the (1, 2)-spot equals 1, then we perform the row operation
L1 ← L1 + L3 and then we replace M3 with M3 + M2. This shows that no
generality is lost in assuming that
M1 =
0 0 0
? ? 0
0 1 0
.
As M1 has rank 2, we find
M1 =
0 0 0
1 a 0
0 1 0
for some a ∈ F2.
20
From there, we lose no generality in adding a linear combination of M1 and J2
to M3, whence we may assume that
M3 =
0 c 0
0 d 0
b
e 1
for some (b, c, d, e) ∈ (F2)4.
As det(J2 + M3) = 0, det(M1 + M3) = 0 and det(J2 + M1 + M3) = 0, we find
d = 1, c = 0 and a = 0, successively. Finally, using det(M2 + M3) = 0 and
det(J2 + M2 + M3) = 0, we find b = e and e = 0. Thus, V is the span of the
matrices
J2 =
1 0 0
0 1 0
0 0 0
, M1 =
0 0 0
1 0 0
0 1 0
, M2 =
0 0 1
1 0 0
0 0 0
and M3 =
0 0 0
0 1 0
0 0 1
,
whence it is associated with the generic matrix
0
a
c
b + c a + d 0
d
b
0
.
Using the column operations C2 ↔ C1 and C3 ↔ C2, we conclude that V is
equivalent to V3(F2), as claimed.
This completes the proof of Theorem 1.5.
3.6 Application to maximal spaces of matrices with upper rank
2
Notation 2. Given integers n′ ∈ [[0, n]] and p′ ∈ [[0, p]] together with a subspace
W of Mn′,p′(K), we denote by fW (n,p) the space of all n × p matrices of the form
[0]n′×(p−p′)
(cid:20) M
[0](n−n′)×p′
[0](n−n′)×(p−p′)(cid:21) with M ∈ W.
Note that fW (n,p) has the same upper rank as W.
Let S be a linear subspace of L(U, V ), where U and V are finite-dimensional
vector spaces. We define the kernel and the range of S as, respectively, Ker S :=
21
Tf ∈S
Ker f and Im S := Pf ∈S
Im f . Then, every operator f ∈ S induces a linear
operator f : U/ Ker S → Im S, and one sees that the operator space S := {f
f ∈ S} is a reduced linear subspace with the same dimension and the same
upper rank as S: It is called the reduced operator space of S. Finally, two
operator subspaces S and T of L(U, V ) are equivalent if and only if dim Ker S =
dim Ker T , dim Im S = dim Im T and the operator spaces S and T are equivalent.
In terms of matrices, this reads as follows:
Proposition 3.4. Let V be a linear subspace of Mn,p(K) with upper rank r.
Then, there is a pair (n′, p′) ∈ [[0, n]] × [[0, p]] and a reduced linear subspace V ′ of
. The pair (n′, p′) is uniquely determined by V,
(n,p)
and the equivalence class of V ′ is uniquely determined by that of V.
Mn′,p′(K) such that V ∼ eV ′
Note that V ′ represents the reduced operator space of V (seen as a space
of linear maps from Kp to Kn).
If V is equivalent to a subspace of R(2, 0)
(respectively, of R(0, 2)), then n′ ≤ 2 (respectively, p′ ≤ 2). Moreover, if V ′ is
equivalent to a subspace of R(1, 1), then V is equivalent to a subspace of R(1, 1).
Conversely, assuming that V ⊂ R(1, 1), then we have a hyperplane H of Kp and
a 1-dimensional subspace D of Kn such that Vx ⊂ D for all x ∈ H. Then, for
H ′ := (H + Ker V)/ Ker V and D′ := D ∩ Im V, we see that f (x) ∈ D for all
x ∈ H ′ and all f ∈ V, and H ′ has codimension at most 1 in Kp/ Ker V whereas D′
has dimension at most 1. It follows that V ′ is equivalent to a subspace of R(1, 1).
From the above considerations combined with Proposition 1.1 and Theorem 1.5,
we deduce the following structure theorem on subspaces of matrices of M3(F2)
with rank at most 2:
Theorem 3.5. Let V be an upper rank 2 subspace of Mn,p(F2), with n ≥ 3 and
p ≥ 3. Then, one and only one of the following cases holds:
(i) V is equivalent to a subspace of R(2, 0);
(ii) V is equivalent to a subspace of R(0, 2);
(iii) V is equivalent to a subspace of R(1, 1);
(iv) V is equivalent to fW (n,p), where W is a primitive linear subspace of J3(F2);
(v) V is equivalent to ^A3(F2)
(n,p)
;
22
(vi) V is equivalent to ^U3(F2)
(n,p)
(vii) V is equivalent to ^V3(F2)
(n,p)
;
.
Moreover, if case (iv) holds, then the equivalence class of W is uniquely deter-
mined by that of V.
As a consequence, we get:
Theorem 3.6 (Classification of maximal spaces of matrices with rank at most
2). Let n > 2 and p > 2. Up to equivalence, there are 6 maximal subspaces of
upper rank 2 matrices of Mn,p(F2):
R(2, 0), R(0, 2), R(1, 1), ^J3(F2)
(n,p)
, ^A3(F2)
(n,p)
and ^V3(F2)
(n,p)
.
The only non-trivial point in the derivation of that theorem from Theorem
3.5 and Proposition 3.4 is to see that A3(F2) is maximal among the subspaces
of M3(F2) with upper rank 2. This is obtained as a special case of the following
general result:
Proposition 3.7. Let n be an odd integer and F be an arbitrary field. Then,
An(F) is a maximal subspace of singular matrices of Mn(F).
Proof. For the case when #F > 2, we refer to [5, Proposition 5]. Thus, we
shall only consider the case when F = F2. We note that the problem is tightly
connected to the representation of quadratic forms. We refer to [12, Chapter
XXXII] for the basics on quadratic forms over fields of characteristic 2. Let
P ∈ Mn(F2)rAn(F2). We have to show that P +An(F2) contains a non-singular
matrix. We consider the non-zero quadratic form q : X ∈ Fn
2 7→ X T P X. The
set of matrices Q ∈ Mn(F2) that represent q, i.e. such that, in some basis of
(F2)n, the map X 7→ X T QX corresponds to q, is precisely Cong(P ) + An(F2),
where Cong(P ) denotes the congruence class of P , that is the set of all matrices
RQRT with R ∈ GLn(F2). As An(F2) is invariant under congruence, it suffices
to find a non-singular matrix which represents q. The rank of the polar form
of q equals 2r for some non-negative integer r. As n is odd, the radical of q
is odd-dimensional, and hence non-zero. The restriction of q to its radical is a
linear form. We shall now distinguish between two cases, whether this linear
form is zero or not. For (a, b) ∈ (F2)2, we denote by [a, b] the quadratic form
23
(x, y) 7→ ax2 + xy + by2 on (F2)2, and by hai the quadratic form x 7→ ax2. The
orthogonal direct sum of two quadratic forms q1 and q2 is denoted by q1⊥q2.
Case 1. The restriction of q to its radical R is non-zero.
Then, we may choose a basis of R in which no vector is q-isotropic (indeed,
the set of q-isotropic vectors in R is a linear hyperplane of R, and hence its
complementary set in R spans R). This yields pairs (a1, b1), . . . , (ar, br) in (F2)2
such that q is equivalent to [a1, b1]⊥ · · · ⊥[ar, br]⊥h1i⊥ · · · ⊥h1i. However, the
contamination lemma [12, Chapter XXXII, Lemma 5.4.2] shows that, for all
(a, b) ∈ (F2)2,
[a, b]⊥h1i ≃ [a + 1, b]⊥h1i ≃ [a, b + 1]⊥h1i ≃ [a + 1, b + 1]⊥h1i.
Using this repeatedly, we deduce that q is equivalent to r.[1, 1]⊥(n − 2r).h1i,
whence the invertible matrix(cid:20)Ir
0
Ir
Ir(cid:21) ⊕ In−2r represents q.
Is
Is
Case 2. The restriction of q to its radical is zero.
Then, r ≥ 1 as q is non-zero. Using the equivalence [1, 1]⊥[1, 1] ≃ [0, 0]⊥[0, 0]
(see [12, Chapter XXXII, Example 4.2.3]), we find that q is equivalent to (r −
1).[1, 1]⊥(n − 2r − 1).h0i⊥ϕ, where ϕ equals either [0, 0]⊥h0i or [1, 1]⊥h0i. As
n is odd, we have n − 2r − 1 = 2s for some non-negative integer s, whence (r −
Ir−1(cid:21)⊕
1).[1, 1]⊥(n−2r−1).h0i is represented by the non-singular matrix(cid:20)Ir−1
(cid:20) 0
0(cid:21). Therefore, it only remains to prove that ϕ is represented by at least
non-singular matrix
. If ϕ equals [1, 1]⊥h0i, then it is represented by
the non-singular matrix
. In any case, the conclusion follows that q
If ϕ equals [0, 0]⊥h0i, then it is represented by the
is represented by at least one non-singular matrix.
1 1 0
0 1 1
0 1 0
0 1 1
0 0 1
1 1 0
one non-singular matrix.
Ir−1
0
4 Primitive linear subspaces of J3(F2)
4.1 A rough result on the primitive subspaces of J3(F2)
Let us prove Proposition 1.4. Let V be a linear subspace of J3(F2), and consider
the space D ⊂ (F2)3 of all diagonal vectors in V, that is the space of all vectors
24
with M ∈ V. We know that D is included in the hyperplane
(cid:2)m1,1 m2,2 m3,3(cid:3)T
c(cid:3)T
write(cid:2)a b
H = {(a, b, c) ∈ (F2)3 : a + b + c = 0}.
If D = {0}, then it is obvious that V is not reduced, whence it is not primitive.
Assume now that dim D = 1. Then, D contains a sole non-zero vector which we
. As D is reduced, we must have a = 1 and c = 1. Thus, b = 0
and, by swapping the first and third columns, we see that V is equivalent to a
linear subspace of R(1, 1).
Now, we assume that D = H and we prove that V is primitive. Note that V T
is equivalent to a linear subspace of J3(F2) for which the space of all diagonal
vectors is H. Now, let X = (cid:2)x y z(cid:3)T ∈ (F2)3 be such that M X = 0 for all
M ∈ V. Looking at the third entry of the matrix M X, we deduce that z = 0.
Then, we successively find y = 0 and x = 0 by looking at the second entry of
M X and then at the first one. Thus, V satisfies condition (i) in the definition
of a primitive space. For the same reason V T also does, whence V is reduced.
Denote by (e1, e2, e3) the standard basis of (F2)3. Let x ∈ (F2)3 r F2e1. We
contend that Vx 6= F2e1. Indeed, if the third entry of x equals 1 then, as we know
that some matrix of V has entry 1 at the (3, 3)-spot, we see that Vx 6= F2e1;
Otherwise, the second entry of x equals 1 and as some matrix of V has entry 1
at the (2, 2)-spot we obtain that Vx 6= F2e1.
Now, assume that V is non-primitive. Then, as it is reduced, it must be
equivalent to a subspace of R(1, 1), which yields a 2-dimensional subspace P of
(F2)3 and a 1-dimensional subspace D of (F2)3 such that Vx ⊂ D for all x ∈ P .
As P 6⊂ F2e1, the above proof yields that D 6= F2e1 whence e1 6∈ P . Therefore,
(F2)3 = F2e1 ⊕ P , which yields Vx ∈ D + F2e1 for all x ∈ (F2)3, contradicting
the fact that V is reduced.
We conclude that V is primitive, which finishes the proof of Proposition 1.4.
4.2 The full classification of primitive subspaces of J3(F2)
Now, we shall give a full classification, up to equivalence, of the primitive sub-
spaces of J3(F2). Of course, we have just seen that J3(F2) is primitive, whence
it only remains to classify its primitive subspaces with dimension 2, 3 or 4 (ob-
viously, a subspace of M3(F2) with upper rank 2 and dimension at most 1 is
non-reduced). This is given in the next three propositions:
Proposition 4.1. Let V be a primitive subspace of J3(F2) with dimension 2.
25
Then, V is equivalent to the space associated with the generic matrix
0
0
a
0 a + b 0
0
b
0
.
Proposition 4.2. Let V be a primitive subspace of J3(F2) with dimension 3.
Then, V is equivalent to one and only one of the four spaces associated with the
generic matrices
M1 :=
M3 :=
0
a
c
0 a + b 0
0
b
0
b
0
a
0 a + b c
0
b
0
, M2 :=
and M4 :=
c
0
a
0 a + b a
0
b
0
c
0
a
0 a + b c
0
b
0
Proposition 4.3. Let V be a primitive subspace of J3(F2) with dimension 4.
Then, V is equivalent to one and only one of the four spaces associated with the
generic matrices
,
.
,
.
c
a
d
0 a + b 0
0
b
0
c
a
d
0 a + b c
0
b
0
N1 :=
N3 :=
c
0
a
0 a + b d
0
b
0
0
a
c
0 a + b d
0
b
0
, N2 :=
and N4 :=
Remark 2. In the prospect of the proofs of Propositions 4.2 and 4.3, the following
remark will be useful: the set of all matrices with rank at most 1 in J3(F2) is
the union of the 2-dimensional subspaces
P1 :=(
0 a b
0 0 0
0 0 0
(a, b) ∈ (F2)2) and P2 :=(
0 0 b
0 0 a
0 0 0
(a, b) ∈ (F2)2).
Indeed, if a matrix M ∈ J3(F2) has its diagonal non-zero, then exactly two of its
diagonal entries equal 1, whence it has rank 2. Thus, a rank 1 matrix of J3(F2)
must have its diagonal zero: From there, the claimed result is obvious.
26
Proof of Proposition 4.2. Using Proposition 1.4 together with the rank theorem,
we see that V contains exactly one non-zero matrix M0 with diagonal zero. We
split the discussion into four cases, according to the value of M0.
Case 1. M0 =
0 0 1
0 0 0
0 0 0
Then, we find scalars α, β, γ, δ such that a generic matrix of V is
a αa + βb
0
0
a + b
0
c
γa + δb
b
.
Performing the operations C3 ← C3 + γC2, L2 ← L2 + (δ + γ)L3, L1 ← L1 + βL2
and C2 ← C2 + (α + β)C1, we reduce the situation to the one where α = β =
γ = δ = 0, and hence V is equivalent to the space associated with M1.
.
.
Then, by performing the column operation C3 ← C3 + C2 if necessary, we see
0 1 0
0 0 0
0 0 0
, whence we find
0 1 ?
0 0 0
0 0 0
Case 2. M0 =
that no generality is lost in assuming that M0 =
.
γa + δb
a
0 a + b αa + βb
0
scalars α, β, γ, δ such that a generic matrix of V is
b
0
c
Using the operations C3 ← C3 + γC1, L1 ← L1 + δL3 and L2 ← L2 + βL3,
we reduce the situation to the one where γ = δ = β = 0. If α = 1, then V is
equivalent to the matrix space associated with M2. If α = 0 then permuting
rows and columns shows that V is equivalent to the matrix space associated with
M1.
With a similar line of reasoning as in Case 2, one finds that V is equivalent to the
.
0 0 ?
0 0 1
0 0 0
Case 3. M0 =
matrix space associated with M3 or to the one associated with
which is easily seen to be equivalent to the one associated with M1.
0
0
a
0 a + b c
0
b
0
,
27
0 1 ?
0 0 1
0 0 0
.
Case 4. M0 =
M0 =
0 1 0
0 0 1
0 0 0
with the generic matrix
Using C3 ← C3+C2 if necessary, we see that no generality is lost in assuming that
. Then, we have scalars α, β, γ, δ, η, ǫ such that V is associated
a αa + βb + c
0
0
a + b
0
ηa + ǫb
γa + δb + c
b
.
Using the operations C2 ← C2 + (α + γ)C1, L2 ← L2 + (β + δ)L3, L1 ← L1 + ǫL3
and C3 ← C3 + ηC1, we finally reduce the situation to the one where V is
associated with the generic matrix M4.
It remains to show that the four cited matrix spaces are pairwise inequivalent.
To do this, we note that the equivalence class of a matrix subspace W of M3(F2)
determines both the number of vectors x ∈ (F2)3 for which dim(Wx) = 1 and
the number of vectors x ∈ (F2)3 for which dim(W T x) = 1. For the above four
matrix spaces, we obtain the following results, which show that they are pairwise
inequivalent:
V associated with the generic matrix . . .
Number of vectors x ∈ (F2)3 such that dim(Vx) = 1
Number of vectors x ∈ (F2)3 such that dim(V T x) = 1
M1 M2 M3 M4
1
2
2
1
2
1
1
2
The proofs of Propositions 4.1 and 4.3 are similar and we shall leave them to
the reader. Let us only explain why the four generic matrices given in Proposition
4.3 yield pairwise inequivalent matrix spaces. We simply look at the structure
of the sets of their rank 1 matrices.
• If V is equivalent to the space associated with N1, then it contains exactly
two rank 1 matrices.
• If V is equivalent to the space associated with N2, then it contains exactly
three rank 1 matrices, and they have the same range.
28
• If V is equivalent to the space associated with N3, then it contains exactly
three rank 1 matrices, and they do not have the same range.
• Otherwise, V contains a sole rank 1 matrix.
5 Applications
5.1 Triples of locally linearly dependent operators over F2
In [7, Section 3], we have shown how minimal reduced locally linearly dependent
operator spaces are connected to semi-primitive operator spaces. Let us recall
the basics: Let U and V be finite-dimensional vector spaces, and S be a reduced
linear subspace of L(U, V ). We define the dual operator space bS of S as the
space of all operators from S to V of the form
We say that S is locally linearly dependent (in abbreviated form: LLD) when,
for every vector x ∈ U , there is a non-zero operator s ∈ S such that s(x) = 0.
bx : f ∈ S 7→ f (x), with x ∈ U .
reduced operator spaces S and T are equivalent if and only if their dual operator
Then, S is a minimal LLD space if and only if bS is semi-primitive. Moreover, two
spaces are equivalent. Noting that S is always equivalent tobbS, this yields a one-
to-one correspondence between the equivalence classes of semi-primitive operator
spaces and the ones of minimal reduced LLD spaces.
We have seen that the semi-primitive subspaces of L(U, V ) with upper rank
2 are the primitive ones for which dim V > 2. Thus, we deduce the following
result from Theorem 1.5 and from Propositions 4.1, 4.2 and 4.3.
Theorem 5.1 (Classification of 3-dimensional minimal LLD spaces over F2).
Let S ⊂ L(U, V ) be a 3-dimensional minimal reduced LLD space over F2. Then,
one and only one of the following situations holds:
(a) dim V = 2;
(b) dim U = 2, dim V = 3 and S is represented by the matrix space associated
with
x 0
y y
0
z
.
29
(c) dim U = 3, dim V = 3, and one and only one of the following generic
matrices is associated with a matrix space that represents S:
0 −x −y
0 −z
x
0
y
z
x x + z
0
x + y z
x
0 y
y + z y 0
0
0
z
x y 0
y y z
0
z 0
x 0 z
y y 0
0
z 0
,
,
y
0
,
,
,
x 0 y
y y z
0
0
z
.
x 0 y z
y y z 0
0 0
0
z
0
x
,
,
,
y 0 z z
0
z x 0
x x 0 y
x 0 y z 0
y y 0 0 z
0
0 0 0
z
.
.
(d) dim U = 4, dim V = 3, and one and only one of the following generic
matrices is associated with a matrix space that represents S:
x 0 y 0
y y 0 z
0 0
0
z
x 0 y z
y y 0 0
0 0
0
z
x 0 z 0
y y 0 z
z 0 0
0
and
(e) dim U = 5, dim V = 3, and S is represented by the matrix space associated
with the generic matrix
Conversely, all the above cited matrix spaces represent 3-dimensional minimal
LLD operator spaces.
In (c), the given matrix spaces represent the dual operator spaces of the
matrix spaces A3(F2), U3(F2), and the four matrix spaces cited in Proposition
4.2. In (d), the given matrix spaces represent the dual operator spaces of the
four matrix spaces cited in Proposition 4.3 and of V3(F2). In (e), the matrix
space represents the dual operator space of J3(F2).
The computation of the dual operator spaces is performed in the same way
as in the proof of Lemma 2.2.
30
5.2 Subspaces of M3(F2) with trivial spectrum
Definition 3. Given a field K, a linear subspace V of Mn(K) is said to have a
trivial spectrum when no matrix of V has an eigenvalue in K r {0}.
In [6] and [14], it was proved that a trivial spectrum subspace V of Mn(K)
has dimension at most(cid:18)n
2(cid:19). In [11], the classification of trivial spectrum sub-
spaces with the maximal dimension was achieved for all fields with more than 3
elements, and it was shown that the classification theorem failed for F2. Our aim
here is to use the classification of semi-primitive subspaces of M3(F2) to obtain
the full classification of 3-dimensional trivial spectrum subspaces of M3(F2).
First of all, we introduce some notation:
Notation 4. Let A and B be linear subspaces, respectively, of Mn(K) and
Mp(K). One denotes by A ∨ B the space of all matrices of the form
(cid:20) A
[0]p×n B(cid:21) with A ∈ A, B ∈ B and C ∈ Mn,p(K).
C
A trivial spectrum subspace V of Mn(K) is called irreducible when there is
no proper and non-zero linear subspace F of Kn such that VX ⊂ F for all X ∈ F .
If the contrary holds we say that V is reducible. We have shown in [11] that if
a trivial spectrum subspace V of Mn(K) has dimension(cid:18)n
(n1, . . . , np) of positive integers such thatPp
k=1 nk = n, together with irreducible
trivial spectrum subspaces V1 ⊂ Mn1(K), V2 ⊂ Mn2(K), . . . , Vp ⊂ Mnp(K), such
that
2(cid:19), then there is a list
V ≃ V1 ∨ V2 ∨ · · · ∨ Vp.
In order to obtain the structure of trivial spectrum spaces with the maximal
dimension, it is therefore essential to classify the irreducible ones up to similarity.
The following result was obtained in [11]:
Theorem 5.2. Assume that #K > 2. The irreducible subspaces of Mn(K) with
trivial spectrum and dimension(cid:18)n
2(cid:19) are the spaces of the form P An(K), where
P ∈ GLn(K) is a non-isotropic matrix, i.e. the quadratic form X 7→ X T P X is
non-isotropic. Two such spaces P An(K) and Q An(K) are similar if and only if
there is a non-zero scalar λ such that Q is congruent to λP , that is Q = λRP RT
for some R ∈ GLn(K).
31
For F2, this result holds for n = 2 as well (see the proof in Section 4.1 of
[11]). In that case, the result is simple: An irreducible subspace of M2(F2) with
dimension 1 is spanned by a matrix M ∈ M2(F2) with no eigenvalue in F2. As
X 2 + X + 1 is the only irreducible polynomial of degree 2 over F2, there is only
Using this together with Proposition 16 of [11] - which holds for all fields -
one such matrix M up to similarity: The companion matrix M =(cid:20)0 1
1 1(cid:21).
Proposition 5.3. Set C :=(cid:20)0 1
1 1(cid:21). Up to similarity, there are three reducible
we deduce the 3-dimensional reducible trivial spectrum subspaces of M3(F2):
3-dimensional trivial spectrum subspaces of M3(F2):
F2C ∨ {0},
{0} ∨ F2C and NT3(F2).
Now, we turn to the 3-dimensional irreducible trivial spectrum subspaces of
M3(F2). We shall prove the following result:
Theorem 5.4 (Classification of 3-dimensional irreducible subspaces of M3(F2)
with trivial spectrum). Up to similarity, there are exactly three irreducible 3-
dimensional subspaces of M3(F2) with trivial spectrum:
Let us start by proving that T1, T2 and T3 all satisfy the claimed properties
and that they are pairwise unsimilar. Remember that the identities a(a + 1) = 0
and ab(a + b) = 0 hold for all (a, b) ∈ (F2)2. For all (a, b, c) ∈ (F2)3, we compute
= (a+1)3+bc(b+c)+a3+ba(a+1)+ca(a+1)+(b+c)a(a+1) = a+1+a = 1,
a
a + 1
b + c
which shows that T1 has a trivial spectrum. Similarly, for all ε ∈ F2 and all
(a, b, c) ∈ (F2)3, we have
= (a+1)2(εa+1)+a3+ca(a+1)+ba(a+1) = ε(a+1)a+(a+1)+a = 1,
32
a b a
a 0 c
0 a a
(a, b, c) ∈ (F2)3).
a
a
b a
a c
b + c a a
T1 :=(
(a, b, c) ∈ (F2)3),
(a, b, c) ∈ (F2)3) and T3 :=(
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
a
c
a b a
a a c
0 a a
b
a + 1
a
a + 1
T2 :=(
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
a + 1
a
0
εa + 1
b
a
a
c
a + 1
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
whence T2 and T3 have trivial spectra.
Moreover, we compute that, for all (a, b, c) ∈ (F2)3,
a
a
b a
a c
b + c a a
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
= a3 + bc(b + c) + a3 + ba2 + ca2 + (b + c)a2 = 0.
Therefore, every matrix in T1 is singular. However, the matrix
longs to T2 and has determinant 1, whereas
be-
belongs to T3 and has
determinant 1. Therefore, T1 is unsimilar to both T2 and T3. To see that T2
and T3 are unsimilar, we simply note that T3 contains only trace zero matrices,
whereas T2 does not.
1 0 1
1 1 1
0 1 1
1 1 1
1 0 1
0 1 1
Finally, let us prove that T1, T2 and T3 are all irreducible. We have seen
that if a 3-dimensional trivial spectrum subspace of M3(F2) is reducible, then it
contains only singular matrices, and it contains at least one rank 1 matrix. The
spaces T2 and T3 are both irreducible because each one of them contains a non-
singular matrix. On the other hand, T1 is irreducible because it contains no rank
M has rank 1, then it has trace 0 since it cannot have a non-zero eigenvalue, and
1 matrix: Indeed, let (a, b, c) ∈ (F2)3 r{0}, and consider M :=
this leads to a = 0; Thus, (b, c) 6= (0, 0), and one sees that M =
has rank 2 because exactly two scalars among b, c, b + c equal 1.
a
a
b a
a c
b + c a a
0
0
b 0
0 c
b + c 0 0
. If
Next, we prove that every irreducible 3-dimensional trivial spectrum sub-
space of M3(F2) is similar to one of the Ti's. To achieve this, we shall use a
new technique, featured in [10], that relates such subspaces to semi-primitive
matrix spaces. We recall the basics now: Let K be an arbitrary field and V be
an irreducible trivial spectrum subspace of Mn(K). For each vector X ∈ Kn, we
obtain a bilinear form
(N, Y ) ∈ V × Kn 7−→ Y T N X.
33
Choosing respective bases of V and Kn, we denote by M the space of all matrices
representing the above bilinear forms in those bases. Using the fact that V
is an irreducible trivial spectrum space, one obtains that M is reduced with
upper rank less than n and with dimension n. If M is not semi-primitive, if
dim V = (cid:18)n
of a semi-primitive subspace of Mm,p(K) implies m ≤ (cid:18)p
2(cid:19) and if, for all integers p ∈ [[1, n − 1]] and m > 1, the existence
2(cid:19), then the chain of
arguments from Section 5 of [10] yields that V is reducible, contradicting our
assumptions. In particular, we know from the classification of spaces of matrices
with rank at most 1 that a semi-primitive subspace of Mm,2(K) exists only if
m = 1, and that there is no semi-primitive subspace of Mm,1(K). Moreover, we
have seen in the present article that the existence of a semi-primitive subspace
of Mm,3(K) implies that m ≤ 3. It follows that M is semi-primitive whenever
n ≤ 4 and dim V =(cid:18)n
2(cid:19).
Now, we assume that n = 3, dim V = 3 and K = F2. Then, M is a semi-
primitive subspace of M3(F2) with dimension 3 and upper rank 2, and hence we
deduce from Proposition 1.1 that it is primitive. Thus, MT is also primitive
From there, Theorem 1.5 yields key information on the structure of MT , and
with upper rank 2. One sees that MT represents the dual operator space bV,
whence V is equivalent to dMT .
hence on that of dMT . Using that information will be of great help to understand
the structure of V. We distinguish between several cases.
5.2.1 Case 1. MT is equivalent to A3(F2).
Then, dMT is also equivalent to A3(F2), yielding a non-singular matrix P ∈
GL3(F2) such that V = P A3(F2) (remember that QT A3(F2)Q = A3(F2) for
all Q ∈ GL3(F2)). As every 3-dimensional quadratic form over a finite field is
isotropic, Proposition 10 of [11] yields that V cannot have a trivial spectrum,
contradicting our assumptions.
5.2.2 Case 2. MT is equivalent to U3(F2).
As we have seen in the course of the proof of Lemma 2.2, the dual operator
space \U3(F2) is equivalent to U3(F2), whence V is equivalent to U3(F2). In that
34
case, we note that every matrix in V has rank 2 and that the matrices of V have
pairwise distinct ranges and pairwise distinct kernels.
For M ∈ V, we write the characteristic polynomial of M as χM (t) = t3 −
tr(M )t2 + q(M )t, and the condition that 1 does not belong to the spectrum of
M reads 1 − tr(M ) + q(M ) = 1, whence
∀M ∈ V, q(M ) = tr(M ).
However, q is a quadratic form with polar form (A, B) 7→ tr(A) tr(B) + tr(AB),
and hence
∀(A, B) ∈ V 2, tr(AB) = tr(A) tr(B).
(3)
The linear subspace H := {M ∈ V :
tr(M ) = 0} has codimension at most 1
in V and consists only of nilpotent matrices. If dim H = 3, then V = H and
Gerstenhaber's theorem (see [13, 15]) would yield that V is reducible. Thus,
dim H = 2. Using identity (3), we see that H is included in the radical of the
symmetric bilinear form (A, B) 7→ tr(AB) on V 2.
Claim 5. We define
H1 :=(
0
0
a 0
0 b
a + b 0 0
(a, b) ∈ (F2)2) and H2 :=(
0 0 a
0 0 b
b a 0
(a, b) ∈ (F2)2).
Then, H is equivalent to H1 or to H2.
More generally, it can be shown that an irreducible subspace of nilpotent
matrices of M3(F2) is always equivalent to H1 or to H2.
Proof. We have just seen that ∀(A, B) ∈ H2, tr(AB) = 0. Take linearly inde-
pendent matrices A1 and A2 in H. We know that A1 and A2 are both rank 2
nilpotent matrices with different kernels and different ranges. We distinguish
between two main cases, whether Ker A2 ⊂ Ker A2
Case a. Ker A2 ⊂ Ker A2
invertible matrix so as to reduce the situation to the one where
1. Then, we see that we can conjugate H with an
1 holds or not.
A1 =
0 1 0
0 0 1
0 0 0
and A2 =
? 0 ?
? 0 ?
? 0 ?
.
Using tr(A1A2) = 0, we deduce that the entry of A2 at the (2, 1)-spot is zero. If
the one at the (3, 1)-spot were zero, then the whole first column of A2 would be
35
zero since A2 is nilpotent, contradicting the fact that A2 has rank 2. Thus, if we
denote by (e1, e2, e3) the standard basis of (F2)3, we see that A2e1 6∈ Ker A2
1 and
A1A2e1 = e2. Thus, we may now use (e1, e2, A2e1) as our new basis, thereby
reducing the situation to the one where
As tr A2 = 0, we have c = 0. The characteristic polynomial of A1 + A2 then
equals t3 + at + (b + 1), whence a = 0 and b = 1. We conclude that H is the
1. Then, we take x ∈ Ker A2r{0} and we work with
1x, A1x, x). Thus, using the relations tr(A2) = 0 and tr(A1A2) = 0,
0 1 0
0 0 1
0 0 0
and A2 =
A1 =
space of all matrices of the form
and A2 =
A1 =
Case b. Ker A2 6⊂ Ker A2
0 1 0
0 0 1
0 0 0
the basis (A2
the situation is reduced to the one where
a c 0
b a 0
d b 0
0 0 a
0 0 b
1 0 c
0
0
x 0
0 y
x + y 0 0
for some (a, b, c) ∈ (F2)3.
with (x, y) ∈ (F2)2.
for some (a, b, c, d) ∈ (F2)4.
The characteristic polynomial of A2 + A1 is t3 + t(a + bc) + (ab + d(c + 1)), and
hence a = bc and ab = d(c + 1). Note that (d, b) 6= (0, 0) since Im A2 6= Im A1
and rk A2 = rk A1.
If b = 0, then we deduce that d = 1, and hence a = 0
and c = 1. In that case, we see once more that H is similar to the same space
as in Case a. Assume now that b = 1. Then, c = a = d(c + 1), and hence
c = c2 = d(c + 1)c = 0.
It follows that a = 0 and d = 0. Using the basis
(e1, e3, e2), we conclude that H is similar to H2.
Now, we aim at discarding the second case in Claim 5. Assume that H is
similar to H2. Then, no generality is lost in assuming that H = H2. As no
matrix A ∈ V satisfies Ae3 = e3, we have dim Ve3 ≤ 2, yielding a non-zero
matrix M ∈ V such that M e3 = 0. Then, M 6∈ H. Let us write
L
M =(cid:20)N [0]2×1
0 (cid:21) with N ∈ M2(F2) and L ∈ M1,2(F2).
all A ∈ H, computing the determinant shows that, for K := (cid:20)0 1
As tr(M A) = 0 for all A ∈ H, we find that L = 0. As M + A is singular for
1 0(cid:21), one has
36
X T N KX = 0 for all X ∈ (F2)2. One deduces that N K is alternating, and
hence N = I2 as N is non-zero. We obtain that tr(M ) = 0, contradicting the
assumption that M 6∈ H.
Thus, H is similar to H1. Without loss of generality, we may assume that
H = H1. Now, let us choose a matrix M in V r H. Adding an appropriate
matrix of H, we may assume that
? 0 ?
? ? 0
? ? ?
.
M =
M =
0
a
d
d a + b + 1 0
c
b
d
.
As tr(M A) = 0 for all A ∈ H, while tr(M ) = 1, we obtain (a, b, c, d) ∈ (F2)4
such that
If d = 0, then we see that a, b and a + b + 1 are all eigenvalues of M , which is
impossible since not all of them are zero. It follows that d = 1. Then, for all
(x, y) ∈ (F2)2, we deduce that
0 =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
a
1
c + x + y
x
1
a + b + 1 y
b
1
= cxy + (a + 1)x + (b + 1)y +(cid:0)(a + b + 1)c + ab + 1(cid:1).
It follows that c = 0, a = 1 and b = 1. Thus, V = T1.
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
5.2.3 Case 3. MT is equivalent to a subspace of J3(F2).
As MT has dimension 3, we lose no generality in assuming that it equals one of
the spaces listed in Proposition 4.2. To the space of all operators M ∈ MT 7→
M x, with x ∈ (F2)2 × {0}, then corresponds a linear subspace H of V with one
of the following properties, whether MT is represented by one of the generic
matrices M2, M4 or by one of the generic matrices M1, M3:
• Subcase 3.1. There is a basis (A1, A2) of H in which rk A1 = 1, rk A2 = 2,
Im A1 ⊂ Im A2 and Ker A1 ⊕ Ker A2 = (F2)3.
• Subcase 3.2. H contains two rank 1 matrices A1 and A2 such that Im A1 6=
Im A2 and Ker A1 6= Ker A2.
37
Before we can tackle each case separately, we need a simple lemma:
Lemma 5.5. Let T be a trivial spectrum linear subspace of M3(F2) in which all
the elements have their range included in (F2)2 × {0}. Then, there is a matrix
N ∈ M2(K) whose spectrum does not contain 1 and such that every matrix of T
splits up as
Proof. We can write every matrix of T as
[0]1×2
for some λ ∈ F2.
M =(cid:20) λN [?]2×1
0 (cid:21)
M =(cid:20)K(M )
0 (cid:21) with K(M ) ∈ M2(F2).
[0]1×2
[?]2×1
Then, K(H) is a trivial spectrum subspace of M2(F2). By Theorem 9 of [14], we
have dim K(H) ≤ 1. The result follows by taking N as the sole non-zero vector
of K(H) if dim K(H) = 1, and N = 0 otherwise.
We seek to discard Subcase 3.1. Assume that it holds and note that the 2-
dimensional space Im A2 contains the range of every matrix in H. Conjugating
V with a well-chosen invertible matrix, we lose no generality in assuming that
Im A2 = K2 × {0}. The above lemma yields some N ∈ M2(F2) for which 1 is not
an eigenvalue and such that every matrix M of H splits up as
M =(cid:20) λN [?]2×1
0 (cid:21)
[0]1×2
for some λ ∈ F2.
If N were singular, we would find a non-zero vector that belongs to the kernel
of all the matrices in H, contradicting the fact that Ker A1 ⊕ Ker A2 = (F2)3.
Therefore, N ∈ GL2(F2), and hence the characteristic polynomial of N must be
t2 + t + 1. As A2 has rank 2, we must have
A2 =(cid:20) N
[0]1×2
[?]2×1
0 (cid:21) .
In turn, this shows that Ker A2 ⊕ Im A2 = (F2)3, whence an additional conjuga-
. On the other
tion allows one to assume that Ker A2 is the span of(cid:2)0 0 1(cid:3)T
hand, as A1 has rank 1, we must have
A1 =(cid:20)[0]2×2 C
0(cid:21)
[0]1×2
for some C ∈ (F2)2 r {0}.
38
Since N has no eigenvalue in F2, we see that C and N C are linearly independent,
and we note that N (N C) = C +N C. Conjugating by the change of bases matrix
we reduce the situation further to the point where
0
0
0 0 1
0 0 0
0 0 0
1 (cid:21) ∈ GL3(F2),
Q :=(cid:20)C N C [0]2×1
A1 =
A3 =
.
and A2 =
for some (a, b, c, d, e, f, g) ∈ (F2)7.
a 0 0
b
c d
g
e f
0 1 0
1 1 0
0 0 0
Then, we extend (A1, A2) into a basis (A1, A2, A3) of V. Choosing A3 well, we
may assume that
Note that a = 0 since A3 has no non-zero eigenvalue. Denote by (e1, e2, e3) the
standard basis of (F2)3. Recall that, for all non-zero vectors x ∈ (F2)3, none of
the spaces V T x and Vx contains x, whence dim V T x < 3 and dim Vx < 3. In
particular, dim V(e2 + e3) < 3 and dim V(e1 + e3) < 3 yield f = g and e = g,
respectively. As V is irreducible, we deduce that e = f = g = 1. Then we obtain
V T (e1 + e3) = (F2)3, contradicting the above remarks.
We have just shown that Subcase 3.1 cannot hold. Thus, we obtain two
rank 1 matrices A1 and A2 in V with distinct kernels and ranges. Setting P :=
Im A1 + Im A2, we lose no generality in assuming that P = (F2)2 × {0}. Let us
consider a matrix N ∈ M2(F2) obtained by applying Lemma 5.5 to the trivial
spectrum space H := span(A1, A2). As Ker A1 6= Ker A2, we must have N 6= 0,
and, without loss of generality, we may assume that
A1 =(cid:20) N
[0]1×2
[?]2×1
0 (cid:21) .
As A1 has rank 1 and has no non-zero eigenvalue in F2, it is nilpotent, whence
N is nilpotent. Thus, as A1 has rank 1, no further generality is lost in assuming
that
0 0(cid:21)
N =(cid:20)0 1
and A1 =
0 1 α
0 0 0
0 0 0
for some α ∈ F2.
39
As A2 has rank 1 and Im A2 6= Im A1, the only option is that
A2 =
0 0 β
0 0 1
0 0 0
for some β ∈ F2.
0 x 0
0 0 y
0 0 0
(x, y) ∈ (F2)2).
H =(
a 0 b
c d 0
e f
g
From there, conjugating V with Q :=
one where
1 −β 0
0
α
1
0
1
0
reduces the situation to the
Now, we find a matrix A3 of V r H of the form
A3 =
with (a, b, c, d, e, f, g) ∈ (F2)7.
Noting that V = span(A1, A2, A3), we find 1 = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(x, y) ∈ (F2)2. Expanding the right-hand side of this equality, we find a polyno-
mial with degree at most 1 in each variable x and y, of the form
x
d + 1
b
y
f
g + 1
a + 1
c
e
for all
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
e(xy) + c(g + 1)x + f (a + 1)y+?.
It follows that e = 0, c(g + 1) = 0 and f (a + 1) = 0. If c = 0, then a is an
eigenvalue of A3, and hence the first column of A3 is zero. It would follow that
V is reducible, contradicting our assumptions. Thus, c = 1, and one finds f = 1
with the same line of reasoning. One deduces from the above equalities that
g = 1 and a = 1. As 1 is not an eigenvalue of A3, we must have b = 1. Finally,
we conclude that V = T2 or V = T3, whether d = 1 or d = 0. This finishes the
proof of Theorem 5.4.
Remark 3. Using the elementary operation L2 ← L2 + L3, one sees that T2 and
T3 are equivalent. Moreover, one can check that the dual operator space of T3
is equivalent to the matrix space associated with M3.
40
5.3 Affine subspaces of non-singular matrices of M3(F2)
Using the classification of 3-dimensional trivial spectrum subspaces of M3(F2),
we are now able to classify, up to equivalence, the 3-dimensional affine subspaces
of M3(F2) that are included in GL3(F2).
We need only classify those affine subspaces that contain I3. Given a linear
subspace H of M3(F2), the affine subspace I3 + H is included in GL3(F2) if and
only if H is a trivial spectrum space. Note that if H and H ′ are similar linear
subspaces, then I3 + H and I3 + H ′ are equivalent affine spaces (the converse
does not hold in general). Therefore, using Proposition 5.3 and Theorem 5.4,
we obtain that each 3-dimensional affine subspace that is included in GL3(F2)
is equivalent to one of the following six affine subspaces, where we have set
C :=(cid:20)0 1
1 1(cid:21):
I3+NT3(F2),
I3+(F2C∨{0}),
I3+({0}∨F2C),
I3+T1,
I3+T2,
and I3+T3.
It remains to investigate potential equivalences between those six spaces. This
involves the following lemma:
Lemma 5.6. Let K be an arbitrary field, and H1 and H2 be linear subspaces of
Mn(K) for which the affine spaces In + H1 and In + H2 are equivalent. Then,
H1 is irreducible if and only if H2 is irreducible.
Proof. Assume that H1 is reducible, and consider a non-zero proper linear sub-
space P of Kn such that M x ∈ P for all M ∈ H1 and x ∈ P . It follows that
M x ∈ P for all M ∈ (In + H1) and x ∈ P . Using the assumed equivalence
between In + H1 and In + H2, we recover non-zero proper linear subspaces P1
and P2 of Kn such that M x ∈ P2 for all M ∈ (In + H2) and x ∈ P1.
In
particular, this holds for M = In, whence P2 = P1. Thus, we conclude that
M x = (In + M )x − Inx ∈ P1 for all x ∈ P1 and all M ∈ H2, whence H2
is reducible. Symmetrically, one obtains that H1 is reducible whenever H2 is
reducible.
Using the above lemma, we deduce that any one of the spaces I3 + NT3(F2),
I3 + (F2C ∨ {0}), I3 + ({0} ∨ F2C) is inequivalent to any one of the spaces I3 + Ti
for i ∈ {1, 2, 3}. Moreover, by Proposition 17 of [11], the spaces I3 + NT3(F2),
I3 + (F2C ∨ {0}) and I3 + ({0} ∨ F2C) are pairwise inequivalent.
It remains
only to investigate possible equivalences between the spaces I3 + T1, I3 + T2 and
I3 + T3.
41
Note that equivalent affine spaces have equivalent translation vector spaces.
However, T1 is inequivalent to both T2 and T3 as T1 contains only rank 2 matrices,
whereas T2 and T3 both contain rank 1 matrices.
1 0 1
1 0 0
0 1 1
Finally, we show that I3 + T2 is equivalent to I3 + T3. To see this, we choose
so that the characteristic polynomial of A is t3 + t + 1. Thus, the characteristic
),
an arbitrary matrix A ∈ T3 ∩ GL3(F2) (an obvious choice is A =
polynomial of I3 + A is t3 + t2 + 1, so that tr(cid:0)(I3 + A)−1(cid:1) = 0. We note that
I3 + T3 = (I3 + A) + T3 = (I3 + A)(cid:0)I3 + (I3 + A)−1T3(cid:1) ∼ I3 + (I3 + A)−1T3.
Using Lemma 5.6, we see that (I3 + A)−1T3 is an irreducible trivial spectrum
space with dimension 3; As it is equivalent to T3, it cannot be equivalent to
T1, whence it is similar to T2 or to T3. However, since tr((I3 + A)−1A) =
tr(I3) + tr((I3 + A)−1) = 1, we see that (I3 + A)−1T3 contains a matrix with
trace 1, whence (I3 + A)−1T3 is unsimilar to T3. We conclude that (I3 + A)−1T3
is similar to T2, whence I3 + T2 ∼ I3 + T3. Let us sum up our results:
Theorem 5.7 (Classification of 3-dimensional affine subspaces of non-singu-
lar matrices of M3(F2)). Set C := (cid:20)0 1
1 1(cid:21). Up to equivalence, exactly five 3-
dimensional affine subspaces of M3(F2) are included in GL3(F2): They are the
ones which contain I3 and with respective translation vector spaces NT3(F2),
F2C ∨ {0}, {0} ∨ F2C, T1 and T2.
References
[1] M. D. Atkinson, Primitive spaces of matrices of bounded rank II. J. Austral.
Math. Soc. (Ser. A) 34 (1983), 306 -- 315.
[2] M. D. Atkinson, S. Lloyd, Primitive spaces of matrices of bounded rank. J.
Austr. Math. Soc. (Ser. A) 30 (1980), 473 -- 482.
[3] L. B. Beasley, Spaces of rank 2 matrices over GF (2). Electron. J. Linear
Algebra 5 (1999), 11 -- 18.
[4] D. Eisenbud, J. Harris, Vector spaces of matrices of low rank. Adv. Math.
37(1988), 135 -- 155.
42
[5] P. Fillmore, C. Laurie, H. Radjavi, On matrix spaces with zero determinant.
Lin. Multilin. Algebra 18-3 (1985), 255 -- 266.
[6] R. Quinlan, Spaces of matrices without non-zero eigenvalues in their field of
definition, and a question of Szechtman. Linear Algebra Appl. 434 (2011),
1580 -- 1587.
[7] C. de Seguins Pazzis, Local linear dependence seen through duality II. Linear
Algebra Appl. 462 (2014), 133 -- 185.
[8] C. de Seguins Pazzis, Spaces of matrices with few eigenvalues. Linear Algebra
Appl. 449 (2014), 210 -- 311.
[9] C. de Seguins Pazzis, The classification of large spaces of matrices with
bounded rank. Israel J. Math. in press (2015).
[10] C. de Seguins Pazzis, From primitive spaces of bounded rank matrices to a
generalized Gerstenhaber theorem. Quart. J. Math. 65-2 (2014), 319 -- 325.
[11] C. de Seguins Pazzis, Large affine spaces of non-singular matrices. Trans.
Amer. Math. Soc. 365 (2013), 2569 -- 2596.
[12] C. de Seguins Pazzis, Invitation aux formes quadratiques. Calvage &
Mounet, Paris, 2011.
[13] C. de Seguins Pazzis, On Gerstenhaber's theorem for spaces of nilpotent
matrices over a skew field. Linear Algebra Appl. 438-11 (2013), 4426 -- 4438.
[14] C. de Seguins Pazzis, On the matrices of given rank in a large subspace.
Linear Algebra Appl. 435-1 (2011), 147 -- 151.
[15] V. N. Serezhkin, Linear transformations preserving nilpotency (in Russian).
Izv. Akad. Nauk BSSR, Ser. Fiz.-Mat. Nauk 125 (1985), 46 -- 50.
43
|
1001.0851 | 1 | 1001 | 2010-01-06T10:38:19 | Left ideals in an enveloping algebra, prelie products and applications to simple complex Lie algebras | [
"math.RA"
] | We characterize prelie algebras in words of left ideals of the enveloping algebras and in words of modules, and use this result to prove that a simple complex finite-dimensional Lie algebra is not prelie, with the possible exception of f4. | math.RA | math |
Left ideals in an enveloping algebra, prelie products and
applications to simple complex Lie algebras
L. Foissy
Laboratoire de Mathématiques, Université de Reims
Moulin de la Housse - BP 1039 - 51687 REIMS Cedex 2, France
e-mail : [email protected]
ABSTRACT. We characterize prelie algebras in words of left ideals of the enveloping algebras
and in words of modules, and use this result to prove that a simple complex finite-dimensional
Lie algebra is not prelie, with the possible exception of f4.
Keywords. Prelie algebras, simple complex finite dimensional algebras.
AMS classification. 17B20; 16S30; 17D25.
Contents
1 Prelie products on a Lie algebra
1.1 Preliminaries and recalls on symmetric coalgebras . . . . . . . . . . . . . . . . . .
1.2 Extension of a prelie product
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Prelie product associated to a left ideal . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Prelie products on a Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Good and very good pointed modules
. . . . . . . . . . . . . . . . . . . . . . . .
2 Are the complex simple Lie algebras prelie?
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1 General results
2.2 Representations of small dimension . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Proof in the generic cases
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Case of so2n+1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Proof for sl6 and g2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 Dendriform products on cofree coalgebra
3.1 Preliminaries and results on tensor coalgebras . . . . . . . . . . . . . . . . . . . .
3.2 Left ideal associated to a dendriform Hopf algebra . . . . . . . . . . . . . . . . .
3.3 Dendriform products on a tensorial coalgebra . . . . . . . . . . . . . . . . . . . .
3.4 Dendriform structures on a cofree coalgebra . . . . . . . . . . . . . . . . . . . . .
4 MuPAD computations
4.1 Dimensions of simple modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Computations for sl6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Computations for g2
3
3
6
8
9
10
12
12
12
13
14
16
17
17
18
19
21
22
22
24
26
1
Introduction
A (left) prelie algebra, or equivalently a left Vinberg algebra, or a left-symmetric algebra [1, 2, 13]
is a couple (V, ⋆), where V is a vector space and ⋆ a bilinear product on V such that for all
x, y, z ∈ V :
(x ⋆ y) ⋆ z − x ⋆ (y ⋆ z) = (y ⋆ x) ⋆ z − y ⋆ (x ⋆ z).
This axiom implies that the bracket defined by [x, y] = x ⋆ y − y ⋆ x satisfies the Jacobi identity,
so V is a Lie algebra; moreover, (V, ⋆) is a left module on V . The free prelie algebra is described
in [1] in terms of rooted trees, making a link with the Connes-Kreimer Hopf algebra of Renor-
malization [3, 6].
The aim of this text is to give examples of Lie algebras which are not prelie, namely the
simple complex Lie algebras of finite dimension. For this, we use the construction of [11] of the
extension of the prelie product of a prelie algebra (g, ⋆) which gives to the symmetric coalgebra
S(g) an associative product ∗, making it isomorphic to U (g). This construction is also used in
[9] to classify right-sided cocommutative graded and connected Hopf algebras. A remarkable
corollary of this construction is that S(g)≥2 = Ln≥2 Sn(g) is a left ideal of (S(g), ∗). As a
consequence, there exists a left ideal I of U (g), such that the augmentation ideal U (g)+ of U (g)
can be decomposed as U (g)+ = g ⊕ I. We prove here the converse result: more precisely, given
any Lie algebra g, we prove in theorem 11 that there exists a bijection into these two sets:
• PL(g) = {⋆ ⋆ is a prelie product on g inducing the bracket of g}.
• LI(g) = {I left ideal of U (g) U+(g) = g ⊕ I}.
Let then a prelie algebra g, and I the ideal corresponding to the prelie product of g. The
g-module U (g)/I contains a submodule of codimension 1, isomorphic to (g, ⋆), and the special
element 1 + I. This leads to the definition of good pointed modules (definition 13). The sets
PL(g) and LI(g) are in bijection with the set GM(g) of isoclasses of good pointed modules, as
proved in theorem 14.
If g is semisimple, then any good pointed module is isomorphic to K ⊕ (g, ⋆) as a g-module.
This leads to the definition of very good pointed module, and we prove that if g is a semisimple
Lie algebra, then it is prelie if, and only if g has a very good pointed module. If g is a simple
complex lie algebra, with the possible exception of f4, we prove that it has no very good pointed
module, so g is not prelie. The proof is separated into three cases: first, the generic cases, then
so2n+1, and finally sl6 and g2, which are proved by direct computations using the computer al-
gebra system MuPAD pro 4. We conjecture that this is also true for f4 (the computations would
be similar, though very longer).
We also give in this text a non cocommutative version of theorem 11, replacing enveloping
algebras (which are symmetric coalgebras, as the base field is of characteristic zero) by cofree
coalgebras, and prelie algebras by dendriform algebras [8, 10].
If A is a cofree coalgebra, we
prove in theorem 32 that there is a bijection between these two sets:
• DD(A) = {(≺, ≻) (A, ≺, ≻, ∆) is a dendriform Hopf algebra}.
• LI(A) =(cid:26)(∗, I)
(A, ∗, ∆) is a Hopf algebra and
I is a left ideal of A such that A+ = P rim(A) ⊕ I (cid:27).
The text is organized as follows: the first section deals with the general results on prelie
algebras: after preliminaries on symmetric coalgebras, theorem 11 is proved and good pointed
modules are introduced. The second section is devoted to the study of good pointed modules
over simple complex Lie algebras. The results on dendriform algebras are exposed in the third
section and the MuPAD procedures used in this text are written in the last section.
2
Notations.
1. K is a commutative field of characteristic zero. Any algebra, coalgebra, bialgebra, etc, of
this text will be taken over K.
2. Let g be a Lie algebra. We denote by U (g) its enveloping algebra and by U+(g) the
augmentation ideal of U (g).
1 Prelie products on a Lie algebra
1.1 Preliminaries and recalls on symmetric coalgebras
Let V be a vector space. The algebra S(V ) is given a coproduct ∆, defined as the unique algebra
morphism from S(V ) to S(V ) ⊗ S(V ), such that ∆(v) = v ⊗ 1 + 1 ⊗ v for all v ∈ V . Let us recall
the following facts:
• Let us fix a basis (vi)i∈Λ of V . We define SΛ as the set of sequences a = (ai)i∈Λ of elements
ai. For all a ∈ SΛ,
of N, with a finite support. For an element a ∈ SΛ, we put l(a) =Xi∈Λ
we put:
vai
i
ai!
.
va =Yi∈Λ
Then (va)a∈SΛ is a basis of S(V ), and the coproduct is given by:
∆(va) = Xb+c=a
vb ⊗ vc.
• Let S+(V ) be the augmentation ideal of S(V ). It is given a coassociative, non counitary
coproduct ∆ defined by ∆(x) = ∆(x) − x ⊗ 1 − 1 ⊗ x for all x ∈ S+(V ). In other terms,
putting S+Λ = SΛ − {(0)}, (va)a∈S+Λ is a basis of S+(V ) and:
∆(va) = Xb+c=a
b,c∈S+Λ
vb ⊗ vc.
• Let ∆(n) : S+(V ) −→ S+(V )⊗(n+1) be the n-th iterated coproduct of S+(V ). Then:
Ker(cid:16) ∆(n)(cid:17) =
n
Mk=1
Sk(V ).
In particular, P rim(S(V )) = V .
• (Sn(V ))n∈N is a gradation of the coalgebra S(V ).
Lemma 1 In S+(V ) ⊗ S+(V ), Ker( ∆ ⊗ Id − Id ⊗ ∆) = Im( ∆) + V ⊗ V .
Proof. ⊇. Indeed, if y = ∆(x) + v1 ⊗ v2 ∈ Im( ∆) + V ⊗ V , as ∆ is coassociative:
( ∆ ⊗ Id)(y) − (Id ⊗ ∆)(y) = ( ∆ ⊗ Id) ◦ ∆(x) − (Id ⊗ ∆) ◦ ∆(x)
+ ∆(v1) ⊗ v2 − v1 ⊗ ∆(v2)
= 0.
⊆. Let X ∈ Ker( ∆ ⊗ Id − Id ⊗ ∆). Choosing a basis (vi)i∈Λ of V , we put:
xa,bva ⊗ vb.
X = Xa,b∈S+Λ
3
By homogeneity, we can suppose that X is homogeneous of a certain degree n ≥ 2 in S+(V ) ⊗
S+(V ). If n = 2, then X ∈ V ⊗ V . Let us assume that n ≥ 3. Then:
( ∆ ⊗ Id) ◦ ∆(X) = Xa,b,c∈S+Λ
xa+b,cva ⊗ vb ⊗ vc = (Id ⊗ ∆) ◦ ∆(X) = Xa,b,c∈S+Λ
xa,b+cva ⊗ vb ⊗ vc.
So, for all a, c ∈ S+Λ, b ∈ SΛ, xa+b,c = xa,b+c.
Let a, b, a′, b′ ∈ S+Λ, such that a + b = a′ + b′. Let us show that xa,b = xa′,b′.
First case. Let us assume that the support of a and a′ are not disjoint. For all i ∈ Λ, we put:
if a′
0 if a′
i − ai
ci = (cid:26) a′
i = (cid:26) ai − a′
c′
i − ai > 0,
i − ai ≤ 0,
i > 0,
i ≤ 0.
if ai − a′
i
0 if ai − a′
Then c and c′ belong to SΛ. Moreover, for all i ∈ Λ, ai + ci = a′
a′ = a + c − c′. As a + b = a′ + b′, b′ = b − c + c′.
i + c′
i, so a + c = a′ + c′, or
For all i ∈ Λ:
ai − c′
i > 0,
i ≤ 0.
So a − c′ ∈ SΛ. Moreover, if a − c′ = 0, then if ai > a′
i = 0, so the support of a′ is included
in {i / ai ≤ a′
i}. As a
consequence, the supports of a and a′ are disjoint: this is a contradiction. So a − c′ ∈ S+Λ. As
a − c′ and b′ ∈ S+Λ:
i, then ai = 0, so the support of a is included in {i / ai > a′
i}. If ai ≤ a′
if ai − a′
if ai − a′
i =(cid:26) a′
i
ai
i, a′
xa′,b′ = xa−c′+c,b′ = xa−c′,b′+c = xa−c′,b+c′.
As a − c′ and b ∈ S+Λ:
xa−c′,b+c′ = xa,b.
The proof is similar if the support of b and b′ are not disjoint, permuting the roles of (a, a′) and
(b, b′).
Second case. Let us assume that the supports of a and a′, and the supports of b and b′ are
disjoint. We then denote, up to a permutation of the index set Λ:
a = (a1, . . . , ak, 0, . . . , 0, . . .),
a′ = (0, . . . , 0, a′
k+1, . . . , a′
k+l, 0, . . .),
where the ai's and the a′
j's are non-zero. As a + b = a′ + b′, necessarily bk+1 6= 0.
First subcase.
l(a) or l(a′) > 1. We can suppose that l(a) > 1. Then a and (a1 −
1, a2, . . . , ak, 0, . . .) are elements of S+Λ because l(a) > 1 and their support are non disjoint.
By the first case:
xa,b = x(a1−1,a2,...,ak,0,...),b+(1,0,...).
Moreover, the supports of (a1−1, a2, . . . , ak, 0, . . .) and (a1−1, a2, . . . , ak, 1, 0, . . .) are not disjoint.
As bk+1 > 0, b + (1, 0, . . .) − (0, . . . , 0, 1, 0, . . .) belongs to S+Λ. By the first case:
x(a1−1,a2,...,ak,0,...),b+(1,0,...) = x(a1−1,a2,...,ak,1,0,...),b+(1,0,...,0,−1,0,...).
Finally, the support of (a1 − 1, a2, . . . , ak, 1, 0, . . .) and a′ are not disjoint, so:
x(a1−1,a2,...,ak,1,0,...),b+(1,0,...,0,−1,0,...) = xa′,b′.
4
If l(b) or l(b′) > 1, the proof is similar, permuting the roles of (a, a′) and (b, b′).
Second subcase.
l(a) = l(a′) = l(b) = l(b′) = 1. Then va ⊗ vb, va′ ⊗ vb′ ∈ V ⊗ V . By the
homogeneity condition, xa,b = xa′,b′ = 0.
Hence, we put xa+b = xa,b for all a, b ∈ S+Λ: this does not depend of the choice of a and b.
So:
so X ∈ Im( ∆).
X = Xc∈S+Λ
xc Xa+b=c
a,b∈S+Λ
va ⊗ vb = Xc∈S+Λ
xc ∆(vc),
Lemma 2 In S(V ), Im( ∆) ∩ (V ⊗ V ) = S2(V ).
Proof. ⊆. By cocommutativity of ∆. ⊇. If v, w ∈ V , then v ⊗ w + w ⊗ v = ∆(vw).
✷
✷
Lemma 3 Let W be a subspace of S+(V ), such that S+(V ) = V ⊕ W . There exists a unique
coalgebra endomorphism φ of S(V ), such that φV = IdV and φ
Mn≥2
φ is an automorphism.
Sn(V )
= W . Moreover,
We shall denote from now:
S≥2(V ) =
Sn(V ).
∞
Mn=2
Proof. Existence. We denote by πW the projection on W in the direct sum S(V ) = (1)⊕V ⊕
W . We define inductively φSn(V ). If n = 0, it is defined by φ(1) = 1. If n = 1, it is defined by
φV = IdV . Let is assume that φ is defined on the subcoalgebra Cn−1 = K ⊕ V ⊕ . . . ⊕ Sn−1(V )
with n ≥ 2 and let us define φ on Sn(V ). Let va ∈ Sn(V ). Then ∆(va) ∈ Cn−1 ⊗ Cn−1, so
(φ ⊗ φ) ◦ ∆(va) is already defined. Moreover:
( ∆ ⊗ Id) ◦ (φ ⊗ φ) ◦ ∆(va) = (φ ⊗ φ ⊗ φ) ◦ ( ∆ ⊗ Id) ◦ ∆(va)
= (φ ⊗ φ ⊗ φ) ◦ (Id ⊗ ∆) ◦ ∆(va)
= (Id ⊗ ∆) ◦ (φ ⊗ φ) ◦ ∆(va).
By lemma 1, (φ ⊗ φ) ◦ ∆(va) ∈ Im( ∆) + V ⊗ V . Moreover, as ∆ is cocommutative, using lemma
2:
(φ ⊗ φ) ◦ ∆(va) ∈ (Im( ∆) + V ⊗ V ) ∩ S2(S(V ))
∈ Im( ∆) + (V ⊗ V ) ∩ S2(S(V ))
∈ Im( ∆) + S2(V )
∈ Im( ∆).
Let wa ∈ S+(V ), such that (φ ⊗ φ) ◦ ∆(va) = ∆(wa). We put then φ(va) = πW (wa). So
φ(va) ∈ W . Moreover, wa − πW (wa) ∈ V ⊆ Ker( ∆), so:
(φ ⊗ φ) ◦ ∆(va) = φ(va) ⊗ φ(1) + φ(1) ⊗ φ(va) + (φ ⊗ φ) ◦ ∆(va)
= φ(va) ⊗ 1 + 1 ⊗ φ(va) + ∆(wa)
= φ(va) ⊗ 1 + 1 ⊗ φ(va) + ∆(πW (wa))
= ∆(φ(va)).
So φCn is a coalgebra morphism.
5
Unicity. Let φ be another coalgebra endomorphism satisfying the required properties. Let
us show that φ(va) = φ(va) by induction on n = l(a). If n = 0 or 1, this is immediate. Let us
assume the result at all rank < n, n ≥ 2. Then:
∆(φ(va) − φ(va)) = (φ ⊗ φ − φ ⊗ φ)
Xb+c=a∈S+Λ
vb ⊗ vc
= 0,
by the induction hypothesis. So φ(va) − φ(va) ∈ P rim(S(V )) = V . Moreover, it belongs to
φ(S≥2(V )) + φ(S≥2(V )) ⊆ W . As V ∩ W = (0), φ(v1 . . . vn) = φ(v1 . . . vn).
We have defined in this way an endomorphism φ, such that φV = IdV and φ(Sn(V )) ⊆ W
for all n ≥ 2. We now show that φ is an automorphism. Let us suppose that Ker(φ) 6= (0). As φ
is a coalgebra morphism, Ker(φ) is a non-zero coideal, so it contains primitive elements, that is
to say elements of V : impossible, as φV is monic. So φ is monic. Let us prove that va ∈ Im(φ)
by induction on l(a). If l(a) = 0 or 1, then φ(va) = va. We can suppose that va = λv1 . . . vk,
where λ is a non-zero scalar. Then:
∆k−1(va) = λ Xσ∈Sk
vσ(1) ⊗ . . . ⊗ vσ(k) = λ Xσ∈Sk
φ(vσ(1)) ⊗ . . . ⊗ φ(vσ(k)) = ∆k−1(φ(va)).
So va − φ(va) ∈ Ker( ∆k−1) = K ⊕ V ⊕ . . . ⊕ Sk−1(V ) ⊆ Im(φ) by the induction hypothesis. So
va ∈ Im(φ) and φ is epic. As a consequence, φ(S≥2(V )) = W .
✷
1.2 Extension of a prelie product
Definition 4 A (left) prelie algebra is a vector space g, with a product ⋆ satisfying the
following property: for all x, y, z ∈ g,
(x ⋆ y) ⋆ z − x ⋆ (y ⋆ z) = (y ⋆ x) ⋆ z − y ⋆ (x ⋆ z).
Remark. A prelie algebra g is also a Lie algebra, with the bracket given by:
This bracket will be called the Lie bracket induced by the prelie product.
[x, y] = x ⋆ y − y ⋆ x.
Let (g, ⋆) be a prelie algebra. By [5] and [11], permuting left and right, the prelie product ⋆
can be extended to the coalgebra S(g) in the following way: for all x, y ∈ g, P, Q, R ∈ S(g),
Then S(g) is given an associative product ∗ defined by P ∗ Q = P P (1)(cid:0)P (2) ⋆ Q(cid:1). Moreover,
(S(g), ∗, ∆) is isomorphic, as a Hopf algebra, to U (g). There exists a unique isomorphism ξ⋆ of
Hopf algebras:
1 ⋆ P = P,
(xP ) ⋆ y = x ⋆ (P ⋆ y) − (x ⋆ P ) ⋆ y,
P ⋆ (QR) = P(cid:0)P (1) ⋆ Q(cid:1)(cid:0)P (2) ⋆ R(cid:1) .
ξ⋆ :(cid:26) U (g) −→ (S(g), ∗, ∆)
x ∈ g −→ x ∈ g.
Lemma 5 For all n ≥ 1, g ⋆ Sn(g) ⊆ Sn(g).
Proof. By induction on n. This is immediate for n = 1. Let us assume the result at rank
n − 1. Let P = yQ ∈ Sn(g), with y ∈ g and Q ∈ Sn−1(g). For all x ∈ g, as x is primitive:
Note that x ⋆ y ∈ g and x ⋆ Q ∈ Sn−1(g) by the induction hypothesis. So x ⋆ P ∈ Sn(g).
✷
x ⋆ P = (x ⋆ y)Q + y(x ⋆ Q).
6
Proposition 6 Let g be a prelie algebra. We denote S≥2(g) =Mn≥2
Sn(g). Then:
1. S≥2(g) is a left ideal for ∗.
2. S≥2(g) is a bilateral ideal for ∗ if, and only if, ⋆ is associative on g.
Proof. 1. Let x ∈ g and Q ∈ Sn(g), with n ≥ 2. Then, by lemma 5:
x ∗ Q = xQ + x ⋆ Q ∈ Sn+1(V ) + Sn(V ).
As a consequence, S≥2(g) is stable by left multiplication by an element of g. As g generates
(S(g), ∗) (because it is isomorphic to U (g)), S≥2(g) is a left ideal.
2, ⇐=. Let us first show that Sn(g) ⋆ g ⊆ S≥2(g) for all n ≥ 2 by induction on n. For
n = 2, take x, y, z ∈ g. Then (xy) ⋆ z = x ⋆ (y ⋆ z) − (x ⋆ y) ⋆ z = 0, as ⋆ is associative
on g. Let us assume the result at rank n − 1 (n ≥ 3). Let x ∈ g, P ∈ Sn−1(V ), y ∈ g.
Then (xP ) ⋆ z = x ⋆ (P ⋆ z) − (x ⋆ P ) ⋆ z. By the induction hypothesis, P ⋆ z ∈ S≥2(g). By
lemma 5, x ⋆ (P ⋆ z) ∈ S≥2(g). By lemma 5, x ⋆ P ∈ Sn−1(g). By the induction hypothesis,
(x ⋆ P ) ⋆ z ∈ S≥2(g). So Sn(g) ⋆ g ⊆ S≥2(g) for all n ≥ 2.
Let us now prove that S≥2(g) ∗ g ⊆ S≥2(g). Let P ∈ S≥2(g) and y ∈ g. We put ∆(P ) =
P ⊗ 1 + 1 ⊗ P +P P ′ ⊗ P ′′. Then:
P ∗ y = P y + P ⋆ y +X P ′(P ′′ ⋆ y).
Note that P y and P P ′(p′′ ⋆ y) belong to S≥2(g). We already proved P ⋆ y ∈ S≥2(g). As g
generates (S(g), ∗), S≥2(g) is a right ideal. By the first point, it is a bilateral ideal.
2, =⇒. Let us assume that ⋆ is not associative on g. There exists x, y, z ∈ g, such that
x ⋆ (y ⋆ z) − (x ⋆ y) ⋆ z 6= 0. Then:
(xy) ∗ z = xyz + x(y ⋆ z) + y(x ⋆ z) + (xy) ⋆ z
= xyz + x(y ⋆ z) + y(x ⋆ z)
+ x ⋆ (y ⋆ z) − (x ⋆ y) ⋆ z
,
so (xy) ∗ z /∈ S≥2(g), which is not a right ideal.
∈S≥2(g)
{z
}
∈V −{0}
{z
}
✷
Definition 7 Let g be a Lie algebra. We define:
1. PL(g) = {⋆ ⋆ is a prelie product on g inducing the bracket of g}.
2. LI(g) = {I left ideal of U (g) U+(g) = g ⊕ I}.
Proposition 8 There exists an application:
Φg :(cid:26) PL(g) −→ LI(g)
⋆ −→ ξ−1
⋆ (S≥2(g))
Proof. Φg(⋆) is indeed an element of LI(g), as S≥2(g) is a left ideal of (S(g), ∗) and ξ⋆ is an
✷
isomorphism of algebras.
7
1.3 Prelie product associated to a left ideal
Proposition 9 Let I ∈ LI(g). We denote by g the projection on g in the direct sum
U (g) = (1) ⊕ g ⊕ I. Then the product ⋆ defined by x ⋆ y = g(xy) is an element of PL(g). This
defines an application Ψg : LI(g) −→ PL(g).
Proof. Let x, y ∈ g. In U (g), xy − yx = [x, y] ∈ g, so:
x ⋆ y − y ⋆ x = g(xy − yx) = g([x, y]) = [x, y],
so ⋆ induces the Lie bracket of g. It remains to prove that it is prelie. Let us fix a basis (vi)i∈Λ
of g. By the Poincaré-Birkhoff-Witt theorem, U (g) is isomorphic to S(g) as a coalgebra. Using
lemma 3, there exists an isomorphism of coalgebras:
φI :(cid:26) S(g) −→ U (g)
v ∈ g −→ g,
such that φI (S≥2(g)) = I. We denote by (va)a∈SΛ the basis of U (g), image of the basis (va)a∈SΛ
of S(g). As a consequence:
• A basis of I is given by (va)a∈SΛ, l(a)≥2.
• For all a ∈ SΛ, ∆(va) = Xb+c=a
vb ⊗ vc.
We put δj = (δi,j)i∈I for all j ∈ I (so vi = vδi for all i ∈ I) and, for all a, b ∈ SΛ:
By definition, for all i, j ∈ I:
vavb = Xc∈SΛ
vi ⋆ vj =Xk∈I
xc
a,bvc.
xδk
δi,δj
vk.
Let us prove the prelie relation for vi, vj, vk. It is obvious if i = j: let us assume that i 6= j.
Then, in U (g):
∆(cid:16)vivj − vδi+δj(cid:17) = (cid:16)vivj − vδi+δj(cid:17) ⊗ 1 + 1 ⊗(cid:16)vivj − vδi+δj(cid:17)
= (cid:16)vivj − vδi+δj(cid:17) ⊗ 1 + 1 ⊗(cid:16)vivj − vδi+δj(cid:17) .
+vi ⊗ vj + vj ⊗ vi − vδi ⊗ vδj − vδj ⊗ vδi
So vivj − vδi+δj ∈ P rim(S(g)) = g. So:
vδk
δi,δj
vδk = vδi+δj + vi ⋆ vj.
Hence:
vivj = vδi+δj +Xk∈I
g(vi(vjvk)) = g Xc∈SΛ
= g
Xc∈SΛ
l(c)≥2
cc
δj ,δk
vivc!
cc
δj ,δk
∈I, left ideal
vivc
{z}
= 0 + vi ⋆ (vj ⋆ vk)
= g((vivj)vk)
= g(vδi+δj vk) + g((vi ⋆ vj)vk)
= g(vδi+δj vk) + (vi ⋆ vj) ⋆ vk.
8
+ g(vi(vj ⋆ vk))
So:
vi ⋆ (vj ⋆ vk) − (vi ⋆ vj) ⋆ vk = g(vδi+δj vk) = vj ⋆ (vi ⋆ vk) − (vj ⋆ vi) ⋆ vk.
Hence, the product ⋆ is prelie.
✷
1.4 Prelie products on a Lie algebra
Proposition 10 The applications Φg and Ψg are inverse bijections.
Proof. Let ⋆ ∈ PL(g). We put I = Φg(⋆) and • = Ψg(I). Let πg be the canonical surjection
on g in S(g). Then, as ξ⋆(I) = S≥2(g), ξ⋆ ◦ g = πg ◦ ξ⋆. So, for all x, y ∈ g:
ξ⋆(x • y) = ξ⋆ ◦ g(xy) = πg ◦ ξ⋆(xy) = πg(x ∗ y) = πg(xy + x ⋆ y) = x ⋆ y = ξ⋆(x ⋆ y).
As ξ⋆ is monic, x • y = x ⋆ x, so Ψg ◦ Φg(⋆) = ⋆.
Let I ∈ LI(g). We put ⋆ = Ψg(I). We have to prove that ξ−1
⋆ (S≥2(g)) = I, that is to say
ξ⋆(I) = S≥2(g). Because they are both complements of (1) ⊕ g and ξ⋆ is bijective, it is enough to
prove ξ⋆(I) ⊆ S≥2(g). With the preceding notations, let va ∈ I, l(a) ≥ 2, and let us prove that
ξ⋆(va) ∈ S≥2(g). If l(a) = 2, we put a = δi + δj. Then we saw that va = vivj − vi ⋆ vj if i 6= j. If
i = j, in the same way, 2va = vivj − vi ⋆ vj. So, up to a non-zero mutiplicative constant λ:
ξ⋆(va) = λ(vi ∗ vj − vi ⋆ vj) = λ(vivj + vi ⋆ vj − vi ⋆ vj) = λvivj ∈ S≥2(g).
If l(a) ≥ 3, we put a = a′ +δi for a well-chosen i, with l(a′) ≥ 2. Then, there exists a non-zero
). So, as I is a left ideal, va − λviva′ ∈ g ∩ I = (0),
) belongs to S≥2(g),
✷
constant λ such that, in U (g), ∆(va) = λ(viva′
. Then ξ⋆(va) = vi ∗ φ(va′
so va = λviva′
left ideal for ∗, so va belongs to S≥2(g).
). By the induction hypothesis, φ(va′
As a conclusion:
Theorem 11 Let g be a Lie algebra. There exists a prelie product on g inducing its Lie
bracket, if, and only if, there exists a left ideal I of U (g) such that U+(g) = g ⊕ I. More precisely,
there exists a bijection between the sets:
• PL(g) = {⋆ ⋆ is a prelie product on g inducing the bracket of g}.
• LI(g) = {I left ideal of U (g) U+(g) = g ⊕ I}.
It is given by Ψg : LI(g) −→ PL(g), associating to a left ideal I the prelie product defined by
x ⋆ y = g(xy), where g is the canonical projection on g in the direct sum U+(g) = g ⊕ I. The
inverse bijection is given by Φg : PL(g) −→ LI(g), associating to a prelie product ⋆ the left ideal
U (g)V ect(xy − x ⋆ y, x, y ∈ g).
Proof. It only remains to prove that Φg(⋆) is generated by the elements xy − x ⋆ y, x, y ∈ g.
Using the isomorphism ξ⋆, it is equivalent to prove that the left ideal S≥2(V ) of (S(V ), ∗) is
generated by the elements x ∗ y − x ⋆ y, x, y ∈ g. By definition of ∗, for all x, y ∈ g:
x ∗ y − x ⋆ y = xy + x ⋆ y − x ⋆ y = xy,
so it is equivalent to prove that the left ideal S≥2(V ) is generated by S2(V ). Let us denote by
J the left ideal generated by S2(V ). As S≥2(V ) is a left ideal, J ⊆ S≥2(V ). Let v1, . . . , vn ∈ g,
with n ≥ 2. Let us prove that v1 . . . vn ∈ J by induction on n. This is obvious if n = 2. If n ≥ 3:
v1 ∗ (v2 . . . vn) = v1 . . . vn + v1 ⋆ (v2 . . . vn).
By lemma 5, v1 ⋆ (v2 . . . vn) ∈ Sn−1(V ). By the induction hypothesis, as n ≥ 3, v1 ∗ (v2 . . . vn)
and v1 ⋆ (v2 . . . vn) belong to J, so v1 . . . vn ∈ J. As a conclusion, S≥2(V ) = J.
✷
9
Corollary 12 Let g be a Lie algebra. There exists an associative product on g inducing its
Lie bracket, if, and only if, there exists a bilateral ideal I of U (g) such that U+(g) = g ⊕ I. More
precisely, there exists a bijection between the sets:
1. AS(g) = {⋆ ⋆ is an associative product on g inducing the bracket of g}.
2. BI(g) = {I bilateral ideal of U (g) / U+(g) = g ⊕ I}.
It is given by Ψg : LI(g) −→ PL(g), associating to a left ideal I the associative product defined
by x ⋆ y = g(xy), where g is the canonical projection on g in the direct sum U+(g) = g ⊕ I.
Proof.
⋆ ∈ AS(g). By proposition 6, S≥2(g) is a bilateral ideal, so is Φg(⋆) = ξ−1
I ∈ AS gr(g)). Then g ≈ U+(g)/I inherits an associative product, which is ⋆ = Ψg(I).
It is enough to verify that Ψg(BI(g)) ⊆ AS(g) and Φg(AS(g)) ⊆ BI(g). Let
⋆ (S≥2(g)). Let
✷
1.5 Good and very good pointed modules
Definition 13 Let g be a Lie algebra.
1. A pointed g-module is a couple (M, m), where M is a g-module and m ∈ M .
2. Let (M, m) be a pointed g-module. The application Υm is defined by:
Υm :(cid:26) g −→ M
x −→ x.m.
3. A pointed g-module (M, m) is good if the following assertions hold:
• Υm is injective.
• Im(Υm) is a submodule of M which does not contains m.
• M/Im(Υm) ≈ K (trivial g-module) as a g-module.
4. A pointed g-module (M, m) is very good if Υm is bijective.
Note that all good pointed g-modules have the same dimension, so we can consider the set
GM(g) of isomorphism classes of good pointed g-modules. Similarly, we can consider the set
VGM(g) of very good pointed g-modules.
Remark. If (M, m) is good, then it is cyclic, generated by m. Moreover, M/Im(Υm) is
one-dimensional, trivial, generated by m = m + Im(Υm).
Theorem 14 Let g be a Lie algebra. The following application is a bijection:
Θ :(cid:26) LI(g) −→ GM(g)
I −→ (U (g)/I, 1).
Proof. Let us first proove that Θ is well-defined. As I ∈ LI(g), as a vector space
U (g)/I = (1) ⊕ g. For all x, y ∈ g, xy ∈ U+(g), so x.y ∈ U+(g)/I = g in U (g)/I, as
I ⊆ U+(g). As a conclusion, the subspace g of U (g)/I is a submodule. Moreover, Υ1(x) = x for
all x ∈ g. So Υ1 is injective, and its image is the submodule g, so does not contain 1. Finally,
(U (g)/I)/Im(Υ1) ≈ U (g)/U+(g) ≈ K as a g-module. So (U (g)/I, 1) is a good pointed g-module.
Let us consider:
Θ′ :(cid:26) GM(g) −→ LI(g)
(M, m) −→ Ann(m) = {x ∈ U (g) x.m = 0}.
10
Let us prove that Θ′ is well-defined. Let (M, m) be a good pointed g-module. Then Ann(m)
is a left-ideal of U (g). Let x ∈ Ann(m). Then x.m = 0, so x.m = 0 in M/Im(Υm), so
x.m = ε(x)m = 0. As a conclusion, ε(x) = 0, so Ann(m) ⊆ U+(g). Let us now show that
U+(g) = g ⊕ Ann(m). First, if x ∈ g ∩ Ann(m), then Υm(x) = x.m = 0. As Υm is monic,
x = 0. If y ∈ U+(g), then y.m = ε(y)m = 0 in M/Im(Υm), so y.m ∈ Im(Υm): there exists
x ∈ g, such that y.m = x.m. Hence, y − x ∈ Ann(m), so y = x + (y − x) ∈ g + Ann(m). Finally,
Ann(m) ∈ LI(g).
Moreover, if (M, m) and (M ′, m′) are isomorphic (that is to say there is an isomorphism of
g-modules from M to M ′ sending m to m′), then Ann(m) = Ann(m′). So Θ′ is well-defined.
Let (M, m) be a good pointed g-module. Then Θ ◦ Θ′(M, m) = (U (g)/Ann(m), 1) ≈ (M, m),
as M is cyclic, generated by m. Let now I ∈ LI(g). Then Θ′ ◦ Θ(I) is the annihilator of 1 in
U (g)/I, so is equal to I. As a conclusion, Θ and Θ′ are inverse bijections.
✷
Definition 15 The set GM′(g) is the set of isomorphism classes of couple ((M, m), V ),
where:
• (M, m) is a good pointed g-module.
• V is a submodule of M such that M = V ⊕ Im(Υm).
Proposition 16 For any Lie algebra g, GM′(g) is in bijection with VGM(g).
Proof. Let ((M, m), V ) be an element of GM′(g). As M/Im(Υm) ≈ K, the complement V
of Im(Υm) is trivial and one-dimensional. As m /∈ Im(Υm), V admits a unique element m′ =
m+x.m, where x ∈ g. Let us show that (Im(Υm), −x.m) is very good. Let m′′ ∈ Im(Υm). There
exists y ∈ g, such that y.m = m′′. Then y.m′ = ε(y)m′ = 0 = y.m + y.(x.m) = m′′ + y.(x.m),
so m′′ = y.(−x.m) = Υ−x.m(y). Hence, Υ−x.m is epic.
Let us assume that Υ−x.m(y) = 0. Then y.m′ = ε(y)m′ = 0 = y.m+y.(x.m) = y.m = Υm(x).
As Υm is monic, y = 0, so Υ−x.m is also monic.
Moreover, if ((M, m), V ) ≈ ((M ′, m′), V ′), then (Im(Υm), −x.m) and (Im(Υm′, −x′.m′) are
isomorphic, so the following application is well-defined:
Λ :(cid:26)
GM′(g) −→ VGM(g)
((M, m), V ) −→ (Im(Υm), −x.m).
Let now (M, m) be a very good pointed g-module. Let us prove that ((K ⊕ M, 1 + m), K)
is an element of GM′(g). First, for all x ∈ g, Υ1+m(x) = ε(x)1 + x + m = 0 + Υm(x), so
Υ1+m = Υm. As a consequence, Im(Υ1+m) = M is a submodule which does not contains 1 + m,
Υ1+m is monic, and K is a complement of Im(Υ1+m). Hence, ((K ⊕ M, 1 + m), K) ∈ GM′(g).
Moreover, if (M, m) ≈ (M ′, m′), then ((K ⊕ M, 1 + m), K) ≈ ((K ⊕ M ′, 1 + m′), K), so the
following application is well-defined:
Λ′ :(cid:26) VGM(g) −→ GM′(g)
(M, m) −→ ((K ⊕ M, 1 + m), K).
Let (M, m) ∈ VGM(g). Then Λ ◦ Λ′((M, m)) = (M, m). Let ((M, m), V ) ∈ GM′(g). Then
the following application is an isomorphism of pointed g-modules and sends K to V :
(cid:26) (K ⊕ Im(Υm), 1 − x.m) −→ M
λ + m′′ −→ λm′ + m′′.
So Λ and Λ′ are inverse bijections.
✷
11
2 Are the complex simple Lie algebras prelie?
The aim of this section is to prove that the complex finite-dimensional simple Lie algebras are
not prelie. In this section, the base field is the field of complex C. We shall use the notations
of [4] in the whole section. In particular, if g is a simple complex finite-dimensional Lie algebra,
Γ(a1,...,an) is the simple g-module of hightest weight (a1, . . . , an).
2.1 General results
Proposition 17 Let g be a finite-dimensional semi-simple Lie algebra. Then there exists a
prelie product on g inducing its bracket if, and only if, there exists a very good pointed g-module.
Proof. As the category of finite-dimensional modules over g is semi-simple, for any good
pointed g-module (M, m), there exists a V such that ((M, m), V ) ∈ GM(g). Using the bijections
of the preceding section, PL(g) is non-empty if, and only if, LI(g) is non-empty, if, and only
if, the set GM(g) is non-empty, if, and only if, the set GM′(g) is non-empty, if, and only if,
VGM(g) is non-empty.
✷
Let g be a finite-dimensional complex simple Lie algebra. In order to prove that g is not
prelie, we have to prove that g has no very good pointed modules.
Lemma 18 Let (M, m) be a very good g-module and let M ′ be a submodule of M . There
exists m′ ∈ M , such that the following application is epic:
Υm′ :(cid:26) M ′ −→ M ′
x −→ x.m′.
Proof. As g is semi-simple, let M ′′ be a complement of M ′ in M and let m = m′ + m′′ the
decomposition of m in M = M ′ ⊕ M ′′. Let u ∈ M ′. As Υm is bijective, there exists a (unique)
x ∈ g, such that Υm(x) = u. So u + 0 = x.m′ + xm′′ in the direct sum M = M ′ ⊕ M ′′. Hence,
Υm′(x) = u and Υm′ is epic.
✷
Corollary 19 Let (M, m) be a very good g-module. Then M is not isomorphic to (g, ad)
and its trivial component is (0).
Proof. First, note that for all x ∈ g, ((g, ad), x) is not very good: it is obvious if x = 0 and
if x 6= 0, Υx(x) = [x, x] = 0, so Υx is not monic.
If (M, m) is very good, let us consider its trivial component M ′. By the preceding lemma,
there exists m′ ∈ M ′, such that Υm′ : M ′ −→ M ′ is epic. As M ′ is trivial, Υm′ = 0, and finally
M ′ = (0).
✷
2.2 Representations of small dimension
We first give the representations of g with dimension smaller than the dimension of g. We shall
say that a g-module M is small if:
• M is simple.
• The dimension of M is smaller than the dimension of g.
• M is not trivial and not isomorphic to (g, ad).
Remark. By corollary 19, any very good module is the direct sum of small modules.
Let n be the rank of g and let Γ(a1,...,an) be the simple g-module of highest weight (a1, . . . , an).
The dimension of Γ(a1,...,an) is given in [4] in chapter 15 for sln+1, (formula 15.17), in chapter 24
12
for sp2n, so2n and so2n+1 (exercises 24.20, 24.30 and 24.42). It turns out from these formulas
that if b1 ≤ a1, . . . bn ≤ an, then dim(Γ(b1,...,bn)) ≤ dim(Γ(a1,...,an)), with equality if, and only
if, (b1, . . . , bn) = (a1, . . . , an). Direct computations of dim(Γ(0,...,0,3,0,...,0)) then shows that it is
enough to compute the dimensions of Γ(a1,...,an) with all a1 ≤ 2. With the help of a computer,
we obtain:
g
sln, n ≥ 9
dim(g)
n2 − 1
sln, 3 ≤ n ≤ 8
n2 − 1
sl2
3
sp2n, n = 2 or n ≥ 4
n2n + 1
sp6
21
so2n, n ≥ 8
so2n, 4 ≤ n ≤ 7
n(2n − 1)
n(2n − 1)
so6
15
so2n+1, n ≥ 7
so2n+1, 2 ≤ n ≤ 6
n2n + 1
n2n + 1
e6
e7
e8
f4
g2
78
133
78
52
14
small modules V
Γ(1,0,...,0), Γ(0,...,0,1)
Γ(0,1,0,...,0), Γ(0,...,0,1,0)
Γ(2,0,...,0), Γ(0,...,0,2)
Γ(0,...,0,1,...,0), the 1 in position i
Γ(2,0,...,0), Γ(0,...,0,2)
Γ1
Γ(1,0,...,0)
Γ(0,1,0,...,0)
Γ(1,0,0)
Γ(0,1,0), Γ(0,0,1)
Γ(1,0,...,0)
Γ(1,0,...,0)
Γ(0,0,1,0,...,0), Γ(0,...,0,1)
Γ(1,0,0)
Γ(0,1,0), Γ(0,0,1)
Γ(0,2,0), Γ(0,0,2)
Γ(1,0,...,0)
Γ(1,0,...,0)
Γ(0,...,0,1)
Γ(1,0,0,0,0,0)
Γ(1,0,0,0,0,0,0)
×
Γ(1,0,0,0)
Γ(1,0)
dim(V )
n
n(n − 1)/2
(n + 1)/2
(n + 1)/2
(cid:0)n
i(cid:1)
2
2n
2n + 1(n − 1)
6
14
2n
2n
2n−1
6
4
10
2n + 1
2n + 1
2n
27
56
×
26
7
2.3 Proof in the generic cases
Proposition 20 Let g be sln, with n 6= 6, or so2n, with n ≥ 2, or sp2n, with n ≥ 2, or en,
with 6 ≤ n ≤ 8. Then g is not prelie.
Proof. Let g be a simple finite-dimensional Lie algebra, with a prelie product inducing its
Lie bracket. From the preceding results, g has a very good pointed g-module (M, m), and the
submodules appearing in the decomposition of M into simples modules are all small. Let us
show that it is not possible in these different cases.
First case. If g = sln, with n ≥ 9, then the decomposition of a very good pointed g-module
into simples would have a submodules of dimension n, b submodules of dimensions n(n − 1)/2
and c submodules of dimension n(n + 1)/2. So:
n2 − 1 = an + b
n(n − 1)
2
+ c
n(n + 1)
2
.
Let us assume that n has a prime factor p 6= 2. Then p n, p n(n ± 1) and 2 n(n ± 1), so
p n(n±1)
. As a consequence, p n2 − 1, so p 1: this is absurd. So n = 2k for a certain k ≥ 4,
as n ≥ 9. Then:
2
a2k + b2k−1(2k − 1) + c2k−1(2k + 1) = 22k − 1,
13
so 2k−1 22k − 1 and k = 1: contradiction.
Second case. Similarly, if g = sp2n with n 6= 3, we would have a, b ∈ N such that a2n +
b2n + 1(n − 1) = n2n + 1. If b ≥ 2, then n2n + 1 ≥ 22n + 1(n − 1), so 2n + 1(n − 2) ≤ 0 and
2 : absurd. If b = 1, then a = 1 + 1
n ≤ 2: contradiction. So b = 0 or 1. If b = 0, then a = n + 1
2n :
absurd.
Third case. If g = so2n, with n ≥ 8, then we would have 2n n(2n − 1), so n − 1
2 ∈ N: absurd.
Other cases. Let us sum up the possible dimensions of the small modules in an array:
g
sl2
sl3
sl4
sl5
sl7
sl8
so6
so8
so10
so12
so14
sp6
e6
e7
e8
dimensions of the small modules dim(g)
2
3, 6
4, 6, 10
5, 10, 15
7, 21, 28, 35
8, 28, 36, 56
4, 6, 10
8
10, 16
12, 32
14, 64
6, 14
27
56
×
3
8
15
24
48
63
15
28
45
66
91
21
78
133
248
In all cases except so6, (there is a prime integer which divides all the dimensions of the small
submodules but not the dimension of g), or (there is a unique dimension of small modules and it
does not divide the dimension of g), or (there is no small modules). So g is not prelie. For so12,
it comes from the fact that 4 divides 12 and 32 but divides not 66.
So none of these Lie algebras is prelie.
✷
2.4 Case of so2n+1
We assume that n ≥ 2. Let us recall that:
so2n+1 =
A B −tF
C −tA −tE
E
F
0
B, C skew symmetric
,
where A, B, C are n × n matrices, E, F are 1 × n matrices.
Lemma 21 Let (M, m) be a very good pointed module over g = so2n+1. Then M is isomor-
phic to M2n+1,n(C) as a g-module, with the action of g given by the (left) matricial product.
Proof. Let us first assume that n ≥ 7. Then g has a unique small module of dimension 2n+1,
which is the standard representation C2n+1. So if M is a direct sum of copies of this module;
comparing the dimension, there are necessarily n copies of this module, so M ≈ M2n+1,n(C).
14
If n ≤ 6, let us sum up the possible dimensions of the small modules in an array:
g
so5
so7
so9
so11
so13
dimensions of the small modules dim(g)
5, 4
7, 8
9, 16
11, 32
13, 64
10
21
36
55
78
In all these cases, the only possible decomposition of M is n copies of the standard representation
of dimension 2n + 1, so the conclusion also holds.
✷
The elements of M2n+1,n(C) will be written as (cid:16)X
Proposition 22 For any m =(cid:16)X
Y
Y
Proof. Let us denote:
z(cid:17), where X, Y ∈ Mn(C) and z ∈ M1,n(C).
z(cid:17) ∈ M2n+1,n(C), (M2n+1,n(C), m) is not very good.
M = {A ∈ M2n+1,n((C) (M2n+1,n(C), A) is very good}.
Let us recall that:
SO2n+1 =
B ∈ M2n+1,2n+1(C) tB
0
0 I
I
0 0
0 0 I
B =
0
0 I
I
0 0
0 0 I
It is well-known that if B ∈ SO2n+1 and A ∈ so2n+1, then BAB−1 ∈ so2n+1.
.
First step. Let us prove that m ∈ M if, and only if, Bm ∈ M for all B ∈ SO2n+1. Let us
assume that m ∈ M and let B ∈ SO2n+1. Let us take A ∈ Ker(ΥBm). Then ABm = 0, so
B−1ABm = 0 and B−1AB ∈ Ker(Υm). As m ∈ M, Υm is bijective, so A = 0: ΥBm is monic.
As M and so2n+1 have the same dimension, ΥBm is bijective.
Second step. Let us prove that m ∈ M if, and only if, mP ∈ M for all P ∈ GL(n). Indeed,
m −→ mP is an isomorphism of g-modules for all P ∈ GL(n).
Third step. Let us prove that(cid:16)X
Indeed, if (cid:16)X
Y
Y
z(cid:17) ∈ M and if P, Q ∈ GL(n), by the first and second steps:
z(cid:17) ∈ M if, and only if,(cid:18) QXP
zP (cid:19) ∈ M for all P, Q ∈ GL(n).
tQ−1Y P
0
Q
0
0 tQ−1 0
1
0
0
X
Y
z
P ∈ M,
as the left-multiplying matrix is an element of SO2n+1.
Y
tQ−1Y P
only a 1 in position (i, j): this is a contradiction. So X is invertible.
good choice of P and Q, QXP has its first row and first column equal to 0. By the third step,
Fourth step. Let (cid:16)X
m′ =(cid:18) QXP
Last step. Let us assume that M is not empty and let m =(cid:16)X
For a good choice of P and Q, we obtain an element m′ = (cid:16)In
z(cid:17) ∈ M, let us prove that X is invertible. If X is not invertible, for a
zP (cid:19) ∈ M, but Υm′(En+1,2 − En+2,1) = 0, where Ei,j is the elementary matrix with
z(cid:17) ∈ M. Then X is invertible.
z′(cid:17) ∈ M. Let us then choose a
Y ′
Y
15
non-zero skew-symmetric matrix B ∈ Mn(C) (this exists as n ≥ 2), then the following element
is in so2n+1:
A =
∈ so2n+1.
B
−BY
0
tY BY −tY B 0
0
0
0
An easy computation shows that A.m′ = 0, so A ∈ Ker(Υm′): contradiction, m′ /∈ M. So M is
empty.
✷
Corollary 23 For all n ≥ 2, so2n+1 is not prelie.
2.5 Proof for sl6 and g2
The Lie algebra sl6 has dimension 35, and the possible dimensions of its small modules are 6, 15,
20. So if (M, m) is a very good pointed module, M is the direct sum of a simple of dimension 15
and a simple of dimension 20. As there is only one simple of dimension 20, that is to say Λ3(V )
where V is the standard representation, M contains Λ3(V ). By lemma 18, in order to prove that
sl6 is not prelie, it is enough to prove that for any m ∈ Λ3(V ), Υm is not epic. This essentially
consists to show that the rank of a certain 20 × 35 matrix is not 20 and this can be done directly
using MuPAD pro 4, see section 4.2.
The proof for g2 is similar: if (M, m) is a very good module, then M ≈ V ⊕ V , where V is
the only small module of g, that is to say its standard representation. So M ≈ M7,2(C) as a
g-module. It remains to show that for any m, (M7,2(C), m) is not very good. This essentially
consists to show that a certain 14 × 14 matrix is not invertible and this can be done directly
using MuPAD pro 4, see section 4.3.
The proof for f4 would be similar: if (M, m) is a very good module, then M ≈ V ⊕ V , where
V is the only small module of g, that is to say its standard representation. So M ≈ M26,2(C)
as a f4-module. It would remain to show that for any m, (M26,2(C), m) is not very good. This
would consist to show that a certain 52 × 52 matrix is not invertible.
We finally prove:
Theorem 24 Let g be a simple, finite-dimensional complex Lie algebra. If it is not isomor-
phic to f4, it is not prelie.
We conjecture:
Theorem 25 Let g be a simple, finite-dimensional complex Lie algebra. Then it is not prelie.
Remark. As a corollary, we obtain that g is not associative. This result is also proved
in a different way in [7], with the help of compatible products. Indeed, we have the following
equivalences:
the prelie product ⋆ is compatible
⇐⇒ ∀x, y, z ∈ g, [x, y ⋆ z] = [x, y] ⋆ z + y ⋆ [x, z]
⇐⇒ ∀x, y, z ∈ g, x ⋆ (y ⋆ z) − (y ⋆ z) ⋆ x − (x ⋆ y) ⋆ z + (y ⋆ x) ⋆ z − y ⋆ (x ⋆ z) + y ⋆ (z ⋆ x)
⇐⇒ ∀x, y, z ∈ g, −(y ⋆ z) ⋆ x − +y ⋆ (z ⋆ x) = 0
⇐⇒ ⋆ is associative.
As from [7], the only admissible associative product on g is 0, g is not associative.
16
3 Dendriform products on cofree coalgebra
3.1 Preliminaries and results on tensor coalgebras
Let V be a vector space. The tensor algebra T (V ) has a coassociative coproduct, given for all
v1, . . . , vn ∈ V by:
v1 . . . vi ⊗ vi+1 . . . vn.
∆(v1 . . . vn) =
Let us recall the following facts:
n
Xi=0
1. Let us fix a basis (vi)i∈I of V . We define T I as the set of words w in letters the elements
of I. For an element w = i1 . . . ik ∈ SΛ, we put l(w) = k, and:
vw =Yi∈I
vi1 . . . vik .
Then (vw)w∈T I is a basis of T (V ), and the coproduct is given by:
∆(vw) = Xw1w2=w
vw1 ⊗ vw2.
2. Let T+(V ) be the augmentation ideal of T (V ). It is given a coassociative, non counitary
coproduct ∆ defined by ∆(x) = ∆(x) − x ⊗ 1 − 1 ⊗ x for all x ∈ T+(V ). In other terms,
putting T+I = T I − {∅}, (vw)w∈T+I is a basis of T+(V ) and:
∆(vw) = Xw1w2=w
w1,w2∈T+I
vw1 ⊗ vw2.
3. Let ∆(n) : T+(V ) −→ T+(V )⊗(n+1) be the n-th iterated coproduct of T+(V ). Then:
Ker(cid:16) ∆(n)(cid:17) =
n
Mk=1
T k(V ).
In particular, P rim(T (V )) = V .
4. (T n(V ))n∈N is a gradation of the coalgebra T (V ).
Lemma 26 In T+(V ) ⊗ T+(V ), Ker( ∆ ⊗ Id − Id ⊗ ∆) = Im( ∆).
Proof. ⊇ comes from the coassociativity of ∆.
⊆: let us consider X = Xw1,w2∈T+I
xw1,w2vw1 ⊗ vw2 ∈ Ker( ∆ ⊗ Id − Id ⊗ ∆) = Im( ∆). Then:
( ∆ ⊗ Id) ◦ ∆(X) = Xw1,w2,w3∈T+I
= (Id ⊗ ∆) ◦ ∆(X) = Xw1,w2,w3∈T+I
xw1w2,w3vw1 ⊗ vw2 ⊗ vw3
xw1,w2w3vw1 ⊗ vw2 ⊗ vw3.
So, for all w1, w3 ∈ T+I, w2 ∈ T I, xw1w2,w3 = xw1,w2w3. In particular, xi1...ik,j1...jk = xi1,i2...ikj1...jl
for all i1, . . . , ik, j1, . . . , jl ∈ I. We denote by xi1...ikj1...jl this common value. Then:
∆(X) = Xw∈T+I
xw Xw1w2=w
w1,w2∈SΛ+
vw1 ⊗ vw2 = Xw∈T+I
xw ∆(xw).
So X ∈ Im( ∆).
✷
17
Lemma 27 Let W be a subspace of T+(V ), such that T (V ) = (1) ⊕ V ⊕ W . There exists a
coalgebra endomorphism φ of T (V ), such that φV = IdV and φ
Mn≥2
φ is an automorphism.
T n(V )
= W . Moreover,
Proof. Similar to the proof of lemma 3.
✷
Corollary 28 Let C be a cofree coalgebra and let W ⊂ C+, such that C = (1)⊕P rim(C)⊕W .
There exists a unique coalgebra isomorphism φ : T (P rim(C)) −→ C such that φP rim(C) =
IdP rim(C) and φ(T≥2(P rim(C)) = W .
3.2 Left ideal associated to a dendriform Hopf algebra
Let A be a dendriform Hopf algebra [8, 10], that is to say the product (denoted by ∗) of A can
be split on A+ as ∗ =≺ + ≻, with:
1. For all x, y, z ∈ A+:
2. For all a, b ∈ A+:
(x ≺ y) ≺ z = x ≺ (y ∗ z),
(x ≻ y) ≺ z = x ≻ (y ≺ z),
(x ∗ y) ≻ z = (x ≻ y) ≻ z.
(1)
(2)
(3)
∆(a ≺ b) = a′ ∗ b′ ⊗ a′′ ≺ b′′ + a′ ∗ b ⊗ a′′ + b′ ⊗ a ≺ b′′ + a′ ⊗ a′′ ≺ b + b ⊗ a,
∆(a ≻ b) = a′ ∗ b′ ⊗ a′′ ≻ b′′ + a ∗ b′ ⊗ b′′ + b′ ⊗ a ≻ b′′ + a′ ⊗ a′′ ≻ b + a ⊗ b.
(4)
(5)
We used the following notations: A+ is the augmentation ideal of A and ∆ : A+ −→
A+ ⊗ A+ is the coassociative coproduct defined by ∆(a) = ∆(a) − a ⊗ 1 − 1 ⊗ a for all
a ∈ A+.
Proposition 29 Let A be a dendriform Hopf algebra. Then the following application is a
monomorphism of coalgebras:
ΘA :(cid:26) T (P rim(A)) −→ A
v1 . . . vn −→ vn ≺ (vn−1 ≺ (. . . ≺ (v2 ≺ v1) . . .)
Moreover, if A is connected as a coalgebra, then ΘA is an isomorphism, so A is a cofree coalgebra.
Proof. For all v1, . . . , vn ∈ P rim(A), we put:
ω(v1, . . . , vn) = vn ≺ (vn−1 ≺ (. . . ≺ (v2 ≺ v1) . . .).
An easy induction using (4) proves that:
∆(ω(v1, . . . , vn)) =
n
Xi=0
ω(v1, . . . , vi) ⊗ ω(vi+1, . . . , vn).
So ΘA is a morphism of coalgebras. Let us prove that this morphism is injective. If not, its kernel
would contain primitive elements of T (P rim(A)), that is to say elements of P rim(A): absurd.
If A is connected, this morphism is surjective: let us take x ∈ A, let us prove that x ∈ Im(Θ).
As A is connected, for all x ∈ A, there exists n ≥ 1 such that ∆(n)(x) = 0. Let us proceed by
induction on n. If n = 1, then x ∈ g ⊆ Im(Θ). If n ≥ 2, then ∆(n−1)(x) ∈ g⊗n. Let us put:
∆(n−1)(x) = Xi1,...,in∈I
ai1...invi1 ⊗ . . . ⊗ vin.
18
Then:
∆(n−1)(x) = ∆(n−1)
Xi1,...,in∈I
ai1...invi1...in
.
By the induction hypothesis, x − Xi1,...,in∈I
Θ is an isomorphism, so (vw)w∈T I is a basis of A.
ai1...invi1...in ∈ Im(Θ), so x ∈ Im(Θ). As a conclusion,
✷
Proposition 30 Let A be a dendriform Hopf algebra, connected as a coalgebra. Let us put
A≺2 = P rim(A) ≺ A+. Then:
1. A+ = P rim(A) ⊕ A≺2.
2. A≺2 = A+ ≺ A+.
3. A≺2 is a left ideal of A.
Proof.
1. Using ΘA:
ΘA(P rim(A)) = P rim(A), ΘA
Mn≥2
P rim(A)⊗n
= A≺2, ΘA
Mn≥1
P rim(A)⊗n
= A+.
As ΘA is an isomorphism, we obtain the result.
2. ⊆. As P rim(A) ⊆ A+, A≺2 ⊆ A+ ≺ A+.
⊇. Note that A+ = V ect (ω(v1, . . . , vn) n ≥ 1, v1, . . . , vn ∈ P rim(A)). Let x = ω(v1, . . . , vm)
and y = ω(w1, . . . , wn) ∈ A+. We put x′ = ω(v1, . . . , vm−1). Then, by (1):
x ≺ y = (vm ≺ x′) ≺ y = vm ≺ (x′ ∗ y).
As x′ ∗ y ∈ A+, x ≺ y ∈ A≺2.
3. It is enough to prove that A+ ∗ A≺2 ⊆ A≺2. Let x, y, z ∈ A+.
x ∗ (y ≺ z) = x ≺ (y ≺ z) + (x ≻ y) ≺ z ∈ A+ ≺ A+ = A≺2.
So A≺2 is a left ideal of A.
✷
3.3 Dendriform products on a tensorial coalgebra
Let us now consider an associative product ∗ on the coalgebra T (V ), such that:
1. (T (V ), ∗, ∆) is a Hopf algebra.
2. T (V )≥2 =Mn≥2
T n(V ) is a left ideal of (T (V ), ∗).
Proposition 31 We define a product ≺ on T+(V ) in the following way: for all v, v1, . . . , vn ∈
V , w ∈ T+(V ),
(cid:26)
v ≺ w = wv,
(v1 . . . vn) ≺ w = ((v1 . . . vn−1) ∗ w)vn.
We also put ≻= ∗− ≺. Then (T (V ), ≺, ≻, ∆) is a dendriform Hopf algebra.
19
Proof. Let us first prove (1). Let u1, . . . , uk, v1, . . . , vl, w1, . . . , wm ∈ V .
(u1 . . . uk ≺ v1 . . . vl) ≺ w1 . . . wm = ((u1 . . . uk−1 ∗ v1 . . . vl)uk) ≺ w1 . . . wm
= (u1 . . . uk−1 ∗ v1 . . . vl ∗ w1 . . . wm)uk
= u1 . . . uk ≺ (v1 . . . vl ∗ w1 . . . wm).
Let us now prove (4). We take a = u1 . . . uk, b = v1 . . . vl ∈ T+(V ). We denote a = u1 . . . uk−1.
∆(a ≺ b) = ∆((a ∗ b)uk)
= a ∗ b ⊗ uk + ∆(a ∗ b)(1 ⊗ uk)
= a ∗ b ⊗ uk + a ⊗ buk + b ⊗ auk + a′ ∗ b ⊗ a′′uk
+a′ ⊗ (a′′ ∗ b)uk + a ∗ b′ ⊗ b′′uk + b′ ⊗ (a ∗ b′′)uk + a′ ∗ b′ ⊗ (a′′ ∗ b′′)uk.
Moreover, using (1):
a′ ∗ b′ ⊗ a′′ ≺ b′′ = a ∗ b′ ⊗ b′′uk + a′ ∗ b′ ⊗ (a′′ ∗ b′′)uk,
a′ ∗ b ⊗ a′′ = a ∗ b ⊗ uk + a′ ∗ b ⊗ a′′uk,
b′ ⊗ a ≺ b′′ = b′ ⊗ (a ∗ b′′)uk,
a′ ⊗ a′′ ≺ b = a ⊗ buk + a′ ⊗ (a′′ ∗ b)uk,
b ⊗ a = b ⊗ auk.
So (4) is satisfied. As T (V ) is a Hopf algebra, (4)+(5) is satisfied, so (5) also is.
Let us prove (2). For all x, y, z ∈ T+(V ), we put φ(x, y, z) = (x ≻ y) ≺ z − x ≻ (y ≺ z). A
direct computation using (4) and (5) shows that:
∆(φ(x, y, z)) = x′y′z′ ⊗ Φ(x′′, y′′, z′′) + y′z′ ⊗ Φ(x, y′′, z′′) + x′z′ ⊗ Φ(x′′, y, z′′)
+x′y′ ⊗ Φ(x′′, y′′, z) + z′ ⊗ Φ(x, y, z′′) + y′ ⊗ Φ(x, y′′, z) + x′ ⊗ Φ(x′′, y, z).
Moreover:
φ(x, y, z) = (x ≻ y) ≺ z − x ∗ (y ≺ z) + x ≺ (y ≺ z).
By definition of ≺, (x ≻ y) ≺ z and x ≺ (y ≺ z) ∈ T≥2(V ). In the same way, y ≺ z ∈ T≥2(V ),
left ideal of T (V ), so x ∗ (y ≺ z) ∈ T≥2(V ). finally, φ(x, y, z) ∈ T≥2(V ).
Let us now prove that φ(x, y, z) = 0 by induction on n = l(x) + l(y) + l(z).
If n = 3,
then x, y, z ∈ V , so are primitive. So ∆(φ(x, y, z)) = 0, and φ(x, y, z) ∈ V . By the preced-
ing point, φ(x, y, z) ∈ V ∩ T≥2(V ) = (0), so φ(x, y, z) = 0. Let us assume the result at all
rank < n. By the induction hypothesis applied to x′, y′, z′ and others, ∆(φ(x, y, z)) = 0, so
φ(x, y, z) ∈ V ∩ T≥2(V ) = (0). Hence, (2) is satisfied. As ∗ is associative, (1)+(2)+(3) is satis-
fied, so (3) also is.
✷
Remark. In the dendriform Hopf algebra T (V ), T (V )≺2 = T (V )≥2.
By the dendriform Cartier-Quillen-Milnor-Moore theorem, T (V ) is now the dendriform en-
veloping algebra of the brace algebra V = P rim(T (V )). By [12], the brace structure on V
induced by the dendriform structure of A is given, for all a1, . . . , an ∈ V , by:
(−1)n−1−i(a1 ≺ (a2 ≺ (. . . ≺ ai) . . .) ≻ an ≺ (. . . (ai+1 ≻ ai+2) ≻ . . .) ≻ an−1)
ha1, . . . , ani
n−1
=
=
n−1
Xi=0
Xi=0
(−1)n−1−i(a1 . . . ai) ≻ an ≺ (. . . (ai+1 ≻ ai+2) ≻ . . .) ≻ an−1)
= πV n−1
Xi=0
(−1)n−1−i(a1 . . . ai) ≻ an ≺ (. . . (ai+1 ≻ ai+2) ≻ . . .) ≻ an−1)! ,
20
where we denote by πV the canonical projection on V in T (V ). We obtain,as T (V )≺2 = Ker(πV ):
ha1, . . . , ani = πV ((a1 . . . an−1) ≻ an)
= πV ((a1 . . . an−1) ∗ an) − πV ((a1 . . . an−1) ≺ an)
= πV ((a1 . . . an−1) ∗ an).
In other terms, identifying V and T+(V )/T (V )≥2, V becomes a left (T (V ), ∗)-module, and the
brace structure of V is given by this module structure.
3.4 Dendriform structures on a cofree coalgebra
Theorem 32 Let A be a cofree coalgebra. We define:
1. DD(A) = {(≺, ≻) (A, ≺, ≻, ∆) is a dendriform Hopf algebra}.
2. LI(A) =(cid:26)(∗, I)
(A, ∗, ∆) is a Hopf algebra and
I is a left ideal of A such that A+ = P rim(A) ⊕ I (cid:27).
There is a bijection between these two sets, given by:
ΦA :(cid:26) DD(A) −→ LI(A)
(≺, ≻) −→ (≺ + ≻, A+ ≺ A+).
Proof. By proposition 30, ΦA is well-defined. We now define the inverse bijection ΨA. Let
(∗, I) ∈ LI(A). In order to lighten the notation, we put V = P rim(A). By corollary 28, there
exists a isomorphism of coalgebras φI : T (V ) −→ A, such that φI (v) = v for all v ∈ V and
φI (T (V )≥2) = I. Let ∗ be the product on T (V ), making φI an isomorphism of Hopf algebras.
Then T (V )≥2 is a left ideal of T (V ). By proposition 31, there exists a dendriform structure
(T (V ), ≺, ≻) on T (V ). Let (A, ≺, ≻) be the dendriform Hopf algebra structure on A, making
φI an isomorphism of dendriform Hopf algebras. We then put ΨA(∗, I) = (≺, ≻). Note that
≺ + ≻= ∗, as ≺ + ≻ = ∗.
Let us show that ΦA ◦ ΨA = IdLI(A). Let (∗, I) ∈ LI(A). We put ΦA ◦ ΨA(∗, I) = (∗′, I ′).
Then, as ΦI : (T (V ), ≺, ≻) −→ (T (V ), ≺, ≻) is an isomorphism of dendriform algebras:
I = ΦI (T (V )≥2) = φI (T (V ) ≺2) = A≺2.
By definition of ΦA, I ′ = A≺2 = I. Moreover, ∗′ =≺ + ≻= ∗.
Let us show that ΨA◦ΦA = IdDD(A). Let (≺, ≻) ∈ DD(A). We put ΨA◦ΦA(≺, ≻) = (≺′, ≻′),
and ΦA(≺, ≻) = (∗, I). As φI : (T (V ), ≺, ≻) −→ (A, ≺′, ≻′) is an isomorphism of dendriform
algebras, we have to prove that φI : (T (V ), ≺, ≻) −→ (A, ≺, ≻) is an isomorphism of dendriform
algebras. Let a = u1 . . . uk, b = v1 . . . vl ∈ T+(V ). First:
φI (a ≺b) − φI (a) ≺ φI (b) ∈ φI (T≥2(V )) + A+ ≺ A+ = I + I = I.
Let us prove that φI (a ≺b) = φI (a) ≺ φI (b) by induction on k. For k = 1, let us proceed by
induction on l. For l = 1, φI (a ≺b) = φI (v1u1). Moreover, in A:
∆(φI (u1) ≺ φI (v1)) = ∆(u1 ≺ v1)
= v1 ⊗ u1,
∆(φI (u1 ≺v1)) = (φI ⊗ φI ) ◦ ∆(u1 ≺ v1)
= φI (v1) ⊗ φI (u1)
= v1 ⊗ u1.
21
So φI (u1 ≺v1) − φI (u1) ≺ φI (v1) ∈ P rim(A) ∩ I = (0). Let us suppose the result for all l′ < l.
Then, by the induction hypothesis applied to b′′:
∆(φI (u1 ≺b)) = ∆(φI (bu1))
= (φI ⊗ φI )(b ⊗ u1 + b′ ⊗ u1 ≺b′′)
= φI (b) ⊗ φI (u1) + φI (b)′ ⊗ φI (u1) φI (b′′)
= ∆(φI (u1) ≺ φI (b)).
So φI (u1 ≺b)) − φI (u1) ≺ φI (b) ∈ P rim(A) ∩ I = (0). This prove the result for k = 1. Let us
assume the result at all rank k′ < k. We put u1 . . . uk−1 = a. Then, using the first step:
φI (a ≺b) = φI (uk ≺(a∗b))
= φI (uk) ≺ (φI (a) ∗ φI (b))
= (φI (uk) ≺ φI (a)) ≺ φI (b)
= (φI (uk ≺ a)) ≺ φI (b)
= φI (a) ≺ φI (b).
So ΨA ◦ ΦA = IdDD(A).
✷
4 MuPAD computations
We here give the different MuPAD procedures we used in this text.
4.1 Dimensions of simple modules
The following procedures compute the dimension of the simple module Γ(a1,...,an) for sln, sp2n,
so2n and so2n+1.
dimsl:=proc(a)
local n,res;
begin
n:=nops(a)+1;
res:=product(product((sum(a[k],k=i..j-1)+j-i)/(j-i),j=i+1..n),i=1..n-1);
return(res);
end_proc;
dimsp:=proc(a)
local n,k,s,l,res;
begin
n:=nops(a);l:=[];
for k from 1 to n do l:=l.[sum(a[x],x=k..n)+n-k]; end_for;
res:=product(product((l[i]-l[j])*(l[i]+l[j]+2),j=i+1..n),i=1..n-1);
res:=res*product(l[j]+1,j=1..n)/product((2*n-2*j-1)!,j=0..n-1);
return(res);
end_proc;
dimsoodd:=proc(a)
local n,k,l,res;
begin
n:=nops(a); l:=[];
for k from 1 to n do l:=l.[sum(a[x],x=k..n)-a[n]/2+n-k]; end_for;
22
res:=product(product((l[i]-l[j])*(l[i]+l[j]+1),j=i+1..n),i=1..n-1);
res:=res*product(2*l[j]+1,j=1..n)/product((2*n-2*j-1)!,j=0..n-1);
return(res);
end_proc;
dimsoeven:=proc(a)
local n,k,l,res;
begin
n:=nops(a); l:=[];
for k from 1 to n-2 do l:=l.[sum(a[x],x=k..n-2)+a[n-1]/2+a[n]/2+n-k];end_for;
l:=l.[a[n-1]/2+a[n]/2+1,-a[n-1]/2+a[n]/2];
res:=product(product((l[i]-l[j])*(l[i]+l[j]),j=i+1..n),i=1..n-1);
res:=res/product((2*n-2*j)!,j=1..n-1)*2^(n-1);
return(res);
end_proc;
The following procedures give the representations of dimension smaller than dim(g) for g =
sln, sp2n, so2n and so2n+1.
smallsl:=proc(n)
local ens,prod,i,d,dg;
begin
dg:=n^2-1;
print(Unquoted,"dimension of g: ".expr2text(dg));
ens:=[];
for i from 1 to n-1 do ens:=ens.[{0,1,2}]; end_for;
prod:=combinat::cartesianProduct::list(ens[x]$x=1..n-1);
for i from 1 to 3^(n-1) do
d:=dimsl(prod[i]);
if d<=dg then print(Unquoted,"heighest weight ".expr2text(prod[i]).
" of dimension ".expr2text(d));
end_if;
end_for;
end_proc;
smallsp:=proc(n)
local ens,prod,i,d,dg;
begin
dg:=n*(2*n+1);
print(Unquoted,"dimension of g: ".expr2text(dg));
ens:=[];
for i from 1 to n do ens:=ens.[{0,1,2}]; end_for;
prod:=combinat::cartesianProduct::list(ens[x]$x=1..n);
for i from 1 to 3^n do
d:=dimsp(prod[i]);
if d<=dg then print(Unquoted,"heighest weight ".expr2text(prod[i]).
" of dimension ".expr2text(d));
end_if;
end_for;
end_proc;
smallsoodd:=proc(n)
local ens,prod,i,d,dg;
23
begin
dg:=n*(2*n+1);
print(Unquoted,"dimension of g: ".expr2text(dg));
ens:=[];
for i from 1 to n do ens:=ens.[{0,1,2}]; end_for;
prod:=combinat::cartesianProduct::list(ens[x]$x=1..n);
for i from 1 to 3^n do
d:=dimsoodd(prod[i]);
if d<=dg then print(Unquoted,"heighest weight ".expr2text(prod[i]).
" of dimension ".expr2text(d));
end_if;
end_for;
end_proc;
smallsoeven:=proc(n)
local ens,prod,i,d,dg;
begin
dg:=n*(2*n-1);
print(Unquoted,"dimension of g: ".expr2text(dg));
ens:=[];
for i from 1 to n do ens:=ens.[{0,1,2}]; end_for;
prod:=combinat::cartesianProduct::list(ens[x]$x=1..n);
for i from 1 to 3^n do
d:=dimsoeven(prod[i]);
if d<=dg then print(Unquoted,"heighest weight ".expr2text(prod[i]).
" of dimension ".expr2text(d));
end_if;
end_for;
end_proc;
4.2 Computations for sl6
The procedure testsl6 produces the matrix used in section 2.4, corresponding to the action of
sl6 over Λ3(V ), where V is the standard representation of sl6.
basis:=[[1,2,3],[1,2,4],[1,2,5],[1,2,6],[1,3,4],[1,3,5],[1,3,6],[1,4,5],[1,4,6],
[1,5,6],[2,3,4],[2,3,5],[2,3,6],[2,4,5],[2,4,6],[2,5,6],
[3,4,5],[3,4,6],[3,5,6],[4,5,6]]:
indexbasis:=proc(B)
local i;
begin
for i from 1 to 20 do if B=basis[i] then return(i); end_if; end_for;
end_proc:
actionbasis:=proc(A,B)
local vec,vec1,liste,liste2,coef,numero,i,j,k,n;
begin
liste:=[]; i:=B[1]; j:=B[2]; k:=B[3];
vec:=matrix(6,1); vec[i]:=1; vec:=A*vec;
for n from 1 to 6 do liste:=liste.[[vec[n],[n,j,k]]]; end_for;
vec:=matrix(6,1); vec[j]:=1; vec:=A*vec;
for n from 1 to 6 do liste:=liste.[[vec[n],[i,n,k]]]; end_for;
24
vec:=matrix(6,1); vec[k]:=1; vec:=A*vec;
for n from 1 to 6 do liste:=liste.[[vec[n],[i,j,n]]]; end_for;
liste2:=[];
for n from 1 to nops(liste) do
coef:=(liste[n])[1];
i:=((liste[n])[2])[1]; j:=((liste[n])[2])[2]; k:=((liste[n])[2])[3];
if (coef<>0) then
if (i<j) and (j<k) then liste2:=liste2.[[coef,[i,j,k]]]; end_if;
if (i<k) and (k<j) then liste2:=liste2.[[-coef,[i,k,j]]]; end_if;
if (j<i) and (i<k) then liste2:=liste2.[[-coef,[j,i,k]]]; end_if;
if (j<k) and (k<i) then liste2:=liste2.[[coef,[j,k,i]]]; end_if;
if (k<i) and (i<j) then liste2:=liste2.[[coef,[k,i,j]]]; end_if;
if (k<j) and (j<i) then liste2:=liste2.[[-coef,[k,i,j]]]; end_if;
end_if;
end_for;
vec:=matrix(20,1);
for n from 1 to nops(liste2) do
coef:=(liste2[n])[1];
numero:=indexbasis((liste2[n])[2]);
vec1:=matrix(20,1); vec1[numero]:=coef; vec:=vec+vec1;
end_for;
return(vec);
end_proc:
actionvector:=proc(A,vec)
local i,res;
begin
res:=matrix(20,1);
for i from 1 to 20 do res:=res+vec[i]*actionbasis(A,basis[i]); end_for;
return(res);
end_proc:
testsl6:=proc()
local vec,res,i,j,mat;
begin
res:=[]; vec:=[];
for i from 1 to 20 do vec:=vec.[a[i]]; end_for;
for i from 1 to 5 do
mat:=matrix(6,6); mat[i,i]:=1; mat[i+1,i+1]:=-1;
res:=res.actionvector(mat,vec);
end_for;
for i from 1 to 5 do
for j from i+1 to 6 do
mat:=matrix(6,6); mat[i,j]:=1;
res:=res.actionvector(mat,vec);
mat:=matrix(6,6); mat[j,i]:=1;
res:=res.actionvector(mat,vec);
end_for;
end_for;
return(res);
end_proc;
25
4.3 Computations for g2
The procedure testg2 produces the matrix used in section 2.4, corresponding to the action of
g2 on M7,2(C).
H1:=matrix([[1,0,0,0,0,0,0],[0,-1,0,0,0,0,0],[0,0,2,0,0,0,0],[0,0,0,0,0,0,0],
[0,0,0,0,-2,0,0],[0,0,0,0,0,1,0],[0,0,0,0,0,0,-1]]):
H2:=matrix([[0,0,0,0,0,0,0],[0,1,0,0,0,0,0],[0,0,-1,0,0,0,0],[0,0,0,0,0,0,0],
[0,0,0,0,1,0,0],[0,0,0,0,0,-1,0],[0,0,0,0,0,0,0]]):
Y1:=matrix([[0,0,0,0,0,0,0],[1,0,0,0,0,0,0],[0,0,0,0,0,0,0],[0,0,1,0,0,0,0],
[0,0,0,2,0,0,0],[0,0,0,0,0,0,0],[0,0,0,0,0,-1,0]]):
Y2:=matrix([[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],[0,0,0,0,1,0,0],[0,0,0,0,0,0,0]]):
Y3:=-Y1*Y2+Y2*Y1: Y4:=-1/2*(Y1*Y3-Y3*Y1):
Y5:=1/3*(Y1*Y4-Y4*Y1): Y6:=Y2*Y5-Y5*Y2:
X1:=matrix([[0,1,0,0,0,0,0],[0,0,0,0,0,0,0],[0,0,0,2,0,0,0],[0,0,0,0,1,0,0],
[0,0,0,0,0,0,0],[0,0,0,0,0,0,-1],[0,0,0,0,0,0,0]]):
X2:=matrix([[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],
[0,0,0,0,0,1,0],[0,0,0,0,0,0,0],[0,0,0,0,0,0,0]]):
X3:=X1*X2-X2*X1: X4:=1/2*(X1*X3-X3*X1):
X5:=-1/3*(X1*X4-X4*X1): X6:=-(X2*X5-X5*X2):
g2:=[H1,H2,X1,X2,X3,X4,X5,X6,Y1,Y2,Y3,Y4,Y5,Y6]:
testg2:=proc()
local i,j,res,A,M;
begin
A:=matrix(7,2);
for i from 1 to 7 do for j from 1 to 2 do A[i,j]:=a[i,j]; end_for; end_for;
res:=[];
for i from 1 to 14 do
M:=g2[i]*A;
res:=res.linalg::stackMatrix(M[1..7,1],M[1..7,2]);
end_for;
return(res);
end_proc;
References
[1] 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.
[2] 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.
[3] 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.
[4] William Fulton and Joe Harris, Representation theory, Graduate Texts in Mathematics, vol.
129, Springer-Verlag, New York, 1991, A first course, Readings in Mathematics.
[5] Wee Liang Gan and Travis Schedler, The necklace Lie coalgebra and renormalization alge-
bras, J. Noncommut. Geom. 2 (2008), no. 2, 195 -- 214.
[6] Dirk Kreimer, Combinatorics of (pertubative) Quantum Field Theory, Phys. Rep. 4 -- 6
(2002), 387 -- 424, arXiv:hep-th/0010059.
26
[7] F. Kubo, Compatible algebra structures of Lie algebras, Ring theory 2007, World Sci. Publ.,
Hackensack, NJ, 2009, pp. 235 -- 239.
[8] Jean-Louis Loday, Dialgebras, Lecture Notes in Math., vol. 1763, Springer, Berlin, 2001.
[9] Jean-Louis Loday and Maria Ronco, Combinatorial hopf algebras, arXiv:0810.0435, 2009.
[10] Jean-Louis Loday and Maria O. Ronco, Hopf algebra of the planar binary trees, Adv. Math.
139 (1998), no. 2, 293 -- 309.
[11] Jean-Michel Oudom and Daniel Guin, Sur l'algèbre enveloppante d'une algèbre pré-Lie, C.
R. Math. Acad. Sci. Paris 340 (2005), no. 5, 331 -- 336, arXiv:math/0404457.
[12] Maria Ronco, A Milnor-Moore theorem for dendriform Hopf algebras, C. R. Acad. Sci. Paris
Sér. I Math. 332 (2001), no. 2, 109 -- 114.
[13] Pepijn van der Laan and Ieke Moerdijk, Families of Hopf algebras of trees and pre-Lie
algebras, Homology, Homotopy Appl. 8 (2006), no. 1, 243 -- 256, arXiv:math/0402022.
27
|
1804.11100 | 1 | 1804 | 2018-04-30T09:40:40 | Identities in unitriangular and gossip monoids | [
"math.RA"
] | We establish a criterion for a semigroup identity to hold in the monoid of $n \times n$ upper unitriangular matrices with entries in a commutative semiring $S$. This criterion is combinatorial modulo the arithmetic of the multiplicative identity element of $S$. In the case where $S$ is idempotent, the generated variety is the variety $\mathbf{J_{n-1}}$, which by a result of Volkov is generated by any one of: the monoid of unitriangular Boolean matrices, the monoid $R_n$ of all reflexive relations on an $n$ element set, or the Catalan monoid $C_n$. We propose $S$-matrix analogues of these latter two monoids in the case where $S$ is an idempotent semiring whose multiplicative identity element is the `top' element with respect to the natural partial order on $S$, and show that each generates $\mathbf{J_{n-1}}$. As a consequence we obtain a complete solution to the finite basis problem for lossy gossip monoids. | math.RA | math |
IDENTITIES IN UNITRIANGULAR AND GOSSIP
MONOIDS
MARIANNE JOHNSON1 AND PETER FENNER2
Abstract. We establish a criterion for a semigroup identity to hold
in the monoid of n × n upper unitriangular matrices with entries in a
commutative semiring S. This criterion is combinatorial modulo the
arithmetic of the multiplicative identity element of S. In the case where
S is idempotent, the generated variety is the variety Jn−1, which by a
result of Volkov is generated by any one of: the monoid of unitriangular
Boolean matrices, the monoid Rn of all reflexive relations on an n ele-
ment set, or the Catalan monoid Cn. We propose S-matrix analogues of
these latter two monoids in the case where S is an idempotent semiring
whose multiplicative identity element is the 'top' element with respect to
the natural partial order on S, and show that each generates Jn−1. As
a consequence we obtain a complete solution to the finite basis problem
for lossy gossip monoids.
1. Introduction
The finite basis problem for semigroups asks: which semigroups have an
equational theory admitting a finite basis of identities? Such semigroups are
called finitely based. In contrast to the situation for finite groups [8], it has
long been known that there exist finite semigroups which are non-finitely
based [9], and there is a rich literature studying the finite basis problem from
viewpoint of finite semigroups (see the survey [13]). As observed by Volkov
[15], infinite semigroups are far less frequently studied in the context of the
finite basis problem, due to the fact that many natural infinite semigroups
are in some sense 'too big' to allow for the kind of universal coincidences
demanded by identities. For example, if S is a commutative semiring into
which the semiring of natural numbers can be embedded, then for n > 1 the
monoid of all n×n (upper triangular) matrices over S satisfies no non-trivial
identities, since the free monoid of rank 2 embeds into all such semigroups
(see [15] for example). The finite basis problem is increasingly studied for
families of infinite semigroups of combinatorial interest for which identities
are known to exist, with complete results available for one-relator semigroups
[10] and Kauffman monoids [1], and several recent partial results for various
semigroups of upper triangular matrices with restrictions on the size of the
matrices and the entries permitted on the diagonals [3, 4, 15, 16].
In this work we consider the identities satisfied by several families of
matrix semigroups, beginning with upper triangular matrices with entries
1School of Mathematics, University of Manchester, Manchester M13 9PL, UK.
Email [email protected]
2School of Mathematics, University of Manchester, Manchester M13 9PL, UK.
Email [email protected].
Peter Fenner's research is supported by an EPSRC Doctoral Training Award.
1
2
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
in a commutative semiring. In Section 2 we show how the analysis of [5]
may be generalised to the setting of commutative semirings S to provide
necessary and sufficient conditions for a semigroup identity to hold in the
monoid of n × n upper triangular matrices with entries in S. This result is
then applied in Section 3 to establish a criterion for a semigroup identity
to hold in the submonoid of n × n upper unitriangular matrices, showing
that the generated variety depends only upon the isomorphism type of the
subsemiring generated by the multipllicative identity element of S. In the
case where S is an idempotent semiring our result together with a result of
Volkov [14] yields that the generated variety is Jn−1, that is, the variety of
semigroups generated by the monoids of height n − 1 in Simon's hierarchy
of finite J -trivial monoids [11]. In Sections 4 we introduce the submonoid
Rn(S) of the full matrix monoid over an interval semiring S, and show
that this generates the same variety as its finite (Boolean) counterpart, the
reflexive monoid. In Section 5 we consider several monoids related to the
Catalan monoid, including the so-called lossy gossip monoid Gn (that is, the
monoid generated by all "metric" matrices in the full matrix monoid over
the tropical semiring [2]). By [14] this common variety is once again seen
to be Jn−1. Since the variety Jn−1 is known to be finitely based for n ≤ 4,
and non-finitely based otherwise, this settles the finite basis problem for the
above mentioned families of monoids.
We conclude this introduction by briefly recalling the necessary defini-
tions, notation and background.
Semigroup identities. We write N0 and N respectively for the natural
numbers with and without 0. If Σ is a finite alphabet, then Σ+ will denote
the free semigroup on Σ, that is, the set of finite, non-empty words over
Σ under the operation of concatenation. Likewise, Σ∗ will denote the free
monoid on Σ. Thus Σ∗ = Σ+ ∪ {1} where 1 denotes the empty word. For
w ∈ Σ+ and s ∈ Σ we write w for the length of w and ws for the number
of occurrences of the letter s in w. For 1 ≤ i ≤ w we write wi to denote
the ith letter of w. The content of w is the map Σ → N0, s 7→ ws.
Recall that a (semigroup) identity is a pair of words, usually written
"u = v", in the free semigroup Σ+ on an alphabet Σ. The identity is said
to be balanced if ua = va for all a ∈ Σ. We say that the identity holds
in a semigroup U (or that U satisfies the identity) if every morphism from
Σ+ to U maps u and v to the same element of U . If a morphism maps u
and v to the same element we say that it satisfies the given identity in U ;
otherwise it falsifies it. We write Id(U ) to denote the set of all identities
satisfied by the semigroup U .
Semirings. Let S be a commutative semiring with additive identity 0S and
multiplicative identity 1S. We say that S is idempotent if a + a = a for all
a ∈ S. Examples include the Boolean semiring B = {0, 1} in which the only
undetermined operation is defined by 1 + 1 = 1, and the tropical semifield
(R ∪ {−∞}, ⊕, ⊗), where a ⊕ b = max(a, b) and a ⊗ b = a + b, and in which
−∞ is the 'zero' element, and 0 is the 'one'. There is a natural partial order
on every idempotent semiring S given by a ≤ b if and only if a + b = b;
it is clear from definition that a + b ≥ a, b for all a, b ∈ S. Thus 0S is the
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
3
least element of S with respect to this order. Moreover, if a ≤ b in S, then
cad ≤ cbd and a + c ≤ b + c for all c, d ∈ S.
We say that a commutative semiring S is an interval semiring if S is idem-
potent, and 1S is the greatest element of S with respect to the natural partial
order on S. Examples of interval semirings include: the Boolean semiring
B; the semiring I = ([0, 1], ·, ⊕) with usual multiplication of numbers and
addition given by taking the maximum; the semiring (R≤0 ∪ {−∞}, ⊗, ⊕)
with multiplication given by usual addition of numbers and addition given
by taking the maximum; any distributive lattice L with addition ∨ and
multiplication ∧.
Matrix semigroups. It is easy to see that the set of all n×n matrices with
entries in S forms a monoid under the matrix multiplication induced from
the operations in S. We denote this semigroup by Mn(S) and write U Tn(S)
to denote the subsemigroup of Mn(S) consisting of the upper-triangular
matrices in Mn(S) whose entries below the main diagonal are zero. We
also write Un(S) to denote the semigroup of unitriangular matrices, namely
those elements of U Tn(S) whose diagonal entries are all equal to 1S.
In the case where S is an idempotent semiring we define a partial order
(cid:22) on Mn(S) by A (cid:22) B if and only if Ai,j ≤ Bi,j for all i and j. It is easy
to see that matrix multiplication respects the partial order (cid:22) (i.e. Mn(S) is
an ordered monoid). Indeed, for A, B, C ∈ Mn(S) with A (cid:22) B, for all i, j
we have
(CA)i,j =
n
Xk=1
Ci,kAk,j ≤
n
Xk=1
Ci,kBk,j = (CB)i,j.
Polynomials. By a formal polynomial in variables from a set X we mean
an element of the commutative polynomial semiring S[X], that is, a finite
formal sum in which each term is a formal product of a non-zero coefficient
from S and formal powers of finitely many of the variables of X, considered
up to the commutative and distributive laws in S. If S is idempotent, we
consider the summation up to idempotency of addition.
Each formal polynomial naturally defines a function from SX to S, by
interpreting all formal products and formal sums as products and sums
within S. Two distinct formal polynomials may define the same function.
For example, x⊗2 ⊕x⊕1 and x⊗2 ⊕1 are distinct formal tropical polynomials
defining the same function, since x can never exceed both x⊗2 and 1. We
say that two formal polynomials are functionally equivalent over S if they
represent the same function from SX to S.
2. The identities of triangular matrices
We begin by providing a generalisation of [5, Theorem 5.1] to the setting
of commutative semirings.
Let [n] = {1, 2, . . . , n}. By a k-vertex walk (or walk of vertex length k) in
[n] we mean a k-tuple (v1, . . . , vk) such that v1 ≤ v2 ≤ · · · ≤ vk. A k-vertex
path (or path of vertex length k) is a k-vertex walk in which consecutive
vertices (and hence all vertices) are distinct.
4
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
Let w be a word over the alphabet Σ. For 0 ≤ p < q ≤ w + 1 and s ∈ Σ
we define
βw
s (p, q) = {i ∈ N p < i < q, wi = s}
to be the number of occurrences of s lying strictly between wp and wq.
For each u ∈ Σ∗ with u ≤ n − 1 and each (u + 1)-vertex path ρ =
(ρ0, ρ1, . . . , ρu) in [n], we define a formal polynomial (over an arbitrary, but
fixed, commutative semiring S) having variables x(s, i) for each letter s ∈ Σ
and vertex i ∈ [n] as follows:
f w
u,ρ =XYs∈Σ
u
Yk=0
x(s, ρk)βw
s (αk ,αk+1),
where the sum ranges over all 0 = α0 < α1 < · · · < αu < αu+1 = w + 1
such that wαk = uk for k = 1, . . . , u. Thus it is easy to see that f w
u,ρ 6= 0S
if and only if u is a scattered subword of w of length equal to the path ρ.
Note that taking u to be the empty word forces ρ = (ρ0) for some ρ0 ∈ [n]
and hence f w
the content of w.
u,ρ =Qs∈Σ x(s, ρ0)ws is a monomial completely determined by
Lemma 2.1. Let S be a commutative semiring, and let φ : Σ+ → U Tn(S)
be a morphism. Define x ∈ SΣ×[n] by
Then for any word w ∈ Σ+ and vertices i, j ∈ [n] we have
x(s, i) = φ(s)i,i.
φ(w)i,j = Xu∈Σ∗,
u≤n−1 Xρ∈[n]u
i,j
u
Yk=1
φ(uk)ρk−1,ρk
· f w
u,ρ(x),
(1)
where [n]u
i,j denotes the set of all (u + 1)-vertex paths from i to j in [n].
Proof. We follow the proof given in [5].
Let i and j be vertices. Using the definition of the functions f w
u,ρ, the
value given to x and the distributivity of multiplication over addition, the
right-hand-side of (1) is equal to
Xu∈Σ∗,
u≤n−1 Xα∈Aw
u Xρ∈[n]u
i,j
u
Yk=1
φ(uk)ρk−1,ρk
·
Ys∈Σ
u
Yk=0
(φ(s)ρk ,ρk )βw
s (αk ,αk+1)
where
Aw
u = {(α0, . . . , αu+1) : 0 = α0 < α1 < · · · < αu < αu+1 = w+1 with wαk = uk}.
Notice that we are summing over all possible words u of length less than n,
and then over all scattered subwords of w equal to u. Thus, we are simply
summing over all scattered subwords of w of length less than n, so the above
is equal to:
n−1
Xl=0 Xα∈Al Xρ∈[n]l
i,j
l
Yk=1
φ(wαk )ρk−1,ρk! · Ys∈Σ
·
l
Yk=0
(φ(s)ρk ,ρk )βw
s (αk ,αk+1)!
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
5
where Al = {(α0, . . . , αl+1) : 0 = α0 < α1 < · · · < αl < αl+1 = w + 1}.
Now to each term in the above sum, defined by a choice of αi's and a
ρ ∈ [n]l
i,j, we can associate a (w + 1)-vertex walk (σ0 = i, . . . , σw = j) in
[n] whose underlying path is ρ and which transitions to vertex ρk after αk
steps. Clearly every (w + 1)-vertex walk from i to j arises exactly once in
this way, and so we are summing over all such walks. In each term, the first
bracket gives a factor φ(wq)σq−1,σq when q = αk for some k, while from the
definition of the functions βw
s , the second bracket gives a factor φ(wq)σq−1,σq
for each q not of this form. Thus, the above is simply equal to:
w
Yq=1
X
φ(wq)σq−1,σq
where the sum is taken over all (w + 1)-vertex walks (i = σ0, σ1, . . . , σw =
j) in [n]. But by the definition of multiplication in U Tn(S), this is easily
(cid:3)
= φ(w)i,j .
seen to be equal to (cid:0)φ(w1) · · · φ(ww)(cid:1)i,j
Let f w
u denote the polynomial f w
u,ρ with ρ = (1, 2, . . . , u + 1) in variables
x(s, i), with s ∈ Σ and 1 ≤ i ≤ u + 1. We are now ready to prove the main
theorem of this section.
Theorem 2.2. Let S be a commutative semiring. The identity w = v
over alphabet Σ is satisfied in U Tn(S) if and only if for every u ∈ Σ∗ with
u ≤ n − 1 the polynomials f w
u and f v
u are functionally equivalent over S.
u (x) for some word u ∈ Σ+ of length at
Proof. Suppose first that f w
most n − 1 and x ∈ SΣ×[n]. Define a morphism φ : Σ+ → U Tn(S) by
u (x) 6= f v
φ(s)p,p = x(s, p) ∈ S, for all p ∈ [n] and s ∈ Σ; and
φ(s)p,q = (1S
0S
if s = ui, p = i, q = i + 1
otherwise.
Then by Lemma 2.1,
φ(v)i,j = f v
u (x) 6= f w
u (x) = φ(w)i,j ,
and so the morphism φ falsifies the identity in U Tn(S).
u and f v
Conversely, suppose that f w
u are functionally equivalent over S for
all u ∈ Σ∗ of length at most n − 1. Noting that for any path ρ of vertex
u,ρ and f w
length u, the polynomials f w
u differ only in the labelling of their
u,ρ and f v
variables, it is then easy to see that f w
u,ρ are functionally equivalent
for all pairs u, ρ with u ∈ Σ∗ of length at most n − 1, and ρ a path of vertex
length u + 1 through [n].
It suffices to show that the identity w = v is satisfied by every morphism
φ : Σ+ → U Tn(S), so let φ be such a morphism and define x ∈ SΣ×[n] by
x(s, i) = φ(s)i,i. Since φ is a morphism to U Tn(S), we know that φ(w)i,j =
0S = φ(v)i,j whenever i > j. On the other hand, if i ≤ j then Lemma 2.1
gives
φ(w)i,j = Xu∈Σ∗,
u≤n−1 Xρ∈[n]u
i,j
u
Yk=1
φ(uk)ρk−1,ρk
· f w
u,ρ(x) = φ(v)i,j .
6
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
Lemma 2.3. Let S be a semiring whose multiplicative monoid contains an
element α generating a free submonoid of rank 1, and let w, v ∈ Σ+.
(cid:3)
(i) The polynomials f w
1 and f v
1 are functionally equivalent if and only if
w and v have the same content.
(ii) Suppose further that the partial sumsPj
distinct. If f w
1 is functionally equivalent to f v
f w
a is functionally equivalent to f v
1 .
i=0 αi for j ∈ N0 are pairwise
a for all a ∈ Σ, then
1 is the monomialQs∈Σ x(s, 1)ws, and it is clear
Proof. (i) By definition, f w
that w and v have the same content if and only if the formal polynomials
1 and f v
f w
1 are identical. In particular, if the content of the two words agree,
then these polynomials are functionally equivalent. Suppose then that f w
1
and f v
1 are functionally equivalent. Setting x(s, 1) = α and x(t, 1) = 1S for
all t 6= s then yields αws = αvs, and hence ws = vs. Repeating this
argument for each s ∈ Σ yields that the two words have the same content.
a for all
a ∈ Σ, then the content of the two words must be equal. Evaluating the
polynomials f w
a at x(a, 1) = α and x(z, i) = 1S for all other choices
i=0 αi, and hence wa = va. Repeating this
argument for each a ∈ Σ gives that the two words have the same content. (cid:3)
a is functionally equivalent to f v
(ii) It suffices to show that if f w
a and f v
of z, i yieldsPwa−1
αi =Pva−1
i=0
Under the hypothesis of the previous lemma, we note that identities must
be balanced.
Corollary 2.4. Let S be a semiring whose multiplicative monoid contains
a free submonoid of rank 1. Any non-trivial semigroup identity satisfied by
U Tn(S) or Mn(S) must be balanced.
Under the stronger hypothesese of the Lemma 2.3 (ii), we may reduce the
number of polynomials to be checked by 1.
Corollary 2.5. Let S be a semiring whose multiplicative monoid contains
an element α generating a free submonoid of rank 1, and suppose that the
i=1 αi for j ∈ N are pairwise distinct. The identity w = v
over alphabet Σ is satisfied in U Tn(S) if and only if for every u ∈ Σ+ with
1 ≤ u ≤ n − 1 the polynomials f w
partial sums Pj
u are functionally equivalent.
u and f v
3. The identities of unitriangular matrices
Say that the scattered multiplicity of u ∈ Σ+ in w ∈ Σ+ is the number
of distinct ways in which u occurs as a scattered subword of w, and denote
this by mw
u ∈ N0. For m ∈ N0 write ⌊m⌋S :=Pm
j=1 1S .
Theorem 3.1. Let S be a commutative semiring. The identity w = v over
alphabet Σ is satisfied in Un(S) if and only if ⌊mw
u⌋S for each word
u ∈ Σ+ of length at most n − 1.
u ⌋S = ⌊mv
Proof. Let φ : Σ+ → Un(S) be a morphism. Since every element of the
image of φ has all diagonal entries equal to 1S it follows from Lemma 2.1
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
7
and the definition of the polynomials f w
u,ρ that for all 1 ≤ i < j ≤ n, we have
φ(w)i,j = Xu∈Σ+,
u≤n−1 Xρ∈[n]u
i,j
u
Yk=1
φ(uk)ρk−1,ρk
u ⌋S,
· ⌊mw
where mw
u denotes the scattered multiplicity of u in w. Since these mul-
tiplicities account for the only part of the formula which directly depends
upon w, it is then clear that if each of the equalities ⌊mw
u⌋S holds,
then we must have w = v in Un(S).
u ⌋S = ⌊mv
Now suppose w = v is satsified in Un(S) and let u be a word of length
u in w and v respectively.
l < n with scattered multiplicities mw
Consider the morphism φ : Σ+ → Un(S) defined by
u and mv
φ(s)p,p = 1S , for all p ∈ [n] and s ∈ Σ; and
φ(s)p,q = (1S
0S
if s = ui, p = i, q = i + 1
otherwise.
Notice that Lemma 2.1 then yields ⌊mw
u ⌋S = φ(w)1,l+1 = φ(v)1,l+1 = ⌊mv
u⌋S.
(cid:3)
Proposition 3.2. Let S and T be commutative semirings. The semigroups
Un(S) and Un(T ) generate the same variety of semigroups if and only if 1S
and 1T generate isomorphic semirings.
Proof. If 1S and 1T generate isomorphic semirings, then for all j, k ∈ N0 we
have ⌊j⌋S = ⌊k⌋S if and only if ⌊j⌋T = ⌊k⌋T . It then follows immediately
from Theorem 3.1 that Un(S) and Un(T ) satisfy exactly the same semigroup
identities.
u ⌋T = ⌊mv
Conversely, if Un(S) and Un(T ) satisfy the same identities, it follows that
for all words w, v, u ∈ Σ+ we must have ⌊mw
u⌋S if and only if
u⌋T . Consideration of all pairs of words w = aj, v = ak with
⌊mw
respect to the fixed word u = a of length 1 allows us to determine all relations
of the form ⌊j⌋R = ⌊k⌋R for j, k ∈ N and R = S, T . Since the same set of
relations holds for R = S and R = T , it follows that 1S and 1T generate
isomorphic semirings.
(cid:3)
u ⌋S = ⌊mv
Corollary 3.3. Let S be an idempotent semiring. The identity w = v over
alphabet Σ is satisfied in Un(S) if and only if w and v admit the same set
of scattered subwords of length at most n − 1.
Proof. If S is idempotent then it is easy to see that
⌊mv
u⌋S =(1S
0S
if u is a scattered subword of w
otherwise.
(cid:3)
The previous results generalise a result of Volkov [14], who proved that
w = v is a semigroup identity for Un(B) if and only if w and v have the
same scattered subwords of length at most n − 1. Since the results of that
paper also show that the unitriangular Boolean matrices Un(B), the monoid
Rn of reflexive binary relations on a set of cardinality n, and the Catalan
8
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
monoid Cn all satisfy exactly the same set of identities, we get the following
immediate corollary.
Corollary 3.4. Let S be an idempotent semiring. The semigroup Un(S)
satisfies exactly the same semigroup identities as the semigroup of reflexive
relations Rn or the Catalan monoid Cn.
The monoid Un(S) can be viewed as an oversemigroup of Un(B) allowing
for entries over the idempotent semiring S, and so it is natural to ask if there
are analogous extensions of Rn and Cn. We note that there is an obvious
Boolean matrix representation of Rn, formed by sending a relation R to the
Boolean matrix whose i, jth entry is 1 if and only if i and j are related
by R. In the following section we shall consider a natural analogue of Rn
consisting of matrices over a semiring S with diagonal entries all equal to
the multiplicative identity of S. It is clear that, in general, the set of all such
matrices need not form a semigroup (e.g. over the tropical semiring such
matrices are not closed under multiplication). We shall therefore restrict
our attention to a particular class of idempotent semirings.
4. Generalised reflexive monoids
Lemma 4.1. Let S be an idempotent semiring, and let V be a subsemigroup
of Mn(S) with the property that every element of V has all diagonal entries
equal to 1S.
(i) If A = U (1)X(1)U (2) · · · U (L)X(L)U (L+1) and B = X(1) · · · X(L)
for some U (i), X(i) ∈ V , then B (cid:22) A.
(ii) For all A ∈ V we have
In (cid:22) A (cid:22) A2 (cid:22) A2 (cid:22) · · · (cid:22) An (cid:22) · · ·
where In denotes the identity matrix of Mn(S).
(In particular, V is J -trivial and so every regular element of V is
idempotent.)
Proof. (i) Suppose that A = U (1)X(1)U (2) · · · U (L)X(L)U (L + 1) and B =
X(1) · · · X(L). Since every element of V has only ones on its diagonal, for
all i, j ∈ [n] this gives
Ai,j = X U (1)ρ0,ρ1 X(1)ρ1,ρ2 U (2)ρ2,ρ3 · · · X(L)ρ2L−1 ,ρ2LU (L + 1)ρ2L,ρ2L+1
where the sum ranges over all choices of ρi ∈ [n], with ρ0 = i and ρ2L+1 = j.
Since a + b ≥ a, b for all a, b ∈ S, it follows that by restricting the choices
for the ρi we will obtain a partial sum that must be less than or equal to
Ai,j. In particular, we have
Ai,j ≥ X U (1)ρ0,ρ0X(1)ρ0,ρ1 U (2)ρ1,ρ1 · · · X(L)ρL−1,ρL U (L + 1)ρL,ρL
where the sum ranges over all choices of ρi ∈ [n], with ρ0 = i and ρL = j.
Since all diagonal entries of elements of V are equal to 1S , this gives
Ai,j ≥ X X(1)ρ0 ,ρ1 · · · X(L)ρL−1,ρL ,
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
9
where the sum ranges over all choices of ρi ∈ [n], with ρ0 = i and ρL = j.
By the definition of matrix multiplication, the latter is equal to Bi,j. Thus
for all i, j ∈ [n] we have Ai,j ≥ Bi,j, and hence B (cid:22) A.
(ii) It follows immediately from part (i) that the powers are non-decreasing.
In particular, if AJ B in V then there exist P, Q, X, Y ∈ V with A = P BQ
and B = XAY . Now by part (i) this gives A (cid:22) B and B (cid:22) A, and hence
A = B. Recalling that an element A ∈ V is regular if and only if it is
D-related to an idempotent, it follows immediately that A is regular if and
only if it is idempotent.
(cid:3)
From now on let S be an interval semiring and define
Rn(S) = {A ∈ Mn(S) : Ai,i = 1S}.
It is easily verified that Rn(S) is a semigroup satisfying the conditions of
Lemma 4.1. Let Z be the element of Rn(S) given by Zi,j = 1S for all
i and j. Then it is easy to see that In (cid:22) A (cid:22) Z for all A ∈ Rn(S), with
AZ = Z = ZA. In the case where S = B, it is clear that Rn(B) is isomorphic
to the monoid Rn of reflexive binary relations on a set of cardinality n.
Let ρ := (ρ0, . . . , ρL) be an L + 1-tuple of elements from [n]. We shall say
that ρ is a block chain of length L + 1 if ρ has the form:
ρ := (i0, . . . , i0, i1, . . . , i1, . . . , ik . . . , ik),
where i0, . . . , ik are distinct elements of [n] and thus, k ≤ n − 1.
Lemma 4.2. Let S be an interval semiring.
(i) If A = X(1) · · · X(L) in Rn(S), then for all i, j ∈ [n] we have
Ai,j =X X(1)ρ0,ρ1X(2)ρ1 ,ρ2 · · · X(L)ρL−1,ρL,
where the sum ranges over all block chains ρ := (ρ0, . . . , ρL) with
ρ0 = i and ρL = j.
(ii) For all A ∈ Rn(S) and all N ≥ n − 1 we have AN = An−1.
(In particular, An−1 is idempotent and Rn(S) is aperiodic.)
Proof. (i) Let A = X(1) · · · X(L) in Rn(S). Then, by the definition of
matrix multiplication,
Ai,j =X X(1)ρ0,ρ1X(2)ρ1 ,ρ2 · · · X(L)ρL−1,ρL,
where the sum ranges over all L + 1-tuples ρ := (ρ0, . . . , ρL), with ρk ∈ [n]
and ρ0 = i, ρL = j. Let ρ be such a tuple, and suppose that ρ is not a block
chain. Then for some s, t with s + 1 < t we must have ρs 6= ρs+1 and ρs = ρt.
Consider the tuple ρ′ := (ρ′
L) obtained from ρ by replacing each ρk
with s < k < t by ρs. Since each diagonal entry is equal to 1S and 1S ≥ a
for all a ∈ S, it is easy to see that:
0, . . . , ρ′
X(1)ρ′
0,ρ′
1
X(2)ρ′
1,ρ′
2
· · · X(L)ρ′
L−1,ρ′
L
≥ X(1)ρ0 ,ρ1X(2)ρ1,ρ2 · · · X(L)ρL−1,ρL.
By repeated application of the above argument, it is clear that
X(1)σ0 ,σ1 X(2)σ1,σ2 · · · X(L)σL−1,σL ≥ X(1)ρ0,ρ1 X(2)ρ1,ρ2 · · · X(L)ρL−1,ρL,
10
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
for some block chain σ. Since a ≤ b in S if and only if a + b = b, it follows
from the previous observation that taking the sum over all block chains must
give the same result as taking the sum over all tuples. Thus
Ai,j =X X(1)ρ0,ρ1X(2)ρ1 ,ρ2 · · · X(L)ρL−1,ρL,
where the sum ranges over all block chains ρ := (ρ0, . . . , ρL) with ρ0 = i and
ρL = j.
(ii) Let A ∈ Rn(S) and N ∈ N. Then by part (i)
(AN )i,j =X Aρ0,ρ1Aρ1,ρ2 · · · AρN−1,ρN ,
where the sum ranges over all N + 1-tuples of the form
ρ := (i, . . . , i, i1, . . . , i1, . . . , ik . . . , ik, j . . . , j),
where i, i1, . . . , ik, j are distinct elements of [n]. Moreover, for such an N +1-
tuple ρ, the fact that the diagonal entries of A are all equal to 1S means
that the corresponding term of the summation is equal to
Ai,i1 Ai1,i2 · · · Aik−1,ik Aik,j.
Thus for each N ≥ n − 1 we see that every term occurring in the summation
above also occurs as a term in the corresponding summation for An−1, and
hence AN (cid:22) An−1. On the other hand, by Lemma 4.1, we know that
An−1 (cid:22) AN for all N ≥ n − 1. Thus we may conclude that An−1 = AN for
all N ≥ n − 1. In particular,
An−1An−1 = A2n−2 = An−1.
(Recall that a semigroup V is aperiodic if for every a ∈ V there exists a
positive integer m such that am+1 = am.)
(cid:3)
We note that in the case where Rn(S) is finite, the fact that Rn(S) is
aperiodic follows directly from Lemma 4.1, since every finite H-trivial semi-
group is aperiodic. For infinite semigroups, J -triviality is not sufficient to
deduce aperiodicity (for example, the semigroup of natural numbers under
addition is an infinite J -trivial semigroup which is clearly not aperiodic).
Theorem 4.3. Let S be an interval semiring. The identity w = v over
alphabet Σ is satisfied in Rn(S) if and only if w and v have the same scattered
subwords of length at most n − 1.
Proof. Noting that Un(S) ⊆ Rn(S), it suffices to show that if w and v have
the same scattered subwords of length at most n − 1, then w = v holds in
Rn(S).
Let φ : Σ+ → Rn(S) be a morphism and let w = w1 · · · wq ∈ Σ+. By
Lemma 4.2 for each i, j ∈ [n] we have
φ(w)i,j = (φ(w1) · · · φ(wq))i,j
= X φ(w1)ρ0,ρ1 · · · φ(wq)ρq−1,ρq ,
where the sum ranges over all block chains ρ of total length q + 1, with
first entry i and last entry j. To each choice of t = (t0, t1, . . . , tp, tp+1) with
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
11
0 = t0 < t1 < · · · < tp < tp+1 = q + 1 and p ≤ n − 1 we may associate the
set Bi,j
t of all block chains of the form:
, i2, . . . , i2
(i0, . . . , i0
, i1, . . . , i1
t1−t0
t2−t1
t3−t2
{z
}
{z
}
{z
}
. . . , ip−1, . . . , ip−1
ip, . . . , ip
)
tp−tp−1
tp+1−tp
{z
}
{z
}
with i0 = i, ip = j. It is easy to see that the set of all block chains of total
length q + 1 with first entry i and last entry j is the disjoint union of the
sets Bi,j
. Thus the summation above can be viewed as summing over all
t
block chains in Bi,j
t
for all choices 0 = t0 < t1 < · · · < tp < tp+1 = q + 1.
Fix t and consider the term of the summation corresponding to the block
chain
(i0, . . . , i0
, i1, . . . , i1
, i2, . . . , i2
. . . , ip−1, . . . , ip−1
ip, . . . , ip
).
{z
t1−t0
t2−t1
t3−t2
}
{z
}
{z
}
tp−tp−1
tp+1−tp
{z
}
{z
}
The fact that all diagonal entries are equal to 1S means that the correspond-
ing term is equal to
φ(wt1 )i0,i1φ(wt2 )i1,i2 · · · φ(wtp)ip−1,ip.
It is then clear that the above expression depends only upon the choice
of scattered subword u = wt1 · · · wtp of w of length p ≤ n − 1, and the
intermediate vertices i1, . . . , ip−1. Since addition in S is idempotent, we
may therefore conclude that
φ(w)i,j =X φ(u1)i0,i1φ(u2)i1,i2 · · · φ(up)ip−1,ip,
where the sum ranges over all scattered subwords u of w of length at most
n − 1, and over all choices of distinct i0, . . . , ip ∈ [n] with i0 = i and ip = j.
It then follows that if w and v contain the same scattered subwords of length
at most n − 1 then φ(w) = φ(v).
(cid:3)
5. Catalan monoids and gossip
The Catalan monoid Cn [12] is the monoid given by the presentation with
generators e1, . . . , en−1 and relations
eiei = ei,
eiej = ej ei
eiei+1ei = ei+1eiei+1 = eiei+1
(2)
for all appropriate i, j with i − j > 1. The name comes from the fact that
Cn = 1
n+1(cid:0)2n
n(cid:1) is the nth Catalan number.
Say that a matrix A ∈ Mn(B) is convex if:
(1) Ai,l = Ai,r = 1 with l ≤ r implies Ai,k = 1 for all l ≤ k ≤ r,
(2) Au,j = Ad,j = 1 with u ≤ d implies Ak,j = 1 for all u ≤ k ≤ d, and
(3) Ai,i = 1 for all i.
By [7, Proposition 3] the set Convn of all convex Boolean matrices is a
submonoid of Rn. Let C U
n = Convn ∩ Un denote the monoid of all convex
upper unitriangular matrices, and for 1 ≤ i ≤ n − 1 let D(i) ∈ C U
n be
the matrix with 1's on the diagonal and a single off-diagonal 1 in position
(i, i + 1).
Lemma 5.1. The matrices D(1), . . . , D(n − 1) generate the monoid C U
all convex upper unitriangular Boolean matrices. Moreover, C U
n
∼= Cn.
n of
12
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
Proof. Since each D(i) is a convex upper unitriangular matrix, these matri-
ces clearly generate a submonoid of C U
n . Let
mi = max{j : Ai,j = 1}. Since Ai,i = 1 we note that mi ≥ i. Convexity of
A yields that mi ≤ mj whenever i ≤ j. Define
n . Suppose then that A ∈ C U
F (i) =(In
D(i) · · · D(mi − 1)
if mi = i
if mi > i
It is straightforward to verify that if F (i)i,j = 1 if and only if i ≤ j ≤ mi,
and similarly for all k > i, we have F (k)i,j = 1 if and only if i = j. Thus
the (i, j)th coordinate of B := F (n − 1) · · · F (i) is non-zero if and only if
i ≤ j ≤ mi. Let M = BF (i−1) · · · F (1). We claim A = M . Since M (cid:23) B, it
is clear from the observations above that Mi,j ≥ Bi,j = 1 for all i ≤ j ≤ mi.
Since M is clearly upper triangular, it remains to show that Mi,j = 0 for
all j > mi. To see this, notice that the right action of D(k) on any Boolean
matrix X results in the matrix obtained from X by taking the Boolean sum
of columns k and k + 1. By definition, all factors D(k) occurring in F (j)
satisfy j ≤ k ≤ mj − 1. For j < i the only factors D(k) occurring in F (j)
therefore satisfy j ≤ k ≤ mj − 1 ≤ mi − 1. This means that M is obtained
from the matrix B by the right action of some collection of matrices D(k)
with k ≤ mi − 1, and hence columns j > mi of M and B agree.
It is straightforward to verify that the matrices D(i) satisfy the relations
n are in one to one correspondence with the
n = Cn, and so these two
(cid:3)
(2). Since the elements of C U
Dyck paths from (0, 0) to (n, n), we see that C U
monoids must be isomorphic.
Let E(i) denote the product D(i)D(i)T ∈ Convn. The double Catalan
monoid DCn of Mazorchuk and Steinberg [7] is the submonoid of Convn
generated by the matrices E1, . . . , En−1. Define U : DCn → C U
n to be the
map sending a matrix to its upper profile, namely U (A)i,j = Ai,j if i ≤ j
and U (A)i,j = 0 otherwise.
Lemma 5.2. The map U : DCn → C U
phism.
n is a surjective monoid homomor-
Proof. Let A, B ∈ DCn. By definition U (AB)i,j = (AB)i,j if i ≤ j and
0 otherwise. Thus the non-zero entries occur in positions i ≤ j for which
there exists k with Ai,k = Bk,j = 1. Note that if there exists such a k with
k < i, then by the convexity of B we must have Ai,i = Bi,j = 1, whilst if
there exists such a k with k > j, then by the convexity of A we must have
Ai,j = Bj,j = 1. The non-zero entries of U (AB) therefore occur in positions
i, j for which there exists i ≤ k ≤ j with Ai,k = Bk,j = 1, and it is easy to
see that these coincide with the non-zero entries of U (A)U (B).
Now let A ∈ DCn. By definition we may write A = Ei1 · · · Eim for some
1 ≤ i1, . . . , im ≤ n. Applying the morphism U then yields
U (A) = U (Ei1) · · · U (Eim) = Di1 · · · Dim ,
and the result follows from Lemma 5.1.
(cid:3)
Now let D(i, j) denote the n × n Boolean matrix with 1's on the diagonal
and a single off-diagonal 1 in position (i, j), and let E(i, j) = D(i, j)D(j, i).
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
13
The gossip monoid [2, 6] is the submonoid of Mn(B) generated by the set
{E(i, j) : 1 ≤ i < j ≤ n}. The one directional gossip monoid Gn is the
submonoid of Mn(B) generated by the set {D(i, j) : 1 ≤ i 6= j ≤ n}. It
is clear from the definition that Gn is a submonoid of Gn. Moreover, since
E(i) = E(i, i + 1) we see that the double Catalan monoid is a submonoid of
Gn. The names 'one-directional gossip monoid' and 'gossip monoid' come
from the following interpretation of the matrices D(i, j) and E(i, j). Con-
sider a group of n people, each with a unique piece of information or 'gossip'
they would like to spread. It is clear that we can record the state of knowl-
edge amongst the n people at any given time by means of a Boolean matrix,
putting a 1 in the (i, j)th position if and only if person j has learned the
piece of gossip originally known only to person i. The right action of the ma-
trix D(i, j) on Mn(B) then corresponds to a one-way communication from
person i to person j, in which person i recounts to person j all of the gossip
that they know. The right action of the matrix E(i, j) on Mn(B) corre-
sponds to a two-way communication between person i and person j, at the
end of which both parties have learned the sum total of gossip known to
either i or j. The double Catalan monoid can therefore be thought of as an
algebraic model of gossip in a network in which person i can communicate
only with their nearest neighbours, i − 1 and i + 1.
Proposition 5.3. Let n ∈ N. The gossip monoid Gn, the one-directional
gossip monoid ¯Gn, and the double Catalan monoid DCn, all satisfy the same
set of identities as the reflexive monoid Rn.
Proof. It is clear from the above definitions that DCn ⊆ Gn ⊆ Gn ⊆ Rn.
Thus Id(DCn) ⊇ Id(Gn) ⊇ Id(Gn) ⊇ Id(Rn). By Lemma 5.2, there is
a surjective monoid homomorphism from U : DCn → C U
n , from which it
follows that Id(C U
n ) ⊇ Id(DCn). The result then follows from the fact that
Rn and Cn satisfy the same identities [14].
(cid:3)
Now let S be an interval semiring and for each s ∈ S define: D(i, j; s)
to be the matrix with 1's on the diagonal and a single off-diagonal entry s
in position (i, j); and E(i, j; s) = D(i, j; s)D(j, i; s). Then we may define
monoids:
C U
n (S)
DCn(S)
Gn(S)
Gn(S)
:= hD(i, i + 1; s) : 1 ≤ i ≤ n − 1, s ∈ Si
:= hE(i, i + 1; s) : 1 ≤ i ≤ n − 1, s ∈ Si
:= hE(i, j; s) : 1 ≤ i, j ≤ n − 1, s ∈ Si
:= hD(i, j; s) : 1 ≤ i, j ≤ n − 1, s ∈ Si
Since S is an interval semiring, we note that each is a submonoid of Rn(S).
Proposition 5.4. Let S be an interval semiring. The monoids C U
n (S),
DCn(S), Gn(S) and Gn(S) satisfy the same identities as the monoid Rn(S).
Proof. It is clear from the definitions that
DCn ⊆ DCn(S) ⊆ Gn(S) ⊆ Gn(S) ⊆ Rn(S)
and
C U
n ⊆ C U
n (S) ⊆ Rn(S).
14
IDENTITIES IN UNITRIANGULAR AND GOSSIP MONOIDS
Thus by Proposition 5.4 and Theorem 4.3 we deduce that each of these
monoids satisfies the same set of identities.
(cid:3)
In the case where S is the subsemiring [0, +∞] of the (min-plus) tropical
semiring, it is straightforward to verify that the monoid Gn(S) is precisely
the lossy gossip monoid Gn of [2].
Corollary 5.5. The lossy gossip monoid is finitely based for n ≤ 4 and
non-finitely based otherwise.
Acknowledgements
The authors thank Professor Volkov for suggesting the topic of this article.
References
[1] K. Auinger, Yuzhu Chen, Xun Hu, Yanfeng Luo, and M. V. Volkov. The finite basis
problem for Kauffman monoids. Algebra Universalis, 74(3-4):333 -- 350, 2015.
[2] Andries E. Brouwer, Jan Draisma, and Bart J. Frenk. Lossy gossip and composition
of metrics. Discrete Comput. Geom., 53(4):890 -- 913, 2015.
[3] Yuzhu Chen, Xun Hu, and Yanfeng Luo. The finite basis property of a certain semi-
group of upper triangular matrices over a field. J. Algebra Appl., 15(9):1650177, 14,
2016.
[4] Yuzhu Chen, Xun Hu, Yanfeng Luo, and Olga Sapir. The finite basis problem for
the monoid of two-by-two upper triangular tropical matrices. Bull. Aust. Math. Soc.,
94(1):54 -- 64, 2016.
[5] Laure Daviaud, Marianne Johnson, and Mark Kambites. Identities in upper triangular
tropical matrix semigroups and the bicyclic monoid. J. Algebra, 501:503 -- 525, 2018.
[6] P. Fenner, M. Johnson, and M. Kambites. NP-completeness in the gossip monoid.
International Journal of Algebra and Computation.
[7] Volodymyr Mazorchuk and Benjamin Steinberg. Double Catalan monoids. J. Alge-
braic Combin., 36(3):333 -- 354, 2012.
[8] Sheila Oates and M. B. Powell. Identical relations in finite groups. J. Algebra, 1:11 -- 39,
1964.
[9] Peter Perkins. Bases for equational theories of semigroups. J. Algebra, 11:298 -- 314,
1969.
[10] L. M. Shneerson. On the axiomatic rank of varieties generated by a semigroup or
monoid with one defining relation. Semigroup Forum, 39(1):17 -- 38, 1989.
[11] Imre Simon. Hierarchies of events with dot -- depth one. Ph. D. Thesis, University of
Waterloo (1972).
[12] Andrew Solomon. Catalan monoids, monoids of local endomorphisms, and their pre-
sentations. Semigroup Forum, 53(3):351 -- 368, 1996.
[13] M. V. Volkov. The finite basis problem for finite semigroups. Sci. Math. Jpn.,
53(1):171 -- 199, 2001.
[14] M. V. Volkov. Reflexive relations, extensive transformations and piecewise testable
languages of a given height. Internat. J. Algebra Comput., 14(5-6):817 -- 827, 2004.
International Conference on Semigroups and Groups in honor of the 65th birthday of
Prof. John Rhodes.
[15] M. V. Volkov. A nonfinitely based semigroup of triangular matrices. In Semigroups,
algebras and operator theory, volume 142 of Springer Proc. Math. Stat., pages 27 -- 38.
Springer, New Delhi, 2015.
[16] Wen Ting Zhang, Ying Dan Ji, and Yan Feng Luo. The finite basis problem for infinite
involution semigroups of triangular 2 × 2 matrices. Semigroup Forum, 94(2):426 -- 441,
2017.
|
1711.05364 | 2 | 1711 | 2017-11-21T00:22:35 | Classification of $2$-dimensional evolution algebras, their groups of automorphisms and derivation algebras | [
"math.RA"
] | In the paper we give a complete classification of $2$-dimensional evolution algebras over algebraically closed fields, describe their groups of automorphisms and derivation algebras. | math.RA | math |
In the name of Allah, Most Gracious, Most Merciful.
CLASSIFICATION OF 2-DIMENSIONAL EVOLUTION ALGEBRAS, THEIR
GROUPS OF AUTOMORPHISMS AND DERIVATION ALGEBRAS
H.AHMED1, U.BEKBAEV2, I.RAKHIMOV3
1Department of Math., Faculty of Science, UPM, Selangor, Malaysia & Depart. of Math., Faculty of
2Department of Science in Engineering, Faculty of Engineering, IIUM, Kuala Lumpur, Malaysia
3Department of Math., Faculty of Science & Institute for Mathematical Research (INSPEM), UPM,
Science, Taiz University, Taiz, Yemen
Serdang, Selangor, Malaysia
Abstract. In the paper we give a complete classification of 2-dimensional evolution algebras over
algebraically closed fields, describe their groups of automorphisms and derivation algebras.
Keywords: evolution algebra, structure constants, automorphism, derivation.
MSC(2010): Primary: 15A72; Secondary: 22F50, 20H20, 17A60.
1. Introduction
The classification problem of finite dimensional algebras and description of their invariants with
respect to the basis changes is one of the important problems of algebra. One of the interesting class
of algebras is the class of evolution algebras. In the present paper we give a complete classification
of 2-dimensional evolution algebras over any algebraically closed field, describe their groups of auto-
morphisms and algebras of derivations. For the further information, related to similar problems, the
reader is refereed to [1, 2].
2. Classification of 2-dimensional evolution algebras
Let A be an n-dimensional algebra over an algebraic closed field F with a multiplication · given by
a bilinear map (u, v) 7→ u · v whenever u, v ∈ A. If e = (e1, e2, ..., en) is a basis of A over F then one
can represent this bilinear map by a matrix A =(cid:16)Ak
i,j(cid:17) ∈ M at(n × n2; F) as follows
u · v = eA(u ⊗ v)
for any u = eu, v = ev, where u = (u1, u2, ..., un), and v = (v1, v2, ..., vn) are column vectors of
u and v, respectively, ei · ej = A1
i,jen whenever i, j = 1, 2, ..., n. The matrix
A ∈ M at(n × n2; F) is called the matrix of structure constants (MSC) of A with respect to the basis
e. Further we do not differentiate A and its MSC A.
i,je2 + ... + An
i,j e1 + A2
It is known that under change of the basis e = (e1, e2, ..., en) by g ∈ GL(n, F) the matrix A changes
according to the rule B = gA(g−1)⊗2 that motivates to give the following definition.
Definition 2.1. n-dimensional algebras A, B, given by their matrices of structural constants A, B,
are said to be isomorphic if B = gA(g−1)⊗2 holds true for some g ∈ GL(n, F).
emails: 1houida [email protected]; [email protected]; [email protected].
1
2
H. Ahmed, U.Bekbaev, I.Rakhimov.
Let us recall definition of evolution algebras which are the object on focus of the paper.
Definition 2.2. An n-dimensional algebra E is said to be an evolution algebra if it admits a basis
{e1, e2, ..., en} such that ei · ej = 0, whenever i 6= j and i, j = 1, 2, ..., n.
The observations above in 2-dimensional case is viewed as follows. Let A be a 2-dimensional algebra
then on a basis e = (e1, e2) we have
A = A1
A2
1,1 A1
1,1 A2
1,2 A1
1,2 A2
2,1 A1
2,2
2,1 A2
2,2 ! ∈ M at(2 × 4; F)
ξ2 η2 (cid:19) ∈ GL(2, F) one has
and B = gA(g−1)⊗2, where for g−1 =(cid:18) ξ1 η1
(g−1)⊗2 = g−1 ⊗ g−1 =
ξ2
1
ξ1ξ2
ξ1ξ2
ξ2
2
ξ1η1
ξ1η2
ξ2η1
ξ2η2
η2
ξ1η1
1
ξ2η1 η1η2
ξ1η2 η1η2
η2
ξ2η2
2
.
Here is a theorem on description of all evolution algebra structures on a 2-dimensional vector space
over F.
Theorem 2.3. Over any algebraically closed field F every nontrivial 2-dimensional evolution algebra
is isomorphic to only one of the algebras listed below by MSC:
c 0 0 1 (cid:19) , where bc 6= 1, (b, c) ∈ F2,
• E1(c, b) ≃ E1(b, c) =(cid:18) 1 0 0 b
1 0 0 0 (cid:19) , where b ∈ F,
• E2(b) =(cid:18) 1 0 0 b
1 0 0 0 (cid:19) ,
• E3 =(cid:18) 0 0 0 1
• E4 =(cid:18) 1 0 0 1
0 0 0 0 (cid:19) ,
• E5 =(cid:18) 1
• E6 =(cid:18) 0 0 0 1
0 0 0 0 (cid:19) .
1 (cid:19) ,
−1 0 0
0 0 −1
Proof. Let E be a nontrivial evolution algebra given by E =(cid:18) a 0 0 b
E′ =(cid:18) α′1 α′2 α′3 α′4
β′1 β′2 β′3 β′4 (cid:19) = gE(g−1)⊗2, where g−1 =(cid:18) ξ1 η1
c 0 0 d (cid:19) and
ξ2 η2 (cid:19) . For the entries of E′ we have
(2.1)
2(bη2 − dη1)),
1(aη2 − cη1) + ξ2
1
∆ (ξ2
α′1 =
α′2 = α′3 = 1
∆ (ξ1η1(aη2 − cη1) + ξ2η2(bη2 − dη1)),
1
∆ (η2
α′4 =
1
∆ (ξ2
β′1 =
β′2 = β′3 = 1
∆ (ξ1η1(−aξ2 + cξ1) + ξ2η2(−bξ2 + dξ1)),
1
∆ (η2
β′4 =
1(aη2 − cη1) + η2
1(−aξ2 + cξ1) + ξ2
2(bη2 − dη1)),
2(−bξ2 + dξ1)),
1(−aξ2 + cξ1) + η2
2(−bξ2 + dξ1)),
Evolution algebras
3
where ∆ = ξ1η2 − ξ2η1. In particular, one has
Note also that
(cid:18) α′2
β′2 (cid:19) =(cid:18) ξ1 η1
c d (cid:19)(cid:18) ξ1η1
(cid:18) a b
ξ2η2 (cid:19) .
ξ2 η2 (cid:19)−1
ξ2 η2 (cid:19)−1
(cid:18) α′1 α′4
β′1 β′4 (cid:19) =(cid:18) ξ1 η1
c d (cid:19)(cid:18) ξ2
(cid:18) a b
1
ξ2
2
η2
1
η2
2 (cid:19) ,
which shows that α′1β′4 − α′4β′1 = 0 whenever ad − bc = 0.
Now we are searching possibilities to choose the base changes that make α′1, α′4, β′1, β′4 as simple
as possible and
that means
(2.2)
α′2 = α′3 = β′2 = β′3 = 0
ξ2 η2 (cid:19)−1
(cid:18) ξ1 η1
c d (cid:19)(cid:18) ξ1η1
(cid:18) a b
ξ2η2 (cid:19) =(cid:18) 0
0 (cid:19) .
We make use the following case by case considerations.
Case 1. ad − bc 6= 0. In this case (2.2) is equivalent to ξ1η1 = ξ2η2 = 0. Let us consider g =(cid:18) ξ1
0
Then ∆ = ξ1η2 and
0
η2 (cid:19) .
η2
2
ξ1
Due to ad − bc 6= 0 one has the following cases:
Case 1.1. a 6= 0, d 6= 0. In this case one can make α′1 = 1, β′4 = 1 to get
α′1 = aξ1, α′4 = b
, β′4 = dη2.
, β′1 = c
ξ2
1
η2
E1(b, c) =(cid:18) 1 0 0 b
c 0 0 1 (cid:19) , where bc 6= 1.
Case 1.2. a 6= 0, d = 0. In this case β′4 = 0 and one can make α′1 = 1, β′1 = 1 to get
E2(b) =(cid:18) 1 0 0 b
1 0 0 0 (cid:19) , where b 6= 0.
Case 1.3. a = 0, d 6= 0. In this case α′1 = 0 and one can make β′4 = 1, α′4 = 1 to get E′ =
c 0 0 1 (cid:19) , where c 6= 0. But E′ is isomorphic to E2(c).
(cid:18) 0 0 0 1
Case 1.4 a = 0, d = 0. In this α′1 = 0, β′4 = 0 and one can make β′1 = α′4 = 1 to get
E3 =(cid:18) 0 0 0 1
1 0 0 0 (cid:19) .
Case 2. ad − bc = 0.
Case 2.1. Both (a, b), (c, d) are nonzero and (c, d) = λ(a, b). In this case (2.1) is equivalent to
aξ1η1 + bξ2η2 = 0, α′1 =
η2 − λη1
∆
(aξ2
1 + bξ2
2), α′4 =
η2 − λη1
∆
(aη2
1 + bη2
2),
β′1 = −
ξ2 − λξ1
∆
(aξ2
1 + bξ2
2), β′4 = −
ξ2 − λξ1
∆
(aη2
1 + bη2
2).
4
H. Ahmed, U.Bekbaev, I.Rakhimov.
Case 2.1.1. a + bλ2 6= 0. Put ξ2−λξ1 = 0. Then ξ1 6= 0, the equation aξ1η1+bξ2η2 = ξ1(aη1+bλη2)
implies aη1 + bλη2 = 0. If b 6= 0 then η2
η1
= − a
bλ , ∆ = ξ1(η2 − λη1) and
β′1 = β′4 = 0, α′1 = (a + bλ2)ξ1, α′4 =
η2
1
ξ1
a(a + bλ2)
bλ2
.
It implies that in this case one can make α′1 = 1, α′4 equal to one or zero, depending on a, to get
0 0 0 0 (cid:19) or E′ =(cid:18) 1 0 0 0
E4 =(cid:18) 1 0 0 1
0 0 0 0 (cid:19) , which is isomorphic to E2(0).
0 0 0 0 (cid:19) .
α′1 = 1 one gets E′ =(cid:18) 1 0 0 0
The last E′ is isomorphic to E2(0). If b = 0 then η1 has to be zero, α′1 = aξ1, α′4 = 0, so by making
Case 2.1.2. a + bλ2 = 0. Note that in this case a, b, λ have to be nonzero and therefore one can
make ξ2 = η1 = 0. Then ∆ = ξ1η2, and
α′1 = aξ1, α′4 =
aλξ2
1
η2
bλ2 = −1, β′1 = bλ2
It implies that one can make α′1 = 1, β′4 = 1 to get α′4 = a
1 (cid:19) .
E5 =(cid:18) 1
, β′4 = bλη2.
0 0 −1
−1 0 0
bη2
2
ξ1
, β′1 =
a = −1 and
Case 2.2. c = d = 0. In this case
α′1 =
β′1 = −
η2
∆
ξ2
∆
(aξ2
1 + bξ2
2), α′2 = α′3 =
(aξ1η1 + bξ2η2), α′4 =
η2
∆
(aη2
1 + bη2
2),
(aξ2
1 + bξ2
2), β′2 = β′3 = −
(aξ1η1 + bξ2η2), β′4 = −
ξ2
∆
(aη2
1 + bη2
2).
η2
∆
ξ2
∆
Taking ξ2 = 0, η1 = 0 results in
α′1 = aξ1, α′2 = α′3 = 0, α′4 =
bη2
2
ξ1
, β′1 = β′2 = β′3 = β′4 = 0.
Case 2.2.1. a 6= 0. Then one can make α′1 = 1, α′4 = 1 or 0, depending on b to get
E4 =(cid:18) 1 0 0 1
0 0 0 0 (cid:19) or E′ =(cid:18) 1 0 0 0
0 0 0 0 (cid:19) ,
respectively. The last E′ is isomorphic to E2(0).
Case 2.2.2. a = 0. Then
α′1 = 0, α′2 = α′3 = 0, α′4 =
bη2
2
ξ1
, β′1 = β′2 = β′3 = β′4 = 0,
and one can make α′4 = 1 to get
Case 2.3. a = b = 0. In this case we have
E6 =(cid:18) 0 0 0 1
0 0 0 0 (cid:19) .
α′1 = −
(cξ2
1 + dξ2
2), α′2 = α′3 = −
(cξ1η1 + dξ2η2), α′4 = −
η1
∆
(cη2
1 + dη2
2),
β′1 =
(cξ2
1 + dξ2
2), β′2 = β′3 =
(cξ1η1 + dξ2η2), β′4 =
ξ1
∆
(cη2
1 + dη2
2),
η1
∆
ξ1
∆
η1
∆
ξ1
∆
Evolution algebras
5
which is similar to that of c = d = 0 case. A justification, similar to the case of c = d = 0 shows that
such algebras are isomorphic to those considered earlier.
(cid:3)
Remark 2.4. The following classification theorem on complex evolution algebras has been stated in
[1]. By the next theorem we restated the result by MSC.
Theorem 2.5. Every nontrivial 2-dimensional complex evolution algebra is isomorphic to exactly one
evolution algebra presented below by its MSC:
E1 :(cid:18) 1 0 0 0
E3 :(cid:18) 1 0 0 −1
E5a,b :(cid:18) 1 0 0 b
0 0 0 0 (cid:19) ,
0 0 0 0 (cid:19) , E2 :(cid:18) 1 0 0 1
1 0 0 0 (cid:19) ,
1 0 0 −1 (cid:19) , E4 :(cid:18) 0 0 0 0
1 0 0 c (cid:19) ,
a 0 0 1 (cid:19) , E6c :(cid:18) 0 0 0 1
where ab 6= 1, c 6= 0 and E5a,b ≃ E5b,a , E6c ≃ E6c′ ⇔ c
c′ = cos 2kπ
3 + isin 2kπ
3
for some k ∈ {0, 1, 2}.
Let us compare the list given above and that of Theorem 2.3:
E2(0) ≃ E1, E4 ≃ E2, E5 ≃ E3, E6 ≃ E4, E1(a, b) = E5a,b , E2(c−3) ≃ E6c,
where the last isomorphism is due to (cid:18) 1 0 0 c−3
1 0 0
conclude that in Theorem 2.5 the algebra E3 = E60 is missed.
0 (cid:19) = gE6c (g−1)⊗2 at g = (cid:18) 0
c2 0 (cid:19) . So we
c
3. The groups of automorphisms of 2-dimensional evolution algebras
Let i ∈ F stand for an element with i2 = −1, I =(cid:18) 1 0
0 1 (cid:19) and g =(cid:18) x y
t (cid:19) .
z
If E is an algebra given by MSC E then its group of automorphisms Aut(E) is presented as follows
(3.1)
Aut(E) = {g ∈ GL(2, F) : gE − E(g ⊗ g) = 0}.
Theorem 3.1. Automorphism groups of all evolution algebra structures on 2-dimensional vector space
over an algebraically closed field F of characteristic not 2 are given as follows.
• Aut(E1(b, c)) = {I}, if b 6= c,
• Aut(E2(b)) = {I}, if b 6= 0,
• Aut(E1(b, b)) =(cid:26)I,(cid:18) 0 1
• Aut(E2(0)) =(cid:26)(cid:18) 1
• Aut(E3) =(cid:26)I,(cid:18) 0 1
1 0 (cid:19)(cid:27) , if b2 6= 1,
t 1 − t (cid:19) : t 6= 1(cid:27) ,
1 0 (cid:19) ,(cid:18) t
0 t2 (cid:19)(cid:18) t2 0
0
0
0
t
t (cid:19) ,(cid:18) 0
t2 0 (cid:19) ,(cid:18) 0 t2
0 (cid:19)(cid:27) ,
t
0
• Aut(E4) =(cid:26)I,(cid:18) 1
• Aut(E5)) =(cid:26)(cid:18) t
0 −1 (cid:19)(cid:27) ,
t (cid:19) :
1 − t
1 − t
where t = − 1
2 + i
t 6= 1
2(cid:27) ,
√3
2 ,
6
H. Ahmed, U.Bekbaev, I.Rakhimov.
s
0
t (cid:19) :
• Aut(E6) =(cid:26)(cid:18) t2
t 6= 0, s ∈ F(cid:27) .
c 0 0 1 (cid:19) . Then
Proof. Indeed, let E = E1(b, c) =(cid:18) 1 0 0 b
gE1(b, c) − E1(b, c)(g ⊗ g) =(cid:18) x − x2 + cy − bz2 −xy − btz −xy − btz −bt2 + bx + y − y2
ct − cx2 + z − z2 −cxy − tz −cxy − tz
t − t2 − cy2 + bz (cid:19) .
Therefore to describe the automorphisms we have to solve the system of equations:
(3.2)
x − x2 + cy − bz2 = 0,
ct − cx2 + z − z2 = 0,
−xy − btz = 0,
−cxy − tz = 0,
−bt2 + bx + y − y2 = 0,
t − t2 − cy2 + bz = 0.
The equations 3 and 4 of the system of equations (3.2) imply that tz(bc − 1) = 0.
Case 1. b 6= c. In this case, the system above has only one solution g = I due to bc − 1 6= 0.
Case 2. b = c. In this case also the system of equations (3.2) has only one solution: (cid:18) 0 1
Let now E = E2(b) =(cid:18) 1 0 0 b
1 0 0 0 (cid:19) . Then
1 0 (cid:19) .
gE2(b) − E2(b)(g ⊗ g) =(cid:18) x − x2 + y − bz2 −xy − btz −xy − btz −bt2 + bx − y2
t − x2 + z
−y2 + bz
−xy
−xy
(cid:19) .
To find g one has to solve the following system of equations with respect to x, y, z and t:
(3.3)
x − x2 + y − bz2 = 0,
t − x2 + z = 0,
−xy − btz = 0,
−xy = 0,
−bt2 + bx − y2 = 0,
−y2 + bz = 0.
We make the following case by case consideration:
Case 1. b 6= 0. Due to xy = zt = 0 one has only two cases:
Case 1.1. x = t = 0, yz 6= 0. In this case the equation 2 of the system of equations (3.3) implies
z = 0. So there is no nontrivial g, Aut(E2(b)) = {I}.
Case 1.2. xt 6= 0, y = z = 0. In this case we have x = t = 1, hence, g = I.
Case 2. b = 0. One has y = x2 − x, t = x2 − z, y = 0 therefore x has to be 1, t = 1 − z and
0
z 1 − z (cid:19) , where z 6= 1.
g =(cid:18) 1
Let E = E3 =(cid:18) 0 0 0 1
1 0 0 0 (cid:19) . Then
gE3 − E3(g ⊗ g) =(cid:18) y − z2 −tz −tz −t2 + x
t − x2 −xy −xy −y2 + z (cid:19) .
Evolution algebras
7
√3
2 , t =
2 +i
Due to (3.1) we have two cases:
Case 1. xt 6= 0, y = z = 0. In this case due to t = x2, x = t2 one has x = 1, t = 1 or x = − 1
− 1
Case 2. x = t = 0, yz 6= 0. Similarly in this case one comes to y = 1, z = 1 or y = − 1
− 1
√3
2 or x = − 1
√3
2 , t = − 1
√3
2 .
2 + i
2 − i
2 − i
√3
2 or y = − 1
√3
2 , z = − 1
√3
2 .
2 − i
2 + i
2 − i
√3
2 , z =
2 + i
Let E = E4 =(cid:18) 1 0 0 1
0 0 0 0 (cid:19) . Then
gE4 − E4(g ⊗ g) =(cid:18) x − x2 − z2 −xy − tz −xy − tz −t2 + x − y2
0
0
z
z
(cid:19) .
In this case g = I or g =(cid:18) 1
Let E = E5 =(cid:18) 1
−1 0 0
0
0 −1 (cid:19) .
0 0 −1
1 (cid:19) . Then
gE5 − E5(g ⊗ g) =(cid:18) x − x2 − y + z2 −xy + tz −xy + tz
−t + x2 + z − z2
xy − tz
xy − tz
t2 − x + y − y2
t − t2 + y2 − z (cid:19) .
Due to (3.1) one has the system of equations:
(3.4)
which can be rewritten as follows
x − x2 − y + z2 = 0,
−t + x2 + z − z2 = 0,
xy − tz = 0,
xy − zt = 0,
t2 − x + y − y2 = 0,
t − t2 + y2 − z = 0
y = x − x2 + z2,
t = x2 + z − z2,
x(x2 − x − z2) + z(x2 + z − z2) = 0,
x2 − z2 = −(x2 − x − z2)2 + (x2 − (z2 − z))2.
Case 1. z 6= 0. Then x2 + z − z2 = −x(x2−x−z2)
of equations (3.4) implies that
z
and substitution it into the last equation of the system
z2(x2 − z2) = (x2 − z2)(x2 − x − z2)2,
(x2 − z2)((x2 − x − z2)2 − z2) = 0.
Case 1.1. x2 − z2 = 0. Then x = ±z, y = ±z, t = z, i.e., g is singular.
Case 1.2. (x2 − x − z2)2 − z2 = 0. Then one has x2 −x−z2 = ±z, y = ∓z, t = x±z +z, x±z +z =
z = ∓x. Therefore there are two cases:
Case 1.2.1. x2 − x − z2 = z, y = −z, t = x + 2z, 2x + 2z = 0. One has x = −z, y = −z, t = z
−x(±z)
and g is singular.
Case 1.2.2. x2 − x − z2 = −z, y = z, t = x. This case implies that z = 1−x and g =(cid:18) x
is an automorphism, where x 6= 1
2 .
Case 2. z = 0. Then one has y = −(x2 − x), t = x2, x2(x − 1) = 0 and x2 = −(x2 − x)2 + x4. So
x = 1, y = 0, t = 1 and one gets the trivial automorphism.
1 − x
1 − x
x (cid:19)
8
H. Ahmed, U.Bekbaev, I.Rakhimov.
Let E = E6 =(cid:18) 0 0 0 1
0 0 0 0 (cid:19) . Then
gE6 − E6(g ⊗ g) =(cid:18) −z2 −tz −tz −t2 + x
0
0
0
z
so g =(cid:18) t2 y
0
t (cid:19) , where t 6= 0.
(cid:19) ,
(cid:3)
In the case of characteristic 2 the corresponding result is as follows.
Theorem 3.2. Automorphism groups of all 2-dimensional evolution algebras over an algebraically
closed field F of characteristic 2 are given as follows.
• Aut(E1(b, c)) = {I}, if b 6= c,
• Aut(E2(b)) = {I}, if b 6= 0,
• Aut(E1(b, b)) =(cid:26)I,(cid:18) 0 1
• Aut(E2(0)) =(cid:26)(cid:18) 1
• Aut(E3) =(cid:26)I,(cid:18) 0 1
1 0 (cid:19)(cid:27) , if b2 6= 1,
t 1 − t (cid:19) : t 6= 1(cid:27) ,
1 0 (cid:19) ,(cid:18) t
0 t2 (cid:19)(cid:18) t2 0
0
0
0
t
t (cid:19) ,(cid:18) 0
t2 0 (cid:19) ,(cid:18) 0 t2
0 (cid:19)(cid:27) ,
t
• Aut(E4) = {I},
where t2 + t + 1 = 0,
• Aut(E5) =(cid:26)(cid:18) t
• Aut(E6) =(cid:26)(cid:18) t2
0
1 − t
s
t (cid:19) :
1 − t
t (cid:19) :
t ∈ F(cid:27) ,
t 6= 0, s ∈ F(cid:27) .
4. Derivation algebras of 2-dimensional evolution algebras
If E is an algebra given by MSC E then the algebra of its derivations Der(E) is presented as follows
Der(E) = {D ∈ M (2; F) : E(D ⊗ I + I ⊗ D) − DE = 0}.
Theorem 4.1. Derivations of all 2-dimensional evolution algebras over an algebraically closed field
F of characteristic not 2, 3 are given as follows.
• Der(E1(b, c)) = {0},
• Der(E2(b)) = {0}, if b 6= 0,
0
• Der(E3) = Der(E4) = {0},
• Der(E2(0)) =(cid:26)(cid:18) 0
• Der(E5) =(cid:26)(cid:18) −t
• Der(E6) =(cid:26)(cid:18) 2t s
t −t (cid:19) : t ∈ F(cid:27) ,
t −t (cid:19) : t ∈ F(cid:27) ,
t (cid:19) : t, s ∈ F(cid:27) .
0
t
Proof. Let D =(cid:18) x y
If E = E1(b, c) =(cid:18) 1 0 0 b
t (cid:19) be any element in M (2; F).
c 0 0 1 (cid:19) then
z
Evolution algebras
E1(b, c)(D ⊗ I + I ⊗ D) − DE1(b, c) =(cid:18)
and one has to solve the system of equations:
x − cy
−ct + 2cx − z
(4.1)
x − cy = 0,
−ct + 2cx − z = 0,
y + bz = 0,
cy + z = 0,
2bt − bx − y = 0,
t − bz = 0
y + bz y + bz 2bt − bx − y
cy + z
cy + z
t − bz
9
(cid:19)
to find the derivations. The equations 3, 4 of the system of equations (4.1) imply z(1 − bc) = 0.
Therefore due to bc 6= 1 one has x = y = t = z = 0 and D = 0, which implies that Der(E1(b, c)) = {0}.
Let E = E2(b) =(cid:18) 1 0 0 b
1 0 0 0 (cid:19) . Then
which implies due to (3.1) that x = y = 0, t = −z, bz = 0. So E2(b) has a nontrivial derivation
x − y
y + bz y + bz 2bt − bx
−t + 2x − z
y
y
−bz (cid:19) ,
E2(b)(D ⊗ I + I ⊗ D) − DE2(b) =(cid:18)
z −z (cid:19) if and only if b = 0.
0
D =(cid:18) 0
Let E = E3 =(cid:18) 0 0 0 1
1 0 0 0 (cid:19) . Then
E3(D ⊗ I + I ⊗ D) − DE3 =(cid:18) −y
0 0 0 0 (cid:19) . Then
E4(D ⊗ I + I ⊗ D) − DE4 =(cid:18) x
−z
and one gets D = 0.
Let E = E4 =(cid:18) 1 0 0 1
−t + 2x y y
z
z 2t − x
−z (cid:19)
y + z y + z 2t − x
0
0
−z (cid:19)
and we get D = 0.
0 0 −1
−1 0 0
Let E = E5 =(cid:18) 1
1 (cid:19) . Then
E5(D ⊗ I + I ⊗ D) − DE5 =(cid:18)
z −z (cid:19) .
0 0 0 0 (cid:19) . Then
and one easily comes to D =(cid:18) −z
Let E = E6 =(cid:18) 0 0 0 1
z
x + y
y − z
y − z −2t + x − y
t − 2x − z −y + z −y + z
t + z
(cid:19)
E6(D ⊗ I + I ⊗ D) − DE6 =(cid:18) 0 z
0 0 0
z 2t − x
−z (cid:19)
and one obtains D =(cid:18) 2t y
t (cid:19) .
0
(cid:3)
10
H. Ahmed, U.Bekbaev, I.Rakhimov.
Here are the corresponding results in the case of characteristic 2 and 3.
Theorem 4.2. Derivations of all 2-dimensional evolution algebras over an algebraically closed field
F of characteristic 2 are given as follows.
• Der((E1(b, c)) = {0} ,
• Der(E2(b)) = {0}, if b 6= 0,
0
• Der(E3) = {0},
• Der(E2(0)) =(cid:26)(cid:18) 0
• Der(E4) =(cid:26)(cid:18) 0 0
• Der(E5) =(cid:26)(cid:18) t −t
• Der(E6) =(cid:26)(cid:18) 0 s
t −t (cid:19) : t ∈ F(cid:27) ,
0 t (cid:19) : t ∈ F(cid:27) ,
t −t (cid:19) : t ∈ F(cid:27) ,
0 t (cid:19) : t, s ∈ F(cid:27) .
Theorem 4.3. Derivations of all 2-dimensional evolution algebras over an algebraically closed field
F of characteristic 3 are given as follows.
• Der(E1(b, c)) = {0},
• Der(E2(b)) = {0}, if b 6= 0,
0
• Der(E4) = {0},
• Der(E2(0)) =(cid:26)(cid:18) 0
• Der(E3) =(cid:26)(cid:18) 2t 0
• Der(E5) =(cid:26)(cid:18) −t
• Der(E6) =(cid:26)(cid:18) 2t s
t −t (cid:19) : t ∈ F(cid:27) ,
t (cid:19) : t ∈ F(cid:27) ,
t −t (cid:19) : t ∈ F(cid:27) ,
t (cid:19) : t, s ∈ F(cid:27) .
t
0
0
The second author's research is supported by FRGS14-153-0394, MOHE and the third author ac-
knowledges MOHE for a support by grant 01-02-14-1591FR.
Acknowledgments
[1] J.M. Casas, M. Ladra, B.A. Omirov, U.A. Rozikov, On evolution Algebras, Algebra Colloq., 21, 2014, 331 -- 342.
[2] J.P. Tian, Evolution Algebras and their Applications, Springer, Berlin, Heidelberg, 2008, doi.org/10.1007/978-3-540-
74284-5.
References
|
1901.04261 | 1 | 1901 | 2019-01-14T12:34:19 | 2-local derivations on infinite-dimensional Lie algebras | [
"math.RA"
] | The present paper is devoted to study 2-local derivations on infinite-dimensional Lie algebras over a field of characteristic zero. We prove that all 2-local derivations on the Witt algebra as well as on the positive Witt algebra are (global)derivations, and give an example of infinite-dimensional Lie algebra with a 2-local derivation which is not a derivation. | math.RA | math |
2-LOCAL DERIVATIONS ON INFINITE-DIMENSIONAL LIE
ALGEBRAS
SHAVKAT AYUPOV1
,
2, BAXTIYOR YUSUPOV3
Abstract. The present paper is devoted to study 2-local derivations on infinite-
dimensional Lie algebras over a field of characteristic zero. We prove that all 2-local
derivations on the Witt algebra as well as on the positive Witt algebra are (global)
derivations, and give an example of infinite-dimensional Lie algebra with a 2-local
derivation which is not a derivation.
Keywords: Lie algebras, Witt algebra, positive Witt algebra, thin Lie algebra,
derivation, 2-local derivation.
AMS Subject Classification: 17A32, 17B30, 17B10.
1. Introduction
In 1997, Šemrl [8] introduced the notion of 2-local derivations and 2-local automor-
phisms on algebras. Namely, a map ∆ : L → L (not necessarily linear) on an algebra
L is called a 2-local derivation if, for every pair of elements x, y ∈ L, there exists a
derivation Dx,y : L → L such that Dx,y(x) = ∆(x) and Dx,y(y) = ∆(y). The notion of
2-local automorphism is given in a similar way. For a given algebra L, the main prob-
lem concerning these notions is to prove that they automatically become a derivation
(respectively, an automorphism) or to give examples of local and 2-local derivations or
automorphisms of L, which are not derivations or automorphisms, respectively. Solu-
tion of such problems for finite-dimensional Lie algebras over algebraically closed field
of zero characteristic were obtained in [1, 2] and [3]. Namely, in [2] it is proved that
every 2-local derivation on a semi-simple Lie algebra L is a derivation and that each
finite-dimensional nilpotent Lie algebra, with dimension larger than two admits 2-local
derivation which is not a derivation. Concerning 2-local automorphism, Chen and Wang
in [3] prove that if L, is a simple Lie algebra of type Al, Dl or Ek, (k = 6, 7, 8) over an
algebraically closed field of characteristic zero, then every 2-local automorphism of L,
is an automorphism. Finally, in [1] Ayupov and Kudaybergenov generalized this result
of [3] and proved that every 2-local automorphism of a finite-dimensional semi-simple
Lie algebra over an algebraically closed field of characteristic zero is an automorphism.
Moreover, they show also that every nilpotent Lie algebra with finite dimension larger
than two admits 2-local automorphisms which is not an automorphism.
In the present paper we study 2-local derivations on infinite-dimensional Lie algebras
over a field of characteristic zero.
In Section 2 we give some preliminaries concerning Witt and positive Witt algebras.
In Section 3 we give a general form of derivations on the positive Witt algebra. In Sec-
tion 4 we prove that every 2-local derivations on Witt algebra and on the positive Witt
algebra are automatically derivations. We also show that so-called thin Lie algebras
admit 2-local derivations which are not derivations.
1
2
SH. A. AYUPOV, B. B. YUSUPOV
2. Preliminaries
In this section we give some necessary definitions and preliminary results.
A derivation on a Lie algebra L is a linear map D : L → L which satisfies the Leibniz
law, that is,
D([x, y]) = [D(x), y] + [x, D(y)]
for all x, y ∈ L. The set of all derivations of L with respect to the commutation operation
is a Lie algebra and it is denoted by Der(L). For all a ∈ L, the map ad(a) on L defined
as ad(a)x = [a, x], x ∈ L is a derivation and derivations of this form are called inner
derivation. The set of all inner derivations of L, denoted ad(L), is an ideal in Der(L).
Let A = C[x, x−1] be the algebra of all Laurent polynomials in one variable over a
field of characteristic zero F. The Lie algebra of derivations
Der(A) = span(cid:26)f (x)
d
dx
: f ∈ C(cid:2)x, x−1(cid:3)(cid:27)
with the Lie bracket is called a Witt algebra and denoted by W . Then [4] W is an
infinite-dimensional simple algebra which has the basis (cid:8)ei : ei = xi+1 d
the multiplication rule
dx , i ∈ Z(cid:9) and
[ei, ej] = (j − i)ei+j, i, j ∈ Z.
We also consider the infinite-dimensional positive part W + of the Witt algebra. The
positive Witt algebra W + is an infinite-dimensional Lie algebra [7] which has the basis
(cid:8)ei : ei = xi+1 d
dx , i ∈ N(cid:9) and the multiplication rule
[ei, ej] = (j − i)ei+j, i, j ∈ N.
Recall that a map ∆ : L → L (not liner in general) is called a 2-local derivation if
for every x, y ∈ L, there exists a derivation Dx,y : L → L (depending on x, y) such that
∆(x) = Dx,y(x) and ∆(x) = Dx,y(y).
Since any derivation on the infinite-dimensional Witt algebra W is inner [6], it follows
that for this algebra the above definition of the 2-local derivation can be reformulated
as follows. A map ∆ on W is called a 2-local derivation on W, if for any two elements
x, y ∈ W there exists an element ax,y ∈ W (depending on x, y) such that
∆(x) = [ax,y, x], ∆(y) = [ax,y, y].
Henceforth, given a 2-local derivation on W, the symbol ax,y will denote the element
from W satisfying ∆(x) = [ax,y, x] and ∆(y) = [ax,y, y].
3. Derivations on the positive Witt algebra
Let us consider the following algebra W + + he0i = span{en : n = 0, 1, 2, ...} with the
multiplication rule
[en, em] = (m − n)en+m,
n, m ≥ 0.
It is clear that W + is an ideal in W + + he0i. Hence, any element a ∈ W + + he0i
defines a spatial derivation La on W + by the following way:
La(x) = [a, x], x ∈ W +.
Theorem 3.1. Let D be a derivation on W +. Then there exists an element a ∈ W + +
he0i such that
D = La.
(3.1)
2-LOCAL DERIVATIONS ON INFINITE-DIMENSIONAL LIE ALGEBRAS
3
Proof. Let D(e1) = α1e1 + α2e2 + ... + αnen. Take an element
a1 = −α3e2 −
α4
2
e3 − ... −
αn
n − 2
en−1.
Then
La1(e1) = [a1, e1] = α3e3 + α4e4 + ... + αnen = D(e1) − α1e1 − α2e2.
Setting D1 = D − La1 , we have a derivation D1 such that D1(e1) = α1e1 + α2e2.
Now we shall show that α2 = 0. Suppose that D1(e2) = Pk≥1
D1(e3) = D1([e1, e2]) = [D1(e1), e2] + [e1, D1(e2)] =
βkek. Then
(α2 + 6β3)e5 +
(k2 + k)βk+1ek+3,
[D1(e1), e4] +
[e1, D1(e4)] =
(4α2 + 12β3)e6 +
(k3 + 3k2 + 2k)βk+1ek+4,
1
2
1
1
3
2Xk≥3
1
6Xk≥3
= (α1 + β2)e3 +Xk≥2
kβk+1ek+2,
1
2
D1(e4) =
D1([e1, e3]) =
[D1(e1), e3] +
[e1, D1(e3)] =
= (2α1 + β2)e4 +
D1(e5) =
1
3
D1([e1, e4]) =
= (3α1 + β2)e5 +
1
2
1
2
1
3
1
3
D1(e5) = D1([e2, e3]) = [D1(e2), e3] + [e2, D1(e3)] =
= 2β1e4 + (α1 + 2β2)e5 + 4β3e6 +Xk≥3
(k2 − k + 2)βk+1ek+4.
Comparing the last two equalities we obtain that
α2 = β1 = 0, β2 = 2α1 and βk = 0, k ≥ 4.
Therefore
Set a2 = α1e0. Then
D1(e1) = α1e1 and D1(e2) = 2α1e2 + β3e3.
Setting D2 = D1 − La2, we have a derivation D2 such that D2(e1) = 0.
Now by induction we shall show that
La2(e1) = [α1e0, e1] = α1e1.
D2(ek) = (k − 1)β3ek+1, for all k ≥ 2.
We have
Assume that
Then
D2(e2) = D1(e2) − La2(e2) = 2α1e2 + β3e3 − 2α1e2 = β3e3.
D2(ek) = (k − 1)βkek+1.
D2(ek+1) =
1
k − 1
D2([e1, ek]) =
1
k − 1
[D2(e1), ek] +
1
k − 1
[e1, D2(ek)] = kβ3ek+2.
So
D2(ek) = (k − 1)β3ek+1, for all k ≥ 2.
4
SH. A. AYUPOV, B. B. YUSUPOV
Take the element a3 = β3e1, then
La3(ek) = [a3, ei] = [β3e1, ek] = (k − 1)β3ek+1, k ≥ 2.
This means that D2 = La3 . Thus
D = D1 + La1 = D2 + La1 + La2 = La3 + La2 + La3 = La,
where a = a1 + a2 + a3. The proof is complete.
Remark 3.2. Let D = La be a derivation of the form (3.1), where a = −
computations show that
(cid:3)
αiei. Direct
n
Pi=0
D(ej) =
n
Xi=1
αi(j + 1 − i)ei+j−1,
j ≥ 1.
4. 2-Local derivations on some infinite-dimensional Lie algebras
Now we shall give the main result concerning 2-local derivations on infinite-
dimensional Lie algebras.
4.1. 2-Local derivations on the Witt algebra.
Theorem 4.1. Let W be the Witt algebra over a field of characteristic zero. Then any
2-local derivation on W is a derivation.
For the proof of this Theorem we need several Lemmas.
Lemma 4.2. Let ∆ be a 2-local derivation on W such that
∆(e0) = ∆(e1) = 0.
Then ∆(ei) = 0 for all i ∈ Z.
Proof. Let i ∈ Z be a fixed index except 0, 1. There exists an element ae0,ei ∈ W such
that
∆(e0) = [ae0,ei, e0], ∆(ei) = [ae0,ei, ei].
Then
0 = ∆(e0) = [ae0,ei, e0] ="Xj∈Z
αjej, e0# =Xj∈Z
αjjej.
Thus αj = 0 for all j ∈ Z with j 6= 0. This means that ae0,ei = α0e0. Thus
i.e.,
∆(ei) = [ae0,ei, ei] = [α0e0, ei] = iα0ei,
∆(ei) = iα0ei.
(4.1)
Now take an element ae1,ei ∈ W such that
∆(e1) = [ae1,ei, e1], ∆(ei) = [ae1,ei, ei].
Then
0 = ∆(e1) = [ae1,ei, e1] ="Xj∈Z
βjej, e1# =Xj∈Z
βj(1 − j)ej,
and therefore βj = 0 for any j ∈ Z, j 6= 1. This means that ae1,ei = β1e1. Hence
∆(ei) = [ae1,ei, ei] = [β1e1, ei] = (1 − i)β1ei+1,
2-LOCAL DERIVATIONS ON INFINITE-DIMENSIONAL LIE ALGEBRAS
5
i.e.,
(4.2)
Taking into account that i 6= 0, 1, and comparing (4.1) and (4.2) we obtain that α0 =
β1 = 0, i.e.,
∆(ei) = (1 − i)β1ei+1.
∆(ei) = 0.
The proof is complete.
(cid:3)
Lemma 4.3. Let ∆ be a 2-local derivation on W such that ∆(ei) = 0 for all i ∈ Z.
Then ∆ ≡ 0.
Proof. Take an arbitrary element x = Pj∈Z
all i > nx. Then x =
nx
Pj=−nx
cjej. Let nx be an index such that ci = 0 for
cjej. There exists an element ae0,x ∈ W such that
∆(e0) = [ae0,x, e0], ∆(x) = [ae0,x, x].
Since 0 = ∆(e0) = [ae0,x, e0], it follows that
ae0,x = α0e0.
Therefore
∆(x) = [ae0,x, x] ="α0e0,
nx
Xj=−nx
cjej# = −
nx
Xj=−nx
cjα0jej.
(4.3)
Now take an index n > 2nx. Let aen,x ∈ W be an element such that
∆(en) = [aen,x, en], ∆(x) = [aen,x, x].
By lemma 4.2, ∆(en) = 0. Since
it follows that
Then
0 = ∆(en) = [aen,x, en],
aen,x = αnen.
nx
nx
∆(x) = [aen,x, x] =
[αnen, ciei] =
(n − i)αncien+i.
(4.4)
Xi=−nx
X−nx
Since n > 2nx, it follows that n+ i > nx for all i ∈ {−nx, . . . , nx}. Thus comparing (4.3)
and (4.4) we obtain that αn = 0. This means that ∆(x) = 0. The proof is complete. (cid:3)
Now we are in position to prove Theorem 4.1.
Proof of Theorem 4.1 Let ∆ be a 2-local derivation on W . Take a derivation De0,e1
such that
∆(e0) = De0,e1(e0) and ∆(e1) = De0,e1(e1).
Set ∆1 = ∆ − De0,e1. Then ∆1 is a 2-local derivation such that ∆1(e0) = ∆1(e1) = 0.
By lemma 4.2, ∆1(ei) = 0 for all i ∈ Z. By lemma 4.3, it follows that ∆1 ≡ 0. Thus
∆ = De0,e1 is a derivation. The proof is complete.
✷
6
SH. A. AYUPOV, B. B. YUSUPOV
4.2. 2-Local derivations on the positive Witt algebra.
Theorem 4.4. Any 2-local derivation on W + is a derivation.
For the proof of this Theorem we need several Lemmas.
Lemma 4.5. Let ∆ be a 2-local derivation on W + such that
∆(e1) = ∆(e2) = 0.
Then ∆(ej) = 0 for all j ∈ N.
Proof. By the definition of 2-local derivations, we can find a derivation D1 such that
Then
and therefore
Thus
∆(e1) = D1(e1),
∆(ej) = D1(ej).
0 = ∆(e1) = D1(e1) = a1e1 +Xi≥3
(2 − i)aiei,
ai = 0,
i 6= 2.
∆(ej) = D1(ej) = (j − 1)a2ej+1,
j ≥ 3.
(4.5)
Now take a derivation D2 such that
∆(e2) = D2(e2) = 0, ∆(ej) = D2(ej).
Thus
i.e.,
0 = ∆(e2) = D2(e2) = 2a(2)
1 e2 + a(2)
2 e3 +Xi≥4
(3 − i)a(2)
i ei+1,
a(2)
i = 0,
i 6= 3.
This means that
∆(ej) = (j − 2)a(2)
3 ej+2,
j ≥ 3.
(4.6)
Comparing (4.5) and (4.6) we obtain that
j ≥ 3.
Thus a2 = 0, and therefore ai = 0 for all i. So ∆(ej) = 0,
complete.
(j − 1)a2ej+1 = (j − 2)a(2)
3 ej+2,
Lemma 4.6. Let ∆(ei) = 0 for all i ∈ N. Then ∆ ≡ 0.
j ≥ 3. The proof is
(cid:3)
n
Proof. Let x =
xkek be a non zero element from W +. By (3.1) there exists an element
ae1,x ∈ W + + he0i such that ∆(e1) = [ae1,x, e1] and ∆(x) = [ae1,x, e1]. Since ∆(e1) = 0,
it follows that ae1,x = α(1)
Pk=1
1 e1. Then
n
∆(x) = [ae1,x, x] = α(1)
1
(k − 1)xkek+1.
(4.7)
Xk=2
Now take an number m such that m > 2n. Again by (3.1) there exists an element
aem,x ∈ W + + he0i such that ∆(em) = [aem,x, em] and ∆(x) = [aem,x, em]. Since ∆(em) =
0, it follows that aem,x = α(m)
m em. Then
∆(x) = α(m)
em,x
n
Xk=1
(k − m)xkek+m.
(4.8)
2-LOCAL DERIVATIONS ON INFINITE-DIMENSIONAL LIE ALGEBRAS
7
Taking into account that m > 2n, and comparing the right sides of the equalities (4.7)
and (4.8), we obtain that α(m)
(cid:3)
em,x = 0. Thus ∆(x) = 0. The proof is complete.
Now we are in position to prove Theorem 4.4.
Proof of Theorem 4.4 Let ∆ be a 2-local derivation on W +. Take a derivation De1,e2
such that
∆(e1) = De1,e2(e1) and ∆(e2) = De1,e2(e2).
Set ∆1 = ∆ − De1,e2. Then ∆1 is a 2-local derivation such that ∆1(e1) = ∆1(e2) = 0.
By lemma 4.5, ∆1(ei) = 0 for all i ∈ N. By lemma 4.6, it follows that ∆1 ≡ 0. Thus
∆ = De1,e2 is a derivation. The proof is complete.
✷
4.3. An example of a 2-local derivation which is not a derivation on an
infinite-dimensional Lie algebra. Let us consider the following (see [5]) so-called
thin Lie algebra L with a basis {en : n ∈ N}, which is defined by the following table of
multiplications of the basis elements:
[e1, en] = en+1, n ≥ 2.
and other products of the basis elements being zero.
Theorem 4.7. Any derivation D on the algebra L has the following form:
D(e1) =
D(e2) =
αiei,
βiei,
n
n
Xi=1
Xi=2
D(ej) = ((j − 2)α1 + β2)ej +
where αi, βi ∈ C, i = 1, . . . , n, and n ∈ N.
Proof. Let D be a derivation on L. We set
(4.9)
βi+2ei+j,
j ≥ 3,
n
Xi=1
We have
D(e1) =
n
Xi=1
αiei, D(e2) =
βiei.
n
Xi=1
D(e3) = D([e1, e2]) = [D(e1), e2] + [e1, D(e2)] =
=" n
Xi=1
αiei, e2# +"e1,
Xi=1
n
= (α1 + β2)e3 +
βi+2ei+3.
βiei# =
n
Xi=1
Using [e2, e3] = 0, we have
0 = D([e2, e3]) = [D(e2), e3] + [e2, D(e3)] =" n
Xi=1
βiei, e3# +
βi+2ei+3# = β1e4,
+"e2, (α1 + β2)e3 +
n
Xi=1
8
SH. A. AYUPOV, B. B. YUSUPOV
Thus β1 = 0. Further
D(e4) = D([e1, e3]) = [D(e1), e3] + [e1, D(e3)] =
=" n
Xi=1
αiei, e3# +"e1, (α1 + β2)e3 +
n
βi+2ei+3# =
n
Xi=1
= (2α1 + β2)e4 +
βi+2ei+4.
Xi=1
With similar arguments applied to the products [e1, ej] = ej+1 and by the induction
on j, it is easy to check that the following identities hold for j ≥ 3:
D(ej) = ((j − 2)α1 + β2)ej +
The proof is complete.
βi+2ei+j,
j ≥ 3.
n
Xi=1
(cid:3)
Theorem 4.8. Let L be the thin Lie algebra. Then L admits a 2-local derivation which
is not a derivation.
Proof. For x =
n
Pi=1
xiei ∈ L set
if x1 = 0,
if x1 6= 0.
xiei,
∆(x) =
0,
n
Pi=2
We shall show that ∆ is a 2-local derivation on L, which is not a derivation.
Firstly, we show that ∆ is not a derivation. Take the elements x = e1 + e2 and
y = −e1 + e2. We have
and
Thus
∆(x + y) = ∆(2e2) = 0
∆(x) + ∆(y) = ∆(e1 + e2) + ∆(−e1 + e2) = 2e2.
∆(x + y) 6= ∆(x) + ∆(y).
So, ∆ is not additive, and therefore is not a derivation.
Let us consider the linear maps D1 and D2 on L defined as:
where αk ∈ C, k = 2, . . . , m, and m ∈ N and
m
0,
Pk=2
D1(en) =
D2(en) =(0,
en,
αkek,
if n = 1,
if n ≥ 2,
if n = 1,
if n ≥ 2.
(4.10)
(4.11)
By (4.9), it follows that both D1 and D2 are derivations on L.
For any x =
that
nx
Pk=1
ny
Pk=1
xkek, y =
ykek ∈ L we need to find a derivation D = Dx,y such
∆(x) = D(x) and ∆(y) = D(y).
It suffices to consider the following three cases.
Case 1. Let x1 = y1 = 0. In this case, we take D ≡ 0, because ∆(x) = ∆(y) = 0.
2-LOCAL DERIVATIONS ON INFINITE-DIMENSIONAL LIE ALGEBRAS
9
Case 2. Let x1 = 0, y1 6= 0. In this case we take the derivation D1 of the form (4.10)
with α1 = 0, αk =
and
yk
y1
, 2 ≤ k ≤ ny. Then
∆(x) = 0 = D1(x)
ny
ny
∆(y) =
ykek = y1
Xk=2
yk
y1
ek = D1(y).
Xk=2
So, D is a derivation such that ∆(x) = D(x), ∆(y) = D(y).
Case 3. Let x1 6= 0, y1 6= 0. In this case we take the derivation D2 of the form (4.11).
Then
and
∆(x) =
∆(y) =
xkek = D2(x)
ykek = D2(y).
nx
Xk=2
Xk=2
ny
Therefore in all cases we constructed a derivation on L such that ∆(x) =
D(x), ∆(y) = D(y), i.e. ∆ is a 2-local derivation which is not a derivation. The
proof is complete.
(cid:3)
References
[1] Ayupov Sh.A., Kudaybergenov K.K., 2-Local automorphisms on finite dimensional Lie algebras,
Linear Algebra and its Applications, 507 , 121-131 (2016).
[2] Ayupov Sh.A., Kudaybergenov K.K., Rakhimov I.S., 2-Local derivations on finite-dimensional Lie
algebras, Linear Algebra and its Applications, 474, 1-11 (2015).
[3] Chen Z., Wang D., 2-Local automorphisms of finite-dimensional simple Lie algebras, Linear Al-
gebra and its Applications, 486, 335 -- 344 (2015).
[4] Kac V., Raina A., Bombay lectures on highest weight representations of infinite-dimensional Lie
algebras, World Sci.Singapore, 1987.
[5] Khakimdjanova K., Khakimdjanov Yu., Sur une classe d'algebres de Lie de dimension infinie,
Comm. Algebra, Vol.29(1), 177 -- 191 (2001).
[6] Ikeda T., Kawamoto N., On the derivations of generalized Witt algebras over a field of character-
istic zero, Hiroshima Math.J., 20, 47 -- 55 (1990).
[7] Millionshchikov D.V., Naturally graded Lie algebras (Carnot algebras) of slow growth, Mat. Sb.,
Forthcoming paper (Mi msb9055)
[8] Šemrl P., Local automorphisms and derivations on B(H), Proc. Amer. Math. Soc., 125, 2677 -- 2680
(1997).
1 V.I.Romanovskiy Institute of Mathematics, Uzbekistan Academy of Sciences, 81,
Mirzo Ulughbek street, 100170, Tashkent, Uzbekistan
2 National University of Uzbekistan, 4, University street, 100174, Tashkent, Uzbek-
istan
E-mail address: sh−[email protected]
3 National University of Uzbekistan, 4, University street, 100174, Tashkent, Uzbek-
istan
E-mail address: [email protected]
|
1604.02923 | 2 | 1604 | 2017-01-12T15:39:28 | Free nilpotent and nilpotent quadratic Lie algebras | [
"math.RA"
] | In this paper we introduce an equivalence between the category of the t-nilpotent quadratic Lie algebras with d generators and the category of some symmetric invariant bilinear forms on the t-nilpotent free Lie algebra with d generators. Taking into account this equivalence, t-nilpotent quadratic Lie algebras with d generators are classified (up to isometric isomorphism, and over any field of characteristic zero), in the following cases: d=2 and t<6, d=3 and t<4. | math.RA | math | 7 Free nilpotent and nilpotent quadratic Lie algebras
1
0
2
P. Benito, D. de-la-Concepci´on, J. Laliena
n
a
J
2
1
]
.
A
R
h
t
a
m
[
2
v
3
2
9
2
0
.
4
0
6
1
:
v
i
X
r
a
Abstract
In this paper we introduce an equivalence between the category of the t-
nilpotent quadratic Lie algebras with d generators and the category of some
symmetric invariant bilinear forms over the t-nilpotent free Lie algebra with
d generators. Taking into account this equivalence, t-nilpotent quadratic Lie
algebras with d generators are classified (up to isometric isomorphisms, and
over any field of characteristic zero), in the following cases: d = 2 and t ≤ 5,
d = 3 and t ≤ 3.
Keywords: Nilpotent Lie algebras, invariant nondegenerate symmetric
bilinear forms, free nilpotent Lie algebras.
Classification MSC 2010: 17B01, 17B30
1
Introduction
Let n be a Lie algebra over an arbitrary field K of characteristic zero. The Lie
algebra n is said to be nilpotent if nt+1 = 0, where nt is defined inductively as
n1 = n, ni = [ni−1, n]. In this case we call t the index of nilpotency of n and we say
that n is t-nilpotent or also t-step nilpotent (nt 6= 0). The chain of ideals of n:
n ⊇ n2 ⊇ · · · ⊇ nt ⊇ nt+1 ⊇ . . .
(1)
is the well-known lower central series of n. Hence, if n is t-nilpotent the lower central
series finishes after t + 1 steps.
The type of a nilpotent Lie algebra n is defined as the codimension of n2 in n.
Following M. A. Gauger [5, Section 1, Corollary 1.3], a set m = {x1, x2, . . . , xd}
generates n if and only if {x1 + n2, . . . xd + n2} is a basis of n/n2. So, the type of a
Lie algebra is the cardinal of every K-linearly independent set, m = {x1, x2, . . . xd},
such that t, the subspace generated by m, satisfies t⊕ n2 = n. The above conditions
imply that m = {x1, . . . , xd} generates n as K-algebra and therefore, we can see the
elements xi ∈ m as a minimal set of generators of n.
1
Let FL(d) be the free Lie algebra on a set of d generators. The free t-nilpotent
Lie algebra on d generators is denoted nd,t and defined as the quotient algebra
nd,t = FL(d)/FL(d)t+1
(2)
Any t-nilpotent Lie algebra n of type d is a homomorphic image of nd,t. According
to [5, Section 1, Propositions 1.4 and 1.5], n ∼= nd,t/I, with I an ideal of nd,t such
that I ⊆ n2
d,t 6⊆ I.
d,t and nt
The Lie algebra, n is said to be quadratic (also known as metric) if it is endowed
with a nondegenerate symmetric bilinear form B which is invariant, that is,
B([x, y], z) = B(x, [y, z])
(3)
The class of quadratic Lie algebras is related to physics and Riemannian geometry
(see M. Bordemann [2, Section 1] for a explicit description of several connections).
Any semisimple Lie algebra is quadratic by means of the Killing form, B(x, y) =
T r(ad x ◦ ad y). In fact, the non degeneration of this trace form characterizes (in
characteristic zero) the class of semisimple Lie algebras according to Cartan's Crite-
rion. In the early 1980s, by the independent work of several authors (Kac, Favre and
Santharouban, Medina and Revoy, Hofmann and Keith), a recursive method to con-
struct quadratic Lie algebras known as the double extension method was developed
(see [2, Theorem 2.2]). Starting with an abelian Lie algebra of dimension 0 or 1,
and using one-dimensional central extensions and skew symmetric derivations, the
method let us construct every finite-dimensional quadratic solvable Lie algebra. But
this method seems to be difficult in high dimensions due to the multistep procedure.
In recent years, we can find different research papers on the structure of nilpotent
quadratic Lie algebras and also about partial classifications of them. For instance,
the complete list of Lie algebras in this class has been given by G. Favre and I.J.
Santharoubane [4] in 1987 up to dimension 7 and in 2007 by I. Kath [9] up to
dimension 10 (only over the real field). The classification of solvable quadratic Lie
algebras over algebraically closed fields given by M.T. Duong and R. Ushirobira [3]
in 2014 includes the complete list of nilpotent quadratic of dimension 8. In [12], G.
Ovando showed that there are 2-step nilpotent quadratic Lie algebras of any type
d ≥ 3 with the exception of d = 4; this fact was previously pointed out by Tsou and
Walker [14] in 1957. Following L. Noui and Ph. Revoy, [11], the classification of
2-step nilpotent quadratic Lie algebras is reduced to the classification of alternating
trilinear forms. In 2012, V. del Barco and G. Ovando [1] proved that n3,2 and n2,3
are the unique free nilpotent quadratic Lie algebras using elementary properties of
quadratic algebras. The classifications we have mentioned are mainly based on the
double extension method. In this work we develop a general classification scheme
2
for quadratic nilpotent Lie algebras based on the use of invariant bilinear forms on
free nilpotent Lie algebras.
The paper is divided into 5 sections from number 2. Section 2 is devoted to
give some examples of quadratic Lie algebras and examples of invariant bilinear
forms on free nilpotent Lie algebras nd,t of small dimensions. The group Aut nd,t
of automorphisms of nd,t is described in Section 3. A categorical approach between
free nilpotent Lie algebras and their invariant bilinear forms and quadratic nilpotent
Lie algebras is explained in Section 4. This approach let us identify the isometric
isomorphism classes of quadratic t-nilpotent Lie algebras of type d with the union of
spaces of orbits of the natural action of the group Aut nd,t over some sets of invariant
bilinear forms. The results given in Section 4 are used in Section 5 to classify up to
isometric isomorphisms all the t-nilpotent quadratic Lie algebras in characteristic
zero, with d generators for d = 2 and t ≤ 5, d = 3 and t ≤ 3. The final Section
includes the complete list of these algebras (basis and associated invariant bilinear
form) over an algebraically closed field. Apart from the abelian 1-dimensional, there
are only six indecomposable quadratic algebras. The list includes the classification
of the nilpotent quadratic Lie algebras up to dimension 7 and most of nilpotent
quadratic Lie algebras of dimension 8. The section ends with the complete list of
these algebras over the real field.
Throughout the paper K will denote an arbitrary field of characteristic zero
except where otherwise specified.
2 Preliminaries and examples
Let nd,t be the free t-nilpotent Lie algebra on the set of generators m = {x1, . . . , xd}.
If si is the vector subspace generated by mi, and mi denotes the set of all possible
products made with i elements of m, we have that
nd,t = ⊕t
k=1sk,
(4)
where ⊕ denotes the direct sum as vector spaces. This decomposition gives us a
natural graduation on nd,t that provides many structure results of free nilpotent Lie
algebras in an easy way. Following M. Hall [7], we can get the well-known Hall basis
of nd,t whose elements are monomials in the generators. From M. Grayson and R.
Grossman [6, Definition 1.1] each element in the Hall basis is defined recursively as:
1. x1, . . . , xd are (ordered) elements of the basis of length 1.
2. The elements of lengths 1, . . . , r − 1 are simply ordered so that a < b if
length(a) < length(b).
3
3. If length(a) = s and length(b) = t and r = s + t, [a, b] is a basis element if:
a, b are basis elements and a > b; and in case a = [c, d], then b ≥ d.
From the Hall basis, inductively we get the dimension of any subspace sl, l = 1, . . . , t
as
µ(a)dl/a,
(5)
1
l Xal
where µ is the Moebius function.
Example 2.1. For d = 2, 3 and t ≤ 5, the Hall basis (ordered), Hd,t, of nd, t are
given as:
• H2,2 = {x1, x2, [x2, x1]},
• H2,3 = H2,2 ∪ {[[x2, x1], x1], [[x2, x1], x2]};
• H2,4 = H2,3 ∪ {[[[x2, x1], x1], x1], [[[x2, x1], x1], x2], [[[x2, x1], x2], x2]};
• H2,5 = H2,4∪{[[[[x2, x1], x1], x1], x1], [[[[x2, x1], x1], x1], x2], [[[x2, x1], x1], [x2, x1]],
[[[[x2, x1], x1], x2], x2], [[[x2, x1], x2], [x2, x1]], [[[[x2, x1], x2], x2], x2]};
• H3,2 = {x1, x2, x3, [x2, x1], [x3, x1], [x3, x2]};
• H3,3 = H3,2 ∪ {[[x2, x1], xj], [[x3, x1], xj], [[x3, x2], xk] : j = 1, 2, 3, k = 2, 3}.
For every Lie algebra g we can define recursively the upper central series:
Z1(g) = 0
and Zi(g) = {x ∈ g : [x, g] ⊆ Zi−1(g)}.
Note that the center of g, Z(g) = {x ∈ g : [x, g] = 0} is just Z2(g).
In the case that (g, B) be quadratic (then B is a nondegenerate symmetric invari-
ant bilinear form), we have the following relation of orthogonality (see A. Meedina
and Ph. Revoy [10, Proposition 1.2]) between the ideals in the lower central series
of g defined in (1) and those of its upper central series (a⊥ = {x ∈ g : B(x, a) = 0}):
(6)
(gi)⊥ = Zi(g)
Hence, in any Lie algebra equipped with an invariant nondegenerate symmetric
bilinear form the next equality holds:
dim g = dim gi + dim Zi(g)
(7)
For nilpotent Lie algebras, both series have the same length and this length deter-
mines the nilpotency index of the algebra. In fact, g is t-nilpotent if and only if
gt+1 = 0 if and only if Zt+1(g) = g. The upper central series of the free nilpotent
Lie algebra nd,t is:
(8)
Zi(nd,t) = ⊕k≥t+2−i nk
d,t .
4
Example 2.2. Any abelian Lie algebra n endowed with a nondegenerate bilinear
form B is a quadratic Lie algebra. Moreover, taking a fixed basis {x1, . . . , xd} for
n, two quadratic Lie algebras (n, B) and (n, B′) are (isometric) isomorphic if and
only if the matrices AB = (B(xi, xj)) and AB′ = (B′(xi, xj)) are congruents, i.e.:
AB′ = P tABP for some regular matrix P .
Example 2.3. From (7), in any quadratic Lie algebra g the following equality holds
dim Z(g) = dim g − dim g2. So, filiform algebras are not quadratic and the unique
quadratic quasifiliform Lie algebra is n2,3. Generalized Heissenberg algebras are not
quadratic (the general definition for GH is hn such that h2
n = Z(hn) = K · z).
The following example summarizes part of the results on quadratic free nilpotent
Lie algebras therein V. del Barco and G. Ovando [1]:
d,t. According to formulas
and dZ(d, 3) = d(d2−1)
Example 2.4. The center of nd,t is exactly, Z(nd,t) = nt
(5) and (7), the dimension of Z(nd,t) is equal to dZ(d, t) = 1
t Pat µ(a)dt/a. From [1,
Proposition 3.7], 2 ≤ d < dZ(d, t) in case t ≥ 4. On the other hand, dZ(d, 1) = d,
dZ(d, 2) = d(d−1)
. So, the only possible free nilpotent Lie
algebras are nd,1, n3,2 and n2,3. From Example 2.2, nd,1 is quadratic; several noniso-
morphic possibilities can occur depending on the base field. So are the Lie algebras
n2,3 and n3,2. In fact from G. Favre and L.J. Santharouban [4], over algebraically
closed fields of characteristic zero, any quadratic Lie algebra (n2,3, B) or (n3,2, B′) is
isometric to (the matrices are given in the basis H3,2 and H2,3):
2
3
(n3,2, ϕ3,2) :
0
0
0
0
0
0
0 0
1
0
0 0 −1 0
0
0
0 1
0
0
1 0
0
0 −1 0 0
0 0
1
0
0
0
0
0
0
,
(n2,3, ϕ2,3) :
0
0
0
0
0
0
1
0
0 −1 0
0
0
1
0
0 −1 0
0
1
0
0
0
0
0
(9)
The results in Section 5 of this paper show that (n3,2, ϕ3,2) is, up to isometries, the
unique quadratic 2-nilpotent Lie algebra of type 3 over any field of characteristic zero.
Over the real field, (n2,3, ϕ2,3) is one of the two nonisometric quadratic 3-nilpotent
Lie algebra of type 2. The other one is (n2,3,−ϕ2,3)
The next examples provide the whole vector space of invariant symmetric bilinear
forms of n2,t and n3,t for small nilpotent index. The matrices are given in the Hall
basis Hd,t; they follow from a straightforward computation by using the definition
of invariant bilinear form, and the Jacobi identity (this method is used in the proof
of Theorem 3.8 in [1] for n3,2).
Example 2.5. Any invariant symmetric bilinear form on n2,t for t ≤ 5 is of the
form:
5
BA1
2,1 =(cid:18)α β
β δ(cid:19) , BA1
2,2 =
α β 0
0
β δ
0 0
0
2,3 =
, BA1;γ
0
β
δ
0
0
α
γ
0 −γ 0
β
0
0
γ
0
0 −γ 0
0
0
γ
0
0
0
0
BA1;γ
2,4 =
0
β
δ
0
0
γ 0 0 0
α
0 −γ 0 0 0 0
β
0 0 0 0
0
γ
0 −γ 0
0 0 0 0
0 0 0 0
0
γ
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
(10)
(11)
(12)
2,5
=
0
0
0
0
BA1;γ;A2
0 γ
0
α
β
0
0 −γ
β
δ
0
0
0 −d −e −f
0
γ
0 −γ
0
d
e
0
0
e f
γ
0
0 −d
0
0
0 −e
0
0
0
0
0
0 −f
0
0
0
0
0
d
0
0
0
−d
0
0
0
0
e
d
0
0
0
0
−e
0
0
0
f
e
0
0
0
−f
0
β δ(cid:19) and A2 = (cid:18)d e
where A1 = (cid:18)α β
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 0 −d d −e
e −f
0
0 f
0 d
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 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
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 0
This implies that the quadratic dimension of n2,1, n2,2, n2,3, n2,4 and n2,5 are 3, 3, 4, 4
and 7 respectively.
e f(cid:19) are 2 × 2 symmetric matrices and γ ∈ K.
Example 2.6. Any invariant symmetric bilinear form of n3,t for t ≤ 3 is of the
form:
BA1
3,1 =
α β γ
β δ
ǫ
ǫ ω
γ
3,2 =
, BA1;λ
6
0
β
δ
ǫ
0
α
β
γ
0
0 −λ 0
0
λ
γ
ǫ
ω λ
λ 0
0
0
0
λ
0 −λ 0
0
0
0
0
0
0
0
0
0
(13)
b d
0 e
0
a b
0 c −b
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
0 0
0 0
0
0 −b −c 0 −d −e 0 −f
0
0
c
b
0
d
e
0
e f
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
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
c
e
0 f
0
0
0
0
0
0
0
0
0
0
0
0
0 λ
0
0
0 −λ 0 −a
0
0
0
0
0
0
0
0
0
0
0
α
β
γ
β
δ
ǫ
γ
ǫ
ω λ
0
0
λ a
0 −λ
0
b
0
0
λ
c
0
0 −a −b
0 −c 0
a
0 0
c
b
0 −b −d 0
0 −e 0
b
0 0
d
e
0 −f
0
c
0 0
f
e
and A2 =
BA1;λ;A2
3,3
=
(14)
where A1 =
λ ∈ K. So the quadratic dimension of n3,1, n3,2 and n3,3 are 6, 7 and 13 respectively.
are 3 × 3 symmetric matrices and
α β γ
ǫ
β δ
γ
ǫ ω
a b
c
b d e
c e f
3 The group of automorphisms of nd,t
For every a ∈ nd,t we have a unique decomposition a = a1 + · · · + at where ai ∈ si
and we can consider, for each k = 1, . . . , t, the projection endomorphism of nd,t,
which is a linear endomorphism:
ek : nd,t → nd,t
a
7→ ek(a) = ak
(15)
We notice that ek are idempotent endomorphisms of nd,t and:
e1 + · · · + et = Ind,t,
where Ind,t is the identity homomorphism of nd,t.
Following T. Sato [13, Proposition 3], any linear map ϕ : s1 → nd,t may be
extended to a (unique) endomorphism as an algebra Φϕ : nd,t → nd,t in the following
for any monomial of length k, a = [xi1 . . . xik ], Φϕ(a) = [ϕ(xi1) . . . ϕ(xik )].
way:
Moreover Φϕ is an automorphism if and only if {e1(ϕ(x1)), . . . e1(ϕ(xd))} is a set
In this
of linearly independent elements of s1 if and only if e1 ◦ ϕ ∈ GL(s1).
= Ind,t if Is1 : s1 → nd,t is defined over the generator elements of nd,t as
way, ΦIs1
Is1(xi) = xi.
7
Proposition 3.1. The group Aut nd,t decomposes as a semidirect product of the
subgroups H(d, t) = {Φϕ : ϕ ∈ GL(s1)} and N(d, t) = {Φρ : ρ = Is1 + σ, σ ∈
End (s1, n2
d,t)}. Even more, N(d, t) is a normal subgroup which is nilpotent and
H(d, t) is isomorphic to the group of regular d × d matrices with entries in K.
Proof. Firstly we note that the elements of H(d, t) are graded automorphisms and
H(d, t) ∩ N(d, t) = 1. For ϕi ∈ GL(s1), we have Φϕ1(Φϕ2)−1 = Φϕ1Φϕ−1
,
= Φϕ1ϕ−1
so H(d, t) is a subgroup. Analogously, given ρi = Is1 + σi where σi ∈ End (s1, n2
d,t),
−1σ2. In addition, Φρ1Φρ2 = Φρ with ρ = Is1 +σ1+Φρ1σ2
Φ−1
ρ2 = Φρ where ρ = Is1−Φρ2
and Φϕ1Φρ1(Φϕ1)−1 = Φρ where ρ = Is1 + Φϕ1σ1ϕ−1
1 . Hence, N(d, t) is a subgroup
and H(d, t) is contained in NAut nd,t(N(d, t)), the normaliser of N(d, t) in Aut nd,t.
Now every ϕ : s1 → nd,t that satisfies e1 ◦ ϕ ∈ GL(s1) decomposes in the form:
2
2
Φϕ = Φe1ϕΦρ, ρ = Is1 + e2(Φe1ϕ)−1e2ϕ + . . . et(Φe1ϕ)−1etϕ.
Hence N(d, t) is a normal subgroup and Aut nd,t is a semidirect product of H(d, t)
and N(d, t). The last assertion is easily checked.
For technical purposes, we give the following alternative description for the
If ϕ ∈ Aut nd,t, we have that
subgroups H(d, t) and N(d, t) in Proposition 3.1.
ϕ(si) ⊆ ⊕t
j=isj for i = 1, . . . , t. Therefore for j < k, ejϕek = 0. So
(16)
(17)
(18)
t
ϕ =
Xj,k=1
ejϕek = Xt≥j≥jk≥1
ejϕek,
and we have that
t
eiϕei
Xi=1
H(d, t) = {
N(d, t) = { Xt≥j>k≥1
ejϕek + Id / ϕ ∈ Aut nd,t}
/ ϕ ∈ Aut nd,t}
Using Proposition 3.1 and the Hall basis Hd,t described in Example 2.1, from
straightforward computations we get the following examples:
Example 3.2. If we consider in n2,3 the basis H2,3, the elements of Aut n2,3 can be
identified with matrices in GL(5). In fact, the elements of H(2, 3) and N(2, 3)are
given by matrices
0
A
0 det(A)
0
0
0
0
det(A)A
and
8
0
0
1
β
1 0
0 0
0 1
0 0
α β
0 0
δ γ
1 0
µ ǫ −α 0 1
,
(19)
where A ∈ GL(2), and α, β, δ, γ, ǫ, µ ∈ K. It is easy to check that N(2, 3) satisfies
N(2, 3)2 = 1 (here N(2, 3)2 means [N(2, 3), [N(2, 3), N(2, 3)]] with [N(2, 3), N(2, 3)]
the commutator of two subgroups, i.e, [P, Q] = {[p, q] = p−1q−1pq : p ∈ P, q ∈ Q}).
Example 3.3. Consider now n3,2 and the basis H3,2. The matrices representing the
elements of Aut n3,3 in relation to H3,2 are given by
A
B
0
A3,3 A3,2 A3,1
A2,3 A2,2 A2,1
A1,3 A1,2 A1,1
where A ∈ GL(3), B ∈ Mat3,3(K) (where Mat3,3(K) denotes the set of 3×3 matrices
with entries in K), and the Ai,j matrices are obtained deleting row i and column j
in A and taking the determinant of the resulting matrix. In this case H(3, 2) and
N(3, 2) are given respectively by matrices of the types
A
0
0
A3,3 A3,2 A3,1
A2,3 A2,2 A2,1
A1,3 A1,2 A1,1
where I3 denotes the 3 × 3 identity matrix.
and (cid:18) I3
B I3 (cid:19)
0
(20)
4 Free and quadratic nilpotent Lie algebras
In this section we shall introduce a new technique of constructing quadratic nilpotent
Lie algebras out of free nilpotent Lie algebras endowed with an invariant symmetric
bilinear form.
4.1. Categorical approach. Following [5], any t-step nilpotent Lie algebra n of
type d is a homomorphic image of nd,t. In fact, given an arbitrary isomorphism of
Lie algebras ϕ : nd,t/I → n the ideal I satisfies that nt
d,t. Indeed, if x ∈ I
and x /∈ n2
d,t, then from the introduction, x can be seen as a generator of nd,t, but
this is a contradiction because nd,t/I ∼= n and n has d generators. On the other
hand, if nt
d,t ⊆ I, we have (nd,t/I)t = 0 and then nt = 0, a contradiction with the
t-nilpotency of n.
Let (n, B) be a t-nilpotent quadratic Lie algebra of type d and ϕ : nd,t/I → n
be any isomorphism of Lie algebras. We can define on nd,t the following symmetric
bilinear form:
d,t 6⊆ I ⊆ n2
B1(x, y) = B(ϕ(x + I), ϕ(y + I)).
(21)
9
Using that ϕ is isomorphism and that B is an invariant form on n, we easily check
that B1 is an invariant form:
B1([x, y], z) = B(ϕ([x + I, y + I]), ϕ(z + I)] = B[(ϕ(x + I), ϕ(y + I)], ϕ(z + I))
= B(ϕ(x + I), [ϕ(y + I), ϕ(z + I)]) = B1(x, [y, z]).
We also note that if B1(x, y) = 0 for every y ∈ nd,t, then 0 = B(ϕ(x + I), z) for
every z ∈ n because ϕ is surjective. Since B is nondegenerate we get that x + I = 0,
that is Ker(B1) = I. Hence nt
Suppose now that nd,t is endowed with an invariant symmetric bilinear form U
d,t = Ker(U). For every x ∈ Ker(U) and
d,t 6⊆ Ker(B1) ⊆ n2
d,t.
and consider the orthogonal subspace n⊥
y, z ∈ nd,t
U([x, y], z) = U(x, [y, z]) = 0.
So, [Ker(U), nd,t] ⊆ Ker(U) which proves that Ker(U) is an ideal. Then we can de-
fine on the Lie algebra nd,t/Ker(U) the invariant symmetric nondegenerate bilinear
form
U(x + Ker(U), y + Ker(U)) = U(x, y).
(22)
Following the previous approach, if we start with a t-nilpotent quadratic Lie algebra
(n, B) of type d and ϕ : nd,t/I → n is an isomorphism of Lie algebras and B1 is the
invariant symmetric bilinear form defined in (21), then we have that (nd,t/I, B1) is
a nilpotent quadratic Lie algebra. Even more, ϕ is an isomorphism of Lie algebras
and an isometry from (nd,t/I, B1) onto (n, B), because
B1(x + I, y + I) = B1(x, y) = B(ϕ(x + I), ϕ(y + I)).
Previous discussion can be settled in the following result:
Proposition 4.1. Let (n, B) be a quadratic t-nilpotent Lie algebra of type d and
ϕ : nd,t/I → n be an isomorphism of Lie algebras. Then:
(i) The map B1 : nd,t × nd,t → K given by B1(x, y) = B(ϕ(x + I), ϕ(y + I)) is an
invariant symmetric bilinear form on nd,t.
(ii) The orthogonal subspace of B1 is exactly n⊥
d,t = Ker(B1) = I and satisfies that
d,t * I ⊆ n2
nt
d,t.
(iii) The map B1 : nd,t/I × nd,t/I → K defined as B1(x + I, y + I) = B1(x, y) is an
invariant nondegenerate symmetric bilinear form on nd,t/I.
(iv) ϕ is an isometry from (nd,t/I, B1) onto (n, B).
10
Let us define NilpQuadd,t the category whose objects are the t-nilpotent quadratic
Lie algebras (n, B) of type d, and whose morphisms are Lie homomorphisms
ϕ : (n, B) → (n′, B′)
such that B(x, y) = B′(ϕ(x), ϕ(y)). We will call these Lie homomorphisms, metric
Lie homomorphisms.
We define also Sym0(d, t) the category whose objects are the symmetric in-
variant bilinear forms B on the free Lie algebra nd,t for which Ker(B) ⊆ n2
d,t and
nt
d,t * Ker(B), and whose morphisms are defined as follows:
for any B1, B2 ∈
Obj(Sym0(d, t)), we introduce the set of metric Lie endomorphisms of nd,t that
respect the kernel of the bilinear form, i.e.:
MEnd⊥(B1, B2) :=
{ϕ ∈ End(nd,t) : B1(x, y) = B2(ϕ(x), ϕ(y)), ϕ(Ker(B1)) ⊆ Ker(B2)}.
The whole set of morphisms from B1 to B2 is defined as the quotient set
Hom(B1, B2) := MEnd⊥(B1, B2)/ ∼
where ∼ is the equivalence relation,
ϕ1 ∼ ϕ2 ⇐⇒ (ϕ1 − ϕ2)(nd,t) ⊆ Ker(B2) ∀ϕ1,ϕ2∈M End⊥(B1,B2)).
(23)
(24)
The morphisms of Sym0(d, t) are well-defined: if ϕ ∈ MEnd⊥(B1, B2) and ψ ∈
MEnd⊥(B2, B3) for B1, B2, B3 ∈ Obj(Sym0(d, t)), ψ◦ϕ(Ker(B1)) ⊆ (Ker(B3)), so
ψ◦ϕ ∈ MEnd⊥(B1, B3). Moreover, if [ϕ] = [ϕ′] and [ψ] = [ψ′], since (ϕ−ϕ′)(nd,t) ⊆
Ker(B2) and (ψ − ψ′)(nd,t) ⊆ Ker(B3) and
ψϕ − ψ′ϕ′ = ψ(ϕ − ϕ′) − (ψ′ − ψ)ϕ′,
we get (ψϕ − ψ′ϕ′)(nd,t) ⊆ Ker(B3). Hence, [ψϕ] = [ψ′ϕ′]
Now we define the functor
Qd,t : Sym0(d, t) → NilpQuadd,t
(25)
such that for every B ∈ Obj(Sym0(d, t))
Qd,t(B) = (nd,t/Ker(B), B),
where B : nd,t/Ker(B)×nd,t/Ker(B) → K is defined by B(x+Ker(B), y+Ker(B)) =
B(x, y). And, for every morphism [ϕ] ∈ Hom(B1, B2),
Qd,t([ϕ]) : (nd,t/Ker(B1), B1) → (nd,t/Ker(B2), B2)
7→ ϕ(x) + Ker(B2).
x + Ker(B1)
(26)
11
We remark that Qd,t is a well-defined functor for objects, because of Proposition
4.1. Also Qd,t is well-defined for morphisms. If ϕ, ψ ∈ MEnd⊥(B1, B2) such that
[ϕ] = [ψ], then
Qd,t([ϕ])(x + Ker(B1)) = ϕ(x) + Ker(B2)
and
Qd,t([ψ])(x + Ker(B1)) = ψ(x) + Ker(B2)
for every x ∈ nd,t. But, from (24), (ϕ−ψ)(nd,t) ⊆ Ker(B2), and therefore, Qd,t([ϕ])(x+
Ker(B1)) = Qd,t([ψ])(x + Ker(B1)).
The following theorem stablishes that the functor Qd,t defined in (25) provides
an equivalence between both categories, Sym0(d, t) and NilpQuadd,t.
Theorem 4.2. The categories Sym0(d, t) and NilpQuadd,t are equivalent.
Proof. We will show that the functor Qd,t is faithful and full, and that for every
object (n, B) in NilpQuadd,t there exists an object U in Sym0(d, t) such that
Qd,t(U) and (n, B) are isomorphic in NilpQuadd,t (that is, the functor is essentially
surjective or dense). So, according to [8, Proposition 1.3], we will have that both
categories are equivalent.
From Proposition 4.1 (iv), we get that Qd,t is dense. Now we prove that Qd,t is
a faithful functor. Indeed, if Qd,t([ϕ]) = Qd,t([ψ]), with [ϕ], [ψ] ∈ Hom(B1, B2) and
B1, B2 ∈ Obj(Sym0(d, t)) then ϕ(x) + Ker(B2) = ψ(x) + Ker(B2) for all x ∈ nd,t.
That is, (ϕ − ψ)(nd,t) ⊆ Ker(B2) and therefore [ϕ] = [ψ] by using (24).
Finally we check that Qd,t is a full functor. Let τ be any element in the set of
morphisms Hom(Qd,t(B1), Qd,t(B2)). For a fixed {x1, . . . , xd} set of generators of
nd,t, take yi ∈ nd,t such that yi + Ker(B2) = τ (xi + Ker(B1)) for i = 1, . . . , d. From
the universal mapping property of nd,t the correspondence xi → yi extends uniquely
to a Lie homormorphism ϕ : nd,t → nd,t. Since ϕ(xi) + Ker(B2) = yi + Ker(B2) =
τ (xi + Ker(B1)) and since the elements xi generate nd,t, we have τ π1 = π2ϕ where
πi : nd,t → nd,t/Ker(Bi) is the canonical surjection. This implies ϕ(Ker(B1)) ⊆
Ker(B2) and ϕ(x) + Ker(B2) = τ (x + Ker(B1)), for all x ∈ nd,t. Moreover, ϕ is a
metric endomorphism:
B2(ϕ(x), ϕ(y)) = B2(ϕ(x) + Ker(B2), ϕ(y) + Ker(B2)) =
B2(τ (x + Ker(B1), τ (y + Ker(B1)) =
B1(x + Ker(B1), y + Ker(B1)) = B1(x, y).
Therefore [ϕ] ∈ Hom(B1, B2) and Qd,t([ϕ]) = τ .
As a consequence of this equivalence of categories we have:
Corollary 4.3. For all B1, B2 ∈ Obj(Sym0(d, t)), the following assertions are
equivalent:
12
(i) B1 and B2 are isomorphic in Sym0(d, t).
(ii) Qd,t(B1) and Qd,t(B2) are isometrically isomorphic Lie algebras.
(iii) There exists a metric automorphism θ : (nd,t, B1) → (nd,t, B2).
Proof. The equivalence (i) ⇐⇒ (ii) follows from Theorem 4.2 and (ii) ⇐⇒ (iii)
follows from Proposition 1.6 in [5] and the proof therein.
Since Sym0(d, t) and NilpQuadd,t are equivalent categories, there is a bijection
between the isomorphism types of objects of each category. So, the classification of
t-nilpotent quadratic Lie algebras with d generators up to isometric isomorphisms
is the classification of objects in the category NilpQuadd,t up to isomorphism, and
this classification, in turn, is the one of objects in the category Sym0(d, t) up to
isomorphism.
4.2. The group Aut nd,t acting on Sym0(d, t). The group of automorphisms of
nd,t acts on the set Obj(Sym0(d, t)) in the natural way:
Aut nd,t × Obj(Sym0(d, t)) → Obj(Sym0(d, t)), (θ, B) → Bθ
(27)
where Bθ(x, y) = B(θ(x), θ(y)).
(nd,t, B) is a metric Lie isomorphism. Even more:
Indeed, for every θ ∈ Aut nd,t, θ : (nd,t, Bθ) →
Lemma 4.4. For all B ∈ Obj(Sym0(d, t)), the set OrbAut nd,t(B) = {Bθ : θ ∈
Aut nd,t} is equal to the set of bilinear invariant symmetric forms isomorphic to B
in the category Sym0(d, t). Therefore the number of orbits of the action described
in (27) is exactly the number of isomorphism types in the classification of t-nilpotent
quadratic Lie algebras of type d up to isometries.
Proof. For the first assertion, we remember that B and B′ are isomorphic objects
in Sym0(d, t) if and only if there exist ϕ ∈ MEnd⊥(B, B′) and τ ∈ MEnd⊥(B′, B)
such that [ϕτ ] = [τ ϕ] = [Ind,t].
It is clear that B is isomorphic to Bθ for every
θ ∈ Aut nn,t. Conversely, if B′ is isomorphic to B, from Corollary 4.3, there exists
a metric automorphism θ : (nd,t, B′) → (nd,t, B). So, B(θ(x), θ(x)) = B′(x, y) and
therefore, B′ = Bθ ∈ OrbAut nd,t(B).
From now on, we will denote as S2
0(d, t) the K-vector space
S2
0(d, t) = {B : B is an invariant symmetric bilinear form on nd,t}.
(28)
We note that the elements in Obj(Sym0(d, t)) are not closed by the sum of bilinear
forms. If B1, B2 ∈ Obj(Sym0(d, t)), it can happen that Ker(B1 + B2) * n2
d,t, and
also that nt
d,t ⊆ Ker(B1 + B2). On these occasions we can see B1 + B2 as a invariant
bilinear form on nd1,t with 1 ≤ d1 ≤ d, or on nd,t1, with 1 ≤ t1 ≤ t, respectively.
13
Also we notice that it could be interesting to consider the more general action:
Aut nd,t × S2
0(d, t) → S2
7→ Bθ
(θ, B)
0(d, t)
We observe that if B ∈ S2
0(d, t), and we define
B(ei, ej) : nd,t × nd,t →
K
(x, y) 7→ B(ei(x), ej(y))
(29)
then
B =
t
Xi,j=1
B(ei, ej),
where ek are the projections defined in (15). We also observe that B(ei, ej) = 0 if
i + j > t + 1. Defining for k = 1, . . . , t,
Bk =
t−k+1
Xi=1
B(ei, et−i−k+2),
(30)
we have the decomposition
t
where Bk for k = 1, . . . , t are invariant symmetric bilinear forms, because:
B =
Bk,
Xk=1
t−k+1
Bk(a, [b, c]) =
B(ei(a), et−i−k+2([b, c])) =
B(ei(a), Xl+m=t−i−k+2
[el(b), em(c)]) =
B(ei(a), [el(b), em(c)]) =
B([ei(a), el(b)], em(c)) =
t−k+1
t−k+1
Xi=1
Xi=1
Xi=1 Xl+m=t−i−k+2
Xm=1 Xl+i=t−m−k+2
Xm=1
t−k+1
t−k+1
14
B(et−m−k+2([a, b]), em(c)) = Bk([a, b], c)
Example 4.5. In Example 2.5, the sum decomposition BA;γ
be (matrices are given in the Hall basis H2,3)
2,3 = B1 + B2 + B3 will
0
0
0
γ
0
0 −γ 0
0
0
γ
0
0 −γ 0
γ
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 0
0 0 0 0 0
+
+
α β 0 0 0
β δ 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
.
The k-components related to invariant bilinear symmetric that we have intro-
duced in (30) have special features:
forms.
0(d, t), the following assertions holds:
determines completely the whole set of addends.
i=1 B(ei, et−i−k+2), any fixed addend B(ei, et−i−k+2),
i=1 B(ei, et−i−k+2) for k = 1, . . . , t are invariant symmetric bilinear
Proposition 4.6. For any B ∈ S2
(i) Bk = Pt−k+1
(ii) In the sum Bk = Pt−k+1
(iii) If B ∈ Obj(Sym0(d, t)) then B1 6= 0. In fact, B(ei, ej) 6= 0 for i + j = t + 1.
Proof. Item (i) has been already proved and item (ii) is a consequence of the in-
variance of B. To show (iii), we take into account that nt
d,t * Ker(B) if B ∈
Obj(Sym0(d, t)). So, as B(ei, et) = 0 for i = 2, . . . , t, B(e1, et) 6= 0; otherwise,
st = nt
d,t ⊆ Ker(B), a contradiction. So B1 6= 0, and from (ii) we get B(ei, ej) 6= 0
for i, j = 1, . . . , t such that i + j = t + 1.
Next, we consider the set of B1-components of every invariant symmetric bilinear
form B in S2
0 (d, t), that is:
S2
00(d, t) = {
t
Xi=1
B(ei, et−i+1) : B ∈ S2
0(d, t)}.
We note that S2
of S2
00(d, t) is a K-vector subspace of invariant symmetric bilinear forms
0(d, t). Moreover, if B ∈ S2
00(d, t) and ϕ ∈ H(d, t), from (17) we have
Bϕ =
t
t
Xi=1
Xi=1
t
t
B(ei
Xi=1
Bϕ(ei, et−i+1) =
Xj=1
B(eiϕei, et−i+1ϕet−i+1) ∈ S2
00(d, t)
ejϕej, et−i+1
ejϕej) =
t
Xj=1
15
So, we can conclude that S2
using (18) we get
00(d, t) is a H(d, t)-module. In the case that ϕ ∈ N(d, t),
Bϕ(ei, et−i+1) =
t
Xi=1
ejϕek), et−i+1(Id +Xj>k
ejϕek)) =
eiϕek, et−i+1) +
et−i+1ϕek) +
Bϕ =
t
t
Xi=1
Xi=1
Xi=1
t
B(Xi>k
B(Xi>k
B(ei(Id +Xj>k
Xi=1
B(ei, Xt−i+1>k
t
eiϕek, Xt−i+1>k
et−i+1ϕek) + B
and therefore (Bϕ)1 = B. That is, the action on B ∈ S2
00(d, t) of the elements of
N(d, t) provides invariant bilinear forms that have B as 1-component. Summarizing:
Lemma 4.7. For any B ∈ S2
ϕ ∈ N(d, t) then (Bϕ)1 = B.
00(d, t), Bϕ ∈ S2
00(d, t) for every ϕ ∈ H(d, t) and if
From S2
00(d, t) it is possible to obtain a large amount of elements in Obj(Sym0(d, t)).
In fact, for small d and t, we will prove in the next section that the orbits of S2
provide the whole set of elements in Sym0(d, t).
00(d, t)
5 Small type quadratic nilpotent Lie algebras
A quadratic Lie algebra (L, ϕ) is called decomposable if it contains a proper ideal I
such that ϕI×I is nondegenerate, otherwise L is called indecomposable. In this case,
L = I ⊕ I ⊥, ϕI ⊥×I ⊥ is nondegenerate and I ⊥ is an ideal. Hence, any quadratic Lie
algebra is the direct sum of orthogonal indecomposable Lie algebras. If L = I1 ⊕ I2
is a decomposable quadratic nilpotent Lie algebra, the type of L is the sum of the
types of I1 and I2 and the nilpotent index of L is equal to max{t1, t2}, where ti is
the nilpotent index of Ii. Then, any quadratic abelian Lie algebra of dimension ≥ 2
is decomposable and the quadratic nonabelian nilpotent Lie algebras of type 2 are
indecomposable.
We recall that two matrices, A and B, are congruent if there exists a regular
matrix P such that B = P tAP .
Theorem 5.1. Over any arbitrary field of characteristic 0, the objects in the cat-
egory Sym0(d, 1), d ≥ 1, are the nondegenerate symmetric bilinear forms BA
determined by a d × d symmetric and regular matrix A. Moreover:
2,1
(i) Two bilinear forms BA1
d,1 and BA2
d,1 are isomorphic in Sym0(d, 1) if and only if
A1 and A2 are congruent.
16
(ii) Any quadratic Lie algebra of the form Qd,1(BA
d,1) decomposes
as an orthogonal direct sum of 1-dimensional ideals which are quadratic Lie
algebras.
d,1) = (nd,1, BA
(iii) (nd,1, BA1
d,1) and (nd,1, BA2
d,1) are isomorphic as metric Lie algebras if and only if
the matrices A1 and A2 are congruent.
Proof. The kernel of any bilinear invariant form B ∈ Sym0(d, 1) is contained in
n2
2,1 = 0. So, these forms are of maximal rank d. The assertion (i) follows from the
fact that Aut nd,1 = GL(d). Then the bilinear forms in the same orbit as BA1
2,1 are
BP tA1P
, with P ∈ GL(d). Finally, (ii) follows from the fact that any bilinear form
on a finite dimensional vector space (characteristic 6= 2) has an orthogonal basis and
(iii) follows from Corolllary 4.3.
2,1
Theorem 5.2. Over any field K of characteristic 0, the set Obj(Sym0(2, t)) is
empty if t = 2, 4. For t = 3, 5 we have the following:
(i) The objects in Sym0(2, 3) are the invariant forms BA;γ
2,3 described in Example
2.5 and given by any 2× 2 symmetric matrix A and any 0 6= γ ∈ K. Moreover,
BA1;γ
2,3
2,3 are isomorphic in Sym0(2, 3) if and only if
and BA2;δ
δ
γ ∈ K2.
(ii) The objects in Sym0(2, 5) are the invariant forms BA1;γ;A2
described in Ex-
ample 2.5 for any γ ∈ K and any 2 × 2 symmetric matrices A1 and A2 with
rank A2 ≥ 1. Moreover, BA1;γ;A2
are isomorphic in Sym0(2, 5)
if and only if B2 = (det P )2P tA2P for some 2 × 2 regular matrix P .
and BB1;δ;B2
2,5
2,5
2,5
Proof. According to Proposition 4.6, any B ∈ Obj(Sym0(2, t)) satisfies that 0 6= B1.
We will prove that for t = 2, 4 any invariant bilinear form B satisfies B1 = 0. The
invariant form B1 is completely determined by B(ei, ej), as defined in (29), for any
pair (i, j) such that i + j = t + 1. If t = 2, we consider the pair (i, j) = (1, 2) and
note that: s1 = span < x1, x2 > and s2 = span < [x2, x1] >. From the invari-
ance of B, we have that B(x2, [x2, x1]) = B([x2, x2], x1]) = 0 and B(x1, [x2, x1]) =
−B([x1, x1], x2]) = 0. For t = 4, we take the pair (1, 4); in this case, s4 = span <
[[[xr, xs], xp], xq] : (r, s) = (2, 1), (1, 2), p, q ∈ {1, 2} > and B(xr, [[[xr, xs], xp], xr]) =
0 and B(xr, [[[xr, xs], xp], xs]) = −B([xr, xs], [[xr, xs], xp]) = 0. Hence, for t = 2, 4
the set Obj(Sym0(2, t)) is empty because of (iii) in Proposition 4.6.
if and only if γ 6=
2,3 ) is a quadratic Lie algebra.
∈ Obj(Sym0(2, 3)) and consider θP ∈ Aut n2,3 the
2,3 is not contained in the kernel of BA;γ
is nondegenerate and (n2,3, BA;γ
(cid:16) 0
0
2,3
It is clear that n3
0. In this case BA;γ
2,3
v w (cid:17),γ
Now, let B
(cid:16) u
2,3
0 (cid:17),γ
, B
2,3
v
0
17
automorphism given by the matrix (in the Hall basis H2,3):
1
0
0 0 0
0
1
0 0 0
0
0
1 0 0
γ v − 1
− 1
2γ w 0 1 0
1
2γ u
0
0 0 1
P =
.
We have that
v
0
0
0
0
0
0
0
0
0
P =
P t
0
0
γ
0
0 −γ 0
0
γ
0
0
0 −γ 0
0
γ
0
u
γ
v w 0 −γ 0
0
0
γ
0
0 −γ 0
0
γ
0
On the other hand, for γ, δ ∈ K such that γδ 6= 0 and δ
automorphism θQ ∈ Aut n2,3 given by the matrix:
0
0
0
0
0
0
Q =
0
0
0
ǫ 0 0
0
0
0 1 0
0 0 ǫ
0
0 0 0 ǫ2 0
0 0 0
ǫ
0
γ = ǫ2 with ǫ ∈ K, the
satisfies:
Qt
0
0
0
0
0
0
γ
0
0 −γ 0
0
0
γ
0
0 −γ 0
0
0
γ
0
0
0
0
This implies from Lemma 4.4 that B
the category Sym0(2, 3) and so are B
δ
γ = ǫ2 with ǫ ∈ K.
Finally, let as assume that BA1;γ
Q =
v w (cid:17),γ
v
(cid:16) u
2,3
(cid:16) 0
0
2,3
0
0 (cid:17),γ
.
0
0
0
0
0
0
δ
0
0 −δ 0
0
0
δ
0
0 −δ 0
0
0
δ
(cid:16) 0
0
2,3
(cid:16) 0
0
2,3
0
0
0
0 (cid:17),γ
and B
0 (cid:17),δ
and B
0
0
0
are isomorphic in
in case γδ 6= 0 and
from previous paragraph, B 02×2;γ
Lemma 4.4, we can find θR ∈ Aut n2,3 where (ǫ = ad − bc 6= 0)
2,3
2,3
2,3 and BA2;δ
and B 02×2;δ
2,3 are isomorphic in Sym0(2, 3). Then,
are also isomorphic and because of
R =
0
b
a
d
c
0
α1 α2
ǫ
α3 α4
α2
α5 α6 −α1
0
0
0
0
0
0
ǫa ǫb
ǫc
ǫd
18
such that RtB 02×2;γ
equivalence in (i) follows.
2,3 R = B 02×2;δ
2,3
. The last matrix equation implies γǫ2 = δ and the
In the sequel we consider the objects in the category Sym0(2, 5). For any
invariant bilinear form BA1;γ;A2
2,5 is not contained
in the kernel of this form if and only if rank A2 ≥ 1 if and only if the kernel is
contained in n2
on n2,5, it is easily checked that n3
, B 02×2;0;A2
2,5. Let BA1;γ;A2
2,5
2,5
2,5
∈ Obj(Sym0(2, 5)) where
A1 =(cid:18) p q
r s (cid:19) , A2 =(cid:18) d e
e f (cid:19) 6= 02×2.
We are interested in proving that both objects are isomorphic. For this, we will use
the automophism subgroup N(2, 5). The matrix elements in this subgroup (with
respect to the Hall basis H2,5) are of the form:
I2
0
0
0
0
0
A 1
C A′
0
I2
D C ′ A′′
I3
E D′ C ′′ A′′′
X(A, C, D, E) =
0
0
0
0
I6
,
where A ∈ M1×2(K), C ∈ M2×2(K), D ∈ M3×2(K), E ∈ M6×2(K), but A′ ∈
M2×1(K), A′′ ∈ M3×2(K), A′′′ ∈ M6×3(K) are determined by A, and C ′ ∈ M3×1(K),
C ′′ ∈ M6×2(K) are determined by C and D′ ∈ M6×1(K) is determined by D. Let
us assume firstly that f 6= 0 and consider the matrices
A = 01×2, C =(cid:18) 0
γ
2f
0
0 (cid:19) , D = 03×2, Et = 0 0 0 0
0 0 0 0
0
r
2f
γ2−4pf
8f 2
2f 2 !
er−2qf
For the automorphism θX(A,C,D,E) we have that A′, A′′, A′′′ and D′ are zero matrices
and (C ′)t = (0 0 − γ
2f ) and
X(A, C, D, E)tB 02×2;0;A2
2,5
X(A, C, D, E) = BA1;γ;A2
2,5
.
(We note that, denoting by BθX the invariant bilinear form on the left part and
using Proposition 4.6, for checking the identity, we only need to compute BθX (ei, ei)
for i = 1, 2, 3; this computations involves the matrices C ′, A2, A′
2 and E.)
2, A′′
Next let us suppose that f = 0, d 6= 0 and consider the matrices
A = 01×2, C =(cid:18) 0 − γ
0
In this case, (C ′)t = (− γ
0 (cid:19) , D = 03×2, Et =(cid:18)
2d 0 0) and
4dr−γ2
8d2
q
d
2d
− p
2d 0 0 0 0
0
0 0 0 0 (cid:19) .
X(A, C, D, E)tB 02×2;0;A2
2,5
X(A, C, D, E) = BA1;γ;A2
2,5
.
19
Finally, if f = d = 0, we have e 6= 0 and we can consider the automorphism
θ(A, C, D, E) where
A = 01×2, C =(cid:18) γ
2e 0
0
0 (cid:19) , D = 03×2, Et =(cid:18) 0 0
0 0
q
0
e
r
2e 0
p
2e 0
0
0 (cid:19) .
Then, (C ′)t = (0 − γ
2e 0) and
X(A, C, D, E)tB 02×2;0;A2
2,5
X(A, C, D, E) = BA1;γ;A2
2,5
.
From the previous reasoning, the final assertion in item (ii) of this theorem is equiv-
alent to the following statement: B 02×2;0;A2
are isomorphic if and only
if B2 = (det P )2P tA2P . The necessary condition follows from the general form of
any automorphism of n2,5:
and B 02×2;0;B2
2,5
2,5
0
0
0
0
P
A P ′
0
0
0
C A′ P ′′
0
0
D C ′ A′′ P ′′′
0
E D′ C ′′ A′′′ P ′′′′
,
X(P ; A, C, D, E) =
where P is a regular matrix, P ′ = det P and P ′′ = det P · P ; the rest of the
submatrices under the principal diagonal are as in the case N(2, 5). From the notion
of isomorphism we get
X(P ; A, C, D, E)tB 02×2;0;A2
2,5
X(P ; A, C, D, E) = B 02×2;0;B2
2,5
and so we get the relation B2 = (det P )2P tA2P . The converse follows by using the
automorphism θX(P ;0,0,0,0).
If K is a field, we can consider the group K∗ with the product of the field, where
K∗ = K−{0}. Then (K∗)2 = {k2 : k ∈ K∗} is a subgroup of K∗, and we can consider
the quotient group K∗/(K∗)2.
Corollary 5.3. Over any field K of characteristic 0, there are no nilpotent quadratic
Lie algebras of type 2 and nilpotent index 2 or 4. In addition, up to metric isomor-
phism:
(i) The nilpotent quadratic Lie algebras of type 2 and nilpotent index 3 are of the
2,3) are
form (n2,3, B0;γ
isometrically isomorphic if and only if δ and γ are congruent module (K∗)2.
2,3 ) for any 0 6= γ ∈ K. Moreover, (n2,3, B0;γ
2,3 ) and (n2,3, B0;δ
20
n2,5
form (
Ker B0;0;A
, B0;0;A
(ii) The nilpotent quadratic Lie algebras of type 2 and nilpotent index 5 are of the
) for any arbitrary 2 × 2 nonzero symmetric matrix A.
, B0;0;A
) are isometrically isomorphic
Moreover, (
if and only if B = (det P )2P tAP for some P regular 2 × 2 matrix.
, B0;0;B
Ker B0;0;A
Ker B0;0;B
) and (
n2,5
n2,5
2,5
2,5
2,5
2,5
2,5
2,5
All the quadratic algebras described in (i) and (ii) are indecomposable. We notice that
in (i) the number of nonisomorphic algebras is the cardinal of the group K∗/(K∗)2.
Also we notice that in (ii), if K satisfies that for every α ∈ K there exists δ ∈ K such
that δ3 = α, then (
) are isomorphic as metric
, B0;0;B
, B0;0;A
) and (
n2,5
n2,5
Ker B0;0;A
2,5
2,5
Ker B0;0;B
2,5
2,5
Lie algebras if and only if A and B are congruent.
Proof. We just need to apply the previous Therorem 5.2 and Corollary 4.3 to prove
(i) and (ii). For the last statement about (ii), we notice that if B = P tAP with P
regular, then R = ( 3pdet(P ))−1P satisfies that B = det(R)2RtAR.
From now on, we will tackle the case Sym0(3, t). From Example 2.6, the invari-
ant bilinear forms on n3,2 and n3,3 are given by the matrices on the Hall basis:
0
Wγ
0 (cid:19) ,
A1 Wγ A′
2
Wγ A2
0
A′t
0
0
2
(n3,2, BA1;γ) :(cid:18) A1 Wγ
(n3,3, BA1;γ;A2) :
where A1 and A2 are 3 × 3 symmetric matrices, δ ∈ K and
= γ · C, where C =
0
1
0 −1 0
;
0
1
A′
2 is a 3 × 8 matrix completely determined by A2 as follows:
A2 =
2 =
0
c
−a
0
−b −c 0 −d −e 0 −f
and A′
Wδ =
0
γ
0 −γ 0
0
γ
c
a b
b d e
c
e f
b
0
c −b
a
0
d
e
b
0
e
f
0
0
0
0
.
(31)
(32)
(33)
In the following lemma, we discuss the indecomposibility of Q3,t(B) for B ∈
Sym0(3, t).
Lemma 5.4. Let γ be any scalar and A1, A2 symmetric 3×3 matrices over a field of
characteristic 0. The objects in Sym0(3, 2) are the bilinear forms BA1;γ
3,3 with γ 6= 0
and for the bilinear forms BA1;γ;A2
, we have that:
3,3
21
(i) rank A2 ≥ 2 if and only if rank A′
object in Sym0(3, 3).
2 = 3. In this case, BA1;γ;A2
3,3
is a well-defined
(iii) rank A2 = 1 if and only if rank A′
2 = 2.
Moreover, Q3,3(B) is an indecomposable quadratic Lie algebra except for the objects
in Sym0(3, 3) of the form BA1;γ;A2
with rank A2 = 1.
3,3
b
0
Proof. The first assertion is clear. To prove (i) and (ii), let A2 and A′
2 be as in
(33). Denoting as ci the columns of A′
2 and taking the matrix D with columns ci(D)
given by c1(D) = c2, c2(D) = c5, c3(D) = c7, c4(D) = −c1, c5(D) = −c4, c6(D) =
c3 − c5, c7(D) = c6, c8(D) = c8 and D′ obtained form D by reordering rows and
columns:
D =
a
c
0
0
−c −e −f
2 = rank D = rank D′. If either rank A2 = 3, or rank A2 = 2
We observe that rank A′
with (a, b, c) 6= (0, 0, 0), it is clear that rank D′ = 3. If rank A2 = 2 and (a, b, c) =
(0, 0, 0), then df 6= e2 and therefore rank D′ = 3. Let us assume that rank A2 = 1.
If (a, b, c) = (0, 0, 0), using df = e2 we easily get rank A′
2 = rank D′ = 2. Otherwise
(a, b, c) 6= (0, 0, 0) and λ(a, b, c) = (b, d, e), µ(a, b, c) = (c, e, f ). Then, the matrix D′′
obtained from D by changing the third row r3 by r′
1 = r1
and r′
a b
e f
b d e −c −e −f 0 0
d e
0 0 0
, D′ =
0 0 0 d e
a b
c e f
b d e 0 0
3 = r3 + µr1 − λr2 and r′
a
0
0
c
2 = r2 is:
c
0
b
c 0 0 0
c
D′′ =
a b
0 0 0 a b
0 0 0 0 0 0 µd − λe = 0 µe − λf = 0
2 = rank D′ = rank D′′ = 2. This proves (i) and (ii).
e
f
d
e
Hence, rank A′
(34)
Finally, let B ∈ Sym0(3, t) such that Q3,t(B) = (n, B) is a decomposable
quadratic Lie algebra. Then, n = n3,t
Ker(B) is the orthogonal sum of two quadratic
Lie algebras of types 2 and 1.
In this case, there exists a minimal generator set
{x1, x2, x3} of n3,t such that [x3, x1], [x3, x2] ∈ Ker(B), so the rank of the form
B(e2, e2) is 1. Therefore, t = 3 and B = BA1;γ;A2
with rank A2 = 1. On the
other hand, any quadratic Lie algebra Q3,3(B) with B = BA1;γ;A2
∈ Sym0(3, 3)
and rank A2 = 1 is of type 3 and dimension 5. Thus the derived algebra n2 has
dimension 2. If n is indecomposable, from [14, section 6.2], Z(n) ⊆ n2, but using
(7), Z(n) must be 3-dimensional, which implies dim n ≥ 6, a contradiction.
3,3
3,3
Theorem 5.5. Over any field K of characteristic zero, the categories Sym0(3, 2)
and Sym0(3, 3) satisfy the following features:
22
(i) The objects in Sym0(3, 2) are the invariant forms BA;γ
3,2 described in Example
2.6 and given by any 3× 3 symmetric matrix A and any 0 6= γ ∈ K. Moreover,
any BA;γ
3,2 is isomorphic to B 03×3;1
in Sym0(3, 2).
3,2
3,3
3,3
3,3
and BB1;γ;B2
(ii) The objects B ∈ Sym0(3, 3) such that Q3,3(B) is an indecomposable quadratic
Lie algebra are the invariant forms BA1;γ;A2
described in Example 2.6, and
given by any scalar γ ∈ K and any pair of 3 × 3 symmetric matrices (A1,A2)
with rank A2 ≥ 2. Moreover, for symmetric matrices A2, B2 of rank ≥ 2,
BA1;δ;A2
are isomorphic in Sym0(3, 3) if and only if CB2C =
(Adj P )tCA2C(Adj P ) for some 3×3 regular matrix P and C the matrix given
in (32).
Proof. The first assertion in (i) follows from Lemma 5.4. Now, let A1 =
t u
and γ 6= 0 be and consider θP ∈ Aut n2,3 the automorphism given by the matrix (in
the Hall basis H3,2):
p q
q
s
r
r
t
We have that P tB 03×3;1
3,2
3,2 , which proves the second assertion in (ii).
Let us assume now that BA1;γ;A2
∈ Obj(Sym0(3, 3)). So the matrix of this
form is as in (31) and (33). The elements in the automorphism group Aut n3,3 with
respect to the Hall basis H3,3 have the matrix form (the 3 × 3 matrix C is defined
in (32)):
3,3
X(P ; U, V ) =
0
0
P
U C(Adj P )C 0
V
U ′
P ′′
where P ∈ GL(3), U ∈ M3×3(K), V ∈ M8×3(K), P ′′ ∈ M8×8 is completely deter-
mined by P and U ′ ∈ M8×3(K) is related to U as follows
U =
x5 x8 − x4
−x9 − x1 x6
0
, U ′t =
−x7
0
−x7
0
x9 −x8
x1 x2 x3
x4 x5 x6
x7 x8 x9
x2 −x1
0
x3
0
x3
0
−x4
−x5
−x8
−x2
The automorphisms θX(P ;0,0) ∈ Aut (n3,3) given by the matrix X(P ; 0, 0) are those
corresponding to the subgroup H(3, 3). For any arbitrary θX(P ;0,0), the matrix of
.
x9
x6
23
P =
p
2γ
P = BA1;γ
0
0
1
u
2
γ
0
0 0 0
0
1
0 0 0
0
0
0 0 0
0
0
γ 0 0
0 − s
2 −t 0 γ 0
0 0 1
q
γ
r
γ
.
(BA1;γ;A2
3,3
)θX(P ;0,0) is
0
P t
0
0 C(Adj P t)C 0
0
0
P ′′t
A1 Wγ A′
2
Wγ A2
0
A′t
0
0
2
0
P
0
0 C(Adj P )C 0
0
0
P ′′
=
P tA1P
Wγdet P C(Adj P t)CA2C(Adj P )C
P ′′tA′t
Wγdet P
0
2 P
P tA′
2P ′′
0
0
.
The automorphisms θX(I3;U,V ) belong to the subgroup N(3, 3). For any arbitrary
θX(I3;U,V ), the matrix of the bilinear form (BA1;γ;A2
3,3
)θX(I3;U,V ) is
I3 U t V t
I3 U ′t
0
I8
0
0
A1 Wγ A′
2
Wγ A2
0
A′t
0
0
2
0
I3
U I3
V U ′
0
0
I8
=
A1 + U tWγ + V tA′t
2 + WγU + U tA2U + A′
2V Wγ + U tA2 + A′
Wγ + A2U + U ′tA′t
2
A′t
2
The matrix equation in U,
2U ′ A′
2
0
0
A2
0
.
(35)
Wγ + U tA2 + A′
2U ′ = 03×3
can be viewed as a homogeneous system of 9 equations in the variables x1, . . . , x9,
that that in fact are always repeated the following equation:
γ + cx1 − bx2 + ax3 + ex4 − dx5 + bx6 + f x7 − ex8 + cx9 = 0.
(36)
3,3
Since A2 6= 0, it is immediate that, for any γ ∈ K, we can find adequate entries
for U so that the identity (36) follows. Hence, any form BA1;γ;A2
is isomorphic to a
form BY ;0;A2
for some symmetric 3 × 3 matrix Y .
We also claim that, if rank A2 ≥ 2, for any 3× 3 symmetric matrix Y there exits
an automorphism ϕ ∈ N(3, 3) such that (B 03×3;0;A2
. This assertion is
equivalent to find U ∈ M3×3(K), V ∈ M8×3(K) that are solutions to the system of
matrix equations:
)ϕ = BY ;0;A2
3,3
3,3
3,3
03,3 = U tA2 + A′
(37)
From the above reasoning on the equation (35), for γ = 0 there exist a 3 × 3
0A2U0 is symmetric, the
0A2 + A′
2 + U tA2U + A′
and Y = V tA′t
2U ′
2V
2(U0)′. Since Y − U t
matrix U0 such that 03,3 = U t
equation
Y − U t
0A2U0 = V tA′t
2 + A′
2V
24
(38)
2 + A′
is reduced to find a matrix V ∈ M8×3(K) such that V tA′t
2V is any symmetric
3 × 3 matrix. This is equivalent to find a 8 × 3 matrix V such that A′
2V is any
3 × 3 matrix. For a 3 × 1 vector v, the first column X of the matrix V must satisfy
A′
2X = v. But this equation has solution because the rank of A′
2 is 3 according to
Lemma restricciones. In the same vein, we can obtain the second and third columns
of V and so A′
2V is any 3 × 3 matrix. This proves our claim.
Now we will prove the equivalence in (ii). Let BA1;δ;A2
be isomorphic
objects in Sym0(3, 3). Then, there exists τ ∈ Aut n2,3 = H(3, 3)N(3, 3) such that
(BA1;δ;A2
. From the previous reasoning, we can assume that δ = γ = 0.
Decomposing, τ = σϕ with σ ∈ H(3, 3) and ϕ ∈ N(3, 3), we have BB1;0;B2
=
((BA1;0;A2
)σ)ϕ and therefore B2 = C(Adj P )tCA2C(Adj P )C.
)τ = BB1;γ;B2
and BB1;γ;B2
3,3
3,3
3,3
3,3
3,3
For the converse, suppose B2 = C(Adj P )tCA2C(Adj P )C and rank A2 ≥ 2.
From the automorphism σ ∈ H(3, 3) given by the matrix X(P ; 0, 0),
3,3
(BA1;0;A2
3,3
)σ = BP tA1P ;0;B2
3,3
.
As rank B2 = rank A2 ≥ 2, there exist ϕ1, ϕ2 ∈ N(3, 3) such that ((BP tA1P ;0;B2
(B0;0;B2
phism.
)σϕ1ϕ2 = BB1;0;B2
. Hence (BA1;0;A2
)ϕ2 = BB1;0;B2
3,3
3,3
3,3
3,3
3,3
, which proves the isomor-
)ϕ1)ϕ2 =
Remark 5.6. In item (ii) of Theorem 5.5, if rank A2 = 1, the equality CB2C =
(Adj P )tCA2C(Adj P ) is a necessary condition of isomorphism, but it is not suffi-
cient. The invariant forms BA1;0;A2
and B 0;0;A2
with
3,3
3,3
A1 =
0 0 0
0 0 0
0 0 1
and
A2 =
1 0 0
0 0 0
0 0 0
+are not isomorphic elements in Sym0(3, 3)
Corollary 5.7. Over any field K of characteristic 0, up to metric isomorphism the
indecomposable quadratic Lie algebras of type 3 and nilpotent indices 2 or 3 are:
(i) The quadratic Lie algebra (n3,2, B 03×3;1
3,2
).
n3,3
, B0;0;A
(ii) The quadratic Lie algebras (
) for any 3 × 3 symmetric matrix
) are isomet-
) and (
A of rank ≥ 2. Moreover, (
rically isomorphic if and only if CBC = (Adj P )tCAC(Adj P ) for some 3 × 3
regular matrix P .
, B0;0;B
, B0;0;A
Ker B0;0;B
Ker B0;0;A
Ker B0;0;A
n3,3
n3,3
3,3
3,3
3,3
3,3
3,3
3,3
In (ii), if K2 = K or K = R, then (
isometrically isomorphic if and only if A and B are congruent.
) and (
Ker B0;0;A
3,3
, B0;0;A
n3,3
3,3
n3,3
Ker B0;0;B
3,3
, B0;0;B
3,3
) are
25
Proof. We just need to apply the previous Therorem 5.5 and Corollary 4.3 to prove
(i) and (ii). For the last statement, since C t = C we only need to show that if A
and B are congruent then (
) are isometrically
isomorphic. First we notice that if D is a regular matrix, Adj(D) = det(D)(D−1)t.
We apply this now. We suppose that B = RtAR, and we want to find a regular
matrix P such that R = CAdj(P )C. Since C = C −1 this implies that Adj(P ) =
CRC. So if we find a matrix P such that Adj(P ) = CRC, then we have finished. In
, B0;0;B
) and (
, B0;0;A
3,3
Ker B0;0;A
3,3
n3,3
n3,3
Ker B0;0;B
3,3
3,3
the case K2 = K, if we take P = pdet(R)C(R−1)tC, we can check that Adj(P ) =
CRC. Indeed, as we have shown Adj(P ) = det(P )(P −1)t and then
Adj(P ) = (pdet(R))3det(R)−1(pdet(R))−1CRC = CRC
In the case K = R, we observe that if R is such that det(R) < 0 we can consider
S = −R and then also StAS = B and det(S) > 0. So we can suppose that det(R) >
0 and then taking P =pdet(R)C(R−1)tC, we have also Adj(P ) = CRC.
6 K algebraically closed and K = R
In this final section, first we restrict ourshelves to algebraically closed fields to reach,
up to isomorphism, the nilpotent quadratic Lie algebras on 2 and 3 generators and
small nilpotent index. In this case, we note that for a given B ∈ GL(3), B = Adj(A)
with A = √det B(B−1)t.
Theorem 6.1. Up to isomorphism, the indecomposable quadratic nilpotent Lie al-
gebras over any algebraically closed filed K of characteristic zero K of type 1 or of
type 2 and nilindex ≤ 5 or of type 3 and nilindex ≤ 3 are isomorphic to one of the
following Lie algebras (where φ, ϕ, ϕi, ψ and ψi are symmetric bilinear forms):
(i) The 1-dimensional abelian Lie algebra (n1,1, φ) with basis {a1} and φ(a1, a1) =
1.
(ii) The 5-dimensional free nilpotent Lie algebra (n2,3, ϕ) with basis {ai}5
i=1 and
nonzero products [a2, a1] = −[a1, a2] = a3, [a3, a1] = −[a1, a3] = a4 and
[a3, a2] = −[a2, a3] = a5 where ϕ(ai, aj) = (−1)i−1 for i ≤ j and i + j = 6 and
ϕ(ai, aj) = 0 otherwise.
(iii) The 7-dimensional Lie algebra (n1
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a3, [a3, a1] = −[a1, a3] = a4, [a4, a1] = −[a1, a4] =
a5, [a5, a1] = −[a1, a5] = a6, [a5, a2] = −[a2, a5] = a7 and [a3, a4] = −[a4, a3] =
a7 where ϕ1(ai, aj) = (−1)i for i ≤ j and i+j = 8 and ϕ1(ai, aj) = 0 otherwise.
2,5, ϕ1) with basis {ai}7
26
(iv) The 8-dimensional Lie algebra (n2
2,5, ϕ2) with basis {ai}8
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a3, [a3, a1] = −[a1, a3] = a4,[a3, a2] = −[a2, a3] = a5
[a4, a1] = −[a1, a4] = a6, [a6, a1] = −[a1, a6] = a7, [a6, a2] = −[a2, a6] = a8,
[a2, a5] = −[a5, a2] = −a6, [a4, a3] = −[a3, a4] = −a8, and [a5, a3] = −[a3, a5] =
a7, where ϕ2(ai, aj) = (−1)i for 0 ≤ i ≤ 3 and i + j = 9, ϕ2(a4, a4) =
ϕ2(a5, a5) = 1 and ϕ2(ai, aj) = 0 otherwise.
(v) The 6-dimensional free nilpotent Lie algebra (n3,2, ψ) with basis {ai}6
i=1 and
nonzero products [a2, a1] = −[a1, a2] = a4, [a3, a1] = −[a1, a3] = a5 and
[a3, a2] = −[a2, a3] = a6 where ψ(ai, aj) = (−1)i−1 for i ≤ j and i + j = 7 and
ψ(ai, aj) = 0 otherwise.
(vi) The 8-dimensional Lie algebra (n1
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a4, [a3, a1] = −[a1, a3] = a5, [a4, a1] = −[a1, a4] =
a6, [a4, a2] = −[a2, a4] = a7, [a5, a1] = −[a1, a5] = a8 and [a5, a3] = −[a3, a5] =
a7 where ψ1(a4, a4) = ψ1(a5, a5) = 1, ψ1(a1, a7) = 1 = −ψ1(a2, a6) = −ψ1(a3, a8)
and ψ1(ai, aj) = 0 otherwise.
3,3, ψ1) with basis {ai}8
(vii) The 9-dimensional Lie algebra (n2
3,3, ψ2) with basis {ai}9
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a4, [a3, a1] = −[a1, a3] = a5 [a3, a2] = −[a2, a3] = a6,
[a4, a1] = −[a1, a4] = a7, [a4, a2] = −[a2, a4] = a8, [a5, a1] = −[a1, a5] = a9,
[a5, a3] = −[a3, a5] = a8, [a6, a3] = −[a3, a6] = −a7 and [a6, a2] = −[a2, a6] = a9
where ψ2(a4, a4) = ψ2(a5, a5) = ψ2(a6, a6) = 1, ψ2(a1, a8) = 1 = −ψ2(a2, a7) =
−ψ2(a3, a9) and ψ2(ai, aj) = 0 otherwise.
Any nonabelian quadratic Lie algebra of type ≤ 2 is indecomposable. The nonabelian
decomposable Lie algebras of type 3 and nilpotent index ≤ 3 are the orthogonal sum
(n1,1, φ) ⊕ L where L is a quadratic Lie algebra as in item (ii), (iii) or (iv).
Proof. Let us assume now that (n, ϕ) is an indecomposable nilpotent quadratic
nonabelian Lie algebra. If n has type 2, according to Corollary 5.3, either (n, ϕ) is
) where, up to isometric isomorphism, A
as in item (ii) or (n, ϕ) = (
, B0;0;A
n2,5
Ker B0;0;A
2,5
2,5
is one of the following matrices:
A1 =(cid:18)1 0
0 0(cid:19) or A2 =(cid:18)1 0
0 1(cid:19)
and B0;0;A
of the form B0;0;A1
2,5
2,5
are both 7, and moreover
is as described in Example 2.5. The rank and the dimension of the kernel
Ker B0;0;A1
2,5 = span < [[x2, x1], x2], [[[x2, x1], x1], x2], [[[x2, x1], x2], x2],
[[[x2, x1], x1], x1], x2]+[[[x2, x1], x1], [x2, x1]], [[[[x2, x1], x1], x2], x2], [[[x2, x1], x2], [x2, x1]],
27
[[[[x2, x1], x2], x2], x2] > .
The rank and the dimension of the kernel of the form B0;0;A2
and
2,5
are 8 and 6, respectively,
Ker B0;0;A2
2,5 = span < [[[x2, x1], x1], x2], [[[x2, x1], x2], x2] − [[[x2, x1], x1], x1],
[[[x2, x1], x1], [x2, x1]] + [[[x2, x1], x1], x1], x2], [[[[x2, x1], x1], x2], x2],
[[[x2, x1], x2], [x2, x1]]−[[[[x2, x1], x1], x1], x1], [[[[x2, x1], x2], x2], x2]−[[[[x2, x1], x1], x1], x2] > .
In this way we obtain the quadratic algebras in items (iii) and (iv).
If n has type 3, from Corollary 5.7 either (n, ϕ) is as in item (v) or (n, ϕ) =
n3,3
) where, up to isometric isomorphism, A is one of the following
, B0;0;A
(
Ker B0;0;A
matrices:
3,3
3,3
1 0 0
0 1 0
0 0 0
B1 =
or B2 =
1 0 0
0 1 0
0 0 1
,
and B0;0;A
of the form B0;0;B1
3,3
3,3
are 8 and 6 respectively, and
is described in Example 2.6. The rank and the dimension of the kernel
Ker B0;0;B1
3,3 = span < [x3, x2], [[x2, x1], x3], [[x3, x1], x2],
[[x3, x1], x3] − [[x2, x1], x2], [[x3, x2], x2], [[x3, x2], x3] > .
The rank and the dimension of the form B0;0;B2
3,3
are 9 and 5 respectively, and
Ker B0;0;B2
3,3 = span < [[x2, x1], x3], [[x3, x1], x2], [[x3, x1], x3] − [[x2, x1], x2],
[[x3, x2], x2] − [[x3, x1], x1], [[x3, x2], x3] − +x2, x1], x1] >
So we get the quadratic algebras in items (vi) and (vii). The final assertion is
straightforeword.
Now, we consider the case K = R.
Theorem 6.2. Up to isomorphism, the indecomposable quadratic nilpotent real Lie
algebras of type 1 or of type 2 and nilindex ≤ 5 or of type 3 and nilindex ≤ 3 are
the following (where ϕ3, ψ3 and ψ4 are symmetric bilinear forms):
(i) The quadratic Lie algebras (n,±ϕ) where n and ϕ is as described in items (i),
(ii), (iii), (iv), (vi), (vii) of Theorem 6.1 and (n3,2, ψ) as in item (v).
28
(ii) The 8-dimensional Lie algebra (n3
2,5, ϕ3) with basis {ai}8
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a3, [a3, a1] = −[a1, a3] = a4, [a4, a1] = −[a1, a4] =
a6, [a6, a1] = −[a1, a6] = a7, ,[a3, a2] = −[a2, a3] = a5, ,[a5, a2] = −[a2, a5] =
−a6, [a6, a2] = −[a2, a6] = a8, ,[a4, a3] = −[a3, a4] = −a8 and ,[a5, a3] =
−[a3, a5] = −a7 where ϕ3(ai, aj) = (−1)i for i ≤ j, i = 1, 2, 3 and i + j = 9,
ϕ3(a4, a4) = 1 = −ϕ3(a5, a5) and ϕ3(ai, aj) = 0 otherwise.
3,3, ψ3) with basis {ai}8
i=1 and nonzero prod-
ucts [a2, a1] = −[a1, a2] = a4, [a3, a1] = −[a1, a3] = a5, [a4, a1] = −[a1, a4] =
a6, [a5, a1] = −[a1, a5] = a8, [a4, a2] = −[a2, a4] = a7 and [a5, a3] = −[a3, a5] =
−a7 where ψ3(a4, a4) = −ψ3(a5, a5) = 1, ψ3(a1, a7) = 1 = −ψ3(a2, a6) =
ψ3(a3, a8) and ψ3(ai, aj) = 0 otherwise.
(iii) The 8-dimensional Lie algebra (n3
(iv) The 9-dimensional quadratic Lie algebras (n4
i=1 and
nonzero products [a2, a1] = −[a1, a2] = a4, [a3, a1] = −[a1, a3] = a5, [a4, a1] =
−[a1, a4] = a7, [a5, a1] = −[a1, a5] = a9 [a3, a2] = −[a2, a3] = a6, [a4, a2] =
−[a2, a4] = a8, [a6, a2] = −[a2, a6] = −a9, [a5, a3] = −[a3, a5] = a8 and
[a6, a3] = −[a3, a6] = a7, where ψ4(a4, a4) = ψ4(a5, a5) = 1 = −ψ4(a6, a6),
ψ4(a1, a8)1 = −ψ4(a2, a7) = −ψ4(a3, a9) and ψ2(ai, aj) = 0 otherwise.
3,3,±ψ4) with basis {ai}9
Any nonabelian quadratic Lie algebra of type ≤ 2 is indecomposable. The nonabelian
decomposable Lie algebras of type 3 and nilpotent index ≤ 3 are the orthogonal sum
(n1,1, φ) ⊕ L where L is a quadratic Lie algebra as in items (ii), (iii) or (iv).
Proof. According to Corollaries 5.3 and 5.7, over the real field the classification is
given by the rank and the signature of n× n symmetric matrices for n = 1, 2, 3. So,
the cases to be considered are:
• (n1,1,±φ) with basis {a1} and φ(a1, a1) = 1 and (n3,2, ψ) as described in item
(v) of Therorem 6.1.
• (n,±ϕ) where (n,±ϕ) is as in item (ii) or Theorem 6.1, (n, ϕ) = (
Ker B0;0;A
where, up to isometric isomorphism, Ai is one of the following matrices:
2,5
n2,5
, B0;0;Ai
2,5
)
±A1 =(cid:18)1 0
0 0(cid:19) ,±A2 =(cid:18)1 0
0 1(cid:19) or A3 =(cid:18)1
0 −1(cid:19)
0
and B0;0;Ai
2,5
is as described in Example 2.5.
• (n, ϕ) = (
3,3
the following matrices:
Ker B0;0;A
3,3
, B0;0;Ci
n3,3
) where, up to isometric isomorphism, Ci is one of
±C1 =
1 0 0
0 1 0
0 0 0
0
, C2 =
1
0
0 −1 0
0
0
0
,±C3 =
1 0 0
0 1 0
0 0 1
,±C4 =
1 0
0
0 1
0
0 0 −1
29
and B0;0;Ci
3,3
is as described in Example 2.6.
The matrices ±A1,±A2,±C1,±C3, and (n1,1,±φ), (n2,3,±φ) and (n3,2, ψ) yield
to the quadratic algebras in item (i) of this theorem. A straightforward calculation
gives us the rank and the dimension of the kernel of the form B0;0;A3
, which are 8
and 6 respectively, and
2,5
Ker B0;0;A3
2,5 = span < [[[x2, x1], x1], x2], [[[x2, x1], x1], x1] + [[[x2, x1], x2], x2],
[[[[x2, x1], x1], x1], x2]+[[[x2, x1], x1], [x2, x1]], [[[[x2, x1], x1], x2], x2], [[[[x2, x1], x1], x1], x1]+
[[[x2, x1], x2], [x2, x1]], [[[[x2, x1], x1], x1], x2] + [[[[x2, x1], x2], x2], x2] > .
The corresponding quotient provides the quadratic algebra described in (ii). The
rank and the dimension of the kernel of the forms B0;0;C2
are 8 and 6 and
9 and 5 respectively. Moreover,
and B0;0;C4
3,3
3,3
Ker B0;0;C2
3,3 = span < [x3, x2], [[x2, x1], x3], [[x3, x1], x2], [[x2, x1], x2] + [[x3, x1], x3],
and
[[x3, x2], x2], [[x3, x2], x3] > .
Ker B0;0;±C4
3,3
= span < [[x2, x1], x3], [[x3, x1], x2],−[[x2, x1], x2] + [[x3, x1], x3],
[[x3, x1], x1] + [[x3, x2], x2],−[[x2, x1], x1] + [[x3, x2], x3] > .
The corresponding two quotients give the quadratic algebras in items (iii) and
(iv).
Remark 6.3. If we consider the general problem of building a t-nilpotent quadratic
Lie algebra with d generators beginning with a B(ei, ej) with i + j = t + 1 in nd,t,
one can think that this seems easy, but one has really to check some facts about the
final B. It takes some work to prove what is needed in a particular case. In fact, we
do not know what is necessary in general. This can be a line of research to address
in the future.
ACKNOWLEDGEMENTS
The authors thank the referee for very valuable suggestions that have improved
the paper.
30
References
[1] V. J. del Barco, G. P. Ovando, Free nilpotent Lie algebras admitting ad-
invariant metrics, J. Algebra 366 (2012), 205-216.
[2] M. Bordemann, Nondegenerate invariant bilinear forms on Nonassociative al-
gebras, Acta Math. Univ. Comenianae, Vol. LXVI, 2 (1997), 151-201.
[3] M. T. Duong, R. Ushirobira, Solvable quadratic Lie algebras of dimension ≤ 8,
arXiv: 1407.6775v1.
[4] G. Favre, L. J. Santharoubane, Symmetric, Invariant, Non-degenerate Bilinear
Form on a Lie Algebra, J. Algebra 105 (1987), 451-464.
[5] M. A. Gauger, On the classification of metabelian Lie algebras, Trans. Amer.
Math. Soc., 179 (1973), 293-329.
[6] M. Grayson, R. Grossman: Models for free nilpotent Lie algebras, J. of Algebra,
135 (1990), 177-191.
[7] M. Hall, A basis for free Lie rings and higher commutators in free groups, Proc.
Amer. Math. Soc., 1 (1950), 575-581.
[8] N. Jacobson, "Basic Algebra II", W.H. Freeman and Company, New York,
1989.
[9] I. Kath, Nilpotent metric Lie algebras of small dimension, J. Lie Theory 17
(2007), no. 1, 41-61. (see also, arXiv:math/0505688)
[10] Medina, Ph. Revoy, Alg`ebres de Lie et produit escalaire invariant, Ann. Sci.
`Ec. Norm. super. (4) 18 (3) (1985) 553-561.
[11] L. Noui, Ph. Revoy, Alg`ebres de Lie orthogonales et formes trilineaires altern´ees,
Comm. Algebra 25 (2) (1997), 617-622.
[12] G. Ovando, Two-step nilpotent Lie algebras with ad-invariant metrics and a
special kind of skew-symmetric maps, J. Algebra Appl. 6 (2007), no. 6, 897-
917.
[13] T. Sato, The derivations of the Lie algebras, Tohoku Math. Journ. 23 (1971),
21-36.
[14] T-S Tsou, A.G. Walker, Metrisable Lie groups and algebras, Proc. Roy. Soc.
Edinburgh. Sect. A 64, 1955/1957, 290-304.
31
Pilar Benito, Daniel de-la-Concepci´on, Jes´us Laliena 1
Departamento de Matem´aticas y Computaci´on
Universidad de La Rioja
26004, Logrono. Spain
[email protected] , [email protected], [email protected]
1The authors have been supported by the Spanish Ministerio de Econom´ıa y Competitividad
(MTM 2013-45588-CO3-3), and the second author also by a FPU grant (12/03224) of the Spanish
Ministerio de Educaci´on y Ciencia
32
|
1808.06242 | 1 | 1808 | 2018-08-19T18:51:36 | Non-isomorphism of categories of algebras | [
"math.RA",
"math.CT",
"math.LO"
] | This paper was accepted for Comment. Math. Univ. Carolinae in 1968 but then got lost during the military occupation of Prague and surrounding events. My own long-lost carbon copy of it turned up in my Columbia office. The only changes to the text are addition of a footnote and some commas. | math.RA | math |
NON-ISOMORPHISM OF CATEGORIES OF ALGEBRAS
WALTER D. NEUMANN
Abstract. This paper was accepted for Comment. Math. Univ. Caroli-
nae in 1968 but then got lost during the military occupation of Prague
and surrounding events. A carbon copy of it turned up in my Columbia
office. The only changes to the text are addition of a footnote and some
commas.
We recall that if ∆ = (Ki)i∈I is a collection of sets, then an algebra
(A, (fi){i∈I}) of type ∆ consists of a set A and a collection (fi)i∈I of op-
erations on A, such that for each i ∈ I fi is a Ki -- ary operation; that is,
fi : AKi → A. We shall use the same symbol for an algebra A and its
underlying set.
We say that the types ∆ = (Ki)i∈I and ∆′ = (Lj)j∈J are equivalent if
there is a bijection φ : I → J such that for each i the cardinalities Ki and
Lφ(i) are equal.
We denote by A(∆) the category of all algebras of type ∆ and all homo-
morphisms between them. Clearly, if ∆ and ∆′ are equivalent types, then
A(∆) and A(∆′) are isomorphic categories. The purpose of this note is to
prove the converse, thereby answering a question posed by A. Pultr.
This question was motivated by the result of Z. Hedrl´ın and A. Pultr [2]
which states that even if ∆ and ∆′ are not equivalent, A(∆) and A(∆′) are
embeddable as full subcategories in each other, so long as ∆ and ∆′ are not
too small (the sums of the cardinalities of the sets in ∆ and ∆′ should each
exceed 1).
In §1 we recall the necessary properties of algebraic operations, and in §2
we prove the theorem.
1. Operations
Let f : F K → K be a K -- ary operation on F (F, K any sets). The support
of f (Felscher [1]) is defined by
supp(f ) = {A ⊆ K for all α, β ∈ F K , αA = βA ⇒ f (α) = f (β)}
That is, A ∈ supp(f ) means that the value of f on any α ∈ F K is already
determined by the restriction of α to A.
The essential rank of f is defined as min{A A ∈ supp(f )}. If A is a
primitive class (variety) of algebras, define its rank to be the supremum of
the essential ranks of all A -- algebraic operations. Since the essential rank of
an algebraic operation is always less than the dimension (S lomi´nski [3]) of
A, this supremum exists.
1
2
WALTER D. NEUMANN
Now let f be any element of the free algebra F (X, A) with basis X of
the primitive class A, and let A be any algebra in A. One can define an
X -- ary operation f A on A by f A(α) = ¯α(f ) for any α ∈ AX . Here ¯α is the
homomorphic extension of α to a homomorphism of F (X, A) to A. f A is
an algebraic operation on A; in fact f 7→ f A is a surjective homomorphism
of F (X, A) onto the algebra H X(A) of all X -- ary algebraic operations on A
(see for instance [1]).
2. Non-isomorphism of categories
Theorem. If the categories A(∆) and A(∆′) are isomorphic, then ∆ and
∆′ are equivalent types.
We shall prove this theorem in two steps. Let s denote the canonical
underlying set functors on A(∆) and A(∆′).
Lemma 1. If the concrete categories (A(∆), s) and (A(∆′), s) are concretely
isomorphic (that is, isomorphic by a functor which preserves underlying
sets), then ∆ is equivalent to ∆′.
Proof. Let ∆ = (Ki)i∈I .
It suffices to show that the knowledge of the
category A(∆) and its underlying set functor is sufficient to recover ∆ up to
equivalence. Now the definition of free algebra of A(∆) over a basis involves
only the underlying sets and homomorphisms, so we know the free algebras
of A(∆). Hence by the last paragraph of §1 we know the algebraic operations
for A(∆) so we can calculate the rank δ of A(∆).
Let F be a free algebra of A(∆) with basis X such that X ≥ δ. Let
Y = F − X. Define a relation R on Y by y1Ry2 if and only if there is
an endomorphism of F which maps y1 onto y2. Let S be the smallest
equivalence relation on Y containing R, and let J = Y /S be the set of
equivalence classes of Y under S.
We now consider the meaning of this construction in terms of the actual
algebraic structure of F . The set Y is the set of all fi(α), i ∈ I, α ∈ F Ki,
where the fi are the defining operations of the class. Under an endomor-
phism of F an element fi(α) cannot be mapped onto an element fj(β) with
i 6= j. Further, if α0 ∈ F Ki is injective with α0(Ki) ⊆ X, then every element
of the form fi(α), α ∈ F Ki, is the image of fi(α0) under suitable endomor-
phisms of F . Hence the equivalence classes under S are just the subsets of
Y of the form φ(i) = {fi(α) α ∈ F Ki}, and φ : i 7→ φ(i) is a bijective map
from I to J.
It remains only to show that to each φ(i) ∈ J we can recover Ki. We
can choose a y ∈ φ(i) with the property that every element of φ(i) is the
image of y under some endomorphism of F . We then have the corresponding
X -- ary algebraic operation y = yF in H X(F ). We claim that y has essential
rank Ki, completing the proof. Indeed, y is of the form fi(α0) for some
injective α0 ∈ F Ki with α0(Ki) ⊆ X. supp(y) is just the set of all subsets
NON-ISOMORPHISM OF CATEGORIES OF ALGEBRAS
3
of X which contain α0(Ki), so the essential rank of y is α0(Ki). Since α0
is injective, α0(Ki) = Ki.
(cid:3)
The second part of the proof of the theorem is given by the following
lemma:
Lemma 2. If the categories A(∆) and A(∆′) are isomorphic, then the con-
crete categories (A(∆), s) and (A(∆′), s) are concretely isomorphic.
Proof. We first determine a free algebra P of rank 1 (that is, basis cardinality
1) in A(∆). This can be done in many ways, For instance, P ∈ A(∆) is a
free algebra if and only if every epimorphism to P has a section, and it
furthermore has rank 1 if and only if every morphism P → P is mono.
The functor mor(P, −) : A(∆) → Set is a "new underlying set functor"
which is naturally equivalent to the standard underlying set functor s on
A(∆).
Let T : A(∆) → A(∆′) be an isomorphism. Then T (P ) is a free alge-
bra of rank 1 in A(∆′), so we also have a "new underlying set functor"
mor(T (P ), −) on A(∆′) which is naturally equivalent to the standard one.
If one identifies the sets mor(P, A) and mor(T (P ), T (A)) by means of T for
each A ∈ A(∆), then T commutes with these "new underlying set functors",
and it is not difficult, using the properties of the standard underlying set
functors, to deduce that A(∆) and A(∆′) are also concretely isomorphic
with respect to the standard underlying set functors.
(cid:3)
Lemma 2 states that two categories of the form A(∆) are abstractly iso-
morphic if and only if they are concretely isomorphic, or in more algebraic
terminology: rationally equivalent. This in fact holds for more general prim-
itive classes of algebras; for instance a slight modification of the above proof
shows that two Schreier primitive classes (subalgebras of free algebras are
free) whose free algebras of rank 1 are not isomorphic to any of higher rank1
are abstractly isomorphic as categories if and only if they are rationally
equivalent. Some restriction on the classes considered is however necessary,
for it is known that to any primitive class one can find primitive classes not
rationally equivalent to the given one, such that the categories are isomor-
phic.
References
[1] W. Felscher: Equational maps, Contributions to Mathematical Logic, Amsterdam
1967.
[2] Z. Hedrl´ın and A. Pultr: On the full embeddings of categories of algebras, Illinois J.
Math. 10,3 (1966), 392 -- 406.
[3] J. S lomi´nski: The theory of abstract algebras with infinitary operations, Rozprawy
Mat. 18 (1959).
1The phrase "whose free algebras . . . higher rank" was a handwritten addition in the
1968 typescript -- apparently an afterthought.
|
1402.2212 | 2 | 1402 | 2017-12-17T22:46:15 | Inner derivations of exceptional Lie algebras in prime characteristic | [
"math.RA"
] | It is well-known that every derivation of a semisimple Lie algebra $L$ over an algebraically closed field $F$ with characteristic zero is inner. The aim of this paper is to show what happens if the characteristic of $F$ is prime with $L$ an exceptional Lie algebra. We prove that if $L$ is a Chevalley Lie algebra of type $\{G_2,F_4,E_6,E_7,E_8\}$ over a field of characteristic $p$ then the derivations of $L$ are inner except in the cases $G_2$ with $p=2$, $E_6$ with $p=3$ and $E_7$ with $p=2$. | math.RA | math |
INNER DERIVATIONS OF EXCEPTIONAL LIE ALGEBRAS IN
PRIME CHARACTERISTIC
PABLO ALBERCA BJERREGAARD, DOLORES MART´IN BARQUERO,
AND C ´ANDIDO MART´IN GONZ ´ALEZ
Abstract. It is well-known that every derivation of a semisimple Lie algebra
L over an algebraically closed field F with characteristic zero is inner. The aim
of this paper is to show what happens if the characteristic of F is prime with
L an exceptional Lie algebra. We prove that if L is a Chevalley Lie algebra
of type {g2, f4, e6, e7, e8} over a field of characteristic p then the derivations
of L are inner except in the cases g2 with p = 2, e6 with p = 3 and e7 with
p = 2.
1. Introduction
This paper deals with the derivation algebra of the Chevalley Lie algebras of
exceptional type {g2, f4, e6, e7, e8}. As it is well-known Chevalley constructed Lie
algebras over arbitrary fields F starting from any semisimple (finite-dimensional)
Lie algebra over an algebraically closed field of characteristic zero. The key point
was to realize that on such algebras one can find a suitable basis whose structure
constants are integers. Then, by an scalar extension process he constructed those
Lie algebras which nowadays are called Chevalley algebras.
By gathering results of Zassenhaus (1939), Seligman (1979), Springer and Stein-
berg (we will give full details in the next section) one can get convinced that
any derivation of a Chevalley F -algebra of any of the types {g2, f4, e6, e7} is in-
ner for char(F ) ≥ 5. Also the derivations of a Chevalley algebra of type e8 with
char(F ) ≥ 7, are inner.
In [1], the case f4 over fields of characteristic 6= 2 is considered. There, it is
proved that all derivations are inner. In this paper we study the cases not covered
previously: the derivations of the algebras of type {g2, f4, e6, e7} over fields of char-
acteristic 2 or 3, and algebras of type e8 over fields of characteristic 2, 3 or 5 (we
overlap with [1] only in the case f4 for characteristic 3 though our methodology in
this case is different).
Returning to the reference [1], the key idea in this paper is the formula stating
that any derivation is the sum of an inner derivation plus a derivation annihilating
a fixed Cartan subgalgebra ([1, Proof of Proposition 8.1]). We have proved that
a certain extension of this idea is also true for Chevalley algebras of exceptional
type: any derivation is the sum of an inner derivation plus a derivation mapping a
fixed Cartan subalgebra into the center. Thus, when the center is null, we recover
the formula in [1]. This formula reduces the computation of the dimension of the
derivation algebra to the computation of the dimension of the vector space V of
derivations mapping a fixed Cartan subalgebra into the center.
For the exceptional algebras of low rank g2 and f4, most of the computational
algebra packages with support for Lie algebras gives us directly the dimension of
Date: 1 abril 2013.
2000 Mathematics Subject Classification. Primary 16D70.
Key words and phrases. Derivations, Lie algebras, Exceptional Lie algebras.
1
2
P. ALBERCA, D. MART´IN, AND C. MART´IN
the derivation algebra. However this computational approach does not work for the
algebras ei, i = 6, 7, 8 which have a higher rank.
2. Previous definitions and results
Let k be an algebraically closed field k of characteristic 0. Recall that the Cartan
matrix of a finite-dimensional semi-simple Lie algebra L over k is a matrix
A = 2
(αi, αj)
(αj, αj)!i,j=1,...,r
where α1, . . . , αr is some system of simple roots of L with respect to a fixed Cartan
subalgebra t and ( , ) is the scalar product on the dual space of t defined by the
Killing form on L. The entries aij = 2 (αi,αj )
(αj ,αj ) of a Cartan matrix have the following
properties:
(1)
aii = 2;
aij ≤ 0 and aij ∈ Z for i 6= j,
aij = 0 ⇒ aji = 0.
(cid:27)
The Cartan matrix is a key tool in the description of L by generators and rela-
tions. Denote by Φ the root system of L relative to K. As it is known, there exist
in L a basis {vα}α∈Φ ∪ {hi}r
i=1 whose structure constants are in Z:
[hi, vα] = hα, αiivα ∈ Zvα for α ∈ Φ, 1 ≤ i ≤ r and hα, βi := 2 (α,β)
(β,β) .
(1) [hi, hj] = 0, for any i, j ∈ {1, . . . , r}.
(2)
(3) [vα, v−a] = hα a Z-linear combination of h1, . . . , hr.
(4) If α and β are independent roots, β − rα, . . . , β + qα the α-string through
β then [vα, vβ] = 0 if q = 0 while [vα, vβ] = ±(r + 1)vα+β if α + β ∈ Φ.
Such a basis {vα}α∈Φ ∪ {hi}r
proved for instance in [6, Proposition, p. 146 and Theorem, p. 147].
i=1 is called a Chevalley basis of L and its existence is
Following [6, p. 149], given a Chevalley basis {vα}α∈Φ∪{hi}r
i=1 of the semisimple
Lie algebra L over k we can consider its linear envelope LZ which is a Lie algebra
over Z in the obvious sense. Now if Fp is the field of integers module the prime
p we can construct the Lie Fp-algebra LFp := LZ ⊗Z Fp. The multiplication table
of this algebra in a suitable basis is the one for L reduced module p. Now, if F
is any extension field of Fp we can construct the Lie algebra (over F ) given by
LF := LFp ⊗Fp F . This algebra inherits both basis and Lie algebra structure from
LFp . In this way we obtain from each couple (L, F ) a Lie algebra LF over F , called
a Chevalley algebra, which depends up to isomorphism only on L and the field F .
If L is simple of Dynkin diagram D, the we will say that LF is a Lie algebra of
type D. A common notation of this algebra will be D(F ) (for instance f4(F3) or
e8(F2)).
Let L be a Lie algebra over a field F .
If x ∈ L, we have ad(x) : L → L
with ad(x)(y) := [x, y]. It is easy to check that ad(x) ∈ Der(L). We will denote
ad(L) := {ad(x) : x ∈ L} ⊂ Der(L). In fact,
(1)
ad(L) ⊳ Der(L).
Theorem 1. Let F be an algebraically closed field of characteristic zero. If L is a
finite-dimensional semisimple Lie algebra over F , then
(2)
ad(L) = Der(L)
If L is finite-dimensional, we have the Killing form of L:
(3)
k : L × L → F, k(x, y) := tr(ad(x), ad(y)),
and we have the following results
INNER DERIVATIONS OF EXCEPTIONAL LIE ALGEBRAS IN PRIME CHARACTERISTIC 3
Theorem 2 (Zassenhaus [8]).
degenerate Killing form, then every derivation is inner. That is to say
If L is a finite-dimensional Lie algebra with non-
(4)
ad(L) = Der(L).
Let L be a Chevalley Lie algebra of exceptional type {g2, f4, e6, e7, e8}. A very
fundamental result ([5, p.49]) in what follows, due to Seligman and independently
to Springer and Steinberg, says that (relative to a Chevalley basis)
(5)
(6)
det k = 2α3β,
if L is of type g2, f4, e6 or e7,
if L is of type e8 ,
det k = 2α3β5γ,
for some nonzero scalars α, β and γ. It is easy now to obtain
Corollary 1. Let L be a Chevalley F -algebra of any of the types {g2, f4, e6, e7},
then if char(F ) ≥ 5 every derivation is inner. If L is of type e8 and char(F ) ≥ 7,
every derivation in L is inner.
We have defined ad : L → Der(L) as x 7→ ad(x). An important fact is that
(7)
dim ad(L) = dim L − dim ker ad = dim L − dim Z(L),
where Z(L) is the center of the Lie algebra. If, for example, Z(L) = 0 then ad is a
monomorphism and dim L = dim ad(L) ≤ dim Der(L). Now if dim L = dim Der(L)
then ad(L) = Der(L), and every derivation will be inner.
Take now a simple Lie algebra L over an algebraically closed field k of charac-
teristic zero. Fix a Chevalley basis {vα} ∪ {hi} of L. The different hi generate a
Cartan subalgebra H and each vα the root space Lα. Consider now an arbitrary
field F and consider the Chevalley algebra LF introduced above. The properties of
the new algebra LF may have changed dramatically, for instance LF may happen to
be non-simple. Also while L is simple, the algebra LF may have a nonzero center.
In any case LF = HF ⊕ (⊕α(Lα)F ) where HF is the F -linear span of the hi's and
(Lα)F is the one-dimensional space F vα.
Remark 1. If α : H → k is a nonzero root of L then we can define a root ¯α of
LF taking into account that HF is generated as F -vector space by the hi's of the
Chevalley basis {vα}α∈Φ ∪ {hi}r
i=1. We must realize that α(hi) ∈ Z by (2) of the
definition of Chevalley basis. Thus, we define ¯α : HF → F as the F -linear extension
¯α(Pi λihi) := Pi λiα(hi). Since α is nonzero ¯α is also a nonzero. Furthermore it
is straightforward that for any h ∈ HF one has [h, vα] = ¯α(h)vα in LF .
The result in the remark below is a corollary of the main result in [2].
Remark 2. Let LF be a Chevalley algebra of exceptional type over a field F of
prime characteristic. Then the center of LF is zero except for e6 in characteristic
3 and e7 in characteristic 2.
In this cases the center Z is one-dimensional and
Z ⊂ HF .
The following results takes the idea of [1, Proposition 8.1] to whom we are
emdebted. Though in [1] the result appears in a particular context, the idea can
be extended to the following setting.
Theorem 3. Let L be a semisimple Lie algebra over an algebraically closed field k
of characteristic zero and let LF be the corresponding Chevalley F -algebra (where F
is a field of prime characteristic). Denote by V the F -vector space of all derivations
d of LF such that d(HF ) ⊂ Z := Z(LF ), then Der(LF ) = ad(LF ) + V hence we
4
P. ALBERCA, D. MART´IN, AND C. MART´IN
have the following Hesse diagram of subspaces
Der(LF )
ad(LF )
8qqqqqqqqqq
f▼▼▼▼▼▼▼▼▼▼
d■■■■■■■■■■
:✉✉✉✉✉✉✉✉✉✉
V
ad(LF ) ∩ V
Proof. Consider first the Cartan decomposition L = H ⊕ (⊕αLα) of L with
relation to H. As it is known this decomposition is a (fine) grading on L whose zero
component is H and the rest of the homogeneous components are the root spaces
Lα. Moreover, it is a group grading with grading group Zr being r = dim H.
Next we take the induced Zr-grading on the Chevalley algebra (zero component
HF and homogeneous components (Lα)F ) which will play an important role in the
forthcoming argument. Now we consider the Zr-grading on Der(LF ) such that a
derivation d is of degree z if and only if d takes the component of degree w of LF to
the component of degree w + z (for any z, w ∈ Zr). In order to prove the formula
Der(LF ) = ad(LF ) + V it suffices to prove that for any homogeneous derivation d
we can decompose d as d = ad (x) + v for some x ∈ LF and v ∈ V . We analyze
first the case in which d is of degree 0. Then d(HF ) ⊂ HF and d(vα) = λαvα
for some scalar λα ∈ F . So starting from [h, vα] = α(h)vα and applying d we
get [d(h), vα] + λα[h, vα] = α(h)λαva (for arbitrary α). Thus [d(h), vα] = 0 and
d(h) ∈ Z(LF ) so that in this case d ∈ V . Now assume that d is a derivation
of degree z 6= 0 and that d(HF ) 6= 0 (if d annihilates HF then d ∈ V ).
If the
component of degree z of LF is (Lα)F then for any h ∈ HF we have d(h) = λ(h)vα
where λ : HF → F is a linear map. If we now take arbitrary elements h, k ∈ HF ,
since [h, k] = 0 we get
λ(h)[vα, k] + λ(k)[h, vα] = 0,
implying λ(h)¯α(k) = λ(k)¯α(h). The fact that d does not annihilates HF implies
λ 6= 0 hence there is some h ∈ Hf such that ¯α = cλ being c = λ(h)−1 ¯α(h). Take
into account also that c 6= 0 since ¯α 6= 0 as pointed out in Remark 1. Finally the
reader can check that d + c−1ad (vα) is a derivation of LF in V (more precisely it
annihilates HF ).
Corollary 2.
If Z is the center of LF then
dim Der(LF ) = dimk L + dim V − dimk H.
(we write dim for dimF ).
Proof. We know that dim ad(LF ) = dim LF − dim Z = dimk L − dim Z. On
the other hand ad(LF ) ∩ V = {ad(x) : [x, HF ] ⊂ Z} therefore dim ad(LF ) ∩ V =
dim{x : [x, HF ] ⊂ Z} − dim Z. Next we prove that {x : [x, HF ] ⊂ Z} = H. Take
x = h + Pα λαva with h ∈ HF and λα ∈ F , satisfying [x, HF ] ⊂ Z. Then
Pα λαα(k)vα ∈ Z for any k ∈ H. Since Z ⊂ HF we conclude Pα λαα(k)vα = 0
hence if some λα 6= 0 then α(H) = 0 and so α = 0 a contradiction.
So dim Der(LF ) = dimk L − dim Z + dim V − dimk H + dim Z whence the result.
Theorem 4. (Referee's private communication) Under the hypothesis of Theorem
3, if the root system of L is simply laced, then dim(V ) = dim(H) and so
dim(ad (LF )) = dim(Der(LF )) − dim Z.
Thus if LF is centerless, all its derivations are inner.
8
d
:
f
INNER DERIVATIONS OF EXCEPTIONAL LIE ALGEBRAS IN PRIME CHARACTERISTIC 5
Proof. We consider a Chevalley basis {vα}α∈Φ ∪ {hi}r
i=1 of L as described in
section 2. We write [vα, vβ] = nα,βvα+β where nα,β ∈ F . Since the root system
is simply laced (for example of type E), we have nα,β = ±1, and in particular it
is nonzero modulo p. Let d be a derivation of LF such that d(HF ) ⊂ Z. Then,
the root spaces of LF with respect to HF are d-invariant. Now in type E, the root
spaces are either 1-dimensional, or 2-dimensional. The latter case only happens in
characteristic 2, where α and −α agree. In any case, there are scalars λα,i and µα,i,
such that
(d(vα) = λα,1vα + λα,2v−α
d(v−α) = µα,1vα + µα,2v−α.
Let now α = αi be a simple root, then by applying d to [vα, v−α] = hi we get
d(hi) = (λα,1 + µα,2)hi . But also d(hi) ∈ Z. Observe that no hi lies in the centre
(it would force a row of the Cartan matrix to be zero, for example). It follows that
d(hi) = 0 and µα,2 = −λα,1. In particular, we get d(HF ) = 0.
Let β be another simple root such that α + β ∈ Φ. Applying d to the equality
[vα, vβ] = nα,βvα+β we get d(vα+β ) = (λα,1 + λβ,1)vα+β . Also applying d to
[v−α, vα+β] = n−α,α+βvβ we get d(vβ ) = λβ,1vβ, that is, λβ,2 = 0. Similarly, we
get µβ,1 = 0. It follows that d(vα) = λαvα , d(vα) = −λαv−α for all simple roots
α.
Conversely, fix a scalar λα for all simple roots α, and define a linear map d : LF →
LF by d(HF ) = 0 and d(vγ ) = avγ, where a =Pα cαλα, being γ =Pα cαα. Then
by using the multiplication table of LF we see that d is a derivation of LF . The
conclusion is that the space V of all derivations of LF mapping HF into Z has
dimension exactly r (the rank of the root system).
Thus, the formula in Corollay 2 reduces to dim Der(LF ) = dimk L and since
dim(L) − dim(Z) = dim(ad (LF )) we finally get dim(ad (LF )) = dim Der(LF ) −
dim(Z). In particular when Z = 0 all the derivations are inner.
Corollary 3. For i ∈ {6, 7, 8} the derivations of Chevalley algebras of exceptional
type ei are all inner except in the cases e6 in characteristic 3 and e7 in characteristic
2.
Proof. Let LF be a Chevalley algebra of type ei with i = 6, 7 or 8. Taking
into account Remark 2 we have Z(LF ) = 0 except for LF of type e6 and the
characteristic of the ground field is 3, or LF of type e7 and the characteristic of the
ground field is 2. Then applying the formula in Theorem 4 we get the announced
result.
3. Exceptional lie algebras g2 and f4
With these two algebras, g2 and f4, we have used the software GAP to compute
the dimension of their derivation algebras. All the following routines have run
reasonably rapidly in a personal computer.
3.1. Lie algebra g2. Let us consider a Chevalley Lie algebra g2. If the character-
istic of the base field is other than 2 or 3 then every derivation is inner (Corollary
1). We present a computational approach to the problem and extend this result to
the cases not covered. We are going to prove the following result:
Theorem 5. Let L = g2 over a field F of prime characteristic p. Every derivation
in g2 is inner if, and only if, p 6= 2. Thus
(8)
ad(g2) = Der(g2) if char(F ) 6= 2.
If the characteristic is 2 then dim Der(g2) = 21 while dim g2 = 14.
6
P. ALBERCA, D. MART´IN, AND C. MART´IN
Proof.
If the characteristic of the base field is p ≥ 5 we can use Corollary
1 and (5), and then every derivation is inner. Let us now consider p = 3, since
dim Der(g2(F )) = dim Der(g2(F3)), we may take without loss of generality F = F3.
We use the software GAP. The next lines
gap> F:=GF(3);
gap> L:=SimpleLieAlgebra("G",2,F);
<Lie algebra of dimension 14 over GF(3)>
define the base field F of characteristic 3 and the Lie algebra g2 over F , which is
14-dimensional. We can also compute de center of the Lie algebra by doing
gap> LieCenter(L);
<Lie algebra of dimension 0 over GF(3)>
which is 0-dimensional. We determine the dimension of the Lie algebra Der(g2):
gap> B:=Basis(L);
CanonicalBasis( <Lie algebra of dimension 14 over GF(3)> )
gap> Derivations(B);
<Lie algebra of dimension 14 over GF(3)>
and then dim Der(g2) = 14. We have dim ad(g2) = dim g2 − dim Z(g2) = 14 − 0 =
14 = dim Der(g2). Thus every derivation is inner also if char(F ) = 3 as we have
confirmed that
(9)
ad(g2) = Der(g2), char(F ) = 3,
in this case by using a dimensional reasoning.
Now, if p = 2, we do first
gap> F:=GF(2);
gap> L:=SimpleLieAlgebra("G",2,F);
<Lie algebra of dimension 14 over GF(2)>
in order to work with g2 with char(F ) = 2, which is also 14-dimensional. We have
now to define a basis and then we can compute the dimension of Der(g2):
gap> B:=Basis(L);
CanonicalBasis( <Lie algebra of dimension 14 over GF(2)> )
gap> Derivations(B);
<Lie algebra of dimension 21 over GF(2)>
and it has dimension 21. Then
(10)
ad(g2) ( Der(g2), char(F ) = 2,
because dim ad(g2) = dim g2 = 14 6= 21, where we have used that the center is
again 0-dimensional, which can be confirmed with the next code line:
gap> LieCenter(L);
<Lie algebra of dimension 0 over GF(2)>
Finally, we present a table that summarizes the previous computations:
L
g2
g2
char.
dim L dim Z(L)
dim ad(L)
dim Der(L)
ad(L) = Der(L)
2
3
14
14
0
0
14
14
21
14
no
yes
(cid:3)
INNER DERIVATIONS OF EXCEPTIONAL LIE ALGEBRAS IN PRIME CHARACTERISTIC 7
3.2. Lie algebra f4. Let us work with the Chevalley Lie algebra f4. If the char-
acteristic of the base field is other than 2 or 3, we can use Corollary 1 to conclude
that every derivation is inner. In spite of the fact that A. Elduque and M. Kochetov
have recently proved that the result is also true if the characteristic of the base field
is 3 (see [1, Proposition 8.1]) we present an extension that confirms this result is
also true at any prime characteristic.
If the characteristic of the base field is 2 or 3 we have repeated the previous
strategy using GAP with the results summarized in the following table:
L
f4
f4
char.
dim L dim Z(L)
dim ad(L)
dim Der(L)
ad(L) = Der(L)
2
3
52
52
0
0
52
52
52
52
yes
yes
Then we have the next result:
Theorem 6. If L = f4 over a field F of prime characteristic then
(11)
ad(f4) = Der(f4),
that is to say, every derivation of f4 is inner.
4. Further considerations
We summarize all the results and the software involved at the following table:
L
g2
g2
f4
f4
e6
e6
e7
e7
e8
e8
e8
char.
2
3
2
3
2
3
2
3
2
3
5
dim Z(L)
14
14
52
52
78
78
133
133
248
248
248
0
0
0
0
0
1
1
0
0
0
0
ad(L) Der(L)
inner
14
14
52
52
78
77
132
133
248
248
248
21
14
52
52
78
78
133
133
248
248
248
no
yes
yes
yes
yes
no
no
yes
yes
yes
yes
Table 1. Dimensions and results.
And as a corollary
Theorem 7. The derivations of Chevalley algebras of exceptional type are all inner
except in the cases g2 in characteristic 2, e6 in characteristic 3 and e7 in charac-
teristic 2.
Acknowledgments
All the authors have been partially supported by the Spanish MEC and Fondos
FEDER through project MTM2010-15223, by the Junta de Andaluc´ıa and Fondos
FEDER, jointly, through projects FQM-336, FQM-02467 and FQM-3737.
The author thankfully acknowledges the computer resources, technical expertise
and assistance provided by the SCBI (Supercomputing and Bioinformatics) center
of the University of M´alaga.
8
P. ALBERCA, D. MART´IN, AND C. MART´IN
References
[1] A. Elduque and M. Kochetov, Gradings on the exceptional Lie algebras F4 and G2 revis-
ited. Revista matemtica iberoamericana. ISSN 0213-2230, Vol. 28, n 3, 2012, pgs. 773-813.
[2] J. Dieudonn´e, Les alg`ebres de Lie simples associ´ees aux groupes simples alg´ebriques sur un
corps de caract´eristique p > 0. Rendiconti del Circolo Matematico di Palermo. Volume 6,
Issue 2, 198-204, 1957.
[3] The GAP Group, GAP -- Groups, Algorithms, and Programming, Version 4.6.5 ; 2013,
(\protect\vrule width0pt\protect\href{http://www.gap-system.org}{http://www.gap-system.org}).
[4] R.B. Howlett, L.J. Rylands and D.E. Taylor, Matrix generators for exceptional groups
of Lie type. J. Symbolic Computation (2000) 11, 1-000.
[5] J. E. Humphreys, Conjugacy Classes in Semisimple Algebraic Groups. Mathematical Sur-
veys and Monographs, vol. 43, AMS.
[6] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory. Graduated Texts
in Mathematics vol. 9, Springer. 1972
[7] G.B. Seligman, Modular Lie algebras. Ergebnisse der Mathematik und ihrer Grenzgebiete,
Band 40, Springer-Verlag, New York, 1979.
[8] H. Zassenhaus, Uber Lie'sche Ringe mit Primzahlcharacteristik. Abh. Math. Sem. Univ.
Hamburg 13 (1939), 1-100.
P. Alberca Bjerregaard: Departamento de Matem´atica Aplicada, Escuela T´ecnica
Superior de Ingenieros Industriales, Universidad de M´alaga. 29071 M´alaga. Spain.
E-mail address: [email protected]
D. Mart´ın Barquero: Departamento de Matem´atica Aplicada, Escuela T´ecnica Su-
perior de Ingenieros Industriales, Universidad de M´alaga. 29071 M´alaga. Spain.
E-mail address: [email protected]
C. Mart´ın Gonz´alez: Departamento de ´Algebra Geometr´ıa y Topolog´ıa, Facultad de
Ciencias, Universidad de M´alaga, Campus de Teatinos s/n. 29071 M´alaga. Spain.
E-mail address: [email protected]
|
1207.4658 | 4 | 1207 | 2018-04-18T08:49:19 | Weakly Hyperbolic Involutions | [
"math.RA"
] | Pfister's Local-Global Principle states that a quadratic form over a (formally) real field is weakly hyperbolic (i.e. represents a torsion element in the Witt ring) if and only if its total signature is zero. This result extends naturally to the setting of central simple algebras with involution. The present article provides a new proof of this result and extends it to the case of signatures at preorderings. Furthermore the quantitative relation between nilpotence and torsion is explored for quadratic forms as well as for central simple algebras with involution. | math.RA | math |
Weakly hyperbolic involutions
Karim Johannes Bechera, Thomas Ungerb
aUniversiteit Antwerpen, Departement Wiskunde -- Informatica, Middelheimlaan 1, 2020 Antwerpen, Belgium
bSchool of Mathematics and Statistics, University College Dublin, Belfield,
Dublin 4, Ireland
Abstract
Pfister's Local-Global Principle states that a quadratic form over a (formally) real field is
weakly hyperbolic (i.e. represents a torsion element in the Witt ring) if and only if its total
signature is zero. This result extends naturally to the setting of central simple algebras with
involution. The present article provides a new proof of this result and extends it to the case
of signatures at preorderings. Furthermore the quantitative relation between nilpotence
and torsion is explored for quadratic forms as well as for central simple algebras with
involution.
Keywords: Real field, Quadratic form, Signature, Localglobal, Quaternion algebra, Witt
group
2010 MSC: primary 11E10, secondary 11E04, 11E81, 12D15, 16K20, 16W10
1. Introduction
Pfister's Local-Global Principle says that a regular quadratic form over a (formally) real
field represents a torsion element in the Witt ring if and only if its signature at each ordering
of the field is zero. This result has been extended in [13] to central simple algebras with
involution.
The theory of central simple algebras with involution is a natural extension of quadratic
form theory. On the one hand many concepts and results from quadratic form theory have
been extended to algebras with involution. On the other hand quadratic forms are used as
tools in the study of algebras with involution. Examples include involution trace forms and
spaces of similitudes.
In this article we are interested in weakly hyperbolic algebras with involution, a natural
generalization of torsion quadratic forms, which was considered first in [20, Chap. 5]. In
[13] such algebras with involution were characterized as those having trivial signature at
all orderings of the base field, thus generalizing Pfister's Local-Global Principle.
We give a new exposition of this result, highlighting several new aspects, and ob-
tain some extensions. In particular, we provide bounds for the torsion order of nilpotent
quadratic forms and extend this result to involutions. We attempt to minimize the use of
hermitian forms and treat algebras with involution as direct analogues of quadratic forms.
We only consider fields of characteristic different from 2 since the notion of weak hyper-
bolicity is only interesting in this case.
The structure of this article is as follows. In Section 2 we recall the necessary back-
ground material from the theory of quadratic forms and ordered fields as well as Pfister's
Local-Global Principle for quadratic forms in a generalized version, relative to preorder-
ings. We also touch on the quantitative aspect of the relation between nilpotence and tor-
sion, using Lewis' annihilating polynomials.
Email addresses: [email protected] (Karim Johannes Becher),
[email protected] (Thomas Unger)
August 8, 2017
K.J. Becher, T. Unger: Weakly hyperbolic involutions
2
In Section 3 we recall the basic terminology for algebras with involution, consider their
relations to quaternion algebras and quadratic forms and study involution trace forms.
In Section 4 we treat the notion of hyperbolicity for algebras with involution and cite
the relevant results about its behaviour under field extensions.
In Section 5 we turn to the study of algebras with involution over real fields. In The-
orem 5.2 we obtain a classification over real closed fields. We then provide a uniform
definition of signatures for involutions of both kinds with respect to an ordering. Signa-
tures were introduced in [12] for involutions of the first kind and in [16] for involutions of
the second kind, and both cases are treated in [7, (11.10), (11.25)].
In Section 6 we give a new proof of the main result of [13], an analogue of Pfister's
Local-Global Principle for algebras with involution (Theorem 6.5). In Theorem 6.7 we
extend this result to a local-global principle for T -hyperbolicity with respect to a preorder-
ing T .
In its original version for quadratic forms as well as in the generalized version for al-
gebras with involution Pfister's Local-Global Principle relates the hyperbolicity of tensor
powers to the hyperbolicity of multiples. For quadratic forms this corresponds to the re-
lation between nilpotence and torsion for an element of the Witt ring. In Section 7 we
consider the quantitative aspect of this relation in the setting of algebras with involution.
Some of the essential ideas contained in Sections 5 and 6 germinated in the MSc thesis
of Beatrix Bernauer [2], prepared under the guidance of the first named author.
2. Pfister's Local-Global Principle
We refer to [8] and [17] for the foundations of quadratic form theory over fields and the
relevant terminology. Let K be a field of characteristic different from 2. We denote by K×
the multiplicative group of K, by K×2 the subgroup of nonzero squares, and by PK2 the
subgroup of nonzero sums of squares in K. IfPK2 = K×2 then K is said to be pythagorean.
By a quadratic form over K (or just a form) we mean a pair (V, B) consisting of a finite-
dimensional K-vector space V and a regular symmetric K-bilinear form B. We consider
quadratic forms up to isometry and use the equality sign to indicate that two forms are
isometric. For a form ϕ = (V, B) over K we write DK (ϕ) = {a ∈ K× a = B(x, x)}. We
further abbreviate DK(m) = DK (m × h1i), which is the set of nonzero sums of m squares
in K. Given n ∈ N and a1, . . . , an ∈ K×, we use the standard notations ha1, . . . , ani and
hha1, . . . , anii = h1, −a1i⊗· · ·⊗h1, −ani to denote diagonalized quadratic forms of dimension
n and n-fold Pfister forms, respectively. Given a quadratic form ϕ over K and m ∈ N, we
write m × ϕ for the m-fold orthogonal sum ϕ ⊥ · · · ⊥ ϕ and further ϕ⊗m for the m-fold
tensor product ϕ ⊗ · · · ⊗ ϕ.
Let WK denote the Witt ring of K and IK its fundamental ideal. We sometimes write
[ϕ] to denote the class in WK given by a form ϕ over K.
An ordering of K is a set P ⊆ K that is additively and multiplicatively closed and that
satisfies P ∪ −P = K and P ∩ −P = {0}. Any such set P is the positive cone {x ∈ K x > 0}
for a unique total order relation 6 on K that is compatible with the field operations. Let
XK denote the set of orderings of K. If the field K has an ordering then we say that it is
real, otherwise nonreal. By the Artin-Schreier Theorem the field K is real if and only if
−1 < PK2.
Let T ⊆ K be additively and multiplicatively closed with K×2 ∪ {0} ⊆ T . Then the
set T + xT = {s + xt s, t ∈ T} is additively and multiplicatively closed for any x ∈ K.
Moreover T × = T \ {0} is a subgroup of K× containing PK2. If further −1 < T , then T is
called a preordering of K. Any ordering is a preordering. Furthermore, if T is a preordering
of K, then so is T + xT for any x ∈ K \ −T . It follows from this that a preordering of K is
maximal with respect to inclusion if and only if it is an ordering. Hence, every preordering
is contained in an ordering. For a preordering T of K we set
XT = {P ∈ XK T ⊆ P} .
K.J. Becher, T. Unger: Weakly hyperbolic involutions
3
Given a quadratic form ϕ over K, we denote the signature of ϕ at an ordering P ∈ XK
by signP(ϕ) and obtain a map bϕ : XK → Z, P 7−→ signP(ϕ). This gives rise to a ring
homomorphism
called the total signature. If K is nonreal, then XK = ∅ and ZXK is the ring with one element.
Let T be a fixed preordering of K. We write
sign : WK → ZXK , ϕ 7→bϕ
and we denote the kernel of this homomorphism by IT K.
signT : WK → ZXT , ϕ 7→bϕXT
Let ϕ be a quadratic form over K. We say that ϕ is T -positive if ϕ is nontrivial and
DK (ϕ) ⊆ T ×. If a1, . . . , an ∈ K× are such that ϕ = ha1, . . . , ani, then ϕ is T -positive if
and only if a1, . . . , an ∈ T ×. Hence, orthogonal sums and tensor products of T -positive
forms are again T -positive. We say that ϕ is T -isotropic or T -hyperbolic if there exists a
T -positive form ϑ over K such that ϑ ⊗ ϕ is isotropic or hyperbolic, respectively.
to a preordering (cf. [9, (1.26)]).
The following statement is a generalization of Pfister's Local-Global Principle, relative
Theorem 2.1 (Pfister). Let T be a preordering of K. The ideal IT K is generated by the
classes of binary forms h1, −ti with t ∈ T ×. Moreover, for a quadratic form ϕ over K the
following statements are equivalent:
(i) signT (ϕ) = 0.
(ii) The form ϕ is T -hyperbolic.
(iii) There exists a T -positive Pfister form τ over K such that τ ⊗ ϕ is hyperbolic.
(iv) There exist r > 0, a1, . . . , ar ∈ K× and t1, . . . , tr ∈ T × such that ϕ is Witt equivalent to
ha1, −a1t1i ⊥ · · · ⊥ har, −artri.
A quadratic form ϕ over K is said to be torsion or weakly hyperbolic if m × ϕ is hyper-
bolic for some positive integer m. The following is [15, Satz 10 and Satz 22].
Corollary 2.2 (Pfister). Assume that K is real. For a quadratic form ϕ over K the following
statements are equivalent:
(i) sign(ϕ) = 0.
(ii) The quadratic form ϕ is weakly hyperbolic.
(iii) There exists m ∈ N such that 2m × ϕ is hyperbolic.
(iv) There exists n ∈ N such that ϕ⊗n is hyperbolic.
(v) There exist r > 0, a1, . . . , ar ∈ K× and s1, . . . , sr ∈ PK2 such that ϕ is Witt equivalent
to ha1, −a1s1i ⊥ · · · ⊥ har, −ar sri.
Corollary 2.3 (Pfister). The order of any torsion element in WK is a 2-power.
For n ∈ N let
Ln(X) =
nY
i=0
(X − n + 2i) ∈ Z[X].
Note that Ln(−X) = (−1)n+1 · Ln(X), so that the polynomial Ln(X) is either even or odd.
In [10], Lewis showed that these polynomials have a crucial property relating to quadratic
forms. We include the short proof due to K.H. Leung, which is mentioned in [10].
K.J. Becher, T. Unger: Weakly hyperbolic involutions
4
Theorem 2.4 (Lewis). Let n ∈ N and let ϕ be a quadratic form of dimension n over K.
Then Ln([ϕ]) = 0 in WK.
Proof. As (aϕ)⊗2 = ϕ⊗2 for all a ∈ K× and Ln(X) is either even or odd, we may scale ϕ
and assume that ϕ = ϕ′ ⊥ h1i where ϕ′ is a form of dimension n − 1. Using the induction
hypothesis for ϕ′ we obtain that Ln−1([ϕ] − 1) = Ln−1([ϕ′]) = 0. In view of the equality
Ln(X) = (X + n) · Ln−1(X − 1) we conclude that Ln([ϕ]) = 0.
As shown in [10] and [11] the last statement can be used to prove some statements about
the Witt ring of quadratic forms which were obtained earlier by different methods, usually
involving field extensions. This holds in particular for Theorem 2.1 and Corollary 2.2.
Another such example is the following statement.
Corollary 2.5 (Scharlau). Any zero-divisor of WK lies in IK.
i=0(2i + 1)2, thus β = 0 by Corollary 2.3.
Proof. Let α ∈ WK \ IK. By Theorem 2.4 there exists n ∈ N such that α is a zero of
L2n+1(X) = Qn
i=0(X2 − (2i + 1)2). Hence, for β ∈ WK with αβ = 0 we have mβ = 0 for the
odd integer m = Qn
Corollary 2.6. Let n ∈ N and let ϕ be a quadratic form of dimension 2n over K. Then
22n−1n!(n − 1)! · [ϕ] is a multiple of [ϕ]2 in WK.
Proof. We may scale ϕ and assume that ϕ = h1i ⊥ ϕ′ where ϕ′ is a form of dimension
2n− 1. Then L2n−1([ϕ′]) = 0 by Theorem 2.4. It follows that [ϕ] is a zero of the polynomial
L2n−1(X − 1) = (X − 2n)X
This implies the statement.
(X2 − 4i2).
n−1Y
i=1
For n ∈ N we denote by d(n) the number of occurrences of the digit 1 in the binary
representation of n. Note that d(2n) = d(n) and d(2n + 1) = d(n) + 1. In [4, §4.4] the
following observation is attributed to Legendre.
Proposition 2.7. For n ∈ N the largest 2-power dividing n! is equal to 2n−d(n).
Proof. Let n ∈ N. The largest 2-power dividing n is 2m where m is the number of consecu-
tive digits 1 at the end of the binary representation of n− 1, whereby m = d(n− 1)− d(n) + 1.
Hence the largest 2-power dividing n! is 2k, where k = Pn
For n > 1 we set ∆(n) = 2n − 1 − d(n) − d(n − 1).
i=1(cid:0)d(i−1)−d(i)+1(cid:1) = n−d(n).
Theorem 2.8. Let ϕ and π be quadratic forms over K such that ϕ ⊗ ϕ ⊗ π is hyperbolic.
Then 2∆(n) × ϕ ⊗ π is hyperbolic for n = dim(ϕ).
Proof. Assume that π is not hyperbolic, as otherwise the statement is trivial. Then n =
dim(ϕ) = 2k for some k ∈ N, by Corollary 2.5. It follows from Corollary 2.6 that the form
22k−1k!(k − 1)! × ϕ ⊗ π is hyperbolic. By Corollary 2.3 the order of [ϕ ⊗ π] in W K is a
2-power and in view of Proposition 2.7 the largest 2-power dividing 22k−1k!(k − 1)! is 2m
where
m = (2k − 1) + k − d(k) + k − 1 − d(k − 1) = 4k − 1 − d(2k) − d(2k − 1) = ∆(n).
Hence 2∆(n) × ϕ ⊗ π is hyperbolic.
Remark 2.9. In view of Theorem 2.8 we may define a function g : N → N in the following
way: for k ∈ N, let g(k) be the smallest number m ∈ N such that, for any quadratic form
ϕ of dimension 2k over an arbitrary field of characteristic different from 2 for which ϕ ⊗ ϕ
is hyperbolic, also 2m × ϕ is hyperbolic. Applying Theorem 2.8 with π = h1i yields that
g(k) 6 ∆(2k) = 4k − 2 − d(k)− d(k − 1). This bound, however, does not seem to be optimal
for k > 1. In fact, it is not difficult to show that g(2) = g(3) = 2.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
5
3. Algebras with involution
Our general references for the theory of central simple algebras and their involutions
are [7] and [17, Chap. 8]. We fix some terminology.
Let K be a field and let A be a K-algebra. We call A a K-division algebra if every non-
zero element in A is invertible. We denote by Z(A) the centre of A. We call the K-algebra A
central simple if it is finite-dimensional over K, simple as a ring and such that Z(A) = K. A
K-involution on A is a K-linear map σ : A → A such that σ(xy) = σ(y)σ(x) for all x, y ∈ A
and σ ◦ σ = idA.
Assume now that A is a central simple K-algebra. Wedderburn's Theorem says that in
this case dim(A) = n2 for a positive integer n, called the degree of A and denoted deg(A),
and further that A is isomorphic to a matrix algebra over a central simple K-division algebra
D unique up to K-isomorphism; one says that A is split if D = K. If A and A′ are matrix
algebras over the same central simple K-division algebra, then A and A′ are called Brauer
equivalent and we write A ∼ A′.
A K-algebra with involution is a pair (A, σ) where A is a finite-dimensional K-algebra
and σ is a K-involution on A such that K = {x ∈ Z(A) σ(x) = x} and such that either A is
simple or A is a product of two simple K-algebras that are mapped to each other by σ. We
will often denote a K-algebra with involution by a single capital Greek letter.
Let (A, σ) be a K-algebra with involution. We have either Z(A) = K or Z(A) is a
quadratic ´etale extension of K. We say that (A, σ) is of the first or second kind, depending
on whether [Z(A) : K] is 1 or 2, respectively. Note that A is simple if and only if Z(A) is a
field. If Z(A) is not a field then it is isomorphic to F × F; this can only occur if (A, σ) is of
the second kind. We have dimK(A) = [Z(A) : K] · n2 for a positive integer n ∈ N, which we
call the degree of (A, σ); if A is simple, then deg(A, σ) is just the degree of A as a central
simple Z(A)-algebra.
We say that x ∈ A is symmetric or skew-symmetric (with respect to σ) if σ(x) = x or
σ(x) = −x, respectively. We let
Sym(A, σ) = {x ∈ A σ(x) = x}
and Skew(A, σ) = {x ∈ A σ(x) = −x} .
These are K-linear subspaces of A.
Assume from now on that the characteristic of K is different from 2. Then
and there exists ε ∈ {−1, 0, +1} such that
A = Sym(A, σ) ⊕ Skew(A, σ)
dimK(Sym(A, σ)) = 1
2 n(n + ε) and dimK (Skew(A, σ)) = 1
2 n(n − ε) .
If (A, σ) is of the first kind, then ε = ±1, and we say that (A, σ) is orthogonal if ε = 1 and
symplectic if ε = −1. If (A, σ) is of the second kind, then ε = 0, and we say that (A, σ) is
unitary. The integer ε is called the type of (A, σ) and denoted type(A, σ). We say that (A, σ)
is unitary of inner type when Z(A) ≃ F × F. (The term is motivated by a corresponding
notion for algebraic groups.)
Following [7, §12], given a K-algebra with involution (A, σ), we denote
Sim(A, σ) = {x ∈ A× σ(x)x ∈ K×}
G(A, σ) = {σ(x)x x ∈ Sim(A, σ)} ;
and
note that these are subgroups of A× and of K×, respectively.
Let Ψ = (A, σ) be a K-algebra with involution. In this notational setting we denote the
underlying K-algebra A by Ψ. We say that Ψ is split if Ψ is isomorphic to a matrix algebra
over Z(Ψ).
For any field extension L/K the L-algebra with involution (A ⊗K L, σ ⊗ idL) is denoted
by ΨL. Note that type(ΨL) = type(Ψ) and that ΨL is simple if and only if Ψ is simple and
L is linearly disjoint to Z(Ψ) over K.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
6
We now consider two K-algebras with involution Ψ = (A, σ) and Θ = (B, ϑ). A homo-
morphism of algebras with involution Ψ → Θ is a K-homomorphism f : A → B satisfying
ϑ◦ f = f ◦σ. A homomorphism is called an embedding if it is injective and an isomorphism
if it is bijective. We write Ψ ≃ Θ if there exists and isomorphism Ψ → Θ. (This occurs
if and only if either Ψ and Θ are adjoint to two similar hermitian or skew-hermitian forms
over some K-division algebra with involution or Ψ and Θ are both unitary of inner type
with Ψ ≃ Θ.)
Except when Ψ and Θ are both unitary with different centres, we can define their tensor
product Ψ ⊗ Θ. If Ψ and Θ are not both unitary, let Ψ ⊗ Θ denote the K-algebra with
involution (A ⊗K B, σ ⊗ ϑ). If Ψ and Θ are both unitary and with same centre L, then Ψ ⊗ Θ
is the unitary K-algebra with involution (A⊗L B, σ⊗ϑ), whose centre is also L. In particular,
for a positive integer n the tensor power Ψ⊗n of a K-algebra with involution Ψ is defined.
Furthermore, in each of the cases where we defined Ψ ⊗ Θ we have
deg(Ψ ⊗ Θ) = deg(Ψ) · deg(Θ)
and
type(Ψ ⊗ Θ) = type(Ψ) · type(Θ) .
Let ϕ = (V, B) be a quadratic form over K. Consider the split central simple K-algebra
EndK (V). Let σ : EndK (V) → EndK(V) denote the involution determined by the formula
B( f (u), v) = B(u, σ( f )(v))
for all u, v ∈ V and f ∈ EndK(V).
We denote this involution σ by adB and call it the adjoint involution of ϕ. Furthermore, we
call (EndK(V), adB) the adjoint algebra with involution of ϕ and denote it by Ad(ϕ). Note
that Ad(ϕ) is split orthogonal and determines ϕ up to similarity.
Example 3.1. Let n be a positive integer and ϕ = n×h1i, the n-dimensional form h1, . . . , 1i
over K. Then Ad(ϕ) ≃ (Mn(K), t) where Mn(K) is the K-algebra of n× n-matrices over K
and t is the transpose involution.
Proposition 3.2. For quadratic forms ϕ and ψ over K we have
Ad(ϕ ⊗ ψ) ≃ Ad(ϕ) ⊗ Ad(ψ) .
Proof. Denoting V and W the underlying vector spaces of ϕ and ψ, respectively, the natu-
ral K-algebra isomorphism EndK (V) ⊗K EndK(W) → EndK(V ⊗K W) yields the required
identification for the adjoint involutions.
map (cf. [7, p. 5 and p. 22]).
For a finite-dimensional K-algebra A we denote by TrdA : A → Z(A) its reduced trace
A K-quaternion algebra is a central simple K-algebra of degree 2. Given a K-quaternion
algebra Q, the map σ : Q → K, x 7→ TrdQ(x) − x is a K-involution, called the canonical
involution on Q and denoted by canQ; this is the unique symplectic K-involution on Q.
If L is a quadratic ´etale extension of K we denote by canL/K the unique nontrivial K-
automorphism of L. We further set canK = idK . We call (A, σ) a K-algebra with canonical
involution if it is of one of the forms (K, idK ), (L, canL/K ) for a quadratic ´etale extension
L/K, or (Q, canQ) for a K-quaternion algebra Q.
Proposition 3.3. Let (A, σ) be a K-algebra with involution. We have that Sym(A, σ) = K
if and only if (A, σ) is a K-algebra with canonical involution.
Proof. If Sym(A, σ) = K then dimK (A) = 21−ε for ε = type(A, σ), so that A is either K, a
quadratic ´etale extension of K, or a K-quaternion algebra. In any of these three cases, the
canonical involution of A is the unique K-involution of type ε, hence it is therefore equal
to σ and Sym(A, σ) = K.
The following statement goes back to Jacobson [5].
K.J. Becher, T. Unger: Weakly hyperbolic involutions
7
Proposition 3.4. Let Ψ be a K-algebra with involution. Then Ψ ≃ Ad(ϕ) ⊗ Φ for a K-
algebra with canonical involution Φ and a quadratic form ϕ over K if and only if Ψ is
either split or symplectic of index 2.
Proof. Clearly, any K-algebra with canonical involution Φ is either split or symplectic of
index 2, and thus so is Φ ⊗ Ad(ϕ) for any quadratic form ϕ over K. Assume now that Ψ
is either split or symplectic of index 2. Then Ψ is adjoint to a hermitian form over a K-
algebra with canonical involution Φ. Since such a form has a diagonalisation with entries
in Sym(Ψ) = K, we obtain that Ψ ≃ Ad(ϕ) ⊗ Φ for a form ϕ over K.
For computational purposes we augment the classical notation for quaternion algebras
in terms of pairs of field elements to take into account an involution. Let a, b ∈ K× and
let Q be the K-algebra with basis (1, i, j, k), where i2 = a, j2 = b and i j = − ji = k. This
quaternion algebra is denoted by (a, b)K. For δ, ε ∈ {+1, −1} there is a unique K-involution
σ on Q such that σ(i) = δi and σ( j) = ε j. We denote the pair (Q, σ) by
(a b)K
(a · b)K
(a · b)K
(a ·· b)K
if
if
if
if
δ = +1, ε = +1,
δ = −1, ε = +1,
δ = +1, ε = −1,
δ = −1, ε = −1.
In particular, (a ·· b)K denotes the quaternion algebra (a, b)K together with its canonical
involution. Any K-quaternion algebra with orthogonal involution is isomorphic to (a · b)K
for some a, b ∈ K×. Note that (a b)K ≃ (−ab · b)K and (a · b)K ≃ (b · a)K for any a, b ∈
K×. Hence the presence or absence of a dot indicates on which quadratic ´etale K-subalgebra
the involution restricts to the nontrivial automorphism or to the identity, respectively.
Proposition 3.5. Let Q be a K-quaternion algebra. Let a, b ∈ K× be such that Q ≃ (a, b)K.
Then (Q, canQ) ≃ (a ·· b)K. Moreover, for i ∈ Q× \ K× with i2 = a and τ = Int(i) ◦ canQ we
have (Q, τ) ≃ (a · b)K.
Proof. Since canQ is the only symplectic involution on Q, any K-isomorphism Q → (a, b)K
is also an isomorphism of K-algebras with involution. Hence, (Q, canQ) ≃ (a ·· b)K. We
choose an element i ∈ Q× \ K× such that i2 = a. Then V = { j ∈ Q i j + ji = 0} is
the orthogonal complement of K[i] in Q with respect to the symmetric K-bilinear form
B : Q × Q → K, (x, y) 7→ 1
2 TrdQ(canQ(x) · y). By [17, Chap. 2, (11.4)] we obtain that
(Q, B) ≃ hha, bii. Since (K[i], BK[i]) ≃ hhaii it follows that (V, BV) ≃ −bhhaii. As for any
j ∈ V we have B( j, j) = − j2, there exists an element j ∈ V with j2 = b. For τ = Int(i)◦canQ
we obtain that τ(i) = −i and τ( j) = j, whereby (Q, τ) ≃ (a · b)K.
For a ∈ K×, let (a)K denote the unitary K-algebra with canonical involution (L, canL/K )
where L = K[X]/(X2 − a); it is of inner type if and only if a ∈ K×2.
Corollary 3.6. Let Φ be a K-quaternion algebra with involution. If Φ is orthogonal there
exist a, b ∈ K× such that Φ ≃ (a · b)K. If Φ is symplectic there exist a, b ∈ K× such that
Φ ≃ (a ·· b)K. If Φ is unitary, there exist a, b, c ∈ K× such that Φ ≃ (a ·· b)K ⊗ (c)K.
Proof. Assume that Φ is of the first kind and let Φ = (Q, σ). If Φ is symplectic, then
σ = canQ and we choose a, b ∈ Q× such that Q ≃ (a, b)K to obtain by Proposition 3.5 that
Φ ≃ (a ·· b)K. If Φ is orthogonal, we choose i ∈ Skew(Q, σ)∩ Q× and a, b ∈ K× with a = i2
and Q ≃ (a, b)K, and obtain that σ = Int(i) ◦ canQ, so that Φ ≃ (a · b)K by Proposition 3.5.
Assume now that Φ is of the second kind. From [7, (2.22)] we obtain that Φ ≃
(Q, canQ) ⊗ (c)K for a K-quaternion algebra Q and an element c ∈ K×, and by the above
there exist a, b ∈ K× such that (Q, canQ) ≃ (a ·· b)K.
Proposition 3.7. Let a, b, c, d ∈ K×. We have (a · b)K ≃ (c · d)K if and only if aK×2 = cK×2
and bd ∈ DKhhaii.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
8
Proof. We set (Q, τ) = (a · b)K. There exist i ∈ Skew(Q, τ) and j ∈ Sym(Q, τ) with i2 = a,
j2 = b, and i j + ji = 0.
Assuming that bd ∈ DKhhaii, we may write d = b(u2 − av2) with u, v ∈ K and obtain for
g = u j + vi j that g ∈ Sym(Q, τ), gi + ig = 0, and g2 = d. If further cK×2 = aK×2, then
c = f 2 for some f ∈ iK×, and we have f ∈ Skew(Q, τ) and g f + f g = 0, and conclude that
(Q, τ) ≃ (c · d)K.
For the converse, suppose that (Q, τ) ≃ (c · d)K. There exist f ∈ Skew(Q, τ) and
g ∈ Sym(Q, τ) with f 2 = c, g2 = d, and f g+g f = 0. It follows that iK = Skew(Q, τ) = f K,
so that aK×2 = cK×2. Moreover, ig + gi = 0 and jgi = i jg. As K[i] is a maximal
commutative K-subalgebra of Q, we obtain that jg ∈ K[i]. Writing jg = x + iy with
x, y ∈ K, we obtain that
bd = j2g2 = j(x + iy)g = (x − iy) jg = (x − iy)(x + iy) = x2 − ay2 ,
whence bd ∈ DKhhaii.
Remark 3.8. For a K-algebra with orthogonal involution (A, σ) of degree 2m the class in
K×/K×2 given by the reduced norm of an arbitrary element u ∈ A× ∩ Skew(A, σ) times
(−1)m yields an invariant of (A, σ), called the discriminant (cf. [7, (7.1) and (7.2)]). For
a, b ∈ K× the discriminant of (a · b)K is aK×2. Hence Proposition 3.7 contains the obser-
vation that this is an invariant in the case where A is a K-quaternion algebra.
Proposition 3.9. For a, b, c, d ∈ K× we have
(a · b)K ⊗ (c · d)K ≃ (a ·· bc)K ⊗ (c ·· ad)K
(a · b)K ⊗ (c ·· d)K ≃ (a ·· bc)K ⊗ (c · ad)K .
and
Proof. Let (A, σ) = (a · b)K ⊗ (c · d)K. Then there exist elements i, j, f , g ∈ A× such
f 2 = c, g2 = d,
that σ(i) = −i, σ( j) = j, σ( f ) = − f , σ(g) = g, i2 = a,
i j + ji = f g + g f = 0, and each of i and j commutes with each of f and g. Set j′ = f j
and g′ = ig. Then σ(i) = −i, σ( j′) = − j′, σ( f ) = − f , σ(g′) = −g′, i2 = a, j′2 = bc,
f 2 = c, g′2 = ad, i j′ + j′i = f g′ + g′ f = 0, and each of i and j′ commutes with each of
f and g′. The K-subalgebra Q of A generated by i and j′ commutes elementwise with the
K-subalgebra Q′ of A generated by f and g′, and Q and Q′ are σ-stable. Hence
j2 = b,
(A, σ) ≃ (Q, σQ) ⊗ (Q′, σQ′ ) ≃ (a ·· bc)K ⊗ (c ·· ad)K .
This shows the first isomorphism. The proof of the second isomorphism is almost identical,
with the only difference that σ(g) = −g and σ(g′) = g′.
Proposition 3.10. For a, b ∈ K×, we have
G (a ·· b)K = DKhha, bii
and G (a · b)K = DKhhaii ∪ bDKhhaii.
Proof. Let Q = (a, b)K and u, v ∈ Q× with u2 = a, v2 = b and uv + vu = 0. Then
Sim(Q, canQ) = Q× and thus G(Q, canQ) = DKhha, bii. For τ = Int(u) ◦ canQ we obtain
Sim(Q, τ) = K(u)× ∪ vK(u)× and thus G(Q, τ) = DKhhaii ∪ bDKhhaii.
A K-algebra with involution (A, σ) with centre L gives rise to a regular hermitian form
T(A,σ) : A× A → L over (L, canL/K ) defined by T(A,σ)(x, y) = TrdA(σ(x)y); this follows from
[7, (2.2) and (2.16)]. We further obtain a regular symmetric K-bilinear form Tσ : A×A → K
defined by Tσ(x, y) = 1
2 TrdA(σ(x)y + σ(y)x). Note that if L = K then Tσ = T(A,σ), otherwise
2Tσ = T ◦ T(A,σ) where T is the trace of L/K.
(A, Tσ) over K. Note that dim(Tr(Ψ)) = dimK(A).
involution, then Tr(Ψ) is the norm form of the K-algebra Ψ.
Given a K-algebra with involution Ψ = (A, σ), we denote by Tr(Ψ) the quadratic form
If Ψ is a K-algebra with canonical
Note that for a quadratic form ϕ the notation 2ϕ refers to h2i ⊗ ϕ (the form obtained by
scaling ϕ by 2), which needs to be distinguished from 2 × ϕ = ϕ ⊥ ϕ.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
9
Example 3.11. For a ∈ K× we have Tr (a)K = hhaii. For a, b ∈ K× we have Tr (a · b)K =
2hha, −bii and Tr (a ·· b)K = 2hha, bii.
Proposition 3.12. Let Ψ and Θ be K-algebras with involution. If Ψ is of the first kind,
then Tr(Ψ ⊗ Θ) ≃ Tr(Ψ) ⊗ Tr(Θ).
If Ψ and Θ are both unitary with same centre, then
2 × Tr(Ψ ⊗ Θ) ≃ Tr(Ψ) ⊗ Tr(Θ).
Proof. Write Ψ = (A, σ) and Θ = (B, τ). Let K′ = Z(A) and L = Z(B). In view of the
claims we may assume that K′ ⊆ L. For a ∈ A and b ∈ B we have TrdA⊗K′ B(a ⊗ b) =
TrdA(a) · TrdB(b), as one verifies by reduction to the split case. Hence, TΨ⊗Θ and TΨ ⊗ TΘ
coincide as hermitian forms on A ⊗K′ B with respect to (L, canL/K ). If L = K then we are
done. Assume now that (L, canL/K ) ≃ (c)K where c ∈ K×. Then Tr(Θ) ≃ hhcii ⊗ ϑ for a
form ϑ over K. If now K′ = K then Tr(Ψ) ⊗ Tr(Θ) ≃ hhcii ⊗ (Tr(Ψ) ⊗ ϑ) ≃ Tr(Ψ ⊗ Θ). In the
remaining case K′ = L and Tr(Ψ) ≃ hhcii ⊗ ψ for a quadratic form ψ over K, and we obtain
that Tr(Ψ) ⊗ Tr(Θ) ≃ hhc, cii ⊗ ψ ⊗ ϑ ≃ 2 × hhcii ⊗ (ψ ⊗ ϑ) ≃ 2 × Tr(Ψ ⊗ Θ).
Proposition 3.13. For any form ϕ over K we have Tr(Ad(ϕ)) ≃ ϕ ⊗ ϕ.
Proof. See [7, (11.4)].
Let A be a finite-dimensional K-algebra. For a ∈ A let λa ∈ EndK (A) be given by
λa(x) = ax for x ∈ A. The K-algebra homomorphism λ : A → EndK (A), a 7→ λa thus
obtained is called the left regular representation of A.
Proposition 3.14. Let Ψ be a K-algebra with involution. The left regular representation of
Ψ yields an embedding of Ψ into Ad(Tr(Ψ)).
Proof. Write Ψ = (A, σ). For a, x, y ∈ A we have Tσ(x, λσ(a)(y)) = Tσ(λa(x), y). Thus λ
identifies σ with the restriction to λ(A) of the involution adjoint to Tσ.
Proposition 3.15. Let Ψ be a K-algebra with involution of the first kind. Then
Ψ ⊗ Ψ ≃ Ad(Tr(Ψ)) .
Proof. We expand the proof of [7, (11.1)]. Let Ψ = (A, σ). Consider the K-algebra homo-
morphism σ∗ : A⊗K A → EndK (A) determined by σ∗(a⊗ b)(x) = axσ(b) for all a, b, x ∈ A.
As A ⊗K A is simple and of the same dimension as EndK (A), σ∗ is an isomorphism. For
a, b, x, y ∈ A we have
Tσ(x, σ∗(σ(a) ⊗ σ(b))(y)) = Tσ(σ∗(a ⊗ b)(x), y) .
Thus σ∗ identifies the involution σ ⊗ σ with the adjoint involution of Tσ.
4. Hyperbolicity
Following [1, (2.1)], we say that the K-algebra with involution (A, σ) is hyperbolic if
there exists an element e ∈ A with e2 = e and σ(e) = 1 − e. If (A, σ) is adjoint to a
hermitian form over a K-division algebra with involution, then it is hyperbolic if and only
if the hermitian form is hyperbolic.
Proposition 4.1. The K-algebra with involution (A, σ) is hyperbolic if and only if there
exists f ∈ Skew(A, σ) with f 2 = 1, if and only if (1)K embeds into (A, σ).
Proof. The second equivalence is obvious. To prove the first equivalence, given e ∈ A with
e2 = e and σ(e) = 1 − e, we see that f = 2e − 1 satisfies σ( f ) = − f and f 2 = 1, and
conversely, for f ∈ A with these properties, the element e = 1
2 ( f − 1) satisfies e2 = 1 and
σ(e) = 1 − e.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
10
Corollary 4.2. Let Ψ be a K-algebra with involution which is either split symplectic or
unitary of inner type. Then Ψ is hyperbolic.
Proof. Using Proposition 3.4 we have that Ψ ≃ Ad(ϕ) ⊗ Φ for a K-algebra with canonical
involution Φ, and conclude that Φ ≃ (1 ·· 1)K or Φ ≃ (1)K. In either case Ψ contains (1)K
and thus is hyperbolic by Proposition 4.1.
Proposition 4.3. Let Ψ and Θ be K-algebras with involution.
(a) If Ψ and Θ are hyperbolic, type(Φ) = type(Ψ) and Ψ ≃ Θ, then Ψ ≃ Θ.
(b) If Ψ is hyperbolic, then Ψ ⊗ Θ is hyperbolic.
(c) If Ψ is hyperbolic, then Tr(Ψ) is hyperbolic.
Proof. (a) If Ψ and Θ are unitary of inner type, the statement follows from [7, (2.14)].
Otherwise Ψ and Θ are adjoint to hyperbolic hermitian or skew-hermitian forms of the same
dimension over a common K-division algebra with involution, and these are necessarily
isometric.
(b) This is obvious.
(c) By Proposition 3.14, Ψ embeds into Ad(Tr(Ψ)). This yields the statement.
Proposition 4.4. Let Ψ be a K-algebra with involution and a ∈ K×. We have that a ∈ G(Ψ)
if and only if Adhhaii ⊗ Ψ is hyperbolic.
Proof. See [7, (12.20)].
Theorem 4.5 (Bayer-Fluckiger, Lenstra). Let Ψ be a K-algebra with involution and let
L/K be a finite field extension of odd degree. Then ΨL is hyperbolic if and only if Ψ is
hyperbolic.
Proof. See [7, (6.16)].
The following is a reformulation of the main result in [5].
Theorem 4.6 (Jacobson). Let Φ be a K-algebra with canonical involution and ϕ a quadratic
form over K. Then Ad(ϕ) ⊗ Φ is hyperbolic if and only if ϕ ⊗ Tr(Φ) is hyperbolic.
Proof. Let ϕ = (V, B) and Φ = (A, σ). We may assume that A has no zero-divisors since
otherwise Φ and Tr(Φ) are hyperbolic. The K-algebra with involution Ad(ϕ)⊗Φ is adjoint to
the hermitian form (VA, h) over Φ obtained from ϕ, with VA = V ⊗K A and h : VA × VA → A
determined by h(v ⊗ a, w ⊗ b) = σ(a)B(v, w)b for a, b ∈ A and v, w ∈ V. Let T : A → K,
x 7→ x + σ(x). Note that T ◦ h coincides with B ⊗ Tσ on VA × VA up to a scalar factor.
The isotropic vectors of the hermitian form h and of the quadratic forms T ◦ h and B ⊗ Tσ
therefore coincide. In particular, a maximal totally isotropic A-subspace for h is the same
as a maximal totally isotropic K-subspace for B ⊗ Tσ. This yields the statement.
Theorem 4.7 (Bayer-Fluckiger, Shapiro, Tignol). Let Ψ be a K-algebra with involution
and a ∈ K×\K×2. Then ΨK( √a) is hyperbolic if and only if (a)K embeds into Ψ or Ψ ≃ Ad(ϕ)
for a quadratic form ϕ over K whose anisotropic part is a multiple of hhaii.
Proof. Assume first that Ψ is split orthogonal, so that Ψ ≃ Ad(ϕ) for a form ϕ over K. Then
ΨK( √a) is hyperbolic if and only if ϕK( √a) is hyperbolic, which by [8, Chap. VII, (3.2)] is
the case if and only if the anisotropic part of ϕ is a multiple of hhaii. In the remaining cases,
the statement is proven in [1, (3.3)] for involutions of the first kind, and an adaptation of
the argument for involutions of the second kind is provided in [13, (3.6)].
K.J. Becher, T. Unger: Weakly hyperbolic involutions
11
Remark 4.8. There is an overlap in the two cases of the characterization given in Theo-
rem 4.7. Assume that a ∈ K× \ K×2 and ϕ is a form over K. Then (a)K embeds into Ad(ϕ)
if and only if ϕ is a multiple of hhaii, if and only if the anisotropic part ϕan of ϕ is a multiple
of hhaii and ϕ ≃ ϕan ⊥ 2m × H for some m ∈ N.
Corollary 4.9. Let Ψ be a K-algebra with involution. For any a ∈ K× \ K×2 such that
ΨK( √a) is hyperbolic, we have that DKhhaii ⊆ G(Ψ).
Proof. As DKhhaii = G((a)K), the statement follows immediately from Theorem 4.7.
The following was already observed in [7, p. 91, Examples (b) and (c)].
Proposition 4.10. Let Q1 and Q2 be K-quaternion algebras. The K-algebra with involution
(Q1, canQ1 ) ⊗ (Q2, canQ2 ) is hyperbolic if and only if one of Q1 and Q2 is split.
Proof. Let (A, σ) = (Q1, canQ1 )⊗ (Q2, canQ2 ). If one of the factors is split, it is hyperbolic,
and thus (A, σ) is hyperbolic. Assume now that (A, σ) is hyperbolic. Then by Proposi-
tion 4.1 there exists f ∈ Skew(σ) with f 2 = 1. We identify Q1 and Q2 with K-subalgebras
of A that commute with each other elementwise and such that σQi = canQi for i = 1, 2.
Then Skew(σ) = Q′1⊕ Q′2 where Q′i is the K-subspace of pure quaternions of Qi for i = 1, 2.
Writing f = f1 + f2 with fi ∈ Q′i for i = 1, 2, we obtain that 1 = f 2 = f 2
2 + 2 f1 f2. As
1 , f 2
f 2
2 ∈ K, we conclude that f1 f2 ∈ K. This is only possible if f1 f2 = 0, that is, if either
f1 = 0 or f2 = 0. If, say, f2 = 0, then f = f1, which then is a hyperbolic element with
respect to σ contained in Q1, whereby Q1 is split. Hence, one of Q1 and Q2 is split.
1 + f 2
Theorem 4.11 (Karpenko, Tignol). Let Ψ be a K-algebra with involution such that Ψ ⊗ Ψ
is split. If Ψ is not hyperbolic, then there exists a field extension L/K such that ΨL is not
hyperbolic and, either ΨL is split, or Ψ is symplectic and ΨL is Brauer equivalent to an
L-quaternion algebra.
Proof. See [6, (1.1)] for the orthogonal case and [19, (A.1) and (A.2)] for the other cases.
a K-algebra with involution of the first kind.
Note that the condition in Theorem 4.11 that Ψ ⊗ Ψ be split is trivially satisfied if Ψ is
We mention separately the following special case of Theorem 4.11, which was obtained
earlier by more classical methods. It will be used in Theorem 7.4.
Theorem 4.12 (Dejaiffe, Parimala, Sridharan, Suresh). Let a, b ∈ K× and let L be the
function field of the conic aX2 + bY 2 = 1 over K. Let Ψ be a K-algebra with orthogonal
involution such that Ψ ∼ (a, b)K. Then Ψ is hyperbolic if and only if ΨL is hyperbolic.
Proof. If the conic aX2 + bY 2 = 1 is split over K, then L is a rational function field over
K and the statement is obvious. Otherwise Φ = (a ·· b)K is a K-quaternion division alge-
bra with involution and Ψ is adjoint to a skew-hermitian form over Φ, in which case the
statement follows alternatively from [3] or [14, (3.3)].
5. Algebras with involution over real closed fields
Let Ψ be a K-algebra with involution. For n > 1 we set n × Ψ = Ad(n × h1i) ⊗ Ψ.
Proposition 5.1. Assume that K is pythagorean and that Ψ is orthogonal and such that
Ψ ∼ (−1, −1)K. Then Ψ ≃ Ad(ϕ) ⊗ (−1 · −1)K for a form ϕ over K. Moreover, 2 × Ψ is
hyperbolic.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
12
Proof. Let Q = (−1, −1)K. We may identify Ψ with (EndQ(V), σ) where V is a finite-
dimensional right Q-vector space and σ is the involution adjoint to a regular skew-hermitian
form h : V × V → Q with respect to canQ. Since K is pythagorean, any maximal subfield of
Q is K-isomorphic to K( √
−1). We fix a pure quaternion i ∈ Q with i2 = −1 and obtain that
any invertible pure quaternion in Q is conjugate to an element of iK×. This yields that h has
a diagonalization with entries in iK×. It follows that ih : V × V → Q is a hermitian form
with respect to the involution τ = Int(i)◦ canQ and has a diagonalization with entries in K×.
This yields that Ψ ≃ Ad(ϕ) ⊗ (Q, τ) for a form ϕ over K. Moreover, (Q, τ) ≃ (−1 · −1)K.
This shows the first claim.
As Adh1, 1i ≃ (−1 · 1)K we obtain using Proposition 3.9 that
2 × (Q, τ) ≃ (−1 · 1)K ⊗ (−1 · −1)K ≃ (−1 ·· −1)K ⊗ (−1 ·· 1)K .
By Proposition 4.1 this K-algebra with involution is hyperbolic, and thus so is 2 × Ψ.
Theorem 5.2. Assume that K is real closed.
(a) If Ψ is split orthogonal, then Ψ ≃ Ad(r × h1i ⊥ η) for a hyperbolic form η over K and
r ∈ N such that sign(Tr(Ψ)) = r2.
(b) If Ψ is non-split orthogonal, then Ψ ≃ r × (−1 · −1)K for a positive integer r, the form
Tr(Ψ) is hyperbolic, and Ψ is hyperbolic if and only if r is even.
(c) If Ψ is split symplectic, then Ψ ≃ r × (1 ·· 1)K for a positive integer r, and both Ψ and
Tr(Ψ) are hyperbolic.
(d) If Ψ is non-split symplectic, then Ψ ≃ Ad(r × h1i ⊥ η) ⊗ (−1 ·· − 1)K for a hyperbolic
form η over K and r ∈ N such that sign(Tr(Ψ)) = 4r2.
(e) If Ψ is unitary and Ψ is simple, then Ψ ≃ Ad(r×h1i ⊥ η)⊗ (−1)K for a hyperbolic form
η over K and r ∈ N such that sign(Tr(Ψ)) = 2r2.
( f ) If Ψ is unitary of inner type, then Ψ ≃ r × (1)K for a positive integer, and both Ψ and
Tr(Ψ) are hyperbolic.
These cases are mutually exclusive and cover all possibilities, and the integer r is unique
in each case.
Proof. It is clear that exactly one of the conditions in (a) -- ( f ) is satisfied. As K is real
closed, the only finite-dimensional K-division algebras are K, K( √
−1), and (−1, −1)K. We
set
Note that type(Φ) = type(Ψ) and either Φ and Ψ are both unitary of inner type or the
algebras Φ and Ψ are both simple with the same centre and Brauer equivalent.
Suppose that Ψ is split-symplectic or unitary of inner type. Then Ψ is hyperbolic by
Corollary 4.2. By Proposition 4.3 it follows that Tr(Ψ) is hyperbolic and that Ψ ≃ r × Φ for
some r ∈ N. This shows (c) and ( f ).
Next, suppose that Ψ is non-split orthogonal. Then by Proposition 5.1 we have that
Ψ ≃ Ad(ϕ) ⊗ Φ for a form ϕ over K, and as G (−1 · −1)K = K×2 ∪ −K×2 = K× we may
choose ϕ to be r × h1i for some r ∈ N. We thus have Ψ ≃ r × Φ with r ∈ N such that
deg(Ψ) = 2r. With this follows from Proposition 5.1 that Ψ is hyperbolic if and only if r is
Φ =
(K, idK)
(−1 · −1)K
(1 ·· 1)K
(−1 ·· − 1)K
(−1)K
(1)K
if Ψ is split orthogonal,
if Ψ is non-split orthogonal,
if Ψ is split symplectic,
if Ψ is non-split symplectic,
if Ψ is unitary and Ψ is simple,
if Ψ is unitary of inner type.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
13
even. Furthermore, Tr(Φ) is hyperbolic by Example 3.11, and thus so is Tr(Ψ) ≃ r2 × Tr(Φ).
This shows (b).
In each of the remaining cases (a), (d), and (e), by Proposition 3.4 we have that Ψ ≃
Ad(ϕ) ⊗ Φ for a form ϕ over K. Since K is real closed and ϕ is determined up to a scalar
factor, we choose ϕ to be r × h1i ⊥ η for some r ∈ N and a hyperbolic form η over K. It
further follows that Tr(Ψ) ≃ ϕ⊗ϕ⊗Tr(Φ) by Proposition 3.12 and Proposition 3.13 and thus
sign(Tr(Ψ)) = r2 · sign(Tr(Φ)). As in either case Tr(Φ) is positive definite by Example 3.11,
we have that sign(Tr(Φ)) = dimK (Φ). This establishes the cases (a), (d), and (e).
Finally, note that in each case the non-negative integer r is determined by deg(Ψ) or by
dim(Tr(Ψ)), respectively.
Corollary 5.3. Assume K is real closed. Then Tr(Ψ) is hyperbolic if and only if 2 × Ψ is
hyperbolic, if and only if either Ψ is hyperbolic or Ψ ≃ r × (−1 · −1)K where r ∈ N is odd.
Proof. We shall refer to the cases in Theorem 5.2. In each of the cases (b), (c), or ( f ), both
Tr(Ψ) and 2 × Ψ are hyperbolic. Assume that we are in one of the cases (a), (d), or (e), and
let r be the integer occurring in the statement for that case. Then Tr(Ψ) is hyperbolic if and
only if r = 0, if and only if Ψ is hyperbolic.
Corollary 5.4. Let P be an ordering of K and Ψ a K-algebra with involution. Then
signP(Tr(Ψ)) = [Z(Ψ) : K] · s2 for some s ∈ N.
Proof. By Theorem 5.2 the statement holds in the case where K is real closed and P is the
unique ordering of K. The general case follows immediately by extending scalars to the
real closure of K at P.
Let P be an ordering of K. The integer s occurring in Corollary 5.4 is called the signa-
ture of Ψ at P and denoted signP(Ψ). With k = [Z(Ψ) : K] we thus have
signP(Ψ) = q 1
k signP(Tr(Ψ)) .
Proposition 5.5. Let Ψ and Θ be two K-algebras with involution. If Ψ and Θ are both
unitary, assume that they have the same centre. For every ordering P of K we have that
signP(Ψ ⊗ Θ) = signP(Ψ) · signP(Θ).
Proof. This follows immediately from Proposition 3.12.
Proposition 5.6. Let ϕ be a quadratic form over K. For every ordering P of K we have
that signP(Ad(ϕ)) = signP(ϕ).
Proof. This is clear as Tr(Ad(ϕ)) ≃ ϕ ⊗ ϕ by Proposition 3.13.
6. Local-global principle for weak hyperbolicity
Let Ψ be a K-algebra with involution. We call Ψ weakly hyperbolic if there exists a
positive integer n such that n × Ψ is hyperbolic. We say that Ψ has trivial signature and
write sign(Ψ) = 0 to indicate that signP(Ψ) = 0 for every P ∈ XK .
Lemma 6.1. Assume that there exists a ∈ K× \ ±K×2 such that ΨK( √a) and ΨK( √−a) are
hyperbolic. Then 2 × Ψ is hyperbolic.
Proof. Let a ∈ K×\±K×2 be such that ΨK( √a) and ΨK( √−a) are hyperbolic. By Corollary 4.9
a and −a both belong to G(Ψ). As G(Ψ) is a group, we conclude that −1 ∈ G(Ψ), so
2 × Ψ ≃ Adhh−1ii ⊗ Ψ is hyperbolic by Proposition 4.4.
Proposition 6.2. Assume that K is nonreal and let n ∈ N be such that −1 is a sum of 2n
squares in K. Then 2n+1 × Ψ is hyperbolic.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
14
Proof. By the assumption, the Pfister form π = 2n+1 × h1i over K is isotropic, whereby it is
hyperbolic. Hence, 2n+1 × Ψ ≃ Ad(π) ⊗ Ψ is hyperbolic.
Lemma 6.3. Assume that 2n × Ψ is not hyperbolic for any n ∈ N, and that for every proper
finite extension L/K there exists n ∈ N such that 2n × ΨL is hyperbolic. Then K is real
closed and sign(Ψ) , 0.
Proof. By Proposition 6.2 the field K is real, by Lemma 6.1 its only quadratic field ex-
tension is K( √
−1), and by Theorem 4.5 K has no proper finite field extension of odd
degree. Thus K is real closed by [17, Chap. 3, (2.3)]. It follows from Theorem 5.2 that
Ψ ≃ Ad(ϕ) ⊗ Φ for a form ϕ over K and a K-division algebra with canonical involution Φ.
As K is real closed, it follows with Example 3.11 that Tr(Φ) is positive definite, and thus
sign(Ψ) is equal to sign(ϕ) or to 2 · sign(ϕ). As Ψ is not hyperbolic, ϕ is not hyperbolic,
and we conclude that sign(Ψ) , 0.
Lemma 6.4. Assume that Ψ is split and let r ∈ N. Then Ψ⊗2r is hyperbolic if and only if
Tr(Ψ)⊗r is hyperbolic.
Proof. Replacing Ψ by Ψ⊗r we may in view of Proposition 3.12 assume that r = 1. If Ψ
is symplectic then Ψ and Tr(Ψ) are hyperbolic by Corollary 4.2 and Proposition 4.3. If Ψ
is orthogonal, then Ψ⊗2 ≃ Ad(Tr(Ψ)) by Proposition 3.15, implying the statement. Assume
now that Ψ is unitary. Then Ψ ≃ Ad(ϕ) ⊗ (a)K for a form ϕ over K and some a ∈ K×.
We obtain that Ψ⊗2 ≃ Ad(ϕ ⊗ ϕ) ⊗ (a)K and Tr(Ψ) ≃ ϕ ⊗ ϕ ⊗ hhaii. Using Theorem 4.6 we
conclude that Ψ⊗2 is hyperbolic if and only if Tr(Ψ) is hyperbolic.
Theorem 6.5. The following are equivalent:
(i) sign(Ψ) = 0;
(ii) Ψ is weakly hyperbolic;
(iii) 2n × Ψ is hyperbolic for some n ∈ N;
(iv) either Ψ⊗m is hyperbolic for some m > 1, or K is nonreal and Ψ is split orthogonal
of odd degree.
These conditions are trivially satisfied if K is nonreal.
Proof. Trivially (iii) implies (ii), and by Proposition 5.5 any of the conditions implies (i).
Suppose that 2n × Ψ is not hyperbolic for any n ∈ N. By Zorn's Lemma there exists a
maximal algebraic extension L/K such that 2n × ΨL is not hyperbolic for any n ∈ N. By
Lemma 6.3 L is real closed and ΨL has nonzero signature at the unique ordering of L. For
the ordering P = L2 ∩ K of K we obtain that signP(Ψ) , 0. This shows that (i) implies (iii).
To finish the proof we show that (i) implies (iv). We may assume that Ψ is simple
as otherwise Ψ is hyperbolic. We may choose a positive integer e such that Ψ⊗e is split.
Note that if Ψ⊗e is orthogonal of odd degree, then so is Ψ and in particular Ψ is split. In
view of Proposition 5.5, we may now replace Ψ by Ψ⊗e and assume that Ψ is split. From the
assumption (i) we obtain that sign(Tr(Ψ)) = 0. Note further that dimK (Ψ) = dimK (Tr(Ψ)). If
dim(Tr(Ψ)) is odd, we conclude that K is nonreal and Ψ is split orthogonal of odd degree. If
dim(Tr(Ψ)) is even, we obtain by Corollary 2.2 that Tr(Ψ)⊗r is hyperbolic for some positive
integer r, and then Ψ⊗2r is hyperbolic by Lemma 6.4.
The equivalence (ii ⇔ iii) in Theorem 6.5 is equivalent to Scharlau's result [18, Theo-
rem 5.1] that the torsion of the Witt group of hermitian forms over an algebra with involu-
tion is 2-primary. The equivalence (ii ⇔ iv) was observed in [20, Proposition 5.41].
Corollary 6.6. Assume that Z(Ψ) ≃ K( √a) with a ∈ PK2 in case Ψ is unitary, and
otherwise that ΨR ∼ (−1, −type(Ψ))R for every real closure R of K. Then Ψ is weakly
hyperbolic.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
15
Proof. In view of the hypothesis, we obtain from Theorem 5.2 that Tr(Ψ) becomes hyper-
bolic over every real closure of K. Therefore sign(Ψ) = 0, and it follows by Theorem 6.5
that Ψ is weakly hyperbolic.
Let T be a preordering of K. We say that a K-algebra with involution Ψ is T -hyperbolic
if there exists a T -positive quadratic form τ over K such that Ad(τ) ⊗ Ψ is hyperbolic. It
is clear from Proposition 3.2 that a quadratic form ϕ over K is T -hyperbolic if and only if
Ad(ϕ) is T -hyperbolic.
Theorem 6.7. Let T be a preordering of K. We have signP(Ψ) = 0 for every P ∈ XT (K) if
and only if Ψ is T -hyperbolic. Moreover, in this case there exists a T -positive Pfister form
ϑ over K such that Ad(ϑ) ⊗ Ψ is hyperbolic.
Proof. Assume first that Ψ is T -hyperbolic. Let ϑ be a T -positive form over K such that
Ad(ϑ) ⊗ Ψ is hyperbolic. Using Proposition 5.5 it follows that signP(ϑ) · signP(Ψ) = 0 for
any ordering P of K. For P ∈ XT we have signP(ϑ) > 0 as ϑ is T -positive, and we conclude
that signP(Ψ) = 0.
Assume now that signP(Ψ) = 0 for every P ∈ XT . Then ϑ ⊗ Tr(Ψ) is hyperbolic
for some T -positive Pfister form ϑ over K, by Theorem 2.1. By Proposition 3.12 and
Proposition 3.13 we have Tr(Ad(ϑ) ⊗ Ψ) ≃ ϑ ⊗ ϑ ⊗ Tr(Ψ). We conclude that Ad(ϑ) ⊗ Ψ
has trivial total signature. By Theorem 6.5 there exists n ∈ N such that 2n × Ad(ϑ) ⊗ Ψ is
hyperbolic. Hence, the isomorphic K-algebra with involution Ad(2n × ϑ)⊗ Ψ is hyperbolic.
As 2n × ϑ is a T -positive Pfister form, this shows the statement.
7. Bounds on the torsion order
Let Ψ be a K-algebra with involution. By Theorem 6.5, if Ψ⊗n is hyperbolic for some
n ∈ N, we have that 2m × Ψ is hyperbolic for some m ∈ N. In this situation, one may want
to bound m in terms of n and the degree of Ψ. We restrict to the case n = 2, that is, where
Ψ⊗2 is hyperbolic, and use the function ∆ : N → N introduced in Section 2 to bound m.
Theorem 7.1. Assume that Ψ⊗2 is split hyperbolic. Let m = deg(Ψ) if σ is orthogonal or
unitary, and m = 1
2 deg(Ψ) if σ is symplectic. Then 2∆(m) × Ψ is hyperbolic.
Proof. In view of Theorem 4.11 it suffices to consider the situation where Ψ is either split
orthogonal, or split unitary, or symplectic of index 2. Then by Proposition 3.4 we have
that Ψ ≃ Ad(ϕ) ⊗ Φ for a form ϕ over K with dim(ϕ) = m and a K-algebra with canonical
involution Φ. As Ψ⊗2 is hyperbolic, it follows from Proposition 3.15 if Ψ is orthogonal
or symplectic and from Lemma 6.4 if Ψ is unitary that Tr(Ψ) is hyperbolic. By Propo-
sition 3.12 and Proposition 3.13 we have Tr(Ψ) ≃ ϕ ⊗ ϕ ⊗ Tr(Φ). Hence, Theorem 2.8
yields that the form (2∆(m) × ϕ) ⊗ Tr(Φ) is hyperbolic. We conclude using Theorem 4.6 that
2∆(m) × Ψ ≃ Ad(2∆(m) × ϕ) ⊗ Φ is hyperbolic.
Theorem 7.2. Assume that Ψ is a K-quaternion algebra and let m ∈ N be such that 2m×Ψ⊗2
is hyperbolic. Then 2m+1 × Ψ is hyperbolic. Furthermore, if Ψ is symplectic, then 2m × Ψ is
hyperbolic.
Proof. Assume that Ψ is split and orthogonal or unitary. Then either Ψ ≃ Adhhaii for some
a ∈ K× or Ψ ≃ Adhhaii⊗ (b)K for some a, b ∈ K×. Either way, as hha, aii ≃ 2×hhaii it follows
that Ψ⊗2 ≃ 2 × Ψ, whereby 2m × Ψ⊗2 ≃ 2m+1 × Ψ and the statement follows. We derive the
implication claimed in general for Ψ orthogonal or unitary by reduction to the split case by
means of Theorem 4.11.
Assume now that Ψ is symplectic. If Ψ is split then it is hyperbolic and so is 2m × Ψ.
Assume that Ψ is not split. Then we have Ψ⊗2 ≃ Ad(Tr(Ψ)) by Proposition 3.15 and hence
2m × Ψ⊗2 ≃ Ad(2m × Tr(Ψ)) by Proposition 3.2. Hence, if 2m × Ψ⊗2 is hyperbolic, then
2m × Tr(Ψ) is hyperbolic, and it follows by Theorem 4.6 that 2m × Ψ is hyperbolic.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
16
The following example shows that the converse in Theorem 7.2 does not hold in gen-
eral.
K ≃ Ad(2m × hha, −tii) is anisotropic, whereas 2m+1 × (a · t)K is hyperbolic.
Example 7.3. Let m ∈ N. Assume that K is either k(t) or k((t)) for some field k. Let
a ∈ Dk(2m+1) \ Dk(2m). Then the form 2m × hha, −tii over K is anisotropic. Therefore
2m × (a · t)⊗2
Theorem 7.4. Let a, b ∈ K×. The K-algebra with involution 2 × (a · b)K is hyperbolic if
and only if a ∈ DKh1, 1i ∪ DKh1, bi. For n ∈ N with n > 2 we have that 2n × (a · b)K is
hyperbolic if and only if a = x(y + b) with x ∈ DK (2n − 1) and y ∈ DK (2n − 1) ∪ {0}.
Proof. Note that 2 × (a · b)K ≃ (−1 · 1)K ⊗ (a · b)K ≃ (−1 ·· a)K ⊗ (a ·· − b)K by
Proposition 3.9. Hence, by Proposition 4.10, 2 × (a · b)K is hyperbolic if and only if one of
(−1, a)K and (a, −b)K is split, which happens if and only if a ∈ DKh1, 1i ∪ DKh1, bi.
Let n > 2. Let L denote the function field of the conic aX2 + bY 2 = 1 over K. Note that
(a, b)L is split and thus 2n × (a · b)L ≃ Ad(2n × hhaiiL). Using Theorem 4.12 we obtain that
2n × (a · b)K is hyperbolic if and only if 2n × (a · b)L is hyperbolic, which is the case if and
only if 2n × hhaiiL is hyperbolic. Using [8, Chap. X, (4.28)] we conclude that this happens
if and only if h1, −a, −bi is a subform of 2n × hhaii over K. (Note that, since n > 2 both
conditions hold in particular when 2n × hhaii is hyperbolic; for n = 1 this would fail.) This
is the case if and only if (2n − 1) × hhaii ⊥ hbi is isotropic. Finally, this occurs if and only if
a = x(y + b) for some x ∈ DK (2n − 1) and y ∈ DK(2n − 1) ∪ {0}.
Question 7.5. If K is pythagorean and sign(Ψ) = 0, is 2 × Ψ then necessarily hyperbolic?
Acknowledgements
This work was supported by the Deutsche Forschungsgemeinschaft (project Quadratic
Forms and Invariants, BE 2614/3), by the Zukunftskolleg, Universitat Konstanz, by the
Odysseus Programme funded by the Fonds Wetenschappelijk Onderzoek -- Vlaanderen
(project Explicit Methods in Quadratic Form Theory), and by the Science Foundation Ire-
land Research Frontiers Programme (project no. 07/RFP/MATF191).
References
References
[1] E. Bayer-Fluckiger, D.B. Shapiro, J.-P. Tignol. Hyperbolic involutions. Math. Z. 214
(1993), 461 -- 476.
[2] B. Bernauer. Ein Lokal-Global-Prinzip fur Involutionen und hermitesche Formen.
Diplomarbeit, Universitat Konstanz, 2004.
Online: http://kops.ub.uni-konstanz.de/handle/urn:nbn:de:bsz:352-opus-13143
[3] I. Dejaiffe. Formes antihermitiennes devenant hyperboliques sur un corps de
d´eploiement. C. R. Acad. Sci., Paris, S´er. I, Math. 332 (2001), 105 -- 108.
[4] R.L. Graham, D.E. Knuth, O. Patashnik. Concrete mathematics. A foundation for
computer science. Addison-Wesley Publishing Company, Advanced Book Program,
Reading, MA, 1989.
[5] N. Jacobson. A note on hermitian forms. Bull. Amer. Math. Soc. 46 (1940), 264 -- 268.
[6] N.A. Karpenko. Hyperbolicity of orthogonal involutions. Documenta Math., Extra
Volume Suslin (2010), 371 -- 389.
[7] M.-A. Knus, A.S. Merkurjev, M. Rost, J.-P. Tignol, The Book of Involutions, Coll.
Pub. 44, Amer. Math. Soc., Providence, RI, 1998.
K.J. Becher, T. Unger: Weakly hyperbolic involutions
17
[8] T.Y. Lam, Introduction to quadratic forms over fields, Graduate Studies in Mathemat-
ics 67, American Mathematical Society, Providence, RI, 2005.
[9] T.Y. Lam. Orderings, valuations and quadratic forms. CBMS Regional Conf. Ser.
Math., 52, published by AMS, 1983.
[10] D.W. Lewis. Witt rings as integral rings. Invent. Math. 90 (1987), 631 -- 633.
[11] D.W. Lewis. New proofs of the structure theorems for Witt rings. Expo. Math. 7
(1989), 83 -- 88.
[12] D.W. Lewis, J.-P. Tignol, On the signature of an involution, Arch. Math. 60 (1993),
128 -- 135.
[13] D.W. Lewis, T. Unger. A local-global principle for algebras with involution and her-
mitian forms. Math. Z. 244 (2003), 469 -- 477.
[14] R. Parimala, R. Sridharan, V. Suresh. Hermitian analogue of a theorem of Springer. J.
Algebra 243 (2001), 780 -- 789.
[15] A. Pfister. Quadratische Formen in beliebigen Korpern. Invent. Math. 1 (1966), 116 --
132.
[16] A. Qu´eguiner. Signature des involutions de deuxi`eme esp`ece. Arch. Math. 65 (1995),
408 -- 412.
[17] W. Scharlau. Quadratic and Hermitian forms. Grundlehren 270, Springer, Berlin,
1985.
[18] W. Scharlau. Induction theorems and the structure of the Witt group. Invent. Math. 4
(1970), 37 -- 44.
[19] J.-P. Tignol. Hyperbolicity of symplectic and unitary involutions. Appendix to a paper
of N. Karpenko. Documenta Math., Extra Volume Suslin (2010), 389 -- 392.
[20] T. Unger, Quadratic Forms and Central Simple Algebras with Involution, Ph.D. The-
sis, National University of Ireland, Dublin, 2000.
|
1304.1922 | 1 | 1304 | 2013-04-06T18:50:36 | On the simplicity of Lie algebra of Leavitt path algebra | [
"math.RA"
] | For a field $F$ and a row-finite directed graph $\Gamma$ let $L(\Gamma)$ be the Leavitt path algebra. We find necessary and sufficient conditions for the Lie algebra $[L(\Gamma),L(\Gamma)]$ to be simple. | math.RA | math |
ON THE SIMPLICITY OF LIE ALGEBRA OF LEAVITT PATH
ALGEBRA
ADEL ALAHMEDI AND HAMED ALSULAMI
Abstract. For a field F and a row-finite directed graph Γ let L(Γ) be the
Leavitt path algebra. We find necessary and sufficient conditions for the Lie
algebra [L(Γ), L(Γ)] to be simple.
1. Introduction.
In [3] G. Abrams and Z. Mesyan found necessary and sufficient conditions for a
simple Leavitt path algebra L(Γ) to give rise to a simple Lie algebra [L(Γ), L(Γ)].
This result is based on a simple easily checkable criterion for a linear combination
of vertices X
αivi, αi ∈ F, vi ∈ V, to lie in [L(Γ), L(Γ)]. In this paper we extend
i
the result of G. Abrams and Z. Mesyan to not necessarily simple algebras and find
the necessary and sufficient conditions for a Lie algebra [L(Γ), L(Γ)] to be simple.
2. Definitions and Terminology
A (directed) graph Γ = (V, E, s, r) consists of two sets V and E that are respec-
tively called vertices and edges, and two maps s, r : E → V .The vertices s(e) and
r(e) are referred to as the source and the range of the edge e, respectively. The
graph is called row-finite if for all vertices v ∈ V ,card(s−1(v)) < ∞. A vertex v
for which s−1(v) = ∅ is called a sink. A vertex v such that r−1(v) = ∅ is called a
source. A path p = e1.....en in a graph Γ is a sequence of edges e1.....en such that
r(ei) = s(ei+1) for i = 1, ..., n − 1. In this case we say that the path p starts at the
vertex s(e1) and ends at the vertex r(en). If s(e1) = r(en), then the path is closed.
If p = e1.....en is a closed path and the vertices s(e1), ...., s(en) are distinct, then
the subgraph ( s(e1), ..., s(en); e1, ..., en) of the graph Γ is called a cycle. A cycle of
length 1 is called a loop.
Definition 1. Let W be a subset of V . We say that
• W is hereditary if v ∈ W implies w ∈ W for every vertex w connects to v.
• W is saturated if {r(e) : s(e) = v} ⊆ W implies that v ∈ W, for every
non-sink vertex v ∈ V.
Definition 2. We call an edge e ∈ E a fiber if s(e) is source, r(e) is sink and
E(V, r(e)) = {e}.
Key words and phrases. Leavitt Path Algebra. Cuntz-Krieger C*-Algebras. Simple Lie
Algebra.
1
2
ADEL ALAHMEDI AND HAMED ALSULAMI
Definition 3. We call a vertex v in a connected graph Γ(V, E) a balloon over
a nonempty subset W of V if (i) v /∈ W, (ii) there is a loop C ∈ E(v, v), (iii)
E(v, W ) 6= ∅, (iv) E(v, V ) = {C} ∪ E(v, W ), and (v) E(V, v) = {C}.
Let Γ be a row-finite graph and let F be a field. The Leavitt path F -algebra
L(Γ) is the F -algebra presented by the set of generators {vv ∈ V }, {e, e∗ e ∈ E}
and the set of relators (1) vivj = δvi,vj vi for all vi, vj ∈ V ; (2) s(e)e = er(e) = e,
r(e)e∗ = e∗s(e) = e∗ for all e ∈ E; (3) e∗f = δe,f r(e), for all e, f ∈ E; (4)
v = X
ee∗, for an arbitrary vertex v which is not a sink. The mapping which
s(e)=v
sends v to v for v ∈ V, e to e∗ and e∗ to e for e ∈ E, extends to an involution of
the algebra L(Γ). If p = e1.....en is a path, then p∗ = e∗
1. In what follows we
consider only row-finite directed graphs. We call a graph Γ simple if the Leavitt
path algebra L(Γ) is simple. The conditions for a graph to be simple are given in
[1].
n....e∗
Let A be an associative F −algebra. For elements a, b ∈ A, let [a, b] = ab − ba be
their the commutator. Then A(−) = (A, [, ]) is a Lie algebra. If A is an associative
algebra and S is a subset of A, we will denote the ideal of A generated by S as
idA(S).
3. Lie algebra of Leavitt path algebra
We start with theorem by G. Abrams and Z. Mesyan in [3].
Theorem 1. ([3]) Let Γ(V, E) be a directed graph. Let L(Γ) be a simple algebra.
(i) If V is infinite then the Lie algebra [L(Γ), L(Γ)] is simple;
(ii) If V is finite, then [L(Γ), L(Γ)] is simple if and only if 1L(Γ) = X
v /∈
v∈V
[L(Γ), L(Γ)].
There exist however non-simple Leavitt path algebras having the Lie algebra
[L(Γ), L(Γ)] simple.
. The Lie algebra [L(Γ), L(Γ)] is isomorphic
Example 1. Let Γ =
to the Lie algebra of infinite finitary matrices over the Leavitt algebra L(2) and
therefore is simple.
The following theorem gives a classification of directed graph having [L(Γ), L(Γ)]
simple.
Theorem 2. Let Γ(V, E) be a directed row-finite graph. The Lie algebra [L(Γ), L(Γ)]
is simple if and only if either L(Γ) is simple- this case is covered by Theorem1 - or
Γ contains a simple subgraph W such that every point v ∈ V \ W is a balloon over
W, and X
w ∈ [L(W ), L(W )].
w∈r(E(v,W ))
We will prove the theorem by proving a series of lemmas. The first lemma is due
to G. Abrams and Z. Mesyan, [3]. We will state it without proof.
ON THE SIMPLICITY OF LIE ALGEBRA OF LEAVITT PATH ALGEBRA
3
Lemma 1. ([3]) Let Γ(V, E) be a directed graph. Then [L(Γ), L(Γ)] = (0) if and
only if Γ is a disjoint union of
,
( vertices and loops).
Lemma 2. Let Γ(V, E) be a row-finite graph. If the Lie algebra [L(Γ), L(Γ)] is
nonzero simple, then every cycle has an exit.
Proof. Let C be a no exist cycle of Γ of length d. Then L(C) ∼= Md(F [t, t−1]). Let a
be the sum of all vertices on the cycle C. The element a is the identity of L(C) and
L(C) = aL(Γ)a. Consider the ideal Jn = (1 − t)nF [t, t−1] of F [t, t−1]. Now, if d ≥ 2,
then [Md(Jn), Md(Jn)] 6= (0) for all n ≥ 1, see [5]. Let In = idL(Γ)(Md(Jn)). Then
[In, In]⊳[L(Γ), L(Γ)] and because of simplicity of [L(Γ), L(Γ)] we have [L(Γ), L(Γ)] =
[In, In] ⊆ In. Hence [L(Γ), L(Γ)] ∩ L(C) ⊆ In ∩ L(C) = Md(Jn). Since ∩nJn = (0),
it follows that [L(Γ), L(Γ)] ∩ L(C) = (0), but (0) 6= [Md(F [t, t−1]), Md(F [t, t−1])] ⊆
[L(Γ), L(Γ)]∩L(C). A contradiction. Hence d = 1. Thus C is a loop. Since C has no
exit and can not be isolated there exist an edge e ∈ E, such that s(e) /∈ V (C) = {v}.
Let Jn = (v − C)nL(C), In = idL(Γ)(Jn), vInv ⊆ Jn. Now, [eJn, Jn] = eJn 6= (0).
Hence [In, In] 6= (0), [L(Γ), L(Γ)] = [In, In] ⊆ In and therefore v[L(Γ), L(Γ)]v ⊆ Jn.
Since ∩nJn = (0) it follows that v[L(Γ), L(Γ)]v = (0), but [e∗, e] = v − ee∗, and
v[e∗, e]v = v 6= 0. A contradiction.
(cid:3)
The algebra L(Γ) is graded: dg(v) = 0, dg(e) = 1, dg(e∗) = −1 for all v ∈ V, e ∈
E. In [9] it is shown that every graded ideal I of L(Γ) is generated (as an ideal
) by I ∩ V. Thus there is a one-to-one correspondence between graded ideals and
hereditary saturated subsets of V.
Lemma 3. Let Γ(V, E) be a row-finite graph. Let W be nonempty hereditary and
saturated subset of V. Let I = idL(Γ)(W ). If [L(Γ), L(Γ)] is nonzero simple, then
[I, I] 6= (0).
Proof. If [I, I] = (0), then, in particular, [L(W ), L(W )] = (0), W is a disjoint
, and
. This implies that for every vertex w ∈ W there exist an
union of
edge e ∈ E such that r(e) = w, s(e) /∈ W otherwise w is isolated in Γ. Now, e, e∗ ∈ I
and [e, e∗] 6= 0. Lemma is proved.
(cid:3)
Lemma 4. Let Γ(V, E) be a a row-finite graph. If [L(Γ), L(Γ)] is nonzero simple,
then there exists a minimal hereditary saturated subset in V.
Proof. We need to show that the intersection of all nonzero graded ideals in L(Γ) is
nonzero. If I is a nonzero graded ideal of L(Γ) then by Lemma 3 [I, I] 6= (0). Since
[L(Γ), L(Γ)] is simple, then [L(Γ), L(Γ)] = [I, I] and therefore [L(Γ), L(Γ)] lies in
the intersection of all nonzero graded ideals of L(Γ).
(cid:3)
Let Γ(V, E) be a a row-finite graph. Suppose [L(Γ), L(Γ)] is nonzero simple. Let
W be a minimal hereditary saturated subset in V. Let I = idL(Γ)(W ), Γ′ = (V \
W, E \ E(V, W )). We assume that W 6= V, that is L(Γ) is not simple. Since L(Γ′) ∼=
L(Γ)/I and [L(Γ), L(Γ)] ⊆ I it follows that [L(Γ′), L(Γ′)] = (0). By Lemma 1 Γ′ is
a disjoint union of
, and
.
4
ADEL ALAHMEDI AND HAMED ALSULAMI
Lemma 5. Γ′ does not have components
.
Proof. Let a vertex v ∈ V \ W be isolated in Γ′. Then E(V \ W, v) = ∅. Since W is
hereditary and v /∈ W we conclude that E(V, v) = ∅. Since v can not be isolated in
Γ it can not be a sink, E(v, V ) 6= ∅. But E(v, V \ W ) = ∅, hence all descendants of
v lie in W. Since W is saturated we conclude that v ∈ W, a contradiction.
(cid:3)
Lemma 6. Every vertex v ∈ V \ W is a balloon over W.
Proof. By what we have shown Γ′ is a disjoint union of loops
c
v
. It is easly to
c
see that E(V, v) = {c} and E(v, V \ W ) = {c}. If E(v, W ) = ∅ then the loop
is isolated in Γ. Hence E(v, W ) 6= ∅. Thus v is a balloon over W.
v
(cid:3)
Let S0 be the span of all elements pp∗, where p is a path on Γ including pathes
of length zero(that is vertices). Let S1 be the span of all elements pq∗, where p, q
are pathes on Γ, r(p) = r(q), p 6= q. It follows from the description of a Groebner
- Shirshov basis of L(Γ) [4] that L(Γ) = S0 + S1 is a direct sum of vector spaces.
Let M be the semigroup generated by V ∪ E ∪ E ∗. It is easily to see that (i)
M = (M ∩ S0) ∪ (M ∩ S1), (ii) for arbitrary elements a, b ∈ M if 0 6= ab ∈ Si, then
ba ∈ Si or ba = 0, for i = 0, 1.
Lemma 7. [I, I] ∩ S0 = span{[p, p∗] p is a path on Γ , r(p) ∈ W }.
1 and p2q∗
2, 0 6= p1q∗
1 p2q∗
Proof. The ideal I is spanned by elements pq∗; p, q are paths, r(p) = r(q) ∈ W.
Consider two such elements p1q∗
1p2 6= 0
it follows that p2 = q1u or q1 = p2u, where u is a path on Γ. Consider the
first case, p2 = q1u. Then p1q∗
2. Since this element lies in S0 we
conclude that q2 = p1u. Now, p2q∗
1 = (q1u)(q1u)∗ and there-
fore [p1q∗
2] = (p1u)(p1u)∗ − (q1u)(q1u)∗ = [p1u, (p1u)∗] − [q1u, (q1u)∗]. Re-
member that r(u) = r(q2) ∈ W. Let q1 = p2u. Then p1q∗
2 =
p1(q2u)∗. Again p1q∗
2 ∈ S0 implies p1 = q2u. Now, p2q∗
2 =
(p2u)(p2u)∗. Therefore, [p1q∗
2] = (q2u)(q2u)∗ − (p2u)(p2u)∗ = [q2u, (q2u)∗] −
1 , p2q∗
[p2u, (p2u)∗] and r(u) = r(p1) ∈ W.
(cid:3)
2 = p1u∗p∗
1 = p2q∗
2p2q∗
2q2uu∗p∗
2 ∈ S0. Since q∗
2p1q∗
1 = q1uu∗p∗
1p2q∗
2 = p1uq∗
1p2q∗
2p1q∗
1, p2q∗
1 p2q∗
1p1q∗
Let v ∈ V \ W, E(v, W ) = {e1, . . . , en}, r(ei) = wi for 1 ≤ i ≤ n. Let w =
n
X
i=1
wi.
Lemma 8. w ∈ [L(W ), L(W )].
Proof. Since v is a balloon over W, let c be the loop from E(v, v), we have v =
n
n
n
n
X
i=1
eie∗
i . Hence c∗c−cc∗ = v−(v−
n
eie∗
i ) =
X
i=1
X
i=1
eie∗
i =
[ei, e∗
i ]+
X
i=1
e∗
i ei =
X
i=1
n
cc∗+
n
X
i=1
[ei, e∗
i ]+ w. Thus w = c∗c− cc∗ −
X
[ei, e∗
i ] = [c∗, c]−
i=1
[ei, e∗
i ] ∈ [L(Γ), L(Γ)] =
[I, I]. Hence w ∈ [I, I] ∩ S0. By Lemma 7 w = X
αi[pi, p∗
i ], αi ∈ F, r(pi) ∈ W.
We will distinguish between pathes that start with an edge from E(V \ W, W ) and
paths that lie entirely on W, w = X
e,ie∗]+X β[q, q∗], where e runs over
αe,i[epe,i, p∗
i
i
n
X
i=1
ON THE SIMPLICITY OF LIE ALGEBRA OF LEAVITT PATH ALGEBRA
5
E(V \W, W ), αe,i ∈ F, pe,i and q are paths on W. We have, w = X
αe,i(epe,ip∗
e,ie∗−
i
r(pe,i)) + X β[q, q∗]. Fix e ∈ E(V \ W, W ). From the description of the basis of
L(Γ) in [4] it follows that X
e,i = 0.
e,ie∗ = 0 and therefore X
αe,iepe,ip∗
αe,ipe,ip∗
Now X
αe,i(epe,ip∗
e,ie∗ − r(pe,i)) = X
αe,i[pe,i, p∗
e,i] ∈ [L(W ), L(W )].
i
i
i
i
Hence w = X αe,i[pe,i, p∗
e,i] + X β[q, q∗] ∈ [L(W ), L(W )].
(cid:3)
We proved Theorem 2 in one direction.
4. Simplicity of the Lie algebra of Leavitt path algebra
Let Γ(V, E) be a graph. Suppose that W $ V is a simple subgraph, every vertex
w lies in [L(W ), L(W )]. We will
v ∈ V \ W is a balloon over W and X
show that the algebra [L(Γ), L(Γ)] is simple. As above, denote I = idL(Γ)(W ). The
following lemma was proved in [5].
w∈r(E(v,W ))
Lemma 9. I is a simple algebra.
Lemma 10. Let A be an arbitrary simple algebra with two orthogonal idempotents
e1, e2. Then A = [A, A] + eiAei, i = 1, 2.
Proof. We have A = Ae1A. For arbitrary elements a, b ∈ A, ae1b = [a, e1b] +
e1ba. Similarly, A = Ae2A. For arbitrary elements a, b ∈ A, we have e1ae2b =
[e1ae2, e2b] + e2be1ae2. We proved that A = [A, A] + e2Ae2. The equality A =
[A, A] + e1Ae1 is proved similarly.
(cid:3)
Lemma 11. [L(Γ), L(Γ)] = [I, I] .
Proof. We have L(Γ) = I + span{cn
V \ W }. Let w ∈ W. Then, by Lemma 10, I = [I, I] + wIw.
Hence [cn
remains to show that [cn
Then
v , [I, I] + wIw] = [cn
v , I] = [cn
v , (c∗
v)n, I] ⊆ [I, I]. It
v)m] ∈ [I, I]. Let c = cv. Suppose at first that m > n.
v , [I, I]] ⊆ [I, I]. Similarly, [(c∗
v n ≥ 0, v ∈ V \ W } + span{(c∗
v)nn ≥ 1, v ∈
[cn, (c∗)m] = cn(c∗)m − (c∗)m−n
= cn−1(cc∗)(c∗)m−1 − (c∗)m−n
= cn−1(v − X eie∗
= (cn−1(c∗)m−1 − (c∗)m−n) − cn−1 X eie∗
i )(c∗)m−1 − (c∗)m−n
i (c∗)m−1.
The first summand cn−1(c∗)m−1 − (c∗)m−n = [cn−1, (c∗)m−1] and we can apply the
induction assumption. Furthermore,
cn−1eie∗
since e∗
i (c∗)m−1 = [cn−1ei, e∗
i (c∗)m−1cn−1ei = e∗
i (c∗)m−nei = 0. Now, let n > m. Then
i (c∗)m−1cn−1ei = [cn−1ei, e∗
i (c∗)m−1] + e∗
i (c∗)m−1],
[cn, (c∗)m] = cn(c∗)m − cn−m
= cn−1(v − X eie∗
= [cn−1, (c∗)m−1] − cn−1 X eie∗
i )(c∗)m−1 − cn−m
i (c∗)m−1.
6
ADEL ALAHMEDI AND HAMED ALSULAMI
As above,
cn−1eie∗
i (c∗)m−1 = [cn−1ei, e∗
= [cn−1ei, e∗
i (c∗)m−1] + e∗
i (c∗)m−1] + e∗
i (c∗)m−1cn−1ei
i cn−mei = [cn−1ei, e∗
i (c∗)m−1].
Finally, let n = m. As above we conclude that
[cn, (c∗)n] = [cn−1, (c∗)n−1] − cn−1 X eie∗
i (c∗)n−1 = X[cn−1ei, e∗
i (c∗)n−1] + X e∗
i (c∗)n−1,
i ei ∈ [I, I] by our assumption.
X cn−1eie∗
Lemma is proved.
(cid:3)
Lemma 12. The algebra [I, I] has zero center.
Proof. I. Herstein [7] proved that in a simple associative algebra A of dimension
bigger than 4 over its center, [A, A] generates A. Hence an elements from I, that
commutes with [I, I], lies in the center of I. An arbitrary element from I looks
as z = a0 + X
eae,f f ∗, a0, ae, be, ae,f ∈
ee∗ + X
b∗
eae + X
e∈E(vi,W )
e∈E(vi,W )
e∈E(vi ,W )
f ∈E(vj ,W )
L(W ). Suppose that z lies in the center of I. Commuting z with idempotents
w ∈ W, ee∗, e ∈ E(vi, W ) we see that z = a0 + X
eaee∗. This implies that a0
e∈E(vi,W )
lies in the center of W. Therefore by [2], W < ∞ and a0 = α X
w, α ∈ F.
Multiplying z on the left by e∗ and on the right by e, e ∈ E(vi, W ), we get
r(e)z = ae = αr(e). We proved that z = α( X
ee∗). Now choose
w + X
w∈W
a vertex vi ∈ V \ W and an edge f ∈ E(vi, W ). We have zcif = 0, whereas
cif z = αcif. Hence α = 0. Lemma is proved.
w∈W
e∈E(vi,W )
(cid:3)
Now it remans to refer to Herstein's theorem about simplicity of [I, I]/center,
see [8]. Hence [I, I] is simple and therefore [L(Γ), L(Γ)] is simple.
Acknowledgement
The authors would like to thank professor Efim Zelmanov for his constant ad-
vise and valuable help during the preparation of this work. The authors would
also like to express their appreciation to professor S. K. Jain for carefully reading
the manuscript and for offering his comments. This paper was funded by King
Abdulaziz University, under grant No. (7-130/1433 HiCi). The authors, therefore,
acknowledge technical and financial support of KAU.
References
1. G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005),
319-334
2. G. Aranda Pino, K. Crow The center of a Leavitt path algebra, Rev. Mat. Iberoamericana
Volume 27, Number 2 (2011), 621-644
3. G. Abrams, Z. Mesyan, Simple Lie algebra arising from Leavitt path algebra, Journal of pure
and applied algebra, 216(2012), 2303-2313.
4. A. Alahmadi, H. Alsulami, S.K. Jain, E. Zelmanov, Leavitt path algebras of finite Gelfand
Kirillov dimension, Journal of Algebra and Its Applications, 171(2012)
ON THE SIMPLICITY OF LIE ALGEBRA OF LEAVITT PATH ALGEBRA
7
5. A. Alahmadi, H. Alsulami, Simplicity of Lie algebra of skew elements of Leavitt path algebra,
submitted
6. P. Colak, Two-sided ideals in Leavitt path algebras, Journal of Algebra and Its Applications
10-5 (2011)
7. I.N.Herstein, Topics in Ring Theory, Mathematics Lecture Notes, University of Chicago,1965
8. I.N.Herstein, Rings with Involution, Mathematics Lecture Notes, University of Chicago,1976.
9. Mark Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras, Journal
of Algebra 318 (2007), 270299
Department of Mathematics, King Abdulaziz University, P.O.Box 80203, Jeddah,
21589, Saudi Arabia
Current address: Department of Mathematics, King Abdulaziz University, P.O.Box 80203,
Jeddah, 21589, Saudi Arabia
E-mail address: [email protected]
Department of Mathematics, King Abdulaziz University, P.O.Box 80203, Jeddah,
21589, Saudi Arabia
E-mail address: [email protected]
|
1302.2062 | 1 | 1302 | 2013-02-08T15:16:35 | Quotients of one-sided trianglated categories by rigid subcategories as module categories | [
"math.RA",
"math.RT"
] | We prove that some subquotient categories of one-sided triangulated categories are abelian. This unifies a result by Iyama-Yoshino in the case of triangulated categories and a result by Demonet-Liu in the case of exact categories. | math.RA | math |
QUOTIENTS OF ONE-SIDED TRIANGULATED
CATEGORIES BY RIGID SUBCATEGORIES AS
MODULE CATEGORIES
ZENGQIANG LIN AND YANG ZHANG
Abstract. We prove that some subquotient categories of one-
sided triangulated categories are abelian. This unifies a result by
Iyama-Yoshino in the case of triangulated categories and a result
by Demonet-Liu in the case of exact categories.
1. Introduction
Cluster tilting theory gives a way to construct abelian categories
from some triangulated categories. Let H be a hereditary algebra over
a field k, and C be the cluster category defined in [1] as the factor
category Db(modH)/τ −1Σ, where τ and Σ be the Auslander-Reiten
translation and shift functor of Db(modH) respectively. For a clus-
ter tilting object T in C, Buan, Marsh and Reiten [2] showed that
C/addτ T ∼= mod EndC(T )op. Keller and Reiten [3] generalized this re-
sult in the case of 2-Calabi-Yau triangulated categories by showing that
C/ΣT ∼= modT , where T is a cluster tilting subcategory of C. A gen-
eral framework for cluster tilting is set up by Koenig and Zhu. They [4]
showed that any quotient of a triangulated category modulo a cluster
titling subcategory carries an abelian structure . Let C be a triangu-
lated category and M be a rigid subcategory, i.e. HomC(M, ΣM) = 0.
Iyama and Yoshino [5] showed that M ∗ ΣM/ΣM ∼= modM. In par-
ticular, if M is a cluster tilting subcategory, then M ∗ ΣM = C, thus
the work generalized some former results in [2,3,4].
Recently, Cluster tilting theory is also permitted to construct abelian
categories from some exact categories. Let B be an exact category with
enough projectives and M be a cluster tilting subcategory. Demonet
and Liu [6] showed that B/M ∼= modM, which generalized the work
of Koenig and Zhu in the case of Frobenius categories.
The main aim of this article is to unify the work of Iyama-Yoshino
and Demonet-Liu, and give a framework for construct abelian cate-
gories from triangulated categories and exact categories. Our setting
is one-sided triangulated category, which is a natural generalization
2000 Mathematics Subject Classification. 18E10, 18E30.
Key words and phrases. Quotient category; rigid subcategory; right triangulated
category; left triangulated category.
Supported by the NSF of China (Grants 11126331, 11101084).
1
2
ZENGQIANG LIN AND YANG ZHANG
of triangulated category. Left and right triangulated categories were
defined by Beligiannis and Marmaridsis in [7]. For details and more
information on one-sided triangulated categories we refer to [7-9].
The paper is organized as follows. In Section 2, we review some basic
material on module categories over k-linear categories and quotient
categories etc. In Section 3, we prove that some subquotient categories
of right triangulated categories are module categories, which unifies
the Proposition 6.2 in [4] and the Theorem 3.5 in [5]. In Section 4, we
prove that some subquotient categories of left triangulated categories
are module categories, which unifies the Proposition 6.2 in [4] and the
Theorem 3.2 in [5]. And we will see that the case of right triangulated
categories and the case of left triangulated categories are not dual.
2. Preliminaries
Throughout this paper, k denotes a field. When we say that C is
a category, we always assume that C is a Hom-finite Krull-Schmidt k-
linear category. For a subcategory M of category C, we mean M is
an additive full subcategory of C which is closed under taking direct
summands. Let f : X → Y , g : Y → Z be morphisms in C, we denote
by gf the composition of f and g, and f∗ the morphism HomC(M, f ) :
HomC(M, X) → HomC(M, Y ) for any M ∈ C.
Let C be a category and X be a subcategory of C. A right X -
approximation of C in C is a map f : X → C, with X in X , such that
for all objects X ′ in X , the sequence HomC(X ′, X) →HomC(X ′, C) → 0
is exact. If for any object C ∈ C, there exists a right X -approximation
f : X → C, then X is called a contravariantly finite subcategory of
C. Dually we have the notions of left X -approximation and covariantly
finite subcategory. X is called functorially finite if X is contravariantly
finite and covariantly finite.
Let C be a category. A pseudokernel of a morphism v : V → W in
C is a morphism u : U → V such that vu = 0 and if u′ : U ′ → V
is a morphism such that vu′ = 0, there exists f : U ′ → U such that
u′ = uf . Pseudocokernels are defined dually.
Let C be a category. A C-module is a contravariant k-linear functor
F : C → Modk. Then C-modules form an abelian category ModC.
By Yoneda's lemma, representable functors HomC(−, C) are projective
objects in ModC. We denote by mod C the subcategory of ModC con-
sisting of finitely presented C-modules. One can easily check that modC
is closed under cokernels and extensions in ModC. Moreover, modC is
closed under kernels in ModC if and only if C has pseudokernels. In
this case, modC forms an abelian category (see [10]). For example, if C
is a contravariantly finite subcategory of a triangulated category, then
modC forms an abelian category.
Let C be an additive category and B be a subcategory of C. For
any two objects X, Y ∈ C, denote by B(X, Y ) the additive subgroup
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
3
of HomC(X, Y ) such that for any morphism f ∈ B(X, Y ), f factors
through some object in B. We denote by C/B the quotient category
whose objects are objects of C and whose morphisms are elements of
HomC(M, N)/B(M, N). The projection functor π : C → C/B is an
additive functor satisfying π(B) = 0, and for any additive functor F :
C → D satisfying F (B) = 0, there exists a unique additive functor
G : C/B → D such that F = Gπ. We have the following two easy and
useful facts.
Lemma 2.1. Let F : C → D be an additive functor. If F is full and
dense, and there exists a subcategory B of C such that any morphism
f : X → Y in C with F (f ) = 0 factors through some object in B, then
F induces an equivalence C/B ∼= D.
Lemma 2.2. Let A be an additive category, B and C be two subcate-
gories of A with C ⊂ B. Then there exists an equivalence of categories
(A/C)/(B/C) ∼= A/B.
Proof. Let πB : A → A/B and πC : A → A/C be the projection
functors. Note that C ⊂ B, we have πB(C) = 0, thus there exists a
unique functor F : A/C → A/B such that F πC = πB. Since πB is full
and dense, F is full and dense too.
Let f : X → Y be a morphism in A such that F (πC(f )) = 0, that
is πB(f ) = 0. Then f factors through some object in B, thus πC(f )
factors through some object in B/C. According to Lemma 2.1, we have
an equivalence of categories (A/C)/(B/C)
(cid:3)
∼
−→ A/B.
3. Subquotient categories of right triangulated
categories
Firstly, we recall some basics on right triangulated categories from
[8].
Definition 3.1. A right triangulated category is a triple (C, Σ, ✄), or
simply C, where:
(a) C is an additive category.
(b) Σ : C → C is an additive functor, called the shift functor of C.
(c) ✄ is a class of sequences of three morphisms of the form U u−→
V v−→ W w−→ ΣU, called right triangles, and satisfying the following
axioms:
(RTR0)
−→ W ′ w′
If U u−→ V v−→ W w−→ ΣU is a right triangle, and U ′ u′
−→
−→ ΣU ′ is a sequence of morphisms such that there exists a
V
g
v
v′
W w /
ΣU
h
Σf
/ W ′
w′
/ ΣU ′,
V ′ v′
commutative diagram in C
U u
f
U ′
u′
/ V ′
/
/
/
/
/
/
/
/
4
ZENGQIANG LIN AND YANG ZHANG
where f, g, h are isomorphisms, then U ′ u′
right triangle.
−→ V ′ v′
−→ W ′ w′
−→ ΣU ′ is also a
(RTR1) For any U ∈ C, the sequence 0 −→ U
1U−→ U −→ 0 is a right
triangle. And for any morphism u : U → V in C, there exists a right
triangle U u−→ V v−→ W w−→ ΣU.
(RTR2)
If U u−→ V v−→ W w−→ ΣU is a right triangle, then so is
V v−→ W w−→ ΣU
−P u
−−−→ ΣV .
(RTR3) For any two right triangles U u−→ V v−→ W w−→ ΣU and
U ′ u′
−→ ΣU ′, and any two morphisms f : U → U ′,
−→ V ′
g : V → V ′ such that gu = u′f , there exists h : W → W ′ such that the
following diagram is commutative
v′
−→ W ′ w′
U u
f
U ′
u′
/ V ′
V
g
v
v′
W w /
ΣU
h
/ W ′
Σf
w′
/ ΣU ′.
(RTR4) For any two right triangles U u−→ V v−→ W w−→ ΣU and
−→ U v′
−→ ΣU ′, there exists a commutative diagram
−→ W ′ w′
U ′ u′
v′
p
U ′
u′
/ U
U ′
u·u′
/ V
u
v
W
w
w′
q
ΣU
ΣU ′
W ′
f
V ′
g
W
Σv′·w
ΣU Σv′
/ ΣW ′,
where the second row and the third column are right triangles.
Example 3.2. A triangulated category C is a right triangulated cat-
egory, where the shift functor Σ is an equivalence. In this case, right
triangles in C are called triangles.
Example 3.3. (cf.[7,11]) Let B be an exact category which contains
enough injectives. The subcategory of injectives is denoted by I. Then
the quotient category B = B/I is a right triangulated category. For
any morphism f ∈ HomB(X, Y ), we denote its image in HomB(X, Y )
by f . Let us recall the definitions of the shift functor Σ and of the
distinguished right triangles. For any X ∈ B, there is a short exact
dX−→ CX → 0 with IX ∈ I. For any morphism
sequence 0 → X iX−→ IX
f : X → Y , we have the following commutative diagram with exact
/
/
/
/
/
✤
✤
✤
/
/
/
/
/
/
/
/
/
/
/
/
/
/
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
5
rows
0
0
/ X
f
/ Y
iX /
/ IX
dX /
CX
if
/ IY
iY
cf
dY /
/ CY
0
/ 0,
where IX , IY ∈ I. Define Σ(X) = CX and Σf = cf . We can show that
the functor Σ is well defined. For any morphism f : X → Y , we have
the following commutative diagram with exact rows
0
0
/ X
f
/ Y
iX /
/ IX
dX /
CX
if
/ Z
g
h /
/ CX
/ 0
/ 0,
where Z is the pushout of f and iX . Then X
f
−→ Y
g
−→ Z h−→ ΣX, or
equivalently X
iX (cid:19)
(cid:18) f
−−−−−→ Y ⊕IX
(g,−if )
−−−−→ Z h−→ ΣX is a distinguished right
(cid:18) f
iX (cid:19)
−−−−−→
triangle. In this case, there is a short exact sequence 0 → X
(g,−if )
−−−−→ Z → 0. And we have the following commutative diagram
Y ⊕IX
of short exact sequences
(cid:18) f
iX (cid:19)
/ Y ⊕ IX
(g,−if )
0
0
/ X
/ X
iX
(0,1)
/ IX
/ 0
/ Z
−h
dX
/ ΣX
/ 0.
So a distinguished right triangle in B give rise to a short exact sequence
g
in B. On the other hand, Let 0 → X
−→ Z → 0 be a short exact
sequence in B, then we have the following commutative diagram with
exact rows
f
−→ Y
0
0
/ X
/ X
f
iY
/ Y
iY
/ IY
g
p
0
Z
h
/ ΣX
/ 0,
where IY ∈ I, and X
Thus, a short exact sequence in B give rise to a right triangle in B.
−h−→ ΣX is a right triangle in B [11].
f
−→ Y
g
−→ Z
The following lemma can be found in [7].
Lemma 3.4. Let C be a right triangulated category, and U u−→ V v−→
W w−→ ΣU be a right triangle.
(a) v is a pseudocokernel of u, and w is a pseudocokernel of v.
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
6
ZENGQIANG LIN AND YANG ZHANG
(b) If Σ is fully faithful, then u is a pseudokernel of v, and v is a
pseudokernel of w.
Definition 3.5. Let C be a right triangulated category. A subcategory
M of C is called a rigid subcategory if HomC(M, ΣM) = 0.
Let M be a rigid subcategory of C. Denote by M ∗ ΣM the sub-
category of C consisting of all such X ∈ C with right triangles M0 →
M1 → X → ΣM0, where M0, M1 ∈ M.
Now we can state the main theorem of this section.
Theorem 3.6. Let C be a right triangulated category, and M be a rigid
subcategory of C satisfying:
(RC1) Σ is fully faithful when it is restricted to M.
f
(RC2) For any two objects M0, M1 ∈ M, if M0
−→ M1
−→ X h−→
ΣM0 is a right triangle in C, then g is a right M-approximation of X.
Then there exists an equivalence of categories M ∗ ΣM(cid:14)ΣM ∼=
g
modM.
Before prove the theorem, we prove the lemma as follow.
Lemma 3.7. Under the same assumption as in Theorem 3.6, for any
−→ X h−→ ΣM0 where M0, M1 ∈ M, there is an
right triangle M0
exact sequence in ModM
f
−→ M1
g
HomM(−, M0)
Thus, HomC(−, X)M ∈ modM.
HomM(−,f )
−−−−−−−→ HomM(−, M1)
HomC (−,g)
−−−−−−→ HomC(−, X)M → 0.
Proof. Let M0
M. For any M ∈ M, we claim that the following sequence is exact
−→ X h−→ ΣM0 be a right triangle with M0, M1 ∈
f
−→ M1
g
HomC(M, M0)
f∗−→ HomC(M, M1)
g∗−→ HomC(M, X) → 0.
(⋆)
In fact, by Lemma 3.4 (a), we have gf = 0, hence Imf∗ ⊆Kerg∗. For
any t ∈Kerg∗, we have the following commutative diagram of right
triangles by (RTR3)
M
t
M1
0
ΣM
−Σ1M /
ΣM
m′
Σt
g
/ X h /
/ ΣM0
−Σf
/ ΣM1.
Since ΣM is full, there exists a morphism m : M → M0 such that m′ =
Σm, so Σt = Σ(f m). Since ΣM is faithful, t = f m = f∗(m) ∈Imf∗,
then Imf∗ ⊇Kerg∗. Hence Imf∗=Kerg∗. On the other hand, by (RC2),
g∗ is surjective. So (⋆) is exact. Since M is arbitrary in M, there exists
an exact sequence
HomM(−, M0)
HomM(−,f )
−−−−−−−→ HomM(−, M1)
HomC(−,g)
−−−−−−→ HomC(−, X)M → 0.
(cid:3)
/
/
/
/
/
✤
✤
✤
/
/
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
7
Proof of Theorem 3.6. By Lemma 3.7, we have an additive functor
F : M ∗ ΣM → ModM, which is defined by F (X) = HomC(−, X)M.
Firstly, we show that F is dense.
For any object G ∈ modM, there exists an exact sequence
HomM(−, M ′) α−→ HomM(−, M ′′) → G → 0
with M ′, M ′′ ∈ M. By Yoneda's Lemma, there exists a morphism
f : M ′ → M ′′ such that α=HomM(−, f ). Then by (RTR1), there exists
a right triangle M ′ f
−→ Z h−→ ΣM ′. By Lemma 3.7, there exists
an exact sequence HomM(−, M ′) α−→ HomM(−, M ′′) → F (Z) → 0,
thus G=Cokerα ∼= F (Z). Hence F is dense.
−→ M ′′ g
Secondly, we show that F is full.
For any morphism β : F (X) → F (Y ) in mod M, because HomM(−, M1)
is a projective object in modM, we have the following commutative
diagram with exact rows in ModM
HomM(−, M0)
HomM(−,f1)
HomM(−, M1)
F (X)
/ 0
γ0
γ1
β
HomM(−, N0)
HomM(−,f2)
/ HomM(−, N1)
/ F (Y )
/ 0.
By Yoneda's Lemma, for i = 0, 1, there exists a morphism mi : Mi →
Ni such that γi=HomM(−, mi) and m1f1=f2m0. Hence by (RTR3) we
have the following commutative diagram of right triangles
f1
f2
M0
m0
N0
g1
g2
M1
m1
/ N1
X
s
/ Y
h1 /
ΣM0
Σm0
h2 /
/ ΣN0.
Then by Lemma 3.7, we have the following commutative diagram with
exact rows in ModM
HomM(−, M0)
HomM(−,f1)
HomM(−, M1)
F (X)
/ 0
γ0
γ1
F (s)
HomM(−, N0)
HomM(−,f2)
/ HomM(−, N1)
/ F (Y )
/ 0.
So β = F (s). Hence F is full.
At last, in order to show M ∗ ΣM/ΣM ∼= modM, by Lemma 2.1
we only need to prove that any morphism t : X → Y in M ∗ ΣM
satisfying F (t) = 0 factors through some object in ΣM.
In fact, let M0
g1−→ X h1−→ ΣM0 be a right triangle with
M0, M1 ∈ M, then tg1 = 0 since F (t) = 0. Thus by Lemma 3.4(a), t
factors through h1, so t factors through ΣM0 ∈ ΣM. ✷
f1−→ M1
Applying Theorem 3.6, we can get the following two corollaries.
/
/
✤
✤
✤
/
/
✤
✤
✤
/
/
/
/
/
/
/
/
/
✤
✤
✤
/
/
/
/
/
/
/
/
/
/
8
ZENGQIANG LIN AND YANG ZHANG
Corollary 3.8. ([4, Proposition 6.2]) Let C be a triangulated category
with the shift functor Σ and M be a rigid subcategory of C. Then there
exists an equivalence of categories M ∗ ΣM(cid:14)ΣM ∼= modM.
Proof. Since the shift functor Σ is an equivalence, we know that ΣM
−→ X h−→ ΣM0 be a triangle in C,
is fully faithful. Let M0
where M0, M1 ∈ M. Since M is rigid, we know that g is a right M-
approximation of X by Lemma 3.4(b). Thus, condition (RC1) and
(RC2) hold.
(cid:3)
f
−→ M1
g
Definition 3.9. Let B be an exact category and M be a full subcat-
egory of B. M is called rigid if Ext1
B(M, M) = 0.
Corollary 3.10. ([6, Theorem 3.5]) Let B be an exact category which
contains enough injectives, and M be a rigid subcategory of B contain-
ing all injectives. Denote by I the subcategory of injectives, and by M
the quotient category M/I. Denote by MR the subcategory of objects X
in B such that there exist short exact sequences 0 → M0 → M1 → X →
0, where M0, M1 ∈ M. Denote by ΣM the subcategory of objects Y in
B such that there exist short exact sequences 0 → M → I → Y → 0,
where M ∈ M, I ∈ I. Then MR(cid:14)ΣM ∼= modM.
Proof. According to Theorem 3.6, we prove the corollary by several
steps.
(a) M is a rigid subcategory of the right triangulated category B =
B/I.
Let Σ be the shift functor of B, then it is easy to see that ΣM = ΣM.
We claim that HomB(M, ΣM) = 0. In fact, for any f ∈ HomB(M, Y ),
where M ∈ M and Y ∈ ΣM. There is a short exact sequence 0 →
i−→ I d−→ Y → 0, where M ′ ∈ M, I ∈ I. Since M is rigid in B,
M ′
applying HomB(M, −) to the short exact sequence, we have an exact
sequence
0 → Hom(M, M ′) i∗−→ Hom(M, I) d∗−→ Hom(M, Y ) → 0.
So d is a right M-approximation of Y . Thus, f factors through I,
hence f = 0.
(b) MR = M ∗ ΣM.
It follows from Example 3.3.
(c) MR(cid:14)ΣM ∼= MR(cid:14)ΣM.
It follows from Lemma 2.2 since I ⊂ ΣM ⊂ MR and ΣM = ΣM.
(d) ΣM is fully faithful.
For any M ′, M ′′ ∈ M, there exist two short exact sequences 0 →
dM ′′
−−→ ΣM ′′ → 0,
dM ′
−−→ ΣM ′ → 0 and 0 → M ′′
M ′
where IM ′, IM ′′ ∈ I, and dM ′,dM ′′ are right M-approximations.
iM ′
−−→ IM ′
iM ′′
−−→ IM ′′
For any morphism α : ΣM ′ → ΣM ′′ in B, since dM ′′ is a right M-
approximation and IM ′ ∈ I ⊂ M , we have the following commutative
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
9
diagram with exact rows in B
0
0
/ M ′
iM ′
/ IM ′
dM ′
ΣM ′
m
j
α
/ M ′′
iM ′′
/ IM ′′
dM ′′
/ ΣM ′′
0
/ 0.
Hence we have α = Σm by the definition of Σ, thus ΣM is full.
For any morphism f : M ′ → M ′′ in B, Since IM ′ is an injective object,
we have the following commutative diagram of short exact sequences
0
0
/ M ′
f
iM ′
/ IM ′
dM ′
ΣM ′
if
Σf
/ M ′′
iM ′′
/ IM ′′
dM ′′
/ ΣM ′′
0
/ 0.
is right M-approximation, Σf factors through IM ′′, i.e.
Suppose Σf = 0, then Σf factors through some object in I. Because
there
dM ′′
exists a morphism a : ΣM ′ → IM ′′ such that Σf = dM ′′a. Then
dM ′′(if − adM ′) = dM ′′if − (Σf )dM ′ = 0, thus there exists a morphism
b : IM ′ → M ′′ such that iM ′′b = if − adM ′, so iM ′′(f − biM ′) = iM ′′f −
if iM ′ + adM ′iM ′ = 0. Since iM ′′ is a monomorphism, f = biM ′, thus f
factors through IM ′. Hence f = 0 and ΣM is faithful.
(e) Let M ′
f
−→ M ′′
g
−→ X h−→ ΣM ′ be a right triangle in B with
M ′, M ′′ ∈ M, then g is a right M-approximation of X.
According to Example 3.3 and I ⊂ M, we can assume that there is a
short exact sequence 0 → M ′ f
−→ X → 0. Since M is rigid, there
exists an epimorphism HomB(M, g) : HomB(M, M ′′) →HomB(M, X)
for any M in M. Thus we have an epimorphism HomB(M, g) : HomB(M,
M ′′) → HomB(M, X), i.e. g is a right M-approximation of X.
(cid:3)
−→ M ′′ g
4. Subquotient categories of left triangulated
categories
The definition of left triangulated category is dual to right trian-
gulated category. For convenience, we recall the definition and some
facts.
Definition 4.1. ([7]) A left triangulated category is a triple (C, Ω, ✁),
or simply C, where:
y
(a) C is an additive category.
(b) Ω : C → C is an additive functor, called the shift functor of C.
(c) ✁ is a class of sequences of three morphisms of the form ΩZ x−→
−→ Y z−→ Z, called left triangles, and satisfying the following axioms:
−→ Y z−→ Z is a left triangle, and ΩZ ′ x′
If ΩZ x−→ X
(LTR0)
−→
z′
−→ Z ′ is a sequence of morphisms such that there exists a
−→ Y ′
y
X
X ′ y′
/
✤
✤
✤
/
/
/
✤
✤
✤
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
10
ZENGQIANG LIN AND YANG ZHANG
commutative diagram in C
ΩZ x
Ωh
X
f
ΩZ ′
x′
/ X ′
y
y′
Y
g
/ Y ′
z
z′
Z
h
/ Z ′,
where f, g, h are isomorphisms, then ΩZ ′ x′
left triangle.
−→ X ′ y′
−→ Y ′ z′
−→ Z ′ is also a
(LTR1) For any X ∈ C, the sequence 0 −→ X 1X−→ X −→ 0 is a left
triangle. And for every morphism z : Y → Z in C, there exists a left
triangle ΩZ x−→ X
−→ Y z−→ Z.
y
If ΩZ x−→ X
y
−→ Y z−→ Z is a left triangle, then so is ΩY
−Ωz−−→
(LTR2)
ΩZ x−→ X
y
−→ Y .
(LTR3) For any two left triangles ΩZ x−→ X
−→ Y ′ z′
−→ Y z−→ Z and ΩZ ′ x′
−→
X ′ y′
−→ Z ′, and any two morphisms g : Y → Y ′, h : Z → Z ′ such
that hz = z′g, there exists f : X → X ′ making the following diagram
commutative
y
ΩZ x
Ωh
X
f
ΩZ ′
x′
/ X ′
y
y′
Y
g
/ Y ′
z
z′
Z
h
(LTR4) For any two left triangles ΩZ x−→ X
−→ Z z′
−→ Z ′, there exists a commutative diagram
X ′ y′
/ Z ′
−→ Y z−→ Z and ΩZ ′ x′
−→
y
ΩY ′ Ωy′
x·Ωy′
X
g
/ X ′
h
/ Y ′
v
y′
ΩZ
x
X
Y
y
z
z′·z /
Z ′
/ Z
z′
/ Z ′,
ΩZ ′
ΩZ ′
u
x′
where the third row and the second column are left triangles.
Example 4.2. A triangulated category is a left triangulated category.
Example 4.3. Let B be an exact category with enough projectives.
Denote by P the subcategory of B consisting of projectives. Then the
quotient category B = B/P is a left triangulated category.
By (LTR0) and (LTR2), we have the following easy lemma.
/
/
/
/
/
/
/
/
/
/
/
/
/
✤
✤
✤
/
/
/
/
/
/
/
/
/
/
/
/
/
/
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
11
Lemma 4.4. Let ΩZ x−→ X
ΩY Ωz−→ ΩZ x−→ X
−y
−→ Y .
y
−→ Y z−→ Z be a left triangle, then so is
Lemma 4.5. (cf. [8]) Let C be a left triangulated category. Then for
any left triangle ΩZ x−→ X
−→ Y z−→ Z and any object U of C, there
exists an exact sequence
y
· · · →HomC(U, ΩZ) x∗−→HomC(U, X)
y∗−→ HomC(U, Y ) z∗−→HomC(U, Z).
Definition 4.6. Let C be a left triangulated category. A subcategory
M of C is called a rigid subcategory if HomC(ΩM, M) = 0.
Let M be a rigid subcategory of C. Denote by ΩM ∗ M the sub-
category of objects X in C such that there exist left triangles ΩM1 →
X → M0 → M1, where M0, M1 ∈ M. Now we consider the functor
H : ΩM ∗ M → Mod M defined by H(X) = HomC(Ω(−), X)M.
Lemma 4.7. Let (C, Ω, ✁) be a left triangulated category and M be
If ΩM is fully faithful, then for any left
a rigid subcategory of C.
f
g
h−→ M1 where M0, M1 ∈ M, there is an exact
triangle ΩM1
−→ X
−→ M0
sequence in ModM
HomM(−, M0)
HomM(−,h)
−−−−−−−→ HomM(−, M1) → H(X) → 0.
Thus, H(X) ∈ modM.
h−→ M1, where M0, M1 ∈ M. Then ΩM0
f
g
Proof. For any X ∈ ΩM ∗ M, there exists a left triangle ΩM1
−→ X
−→
−g
M0
−→ M0 is
a left triangle by Lemma 4.4. Thus there exists an exact sequence by
Lemma 4.5
Ωh−→ ΩM1
f
−→ X
HomC(ΩM, ΩM0)
(Ωh)∗−−−→ HomC(ΩM, ΩM1)
f∗−→
HomC(ΩM, X) → HomC(ΩM, M0) = 0.
Since ΩM is fully faithful, we have the following commutative diagram
with exact rows
HomC(M, M0)
h∗
/ HomC(M, M1)
/ HomC(ΩM, X)
/ 0
HomC(ΩM, ΩM0)
(Ωh)∗
/ HomC(ΩM, ΩM1)
/ HomC(ΩM, X)
/ 0,
where M ∈ M and the vertical morphisms are isomorphisms. Thus we
have an exact sequence in ModC
HomM(−, M0)
HomM(−,h)
−−−−−−−→ HomM(−, M1) → H(X) → 0.
So H(X) ∈ modM.
(cid:3)
/
/
/
/
/
/
12
ZENGQIANG LIN AND YANG ZHANG
Theorem 4.8. Let C be a left triangulated category, and M be a rigid
subcategory of C satisfying:
(LC1) Ω is fully faithful when it is restricted to M.
(LC2) Let ΩM1
h−→ M1 be a left triangle, where
M0, M1 ∈ M. Let Y ∈ ΩM ∗ M and a morphism t : X → Y such that
tf = 0, then t factors through g.
g
−→ M0
f
−→ X
Then there exists an equivalence of categories ΩM∗M/M ∼= modM.
Proof. According to Lemma 4.7, we have a functor H : ΩM ∗ M →
mod M.
Firstly, we show that H is dense.
For any object G ∈ mod M, there exists an exact sequence
HomM(−, M ′) α−→ HomM(−, M ′′) → G → 0
with M ′, M ′′ ∈ M. By Yoneda's Lemma, there exists a morphism
h : M ′ → M ′′ such that α=HomM(−, h). Then by (LTR1), there
exists a left triangle ΩM ′′ f
−→ M ′ h−→ M ′′. Hence by Lemma 4.7,
there exists an exact sequence
−→ Z
g
HomM(−, M ′) α−→ HomM(−, M ′′) → H(Z) → 0,
so G=Cokerα ∼= H(Z). Hence H is dense.
Secondly, we show that H is full.
For any morphism β : H(X) → H(Y ) in modM. By Lemma 4.7
and because HomM(−, M1) is a projective object of modM, we have
the following commutative diagram with exact rows in ModM
HomM(−, M0)
HomM(−,h1)
HomM(−, M1)
H(X)
/ 0
γ0
γ1
β
HomM(−, N0)
HomM(−,h2)
/ HomM(−, N1)
/ H(Y )
/ 0.
By Yoneda's Lemma, for i = 0, 1, there exists a morphism mi : Mi →
Ni such that γi=HomM(−, mi) and m1h1=h2m0. Hence by (LTR3),
we have the following commutative diagram of left triangles
f1
ΩM1
Ωm1
X
s
g1 /
M0
m0
ΩN1
f2
/ Y
g2
/ N0
h1
h2
M1
m1
/ N1.
/
/
/
/
/
/
/
/
/
/
/
✤
✤
✤
/
/
/
/
/
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
13
According to the proof of Lemma 4.7, for any object M ∈ M, we have
the following commutative diagram with exact columns.
HomC (M, M0)
/ HomC(ΩM, ΩM0)
m0∗
v♥♥♥♥♥♥♥♥♥♥♥♥
HomC(M, N0)
h1∗
(Ωm0)∗
u❧❧❧❧❧❧❧❧❧❧❧❧❧
HomC (ΩM, ΩN0)
(Ωh1)∗
h2∗
HomC (M, M1)
(Ωh2)∗
HomC(ΩM, ΩM1)
m1∗
v♥♥♥♥♥♥♥♥♥♥♥♥
HomC(M, N1)
(Ωm1)∗
u❧❧❧❧❧❧❧❧❧❧❧❧❧
HomC (ΩM, ΩN1)
HomC(ΩM, X)
HomC(ΩM, X)
s∗
v♥♥♥♥♥♥♥♥♥♥♥♥
HomC(ΩM, Y )
s∗
u❧❧❧❧❧❧❧❧❧❧❧❧❧
HomC(ΩM, Y )
0
0
0
0.
Thus we have the following commutative diagram with exact rows in
Mod M
HomM(−, M0)
HomM(−,h1)
HomM(−, M1)
H(X)
/ 0
γ0
γ1
H(s)
HomM(−, N0)
HomM(−,h2)
/ HomM(−, N1)
/ H(Y )
/ 0.
So β = H(s). Hence H is full.
f
−→ X
At last, let X, Y be objects of ΩM ∗ M. We have a left triangle
h−→ M1, where M0, M1 ∈ M. Let t : X → Y be a
ΩM1
morphism with H(t) = 0, then tf = 0. Thus t factors through M0 by
(LC2). So ΩM ∗ M/M ∼= modM by Lemma 2.2.
(cid:3)
g
−→ M0
Since a triangulated category is a left triangulated category such that
the shift functor is an equivalence, the conditions (LC1) and (LC2)
holds automatically. Thus we have the following corollary.
Corollary 4.9. Let C be a triangulated category with the shift functor
T and M be a rigid subcategory of C, then T −1M ∗ M/M ∼= modM.
Corollary 4.10. ([5], Theorem 3.2) Let B be an exact category which
contains enough projectives, and M be a rigid subcategory of B con-
taining all projectives. Denote by P the subcategory of projectives,
and by M the quotient category M/P. Denote by ML the subcat-
egory of objects X in B such that there exist short exact sequences
0 → X → M0 → M1 → 0, where M0, M1 ∈ M. Then ML(cid:14)M ∼=
modM.
v
/
u
/
/
v
/
/
u
/
/
v
u
/
/
/
/
/
/
/
/
14
ZENGQIANG LIN AND YANG ZHANG
f1
−→ X
In fact, let ΩM ′′
Proof. Similar to the proof of Corollary 3.10, we can prove that M
is a rigid subcategory of the left triangulated category B, and ML =
ΩM ∗ M, and ML(cid:14)M ∼= ML(cid:14)M, and ΩM is fully faithful. To end
the proof, we only need to show that M satisfies the condition (LC2).
−→ M ′ h1−→ M ′′ be a left triangle in B, where
g1−→ M ′ h1−→
M ′, M ′′ ∈ M. Since P ⊂ M, we can assume that 0 → X
M ′′ → 0 is a short exact sequence. Let t : X → Y be a morphism
satisfying tf1 = 0, where Y ∈ ML. Then there exists a short exact
g2−→ N ′ h2−→ N ′′ → 0, where N ′, N ′′ ∈ M. Since M is
sequence 0 → Y
rigid, it is easy to see that g1 is a left M-approximation, then we have
the following commutative diagram with exact rows in B
g1
0
0
/ X
t
/ Y
g1 /
/ M ′
h1
M ′′
m1
m2
g2
/ N ′
h2
/ N ′′
0
/ 0.
h2
The lower exact sequence induces a left triangle ΩN ′′
−→
N ′′. We claim that tf1 = f2Ωm2. In fact, we have the following diagram
with exact rows in B
g2
−→ N ′
f2
−→ Y
0
0
0
0
/ X
t
/ Y
/ ΩM ′′
f1
③③③③③③③③③
Ωm2
g1
iM ′′
M ′
/ PM ′′
pM
②②②②②②②②
h1
p
/ ΩN ′′
f2
}③③③③③③③③③
g2
iN ′′
m1
/ N ′
/ PN ′′
pN
③③③③③③③③
h2
dM ′′
M ′′
dN ′′
m2
/ N ′′
M ′′
③③③③③③③③
③③③③③③③③
m2
/ N ′′
③③③③③③③③
③③③③③③③③
0
0
/ 0
/ 0,
where PM ′′, PN ′′ ∈ P, and all squares are commutative except the left
one and the middle one. Since h2(m1pM −pN p) = m2dM ′′ −m2dM ′′ = 0,
there exists a morphism q : PM ′′ → Y such that g2q = m1pM − pN p.
Then g2(tf1−f2Ωm2 −qiM ′′) = (m1pM −pN p)iM ′′ −(m1pM −pN p)iM ′′ =
0. Since g2 is a monomorphism, we get tf1 − f2Ωm2 = qiM ′′. Thus
tf1 = f2Ωm2. Hence we have the following commutative diagram of
left triangles in B
f1
f2
ΩM ′′
Ωm2
ΩN ′′
/ X
t
/ Y
g1
g2
/ M ′
m1
/ N ′
h1
h2
M ′′
m2
/ N ′′.
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
}
/
/
/
/
/
/
/
/
/
/
/
/
/
/
QUOTIENTS OF ONE-SIDED TRIANGULATED CATEGORIES
15
By Lemma 4.4, we have the following commutative diagram of left
triangles in B
ΩM ′
Ωh1
/ ΩM ′′
Ωm1
Ωm2
ΩN ′
Ωh2
/ ΩN ′′
f1
f2
/ X
t
/ Y
−g1
M ′
m1
−g2
/ N ′.
Since f2Ωm2 = tf1 = 0, there exists a morphism n′ : ΩM ′′ → ΩN ′
such that Ωm2 = (Ωh2)n′. Because ΩM is fully faithful, there exists a
morphism n1 : M ′′ → N ′ such that n′ = Ωn1 and m2 = h2n1. Hence
m2 − h2n1 factors through P ∈ P. Since h2 is a epimorphism, we have
the following commutative diagram in B:
M ′′
a
}④④④④④④④④
!❈❈❈❈❈❈❈❈
b
m2−h2n1
/ N ′′.
P
h2
c
~⑥⑥⑥⑥⑥⑥⑥
N ′
Let n = ca + n1. Then m2 = h2n1 + ba = h2n1 + h2ca = h2n and
n = n1. Since h2(m1 − nh1) = h2m1 − m2h1 = 0, there exists a
morphism s : M ′ → Y such that g2s = m1 − nh1. Hence g2(t − sg1) =
g2t − m1g1 + nh1g1 = 0. Because g2 is a monomorphism, t = sg1, i.e. t
factors through g1. Hence t factors through g1 in B.
(cid:3)
References
[1] A Buan, R Marsh, M Reineke, I Reiten, G Todorov. Tilting theory and cluster
combinatorics, preprint math. RT/0402054, (2004)
[2] A B Buan, R J Marsh, I Reiten. Cluster-tilted algebras, Trans. Amer. Math.
Soc. 359 (2007), no. 1, 323-332.
[3] B Keller, I Reiten. Cluster-tilted algebras are Gorenstein and stably Calabi-
Yau, Adv. Math. 211 (2007), no. 1, 123-151.
[4] S Koenig, B Zhu. From triangulated categories to abelian categories: cluster
tilting in a general framework. Math. Z. 258 (2008), no. 1, 143-160.
[5] O Iyama, Y Yoshino. Mutation in triangulated categories and rigid Cohen-
Macaulay modules, Invent. Math. 172 (2008), no. 1, 117-168.
[6] L Demonet, Y Liu. Quotients of exact categories by cluster tilting subcategories
as module categories. arXiv: 1208.0639v1.
[7] A Beligiannis, N Marmaridis. Left triangulated categories arising from con-
travariantly finite subcategories. Communications in algebra. 22(12), 5021-
5036, 1994.
[8] I Assem, A Beligiannis, N Marmaridis. Right Triangulated Categories with
Right Semi-equivalences. Canandian Mathematical Society Conference Pro-
ceeding. Volume 24, 1998.
[9] A Beligiannis, I. Reiten. Homological and homotopical aspects of torsion the-
ories (English summary), Mem. Amer. Math. Soc. 188 (2007), no. 883.
/
/
/
/
/
/
/
}
~
!
/
16
ZENGQIANG LIN AND YANG ZHANG
[10] M Auslander. Coherent functors. 1966 Proc. Conf. Categorical Algebra (La
Jolla, Calif., 1965) pp. 189-231 Springer, New York.
[11] D Happel. Triangulated Categories in the Representation Theory of Finite
Dimensional Algebras. London Mathematical Society, LMN 119, Cambridge,
1988.
School of Mathematical sciences, Huaqiao University, Quanzhou
362021, China.
E-mail address: [email protected]
School of Mathematical sciences, Huaqiao University, Quanzhou
362021, China.
E-mail address: [email protected]
|
1311.5741 | 1 | 1311 | 2013-11-22T13:03:02 | The commutant of simple modules over almost commutative algebras | [
"math.RA"
] | Let $B$ be a finitely generated algebra over a field $k$. Then $B$ is called a Jacobson algebra if every semiprime ideal of $B$ is semiprimitive. We will discuss several conditions, all involving the commutant of simple $B$-modules, which imply that $B$ is Jacobson. In particular, we will recover the well-known result that every finitely generated almost commutative algebra is Jacobson. The same holds true for $\mathbb{N}$-filtered $k$-algebras $B$ with a locally finite filtration such that the associated graded $k$-algebra is left-noetherian. | math.RA | math |
The commutant of simple modules over
almost commutative algebras
1
Oliver Ungermann
Ernst-Moritz-Arndt-Universitat Greifswald
Institut fur Mathematik und Informatik
Friedrich-Ludwig-Jahn-Strasse 15a
17487 Greifswald, Germany
October 2009
Abstract
Let B be a finitely generated algebra over a field k. Then B is called a Jacobson
algebra if every semiprime ideal of B is semiprimitive. We will discuss several
conditions, all involving the commutant of simple B-modules, which imply that B
is Jacobson. In particular, we will recover the well-known result that every finitely
generated almost commutative algebra is Jacobson. The same holds true for N-
filtered k-algebras B with a locally finite filtration such that the associated graded
k-algebra is left-noetherian.
Introduction
Following Dixmier [3] we shall recall the proofs of the following three well-known results
which are fundamental for the representation theory of universal enveloping algebras:
Let k be a field and g a finite-dimensional Lie algebra over k. Let U (g) denote the
universal enveloping algebra of g. Then we have:
1. If M is a simple U (g)-module, then its commutant EndU (g)(M ) is algebraic over k.
(This assertion is not obvious if M is infinite-dimensional as a k-vector space.)
2. The associative k-algebra U (g) is a Jacobson algebra which means that every
prime ideal is an intersection of primitive ideals.
3. In particular U (g) is Jacobson-semisimple.
These assertions can be proved using variants of Grothendieck's generic flatness lemma.
In algebraic geometry generic flatness means the following: If A is a noetherian integral
domain, B a finitely generated commutative A−algebra, and M a finitely generated B-
module, then there exists a non-zero f ∈ A such that the localization Mf = Af ⊗A M
2
O. Ungermann
of M is free and hence flat. Duflo [7] and Quillen [12] showed that generic freeness
over k[X] is satisfied for simple k[X] ⊗k B-modules where B is a finitely generated al-
most commutative k-algebra, compare Theorem 14 and Theorem 15. This includes all
quotients B of universal enveloping algebras of finite-dimensional Lie algebras. Artin,
Small, and Zhang [2] proved that generic freeness holds for simple k[X] ⊗k B-modules
where B is a k-algebra endowed with a locally finite N-filtration such that the associ-
ated graded algebra is noetherian.
These results about generic freeness can be used to prove that algebras over com-
mutative Jacobson rings are also Jacobson. This was done by Duflo [7] in the almost
commutative case and by Szczepanski [14] in the N-filtered case.
As a motivation we give an application of the above results. Let k be an algebraically
closed field. Let P be a primitive ideal of U (g) and M a simple U (g)-module such that
P = AnnU (g)(M ), keeping in mind that this equality does not determine M up to equiv-
alence. In this case it follows EndU (g)(M ) = k·1 because EndU (g)(M ) is algebraic over k
and k is algebraically closed. Let Z(g) denote the center of U (g). Since Z(g)/(Z(g)∩P )
can be regarded as a subalgebra of EndU (g)(M ), we conclude Z(g)/(Z(g)∩P ) ∼= k. Thus
there exists a unique homomorphism χP : Z(g) −→ k such that ker χP = Z(g) ∩ P ,
which is called the central character of P . Since U (g) is semiprimitive, we know that
W ∈ Z(g) and χP (W ) = 0 for all P ∈ Prim(U (g)) implies W = 0.
Let S(g) be the symmetric algebra of g and ¯γ : S(g) −→ U (g) the Duflo map which is,
in a certain sense not to be explained here, a modification of the symmetrization map
β(X1 . . . Xn) =
1
n! X
σ∈Sn
Xσ(1) . . . Xσ(n).
The Poincar´e-Birkhoff-Witt theorem implies that β and ¯γ are ad(g)-invariant linear
isomorphisms. Let I(g) denote the subalgebra of all ad(g)-invariants of S(g). In 1969
Duflo proved that the restriction γ : I(g) −→ Z(g) of ¯γ is an isomorphism of algebras,
the so-called Duflo isomorphism. See [5] for the case of solvable and semisimple Lie
algebras, and [6] for the general case. Duflo's idea can be described as follows: In order
to prove γ(X1X2) = γ(X1)γ(X2) for all X1, X2 ∈ I(g), it suffices, in view of the above
considerations, to verify
χP (γ(X1X2)) = χP (γ(X1)) χP (γ(X2))
for all P ∈ Prim(U (g)). The latter identity is an immediate consequence of Duflo's
∗, P is the primitive ideal of U (g) associated
character formula which states: If f ∈ g
to f via the Kirillov map, and χf = χP is its central character, then
χf (γ(X)) = X(f )
for all X ∈ I(g). But the proof of Duflo's character formula is demanding and requires
a detailed knowledge of the representation theory and the primitive ideal theory of U (g).
3
The importance of the results U (g) semiprimitive and EndU (g)(M ) algebraic over k
in the context of Duflo isomorphism motivated us to elaborate on their proofs.
Jacobson algebras
Let B be a unital ring. An ideal P of B is called primitive if there exists a (non-zero)
simple B-module M with annihilator
P = AnnB(M ) = {a ∈ B : b·m = 0 for all m ∈ M} .
An ideal I of B is said to be semiprimitive if it is an intersection of primitive ones. The
Jacobson radical J(B) is defined as the intersection of all primitive ideals of B.
We shall recall some facts about simple modules. If L is a maximal left ideal of B, then
M = B/L is a simple B-module whose annihilator is given by
AnnB(M ) = {a ∈ B : a·B ⊂ L} .
Conversely, if M is a simple B-module and ξ ∈ M is non-zero, then L = {a ∈ B :
a·ξ = 0} is a maximal left ideal and M ∼= B/L in a canonical way. Since maximal left
ideals exist by Zorn's Lemma, the existence of simple modules and primitive ideals is
guaranteed.
It is known that maximal ideals are primitive: Let B be a ring and I a maximal
ideal of B. By Zorn's Lemma there exists a maximal left ideal L of B with I ⊂ L. Now
M = B/L is a simple B-module with I ⊂ AnnB(M ). Since I is maximal, we see that
I = AnnB(M ) is primitive.
Lemma 1. If B is commutative, then every primitive ideal of B is maximal.
Proof. Let I be a primitive ideal of B, I = AnnB(M ) for some simple B-module M .
Let ξ ∈ M be non-zero. Then L = {a ∈ B : a·ξ = 0} is a maximal (left) ideal of B.
Clearly I ⊂ L. If a ∈ I, then it follows a·M = a·(B·ξ) = B·(a·ξ) = 0. Thus I = L is
maximal.
We refrain from giving the various characterizations of the Jacobson radical and content
ourselves with the following observation.
Lemma 2. If a ∈ J(B), then 1 + xa is left invertible for all x ∈ B.
Proof. Suppose that 1+xa is not left invertible. Then B(1+xa) 6= B. By Zorn's Lemma
there exists a maximal left ideal L with 1 + xa ∈ L and 1 6∈ L. Since B/L is a simple
B-module and a ∈ J(B), it follows a ∈ aB ⊂ L. This implies 1 = (1 + xa) − xa ∈ L, a
contradiction.
4
O. Ungermann
Next we shall introduce the Baer radical of B. Let I be an ideal of B with I 6= B. We
say that I is prime if the following condition is satisfied: If J1 and J2 are ideals of B
with J1J2 ⊂ I, then J1 ⊂ I or J2 ⊂ I. The radical √I of I is defined as the intersection
of all prime ideals containing I. Further we say that I is semiprime if I = √I. By
Zorn's Lemma there exist maximal ideals containing I. And since maximal ideals are
prime, this implies that √I 6= B is well-defined. Clearly I ⊂ √I and √I = p√I. This
shows that √I is the smallest semiprime ideal of B containing I. Finally the Baer
radical √0 is defined to be the intersection of all prime ideals of B. In the literature
√0 is also known as the prime radical, the lower nilradical or the Baer-McCoy radical
of B.
It is easy to see that primitive ideals are prime: Let M be a simple module such
that I = AnnB(M ). Let J1, J2 be ideals of B such that J1J2 ⊂ I. If J2 6⊂ I, then
J2·M is a non-zero submodule and hence J2·M = M because M is simple. This yields
J1·M = J1J2·M = I·M = 0 and hence J1 ⊂ I. Thus I is prime. In particular we see
that the Jacobson radical contains the Baer radical: √0 ⊂ J(B).
In the proof of the next lemma we will need the following fact: If I is a semiprime
ideal of B and J is an ideal of B with J 2 ⊂ I, then it follows J ⊂ I. This can be seen
as follows: If P is prime ideal with J 2 ⊂ I ⊂ P , then it follows J ⊂ P . Intersecting all
these P gives J ⊂ √I = I.
Lemma 3. If I is a nilpotent ideal of B, then I ⊂ √0.
Proof. Suppose that I is nilpotent and I 6⊂ √0. Then there is some n ≥ 1 such
that I n 6⊂ √0 and I n+1 ⊂ √0. Let k be the smallest integer ≥ (n + 1)/2. Clearly
(I k)2 ⊂ I n+1 ⊂ √0. Since √0 is semiprime, this implies I k ⊂ √0, a contradiction.
A ring B is called (semi-)prime or (semi-)primitive if its zero ideal is (semi-)prime or
(semi-)primitive respectively. In the literature semiprimitive rings are also known as
Jacobson-semisimple rings. Clearly I is a semiprime or semiprimitive ideal of B if and
only if B/I is a semiprime or semiprimitive ring respectively.
For noetherian rings we obtain additional results.
Proposition 4. Let B be a left noetherian ring. Then the Baer radical √0 is a nilpotent
ideal of B. Thus B has a maximal nilpotent ideal.
Proof. The first step is to prove that B contains a maximal nilpotent ideal. Suppose
that this is not the case. Let J0 be an arbitrary nilpotent ideal of B. Since J0 is not
maximal, there exists a nilpotent ideal J1 of B such that J1 6⊂ J0. Clearly J0 + J1 is a
nilpotent ideal which properly contains J0. By induction we obtain a strictly increasing
sequence of nilpotent (two-sided) ideals, in contradiction to the assumption that B left
noetherian.
Let J denote the maximal nilpotent ideal of B. Clearly J ⊂ √0 by Lemma 3. Suppose
5
that J is not semiprime. Then Proposition 33.(ii) of the Appendix shows that there
exists an ideal J1 of B with J 2
1 ⊂ J and J1 6⊂ J. Since J is nilpotent, it follows that
J + J1 is also nilpotent, in contradiction to the maximality of J. Thus J is semiprime.
This shows that √0 ⊂ √J = J so that J = √0 is nilpotent.
The next proposition is crucial for the proof of Proposition 7.
Proposition 5. Let B be a left noetherian ring. Then every nil right ideal of B is
nilpotent.
Proof. Let R be a nil right ideal of B. Let J = √0 denote the Baer radical of B which
is known to be the maximal nilpotent ideal of B by Proposition 4. Consequently we
must prove R ⊂ J. Suppose that R 6⊂ J. Since B is right noetherian, there exists
some a ∈ R \ J such that the left ideal l(a) = {x ∈ B : xa ∈ J} is maximal among
all left ideals of this form. Let y ∈ B be arbitrary. Our aim ist to prove aya ∈ J. If
ay ∈ J, then we are done. So let us assume ay 6∈ J. For ay ∈ L and L is nil, there
exists some k > 1 such that (ay)k−1 6∈ J and (ay)k ∈ J. Since l(a) ⊂ l( (ay)k−1 ), it
follows l( (ay)k−1 ) = l(a) by the maximality of l(a). This implies ay ∈ l(a) and hence
aya ∈ J. Since aBa ⊂ J and J is semiprime, it follows a ∈ J. This contradiction
proves R ⊂ J.
The preceding two propositions as well as the subsequent theorem remain valid after
exchanging the words left and right.
The next result is fundamental.
Lemma 6 (Schur). If M is a simple left B-module, then the commutant
D = EndB(M ) = {ϕ : M −→ M : ϕ is B-linear}
is a division ring. In particular the center Z(D) of D is a field.
Proof. First of all, D = EndB(M ) is a (not necessarily commutative) unital ring. If
ϕ ∈ D, ϕ 6= 0, then it follows im ϕ = M and ker ϕ = 0 because im ϕ 6= 0 and ker ϕ 6= M
are submodules of the simple B-module M . This means that ϕ is bijective so that ϕ−1
exists. Clearly ϕ−1 ∈ D. This proves D to be a division ring. Obviously Z(D) is a
field.
In order to obtain information about the given ring B we will introduce an additional
variable X and consider (simple) modules M over the ring B[X] of all polynomials in
one indeterminate with coefficients in B. This striking idea can be traced back to the
famous one-sided article [13] of Rabinovich from 1929. Note that any B[X]-module
structure boils down to a B-module M together with a map ϕ(m) = X·m in its com-
mutant. Clearly M is B[X]-simple if and only if M does not admit proper ϕ-invariant
B-submodules.
Now let A be a commutative unital ring and B a unital A-algebra. One might think of
6
O. Ungermann
A as a subring of the center Z(B) of the ring B. Note that the commutant EndB(M )
of any B-module M becomes an A-algebra via (z · ϕ)(m) = z · ϕ(m). We say that
ϕ ∈ EndB(M ) is integral over A if there exist elements z0, . . . , zn−1 ∈ A such that
ϕn +Pn−1
k=0 zkϕk = 0. In the sequel we shall investigate whether the following condition
is satisfied:
(E) If M is a simple B[X]-module, then ϕ(m) = X·m is integral over A.
Let P = AnnB[X](M ) denote the primitive ideal of B[X] associated to M . Since
A∩ P = AnnA(M ) is a prime ideal of A, it follows that A/A∩ P is an integral domain.
Note that M can be regarded as an A/A∩ P -module. We point out that (E) is satisfied
if and only if ϕ(m) = X·m is integral over A/A ∩ P . This is the case if and only if
X is integral as an element of the integral domain and A/A∩ P -algebra A[X]/A[X]∩ P .
The next proposition is due to Duflo, see Th´eor`eme 1 of [7]. This result seems to
be inspired by an earlier work of Amitsur on the radical of polynomial rings, see [1].
Proposition 7. Let A be a commutative ring and B be a left noetherian A-algebra. If
B satisfies (E), then the Jacobson radical and the Baer radical of B coincide.
Proof. It suffices to show that J(B) ⊂ √0. Let a ∈ J(B) be arbitrary. First we shall
prove that B[X](1− aX) = B[X]. Suppose that this is not the case so that there exists
a maximal left ideal L of B[X] such that 1 − aX ∈ L and 1 6∈ L. Then M = B[X]/L
is a simple B[X]-module and 1 + L ∈ M is non-zero. Since X is in the center of B[X],
the map
ϕ : M −→ M, ϕ(p + L) = X·(p + L) = Xp + L,
is B[X]-linear. Clearly ϕ(1 + L) = X + L 6= 0. As M is simple, Lemma 6 implies
that ϕ is invertible. By assumption ϕ ∈ EndB[X](M ) is integral over A. Thus there
exist z0, . . . , zn−1 in A such that ϕn + Pn−1
k=0 zk ϕk = 0. Dividing by ϕn we obtain
1 + f (ϕ−1) = 0 where f = Pn−1
k=0 zkY n−k is a polynomial with coefficients in A. From
aX + L = 1 + L we deduce
ϕk(ak·(1 + L)) = akX k + L = (aX)k + L = (aX + L)k = 1 + L
for all k ≥ 0 which shows that
(1 + f (a))·(1 + L) = (1 + f (ϕ−1))(1 + L) = 0 .
On the other hand, f (a) ∈ J(B) because f has constant term zero. Hence 1 + f (a) is
left invertible in B by Lemma 2. This contradiction proves B[X](1 − aX) = B[X].
Thus we know that there exists a polynomial p = Pm
k=0 ak X k in B[X] such that
1 = p(1 − aX) = a0 +
X
k=1
(ak − ak−1a) X k + ama X m+1 .
m
7
Comparing the coefficients we find ak = ak for 0 ≤ k ≤ m and am+1 = ama = 0, which
proves a to be nilpotent. Thus J(B) is a nil ideal. As B is left-noetherian, Proposition 5
shows that J(B) is a nilpotent ideal. Finally Lemma 3 implies that J(B) is contained
in √0.
Next we recall the definition of a Jacobson algebra which goes back to Duflo [7].
The motivation is to generalize the assertion of Hilbert's Nullstellensatz to the non-
commutative setting. To this end we recall the following version of the Nullstellensatz:
Let k be an algebraically closed field and I a proper ideal of the commutative polynomial
algebra B = k[X1, . . . , Xn]. Let V(I) = { λ ∈ kn : p(λ) = 0 for all p ∈ I } be the set of
common zeros of I and r(I) the ideal of all p ∈ B such that pn ∈ I for some n ≥ 1.
The Nullstellensatz asserts
r(I) = { p ∈ B : p(λ) = 0 for all λ ∈ V(I) } .
This means that r(I) is equal to the intersection of the maximal ideals { Iλ : λ ∈ V(I)}
of B where Iλ = { p ∈ B : p(λ) = 0}. In particular it follows r(I) = √I, in accordance
with the usual notation. If k is an arbitrary field, then one might expect that √I equals
the intersection of all maximal ideals containing I, whereas these maximal ideals need
not be of the form Iλ for some λ ∈ k. Finally one might ask for a generalization of
the assertion of the Nullstellensatz to non-commutative rings B. The decisive idea
is to replace maximal ideals by primitive ones. This step is encouraged by the fact
that primitive ideals play a very prominent role in representation theory. Taking into
account that I = √I semiprime is necessary for I to be an intersection of primitive
ideals we end up with
Definition 8. A ring B is called a Jacobson ring if every semiprime ideal of B is
semiprimitive.
The next theorem is an immediate consequence of Proposition 7.
Theorem 9. Let A be a commutative ring and B a left noetherian A-algebra satisfying
condition (E). Then B is a Jacobson ring.
Proof. Let I be a semiprime ideal of B. Then ¯B = B/I is a left notherian semiprime
ring. Note that ¯B[X] is a quotient of B[X] so that every ¯B[X]-module can be regarded
as a B[X]-module. Thus condition (E) implies that ϕ = X·m is integral over A for
every simple ¯B[X]-module M . Now Proposition 7 yields J( ¯B) = √ ¯B = 0 which shows
that I is semiprimitive.
Duflo proved that finitely generated almost commutative algebras over commutative
Jacobson rings are Jacobson, see Th´eor`eme 3 of [7]. As a first step in this direction
he considers almost commutative algebras over arbitrary fields.
In this context the
following condition plays a crucial role: Let k be a field and B an arbitrary k-algebra.
(D) If M is a simple B[X]-module, then ϕ(m) = X·m is algebraic over k.
8
O. Ungermann
According to Theorem 9 this condition implies the Jacobson property. By contrast
Irving prefered the following stronger requirement.
(I) If M is a simple B[X]-module, then EndB[X](M ) is algebraic over k.
However, we do not follow Irving [10] in saying that B[X] satisfies the Nullstellensatz
if condition (I) holds true.
In [12] Quillen discovered that the generic flatness lemma can be used to prove that
the commutant of a simple module is algebraic. We shall explain this important fact
in detail.
First we recall some definitions. Let A be a commutative ring and f ∈ A not nilpotent
defining the multiplicative subset Sf = {f k : k ≥ 0} of A. Then Af = S−1
f A is called
a simple localization. Roughly speaking, Af is generated by A and f −1. Note that
Af 6= 0 because f is not nilpotent. We say that an A-module M is generically free
if there exists a non-nilpotent f ∈ A such that Mf = Af ⊗A M is a free Af -module. In
the sequel we will use the fact that the localization functor is exact which means that
Af is a flat A-module.
Returning to the original situation we suppose that B is a k-algebra and that M
is a simple B[X]-module. Since k[X] is contained in the center of B[X], M becomes a
module over the principal ideal domain A = k[X] such that the actions of A and B[X]
commute. Hence the set
TA(M ) = {m ∈ M : a·m = 0 for some a ∈ A}
of A-torsion elements is a B[X]-submodule of M .
non-zero elements m ∈ M and f ∈ A such that f·m = 0. Then
f·M = f·(B[X]·m) = B[X]·(f·m) = 0
If TA(M ) 6= 0, then there exist
which means f (ϕ) = 0. Thus ϕ(m) = X·m is algebraic over k. In order to establish
property (D) we must exclude the case TA(M ) = 0. (In this case M is flat, but not
necessarily free over A.)
According to Quillen we introduce the following condition.
(Q0) Every simple B[X]-module is generically free over k[X].
In the proof of Theorem 11 we will need the following auxiliary result.
Lemma 10. Let k be a field. Then k[X] contains infinitely many monic irreducibles.
Proof. If k is infinite, then J = {X − λ : λ ∈ k} is an infinite set of irreducibles with
leading coefficient 1. Next we assume that k is finite. Suppose that the set J of all
9
irreducible and monic polynomials in k[X] is finite. Set n = max{deg(q) : q ∈ J}
and define Q as the product of all q ∈ J. Since deg(Q) > n, it follows that 1 + Q is
reducible. Hence there exist p, q ∈ k[X] with 1 + Q = pq and p irreducible. Clearly
p Q so that there is a q′ ∈ k[X] with Q = pq′. This implies pq = 1 + Q = 1 + pq′ and
p(q − q′) = 1, a contradiction. Thus J is infinite.
The next result is due to Quillen, see [12]. Since the proof in [12] is quite succinct, we
reproduce the elaborate argument given by Dixmier in Lemme 2.6.4 of [3].
Theorem 11. Let B be a unital k-algebra. Then (Q0) implies (D).
Proof. Let M be a simple B[X]-module. As above we regard M as a module over
A = k[X]. Suppose that ϕ(m) = X·m is transcendental over k. By (Q0) there exists
a non-zero f ∈ A such that Mf = Af ⊗A M is a free Af -module.
First of all we observe that the natural map η : M −→ Mf , η(m) = 1 ⊗ m, is an
injection: Its kernel ker η = TSf (M ) = {m ∈ M : f n·m = 0 for some n ≥ 0} is trivial
because ϕ is transcendental. Thus Mf 6= 0.
By Lemma 10 we can choose an irreducible polynomial g ∈ A not dividing f . Then
Af −→ Af , p 7→ gp is not surjective because f n 6∈ gA for all n ≥ 0.
Next we define Ψ : Mf −→ Mf , Ψ(p ⊗ m) = (gp) ⊗ m = p ⊗ (g · m). On the
one hand, Mf ∼= A(I)
f has the form Ψ(p)i = gpi.
This shows that Ψ is not surjective. On the other hand, im g(ϕ) is a non-zero B[X]-
submodule of M because ϕ ∈ EndB[X](M ) is transcendental over k. As M is simple,
it follows im g(ϕ) = M . This shows im Ψ = Mf because Ψ(p ⊗ m) = p ⊗ g(ϕ)(m) and
Mf = Af ⊗A M is generated by elementary tensors. This contradiction proves (D).
6= 0 is free and Ψ : A(I)
f −→ A(I)
f
Next we will prove that (Q0) holds for all finitely generated almost commutative
algebras. Let A be a commutative ring and B a unital A-algebra. Suppose that there
exists an A-submodule g of B such that xy − yx ∈ g for all x, y ∈ g and B = P∞
n.
In particular this means that g is a Lie subalgebra of B with respect to the canonical
Lie algebra structure [x, y] = xy − yx of B. Further we assume that g is a finitely gen-
n−1 and as an
erated A-module. If x1, . . . , xd are generators of g, then g
A-module, generated by the set of all products of the form xν = xν1
d with ν ∈ Nd
and ν = Pd
j=0 νj = n. We say that (B, g) is a finitely generated almost commutative
A-algebra if the above conditions are satisfied.
n is, modulo g
1 . . . xνd
n=0 g
By means of the universal property of universal enveloping algebras we obtain the
following characterization.
Lemma 12. Let A be a commutative ring. If g is a Lie algebra over A which is finitely
generated as an A-module, then every quotient of its universal enveloping algebra U (g)
is a finitely generated almost commutative A-algebra. Conversely, if (B, g) is finitely
generated and almost commutative, then B is isomorphic to a quotient of U (g).
10
O. Ungermann
Theorem 14 generalizes the generic flatness lemma of algebraic geometry, compare
Lemma 6.7 in [9] Expos´e IV of SGA 1960-61. Grothendieck's proof of generic freeness
for modules over finitely generated commutative algebras uses the Noether normaliza-
tion Lemma and Krull dimension theory. Here we restate a more elementary proof due
to Dixmier [4] and Duflo [7] only relying on the fact that the localization functor is exact.
n=0 Mn.
A filtration of an A-module M is a sequence Mn of A-submodules of M such that
M−1 = {0}, Mn−1 ⊂ Mn for all n ≥ 0, and M = P∞
Lemma 13. Let A be a commutative ring and M a left A-module endowed with a
filtration {Mn : n ≥ −1}. Suppose that f ∈ A is non-nilpotent such that f·Mn ⊂ Mn−1
whenever Mn/Mn−1 is not free. Then Af ⊗A M is a free Af -module.
Proof. The images Ln of the canonical injections ιn : Af ⊗A Mn −→ Af ⊗A M form a
filtration of the Af -module Af ⊗A M . We claim that Ln/Ln−1 is free over Af for all
n ≥ 0: First we treat the case Mn/Mn−1 not free. Then
r ⊗ m = (rf −1) ⊗ (f·m) ∈ Ln−1
for all r ∈ Af and m ∈ Mn. This proves Ln = Ln−1 so that Ln/Ln−1 = 0. Next
we assume that Mn/Mn−1 ∼= A(I) is free. As Af ⊗A − commutes with direct sums, it
follows that Af ⊗A (Mn/Mn−1) ∼= Af ⊗A A(I) ∼= A(I)
is a free Af -module. Since Af is
flat, it follows that Ln/Ln−1 ∼= Af ⊗A (Mn/Mn−1) is free over Af for all n ≥ 0. This
shows that Af ⊗A M is free.
f
The next theorem is due to Duflo, see Th´eor`eme 2 of [7]. The idea of the proof goes
back to Dixmier [4]. Compare also Lemme 2.6.3 of Dixmier's book [3]. The proof is a
bit technical and needs some preparation. For d ≥ 1 we interpret Nd as the set of all
multi-indices. The order of ν ∈ Nd is given by ν = Pd
j=1 νj. Let ≤ denote the unique
total order on Nd satisfying the following two properties:
1. If µ, ν ∈ Nd and µ < ν, then µ ≤ ν.
2. If µ, ν ∈ Nd such that µ = ν and if there exists 1 ≤ k ≤ n such that µj = νj
for 1 ≤ j ≤ k − 1 and µk < νk, then µ ≤ ν.
As a poset (Nd,≤) is isomorphic to (N,≤) in a canonical way. By abuse of notation
we write ν − 1 for the maximum of the set of all µ ∈ Nd such that µ ≤ ν and µ 6= ν,
provided that ν 6= 0.
Theorem 14. Let A be an integral domain and (B, g) a finitely generated almost
commutative A-algebra. Then the following enhancement of (Q0) holds true: If M a
cyclic B-module, then there is a non-zero f ∈ A such that Af ⊗ M is a free Af -module.
Proof. Let M be a cyclic B-module and ξ ∈ M with M = B· ξ. Let x1, . . . , xd be
generators of the A-submodule g of B. As usual we write xν = xν1
d for ν ∈ Nd.
Since B = P∞
n = Pν∈Nd A·xν , it is evident that Mν = Pµ≤ν A·xµ·ξ defines a
1 . . . xνd
n=0 g
11
filtration of the A-module M such that Mν /Mν−1 is cyclic for all ν. Let us consider
the ideals
Iν = {a ∈ A : a·Mν ⊂ Mν−1} = {a ∈ A : a·xν·ξ ∈ Mν−1}
of A. Clearly Mν/Mν−1 ∼= A is free provided that Iν = 0. Let Λ = {ν ∈ Nd : Iν 6= 0}.
In view of Lemma 13 it now suffices to prove that the ideal J = T{Iν : ν ∈ Λ} is
non-zero. The problem is to treat the case where Λ happens to be infinite.
If ν ∈ Nd is arbitrary, ej ∈ Nd is the jth canonical basis vector, and a ∈ Iν, then
a·xν+ej · ξ = xj·a·xν· ξ + a· [xν1
1 . . . x
νj−1
j−1 , xj] x
νj
j
. . . xνd
d · ξ ∈ Mν+ej −1
which shows a ∈ Iν+ej . Thus Iν ⊂ Iν+α for all α ∈ Nd and hence Λ + Nd = Λ. Let
Λ0 denote the set of all γ ∈ Λ such that γ 6= ν + ej for all ν ∈ Λ and 1 ≤ j ≤ n. It
is easy to see that Λ0 is a finite subset of Λ such that Λ = Λ0 + Nd. Since A is an
integral domain and Iν ⊂ Iν+α, it follows that J = T{Iγ : γ ∈ Λ0} is non-zero. For
every 0 6= f ∈ J we have f·Mν ⊂ Mν−1 for all ν ∈ Λ. This completes the proof.
The preceding theorem can be generalized to the case of finitely generated B-
modules. We omit the details.
The next result appears as a corollary of the preceding achievements.
It is a gen-
eralization of a result of Krull in the commutative case, see Satz 1 and Satz 2 of [11].
Compare also Theorem 3 of Goldmann [8].
Theorem 15. Let k be a field and (B, g) a finitely generated almost commutative k-
algebra. Then B satisfies condition (D). In particular B is a Jacobson algebra and the
commutant EndB(M ) of any simple B-module is algebraic over k.
Proof. By Theorem 14 we know that B has property (Q0). Theorem 11 implies (D).
Thus B is Jacobson by Theorem 9. Since any ϕ ∈ EndB(M ) gives rise to a simple
B[X]-module, it follows by (D) that ϕ is algebraic over k.
Lemma 16. Let k be an algebraically closed field and B be a finitely generated com-
mutative k-algebra. If I is a primitive (i.e. maximal) ideal of B, then there exists a
homomorphism χ : B −→ k such that I = ker χ.
Proof. Let M be a simple B-module with I = AnnB(M ). Observe that B/I can be
regarded as a subring of EndB(M ) because B is commutative. Since k is algebraically
closed and EndB(M ) is algebraic over k by Theorem 15, it follows B/I ∼= k. Thus the
canonical projection χ : B −→ B/I ∼= k has the desired properties.
Finally, applying these results to the commutative algebra B = k[X1, . . . , Xn],
where k algebraically closed, we rediscover Hilbert's Nullstellensatz: Any semiprime
ideal I = √I of B is equal to the intersection of the maximal ideals containing it. As
the homomorphisms χ : B −→ k are known in this case, every maximal ideal is of the
form Iλ = { p ∈ B : p(λ) = 0 } for some λ ∈ kn.
12
O. Ungermann
Theorem 17. Let A be a commutative noetherian Jacobson ring and B a finitely gener-
ated almost commutative A-algebra. Then B is a Jacobson algebra and the commutant
EndB(M ) of any simple B-module is integral over A.
Proof. First of all, B is known to be left noetherian. According to Theorem 9 we
must verify (E). Let M be a simple B[X]-module with annihilator P = AnnB(M ) and
ϕ(m) = X·m. As we have already seen, it suffices to prove that ϕ is integral over the
integral domain ¯A = A/A ∩ P . For the convenience of the reader we reproduce the
proof of Proposition 1 of [7] which states that ¯A is a field.
By Theorem 14 there exists a non-zero f ∈ ¯A such that Mf = ¯Af ⊗ ¯A M is a free
¯Af -module. Suppose that the natural map η : M −→ Mf , η(m) = 1 ⊗ m, has a non-
trivial kernel ker η = TSf (M ) 6= 0. Then there exists a k ≥ 1 and a non-zero m ∈ M
such that f k·m = 0. Since M is simple, it follows f k·M = 0. Hence f k = 0. For ¯A has
no zero divisors, we get f = 0, a contradiction. Thus η is injective. In particular Mf is
non-zero.
First we prove that ¯Af is a field. Let a ∈ ¯Af be non-zero. Clearly µa : Mf −→ Mf ,
µa(m) = a·m, is a non-zero element of EndB(M ). By Schur's Lemma µa is invertible.
Let {mi : i ∈ I} be a basis of Mf and choose i ∈ I. One can find b, bj ∈ ¯Af such that
a (mi) = b·mi + X
µ−1
j6=i
bj·mj .
Applying µa to both sides and comparing coefficients, we obtain ab = 1 so that a is
invertible in ¯Af . Consequently ¯Af is a field.
Since ¯A is a Jacobson ring, there exists a maximal ideal I of ¯A such that f 6∈ I.
As f + I is invertible in ¯A/I, there exists a unique ring homomorphism ψ : ¯Af −→ ¯A/I
such that ψ(x) = x + I for all x ∈ ¯A. It follows that ψ is injective because ¯Af is a field.
This proves I = 0. Thus ¯A itself is a field.
Since ¯B = B/P is an almost commutative ¯A-algebra, Theorem 14 implies that ϕ
is algebraic over ¯A. This proves (E).
In respect of Irving's condition (I) we state the following generic freeness property.
(Q1) If M is a simple B[X]-module and ϕ ∈ EndB[X](M ), then Y ·m = ϕ(m) defines a
generically free k[Y ]-module.
In the situation of (Q1) we can regard M as a simple module over the k-algebra
B[X, Y ] = B[X][Y ] = B[X] ⊗k k[Y ]. Applying Theorem 11 to B[X][Y ], we obtain
Lemma 18. Condition (Q1) implies (I).
Finally we add the conditions (A) the commutant EndB(M ) of any simple B-module
M is algebraic over k and (J) B is a Jacobson algebra. Altogether these conditions are
related as follows:
13
(Q1)
3 (Q0)
(I)
3 (D)
(J)
(A)
Here the implication (D)⇒(J) is valid only if B is left (or right) noetherian. The
others hold true in general.
Modules over filtered algebras
Some results of the preceding section can be generalized to filtered k-algebras such that
the associated graded algebra is noetherian. To begin with, let A be a commutative ring
and B an A-algebra. An increasing sequence F = {Bn : n ≥ −1} of A-submodules of B
is called a filtration of B if the following conditions are satisfied: B−1 = 0, A · 1 ⊂ B0,
P∞
n=0 Bn = B, and BmBn ⊂ Bm+n. We say that (B,F) is an N-filtered A-algebra.
The filtration is locally finite if Bn is a finitely generated A-module for all n ≥ 0.
Subject to a given filtration F of B we define the A-modules ¯Bn = Bn/Bn−1 and
¯B = L∞
¯Bn. It is easy to see that
¯Bm × ¯Bn −→ ¯Bm+n, (x + Bm−1)(y + Bn−1) = xy + Bm+n−1
n=0
is well-defined and turns ¯B into an N-graded A-algebra. We call ¯B = gr(B,F) the
associated graded algebra of B.
n=0 g
n, then Bn = g
If g is an A-submodule of B such that [g, g] ⊂ g and
Let B be an A-algebra.
B = P∞
n + Bn−1 yields a filtration F of B such that the asso-
ciated graded algebra gr(B,F) is commutative, i.e., such that [Bn, Bm] ⊂ Bn+m−1 for
all m, n ≥ 0. Generalizing the notion of finitely generated almost commutative algebras
given in the previous section we state
Definition 19. A filtered A-algebra (B,F) is called almost commutative if gr(B,F)
is commutative. If in addition gr(B,F) is finitely generated, then (B,F) is said to be
of finite type.
Conversely, if (B,F) is almost commutative, then B1 is a Lie subalgebra of B, but
1 is not apparent. If in addition (B,F) is of finite type, then B1 is finitely
n=0 Bn
B = P∞
generated.
Suppose that A is a noetherian commutative ring and B is an almost commutative
A-algebra of finite type. Then it follows by the Hilbert basis theorem that gr(B,F) is
+
+
+
3
14
O. Ungermann
a noetherian A-algebra.
Let (B,F) be a filtered A-algebra. For every finitely generated B-module M with
generators ξ1, . . . , ξr we consider the filtration
Mn =
r
X
j=1
Bn·ξj
of M . This means that E = {Mn : n ≥ −1} is an increasing sequence of A-submodules
of M such that M−1 = 0, P∞
n=1 Mn = M , and Bm· Mn ⊂ Mm+n. Here the action
of A on M is given by a · ξ = (a · 1)· x. Note that the definition of E depends on
the choice of the generators ξ1, . . . , ξr of M and the filtration of F of B. Now we set
¯Mn = Mn/Mn−1 and ¯M = L∞
¯Mn. One verifies easily that
n=0
¯Bm × ¯Mn −→ ¯Mm+n, (x + Bm−1)·(m + Lν−1) = x·m + Lm+n−1
turns ¯M = gr(M,E) into a finitely generated graded ¯B-module.
As from now let k be a field and B a k-algebra with filtration F = {Bn : n ≥ −1}. The
k[X]-algebra B′ = B[X] carries a natural filtration F ′ given by
B′
n = Bn[X] = {p ∈ B[X] : p(k) ∈ Bn for all k ≥ 0}
and the associated graded k[X]-algebra gr(B′,F ′) is isomorphic to ¯B[X] = gr(B,F)[X].
We remind the reader that we are interested in finding convenient conditions which
are sufficient for B to be a Jacobson algebra. To this end we introduce
(G0) Every cyclic graded ¯B[X]-module is generically free over k[X].
The next proposition can be found (more or less explicitly) in the work of Quillen [12],
Dixmier [3], and Artin-Small-Zhang [2].
Proposition 20. Let (B,F) be an N-filtered k-algebra. Then (G0) implies (Q0).
Proof. Let M be a simple B[X]-module and ξ ∈ M non-zero. As above we define a
filtration Mn = Bn[X]· ξ of M and consider the associated graded ¯B[X]-module ¯M .
Note that ¯M is cyclic over ¯B[X]. Put A = k[X]. By (G0) there exists a non-zero
f ∈ A such that
∞
¯Mf = Af ⊗A ¯M ∼=
M
n=0
Af ⊗A (Mn/Mn−1)
is free. (Here we used the fact that Af ⊗A − commutes with direct sums.) Since Af is
a principal ideal domain, it follows that the Af -submodules Af ⊗A (Mn/Mn−1) of ¯Mf
are free for all n. We know that
0 −→ Af ⊗A Mn−1
ηn−1−→ Af ⊗A Mn −→ Af ⊗A (Mn/Mn−1) −→ 0
15
is exact because Af is flat over A. Consequently the canonical images Ln of Af ⊗A Mn
in Mf = Af ⊗A M form a filtration of Mf such that
Ln/Ln−1 ∼= (Af ⊗A Mn) / ηn−1(Af ⊗A Mn−1) ∼= Af ⊗A (Mn/Mn−1)
is free over Af for all n ≥ 0. Hence it follows that Af ⊗A Mf is a free Af -module. This
proves (Q0).
In the sequel we shall restrict ourselves to filtered algebras (B,F) such that gr(B,F)
is left noetherian. As we will see next, a necessary condition for the latter is that
B itself is left noetherian: Let A be a ring and (B,F) an A-algebra with filtration
F = {Bn : n ≥ −1}. Let grn : Bn −→ Bn/Bn−1 denote the canonical maps. If L is a
left ideal of B, then
∞
gr(L) =
M
n=0
grn(L ∩ Bn)
is a left ideal of gr(B). Clearly L1 ⊂ L2 implies gr(L1) ⊂ gr(L2).
Lemma 21. If L1 and L2 are left ideals of B such that L1 ⊂ L2 and L1 6= L2, then
gr(L1) 6= gr(L2).
Proof. Suppose that gr(L1) = gr(L2). Let n ≥ 0 be minimal with L1 ∩ Bn 6= L2 ∩ Bn.
Choose b ∈ L2 ∩ Bn such that b 6∈ L1. Since gr(L1) = gr(L2), there is a c ∈ L1 ∩ Bn
such that grn(c) = grn(b). By the minimality of n we conclude that b − c ∈ L1 ∩ Bn−1.
But this implies b = (b − c) + c ∈ L1, a contradiction.
Suppose {Ln : n ≥ 0} is a chain of left ideals of B. Then {gr(Ln) : n ≥ 0} is a
chain of left ideals of gr(B,F) which becomes stationary provided that gr(B,F) is left
noetherian. By Lemma 21 it follows that {Ln : n ≥ 0} is stationary. This proves B to
be left noetherian.
Theorem 22. Let B be a k-algebra endowed with a locally finite filtration F such
that the associated graded k-algebra ¯B = gr(B,F) is left noetherian. Then B satisfies
condition (G0). In particular B is a Jacobson algebra and EndB(M ) is algebraic over k
for every simple B-module M .
Proof. As in (G0) let ¯M be a cyclic graded ¯B[X]-module. If ξ ∈ ¯M0 is a cyclic vector,
then L = {p ∈ ¯B[X] : p· ξ = 0} is a left ideal of ¯B[X] such that ¯M ∼= ¯B[X]/L. In
particular it follows that ¯M is noetherian. Furthermore the summands of the grading
¯M =
∞
M
k=0
¯Mk =
∞
M
k=0
¯Bk[X]·ξ
are finitely generated k[X]-modules. Set A = k[X]. Obviously the set
¯T = TA( ¯M ) = { ¯m ∈ ¯M : there exists a non-zero g ∈ A such that g· ¯m = 0 }
16
O. Ungermann
j=0
of A-torsion elements of ¯M is a ¯B[X]-submodule. Since ¯M is noetherian, we know that
¯B[X]·ηj is finitely generated. As ηj ∈ ¯T , there exist 0 6= fj ∈ A such that
¯T = PN
j=0 fj 6= 0 we find that f · ¯T = 0. We consider the simple
fj ·ηj = 0. Setting f = QN
localization Af = S−1A of A by S = {f n : n ≥ 0}. Next we observe that the Af -torsion
of the localization ¯Mf = Af ⊗A ¯M of M is zero:
TAf ( ¯Mf ) = {
¯m
s
: ¯m ∈ TA( ¯M ) and s ∈ S } = 0 .
For Af is a Prufer Domain, it follows that ¯Mf is a flat Af -module. Consequently all
summands of the decomposition ¯Mf = L∞
k=0 Af ⊗A ¯Mn are flat. Further the Af ⊗A ¯Mn
are projective because they are finitely generated over Af . Moreover, they are even free
because Af is a principal ideal domain. Altogether we see that ¯Mf is a free Af -module.
This proves (G0). Since gr(B,F) and hence B are noetherian, Proposition 20 and
Theorem 11 imply that B is a Jacobson algebra satisfying (A).
We sustain our search for sufficient conditions for B to be Jacobson. The aim is
to introduce a condition (G1) on the level of associated graded algebras and modules
which is stronger than (Q1). To this end we suppose that the k-algebra B′ = B[X]
carries a filtration F ′ = {B′
n : n ≥ −1} which need not be induced by a filtration F
of B as above. However, the k[Y ]-algebra B′[Y ] is endowed with the natural filtration
F ′′ induced by F ′. In particular gr(B′[Y ],F ′′) ∼= gr(B′,F ′)[Y ].
Let M be a simple B′-module and ϕ ∈ EndB ′(M ). As usual we regard M as a B′[Y ]-
module via Y ·m = ϕ(m) and form the associated graded gr(B′F ′)[Y ]-module ¯M . Note
that ¯M is cyclic over gr(B′,F ′). As a variant of (G0) we implement
(G1) Every graded gr(B′,F ′)[Y ]-module ¯M which is cyclic over gr(B′,F ′) is generically
free over k[Y ].
Applying Proposition 20 to (B′[Y ],F ′′) we obtain
Corollary 23. Condition (G1) implies (Q1).
Remark 24. It is an interesting question whether the preceding results can be gener-
alized to filtered algebras over arbitrary Jacobson rings. In [14] it is proven that if A
is a noetherian Jacobson ring and (B,F) is an A-algebra with a locally finite filtration
such that gr(B,F) is noetherian, then B is a Jacobson algebra. The crucial step in the
proof is to show that A/A ∩ P is a field for every primitive ideal P of B.
17
Appendix A: More on filtered algebras
Let I be an ideal of an algebra B with filtration F = {Bn : n ≥ −1}. Let π : B −→ B/I
denote the canonical map. Then F = {π(Bn) : n ≥ −1} is a filtration of B/I. Further
the maps πn : Bn/Bn−1 −→ π(Bn)/π(Bn−1) define a homomorphism of gr(B,F) onto
F ) with kernel gr(I). From this we deduce that B/I is finitely generated and
gr(B/I,
almost commutative whenever B is. Further if B is an almost commutative algebra of
finite type, so is B/I.
The following observation seems to be appropriate: If (B,F) is a finitely generated,
almost commutative A-algebra, then gr(B,F) is a finitely generated commutative A-
algebra, but the converse fails. The notion of an almost commutative algebra (B,F)
of finite type is more general.
Under the additional assumption that A is a principal ideal domain, the preceding re-
sult can be generalized to the case of almost commutative A-algebras B of finite type.
The next proposition is contained implicitly in Quillen [12]. See also Lemme 2.6.4 in [3].
Let k be a field, A a commutative k-algebra, and (B,F) an A-algebra with filtration
F = {Bn : n ≥ −1}. Clearly B′ = A ⊗k B becomes an A-algebra via
(a1 ⊗ b1)(a2 ⊗ b2) = (a1a2) ⊗ (b1b2) and a·(a1 ⊗ b1) = (aa1) ⊗ b1 .
Further B′
n = A⊗k Bn defines a filtration F ′ of the A-algebra B′. Note that B′ can also
be regarded as a filtered k-algebra. Since A ⊗k − is exact and commutes with direct
sums, it follows that
B′
n / B′
n−1 = (A ⊗k Bn) / (A ⊗k Bn−1) ∼= A ⊗k (Bn / Bn−1)
are isomorphic as A-modules, and
gr(B′,F ′) =
∞
M
n=0
B′
n / B′
n−1 = A ⊗k (
∞
M
n=0
Bn / Bn−1 ) ∼= A ⊗k gr(B,F)
as A-algebras. From the last equation we deduce
1. If (B,F) is a (finitely generated) almost commutative k-algebra, then (B′,F ′) is
also a (finitely generated) almost commutative A-algebra.
2. If gr(B,F) is a finitely generated k-algebra, then gr(B′,F ′) is a finitely generated
A-algebra. If in addition A is finitely generated as a k-algebra, then gr(B′,F ′) is
also a finitely generated k-algebra.
Appendix B: Localizations
In this section we collect some results about localizations. In particular we prove that
every localization S−1A of a commutative ring A is a flat (right) A-module.
18
O. Ungermann
Let A be a commutative ring and S a multiplicative subset of A which means 1 ∈ S,
0 6∈ S, and s, t ∈ S implies st ∈ S. In this section we shall discuss the notion of a local-
ization of A with respect to S. Generalizing the definition of the field of fractions of a
integral domain we consider the following equivalence relation on S × A: (s, a) ∼ (t, b)
if and only if there exists v ∈ S such that vta = vsb. Let a
s denote the equivalence class
of (s, a), and S−1A the set of all equivalence classes. One verifies easily that
a
s
+
b
t
=
ta + sb
st
and
a
s ·
b
t
=
ab
st
give a well-defined addition and multiplication on S−1A. Further we define a map
ι : A −→ S−1A, ι(a) = a
1 . Then the localization (S−1A, ι) of A with respect to S has
the following properties:
• S−1A is a ring and ι : A −→ S−1A is a ring homomorphism.
• ι(s) is invertible for all s ∈ S.
• If B is a ring and ϕ : A −→ B a ring homomorphism such that ϕ(s) is invertible
for all s ∈ S, then there exists a unique homomorphism ¯ϕ : S−1A −→ B such
that ϕ = ¯ϕ ◦ ι.
• S−1A is generated by ι(A) and ι(S)−1.
• ker ι = {a ∈ A : as = 0 for some s ∈ S}.
Clearly the first three properties determine (S−1A, ι) up to isomorphisms. Next we
shall discuss the ideal theory of A and S−1A. If J is an ideal of A, then
ext(J) = (S−1A)·ι(J) = {
a
s
: a ∈ J and s ∈ S }
is an ideal of S−1A called the extension of J in S−1A. If I is an ideal of S−1A, then
res(I) = ι−1(I) = { a ∈ A :
is an ideal of A, the restriction of I to A. Obviously
a
1 ∈ I }
I = ext(res(I))
and J ⊂ res(ext(J)).
Note that I 6= S−1A is proper if and only if res(I) ∩ S = ∅. If I is prime, so is res(I).
If J is prime and J ∩ S = ∅, then ext(J) is also prime. In this case J = res(ext(J)).
Furthermore, if J1 ⊂ J2, then ext(J1) ⊂ ext(J2), and if I1 ⊂ I2, then res(I1) ⊂ res(I2).
Definition 25. Let A be a commutative ring. Let Spec(A) denote the set of all prime
ideals of A. If J is an ideal of A, then h(J) = {P ∈ Spec(A) : J ⊂ P} is called the hull
of J. Conversely, if X ⊂ Spec(A), then the ideal k(X) = T{P : P ∈ X} is called the
kernel of X. We say that a subset X of Spec(A) is closed if and only if X = h(J) for
some ideal J of A. The space Spec(A) endowed with this so-called hull-kernel topology
is the spectrum of A.
19
The preceding observations show that the spectrum Spec(S−1A) of the localization of
A with respect to S can be identified with the subset {P ∈ Spec(A) : P ∩ S = ∅} of
the spectrum of A by means of the homeomorphisms ext and res.
Lemma 26. If A is a principal ideal domain, so is S−1A.
Proof. Obviously S−1A is an integral domain. Let I be an ideal of S−1A. Since A is a
principal ideal ring, there exists some b ∈ A such that res(I) = A·b. This implies
I = ext(res(I)) = (S−1A) ·
b
1
.
Now let P be a left A-module. Then the functors HomA(−, P ) and HomA(P,−)
are left exact. One can prove that P ⊗A − is also right exact.
Definition 27. Let A be a ring and P a right A-module. We say that P is a flat
A-module if the functor P ⊗A − is exact.
Since P ⊗A − is already known to be right exact, it follows that P is A-flat if and
only if the following condition is satisfied: If ϕ : L −→ M is an injective homomorphism
of left A-modules, then 1 ⊗ ϕ : P ⊗A L −→ P ⊗A M is injective.
To prove that localizations S−1A are A-flat, i.e., that the functor (S−1A) ⊗A − is
exact, we introduce localizations of A-modules. Let A be a commutative ring, S a
multiplicative subset of A, and M a left A-module.
In analogy to the definition of
S−1A we consider the following equivalence relation on S × M : (s, m) ∼ (t, n) if and
only if there exists a v ∈ S such that vt·m = vs·n. Let m
s denote the equivalence class
of (s, m), and S−1M the set of all such equivalence classes. It is easy to see that
a
r ·
gives a well-defined S−1A-module structure on S−1M .
tm + sn
m
s
and
st
+
=
n
t
m
s
=
a·m
rs
Lemma 28. Let A be a commutative ring and S a multiplicative subset of A. If M is
a left A-module, then (S−1A) ⊗A M and S−1M are isomorphic as left S−1A-modules.
Proof. By the universal property of the balanced tensor product we obtain a map
Φ : (S−1A) ⊗A M −→ S−1M , Φ(
a
r ⊗ m ) =
a·m
r
.
One checks that Φ is a homomorphism of left S−1A-modules. On the other hand we
have a map
Ψ : S−1M −→ (S−1A) ⊗A M , Ψ(
m
s
) =
1
s ⊗ m .
20
O. Ungermann
We check that Ψ is well-defined: Let m, n ∈ M and s, t ∈ S such that m
there exists a v ∈ S such that vt·m = vs·n. This implies
s = n
t . Hence
) .
Ψ(
) =
m
s
1
s ⊗ m =
1
vts ⊗ (vt·m) =
1
vts ⊗ (vs·n) =
1
t ⊗ n = Ψ(
n
t
Obviously Ψ is also a homomorphism of S−1A-modules. Further Ψ ◦ Φ = Id and
Φ ◦ Ψ = Id.
Proposition 29. Let A be a commutative ring and S a multiplicative subset of A.
Then S−1A is a flat A-module.
Proof. Let ϕ : L −→ M be an injective homomorphism of left A-modules. By the
preceding considerations it suffices to prove that 1⊗ ϕ : (S−1A)⊗A L −→ (S−1A)⊗A M
is injective. By means of the isomorphisms Φ and Ψ given in Lemma 28 we see that
1 ⊗ ϕ corresponds to the homomorphism
¯ϕ : S−1L −→ S−1M , ¯ϕ(
l
s
) =
ϕ(l)
s
.
s ) = ϕ(l)
It suffices to check that ¯ϕ is injective: Let l ∈ L and s ∈ S such that 0 = ¯ϕ( l
s .
Then there exists v ∈ S such that 0 = v·ϕ(l) = ϕ(v·l). Hence 0 = v·l because ϕ is
injective. This proves l
s = 0
1 .
Appendix C: Semiprimitive and semiprime ideals
Here we collect some properties of (semi-)primitive and (semi-)prime ideals in non-
commutative rings.
Definition 30. Let B be a ring and Λ a subset of B. We say that Λ is m-closed if, for
any a, b ∈ Λ, there exists some x ∈ B such that axb ∈ Λ. Further Λ is called p-closed
if, for any a ∈ Λ, there exists some x ∈ B such that axa ∈ Λ.
Prime ideals are characterized easily as follows.
Proposition 31. Let B be a ring and I an ideal of B with I 6= B. Then there are
equivalent:
(i) I is prime.
(ii) a1Ba2 ⊂ I implies a1 ∈ I or a2 ∈ I.
(iii) B \ I is an m-closed subset of B.
Proof. First we prove (i)⇒(ii). Suppose that I is prime. Let a1, a2 be in B such that
a1Ba2 ⊂ I. Then (Ba1B)(Ba2B) ⊂ I. Since I is prime, it follows Ba1B ⊂ I or
Ba2B ⊂ I, and hence a1 ∈ I or a2 ∈ I because B is unital. This proves (ii). Obviously
(iii) is the contraposition of (ii). Thus (ii)⇔(iii). It remains to prove (ii)⇒(i). Let J1
and J2 be ideals of B such that J1J2 ⊂ I. Suppose that J2 6⊂ I and choose b ∈ J2 \ I.
For every a ∈ J1 we have aBb ⊂ J1J2 ⊂ I and thus a ∈ I by (ii). This proves J1 ⊂ I.
Hence I is prime.
21
Before we state a similar characterization for semiprime ideals, we prove the following
auxiliary result.
Lemma 32. Let B be a ring. If Λ is a p-closed subset of B and x ∈ Λ, then there
exists a countable m-closed subset Λ0 of B with Λ0 ⊂ Λ and x ∈ Λ0.
Proof. By induction we define a sequence {xn : n ≥ 0} of elements of Λ as follows: Set
x0 = x. If x0, . . . , xn are defined, then there is a y ∈ B such that xnyxn ∈ Λ because
Λ is p-closed. We set xn+1 = xnyxn. Finally we define Λ0 = {xn : n ≥ 0}. Now
we must prove that xmBxn ∩ Λ0 6= ∅ for all m, n ≥ 0: Suppose that m ≤ n. Then
xn+1 ∈ xnBxn ⊂ xmBxn and xn+1 ∈ Λ0. The case m ≥ n is similar.
Proposition 33. Let B be a ring and I 6= B an ideal of B. Then there are equivalent:
(i) I is semiprime.
(ii) If J is an ideal of B such that J 2 ⊂ I, then J ⊂ I.
(iii) aBa ⊂ I implies a ∈ I.
(iv) B \ I is a p-closed subset of B.
Proof. First we prove (i)⇒(ii). Suppose that I is semiprime and let J be an ideal of B
such that J 2 ⊂ I. Let P be an arbitrary prime ideal of B such that I ⊂ P . From
J 2 ⊂ P it follows J ⊂ P . Intersecting all these P we conclude J ⊂ √I = I which
proves (ii). Next we verify (ii)⇒(iii). Let a ∈ B such that aBa ⊂ I. Then it follows
(BaB)(BaB) ⊂ I and hence BaB ⊂ I by (ii). Thus a ∈ I because B is unital. This
proves (iii). Obviously (iv) is the contrapostion of (iii) so that (iii)⇔(iv). Finally we
establish (iv)⇒(i). Suppose that B \ I is p-closed. We must show that √I ⊂ I. Let
a ∈ B \ I. From Lemma 32 we deduce that there exists an m-closed subset Λ0 of B \ I
such that a ∈ Λ0. By Proposition 31 we know that P = B \ Λ0 is prime. Since a 6∈ P ,
we conclude a 6∈ √I.
References
[1] S. A. Amitsur, Radicals of polynomial rings. Canad. J. Math. 8 (1956), pp.
355 -- 361.
[2] M. Artin, L. W. Small, J. J. Zhang, sl Generic flatness for strongly Noethe-
rian algebras. J. Algebra 221 (1999), no. 2, pp. 579 -- 610.
[3] J. Dixmier, Alg`ebres enveloppantes. Editions Jacques Gabay, Paris, 1996.
[4] J. Dixmier, Sur les repr´esentations induites des alg`ebres de Lie. J. Math.
Pures Appl. 9, vol. 50, (1971), pp. 1 -- 24.
[5] M. Duflo, Caract`eres des groupes et des alg`ebres de Lie r´esolubles. Ann. Sci.
´Ecole Norm. Sup. 4, vol. 3 (1970), pp. 23 -- 74.
22
O. Ungermann
[6] M. Duflo, Repr´esentations induites d'alg`ebres de Lie. C. R. Acad. Sci. Paris
S´er. A-B 272 (1971), pp. A1157 -- A1158.
[7] M. Duflo, Certaines alg`ebres de type fini sont des alg`ebres de Jacobson. J.
Algebra, 27 (1973), pp. 358 -- 365.
[8] O. Goldman, Hilbert rings and the Hilbert Nullstellensatz. Math. Z. 54
(1951), pp. 136 -- 140.
[9] A. Grothendieck, Revetements ´etales et groupe fondamental. Fasc. I: Ex-
pos´es 1 `a 5. S´eminaire de G´eom´etrie Alg´ebrique, 1960/61, Institut des
Hautes ´Etudes Scientifiques, Paris, 1963.
[10] R. S. Irving, Generic flatness and the Nullstellensatz for Ore extensions.
Comm. Algebra, 7 (1979), no. 3, pp. 259 -- 277.
[11] W. Krull, Jacobsonsche Ringe, Hilbertscher Nullstellensatz, Dimensionsthe-
orie. Math. Z. 54 (1951), pp. 354 -- 387.
[12] D. Quillen, On the endomorphism ring of a simple module over an enveloping
algebra. Proc. Amer. Math. Soc. 21 (1969), pp. 171 -- 172.
[13] J. L. Rabinovich, Zum Hilbertschen Nullstellensatz. Math. Ann. 102 (1929),
p. 502.
[14] A. F. Szczepa´nski, Generic flatness and the Jacobson conjecture. Comm.
Algebra 33 (2005), no. 11, pp. 4159 -- 4169.
|
1510.00681 | 1 | 1510 | 2015-10-02T18:52:13 | Valuation on filtered module and relations | [
"math.RA"
] | In this paper we show if R is a filtered ring and M a filtered R module then we can define a valuation on a module for M. Then we show that we can find an skeleton of valuation on M, and we prove some properties such that derived form it for a filtered module. | math.RA | math |
VALUATION ON A FILTERED MODULE AND
RELATIONS
M.H. Anjom SHoa∗, M.H. Hosseini†
Abstract
In this paper we show if R is a filtered ring and M a filtered
R module then we can define a valuation on a module for M . Then
we show that we can find an skeleton of valuation on M , and we prove
some properties such that derived form it for a filtered module.
Key Words: Filtered module, Filtered ring, Valuation on module, skele-
ton of valuation.
1
Introduction
In algebra valuation module and filtered R module are two most impor-
tant structures. We know that filtered R module is the most important
structure since filtered module is a base for graded module especially associ-
ated graded module and completion and some similar results([1],[2],[3],[7],[8]).
So, as these important structures, the relation between these structure is
useful for finding some new structures, and if M is a valuation module
then M has many properties that have many usage for example, Rees val-
uations and asymptotic primes of rational powers in Noetherian rings and
lattices([4],[10]).
In this article we investigate the relation between filtered R module and
valuation module. We prove that if we have filtered R module then we can
find a valuation R module on it. For this we define ν : M → Z such that
for every t ∈ M , and by lemma(3.1), lemma(3.2), lemma(3.3), lemma(3.4)
and theorem(3.1) we show ν has all properties of valuation on R-module M .
Also we show if M is a filtered R module then it has a skeleton of valuation,
continuously we prove some properties for M that derived from skeleton of
valuation([6],[9]).
∗Mohammad Hassan Anjom SHoa, University of Biirjand, [email protected],
[email protected]
†Mohammad Hossein Hosseini, University of Birjand, [email protected]
1
2 Preliminaries
Definition 2.1. A filtered ring R is a ring together with a family {Rn}n≥0
of additive subgroups of R satisfying in the following conditions:
i) R0 = R;
ii) Rn+1 ⊆ Rn for all n ≥ 0;
iii) RnRm ⊆ Rn+m for all n, m ≥ 0.
Definition 2.2. Let R be a ring together with a family {Rn}n≥0 of additive
subgroups of R satisfying the following conditions:
i) R0 = R;
ii) Rn+1 ⊆ Rn for all n ≥ 0;
iii) RnRm = Rn+m for all n, m ≥ 0,
Then we say R has a strong filtration.
Definition 2.3. Let R be a filtered ring with filtration {Rn}n≥0 and M be
a R module with family {Mn}n≥0 of subgroups of M satisfying the following
conditions:
i) M0 = M ;
ii) Mn+1 ⊆ Mn for all n ≥ 0;
iii) RnMm ⊆ Mn+m for all n, m ≥ 0,
Then M is called filtered R module.
Definition 2.4. Let R be a filtered ring with filtration {Rn}n≥0 and M be a
R module together with a family {Mn}n≥0 of subgroups of M satisfying the
following conditions:
i) M0 = R;
ii) Mn+1 ⊆ Mn for all n ≥ 0;
iii) RnMm = Mn+m for all n, m ≥ 0,
Then we say M has a strong filtration.
Definition 2.5. Let M be an R module where R is a ring, and ∆ an ordered
set with maximum element ∞ and ∆ 6= {∞}. A mapping υ of M onto ∆ is
called a valuation on M , if the following conditions are satisfied:
i) For any x, y ∈ M , υ(x + y) ≥ min{υ(x), υ(y)};
2
ii) If υ(x) ≤ υ(y), x, y ∈ M , then υ(ax) ≤ υ(ay) for all a ∈ R;
iii) Put υ−1(∞) := {x ∈ M υ(x) = ∞}. If υ(az) ≤ υ(bz), where a, b ∈ R,
and z ∈ M \ υ−1(∞), then υ(ax) ≤ υ(bx) for all x ∈ M
iv) For every a ∈ R \ (υ−1(∞) : M ), there is an a′ ∈ R such that
υ((a′ a)x) = υ(x) for all x ∈ M
Definition 2.6. Let M be an R module where R is a ring, and let ν be a
valuation on M . A representation system of the equivalence relation ∼ν is
called a skeleton of ν.
Definition 2.7. A subset S of M is said to be ν-independent if S∩ν −1(∞) =
φ, and ν(x) /∈ ν(Ry) for any pair of distinct elements x, y ∈ S. Here, we
adopt the convention that the empty subset φ is ν-independent.
Proposition 2.1. Let M be an R module where R is a ring, and let
υ : M → ∆ be a valuation on M . Then the following statements are true:
i) If ν(x) = ν(y) for x, y ∈ M , then ν(ax) = ν(ay) for all a ∈ R;
ii) ν(−x) = ν(x) for all x ∈ M ;
iii) If ν(x) 6= ν(y), then ν(x + y) = min{ν(x), ν(y)};
iv) If ν(az) = ν(bz) for some a, b ∈ R and z ∈ M \ ν −1(∞),then
ν(ax) = ν(bx) for all x ∈ M ;
v) If ν(az) < ν(bz) for some a, b ∈ R and z ∈ M , then ν(ax) < ν(bx) for
all x ∈ M \ ν −1(∞);
vi) The core ν −1 of ν is prime submodule of M ;
vii) The following subsets constitute a valuation pair of R with core
(M : ν −1(∞)):
Aν = {a ∈ Aν(ax) ≥ ν(x) f or all x ∈ M },
Pν = {a ∈ Aν(ax) ≥ ν(x) f or all x ∈ M \ ν −1(∞)}
Proof. see proposition 1.1 [6]
Definition 2.8. The pair (Aν , Pν ) as in Proposition (2.1) is called the val-
uation pair of R induced by ν or the induced valuation pair of ν.
3
3 Valuation derived from filtered module
In this section we use the four following lemmas for showing the existence
of valuation on filtered module. Let R be a ring with unit and R a filtered
ring with filtration {Rn}n>0 and M be filtered R module with filtration
{Mn}n>0.
Lemma 3.1. Let M be filtered R module with filtration {Mn}n>0. Now we
define ν : M → Z such that for every t ∈ M and ν(t) = min {i t ∈ Mi\Mi+1 }.
Then for all x, y ∈ M we have ν(x + y) ≥ min{ν(x), ν(u)}.
Proof. For any x, y ∈ M such that ν(x) = i also ν(y) = j, and ν(x + y) = k,
so we have x + y ∈ Mk\Mk+1. Without losing the generality, let i < j so
Mj ⊂ Mi hence y ∈ Ri. Now if k < i, then k + 1 ≤ i and Mi ⊂ Mk+1
so x + y ∈ Mi ⊂ Mk+1 it is contradiction. Hence k ≥ i and so we have
ν(x + y) ≥ min {ν(x), ν(y)}.
Lemma 3.2. Let M be filtered R module with filtration {Mn}n>0. Now we
define ν as lemma(3.1). If υ(y) ≤ υ(x), x, y ∈ M , then υ(ay) ≤ υ(ax) for
all a ∈ R;
Proof. Let ν(x) = i and ν(y) = j, since ν(x) ≥ ν(y) then Mj ⊇ Mi. Since
R is filtered ring, there exists k ∈ Z such that a ∈ Rk so
ax ∈ RkMi ⊆ Mk+i
ay ∈ RkMj ⊆ Mk+j
we have i + k ≥ j + k by i ≥ j, then ν(ax) ≥ ν(ay) for all a ∈ R.
Lemma 3.3. Let M be filtered R module with filtration {Mn}n>0. Now we
define ν as lemma(3.1). Put υ−1 := {x ∈ M υ(x) = ∞}. If υ(az) ≤ υ(bz),
where a, b ∈ R, and z ∈ M \ υ−1(∞), then υ(ax) ≤ υ(bx) for all x ∈ M .
Proof. Since a, b ∈ R and z ∈ M then there exist i, j, k ∈ Z such that
a ∈ Ri , b ∈ Rj and z ∈ Mk hence
az ∈ RiMk ⊆ Mi+k
bz ∈ RjMk ⊆ Mj+k
Now if ν(az) ≤ ν(bz) then
k + i ≤ k + j =⇒ i ≤ j =⇒ Rj ⊆ Ri
So we have ν(ax) ≤ ν(bx) for all x ∈ M
4
Lemma 3.4. Let M be filtered R module with filtration {Mn}n>0. Now we
define ν as lemma(3.1). For every a ∈ R \ (υ−1(∞) : M ), there is an a′ ∈ R
such that υ((a′a)x) = υ(x) for all x ∈ M .
Proof. Let x ∈ ν −1(∞) then for all a′, a ∈ R υ((a′ a)x) = υ(x) = ∞.
Now let x /∈ ν −1(∞) and for all a
∈
R \ (ν −1(∞) : M ), then a′a ∈ R \ (ν −1(∞) : M ) and hence ν((a′a)x) 6= ∞.
′ and x ∈ Mi, then a
Let a ∈ Rk , a
We may have one of following conditions:
′
∈ R we have ν((a
a)x) 6= ν(x). So if a
a)x ∈ Mi+k+k
′ .
a ∈ Rk+k
′ so (a
′
∈ Rk
′
′
′
′
1) ν((a
′
a)x) < ν(x).
2) ν(x) < ν((a
′
a)x)
Now if we have (1) then i + k + k′ < i, it is contradiction .
Consequently a′ ∈ Rk′ and a ∈ Rk for k ∈ Z then
′
a
a ∈ Rk
′
+k =⇒ (a
′
a)x ∈ Rk+k
′ Mi ⊆ Mi+k
′
+k.
+k+i ⊆ Mi hence (a′a)x ∈ Mi. So we have ν((a′a)x) < i therefore
Since Mk′
ν(x) > ν((a′a)x), it is contradiction with (2). By now we have
ν(x) = ν((a′a)x).
Theorem 3.1. Let R be a filtered ring with filtration {Rn}n>0, and M be
a filtered R module with filtration {Mn}n>0. Now we define ν : M → Z
such that for every t ∈ M and ν(t) = min {i t ∈ Mi\Mh+1 }. Then ν is a
valuation on M .
Proof.
i) By lemma (3.1) we have For any x, y ∈ M , υ(x+y) ≥ min{υ(x), υ(y)};
ii) We have If υ(x) ≤ υ(y), x, y ∈ M , then υ(ax) ≤ υ(ay) for all a ∈ R
by lemma(3.2);
iii) Put υ−1 := {x ∈ M υ(x) = ∞}. If υ(az) ≤ υ(bz), where a, b ∈ R,
and z ∈ M \ υ−1(∞), then then by lemma (3.3) υ(ax) ≤ υ(ay) for all
x ∈ M ;
iv) For every a ∈ R\(υ−1(∞) : M ), then by lemma(3.4) there is an a′ ∈ R
such that υ((a′ a)x) = υ(x) for all x ∈ M .
So by definition(2.5) ν is a valuation onM if has those conditions.
Corollary 3.1. If M be a filtered R module, then ν : M → Z has all of
properties that explained in Proposition(2.1).
Proposition 3.1. I R is a strongly filtered ring and M is a strongly filtered
R module and there exist valuation ν : M → Z on M , then R should be a
trivial filtered R module.
5
Proof. By definition(2.5)(iv) and theorem(3.1) we have for every a ∈ R \
(υ−1(∞) : M ), there is an a
a)x) = υ(x). Now if
ν(a) = i , ν(a′) = j and ν(x) = k then i + j + k = k so i + j = 0,
consequently Ri = R for every i > 0.
∈ R such that υ((a
′
′
Proposition 3.2. Let M be an R module, where R is a ring. Then there
is a valuation on M , if and only if there exists a prime ideal P of R such
that P MP 6= MP , where MP is the localization of M at P .
Proof. see (Proposition 1.3 [6])
Corollary 3.2. Let M be an filtered R module, where R is a filtered ring.
Then there exists a prime ideal P of R such that P MP 6= MP , where MP is
the localization of M at P .
Proof. By theorem(3.1) there is an valuation on M , then by proposition(3.2)
there exists a prime ideal P of R such that P MP 6= MP , where MP is the
localization of M at P .
Corollary 3.3. Let M be an filtered R module, where R is a filtered ring.
Then there is a skeleton on M .
Proof. By theorem(3.1) there is a valuation on M ,then by definition(2.6) we
have there is a skeleton on M .
Proposition 3.3. Let M be an filtered R module where R is a filtered ring,
and ν a valuation on M . If Λ is a skeleton of ν, then the following conditions
are satisfied:
i) Λ is a ν-independent subset of M ;
ii) For every x ∈ M \ ν −1(∞), there exists a unique λ ∈ Λ such that
ν(x) = ν(Rλ).
Proof. By corollary(3.3) Λ is a skeleton of ν and by proposition(1.4, [6]) we
have the above conditions.
Proposition 3.4. Let M be an filtered R module where R is a filtered ring,
and ν a valuation on M . If Λ is a skeleton of ν. If a1λ1 + · · · + anλn = 0
where a1, · · · an ∈ R and λ1 · · · λn ∈ Λ are mutually distinct, then
ai ∈ (ν −1(∞) : M ),i = 1, · · · , n.
Proof. By corollary(3.3) Λ is a skeleton of ν and by proposition(1.5, [6]) we
have If a1λ1 + · · · + anλn = 0 where a1, · · · an ∈ R and λ1 · · · λn ∈ Λ are
mutually distinct, then ai ∈ (ν −1(∞) : M ),i = 1, · · · , n.
6
References
[1] J. Alajbegovic, Approximation theorems for Manis valuations with the
inverse, (1984). property. Comm. Algebra 12:13991417.
[2] M. F. Atiyah, I. G. Macdonald, Introduction to Commutative Algebra.
Massachusetts:(1969). Addison-Wesley.N.
[3] O. Endler, Valuation Theory. New York: Springer-Verlag.Algebras,
Rings
and Modules by Michiel Hazewinkel CWI,Amsterdam,
The Netherlands Nadiya Gubareni Technical University of Czsto-
chowa,Poland and V.V. KirichenkoKiev Taras Shevchenko Univer-
sity,Kiev, Ukraine KLUWER.(1972).
[4] Fuchs, L. Partially Ordered Algebraic Systems. New York: Pergammon
Press.S.(1963).
[5] Gopalakrishnan, Commutative algebra,oxonian press,1983.
[6] Z. Guangxing, Valuations on a Module, Communications in Algebra,
2341-2356, 2007.
[7] Huckaba, J. A. Commutative Rings with Zero Divisors. New York: Mar-
cel Dekker, Inc.(1988).
[8] T.Y. Lam , A First Course in Noncommutative Rings,Springer-Verlag
,1991.
[9] C. P. Lu, Spectra of modules. Comm. Algebra 175:37413752. (1995).
[10] O. F. G. Schilling, The Theory of Valuations. New York: Amer. Math.
Soc. (1952).
7
|
math/0610921 | 3 | 0610 | 2012-02-14T19:27:28 | Spectral calculations in rings | [
"math.RA",
"math.SP"
] | We examine the validity of certain spectral integral formulas in topological rings. We consider the sign and square-root functions in polymetric rings containing $\frac12$. It turns out that formal analogues of classical transformation kernels and the resolvent identity can be used to understand the situation. In the lack of $\frac12$, the functions $\frac12-\frac12\sgn(\frac12-z)$ and $\frac12-\sqrt{\frac14-z}$ can be generalized, respectively. | math.RA | math |
SPECTRAL CALCULATIONS IN RINGS
GYULA LAKOS
Abstract. We examine the validity of certain spectral integral formulas in topological
rings. We consider the sign and square-root functions in polymetric rings containing 1
2 .
It turns out that formal analogues of classical transformation kernels and the resolvent
identity can be used to understand the situation.
2 sgn(cid:0) 1
4 − z can be generalized, respectively.
2 , the functions 1
2 − z(cid:1) and 1
In the lack of 1
2 − q 1
2 −
1
Spectral integrals like
0. Introduction
(1)
and
(2)
1−z
2 + 1+z
1+z
2 + 1−z
2 Q
2 Q
dz
2π
sgn Q =Z{z∈C : z=1}
z(cid:16)(cid:0) 1+z
2 (cid:1)2
√S =Z{z∈C : z=1}
1
S
−(cid:0) 1−z
2 (cid:1)2
S(cid:17) dz
2π
are often useful. They extend the complex functions sgn Q, which is the sign function of
(the real part of) Q, and √S, which is the square root function cut along the negative real
axis, respectively. Definitions like above are justified if they are supported by appropriate
algebraic identities and spectral properties. This is the situation in linear analysis, where
the formulas above can be established for elements with appropriate spectral properties in
great generality, even if the resolvent terms are not necessarily continuous, cf. Haase [1],
Mart´ınez Carracedo -- Sanz Alix [3]. We refer to this case as the "analytic case".
However, in the analytic case, if the resolvent terms are continuous, then they are also
smooth, and one can expand everything in terms of Fourier series, or rather Laurent series
in z. One can naturally ask if similar computations can be done in more general rings, in
particular, in rings without a natural R-action. It is natural to check these ideas for formal
Laurent series on polymetric rings. We refer to this case as the "algebraic case".
p(0) = 0, p(−X) = p(X), p(X + Y ) ≤ p(X) + p(Y ),
We call a ring A polymetric if
(a) its topology is induced by a family of "seminorms" p : A → [0, +∞) such that
(b) for each "seminorm" p there exists a "seminorm" p such that p(XY ) ≤ p(X)p(Y );
i. e., if it is a polymetric space whose multiplication is compatible with the topology. Dealing
with Laurent series in z, integration over the unit circle becomes a formal process. It is
nothing else but detecting the coefficient of z0. On the other hand, for the multiplication of
Laurent series, some sort of convergence control is required. Primarily, we will be interested
in Laurent series with rapidly decreasing coefficients. A sequence is rapidly decreasing if it is
rapidly decreasing in each seminorm. Furthermore, we consider only sequentially complete,
Hausdorff polymetric rings. We also assume that 1
2 ∈ A. Then (1) and (2) are meaningful.
2000 Mathematics Subject Classification. Primary: 46H30, Secondary: 13J99.
Key words and phrases. Sign operation, square root operation, formal transformation kernels.
1
(3)
and
(4)
dz
2π
1 − P + P z
P z
idem P =Z{z∈C : z=1}
F√T =Z{z∈C : z=1}
2 sgn(cid:0) 1
2 − 1
1. Laurent series
(1 + z)T
dz
2π
,
1 + (z − 2 + z−1)T
2 − P(cid:1) and F√T = 1
2 −q 1
2
GYULA LAKOS
Indeed, applied to elements with appropriate "spectral" properties, the expressions sgn Q
and √S will have good properties justifying the notation. This can be proved by an analysis
of the coefficients. The algebraic approach is particularly manageable in the case of (1), it
is essentially shown in Karoubi [2], by a direct analysis of coefficients, that the expression
sgn Q yields an involution compatible with factorization of affine loops. Nevertheless, such
computations are not necessarily very enlightening. The objective of this paper is to prove
our statements regarding the algebraic case and to do this in a manner which brings the
algebraic and analytic cases together, at least formally. It turns out that the basic tool of
the analytic case, the resolvent identity, works generally. Another natural question is what
happens in the lack of 1
2 . Then the sign and square root functions are not really appropriate.
Instead, we can generalize
extending the functions idem P = 1
4 − T , repectively.
1.1. If A is a polymetric ring, then we may consider formal Laurent series a =Pn∈Z anzn.
pα(a) = Pn∈Z α(n)p(an). Then we may consider the polymetric spaces (a) A[z−1, z]f of
If p is a seminorm on A and α : Z → R+ is a non-negative function, then we may define
essentially finite Laurent series, (b) A[z−1, z]∞ of rapidly decreasing Laurent series, (c)
A[z−1, z]b of "summable" Laurent series, (d) A[[z−1, z]]b of bounded Laurent series, (e)
A[[z−1, z]]∞ of polynomially growing Laurent series, (f) A[[z−1, z]]f of formal Laurent series,
as the spaces which contain series bounded for each pα such that (a) α is unrestricted, (b) α
is polynomially growing, (c) α is bounded, (d) α is summable, (e) α is rapidly decreasing, (f)
α is vanishing except at finitely many places, respectively. We have continuous inclusions
A[z−1, z]f ֒→ A[z−1, z]∞ ֒→ A[z−1, z]b ֒→ A[[z−1, z]]b ֒→ A[[z−1, z]]∞ ֒→ A[[z−1, z]]f .
Of these spaces, A[z−1, z]f , A[z−1, z]∞, A[z−1, z]b will remain polymetric rings. Indeed,
g(pα) can be chosen as p α, where α(n) = 1 ∨ max−2n≤m≤2n α(m). We have compatible
continuous module actions A[z−1, z]• × A[[z−1, z]]• → A[[z−1, z]]•. Essentially the same
applies to the spaces of power series A[z]f , A[z]∞, A[z]b, A[[z]]b, A[[z]]∞, A[[z]]f , except
here even A[[z]]∞, A[[z]]f are polymetric rings. Indeed, for them, g(pα) can be chosen as p `α,
where `α(n) =pmaxn≤m α(m). An element like 1+Q
1.2. If a(z) =Pn∈Z anzn ∈ A[[z−1, z]]f , then we define formally
2 z may be considered either as
an element of A[z−1, z]f , A[z−1, z]∞, or A[z−1, z]b, etc. Practically, the difference is that
the larger the ring is the easier is to find a multiplicative inverse of the element given.
2 + 1−Q
Z a(z) dz
2π
= a0.
The Hilbert kernel ("up to multiplication by i") is defined as
(cid:20) 1 + z
1 − z(cid:21) =Xs∈Z
(sgn s)zs ∈ A[[z−1, z]]b.
SPECTRAL CALCULATIONS IN RINGS
3
1.3. Some further terminology is as follows. For a(z) =Pn∈Z anzn ∈ A[z−1, z]b we let
an.
lim
zր1
a(z) =Xn∈Z
Naturally, this notation also applies to power series.
In what follows, we let A[z−1, z], A[[z−1, z]] A[z], A[[z]] denote A[z−1, z]∞, A[[z−1, z]]∞,
A[z]∞, A[[z]]∞, respectively, but similar statements hold for f and b, too.
Proposition 1.4. For a(t, z) ∈ A[t] [[z−1, z]] and b(z) ∈ A[z−1, z],
tր1Z (a(t, z)) b(z) dz
a(t, z)(cid:19) b(z) dz
Z (cid:18)lim
= lim
tր1
2π
.
2π
Proof. This is just the generalized associativity of the rapidly decreasing (hence absolute
(cid:3)
convergent) sumPn∈N,s∈Z an,sb−s.
1.5. For the sake of brevity, we call the elements of A[t] [[z−1, z]] as transformation kernels.
Practically, the convenient thing is to consider those elements of the ring A[[t]]f [z−1, z]b
which can be thought to be transformation kernels. (This is advantageous from compu-
tational viewpoint, because the product of a(t, z) and b(z) ∈ A[z−1, z] formally yields the
same element of A[[t]]f [[z−1, z]]b either we interpret a(t, z) ∈ A[[t]]f [z−1, z]b or a(t, z) ∈
A[t] [[z−1, z]], but the first case is often easier to compute with.)
Such elements are the Poisson kernel
P(t, z) =
1 − t2
(1 − tz)(1 − tz−1)
the Hilbert-Poisson kernel
H(t, z) =
t(z − z−1)
(1 − tz)(1 − tz−1)
the 1
2 -shifted odd Poisson kernel
L(t, z) =
(1 − t)(1 + z)
(1 − tz)(1 − tz−1)
and the variant regularization kernel
(1 + t)t(1 − z)(1 − z−1)
R(t, z) =
2(1 − tz)(1 − tz−1)
This latter one has the property
= Xs∈N,s>0
R(t, z) = 1
lim
tր1
tszs,
(sgn s)tszs,
=Xs∈Z
=Xs∈Z
tsz−s +Xs∈N
=Xs∈N
z−s + t + Xs∈N,s>0
ts+1 − ts
tszs+1,
2
ts+1 − ts
2
zs.
(here we think of R(t, z) as an element of A[t] [[z−1, z]]). The ordinary regularization kernel
R(t, z) =
t(1 − z)(1 − z−1)
(1 − tz)(1 − tz−1)
is just an element of A[[t]]f [z−1, z]b, hence it is not so convenient algebraically. On the other
hand, the use of variant regularization makes the variant Hilbert-Poisson kernel
H(t, z) =
1 + t
2 H(t, z) =
(1 + t)t(z − z−1)
2(1 − tz)(1 − tz−1)
=Xs∈Z
(sgn s)
1 + t
2
tszs
useful.
4
GYULA LAKOS
It is not hard to see that the definitions and the proposition above can be formulated
in the case when we have many variables z, w, . . . instead of just z. For example, we may
consider the Hilbert kernel
z − w(cid:21) =(cid:20) 1 + wz−1
(cid:20) z + w
1 − wz−1(cid:21) =Xs∈Z
(sgn s)wsz−s.
2.1. In order to save some space we use the short-hand notation
2. Spectral classes
Λ(a) = 1
Λ(a, b) = 1
Λ(a, b, c) = 1
Λ(a, b, c, d) = 1
2 (1 + a),
2 (1 + a + b − ab),
4 (1 + a + b + c − ab + ac − bc + abc),
4 (1 + a + b + c + d − ab − bc − cd + ac + ad + bd
+ abc − acd − abd + bcd − abcd),
etc., following the scheme
Λ(c1, c2, c3, . . . , cn) = 2−⌈ n
2 ⌉ Xε∈{0,1}n Y1≤j<n
(−1)εj εj+1 Y1≤k≤n
such that the order of the symbols ck is preserved in the products.
2.2. Let C = C ∪ {∞} denote the Riemann sphere. Some subsets are:
iR = iR ∪ {∞},
R− = (−∞, 0] ∪ {∞}, C− = {s ∈ C : Re s ≤ 0} ∪ {∞}, D1 = {z ∈ C : z < 1}. We define
the functions pol J = −i sgn iJ, Ji = √−J 2, Qr =pQ2, PF = 1
2 −(cid:12)(cid:12) 1
the following commutative diagram on certain subsets of the complex plane:
k
2 − P(cid:12)(cid:12)r. We have
cεk
(5)
S ∈ C \ R−
J7→−J 2
Q ∈ C \ C−
√S
S7→
Q7→Q2
T7→1−4T
T ∈ C \(cid:0) 1
4
S7→ 1−S
4 − R−(cid:1)T7→ F√T
Q7→ 1−Q
2
2 − C−(cid:1)
P7→1−2P
P ∈ C \(cid:0) 1
W ∈ D1
P7→P (1−P )
W7→−W (1−W )−1
P7→−P (1−P )−1
J7→pol J
J ∈ C \ R
J ∈ {i,−i}
❦
❦
❦
❦
❦
❦
J7→Ji
❦
❦
❦
❦
❦
❦
❦
u❦
Q7→Qr
Q ∈ C \ iR
Q7→sgn Q
Q ∈ {1,−1}
P7→PF
P7→1−2P
Q7→ 1−Q
2
P ∈ C \(cid:0) 1
2 − iR(cid:1)
P7→1−2P
P7→idem P
Q7→ 1−Q
2
P ∈ {0, 1}
such that pol J, sgn Q, idem P , Ji, Qr, PF yield idempotent operations and they yield
decompositions
J = Ji pol J,
Q = Qr sgn Q,
P = idem P + PF − 2PF idem P.
Definition 2.3. Suppose that A is a locally convex algebra and R is a compact subset of
C. We define SpecR(A) as the set containing all elements X ∈ A such that the functions
fR : z ∈ R \ {∞} 7→ (z − X)−1
and
gR : z ∈ R \ {0} 7→ X(1 − z−1X)−1
are well-defined and continuous. The topology of SpecR(A) is induced from the compact-
open topology of the continuous functions (fR)D1∩R and (gR)R\D1 . (If R is symmetric for
conjugation then we may use the functions fR(z)fR(¯z) and gR(z)gR(¯z) in order to get a
formally real characterization.)
-
-
u
u
u
-
-
o
o
n
n
.
.
q
q
.
.
J
J
o
o
J
J
n
n
-
-
p
p
J
J
J
J
J
J
SPECTRAL CALCULATIONS IN RINGS
5
Classes of interest are like SpeciR(A), etc., i. e. the elements spectrally avoiding iR, etc.
Another way to specify spectral conditions is to ask for skew-involutions, involutions, or
idempotents. The main spectral classes correspond to the sets in (5) for A = C.
Definition 2.4. We define the corresponding formal spectral classes by
+(cid:0) 1−z
2 (cid:1)2
J 2(cid:17) is invertible in A[z, z−1],
2 Q = Λ(z, Q) is invertible in A[z, z−1],
2 Q = Λ(z, Q) is invertible in A[z],
z(cid:16)(cid:0) 1+z
2 (cid:1)2
J ∈ ]SpecR(A) ⇔ 1
J ∈ Skvol(A) ⇔ J 2 = −1,
Q ∈ ]SpeciR(A) ⇔ 1+z
2 + 1−z
Q ∈ ]SpecC−(A) ⇔ 1+z
2 + 1−z
Q ∈ Invol(A) ⇔ Q2 = 1,
z(cid:16)(cid:0) 1+z
2 (cid:1)2
S ∈ ]SpecR−(A) ⇔ 1
P ∈ ]Spec 1
P ∈ ]Spec 1
P ∈ Idem(A) ⇔ P 2 = P ,
−(cid:0) 1−z
2 (cid:1)2
S(cid:17) = Λ(z, S, z−1) is invertible in A[z, z−1]
2 +iR(A) ⇔ (1 − P ) + P z is invertible in A[z, z−1],
2−C−(A) ⇔ (1 − P ) + P z is invertible in A[z],
4−R−(A) ⇔ 1 − (1 − z)(1 − z−1)T is invertible in A[z, z−1],
T ∈ ]Spec 1
W ∈ ]SpecC\D1(A) ⇔ 1 + zW is invertible in A[z].
2.5. If A is a locally convex algebra, then the formal spectral classes and their ordinary
counterparts are the same. Indeed, the continuity of the resolvent terms implies smooth-
ness by the resolvent identity, hence the existence of the appropriate Fourier series, and,
conversely, the existence of the expansions implies continuity.
Objective 2.6. We want to establish the spectral correspondences and decompositions as
in point 2.2 for the formal spectral classes.
A. Sign and square root.
3. Calculations with 1
2
Definition 3.1. For Q ∈ ]SpeciR(A), we define
2 Q
dz
2π
2 Q
sgn Q =Z 1−z
2 + 1+z
2 + 1−z
1+z
=Z Λ(−z, Q)
Λ(z, Q)
dz
2π
.
Proposition 3.2. If Q ∈ ]SpeciR(A), then −Q, Q−1 ∈ ]SpeciR(A). Q commutes with sgn Q.
sgn−Q = − sgn Q and sgn Q−1 = sgn Q. Moreover,
(sgn Q)2 = 1.
Proof. Substituting z = −1 we see that Q−1 exists. The first statement follows from the
identities Λ(z,−Q) = zΛ(z−1, Q) and Λ(z, Q−1) = Q−1Λ(−z, Q). Furthermore, Q and
sgn Q commute, because Q commutes with the integrand in sgn Q. The identities
Λ(−z,−Q)
Λ(z,−Q)
Λ(−z, Q−1)
Λ(z, Q−1)
Λ(−z−1, Q)
Λ(z−1, Q)
Λ(z, Q)
Λ(−z, Q)
d(−z)
2π
d(z−1)
2π
dz
2π
=
dz
2π
= −
and
integrated prove the first and second equalities, respectively.
The critical one is the involution property. We give several proofs.
dw
2π
,
"Resolvent algebraic" proof. As
(6)
2π
Λ(−w, Q)
Λ(−z, Q)
Λ(z, Q)
Λ(−w, Q)
Λ(w, Q)
Λ(w, Q) (cid:19) dz
1 − (sgn Q)2 =ZZ (cid:18)1 −
=(cid:18) z + w
2 (cid:19)
2(cid:20) z + w
z − w(cid:21)(cid:18) (z − 1)(1 − Q2)
2π −ZZ (cid:20) z + w
Λ(z, Q)Λ(w, Q)
1 − Q2
dz
2π
Λ(z, Q)
dw
=
=
1
Λ(−z, Q)
Λ(z, Q)
(7) 1 −
1
=
Λ(z, Q)Λ(w, Q)
2(cid:20) z + w
z − w(cid:21) (z − w)(1 − Q2)
z − w(cid:21) (z − 1)(1 − Q2)
=ZZ (cid:20) z + w
=ZZ 0 ·
Evaluating the integrals we find
(z − 1)(1 − Q2)
2Λ(z, Q)
2Λ(z, Q)
dz
2π −ZZ 0 ·
(w − 1)(1 − Q2)
2Λ(w, Q)
dw
2π
= 0 − 0 = 0.
(w − 1)(1 − Q2)
Λ(w, Q)
−
(cid:19) ,
z − w(cid:21) (w − 1)(1 − Q2)
2Λ(w, Q)
dz
2π
dw
2π
.
which does make sense in A[[z−1, z]][[w−1, w]]. Indeed, (6) can be continued as
6
GYULA LAKOS
2
1+Q
1−Q
2 es+ 1
2 z−1/2 + 1−Q
2(cid:1) ×(cid:0)Z + 1
(sgn s)es,s be the (cid:0)Z + 1
2(cid:1) ma-
"Matrix algebraic" proof. Let H1/2 =Ps∈Z+ 1
2 z1/2(cid:17) = Ps∈Z+ 1
trix of the odd Hilbert transform. Let U(cid:16) 1+Q
2(cid:1) matrix of the action of multiplication by(cid:16) 1+Q
2 z(cid:17) z−1/2.
2 ,s be the Z×(cid:0)Z + 1
According to our assumption, this has a(cid:0)Z + 1
2(cid:1)× Z inverse matrix representing the action
of multiplication by z1/2(cid:16) 1+Q
B(cid:16) 1+Q
2 z1/2(cid:17)−1
2 z1/2(cid:17) H1/2U(cid:16) 1+Q
which is the same as the even Hilbert transform H = Ps∈Z(sgn s)es,s, except in the 0th
column. This special shape implies that the diagonal element in the 0th column is an
involution. On the other hand, it is easy to see that this diagonal element is exactly sgn Q.
Due to the special shape of the matrices involved, it is easy to see that this is an involution
2 z1/2(cid:17) = U(cid:16) 1+Q
. So, we can consider the matrix
2 z(cid:17)−1
2 z−1/2 + 1−Q
2 z−1/2 + 1−Q
2 z−1/2 + 1−Q
2 + 1−Q
2 + 1−Q
2 es− 1
2 ,s −
.
2
we should show that this integral is 0. This, however, follows from the key identity
This proof, like the previous one, relies heavily on the nature of Laurent series. Nevertheless
the argument can be modified so that formally it makes sense in the analytical and the
algebraic cases as well.
"Resolvent analytic" proof. According to the discussion about transformation kernels,
(6) can be continued as follows:
= lim
tր1ZZ R(t, wz−1)(cid:18)1 −
Λ(−z, Q)
Λ(z, Q)
Λ(−w, Q)
Λ(w, Q) (cid:19) dz
2π
dw
2π
.
By simple arithmetic in the integrand, this yields
(z − 1)(1 − Q2)
2Λ(z, Q)
dz
2π
Executing the integrals we find
= lim
tր1(cid:18)ZZ H(t, wz−1)
tր1(cid:18)Z 0 ·
= lim
(z − 1)(1 − Q2)
Λ(z, Q)
dz
2π −Z 0 ·
dw
2π −ZZ H(t, wz−1)
(w − 1)(1 − Q2)
Λ(w, Q)
(w − 1)(1 − Q2)
2Λ(w, Q)
dz
2π
dw
2π (cid:19).
dw
2π (cid:19) = 0 + 0 = 0,
yielding, ultimately, the identity. We remark that in the analytic case, it would actually be
simpler to use the kernel R(t, wz−1).
(cid:3)
SPECTRAL CALCULATIONS IN RINGS
7
S
S
zS
dz
2π
(cid:0) 1+z
2 (cid:1)2
√S =Z
Definition 3.3. If S ∈ ]SpecR−(A), then we define the inverse square root operation as
=Z
−(cid:0) 1−z
2 (cid:1)2
Proposition 3.4. Suppose that S ∈ ]SpecR−(A). Then S−1 ∈ ]SpecR−(A). The elements
√S and S commute with each other. √S−1 = √S−1
(√S)2 = S.
(8)
Proof. Substituting z = −1 into the resolvent term, we see that S−1 exists. The identity
Integrated, it yields
Λ(z, S−1, z−1)
S√S−1 = √S. If the square-root identity (8) holds, then this implies √S−1 = √S−1
shows that S−1 ∈ ]SpecR−(A).
Λ(−z, S, (−z)−1)
. Furthermore,
Λ(z, S, z−1)
dz
2π
. So,
=
S
1
.
it remains to prove (8). As
(√S)2 − S =ZZ (cid:18)
S
Λ(z, S, z−1)
S
Λ(w, S, w−1) − S(cid:19) dz
2π
dw
2π
,
we have to show that this integral is 0. This follows using the key identities
(10)
S
Λ(z, S, z−1)
S
Λ(w, S, w−1) − S =
= S
1
+ S
S − Λ(z, S, z−1, 1, w, S, w−1)
2Λ(z, S, z−1)Λ(w, S, w−1)
2(cid:20) zw−1 + 1
zw−1 − 1(cid:21)(cid:18) Λ(−z, S, z−1)
2(cid:20) zw + 1
zw − 1(cid:21)(cid:18)Λ(−z, S, z−1)
S − Λ(z, S, z−1, 1, w−1, S, w)
2Λ(z, S, z−1)Λ(w, S, w−1)
Λ(w, S, w−1) (cid:19) ;
Λ(−w, S, w−1)
Λ(w, S, w−1) (cid:19) .
Λ(w, S,−w−1)
Λ(z, S, z−1) −
Λ(z, S, z−1) −
=
=
1
;
S − Λ(z, S, z−1, 1, w, S, w−1)
2Λ(z, S, z−1)Λ(w, S, w−1)
S − Λ(z, S, z−1, 1, w−1, S, w)
2Λ(z, S, z−1)Λ(w, S, w−1)
Indeed, after we decomposed the integrand in (9) according to (10 -- 12), we can show that
both parts are 0 as we did in the previous proof.
(cid:3)
Proposition 3.5. Q ∈ ]SpeciR(A) if and only if Q2 ∈ ]SpecR−(A). In this case
(9)
(11)
(12)
sgn Q = Q−1pQ2.
Proof. The first statement follows from the equality Λ(z, Q2, z−1) = Λ(z, Q)Λ(z−1, Q). The
identity statement follows from
1
1
2
2
+
=
+
+
sgn Q
sgn Q
Λ(z, Q)
sgn Q =
dz
2π
Λ(z−1, Q)
=Z 1
2(cid:18) Λ(−z, Q)
2Z Λ(−z, Q)
Λ(z, Q)
Λ(z−1, Q) (cid:19) dz
Λ(−z−1, Q)
2Z Λ(−z−1, Q)
=Z
Definition 3.6. (a) We define Qr =pQ2 = Q sgn Q.
2 (cid:21) .
B. Finer analysis of the resolvent terms.
1−F
2 A 1+F
1+F
2 A 1+F
2 A 1−F
2 A 1−F
(cid:20) 1−F
Λ(z, Q2, z−1)
1+F
2π
Q
2
2
2
(b) If F ∈ A is an involution, then an element A ∈ A can be written in matrix form
2π
d(z−1)
= Q−1pQ2.
dz
2π
(cid:3)
8
GYULA LAKOS
Suppose that Q ∈ ]SpeciR(A). In the decomposition of A along the involution sgn Q, the
various components are denoted according to
sgn Q =(cid:20)−1− sgn Q
Q+(cid:21) .
Qr =(cid:20)Q−
Proposition 3.7. Q± ∈ ]SpeciR(A1± sgn Q), sgn Q± = 1± sgn Q, and sgnQr = 1.
Q =(cid:20)−Q−
1sgn Q(cid:21) ,
Q+(cid:21) ,
Proof. The decomposition of Q along sgn Q allows us to consider Q separately in the direct
sum components of A. In particular, the sign integral splits, too, and it necessarily yields
sgn Q+ = 1sgn Q and sgn−Q− = −1sgn −Q. The statement follows from this immediately. (cid:3)
Proposition 3.8. If Q ∈ ]SpeciR(A), then
(a)
1
1+z
2 + 1−z
2 Q
=
1
Λ(z, Q)
is given by
Q−+1
2
. . . +"(cid:16) Q−−1
Q−+1(cid:17)2
+(cid:20)0
(b)
Q−+1
0# z−3 +"(cid:16) Q−−1
Q++1(cid:21) 1 +"0
Q−+1(cid:17) 2
Q++1(cid:16) Q+−1
0# z−2 +(cid:20) 2
Q++1(cid:17)# z +"0
0(cid:21) z−1+
Q++1(cid:17)2# z2 + . . .
Q++1(cid:16) Q+−1
Q−+1
2
2
2
2 + 1+z
1−z
2 + 1−z
1+z
2 Q
2 Q
=
Λ(−z, Q)
Λ(z, Q)
is given by
. . . +"−2(cid:16) Q−−1
Q−+1(cid:17)3
1sgn Q(cid:21) 1 +"0
+(cid:20)−1− sgn Q
(c) In particular,
0# z−2 +"−2(cid:16) Q−−1
Q++1(cid:17)2# z2 +"0
2(cid:16) Q+−1
Q−+1(cid:17) 0# z−1+
Q++1(cid:17)3# z3 + . . .
2(cid:16) Q+−1
0# z−3 +"−2(cid:16) Q−−1
Q−+1(cid:17)2
Q++1(cid:17)# z +"0
2(cid:16) Q+−1
=Z 1
1 + Qr
1
2
1 + z
Λ(z, Q)
dz
2π
.
We try to figure out the coefficients in the expansion
Proof. (a) It is enough to consider the Q+ part of the decomposition, because the other
part follows from changing z to z−1. So, we can suppose that Q = Q and sgn Q = 1.
sgn Q = 1 means that
1 − sgn Q
Λ(z,Q) =Pn∈Z anzn. The equality
Λ(−z, Q)
1 − Q
(13)
0 =
=
=
z
1
dz
2π
Λ(z, Q)
a−1.
2
2
The product of Λ(z, Q) = 1+Q
(14)
1
2Z (cid:18)1 −
Λ(z, Q) (cid:19) dz
=Z 1 − Q
2 z andPn∈Z anzn gives 1, so
1 + Q
2π
2
an = δ0,n1.
an−1 +
2
2 + 1−Q
1 − Q
2
From (13) and the case n = 0 in (14), we obtain that a0 = 2
find an+1 = Q−1
already inverts Λ(z, Q), hence, from the uniqueness of the inverse, a(z) = F (z).
Q+1 an, yielding the positive-numbered coefficients. Then F (z) =Pn∈N anzn
1+Q . After that, from (14), we
SPECTRAL CALCULATIONS IN RINGS
(b) and (c) follow from part (a) by simple algebra.
9
(cid:3)
Proposition 3.9. (a) If S ∈ ]SpecR−(A), then
−(cid:0) 1−z
2 (cid:1)2
(cid:0) 1+z
2 (cid:1)2
√S + 1! (z + z−1) + √S−1 √S − 1
1 + √S−1 √S − 1
√S + 1!2
yields the expansion
Λ(z, S, z−1)
√S−1
=
S
1
z
(z2 + z−2) + . . .
(b) In particular,
1
√S + 1
=Z 1
2
1 + z
Λ(z, S, z−1)
dz
2π
.
Proof. (a) Take Q = √S. Proposition 3.5 yields sgn Q = 1. Applying the identity
and Proposition 3.8, it follows that the coefficient of z±n
=
1
1
Λ(z, S, z−1)
(n ≥ 0) in the expansion is
Λ(z, Q)Λ(z−1, Q)
,
(cid:3)
Q + 1(cid:19)n+2m
Q + 1(cid:19)2Xm∈N(cid:18) Q − 1
(cid:18) 2
Q + 1(cid:19)n 1 −(cid:18) Q − 1
Q + 1(cid:19)2(cid:18) Q − 1
which simplifies as above. The rapid decrease of(cid:16) Q−1
Q+1(cid:17)s
Proposition 3.10. (a) Q ∈ ]SpecC−(A) if and only if Q ∈ ]SpeciR−(A) and sgn Q = 1.
Q + 1(cid:19)2!−1
=(cid:18) 2
(b) follows from part (a).
makes our computations legal.
1+Q ∈ ]SpecC\D1(A). Conversely, if W ∈ ]SpecC\D1(A), then
(b) W ∈ ]SpecC\D1(A) if and only if W n is rapidly decreasing.
(c) If Q ∈ ]SpecC−(A), then 1−Q
1+W ∈ ]SpecC−(A). This establishes a bijection.
1−W
Proof. (a) follows from Proposition 3.8.a. (b) holds because in those cases (1 − W z)−1 =
1+Q . (cid:3)
Pn∈N W nzn must hold. (c) follows from Λ(cid:16)z, 1−W
all the spectral correspondences asked in 2.6. We merely define Ji = √−J 2, pol J =
J − J 2−1
4 − T .
Hence our objective is established.
3.11. Now, it is easy to see that the propositions proven above are sufficient to establish
1+W(cid:17) = 1+zW
1+W and 1 + 1−Q
1+Q z = 2Λ(z,Q)
C. On our objective.
2 − P(cid:1), PF = 1
2 − P(cid:12)(cid:12)r, and F√T = 1
This is, however, not to say that everything is just like for locally convex algebras:
2 −q 1
2 sgn(cid:0) 1
2 −(cid:12)(cid:12) 1
, idem P = 1
2 − 1
i
D. Comparison to the case of locally convex algebras.
Proposition 3.12. If A is a locally convex algebra, then the condition that S ∈ SpecR−(A)
is equivalent to the condition that the function
1 + t
2
+
1 − t
2
S
has a continuous inverse on [−1, 1]. (This is the same thing as to say that the segment
connecting 1 and S is continuously invertible.) The square root can be expressed as
√S =Zt∈[−1,1]
S
1+t
2 + 1−t
2 S
dt
π√1 − t2
.
10
GYULA LAKOS
Proof. It follows by change of variables using t = z+z−1
3.13. If S ∈ SpecC−(A), then 1+t
decreasing power series in t, using t = z+z−1
(cid:16) z+z−1
2 (cid:17)n
2 S (t ∈ [−1, 1]) is clearly invertible. As a rapidly
and considering the coefficients of zk in
, it follows that S ∈ SpecR−(A). Hence the inclusion SpecC−(A) ⊂ SpecR−(A) is
true. In the general context, this cannot be done so, because the boundedness of the elements
2n ( n
1
k ) is not always clear. Similar comment applies for Spec 1
.
(cid:3)
2 + 1−t
2
2
2 +C−(A) ⊂ Spec 1
2 +R−(A).
4. Formal homotopies
4.1. One expects certain natural behaviour from the operations above. For example, one
expects to have a homotopy from Q to sgn Q inside ]SpeciR(A). In general algebras, one
cannot use continuous variables, but one can come up with homotopies using formal vari-
ables. Let us remind that an element Q ∈ ]SpeciR(A) can be decomposed to a commuting
pair, sgn Q and a perturbation of 1 which is Qr. But we may also consider this as a
decomposition to the commuting pair sgn Q and a perturbation of 0 which is
dz
2π
dz
2π
1+z
= −Z Λ(z,−Q)
Λ(z, Q)
pertr Q = Qr − 1
Qr + 1
= −Z 1+z
2 − 1−z
2 Q
2 + 1−z
2 Q
If we replace pertr Q by t pertr Q in the decomposition, then we obtain a homotopy, this
appears as K(t,−1, Q) in what follows.
Definition 4.2. We define
.
K(t, z, Q) =
1 + sgn Q
2
2
H(t, z, Q) =
1 + sgn Q
2
L(t, z, Q) =
1 + sgn Q
G(t, z, Q) =
1 + sgn Q
2
+
Λ(tz,Qr)
Λ(t,Qr)
Λ(t,Qr)
Λ(tz,Qr)
Λ(tz,Qr) +
+
+
1
Λ(tz,Qr)
1 − sgn Q
2
1 − sgn Q
2
1 − sgn Q
1 − sgn Q
2
2
,
z
Λ(tz−1,Qr)
Λ(t,Qr)
z−1 Λ(t,Qr)
Λ(tz−1,Qr)
zΛ(tz−1,Qr),
z−1
1
Λ(tz−1,Qr)
,
.
Proposition 4.3. The expressions K(t, z, Q) and H(t, z, Q) are multiplicative inverses of
each other.
sgn Q(cid:19) .
K(t, 1, Q) = 1, K(1,−1, Q) = Q, K(0,−1, Q) = sgn Q, K(−1,−1, Q) = Q−1.
Λ(−t,Qr)
Λ(t,Qr)
Λ(z, sgn Q, t,Qr)
= Λ(cid:18)z,
Λ(t,Qr)
K(t, z, Q) =
H(t, z, Q) =
Λ(t,Qr)Λ(z−1, sgn Q, t,Qr)
.
Λ(tz,Qr)Λ(tz−1,Qr)
Similarly, the expressions L(t, z, Q) and G(t, z, Q) are inverses.
L(t, z, Q) = Λ(z, sgn Q, t,Qr) = Λ(t, sgn Q, z, Q).
L(t, 1, Q) = Λ(t,Qr), L(1,−1, Q) = Q, L(0,−1, Q) = 1
2 (Q+sgn Q), L(−1,−1, Q) = sgn Q.
G(t, z, Q) =
Λ(z−1, sgn Q, t,Qr)
Λ(tz,Qr)Λ(tz−1,Qr)
.
Proof. The computation is easy if we notice that in the defining formulas the coefficients of
1+sgn Q
(cid:3)
live separate lives because sgn Q is an involution.
and 1−sgn Q
2
2
SPECTRAL CALCULATIONS IN RINGS
11
The properties of K show that Λ(−t,Qr)
Λ(t,Qr) sgn Q is a homotopy from Q (t = 1) to sgn Q
(t = 0) inside ]SpeciR(A). Here the meaning of "inside" is that the whole expression satisfies
the appropriate formal spectral condition.
Proposition 4.4.
K(t, w, Q) =Z P(t, z)
H(t, w, Q) =Z P(t, z)
G(t, w, Q) =Z L(t, z)
Λ(zw, Q)
Λ(z, Q)
Λ(z, Q)
Λ(zw, Q)
1
Λ(zw, Q)
,
,
dz
2π
dz
2π
dz
2π
.
Proof. This follows from the series expansion in Proposition 3.8.
(cid:3)
Remark 4.5. In locally convex algebras, the controllability of the powers of z+z−1
possible to consider
2 makes
Λ (z, Λ(t,Qr) sgn Q) =
whose inverse turns out to be
1 + sgn Q
2
Λ(Λ(z, t),Qr) +
1 − sgn Q
2
zΛ(Λ(z−1, t),Qr),
Λ(z−1, Λ(t,Qr) sgn Q)
Λ(Λ(z, t),Qr)Λ(Λ(z−1, t),Qr)
.
This shows that Λ(t,Qr) sgn Q = 1+t
(t = −1) and sgn Q (t = 1) inside ]SpeciR(A).
4.6. Similarly, we can contract elements inside ]SpecR−(A) to 1. For S ∈ ]SpecR−(A) consider
2 Q is also a formal homotopy between Q
2 sgn Q + 1−t
C(t, S) =
√S−1√S+1
√S−1√S+1
1 + t
1 − t
2
,
pC(t, S) =
1 + t
1 − t
√S−1√S+1
√S−1√S+1
.
The substitution t 7→ −t inverts them multiplicatively.
invert:
In fact, the corresponding loops
Proposition 4.7. For S ∈ ]SpecR−(A), we have
1
Λ(w, C(t, S), w−1)
dz
2π
.
Λ(zw, S, (zw)−1)
√S
1
1
1
=
pC(t, S)Z P(t, z)
z(cid:16)(cid:0) 1+z
B−1(cid:17)
2 (cid:1)2
B−1(cid:17) dz
A−1 −(cid:0) 1−z
2 (cid:1)2
A−1 −(cid:0) 1−z
2 (cid:1)2
2π
1
z(cid:16)(cid:0) 1+z
2 (cid:1)2
1
.
√A · B =Z
Remark 4.8. In locally convex algebras, alternative contracting paths are rather trivial to
find. It is more interesting to see that the class of loops of type
remains invariant with respect to the Poisson kernel. For t = 0, they contract to the
geometric mean
12
GYULA LAKOS
5. Calculations without 1
2
As we have seen, much can be generalized to the case 1
2 ∈ A. It is natural to ask what
2 . Then only the lower portion of (5) can be generalized. Again, the
happens in the lack of 1
idempotent and the F-square-root identities are the key properties.
Definition 5.1. For P ∈ Spec 1
idem P =Z
2 +iR(A), then 1−2P is invertible, 1−P, −P
Proposition 5.2. If P ∈ Spec 1
and idem (1 − P ) = 1 − idem P , idem −P
1−2P = idem P . Furthermore,
1 − P + P z
2 +iR(A), we define
P z
dz
2π
.
(idem P )2 = idem P.
1−2P ∈ Spec 1
2 +iR(A),
Proof. The invertibility statement follows from the substitution z = −1. The identities
P + (1 − P )z = (P + (1 − P )z−1)z and (1 − −P
1−2P z = (1 − 2P )−1(1 − P + P (−z))
imply the spectral statements. The identities
=(cid:18)1 −
1−2P ) + −P
P + (1 − P )z−1(cid:19) d(z−1)
(1 − P )z−1
dz
2π
P z
2π
,
(1 − P ) + P z
−P
1−2P z
(1 − −P
1−2P ) + −P
1−2P z
dz
2π
=
P (−z)
1 − P + P (−z)
d(−z)
2π
integrated prove the first and second equalities, respectively. We can prove the idempotent
identity in several ways:
Matrix algebraic proof. We can proceed as before, but have to conjugate not the Hilbert
transform involution but the idempotentPs∈−N− 1
A direct algebraic proof. See Karoubi [2], Lemma III.1.23 -- 24.
A resolvent algebraic proof. We should prove that
es,s.
2
1 − P + P z
is equal to 0. It is natural try the proof along the steps
1 − P + P w
P z
1 − P
dz
2π
dw
2π
(15)
(16)
idem P (1 − idem P ) =ZZ
1 − P + P w ∼(cid:20) z + w
1 − P
(cid:21)
P z
P (1 − P )
=
1 − P + P z
=(cid:20) 1
2
(1 − P + P z)(1 − P + P w)
z + w
z − w(cid:21)(cid:18) (z − 1)P (1 − P )
1 − P + P z −
(w − 1)P (1 − P )
1 − P + P w (cid:19) ∼ 0,
2
except it seems to be plagued by 1
division by 2 is of superficial nature in the proof. This can be done as follows.
For a Laurent series a(z, w), we define : a(z, w) :z,w by linear extension from
2 's as before. We have to demonstrate that the use of
: znwm :z,w= zmax(n,m)wmin(n,m).
Lemma 5.3. For any Laurent series a(z, w), we have
For a symmetric Laurent series a(z, w), i. e.
such that a(z, w) = a(w, z), we define
formally
2π
ZZ a(z, w)dz
:(cid:20) z + w
dw
2π
=ZZ : a(z, w) :z,w dz
(cid:21) a(z, w) :z,w=: za(z, w) :z,w .
2π
2
dw
2π
.
SPECTRAL CALCULATIONS IN RINGS
13
Suppose that a(z, w) is anti-symmetric in its variables, included that the coefficient of
to our natural expectations. We have to check that the resulting expression is integral in
terms of the coefficients. In the present case, the definition yields by linear extension from
z+w
z−wi a(z, w) :z,w formally to be as it should be according
znwn is always 0. We define :h 1
:(cid:20) 1
z + w
2
2
z − w(cid:21) (znwm − zmwn) :z,w= znwm + 2 X0<k< n−m
2
zn−kwm+k + δ n+m
2 ∈Zz
n+m
2 w
n+m
2
,
where n > m. Checking for elements of suitable bases it is easy to see the following lemmas:
Lemma 5.4. For any symmetric Laurent series a(z, w),
Lemma 5.5. For any Laurent series a(z), we have
z + w
z − w(cid:21) (z − w)a(z, w) :z,w .
2
:(cid:20) z + w
ZZ :(cid:20) 1
(cid:21) a(z, w) :z,w=:(cid:20) 1
z − w(cid:21) (a(z) − a(w)) :z,w dz
z + w
2π
2
2
dw
2π
= 0.
Now it is easy to carry out the proof. From (15) we should pass to the "normal ordered"
(cid:3)
form, after which the subsequent manipulations as in (16), leading to 0, make sense.
Definition 5.6. For T ∈ Spec 1
4−R−(A), we define
(1 + z)T
F√T =Z
1 + (z − 2 + z−1)T
dz
2π
.
,
2
zk+z−k
Definition 5.7. Let Ahzi+ be the space of formal Laurent series a(z) = a0+P∞k=1 ak
and let A[z]+ be the space of formal Laurent series b(z) = b0 +P∞k=1 bk(zk + z−k). Similarly,
let Ahzi− be the space of formal Laurent series a(z) =P∞k=1 a−k
, and let A[z]− be
2
the space of formal Laurent series b(z) = P∞k=1 b−k(zk − z−k).
It is easy to see that
Ahzi = Ahzi+ ⊕ Ahzi− is a natural A[z]± = A[z]+ ⊕ A[z]−-module, in fact, this action is
Z2-graded. Multiplication of 1 ∈ Ahzi yields a natural map from A[z]± into Ahzi:
1 · ∞Xk=1
2bk
bk(zk + z−k)! =
b−k(zk − z−k) + b0 +
In fact, this notation can be extended to b(z) ∈ A[z, z−1] in a compatible way, by 1 · z =
, etc. Integration can be defined for elements of Ahzi or A[z]±. Again, it
singles out the 0th coefficient. We see that if a(z) ∈ Ahzi and b(z) ∈ A[z]± as above, then
2 + z−z−1
2 + b0 +
∞Xk=1
∞Xk=1
∞Xk=1
zk−z−k
zk−z−k
2b−k
zk+z−k
z+z−1
2
2
.
Z a(z) · b(z)dz
2π
=Xn∈Z
akbk.
For example,R(cid:16)1 + z+z−1
We can extend this formalism to multiple variables. The spaces Ahzihwi and A[z]±[w]±
2 (cid:17) · b(z)dz2π = b0 + b1. Furthermore,R b(z)dz2π =R 1 · b(z)dz2π .
can be considered. In the case of Ahzihwi, colloquial notation like
w−w−1
zw−1+wz−1
w+w−1
≡ z+z−1
2
2 − z−z−1
2
2
2
is allowed. However, an other space between Ahzihwi and A[z]±[w]± can be considered.
Indeed, let A{z, w} be the space of the formal combinations of the basis elements
(zn+z−n)(wn+w−n)
2
, zn−z−n
1, zn+z−n
, (zn+z−n)(wn−w−n)
2
2
2
, wn+w−n
, wn−w−n
, (zn−z−n)(wn+w−n)
2
2
2
,
, (zn−z−n)(wn−w−n)
2
,
14
GYULA LAKOS
where n, m ≥ 1. In this case, colloquial notation like
zw−1 − wz−1 ≡ (z−z−1)(w+w−1)
2
− (z+z−1)(w−w−1)
2
is allowed. There are natural A[z]±[w]±-module homomorphisms A[z]±[w]± → A{z, w} →
Ahzihwi respecting the grading.
5.8. The advantage of the terminology above is that it allows us to rewrite the definition
of the F-square-root as the "manifestly real" expression
Proposition 5.9. If T ∈ Spec 1
Furthermore,
)T
2
(1 + z+z−1
F√T =Z
4−R−(A), then −T
1 + (z − 2 + z−1)T
dz
2π
.
1−4T is invertible and Fq −T
1−4T = − F√T
1−2 F√T
.
F√T (1 − F√T ) = T.
Spec 1
(17)
Proof. Substituting z = −1 into the resolvent term, we see that (1 − 4T )−1 exists. The
identity 1 + (z − 2 + z−1) −T
1−4T ∈
4−R−(A). If (17) holds, then (1 − 2 F√T )2 = 1 − 4T , and it is sufficient to prove that
Fq −T
1−4T =
F√T−2T
1−4T . This, however, follows from the identity
(1+ (−z)+(−z)−1
1+((−z)−2+(−z)−1)T − 2T
1−4T = (1 − 4T )−1(1 + ((−z) − 2 + (−z)−1)T ) shows that −T
(1 + z+z−1
)T
2
2
=
) −T
1−4T
1 + (z − 2 + z−1) −T
1−4T
1 − 4T
integrated. So, what we have to show is the F-square-root identity (17). Now, as
2
)T
(18) F√T (1− F√T )−T =ZZ
1 + (z − 2 + z−1)T 1 −
(1 + z+z−1
(1 + z+z−1
1 + (z − 2 + z−1)T 1 −
1 + (w − 2 + w−1)T! − T =
(1 + w+w−1
(19)
)T
)T
2
2
(1 + w+w−1
1 + (w − 2 + w−1)T!−T dz
2π
)T
2
dw
2π
,
we should show that this latter term is 0. It would be natural to proceed along the steps
2
2
2 + w+w−1
2 − z+z−1
= (cid:16) z+z−1
∼ (cid:18)(cid:20) z+z−1
z − w(cid:21)(cid:16)(cid:16) z−z−1
=(cid:20) 1
z − w(cid:21)
=(cid:20) 1
z + w
z + w
2
2
2
2
2
w+w−1
(1 + (z − 2 + z−1)T )(1 + (w − 2 + w−1)T )
T + T + z+z−1
2 T − w+w−1
(cid:21) (1 − 2T ) +(cid:16)1 + z−1w+zw−1
T(cid:17) T (1 − 4T )
(cid:17) T(cid:19) T (1 − 4T )
(1 + (z − 2 + z−1)T )(1 + (w − 2 + w−1)T )
2 − w−w−1
(cid:17) (1 − 2T ) + (zw−1 − z−1w)T(cid:17) T (1 − 4T )
(1 + (z − 2 + z−1)T )(1 + (w − 2 + w−1)T )
2
2
z−z−1
2
1 + (z − 2 + z−1)T ) −
w−w−1
1 + (w − 2 + w−1)T )! T (1 − 4T ) ∼ 0,
2
except we have to demonstrate that the use of 1
2 is superficial.
Another reordering operation can be defined according to
:: znwm ::z,w= zmax(n,m)+z− max(n,m)
2
wmin(n,m)+w− min(n,m)
2
.
SPECTRAL CALCULATIONS IN RINGS
15
This applies to our standard A[z]±[w]±-modules. In the context of Ahzihwi, it leaves only
the (++)-graded parts and reorders them.
Lemma 5.10. (a) For any Laurent series a(z, w) ∈ Ahzihwi, we have
dw
.
2π
=ZZ :: a(z, w) ::z,w dz
ZZ a(z, w)dz
b(z, w) = b(w, z), then :: a1(z, w)b(z, w) ::z,w=:: a2(z, w)b(z, w) ::z,w.
(b) If :: a1(z, w) ::z,w=:: a2(z, w) ::z,w and b(z, w) ∈ A[z]+[w]+ is symmetric, i. e.
For any symmetric Laurent series b(z, w) ∈ A[z]+[w]+, i. e. such that b(z, w) = b(w, z),
dw
2π
2π
2π
we define formally
::" z+z−1
2
2 + w+w−1
2
# b(z, w) ::z,w=:: z+z−1
2
b(z, w) ::z,w .
2
2
2
2
2
z+w
z+w
cn,0 = zn−z−n
− wn−w−n
Let us consider a Laurent series c(z, w) ∈ (A{z, w})− such that it is antisymmetric, i. e.
and cn,m = (zn−z−n)(wm+w−m)
c(z, w) = −c(w, z). Then c(z, w) is a formal linear combination of the basis elements
− (wn−w−n)(zn+z−n)
,
z−wi c(z, w) ::z,w according
to our natural expectations. Again, we have to check that the result is integral in terms of
coefficients of c(z, w). We just give some samples in the table
where n, m ≥ 1. For such c(z, w), we can formally define ::h 1
::h 1
z−wi cn,m ::z,w m = 0
n = 1
1 + 2d1
n = 2
0
1 + d0
2 + 2d1
n = 3
0
where dn
m =:: znwm ::z,w. In fact, at first sight, it looks more natural to choose c(z, w) from
the antisymmetric elements of (Ahzihwi)− = Ahzi+hwi−⊕ Ahzi−hwi+, but it turns out that
the coefficients in ::h 1
Lemma 5.11. For any symmetric Laurent series b(z, w) ∈ A[z]+[w]+, we have
m = 1
1 + d0
d1
0
1 + 2d1
2d2
0
1 + 2d2
2 + d1
1 + d2
2d1
0
1 + d0
2 + 2d1
d2
0
1 + 2d1
2 + 2d2
0 2d3
m = 3
2 + d1
−d2
2d2
3 + 2d2
d1
0
0 + d1
d2
1
1 2d3
d3
0 + 2d2
1 + 2d2
0
0 d3
m = 2
2
2
2
z+w
z + w
z−wi c(z, w) ::z,w would fail to be integral.
# b(z, w) ::z,w=::(cid:20) 1
(cid:19) b(z, w) ::z,w=::(cid:20) 1
z − w(cid:21) (a(z) − a(w)) ::z,w dz
2 − w−w−1
z − w(cid:21)(cid:16) z−z−1
(cid:17) b(z, w) ::z,w,
z − w(cid:21)(cid:0)zw−1 − wz−1(cid:1) b(z, w) ::z,w .
dw
2π
z + w
z + w
= 0.
2π
2
2
ZZ ::(cid:20) 1
2
2
Lemma 5.12. For any Laurent series a(z) ∈ Ahzi−, we have
::" z+z−1
::(cid:18)1 +
Now, it is easy to carry out the proof. From (18), we should pass to the "normal ordered"
(cid:3)
form, after which the subsequent manipulations leading to 0 make sense.
Proposition 5.13. P ∈ Spec 1
2 +iR(A) if and only if P (1−P ) ∈ Spec 1
4−R−(A). Furthermore,
2 + w+w−1
2
2
zw−1 + wz−1
and
Consequently,
FpP (1 − P ) = P + idem P − 2P idem P.
idem P =
FpP (1 − P ) − P
1 − 2P
.
16
GYULA LAKOS
Proof. The decomposition 1 + (z − 2 + z−1)P (1− P ) = (1− P + P z)(1− P + P z−1) implies
the spectral statement. The equality follows from the identity
1 + (z − 2 + z−1)P (1 − P )
integrated.
= P +
P z
1 − P + P z
(1 − 2P )
2 (cid:17) P (1 − P )
+
z−z−1
2 P (1 − P )(1 − 2P )
1 + (z − 2 + z−1)P (1 − P )
(cid:16)1 + z+z−1
Then we let PF = FpP (1 − P ). Further statements can be proven parallel to the case
(z2 + z−2) + . . .
1 − 2 F√T 1 + − F√T
(z + z−1) + − F√T
1 − F√T!2
2 , except the formulas are less customary. E. g., the analogue of Proposition 3.9 is
with 1
Proposition 5.14. For T ∈ Spec 1
1 + (z − 2 + z−1)T
and
4−R−(A), we have
1
1
=
(cid:3)
1 − F√T
=Z
1
1 − F√T
1 + z
1 + (z − 2 + z−1)T
References
dz
2π
.
[1] M. Haase: The Functional Calculus for Sectorial Operators. Operator Theory: Advances and Appli-
cations 169. Birkhauser Verlag, Basel, 2006.
[2] M. Karoubi: K-theory. An introduction. Grundlehren der mathematischen Wissenschaften, 226.
Springer Verlag, Berlin, Heidelberg, New York, 1978.
[3] C. Mart´ınez Carracedo, M. Sanz Alix: The theory of fractional powers of operators. North-Holland
Mathematics Studies, 187. North-Holland, Amsterdam, 2001.
Department of Geometry, Eotvos University, P´azm´any P´eter s. 1/C, Budapest, H -- 1117,
Hungary
E-mail address: [email protected]
|
1611.05718 | 1 | 1611 | 2016-11-17T14:53:24 | Linear commuting maps and skew-symmertric biderivations of the deformative Schrodinger-Virasoro Lie algebras | [
"math.RA"
] | In this paper, we investigate the skew-symmertric biderivations of the deformative Schrodinger-Virasoro Lie algebras which contain the twisted and original deformative Schrodinger-Virasoro Lie algebras. As an application, we give the explicit form of each linear commuting map on the deformative Schrodinger-Virasoro Lie algebras. In particular, we obtain that there exist non-inner biderivations and non-standard linear commuting maps for the certain deformative Schrodinger-Virasoro Lie algebras. | math.RA | math |
Linear commuting maps and skew-symmertric biderivations of the
deformative Schrodinger-Virasoro Lie algebras
Guangzhe Fan ∗,†, Yucai Su ∗,¶, Kun Xu ∗,§
†[email protected]
¶[email protected]
§[email protected]
∗School of Mathematical Sciences, Tongji University, Shanghai 200092, P. R. China
Abstract: In this paper, we investigate the skew-symmertric biderivations of the deformative Schrodinger-Virasoro
Lie algebras which contain the twisted and original deformative Schrodinger-Virasoro Lie algebras. As an application,
we give the explicit form of each linear commuting map on the deformative Schrodinger-Virasoro Lie algebras. In
particular, we obtain that there exist non-inner biderivations and non-standard linear commuting maps for the certain
deformative Schrodinger-Virasoro Lie algebras.
Key words: biderivations, commuting maps, Schrodinger-Virasoro Lie algebra, deformative Schrodinger-Virasoro
Lie algebras
Mathematics Subject Classification (2010): 17B05, 17B40, 17B65, 17B68.
1
Introduction
Throughout this paper, we denote by Z, C the sets of integers, complex numbers respectively. We
assume that all vector spaces are based on C, unless otherwise stated.
It is well known that the infinite-dimensional Schrodinger Lie algebras and the Virasoro al-
gebra play important roles in many areas of mathematics and physics.
In order to investigate
the free Schrodinger equations, the twisted and original Schrodinger-Virasoro Lie algebras were
introduced by [10] in the context of non-equilibrium statistical physics. In [19], the author intro-
duced the deformations of the Schrodinger-Virasoro Lie algebras. In this paper, we consider the
following Lie algebras, which are referred to as deformative Schrodinger-Virasoro Lie algebras
L (l , m , s). Fix the complex numbers l , m and s = 0 or 1
2, the Lie algebra L (l , m , s) has C-basis
{Ln, Mn,Yn+s n ∈ Z} with the following nontrivial Lie brackets:
[Ln, Lm] = (m − n)Ln+m,
[Ln, Mm] = (m − l n + 2m )Mn+m,
[Ln,Ym+s] = (m + s −
l + 1
2
n + m )Yn+m+s,
(1.1)
(1.2)
(1.3)
[Yn+s,Ym+s] = (m − n)Mn+m+2s,
(1.4)
where n, m ∈ Z. Obviously, L (l , m , s) is a generalization of the Schrodinger-Virasoro Lie alge-
bras. Note that L (l , m , s) contains the Schrodinger-Virasoro Lie algebras and their deformations.
For instance,
Supported by the National Natural Science Foundation of China (No. 11431010, 11371278).
§Corresponding author: Kun Xu.
1
• L (0, 0, s) is the well-known Schrodinger-Virasoro Lie algebra introduced in [10]. Its struc-
tures and representations have been widely studied in [8, 15, 21].
• L (l , m , 0) is called the twisted deformative Schrodinger-Virasoro Lie algebras whose struc-
ture theories studied in [17, 22].
• L (l , m , 1
2) is called the original deformative Schrodinger-Virasoro Lie algebras. In [13,16],
the authors obtained some results about structures.
• L (l , 0, s) is also a class of the deformative Schrodinger-Virasoro Lie algebras. In [5], the
author investigated the Lie bialgebra structures.
As is well known, derivations and generalized derivations are very important subjects in the re-
search of both algebras and their generalizations. In recent years, biderivations have aroused many
scholars' great interests in [1 -- 4, 6, 7, 9, 20, 21, 23, 24]. In [2], Bre sar showed that all biderivations
on commutative prime rings are inner bidrivations and they also determined the biderivations of
semiprime rings. The notation of biderivation of Lie algebras was introduced in [20]. In addition,
in [3, 21] the authors obtained that the skew-symmetric biderivations of the Schrodinger-Virasoro
algebra and a simple generalized Witt algebra are inner biderivations. Furthermore, in [9] the
authors determined all the skew-symmetric biderivations of W (a, b) and found that there exist
non-inner biderivations. In 1957, the first important result on linear (or additive) commuting maps
was introduced by Posners in [18]. The author of [2] described that commuting maps on an as-
sociative algebra have significant application to other important problems (e.g., Lie derivations,
biderivations, etc). Moreover, linear commuting maps of some Lie algebras were investigated
in [3, 4, 9, 21]. Recently, the theory of biderivations and linear commuting maps have become one
of research focuses in the Lie theory. Motivated by this reason, we attempt to investigate the linear
commuting maps and skew-symmetric biderivations of some important Lie algebras. We know that
the linear commuting maps and skew-symmetric biderivations of L (0, 0, 0) were studied in [21].
Thus, the results of this paper are more generic.
This paper is organized as follows.
In Section 2, we review some definitions and conclu-
sions on biderivations of Lie algebras. In Section 3, we compute the skew-symmetric bideriva-
tions of L (l , m , s). In particular, we obtain that there exist non-inner biderivations for the certain
L (l , m , s). In Section 4, we give the explicit form of each linear commuting map on L (l , m , s).
2 Preliminaries and main results
Firstly, we shall recall some definitions and conclusions about biderivations of Lie algebras in
[3, 21]. In this section, we assume that L is a Lie algebra over C.
Definition 2.1. A bilinear map f : L × L −→ L is called skew-symmetric if f (x, y)= −f (y, x) for all
x, y ∈ L.
2
Definition 2.2. A bilinear map f : L × L −→ L ia called a biderivation if it satisfies the following
two axioms:
f ([x, y], z) = [x,f (y, z)] + [f (x, z), y]
f (x, [y, z]) = [f (x, y), z] + [y,f (x, z)]
for any x, y, z ∈ L.
Example 2.3. Let l ∈ C, then the map f l : L × L −→ L, sending (x, y) to l [x, y], is a biderivation
of L. All biderivations of this kind are called inner biderivations of L.
Remark 2.4. It is straight to check that every inner biderivation f l
tion.
is a skew-symmetric bideriva-
By [3, 21], the following two results are straightforward to verify.
Lemma 2.5. Let f be a skew-symmetric biderivation on L, then
[f (x, y), [u, v]] = [[x, y],f (u, v)]
for any x, y, u, v ∈ L. In particular, [f (x, y), [x, y]] = 0.
Lemma 2.6. Let f be a skew-symmetric biderivation on L. If [x, y] = 0, then f (x, y) ∈ CL([L, L]),
where CL([L, L]) is the centralizer of [L, L].
3 Skew-symmertric biderivations of L
In this section, we would like to compute the skew-symmetric biderivations of L (l , m , s), which
is denoted by L for simplicity.
Lemma 3.1.
(1) The center Z(L ) of L is given by
Z(L ) =( CM−2m
0
if l = 0, m ∈ 1
2
otherwise.
Z;
(2) For x ∈ L , denote x = x + [L , L ]. Then
L /[L , L ] =( CY −m
0
if l = −3, m ∈ s + Z;
otherwise.
(3) The centralizer of [L , L ] coincides with Z(L ), i.e., CL ([L , L ]) = Z(L ).
Proof. It can be easily obtained by the Lie brackets of L .
3
For convenience, we first introduce two kinds of skew-symmetric biderivations.
(1) For L with l = 1, m ∈ s + 1
2
Z, define the following skew-symmetric bilinear map:
f 0 : L × L → L ,
(Ln, Lm) 7→ (m − n)Mn+m−2m ,
(3.1)
the others map to 0.
(2) For L with l = 1, m ∈ s + Z, define the following skew-symmetric bilinear map:
f 1 : L × L → L
(Ln, Lm) 7→ (m − n)Yn+m−m ,
(Ln,Ym+s) 7→ (m + s − n + m )Mn+m+s−m ,
(Ym+s, Ln) 7→ (n − m − s − m )Mn+m+s−m ,
(3.2)
the others map to 0.
Obviously, we see that f 0 and f 1 are skew-symmetric non-inner biderivations of L .
Theorem 3.2. Let f be a skew-symmetric biderivation of L , then we have
a [x, y] + bf
a [x, y] + bf
a [x, y]
0(x, y)
0(x, y) + gf 1(x, y)
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
otherwise,
2 + Z;
f (x, y) =
for all x, y ∈ L , where a ,b ,g ∈ C, f 0 and f 1 are given by (3.1) and (3.2).
Proof. We shall complete the proof by verifying the following ten claims.
Claim 1. There exist a ,b ,g ∈ C such that
f (Ln, Lm) ≡
for all n, m ∈ Z, where c−m ∈ C.
a [Ln, Lm] + c−m Y−m (modZ(L ))
a [Ln, Lm] + b (m − n)Mn+m−2m (modZ(L ))
a [Ln, Lm] + b (m − n)Mn+m−2m
+g (m − n)Yn+m−m (modZ(L ))
a [Ln, Lm](modZ(L ))
if l = −1, m ∈s + Z;
if l = 1, m ∈ s + 1
2 + Z;
if l = 1, m ∈s + Z;
otherwise,
If n = m, then the fact [Ln, Lm] = 0 gives that f (Ln, Lm) ∈ Z(L ) by Lemmas 2.6 , 3.1(3).
For n 6= m, suppose that f (Ln, Lm) = (cid:229)
i∈Z(aiLi + biMi + ci+sYi+s) for some ai, bi, ci+s ∈ C.
According to Lemma 2.5, one has
0 = 1
m−n[[Ln, Lm],f (Ln, Lm)]
=hLn+m, (cid:229)
i∈Z(cid:16)ai(i − n − m)Ln+m+i + bi(i − l (n + m) + 2m )Mn+m+i
(aiLi + biMi + ci+sYi+s)i
= (cid:229)
i∈Z
+ci+s(i + s −
l +1
2 (n + m) + m )Yn+m+i+s(cid:17),
4
which follows that
i f i 6= n + m,
i f i 6= l (n + m) − 2m ,
ai = 0
bi = 0
ci+s = 0 i f i + s 6=
l +1
2 (n + m) − m .
Furthermore, by Lemma 2.5 and the equation (3.3), for any k 6= 0, then we get
0 = [f (Ln, Lm), [Lk, L0]] − [[Ln, Lm],f (Lk, L0)]
= [an+mLn+m + bl (n+m)−2m Ml (n+m)−2m + c l +1
−[(m − n)Ln+m, akLk + bkl −2m Mkl −2m + c l +1
2 (n+m)−m Yl +1
2 k−m Yl +1
2 k−m ].
2 (n+m)−m , −kLk]
Thus,
k(n + m − k)an+mLn+m+k = (m − n)(k − n − m)akLn+m+k,
kl (n + m − k)bl (n+m)−2m Ml (n+m)−2m +k = l (m − n)(k − n − m)bkl −2m Mn+m+kl −2m ,
l + 1
k(n + m − k)c l +1
2 (n+m)−m Yl +1
2 (n+m)−m +k
2
l + 1
2
=
(m − n)(k − n − m)c l +1
2 k−m Yn+m+
2 k−m .
l +1
(3.3)
(3.4)
(3.5)
(3.6)
Due to the arbitrariness of k, we can conclude an+m = a (m − n) from (3.4) for some a ∈ C.
In the following we shall consider the coefficients of M and Y in (3.5) and (3.6). It can be
divided into four cases by choosing different l .
Case 1. l = 0. One can deduce that c l +1
2 (n+m)−m = 0 by the arbitrariness of k. If m ∈ 1
2
Z, then
we get f (Ln, Lm) = a [Ln, Lm] + b−2m M−2m ; otherwise, f (Ln, Lm) = a [Ln, Lm].
Case 2. l = −1. It is easy to get that bl (n+m)−2m = 0. If m /∈ s + Z, then c−m = 0. Furthermore,
we obtain that f (Ln, Lm) = a [Ln, Lm] + c−m Y−m
if m ∈ s + Z.
Case 3.
l = 1.
If m ∈ s + Z, then we have f (Ln, Lm) = a [Ln, Lm] + bn+m−2m Mn+m−2m +
cn+m−m Yn+m−m . Moreover, if m ∈ s + 1
2 + Z, one has f (Ln, Lm) = a [Ln, Lm] + bn+m−2m Mn+m−2m .
(n − m) and cn+m−m =
Comparing the coefficients in (3.5) and (3.6), we conclude bn+m−2m =
k (n −m) by the arbitrariness of k. Hence, we could assume bn+m−2m = b (m −n) and cn+m−m =
ck−m
g (m − n) for some b ,g ∈ C. We have obtained the desired result.
6= 0, ±1. We also obtain that f (Ln, Lm) = a [Ln, Lm].
Case 4. l
bk−2m
k
Claim 2.
f (Ln, Mm) ≡( a [Ln, Mm] + f−m Y−m (modZ(L ))
a [Ln, Mm](modZ(L ))
if l = −1,
otherwise,
m ∈ s + Z;
for all n, m ∈ Z, where f−m ∈ C.
5
For any fixed n, m ∈ Z, suppose f (Ln, Mm) =(cid:229)
i∈Z(diLi + eiMi + fi+sYi+s) for some di, ei, fi+s ∈
C. By Lemma 2.5 and Claim 1, for any k 6= 0, we have
0 = [f (Ln, Mm), [L0, Lk]] − [[Ln, Mm],f (L0, Lk)]
=h (cid:229)
i∈Z(cid:16)k(k − i)diLi+k − k(i − kl + 2m )eiMi+k − k(i + s −
+ka (m − l n + 2m )(n + m − kl + 2m )Mn+m+k,
(diLi + eiMi + fi+sYi+s), kLki − [(m − l n + 2m )Mn+m, ka Lk]
= (cid:229)
i∈Z
l +1
2 k + m ) fi+sYi+s+k(cid:17)
which implies that
di = 0 if i 6= k,
(i − kl + 2m )ei = 0 if i 6= n + m,
(n + m − kl + 2m )(ei − a (m − l n + 2m )) = 0 if i = n + m,
fi+s = 0 if i + s 6=
l + 1
2
k − m .
(3.7)
(3.8)
(3.9)
(3.10)
Firstly, one can easily get that di = 0 by the arbitrariness of k in (3.7). If l
ei = 0 if i 6= n + m and en+m = a (m − l n + 2m ) by (3.8) and (3.9). If l
fi+s = 0 from (3.10). Thus,
6= 0, one shows that
6= −1, we obtain that
Case 1. l = −1. We have f (Ln, Mm) = a [Ln, Mm] + f−m Y−m . If m /∈ s + Z, then f−m = 0.
Case 2. l = 0. According to (3.8) and (3.9), we have f (Ln, Mm) = e−2m M−2m
if n + m = −2m .
if n + m 6= −2m . Thanks to Lemma 3.1(1), this
Furthermore, f (Ln, Mm) = a [Ln, Mm] + e−2m M−2m
claim holds.
Case 3.
l
6= 0, −1. One shows that f (Ln, Mm) = a (m − l n + 2m )Mn+m = a [Ln, Mm]. This
completes the claim.
Claim 3.
f (Ln,Ym+s) ≡
l−m Y−m (modZ(L )) when n + m + s = −m
a [Ln,Ym+s]+l−m Y−m (modZ(L )) when n+m+s 6= −m
a [Ln,Ym+s] + g (m + s − n + m )Mn+m+s−m (modZ(L ))
a [Ln,Ym+s](modZ(L ))
if l = −1, m ∈ s + Z;
if l = −1, m ∈ s + Z;
if l = 1, m ∈ s + Z;
otherwise,
for n, m ∈ Z, where l−m ∈ C.
For any fixed n, m ∈ Z, suppose f (Ln,Ym+s) = (cid:229)
i∈Z(giLi +hiMi +li+sYi+s) for some gi, hi, li+s ∈
C.
6
Case 1. l = −1, m ∈ s + Z. According to Lemma 2.5 and Claim 1, for any k 6= 0, then we get
0 = [f (Ln,Ym+s), [L0, Lk]] − [[Ln,Ym+s],f (L0, Lk)]
=h (cid:229)
i∈Z(cid:16)k(k − i)giLi+k − k(i + k + 2m )hiMi+k − k(i + s + m )li+sYi+s+k(cid:17)
+c−m (m + s + m )(n + m + s + m )Mn+m+s−m + ka (m + s + m )(n + m + s + m )Yn+m+s+k.
(giLi + hiMi + li+sYi+s), kLki − [(m + s + m )Yn+m+s, ka Lk + c−m Y−m ]
= (cid:229)
i∈Z
Firstly, by the arbitrariness of k, n and m, one can easily get from the above formula that
c−m = gi = hi = 0,
(i + s + m )li+s = 0 if i 6= n + m,
(n + m + s + m )(li+s − a (m + s + m )) = 0 if i = n + m.
(3.11)
(3.12)
(3.13)
if n + m + s = m , and f (Ln,Ym+s) =
By (3.12) and (3.13), we conclude that f (Ln,Ym+s) = l−m Y−m
a [Ln,Ym+s] + l−m Y−m
if n + m + s 6= −m .
l = 1, m ∈ s + Z. Using the similar method of Case 1, we conclude the following
Case 2.
equalities:
gi = 0 if i 6= k,
(i − k + 2m )hi = 0 if i 6= n + m + s − m ,
(n + m + s − k + m )(hi − g (m − n + s + m )) = 0 if i = n + m + s − m ,
(i + s − k + m )li+s = 0 if i 6= n + m,
(n + m + s − k + m )(li+s − a (m − n + s + m )) = 0 if i = n + m.
(3.14)
(3.15)
(3.16)
(3.17)
(3.18)
Obviously, it can be easily obtained that gi = 0 by (3.14). According to (3.15)-(3.18), we have
f (Ln,Ym+s) = a [Ln,Ym+s] + g (m − n + s + m )Mn+m+s−m .
Case 3. Otherwise. By Lemma 2.5 and Claim 1, for any k 6= 0, we have
0 = [f (Ln,Ym+s), [L0, Lk]] − [[Ln,Ym+s],f (L0, Lk)]
l +1
2 n + s + m )Yn+m+s, ka Lk]
=h (cid:229)
(giLi + hiMi + li+sYi+s), kLki − [(m −
i∈Z(cid:16)k(k − i)giLi+k − k(i − l k + 2m )hiMi+k − k(i + s −
+ka (m −
l +1
l +1
2 k + m )Yn+m+s+k,
l +1
2 n + s + m )(n + m + s −
2 k + m )li+sYi+s+k(cid:17)
= (cid:229)
i∈Z
which implies that
gi = 0 if i 6= k,
(i − kl + 2m )hi = 0 if i 6= kl − 2m ,
(i + s −
l + 1
2
(n + m + s −
2
k + m )li+s = 0 if i 6= n + m,
l + 1
k + m )(li+s − a (m −
l + 1
2
7
(3.19)
(3.20)
(3.21)
n + s + m )) = 0 if i = n + m.
(3.22)
If l = 0, we obtain that f (Ln,Ym+s) =
We get that gi = 0 by the arbitrariness of k in (3.19).
a [Ln,Ym+s] + h−2m M−2m . If l
6=
−1 or m /∈ s + Z, we get li+s = 0 if i 6= n +m by (3.21) and (3.22). Thus, f (Ln,Ym+s) = a [Ln,Ym+s].
Claim 4.
6= 0, by the arbitrariness of k in (3.20), it follows hi = 0. Since l
f (Yn+s,Ym+s) ≡( a [Yn+s,Ym+s] + r−m Y−m (modZ(L ))
a [Yn+s,Ym+s](modZ(L ))
if l = −1, m ∈ s + Z;
otherwise,
for all n, m ∈ Z, where r−m ∈ C.
For any fixed n, m ∈ Z, assume f (Yn+s,Ym+s) = (cid:229)
i∈Z(piLi +qiMi +ri+sYi+s) for some pi, qi, ri+s ∈
C. By Lemma 2.5 and Claim 1, for any k 6= 0, we have
0 = [f (Yn+s,Ym+s), [L0, Lk]] − [[Yn+s,Ym+s],f (L0, Lk)]
=h (cid:229)
i∈Z(cid:16)k(k − i)piLi+k − k(i − kl + 2m )qiMi+k − k(i + s −
+ka (m − n)(n + m + s − kl + 2m )Mn+m+2s+k,
(piLi + qiMi + ri+sYi+s), kLki − [(m − n)Mn+m+2s, ka Lk]
= (cid:229)
i∈Z
l +1
2 k + m )ri+sYi+s+k(cid:17)
which implies that
pi = 0 if i 6= k,
(i − kl + 2m )qi = 0 if i 6= n + m + 2s,
(n + m + 2s − kl + 2m )(qi − a (m − n)) = 0 if i = n + m + 2s,
ri+s = 0 if i + s 6=
l + 1
2
k − m .
(3.23)
(3.24)
(3.25)
(3.26)
Hence, pi = 0 by (3.23). If l
from (3.24) and (3.25)). If l
6= 0, then we have qi = 0 if i 6= n + m + 2s and qn+m+2s = a (m − n)
6= −1, we obtain that ri+s = 0 by the arbitrariness of k in (3.26). Thus,
Case 1. l = −1. It has f (Yn+s,Ym+s) = a [Yn+s,Ym+s] + r−m Y−m . If m /∈ s + Z, then r−m = 0.
Case 2. l = 0. We have f (Yn+s,Ym+s) = q−2m M−2m
if n +m +2s = −2m , and f (Yn+s,Ym+s) =
a [Yn+s,Ym+s] + q−2m M−2m
if n + m + 2s 6= −2m .
6= 0, −1. Thus, f (Yn+s,Ym+s) = a [Yn+s,Ym+s]. This claim holds.
Case 3. l
Now we would like to simplify the forms of Claims 1 to 4.
If l = −1 and m ∈ s + Z. On one hand, for any n, m ∈ Z, since f
is a biderivation, we have
f (Ln, [Ys,Ym+s]) = [f (Ln,Ys),Ym+s] + [Ys,f (Ln,Ym+s)].
(3.27)
If m 6= 0, n + s 6= −m and n + m + s 6= −m , from Claims 2, 3 and (3.27), then we get that
a m(n + m + 2s + 2m )Mn+m+2s + m f−m Y−m
= a (s + m )(m − n)Mn+m+2s + (m + s + m )l−m Mm+s−m
+a (n + m)(m + s + m )Mn+m+2s − (s + m )l−m Ms−m ,
8
which infers that
f−m = l−m = 0.
(3.28)
If m 6= 0 and n + m + s = −m , by Claims 2, 3 and (3.27), it is enough to see that
a m(s + m )Ms−m = a (s + m )(m − n)Ms−m − (s + m )l−m Ms−m ,
which implies that l−m = −a n if m
6= −s. Thus, one shows that
f (Ln,Ym+s) =( lsYs
a [Ln,Ym+s]
if n + m = 0;
if n + m 6= 0,
for any n, m ∈ Z.
On the other hand, for any n, m ∈ Z, we have
f ([L0,Yn+s],Ym+s) = [L0,f (Yn+s,Ym+s)] + [f (L0,Ym+s),Yn+s].
If m 6= 0 and n 6= −s − m , according to Claim 4, (3.29) and (3.30), we can easily get that
r−m = 0.
If m = 0 and n 6= 0, by Claim 4, (3.29)-(3.31), one shows that
−a n(n + s + m )Mn+2s = −a n(n + 2s + 2m )Mn+s + nlsMn+s.
which implies that ls = a (s + m ). Hence, for n, m ∈ Z, we have
f (Ln,Ym+s) = a [Ln,Ym+s].
Claim 5. There exist a ,b ,g ∈ C such that
(3.29)
(3.30)
(3.31)
(3.32)
a [Ln, Lm] + bf
a [Ln, Lm] + bf
a [Ln, Lm](modZ(L ))
0(Ln, Lm)(modZ(L ))
0(Ln, Lm) + gf 1(Ln, Lm)(modZ(L ))
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
otherwise,
2 + Z;
f (Ln, Lm) ≡
for all n, m ∈ Z, where f 0 and f 1 are given by (3.1) and (3.2).
It is obvious by Claim 1 and (3.11).
Claim 6. f (Ln, Mm) ≡ a [Ln, Mm](modZ(L )) for all n, m ∈ Z.
It is easily obtained by Claim 2 and (3.28).
Claim 7.
f (Ln,Ym+s) ≡( a [Ln,Ym+s] + gf 1(Ln,Ym+s)(modZ(L ))
a [Ln,Ym+s](modZ(L ))
if l = 1, m ∈ s + Z;
otherwise,
for all n, m ∈ Z, where f 1 is given by (3.2).
9
It can be easily obtained by Claim 3 and (3.32).
Claim 8. f (Yn+s,Ym+s) ≡ a [Yn+s,Ym+s](modZ(L )) for n, m ∈ Z.
It is obvious by Claim 4 and (3.31).
Claim 9. f (Mn, Mm) ≡ 0(modZ(L )) for n, m ∈ Z.
For any fixed n, m ∈ Z, since [Mn, Mm] = 0, it is enough to see that f (Mn, Mm) ∈ CL ([L , L ])
by Lemma 2.6. Thus, this claim holds by Lemma 3.1(3).
Claim 10. f (Mn,Ym+s) ≡ 0(modZ(L )) for n, m ∈ Z.
It can be obtained by using the similar method of claim 9.
Now, by Claims 5-10, for any x, y ∈ L , we have
f (x, y) =
6= 0 or m /∈ 1
2
If l
consider the case l = 0, m ∈ 1
2
a [x, y] + q (x, y)M−2m , where q
is the zero function. In fact, by
a [x, y] + bf
a [x, y] + bf
a [x, y](modZ(L ))
0(x, y)(modZ(L ))
0(x, y) + gf 1(x, y)(modZ(L ))
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
otherwise.
2 + Z;
(3.33)
Z, we get Z(L ) = 0 by Lemma 3.1(1). Hence, Theorem 3.2 holds. Then,
Z. By Lemma 3.1(1) and (3.33), we may assume that f (x, y) =
is a bilinear function from L × L to C. We need to show that q
f ([x, y], z) = [x,f (y, z)] + [f (x, z), y],
we have that q ([x, y], z) = 0 for all x, y, z ∈ L . Note that L = [L , L ] in this case by Lemma
3.1(2). It follows that q
is exactly the zero function, as desired.
4 Linear commuting maps on L
Let A be an associative algebra, a map y
: A → A is called a commuting map if y (x)x = xy (x)
for all x ∈ A . If we denote [x, y] = xy − yx for x, y ∈ A , then a commuting map y on A can also
be defined as [y (x), x] = 0 for all x ∈ A .
Definition 4.1. Let L be a Lie algebra, a map y
: L → L is called commuting if
[y (x), x] = 0 ∀ x ∈ L.
Definition 4.2. Define the following map
y (x) = a x + f (x),
∀ x ∈ L
on L , where a ∈ C,
f : L → Z(L ). Obviously, y (x) is a linear commuting map. We call
such a map a standard linear commuting map on L . Other linear commuting maps are called
non-standard.
10
Now we apply Theorem 3.2 to describe linear commuting maps on the deformative Schrodinger-
Virasoro Lie algebras L . Obviously, the identity map is a standard linear commuting map. For
L with l = 0, m ∈ 1
Z and some linear function f from L to C, by Lemma 3.1(1), the map
2
x 7→ f (x)M−2m
is also a standard linear commuting map.
For convenience, we first introduce two kinds of linear commuting maps.
(1) For L with l = 1, m ∈ s + 1
2
Z, define the following linear map:
y 0 : L → L , Ln 7→ Mn−2m ,
the others map to 0.
(2) For L with l = 1, m ∈ s + Z, define the following linear map:
y 1 : L → L , Ln 7→ Yn−m , Yn+s 7→ Mn+s−m ,
(4.1)
(4.2)
others map to 0.
It can be easily checked that y 0 and y 1 are non-standard linear commuting maps.
Theorem 4.3. Each linear commuting map y on L is one of the following forms:
a (x) + by
0(x)
0(x) + gy
a (x) + by
a (x) + f (x)M−2m
a (x)
1(x)
2 + Z;
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
if l = 0, m ∈ 1
2
otherwise,
Z;
y (x) =
for any x ∈ L , where a ,b ,g ∈ C, y 0 and y 1 are given by (4.1) and (4.2), and f is a linear function
form L to C.
Proof. We assume that y
is a linear commuting map on L . Define
f
: L × L → L (x, y) 7→ [y (x), y] ∀ x, y ∈ L .
(4.3)
Thus, one has
f (x, [y, z]) = [f (x, y), z] + [y,f (x, z)] ∀ x, y, z ∈ L .
Obviously, f
conclude that f
of L . From (4.3), we know that f
a ,b ,g ∈ C such that
is a derivation with respect to the second component. Since [y (x), y] = [x,y (y)], we
is a biderivation
is skew-symmetric. According to Theorem 3.2, there exist
is also a derivation with respect to the first component. Thus, f
a [x, y] + bf
a [x, y] + bf
a [x, y]
f (x, y) =
0(x, y)
0(x, y) + gf 1(x, y)
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
otherwise,
2 + Z;
11
for all x, y ∈ L , where f 0 and f 1 are given by (3.1) and (3.2). Furthermore, by (4.3), then we
obtain that
bf
bf
0
0(x, y)
0(x, y) + gf 1(x, y)
[y (x) − a x, y] =
2 + Z. By the definition of f 0 given by (3.1), we conclude that y (x) −
if l = 1, m ∈ s + 1
if l = 1, m ∈ s + Z;
otherwise.
2 + Z;
Case 1. l = 1, m ∈ s + 1
a x = by
0(x). The conclusion holds.
Case 2. l = 1, m ∈ s + Z. By the definition of f 0 and f 1 given by (3.1) and (3.2), one follows
that y (x) − a x = by
0(x) + gy
Case 3. l
6= 1 or m /∈ s + 1
2
1(x).
Z. Obviously, we have y (x) − a x ∈ Z(L ). By Lemma 3.1(1), we
Z. Define f by letting y (x) − a x = f (x)M−2m , then f
only need to consider the case l = 0, m ∈ 1
2
is a linear function from L to C, and y (x) = a x + f (x)M−2m . This completes the proof.
References
[1] D. Benkovi c. "Biderivations of triangular algebras," Lin. Alg. Appl. 431, 1587 -- 1602 (2009).
[2] M. Bre sar. "Commuting maps: A survey," Taiwanese J. Math. 8, 361 -- 397 (2004).
[3] Z. Chen. "Biderivations and linear commuting maps on simple generalized Witt algebras over a
field," Elec. J. Lin. Alg. 31, 1 -- 12 (2016).
[4] Y. Du, Y. Wang. "Biderivations of generalized algebras," Lin. Alg. Appl. 438, 4483 -- 4499 (2013).
[5] H. Fa, J. Li, Y. Zheng. "Lie bialgebra structures on the deformative Schrodinger-Virasoro algebras,"
J. Math. Phys. 56, 111706 (2015).
[6] G. Fan, X. Dai. "Super-biderivations of Lie superalgebras," Lin. Mult. Alg. 1 -- 9 (2016).
[7] N. M. Ghosseiri. "On biderivations of upper triangular matrix ring ," Lin. Alg. Appl. 438, 250 -- 260
(2013).
[8] J. Han, J. Li, Y. Su. "Lie bialgebra structures on the Schrodinger-Virasoro Lie algebra," J. Math.
Phys. 50, 083504 (2009).
[9] X. Han, D. Wang, C. Xia. "Linear commuting maps and biderivations on the Lie algebras W (a,b),"
J. Lie. Theo. 26, 777-786 (2016).
[10] M. Henkel. "Schrodinger invariance and strongly anisotropic critical systems," J. Stat. Phys. 75,
1023 -- 1029 (1994).
[11] M. Henkel. "Phenomenology of local scale invariance: From conformal invariance to dynamical
scaling," Nucl. Phys. B641, 405 -- 410 (2002).
[12] M. Henkel. "Schrodinger invariance and space-time symmetries," Nucl. Phys. B660, 407 -- 412
(2003).
[13] Q. Jiang, S. Wang,S. "Derivations and automorphism groups of the original deformative
Schrodinger-Virasoro algebras," Alg. Colloq. 22(03), 517 -- 540 (2015). Algebra Colloquium
12
[14] J. Li, Y. Su. "The derivation algebra and automorphism group of the twisted Schrodinger-Virasoro
Lie algebra," arXiv:0801.2207v1 (2008).
[15] J. Li, Y. Su. "Representations of the Schrodinger-Virasoro Lie algebra," J. Math. Phys. 49,
053512(14 pages) (2008).
[16] J. Li. "2-cocycles of original deformative Schrodinger-Virasoro Lie algebra," Sci. China. Ser. A-
Math. 51, 1989-1999 (2008).
[17] J. Li, Y. Su, L, Zhu. "2-cocycles of twisted deformative Schrodinger-Virasoro algebras," Comm. Alg.
40, 1933-1950 (2012).
[18] E. C. Posner. "Derivations in prime rings," Proc. Amer. Math. Soc. 8, 1093 -- 1100 (1957).
[19] C. Roger, J. Unterberger. "The Schrodinger-Virasoro Lie group and algebra: From geometry to
representation theory," Ann. Henri poincar ´e. 7, 1477 -- 1529 (2006).
[20] D. Wang, X. Yu, Z. Chen. "Biderivations of the parabolic subalgebras of simple Lie algebras,"
Comm. Alg. 39, 4097 -- 4104 (2011).
[21] D. Wang, X. Yu. "Biderivations and linear commuting maps on the Schrodinger-Virasoro Lie alge-
bra," Comm. Alg. 41, 2166 -- 2173 (2013).
[22] W. Wang, J. Li, Y. Xu. "Derivations and automorphisms of twisted deformative Schrodinger-
Virasoro Lie algebras," Comm. Alg. 40, 3365 -- 3388 (2012).
[23] C. Xia, D. Wang, X. Han. "Linear super-commuting maps and super-biderivations on the super-
Virasoro algebras," Comm. Alg. 42, 5342 -- 5350 (2016).
[24] J. Zhang, S. Feng, H. Li, R. Wu. "Generalized biderivations of nest algebras," Lin. Alg. Appl. 418,
225 -- 233 (2006).
13
|
1307.0300 | 1 | 1307 | 2013-07-01T08:40:12 | A few remarks on Euler and Bernoulli polynomials and their connections with binomial coefficients and modified Pascal matrices | [
"math.RA",
"math.CA"
] | We prove certain identities involving Euler and Bernoulli polynomials that can be treated as recurrences. We use these and also other known identities to indicate connection of Euler and Bernoulli numbers with entries of inverses of certain lower triangular built of binomial coefficients. Another words we interpret Euler and Bernoulli numbers in terms of modified Pascal matrices. | math.RA | math |
A few remarks on Euler and Bernoulli polynomials and their
connections with binomial coefficients and modified Pascal
matrices
Pawe l J. Szab lowski
Abstract. We prove certain identities involving Euler and Bernoulli polyno-
mials that can be treated as recurrences. We use these and also other known
identities to indicate connection of Euler and Bernoulli numbers with entries
of inverses of certain lower triangular built of binomial coefficients. Another
words we interpret Euler and Bernoulli numbers in terms of modified Pascal
matrices.
1. Introduction and notation
The aim of the paper is to indicate close relationship between Euler and
Bernoulli polynomials and certain lower triangular matrices with entries depending
on binomial coefficients and some other natural numbers. In this way we point out
new interpretation of Euler and Bernoulli numbers.
In the series of papers [2], [3], [5], [4] Zhang, Kim and their associates have
studied various generalizations of Pascal matrices and examined their properties.
The results of this paper can be interpreted as the next step in studying properties
of various modifications of Pascal matrices.
We do so by studying known and indicating new identities involving Euler and
Bernoulli polynomials. One of them particularly simple involves these polynomials
of either only odd or only even degrees.
More precisely we will express these numbers as entries of inverses of certain
matrices build of almost entirely of binomial coefficients.
Throughout the paper we will use the following notation. Let a sequences of
lower triangular matrices {An}n≥0 be such that An is (n + 1) × (n + 1) matrix and
matrix say Ak is a submatrix of every An, for n ≥ k. Notice that the same property
can be attributed to inverses of matrices An (of course if they exist) and to products
of such matrices. Hence to simplify notation we will denote entire sequence of such
matrices by one symbol. Thus e.g. sequence {An}n≥0 will be dented by A or [aij]
if aij denotes (i, j)−th entry of matrices An, n ≥ i. The sequence {A−1
n }n≥0 will
be denoted by A−1 or [aij ]−1. Analogously if we have two sequences say A and B
then AB would mean sequence {AnBn}n≥0 . It is easy to notice that all such lower
2000 Mathematics Subject Classification. Primary 11B68, 11B65; Secondary 11B37, 15B36.
Key words and phrases. Euler polynomials, Bernoulli polynomials, Binomial Coefficients,
Pascal matrices.
1
2
PAWE L J. SZAB LOWSKI
triangular matrices form a non-commutative ring with unity. Moreover this ring is
also a linear space over reals as far as ring's addition is concerned. Diagonal matrix
I with 1 on the diagonal is this ring's unity.
Let us consider also (n+1) vectors X(n) df
= (1, x, . . . , xn)T , f (X)(n) df
= (1, f (x), . . . , f (x)n),
By X or by [xi] we will mean sequence of vectors (cid:0)X(n)(cid:1)n≥0 and by f (X) or by
[f (x)i] the sequence of vectors (f (X)(n))n≥0.
Let En(x) denote n−th Euler polynomial and Bn(x) n−th polynomial. Let
us introduce sequences of vectors E(n)(x) = (1, 2E1(x), . . . , 2nEn(x))T and B(n)(x)
= (1, 2Bn(x), . . . , 2nBn(x))T . These sequences will be denoted briefly E and B
respectively.
⌊x⌋ will denote so called 'floor' finction of x that is the largest integer not
exceeding x.
Since we will use in the sequel often Euler and Bernoulli polynomials we will
recall now briefly the definition of these polynomials. Their characteristic functions
are given by formulae (23.1.1) of [7] and respectively:
(1.1)
(1.2)
tj
j!
tj
j!
Xj≥0
Xj≥0
Ej(x) =
Bj(x) =
2 exp(xt)
exp(t) + 1
,
t exp(xt)
exp(t) − 1
.
Numbers En = 2nEn(1/2) and Bn = Bn(0) are called respectively Euler and
Bernoulli numbers.
By standard manipulation on characteristic functions we obtain for example
the following identities some of which are well known ∀k ≥ 0 :
(1.3)
2kEk(x) =
(1.4)
Bk(x) =
(1.5)
Ek(x) =
(1.6)
Ek(x) =
(1.7)
Bk(x) =
k
Xj=0
k
Xj=0
k
Xj=0
k
Xj=0
k
Xj=0
(cid:18)k
j(cid:19)Ek−j × (2x − 1)j,
(cid:18)k
j(cid:19)Bk−j × xj, xk =
k
Xj=0
(cid:18)k
j(cid:19)
1
k − j + 1
Bj(x),
(cid:18)k
j(cid:19)2jBj(x)
(1 − x)n−j+1 − (−x)n−j+1
(n − j + 1)
,
(cid:18)k
j(cid:19)2jBj(
x
2
)
1
k − j + 1
,
(cid:18)k
j(cid:19)2jBj(x)
(1 − x)n−j + (−x)n−j
2
,
(1.8)
Bk(x) = 2kBk(
x
2
) +
k
Xj=1
(cid:18)k
j(cid:19)2k−j−1Bk−j (
x
2
),
EULER AND BERNOULLI POLYNOMIALS, PASCAL MATRIX
3
which are obtained almost directly from the following trivial identities respectively:
t exp(xt)
exp(t) − 1
×
exp(t) − 1
t
,
× (exp((1 − x)t) − exp(−xt))/t,
1
t
× exp(xt), exp(xt) =
=
=
=
t
2
× exp(
cosh(t/2)
(2x − 1)),
2 exp(xt)
exp(t) + 1
t exp(xt)
exp(t) − 1
2 exp(xt)
exp(t) + 1
2 exp(xt)
exp(t) + 1
t exp(xt)
exp(t) − 1
t exp(xt)
exp(t) − 1
By direct calculation one can easily check that:
exp(t) − 1
2t exp(2xt))
exp(2t) − 1
2t exp( x
2 (2t)))
exp(2t) − 1
2t exp(2xt))
exp(2t) − 1
2t exp( x
2 2t))
exp(2t) − 1
× (exp(t) + 1)/2.
× (exp(t) − 1)/t,
=
=
=
× (exp((1 − x)t) + exp(−xt))/2,
j(cid:19)]−1 = [(−1)i−j(cid:18)i
[(cid:18)i
j(cid:19)], [λi−j(cid:18)i
j(cid:19)]−1 = [(−λ)i−j(cid:18)i
j(cid:19)],
for any λ. The above mentioned identities are well known. They expose properties
of Pascal matrices discussed in [2]. Similarly by direct application of (1.4) we have:
(1.9)
[(cid:18)i
j(cid:19)
1
i − j + 1
]−1 = [(cid:18)i
j(cid:19)Bi−j ]
giving new interpretation of Bernoulli numbers. Now notice that we can multiply
both sides of (1.4) by say λk and define new vectors [(λx)i] and [λiBi(x)]. Thus
(1.9) can be trivially generalized to
j(cid:19) λi−j
[(cid:18)i
i − j + 1
]−1 = [(cid:18)i
j(cid:19)λi−j Bi−j]
for all λ ∈ R, presenting first of the series of modifications of Pascal matrices and
their properties that we will present in the sequel.
To find inverses of other matrices built of binomial coefficients we will have to
refer to the results of the next section.
Theorem 1. ∀n ≥ 1 :
2. Main results
(2.1)
(2.2)
⌊n/2⌋
Xj=0
n
(cid:18)
2 ⌊n/2⌋ − 2j(cid:19)22j+n−2n⌊n/2⌋E2j+n−2⌊n/2⌋(x) = (2x − 1)n,
⌊n/2⌋
Xj=0
n
2 ⌊n/2⌋ − 2j(cid:19)22j+n−2n⌊n/2⌋ B2j+n−2⌊n/2⌋(x)
(cid:18)
2j + 1
= (2x − 1)n
Proof. We start with the following identities:
cosh(t/2)
t exp(xt)
exp(t) − 1
2 exp(xt)
exp(t) + 1
2 sinh(t/2)
t
= exp(t(x − 1/2)),
= exp(t(x − 1/2)).
4
PAWE L J. SZAB LOWSKI
Reacall that we also have:
cosh(t/2) = Xj≥0
t2j
22j(2j)!
,
2 sinh(t/2)
t
= Xj≥0
t2j
22j(2j)!(2j + 1)
.
So applying the standard Cauchy multiplication of two series we get respectively:
Xn≥0
Xn≥0
(2.3)
(2.4)
(2.5)
(2.6)
tn
n!2n (2x − 1)n = Xj≥0
= Xn≥0
n!2n (2x − 1)n = Xj≥0
= Xn≥0
tn
t2j
tj
j!
Ej(x)
22j(2j)! Xj≥0
(cid:18)n
j(cid:19)cjEn−j (x),
tn
n!
n
Xj=0
t2j
22j(2j)!(2j + 1) Xj≥0
tj
j!
Bj(x)
tn
n!
n
Xj=0
(cid:18)n
j(cid:19)c
′
jBn−j(x),
where we denoted by cn and c
′
n the following numbers:
cn = (cid:26) 1
2n
0
if n = 2 ⌊n/2⌋
if
otherwise
, c
′
n = (cid:26)
1
2n(n+1)
0
if n = 2 ⌊n/2⌋
if
otherwise
.
Making use of uniqueness of characteristic functions we can equate functions of x
standing by tn. Finally let us multiply both sides so obtained identities by 2n. We
have obtained (2.1) and (2.2).
(cid:3)
We have the following other result:
1
if
if
Theorem 2. Let e(i) = (cid:26) 0
[e(i − j)(cid:18)i
[e(i − j)(cid:18)i
j(cid:19)
i is odd
i is even , then
j(cid:19)]−1 = [(cid:18)i
j(cid:19)Ei−j],
]−1 = [(cid:18)i
j(cid:19)
i − j + 1
i−j
1
(2.7)
(2.8)
(cid:18)i − j
k (cid:19)2kBk].
Xk=0
Proof. Let us define by Wn(x) = 2nEn((x + 1)/2) and Vn(x) = 2nBn((x +
1)/2). Notice that characteristic function of polynomials Wn and Vn are given by
tj
j!
Xj≥0
Wj (x) = Xj≥0
(2t)j
j!
Ej ((x + 1)/2)
2 exp(2t(x + 1)/2)
=
=
exp(tx)
cosh(t)
,
tj
j!
Xj≥0
Vj (x) = Xj≥0
exp(2t) + 1
(2t)j
j!
Bj((x + 1)/2)
=
exp(2t(x + 1)/2)2t
exp(2t) − 1
=
t exp(tx)
sinh(t)
Now recall that
cosh(t) is a characteristic function of Euler numbers while
1
equal to the characteristic function of numbers nPn
j=0(cid:0)n
j(cid:1)2jBjon≥0
t
sinh t
. Hence on one
EULER AND BERNOULLI POLYNOMIALS, PASCAL MATRIX
5
hand see that
Wn(x) =
Vn(x) =
n
Xj=0
n
Xj=0
(cid:18)n
j(cid:19)xn−jEj ,
(cid:18)n
j(cid:19)xn−j
j
Xk=0
(cid:18)j
k(cid:19)2kBk
On the other substituting x by (x + 1)/2 in (2.1) and (2.2) we see that
xn =
xn =
n
Xj=0
Xj=0
n
e(n − j)(cid:18)n
j(cid:19)Wj(x),
e(n − j)(cid:18)n
j(cid:19)
1
n − j + 1
Vj
By uniqueness of the polynomial expansion we deduce (2.7) and (2.8).
(cid:3)
As a corollary we get the following result following also well known properties
of lower triangular matrices (see e.g. : [6]).
Corollary 1.
[(cid:18)2i
2j(cid:19)
[(cid:18)2i
2j(cid:19)]−1 = [(cid:18)2i
2j(cid:19)E2(i−j)],
]−1 = [(cid:18)2i
2j(cid:19)
2i−2j
1
Xk=0
2(i − j) + 1
(cid:18)2i − 2j
k (cid:19)2kBk].
As in Section 1 we can multiply both sides of (2.1) and (2.2) by λn and redefine
appropriate vectors and rephrase out results in terms of modified Pascal matrices.
Corollary 2. Forr all λ ∈ R:
(2.9)
(2.10)
(2.11)
(2.12)
[e(i − j)(cid:18)i
[e(i − j)(cid:18)i
i−j
i − j + 1
j(cid:19)λi−j]−1 = [(cid:18)i
j(cid:19) λi−j
]−1 = [(cid:18)i
[(cid:18)2i
2j(cid:19)λi−j]−1 = [(cid:18)2i
λi−j
]−1 = [(cid:18)2i
j(cid:19)λi−jEi−j ],
j(cid:19)λi−j
(cid:18)i − j
2j(cid:19)λi−j E2(i−j)],
2j(cid:19)λi−j
2(i − j) + 1
Xk=0
2i−2j
Xk=0
[(cid:18)2i
2j(cid:19)
k (cid:19)2kBk],
(cid:18)2i − 2j
k (cid:19)2kBk].
References
[1] Srivastava, H. M.; Pint´er, ´A. Remarks on some relationships between the Bernoulli and Euler
polynomials. Appl. Math. Lett. 17 (2004), no. 4, 375 -- 380. MR2045740 (2004m:33020)
[2] Zhang, Zhizheng. The linear algebra of the generalized Pascal matrix. Linear Algebra Appl.
250 (1997), 51 -- 60. MR1420570 (97g:15014)
[3] Zhang, Zhizheng; Liu, Maixue. An extension of the generalized Pascal matrix and its algebraic
properties. Linear Algebra Appl. 271 (1998), 169 -- 177. MR1485166 (98m:15045)
[4] Zhang, Zhizheng; Wang, Xin. A factorization of the symmetric Pascal matrix involving the Fi-
bonacci matrix. Discrete Appl. Math. 155 (2007), no. 17, 2371 -- 2376. MR2360667 (2008h:11013)
6
PAWE L J. SZAB LOWSKI
[5] Cheon, Gi-Sang; Kim, Jin-Soo. Stirling matrix via Pascal matrix. Linear Algebra Appl. 329
(2001), no. 1-3, 49 -- 59. MR1822221 (2002b:11030)
[6] Lutkepohl, H. Handbook of matrices. John Wiley & Sons, Ltd., Chichester, 1996. xvi+304 pp.
ISBN: 0-471-97015-8 MR1433592 (97i:15001)
[7] Gradshteyn, I. S.; Ryzhik, I. M. Table of integrals, series, and products. Translated from
the Russian. Translation edited and with a preface by Alan Jeffrey and Daniel Zwillinger.
With one CD-ROM (Windows, Macintosh and UNIX). Seventh edition. Elsevier/Academic
Press, Amsterdam, 2007. xlviii+1171 pp. ISBN: 978-0-12-373637-6; 0-12-373637-4 MR2360010
(2008g:00005)
Department of Mathematics and Information Sciences,, Warsaw University of Tech-
nology, ul. Koszykowa 75, 00-662 Warsaw, Poland
E-mail address: [email protected]
|
1109.3003 | 1 | 1109 | 2011-09-14T06:46:08 | Characterisation of PF rings by the Finite Topology on duals of R-Modules | [
"math.RA"
] | In this paper we study the properties of the finite topology on the dual of a module over an arbitrary ring. We aim to give conditions when certain properties of the field case are can be still found here. Investigating the correspondence between the closed submodules of the dual $M^{*}$ of a module $M$ and the submodules of $M$, we prove some characterisations of PF rings: the up stated correspondence is an anti isomorphism of lattices iff $R$ is a PF ring. | math.RA | math |
Characterisation of PF rings by the Finite Topology on
duals of R Modules
Miodrag Cristian Iovanov
Department of Algebra, Faculty of Mathematics, University of Bucharest
Academiei 14, Bucharest, Romania
Abstract. In this paper we study the properties of the finite topology on the dual of a module
over an arbitrary ring. We aim to give conditions when certain properties of the field case are can be
still found here. Investigating the correspondence between the closed submodules of the dual M ∗ of
a module M and the submodules of M , we prove some characterisations of PF rings: the up stated
correspondence is an anti isomorphism of lattices iff R is a PF ring.
1
Introduction and preliminaries
Let R be an arbitrary (non commutative) ring. We will use the notations HomR(M, N ) for the set
of R module morphisms from M to N for right modules M, N and RHom(M, N ) respectively for left
modules M, N . Also we use M ∗ = HomR(M, R) for any right module M and ∗M = RHom(M, R) for
a left module M .
Given two right R modules M and N , recall that the finite topology on HomR(M, N ) is the linear
topology for which a basis of open neighborhoods for 0 is given by the sets {f ∈ HomR(M, N ) f (xi) =
0, ∀ i ∈ {1, . . . , n}}, for all finite sets {x1, . . . , xn} ⊆ M . This is actually the topology induced on
HomR(M, N ) from HomSet(M, N ) = N M which is a product of topological spaces, where N is the
topological discrete space on the set N . For an arbitrary set X ⊆ M we denote by X ⊥ = {f ∈
HomR(M, N ) f X = 0}. Denoting by < X >R the R submodule generated by X, we obviously have
(< X >R)⊥ = X ⊥, so we will work with finitely generated submodules F ≤ M and the basis of open
neighborhoods {F ⊥ F ≤ M finitely generated}. Also for left R modules X and Y and U ≤ X a
submodule of X we will denote U ⊥
RHom(M,N ) or simply U ⊥ = {g ∈ RHom(X, Y ) gX = 0} when there
is no danger of confusion. If W ≤ HomR(M, N ) is a subgroup with M and N left R modules we denote
W ⊥ = {x ∈ N f (x) = 0, ∀ f ∈ W }. If N is an R bimodule then we consider the left R module
structure on HomR(M, N ) given by (r · f )(x) = rf (x), for all x ∈ M, f ∈ HomR(M, N ), r ∈ R. If W
is a (left) submodule in HomR(M, N ), then W ⊥ is a (right) submodule of M.
For any right module M we denote by ΦM the right R modules morphism
M
ΦM−→ ∗(M ∗)
1
defined by ΦM (m)(f ) = f (m), for all f ∈ M ∗ and all m ∈ M . Then Φ is a functorial morphism from
idMR to the functor ∗((−)∗).
Over a field, there is a series of properties involving the orthogonal F ⊥ for a vector space V and its
dual V ∗ which we will state in a more general setting.
Proposition 1.1 Let M, N be R modules.
(i) If X ⊆ Y are submodules of M then Y ⊥ ≤ X ⊥.
(ii) If U ⊆ V are subgroups of HomR(M, N ) then V ⊥ ≤ U ⊥.
Lemma 1.2 For M, N right R modules we have:
(i) If X ≤ M is a submodule of M then (X ⊥)⊥ ⊇ X and if we denote 0 the class of 0 in M/X then
we have ({0}⊥)⊥ = (X ⊥)⊥/X. If N is an injective cogenerator of MR then the equality (X ⊥)⊥ = X
holds.
(ii) If Y ≤ HomR(M, N ) is a (left) submodule of HomR(M, N ) then (Y ⊥)⊥ ⊇ Y (Y is the closure of Y
in HomR(M, N )). If N = R and R is a left PF ring (RR is injective and a cogenerator of RM) then
the equality (Y ⊥)⊥ = Y holds for all modules M and (left) submodules Y ≤ M ∗.
Proof. (i) If x ∈ X then take f ∈ X ⊥; then f (x) = 0 as f X = 0. We get that f (x) = 0, ∀ f ∈ X ⊥
so x ∈ (X ⊥)⊥. Moreover, x ∈ ({0}⊥)⊥ if and only if h(x) = 0, ∀ h : M/X −→ N , equivalent to
h(x) = 0, ∀h ∈ X ⊥, i.e. x ∈ (X ⊥)⊥.
Suppose now N is an injective cogenerator of MR and take x ∈ (X ⊥)⊥.
If x /∈ X then there is
f : M/X −→ N such that f (x) 6= 0 (x is the image of x in M/X via the canonic morphism π : M −→
M/X). Then there is g = f ◦ π, g ∈ HomR(M, N ) such that gX = 0 (g ∈ X ⊥) and g(x) 6= 0, showing
that x /∈ (X ⊥)⊥, a contradiction.
(ii) Let f ∈ Y and take x ∈ Y ⊥. Then there is g ∈ Y such that f (x) = g(x). But g(x) = 0 because
x ∈ Y ⊥ so f (x) = 0. Thus f Y ⊥ = 0 and f ∈ (Y ⊥)⊥.
For the converse, first we see that RR injective implies that for all finitely generated right R modules F
ΦF−→ ∗(F ∗) is an epimorphism. Take π : P = Rn −→ F an epimorphism in MR. Then
we have that F
we have a monomorphism 0 −→ P ∗ −→ F ∗ in RM, and as RR is injective we obtain an epimorphism
∗(p∗)
of right modules ∗(P ∗)
−→ ∗(F ∗) −→ 0. Because Φ is a functorial morphism then we have the
commutative diagram
P
ΦP
π
F
ΦF
∗(P ∗) ∗(π∗)
/ ∗(F ∗)
0
/ 0
showing that ΦF is surjective, as ΦP = ΦRn is an isomorphism. Now to prove the desired equality,
take f ∈ (Y ⊥)⊥, (fi)i∈I a family of generators of the left R module Y , and F < M a finitely generated
submodule of M . Then fiM ∈ F ∗ and if f F /∈ R < fiF i ∈ I > then as RR is an injective
2
/
/
/
/
/
/
cogenerator of RM we can find a morphism of left R modules φ : F ∗ −→ R such that φ(fi) = 0, ∀i ∈ I
and φ(f ) 6= 0. But as ΦF is surjective, we can then find x ∈ F such that φ = Φ(x) and then
fi(x)Φ(x)(fi) = φ(fi) = 0, ∀i ∈ I, showing that x ∈ Y ⊥ and f (x) = Φ(x)(f ) = φ(f ) 6= 0 which
contradicts the fact that f belongs to (Y ⊥)⊥. Thus we must have f F ∈ R < fiF i ∈ I > so there
is (ri)i∈I a family of finite support such that f F = P
rifi)F . This last relation shows
that f ∈ Y .
ri(fiF ) = (P
i∈I
i∈I
✷
Corollary 1.3 If R is a PF ring (left and right) then for any right (or left) R module M and Y < M ∗
we have that Y is dense in M ∗ if and only if Y ⊥ = 0.
Proposition 1.4 Let M be a right R module.
(i) If X ≤ M then we have ((X ⊥)⊥)⊥ = X ⊥ and X ⊥ is closed.
(ii) If Y ≤ HomR(M, N ) then ((Y ⊥)⊥)⊥ = Y ⊥.
Proof. "⊆" from (i) and (ii) follow from Proposition 1.1 and Lemma 1.2.
(i) "⊇" Let f ∈ X ⊥. Take x ∈ (X ⊥)⊥; then f (x) = 0 so f ∈ ((X ⊥)⊥)⊥. To show that X ⊥ is closed
take f ∈ X ⊥ and x ∈ X. Then there is g ∈ X ⊥ such that g(x) = f (x) so f (x) = 0 (x ∈ X). We obtain
that f X = 0 so f ∈ X ⊥.
(ii) "⊇" Let x ∈ Y ⊥. If f ∈ (Y ⊥)⊥ then f Y ⊥ = 0 so f (x) = 0 showing that x ∈ ((Y ⊥)⊥)⊥.
✷
Proposition 1.5 Let M, N be right R modules and (Xi)i∈I a family of submodules of M . Then
(i) (P
i∈I
(ii) ( T
i∈I
i . If I is finite and N is injective then equality holds.
Xi)⊥ = T
Xi)⊥ ⊇ P
X ⊥
i .
X ⊥
i∈I
i∈I
Proof. (i) f ∈ (P
i∈I
Xi)⊥ ⇔ f Pi∈I
Xi = 0 ⇔ f Xi = 0, ∀i ∈ I ⇔ f ∈ X ⊥
i , ∀i ∈ I ⇔ f ∈ T
X ⊥
i .
i∈I
(ii) "⊇" is obvious, for Proposition 1.1 shows that X ⊥
i ⊆ T
j∈I
Xj
⊥, ∀i ∈ I. For the converse it is enough
to prove the equality for two submodules X, Y of M . Denote π : M −→ M/X ∩ Y , p : M −→ M/X,
q : M −→ M/Y the canonical morphisms. If f ∈ HomR(M, N ) such that f X∩Y = 0 then denote
f : M/X ∩ Y −→ N the factorisation of f (f = f ◦ π) and i : M/X ∩ Y −→ M/X ⊕ M/Y the injection
i(π(x)) = (p(x), q(x)), ∀x ∈ M . Then the diagram
0
/ M
X∩Y
i
M
X ⊕ M
Y
u
u
u
u
u
u
u
h=u⊕v
u
zu
f
N
is completed commutatively by h. Then h = u ⊕ v, with u ∈ HomR(M/X, N ) and HomR(M/Y, N ),
such that h(p(x), q(x)) = u(p(x)) + v(q(x)). Taking u = u ◦ p and v = v ◦ q we have u ∈ X ⊥, v ∈ Y ⊥
and f (x) = f (π(x)) = h(i(π(x))) = h(p(x), q(x)) = u(p(x)) + v(q(x)) = u(x) + v(x), ∀x ∈ M , so
f ∈ X ⊥ + Y ⊥.
✷
3
/
/
/
z
Proposition 1.6 Let M, N be right R modules and (Yi)i∈I a family of submodules of HomR(M, N ).
Then:
(i) (P
i∈I
(ii) ( T
i∈I
of M ∗ = HomR(M, R) then the equality holds: ( T
i . If N = R and R is a PF ring (both left and right PF) and Yi are closed subsets
Yi)⊥ = T
Yi)⊥ ⊇ P
Yi)⊥ = P
Y ⊥
i .
Y ⊥
i .
Y ⊥
i∈I
i∈I
i∈I
i∈I
Proof. (i) Obvious.
(ii) "⊇" similar to (ii)"⊇" of the previous proposition. For the converse inclusion, take (Yi)i∈I a family
of submodules of M ∗. Then
X
i∈I
Y ⊥
i = ((X
i∈I
= (\
i∈I
= (\
i∈I
Y ⊥
i )⊥)⊥ (from Lemma 1.2 : R is right PF)
(Y ⊥
i )⊥)⊥ (from Proposition 1.5)
Yi)⊥ (Lemma 1.2 : Yi are closed and R is left PF)
✷
Example 1.7 (i) We show that the equality in Proposition 1.5 does not hold for infinite sets. Let V
be an infinite dimensional space with a countable basis indexed by the set of natural numbers: (en)n∈N.
Vn = 0)⊥ = V ∗. Let f ∈ V ∗
Put Vn =< ek k ≥ n >. Then we can easily see that T
be the function equal to 1 on all the en-s. Then as V ⊥
Vn = 0 so ( T
n∈N
n∈N
n < V ⊥
V ⊥
n ⇔
m , ∀ n < m, we have that f ∈ P
n∈N
V ⊥
n a strict
Vn ⊃ P
n∈N
n∈N
∃ n ∈ N such that f ∈ V ⊥
n which is impossible as f (en) = 0, ∀ n. We obtain T
inclusion.
(ii) We show now that the equality in Proposition 1.6 does not hold for non-closed sets. Let again V
be a vector space with a countable basis B = (en)n∈N. Denote by e∗
n the linear map equal to 1 on en
and 0 on the other elements of the basis B and by f ∗ the linear map equal to 1 on all the en-s. Take
n n ∈ N∗ >. Then we can easily see that H ⊥ = 0, L⊥ = 0
H =< e∗
and H ∩ L =< e∗
n n ∈ N >⊥=< e0 >, thus
H ⊥ + L⊥ 6= (H ∩ L)⊥.
n n ∈ N∗ >, so H ⊥ + L⊥ = 0, but (H ∩ L)⊥ =< f ∗, e∗
n n ∈ N > and L =< f ∗, e∗
(iii) Given the same vector space, we give an example of a family of dense subspaces of V ∗ whose
intersection is 0. For p ∈ N let Hp =< e∗
n+p n ∈ N >. Then a short computation shows
Hn ⊂ H0,
Hn = 0, because if f =
n = 0 showing that Hn is closed in V ∗. But T
n+1+. . .+e∗
that H ⊥
i ∈ T
n+e∗
λie∗
n∈N
then f ∈ Hm+1 which shows that if f 6= 0, than it can be written as a linear combination of e∗
at least one of the e∗
i has i > m. This is impossible as the e∗
n-s are independent.
n∈N
i in which
m
P
i=1
4
2 The Finite Topology vs PF Rings
If R is a ring then we have (Rn)∗ = HomR(R, R) ≃ RRn. So we can identify R submodules of the right
dual of Rn with left submodules of RR and vice versa. For all x = (x1, . . . , xn) ∈ Rn we denote by
ϕx : Rn −→ R the morphism of right R modules ϕx(r1, . . . , rn) =
xiri and by ψx the morphism of
n
P
i=1
n
P
i=1
left modules defined by ψx(r1, . . . , rn) =
rixi, ∀ (r1, . . . , rn) ∈ Rn. Also because of the isomorphism
(Rn)∗ ≃ RRn, x 7→ ϕx, we will denote by I ⊥ = {x ∈ Rn ϕx(r) = 0, ∀ r ∈ I} if I is a right submodule
of Rn and similarly for left submodules X of Rn, X ⊥ = {x ∈ Rn ψx(r) = 0, ∀ r ∈ X}.
Over a vector space V there is an anti isomorphism of lattices between the lattice of closed subspaces
of V ∗ and the subspaces of V given by X 7→ X ⊥, ∀ X ≤ V . We have the obvious
Proposition 2.1 For a right module M the following are equivalent:
(i) The applications M ≥ X 7→ X ⊥ ≤ M ∗ and M ∗ ≥ Y 7→ Y ⊥ ≤ M between the lattice of the
submodules of M and the lattice of the closed submodules of M ∗ are inverse anti isomorphism of
lattices.
(ii) (X ⊥)⊥ = X, ∀ X ≤ M and (Y ⊥)⊥ = Y , ∀ Y ≤ M ∗.
(iii) (X ⊥)⊥ = X, ∀ X ≤ M and (Y ⊥)⊥ = Y, ∀ Y ≤ M ∗, Y closed.
(iv) The applications of (i) are inverse to each other.
If F is a finitely generated right R module then every submodule of F ∗ is closed, as if Y is a left
submodule of F ∗ and f ∈ Y , taking {x1, . . . , xn} the a system of generators of F , there is g ∈ Y such
that g(xi) = f (xi), for all i, so f = g ∈ Y . Also it is easy to see that Rn has orthogonal equivalence
as right module if and only if it has orthogonal equivalence as left module, and this is equivalent to
(I ⊥)⊥ = I, ∀ I ≤ Rn
R and (X ⊥)⊥ = X, ∀ X ≤ RRn.
Definition 2.2 We will say that a right R module M has orthogonal equivalence (or orthogonal isomor-
phism, or shortly M has ⊥ equivalence) if the equivalent statements of Proposition 2.1 hold. The ring
R will be called with ⊥ equivalence if RR (or equivalently RR) is a module with orthogonal equivalence.
Proposition 2.3 Let M be a right R module and X a submodule of M . Then we have the exact
sequence
0 −→ (0⊥)⊥ −→ M
ΦM−→ ∗(M ∗)
Proof. For x ∈ M we have ΦM (x) = 0 ⇔ f (x) = 0, ∀ f ∈ M ∗ and this equivalent to x ∈ (M ∗)⊥ =
(0⊥)⊥, thus ker ΦM = (0⊥)⊥.
✷
Proposition 2.4 (i) For an R module M we have (0⊥)⊥ = 0 if and only if M is R cogenerated, i.e.
there is a monomorphism M ֒→ RI for some set I.
(ii) If C is a class of right R modules which is closed under quotients then the following are equivalent:
(a) (X ⊥)⊥ = X for all M in C, X < M .
(b) (0⊥)⊥ = 0 for all M in C.
5
(c) Any M ∈ C is cogenerated by R.
(d) ΦM is a monomorphism for every M in C.
Proof. (i) If (0⊥)⊥ = 0 then take I = M ∗ and M i−→ RI , i(x) = (f (x))f ∈I ; then of course i is a
monomorphism as i(x) = 0 if and only if f (x) = 0, ∀f ∈ I = M ∗ i.e. x ∈ (0⊥)⊥ = 0. Conversely, given
i
֒→ RI , taking πj the canonical projections for all j ∈ I, we obtain the morphisms
a monomorphism M
fj = πj ◦ i ∈ M ∗ and then x ∈ (0⊥)⊥ = (M ∗)⊥ implies fj(x) = 0, ∀ j ∈ I, i.e. i(x) = 0 so x = 0, as i is
injective. Thus (0⊥)⊥ = 0.
(ii) (b) ⇔ (c) by (i). (a) ⇔ (b) follows as C is closed under quotient objects and denoting 0 the zero
element of M/X ∈ C we have ({0}⊥)⊥ = (X ⊥)⊥ from Lemma 1.2. Equivalence with (d) follows from
Proposition 2.3
✷
Proposition 2.5 Suppose RR is a module with ⊥ equivalence. Then R contains all left simple modules
and all right simple modules (up to an isomorphism; this is called a right - and left- Kasch ring).
Proof. It is easy to see that for every right ideal I of R we have the isomorphism of left R modules
I )∗ ≃ I ⊥, given by I ⊥ ∋ f 7→ f ◦ π ∈ ( R
( R
I )∗, with π : R −→ R/I the canonical projection. Then if
S is simple right module there is a maximal right ideal M < R and an isomorphism S ≃ R
M . Then
S∗ ≃ ( R
M )∗ ≃ M ⊥ 6= 0 because if M ⊥ = 0 then M = (M ⊥)⊥ = 0⊥ = R, which contradicts the
maximality of M . In a similar way one can see that R contains all the isomorphism types of left R
modules.
✷
We shall say a right (or left) R module is n generated if it has a system of n generators.
Lemma 2.6 Let X be a right R module such that every monomorphism i : X ֒→ M with the property
that M/Im i is 1-generated splits. Then X is an injective module.
Proof. Let M be a right R module such that X < M (we identify X with its image in M ) and suppose
X 6= M . Let L = {Y < M Y 6= 0 and X ∩ Y = 0}. Then L 6= ∅, because if x ∈ M \ X then as
(X + xR)/X 6= 0 is finitely generated then the hypothesis shows that there is Y < X + xR such that
.
+ Y = X + xR and then Y 6= 0 as x /∈ X, so Y ∈ L. We can easily see that L is inductive, because if
X
Yi is its majorant in L. Take N a maximal
(Yi)i∈I is a totally ordered family of elements of L then S
element of L and suppose X + N 6= M . Then there is x ∈ M \ (X + N ) and as (X + N + xR)/(X + N )
is finitely generated, by the hypothesis we can find Y < M such that X + N + Y = X + N + xR and
(X + N ) ∩ Y = 0. An easy computation shows now that (N + Y ) ∩ X = 0 and so N + Y = N by the
maximality of N . Thus we obtain X + N + Y = X + N = X + N + xR which is a contradiction, because
x /∈ X + N . We find that X is a direct summand in M for every module M such that X ֒→ M , so X
is injective in MR.
i∈I
✷
Proposition 2.7 Let R be a ring with ⊥ equivalence. If R
modules and X is R cogenerated then j splits.
j
֒→ X is a monomorphism of right (left) R
6
Rx⊥
i = 0. Then we have 0 = T
i∈I
Rx⊥
i = (P
i∈I
Rxi)⊥)⊥ = 0⊥ = R. Then we find that there is F a finite subset of I such that P
σ
֒→ RI a monomorphism and let (xi)i∈I = σ(j(1)). Then we have (xir)i∈I =
Proof. Consider X
σ(j(1))r = σ(j(r)) and as j, σ are injective we see that xir = 0, ∀ i ∈ I if and only if r = 0. This
Rxi =
shows that T
i∈I
((P
i∈I
there are (yi)i∈F ∈ R such that P
πF ((ri)i∈I ) = (ri)i∈F and by y = (yi)i∈F ∈ RF = R(F ), then ϕy(πF (σ(j(r)))) = ϕy(πF ((xir)i∈I )) =
yixir = r, so ϕy ◦ πF ◦ σ ◦ j = idR, showing that the morphism of right modules
ϕy((xir)i∈F ) = P
i∈F
ϕy ◦ πF ◦ σ : X −→ R is a split for j.
yixi = 1. Now if we denote by πF the projection of RI on RF ,
Rxi)⊥ (by Proposition 1.5), so P
Rxi = R, thus
i∈F
i∈I
i∈F
✷
Lemma 2.8 Rn has orthogonal equivalence (as left or right R module) if and only if every n generated
right (or left) module has orthogonal equivalence.
Proof. Suppose Rn has ⊥ equivalence. Let F = Rn/X be a right n generated R modules and
π : Rn −→ F the canonical projection. For each g ∈ X ⊥ (X < Rn) we denote by g ∈ F ∗ the (unique)
morphism for which g ◦ π = g and with x = π(x) - the class of an element x ∈ Rn. Now we see that
if Y < F ∗ and Z = {α ◦ π α ∈ Y }, then Y = {g g ∈ Z}, Y ⊥ = {x g(x) = 0, ∀ g ∈ Z} = Z ⊥/X
(Z ⊆ X ⊥ so Z ⊥ ⊇ (X ⊥)⊥ = X) and (Y ⊥)⊥ = {g g(x) = 0, ∀ x ∈ Z ⊥/X} = {g g(x) = 0, ∀ x ∈
Z ⊥} = {g g ∈ (Z ⊥)⊥ = Z} = Y .
Now if Y < F and Z = π−1(Y ) then Y ⊥ = {g g(x) = 0, ∀ x ∈ Y } = {g g(x) = 0, ∀ x ∈ Z} = {g
g ∈ Z ⊥} and (Y ⊥)⊥ = {x g(x) = g(x) = 0, ∀ g ∈ Z ⊥} = {x x ∈ (Z ⊥)⊥ = Z} = Y .
✷
Theorem 2.9 The following assertions are equivalent:
(i) Every right R module has ⊥ equivalence.
(ii) Every finitely generated module has ⊥ equivalence.
(iii) Every left R module has ⊥ equivalence.
(iv) Every finitely generated module has ⊥ equivalence.
(v) R is a PF ring (both left and right).
(vi) (X ⊥)⊥ = X for all X < M in MR or in RM.
(vii) R2 has ⊥ equivalence.
Proof. • (v) ⇒ (i) and (v) ⇒ (vi) follow from Lemma 1.2 so we have the implications (v) ⇒ (i) ⇒
(ii) ⇒ (vii) and (v) ⇒ (vi) ⇒ (vii)
• (v) ⇒ (iii) ⇒ (iv) ⇒ (vii) is the left symmetric of (v) ⇒ (i) ⇒ (ii) ⇒ (vii).
• (vii) ⇒ (v) If R2 has ⊥ equivalence, then by Lemma 2.8 we have that any 2 generated right (and
i
any left) module has ⊥ equivalence, in particular R has orthogonal equivalence. Now let R
֒→ X be
a monomorphism in MR such that X/i(R) is 1 generated. Then as X has ⊥ equivalence, Proposition
2.4 shows that X is R cogenerated as right R module. Now by Proposition 2.7 i splits, as X is R
cogenerated and R has ⊥ equivalence. Then we can apply Lemma 2.6 and obtain that RR is injective.
Because R has ⊥ equivalence, by Proposition 2.5 we obtain that RR contains all isomorphism types of
7
simple right modules, and as RR is injective, we obtain that RR is an injective cogenerator of MR, i.e.
a right RF ring. Similarly we can show that R is also a left PF ring.
✷
Corollary 2.10 If R is a PF ring, then F ≃ ∗(F ∗) by ΦF for every finitely generated left module (the
analogue holds for right modules).
Proof. Proposition 2.3 shows that ΦF is injective. By the same argument as in the proof of Lemma
1.2 we have that RR injective implies that ΦF is an epimorphism and the conclusion is proved.
✷
Corollary 2.11 R is a PF ring if and only if for every finitely generated right (or left) R module F ,
the lattice of the submodules of F is anti isomorphic to the lattice of the submodules of F ∗ via the ⊥
applications of Proposition 2.1, equivalently, the dual lattice of the submodules of any finitely generated
right module is isomorphic (via ⊥ applications) to the lattice of the submodules of the dual of that
module.
References
[1] R. B. Ash, Abstract Algebra , University of Illinois, 2000.
[2] X. Chen and Q. Xu, Toeplitz operators on discrete abelian groups, Mathematical Proceeding of the Royal Irish
Academy, 100A(2), (2000), 139-148.
[3] E. H. Connel, Elements of Abstract and Linear Algebra, University of Miami, 2002.
[4] I. D. Ion, Algebra, Bucharest, 1974.
[5] I. M. Isaac, Algebra, Belmont, California, 1994.
[6] L. Lady, Finite Rank Torsion Free Modules over Dedekind Domains , University Hawaii, 1998.
[7] J. D. Lewis, Commutative ring theory, Alberta University, 2000.
[8] J. Rada and M. Saorin, On semiregular rings whose finitely generated modules embed in free modules, Canad. Math.
Bull. , Vol. 40(2), (1997).
[9] T. M. Viswanathan, Ordered modules of fractions, J. Reine Sngew. Math., (1969).
8
|
1101.6004 | 1 | 1101 | 2011-01-31T16:17:15 | Weight Ideals Associated to Regular and Log-Linear Arrays | [
"math.RA",
"math.LO"
] | Certain weight-based orders on the free associative algebra $R = k<x_1, ..., x_t >$ can be specified by $t \times \infty$ arrays whose entries come from the subring of nonnegative elements in a totally ordered field. Such an array $A$ satisfying certain additional conditions produces a partial order on $R$ which is an admissible order on the quotient $R/I_A$, where $I_A$ is a homogeneous binomial ideal called the {\em weight ideal} associated to the array and whose structure is determined entirely by $A$. This article discusses the structure of the weight ideals associated to two distinct sets of arrays whose elements define admissible orders on the associated quotient algebra. | math.RA | math |
Weight Ideals Associated to Regular and
Log-Linear Arrays
Jeremiah W. Johnson
Dept. of Mathematics and Statistics, Penn State Harrisburg
Middletown, PA 17057
Abstract
Certain weight-based orders on the free associative algebra R = khx1, . . . , xti can
be specified by t × ∞ arrays whose entries come from the subring of nonnegative
elements in a totally ordered field. Such an array A satisfying certain additional
conditions produces a partial order on R which is an admissible order on the quotient
R/IA, where IA is a homogeneous binomial ideal called the weight ideal associated
to the array and whose structure is determined entirely by A. This article discusses
the structure of the weight ideals associated to two distinct sets of arrays whose
elements define admissible orders on the associated quotient algebra.
Key words: Noncommutative Grobner Bases, Grobner Bases, Admissible Orders
1
Introduction
Work over the past two decades has extended the theory of Grobner bases
to various noncommutative algebras (Green, 2000; Madlener & Reinert, 1997;
Nordbeck, 2001; Mora, 1994). Before a Grobner basis for an ideal of a k-algebra
A can be constructed, where k is a field, an admissible order on a multiplicative
basis of A is required. Following Green (1996), we say that A has a Grobner
basis theory when an admissible order exists on a multiplicative basis of A. In
Hinson (2010), E. Hinson adapted the theory of position-dependent weighted
orders to define a length-dominant partial order on the set of words in the free
associative algebra R = khx1, . . . , xti, including the trivial word, which pro-
duces an admissible order on a quotient of R. In this construction, the partial
order on R is specified by a t×∞ array A whose entries come from the subring
consisting of the positive elements of a totally ordered field, and the quotient
Email address: [email protected] (Jeremiah W. Johnson).
Preprint submitted to Elsevier
1 August 2018
is by a homogenous binomial ideal IA whose elements are determined by the
partial order given by A. This gives rise to two immediate questions. First,
given an array A that defines an admissible order on a quotient R/IA, what
is the algebra that is determined, or more specifically, what is the structure
of the ideal IA? Second, given two arrays A and B which define orders ≻A
and ≻B on R/IA and R/IB respectively, even when R/IA = R/IB it is not
necessarily the case that ≻A=≻B. Under what circumstances does ≻A=≻B?
This paper describes results concerning the first of these two questions for two
distinct families of admissible arrays. In this introductory section, we review
the relevant definitions and results from Hinson (2010) and we make the pre-
ceding general statements precise. Our primary objects of interest are defined
in Definitions 5 and 6. The results on which the remainder of the paper relies
are given in Theorems 7 and 8. In what follows, let R = khx1, . . . , xti denote
the free associative algebra, let S>0 denote the positive elements of a totally
ordered field, and let Mt×∞ (S>0) denote the set of t × ∞ arrays with entries
in S>0. The following two definitions are adopted from Green (1996).
Definition 1 Let B be a k -- basis of an algebra A. B is a multiplicative basis
for A if
b, b′ ∈ B ⇒ b · b′ ∈ B or b · b′ = 0.
We will have occasion to refer to the nontrivial elements of B, which we denote
by B×.
Definition 2 A total order ≻ on a multiplicative basis B of A is an admissible
order on B if
• ≻ is a well-order on B,
• for all b1, b2, b3 ∈ B such that b1b3 6= 0 and b2b3 6= 0, if b1 ≻ b2, then
b1b3 ≻ b2b3,
• for all b1, b2, b3 ∈ B such that b3b1 6= 0 and b3b2 6= 0, if b1 ≻ b2, then
b3b1 ≻ b3b2, and
• for all b1, b2, b3, b4 ∈ B, if b1 = b2b3b4, then b1 (cid:23) b3.
Commonly used admissible orders for Grobner basis calculations on noncom-
mutative algebras are the left length-lexicographic order or the right length-
lexicographic order (Green, 1996). We specify a position-dependent weighted
order on words in the free algebra using a t × ∞ array to define a weight
function as described in the following definition.
Definition 3 Let A = (ai,j) ∈ Mt×∞(S>0). A gives a monomial weighting
σA : B× → S>0 by
l−1
Y
j=0
auj ,j
σA(xu0xu1 · · · xul−1) =
2
for a given monomial xu0xu1 · · · xul−1 ∈ R. The function σA is the weight
function associated to A.
Note that for computational convenience we index the columns of an array
starting with 0 rather than 1. When the array A is clear, we will suppress it
from the notation and write the associated weight function σA simply as σ.
For the remainder of this section, fix an array A ∈ Mt×∞(S>0) and associated
weight function σ. In order to discuss the weight of the product of two words,
we identify a translated version of the weight function associated to A by
σk(xu0xu1 · · · xul−1) =
l−1
Y
j=0
auj,j+k,
where k ∈ N. We consider σ(ω) = σA(ω) = σA,0(ω). Let ω denote the length
of ω. Given ω and λ such that ω = k,
σ(ωλ) = σ(ω) · σk(λ).
This gives rise to the following equivalence relation.
Definition 4 Define the relation ≻σ on B× by
ω1 ≻σ ω2 ⇐⇒ ω1 > ω2, or ω1 = ω2 and σ(ω1) > σ(ω2).
Let Γ denote the set of pure homogeneous binomial differences ω1 − ω2, where
ω1, ω2 ∈ B×, ω1 = ω2, and σ(ω1) = σ(ω2).
Definition 5 The ideal IA = hΓi is the weight ideal associated to A.
Definition 6 A is an admissible array if for every pair ω1, ω2 ∈ B× with
ω1 = ω2,
(1) for all k ≥ 0, if σk(ω1) > σk(ω2), then σk+1(ω1) > σk+1(ω2), and
(2) for all k ≥ 0, if σk(ω1) = σk(ω2), then σk+1(ω1) = σk+1(ω2).
The following theorem illustrates that the second part of Definition 6 is in fact
unnecessary.
Theorem 7 Let A ∈ Mt×∞ (S>0) be an array with associated weight function
σ. The following are equivalent:
(1) A is an admissible array;
(2) for all k ≥ 0 and for all ω1, ω2 ∈ B× such that ω1 = ω2, σk(ω1) > σk(ω2)
if and only if σk+1(ω1) > σk+1(ω2).
3
Admissible arrays define an admissible order on the quotient R/IA.
Theorem 8 An array A ∈ Mt×∞ (S>0) with associated weight function σ is
an admissible array if and only if ≻σ is an admissible order on Bσ ⊆ R/IA,
where Bσ is the image of B in R/IA under the projection R → R/IA.
Definition 9 A is said to be degenerate if there exists i, j, 1 ≤ i 6= j ≤ t,
such that σ(xi) = σ(xj).
We will assume in what follows that all arrays considered are nondegenerate,
for if σ(xi) = σ(xj) for some i, j ∈ {1, . . . , t} where i 6= j, then xi − xj ∈ IA
and khx1, . . . , xti/IA ≃ khx1, . . . , xi−1, xi+1, . . . , xti/hI ′
Ai where A′ is the array
obtained from A by deleting the ith row.
2 Weight Ideals Associated to Regular Arrays
In Hinson (2010), E. Hinson described two sets of admissible arrays. We begin
by studying the first of these, the set of regular arrays.
Definition 10 An array A is regular if A has rank 1.
The set of linear arrays is a subset of the set of regular arrays which will be
used later on to construct the set of log-linear arrays.
Definition 11 An array A is linear if for all i ≥ 1, A(i) = d · A(i−1) for some
fixed d ∈ S>0. The fixed scalar d is referred to as the slope of the array.
Example 12 The array
A =
2 6 18 · · ·
3 9 27 · · ·
4 12 36 · · ·
is a linear array with slope d = 3.
The weight ideal associated to a regular array contains the commutator ideal
C = hxixj − xjxi1 ≤ i 6= j ≤ ti, and thus is never trivial (Hinson, 2010).
Definition 13 The support of a word ω is the set
supp(ω) = {xii ∈ {1, . . . , t} and xi occurs in ω}.
Definition 14 The frequency of x in ω is the number of times that x occurs
in ω and is written #(x, ω).
4
Definition 15 Let f ∈ R and G = {g1, g2, . . . } ⊂ R. We say that f is an
algebraic consequence of G if f = Pg∈G ciuigivi, where ci ∈ k, ui, vi ∈ R, and
only finitely many ci 6= 0.
Suppose A is a regular array with first column [a1,0, . . . , at,0]T , where ai,0 ∈
N and at least one of (ai,0, aj,0) 6= 1, where (ai,0, aj,0) denotes the greatest
common divisor of ai,0 and aj,0 and 1 ≤ i 6= j ≤ t. Let ω1 = xu0 · · · xul−1 and
ω2 = xv0 . . . xvl−1 ∈ B such that ω1 − ω2 ∈ IA. Then we have
l−1
Y
i=0
auii =
l−1
Y
i=0
avii,
and each auii and avii can be written as scalar multiples of aui0 and avi0
respectively:
l−1
Y
i=0
diaui0 =
l−1
Y
i=0
diavi0.
Factoring out and canceling the common di's reduces the equation to
l−1
Y
i=0
aui0 =
l−1
Y
i=0
avi0.
(1)
Equation (1) does not depend on the how the variables were ordered in ω1
and ω2; in particular, by factoring out and canceling any terms aui0 = avj 0
common to both sides of the equation, one obtains the reduced expression
n
Y
k=0
auk0 =
n
Y
k=0
avk0.
(2)
In this expression, aum0 6= avm′ 0 for all um and vm′. Note that we have not
cancelled any common divisors of the aui,0, we have only cancelled those aui,0's
and avj ,0's for which aui,0 = avj ,0. Since each aum0 and avm′ 0 corresponds to the
weight assigned to an individual letter in {x1, . . . , xt}, this equation describes
2 ∈ IA in which no letter that occurs
a homogeneous binomial difference ω ′
in ω ′
1 will occur in ω ′
2.
1 − ω ′
Definition 16 A homogeneous binomial difference ω1 − ω2 ∈ IA for which
supp(ω1)\ supp(ω2) = ∅
will be referred to as a homogeneous binomial difference of disjoint support.
Homogeneous binomial differences of disjoint support may arise as algebraic
consequences of other homogeneous binomial differences of disjoint support.
5
For example, suppose
A =
2 4 8 · · ·
3 6 12 · · ·
4 8 16 · · ·
6 12 24 · · ·
.
This array A is linear with slope 2. Consider the homogeneous binomial dif-
ference x3x2x3x2 − x4x1x4x1. Since
σ(x3x2x3x2) = σ(x4x1x4x1) = 9216,
we must have x3x2x3x2 − x4x1x4x1 ∈ IA. Neither word in this homogeneous
difference shares a letter with the other, so x3x2x3x2 − x4x1x4x1 is a homoge-
neous binomial difference of disjoint support. Furthermore,
x3x2x3x2 − x4x1x4x1 = (x3x2 − x4x1)x3x2 + x4x1(x3x2 − x4x1),
so x3x2x3x2 − x4x1x4x1 is a homogeneous binomial difference of disjoint sup-
port which arises as an algebraic consequence of a homogeneous binomial
difference of disjoint support consisting of words of lesser length.
Definition 17 Let ω1 − ω2 be a homogeneous binomial difference of disjoint
support. ω1 − ω2 is minimal if any expression
ω1 − ω2 =
n
X
i=1
αi(ui − vi)βi
for ω1 − ω2 as a sum of homogeneous binomial differences has at least one
difference ui − vi such that ui = vi = w1. MA will be used to denote
the set of minimal length homogeneous binomial differences of disjoint support
associated to A.
In other words, a minimal homogeneous binomial difference of disjoint support
is one which cannot be realized as an algebraic consequence of homogeneous
binomial differences of disjoint support consisting of words of lesser length.
Any element of IA may be decomposed over the set of commutators {xixj −
xjxi1 ≤ i 6= j ≤ t} and the set of homogeneous binomial differences of disjoint
support.
Lemma 18 Let ω1 − ω2 be a homogeneous binomial difference in IA. Then
ω1 − ω2 = Pn
i=1 αi(ui − vi)βi, where each homogeneous binomial difference
ui − vi, 1 ≤ i < n is a commutator and un − vn is a homogeneous binomial
difference of disjoint support.
6
PROOF. Suppose ω1 − ω2 ∈ IA. We proceed by induction. The base case
when l = 2 is established trivially. Assume now that the induction hypothesis
holds for homogeneous binomial differences consisting of words of length l − 1
and suppose ω1 = ω2 = l. Write ω1 = xu0 . . . xul−1 and ω2 = xv0 . . . xvl−1.
Let i ∈ {0, . . . , l − 1} be the least value for which xu0 = xvi (if no such value
exists, we are done). By inserting the expression
−xv0 . . . xvi−2xvixvi−1xvi+1 . . . xvl−1 + xv0 . . . xvi−2xvixvi−1xvi+1 . . . xvl−1,
we obtain
ω1 − xv0 . . . xvi−2xvixvi−1xvi+1 . . . xvl−1 + xv0 . . . xvi−2xvixvi−1xvi+1 . . . xvl−1 − ω2,
which is equal to
ω1 − xv0 . . . xvi−2xvixvi−1xvi+1 . . . xvl−1+
xv0 . . . xvi−2 (cid:16)xvixvi−1 − xvi−1xvi(cid:17) xvi+1 . . . xvl−1.
(3)
In the second term in expression (3), xvi occurs in the i − 1st position. The
third and fourth terms in Equation 3 have been expressed as (left and right)
multiples of the commutator xvixvi−1 − xvi−1xvi. Iterating this process i times
results in the expression
ω1 − xvixv0 . . . xvi−1xvi+1 . . . xvl−1 +
i−1
X
k=1
αk(xvixvi−k − xvi−k xvi)βk,
(4)
where αk = xv0 . . . xvi−k−1 and βk = xvi−k+1 . . . xvl−1.
Since xu0 = xvi, the difference of the first two terms in 4 can be rewritten as
xu0 (cid:16)xu1 . . . xul−1 − xv1 . . . xvl−1(cid:17) .
The expression in parentheses consists of monomials of length l − 1 which
is an algebraic consequence of the commutators and a homogeneous binomial
difference of disjoint support. Rearranging and renaming terms as needed gives
the desired result. ✷
Theorem 19 Let A be a regular array. The weight ideal IA associated to a
regular array A is generated by the union of the set of commutators {xixj −
xjxi1 ≤ i 6= j ≤ t} and MA.
7
PROOF. Fix a homogeneous binomial difference ω1 − ω2 ∈ IA. By iterating
the algorithm described in the proof of Lemma 18, ω1−ω2 ∈ IA can be reduced
to an algebraic consequence of the commutators plus a single, perhaps trivial,
homogeneous binomial difference of disjoint support ω ′
2. To see this,
note that each iteration of the algorithm produces in the sum a difference of
commutators and a homogeneous binomial difference of shorter length than
in the previous iteration in which a letter common to each word has been
extracted. We may continue the algorithm until either the next iteration is
over a commutator or there are no common letters to extract. In the first
case, we are done, and in the second case, if ω ′
2 is minimal, we are also
2 is not a minimal homogeneous binomial difference, then by
done. If ω ′
definition it is an algebraic consequence of minimal homogeneous binomial
differences of disjoint support. ✷
1 − ω ′
1 − ω ′
1 − ω ′
Having obtained a description of the generators of IA, we will next show that
when A is regular, IA is finitely generated. We include the following lemma
to describe the means by which a disjoint homogeneous binomial difference
which contains another difference as scattered subwords can be decomposed
over that subdifference.
Lemma 20 Let ω1−ω2 ∈ MA and suppose λ1−λ2 is a homogeneous binomial
difference of disjoint support such that ω1 occurs as a scattered subword in λ1
and ω2 occurs as a scattered subword in λ2. Then
λ1 − λ2 = (ω1 − ω2)α + ω2(α − β) +
n
X
i=1
αi(γi − ζi)βi,
where α−β is a homogeneous binomial difference of disjoint support and γi−ζi
is a commutator for each i, 1 ≤ i ≤ n.
PROOF. The algorithm of Lemma 18 may be modified to move any letter
that occurs in a word in a homogeneous binomial difference in IA either forward
or backwards to the desired position, resulting in a decomposition
λ1 − λ2 = ω1α − ω2β +
n
X
i=1
αi(γi − ζi)βi,
where γi − ζi is a commutator, 1 ≤ i ≤ n. The result then follows.
Theorem 21 Let A be a regular array. The associated weight ideal IA is
finitely generated.
PROOF. By Theorem 19, IA is generated by the union of the set of commu-
tators and the set MA of minimal homogeneous binomial differences of disjoint
8
support. The set of commutators is clearly finite. It remains to demonstrate
that MA is also finite. Assume the contrary. Then there exists some parti-
tion of X = {x1, . . . , xt} into two sets X1, X2 such that there are infinitely
many minimal disjoint homogeneous binomial differences ω1 − ω2 in which
supp(ω1) ⊆ X1 and supp(ω2) ⊆ X2. Let D = {ω1 − ω2 ∈ MAsupp(ω1) ∈
X1, supp(ω2) ∈ X2} and let ω1 − ω2 ∈ D such that ω1 ≤ λ1 for any λ1 that
occurs in a homogeneous binomial difference λ1 −λ2 ∈ D. Consider the follow-
ing three sets: D(ω1) = {λ1−λ2 ∈ D : ω1 occurs as a scattered subword in λ1},
D(ω2) = {λ1 − λ2 ∈ D : ω2 occurs as a scattered subword in λ2}, and D(0) =
{λ1 − λ2 ∈ D : neither ω1 nor ω2 occur as scattered subwords in λ1 and λ2}.
Note that D = {ω1 − ω2} ∪ D(ω1) ∪ D(ω2) ∪ D(0). Furthermore, these sets are
disjoint. If ω1 were to occur as a scattered subword in λ1 and ω2 occurs as a
scattered subword of λ2, then Lemma 20 shows that λ1 − λ2 is an algebraic
consequence of commutators, ω1 − ω2, and perhaps some other homogeneous
binomial difference in D consisting of words of length less than λ1; that is,
λ1 − λ2 is not minimal. Thus, these sets form a partition of D and so at least
one of D(ω1), D(ω2), and D(0) must be infinite.
Now let λ1 − λ2 ∈ D(ω2) and suppose λ1 > ω1. Since λ1 does not contain ω1
as a scattered subword, the number of occurrences ki of some variable xi in λ1
must be less than in ω1, so the number of occurrences kj of some other variable
xj must be greater than the number of occurrences in ω1. Suppose D(ω2) is
1 > λ1,
infinite. Then there exists a difference λ′
1.
and furthermore, neither λ1 nor ω1 can occur as scattered subwords in λ′
Thus the number of occurrences ki′ of another variable xi′ must be less than
in ω1, and so the number of occurrences kj ′ of another variable xj ′ must be
greater than in ω1. This indicates that D(ω2) cannot be infinite: for some l,
any homogeneous binomial difference γ1 − γ2 ∈ D(ω2) such that γ1 > l must
have a first word which contains as a scattered subword some word ¯λ1 which
previously occurred in a homogeneous binomial difference ¯λ1 − ¯λ2 ∈ D(ω2)
and is thus not minimal. The same argument, mutatis mutandis, shows that
D(ω1) is also finite.
2 ∈ D(ω2) with λ′
1 − λ′
1 − ω ′
1 − ω ′
1 and ω ′
1), D(ω ′
2 ∈ D(0) be such that ω ′
1 ≤ λ1 for
Consider, then, the set D(0). Let ω ′
any λ1 − λ2 ∈ D(0). Note that ω ′
2 must consist of words at least as long
2 some variables xk1 and xk2 must
as ω1, and furthermore, in both ω ′
occur less often than in ω1 and ω2 respectively. We may partition D(0) into
2), and D(0′) which form a partition of D(0). As above, these
sets D(ω ′
sets form a partition of D(0), and following the argument above, both D(ω ′
1)
2) are finite. Consider then D(0′), which must be infinite, and select a
and D(ω ′
1 ≤ λ1 for any λ1 − λ2 ∈ D(0′). Again
difference ω ′′
ω ′′
1, and furthermore, in both
1 and ω ′
ω ′′
2
respectively. Continuing this partitioning process ad infinitum is impossible:
for some l, any difference γ1−γ2 such that γ1 > l must contain the occurrence
2 must consist of words at least as long as ω ′
must occur less often than in ω ′
1 − ω ′′
2 ∈ D(0′) such that ω ′′
2 some variables xk′
1
1 − ω ′′
1 and ω ′′
and xk′
2
9
of some ¯λi, i ∈ {1, 2}, which previously occurred in a homogeneous binomial
difference in ¯λ1 − ¯λ2 ∈ D(0) as a scattered subword. Thus D(0) cannot be
infinite, and so MA is finite and IA must be finitely generated. ✷
We have the following corollaries. Corollary 23 gives a description of those ho-
mogeneous binomial differences in the commutator ideal. Note that necessity
in Corollary 23 was proved in Hinson (2010).
Corollary 22 Let A be a regular array with pairwise-coprime first column
entries. Then IA = C, where C denotes the commutator ideal.
PROOF. Since the entries in the first column of A are pairwise-coprime, MA
is trivial. ✷
Corollary 23 Let A be a regular array with pairwise-coprime first column
entries, and suppose ω1, ω2 ∈ B with ω1 = ω2 = l. Then ω1 − ω2 ∈ IA ⇐⇒
supp(ω1) = supp(ω2) and #(xi, ω1) = #(xi, ω2) for all xi ∈ supp(ω1) =
supp(ω2).
PROOF. To prove sufficiency, let
A =
a1,0 d1a1,0 · · · dna1,0 · · ·
a2,0 d1a2,0 · · · dna2,0 · · ·
...
at,0 d1at,0 · · · dnat,0 · · ·
...
...
.
Assume that ω1 − ω2 ∈ IA. Then
σ(ω1) = σ(ω2) ⇒
l−1
Y
i=0
auii =
l−1
Y
i=0
avii.
Expressing each weight as a multiple of a first-column entry and canceling the
di's common to each side of the equation gives
l−1
Y
i=0
aui0 =
l−1
Y
i=0
avi0.
(5)
Since the first column entries of A are pairwise-coprime, equality can only hold
i=0 and {avi0}l−1
in Equation (5) when there exists a bijection between {aui0}l−1
i=0.
10
Each aui0 corresponds to an occurrence of the letter xui in ω1; thus, supp(ω1) =
supp(ω2), and because the weights are equal, #(xi, ω1) = #(xi, ω2) for each
xi ∈ supp(ω1) = supp(ω2). ✷
In particular, when a regular array A has pairwise-coprime first column entries,
then the algebra R/IA on which it defines an order is in fact isomorphic to
the commutative polynomial algebra k[x1, · · · , xt]. The preceding construction
can be viewed as an alternative way to specify an admissible length-dominant
weight order on the monomials in k[x1, . . . , xt]. L. Robbiano has proven that
any such order is a lexicographic product of weight orders (Robbiano, 1986).
More generally, in Gilmer (1984), R. Gilmer proved that monoid algebras of
commutative monoids are precisely the homomorphic images of polynomial
rings by ideals which are generated by pure binomial differences. Of course,
the algebra R/IA is precisely such an algebra when A is regular. Thus, we
can view R/IA as a monoid algebra of a commutative monoid for which an
admissible order on the basis of R/IA can be obtained.
While Theorem 19 shows how to decompose a homogeneous binomial differ-
ence over the set of commutators and MA and Theorem 21 demonstrates that
MA is finite, constructing MA may present significant computational difficul-
ties. Neglecting minimality, the process of directly identifying a homogeneous
binomial difference of disjoint support consisting of words of length l by calcu-
lating weights is easily seen to be equivalent to the Subset Product problem,
which is known to be NP-complete (Garey & Johnson, 1979). Furthermore,
the above proofs do not give a bound on the lengths of the words that may
occur in a minimal homogeneous binomial difference of disjoint support, sug-
gesting that even if one is able to devise an algorithm to efficiently identify
minimal homogeneous binomial differences of disjoint support, one could not
terminate the algorithm and be satisfied that all the minimal homogeneous
binomial differences of disjoint support had been enumerated.
3 Weight Ideals Associated to Log-Linear Arrays
Bijectively related to the family of linear arrays is the family of log-linear
arrays.
Definition 24 An array A = (aij) ∈ Mt×∞ (S>0) is log-linear if the array
log A = (log (aij)) is linear.
11
Example 25 The array
B =
e2 e6 e18 · · ·
e3 e9 e27 · · ·
e4 e12 e36 · · ·
is a log-linear array, because
log B =
2 6 18 · · ·
3 9 27 · · ·
4 12 36 · · ·
is a linear array (with slope d = 3).
Note that an array A for which log A is regular but not linear need not be
admissible. Because every log-linear array with slope 1 is in fact a (constant)
regular array, we will assume without comment in the remainder that any log-
linear array considered has slope d 6= 1. We point out also that the base of a
log-linear array is immaterial in determining the order given by the array. To
see this, let A ∈ Mt×∞ (S>0) be a given linear array with i, jth entry djai,0 and
consider the arrays B = (bij) and C = (cij), where bij = bdj ai,0 and cij = cdjai,0
for elements b, c ∈ S>0 with b 6= c. Suppose that ω1 = xu0xu1 . . . xul−1 and
ω2 = xv0xv1 . . . xvl−1 are words in X ∗ such that ω1 ≻B ω2. Then
σB(ω1) > σB(ω2)
which means that
l−1
Y
k=0
buk,k >
l−1
Y
i=0
bvk,k.
Of course, this is equivalent to the inequality
k=0 dkauk > bPl−1
bPl−1
k=0 dkavk ,
and replacing b with c does not change the direction of the inequality. Thus,
we will typically assume without comment that the base of a log-linear array
is e. As the preceding discussion indicates, the significant distinction between
regular arrays and log-linear arrays is that when working with regular arrays,
the weight associated to a word is calculated by multiplying the weights given
to each variable in their respective positions, while when working with log-
linear arrays, the weight associated to a word is calculated by adding the
weights given to each variable in their respective positions. In particular, when
working with a log-linear array A, the equation Pl−1
k=0 dkavk must
be satisfied for a homogeneous binomial difference ω1 − ω2 = xu0xu1 . . . xul−1 −
xv0xv1 . . . xvl−1 to be a member of IA.
k=0 dkauk = Pl−1
12
The structure of log-linear arrays is much less uniform than that of regular
arrays. Theorem 26 is the main result of this section and is proved via the
examples that follow. We will also demonstrate that log-linear arrays can be
constructed to give orders on R which are equivalent to the familiar left and
right length-lexicographic orders.
Theorem 26 There exist log-linear arrays whose associated weight ideals are
trivial, log-linear arrays whose associated weight ideals admit a finite gener-
ating set, and log-linear arrays whose associated weight ideals do not admit a
finite generating set.
It would be of interest to find necessary and sufficient conditions on a log-linear
array A such that IA is trivial, is nontrivial but admits a finite generating set,
or is nontrivial and does not admit a finite generating set.
The following two lemmas describe distinct arrays that define orders on R
which are equivalent to left and right length-lexicographic order respectively.
The hypotheses on the array A is sufficient to insure in each case that IA is
trivial.
Lemma 27 Suppose the variables x1, . . . , xt are ordered. Let A ∈ Mt×∞ (S>0)
be log-linear with first column A(0) = [ea1,0, . . . , eat,0]T for which the values of the
first-column entries of A reflect the order given to x1, . . . , xt; that is, xi1 < xi2
if and only if ai1,0 < ai2,0 also. Let α and β denote the minimum and maximum
nonzero first column differences of log A respectively; that is,
α = min{ai,0 − aj,0 : 1 ≤ i, j ≤ t, i 6= j},
and
β = max{ai,0 − aj,0 : 1 ≤ i, j ≤ t, i 6= j}.
Let d be the slope of log A. If d < 1 and α > dβ/(1 − d), then IA is trivial and
the order given by A is the left length-lexicographic order.
PROOF. Assume the hypotheses, and assume that ω1 = xu0xu1 . . . xul−1 and
ω2 = xv0xv1 . . . xvl−1 are two words of equal length in R such that ω1 −ω2 ∈ IA.
This implies that
eau0 ,0+dau1 ,0+···+dl−1aul−1,0 = eav0 ,0+dav1 ,0+···+dl−1avl−1 ,0,
(6)
and thus
au0,0 + dau1,0 + · · · + dl−1aul−1,0 = av0,0 + dav1,0 + · · · + dl−1avl−1,0.
13
This in turn implies that
au0,0 − av0,0 =
l−1
X
i=1
di (aui,0 − avi,0) .
(7)
Suppose now that the first letters of ω1 and ω2 differ. The largest in absolute
value that the right-hand side of Equation 7 can be is when each difference
aui,0 −avi,0 = β, so the right-hand side has an upper bound at β(d−dl)/(1−d).
The smallest the left-hand side of Equation 7 can be in absolute value is when
au0,0 − av0,0 = α, and by hypothesis, α > dβ/(1 − d) > (d − dl)β/(1 − d) for
all l. This contradicts the assumption that ω1 − ω2 ∈ IA, so in fact IA = {0}.
Furthermore, the difference au0,0 − av0,0 is greater in absolute value than any
possible subsequent sum and hence determines the order between ω1 and ω2.
Now, note that if the first k letters of ω1 and ω2 are the same, then those first
k letters contribute the same expression to either side of Equation 6, and thus
play no role in determining the order between ω1 and ω2. Thus, when the first
k letters are the same, we may determine the order between ω1 and ω2 by
simply applying the above argument to the truncated words xuk xuk+1 · · · xul−1
and xvk xvk+1 · · · xvl−1 to note that the order on ω1 and ω2 is determined solely
by the order between auk,0 and avk,0.
Because the order on the first column entries of log A is equivalent to the
order on the variables x1, . . . , xt the order given by A is thus the left length-
lexicographic order. ✷
Lemma 28 Suppose the variables x1, . . . , xt are ordered. Let A ∈ Mt×∞ (S>0)
be log-linear with first column A(0) = [ea1,0, . . . , eat,0]T for which the values of the
first column entries of A reflect the order given to x1, . . . , xt; that is, xi1 < xi2
if and only if ai1,0 < ai2,0 also. Let α and β denote the minimum and maximum
nonzero first column differences of log A respectively; that is,
α = min{ai,0 − aj,0 : 1 ≤ i, j ≤ t, i 6= j},
and
β = max{ai,0 − aj,0 : 1 ≤ i, j ≤ t, i 6= j}.
If d > 1 and α > β/(d − 1), then IA is trivial and the order given by A is the
right length-lexicographic order.
PROOF. The same argument as in Lemma 27 applies, mutatis mutandis.
Orders constructed via admissible arrays with trivial weight ideals are sim-
ply admissible length-dominant orders on R. A set of invariants that fully
characterize the admissible orders that can be defined on a noncommutative
14
k-algebra such as R has not yet been described, though results in this direc-
tion have been obtained (Scott, 1994; Perlo-Freeman & Prohle, 1997). It is
possible that array-based admissible orders may be of use in defining such a
set of invariants.
We turn our attention next to an example of a log-linear array A for which IA
is nontrivial and admits a finite generating set.
Example 29 Let A be the log-linear array such that
log A =
2 4 8 . . .
3 6 12 . . .
4 8 16 . . .
6 12 24 . . .
.
The weight ideal IA associated to A is nontrivial and is finitely generated.
Clearly the weight ideal associated to A is nontrivial; for example, x1x2 −
x3x1 ∈ IA. Interestingly, any homogeneous binomial difference of length l > 2
in IA can be reduced in at most two steps to an algebraic consequence of
homogeneous binomial differences consisting of words whose maximum length
is l − 1. It follows inductively that any homogeneous binomial difference of
length l can be reduced to an algebraic consequence of homogeneous binomial
differences of length 2; that is, for this particular A, the homogeneous binomial
differences of length 2 in fact generate IA. The proof of this proposition is
straightforward but relies on a lengthy case-by-case analysis, which is included
as an appendix.
The next example demonstrates that there exists log-linear arrays with a non-
trivial associated weight ideal which admits no finite generating set.
Example 30 Let A be the log-linear array given by
log A =
2 4 8 . . .
4 8 16 . . .
7 14 28 . . .
.
The weight ideal IA associated to A is nontrivial and does not admit a finite
generating set.
In the proof we will use the term factor to indicate a subword in which the
letters occur consecutively, in order to alleviate any potential confusion with
scattered subwords, which some authors refer to simply as subwords.
15
PROOF. Consider the homogeneous binomial difference
ω1 − ω2 := x2xn
3 x2 − x1(x2x3)(n−2)/2x1x2x3,
where n ≥ 4 is an even integer and l = n + 2 is the length of ω1 and ω2. We
will show first that for any such n, the difference given above is a member of
IA, and then we will demonstrate that ω1 − ω2 is not an algebraic consequence
of the shorter length differences in IA and so must belong to any generating
set for IA. Since this holds for all n ≥ 4, this will prove that IA does not admit
a finite generating set.
To demonstrate that x2xn
3 x2 − x1(x2x3)(n−2)/2x1x2x3 ∈ IA for any n ≥ 4, let us
calculate the difference of the weights associated to ω1 −ω2 respectively. Fix an
even integer n ≥ 4. The difference in weights associated to any homogeneous
binomial difference by A can be expressed as a polynomial:
∆ := a0 + da1 + d2a2 + · · · + dl−1al−1,
where ak is the difference of the first-column entries associated to the letter
in position k in ω1 and ω2 respectively. For the given difference,
∆ω1−ω2 = 2+2·3+22 ·0+23 ·3+24 ·0+· · ·+2l−4 ·0+2l−3 ·4+2l−2 ·3+2l−1 ·(−3).
To show that ∆ω1−ω2 = 0 regardless of the value of n, it is easiest to work in
binary. We have
10 + 10 · 11 + 1000 · 11 + 100000 · 11 + · · ·
·11 + 1 0 . . . 0
{z }
+ 1 0 . . . 0
{z }
Multiplying simplifies this to
l−5
l−3
·101 + 1 0 . . . 0
{z }
l−2
·11 − 1 0 . . . 0
{z }
l−1
·11.
10 + 110 + 11000 + 1100000 + · · · + 11 0 . . . 0
{z }
l−3
+101 0 . . . 0
{z }
l−3
+11 0 . . . 0
{z }
l−2
−11 0 . . . 0
{z }
l−1
.
This expression is equal to 0:
1 0 . . . 0
{z }
l−3
+101 0 . . . 0
{z }
l−3
+11 0 . . . 0
{z }
l−2
−11 0 . . . 0
{z }
l−1
and so
11 0 . . . 0
{z }
l−2
11 0 . . . 0
{z }
3 x2 − x1(x2x3)(n−2)/2x1x2x3 ∈ IA. To show that
3 x2 − x1(x2x3)(n−2)/2x1x2x3 must be contained in any generating set for
+11 0 . . . 0
{z }
−11 0 . . . 0
{z }
This demonstrates that x2xn
x2xn
−11 0 . . . 0
{z }
l−1
= 0.
=
l−2
l−1
l−1
16
IA, we will show that the word x2xn
3 x2 contains no factor that occurs as a word
in a homogeneous binomial difference of shorter length in IA. In particular, this
implies that x2xn
3 x2 − x1(x2x3)(n−2)/2x1x2x3 cannot be written as an algebraic
consequence of strictly shorter length homogeneous differences in IA.
3x2, and xk
3, factors xk
Consider the possible factors of the word x2xn
3 x2. For each k, 4 ≤ k ≤ n, we
have factors x2xk
3. No factor that occurs in a homogeneous
binomial difference in IA can begin with x3, because of the parity of the weight
that results, so we can rule out as possibilities any factors of the form xk
3 and
xk
3x2. The weight given the factor x2xk
3 will be greater than the weight assigned
any other word of equal length except xk
3, and it will not equal this weight.
Thus, x2xn
3 x2 contains no factors that occur as a word in a homogeneous
binomial difference in IA. Because this holds for each n ≥ 4, the difference
x2xn
3 x2 − x1(x2x3)(n−2)/2x1x2x3 must be included in any generating set for IA,
and thus any generating set for IA is infinite. ✷
A Proof that the Array in Example 29 is Finitely Generated
PROOF. Let us first list the the homogeneous binomial differences of length
two that occur in IA. They are x1x2 − x3x1, x1x3 − x3x2, x1x3 − x4x1, x3x2 −
x4x1, x3x3 − x4x2, and x1x4 − x4x3. Given a homogeneous binomial difference
consisting of words of length l in IA, we will show that it either decomposes
over the homogeneous binomial differences in IA of length two or fails to belong
to IA. Proof of the latter claim requires us to consider the existence of solutions
to the polynomial
a0 + da1 + · · · + dl−1al−1 = 0
which corresponds to a given homogenous binomial difference in IA. Any so-
lution to this equation is an element of the set `l−1
i=0 Ad = {(a0, . . . , al−1)ai =
aj,0 − ak,0,1 ≤ j, k ≤ 4}.
∗←→ λ1
∗←→ · · ·
∗←→ λn
To simplify exposition, we use the notation of rewriting relations on words in
the free monoid X ∗ := hx1, . . . , xti. We define the following rewriting relation:
∗←→ ω2 to denote that ω1 − ω2 ∈ IA. A particular
for ω1, ω2 ∈ X ∗, we write ω1
∗←→ ω2 corresponds to a
chain of rewritings ω1
unique decomposition of ω1 −ω2 in R (Madlener & Reinert, 1997), though this
decomposition need not be over homogeneous binomial differences consisting
∗←→ λ where at
of words of lesser length. However, a chain of rewritings ω1
least the first letter of λ is the same as the first letter of ω2 does correspond
to a unique decomposition of ω1 − ω2 over the set of homogeneous binomial
differences in IA whose words are of lesser length. Rather than calculate the
decomposition precisely, we will rewriting to indicate that a decomposition is
possible.
17
Organizing by weight, the length two homogeneous binomial differences in IA
give rise to the following rewriting relations:
x1x2
∗←→ x3x1,
x1x3
∗←→ x3x2
∗←→ x4x1,
x3x3
∗←→ x4x2,
x1x4
∗←→ x4x3.
Suppose that ω1−ω2 ∈ IA, with ω1 = ω2 = l. Let ω1 = xu0 . . . xul−1 and ω2 =
xv0 . . . xvl−1. By parity, if either ω1 or ω2 start with x2, then so must the other,
in which case ω1 − ω2 is immediately reducible to a homogeneous binomial
difference of length l − 1. Thus we may assume without loss of generality that
neither ω1 nor ω2 start with x2.
Assume next that ω1 begins with x1. We need to consider the cases when ω2
starts with x3 or x4, as well as the case when ω1 begins with x3 and ω2 begins
with x4. All other cases will then be captured by symmetry.
If ω1 − ω2 is not immediately reducible, then ω2 must begin with either x3 or
x4.
Case 1 : ω2 begins with x3.
Subcase 1.1 : ω2 begins with x3x1.
x3x1
∗←→ x1x2, so ω1 − ω2 is reducible after this single rewrite.
Subcase 1.2 : ω2 begins with x3x2.
x3x2
∗←→ x1x3, so ω1 − ω2 is reducible after this single rewrite.
Subcase 1.3 : ω2 begins with x3x3.
In this case, reduction with a single rewrite of ω2 is not always possible. If
the second letter of ω1 is x2 or x3, then we can rewrite ω1 to reduce ω1 − ω2.
Suppose then that the second letter of ω1 is x4. Rewrite as follows:
x3x3
∗←→ x4x2 and x1x4
∗←→ x4x3.
Now first letters agree and ω1 − ω2 is reducible. Finally, suppose that the
second letter of ω1 is x1. Then ω1 − ω2 is of the form
x1x1xu2 . . . xul−1 − x3x3xv2 . . . xvl−1.
Since ω1 − ω2 ∈ IA, this gives rise to the equation a0 + 2a1 + · · · + 2l−1al−1 = 0,
where a0 is a difference of first-column entries of IA determined by the letters
18
of ω1 and ω2. In particular, a0 = −2 and a1 = −2, so
−2 − 4 + 22a2 + · · · + 2l−1al−1 = 0,
or equivalently,
2a2 + · · · + 2l−2al−1 = 3,
(A.1)
but (A.1) does not have a solution over `l−1
Subcase 1.4 : ω2 begins with x3x4.
i=0 Ad.
Again, if the second letter of ω1 is x2 or x3, then ω1 − ω2 is immediately
reducible. Suppose the second letter of ω1 is x1. Then ω1 − ω2 is of the form
x1x1xu2 . . . xul−1 − x3x4xv2 . . . xvl−1.
Since ω1 − ω2 ∈ IA, this gives rise to the equation a0 + 2a1 + · · · + 2l−1al−1 = 0,
where each ai is a difference of first column entries of A determined by the
letters of ω1 and ω2. In particular, a0 = −2 and a1 = −4, so
−2 − 8 + 22a2 + · · · + 2l−1al−1 = 0,
or equivalently,
2a2 + · · · + 2l−2al−1 = 5,
(A.2)
but (A.2) does not have a solution over `l−1
Finally, if the second letter of ω1 is x4, then ω1 − ω2 is of the form
i=0 Ad.
x1x4xu2 . . . xul−1 − x3x4xv2 . . . xvl−1.
Since ω1 − ω2 ∈ IA, this gives rise to the equation a0 + 2a1 + · · · + 2l−1al−1 = 0,
where each ai is a difference of first column entries of A determined by the
letters of ω1 and ω2. In particular, a0 = −2 and a1 = 0, so
−2 − 0 + 22a2 + · · · + 2l−1al−1 = 0,
or equivalently,
2a2 + · · · + 2l−2al−1 = 1,
(A.3)
but (A.3) does not have a solution over `l−1
Case 2: ω1 begins with x1 and ω2 begins with x4.
i=0 Ad.
Subcase 2.1: ω1 begins with x1 and ω2 begins with x4x1.
x4x1
∗←→ x1x3, so ω1 − ω2 is reducible after this single rewrite.
Subcase 2.2: ω1 begins with x1 and ω2 begins with x4x2.
19
∗←→ x4x3 to reduce ω1−ω2. If ω1 begins with x1x2, rewrite x1x2
∗←→ x4x1 and immediately
If ω1 begins with x1x3, then we can rewrite x1x3
reduce ω1 − ω2. Similarly, if ω1 begins with x1x4, we can immediately rewrite
∗←→ x3x1
x1x4
∗←→ x3x3 to reduce ω1 −ω2. The only remaining possibility is that ω1
and x4x2
begins with x1x1 and ω2 begins with x4x2, but the corresponding polynomial
shows that no such difference that begins with these letters can belong to IA:
−4 + 2 · (−1) + 22a2 + · · · + 2l−1al−1 = 0
reduces to
−3 + 2a2 + · · · + 2l−2al−1 = 0,
and this equation has no solution over `l−1
Subcase 2.3: ω1 begins with x1 and ω2 begins with x4x3.
i=0 Ad.
x4x3
∗←→ x1x4, so ω1 − ω2 is reducible after this single rewrite.
Subcase 2.4: ω1 begins with x1 and ω2 begins with x4x4.
Subsubcase 2.4.1: ω1 begins with x1x1.
Consider the polynomial equation a0 + 2a1 + · · · + 2l−1al−1 = 0 corresponding
to this difference. The choice of first letters for ω1 and ω2 determine a0 = −4
and a1 = −4. Factoring out 4 from the resulting equation gives
−3 + a2 + 2a3 + · · · + 2l−3al−1 = 0.
∗←→ x1x3x1
∗←→ x1x4x3
Each term after the second in the expression on the left -- hand side of the
above equation is congruent to 0 mod 2, thus this equation has a solution
only if a2 = ±3 or ±1 (each ai ∈ {±1, ±2, ±3 ± 4}. There are thus four
possible subcases to consider. If a2 = −3, then the first three letters of ω1 are
∗←→ x4x1x1, so ω1 − ω2 is reducible. If a2 = 3,
x1x1x2, and x2x1x2
∗←→ x4x4x3,
the first three letters of ω1 are x1x1x4, and x1x1x4
so ω1 − ω2 is reducible. If a2 = −1 then either ω1 begins with x1x1x1 or
∗←→
x1x1x2. In the first case, ω2 must therefore begin with x4x4x2, and x4x4x2
∗←→ x1x4x3, so the difference is reducible, and in the second case,
x4x3x3
∗←→ x1x3x1
∗←→ x4x1x1, so the difference is reducible. If a2 = 1, then
x1x1x2
either ω1 begins with x1x1x2 or ω1 begins with x1x1x3. The first case has
already been addressed, and the second case gives rise to a reducible instance
of ω1 − ω2, for x1x1x3
∗←→ x4x1x2.
∗←→ x1x3x2
Subsubcase 2.4.2: ω1 begins with x1x2.
The polynomial equation corresponding to this difference is
−4 + 2(−3) + · · · + 2l−1al−1 = 0,
20
and this equation has no solution over `l−1
Subsubcase 2.4.3: ω1 begins with x1x3.
i=0 Ad.
x1x3
∗←→ x4x3, so ω1 − ω2 is reducible after this single rewrite.
Case 3: ω1 begins with x3 and ω2 begins with x4.
Subcase 3.1: ω1 begins with x3 and ω2 begins with x4x1.
x4x1
∗←→ x3x2, so ω1 − ω2 is reducible after this single rewrite.
Subcase 3.2: ω1 begins with x3 and ω2 begins with x4x2.
x4x2
∗←→ x3x3, so ω1 − ω2 is reducible after this single rewrite.
Subcase 3.3: ω1 begins with x3 and ω2 begins with x4x3.
∗←→
There are a number of cases to consider. If ω1 begins with x3x1, since x3x1
∗←→ x1x4, the difference is reducible. If ω1 begins with x3x2, we
x1x2 and x4x3
∗←→ x1x4 to reduce ω1 − ω2. If ω1 begins
can rewrite x3x2
∗←→ x4x2 to immediately reduce ω1 − ω2. It
with x3x3, we may rewrite x3x3
remains to consider the case when ω1 begins with x3x4 and ω2 begins with x4x3.
The corresponding polynomial shows that a homogeneous binomial difference
that starts with these letters cannot occur in IA:
∗←→ x1x3 and x4x3
−2 + 2 · (2) + 22a2 + · · · + 2l−1al−1 = 0
implies
1 + 2a2 + · · · + 2l−2al−1 = 0,
and this equation does not have a solution over `l−1
Subcase 3.4: ω1 begins with x3 and ω2 begins with x4x4.
i=0 Ad.
The corresponding polynomial equation for a difference with these starting
letters is
−2 + 2a1 + · · · + 2l−1al−1 = 0,
and we can factor out the common 2 to obtain
−1 + a1 + · · · + 2l−1al−1 = 0.
In order for this equation to have a solution, a1 ∈ {±1, ±3}, but because the
second letter of ω2 is x4, a1 6= ±1 and a1 6= 3, so a1 = −3. Thus ω1 begins
with x3x2, and because x3x2
∗←→ x4x1, ω1 − ω2 is reducible. ✷
21
References
Adian, S.I., Durnev, V.G., 2000. Decision Problems for Groups and Semi-
groups. Russian Math. Surveys, vol. 55 no. 2, 207-296.
Gilmer, R., 1984. Commutative Semigroup Rings, Chicago Lect. in Math.,
The University of Chicago Press, Chicago.
Green, E.L., 1990. GRB. Available from: http://www.math.vt.edu/people/
green/grb/.
Green, E.L., 1996. Noncommutative Grobner Bases: A Computational and
Theoretical Tool. Lecture Notes, Holiday Mathematics Symposium, New
Mexico State University.
Green, E.L., 2000. Multiplicative Bases, Grobner Bases, and right Grobner
Bases. Journal of Symbolic Computation, Vol. 29, 601 - 623.
Garey, M., Johnson, D., 1979. Computers and Intractability: A Guide to the
Theory of NP-Completeness. Addison-Wesley.
Hinson, E.K., 2010. Admissibility of Weight-Defined Term Orders on Quo-
tients of Free Associative Algebras. Preprint.
Madlener, K., Reinert, B., 1998. Relating Rewriting Techniques on Monoids
and Rings: Congruences on Monoids and Ideals in Monoid Rings. Journal
of Theoretical Computer Science, Vol. 208, 3 - 31.
Nordbeck, P., 2001. Canonical Bases for Algebraic Computations, Ph. D The-
sis, Lund Institute of Technology.
Mora, T., 1994. An Introduction to Commutative and Non-Commutative
Grobner Bases. Theoretical Computer Science, vol. 134, 131 - 173.
Perlo-Freeman, S., Prohle, P., 1997. Scott's Conjecture is True, Position Sensi-
tive Weights. Proceedings of the 8th International Conference on Rewriting
Techniques and Applications Lecture Notes In Computer Science, Vol. 1232,
217 - 227.
Robbiano, L., 1986. On the Theory of Graded Structures. Journal of Symbolic
Computation, Vol. 2, 139 - 170.
Scott, E.A., 1994. Weights for Total Division Orderings on Strings. Theoretical
Computer Science, Vol. 135, pp.345 - 359.
22
|
1605.08109 | 1 | 1605 | 2016-05-26T00:41:40 | Nilpotence dans les alg\`ebres de Malcev | [
"math.RA"
] | The main result is to prove that if a Malcev algebra $A$ is \textit{right nilpotent} of degree $n$, then $A$ is \textit{strongly nilpotent} of degree less or equals to $4n^2-2n+1$. | math.RA | math |
Nilpotence dans les algèbres de Malcev∗
Côme J. A. BÉRɆ, Nakelgbamba B. PILABRɇ
and Moussa OUATTARA§
Laboratoire T. N. AGATA /UFR-SEA
Département de Mathématiques / Université de Ouagadougou
Adresse 03 B. P. 7021 Ouagadougou, Burkina Faso 03
October 26, 2018
Abstract
The main result is to prove that if a Malcev algebra A is right
nilpotent of degree n, then A is strongly nilpotent of degree less or
equals to 4n2 − 2n + 1.
Résumé
Nous prouvons que toute algèbre de Malcev A nilpotente à droite
d'indice n est fortement nilpotente d'un indice inférieur ou égal à
4n2 − 2n + 1.
Keywords. Algèbre de Malcev, nilpotent à droite, nilpotent, fortement
nilpotent, indice.
2010 Mathematics Subject Classification : 17D10
1
Introduction
Lorsque l'algèbre A est de Lie ou alternative, il est facile de montrer que
pour un idéal B, les trois notions de nilpotence, à savoir : B est nilpotent à
droite, B est nilpotent et B est fortement nilpotent sont équivalentes.
Pour certaines algèbres non associatives, ce n'est pas le cas. Une algèbre
de Jordan peut posséder un idéal nilpotent qui n'est pas fortement nilpotent.
∗À la mémoire des Professeurs Akry Koulibaly et Artibano Micali.
†first author email: [email protected]
‡second author email: [email protected]
§third author email: [email protected]
1
De même l'agèbre de Leibniz de dimension deux est nilpotente à gauche et
non nilpotente à droite ([2, Exemple 3. 2] ou [3, Exemple 3. 3]). Dans [4, 5]
M. Gerlein et A. Micali, ont montré l'équivalence de ces trois notions pour
un idéal B d'une algèbre de Malcev A. Cependant, certaines démonstrations
comportaient quelques points d'ombres (voir [8]).
Grâce à l'introduction de nouveaux outils, nous montrons que si A est
nilpotent à droite d'indice n, alors A est fortement nilpotent d'indice inférieur
ou égal à 4n2 − 2n + 1.
Nous montrons aussi l'équivalence de ces trois notions pour un idéal B
d'une algèbre de Malcev A qui est Jk-nil.
Nous commençons par un rappel de quelques définitions et exemple sur les
algèbres de Malcev, puis dans la section 3 nous établissons quelques résultats
sur les produits droits de longueurs n. La section 4 est dédiée à l'étude des
produits droits de poids n. Dans la section 5, nous énoncons un théorème
dans le cas d'un idéal Jk-nil. Puis nous montrons que les trois types de
nilpotence sont équivalents pour une algèbre de Malcev A (cf.
le corollaire
5.1).
2 Préliminaires
Définition 2.1 Par la suite, sauf mention expresse du contraire, K désign-
era un corps commutatif de caractéristique différente de 2 et tout espace
vectoriel sera de dimension finie sur K. Si A est une K-algèbre, non néces-
sairement associative, l'application K-trilinéaire J : A × A × A −→ A définie
par (x, y, z) 7−→ (xy)z + (yz)x + (zx)y est appelée le jacobien de A.
Soient U, V, W trois sous algèbres de A. J (U, V, W ) est le sous espace
vectoriel de A engendré par les éléments de la forme J (u, v, w) où u ∈ U,
v ∈ V et w ∈ W .
On dira que A est une algèbre de Malcev si l'une quelconque des trois
conditions équivalentes suivantes est vérifiée :
1. x2 = 0 pour tout x dans A et J(x, y, xz) = J(x, y, z)x, quel que soient
x, y, z dans A ;
2. x2 = 0 pour tout x dans A et J(x, xy, z) = J(x, y, z)x, quel que soient
x, y, z dans A ;
3. x2 = 0 pour tout x dans A et (xy)(xz) = ((xy)z)x + ((yz)x)x + ((zx)x)y,
quels que soient x, y, z dans A.
Il est clair que ces définitions et équivalences ne dépendent pas de la
caractéristique de K mais si celle-ci est différente de 2, les conditions
2
ci-dessus mentionnées sont encore équivalentes à la suivante :
4. x2 = 0 et (xz)(yt) = ((xy)z)t + ((yz)t)x + ((zt)x)y + ((tx)y)z, quels que
soient x, y, z, t dans A.
Si la caractéristique de K est différente de 2, on a l'égalité (cf. [1]) :
5. J (x, y, z) t = J (t, x, zy)+J (t, y, xz)+J (t, z, yx) , quels que soient x, y, z, t
dans A.
Si A est une K-algèbre, considérons l'algèbre notée A− dont l'espace
vectoriel sous-jacent coïncide avec A et dont la multiplication est définie par
[x, y] = xy − yx pour x, y parcourant A. On vérifie sans peine, que si A
est une algèbre associative, l'algèbre A− est de Lie et si A est une algèbre
alternative, l'algèbre A− est de Malcev. De plus, le jacobien d'une algèbre
de Lie étant nul, toute algèbre de Lie est de Malcev.
Exemple 2.1 Soit A la K-algèbre de dimension 4 dont la table de mul-
tiplication relative à une base {e1, e2, e3, e4} s'écrit e1e2 = e1 = −e2e1,
e1e2 = e1 = −e2e1, e3e1 = e4 = −e1e3, e3e2 = e3 = −e2e3, e2e4 = e4 = −e4e2
et tous les autres produits étant nuls. On vérifie que A est une algèbre de
Malcev (par calcul direct), non de Lie si la caractéristique de K est aussi dif-
férente de 3 (car J(e1, e2, e3) = −3e4). Si K est un corps de caractéristique
3, cette algèbre est de Lie.
On renvoie à [6, 7] pour les renseignements complémentaires concernant les
algèbres de Malcev.
Trois types de nilpotence d'un idéal B d'une algèbre de Malcev A ont été
introduits par A. Micali et Ch. Gerlein dans [4, 5]. Rappelons les :
Définitions 2.2 Soit un idéal B d'une algèbre de Malcev A, on introduit les
notations et terminologies suivantes :
Soit P un produit de m facteurs sm, sm−1, · · · , s1 distincts ou non, asso-
ciés d'une manière quelconque et dont n (n ≤ m) ou plus de ses facteurs
appartiennent à B. On dira que le produit P est de longueur m et de poids
n relativement à B ou plus simplement que P est de longueur m et de poids
n dans B. La longueur m de P sera notée # (P ) et son poids n relativement
à B sera notée #B (P ).
Lorsque P = ((· · · ((smsm−1) sm−2 · · · ) s3) s2) s1 où l'association est faite
à droite systématiquement, on dira que P est un produit droit et on écrira
alors simplement P = smsm−1sm−2 · · · s1, sans parenthèses.
Soient S1, S2, · · · , Sp des produits droits. On peut en faire le produit à droite
N = S1S2 · · · Sp. On dira que N est un produit normal.
3
• Bn le sous-espace vectoriel de l'algèbre A engendré par tous les produits
droits de longueur n et de poids n relativement à B. Par convention,
on posera B0 = A. Et on dit que l'idéal B est nilpotent à droite s'il
existe un entier n ≥ 1 tel que Bn = {0}.
En particulier An est engendré par tous les produits droits de longueur
n et c'est un idéal de l'algèbre A.
• B{n} le sous-espace vectoriel de A engendré par les produits de n élé-
ments de B associés d'une façon quelconque. Et on dit que l'idéal B
est nilpotent s'il existe un entier n ≥ 1 tel que B{n} = {0}.
• Bhni l'ensemble des sommes de produits d'éléments de A avec au moins
n éléments de B. On voit que Bhni est un idéal de A et on a
A ⊇ Bh1i ⊇ Bh2i ⊇ · · · ⊇ Bhni ⊇ · · · et BhiiBhji ⊆ Bhi+ji quels que
soient les entiers i, j ≥ 1. On dit que l'idéal B est fortement nilpotent
s'il existe un entier n ≥ 1 tel que Bhni = {0}.
D'une manière générale si A est une algèbre non associative qui n'est ni
commutative, ni anticommutative, nous ajoutons la définition suivante
:
Lorsque P = s1 (s2 (s3 (· · · sm−2 (sm−1sm)) · · · )) où l'association est faite
à gauche systématiquement, on dira que P est un produit gauche.
• nB le sous-espace vectoriel de l'algèbre A engendré par tous les produits
gauches de longueur n et de poids n relativement à B. Par convention,
on posera 0B = A. Et on dit que l'idéal B est nilpotent à gauche s'il
existe un entier n ≥ 1 tel que nB = {0}.
En particulier si A est commutative ou anticommutative, on a nB coïn-
cide avec Bn.
La preuve qu'une algèbre A qui est nilpotente à droite est une algèbre
fortement nilpotente donnée dans [4] n'est pas assez convainquante (cf. [8]).
C'est pourquoi nous développons d'autres outils qui vont nous permettre de
reécrire une preuve plus rigoureuse.
Définition 2.3 Soit D un sous espace vectoriel d'une algèbre de Malcev A.
Soit k un entier supérieur ou égal à 1. Notons D(A,k) = DAAA . . . A
le sous
espace vectoriel engendré par tous les produits droits de la forme dak · · · a3a2a1
où d ∈ D et ai ∈ A pour tout entier i, tel que 1 ≤ i ≤ k.
{z
k facteurs
}
4
Définition 2.4 Soit B un idéal non nul d'une algèbre de Malcev A. Si
l'idéal J (B, A, A) satisfait pour un entier k supérieur ou égal à 1, la relation
J (B, A, A)(A,k) = {0},on dira que B est Jk-nil dans A.
Définition 2.5 Soient A une algèbre de Malcev et B un idéal de A. Soient
a, b deux éléments de A. On dira que a et b sont égaux modulo B si leur
différence est dans B.
Définition 2.6 Soit n un entier naturel non nul. On pose par définition
I(n) = {1, 2, · · · , n}.
3 Produits droits dans l'algèbre de Malcev A
La preuve du lemme 3.1 se trouve dans [1], mais nous la reprenons.
Lemme 3.1 Soit A une algèbre de Malcev de dimension finie et B un idéal
de A. Posons B0 = A, B1 = B et Bk = Bk + J(B, A, A) pour tout entier
naturel k supérieur ou égal à 2 ; Bk est un idéal de A vérifiant Bk ⊇ Bk+1.
Il est bien connu que J(B, A, A), B0, B1 sont des idéaux. Sup-
Preuve :
posons que pour un entier k ≥ 2, Bk soit un idéal, alors on a Bk.A ⊆
Bk + J(B, A, A). Montrons que Bk+1 est aussi un idéal. En effet on a ;
⊆ Bk+1.A + J(B, A, A)
Bk+1.A =(cid:0)Bk+1 + J(B, A, A)(cid:1) .A
⊆(cid:0)Bk.B(cid:1) .A + J(B, A, A)
⊆ J(Bk, B, A) + (B.A) .Bk +(cid:0)A.Bk(cid:1) .B + J(B, A, A)
⊆ B.Bk +(cid:0)Bk + J(B, A, A)(cid:1) .B + J(B, A, A)
⊆ Bk+1 + J(B, A, A) = Bk+1.
Proposition 3.1 Soient un produit droit P0 = amam−1am−2 · · · a3a2a1 et Q0
un produit quelconque de longueur m′. Posons pour tout entier 0 ≤ i ≤ m−1,
Pi = amam−1 · · · ai+1 =
Qr+1 = ar+1Qr. Alors :
m−i
Yj=1
am−j+1 et pour tout entier 0 ≤ r ≤ m − 2,
(cid:3)
Tm = Q0P0 =
m−1
Xi=1
Qi−1Piai + Qm−1am −
J (Qi−1, Pi, ai) .
m−1
Xi=1
5
Preuve : Avant de faire la preuve fixons quelques définitions de produits :
m désignera la longueur de P .
Posons pour 1 ≤ i ≤ m − 1, Q′
i−1 = Qi, a′
m−i+1 = am−i+2. Nous avons alors,
P ′
i =
m−i
Yj=1
a′
m−j+1 =
m−i
Yj=1
am−j+2 =
m+1−(i+1)
Yj=1
am+1−j+1 = Pi+1.
Il est clair que si m = 2, on a :
T2 = Q0 (a2a1) = Q0a2a1 − Q0a1a2 − J (Q0, a2, a1) , et si m = 3 :
T3 = Q0 (a3a2a1)
= Q0 (a3a2) a1 + a1Q0 (a3a2) − J (Q0, a3a2, a1)
= Q0 (a3a2) a1 + Q1 (a3a2) − J (Q0, a3a2, a1)
= Q0 (a3a2) a1 + Q1a3a2 + a2Q1a3 − J (Q1, a3, a2) − J (Q0, a3a2, a1)
= Q0P1a1 + Q1a3a2 + Q2a3 − J (Q1, a3, a2) − J (Q0, a3a2, a1) .
Posons en hypothèse que pour un produit P de longueur m on a :
Alors on aura pour un produit P = am+1amam−1 · · · a3a2a1 de longueur
(1)
m−1
Tm =
m−1
J (Qi−1, Pi, ai) .
Qi−1Piai + Qm−1am −
Xi=1
Xi=1
am−k+2 = Q0" m
am−k+2a1#
Yk=1
am−k+2, a1!
am−k+2 − J Q0,
Yk=1
Yk=1
am−k+2 − J Q0,
am−k+2, a1!
Yk=1
Yk=1
am−k+2a1 + a1Q0
am−k+2a1 + Q1
m
m
m
m
m+1
m
Yk=1
Yk=1
Yk=1
m
m + 1 :
Tm+1 = Q0
= Q0
= Q0
Il s'ensuit alors que le produit Tm+1 vaut,
Tm+1 = Q0
m+1
Yk=1
am−k+2 = Q0" m
Yk=1
am−k+2a1#
= Q0 (P ′
= Q0P ′
= Q0P1a1 + Q′
0a1)
0a1 + Q1P ′
0P ′
0 − J (Q0, P ′
0, a1)
0 − J (Q0, P1, a1)
(2)
(3)
(4)
(5)
Par application de l'hypothèse de récurrence (1) à Q′
0P ′
0 (car P ′
0 est de
longueur m) :
6
Q′
i−1P ′
i a′
i + Q′
m−1a′
m −
i−1, P ′
i , a′
QiPi+1ai+1 + Qmam+1 −
J (Qi, Pi+1, ai+1)
i(cid:1)
m−1
Xi=1
J(cid:0)Q′
Xi=1
m−1
Q′
0P ′
0 =
=
m−1
m−1
Xi=1
Xi=1
De l'équation (5), il vient que :
Tm+1 = Q0P1a1 + Q′
0P ′
0 − J (Q0, P1, a1)
= Q0P1a1 +"m−1
Xi=1
QiPi+1ai+1 + Qmam+1
−
m−1
Xi=1
J (Qi, Pi+1, ai+1)# − J (Q0, P1, a1)
Xi=1
J (Qi−1, Pi, ai) .
m
(cid:3)
=
m
Xi=1
Qi−1Piai + Qmam+1 −
Remarque 3.1 Sous les hypotèses de la proposition 3.1 et en considérant le
tableau suivant, il vient que :
Pm−1 = am
Pm−2 = Pm−1am−1
Qm−1 = am−1Qm−2 Qm−2 = am−2Qm−3
Pi−1 = Piai
· · ·
· · · Qi−1 = ai−1Qi−2
· · ·
· · ·
Pi+j = Pi+j+1ai+j+1
· · ·
· · · Qi+j = ai+jQi+j−1
P2 = P3a3
· · ·
· · · Q2 = a2Q1 Q1 = a1Q0 Q0 = Q0
P1 = P2a2
P0 = P1a1
L'ensemble Λ = {(ai)1≤i≤m, (bj)1≤j≤m′} est l'ensemble des facteurs dis-
tincts ou non qui donne le produit P . On vérifie facilement qu'il donne
également les produits Qk−1Pkak, Qm−1am où k ∈ I(m − 1). Il vient alors que
pour k ∈ I(m − 1) :
= # (Qk−1) + # (Pk) + 1
# (P )
#B (P ) = #B (Qk−1) + #B (Pk) + #B (ak)
# (P )
#B (P )
= #B (Qm−1) + #B (am) .
= # (Qm−1) + 1
(6)
(7)
(8)
(9)
Lemme 3.2 Tout produit T de longueur m dans une algèbre de Malcev A
de dimension finie est combinaison linéaire de produits droits de longueur m
modulo l'idéal J(A, A, A).
7
Preuve :
Procédons par récurrence sur la longueur m. Lorsque la
longueur m est inférieure ou égale à 3, c'est évident que le résultat est vrai.
Supposons alors la relation vraie pour tout produit de longueur strictement
inférieure à m ≥ 4.
Nous ferons la démonstration en considérant T comme un produit Q0P0
avec P0 un produit droit de longueur n > 1.
Regardons ce qui se passe avec un produit T = Q0P0 de longueur m > n.
Alors d'après la proposition 3.1,
T = Q0P0 =
n−1
Xi=1
Qi−1Piai + Qn−1an −
J (Qi−1, Pi, ai) .
n−1
Xi=1
Les produits suivants Qi−1Pi pour 1 ≤ i ≤ n − 1 et Qn−1 sont des produits de
longueur égale à m − 1 et sont donc des combinaisons linéaires de produits
droits de longueur m − 1 modulo l'idéal J(A, A, A). Ceci montre que T
est combinaison linéaire de produits droits de longueur m modulo l'idéal
J(A, A, A).
(cid:3)
4 Produits droits de poids n dans un idéal B
Lemme 4.1 Tout produit de longueur m et de poids n relativement à B
dans une algèbre de Malcev est combinaison linéaire de produits normaux de
longueur m et de poids n.
Preuve : Le lemme est évident si la longueur m du produit est inférieure
ou égale à trois. Si m = 4, alors il existe quatre éléments a1, a2, a3, a4 dans
A tels que P est égal à l'une des combinaisons de produits droits suivants :
a4a3a2a1 et (a4a3) (a2a1) = a4a3a2a1 + a1a4a3a2 + a2a1a4a3 + a3a2a1a4.
Supposons le lemme vrai pour tout produit de longueur inférieure ou égal
à m. Soit P un produit de longueur m + 1. Alors P s'écrit :
Soit P0a0 avec la longueur de P0 égal à m,
Soit (P4P3) (P2P1) = P4P2P3P1 + P1P4P2P3 + P3P1P4P2 + P2P3P1P4 avec la
longueur de chaque Pi dans I(m − 1).
L'hypothèse de récurrence montre que P0, P4P2P3, P1P4P2, P3P1P4, P2P3P1
sont combinaisons linéaires de produits normaux conservant les longueur et
poids initiaux. Ainsi P l'est aussi. Il est évident que par construction des
termes la longueur et le poids de P sont conservés par chacun des termes. (cid:3)
8
Lemme 4.2 Soient A une algèbre de Malcev et B un idéal non nul de A.
Tout produit T de A de longueur m ≥ 1 et de poids n ≥ 1 relativement à
B est combinaison linéaire de produits droits de longueur m et de poids n
relativement à B, modulo l'idéal J(B, A, A).
Preuve : Le lemme est évident si T est de longueur m ≤ 3.
Posons l'hypothèse suivante :
la relation est vraie pour tout produit T
dont la longueur est strictement inférieure à m.
Soit Q0P0 un produit normal de poids n relativement à B et de longueur
m = p′ + p où Q0 (respectivement P0) est un produit droit de longueur p′
(respectivement p). Alors d'après la proposition 3.1,
Q0P0 =
p−1
Xi=1
Qi−1Piai + Qp−1ap −
J (Qi−1, Pi, ai) .
p−1
Xi=1
Les produits Qi−1Pi , Qp−1 ( où 1 ≤ i ≤ p − 1) sont des produits de
longueur égale à m − 1.
Ils sont donc des combinaisons linéaires de produits droits de longueur m − 1
modulo l'idéal J(B, A, A).
D'où Q0P0 est combinaison linéaire de produits droits de longueur m modulo
l'idéal J(B, A, A).
Grace aux équations 6-9, il est clair que pour i ∈ I(p − 1), le poids
relativement à B du produit normal Qi−1Piai est n. Il en est de même pour
le poids du produit Qp−1ap, relativement à B.
(cid:3)
Lemme 4.3 Soient A une algèbre de Malcev A de dimension finie et B un
idéal non nul de A. Soit P0 = amam−1am−2 · · · a3a2a1 un produit droit de
longueur m et de poids supérieur ou égal à n ≥ 1, relativement à B. Alors
P0 appartient à l'idéal Bn.
Preuve :
Soit σ une injection croissante de I(n) dans I(m) telle que
pour tout j ∈ I(n), aσ(j) ∈ B. Posons pour tout entier 1 ≤ j ≤ n − 1,
0 =amam−1 · · · aσ(n)+1,
j−1aσ(n−j+1),
Q′
Qj =Q′
Q′
Qn =Q′
j =Qjaσ(n−j+1)−1 · · · aσ(n−j)+1,
n−1aσ(1) · · · a1.
Il est clair que, Q1 ∈ B1 ⊆ B1 + J(B, A, A) = B1 et Q′
1 ∈ B ;
Q2 = Q′
Supposons que pour 1 ≤ j < n, Qj ∈ Bj = Bj + J(B, A, A) et montrons que
1aσ(n−j+1) ∈ B.B ⊆ B2 + J(B, A, A) = B2.
9
Qj+1 ∈ Bj+1 = Bj+1 + J(B, A, A).
En exploitant l'hypothèse, on a que
Q′
Qj+1 = Q′
Alors P0 = Qn ∈ Bn.
j = Qjaσ(n−j+2)−1 · · · aσ(n−j+1)+1 appartient à l'idéal Bj. Par suite
jaσ(n−j+1) ∈ Bj.B ⊆ Bj+1 + J(B, A, A) = Bj+1.
(cid:3)
(cid:3)
Lemme 4.4 Soient A une algèbre de Malcev de dimension finie et B un
idéal Jk-nil dans A. Soit ℓ un entier supérieur ou égale à k. Soit un produit
droit P = amam−1am−2 · · · a3a2a1 de longueur m et de poids n supérieur ou
égal à 2ℓ, relativement à B. Alors P ∈ Bℓ
(A,k).
Preuve :
Le produit droit Q = amam−1am−2 · · · ak+1 est de poids
supérieur ou égal à ℓ, en effet posons n′ le poids de Q et n′′ le poids de
ak · · · a3a2a1. On a 0 ≤ n′′ ≤ k et l'égalité P = Qak · · · a3a2a1 nous donne
n′ ≤ n ≤ k + n′. Ainsi n′ ≥ n − k ≥ 2ℓ − k ≥ ℓ. Le lemme 4.3 nous dit que
Q ∈ Bℓ. Ainsi P = Qak · · · a3a2a1 ∈ (Bℓ)(A,k) = (cid:0)Bℓ + J(B, A, A)(cid:1)(A,k) =
(A,k) car B est un idéal Jk-nil dans A.
Bℓ
(cid:3)
Lemme 4.5 Soit B un idéal Jk-nil de l'algèbre de Malcev A. Soit P un
produit de poids t ≥ 4k2 − 2k + 1, relativement à l'idéal B. Alors P est
µjQj) tels que, pour
combinaison linéaire de produits normaux Qj (P = Xj fini
j fixé on a Qj est dans (cid:0)Bk(cid:1)(A,k) ou comporte au moins un facteur dans
(cid:0)Bk(cid:1)(A,k).
normaux de poids supérieur ou égal à t. Soit P = Xj
Preuve : Soient k > 1 et t ≥ 4k2 − 2k + 1. D'après le lemme 4.1, tout
produit P de poids supérieur ou égal à t est combinaison linéaire de produits
µjQj où Qj est un
produit normal de poids supérieur ou égal à t.
Pour j fixé on a Qj = Sj,pSj,p−1 · · · Sj,1 où Sj,i est un produit droit (avec
p le nombre de facteurs Sj,i de Qj).
- Si un des produits droits Sj,i0 possèdent un poids supérieur ou égal à 2k,
alors Sj,i0 appartient à (cid:0)Bk(cid:1)(A,k) d'après le lemme 4.4. Et alors Qj
possède un facteur dans(cid:0)Bk(cid:1)(A,k).
- Sinon, tous les facteurs Sj,i possèdent un poids strictement inférieur à 2k.
Soit q le nombre de facteurs Sj,i ayant un poids supérieur ou égal à 1.
On a alors q (2k − 1) ≥ t = 4k2 − 2k + 1
Par suite q > 2k. Remplaçons chacun des produits Sj,i par sa valeur
10
sj,i = Sj,i ∈ A. Lorsque le poids de Sj,i est supérieur ou égal à 1, on
a sj;i ∈ B. Ainsi Qj se met sous la forme Qj se met sous la forme
d'un produit droit sj,psj,p−1 · · · sj,1 de longueur p et de poids q dans B.
Cela achève la démonstration.
si (cf. les cas cités dans la preuve précédente) :
Comme q ≥ 2k, on a Qj ∈(cid:0)Bk(cid:1)(A,k) d'après le lemme 4.4.
Par un abus de language, on dira que Qj possède un facteur dans(cid:0)Bk(cid:1)(A,k)
- un des produits droits Sj,i0 appartient à(cid:0)Bk(cid:1)(A,k),
- on a Qj ∈(cid:0)Bk(cid:1)(A,k).
(cid:3)
5 Théorème principal
Théorème 5.1 Soient K un corps commutatif de caractéristique différente
de 2, A une K-algèbre de Malcev et B un idéal Jk′-nil de Malcev A. Alors
les conditions suivantes suivantes sont équivalentes :
(i) B est nilpotent à droite ;
(ii) B est nilpotente ;
(iii) B est fortement nilpotente. ;
Preuve : En effet, pour tout entier k > 1, les inclusions d'espaces vectoriels
Bk ⊆ B{k} ⊆ Bhki nous montrent que (iii) ⇒ (ii) ⇒ (i).
Par ailleurs, supposons qu'il existe un entier ℓ′ ≥ 1 tel que Bℓ′ = {0}. Posons
k = max {k′, ℓ′}, alors pour ℓ = 4k2 − 2k + 1, le lemme 4.5 nous dit que
tout produit P de poids ≥ ℓ = 4k2 − 2k + 1, relativement à l'idéal B est
combinaison linéaire de produits normaux dont chacun comporte au moins
un facteur dans (cid:0)Bk(cid:1)(A,k) ⊆(cid:0)Bk′(cid:1)(A,k) = {0}. Il s'ensuit que P = 0 et ainsi
Bhℓi = 0. Prouvant ainsi que (i) ⇒ (iii).
(cid:3)
Corollaire 5.1 Soient K un corps commutatif de caractéristique différente
de 2, A une K-algèbre de Malcev, Les conditions suivantes suivantes sont
équivalentes :
(i) A est nilpotent à droite ;
(ii) A est nilpotente ;
(iii) A est fortement nilpotente.
11
Preuve : Remarquons d'abord que l'idéal J (A, A, A) de A est inclus dans
A3. Supposons que la condition (i) est vérifiée et montrons (iii).
Il existe donc un entier ℓ′ > 1 tel que {0} = Aℓ′, par suite
J(A, A, A)AAA . . . A
⊆ Aℓ′+3 = {0}. Ainsi A est Jℓ′ − nil.
⊆ (A3)AAA . . . A
ℓ′ facteurs
ℓ′ facteurs
En s'appuiant sur la preuve du théorème 5.1, il vient que (i) ⇒ (iii).
Bien entendu, (iii) ⇒ (ii) ⇒ (i) du fait des inclusion de sous espaces vecto-
riels : Bk ⊆ B{k} ⊆ Bhki.
(cid:3)
{z
}
{z
}
References
[1] Côme J. A. BÉRÉ, Superalgèbres de Malcèv, thèse de 3i`eme cycle,
Université de Ougadougou, 1997.
[2] Côme J. A. BÉRÉ, M. Françoise OUEDRAOGO and Nakelgbamba
B. PILABRÉ, On the existence of ad-nilpotent elements, Afr. Mat.
(2015) 26:813-823 DOI 10. 1007/s13370-014-0246-y.
[3] J. A. BÉRÉ Côme, PILABRÉ N. Boukary and KOBMBAYE Aslao,
Lie's theorems on soluble Leibniz algebras. British journal of Mathe-
matics & Computer Science (2014) 4(18): p. 2570-2581.
[4] Ch. GERLEIN et A. MICALI, Sur la nilpotence dans les algèbres de
Malcev, C. R. Ac. Sc. Paris, 286(1978), 1091-1094.
[5] Ch. GERLEIN et A. MICALI, Sur la nilpotence dans les algèbres de
Malcev II (non publié).
[6] A. KOULIBALY, Contribution à la théorie des algèbres de Malcev,
thèse de Doctorat d'Etat, Université de Montpellier, 1984.
[7] A. K. ABD EL MALEK, Sur les algèbres de Malcev, thèse de Doctorat
d'Etat, Université de Montpellier, 1985.
[8] I. SHESTAKOV, Lettre du 19 juin 1979 à Ch. GERLEIN et A. MI-
CALI.
12
|
1508.01850 | 2 | 1508 | 2016-10-03T12:00:39 | Extensions and automorphisms of Lie algebras | [
"math.RA"
] | Let $0 \to A \to L \to B \to 0$ be a short exact sequence of Lie algebras over a field $F$, where $A$ is abelian. We show that the obstruction for a pair of automorphisms in $\Aut(A) \times \Aut(B)$ to be induced by an automorphism in $\Aut(L)$ lies in the Lie algebra cohomology $\Ha^2(B;A)$. As a consequence, we obtain a four term exact sequence relating automorphisms, derivations and cohomology of Lie algebras. We also obtain a more explicit necessary and sufficient condition for a pair of automorphisms in $\Aut\big(L_{n,2}^{(1)}\big) \times \Aut\big(L_{n,2}^{ab}\big)$ to be induced by an automorphism in $\Aut\big(L_{n,2}\big)$, where $L_{n,2}$ is a free nilpotent Lie algebra of rank $n$ and step $2$. | math.RA | math |
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
VALERIY G. BARDAKOV AND MAHENDER SINGH
Abstract. Let 0 → A → L → B → 0 be a short exact sequence of Lie algebras over a
field F , where A is abelian. We show that the obstruction for a pair of automorphisms in
Aut(A)×Aut(B) to be induced by an automorphism in Aut(L) lies in the Lie algebra cohomology
H2(B; A). As a consequence, we obtain a four term exact sequence relating automorphisms,
derivations and cohomology of Lie algebras. We also obtain a more explicit necessary and
sufficient condition for a pair of automorphisms in Aut (cid:0)L(1)
n,2(cid:1) to be induced by an
automorphism in Aut (cid:0)Ln,2(cid:1), where Ln,2 is a free nilpotent Lie algebra of rank n and step 2.
n,2(cid:1) × Aut (cid:0)Lab
1. Introduction
Let A and B be Lie algebras over a field F where A is abelian. We say that A is a left
B-module if there is a F -homomorphism B ⊗ A → A written as b ⊗ a 7→ ba such that
[b1, b2]a = b1(b2a) − b2(b1a) for all b1, b2 ∈ B and a ∈ A.
Let End(A) be the Lie algebra of all F -endomorphisms of A equipped with the Lie bracket
Here f g(a) = f(cid:0)g(a)(cid:1) for a ∈ A. Then a left B-module structure on A is equivalent to existence
of a Lie algebra homomorphism
[f, g] = f g − gf for all f, g ∈ End(A).
B → End(A).
Let A and B be Lie algebras. Then an extension of B by A is a short exact sequence of Lie
algebras
0 → A
i
→ L
p
→ B → 0,
where L is a Lie algebra. Without loss of generality, we may assume that i is the inclusion map
and we omit it from the notation. It follows from the exactness that A is an ideal of L. This
together with the Jacobi identity gives a left L-module structure on A given by
xa = [x, a] for x ∈ L and a ∈ A.
Let s : B → L be a section of p, that is, s is a F -linear map such that ps = 1. If A is abelian,
then this induces a left B-module structure on A given by
ba = (cid:2)s(b), a(cid:3) for b ∈ B and a ∈ A.
In fact, the converse is also true. Note that the above B-module structure on A does not depend
on the choice of the section. We denote the above B-module structure on A by
α : B → End(A).
Date: October 4, 2016.
2010 Mathematics Subject Classification. Primary 17B40, 17B56; Secondary 17B01, 13N15.
Key words and phrases. Automorphism of Lie algebra, Extension of Lie algebras, Free nilpotent Lie algebra,
Cohomology of Lie algebra.
1
2
VALERIY G. BARDAKOV AND MAHENDER SINGH
Let Aut(A), Aut(L) and Aut(B) denote the groups of all Lie algebra automorphisms of A, L
and B, respectively. Let AutA(L) denote the group of all Lie algebra automorphisms of L which
keep A invariant as a set. Note that an automorphism γ ∈ AutA(L) induces automorphisms
γA ∈ Aut(A) and γ ∈ Aut(B) given by γA(a) = γ(a) for all a ∈ A and γ(b) = p(cid:0)γ(s(b))(cid:1) for
all b ∈ B. This gives a group homomorphism
given by
τ : AutA(L) → Aut(A) × Aut(B)
τ (γ) = (γA, γ).
A pair of automorphisms (θ, φ) ∈ Aut(A) × Aut(B) is called inducible if there exists a γ ∈
AutA(L) such that τ (γ) = (θ, φ). Our main aim in this paper is to investigate the following
natural problem.
Problem 1. Let 0 → A → L → B → 0 be an extension of Lie algebras over a field F . Under
what conditions is a pair of automorphisms (θ, φ) ∈ Aut(A) × Aut(B) inducible?
Various aspects of automorphisms of Lie algebras have been investigated extensively in the
literature, see for example [3, 7, 12]. Automorphisms of real Lie algebras of dimension five or
less have been classified in [4]. Further, this work has been extended to automorphisms of six
dimensional real Lie algebras in the recent thesis by Gray [5]. Recall that, the derived series
of a Lie algebra L is given by L(0) = L and L(k) = [L(k−1), L(k−1)] for k ≥ 1. In [1], Bahturin
and Nabiyev investigated the structure of automorphism groups of Lie algebras of the form
L/[R, R], where L a free Lie algebra over a commutative and associative ring k and R is an ideal
of L such that L/R is a free k-module. As an application, they obtained the structure of the
automorphism group of free metabelian Lie algebra L/L(2) of finite rank, and showed that it
has automorphisms which cannot be lifted to automorphisms of L. However, to our knowledge,
almost nothing seems to be known about Problem 1. For extensions of groups, the analogous
problem has been investigated recently in [8, 9, 10, 11], wherein using cohomological methods
the problem for finite groups has been reduced to p-groups.
We would like to remark that our approach in this paper is purely algebraic and we make no
reference to Lie groups. However, it would be interesting to explore connections of our results
to Lie groups, and we leave it to the curious reader.
The paper is organized as follows. In Section 2, we obtain a necessary and sufficient condition
for a pair of automorphisms to be inducible. Our main theorem is the following.
Theorem 1.1. Let 0 → A → L → B → 0 be an abelian extension of Lie algebras over a field F
and (θ, φ) ∈ Aut(A) × Aut(B). Then the pair (θ, φ) is inducible if and only if the following two
conditions hold:
(1) There exists an element λ ∈ Hom(B, A) such that
θ(cid:0)µ(b1, b2)(cid:1) − µ(cid:0)φ(b1), φ(b2)(cid:1) = φ(b1)λ(b2) − φ(b2)λ(b1) − λ(cid:0)[b1, b2](cid:1) for all b1, b2 ∈ B.
(2) αφ(b) = θα(b)θ−1 for all b ∈ B.
Here, Hom(B, A) is the group of all F -linear maps from B to A and µ is the 2-cocycle
corresponding to the extension 0 → A → L → B → 0 (see Section 2 for details). In Section
3, we use this condition to obtain an obstruction to inducibility of automorphisms and derive
an exact sequence relating automorphisms, derivations and cohomology of Lie algebras. More
precisely, we prove the following theorem.
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
3
Theorem 1.2. Let E : 0 → A → L → B → 0 be an abelian extension of Lie algebras over a
field F . Let α : B → End(A) denote the induced B-module structure on A. Then there is an
exact sequence
0 → Z1(B; A)
i
−→ AutA(L)
τ
−→ Cα
ωE−→ H2(B; A).
See Section 3 for unexplained notation. Finally, in Section 4, we discuss Problem 1 for free
nilpotent Lie algebras of rank n and step 2.
2. Condition for extension and lifting of automorphisms
Let 0 → A → L
p
→ B → 0 be an abelian extension of Lie algebras and s : B → L be a section.
For any two elements b1, b2 ∈ B, we have
p(cid:0)s([b1, b2])(cid:1) = [b1, b2] = (cid:2)p(s(b1)), p(s(b2))(cid:3) = p(cid:0)[s(b1), s(b2)](cid:1).
Thus there exists a unique element, say µ(b1, b2) ∈ A, such that
µ(b1, b2) = (cid:2)s(b1), s(b2)(cid:3) − s(cid:0)[b1, b2](cid:1).
Note that µ gives a F -bilinear map from B × B to A such that µ(b, b) = 0 for all b ∈ B. Hence
µ determines an element in Hom(Λ2B, A), which we again denote by µ.
Next, we give a necessary and sufficient condition for a pair of automorphisms to be inducible.
This provides a solution of Problem 1 when A is an abelian Lie algebra.
Proof of Theorem 1.1. Suppose that there exists a γ ∈ AutA(L) such that τ (γ) = (θ, φ). Let
s : B → L be a section. Then any element of L can be written uniquely as a+s(b) for some a ∈ A
and b ∈ B. Now, φ(b) = pγs(b). This implies psφ(b) = pγs(b), and hence γs(b) = sφ(b) + λ(b)
for some λ(b) ∈ A. Since all the maps involved are F -linear, it follows that λ ∈ Hom(B, A).
consider
To derive condition (1), we use the fact that γ is a Lie algebra homomorphism. Let l1 =
a1 + s(b1) and l2 = a2 + s(b2) be two elements of L. Note that γ(cid:0)[l1, l2](cid:1) = (cid:2)γ(l1), γ(l2)(cid:3). First,
γ(cid:0)[l1, l2](cid:1) = γ(cid:0)[a1 + s(b1), a2 + s(b2)](cid:1)
= γ(cid:0)[s(b1), a2] − [s(b2), a1] + [s(b1), s(b2)](cid:1)
= γ(cid:0)[s(b1), a2] − [s(b2), a1] + µ(b1, b2) + s([b1, b2])(cid:1)
= γ(cid:0)[s(b1), a2](cid:1) − γ(cid:0)[s(b2), a1](cid:1) + γ(cid:0)µ(b1, b2)(cid:1) + γ(cid:0)s([b1, b2])(cid:1)
= (cid:2)γ(s(b1)), γ(a2)(cid:3) −(cid:2)γ(s(b2)), γ(a1)(cid:3) + θ(cid:0)µ(b1, b2)(cid:1) + s(cid:0)φ([b1, b2])(cid:1) + λ([b1, b2])
= (cid:2)s(φ(b1)) + λ(b1), θ(a2)(cid:3) −(cid:2)s(φ(b2)) + λ(b2), θ(a1)(cid:3)
= (cid:2)s(φ(b1)), θ(a2)(cid:3) −(cid:2)s(φ(b2)), θ(a1)(cid:3) + θ(cid:0)µ(b1, b2)(cid:1) + s(cid:0)[φ(b1), φ(b2)](cid:1) + λ([b1, b2]).
+θ(cid:0)µ(b1, b2)(cid:1) + s(cid:0)[φ(b1), φ(b2)](cid:1) + λ([b1, b2])
4
VALERIY G. BARDAKOV AND MAHENDER SINGH
Next, consider
(cid:2)γ(l1), γ(l2)(cid:3) = (cid:2)γ(a1 + s(b1)), γ(a2 + s(b2))(cid:3)
= (cid:2)θ(a1) + s(φ(b1)) + λ(b1), θ(a2) + s(φ(b2)) + λ(b2)(cid:3)
= (cid:2)θ(a1), s(φ(b2))(cid:3) +(cid:2)s(φ(b1)), θ(a2)(cid:3) +(cid:2)s(φ(b1)), s(φ(b2))(cid:3)
+(cid:2)s(φ(b1)), λ(b2)(cid:3) +(cid:2)λ(b1), s(φ(b2))(cid:3)
= (cid:2)s(φ(b1)), θ(a2)(cid:3) −(cid:2)s(φ(b2)), θ(a1)(cid:3) +(cid:2)s(φ(b1)), s(φ(b2))(cid:3)
+(cid:2)s(φ(b1)), λ(b2)(cid:3) −(cid:2)s(φ(b2)), λ(b1)(cid:3)
= (cid:2)s(φ(b1)), θ(a2)(cid:3) −(cid:2)s(φ(b2)), θ(a1)(cid:3) + µ(cid:0)φ(b1), φ(b2)(cid:1)
+s(cid:0)[φ(b1), φ(b2)](cid:1) +(cid:2)s(φ(b1)), λ(b2)(cid:3) −(cid:2)s(φ(b2)), λ(b1)(cid:3)
= (cid:2)s(φ(b1)), θ(a2)(cid:3) −(cid:2)s(φ(b2)), θ(a1)(cid:3) + µ(cid:0)φ(b1), φ(b2)(cid:1)
+s(cid:0)[φ(b1), φ(b2)](cid:1) + φ(b1)λ(b2) − φ(b2)λ(b1).
By comparing the LHS and the RHS we obtain condition (1).
To derive condition (2), we use the fact that γ is an isomorphism. Let b ∈ B and a ∈ A.
Then
α(cid:0)φ(b)(cid:1)(a) = (cid:2)s(φ(b)), a(cid:3)
= (cid:2)γ(s(b)) − λ(b), a(cid:3)
= (cid:2)γ(s(b)), a(cid:3)
= (cid:2)γ(s(b)), γ(γ−1(a))(cid:3)
= γ(cid:0)[s(b), γ−1(a)](cid:1)
= θ(cid:0)[s(b), θ−1(a)](cid:1)
= θ(cid:0)α(b)(θ−1(a))(cid:1)
= (cid:0)θα(b)θ−1(cid:1)(a).
Conversely, suppose that conditions (1) and (2) are given. Let l = a + s(b) ∈ L. Then define
γ : L → L by
γ(l) = θ(a) + λ(b) + s(φ(b)).
Since all the maps are F -linear, it follows that γ is F -linear. Clearly, γ(a) = θ(a) for all a ∈ A.
Suppose that γ(a + s(b)) = 0. This implies that s(φ(b)) = 0. Since s and φ are both injective,
it follows that b = 0. This further implies that a = 0, and hence γ is injective. Let a + s(b) ∈ L.
Further, γ(b) = p(cid:0)γ(s(b))(cid:1) = p(cid:0)λ(b) + s(φ(b))(cid:1) = p(cid:0)s(φ(b))(cid:1) = φ(b) for all b ∈ B.
Taking a′ = θ−1(cid:0)a − λ(φ−1(b))(cid:1) and b′ = φ−1(b) yields
γ(cid:0)a′ + s(b′)(cid:1) = θ(a′) + λ(b′) + s(φ(b′))
= θ(cid:0)θ−1(cid:0)a − λ(φ−1(b))(cid:1)(cid:1) + λ(cid:0)φ−1(b)(cid:1) + s(cid:0)φ(φ−1(b))(cid:1)
= a − λ(φ−1(b)) + λ(cid:0)φ−1(b)(cid:1) + s(b)
= a + s(b).
Hence γ is surjective.
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
5
Let l1 = a1 + s(b1) and l2 = a2 + s(b2) be two elements of L.
It remains to prove that
γ([l1, l2]) = [γ(l1), γ(l2)].
(cid:2)γ(l1), γ(l2)(cid:3) = (cid:2)γ(a1 + s(b1)), γ(a2 + s(b2))(cid:3)
+φ(b1)λ(b2) − φ(b2)λ(b1), using (2)
+(cid:2)s(φ(b1)), λ(b2)(cid:3) +(cid:2)λ(b1), s(φ(b2))(cid:3)
= (cid:2)θ(a1) + λ(b1) + s(φ(b1)), θ(a2) + λ(b2) + s(φ(b2))(cid:3)
= (cid:2)θ(a1), s(φ(b2))(cid:3) +(cid:2)s(φ(b1)), θ(a2)(cid:3) +(cid:2)s(φ(b1)), s(φ(b2))(cid:3)
= θ(cid:0)[s(b1), a2](cid:1) − θ(cid:0)[s(b2), a1](cid:1) + µ(cid:0)φ(b1), φ(b2)(cid:1) + s(cid:0)[φ(b1), φ(b2)](cid:1)
= θ(cid:0)[s(b1), a2](cid:1) − θ(cid:0)[s(b2), a1](cid:1) + s(cid:0)[φ(b1), φ(b2)](cid:1)
= θ(cid:0)[s(b1), a2] − [s(b2), a1] + µ(b1, b2)(cid:1) + λ([b1, b2]) + s(cid:0)φ([b1, b2])(cid:1)
= γ(cid:0)[s(b1), a2] − [s(b2), a1] + µ(b1, b2) + s([b1, b2])(cid:1)
= γ(cid:0)[a1, s(b2)] + [s(b1), a2] + [s(b1), s(b2)](cid:1)
= γ(cid:0)[a1 + s(b1), a2 + s(b2)](cid:1)
= γ(cid:0)[l1, l2](cid:1).
+θ(cid:0)µ(b1, b2)(cid:1) + λ([b1, b2]), using (1)
This completes the proof of the theorem. ✷
Remark 2.1. To set notation, we briefly recall the definition of cohomology of Lie algebras.
Let B be a Lie algebra and A be a left B-module. For each 0 ≤ k ≤ dim(B), define Ck(B; A) =
Hom(ΛkB, A) and ∂k : Ck(B; A) → Ck+1(B; A) by
∂k(ν)(b0, . . . , bk) =
(−1)i bi ν(. . . , bi, . . . )
(−1)i+j ν(cid:0)[bi, bj], . . . , bi, . . . , bj, . . .(cid:1)
kX
+ X
i=0
0≤i<j≤k
for all ν ∈ Ck(B; A).
It is straightforward to verify, using the Jacobi Identity and B-action
on A, that ∂k+1∂k = 0. Let Zk(B; A) = ker(∂k) be the group of k-cocycles and Bk(B; A) =
image(∂k−1) be the group of k-coboundaries. Then Hk(B; A) = Zk(B; A)/ Bk(B; A) is the k-
dimensional Lie algebra cohomology of B with values in A.
Let α : B → End(A) be the B-module structure on A and Extα(B, A) denote the set of
equivalence classes of extensions of B by A inducing α. Let E : 0 → A → L → B → 0 be
an extension inducing α and µ ∈ Hom(Λ2B, A) be the F -bilinear map associated to section
s : B → L. Then it is easy to see that µ is a 2-cocycle and 2-cocyles corresponding to different
sections differ by a 2-coboundary. Thus the map [E] 7−→ [µ] gives a bijection
Extα(B, A) ←→ H2(B; A).
See [6, p.238] for a proof and more details.
Now, let (θ, φ) ∈ Aut(A) × Aut(B). Since φ ∈ Aut(B), we replace φ(bi) by bi in condition (1)
and obtain the following condition:
6
VALERIY G. BARDAKOV AND MAHENDER SINGH
θ(cid:0)µ(φ−1(b1), φ−1(b2))(cid:1) − µ(cid:0)b1, b2(cid:1) = b1λ(cid:0)φ−1(b2)(cid:1) − b2λ(cid:0)φ−1(b1)(cid:1) − λ(cid:0)[φ−1(b1), φ−1(b2)](cid:1)
= b1λ′(b2) − b2λ′(b1) − λ′(cid:0)[b1, b2](cid:1)
= ∂1(λ′)(b1, b2),
where λ′ = λφ−1 ∈ C1(B; A) and b1, b2 ∈ B. Thus condition (1) is equivalent to saying that the
LHS of the above equation is a 2-coboundary.
Remark 2.2. A pair (θ, φ) ∈ Aut(A) × Aut(B) is called compatible if
for all b ∈ B. Equivalently, the following diagram commutes.
αφ(b) = θα(b)θ−1
B
α
φ
/ B
α
End(A)
/ End(A)
f 7→θf θ−1
It is easy to see that the set Cα of all compatible pairs is a subgroup of Aut(A) × Aut(B).
Condition (2) of the above theorem shows that every inducible pair is compatible.
Note that αφ also gives a B-module structure on A. Then the compatibility condition is
equivalent to saying that θ : A → A is a B-module homomorphism from the B-module structure
α to the B-module structure αφ on A.
3. An exact sequence for extensions of Lie algebras
Let 0 → A → L
p
→ B → 0 be an abelian extension of Lie algebras over a field F and s : B → L
be a section of p. We show that the obstruction to inducibility of a pair of automorphisms in
Aut(A) × Aut(B) lies in the Lie algebra cohomology H2(B; A).
In fact, we derive an exact
sequence (see (3.1)) corresponding to the above extension, and relating the group of derivations
Z1(B; A), the automorphism group AutA(L) and the cohomology H2(B; A). Here, we consider
the abelian additive group structure on Z1(B; A). The sequence is similar to the one derived for
extensions of groups by Wells in [13] and studied subsequently in [8, 9, 10, 11].
Let AutA,B(L) = (cid:8)γ ∈ Aut(L) τ (γ) = (1A, 1B)(cid:9). Recall that
Z1(B; A) = (cid:8)λ ∈ C1(B; A) ∂1(λ) = 0(cid:9)
= (cid:8)λ ∈ C1(B; A) λ(cid:0)[b0, b1](cid:1) = (cid:2)s(b0), λ(b1)(cid:3) −(cid:2)s(b1), λ(b0)(cid:3) for all b0, b1 ∈ B(cid:9).
Proposition 3.1. Let E : 0 → A → L → B → 0 be an abelian extension of Lie algebras over a
field F . Then Z1(B; A) ∼= AutA,B(L) as groups.
Proof. Define ψ : Z1(B; A) → Aut(L) by ψ(λ) = γλ, where γλ : L → L is given by
γλ(cid:0)a + s(b)(cid:1) = a + λ(b) + s(b) for all a ∈ A and b ∈ B.
Since s and λ are F -linear maps, it follows that γλ is also F -linear. Let l1 = a1 + s(b1) and
l2 = a2 + s(b2). Then
/
/
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
7
γλ(cid:0)[l1, l2](cid:1) = γλ(cid:0)[a1 + s(b1), a2 + s(b2)](cid:1)
= γλ(cid:0)[a1, s(b2)] + [s(b1), a2] + [s(b1), s(b2)](cid:1)
= γλ(cid:0) − [s(b2), a1] + [s(b1), a2] + µ(b1, b2) + s([b1, b2])(cid:1)
= −(cid:2)s(b2), a1(cid:3) +(cid:2)s(b1), a2(cid:3) + µ(b1, b2) + λ(cid:0)[b1, b2](cid:1) + s(cid:0)[b1, b2](cid:1)
= −(cid:2)s(b2), a1(cid:3) +(cid:2)s(b1), a2(cid:3) + µ(b1, b2) +(cid:2)s(b1), λ(b2)(cid:3) −(cid:2)s(b2), λ(b1)(cid:3) + s(cid:0)[b1, b2](cid:1)
= −(cid:2)s(b2), a1(cid:3) +(cid:2)s(b1), a2(cid:3) +(cid:2)s(b1), λ(b2)(cid:3) −(cid:2)s(b2), λ(b1)(cid:3) +(cid:2)s(b1), s(b2)(cid:3)
= (cid:2)γλ(l1), γλ(l2)(cid:3).
Thus γλ is a Lie algebra homomorphism of L. Since s is injective, γλ(a + s(b)) = 0 implies that
a = 0 and b = 0. Finally, if a + s(b) ∈ L, then γλ(a − λ(b) + s(b)) = a + s(b). Hence γλ is a Lie
algebra automorphism of L. Clearly, τ (γλ) = (1A, 1B), and hence ψ(γ) = γλ ∈ AutA,B(L).
Let λ1, λ2 ∈ Z1(B; A). Then
γλ1+λ2(cid:0)a + s(b)(cid:1) = a + λ1(b) + λ2(b) + s(b)
= γλ2(cid:0)a + λ1(b) + s(b)(cid:1)
= γλ2(cid:0)γλ1(a + s(b))(cid:1).
Thus ψ is a group homomorphism. It is easy to see that ψ is injective. Finally, we show that
ψ is surjective onto AutA,B(L). Let γ ∈ AutA,B(L). Since γ = 1B, we have p(cid:0)γ(s(b))(cid:1) = b =
p(cid:0)s(b)(cid:1) of all b ∈ B. This implies that γ(cid:0)s(b)(cid:1) = λ(b) + s(b) for some element λ(b) ∈ A. We
F -linear. Since γ is a Lie algebra homomorphism, we have γ(cid:0)[s(b1), s(b2)](cid:1) = (cid:2)γ(s(b1)), γ(s(b2))(cid:3)
claim that λ ∈ Z1(B; A). Since γ and s are F -linear maps, it follows that λ : B → A is also
for all b1, b2 ∈ B. But
γ(cid:0)[s(b1), s(b2)](cid:1) = γ(cid:0)µ(b1, b2) + s([b1, b2])(cid:1)
= µ(b1, b2) + λ(cid:0)[b1, b2](cid:1) + s(cid:0)[b1, b2](cid:1)
= λ(cid:0)[b1, b2](cid:1) +(cid:2)s(b1), s(b2)(cid:3).
On the other hand, we have
(cid:2)γ(s(b1)), γ(s(b2))(cid:3) = (cid:2)λ(b1) + s(b1), λ(b2) + s(b2)(cid:3)
= (cid:2)λ(b1), s(b2)(cid:3) +(cid:2)s(b1), λ(b2)(cid:3) +(cid:2)s(b1), s(b2)(cid:3).
Equating these two expressions, we obtain λ(cid:0)[b1, b2](cid:1) = (cid:2)λ(b1), s(b2)(cid:3) + (cid:2)s(b1), λ(b2)(cid:3). Hence
λ ∈ Z1(B; A). This completes the proof of the proposition.
(cid:3)
In view of the above proposition, we can view Z1(B; A) as a subgroup of AutA(L). By Remark
2.2, the image of τ lies in Cα. For each (θ, φ) ∈ Cα, we define µθ,φ : B ⊗ B → A by
µθ,φ(b1, b2) = θ(cid:0)µ(φ−1(b1), φ−1(b2))(cid:1) − µ(cid:0)b1, b2(cid:1) for b1, b2 ∈ B.
It can be easily seen that µθ,φ is a 2-cocycle. Since µ is the 2-cocycle corresponding to the
extension E, this defines a map
ωE : Cα → H2(B; A)
8
given by
VALERIY G. BARDAKOV AND MAHENDER SINGH
ωE (θ, φ) = [µθ,φ],
the cohomology class of µθ,φ.
By Theorem 1.1 and Remark 2.1, the pair (θ, φ) ∈ Cα is inducible if and only if µθ,φ is a
2-coboundary. Thus [µθ,φ] ∈ H2(B; A) is an obstruction to inducibility of the pair (θ, φ) ∈ Cα.
With the preceding discussion, we have derived the following exact sequence of groups associated
to an abelian extension of Lie algebras and proving Theorem 1.2.
(3.1)
0 → Z1(B; A)
i−→ AutA(L)
τ−→ Cα
ωE−→ H2(B; A).
The following are some immediate consequences of the above theorem.
Corollary 3.2. Let E : 0 → A → L → B → 0 be a split extension. Then every compatible pair
is inducible.
Proof. By definition, if 0 → A → L → B → 0 is a split extension, then ωE is trivial. The result
now follows from the exactness of sequence (3.1).
(cid:3)
Corollary 3.3. Let 0 → A → L → B → 0 be an extension of finite dimensional Lie algebras
over a field F of characteristic 0. Suppose that B is semisimple. Then every compatible pair is
inducible.
Proof. By Whitehead's Second Lemma [6, Proposition 6.3, p. 249], we have H2(B; A) = 0, and
hence the result follows.
(cid:3)
Recall that a finite dimensional Lie algebra L is said to be perfect if L = [L, L]. Also, the
Schur multiplier of L is defined as M(L) = H2(L, F ), where F is viewed as a trivial L-module.
Corollary 3.4. Let 0 → A → L → B → 0 be a central extension of finite dimensional Lie
algebras over a field F . Suppose that B is perfect and M(B) = 0. Then every pair in Aut(A) ×
Aut(B) is inducible.
Proof. Since 0 → A → L → B → 0 is a central extension, it follows that Cα = Aut(A)× Aut(B).
Further, B being perfect and M(B) = 0 implies that H2(B, A) = 0 by [2, Theorem 6.12]. Hence
it follows from the exact sequence (3.1) that every pair in Aut(A) × Aut(B) is inducible.
(cid:3)
Let E : 0 → A → L → B → 0 be an abelian extension of Lie algebras over a field F and
Then the exact sequence (3.1) becomes the following short exact sequence
(3.2)
0 → Z1(B; A)
i
fCα = (cid:8)(θ, φ) ∈ Cα ωE (θ, φ) = 0(cid:9).
−→ fCα → 0.
−→ AutA(L)
τ
It is natural to investigate splitting of this short exact sequence, and we prove the following.
Theorem 3.5. Let 0 → A → L → B → 0 be a split and abelian extension of Lie algebras over
a field F . Then the associated short exact sequence 0 → Z1(B; A)
also split.
τ−→ fCα → 0 is
i−→ AutA(L)
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
9
Proof. Let A ⋊ B = A ⊕ B as a F -vector space and equipped with the Lie algebra structure
given by
(cid:2)(a1, b1), (a2, b2)(cid:3) = (cid:0)b1a2 − b2a1, [b1, b2](cid:1) for a1, a2 ∈ A and b1, b2 ∈ B.
Then the split extension 0 → A → L → B → 0 is equivalent to the extension
0 → A
a7→(a,0)
−→ A ⋊ B
(a,b)7→b
−→ B → 0,
and hence AutA(L) ∼= AutA(A⋊B). Note that, for a split extension, the corresponding 2-cocycle
is zero, and hence fCα = Cα. Now we define a section σ : fCα → AutA(A ⋊ B) by σ(θ, φ) = γ,
where γ(a, b) = (cid:0)θ(a), φ(b)(cid:1) for a ∈ A and b ∈ B. Clearly, γ is F -linear. Further, for a1, a2 ∈ A
γ(cid:0)[(a1, b1), (a2, b2)](cid:1) = γ(cid:0)b1a2 − b2a1, [b1, b2](cid:1)
and b1, b2 ∈ B, we have
= (cid:0)θ(b1a2 − b2a1), φ([b1, b2])(cid:1)
= (cid:0)θ(b1a2) − θ(b2a1), [φ(b1), φ(b2)](cid:1)
= (cid:0)φ(b1)θ(a2) − φ(b2)θ(a1), [φ(b1), φ(b2)](cid:1) by compatibility of (θ, φ)
= (cid:2)(θ(a1), φ(b1)), (θ(a2), φ(b2))(cid:3)
= (cid:2)γ(a1, b1), γ(a2, b2)(cid:3).
Hence γ ∈ AutA(A ⋊ B). It is clear that σ is a group homomorphism, and hence the sequence
(3.2) splits.
(cid:3)
We conclude by discussing the map ωE in more detail. We show that there is a left action of
Cα on H2(B; A) with respect to which ωE is an inner derivation. Hence ωE = 0 if and only if
this action of Cα on H2(B; A) is trivial.
Let (θ, φ) ∈ Cα and ν ∈ Z2(B; A). For b1, b2 ∈ B, define
(θ,φ)ν(b1, b2) = θ(cid:0)ν(φ−1(b1), φ−1(b2))(cid:1).
Compatibility of (θ, φ) implies that (θ,φ)ν ∈ Z2(B; A). Further, if ν ∈ B2(B; A), then ν = ∂1(λ)
for some λ ∈ C1(B; A). Again compatibility of (θ, φ) implies that (θ,φ)ν = ∂1(θλφ−1), where
θλφ−1 ∈ C1(B; A). Thus [(θ,φ)ν] ∈ H2(B; A). Clearly, this defines a left action of Cα on H2(B; A)
given by
(θ,φ)[ν] = [(θ,φ)ν].
Proposition 3.6. Let E be an extension inducing α. Then ωE is an inner derivation with respect
to the action of Cα on H2(B; A).
Proof. Let E be an extension inducing α and ωE : Cα → H2(B; A) be the corresponding map.
Then for (θ, φ) in Cα, we have
µθ,φ(b1, b2) = θ(cid:0)µ(φ−1(b1), φ−1(b2))(cid:1) − µ(b1, b2)
= (θ,φ)µ(b1, b2) − µ(b1, b2) for all b1, b2 ∈ B.
This implies ωE (θ, φ) = [µθ,φ] = [(θ,φ)µ − µ] =(θ,φ) [µ] − [µ]. Hence ωE is an inner derivation with
respect to the action of Cα on H2(B; A).
(cid:3)
10
VALERIY G. BARDAKOV AND MAHENDER SINGH
Corollary 3.7. Let α : B → End(A) be a B-module structure on A and (θ, φ) ∈ Cα. Then
(θ, φ) is inducible in each extension inducing α if and only if (θ, φ) acts trivially on H2(B; A).
4. Automorphisms of free nilpotent Lie algebras
In this section, we focus on automorphisms of free step 2 nilpotent Lie algebras. Let
Ln,2 = (cid:10)x1, . . . , xn (cid:2)[xi, xj], xk(cid:3) = 0 for all 1 ≤ i, j, k ≤ n(cid:11)
Let L(1)
be the free nilpotent Lie algebra of rank n and step 2. Since every Lie algebra on one generator
is abelian, we assume that n ≥ 2.
n,2 = (cid:10)[xi, xj] 1 ≤ j < i ≤ n(cid:11) be the derived subalgebra of Ln,2. Set Z = {zi,j zi,j =
[xi, xj] for 1 ≤ j < i ≤ n}. Since Ln,2 is step 2 nilpotent, it follows that L(1)
algebra with basis Z and rank n(n−1)
by
n,2 is a free abelian Lie
. If we take the lexicographic order on the basis Z given
2
then Aut(cid:0)L(1)
n,2(cid:1) ∼= GL(cid:0) n(n−1)
2
z2,1 < z3,1 < z3,2 < · · · < zn,n−1,
, F(cid:1). Let θ ∈ Aut(cid:0)L(1)
n,2(cid:1) given by
θ :
...
z2,1 7−→ b2,1;2,1z2,1 + b2,1;3,1z3,1 + · · · + b2,1;n,n−1zn,n−1
...
zi,j 7−→ P1≤l<k≤n bi,j;k,lzk,l
...
zn,n−1 7−→ bn,n−1;2,1z2,1 + bn,n−1;3,1z3,1 + · · · + bn,n−1;n,n−1zn,n−1.
...
...
...
...
...
n,2 = Ln,2/L(1)
n,2(cid:1) ∼= GL(n, F ). Let φ ∈ Aut(cid:0)Lab
n,2 = (cid:10)x1, . . . , xn(cid:11) is also a free
n,2(cid:1) given by
2
Then the matrix [θ] ∈ GL(cid:0) n(n−1)
, F(cid:1). Similarly, Lab
abelian Lie algebra of rank n, and hence Aut(cid:0)Lab
Theorem 4.1. Let (θ, φ) ∈ Aut(cid:0)L(1)
n,2(cid:1) × Aut(cid:0)Lab
x1 7−→ a11x1 + · · · + a1nxn
...
xi 7−→ ai1x1 + · · · + ainxn
...
xn 7−→ an1x1 + · · · + annxn.
φ :
...
...
...
...
...
...
corresponding matrices, then the pair (θ, φ) is inducible if and only if
Then the matrix [φ] ∈ GL(n, F ). We can now formulate the main theorem of this section.
n,2(cid:1). If [θ] = (bi,j;k,l) and [φ] = (aij) are the
(4.1)
aikajl − ailajk = bi,j;k,l for all 1 ≤ j < i ≤ n and 1 ≤ l < k ≤ n.
Proof. For the free nilpotent Lie algebra Ln,2, we have the following central extension
0 → L(1)
n,2 → Ln,2 → Lab
n,2 → 0.
Let s : Lab
n,2 → Ln,2 be the section given by s(xi) = xi on the generators. Then µ(xi, xj) = [xi, xj]
for all 1 ≤ i, j ≤ n. Since Ln,2 is step 2 nilpotent, it follows that RHS of Theorem 1.1(1) is 0
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
11
and α of Theorem 1.1(2) is trivial. Hence the necessary and sufficient condition of Theorem 1.1
becomes
θ(cid:0)µ(xi, xj)(cid:1) = µ(cid:0)φ(xi), φ(xj)(cid:1) for all 1 ≤ i, j ≤ n.
In what follows, we show that this is precisely the condition (4.1).
Recall that, Ln,2 is generated as a Lie algebra by {x1, x2, . . . , xn}. Further, the set
{x1, x2, . . . , xn, z2,1, z3,1, . . . , zn,n−1}
is an ordered basis for Ln,2 as a vector space over F . Thus, if γ ∈ Aut(Ln,2), then γ is given by
x1 7−→ α11x1 + α12x2 + · · · + α1nxn + β1;2,1z2,1 + β1;3,1z3,1 + · · · + β1;n,n−1zn,n−1
...
xn 7−→ αn1x1 + αn2x2 + · · · + αnnxn + βn;2,1z2,1 + βn;3,1z3,1 + · · · + βn;n,n−1zn,n−1
...
...
...
γ :
for some αij, βi;k,l ∈ F . Now, suppose that (θ, φ) is inducible by γ. Then
γ :
Since γ = φ, we obtain
x1 7−→ α11x1 + α12x2 + · · · + α1nxn
...
xn 7−→ αn1x1 + αn2x2 + · · · + αnnxn.
...
...
...
(4.2)
αij = aij for all 1 ≤ i, j ≤ n.
Next, we consider γ
L
as follows
(1)
n,2
γ(zi,j) = γ(cid:0)[xi, xj](cid:1)
= (cid:2)γ(xi), γ(xj )(cid:3)
= (cid:2)αi1x1 + αi2x2 + · · · + αinxn + βi;2,1z2,1 + βi;3,1z3,1 + · · · + βi;n,n−1zn,n−1,
αj1x1 + αj2x2 + · · · + αjnxn + βj;2,1z2,1 + βj;3,1z3,1 + · · · + βj;n,n−1zn,n−1(cid:3)
= (cid:2)αi1x1 + αi2x2 + · · · + αinxn, αj1x1 + αj2x2 + · · · + αjnxn], since Ln,2 is step 2 nilpotent
= X
(cid:0)αikαjl − αilαjk(cid:1)zk,l.
1≤l<k≤n
Since γ
L
(1)
n,2
= θ, we obtain
(4.3)
αikαjl − αilαjk = bi,j;k,l for all 1 ≤ j < i ≤ n and 1 ≤ l < k ≤ n.
Combining equations (4.2) and (4.3), we get (4.1).
(cid:3)
As a consequence, we obtain the following result for n = 2.
matrices. Then the pair (θ, φ) is inducible if and only if r = det[φ].
2,2(cid:1) × Aut(cid:0)Lab
2,2(cid:1) and (cid:0)r, [φ](cid:1) be the corresponding pair of
Corollary 4.2. Let (θ, φ) ∈ Aut(cid:0)L(1)
Proof. In this case, the derived subalgebra L(1)
ated by [x2, x1] and the abelianization Lab
{x1, x2}. The proof now follows from (4.1).
2,2 is a one-dimensional free abelian algebra gener-
2,2 is a two-dimensional free abelian algebra with basis
(cid:3)
We conclude by giving an example to show that the converse of Theorem 3.5 is not true in
general.
12
VALERIY G. BARDAKOV AND MAHENDER SINGH
Example 4.3. Consider the central extension of Lie algebras
0 → L(1)
2,2 → L2,2 → Lab
2,2 → 0.
Since L2,2 is non-abelian, the sequence does not split. We show that the associated short exact
sequence
(4.4)
2,2(cid:1)
2,2; L(1)
splits. We define a section σ : fCα → Aut
0 → Z1(cid:0)Lab
that the sequence (4.4) splits. Define σ(θ, φ) = γ, where
i
−→ Aut
2,2 (cid:0)L2,2(cid:1) τ
(1)
L
−→ fCα → 0
2,2 (cid:0)L2,2(cid:1) which is a group homomorphism, showing
(1)
L
γ :
x1 7−→ s(cid:0)φ(x1)(cid:1)
x2 7−→ s(cid:0)φ(x2)(cid:1)
[x1, x2] 7−→ θ(cid:0)[x1, x2](cid:1).
Then
(cid:2)γ(x1), γ(x2)(cid:3) = (cid:2)s(cid:0)φ(x1)(cid:1), s(cid:0)φ(x2)(cid:1)(cid:3)
= (cid:2)s(a11x1 + a12x2), s(a21x1 + a22x2)(cid:3)
= (cid:2)a11s(x1) + a12s(x2), a21s(x1) + a22s(x2)(cid:3)
= (cid:2)a11x1 + a12x2, a21x1 + a22x2(cid:3)
= θ(cid:0)[x1, x2](cid:1) by Corollary 4.2, since (θ, φ) is inducible
= γ(cid:0)[x1, x2](cid:1).
= (a11a22 − a12a21)[x1, x2]
It follows that γ ∈ Aut
L
2,2 (cid:0)L2,2(cid:1). Since τ (γ) = (θ, φ), σ is a section. It is easy to see that σ is
(1)
a group homomorphism, and hence the sequence (4.4) splits.
Acknowledgement. The authors thank the referee for many useful comments and for the
reference [1]. Support from the DST-RSF project INT/RUS/RSF/2 is gratefully acknowledged.
Bardakov is partially supported by Laboratory of Quantum Topology of Chelyabinsk State
University, and grants RFBR-16-01-00414, RFBR-14-01-00014 and RFBR-15-01-00745. Singh
is also supported by DST INSPIRE Scheme IFA-11MA-01/2011 and DST Fast Track Scheme
SR/FTP/MS-027/2010.
References
[1] Y. Bahturin and S. Nabiyev, Automorphisms and derivations of abelian extensions of some Lie algebras,
Abh. Math. Sem. Univ. Hamburg 62 (1992), 43–57.
[2] P. G. Batten, Multipliers and covers of Lie algebras, PhD Thesis, North Carolina State University, 1993.
[3] A. Borel and G. D. Mostow, On semi-simple automorphisms of Lie algebras, Ann. of Math. 61 (1955),
389–405.
[4] D. J. Fisher, R. J. Gray and P. E. Hydon, Automorphisms of real Lie algebras of dimension five or less, J.
Phys. A 46 (2013), 225204, 18 pp.
[5] R. J. Gray, Automorphisms of Lie Algebras, PhD Thesis, University of Surrey, 2013.
[6] P. J. Hilton and U. Stammbach, A course in homological algebra, Second edition. Graduate Texts in Math-
ematics, 4. Springer-Verlag, New York, 1997.
EXTENSIONS AND AUTOMORPHISMS OF LIE ALGEBRAS
13
[7] N. Jacobson, A note on automorphisms of Lie algebras, Pacific J. Math. 12 (1962), 303–315.
[8] P. Jin, Automorphisms of groups, J. Algebra 312 (2007), 562–569.
[9] P. Jin and H. Liu, The Wells exact sequence for the automorphism group of a group extension, J. Algebra
324 (2010), 1219–1228.
[10] I. B. S. Passi, M. Singh and M. K. Yadav, Automorphisms of abelian group extensions, J. Algebra 324
(2010), 820–830.
[11] D. J. S. Robinson, Automorphisms of group extensions, Note Mat. 33 (2013), 121–129.
[12] R. Steinberg, Automorphisms of classical Lie algebras, Pacific J. Math. 11(1961), 1119–1129.
[13] C. Wells, Automorphisms of group extensions, Trans. Amer. Math. Soc. 155 (1971), 189–194.
Sobolev Institute of Mathematics, Novosibirsk 630090, Russia
Novosibirsk State University, Novosibirsk 630090, Russia
Laboratory of Quantum Topology, Chelyabinsk State University, Brat'ev Kashirinykh street
129, Chelyabinsk 454001, Russia.
Novosibirsk State Agrarian University, Dobrolyubova street, 160, Novosibirsk, 630039, Russia.
E-mail address: [email protected]
Indian Institute of Science Education and Research (IISER) Mohali, Sector 81, S. A. S. Nagar,
P. O. Manauli, Punjab 140306, India.
E-mail address: [email protected]
|
1204.4123 | 3 | 1204 | 2013-10-17T15:08:51 | On a spectral sequence for the cohomology of a nilpotent Lie algebra | [
"math.RA"
] | Given a nilpotent Lie algebra $\mathfrak n$ we construct a spectral sequence which is derived from a filtration of its Chevalley-Eilenberg differential complex and converges to the Lie algebra cohomology of $\mathfrak n$. The limit of this spectral sequence gives a grading for the Lie algebra cohomology, except for the cohomology groups of degree 0, 1, $\dim \mathfrak n-1$ and $\dim\mathfrak n$ as we shall prove. We describe the spectral sequence associated to a nilpotent Lie algebra which is a direct sum of two ideals, one of them of dimension one, in terms of the spectral sequence of the co-dimension one ideal. Also, we compute the spectral sequence corresponding to each real nilpotent Lie algebra of dimension less than or equal to six. | math.RA | math |
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A
NILPOTENT LIE ALGEBRA
VIVIANA DEL BARCO
Abstract. Given a nilpotent Lie algebra n we construct a spectral sequence which is
derived from a filtration of its Chevalley-Eilenberg differential complex and converges to
the Lie algebra cohomology of n. The limit of this spectral sequence gives a grading for
the Lie algebra cohomology, except for the cohomology groups of degree 0, 1, dim n − 1
and dim n as we shall prove. We describe the spectral sequence associated to a nilpotent
Lie algebra which is a direct sum of two ideals, one of them of dimension one, in terms
of the spectral sequence of the co-dimension one ideal. Also, we compute the spectral
sequence corresponding to each real nilpotent Lie algebra of dimension less than or equal
to six.
MSC (2010): 17B30, 17B56, 55T05.
Keywords: Nilpotent Lie algebras, Lie algebra cohomology, spectral sequences.
1. Introduction
In the present work we associate a cohomological spectral sequence to each nilpotent
Lie algebra n. This association is achieved by considering a natural filtration of the
Chevalley-Eilenberg differential complex of n, namely that one defined by the annihilator
spaces of the central descending series. The spectral sequence {Ep,q
r } induced by this
filtration converges to the Lie algebra cohomology with trivial coefficients H ∗(n), yielding
the formula
H i(n) ≃ Mp+q=i
Ep,q
∞ .
(∗)
Thus one obtains a refinement of the Lie algebra cohomology of n by using a natural
filtration on the space of 1-forms. The dimensions of the limit terms of the spectral
sequence are invariants of n from which its Betti numbers can be recovered. In this work
we investigate this spectral sequence in the aim to contribute on a further study of the
cohomology of nilpotent Lie algebras. Also, an application we pursue is to establish how
this spectral sequence constraints the existence of certain geometric structures on n.
It is known that, contrary to the semisimple case, the explicit calculation of the coho-
mology of a nilpotent Lie algebra is a hard task in general. Nevertheless a few particular
cases are well understood, such as for Heisenberg Lie algebras [18], free 2-step nilpotent Lie
algebras [7, 19] and small dimensional Lie algebras, among others. Also, the cohomology
of the nilradicals of parabolic subalgebras in semisimple Lie algebras has been described
by Kostant [9].
Recall that nilpotent Lie algebras are classified up to dimension seven (see [3, 6, 11] and
references therein). An outstanding conjecture involving the cohomology of nilpotent Lie
algebras is the Toral Rank Conjecture [2, 5, 15, 16, 20], which is still open in the general
case. It was posed by S. Halperin in 1968 [8] and states that the total dimension of the
Lie algebra cohomology group of n is greater than the total dimension of the Lie algebra
Partially supported by CONICET, SCyT-UNR and Secyt-UNC.
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
2
cohomology group of the center of n. Introducing new tools to study the cohomology of
nilpotent Lie algebras would feasibly lead to future advances on these problems.
The construction of the spectral sequence under consideration was proposed by S. Sala-
mon in his private correspondence with I. Dotti. In this paper we introduce its formal
definition and we investigate properties of this spectral sequence in relation to the structure
of the nilpotent Lie algebra.
We include here the computation of the spectral sequence of each real nilpotent Lie
algebra of dimension ≤ 6. Despite the fact that their real cohomology is well known,
we find this a necessary step to get conclusions on the properties of n that this spectral
sequence might reveal. One sees that there are nilpotent Lie algebras having the same
Betti numbers but different gradings in (*), showing that the grading given by the spectral
sequence is, in fact, more refined that the usual cohomology degree. Also we were able
to observe that non-symplectic Lie algebras have zero term E0,2
∞ . This yields to a general
obstruction to the existence of symplectic structures for nilmanifolds, a result which is
fully developed by the author in [4].
Throughout this work elementary knowledge of spectral sequences will be assumed.
Acknowledgments. This work is part of my Ph.D. Thesis written at FCEIA, Univer-
sidad Nacional de Rosario, Argentina and under the supervision of Isabel Dotti. I am very
grateful to Isabel Dotti for all her dedication to the guidance work. I am also thankful to
Simon Salamon for suggestions which helped me improve the results presented here.
2. Definition of the spectral sequence
Let n denote a real Lie algebra. The central descending series of n, {ni} is given by
n0 = n,
ni = [n, ni−1],
i ≥ 1.
A Lie algebra n is k-step nilpotent if nk = 0 and nk−1 6= 0; this number k is called the
nilpotency index of n. For example, abelian Lie algebras are 1-step nilpotent. For a k-step
nilpotent Lie algebra it holds nk−1 ⊆ z(n), where z(n) denotes the center of n.
The Chevalley-Eilenberg complex of n is
d=0
−→ n∗ = C 1(n)
d
−→ C 2(n)
R
(1)
where C p(n) denotes the vector space of skew-symmetric p-linear forms on n identified
with the exterior product Λpn∗, and the differential d : Λpn∗ −→ Λp+1n∗ is defined by:
−→ . . .
−→ . . .
d
d
−→ C p(n)
d
dpx (u1, . . . , up+1) = X1≤i<j≤p+1
(−1)i+j−1x([ui, uj], u1, . . . , ui, . . . , uj, . . . , up+1).
Notice that the differential d : n −→ Λ2n∗ coincides with the dual mapping of the Lie
bracket [ , ] : Λ2n −→ n and for x, y ∈ Λ∗(n∗) we have
d(x ∧ y) = dx ∧ y + (−1)deg xx ∧ dy.
The cohomology of (C ∗(n), d) is called the Lie algebra cohomology (with trivial coeffi-
cients) of n and it is denoted by H ∗(n).
The following subspaces of the dual of a nilpotent Lie algebra were considered in [17]
and they lead us to a filtration of the complex in (1). Set
(2)
Notice that V1 is the space of closed 1-forms and that V0 ⊆ V1 ⊆ · · · ⊆ Vi ⊆ · · · ⊆ n∗.
These spaces are dual to those in the central descending series [17]. That is, for each i ≥ 0,
Vi = {x ∈ n∗ : dx ∈ Λ2Vi−1}
V0 = 0
i ≥ 1.
and
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
3
Vi = {x ∈ n∗ : x(u) = 0, ∀ u ∈ ni} = (ni)◦. In particular, n is k-step nilpotent if and only
if Vk = n∗ and Vk−1 6= n∗; we define Vi = 0 for i < 0.
Let n be a k-step nilpotent Lie algebra of dimension m. The sequence in (2) defines a
filtration in the space of skew symmetric q-forms, Λqn∗, for all q = 0, . . . , m,
0 = ΛqV0 ( ΛqV1 ( . . . ( ΛqVk−1 ( ΛqVk = Λqn∗.
(3)
Also d(ΛqVi) ⊂ Λq+1Vi so each of these subspaces is invariant under the Lie algebra
differential. Thus for each p
(4)
F pC ∗ :
0 −→ R −→ Vk−p −→ Λ2Vk−p −→ · · · −→ ΛmVk−p −→ 0
is a subcomplex of the Chevalley-Eilenberg complex. Observe that F pC = 0 (the zero
complex) if p > k and F pC = C ∗ if p < 0, hence
. . . ⊆ 0 ⊆ F k−1C ∗ ⊆ . . . ⊆ F p+1C ∗ ⊆ F pC ∗ ⊆ . . . ⊆ F 1C ∗ ⊆ C ∗ ⊆ . . .
.
Therefore {F pC ∗} constitutes a bounded filtration of the complex (1) which gives rise to
a spectral sequence {Ep,q
r } where for p, q ∈ Z it holds
(5)
Ep,q
0 =
F pC p+q
F p+1C p+q ,
and dp,q
0
further details on spectral sequences we refer the reader to [21, Sect. 5.4].
is the quotient map induced by the Lie algebra differential. For
0
: Ep,q
0 −→ Ep,q+1
For any nilpotent Lie algebra n the filtration and the spectral sequence that arise in this
natural way are bounded because the dimension of n and its nilpotency index are finite.
It is well known that given a cohomological complex and a filtration of it, the associated
spectral sequence converges to the cohomology of the original complex. In this case the
spectral sequence in (5) converges to the Lie algebra cohomology of n. This fact is denoted
by
Ep,q
r ⇒ H ∗(n).
An isomorphism between two nilpotent Lie algebras preserves the filtrations and induces
a spectral sequence isomorphism. This implies that there is a spectral sequence associated
to each isomorphism class of nilpotent Lie algebras.
The zero page of the spectral sequence defined here satisfies
(6)
Ep,q
0 =
Λp+qVk−p
Λp+qVk−(p+1)
if
0 ≤ p ≤ k − 1,
while Ep,q
in the r-th page are E0,n
r
0 = 0 if p < 0 or p ≥ k. Thus the (possibly) non-zero elements of total degree n
, E1,n−1
r
, . . . , Ek−1,n−k+1
r
where k is the nilpotency index of n.
Each term of the spectral sequence can be computed directly by calculating the Lie
algebra differential d paying attention to the filtration. Explicitly,
(7)
Ep,q
r ≃
Ap,q
r
d(Ap−r+1, q+r−2
r−1
) + Ap+1, q−1
r−1
where Ap,q
sequence are
r = {x ∈ Λp+qVk−p : dx ∈ Λp+q+1Vk−p−r}. The limit terms of the spectral
(8)
Ep,q
∞ ≃
d({x ∈ Λp+q−1n∗ : dx ∈ Λp+qVk−p}) + {x ∈ Λp+qVk−p−1 : dx = 0}
{x ∈ Λp+qVk−p : dx = 0}
.
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
4
In the next example we deal with a family of filiform Lie algebras. For each Lie algebra
of the family we compute the limit terms of the spectral sequence associated to it with total
degree 0,1 and 2; for total degree 2 this is done using previous results on their cohomology.
Example 2.1. Let m ∈ N and denote m0(m) the real Lie algebra having a basis of 1-forms
{e1, . . . , em} with differential d : n∗ −→ Λ2n∗ determined by the following formula
(9)
dei =(cid:26)
0
e1 ∧ ei−1
if i = 1, 2,
if i = 3, . . . , m.
The Lie algebra m0(m) is filiform, that is, its nilpotency index k is m − 1, and its coho-
mology is studied by D. Millionschikov in [12].
Let {Ep,q
r } be the spectral sequence associated to m0(m). The filtration of the dual of
m0(m) is Vi = span{e1, . . . , ei+1} for i = 1, . . . , m − 1. Using formula in (8) it is easy to
see that Ep,−p
∞ ≃ R if p = k − 1 = m − 2 and Ep,−p
∞ = 0 otherwise. Also, there is only one
non-zero term of total degree 1 in the limit of {Ep,q
r }. Indeed,
Ep,1−p
∞
≃
{x ∈ Vk−p : dx = 0}
{x ∈ Vk−p−1 : dx = 0}
,
which is zero if k − p ≥ 2 while Ek−1,2−k
≃ V1. For any nilpotent Lie algebra the limit of
the spectral sequence associated to it has only one non-zero element of total degree 0 and
only one of total degree 1 as we shall see in the next section.
∞
According to [12, Lemma 5.2], the second cohomology group H 2(m0(m)) admits a basis
composed by the cohomology class of e1 ∧ em and those of ωs, where
ωs =
1
2
2s−1Xi=2
(−1)i ei ∧ e2s+1−i, s = 2, . . . , [(m + 1)/2];
the brackets denote the integer part. Each ωs is closed and non-exact.
This basis of H 2(m0(m)) can be used to compute the degree 2 terms of the limit of
spectral sequence {Ep,q
r }; namely, E0,2
∞ , E1,1
∞ , . . . , Em−2,4−m
∞
. Recall from Eq. (8)
Ep,2−p
∞
≃
{x ∈ Λ2Vm−1−p : dx = 0}
d(Vm−p) + {x ∈ Λ2Vm−p−2 : dx = 0}
for p = 0, . . . , m − 2.
Since Vi = span{e1, . . . , ei+1} for i = 1, . . . , m − 1, it holds
e1 ∧ em ∈ Λ2Vm−1, while ωs ∈ Λ2V2s−2,
s = 2, . . . , [(m + 1)/2];
also ωs /∈ Λ2V2s−3. Hence e1 ∧ em defines a non-trivial element in E0,2
defines a non-zero element in Ep,2−p
situations depending on whether m is even or odd. In fact
∞ . In addition, ωs
if and only if p = m − 2s − 1. This gives different
∞
(10)
dim E0,2
∞ =(cid:26) 1 if m is even,
2 if m is odd,
and dim Ep,2−p
∞
=(cid:26) 0 if p ≡ m (mod 2)
1 if p 6≡ m (mod 2)
for p = 1, . . . , m − 2.
3. General properties
In the previous section, to each nilpotent Lie algebra n was associated a natural filtration
of the Chevalley-Eilenberg differential complex. This filtration gives rise to a spectral
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
5
sequence that converges to the Lie algebra cohomology. Thus each cohomology group
H i(n) can be written as a direct sum of the limit terms of the spectral sequence. Namely
(11)
Ep,q
∞
for all i = 0, . . . , dim n.
H i(n) ≃ Mp+q=i
Thus the spaces Ep,q
∞ refine the Lie algebra cohomology of n and, in view of Nomizu's
theorem [14], the de Rham cohomology of a nilmanifold Γ\N where n is the Lie algebra
of N . Because of this essential property, we might call them the intermediate cohomology
groups of n (or N ).
Computing all terms of the spectral sequence associated to a nilpotent Lie algebra n
using the definition or Eqns. (7), (8) might take a large amount of work. For this reason
we introduce properties that intend to facilitate this work. First, we review results on
closed and exact m − 1-forms of an m dimensional nilpotent Lie algebra.
Lemma 3.1. Let n be an m dimensional (r + 1)-step nilpotent Lie algebra with differential
d : n∗ −→ Λ2n∗. Let n0 be the dimension of V1 = ker d and denote β1, β2, . . . , βn0 a basis
of V1. Then every σ ∈ Λm−1n∗ is closed. Moreover σ is exact if and only if it is divisible
by β1 ∧ β2 ∧ · · · ∧ βn0.
Remark. This lemma joins a result due to Benson and Gordon in [1] and its converse
proved by Yamada in [22].
In both works, the authors add the hypothesis that n is a
symplectic Lie algebra. Nevertheless we noticed that this is valid for any nilpotent Lie
algebra. We include the proof here to make this fact explicit and for completeness of the
exposition.
Proof of Lemma 3.1. Let ai denote a vector space complementary to ni+1 in ni where ni
is the i-th term of the central descending series of n. For i = 0, 1, . . . , r
define ni = dim ai. Denote
ni = ni+1 + ai;
Λi0,i1,...,ir = Λi0(a0)∗ ∧ Λi1(a1)∗ ∧ · · · ∧ Λir (ar)∗ ⊆ Λi0+i1+...+ir n∗.
Then for s = 0, . . . , m
Λsn∗ = Xi0+i1+...+ir=s
Λi0,i1,...,ir .
Notice that it is possible to choose a0 such that (a0)∗ = V1. Let β1, β2, . . . , βn0 be a basis
of (a0)∗.
Assume η ∈ Λi0,...,ir then each term of dη belongs to a subspace Λt0,t1,...,tr such that
there exists an index j ≥ 1 with tj = ij − 1 and t0 + t1 + . . . + tj−1 = i0 + i1 + . . . + ij−1 + 2.
Suppose in particular that η ∈ Λm−1n∗, this implies i0 + i1 + . . . + ir = m − 1. Then for
any j ≥ 1 we have i0 + i1 + · · · + ij−1 ≥ n0 + n1 + · · · + nj−1 − 1. In fact if for some j holds
l=1 nl − 1 = 2m − 1 leading
to a contradiction. Fixed the term of dη in Λt0,t1,...,tr , let j be the index specified in the
previous paragraph. It verifies t0+t1+. . .+tj−1 = i0+i1+. . .+ij−1+2 > n0+n1+· · ·+nj−1
implying dη = 0. Therefore any m − 1-form is closed.
i0 + i1 + · · · + ij−1 < n0 + n1 + · · · + nj−1 − 1, thenPr
l=1 il <Pr
Let σ = dα be an m − 1-exact form. Clearly each component η of α belongs to some
Λi0,...,ir with i0 + . . . + ir = m − 2. As before, the term of dη belonging to Λt0,t1,...,tr satisfies
t0 + t1 + · · · + tj = n0 + n1 + · · · + nj for all j ≥ 1. In particular, t0 = n0 which implies
that σ is divisible by β0 ∧ · · · ∧ βn0.
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
6
To prove that any m − 1-form divisible by β0 ∧ · · · ∧ βn0 is exact, a dimensional ar-
gument is used. Denote Bm−1 the set of exact m − 1-forms, then Bm−1 is a subset of
Pn0+i1+···+ir=m−1 Λn0,i1,...,ir . From Poincar´e duality dim H m−1(n) = dim H 1(n) = n0.
Bm−1 =Pn0+i1+···+ir=m−1 Λn0,i1,...,ir .
Moreover, the result above states dim H m−1(n) = dim Λm−1n∗ − dim Bm−1. Combining
to dim Bm−1 = n1 + · · · + nr and therefore
these two formulas one leads
(cid:4)
The next result states that the limit terms of the spectral sequence of total degree
0, 1, dim n − 1 and dim n are mostly zero.
Theorem 3.2. Let n be an m dimensional k-step nilpotent Lie algebra with differential
d : n∗ −→ Λ2n∗ and let {Ep,q
r } be the spectral sequence associated to n. Then
∞
(1) Ek−1,1−k
(2) Ek−1,2−k
(3) E0,m−1
(4) E0,m
∞
∞
= H 0(n) = R and hence Ep,−p
= H 1(n) = ker d and hence Ep,1−p
∞ (n) = 0 for all p = 0, . . . , k − 2.
∞ (n) = 0 for all p = 0, . . . , k − 2.
= H m−1(n) and hence Ep,m−1−p
∞
= 0 for all p = 1, . . . , k − 1.
∞ ≃ H m(n) ≃ R and hence Ep,m−p
∞
= 0 for all p =, 1 . . . , k − 1.
Proof. The first two assertions follow straight from the definition of the spectral sequence
of n. Moreover, it is well known that the top cohomology group H m(n) of a nilpotent Lie
algebra n is spanned by a class of the form e1 ∧ . . . ∧ em if {e1, . . . , em} is a basis of n∗ [10].
This implies E0,m
∞ ≃ H m(n) ≃ R which combined with Eq. (11) proves (4).
To prove (3), we make use of the lemma above and Eq. (8) which together imply
E0,m−1
∞
=
Λm−1n∗
Bm−1 + Λm−1Vk−1
.
It is sufficient to show that Λm−1Vk−1 ⊆ Bm−1 since it implies
E0,m−1
∞
=
Λm−1n∗
Bm−1 = H m−1(n).
Notice that if dim Vk−1 < m − 1 then Λm−1Vk−1 = 0 and it is clearly contained in
Bm−1. When dim Vk−1 = m − 1, dim ak−1 = 1 (in the notation of Lemma 3.1) and the
subspace Λm−1Vk−1 coincides with Λn0,n1,...,nk−2,0 also a subset of Bm−1 because of the
lemma above. Thus E0,m−1
= H m−1(n) in any case. Using Eq. (11) one concludes that
Ep,m−1−p
(cid:4)
= 0 for all 1 ≤ p ≤ k − 1.
∞
∞
In Lie algebra cohomology the well known Kunneth formula relates the Lie algebra
cohomology of a Lie algebra which is a direct sum of ideals, with the cohomology of its
summands.
When the nilpotent Lie algebra n can be decomposed as a direct sum of two ideals being
one of them of dimension one, a similar formula relates the spectral sequence of n with
those of its summands.
Theorem 3.3. Let n be a k-step nilpotent Lie algebra which can be decomposed as a direct
sum of ideals n = R ⊕ h. Denote by {Ep,q
r } the spectral sequences associated to
n and to h respectively. Then it holds
r } and { Ep,q
r = 0 if p + q < 0,
≃ R and Ep,−p
r
≃ Ek−1,2−k
⊕ R,
r
r
(1) Ep,q
(2) Ek−1,1−k
(3) Ek−1,2−k
(4) Ep,1−p
(5) Ep,q
r
≃ Ep,1−p
if 0 ≤ p ≤ k − 2 and Ep,1−p
r
= 0 if p < 0 or p ≥ k.
r
r
r ≃ Ep,q
r ⊕ Ep,q−1
r
if p + q ≥ 2.
= 0 for all p 6= k − 1.
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
7
Proof. Notice that under the hypothesis of the theorem h is k-step nilpotent. We will only
show some equalities that lead to the complete proof.
The first two assertions follow from the definition of the spectral sequence and Eq. (6).
Suppose n = Rx ⊕ h and denote by x∗ the element in n∗ such that x∗(x) = 1, x∗(h) = 0
and identify h∗ with a subset of n∗. Clearly dx∗ = 0 and h∗ is invariant by d. Also the
restriction of d to h∗ is the differential of the Lie algebra h.
The subsets V0 ⊆ V1 ⊆ · · · ⊆ Vk = n∗ and eV0 ⊆ eV1 ⊆ · · · ⊆ eVk = h∗ which filter n∗ and
h∗ respectively satisfy:
(12)
Eq. (6) gives the initial term of {Ep,q
eV0 = V0 = 0,
Vi = eVi ⊕ Rx∗, i = 1, . . . , k.
r }: for total degree one, Ep,1−p
0
= Vk−p/Vk−p−1
which by (12) is
0
Ek−1,2−k
Ep,1−p
0
= eV1 ⊕ Rx∗ and
≃ eVk−p/eVk−p−1 = Ep,1−p
0
if p 6= k − 1.
If p + q ≥ 2 the equality Λp+q Vk−p = Λp+q eVk−p ⊕ (Rx∗ ∧ Λp+q−1 eVk−p) used in Eq. (6)
yields to
Ep,q
0
=
Λp,qeVk−p ⊕ (Rx∗ ∧ Λp+q−1eVk−p)
Λp,qeVk−p−1 ⊕ (Rx∗ ∧ Λp+q−1eVk−p−1)
≃ Ep,q
0 ⊕ Ep,q−1
0
.
≃
Λp,qeVk−p
Λp,qeVk−p−1
Λp+q−1eVk−p
Λp+q−1eVk−p−1
⊕
.
When r ≥ 1 the key is to use Eq. (7). Suppose p = k − 1, then
(13)
Ek−1,q
r
=
{y ∈ Λk+q−1V1 : dy = 0}
d({y ∈ Λk+q−2Vr : dy ∈ Λk+q−1V1})
If q = 2 − k then Ek−1,2−k
k + q − 1 ≥ 2)
r
= eV1 ⊕ Rx∗ ≃ Ek−1,2−k
r
⊕ R; if q ≥ 3 − k (or equivalently
Λk+q−1V1 = Λk+q−1eV1 ⊕ (Rx∗ ∧ Λk+q−2eV1),
therefore ω ∈ Λk+q−1V1 if and only if ω = ω1 + x∗ ∧ ω2 with ω1 ∈ Λk+q−1eV1 and ω2 ∈
Λk+q−2eV1. Moreover ω is closed if and only if dω1 + x∗ ∧ dω2 = 0, condition only satisfied
in the case that ω1 and ω2 are simultaneously closed. So it is possible to rewrite the
numerator in (13) as
{x ∈ Λk+q−1eV1 : dy = 0} ⊕ (Rx∗ ∧ {x ∈ Λk+q−2eV1 : dy = 0}).
To describe the denominator of the same equation it is necessary to consider two cases:
k + q − 2 = 1 (or equivalently q = 3 − k) and k + q − 2 ≥ 2. In the first situation
d({y ∈ Λk+q−2Vr : dy ∈ Λk+q−1V1}) = d({y ∈ Vr : dy ∈ Λ2V1}) = d(V2) = d(eV2),
reaching
Ek−1,3−k
r
=
{x ∈ Λ2eV1 : dy = 0} ⊕ (Rx∗ ∧ {x ∈ eV1 : dy = 0})
≃ Ek−1,3−k
r
) ≃ Ek−1,3−k
r
⊕ Ek−1,2−k
r
.
d(eV2)
⊕ (Rx∗ ∧ eV1
{z
}
≃ eV1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
8
In the case k + q − 2 ≥ 2 (or equivalently q ≥ 4 − k) one has
Λk+q−2Vr = Λk+q−2eVr ⊕ (Rx∗ ∧ Λk+q−3eVr),
Λk+q−1V1 = Λk+q−1eV1 ⊕ (Rx∗ ∧ Λk+q−2eV1).
Thus any element ω of Λk+q−2Vr can be written as ω = ω1 + x∗ ∧ ω2 where
In addition dω = dω1 + x∗ ∧ dω2 is an element
ω1 ∈ Λk+q−2eVr and ω2 ∈ Λk+q−3eVr.
into Λk+q−1V1 if and only if dω1 ∈ Λk+q−1eV1 and dω2 ∈ Λk+q−2eV1. All this imply
d({y ∈ Λk+q−2Vr : dy ∈ Λk+q−1V1}) = U ⊕ W
where
U = d({y ∈ Λk+q−2eVr : dy ∈ Λk+q−1eV1}),
W = Rx∗ ∧ d({y ∈ Λk+q−3eVr : dy ∈ Λk+q−2eV1}).
Combining the formulas obtained for the numerator and denominator in Eq. (13) it
follows
≃ Ek−1,q
For those p 6= k − 1 the proof is analogous.
Ek−1,q
r
r
⊕ Ek−1,q−1
r
.
(cid:4)
Applying an inductive procedure, one proves a similar formula for Lie algebras having
an abelian direct factor of dimension s, s ≥ 1.
Corollary 3.4. Let n be a nilpotent Lie algebra that decomposes as a direct sum of ideals
Rs ⊕ h for some s ≥ 1. Let {Ep,q
r } be the spectral sequences associated to n and
h respectively. Then each term of the spectral sequence {Ep,q
r } can be written as a sum of
certain terms of the spectral sequence { Ep,q
r } degenerates at r0 if and
only if { Ep,q
r }. In particular {Ep,q
r } and { Ep,q
r } does.
Example 3.5. Let n be the four dimensional Lie algebra with basis {e1, e2, e3, e4} and
non-zero bracket [e2, e3] = e4. This Lie algebra is 2-step nilpotent and it is isomorphic to
R ⊕ h3; we apply Theorem 3.3 to compute the limit of the spectral sequence of n, {Ep,q
r },
by means of the spectral sequence associated to h3, { Ep,q
r }.
Recall that, Ep,q
∞ = 0 if p < 0 or p ≥ 2. Using the formulas in the previous theorem for
the limit of the spectral sequence of n, we have
(1) Ep,q
(2) E0,0
(3) E1,0
(4) E0,1
(5)
∞ = R,
∞ = 0 if p + q < 0,
∞ = 0, E1,−1
∞ ≃ E1,0
∞ ≃ E0,1
(a) E0,2
(b) E1,1
(c) E0,3
∞ ⊕ R = span{[e2], [e3]} ⊕ R,
∞ = 0,
∞ ≃ E0,2
∞ ≃ E1,1
∞ ≃ E0,3
∞ ⊕ E0,1
∞ ⊕ E1,0
∞ ⊕ E0,2
∞
∞ ≃ span{[e2 ∧ e4], [e3 ∧ e4]},
∞ ≃ span{[e2], [e3]},
≃ span{[e2 ∧ e3 ∧ e4]} ⊕ span{[e2 ∧ e4], [e3 ∧ e4]},
(d) E1,2
(e) E0,4
∞ ≃ E1,2
∞ ≃ E0,4
∞ ⊕ E1,1
∞ ⊕ E0,3
∞ = 0 since E1,1
∞ = E1,2
∞ ≃ span{[e2 ∧ e3 ∧ e4]}.
∞ = 0,
It is not hard to verify the isomorphisms established for the non-zero limit terms of the
spectral sequence { Ep,q
r }.
We conclude this section with some comments on the behavior of the number r0 for
r = 0 and
which the spectral sequence degenerates, i.e. r0 is the minimum r for which dp,q
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
9
consequently Ep,q
nilpotency index and the dimension of the nilpotent Lie algebra n.
∞ for each r ≥ r0 and p, q ∈ Z. This number is related to the
r = Ep,q
When the Lie algebra n is k-step nilpotent it is easy to see that for any r ≥ k, the
has the zero space as either its domain or target
r −→ Ep+r,q−r+1
: Ep,q
differential dp,q
r
space (or both). Hence, the minimum r0 is at most k. Namely,
Proposition 3.6. Let n be a k-step nilpotent Lie algebra and {Ep,q
associated to n. For any r ≥ k it holds Ep,q
∞ for all p, q ∈ Z.
r = Ep,q
r
r } the spectral sequence
The family of nilpotent Lie algebras in Example 2.1 shows that it is not possible to find
for any m ∈ N the spectral sequence
a common r0 valid for all nilpotent Lie algebras:
{Ep,q
r } associated to the nilpotent Lie algebra m0(2m) does not degenerate in the m − 1
step. Indeed, denote with Vi, i = 1, . . . , 2m − 1 the filtration of m0(2m)∗. Recall that
Vi = {e1, . . . , ei+1}, then the 2-form
σ = e2 ∧ e2m − e3 ∧ e2m−1 + · · · + (−1)m−2em ∧ em+2
is not an element of Λ2V2m−2 and by Eq. (9) its differential is
dσ = (−1)m−2 e1 ∧ em ∧ em+1.
Thus dσ ∈ Λ3Vm since Vm = span{e1, . . . , em+1}. Hence σ and e1 ∧ e2m define linearly
independent elements in
E0,2
m−1(m0(2m)) ≃
{x ∈ Λ2n∗ : dx ∈ Λ3Vm}
{x ∈ Λ2V2m−2 : dx ∈ Λ3Vm}
(see Eq. (7)) so dim E0,2
m−1(m0(2m)) ≥ 2. Instead, by Eq. (10)
therefore Ep,q
m−1 is not the limit of the spectral sequence.
dim E0,2
∞ (m0(2m)) = 1,
Remark. It would be interesting to find numbers r0(m) for which the spectral sequence
of any nilpotent Lie algebra n of dimension m degenerates at r ≤ r0(m). From the last
example, this number r0(m) must be at least m/2. We have examples to believe that
r0(m) would be ⌈m/2⌉, the least integer not smaller than m/2.
4. The spectral sequence in low dimensions
Nilpotent Lie algebras over the real numbers are classified up to dimension seven.
Though six is the highest dimension in which there do not exist continuous families [6, 11].
In this section we present the limit of the spectral sequences associated to nilpotent Lie
algebra up to dimension six. The computations were made by hand using, when possible,
the properties proved in the previous section. Later, we checked the calculations with a
computational program we developed. The information obtained is presented in tables as
we explain next.
Given a k-step nilpotent Lie algebra n of dimension m and with spectral sequence {Ep,q
r }
we build up tables to display the dimensions of Ep,q
∞ , precisely, r + 1
tables having k rows and m + 1 columns. In the first column we put the terms of total
degree 0, in the second column the ones of total degree 1, and so on. Set ep,q
r = dim Ep,q
,
thus the table corresponding to r-th term of the spectral sequence of n is as the one below.
1 , . . . , Ep,q
r = Ep,q
0 , Ep,q
r
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
10
ek−1,1−k
r
ek−1,2−k
r
ek−1,3−k
r
...
e1,−1
r
e0,0
r
...
e1,0
r
e0,1
r
...
e1,1
r
e0,2
r
· · ·
· · ·
· · ·
· · ·
ek−1,m+1−k
r
...
e1,m−1
r
e0,m
r
Some of the properties of the spectral sequence established along the previous section
can be read off the diagram corresponding to Ep,q
∞ .
Indeed, because of Theorem 3.2
the first two columns and the last two columns consist of zeros except for the top or
bottom numbers. In fact, the first column (resp. last column) has a number 1 at the top
(resp. bottom) of the table and the second column (resp. penultimate column) has the
dim ker(d : n∗ −→ Λ2n∗) at the top (resp. bottom) of the table.
Furthermore a consequence of Eq.
(11) is that the sum of the elements of the i-th
column gives as result the i-th Betti number of the Lie algebra n, i.e.
βi = dim H i(n) = Xp+q=i
ep,q
∞ .
For nilpotent Lie algebras the Poincar´e duality holds (see [10]) hence the sum of the
elements of the i-th column coincides with the sum of the elements in the (m − i)-th
column.
Theorem 3.3 gives instructions to construct the table of a Lie algebra of the form
n = R ⊕ h once we know the table of h: the i-th column of the table corresponding to n is
obtained by adding up the i-th with the (i − 1)-th columns of the table corresponding to
h with the only exception of the first column.
The notation we use to describe the Lie algebras is that one employed by S. Salamon in
[17]. That is, a Lie algebra denoted as n = (0, 0, 12, 13, 23, 14 + 25) is that one admitting
a basis of 1-forms {e1, e2, e3, e4, e5, e6} where the Lie algebra differential satisfies de1 =
de2 = 0 and
de3 = e1 ∧ e2,
de4 = e1 ∧ e3,
de5 = e2 ∧ e3,
de6 = e1 ∧ e4 + e2 ∧ e5.
The comparison between this notation and that one used by Magnin [11] and Morozov
[13] in their classifications is available in [17] for dimension six.
If a listed nilpotent Lie algebra n admits a decomposition as a direct sum of ideals of
the form n = Rs ⊕ h we will make this fact clear for the reader to compare the tables of n
and h according to Theorem 3.3.
Dimension three
h3 = (0, 0, 12)
E0
2
1
1
2
0
1
1
0
E1
2
1
1
2
0
1
1
0
E2 = E∞
1
0
2
0
0
2
0
1
Dimension four
(1) n = (0, 0, 0, 12), n ≃ R ⊕ h3.
1
0
3
1
E0
3
3
1
3
0
1
1
0
3
1
E1
3
3
1
3
0
1
E2 = E∞
0
3
2
0
2
3
1
0
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
11
(2) n = (0, 0, 12, 13)
E0
1
2
3
1
0
0
2
1
1
0
1
3
0
0
1
E1
1
2
2
1
0
0
2
1
1
0
1
2
0
0
1
E2 = E∞
0
2
0
0
0
1
1
0
2
1
0
0
0
0
1
Dimension five
(1) • n = (0, 0, 12, 13, 14 + 23):
E0
E1
E2
1
0
0
0
2
1
1
1
1
2
3
4
0
1
3
6
(2) • n = (0, 0, 12, 13, 14):
E0
1
0
0
0
2
1
1
1
1
2
3
4
0
1
3
6
(3) • n = (0, 0, 12, 13, 23):
E0
1
0
0
2
1
2
1
2
7
0
1
9
0
0
1
4
0
0
1
4
0
0
5
0
0
0
1
0
0
0
1
0
0
1
(4) • n = (0, 0, 0, 12, 14 + 23):
E0
1
0
0
3
1
1
3
3
4
1
3
6
0
1
4
0
0
1
1
0
0
0
1
0
0
0
1
0
0
1
0
0
2
1
1
1
2
1
1
1
2
1
1
3
1
1
1
2
2
2
1
2
2
2
1
2
4
3
3
3
E1
E1
E1
0
1
2
2
0
1
2
2
0
1
3
1
3
4
0
0
1
2
0
0
1
2
0
0
2
0
1
3
0
0
0
1
0
0
0
1
0
0
1
0
0
1
1
0
0
0
1
0
0
0
1
0
0
1
0
0
2
0
0
0
2
0
0
0
2
0
0
3
0
0
0
1
0
2
0
1
0
2
0
0
3
2
1
1
E2
E2
E2
0
0
2
1
0
0
2
1
0
0
3
0
1
3
0
0
0
2
0
0
0
2
0
0
2
0
0
3
0
0
0
1
0
0
0
1
0
0
1
0
0
1
(5) • n = (0, 0, 0, 12, 24): n = Re3 ⊕ h where h is the 3-step nilpotent Lie algebra of
dimension 4.
E0
E1
E2
1
0
0
3
1
1
3
3
4
1
3
6
0
1
4
0
0
1
1
0
0
3
1
1
3
3
3
1
3
4
0
1
3
0
0
1
1
0
0
3
0
0
2
1
1
0
1
3
0
0
3
0
0
1
(6) • n = (0, 0, 0, 12, 13):
E0
E1
E2
1
0
3
2
3
7
1
9
0
5
0
1
1
0
3
2
3
6
1
6
0
3
0
1
1
0
3
0
1
5
0
6
0
3
0
1
(7) • n = (0, 0, 0, 0, 12): n = Re3 ⊕ Re4 ⊕ h3.
E0
E1
E2
1
0
4
1
6
4
4
6
1
4
0
1
1
0
4
1
6
4
4
6
1
4
0
1
1
0
4
0
5
2
2
5
0
4
0
1
(8) • n = (0, 0, 0, 0, 12 + 34):
E0
E1
E2
1
0
4
1
6
4
4
6
1
4
0
1
1
0
4
1
6
4
4
6
1
4
0
1
1
0
4
0
5
0
0
5
0
4
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
12
Dimension six
(1) • n = (0, 0, 12, 13, 14 + 23, 34 + 52):
E0
0
1
3
6
0
0
1
4
10
10
1
0
0
0
0
2
1
1
1
1
1
2
3
4
5
0
0
0
1
5
0
0
0
0
1
1
0
0
0
0
2
1
1
1
1
1
2
2
2
2
E1
0
1
2
2
3
(2) • n = (0, 0, 12, 13, 14, 34 + 52):
E0
0
1
3
6
0
0
1
4
10
10
1
0
0
0
0
2
1
1
1
1
1
2
3
4
5
0
0
0
1
5
0
0
0
0
1
1
0
0
0
0
2
1
1
1
1
(3) • n = (0, 0, 12, 13, 14, 15):
E0
0
1
3
6
0
0
1
4
10
10
1
0
0
0
0
2
1
1
1
1
1
2
3
4
5
0
0
0
1
5
0
0
0
0
1
1
0
0
0
0
2
1
1
1
1
E1
0
1
2
2
3
E1
0
1
2
2
3
1
2
2
2
2
1
2
2
2
2
0
0
1
2
3
0
0
1
2
3
0
0
1
2
3
0
0
0
1
2
0
0
0
1
5
0
0
0
1
2
(4) • n = (0, 0, 12, 13, 14 + 23, 15 + 24):
E0
0
1
3
6
0
0
1
4
10
10
1
0
0
0
0
2
1
1
1
1
1
2
3
4
5
0
0
0
1
5
0
0
0
0
1
1
0
0
0
0
2
1
1
1
1
1
2
2
2
2
E1
0
1
2
2
3
(5) • n = (0, 0, 12, 13, 14, 15 + 23):
E0
0
1
3
6
0
0
1
4
10
10
1
0
0
0
0
2
1
1
1
1
1
2
3
4
5
0
0
0
1
5
0
0
0
0
1
1
0
0
0
0
2
1
1
1
1
1
2
2
2
2
E1
0
1
2
2
3
(6) • n = (0, 0, 12, 13, 23, 14):
0
0
1
2
2
0
0
1
2
3
0
0
0
1
2
0
0
0
1
2
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
E2
0
0
2
1
1
E2
0
0
2
1
1
E2
0
0
2
1
2
E2
0
0
2
1
2
E2
0
0
2
1
2
0
1
0
1
2
0
1
0
1
2
0
1
0
1
2
0
1
0
1
2
0
1
0
1
2
0
0
0
0
2
0
0
0
0
2
0
0
0
1
2
0
0
0
1
2
0
0
0
1
2
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
E3 = E∞
0
1
0
1
0
0
0
0
1
1
0
0
0
0
2
E3 = E∞
0
1
0
1
0
0
0
0
1
1
0
0
0
0
2
E3 = E∞
0
1
0
1
1
0
0
1
1
2
0
0
0
1
2
E3 = E∞
0
1
0
1
1
0
0
1
1
2
0
0
0
1
2
E3 = E∞
0
1
0
1
1
0
0
1
1
2
0
0
0
1
2
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
2
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
E0
0
1
9
0
0
5
10
10
1
0
0
0
2
1
2
1
1
2
7
5
0
0
1
5
0
0
0
1
1
0
0
0
(7) • n = (0, 0, 12, 13, 23, 14 − 25):
E0
0
1
9
0
0
5
10
10
1
0
0
0
2
1
2
1
1
2
7
5
0
0
1
5
0
0
0
1
1
0
0
0
E1
0
1
3
3
E1
0
1
3
3
1
2
4
2
1
2
4
2
0
0
2
3
0
0
2
3
2
1
2
1
2
1
2
1
0
0
1
2
0
0
1
2
0
0
0
1
0
0
0
1
E2 = E∞
0
0
2
2
0
0
3
3
0
0
2
2
E2 = E∞
0
0
2
2
0
0
3
3
0
0
2
2
1
0
0
0
1
0
0
0
2
0
0
0
2
0
0
0
0
0
0
2
0
0
0
2
0
0
0
1
0
0
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
13
(8) • n = (0, 0, 12, 13, 23, 14 + 25):
E0
0
1
9
0
0
5
10
10
1
0
0
0
2
1
2
1
1
2
7
5
0
0
1
5
0
0
0
1
1
0
0
0
2
1
2
1
1
2
4
2
E1
0
1
3
3
E2 = E∞
0
0
2
3
0
0
1
2
0
0
0
1
1
0
0
0
2
0
0
0
0
0
2
2
0
0
3
3
0
0
2
2
0
0
0
2
0
0
0
1
(9) • n = (0, 0, 0, 12, 14 − 23, 15 + 34):
E0
1
3
6
0
1
4
10
10
1
0
0
0
3
1
1
1
3
3
4
5
0
0
1
5
0
0
0
1
1
0
0
0
3
1
1
1
3
3
3
3
E1
1
3
4
4
0
1
3
4
0
0
1
3
0
0
0
1
1
0
0
0
3
0
0
0
2
1
0
2
E2
0
1
2
2
E3 = E∞
0
0
1
3
0
0
0
3
0
0
0
1
1
0
0
0
3
0
0
0
2
1
0
1
0
0
2
2
0
0
1
3
0
0
0
3
0
0
0
1
(10) • n = (0, 0, 0, 12, 14, 15 + 23):
E0
1
3
6
0
1
4
10
10
1
0
0
0
3
1
1
1
3
3
4
5
0
0
1
5
0
0
0
1
1
0
0
0
3
1
1
1
(11) • n = (0, 0, 0, 12, 14, 15 + 23 + 24):
E0
1
3
6
0
1
4
10
10
1
0
0
0
3
1
1
1
3
3
4
5
0
0
1
5
0
0
0
1
1
0
0
0
3
1
1
1
E1
1
3
4
4
E1
1
3
4
4
3
3
3
3
3
3
3
3
0
1
3
4
0
1
3
4
0
0
1
3
0
0
1
3
0
0
0
1
0
0
0
1
E2 = E∞
2
1
0
2
0
1
2
3
0
0
2
3
E2 = E∞
2
1
0
2
0
1
2
3
0
0
2
3
1
0
0
0
1
0
0
0
3
0
0
0
3
0
0
0
0
0
0
3
0
0
0
3
0
0
0
1
0
0
0
1
(12) • n = (0, 0, 0, 12, 14, 15+24): n = Re3⊕h where h is the Lie algebra (1) of dimension
5.
E0
1
3
6
0
1
4
10
10
1
0
0
0
3
1
1
1
3
3
4
5
0
0
1
5
0
0
0
1
1
0
0
0
3
1
1
1
3
3
3
3
E1
1
3
4
4
E2 = E∞
0
1
3
4
0
0
1
3
0
0
0
1
1
0
0
0
3
0
0
0
2
1
0
2
0
1
2
3
0
0
2
3
0
0
0
3
0
0
0
1
(13) • n = (0, 0, 0, 12, 14, 15): n = Re3 ⊕ h where h is the Lie algebra (2) of dimension 5.
E0
1
3
6
0
1
4
10
10
1
0
0
0
3
1
1
1
3
3
4
5
0
0
1
5
0
0
0
1
1
0
0
0
(14) • n = (0, 0, 0, 12, 13, 14 + 35):
E0
1
9
0
5
10
10
1
0
0
3
2
1
3
7
5
0
1
5
0
0
1
1
0
0
E1
1
3
4
4
E1
1
6
6
3
3
3
3
3
6
3
0
1
3
4
0
3
6
3
1
1
1
3
2
1
0
0
1
3
0
1
3
0
0
0
1
0
0
1
E2 3 = E∞
2
1
0
2
0
1
2
3
0
0
2
3
E2 = E∞
1
4
0
0
3
3
0
0
5
1
0
0
0
1
0
0
3
0
0
0
3
0
0
0
0
0
3
0
0
3
0
0
0
1
0
0
1
(15) • n = (0, 0, 0, 12, 23, 14 + 35): n is the Lie algebra of upper 4 × 4 real matrices.
E0
1
9
0
5
10
10
1
0
0
3
2
1
3
7
5
0
1
5
0
0
1
1
0
0
3
2
1
3
6
3
E1
1
6
6
E2 = E∞
0
3
6
0
1
3
0
0
1
1
0
0
3
0
0
1
4
0
0
3
3
0
0
5
0
0
3
0
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
14
(16) • n = (0, 0, 0, 12, 23, 14 − 35):
E0
1
9
0
5
10
10
1
0
0
3
2
1
3
7
5
0
1
5
0
0
1
1
0
0
3
2
1
3
6
3
E1
1
6
6
E2 = E∞
0
3
6
0
1
3
0
0
1
1
0
0
3
0
0
1
4
0
0
3
3
0
0
5
0
0
3
0
0
1
(17) • n = (0, 0, 0, 12, 14, 24):n = Re3 ⊕ h where h is the Lie algebra (3) of dimension 5.
E0
1
3
0
1
16
14
1
0
0
3
1
2
3
3
9
0
0
6
0
0
1
1
0
0
3
1
2
(18) • n = (0, 0, 0, 12, 13 − 24, 14 + 23):
E0
1
3
0
1
16
14
1
0
0
3
1
2
3
3
9
0
0
6
0
0
1
1
0
0
(19) • n = (0, 0, 0, 12, 14, 13 − 24):
E0
1
3
0
1
16
14
1
0
0
3
1
2
3
3
9
0
0
6
0
0
1
1
0
0
(20) • n = (0, 0, 0, 12, 13 + 14, 24):
E0
1
3
0
1
16
14
1
0
0
3
1
2
3
3
9
0
0
6
0
0
1
1
0
0
(21) • n = (0, 0, 0, 12, 13, 14 + 23):
E0
1
9
0
5
10
10
0
1
5
0
0
1
1
0
0
3
2
1
3
7
5
(22) • n = (0, 0, 0, 12, 13, 24):
E0
1
9
0
5
10
10
0
1
5
0
0
1
1
0
0
3
2
1
3
7
5
(23) • n = (0, 0, 0, 12, 13, 14):
E0
1
9
0
5
10
10
0
1
5
0
0
1
1
0
0
3
2
1
3
7
5
(24) • n = (0, 0, 0, 12, 13, 23):
1
0
0
1
0
0
1
0
0
3
1
2
3
1
2
3
1
2
3
2
1
3
2
1
3
2
1
3
3
6
3
3
6
3
3
6
3
3
6
3
6
3
3
6
3
3
6
3
E0
1
19
1
0
3
3
3
12
0
15
0
6
0
1
1
0
3
3
3
9
E1
1
3
7
E1
1
3
7
E1
1
3
7
E1
1
3
7
E1
1
6
6
E1
1
6
6
E1
1
6
6
E1
1
12
0
1
5
0
1
5
0
1
5
0
1
5
0
3
6
0
3
6
0
3
6
0
0
3
0
0
3
0
0
3
0
0
3
0
1
3
0
1
3
0
1
3
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
E2 = E∞
2
0
3
0
0
6
0
0
5
E2 = E∞
2
0
3
0
0
6
0
0
5
E2 = E∞
2
0
3
0
0
6
0
0
5
E2 = E∞
2
0
3
0
0
6
0
0
5
E2 = E∞
1
4
1
0
4
4
0
1
5
E2 = E∞
1
4
1
0
4
4
0
1
5
E2 = E∞
1
4
1
0
4
4
0
1
5
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
0
0
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
3
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
E2 = E∞
0
8
0
3
0
1
1
0
3
0
0
8
0
12
0
8
0
3
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
15
(25) • n = (0, 0, 0, 0, 12, 15 + 34):
E0
4
6
1
4
10
10
1
0
0
4
1
1
6
4
4
0
1
4
0
0
1
1
0
0
4
1
1
6
4
4
E1
4
6
7
1
4
7
0
1
4
0
0
1
1
0
0
4
0
0
5
1
1
E2
2
2
3
E3 = E∞
0
0
6
0
0
4
0
0
1
1
0
0
4
0
0
5
1
0
1
2
3
0
0
6
0
0
4
0
0
1
(26) • n = (0, 0, 0, 0, 12, 15): n = Re3 ⊕ Re4 ⊕ h where h is the 3-step nilpotent Lie
algebra of dimension four.
E0
4
6
1
4
10
10
0
1
4
0
0
1
1
0
0
4
1
1
6
4
5
(27) • n = (0, 0, 0, 0, 12, 14 + 25):
E0
4
6
1
4
10
10
0
1
5
0
0
1
1
0
0
4
1
1
6
4
5
E1
4
6
7
E1
4
6
7
6
4
4
6
4
4
1
4
7
1
4
7
1
0
0
1
0
0
4
1
1
4
1
1
0
1
4
0
1
4
0
0
1
0
0
1
E2 = E∞
5
1
1
2
2
4
0
1
6
E2 = E∞
5
1
1
2
2
4
0
1
6
1
0
0
1
0
0
4
0
0
4
0
0
0
0
4
0
0
4
0
0
1
0
0
1
(28) • n = (0, 0, 0, 0, 13 + 42, 14 + 23):
E0
1
0
4
2
6
9
4
16
1
14
0
6
0
1
1
0
4
2
6
8
(29) • n = (0, 0, 0, 0, 12, 14 + 23):
E0
1
0
4
2
6
9
4
16
1
14
0
6
0
1
1
0
4
2
6
8
(30) • n = (0, 0, 0, 0, 12, 34):
E0
1
0
4
2
6
9
4
16
1
14
0
6
0
1
1
0
4
2
6
8
E1
4
11
E1
4
11
E1
4
11
E2 = E∞
1
8
0
4
0
1
1
0
4
0
4
4
0
10
0
8
0
4
0
1
E2 = E∞
1
8
0
4
0
1
1
0
4
0
4
4
0
10
0
8
0
4
0
1
E2 = E∞
1
8
0
4
0
1
1
0
4
0
4
4
0
10
0
8
0
4
0
1
(31) • n = (0, 0, 0, 0, 12, 13): n = Re4 ⊕ h where h is the Lie algebra (6) of dimension 5.
E0
1
0
4
2
6
9
4
16
1
14
0
6
0
1
1
0
4
2
6
8
E1
4
12
E2 = E∞
1
9
0
4
0
1
1
0
4
0
4
5
1
11
0
9
0
4
0
1
(32) • n = (0, 0, 0, 0, 0, 12 + 34): n = Re5 ⊕ h where h is the Lie algebra (8) of dimension
5.
E0
10
10
1
0
5
1
10
5
5
10
1
5
0
1
1
0
5
1
10
5
E1
10
10
E2 = E∞
5
10
1
5
0
1
1
0
5
0
9
0
5
5
0
9
0
5
0
1
(33) • n = (0, 0, 0, 0, 0, 12):n = Re3 ⊕ Re4 ⊕ h where h is the nilpotent Lie algebra or
dimension 3.
E0
10
10
1
0
5
1
10
5
5
10
1
5
0
1
1
0
5
1
10
5
E1
10
10
E2 = E∞
5
10
1
5
0
1
1
0
5
0
9
2
7
7
2
9
0
5
0
1
ON A SPECTRAL SEQUENCE FOR THE COHOMOLOGY OF A NILPOTENT LIE ALGEBRA
16
From simple inspection of the tables of the limit of the spectral sequence showed above
we can see that in dimension five there are 6 different configurations of tables that corre-
spond to 8 isomorphisms classes of nilpotent Lie algebras in that dimension. Meanwhile,
there are 15 different configurations of tables that correspond to 33 isomorphisms classes
of nilpotent Lie algebras in dimension six. Thus, there are non-isomorphic nilpotent Lie
algebras with the same table. Nevertheless, notice that the Lie algebras 16 and 17 of
dimension six have the same Betti numbers but different tables.
References
[1] C. Benson and C. Gordon, Kahler and symplectic structures on nilmanifolds, Topology 27 No. 4
(1988), 513 -- 518.
[2] G. Cairns and B. Jessup, New bounds on the Betti numbers of nilpotent Lie algebras, Commun.
Algebra 25 No. 2 (1997), 415 -- 430.
[3] S. Cical`o, W. de Graaf and C. Schneider, Six-dimensional nilpotent Lie algebras, Linear Algebra
Appl. 436 No. 1 (2012), 163 -- 189.
[4] V. del Barco, Symplectic structures on nilmanifolds: an obstruction for its existence, preprint
(2012), arXiv:1210.6296.
[5] Ch. Deninger and W. Singhof, On the cohomology of nilpotent Lie algebras, Bull. Soc. Math. Fr.
116 No. 1 (1988), 3 -- 14.
[6] M. Goze and Y. Khakimdjanov, Nilpotent Lie algebras, Kluwer Academic, Dordrecht, 1996.
[7] J. Grassberger, A. King and P. Tirao, On the homology of free 2-step nilpotent Lie algebras, J.
Algebra 254 No. 2 (2002), 213 -- 225.
[8] S. Halperin, Le complexe de Koszul en alg`ebre et topologie, Ann. l'Inst. Fourier 37 (1987), 77 -- 97.
[9] B. Kostant, Lie algebra cohomology and the generalized Borel-Weil theorem, Ann. of Math. 72
(1961), 329 -- 390.
[10] J.-L Koszul, Homologie et cohomologie des alg`ebres de Lie, Bull. Soc. Math. France 78 (1950),
65 -- 127.
[11] L. Magnin, Sur les alg`ebres de Lie nilpotentes de dimension ≤ 7, J. Geom. Phys. 3 (1986), 119 -- 131.
[12] D. Millionschikov, Graded filiform Lie algebras and symplectic nilmanifolds, Amer. Math. Soc.
Transl. Ser. 2 212 (2004), 259 -- 279.
[13] V. Morozov, Classification of nilpotent Lie algebras of sixth order, Izv. Vyss. Ucebn. Zaved. Matem-
atika 4 No. 5 (1958), 161 -- 171.
[14] K. Nomizu, On the cohomology of compact homogeneous spaces of nilpotent Lie groups, Ann. Math.
(2), 59 (1954), 531 -- 538.
[15] H. Pouseele, Betti number behavior for nilpotent Lie algebras, Geom. Dedicata 122 (2006), 77 -- 88.
[16] H. Pouseele and P. Tirao, Constructing Lie algebra homology classes, J. Algebra 292 No. 2 (2005),
585 -- 591.
[17] S. Salamon, Complex structures on nilpotent Lie algebras, J. of Pure Appl. Algebra 157 (2001),
311 -- 333.
[18] L. Santharoubane, Cohomology of Heisenberg Lie algebras, Proc. Am. Math. Soc. 87 (1993), 23 -- 28.
[19] S. Siggs, Laplacian and homology of free two-step nilpotent Lie algebras, J. Algebra 185 No. 1 (1996),
144 -- 161.
[20] P. Tirao, A refinement of the Toral Rank Conjecture for 2-Step nilpotent Lie algebras, Proc. Am.
Math. Soc. 128 No. 10 (2000), 2875 -- 2878.
[21] C. Weibel, An introduction to homological algebra, Cambridge University Press, Cambridge, 1994.
[22] T. Yamada, Harmonic cohomology groups on compact symplectic nilmanifolds, Osaka J. Math. 39
(2002), 363 -- 381.
Viviana del Barco. Facultad de Ciencias Exactas Ingenier´ıa y Agrimensura, Universidad
Nacional de Rosario, Av. Pellegrini 250, (2000) Rosario, Argentina.
E-mail address: [email protected]
|
1707.05526 | 2 | 1707 | 2018-02-12T09:07:36 | Classification of involutions on graded-division simple real algebras | [
"math.RA"
] | We classify, up to isomorphism and up to equivalence, involutions on graded-division finite-dimensional simple real (associative) algebras, when the grading group is abelian. | math.RA | math |
CLASSIFICATION OF INVOLUTIONS ON GRADED-DIVISION
SIMPLE REAL ALGEBRAS
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
Abstract. We classify, up to isomorphism and up to equivalence, involutions
on graded-division finite-dimensional simple real (associative) algebras, when
the grading group is abelian.
Contents
Introduction
1.
2. Background on involutions
3. Background on gradings
4. Quadratic forms
5. Division gradings and quadratic forms
6. Classification in the one-dimensional case
7. Classification in the two-dimensional non-complex case
8. Classification in the two-dimensional complex case
9. Classification in the four-dimensional case
10. Semisimple algebras with involution
11. Distinguished involutions
Acknowledgements and license
References
1
3
4
6
8
11
14
19
22
26
27
28
29
1. Introduction
The study of gradings on various algebras has recently become an active research
field -- see the monograph [9] and the references therein for an overview of this
topic. One of the milestone results in that monograph (following [5, 2, 8]) is the
classification of gradings on classical simple Lie algebras over algebraically closed
fields of characteristic different from 2. It was achieved by first reducing the problem
to the classification of gradings on finite-dimensional simple associative algebras
with involution (or, more generally, an antiautomorphism).
This was the main reason to write this article: ultimately, we want to classify
gradings on real Lie algebras, and the first step in our approach is to study involu-
tions on graded-division real associative algebras. In fact, we have already finished
the classification of gradings on classical central simple real Lie algebras (except
those of type D4). The results are to appear in a separate article (see preprint
[3]), in which some of the arguments rely on this paper. On the other hand, the
classification of involutions (and related objects) may be of independent interest.
Date: 12 February 2018.
2010 Mathematics Subject Classification. Primary 16W50, 16W10; secondary 16K20, 16S35.
Key words and phrases. Graded algebra;
involution; division grading; simple real algebra;
classification.
https://doi.org/10.1016/j.laa.2018.01.040 This is the accepted manuscript (accommodat-
ing the referee's suggestions) of the article published in Linear Algebra and its Applications.
1
2
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
Involutions on graded-division finite-dimensional simple complex algebras are
classified in [9, Propositions 2.51 and 2.53] (see also [4]). In this paper we solve the
real case. As a prerequisite, we need to know the classification of division gradings
on finite-dimensional simple real algebras (without involution). This classification
has been done in [13], both up to isomorphism and up to equivalence, and indepen-
dently in [6], up to equivalence (but note that one of the equivalence classes was
overlooked). A classification up to equivalence has been obtained in [7] without
assuming simplicity.
The main objective of this work is to classify, up to isomorphism and up to
equivalence, involutions on graded-division simple real associative algebras of finite
dimension, when the grading group is abelian. We consider only abelian grading
groups here because of our intended applications: the support of a grading on a
simple Lie algebra always generates an abelian subgroup of the grading group (see
for example [9, Proposition 1.12]). Our main classification results are achieved in
Sections 6, 7, 8 and 9.
The paper is structured as follows. We have collected the properties that charac-
terize involutions on finite-dimensional simple real algebras in Section 2. Other pre-
liminaries, such as the definitions of isomorphism, equivalence and division grading,
can be found in Section 3, together with the rest of terminology related to gradings
that we use in the paper. Our main classification results are presented in terms of
quadratic forms on certain abelian groups and a similar kind of maps (which we
call "nice maps"). These objects are introduced in Section 4.
All homogeneous components of finite-dimensional graded-division real algebras
have the same dimension, which can be 1, 2 or 4, according to the identity com-
ponent being the field of real numbers R, the field of complex numbers C or the
division algebra of quaternions H. In the case of dimension 2, the identity com-
ponent may or may not be contained in the center of the algebra. Consequently,
our classification results are arranged into four sections. In Section 6, we classify
involutions on graded-division algebras whose homogeneous components have di-
mension 1. In Section 8, we consider the case of dimension 2 where the identity
component is contained in the center, or, equivalently, the center is C with the triv-
ial grading; in this situation the algebra can be regarded as a graded algebra over
C. In Section 7, we also study the case of dimension 2, but the identity component
is not contained in the center. Finally, the case of dimension 4 is reduced to the
case of dimension 1 thanks to the Double Centralizer Theorem, as stated in Section
9. Note that these four sections are written as if they were very long theorems; we
have made an effort to compile the classification to serve as a reference.
Section 5 is written in the same style, that is, as if it were a very long theorem,
but its motivation is different. Instead of classifying involutions, we classify division
gradings. Moreover, the underlying algebra is not necessarily simple here. The main
goal of this section is to classify all quadratic forms that will appear in the following
sections and, in particular, establish their existence. Thus, the logic of this section
has the opposite direction as compared to the rest of the text.
As mentioned above, we use the results of this paper to classify gradings on
classical real Lie algebras in [3]. There, in the case of outer gradings on special
linear Lie algebras (which belong to series A), we have to deal with associative
algebras that are not simple, but simple as algebras with involution. So, in Section
10 of this paper, we extend a part of the results of the previous sections to algebras
whose center is isomorphic to R × R.
Finally, in Section 11, we discuss involutions with special properties, which we
call "distinguished involutions". We use them in our preprint [3], but they may also
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
3
be of independent interest. For example, in the situation of Section 8, they allow
us to construct a special basis for a part of the graded-division algebra.
2. Background on involutions
In this section we review the basic properties of involutions on finite-dimensional
simple real algebras. We will use [12] as a reference.
An antiautomorphism of an algebra D is a map ϕ : D → D which is an isomor-
phism of vector spaces and such that ϕ(xy) = ϕ(y)ϕ(x) for all x, y ∈ D. If it also
satisfies ϕ2(x) = x for all x ∈ D, ϕ is called an involution.
Let ϕ be an involution on a real algebra D. The center Z(D) is preserved under
ϕ, so either the restriction of ϕ to Z(D) is the identity and the involution is said
to be of the first kind, or this restriction has order 2 and the involution is said to
be of the second kind.
Let F be either R or C, and let V be an F-vector space of dimension n. An
F-bilinear form b : V × V → F is called nonsingular (or nondegenerate) if the only
element x ∈ V such that b(x, y) = 0 for all y ∈ V is x = 0. The following is
well known (see [12, p. 1]). First, given one such b, there exists a unique map
σb : EndF(V ) → EndF(V ) that satisfies the equation
b(x, f(y)) = b(σb(f)(x), y)
for all x, y ∈ V and f ∈ EndF(V ). Second, the map b 7→ σb induces a bijective
correspondence between the classes of nonsingular F-bilinear forms on V that are
either symmetric or skew-symmetric, up to a factor in F×, and involutions (of
the first kind in the case F = C) on EndF(V ) (∼= Mn(F)). The involutions that
are adjoint to symmetric bilinear forms are called orthogonal, while those that are
adjoint to skew-symmetric bilinear forms are called symplectic.
Let ϕ be an orthogonal involution on Mn(R), and take a nonsingular symmetric
bilinear form b on a real vector space V such that ϕ corresponds to σb via some
isomorphism Mn(R) ∼= EndR(V ). The number m+ (respectively m−) of positive
(respectively negative) entries in a diagonalization of b does not depend on the
choice of the orthogonal basis. Therefore, m+ − m− is an invariant of ϕ, called
its signature.
An involution ϕ on Mn(H) is called orthogonal or symplectic if so is its complexi-
fication ϕ ⊗R idC. We will use the following characterization ([12, Proposition 2.6]).
Let D be a finite-dimensional simple real algebra, and let ϕ be an involution on D
(of the first kind if D ∼= Mn(C)); then ϕ is orthogonal if and only if the dimension
of {x ∈ D ϕ(x) = +x} is greater than the dimension of {x ∈ D ϕ(x) = −x},
while it is symplectic if and only if it is smaller.
Let D be either H or C, let V be a right D-vector space of dimension n, and
denote by x the conjugate of x in D. A hermitian form on V is an R-bilinear
map h : V × V → D such that, for all x, y ∈ V and a, b ∈ D, we have: (1)
h(xa, yb) = ah(x, y)b and (2) h(y, x) = h(x, y). The form is called skew-hermitian
if condition (2) is replaced by: (2') h(y, x) = −h(x, y). Thus, these forms are
sesquilinear:
linear in the second variable and semilinear in the first. If we take
D = R (with x = x) then we recover the definitions of symmetric and skew-
symmetric forms.
A hermitian or skew-hermitian form h is called nonsingular if the only element
x ∈ V such that h(x, y) = 0 for all y ∈ V is x = 0. It is well known (see [12,
Proposition 4.1]) that, given one such h, there exists a unique map σh : EndD(V ) →
EndD(V ) that satisfies the equation
(1)
for all x, y ∈ V and f ∈ EndD(V ). Also, by [12, Theorem 4.2], we have the following.
h(x, f(y)) = h(σh(f)(x), y)
4
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
• In the case D = H, the map h 7→ σh defines a bijective correspondence
between the classes of nonsingular hermitian (respectively skew-hermitian)
forms on V , up to a factor in R×, and symplectic (respectively orthogonal)
involutions on EndH(V ) (∼= Mn(H)).
• In the case D = C, the map h 7→ σh defines a bijective correspondence
between the classes of nonsingular hermitian forms on V , up to a factor in
R×, and involutions of the second kind on EndC(V ) (∼= Mn(C)).
For a symplectic involution on Mn(H) or an involution of the second kind on
Mn(C), we define, in the same way as in the case of orthogonal involutions on
Mn(R), the signature as the absolute value of the difference between the number of
positive and negative entries in any diagonalization of any adjoint hermitian form.
Finally, let us also state a couple of lemmas for future reference.
Lemma 1. Let ϕ1 be an orthogonal involution on Mn1(R). Let D be R (respectively
H, C), and let ϕ2 be an orthogonal (respectively symplectic, second kind) involution
on Mn2(D). Then ϕ1 ⊗R ϕ2 is an orthogonal (respectively symplectic, second kind)
involution on Mn1(R) ⊗R Mn2(D), and its signature is the product of the signatures
of ϕ1 and ϕ2.
Proof. Assume that ϕ1 is adjoint to the bilinear form b1 : V1 × V1 → R and ϕ2 is
adjoint to the hermitian form h2 : V2 × V2 → D. Note that we have the natural
isomorphism of real algebras:
EndR(V1) ⊗R EndD(V2) ∼= EndD(V1 ⊗R V2).
Through these identifications, b1 ⊗R h2 is a hermitian form on V1 ⊗R V2 adjoint
to ϕ1 ⊗R ϕ2. Picking orthogonal bases in V1 and V2, we reduce the proof to a
straightforward combinatorial fact.
(cid:3)
Lemma 2. Let ϕ1 and ϕ2 be second kind involutions on Mn1(C) and Mn2(C).
Then there is a unique second kind involution on Mn1(C) ⊗C Mn2(C) that sends
X1 ⊗C X2 to ϕ1(X1) ⊗C ϕ2(X2); we denote this map by ϕ1 ⊗C ϕ2. Moreover, its
signature is the product of the signatures of ϕ1 and ϕ2.
Proof. Let us just recall the well known construction of ϕ1 ⊗C ϕ2, because the rest
of the proof is analogous to the proof of Lemma 1. We can consider the C-vector
space Mni(C), which has the same underlying abelian group as Mni(C), but a
twisted scalar multiplication ∗ given by α ∗ X := αX. If we denote by ϕi the map
ϕi viewed as a map from Mni(C) to Mni(C), then ϕi is C-linear. Therefore, we
have the C-linear map:
ϕ1 ⊗C ϕ2 : Mn1(C) ⊗C Mn2(C) −→ Mn1(C) ⊗C Mn2(C).
On the other hand, we have a natural C-linear isomorphism:
Mn1(C) ⊗C Mn2(C) −→ Mn1(C) ⊗C Mn2(C).
Finally, ϕ1 ⊗C ϕ2 is the C-semilinear map corresponding to the composition of the
two maps above, and it sends X1 ⊗C X2 to ϕ1(X1) ⊗C ϕ2(X2).
(cid:3)
3. Background on gradings
In this section we review, following [9], the basic definitions and properties of
gradings that will be used in the rest of the paper. Here we only deal with associative
algebras.
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
5
Definition 3. Let D be an algebra over a field F, and let G be a group. A G-
grading Γ on D is a decomposition of D into a direct sum of subspaces indexed by
G,
Γ : D = Mg∈G
Dg,
such that, for all g, h ∈ G, we have
DgDh ⊆ Dgh.
If such a decomposition is fixed, we refer to D as a G-graded algebra. The support
of Γ (or of D) is the set supp(Γ) := {g ∈ G Dg 6= 0}. If X ∈ Dg, then we say
that X is homogeneous of degree g, and we write deg(X) = g. The subspace Dg is
called the homogeneous component of degree g.
Note that, if D is a G-graded algebra and D′ is an H-graded algebra, then the
h,
tensor product D⊗D′ has a natural G×H-grading given by (D⊗D′)(g,h) = Dg ⊗D′
for all g ∈ G, h ∈ H. This will be called the product grading.
A subspace F (in particular, a subalgebra or an ideal) of a G-graded algebra D
is said to be graded if F = Lg∈G(Dg ∩ F).
There are two natural ways to define an equivalence relation on group gradings,
depending on whether the grading group plays a secondary role or is a part of the
definition.
Definition 4. Let Γ be a G-grading on the algebra D and let Γ′ be an H-grading
on the algebra D′. We say that Γ and Γ′ are equivalent if there exist an isomorphism
of algebras ψ : D → D′ and a bijection α : supp(Γ) → supp(Γ′) such that ψ(Dt) =
D′
α(t) for all t ∈ supp(Γ).
Definition 5. Let Γ and Γ′ be G-gradings on the algebras D and D′, respectively.
We say that Γ and Γ′ are isomorphic if there exists an isomorphism of algebras
ψ : D → D′ such that ψ(Dg) = D′
g for all g ∈ G.
Definition 6. Given gradings on the same algebra, Γ : D = Lg∈G Dg and Γ′ :
h, we say that Γ′ is a coarsening of Γ, or that Γ is a refinement of
D = Lh∈H D′
h. If, for some g ∈ G,
Γ′, if, for any g ∈ G, there exists h ∈ H such that Dg ⊆ D′
this inclusion is strict, then we will speak of a proper refinement or coarsening. A
grading is said to be fine if it does not admit a proper refinement.
Definition 7. A graded algebra is said to be a graded division algebra if it is unital
and every nonzero homogeneous element has an inverse. In this case, the grading
will be called a division grading.
If D is a G-graded division algebra, then I ∈ De, where e is the identity element
of G and I the unity of D. Also, if 0 6= X ∈ Dg, then X −1 ∈ Dg−1. Therefore,
the support of D is a subgroup of G, since whenever Dg 6= 0 and Dh 6= 0, we also
have 0 6= DgDh ⊆ Dgh and Dg−1 6= 0. This also shows that, in the situation of
Definition 4, if Γ and Γ′ are division gradings, then α : supp(Γ) → supp(Γ′) is a
homomorphism of groups.
The identity component De of a graded division algebra D is a division algebra.
Also, if Xt ∈ Dt is nonzero, then Dt = DeXt. Therefore, all the (nonzero) homoge-
neous components of the grading have the same dimension. In our case D will be
finite-dimensional and the ground field will be R, so this dimension must be 1, 2 or
4 depending on whether De is isomorphic to R, C or H.
Definition 8. Let D be a G-graded algebra. A map ϕ : D → D is said to be
an antiautomorphism of the G-graded algebra D if it is an isomorphism of vector
spaces such that ϕ(XY ) = ϕ(Y )ϕ(X) for all X, Y ∈ D and ϕ(Dg) = Dg for all
6
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
g ∈ G. If it also satisfies ϕ2(X) = X for all X ∈ D, ϕ is called an involution of the
G-graded algebra D.
Definition 9. Let Γ and Γ′ be gradings on the algebras D and D′. Let ϕ : D → D
and ϕ′ : D′ → D′ be antiautomorphisms of graded algebras. We say that (Γ, ϕ)
is isomorphic (respectively equivalent) to (Γ′, ϕ′) if there exists an isomorphism
(respectively equivalence) of graded algebras ψ : D → D′ such that ϕ′ = ψϕψ−1.
We will use the following result [8, Lemma 3.3].
Lemma 10. Let D be an algebra endowed with a division grading by an abelian
group G. Let X ∈ D× and consider the corresponding inner automorphism Int(X) :
D → D given by Int(X)(Y ) = XY X −1. If Int(X)(Dg) ⊆ Dg for all g ∈ G, then
there exists a nonzero homogeneous X0 ∈ D such that Int(X) = Int(X0).
Proof. Let ψ := Int(X) and note that ψ(Y )X = XY for all Y ∈ D. Write X =
X0 + . . .+ Xn where the Xi are nonzero homogeneous elements of pairwise different
degrees gi. If Y ∈ D is homogeneous of degree g, so is ψ(Y ). Since G is abelian, if we
consider the terms of degree gg0 in the equation ψ(Y )X = XY , we get ψ(Y )X0 =
X0Y . But X0 is invertible because it is homogeneous, so ψ(Y ) = X0Y X −1
for all
Y ∈ Dg and g ∈ G.
(cid:3)
0
4. Quadratic forms
In this section we introduce the necessary terminology concerning quadratic
forms on certain abelian groups, mainly following [13, Section 4]. That article
established a correspondence between isomorphism classes of division gradings and
quadratic forms that are regular in the sense that usually appears in the literature.
In this paper, however, we deal with quadratic forms that satisfy less restrictive
conditions of regularity. The notation here is congruent with [13], but note that
now we do not require quadratic forms and nice maps to be defined on elementary
abelian 2-groups.
Definition 11. Let T be a finite abelian group. In this article, by an alternating
bicharacter on T we will mean a map β : T × T → R× that satisfies β(uv, w) =
β(u, w)β(v, w), β(u, vw) = β(u, v)β(u, w), and β(u, u) = 1 for all u, v, w ∈ T .
(It follows that β takes values in {±1}.)
If we have to consider an alternating
bicharacter β that takes values in C× instead of R×, we will explicitly say that β
is C-valued. A quadratic form on T is a map µ : T → {±1} such that βµ is an
alternating bicharacter, where βµ : T × T → {±1}, called the polarization of µ, is
defined by
(2)
(The inverses above have no effect, but this way the equation is more similar to the
usual definition of quadratic forms on a vector space.)
βµ(u, v) := µ(uv)µ(u)−1µ(v)−1.
Definition 12. Let β be an alternating bicharacter on T , and consider its radical:
rad(β) := {t ∈ T β(u, t) = 1, ∀u ∈ T }. We say that β has type I if rad(β) = {e},
and that it has type II if rad(β) = {e, f } for some f ∈ T (of order 2). In the latter
case, as f is determined by β, we denote it by fβ.
We will say that a family {a1, b1, . . . , am, bm} in T is symplectic if β(ai, bi) =
β(bi, ai) = −1 (i = 1, . . . , m) and the value of β on all other pairs is +1. We will say
that it is a basis if T is the direct product of the subgroups ha1i, hb1i, . . . , hami, hbmi.
The following result [13, Proposition 9] describes alternating bicharacters satis-
fying Definition 12.
Proposition 13. Let β be an alternating bicharacter on a finite abelian group T .
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
7
(1) If β has type I, then T ∼= (Z2
2)m and there exists a symplectic basis {a1, b1,
. . . , am, bm} of T .
(2) If β has type II, then either T ∼= (Z2
2)m × Z2 or (Z2
2)m−1 × (Z2 × Z4), and
there exists a symplectic family {a1, b1, . . . , am, bm} such that:
(a) In the first case, {a1, b1, . . . , am, bm, fβ} is a basis of T .
(b) In the second case, b2
m = fβ and {a1, b1, . . . , am, bm} is a basis of T .
(cid:3)
Lemma 14. Let µ and η be two different quadratic forms on a finite abelian group
T such that βµ = βη. Then {t ∈ T µ(t) = η(t)} is a subgroup of T of index 2.
Proof. Let S := {t ∈ T µ(t) = η(t)}. By Equation (2), we have µ(e) = 1 = η(e),
so e ∈ S. Also, u, v ∈ S implies uv ∈ S, hence S is a subgroup. Since µ and η
take values in {±1}, u, v ∈ T \ S implies µ(u)µ(v) = η(u)η(v), hence uv ∈ S by
Equation (2). Thus the quotient group T /S can have only two elements.
(cid:3)
Lemma 15. Let β be an alternating bicharacter of type I on a finite abelian group
T . Then the following map is a bijection:
T −→ {S S is a subgroup of T, [T : S] ≤ 2}
u 7−→ u⊥ = {v ∈ T β(u, v) = 1}
Proof. It is enough to interpret β as a nondegenerate alternating bilinear form over
the field of two elements.
(cid:3)
Lemma 16. Let β be an alternating bicharacter of type II on a finite abelian group
T . Then the following map is a bijection:
T /hfβi −→ {S S is a subgroup of T, fβ ∈ S, [T : S] ≤ 2}
[u] 7−→ u⊥ = {v ∈ T β(u, v) = 1}
Proof. Consider the alternating bicharacter ¯β on T /hfβi such that β = ¯β ◦ (π × π),
where π : T → T /hfβi is the natural projection. Then ¯β has type I and we can
apply Lemma 15.
(cid:3)
Notation 17. For any natural number n and abelian group T , we define T[n] =
{t ∈ T tn = e} and T [n] = {tn t ∈ T }.
(We will primarily need the case n = 2.)
2
Notation 18. Let β be an alternating bicharacter of type II on T ∼= Z2m−1
× Z4.
Then T [2] has order 2 and we denote by fT its generator. By Proposition 13,
fT = fβ. We set rad′(β) := rad(βT[2] ×T[2]) \ rad(β) (which equals {am, amfT } with
the notation of Proposition 13).
Remark 19. If η is a quadratic form on T ∼= Z2m−1
× Z4 such that βη has type
II, then η(fT ) = +1. Indeed, if g is an element of T of order 4, then fT = g2, so
η(fT ) = η(g2) = η(g)2βη(g, g) = +1. Also note that η takes the same value on the
two elements of rad′(βη), because one is the other multiplied by fT . Finally, if β
is an alternating bicharacter of type II on T ∼= Z2m−1
× Z4, and µ is a quadratic
form defined only on T[2] such that βµ = βT[2]×T[2] and µ(fT ) = +1, then there
exist exactly two quadratic forms on T that extend µ and whose polarization is β.
2
2
Proposition 20. Let T be a finite abelian group, K a subgroup of T of index 2,
and ν : T \ K → {±1} a map. Consider the family of maps µg : K → {±1} defined
by µg(k) := ν(gk)ν(g)−1, as g runs through T \ K. Then, if a member of this
family is a quadratic form, so are the others, and all have the same polarization
β. Moreover, if β has type II, the value µg(fβ) does not depend on the choice of
g ∈ T \ K.
8
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
Proof. Let g, h ∈ T \ K, and assume that µg is a quadratic form. Call β its
polarization. The assertions follow from the following formula (k ∈ K):
µh(k) =
ν(hk)
ν(h)
=
µg(g−1hk)
µg(g−1h)
= µg(k)β(g−1h, k).
(3)
Indeed, as β is multiplicative in k, µh is also a quadratic form with the same
polarization as µg. And if β has type II, then µh(fβ) = µg(fβ)β(g−1h, fβ) =
µg(fβ).
(cid:3)
Definition 21. In the situation of Proposition 20, we say that ν is a nice map on
T \ K, and we denote by βν the common polarization β of the quadratic forms µg.
If β has type II, we also define ν(fβ) := µg(fβ), where g is any element of T \ K.
Lemma 22. Under the conditions of Proposition 20, suppose that T ∼= Z2m−2
K ∼= Z2m−3
element of T \ K of order 2, does not depend on the choice of g.
× Z4,
× Z4, and β = βν has type II. Then the set µg(rad′(β)), where g is any
2
2
Proof. Let g, h ∈ T \ K be elements of order 2 and let a ∈ rad′(β) ⊆ rad(βT[2]×T[2]).
By Equation (3), µh(a) = µg(a)β(g−1h, a) = µg(a).
(cid:3)
Notation 23. In the situation of Lemma 22, we define ν(rad′(β)) := µg(rad′(β)),
where g is any element of T \ K of order 2. Note that, by Remark 19 (applied to
K), µg takes the same value on the two elements of rad′(β), so the set ν(rad′(β))
actually consists of one element.
Notation 24 (Arf invariant). Let T be a finite set and let µ : T → {±1} be a map.
If the cardinality of µ−1(+1) is greater than the cardinality of µ−1(−1), we write
Arf(µ) = +1. If it is smaller, we write Arf(µ) = −1. Finally, if both cardinalities
are equal, Arf(µ) is not defined.
5. Division gradings and quadratic forms
As mentioned in the Introduction, the main purpose of this section is to classify
all quadratic forms whose polarization has type I or II. We establish a correspon-
dence with gradings in order to prove their existence.
The following division gradings will be our building blocks.
Example 25. Two division gradings by the group Z2:
C = R1 ⊕ Ri and R × R = R(1, 1) ⊕ R(1, −1).
Two division gradings by the group Z2
2:
0 1(cid:19) ⊕ R(cid:18)0
M2(R) = R(cid:18)1 0
1
1
0(cid:19) ⊕ R(cid:18)−1 0
1(cid:19) ⊕ R(cid:18)0 −1
0 (cid:19)
0
1
and
H = R1 ⊕ Ri ⊕ Rj ⊕ Rk.
And the three division gradings by the group Z2 × Z4 = {e, a; b, ab; b2, ab2; b3, ab3}
presented in Figure 1.
Let D be a finite-dimensional real (associative) algebra whose center Z(D) has
dimension 1 or 2. Let G be an abelian group and let Γ be a division G-grading on
D with support T and homogeneous components of dimension 1. Note that D must
be unital, but we do not assume that it is a simple algebra. By a generalization
of Maschke's Theorem (see for example [11, Corollary 10.2.5 on p. 443]), D is
necessarily semisimple, which implies that it is simple if Z(D) is R or C, and the
direct product of two simple algebras if Z(D) is R × R.
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
9
M2(C) =R(cid:18)1
0
0
1(cid:19) ⊕ R(cid:18)0 1
1 0(cid:19) ⊕
0
−eπi/4(cid:19) ⊕ R(cid:18) 0
eπi/4
−eπi/4
0 (cid:19) ⊕
i(cid:19) ⊕ R(cid:18)0
i
i
0(cid:19) ⊕
0
0
0
0
(1, 1)
(0, 0)
R(cid:18)eπi/4
R(cid:18)i
R(cid:18)e3πi/4
M2(R × R) =R(cid:18)(1, 1)
R(cid:18)(0, 0)
R(cid:18)(1, −1)
R(cid:18) (0, 0)
M2(R) × H =R(cid:18)(cid:18)1
R(cid:18)(cid:18)1
R(cid:18)(cid:18)1
R(cid:18)(cid:18)1
(1, −1)
(0, 0)
0
0
0
(0, 0)
(0, 0)
(1, −1)
−e3πi/4(cid:19) ⊕ R(cid:18) 0
e3πi/4
(1, 1)(cid:19) ⊕ R(cid:18)(1, 1)
(0, 0) (cid:19) ⊕ R(cid:18) (0, 0)
(1, −1)(cid:19) ⊕ R(cid:18)(1, −1)
(0, 0)(cid:19) ⊕ R(cid:18) (0, 0)
1(cid:19) , 1(cid:19) ⊕ R(cid:18)(cid:18)0 −1
(−1, −1)
(−1, 1)
(0, 0)
(1, 1)
(0, 0)
0
1
0
0 −1(cid:19) , j(cid:19) ⊕ R(cid:18)(cid:18)0
1
0
1(cid:19) , −1(cid:19) ⊕ R(cid:18)(cid:18)0 −1
1
0
0 −1(cid:19) , −j(cid:19) ⊕ R(cid:18)(cid:18)0
1
−e3πi/4
0 (cid:19)
(−1, −1)(cid:19) ⊕
(0, 0)
(1, 1)
(0, 0)
(1, −1)
(0, 0) (cid:19) ⊕
(−1, 1)(cid:19) ⊕
(0, 0)(cid:19)
0 (cid:19) , i(cid:19) ⊕
0(cid:19) , k(cid:19) ⊕
0 (cid:19) , −i(cid:19) ⊕
0(cid:19) , −k(cid:19)
1
1
Figure 1. Division gradings of Example 25; degrees in the group
hai × hbi ∼= Z2 × Z4 are assigned line-by-line in the following order:
e, a; b, ab; b2, ab2; b3, ab3.
We claim that the graded algebra D is equivalent to one, and only one, tensor
product on the list below, equipped with the product grading where each factor is
graded as in Example 25. The isomorphism classes are in bijective correspondence
with the triples (T, β, µ), where β is an alternating bicharacter on T of type I or II,
and µ is a quadratic form on T[2] such that βµ = βT[2]×T[2]. Namely, β : T × T →
{±1} is defined by the commutation relations of homogeneous elements,
(4)
for all Xu ∈ Du and Xv ∈ Dv, and µ : T[2] → {±1} is defined by the signs of the
squares of homogeneous elements,
XuXv = β(u, v)XvXu
X 2
t = µ(t)I
(5)
for all Xt ∈ Dt such that X 4
t = I, where I is the unity of D and t ∈ T[2]. Conversely,
given such (T, β, µ), we can construct D as the algebra with generators Xu, with u
ranging over a basis of T , and defining relations given by Equations (4) for all basis
elements u and v, Equations (5) for all basis elements t of order 2, and X 4
t = µ(fT )I
for the basis element t of order 4 (if it is present). The grading on D is defined
by declaring the generator Xu to be of degree u. Now we give the list of the
10
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
equivalence classes, and we also compile the classification up to isomorphism to
serve as a reference:
2
2
2
(1-a) Mn(R) ∼= M2(R) ⊗ . . .⊗ M2(R), n = 2m ≥ 1 (if n = 1, Mn(R) = R with the
trivial grading). The grading Γ is determined up to isomorphism by (T, µ), where
T is a subgroup of G isomorphic to Z2m
2 , and µ is a quadratic form on T such that
βµ has type I and Arf(µ) = +1.
(1-b) Mn/2(H) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ H, n = 2m ≥ 2. The grading Γ is
determined up to isomorphism by (T, µ), where T is a subgroup of G isomorphic
to Z2m
2 , and µ is a quadratic form on T such that βµ has type I and Arf(µ) = −1.
(1-c) Mn(C) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ C, n = 2m ≥ 1. The grading Γ is
determined up to isomorphism by (T, µ), where T is a subgroup of G isomorphic
to Z2m+1
, and µ is a quadratic form on T such that β := βµ has type II and
µ(fβ) = −1.
(1-d) Mn(C) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ M2(C), n = 2m ≥ 2. The grading Γ is
determined up to isomorphism by (T, β, µ), where T is a subgroup of G isomorphic
to Z2m−1
× Z4, β is an alternating bicharacter on T of type II, and µ is a quadratic
form on T[2] such that βµ = βT[2]×T[2] and µ(fT ) = −1.
(1-e) Mn(R) × Mn(R) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ [R × R], n = 2m ≥ 1. The
grading Γ is determined up to isomorphism by (T, µ), where T is a subgroup of G
isomorphic to Z2m+1
, and µ is a quadratic form on T such that β := βµ has type
II, µ(fβ) = +1 and Arf(µ) = +1.
(1-f) Mn/2(H) × Mn/2(H) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ H ⊗ [R × R], n = 2m ≥ 2.
The grading Γ is determined up to isomorphism by (T, µ), where T is a subgroup
of G isomorphic to Z2m+1
, and µ is a quadratic form on T such that β := βµ has
type II, µ(fβ) = +1 and Arf(µ) = −1.
(1-g) Mn(R) × Mn(R) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ M2(R × R), n = 2m ≥ 2. The
grading Γ is determined up to isomorphism by (T, β, µ), where T is a subgroup of G
isomorphic to Z2m−1
× Z4, β is an alternating bicharacter on T of type II, and µ is
a quadratic form on T[2] such that βµ = βT[2] ×T[2], µ(fT ) = +1, µ(rad′(β)) = {+1}
and Arf(µ) = +1.
(1-h) Mn/2(H) × Mn/2(H) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ H ⊗ M2(R × R), n = 2m ≥ 4.
The grading Γ is determined up to isomorphism by (T, β, µ), where T is a subgroup
of G isomorphic to Z2m−1
× Z4, β is an alternating bicharacter on T of type II, and
µ is a quadratic form on T[2] such that βµ = βT[2]×T[2], µ(fT ) = +1, µ(rad′(β)) =
{+1} and Arf(µ) = −1.
(1-i) Mn(R) × Mn/2(H) ∼= M2(R) ⊗ . . . ⊗ M2(R) ⊗ [M2(R) × H], n = 2m ≥ 2.
The grading Γ is determined up to isomorphism by (T, β, µ), where T is a subgroup
of G isomorphic to Z2m−1
× Z4, β is an alternating bicharacter on T of type II,
and µ is a quadratic form on T[2] such that βµ = βT[2]×T[2], µ(fT ) = +1 and
µ(rad′(β)) = {−1}.
2
2
2
2
Proof. Denote by I the unity of D. If Z(D) has dimension 2, denote Z = iI in
the case Z(D) = C (where i is the imaginary unit) and Z = (1, −1)I in the case
Z(D) = R × R. Then Z(D) is either RI or RI ⊕ RZ, and it is easy to show that
Z is homogeneous and its degree f is an element of order 2 (see for example [13,
Lemma 14]).
Now the classification is obtained by repeating the arguments of [13, Theorems
15 and 16]. Remark that fβ = f and that now the case µ(fβ) = +1 is possible.
As an example, let us recall the construction of a graded division algebra D for a
given datum (T, µ) in the case (1-c), that is, T ∼= Z2m+1
, βµ of type II and µ(fβ) =
−1. We pick a basis {a1, b1, . . . , am, bm, fβ} of T as in Proposition 13. Changing
if necessary aj or bj to ajfβ or bjfβ, respectively, we can get µ(aj ) = µ(bj) = +1
2
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
11
(j = 1, . . . , m). Hence we can choose as the graded algebra M2(R)⊗. . .⊗M2(R)⊗C,
as it is stated in the list, where the support of the j-th factor is haj, bji (j = 1, . . . , m)
and the support of the last factor is hfβi.
The only new argument is the fact that M2(R)⊗[M2(R)×H] and H⊗[M2(R)×H]
are in the same equivalence class, (1-i).
Indeed, consider the symplectic basis
{a1, b1, a2, b2} of Z2
2 × (Z2 × Z4) relative to the grading of H ⊗ [M2(R) × H], that is,
a1 and b1 generate the support of the grading on H, while a2 and b2 generate the
support of the grading on M2(R) × H playing the roles of a and b in Example 25, so
b2
2 6= e. The quadratic form is determined by µ(a1) = µ(b1) = −1 and µ(a2) = −1,
µ(b2
2 = a1b1b2 form another
2)2) = +1.
symplectic basis, but now µ(a′
Therefore, we can rewrite H ⊗ [M2(R) × H] as M2(R) ⊗ [M2(R) × H] by renaming
the elements of the group, so these graded algebras are equivalent.
(cid:3)
1) = +1 and still µ(a′
2) = +1. Then a′
2) = −1, µ((b′
1) = µ(b′
1 = a1a2, b′
1 = b1a2, a′
2 = a2, b′
Remark 26. If Z(D) is R× R then it must be nontrivially graded, so it is isomorphic
to the group algebra of a subgroup of G of order 2, namely, {e, f } where f = fβ.
Hence, D with its G-grading can be obtained from a simple algebra with a grading
by the quotient group G/hf i by means of the loop construction (see [1]): (1-e) and
(1-g) from (1-a); (1-f) and (1-h) from (1-b); and (1-i) from either (1-a) or (1-b).
6. Classification in the one-dimensional case
Example 27. We define involutions on M2(R), H, C and M2(C) that respect the
gradings of Example 25. The notation with subscripts will make sense later on,
when the classification of this section is stated.
• Let ϕ(1-a-1) be the matrix transpose on M2(R). It is an orthogonal involu-
• Let ϕ(1-a-2) be the involution on M2(R) given by ϕ(1-a-2)(X) = A−1X T A,
tion with signature 2.
where
A = (cid:18)1
0 −1(cid:19) .
0
(6)
It is an orthogonal involution with signature 0.
• Let ϕ(1-a-3) be the involution on M2(R) that acts as minus the identity on
the matrices of trace zero, and acts as the identity on the center of M2(R).
It is a symplectic involution.
• Let ϕ(1-b-1) be the standard conjugation on H. It is a symplectic involution
with signature 1.
• Let ϕ(1-b-3) be the involution on H that acts as the identity on 1, i and j,
and acts as minus the identity on k. It is an orthogonal involution.
• Let ϕ(1-c-1) be the conjugation on C. It is an involution of the second kind
and has signature 1.
• Let ϕ(1-c-3) be the identity on C. It is an involution of the first kind and
orthogonal.
• Let ϕ(1-d-1) be the matrix transpose on M2(C). It is an involution of the
first kind and orthogonal.
• Let ϕ(1-d-3) be the involution on M2(C) given by ϕ(1-a-2)(X) = A−1X T A,
with A as in Equation (6). It is an involution of the first kind and orthog-
onal.
• Let ϕ(1-d-4) be the involution on M2(C) that acts as minus the identity on
the matrices of trace zero, and acts as the identity on the center of M2(C).
It is an involution of the first kind and symplectic.
The involution ϕ(1-a-1) will occur most frequently, so we will abbreviate it as ϕ∗.
12
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
Let G be an abelian group, D a finite-dimensional simple real (associative) alge-
bra, and Γ a division G-grading on D with homogeneous components of dimension
1. These gradings are classified in [13, Theorems 15 and 16]; there are four families
of equivalence classes: (1-a), (1-b), (1-c) and (1-d). We keep the same notation, so
let T be the support of Γ, and let β : T × T → {±1} be the alternating bicharacter
given by the commutation relations.
Then, any antiautomorphism ϕ of the G-graded algebra D is an involution. We
want to classify the pairs (Γ, ϕ), up to isomorphism and up to equivalence. The
isomorphism classes are in bijective correspondence with the quadratic forms η on
T such that βη = β. Namely, the correspondence is given by the equation
(7)
for all Xt ∈ Dt. Now we give a list of the equivalence classes together with a rep-
resentative of every class, and we also compile the classification up to isomorphism
to serve as a reference:
ϕ(Xt) = η(t)Xt
(1-a) The grading Γ on D ∼= Mn(R) (n = 2m ≥ 1) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m
2 , and µ
was a quadratic form on T such that β := βµ had type I and Arf(µ) = +1. Now
(Γ, ϕ) is determined up to isomorphism by (T, µ, η), where η is a quadratic form
on T such that βη = β. These isomorphism classes belong to one of the following
three equivalence classes:
(1) η = µ (n = 2m ≥ 1).
The involution ϕ is orthogonal with signature n.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ (if n = 1, ϕ is just the identity on R).
(2) Arf(η) = +1 but η 6= µ (n = 2m ≥ 2).
The involution ϕ is orthogonal with signature 0.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(3) Arf(η) = −1 (n = 2m ≥ 2).
The involution ϕ is symplectic.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
(1-b) The grading Γ on D ∼= Mn/2(H) (n = 2m ≥ 2) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m
2 , and µ
was a quadratic form on T such that β := βµ had type I and Arf(µ) = −1. Now
(Γ, ϕ) is determined up to isomorphism by (T, µ, η), where η is a quadratic form
on T such that βη = β. These isomorphism classes belong to one of the following
three equivalence classes:
(1) η = µ (n = 2m ≥ 2).
The involution ϕ is symplectic with signature n/2.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-1).
(2) Arf(η) = −1 but η 6= µ (n = 2m ≥ 4).
The involution ϕ is symplectic with signature 0.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-b-1).
(3) Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is orthogonal.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-3).
(1-c) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 1) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m+1
, and µ
was a quadratic form on T such that β := βµ had type II and µ(fβ) = −1. Now
(Γ, ϕ) is determined up to isomorphism by (T, µ, η), where η is a quadratic form on
T such that βη = β. These isomorphism classes belong to one of the following four
equivalence classes:
2
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
13
(1) η = µ (n = 2m ≥ 1).
The involution ϕ is of the second kind and has signature n.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(2) η(fβ) = −1 but η 6= µ (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-c-1).
(3) η(fβ) = +1 and Arf(η) = +1 (n = 2m ≥ 1).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(4) η(fβ) = +1 and Arf(η) = −1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-c-3).
(1-d) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 2) was determined up to
isomorphism by (T, β, µ), where T was a subgroup of G isomorphic to Z2m−1
× Z4,
β was an alternating bicharacter on T of type II, and µ was a quadratic form on
T[2] such that βµ = βT[2]×T[2] and µ(fT ) = −1. Now (Γ, ϕ) is determined up to
isomorphism by (T, β, µ, η), where η is a quadratic form on T such that βη = β
(so η(fT ) = +1 by Remark 19). These isomorphism classes belong to one of the
following four equivalence classes:
2
(1) η(rad′(β)) = {+1} and Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-1).
(2) η(rad′(β)) = {+1} and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-d-1).
(3) η(rad′(β)) = {−1} and Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-3).
(4) η(rad′(β)) = {−1} and Arf(η) = −1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-4).
Proof. Define η : T → R× by Equation (7). For all t ∈ T , we can pick Xt so
that X 8
t = +I, hence η(t)8 = +1; therefore η takes values in {±1} and ϕ is an
involution. The fact that ϕ reverses the order of the product is equivalent to η
being a quadratic form with βη = β; indeed:
η(uv)XvXu = ϕ(XvXu) = ϕ(Xu)ϕ(Xv) = η(u)η(v)β(u, v)XvXu.
Thus we have a bijective correspondence between the isomorphism classes of pairs
(Γ, ϕ) and the quadratic forms η on T such that βη = β.
Involutions belonging to (1-a-1) or (1-a-2) are not equivalent to those in (1-a-3),
because of the Arf invariant. The involution (1-a-1) is determined by Equation (5),
so it is not equivalent to the involutions that belong to (1-a-2), in other words, it
is a distinguished involution of the grading. Considering also η(fβ) and η(rad′(β)),
we see that the rest of the equivalence classes of the list do not overlap.
We know that there exist quadratic forms η for the indicated values of n because
of Section 5.
The tricky point is to prove that involutions that lie in the same item of the
list are equivalent. The idea is to write any ϕ in a given equivalence class as the
representative that we indicated in the list. Let us start with the case (1-a-2),
so assume that T ∼= Z2m
2 , β has type I, Arf(µ) = Arf(η) = +1, and µ 6= η. By
1 = {t ∈ T µ(t) = η(t)}. We
Lemmas 14 and 15, there exists b1 ∈ T such that b⊥
14
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
are going to prove that µ(b1) = +1. Take c ∈ T \ b⊥
1 . Because of
Equation (2), µ takes the value −1 either once or three times on hc, b1i. Since η
has the same Arf invariant as µ and T ∼= hc, b1i × hc, b1i⊥, η takes the value −1 on
hc, b1i as many times as µ. This number cannot be three, because µ(c) 6= η(c), so
µ(b1) = η(b1) = +1.
1 , so T ∼= hci × b⊥
We can take a1 ∈ T , and then inductively a2, b2, . . . , am, bm ∈ T so that {a1,
b1, . . . , am, bm} is a symplectic basis as defined before Proposition 13 (follow, for
example, the arguments in [9, Equation (2.6) on p. 36]). Moreover, since Arf(µ) =
+1 and µ(b1) = +1, we can argue as in the last paragraph of [13, proof of Theorems
15 and 16] and assume that our symplectic basis satisfies
µ(a1) = µ(b1) = . . . = µ(am) = µ(bm) = +1.
By construction, this implies
η(a1) = −1 and η(b1) = . . . = η(am) = η(bm) = +1.
We have shown that any ϕ in (1-a-2) can be written as ϕ(1-a-2) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗, thus
they are all equivalent.
The same reasoning works for (1-a-3), but note that now µ(b1) = η(b1) = −1.
Analogously for (1-b-2), (1-b-3) and (1-c-2), but in this last case we use Lemma 16
instead of Lemma 15, and we may replace b1 by b1fβ so that µ(b1) = +1. In the
cases (1-c-3), (1-c-4), (1-d-1), (1-d-2), (1-d-3) and (1-d-4), we cannot apply Lemma
16, but in fact they are easier, because µ(fβ) = −1 whereas η(fβ) = +1. We can
first pick a1, b1, . . . , am, bm ∈ T so that η takes the values that we want on them.
Then, changing, if necessary, the ai and bj to aifβ and bjfβ, we can also select
the values taken by µ. For example, in the case (1-c-3), η is a quadratic form on
T ∼= Z2m+1
such that βη has type II, η(fβ) = +1 and Arf(η) = +1. This means
that η belongs to the item (1-e) of the list of Section 5, hence there exists a basis
{a1, b1, . . . , am, bm, fβ} of T as in Proposition 13 such that η(a1) = η(b1) = · · · =
η(am) = η(bm) = 1 (and η(fβ) = 1). We can assume without loss of generality that
also µ(a1) = µ(b1) = · · · = µ(am) = µ(bm) = 1 (and µ(fβ) = −1). Therefore any ϕ
can be written as ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
Finally, in order to compute the signatures it is enough to apply Lemma 1 to the
representative of every equivalence class, since we already calculated the signature
of each factor in Example 27.
(cid:3)
2
Remark 28. Recall from Lemma 10 that, given an involution ϕ on the graded
algebra D, we can obtain the rest of the involutions (of the same kind, if D ∼=
Mn(C)) as Int(Xu) ◦ ϕ, where u runs through T . If η is the quadratic form on
T corresponding to ϕ, then the quadratic form ηu : T → {±1} corresponding to
Int(Xu) ◦ ϕ is given by ηu(v) = β(u, v)η(v) = η(uv)η(u). In particular, Arf(ηu) =
Arf(η)η(u) if the Arf invariant is defined.
7. Classification in the two-dimensional non-complex case
Example 29. Consider the division grading by the group Z2 on the algebra M2(R)
obtained by coarsening of the grading of Example 25:
M2(R) = (cid:20)R(cid:18)1 0
0 1(cid:19) ⊕ R(cid:18)0 −1
0 (cid:19)(cid:21) ⊕(cid:20)R(cid:18)0 1
1 0(cid:19) ⊕ R(cid:18)−1 0
1
1(cid:19)(cid:21) .
0
When M2(R) is endowed with this grading, we denote the involutions of Example
27 as:
• ϕ(2-a-1) = ϕ(1-a-2);
• ϕ(2-a-3) = ϕ(1-a-1);
• ϕ(2-a-5) = ϕ(1-a-3).
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
15
Analogously, if H is Z2-graded as H = [R1 ⊕ Ri] ⊕ [Rj ⊕ Rk], we denote:
• ϕ(2-b-2) = ϕ(1-b-3);
• ϕ(2-b-3) = ϕ(1-b-1);
• we also denote ϕ(2-b-5) the involution on H that acts as the identity on 1, j
and k, and acts as minus the identity on i; it is an orthogonal involution.
Finally, consider the division grading by the group Z4 on the algebra M2(C) ob-
tained by the coarsening of the grading of Figure 1 that joins, for all t ∈ hai × hbi ∼=
Z2 × Z4, the homogeneous component of degree t to the homogeneous component of
degree ab2t. When M2(C) is endowed with this grading, we denote the involutions
of Example 27 as:
• ϕ(2-e-1) = ϕ(1-d-1);
• ϕ(2-e-3) = ϕ(1-d-3);
• ϕ(2-e-4) = ϕ(1-d-4).
Let G be an abelian group, D a finite-dimensional simple real (associative) alge-
bra, and Γ a division G-grading on D with homogeneous components of dimension
2 such that the identity component does not coincide with the center of D. These
gradings are classified in [13, Theorems 22 and 23]; there are five families of equiv-
alence classes: (2-a), (2-b), (2-c), (2-d) and (2-e). We keep the same notation, so
write De = RI ⊕ RJ (∼= C), where I is the unity of D and J 2 = −I; and let T be
the support of Γ, K the support of the centralizer of the identity component, and
β : K × K → {±1} the alternating bicharacter given by the commutation relations
in the centralizer of the identity component.
Then, for any antiautomorphism ϕ of the G-graded algebra D, either ϕ(J) = +J
or ϕ(J) = −J. We want to classify the pairs (Γ, ϕ), up to isomorphism and up to
equivalence, when ϕ is an involution. In the case ϕ(J) = +J, any antiautomorphism
is an involution, and there is exactly one proper refinement of Γ compatible with a
given involution; the isomorphism classes are in bijective correspondence with the
quadratic forms η on K such that βη = β (and, in the case (2-e), η(fT ) = +1) by
means of the equation:
ϕ(Xk) = η(k)Xk
(8)
for all k ∈ K and Xk ∈ Dk. In the case ϕ(J) = −J, there are antiautomorphisms
that are not involutions, but any proper refinement of Γ is compatible with a given
involution; the isomorphism classes of involutions are in bijective correspondence
with the nice maps ω on T \K such that βω = β (and, in the case (2-e), ω(fT ) = +1)
by means of the equation:
(9)
for all t ∈ T \ K and Xt ∈ Dt. Now we give a list of the equivalence classes together
with a representative of every class, and we also compile the classification up to
isomorphism to serve as a reference:
ϕ(Xt) = ω(t)Xt
(2-a) The grading Γ on D ∼= Mn(R) (n = 2m ≥ 2) was determined up to
isomorphism by (T, K, ν), where T was a subgroup of G isomorphic to Z2m−1
, K
was a subgroup of T of index 2, and ν was a nice map on T \ K such that β := βν
had type I and Arf(ν) = +1. Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined up
to isomorphism by (T, K, ν, η), where η is a quadratic form on K such that βη = β.
These isomorphism classes belong to one of the following two equivalence classes:
2
(1) Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is orthogonal with signature 0.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(2) Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is symplectic.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
16
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up to isomorphism
by (T, K, ν, ω), where ω is a nice map on T \K such that βω = β. These isomorphism
classes belong to one of the following four equivalence classes:
(3) ω = ν (n = 2m ≥ 2).
The involution ϕ is orthogonal with signature n.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(4) Arf(ω) = +1 but ω 6= ν (n = 2m ≥ 4).
The involution ϕ is orthogonal with signature 0.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(5) ω = −ν (n = 2m ≥ 2).
The involution ϕ is symplectic.
A representative is ϕ(2-a-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(6) Arf(ω) = −1 but ω 6= −ν (n = 2m ≥ 4).
The involution ϕ is symplectic.
A representative is ϕ(2-a-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(2-b) The grading Γ on D ∼= Mn/2(H) (n = 2m ≥ 2) was determined up to
isomorphism by (T, K, ν), where T was a subgroup of G isomorphic to Z2m−1
, K
was a subgroup of T of index 2, and ν was a nice map on T \ K such that β := βν
had type I and Arf(ν) = −1. Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined up
to isomorphism by (T, K, ν, η), where η is a quadratic form on K such that βη = β.
These isomorphism classes belong to one of the following two equivalence classes:
2
(1) Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is symplectic with signature 0.
A representative is ϕ(2-b-2) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
(2) Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is orthogonal.
A representative is ϕ(2-b-2) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up to isomorphism
by (T, K, ν, ω), where ω is a nice map on T \K such that βω = β. These isomorphism
classes belong to one of the following four equivalence classes:
(3) ω = ν (n = 2m ≥ 2).
The involution ϕ is symplectic with signature n/2.
A representative is ϕ(2-b-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(4) Arf(ω) = −1 but ω 6= ν (n = 2m ≥ 4).
The involution ϕ is symplectic with signature 0.
A representative is ϕ(2-b-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(5) ω = −ν (n = 2m ≥ 2).
The involution ϕ is orthogonal.
A representative is ϕ(2-b-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(6) Arf(ω) = +1 but ω 6= −ν (n = 2m ≥ 4).
The involution ϕ is orthogonal.
A representative is ϕ(2-b-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(2-c) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 2) was determined up to
isomorphism by (T, K, ν), where T was a subgroup of G isomorphic to Z2m
2 , K was
a subgroup of T of index 2, and ν was a nice map on T \ K such that β := βν had
type II and ν(fβ) = −1. Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined up to
isomorphism by (T, K, ν, η), where η is a quadratic form on K such that βη = β.
These isomorphism classes belong to one of the following three equivalence classes:
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
17
(1) η(fβ) = −1 (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(2) η(fβ) = +1 and Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(3) η(fβ) = +1 and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-c-3).
On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up to isomorphism
by (T, K, ν, ω), where ω is a nice map on T \K such that βω = β. These isomorphism
classes belong to one of the following five equivalence classes:
(4) ω = ν (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature n.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(5) ω = −ν (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(2-a-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(6) ω(fβ) = −1 but ω 6= ±ν (n = 2m ≥ 4).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-c-1).
(7) ω(fβ) = +1 and Arf(ω) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(8) ω(fβ) = +1 and Arf(ω) = −1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-5) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(2-d) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 4) was determined up to
isomorphism by (T, K, β, ν), where T was a subgroup of G isomorphic to Z2m−2
×Z4,
K was a subgroup of T of index 2 but different from T[2], β was an alternating
bicharacter on K of type II, and ν was a nice map on T[2] \ K[2] such that βν =
βK[2]×K[2] and ν(fT ) = −1. Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined
up to isomorphism by (T, K, β, ν, η), where η is a quadratic form on K such that
βη = β (so η(fT ) = +1). These isomorphism classes belong to one of the following
four equivalence classes:
2
(1) η(rad′(β)) = {+1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-1).
(2) η(rad′(β)) = {+1} and Arf(η) = −1 (n = 2m ≥ 8).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-d-1).
(3) η(rad′(β)) = {−1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-3).
(4) η(rad′(β)) = {−1} and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-4).
On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up to isomorphism
by (T, K, β, ν, ω), where ω is a nice map on T \K such that βω = β (so ω(fT ) = +1).
These isomorphism classes belong to one of the following four equivalence classes:
18
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
(5) ω(rad′(β)) = {+1} and Arf(ω) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-1).
(6) ω(rad′(β)) = {+1} and Arf(ω) = −1 (n = 2m ≥ 8).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-d-1).
(7) ω(rad′(β)) = {−1} and Arf(ω) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-3).
(8) ω(rad′(β)) = {−1} and Arf(ω) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-a-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-4).
(2-e) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 2) was determined up to
isomorphism by (T, [ν]), where T was a subgroup of G isomorphic to Z2m−2
× Z4
(K = T[2]), and [ν] was an equivalence class of nice maps ν on T \ T[2] such that
β := βν had type II, fβ = fT and ν(fT ) = −1, with the equivalence relation ν ∼ ν ′
if either ν ′ = ν or ν ′ = −ν. Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined up
to isomorphism by (T, [ν], η), where η is a quadratic form on T[2] such that βη = β
and η(fT ) = +1. These isomorphism classes belong to one of the following two
equivalence classes:
2
(1) Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-e-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(2) Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-e-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up to isomorphism
by (T, [ν], ω), where ω is a nice map on T \ T[2] such that βω = β and ω(fT ) = +1.
These isomorphism classes belong to one of the following two equivalence classes:
(3) Arf(ω) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-e-3) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(4) Arf(ω) = −1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-e-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
Proof. Let g ∈ T \ K. We start with the case (2-a). We know from [13, Theorem
22] that we can write D as follows:
(De ⊕ Dg) ⊗R CD(De ⊕ Dg).
(10)
Recall that CD(De ⊕ Dg) is a subalgebra isomorphic to Mn/2(R) or Mn/4(H), en-
dowed with a division grading whose homogeneous components have dimension 1.
Since ϕ(De ⊕ Dg) = De ⊕ Dg, also ϕ(CD(De ⊕ Dg)) = CD(De ⊕ Dg). Therefore we
have reduced the problem to the study of antiautomorphisms on De ⊕ Dg, which is
isomorphic either to M2(R) if ν(g) = +1, or to H if ν(g) = −1.
If ϕ(J) = +J, then ϕ is De-semilinear on Dg, hence there exists X ∈ Dg such
that ϕ(X) = X (and ϕ(JX) = −JX). Therefore, ϕ is an involution and it is
only compatible with the proper refinement that splits Dg as RX ⊕ RJX. It is
straightforward to check the assertions about the isomorphisms classes. Let us see
that involutions that lie in the same item of the list are equivalent. If n ≥ 4, we can
always choose g such that ν(g) = +1 and the quadratic form µg(k) := ν(gk)ν(g)−1
is different from η, so we can write any involution in (2-a-1) (respectively (2-a-2))
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
19
as the tensor product of an involution on M2(R) and an involution on Mn/2(R)
that lies in (1-a-2) (respectively (1-a-3)), hence they are all equivalent.
If ϕ(J) = −J, then ϕDg = λ idDg, where λ ∈ De. Therefore, ϕ is an involution
if and only if λ = ±1 and, in that case, every refinement is compatible with ϕ.
Again, we can always choose g such that ν(g) = +1 and ω(g) = +1 (respectively
ω(g) = −1), so we can write any involution in (2-a-4) (respectively (2-a-6)) as the
tensor product of an involution on M2(R) and an involution on Mn/2(R) that lies
in (1-a-2), hence they are all equivalent.
The same arguments work for (2-b) and (2-c), and also for the case (2-d), which
is, in fact, easier because there is no distinguished involution.
k ⊕ D−
k (respectively D−
Let us now consider the remaining case (2-e). Any proper refinement of the
grading has to split Dk, for all k ∈ T[2], as D+
k , where the squares of the
elements in D+
k ) are positive (respectively negative) multiples of
I. Also recall from [13, Remark 21] that, if X ∈ Dg, then there exists a proper
refinement of the grading such that the element X is still homogeneous; this implies
that Dgk splits as XD+
k ) = D+
k for all k ∈ T[2]. Assume that ϕ(J) =
+J. As before, ϕ is De-semilinear on Dg and there exists X ∈ Dg such that ϕ(X) =
X (and ϕ(JX) = −JX). This implies that ϕ(XD+
k ) = XD−
k
for all k ∈ T[2], that is, ϕ is an involution and there is exactly one proper refinement
compatible with ϕ. Assume that ϕ(J) = −J. Then ϕDg = λ idDg , where λ ∈ De,
thus ϕDgk = ±λ idDgk for all k ∈ T[2]. Therefore ϕ is an involution if and only if
λ = ±1, and, in that case, every refinement is compatible with ϕ.
k and ϕ(D−
k ) = D−
k ⊕ XD−
k for all k ∈ T[2].
k ) = XD+
k and ϕ(XD−
We have ϕ(D+
Now that we know that every involution is compatible with at least one proper
refinement, we can use this fact to prove the rest of the assertions of the theorem
(see Remark 19). Unlike in the previous cases, in (2-e), if ψ is any isomorphism or
equivalence between two refinements with supports hh1i × T1 and hh2i × T2, then
ψ will continue to be an isomorphism or equivalence with respect to the original
gradings, with supports T1 and T2. Indeed, ψ has to send h1 to h2, because they
are distinguished elements.
Finally, the computation of signature of ϕ can be done similarly to Section 6
or, alternatively, we can take a compatible refinement and see the signature of the
corresponding isomorphism class already on the list of Section 6.
(cid:3)
8. Classification in the two-dimensional complex case
Example 30. Let ε = e2πi/l ∈ C and consider the generalized Pauli Matrices
Xa, Xb ∈ Ml(C) of Figure 2. Note that
XaXb = εXbXa
and X l
a = X l
b = I.
Therefore, we can construct a division grading on Ml(C) by the group Zl × Zl if we
define the homogeneous component of degree (j, k) to be CX j
b . Let ϕA and ϕB
be the second kind antiautomorphisms on Ml(C) given by ϕA(X) = A−1X ∗A and
ϕB(X) = B−1X ∗B, where X ∗ := X T and the matrices A, B ∈ Ml(C) are those
of Figure 2. Since A∗ = A and B∗ = B, both ϕA and ϕB are involutions. The
signatures of ϕA and ϕB are, respectively, 2 and 0 if l is even, and 1 and 1 if l is
odd. Both involutions respect the grading because:
aX k
ϕA(Xa) = Xa, ϕA(Xb) = Xb;
ϕB(Xa) = εXa, ϕB(Xb) = Xb.
We will write:
• ϕl = ϕA,
• φl = ϕB.
20
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
Xa =
εl−1
0
0
...
0
0
εl−2
0
...
0
0
0
εl−3
...
0
· · ·
· · ·
· · ·
. . .
· · ·
0
0
0
...
1
Xb =
0 1
0 0
...
...
0 0
1 0
0 · · ·
1 · · ·
...
. . .
0 · · ·
0 · · ·
0
0
...
1
0
A =
0
...
0
1
0
0
...
1
0
0
· · ·
. . .
· · ·
· · ·
· · ·
1
...
0
0
0
0
...
0
0
1
B =
0
...
0
0
1
0 0
...
...
0 1
1 0
0 0
· · ·
. . .
· · ·
· · ·
· · ·
1
...
0
0
0
Figure 2. Matrices in Ml(C) of Example 30; ε = e2πi/l.
Example 31. We introduce two involutions on M2(C) that respect the division
grading induced by the Pauli matrices in the case l = 2:
• Let ϕ(2-f-1-1) be the matrix transpose on M2(C). It is an involution of the
first kind and orthogonal.
• Let ϕ(2-f-1-2) be the involution on M2(C) that acts as minus the identity on
the matrices of trace zero, and acts as the identity on the center of M2(C).
It is an involution of the first kind and symplectic.
Let G be an abelian group, D a real algebra isomorphic to Mn(C), and Γ a
division G-grading on D with homogeneous components of dimension 2 such that
the identity component coincides with the center of D. Γ can be regarded as a
grading of the complex algebra Mn(C), and these gradings are classified in [9,
Theorem 2.15]; there is one family of equivalence classes: (2-f). The isomorphism
and equivalence classes in this classification remain the same over R, because the
invariants that differentiate them, namely, the pair (T, β) and the isomorphism class
of T respectively, are also preserved by isomorphisms of real algebras. As always, T
is the support of Γ and β is the alternating bicharacter given by the commutation
relations, XuXv = β(u, v)XvXu (where 0 6= Xt ∈ Dt for all t ∈ T ), but, in contrast
with Section 6, β is now C-valued.
Any antiautomorphism ϕ of the G-graded algebra D is an involution, and satisfies
either ϕ(iI) = +iI or ϕ(iI) = −iI (where I is the unity of D). We classify the
pairs (Γ, ϕ), up to isomorphism and up to equivalence, and we give a representative
of every equivalence class:
(2-f) The grading Γ on D ∼= Mn(C) (n ≥ 1) was determined up to isomorphism
by (T, β), where T was a subgroup of G isomorphic to Z2
(l1 · · · lr = n),
l1
and β was a C-valued alternating bicharacter on T such that rad(β) = {e}. The
equivalence class of the grading Γ was determined by the isomorphism class of
the group T , which we fix henceforth. Now, in the case ϕ(iI) = +iI, we have
l1 = . . . = lr = 2 (so β takes values in {±1} ⊆ R×) and (Γ, ϕ) is determined up to
isomorphism by (T, β, η), where η is a quadratic form on T such that βη = β, and
η is defined by the equation
× . . . × Z2
lr
ϕ(Xt) = η(t)Xt
(11)
for all Xt ∈ Dt. These isomorphism classes belong to one of the following two
equivalence classes:
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
21
(1-1) Arf(η) = +1 (n = 2m ≥ 1).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(2-f-1-1) ⊗C · · · ⊗C ϕ(2-f-1-1) (if n = 1, we take ϕ = idC).
(1-2) Arf(η) = −1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(2-f-1-1) ⊗C · · · ⊗C ϕ(2-f-1-1) ⊗C ϕ(2-f-1-2).
On the other hand, in the case ϕ(iI) = −iI, (Γ, ϕ) is determined up to isomorphism
by (T, β, S), where S is a subgroup of T[2] of index 1 or 2, and S is defined as
S := {t ∈ T[2] ∃X ∈ Dt such that X 2 = +I and ϕ(X) = X}.
(12)
These isomorphism classes belong to one of the following equivalence classes:
(2-0) S = T[2] (n ≥ 1).
The involution ϕ is of the second kind and has signature pT[2].
A representative is ϕl1 ⊗C · · · ⊗C ϕlr .
(2-p) S 6= T[2], and any t ∈ T of order 2p+1 satisfies t2p
∈ T[2] \ S (p ≥ 1).
t ∈ T of order 2p such that t2p−1
The involution ϕ is of the second kind and has signature 0.
A representative is φl1 ⊗C · · · ⊗C φls ⊗C ϕls+1 ⊗C · · · ⊗C ϕlr , provided that
l1 l2 · · · lr and 2p+1 divides ls+1 but does not divide ls.
∈ S, but there exists
t
Proof. The case ϕ(iI) = +iI was proved in [9, Propositions 2.51 and 2.53]. So
assume that ϕ(iI) = −iI. For any element t ∈ T , denote its order by o(t) and
define:
D[+]
:= {X ∈ Dt X o(t) = +I} and D[−]
:= {X ∈ Dt X o(t) = −I}.
and ε ∈ C is a primitive o(t)-th root of unity, then D[+]
t = {X,
) =
; in particular, ϕ is an involution. We define the following subsets of T , which
Note that, if X ∈ D[+]
εX, . . . , εo(t)−1X}, and similarly for D[−]
D[−]
are invariants of the isomorphism class of (Γ, ϕ):
. Besides, ϕ(D[+]
and ϕ(D[−]
) = D[+]
(13)
t
t
t
t
t
t
t
t
S ′ := {t ∈ T ∃X ∈ D[+]
(14)
If o(t) is odd, then t ∈ S ′, while if o(t) is even, then t ∈ T \ S ′ if and only if there
[−]
exists X ∈ D
t
such that ϕ(X) = X} and S := S ′ ∩ T[2].
such that ϕ(X) = X.
Write T as U × V , where U is the subgroup of T formed by the elements whose
order is a power of 2, and V is the subgroup of T formed by the elements of odd
order. We know that V ⊆ S ′. Moreover, if u ∈ U and v ∈ V , then u ∈ S ′ if and
only if uv ∈ S ′, because β(u, v) = 1. Finally, if u ∈ U \ T[2], then u ∈ S ′ if and only
if u2 ∈ S ′. Therefore, S determines S ′.
The restriction of β to T[2] × T[2] takes values in {±1}. Hence, u, v ∈ S implies
uv ∈ S, and also u, v ∈ T[2] \ S implies uv ∈ S. Therefore, S is a subgroup of T[2]
of index 1 or 2.
T = ha1i × hb1i × . . . × hari × hbri,
We know, for example from [9, Equation (2.6)], that we can write T as follows:
(15)
where ai, bi ∈ T , haii × hbii ∼= Z2
, li is a power of a prime, β(ai, bi) = β(bi, ai)−1 =
li
e2πi/li, and the value of β on all other pairs is 1. We claim that (T, β, S) determines
(Γ, ϕ) up to isomorphism. We can pick Xai ∈ Dai such that ϕ(Xai ) = Xai, and
either Xai ∈ D[+]
if ai ∈ T \ S ′. We pick Xbi ∈ Dbi
in the same way. The elements Xai, Xbi generate D, with defining relations of
two kinds: the commutation relations are determined by β and the power relations
are determined by S through Equation (13). This proves the claim; in fact, the
isomorphism can be chosen to be an isomorphism of complex algebras. Conversely,
if ai ∈ S ′, or Xai ∈ D[−]
ai
ai
22
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
let us find an involution ϕ for a given subgroup S of T[2] of index 1 or 2. Thanks
to Lemma 2, it is enough to construct it for every factor haii × hbii of T , but we
have already done it in Example 30.
Let us see that, for a fixed p ≥ 1, all the involutions that lie in (2-f-2-p) are
equivalent.
In fact, we will show that a1, b1, . . . , ar, br in Equation (15) may be
chosen so that they also satisfy: ai, bi ∈ S ′ for all i, except in the case li = 2mi ≤ 2p,
when ai ∈ S ′ but bi ∈ T \ S ′. We can follow the same induction process as the
one leading to [9, Equation (2.6)], until we arrive to a situation in which T is a
2-group and there are elements in T of maximal order, 2p, that do not belong to
S ′. Rearranging, we may assume that l1 = 2m1 ≥ l2 = 2m2 ≥ . . . ≥ lr = 2mr. If
r = 1, the statement is clear, so suppose that r ≥ 2. Then we choose the next a, b
in the following way.
We want to take a, b ∈ T such that o(a) = o(b) = 2m1, β(a, b) = e2πi/l1, a ∈ S ′,
b ∈ T \S ′, and such that there are elements in ha, bi⊥ of maximal order, 2m2, that do
not belong to S ′, because then T = ha, bi × ha, bi⊥ and we will be able to continue
the induction process with ha, bi⊥. We know the existence of a decomposition
T = ha1i × hb1i × . . . × hari × hbri as in Equation (15), but we cannot assure that
ai ∈ S ′ and bi ∈ T \ S ′. Without loss of generality, a1, a2 ∈ S ′ and b1 ∈ T \ S ′,
hence a1a2 ∈ S ′. If b2 ∈ T \ S ′, simply take a = a1 and b = b1. If b2 ∈ S ′, take
b−1
a = a1a2 and b = b1, and note that bl1/l2
2 has order 2m2 and belongs both to
ha, bi⊥ and to T \ S ′.
1
Finally, the computation of signature is analogous to Section 6, but using Ex-
ample 30 instead of Example 25 and Lemma 2 instead of Lemma 1. For involutions
in (2-f-2-p), we pick up a zero factor. For involutions in (2-f-2-0), we may assume
that l1, . . . , ls are even and ls+1, . . . , lr are odd, then s is the number of factors 2,
(cid:3)
so the signature equals 2s = pT[2].
Remark 32. Consider an involution ϕ of the second kind on the graded algebra D.
By Lemma 10, all such involutions can be obtained as Int(Xu) ◦ ϕ where u runs
through T . Since β is nondegenerate, it is easy to see that u ∈ T [2] (recall Notation
17) if and only if β(u, v) = 1 for all v ∈ T[2]. Therefore, Int(Xu) ◦ ϕ and ϕ are in
the same isomorphism class if and only if u ∈ T [2], because (Int(Xu) ◦ ϕ)(Xv) =
β(u, v)ϕ(Xv). Now assume that ϕ lies in (2-f-2-0). We have just shown that
Int(Xu) ◦ ϕ lies in (2-f-2-0) if and only if u is a square in T . Now we claim that
Int(Xu) ◦ ϕ lies in (2-f-2-p) (p ≥ 1) if and only if uT[2p] is a square in T /T[2p]
but uT[2p−1] is not a square in T /T[2p−1]. Indeed, using the nondegeneracy of β
(or explicitly using its values on the pairs of generators in Equation (15)), it is
straightforward to show that uT[2p] is a square in T /T[2p] if and only if β(u, v2p
) = 1
for all v ∈ T[2p+1].
9. Classification in the four-dimensional case
Let G be an abelian group, D a finite-dimensional simple real (associative) alge-
bra, and Γ a division G-grading on D with homogeneous components of dimension
4. We can apply the Double Centralizer Theorem (see for example [10, Theorem
4.7]) to the identity component De, which is isomorphic to H, to conclude that D is
isomorphic, as a graded algebra, to De ⊗R CD(De) (see [13, Theorem 19] for more
details). Note that CD(De) is again a finite-dimensional simple real graded-division
algebra, but with homogeneous components of dimension 1. Any antiautomorphism
ϕ of the G-graded algebra D is the tensor product of its restrictions to De and to
CD(De).
The following result is well known and easily follows from Skolem -- Noether The-
orem.
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
23
Proposition 33. Any antiautomorphism ϕ of the real algebra H can be written as
ϕ(X) = A−1XA, for some A = a + bi + cj + dk ∈ H×. So ϕ is an involution if
and only if either b = c = d = 0 or a = 0. In the first case ϕ is symplectic with
signature 1 (ϕ is the conjugation), while in the second ϕ is orthogonal.
(cid:3)
For us, this means the following: if Γ is the trivial grading on H, then there are
exactly two isomorphism classes of pairs (Γ, ϕ), where ϕ is an involution, and they
coincide with the equivalence classes. When H is endowed with the trivial grading,
we denote:
• Let ϕ(3-b-1) be the conjugation on H, that is, ϕ(3-b-1) = ϕ(1-b-1).
• Let ϕ(3-b-4) be a representative of the orthogonal equivalence class, for
example, ϕ(3-b-4) = ϕ(1-b-3).
Now, the classification of pairs (Γ, ϕ), where Γ is a division grading on D as
above and ϕ is an involution, is easily obtained from Proposition 33 and Section
6. Therefore, the isomorphism classes are in two-to-one correspondence with the
quadratic forms η on T such that βη = β, where T is the support of Γ and β is the
alternating bicharacter on T given by the commutation relations in the centralizer
of the identity component. Each quadratic form η corresponds to two isomorphism
classes, one for every class of involutions on De ∼= H, by means of the equation:
ϕ(Xt) = η(t)Xt
(16)
for all Xt ∈ Dt ∩ CD(De). Note that, if we have to compute the signature of ϕ, we
take a compatible refinement and check the signature of the corresponding isomor-
phism class in the list of Section 6. We compile the classification, up to isomorphism
and up to equivalence, and we give a representative of every equivalence class, to
serve as a reference:
(3-a) The grading Γ on D ∼= Mn(R) (n = 2m ≥ 4) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m−2
, and µ
was a quadratic form on T such that βµ had type I and Arf(µ) = −1. Now, if ϕ is
the conjugation on De, (Γ, ϕ) is determined up to isomorphism by (T, µ, η), where
η is a quadratic form on T such that βη = βµ. These isomorphism classes belong
to one of the following three equivalence classes:
2
(1) η = µ (n = 2m ≥ 4).
The involution ϕ is orthogonal with signature n.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-1).
(2) Arf(η) = −1 but η 6= µ (n = 2m ≥ 8).
The involution ϕ is orthogonal with signature 0.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-b-1).
(3) Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is symplectic.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-3).
On the other hand, if ϕ is orthogonal on De, (Γ, ϕ) is determined up to isomorphism
by (T, µ, η), where again η is a quadratic form on T such that βη = βµ. These
isomorphism classes belong to one of the following three equivalence classes:
(4) η = µ (n = 2m ≥ 4).
The involution ϕ is symplectic.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-1).
(5) Arf(η) = −1 but η 6= µ (n = 2m ≥ 8).
The involution ϕ is symplectic.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-b-1).
24
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
(6) Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is orthogonal with signature 0.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-b-3).
(3-b) The grading Γ on D ∼= Mn/2(H) (n = 2m ≥ 2) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m−2
, and µ
was a quadratic form on T such that βµ had type I and Arf(µ) = +1. Now, if ϕ is
the conjugation on De, (Γ, ϕ) is determined up to isomorphism by (T, µ, η), where
η is a quadratic form on T such that βη = βµ. These isomorphism classes belong
to one of the following three equivalence classes:
2
(1) η = µ (n = 2m ≥ 2).
The involution ϕ is symplectic with signature n/2.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(2) Arf(η) = +1 but η 6= µ (n = 2m ≥ 4).
The involution ϕ is symplectic with signature 0.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(3) Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is orthogonal.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
On the other hand, if ϕ is orthogonal on De, (Γ, ϕ) is determined up to isomorphism
by (T, µ, η), where again η is a quadratic form on T such that βη = βµ. These
isomorphism classes belong to one of the following three equivalence classes:
(4) η = µ (n = 2m ≥ 2).
The involution ϕ is orthogonal.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗.
(5) Arf(η) = +1 but η 6= µ (n = 2m ≥ 4).
The involution ϕ is orthogonal.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2).
(6) Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is symplectic with signature 0.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3).
(3-c) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 2) was determined up to
isomorphism by (T, µ), where T was a subgroup of G isomorphic to Z2m−1
, and µ
was a quadratic form on T such that β := βµ had type II and µ(fβ) = −1. Now,
if ϕ is the conjugation on De, (Γ, ϕ) is determined up to isomorphism by (T, µ, η),
where η is a quadratic form on T such that βη = β. These isomorphism classes
belong to one of the following four equivalence classes:
2
(1) η = µ (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature n.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(2) η(fβ) = −1 but η 6= µ (n = 2m ≥ 4).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-c-1).
(3) η(fβ) = +1 and Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(4) η(fβ) = +1 and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-c-3).
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
25
On the other hand, if ϕ is orthogonal in De, (Γ, ϕ) is determined up to isomorphism
by (T, µ, η), where again η is a quadratic form on T such that βη = β. These
isomorphism classes belong to one of the following four equivalence classes:
(5) η = µ (n = 2m ≥ 2).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-1).
(6) η(fβ) = −1 but η 6= µ (n = 2m ≥ 4).
The involution ϕ is of the second kind and has signature 0.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-2) ⊗ ϕ(1-c-1).
(7) η(fβ) = +1 and Arf(η) = +1 (n = 2m ≥ 2).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-c-3).
(8) η(fβ) = +1 and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-c-3).
(3-d) The grading Γ on D ∼= Mn(C) (n = 2m ≥ 4) was determined up to
isomorphism by (T, β, µ), where T was a subgroup of G isomorphic to Z2m−3
× Z4,
β was an alternating bicharacter on T of type II, and µ was a quadratic form on
T[2] such that βµ = βT[2]×T[2] and µ(fT ) = −1. Now, if ϕ is the conjugation on De,
(Γ, ϕ) is determined up to isomorphism by (T, β, µ, η), where η is a quadratic form
on T such that βη = β (so η(fT ) = +1). These isomorphism classes belong to one
of the following four equivalence classes:
2
(1) η(rad′(β)) = {+1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-1).
(2) η(rad′(β)) = {+1} and Arf(η) = −1 (n = 2m ≥ 8).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-d-1).
(3) η(rad′(β)) = {−1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-3).
(4) η(rad′(β)) = {−1} and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-1) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-4).
On the other hand, if ϕ is orthogonal on De, (Γ, ϕ) is determined up to isomorphism
by (T, β, µ, η), where again η is a quadratic form on T such that βη = β (so η(fT ) =
+1). These isomorphism classes belong to one of the following four equivalence
classes:
(5) η(rad′(β)) = {+1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-1).
(6) η(rad′(β)) = {+1} and Arf(η) = −1 (n = 2m ≥ 8).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-a-3) ⊗ ϕ(1-d-1).
(7) η(rad′(β)) = {−1} and Arf(η) = +1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and orthogonal.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-3).
(8) η(rad′(β)) = {−1} and Arf(η) = −1 (n = 2m ≥ 4).
The involution ϕ is of the first kind and symplectic.
A representative is ϕ(3-b-4) ⊗ ϕ∗ ⊗ · · · ⊗ ϕ∗ ⊗ ϕ(1-d-4).
26
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
10. Semisimple algebras with involution
The results of the previous sections can be extended to finite-dimensional real
algebras that are not necessarily simple, but do not have nontrivial ideals preserved
by the involution. As mentioned in the Introduction, our purpose is to apply these
results for the classification of gradings on classical real Lie algebras in a forthcoming
article [3], so here we restrict ourselves to the situation relevant for that application.
Let G be an abelian group, D a finite-dimensional non-simple real (associative)
algebra whose center has dimension 2, and Γ a division G-grading on D. Recall
from Section 5 that this implies that D is the direct product of two central simple
algebras over R. Let ϕ be a second kind involution on the G-graded algebra D. We
want to classify the pairs (Γ, ϕ), up to isomorphism (but not up to equivalence).
In fact, we can repeat the arguments in [13] and in the previous sections, because
they do not depend on the simplicity of the underlying algebra.
Let us start by considering the grading Γ and disregarding the involution ϕ. As
always, we denote by T the support of Γ, by K the support of the centralizer of the
identity component, and by β : K × K → {±1} the alternating bicharacter given by
the commutation relations in the centralizer of the identity component. Also, if the
homogeneous components have dimension 2, we write De = RI ⊕ RJ (∼= C), where
I is the unity of D and J 2 = −I. By [13, Proposition 20], if the homogeneous
components have dimension 2 or 4, then there exists a proper refinement of the
grading.
The existence of a second kind involution ϕ prevents T from having a factor
Z4, in other words, T is an elementary abelian 2-group. Indeed, Remark 19 can be
invoked if the homogeneous components have dimension 1. As in Section 9, the case
of dimension 4 reduces to dimension 1 using the Double Centralizer Theorem (note
that [10, Theorem 4.7] does not require the ambient algebra to be simple). Finally,
in the case where the homogeneous components have dimension 2, if there existed
an element g ∈ T \ K of order 4, then, by [13, Remark 21], any 0 6= X, X ′ ∈ Dg
would satisfy X 2 ∈ Z(D) and (X ′)2 ∈ R>0X 2, so ϕ(X 2) = ϕ(X)2 ∈ R>0X 2 would
give us a contradiction with ϕ being of the second kind.
Looking at the list in Section 5, we see that D must be isomorphic to either
Mn(R) × Mn(R) (n = 2m ≥ 1) or Mn/2(H) × Mn/2(H) (n = 2m ≥ 2), both with
a grading whose support is an elementary 2-group of rank 2m + 1, 2m or 2m − 1,
according to the homogeneous components being of dimension 1, 2 or 4 respectively.
• If the homogeneous components have dimension 1, then (Γ, ϕ) is determined
up to isomorphism by (T, µ, η), where T is a subgroup of G isomorphic
to Z2m+1
, µ is a quadratic form on T such that β := βµ has type II
and µ(fβ) = +1, and η is a quadratic form on T such that βη = β and
η(fβ) = −1.
If Arf(µ) = +1, then D ∼= Mn(R) × Mn(R) (n = 2m ≥ 1), whereas if
Arf(µ) = −1, then D ∼= Mn/2(H) × Mn/2(H) (n = 2m ≥ 2).
2
• If the homogeneous components have dimension 4, then the classification
is again reduced to the case of dimension 1 (see Section 9).
• If the homogeneous components have dimension 2, then the grading Γ is
determined up to isomorphism by (T, K, ν), where T is a subgroup of G
isomorphic to Z2m
2 , K is a subgroup of T of index 2, and ν is a nice map
on T \ K such that β := βν has type II and ν(fβ) = +1.
If Arf(ν) = +1, then D ∼= Mn(R) × Mn(R) (n = 2m ≥ 2), whereas if
Arf(ν) = −1, then D ∼= Mn/2(H) × Mn/2(H) (n = 2m ≥ 2).
Now, in the case ϕ(J) = +J, (Γ, ϕ) is determined up to isomorphism by
(T, K, ν, η), where η is a quadratic form on K such that βη = β and η(fβ) =
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
27
−1. On the other hand, in the case ϕ(J) = −J, (Γ, ϕ) is determined up
to isomorphism by (T, K, ν, ω), where ω is a nice map on T \ K such that
βω = β and ω(fβ) = −1.
11. Distinguished involutions
Let D be as in Sections 6, 7, 8 or 9, that is, a finite-dimensional simple real
algebra with a division grading Γ by an abelian group G such that D admits an
involution as a graded algebra. Let T be the support of Γ.
We already observed (see Remarks 28 and 32) that, given one such involution ϕ,
we can obtain all involutions (of the same kind in the case Z(D) = C) as Int(X)◦ ϕ,
where X runs through nonzero homogeneous elements of D. Over an algebraically
closed field such as C, which appears in this paper in Section 8 when ϕ is of the first
kind, there is no special choice of ϕ. Over the field R, however, we conclude from
our results that there is often a special choice, which we refer to as a distinguished
involution. Here we collect some of the properties of these involutions.
First assume that T is an elementary 2-group and, if the identity component
De has dimension 2, it does not coincide with Z(D). Then, looking at the lists in
Sections 6, 7 and 9, we can see that there is a unique involution ϕ characterized by
any of the following equivalent properties:
(i) ϕ has a nonzero signature;
(ii) ϕ has the maximal possible signature;
(iii) Xϕ(X) ∈ R>0 for all nonzero homogeneous X ∈ D.
This distinguished involution appears in items (1-a-1), (1-b-1), (1-c-1), (2-a-3), (2-
b-3), (2-c-4), (3-a-1), (3-b-1) and (3-c-1).
Let us now turn to the case of Section 8, that is, D ∼= Mn(C) and Γ is a division
grading on D as a complex algebra, and consider involutions of the second kind.
Then there is a unique isomorphism class, (2-f-2-0), of distinguished involutions ϕ
characterized by any of the following equivalent properties:
(i′) ϕ has a nonzero signature;
(ii′) ϕ has signature pT[2];
(iii′) for any t ∈ T of even order o(t), we have that ϕ(X) = X implies X o(t) ∈
R>0 for all nonzero X ∈ Dt.
Note that the signature of distinguished involutions reaches the maximal possible
value, n, if and only if T is an elementary 2-group. This latter condition is also
necessary and sufficient for the uniqueness of a distinguished involution (see Remark
32). Moreover, if it is satisfied, then property (iii′) is equivalent to property (iii).
If T is not an elementary 2-group then the presence of a (fixed) distinguished
involution ϕ allows us to construct a special basis in the graded subalgebra
D[2] := Ms∈T [2]
Ds.
(If T is an elementary 2-group then D[2] = De = C.) The construction is as follows.
In each component Dt, t ∈ T , we can find a nonzero element Xt such that
ϕ(Xt) = Xt, and this element is determined up to multiplication by a real scalar.
If o(t) is odd, then we can scale Xt so that X o(t)
= 1, and this determines the
element Xt uniquely. If o(t) is even, then we can also scale Xt so that X o(t)
t = 1
because ϕ is distinguished, but such an element Xt is unique only up to sign. For
t ∈ T [2], we have a way to choose the sign, which is given by the following result.
Lemma 34. Fix an isomorphism D ∼= EndC(V ) and a hermitian form h on V that
defines ϕ, that is, ϕ = σh as in Equation (1). For any X ∈ D, set hX (v, w) :=
h(v, Xw) for all v, w ∈ V . Then, for any s ∈ T [2], we have:
t
28
YURI BAHTURIN, MIKHAIL KOCHETOV, AND ADRIÁN RODRIGO-ESCUDERO
(1) If o(s) is odd, then (a) for any t ∈ T , t2 = s implies X 2
t = Xs and (b) the
signature of hXs equals the signature of h.
(2) If o(s) is even, then there exists ǫ ∈ {±1} such that (a) for any t ∈ T ,
t = ǫXs and (b) the signature of hXs equals the signature
t2 = s implies X 2
of ǫh.
t )o(s) = 1, we have X 2
Proof. Suppose t2 = s. Since X 2
t and
(X 2
t = ǫXs where ǫ = 1 if o(s) is odd and ǫ ∈ {±1} if
o(s) is even. Next, since ϕ(Xt) = Xt, we can write hXs(v, w) = h(v, Xsw) =
ǫh(Xtv, Xtw), which shows that the hermitian forms ǫh and hXs are isometric. (cid:3)
t belongs to Ds and satisfies ϕ(X 2
t ) = X 2
Note that, since ǫ is determined by each of the conditions (a) and (b), it depends
only on Xs, and neither on the choice of t ∈ T satisfying t2 = s nor on the choice
of the isomorphism D ∼= EndC(V ) and hermitian form h. If ǫ = −1, we replace Xs
by −Xs. We have proved the existence and uniqueness of a basis {Xs s ∈ T [2]}
for the complex algebra D[2] with the following properties: ϕ(Xs) = Xs, X o(s)
s = 1,
and, for any t ∈ T with t2 = s, we have that ϕ(X) = X implies X 2 ∈ R>0Xs for
all nonzero X ∈ Dt. We will refer to it as the distinguished basis. The following
result gives explicit formulas for the products of the elements of the distinguished
basis of D[2] and also for the quadratic Jordan operators of the basis elements of D
acting on the distinguished basis of D[2].
Proposition 35. Let {Xs s ∈ T [2]} be the distinguished basis of D[2]. Then:
(1) Xu2Xv2 = β(u, v)2Xu2v2 for all u, v ∈ T .
(2) For any t ∈ T , we have that ϕ(X) = X implies XXsX ∈ R>0Xst2 for
all s ∈ T [2] and nonzero X ∈ Dt. In particular, if X is scaled to satisfy
X o(t) = 1 then XXsX = Xst2 .
Proof. Recall that we chose Xt for all t ∈ T such that ϕ(Xt) = Xt and X o(t)
Then X 2
t = Xt2 for all t ∈ T .
t = 1.
We have XuXv = λXuv for some λ ∈ C with λ = 1. Applying ϕ to both sides
of this equation, we get XvXu = ¯λXuv and hence β(u, v) = λ2. Then, on the
one hand, (XuXv)2 = (λXuv)2 = λ2Xu2v2 = β(u, v)Xu2v2 and, on the other hand,
(XuXv)2 = XuXvXuXv = β(v, u)X 2
v = β(v, u)Xu2 Xv2. This proves (1).
For (2), it is necessary and sufficient to prove that XtXsXt = Xst2. Indeed, pick
uX 2
u ∈ T such that u2 = s and compute:
XtXsXt = β(t, s)XsX 2
t = β(t, s)XsXt2 = β(t, s)β(u, t)2Xst2 = Xst2 ,
where in the second last step we have used (1).
(cid:3)
Acknowledgements and license
The authors are thankful to Alberto Elduque, for numerous fruitful discussions
about gradings and mathematics in general, and to the anonymous referee, for
taking the time to read in detail such a technical text and suggest significant im-
provements of exposition. Adrián also wants to thank Alberto for his support as
his thesis supervisor.
Yuri Bahturin was supported by Discovery Grant 227060-2014 of the Natural
Sciences and Engineering Research Council (NSERC) of Canada.
Mikhail Kochetov was supported by Discovery Grant 341792-2013 of the Natural
Sciences and Engineering Research Council (NSERC) of Canada and a grant for
visiting scientists by Instituto Universitario de Matemáticas y Aplicaciones, Uni-
versity of Zaragoza.
Adrián Rodrigo-Escudero was supported by a doctoral grant of the Diputación
General de Aragón. His three-month stay with Atlantic Algebra Centre at Memorial
INVOLUTIONS ON GRADED-DIVISION SIMPLE REAL ALGEBRAS
29
University of Newfoundland was supported by the Fundación Bancaria Ibercaja
and Fundación CAI (reference number CB 4/15), and by the NSERC of Canada
(Discovery Grant 227060-2014). He was also supported by the Spanish Ministerio
de Economía y Competitividad and Fondo Europeo de Desarrollo Regional FEDER
(MTM2013-45588-C3-2-P); and by the Diputación General de Aragón and Fondo
Social Europeo (Grupo de Investigación de Álgebra).
This work is licensed under the Creative Commons Attribution-NonCommercial-
NoDerivatives 4.0 International License. To view a copy of this license, visit
http://creativecommons.org/licenses/by-nc-nd/4.0/ or send a letter to Creative
Commons, PO Box 1866, Mountain View, CA 94042, USA.
References
[1] B. Allison, S. Berman, J. Faulkner and A. Pianzola, Realization of graded-simple algebras as
loop algebras, Forum Math. 20 (2008), no. 3, 395 -- 432.
[2] Y. Bahturin and M. Kochetov, Classification of group gradings on simple Lie algebras of
types A, B, C and D, J. Algebra 324 (2010), no. 11, 2971 -- 2989.
[3] Y. Bahturin, M. Kochetov and A. Rodrigo-Escudero, Gradings on classical central simple real
Lie algebras, arXiv:1707.07909 [math.RA].
[4] Y. Bahturin and M. Zaicev, Group gradings on simple Lie algebras of type "A", J. Lie Theory
16 (2006), no. 4, 719 -- 742.
[5] Y. Bahturin and M. Zaicev, Involutions on graded matrix algebras, J. Algebra 315 (2007),
no. 2, 527 -- 540.
[6] Y. Bahturin and M. Zaicev, Simple graded division algebras over the field of real numbers,
Linear Algebra Appl. 490 (2016), 102 -- 123.
[7] Y. Bahturin and M. Zaicev, Graded division algebras over the field of real numbers,
arXiv:1611.04000 [math.RA].
[8] A. Elduque, Fine gradings on simple classical Lie algebras, J. Algebra 324 (2010), no. 12,
3532 -- 3571.
[9] A. Elduque and M. Kochetov, Gradings on simple Lie algebras, Mathematical Surveys and
Monographs, 189, American Mathematical Society, Providence, RI, 2013.
[10] N. Jacobson, Basic algebra. II, second edition, W. H. Freeman and Company, New York,
1989.
[11] G. Karpilovsky, Group representations. Vol. 1. Part A, North-Holland Mathematics Studies,
175, North-Holland Publishing Co., Amsterdam, 1992.
[12] M.-A. Knus, A. Merkurjev, M. Rost and J.-P. Tignol, The book of involutions, American
Mathematical Society Colloquium Publications, 44, American Mathematical Society, Provi-
dence, RI, 1998.
[13] A. Rodrigo-Escudero, Classification of division gradings on finite-dimensional simple real
algebras, Linear Algebra Appl. 493 (2016), 164 -- 182.
Department of Mathematics and Statistics, Memorial University of Newfoundland,
St. John's, NL, A1C5S7, Canada.
E-mail address: [email protected]
Department of Mathematics and Statistics, Memorial University of Newfoundland,
St. John's, NL, A1C5S7, Canada.
E-mail address: [email protected]
Departamento de Matemáticas e Instituto Universitario de Matemáticas y Aplica-
ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain.
E-mail address: [email protected]
|
1803.06949 | 2 | 1803 | 2018-03-26T13:13:07 | Graded Identities and Isomorphisms on Algebras of Upper Block-Triangular Matrices | [
"math.RA"
] | Let $G$ be an abelian group and $\mathbb{K}$ an algebraically closed field of characteristic zero. A. Valenti and M. Zaicev described the $G$-gradings on upper block-triangular matrix algebras provided that $G$ is finite. We prove that their result holds for any abelian group $G$: any grading is isomorphic to the tensor product $A\otimes B$ of an elementary grading $A$ on an upper block-triangular matrix algebra and a division grading $B$ on a matrix algebra. We then consider the question of whether graded identities $A\otimes B$, where $B$ is an algebra with a division grading, determine $A\otimes B$ up to graded isomorphism. In our main result, Theorem 3, we reduce this question to the case of elementary gradings on upper block-triangular matrix algebras which was previously studied by O. M. Di Vincenzo and E. Spinelli. | math.RA | math |
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS
OF UPPER BLOCK-TRIANGULAR MATRICES
ALEX RAMOS AND DIOGO DINIZ
Abstract. Let G be an abelian group and K an algebraically closed field
of characteristic zero. A. Valenti and M. Zaicev described the G-gradings on
upper block-triangular matrix algebras provided that G is finite. We prove that
their result holds for any abelian group G: any grading is isomorphic to the
tensor product A ⊗ B of an elementary grading A on an upper block-triangular
matrix algebra and a division grading B on a matrix algebra. We then consider
the question of whether graded identities A ⊗ B, where B is an algebra with
a division grading, determine A ⊗ B up to graded isomorphism. In our main
result, Theorem 3, we reduce this question to the case of elementary gradings
on upper block-triangular matrix algebras which was previously studied by O.
M. Di Vincenzo and E. Spinelli.
1. Introduction
Algebras of upper block-triangular matrices are important examples of p.i.-
algebras. The verbal ideal of ordinary identities of an algebra upper block-triangular
matrices is the product of the ideals of identities of each block (see [8]). Minimal
varieties of finite basic rank were classified by A. Giambruno and M. Zaicev in
terms of such algebras. The general result for arbitrary minimal varieties is in
terms of of mimimal superalgebras, which may be realized as subalgebras of upper
block-triangular matrix algebras with a Z2-grading (see [7, Chapter 8]).
In this
context it is of interest to classify the gradings by a group G on an algebra of
upper block-triangular matrices over a field K and to study its graded identities.
For a finite abelian group and an algebraically closed field of characteristic zero
the gradings were classified in [11]. In Section 3 we prove, as a corollary of the
result of A. Valenti and M. Zaicev, that given an algebra U T (d1, . . . , dm) of up-
per block-triangular matrices with a grading by an abelian group G there exist an
algebra U T (p1, . . . , pm) with an elementary grading and a matrix algebra Mt(K)
with a division grading such that U T (d1, . . . , dm) is isomorphic as a graded algebra
to U T (p1, . . . , pm) ⊗ Mt(K), here K is an algebraically closed field of characteristic
zero. It was conjectured in [12] that this holds for any group G and any field K. This
conjecture was recently solved in [14] with a minor restriction on the characteristic
of the field.
In Section 4 we consider the following question: Let R = A⊗B and R′ = A′ ⊗B ′,
where A and A′ are algebras of upper block-triangular matrices with elementary
gradings and B and B ′ are algebras with division gradings. If R and R′ satisfy
the same graded identities then are R and R′ isomorphic as graded algebras? If
2010 Mathematics Subject Classification. 16W50, 16R50, 16R10.
Key words and phrases. Graded algebra, Graded polynomial identity, Algebra of upper block-
triangular matrices.
1
2
ALEX RAMOS AND DIOGO DINIZ
A is a matrix algebra then R and R′ are graded simple algebras. The graded
Wedderburn-Artin theorem implies that any graded simple algebra that satisfies
the d.c.c. on graded left ideals is the tensor product of a matrix algebra with an
elementary grading and an algebra with a division grading. In [1], [9] it was proved
that graded simple algebras are determined up to isomorphism by their graded
identities. In [13] the authors consider algebras of upper block-triangular matrices
with elementary gradings, it is proved in this paper that the answer is positive
under suitable conditions. In our main result we reduce the previous question to
the case of elementary gradings. As a consequence of the results of Di Vincenzo
and Spinelli we conclude, for example, that the algebras R and R′ are isomorphic
as graded algebras if the number of blocks in A is 1 or 2.
The main results of Section 4 are that if IdG(R) = IdG(R′) then supp B =
supp B ′ and IdH (B) = IdH (B ′), where H = supp B (see Corollary 3 and Propo-
sition 3). Now we assume that H is a finitelly generated subgroup of G, since the
field is algebraically closed this implies that B is isomorphic to B ′ as a graded
algebra. It was proved by D. S. Passmann in [10] that a twisted group algebra of a
solvable group over a field of characteristic zero is semisimple. We use this result
to conclude that if a graded division algebra satisfies an ordinary polynomial iden-
tity then it satisfies the same ordinary identities as a matrix algebra (see Corollary
4). As a consequence of this result we prove that if B and B ′ satisfy an ordinary
polynomial identity then the coarsenings αR and αR′, induced by the canonical
quotient map α : G → G/H, satisfy the same identities as suitable algebras U
and U ′ (described in Theorem 3), respectively, of upper block-triangular matrices
with elementary gradings (see Lemma 6). In our main result, Theorem 3, we prove
that IdG/H (U ) = IdG/H (U ′) and that if U and U ′ are isomorphic as G/H-graded
algebras then R and R′ are isomorphic as G-graded algebras.
2. Preliminaries
Let K be a field and let G be a group with identity element e. A grading by the
group G on a K-algebra A is a vector space decomposition A = ⊕g∈GAg such that
AgAh ⊆ Agh for every g, h ∈ G. An element a of A is homogeneous in the grading
if a ∈ Ag for some g ∈ G, if a 6= 0 we say that g is the degree of a in this grading
and denote it degG a. The subspace Ae is the neutral component of A, clearly it is
a subalgebra of A. The support of the grading is the set
supp A = {g ∈ G Ag 6= 0}.
The grading on A is a division grading if A has a unit and every non-zero
homogeneous element is invertible. If K is an algebraically closed field then any
graded division algebra is isomorphic to a twisted group algebra KαG, where α is
a 2-cocycle (see [6, Theorem 2.13]). The map βα : G × G → K× given by
βα(g, h) =
α(g, h)
α(h, g)
is an alternating bicharacter. If G is a finitely generated abelian group it is easy
to verify that two twisted group algebras KαG and Kα′
G with the same associated
bicharacter are isomorphic. Moreover, it is clear that two graded division algebras
with the same graded identities have the same associated bicharacter. This proves
the following remark.
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES3
Remark 1. Let G be a finitely generated abelian group and let K be an alge-
braically closed field. If D1 and D2 be two K-algebras with division G-gradings and
IdG(D1) = IdG(D2) then D1 is isomorphic to D2 as a graded algebra.
Given a homomorphism α : G → H of groups and a G-grading A = ⊕g∈GAg
on the algebra A it follows that the decomposition A = ⊕h∈HAh, where Ah =
⊕g∈α−1(h)Ag, is an H-grading on A. We refer to this as the grading induced by α
and denote it αA
Let A and B be algebras graded by the group G. A map f : A → B is a
homomorphism of graded algebras if it is a homomorphism of algebras such that
f (Ag) ⊆ Bg for every g ∈ G. If f is also bijective then we say it is an isomorphism
of graded algebras.
Given an m-tuple of positive integers (d1, . . . , dm) we denote U T (d1, . . . , dm) the
subalgebra of Mn(K), where n = d1 + · · · + dm, of the matrices of the form
A11
...
0
· · · A1m
...
. . .
· · · Amm
,
j
1, xg
where Aij is a block of size di × dj. We refer to U T (d1, . . . , dm) as an algebra of
upper block-triangular matrices. A grading by a group G on A = U T (d1, . . . , dm)
is elementary if every elementary matrix in A is homogeneous in the grading. In
this case there exist an n-tuple (g1, . . . , gn) ∈ Gn such that the elementary matrix
eij ∈ A is homogeneous of degree gig−1
. Moreover if B is any G-graded algebra
then A ⊗ B has a G-grading such that (A ⊗ B)g is the subspace spanned by the
elements eij ⊗ b such that b is a homogeneous element in B and gi(degG b)g−1
j = g.
The free G-graded algebra is the free algebra KhXGi, freely generated by the
set XG which is the disjoint union of the sets Xg = {xg
2, · · · }, g ∈ G, with the
grading in which the variables in Xg are homogeneous of degree g.
In order to
simplify the notation we may omit the neutral element e in the notation, thus we
1 , . . . , xgn
denote by X the set Xe and by xi the indeterminate xe
n )
be a polynomial in KhXGi and A = ⊕g∈GAg be a G-graded algebra. An admissible
substitution for f is an n-tuple (a1, . . . , an) of elements of A such that ai ∈ Agi for
i = 1, . . . , n. We say that f is a graded polynomial identity for A if f (a1, . . . , an) = 0
whenever (a1, . . . , an) is an admissible substitution for f . Note that f ∈ KhXGi is
a graded polynomial identity for A if and only if f lies in the kernel of every graded
homomorphism KhXGi → A. The set of graded polynomial identities for A is
denoted IdG(A). This is an ideal of KhXGi invariant for graded endomorphisms of
KhXGi. The subalgebra KhXi of KhXGi is the ordinary free algebra. A polynomial
f (x1, . . . , xn) ∈ KhXi is a polynomial identity for an algebra B if f (b1, . . . , bn) = 0
for every b1, . . . , bn ∈ B. We denote Id(B) the ideal of identities of B.
i in X. Let f (xg1
3. Gradings on Upper Block-Triangular Matrix Algebras
In this section we present the classification of the group gradings on upper block-
triangular matrix algebras up to isomorphism. The base field is algebraically closed
of characteristic zero and the group is abelian. The main result of [11] is the
following theorem.
4
ALEX RAMOS AND DIOGO DINIZ
Theorem 1. [11, Theorem 3.2] Let G be a finite abelian group and U T (d1, . . . , dm)
an upper block-triangular matrix algebra over an algebraically closed field K of char-
acteristic zero. Then there exist a decomposition d1 = p1t, . . . , dm = pmt, a sub-
group H ⊂ G, and an n-tuple (g1, . . . , gn) ∈ Gn , where n = p1 + · · · + pm, such
that U T (d1, . . . , dm) is isomorphic to U T (p1, · · · , pm) ⊗ Mt(K) as a G-graded alge-
bra where Mt(K) is an H-graded algebra with a "fine" grading with support H and
U T (p1, · · · , pm) has an elementary grading defined by (g1, . . . , gn).
We prove, as a corollary, that the above theorem holds for arbitrary abelian
groups.
Lemma 1. Let ϕ : G → H be a homomorphism of groups and let S be a subset
of G such that the restriction of ϕ to S ∪ (S · S) is injective. Let A be a G-
graded algebra such that (supp A) ⊆ S and let B be an H-graded algebra such that
(supp B) ⊆ ϕ(S).
(i) If Bs = Bϕ(s) for any s ∈ S and Bg = 0 for any g ∈ G\S then B = ⊕g∈GBg
is a G-grading on B. A subspace of B is homogeneous in this G-grading if
and only if it is homogeneous in the H-grading.
(ii) An isomorphism of algebras f : A → B is an isomorphism of G-graded
algebras from A to B with this G-grading if and only if f is an isomorphism
of H-graded algebras form ϕA to B with its H-grading.
Proof. If g ∈ G \ S or h ∈ G \ S then BgBh = 0 ⊆ Bgh. It remains to prove that
BsBt ⊆ Bst for every s, t ∈ S. In this case Bs = Bϕ(s) and Bt = Bϕ(t), therefore
(1)
BsBt = Bϕ(s)Bϕ(s) ⊆ Bϕ(st).
If Bϕ(st) = 0 then BsBt = 0 and therefore BsBt ⊆ Bst. Now we assume that
Bϕ(st) 6= 0. In this case ϕ(st) ∈ supp B ⊆ ϕ(S). Therefore there exists an s′ ∈ S
such that ϕ(st) = ϕ(s′). Since the restriction ϕ to S ∪(S ·S) is injective we conclude
that st = s′. Hence
Bϕ(st) = Bϕ(s′) = Bs′ = Bst.
Then it follows from (1) that BsBt ⊆ Bst. An element of b ∈ B is homogeneous in
the G-grading if and only if b is homogeneous in the H-grading, this implies that a
subspace V is homogeneous in the G-grading of B if and only if it is homogeneous
in the H-grading. This proves [(i)]. Note that if b is a homogeneous element of B
we have degH b = ϕ(degG b). Since (supp A) ⊆ S and the restriction of ϕ to S
is injective we conclude that an element a ∈ A is homogeneous in the G-grading if
and only if a is homogeneous in the H-grading ϕA and that degH a = ϕ(degG a).
This implies [(ii)].
(cid:3)
Next we prove that the above theorem holds for arbitrary abelian groups. We
remark that in [14] it was proved that this result holds in general provided that
the base field K is of characteristic zero or that char K is strictly greater than
dim U T (d1, . . . , dm).
Corollary 1. The previous theorem holds for abelian groups.
Proof. Let G be an abelian group and K an algebraically closed field of character-
istic zero. Let A = U T (d1, . . . , dm) be an upper block triangular matrix algebra over
K with a G-grading. We assume without loss of generality that G is generated by
S = supp A. Since A has finite dimension the support S is finite, therefore G is iso-
morphic to a direct product of cyclic groups Zp × Zn1 ×· · ·× Znq , where p, q ≥ 0 and
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES5
n1, . . . , nq > 0 are integers. Hence we may assume without loss of generality that
G = Zp × Zn1 × · · · × Znq . Given an integer m > 0 let H = (Zm)p × Zn1 × · · · × Znq .
Denote ϕ : G → H the homomorphism given by
ϕ(z1, . . . , zp, w1, . . . , wq) = (z1, . . . , zp, w1, . . . , wq),
where z1, . . . , zp ∈ Z, wi ∈ Zni , for i = 1, . . . , q and z 7→ z is the canonical
homomorphism Z → Zm. Given any finite subset T of G we may choose m large
enough so that the restriction of ϕ to T is injective. Since the support S is finite
this implies that we may choose m large enough so that the restriction of ϕ to
T = S ∪ (S · S) is an injective map. Let ϕA be the H-grading induced by ϕ.
Theorem 1 implies that the coarsening ϕA is isomorphic as an H-graded algebra, to
B = U T (p1, · · · , pm) ⊗ Mt(K), where U T (p1, · · · , pm) has an elementary grading
and Mt(K) has a fine grading. Denote B = ⊕h∈HBh the H-grading on B. Let
Bs = Bϕ(s) for any s ∈ S and Bg = 0 for any g ∈ G \ S. Lemma 1 implies
that B = ⊕g∈GBg is a G-grading on B. Moreover any isomorphism of H-graded
algebras from ϕA to B is also an isomorphism of G-graded algebras from A to B
with this G-grading, hence A with its G-grading is isomorphic as a graded algebra
to B with this G-grading. A subspace of B is homogeneous in this G-grading if and
only if it is homogeneous in the H-grading, therefore we obtain a fine G-grading
on Mt(K) and an elementary G-grading on U T (p1, . . . , pm) such that B with the
G-grading is U T (p1, . . . , pm) ⊗ Mt(K) with the canonical G-grading on the tensor
product.
(cid:3)
In [4] the isomorphism classes of upper block-triangular matrix algebras graded
by an abelian group were described.
Corollary 2. [4, Corollary 4] Let G be an abelian group and let K be an arbitrary
field. Let R and R′ denote the algebras U T (p1, . . . , pm)⊗KB and U T (p′
m)⊗K
B ′ respectively, where B, B ′ are graded division algebras and U T (p1, . . . , pm),
U T (p′
m) have elementary gradings defined by the tuples g, g′ of elements
of G, respectively. The algebras R and R′ are isomorphic if and only if B ∼= B ′,
(p1, . . . , pm) = (p′
m) and there exist a g ∈ G, h1, . . . , hn ∈ supp B and
σ ∈ Sp1 × · · · × Spm such that g′
i = gσ(i)hσ(i)g for i = 1, . . . , n.
1, . . . , p′
1, . . . , p′
1, . . . , p′
We remark that in [4] the previous corollary was sated for B = Mt(K) and
B ′ = Mt′(K), this however is not used in the proof.
Let R = A ⊗ B, where A = U T (d1, . . . , dm) is an upper block-triangular ma-
trix algebra with an elementary grading and B is a graded division algebra with
support H. Let g = (g1, . . . , gn) be a tuple of elements of G that determines
the elementary grading in A. Now let g′
n representatives of the cosets
g1H, . . . , gnH, respectively, such that, giH = g′
i = g′
j
jH. Let A′ denote the algebra U T (d1, . . . , dm) with the elemen-
whenever g′
tary grading induced by g′ = (g′
It follows from Corollary 2 that the
algebras R and R′ = A′ ⊗ B are isomorphic. Since (g′
j ∈ H if and only if
g′
i = g′
iH for i = 1, . . . , n and g′
j we conclude that R′
e ⊗ 1 ∼= A′
e.
1, . . . , g′
1, . . . , g′
iH = g′
n).
i)−1g′
e = A′
Remark 2. Let R = A⊗B, where A = U T (d1, . . . , dm) is an upper block-triangular
matrix algebra with an elementary grading and B is a graded division algebra. The
algebra U T (d1, . . . , dm) may be equipped with an elementary grading A′ such that
R is isomorphic to R′, where R′ = A′ ⊗ B and R′
e = A′
e ⊗ 1.
6
ALEX RAMOS AND DIOGO DINIZ
4. Graded Identities for Algebras of Upper Block-Triangular
Matrices
In this section we present the main results of the paper. Next we define the
Capelli polynomials, these polynomials will play an important role in this section.
Definition 1. The Capelli polynomial of rank t is the polynomial
Capt(x1, . . . , xt, y1, . . . , yt+1) = X
σ∈St
(sgn σ)y1xσ(1)y2xσ(2) · · · ytxσ(t)yt+1,
in KhXi.
Lemma 2. [7, Lemma 9.1.4] If d = d2
the Capelli identity Capd+m ≡ 0 but does not satisfy Capd+m−1 ≡ 0.
1 +· · · d2
m, the algebra U T (d1, . . . , dm) satisfies
Remark 3. As noted in [13, Lemma 3.2] it follows from the proof of Lemma 9.1.4
in [7] that the result of any substitution in Capd+m−1 lies in J(A)m−1, where A =
U T (d1, . . . , dm) and J(A) denotes the Jacobson radical of A.
Proposition 1. Let A = U T (d1, . . . , dm) denote an upper-block triangular matrix
algebra with an elementary grading induced by g = (g1, . . . , gn), where n = d1 +
· · · + dm. Let r denote the number of distinct elements that appear in g. Then the
neutral component Ae is a direct sum of r ideals, each isomorphic to an upper-block
triangular matrix algebra of length ≤ m.
Proof. Let I1 = {1, . . . , d1} and
It = {d1 + · · · + dt−1 + 1, . . . , d1 + · · · + dt−1 + dt},
for t = 2, . . . , m. The sets I1, . . . , Im form a partition of N = {1, . . . , n}. A basis for
A consists of the matrix units eij where i ∈ Is, j ∈ It and s ≤ t. Denote h1, . . . , hr
the elements of G that appear in the n-tuple g, moreover let Jk = {i ∈ N gi = hk}.
It is clear that a basis for Ae consists of the matrix units eij in the basis for A such
that i, j ∈ Jk for some k. Therefore
Ae = B1 ⊕ · · · ⊕ Br,
where Bk is the subspace of Ae generated by the matrix units eij in the basis for
A such that i, j ∈ Jk. It is easy to verify that Bk is an ideal in Ae. A basis for Bk
is the set of matrix units eij where i ∈ Is ∩ Jk, j ∈ It ∩ Jk and s ≤ t, therefore Bk
is isomorphic an algebra of upper block-triangular matrices of length ≤ m.
(cid:3)
The following remarks will be used in the proof of Lemma 3.
Remark 4. Let a1, . . . , at, v1, . . . , vt−1 be matrix units in a matrix algebra and
assume that v1, . . . , vt−1 are pairwise distinct. If
a1v1 · · · at−1vt−1at 6= 0
then for every permutation τ ∈ St−1 different from Id we have
a1vτ (1) · · · at−1vτ (t−1)at = 0.
Lemma 3. Let A be an upper-block triangular matrix algebra with an elementary
G-grading. Let t be the natural number such that Capt−1 is not an identity for Ae
and Capt is an identity for Ae. Then Capt−1xg
Capt(x1, . . . , xt−1, xg
t /∈ IdG(A) and
t , y1, . . . , yt+1) ∈ IdG(A)
if and only if g = e.
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES7
t /∈ IdG(A). To prove the converse we assume that Capt−1xg
Proof. It is clear, from the choice of t, that if g = e then Capt is an identity for A
and Capt−1xg
t is not an
identity for A and that g 6= e and prove that Capt(x1, . . . , xt−1, xg
t , y1, . . . , yt+1) is
not an identity for A. Let Ae = B1 ⊕ · · · ⊕ Br be the decomposition in Proposition
1. Let a1, . . . , at−1, at, v1, . . . , vt−1, vt be an admissible substitution by matrix units
such that
Capt−1(v1, . . . , vt−1, a1, . . . , at)vt 6= 0.
This implies that elements v1, . . . , vt−1 are pairwise distinct, moreover there exists
i such that v1, . . . , vt−1, a1, . . . , at ∈ Bi. We may reorder the elements v1, . . . , vt−1,
if necessary, and assume that
c := a1v1 · · · at−1vt−1atvt 6= 0.
Clearly there exists at+1 ∈ Ae such that cat+1 6= 0. Since degG(vt) 6= e and
degG(vi) = e, for i = 1, . . . , t − 1, we conclude that the elements v1, . . . , vt are
pairwise distinct. Remark 4 implies that
(2)
a1vσ(1)a2 · · · at−1vσ(t)at+1 = 0
whenever σ 6= 1. Therefore we conclude that Capt(v1, . . . , vt, a1, . . . , at+1) 6= 0. (cid:3)
Corollary 3. Let R = A ⊗ B, where A is an upper block-triangular matrix algebra
with an elementary grading and B is a graded division algebra. Let t be the natural
number such that Capt−1 is not an identity for Re and Capt is an identity for Re.
The element g lies in supp B if and only if Capt−1xg
t /∈ IdG(R) and the polynomial
Capt(x1, . . . , xt−1, xg
t , y1, . . . , yt+1) lies in IdG(R).
Proof. Remark 2 implies that we may assume without loss of generality that
Re = Ae ⊗ 1.
(⇒)
Let g be an element of supp B. Since Capt−1 is not an identity for Re exist
c1, . . . , ct−1, d1, . . . , dt ∈ Ae such that
Let b be a non-zero element in Bg. We have
Capt−1(c1, . . . , ct−1, d1, . . . , dt) 6= 0.
(cid:0)Capt−1(c1 ⊗ 1, . . . , ct−1 ⊗ 1, d1 ⊗ 1, . . . , dt ⊗ 1)(cid:1) (1 ⊗ b) 6= 0.
Hence Capt−1xg
t /∈ IdG(R). Since 1 ⊗ b is a homogeneous invertible element of
degree g in R and Re = Ae ⊗ 1 we conclude that Rg = {a ⊗ b a ∈ Ae}.
Therefore given a homomorphism of graded algebras ϕ : F hXGi → R there ex-
ist v1, . . . , vt−1, vt, w1, . . . , wt+1 ∈ Ae such that ϕ(xi) = vi ⊗ 1, i = 1, . . . , t − 1,
ϕ(xg
t ) = vt ⊗ b and ϕ(yj) = wj ⊗ 1, j = 1, . . . , t + 1. Hence
ϕ (Capt(x1, . . . , xt−1, xg
Capt(v1, . . . , vt−1, vt, w1, . . . , wt+1) ⊗ b = 0.
t , y1, . . . , yt+1)) =
Therefore Capt(x1, . . . , xt−1, xg
t , y1, . . . , yt+1) lies in IdG(R).
(⇐)
Since Capt−1xg
t /∈ IdG(R) there exists a homomorphism of graded algebras ϕ :
i=1 ai ⊗ bi, where
t ) 6= 0. We write ϕ(xg
F hXGi → R such that ϕ(Capt−1xg
ai ∈ A and bi ∈ B are homogeneous elements and
t ) = Pn
(3)
(degGai) · (degGbi) = g,
8
ALEX RAMOS AND DIOGO DINIZ
for i = 1, . . . , n. Since Re = Ae ⊗ 1 there exist c1, . . . , ct−1, d1, . . . , dt ∈ Ae such
that
ϕ(Capt−1) = Capt−1(c1, . . . , ct−1, d1, . . . , dt) ⊗ 1.
Note that ϕ(Capt−1) · (aj ⊗ bj) 6= 0 for some j. This implies that
Capt−1(c1, . . . , ct−1, d1, . . . , dt) · aj 6= 0,
hence the polynomial Capt−1xg1
for A. Given v1, . . . , vt−1, w1, . . . , wt+1 ∈ Ae and vt ∈ Ag1 we have
t , where g1 = (degG aj), is not a graded identity
0 = Capt(v1 ⊗ 1, . . . , vt−1 ⊗ 1, vt ⊗ bj, w1 ⊗ 1, . . . , wt+1 ⊗ 1)
= Capt(v1, . . . , vt−1, vt, w1, . . . , wt+1) ⊗ bj.
The element bj is a non-zero homogeneous element in B, hence it is invertible. This
implies that Capt(v1, . . . , vt−1, vt, w1, . . . , wt+1) = 0. Therefore the polynomial
Capt(x1, . . . , xt−1, xg1
t , y1, . . . , yt+1) is a graded identity for A. The previous lemma
implies that g1 = e, therefore it follows from (3) that g = (degGbi) ∈ supp B. (cid:3)
Lemma 4. Let A = U T (d1, . . . , dm) be an upper block-triangular matrix algebra
and t = d2
m + m. Let f be a polynomial such that f Capt−1 is a multilinear
If the result of a substitution is a non-zero element in A then the
polynomial.
variables in f are replaced by elements in the first block of A.
1 + · · · d2
Proof. Since f Capt−1 is a multilinear polynomial we only need to consider sub-
stitutions of the variables by matrix units. Let ϕ : F hXi → A be a homomorphism
such that ϕ(x) is a matrix unit for every x ∈ X and ϕ(f Capt−1) 6= 0. Remark 3
implies that ϕ(Capt−1) ∈ J(A)m−1. Therefore ϕ(Capt−1) = ers with 1 ≤ r ≤ d1.
Since J(A)m = 0, if any variable in f is replaced by an element in J(A) we have
ϕ(f ) ∈ J(A). Therefore ϕ(f Capt−1) lies in J(A)m = 0, a contradiction. Hence
the variables in f are replaced by elements in the diagonal blocks of A. If vari-
ables are replaced by elements in distinct blocks then ϕ(f ) = 0. Thus the variables
in f are replaced by elements in a single diagonal block. Hence ϕ(f ) = etu with
d1 + · · · + di−1 + 1 ≤ t, u ≤ d1 + · · · + di. We have
0 6= ϕ(f Capt−1) = etuers,
hence u = r. Therefore we conclude that i = 1.
(cid:3)
Proposition 2. Let U = U1 ⊕· · ·⊕Ur, where U1, . . . , Ur are ideals of U isomorphic
to algebras of upper block-triangular matrices. Denote t the least integer such that
Capt ∈ Id(U ). There exists a polynomial f in KhXi such that:
(i) f Capt−1 is a multilinear polynomial;
(ii) f Capt−1 is not an identity for U ;
(iii) If ϕ : KhXi → U is a homomorphism and ϕ(f Capt−1) 6= 0 then ϕ(f ) is in
the center of U .
mi),
, . . . , d(js)
1
1 )2 + · · · + (d(i)
Proof. We may assume without loss of generality that have Ui = U T (d(i)
1 , . . . , d(i)
i = 1, . . . , r. Denote ti = (d(i)
mi)2 + mi. Note that t = max {t1, . . . , tr}.
Denote k = max {d(j1)
1 }, where {j1, . . . , js} = {j tj = t}. Let f be a
central polynomial for Mk(K) such that f Capt−1 is a multilinear polynomial. It is
clear that is not an identity for U . We claim that (iii) holds. Let ϕ : F hXi → U be
a homomorphism such that ϕ(f Capt−1) 6= 0. Note that ϕ(f Capt−1) is the result
of the substitution x 7→ ϕ(x) in f Capt−1. Since ϕ(f Capt−1) 6= 0 there exists an i
such that each indeterminate in f Capt−1 is replaced by an element in Ui. Lemma
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES9
4 implies that the variables in f are replaced by elements in the first block of Ui.
Recall that f is an identity for Ml(K) if k > l (see [5, pg. 150, Lemma 3.5]). Since
ϕ(Capt−1) 6= 0 we conclude that ti = t. Therefore the choice of k implies that
d(i)
1 = k. Hence ϕ(f ) is a central element.
(cid:3)
Definition 2. Let A be an upper block-triangular matrix algebra with an elementary
grading and let t be the least integer such that Capt ∈ Id(Ae). We say a multilinear
polynomial f is A-good if the polynomial f Capt−1 is multilinear, f Capt−1 is not a
polynomial identity for Ae and for every homomorphism ϕ : KhXi → Ae such that
ϕ(f Capt−1) 6= 0 the element ϕ(f ) is in the center of Ae.
Remark 5. Proposition 1 and Proposition 2 imply that for any upper block-triangular
matrix algebra A with an elementary grading there exists an A-good polynomial.
Proposition 3. Let R = A ⊗ B, where A is an upper block-triangular matrix
algebra with an elementary grading and B is a graded division algebra. Let t the
least integer such that Capt ∈ Id(Ae) and let f be an A-good polynomial. Let
S1 = {yh1
m} be u pairwise disjoint sets of
variables and let
m}, . . . , Su = {zhu
2, . . . , ye
2, . . . , ze
1 , ze
1 , ye
f h1 = f (yh1
m), . . . , f hu = f (zhu
1 , ye
A multilinear polynomial g(xh1
u ) ∈ F hXH i, where H = supp B, is a graded
identity for B if and only if the multilinear polynomial g(f h1, . . . , f hu)Capt−1 is a
graded identity for R = A ⊗ B.
2, . . . , ye
1 , . . . , xhu
2, . . . , ze
1 , ze
m).
Proof.
As a consequence of Remark 2 we may assume without loss of generality that
Re = Ae ⊗ 1.
(⇒)
For each h ∈ H we fix a non-zero element bh ∈ Bh. Since Rh = Re(1 ⊗ bh) any
element a in Rh is of the form ae ⊗ bh, where ae ∈ Ae. Let ϕ : F hXGi → R
be a homomorphism of graded algebras. Note that for each l there exists an
fl ∈ f (Ae, . . . , Ae) such that ϕ(f hl) = fl ⊗ bhl. Moreover ϕ(Capt−1) = d ⊗ 1
for some d ∈ Capt−1(Ae, . . . , Ae). The polynomial f is an A-good polynomial,
therefore either fl is a central element in Ae or fld = 0. Hence we conclude that
ϕ(g(f h1, . . . , f hu)Capt−1) = 0 if fld = 0 for some l. It remains to consider the case
In this case ϕ(g(f h1 , . . . , f hu)) =
where f1, . . . , fu are central elements of Ae.
f1 · · · fu ⊗ g(bh1, · · · , bhu) = 0, because g(xh1
u ) ∈ IdH (B). Therefore
g(f h1, . . . , f hu)Capt−1 is a graded identity for R = A ⊗ B.
1 , . . . , xhu
(⇐)
For each h ∈ H we choose an element bh in Bh. There exists a substitution
xi 7→ ai ∈ Ae in f Capt−1 that results in a non-zero element in Ae. Denote a
the result of this substitution in Capt−1. Then ca 6= 0, where c = f (a1, . . . , am).
We now obtain substitutions in each f hi that result in the element c ⊗ bhi, for
i = 1, . . . , u. Hence we obtain a substitution in g(f h1, . . . , f hu)Capt−1 that results
in
g(c ⊗ bh1 , · · · , c ⊗ bhu)(a ⊗ 1) = cua ⊗ g(bh1, · · · , bhu).
Since f is an A-good polynomial the element c lies in Z(Ae). Therefore c a diagonal
matrix, hence from ca 6= 0 we conclude that cua 6= 0. Therefore g(bh1, · · · , bhu) = 0.
Since the elements bh1, . . . , bhu are arbitrary we conclude that g(xh1
u ) is a
graded identity for R.
(cid:3)
1 , . . . , xhu
10
ALEX RAMOS AND DIOGO DINIZ
Lemma 5. Let K be a field of characteristic zero and let A = ⊕g∈GAg be a K-
algebra with a G-grading. If the algebras B and B ′ satisfy the same ordinary iden-
tities then A ⊗ B and A ⊗ B ′ satisfy the same graded identities.
Proof. Let F = KhXi/Id(B), we prove first that IdG(A ⊗ B) = IdG(A ⊗ F ). Let
f (xg1
1 , · · · , xgn
n ) be a multilinear graded identity for A⊗F and let a1 ⊗b1, . . . , an⊗bn
be elements of A ⊗ B such that ai ⊗ bi ∈ (A ⊗ B)gi , i = 1, . . . , n. Let ϕ : A ⊗ F →
A ⊗ B be a graded homomorphism such that ϕ(ai ⊗ xi) = ai ⊗ bi. We have
f (a1 ⊗ x1, . . . , an ⊗ xn) = 0, therefore
0 = ϕ(f (a1 ⊗ x1, . . . , an ⊗ xn)) = f (a1 ⊗ b1, . . . , an ⊗ bn).
Hence we conclude that f is a graded identity for A ⊗ B. Since char K = 0 this
implies that IdG(A ⊗ F ) ⊆ IdG(A ⊗ B). Now let f (xg1
n ) be a multilinear
graded identity for A ⊗ B and let ϕ : KhXGi → A ⊗ F be a graded homomorphism.
We write
1 , · · · , xgn
ϕ(f ) = a1 ⊗ f1 + · · · + am ⊗ fm,
where a1, . . . , am are linearly independent elements of A. If fi 6= 0 for some i then
there exists a homomorphism ψ0 : F → B such that ψ0(fi) 6= 0. Then ψ = IdA ⊗ψ0
is a graded homomorphism from A ⊗ F to A ⊗ B and
ψ ◦ ϕ(f ) = a1 ⊗ ψ0(f1) + · · · + am ⊗ ψ0(fm) 6= 0,
this is a contradiction because ψ ◦ ϕ : KhXGi → A ⊗ B is a graded homomorphism
and f is a graded polynomial identity for A⊗ B. Therefore f1 = f2 = · · · = fm = 0.
Hence f is a graded identity for A⊗ F . This proves that IdG(A⊗ B) ⊆ IdG(A⊗ F ).
Therefore IdG(A ⊗ B) = IdG(A ⊗ F ). Clearly we also have IdG(A ⊗ B ′) = IdG(A ⊗
F ).
(cid:3)
Lemma 6. Let R = A ⊗ B, where A = U T (d1, . . . , dm) has the elementary grading
induced by the n-tuple (g1, . . . , gn) ∈ Gn and B is an algebra that satisfies the same
ordinary identities as Mt(K) and has a G-grading with support the subgroup H of
G. Let α : G → G/H denote the canonical epimorphism. The coarsening αR
satisfies the same G/H-graded identities as the algebra U T (d1t, . . . , dmt) with the
elementary grading induced by (g1H, g1H, . . . , gnH, gnH), where each lateral class
is repeated t times.
Proof. Note that αR = (αA) ⊗ B, where B has the trivial G/H-grading. Lemma
5 implies that αR satisfies the same graded identities as (αA) ⊗ Mt(K). Let ϕ :
Mn(K) ⊗ Mt(K) → Mnt(K) denote the canonical isomorphism:
if we write the
matrices in Mnt(K) as n2 blocks of size t × t then given a = (aij) ∈ Mn(K)
and b ∈ Mt(K) the (i, j)-th block of the matrix ϕ(a ⊗ b) is aijb. We consider
A = U T (d1, . . . , dm) as an ungraded subalgebra of Mn(K), where n = d1 + · · ·+ dm.
The restriction of ϕ to A⊗Mt(K) is an isomorphism (of ungraded algebras) onto the
subalgebra U T (d1t, . . . , dmt) of Mnt(K). If U T (d1t, . . . , dmt) has the elementary
grading induced by (g1H, g1H, . . . , gnH), where each lateral class is repeated t
times, then this restriction is an isomorphism of graded algebras from (αA)⊗Mt(K)
to U T (d1t, . . . , dmt).
(cid:3)
The next two results will be used in the proof of Theorem 3.
Proposition 4. [3, Proposition 1] Let A and B be two algebras G-graded algebras
and α : G → H a homomorphism of groups. If IdG(A) = IdG(B) then IdH (αA) =
IdH (αB).
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES11
In order to proceed to our main result we need to describe the ordinary identities
of graded division algebras.
Theorem 2. [13, Theorem 3] Let G be a solvable group and suppose that either K
has characteristic 0 or K has characteristic p > 0 and G has no elements of order
p. Then K αG is semisimple.
A semisimple algebra is the subdirect product of primitive algebras. It follows
from Kaplansky's Theorem that it satisfies the same identities of a matrix algebra
Mt(K), for some t (see, for example, the remarks after Theorem 3.2 in [2]). This
discussion proves the following corollary.
Corollary 4. Let G be an abelian group and K an algebraically closed field of
If D is a graded division algebra that satisfies an ordinary
characteristic zero.
polynomial identity then there exists a natural number t such that D satisfies the
same ordinary identities as Mt(K).
We recall that the p.i. degree of a p.i. algebra D is the largest integer d such
that the multinear identities of B follow from the multilinear identities of Md(K)
(see [5, pg. 90, Definition 7.1.8]). Therefore the natural number t in the previous
corollary is the p.i. degree of D.
1, . . . , d′
Theorem 3. Let G be an abelian group. Denote A = U T (d1, . . . , dm) and A′ =
U T (d′
m) upper block-triangular matrix algebras with elementary gradings in-
duced by the tuples g = (g1, . . . , gn) and g′ = (g′
n), respectively. Let K be
a finitely generated subgroup of G and let B and B ′ be K-graded division algebras
that satisfy an ordinary polynomial identity and let R = A ⊗ B, R′ = A′ ⊗ B ′.
If IdG(R) = IdG(R′) then supp B = supp B ′ and B ∼= B ′. If B and B ′ have
p.i. degree t then IdG/H (αR) = IdG/H (U ) = IdG/H (U ′) = IdG/H (αR′), where
α : G → G/H is the canonical epimorphism, and U , U ′ denote the elementary
gradings induced by (g1H, g1H, . . . , gnH, gnH) and (g′
nH), re-
spectively, on the algebra U T (d1t, . . . , dmt), where each lateral class is repeated t
times. Moreover, if U ∼= U ′ then R ∼= R′.
1H, . . . , g′
1, . . . , g′
1H, g′
nH, g′
Proof. Corollary 3 and Proposition 3 imply that supp B = supp B ′ and
IdH (B) = IdH (B ′), therefore Remark 1 implies that B ∼= B ′. Corollary 4 and
Lemma 6 imply that αR, αR′ satisfy the same G/H-graded identities as U and U ′
respectively. The equality IdG/H (αR) = IdG/H (αR′) follows from Proposition 4.
Therefore IdG/H (αR) = IdG/H (U ) = IdG/H (U ′) = IdG/H (αR′). Now we assume
that U and U ′ are isomorphic as G/H-graded algebras. Corollary 2 imply that
there exist a gH ∈ G/H, σ ∈ Std1 × · · · × Stdm such that g′
iH = gσ(i)H(gH) for i =
1, . . . , tn. Therefore there exists τ ∈ Sd1 × · · · × Sdm such that g′
iH = gτ (i)H(gH),
for i = 1, . . . , n. Clearly this implies that that there exists h1, . . . , hn ∈ H such that
i = gτ (i)hτ (i)g, for i = 1, . . . , n. Hence it follows from Corollary 2 that R ∼= R′. (cid:3)
g′
In [13] the authors provide sufficient conditions to ensure that the algebras U and
U ′ in the previous theorem are isomorphic as graded algebras. To state their result
we need the notion of invariance subgroup of an elementary grading introduced in
that paper.
Definition 3. Let A denote the matrix algebra Mn(K) with the elementary grading
induced by (g1, . . . , gn). The map ωA : G → N associates to each g ∈ G the number
of indices i such that gi = g. The invariance subgroup of this elementary grading is
HA = {h ∈ G ωA(hg) = ωA(g), ∀g ∈ G}.
12
ALEX RAMOS AND DIOGO DINIZ
Proposition 5. [13, Corollary 3.4] Let G be an abelian group and let A and B be
upper block-triangular matrix algebras endowed with an elementary G-grading such
that A has either at most two block components or finitely many block components
but at least one of them, let us say Ad, with invariance subgroup H (d)
A = h1Gi. Then
A and B are isomorphic as G-graded algebras if, and only if, IdG(A) = IdG(B).
The previous proposition and Theorem 3 imply the following result.
1, . . . , d′
Corollary 5. Let G be an abelian group. Denote A = U T (d1, . . . , dm) and A′ =
U T (d′
m) upper block-triangular matrix algebras with elementary gradings in-
duced by the tuples g ∈ Gn and g′, respectively. Let K be a finitely generated sub-
group of G, let B and B ′ be G-graded division algebras that satisfy an ordinary p.i.
and have the same p.i. degree. Let R = A ⊗ B, R′ = A′ ⊗ B ′. The algebra A
has either at most two block components or finitely many block components but at
least one of them has trivial invariance subgroup for αA, where α : G → G/H is
the canonical epimorphism. Then R and R′ are isomorphic as G-graded algebras
if, and only if, IdG(R) = IdG(R′).
Proof. Let U and U ′ be the elementary graded algebras in the previous theo-
rem. The invariance subgroup of the i-th block of αA coincides with the invariance
subgroup of the i-th block U . The number of blocks in αA also coincide with the
number of blocks in U . Therefore Proposition 5 implies that U ∼= U ′. The result
now follows from Theorem 3.
(cid:3)
References
[1] E. Aljadeff, D. Haile, Simple G-graded algebras and their polynomial identities, Trans. Amer.
Math. Soc. 366 (2014), 1749 -- 1771 .
[2] E. Aljadeff, A. Kanel-Belov, Y. Karasik, Kemer's theorem for affine PI algebras over a field
of characteristic zero, Journal of Pure and Applied Algebra 220 (2016) 2771 -- 2808.
[3] Y. Bahturin, D. Diniz, Graded identities of simple real graded division algebras, Journal of
Algebra, In Press.
[4] A. R. Borges, C. Fidelis and D. Diniz, Graded isomorphisms on upper block triangular matrix
algebras, Linear Algebra and its Applications 543 (2018) 92 -- 105.
[5] V. Drensky, E. Formanek, Polynomial Identity Rings, Springer Basel AG (2000), 196 p.
[6] A. Elduque, M. Kochetov, Gradings on Simple Lie Algebras, AMS Math. Surv. Monographs
189 (2013), 336p.
[7] A. Giambruno, M. Zaicev, Polynomial Identities and Asymptotic Methods, Mathematical Sur-
veys and Monographs 122 (2005) American Mathematical Society.
[8] A. Giambruno, M. Zaicev, Minimal varieties of exponential growth, Adv. Math. 174 (2003),
310 -- 323.
[9] P. Koshlukov, M. Zaicev, Identities and isomorphisms of graded simple algebras, Linear Alge-
bra and its Applications 432 (2010) 3141 -- 3148.
[10] D. S. Passman, On the Semisimplicity of Twisted Group Algebras, Proceedings of the Amer-
ican Mathematical Society 25 (1970), 161 -- 166.
[11] A. Valenti, M. Zaicev, Abelian gradings on upper block triangular matrices, Canad. Math.
Bull 55 (2012), 208 -- 213.
[12] A. Valenti, M. Zaicev, Group gradings on upper triangular matrices, Arch. Math. 89 (2007),
33 -- 40.
[13] O. M. Di Vincenzo, E, Spinelli, Graded polynomial identities on upper block triangular matrix
algebras, Journal of Algebra 415 (20014) 50 -- 64.
[14] F.Y. Yasumura, Group gradings on upper block triangular matrices, Arch. Math. 4 (2018)
327 -- 332.
GRADED IDENTITIES AND ISOMORPHISMS ON ALGEBRAS OF UPPER BLOCK-TRIANGULAR MATRICES13
Unidade Academica de Matem´atica, Universidade Federal de Campina Grande, Camp-
ina Grande, PB, 58429-970, Brazil
E-mail address: [email protected]
Unidade Academica de Matem´atica, Universidade Federal de Campina Grande, Camp-
ina Grande, PB, 58429-970, Brazil
E-mail address: [email protected]
|
1110.1199 | 4 | 1110 | 2012-12-28T20:28:13 | Factorial cluster algebras | [
"math.RA"
] | We show that cluster algebras do not contain non-trivial units and that all cluster variables are irreducible elements. Both statements follow from Fomin and Zelevinsky's Laurent phenomenon. As an application we give a criterion for a cluster algebra to be a factorial algebra. This can be used to construct cluster algebras, which are isomorphic to polynomial rings. We also study various kinds of upper bounds for cluster algebras, and we prove that factorial cluster algebras coincide with their upper bounds. | math.RA | math |
FACTORIAL CLUSTER ALGEBRAS
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Abstract. We show that cluster algebras do not contain non-trivial units and
that all cluster variables are irreducible elements. Both statements follow from
Fomin and Zelevinsky's Laurent phenomenon. As an application we give a crite-
rion for a cluster algebra to be a factorial algebra. This can be used to construct
cluster algebras, which are isomorphic to polynomial rings. We also study various
kinds of upper bounds for cluster algebras, and we prove that factorial cluster
algebras coincide with their upper bounds.
Contents
Introduction and main results
Invertible elements in cluster algebras
Irreducibility of cluster variables
1.
2.
3.
4. Factorial cluster algebras
5. The divisibility group of a cluster algebra
6. Examples of non-factorial cluster algebras
7. Examples of factorial cluster algebras
8. Applications
References
1
6
8
9
11
12
14
21
23
1. Introduction and main results
1.1. Introduction. The introduction of cluster algebras by Fomin and Zelevinsky
[FZ1] triggered an extensive theory. Most results deal with the combinatorics of seed
and quiver mutations, with various categorifications of cluster algebras, and with
cluster phenomena occuring in various areas of mathematics, like representation
theory of finite-dimensional algebras, quantum groups and Lie theory, Calabi-Yau
categories, non-commutative Donaldson-Thomas invariants, Poisson geometry, dis-
crete dynamical systems and algebraic combinatorics.
On the other hand, there are not many results on cluster algebras themselves.
As a subalgebra of a field, any cluster algebra A is obviously an integral domain.
It is also easy to show that its field of fractions Frac(A) is isomorphic to a field
K(x1, . . . , xm) of rational functions. Several classes of cluster algebras are known to
be finitely generated, e.g. acyclic cluster algebras [BFZ, Corollary 1.21] and also a
class of cluster algebras arising from Lie theory [GLS2, Theorem 3.2]. Berenstein,
Fomin and Zelevinsky gave an example of a cluster algebra which is not finitely
2010 Mathematics Subject Classification. Primary 13F60; Secondary 13F15,17B37.
1
2
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
generated. (One applies [BFZ, Theorem 1.24] to the example mentioned in [BFZ,
Proposition 1.26].) Only very little is known on further ring theoretic properties
of an arbitrary cluster algebra A. Here are some basic questions we would like to
address:
• Which elements in A are invertible, irreducible or prime?
• When is A a factorial ring?
• When is A a polynomial ring?
In this paper, we work with cluster algebras of geometric type.
1.2. Definition of a cluster algebra. In this section we repeat Fomin and Zelevin-
sky's definition of a cluster algebra.
A matrix A = (aij) ∈ Mn,n(Z) is skew-symmetrizable (resp.
symmetrizable)
if there exists a diagonal matrix D = Diag(d1, . . . , dn) ∈ Mn,n(Z) with positive
diagonal entries d1, . . . , dn such that DA is skew-symmetric (resp. symmetric), i.e.
diaij = −djaji (resp. diaij = djaji) for all i, j.
Let m,n and p be integers with
m ≥ p ≥ n ≥ 1
and m > 1.
Let B = (bij) ∈ Mm,n(Z) be an (m×n)-matrix with integer entries. By B◦ ∈ Mn,n(Z)
we denote the principal part of B, which is obtained from B by deleting the last
m − n rows.
Let ∆(B) be the graph with vertices 1, . . . , m and an edge between i and j pro-
vided bij or bji is non-zero. We call B connected if the graph ∆(B) is connected.
Throughout, we assume that K is a field of characteristic 0 or K = Z. Let
F := K(X1, . . . , Xm) be the field of rational functions in m variables.
A seed of F is a pair (x, B) such that the following hold:
(i) B ∈ Mm,n(Z),
(ii) B is connected,
(iii) B◦ is skew-symmetrizable,
(iv) x = (x1, . . . , xm) is an m-tuple of elements in F such that x1, . . . , xm are
algebraically independent over K.
For a seed (x, B), the matrix B is the exchange matrix of (x, B). We say that B
has maximal rank if rank(B) = n.
Given a seed (x, B) and some 1 ≤ k ≤ n we define the mutation of (x, B) at k as
µk(x, B) := (x′, B′),
FACTORIAL CLUSTER ALGEBRAS
3
where B′ = (b′
ij) is defined as
b′
−bij
bij +
ij :=
s :=(x−1
k Qbik>0 xbik
m) is defined as
xs
x′
2
and x′ = (x′
1, . . . , x′
The equality
(1)
bikbkj + bikbkj
if i = k or j = k,
otherwise,
i + x−1
k Qbik<0 x−bik
i
if s = k,
otherwise.
xkx′
k = Ybik>0
xbik
i + Ybik<0
x−bik
i
is called an exchange relation. We write
and
µ(x,B)(xk) := x′
k
µk(B) := B′.
It is easy to check that (x′, B′) is again a seed. Furthermore, we have µkµk(x, B) =
(x, B).
Two seeds (x, B) and (y, C) are mutation equivalent if there exists a sequence
(i1, . . . , it) with 1 ≤ ij ≤ n for all j such that
µit · · · µi2µi1(x, B) = (y, C).
In this case, we write (y, C) ∼ (x, B). This yields an equivalence relation on all
seeds of F . (By definition (x, B) is also mutation equivalent to itself.)
For a seed (x, B) of F let
X(x,B) := [(y,C)∼(x,B)
{y1, . . . , yn},
where the union is over all seeds (y, C) with (y, C) ∼ (x, B). By definition, the
cluster algebra A(x, B) associated to (x, B) is the L-subalgebra of F generated by
X(x,B), where
L := K[x±1
n+1, . . . , x±1
p , xp+1, . . . , xm]
is the localization of the polynomial ring K[xn+1, . . . , xm] at xn+1 · · · xp. (For p = n
we set xn+1 · · · xp := 1.) Thus A(x, B) is the K-subalgebra of F generated by
p , xp+1, . . . , xm} ∪ X(x,B).
The elements of X(x,B) are the cluster variables of A(x, B).
n+1, . . . , x±1
{x±1
We call (y, C) a seed of A(x, B) if (y, C) ∼ (x, B). In this case, for any 1 ≤ k ≤ n
we call (yk, µ(y,C)(yk)) an exchange pair of A(x, B). Furthermore, the m-tuple y is
a cluster of A(x, B), and monomials of the form ya1
m with ai ≥ 0 for all i
are called cluster monomials of A(x, B).
2 · · · yam
1 ya2
Note that for any cluster y of A(x, B) we have yi = xi for all n+1 ≤ i ≤ m. These
m − n elements are the coefficients of A(x, B). There are no invertible coefficients
if p = n.
4
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Clearly, for any two seeds of the form (x, B) and (y, B) there is an algebra iso-
morphism η : A(x, B) → A(y, B) with η(xi) = yi for all 1 ≤ i ≤ m, which respects
the exchange relations. Furthermore, if (x, B) and (y, C) are mutation equivalent
seeds, then A(x, B) = A(y, C) and we have K(x1, . . . , xm) = K(y1, . . . , ym).
1.3. Trivial cluster algebras and connectedness of exchange matrices. Note
that we always assume m > 1. For m = 1 we would get the trivial cluster algebra
A(x, B) with exactly two cluster variables, namely x1 and x′
1 (1+
1). In particular, for K 6= Z, both cluster variables are invertible in A(x, B), and
A(x, B) is just the Laurent polynomial ring K[x±1
1 := µ(x,B)(x1) = x−1
1 ].
Furthermore, for any seed (x, B) of F the exchange matrix B is by definition
connected. For non-connected B one could write A(x, B) as a product A(x1, B1) ×
A(x2, B2) of smaller cluster algebras and study the factors A(xi, Bi) separately. The
connectedness assumption also ensures that there are no exchange relations of the
form xkx′
k = 1 + 1.
1.4. The Laurent phenomenon. It follows by induction from the exchange rela-
tions that for any cluster y of A(x, B), any cluster variable z of A(x, B) is of the
form
z =
f
g
,
where f, g ∈ N[y1, . . . , ym] are integer polynomials in the cluster variables y1, . . . , ym
with non-negative coefficients. For any seed (x, B) of F let
Lx := K[x±1
p , xp+1, . . . , xm]
1 , . . . , x±1
n , x±1
n+1, . . . , x±1
be the localization of K[x1, . . . , xm] at x1x2 · · · xp, and let
Lx,Z := Z[x±1
1 , . . . , x±1
n , xn+1, . . . , xm]
be the localization of Z[x1, . . . , xm] at x1x2 · · · xn. We consider Lx and Lx,Z as
subrings of the field F . The following remarkable result, known as the Laurent
phenomenon, is due to Fomin and Zelevinsky and is our key tool to derive some ring
theoretic properties of cluster algebras.
Theorem 1.1 ([FZ1, Theorem 3.1], [FZ2, Proposition 11.2]). For each seed (x, B)
of F we have
and
A(x, B) ⊆ A(x, B) := \(y,C)∼(x,B)
Ly
X(x,B) ⊂ \(y,C)∼(x,B)
Ly,Z.
The algebra A(x, B) is called the upper cluster algebra associated to (x, B), com-
pare [BFZ, Section 1].
FACTORIAL CLUSTER ALGEBRAS
5
1.5. Upper bounds. For a seed (x, B) and 1 ≤ k ≤ n let (xk, Bk) := µk(x, B).
Berenstein, Fomin and Zelevinsky [BFZ] called
n
the upper bound of A(x, B). They prove the following:
U(x, B) := Lx ∩
Lxk
\k=1
Theorem 1.2 ([BFZ, Corollary 1.9]). Let (x, B) and (y, C) be mutation equivalent
seeds of F .
In
particular, we have A(x, B) = U(x, B).
If B has maximal rank and p = m, then U(x, B) = U(y, C).
For clusters y and z of A(x, B) define
U(y, z) := Ly ∩ Lz.
1.6. Acyclic cluster algebras. Let (x, B) be a seed of F with B = (bij). Let
Σ(B) be the quiver with vertices 1, . . . , n, and arrows i → j for all 1 ≤ i, j ≤ n
with bij > 0, compare [BFZ, Section 1.4]. So Σ(B) encodes the sign-pattern of the
principal part B◦ of B.
The seed (x, B) and B are called acyclic if Σ(B) does not contain any oriented
cycle. The cluster algebra A(x, B) is acyclic if there exists an acyclic seed (y, C)
with (y, C) ∼ (x, B).
1.7. Skew-symmetric exchange matrices and quivers. Let B = (bij) be a
matrix in Mm,n(Z) such that B◦ is skew-symmetric. Let Γ(B) be the quiver with
vertices 1, . . . , m and bij arrows i → j if bij > 0, and −bij arrows j → i if bij < 0.
Thus given Γ(B), we can recover B. In the skew-symmetric case one often works
with quivers and their mutations instead of exchange matrices.
1.8. Main results. For a ring R with 1, let R× be the set of invertible elements in
R. Non-zero rings without zero divisors are called integral domains. A non-invertible
element a in an integral domain R is irreducible if it cannot be written as a product
a = bc with b, c ∈ R both non-invertible. Cluster algebras are integral domains,
since they are by definition subrings of fields.
Theorem 1.3. For any seed (x, B) of F the following hold:
(i) We have A(x, B)× =(cid:8)λxan+1
(ii) Any cluster variable in A(x, B) is irreducible.
p λ ∈ K ×, ai ∈ Z(cid:9).
n+1 · · · xap
For elements a, b in an integral domain R we write ab if there exists some c ∈ R
with b = ac. A non-invertible element a in a commutative ring R is prime if whenever
abc for some b, c ∈ R, then ab or ac. Every prime element is irreducible, but the
converse is not true in general. Non-zero elements a, b ∈ R are associate if there is
some unit c ∈ R× with a = bc. An integral domain R is factorial if the following
hold:
(i) Every non-zero non-invertible element r ∈ R can be written as a product
r = a1 · · · as of irreducible elements ai ∈ R.
6
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
(ii) If a1 · · · as = b1 · · · bt with ai, bj ∈ R irreducible for all i and j, then s = t
and there is a bijection π : {1, . . . , s} → {1, . . . , s} such that ai and bπ(i) are
associate for all 1 ≤ i ≤ s.
For example, any polynomial ring is factorial.
elements are prime.
In a factorial ring, all irreducible
Two clusters y and z of a cluster algebra A(x, B) are disjoint if {y1, . . . , yn} ∩
{z1, . . . , zn} = ∅.
The next result gives a useful criterion when a cluster algebra is a factorial ring.
Theorem 1.4. Let y and z be disjoint clusters of A(x, B). If there is a subalgebra
U of A(x, B), such that U is factorial and
{y1, . . . , yn, z1, . . . , zn, x±1
n+1, . . . , x±1
p , xp+1, . . . , xm} ⊂ U,
then
In particular, A(x, B) is factorial and all cluster variables are prime.
U = A(x, B) = U(y, z).
We obtain the following corollary on upper bounds of factorial cluster algebras.
Corollary 1.5. Assume that A(x, B) is factorial.
(i) If y and z are disjoint clusters of A(x, B), then A(x, B) = U(y, z).
(ii) For any (y, C) ∼ (x, B) we have A(x, B) = U(y, C).
In Section 7 we apply the above results to show that many cluster algebras are
In Section 8 we discuss some further applications concerning
polynomial rings.
the dual of Lusztig's semicanonical basis and monoidal categorifications of cluster
algebras.
1.9. Factoriality and maximal rank. In Section 6.1 we give examples of cluster
algebras A(x, B), which are not factorial.
In these examples, B does not have
maximal rank.
After we presented our results at the Abel Symposium in Balestrand in June 2011,
Zelevinsky asked the following question:
Problem 1.6. Suppose (x, B) is a seed of F such that B has maximal rank. Does
it follow that A(x, B) is factorial?
After we circulated a first version of this article, Philipp Lampe [La] discovered
an example of a non-factorial cluster algebra A(x, B) with B having maximal rank.
With his permission, we explain a generalization of his example in Section 6.2.
2. Invertible elements in cluster algebras
In this section we prove Theorem 1.3(i), classifying the invertible elements of
cluster algebras.
The following lemma is straightforward and well-known.
FACTORIAL CLUSTER ALGEBRAS
7
Lemma 2.1. For any seed (x, B) of F we have
L ×
x = {λxa1
1 · · · xap
p λ ∈ K ×, ai ∈ Z}.
Theorem 2.2. For any seed (x, B) of F we have
A(x, B)× =(cid:8)λxan+1
n+1 · · · xap
p λ ∈ K ×, ai ∈ Z(cid:9) .
Proof. Let u be an invertible element in A(x, B), and let (y, C) be any seed of
A(x, B). By the Laurent phenomenon Theorem 1.1 we know that A(x, B) ⊆ Ly. It
follows that u is also invertible in Ly. Thus by Lemma 2.1 there are a1, . . . , ap ∈ Z
and λ ∈ K × such that u = λM, where
M = ya1
1 · · · yak
k · · · yap
p .
If all ai with 1 ≤ i ≤ n are zero, we are done. To get a contradiction, assume that
there is some 1 ≤ k ≤ n with ak 6= 0.
Let y∗
k := µ(y,C)(yk). Again the Laurent phenomenon yields b1, . . . , bp ∈ Z and
ν ∈ K × such that
u = νyb1
1 · · · ybk−1
k−1 (y∗
k)bk ybk+1
k+1 · · · ybp
p .
Without loss of generality let bk ≥ 0. (Otherwise we can work with u−1 instead of
u.)
If bk = 0, we get
λya1
1 · · · yak
k · · · yap
p = νyb1
1 · · · ybk−1
k−1 ybk+1
k+1 · · · ybp
p ,
where λ, ν ∈ K ×. This is a contradiction, because ak 6= 0 and y1, . . . , ym are alge-
braically independent, and therefore Laurent monomials in y1, . . . , ym are linearly
independent in F .
Next, assume that bk > 0. By definition we have
y∗
k = M1 + M2
with
M1 = y−1
ycik
i
k Ycik>0
and M2 = y−1
y−cik
i
,
k Ycik<0
where the products run over the positive, respectively negative, entries in the kth
column of the matrix C.
Thus we get an equality of the form
(2)
u = λM = ν(yb1
1 · · · ybk−1
k−1 )(M1 + M2)bk (ybk+1
k+1 · · · ybp
p ).
We know that M1 6= M2. (Here we use that m > 1 and that exchange matrices are
by definition connected. Otherwise, one could get exchange relations of the form
xkx′
k = 1 + 1.) Thus the right-hand side of Equation (2) is a non-trivial linear
combination of bk + 1 ≥ 2 pairwise different Laurent monomials in y1, . . . , ym. This
is again a contradiction, since y1, . . . , ym are algebraically independent.
(cid:3)
Corollary 2.3. For any seed (x, B) of F the following hold:
8
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
(i) Let y and z be non-zero elements in A(x, B). Then y and z are associate if
and only if there exist an+1, . . . , ap ∈ Z and λ ∈ K × with
y = λxan+1
n+1 · · · xap
p z.
(ii) Let y and z be cluster variables of A(x, B). Then y and z are associate if
and only if y = z.
Proof. Part (i) follows directly from Theorem 2.2. To prove (ii), let y and z be
clusters of A(x, B). Assume yi and zj are associate for some 1 ≤ i, j ≤ n. By (i)
there are an+1, . . . , ap ∈ Z and λ ∈ K × with yi = λxan+1
p zj. By Theorem 1.1
we know that there exist b1, . . . , bn ∈ Z and a polynomial f in Z[z1, . . . , zm] with
n+1 · · · xap
yi =
f
zb1
1 · · · zbn
n
n+1
· · · x−ap
and f is not divisible by any z1, . . . , zn. The polynomial f and b1, . . . , bn are uniquely
determined by yi. It follows that λ ∈ Z and an+1, . . . , ap ≥ 0. But we also have
zj = λ−1x−an+1
yi. Reversing the role of yi and zj we get −an+1, . . . , −ap ≥ 0
and λ−1 ∈ Z. This implies yi = zj or −yi = zj. By the remark at the beginning
of Section 1.4 we know that zj = f /g for some f, g ∈ N[y1, . . . , ym]. Assume that
−yi = zj. We get zj = −yi = f /g and therefore f + yig = 0. This is a contradiction
to the algebraic independence of y1, . . . , ym. Thus we proved (ii).
(cid:3)
p
We thank Giovanni Cerulli Irelli for helping us with the final step of the proof of
Corollary 2.3(ii).
Two clusters y and z of a cluster algebra A(x, B) are non-associate if there are
no 1 ≤ i, j ≤ n such that yi and zj are associate.
Corollary 2.4. For clusters y and z of A(x, B) the following are equivalent:
(i) The clusters y and z are non-associate.
(ii) The clusters y and z are disjoint.
Proof. Non-associate clusters are obviously disjoint. The converse follows directly
from Corollary 2.3(ii).
(cid:3)
3. Irreducibility of cluster variables
In this section we prove Theorem 1.3(ii). The proof is very similar to the proof
of Theorem 2.2.
Theorem 3.1. Let (x, B) be a seed of F . Then any cluster variable in A(x, B) is
irreducible.
Proof. Let (y, C) be any seed of A(x, B). We know from Theorem 2.2 that the
cluster variables of A(x, B) are non-invertible in A(x, B).
Assume that yk is not irreducible for some 1 ≤ k ≤ n. Thus yk = y′
k for some
k in A(x, B). Since yk is invertible in Ly, we know
non-invertible elements y′
k and y′′
ky′′
FACTORIAL CLUSTER ALGEBRAS
9
k and y′′
that y′
and λ′, λ′′ ∈ K × with
k are both invertible in Ly. Thus by Lemma 2.1 there are ai, bi ∈ Z
Since yk = y′
1 · · · yas
k = λ′ya1
y′
ky′′
k, we get as + bs = 0 for all s 6= k and ak + bk = 1.
k = λ′′yb1
y′′
s · · · yap
p
1 · · · ybs
and
s · · · ybp
p .
Assume that as = 0 for all 1 ≤ s ≤ n with s 6= k. Then y′
and y′′
then y′′
k ybn+1
k = λ′′ybk
k is invertible in A(x, B). In both cases we get a contradiction.
p . If ak ≤ 0, then y′
n+1 · · · ybp
n+1 · · · yap
p
k is invertible in A(x, B), and if ak > 0,
k = λ′yak
k yan+1
Next assume as 6= 0 for some 1 ≤ s ≤ n with s 6= k. Let y∗
s := µ(y,C)(ys). Thus
we have
with
y∗
s = M1 + M2
M1 = y−1
ycis
i
and M2 = y−1
s Ycis>0
y−cis
i
,
s Ycis<0
where the products run over the positive, respectively negative, entries in the sth
column of the matrix C.
Since s 6= k, we see that yk and therefore also y′
k and y′′
k are invertible in Lµs(y,C).
Thus by Lemma 2.1 there are ci, di ∈ Z and ν ′, ν ′′ ∈ K × with
k = ν ′yc1
y′
1 · · · ycs−1
s−1 (y∗
s)csycs+1
s+1 · · · ycp
p
and
k = ν ′′yd1
y′′
1 · · · yds−1
s−1 (y∗
s )dsyds+1
s+1 · · · ydp
p .
Note that cs + ds = 0. Without loss of generality we assume that cs ≥ 0. (If cs < 0,
we continue to work with y′′
k = λ′ya1
y′
k.) If cs = 0, we get
p = ν ′yc1
k instead of y′
1 · · · yas
s · · · yap
s · · · ycp
p .
1 · · · y0
This is a contradiction, since as 6= 0 and y1, . . . , ym are algebraically independent.
If cs > 0, then
s+1 · · · yap
p
k = λ′ya1
y′
= ν ′yc1
= ν ′yc1
s−1 yas
s+1 · · · ycp
s−1 (y∗
p
s−1 (M1 + M2)csycs+1
s+1 · · · ycp
p .
We know that M1 6= M2. Thus the Laurent monomial y′
k is a non-trivial linear
combination of cs + 1 ≥ 2 pairwise different Laurent monomials in y1, . . . , ym, a
contradiction.
(cid:3)
1 · · · yas−1
1 · · · ycs−1
1 · · · ycs−1
s yas+1
s )csycs+1
Note that the coefficients xp+1, . . . , xm of A(x, B) are obviously irreducible in Lx.
Since A(x, B) ⊆ Lx, they are also irreducible in A(x, B).
4. Factorial cluster algebras
4.1. A factoriality criterion. This section contains the proofs of Theorem 1.4 and
Corollary 1.5.
Theorem 4.1. Let y and z be disjoint clusters of A(x, B), and let U be a factorial
subalgebra of A(x, B) such that
{y1, . . . , yn, z1, . . . , zn, x±1
n+1, . . . , x±1
p , xp+1, . . . , xm} ⊂ U.
10
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Then we have
U = A(x, B) = U(y, z).
Proof. Let u ∈ U(y, z) = Ly ∩ Lz. Thus we have
u =
f
2 · · · yap
1 ya2
ya1
p
=
g
2 · · · zbp
p
1 zb2
zb1
,
where f is a polynomial in y1, . . . , ym, and g is a polynomial in z1, . . . , zm, and
ai, bi ≥ 0 for all 1 ≤ i ≤ p. By the Laurent phenomenon it is enough to show that
u ∈ U.
Since yi, zi ∈ U for all 1 ≤ i ≤ m, we get the identity
2 · · · yan
n+1 · · · zbp
p = gya1
2 · · · zbn
n zbn+1
1 ya2
1 zb2
f zb1
n yan+1
n+1 · · · yap
p
in U.
By Theorem 3.1 the cluster variables yi and zi with 1 ≤ i ≤ n are irreducible in
A(x, B). In particular, they are irreducible in the subalgebra U of A(x, B). The
elements yan+1
p are units in U. (Recall that xi = yi = zi for
all n + 1 ≤ i ≤ m.)
p and zbn+1
n+1 · · · yap
n+1 · · · zbp
The clusters y and z are disjoint. Now Corollary 2.4 implies that the elements
yi and zj are non-associate for all 1 ≤ i, j ≤ n. Thus, by the factoriality of U, the
monomial ya1
n divides f in U. In other words there is some h ∈ U with
f = hya1
2 · · · yan
1 ya2
1 ya2
u =
2 · · · yan
n . It follows that
1 ya2
hya1
f
2 · · · yap
ya1
1 ya2
ya1
1 ya2
n+1, . . . , y±1
=
p
Since h ∈ U and y±1
2 · · · yan
n
2 · · · yap
p
=
h
n+1 · · · yap
yan+1
p
= hy−an+1
n+1
· · · y−ap
p
.
p ∈ U, we get u ∈ U. This finishes the proof.
(cid:3)
Corollary 4.2. Assume that A(x, B) is factorial.
(i) If y and z are disjoint clusters of A(x, B), then A(x, B) = U(y, z).
(ii) For any (y, C) ∼ (x, B) we have A(x, B) = U(y, C).
Proof. Part (i) follows directly from Theorem 1.4. To prove part (ii), assume
(y, C) ∼ (x, B) and let u ∈ U(y, C). For 1 ≤ k ≤ n let (yk, Ck) := µk(y, C)
and y∗
k := µ(y,C)(yk). We get
u =
f
k · · · yap
1 · · · yak
ya1
p
=
fk
k)bk · · · ybp
yb1
1 · · · (y∗
p
for a polynomial f in y1, . . . , yk, . . . , ym, a polynomial fk in y1, . . . , y∗
ai, bi ≥ 0. This yields an equality
1 · · · (y∗
p = fkya1
k)bk · · · ybp
1 · · · yak
k · · · yap
f yb1
(3)
p
k, . . . , ym, and
in A(x, B). Now we argue similarly as in the proof of Theorem 4.1. The cluster
variables y1, . . . , yn, y∗
n are obviously pairwise different. Now Corollary 2.3(ii)
implies that they are pairwise non-associate, and by Theorem 3.1 they are irreducible
in A(x, B). Thus by the factoriality of A(x, B), Equation (3) implies that yak
k divides
1, . . . , y∗
FACTORIAL CLUSTER ALGEBRAS
11
f in A(x, B). Since this holds for all 1 ≤ k ≤ n, we get that ya1
f in A(x, B). It follows that u ∈ A(x, B).
n divides
(cid:3)
1 · · · yak
k · · · yan
4.2. Existence of disjoint clusters. One assumption of Theorem 4.1 is the exis-
tence of disjoint clusters in A(x, B). We can prove this under a mild assumption.
But it should be true in general.
Proposition 4.3. Assume that the cluster monomials of A(x, B) are linearly inde-
pendent. Let (y, C) be a seed of A(x, B), and let
Then the clusters y and z are disjoint.
(z, D) := µn · · · µ2µ1(y, C).
Proof. Set (y[0], C[0]) := (y, C), and for 1 ≤ k ≤ n let (y[k], C[k]) := µk(y[k −
1], C[k − 1]) and (y1[k], . . . , ym[k]) := y[k]. We claim that
{y1[k], . . . , yk[k]} ∩ {y1, . . . , yn} = ∅.
For k = 1 this is straightforward. Thus let k ≥ 2, and assume that our claim is
true for k − 1. To get a contradiction, assume that yk[k] = yj for some 1 ≤ j ≤ n.
(By the induction assumption we know that {y1[k], . . . , yk−1[k]} ∩ {y1, . . . , yn} = ∅,
since yi[k] = yi[k − 1] for all 1 ≤ i ≤ k − 1.)
We have y[k] = (y1[k], . . . , yk[k], yk+1, . . . , ym). Since y1[k], . . . , yk[k], yk+1, . . . , ym
are algebraically independent and yk[k]
6= yk, we get 1 ≤ j ≤ k − 1. Since
(y[j], C[j]) = µj(y[j − 1], C[j − 1]), it follows that (yj[j − 1], yj[j]) is an exchange
pair of A(x, B). Next, observe that yk[k] = yj = yj[j − 1] and yj[k] = yj[j]. Thus
yj[j −1] and yj[j] are both contained in {y1[k], . . . , ym[k]}, and therefore yj[j −1]yj[j]
is a cluster monomial. The corresponding exchange relation gives a contradiction to
the linear independence of cluster monomials.
(cid:3)
Fomin and Zelevinsky [FZ3, Conjecture 4.16] conjecture that the cluster monomi-
als of A(x, B) are always linearly independent. Under the assumptions that B has
maximal rank and that B◦ is skew-symmetric, the conjecture follows from [DWZ,
Theorem 1.7].
5. The divisibility group of a cluster algebra
Let R be an integral domain, and let Frac(R) be the field of fractions of R. Set
Frac(R)∗ := Frac(R) \ {0}. The abelian group
is the divisibility group of R.
G(R) := (Frac(R)∗/R×, ·)
For g, h ∈ Frac(R)∗ let g ≤ h provided hg−1 ∈ R. This relation is reflexive and
transitive and it induces a partial ordering on G(R).
Let I be a set. The abelian group (Z(I), +) is equipped with the following partial
ordering: We set (xi)i∈I ≤ (yi)i∈I if xi ≤ yi for all i. (By definition, the elements in
Z(I) are tuples (xi)i∈I of integers xi such that only finitely many xi are non-zero.)
12
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
There is the following well-known criterion for the factoriality of R, see for example
[C, Section 2].
Proposition 5.1. For an integral domain R the following are equivalent:
(i) R is factorial.
(ii) There is a set I and a group isomorphism
φ : G(R) → Z(I)
such that for all g, h ∈ G(R) we have g ≤ h if and only if φ(g) ≤ φ(h).
Not all cluster algebras A(x, B) are factorial, but at least one part of the above
factoriality criterion is satisfied:
Proposition 5.2. For any seed (x, B) of F the divisibility group G(A(x, B)) is
isomorphic to Z(I), where
I := {f ∈ K[x1, . . . , xm] f is irreducible and f 6= xi for n + 1 ≤ i ≤ p} /K ×
is the set of irreducible polynomials unequal to any xn+1, . . . , xp in K[x1, . . . , xm] up
to non-zero scalar multiples.
Proof. By the Laurent phenomenon and the definition of a seed we get
Frac(A(x, B)) = Frac(Lx) = K(x1, . . . , xm).
Furthermore, by Theorem 2.2 we have
A(x, B)× = {λxan+1
n+1 · · · xap
p λ ∈ K ×, ai ∈ Z}.
Any element in K(x1, . . . , xm) is of the form f1 · · · fsg−1
t with fi, gj irreducible
1
in K[x1, . . . , xm]. Using that the polynomial ring K[x1, . . . , xm] is factorial, and
working modulo A(x, B)× yields the result.
(cid:3)
· · · g−1
6. Examples of non-factorial cluster algebras
6.1. For a matrix A ∈ Mm,n(Z) and 1 ≤ i ≤ n let ci(A) be the ith column of A.
Proposition 6.1. Let (x, B) be a seed of F . Assume that ck(B) = cs(B) or ck(B) =
−cs(B) for some k 6= s with bks = 0. Then A(x, B) is not factorial.
Proof. Define (y, C) := µk(x, B) and (z, D) := µs(y, C). We get
yk = zk = x−1
k (M1 + M2),
where
M1 := Ybik>0
xbik
i
and M2 := Ybik<0
x−bik
i
.
By the mutation rule, we have ck(C) = −ck(B), and since bks = 0, we get cs(C) =
cs(B). Since cs(B) = ck(B) or cs(B) = −ck(B), this implies that
zs = x−1
s (M1 + M2).
FACTORIAL CLUSTER ALGEBRAS
13
The cluster variables xk, xs, zk, zs are pairwise different. Thus they are pairwise non-
associate by Corollary 2.3(ii), and by Theorem 3.1 they are irreducible in A(x, B).
Obviously, we have
xkzk = xszs.
Thus A(x, B) is not factorial.
(cid:3)
To give a concrete example of a cluster algebra, which is not factorial, assume
m = n = p = 3, and let B ∈ Mm,n(Z) be the matrix
B =
0 −1
1
0
0
0 −1
1
0
.
The matrix B obviously satisfies the assumptions of Proposition 6.1. Note that
B = B◦ is skew-symmetric, and that Γ(B) is the quiver
Thus A(x, B) is a cluster algebra of Dynkin type A3. (Cluster algebras with finitely
many cluster variables are classified via Dynkin types, for details see [FZ2].)
3
/ 2
/ 1.
Define (z, D) := µ3µ1(x, B). We get z1 = x−1
1 (1 + x2), z3 = x−1
3 (1 + x2) and
therefore x1z1 = x3z3.
Clearly, the cluster variables x1, x3, z1, z3 are pairwise different. Using Corol-
lary 2.3(ii) we get that x1, x3, z1, z3 are pairwise non-associate, and by Theorem 3.1
they are irreducible. Thus A(x, B) is not factorial.
6.2. The next example is due to Philipp Lampe.
Zelevinsky's Question 1.6.
It gives a negative answer to
Proposition 6.2 ([La]). Let K = C, m = n = 2 and
Then A(x, B) is not factorial.
B =(cid:18)0 −2
0 (cid:19) .
2
The proof of the following result is a straightforward generalization of Lampe's
proof of Proposition 6.2.
Proposition 6.3. Let (x, B) be a seed of F . Assume that there exists some 1 ≤
k ≤ n such that the polynomial X d + Y d is not irreducible in K[X, Y ], where d :=
gcd(b1k, . . . , bmk) is the greatest common divisor of b1k, . . . , bmk. Then A(x, B) is
not factorial.
Proof. Let X d + Y d = f1 · · · ft, where the fj are irreducible polynomials in K[X, Y ].
Since X d + Y d is not irreducible in K[X, Y ], we have t ≥ 2. Let yk := µ(x,B)(xk).
The corresponding exchange relation is
xkyk = Ybik>0
xbik
i + Ybik<0
x−bik
i
= M d + N d =
fj(M, N),
t
Yj=1
/
/
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
14
where
xbik/d
i
M := Ybik>0
and N := Ybik<0
x−bik/d
i
.
Clearly, each fj(M, N) is contained in A(x, B). To get a contradiction, assume that
A(x, B) is factorial. By Theorem 2.2 none of the elements fj(M, N) is invertible in
A(x, B). Since A(x, B) is factorial, each fj(M, N) is equal to a product f1j · · · faj j,
where the fij are irreducible in A(x, B) and aj ≥ 1. By Theorem 3.1 the cluster
variables xk and yk are irreducible in A(x, B). It follows that a1 + · · · + at = 2, since
A(x, B) is factorial. This implies t = 2 and a1 = a2 = 1. In particular, f1(M, N)
and f2(M, N) are irreducible in A(x, B), and we have xkyk = f1(M, N)f2(M, N).
For j = 1, 2 the elements xk and fj(M, N) cannot be associate, since fj(M, N) is
just a K-linear combination of monomials in {x1, . . . , xm} \ {xk}.
(Here we use
Corollary 2.3(i) and the fact that bkk = 0.) This is a contradiction to the factoriality
of A(x, B).
(cid:3)
Note that a polynomial of the form X d + Y d is irreducible if and only if X d + 1
is irreducible.
Corollary 6.4. Let K = C, m = n = 2 and
with c ≥ 1 and d ≥ 2. Then A(x, B) is not factorial.
B =(cid:18)0 −c
0 (cid:19)
d
Proof. For k = 1 the assumptions of Proposition 6.3 hold. (We have gcd(0, d) = d,
and the polynomial X d + 1 is not irreducible in C[X].)
(cid:3)
Corollary 6.5. Let m = n = 2 and
B =(cid:18)0 −c
0 (cid:19)
d
with c ≥ 1 and d ≥ 3 an odd number. Then A(x, B) is not factorial.
Proof. For k = 1 the assumptions of Proposition 6.3 hold. (We have gcd(0, d) = d,
and for odd d we have
X d + 1 = (X + 1) d−1
Xj=0
(−1)jX j! .
Thus X d + 1 is not irreducible in K[X].)
(cid:3)
7. Examples of factorial cluster algebras
FACTORIAL CLUSTER ALGEBRAS
15
7.1. Cluster algebras of Dynkin type A as polynomial rings. Assume m =
n + 1 = p + 1, and let B ∈ Mm,n(Z) be the matrix
0 −1
1
0 −1
1
0
1
. . .
. . . −1
. . .
0 −1
0
1
1
B =
.
Obviously, B◦ is skew-symmetric, Γ(B) is the quiver
m
/ · · ·
/ 2
/ 1,
and A(x, B) is a cluster algebra of Dynkin type An. Note that A(x, B) has exactly
one coefficient, and that this coefficient is non-invertible.
Let (x[0], B[0]) := (x, B). For each 1 ≤ i ≤ m − 1 we define inductively a seed by
(x[i], B[i]) := µm−i · · · µ2µ1(x[i − 1], B[i − 1]).
For 0 ≤ i ≤ m − 1 set (x1[i], . . . , xm[i]) := x[i].
For simplicity we define x0[i] := 1 and x−1[i] := 0 for all i.
Lemma 7.1. For 0 ≤ i ≤ m − 2, 1 ≤ k ≤ m − 1 − i and 0 ≤ j ≤ i we have
(4)
µ(x[i],B[i])(xk[i]) =
xk−1[i] + xk+1[i]
xk[i]
=
xk−1+j[i − j] + xk+1+j[i − j]
xk+j[i − j]
.
Proof. The first equality follows from the definition of (x[i], B[i]) and the mutation
rule. The second equality is proved by induction on i.
(cid:3)
Corollary 7.2. For 0 ≤ i ≤ m − 2 we have
(5)
(6)
xi+2 = x1[i + 1]xi+1 − xi,
xi+1[1] = x1[i + 1]xi[1] − xi−1[1].
Proof. Equation (5) follows from (4) for k = 1 and j = i. The case k = 1 and
j = i − 1 yields Equation (6).
(cid:3)
Proposition 7.3. The elements x1[0], x1[1], . . . , x1[m − 1] are algebraically indepen-
dent and
K[x1[0], x1[1], . . . , x1[m − 1]] = A(x, B).
In particular, the cluster algebra A(x, B) is a polynomial ring in m variables.
Proof. It follows from Equation (5) that
x1[i] ∈ K(x1, . . . , xi+1) \ K(x1, . . . , xi)
for all 1 ≤ i ≤ m − 1. Since x1, . . . , xm are algebraically independent, this implies
that x1[0], x1[1], . . . , x1[m − 1] are algebraically independent as well. Thus
U := K[x1[0], x1[1], . . . , x1[m − 1]]
/
/
/
16
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
In particular, U is factorial. Equation (5)
is a polynomial ring in m variables.
implies that x1, . . . , xm ∈ U, and Equation (6) yields that x1[1], . . . , xm[1] ∈ U.
Clearly, the clusters x and x[1] are disjoint. Thus the assumptions of Theorem 4.1
are satisfied, and we get U = A(x, B).
(cid:3)
The cluster algebra A(x, B) as defined above has been studied by several people.
It is related to a T -system of Dynkin type A1 with a certain boundary condition, see
[DK]. Furthermore, for K = C the cluster algebra A(x, B) is naturally isomorphic to
the complexified Grothendieck ring of the category Cn of finite-dimensional modules
of level n over the quantum loop algebra of Dynkin type A1, see [HL, N2]. It is well
known, that A(x, B) is a polynomial ring. We just wanted to demonstrate how to
use Theorem 4.1 in practise.
7.2. Acyclic cluster algebras as polynomial rings. Let C = (cij) ∈ Mn,n(Z)
be a generalized Cartan matrix, i.e. C is symmetrizable, cii = 2 for all i and cij ≤ 0
for all i 6= j.
Assume that m = 2n = 2p, and let (x, B) be a seed of F , where B = (bij) ∈
M2n,n(Z) is defined as follows: For 1 ≤ i ≤ 2n and 1 ≤ j ≤ n let
bij :=
Thus we have
0
−cij
cij
1
ci−n,j
0
B =
if i = j,
if 1 ≤ i < j ≤ n,
if 1 ≤ j < i ≤ n,
if i = n + j,
if n + 1 ≤ i ≤ 2n and i − n < j,
if n + 1 ≤ i ≤ 2n and i − n > j.
0
b21
b12
0
b13
b23
b32
...
bn2
0
b31
...
. . .
bn1
· · ·
1 −b12 −b13
0
−b23
1
0
...
0
0
...
0
1
. . .
0
· · ·
· · ·
. . .
. . .
bn,n−1
b1n
b2n
...
bn−1,n
0
−b1n
−b2n
...
· · ·
· · ·
. . .
. . . −bn−1,n
0
1
.
Clearly, (x, B) is an acyclic seed. Namely, if i → j is an arrow in Σ(B), then i < j.
Up to simultaneous reordering of columns and rows, each acyclic skew-symmetrizable
matrix in Mn,n(Z) is of the form B◦ with B defined as above. Note that A(x, B)
has exactly n coefficients, and that all these coefficients are non-invertible.
For 1 ≤ i ≤ n let
(x[1], B[1]) := µn · · · µ2µ1(x, B)
FACTORIAL CLUSTER ALGEBRAS
17
and (x1[1], . . . , x2n[1]) := x[1]. Let B0 := B, and for 1 ≤ i ≤ n let Bi := µi(Bi−1).
Thus we have Bn = B[1]. It is easy to work out the matrices Bi explicitly: The
matrix Bi is obtained from Bi−1 by changing the sign in the ith row and the ith
column of the principal part B◦
i−1. Furthermore, the (n + i)th row
(0, . . . , 0, 1, −bi,i+1, −bi,i+2, . . . , −bin)
of Bi−1 gets replaced by
(−bi1, −bi2, . . . , −bi,i−1, −1, 0, . . . , 0).
If we write N+ (resp. N−) for the upper (resp. lower) triangular part of B◦, we
get
B =
1
N−
1
N+
1
−N+
. . .
. . .
0
1
and B[1] =
1
N−
−1
−N−
. . .
. . .
N+
1
0
−1
.
In particular, the principal part B◦ of B is equal to the principal part B[1]◦ of B[1].
Now the definition of seed mutation yields
(7)
for 1 ≤ k ≤ n.
xk[1] = x−1
k xn+k +
xi[1]bik
k−1
Yi=1
x−bik
i !
n
Yi=k+1
Proposition 7.4. The elements x1, . . . , xn, x1[1], . . . , xn[1] are algebraically inde-
pendent and
K[x1, . . . , xn, x1[1], . . . , xn[1]] = A(x, B).
In particular, the cluster algebra A(x, B) is a polynomial ring in 2n variables.
Proof. By Equation (7) and induction we have
xk[1] ∈ K(x1, . . . , xn+k) \ K(x1, . . . , xn+k−1)
for all 1 ≤ k ≤ n. It follows that x1, . . . , xn, x1[1], . . . , xn[1] are algebraically inde-
pendent, and that the clusters x and x[1] are disjoint. Let
U := K[x1, . . . , xn, x1[1], . . . , xn[1]].
Thus U is a polynomial ring in 2n variables. In particular, U is factorial. It follows
from Equation (7) that
k−1
n
(8)
xn+k = xk[1]xk −
Yi=1
xi[1]bik
x−bik
i
.
Yi=k+1
This implies xn+k ∈ U for all 1 ≤ k ≤ n. Thus the assumptions of Theorem 4.1 are
satisfied, and we can conclude that U = A(x, B).
(cid:3)
18
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Proposition 7.4 is a special case of a much more general result proved in [GLS2].
But the proof presented here is new and more elementary.
Next, we compare the basis
PGLS :=(cid:8)x[a] := xa1
1 · · · xan
n x1[1]an+1 · · · xn[1]a2n a = (a1, . . . , a2n) ∈ N2n(cid:9)
of A(x, B) resulting from Proposition 7.4 with a basis constructed by Berenstein,
Fomin and Zelevinsky [BFZ]. For 1 ≤ k ≤ n let
k := µ(x,B)(xk) = x−1
x′
k xn+k
xbik
i +
k−1
Yi=1
x−bik
i
n
Yi=k+1
xbik
n+i!
k−1
Yi=1
(9)
and set
PBFZ := {x′[a] := xa1
1 · · · xan
n+1 · · · xa2n
n xan+1
a = (a1, . . . , a3n) ∈ N3n, aka2n+k = 0 for 1 ≤ k ≤ n}.
1)a2n+1 · · · (x′
n)a3n
2n (x′
Proposition 7.5 ([BFZ, Corollary 1.21]). The set PBFZ is a basis of A(x, B).
Note that the basis PGLS is constructed by using cluster variables from two seeds,
namely (x, B) and µn · · · µ1(x, B), whereas PBFZ uses cluster variables from n + 1
seeds, namely (x, B) and µk(x, B), where 1 ≤ k ≤ n.
Now we insert Equation (8) into Equation (9) and obtain
(10) xkx′
k = xk[1]xk −
k−1
Yi=1
xi[1]bik
n
n
Yi=k+1
x−bik
i
Yi=k+1
x−bik
i ! k−1
Yi=1
Yi=1 xi[1]xi −
k−1
xbik
i +
xj[1]bji
i−1
Yj=1
x−bji
j !bik
.
n
Yj=i+1
Then we observe that the right-hand side of Equation (10) is divisible by xk and
that x′
k is a polynomial in x1, . . . , xn, x1[1], . . . , xn[1]. Thus we can express every
element of the basis PBFZ explicitely as a linear combination of vectors from the
basis PGLS.
One could use Equation (10) to get an alternative proof of Proposition 7.4 as
pointed out by Zelevinsky [Z]. Vice versa, using Propostion 7.4 yields another proof
that PBFZ is a basis.
FACTORIAL CLUSTER ALGEBRAS
19
As an illustration, for n = 3 the matrices Bi look as follows:
B0 =
B2 =
b13
b23
0
b12
0
b32
0
b21
b31
1 −b12 −b13
0
−b23
0
1
0
1
b12 −b13
0
−b23
0
b21
−b31 −b32
−1
0
−b21 −1
0
0
0
0
0
1
,
,
B1 =
B3 =
0
−b21
−b31
−1
0
0
−b12 −b13
b23
0
0
0
b32
0
1
0
−b23
1
b13
0
b12
b23
b21
0
0
b31
b32
0
−1
0
−b21 −1
0
−b31 −b32 −1
,
.
For example, for
B = B0 =
2
0
−2
0
0 −1
1 −2
0
0
0
1
0
0
1 −1
0
1
Γ(B3) :
the quivers Γ(B0) and Γ(B3) look as follows:
Γ(B0) :
4
5
_❄❄❄❄❄❄❄
/ 3
6
/ 1
⑧⑧⑧⑧⑧⑧⑧
⑧⑧⑧⑧⑧⑧⑧
_❄❄❄❄❄❄❄
_❄❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧⑧
/ 2
1
4
4❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥
4❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥
⑧⑧⑧⑧⑧⑧⑧
⑧⑧⑧⑧⑧⑧⑧
2
5
4❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥
⑧⑧⑧⑧⑧⑧⑧
3
6
The cluster algebra A(x, B) is a polynomial ring in the 6 variables x1, x2, x3 and
x1[1] =
x2[1] =
x3[1] =
,
x2
2 + x4
x1
x4
2x3 + 2x2
2x3x4 + x3x2
4 + x2
1x5
x2
1x2
,
x4
2x3 + 2x2
2x3x4 + x3x2
x2
1x2x3
4 + x2
1x5 + x2
1x2x6
.
7.3. Cluster algebras arising in Lie theory as polynomial rings. The next
class of examples can be seen as a fusion of the examples discussed in Sections 7.1
and 7.2. In the following we use the same notation as in [GLS2].
Let C ∈ Mn,n(Z) be a symmetric generalized Cartan matrix, and let g be the
associated Kac-Moody Lie algebra over K = C with triangular decomposition g =
n− ⊕ h ⊕ n, see [K].
/
o
o
/
_
_
4
4
o
o
/
_
4
o
o
20
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Let U(n)∗
gr be the graded dual of the enveloping algebra U(n) of n. To each element
w in the Weyl group W of g one can associate a subalgebra R(Cw) of U(n)∗
gr and a
cluster algebra A(Cw), see [GLS2]. Here Cw denotes a Frobenius category associated
to w, see [BIRS, GLS2].
In [GLS2] we constructed a natural algebra isomorphism
This yields a cluster algebra structure on R(Cw).
A(Cw) → R(Cw).
Let i = (ir, . . . , i1) be a reduced expression of w. In [GLS2] we studied two cluster-
tilting modules Vi = V1 ⊕· · ·⊕Vr and Ti = T1 ⊕· · ·⊕Tr in Cw, which are associated to
i. These modules yield two disjoint clusters (δV1, . . . , δVr ) and (δT1, . . . , δTr ) of R(Cw).
The exchanges matrices are of size r × (r − n). In contrast to our conventions in
this article, the n coefficients are δVk = δTk with k+ = r + 1, where k+ is defined
as in [GLS2], and none of these coefficients is invertible. Furthermore, we studied
a module Mi = M1 ⊕ · · · ⊕ Mr in Cw, which yields cluster variables δM1, . . . , δMr of
R(Cw). (These do not form a cluster.) Using methods from Lie theory we obtained
the following result.
Theorem 7.6 ([GLS2, Theorem 3.2]). The cluster algebra R(Cw) is a polynomial
ring in the variables δM1, . . . , δMr .
To obtain an alternative proof of Theorem 7.6, one can proceed as follows:
(i) Show that the cluster variables δM1, . . . , δMr are algebraically independent.
(ii) Show that for 1 ≤ k ≤ r the cluster variables δVk and δTk are polynomials in
δM1, . . . , δMr.
(iii) Apply Theorem 4.1.
Part (i) can be done easily using induction and the mutation sequence in [GLS2,
Section 13]. Part (ii) is not at all straightforward.
Let us give a concrete example illustrating Theorem 7.6. Let g be the Kac-Moody
Lie algebra associated to the generalized Cartan matrix
C =(cid:18) 2 −2
2 (cid:19) ,
−2
and let i = (2, 1, 2, 1, 2, 1, 2, 1). Then A(Cw) = A(x, Bi), where r = n + 2 = 8, x7
and x8 are the (non-invertible) coefficients, and
0
2 −1
−2
0
1 −2
2 −1
0
1 −2
2 −1
0
1 −2
Bi =
.
2 −1
0
2
0
1 −2
1 −2
1
FACTORIAL CLUSTER ALGEBRAS
21
The principal part B◦
i of Bi is skew-symmetric, and the quiver Γ(Bi) looks as follows:
8
/ 6
✁✁✁✁✁✁✁
✁✁✁✁✁✁✁
^❂❂❂❂❂❂❂
^❂❂❂❂❂❂❂
/ 5
7
/ 4
✁✁✁✁✁✁✁
✁✁✁✁✁✁✁
^❂❂❂❂❂❂❂
^❂❂❂❂❂❂❂
/ 3
/ 2
✁✁✁✁✁✁✁
✁✁✁✁✁✁✁
^❂❂❂❂❂❂❂
^❂❂❂❂❂❂❂
/ 1
Define
(x[0], B[0]) := (x, Bi),
(x[1], B[1]) := µ5µ3µ1(x[0], B[0]),
(x[3], B[3]) := µ3µ1(x[2], B[2]),
(x[5], B[5]) := µ1(x[4], B[4]),
(x[2], B[2]) := µ6µ4µ2(x[1], B[1]),
(x[4], B[4]) := µ4µ2(x[3], B[3]),
(x[6], B[6]) := µ2(x[5], B[5]),
and for 0 ≤ k ≤ 6 let (x1[k], . . . , x8[k]) := x[k].
Under the isomorphism A(Cw) → R(Cw) the cluster x[0] of A(Cw) = A(x, Bi)
corresponds to the cluster (δV1, . . . , δV8) of R(Cw), the cluster x[6] corresponds to
(δT1, . . . , δT8), and we have
x1[0] 7→ δM1,
x1[4] 7→ δM5,
x2[0] 7→ δM2,
x2[4] 7→ δM6,
x1[2] 7→ δM3,
x1[6] 7→ δM7,
x2[2] 7→ δM4,
x2[6] 7→ δM8.
By Theorem 7.6 we know that the cluster algebra A(x, Bi) is a polynomial ring in
the variables x1[0], x2[0], x1[2], x2[2], x1[4], x2[4], x1[6], x2[6].
8. Applications
8.1. Prime elements in the dual semicanonical basis. As in Section 7.3 let
C ∈ Mn,n(Z) be a symmetric generalized Cartan matrix, and let g = n− ⊕ h ⊕ n be
the associated Lie algebra.
As before let W be the Weyl group of g. To C one can also associate a prepro-
jective algebra Λ over C, see for example [GLS2, R]
Lusztig [Lu] realized the universal enveloping algebra U(n) of n as an algebra of
constructible functions on the varieties Λd of nilpotent Λ-modules with dimension
vector d ∈ Nn. He also constructed the semicanonical basis S of U(n). The elements
of S are naturally parametrized by the irreducible components of the varieties Λd.
An irreducible component Z of Λd is called indecomposable if it contains a Zariski
dense subset of indecomposable Λ-modules, and Z is rigid if it contains a rigid
Λ-module M, i.e. M is a module with Ext1
Λ(M, M) = 0.
Let S ∗ be the dual semicanonical basis of the graded dual U(n)∗
gr of U(n). The
elements ρZ in S ∗ are also parametrized by irreducible components Z of the varieties
Λd. We call ρZ indecomposable (resp. rigid ) if Z is indecomposable (resp. rigid).
An element b ∈ S ∗ is called primitive if it cannot be written as a product b = b1b2
with b1, b2 ∈ S ∗ \ {1}.
Theorem 8.1 ([GLS1, Theorem 1.1]). If ρZ is primitive, then Z is indecomposable.
/
/
/
/
^
^
/
^
^
/
^
^
22
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
Theorem 8.2 ([GLS2, Theorem 3.1]). For w ∈ W all cluster monomials of the
cluster algebra R(Cw) belong to the dual semicanonical basis S ∗ of U(n)∗
gr. More
precisely, we have
{cluster variables of R(Cw)} ⊆ {ρZ ∈ S ∗ Z is indecomposable and rigid},
{cluster monomials of R(Cw)} ⊆ {ρZ ∈ S ∗ Z is rigid}.
Combining Theorems 3.1, 7.6 and 8.2 we obtain a partial converse of Theorem 8.1.
Theorem 8.3. The cluster variables in R(Cw) are prime, and they are primitive
elements of S ∗.
Conjecture 8.4. If ρZ ∈ S ∗ is indecomposable and rigid, then ρZ is prime in U(n)∗
gr.
8.2. Monoidal categorifications of cluster algebras. Let C be an abelian tensor
category with unit object IC. We assume that C is a Krull-Schmidt category, and that
all objects in C are of finite length. Let M(C) := K0(C) be the Grothendieck ring of
C. The class of an object M ∈ C is denoted by [M]. The addition in M(C) is given
by [M]+[N] := [M ⊕N] and the multiplication is defined by [M][N] := [M ⊗N]. We
assume that [M ⊗ N] = [N ⊗ M]. (In general this does not imply M ⊗ N ∼= N ⊗ M.)
Thus M(C) is a commutative ring.
Tensoring with K over Z yields a K-algebra MK(C) := K ⊗Z K0(C) with K-basis
the classes of simple objects in C. Note that the unit object IC is simple.
A monoidal categorification of a cluster algebra A(x, B) is an algebra isomorphism
Φ : A(x, B) → MK(C),
where C is a tensor category as above, such that each cluster monomial y = ya1
1 · · · yam
m
of A(x, B) is mapped to a class [Sy] of some simple object Sy ∈ C. In particular, we
have
[Sy] = [Sy1]a1 · · · [Sym]am = [S ⊗a1
y1 ⊗ · · · ⊗ S ⊗am
ym ].
For an object M ∈ C let xM be the element in A(x, B) with Φ(xM ) = [M].
The concept of a monoidal categorification of a cluster algebra was introduced in
[HL, Definition 2.1]. But note that our definition uses weaker conditions than in
[HL].
An object M ∈ C is called invertible if [M] is invertible in MK(C). An object
M ∈ C is primitive if there are no non-invertible objects M1 and M2 in C with
M ∼= M1 ⊗ M2.
Proposition 8.5. Let Φ : A(x, B) → MK(C) be a monoidal categorification of a
cluster algebra A(x, B). Then the following hold:
(i) The invertible elements in MK(C) are
MK(C)× =(cid:8)λ[IC][Sxn+1]an+1 · · · [Sxp]ap λ ∈ K ×, ai ∈ Z(cid:9) .
(ii) Let M be an object in C such that the element xM is irreducible in A(x, B).
Then M is primitive.
FACTORIAL CLUSTER ALGEBRAS
23
Proof. Part (i) follows directly from Theorem 2.2. To prove (ii), assume that M is
not primitive. Thus there are non-invertible objects M1 and M2 in C with M ∼= M1⊗
M2. Thus in MK(C) we have [M] = [M1][M2]. Since Φ is an algebra isomorphism,
we get xM = xM1xM2 with xM1 and xM2 non-invertible in A(x, B). Since xM is
irreducible, we have a contradiction.
(cid:3)
Combining Proposition 8.5 with Theorem 3.1 we get the following result.
Corollary 8.6. Let Φ : A(x, B) → MK(C) be a monoidal categorification of a clus-
ter algebra A(x, B). For each cluster variable y of A(x, B), the simple object Sy is
primitive.
Examples of monoidal categorifications of cluster algebras can be found in [HL,
N1], see also [Le].
Acknowledgements. We thank Giovanni Cerulli Irelli, Sergey Fomin, Daniel
Labardini Fragoso, Philipp Lampe and Andrei Zelevinsky for helpful discussions.
We thank Giovanni Cerulli Irelli for carefully reading several preliminary versions
of this article. The first author likes to thank the Max-Planck Institute for Mathe-
matics in Bonn for a one year research stay in 2010/2011. The third author thanks
the Sonderforschungsbereich/Transregio SFB 45 for financial support, and all three
authors thank the Hausdorff Institute for Mathematics in Bonn for support and
hospitality.
References
[BFZ] A. Berenstein, S. Fomin, A. Zelevinsky, Cluster algebras III: Upper bounds and double
Bruhat cells, Duke Math. J. 126 (2005), no. 1, 1 -- 52.
[BIRS] A. Buan, O. Iyama, I. Reiten, J. Scott, Cluster structures for 2-Calabi-Yau categories and
unipotent groups, Compos. Math. 145 (2009), no. 4, 1035 -- 1079.
[C] P.M. Cohn, Unique factorization domains, Amer. Math. Monthly 80 (1973), 1 -- 18.
[DWZ] H. Derksen, J. Weyman, A. Zelevinsky, Quivers with potentials and their representations
II: applications to cluster algebras, J. Amer. Math. Soc. 23 (2010), no. 3, 749 -- 790.
[DK] P. Di Francesco, R. Kedem, Q-systems as cluster algebras. II. Cartan matrix of finite type
and the polynomial property, Lett. Math. Phys. 89 (2009), no. 3, 183 -- 216.
[FZ1] S. Fomin, A. Zelevinsky, Cluster algebras. I. Foundations, J. Amer. Math. Soc. 15 (2002),
no. 2, 497 -- 529.
[FZ2] S. Fomin, A. Zelevinsky, Cluster algebras. II. Finite type classification, Invent. Math. 154
(2003), no. 1, 63 -- 121.
[FZ3] S. Fomin, A. Zelevinsky, Cluster algebras: notes for the CDM-03 conference, Current devel-
opments in mathematics, 2003, 1 -- 34, Int. Press, Somerville, MA, 2003.
[GLS1] C. Geiss, B. Leclerc, J. Schroer, Semicanonical bases and preprojective algebras, Ann. Sci.
´Ecole Norm. Sup. (4) 38 (2005), no. 2, 193 -- 253.
[GLS2] C. Geiss, B. Leclerc, J. Schroer, Kac-Moody groups and cluster algebras, Adv. Math. 228
(2011), 329 -- 433.
[HL] D. Hernandez, B. Leclerc, Cluster algebras and quantum affine algebras, Duke Math. J. 154
(2010), no. 2, 265 -- 341.
[K] V. Kac, Infinite-dimensional Lie algebras. Third edition. Cambridge University Press, Cam-
bridge, 1990. xxii+400pp.
[La] P. Lampe, Email from October 8th, 2011.
24
CHRISTOF GEISS, BERNARD LECLERC, AND JAN SCHR OER
[Le] B. Leclerc, Quantum loop algebras, quiver varieties, and cluster algebras, Representations of
algebras and related topics, 117 -- 152, EMS Ser. Congr. Rep., Eur. Math. Soc., Zurich, 2011.
[Lu] G. Lusztig, Semicanonical bases arising from enveloping algebras, Adv. Math. 151 (2000),
no. 2, 129 -- 139.
[N1] H. Nakajima, Quiver varieties and cluster algebras, Kyoto J. Math. 51 (2011), no. 1, 71 -- 126.
[N2] H. Nakajima, Lecture at the workshop "Cluster algebras, representation theory, and Poisson
geometry", Banff, September 2011.
[R] C.M. Ringel, The preprojective algebra of a quiver. Algebras and modules, II (Geiranger,
1996), 467 -- 480, CMS Conf. Proc., 24, Amer. Math. Soc., Providence, RI, 1998.
[Z] A. Zelevinsky, Private communication, Balestrand, June 2011, and Email exchange, July
2011.
Christof Geiss
Instituto de Matem´aticas
Universidad Nacional Aut´onoma de M´exico
Ciudad Universitaria
04510 M´exico D.F.
M´exico
E-mail address: [email protected]
Bernard Leclerc
Universit´e de Caen
LMNO, CNRS UMR 6139
Institut Universitaire de France
14032 Caen cedex
France
E-mail address: [email protected]
Jan Schroer
Mathematisches Institut
Universitat Bonn
Endenicher Allee 60
53115 Bonn
Germany
E-mail address: [email protected]
|
1111.6141 | 1 | 1111 | 2011-11-26T08:12:19 | Prime Ideals in Noetherian Rings | [
"math.RA"
] | In this short note we study the links of certain prime ideals of a noetherian ring R. We first give the definition of a link krull symmetric noetherian ring R. We then prove theorem 9 that states that for any linked prime ideals P' and Q' of the polynomial ring R[X] where R is a link krull symmetric noetherian ring, if The prime ideal P' is extended then Q' is also an extended prime ideal of R[X]. An application of theorem 9 is then given in theorem 12 for the ring R[X] when R is assumed to be a fully bounded noetherian ring. | math.RA | math | 1
Prime Ideals in Noetherian Rings
C. L. Wangneo
# A-1, Jammu university,Rail Road Jammu,Jammu & Kashmir,India(180006).
Abstract: In this short note we study the links of certain prime ideals of a noetherian ring R. We first give
the definition of a link krull symmetric noetherian ring R. We then prove theorem 9 that states that for any
linked prime ideals P ′ and Q′ of the polynomial ring R[X] where R is a link krull symmetric noetherian ring,
if The prime ideal P ′ is extended then Q′ is also an extended prime ideal of R[X]. An application of
theorem 9 is then given in theorem 12 for the ring R[X] when R is assumed to be a fully bounded noetherian
ring.
1. Introduction: In a noetherian ring R there is a link from Q → P , for Q, P prime ideals of R if there is an
ideal A of R Such that QP ≤ A < Q ∩ P and Q ∩ P/A is torsionfree as a left R/Q and as a right R/P module. In
this paper we study the links between certain prime ideals of the polynomial ring R[X], where R is a link k-
symmetric noetherian ring. We first state the definition of a link krull symmetric(link k-symmetric for short)
noetherian ring R. We then prove the main theorem of this paper, namely theorem 9, which states that for the
linked prime ideals Q and P of the the polynomial ring R[X], where R is a link k-symmetric noetherian ring,
if the ideal P is an extended prime ideal of the ring R[X] then Q is also an extended prime ideal of the ring
R[X]. Theorem 12 then is an application of theorem 9 to the case of a polynomial ring R[X], when R is
considered to be a fully bounded noetherian ring. Moreover theorem 9 also provides easily in theorem 12 a
description of all the links between any two prime ideals of the polynomial ring R[X] over a link k-
symmetric noetherian ring.
2. Definitions and Notation: We mention that throughout all our rings
are with identity and all modules are unitary. A ring R is noetherian means that R is a right as well as a left
noetherian ring. For the definition of right Krull dimension of a right R module M we refer the reader to [3 ]
or [4]. A right module M is said to be right k-homogenous (k-homogenous for short) if every non zero sub
module of M has same krull dimension as that of M . A semi- prime ideal S of a noetherian ring R is said to
be right krull homogenous if the ring R/S is a right k-homogenous ring. A noetherian bimodule M over a ring
R is said to be krull symmetric if the right krull dimension of M equals its left krull dimesion.For the
definition of weak ideal invariance of an ideal A of a noetherian R , see[1]. For the definition of a fully
bounded noetherian (FBN for short) ring, see [3]. We now mention a few words regarding our terminology
in the present paper. For a ring R, Spec.R denotes the set of prime ideals of R and minSpec.R denotes the set
of minimal prime ideals of R. For a krull symmetric bimodule M over a ring R, M denotes its left or right
krull dimension. Finally if A is an ideal of a ring R then c(A) denotes the set of elements of R that are regular
modulo A. For a ring R we denote by N (R) the nil radical of R.
3. The Main Theorem: We start with some Lemmas which help us in proving our main theorem.
Lemma 1. Let R be a Noetherian ring and let R[X] be the polynomial ring over R in a commuting
indeterminate X. Let Q′ε spec.R[X] be a prime ideal such that Q′ ∩ R = Q. If c ε c(Q), then c ε c(Q′).
Proof. First observe that Q is a prime ideal of R. Next we consider two cases;
Case 1 Q′ = Q[X]. In this case it is not difficult to see that c ε c(Q) implies that c ε c(Q′) also.
2
Case 2 Q′ > Q[X]. In this case consider the ring R[X]/Q[X]. Then Q′/Q[X] is a nonzero prime ideal
of the ring R[X] . If we prove that c+Q[X], where c ε c(Q), is an element such that c ε c(Q′), then the
proof follows. Hence we may assume that Q[X]=0 and that R as well as R[X] are prime rings. Thus
Q′ is then a nonzero prime ideal of the ring R[X] such that Q′ ∩ R = 0. In this context then c is a nonzero
regular element of R. We now prove that c ε c(Q′). To see this, suppose f(X) ε R[X] is an element such that
f (x) is not an element of Q′ and cf(X) ε Q′. Now it follows from Goldie [2], theorem (13) that since R is a
Noetherian prime ring then the set c(0) = set of regular elements of R, forms a left and a right ore-set in R. It
is not difficult to see that c(0) is a left and a right ore- sub set of the ring R[X] too. Thus for any g(X) ε R[X],
we have that there exists an element d ε c(0) such that dg(X) ε R[X].c (after using the fact that c(0) is a left
ore-set of R[x]). Let dg(X) = h(X).c, for some h(X) ε R[X]. So we get that dg(X)f (X) = h(X)cf (X) , for all
g(X) ε R[X]. Since it is given that c f(X) ε Q′,this thus implies that dg(X)f (X)R[X] ≤ Q′, for all
g(X)εR[X]. This immediately implies that the R[X] bi-module say,
A = R[X]f [X]R[X] + Q′/Q′ is left c(0) torsion module. Since R[X] is a Noethrian ring because R is a
Noetherian ring, so the bi-module A is left and right finitely generated over R[X]. Since R[X] is a prime
ring, and A is c(0) left torsion bi-module which is also finitely generated on either side, hence there exists
an element d1εc(0) such that that d1A=0. It is not difficult to see that this implies that
d1R[X]f [X] εQ′. Since Q′ is a prime ideal of R[X], so either d1ε Q′ or f (X) ε Q′ which is not true. Thus
cf (X) εQ′ implies that f (X) ε Q′. Hence c ε c(Q′).
Lemma 2. Let R be a Noetherian ring and let R[X] be the usual polynomial ring over R in a
commuting indeterminate X. Let P be a prime ideal of R. Let Q′ ε spec.R[X] and B′ be an ideal of R[X]
such that Q′P [X] ≤ B′ < Q′ ∩ P [X]. Let Q = Q′ ∩ R and B = B′ ∩ R. Suppose further that
Q′ ∩ P [X]/B′ is a nonzero torsionfree left R[X]/Q′ bi-module. Then the following hold:
(a) If Q′ ≥ Q[X], then Q ∩ P/B is a left-R/Q torsionfree module provided Q∩P/B is a nonzero R- bi-
module.
(b) If Q′ = Q[X], then Q ∩ P/B is a nonzero R-bi-module that is left R/Q- torsionfree.
Proof. (a) First note that B is an ideal of R and Q is a prime ideal of R. We now prove (a) given that
Q ∩ P/B ≠ 0. Suppose Q ∩ P/B is left R/Q torsion module. Then there exists an element a ε Q ∩ P ( a is not
an element of B) and an element d ε c(Q), such that da ε B. Since a is not an element of B, so a is not an
element of B′ also. Hence we get that there is an element a ε Q′ ∩ P [X] and a is not an element of B′. By
the previous lemma we have that d ε c(Q′). So da ε B and B ≤ B′ yields that a + B′ is a c(Q) torsion element
of the bi- module Q′∩P[X]/B′. This ontradicts our hypothesis in (a). Hence Q ∩ P/B is a left-R/Q
torsionfree module.
(b) For the proof of (b) apply (a) above after observing that Q′ = Q[X] and B′ < Q′ ∩ P[X] implies
immediately that B < Q ∩ P and hence Q ∩ P/B is a nonzero R bi-module.
Lemma 3. Let R be a Noetherian ring and let R[X] be the usual Polynomial ring. Let Q′ ε spec.R[X]. Let f
(X) = a0 + a1x1 + . . . + anXn be a polynomial such that f (X) is not an element of Q′. Assume an is not an
element of Q where Q =Q′∩R.Then the following hold;
(a) If f (X) ε c(Q′), then there exists an element r ε R such that an rεc(Q)
(b) If an ε c(Q), then f (X) ε c(Q)[X] .
Proof. (a). The proof of (a) is on the same lines as that of Small [6,Lemma4.2].
(b). The proof of (b) is obvious.
Lemma 4. Let R be a Noetherian ring and let R[X] be the usual Polynomial ring. Let Q and P be prime
ideals of R and let B be an ideal of R such that B<Q∩P and QP ≤ B. If Q∩P/B is a nonzero R-bi-module
that is left R/Q torsionfree module then Q[X] ∩ P [X]/B[X] is a nonzero R[X]-bi-module that is left
R[X]/Q[X] torsionfree module and conversely.
Proof. First note that since Q∩P/B is a nonzero R-bi-module hence
3
Q[X]∩P [X]/B[X] is also a nonzero R[X]-bi-module.Suppose Q[X]∩P[X]/B[X] is not a left R[X]/Q[X]
torsionfree module. Then there exists an element f (X) in Q[X] ∩ P [X](such that f(X) is not an element of
B[X]) and an element d(X) of c ((Q[X]) such that d(X)f (X) ε B[X]. Let a and d be the leading terms of f
(X) and d(X) such that a and d do not belong to B and Q respectively . By the left version of Lemma (3)
there exists an element r in R such that rd is an element of c(Q). Now d(X)f (X) Ɛ B[X] implies that rda ε B.
Note that a ε Q ∩ P and a is not an element of B. Since rd ε c(Q) we get thus that Q ∩ P/B is not a left R/Q
torsionfree module. This is a contradiction to our hypothesis. Hence Q[X]∩P[X]/B[X] must be a left
R[X]/Q[X] torsionfree module. The converse follows from Lemma 2.
Definition 5. Two prime ideals P and Q in a Noetherian ring R are said to have a second layer link, written
Q → P , if there exists an ideal A of R with QP ≤ A and A < Q ∩ P such that Q ∩ P/A is a torsionfree
R/Q − R/P bimodule. In this case we also say Q is linked to P or that Q is a second layer link to P . We also
say in this case that Q ∩P/A is a linking bimodule for the link Q → P or Q is linked to P via the ideal A.
Definition 6. We say a Noetherian ring R is link krull symmetric (link k-symmetric for short) if for any
prime ideals Q and P of R such that Q is linked to P we have R/P = R/Q.
Theorem 7. Let R be a Noetherian ring. Let R[X] be the polynomial ring over R. Then Q → P is a link
between the prime ideals Q and P of R if and only if Q[X] → P [X] is a link between the corresponding prime
ideals Q[X] and P [X] of R[X].
Proof. Let Q → P be a link between the prime ideals P and Q of R with Q∩P/B the corresponding linking
bimodule. Then by Lemma (4) we get that Q[X] ∩ P [X]/B[X] is a linking bimodule for the prime ideals
Q[X] and P[X] of R[X]. Hence Q[X] is linked to P [X]. Conversely let Q[X] → P [X] be a link of the prime
ideals Q[X] and P [X] of R[X] via the bimodule Q[X]∩ P [X]/B′, where B′ is an ideal of R[X]. Suppose B
= B′ ∩ R. It is clear that Q[X] → P [X] is also a link of the prime ideal Q[X] to P[X] via the via the ideal
B[X]. Using Lemma 2 we get that Q→P is a link of the prime ideals Q and P via the ideal B.
Theorem 8. Let R be a link k-symmetric, Noetherian ring. Let Q → P be a link of the prime ideals Q and P
of R via an ideal B of R. Then R/B has an artinian quotient ring.
Proof. Given that QP ≤ B < Q ∩ P . Also R/P = R/Q because R is a link k-symmetric ring. Thus P/B and
Q/B are distinct incomparable prime ideals of R/B unless Q = P.Now it is also obvious that Q/B and P/B are
the minimal prime ideals of ring R/B. Two cases arise:
Case (1): P = Q. In this case N (R/B) = P/B. It is clear now that if c+B is an element of c(N(R/B)),then
c+B is a regular element of R/B . Hence by Goldie [2, Theorem 13], R/B has an artinian quotient ring.
Case (2): Q ≠ P . In this case observe that Q/B and P/B are distinct minimal prime ideals of the ring R/B. It
is not difficult to see that in this case also that R/B has an artinian quotient ring.
Theorem 9. Let R be a Noetherian, link k-symmetric ring. Let R[X] be the polynomial ring over R. Let P
be a prime ideal of R. Let P′ = P [X] be the extended prime ideal of P in the ring R[X]. Let Q′ be a prime
ideal of R[X] such that Q′ → P ′ is a link of the prime ideal Q′ to P ′. Then Q′= Q[X] where Q = Q′ ∩ R.
Moreover Q → P is a link of prime ideal Q of R to the prime ideal P of R.
Proof. Let Q′ ∩ P ′/B′ be the linking bimodule for the link Q′ → P ′, where B′ is an ideal of R[X]. Let B =
B′ ∩ R. By hypothesis Q = Q′ ∩R. Clearly Q is a prime ideal of R whereas B is an ideal of R. Observe that
by Lemma 1 any element d in c(Q) implies that d in c(Q′) also. Also it is obvious that any element c in c(P )
implies that c in c(P ′). We now have two cases.
Case (1) B < Q ∩ P . In this case by Lemma 2 Q ∩ P/B is a R/Q − R/P is a torsionfree bimodule. So Q →
P is a right link between prime ideals Q and P of R. Again by Lemma 4 this induces a link Q[X] → P [X] of
4
the prime ideals Q[X] and P [X] of R[X] via the ideal B[X]. Moreover R is right k-symmetric ring implies
that R/P = R/Q. Hence by [3,Theorem13.17] R[X]/P [X] = R[X]/Q[X]. Now by theorem (8) we get
that R[X]/B[X] has an Artinian quotient ring and since N (R[X]/B[X]) is a right k-homogeneous semiprime
ideal of the ring R[X]/B[X], so by [1,Proposition 3.1], we get that N (R[X]/B[X]) is a right weakly ideal
invariant ideal of R[X]/B[X]. Suppose Q′ > Q[X]. Then since N(R[X]/B[X]) is right weakly ideal
invariant ideal of the ring R[X]/B[X], therefore Q′ ∩P ′/Q′P ′ < R[X]/P ′ because R[X]/Q′<R[X]/P [X].
This would immediately imply that Q′ ∩ P ′/Q′P′ is a right R[X]/P ′ torsion module. Hence in particular
Q′ ∩ P ′/B′ is a right torsion R[X]/P ′ module. This is a contradiction to our hypothesis. Hence Q′ = Q[X] in
this case.
Case (2) B = Q ∩ P . In this case B[X] = Q[X] ∩ P [X] and hence B[X] is a semiprime ideal of the ring
R[X]. Let S = set of regular elements of R[X]/B[X], then by Goldie [2, Theorem 13], we get that S is a left
and a right ore-set of the ring R[X]/B[X]. Three subcases arise
Subcase (a) Suppose Q[X] and P [X] are distinct incomparable prime ideals. Then in this subcase it is
clear that Q[X]/B[X] and P[X]/B[X] are the minimal prime ideals of the ring R[X]/B[X]. So S = c((P[X]) ∩
(Q[X])). Now the link Q′ → P ′ of the prime ideals Q′ and P ′ via the linking bimodule Q′ ∩ P ′/B′ can be
thought of as a link in the ring R[X]/B[X] . But this is impossible by [3, Lemma 11.17].
Subcase (b) Q[X]< P [X]. In this subcase B[X] = Q[X]. Let S1 = Set of regular elements of R/Q . Then it
is not difficult to see that S1 is a left and a right ore-set of R/B and hence also of the ring R[X]/B[X]
(=R[X]/Q[X]). Now by Lemma 1, S1 is a subset of the set c((Q′)/B[X]). Now considering Q′ → P ′ as a link
in the ring R[X]/B[X] we must have by [3,Lemma 12.17] that S1 = c((P ′)/B[X]). But that is impossible
because Q[X] < P [X]. Hence Q′ = Q[X].
The other subcase, that is when P [X] < Q[X], is on the same lines as subcase (b) above.
We must state that the proof is quite obvious in the case when Q[X]=P[X].
Proposition 10. Let R be a fully bounded Noetherian ring. Then R is a link k-symmetric ring.
Proof. See [3, Theorem 13.15].
Theorem 11. Let R be a fully bounded Noetherian ring. Let P be a fixed prime ideal of R. Then the
following hold;
(a) If Q′ → P [X] is a link of the prime ideal Q′ of R[X] to the extended prime ideal P [X] of R[X], then Q′ =
Q[X], where Q = Q′ ∩ R.
(b) There are only finitely many prime ideals of R[X] right linked to P[X].
Proof. (a) The proof of (a) follows by using Proposition 11 and Theorem 9.
(b) This follows immediately by [3, Theorem 13.22].
Theorem 12. Let R be a link k-symmetric Noetherian ring and let R[X] be the polynomial ring. Let P1
and Q1 be prime ideals of R[X] and
Q1 →P1 be a link of these prime ideals. Let Q = Q1 ∩ R and P = P1 ∩ R. Then the following hold:
(a) If P1 = P [X], then Q1 = Q[X] and moreover Q → P is a link of the
prime ideals Q and P of R.
(b) If P1 > P [X], then Q1 > Q[X].
Proof. (a) Proof of (a) follows directly from Theorems 7 and 9.
(b) Proof of (b) is immediate from (a) above.
5
Remark:- We have come to know that our Theorem 9 is similar to Theorem 3.15 of the paper
“Projective Prime ideals and localization in P.I rings” by A.W.chatters,C.R. Hajarnavis and
R.M.Lissaman which appeared in J.London Math. Soc.(2)64 (2001)
( 1-12),L.M.S-2001 and in which the authors prove their result in the case when the base ring R is a P.I
ring .In this respect we mention that our Theorem 9 is more general than the above mentioned Theorem
3.15. In fact the base ring R of our Theorem 9 which is a link krull symmetric noetherian ring generalizes
noetherian P.I.rings and it is a significant open question whether or not a two sided noetherian ring is a link
krull symmetric ring.
6
REFERENCES
[1] K.A Brown, T.H Lenagan and J.T Stafford, Weak ideal invariance and
Localization, J. London Math. Soc. 21 (1980), 53-61.
[2] A.W.Goldie, The Structure of prime rings under ascending chain
Conditions, Proc. London Math. Soc, 8(1958),
[3] K.R Goodearl and R.B.Warfield, An Introduction to Non-Commutative
Noetherian Rings, L.M.S., Student texts, 16 589-631.
[4] J.C McConell; J.C.Robson, Non-Commutative Noetherian Rings, John
Wiley and Sons Ltd., 1987.
[5] L.W.Small, Orders in Artinian Rings, J.Alg. 4 (1966) 13-41.
[6] L.W Small, Orders in Artinian Rings II, J Alg. 9 (1968), 266-273.
|
1011.3173 | 2 | 1011 | 2011-06-14T21:09:08 | Some isomorphism invariants for Lie tori | [
"math.RA"
] | In this paper we study the isomorphism problem for centreless Lie tori that are fgc (finitely generated as modules over their centroid). These Lie tori play a important role in the theory of extended affine Lie algebras and of multiloop Lie algebras. We introduce four isomorphism invariants for fgc centreless Lie tori, and use them together with known structural results to investigate the classification problem for fgc centreless Lie tori up to isomorphism. | math.RA | math |
SOME ISOMORPHISM INVARIANTS FOR LIE TORI
BRUCE ALLISON
Abstract. In this paper we study the isomorphism problem for centreless Lie
tori that are fgc (finitely generated as modules over their centroid). These Lie
tori play a important role in the theory of extended affine Lie algebras and
of multiloop Lie algebras. We introduce four isomorphism invariants for fgc
centreless Lie tori, and use them together with known structural results to
investigate the classification problem for fgc centreless Lie tori up to isomor-
phism.
Suppose that k is a field of characteristic 0, Λ is a finitely generated free abelian
group, and ∆ is an irreducible finite root system. A Lie torus of type (∆, Λ) is a
Lie algebra L over k that has two compatible gradings, one by the root lattice Q of
∆ and the other by Λ, such that a list of natural axioms hold (see Definition 3.1).
In that case the rank of Λ is called the nullity of L. Lie tori were introduced by
Yoshii in [37, 38] and, in an equivalent form that we use here, by Neher in [26].
Centreless (zero centre) Lie tori are of fundamental importance in the theory of
extended affine Lie algebras (EALAs), where they are used as the starting point
for the construction of all EALAs [27]. Perhaps the best known example occurs in
nullity 1. In that case, any centreless Lie torus is isomorphic to the derived algebra
modulo its centre of an affine Kac-Moody Lie algebra g [3], and the full affine
algebra g is constructed from this Lie torus by the familiar process of affinization.
In this article, we focus our attention on centreless Lie tori that are fgc (finitely
generated as modules over their centroids). We do this for two reasons. First, it is
these Lie tori that play an important role in the study of multiloop Lie algebras;
and vice versa (see more about this in Section 3). Second, it is known that the
fgc assumption excludes only one family of centreless Lie tori (see the discussion
preceding Theorem 8.5).
The structure of fgc centreless Lie tori is now quite well understood, using work
of a number of authors over a period of almost 15 years. However, the isomorphism
problem, by which we mean the problem of determining when two such Lie tori
are isomorphic, is much less understood. Note that here and subsequently, the
term isomorphic means isomorphic as (ungraded) algebras, unless mentioned to
the contrary.
The isomorphism problem for fgc centreless Lie tori has been solved in nullities
0, 1 and 2. Indeed, in nullities 0 and 1, a solution follows from classical conjugacy
theorems for maximal split toral k-subalgebras of finite dimensional simple Lie
algebras and affine Kac-Moody Lie algebras respectively. (See Sections 5.4 and 6.3
in [7].) In nullity 2, the problem was solved in [7] as part of the classification of
nullity 2 multiloop Lie algebras. (See [7, Cor. 10.1.3 and Thm. 13.3.1].) In this
Date: June 11, 2011.
2000 Mathematics Subject Classification. 17B65, 17B67.
Key words and phrases. Lie tori, extended affine Lie algebras.
1
2
BRUCE ALLISON
paper, we consider the problem for arbitrary nullity. As one might expect, our
approach is to look for isomorphism invariants.
In order to describe some of our results, we briefly outline the structure of this
paper, which begins in Sections 1 -- 4 with some basic definitions and properties of
Lie tori.
type (∆, Λ), which is obtained from L by extending the base ring from the centroid
In Section 5, we investigate the central closure eL of an fgc centreless Lie torus L of
C of L to its quotient field eC. It is known that eL is a finite dimensional isotropic
central simple Lie algebra over eC, and hence the theory of such Lie algebras can
describes an explicit maximal split toral eC-subalgebraeh of eC. From this we deduce
Corollary 5.6, which asserts that the relative type of eL is the type of the given root
system ∆. We note that Corollary 5.6 was a basic tool in the article [7] mentioned
above, but its proof was left to be presented in this article.
be brought to bear on our problem. The main result in this section, Theorem 5.4,
In Section 6, we show that an fgc centreless Lie torus L of type (∆, Λ) has four
isomorphism invariants: (i) the type of the root system ∆, which is called the
root-grading type of L; (ii) the nullity of L; (iii) the rank of L as a module over
its centroid C, which is called the centroid rank of L; and (iv) a vector of positive
integers, called the root-space rank vector of L, that lists the ranks over C of the
root spaces of L in the Q-grading. Indeed, the invariance of the centroid rank is
clear. However, the other three quantities are defined using the graded structure of
L and hence their invariance requires more argument. We establish the invariance of
the root-grading type and the root-space rank vector using the results of Section 5.
We also see that invariance of the nullity follows easily from known facts about Lie
tori.
We note that the four invariants just discussed are rational, by which we mean,
as in [32], that they are defined without using base ring extension. We also note
that, up to this point in the paper, our methods are elementary, using for the most
part linear algebra, sl2-theory and facts from [32, Chap. I] about finite dimensional
central simple Lie algebras. For another approach, see [31], [19] and [20], where tools
from Galois cohomology are used to study the isomorphism problem for forms of
algebras over Laurent polynomial rings and in particular for multiloop Lie algebras.
In Section 7, we recall an equivalence relation for Lie tori, called isotopy, that is
finer than isomorphism as it takes into account the grading [5, 8]. We observe that
the group Λ/Γ(L) is an isotopy invariant (but not yet an isomorphism invariant)
of a centreless Lie torus L, where Γ(L) denotes the Λ-support of the centroid of L.
The main result of the section is a simple characterization of isotopy for centreless
Lie tori.
In the rest of the paper, we assume that k is algebraically closed and we apply
the invariants from Sections 6 and 7 to study classification and the isomorphism
problem for fgc centreless Lie tori. First in Section 8 we summarize in one theorem
the known structure theorems for fgc centreless Lie tori. It states that any such
Lie torus is either classical, which means roughly that it can be constructed as a
special linear Lie algebra, a special unitary Lie algebra, a special symplectic Lie
algebra, or an orthogonal Lie algebra over an associative torus; or it is one of 27
Lie tori (defined for each sufficiently large nullity) that we call exceptional. Since
the statements of the structure theorems are spread over many papers, we hope
that our summary will be of independent interest to the reader. Included in this
LIE TORI
3
section is a table, numbered as Table 1, of our invariants for exceptional Lie tori,
with references to the literature.
In Section 9, we show how to calculate the invariants for classical Lie tori, and
list the results in two tables, numbered as Tables 2 and 3. The three tables are
then applied in Section 10 to obtain results about the isomorphism problem for fgc
centreless Lie tori. We show that the classes of exceptional and classical Lie tori have
no overlap and that the four classes of classical Lie tori are similarly disjoint. We
then solve the isomorphism problem for special symplectic Lie tori and orthogonal
Lie tori (the latter is easy), and we reduce the problem for exceptional Lie tori to
consideration of at most five particular algebras (in each nullity). This reduces the
classification of fgc centreless Lie tori to the separate isomorphism problems for
(1) five particular exceptional Lie tori, (2) special linear Lie tori, and (3) special
unitary Lie tori.
In the final section, we discuss these three problems under an additional con-
jugacy assumption for certain (but not all) maximal split toral k-subalgebras of
an fgc centreless Lie torus. The additional assumption is reasonable since work
in progress by Chernousov, Gille and Pianzola [16] will show that it always holds
(see Remark 11.1). Under the conjugacy assumption, we show that isotopy and
isomorphism coincide for fgc centreless Lie tori and use this to complete the classi-
fication of exceptional Lie tori. Also under the conjugacy assumption, we complete
the classification of special linear Lie tori, leaving only the isomorphism problem
for special unitary Lie tori to be solved.
Finally, we note that the conjugacy assumption could have been used earlier in
the paper to demonstrate the invariance of the root-grading type and the root-space
rank vector. However, we did not do that since we understand that [16] uses deep
results from the theory of group-schemes, whereas our goal has been to deduce as
much as possible about the isomorphism problem for Lie tori using self-contained
and elementary methods.
Acknowledgments. First, we thank Arturo Pianzola for carefully reading an ear-
lier version of this paper and making several suggestions that substantially improved
its presentation. We also thank him for keeping us informed of the work in [16] on
conjugacy. Second, we thank the referee who noticed and filled a small gap in the
proof of Theorem 7.2. The referee also very helpfully suggested the expansion, from
the first version of the paper, of the material now included in Sections 9 and 10.
1. Preliminaries
Throughout the paper, we assume that k is a field of characteristic 0. Unless
mentioned to the contrary, algebra will mean algebra over k.
The centroid. Suppose that A is an algebra over k. The centroid of A is the
subalgebra of Endk(A) defined by
Ck(A) := {c ∈ Endk(A) c(x · y) = c(x) · y = x · c(y) for x, y ∈ A}.
Then k idA is a subalgebra of Ck(A), which we identify with k in the evident fashion
when A 6= 0. The algebra A is said to be central if Ck(A) = k idA.
Note that A is naturally a left Ck(A)-module; and we say that A is fgc if this
module is finitely generated.
4
BRUCE ALLISON
The algebra A is said to be perfect if A · A = A, where · denotes the product in
A. If A is perfect, then Ck(A) is commutative. If A is simple (and hence perfect),
then Ck(A) is a field and A is a central simple algebra as an algebra over Ck(A).
If A is a unital associative algebra, we denote the centre of A by Z(A). Then
the map z 7→ ℓz is an isomorphism of Z(A) onto Ck(A), where ℓz ∈ Endk(A) is left
multiplication by z.
Remark 1.1. (i) If A is an algebra over an extension field F of k and A is perfect
(over F or equivalently over k), then Ck(A) = CF (A).
(ii) Any isomorphism ϕ : A → A′ of algebras induces a unique isomorphism
χ : Ck(A) → Ck(A′) such that ϕ(cx) = χ(c)ϕ(x) for c ∈ Ck(A), x ∈ A.
Involutions. If A is an algebra, an involution of A is an anti-automorphism "−"
of A (so xy = ¯y ¯x for x, y ∈ A) of period 2. In that case, we call (A, −) an algebra
with involution. If the involution is fixed, we often use the notation
A+ = {x ∈ A ¯x = x}
and A− = {x ∈ A ¯x = −x},
in which case A = A+ ⊕ A−. If A is unital and associative, the centre of (A, −) is
defined as Z(A, −) := {x ∈ Z(A) ¯x = x} = Z(A) ∩ A+.
Graded algebras. If Λ be an abelian group and A = Lλ∈Λ Aλ is a Λ-graded
algebra, we use the notation suppΛ(A) := {λ ∈ Λ Aλ 6= {0}} for the Λ-support
of A.
If A is a Λ-graded algebra and A′ is a Λ′-graded algebra we say that A and A′
are isograded-isomorphic if there exists an algebra isomorphism ϕ : A → A′ and a
group isomorphism ϕgr : Λ → Λ′ such that ϕ(Aλ) = A′ϕgr(λ) for λ ∈ Λ.
There is an evident definition of a graded algebra with involution (the involution
is assumed to be graded) and of isograded-isomorphism for graded algebras with
involution (the map is assumed to preserve the involutions).
Irreducible finite root systems. As in [1] and [26], it will be convenient for us
to work with root systems that contain 0. So, if X is a finite dimensional vector
space over k, by an irreducible finite root system in X we will mean a finite subset
∆ of X such that 0 ∈ ∆ and ∆× := ∆ \ {0} is an irreducible finite root system in
X in the usual sense (see [15, chap. VI, §1, D´efinition 1]). We say that ∆ is reduced
if 2α /∈ ∆× for α ∈ ∆×.
An irreducible finite root system ∆ has one of the following types: Aℓ (ℓ ≥ 1),
Bℓ (ℓ ≥ 2), Cℓ (ℓ ≥ 3), Dℓ (ℓ ≥ 4), E6, E7, E8, F4 or G2 if ∆ is reduced; or
BCℓ (ℓ ≥ 1) if ∆ is not reduced.
We will use the following notation for an irreducible finite root system ∆ in X .
Let
Q(∆) := spanZ(∆)
be the root lattice of ∆. Let X ∗ denote the dual space of X , let h , i : X × X ∗ → k
denote the natural pairing, and, if α ∈ ∆×, let α∨ denote the coroot of α in X ∗.
Finally, let
∆×
ind := ∆× \ 2∆×
denote the set of indivisible nonzero roots in ∆, and let ∆ind := ∆×
ind ∪ {0}. Then
∆ind is a reduced irreducible finite root system in X ; and, if ∆ is reduced, we
have ∆ind = ∆.
LIE TORI
5
2. Split toral subalgebras and relative type
Suppose that L is a Lie algebra over k.
A split toral k-subalgebra of L is an abelian1 k-subalgebra h of L such that
there is a k-basis for L consisting of simultaneous eigenvectors (with corresponding
eigenvalues in k) for all of the operators ad(h), h ∈ h.
If h is a split toral k-subalgebra of L, then we have the decomposition L =
Lα∈h∗ Lα, called the root-space decomposition of L relative to h, where
Lα = {x ∈ L [h, x] = α(h)x for h ∈ h}
for α ∈ h∗. We set
∆k(L, h) := {α ∈ h∗ Lα 6= 0},
and we call ∆k(L, h) the root system of L relative to h.
The following formal result is well-known and easily checked using Remark 1.1.
Lemma 2.1. Suppose that L (resp. L′) is a central perfect Lie algebra over a field
F (resp. F ′) that is an extension field of k. Suppose that ϕ : L → L′ is a k-algebra
isomorphism, h is a split toral F -subalgebra of L, and h′ = ϕ(h). Then h′ is a split
toral F ′-subalgebra of L′, which is maximal if and only if h is maximal. Moreover,
setting ∆ = ∆F (L, h), Q = spanZ(∆), ∆′ = ∆F ′ (L′, h′) and Q′ = spanZ(∆′), there
exists a unique group isomorphism ρ : Q → Q′ such that ϕ(Lα) = L′
ρ(α) for α ∈ Q.
Furthermore, ρ(∆) = ∆′ and dimF (Lα) = dimF ′ (L′
ρ(α)) for α ∈ Q.
A finite dimensional central simple Lie algebra over k is said to be isotropic if it
contains a nonzero split toral k-subalgebra.
Theorem 2.2. [32, §I.2] Suppose that L is an isotropic finite dimensional central
simple Lie algebra over k and h is a maximal split toral k-subalgebra of L. Then
(i) ∆k(L, h) is an irreducible finite root system in h∗.
(ii) If h′ is another maximal split toral k-subalgebra of L, there exists an auto-
morphism ϕ of L such that ϕ(h) = h′.
If L is an isotropic finite dimensional central simple Lie algebra, the relative type
of L is defined to be the type of the root system ∆k(L, h), where h is a maximal
split toral k-subalgebra of L. By Theorem 2.2 and Lemma 2.1 (with F = F ′ = k)
this is independent of the choice of h.
3. Lie tori
For the rest of the paper we assume that ∆ is an irreducible finite root system
with Q = Q(∆), and that Λ is a finitely generated free abelian group.
This section contains the definition and some basic properties of Lie tori. We
restrict ourselves to the properties that we will need. For the reader wanting to
learn more about this topic, two recent articles by Neher [28, 29] are recommended.
In order to recall the definition of a Lie torus, we first introduce some notation
for Q × Λ-gradings. Let
L = M(α,λ)∈Q×Λ
Lλ
α
1It is not difficult to show that the abelian assumption is superfluous (although we will not
use this fact).
6
BRUCE ALLISON
Lλ
α
with
be a Q × Λ-grading on a Lie algebra L.2 Then L =Lα∈Q Lα is a Q-grading of L
Lα :=Mλ∈Λ
L =Lλ∈Λ Lλ is a Λ-grading of L with
Lλ :=Mα∈Q
for α ∈ Q;
for λ ∈ Λ;
and we have Lλ
α = Lα ∩ Lλ. Conversely if L has a Q-grading and a Λ-grading
that are compatible (which means that each Lα is a Λ-graded subspace of L or
equivalently that each Lλ is a Q-graded subspace of L), then L is Q × Λ-graded
with Lλ
α = Lα ∩ Lλ. From either point of view, we can simultaneously regard L
as a Q × Λ-graded algebra, a Q-graded algebra and a Λ-graded algebra; and we
correspondingly have the support sets suppQ×Λ(L), suppQ(L) and suppΛ(L). We
refer to the Q-grading as the root grading of L, and we refer to the Λ-grading as
the external grading of L.
Lλ
α
Definition 3.1. [26] A Lie torus of type (∆, Λ) is a Lie algebra L which has the
following properties:
(LT1) L has a Q × Λ-grading L =L(α,λ)∈Q×Λ Lλ
ind, 0) ⊆ suppQ×Λ(L).
(i) (∆×
(ii) If (α, λ) ∈ suppQ×Λ(L) with α ∈ ∆×, then there exist elements eλ
α such that suppQ(L) = ∆.
(LT2)
α ∈
(1)
Lλ
α and f λ
α ∈ L−λ
α = keλ
α, L−λ
−α = kf λ
α and
−α such that Lλ
[[eλ
α, f λ
α ], xβ ] = hβ, α∨ixβ
for xβ ∈ Lβ, β ∈ Q.
(LT3) L is generated as an algebra by the spaces Lα, α ∈ ∆×.
(LT4) Λ is generated as a group by suppΛ(L).
In the definition given in [26], it is only assumed that suppQ(L) ⊆ ∆ in (LT1).
However, our stronger assumption is more convenient here and it results in no loss
of generality (see [5, Remark 1.1.11]).
α = [eλ
α as in (LT2)(ii). Thus if (α, λ) ∈ suppQ×Λ(L) with α ∈ ∆×, then (eλ
made a fixed choice of a grading L =L(α,λ)∈Q×Λ Lλ
If L is a Lie torus, we assume (unless mentioned to the contrary) that we have
α as in (LT1) and elements eλ
α
α, f λ
α )
α spanned by this triple
and f λ
is an sl2-triple in L, where hλ
is a 3-dimensional split simple Lie subalgebra of L.
Remark 3.2. If (α, λ) ∈ suppQ×Λ(L) with α ∈ ∆×, then L is a locally finite
dimensional S λ
α-module under the adjoint action. Indeed, to see this it suffices to
show that U (S λ
α)xβ is finite dimensional for xβ ∈ Lβ, β ∈ ∆, where U (S λ
α) is the
universal enveloping algebra of S λ
α. This fact in turn follows from the Poincar´e-
Birkhoff-Witt theorem for S λ
α, (1) and the assumption that ∆ is finite.
α ]. Hence the space S λ
α, f λ
α, hλ
Definition 3.3. If L is a Lie torus of type (∆, Λ), we define the nullity of L to be
rankZ(Λ) and the root-grading type of L to be the type of ∆.
We note that a Lie torus is perfect by (1) and (LT3).
2As is usual in the study of Lie tori, it is convenient to use the notation Lλ
α rather than L(α,λ)
or L(α,λ) for the homogeneous component of degree (α, λ).
LIE TORI
7
Example 3.4. Suppose that g is a finite dimensional split simple Lie algebra with
splitting Cartan subalgebra h over k. Let ∆ = ∆k( g, h) and Q = Q(∆); and let
g =Lα∈Q gα be the corresponding root-space decomposition. For n ≥ 0, let
grading Rn =Lλ∈Zn Rλ
be the algebra of Laurent polynomials in n variables over k with its natural Zn-
n. Then g⊗ Rn is an fgc centreless Lie torus of type (∆, Zn)
n for (α, λ) ∈ Q × Zn. We call g ⊗ Rn the untwisted Lie
α = gα ⊗ Rλ
1 , . . . , t±1
n ]
Rn := k[t±1
with ( g ⊗ Rn)λ
torus of type (∆, Zn).
When k is algebraically closed, there is a twisted version of the above example
which constructs a subalgebra L( g, σ) of g ⊗ Rn from a finite dimensional (split)
simple Lie algebra g and an n-tuple σ of commuting finite order automorphisms of
g.3 The algebra L( g, σ) is called a nullity n multiloop Lie algebra. If the common
fixed point algebra gσ is nonzero, then L( g, σ) is an fgc centreless Lie torus of
nullity n relative to some Q × Λ grading on L( g, σ) [24, Thm. 5.1.4]. Conversely,
any fgc centreless Lie torus of nullity n is isomorphic to L( g, σ) for some g and σ
as above with gσ 6= 0 [5, Thm. 3.3.1].
We will recall some other constructions of Lie tori in Section 8.
We now prove three lemmas about Lie tori using sl2-theory. In each lemma we
assume that L is a Lie torus of type (∆, Λ), where we recall that we are assuming
that Λ is a finitely generated free abelian group.
The first lemma is an analogue for Lie tori of the well-known fact that any asso-
ciative Λ-torus is a domain. (See Section 8 to recall the definition of an associative
torus.)
Lemma 3.5. If α, β ∈ ∆× with hβ, α∨i < 0, 0 6= xα ∈ Lα and 0 6= yβ ∈ Lβ, then
ad(xα)−hβ,α∨iyβ 6= 0.
Proof. Because of our assumptions on Λ, we know that we can give Λ a linear order
(for example the lexicographic order relative to some Z-basis of Λ). Given nonzero
x ∈ L, this order on Λ allows us to speak of the nonzero component of highest
degree of x.
α and yβ ∈ Lµ
Suppose for contradiction that ad(xα)−hβ,α∨iyβ = 0. Then replacing xα and yβ
by their nonzero components of highest degree in the Λ-grading, we can assume
that xα ∈ Lλ
β are
1-dimensional, we have ad(eλ
β lies in a finite
dimensional S λ
β is an eigenvector for ad(hλ
α)
with eigenvalue hβ, α∨i < 0. Therefore from the classification of finite dimensional
irreducible S λ
(cid:3)
β, where λ, µ ∈ Λ. Thus, since the spaces Lλ
α-submodule of L. Further, by (1), eµ
β = 0. But, by Remark 3.2, eµ
α-modules, we have ad(eλ
α)−hβ,α∨ieµ
α)−hβ,α∨ieµ
α and Lµ
β 6= 0.
The second lemma is an analogue for Lie tori of the well-known fact that any
invertible element in an associative Λ-torus is homogeneous.
Lemma 3.6. Suppose [x, y] ∈ L0
Then x ∈ Lλ
α and y ∈ L−λ
−α for some λ ∈ Λ.
0, where 0 6= x ∈ Lα, 0 6= y ∈ L−α and α ∈ ∆×.
Proof. We order Λ as in the previous proof. Let xµ(x)
Λ-homogeneous component of x of highest degree µ(x), and let yµ(y)
∈ Lµ(x)
α
α
be the nonzero
−α ∈ Lµ(y)
−α be
3In [5] and [24], L(g, σ) is denoted by Mm(g, σ).
8
BRUCE ALLISON
α
the nonzero Λ-homogeneous component of y of highest degree µ(y). Then, [x, y] −
[xµ(x)
, yµ(y)
−α ] is the sum of Λ-homogeneous terms of degree less than µ(x) + µ(y).
But [xµ(x)
−α ] 6= 0 by Lemma 3.5 with β = −α. So µ(x) = −µ(y). Similarly if
we use lowest degrees ν(x) and ν(y), we get ν(x) = −ν(y). So µ(x) = −µ(y) ≤
−ν(y) = ν(x), which implies that x = xµ(x)
(cid:3)
. Similarly, y = yµ(y)
−α .
, yµ(y)
α
α
i=1 (Lαi ∪ L−αi ).
Lemma 3.7. Suppose L is a Lie torus of type (∆, Λ). If {α1, . . . , αℓ} is a base for
the root system ∆, then the algebra L is generated bySℓ
Proof. Let M be the subalgebra of L that is generated by the indicated set, and let
E× = {α ∈ ∆× Lα ⊆ M}. In view of (LT3), it suffices to show that E× = ∆×.
Now it follows from [5, (4)] that E× is stable under the action of the Weyl group
of ∆. Hence, ∆ind ⊆ E×, and we are done if ∆ is reduced. Assume now that ∆
is not reduced, and let α be a root of smallest length in ∆×. It remains to show
2α ∈ ad(e0
that 2α ∈ E×. To verify this, it is enough to show that eσ
α)Lα for all
σ ∈ Λ. This is an easy exercise using representations of the algebra S0
α. We leave
the details to the reader.
(cid:3)
4. Centreless Lie tori
In this section, we assume that L is a centreless Lie torus of type (∆, Λ) and we
recall the basic facts that we will need about L. All of these facts were announced
by Neher in [26] or [28, §5.8(c)]. For the convenience of the reader, we provide a
proof or a reference for a proof in each case.
Set
g = L0
and h = L0
0.
Then, by [5, Prop. 1.2.2], g is a finite dimensional split simple Lie algebra with
splitting Cartan subalgebra h. Moreover [ibid], ∆ can be uniquely identified (by
means of a linear isomorphism of spank(∆) onto h∗) as a root system in h∗ in such
a way that
∆ind = ∆k(g, h)
α, f 0
and [e0
tion. In that case we have [ibid]
α] = α∨ for α ∈ ∆×
ind. We will subsequently always make this identifica-
[eλ
α, f λ
α ] = α∨ for (α, λ) ∈ suppQ×Λ(L), α ∈ ∆×
and
(2)
(Here α∨ ∈ (h∗)∗ = h.)
Lα = {x ∈ L [h, x] = α(h)x for h ∈ h}
for α ∈ Q.
Note that (2) tells us that h is a split toral k-subalgebra of L and that the root
grading of L is the root-space decomposition of L relative to h.
Recall that an algebra A is said to be prime if the product of any two nonzero
ideals of A is nonzero.
Proposition 4.1. L is prime.
Proof. The main tool in the argument is Lemma 3.5, which tells us that if α, β ∈ ∆×
with hβ, α∨i < 0, 0 6= xα ∈ Lα and 0 6= yβ ∈ Lβ, then
(3)
0 6= ad(xα)−hβ,α∨iyβ ∈ Lwα (β),
where wα is the reflection along α in the Weyl group W of ∆.
Suppose now that I is a nonzero ideal of L. By (2), I is Q-graded; that is
LIE TORI
9
that ∆×(I) = ∆×.
I =Lα∈∆ Iα, where Iα = I ∩ Lα. Let ∆×(I) = {α ∈ ∆× Iα 6= 0}. We will see
Note first that ∆×(I) 6= ∅. Indeed otherwise we have I ⊆ L0, which implies
[I, Lα] = 0 for α ∈ ∆× and hence [I, L] = 0 by (LT3), contradicting our assumption
that L is centreless.
We now claim that W ∆×(I) ⊆ ∆×(I). To see this, it is enough to show that
wα(β) ∈ ∆×(I) for α ∈ ∆× and β ∈ ∆×(I). For this we can assume that hβ, α∨i <
0 in which case our claim follows taking yβ ∈ Iβ in (3). Note that in particular, if
β ∈ ∆×(I), we have −β = wβ(β) ∈ ∆×(I).
Next we claim that ∆×(I) and ∆× \ ∆×(I) are orthogonal. Indeed, if not, we
can choose α ∈ ∆×(I) and β ∈ ∆× \ ∆×(I) with hβ, α∨i 6= 0. Replacing, α by
−α if necessary, we can assume that hβ, α∨i < 0. But then taking xα ∈ Iα in (3),
we see that wα(β) ∈ ∆×(I) and hence (by the previous claim) β ∈ ∆×(I). This
contradiction proves the claim. It then follows from the irreducibility of ∆ that
∆×(I) = ∆×.
To prove the proposition, suppose for contradiction that I and J are nonzero
ideals of L with [I, J ] 6= 0. Then ∆×(I) = ∆× and ∆×(J ) = ∆×. Hence, for any
α ∈ ∆×, we have α ∈ ∆×(I) and −α ∈ ∆×(J ). So Iα 6= {0} and J−α 6= {0}.
Since [Iα, J−α] = 0, this contradicts (3) (with β = −α).
(cid:3)
Let C = Ck(L). Then C = Lλ∈Λ C λ is a Λ-graded commutative associative
algebra, where C λ := {c ∈ C c(Lµ) ⊆ Lµ+λ for µ ∈ Λ} [12, Lemma 3.11(1)].
Set
Γ = Γ(L) := suppΛ(C).
Then Γ is a subgroup of Λ [ibid], and
(4)
C ≃ k[Γ],
as graded algebras, where k[Γ] is the group algebra of Γ with its natural Λ-grading
[12, Prop. 3.13(ii)].
Recall (see Section 3) that Λ is called the external-grading group of L. Note also
that L it is naturally graded by the quotient group Λ/Γ, and we call the group Λ/Γ
the quotient external-grading group of L.
The following proposition follows from [5, Lemma 1.3.7 and Prop. 1.4.1]:
Proposition 4.2. Suppose that L is fgc. Then
(i) Lλ is finite dimensional for λ ∈ Λ.4
(ii) Λ/Γ is finite.
5. The central closure of an fgc centreless Lie torus
In this section we assume that L is an fgc centreless Lie torus of type (∆, Λ)
and we discuss the central closure of L. We continue using the notation h = L0
0,
C = Ck(L) and Γ = Γ(L) introduced in Section 4.
Taking into account Proposition 4.2, we now fix a list λ1, . . . , λm of represen-
tatives of the cosets of Γ in Λ, with λ1 = 0. For α ∈ ∆ and 1 ≤ i ≤ m, we
4Part (i) is true without the assumptions that L is fgc and centreless [27, Thm. 5], but the
proposition as stated is all that we need.
10
BRUCE ALLISON
choose a (finite) k-basis Bi
α . For α ∈ ∆ we let Bα = ∪m
B = ∪α∈∆Bα. Note that B is finite since ∆ = suppQ(L) is finite.
α for Lλi
i=1Bi
α; and we let
Proposition 5.1.
(i) If α ∈ Q, Lα is a C-submodule of L and Bα is a Λ-homogeneous C-basis
for Lα. Hence Lα is a free C-module of finite rank.
(ii) B is a Q × Λ-homogeneous C-basis for L. Hence L is a free C-module of
finite rank.
Proof. Since (ii) follows from (i), so we only need to check (i). First, the fact that
Lα is a C-submodule of L follows from (2). Also Bα is Λ-homogeneous by definition.
Finally, the fact that Bα is a C-basis for Lα is easily checked directly using (4). (cid:3)
eL := eC ⊗C L.
The centroid C of L is an integral domain (for example by (4)). Let eC be the
quotient field of C, in which case eC is an extension field of k. Let
Then eL is a algebra over eC which we call the central closure of L.
Now L is prime (by Proposition 4.1), perfect and fgc. So eL is a finite dimensional
central simple algebra over eC, and the map x 7→ x⊗1 identifies L as a C-subalgebra
of eL. (See for example [6, §3], which uses results from [17, §1].)
It follows from Proposition 5.1(ii) that B is a eC-basis of eL and hence
algebras), it follows easily using Remark 1.1(ii) that eL and eL′ are isomorphic (as
Remark 5.2. If L and L′ are fgc centreless Lie torus that are isomorphic (as k-
dim eC(eL) = rankC (L).
k-algebras).
(5)
Next let
eh = eCh
(i) B0
Proposition 5.3.
0 is a k-basis for h = L0
in eC. It is clear thateh is a nonzero split toral eC-subalgebra of eL, and hence eL is
isotropic (see Section 2). We will show in Theorem 5.4 thateh is a maximal split
toral eC-subalgebra of eL.
We first look at the root space decomposition of eC relative to eh. For this, let
h∗ = Homk(h, k) be the dual space of h over k (as before), andeh∗ = Hom eC(eh,eC)
be the dual space ofeh over eC.
0 is a eC-basis for eh. Hence dim eC (eh) =
dimk(h), and any k-basis for h is a eC-basis foreh.
(ii) There exists a unique k-linear map α 7→ α of h∗ intoeh∗ with αh = α for
α ∈ h∗. Under this map, any k-basis for h∗ is sent to a eC-basis foreh∗; and
h = {h ∈eh α(h) ∈ k for α ∈ ∆}.
(iii) Let e∆ = { α α ∈ ∆} and eQ = { α α ∈ Q}. Then e∆ is an irreducible finite
root system over eC ineh∗ of the same type as ∆,5 and we have eQ = Q(e∆).
5 In fact, one can check that e∆ is isomorphic to the root system obtained from ∆ by base field
0 and B0
we have
(6)
extension from k to eC (as described in [15, Chap. VI, §1, Remark 1]).
LIE TORI
11
Proof. B0
0 was chosen as a k-basis for h = L0
(iv) Let eL α := {x ∈ eL [h, x] = α(h)x for h ∈eh} for α ∈ Q. Then eL α = eCLα
for α ∈ Q and eL =L α∈ e∆ eL α.
(v) ∆ eC (eL,eh) = e∆.
(vi) If α ∈ ∆, then Bα is a eC-basis for eL α and hence rankC (Lα) = dim eC (eL α).
0 is part of the eC-basis B for
eL. This implies (i); (ii) follows from (i) and the fact that ∆ contains a k-basis of
Next eL = Pα∈Q eCLα and eCLα ⊆ eL α for α ∈ Q. Since the sum P α∈ eQ eL α is
direct, this implies (iv). Also, if α ∈ Q, we have eL α 6= {0} ⇐⇒ eCLα 6= 0 ⇐⇒
Lα 6= 0 ⇐⇒ α ∈ ∆. (Here we have used the equality ∆ = suppQ(L) from (LT1).)
So we have (v). Finally, if α ∈ Q, then Bα is part of a C-basis for L by Proposition
5.1, so (vi) follows from (iv).
(cid:3)
h∗; and (iii) follows from (ii).
0, and B0
Theorem 5.4. Suppose that L is an fgc centreless Lie torus of type (∆, Λ) with
is injective. For this, we argue by
the claim it is enough to show that ad(x)2 eL− α
by Proposition 5.3(iv), x = c−1x and y = d−1y, where c and d are nonzero elements
of C, 0 6= x ∈ Lα and 0 6= y ∈ L−α. Then ad(x)2y = 0. But this contradicts Lemma
3.5 (with β = −α), so we have the claim.
0 andeh = eCh. Then,eh is a maximal split toral
central closure eL = eCL. Let h = L0
eC-subalgebra of eL.
Proof. We first claim that if α ∈ ∆× and x is a nonzero element of eL α, then
ad(x)2 maps eL− α bijectively onto eL α. Now eL− α and eL α have the same dimension
over eC, since they are paired by the Killing form of eL over eC. Hence to prove
contradiction. Suppose that ad(x)2 y = 0 for some nonzero element y of eL− α. Now,
To prove the theorem, let t be a maximal split toral eC-subalgebra of eL containing
eh, and let E = ∆ eC (eL, t). By Theorem 2.2(i), E is an irreducible finite root system
over eC in t∗. We choose a Z-basis for the root lattice Q(E) of E and order Q(E)
Since [t,eh] = 0, we have [t, eL α] ⊆ eL α for α ∈ ∆. Then, sinceeh ⊆ t, it follows easily
using the corresponding lexicographic order. For α ∈ ∆ we let
E α = {ε ∈ E εeh = α}.
that
(7)
eL α = Mε∈E α eLε.
space relative to t.)
Now let α ∈ ∆×. Then, E α 6= ∅ by (7). Let ε be the maximum root in E α,
for α ∈ ∆. (Here eL α denotes a root space relative toeh, whereas eLε denotes a root
and fix nonzero x ∈ eLε. Then, again by (7), x ∈ eL α. So, as we saw above, ad(x)2
maps eL− α bijectively onto eLα. It follows from this that E α = E− α + 2ε. Since
Finally, to show that t ⊆eh, let t ∈ t. Let {α1, . . . , αℓ} be a base for the root
E− α = −E α, we have E α = −E α + 2ε. Hence, if ζ ∈ E α, we have ζ = −η + 2ε for
some η ∈ E α, which gives 2ε = ζ + η. But if ζ < ε this forces 2ε < ε + η ≤ 2ε, a
contradiction. Therefore E α = {ε}; that is E α is a singleton.
system ∆, and choose ε1, . . . , εℓ in E with E αi = {εi} for 1 ≤ i ≤ ℓ. But, by
12
BRUCE ALLISON
Corollary 5.5. h is a maximal split toral k-subalgebra of L.
such that αi(h) = εi(t) for 1 ≤ i ≤ ℓ. Then it follows from (7) (with α = αi)
Proposition 5.3(ii), α1, . . . , αℓ is a eC-basis for eh∗, and so we can choose h ∈ eh
that ad(h) = ad(t) on eL αi for each i. Similarly, since E− αi = −E αi = {−εi},
ad(h) = ad(t) on eL− αi for each i. So, by Lemma 3.7, ad(h − t) = 0 on eL. Since eL
is centreless, t = h ∈eh.
Proof. Suppose that t is a split toral k-subalgebra of L containing h. Then t := eCt is
a split toral eC-subalgebra of eL containingeh. Consequently, by Theorem 5.4, t =eh.
Now let t ∈ t. So t ∈ t = eh. But adL(t) is diagonalizable linear operator on
L over k, and hence ad eL(t) is a diagonalizable linear operator on eL over eC with
eigenvalues lying in k. So α(t) ∈ k for α ∈ ∆. Thus, by (6), t ∈ h.
(cid:3)
(cid:3)
The next corollary was announced in [7] as Theorem 5.5.1 and used there as one
of the main tools in the classification of nullity 2 multiloop Lie algebras.6 7
Corollary 5.6. The relative type of eL is the root-grading type of L.
Proof. By Theorem 5.4, the relative type of eL is the type of the root system
∆ eC (eL,eh), which, by Proposition 5.3(iii) and (v), has the same type as ∆.
6. Some isomorphism invariants
(cid:3)
Suppose that L is an fgc centreless Lie torus L of type (∆, Λ) with centroid C.
We now describe four entities that we then show are isomorphism invariants of L.
Recall first that we defined the root-grading type of L and the nullity of L in
Definition 3.3. Next, we define the centroid rank of L to be
crk(L) := rankC (L).
Finally, it follows from [5, (4)] that if α, β ∈ ∆× are in the same orbit under
the Weyl group of L, then rankC (Lα) = rankC (Lβ). Consequently, this equality of
rank holds whenever α, β have the same length. So, we may define rksh(L) to be
rankC(Lα), where α is a short root8 in ∆×. If there exists a long root (resp. an
extra long root) α in ∆× we define rklg(L) (resp. rkex(L)) to be rankC (Lα). Putting
these quantities together, we define a vector of positive integers
rkv(L) =
(rksh(L))
(rksh(L), rklg(L))
(rksh(L), rkex(L))
(rksh(L), rklg(L), rkex(L))
if ∆ is reduced and simply laced,
if ∆ is reduced and not simply laced,
if ∆ is of type BC1,
if ∆ is of type BCℓ, ℓ ≥ 2,
which we call the root-space rank vector of L.
Proposition 6.1. Suppose L and L′ are fgc centreless Lie tori with central closures
eL and eL′ respectively. If eL and eL′ are isomorphic as Lie algebras over k, then
6In [7], each result in the sequence Theorem 5.5.1, Corollary 5.5.2, Theorem 9.2.1, Theorem
12.2.1, Table 2, Theorem 13.2.1(b) and the classification Theorem 13.3.1 uses its predecessor.
7The classification of nullity 2 multiloop Lie algebras has subsequently also been obtained by
Gille and Pianzola in [20] as a consequence of their classification of R2-loop simple adjoint groups
and algebras using cohomological methods .
8Our root length terminology follows [1]. Roots of minimum length in ∆× are called short,
roots in ∆× ∩ (2∆×) are called extra-long, and all other roots in ∆× are called long.
LIE TORI
13
(i) The root-grading type of L equals the root-grading type of L′.
(ii) The nullity of L equals the nullity of L′.
(iii) crk(L) = crk(L′).
(iv) rkv(L) = rkv(L′).
0) of Sections 4 and 5; and we use
(iii): This is clear (and does not use Theorem 5.4).
the same relative type. Hence, by Corollary 5.6, we have (i).
Proof. We use the notation (for example h = L0
corresponding primed notation (for example h′ = L′0
k-algebra isomorphism.
(ii): This is easy to see (and does not require the results of Section 5). Indeed,
by Proposition 4.2(ii), rankZ(Λ) = rankZ(Γ) and similarly rankZ(Λ′) = rankZ(Γ′).
in rankZ(Γ) (resp. rankZ(Γ′)) variables over k, so rankZ(Γ) = rankZ(Γ′) as desired.
Indeed, it follows easily
0) for L′. Let ϕ : eL → eL′ be a
(i): It follows from Lemma 2.1 (with F = C and F ′ = C′) that eL and eL′ have
So it suffices to show that rankZ(Γ) = rankZ(Γ′). Now eC = Ck(eL) andfC′ = Ck(eL′),
so eC ≃fC′. But, by (4), eC (resp.fC′) is isomorphic to the field of rational functions
from Remark 1.1(ii) (applied to eL and eL′) that dim eC(eL) = dimfC ′(eL′). So, by (5),
(iv): By Theorem 5.4, eh is a maximal split toral eC-subalgebra of eL. So, by
Lemma 2.1 applied to eL and eL′, ϕ(eh) is a maximal split toral fC′-subalgebra of
eL′. Thus, by Theorem 2.2(ii), we can assume that ϕ(eh) = eh′. Now by Proposition
5.3(iii) and (v), we have eQ = Q(e∆) and ∆ eC (eL,eh) = e∆, as well as corresponding
equations for L′. Thus, by Lemma 2.1 applied to eL and eL′, there exists a group
isomorphism ρ : eQ →fQ′ such that ρ(e∆) = f∆′ and dim eC (eL α) = dimfC ′(eL′
α ∈ eQ. Finally, we let τ : Q → Q′ be the group isomorphism such that the following
ρ( α)) for
rankC(L) = rankC ′(L′).
diagram commutes:
Q
ye
eQ
τ−−−−→ Q′
ye
−−−−→ fQ′
ρ
Then τ (∆) = ∆′; and we have rankC (Lα) = rankC ′(L′
τ (α)) for α ∈ ∆ by Proposition
5.3(vi). Finally, τ extends to a k-linear isomorphism h∗ → h′∗ which maps ∆ onto
∆′. This extension is an isomorphism of root systems, and so it maps short roots,
long roots and extra long roots in ∆× to roots of corresponding length in ∆′×. (cid:3)
By Remark 5.2, the following result follows immediately from Proposition 6.1.
Theorem 6.2. If L and L′ are fgc centreless Lie tori that are isomorphic as k-
algebras, then (i), (ii), (iii) and (iv) in Proposition 6.1 hold. That is, the root-
grading type, the nullity, the centroid rank, and the root-space rank vector are iso-
morphism invariants of an fgc centreless Lie torus.
The above proofs also show that the rank of L0 over C is an isomorphism in-
variant. However, this invariant is redundant, since it can be computed from the
root-grading type, the centroid rank and the root-space rank vector of L.
If L is an fgc centreless Lie algebra that possesses the graded structure of a Lie
torus, we can now unambiguously speak of the root-grading type, the nullity, the
centroid rank and the root-space rank vector of L, since these entities do not depend
on the graded structure.
14
BRUCE ALLISON
Remark 6.3. If L is an fgc centreless Lie torus (or more generally any prime
perfect fgc Lie algebra), the (Tits) index of L is the index, as defined in [34, §2.3],
of the connected component of the automorphism group of the finite dimensional
central simple Lie algebra eL over eC. (See Section 5 for the notation.) The index of
L is a (non-rational) isomorphism invariant of L [7, Lemma 14.1.5]. We won't use
the index in this article. However, to provide a link to recent work on multiloop
algebras [5, 7, 19, 20], we will later display without proof the index of each fgc
centerless Lie torus (see Table 1 and Remark 9.3).
Suppose that L is a Lie torus of type (∆, Λ) and L′ is a Lie torus of type (∆′, Λ′).
An isotopy of L onto L′ is an algebra isomorphism ϕ : L → L′ such that
7. Isotopy
ϕ(Lλ
α) = L′ϕe(λ)+ϕs(α)
ϕr(α)
,
for α ∈ Q and λ ∈ Λ, where ϕr : Q → Q′ and ϕe : Λ → Λ′ are group isomorphisms
and ϕs : Q → Λ′ is a group homomorphism. In that case, it is easy to check using
(LT2)(i) and (LT4) that the maps ϕr, ϕe and ϕs are uniquely determined. It is also
easy to check that the composite of two isotopies is an isotopy and that the inverse
of an isotopy is an isotopy. We say that L and L′ are isotopic9 if there exists an
isotopy from L onto L′.
Finally, we define a bi-isomorphism 10 of L onto L′ to be an isotopy ϕ : L → L′
with ϕs = 0. If such a bi-isomorphism exists we say that L and L′ are bi-isomorphic.
If L is bi-isomorphic to L′, then by definition L is isotopic to L′; however the
converse is not true [5, Example 4.3.1]. Also, if L is isotopic to L′, then by definition
L is isomorphic to L′. We will consider the converse statement in Section 11.
We next show that Λ/Γ(L) is an isotopy invariant of a centreless Lie torus.
Proposition 7.1. Suppose that L and L′ are centreless Lie tori of type (∆, Λ) and
(∆′, Λ′) respectively. If L is isotopic to L′, then Λ/Γ(L) ≃ Λ′/Γ(L′).
Proof. Let ϕ : L → L′ be an isotopy, C = C(L) and C′ = C(L′). Since ϕ is an
isomorphism, we have an induced isomorphism χ : C → C′ as in Remark 1.1(ii).
Then for λ, µ ∈ Λ and α ∈ Q, we have, setting λ′ = ϕe(µ) + ϕs(α), that
χ(C λ)(L′λ′
ϕr(α)) = χ(C λ)ϕ(C µ
α) = ϕ(C λLµ
α) = ϕ(Lµ+λ
α
) = L′ϕe(λ)+λ′
ϕr(α)
.
But for α ∈ Q, ϕe(Λ) + ϕs(α) = Λ′. Hence χ(C λ) ⊆ (C′)ϕe(λ) for λ ∈ Λ. Thus,
since ϕe is invertible, χ(C λ) = (C′)ϕe(λ) for λ ∈ Λ. Hence ϕe(Γ(L)) = Γ(L′), and
therefore ϕe induces the desired isomorphism.
(cid:3)
It does not follow from Proposition 7.1 that Λ/Γ(L) is an isomorphism invariant.
We will consider this issue later in Section 11 for fgc centreless Lie tori.
We have the following simple characterization of isotopies of centreless Lie tori.
9The term isotopic was defined in a different way in [5, Def. 2.2.9] and [8, Def. 5.5], but it is
easy to check that the definitions are equivalent.
10Bi-isomorphism is short for the more suggestive but cumbersome term bi-isograded-
isomorphism.
LIE TORI
15
Theorem 7.2. Suppose that L and L′ are centreless Lie tori of type (∆, Λ) and
(∆′, Λ′) respectively. Let h = L0
If ϕ : L → L′ is an algebra
isomorphism, then
0 and h′ = L′0
0.
ϕ is an isotopy ⇐⇒ ϕ(h) = h′.
Proof. The implication "⇒" is trivial. To prove the reverse implication, suppose
that ϕ(h) = h′.
We use the notation of Section 4 for L, and we set
Λα = {λ ∈ Λ Lλ
α 6= 0}
for α ∈ ∆×. We also use primed versions of this notation for L′. Note that if
α ∈ ∆×, then Λ−α = −Λα by LT2(ii).
Let ϕ : h∗ → (h′)∗ be the transpose of ϕ−1h′ : h′ → h. Then, by (2), ϕ(Lα) =
L′
ϕ(α) for α ∈ h∗. So ϕ(∆) = ∆′ and hence ϕ(Q) = Q′. Let ϕr = ϕQ : Q → Q′.
Then ϕr : Q → Q′ is a group isomorphism such that ϕr(∆) = ∆′ (and hence also
ϕr(∆×) = ∆′×) and
ϕ(Lα) = L′
ϕr(α)
for α ∈ Q.
α), ϕ(f λ
Next let α ∈ ∆×. If λ ∈ Λα, then 0 6= eλ
α )] ∈ h′. So, by Lemma 3.6, we have ϕ(eλ
Thus, since ϕ(h) = h′, we have 0 6= ϕ(eλ
[ϕ(eλ
L′−ρα(λ)
L′ρα(λ)
ρα : Λα → Λ′
−ϕr(α) for some ρα(λ) ∈ Λ′
−α) = L′−ρα(λ)
ϕr(α) and ϕ(L−λ
α ∈ Lα, 0 6= f λ
α) ∈ L′
α, f λ
α ] ∈ h.
−ϕr(α) and
α ) ∈
α) =
−ϕr(α). Since ϕ is an isomorphism, we have a bijection
α ) ∈ L′
ϕr(α) and ϕ(f λ
ϕr(α). So counting dimensions, we have ϕ(Lλ
α ∈ L−α and [eλ
ϕr(α), 0 6= ϕ(f λ
α) ∈ L′ρα(λ)
ϕr(α) such that
(8)
for λ ∈ Λα.
ϕ(Lλ
α) = L′ρα(λ)
ϕr(α)
and ϕ(L−λ
−α) = L′−ρα(λ)
−ϕr(α)
If α ∈ ∆× and λ ∈ Λα, we have ϕ(L−λ
−α) = L′ρ−α(−λ)
ϕr(−α)
since −λ ∈ −Λα = Λ−α.
Comparing this with the second equation in (8), we obtain
(9)
ρ−α(−λ) = −ρα(λ)
We next claim that if α, β ∈ ∆×, λ ∈ Λα and µ ∈ Λβ, we have11
(10)
and
µ − hβ, α∨iλ ∈ Λwα(β)
ρwα(β)(µ − hβ, α∨iλ) = ρβ(µ) − hβ, α∨iρα(λ).
(11)
Indeed, this is clear if hβ, α∨i = 0. Next, suppose hβ, α∨i < 0. Then, by Lemma
3.5, we have 0 6= ad(Lλ
, which implies (10). Moreover,
wα(β)
counting dimensions, we see that Lµ−hβ,α∨iλ
get
β ⊆ Lµ−hβ,α∨iλ
β. Applying ϕ we
α)−hβ,α∨iLµ
α)−hβ,α∨iLµ
= ad(Lλ
wα(β)
L
ρwα (β)(µ−hβ,α∨iλ)
ϕr (wα(β))
= ad(Lρα(λ)
ϕr(α))−hβ,α∨iLρβ (µ)
ϕr(β),
which implies (11). Finally, if hβ, α∨i > 0, then hβ, (−α)∨i < 0 and −λ ∈ −Λα =
Λ−α. Hence, by our previous case, we have (10) and (11) with α replaced by −α
11The equalities (10) and (12) are well-known (see for example [5, §1.1] and the earlier refer-
ences there), but they arise naturally here so we give the arguments.
16
BRUCE ALLISON
and λ replaced by −λ, which gives (10) and (11) for α and λ using (9). So we have
the claim.
To simplify notation, we now denote the reduced irreducible finite root system
∆ind by E. Let W denote the Weyl group of ∆ (= the Weyl group of E).
If
α ∈ E×, then 0 ∈ Λα by LT(i). So by (10) (with λ = 0), we see that Λβ ⊆ Λwα(β)
for α ∈ E×, β ∈ ∆×. Hence Λβ = Λw(β) for β ∈ ∆× and w ∈ W . Thus
(12)
if α, β ∈ ∆× have the same length.
Λα = Λβ
Define σ : E× → Λ′ by σ(α) = ρα(0). Putting λ = µ = 0 in (11), we obtain
σ(wα(β)) = σ(β) − hβ, α∨iσ(α)
(13)
for α, β ∈ E×. Let {α1, . . . , αr} be a base for the root system ∆, and choose
ϕs ∈ HomZ(Q, Λ′) such that ϕs(αi) = σ(αi) for 1 ≤ i ≤ r. Define δ : E× → Λ by
δ(α) = σ(α) − ϕs(α). Then, since ϕs is Z-linear, it follows from (13) that
δ(wα(β)) = δ(β) − hβ, α∨iδ(α)
(14)
for α, β ∈ E×. Now the set X := {α ∈ E× δ(α) = 0} contains {α1, . . . , αr}; and
so, by (14), X is stable under the action of W . Since E is reduced, this implies
that X = E×, so σ(α) = ϕs(α) for α ∈ E×. Hence
ρα(0) = ϕs(α)
for α ∈ E×.
Next for α ∈ E×, we define τα : Λα → Λ′ by
(15)
τα(λ) = ρα(λ) − ϕs(α).
Observe that τα(0) = 0.
Suppose that α, β ∈ E×. Then, since ϕs is Z-linear, we have ϕs(wα(β)) =
ϕs(β) − hβ, α∨iϕs(α). Subtracting this from (11) we see that
(16)
τwα(β)(µ − hβ, α∨iλ) = τβ(µ) − hβ, α∨iτα(λ)
for λ ∈ Λα, µ ∈ Λβ. Taking λ = 0, we have τwα(β)(µ) = τβ(µ) for µ ∈ Λβ. Hence
(17)
for β ∈ E× and w ∈ W .
τw(β) = τβ
Now fix a short root γ in E×, and let S = Λγ, which does not depend on the
choice of γ by (12). It is known that 0 ∈ S, −S = S, S + 2Λ ⊆ S, Λα ⊆ S for
α ∈ E× and S generates the group Λ (see for example [5, Lemma 1.1.12]). Hence
S contains a Z-basis {ν1, . . . , νn} for Λ [1, Prop. II.1.11].
We define τ : S → Λ′ by τ = τγ, which does not depend on the choice of γ by
(17). We claim next that
(18)
for α in E×. Indeed, if α has the same length as γ, we already know that (18)
holds. So we can assume that α is long and hγ, α∨i = −1. But then taking β = γ
and µ = 0 in (16), we see that τwα(γ)(λ) = τα(λ) for λ ∈ Λα, and so τ (λ) = τα(λ)
for λ ∈ Λα.
τα = τ Λα
Next taking α = γ and β = −γ in (16), we see using (17) that
for µ, λ ∈ S.
τ (µ + 2λ) = τ (µ) + 2τ (λ)
LIE TORI
17
Define ϕe ∈ Hom(Λ, Λ′) by ϕe(νi) = τ (νi) for 1 ≤ i ≤ n. Further, define
ε : S → Λ′ by ε(λ) = τ (λ) − ϕe(λ). Then ε(νi) = 0 for 1 ≤ i ≤ n and
(19)
ε(µ + 2λ) = ε(µ) + 2ε(λ)
for µ, λ ∈ S. So, taking µ = −λ, we have ε(−λ) = −ε(λ) for λ ∈ S. Hence
ε(±νi) = 0 for 1 ≤ i ≤ n.
It follows by induction on k using (19) that ε(µ+2Pk
i=1 ε(λi)
for µ, λ1, . . . , λk ∈ S. But each λ ∈ S is the sum of elements from {±ν1, . . . , ±νn}
and ε vanishes on the elements of this set. So we have ε(µ+ 2λ) = ε(µ) for µ, λ ∈ S.
Therefore by (19), 2ε(λ) = 0 for λ ∈ S, and hence, since Λ has no 2-torsion, ε = 0.
So τ (λ) = ϕe(λ) for λ ∈ S. Thus, by (15) and (18), we have
i=1 λi) = ε(µ)+2Pk
(20)
ρα(λ) = ϕe(λ) + ϕs(α).
for α ∈ E×, λ ∈ Λα. So by (8), we have
(21)
ϕ(Lλ
α) ⊆ L′ϕe(λ)+ϕs(α)
ϕr(α)
for α ∈ E×, λ ∈ Λα. But, by Lemma 3.7, every element of L is the sum of products
of elements chosen from Lλ
α, α ∈ E×, λ ∈ Λ. So (21) holds for α ∈ Q, λ ∈ Λ.
Finally, the isomorphism ϕ−1 : L′ → L satisfies an inclusion of exactly the same
form as (21). Using this it is easy to check that ϕe : Λ → Λ′ is an isomorphism and
hence that equality holds in (20) for α ∈ Q, λ ∈ Λ. We leave these arguments to
the reader.
(cid:3)
8. The structure of fgc centreless Lie tori
For the rest of the article we assume that k is algebraically closed.
In this section, we recall the structure theorems for fgc centreless Lie tori. We
combine these results into one theorem, which states that any fgc centreless Lie
torus is either classical or exceptional.
Classical Lie tori and, in several cases, exceptional Lie tori are constructed from
associative tori. So we begin the section with a discussion of these graded algebras.
Associative tori. Recall [35] that an associative Λ-torus (or simply an associative
torus) is a Λ-graded unital associative algebra A =Lλ∈Λ Aλ such that every Aλ
is spanned by an invertible element for λ ∈ Λ. (Equivalently, A is a twisted group
algebra of Λ over k.) In that case, we call the rank of the group Λ the nullity of A.
It is easy to check that if A is an associative Λ-torus, A′ is an associative Λ′-
torus, and ϕ : A → A′ is an algebra isomorphism, there exists a group isomorphism
ϕgr : Λ → Λ′ such that ϕ(Aλ) = A′ϕgr(λ) for λ ∈ Λ. Thus it is not necessary to
distinguish between isomorphism and isograded-isomorphism for associative tori.
If A is an associative Λ-torus, we set Γ(A) := suppΛ(Z(A)). Then Γ(A) is a
subgroup of Λ and Z(A) is a commutative associative Γ(A)-torus.
It is easily checked (and well-known) that any associative torus A is a domain
and hence prime (as a k-algebra or equivalently as a ring).
1 , . . . , t±1
The simplest example of an fgc associative torus is the Zn-associative torus
Rn = F [t±1
n ] with its natural Zn-grading. (If n = 0, Rn = k is graded by
Z0 = {0}.) Another important example is obtained as follows. Let ζ ∈ k× and
let Q(ζ) be the algebra presented by the generators x±1
subject to the inverse
relations xix−1
i xi = 1, i = 1, 2, and the relation x1x2 = ζx2x1. Then Q(ζ),
i = x−1
1 , x±1
2
18
BRUCE ALLISON
with its natural Z2-grading, is an associative Z2-torus which is fgc if and only if ζ
is a root of unity. We call Q(ζ) the quantum torus determined by ζ.
If Ai is an associative Λi-torus for 1 ≤ i ≤ k, then A1 ⊗ · · · ⊗ Ak is an associative
Λ-torus with Λ = Λ1 ⊕ · · · ⊕ Λk. Moreover,
Z(A1 ⊗ · · · ⊗ Ak) = Z(A1) ⊗ · · · ⊗ Z(Ak),
and A1 ⊗ · · · ⊗ Ak is fgc if and only if each Ai is fgc.
Any fgc associative torus is isomorphic to a tensor product
A1 ⊗ · · · ⊗ Ak ⊗ Rq,
(22)
where k ≥ 0, q ≥ 0 and Ai ≃ Q(ζi) with ζi a root of unity 6= 1 in k× for i = 1, . . . , k.
Moreover, the ζi's can be chosen satisfying further restrictions, and under those
restrictions Neeb has given necessary and sufficient conditions for isomorphism (or
equivalently isograded-isomorphism) of two such tensor products [25, Thm. 4.5]
(although a subtle point about determinants of certain integral matrices is not
resolved -- see [25, Conjecture 4.2]).
Associative tori with involution. An associative Λ-torus with involution is a
Λ-graded associative algebra with involution (A, −) such that A is an associative
Λ-torus.
If (A, −) is an associative Λ-torus with involution, we use the notation Γ(A, −) :=
suppΛ(Z(A, −)). Then Γ(A, −) is a subgroup of Λ and Z(A, −) is a commutative
associative Γ(A, −)-torus. Also we have
(23)
Z(A) = Z(A, −) ⊕ (Z(A) ∩ A−),
and we say that (A, −) is of first kind (resp. second kind ) if Z(A) = Z(A, −)
(resp. Z(A) 6= Z(A, −)). If (A, −) is of second kind, then there exists a nonzero
homogeneous element s0 ∈ Z(A) ∩ A−, and for any such s0 we have
(24)
Hence
(25)
Z(A) ∩ A− = s0Z(A, −) and A− = s0A+.
[Γ(A) : Γ(A, −)] = 1 or 2
according as (A, −) is of first or second kind.
Four basic examples of associative tori with involution are
(Rn, 1),
(R1, ♮),
(Q(−1), ♮)
and (Q(−1), ∗),
graded by Zn, Z1, Z2 and Z2 respectively, where the standard involution ♮ of R1
anti-fixes the generator x1 (x♮
1 = −x1); the standard involution ♮ of Q(−1) anti-
fixes the generators x1 and x2; and the reversal involution ∗ of Q(−1) fixes the
generators x1 and x2.12
If (Ai, −) is an associative Λi-torus with involution for 1 ≤ i ≤ k, then (A1, −) ⊗
· · · ⊗ (Ak, −) is an associative Λ-torus with involution, where Λ = Λ1 ⊕ · · · ⊕ Λk;
and we have
Z((A1, −) ⊗ · · · ⊗ (Ak, −)) = Z(A1, −) ⊗ · · · ⊗ Z(Ak, −).
Any associative torus with involution (A, −) is isomorphic (or equivalently iso-
graded-isomorphic) to a unique tensor product of the form
(26)
(A1, −) ⊗ · · · ⊗ (Ak, −) ⊗ (Ak+1, −) ⊗ (Rq, 1),
12The term reversal involution is used since ∗ reverses the order of products of the generators
x±1
1 , x±1
2 . So (xi1
1 xi2
2 )∗ = xi2
2 xi2
1 = (−1)i1 i2 xi1
1 xi2
2 for i1, i2 ∈ Z.
LIE TORI
19
where k ≥ 0, q ≥ 0, (Ai, −) ≃ (Q(−1), ♮) for i = 1, . . . , k, and (Ak+1, −) is
isomorphic to one of the associative tori with involution (k, 1), (R1, ♮) or (Q(−1), ∗)
(see [36, Thm. 2.7] or [9, Remark 5.20]). In that case (A, −) is of second kind if
and only if (Ak+1, −) ≃ (R1, ♮).
We will use the following lemmas about associative tori.
Lemma 8.1. Suppose that (A, −) is an associative torus with involution. If (A, −)
is not isomorphic to (Q(−1), ♮) ⊗ (Rq, 1) for q ≥ 0, then [A−, A−] ⊆ A+A+.
Proof. Now (A, −) is isomorphic to an associative torus with involution of the
form (26). If (Ak+1, −) ≃ (R1, ♮), then (A, −) is of second kind, and choosing s0
as in (24), we have [A−, A−] = [s0A−, s−1
0 A−] ⊆ A+A+. Also, if k = 0, then
[A−, A−] = 0.
To complete the proof we assume that k ≥ 1, (Ak+1, −) ≃ (k, 1) or (Q(−1), ∗),
and, if k = 1, (Ak+1, −) ≃ (Q(−1), ∗). We show by induction that
(27)
A−A− = A and A+A+ = A.
First, if k = 1, then (A, −) ≃ (Q(−1), ♮) ⊗ (Q(−1), ∗) ⊗ Rq and (27) is easily
checked. Suppose next that k ≥ 2. When (A, −) ≃ (Q(−1), ♮) ⊗ (Q(−1), ♮) ⊗ Rq,
(27) is again easily checked. Otherwise, we can identify (A, −) = (B, −) ⊗ (C, −),
where (B,−) ≃ (Q(−1), ♮) and (C, −) is of the form needed to apply our induction
hypothesis. Thus, A+A+ ⊇ (B− ⊗ C−)(B− ⊗ C−) = B−B− ⊗ C−C− = B ⊗ C = A
and A−A− ⊇ (B− ⊗ C+)(B− ⊗ C+) = B−B− ⊗ C+C+ = B ⊗ C = A.
(cid:3)
Lemma 8.2.
(i) Suppose that A is an fgc associative Λ-torus. Then [Λ : Γ(A)] is finite.
Further, if X is a graded Z(A)-submodule of A, then X is a free Z(A)-
module of rank ≤ [Λ : Γ(A)], with equality holding if X = A.
(ii) Suppose that (A, −) is an associative Λ-torus with involution. Then A is
fgc and [Λ : Γ(A, −)] is finite. Further, if X is a graded Z(A, −)-submodule
of A, then X is a free Z(A, −)-module of rank ≤ [Λ : Γ(A, −)], with equality
holding if X = A.
Proof. i): This is well-known (see [4, Remark 4.4.2] and the earlier references there),
but we indicate a proof for the convenience of the reader and as a model for the
proof of (ii). Let X be a graded Z(A)-submodule of A, and let X = suppΛ(X ).
Then Γ(A) + X ⊆ X. Thus, X is the union of cosets of Γ(A) in Λ, so we can choose
a set of representatives {µi}i∈I of these cosets. Further, choose 0 6= mi ∈ Aµi
for i ∈ I. Then {mi}i∈I is a Z(A)-basis for X , so X is a free Z(A)-module of
rank equal to the cardinality of I. In particular, A is a free Z(A)-module of rank
[Λ : Γ(A)], which must therefore be finite since A is fgc.
(ii): The component associative tori in the tensor product decomposition (26) of
(A, −) are fgc, and hence so is A. So by (i), [Λ : Γ(A)] is finite, and hence, by (25)
[Λ : Γ(A, −)] is finite. The rest of the proof of (ii) is similar to the proof of (i). (cid:3)
Classical Lie tori. We next recall constructions of some fgc centreless Lie tori
of root-grading type Ar, r ≥ 1; BCr or Br, r ≥ 1; Cr, r ≥ 1; and Dr, r ≥ 4
respectively. Here types B1 and C1 should be interpreted as A1, and type C2
should be interpreted as B2.
In each of these constructions, we use Ms(A) to denote the associative algebra
of s × s matrices over A if s ≥ 1 and A an associative algebra. Note that Ms(A) is
20
BRUCE ALLISON
therefore also a Lie algebra under the commutator product. Furthermore Ms(A) is
a free left A-module with basis {eij}1≤i,j≤s, where the action of A on Ms(A) is by
left multiplication on entries and where eij denotes the (i, j)-matrix unit.
In the last three constructions we will use the notation Jp := (δi,p+1−j) ∈ Mp(k),
for p ≥ 1. In other words, Jp is the p × p matrix with ones on the anti-diagonal
and zeroes elsewhere.
Constructions 8.3.
(A): [13, §2], [8, §10], [29, §4.4].13 Suppose that r ≥ 1 and A is an fgc associative
Λ-torus. Let L = slr+1(A) be the derived algebra of the Lie algebra Mr+1(A) under
the commutator product. More explicitly, one easily checks that
(28)
L = slr+1(A) = {X ∈ Mr+1(A) tr(X) ∈ [A, A]},
where [A, A] is the space spanned by commutators in A. Let h = Pr
i=1 k(eii −
ei+1,i+1). Then h is a split toral k-subalgebra of L with irreducible finite root
system ∆ = ∆k(L, h) of type Ar. Moreover L is an fgc centreless Lie torus of type
(∆, Λ), where the Q-grading of L is the root-space decomposition relative to h and
the Λ-grading of L is induced by the Λ-grading of A. We call L the (r + 1) × (r + 1)-
special linear Lie torus over A.
(BC -- B): [1, §III.3], [2, §7.2]. Suppose that r ≥ 1, L is a finitely generated free
abelian group (which we will embed in a larger group Λ of the same rank below), and
(A, −) is an associative L-torus with involution. Suppose also that m ≥ 1 and D =
diag(d1, . . . , dm) ∈ Mm(A), where d1, . . . , dm are nonzero homogeneous hermitian
elements of A whose respective degrees δ1, . . . , δm in L are distinct modulo 2L
with d1 = 1 and δ1 = 0. To eliminate overlap with the other constructions, we
assume that if r = 1 and − = 1, then m ≥ 5. Let G = diag(J2r, D) in block
diagonal form, and let L = su2r+m(A, −, D) be the derived algebra of the Lie
algebra {X ∈ M2r+m(A) G−1 ¯X tG = −X} under the commutator product. More
explicitly we have [2, §7.2.3]
L = su2r+m(A, −, D) = {X ∈ M2r+m(A) G−1 ¯X tG = −X, tr(X) ∈ [A, A]}.
2 δ1, . . . , 1
2 τi − 1
To describe the external grading on L, we first embed L in the rational vector space
Q ⊗Z L and let Λ be the subgroup of Q ⊗Z L generated by L and 1
2 δm.
Further we define τi ∈ L for 1 ≤ i ≤ 2r + m by τi = 0 for 1 ≤ i ≤ 2r and τ2r+i = δi
for 1 ≤ i ≤ m. Then the associative algebra M2r+m(A) is Λ-graded by assigning
the degree λ + 1
2 τj to each element in Aλeij for λ ∈ L, 1 ≤ i, j ≤ 2r + m;
and one checks directly that the involution X 7→ G−1 ¯X tG of M2r+m(A) is Λ-
graded. Consequently, the Lie algebra M2r+m(A) under the commutator product
is Λ-graded, and L is a Λ-graded subalgebra of this algebra. To describe the root
i=1 k(eii − e2r+1−i,2r+1−i). Then h is a split toral k-
subalgebra of L with irreducible finite root system ∆ = ∆k(L, h), and the type
of ∆ is BCr if − 6= 1 and Br if − = 1. Also, the root-space decomposition of
L relative to h is a Q-grading of L which is compatible with the Λ-grading just
described. With the resulting Q × Λ-grading, L is an fgc centreless Lie torus of
grading on L, let h = Pr
13In some of the references cited in this section, additional assumptions (such as k = C or
r ≥ 2) are made that can be checked to be unnecessary for our purposes.
LIE TORI
21
type (∆, Λ).14 We call L the (2r + m) × (2r + m)-special unitary Lie torus over
(A, −) determined by D.
(C): [1, §III.4], [8, §11]. Suppose that r ≥ 1 and (A, −) is an associative Λ-torus
with involution. To avoid degenerate cases and eliminate overlap with the other
constructions, we assume that if r = 1 or 2, then (A, −) is not isomorphic to (Rq, 1),
(R1, ♮) ⊗ (Rq, 1) or (Q(−1), ♮) ⊗ (Rq, 1) for q ≥ 0. Let G =(cid:2) 0 Jr
block form, and let L = ssp2r(A, −) be the derived algebra of the Lie algebra
{X ∈ M2r(A) G−1 ¯X tG = −X} under the commutator product. Once again, we
have more explicitly that
−Jr 0 (cid:3) ∈ M2r(k) in
L = ssp2r(A, −) = {X ∈ M2r(A) G−1 ¯X tG = −X, tr(X) ∈ [A, A]}.
using Lemma 8.1. Let h = Pr
Indeed, if r ≥ 2 this is easily checked directly, whereas if r = 1 it is easily checked
i=1 k(eii − e2r+1−i,2r+1−i). Then h is a split toral
k-subalgebra of L with irreducible finite root system ∆ = ∆k(L, h) of type Cr (see
the proof of Proposition 9.2 below for this calculation), and L is an fgc centreless
Lie torus of type (∆, Λ) with gradings determined by h and A as in (A) above. We
call L the (2r) × (2r)-special symplectic Lie torus over (A, −).
(D): Suppose that r ≥ 4 and A = Rn with its natural grading by Λ = Zn. Let
L = o2r(A) := {X ∈ M2r(A) J −1
2r X tJ2r = −X}.
Then L is a Lie algebra under the commutator product. Let h = Pr
i=1 k(eii −
e2r+1−i,2r+1−i). Then h is a split toral k-subalgebra of L with irreducible finite root
system ∆ = ∆k(L, h) of type Dr, and L is an fgc centreless Lie torus of type (∆, Λ)
with gradings determined by h and A as in (A) above. In fact, L ≃ o2r(k) ⊗ Rn
is just the untwisted Lie torus of type (∆, Zn) (see Example 3.4), viewed as an
algebra of matrices. We call L the (2r) × (2r)-orthogonal Lie torus over A.
maximal split toral k-subalgebra L0
We note that in each of the constructions, the indicated subalgebra h is the
0 of L that was denoted by h in Sections 4 to 7.
We call an fgc centreless Lie torus that arises from any one of the Constructions
(A), (BC -- B), (C) or (D) a classical Lie torus.
Remark 8.4. If we allow r = 0 in Constructions (A) and (BC -- B), we obtain
multiloop Lie algebras that are not Lie tori.15 Indeed, one can show that they
are multiloop Lie algebras (see the discussion following Example 3.4) using the
multiloop realization theorem [4, Cor. 8.3.5]. (The hypotheses of that theorem can
be checked using a base ring extension argument as in Proposition 9.1 below.) Also,
one can show that they do not contain nonzero split toral k-subalgebras (using [7,
§4.5.9] and [2, Prop. 5.2.5]), which shows that they are not Lie tori. Since our
interest in this article is in Lie tori, we omit the details in this remark and we do
not consider the r = 0 case further.
Exceptional Lie tori. We next display in Table 1 a list of fgc centreless Lie
tori that we call exceptional Lie tori. For convenience of reference we have labeled
these Lie tori as #1 -- 27 in the column labeled #. Each row of the table represents
14As an example, if we take (A, −) = (R2, 1), r = 1 and D = diag(1, t1, t2), su5(A, −, D) is
the centreless Lie torus whose universal central extension is called the baby TKK algebra in [33].
15A few low rank cases must be excluded but these are easy to identify.
22
BRUCE ALLISON
exactly one Lie torus of nullity n for each n ≥ n0, where the minimum nullity n0 is
displayed in the second column (not counting the # column) of the table.16
crk(L) rkv(L)
Root-
grading
type
#
n0
"
"
"
"
"
"
A1
1
A2
2
C3
3
E6
4
E7
5
E8
6
7 G2
8
9
10
11 F4
12
13
14
15 BC1
16
17
18
19
20
21
22
23
24
25 BC2
26
27
"
"
"
"
"
"
"
"
"
"
"
3
3
3
0
0
0
0
1
2
3
0
1
2
3
3
4
5
6
5
6
7
5
3
3
3
4
5
133
78
133
78
133
248
14
28
78
248
52
78
133
248
52
78
133
248
78
133
248
133
133
248
78
133
248
7,7
6,6
6,4
Λ/Γ(L)
Z3
(27)
3
Z3
(8)
2
Z3
(8, 1)
2
{0}
(1)
{0}
(1)
{0}
(1)
{0}
(1, 1)
Z3
(3, 1)
Z2
(9, 1)
3
Z3
(27, 1)
3
{0}
(1, 1)
Z2
(2, 1)
Z2
(4, 1)
2
Z3
(8, 1)
2
Z3
(8, 1)
2
Z4
(16, 8)
2
Z5
(32, 10)
2
Z6
(64, 14)
2
Z5
(20, 1)
2
Z6
(32, 1)
2
Z7
(56, 1)
2
Z5
(32, 1)
2
Z2 ⊕ Z2
(32, 1)
Z3
(56, 1)
4
Z3
(8, 12, 1)
2
(16, 16, 1) Z4
2
(32, 24, 1) Z5
2
Index Reference
E78
7,1
1E28
6,2
E28
7,3
1E0
E0
E0
8,8
G0
2,2
3D2
4,2
1E16
6,2
E78
8,2
F0
4,4
2E2
E9
7,4
E28
8,4
F21
4,1
2E29
6,1
E48
7,1
E91
8,1
2E35
6,1
E66
7,1
E133
8,1
E66
7,1
4 E66
7,1
E133
8,1
2E16′
6,2
E31
7,2
E66
8,2
[35, Example 6.8(3)]
[8, Example 9.2]
[10, Thm. 4.87(ii)]
untwisted
untwisted
untwisted
untwisted
[10, Thm. 5.63, p=1]
[10, Thm. 5.63, p=2]
[10, Thm. 5.63, p=3]
untwisted
[10, Thm. 5.50, p=1]
[10, Thm. 5.50, p=2]
[10, Thm. 5.50, p=3]
[9, Thm. 5.19(b), k=0]
[9, Thm. 5.19(b), k=1]
[9, Thm. 5.19(b), k=2]
[9, Thm. 5.19(b), k=3]
[9, Thm. 10.6(a), case 1]
[9, Thm. 10.6(a), case 2]
[9, Thm. 10.6(a), case 3]
[9, Remark 10.6(a)]
[9, Thm. 13.3, case 1]
[9, Thm. 13.3, case 2]
[18, Lem. 7, n = 0]
[18, Lem. 7, n = 1]
[18, Lem. 7, n = 2]
Table 1. Exceptional Lie tori and their invariants
We do not provide here precise definitions of the exceptional Lie tori, because
to do so would take us rather far afield into the fascinating world of nonassociative
16For the Lie tori numbered 25, 26 and 27, there are parameters σ0 and µ in the description
given in [18]. However, one can argue as in [8, §10], that the Lie torus does not depend on these
choices up to bi-isomorphism.
LIE TORI
23
tori. Instead, in each case we have given a reference for the definition in the last
column of the table.17 If the Lie torus is the untwisted Lie torus with the indicated
root-grading type and nullity n (see Example 3.4), we indicate this simply with
the word untwisted. In the case of the Lie tori numbered 15 -- 24 (resp. 25 -- 27), the
Lie torus is constructed using the Kantor construction from a structurable torus
(resp. quasi-torus) that is defined in the indicated reference. (See [11, Thm. 5.6]
and [18, Thm. 3].) Also, for the Lie tori numbered 25 -- 27, the quantity n used in
the last column is the integer denoted by n in [18, Lemma 7].
For each exceptional Lie torus L, Columns 1, 3 and 4 of the table contain the
isomorphism invariants of L (besides the nullity) that are described in Theorem
6.2, namely the root-grading type of L, the centroid rank of L and the root-space
rank vector of L respectively. Column 5 contains the isotopy invariant described
in Proposition 7.1, namely the quotient external-grading group Λ/Γ(L) (up to iso-
morphism) of L.18
Finally, Column 6 contains the index of L (see Remark 6.3). These indices were
calculated using Tits' classification of indices [34, Table II], Theorem 5.4, and the
entries in Columns 1, 3 and 4, together with some special arguments in a few cases.
(See [7, §14.2] for some similar calculations.)
The structure theorem. In about the last 15 years, structure theorems (coordi-
natization theorems) have been proved for centreless Lie tori of each root-grading
type. This is work of (in alphabetical order) Allison, Benkart, Berman, Faulkner,
Gao, Krylyuk, Neher and Yoshii in various combinations beginning with [13]. The
reader can consult Section 7 of the survey article [8] for precise references.
It turns out from these theorems that the only centreless Lie tori that are not
fgc are the Lie tori slr+1(A) defined exactly in (A) using an associative torus A
that is not fgc.
The following theorem summarizes the results of the structure theorems for fgc
centreless Lie tori. There is some work needed to translate the known results into
our form, but it is not difficult to supply these arguments and we omit them.
Theorem 8.5. If k is algebraically closed, every fgc centreless Lie torus is bi-
isomorphic and hence isotopic and isomorphic to either a classical Lie torus or an
exceptional Lie torus.
9. Invariants of classical Lie tori
In this section, we calculate the invariants described in Theorem 6.2 and Propo-
sition 7.1 for classical Lie tori. For this, we first need to calculate the centroid in
each case.
Proposition 9.1. Let L be slr+1(A) as in (A), su2r+m(A, −, D) as in (BC -- B),
ssp2r(A, −) as in (C), or o2r(A) as in (D). Correspondingly let M be Mr+1(A),
M2r+m(A), M2r(A) or M2r(A), in which case L is a Lie subalgebra of M under
the commutator product. Also correspondingly, let Z be Z(A), Z(A, −), Z(A, −)
or A, and regard M as a Lie algebra over Z, where the action of Z on M is
17We only cite the reference that we find most convenient in our context. Additional and
sometimes earlier references can be found in the cited articles as well as in Section 7 of the survey
article [8].
18In this column and subsequently, we denote the direct sum of s copies of the group Zℓ of
integers mod ℓ by Zs
ℓ for ℓ ≥ 1 and s ≥ 0. (If s = 0 this direct sum is 0.)
24
BRUCE ALLISON
by left multiplication on entries. Then L is a Z-subalgebra of M and the map
ρ : Z → Ck(L) defined by ρ(z)(x) = zx is a Λ-graded algebra isomorphism.
Proof. This can be proved using Corollary 5.16 and Theorem 4.18 of [12], although
care must be taken in low rank. Instead, we present an argument using base ring
extension and results about finite dimensional simple Lie algebras from [22, Chap.
X]. We record this for the algebra L = slr+1(A) as in (A), with the other cases
being similar.
It is clear that L is a Z-subalgebra of M and that that ρ is an injective Λ-graded
algebra homomorphism. So it remains to show that ρ is surjective. Let C = Ck(L)
and use ρ to regard C as an algebra over Z.
Now Z ≃ k[Ω] and C ≃ k[Γ] as Λ-graded algebras, where Ω and Γ are subgroups
of Λ. (The first statement is clear and the second is (4).) Hence, Ω is a subgroup
of Γ and C is a free Z-module of rank [Γ : Ω]. So, to show that ρ is surjective, it
suffices to show that rankZ (C) ≤ 1.
Note that L is a free Z-module (for example since C is a free Z-module and L
is a free C-module by Proposition 5.1(ii)).
Next, since L is perfect, we have C = CZ(L), where CZ (L) = {c ∈ EndZ (L)
c[x, y] = [c(x), y] = [x, c(y)] for x, y ∈ L}. So we have a natural Z-algebra homo-
morphism
Z ⊗Z C = Z ⊗Z CZ (L) 7→ C Z ( Z ⊗Z L),
(29)
where Z is the quotient field of Z. We claim that this map is injective. Indeed, any
element of Z ⊗Z C is of the form z−1 ⊗ c, where 0 6= z ∈ Z and c ∈ C. But if this
element is in the kernel of the map (29) then so is 1 ⊗ c. So 1 ⊗ cx = 0 for x ∈ L,
which implies that cx = 0 for x ∈ L, since L is a free Z-module. Thus c = 0, and
we have proved the claim. So it suffices to show that dim Z (C Z ( Z ⊗Z L)) ≤ 1, or
in other words that Z ⊗Z L is central over Z.
Since A is a free Z-module by Lemma 8.2, A embeds naturally in the Z-algebra
Z ⊗Z A. Moreover, since A is an fgc domain, it is easily checked that Z ⊗Z A is a
finite dimensional central division algebra over Z. Also, by definition, L = [M, M],
so Z ⊗Z L = Z ⊗Z [M, M] ≃ [ Z ⊗Z M, Z ⊗Z M] as Z-algebras, where the last holds
since Z/Z is a flat extension. But Z ⊗Z M = Z ⊗Z Mr+1(A) ≃ Mr+1( Z ⊗Z A), so
Z ⊗Z L = [Mr+1( Z ⊗Z A), Mr+1( Z ⊗Z A)].
Thus by [22, Thm. X.8], Z ⊗Z L is a finite dimensional central simple Lie algebra
over Z.
(cid:3)
Parameterization of classical Lie tori. To tabulate the invariants of classical
Lie tori, we need to view each of the Constructions (A), (BC -- B), (C) and (D) as a
construction from a list of parameters. We now do this using for the most part the
tensor product decomposition of associative tori.
(A): Suppose L = slr+1(A) is a special linear Lie torus as in Construction 8.3(A).
Then, as noted in Section 8, we can assume that
A = A1 ⊗ · · · ⊗ Ak ⊗ Rq,
where k ≥ 0, q ≥ 0 and Ai = Q(ζi) with ζi a root of unity 6= 1 in k× for i = 1, . . . , k.
So we can view L as being constructed from the parameters
r ≥ 1, k ≥ 0, ζ1, . . . , ζk ∈ k× and q ≥ 0.
LIE TORI
25
The only restrictions on the integral parameters r, k and q are those indicated, and
the only restrictions on the parameters ζ1, . . . , ζk are
where ζi denotes the order of ζi in the group k×.
2 ≤ ζi < ∞ for 1 ≤ i ≤ k,
(BC -- B): Suppose L = su2r+m(A, −, D) is a special unitary Lie torus as in
Construction 8.3(BC -- B). Then as noted in Section 8 we can assume that
(A, −) = (A1, −) ⊗ · · · ⊗ (Ak, −) ⊗ (Ak+1, −) ⊗ (Rq, 1),
where k ≥ 0, q ≥ 0,
(30)
and
(31)
(Ai, −) = (Q(−1), ♮)
for i = 1, . . . , k,
(Ak+1, −) = (k, 1), (R1, ♮) or (Q(−1), ∗)
(as associative tori). Corresponding to these 3 choices for (Ak+1, −) we set
p := 0, 1 or 2.
Note that the grading group for (A, −) is L = L1 ⊕ · · · ⊕ Lk+2, where L1, . . . , Lk =
Z2, Lk+1 = Zp and Lk+2 = Zq. Moreover the set L+ consisting of the degrees
of nonzero homogeneous elements in A+ is then determined by k, p and q. (For
example if k = 1, p = 1 and q ≥ 0, we have L+ = 2L + Lk+2 + {0, ε11 + ε21, ε12 +
ε21, ε11 + ε12 + ε21}, where {ε11, ε12} is a Z-basis for L1, and {ε21} is a Z-basis for
L2.) Recall that D = diag(d1, . . . , dm) and that δi is the degree of di in L. We note
that if the elements d2, . . . , dm are replaced by nonzero scalar multiples (d1 = 1 is
fixed), then L is not changed up to isomorphism (in fact bi-isomorphism) [2, Cor.
6.6.4]. Hence, we can view L as being constructed from the parameters
(32)
r ≥ 1, k ≥ 0, p ∈ {0, 1, 2}, q ≥ 0, m ≥ 1 and δ1, . . . , δm ∈ L+.
The restrictions on the integral parameters r, k, p, q, m are those indicated as well
as the additional restriction
(33)
m ≥ 5 if (r, k, p) = (1, 0, 0)
imposed in Construction 8.3(BC -- B). The restrictions on the δi's in L+ are that
δ1 = 0
and δi + 2L 6= δj + 2L for i 6= j.
(C): Suppose L = ssp2r(A, −) is a special symplectic Lie torus as in Construction
8.3(C). Then we can assume as in (BC -- B) above that
(34)
(A, −) = (A1, −) ⊗ · · · ⊗ (Ak, −) ⊗ (Ak+1, −) ⊗ (Rq, 1)
where k, q ≥ 0 and (A1, −), . . . , (Ak+1, −) satisfy (30) and (31). Again, we define
(35)
p = 0, 1 or 2
corresponding to the choice of (Ak+1, −) in (31). Then we can view L as constructed
from the integral parameters
(36)
r ≥ 1, k ≥ 0, p ∈ {0, 1, 2} and q ≥ 0,
subject to the indicated restrictions as well as the additional restriction
(37)
(k, p) 6= (0, 0), (0, 1), (1, 0)
if r = 1 or 2
imposed in Construction 8.3(C).
26
BRUCE ALLISON
(D): Suppose finally that L = o2r(Rq) is an orthogonal Lie torus as in Construc-
tion 8.3(D), where q ≥ 0. Then L is constructed from the integral parameters
r ≥ 4 and q ≥ 0.
The invariants of classical Lie tori. We can now calculate the invariants of
classical Lie tori. These will appear in Tables 2 and 3, where we use the parameter-
izations described above. In the tables and subsequently, we also use the following
additional notation:
• For each of the four constructions, we define two additional positive integers
d and s in the Construction column of Table 2. In the definition of d in
(BC -- B) and (C), ⌊ ⌋ is the floor function, so that d = 2k if p = 0 or 1, and
d = 2k+1 if p = 2.
indicates that the entry is to be omitted when r = 1.
• In the last column of Table 2, the symbol b above an entry of a vector
• In Construction (BC -- B), L/2L (resp. L/(2L + Lk+2)) is a vector space of
dimension 2k + p + q (resp. 2k + p) over the field Z2 of integers modulo 2.
In Table 3, we let a (resp. b) denote the dimension of the Z2-vector space
generated by the cosets represented by δ1, . . . , δm in L/2L (resp. L/(2L +
Lk+2)).
Proposition 9.2. If L is a classical Lie torus depending on parameters as described
above, then the root-grading type, the nullity, the centroid rank and the root-space
rank vector of L are listed in Table 2, and the quotient external-grading group of L
is listed in Table 3.
Proof. We outline the proof for special symplectic Lie tori. The interested reader
will be able to supply the missing details in this case and provide the arguments
in the other three cases. (Admittedly more work is involved for special unitary Lie
tori, but the approach is the same.)
Let L = ssp2r(A, −) with the assumptions and notation as in (C) above, and let
Z = Z(A, −).
Then, by Proposition 9.1, M2r(A) is a Lie algebra over Z under the commutator
product and L is a Z-subalgebra of M2r(A).
To compute some of the invariants of L, it will be helpful to work in a larger
Z-subalgebra U of the Lie algebra M2r(A). Let G =(cid:2) 0 Jr
−Jr 0 (cid:3) ∈ M2r(k) and
C −Jr ¯AtJr(cid:21) A, B, C ∈ Mr(A), Jr ¯BtJr = B, Jr ¯C tJr = C}.
U = {X ∈ M2r(A) G−1 ¯X tG = −X}
= {(cid:20)A
(38)
B
Then, as we saw in Construction 8.3(C),
(39)
L = {X ∈ U tr(X) ∈ [A, A]}
Also, it is known that A = Z(A) ⊕ [A, A] (see [2, Lemma 5.1.3] for a proof). Hence
(40)
Thus
(41)
A− = (Z(A) ∩ A−) ⊕ ([A, A] ∩ A−).
U = (Z(A) ∩ A−)I2r ⊕ L
LIE TORI
Nullity
of L
2k + q
crk(L)
s2 − 1
rkv(L)
(d2)
27
2k + p+ q
s(s − (−1)k)
2
if p 6= 1;
s2 − 1
if p = 1
2k + p+ q
n
s(s + (−1)k)
2
if p 6= 1;
s2 − 1
if p = 1
s(s − 1)
2
if (k, p) = (0, 0);
d(d − (−1)k)
)
2
if (k, p) 6= (0, 0)
and p 6= 1;
if p = 1
if p 6= 1;
if p = 1
(m,b1)
(md2, bd2,
(2md2,d2d2, d2)
(bd2,
(d2d2, d2)
2
(d2) = (1)
d(d + (−1)k)
)
Construction
Root-grading
type of L
(A)
with
d =Qk
i=1 ζi,
s = (r + 1)d
(BC -- B)
with
d = 2k+⌊ p
s = (2r + m)d
2 ⌋,
Ar
Br
if (k, p) = (0, 0);
BCr
if (k, p) 6= (0, 0)
Cr
Dr
(C)
with
d = 2k+⌊ p
s = 2rd
2 ⌋,
(D)
with
d = 1,
s = 2r
Table 2. Invariants of classical Lie tori -- Part 1
Construction
Λ/Γ(L)
(A)
(BC -- B)
(C)
(D)
i=1 Z2
ζi
Z2k+p+a−2b
⊕ Zb
4
Lk
2
Z2k+p
2
{0}
Table 3. Invariants of classical Lie tori -- Part 2
Indeed the inclusion from right to left is clear, and the reverse inclusion follows
easily using (38), (39) and (40).
i=1 k(eii − e2r+1−i,2r+1−i). Recall from Construction 8.3(C) that
h is a split toral k-subalgebra of L with irreducible finite root system ∆ = ∆k(L, h)
Next let h =Pr
28
BRUCE ALLISON
of type Cr. In fact if we define εi ∈ h∗ for 1 ≤ i ≤ r by εi(ejj −e2r+1−j,2r+1−j) = δij
we have
(42)
with
(43)
∆ = {εi − εj 1 ≤ i 6= j ≤ r} ∪ {±(εi + εj) 1 ≤ i ≤ j ≤ r},
Lεi−εj = {a eij − ¯a e2r+1−j,2r+1−i a ∈ A}
Lεi+εj = {a ei,2r+1−j − ¯a ej,2r+1−i a ∈ A}
L−εi−εj = {a e2r+1−i,j − ¯a e2r+1−j,i a ∈ A}
for 1 ≤ i 6= j ≤ r,
for 1 ≤ i < j ≤ r,
for 1 ≤ i < j ≤ r,
L2εi = {h ei,2r+1−i h ∈ A+}
for 1 ≤ i ≤ r.
(L0 is the set of diagonal matrices in L.) Recall also that L is an fgc centreless Lie
torus of type (∆, Λ) with gradings determined by h and A, where Λ is the grading
group of A. So, as already observed in Construction 8.3(C), the root-grading type
of L is Cr.
Next we have
(A, −) = (A1, −) ⊗ · · · ⊗ (Ak+2, −),
where for convenience we have set (Ak+2, −) = (Rq, 1). Hence, by definition, the
grading group of A is
Λ = Λ1 ⊕ · · · ⊕ Λk+2,
where Λi is the grading group of Ai for 1 ≤ i ≤ k + 2. But Λ1, . . . , Λk = Z2,
Λk+1 = Zp and Λk+2 = Zq, so Λ ≃ Z2k+p+q. Hence the nullity of L is 2k + p + q.
Now by Proposition 9.1, we have a graded isomorphism ρ : Z → Ck(L), which
we now use to identify
(44)
Hence, we have
Ck(L) = Z.
Γ := Γ(L) = Γ(A, −),
where, as in Section 8, Γ(A, −) := suppΛ(Z).
As we observed in Section 8, we have
(45)
Z = Z1 ⊗ · · · ⊗ Zk+2,
where Zi = Z(Ai, −). Hence Γ = Γ1 ⊕ · · · ⊕ Γk+2, where Γi = suppΛi (Zi) for
1 ≤ i ≤ k + 2. One checks that Γi = 2Λi for 1 ≤ i ≤ k + 1, and clearly Γk+2 = Λk+2,
so
Γ = 2Λ1 ⊕ · · · ⊕ 2Λk+1 ⊕ Λk+2,
Thus, Λ/Γ(L) = Λ/Γ ≃ Z2k+p
.
2
Observe next that by Lemma 8.2(ii), A is a free Z-module with
(46)
rankZ (A) = [Λ : Γ] = 22k+p =(d2
2d2
if p 6= 1,
if p = 1,
LIE TORI
29
where recall that d = 2k+⌊ p
free Z-modules of finite rank. Moreover,
2 ⌋. Also, by Lemma 8.2(ii), Z(A) ∩ A−, A+ and A− are
(47)
(48)
(49)
rankZ (Z(A) ∩ A−) = δ1p,
2
rankZ (A+) =( d(d+(−1)k)
rankZ (A−) =( d(d−(−1)k)
d2
d2
2
if p 6= 1,
if p = 1,
if p 6= 1,
if p = 1.
We will justify these equalities at the end of the proof, but for the moment we
assume that they hold and use them to calculate the remaining invariants.
First
crk(L) = rankCk(L)(L) = rankZ(L)
= rankZ(M) − rankZ (Z(A) ∩ A−)
= rankZ(M) − δ1p
= r2 rankZ (A) + 2(cid:0) r(r − 1)
2
rankZ(A) + r rankZ (A+)(cid:1) − δ1p
= (2r2 − r) rankZ(A) + 2r rankZ(A+) − δ1p.
by (44)
by (41)
by (47)
by (38)
If we plug in the expressions (46) and (48) for rankZ (A) and rankZ (A+) into this
last expression and simplify, we obtain the values of crk(L) appearing in Table 2.
Also,
rkv(L) = ( \rankZ(Lε1−ε2), rankZ (L2ε1 )) = ( \rankZ (A), rankZ(A+))
using (43). Again, plugging in our expressions for rankZ (A) and rankZ(A+), we
obtain the values of rkv(L) appearing in Table 2.
We conclude the proof by justifying (47), (48) and (49).
Suppose first that p 6= 1. Then (A, −) is of first kind, so we have (47). Also, if
p = 0, we have
(A, −) = (Q(−1), ♮) ⊗ · · · ⊗ (Q(−1), ♮)
⊗(Rq, 1),
and (48) and (49) and be proved simultaneously by induction on k using (45). We
leave the details to the reader. Moreover, the equations (48) and (49) for the case
p = 2 can easily be deduced from the equations (48) and (49) for the case p = 0
(tensor with (Q(−1), ∗)). Again we leave the details to the reader.
Finally, if p = 1, then (A, −) is of second kind and our equalities follow from
(cid:3)
(24).
Remark 9.3. In this remark, we list the index of each classical Lie torus L (see
Remark 6.3) using the notation of Table 2. We computed these indices using
the methods outlined in the proofs of Propositions 9.1 and 9.2, together with the
detailed information about the indices of classical algebraic groups found in [34,
Table II]. We omit any details since, as we have mentioned, we do not use the index
in this article.
k factors
{z
}
30
BRUCE ALLISON
If L = slr+1(A) is as in (A) above, then L has index 1A(d)
s−1,r. Next, if L =
su2r+m(A, −, D) as in (BC -- B), then L has index
C(d)
B s−1
s
2 ,r if p 6= 1 and k is odd;
2 ,r if p 6= 1, (k, p) = (0, 0) and m is odd;
2 ,r if p 6= 1, k is even and either (k, p) 6= (0, 0) or m is even; and
s−1,r if p = 1.
tD(d)
s
2A(d)
(In the second last case t = 1 or 2, and one can write down necessary and sufficient
conditions involving the parameters (32) for t to be 1.) Further, if L = ssp2r(A, −)
as in (C), then L has index
1D(d)
C(d)
s
s
2 ,r if p 6= 1 and k is odd;
2 ,r if p 6= 1 and k is even; and
s−1,r if p = 1.
2A(d)
Finally, if L = o2r(A) as in (D), then L has index 1D(1)
s
2 ,r = 1D(1)
r,r .
10. The isomorphism problem
To provide a classification of fgc centreless Lie tori up to isomorphism, it remains
to solve the isomorphism problem for fgc centreless Lie tori as they are described
in Theorem 8.5.
In this section, we describe the results about the isomorphism problem that we
can deduce using our isomorphism invariants and their values listed in Tables 1 and
2.
Theorem 10.1.
(i) The classes of classical Lie tori and exceptional Lie tori are disjoint. That
is, there is no Lie algebra that is isomorphic to a classical Lie torus and to
an exceptional Lie torus.
(ii) The classes of classical Lie tori obtained using constructions (A), (BC -- B),
(C) and (D) are pairwise disjoint.
Proof. (i): Suppose for contradiction that L is a Lie algebra that is isomorphic to a
classical Lie torus and an exceptional Lie torus. Then, comparing Tables 1 and 2,
we see that L has root-grading type A1, A2 or C3. Hence, by Table 1, crk(L) = 78
or 133. But from Table 2, we see that crk(L) has the form s2 − 1 or s(s±1)
for some
positive integer s. This rules out crk(L) = 133, so crk(L) = 78. Hence by Table 1,
L has root grading type A2. Thus, by Table 2, crk(L) = s2 − 1 for some positive
integer s, so 78 = s2 − 1. This is a contradiction.19
2
(ii): Suppose for contradiction that there is a Lie algebra that is isomorphic to
Lie tori L and L′ coming from two different constructions from the list (A), (BC --
B), (C) and (D). We will use the notation of Section 9 for L and corresponding
primed notation for L′. Since the root-grading type is an isomorphism invariant,
we see from Table 2 that we must have one of the following (up to an exchange of
L and L′):
19(i) can also be seen by comparing indices or, in all but one case, absolute types (see for
example [7, §3.3] for this terminology).
LIE TORI
31
(a) L arises from (A) with r = 1, and L′ arises from (BC -- B) with (r′, k′, p′) =
(1, 0, 0);
(b) L arises from (A) with r = 1, and L′ arises from (C) with r′ = 1; or
(c) L arises from (BC -- B) with (r, k, p) = (2, 0, 0), and L′ arises from (C) with
r′ = 2.
Suppose first that (a) holds. Then we have r = 1 and s = 2d in construction (A);
and we have d′ = 1 and s′ = m′ + 2 in construction (BC -- B). Comparing the root-
space rank vectors in Table 2 we see that d2 = m′, whereas comparing centroid-rank
. So 4m′ − 1 = (m′+2)(m′+1)
vectors we see that 4d2 − 1 = s′(s′−1)
which forces m′ = 1 or 4. But these values of m′ were excluded in (33).
= (m′+2)(m′+1)
2
2
2
Suppose next that (b) holds. Then arguments (which we leave to the reader)
that are similar to the one in (a) handle all but one case:
L = sl2(A) and L′ = ssp2(A′, −) with p′ = 1.
In this case, we have k′
6= 0 by (37), so d′ ≥ 2. We now indicate, omitting the
details, how this leads to a contradiction using base-ring extension and a theorem
about finite dimensional central simple Lie algebras from [22]. (Alternatively, one
could compare the indices of both sides using Remark 9.3.) Let Z = Z(A) and
Proposition 9.2, we see that A has rank d2 over Z and A′ has rank 2d′2 over Z ′.
Moreover, as in the proof of Proposition 9.1, we see that that Z ⊗Z A is a division
Z ′ = Z(A′, −), with quotient fields Z andfZ ′ respectively. Now, as in the proof of
algebra of dimension d2 over its centre Z, and (fZ ′ ⊗Z ′ A, −) is a division algebra
with involution of dimension 2d′2 over its centrefZ ′ (as an algebra with involution).
L and L′ are respectively sl2( Z ⊗Z A) and ssp2(fZ ′ ⊗Z ′ A, −). (These last two
we have sl2( Z ⊗Z A) ≃ ssp2(fZ ′ ⊗Z ′ A, −). Since d′ ≥ 2, this contradicts the last
Also, again as in the proof of Proposition 9.1, we see that the central closures of
Lie algebras are defined as in Constructions 8.3(A) and (C).) So, by Remark 5.2,
Finally, case (c) is handled easily using the method in (a) and the exclusion (37).
(cid:3)
statement in Theorem X.11 of [22].
Combining Theorems 8.5 and 10.1, we see that the isomorphism problem for fgc
centreless Lie tori reduces to 5 separate problems, one for exceptional Lie tori and
one for each of the four Constructions (A), (BC -- B), (C) and (D) of classical Lie
tori.
The next theorem solves the isomorphism problem for Constructions (C) and (D).
Theorem 10.2.
(i) Let PC be the set of all vectors (r, k, p, q) in Z4 such that r ≥ 1, k ≥ 0,
p ∈ {0, 1, 2}, q ≥ 0 and if, r = 1 or 2, (k, p) /∈ {(0, 0), (0, 1), (1, 0)}.
Then the map that sends (r, k, p, q) to the isomorphism class represented
by ssp2r(A, −), where (A, −) is the tensor product of basic associative tori
with involution constructed from (k, p, q) as in (34) -- (36), is a bijection
from PC onto the set of isomorphism classes of special symplectic Lie tori.
Moreover, special symplectic tori are classified by their root-grading type,
nullity and root-space rank vector. That is, two special symplectic Lie tori
are isomorphic if and only if they have same root-grading type, the same
nullity and the same root-space rank vector.
32
BRUCE ALLISON
(ii) Let PD = {(r, n) ∈ Z2 r ≥ 4, n ≥ 0}. Then the map that sends (r, n) to
the isomorphism class represented by o2r(Rn) is a bijection from PD onto
the set of isomorphism classes of orthogonal Lie tori. Moreover, orthogonal
Lie tori are classified by their root-grading type and nullity.
Proof. (i): Before beginning, let N = {k ∈ Z k ≥ 1}, S = N × {0, 1, 2}, and define
f : S → N by
f (k, p) =(2k+ p
22k
2 −1(2k+ p
2 + (−1)k)
if p 6= 1,
if p = 1.
It is not difficult to show that if (k, p) ∈ S then
(50)
and that
(51)
f (k, p) = 1 ⇐⇒ (k, p) ∈ {(0, 0), (1, 0), (0, 1)};
f S\{(0,0),(1,0),(0,1)} is one-to-one.
We leave these facts for the reader to check.
Now to begin the proof of (i), observe that the map described in the first state-
ment of (i) is surjective by the discussion of parameterization in Section 9.
Next suppose that (r, k, p, q) and (r′, k′, p′, q′) are in PC . We let L = ssp2r(A, −),
where (A, −) is the tensor product of basic associative tori with involution con-
structed from (k, p, q) as in (34) -- (36), and we let L′ = ssp2r′(A′, −), where (A′, −)
is obtained in the same way from (k′, p′, q′). Observe that by Table 2, we have
rkv(L) = (\22k+p, f (k, p)),
and we have a similar expression for rkv(L′).
We will show that the following statements are equivalent:
(a) L ≃ L′,
(b) L and L′ have the same root-grading type, the same nullity and the same
root-space rank vector,
(c) (r, k, p, q) = (r′, k′, p′, q′).
Note that this will compete the proof of both of the statements in (i).
Now "(a) ⇒ (b)" holds by Theorem 6.2, and "(c) ⇒ (a)" is trivial. Thus it
suffices to show that "(b) ⇒ (c)". So, suppose that (b) holds. Then r = r′,
(52)
(53)
(54)
2k + p + 1 = 2k′ + p′ + q′,
f (k, p) = f (k′, p′), and
22k+p = 22k′+p′
if r ≥ 2.
Suppose first that (k, p) ∈ {(0, 0), (1, 0), (0, 1)}. Then, f (k, p) = 1 by (50), so
(k′, p′) ∈ {(0, 0), (1, 0), (0, 1)} by (50) and (53). But r ≥ 3 by definition of PC , so
by (54) we have 2k + p = 2k′ + p′. Hence (k, p) = (k′, p′). Finally, by (52), we have
q = q′.
Lastly, suppose that (k, p) /∈ {(0, 0), (1, 0), (0, 1)}, so by the argument just given
(k′, p′) /∈ {(0, 0), (1, 0), (0, 1)}. Thus, by (53) and (51), we have (k, p) = (k′, p′), and
therefore also q = q′ as above.
(ii): Since o2r(Rn) ≃ o2r(k) ⊗ Rn, this follows from well-known facts about
multiloop algebras (see for example [7, Cor. 8.19]). From our point of view here, it
also follows immediately using the argument in (i) and Table 2.
(cid:3)
LIE TORI
33
Corollary 10.3. Fix r ≥ 3. Then the fgc centreless Lie tori of type Cr are classified
by their nullity and root-space rank vector.
Proof. Suppose that L and L′ are fgc centreless Lie tori of type Cr with the same
If L and L′ are classical, we have
nullity and the same root-space rank vector.
L ≃ L′ by Theorem 10.2(i). On the other hand if L and L′ are exceptional, then
r = 3 and L ≃ L′ by Table 1. Finally, if L is exceptional and L′ is classical, then
r = 3 and rkv(L′) = rkv(L) = (8, 1) by Table 1, which contradicts Table 2.
(cid:3)
The remaining isomorphism problems. The 3 remaining isomorphism prob-
lems are now listed, together with some comments. We will say more about each
of these problems in the next section.
(1) The isomorphism problem for exceptional Lie tori. It follows looking at root-
space rank vectors in Table 1 that the only possible isomorphisms between Lie
tori of a given root-grading type and nullity are between the tori numbered 20,
22 and 23, or between the tori numbered 21 and 24. So it remains to decide if
any such isomorphisms exist. Note however that the tori numbered 20, 22 and 23
have distinct quotient external-grading groups, as do the tori numbered 21 and 24.
Therefore, the exceptional Lie tori listed in Table 1 are pairwise not isotopic.
(2) The isomorphism problem for special linear Lie tori. The Lie tori in con-
struction (A) are not classified by the four isomorphism invariants from Theorem
6.2, even in nullity 2.20 (The index adds no extra information; that is, if the four
invariants match for two special linear Lie tori, one can check that the indices match
as well.) However, it is clear that two special linear Lie tori that have the same
root-grading type, the same nullity and the same root-grading rank vector, have a
common r, a common nullity for their coordinate associative tori A and A′, and a
common rank for A and A′ as modules over their centres. So these quantities can
be fixed when considering the problem.
(3) The isomorphism problem for special unitary Lie tori. The Lie tori in con-
struction (BC -- B) are not classified by the four isomorphism invariants from The-
orem 6.2, even in nullity 3. (Again, including the index does not provide enough
information for classification.) However, one can check (using the argument in the
proof of Theorem 10.2(i))) that with one exception, two special unitary Lie tori
that have the same root-grading type, nullity, centroid rank and root-grading rank
vector have a common r, a common coordinate associative torus with involution
(A, −) up to isomorphism, and a common value for m = D. So these entities can
be fixed when considering the problem.
In summary, the classification of fgc centreless Lie tori up to isomorphism is
reduced to the separate isomorphism problems for (1) five particular exceptional
Lie tori, (2) special linear Lie tori, and (3) special unitary Lie tori.
11. Conjugacy and its implications
In this section we consider the three isomorphism problems just discussed under
a conjugacy assumption for certain maximal split toral k-subalgebras.
A conjugacy assumption. We say that an fgc centreless Lie algebra L satisfies
Assumption (C) if the following holds:
20For example, it follows from [7, Thm. 11.3.2] that if ζ is a 5th root of unity, then slr+1(Q(ζ))
and slr+1(Q(ζ 2)) are not isomorphic. However, by Table 2, they each have nullity 2, root-grading
type Ar, centroid rank 25(r + 1)2 − 1 and root-space rank vector (25).
34
BRUCE ALLISON
(C) If L has the graded structure L = L(α,λ)∈Q(∆)×Λ Lλ
α of a Lie torus of
0 and if (the same) L has the graded structure
α′ of a Lie torus of type (∆′, Λ′) with h′ = L′0
0,
then there exists ϕ ∈ Aut(L) such that ϕ(h) = h′.
type (∆, Λ) with h = L0
L = L(α′,λ′)∈Q(∆′)×Λ′ L′λ′
Less precisely, this assumption says that two maximal split toral k-subalgebras
of L that arise from Lie torus structures on L are conjugate under the action of
Aut(L).
Remark 11.1. Our motivation for making Assumption (C) in the theorems below
is work in progress by V. Chernousov, P. Gille and A. Pianzola [16]. This work will
show that Assumption (C) holds for any fgc centreless Lie torus and therefore the
assumption is superfluous. It is already known that this is the case for untwisted
Lie tori [30].
An immediate consequence of Assumption (C) together with Theorem 7.2 is the
following:
Theorem 11.2. Suppose that L and L′ are fgc centreless Lie tori satisfying As-
sumption (C). Then L is isomorphic to L′ if and only if L is isotopic to L′ .
Proof. Suppose that ϕ : L → L′ is an isomorphism. By Assumption (C) we can
assume that ϕ(h) = h′, where h = L0
0. Then ϕ is an isotopy by
Theorem 7.2.
(cid:3)
0 and h′ = L′0
Putting this result together with Proposition 7.1, we have:
Corollary 11.3. The quotient external-grading group is an isomorphism invariant
for an fgc centreless Lie torus satisfying Assumption (C).
In the remaining sections, we discuss the implications of Assumption (C) and
our results for the three isomorphism problems listed at the end of Section 10.
Isomorphism of exceptional Lie tori. In the discussion of Problem 1 at the
end of Section 10, we used the quotient external-grading group to show that the
exceptional Lie tori in Table 1 are pairwise not isotopic. Therefore, it follows from
Theorem 11.2 that if exceptional Lie tori satisfy Assumption (C), then they are
listed up to isomorphism without redundancy in Table 1.
Isomorphism of special linear Lie tori. For special linear Lie tori, the quotient
external-grading group Λ/Γ adds no further information. That is, if the four invari-
ants in Theorem 6.2 match for two special linear Lie tori, one can check that the
quotient external-grading groups also match. However, under assumption (C) we
can prove the following theorem, which was proved in nullity 2 in [7, Thm. 11.3.2]
without Assumption (C).
Theorem 11.4. Suppose that A and A′ are fgc associative tori of nullity n, r ≥ 1,
and the Lie tori slr+1(A) and slr+1(A′) satisfy Assumption (C). Then slr+1(A)
and slr+1(A′) are isomorphic if and only A and A′ are isomorphic.
Proof. The implication from right to left is clear and so we consider only the con-
verse. This can be seen to follow using Assumption (C) from Theorems 8.6(ii),
9.11 and 10.6 of [8] in the cases r = 1, r = 2 and r ≥ 3 respectively. However for
the readers convenience we give a uniform argument that follows the approach in
Section 9 of [8].
LIE TORI
35
We begin arguing that (as noted in [8, Remark 9.12]) A ≃ Aop, where Aop
denotes the opposite algebra of A with product (x, y) 7→ yx.
Indeed, using the
tensor product decomposition (22) of A, it is sufficient to consider the case when
A = Q(ζ), where ζ is a root of unity. But in this case we have A ≃ Aop under the
homomorphism exchanging x1 and x2.
Let L = slr+1(A), L′ = slr+1(A′) and L′′ = slr+1(A′op). Then h = L0
0, h′ = L′0
0
and h′′ = L′′0
i=1 k(eii−e2r+1−i,2r+1−i)
in L, L′ and L′′ respectively. So we can identify h, h′ and h′′ in the evident fashion.
i=1 k(eii − e2r+1−i,2r+1−i) is a subalgebra of L, L′ and L′′; and
0 are identified in Construction 8.3(A) withPr
In this way, h =Pr
we have
∆ := ∆k(L, h) = ∆k(L′, h) = ∆k(L′′, h).
Assume now that slr+1(A) ≃ slr+1(A′). Then, by Assumption (C) applied to
L′, we have an isomorphism ϕ : L → L′ such that ϕ(h) = h. Thus, as in the proof
of Theorem 7.2, ϕ induces a linear automorphism ϕ of h∗ such that ϕ(Lα) = L′
for α ∈ h∗. So ϕ is an automorphism of the root system ∆.
ϕ(α)
Next define ψ : L′ → L′′ by ψ(x) = −xt for x ∈ L′. Then ψ is an algebra
isomorphism with ψ(h) = h, and we have ψ = −1. Thus, replacing A′ by A′op and
ϕ by ψϕ if necessary, we can assume that ϕ is in the Weyl group of ∆.
So, replacing ϕ by πϕ, where π is conjugation by an appropriate permutation
matrix over k, we can assume that ϕ = 1. Therefore, for 1 ≤ i 6= j ≤ r + 1, we have
a linear bijection ϕij : A → A′ with
for a ∈ A.
Now, if a, b, c ∈ A, we have
ϕ(aeij) = ϕij (a)eij
[[ae12, be21], ce12] = [abe11 − bae22, ce12] = (abc + cba)e12.
Applying ϕ to this equation, yields
(55)
ϕ12(a)ϕ21(b)ϕ12(c) + ϕ12(c)ϕ21(b)ϕ12(a) = ϕ12(abc + cba).
If we put a = c = 1 in this equation we get
ϕ12(1)ϕ21(b)ϕ12(1) = ϕ12(b)
(56)
for b ∈ A. So ϕ12 = ℓϕ12(1)rϕ12(1)ϕ21, where ℓϕ12(1) and rϕ12(1) in Endk(A′) are left
and right multiplication respectively by ϕ12(1). Thus, ℓϕ12(1)rϕ12(1) = rϕ12(1)ℓϕ12(1)
is invertible, and so ϕ12(1) is a unit in A′.
Replacing ϕ by µϕ, where µ is conjugation by diag(ϕ12(1)−1, 1, . . . , 1), we can
assume that ϕ12(1) = 1. Hence, by (56), ϕ21 = ϕ12. So putting b = 1 in (55), we
have ϕ12(a)ϕ12(c) + ϕ12(c)ϕ12(a) = ϕ12(ac + ca) for a, c ∈ A. That is ϕ12 : A → A′
is a Jordan isomorphism. Hence, since A′ is a prime ring, a theorem of Herstein
[21, Thm. 3.1] tells us that ϕ12 is either an isomorphism or an anti-isomorphism of
A onto A′. Since A′ ≃ A′op it follows that A ≃ A′.
(cid:3)
If all special linear Lie tori satisfy Assumption (C), Theorem 11.4 reduces their
classification to Neeb's classification of fgc associative tori mentioned in Section 8.
Isomorphism of special unitary Lie tori. If all special unitary Lie tori satisfy
Assumption (C), Theorem 11.2 tells us that their classification up to isomorphism
is reduced to determining when two such Lie tori are isotopic. We are optimistic
that the latter can be accomplished along the lines of [2, §7], perhaps using a notion
36
BRUCE ALLISON
of isotope for graded hermitian forms (following the philosopy of [8]). However, at
this point the isotopy problem for special unitary Lie tori is open.
Summary. If all fgc centreless Lie tori satisfy Assumption (C), our work in this
article has reduced their classification up to isomorphism to solving the isotopy
problem for special unitary Lie tori.
References
[1] Allison, B., S. Azam, S. Berman, Y. Gao and A. Pianzola, Extended affine Lie algebras and
their root systems, Mem. Amer. Math. Soc. 126 (603), 1997.
[2] Allison, B., and G. Benkart, Unitary Lie algebras and Lie tori of type BCr , r ≥ 3, in
Proceedings of the Conference on Quantum Affine Algebras, Extended Affine Lie Algebras
and Applications, Banff, Canada, 2008, Contemporary Mathematics 506 (2010), 1 -- 47, Amer.
Math Soc., Providence, RI.
[3] Allison, B., S. Berman, Y. Gao and A. Pianzola, A characterization of affine Kac-Moody Lie
algebras, Comm. Math. Phys. 185 (1997), 671 -- 688.
[4] Allison, B., S. Berman, J. Faulkner, A. Pianzola, Realization of graded-simple algebras as
loop algebras, Forum Mathematicum, 20 (2008), 395 -- 432.
[5] Allison, B., S. Berman, J. Faulkner, A. Pianzola, Multiloop realization of extended affine Lie
algebras and Lie tori, Trans. Amer. Math. Soc. 361 (2009), 4807 -- 4842.
[6] Allison, B., S. Berman and A. Pianzola, Iterated loop algebras, Pacific J. Math. 227 (2006),
1 -- 42.
[7] Alliso,n B., S. Berman and A. Pianzola Multiloop algebras, iterated loop algebras and extended
affine Lie algebras of nullity 2, arXiv:1002.2674v1 [math.RA].
[8] Allison, B., and J. Faulkner, Isotopy for extended affine Lie algebras and Lie tori, in De-
velopments and Trends in Infinite-Dimensional Lie Theory, 3 -- 43, Progress in Mathematics,
Vol. 288, Birkhauser, Boston, MA, 2011.
[9] Allison, B., J. Faulkner and Y. Yoshii, Structurable tori, Comm. Algebra 36 (2008), 2265 --
2332.
[10] Allison, B., and Y. Gao, The root system and the core of an extended affine Lie algebra,
Selecta Math. (N.S.) 7 (2001), no. 2, 149 -- 212.
[11] Allison, B., and Y. Yoshii, Structurable tori and extended affine Lie algebras of type BC1,
J. Pure and Appl. Algebra 184 (2003), 105-138.
[12] Benkart, G., and E. Neher, The centroid of extended affine and root graded Lie algebras,
J. Pure Appl. Algebra 205 (2006), 117 -- 145.
[13] Berman, S., Y. Gao and Y. Krylyuk, Quantum tori and the structure of elliptic quasi-simple
Lie algebras, J. Funct. Anal. 135 (1996), 339 -- 389.
[14] Berman, S., Y. Gao, Y. Krylyuk and E. Neher, The alternative torus and the structure of
elliptic quasi-simple Lie algebras of type A2, Trans. Amer. Math. Soc. 347 (1995), no. 11,
4315 -- 4363.
[15] Bourbaki, N., "Lie groups and Lie algebras", Chapters 4 -- 6, Translated from the French,
Elements of Mathematics, Springer-Verlag, Berlin, 2002.
[16] Chernousov, V., P. Gille and A. Pianzola, Conjugacy theorems for twisted forms of simple
Lie algebras extended over Laurent polynomial rings, work in progress.
[17] Erickson, T.S., W.S. Martindale, 3rd and J.M. Osborn, Prime nonassociative algebras, Pacific
J. Math. 60 (1975), 49 -- 63.
[18] Faulkner, J., Lie tori of type BC2 and structurable quasitori, Comm. Algebra 36 (2008),
2593 -- 2618.
[19] Gille, P., and A. Pianzola, Galois cohomology and forms of algebras over Laurent polynomial
rings, Math. Ann. 338 (2007), 497 -- 543.
[20] Gille, P., and A. Pianzola, Galois cohomology and forms of algebras over Laurent polynomial
rings II, in preparation.
[21] Herstein, I.N., "Topics in ring theory", The University of Chicago Press, Chicago, 1969
[22] Jacobson, N., "Lie algebras", Dover, New York, 1979.
[23] Lam, T.Y., "A first course in noncommutative rings", Second edition. Graduate Texts in
Mathematics, 131. Springer-Verlag, New York, 2001.
LIE TORI
37
[24] Naoi, K., Multiloop Lie algebras and the construction of extended affine Lie algebras, J.
Algebra 323 (2010), 2103 -- 2129.
[25] Neeb, K.-H., On the classification of rational quantum tori and the structure of their auto-
morphism groups, Canad. Math. Bull. 51 (2008), 261 -- 282.
[26] Neher, E., Lie tori, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 84 -- 89.
[27] Neher, E., Extended affine Lie algebras, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004),
90 -- 96.
[28] Neher, E., Extended affine Lie algebras and other generalizations of affine Lie algebras - a
survey, in Developments and Trends in Infinite-Dimensional Lie Theory, 53-125, Progress in
Mathematics, Vol. 288, Birkhauser, Boston, MA, 2011.
[29] Neher, E., Lectures on extended affine Lie algebras, arXiv:1003.2352v1 [math.RA].
[30] Pianzola, A., Locally trivial principal homogeneous spaces and conjugacy theorems for Lie
algebras, J. Algebra 275 (2004), 600-614.
[31] Pianzola, A., Vanishing of H 1 for Dedekind rings and applications to loop algebras, C.R.
Acad. Sci. Paris, Ser I 340 (2005), 633-638.
[32] Seligman, G.B., "Rational methods in Lie algebras", Lect. Notes in Pure and Applied Math.
27, Marcel Dekker, New York, 1976.
[33] Tan, S., TKK algebras and vertex operator representations, J. Algebra 211 (1999), 298 -- 342.
[34] Tits, J., Classification of algebraic semisimple groups, Algebraic Groups and Discontinuous
Subgroups (Proc. Sympos. Pure Math., Boulder, Colo., 1965), pp. 33 -- 62, Amer. Math. Soc.,
Providence, R.I., 1966.
[35] Yoshii, Y., Coordinate algebras of extended affine Lie algebras of type A1, J. Algebra 234
(2000), 128 -- 168.
[36] Yoshii, Y., Classification of quantum tori with involution, Canad. Math. Bull. 45 (2002), no.
4, 711 -- 731.
[37] Yoshii, Y., Root systems extended by an abelian group and their Lie algebras, J. Lie Theory
14 (2004), pp. 371 -- 374.
[38] Yoshii, Y., Lie tori -- A simple characterization of extended affine Lie algebras, Publ. Res.
Inst. Math. Sci. 42 (2006), 739 -- 762.
(Bruce Allison) Department of Mathematical and Statistical Sciences, University of
Alberta, Edmonton, Alberta T6G 2G1, Canada
E-mail address: [email protected]
|
1306.5967 | 1 | 1306 | 2013-06-25T13:53:41 | Geometry of totally real Galois fields of degree 4 | [
"math.RA",
"math.AG"
] | We will consider a totally real Galois field $K$ of degree 4 as the linear coordinate space $\mathbb{Q}^4\subset\mathbb{R}^4$. An element $k\in K$ is called strictly positive, if all its conjugates are positive. The set of strictly positive elements is a convex cone in $K$. The convex hull of strictly positive integral elements is a convex subset of this cone and its boundary $\Gamma$ is an infinite union of 3-dimensional polyhedrons. The group $U$ of strictly positive units acts on $\Gamma$: the action of a strictly positive unit permutes polyhedrons. Fundamental domains of this action are the object of study in this work. We mainly present some interesting examples. | math.RA | math |
GEOMETRY OF TOTALLY REAL GALOIS FIELDS
OF DEGREE 4
YURY KOCHETKOV
Abstract. We will consider a totally real Galois field K of degree 4 as the
linear coordinate space Q4
⊂ R4. An element k ∈ K is called strictly positive,
if all its conjugates are positive. The set of strictly positive elements is a convex
cone in K. The convex hull of strictly positive integral elements is a convex
subset of this cone and its boundary Γ is an infinite union of 3-dimensional
polyhedrons. The group U of strictly positive units acts on Γ: the action of
a strictly positive unit permutes polyhedrons. Fundamental domains of this
action are the object of study in this work. We mainly present some interesting
examples.
1. Introduction
Let K be a cubic totally real Galois field, defined by polynomial p = x3 +ax+b ∈
Q[x]. Elements of K will be written in the form kx2 + lx + m, k, l, m ∈ Q. This
notation allows us to identify K with 3-dimensional space Q3 ⊂ R3. . An element
if x1, x2, x3
k ∈ K is called strictly positive, if all its conjugates are positive, i.e.
are (real) roots of p then an element kx2 + lx + m is strictly positive, if
kx2
kx2
kx2
1 + lx1 + m > 0
2 + lx2 + m > 0
3 + lx3 + m > 0
These three conditions define a convex cone C in R3. Let O ⊂ C be the set of
strictly positive integral elements of K, O -- its convex hull (an infinite polyhedral
subset of C) and Γ be the boundary of this convex hull -- an infinite polygonal
2-dimensional complex in R3. Let U be the group of strictly positive units (a free
Abelian group of rank 2). The action of a positive unit on Γ induces a permutation
of its faces. The fundamental domain of the action of U on Γ is a finite union
of polygons with pairwise identified sides -- a torus. In [1] the structure of this
polygonal complex was studied.
In this work we will study a 4-dimensional analogue of this problem.
2. Totally real Galois fields of degree 4
Let K be a totally real Galois field of degree 4 defined by a polynomial p ∈ Q[x].
Elements of the field will be written in the form kx3 + lx2 + mx + n, k, l, m, n ∈ Q,
and the field will be considered as 4-dimensional space Q4 ⊂ R4 with coordinates
k, l, m, n. By O will be denoted the set of integral strictly positive elements from
1
2
YURY KOCHETKOV
K: an element kx3 + lx2 + mx + n is called strictly positive if
kx3
kx3
kx3
kx3
1 + lx2
2 + lx2
3 + lx2
4 + lx2
1 + mx1 + n > 0
2 + mx2 + n > 0
3 + mx3 + n > 0
4 + mx4 + n > 0
where x1, x2, x3, x4 are (real) roots of p. These four conditions define a convex cone
C ⊂ R4. The convex hull O of the set O is an infinite polyhedral 4-dimensional
subset of C. Its boundary Γ is an infinite 3-dimensional polyhedral complex. Let
U ⊂ O be the group of strictly positive units -- a free Abelian group of rank 3. The
action of a positive unit on Γ induces a permutation of its 3-dimensional polyhedra.
We will study fundamental domain of action of U on Γ. This domain -- a finite
union of 3-dimensional polyhedrons with pairwise identified faces is homeomorphic
to 3-dimensional torus [2].
Galois group G of our field K is either Z4, or Z2 ⊕ Z2. In both cases G contains
an invariant subgroup of order 2, so K contains a subfield L of degree 2. L is a
decomposition field of a second degree polynomial with rational coefficients and
positive discriminant d. Hence, K is a second degree algebraic extension of L,
defined by polynomial r ∈ Q(√d)[x].
Let r = x2 + 2sx + t, s, t ∈ Q(√d), then roots x1, x2 of r are of the form
x1,2 = −s ±ps2 − t ∈ K ⇒ ps2 − t ∈ K.
If s2 − t = m + n√d, m, n ∈ Q, then pm + n√d ∈ K,
±pm ± n√d belong to K. Thus, the field K is the decomposition field of the
q = (x −qm + n√d)(x +qm + n√d)(x −qm − n√d)(x +qm − n√d) =
= x4 − 2mx2 + m2 − n2d = x4 − 2ax2 + b.
if a > 0, b > 0 and d = a2 − b > 0; 2) a root y1 = pa + √d of q
√d,
1+ly1. This gives us two possibilities:
2 + ly2 = −y1, then G = Z4, a2b − b2 = bd = c2, l = (2a2 − b)/c =
Such biquadratic polynomial defines a totally real Galois field if: 1) all its roots
are real, i.e.
polynomially generates all other roots of q, in particular, the root y2 = pa −
which can represented only in the form y2 = ky3
biquadratic polynomial
i.e.
all conjugates
2 + ly2 = y1, then G = Z2 ⊕ Z2, b = c2, l = 2a/c and k = −1/c.
Remark 1. If b is a full square and a2b − b2 is also a full square, then polynomial
x4 − 2ax2 + b is reducible.
In what follows we'll assume that 2a and b are positive integers. Let us describe
at first the structure of the fundamental domain for fields with smallest a and b:
for the field K1 = Q[x]/(x4 − 4x2 + 1) and for the field K2 = Q[x]/(x4 − 4x2 + 2)
(polynomial x4 − 3x2 + 1 is reducible).
3. The field K1 = Q[x]/(x4 − 4x2 + 1)
Galois group here is Z2 ⊕ Z2. Transformations x 7→ x3 − 4x and x 7→ −x are
generators of Galois group. Elements kx3 + lx2 + mx + n, k, l, m, n ∈ Z are integral.
• either ky3
• or ky3
(a2 + d)/c and k = −a/c;
GEOMETRY OF TOTALLY REAL GALOIS FIELDS
OF DEGREE 4
3
Hyperplane k + l + n − 1 = 0 is a support hyperplane for the set O. It contains
9 integral strictly positive elements:
A1 = (−2, 0, 6, 3), A2 = (−2, 3, 2, 0), A3 = (−1, 0, 3, 2),
A4 = (−1, 0, 4, 2), A5 = (−1, 1, 2, 1), A6 = (−1, 2, 0, 0),
A9 = (0, 1, 0, 0).
A7 = (−1, 2, 1, 0), A8 = (0, 0, 0, 1),
Elements A1, A2, A4, A6, A8 and A9 are units. Element A3 is a midpoint of segment
[A1, A8] and has norm 4; element A7 is a midpoint of segment [A2, A9] and has norm
4; element A5 is a midpoint of segment [A4, A6] and has norm 9. The convex hull
of these points in our hyperplane is an octahedron:
A1
q
❝
❝
❝
A4
q
❝
❝
❝
❝❝
✪✪
q
A8
A2
❅
q
❍❍❍❍❍❍
❅
❅
✟✟✟✟✟✟✟✟✟
✪
✪
✪
✪
✪
❅
q
❇
A6
❅
❅
✪
✪
❇
❇
❅❅
❇❇
A9
✪
q
Faces of this octahedron are not identified by the action of group U . Hence, the
construction of fundamental domain is not finished.
Indeed, hyperplane 4k + 5l/2 + m + n − 1 = 0 is also a support hyperplane for
the set O. It contains 5 integral strictly positive elements:
A1, A3, A6, A8 and also an element A10 = (0, 0,−1, 2),
which is a unit. The convex hull of these points is a tetrahedron, which is glued
to our octahedron (by the triangular face A1A6A8). In thus obtained polyhedron
some faces (not all!) are identified:
(1) multiplication by the unit A9 identifies the face A1A8A10 with the face
(2) multiplication by the unit A4 identifies the face A6A8A10 with the face
(3) multiplication by the unit A10 identifies the face A4A8A9 with the face
A2A6A9: A1 → A2, A8 → A9, A10 → A6;
A1A2A4: A6 → A2, A8 → A4, A10 → A1;
A1A6A10: A4 → A1, A8 → A10, A9 → A6.
Faces A1A4A8, A1A2A6, A6A8A9 A2A4A9 are "free". Hence, the construction of
fundamental domain is not finished.
Indeed, hyperplane k + l/2 + n − 1 = 0 is also a support hyperplane for the set
O. It contains 5 integral strictly positive elements:
A1, A3, A4, A8 and also an element A11 = (−4,−2, 15, 8),
which is a unit. The convex hull of these points is a tetrahedron, which is glued to
our octahedron (by the triangular face A1A4A8). In thus obtained polyhedron all
its 2-dimensional faces are pairwise identified:
4
YURY KOCHETKOV
(1) multiplication by the unit A9 identifies the face A1A8A11 with the face
(2) multiplication by the unit A6 identifies the face A4A8A11 with the face
(3) multiplication by the unit A11 identifies the face A6A8A9 with the face
A2A4A9: A1 → A2, A8 → A9, A11 → A4;
A1A2A6: A4 → A2, A8 → A6, A11 → A1;
A1A4A11: A6 → A1, A8 → A11, A9 → A4.
The construction of the fundamental domain is finished.
4. The field K2 = Q[x]/(x4 − 4x2 + 2)
Galois group here is Z4. Transformation x 7→ x3 − 3x is a generator of Galois
Hyperplanes
group. Elements kx3 + lx2 + mx + n, k, l, m, n ∈ Z are integral.
Π1 : k + 2l + m + 2n − 2 = 0,
Π3 : k + 2l + 2n − 2 = 0,
Π5 : k − 2l + m − 2n + 2 = 0,
Π7 : k − 2l − 2n + 2 = 0,
Π2 : 3k + 4l + 2m + 4n − 4 = 0,
Π4 : k + 4l − m + 4n − 4 = 0,
Π6 : 3k − 4l + 2m − 4n + 4 = 0,
Π8 : k − 4l − m − 4n + 4 = 0
are support hyperplanes for the set O. Each hyperplane contains 5 integral ele-
ments. Three elements A = (0, 0, 0, 1), B = (0,−2, 0,−1) and (A + B)/2 belong to
each hyperplane. Thus, the convex hull of the set O ∩ Πi is an tetrahedron.
Besides A, B and (A + B)/2, hyperplane Π1 contains elements C = (1, 0,−3, 2
and D = (0, 1,−2, 1), hyperplane Π2 -- D and E = (−2, 4, 1,−2), hyperplane Π3
-- E and F = (−2, 3, 2,−1), hyperplane Π4 -- F and G = (−1, 0, 3, 2), hyperplane
Π5 -- G and H = (0, 1, 2, 1), hyperplane Π6 -- H and K = (2, 4,−1,−2), hyper-
plane Π7 -- K and L = (2, 3,−2,−1), hyperplane Π8 -- L and C. The union of
these tetrahedrons is something like two octagonal pyramids ACDEF GHKL and
BCDEF GHKL with common base CDEF GHKL. Here elements A, B, D, F , H
and L are units and elements C, E, G and K have norm 2.
In thus constructed polyhedron some faces (not all!) are identified:
• the multiplication by unit H identifies the face ACD with the face BHK:
• the multiplication by unit L identifies the face AF G with the face BKL:
• the multiplication by unit D identifies the face AGH with the face BDE:
• the multiplication by unit F identifies the face ALC with the face BEF :
A → H, C → K, D → B;
A → L, F → B, G → K;
A → D, G → E, H → B;
A → F , C → E, L → B.
Faces ADE, AEF , AHK, AKL, BCD, BF G, BGH and BLC are "free". The
construction of the fundamental domain is not finished.
3k + 2l + m + n − 1 = 0 also is a support hyperplane for the set O. It contains 7
integral strictly positive elements: A, D, E, F , M = (0, 0,−1, 2), N = (−2, 4, 0,−1)
and (−1, 2, 0, 0) = (A + N )/2 = (D + F )/2 = (E + M )/2. The convex hull of these
points is an octahedron, where (A, N ), (D, F ) and (E, M ) -- pairs of opposite
vertices. Here element N is a unit and element M has norm 2.
In thus obtained polyhedron -- two octagonal pyramids with a common base
and octahedron glued to them by the faces ADE and AF E, the following faces are
identified:
GEOMETRY OF TOTALLY REAL GALOIS FIELDS
OF DEGREE 4
5
• the multiplication by unit N identifies the face AHK with the face EF N :
• the multiplication by unit N identifies the face AKL with the face N DE:
• the multiplication by unit H identifies the face ADM with the face BGH:
• the multiplication by unit L identifies the face AF M with the face BCL:
• the multiplication by unit B−1F identifies the face BCD with the face
• the multiplication by unit B−1D identifies the face BF G with the face
A → N , H → F , K → E;
A → N , K → E, L → D;
A → H, D → B, M → G;
A → L, F → B, M → C;
F M N : B → F , C → M , D → N ;
DM N : B → D, F → N , G → M .
The construction of the fundamental domain is finished.
5. The simplest fundamental domains in the case of cyclic Galois
group
The combinatorics of a fundamental domain can be quite complex. The natural
question here is: how simple it can be? It turns out that there exists an one
parameter family of fields with a simple domains and also one field apart. At first
we will study the special field.
5.1. The field K = Q[x]/(x4 − 15x2 + 45). The transformation x 7→ x3/3 − 3x is
a generator of Galois group. Elements of the form
(k, l, m, n) + (
i
6
, 0, 0,
i
2
) + (0,
j
6
,
j
2
,
j
2
),
k, l, m, n ∈ Z,
i, j = 0, . . . , 5,
are integral.
Hyperplane Π : 3k + 6l + m + n = 1 is a support hyperplane for the set O. It
contains 8 integral strictly positive elements -- all of them are units:
6
13
7
2
1
3
9
1
2
,−2,
1
6
2(cid:17), C = (cid:16) 1
,−
,−
2(cid:17), C1 = (cid:16) 1
1
2
1
3
6
,
1
3
,−
, 0,−1(cid:17), B1 = (cid:16)0,
A = (0, 0, 0, 1), B = (cid:16) 1
A1 = (cid:16)0,
,−1,−
The convex hull of the set O∩ Π is a decahedron with 2 quadrangular faces ABCD
and A1B1C1D1 -- bases, and 8 triangular faces. Bases are parallelograms, that are
parallel to each other. In the picture below is presented the pattern of triangular
faces of this "prism":
2 (cid:17), D = (cid:16) 1
,−
, 1(cid:17), D1 = (cid:16) 1
,−
1
6
, 3(cid:17),
2(cid:17).
,−
,−
1
6
3
2
3
2
1
3
,
6
6
1
,
,
r
A
❏
❏
r
A1
❏
❏
❏
B
r
✡
❏❏✡
B1
r
✡
✡
✡
C
✡✡❏
r
❏
r
C1
D
r
✡
❏❏✡
D1
r
❏
❏
❏
✡
✡
✡
A
✡✡
r
r
A1
Faces of ABCDA1B1C1D1 are identified in the following way:
6
YURY KOCHETKOV
• the multiplication by unit A1 identifies the base ABCD with the face
• the multiplication by unit D identifies the face AA1B1 with the face CDD1:
• the multiplication by unit D1 identifies the face ABB1 with the face CC1D1:
• the multiplication by unit B identifies the face AA1D1 with the face BCB1:
• the multiplication by unit B1 identifies the face ADD1 with the face CB1C1:
A1B1C1D1: A → A1, B → B1, C → C1, D → D1;
A → D, A1 → D1, B1 → C;
A → D1, B → C, B1 → C1;
A → B, A1 → B1, D1 → C;
A → B1, D → C, D1 → C1.
The construction of the fundamental domain is finished.
5.2. Fields Kn. Let pn = x4 − (n2 + 4)x2 + n2 + 4, where n is an odd positive
integer, and let Kn be the field, defined by pn.
Lemma 1. Polynomials pn are irreducible.
Proof. If pn is reducible, then either it has a positive integral root, or it has an
irreducible factor of degree 2.
Let a be a positive integral root of pn: a4 − (n2 + 4)a2 + n2 + 4 = 0. Each prime
factor of a is a factor of the number n2 + 4 and vice versa. Let q be such prime
factor, α > 0 be an exponent of a and β > 0 be an exponent of n2 + 4. In the set
of 3 numbers {4α, 2α + β, β} cannot be exactly one minimal, thus, 4α = β. As this
is true for each prime factor of a, then n2 + 4 is a full square -- a contradiction.
Let pn be a product of two irreducible polynomials of degree 2 with integral
coefficients, then these factors are x2 − ax + b and x2 + ax + b:
(x2 − ax + b)(x2 + ax + b) = x4 − (n2 + 4)x2 + n2 + 4.
But then again n2 + 4 is a full square.
(cid:3)
Remark 2. Fields Kn are not pairwise different. Indeed, fields K1 and K11 coincide:
if y = −3x3 + 5x ∈ K1, then y4 − 125y2 + 125 = 0.
5.3. The field K1 ≃ Q[x]/(x4 − 5x2 + 5). The transformation x 7→ x3 − 3x is a
generator of Galois group. Elements of the form kx3 + lx2 + mx + n, k, l, m, n ∈ Z
are integral.
Hyperplane Π : 2k + 2l + m + n = 1 is a support hyperplane for the set O. Π
contains 10 integral strictly positive elements:
A1 = (0, 1, 0,−1),
A = (0, 0, 0, 1),
B = (0, 0,−1, 2),
B1 = (−1, 2, 1,−2),
C = (2,−2,−8, 9), C1 = (0, 1,−2, 1),
D = (2,−2,−7, 8), D1 = (1, 0,−3, 2),
E = (1,−1,−4, 5), E1 = (0, 1,−1, 0).
The convex hull of this points in Π is a decahedron, combinatorially the same, as
in subsection 5.1. The only difference is that centers E and E1 of bases ABCD
and A1B1C1D1, respectively, also are integral and strictly positive.
GEOMETRY OF TOTALLY REAL GALOIS FIELDS
OF DEGREE 4
7
5.4. The field K3 ≃ Q[x]/(x4 − 13x2 + 13). The transformation x 7→ x3/3− 11x/3
is a generator of Galois group. Elements of the form
(k, l, m, n) +(cid:18) i
3
, 0,
i
3
, 0(cid:19) +(cid:18)0,
j
3
, 0,
j
3(cid:19) ,
k, l, m, n ∈ Z,
i, j = 0, 1, 2,
are integral.
Hyperplane Π : 2k + 2l + m + n = 1 is a support hyperplane for the set O. Π
contains 36 integral strictly positive elements. The convex hull of this points in Π
is a decahedron, combinatorially the same, as in subsection 5.1. Besides vertices
A, B, C, D, A1, B1, C1, D1 of decahedron the intersection Π ∩ O contains:
(1) centers of bases ABCD and A1B1C1D1;
(2) two points on each segment [A, A1], [B, B1], [C, C1] and [D, D1], which
divide each segment into 3 equal parts;
(3) 18 points inside the decahedron.
5.5. General case. Let assume that a field Kn is not isomorphic to any field Km,
m < n. Put n = 2a + 1, then transformation
x3
2a + 1 −
x 7→
4a2 + 4a + 3
2a + 1
x
is a generator of Galois group. Elements of the form
(k, l, m, n) + i(cid:18) 1
2a + 1
, 0,
2a − 1
2a + 1
, 0(cid:19) + j(cid:18)0,
1
2a + 1
, 0,
2a + 1(cid:19) ,
2a − 1
are integral.
k, l, m, n ∈ Z,
i, j = 0, . . . , 2a,
Hyperplane Π : 2k + 2l + m + n = 1 is a support hyperplane for the set O.
The convex hull of the set Π ∩ O in Π is the same decahedron with vertices
A, B, C, D, A1, B1, C1, D1 and bases -- parallelograms ABCD and A1B1C1D1. In
internal coordinates k, l, m of the hyperplane Π bases lie in parallel planes k + l = 0
and k + l = 1. All vertices are units: A = 1,
B =
2(a + 1)x3 − 2(a + 1)x2 − (8a3 + 16a2 + 16a + 7)x + 8a3 + 16a2 + 18a + 8
2a + 1
,
C = 2x3 − 2x2 − (8a2 + 8a + 8)x + 8a2 + 8a + 9,
D =
2ax3 − 2ax2 − (8a3 + 8a2 + 8a + 1)x + 8a3 + 8a2 + 10a + 2
2a + 1
,
A1 = x2 − 1, B1 =
x3 + 2ax2 − (2a + 3)x + 2
2a + 1
,
C1 = x2 − 2x + 1, D1 = −x3 + (2a + 2)x2 − (2a − 1)x − 2
2a + 1
.
8
YURY KOCHETKOV
The multiplication by unit A1 identifies bases ABCD and A1B1C1D1. Other 8
triangular faces of the "prism" are pairwise identified in the following way:
△AA1B1 ∼ △CDD1, △ABB1 ∼ △CC1D1,
△BB1C ∼ △D1AA1, △B1CC1 ∼ △DD1A.
Besides vertices, the intersection Π∩ O contains centers of bases, 2a points on each
segment [A, A1], [B, B1], [C, C1] and [D, D1], which divide each segment into 2a + 1
equal parts, and points inside the decahedron.
Remark 3. Integers of the field K11 are of the form
(k, l, m, n) + i(cid:18) 1
275
, 0,
3
11
, 0(cid:19) + j(cid:18)0,
1
55
, 0,
4
11(cid:19) .
Therefore, the described above "prism" contains several fundamental decahedrons
of the field K11.
6. Simple fundamental domains in the case of Galois group Z2 ⊕ Z2
In the case of Klein group we don't know fields, where fundamental domain
consists of one 3-dimensional face of Γ, but there are examples, where this domain
consists of two.
6.1. The field K = Q[x]/(x4 − 9x2 + 9). Transformations x 7→ x3/3 − 3x and
x 7→ −x are generators of Galois group. An element kx3 + lx2 + mx + n ∈ K is
integral, if 3k, 3l, m, n ∈ Z.
Hyperplane Π1 : 6k + 3l + m + n − 1 = 0 is a support hyperplane for the set O.
It contains 14 integral strictly positive elements. This elements belong to 2 parallel
2-dimensional planes in Π1: P0 : 9k + 3l + m = 0 and P1 : 9k + 3l + m = 1 (seven in
each). Elements A = (0, 0, 0, 1), B = (0, 1/3,−1, 1), C = (−1/3, 4/3,−1, 0), D =
(−2/3, 2, 0, 1), E = (−2/3, 5/3, 1,−1), F = (−1/3, 2/3, 1, 0) and G = (−1/3, 1, 0, 0
belong to the plane P0. Points A, B, C, D, E and F are vertices of centrally-
symmetric hexagon; A, C, D and F are units; B and E have norm 4; G is the
center of hexagon and has norm 9.
Elements A1 = (0, 1/3, 0, 0), B1 = (−1/3, 4/3, 0,−1), C1 = (−4/3, 4, 1,−4),
D1 = (−2, 17/3, 2,−6), E1 = (−5/3, 14/3, 2,−5), F1 = (−2/3, 2, 1,−2) and G =
(−1, 3, 1,−3 belong to the plane P1. Points A1, B1, C1, D1, E1 and F1 are vertices
of centrally-symmetric hexagon; A1, C1, D1 and F1 are units; B1 and E1 have norm
4; G1 is the center of hexagon and has norm 9.
Polyhedron M1 -- the convex hull in Π1 of the set O∩ Π1, is a "prism" with two
parallel hexagon bases and ten "side" faces: ABA1, A1B1CB, B1C1C, C1D1C,
CDD1, DED1, D1E1F E, E1F1F , F AA1, F1A1F . The multiplication by unit A1
identifies bases. Also
A → F , B → E1, A1 → F1;
D1E1F E: A1 → F , B1 → E1, C → D1, B → E;
DED1: B1 → E, C1 → D1, C → D.
• the multiplication by unit F identifies the face ABA1 with the face E1F1F :
• the multiplication by unit A−1
1 F identifies the face A1B1CB with the face
• the multiplication by unit C −1D identifies the face B1C1C with the face
Hyperplane Π2 : 3k+3l+n−1 = 0 also is a support hyperplane for the set O. The
intersection Π2 ∩ O contains 4 elements: A, A1, F and a unit H = (−1/3, 0, 2, 2).
GEOMETRY OF TOTALLY REAL GALOIS FIELDS
OF DEGREE 4
9
Polyhedron M2 -- the convex hull in Π2 of the set O ∩ Π2, is a tetrahedron, glued
to triangular "side" face of M1. The "side" surface of M1 ∪ M2 is presented below:
F1
r
A1
❏
r
❏
❏
r
A
B1
❏
r
❏
❏
❏
❏❏
B
r
C1
r
✡
❏❏✡
C
r
❏
❏
❏
✡
✡
✡
D1
✡✡❏
r
❏
r
D
❏
❏
❏
E1
❏
r
❏
❏❏
E
r
❏
❏
❏
r
A1
✡✡
✄
✄
✡
✡
✄
✡
✄
H
✑❆
✡
✑
✑
❏❏✡
F
r
r
❆
A
r
Faces C1D1C, CDD1, F1A1F , AA1H, A1F H and F AH are pairwise identified in
the following way:
• the multiplication by unit C −1A identifies the face C1D1C with the face
• the multiplication by unit C −1A1 identifies the face CDD1 with the face
• the multiplication by unit A−1
1 A identifies the face F1A1F with the face
AA1H: C1 → A1, D1 → H, C → A;
A1F H: C → A1, D → H, D1 → F ;
F AH: F1 → F , A1 → A, F → H.
The construction of the fundamental domain is finished.
6.2. The field K = Q[x]/(x4 − 25x2 + 25). Transformations x 7→ x3/5 − 5x and
x 7→ −x are generators of Galois group. An element kx3 + lx2 + mx + n ∈ K is
integral, if 5k, 5l, m, n ∈ Z. Hyperplanes Π1 : 5k + 5l + m + n = 1 and Π2 : −75k +
20l − 3m + n = 1 are support hyperplanes for the set O. The intersection Π1 ∩ O
contains 14 elements and the intersection Π2 ∩ O -- 4. The combinatorial structure
of the fundamental polyhedron here is the same as in the previous example.
Remark 4. The case of the field K = Q[x]/(x4 − 49x2 + 49) is significantly more
complex.
References
[1] Kochetkov Yu. On geometry of cubic Galois fields, Math. Notes, 2011, v. 89(1), 150-155.
[2] Vinberg E. Private communication.
[3] Borevich Z., Shafarevich I., Number Theory, M., Nauka, 1985.
E-mail address: [email protected]
|
1510.01481 | 1 | 1510 | 2015-10-06T08:55:36 | Restricted hom-Lie algebras | [
"math.RA"
] | The paper studies the structure of restricted hom-Lie algebras. More specifically speaking, we first give the equivalent definition of restricted hom-Lie algebras. Second, we obtain some properties of $p$-mappings and restrictable hom-Lie algebras. Finally, the cohomology of restricted hom-Lie algebras is researched. | math.RA | math |
Restricted hom-Lie algebras
Baoling Guan1,2, Liangyun Chen1
1 School of Mathematics and Statistics, Northeast Normal University,
2 College of Sciences, Qiqihar University, Qiqihar 161006, China
Changchun 130024, China
Abstract: The paper studies the structure of restricted hom-Lie algebras. More
specifically speaking, we first give the equivalent definition of restricted hom-Lie
algebras. Second, we obtain some properties of p-mappings and restrictable hom-
Lie algebras. Finally, the cohomology of restricted hom-Lie algebras is researched.
Keywords: Restricted hom-Lie algebras; restrictable hom-Lie algebras; p-mapping;
cohomology
Mathematics Subject Classification 2010: 17A60, 17B56, 17B50
1
Introduction
The concept of a restricted Lie algebra is attributable to N. Jacobson in 1943. It is well
known that the Lie algebras associated with algebraic groups over a field of characteristic p
are restricted Lie algebras [14]. Now, restricted theories attract more and more attentions.
For example: restricted Lie superalgebras [6], restricted Lie color algebras [2], restricted
Leibniz algebras [4], restricted Lie triple systems [8] and restricted Lie algebras [5] were
studied, respectively.
However, The notion of hom-Lie algebras was introduced by Hartwig, Larsson and
Silvestrov in [7] as part of a study of deformations of the Witt and the Virasoro algebras.
In a hom-Lie algebra, the Jacobi identity is twisted by a linear map, called the hom-Jacobi
identity. Some q-deformations of the Witt and the Virasoro algebras have the structure of
a hom-Lie algebra [7]. Because of close relation to discrete and deformed vector fields and
differential calculus [7, 9, 10], hom-Lie algebras are widely studied recently [1, 3, 11, 12, 16–
18]. As a natural generalization of a restricted Lie algebra, it seems desirable to investigate
the possibility of establishing a parallel theory for restricted hom-Lie algebras. As is
well known, restricted Lie algebras play predominant roles in the theories of modular Lie
Corresponding author(L. Chen): [email protected].
Supported by NNSF of China (No.11171055), Natural Science Foundation of Jilin province (No.
201115006), Scientific Research Foundation for Returned Scholars Ministry of Education of China and
the Fundamental Research Funds for the Central Universities(No. 11SSXT146).
1
algebras [15]. Analogously, the study of restricted hom-Lie algebras will play an important
role in the classification of the finite-dimensional modular simple hom-Lie algebras.
The paper study the structure of restricted hom-Lie algebras. Let us briefly describe
the content and setup of the present article.
In Sec. 2, the equivalent definition of
restricted hom-Lie algebras is given. In Sec. 3, we obtain some properties of p-mappings
and restrictable hom-Lie algebras. In Sec. 4, we research the cohomology of restricted
hom-Lie algebras.
In the paper, F is a field of prime characteristic. Let L denote a finite-dimensional
restricted hom-Lie algebra over F.
Definition 1.1. [14] Let L be a Lie algebra over F. A mapping [p] : L → L, a 7→ a[p] is
called a p-mapping, if
(1) ada[p] = (ada)p, ∀a ∈ L,
(2) (αa)[p] = kpa[p], ∀a ∈ L, k ∈ F,
(3) (a + b)[p] = a[p] + b[p] +
si(a, b),
p−1Pi=1
p−1Pi=1
where (ad(a ⊗ X + b ⊗ 1))p−1(a ⊗ 1) =
isi(a, b) ⊗ X i−1 in L ⊗F F[X], ∀a, b ∈ L, The
pair (L, [p]) is referred to as a restricted Lie algebra.
Definition 1.2. [13] (1) A hom-Lie algebra is a triple (L, [., .]L, α) consisting of a linear
space L, a skew-symmetric bilinear map [., .]L : Λ2L → L and a linear map α : L → L
satisfying the following hom-Jacobi identity:
[α(x), [y, z]L]L + [α(y), [z, x]L]L + [α(z), [x, y]L]L = 0
for all x, y, z ∈ L;
(2) A hom-Lie algebra is called a multiplicative hom-Lie algebra if α is an algebraic
morphism, i.e., for any x, y ∈ L, we have α([x, y]L) = [α(x), α(y)]L;
(3) A sub-vector space η ⊂ L is called a hom-Lie subalgebra of (L, [., .]L, α) if α(η) ⊂ η
and η is closed under the bracket operation [., .]L, i.e., [x, y]L ∈ η for all x, y ∈ η;
(4) A sub-vector space η ⊂ L is called a hom-Lie ideal of (L, [., .]L, α) if α(η) ⊂ η and
[x, y]L ∈ η for all x ∈ η, y ∈ L.
2 The equivalent definition of restricted hom-Lie al-
gebras
Let (L, [, ]L, α) be a multiplicative hom-Lie algebra over F. For c ∈ L satisfying α(c) = c,
we define adc(a) := [α(a), c]. Put L0 := {xα(x) 6= x} ∪ {0} and L1 := {xα(x) = x}.
Then L = L0 ∪ L1 and L1 is a hom-Lie subalgebra of L.
2
Definition 2.1. Let (L, [, ]L, α) be a multiplicative hom-Lie algebra over F. A mapping
[p] : L1 → L1, a 7→ a[p] is called a p-mapping, if
(1) [α(y), x[p]] = (adx)p(y), ∀x ∈ L1, y ∈ L,
(2) (kx)[p] = kpx[p], ∀x ∈ L1, k ∈ F,
(3) (x + y)[p] = x[p] + y[p] +
si(x, y),
p−1Pi=1
p−1Pi=1
where (ad(x ⊗ X + y ⊗ 1))p−1(x ⊗ 1) =
isi(x, y) ⊗ X i−1 in L ⊗F F[X], ∀x, y ∈ L1, α(x ⊗
X) = α(x) ⊗ X. The pair (L, [, ]L, α, [p]) is referred to as a restricted hom-Lie algebra.
From the above definition, we may see that (i) α(x[p]) = (α(x))[p] for all x ∈ L1, i.e.,
α ◦ [p] = [p] ◦ α; (ii) By (1) of the definition, one gets adx[p] = (adx)p for all x ∈ L1.
Let (L, α) be a hom-Lie algebra over F and f : L → L be a mapping. f is called a
p-semilinear mapping, if f (kx + y) = kpf (x) + f (y), ∀x, y ∈ L, ∀k ∈ F. Let S be a subset
of a hom-Lie algebra (L, α). We put CL(S) := {x ∈ L [α(y), x] = 0, ∀y ∈ S}. CL(S) is
called the centralizer of S in L. Put C(L) := {x ∈ L [α(y), x] = 0, ∀y ∈ L}. C(L) is
called the center of L.
Definition 2.2. Let (L, [., .]L, α) be a restricted hom-Lie algebra over F. A hom-Lie sub-
algebra H of L is called a p-subalgebra, if x[p] ∈ H1 for all x ∈ H1, where H1 = {x ∈
Hα(x) = x}.
Proposition 2.3. Let L be a hom-Lie subalgebra of a restricted hom-Lie algebra (G, [, ]G, α, [p])
and [p]1 : L1 → L1 a mapping. Then the following statements are equivalent:
(1) [p]1 is a p-mapping on L1.
(2) There exists a p-semilinear mapping f : L1 → CG(L) such that [p]1 = [p] + f.
Proof.
(1)⇒(2). Consider f : L1 → G, f (x) = x[p]1 − x[p]. Since adf (x)(y) =
[α(y), f (x)] = 0, ∀x ∈ L1, y ∈ L, f actually maps L1 into CG(L). For x, y ∈ L1, k ∈ F, we
obtain
f (kx + y)
= kpx[p]1 + y[p]1 +
= kpf (x) + f (y),
p−1Xi=1
si(kx, y) − kpx[p] − y[p] −
si(kx, y)
p−1Xi=1
which proves that f is p-semilinear.
(2)⇒(1). We next will check three conditions of the definition step and step. For
x, y ∈ L1, we have
(x + y)[p]1 = (x + y)[p] + f (x + y)
3
si(x, y)
p−1Xi=1
= x[p] + f (x) + y[p] + f (y) +
= x[p]1 + y[p]1 +
si(x, y)
p−1Xi=1
and
For x ∈ L1, z ∈ L, one gets
(kx)[p]1 = (kx)[p] + f (kx)
= kpx[p] + kpf (x)
= kp(x[p] + f (x))
= kpx[p]1.
adx[p]1(z) = ad(x[p] + f (x))(z)
= adx[p](z) + adf (x)(z)
= adx[p](z)
= (adx)p(z).
The proof is complete.
Corollary 2.4. The following statements hold.
(1) If C(L) = 0, then L admits at most one p-mapping.
(2) If two p-mappings coincide on a basis, then they are equal.
(3) If (L, [, ]L, α, [p]) is restricted, then there exists a p-mapping [p]′ of L such that
= 0, ∀x ∈ C(L1).
′
x[p]
Proof.
(1) We set G = L. Then CG(L) = C(L), the only p-semilinear mapping
occurring in Proposition 2.3 is the zero mapping.
(2) If two p-mappings coincide on a basis, their difference vanishes since it is p-
semilinear.
(3) [p]C(L1) defines a p-mapping on C(L1). Since C(L1) is abelian, it is p-semilinear.
Extend this to a p-semilinear mapping f : L1 → C(L1). Then [p]′ := [p]−f is a p-mapping
of L, vanishing on C(L1).
From the proof of Theorem 2 in [18], we see the following definition:
Definition 2.5. Let (L, [., .]L, αL) be a hom-Lie algebra, and let j : L → UHLie(L) be the
composition of the maps L ֒→ FHN As(L) ։ UHLie(L). The pair (UHLie(L), j) is called
a universal enveloping algebra of L if for every hom-associative algebra (A, µA, αA) and
every morphism f : L → HLie(A) of hom-Lie algebras, there exists a unique morphism
h : UHLie(L) → A of hom-associative algebras such that f = h ◦ j (as morphisms of
F-modules).
4
In the special case of G = UHLie(L)− ⊃ L, where UHLie(L) is the universal enveloping
algebra of hom-Lie algebra L (see [18]) and UHLie(L)− denotes a hom-Lie algebra given
by hom-associative algebra UHLie(L) via the commutator bracket. We have the following
theorem:
Theorem 2.6. Let (ej)j∈J be a basis of L1 such that there are yj ∈ L1 with (adej)p = adyj.
Then there exists exactly one p-mapping [p] : L1 → L such that e[p]
j = yj, ∀j ∈ J.
Proof. For z ∈ L1, we have 0 = ((adej)p − adyj)(z) = [α(z), ep
j − yj ∈
CUHLie(L1)(L1), ∀j ∈ J. We define a p-semilinear mapping f : L1 → CUHLie(L1)(L1) by
means of
j − yj]. Then ep
j (yj − ep
j ).
f (X αjej) :=X αp
p−1Xi=1
Consider V := {x ∈ L1xp + f (x) ∈ L1}. The equation
(kx + y)p + f (kx + y) = kpxp + yp +
si(kx, y) + kpf (x) + f (y)
ensures that V is a subspace of L1. Since it contains the basis (ej)j∈J, we conclude that
xp + f (x) ∈ L1, ∀x ∈ L1. By virtue of Proposition 2.3, [p] : L1 → L, x[p] := xp + f (x)
is a p-mapping on L1. In addition, we obtain e[p]
j + f (ej) = yj, as asserted. The
uniqueness of [p] follows from Corollary 2.4.
j = ep
Definition 2.7. A multiplicative hom-Lie algebra (L, [., .]L, αL) is called restrictable, if
(adx)p ∈ adL1 for all x ∈ L1, where adL1 = {adxx ∈ L1}.
Theorem 2.8. L is a restrictable hom-Lie algebra if and only if there is a p-mapping
[p] : L1 → L1 which makes L a restricted hom-Lie algebra.
Proof.
(⇐) By the definition of p-mapping [p], for x ∈ L1, there exists x[p] ∈ L1
such that (adx)p = adx[p] ∈ adL1. Hence L is restrictable.
(⇒) Let L be restrictable. Then for x ∈ L1, we have (adx)p ∈ adL1, that is, there
exists y ∈ L1 such that (adx)p = ady. Let (ej)j∈J be a basis of L1. Then there exists
yj ∈ L1 such that (adej)p = adyj(j ∈ J). By Theorem 2.6, there exists exactly one p-
mapping [p] : L1 → L1 such that e[p]
j = yj, ∀j ∈ J, which makes L a restricted hom-Lie
algebra.
3 Properties of p-mappings and restrictable hom-Lie
algebras
In the section, we will discuss some properties of p-mappings and restrictable hom-Lie
algebras.
5
Definition 3.1. [13] Let (L, [·, ·]L, α) and (Γ, [·, ·]Γ, β) be two hom-Lie algebras. A linear
map φ : L → Γ is said to be a morphism of hom-Lie algebras if
φ[u, v]L = [φ(u), φ(v)]Γ,
∀u, v ∈ L,
φ ◦ α = β ◦ φ.
(1)
(2)
Denote by Gφ = {(x, φ(x))x ∈ L} ⊆ L ⊕ Γ the graph of a linear map φ : L → Γ.
Definition 3.2. A morphism of hom-Lie algebras φ : (L, [·, ·]L, α, [p]1) → (Γ, [·, ·]Γ, β, [p]2)
is said to be restricted if φ(x[p]1) = (φ(x))[p]2 for all x ∈ L.
Proposition 3.3. Given two restricted hom-Lie algebras (L, [·, ·]L, α, [p]1) and (Γ, [·, ·]Γ, β,
[p]2), there is a restricted hom-Lie algebra (L ⊕ Γ, [·, ·]L⊕Γ, α + β, [p]), where the bilinear
map [·, ·]L⊕Γ : ∧2(L ⊕ Γ) → L ⊕ Γ is given by
[u1 + v1, u2 + v2]L⊕Γ = [u1, u2]L + [v1, v2]Γ,
∀u1, u2 ∈ L, v1, v2 ∈ Γ,
and the linear map (α + β) : L ⊕ Γ → L ⊕ Γ is given by
(α + β)(u + v) = α(u) + β(v),
∀u ∈ L, v ∈ Γ,
the p-mapping [p] : L ⊕ Γ → L ⊕ Γ is given by
(u + v)[p] = u[p]1 + v[p]2,
∀u ∈ L, v ∈ Γ.
Proof. Recall that L1 = {x ∈ Lα(x) = x} and Γ1 = {x ∈ Γβ(x) = x}. For any
u1, u2 ∈ L, v1, v2 ∈ Γ, we have
[u2 + v2, u1 + v1]L⊕Γ = [u2, u1]L + [v2, v1]Γ
= −[u1, u2]L − [v1, v2]Γ
= −[u1 + v1, u2 + v2]L⊕Γ.
The bracket is obviously skew-symmetric. By a direct computation we have
[(α + β)(u1 + v1), [u2 + v2, u3 + v3]L⊕Γ]L⊕Γ
+c.p.((u1 + v1), (u2 + v2), (u3 + v3))
= [α(u1) + β(v1), [u2, u3]L + [v2, v3]Γ]L⊕Γ + c.p.
= [α(u1), [u2, u3]L]L + c.p.(u1, u2, u3) + [β(v1), [v2, v3]Γ]Γ
+c.p.(v1, v2, v3)
= 0,
where c.p.(a, b, c) means the cyclic permutations of a, b, c. For any u1 ∈ L1, v1 ∈ Γ1, u2 ∈
L, v2 ∈ Γ, we obtain
ad(u1 + v1)[p](u2 + v2) = [(α + β)(u2 + v2), (u1 + v1)[p]]L⊕Γ
6
1 + v[p]2
]L + [β(v2), v[p]2
= [α(u2) + β(v2), u[p]1
= [α(u2), u[p]1
= adu[p]1
= (adu1)p(u2) + (adv1)p(v2)
(u2) + adv[p]2
(v2)
1
1
1
1
1
]L⊕Γ
]Γ
and
(ad(u1 + v1))p(u2 + v2)
= [[[αp(u2) + βp(v2),
p
u1 + v1], u1 + v1], · · · , u1 + v1]L⊕Γ
= [[[αp(u2),
= (adu1)p(u2) + (adv1)p(v2).
u1], u1], · · · , u1]L + [[[βp(v2),
v1], v1], · · · , v1]Γ
z
z
}
{
p
}
z
{
p
}
{
Hence ad(u1 + v1)[p](u2 + v2) = (ad(u1 + v1))p(u2 + v2), thus ad(u1 + v1)[p] = (ad(u1 + v1))p.
Moreover, for any u1, u2 ∈ L1, v1, v2 ∈ Γ1, one gets
((u1 + v1) + (u2 + v2))[p] = ((u1 + u2) + (v1 + v2))[p] = (u1 + u2)[p]1 + (v1 + v2)[p]2
= u[p]
1 + u[p]
2 +
p−1Xi=1
si(u1, u2) + v[p]
1 + v[p]
2 +
si(v1, v2)
= (u[p]
1 + v[p]
1 ) + (u[p]
2 + v[p]
2 ) + (
si(u1, u2) +
p−1Xi=1
si(v1, v2))
p−1Xi=1
p−1Xi=1
p−1Xi=1
p−1Xi=1
= (u1 + v1)[p] + (u2 + v2)[p] +
= (u1 + v1)[p] + (u2 + v2)[p] +
and
(si(u1, u2)) + si(v1, v2))
si((u1, v1) + (u2, v2))
(k(u1 + v1))[p] = (ku1 + kv1)[p] = (ku1)[p]1 + (kv1)[p]2
= kpu[p]1
1 + kpv[p]2
= kp(u1 + v1)[p].
1 = kp(u[p]1
1 + v[p]2
)
1
Therefore, (L ⊕ Γ, [·, ·]L⊕Γ, α + β, [p]) is a restricted hom-Lie algebra.
Proposition 3.4. A linear map φ : (L, [·, ·]L, α, [p]1) → (Γ, [·, ·]Γ, β, [p]2) is a restricted
morphism of restricted hom-Lie algebras if and only if the graph Gφ ⊆ L⊕Γ is a restricted
hom-Lie subalgebra of (L ⊕ Γ, [·, ·]L⊕Γ, α + β, [p]).
Proof. Let φ : (L, [·, ·]L, α) → (Γ, [·, ·]Γ, β) be a restricted morphism of restricted
hom-Lie algebras. By (1), we have
[u + φ(u), v + φ(v)]L⊕Γ = [u, v]L + [φ(u), φ(v)]Γ = [u, v]L + φ[u, v]L.
7
Then the graph Gφ is closed under the bracket operation [·, ·]L⊕Γ. Furthermore, by (2),
we have
(α + β)(u + φ(u)) = α(u) + β ◦ φ(u) = α(u) + φ ◦ α(u),
which implies that (α + β)(Gφ) ⊆ Gφ. Thus, Gφ is a hom-Lie subalgebra of (L ⊕
Γ, [·, ·]L⊕Γ, α + β). Moreover, for u + φ(u) ∈ Gφ, one gets
(u + φ(u))[p] = u[p]1 + (φ(u))[p]2 = u[p]1 + φ(u[p]1) ∈ Gφ.
Thereby, the graph Gφ ⊆ L ⊕ Γ is a restricted hom-Lie subalgebra of (L ⊕ Γ, [·, ·]L⊕Γ, α +
β, [p]).
Conversely, if the graph Gφ ⊆ L ⊕ Γ is a restricted hom-Lie subalgebra of (L ⊕
Γ, [·, ·]L⊕Γ, α + β, [p]), then we have
[u + φ(u), v + φ(v)]L⊕Γ = [u, v]L + [φ(u), φ(v)]Γ ∈ Gφ,
which implies that
[φ(u), φ(v)]Γ = φ[u, v]L.
Furthermore, (α + β)(Gφ) ⊂ Gφ yields that
(α + β)(u + φ(u)) = α(u) + β ◦ φ(u) ∈ Gφ,
which is equivalent to the condition β ◦ φ(u) = φ ◦ α(u), i.e. β ◦ φ = φ ◦ α. Therefore, φ is
a morphism of restricted hom-Lie algebras. Since Gφ is a restricted hom-Lie subalgebra
of (L ⊕ Γ, [·, ·]L⊕Γ, α + β, [p]), we have
(u + φ(u))[p] = u[p]1 + (φ(u))[p]2 ∈ Gφ.
Thus, (φ(u))[p]2 = φ(u[p]1) for u ∈ L, i.e., φ is a restricted morphism.
One advantage in considering restrictable hom-Lie algebras instead of restricted ones
rests on the following theorem.
Theorem 3.5. Let f : (L, [, ]L, α, [p]1) → (L′, [, ]L
phism of hom-Lie algebras. If L is restrictable, so is L′.
′ , β, [p]2) be a surjective restricted mor-
Proof.
It follows from f is a surjective mapping that L′ = f (L). Then for x ∈ L1,
1, where L1 = {x ∈ Lα(x) = x} and
we have β(f (x)) = f (α(x)) = f (x) and f (x) ∈ L′
L′
1 = {x ∈ L′β(x) = x}. For y ∈ L, one gets
(adf (x))p(f (y)) = (adf (x))p−1[β(f (y)), f (x)]
= (adf (x))p−2[[β2(f (y)), β(f (x))], f (x)]
= [[[βpf (y), f (x)], f (x)], · · · , f (x)]
= βp[[[f (y), f (x)], f (x)], · · · , f (x)]
}
}
{z
{z
p
p
8
= βp ◦ f [[[y, x], x], · · · , x]
= f [[[αp(y), x], x], · · · , x]
p
{z
}
p
{z
}
= f ((adx)p(y)) = f ((adx[p]1)(y)) = f [α(y), x[p]1]
= f [α(y), α(x[p]1)] = f ◦ α[y, x[p]1] = β ◦ f [y, x[p]1]
= β[f (y), f (x[p]1)] = [β(f (y)), β(f (x[p]1))]
= [β(f (y)), f (x[p]1)] = adf (x[p]1)(f (y))
= ad(f (x))[p]2(f (y)).
We have (adf (x))p = ad(f (x))[p]2 ∈ adL1
′
. Hence L′ is restrictable.
Theorem 3.6. Let A and B be hom-Lie ideals of hom-Lie algebra (L, [., .]L, α) such that
L = A ⊕ B. Then L is restrictable if and only if A, B are restrictable.
Proof.
(⇐) If A, B are restrictable, for x ∈ L1 with α(x) = x, we may suppose that
x = x1 + x2, where x1 ∈ A, x2 ∈ B. Then α(x1 + x2) = α(x1) + α(x2) = x1 + x2. Since
A and B are hom-Lie ideals, one gets α(x1) ∈ A, α(x2) ∈ B. we obtain α(x1) = x1 and
α(x2) = x2. As A, B are restrictable, then there exists y1 ∈ A1, y2 ∈ B1 with α(y1) = y1
and α(y2) = y2, such that (adx1)p = ady1 and (adx2)p = ady2. Thus,
(ad(x1 + x2))p = (adx1 + adx2)p
= (adx1)p + (adx2)p = ady1 + ady2
= ad(y1 + y2).
Therefore, L is restrictable.
(⇒) If L is restrictable , so are A ∼= L/B, B ∼= L/A by Theorem 3.5.
Corollary 3.7. Let A, B be restrictable hom-Lie ideals of a restricted hom-Lie algebra
(L, [., .]L, α, [p]) such that L = A + B and [A, B] = 0. Then L is restrictable.
Proof. Define a mapping f : A ⊕ B → L, (x, y) 7→ x + y. Clearly, f is a surjection.
For (x1, y1), (x2, y2) ∈ A ⊕ B, by [A, B] = 0, one gets [x1, y2] = [y1, x2] = 0. We have
f [(x1, y1), (x2, y2)] = f ([x1, x2], [y1, y2])
= [x1, x2] + [y1, y2] = [x1, x2] + [x1, y2] + [y1, x2] + [y1, y2]
= [x1 + y1, x2 + y2] = [f (x1, y1), f (x2, y2)].
Moreover, one gets
α ◦ f (x, y) = α(x + y)
= α(x) + α(y) = f ((α(x), α(y)))
= f ◦ α(x, y).
Therefore, α ◦ f = f ◦ α. For x ∈ A, y ∈ B, α(x, y) = (x, y), we have
f ((x, y)[p]) = f ((x[p]1, y[p]2))
9
= x[p]1 + y[p]2 = (x + y)[p]
= (f (x, y))[p].
Thus, f is a restricted morphism. By Theorem 3.6, we have A ⊕ B is restrictable. By
Theorem 3.5, one gets L is restrictable.
Definition 3.8. Let (L, [., .]L, α) be a hom-Lie algebra and ψ be a symmetric bilinear form
on L. ψ is called associative, if ψ(x, [z, y]) = ψ([α(z), x], y).
Definition 3.9. Let (L, [., .]L, α) be a hom-Lie algebra and ψ a symmetric bilinear form
on L. Set L⊥ = {x ∈ Lψ(x, y) = 0, ∀ y ∈ L}. L is called nondegenerate, if L⊥ = 0.
Theorem 3.10. Let L be a subalgebra of the restricted hom-Lie algebra (G, [., .]G, α, [p])
with C(L) = {0}. Assume λ : G × G → F to be an associative symmetric bilinear form,
which is nondegenerate on L × L. Then L is restrictable.
Proof. Since λ is nondegenerate on L × L, every linear form f on L is determined by
a suitably chosen element y ∈ L : f (z) = λ(y, z), ∀z ∈ L. Let x ∈ L1. Then there exists
y ∈ L such that
λ(x[p], z) = λ(y, z), ∀z ∈ L.
This implies that 0 = λ(x[p] − y, [L, L]) = λ([α(L), x[p] − y], L) and [α(L), x[p] − y] = 0.
Therefore, x[p] − y ∈ C(L) = {0} and y = x[p] ∈ L1. Moreover, we obtain
(adxL)p = adx[p]L = adyL,
which proves that L is restrictable.
Proposition 3.11. Let (L, [., .]L, α) be a restrictable hom-Lie algebra and H a subalgebra
of L. Then H is a p-subalgebra for some mapping [p] on L if and only if (adH1L)p ⊆
adH1L.
Proof.
(⇒) If H is a p-subalgebra, then for x ∈ H1, x[p] ∈ H1, and (adx)p = adx[p] ⊆
adH1L. Hence, (adH1L)p ⊆ adH1L.
(⇐) If (adH1L)p ⊆ adH1L, then H is restrictable. By Theorem 2.8, H is restricted.
Thereby, H is a p-subalgebra of L.
4 Cohomology of restricted hom-Lie algebras
In this section, we will discuss the cohomology of restricted hom-Lie algebras in the abelian
case, which is similar to the reference [5]. The following definition is analogous to that of
the restricted universal enveloping algebra in the reference [14].
10
Definition 4.1. Let (L, [., .]L, α, [p]) be a restricted hom-Lie algebra. The pair (u(L), α′, i)
consisting of a hom-associative algebra (u(L), α′) with unity and a restricted hom-morphism
i : L → u(L)− is called a restricted hom-universal enveloping algebra of L if given any
hom-associative algebra (A, β) with unity and any restricted hom-morphism f : L → A−,
there exists a unique morphism ¯f : u(L) → A of hom-associative algebras such that
¯f ◦ i = f.
Definition 4.2. [11] Let A = (V, µ, α) be a hom-associative F-algebra. An A-module is a
triple (M, f, γ) where M is F-vector space and f, γ are F-linear maps, f : M −→ M and
γ : V ⊗ M −→ M, such that the following diagram commutes:
V ⊗ M
γ
−−−→ M
V ⊗ V ⊗ M
−−−→ V ⊗ M
µ⊗f
xα⊗γ
xγ
We let S∗(L) and Λ∗(L) denote the symmetric and alternating algebras of restricted
hom-Lie algebra (L, [., .]L, α, [p]), respectively. Bases for the homogeneous subspaces of
degree k for these spaces consist of monomials eµ = eµ1
n and e~i = ei1 ∧ · · · ∧ eik,
respectively, where
1 · · · eµn
µ = (µ1, · · · , µn) ∈ Zn satisfies µj ≥ 0, µ =Pj µj = k;
~i = (i1, · · · , ik) ∈ Zk satisfies 1 ≤ i1 < · · · < ik ≤ n.
Let α : λ 7→ λp denote the Frobenius automorphism of F. If V is an abelian group with
an F-vector space structure given by F → End(V ), then the composition
F α−1
−−→ F → End(V )
gives another vector space structure on V which we will denote by V . Of course V is
isomorphic to V as an F-vector space (they have the same dimension). We note that if
W is any other F-vector space, then a p-semilinear map V → W is a linear map V → W
and vice versa.
In sequel, (L, [., .]L, α, [p]) denotes a finite-dimensional restricted hom-Lie algebra
over F such that [gi, gj] = 0 for all gi, gj ∈ L and (u(L), α′, i) denotes the restricted
hom-universal enveloping algebra of L. Here we take α = α′ and α satisfies α(u1u2) =
α(u1)α(u2) for u1, u2 ∈ u(L). For s, t ≥ 0, we define
with the u(L)-module structure given by
Cs,t = StL1 ⊗ ΛsL ⊗ u(L)
u(h1 · · · ht ⊗ g1 ∧ · · · ∧ gs ⊗ x) = h1 · · · ht ⊗ g1 ∧ · · · ∧ gs ⊗ α(u)x,
where hi, gj ∈ L and u, x ∈ u(L). If either s < 0 or t < 0, we put Cs,t = 0 and define
Ck = M2t+s=k
Cs,t
11
for all k ∈ Z. Note that each Ck is a free u(L)-module. If not both t = 0 and s = 0, we
then define a map
ds,t : Cs,t → Ct,s−1 ⊕ Ct−1,s+1
by the formulas
dt,s(h1 · · · ht ⊗ g1 ∧ · · · ∧ gs ⊗ x)
=
+
−
sXi=1
tXj=1
tXj=1
(−1)i−1h1 · · · ht ⊗ α(g1) ∧ · · · [α(gi) · · · ∧ α(gs) ⊗ α(gi)x
j ∧ α(g1) ∧ · · · ∧ α(gs) ⊗ α(x)
h1 · · ·bhj · · · ht ⊗ h[p]
h1 · · ·bhj · · · ht ⊗ hj ∧ α(g1) ∧ · · · ∧ α(gs) ⊗ hp−1
j x.
(3)
(4)
(5)
For k ≥ 1, we define the map dk : Ck → Ck−1 by dk =L2t+s=k ds,t. Then we obtain the
following theorem.
Theorem 4.3. The maps dk defined above satisfy dk−1dk = 0 for k ≥ 1, so that C =
(Ck, dk) is an augmented complex of free u(g)-modules.
Proof. The terms in the sum (3) are elements of Ct,s−1 whereas the terms in the
sums (4) and (5) lie in Ct−1,s+1. Therefore, in order to compute dk−1dk = 0, we must
apply dt,s−1 to (3) and dt−1,s+1 to (4) and (5). Applying dt,s−1 to (5), we have
(−1)σh1 · · · ht ⊗ α2(g1) ∧ · · · \α2(gi) · · · \α2(gσ) · · · ∧ α2(gs) ⊗ α2(gσ)(α(gi)x)
(−1)σ−1h1 · · · ht ⊗ α2(g1) ∧ · · · \α2(gσ) · · · \α2(gi) · · · ∧ α2(gs)
⊗α2(gσ)(α(gi)x)
(−1)i−1h1 · · · ht ⊗ α(g1) ∧ · · · [α(gi) · · · ∧ α(gs) ⊗ α(gi)x(cid:1)
dt,s(cid:0) sXi=1
sXi=1
(−1)i−1(cid:0)Xσ<i
+Xσ>i
tXj=1
h1 · · ·bhj · · · ht ⊗ h[p]
tXj=1
h1 · · ·bhj · · · ht ⊗ hj ∧ α2(g1) ∧ · · · \α2(gi) · · · ∧ α2(gs) ⊗ hp−1
sXi=1
(−1)i−1(cid:0)Xσ<i
−
+
j
⊗(α(gσ)α(gi))α(x)
=
=
j ∧ α2(g1) ∧ · · · \α2(gi) · · · ∧ α2(gs) ⊗ α(α(gi)x)
(−1)σ−1h1 · · · ht ⊗ α2(g1) ∧ · · · \α2(gσ) · · · \α2(gi) · · · ∧ α2(gs)
(α(gi)x)(cid:1)
12
+Xσ>i
tXj=1
tXj=1
+
−
(−1)σh1 · · · ht ⊗ α2(g1) ∧ · · · \α2(gi) · · · \α2(gσ) · · · ∧ α2(gs) ⊗ (α(gσ)α(gi))α(x)
j ∧ α2(g1) ∧ · · · \α2(gi) · · · ∧ α2(gs) ⊗ α(α(gi)x)
h1 · · ·bhj · · · ht ⊗ h[p]
h1 · · ·bhj · · · ht ⊗ hj ∧ α2(g1) ∧ · · · \α2(gi) · · · ∧ α2(gs) ⊗ (hp−1
j α(gi))α(x)(cid:1). (7)
(6)
Since α(gi)α(gj) = α(gj)α(gi) in u(g), the terms in the first two sums in the parentheses
cancel in pairs when summed over all i. This leaves the sum over i of (6) and (7). Now
we apply dt−1,s+1 to (4).
j ∧ α(g1) ∧ · · · ∧ α(gs) ⊗ α(x)(cid:1) =
tXj=1
j ∧ α2(g1) ∧ · · · \α2(gσ) · · · ∧ α2(gs) ⊗ α2(gσ)α(x) (8)
dt−1,s+1(cid:0) tXj=1
h1 · · ·bhj · · · ht ⊗ h[p]
(cid:0) sXσ=1
(−1)σh1 · · ·bhj · · · ht ⊗ h[p]
+h1 · · ·bhj · · · ht ⊗ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α(h[p]
+Xτ 6=j
h1 · · · bhτ · · ·bhj · · · ht ⊗ h[p]
−Xτ 6=j
h1 · · · bhτ · · ·bhj · · · ht ⊗ hτ ∧ h[p]
τ ∧ h[p]
j )α(x)
j ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α2(x)
j ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ hp−1
τ α(x)(cid:1).
(9)
(10)
(11)
We note that the terms in (10) cancel in pairs since interchanging the first two terms in
the alternating product multiplies the term by −1. Finally, we apply dt−1,s+1 to (5) to
get
= −
tXj=1
dt−1,s+1(cid:0) −
j x(cid:1)
h1 · · ·bhj · · · ht ⊗ hj ∧ α2(g1) ∧ · · · ∧ α(gs) ⊗ hp−1
tXj=1(cid:0) sXσ=1
(−1)σh1 · · ·bhj · · · ht ⊗ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α2(gσ)(hp−1
+h1 · · ·bhj · · · ht ⊗ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α2(hj)(hp−1
+Xτ 6=j
h1 · · · bhτ · · ·bhj · · · ht ⊗ h[p]
−Xτ 6=j
h1 · · ·bhτ · · ·bhj · · · ht ⊗ hτ ∧ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ hp−1
τ ∧ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α(hp−1
(hp−1
j x)
j x)
j x)(cid:1)
τ
j x)
13
= −
(−1)σh1 · · ·bhj · · · ht ⊗ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α(gσhp−1
tXj=1(cid:0) sXσ=1
+h1 · · ·bhj · · · ht ⊗ α2(g1) ∧ · · · ∧ α2(gs) ⊗ hp
+Xτ 6=j
h1 · · · bhτ · · ·bhj · · · ht ⊗ h[p]
−Xτ 6=j
h1 · · · bhτ · · ·bhj · · · ht ⊗ hτ ∧ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ (hp−1
τ ∧ hj ∧ α2(g1) ∧ · · · ∧ α2(gs) ⊗ α(hp−1
τ hp−1
j α(x)
j
j
j x)
)α(x)(12)
(13)
(14)
)α(x)(cid:1).(15)
This time the terms in (15) cancel in pairs. Moreover, the terms in (6) and (8) are
identical (with σ = i) except for sign and hence they cancel. The terms in (7) and (12)
cancel in pairs since α(hp−1
). The terms in (11) and (14) have the
same sign but are equal apart from interchanging the first two terms in the alternating
part. Finally the terms in (9) and (13) match except for sign since h[p]
j in u(g) and
hence the entire sum is zero as claimed. This completes the proof.
)α(gj) = α(gj)α(hp−1
j = hp
i
i
We next will consider the cohomology of restricted hom-Lie algebras in the case of
simpleness. A basis for the space Ct,s consists of the monomials
eµ ⊗ eI ⊗ er = eµ1
1 · · · eµn
n ⊗ ei1 ∧ · · · ∧ eis ⊗ er1
1 · · · ern
n ,
where µ = (µ1, · · · , µn), I = (i1, · · · , is), r = (r1, · · · , rn) and
µj ≥ 0, µ =Xj
µj = t, 1 ≤ i1 < · · · < is ≤ n, 0 ≤ rj ≤ p − 1.
For each i = 1, · · · , n and ei ∈ L1, we let
ci = 1 ⊗ e[p]
i ⊗ 1 − 1 ⊗ ei ⊗ ep−1
i
and we easily note that ci ∈ C0,1 is a cycle for all i. Now we define
(∂/∂ei ⊗ ci) : Ct,s −→ Ct−1,s+1
by the formula
(
∂
∂ei
⊗ ci)(eµ ⊗ eI ⊗ er) =
∂eµ
∂ei
⊗ e[p]
i ∧ α(eI) ⊗ α(er) −
∂eµ
∂ei
⊗ ei ∧ α(eI) ⊗ ep−1
i α(er).
If µ = (µ1, · · · , µn) satisfies µ = t and I = (i1, · · · , is) is increasing, then by the definition
we write
(−1)Jeµ ⊗ fi1 ∧ · · · ∧ fis ⊗ e
qi1
i1 · · · eqis
is
eµ ⊗ cI = XJ⊂{1,··· ,s}
and
eµ ⊗ α(cI) = XJ⊂{1,··· ,s}
(−1)Jeµ ⊗ α(fi1) ∧ · · · ∧ α(fis) ⊗ α(e
qi1
i1 ) · · · α(eqis
is ),
14
where
fij =( eij ,
e[p]
ij ,
j ∈ J
j /∈ J;
qij =(cid:26) p − 1,
0,
j ∈ J
j /∈ J.
We then define Ct,s to be the F-subspace of Ct,s spanned by the elements {eµ ⊗ α(cI) :
µ = t and I is increasing } and
Ck = M2t+s=k
Ct,s.
The boundary operator ∂k = ∂ : Ck −→ Ck−1 is defined by
∂ =
∂
∂ej
nXj=1
⊗ cj.
Then we may show that ∂2 = 0. In fact,
∂2(eµ ⊗ cI) = ∂(∂(eµ ⊗ cI))
(−1)Jeµ ⊗ fi1 ∧ · · · ∧ fis ⊗ e
qi1
i1 · · · eqis
is ))
(−1)J(
⊗ e[p]
j ∧ α(fi1) ∧ · · · ∧ α(fis) ⊗ α(e
qi1
i1 ) · · · α(eqis
is )
⊗ cj( XJ⊂{1,··· ,s}
∂eµ
∂ej
= ∂(
= ∂(
∂
∂ej
nXj=1
nXj=1 XJ⊂{1,··· ,s}
−
∂eµ
∂ej
⊗ ej ∧ α(fi1) ∧ · · · ∧ α(fis) ⊗ ep−1
j α(e
qi1
i1 ) · · · α(eqis
is )))
=
nXl=1
nXj=1 XJ⊂{1,··· ,s}
(−1)J{
∂
∂el
⊗ cl(
∂eµ
∂ej
⊗ e[p]
j ∧ α(fi1) ∧ · · · ∧ α(fis) ⊗ α(e
qi1
i1 ) · · ·
α(eqis
is )) −
∂
∂el
⊗ cl(
∂eµ
∂ej
⊗ ej ∧ α(fi1) ∧ · · · ∧ α(fis) ⊗ ep−1
j α(e
qi1
i1 ) · · · α(eqis
is ))}
=
nXl=1
nXj=1 XJ⊂{1,··· ,s}
(−1)J{
)
∂( ∂eµ
∂ej
∂el
⊗ e[p]
l ∧ α(e[p]
j ) ∧ α2(fi1) ∧ · · · ∧ α2(fis)
⊗α2(e
qi1
i1 ) · · · α2(eqis
is )
⊗ el ∧ α(e[p]
j ) ∧ α2(fi1) ∧ · · · ∧ α2(fis) ⊗ ep−1
l α2(e
qi1
i1 ) · · · α2(eqis
is )
(16)
(17)
⊗ e[p]
l ∧ α(ej) ∧ α2(fi1) ∧ · · · ∧ α2(fis) ⊗ α(ep−1
j
)α2(e
qi1
i1 ) · · · α2(eqis
is )(18)
)
)
∂( ∂eµ
∂ej
∂el
∂( ∂eµ
∂ej
∂el
−
−
)
∂( ∂eµ
∂ej
∂el
+
⊗ el ∧ α(ej) ∧ α2(fi1) ∧ · · · ∧ α2(fis) ⊗ ep−1
l α(ep−1
j
)α2(e
qi1
i1 ) · · · α2(eqis
is )}.
15
(19)
This time the terms in (16) cancel in pairs, and the terms in (19) cancel in pairs since
ep−1
l α(ep−1
. Moreover, the terms in (17) and (18) are identical except for
sign and hence they cancel, so that C = {Ck, ∂k}k≥0 is a complex.
) = α(ep−1
)ep−1
j
j
l
Theorem 4.4. If C is the complex defined above, we define Hk(C) := Ker∂k/Im∂k. Then
Hk(C) =(cid:26) Ures.(g), k = 0
0,
0 < k < p.
Proof. Define a map D : Ck → Ck+1 by the formula
D(eµ ⊗ α(cI)) =
sXa=1
(−1)a−1eµeia ⊗ ci1 · · ·ccia · · · cis
and compute for any monomial eµ ⊗ α(cI) :
D∂(eµ ⊗ α(cI)) = D(
⊗ cj)(eµ ⊗ α(cI)))
1 · · · eµj −1
j
=
=
( ∂
∂ej
D(µjeµ1
nPj=1
nXj=1,j6=i1,··· ,is
nXj=1,j6=i1,··· ,is
nXj=1,j6=i1,··· ,is
µj(cid:1)eµ ⊗ α(cI)
=(cid:0)
sXa=1
nXj=1,j6=i1,··· ,is
(−1)aµjeµ1
D(µjeµ1
+
1 · · · eµj −1
j
· · · eµn
n ⊗ cjα2(cI))
· · · eµn
n ⊗ α(cj)α2(cI))
1 · · · eµj −1
j
· · · eµia +1
ia
· · · eµn
n
⊗ α(cj)α(ci1) · · ·\α(cia) · · · α(cis)
(20)
and ∂D(eµ ⊗ α(cI)) = ∂(
(−1)a−1eµeia ⊗ ci1 · · ·ccia · · · cis)
(−1)a−1∂(eµ1
1 · · · eµia +1
ia
· · · eµn
=
sPa=1
sXa=1
=(cid:0) sXa=1
sXa=1
−
µia + 1(cid:1)eµ ⊗ α(cI)
nXj=1,j6=i1,··· ,is
(−1)a
16
n ⊗ ci1 · · ·ccia · · · cis)
µjeµ1
1 · · · eµj −1
j
· · · eµia +1
ia
· · · eµn
n
⊗ α(cj)α(ci1) · · ·\α(cia) · · · α(cis).
(21)
Clearly the terms (20) and (21) are identical apart from sign so that we have
(D∂ + ∂D)(eµ ⊗ α(cI)) =(cid:0)
nXj=1,j6=i1,··· ,is
µj +
sXa=1
µia + s(cid:1)(eµ ⊗ α(cI)) = (t + s)(eµ ⊗ α(cI)).
Therefore we see that every cycle in Ck(k = 2t + s) is a boundary provided that t + s 6=
0(modp). In particular, if 0 < k < p, then 0 < t + s < p so that Hk(C) = 0. Moreover,
C1 = C0,1 is spanned by the ci and ∂ci = 0 for all i. Therefore H0(C) = C0 = Ures.(g), the
proof of the theorem is complete.
References
[1] Ammar, F., Ejbehi, Z. and Makhlouf, A., Cohomology and Deformations of Hom-
algebras. J. Lie Theory, 2011, 21: 813-836.
[2] Bahturin, Y., Mikhalev, A., Petrogradski, V. M. and Zaicev, M., Infinite-dimensional
Lie superalgebras, Walter de Gruyter, Berlin, New York, 1992.
[3] Benayadi, S., Makhlouf, A., Hom-Lie Algebras with Symmetric Invariant NonDegen-
erate Bilinear Forms. arXiv:1009.4226 (2010).
[4] Dokas, I., Loday, J. L., On restricted Leibniz algebras. Comm. Algebra, 2006, 34:
4467-4478.
[5] Evans, T. J., Fuchs, D., A complex for the cohomology of restricted Lie algebras. J.
fixed point theory appl., 2008, 3: 159-179.
[6] Farnsteiner, R., Note on Frobenius extensions and restricted Lie superalgebras. J.
Pure Appl. Algebra, 1996, 108: 241-256.
[7] Hartwig, J. T., Larsson D. and Silvestrov, S. D., Deformations of Lie algebras using
σ-derivations. J. Algebra, 2006, 295: 314-361.
[8] Hodge, T. L., Lie triple system, restricted Lie triple system and algebraic groups. J.
Algebra, 2001, 244: 533-580.
[9] Larsson, D., Silvestrov, S., Quasi-hom-Lie algebras, central extensions and 2-cocycle-
like identities. J. Algebra, 2005, 288: 321-344.
[10] Larsson, D., Silvestrov, S., Quasi-Lie algebras. Contemp. Math., 2005, 391: 241-248.
[11] Makhlouf, A., Silvestrov, S., Notes on formal deformations of hom-associative and
hom-Lie algebras. Forum Math., 2010, 22(4): 715-739.
17
[12] Makhlouf, A., Silvestrov, S., Hom-algebra structures. J. Gen. Lie Theory Appl., 2008,
2(2), 51-64.
[13] Sheng, Y. H., Representations of hom-Lie algebras. Algebra Representation Theory,
2012, 15: 1081–1098.
[14] Strade, H., Farnsteiner, R., Modular Lie algebras and their representations. New
York: Dekker. 1988.
[15] Strade, H., The classification of the simple modular Lie algebras. Ann. Math., 1991,
133: 577-604.
[16] Yau, D., Hom-Yang-Baxter equation, hom-Lie algebras, and quasi-triangular bialge-
bras. J. Phys. A: Math. Theory, 2009, 42, 165202.
[17] Yau, D., Hom-algebras and homology. J. Lie Theory, 2009, 19: 409-421.
[18] Yau, D., Enveloping algebras of Hom-Lie algebras. Journal of Generalized Lie Theory
and Applications, 2008, 2: 95-108.
18
|
1011.2588 | 1 | 1011 | 2010-11-11T07:47:23 | A $q$-Identity Related to a Comodule | [
"math.RA"
] | In this paper we show that a certain algebra being a comodule algebra over the Taft Hopf algebra of dimension $n^2$ is equivalent to a set of identities related to the $q$-binomial coefficient, when $q$ is a primitive $n^{th}$ root of 1. We then give a direct combinatorial proof of these identities. | math.RA | math |
A q-IDENTITY RELATED TO A COMODULE
A. JEDWAB AND S. MONTGOMERY
Dedicated to Mia Cohen, coauthor and friend, on the occasion of her retirement
1. Introduction
In this paper we show that a certain algebra being a comodule algebra over the
Taft Hopf algebra of dimension n2 is equivalent to a set of identities related to the
q-binomial coefficient, when q is a primitive nth root of 1. We then give a direct
combinatorial proof of these identities. To be consistent with the usual notation for
the Taft algebra, we will write q = ω for our nth root of 1.
Let k be a field of characteristic 0 which contains a primitive nth root of 1, ω.
Consider the algebra A = An(ω) = k[z]/(zn − ω).
It was proposed by Cohen,
Fischman, and the second author [CFM] that A is a right H-comodule for the Taft
Hopf algebra H = Tn2(ω) of dimension n2, for a particular map ρ : A → A ⊗ H.
[CFM, Proposition 2.2(d)] proved that ρ is a comodule map when n ≤ 4. However
the question for general n was left open, since the general case seemed to lead to
some rather complicated identities.
The comodule problem was later solved for arbitrary n in [MS] by indirect means:
it was shown there that A is a module for the Drinfel'd double D(H), giving an action
of the dual (H ∗)cop on A. This action dualizes exactly to the [CFM] coaction of H.
Moreover [MS] show that A is always a Yetter-Drinfeld module for H; this had been
proved in [CFM] for n ≤ 4.
The question was raised as to whether a direct proof of the comodule property for
A via ρ could be given, by determining precisely the identities involved (see [MS,
p. 357]). In Theorem 3.9 we determine exactly the identities needed, using the q-
binomial coefficient with q = ω. In Theorem 4.2 we then give a combinatorial proof
of the identities. This gives an alternative to the methods of [MS].
In Section 5 we also show directly that our algebra A = An(ω) is in the category
H
H YD of Yetter-Drinfel'd modules for H, for any n, using the form of the comodule
map ρ. As a consequence A is always a commutative algebra in the category H
HYD,
answering another question of [CFM].
Finally in Section 6 we discuss in more detail the dual approach of [MS].
Both authors were supported by NSF grant DMS 07-01291.
1
2
A. JEDWAB AND S. MONTGOMERY
2. Preliminaries
We let H denote the Taft Hopf algebra of dimension n2, that is
H = Tn2(ω) = khx, gxn = 0, gn = 1, xg = ωgxi,
where ω is a fixed primitive nth root of 1, with Hopf structure given by
∆(g) = g ⊗ g, ∆(x) = x ⊗ 1 + g ⊗ x
ǫ(g) = 1, ǫ(x) = 0, S(g) = g−1 and S(x) = −g−1x.
We also need some well-known facts about q-binomial coefficients [K]. Recall that
k(cid:19)q
(cid:18)b
:=
(b)!q
(k)!q (b − k)!q
, where (b)!q :=
(q − 1)(q2 − 1) · · · (qb − 1)
(q − 1)b
.
So for k, s ∈ N,
k (cid:19)q
(cid:18)k + s
=
(1 − q) · · · (1 − qs)(1 − qs+1) · · · (1 − qs+k)
(1 − q) · · · (1 − qs)(1 − q) · · · (1 − qk)
=
(1 − qs+1) · · · (1 − qs+k)
(1 − q) · · · (1 − qk)
.
Lemma 2.1. Given x, g ∈ H as above and b ∈ N,
∆(xb) =
gkxb−k ⊗ xk.
b
k(cid:19)ω
Xk=0(cid:18)b
Proof. Since ∆(xb) = (∆(x))b = (x ⊗ 1 + g ⊗ x)b, the lemma follows from the q-
binomial theorem [K, IV.2.2], as follows: in [K], the theorem is stated for (x + y)b,
where yx = qxy. Here we replace x by g ⊗ x, y by x ⊗ 1 and q by ω.
(cid:3)
Corollary 2.2. For any a, b ∈ N,
∆(xbga) =
b
Xk=0
k(cid:19)ω
w−k(b−k)(cid:18)b
xb−kgk+a ⊗ xkga.
Proof.
∆(xbga) = ∆(xb)∆(ga) =
gkxb−kga ⊗ xkga
b
k(cid:19)ω
Xk=0(cid:18)b
k(cid:19)ω
w−k(b−k)(cid:18)b
=
b
Xk=0
xb−kgk+a ⊗ xkga.
(cid:3)
3. The comodule algebra for H
As noted in the introduction, [CFM] proposed that A will be an H-comodule. We
let u denote the coset z + I, where I = (zn − ω), and thus {1, u, u2, . . . , un−1} will be
a basis for A.
For our given root of unity ω, we define
(3.1)
ai := (ω − 1)iω
i(i+1)
2
.
A q-IDENTITY RELATED TO A COMODULE
3
The explicit coaction ρ : A → A ⊗ H is now defined by
(3.2)
We must prove that
ρ(u) =
n−1
Xi=0
aixig−(i+1) ⊗ ui+1.
(3.3)
(id ⊗ ρ)ρ = (∆ ⊗ id)ρ.
Now [CFM] showed that ρ(u)n = ω1, and thus ρ is a homomorphism since un = ω1.
Since ∆ is also a homomorphism and the powers of u are a basis for A, it will suffice
to check that Equation (3.3) holds when applied to the element u.
Evaluating Equation (3.3) on u, we obtain the new equation:
n−1
n−1
(3.4)
asxsg−(s+1) ⊗ ρ(u)s+1 =
am∆(xmg−(m+1)) ⊗ um+1,
where by Corollary 2.2, the right hand side is
Xm=0
Xs=0
am m
Xk=0
n−1
Xm=0
k(cid:19)ω
ω−k(m−k)(cid:18)m
xm−kgk−(m+1) ⊗ xkg−(m+1)! ⊗ um+1.
In order to compute the left hand side of (3.4), we need to find an explicit formula
for ρ(u)s for any 1 ≤ s ≤ n. We start with an auxiliary lemma:
Lemma 3.5. (i) Given r, s ∈ N, aras = ar+sω−rs and more generally
Πt
i=1ari = a(Pt
i=1 ri)ω− Pj<i rirj
(ii) For all 1 ≤ i ≤ n − 1,
ωj−i)ai + ai−1 = ωi+1ai−1.
i
(
Xj=0
Proposition 3.6. For any 1 ≤ s ≤ n,
ρ(u)s =
n−1
Xk=0
Proof. Let ρ(u)j =
ak
Xij=0
n−1
{0≤i1,...,is≤k Ps
j=1 ij =k}
X
ωPs
j=2 ij (j−1)
xkg−(k+s) ⊗ uk+s.
aij xij g−(ij+1) ⊗ uij +1 denote the j-th copy of ρ(u) in ρ(u)s. As
we multiply one term from each of the s factors ρ(u)j in ρ(u)s, we obtain a sum of
terms of the form
(3.7) ai1 · · · aij · · · aisxi1g−(i1+1) · · · xij g−(ij+1) · · · xisg−(is+1) ⊗ ui1+1 · · · uij+1 · · · uis+1.
j=1 ij. Using Lemma 3.5 (i) and the fact that grxs = ω−rsxsgr, (3.7)
Let k = Ps
becomes
akω−(Pt<r ir it)Πs
j=2ωPj−1
l=1 ij(il+1)xkg−(k+s) ⊗ uk+s.
4
A. JEDWAB AND S. MONTGOMERY
Simplifying, we have
(3.8)
ω−(Pt<r ir it)Πs
j=2ωPj−1
l=1 ij(il+1) = ω−(Pt<r ir it)ωPs
j=2 Pj−1
l=1 (ij il+ij)
=
=
ω−(Pt<r ir it)ωPs
ωPs
j=2 ij (j−1),
j=2(Pj−1
l=1 ij il)ωPs
j=2 ij (j−1)
since the first two powers of ω which appear have opposite exponents.
Finally, since such a term arises whenever i1 + · · · + is = k, by ordering the terms
according to powers of x we have that
ρ(u)s =
n−1
Xk=0
ak
{0≤i1,...,is≤k Ps
j=1 ij=k}
X
ωPs
j=2 ij(j−1)
xkg−(k+s) ⊗ uk+s.
(cid:3)
Using Proposition 3.6 with s + 1 instead of s, we have all the components of our
desired equation (3.4). Substituting them in (3.4), we may compare the coefficients
on both sides:
n−1
n−1
Xs=0
Xk=0
asak
{0≤i1,...,is+1≤k Ps+1
j=1 ij=k}
ωPs+1
j=2 ij(j−1)
l(cid:19)ω
amωl(m−l)(cid:18)m
X
=
n−1
m
Xm=0
Xl=0
xsg−(s+1) ⊗ xkg−(k+s+1) ⊗ uk+s+1
xm−lgl−(m+1) ⊗ xlg−(m+1) ⊗ um+1
By linear independence, the coefficients of each term on both sides should agree.
Thus we have:
Theorem 3.9. Fix a primitive nth root of unity ω in k, and let A = An(ω) and
H = Tn2(ω) be as above. Then A is a right H-comodule algebra via the coaction ρ in
Equation (3.2) ⇐⇒ for all pairs of natural numbers 0 ≤ k, s ≤ n − 1,
{0≤i1,...,is+1≤k Ps+1
j=1 ij =k}
X
ωPs+1
j=2 ij(j−1) =
k (cid:19)ω
(cid:18)k + s
0
if k + s < n
if k + s ≥ n.
4. A proof of the identities
In this section we give a direct combinatorial proof of the identities in Theorem
3.9. We thank Jason Fulman for pointing it out to us.
1
(1 − z)(1 − zω) · · · (1 − zωs)
as a formal power series
We consider the expansion of
in the ring k[[z]]. Write
1
(1 − z)(1 − zω) · · · (1 − zωs)
βkzk.
=Xk≥0
A q-IDENTITY RELATED TO A COMODULE
5
Lemma 4.1. For each k ≥ 0,
βk =
Proof. We know that
{0≤i1,...,is+1≤k Ps+1
j=1 ij =k}
ωPs+1
j=2 ij(j−1).
X
1
s+1
Xk≥0
Xl=1
βkzk = Πs+1
l=1 (
1 − zωl−1 ) = Πs+1
l=1 Xil≥0
(zωl−1)il! .
Whenever
il = k, the last product gives a term zkωPs+1
l=2 il(l−1), where the sum
in the exponent starts at l = 2 because l − 1 = 0 for l = 1. Thus
βk =
{0≤i1,...,is+1≤k Ps+1
l=1 il=k}
X
ωPs+1
l=2 il(l−1)
and thus the left hand side of the identity in Theorem 3.9 is the coefficient of zk in
the power series.
(cid:3)
Theorem 4.2. The identities in Theorem 3.9 hold, for all n > 1, any given primitive
nth root of unity ω in k, and all pairs of natural numbers 0 ≤ k, s ≤ n − 1.
Proof. We evaluate the coefficient βk in a different way, using Theorem 349 in [HW]
which states that given ω ∈ k,
1
(1 − zω)(1 − zω2) · · · (1 − zωj)
Replacing zω by z we get
= 1 + zω
1 − ωj
1 − ω
+ z2ω2 (1 − ωj)(1 − ωj+1)
(1 − ω)(1 − ω2)
+ · · · .
1
(1 − z)(1 − zω) · · · (1 − zωj−1)
and if we choose j = s + 1 then
= 1 + z
1 − ωj
1 − ω
+ z2 (1 − ωj)(1 − ωj+1)
(1 − ω)(1 − ω2)
+ · · ·
1
(1 − z)(1 − zω) · · · (1 − zωs)
= 1 + z
1 − ωs+1
1 − ω
+ z2 (1 − ωs+1)(1 − ωs+2)
(1 − ω)(1 − ω2)
+ · · · .
In particular, the coefficient βk of zk turns out to be
(1 − ωs+1) · · · (1 − ωs+k)
(1 − ω) · · · (1 − ωk)
Since βk is unique, both forms must agree and
{0≤i1,...,is+1≤k Ps+1
j=1 ij=k}
X
ωPs+1
.
k (cid:19)ω
=(cid:18)k + s
k (cid:19)ω
j=2 ij(j−1) =(cid:18)k + s
.
When k + s ≥ n with 0 ≤ k ≤ n − 1, one of the factors in the numerator
(1 − ωs+1) · · · (1 − ωs+k) is 1 − ωn = 0 while the denominator (1 − ω) · · · (1 − ωk) 6= 0,
(cid:3)
making (cid:0)k+s
k (cid:1)ω = 0 as required.
6
A. JEDWAB AND S. MONTGOMERY
Corollary 4.3. The algebra A is an H-comodule algebra, via the coaction in Equation
(3.2).
5. YD-module algebras and H-commutativity
In this section we consider the (left, left) Yetter-Drinfel'd category H
HYD. Recall
H YD if it is both a left H-module, a left H-comodule (via ρ),
that a module M is in H
and
(5.1)
h · ρ(m) =X ρ(h1 · m)(h2 ⊗ 1).
[CFM, Prop 2.2(e)] prove that our algebra A = An is in H
HYD for H = Tn2(ω), for
all n ≤ 4. Here we show this for all n. We use a result from [CFM] which holds for
any H and any A:
Lemma 5.2. [CFM, Lemma 2.10] Let A be a left H-module and a left H-comodule.
(a) Let M be an H-submodule of A. If the Yetter-Drinfel'd condition is satisfied
for all m ∈ M and all algebra generators of H (from some chosen generating set),
then it is satisfied for all m ∈ M and all H ∈ H.
(b) If A is also an H-module algebra and an H-module coalgebra, and if the Yetter-
Drinfeld condition holds for all h ∈ H and all algebra generators of A (from some
generating set), then A ∈ H
HYD.
Proposition 5.3. The algebra A = An(ω) is in H
Tn2(ω), for all n.
HYD for the Taft algebra H =
Proof. By Corollary 4.3, A is a left H-comodule, and so Lemma 5.2 will apply. We
use that A is generated as an algebra by the H-submodule M = k{1, u} and H is
generated as an algebra by the set {g, x}. Thus A will be in H
HYD provided we show
the Yetter-Drinfeld condition (5.1) when a = u and either h = g or h = x.
First assume h = g. Then, using ρ(u) as in (3.2),
n−1
aigxig−(i+1) ⊗ g · ui+1
aiω−ixig−i ⊗ wi+1ui+1
ωaixig−i ⊗ ui+1.
=
=
n−1
g · ρ(u) =
Xi=0
Xi=0
Xi=0
ρ(g · u)(g ⊗ 1) = ω n−1
Xi=0
n−1
Thus the Yetter-Drinfel'd condition holds for g and u.
=
ωaixig−i ⊗ ui+1.
n−1
Xi=0
On the other hand, since g · u = ωu
aixig−(i+1) ⊗ ui+1! (g ⊗ 1)
A q-IDENTITY RELATED TO A COMODULE
7
Now assume that h = x. First, since ∆(x) = x ⊗ 1 + g ⊗ x and x · u = 1, it is easy
j=0 ωj)ui. Thus in (5.1),
n−1
to see that x · ui+1 = (Pi
Xi=0
Xi=0
x · ρ(u) =
n−1
=
= 1 ⊗ 1 +
On the other hand,
n−1
Xi=1 (
i
Xj=0
n−1
n−1
Xi=0
Xi=0
aixxig−(i+1) ⊗ ui+1 +
aigxig−(i+1) ⊗ x · ui+1
aixi+1g−(i+1) ⊗ ui+1 +
i
ωj)ui
ω−iaixig−i ⊗ (
Xj=0
ωj−i)ai + ai−1! xig−i ⊗ ui.
X ρ(x1 · m)(x2 ⊗ 1) = ρ(x · u)(1 ⊗ 1) + ρ(g · u)(x ⊗ 1)
= ρ(1)(1 ⊗ 1) + ω n−1
Xi=0
aixig−(i+1) ⊗ ui+1! (x ⊗ 1)
= 1 ⊗ 1 + ω
ωi+1aixi+1g−(i+1) ⊗ ui+1
n−1
n−1
Xi=0
Xi=1
= 1 ⊗ 1 +
ωi+1ai−1xig−i ⊗ ui,
where in both cases the term corresponding to i = n vanishes because xn = 0. Thus
for A to be a Yetter-Drinfel'd module algebra, we need that
ωj−i)ai + ai−1 = ωi+1ai−1, for all 1 ≤ i ≤ n − 1.
i
(
Xj=0
However this holds by Lemma 3.5 (ii).
(cid:3)
[CFM] also study when A is commutative as an algebra in the category H
Recall that for any braided monoidal category C, with braiding τ : V ⊗W → W ⊗V
HYD.
for V, W ∈ C, an algebra A in C is commutative in C if for all a, b ∈ A,
(5.4)
mA(a ⊗ b) = mA ◦ τ (a ⊗ b).
Several authors have considered this generalized commutativity. In particular Co-
hen and Westreich considered the case when C is the module category of a quasi-
triangular Hopf algebra in [CW].
In our situation H
HYD has the structure of a braided monoidal category, as follows:
for two modules M, N ∈H
H YD, the braiding is given as follows [Y]:
τ : M ⊗ N → N ⊗ M via m ⊗ n 7→ ρ(m)(n ⊗ 1) =X(m−1 · n) ⊗ m0.
8
A. JEDWAB AND S. MONTGOMERY
Thus an algebra A in H
HYD is commutative in H
HYD if
(5.5)
ab =X(a−1 · b)a0.
Corollary 5.6. For the given algebra An = k[u] and H = Tn2(ω), A is commutative
in H
HYD, for any n.
Proof. It is shown in [CFM, Prop 2.2(e)] that if An is in H
in H
algebra; again it suffices to check on generators of A and of H.
H YD, then it is commutative
In fact their argument uses only that An is an H-module H-comodule
(cid:3)
HYD.
6. The dual action
In this section, for the sake of completeness, we sketch the approach of [MS] for
the action of H ∗ on A. As noted in the introduction, it is shown there that A is a
D(H)-module algebra (and thus a Yetter-Drinfeld module algebra).
The Taft Hopf algebras H = Tn2(ω) are known to be self-dual; thus we may write
(6.1)
H ∗ = khG, X Gn = ε, X n = 0, XG = ωGXi,
where ∆(G) = G ⊗ G, ∆(X) = X ⊗ ε + G ⊗ X, hG, 1i = 1, and hX, 1i = εH ∗(X) = 0.
The dual pairing between H and H ∗ is determined by
(6.2)
hG, gi = ω−1, hG, xi = 0, hX, gi = 0, and hX, xi = 1.
Lemma 6.3. As an algebra, D(H) is generated by {x, g, X, G}. The relations among
these generators, in addition to the relations in H and H ∗, are as follows:
gG = Gg, xG = ω−1Gx, Xg = ω−1gX, and xX − Xx = G − g.
One may check that (H ∗)cop = khG−1, XG−1i ⊂ D(H), and that these generators
give the usual relations in D(H). The generators given in Lemma 6.3 are used since
X and x behave similarly when acting as skew derivations. [MS] then use properties
of higher skew derivations and the relations in Lemma 6.3 to prove:
Theorem 6.4. [MS, Theorem 4.5] Let H = Tn2(ω) be the Taft Hopf algebra and H ∗
its dual as above. Then A = An becomes a D(H)-module algebra via the following:
(a) g · u = ωu and G · u = ω−1u, and
(b) x · u = 1 and X · u = (ω−1 − 1)u2.
To see that A is an algebra in the category H
HYD of left, left Yetter-Drinfeld mod-
ules, one may use a theorem of Majid [Mj] that D(H)-modules maybe be identified
with H
HYD-modules. The only difficulty remains in showing that dualizing the left
(H ∗)cop-action in Theorem 6.4 to a left H-comodule action gives the desired coaction.
Theorem 6.5. [MS, Theorem 5.7] Let H = Tn2(ω), A = An, and the H-action on
A be as described in Theorem 6.4. Then there is a unique left H-comodule algebra
structure ρ on A such that A is in H
HYD, given by
ρ(u) =
n−1
Xm=0
amxmg−(m+1) ⊗ um+1,
A q-IDENTITY RELATED TO A COMODULE
9
where the coefficient am is given by
am = ((1 − ω−1)ω)mω
m(m+1)
2 = (ω − 1)mω
m(m+1)
2
.
This coefficient am is exactly our coefficient in Definition (3.1), and so the coaction
in (6.5) is exactly our coaction in Equation (3.2). Thus [MS, Theorem 5.7] gives an
alternate proof of Corollary 4.3 and Proposition 5.3.
The authors wish to thank Jason Fulman for suggesting the proof of Theorem 4.2.
ACKNOWLEDGMENT
References
[CFM] M. Cohen, D. Fischman, and S. Montgomery, On Yetter-Drinfeld categories and H-
commutativity, Comm. in Algebra 27 (1999), 1321 - 1345.
[CW] M. Cohen and S. Westreich, From supersymmetry to quantum commutativity, J. Algebra 168
(1994), 1 - 27.
[HW] G. H. Hardy and E. M. Wright,An introduction to the theory of numbers, Fifth edition, Oxford
University Press, 1979.
[JM] A. Jedwab and S. Montgomery, Representations of some Hopf algebras associated to the
symmetric group Sn, Algebras and Representation Theory 12 (2009), 1 - 17.
[K] C. Kassell, Quantum Groups GTM, 155, Springer-Verlag,1995.
[Mj] S. Majid, Doubles of quasitriangular Hopf algebras, Comm. Algebra 19 (1991), 3061 - 3073.
[M] S. Montgomery, Hopf Algebras and Their Actions on Rings, CBMS Lecture Notes Vol. 82,
American Math Society, Providence 1993.
[MS] S. Montgomery and H.-J. Schneider, Skew-derivations of finite-dimensional algebras and ac-
tions of the double of the Taft Hopf algebra, Tsukuba J. of Math 25 (2001), 337 - 358.
[Y] D. N. Yetter, Quantum groups and representations of monoidal categories, Math. Proc. Cam-
bridge Phil. Soc 108 (1990), 261 - 290.
University of Southern California, Los Angeles, CA90089-1113
E-mail address: [email protected]
University of Southern California, Los Angeles, CA 90089-1113
E-mail address: [email protected]
|
1601.00554 | 1 | 1601 | 2016-01-04T16:20:21 | Asymptotically optimal $k$-step nilpotency of quadratic algebras and the Fibonacci numbers | [
"math.RA"
] | It follows from the Golod--Shafarevich theorem that if R is an associative algebra given by n generators and $d<\frac{n^2}{4}\cos^{-2}(\frac{\pi}{k+1})$ quadratic relations, then R is not k-step nilpotent. We show that the above estimate is asymptotically optimal, and establish number of related results. For example, we show that for any k this estimate is attained for ifinitely many n. | math.RA | math |
Asymptotically optimal k-step nilpotency of quadratic algebras
and the Fibonacci numbers
Natalia Iyudu and Stanislav Shkarin
Abstract
It follows from the Golod -- Shafarevich theorem that if k ∈ N and R is an associative algebra
given by n generators and d < n2
k+1 ) quadratic relations, then R is not k-step nilpotent.
We show that the above estimate is asymptotically optimal. Namely, for every k ∈ N, there is a
sequence of algebras Rn given by n generators and dn quadratic relations such that Rn is k-step
nilpotent and lim
4 cos−2( π
dn
n2 = 1
4 cos−2( π
k+1 ).
n→∞
MSC: 17A45, 16A22
Keywords: Quadratic algebras, Golod -- Shafarevich theorem, Anick's conjecture
1
Introduction
Throughout this paper K is an arbitrary field, Z+ is the set of non-negative integers and N is the
set of positive integers. For a set X, KhXi stands for the free associative algebra over K generated
by X. We deal with quadratic algebras, that is, algebras R given as KhXi/I, where I is the ideal
in KhXi generated by a collection of homogeneous elements (called relations) of degree 2.
Algebras of this class, their growth, their Hilbert series and nil/nilpotency properties have been
extensively studied, see [11, 12, 13] and references therein. One of the most challenging questions in
the area (see [12, 15]) is the Kurosh problem of whether there is an infinite dimensional nil algebra
in this class. A version of this question dealing with algebras of finite Gelfand -- Kirillov dimension
was solved in [8]. The Golod -- Shafarevich type lower estimates for the dimensions of the graded
components of an algebra play a crucial role in the study of quadratic algebras. These estimates
have many other applications, for instance, to p-groups and class field theory [5, 16].
Recall that a K-algebra R defined by the set X of generators and a set of homogeneous relations
inherits the degree grading from the free algebra KhXi. If X is finite, one can consider the Hilbert
series of R:
HR(t) =
(dim K Rq) tq,
∞
X
q=0
where Rq is the qth homogeneous component of R. The original Golod -- Shafarevich theorem pro-
vides a lower estimate for the coefficients of HR. In the case of quadratic algebras the theorem
reads as follows [5, 11]. For two power series a(t) and b(t) with real coefficients we write a(t) > b(t)
if aj > bj for any j ∈ Z+, while a(t) stands for the power series obtained from a(t) by replacing
by zeros all coefficients starting from the first non-positive one.
Theorem GS. Let, n ∈ N, 0 6 d 6 n2 and R be a quadratic K-algebra with n generators and d
relations. Then HR(t) > (1 − nt + dt2)−1.
In particular, Theorem GS provides a lower estimate on the order of nilpotency of R.
Definition 1.1. A graded algebra R is called k-step nilpotent if Rk = {0}.
1
Analysing the series K(t) = (1 − nt + dt2)−1 in a standard way, one can easily see that it is a
polynomial of degree < k if and only if
d
n2
> ϕk, where ϕk =
1
4
cos−2(cid:16) π
k + 1(cid:17).
For the sake of convenience, we outline the argument. If (1− nt + dt2)−1 =
(1.1)
∞
Pm=0
cmtm (the Taylor
series expansion), then K(t) is not a polynomial of degree < k precisely when cm > 0 for 0 6 m 6 k.
Next, if x2 − nx + d = (x − a)(x − b) (a and b are complex numbers in general), then an easy
computation yields that cm = (m + 1)(n/2)m if a = b and cm = am+1−bm+1
otherwise for m ∈ Z+.
It follows that cm > 0 for all m ∈ Z+ if a and b are real, which happens precisely when d 6 n2
4 . If
n2 > d > n2
= dm/2 sin(m+1)α
for m ∈ Z+. Clearly cm for 0 6 m 6 k are positive precisely when (k + 1)α < π. After plugging in
α = arccos n√d
4 , then a, b = √de±iα, where α = arccos n√d
. Hence cm = am+1−bm+1
, (1.1) follows.
a−b
a−b
sin α
Formula (1.1) together with Theorem GS and the obvious fact that the sequence {ϕk} decreases
4 implies the following corollary, which can be found in [11].
and converges to 1
Corollary GS. If R is a quadratic K-algebra given by n generators and d < ϕkn2 relations, then
dim Rk > 0, where ϕk is defined in (1.1). That is, R is not k-step nilpotent.
In particular, if
d 6 n2
4 , then dim Rk > 0 for every k ∈ N and therefore R is infinite dimensional.
Asymptotic optimality of the last statement in Corollary GS was proved by Wisliceny [14].
4
dn
Theorem W. For every n ∈ N, there exists a quadratic K-algebra R given by n generators and dn
relations such that R is finite dimensional and lim
n→∞
n2 = 1
4 .
More specifically, Wisliceny has constructed a quadratic algebra given by n generators and
(cid:6) n2+2n
(cid:7) semigroup relations (that is, every relation is either a degree 2 monomial or a difference of
two degree 2 monomials), which is finite dimensional. Note that here and everywhere below ⌊t⌋ is
the largest integer 6 t, while ⌈t⌉ is the smallest integer > t, where t is a real number. The authors
[7] have improved the last result by showing that the minimal number of semigroup quadratic
relations needed for finite dimensionality of an algebra with n generators is exactly (cid:6) n2+n
4 (cid:7). The
number (cid:6) n2+1
4 (cid:7) remains a conjectural answer to the same question in the class of general quadratic
(not necessarily semigroup) algebras.
1.1 Results
Note that if R is k-step nilpotent, then Rm = {0} for m > k and therefore R is finite dimensional
provided X < ∞, where X is the set of generators of R. Thus R is k-step nilpotent if and only if
HR is a polynomial of degree < k.
In this article we show that the first statement in Corollary GS is asymptotically optimal for
every k > 2. In order to formulate the exact statement, we shall introduce the following numbers.
For n ∈ N and k > 2 let
dn,k =
(1.2)
max
min
(a1 + . . . + aj)(aj + . . . + ak−1),
n=a1+...+ak−1
16j6k−1
where aj are assumed to be non-negative integers. It turns out that the integers dn,k are not too
far from ϕkn2.
Lemma 1.2. For each n, k ∈ N with k > 2,
(1.3)
ϕkn2 6 dn,k 6 ϕkn2 + (1+ϕk)n
2
+ 1
4 .
In particular,
lim
n→∞
dn,k
ϕkn2 = 1 for each k > 2.
2
We have defined the numbers dn,k since they feature in the following theorem.
Theorem 1.3. Let k > 2. Then for every n ∈ N, there exists a quadratic K-algebra R given by n
generators and dn,k relations such that R is k-step nilpotent.
Corollary GS, Theorem 1.3 and Lemma 1.2 imply that the first statement in Corollary GS is
asymptotically optimal. Note that Anick [1, 2] conjectured that for any n ∈ N and 0 6 d 6 n2, there
is a quadratic K-algebra R with n generators and d relations such that HR(t) = (1 − nt + dt2)−1.
The problem whether this conjecture is true remains open. Theorem 1.3 can be considered as an
affirmative solution of its natural asymptotic version. It is also worth noting that for k = 2, the
statement of Theorem 1.3 is trivial, while the case k = 3 was done by Anick [1]. It is also worth
mentioning that the asymptotic optimality of the first statement in Corollary GS for k = 4 and for
k = 5 in the case K = ∞ was earlier obtained by the authors [6] building upon the ideas set in [3]
and using a completely different approach. We refer to [10] for a result on asymptotic optimality
of Theorem GS in a completely different sense.
and ϕ5 = 1
Curiously enough, for some pairs (n, k) the estimate provided by Theorem 1.3 hits the mark.
We illustrate this observation by the following result dealing with the cases k = 4 and k = 5. Note
√5
that ϕ4 = 3−
3 . Recall that Fibonacci numbers are the members of the recurrent
2
sequence defined by F0 = F1 = 1 and Fn = Fn−1 + Fn−2 for n > 2.
Theorem 1.4. The equality dn,4 = (cid:6) 3−√5
equality dn,5 = (cid:6) n2
2 n2(cid:7) holds if and only if n is a Fibonacci number. The
3 (cid:7) holds if and only if n ∈ {1, 2} or n is divisible by 6.
Note that Theorem 1.4, Theorem 1.3 and Corollary GS imply that if k = 4 and n is a Fibonacci
number or if k = 5 and 6 divides n, then the minimal number of quadratic relations needed for the
finite dimensionality of an algebra with n generators is exactly ⌈ϕkn2⌉. The proof of Theorem 1.3
is based upon the following general result. We start by introducing some notation.
Definition 1.5. Let X be the union of pairwise disjoint sets A1, . . . , Ak and
M = M (A1, . . . , Ak) = [
Aq × Aj ⊆ X × X.
16j6q6n
(1.4)
We introduce the following partial ordering on M , generated by the partition {A1, . . . , Ak}. Namely,
for distinct elements (a, b) and (c, d) of M , we write (a, b) ≺ (c, d) if (a, b) ∈ Al × Aj and (c, d) ∈
Am × Ar with m > r > l > j.
Definition 1.6. For a homogeneous degree 2 polynomial g in the free algebra KhXi, the (uniquely
determined) finite subset S of X × X such that g = P(x,y)∈S
cx,yxy with cx,y ∈ K \ {0} is called the
support of g and is denoted S = supp (g).
The next result is one of the main tools in the proof of Theorem 1.3.
Theorem 1.7. Let k ∈ N, {A1, . . . , Ak} be a partition of a set X and M be the set defined in
(1.4). Assume also that {fα}α∈Λ is a family of homogeneous degree 2 elements of the free algebra
KhXi such that Sα∈Λ
supp (fα) = M and each supp (fα) is a chain in M with respect to the partial
ordering ≺ on M , generated by the partition {A1, . . . , Ak} as in Definition 1.5. Then the algebra
R = KhXi/I with I = Id{fα : α ∈ Λ} is (k + 1)-step nilpotent.
We conclude the introduction by providing a specific example of an application of Theorem 1.7.
3
Example 1.8. Let X = {a, b, c, p, q, x, y, z} be an 8-element set partitioned into 3 subsets A1 =
{a, b, c}, A2 = {p, q} and A3 = {x, y, z}. Let M and the partial ordering ≺ on M be as in
Definition 1.5. Consider the following 25 quadratic relations:
f1 = xc,
f6 = yc,
f10 = zc,
f14 = xb,
f18 = yb,
f22 = zb,
f2 = xa,
f7 = ya,
f11 = za,
f15 = xq + ac,
f19 = yq + bc,
f23 = zq + cc,
f3 = xp + ab,
f8 = yp + bb,
f12 = zp + cb,
f16 = xz + pc,
f20 = xy + pb,
f24 = xx + pa,
f5 = pq,
f4 = yz + qc,
f9 = yy + qb,
f13 = yx + qa,
f17 = zz + qq + ca,
f21 = zy + qp + ba,
f25 = zx + pp + aa.
It is straightforward to verify that the support of each fj is a chain in (M,≺) and that the union
of supp (fj) for 1 6 j 6 25 is M . Theorem 1.7 ensures that the algebra given by the 8-element
generator set X and the relations fj with 1 6 j 6 25 is 4-step nilpotent. Incidentally, 25 = (cid:6)ϕ4·82(cid:7),
which means (see Corollary GS) that a quadratic algebra given by 8 generators and 6 24 relations
is never 4-step nilpotent.
2 Combinatorial lemmas
Theorem 1.7 allows us to construct k-step nilpotent quadratic algebras with few relations. In order
to do this, we need an estimate on the number of relations in an algebra featuring in Theorem 1.7.
Recall that the width w(X, <) of a partially ordered set (X, <) is the supremum of the cardinalities
of antichains in X.
Lemma 2.1. Let k ∈ N, {A1, . . . , Ak} be a partition of a finite set X and M ⊆ X 2 be the set
defined in (1.4) with the partial ordering ≺ introduced in Definition 1.5. For 1 6 q 6 k, let
Bq = Sj>q>m
Aj × Am. Then w(M,≺) = max{B1, . . . ,Bk}.
Proof. It is a straightforward exercise to verify that each Bq is an antichain in (M,≺) and that
every antichain is contained in at least one of the sets Bq.
We also need the following observation.
Lemma 2.2. Let k > 2 and α0, α1, . . . , αk−1 > 0 be defined by the fromulae α0 = 0, α1 = ϕk and
αj = ϕk
1−αj−1
for 2 6 j 6 k − 1. Then
0 = α0 < α1 < . . . < αk−1 = 1,
αj(1 − αj−1) = ϕk for 1 6 j 6 k − 1
and max
(2.1)
(2.2)
(2.3)
(αj − αj−1) = ϕk (attained for j = 1 and for j = k − 1).
16j6k−1
Proof. Obviously, (2.2) is a direct consequence of the definition of αj. Next, (2.3) follows easily
from (2.1). Indeed, assuming that (2.1) holds, we have αk−1 = 1, which implies αk−2 = 1 − ϕk.
Since αj − αj−1 = ϕk
1−αj−1 − αj−1 and 0 6 αj−1 6 1 − ϕk for 1 6 j 6 k − 1, (2.3) follows from the
elementary fact that the function ϕk
1−x − x on the interval [0, 1 − ϕk] attains its maximal value at
the end-points.
1−x and for
t = ft ◦ . . . ◦ ft m times for m ∈ N.
Thus it remains to verify (2.1). For 0 < t 6 1 consider the rational function ft(x) = t
be the mth iterate of ft: f [0]
m ∈ Z+ let f [m]
t
t (x) = x and f [m]
We start with an elementary observation
4
if 0 6 t 6 1
and converges to the fixed point wt = 1−√1−4t
4 , then the sequence {f [m]
t
(0)}m∈Z+ is strictly increasing
2
∈ (cid:2)0, 1
2(cid:3) of ft.
(2.4)
For instance, to justify (2.4), one can use induction with respect to m to prove the chain of
inequalities 0 6 f [m]
(0) < f [m+1]
(0) < wt.
t
t
Next, it is easy to verify that if 1
4 < t 6 1, then ft(x) > x for x ∈ [0, 1). Hence,
f [m+1]
t
(0) > f [m]
t
(0) provided 0 6 f [m]
t
(0) < 1.
(2.5)
For each m ∈ Z+, we consider the rational function hm(t) = f [m]
observe that (2.3) follows from the claim
t
(0) of the variable t. Now we
for every m ∈ N, ϕm+1 is the smallest solution of the equation hm(t) = 1 on (cid:0) 1
(2.6)
Indeed, assume that (2.6) holds. By (2.4), 0 < hm(t) < 1
4(cid:3). Since
the sequence {ϕm} is decreasing, hj(t) < 1 whenever j 6 m and 0 6 t < ϕm+1. Using (2.6) with
m = k − 1 and (2.5), we now have
4 , 1(cid:3).
2 for every m ∈ N and t ∈ (cid:0)0, 1
0 = f [0]
ϕk (0) < f [1]
ϕk(0) < . . . < f [k−1]
ϕk
(0) = hk−1(ϕk) = 1.
On the other hand, by definition of αj, αj = f [j]
ϕk (0) for 0 6 j 6 k − 1 and (2.3) follows.
t
Thus it remains to prove (2.6). Using the obvious recurrent relation hj+1(t) =
with the initial data h0 = 0, one can use the induction with respect to m to verify that
1−hj(t) together
hm(t) = t
am−am
am+1−am+1
Hence for t ∈ (cid:2) 1
4 , 1(cid:3),
for m ∈ Z+ and t ∈ (cid:2) 1
4 , 1(cid:3), where a = a(t) = 1+i√4t−1
2
.
hm(t) = 1 ⇐⇒ (a/a)m = (a − t)/(a − t) ⇐⇒ eimα(t) = eiβ(t),
(2.7)
where
α(t) = 2 arccos 1
2√t
and β(t) = 2π − 2 arccos(cid:0) 1
2t − 1(cid:1)
are the arguments of the unimodular complex numbers a/a and (a − t)/(a − t). The case m = 1
is trivial. Assuming that m > 2 and using (2.7), we see that the smallest t ∈ (cid:2) 1
2(cid:3) satisfying
hm(t) = 1 must satisfy mα(t) = β(t). Since the function mα(t) − β(t) on the interval (cid:2) 1
4 , 1
2(cid:3)
is strictly increasing (look at the derivative) and has values of opposite signs at the ends, there
is exactly one tm ∈ (cid:2) 1
2(cid:3) satisfying mα(tm) = β(tm). Then tm is the smallest solution of the
4 , 1
equation hm(t) = t on the interval (cid:2) 1
2(cid:3), (2.6) will follow if we show that
mα(ϕm+1) = β(ϕm+1). This is indeed true: plugging in ϕm+1 =
4 , 1(cid:3). Since ϕm+1 ∈ (cid:2) 1
4 cos2(π/(m+2)) , we have
4 , 1
4 , 1
1
mα(ϕm+1) = 2m arccos(cid:0)cos(cid:0) π
β(ϕm+1) = 2π − 2 arccos(cid:0)2 cos2(cid:0) π
m+2(cid:1)(cid:1) = 2πm
m+2 ;
m+2(cid:1) − 1(cid:1) = 2π − 2 arccos(cid:0)cos(cid:0) 2π
m+2(cid:1)(cid:1) = 2π − 4π
m+2 = 2πm
m+2 .
Hence mα(ϕm+1) = β(ϕm+1), which completes the proof.
3 Proof of Theorem 1.7
For k ∈ N, we denote Nk = {1, 2, . . . , k}. Assume the contrary. Then the set Ω of j = (j1, . . . , jk+1) ∈
Nk+1
such that there are x1 ∈ Aj1, . . . , xk+1 ∈ Ajk+1 for which x1 . . . xk+1 /∈ I is non-empty. We
endow Nk+1
k with the lexicographical ordering < counting from the right-hand side. That is, j < m
k
5
if and only if there is l ∈ Nk+1 such that jl < ml and jr = mr for r > l. Since < is a total ordering
on the finite set Nk+1
is non-empty, Ω has a unique element j minimal with respect
to <. Since j ∈ Ω, there are x1 ∈ Aj1, . . . , xk+1 ∈ Ajk+1 for which x1 . . . xk+1 /∈ I.
degree k + 1 such that
Now we shall construct inductively m1, . . . , mk+1 ∈ Nk and monomials u1, . . . , uk+1 in KhXi of
k
and Ω ⊆ Nk+1
k
ml > ml−1 if l > 2;
ul /∈ I;
ul = vlwlxl+1xl+2 . . . xk+1, where wl ∈ Aml and vl is a monomial of degree l − 1.
(3.1)
(3.2)
(3.3)
We start by setting u1 = x1 . . . xk+1 and m1 = j1 and observing that (3.1 -- 3.3) with l = 1 are
satisfied. Assume now that 2 6 l 6 k + 1 and that m1, . . . , ml−1 and u1, . . . , ul−1 satisfying the
desired conditions are already constructed.
If ml−1 < jl, then we set ml = jl, wl = xl, ul = ul−1 and vl = vl−1wl−1. Using the induction
hypothesis, we see that (3.1 -- 3.3) are satisfied.
In
this case wl−1xl ∈ M and therefore there is α ∈ Λ such that (wl−1, xl) ∈ supp (fα). Let S =
supp (fα) \ {(wl−1, xl)}. Since fα ∈ I,
It remains to consider the case ml−1 > jl.
wl−1xl = X
(a,b)∈S
ca,bab (modI) with ca,b ∈ K.
Using (3.3) for l − 1 and the above display, we get
ul−1 = X
(a,b)∈S
ca,bvl−1abxl+1 . . . xk+1 (modI).
Since supp (fα) is a chain in M with respect to ≺, for every (a, b) ∈ S, either (a, b) ≺ (wl−1, xl) or
(wl−1, xl) ≺ (a, b). If (a, b) ≺ (wl−1, xl), b is contained in Aq with q < jl. Using the definition of Ω
and the minimality of j in Ω, we obtain
vl−1abxl+1 . . . xk+1 ∈ I if (a, b) ∈ S, (a, b) ≺ (wl−1, xl).
According to the last two displays
ul−1 = X
(a,b)∈S
(wl−1 ,xl)≺(a,b)
ca,bvl−1abxl+1 . . . xk+1 (modI).
By (3.2) for l − 1, ul−1 /∈ I. Thus, using the above display, we can pick (a, b) ∈ S such that
(wl−1, xl) ≺ (a, b) and vl−1abxl+1 . . . xk+1 /∈ I. Now we set ul = vl−1abxl+1 . . . xk+1, wl = b,
vl = vl−1a and take ml such that wl = b ∈ Aml.
Since wl−1 ∈ Aml−1 and (wl−1, xl) ≺ (a, b) = (a, wl), we have ml > ml−1. Thus (3.1 -- 3.3) are
satisfied. This completes the inductive procedure of constructing m1, . . . , mk+1 and u1, . . . , uk+1.
By (3.1), mj for 1 6 j 6 k + 1 are k + 1 pairwise distinct elements of the k-element set Nk. We
have arrived to a contradiction, which proves that R is (k + 1)-step nilpotent.
4 Proofs of Theorem 1.3 and Lemma 1.2
Let k > 2, n ∈ N and a1, . . . , ak−1 ∈ Z+ be such that a1 + . . . + ak−1 = n. In order to prove
Theorem 1.3, it suffices to prove that there is a quadratic K-algebra R given by n generators and
d = max
16j6k−1
(a1 + . . . + aj)(aj + . . . + ak−1)
6
relations such that R is k-step nilpotent.
Let X be an n-element set of generators. Since a1 +. . .+ak−1 = n, we can present X as the union
of the pairwise disjoint sets A1, . . . , Ak−1 with Aj = aj for 1 6 j 6 k−1. Consider the set M ⊂ X 2
defined in (1.4) and the partial ordering ≺ on M generated by the partition {A1, . . . , Ak−1}. For
1 6 j 6 k − 1, let Bj = Sq>j>m
Aq × Am. Clearly, Bj = (a1 + . . . + aj)(aj + . . . + ak−1). Hence
d = max{B1, . . . ,Bk−1}. By Lemma 2.1, w(M,≺) = d. According to the Dilworth theorem (see
[4] for a short inductive proof) the width of a finite partially ordered set P is precisely the minimal
number of chains needed to cover P . Hence, we can write M =
d
Sq=1
Cq, where each Cq is a chain in
M . Now we consider the homogeneous degree 2 elements of KhXi given by
fq = X
(a,b)∈Cq
ab for 1 6 q 6 d.
Clearly supp (fq) = Cq. Thus the union of the supports of fq is M and each supp (fq) is a chain
in M . By Theorem 1.7, the algebra R given by the relations fq for 1 6 q 6 d is k-step nilpotent.
This completes the proof of Theorem 1.3.
Now we shall prove Lemma 1.2. By Theorems GS and 1.3, dn,k > ϕkn2 for every k > 2 and
n ∈ N. This proves the first inequality in (1.3). It remains to prove the second one. By Lemma 2.2,
there are α0, . . . , αk−1 ∈ [0, 1] such that 0 = α0 < α1 < . . . < αk−1 = 1 and αj(1 − αj−1) = ϕk for
1 6 j 6 k − 1. Now for 0 6 j 6 k − 1 let bj = ⌈nαj − 1
2⌉. Clearly 0 = b0 6 b1 6 . . . 6 bk−1 = n.
Now we set aj = bj − bj−1 for 1 6 j 6 k − 1. Then aj ∈ Z+ and a1 + . . . + ak−1 = n. Hence
dn,k 6 max
bj(n−bj−1) = max
16j6k−1(cid:6)nαj− 1
2(cid:7)·(cid:4)n(1−αj−1)+ 1
2(cid:5).
16j6k−1
(a1+. . .+aj)(aj+. . .+ak−1) = max
16j6k−1
It is easy to see that for every α, β ∈ [0, 1],
(cid:6)nα − 1
2(cid:7) · (cid:4)nβ + 1
2(cid:5) − αβn2 6 α+β
2 n + 1
4 .
From the last two displays and the equalities αj(1 − αj−1) = ϕk it follows that
dn,k 6 ϕkn2 +
n
2
max
16j6k−1
(1 + αj − αj−1) +
1
4
.
By Lemma 2.2, the maximum in the above display equals ϕk. Thus dn,k 6 ϕkn2 + 1+ϕk
which completes the proof of Lemma 1.2.
2 n + 1
4 ,
5
4-Step nilpotency and the Fibonacci numbers
First, we derive an explicit formula for dn,4.
Lemma 5.1. For every n ∈ N,
dn,4 = min(cid:8)(cid:6)
√5−1
2 n(cid:7)2, n(cid:6) 3−
√5
2 n(cid:7)(cid:9).
(5.1)
Proof. Using (1.2) with k = 4 and denoting a = a1 and b = a3, we obtain
dn,4 = min{max{na, nb, (n − a)(n − b)} : a, b ∈ Z+, a + b 6 n}.
An obvious symmetry consideration yields
dn,4 = min{max{na, nb, (n − a)(n − b)} : a, b ∈ Z+, b 6 a, a + b 6 n}.
7
Since nb 6 na and (n − a)(n − b) > (n − a)2 when a, b ∈ Z+ satisfy b 6 a 6 n, we have
dn,4 = min{max{na, (n − a)2} : a ∈ Z+, 2a 6 n}.
(5.2)
Now, assume that a ∈ Z+ satisfies 2a 6 n. Solving a quadratic inequality we see that na >
(n − a)2 holds precisely when a > ϕ4n. Hence (5.2) can be rewritten as
dn,4 = min{an, bn}, where
an = min{na : a ∈ Z+, ϕ4n 6 a 6 n/2} and bn = min{(n − a)2 : a ∈ Z+, a 6 ϕ4n}.
√5−1
2
and 1 − ϕ4 =
, we see that (5.1) follows from the above display.
If m is a square, we can write m = j2 for some j ∈ N. Now it is easy to see that j = (cid:6)
If m is divisible by n, we can write m = nj for some j ∈ N. Now it is easy to see that j = (cid:6) 3−
Clearly, the minimum in the definition of an is attained for a = ⌈ϕ4n⌉ and the minimum in the
definition of bn is attained for a = ⌊ϕ4n⌋. Hence an = n⌈ϕ4n⌉ and bn = ⌈(1 − ϕ4)n⌉2. Using the
√5
equalities ϕ4 = 3−
2
Corollary 5.2. The equality dn,4 = (cid:6)ϕ4n2(cid:7) holds if and only if either (cid:6)ϕ4n2(cid:7) is divisible by n or
(cid:6)ϕ4n2(cid:7) is a square of a positive integer.
Proof. Let m = (cid:6)ϕ4n2(cid:7). From Lemma 5.1 it follows that dn,4 is always either divisible by n or is a
square. Thus the equality m = dn,4 can only hold if either m is divisible by n or m is a square.
√5
2 n(cid:7)
and therefore, by Lemma 5.1, dn,4 > jn = m. On the other hand, choosing a = j and using (5.2),
we get dn,4 6 max{nj, (n − j)2} = nj. Thus dn,4 = nj = m.
√5−1
2 n(cid:7)
and therefore, by Lemma 5.1, dn,4 > j2 = m. On the other hand, choosing a = n − j and using
(5.2), we get dn,4 6 max{n(n − j), j2} = j2. Thus dn,4 = j2 = m.
Proof of the first part of Theorem 1.4. Let F0, F1, . . . be the Fibonacci sequence and ϕ =
be
the golden ratio number. Using the formula Fn = ϕn−(−ϕ)−n
together with the equality ϕ4 = ϕ−2,
one can easily verify that (cid:6)ϕ4F 2
k(cid:7) = FkFk−2 if k is even. Thus if n
is a Fibonacci number, then (cid:6)ϕ4n2(cid:7) is either divisible by n or is a square.
To show the converse, we use the following criterion of recognizing the Fibonacci numbers due
to Mobius [9]. It says that a positive integer n is a Fibonacci number if and only if the interval
(ϕn − n−1, ϕn + n−1) contains an integer. Furthermore, if m is an integer belonging to (ϕn −
n−1, ϕn + n−1), then m is the next Fibonacci number after n.
First, assume that n ∈ N and (cid:6)ϕ4n2(cid:7) is divisible by n. Then ϕ4n2 + θ = nk, where k ∈ N
and 0 < θ < 1. Since ϕ4 = 2 − ϕ, it follows that ϕn − (2n − k) = θ
n and therefore 2n − k ∈
(ϕn − n−1, ϕn + n−1). By the criterion of Mobius, n is a Fibonacci number. Finally, assume that
(cid:6)ϕ4n2(cid:7) is a square number. Since ϕ4 = ϕ−2, this means that n2
ϕ2 + θ = k2, where k ∈ N and
0 < θ < 1. It immediately follows that k = (cid:6) n
ϕ + α with 0 < α < 1.
Squaring the last equality, we get k2 = n2
ϕ2 + θ = n2
ϕ < θ < 1. Hence
ϕα < ϕ2
ϕ(cid:7). In other words k = n
ϕ2 + 2nα
ϕ + α2. In particular, 2nα
k(cid:7) = F 2
k−1 if k is odd and (cid:6)ϕ4F 2
2n . Thus the equality k = n
√5+1
2
√5
ϕ + α implies n = ϕk − ϕα and
ϕ2
ϕ2
2n
ϕ2
<
=
2(ϕk − ϕα)
2(ϕk − ϕ2/2n)
.
ϕα <
where the last inequality is satisfied for k > 2. Now the above display and the equality n = ϕk− ϕα
imply that n belongs to the interval (ϕk − k−1, ϕk + k−1). By the criterion of Mobius, both k and
Since n > k, we have
ϕα <
2(ϕk − ϕ2/2k)
<
1
k
,
ϕ2
8
n are Fibonacci numbers provided k > 2. If k = 1 or k = 2, a direct computation yields n = 2 or
n = 3 respectively, which are Fibonacci numbers as well.
Thus we have proven that (cid:6)ϕ4n2(cid:7) is either divisible by n or is a square number precisely when n
is a Fibonacci number. By Lemma 5.2, dn,4 = (cid:6)ϕ4n2(cid:7) if and only if n is a Fibonacci number.
6
5-Step nilpotency
In this section we prove the second part of Theorem 1.4. As in the previous section we start by
simplification the formula defining dn,5.
Lemma 6.1. If n ∈ N is even, then dn,5 = n
2 (cid:6) 2n2
dn,5 = n(cid:6) n(n+1)
3n+1(cid:7).
Proof. Using the symmetry in (1.2) with respect to reversing the order of aj, we have
3n+1 (cid:7). If n ∈ N is congruent to 1 or to 3 modulo 6, then dn,5 = n+1
3 (cid:7). If n ∈ N is congruent to −1 modulo 6, then
2(cid:6) 2n
dn,5 = min{S(a) : a ∈ Z4
S(a) = max{na1, na4, (a1 + a2)(a2 + a3 + a4), (a1 + a2 + a3)(a3 + a4)}.
+, a1 + a2 + a3 + a4 = n, a1 6 a4}, where
(6.1)
It is easy to see that the minimum in (6.1) can not be attained when a2 = 0 if n > 1 (the case
n = 1 is trivial anyway). If a1 < a4 and a2 > 0, one can easily check that S(a′) 6 S(a), where
a′ is obtained from a by increasing a1 by 1 with simultaneous decreasing of a2 by 1. Similarly, if
a1 = a4 and a2 − a3 > 1, S(a′) 6 S(a), where a′ is obtained from a by increasing the smaller of
a2 and a3 by 1 with simultaneous decreasing of the bigger one by 1. It follows that among a ∈ Z4
+
for which the minimum in (6.1) is attained there must be at least one point satisfying a1 = a4 and
a2 − a3 6 1. Thus the minimum in (6.1) is attained at a point a of the shape a = (α, β, β, α) if n
is even and it is attained at a point a of the shape a = (α, β + 1, β, α) if n is odd. Substituting this
data into (6.1), we get
dn,5 =
n
2
min{max{2a, n − a} : a ∈ Z+, a 6 n/2} if n is even
(6.2)
and
dn,5 = min{max{na, (n + 1)(n − a)/2} : a ∈ Z+, a 6 n/2} if n is odd.
(6.3)
Since max{2a, n − a} = n − a if 3a 6 n and max{2a, n − a} = 2a if 3a > n, (6.2) implies that
dn,5 = min(cid:8)n(cid:6) n
3 (cid:7) if n is even (the two numbers in the last minimum are equal in
all cases except for the numbers n congruent to −2 modulo 6 in which case the second one is less
by 1).
3 (cid:7)(cid:9) = n
3(cid:7), n
2(cid:6) 2n
2(cid:6) 2n
n(n+1)
na if a >
Next, max{na, (n+1)(n−a)/2} = (n+1)(n−a)/2 if a 6
3n+1 . Plugging this into (6.3), we get dn,5 = min{n(cid:6) n(n+1)
3n+1 and max{na, (n+1)(n−a)/2} =
2 (cid:6) 2n2
3n+1(cid:7)}. Considering
the cases of n being 1, 3 and −1 modulo 6 separately, we see that dn,5 = n(cid:6) n(n+1)
3n+1 (cid:7) if n is congruent
to −1 modulo 6 and dn = n+1
3n+1(cid:7) ff n ∈ N is congruent to 1 or to 3 modulo 6.
3n+1 (cid:7), n+1
2 (cid:6) 2n2
n(n+1)
From Lemma 6.1 it immediately follows that dn,5 = n2
3 = ϕ5n2 if 6 is a factor of n. Considering
the exact formula provided by Lemma 6.1 and treating the possible remainders for the division of
n by 6 as separate cases, one easily sees that dn,5 − n2
> 1 and therefore dn,5 > ⌈ϕ5n2⌉ if n is not
divisible by 6 and n > 3. It is easy to verify that the equality dn,5 = (cid:6)ϕ5n2(cid:7) holds for n = 1 and
for n = 2. This completes the Proof of Theorem 1.4.
3
We conclude by reminding that the following particular cases of the Anick's conjecture [1] remain
unproved.
9
Conjecture 6.2. There is a k-step nilpotent K-algebra given by n generators and d quadratic
relations whenever d > ϕkn2.
Conjecture 6.3. There is a finite dimensional K-algebra given by n generators and d quadratic
relations whenever d > n2
4 .
Acknowledgements
We are grateful to the Max-Planck-Institute for Mathematics in Bonn and to IHES, where parts
of this research have been done, for hospitality, support, and excellent research atmosphere. This
work is funded by the ERC grant 320974, and partially supported by the project PUT9038. We also
would like to thank a referee for valuable suggestions, which helped to improve the presentation.
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] P. Cameron and N. Iyudu, Graphs of relations and Hilbert series, J. Symbolic Comput. 42 (2007),
1066 -- 1078
[4] F. Galvin, A proof of Dilworth's chain decomposition theorem, Amer. Math. Monthly 101 (1994), 352 --
353
[5] E. Golod and I. Shafarevich, On the class field tower (Russian), Izv. Akad. Nauk SSSR Ser. Mat. 28
1964, 261 -- 272
[6] 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
[7] N. Iyudu and S. Shkarin, Finite dimensional semigroup quadratic algebras with minimal number of
relations, Monats. Math. [to appear]
[8] T. Lenagan and A. Smoktunowicz, An infinite dimensional affine nil algebra with finite Gelfand -- Kirillov
dimension, J. Amer. Math. Soc. 20, 989 -- 1001 (2007)
[9] M. Mobius, Wie erkennt Man eine Fibonacci Zahl?, Math. Semesterber 45 (1998), 243 -- 246
[10] M. Newman, C. Schneider and A. Shalev, The entropy of graded algebras, J. Algebra 223 (2000), 85 -- 100
[11] A. Polishchuk and L. Positselski, Quadratic algebras, University Lecture Series 37 American Mathe-
matical Society, Providence, RI, 2005
[12] A. Smoktunowicz, Some results in non-commutative ring theory, International Congress of Mathemati-
cians II, Eur. Math. Soc., Zurich, 259 -- 269 (2006)
[13] V. Ufnarovskii, Combinatorial and asymptotic methods in algebra (Russian) Current problems in math-
ematics. Fundamental directions 57 5 -- 177, Moscow, 1990
[14] I. Wisliceny, Konstruktion nilpotenter associativer Algebren mit wenig Relationen, Math. Nachr.
147(1990), 75 -- 82
[15] E. Zelmanov, Some open problems in the theory of infinite dimensional algebras, J. Korean Math. Soc.
44, 1185 -- 1195 (2007)
[16] E. Zelmanov, Infinite algebras and pro-p groups, In: Infinite groups: geometric, combinatorial and
dynamical aspects, Progress in Mathematics 248 , Birkhauser, Basel, 403 -- 413 (2005)
Natalia Iyudu
School of Mathematics
The University of Edinburgh
James Clerk Maxwell Building
The King's Buildings
Mayfield Road
Edinburgh
10
Scotland EH9 3JZ
E-mail address:
and
[email protected]
Stanislav Shkarin
Queens's University Belfast
Pure Mathematics Research Centre
University road, Belfast, BT7 1NN, UK
E-mail address:
[email protected]
11
|
1507.06290 | 3 | 1507 | 2016-01-03T06:26:47 | Idempotents and structures of rings | [
"math.RA"
] | We study a ring containing a complete set of orthogonal idempotents as a generalized matrix ring via its Peirce decomposition. We focus on the case where some of the underlying bimodule homomorphisms are zero. Upper and lower triangular generalized matrix rings are pertinent examples of the class of rings which we study. The triviality of the particular bimodule homomorphisms motivates the introduction of three new types of idempotents, namely inner Peirce trivial idempotents, outer Peirce trivial idempotents and Peirce trivial idempotents. These idempotents provide the main tools in our investigations. | math.RA | math |
IDEMPOTENTS AND STRUCTURES OF RINGS
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
Abstract. Recall that an n-by-n generalized matrix ring is defined in terms of
sets of rings {Ri}n
i=1, (Ri, Rj)-bimodules {Mij}, and bimodule homomorphisms
θijk : Mij ⊗Rj Mjk → Mik, where the set of diagonal matrix units {Eii} form
a complete set of orthogonal idempotents. Moreover, an arbitrary ring with
a complete set of orthogonal idempotents {ei}n
i=1 has a Peirce decomposition
which can be arranged into an n-by-n generalized matrix ring Rπ which is iso-
morphic to R. In this paper, we focus on the subclass Tn of n-by-n generalized
matrix rings with θiji = 0 for i 6= j. Tn contains all upper and all lower gener-
alized triangular matrix rings. The triviality of the bimodule homomorphisms
motivates the introduction of three new types of idempotents called the inner
Peirce, outer Peirce, and Peirce trivial idempotents. These idempotents are our
main tools and are used to characterize Tn and define a new class of rings called
the n-Peirce rings. If R is an n-Pierce ring, then there is a certain complete
set of orthogonal idempotents {ei}n
i=1 such that Rπ ∈ Tn. We show that every
n-by-n generalized matrix ring R contains a subring S which is maximal with
respect to being in Tn and S is essential in R as an (S, S)-bisubmodule of R.
This allows for a useful transfer of information between R and S. Also, we show
that any ring is either an n-Peirce ring or for each k > 1 there is a complete set
of orthogonal idempotents {ei}k
i=1 such that Rπ ∈ Tk. Examples are provided
to illustrate and delimit our results.
Introduction
Throughout this paper all rings are associative with a unity and modules are
unital unless explicitly indicated otherwise.
Given a complete set of orthogonal idempotents, {ei}n
i=1, of a ring R, we can
form a group direct sum,
R = e1Re1 ⊕ · · · ⊕ e1Ren ⊕ e2Re1 ⊕ · · · ⊕ e2Ren ⊕ · · · ⊕ enRe1 ⊕ · · · ⊕ enRen,
called the Peirce decomposition of R. This decomposition can be arranged into
an n-by-n square array, called Rπ, with
2010 Mathematics Subject Classification. 16S50, 15A33, 16D20, 16D70.
Key words and phrases. idempotent, generalized matrix ring, formal matrix ring, Morita
Context, Peirce trivial, annihilator, bimodule, essential, ideal extending.
Corresponding author: G.F. Birkenmeier.
1
2
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
e1Re1
e1Re2
e2Re1
e2Re2
...
. . .
· · ·
. . .
. . .
e1Ren
...
en−1Ren
enRe1
· · ·
enRen−1
enRen
Rπ =
.
The array Rπ forms a ring, where addition is componentwise and multiplication is
the usual row-column matrix multiplication. Moreover, there is a ring isomorphism
h : R → Rπ defined by h(x) = [eixej] for all x ∈ R. Observe that the eiRei
are rings with unity and the eiRej are (eiRei, ejRej)-bimodules. Note that the
bimodule product eiRej · ejRek, arising in the row-column multiplication, may
be thought of as a bimodule homomorphism θijk : eiRej ⊗ej Rej ejRek → eiRek
determined by the multiplication of R.
The above discussion motivates the following well known definition:
an n-by-n generalized (or formal) matrix ring R is a square array
R =
R1 M12
M21 R2
· · ·
. . .
M1n
...
...
. . .
. . . Mn−1,n
Mn1
· · · Mn,n−1
Rn
where each Ri is a ring, each Mij is an (Ri, Rj)-bimodule and there exist (Ri, Rk)-
bimodule homomorphisms θijk : Mij ⊗Rj Mjk → Mik for all i, j, k = 1, . . . , n (with
Mii = Ri). For mij ∈ Mij and mjk ∈ Mjk, mijmjk denotes θijk(mij ⊗ mjk).
The homomorphisms θijk must satisfy the associativity relation: (mijmjk)mkℓ =
mij(mjkmkℓ) for all mij ∈ Mij, mjk ∈ Mjk, mkℓ ∈ Mkℓ and all i, j, k, ℓ = 1, . . . , n.
Observe that θiii is determined by the ring multiplication in Ri, while θijj and θjjk
are determined by the bimodule scalar multiplications. Further information on
generalized matrix rings can be found in [KT].
With these conditions, addition on R is componentwise and multiplication on R
is row-column matrix multiplication. A Morita context is a 2-by-2 generalized
matrix ring. An n-by-n generalized upper (lower) triangular matrix ring is a gen-
eralized matrix ring with Mij = 0 for j < i (Mij = 0 for i < j). Note that
{Eii ∈ R Eii is the matrix with 1 ∈ Ri in the (i, i)-position and 0 elsewhere,
i = 1, . . . , n} is a complete set {Eii}n
i=1 of orthogonal idempotents in the above
constructed generalized matrix ring R.
The foregoing observations allow us to consider a generalized matrix ring in two
ways:
IDEMPOTENTS AND STRUCTURES OF RINGS
3
(1) given a ring R and a complete set of orthogonal idempotents, {ei}n
i=1, then
Rπ is an ”internal” representation of R as a generalized matrix ring in terms of
substructures of R; whereas
(2) given collections {Ri}, {Mij}, and {θijk}, we construct a new ring from
these ”external” components via the generalized matrix ring notion.
An important problem in the study of generalized matrix rings is: given a
collection of rings {Ri i = 1, . . . , n} and bimodules {Mij i, j = 1, . . . , n, i 6= j,
and each Mij is an (Ri, Rj)-bimodule} determine the θiji (i 6= j) and the θijk
(i, j, k distinct) to produce an n-by-n generalized matrix ring. We can simplify
this problem by trivializing the θijk in the following three ways (note that for
n = 2, all three conditions coincide):
(I) Define θiji = 0, for all i 6= j.
(II) Define θijk = 0, for all i, j, k pairwise distinct.
(III) Define θijk = 0, for all i 6= j and j 6= k (I and II combined).
Two questions immediately arise:
(A) Are there significant examples of generalized matrix rings with trivial-
ized θijk?
(B) How can the theory of generalized matrix rings with trivialized θijk be used
to gain insight into the theory of arbitrary generalized matrix rings?
In this paper, we consider the class of n-by-n (n > 1) generalized matrix rings
satisfying condition (I) (i.e., θiji = 0, for all i 6= j).
We denote this class of rings by Tn.
For each generalized matrix ring R,
we use R to denote the ring in Tn which has the same corresponding Ri, Mij, θijk
as R, except that for all i 6= j the homomorphisms θiji are taken to be 0 in R.
Thus R and R are the same ring if and only if R ∈ Tn. Note that the classes of
n-by-n generalized upper and lower triangular matrix rings form significant proper
subclasses of Tn (see Question A). Further examples are provided throughout this
paper.
Observe that the triviality of the θiji motivates three new types of idempo-
tents which appear in the internal (Peirce decomposition) generalized matrix ring
representation of a ring in T2. For e = e2 ∈ R,
(1) e is inner Peirce trivial if eR(1 − e)Re = 0;
(2) e is outer Peirce trivial if (1 − e)ReR(1 − e) = 0;
(3) e is Peirce trivial if e is both inner and outer trivial.
In [P] B. Peirce introduced the concept of an idempotent, and so we are nam-
ing certain idempotents and rings in this paper in his honor. These idempotents
provide the main tools in our investigations; in particular, they are used to char-
acterize the class Tn and the class of n-Peirce rings.
In Section 1, we develop the basic properties of the inner (outer) Peirce trivial
idempotents. Moreover we show that if R is a subring of a ring T and S is
4
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
the subring of T generated by R and a subset E of inner or outer Peirce trivial
idempotents of T , then there is a useful transfer of information between R and S,
e.g., R is strongly π-regular or has classical Krull dimension n if and only if so
does S (Theorems 1.13, 1.14 and 1.16).
In Section 2, we begin by showing that, for a ring R with a complete set {ei}n
i=1
of orthogonal idempotents, Rπ ∈ Tn if and only if each ei is inner Peirce triv-
ial (Theorem 2.2).
Next, let R be a generalized n-by-n matrix ring and take
D(R) = {[rij] ∈ R rij = 0 for all i 6= j}
and
D(R)− = {[rij] ∈ R rii = 0 for all i = 1, . . . , n}.
We obtain that if R ∈ Tn then D(R)− E R such that (D(R)−)n = 0 and D(R) is
ring isomorphic to R/(D(R)−) (Proposition 2.4).
The transfer of various ring properties (e.g., semilocal, bounded index, having
a polynomial identity) between R and D(R) is considered when R ∈ Tn. In Theo-
rem 2.12 (one of the main results of the paper) we show that every n-by-n gener-
alized matrix ring has subrings S maximal with respect to being in Tn such that
S is essential in R as an (S, S)-bimodule. This fact allows for a two-step transfer
of information from D(R) to S (Theorem 1.16) and from S to R (Theorem 2.12
and Corollary 2.14).
In the remainder of this section, we introduce the notion of an ideal extending
ring and use Theorem 2.12 and its consequences to show how this notion passes
from a ring A to certain generalized matrix rings which are overrings of the n-
by-n upper triangular matrix ring over A. Thus Theorem 2.12 and its corollaries
provide answers to Question B.
The n-Peirce rings are introduced and investigated in Section 3. A ring R is a
1-Peirce ring
if 0 and 1 are the only Peirce trivial idempotents in R. Inductively, for a natural
number n > 1, we say a ring R is an
n-Peirce ring
if there is a Peirce trivial idempotent e such that eRe is an m-Peirce ring for some
1 ≤ m < n and (1 − e)R(1 − e) is an (n − m)-Peirce ring.
In Theorem 3.7, we show that an n-Peirce generalized matrix ring is in Tn
(n > 1) and that if R has a complete set {ei}n
i=1of orthogonal idempotents such
that each eiRei is a 1-Peirce ring, then R is a k-Peirce ring for some 1 ≤ k ≤ n.
Example 3.2 shows that the class of n-Peirce rings is a proper subclass of Tn for
n > 1, and that any n-by-n generalized upper triangular matrix ring with prime
diagonal rings is an n-Peirce ring. The class of n-Peirce rings has an advantange
over Tn in that for n > 1, an n-Peirce ring has a complete set {ei}n
i=1 of orthogonal
idempotents such that each eiRei is a 1-Peirce ring. In Theorem 3.11, it is also
IDEMPOTENTS AND STRUCTURES OF RINGS
5
shown that if R has DCC on {ReR e is a Peirce trivial idempotent}, then R is
an n-Peirce ring for some n.
As indicated in Definition 1.1, the inner (outer) Peirce trivial idempotents can
be defined in a ring without a unity. Hence, many of the results in this paper can
be modified to hold in rings without a unity.
Notation and Terminology
(1) R is Abelian - means every idempotent is central.
(2) B(R), P(R) and J (R) denote the central idempotents of R, the prime
radical of R and the Jacobson radical of R respectively.
(3) Sℓ(R) = {e = e2 ∈ R Re = eRe}, Sr(R) = {e = e2 ∈ R eR = eRe}.
(4) Cen(R) is the center of R.
(5) U(R) is the group of units of R.
(6) < − >R is the subring of R generated by −, and (−)R is the ideal of R
generated by −.
(7) X E R means X is an ideal of R.
(8) rA(B) and ℓA(B) denote the right and left annihilator of B in A, respec-
tively.
(9) Z and Zn denote the ring of integers and the ring of integers modulo n,
respectively.
(10) Z+ means the positive integers.
1. Basic Properties of Peirce trivial Idempotents
Definition 1.1. Let R be a ring, not necessarily with a unity, and let e = e2 ∈ R.
We say e is inner Peirce trivial (respectively, outer Peirce trivial ) if exye = exeye
(respectively, xey + exeye = xeye + exey) for all x, y ∈ R. If e is both inner and
outer Peirce trivial, we say e is Peirce trivial.
For a ring R with a unity, e is inner (respectively, outer) Peirce trivial if and
only if eR(1 − e)Re = {0} (respectively, (1 − e)ReR(1 − e) = {0}); moreover, e is
inner Peirce trivial if and only f = 1 − e is outer Peirce trivial. Let
Pit(R), Pot(R) and Pt(R)
denote the set of all inner Peirce trivial idempotents, all outer Peirce trivial
idempotents and all Peirce trivial idempotents of R, respectively. Note that
B(R) ⊆ Pt(R).
Example 1.2. Inner and outer Peirce trivialities are independent properties of
idempotents. Let R1 = Z, R2 = Z/8Z = Z8, M12 = Z4, M21 = Z2, together with
∼= 4R2,
tensor products M12 ⊗R2 M21
∼= Z2 7→ 0 ∈ R1 and M21 ⊗R1 M12
R1 M12
1 0
respectively. Then in R =
M21 R2
the elements e =
∼= Z2
and f =
0 0
6
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
are idempotents. Moreover, e is inner Peirce trivial, but not outer Peirce
trivial, and f is outer Peirce trivial, but not inner Peirce trivial.
0 1
0 0
Proposition 1.3. Let R =
R1 M12
M21 R2
∈ T2 and assume α =
e m
n f
∈ R.
(1) Then α = α2 if and only if e = e2, f = f 2, em + mf = m and ne + f n = n.
(2) If α = α2 and e and f are central idempotents, then α ∈ Pt(R).
In
particular, if R1 and R2 are commutative, then Pt(R) = {α ∈ R α = α2}.
Proof. The proof is a straightforward calculation using Definition 1.1.
(cid:3)
Note that Proposition 1.3(2) is, in general, not true when R ∈ Tn for n > 2 (see
Example 1.9).
As a consequence of Definition 1.1, one has the following descriptions:
Lemma 1.4. For e2 = e ∈ R the following claims are equivalent:
(1) e is inner Peirce trivial.
(2) eℓR(e) = eR(1 − e) is a right ideal of R.
(3) rR(e)e = (1 − e)Re is a left ideal of R.
(4) ef ge = ef ege for all idempotents f, g ∈ R.
(5) h : R → eRe, defined by h(x) = exe, is a surjective ring homomorphism.
(6) eRtRe = 0 for all t ∈ R such that ete = 0.
(7) ReR ⊆ ℓR((1 − e)Re).
Proof. We show implication 4 ⇒ 1; the remaining implications are routine. For
any x, y ∈ R simple computation shows f := e − ex + exe = ef = f 2 and
g := e − ye + eye = ge = g2, whence one has by assumption e + exye − exeye =
ef ge = ef ege = e, implying exye = exeye. Therefore e is inner Peirce trivial. (cid:3)
Lemma 1.5. For e2 = e ∈ R the following claims are equivalent:
(1) e is outer Peirce trivial.
(2) eℓR(e) = eR(1 − e) is a left ideal of R.
(3) rR(e)e = (1 − e)Re is a right ideal of R.
(4) f eg + ef ege = f ege + ef eg for all idempotents f, g ∈ R.
Proof. Again, we show the implication 4 ⇒ 1; the remaining implications are
routine. For any x, y ∈ R simple computation shows f := e + xe − exe = f e = f 2
and g := e + ey − eye = eg = g2, whence one has by assumption f eg + ef ege =
f g + e = f ege + ef eg = f + g. Therefore we have the equality e + (e + xe −
exe)(e + ey − eye) = e + xe − exe + e + ey − eye, from which one can obtain, after
simplification, that e is outer Peirce trivial.
(cid:3)
Corollary 1.6. For e2 = e ∈ R the following claims are equivalent:
(1) e is Peirce trivial.
(2) eℓR(e) = eR(1 − e) is an ideal of R.
IDEMPOTENTS AND STRUCTURES OF RINGS
7
(3) rR(e)e = (1 − e)Re is an ideal of R.
(4) e, 1 − e ∈ Pit(R).
From the above results, one can see that if R is semiprime, then Pit(R) =
Pot(R) = Pt(R) = B(R).
Lemma 1.7. Let e, f ∈ R such that e = e2 and f = f 2.
(1) e ∈ Pit(R) implies ef e = (ef e)2, (ef )2 = (ef )3 and (f e)2 = (f e)3.
(2) e ∈ Pt(R) implies f ef = (f ef )2.
(3) e, f ∈ Pit(R) implies ef e, f ef ∈ Pit(R).
(4) If R is a generalized matrix ring and [eij] ∈ Pit(R) (resp. Pot(R), Pt(R)),
then eii ∈ Pit(Ri) (resp. Pot(Ri), Pt(Ri)).
Proof. This proof is routine.
(cid:3)
Lemma 1.8. Let c, e ∈ R such that c = c2 and e = e2.
(1) e ∈ Pit(R) if and only if Pit(eRe) = eRe ∩ Pit(R).
(2) Pt(R) ∩ cRc ⊆ Pt(cRc).
(3) Assume I E R. Then Pit(I) = I ∩ Pit(R).
Proof. (1) Clearly, eRe ∩ Pit(R) ⊆ Pit(eRe). Assume e ∈ Pit(R), c ∈ Pit(eRe)
and x, y ∈ R. Then cxyc = c(exye)c = c((exe)(eye))c = c(exe)c(eye)c = cxcyc.
Thus eRe ∩ Pit(R) = Pit(eRe). Conversely, assume eRe ∩ Pit(R) = Pit(eRe).
Since e ∈ Pit(eRe), then e ∈ Pit(R).
(2) The proof of this part is straightforward.
(3) Clearly, I ∩ Pit(R) ⊆ Pit(I). Let f ∈ Pit(I) and x, y ∈ R. Then f xyf =
(cid:3)
f ((f x)(yf ))f = f (f x)f (yf )f = f xf yf . Therefore Pit(I) = I ∩ Pit(R).
Example 1.9. In general, for c ∈ Pt(R), Pt(R) ∩ cRc ( Pt(cRc). Let R be
the 3-by-3 upper triangular matrix ring over a ring A with e = E22 ∈ R and
c = E22 + E33. Then c ∈ Pt(R) and e ∈ Pt(cRc), but e 6∈ Pot(R). Thus,
Pt(R) ∩ cRc ( Pt(cRc).
In [BHKP] (also see [AvW1] and [AvW2]), it is shown that a ring R has a
generalized triangular matrix form if and only if it has a set of left (or right)
triangulating idempotents. Such a set is an ordered complete set of orthogonal
idempotents which are contructed from Sℓ(R) and Sr(R).
Our next result and results from Sections 2 and 3 show that Pit(R) and Pt(R)
can be used to naturally extend the notion of a generalized triangular matrix ring.
Moreover, the inherent symmetry in the definitions of Pit(R) and Pt(R) frees us
from the ”ordered” condition on sets of idempotents when characterizing these
natural extensions.
Proposition 1.10. (1) Sl(R) ∪ Sr(R) ⊆ Pt(R).
(2) Let {e1, . . . , en} be a set of left or right triangulating idempotents of R.
Then {e1, . . . , en} ⊆ Pit(R).
Proof. (1) This part is immediate from the definitions.
(2) From [BHKP, p. 560 and Corollary 1.6] and Lemma 1.8, each ei ∈ Pit(R).
(cid:3)
8
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
From Proposition 1.10, B(R) ⊆ Sl(R) ∪ Sr(R) ⊆ Pt(R) ⊆ Pit(R) (Pot(R)).
If R is semiprime these containment relations become equalities by Lemmas 1.4
and 1.5.
Example 1.11. Let A be a ring whose only idempotents are 0 and 1. Assume
0 6= X, Y E A. Using Proposition 1.3 we obtain:
(1) Let R = (cid:20) A X
(2) Let R = (cid:20) A X
0 A (cid:21). Then Sl(R) ∪ Sr(R) = Pt(R) = {e e = e2 ∈ R}.
Y A (cid:21) and XY = 0 = Y X. Then Sr(R) = Sl(R) = {0, 1} (
Pt(R) = {e e = e2 ∈ R}.
(3) Let B be a subring of A with X E B and R = (cid:20) A X
A/X B (cid:21). Then
Sr(R) = Sl(R) = {0, 1} ( Pt(R) = {e e = e2 ∈ R}.
We conclude Section 1 by showing in the next results (1.12 - 1.16) that given a
base ring R, an overring T , and a set E contained in Pit(T ) ∪ Pot(T ), there is a
significant transfer of information between R and S, where S is the subring of T
generated by R and E. These results indicate the importance of the inner and
outer Peirce trivial idempotents.
Let S be an overring of R. We consider the following properties between prime
ideals of R and S (see [BPR2, pp. 295-296] or [R1, p. 292]).
(1) Lying over (LO). For any prime ideal P of R, there exists a prime ideal Q
of S such that P = Q ∩ R.
(2) Going up (GO). Given prime ideals P1 ⊆ P2 of R and Q1 of S with P1 =
Q1 ∩ R, there exists a prime ideal Q2 of S with Q1 ⊆ Q2 and P2 = Q2 ∩ R.
(3) Incomparable (INC). Two different prime ideals of S with the same contrac-
tion in R are not comparable.
Lemma 1.12. Let T be a ring, R a subring of T ,
EP = {e = e2 ∈ T e + P(T ) is central in T /P(T )},
and S = hR ∪ EiT , where ∅ 6= E ⊆ EP Then:
(1) Pit(T ) ∪ Pot(T ) ⊆ EP.
(2) If K is a prime ideal of S, then R/(K ∩ R) ∼= S/K.
(3) LO, GU and INC hold between R and S.
Proof. (1) This part follows from Lemmas 1.4 and 1.5.
(2) and (3). The proof of these parts is similar to that in [BPR1, Lemma 2.1]
(cid:3)
or [BPR2, Lemma 8.3.26].
Recall that a ring R is strongly π-regular if for each x there is a positive integer n
(depending on x) such that xn ∈ xn+1R.
Theorem 1.13. Let C be a property of rings such that a ring A has property C
if and only if every prime factor of A has property C. Assume T is a ring, R is a
subring of T and S := hR ∪ EiT , where ∅ 6= E ⊆ EP, with EP as in Lemma 1.12.
IDEMPOTENTS AND STRUCTURES OF RINGS
9
Then R has property C if and only if S has property C. In particular, R is strongly
π-regular if and only if S is strongly π-regular.
Proof. Assume R has property C and K is a prime ideal of S. From Lemma 1.12(2),
R/(K ∩ R) is a prime ring, and so R/(K ∩ R) has property C. Hence S/K has
property C. Therefore S has property C.
Conversely, assume S has property C and P is a prime ideal of R. From
Lemma 1.12(3), LO holds between R and S. So there exists a prime ideal Q
of S such that Q ∩ R = P . By Lemma 1.12(2), R/P = R/(Q ∩ R) ∼= S/Q. Hence
R/P has property C. Therefore R has property C.
From [FS], a ring A is strongly π-regular if and only if every prime factor of A
(cid:3)
is a strongly π-regular ring.
See [GW] for the definition of a special radical. Observe that the prime, Jacob-
son, and nil radicals are included in the collection of special radicals.
Theorem 1.14. Let R be a subring of a ring T, ∅ 6= E ⊆ EP, and S = hR ∪ EiT .
Then:
(1) ρ(R) = ρ(S) ∩ R, where ρ is any special radical.
(2) The classical Krull dimensions of both S and R are equal.
(3) If S is a von Neumann regular ring, then so is R.
Proof. Using Lemma 1.12, the proof is similar to [BPR1, Theorem 2.2] or [BPR2,
Theorem 8.3.28].
(cid:3)
Lemma 1.15. Let T ∈ Tn (n > 1) and
Ek = {[tij] ∈ T tkj ∈ Mkj for j 6= k, tkk = 1 ∈ Tk and all other entries are zero}.
Then ∪n
k=1Ek ⊆ Pit(T ).
Proof. The proof of this result is a routine but tedious application of Definition 1.1.
(cid:3)
Theorem 1.16. Let T ∈ Tn (n > 1), R = D(T ), and E = ∪n
Lemma 1.15). Then:
k=1Ek (as in
(1) hR ∪ EiT = S = T .
(2) R has property C (as in Theorem 1.13) if and only if T has property C.
(3) ρ(R) = ρ(T ) ∩ R.
(4) The classical Krull dimension of both R and T are equal.
Proof. Use Lemmas 1.12 and 1.15 and Theorems 1.13 and 1.14.
(cid:3)
This result extends [TLZ, Corollary 3.6].
2. Characterization of Tn
Lemma 2.1. Let {e1, . . . , en} be a complete set of orthogonal idempotents of R.
(1) ei ∈ Pit(R) if and only if eiRejRei = 0 for all j 6= i.
(2) ej ∈ Pot(R) if and only if eiRejRek = 0 for all i 6= j and k 6= j.
(3) {e1, . . . , en} ⊆ Pot(R) if and only if {e1, . . . , en} ⊆ Pt(R).
10
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
Proof. (3) follows obviously from (1) and (2). Furthermore, (1) and (2) are imme-
diate consequences of the definition of Peirce trivial idempotents by observing
0 = eiR(1 − ei)Rei = eiR(Xj6=i
ej)Rei = Xj6=i
eiRejRei ⇔ ∀j 6= i eiRejRei = 0,
and
0 = (1 − ej)RejR(1 − ej) = ⊕i6=j6=keiRejRek ⇔ ∀i 6= j 6= k eiRejRek = 0.
(cid:3)
Lemma 2.1 shows remarkably that inner and outer Peirce trivial idempotents
behave quite differently when they are considered together as a complete set of
idempotents although their definition seems very symmetric! Lemma 2.1 shows
clearly the equivalence of the first three statements in the next result.
Theorem 2.2. Let {e1, . . . , en} be a complete set of orthogonal idempotent ele-
ments of R. The following conditions are equivalent:
(1) Rπ ∈ Tn.
(2) eiRejRei = 0, for all i 6= j.
(3) {e1, . . . , en} ⊆ Pit(R).
(4) D(Rπ)− is a right ideal of Rπ.
Proof. (3) ⇔ (4) Let X = D(Rπ)−. Observe that the (i, i)-position of XRπ is
Pk6=i eiRkRei = eiR(1−ei)Rei; and the (i, j)-position of XRπ isPk6=i, k6=j eiRekRej ⊆
eiRej. Therefore XRπ ⊆ X if and only if each ei ∈ Pit(R).
(cid:3)
Observe that from Lemma 1.4 and Theorem 2.2, any property that is preserved
by a surjective ring homomorphism passes from a ring in Tn to its diagonal rings.
Corollary 2.3. Let R be an n-by-n generalized matrix ring. Then the following
conditions are equivalent:
(1) R ∈ Tn.
(2) Let [aij], [bij] ∈ R with [cij] = [aij][bij]. Then cii = aiibii for all i and j.
(3) {Eii ∈ R i = 1, . . . , n} ⊆ Pit(R).
Thus Tn is exactly the class of n-by-n generalized matrix rings in which the
diagonal entries of the product of two matrices is completely determined by the
corresponding entries of the diagonals of the factor matrices.
Note that the idempotents in a generalized matrix ring are not characterized.
ii and
However, for R ∈ Tn, e = e2 ∈ R if and only if e = [eij], where eii = e2
eij = Pn
k=1 eikekj for i 6= j.
Proposition 2.4. Assume {e1, . . . , en} is a complete set of orthogonal idempo-
tents of R, and X = D(Rπ)−.
(1) If {e1, . . . , en} ⊆ Pit(R), then X n = 0 and ⊕n
i=1Ri is a homomorphic image
of Rπ with kernel X, where Ri = eiRei.
(2) If {e1, . . . , en} ⊆ Pt(R), then X 2 = 0.
IDEMPOTENTS AND STRUCTURES OF RINGS
11
Proof. (1) Observe that the (i, j)-position of X n−1 is a sum of terms where each
term is an element of eiReα1Reα2R · · · eαn−2Rej where αk ∈ {1, . . . , n} − {i, j}.
Since {e1, . . . , en} ⊆ Pit(R), it follows that X n = 0. The second part follows from
Lemma 1.4(5) and Theorem 2.2.
(2) This part follows from Lemma 2.1(2).
(cid:3)
Example 2.5. Let A be a ring and X, Y E A such that X 2 ⊆ Y . Take
R =
Then routine calculation yields:
A X Y
X A X
Y X A
.
(1) E22 ∈ Pt(R) if and only if X 2 = 0.
(2) {E11, E22, E33} ⊆ Pit(R) if and only if X 2 = 0 = Y 2.
(3) {E11, E22, E33} ⊆ Pt(R) if and only if X 2 = 0 = Y 2 and XY = 0 = Y X.
For an illustration of (2) and (3), let B be a ring and A = B[x, y]/(x2, y2) and
A = B[x, y]/(x2, y2, xy), respectively.
The next three results (2.6 - 2.8) indicate a transfer of important ring properties
between D(R) and R ∈ Tn.
Corollary 2.6. Let R ∈ Tn (n > 1). Then D(R) satisfies each of the following
conditions if and only if R does so:
(1) semilocal,
(2) semiperfect,
(3) left (or right) perfect,
(4) semiprimary,
(5) bounded index (of nilpotence).
Proof. Observe that if R satisfies any of (1) - (5), then so does eRe for each
e = e2 ∈ R. Hence, if R satisfies any of (1) - (5), then so does D(R).
Conversely, first assume that D(R) is semilocal. From Proposition 2.4(1) we
have that D(R)− E R, R/(D(R)−) ∼= D(R) and D(R)− ⊆ J (R). Therefore,
R/J (R) ∼= (R/(D(R)−))/(J (R)/(D(R)−)) ∼= D(R)/J (D(R)) is semisimple ar-
tinian. Thus R is semilocal. Parts (2) - (4) are proved similarly. Now assume
D(R) has bounded index k, and let v be a nilpotent element of R. Then v = d + x
where d ∈ D(R) and x ∈ D(R)− and vm = 0. Then 0 = vm = dm + y where
y ∈ D(R)−. So dm = −y ∈ D(R)−. Hence d is nilpotent, so dk = 0. Then
vk = dk + w = w ∈ D(R)−. Hence vkn = 0. Thus R has bounded index less than
or equal to kn.
(cid:3)
Corollary 2.6(3) extends [TLZ, Corollary 3.8]. In [ABP], the authors determine
several generalizations of the condition that a ring satisfies a polynomial identity.
With these generalizations they were able to extend classical theorems by Armen-
dariz and Steenberg, Fisher, Kaplansky, Martindale, Posner and Rowen. Two of
these generalizations are: (1) a ring R is an almost PI-ring if every prime factor
ring of R is a PI-ring; (2) R is an instrinsically PI-ring if every nonzero ideal
contains a nonzero PI-ideal of R.
12
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
Corollary 2.7. Let R ∈ Tn (n > 1). Then:
(1) D(R) satisfies a PI if and only if R does so.
(2) If D(R) is commutative, then R satisfies (xy − yx)n = 0 for all x, y ∈ R.
(3) D(R) is almost PI if and only if R is almost PI.
(4) If D(R) is intrinsically PI, then R is instrinsically PI.
Proof. (1) Since subrings of a PI-ring are PI-rings, if R is a PI-ring then so is D(R).
Conversely, assume D(R) is a PI-ring which satisfies the PI p. Then R satisfies pn.
(2) This part follows from (1).
(3) and (4). By Proposition 2.4 we have D(R)− E R, R/(D(R)−) ∼= D(R) and
(D(R)−)n = 0. Now (3) follows from [ABP, Theorem 1.6(i)], and (4) follows from
[ABP, Theorem 1.6(ii)].
(cid:3)
Let R be an n-by-n generalized matrix ring, and let
UT(R) and LT(R)
be the n-by-n upper and lower generalized triangular matrix rings, respectively,
formed from R. Our next result shows that elements of Tn are subdirect products
of generalized triangular matrix rings.
Proposition 2.8. Let R ∈ Tn (n > 1). Then there is a ring monomorphism
ψ : R → UT(R) × LT(R) such that R is a subdirect product of UT(R) and LT(R).
Proof. Let [mij] ∈ R. Define ψ([mij]) = ([aij], [bij]) where
aij = (cid:26) mij,
0,
for j ≥ i
elsewhere,
and
bij = (cid:26) mij,
0,
for i ≥ j
elsewhere.
A routine argument yields that ψ is a ring monomorphism and that R is a subdirect
product of UT(R) and LT(R).
(cid:3)
Definition 2.9. Let R be an n-by-n generalized matrix ring. Let Rla denote the
lower annihilating subring
R1
M12
rM21(M12) ∩ ℓM21(M12) R2
...
. . .
· · ·
. . .
. . .
M1n
...
Mn−1,n
rMn1(M1n) ∩ ℓMn1(M1n)
· · ·
rMn,n−1(Mn−1,n) ∩ ℓMn,n−1(Mn−1,n)
Rn
of R, and let Rua denote the upper annihilating subring
IDEMPOTENTS AND STRUCTURES OF RINGS
13
R1
rM12(M21) ∩ ℓM12(M21)
· · ·
rM1n(Mn1) ∩ ℓM1n(Mn1)
M21
...
Mn1
R2
. . .
· · ·
. . .
. . .
...
rMn−1,n(Mn,n−1) ∩ ℓMn−1,n(Mn,n−1)
Mn,n−1
Rn
of R.
Note that Rla and Rua are subrings of both R and R. Moreover, if R is the
n-by-n matrix ring over a ring A, then Rla and Rua are the n-by-n upper and
lower triangular matrix rings over A, respectively.
Example 2.10. Let A and B be rings. Let R = (cid:20) A × B A × {0}
Rla = (cid:20) A × B A × {0}
{0} × B A × B (cid:21), and Rua = (cid:20) A × B
A × B A × B (cid:21). Then
A × B A × B (cid:21).
{0}
Lemma 2.11. Let R be an n-by-n generalized matrix ring. Then Cen(R) =
Cen(R) = {[cij] ∈ R cij = 0 if i 6= j, cii ∈ Cen(Ri) for all i, and ciimij =
mijcjj for all mij ∈ Mij if i 6= j}.
Proof. First show the result holds when n = 2. For the n-by-n case, block the
matrix ring into a 2-by-2 generalized matrix ring and use induction.
(cid:3)
Let S be a subring of a ring R. It is well known (see [W, p. 26]) that the (S, S)-
bimodule structure of S and R is equivalent to the right T -module structure of S
and R, respectively, where T = S op ⊗Z S, with S op denoting the opposite ring
of S.
The next three results (2.12 - 2.14) indicate the transfer of significant informa-
tion between an n-by-n generalized matrix ring and certain subrings which are
maximal with respect to being in Tn.
In particular, Theorem 2.12 shows that for any ring R with a complete set of
i=1 (n > 1) there are subrings S containing {ei}n
orthogonal idempotents {ei}n
i=1
which are maximal with respect to Sπ being in Tn and ST is right essential in RT ,
where T = S op ⊗Z S. Moreover, this result and its consequences provide a con-
nection between the structure of an arbitrary generalized matrix ring and the
structure of rings in Tn (see Question B in the introduction).
Theorem 2.12. Let R be an n-by-n generalized matrix ring, and S denotes
either Rla or Rua.
(1) S is a subring of R maximal with respect to being in Tn.
(2) Let 0 6= y ∈ R. Then either 0 6= syEjj ∈ S or 0 6= Eiiyt ∈ S for some
s, t, Eii, Ejj ∈ S.
14
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
(3) Every nonzero (S, S)-bisubmodule of R has nonzero intersection with S.
Thus every nonzero ideal of R has nonzero intersection with S, and ST is
right essential in RT where T = S op ⊗Z S.
(4) Cen(R) = Cen(Rla) ∩ Cen(Rua) ⊆ Cen(D(R)).
(5)
U(R) = {u + x u ∈ U(D(R)) and x ∈ D(R)−}, and
U(S) = {u + y u ∈ U(D(R)) and y ∈ D(S)−} ⊆ U(R).
Proof. (1) Suppose that there is a subring Y of R such that S is properly contained
in Y . Assume S = Rla. Then there exists y ∈ Y with an entry yij for some i, j
with i > j such that yij 6∈ rMij (Mji) ∩ ℓMij (Mji). Then either yij 6∈ ℓMij (Mji), or
yij ∈ ℓMij (Mji) but yij 6∈ rMij (Mji).
If yij 6∈ ℓMij (Mji), there exists k ∈ Mji such that yijk 6= 0. Let t be the n-by-n
matrix with k in the (j, i)-position and zero elsewhere. Then t ∈ S ⊆ Y , and so
0 6= yt ∈ Y . However, since 0 6= yijk ∈ MijMji ∈ Ri, we have that Y 6∈ Tn.
If yij ∈ ℓMij (Mji) but yij 6∈ rMij (Mji), then there exsits h ∈ Mji such that hyij 6=
0. Let s be the n-by-n matrix with h in the (j, i)-position and zero elsewhere. Then
s ∈ S ⊆ Y , and so 0 6= sy ∈ Y . However, since 0 6= hyij ∈ MjiMij ∈ Rj, it follows
that Y 6∈ Tn.
Therefore S is a subring of R maximal with respect to being in Tn. The argument
when S = Rua is similar.
(2) Again let S = Rla and 0 6= y ∈ R. If y ∈ S, we are finished. So assume
y 6∈ S. Then as in part (1) there exists an entry yij of y for some i, j with i > j
such that yij 6∈ rMij (Mji) ∩ ℓMij (Mji). As in part (1), we obtain s, t ∈ S. Then
0 6= syEjj ∈ S or 0 6= Eiiyt ∈ S. The argument when S = Rua is similar.
(3) This part is a consequence of (2).
(4) This part follows from Lemma 2.11.
(5) By Proposition 2.4(1), D(R)− E R such that (D(R)−)n = 0. Let u ∈ D(R)
and x ∈ D(R)−. Then (u + x)(u−1 + x) = 1 + ux+ xu−1 + x2. But ux+ xu−1 + x2 ∈
D(R), hence (u + x)(u−1 + x) = w where w ∈ U(R). Therefore u + x ∈ U(R).
Now assume v ∈ U(R). Then there exist d ∈ D(R) and y ∈ D(R)− such
that v = d + y. So 1 = dv−1 + yv−1. Hence dv−1 = 1 − yv−1. Since yv−1 ∈
D(R)−, dv−1 ∈ U(R). So d ∈ U(R). Therefore U(R) = {u + x u ∈ U(D(R)) and
x ∈ D(R)−}.
The remainder of the proof is due to the above argument and the fact that S is
(cid:3)
a subring of R and R.
Note that if n = 2, then in Theorem 2.12(2), there is no need for the Eii and Ejj.
QUESTION: When is U(R) generated by U(Rla) ∪ U(Rua)?
Corollary 2.13. Let R be an n-by-n generalized matrix ring, S denotes Rla or Rua
and T = S op ⊗Z S. Then:
(1) S is maximal among subrings Y of R for which {Eii}n
(2) The sum of the minimal ideals of S equals Soc(ST ) = Soc(RT ).
(3) The uniform dimension of SSS equals the uniform dimension of SRS equals
i=1 ⊆ Pit(R).
the uniform dimension of ST equals the uniform dimension of RT .
IDEMPOTENTS AND STRUCTURES OF RINGS
Proof. (1) This part is a consequence of Theorems 2.2 and 2.12.
(2) and (3). These parts are consequences of Theorem 2.12(3).
15
(cid:3)
Our next result demonstrates that useful information can be transferred from
the diagonal rings Ri of a generalized matrix ring R to R itself via Theorems 1.16
and 2.12. Recall from [R2] and [CR] that an n-by-n (n > 1) matrix ring over a
strongly π-regular ring is not, in general, a strongly π-regular ring.
Corollary 2.14. Let R be an n-by-n generalized matrix ring, and S = Rla or Rua.
If D(R) is strongly π-regular, then for each 0 6= y ∈ R either:
(1) y ∈ S, in which case yn ∈ yn+1S ⊆ yn+1R for some positive integer n; or
(2) y 6∈ S, in which case either 0 6= syEjj ∈ S and (syEjj)m ∈ (syEjj)m+1S ⊆
(syEjj)m+1R, or 0 6= Eiiyv ∈ S and (Eiiyv)k ∈ (Eiiyv)k+1S ⊆ (Eiiyv)k+1R
for some s, v, Eii, Ejj ∈ S and positive integers k, m, n.
Proof. The proof follows from Theorems 1.16 and 2.12(2).
(cid:3)
Thus if D(R) is strongly π-regular, then R is “almost” strongly π-regular.
Next we introduce the notion of an ideal extending ring. The ideal extending
condition is shown to be a Morita invariant. Moreover, it is shown that important
classes of rings have this property. For example, the semiprime quasi-Baer rings
are ideal extending, so this insures that every semiprime ring has a hull which is
ideal extending (see Proposition 2.16). The class of semiprime quasi-Baer rings
includes the local multiplier C ∗-algebras which means that every C ∗-algebra can
be embedded into its local multiplier C ∗-algebra which is an ideal extending ring.
As applications of the results in 2.12 - 2.14 we show in 2.16 - 2.21 that the ideal
extending property transfers from a ring A to a certain type of overring of A which
is in Tn.
Let X and Y be both left or both right ideals of a ring R with X ⊆ Y . Then
X is ideal essential in Y if for each 0 6= I E R such that I ⊆ Y , then 0 6= X ∩ I.
Note that if R is a semiprime ring and X, Y E R with X ⊆ Y , then X is ideal
essential in Y if and only if X is right or left essential in Y .
Definition 2.15. We say R is ideal extending if for each X E R there is an
e ∈ B(R) such that X is ideal essential in eR.
Note that every nonzero ideal of R is ideal essential in R if and only if B(R) =
{0, 1} and R is ideal extending. Some immediate examples of ideal extending
rings are: R is a prime ring, R is an Abelian (i.e., every idempotent is central)
right extending ring (e.g., R is a direct sum of commutative uniform rings (see
[DHSW])), or R = (cid:20) A M
0 A (cid:21) where A is a prime ring and M E A.
The next result shows that the class of ideal extending rings is quite extensive.
See [BPR1] or [BPR2] for undefined terminology.
Proposition 2.16. Assume R is a semiprime ring. Then:
(1) R is ideal extending if and only if R is quasi-Baer if and only if RR is
FI-extending.
16
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
(2) R has an ideal extending hull.
Proof. (1) See [BMR, Theorem 4.7] or [BPR2, Theorem 3.2.37].
(2) This part follows from (1) and [BPR1, Theorem 3.3] or [BPR2, Theo-
(cid:3)
rem 8.3.17].
From Proposition 2.16 it follows that every von Neumann algebra and every
local multiplier algebra of a C ∗-algebra are ideal extending as rings (see [K] and
[BPR1, pp. 345-347] or [BPR2, pp. 380-407]).
Theorem 2.17. (1) Let {Rii ∈ I} be a set of rings. Then Πi∈IRi is ideal
extending if and only if each Ri is so.
(2) The ideal extending property is a Morita invariant.
Proof. (1) The proof of this part is routine.
(2) Assume that R is ideal extending. Then a straightforward argument shows
that R is ideal extending if and only if the ring of n-by-n matrices over R is ideal
extending. Let e be a full idempotent of R (i.e. ReR = R) and 0 6= K E eRe.
Then RKR is ideal essential in cR for some c ∈ B(R). Hence K ⊆ ec(eRe), where
ec ∈ B(eRe). Let 0 6= X E eRe such that X ⊆ ec(eRe). Then RXR ⊆ cR, so
0 6= Y = RXR ∩ RKR. If eY e 6= 0, there exists y ∈ Y such that 0 6= eye =
Σrαxαsα = Σtβkβvβ, where rα, sα, tβ, vβ ∈ R, xα ∈ X and kβ ∈ K. So xα = exαe
and kβ = ekβe. Hence 0 6= eye = Σerα(exαe)sαe = Σetβ(ekβe)vβe ∈ X ∩ K.
Now assume eY e = 0. Since e is full, 1 = Σajebj for some aj, bj ∈ R. Let
0 6= w ∈ Y. Then w = 1w1 = (Σajebj)w(Σajebj) . So there exists j1, j2 such
that aj1ebj1waj2ebj2 6= 0, otherwise w = 0, a contradiction. Hence ebj1waj2e 6= 0,
contrary to eY e = 0. Thus eRe is ideal extending. By [L, Corollary 18.35], the
ideal extending property is a Morita invariant.
(cid:3)
Proposition 2.18. Let R be an n-by-n generalized matrix ring and S = Rla.
(1) If SXS ≤ SRS and X ∩ S is ideal essential in eS, for some e ∈ Sr(S), then
X essential in eR as an (S, S)-bisubmodule.
(2) If S is an ideal extending ring, then for each X E R there is an e ∈ B(S)
such that X is ideal essential in eR.
Proof. (1) Assume (1 − e)X 6= 0. Since 1 − e ∈ Sℓ(S), (1 − e)X is an (S, S)-
bisubmodule of R. By Theorem 2.12(3), 0 6= (1 − e)X ∩ S ⊆ X ∩ S ⊆ eS, a
contradiction. Then X ⊆ eR. Let 0 6= SYS ≤ SRS such that Y ⊆ eR and
Y ∩ X = 0. Hence 0 6= Y ∩ S ⊆ eS and Y ∩ S E S, a contradiction.
(2) This part is a consequence of (1).
(cid:3)
Example 2.19. (1) This example illustrates Proposition 2.18(1).
Let R = (cid:20) Z × Z4 Z × 2Z4
Take X = (cid:20) {0} × {0} {0} × {0}
Z × {0} Z × Z (cid:21), and let S = Rla. Then S = (cid:20) Z × Z4
{0} × {0} {0} × 2Z (cid:21) E R and e = (cid:20) (0, 0)
(0, 0)
Z × 2Z4
{0} × {0} Z × Z (cid:21).
(0, 1) (cid:21). Then
(0, 0)
e ∈ Sr(S) and SXS is essential as an (S, S)-bisubmodule of eR. Note that
e 6∈ B(S).
IDEMPOTENTS AND STRUCTURES OF RINGS
17
(2) This example shows that in Proposition 2.18(2), S cannot be replaced
tative selfinjective ring, hence it is ideal extending. However B(R) = {0, 1}, but
by D(R). Let R = (cid:20) Z4 2Z4
(cid:20) 2Z4 0
0 (cid:21) and (cid:20) 0 2Z4
Z4 (cid:21) (note that R = S = Rla). Then D(R) is a commu-
0 (cid:21) are ideals of R whose intersection is zero. Therefore
0
0
0
R is not ideal extending.
Lemma 2.20. If A is an ideal extending ring, then R is ideal extending where R
is the n-by-n upper triangular matrix ring over A.
Proof. Let X E R. Then
X =
X11 X12
0 X22
...
. . .
· · ·
0
· · · X1n
X2n
...
. . .
0 Xnn
,
where each Xij E A, Xii ⊆ Xi,i+1 ⊆ · · · ⊆ Xin and Xii ⊆ Xi−1,i ⊆ · · · ⊆ X1i for all
1 ≤ i ≤ n. Observe e ∈ B(R) if and only if e = c1R, for some c ∈ B(A). There
exists f ∈ B(A) such that X1n is ideal essential in f A. Then a routine argument
shows that X is ideal essential in (f 1R)R.
(cid:3)
The following corollary is an application of Proposition 2.18 (hence of Theo-
rem 2.12).
Corollary 2.21. Let A be a ring and R the n-by-n generalized matrix ring of the
form
R =
A
A
A
X21 A
A
...
...
. . .
· · · Xn,n−1 A
Xn1
· · ·
· · ·
. . .
,
where Xij = A for i ≤ j, Xij E A for j < i, Xj1 ⊆ Xj2 ⊆ · · · ⊆ Xjn and
Xni ⊆ Xn−1,i ⊆ · · · ⊆ X1i, for all 1 ≤ i ≤ n and 1 ≤ j ≤ n. Then A is ideal
extending if and only if R is ideal extending.
Proof. (⇒) Assume A is ideal extending. Observe that Rla is the n-by-n upper
triangular matrix ring over A. By Lemma 2.20, Rla is ideal extending. From
Proposition 2.18(2), R is ideal extending.
(⇐) Assume R is ideal extending. Let X E A and Y the set of n-by-n matrices
over X. Then Y E R. So there exists e ∈ B(R) such that Y is ideal essential in
eR and e = c1R where c ∈ B(A). Then X is ideal essential in cA. Therefore A is
ideal extending.
(cid:3)
Assume R is ring isomorphic to Rm and to Rn, where Rm is an m-by-m gener-
alized matrix ring, and Rn is an n-by-n generalized matrix ring, with 0 < m < n.
One may naturally ask:
(1) If Rm ∈ Tm, must Rn ∈ Tn?
(2) If Rn ∈ Tn, must Rm ∈ Tm?
18
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
The following example shows that, in general, neither question has an affirmative
answer.
Example 2.22. Let A be a ring.
(1) Let
R =
A A A
A A A
0
0 A
.
Then R 6∈ T3, because E22 6∈ Pit(R) by Theorem 2.2.
A A (cid:21) , M12 = (cid:20) A
A (cid:21), M21 = [0 0] and R2 = A. Then
Let R1 = (cid:20) A A
R ∼= (cid:20) R1 M21
M21 R2 (cid:21) ∈ T2.
(2) Let
R =
.
A A A A
0 A 0
0
0 A A A
0 A 0 A
0 (cid:21), M21 = (cid:20) 0 A
M21 R2 (cid:21) 6∈ T2, since M12M21 6= 0.
0
Then R ∈ T4.
0 A (cid:21) , M12 = (cid:20) A A
Let R1 = (cid:20) A A
0 A (cid:21). Then R ∼= (cid:20) R1 M12
(cid:20) A A
0 A (cid:21) and R2 =
Proposition 2.23. Let R be a ring with a complete set {ei}n
onal idempotents. If {ei}n
block form is in Tm where m ≤ n.
i=1 (n > 1) of orthog-
i=1 ⊆ Pt(R), then any partition of Rπ into an m-by-m
Proof. The proof is straightforward.
(cid:3)
Note that in Proposition 2.23 if n = 3, then we can replace Pt(R) with Pit(R).
QUESTION: Let R be a ring with a complete set {ei}n
i=1 (n > 1) of orthogonal
idempotents. What are necessary and sufficient conditions so that any partition
of Rπ into m-by-m block form is in Tm for 1 < m ≤ n?
3. n-Peirce Rings
Definition 3.1. A ring R is called a 1-Peirce ring if Pt(R) = {0, 1}, with 0 6= 1.
Inductively, for a natural number n > 1, a ring R is called an n-Peirce ring if
there is an e ∈ Pt(R) such that eRe is an m-Peirce ring for some 1 ≤ m < n and
(1 − e)R(1 − e) is an (n − m)-Peirce ring.
Example 3.2. (1) If RR is indecomposable or R is prime, then R is 1-Peirce. In
fact, if R is semiprime then R is 1-Peirce if and only if R is indecomposable (as a
ring).
IDEMPOTENTS AND STRUCTURES OF RINGS
19
(2) If R is an n-by-n generalized upper (lower) triangular matrix ring with
1-Peirce diagonal rings, then R is a n-Pierce ring.
(3) If R has a complete set of n orthogonal primitive idempotents which are
Peirce trivial, then R is an n-Peirce ring (e.g., see Example 2.5(3)).
(4) Example 2.5(2) is a 3-Peirce ring that has a complete set of three primitive
idempotents which are inner Peirce trivial but not all of them are Peirce trivial.
(5) Let I be be an infinite index set, for each i ∈ I let Ai be a ring with only
trivial idempotents, and A = Πi∈IAi. Assume R = (cid:20) A X
0 X (cid:21) , where X is a
nonzero ideal of A. Then R is in T2, but R is not an n-Peirce ring for any positive
integer n.
From Example 3.2(5) and Theorem 3.7, we see that the class of n-Peirce n-
by-n generalized matrix rings is a proper subclass of the class Tn for n > 1.
Also, due to the symmetry of Peirce idempotents (i.e., e ∈ Pt(R) if and only if
1 − e ∈ Pt(R)) the class of n-Peirce rings exhibits better behavior than Tn with
respect to finiteness conditions.
Theorem 3.3. Let R be a ring. Then either:
(1) R is a 1-Peirce ring;
(2) R is an n-Peirce ring for n > 1, and for each k ∈ Z+ with 1 < k ≤ n there
exists a complete set of orthogonal idempotents, {ei}k
i=1, such that Rπ ∈ Tk; or
(3) for each integer k with k > 1, there exists a complete set of orthogonal
idempotents, {ei}k
i=1, such that Rπ ∈ Tk.
Proof. Observe that {0, 1} ⊆ Pt(R), and {0, 1} = Pt(R) if and only if R is
If {0, 1} 6= Pt(R) (i.e., R is not 1-Peirce), then there exists
a 1-Peirce ring.
{e1, 1 − e1} ⊆ Pt(R) \ {0, 1}. By Theorem 2.2, Rπ ∈ T2. Now if at least one
of e1Re1 or (1 − e1)R(1 − e1) is not 1-Peirce, say e1Re1, then there exists e2 ∈
Pt(e1Re1) \ {0, e1}. By Lemma 1.8, {e2, e1 − e2} ⊆ Pt(e1Re1) ⊆ Pit(e1Re1) =
e1Re1 ∩ Pit(R) ⊆ Pit(R). Hence {e2, e1 − e2, 1 − e1} is a complete set of orthogonal
idempotents contained in Pit(R). By Theorem 2.2, Rπ ∈ T3. If at least one of
e2Re2, (e1 − e2)R(e1 − e2), or (1 − e1)R(1 − e1) is not 1-Peirce, then either this
inductive process will terminate in n steps for some n ∈ Z+ yielding condition (2)
or it will continue indefinitely yielding condition (3).
(cid:3)
Note that in Example 2.22(2), R has a 2-by-2 block form which is not in T2.
Observe that one can show that E11 ∈ Pt(R), so R is not 1-Peirce. Surprisingly,
Theorem 3.3 predicts that there is a 2-by-2 block form for R which is in T2. Indeed,
R can be partitioned into another 2-by-2 block form which is in T2 by taking R1 to
be a 1-by-1 matrix and R2 to be a 3-by-3 matrix (i.e., this corresponds to taking
e1 = E11 and 1 − e1 = E22 + E33 + E44 in the proof of Theorem 3.3).
Proposition 3.4. Let 0 6= e = e2 ∈ R such that eRe is a 1-Peirce ring, c ∈ Pt(R),
c1 ∈ Pt(cRc), 0 6= cec and c1ec1 6= 0. Then:
(1) ece = e = ec1e.
(2) cecRcec is a 1-Peirce ring.
20
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
Proof. (1) By Lemma 1.7, 0 6= cec = (cec)2 = c(ece)c. Thus ece 6= 0. From
Lemma 1.8(1), c1 ∈ Pit(R). So, by Lemma 1.7, c1ec1 = (c1ec1)2 = c1(ec1e)c1.
Hence ec1e 6= 0.
Claim 1. ece ∈ Pit(eRe).
Let x, y ∈ eRe. By Lemma 1.7, ece = (ece)2. Consider (ece)x(e − ece)y(ece) =
e(c[exe(1 − c)eye]c)e = e0e = 0, because c ∈ Pt(R) ⊆ Pit(R). Thus ece ∈
Pit(eRe).
Claim 2. ece ∈ Pt(eRe).
Consider (e − ece)xecey(e − ece) = e[(1 − c)execeye(1 − c)]e = e0e = 0, because
c ∈ Pot(R). Thus ece ∈ Pot(eRe). Hence 0 6= ece ∈ Pt(eRe). Since eRe is
1-Peirce, ece = e.
Claim 3. ec1e ∈ Pit(eRe).
From above, c1 ∈ Pit(R). Let x, y ∈ eRe. Then ec1ex(e − ec1e)yec1e =
e[c1xe(1 − c1)eyc1]e = 0. Hence ec1e ∈ Pit(eRe).
Claim 4. ec1e ∈ Pt(eRe).
Since e = ece, we have
(e − ec1e)xec1ey(e − ec1e) = (ece − ec1e)xc1y(ece − ec1e)
= e(c − c1)exc1ye(c − c1)e
= e[(c − c1)(cxc)c1(cyc)(c − c1)]e = 0,
because c1 ∈ Pt(cRc) ⊆ Pot(cRc). Thus ec1e ∈ Pt(eRe). Since eRe is a 1-Peirce
ring, ec1e = e.
(2) Let 0 6= f ∈ Pt(cecRcec). Observe 0 6= f = c[ecf ce]c = c[ef e]c since
c ∈ Pt(R). So ef e 6= 0.
Claim 5. ef e = (ef e)2.
Observe f = (cec)f = c(cecf ) = cf . Similarly, f = f c. Consider ef e =
e(f cecf )e = ef ef e = (ef e)2.
Claim 6. ef e ∈ Pit(eRe).
Let x, y ∈ eRe. By (1) e = ece, so
ef ex(e − ef e)yef e = ef ex(ece − ef e)yef e = ef exe(c − f )eyef e
= e(f c)exe(c − f )eye(cf )e
= ef [(cec)x(cec)(cec)(c − f )(cec)y(cec)]f e
= ef [(cec)x(cec)f (cec)(c − f )(cec)y(cec)]f e
= ef [(cec)x(cec)f (c − f )(cec)y(cec)]f e
= ef [(cec)x(cec)0(cec)y(cec)]f e
= 0,
since f ∈ Pt(cecRcec) ⊆ Pit(cecRcec).
Claim 7. ef e ∈ Pt(eRe).
Consider
(e − ef e)xef ey(e − ef e) = (ece − ef e)xf y(ece − ef e)
= e(c − f )exf ye(c − f )e
= ece(c − f )xf y(c − f )ece
= ecec(c − f )(cecxcec)f (cecycec)(c − f )ece
= e(cec − f )[(cecxcec)f (cecycec)](cec − f )e
= 0,
IDEMPOTENTS AND STRUCTURES OF RINGS
21
since f ∈ Pot(cecRcec). Thus ef e ∈ Pt(eRe). Hence ef e = e. So 0 6= f =
cecf cec = c(ef e)c = cec. Therefore cecRcec is a 1-Peirce ring.
(cid:3)
Observe that in Proposition 3.4 the conclusions ece = e and cecRcec is a 1-
Peirce ring do not need the conditions c1 ∈ Pt(cRc) and c1ec1 6= 0. Also, this
result can be extended under related hypotheses (e.g. e primitive and c ∈ Pit(R)).
Corollary 3.5. Let 0 6= e = e2 ∈ R such that eRe is a 1-Peirce ring and c ∈ Pt(R).
The following conditions are equivalent:
(1) cec 6= 0.
(2) ece = e.
(3) (1 − c)e(1 − c) = 0.
Proof. (1) ⇒ (2) This implication follows from Proposition 3.4.
(2) ⇒ (3) Since c ∈ Pt(R), e(1 − c)e = ece(1 − c)ece = 0. By Lemma 1.7(1),
(1 − c)e(1 − c) = [(1 − c)e(1 − c)]2 = (1 − c)e(1 − c)e(1 − c) = 0.
(3) ⇒ (1) Assume (1 − c)e(1 − c) = 0 and cec = 0. Then
e = (c + (1 − c))e(c + (1 − c))
= cec + ce(1 − c) + (1 − c)ec + (1 − c)e(1 − c)
= ce(1 − c) + (1 − c)ec.
So e = e2 = (ce(1 − c) + (1 − c)ec)2 = ce(1 − c)ce(1 − c) + ce(1 − c)ec + (1 −
(cid:3)
c)ece(1 − c) + (1 − c)ec(1 − c)ec = 0, a contradiction.
Corollary 3.6. Let {ei}n
such that each eiRei is a 1-Peirce ring and 0 6= c ∈ Pt(R). Then:
i=1 be a complete set of nonzero orthogonal idempotents
(1) c = Pi∈J1 ceic and 1 − c = Pi∈J2(1 − c)ei(1 − c), where ceic 6= 0 for all
i ∈ J1 and (1 − c)ei(1 − c) 6= 0 for all i ∈ J2.
(2) J1 + J2 = n.
(3) {ceic i ∈ J1} ∪ {(1 − c)ei(1 − c) i ∈ J2} is a complete set of orthogonal
idempotents, where each ceicRceic and each (1−c)ei(1−c)R(1−c)ei(1−c)
is a 1-Peirce ring.
c)ei(1 − c), where J1 ∪ J2 ⊆ {1, . . . , n}.
(2) This part follows from Corollary 3.5.
Proof. (1) c = c1c = cPi∈J1 eic = Pi∈J1 ceic, and similarly, 1 − c = Pi∈J2(1 −
(3) Note that 1 = c+1−c = Pi∈J1 ceic+Pi∈J2(1−c)ei(1−c). Also, (ceic)cejc =
ceiejc = 0 for all i 6= j, since c ∈ Pt(R). Similarly, [(1−c)ei(1−c)][(1−c)ej (1−c)] =
0 for all i 6= j. Moreover [(1 − c)ei(1 − c)][cejc] = 0 = [ceic][(1 − c)ej(1 − c)]
for all i, j. By Lemma 1.7, ceic and (1 − c)ei(1 − c) are idempotents for all i.
From Proposition 3.4(2), each ceicRceic and each (1 − c)ei(1 − c)R(1 − c)ei(1 − c)
is a 1-Peirce ring.
Theorem 3.7. (1) If R is an n-Peirce ring (n > 1), then there is a complete set of
orthogonal idempotents {ei}n
i=1 ⊆ Pit(R) (hence Rπ ∈ Tn) such that every eiRei
is a 1-Peirce ring.
(cid:3)
22
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
(2) If a ring R has a complete set {ei}n
i=1 of orthogonal idempotents for some
n ≥ 2 such that every eiRei is a 1-Peirce ring, then R is a k-Peirce ring for some
1 ≤ k ≤ n.
Proof. (1) We use strong induction on n. First, let R be a 2-Peirce ring. Then
there is an e ∈ Pt(R) such that eRe and (1 − e)R(1 − e) are 1-Peirce rings, and
{e, 1 − e} is a complete set of orthogonal idempotents, with e, 1 − e ∈ Pit(R).
i=1 ⊆ Pit(cRc) and {fj}n+1−k
Next, consider a fixed n ≥ 2 and assume that for each k, 2 ≤ k ≤ n, if R is
a k-Peirce ring, then there is a complete set of orthogonal idempotents {ei}k
i=1 ⊆
Pit(R) such that every eiRei is a 1-Peirce ring. Now let R be an (n+1)-Peirce ring.
Then there is a c ∈ Pt(R) such that cRc is a k-Peirce ring for some k, 1 ≤ k < n+1,
and (1 − c)R(1 − c) is an (n + 1 − k)-Peirce ring. Since k ≤ n and n + 1 − k ≤ n,
and assuming for the moment that 2 ≤ k and 2 ≤ n + 1 − k, the induction
hypothesis guarantees the existence of complete sets of orthogonal idempotents
{ei}k
j=1 ⊆ Pit((1 − c)R(1 − c)), such that eicRcei and
fj(1 − c)R(1 − c)fj are 1-Peirce rings for every i and j. Since c, 1 − c ∈ Pit(R),
it follows from Lemma 1.8 that {ei}k
j=1 ⊆ Pit(R). Since (cRc)((1 −
c)R(1 − c)) = 0, we conclude that {ei}k
is an orthogonal set of
fj = c + (1 − c) = 1, and
eicRcei = eiRei (since ei ∈ cRc) and fj(1 − c)R(1 − c)fj = fjRfj for every i and j.
Finally, we consider the case k = 1 or n + 1 − k = 1. Without loss of generality,
let k = 1, i.e., c ∈ Pit(R), cRc is a 1-Peirce ring and (1 − c)R(1 − c) is an n-Peirce
ring. Then we can proceed as in the previous paragraph with c ∈ Pit(R) and a
complete set of orthogonal idempotents {fj}n
j=1 ⊆ Pit((1 − c)R(1 − c)), and then
the set {c, f1, . . . , fn} is a complete set of orthogonal idempotents in Pit(R) such
that cRc and fjRfj are 1-Peirce rings for all j.
idempotents in Pit(R). Moreover, Pk
i=1 ei + Pn+1−k
i=1, {fj}n+1−k
i=1 ∪ {fj}n+1−k
j=1
j=1
(2) We again use strong induction on n. Let R have a complete set of orthogonal
idempotents {e1, e2} such that each eiRei is a 1-Peirce ring. If R is a 1-Peirce
ring, then we are done, since 1 ≤ 2. Otherwise there is a c ∈ Pt(R) such that
c 6∈ {0, 1}. Hence 1−c ∈ Pt(R) and 1−c 6∈ {0, 1}. From Corollary 3.6, c = ceic for
i ∈ {1, 2}. Without loss of generality, assume i = 1. Then, again by Corollary 3.6,
cRc = ce1cRce1c, (1 − c)R(1 − c) = (1 − c)e2(1 − c)R(1 − c)e2(1 − c), and cRc and
(1 − c)R(1 − c) are 1-Peirce rings. Therefore R is a 2-Peirce ring.
Next assume that the result holds for a fixed n ≥ 2. Let R be a ring having a
complete set of orthogonal idempotents {ei}n+1
i=1 such that each eiRei is a 1-Peirce
ring. If R is a 1-Peirce ring, we are done. Otherwise there is a c ∈ Pt(R) such that
c 6∈ {0, 1}. Hence 1 −c ∈ Pt(R) and 1 −c 6∈ {0, 1}. From Corollary 3.6, there exist
J1, J2 ⊆ {1, . . . , n} and complete sets of orthogonal idempotents {ceic i ∈ J1}
and {(1 − c)ei(1 − c) i ∈ J2} for cRc and (1 − c)R(1 − c), respectively. From the
induction hypothesis, there exist positive integers k1 and k2 such that 1 ≤ k1 ≤ J1
and 1 ≤ k2 ≤ J2 such that cRc is a k1-Peirce ring and (1 − c)R(1 − c) is a k2-
Peirce ring. Since J1 + J2 = n + 1, then R is k-Peirce where k = k1 + k2 and
1 ≤ k ≤ n + 1.
(cid:3)
IDEMPOTENTS AND STRUCTURES OF RINGS
23
Corollary 3.8. Let {ei}n
i=1 ⊆ Pit(R) be a complete set of orthogonal idempotents
such that each eiRei is a ki-Peirce ring for some positive integers ki. Then R is a
k-Peirce ring for some 1 ≤ k ≤ Pn
Proof. This result follows from Theorem 3.7.
i=1 ki.
(cid:3)
From Theorem 3.7, R is an n-Peirce (n > 1) generalized matrix ring implies
that R ∈ Tn; and if R has a complete set of n orthogonal primitive idempotents,
then R is k-Peirce for some k with 1 ≤ k ≤ n. Thus it is natural to ask: If R
has a complete set of orthogonal primitive idempotents {ei}n
i=1 ⊆ Pit(R) (hence
Rπ ∈ Tn), must R be n-Peirce? Observe that for n = 2, the question has an
affirmative answer. Our next example provides a negative answer, in general.
Example 3.9. Let
R =
A X 2 X
X A X 2
X 2 X A
,
where A is a ring such that 0 and 1 are the only idempotents of A; and 0 6= X E A
such that X 6= X 2, X 2 6= X 3, and X 3 = 0. Then R ∈ T3, but R is a 1-Peirce ring.
One can construct such rings by letting B be a commutative ring, 0 6= P a prime
ideal of B such that P 6= P 2 and P 2 6= P 3. Then take A = B/P 3 and X = P/P 3.
In particular, let B = F [x] where F is a field and P = xF [x].
Since X 2X = XX 2 = 0, Corollary 2.3 yields R ∈ T3. To show that R is a
1-Peirce ring, we first characterize all nontrivial idempotents of R. Let α ∈ R
such that α 6= 0 and α 6= 1. Then α = α2 if and only if
α =
e1 m12 m13
m21
e2 m23
e3
m31 m32
where m12, m23, m31 ∈ X 2, m13, m21, m32 ∈ X, ei ∈ {0, 1}, and the following
equations are satisfied:
e1m12 + m12e2 + m13m32 = m12
= m13
e1m13 + m13e3
m21e1 + e2m21
= m21
e2m23 + m23e3 + m21m13 = m23
m31e1 + e3m31 + m32m21 = m31
m32e2 + e3m23
= m32.
From the above conditions α must have one of the following six forms:
(i)
(ii)
1 m12 m13
0 m23
0
0
m21
m31
0 m12
m21
m31 m32
0
1 m23
0
with m21m13 = m23;
with m32m21 = m31;
24
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
(iii)
(iv)
(v)
(vi)
0 m12 m13
0 m23
0
1
m31 m32
1 m12 m13
1 m23
0
0
m31 m32
1 m12
m21
m31 m32
0
0 m23
1
0 m12 m13
1 m23
1
0
m21
m31
with m13m32 = m12;
with m13m32 = −m12;
with m32m21 = −m31;
with m21m13 = −m23.
Now assume R is not a 1-Peirce ring. Then there exist c, 1 − c ∈ Pt(R) such
that 0 6= c and c 6= 1.
Then either c has a form of type (i), (ii), or (iii); or 1 − c has such a form.
Without loss of generality, assume c has a form of type (i), (ii) or (iii). Then
1 − c has a form of type (iv), (v), or (vi). We show that no matrix of type (iv),
(v) or (vi) is in Pot(R). Hence 1 − c 6∈ Pt(R), a contradition. Therefore R is a
1-Peirce ring.
Since X 2 6= 0, there exist x, y ∈ X such that 0 6= xy. Observe:
0 m12 m13
1 m23
1
0
m21
m31
= xyE23 6= 0;
1 m12
m21
m31 m32
0
0 m23
1
= xyE31 6= 0;
1 m12 m13
0
1 m23
0
m31 m32
= xyE12 6= 0.
xE21
xE32
xE13
1
−m21
−m31
−m12 −m13
−m23
0
0
0
0
−m23
0
0
−m12
−m21
−m31 −m32
1
0
0
−m12 −m13
−m23
0
−m31 −m32
1
yE13
yE21
yE32
0 m12 m13
1 m23
1
0
m21
m31
1 m12
m12
m13 m32
0
0 m23
1
1 m12 m13
0
1 m23
0
m31 m32
Lemma 3.10. Let 0 6= c = c2 = R and e ∈ Pit(cRc) such that e 6= c. Then
ReR ( RcR.
Proof. Observe that {e, c − e} is set of orthogonal idempotents. Clearly, ReR ⊆
RcR. Assume that ReR = RcR. Then c = P riesi. So c − e = (P riesi) − e.
IDEMPOTENTS AND STRUCTURES OF RINGS
25
Then
since
c − e = (c − e)2 = [(P riesi) − e](c − e)
= (P riesi)(c − e)
= P riesi(c − e)
= (P riesi(c − e))2
= Pj Pi riesi(c − e)rjesj(c − e)
= 0,
ri[esi(c − e)rje]sj(c − e) = ri[esie(c − e)rje]sj(c − e)
= ri0sj(c − e)
= 0,
because e ∈ Pit(cRc). However, this is a contradiction, since e 6= c. Therefore
ReR ( RcR.
(cid:3)
Theorem 3.11. Assume R has DCC on {ReR e ∈ Pt(R)}. Then R is an
n-Peirce ring for some n ∈ Z+.
Proof. Assume R has DCC on {ReR e ⊆ Pt(R)}, but R is a not an n-Peirce
ring for any n ∈ Z+. Observe that Pt(R) 6= {0, 1}. So let 0 6= c1 ∈ Pt(R)
be such that c1 6= 1. Then c1Rc1 is not an n-Peirce ring for any n ∈ Z+, or
(1−c1)R(1−c1) is not an n-Peirce ring for any n ∈ Z+. Without loss of generality,
say c1Rc1 is not an n-Peirce ring for any n ∈ Z+. Then Pt(c1Rc1) 6= {0, c1}. So let
0 6= c2 ∈ Pt(c1Rc1) be such that c2 6= c1. Then c2Rc2 is not an n-Peirce ring for
any n ∈ Z+, or (c1 − c2)R(c1 − c2) is not an n-Peirce ring for any n ∈ Z+. Without
loss of generality, say c2Rc2 is not an n-Peirce ring for any n ∈ Z+. By Lemma 1.8,
c1, c2 ∈ Pit(R). From Lemma 3.10, R ) Rc1R ( Rc2R. We can continue this
process indefinitely, which contradicts the DCC on {ReR e ∈ Pit(R)}. Therefore
R is an n-Peirce ring for some n ∈ Z+.
(cid:3)
Proposition 3.12. If {b1, . . . , bn} ⊆ Pit(R) is a complete set of nonzero orthog-
onal idempotents such that Pit(biRbi) = {0, bi}, then {ReR e ∈ Pit(R)} ≤ 2n.
i=1 bi) e (Pn
if biebi = bi. Let J ⊆ {1, . . . , n} such that biebi
Proof. Let 0 6= e = e2 ∈ Pit(R). By Lemma 1.8, each biebi ∈ Pit(R). Ob-
j=1 biebj, and biebi 6= 0 if and only
6= 0 if and only if i ∈ J.
j=1 ebiebje =
serve that e = (Pn
i=1 bi) = Pn
Since e ∈ Pit(R), e = e(e)e = e(cid:16)Pn
Pn
i=1Pj=1 ebibje = Pi∈J ebie. Then ReR ⊆ Pi∈J RbiR.
Observe that e − Pi6=j biebj = Pi∈J biebi = Pi∈J bi. Let k ∈ J, then bke −
bkebj = bk. Hence RbkR ⊆ ReR. So Pi∈J RbiR ⊆ ReR. Therefore ReR =
Pi∈J RbiR. Since J ⊆ {1, . . . , n}, {ReR e ∈ Pit(R)} ≤ {Pi∈K RbiR K ⊆
{1, . . . , n}} = {K K ⊆ {1, . . . , n}} = 2n, where Pi∈K RbiR corresponds to {0}
i=1Pn
i=1Pn
j=1 biebj(cid:17) e = Pn
i=1Pn
when K = ∅.
(cid:3)
As an illustration and application of several of our previous results, we provide
the following lemma and proposition. Recall that a ring R is ”quasi-Baer” if for
each X E R there is an e = e2 ∈ R such that rR(X) = eR. See [BPR2] and [C]
for further details on the class of quasi-Baer rings.
26
P.N. ´ANH, G.F. BIRKENMEIER, AND L. VAN WYK
Lemma 3.13. R is a prime ring if and only if R is quasi-Baer and a 1-Peirce ring.
Proof. Assume R is prime. From [BHKP, Lemma 4.2] or [BPR2, Proposition 3.2.5],
R is quasi-Baer. From Corollary 1.6, R is a 1-Peirce ring. Conversely, assume
xRy = 0 for some x, y ∈ R with x 6= 0. Then y ∈ rR(xR) = rR(RxR) = eR for
some e = e2 ∈ R. Since rR(xR) is an ideal of R, then e ∈ Sℓ(R). By Proposi-
tion 1.10(1), e = 0. Hence y = 0, so R is prime.
(cid:3)
Proposition 3.14. Assume that R is a quasi-Baer ring. If {e1, . . . , en} is a com-
plete set of orthogonal inner Peirce trivial idempotents and each eiRei is a 1-Peirce
ring, then R is a k-Peirce ring for some 1 ≤ k ≤ n, Rπ ∈ Tn and each eiRei is a
prime ring.
Proof. This proof follows from Theorem 3.7(2), Theorem 2.2, Lemma 3.13, and
[C, Lemma 2].
(cid:3)
For example, any quasi-Baer ring with a complete set of orthogonal primitive
idempotents (e.g., a right hereditary right Noetherian ring) satisfies the hypothesis
of Proposition 3.14.
In a sequel to this paper, we further investigate the properties and structure of
the class of n-Peirce rings.
Acknowledgements:
1. Proposition 2.8 was suggested by Arturo Magidin.
2. Our attention was drawn to [P] by Lance Small.
3. The first author was supported by the Hungarian National Foundation for Sci-
entific Research under Grants no. K-101515 and Institute of Mathematics, Hanoi,
Vietnam.
4. The third 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.
References
[ABP] E.P. Armendariz, G.F. Birkenmeier and J.K. Park, Ideal intrinsic exten-
sions with connections to PI-rings, J. Pure Appl. Algebra 213 (2009), 1756-
1776. Corrigendum 215 (2011), 99-100.
[AvW1] P.N. ´Anh and L. van Wyk, Automorphism groups of generalized triangular
matrix rings, Linear Algebra Appl. 434 (2011), 1018-1025.
[AvW2] P.N. ´Anh and L. van Wyk, Isomorphisms between strongly triangular ma-
trix rings, Linear Algebra Appl. 438 (2013), 4374-4381.
[BHKP] G.F. Birkenmeier, H.E. Heatherly, J.Y. Kim and J.K. Park, Triangular
matrix representations, J. Algebra 230 (2000), 558-595.
[BMR] G.F. Birkenmeier, B.J. Muller and S.T. Rizvi, Modules in which every fully
invariant submodule is essential in a direct summand, Comm. Algebra 30
(2002), 1395-1415.
IDEMPOTENTS AND STRUCTURES OF RINGS
27
[BPR1] G.F. Birkenmeier, J.K. Park and S.T. Rizvi, Hulls of semiprime rings with
applications to C ∗-algebras, J. Algebra 322 (2009), 327-352.
[BPR2] G. Birkenmeier, J.K. Park, S.T. Rizvi, Extensions of Rings and Modules,
Birkhauser/Springer, New York, 2013.
[C] W.E. Clark, Twisted matrix units semigroup algebras, Duke Math. J. 34
(1967), 417-423.
[CR] F. Cedo and L.H. Rowen, Addendum to “Examples of semiperfect rings”,
Israel J. Math. 107 (1998), 343-348.
[DHSW] N.V. Dung, D.V. Huynh, P.F. Smith and R. Wisbauer, Extending Modules,
Longman, Harlow, 1994.
[FS] J.W. Fisher and R.L. Snider, On the Von Neumann regularity of rings with
regular prime factor rings, Pacific J. Math. 54 (1974), 135-144.
[GW] B.J. Gardner and R. Wiegandt, Radical Theory of Rings, Marcel Dekker,
New York, 2004.
[K] I. Kaplansky, Rings of Operators, Benjamin, New York, 1968.
[KT] P.A. Krylov, A.A. Tuganbaev, Modules over formal matrix rings (Russian),
Fundam. Prikl. Mat. 15(8) (2009), 145-211; translation in J. Math. Sci. (N.
Y.) 171(2) (2010), 248-295.
[L] T.Y. Lam, Lectures on Modules and Rings, Springer, New York, 1999.
[P] B. Peirce, Linear Associative Algebra, American J. Math. 4 (1881), 97-229.
[R1] L.H. Rowen, Ring Theory I, Academic Press, Boston, 1988.
[R2] L.H. Rowen, Examples of semiperfect rings, Israel J. Math. 65 (1989),
273-283.
[S] J. Szigeti, Linear algebra in lattices and nilpotent endomorphisms of semisim-
ple modules, J. Algebra 319 (2008), 296-308.
[TLZ] G. Tang, C. Li and Y. Zhou, Study of Morita contexts, Comm. Algebra 42
(2014), 1668-1681.
[W] R. Wisbauer, Modules and Algebras: Bimodule Structure and Group Ac-
tions on Algebras, Addison Wesley Longman, Harlow, 1996.
R´enyi Institute of Mathematics, Hungarian Academy of Sciences, 1364 Bu-
dapest, Pf. 127, Hungary
E-mail address: [email protected]
Department of Mathematics, University of Louisiana at Lafayette, Lafayette,
LA 70504-1010, USA
E-mail address: [email protected]
Department of Mathematical Sciences, Stellenbosch University, P/Bag X1,
Matieland 7602, Stellenbosch, South Africa
E-mail address: [email protected]
|
1608.02837 | 1 | 1608 | 2016-08-09T15:25:22 | Toward Homological Characterization of Semirings by e-Injective Semimodules | [
"math.RA"
] | In this paper, we introduce and study e-injective semimodules, in particular over additively idempotent semirings. We completely characterize semirings all of whose semimodules are e-injective, describe semirings all of whose projective semimodules are e-injective, and characterize one-sided Noetherian rings in terms of direct sums of e-injective semimodules. Also, we give complete characterizations of bounded distributive lattices, subtractive semirings, and simple semirings, all of whose cyclic (finitely generated) semimodules are e-injective. | math.RA | math |
Toward Homological Characterization of
Semirings by e-Injective Semimodules
J. Y. Abuhlail∗, S. N. Il'in†, Y. Katsov‡and T. G. Nam§
Abstract
In this paper, we introduce and study e-injective semimodules, in par-
ticular over additively idempotent semirings. We completely character-
ize semirings all of whose semimodules are e-injective, describe semirings
all of whose projective semimodules are e-injective, and characterize one-
sided Noetherian rings in terms of direct sums of e-injective semimodules.
Also, we give complete characterizations of bounded distributive lattices,
subtractive semirings, and simple semirings, all of whose cyclic (finitely
generated) semimodules are e-injective.
Key words:
(e-)Injective Semimodule, (e-)Projective Semimodule,
Morita Equivalence of Semirings, Simple Semirings.
MSC: 16Y60, 16D99, 06A12; 18A40, 18G05
1
Introduction
Semirings and semimodules, and their applications, arise in various branches of
Mathematics, Computer Science, Physics, as well as many other areas of modern
science (see, for instance, [14] and [13]). In the recent years, there has been a sub-
stantial amount of interest in additively idempotent semirings - among which
the Boolean semifield, tropical semifields, and coordinate semirings of tropical va-
rieties represent a set of well-known examples - originated in several extremely
interesting, "nontraditional" contexts as Tropical Geometry [43], Tropical Alge-
bra [20], F1-Geometry [8], and the Geometry of Blueprints [37], for example.
Also, in the last decade, motivated by the Riemann Hypothesis [8] and tropical
∗Department of Mathematics and Statistics, King Fahd University of Petroleum and Min-
erals, Dhahran, Saudi Arabia. Email: [email protected]
†Lobachevsky Institute of Mathematics and Mechanics, Kazan (Volga Region) Federal Uni-
versity, Kazan, Tatarstan, Russia. Email: [email protected]
‡Department of Mathematics, Hanover College, Hanover, IN 47243–0890, USA. Email:
[email protected]
§Institute of Mathematics, VAST, 18 Hoang Quoc Viet, Cau Giay, Hanoi, Vietnam. Email:
[email protected]
1
varieties [43] and [9], several mathematicians have studied from different points
of view semiring schemes, in particular, in [21], sheaves and homological methods
on semiring schemes have been considered.
In the same time, homological characterization/classification of rings by prop-
erties of suitable classes (categories) of modules over them ( see, e.g., [35]) consti-
tutes one of the most sustained interests and important achievements of Homo-
logical Algebra. Inspired by this, during the last three decades, a good number
of important results related to this genre have been obtained in different non-
abelian settings as, for example, in the homological classification of monoids [34]
and distributive lattices [11]. As algebraic objects, semirings are certainly the
most natural generalization of such (at first glance different) algebraic systems as
rings and bounded distributive lattices, and therefore, they form an extremely in-
teresting, natural, and important, non-abelian/non-additive setting for furthering
of the homological characterization, i.e., characterizing semirings by properties
of suitable classes (categories) of semimodules over them.
In fact, this is an
ongoing project of an substantial interest (see, e.g., [22], [24], [23], [16], [25],
[17], [26], [18], [19], and [3]). In all studies regarding the homological character-
ization, the concepts of 'injectivity' and 'projectivity' of objects - R-modules,
S-acts, S-semimodules, etc.- play the most leading role. Perhaps one of the
most important achievements of homological algebra are due to the fact that for
abelian categories of modules RM over a ring R the both - universal algebra
and homological algebra - approaches lead to the identical classes of modules
(see, for example, [35, Sections 1.2A and 1.3A] or [12, Sect. II.6.9]). However,
as we will see later on, in generally additive, but non-abelian, setting of cate-
gories of semimodules SM over a semiring S, these two approaches lead to two
different classes of semimodules. In non-abelian (even in non-additive) settings,
there have been obtained a good number of quite interesting and important re-
sults connected with the concepts of 'projectivity' and 'injectivity' based on the
universal algebra approach (see, e.g., [34]).
In contrast to this, in the present
paper, we initiate investigations related to injective and projective semimodules
defined by using the second approach, i.e., from the homological algebra point of
view, heavily based on the fundamental concepts of 'extensions' and 'short exact
sequences' of modules, that, in turn, lead us to the concepts of 'e-injectivity' and
'e-projectivity' of semimodules. We should mention that the latter concepts have
been earlier somewise considered by some authors in semimodule settings under
different terminology (see, for example, [49], [46], [47], [48], [42], [1], and [2]), but
our approach, nevertheless, is slightly different and, we hope, better reflects the
"homological" spirit of the matter and based on it proofs of the obtained results.
The paper is organized as follows. In Section 2, for the reader's convenience,
we provide all subsequently necessary notions and facts on semirings and semi-
modules.
2
In Section 3, we introduce the concepts of e-injectivity (-projectivity) of semi-
modules and establish some fundamental facts regarding them we use in a se-
quence. Among results of this section, we single out, in our view quite interesting
and useful, Proposition 3.4 and Corollary 3.5 that provide us with a very useful
tool to construct new (e-)injective semimodules from known ones.
In Section 4, we completely characterize e-injective semimodules over addi-
tively idempotent semiring having only two trivial strong one-side ideals (Propo-
sition 4.3) and, based on this characterization, establish some fundamental facts
about e-injective semimodules and their relationship with injective semimodules
over additively idempotent division semirings (Theorems 4.4 and 4.5) that con-
stitute one of the main goals and results of the paper. Also, we demonstrate
(Proposition-Example 4.6) that the concepts of 'injectivity' and 'e-injectivity' for
semimodules over semirings, in general, are different.
In Section 5, we characterize semirings all of whose semimodules are e-injective
(Theorem 5.3), quasi-Frobenius rings in terms of projective and e-injective semi-
modules (Theorem 5.4), and one-sided Noetherian rings in terms of direct sums
of e-injective semimodules (Theorem 5.5). These results are the 'e-injective' ver-
sions of Theorems 3.4, 3.5, and 3.6 of [17], respectively, and they also are the
main results and another main goal of the paper.
Simple semirings, which are the subject of another important area of research
in the theory of semirings, have quite interesting and promising applications in
various fields (for example, in constructing novel semigroup actions for a poten-
tial use in public-key cryptosystems [39]). In contrast to the varieties of groups
and rings, research on simple semirings started only recently, and therefore not
much on the subject is known (for some recent results on simple semirings, one
may consult [5], [40], [50], [27], [31], [28], [32], [33], and [30]). While complete
characterizations of commutative and finite simple semirings have been given in
[5] and [31], respectively, the classification of simple infinite semirings remains an
important unresolved problem (see [27], [28], [32], and [33] for some recent results
in this regard). In light of this and as a substantial step towards this endeavor,
what constitutes one of the main goals of the paper as well, in Section 6, we give
a complete description of simple semirings all of whose cyclic (finitely generated)
semimodules are e-injective (Theorem 6.10). Moreover, in Section 6, we show
that a bounded distributive lattice all of whose cyclic (finitely generated) semi-
modules are e-injective, in fact, is a finite Boolean algebra (Theorem 6.5) and,
applying this result, completely characterize subtractive semirings all of whose
cyclic (finitely generated) semimodules are e-injective (Theorem 6.7). Certainly,
all these theorems belong to the main results of the paper.
Finally, all notions and facts of categorical algebra, used here without any
comments, can be found in [38] or [7]; for notions and facts from semiring theory
we refer to [14].
3
2 Preliminaries
2.1 Recall [14] that a semiring is a datum (S, +, ·, 0, 1) such that the following
conditions are satisfied:
(1) (S, +, 0) is a commutative monoid with identity element 0;
(2) (S, ·, 1) is a monoid with identity element 1;
(3) Multiplication is distributive over addition from both sides;
(4) 0s = 0 = s0 for all s ∈ S.
A semiring that is not a ring we call a proper semiring. A semiring S is
a division semiring if (S \ {0}, ·, 1) is a group; and S is a semifield if it is a
commutative division semiring. Two well-known important examples of semifields
are the so-called Boolean semifield B = {0, 1} with 1 + 1 = 1, and the tropical
semifield T := (R ∪ {−∞}, max, +, −∞, 0}).
As usual, a left S-semimodule over the semiring S is a commutative monoid
(M, +, 0M ) together with a scalar multiplication (s, m) 7→ sm from S × M to M
which satisfies the following identities for all s, s′ ∈ S and m, m′ ∈ M:
(1) (ss′)m = s(s′m);
(2) s(m + m′) = sm + sm′;
(3) (s + s′)m = sm + s′m;
(4) 1m = m;
(5) s0M = 0M = 0m.
Right semimodules over S and homomorphisms between semimodules and
semirings are defined in the standard manner. And, from now on, let M be the
variety of commutative monoids, and MS and SM denote the categories of right
and left S-semimodules, respectively, over a semiring S.
2.2 An element ∞ ∈ M of an S-semimodule M is infinite if ∞ + m = ∞ for
every m ∈ M; and K ≤ S M means that K is an S-subsemimodule of M. Also,
we will use the following subsets of the elements of an S-semimodule M :
I +(M)
Z(M)
V (M)
:= {m ∈ M m + m = m};
:= {z ∈ M z + m = m for some m ∈ M};
:= {m ∈ M m + m′ = 0 for some m′ ∈ M};
For a semimodule M ∈ SM, it is obvious that I +(M) ∩ V (M) = {0}, and
I +(M) ≤ S Z(M) ≤ S M.
A left S-semimodule M is zeroic (zerosumfree, additively idempotent) if Z(M) =
M (V (M) = 0, I +(M) = M). In particular, a semiring S is zeroic (zerosumfree,
additively idempotent) if SS ∈ SM is a zeroic (zerosumfree, additively idem-
potent) semimodule; and we say that S has an infinite element if SS ∈ SM
4
has it. For example, the Boolean semiring B = {0, 1} is a commutative, zeroic,
zerosumfree, additively idempotent semiring in which ∞ = 1.
2.3 A subsemimodule A ≤ SM of a semimodule M is (strongly) subtractive if
(m+m′ ∈ A ⇒ m, m′ ∈ A) m, m+m′ ∈ A ⇒ m′ ∈ A for all m, m′ ∈ M. For each
subsemimodule A ≤ SM, the subsemimodule A := {m ∈ M m + a ∈ A for some
x ∈ A} is obviously the smallest subtractive subsemimodule of SM containing
the subsemimodule A, and therefore, it is called the subtractive closure of A; and
clearly that A is subtractive iff A = A. A left S-semimodule M is subtractive
iff all S-subsemimodules of M are subtractive.
In particular, the semiring S
is left (right) subtractive iff S is subtractive as a left (right) semimodule over
itself. (For important properties of subtractive semirings and semimodules, the
interested reader may consult [25] and [27].)
2.4 Congruences on an S-semimodule M are defined in the standard manner,
and Cong(M) denotes the set of all congruences on M. This set is non-empty
since it always contains at least two congruences-the diagonal congruence △M :=
{(m, m) m ∈ M } and the universal congruence M 2 := {(m, n) m, n ∈ M }.
Any subsemimodule L ≤ SM of an S-semimodule M induces a congruence ≡L
on M, known as the Bourne congruence, by setting m ≡L m′ iff m + l = m′ + l′
for some l, l′ ∈ L; and M/L denotes the factor S-semimodule M/ ≡L having
the canonical S-surjection πL : M −→ M/L. Following [5], a semiring S is
congruence-simple iff the only congruences on S are the diagonal △S and the
universal S2; and S is ideal-simple iff S has exactly two ideals (namely 0 and
S). Note that these notions are not the same (see, e.g., [28, Examples 3.8]).
Moreover, we say that a S is simple iff it is simultaneously congruence-simple and
ideal-simple. The classification of simple infinite semirings remains an important
unresolved problem (see [27], [28], [32], and [33] for recent related results).
the category SM, a free (left) semimodule Pi∈I Si, Si
2.5 As usual (see, for example, [14, Chapter 17]), if S is a semiring, then in
∼= SS, i ∈ I, with a basis
set I is a direct sum (a coproduct) of I copies of SS; a semimodule P ∈ SM is
projective if it is a retract of a free semimodule. Following [22] and [24], in which
there were introduced and considered in detail the tensor product bifunctors
− ⊗S − : MS× SM −→ M, a semimodule F ∈ SM is called (mono-)flat
iff the functor − ⊗S F : MS −→ M preserves (monomorphisms) finite limits;
and the latter is equivalent to that F is a filtered (directed) colimit of finitely
generated free (projective) semimodules. A semimodule M ∈ SM is finitely
generated (cyclic) iff M is a homomorphic image of a free left S-semimodule with
a finite basis (a homomorphic image of SS); a semimodule M ∈ SM is injective
if for any monomorphism µ : A B of left S-semimodules A and B and every
homomorphism f ∈ SM(A, M), there exists a homomorphism ef ∈ SM(B, M)
such that ef µ = f .
5
2.6 For the reader's convenience, we also recall some fundamental notions and
facts about Morita equivalence of semirings (see, e.g., [24] and [26]) that we will
use in sequence. Recall that two semirings S and T are said to be Morita equiv-
alent iff the semimodule categories SM and T M are equivalent, i.e., there exist
two functors F : SM −→ T M and G : T M −→ SM and natural isomorphisms
η : GF −→ IdSM and ξ : F G −→ IdT M. Following [26], a right semimodule PS
is said to be a generator for the category MS of right S-semimodules iff SS is a
retract of a finite direct sum ⊕iP of PS; and that PS is said to be a progenerator
for MS iff PS is a finitely generated projective generator. Then, by [26, Theorems
4.5 and 4.12], F ≃ P ⊗S − for some (T, S)-bisemimodule P such that PS is a
progenerator, P ∗ := HomS(PS, SS) is a progenerator in MT , T ≃ End(PS) as
semirings and G ≃ P ∗ ⊗T −.
2.7 For any left S-semimodule M, there exists the left R-module of differ-
ences D(M) of M [14, Chapter 16] (see also [29, p. 5083]) defined as the factor
semimodule of the left S-semimodule M × M with respect to the subsemimodule
W = {(m, m) m ∈ M } ⊆ M × M, i.e., D(M) := (M × M)/W . In fact, the
semimodule D(M) is a left S-module since for any (m, m′) ∈ M ×M in D(M) one
has (m, m′) + (m′, m) = (0, 0). Also, there exists the canonical S-homomorphism
ξM : M −→ D(M) given by m 7−→ (m, 0). In the case when M is a cancellative
semimodule, ξM is injective, and therefore, we can consider the elements (m, 0)
and m to be the same and any element (m, m′) ∈ D(M) to be the "difference"
of the elements (m, 0) and (m′, 0), i.e., D(M) = {m − m′ m, m′ ∈ M}. In par-
ticular, the left S-module of differences D(S) of the regular semimodule SS can
be considered as a ring - the ring of differences of S [14, Chapter 8, p. 101] -
with the operation of multiplication defined for all a, b, c, d ∈ S by (a, b)(c, d) =
(ac + bd, ad + cb); and if S is a semiring, then the ring of differences D(S) is
also a semiring with the identity (1, 0). Moreover, it is easy to see that D(M)
becomes a left D(S)-module with (a, b) (m1, m2) = (am1 + bm2, am2 + bm1) for
all a, b ∈ S and m1, m2 ∈ M.
2.8 Any homomorphism f : M −→ N of left S-semimodules induces kernel
congruence ≡f on M such that for any m, m′ ∈ M, we have m ≡f m′ iff f (m) =
f (m′), as well as the following subsemimodules:
Ker(f )
im(f )
Im(f )
:= {m ∈ M f (m) = 0};
:= {f (m) m ∈ M};
:= {n ∈ N n + f (m) = f (m′) for some m, m′ ∈ M},
called the kernel, the image, and the extended image of f , respectively. Notice
that Im(f ) = im(f ) and that ≡ Ker(f ) ⊆ ≡f . Moreover, one can easily see that
N/im(f ) = N/Im(f ) and M/ ≡f ≃ im(f ). However, in general, in our non-
abelian setting, as the following example shows, the S-semimodules M/Ker(f )
and im(f ) are not necessarily isomorphic: Indeed, the tropical semifield T :=
6
(R ∪ {−∞}, max, +, −∞, 0}) contains the Boolean semifield B as a subsemiring,
whence T can be considered as a B-semimodule in a canonical way, and let
f : T −→ B be the B-homomorphism map given by
f (x) =
0,
1,
x = −∞
x 6= −∞
;
Then, clearly, f is surjective and Ker(f ) = 0, but T/Ker(f ) ≃ T ≇ B = im(f ).
3
e-Injectivity (-projectivity) of semimodules
For the readers' convenience, briefly remind some general notions for categories
with zero morphisms (see, for example, [44, Chapters 7, 8, 13]) in the context of
semimodule categories SM. A kernel (K, k) of a homomorphism f : A −→ B
of left S-semimodules is a homomorphism k : K −→ A such that
(i) f k = 0,
(ii) for every homomorphism x : X −→ A with f x = 0 there exists exactly
one homomorphism i : X −→ K such that x = ki; and in this case, we write
(K, k) = ker f . By dualizing, one comes up to the concept of a cokernel (k, K),
k : B −→ K of f : A −→ B, and (k, K) = co ker f . Then, it is obvious (or it
is just readily follows from (co)completeness of semimodule categories SM [38,
Sections 5.1, 5.2]) that there exist ker f and co ker f for every homomorphism
f ∈ SM(A, B) := HomS(A, B). Furthermore, following, for example, [12, Sect.
II.6.2] or [44, Sect.13.2] and without loss of generality, we can define a short exact
sequence in SM as a sequence
0 −→ A
f
−→ B
g
−→ C −→ 0
(∗)
of semimodules A, B, C ∈ SM and homomorphisms f and g such that (A, f ) =
ker g and (g, C) = co ker f .
Let F : SM −→ RM (G : SM −→ RM) be a covariant (contravariant) functor
between the semimodule categories SM and RM. Then, the functor F (G) we
say is an exact functor (or just an e-functor ) if for any exact sequence (*) in SM
the sequence
0 −→ F (A)
F (f )
−→ F (B)
F (g)
−→ F (C) −→ 0 (0 ←− G(A)
G(f )
←− G(B)
G(g)
←− G(C) ←− 0)
is exact in RM as well. The following observations will prove to be useful.
Proposition 3.1 (1) A direct sum ⊕iFi of covariant functors Fi : SM −→ RM,
i ∈ I, is an e-functor iff each summand Fi is an e-functor;
(2) A direct product Πi∈I Gi of contravariant functors Gi : SM −→ RM,
i ∈ I, is an e-functor iff each factor Gi is an e-functor;
7
(3) A retract of an e-functor is an e-functor as well.
Proof (1). It immediately follows from the observation that for an exact sequence
(∗), the sequence
0 −→ ⊕iFi(A)
⊕iFi(f )
−→ ⊕iFi(B)
⊕iFi(g)
−→ ⊕iFi(C) −→ 0
is, in fact, a 'direct sum'
0 −→ ⊕i(Fi(A))
⊕i(Fi(f ))
−→ ⊕i(Fi(B))
⊕i(Fi(g))
−→ ⊕i(Fi(C)) −→ 0
of the exact sequences
0 −→ Fi(A)
Fi(f )
−→ Fi(B)
Fi(g)
−→ Fi(C) −→ 0,
i ∈ I,
and therefore, is an exact sequence itself.
(2). It is established in the same fashion as (1).
(3). Clearly, it is enough to consider only a "covariant" case. So, let Q be a
retract of an exact covariant functor F : SM −→ RM, i.e., there exist natural
transformations of functors µ : Q −→ F and π : F −→ Q such that πµ = 1Q.
Then, a "diagram chase" of a commutative diagram for an exact sequence (∗)
0 −→ Q(A)
↓
↓ µA
0 −→ F (A)
↓
↓ πA
0 −→ Q(A)
Q(f )
−→ Q(B)
↓ µB
F (f )
−→ F (B)
↓ πB
Q(f )
−→ Q(B)
Q(g)
−→ Q(C) −→ 0
↓
↓ µC
F (g)
−→ F (C) −→ 0
↓
↓ πC
Q(g)
−→ Q(C) −→ 0
leads us to the exactness of the sequence 0 −→ Q(A)
Q(f )
−→ Q(B)
Q(g)
−→ Q(C) −→ 0.
(cid:3)
In the case when a semiring S is a ring, the category SM becomes an abelian
category of modules over a ring S. Then, there are two, at the first glance,
different approaches to the concepts of injectivity and projectivity of modules:
the first one is an 'universal algebra' approach presented in 2.5, and the second
- a 'homological algebra' approach that in more general setting of semimodules
over semirings is given by the following definition.
Definition 3.2 A semimodule M ∈ SM over a semiring S is exactly-injective
(exactly-projective), or shortly e-injective (e-projective), if SM(−, M) :=
HomS(−, M) : SM −→ M (SM(M, −) := HomS(M, −) : SM −→ M) is an
exact contravariant (covariant) functor.
Perhaps one of the most important achievements of homological algebra are
due to the fact that for abelian categories of modules SM over a ring S the both
8
- universal algebra and homological algebra - approaches lead to the identical
classes of modules (see, for example, [35, Sections 1.2A and 1.3A] or [12, Sect.
II.6.9]). However, as we will see later on, in generally non-abelian setting of
categories of semimodules SM over a semiring S, these two approaches lead to
two different classes of semimodules. As was mentioned earlier, in non-abelian
(even in non-additive) settings, there have been obtained a good number of quite
interesting and important results connected with the concepts of 'projectivity'
and 'injectivity' based on the universal algebra approach. In contrast to this, in
the present paper, we initiate investigations related to injective and projective
semimodules defined by using the second approach, i.e., from the homological
algebra point of view. Thus, from Proposition 3.1 we right away obtain
Corollary 3.3 (1) A direct sum ⊕iMi of e-projective semimodules Mi ∈ SM,
i ∈ I, is an e-projective semimodule iff each summand Mi is an e-projective
semimodule;
(2) A direct product Πi∈I Mi of e-injective semimodules Mi ∈ SM, i ∈ I, is
an e-injective semimodule iff each factor Mi is an e-injective semimodule;
(3) A retract of an e-injective (e-projective) semimodule is an e-injective (e-
projective) semimodule as well.
Now, let SMT be a category of bisemimodules over semirings T and S (see,
e.g., [24, Sect. 3]) and P ∈ SMT . Then for any semimodule M ∈ SM, on
the set of homomorphisms HomS(SP,S M) := SM(P, M) there exists a natural
structure of a left T -semimodule defined as follows: (tf )(p) = f (pt) for any f ∈
HomS(SP,S M), p ∈ P , and t ∈ T . Moreover, there is a functor HomS(SP, −) :
SM −→ T M. The next result provides us with main tools for constructing
examples of (e-)injective semimodules by using known (e-)injective semimodules
over one semiring to produce (e-)injective semimodules over another. It includes
and/or generalizes the well-known "Injective Producing Lemma" [35, Lemma
1.3.5] and [14, Proposition 17.25].
Proposition 3.4 (1) Let for a semimodule P ∈ SMT the semimodule PT ∈
MT be mono-flat and a semimodule M ∈ SM an injective left S-semimodule,
then the semimodule HomS(SP,S M) ∈ T M is an injective left T -semimodule
as well;
(2) Let for a semimodule P ∈ SMT the functor PT ⊗− : T M −→SM
preserve short exact sequences and a semimodule M ∈ SM an e-injective left
S-semimodule, then the semimodule HomS(SP,S M) ∈ T M is an e-injective left
T -semimodule as well.
(1).
Proof
It is easy to see that a semimodule T I ∈ T M is injective iff
the contravariant functor HomT (−,T I) : T M −→ M moves any monomorphism
µ : A B of left T -semimodules A and B to the surjection HomT (µ,T I) :
9
HomT (B,T I) −→ HomT (A,T I) in M. Therefore, we need only to show that for
any injective semimodule M ∈ SM and any monomorphism µ : A B in T M,
HomT (µ, HomS(SP,S M))
: HomT (B, HomS(SP,S M)) −→
HomT (A, HomS(SP,S M))
is a surjection in M. However, by [24, Theorem 3.3] (or [22, Theorem 3.6]),
the functor PT ⊗− : T M −→SM is a left adjoint to the functor HomS(SP, −) :
SM −→ T M. Whence, from the commutative diagram
HomS(SP ⊗ A,S M)
↓ HomS(1S P ⊗ µ, 1SM )
HomS(SP ⊗ B,S M)
≃ HomT (A, HomS(SP,S M))
↑ HomT (µ, HomS(SP,S M))
≃ HomT (B, HomS(SP,S M))
and the mono-flatness of the semimodule PT , we have that HomS(SP,S M) ∈
T M is an injective semimodule.
(2). In the same fashion as it was done in (1) and assuming that PT ⊗− :
T M −→SM preserves short exact sequences, one gets the e-injectivity of
HomS(SP,S M) ∈ T M.
(cid:3)
Corollary 3.5 (cf.
[35, Corollary 1.3.6B] and [14, Proposition 17.25]) Let f :
S −→ T be a semiring homomorphism. Then the functor HomS(ST, −) : SM −→
T M preserves (e-)injective semimodules.
Proof Taking into consideration that by [22, Proposition 3.8] there is the functor
isomorphism TT ⊗− ≃ Id : T M −→ M, and therefore, TT ⊗− is mono-flat and
preserves all (co)limits as well, the statements follow right away from Proposition
3.4.
(cid:3)
We conclude this section with the following two remarks.
Remark 3.6 By dualization, the interested reader may easily obtain the corre-
sponding analogs of the statements of Proposition 3.4 and Corollary 3.5 regarding
(e-)projective semimodules.
Remark 3.7 It should be mentioned that the concept of e-injectivity for semi-
modules, in some way and using different terminology, have been earlier consid-
ered in [48], [49], [1], and [2], where the authors heavily used the obvious fact
that in categories of semimodules a monomorphism µ : A B is a kernel of
its cokernel iff the subsemimodule A ≤ B is a subtractive one. Moreover, it
is easy to see that a semimodule T I ∈ T M is e-injective iff the contravariant
functor HomT (−,T I) : T M −→ M moves any monomorphism µ : A B with
µ(A) being a subtractive subsemimodule of a left T -semimodule B to the sur-
jection HomT (µ,T I) : HomT (B,T I) −→ HomT (A,T I) in M and (HomT (µ,T I),
HomT (A,T I)) = co ker(HomT (π,T I)), where π : B −→ B/µ(A) is the natural
10
surjection; in other words, T I ∈ T M is e-injective iff, for any T -homomorphism
ϕ : A −→ I, there exists a T -homomorphism ψ : B −→ I such that ϕ = ψµ
and, for any T -homomorphisms ψ1, ψ2 : B −→ I, the equality ψ1µ = ψ2µ implies
ψ1 + χ1 = ψ2 + χ2 for suitable χ1 and χ2 for which χ1µ = 0 = χ2µ. And for that
reason, the terminology used in those works was certainly influenced by these
facts. Our approach and "ideology" are slightly different.
4
e-Injective semimodules over additively idem-
potent semirings
In this section, we consider the structure of e-injective semimodules as well as re-
lations between e-injective and injective semimodules over additively idempotent
semirings. To do so, we need first to establish some useful facts.
Proposition 4.1 Let S be a zerosumfree semiring and M ∈ SM an e-injective
S-semimodule. Then there exists an element z ∈ M such that m + z = z for all
m ∈ M, and in addition, sz = z for every s ∈ S\{0} if S is an entire semiring.
i ∈ M, xi, x′
i=1xim′′
i = s1 and m1 + Σn
i, m′′
i = m2 + Σn
Proof For an e-injective semimodule M ∈ SM consider the relation ∼ on
the left S-semimodule S × M ∈ SM defined as follows: (s1, m1) ∼ (s2, m2)
iff s1 = s2 and there are m′
i ∈ S (i = 1, 2, ..., n) such that
xi + x′
i=1xim′
i for some natural number n. As
was shown in [17, Propositions 1.6 and 1.7], ∼ is an S-congruence on S × M ∈
SM and there exists an S-monomorphism α : M −→ S × M/∼, m 7→ [0S, m]
such that [0S, m] + [1, 0M ] = [1, 0M ] for all m ∈ M, respectively. Obviously,
α(M) is a subtractive subsemimodule of the semimodule S × M/∼. Therefore as
well as taking into considertation Remark 3.7, there exists an S-homomorphism
β : S × M/∼ −→ M such that βα = idM . Let z := β([1, 0M ]); and as
it was
shown in [17, Proposition 1.7], m + z = z for all m ∈ M.
If in addition, S is an entire semiring, it is easy to see that for the natural
embedding µ : M Ext(M), where Ext(M) := M ∪ {∞} with 0∞ = 0 and
s∞ = ∞ for all s ∈ S\{0} and m + ∞ = ∞ for all m ∈ Ext(M), there is a
subtractive semimodule µ(M) ≤ Ext(M). Since SM is e-injective, by Remark
3.7, there exists γ : Ext(M) −→ M such that γµ = idM , and clearly z = γ(∞)
and sz = sγ(∞) = γ(s∞) = γ(∞) = z for every s ∈ S\{0}.
(cid:3)
Lemma 4.2 If a zerosumfree semiring S has only two trivial strongly subtractive
left (right) ideals, then S is entire.
Indeed, suppose that ab = 0 for some a, b ∈ S and b 6= 0. Since S is
Proof
zerosumfree, the left ideal
(0 : Sb) := {s ∈ S sb = 0}
11
is strongly subtractive and 1 /∈ (0 : Sb). Hence, (0 : Sb) = 0 and a = 0.
(cid:3)
The next observation, complementing Proposition 4.1, completely describes e-
injective semimodules over additively idempotent semirings with only two trivial
one-sided strongly subtractive ideals.
Proposition 4.3 A left S-semimodule M ∈ SM over an additively idempotent
semiring S with only two trivial left strongly subtractive ideals is e-injective iff it
possesses the infinite element ∞ ∈ M such that s∞ = ∞ for all s ∈ S\{0}.
Proof =⇒. By Lemma 4.2, S is an additively idempotent, zerosumfree, and
entire, semiring. Whence, the result follows from Proposition 4.1.
⇐=. Let µ : A B be the inclusion map from a subtractive subsemimodule
SA to a semimodule SB in SM. For S is additively idempotent, the additive
reducts of all S-semimodules are idempotent monoids as well. Moreover, SA is a
strongly subtractive subsemimodule of SB: Indeed, if x+y ∈ A for some x, y ∈ B,
then x + (x + y) = x + y ∈ A and y + (x + y) = x + y ∈ A, and hence, x, y ∈ A.
Next, for an S-semimodule homomorphism f : A −→ M, define a map g :
B −→ M as follows:
g(b) =
f (b),
b ∈ A
.
∞,
b /∈ A
It is obvious that f = gµ, and we claim that actually g is an S-semimodule
homomorphism, too.
Indeed, for any b, c ∈ B, if b + c ∈ A, then b, c ∈ A since A is a strongly
subtractive subsemimodule of B, and therefore, g(b+c) = f (b+c) = f (b)+f (c) =
g(b) + g(c); if b + c /∈ A, then we have the following two cases to consider:
Case 1: b ∈ A and c /∈ A. Then, g(b + c) = ∞ = f (b) + ∞ = g(b) + g(c);
Case 2: b /∈ A and c /∈ A. Then, g(b + c) = ∞ = ∞ + ∞ = g(b) + g(c).
Now let b ∈ B and s ∈ S. For A ≤S B is a strongly subtractive subsemimod-
ule, the left ideal
Ib := {s ∈ S sb ∈ A}
is a strongly subtractive left ideal in S. Hence, Ib = 0 or Ib = S; and sb ∈ A iff
s = 0 or b ∈ A, or equivalently, sb /∈ A iff s 6= 0 and b /∈ A. Furthermore, we
always have s∞ = ∞ for all s ∈ S\{0}. From these observations, we immediately
get g(sb) = sg(b), and therefore, g is an S-semimodule homorphism.
Finally, let π : B −→ B/A be the natural surjection.
It is easy to see
that there is a homomorphism γ ∈ HomS(B/A, M) such that the image γ([b])
of any non-zero element [b] ∈ B/A is the infinite element ∞ ∈ M. More-
over, it is quite clear that for any α, α′ ∈ HomS(B, M) such that αµ = α′µ,
we have α + γπ = α′ + γπ with γπµ = 0. Hence, by Remark 3.7, we get
(HomS(µ, M), HomS(A, M)) = co ker(HomS(π, M)), and SM is an e-injective
S-semimodule.
(cid:3)
12
It is obvious that every division semiring has only two trivial strongly left/right
subtractive ideals. So, as a consequence of Proposition 4.3, among other, in our
view, interesting and important observations, in the next theorem we obtain a
complete characterization of e-injective semimodules over additively idempotent
division semirings.
Theorem 4.4 For an additively idempotent division semiring D the following
statements are true:
(1) A left D-semimodule M is e-injective iff M has an infinite element;
(2) Every left D-semimodule can be embedded in an e-injective D-semimodule;
(3) Every injective left D-semimodule is e-injective;
(4) Every finite left D-semimodule is e-injective;
(5) The regular D-semimodule DD (and therefore, every finitely generated left
D-semimodule) is e-injective iff D ≃ B.
Proof So, let D be an additively idempotent, and therefore, zerosumfree, division
semiring.
(1) =⇒. This follows immediately from Proposition 4.3.
⇐=. Let M be a left D-semimodule with the infinite element ∞. For every
s ∈ D\{0}, we have s∞ + ∞ = ∞, and hence, ∞ = ∞ + s−1∞ = s−1∞, whence
s∞ = ∞. From this observation and Proposition 4.3, we conclude that DM is
e-injective.
(2) For every left D-semimodule M, the left D-semimodule Ext(M) := M ∪
{∞} has an infinite element, and therefore by (1), it is e-injective.
(3) Let M be an injective left D-semimodule. By [17, Proposition 1.7], M has
an infinite element ∞, whence M is e-injective by (1).
(4) Let M = {m1, · · · , mn}. Then, ∞ := Σn
i=1mi is an infinite element of
M and by (1) DM is e-injective.
(5) =⇒. Assume that DD is e-injective. Since D is zerosumfree and by
Proposition 4.1, it has an infinite element ∞. It follows that ∞ = ∞ + ∞2 =
(∞ + 1)∞ = ∞2, and hence, 1 = ∞. Then, for every d ∈ D\{0}, we have
d−1 + 1 = 1 and 1 = d + 1 = d(1 + d−1) = d · 1 = d. Thus, D ≃ B.
⇐=. This is clear from (1) and (4).
(cid:3)
Our next observation extends statements (2) and (3) of Theorem 4.4 to arbi-
trary idempotent semirings and actually establishes the e-injective completeness
of categories of S-semimodules SM over additively idempotent semirings S (the
injective completeness of those categories has been established earlier in [22, The-
orem 4.2]), namely:
Theorem 4.5 For an additively idempotent semiring S the following statements
are true:
(1) Every left S-semimodule can be embedded in an e-injective S-semimodule;
13
(2) Every injective left S-semimodule is e-injective.
Proof
(1) Since S is additively idempotent, there is the obvious semiring em-
bedding µ : B S of the Boolean semifield B into a subsemiring S. Therefore,
any semimodule M ∈ SM has the canonical B-semimodule structure, i.e., M ∈
BM. By (2) of Theorem 4.4, there exists an an e-injective B-semimodule Q
such that M ≤ BQ. Then, taking into consideration obvious embeddings
SM ≃ HomS(S, M) ≤ HomB(S, M) ≤ HomB(S, Q),
and applying Corollary 3.5, we have that HomB(SS, Q) is an e-injective left S-
semimodule.
(2) Let M ∈ SM be an injective left S-semimodule. By (1), in the cat-
egory SM there exists an embedding µ : M Q, where Q is an e-injective
S-semimodule. Whence, as SM is an injective S-semimodule, there exists an
S-homomorphism π : Q −→ M such that πµ = idM , i.e., SM is a retract of SQ;
and by (3) of Corollary 3.3, we end the proof.
(cid:3)
Theorems 4.4 and 4.5 show that in the category SM over an additively idem-
potent semiring S, the class of injective semimodules, Inj(SM), is a subclass
of the class of e-injective semimodules, e-Inj(SM). In fact, our next observa-
tion shows that for an additively idempotent division semiring S, the subclass
Inj(SM) is always a proper one of the class e-Inj(SM), namely:
Proposition-Example 4.6 Let D be an additively idempotent division semiring.
Then Inj(DM) ⊂ e-Inj(DM).
Proof For D is a zerosumfree division semiring, there exists a surjective semiring
homomorphism π : D −→ B. By [26, Lemma 5.2], the restriction functor π# :
BM −→ DM preserves non-injective semimodules. Let M ∈ BM be a non-
distributive finite lattice, then BM is non-injective B-semimodule [10, Theorem
4], and therefore, π#(M) ∈ DM is non-injective D-semimodule. However, by
(4) of Theorem 4.4, π#(M) ∈ e-Inj(DM).
(cid:3)
In light of the observations above, we finish this section by posting the fol-
lowing, in our view interesting and perspective, conjecture and problems.
Conjecture 1 Inj(SM) = e-Inj(SM) iff a semiring S is a ring.
Problem 2 Is (1) of Theorem 4.5 true for all additively regular semirings S?
Problem 3 Does an e-injective envelope exist for every semimodule M ∈ SM
over an additively idempotent semiring S?
Problem 4 Describe all semirings S such that Inj(SM) ⊂ e-Inj(SM).
14
5 Characterizations of some special classes of
semirings
In [17], there were obtained semiring analogs of well-known characterizations of
(classical) semisimple, quasi-Frobenius, and one-sided Noetherian rings by means
of injective semimodules over them. Motivated by this, it is quite natural to
consider characterizations of those classes of semirings in terms of e-injective
semimodules over them. And therefore, establishing the 'e-injective' versions of
those characterizations constitutes the main goal of this section.
Proposition 5.1 (cf. [17, Theorem 1.2]) Let S be a zerosumfree semiring and
{Mi}i∈I a family of left nonzero S-semimodules. Then the left S-semimodule
⊕i∈I Mi is e-injective iff all semimodules {Mi}i∈I are e-injective and I < ∞.
Proof =⇒. Let M = ⊕i∈I Mi be an e-injective left S-semimodule. By Corol-
lary 3.3 (3), for every i ∈ I, the S-semimodule Mi is e-injective, and hence, by
Proposition 4.1, there exists the element zi ∈ Mi such that mi + zi = zi for all
mi ∈ Mi. Then, the image of the canonical embedding µ : M M := Πi∈I Mi is
clearly a subtractive subsemimodule, and therefore, there exists a homomorphism
ϕ : M −→ M such that ϕµ = 1M .
Now let z := (zi)i∈I ∈ M . Since ϕ(z) ∈ M, there exists a finite subset J ⊆ I
and a collection of elements (mj)j∈J with mj ∈ Mj, such that ϕ(z) = Σj∈J mj.
We shall show that actually J = I. Indeed, for any index i ∈ I \ J, it is easy to
see that µ(zi) + z = z and
Σj∈J mj = ϕ(z) = ϕ(µ(zi) + z) = ϕ(µ(zi)) + ϕ(z) = zi + Σj∈J mk;
and as i /∈ J, one has that zi = 0. Whence, for any mi ∈ Mi, we have that
0 = zi = zi + mi = 0 + mi = mi, i.e., Mi = 0 what contradicts to Mi 6= 0. Thus,
J = I.
⇐=. It follows straightforwardly from Corollary 3.3.
(cid:3)
For given a semiring congruence θ on a semiring S and a semimodule M ∈
SM, we say that an element m ∈ M is compatible with θ iff s1θs2 =⇒ s1m =
s2m for all s1, s2 ∈ S; and let M(θ) := {m ∈ M m is compatible with θ}.
As was shown in [17, Propositions 1.2 and 1.3], M(θ) is a left S/θ-semimodule
and M(θ) ≃ HomS(S/θ, M) as left S-semimodules. From these observations
and Corollary 3.5, we immediately obtain the following e-injective analog of [17,
Proposition 1.4]:
Proposition 5.2 If M is an e-injective left S-semimodule, then M(θ) is an
e-injective left S/θ-semimodule.
In the next result, we complement [17, Theorem 3.4] and describe the class of
semirings all of whose semimodules are e-injective.
15
Theorem 5.3 The following conditions for a semiring S are equivalent:
(1) All left S-semimodules are e-injective;
(2) S is a classical semisimple ring.
(1) =⇒ (2). By our hypothesis, S(N) is an e-injective left S-semimodule.
Proof
Let θ be the Bourne relation on S corresponding to the ideal V (S) (see 2.2 and
2.4). Then, it is easy to see that S/θ is a zerosumfree semiring, and for the e-
injective left S-semimodule S(N) there is the natural S/θ-semimodule isomorphism
S(N)(θ) ≃ S(θ)(N), and therefore, by Proposition 5.2, S(N)(θ) is an e-injective left
S/θ-semimodule. From the latter and Proposition 5.1, S(θ) = 0.
Also, it is easy to verify that the image of the canonical embedding ξV (S) :
V (S) D(S) (see 2.7) is a subtractive S-subsemimodule. For SS is e-injective,
the natural injection µ : V (S) S can be extended to an S-homomorphism
ψ : D(S) −→ S such that ψξV (S) = µ. Since D(S) is a ring, it is clear that
e := ψ(1D(S)) ∈ V (S). For every s ∈ V (S), we have s = µ(s) = ψ(ξV (S)(s)) =
In particular, e2 = e, and so s(1 − e) = se(1 −
ψ(s1D(S)) = sψ(1D(S)) = se.
e) = s(e − e) = s0 = 0 for all s ∈ V (S), and applying [17, Lemma 1.1], we
conclude that 1 = e ∈ V (S), i.e., S is actually a ring and, by [36, Theorem 1.2.9],
even a semisimple ring as the concepts of injectivity and e-injectivity coincide for
modules over rings.
(2) =⇒ (1). This follows immediately from [36, Theorem 1.2.9] as the concepts
of injectivity and e-injectivity coincide for modules over rings.
(cid:3)
The celebrated Faith-Walker Theorem provides the characterization of quasi-
Frobenius rings as rings over which the classes of projective and injective modules
coincide (e.g., [4, Theorem 31.9], or [35, Theorem 15.9]). In [17, Theorem 3.5],
this characterization has been generalized in the semiring setting, and our next
result is an 'e-injective' version of [17, Theorem 3.5].
Theorem 5.4 The following conditions for a semiring S are equivalent:
(1) S is a quasi-Frobenius ring;
(2) All projective left S-semimodules are e-injective;
(3) All e-injective left S-semimodules are projective, and the S-semimodule
S/V (S) can be embedded in an e-injective left S-semimodule.
Proof
(1) =⇒ (2). This follows from [4, Theorem 31.9], or [35, Theorem 15.9],
and the fact that the notions of injectivity and e-injectivity coincide for modules
over rings.
(2) =⇒ (1). Since the projective left S-semimodule S(N) is e-injective, as in
the proof of the implication (1) =⇒ (2) of Theorem 5.4 we have that S is a ring
and then use [4, Theorem 31.9].
(1) =⇒ (3).
If S is a ring, S/V (S) = 0 is an e-injective S-semimodule.
Therefore, the statement follows from [4, Theorem 31.9].
16
(3) =⇒ (1). Let θ be the Bourne relation on S induced by V (S). By the
hypothesis, S/θ can be embedded in an e-injective left S-semimodule M. It is
easy to see that every element of S/θ is compatible with θ, and hence, S/θ can
be embedded in M(θ). Let I be an infinite set such that I ≥ S. By Corollary
3.3 (2), M I is an e-injective left S-semimodule, and hence, a projective left S-
semimodule; and therefore, by [17, Theorem 2.1], (M, +, 0) is an abelian group.
For SM is e-injective, by Proposition 5.2, M(θ) is an e-injective left semimodule
over the zerosumfree semiring S/θ; and by Proposition 4.1, there exists an element
z ∈ M(θ) such that z + m = z for all m ∈ M(θ). In particular, z + z = z, and,
hence, as M(θ) ⊆ M and (M, +, 0) is a group, z = 0, and therefore, M(θ) = 0
and S/θ = 0, i.e. S is a ring. From the latter, the implication follows from [4,
Theorem 31.9].
(cid:3)
Also, the famous and very important characterization of Noetherian rings as
rings over which direct sum of injective modules is always an injective module
given by H. Bass and Z. Papp (see, e.g., [35, Theorem 3.46]) has been generalized
in the semiring setting in [17, Theorem 3.6], and we conclude this section by an
'e-injective' version of the latter.
Theorem 5.5 The following conditions for a semiring S are equivalent:
(1) S is a left Noetherian ring;
(2) Every direct sum of e-injective left S-semimodules is e-injective, and the
S-semimodule S/V (S) can be embedded in an e-injective left S-semimodule.
Proof (1) =⇒ (2). S/V (S) is obviously the zero e-injective left S-semimodule for
any left Noetherian ring S, and the implication follows from [35, Theorem 3.46].
(2) =⇒ (1). Let θ be the Bourne relation on S induced by V (S). By the
hypothesis, S/θ can be embedded in an e-injective left S-semimodule M. It is
easy to see that every element of S/θ is compatible with θ, and hence, S/θ can
be embedded in M(θ). Then, as M (N) is an e-injective left S-semimodule and
applying Proposition 5.2, we have that M(θ)(N) ≃ M (N)(θ) is an e-injective left
S/θ-semimodule. For S/θ is zerosumfree and applying Proposition 5.1, we have
M(θ) = 0 and, hence, S/θ = 0. Therefore, by [35, Theorem 3.46], S is a left
Noetherian ring.
(cid:3)
6 Semirings all of whose finitely generated semi-
modules are e-injective
As was shown in [19, Corollary 3.2], every semimodule can be represented, in a
canonical way, as a colimit of its cyclic subsemimodules. This observation mo-
tivates studying of semirings over which any semimodule is a colimit of cyclic
17
semimodules possessing some special properties (see, e.g., [3], [19]). Thus, it is
quite natural that the main goal of this section is to present complete character-
izations of semirings, belonging to some important special classes of semirings,
all of whose cyclic (finitely generated) semimodules are e-injective. But first we
make some interesting and useful general observations.
Proposition 6.1 The class of semirings all of whose cyclic (finitely generated)
left semimodules are e-injective is closed under homomorphic images and finite
products.
Proof Let C ( F) be the class of semirings all of whose cyclic (finitely generated)
left semimodules are e-injective. First consider the statement for the class C.
So, let S, T be semirings with S ∈ C and π : S −→ T a surjective semiring
homomorphism. We claim that T ∈ C. By [24, Section 4], π induces two functors
- the restriction of scalars functor π# : T M −→ SM and the extension functor
π# := T ⊗S − : SM −→ T M - such that π# is a left adjoint to the functor π#,
i.e., π# ⊣ π# [24, Proposition 4.1]. Moreover, by [24, Proposition 4.6], π#π# ≃
IdT M; also, it is easy to see that a short sequence 0 −→ A
−→ C −→ 0
is exact in T M iff it is exact in SM, and any semimodule M ∈ T M is cyclic
iff M ∈ SM is cyclic. By using these observations and the assumption S ∈ C,
it is easy to see that the functor HomT (−, M) : T M −→ M preserves short
exact sequences iff the functor HomS(−, M) : SM −→ M does it, and therefore,
T ∈ C, i.e., C is closed under homomorphic images. Using the same arguments,
we obtain the closedness of the class F with respect of homomorphic images.
−→ B
f
g
It is clear that it is enough to show only for two semirings S1, S2 ∈ C ( F) that
S := S1 ⊕ S2 ∈ C ( F) as well. Obviously, any S-semimodule M ∈ SM actually
is the direct sum of its S-subsemimodules S1M := {(s1, 0) M s1 ∈ S1} and
S2M := {(0, s2) M s2 ∈ S2}, i.e., M = S1M ⊕ S2M. Moreover, it is easy to see
that any homomorphism f : M = S1M ⊕ S2M −→ S1N ⊕ S2N = N between S-
semimodules M, N ∈ SM is, in fact, the direct sum of the two corresponding S1-
homomorphism f1 : S1M −→ S1N and S2-homomorphism f2 : S2M −→ S2N be-
tween S1M, S1N ∈ S1M and each S2M, S2N ∈ S2M, respectively. From observa-
tions for any functor HomS(−, M) : SM −→ M, we readily have HomS(−, M) =
HomS(−, S1M ⊕ S2M) ≃ HomS(−, S1M) × HomS(−, S2M) ≃ HomS1(−, S1M) ×
HomS2(−, S2M), and therefore, by Proposition 3.1, the functor HomS(−, M) is
e-injective as soon as the functors HomS1(−, S1M) and HomS2(−, S2M) are e-
injective as well. Now it follows immediately that S ∈ C ( F) if we observe that
the semimodules S1M and S2M are cyclic, or finitely generated, provided that
the semimodule M is itself cyclic, or finitely generated, respectively.
(cid:3)
Proposition 6.2 The following conditions for a semiring S are equivalent:
(1) All cyclic (finitely generated) left S-semimodules are e-injective;
18
(2) S ≃ R ⊕ T , where R is a (classical) semisimple ring, and T is a zero-
sumfree semiring with an infinite element whose all cyclic (finitely generated) left
semimodules are e-injective.
Proof
(1) =⇒ (2). Let ≡V (S) be the Bourne congruence on S. It is clear that
the factor semiring S := S/ ≡V (S) is zerosumfree. So, by Proposition 6.1, every
cyclic (finitely generated) left S-semimodule is e-injective. In particular, SS is
e-injective and, by Proposition 4.1, contains an infinite element whence a zeroic
zerosumfree semiring. Whence, by [18, Proposition 2.9], S = R ⊕ T , where R is
a ring and T is a semiring isomorphic to S. Since R is a homomorphic image of
S, all cyclic (finitely generated) left R-modules are injective by Proposition 6.1,
and hence, by the celebrated Osofsky's Theorem ([41, Theorem, p. 649]) R is a
semisimple ring.
(2) =⇒ (1). This follows immediately from Proposition 6.1 and [41, Theorem;
p. 649].
(cid:3)
As was shown in [16], all finitely generated left S-semimodules over a semiring
S are injective if and only if a semiring S is a (classical) semisimple ring. How-
ever, as we will demonstrate below, in the general semiring setting the e-injective
version of this result is not true. From Proposition 6.2, we see that the prob-
lem of describing semirings all of whose cyclic (finitely generated) semimodules
are e- injective is actually reduced to the corresponding problem for the class of
zerosumfree semirings with infinite elements that, particularly, includes the very
important subclass of bounded distributive lattices. Therefore, it is quite natural
that a characterization of bounded distributive lattices all of whose cyclic (finitely
generated) semimodules are e-injective constitutes our next goal, to achieve which
we need first to establish the following important facts.
Lemma 6.3 If S is a bounded distributive lattice all of whose cyclic semimodules
are e-injective, then it is a Boolean algebra.
Proof First let us show that for any a ∈ S, we have Sa + Ann(a) = S, where
Ann(a) := (0 : a) is the annihilator of the element a. Indeed, suppose that Sa +
Ann(a) S, then Sa+Ann(a) is contained in a maximal ideal I of S. Obviously,
I, as any ideal of any bounded distributive lattice, is strongly subtractive.
It
implies that the set {0} ∪ {sa s ∈ S \ I} is a submonoid of the monoid (S, +, 0).
Moreover, by [14, Corollary 7.13], I is also a prime ideal of S. If sa 6= 0 for all
s ∈ S \ I, then the monoid {0} ∪ {sa s ∈ S \ I} becomes a left S-semimodule by
setting s(ra) = 0 for all s ∈ I and r /∈ I, and s(ra) = sra for all s /∈ I and r /∈ I.
Next let us take an arbitrary element m /∈ S and on the set V := {0} ∪ {sa s ∈
S \ I} ∪ {m} extend the operations of the S-semimodule {0} ∪ {sa s ∈ S \ I}
by setting m + 0 = 0 + m = m + m = m, m + sa = sa + m = sa for all
s ∈ S \ I, and sm = 0 for all s ∈ I, and sm = m for all s ∈ S \ I. One can
easily verify that V becomes a left S-semimodule. Obviously, M := {0, m} is a
19
subtractive S-subsemimodule of V . Clearly, M is cyclic and, hence, e-injective;
therefore, there exists an S-homomorphism f : V −→ M such that f M = idM .
It implies that f (sa) = m for all s ∈ S \ I. In particular, we have that f (a) = m.
On the other hand, since Sa + Ann(a) ⊆ I, we get that a ∈ I, and hence,
f (a) = f (a2) = af (a) = am = 0. Whence, sa = 0 for some s ∈ S \ I, that
is, s ∈ Ann(a) ⊆ I, and hence, s ∈ I; and therefore, this contradiction implies
S = Sa + Ann(a) for all a ∈ S.
Now we show that S is a Boolean algebra. Indeed, for any a ∈ S, we have
S = Sa + Ann(a); and hence, 1 = sa + x for some s ∈ S and x ∈ Ann(a). We
then have that
1 = a + 1 = a + sa + x = a(1 + s) + x = a1 + x = a + x.
For x ∈ Ann(a), we have xa = 0, and therefore, x is the complement to a, and S
is a Boolean algebra.
(cid:3)
Moreover, for the semiring S in the previous lemma, we can make a more
precise observation, namely:
Lemma 6.4 If S is a bounded distributive lattice all of whose cyclic semimodules
are e-injective, then S is a complete Boolean algebra.
Proof By [6, Theorem X.9] (or, [45, Theorem 8.5]) and Lemma 6.3, the Boolean
algebra S can be considered as a Boolean subalgebra of a complete Boolean
algebra B; and hence, S ⊆ B, 0S = 0B, and 1S = 1B, and B is naturally a left
S-semimodule. Define on the left S-semimodule S × B the relation ∼ as follows:
i, s′′
(s1, b1) ∼ (s2, b2) iff 1) b1 = b2 and 2) there exist n ∈ N, xi, x′
i ∈ S,
i=1 s′′
i = 1, . . . , n, such that xi + x′
i xi.
Repeating verbatim the proof of [17, Proposition 1.6], one easily sees that ∼ is a
congruence on S(S × B).
i = b1 for all i, and s1 +Pn
ixi = s2 +Pn
i ∈ B, s′
i=1 s′
Let eS := (S × B)/∼ and [s, b] be the class containing the pair (s, b). It is easy
to see that the image of the embedding µ : S eS, s 7−→ [s, 0], is a subtractive S-
semimodule of SeS. Hence, for SS is e-injective, there exists an S-homomorphism
ϕ : eS → S such that ϕµ = 1S. By the dual of [6, Theorem IX.2], in order to prove
S is complete, it is sufficient to show that every family {si}i∈I of elements in S
has a least upper bound s ∈ S. Since S ⊆ B and B is complete, a family {si}i∈I
has a least upper bound b ∈ B in B, and let s := ϕ([0, b]). For each i ∈ I, it is
easy to see that [si, b] = [0, b]: Indeed, (si, b) ∼ (0, b) by putting n = 1, x1 = b,
1 = 0, s′
x′
1 = si. Therefore,
1 = 0, s′′
si + s = ϕ(µ(si)) + ϕ([0, b]) = ϕ([si, 0] + [0, b]) = ϕ([si, b]) = ϕ([0, b]) = s,
and hence, s is an upper bound for {si}i∈I in S.
If y ∈ S is another upper
bound for {si}i∈I, then b ≤ y as b is the least upper bound for {si}i∈I in B; and
20
therefore, by = b and sy = ϕ([0, b])y = ϕ([0, by]) = ϕ[0, b] = s, i.e., s ≤ y and s
is a least upper bound for {si}i∈I in S.
(cid:3)
Now we are ready to characterize bounded distributive lattices all of whose
cyclic (finitely generated) semimodules are e-injective, namely:
Theorem 6.5 The following conditions for a bounded distributive lattice S are
equivalent:
(1) All finitely generated S-semimodules are e-injective;
(2) All cyclic S-semimodules are e-injective;
(3) S is a finite Boolean algebra.
Proof (1) =⇒ (2). It is obvious.
(2) =⇒ (3). By Lemma 6.4, S is a complete Boolean algebra. Suppose that
S is infinite, then, by [3, Lemma 4.2] for example, it contains a countable set of
orthogonal idempotents {en n ∈ N}, and we have the ideal I := Σn∈N Sen and
the factor algebra S := S/I. For S is a Boolean algebra and by Proposition 6.1,
we have that all cyclic S-semimodules are e-injective and, by Lemma 6.4 again, S
is a complete Boolean algebra. On the other hand, repeating verbatim the proof
of [3, Theorem 4.3], we have that S is not a complete Boolean algebra, and from
this contradiction we conclude that S is a finite Boolean algebra.
(3) =⇒ (1). Let S be a finite Boolean algebra and {e1, e2, ..., en} the set of all
atoms of S. Clearly, S = Se1 ⊕ Se2 ⊕ ... ⊕ Sen and Sei = {0, ei} ≃ B for each
i. Then, applying Theorem 4.4 (5) and Proposition 6.1, we conclude the proof.
(cid:3)
As a corollary of Theorem 6.5, we are able to extend the characterization of
finite Boolean algebras among bounded distributive lattices given in [3, Corollary
4.4]:
Corollary 6.6 The following conditions for a bounded distributive lattice S are
equivalent:
(1) All cyclic S-semimodules are projective;
(2) All subsemimodules of the regular semimodule SS are injective;
(3) All cyclic S-semimodules are e-injective;
(4) All cyclic S-semimodules are injective;
(5) S is a finite Boolean algebra.
As another consequence of Theorem 6.5, we obtain a complete description of
subtractive semirings all of whose finitely generated semimodules are e-injective.
Theorem 6.7 The following conditions for a left subtractive semiring S are
equivalent:
(1) All finitely generated S-semimodules are e-injective;
21
(2) All cyclic S-semimodules are e-injective;
(3) S ≃ R⊕T , where R is a semisimple ring and T is a finite Boolean algebra.
Proof (1) =⇒ (2). It is obvious.
(2) =⇒ (3). By Proposition 6.2, S ≃ R ⊕ T , where R is a semisimple ring and
T a semiring with an infinite element ∞ such that all cyclic left T -semimodules
are e-injective. By [25, Lemma 4.7], T is a left subtractive semiring. Whence, the
left ideal T ∞ is a subtractive ideal of T and it follows from 1T + ∞ = ∞ that
1T ∈ T ∞, i.e., t∞ = 1T for some t ∈ T . From the latter and ∞2 + ∞ = ∞, we
have ∞ = ∞ + 1T = 1T , and hence, x + 1T = 1T for all x ∈ T .
As, for each a ∈ T , the cyclic left T -semimodule T a is both e-injective and
subtractive subsemimodule of T by our hypothesis. Therefore, there exists a
homomorphism ϕ : T −→ T a such that ϕT a = 1T a. In particular, a = ϕ(a) =
ϕ(a1) = aϕ(1) and ϕ(1) ∈ T a. Hence, ϕ(1) = xa for some x ∈ T and a = axa.
From the latter, repeating verbatim the proof of [3, Theorem 4.6], we immediately
obtain that T is a bounded distributive lattice, and therefore, applying Theorem
6.5, we conclude that T is a finite Boolean algebra.
(3) =⇒ (1). This follows immediately from Proposition 6.1, Theorem 6.5, and
[41, Theorem, p. 649].
(cid:3)
By [19, Corollary 3.2], every S-semimodule over a semiring S can be canon-
ically represented as a colimit of a diagram of cyclic S-semimodules. In light of
this observation, Corollary 6.6 and Theorem 6.7, it seems to be reasonable and
interesting the following problem.
Problem 5 Describe all semirings S for which the classes of cyclic injective and
cyclic e-injective semimodules coincide.
Now, in what follows, we use the concepts and notations from 2.4, 2.5 and 2.6,
and following [2], we say that a right S-semimodule PS is e-flat iff the functor
P ⊗S − : SM −→ M preserves short exact sequences or equivalently iff the
functor P ⊗S − preserves subtractive subsemimodules.
Lemma 6.8 Every projective right S-semimodule is e-flat.
Proof As was shown in [22] and [24], any tensor product functor P ⊗S − :
SM −→ M has a right adjoint, and therefore, by [38, The dual Theorem 5.5.1],
preserves colimits, in particular, coproducts. Then, for projective semimodules
are retracts of free ones, the statement follows right away from [22, Proposition
3.8] and Corollary 3.3.
(cid:3)
Lemma 6.9 Let F : SM ⇄ T M : G be an equivalence between the semimodule
categories SM and T M. Then a left S-semimodule SM ∈ SM is e-injective iff
T F (M) ∈ T M is e-injective.
Proof By [26, Theorems 4.5 and 4.12], F ≃ P ⊗S − for some (T, S)-bisemimodule
P such that PS is a progenerator, P ∗ := HomS(PS, SS) is a progenerator in MT
22
and T ≃ End(PS) as semirings, and G ≃ P ∗ ⊗T −. Since projective semimodules
are e-flat by Lemma 6.8, both F and G preserve short exact sequences, and the
statement readily follows from the natural functor isomorphisms F G ≃ IdSM
and GF ≃ IdT M [38, Sect. 4.4] (see also the dual of [26, Lemma 4.10]).
(cid:3)
We conclude these section and paper by giving a complete characterization of
simple semirings all of whose finitely generated left semimodules are e-injective.
Theorem 6.10 The following conditions for a semiring S are equivalent:
(1) S is a simple semiring all of whose finitely generated left semimodules are
e-injective;
(2) S is a simple semiring all of whose cyclic left semimodules are e-injective;
(3) S is isomorphic either to a matrix semiring Mn(D) for some division
ring D and n ≥ 1, or to an endomorphism semiring End(L) of a nonzero finite
distributive lattice L.
Proof (1) =⇒ (2). It is obvious.
(2) =⇒ (3). By Proposition 6.2, S is a simple semisimple ring, or S is a
simple semiring with an infinite element. If S is a simple semisimple ring, then
S ≃ Mn(D) for some division ring D and n ≥ 1 by the classical Wedderburn-
Artin structure theorem for rings (see, e.g., [36, Wedderburn-Artin Theorem
3.5]). Otherwise, S is a simple semiring with an infinite element ∞. In particular,
∞+∞ = ∞, whence ∞ ∈ I +(S) and hence I +(S) is a nonzero ideal of S. For S is
a simple semiring, I +(S) = S and S is an additively idempotent simple semiring
containing an infinite element, and therefore, by [28, Theorem 5.7], S ≃ End(L)
for some nonzero finite distributive lattice L.
(3) =⇒ (1). Case I: Let S ≃ Mn(D) for some division ring D and n ≥ 1.
It follows by the Wedderburn-Artin Theorem and the celebrated Osofsky's result
[41, Theorem], all finitely generated left S-semimodules are (e-)injective.
Case II: Let S ≃ End(L) for some nonzero finite distributive lattice L. Then,
by [28, Theorem 5.7], S is a simple semiring that Morita equivalent to the Boolean
semiring B; and let the functors F : SM ⇄ BM : G establish an equivalence
between the semimodule categories SM and BM. For any finitely generated
left S-semimodule M ∈ SM, by [26, Proposition 4.8] and Theorem 4.4 (5),
the semimodule F (M) ∈ BM is a finitely generated e-injective B-semimodule.
Then, applying Lemma 6.9 and the natural isomorphism M ≃ G(F (M)), we have
that the semimodule M ∈ SM is e-injective as well and end the proof.
(cid:3)
Acknowledgement: The first and fourth authors would like to acknowledge
a support provided by the Deanship of Scientific Research (DSR) at King Fahd
University of Petroleum & Minerals (KFUPM) for funding this work through
project RG1304.
23
References
[1] J. Abuhlail, Exact sequences of commutative monoids and semimodules, Homology
Homotopy Appl. 16 (1) (2014), 199-214.
[2] J. Abuhlail, Some remarks on tensor products and flatness of semimodules, Semi-
group Forum 88 (3) (2014), 732-738.
[3] J. Y. Abuhlail, S. N. Il'in, Y. Katsov, T. G. Nam, On V -Semirings and Semirings
all of whose Cyclic Semimodules are Injective, Communications in Algebra, 43
(2015), 4632–4654.
[4] F. W. Anderson and K. R. Fuller, Rings and Categories of Modules, 2nd ed.,
Springer-Verlag, New York-Berlin, 1979.
[5] R. El Bashir, J. Hurt, A. Jancar´ık, and T. Kepka, Simple Commutative Semirings,
J. Algebra, 236 (2001), 277–306.
[6] G. Birkhoff, Lattice Theory, American Mathematical Society, Providence, 1967.
[7] F. Borceux, Handbook of Categorical Algebra. I, Basic Category Theory,
Cambridge Univ. Press (1994).
[8] A. Connes and C. Consani, Schemes over F1 and zeta functions, Compos. Math.,
146 (2010), 1383–1415.
[9] J. Giansiracusa and N. Giansiracusa, Equations of tropical varieties, to appear in
Duke Mathematical Journal (see also preprint: arXiv:1308.0042v2).
[10] T. S. Fofanova, Injectivity of polygons over Boolean algebras, Siberian Math. J.,
13 (1972), 452–458 (in Russian).
[11] T. S. Fofanova, Polygons over distributive lattices, in: Universal Algebra, Colloq.
Math. Soc. J´anos Bolyai # 29, North-Holland Publishing Co., Amsterdam, 1982,
289–292.
[12] S. I. Gelfand, Y. I. Manin, Methods of homological algebra, Second edition,
Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2003
[13] K. G lazek(2001). A Guide to the Literature on Semirings and their Applica-
tions in Mathematics and Information Science, Kluwer Academic Publishers,
Dordrecht-Boston-London, 2001
[14] J. S. Golan, Semirings and their Applications, Kluwer Academic Publishers,
Dordrecht-Boston-London, 1999.
24
[15] A. Horn, N. Kimura, The category of semilattices, Algebra Universalis, 1 (1971),
26–38.
[16] S. N. Il'in, Semirings over Which all Semimodules are Injective (Projective),
Matem. Vestn. Pedvuzov i Univ. Volgo-Vyatsk. Regiona, 8 (2006), 50–53.
[17] S. N. Il'in, Direct Sums of Injective Semimodules and Direct Products of Projective
Semimodules Over Semirings, Russian Mathematics, 54 (2010), 27 - 37.
[18] S. N. Il'in, V-semirings, Siberian Mathematical Journal, 53 (2012), 222 - 231.
[19] S. N. Il'in, Y. Katsov, T.G. Nam, Toward Homological Structure Theory of Semi-
modules: On Semirings All of Whose Cyclic Semimodules Are Projective, preprint:
arXiv:1509.02997v1.
[20] Z. Izhakian, L. Rowen, Supertropical algebra, Adv. Math., 225(4) (2010), 2222–
2286.
[21] J. Jun, Cech cohomology of semiring schemes, preprint: arXiv:1503.01389v1.
[22] Y. Katsov, Tensor products and injective envelopes of semimodules over additively
regular semirings, Algebra Colloquium, 4 (1997), 121-131.
[23] Y. Katsov, On flat semimodules over semirings, Algebra Universalis, 51 (2004),
287-299.
[24] Y. Katsov, Toward homological characterization of semirings: Serre's conjecture
and Bass's perfectness in a semiring context, Algebra Universalis, 52 (2004), 197-
214.
[25] Y. Katsov, T. G. Nam, N. X. Tuyen, On subtractive semisimple semirings, Algebra
Colloquium, 16 (2009), 415-426.
[26] Y. Katsov and T. G. Nam, Morita Equivalence and Homological Characterization
of Semirings, J. Algebra Appl., 10 (2011), 445 - 473.
[27] Y. Katsov, T. G. Nam, N. X. Tuyen, More on Subtractive Semirings: Simpleness,
Perfectness and Related Problems, Comm. Algebra, 39 (2011), 4342 - 4356.
[28] Y. Katsov, T. G. Nam, J. Zumbragel, On Simpleness of Semirings and Complete
Semirings, J. Algebra Appl., 13: 6 (2014). DOI: 10.1142/S0219498814500157.
[29] Y. Katsov, T. G. Nam, On radicals of semirings and related problems,
Comm. Algebra 42 (2014), no. 12, 5065–5099.
25
[30] Y. Katsov, T. G. Nam, J. Zumbragel, Simpleness of Leavitt path algebras
with coefficients in commutative semiring, Semigroup Forum, 2016. DOI
10.1007/s00233-016-9781-1.
[31] A. Kendziorra and J. Zumbragel, Finite simple additively idempotent semirings,
J. Algebra, 388 (2013), 43–64.
[32] T. Kepka, J. Kortelainen and P. Nemec, Simple semirings with zero, J. Algebra
Appl.. 15: 3 (2016) 1650047. DOI: 10.1142/S021949881650047X.
[33] T. Kepka and P. Nemec, Simple semirings with left muliplicatively absorbing
elements, Semigroup Forum, 91 (2015), 159–170.
[34] M. Kilp, U. Knauer, A. V. Mikhalev, Monoids, Acts and Categories, Walter de
Gruyter, Berlin-New York, 2000.
[35] T. Y. Lam, Lectures on Modules and Rings, Springer-Verlag, New York-Berlin,
1999.
[36] T. Y. Lam, A first course in noncommutative rings, 2nd Ed., Springer-Verlag,
New York-Berlin, 2001.
[37] O. Lorscheid, The geometry of blueprints. Part I: Algebraic background and
scheme theory, Adv. Math., 229(3) (2012), 1804–1846.
[38] S. Mac Lane, Categories for the Working Mathematician, Springer-Verlag, New
York-Berlin, 1971.
[39] G. Maze, C. Monico, and J. Rosenthal, Public key cryptography based on semi-
group actions, Adv. Math. Commun., 1 (2007), 489–507.
[40] C. Monico, On finite congruence-simple semirings, J. Algebra, 271 (2004), 846–
854.
[41] B. L. Osofsky, Rings all of whose finitely generated modules are injective, Pacific
J. Math., 14 (1964), 645 - 650.
[42] A. Patchkoria, Extensions of Semimodules and Takahashi functor ExtΛ(C, A),
Homology, Homotopy and Appl., 5 (2013), 387 - 406.
[43] J. Richter-Gebert, B. Sturmfels, T. Theobald, First steps in tropical geometry,
in: Idempotent Mathematics and Mathematical Physics, in: Contemp. Math., vol.
377, American Mathematical Society, Providence, RI, 2005, pp. 289–317.
[44] H. Schubert, Categories, Springer-Verlag, New York-Heidelberg, 1972.
26
[45] L. A. Skornyakov, L. A. `Elementy teorii struktur (Russian) [Elements of
lattice theory], Second edition, "Nauka", Moscow, 1982.
[46] M. Takahashi, On the bordism categories. II. Elementary properties of semimod-
ules, Kobe J. Math., 9 (1981), 495-530.
[47] M. Takahashi, On the bordism categories. III. Functors Hom and for semi-
modules, Math. Sem. Notes Kobe Univ. 10 (2) (1982), 551-562.
[48] M. Takahashi, Completeness and c-cocompleteness of the category of semimodules,
Kobe J. Math., 10 (1982), 551–562.
[49] M. Takahashi, Extensions of semimodules I, Kobe J. Math., 10 (1982), 563–592.
[50] J. Zumbragel, Classification of finite congruence-simple semirings with zero, J.
Algebra Appl., 7 (2008), 363–377.
27
|
1512.07676 | 2 | 1512 | 2016-01-11T11:10:04 | Cohomology of $\mathbb{N}$-graded Lie algebras of maximal class over $\mathbb{Z}_2$ | [
"math.RA"
] | We compute the cohomology with trivial coefficients of Lie algebras $\mathfrak{m}_0$ and $\mathfrak{m}_2$ of maximal class over the field $\mathbb{Z}_2$. In the infinite-dimensional case, we show that the cohomology rings $H^*(\mathfrak{m}_0)$ and $H^*(\mathfrak{m}_2)$ are isomorphic, in contrast with the case of the ground field of characteristic zero, and we obtain a complete description of them. In the finite-dimensional case, we find the first three Betti numbers of $\mathfrak{m}_0(n)$ and $\mathfrak{m}_2(n)$ over $\mathbb{Z}_2$. | math.RA | math | COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL
CLASS OVER Z2
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
Abstract. We compute the cohomology with trivial coefficients of Lie algebras m0
and m2 of maximal class over the field Z2. In the infinite-dimensional case, we show
that the cohomology rings H ∗(m0) and H ∗(m2) are isomorphic, in contrast with the
case of the ground field of characteristic zero, and we obtain a complete description of
them. In the finite-dimensional case, we find the first three Betti numbers of m0(n) and
m2(n) over Z2.
6
1
0
2
n
a
J
1
1
]
.
A
R
h
t
a
m
[
2
v
6
7
6
7
0
.
2
1
5
1
:
v
i
X
r
a
1. Introduction
A Lie algebra g is said to be N-graded, if it is the direct sum of subspaces gi, i ∈ N
(the homogeneous components), such that [gi, gj] ⊂ gi+j. Obviously, finite-dimensional
N-graded Lie algebras are necessarily nilpotent. A great deal of attention in the literature
has been focused on N-graded Lie algebras for which the homogeneous components gi
are "the smallest possible", that is, all of dimension one or, in the finite-dimensional
case, dim gi = 1, for i ≤ n := dim g, and gi = 0, for i > n. With the additional condition
that g is generated as an algebra by elements e1 and e2, spanning g1 and g2 respectively,
one obtains that the subspaces C0 = g, Ck = ⊕∞
i=k+2gi, k > 0, are the terms of the
central descending series. This defines the N-graded filiform Lie algebras in the finite-
dimensional case [15] and the N-graded Lie algebras of maximal class [12] (also called
narrow algebras). In characteristic zero, these algebras have been completely classified.
In the infinite-dimensional case, one gets just three algebras [7], and independently [12,
Theorem 7.1]. We list them here with their presentations:
m0 = Span(e1, e2, . . . ),
m2 = Span(e1, e2, . . . ),
V = Span(e1, e2, . . . ),
[e1, ei] = ei+1, i > 1,
[e1, ei] = ei+1, i > 1,
[ei, ej] = (j − i)ei+j, i, j ≥ 1.
[e2, ej] = ej+2, j > 2,
(1)
(2)
(3)
In the finite-dimensional case in characteristic zero, the classification of finite-dimensional
N-graded filiform Lie algebras was established in [11]: one obtains the "truncations" of
the above three algebras, in particular,
m0(n) = Span(e1, . . . , en), [e1, ei] = ei+1, 1 < i < n,
(4)
m2(n) = Span(e1, . . . , en), [e1, ei] = ei+1, 1 < i < n, [e2, ej] = ej+2, 2 < j < n − 1, (5)
and V(n), plus another three infinite series, and five one-parameter families of low-
dimensional algebras. The picture is more complicated in positive characteristic: by [5],
2010 Mathematics Subject Classification. 17B56, 17B50, 17B70, 17B65, 17B30.
Key words and phrases. Lie algebra of maximal class, characteristic 2, cohomology, Betti number.
The first named author was partially supported by ARC Discovery grant DP130103485.
1
2
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
there are uncountably many isomorphism classes of Lie algebras of maximal class; the
construction of all such algebras in odd characteristic is given in [6], and in characteristic
two, in [10], with m0 and m2 being the simplest possible cases.
The cohomology of N-graded Lie algebras of maximal class has been studied exten-
sively over a field of characteristic zero [7, 8, 15], and at present is well-understood. In
[8], Fialowski and Millionschikov gave a full description of the cohomology with trivial
coefficients of the algebras m0 and m2; the Betti numbers of V are found in [9]. In the
finite-dimensional case, the cohomology of m0(n) were found in [3] (see also [2] and [8]).
However, already for m2(n) over a field of characteristic zero, our present knowledge is
limited to the first two Betti numbers [11, 15].
The study of the cohomology of Lie algebras of maximal class over fields of positive
characteristic is much less developed. The cohomology of the Heisenberg algebra is found
in [4, 13]. A recent result by Tsartsarflis [14] states that over a field of characteristic
two, the algebras m0(n) and m2(n) have the same Betti numbers (in contrast with the
case of characteristic zero), and furthermore, every algebra of the so called Vergne class
admits a dual, non-isomorphic algebra, with the same Betti numbers.
In this paper we study the cohomology with trivial coefficients of the Lie algebras m0
and m2, and their finite dimensional truncations, m0(n) and m2(n), over the field Z2. Let
V = Span(e1, e2, . . . ) and let {ei} be the dual basis for V ∗. Define the operator D1 on
V ∗ by D1e1 = D1e2 = 0, D1ei = ei−1, for i > 2, and extend it to Λ(V ) as a derivation.
1ω) ∧ ei+l+1 (note that the sum on
For ω ∈ Λ(V ) and ei ∈ V ∗, define F (ω, ei) = P∞
the right-hand side is finite).
l=0(Dl
Our main result in the infinite-dimensional case is as follows.
Theorem 1. The cohomology rings H ∗(m0) and H ∗(m2) over the field Z2 are isomorphic.
The respective cohomology classes of the cocycles
e1, e2, F (ei1 ∧ ei2 ∧ · · · ∧ eiq , eiq ),
(6)
where q ≥ 1, 2 ≤ i1<i2<. . .<iq, form a basis for H ∗(m0) and for H ∗(m2), respectively.
Note that H ∗(m0) over Z2 is "the same" as over a field of characteristic zero (compare
In contrast, the fact that H ∗(m0) and H ∗(m2) over Z2 are
with [8, Theorem 3.4]).
isomorphic (note that m0 and m2 are not isomorphic over any ground field) is specific to
the Z2 case: over a field of characteristic zero, H ∗(m2) is very different [8, Theorem 5.5].
In the finite-dimensional case, which appears to be substantially harder that the
infinite-dimensional one, we compute the first three Betti numbers of m0(n) and the
corresponding bases for H i(m0(n)), i = 1, 2, 3.
Theorem 2. The first three Betti numbers of the Lie algebra m0(n) over Z2 are given
by
(a) b1(m0(n)) = 2,
(b) b2(m0(n)) = ⌊ 1
(c) b3(m0(n)) = 1
2 (n + 1)⌋, where ⌊.⌋ denotes the integer part,
3(2p − 1)(2p−1 − 1) + 1
2(n − 1)⌋, where n = 2p + m and
2m(m − 1) + ⌊ 1
0 < m ≤ 2p.
An explicit form of the basis for H 3(m0(n)) is given in Theorem 4 of Section 3. The-
orem 2 also gives us the first three Betti numbers of m2(n) (Corollary 1 of Section 4),
which in characteristic two are simply the same as those for m0(n), by [14, Theorem 1].
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
3
The paper is organised as follows. We begin with some short preliminaries in Section 2.
We treat the algebras m0 and m0(n) in Section 3. Parts (a) and (b) of Theorem 2 follow
from Proposition 1. After some technical preparation similar to the arguments of [8], we
prove Theorem 3, which is "the m0-part" of Theorem 1. We then proceed to the proof
of Theorem 2(c). This is the longest and most technically involved part of the paper.
Finally, in Section 4 we use a construction similar to [14] to establish the isomorphism
between H ∗(m0) and H ∗(m2), hence completing the proof of Theorem 1.
2. Preliminaries
Given a Lie algebra g over Z2 with a basis elements ei, we denote the dual basis
elements ei. For convenience, we set e0 = 0. For simplicity we write a monomial q-form
ei1 ∧ ei2 ∧ · · · ∧ eiq ∈ Λq(g) as ei1i2...iq. For a monomial ei1i2...iq , its degree is defined to be
k(g) of degree k and of rank q is the span of all
Pq
j=1 ij. The homogeneous component Λq
the monomials of degree k and of rank q. We set Λk(g) := ⊕qΛq
k(g).
As usual, the differential d is defined by dξ(X, Y ) = ξ[X, Y ] for one-forms ξ, where
X, Y ∈ g, and then is extended to the exterior algebra Λ(g) as a derivation (so that
d(ω1∧ω2) = d(ω1)∧ω2+ω1∧d(ω2)). Then d2 = 0 and one define the q-th cohomology group
H q(g) (with trivial coefficients) by H q(g) = ker(d : Λq → Λq+1)/Im (d : Λq−1 → Λq).
Then H q(g) is a linear space over Z2; if its dimension is finite, it is called the q-th Betti
number bq(g). It is immediate from the definition that if dim g = n, then
bq(g) = dim ker(d : Λq → Λq+1) + dim ker(d : Λq−1 → Λq) −(cid:18) n
q − 1(cid:19),
(7)
so to compute the Betti numbers it suffices to know the dimensions of the kernels of d
on the Λq's. Also note that in the graded case (in particular, for the bases {ei} from (1
-- 5)), the operator d maps Λq
(g), and so H q(g) is spanned by the classes of
homogeneous elements; we get a decomposition (a bi-gradation) H q(g) = ⊕kH q
k(g). The
multiplicative structure in H(g) := ⊕qH q(g) is inherited from the wedge product.
k(g) to Λq+1
k
3. Cohomology of m0
In this section, we compute the cohomology of the infinite-dimensional Lie algebra
m0 and also the first three Betti numbers of the finite-dimensional Lie algebras m0(n)
defined as follows (1, 4):
m0 = Span(e1, e2, e3, . . . ),
m0(n) = Span(e1, e2, e3, . . . , en),
[e1, ei] = ei+1,
[e1, ei] = ei+1,
for i ≥ 2,
for 2 ≤ i ≤ n − 1.
In the first few paragraphs, we closely follow the approach and the results of [8,
Section 3], adapting them to the case of the ground field Z2.
In effect, the outcome
is that in the infinite-dimensional case, for g = m0, the cohomology is "the same" as
that for a field of characteristic zero, while in the finite-dimensional case, for g = m0(n),
the situation is more delicate -- not only the Betti numbers are different, but also the
methods of [8, 2] and the very elegant approach of [3, Appendix B] do not work directly.
4
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
For a monomial ei1i2...iq ∈ Λq(g), q ≥ 1, i1, i2, . . . , iq ≥ 1, (for both g = m0 and
g = m0(n)) we have
d(ei1i2...iq) = e1(i1−1)i2...iq + e1i1(i2−1)...iq + · · · + e1i1i2...(iq−1)
= e1 ∧(cid:0)e(i1−1)i2...iq + ei1(i2−1)...iq + · · · + ei1i2...(iq−1)(cid:1).
(8)
It follows from (8) that the subspaces Λk(g) are d-invariant.
Moreover, for any ω ∈ Λ(g) we have d(e1 ∧ ω) = 0 and d(ω) ∈ e1 ∧ Λ(g). Set
h := Span(e2, e3, . . . ) for m0, and h := Span(e2, e3, . . . , en) for m0(n). Then h is abelian
and from (8) it follows that there is a well-defined linear operator D on Λ(h) such that
for ω ∈ Λ(h), we have
dω = e1 ∧ (Dω).
(9)
It is easy to see that
De2 = 0, Dei = ei−1 for i > 2,
D(ξ ∧ η) = D(ξ) ∧ η + ξ ∧ D(η) for ξ, η ∈ Λ(h), (10)
so D is a derivation of Λ(h). Recall that the Lie derivative with respect to e1 is defined
by taking the operator (ade1)∗ on g∗ to be the dual to ade1 on g, and then extending it as
a derivation to Λ(g). Note that D is just the restriction of (ade1)∗ to Λ(h). Furthermore,
D(Λq
for any ω ∈ Λ(h) there exists N =
N(ω) ≥ 0 such that DN ω = 0. For convenience, we define D0 to be the identity map.
k−1(h), so that D is "nilpotent":
k(h)) ⊂ Λq
Since from (8), ker d = e1 ∧ Λ(h) ⊕ ker D, to find the kernel of d we need to find the
kernel of D. This is given by the following lemma.
Lemma 1. (a) Let g = m0. For any ω ∈ Λ(h) and ei ∈ h define
F (ω, ei) = X∞
l=0
Dlω ∧ ei+1+l = XN (ω)−1
l=0
Dlω ∧ ei+1+l.
(11)
Then F (ω, ei) ∈ ker D for ω ∧ ei = 0 and moreover, the elements
F (ei1i2...iq, eiq ) = ei1i2...iq iq+1 + Dei1i2...iq ∧ eiq+2 + · · · ∈ Λq+1
k
(h),
where q ≥ 1, 2 ≤ i1 < i2 < · · · < iq, k = iq + 1 +Xq
j=1
ij,
(12)
form a basis for the kernel of the restriction of D to Λq+1
restriction of D to h∗ is spanned by e2.
k
(h); the kernel of the
(b) Let g = m0(n), viewed as the subspace of m0 spanned by the first n vectors. Then
ker D is the intersection of ker D constructed in (a) for the case g = m0 with m0(n).
Note that in the Introduction we used D1 = (ade1)∗ rather than D to define F .
This yields the same object, since in (6), D only acts on elements of Λ(h) and D is
the restriction on D1 to Λ(h). Notice however that Lemma 1 concerns ker D, which is
different to ker D1.
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
5
Proof. (a) The fact that F (ω, ei) ∈ ker D follows immediately, as from (10), for any
ω ∈ Λ(h) and ei ∈ h we have
DF (ω, ei) = D(cid:16)X∞
l=0
Dlω ∧ ei+1+l(cid:17)
Dlω ∧ ei+l
l=0
Dlω ∧ ei+l
l=0
= X∞
= X∞
l=1
= ω ∧ ei,
Dl+1ω ∧ ei+1+l +X∞
Dlω ∧ ei+l +X∞
l=0
as we are working over Z2. Notice in passing that this also shows that D is surjective.
The fact that the elements given by (12) are linearly independent is also easy, as
from among the monomials ej1j2...jqjq+1, 2 ≤ j1 < j2 < · · · < jq < jq+1 which appear
on the right-hand side of the expansion of F (ei1i2...iq , eiq), there is exactly one with the
property that jq+1 = jq + 1, namely the monomial ei1i2...iq iq+1. The fact that they indeed
span the kernel of the restriction of D to Λq+1
(h) follows from the same observation
and from the dimension count. The elements F (ei1i2...iq , eiq) ∈ Λq+1
(h) with q ≥ 1, 2 ≤
j=1 ij = k, are in one-to-one correspondence with the
elements ej1j2...jqjq+1 ∈ Λq+1
(h) with 2 ≤ j1 < j2 < · · · < jq. On the other hand,
consider the linear operator A : Λq+1
k−1(h) defined on the monomials as follows:
Aej1j2...jqjq+1 = ej1j2...jqjq+1−1. Then A is surjective and its kernel is spanned by the
monomials ej1j2...jqjq+1, so every surjective linear operator from Λq+1
k−1(h) (in
particular, D) has a kernel of the same dimension.
i1 < i2 < · · · < iq, iq + 1 + Pq
(h) to Λq+1
(h) → Λq+1
k
k
k
k
k
(b) easily follows from the fact that for the operator D defined for g = m0, the subspace
Λ(h) defined for m0(n) is D-invariant, and the restriction of D to it is the operator D
defined for m0(n).
(cid:3)
With Lemma 1 we can easily finish the computation of the cohomology for g = m0;
we obtain the same answer as in [8, Theorem 3.4]:
Theorem 3. The cohomology classes of the cocycles
e1, e2, F (ei1i2...iq , eiq),
(13)
where q ≥ 1, 2 ≤ i1<i2<. . .<iq, form a basis for H ∗(m0) over the field Z2.
Furthermore, the dimensions of the homogeneous components of H ∗(m0) over Z2 are
the same as those over a field of characteristic zero, so in particular,
dim H q
k+ q(q+1)
2
(m0) = Pq(k) − Pq(k − 1),
where Pq(k) is the number of partitions of a positive integer k into q parts. The products
of the basis elements also have "the same" decomposition as in [8, Equation (8)], after
reducing the coefficients modulo 2.
Proof of Theorem 3. From Lemma 1(a) we know ker D, and so we know ker d = e1 ∧
Λ(h) ⊕ ker D. The image of d is just e1 ∧ Λ(h), by (9) and from the surjectivity of
D (which has been established in the proof of Lemma 1(a)). Putting these two facts
together we get the claim.
(cid:3)
6
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
We now turn our attention to the case g = m0(n). We view m0(n) as a subspace of m0
spanned by the first n basis elements and for convenience, denote the operator D defined
for m0 by D. The following Proposition easily follows from Lemma 1.
Proposition 1. The space H 1(m0(n)) is spanned by the classes of the elements e1, e2
and so b1(m0(n)) = 2. The space H 2(m0(n)) is spanned by the classes of the elements
e1n, F (ei, ei) = ei,i+1 + ei−1,i+3 + · · · + e2,2i−1, 2 ≤ i ≤ 1
2(n + 1), and so b2(m0(n)) =
⌊ 1
2(n + 1)⌋.
Proof. The claim for H 1(m0(n)) is clear. For the second cohomology, by Lemma 1(a),
the kernel of D is spanned by the elements F (ei, ei) = ei,i+1 + ei−1,i+3 + · · ·+ e2,2i−1. Since
a sum of some number of the F (ei, ei) belongs to m0(n) if and only if each of them does
(no two monomials of the different F (ei, ei) may possibly cancel), we get by Lemma 1(b):
ker D = Span(F (ei, ei) : 2 ≤ i ≤ 1
2 (n + 1)).
(14)
Then ker d = e1 ∧ Λ1(h) ⊕ ker D and so the second coboundary space is spanned by
e1i, F (ei, ei), i = 2, . . . , n − 1. Then, as the image of d on the space of one-forms is
spanned by e1 ∧ ei, for 1 ≤ i ≤ n − 1, the claim follows.
(cid:3)
Proposition 1 establishes parts (a) and (b) of Theorem 2. The first two Betti numbers
of m0(n) over Z2 are the same as those over a field of characteristic zero [2], but b3 is
different, as Theorem 2(c) shows.
Remark 1. Explicitly, for small values of n, Theorem 2(c) gives:
n
10 11 12 13 14 15 16 17 18 19 20
b3(m0(n)) 1 2 3 4 7 10 11 12 15 18 23 28 35 42 43 44 47 50
3 4 5 6 7
8
9
The sequence b3(m0(n)) is the sequence A266540 in [1]1. To see that, we note that by
the formula given in Theorem 2(c), b3(m0(n)) = 1
2(b3(m0(n − 1)) + b3(m0(n + 1))), for odd
n ≥ 3, and so it suffices to show that the even terms of the two sequences coincide, which
is equivalent to the fact that the sequence Al := 1
3(22p−2 − 1) + s2, where
l = 2p−1 + s, 0 < s ≤ 2p−1, coincides with A256249. This is equivalent to the fact that
Al is the (l − 1)-st partial sum of the sequence A006257 given by aj = 2(j − 2⌊log2 j⌋) + 1.
2b3(m0(2l)) = 1
But the latter partial sum equals l2 − 1 − 2(2p−1s +Pp−2
i=0 22i), and the claim follows.
The proof of Theorem 2(c) is based on the following Proposition. For brevity, let us
denote the vector space Λ3(e2, . . . , en−1) by W . Denote h = Span(e2, . . . , en).
Proposition 2. For m as defined in Theorem 2, there exists ωk ∈ W for 2 ≤ k ≤ m
such that
ker DΛ3(h) = ker DW ⊕ Span(en ∧ F (ek, ek) + ωk : 2 ≤ k ≤ m).
We first prove the theorem assuming the Proposition.
1The authors are thankful to Omar E. Pol for pointing this out.
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
7
Proof of Theorem 2(c). For n = 3 the statement is easily verified: H 3(m0(3)) is spanned
by the class of the single element e123, so b1(m0(3)) = 1, as claimed.
Assume n ≥ 4. Denote dn the dimension of the kernel of the operator D constructed
for the algebra m0(n). Then from Proposition 2 we have dn = dn−1 + m − 1. It follows
that for n = 2p + m, 0 < m ≤ 2p, we have dn = d2p + 1
2m(m − 1) and in particular,
d2p+1 = d2p + 2p−1(2p − 1).
(15)
We also have d4 = 1, as for m0(4) the space ker D is spanned by e234. It follows from
(15) that d2p = 1
3(2p − 1)(2p−1 − 1), and so dn = 1
3(2p − 1)(2p−1 − 1) + 1
2m(m − 1).
We have
dim ker(d : Λ3(m0(n)) → Λ4(m0(n))) = dn + dim(e1 ∧ Λ2(m0(n)) = dn +
1
2
(n − 1)(n − 2).
On the other hand, from Proposition 1,
dim ker(d : Λ2(m0(n) → Λ3(m0(n)) = (n − 2) + ⌊ 1
2(n + 1)⌋,
and so the claim follows from (7).
(cid:3)
Proof of Proposition 2. Any ω ∈ Λ3(h) can be uniquely represented as ω = en ∧ ξ + ω′,
with ξ ∈ Λ2(e2, . . . , en−1), ω′ ∈ Λ3(e2, . . . , en−1) = W . For ω to belong to ker D it is nec-
essary that Dξ = 0 (so that Dω does not contain en). From the proof of Proposition 1 it
follows that ξ must be a linear combination of F (ek, ek), k = 2, . . . , ⌊n/2⌋. Extracting the
homogeneous components we obtain that the proposition is equivalent to the following
statement: for 2 ≤ k ≤ ⌊n/2⌋, there exists ωk ∈ W such that en ∧ F (ek, ek) + ωk ∈ ker D,
if and only if k ≤ m.
The next step in the proof is the following lemma.
Lemma 2. For n ≥ 4 and 2 ≤ k ≤ ⌊n/2⌋, define a = ⌈(n+2k +1)/3⌉, b = ⌊n/2⌋+k −1.
There exists ωk ∈ W such that en ∧ F (ek, ek) + ωk ∈ ker D if and only if the linear system
Ax = (1, 0, . . . , 0)t ∈ Zk−1
, where A is the (k − 1) × (b − a + 1)-
matrix given by
has a solution x ∈ Zb−a+1
2
2
Aij = (cid:18)n − (a + j − 1) + 2(i − 1)
and as usual we set (cid:0)N
(a + j − 1) + (i − 1) − k (cid:19) mod 2,
t(cid:1) = 0 if t < 0 or t > N .
1 ≤ i ≤ k − 1, 1 ≤ j ≤ b − a + 1, (16)
Proof. Suppose for some ωk ∈ W , the three-form ω = en ∧ F (ek, ek) + ωk belongs to
ker D (where 2 ≤ k ≤ ⌊n/2⌋). Without loss of generality we can assume that ωk is
homogeneous, of the same degree as en ∧ F (ek, ek), so that ω is homogeneous of degree
n + 2k + 1.
By Lemma 1, the form ω viewed as a three-form on m0, lies in the kernel of D and so
is a linear combination of the forms F (es,r, er), 2 ≤ s < r, where by homogeneity we can
assume that s + 2r + 1 = n + 2k + 1, from which it follows that s = n + 2k − 2r. Then
8
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
2 ≤ s ≤ r − 1 gives a ≤ r ≤ b. Therefore for some µr ∈ Z2, r = a, . . . , b we have
µrF (en+2k−2r,r, er)
ω = F (ek, ek) ∧ en + ωk = Xb
= Xb
= X∞
µrX∞
l=0Xb
r=a
r=a
Dl(en+2k−2r,r) ∧ el+r+1
l=0
µrDl(en+2k−2r,r) ∧ el+r+1.
r=a
(17)
r=a
ξN := Pmin{b,N −1}
As n + 2k − 2r = s < r ≤ b and b = ⌊n/2⌋ + k − 1 ≤ 2⌊n/2⌋ − 1 < n, no
terms Dl(en+2k−2r,r) in the latter expression may possibly contain eN , N ≥ n.
It
follows that the only terms containing eN with N ≥ n in (17) are ξN ∧ eN , where
µrDN −r−1(en+2k−2r,r). In fact, since ω ∈ Λ3(m0(n)), we have ξN = 0
for all N > n and equating the terms containing en we get ξn = F (ek, ek). Conversely,
if ξn = F (ek, ek), then ξN = 0 for all N > n, as ξn+1 = Dξn = DF (ek, ek) = 0, ξn+2 =
D2ξn = D2F (ek, ek) = 0, and so on. Thus a necessary and sufficient condition for the
existence of ωk ∈ W such that the three-form ω = en ∧ F (ek, ek) + ωk belongs to ker D
is the existence of µr ∈ Z2, r = a, . . . , b such that
F (ek, ek) = ξn = Xb
r=a
µrDn−r−1(en+2k−2r,r).
(18)
(the summation on the right-hand side is up to b as b ≤ n − 1). Note that both sides
are homogeneous two-forms of degree 2k + 1. Recall that F (ek, ek) = ek,k+1 + ek−1,k+2 +
· · · + e2,2k−1, and observe that
Dn−r−1(en+2k−2r,r) = Xn−r−1
i=0
(cid:0)n−r−1
i
(cid:1)e2k−r+i+1,r−i.
So expanding and equating coefficients of the corresponding monomials we see that (18)
is equivalent to the following system:
r=a
Xb
Xb
r=a
µr(cid:0)(cid:0)n−r−1
µr(cid:0)(cid:0) n−r−1
r−k (cid:1) +(cid:0) n−r−1
r−(k−1)(cid:1) +(cid:0) n−r−1
r−(k+1)(cid:1)(cid:1) = 1 mod 2,
r−(k+2)(cid:1)(cid:1) = 1 mod 2,
...
r=a
coefficients (cid:0)2s−1
r−(2k−1)(cid:1)(cid:1) = 1 mod 2.
Now the linear combination of the first s ≤ k − 1 of the above equations with the
µr(cid:0)(cid:0)n−r−1
r−2 (cid:1) +(cid:0) n−r−1
1 (cid:1),(cid:0)2s−1
Xb
s−2(cid:1), . . . ,(cid:0)2s−1
s−1(cid:1),(cid:0)2s−1
0 (cid:1) respectively gives
r − k + s − 1 (cid:19)
µr(cid:18)n − r + 2s − 2
r − k − s + i(cid:19)(cid:19) = Xb
i (cid:19)(cid:18) n − r − 1
i=0 (cid:18)2s − 1
µr(cid:18)X2s−1
t+i(cid:1) = Pl
on the left-hand side (as Pl
l−i(cid:1)(cid:0) N
i=0(cid:0) l
i(cid:1)(cid:0) N
i=0(cid:0)l
t+l(cid:1) by Vandermonde's
s−1(cid:1) + (cid:0)2s−1
identity). On the right-hand side we obtain (cid:0)2s−1
0 (cid:1) =
2 × 22s−1 = 22s−2, which is odd when s = 1 and even otherwise. Thus the above system
of equations is equivalent to the following one:
t+i(cid:1) = (cid:0)N +l
s−2(cid:1) + · · · + (cid:0)2s−1
1 (cid:1) + (cid:0)2s−1
Xb
r=a
r=a
1
Xb
r=a
µr(cid:0)n−r
r−k(cid:1) = 1 mod 2, Xb
r=a
µr(cid:0)n−r+2s−2
r−k+s−1(cid:1) = 0 mod 2, for 2 ≤ s ≤ k − 1.
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
This is equivalent to the claim of the lemma if we define x = (µa, µa+1, . . . , µb)t.
9
(cid:3)
In order to use Lemma 2 to conclude the proof of the proposition, we need to show
that the system Ax = (1, 0, . . . , 0)t has a solution if and only if k ≤ m. Even though we
are working over Z2, let us say that vectors x, y are orthogonal if xty = 0.
To prove the necessity we show that, assuming k > m, the first row of A belongs to
the span of the next m − 1 rows, namely that
(cid:0)(cid:0)k−m−1
0
(cid:1),(cid:0)k−m−1
1
(cid:1), . . . ,(cid:0)k−m−1
k−m−1(cid:1), 0, . . . , 0(cid:1)A = 0 mod 2.
Then any x orthogonal to all the rows of A starting from the second one, must also be
orthogonal to the first row, and so the system Ax = (1, 0, . . . , 0)t has no solutions. To
establish (19) we need to show that for every j = 1, . . . , b − a + 1, we have
(19)
i=1 (cid:18)k − m − 1
i − 1 (cid:19)(cid:18)n − (a + j − 1) + 2(i − 1)
(a + j − 1) + (i − 1) − k (cid:19) = 0 mod 2.
Xk−m
which is equivalent (by substitution r = a + j − 1, l = i − 1, N = k − m − 1, n = 2p + m)
to showing that for all r = a, . . . , b,
l=0(cid:18)N
l (cid:19)(cid:18)2p − 1 − (r − k + N − 2l)
r − k + l
(cid:19) = 0 mod 2.
(20)
XN
We require the following Lemma.
Lemma 3. Suppose p ≥ 2 and let x, y ∈ Z.
(a) If 0 ≤ x < y < 2p, then (cid:0)2p+x
(cid:1) = (cid:0)y+x
(b) If x, y ≤ 2p − 2 and y, x + y > 0, then (cid:0)2p−1−x
Proof. By Kummer's Theorem, a binomial coefficient (cid:0)q
y (cid:1) = 0 mod 2.
y
y (cid:1) mod 2.
t(cid:1) with 0 ≤ t is odd if and only
if there is a place in the binary representation where q has 0 and t has 1 and, when
0 ≤ t ≤ q, if and only if there is a place in the binary representation where both q − t
and t have 1.
(a) For (cid:0)2p+x
y (cid:1) = 1 mod 2, the binary representation of 2p + x must have a 1 at all the
places where the binary representation of y does. But as y < 2p, this implies that the
binary representation of x has a 1 at all the places where the binary representation of y
does, which contradicts the fact that y > x.
binary representation where 2p − 1 − x has 0 and y has 1 if and only if there is a place
(cid:1) is even if and only if there is a place in the
(b) First suppose x ≥ 0. Then (cid:0)2p−1−x
in the binary representation where x has 1 and y has 1 if and only if (cid:0)y+x
y (cid:1) = 0. Denote z = −x − 1 ≥ 0. Then (cid:0)2p−1−x
Now let x < 0. So (cid:0)y+x
y (cid:1) and
0 ≤ z < y ≤ 2p − 2 by our assumption. By part (a), (cid:0)2p+z
y (cid:1) = (cid:0)z
y(cid:1) = 0
y (cid:1) = (cid:0)2p+z
as z < y. So (cid:0)y+x
y (cid:1) mod 2.
To apply Lemma 3(b) to the binomial coefficients (cid:0)2p−1−(r−k+N −2l)
(cid:1) from (20) we need
to check few inequalities. We have r−k ≥ a−k = (cid:6) 1
3(n −(cid:4) 1
3(n − k + 1)(cid:7) ≥ (cid:6) 1
2n(cid:5) + 1)(cid:7) =
3((cid:6) 1
(cid:6) 1
2n(cid:7) + 1)(cid:7) ≥ 1 and so r − k + l ≥ 1 and (r − k + l) + (r − k + N − 2l) ≥ 1.
Furthermore, r − k + l, r − k + N − 2l ≤ r − k + N ≤ b − k + N = ⌊ 1
2 n⌋ + N − 1, and
2n⌋ + N − 1 = ⌊ 1
⌊ 1
2m⌋ − m − 2 ≤ 2p − 2.
y (cid:1) is even.
(cid:1) = (cid:0)2p+z
y(cid:1) mod 2, and (cid:0)z
2n⌋ − m − 2 = 2⌊2p−1 + 1
2 n⌋ + k − m − 2 ≤ 2⌊ 1
r−k+l
(cid:3)
y
y
10
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
So the hypotheses of Lemma 3(b) are satisfied with x = r − k + N − 2l, y = r − k + l.
So Lemma 3(b) gives (cid:0)2p−1−(r−k+N −2l)
Vandermonde's identity gives (cid:0)2(r−k)+N −l
hand side of (20) is congruent modulo 2 to
r−k+l
i (cid:19)(cid:18) 2(r − k)
r−k+l
r−k+l
i=0 (cid:0)N −l
(cid:1) mod 2, for every l = 0, . . . , N.
i (cid:1)(cid:0) 2(r−k)
r−k+l−i(cid:1), and hence the left-
(cid:1) = (cid:0)2(r−k)+N −l
(cid:1) = PN −l
i, l, N − l − i(cid:19)(cid:18) 2(r − k)
(cid:18)
r − k + l − i(cid:19) = Xi,l≥0;i+l≤N
r − k + i − l(cid:19)(cid:19)
r − k + l − i(cid:19) +(cid:18) 2(r − k)
i, l, N − l − i(cid:19)(cid:18)(cid:18) 2(r − k)
(cid:18)
r − k (cid:19)
i, i, N − 2i(cid:19)(cid:18)2(r − k)
(cid:18)
r − k + l − i(cid:19)
N
N
N
N
Xl=0
(cid:18)N − l
l (cid:19) N −l
(cid:18)N
Xi=0
= Xi>l≥0;i+l≤N
+ Xi≥0;2i≤N
= 0 mod 2,
as (cid:0) 2(r−k)
r−k+l−i(cid:1) = (cid:0) 2(r−k)
r−k+i−l(cid:1) and (cid:0)2(r−k)
r−k (cid:1) = 2(cid:0)2(r−k)−1
r−k
(cid:1). This completes the proof of necessity.
To prove the sufficiency we explicitly produce, for any 2 ≤ k ≤ m, a vector x ∈ Zb−a+1
2
such that Ax = (1, 0, . . . , 0)t ∈ Zk−1
2
:
s=0(cid:18)
xj = Xp−1
m − k
n − (a + j − 1) − 2s(cid:19),
j = 1, . . . , b − a + 1.
(21)
By Lemma 2 we need to show that for all i = 1, . . . , k − 1,
b−a+1
Xj=1
(a + j − 1) + (i − 1) − k (cid:19)
(cid:18)(cid:18)n − (a + j − 1) + 2(i − 1)
p−1
Xs=0
m − k
n − (a + j − 1) − 2s(cid:19)(cid:19) mod 2 = δ1i. (22)
(cid:18)
We first show that the expression on the left-hand side of (22) can be rewritten as
p−1
Xs=0 Xj∈Z (cid:18)n − (a + j − 1) + 2(i − 1)
(a + j − 1) + (i − 1) − k (cid:19)(cid:18)
m − k
n − (a + j − 1) − 2s(cid:19) mod 2,
so that there is no contribution from the values j ≤ 0 and j ≥ b−a+1. The latter is easy:
for the first binomial coefficient to be nonzero we need to have n − (a + j − 1) + 2(i − 1) ≥
(a + j − 1) + (i − 1) − k which gives 2j ≤ n + k + i + 1 − 2a ≤ n + 2k − 2a, as
i ≤ k − 1, so j ≤ ⌊n/2⌋ + k − a = b − a + 1. To prove the former, we first look
at the second binomial coefficient, from which we get m − k ≥ n − (a + j − 1) − 2s, so
j ≥ n−a+1+k −m−2s ≥ n− 1
3(2p+1 +k −m−3·2s).
Now if s < p − 1 the expression on the right-hand side is positive, as m ≤ 2p, and we are
done. Suppose s = p − 1. Then we have j ≥ 1
3(2p−1 + k − m), which still implies j > 0
unless m = 2p−1 + k + l, l ≥ 0, in which case we have j ≥ − 1
3 +1+k −m−2s = 1
3(n+2k +1)− 2
3l. Then
a = (cid:24)2p + m + 2k + 1
3
(cid:25) = (cid:24)2p + 2p−1 + 3k + l + 1
3
(cid:25) = 2p−1 + k +(cid:24)l + 1
3 (cid:25)
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
11
and the first binomial coefficient has the form (cid:0)2p+x
y (cid:1), where
x = n − (a + j − 1) + 2(i − 1) − 2p = m − (a + j − 1) + 2(i − 1)
= 2p−1 + k + l − (a + j − 1) + 2(i − 1) = l + 1 − ⌈(l + 1)/3⌉ + 2(i − 1) − j,
y = (a + j − 1) + (i − 1) − k = 2p−1 + ⌈(l + 1)/3⌉ + j + i − 2.
Note that as i ≥ 1, we have x ≥ 0 if j ≤ 0. Also if j ≤ 0, then as i ≤ k − 1, we have
y ≤ 2p−1 + ⌈(l + 1)/3⌉+ k − 3 ≤ 2p−1 + l + k − 2 = m − 2 < 2p. Moreover, if j ≤ 0, then as
i ≤ k−1, we have y−x = (2p−1+⌈(l + 1)/3⌉+j+i−2)−(l+1−⌈(l + 1)/3⌉+2(i−1)−j) =
2p−1 + 2 ⌈(l + 1)/3⌉ − (l + 1) + 2j − i ≥ 2p−1 + 2 − (l + 1) − k = 2p − m + 1 > 0. So
the hypotheses of Lemma 3(a) are satisfied, and hence the binomial coefficient (cid:0)2p+x
even. So it remains to establish that
y (cid:1) is
p−1
Xs=0 Xj∈Z
(a + j − 1) + (i − 1) − k (cid:19)(cid:18)
(cid:18)n − (a + j − 1) + 2(i − 1)
m − k
n − (a + j − 1) − 2s(cid:19) mod 2 = δ1i,
(23)
for all i = 1, . . . , k − 1.
A clear advantage of (23) is that it "takes care of itself" -- we do not have to worry
about the limits. Changing the summation variable in (23) to h = n − (a + j − 1) − 2s
we obtain that (23) is equivalent to
p−1
Xs=0 Xh∈Z
(cid:18)
2s + 2(i − 1) + h
n − 2s + (i − 1) − k − h(cid:19)(cid:18)m − k
h (cid:19) mod 2 = δ1i.
(24)
Now for a polynomial P ∈ Z2[t] and l ∈ Z we denote {P }l the coefficient of tl in P .
Consider the polynomial Px,y(t) = (t2 + t)x(t2 + t + 1)y. We have
Px,y(t) = Xh∈Z
h(cid:19)(t2 + t)x+h = Xh,s∈Z
(cid:18)y
h(cid:19)(cid:18)x + h
(cid:18)y
s (cid:19)tx+h+s = Xl∈Z Xh∈Z
l − x − h(cid:19)(cid:18)y
(cid:18) x + h
h(cid:19)tl,
so the left-hand side of (24) equals
p−1
Xs=0
p−1
{P2s+2(i−1),m−k}n+3(i−1)−k = (cid:26)
= (cid:26)
= {(t2p + t)(t2 + t)2(i−1)(t2 + t + 1)m−k}n+3(i−1)−k
(t2 + t)2s+2(i−1)(t2 + t + 1)m−k(cid:27)n+3(i−1)−k
(t2 + t)2s(t2 + t)2(i−1)(t2 + t + 1)m−k(cid:27)n+3(i−1)−k
Xs=0
Xs=0
p−1
modulo 2 (since as (t2 +t)2s = t2s+1 +t2s in Z2[t] and so Pp−1
s=0(t2 +t)2s = t2p+1 +t mod 2).
Now, if in the expansion of the latter polynomial we take t from the first parentheses,
then the maximal degree of t in the resulting terms will be 1 + 4(i − 1) + 2(m − k) ≤
2m − 1 + 3(i − 1) − k < n + 3(i − 1) − k, as i ≤ k − 1 and n = 2p + m, m ≤ 2p. It follows
12
that
p−1
Xs=0
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
{P2s+2(i−1),m−k}n+3(i−1)−k = {t2p(t2 + t)2(i−1)(t2 + t + 1)m−k}n+3(i−1)−k
= {(t + 1)2(i−1)(t2 + t + 1)m−k}m+(i−1)−k
= Xl∈Z
= {(t + 1)2(i−1)}i−1{(t2 + t + 1)m−k}m−k mod 2,
{(t + 1)2(i−1)}i−1+l{(t2 + t + 1)m−k}m−k−l
where the last equality follows from the symmetry: for the polynomial f (t) = (t+1)2(i−1)
we have f (t) = t2(i−1)f (t−1), so {(t + 1)2(i−1)}i−1+l = {(t + 1)2(i−1)}i−1−l, and similarly
{(t2 + t + 1)m−k}m−k−l = {(t2 + t + 1)m−k}m−k+l.
l
m−k−l(cid:1) = Ps(cid:0)m−k
s (cid:1)(cid:0)m−k−s
s
l (cid:1)(t2 + t)l}m−k = {Pl,h(cid:0)m−k
l (cid:1)(cid:0) l
Now if i > 1 we obtain {(t + 1)2(i−1)}i−1 = (cid:0)2(i−1)
i−1 (cid:1) = 0 mod 2, as required. If i = 1
we get {(t2 + t + 1)m−k}m−k = {Pl(cid:0)m−k
h(cid:1)th+l}m−k =
Pl(cid:0)m−k
l (cid:1)(cid:0)
(cid:1), where s = m − k − l. The terms with s < 0
(cid:0)m−k
s (cid:1) to be nonzero, the binary expansion of m − k must have a 1 at the same place, so
the binary expansion of m − k − s will have zero at that place, thus (cid:0)m−k−s
(cid:1) = 0. Hence
vanish, and the term with s = 0 is 1. For s > 0, consider the first place, counting from
the right, where the binary expansion of s has a 1. Then by Kummer's Theorem, for
{(t2 + t + 1)m−k}m−k = 1 mod 2, as required. This concludes the proof of Proposition 2
and hence of Theorem 2(c).
(cid:3)
s
Note that one can extract from the above proof an explicit basis for the space of
three-cocycles of m0(n) (and hence for H 3(m0(n))). We have the following theorem.
Theorem 4. For n ≥ 4, n = 2p + m, 0 < m ≤ 2p and for 2 ≤ k ≤ m, define the
numbers a = ⌈(n + 2k + 1)/3⌉, b = ⌊n/2⌋ + k − 1. Let Bn be the set of elements of m0(n)
of the form
b
Xr=a
p−1
Xs=0
n − r − 2s(cid:19)F (en+2k−2r,r, er) =
(cid:18) m − k
b
Xr=a
p−1
Xs=0
n − r − 2s(cid:19)Xl≥0
(cid:18) m − k
Dl(en+2k−2r∧er)∧er+l+1,
for 2 ≤ k ≤ m, where D is the linear operator defined by (9) and the binomial coefficients
are taken modulo 2. Then classes of the elements of the set
{e1,i−1,i,
2 + ⌊n/2⌋ ≤ i ≤ n} ∪ [4≤t≤n
Bt.
is a basis for the cohomology space H 3(m0(n)), n ≥ 4, over the field Z2.
Proof. We start with the elements e1,i−1,i, 2 + ⌊n/2⌋ ≤ i ≤ n. They are linearly inde-
pendent cocycles and the space spanned by them has the correct dimension, which is the
codimension of the space of coboundaries in the space spanned by e1ij, 1 < i < j ≤ n,
by Proposition 1.
It suffices to show that neither of them is a coboundary. But if
it were so, then by homogeneity we would have had that e1,i−1,i is the coboundary
of a linear combination of the elements ekl, 2 ≤ k < l ≤ n, k + l = 2i, that
is, of the elements ei−k,i+k, k = 1, . . . , n − i (note that as i ≥ 2 + ⌊n/2⌋, we have
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
13
2i − n − 1 ≥ 2). But the coboundary of any such element is the sum of exactly two
monomials, e1,i−k−1,i+k + e1,i−k,i+k−1, so the coboundary of any linear combination of
them is a sum of an even number of monomials, hence cannot be equal to e1,i−1,i.
As to the element from the sets Bt, no linear combination of them is a coboundary (as
any coboundary is a multiple of e1). Moreover, from Proposition 2 (both the statement
and the proof) it follows that they form a basis for the kernel of D, where the form of
the elements given in the statement follows from Lemma 2 and Equation (21).
(cid:3)
Example 1. For n = 4, . . . , 12, the space of 3-cocycles of m0(n) is spanned by the three-
forms e1ij, 1 < i < j ≤ n, and the three-forms from the following table in the rows
labelled by the numbers less than or equal to n.
e234
e245 + e236
e345 + e246 + e237, e356 + e257 + e347
e256 + e247 + e238, e456 + e357 + e258 + e348, e467 + e278 + e368 + e458
4
5
6
7
8
9
10 e267 + e258 + e249 + e23(10)
11 e367 + e268 + e358 + e349 + e24(10) + e23(11),
e378 + e279 + e369 + e35(10) + e25(11) + e34(11)
e467 + e368 + e458 + e269 + e25(10) + e24(11) + e23(12),
e478 + e289 + e379 + e469 + e45(10) + e35(11) + e25(12) + e34(12),
e489 + e38(10) + e47(10) + e28(11) + e46(11) + e27(12) + e36(12) + e45(12)
12
4. Cohomology of m2
In this section, we compute the cohomology of the infinite-dimensional Lie algebra m2
given by (2):
m2 = Span(e1, e2, . . . ),
[e1, ei] = ei+1, i > 1,
[e2, ej] = ej+2, j > 2,
hence completing the proof of Theorem 1. First we state the following result for the
truncation m2(n).
Corollary 1. The first three Betti numbers of the Lie algebra m2(n), n ≥ 5, over Z2 are
given by b1(m2(n)) = 2, b2(m2(n)) = [ 1
2(n + 1)], and
b3(m2(n)) = 1
3(2p − 1)(2p−1 − 1) + 1
2m(m − 1) + [ 1
2 (n − 1)],
where n = 2p + m, 0 < m ≤ 2p.
Proof. By [14, Theorem 1], the Betti numbers of m2(n) and of m0(n) over Z2 are the
same. The claim then follows from Theorem 2.
(cid:3)
Remark 2. It is easy to see that H 1(m2(n)) is spanned by the cohomology classes of
e1 and e2 and that H 2(m2(n)) is spanned by the cohomology classes of the elements
e1n + e2,n−1, ei,i+1 + ei−1,i+3 + · · · + e2,2i−1, where 2 ≤ i ≤ 1
2 (n + 1). A basis for H 3(m2(n))
14
YURI NIKOLAYEVSKY AND IOANNIS TSARTSAFLIS
can be found by applying the map f from [14, Definition 3] (see below) to the elements
of the basis for H 3(m0(n)) constructed in Theorem 4; the resulting basis is the same.
In the infinite-dimensional case, we follow the construction of [14]. As in the Introduc-
tion, let V = Span(e1, e2, . . . ), and define the operator D1 on V ∗ by D1e1 = D1e2 = 0,
D1ei = ei−1, for i > 2, and then extend it to Λ(V ) as a derivation. Note that any
ω ∈ Λq(V ), q ≥ 2, has a unique presentation in the form ω = e1 ∧ ξ + e2 ∧ η + ζ, where
ξ ∈ Λq−1(e2, e3, . . . ), η ∈ Λq−1(e3, e4, . . . ) and ζ ∈ Λq(e3, e4, . . . ). Note that ξ, η and ζ
linearly depend on ω.
Define the linear map f on Λ(V ) by setting f (e1∧ξ+e2∧η+ζ) = e1∧ξ+e2∧(η+D1ξ)+ζ
on the forms of rank at least two, and taking it to be the identity on V ∗. The following
properties of f are easy to check:
• f is an involution, hence a bijection, and f −1 = f ,
• the restriction of f to Λ(e2, e3, . . . ) is the identity,
• f preserves the homogeneous components: f (Λq
k(V )) = Λq
k(V ).
The main feature of f is the fact that it interweaves the differentials of m0 and m2.
More precisely, consider m0 and m2 to have the same underlying linear space V , but to
be defined by the brackets (1) and (2) respectively relative to the same basis {e1, e2, . . . }
for V . Then for all ω ∈ Λ(V ), we have
f d0ω = d2f ω,
f d2ω = d0f ω,
(25)
where d0 and d2 are the differentials on m0 and m2 respectively. The first equation is
easily verified for ω = ei, and the proof for ω ∈ Λq(V ), q ≥ 2, is identical to the proof of
[14, Proposition 1]. The second one follows, as f is an involution.
Proof of Theorem 1. By (25), f bijectively maps cocycles and coboundaries of m0 to
cocycles and coboundaries of m2 respectively. It follows that H ∗(m2) is spanned by the
classes of the images under f of the elements (13). As f acts on all those elements as
the identity, we obtain that the basis for H ∗(m2) is the set of the classes of the same
cocycles.
The fact that the multiplicative structure is preserved follows from the fact that the
restriction of f to Λ(e2, e3, . . . ) is the identity and that multiplication by e1 is trivial in
both H ∗(m0) and H ∗(m2). Multiplication by e1 is trivial in H ∗(m0) because e1 ∧ ω is a
d0-coboundary, for any ω (see the proof of Theorem 3). To see that multiplication by
e1 is trivial in H ∗(m2), notice that for any ω in the list (13), one has Dω = 0 (which
is essentially assertion (a) of Lemma 1), and so f (e1 ∧ ω) = e1 ∧ ω, which is then a
d2-coboundary, as f maps coboundaries to coboundaries.
(cid:3)
Acknowledgements. We are very grateful to Grant Cairns for drawing our attention to
the question, for many useful discussions and for his help which improved the presentation
of this paper.
1. The On-Line Encyclopedia of Integer Sequences, published electronically at http://oeis.org/,
2015.
References
COHOMOLOGY OF N-GRADED LIE ALGEBRAS OF MAXIMAL CLASS OVER Z2
15
2. Grant F. Armstrong and Stefan Sigg, On the cohomology of a class of nilpotent Lie algebras, Bull.
Austral. Math. Soc. 54 (1996), no. 3, 517 -- 527.
3. M. Bordemann, Nondegenerate invariant bilinear forms on nonassociative algebras, Acta Math.
Univ. Comenian. (N.S.) 66 (1997), no. 2, 151 -- 201.
4. Grant Cairns and Sebastian Jambor, The cohomology of the Heisenberg Lie algebras over fields of
finite characteristic, Proc. Amer. Math. Soc. 136 (2008), no. 11, 3803 -- 3807.
5. A. Caranti, S. Mattarei, and M. F. Newman, Graded Lie algebras of maximal class, Trans. Amer.
Math. Soc. 349 (1997), no. 10, 4021 -- 4051.
6. A. Caranti and M. F. Newman, Graded Lie algebras of maximal class. II, J. Algebra 229 (2000),
no. 2, 750 -- 784.
7. Alice Fialowski, On the classification of graded Lie algebras with two generators, Moscow Univ.
Math. Bull. 38 (1983), no. 2, 76 -- 79.
8. Alice Fialowski and Dmitri Millionschikov, Cohomology of graded Lie algebras of maximal class, J.
Algebra 296 (2006), no. 1, 157 -- 176.
9. L. V. Goncarova, Cohomology of Lie algebras of formal vector fields on the line, Functional Anal.
Appl. 7 (1973), no. 2, 91 -- 97.
10. G. Jurman, Graded Lie algebras of maximal class. III, J. Algebra 284 (2005), no. 2, 435 -- 461.
11. Dmitri V. Millionschikov, Graded filiform Lie algebras and symplectic nilmanifolds, Geometry, topol-
ogy, and mathematical physics, Amer. Math. Soc. Transl. Ser. 2, vol. 212, Amer. Math. Soc., Prov-
idence, RI, 2004, pp. 259 -- 279.
12. Aner Shalev and Efim I. Zelmanov, Narrow Lie algebras: a coclass theory and a characterization of
the Witt algebra, J. Algebra 189 (1997), no. 2, 294 -- 331.
13. Emil Skoldberg, The homology of Heisenberg Lie algebras over fields of characteristic two, Math.
Proc. R. Ir. Acad. 105A (2005), no. 2, 47 -- 49.
14. Ioannis Tsartsaflis, On the Betti numbers of filiform Lie algebras over fields of characteristic two,
http://arxiv.org/1511.03132, 2015, [math.RA].
15. Mich`ele Vergne, Cohomologie des alg`ebres de Lie nilpotentes. Application `a l'´etude de la vari´et´e des
alg`ebres de Lie nilpotentes, Bull. Soc. Math. France 98 (1970), 81 -- 116.
Department of Mathematics and Statistics, La Trobe University, Melbourne, Aus-
tralia 3086
E-mail address: [email protected]
E-mail address: [email protected]
|
1810.12612 | 1 | 1810 | 2018-10-30T09:52:32 | On the Morita Reduced Versions of Skew Group Algebras of Path Algebras | [
"math.RA",
"math.CT",
"math.RT"
] | Let R be the skew group algebra of a finite group acting on the path algebra of a quiver. This article develops both theoretical and practical methods to do computations in the Morita reduced algebra associated to R. Reiten and Riedtmann proved that there exists an idempotent e of R such that the algebra eRe is both Morita equivalent to R and isomorphic to the path algebra of some quiver which was described by Demonet. This article gives explicit formulas for the decomposition of any element of eRe as a linear combination of paths in the quiver described by Demonet. This is done by expressing appropriate compositions and pairings in a suitable monoidal category which takes into account the representation theory of the finite group. | math.RA | math |
ON THE MORITA REDUCED VERSIONS OF SKEW GROUP
ALGEBRAS OF PATH ALGEBRAS
PATRICK LE MEUR
Abstract. Let R be the skew group algebra of a finite group acting on the
path algebra of a quiver. This article develops both theoretical and practical
methods to do computations in the Morita reduced algebra associated to R.
Reiten and Riedtmann proved that there exists an idempotent e of R such
that the algebra eRe is both Morita equivalent to R and isomorphic to the
path algebra of some quiver which was described by Demonet. This article
gives explicit formulas for the decomposition of any element of eRe as a linear
combination of paths in the quiver described by Demonet. This is done by ex-
pressing appropriate compositions and pairings in a suitable monoidal category
which takes into account the representation theory of the finite group.
Introduction
In [7], Reiten and Riedtmann initiated the investigation of the skew group alge-
bras of Artin algebras from the viewpoint of homological dimensions and represen-
tation theory. Their article was followed-up by many research works illustrating the
following principle: Artin algebras share many properties, whether of homological
or of representation theoretic nature, with their associated skew group algebras.
Let k be an algebraically closed field, Q be a finite quiver which may have ori-
ented cycles, and G be a finite group with order not divisible by char(k) and acting
on kQ by algebra automorphisms in such a way that both the set of (primitive idem-
potents of) vertices and the vector subspace generated by the arrows are stabilised
by this action. Following [7], the skew group algebra kQ ∗ G is hereditary.
While Riedtmann and Reiten described a quiver whose path algebra is Morita
equivalent to kQ ∗ G when G is cyclic, the description of such a quiver in general
is recent. In [2], Demonet described an idempotent e of kQ ∗ G and a quiver QG
such that e · (kQ ∗ G) · e is Morita equivalent to kQ ∗ G and isomorphic to kQG.
This, however, does not yield a completely described isomorphism, which might be
source of trouble. Here is a situation taken from Calabi-Yau algebras and where
such trouble may occur. Following [6], if W is a G-invariant potential on Q, then G
acts on the Ginzburg dg algebra A(Q, W ), defined in [4], and A(Q, W )∗G is Morita
equivalent to A(QG, WG), where WG is the potential on QG which, as a linear
combination of oriented cycles, corresponds to e·W ·e under any chosen isomorphism
kQG → e · (kQ ∗ G) · e. The problem here is that describing that linear combination
explicitly is not easy. Up to now, the only fairly general explicit description of WG
is due to Giovannini and Pasquali ([5]) when G is cyclic, the stabiliser of each vertex
is either trivial or the whole group, and every cycle appearing in W goes through
fixed vertices only (or, through vertices with trivial stabilisers only, respectively)
as soon as it goes through two of them.
This raises the question of decomposing explicitly the elements of e · (kQ ∗ G) · e
as linear combinations of paths in QG under an isomorphism kQG → e · (kQ ∗ G) · e.
Date: October 31, 2018.
2010 Mathematics Subject Classification. Primary 16S35; Secondary 16W22, 20C15, 18D10.
Key words and phrases. Path algebra; skew group algebra; monoidal category; intertwiner.
1
2
PATRICK LE MEUR
Demonet's description of QG involves the representation theory of the stabilisers
of the vertices of Q. It is hence expectable that answering the above mentioned
question should involve representation theory. Answers of this kind already exist in
particular cases where Q has only one vertex. Here is an example due to Ginzburg
([4]). He proved that, if G is a finite subgroup of SL3(C), then C[x, y, y] ∗ G is
Calabi-Yau in dimension 3. This was done by taking Q to be the quiver with one
vertex and three loops x, y, and z, by taking W = xyz −xzy, and by expressing WG
in terms of the monoidal category of finite dimensional representations of G. See
[1] for a generalisation to C[x1, . . . , xn] ∗ G where G is a finite subgroup of SLn(C),
in which case Q is the quiver with one vertex and n loops.
This article hence gives explicit formulas for the decomposition of the elements
of e · (kQ ∗ G) · e as linear combinations of paths in QG under the isomorphism
kQG → e · (kQ ∗ G) · e. These formulas are obtained by translating explicitly the
result of natural operations and pairings in the monoidal category (mod(Ae), ⊗A) of
finite dimensional A-bimodules, where A is the direct product of the group algebras
of suitably chosen stabilisers of the vertices of Q.
On one hand, these formulas provide an interpretation of the coefficients of these
linear combinations in terms of the representation theory of the stabilisers of the
vertices of Q. On the other hand, they are explicit enough to be implemented by
a computer or a human being able to manipulate the irreducible representations of
the involved groups. In specific cases, for instance when the action of the group
elements transform arrows to scalar multiples of arrows and the stabilisers are
cyclic, these formulas simplify in a combinatorial way. Finally, these formulas give
a complete solution to the problem of computing a potential WG on QG such that
the Ginzburg dg algebra A(QG, WG) is Morita equivalent to A(Q, W ) ∗ G, for any
given G-invariant potential W on Q.
1. Definitions and main results
Throughout the article, k denotes an algebraically closed field. For all finite di-
mensional algebras Λ, their categories of finite dimensional left modules are denoted
by mod(Λ).
This section presents the main results of this text as well as the needed definitions
and notation. These results are illustrated in section 2. The remaining sections are
devoted to the proofs of these results.
Setting 1.1. Let Q0, S, G, and M be as follows.
• Q0 is a finite set and S is the semi-simple k-algebra kQ0.
• G is a finite group with order not divisible by char(k) and acting on Q0.
• M is a finite dimensional S-bimodule, considered as a collection of vector
, endowed with an action of G written exponentially
spaces ( iMj)(i,j)∈Q2
and such that g( iMj) = g·iMg·j for all i, j ∈ Q0 and g ∈ G.
0
The choice of a basis of iMj, for all i, j ∈ Q0 determines a quiver Q such that
TS(M ) ≃ kQ as k-algebras. All finite quivers arise in this way.
The actions of G on S and M induce an action on TS(M ) by algebra auto-
morphisms such that g(m1 ⊗ · · · ⊗ mn) = gm1 ⊗ · · · ⊗ gmn for all g ∈ G and
m1, . . . , mn ∈ M . Recall that the skew group algebra TS(M ) ∗ G is the k-algebra
with underlying vector space TS(M )∗G, where the tensor u⊗v is denoted by u∗v for
all u ∈ TS(M ) and v ∈ kG, and with product such that (u∗g)·(u′∗g′) = (u· gu′)∗gg′
for all u, u′ ∈ TS(M ) and g, g′ ∈ G. It is customary to consider TS(M ) and kG as
subalgebras of TS(M ) ∗ G in the canonical way.
Notation 1.2. The following data is used throughout the article.
• [G\Q0] denotes a complete set of representatives of the orbits in Q0.
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS
3
• For all i ∈ Q0,
-- Gi denotes the stabiliser of i and ei denotes the corresponding primitive
idempotent of S,
-- [G/Gi] denotes a complete set of representatives of cosets modulo Gi,
-- irr(Gi) denotes a complete set of representatives of the isomorphism
classes of the irreducible representations of Gi,
-- and εU denotes a primitive idempotent of kGi such that U ≃ kGi · εU
in mod(kGi), for all U ∈ irr(Gi).
• e denotes the idempotent P(i,U) ei ∗ εU of TS(M ) ∗ G, where i runs through
[G\Q0] and U runs through irr(Gi).
Demonet proved in [2] that the skew group algebra TS(M )∗G is Morita equivalent
to the path algebra of a quiver as follows.
Setting 1.3. Let QG be a quiver with the following properties.
• The vertices are the couples (i, U ), where i ∈ [G\Q0] and U ∈ irr(Gi).
• For all vertices (i, U ) and (j, V ), the arrows (i, U ) → (j, V ) form a basis
over k of HomkGi(U, M (i, j; V )), here M (i, j; V ) = ⊕y∈[G/Gj] iMy·j ⊗k yV ,
where yV stands for y ⊗k V ⊂ kG ⊗k
kGi, see Definition 1.6 for details.
Demonet's result may be reformulated as follows, see Section 5 for more details.
Theorem 1.4 ([2]). The k-algebras kQG and TS(M ) ∗ G are Morita equivalent.
Besides, assuming that U = kGi · εU for all vertices (i, U ) of QG, then there exists
an isomorphism of algebras
(1.0.1)
kQG
∼−→ e · (TS(M ) ∗ G) · e
which maps every arrow f : (i, U ) → (j, V ) of QG to f (εU ).
The present article introduces a k-algebra, called the algebra of intertwiners
relative to M and denoted by Intw, and an isomorphism of k-algebras kQG →
Intwop together with explicit formulas decomposing any given element of Intw in
the basis consisting of the images of the paths in QG under kQG → Intwop. See
Theorem 1.15 for details. Composing this isomorphism with the inverse of (1.0.1)
results in a k-algebra isomorphism e · (TS(M ) ∗ G) · e → Intwop. A description of
it is given in Theorem 1.16 hence providing a decomposition, under (1.0.1), of any
element of e · (TS(M ) ∗ G) · e in the basis of kQG consisting of the paths in QG.
These decompositions involve non degenerate pairings between intertwiners relative
to M and intertwiners relative to the dual vector space M ∗. Both the product in
Intw and the pairings are explicit reformulations of natural operations and pairings
in a suitable monoidal category.
Remark 1.5 ([3]). Let (C, ⊗) be a monoidal k-linear category with finite dimen-
sional morphism spaces.
(1) Given objects X, Y , U , V , and W of C, there is a natural operation between
morphisms U → X ⊗ V and V → Y ⊗ W ,
C(U, X ⊗ V ) × C(V, Y ⊗ W ) −→ C(U, X ⊗ Y ⊗ W )
(f, f ′)
7−→ (X ⊗ f ′) ◦ f .
(2) Given an object X of C, recall that a left dual of X is an object X ′ of C
together with an adjunction (X ⊗ −) ⊢ (X ′ ⊗ −) of endofunctors of C. If X
has a left dual, then there exists a non degenerate pairing, for all objects
U and V of C such that C(U, U ) = k · IdU ,
h−−i : C(U, X ⊗ V ) ⊗ C(V, X ′ ⊗ U ) −→ k
f ⊗ φ
7−→ hf φi
such that hf φi · IdU = φ ◦ f , where φ ∈ C(X ⊗ V, U ) is adjoint to φ.
4
PATRICK LE MEUR
The monoidal category C considered here is the category mod(Ae) of finite di-
mensional A-bimodules with tensor product being ⊗A, where A is as follows,
(1.0.2)
A = k × Πi∈Q0 kGi .
Note that A is a semi-simple k-algebra; accordingly, any object X of C has a left
dual given by the dual vector space X ∗. The isolated factor k in the definition of A
is included so that C contains mod(kGi) as a full additive subcategory, yet not as
a monoidal subcategory, for all i ∈ Q0. Note that a smaller algebra might be used
instead of A; however, the reason for introducing C is to prove the needed properties
of the product in Intw and of the pairing; besides (1.0.2) has the advantage of being
easy to describe.
Here are the base ingredients needed to define Intw.
Definition 1.6. Let n be a non negative integer. Let i = i0, . . . , in be a sequence
in Q0. Let V ∈ mod(kGin ).
(1) For all sequences y = y1, . . . , yn, where yt ∈ [G/Git ] for all t ∈ {1, . . . , n},
let My(i; V ) be the following vector space,
My(i; V ) = i0 My1·i1 ⊗k y1·i1 My1y2·i2 ⊗k · · ·⊗k y1···yn−1·in−1 My1···yn·in ⊗k y1 · · · ynV ,
where y1 · · · ynV stands for the vector subspace y1 · · · yn ⊗k V of kG ⊗k V .
(2) Define the kGi0 -module M (i; V ) as follows. Its underlying vector space is
(1.0.3)
My
My(i; V ) ,
where y runs through all possible sequences such as in (1). For all g ∈ Gi0 ,
the action of g is written exponentially and defined as follows. For all y,
there exist a unique sequence y′ = y′
t ∈ [G/Git ] for all
t ∈ {1, . . . , n}, and a unique h ∈ Gin , such that
n, where y′
1, . . . , y′
• gy1 · · · ytGit = y′
• and gy1 · · · yn = y′
1 · · · y′
tGit for all t ∈ {1, . . . , n},
1 · · · y′
nh;
then, the action of g transforms My(i; V ) into My′ (i; V ) and, for all m1 ⊗
· · · ⊗ mn ⊗ y1 · · · ynv ∈ My(i; V ),
g (m1 ⊗ · · · ⊗ mn ⊗ y1 · · · ynv) = gm1 ⊗ · · · ⊗ gmn ⊗ y′
1 · · · y′
n
hv .
This does define a finite dimensional kGi0 -module. If n = 0, then M (i; V ) = V .
In the monoidal category (mod(Ae), ⊗A) considered previously, M (i; V ) is isomor-
phic to M (i) ⊗A V where M (i) is a kGi0 − kGin -bimodule defined by M and i, see
(4.1.1) for the definition of M (i) and (4.1.7) for the isomorphism.
Note that, in the previous definition, if U = kGi0 · εU and V = kGin · εV
for primitive idempotents εU ∈ kGi0 and εV ∈ kGin , then M (i; U ) is a kGi0 -
submodule of TS(M ) ∗ G.
The main results of this article are based on a description of the quiver kQG in
terms of intertwiners.
Definition 1.7. An intertwiner relative to M is a morphism of kGi0 -modules U →
M (i; V ), for some sequence i = i0, . . . , in in Q0 and some modules U ∈ mod(kGi0 )
and V ∈ mod(kGin ).
Note that, in the previous definition, if U = kGi0 · εU and V = kGin · εV
for primitive idempotents εU ∈ kGi0 and εV ∈ kGin , then all intertwiners U →
M (i; V ) take their values in TS(M ) ∗ G.
The main purpose of this article is to provide explicit formulas in the main
theorems. These formulas use the following notation.
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS
5
Notation 1.8. Let i = i0, . . . , in be a sequence in Q0 where n > 1. Let U ∈
mod(kGi0 ) and V ∈ mod(kGin ). Let f ∈ HomkGi0
(U, M (i; V )). For all u ∈ U ,
write the decomposition of f (u) along (1.0.3) as follows
f (u) = Xy
fy(u) ;
and, for all sequences y = y1, . . . , yn, where yt ∈ [G/Git ] for all t ∈ {1, . . . , n}, write
symbolically the element fy(u) of My(i; V ) as follows, with the sum sign omitted,
fy(u) = f (1)
y (u) ⊗ · · · ⊗ f (n)
y
(u) ⊗ y1 · · · ynf (0)
y (u) ;
hence, f (t)
f (0)
y (u) is meant to lie in V ; as a whole,
y (u) is meant to lie in y1···yt−1·it−1 My1···yt·it , for all t ∈ {1, . . . , n}, and
(1.0.4)
f (u) = Xy
f (1)
y (u) ⊗ · · · ⊗ f (n)
y
(u) ⊗ y1 · · · ynf (0)
y (u) .
Now here is the operation to be used as the multiplication in Intw.
Definition 1.9. Let i′′ = i0, . . . , im+n be a sequence in Q0, where m, n > 1.
Denote the sequences i0, . . . , im and im, im+1, . . . , im+n by i and i′, respectively.
Let U ∈ mod(kGi0 ), V ∈ mod(kGim ), and W ∈ mod(kGim+n ). For all f ∈
(U, M (i; V )) and f ′ ∈ HomkGim (V, M (i′; W )), define a k-linear mapping
HomkGi0
by
(1.0.5)
where
f ′ ⊛ f : U → M (i′′; W )
f ′ ⊛ f = Xy′′
(f ′ ⊛ f )y′′ ,
• y′′ runs through all sequences y1, . . . , ym+n with yt ∈ [G/Git ] for all t
• and (f ′ ⊛ f )y′′ is the mapping U → My′′ (i′′; W ) such that, for all u ∈ U ,
(f ′ ⊛ f )y′′ (u) = f (1)
(u)⊗
y
y (u) ⊗ · · · ⊗ f (m)
y1···ym(f ′(1)
y′ (f (0)
y1 · · · ym+nf ′(0)
y (u))) ⊗ · · · ⊗ y1···ym(f ′(n)
y′ (f (0)
y (u)) ,
y′
(f (0)
y (u)))⊗
where y = y1, . . . , ym and y′ = ym+1, . . . , ym+n.
Note that, in the previous definition, if U = kGi0 · εU , V = kGim · εV , and W =
kGim+n · εW for primitive idempotents εU ∈ kGi0 , εV ∈ kGim , and εW ∈ kGim+n ,
then
(1.0.6)
(f ′ ⊛ f )(εU ) = f (εU ) · f ′(εv) ,
where the product on the right-hand side is taken in TS(M ) ∗ G.
The definition of ⊛ is technical. Actually, it is an explicit reformulation of a
simple operation in the monoidal category (mod(Ae), ⊗A). More precisely, using
the comment which follows Definition 1.6,
• first, M (i′′) ≃ M (i′) ⊗A M (i′), see Lemma 4.2;
• next, if M (i; V ), M (i′; W ), and M (i′′; W ) are identified with M (i) ⊗A V ,
M (i′) ⊗A W , and M (i′′) ⊗A W , respectively, then ⊛ is an explicit reformu-
lation of the operation introduced in part (1) of Remark 1.5, see Lemma 4.4
for a precise statement and proof.
Accordingly, ⊛ does yield intertwiners, see Lemma 4.4, and it is associative, see
Proposition 4.5.
6
PATRICK LE MEUR
Definition 1.10. The algebra of intertwiners relative to M is denoted by Intw
and defined by
(1.0.7)
Intw =
Mn,i,U,V
HomkGi0
(U, M (i; V )), ⊛
,
where n runs through all non negative integers, i = i0, . . . , in runs through all
sequences in [G\Q0], and U and V run through irr(Gi0 ) and irr(Gin ), respectively.
It is hence possible to associate an element of Intw to every path in QG.
Notation 1.11. For all paths in QG
γ : (i0, U0)
f1−→ (i1, U1) → · · · → (in−1, Un−1)
fn−→ (in, Un) ,
denote by fγ the following element of Intw lying in HomkGi0
(1.0.8)
fγ = fn ⊛ fn−1 ⊛ · · · ⊛ f1 .
(U0, M (i0, . . . , in; Un)),
As stated below, see Theorem 1.15, assigning fγ to γ for every path γ in QG
yields an isomorphism of algebras from kQG to Intwop. The purpose of this article
is to explain how to decompose any element of Intw as a linear combination of the
fγ's. This involves non degenerate pairings between certain spaces of intertwiners
relative to M and to M ∗, respectively. The dual vector space M ∗ has a structure
j = ( j Mi)∗ for all i, j ∈ Q0, and G acts on M ∗ in such
of S-bimodule such that iM ∗
a way that, for all ϕ ∈ M ∗ and g ∈ G,
gϕ = ϕ( g−1
•) .
This action is such that g( iM ∗
previous considerations may be applied to M ∗ instead of to M .
j ) = g·iM ∗
g·j for all i, j ∈ Q0 and g ∈ G. Hence, the
Notation 1.12. Let i = i0, . . . , in be a sequence in Q0, where n > 0. Let U ∈
mod(kGi0 ) and V ∈ mod(kGin ).
(1) Denote by io the sequence in, in−1, . . . , i0.
(2) Let ϕ ∈ HomkGin (V, M ∗(io, U )). Proceeding similarly as done in Nota-
tion 1.8,
• denote by ϕx the composition of ϕ with the canonical projection
x (io; U ), for all sequences x = xn−1, xn−2, . . . , x0
M ∗(io; U ) → M ∗
where xt ∈ [G/Git ] for all t ∈ {0, . . . , n − 1},
• and for all such x and all v ∈ V , write symbolically the element ϕx(v)
of M ∗
x (io; U ) as follows,
ϕx(v) = ϕx
(n)(v) ⊗ · · · ⊗ ϕx
(1)(v) ⊗ xn−1 · · · x0ϕx
(0)(v) .
Hence, ϕx
for all t ∈ {1, . . . , n} and ϕx
(t)(v) is meant to lie in (xn−1xn−2···xt)·itM ∗
(0)(v) is meant to lie in U . As a whole,
(xn−1xn−2···xt−1)·it−1
(1.0.9)
ϕ(v) = Xx
(n)(v) ⊗ · · · ⊗ ϕx
ϕx
(1)(v) ⊗ xn−1 · · · x1ϕx
(0)(v) .
The above mentioned pairings are described explicitly as follows.
Proposition 1.13. Let i = i0, . . . , in be a sequence in Q0, where n > 0. Let
U ∈ mod(kGi0 ) and V ∈ mod(kGin ). Assume that U is simple. There exists a
non degenerate pairing
(−−) : HomkGi0
(U, M (i; V )) ⊗k HomkGin (V, M ∗(io; U )) → k
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS
7
(U, M (i; V )) and for all ϕ ∈
with the following property. For all f ∈ HomkGi0
HomkGin (V, M ∗(io; U )), the scalar (f ϕ) is such that, for all u ∈ U , the element
(f ϕ) u of U is equal to
n
ϕx
(t)(f (0)
y (u))((y1···yn)−1
y (u)) h0(ϕx
f (t)
(0)(f (0)
y (u))) ,
(1.0.10)
Xy
Yt=1
where y = y1, . . . , yn runs through all sequences with yt ∈ [G/Git ] for all t and
x = xn−1, . . . , x0 and h0 are uniquely determined by the following conditions.
(a) xt ∈ [G/Git ] for all t, and h0 ∈ Gi0 .
(b) (yt · · · yn)−1 ∈ xn−1xn−2 · · · xt−1Git−1 for all t ∈ {0, . . . , n − 1}.
(c) (y1 · · · yn)−1 = xn−1xn−2 · · · x0h0.
In view of computations, here is a useful explanation regarding x and h0 in the
above proposition. For all sequences y = y1, . . . , yn in G such that yt ∈ [G/Git ] for
all t ∈ {1, . . . , n} and for all non zero tensors
m1 ⊗ · · · ⊗ mn ∈ i0 My1·i1 ⊗k · · · ⊗k (y1···yn−1)·in−1 M(y1···yn)·in ,
there exists a unique sequence x = xn−1, xn−2, . . . , x0 in G such that xt ∈ [G/Git ]
for all t ∈ {0, . . . , n − 1} and
(y1···yn)−1
(m1 ⊗ · · · ⊗ mn) ∈ (xn−1xn−2···x0)·i0M(xn−1···x1)·i1 ⊗k · · · ⊗k xn−1·in−1Min ;
besides, (xn−1xn−2 · · · x0)−1(y1 · · · yn)−1 ∈ Gi0 . This sequence x is the one in the
statement of Proposition 1.13 in which h0 = (xn−1xn−2 · · · x0)−1(y1 · · · yn)−1.
Like the operation ⊛, the pairing (−−) has a simple interpretation in the
monoidal category (mod(Ae), ⊗A). Indeed,
• first, M (i) admits M ∗(io) as a left dual, see (4.1.5);
• next, if M (i; V ) and M ∗(io; U ) are identified with M (i)⊗AV and M ∗(io)⊗A
U , respectively, then (−−) is an explicit reformulation of h−−i as intro-
duced in part (2) of Remark 1.5, see (4.3.8).
Using (−−) yields intertwiners relative to M ∗ associated with the fγ's for paths
γ in QG.
Notation 1.14. Let QG be such as in setting 1.3.
(1) Let f ∈ (i, U ) → (j, V ) be an arrow of QG. Using the basis of the vector
space HomkGi(U, M (i, j; V )) consisting of the arrows (i, U ) → (j, V ) of
QG and using the associated (−−)-dual basis of HomkGj (V, M ∗(j, i; U )),
denote by f ∨ the dual element associated with f . In other words, f ∨ lies
in HomkGj (V, M ∗(j, i; U )) and, for all arrows f ′ : (i, U ) → (j, V ) of QG,
(f ′f ∨) = (cid:26) 1 if f ′ = f
0 otherwise.
(2) For all paths γ in QG
γ : (i0, U0)
f1−→ (i1, U1) → · · · → (in−1, Un−1)
fn−→ (in, Un) ,
denote by ϕγ the following intertwiner
ϕγ = f ∨
2 ⊛ · · · ⊛ f ∨
1 ⊛ f ∨
n ∈ HomkGin (Un, M ∗(in, in−1, . . . , i0; U0)) .
Now, here is the result comparing kQG and Intw.
Theorem 1.15. Keep settings 1.1 and 1.3.
(1) Assigning fγ to every path γ in QG yields an isomorphism of algebras
(1.0.11)
kQG
∼−→ Intwop .
8
PATRICK LE MEUR
(2) For all f ∈ Intw, say lying in HomkGi0
(U0, M (i0, . . . , in; Un)),
(1.0.12)
f = Xγ
(f ϕγ) fγ ,
where γ runs through all paths of QG of the shape (i0, U0) → (i1, •) →
· · · → (in−1, •) → (in, Un).
It is now possible to explain how to decompose any element of e · (TS(M ) ∗ G) · e
−−−−−→ e(TS(M ) ∗ G) · e. This
along the paths in QG via the isomorphism kQG
is done with the following construction and assuming that U = kGi · εU for all
vertices (i, U ) of QG. Given a tensor m1 ⊗ · · · ⊗ mn ∗ g ∈ ei0 · (M ⊗Sn ∗ G) · ein ,
where n > 0 and i0, in ∈ [G\Q0], then
(1.0.1)
m1 ⊗ · · · ⊗ mn ∗ g ∈ (cid:0) i0 My1·i1 ⊗k · · · ⊗k (y1···yn−1)·in−1M(y1···yn)·in(cid:1) ∗ y1 · · · ynkGin
for unique i1, . . . , in−1 ∈ [G\Q0], yt ∈ [G/Git ] for all t ∈ {1, . . . , n}, and h ∈ Gin ;
this defines an intertwiner, for all U ∈ irr(Gi0 ) and V ∈ irr(Gin ),
kGi0 · εU −→
M (i0, . . . , in; kGin · εV )
u · εU
7−→ (ei0 ∗ εU ) · (m1 ⊗ · · · ⊗ mn ∗ g) · (ein ∗ εV ) ,
where the products on the right-hand side are meant in TS(M ) ∗ G; summing these
intertwiners over all U ∈ irr(Gi0 ) and V ∈ irr(Gin ) defines an element of Intw.
This construction may be extended linearly to a linear mapping
e · (TS(M ) ∗ G) · e −→ Intwop .
Denote by Ξ the restriction of this mapping to e · (TS(M ) ∗ G) · e,
(1.0.13)
Ξ : e · (TS(M ) ∗ G) · e −→ Intwop .
Theorem 1.16. Let Q0, S, G, and S as in Setting 1.1. Let QG be as in Setting 1.3.
Assume that U = kGi · εU for all vertices (i, U ) of QG. Then, Ξ, as introduced
in (1.0.13), is an isomorphism of k-algebras such that the following diagram is
commutative,
(1.0.14)
kQG
(1.0.1) 3
3❣❣❣❣❣❣❣❣❣❣❣
+❳❳❳❳❳❳❳❳❳❳❳❳❳❳
(1.0.11)
e · (TS(M ) ∗ G) · e
Ξ
Intwop .
In particular, if (1.0.1) is used as an identification, then, for all θ ∈ e·(TS(M )∗G)·e,
(1.0.15)
θ = Xγ path in QG
(Ξ(θ)ϕγ ) · γ .
2. Practical aspects
Let Q0, S, G, and M be as in setting 1.1.
Applying Theorem 1.16 for computations requires to compute QG first. Actually,
rather than computing explicitly the intertwiners relative to M which form the
arrows of QG, it is simpler to
(1) compute a basis of HomGj (τ, M ∗(j, i; ρ)) for all pairs (i, ρ) and (j, τ ) such
that i, j ∈ [G\Q0], ρ ∈ irr(Gi), and τ ∈ irr(Gj ).
Because of the pairing (−−), this yields a quiver QG such as in setting 1.3 as well
as the intertwiners ϕα for all arrows α of QG. This avoids computing explicitly the
intertwiners corresponding to the arrows of QG, which are not used in the formulas
of Theorem 1.16.
+
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS
9
Given an element θ of e · (TS(M ) ∗ G) · e, say homogeneous of degree n for
the grading induced by the tensor powers of M , the decomposition into a lin-
ear combination of paths in QG of the inverse image of θ under the isomorphism
−−−−−→ e · (TS(M ) ∗ G) · e may be computed as follows using Theorem 1.16,
kQG
(2) compute Ξ(θ), which is a sum of intertwiners relative to M , see (1.0.13);
(3) compute the intertwiners ϕγ for all paths γ of length n in QG, see nota-
(1.0.1)
tion 1.14;
(4) compute the pairing (Ξ(θ)ϕγ ) for all such paths γ, see Proposition 1.13.
The desired linear combination of paths is hence Pγ(Ξ(θ)ϕγ ) · γ (see (1.0.15)).
These computations may be performed by a computer which is able to determine
the irreducible representations of irr(Gi) for all i ∈ [G\Q0]. This section illustrates
these computations.
2.1. How to pair intertwiners? In view of (1.0.6), the main difficulty in the
computations listed in the introduction of section 2 lies in computing pairings for
(−−). This section explains how to compute (1.0.10) in combinatorial terms in
the following specific setting.
Setting 2.1. Let Q0, S, G, and M be as in setting 1.1. Let Q be a quiver with
vertex set being Q0 and such that M is equal to the vector space with basis elements
being the arrows of Q. Let QG be as in setting 1.3. Assume that
• the stabiliser Gi is abelian for all i ∈ Q0,
• U = kGi · εU for all vertices (i, U ) of QG, denote by χU the character
Gi → k
× of U ,
• and, for all g ∈ G and all paths γ in Q, there exists a scalar χg,γ ∈ k
× and
a path g(γ) such that gγ = χg,γ g(γ).
If G is abelian, then it is possible to find a quiver Q and to choose irreducible
representations U of Gi, for all i ∈ [G\Q0], fitting in this setting.
Notation 2.2. The following notation relative to M is useful.
• For all arrows a : i → j of Q, denote by a∗ the element of M ∗ such that
a∗(a) = 1 and a∗(b) = 0 for all arrows b of Q distinct from a.
• Denote by Q∗ the quiver with vertex set being Q0 and whose arrows are
the linear forms a∗ introduced just before. This quiver is isomorphic to the
opposite quiver of Q.
• For all paths γ : t0
a1−→ t1 → · · · → tn−1
an−−→ tn in Q, denote by γ∗ the path
a∗
n−−→ tn−1 → · · · → t1
a∗
1−→ t0 in Q∗.
tn
Note that every path in Q starting in some vertex i0 lying in [G\Q0] is of the
shape i0 → y1 · i1 → · · · → (y1 · · · yn) · in for a unique sequence i1, . . . , in in [G\Q0]
and a unique sequence y1, . . . , yn in G such that yt ∈ [G/Git ] for all t ∈ {1, . . . , n}.
Let i = i0, . . . , in be a sequence in [G\Q0], consider U ∈ irr(Gi0 ) and V ∈
irr(Gin ), and let f : U → M (i; V ) and ϕ : V → M ∗(io; U ) be intertwiners. Since
the representation U is one dimensional, then,
(2.1.1)
f (εu) = Xγ
αγγ ⊗ y1 · · · ynεU ,
where γ runs through all paths in Q with source i0 and of length n, and where, for
all such paths γ, then αγ ∈ k and y = y1, . . . , yn denotes the sequence in G such
that
• yt ∈ [G/Git ] for all t ∈ {1, . . . , n}
• and γ is a path of the shape i0 → y1 · i1 → y1y2 · i2 → · · · → y1 · · · yn · in.
10
Similarly,
(2.1.2)
PATRICK LE MEUR
ϕ(εV ) = Xγ ′
βγ ′γ′ ⊗ xn−1 · · · x0εU ,
where γ′ runs through all paths in Q∗ with source in and of length n, and where,
for all such paths γ′, then βγ ′ ∈ k and x = xn−1, . . . , x0 denotes the sequence in G
such that
• xt ∈ [G/Git ] for all t ∈ {0, . . . , n − 1}
• and γ′ is a path of the shape in → xn−1 · in−1 → · · · → xn−1 · · · x0 · i0.
Lemma 2.3. Using the decompositions (2.1.1) and (2.1.2), then
(f ϕ) = Xγ
(2.1.3)
where
αγ · β(y1···yn)−1(γ)∗ · χ(y1···yn)−1,γ · χU (h0) ,
• γ runs through all paths of Q with source i0 and of length n
• and, for all such paths γ, then
-- y = y1, . . . , yn denotes the sequence in G such that yt ∈ [G/Git ], for
all t ∈ {1, . . . , n} and γ is a path of the shape i0 → y1 · i1 → · · · →
y1 · · · yn · in for some sequence i1, . . . , in ∈ [G\Q0],
-- h0 is the element of Gi0 such as (c) in Proposition 1.13,
-- χ(y1···yn)−1,γ is the element of k
× and (y1 · · · yn)−1(γ) is the path in Q
such that (y1···yn)−1
γ = χ(y1···yn)−1,γ (y1 · · · yn)−1(γ),
-- and ((y1 · · · yn)−1(γ))∗ is the reverse path in Q∗ associated to the path
(y1 · · · yn)−1(γ) in Q.
Proof. For all sequences y = y1, . . . , yn and x = xn−1, . . . , x0 such as before the
statement of the lemma, denote by C(y) and C′(x) the sets of paths in Q and in
Q∗ of the shape i0 → y1 · i1 → · · · → y1 · · · yn · in and in → xn−1 · in−1 → · · · →
xn−1 · · · x0 · i0, respectively. For all y,
fy(εU ) = Xγ∈C(y)
αγγ ⊗ y1 · · · ynεU ;
hence, the following term which serves in the definition of (f ϕ),
(y1···yn)−1
y (εU ) ⊗ · · · ⊗ (y1···yn)−1
f (1)
f (n)
y
(εU ) ⊗ f (0)
y (εU ) ,
is equal to
Xγ∈C(y)
Similarly, for all x,
αγ · χ(y1···yn)−1,γ · (y1 · · · yn)−1(γ)
}
{z
∈M ⊗S n
⊗ εU .
ϕx(εV ) = Xγ ′∈C(x)
βγ ′γ′ ⊗ xn−1 · · · x0εU .
Therefore, given y, the "Πt"-term in (1.0.10) is equal to
Xγ∈C(y), γ ′∈C(x)
(2.1.4)
where
αγ · βγ ′ χ(y1···yn)−1,γ · b∗
n(cn) · · · b∗
2(c2) · b∗
1(c1) · χU (h0) · εU ,
• x = xn−1, . . . , x0 and h0 are the elements of G determined by y and by (a),
(b), and (c) in Proposition 1.13,
1, . . . , b∗
n are the arrows of Q∗ such that γ′ is ·
• b∗
b∗
n−→ ·
b∗
n−1−−−→ · · ·
b∗
1−→ ·,
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 11
• and c1, . . . , cn are the arrows of Q such that (y1 · · · yn)−1(γ) is ·
· · ·
cn−→ ·.
c1−→ ·
c2−→
By definition of the arrows b∗
n, the summand of (2.1.4) with index γ is non
bn−→ · and (y1 · · · yn)−1(γ) in Q are equal, that
b1−→ ·
zero only if the paths ·
is, if and only if γ′ = ((y1 · · · yn)−1(γ))∗. Summing (2.1.4) over all possible y yields
that (f ϕ)εU is equal to
1, . . . , b∗
b2−→ · · ·
Xγ
αγβ(y1···yn)−1(γ)∗χ(y1···yn)−1,γχU (h0)εU .
This proves (2.1.3).
(cid:3)
2.2. An example. In this example, Q is the following quiver
0
@
❃❃❃❃❃❃❃❃
^❃❃❃❃❃❃❃❃
1
/ 2
4
3
and G be the dihedral group of order 10
G = hc, τ c5 , τ 2 , τ cτ ci .
Denote by ε the group homomorphism G → {−1, 1} such that ε(c) = 1 and ε(τ ) =
−1. Denote the arrows of Q by xi,j : i → j. For convenience, denote the arrows of
Q∗ by x′
i,j . The actions of G on Q0 and M are assumed
to be the ones such that, for all i ∈ Q0 and g ∈ G, and for all arrows xi,j of Q,
i,j : i → j, hence (xj,i)∗ = x′
c · i = i + 1 mod 5, τ · i = 5 − i, gxi,j = ε(g)xg·i,g·j .
Let [G\Q0] be {0}. Note that G0 = {Id, τ }. Let [G/G0] be {Id, c, c2, c3, c4}. For
all s ∈ {0, 1}, denote by εs the following primitive idempotent of kG0,
and denote εs · kG0 by ρs. Finally, let irr(G0) be equal to {ρ0, ρ1}. In particular,
εs =
1
2
(Id +(−1)sτ )
Denote by W the following element of kQ,
e = e0 ∗ ε0 + e0 ∗ ε1 .
W = x0,1x1,2x2,3x3,4x4,0 − x0,4x4,3x3,2x2,1x1,0 .
Denote e · W · e by θ.
2.2.1. Computation of QG. By definition, for all t ∈ {0, 1},
M ∗(0, 0; ρt) = Span(x′
0,1 ⊗ cεt, x′
0,4 ⊗ c4εt) ,
where the action of τ is given as follows
(cid:26) x′
0,1 ⊗ cεt
0,4 ⊗ c4εt
x′
7−→ (−1)1+tx′
7−→ (−1)1+tx′
0,4 ⊗ c4εt
0,1 ⊗ cεt .
Hence, for all s, t ∈ {0, 1},
HomkG0(ρs, M ∗(0, 0; ρt)) = Span(ϕs,t) ,
where ϕs,t is the intertwiner such that
ϕs,t(εs) = x0,1 ⊗ cεt + (−1)s+t+1x0,4 ⊗ c4εt .
@
^
O
O
/
o
o
O
O
12
PATRICK LE MEUR
If, fs,t denotes the intertwiner ρs → M (0, 0; ρt) such that (fs,tϕt,s) = 1, for all
s, t ∈ {0, 1}, then QG may be taken equal to the following quiver
f0,0
(1, ρ0)
f1,1
/ (1, ρ1) .
f0,1
f1,0
2.3. The intertwiner Ξ(θ). By definition, Ξ(θ) decomposes as
(2.3.1)
Ξ(θ) = Xr,w∈{0,1}
Ξ(θ)r,w ,
where, for all r, w ∈ {0, 1}, the intertwiner Ξ(θ)r,w : ρr → M (0, 0, 0, 0, 0, 0; ρw) is
such that
(2.3.2)
Ξ(θ)r,w(εr) = εr · e · W · e · εw
= 1
2 (1 + (−1)r+w)x0,1x1,2x2,3x3,4x4,0 ⊗ εw
− 1
2 (1 + (−1)r+w)x0,4x4,3x3,2x2,1x1,0 ⊗ εw .
2.4. Intertwiners relative to M ∗ associated to paths of QG. Consider a path
γ : (0, ρr) → (0, ρs) → (0, ρt) → (0, ρu) → (0, ρv) → (0, ρw) of length 5 in QG. Then
ϕγ is an the following intertwiner ρw → M ∗(0, 0, 0, 0, 0, 0; ρr),
ϕγ = ϕs,r ⊛ ϕt,s ⊛ ϕu,t ⊛ ϕv,u ⊛ ϕw,v .
In view of (1.0.6), if ϕγ(εw) is considered as an element of kQ∗ ∗ G, then it equals
(2.4.1)
(x′
(x′
(x′
0,1 ⊗ c + (−1)v+w+1x′
0,1 ⊗ c + (−1)t+u+1x′
0,1 ⊗ c + (−1)r+s+1x′
0,4 ⊗ c4) · (x′
0,4 ⊗ c4) · (x′
0,4 ⊗ c4) · εt .
0,1 ⊗ c + (−1)u+v+1x′
0,1 ⊗ c + (−1)s+t+1x′
0,4 ⊗ c4)·
0,4 ⊗ c4)·
2.5. Decomposition of θ into paths in QG. For all paths (0, ρr) → (0, ρs) →
(0, ρt) → (0, ρu) → (0, ρv) → (0, ρw) of length 5 in QG, Table 1 describes the terms
appearing in (2.1.3) when f is taken equal to Ξ(θ)r,w (see (2.3.2)) and ϕ is taken
equal to ϕs,r ⊛ ϕt,s ⊛ ϕu,t ⊛ ϕv,u ⊛ ϕw,v (see (2.4.1)). Thus,
(y1 ···y5 )−1 ,γ
γ
y1, . . . , y5
χ
(y1 · · · y5)−1(γ)
β
(y1 ···y5 )−1 (γ)∗
h0
χρr (h0)
x0,1 x1,2x2,3 x3,4x4,0
c, c, c, c, c
1
x′
4,3x′
0,4 x′
(−1)1+r+w
1
1
3,2 x′
2,1x′
1,0
x0,4 x4,3x3,2 x2,1x1,0
c−1 , c−1, c−1, c−1, c−1
1
x′
0,1 x′
1
2,3 x′
3,4x′
1,2x′
4,5
1
1
Table 1. Computation of (Ξ(θ)r,wϕs,r ⊛ ϕt,s ⊛ ϕu,t ⊛ ϕv,u ⊛ ϕw,v)
(Ξ(θ)r,sϕs,r ⊛ ϕt,s ⊛ ϕu,t ⊛ ϕv,u ⊛ ϕw,v) = −(1 + (−1)r+w) .
Applying (1.0.15) yields that the inverse image of e · W · e under the isomorphism
(1.0.1)
kQG
in QG of length 5.
−−−−−→ e · (TS(M ) ∗ G) · e is equal to −2Pγ γ, where γ runs through all paths
3. Details on the monoidal category of bimodules over groups
This section details some isomorphisms and properties in the monoidal cate-
gory (mod(Ae), ⊗A) and related to the operation, say ×, and the pairing h−−i
considered in Remark 1.5.
• Section 3.1 describes the left dual of a tensor product of objects in mod(Ae).
• Section 3.2 details the the adjunction (X ⊗A −) ⊢ (X ∗ ⊗A −) for a given
X ∈ mod(Ae).
/
o
o
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 13
• Section 3.3 details the compatibility of h−−i.
The existence of the isomorphisms and adjunctions discussed in these sections might
be part of the folklore (see [3, Exercise 2.10.16]). However, the proof of Theo-
rem 1.15 is based on their description, which is the reason for detailing these here.
For the sake of simplicity, the presentation is made independently of the monoidal
category (mod(Ae), ⊗A) and of the framework of Theorem 1.15.
Here are the conventions specific to this section. In any group, e denotes the
neutral element; and HomkG and ⊗kG are denoted by HomG and ⊗G, respectively,
for all finite groups G.
Given finite groups G and H, by "a module GXH " is meant a kG− kH-bimodule,
by "a module GX" is meant a left kG-module, and by "by a module XH is meant a
right kH-module.
Given finite groups G and H and given a kG − kH-bimodule X, the dual vector
space X ∗ is a kH − kG-bimodule in a natural way (h · φ · g = φ(g · • · h) for all
g ∈ G, h ∈ H and φ ∈ X ∗). In particular, taking H to be the trivial group, this
defines a functor X 7→ X ∗ from left kG-modules to right kG-modules.
3.1. Tensor product of duals and dual of a tensor product.
Lemma 3.1. Let H be a finite group with order not divisible by char(k). For all
finite dimensional kH-modules XH and H Y , there is a functorial bijective mapping
(3.1.1)
Y ∗ ⊗H X ∗ → (X ⊗H Y )∗
φ2 ⊗ φ1
7→ (x ⊗ y 7→ Ph∈H φ1(x · h)φ2(h−1 · y)) .
Proof. The mapping is well defined. When X = kHH and Y = H kH, it identifies
with the following one, where {δh}h∈H denotes the canonical basis of (kH)∗,
(kH)∗ ⊗H (kH)∗ → (kH)∗
7→ δh1h2 ;
δh2 ⊗ δh1
and the latter is bijective. Since XH and H Y are finite dimensional and projective,
then (3.1.1) is bijective.
(cid:3)
The previous lemma generalises as follows. Given an integer n > 2, a sequence
of groups G0, . . . , Gn and a finite dimensional kGi−1 − kGi-bimodule Ui for every
i ∈ {1, . . . , n}, there is a functorial bijective mapping
(3.1.2)
X ∗
n ⊗Gn−1 · · · ⊗G1 X ∗
1 → (X1 ⊗G1 · · · ⊗Gn−1 Xn)∗ ,
which maps any φn ⊗ · · · ⊗ φ1 to the following linear form on X1 ⊗G1 · · · ⊗Gn−1 Xn,
x1 ⊗ · · · ⊗ xn 7→ Xg1,...,gn−1
n
Yt=1
φt(g−1
t−1 · xt · gt) ,
where gt runs through Gt (1 6 t 6 n − 1) and g0 = gn = e.
3.2. Adjunctions.
Lemma 3.2. Let G and H be finite groups with orders not divisible by char(k).
Let GXH , H V and GU be (bi)modules over kG and kH such that X and V are
finite dimensional. Then, there exists a functorial isomorphism
ΨV,U
X : HomH (V, X ∗ ⊗G U ) → HomG(X ⊗H V, U ) ,
such that, for all φ ∈ HomH (V, X ∗ ⊗G U ), x ∈ X and v ∈ V ,
(3.2.1)
(ΨV,U
X (φ))(x ⊗ v) = Xg∈G
(φ(2)(v))(g−1 · x)g · φ(1)(v) ,
where the following notation is used with sum sign omitted, for all v ∈ V ,
φ(v) = φ(2)(v) ⊗ φ(1)(v) ∈ X ∗ ⊗G U .
14
PATRICK LE MEUR
In particular, the pair of functors (X ⊗H −, X ∗ ⊗G −) between finite dimensional
left kH-modules and finite dimensional left kG-modules is adjoint.
Proof. Using the definition of the structure of H−G-bimodule of X ∗, simple changes
of variable in the sum "Pg∈G" show that ΨV,U
Since H V is finitely generated and projective, it suffices to prove the lemma
X identifies with the mapping λX,U
In this case, ΨV,U
assuming that V = kH.
defined (for all left kG-modules X and U ) by
X is indeed well-defined.
X ∗ ⊗G U → HomG(X, U )
φ ⊗ u 7→ (x 7→ Pg∈G φ(g−1 · x)g · u) .
Now, λkG,kG identifies with the following bijective mapping (recall that G ∈ k
×).
kG∗ ⊗G kG →
kG
φ ⊗ e
7→ Pg∈G φ(g−1)g .
Since GX is finite dimensional and projective, and since GU is projective, it follows
that λX,U is bijective. Thus, ΨV,U
(cid:3)
X is bijective.
Lemma 3.3. Let G, H, K be finite groups with orders not divisible by char(k).
Let GXH , H YK, KW and GU be finite dimensional modules. Then, the following
diagram commutes
(3.2.2) HomK(W, Y ∗ ⊗H X ∗ ⊗G U )
HomK(W, (X ⊗H Y )∗ ⊗G U )
W,X∗⊗G U
Y
Ψ
ΨW,U
X⊗H Y
HomH (Y ⊗K W, X ∗ ⊗G U )
/ HomG(X ⊗H Y ⊗K W, U ) ,
Ψ
Y ⊗K W,U
X
where the top horizontal arrow is obtained upon applying HomK(W, − ⊗G U ) to
(3.1.1).
Proof. In view of the compatibility of the involved mappings with direct sum decom-
positions of the finite dimensional projective modules KW and GU , it is sufficient
to prove the lemma assuming that W = kK and U = kG. In this case, the given
diagram identifies with the following one
(3.2.3)
Y ∗ ⊗H X ∗
(3.1.1)
β
(X ⊗H Y )∗
α
HomH (Y, X ∗)
/ HomG(X ⊗H Y, kG) ,
γ
where
• α is given by φ 7→ (x ⊗ y 7→ Pg∈G φ(g−1 · x ⊗ y)g),
• β is given by φ2 ⊗ φ1 7→ (y 7→ Ph∈H φ2(h−1 · y)h · φ1
{z }
• γ is given by f 7→ (x ⊗ y 7→ Pg∈G(f (y))(g−1 · x)g).
φ1(•h)
) and
It is direct to check that (3.2.3) commutes, which proves the lemma.
(cid:3)
3.3. Pairing of morphisms. Recall that every involved finite group is assumed
to be of order coprime to char(k).
Definition 3.4. Let G and H be finite groups whose orders are not divisible by
char(k). Let GXH , H V , and GU be (bi)modules such that X and V are finite
dimensional. Assume that U is simple. Define a pairing
(3.3.1)
h−−i : HomG(U, X ⊗H V ) ⊗k HomH (V, X ∗ ⊗G U ) → k
/
/
/
/
/
/
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 15
as follows. For all f ∈ HomG(U, X ⊗H V ) and φ ∈ HomH (V, X ∗ ⊗G U ),
(3.3.2)
hf φi · IdU = ΨV,U
X (φ) ◦ f
(following Lemma 3.2, the pairing h−−i is non degenerate).
This pairing is compatible with the following operation on morphisms.
Definition 3.5. Let G, H, and K be finite groups whose orders are not divisible
by char(k). Let GXH , H YK, GU , H V , and KW be (bi)modules. For all f1 ∈
HomG(U, X ⊗H V ) and f2 ∈ HomH (V, Y ⊗K W ), define f2 × f1 ∈ HomG(U, X ⊗H
Y ⊗K W ) as the following composite morphism
(3.3.3)
f2 × f1 : U
f1−→ X ⊗H V
(X⊗H f2)
−−−−−−→ X ⊗H Y ⊗K W .
Here is the above mentioned compatibility.
Lemma 3.6. Let
• G, H, and K be finite groups with orders not divisible by char(k),
• GXH , H YK, GU , H V and KW be (bi)modules such that X, Y , and W are
finite dimensional and U and V are simple,
• f1 ∈ HomG(U, X ⊗H V ) and f2 ∈ HomH (V, Y ⊗K W ) and
• φ1 ∈ HomH (V, X ∗ ⊗G U ) and φ2 ∈ HomK(W, Y ∗ ⊗H V ).
Denote by λ the following isomorphism obtained upon applying − ⊗G U to (3.1.1),
Y ∗ ⊗H X ∗ ⊗G U ∼−→ (X ⊗H Y )∗ ⊗G U .
Then, hf2 × f1λ ◦ (φ1 × φ2)i = hf2φ2i · hf1φ1i.
Proof. Using (3.2.2) yields that
X⊗H Y (λ ◦ (φ1 × φ2)) = ΨY ⊗K W,X ∗⊗GU
ΨW,U
X
◦ ΨW,U
Y
(φ1 × φ2) .
Recall that φ1 × φ2 is the following composite morphism
φ2−→ Y ∗ ⊗H V
Y ∗⊗H φ1
−−−−−−→ Y ∗ ⊗H X ∗ ⊗G U .
Since Ψ•,•
W
X and Ψ•,•
ΨY ⊗K W,U
X
Y
Y are adjunction isomorphisms, then
◦ ΨW,X ∗⊗GU
(φ1 × φ2) = ΨY ⊗K W,U
X
(φ1 ◦ ΨW,V
Y
(φ2))
= ΨV,U
X (φ1) ◦ (X ⊗H ΨW,V
Y
(φ2)) .
Therefore,
ΨW,U
X⊗H Y (λ ◦ (φ1 × φ2)) ◦ (f2 × f1) = ΨV,U
X (φ1) ◦ (X ⊗H ΨW,V
Y
(φ2))◦
(X ⊗H f2) ◦ f1
= ΨV,U
= hf1φ1i · hf2φ2i · IdU .
X (φ1) ◦ (X ⊗H hf2φ2i · IdV ) ◦ f1
The conclusion of the lemma then follows from the definition of h−−i.
(cid:3)
The previous lemma has the following generalisation. Consider
• an integer n > 2,
• finite groups G0, . . . , Gn with orders not divisible by char(k),
• a finite dimensional left kGi-module Ui for every i ∈ {0, . . . , n}, such that
U0, . . . , Un−1 are simple,
• a finite dimensional kGi−1 − kGi-bimodule Xi for every i ∈ {1, . . . , n},
• fi ∈ HomGi−1 (Ui−1, Xi ⊗Gi Ui) and φi ∈ HomGi(Ui, X ∗
i ⊗Gi Ui−1) for every
i ∈ {1, . . . , n}.
16
PATRICK LE MEUR
Denote by λ(n) the isomorphism obtained upon applying − ⊗G0 U0 to (3.1.2),
∼−→ (X1 ⊗G1 · · · ⊗Gn−1 Xn)∗ ⊗G0 U0 .
n ⊗Gn−1 · · · ⊗G1 X ∗
1 ⊗G0 U0
λ(n) : X ∗
Then,
(3.3.4)
hfn × · · · × f1λ(n) ◦ (φ1 × · · · × φn)i =
4. Intertwiners
n
Yt=1
hfiφii .
Let Q0, S, G, and M be as in Setting 1.1. This section establishes properties on
intertwiners relative to M which are needed to prove Theorem 1.15. This includes
the properties of ⊛ mentioned in Section 1 as well as Proposition 1.13 on the
existence of the pairing (−−). This is done in several steps,
• first, by proving that, in Definition 1.6, the kGi0 -module M (i; V ) is iso-
morphic to M (i) ⊗Gin V for a suitable kGi0 − kGin -bimodule M (i);
• next, by proving that, under this isomorphism, ⊛ corresponds to the oper-
ation × of Definition 3.5;
• and finally by defining (−−) so that, under this isomorphism, it corre-
sponds to the pairing h−−i of Definition 3.4.
These steps are proceeded in turn in sections 4.1, 4.2, and 4.3, respectively. They
use the bimodule M G defined as follows. Note that, the action of G on S yields
the skew group algebra S ∗ G.
Definition 4.1. Define M G as the following S ∗ G-bimodule.
• Its underlying vector space is M ⊗k
kG, a tensor m ⊗ g is denoted by m ∗ g
for all m ∈ M and g ∈ G.
• The actions of S and G on the left and on the right are such that, for all
m ∈ M , g, h, k ∈ G and s, s′ ∈ S,
s · (m ∗ g) · s′ = sm gs′ ∗ g
h · (m ∗ g) · k = hm ∗ hgk .
The S ∗ G-bimodule M ∗G is defined similarly after replacing M by M ∗.
4.1. The space M (i). Let n be a positive integer. Let i = i0, . . . , in a sequence in
Q0. Denote by M (i) the following kGi0 − kGin -bimodule,
(4.1.1)
M (i) = ei0 M Gei1 ⊗Gi1
· · · ⊗Gin−1
ein−1M Gein .
Note that this yields the kGin − kGi0 -bimodule M ∗(io). These bimodules are dual
to each other. Here is an explicit isomorphism used later. Consider the isomorphism
(3.1.2) taking Xt = eit−1 M ∗Geit for all t ∈ {1, . . . , n},
(4.1.2)
ein (M G)∗ein−1 ⊗Gin−1
· · · ⊗Gi1
ei1 (M G)∗ei0
(3.1.2)
−−−−−→ M (i)∗ .
Since the following mapping is a non degenerate pairing
[−−] :
M G ⊗k M ∗G → k
(m ∗ g) ⊗ (ϕ ∗ h)
7→ ϕ( g−1
m)δe,hg ,
it induces an isomorphism (of S ∗ G − S ∗ G-bimodules)
(4.1.3)
M ∗G → (M G)∗
•
7→ [−•]
the n-th tensor power of which induces an isomorphism of kGin − kGi0 -bimodules,
(4.1.4)
M ∗(io) ∼−→ ein (M G)∗ein−1 ⊗Gin−1
· · · ⊗Gi1
ei1 (M G)∗ei0 .
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 17
Composing (4.1.2) and (4.1.4) yields an isomorphism of kGin − kGi0 -bimodules
(4.1.5)
Λi : M ∗(io) ∼−→ M (i)∗ ,
which is induced by the mapping from (M ∗ G)⊗kn to ((M G)⊗kn)∗ which assigns
to any tensor (ϕn ∗ gn) ⊗ · · · ⊗ (ϕ1 ∗ g1) the following linear form on (M G)⊗kn,
(4.1.6)
(m1 ∗ g′
1) ⊗ · · · ⊗ (mn ∗ g′
n) 7→ Xk1,...,kn−1
n
Yt=1
ϕt( k−1
t g′−1
t mt)δe,gtk−1
t−1g′
tkt
,
where δ denotes the Kronecker symbol, kt runs through Git and kn = k0 = e.
The bimodules M (i) and M ∗(io) are hence objects of the monoidal category
(mod(Ae), ⊗A) in which they are left dual to each other. The following lemma
describes tensor products of these objects in this monoidal category. The same
statements are valid after replacing M by M ∗.
Lemma 4.2. Let m, n be positive integers. Let i′′ = i0, . . . , im+n be a sequence in
Q0. Denote the sequences i0, . . . , im and im, . . . , im+n by i and i′, respectively.
(1) M (i′′) = M (i) ⊗Gim M (i′).
(2) The following diagram is commutative
M ∗(i′o) ⊗Gim M ∗(io)
Λi′ ⊗Gim
Λi
/ M (i′)∗ ⊗Gim M (i)∗
M ∗(i′′o)
Λi′′
(3.1.1)
/ M (i′′)∗ .
Proof. (1) follows from the definition of the spaces M (•) and (2) follows from the
definition of the morphisms Λ• in terms of (3.1.1) and (4.1.3).
(cid:3)
The following result is the link between the bimodules M (•) and the modules of
Definition 1.6 which makes it possible to translate the results of Section 3 in terms
of intertwiners.
Lemma 4.3. Let i = i0, . . . , in be a sequence in Q0, where n > 1. Let Un ∈
mod(kGin ). There is an isomorphism in mod(kGi0 ),
(4.1.7)
given by
Θi;Un : M (i; Un)
∼
−→ M (i) ⊗Gin Un
m1 ⊗ · · · ⊗ mn ⊗ y1 · · · ynu 7→ m1y1 ⊗ y−1
1 m2y2 ⊗ · · · ⊗ (y1···yn−1)−1
mnyn ⊗ u .
Note that Θi;Un induces a functorial bijection, for all U0 ∈ mod(kGi0 ),
(4.1.8)
HomGi0
(U0, M (i; Un))
f
∼−→ HomGi0
7→
(U0, M (i) ⊗Gin Un)
Θi;Un ◦ f .
4.2. Compositions of intertwiners. It is now possible to compare the operation
⊛ on intertwiners to the operation × of Definition 3.5.
Lemma 4.4. Let i′′ = i0, . . . , im+n be a sequence in Q0, where m, n > 1. De-
note the sequences i0, . . . , im and im, im+1, . . . , im+n by i and i′, respectively. Let
U0 ∈ mod(kGi0 ), Um ∈ mod(kGim ), and Um+n ∈ mod(kGim+n ). Finally, let
f ∈ HomkGi0
(4.2.1)
As a consequence, f ′ ⊛ f ∈ HomGi0
(U, M (i; V )) and f ′ ∈ HomkGim (V, M (i′; W )). Then,
(Θi′;Um+n ◦ f ′) × (Θi;Um ◦ f ) = Θi′′;Um+n ◦ (f ′ ⊛ f ) .
(U0, M (i′′; Um+n)).
/
/
18
PATRICK LE MEUR
Proof. Let u ∈ U0. Since Θi′;Um+n ◦ f ′ lies in HomGim (Um, M (i′′) ⊗Gim+n
and Θi;Um ◦ f lies in HomGi0
((Θi′;Um+n ◦f ′)×(Θi;Um ◦f ))(u) =
(U0, M (i) ⊗Gim Um), then
(M (i)⊗kGim (Θi′;Um+n ◦f ′))◦(Θi;Um ◦f )(u) .
Um+n)
(3.3.3)
Using the definition of Θi;Um, see (4.1.7), the right-hand side is equal to
(M (i) ⊗kGim (Θi′;Um+n ◦ f ′))(Py f (1)
(u))ym ⊗ f (0)
· · · ⊗ (y1···ym−1)−1
y (u)y1 ⊗ y−1
y (u)) .
(f (m)
y
1 f (2)
y (u)y2 ⊗ · · ·
Using the definition of Θi′;Um+n , see (4.1.7), this is equal to
Py,y′ f (1)
f ′(1)
y′ (f (0)
· · · ⊗ (ym+1···ym+n)−1
y (u)y1 ⊗ · · · ⊗ (y1···ym−1)−1
y (u))ym+1 ⊗ y−1
y′ (f (0)
y (u)))ym+n ⊗ f ′(0)
(f ′(n)
m+1(f ′(2)
(f (0)
y (u)))ym+2 ⊗ · · ·
y′ (f (0)
(u))ym⊗
(f (m)
y′
y
y (u)) .
Using the definition of Θi′′;Um+n , see (4.1.7), this is equal to
Θi′′;Um+n(Py′′ f (1)
· · · ⊗ y1···ym(f ′(n)
y (u) ⊗ · · · ⊗ f (m)
(f (0)
y (u))) ⊗ y1 · · · ym+nf ′(0)
y
(u) ⊗ y1···ym(f ′(1)
y′ (f (0)
y (u))) .
y (u))) ⊗ · · ·
y′
y′ (f (0)
Finally, using the definition of ⊛, see (1.0.5), this is equal to (Θi′′;Um+n ◦(f ′ ⊛f ))(u).
Whence (4.2.1). Note that f ′ ⊛ f is a morphism of kGi0 -modules because
• Θi′′;Um+n is an isomorphism of kGi0 -modules,
• f and Θi;Um are morphisms of kGi0 -modules,
• f ′ and Θi′;Um are morphisms of kGim -modules,
• and the operation "×" yields morphisms of kGi0 -modules (see (3.3.3)).
(cid:3)
Lemma 4.4 shows that Definition 1.10 yields a, possibly non associative, algebra.
Proposition 4.5. Intw is an associative and unital algebra.
Proof. Note that Intw is unital. For all i ∈ [G\Q0] and U ∈ irr(Gi), denote by
εi,U the intertwiner IdU ∈ HomkGi(U, M (i; U )), recall that M (i; U ) = U . Then
Pi,U εi,U is a unity for ⊛.
In the framework of Definition 3.5, the equality f3 × (f2 × f1) = (f3 × f2) × f1
holds whenever both sides make sense because of the associativity of the tensor
product. In view of Lemma 4.4, this entails that ⊛ is associative.
(cid:3)
4.3. The pairing on intertwiners. Let n be a positive integer, let i = i0, . . . , in
be a sequence in Q0 and let U0 and Un be left modules over kGi0 and kGin . To
every ϕ ∈ HomGin (Un, M ∗(io; U0)) is associated a k-linear mapping
Φ(ϕ) : M (i; Un) → U0
defined as follows. For every sequence of representatives y1, . . . , yn (yt ∈ [G/Git ])
there exist unique sequences
x = xn−1, . . . , x0 (xt ∈ [G/Git ]) , and hn−1, . . . , h0 (ht ∈ Git )
such that, for all t ∈ {1, . . . , n},
(4.3.1)
(yt · · · yn)−1 = xn−1 · · · xt−1ht−1 ;
these sequences are determined by a decreasing induction: y−1
hty−1
n = xn−1hn−1 and
t = xt−1ht−1 for all t ∈ {1, . . . , n− 1}; then, for every m1 ⊗ · · ·⊗ mt ⊗ y1 · · · ynu
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 19
lying in M (i; Un), define the element Φ(ϕ)(m1 ⊗ · · · ⊗ mt ⊗ y1 · · · ynu) of U0, using
Notation 1.12, as
(4.3.2)
n
Yt=1
ϕx
(t)(u)(cid:16) (y1···yn)−1
mt(cid:17) h−1
0 (ϕx
(0)(u)) .
This does make sense, because mt ∈ y1···yt−1·it−1 My1···yt·it , because ϕx
form on xn−1···xt−1·it−1 Mxn−1···xt·it , and because of (4.3.1).
(t) is a linear
After appropriate identifications, Φ is related to the adjunction isomorphism
M(i) , ΘM ∗
M(i) of section 3.2 as explained in the following result. Recall that ΨUn,U0
io;U0
Ψ•,•
and Λi are defined in (3.2.1), (4.1.7), and (4.1.5), respectively.
Lemma 4.6. Keep the previous setting. Denote by ϕ′ the following composite
morphism of left kGin -modules,
Un
ϕ
−→ M ∗(io; U0)
ΘM ∗
io;U0−−−−→ M ∗(io) ⊗Gi0
U0
Λi⊗Gi0
−−−−−−−→ M (i)∗ ⊗Gi0
U0
U0 .
Then,
(4.3.3)
Φ(ϕ) = ΨUn,U0
M(i) (ϕ′) ◦ Θi;Un .
In particular, Φ(ϕ) ∈ HomGi0
Proof. The composition ΘM ∗
(M (i; Un), U0).
io;U0 ◦ ϕ is given by (see (4.1.7))
u 7→ Xx
(n)(u)xn−1 ⊗ x−1
ϕx
n−1ϕx
(n−1)(u)xn−2 ⊗ · · · ⊗ (xn−1···x1)−1
ϕx
(1)(u)x0 ⊗ ϕx
(0)(u) ,
where x = xn−1, . . . , x0 runs through all sequences of representatives (xt ∈ [G/Git ]
for all 0 6 t 6 n − 1). Hence, following (4.1.6), the morphism ϕ′ is given by
u 7→ Xx
φx(u) ⊗ ϕx
(0)(u)
where φx(u) is the linear form on M (i), which is a quotient of (M G)⊗kn, induced
by the linear form on (M G)⊗kn mapping any (m1 ∗ g1) ⊗ · · · ⊗ (mn ∗ gn) to
(4.3.4) Xk1,...,kn−1
n
Yt=1(cid:16) (xn−1xn−2···xt)−1
(ϕx
(t)(u))(cid:17)(cid:16) k−1
t mt(cid:17) · δe,xt−1k−1
t g−1
t−1gtkt
,
where δ denotes the Kronecker symbol, k0 = kn = e and kt runs through Git for
1 6 t 6 n − 1.
In order to simplify the expression of (4.3.4), consider a tensor (m1 ∗ g1) ⊗
· · · ⊗ (mn ∗ gn) in (M G)⊗kn; for all k1, . . . , kn−1 in G, the following assertions are
equivalent, keeping k0 = kn = e,
(i) (∀t ∈ {1, . . . , n − 1}) e = xt−1k−1
(ii) kt = (xt−1 · · · x1x0g1g2 · · · gt)−1 for all t ∈ {1, . . . , n}, and g1 · · · gn is the
t−1gtkt,
inverse of xn−1 · · · x1x0;
given that each (xn−1xn−2···xt)−1
and each kt is constrained to lie in Git , then (4.3.4) is equal to
(t)(u)) in (4.3.4) is a linear form on xt−1it−1 Mit
(ϕx
n
Yt=1
(ϕx
(t)(u)) ( xn−1···x1x0g1g2···gt−1 mt) δe,g1g2···gnxn−1···x1x0 ,
which simplifies to
(4.3.5)
n
Yt=1
(ϕx
(t)(u))(cid:16) (gtgt+1···gn)−1
mt(cid:17) δ(g1g2···gn)−1,xn−1···x1x0 .
20
PATRICK LE MEUR
Thus, the morphism ϕ′ is given by u 7→ Px φx(u) ⊗ ϕx
(0)(u), where φx(u) is
the linear form on M (i) induced by the mapping assigning (4.3.5) to any element
(m1 ∗ g1) ⊗ · · · ⊗ (mn ∗ gn) of (M G)⊗kn.
Hence (see Lemma 3.2), ΨUn,U0
M(i) (ϕ′) is the morphism from M (i) ⊗Gin Un to U0
induced by the mapping from (M G)⊗kn ⊗k Un to U0 which maps any (m1 ∗ g1) ⊗
· · · ⊗ (mn ∗ gn) ⊗ u to
φx(u)(g−1m1g1 ⊗ m2g2 ⊗ · · · ⊗ mngn) g(ϕx
(t)(u)( (gt···gn)−1
mt)δ(g1···gn)−1g,xn−1···x0
g(ϕx
(0)(u))
(0)(u)) .
Now, let m1 ⊗ · · · ⊗ mn ⊗ y1 · · · ynu ∈ M (i; Un). Recall that its image under
mnyn ⊗ u (see (4.1.7)). Hence, its
1 m2y2 ⊗ · · · ⊗ (y1···yn−1)−1
PxPg∈Gi0
= Pg , xQn
t=1 ϕx
Θi;Un is m1y1 ⊗ y−1
image under ΨUn,U0
M(i) (ϕ′) ◦ Θi;Un is
Xg,x
Yt=1
n
(t)(u)( (y1···yn)−1
ϕx
(4.3.6)
mt)δ(y1···yn)−1g,xn−1···x0
g(ϕx
(0)(u)) .
Note that, given g ∈ Gi0 and x, then, for all t ∈ {1, . . . , n},
• ϕx
• (y1···yn)−1
(t)(u) ∈ ( xn−1···xt−1·it−1 Mxn−1···xt·it)∗ and
mt ∈ (yt···yn)−1·it−1 M(yt+1···yn)−1·it ;
in particular, if for a given index (g, x) in (4.3.6) the product "Qn
t=1" is nonzero
then this product is the unique nonzero term of the sum (4.3.6). More precisely,
let {xt, ht}06t6n−1 be the unique collection of elements in G such that, for all
t ∈ {0, . . . , n − 1},
• xt ∈ [G/Git ],
• ht ∈ Git and
• (yt+1 · · · yn)−1 = xn−1 · · · xtht;
then, the only possibly nonzero term of the sum (4.3.6) is the one with index
(h−1
0 , x). Therefore, (4.3.6) equals
n
Yt=1
(t)(u)( (y1···yn)−1
ϕx
mt) h−1
0 (ϕx
(0)(u)) ,
which is equal to Φ(ϕ)(m1 ⊗ · · · ⊗ mn ⊗ y1 · · · ynu) (see (4.3.2)). Thus, Φ(ϕ) =
ΨUn,U0
M(i) (ϕ′) ◦ Θi;Un . This proves (4.3.3). The second assertion of the lemma is a
direct consequence of (4.3.3).
(cid:3)
It follows from Lemma 4.6 that the following mapping is bijective,
Φ : HomGin (Un, M ∗(io; U0)) → HomGi0
(M (i; Un), U0) .
This yields a non degenerate pairing defined as follows.
Definition 4.7. Let i = i0, . . . , in be a sequence in Q0, where n > 1. Let U0 ∈
mod(kGi0 ) be simple. Let Un ∈ mod(kGin ). Define (−−) to be the non degenerate
pairing
(−−) : HomGi0
(U0, M (i; Un)) ⊗k HomGin (Un, M ∗(io; U0)) → k ,
such that, for all f ∈ HomGi0
(4.3.7)
(U0, M (i; Un)) and ϕ ∈ HomGin (Un, M ∗(io, U0)),
(f ϕ) · IdU0 = Φ(ϕ) ◦ f ;
In the setting of this definition, it follows from (3.3.1) and (4.3.3) that
(4.3.8)
(f ϕ) = hΘi;Un ◦ f (Λi ⊗Gi0
It is now possible to prove Proposition 1.13.
U0) ◦ Θ′
i;U0 ◦ ϕi .
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 21
Proof of Proposition 1.13. The pairing (−−) is the one from Definition 4.7. More-
over, (f ϕ) u is equal to (1.0.10) because of the definition of Φ(ϕ), see (4.3.2). (cid:3)
The pairing (−−) is compatible with the composition of (dual) intertwiners in
the following sense.
Lemma 4.8. Let m, n be positive integers, let i′′ = i0, . . . , im+n be a sequence in
Q0, let U0 and Um be simple left modules over kGi0 and kGim , respectively, and
let Um+n be a left kGim+n -module. Then, for all intertwiners
f1 : U0 → M (i; Um) ,
f2 : Um → M (i′; Um+n) , ϕ2 : Um+n → M ∗(i′o; Um) ,
ϕ1 : Um → M ∗(io; U0) ,
where i = i0, . . . , im and i′ = im, . . . , im+n, the following holds
(4.3.9)
(f2 ⊛ f1ϕ1 ⊛ ϕ2) = (f2ϕ2) · (f1ϕ1) .
Proof. Define f ′
1, f ′
2, ϕ′
1, ϕ′
2 as follows
ϕ′
f ′
1 = (Λi ⊗Gi0
1 = Θi;Um ◦ f1 ,
2 = (Λi′ ⊗Gim Um) ◦ Θ′
f ′
2 = Θi′;Um+n ◦ f2 , ϕ′
U0) ◦ Θ′
i;U0 ◦ ϕ1 ,
i′;Um
◦ ϕ2 .
Then, (4.3.8) entails that (f2 ⊛ f1ϕ1 ⊛ ϕ2) equals
hΘi′′,Um+n ◦ (f2 ⊛ f1)(Λi′′ ⊗Gi0
U0) ◦ Θ′
i′′,U0 ◦ (ϕ1 ⊛ ϕ2)i ,
according to Lemma 4.4 this is equal to
h(Θi′,Um+n ◦ f2)
× (Θi,Um ◦ f1)
(Λi′′ ⊗Gi0
f ′
1
{z
}
f ′
2
{z
}
according to Lemma 4.2 this is equal to
U0) ◦h(Θ′
i,U0 ◦ ϕ1) × (Θ′
i′,Um ◦ ϕ2)ii ,
(4.3.10)
hf ′
1 × f ′
2λ ◦ (Λi′ ⊗Gim Λi ⊗Gi0
i,U0 ◦ ϕ1) × (Θ′
U0) ◦h(Θ′
i′,Um ◦ ϕ2)ii .
Now, since the following composite morphisms are equal due to the associativity of
the tensor product
M ∗ (i′ o) ⊗
Gim
Um
1⊗(Θ′
i;U0
◦ϕ1)
/ M ∗(i′ o ) ⊗
Gim
M ∗(io) ⊗
Gi0
U0
Λ
i′ ⊗Λi⊗1
/ M (i′ )∗ ⊗
Gim
M (i)∗ ⊗
Gi0
U0
and
M ∗(i′ o) ⊗
Gim
Um
Λ
i′ ⊗1
/ M (i′ )∗ ⊗
Gim
Um
1⊗(Θ′
i;U0
◦ϕ1)
/ M (i′ )∗ ⊗
Gim
M ∗ (io ) ⊗
Gi0
U0
1⊗Λi⊗1
/ M (i′ )∗ ⊗
Gim
M (i)∗ ⊗
Gi0
U0 ,
then (Λi′ ⊗Gim Λi ⊗Gi0
U0) ◦h(Θ′
i,U0 ◦ ϕ1) × (Θ′
i′,Um
◦ ϕ2)i is equal to
[(Λi ⊗Gi0
U0) ◦ Θ′
i;U0 ◦ ϕ1
] × [(Λi′ ⊗Gim Um) ◦ Θ′
i′;Um ◦ ϕ1
].
ϕ′
1
{z
}
ϕ′
2
{z
}
Therefore,
(4.3.10)
=
=
Lemma 3.6
=
(4.3.8)
hf ′
hf ′
2 × f ′
2ϕ′
1λ ◦ (ϕ′
1 × ϕ′
1ϕ′
1i
(f2ϕ2) · (f1ϕ1) .
2i · hf ′
2)i
Thus, (f2 ⊛ f1ϕ1 ⊛ ϕ2) = (f1ϕ1) · (f2ϕ2).
(cid:3)
/
/
/
/
/
22
PATRICK LE MEUR
5. Proof of the main results
This section proves Theorems 1.15 and 1.16. Keeping Q0, S, G, and M as in
Setting 1.1 and QG as in Setting 1.3, these proofs use the following notation:
• denote by e the idempotent Pi∈[G\Q0] ei of S, this idempotent is not to be
confused with e which is equal to P(i,U) ei ∗ εU , where (i, U ) runs through
the vertices of QG;
• denote by S1 the k-algebra Πi∈[G\Q0]kGi.
First, it is necessary to check that the quiver QG corresponds to the one given
by Demonet in [2].
Proof of Theorem 1.4. Following [2], see p. 1057, the algebra TS(M ) ∗ G is Morita
equivalent to e · (TS(M ) ∗ G) · e, which is isomorphic to the path algebra of a quiver
having the same vertices as QG and such that the arrows from a vertex (i, U ) to
a vertex (j, V ) form a basis of the vector space (ei ∗ εU ) · (M G) · (ej ∗ εV ). Now,
considering ei, ej, εU , and εV as elements of S ∗ G,
(ei ∗ εU ) · (M G) · (ej ∗ εV )
=
=
≃
≃
(4.1.7)
εU · (ei · (M G) · ej) · εV
εU · M (i, j) · εV
HomkGi (U, M (i, j) ⊗Gj V )
HomkGi (U, M (i, j; V )) .
Whence the first statement of the theorem. Now, if U = kGi ·εU for all (i, U ) ∈ QG,
then f (εU ) does belong to (ei ∗εU )·(M G)·(ej ∗εV ), which is contained in e·(TS(M )∗
G) · e, for all arrows f : (i, U ) → (j, V ) of QG. Whence the isomorphism (1.0.1). (cid:3)
The proof of Theorem 1.15 uses Morita equivalences in the graded sense. Two
(N-)graded finite dimensional k-algebras Λ1 and Λ2 are Morita equivalent in the
graded sense if Λ2 ≃ EndΛ1 (P )op for some graded projective Λ1-module P which
is a generator of the category of graded Λ1-modules, and hence of mod(Λ1). Both
Λ1 and Λ2 are isomorphic as graded algebras to basic graded finite dimensional
algebras; moreover Λ1 and Λ2 are Morita equivalent in the graded sense if and only
if their basic versions are isomorphic as graded algebras.
Proof of Theorem 1.15. Denote by Φ the algebra homomorphism kQG → Intwop
given in the statement of the theorem. Hence, Φ(γ) = fγ for all paths γ in QG.
(1) Let γ and γ′ be parallel paths in QG. It follows from the definition of ϕγ ′ ,
see Notation 1.14, and from (4.3.9) that
(5.0.1)
(fγϕγ ′) = (cid:26) 1
0
if γ = γ′,
otherwise.
Since Φ maps non parallel paths in QG into distinct components of Intw in the de-
composition (1.0.7), then Φ is injective. In order to prove that Φ is an isomorphism,
it is hence sufficient to prove that, for all integers n, the homogeneous components
of degree n of the graded algebras kQG and Intw have the same dimension. Note
that kQG is graded by the length of the paths in QG and Intw is graded so that
every intertwiner lying in a space of the shape HomkGi0
(U, M (i0, . . . , in; V )) is ho-
mogeneous of degree n.
First, following [2], see Theorem 1.4, the algebras kQG and e · (TS(M ) ∗ G) · e are
isomorphic k-algebras both Morita equivalent to TS(M ) ∗ G. Actually, the proof
given in [2] works in the graded setting.
Next, e may be considered as an idempotent of TS(M ) ∗ G. In this sense,
e · (TS(M ) ∗ G) · e = e · e(TS(M ) ∗ G) · e · e .
MORITA REDUCED VERSIONS OF SKEW GROUP ALGEBRAS OF PATH ALGEBRAS 23
Following [2], see p. 1057, the algebra e · (TS(M ) ∗ G) · e is Morita equivalent to
TS1(e·(M G)·e). Again, this is still true in the graded sense. Hence, e·(TS(M )∗G)·e,
which is basic, is isomorphic to e · TS1(e · (M G) · e) · e as a graded algebra.
Finally, by definition of e and e, the homogeneous component of degree n of
e · TS1(e · (M G) · e) · e is equal to
M
i0,in∈[G\Q0], U0∈irr(Gi0 ), Un∈irr(Gin )
This is equal to
(ei0 ∗ εU0 ) · (M G)⊗S1 n · (ein ∗ εUn) .
i0,...,in∈[G\Q0], U0∈irr(Gi0 ), Un∈irr(Gin )
M
εU0 · M (i0, . . . , in) · εUn .
This is isomorphic to
Mi0,...,in,U0,Un
HomkGi0
(U0, M (i0, . . . , in) ⊗Gin V
) .
}
≃M(i0,...,in;V )
{z
These considerations prove that the homogeneous components of degree n of
kQG and Intw are isomorphic. Thus, Φ is an isomorphism.
(2) follows from (5.0.1) and from the fact that Φ is surjective.
(cid:3)
Proof of Theorem 1.16. First, Ξ is bijective.
equality of vector subspaces of TS(M ) ∗ G,
Indeed, on one hand, there is an
e · (TS(M ) ∗ G) · e = Mi,y (cid:0) i0 My1·i1 ⊗k · · · ⊗k y1···yn−1·in−1 My1···yn·in(cid:1) ∗ y1 · · · ynkGi ,
where i runs through all sequences i0, . . . , in of [G\Q0] and y runs through all
sequences y1, . . . , yn such that yt ∈ [G/Git ] for all t ∈ {1, . . . , n}. On the other
hand, for all vertices (i, U ) and (j, V ) of QG and for all kGi − kGj-bimodules N ,
the following mapping is bijective,
HomkGi(U, N · εV ) −→ εU · N · εV
f
7→
f (εU ) .
Next, Ξ is a morphism of k-algebras. Indeed, for all f ∈ HomkGi0
(U, M (i; V ))
lying in Intw, then Ξ−1(f ) = f (εU ). Therefore, with the setting of Definition 1.9,
then (f ′ ⊛ f )(εU ) is equal to the product f (εV ) · f ′(εU ) in TS(M ) ∗ G, see (1.0.5).
Accordingly, Ξ−1, and hence Ξ, is a k-algebra homomorphism.
Finally, the diagram (1.0.14) is commutative. Indeed, all its arrows are k-algebra
homomorphisms. Hence, it suffices to prove the commutativity on the arrows of
QG, which follows from the definitions of Ξ, (1.0.1), and (1.0.11). Note that (1.0.15)
follows from (1.0.12).
(cid:3)
6. Acknowledgements
I thank Edson Ribeiro Alvares for helpful comments on previous versions of this
article.
References
[1] Raf Bocklandt, Travis Schedler, and Michael Wemyss. Superpotentials and higher order deriva-
tions. J. Pure Appl. Algebra, 214(9):1501 -- 1522, 2010.
[2] Laurent Demonet. Skew group algebras of path algebras and preprojective algebras. J. Algebra,
323(4):1052 -- 1059, 2010.
[3] Pavel Etingof, Shlomo Gelaki, Dmitri Nikshych, and Victor Ostrik. Tensor categories, volume
205 of Mathematical Surveys and Monographs. American Mathematical Society, Providence,
RI, 2015.
[4] Victor Ginzburg. Calabi-Yau algebras. arXiv:math/0612139 v3, 2006.
24
PATRICK LE MEUR
[5] S. Giovannini and A. Pasquali. Skew group algebras of Jacobian algebras. ArXiv e-prints, May
2018.
[6] Patrick Le Meur. Crossed-products of Calabi-Yau algebras by finite groups. arXiv:1006.1082v1
[math.RA], 2010.
[7] Idun Reiten and Christine Riedtmann. Skew group algebras in the representation theory of
artin algebras. J. Algebra, 92(1):224 -- 282, 1985.
Université Paris Diderot, Sorbonne Université, CNRS, Institut de Mathématiques
de Jussieu-Paris Rive Gauche, IMJ-PRG, F-75013, Paris, France
E-mail address: [email protected]
|
1706.02518 | 1 | 1706 | 2017-06-08T11:22:46 | Bounds on the number of ideals in finite commutative nilpotent $\mathbb{F}_p$-algebras | [
"math.RA"
] | Let $A$ be a finite commutative nilpotent $\mathbb{F}_p$-algebra structure on $G$, an elementary abelian group of order $p^n$. If $K/k$ is a Galois extension of fields with Galois group $G$ and $A^p = 0$, then corresponding to $A$ is an $H$-Hopf Galois structure on $K/k$ of type $G$. For that Hopf Galois structure we may study the image of the Galois correspondence from $k$-subHopf algebras of $H$ to subfields of $K$ containing $k$ by utilizing the fact that the intermediate subfields correspond to the $\mathbb{F}_p$-subspaces of $A$, while the subHopf algebras of $H$ correspond to the ideals of $A$. We obtain upper and lower bounds on the proportion of subspaces of $A$ that are ideals of $A$, and test the bounds on some examples. | math.RA | math |
BOUNDS ON THE NUMBER OF IDEALS IN FINITE
COMMUTATIVE NILPOTENT Fp-ALGEBRAS
LINDSAY N. CHILDS AND CORNELIUS GREITHER
Abstract. Let A be a finite commutative nilpotent Fp-algebra
structure on G, an elementary abelian group of order pn. If K/k is
a Galois extension of fields with Galois group G and Ap = 0, then
corresponding to A is an H-Hopf Galois structure on K/k of type
G. For that Hopf Galois structure we may study the image of the
Galois correspondence from k-subHopf algebras of H to subfields of
K containing k by utilizing the fact that the intermediate subfields
correspond to the Fp-subspaces of A, while the subHopf algebras
of H correspond to the ideals of A. We obtain upper and lower
bounds on the proportion of subspaces of A that are ideals of A,
and test the bounds on some examples.
Introduction
The motivation for this work is to understand the Galois correspon-
dence for certain Hopf Galois structures on field extensions.
Let K/k be a Galois extension of fields with Galois group G. Then
the Galois correspondence sending subgroups G′ of G to subfields K G′
of K containing k is, by the Fundamental Theorem of Galois Theory,
a bijective correspondence from subgroups of G onto the intermediate
fields between k and K.
In 1969 S. Chase and M. Sweedler [CS69] defined the concept of
a Hopf Galois extension of fields for a field extension K/k and H a
k-Hopf algebra acting on K as an H-module algebra. They proved
a weak version of the FTGT, namely, that there is an injective Ga-
lois correspondence from k-subHopf algebras H ′ of H to intermediate
7→ K H ′, the subfield of elements fixed under the
fields, given by H ′
action of H ′. But surjectivity was not obtained. Greither and Pareigis
[GP87] defined a class of non-classical Hopf Galois structures, the "al-
most classical" structures, for which surjectivity holds, but also gave an
example where it fails. Recent work of Crespo, Rio and Vela ([CRV15]
Date: September 24, 2018.
1
2
LINDSAY N. CHILDS AND CORNELIUS GREITHER
and especially [CRV16]) studied the image of the Galois correspondence
for Hopf Galois structures on separable extensions K/k with normal
closure K and found numerous examples where surjectivity fails. In
nearly all of the cases examined in [CRV16] the Galois group of K/K
is non-abelian.
In this paper we seek to quantify the failure of the FTGT for Hopf
Galois structures of the following type.
Let K/k be a Galois extension of fields with Galois group G, an el-
ementary abelian p-group of order pn. Suppose H is a k-Hopf algebra
of type G (that means, K ⊗k H ∼= KG), and K/k is a H-Hopf Galois
extension. As shown in [Ch15], [Ch16], [Ch17], building on work of
[CDVS06] and [FCC12], every H-Hopf Galois structure of type G on
a Galois extension of fields K/k with Galois group G, an elementary
abelian p-group, arises from a commutative nilpotent Fp-algebra struc-
ture A on the additive group G with Ap = 0. In [Ch17], it was shown
that the sub-K-Hopf algebras of H correspond to ideals of A. For a
Galois extension K/k whose Galois group is an elementary abelian p-
group (or equivalently, an Fp-vector space), the classical FTGT gives
a bijection between Fp-subspaces of G and intermediate fields. So let
i(A) denote the number of ideals of A, and s(A) the number of Fp-
subspaces of A. Then the proportion of intermediate fields k ⊆ E ⊆ K
that are in the image of the Galois correspondence for a H-Hopf Galois
structure on K/k arising from A is equal to i(A)/s(A).
As observed in [Ch17], that comparison implies immediately that if
A2 6= 0, then there are subspaces of A that are not ideals, and hence
the Galois correspondence cannot be surjective.
Let e be the unique integer such that Ae 6= 0 and Ae+1 = 0; we
assume throughout that e > 0 (that is, A is not zero) and e < p.
To quantify the failure of surjectivity of the FTGT for a Hopf Galois
structure corresponding to A, we obtain in section 2 of this paper a
general upper bound, depending only on e, on the ratio i(A)/s(A).
The upper bound implies, for example, that for e ≥ 3 and p ≥ 17,
i(A)/s(A) < 0.01.
Using information on the dimensions of the annihilator ideals of A,
we obtain in section 3 a lower bound on i(A).
The upper bound is based on a lower bound on the fibers of the
"ideal generated by" function G from subspaces of A to ideals of A. In
the final section we examine that lower bound on fibers of G, and the
inequalities of sections 2 and 3, for some examples.
BOUNDS ON IDEALS
3
Let s(n) denote the number of subspaces of an Fp-vector space of
dimension n. Then s(n) is a sum of Gaussian binomial coefficients, also
called q-binomial coefficients (where q = p). The first section of the
paper describes properties of these coefficients and obtains inequalities
relating s(m) and s(n) for m < n.
Throughout the paper, we assume that A has dimension n and that
Ap = 0. Recall that e is the largest number so that Ae 6= 0 (so Ae+1 =
0). All vector spaces are over Fp.
Our thanks go to the University of Nebraska at Omaha and to Griff
Elder for their hospitality and support.
1. Gaussian binomial coefficients
To compare the number of ideals of a commutative nilpotent Fp-
algebra A with the number of subspaces of A, we need to collect some
information concerning the number of subspaces of dimension k of an
Fp-vector space of dimension n. So we begin with Gaussian binomial
coefficients.
The Gaussian binomial coefficient, or q-binomial coefficient (here
q = p), is defined as
k(cid:21) =
(cid:20)n
=
(pn − 1)(pn − p) · · · (pn − pk−1)
(pk − 1)(pk − p) · · · (pk − pk−1)
(pn − 1)(pn−1 − 1) · · · (pn−(k−1) − 1)
(p − 1)(p2 − 1) · · · (pk − 1)
.
It counts the number of k-dimensional subspaces of Fn
p . So
(cid:2)n
0(cid:3) =(cid:2)n
n(cid:3) = 1, and (cid:2)n
n − k(cid:21) for all k,
k(cid:21) =(cid:20) n
(cid:20)n
k(cid:3) = 0 for k > n. Then
k(cid:21)
Xk=0(cid:20)n
s(n) =
n
is the total number of subspaces of Fn
the factors (pn − pr)/(pk − pr) by pn/pk in order to see that
p . Note that it suffices to replace
k(cid:21) ≥ pk(n−k),
(cid:20)n
and that (cid:2)n
k(cid:3) has order of magnitude pk(n−k) for large p.
4
LINDSAY N. CHILDS AND CORNELIUS GREITHER
(In fact, the rational function
k(cid:21)x
(cid:20)n
=
(xn − 1)(xn − x) · · · (xn − xk−1)
(xk − 1)(xk − x) · · · (xk − xk−1)
is a polynomial of degree (n − k)k in Z[x]. For let b(x), a(x) be the
numerator and denominator of (cid:2)n
Z[x]. Dividing b(x) by a(x) in Q[x] gives
k(cid:3)x. Both are monic polynomials in
b(x) = a(x)q(x) + r(x),
where deg(r(x)) < deg(a(x)). Since a(x) is monic, q(x) and r(x) are
in Z[x]. Now b(p)/a(p) = (cid:2)n
so the rational function r(p)/a(p) is also an integer for every prime p.
But
k(cid:3)p is a positive integer for every prime p,
lim
p→∞
r(p)
a(p)
= 0.
So r(p) = 0 for all primes greater than some fixed bound, and hence
r(x) = 0. So b(x)/a(x) = q(x) is in Z[x].)
The Gaussian binomial coefficients satisfy two recursive formulas,
analogous to that satisfied by the usual binomial coefficients:
k(cid:21) =(cid:20)n − 1
(cid:20)n
=(cid:20)n − 1
k (cid:21)
k − 1(cid:21) + pk(cid:20)n − 1
k − 1(cid:21).
k (cid:21) + pn−k(cid:20)n − 1
Using properties of the Gaussian binomial coefficients, we will now
obtain some inequalities relating the number of subspaces of Fp-vector
spaces of dimensions n, n − 1 and n − 2 for all n.
Let δ(n) = ⌊ n2
4 ⌋ =(n2/4
(n2 − 1)/4
if n is even
if n is odd.
Lemma 1.1.
a) For all n ≥ 2 we have s(n) ≥ pn−1s(n − 2).
b) If n > 1 is even, then s(n) ≥ 1
2pn/2s(n − 1).
c) If n > 0 is odd, then s(n) ≥ p(n−1)/2s(n − 1).
d) For n ≥ m ≥ 0 arbitrary, we have s(n) ≥ 1
2pδ(n)−δ(m)s(m). The
factor 1/2 may be omitted if m and n have the same parity or
if n is even.
BOUNDS ON IDEALS
5
Proof. a) Using the two recursion formulas for (cid:2)n
d(cid:3) in turn we find:
d (cid:21)
d − 1(cid:21) + pd(cid:20)n − 1
d − 1(cid:21) + pd(pn−1−d(cid:20)n − 2
d(cid:21) = (cid:20)n − 1
(cid:20)n
= (cid:20)n − 1
d − 1(cid:21).
≥ pn−1(cid:20)n − 2
d (cid:21))
d − 1(cid:21) +(cid:20)n − 2
Summing these for d = 1, . . . , n − 1 gives the required inequality.
b) Let n = 2k. We may calculate as follows:
s(n) ≥ (cid:20)n
k(cid:21)
1(cid:21) + . . . +(cid:20)n
= (pn−1(cid:20)n − 1
0 (cid:21) +(cid:20)n − 1
1 (cid:21)) + (pn−2(cid:20)n − 1
1 (cid:21) +(cid:20)n − 1
2 (cid:21)) + . . .
k (cid:21))
k − 1(cid:21) +(cid:20)n − 1
. . . + (pk(cid:20)n − 1
0 (cid:21) + pk(cid:20)n − 1
k − 1(cid:21)
1 (cid:21) + . . . pk(cid:20)n − 1
≥ pk(cid:20)n − 1
= pks(n − 1)/2.
c) Let n = 2k + 1. Using one recursive formula, then the other, we
get:
1(cid:21) ≥ pn−1(cid:20)n − 1
(cid:20)n
2(cid:21) ≥ pn−2(cid:20)n − 1
(cid:20)n
...
0 (cid:21);
0 (cid:21) ≥ pk(cid:20)n − 1
1 (cid:21);
1 (cid:21) ≥ pk(cid:20)n − 1
k − 1(cid:21) ≥ pn−(k−1)(cid:20)n − 1
(cid:20) n
k − 2(cid:21);
k − 2(cid:21) ≥ pk(cid:20)n − 1
6
LINDSAY N. CHILDS AND CORNELIUS GREITHER
(now we switch to the other recursive formula)
k (cid:21);
k(cid:21) ≥ pk(cid:20)n − 1
(cid:20)n
k + 1(cid:21);
k + 1(cid:21) ≥ pk+1(cid:20)n − 1
(cid:20) n
k + 2(cid:21);
k + 2(cid:21) ≥ pk(cid:20)n − 1
k + 2(cid:21) ≥ pk+2(cid:20)n − 1
(cid:20) n
...
Now observe that
Therefore
n − 1(cid:21) ≥ pn−1(cid:20)n − 1
(cid:20) n
k + 1(cid:21) =(cid:20) 2k
(cid:20)n − 1
k + 1(cid:21) = pk(2(cid:20)n − 1
n − 1(cid:21).
n − 1(cid:21) ≥ pk(cid:20)n − 1
k − 1(cid:21).
k + 1(cid:21) =(cid:20) 2k
k − 1(cid:21)).
k + 1(cid:21) +(cid:20)n − 1
k + 1(cid:21)) = pk((cid:20)n − 1
pk+1(cid:20)n − 1
Thus s(n) is at least as large as the sum of the left sides of the inequal-
ities, which is at least the sum of the right sides of the inequalities, and
in view of the last observation, the sum of the right sides is at least
pks(n − 1).
d) We first note that δ(n) − δ(n − 2) = n − 1, so by a),
s(n) ≥ pn−1s(n − 2) = pδ(n)−δ(n−2)s(n − 2).
Iterating this shows that if n > m and n ≡ m (mod 2) then
s(n) ≥ pδ(n)−δ(m)s(m).
If n is even and m is odd, then n/2 = δ(n) − δ(n − 1) and by b) we
find
s(n) ≥
so
1
2
n
2 s(n − 1) =
p
pδ(n)−δ(n−1)s(n − 1),
1
2
s(n) ≥
1
2
pδ(n)−δ(m)s(m).
If n is odd and m is even, then (n − 1)/2 = δ(n) − δ(n − 1), so by c),
s(n) ≥ p
n−1
2 s(n − 1) = pδ(n)−δ(n−1)s(n − 1),
hence
s(n) ≥ pδ(n)−δ(m)s(m).
(cid:3)
BOUNDS ON IDEALS
7
2. An upper bound on the number of ideals of A
In this section we obtain a general upper bound for the ratio i(A)/s(A)
of the number of ideals of A to the number of subspaces of A, for A an
arbitrary commutative nilpotent Fp-algebra of dimension n. To do so,
we consider the function G from subspaces of A to ideals of A which
associates to each subspace U the ideal G(U) = U + AU generated by
U, and we establish a lower bound on the cardinality of the fiber of
each ideal under this map (which is obviously surjective). But first, we
need to count subspaces with certain properties.
Recall that δ(t) = ⌊t2/4⌋. All vector spaces are over Fp, and the num-
ber of k-dimensional subspaces of W , an Fp-vector space of dimension
d, is (cid:2)d
k(cid:3).
We show:
Proposition 2.1. Let dim(W ) = d and let W0 be a fixed subspace of W
of dimension r. For k ≤ d − r, the number s(d, r; k) of k-dimensional
subspaces U of W with U ∩ W0 = (0) is equal to prk times the number
of k-dimensional subspaces of W/W0:
k (cid:21).
s(d, r; k) = prk(cid:20)d − r
Proof. Let V be a complementary subspace to W0, so that V ⊕W0 = W .
For k ≤ d − r, let U be a k-dimensional subspace of V , with basis
(z1, . . . , zk). For each choice a = (a1, . . . , ak) of elements of W0, the
subspace Ua of W generated by (z1 + a1, . . . , zk + ak) is k-dimensional
and has trivial intersection with W0. For suppose
c1(z1 + a1) + . . . ck(zk + ak) = a
in W0 for some c1, . . . , ck in Fp. Then, since V ⊕ W0 is a direct sum of
Fp-vector spaces,
c1z1 + . . . ckzk = 0.
Since z1, . . . zk are linearly independent, c1, . . . ck = 0, hence a = 0.
The same argument with a = 0 shows that (z1 + a1, . . . , zk + ak) is a
linearly independent set.
Finally, each choice of elements a = (a1, . . . , ak) of W0 gives a differ-
ent subspace Ua of A. For suppose zi + bi is in the space Ua. Then
zi + bi = c1(z1 + a1) + . . . + ci(zi + ai) + . . . + ck(zk + ak).
So
0 = c1(z1 + a1) + . . . + ((ci − 1)zi + ciai − bi) + . . . + ck(zk + ak).
8
LINDSAY N. CHILDS AND CORNELIUS GREITHER
But then
0 = c1z1 + . . . + (ci − 1)zi + . . . + ckzk.
So ci = 1, all other cj = 0, and the equation reduces to
ai − bi = 0.
Thus for each k-dimensional subspace U of W , we obtain prk k-dimensional
subspaces Ua of W with W ∩ W0 = (0).
(cid:3)
Corollary 2.2. Let W be a t-dimensional space and W0 ⊂ W a sub-
space of codimension 1. Then the number of subspaces of W not con-
tained in W0 is at least pδ(t).
Proof. First we remark that via a duality argument, the number of sub-
spaces of dimension k not contained in a fixed subspace of codimension
1 is the same as the number of subspaces of dimension t−k intersecting
a fixed subspace of dimension 1 trivially. Hence the preceding proposi-
tion is applicable; summing over all possible dimensions of U, we find
that the number of subspaces U ⊂ W not contained in W0 is
t−1
s(t, 1):=
=
≥
=
t−1
t−1
Xk=0
Xk=0
Xk=0
Xk=0
s(t, 1; k)
k (cid:21)
pk(cid:20)t − 1
pkpk(t−1−k)
pk(t−k)
t−1
≥ p⌊ t2
4 ⌋ = pδ(t).
(cid:3)
Recall that G is the map from subspaces of A to ideals of A defined
by
G(V ) = V + AV.
To simplify notation, we write G(S) instead of G(hSiFp) for any subset
S of A. To get a sense of the relationship between the number of
subspaces of A and the number of ideals of A, we will count the number
of elements in the fibers of G.
BOUNDS ON IDEALS
9
Assume e > 0 is minimal with Ae+1 = 0. (The zero algebra A = 0
can be safely excluded from our study.) Consider the chain
of annihilator ideals defined by
N1 ⊂ N2 ⊂ . . . ⊂ Ne = A
Nk := Annk(A) = {a ∈ Ax1x2 · · · xka = 0 for all x1, . . . xk in A}.
Let dimFp(Nk) = dk. Then the sequence (dk)k is obviously increasing,
and a little argument shows that 0 < d1 < d2 < . . . < de = n.
The strategy for bounding the number of ideals of A begins with the
following idea. Let Jt be the set of ideals J of A contained in Nt but
not contained in Nt−1. Then, since Nt is an ideal of A for all t, we have
G−1(J) = s(Nt) − s(Nt−1).
XJ∈Jt
The next lemma will help us find a lower bound on G−1(J).
Lemma 2.3. Let W = G({x}) = Fpx + Ax , W0 = Ax as above. Let
U be a subspace of W , not contained in W0. Then G(U) = W .
Proof. Let y be in U, y not in W0. Then G({y}) ⊆ G(U). After
multiplying y by a non-zero element of Fp, we can assume that y =
x − ax for some a in A. Then
y + ay + a2y + . . . + ae−1y = x
is in G({y}). So
W = G({x}) ⊆ G({y}) ⊆ G(U) ⊆ W.
(cid:3)
Let J be an ideal of A of Fp-dimension d, let s(J) (or s(d)) be the
number of subspaces of J, and let i(J) be the number of ideals of A that
are contained in J. Lemma 2.3 enables us to prove a result relating
the number of subspaces and the number of ideals contained in the
annihilator ideal Nt in A for each t.
Proposition 2.4. For each t with 1 ≤ t ≤ e, consider the ideal map
G restricted to the set of subspaces V of Nt that are not contained in
Nt−1. For each x in Nt \ Nt−1, let q(x) = dim(G(x)), and let qt =
minx∈Nt\Nt−1 q(x). Then for all ideals J in Jt,
Hence
G−1(J) ≥ pδ(qt).
pδ(qt)(cid:0)i(Nt) − i(Nt−1)(cid:1) ≤ s(Nt) − s(Nt−1).
10
LINDSAY N. CHILDS AND CORNELIUS GREITHER
Proof. Let J be an ideal contained in Nt, not contained in Nt−1. Let
x be in J, x not in Nt−1. Let W0 = Ax and W = G({x}) = Fpx + Ax.
Then W has dimension at least qt, and W0 has codimension 1 in W .
Let Y be a complement of W in J. Then for every subspace U of W
not contained in W0, we have G(U) = W and thus G(U + Y ) = J.
Whenever U and U ′ are distinct subspaces of W not contained in W0,
we have U +Y 6= U ′+Y . So the number of preimages of J = G({x}+Y )
is at least equal to the number of subspaces of W that are not contained
in W0. Since dim(W ) ≥ qt, that number of subspaces is ≥ pδ(qt) by
Corollary 2.2.
(cid:3)
Dividing both sides of the t-th inequality of Proposition 2.4 by pδ(qt)
and summing them over all t yields an upper bound for the number of
ideals of A:
Corollary 2.5.
i(A) ≤
e−1
Xt=1
(p−δ(qt) − pδ(qt+1))s(Nt) + p−δ(qe)s(Ne).
Omitting the negative terms and applying Lemma 1.1 d) yields the
following upper bound on i(A) in terms of s(A) (recall dt = dim Nt):
Corollary 2.6.
i(A) ≤(cid:0)
e−1
Xt=1
2p−δ(qt)+δ(dt)−δ(de) + p−δ(qe)(cid:1)s(A).
To make it easier to apply this inequality for general A, we show the
following simple lower bound on the quantity qt. (Recall it was defined
by qt = minx∈Nt\Nt−1 q(x) with q(x) = dim(G(x)).)
Proposition 2.7. For all t > 0 we have qt ≥ t.
Proof. This is clear for t = 1.
For t > 1 let x be in Nt and not in Nt−1. Let u1, u2, . . . , ut−1 in A so
that u1u2 · · · ut−1x 6= 0. Then for each k, xk = uk · · · ut−1x is in Nk and
not in Nk−1. So x1, . . . , xt−1, x are linearly independent in A. Thus
G(x) = Fpx + Ax has dimension at least t.
(cid:3)
In the next theorem we will use this lower bound on qt to get a
general, fairly elegant upper bound on i(A)/s(A) that only depends on
e, the length of the annihilator chain in A. However, in some of the
examples treated below it will be worthwhile to have a closer look at
BOUNDS ON IDEALS
11
qt; we will find it to be considerably larger than t, which will enable us
to sharpen the upper bound.
The general bound goes as follows.
Theorem 2.8. With the above hypotheses on A and e we have
i(A)
s(A)
≤
2e − 1
pδ(e)
.
Proof. In the inequality of Corollary 2.6, replace qt by t and observe
that since 1 < d1 < d2 < . . . < de, one has δ(de) −δ(dt) ≥ δ(e) −δ(t). If
we insert this into the inequality, the terms p±δ(t) cancel and we obtain
i(A) ≤
e−1
Xt=1
2p−δ(e)s(Ne) + p−δ(e)s(Ne) = (2e − 1)p−δ(e)s(A).
(cid:3)
For e = 2, 3 the inequalities of Theorem 2.8 are
i(A) ≤
i(A) ≤
s(A)
3
p
5
p2 s(A)
for e = 2;
for e = 3.
We can improve these bounds by some constant factors, (almost) with-
out imposing further conditions on the algebra A. Recall that n =
dim(A).
Proposition 2.9. For e = 2, we have
2
p
i(A) ≤
s(A)
whenever p ≥ 3 and n ≥ 3. For e = 3, we have
i(A) ≤
2
p2 s(A)
whenever p ≥ 3, n ≥ 4.
Proof. Case e = 2: From Corollary 2.5 with δ(qt) replaced by δ(t), we
have
To get the claimed inequality it suffices to assume that dim N1 = n − 1
and show that
1
i(A) ≤(cid:0)1 −
(cid:0)1 −
p(cid:1)s(N1) +
p(cid:1)s(n − 1) ≤
1
p
1
1
p
s(A).
s(n),
(cid:0)1 −
1
1
p
−
1
p(cid:1)s(n − 2) +(cid:0)
(cid:0)
p2(cid:1)
1
p
−
1
2
pn/2 +(cid:0)1 −
1
p2 s(n).
p2(cid:1)s(n − 1) ≤
p(cid:1)
1
pn−1 ≤
1
1
p2 ,
12
LINDSAY N. CHILDS AND CORNELIUS GREITHER
or (p − 1)s(n − 1) ≤ s(n). Using Lemma 1.1b) for n even it suffices to
show that
pn/2
2
> p − 1,
which holds for p ≥ 3, n ≥ 4, while for n odd, it suffices by Lemma
1.1c) to show that
which holds for p ≥ 3, n ≥ 3.
pn/2 > p − 1,
Case e = 3. From Corollary 2.5 we have
1
1
i(A) ≤(cid:0)1 −
p(cid:1)s(N1) +(cid:0)
1
p
−
p2(cid:1)s(N2) +
1
p2 s(A).
Since s(A) = s(n), the right side is maximized when s(A) = s(n), s(N2) =
s(n − 1), s(N1) = s(n − 2). To show that the right side is ≤ 2
suffices to show that
p2 s(A), it
Using Lemma 1.1b) for n even, we are reduced to showing that
which holds for p ≥ 3, n ≥ 4. Using Lemma 1.1c) for n odd, we see it
suffices to show that
1
p
which holds for n ≥ 5 and p ≥ 2.
pn−1 +(cid:0)
(cid:0)1 −
p(cid:1)
1
1
1
p(n−1)/2 ≤
1
p2 ,
−
1
p2(cid:1)
(cid:3)
The bounds of Theorem 2.8 and Proposition 2.9 imply:
Corollary 2.10. Suppose K/k is a Galois extension with elementary
abelian p-group G and is also a H-Hopf Galois extension where H
arises from a commutative nilpotent Fp-algebra structure A on the ad-
ditive group G, where Ae 6= 0, Ae+1 = 0 and e < p. Then i(A)/s(A) is
the proportion of intermediate fields that are in the image of the Galois
correspondence from sub-Hopf algebras of H, and i(A)/s(A) < 0.01 for
• e = 2, p ≥ 200,
• e = 3, p ≥ 17,
• e = 4, p ≥ 7,
• all e, p with 5 ≤ e < p.
BOUNDS ON IDEALS
13
3. A lower bound on the number of ideals
We now obtain a lower bound on the number of ideals of A, by
exhibiting a collection of ideals in A and estimating its size. Recall
that Ae 6= 0 = Ae+1 and that we defined
Nr = Annr(A) = {a ∈ Ax1x2 · · · xra = 0 for all x1, . . . , xr in A}.
Then Nr is an ideal of A, and
(0) ⊂ N1 ⊂ N2 ⊂ . . . ⊂ Ne = A,
all inclusions being proper.
We already defined dr = dim(Nr); let us put tr = dimFp(Nr/Nr−1).
For each r = 1, . . . , e, let Wr be a subspace of A so that
Nr = Wr ⊕ Nr−1.
(In particular, W1 = N1.) Then tr = dim(Wr),
A = W1 ⊕ W2 ⊕ . . . ⊕ We
t1 + t2 + . . . + te = n.
and
Proposition 3.1.
i(A) ≥ λ(A) := s(t1) + (s(t2) − 1) + . . . + (s(te) − 1).
Proof. For each r, 1 ≤ r ≤ e, and each non-zero subspace Vr of Wr,
let J = Nr−1 + Vr. Then J is an ideal of A. Indeed, we have AVr ⊂
ANr ⊂ Nr−1, and therefore AJ ⊂ ANr−1 + Nr−1 ⊂ Nr−1 ⊂ J.
The formula λ(A) of the proposition simply counts the number of
(cid:3)
ideals J just described.
Since for any m,
s(m) =
m
k(cid:21)
Xk=0(cid:20)m
and (cid:2)m
bound for the number of ideals in A:
k(cid:3) ≥ p(m−k)k, we can let tM = max tk and get a rough lower
i(A) ≥ pδ(tM ).
14
LINDSAY N. CHILDS AND CORNELIUS GREITHER
4. Some classes of examples
To see how sharp the bounds on ideals are that we obtained in the
last two sections, we look at some explicit classes of algebras.
Example 4.1. First, consider the "uniserial" e-dimensional algebra A
generated by x with xe+1 = 0.
In this case, for every element u in
Nt \ Nt−1, the dimension q(x) = dim(G({x}) is equal to t. We then see
that the general upper bound
i(A) ≤ (2e − 1)p−δ(e)s(A)
is in fact close to the true number i(A) = e + 1 for large p, since s(A)
is a polynomial in p of degree δ(e).
The lower bound λ(A) in this simple class of examples is e + 1.
Example 4.2. Let A be a "binomial" nilpotent algebra: A = hx1, x2, . . . xei
with x2
k = 0 for all k. Then dim(A) = 2e − 1, Ann(A) = (x1x2 · · · xe)
and dim(Nt/Nt−1) =(cid:0) e
t−1(cid:1).
bounded as follows:
Theorem 2.6 tells us that the ratio of ideals to subspaces for A is
i(A)
s(A)
≤
2e − 1
pδ(e) =
2e − 1
p⌊ e2
4 ⌋
.
But that inequality arose from minorizing qt = minx∈Nt\Nt−1 dim(G(x))
by t throughout. In this class of examples we can do better, having a
closer look at qt.
Proposition 4.3. Let A be the binomial algebra of dimension 2e − 1.
Then for every non-zero u in Nt \ Nt−1 we have
dim(G(u)) ≥ 2t−1.
Proof. For any given u in Nt \ Nt−1, pick a monomial summand y of u
in Nt \ Nt−1. Renumber the variables of A so that
y = x1x2 · · · xe−t+1.
Then we introduce an ordering on the set of all nonzero monomials of A
so that any monomial of Nk−1 comes after any monomial of Nk\Nk−1 for
all k, and the monomials within Nk \Nk−1 are ordered lexicographically.
Strictly speaking, this is a total ordering on the set of all monomials
up to multiplication with a nonzero scalar in Fp.
Call a family of monomials admissible if no two of them are equal
up to a nonzero scalar. Every nonzero z ∈ A has a unique "leading"
monomial m(z), according to the ordering. The following is easy to see:
BOUNDS ON IDEALS
15
if (zi)i∈I is a family of elements of A, such that the family of leading
monomials (m(zi))i is admissible, then (zi)i is Fp-linearly independent.
If w is any monomial, we have m(zw) = m(z)w.
Now consider the family F of monomials that consist only of factors
xe−t+2, . . . , xe; this family has 2t−1 entries, and is of course admissible.
If we multiply every element of this family by u, the leading terms just
get multiplied by the monomial y, so they again are an admissible fam-
ily. Hence the entries of the family uF are again linearly independent,
which shows that the ideal G(u) generated by u has dimension at least
2t−1.
(cid:3)
We illustrate how working with qt ≥ 2t−1 instead of the crude lower
bound qt ≥ t affects the upper bound on the ratio i(A)/s(A) of Theo-
rem 2.8 for a binomial algebra.
Consider the binomial algebra A = hx1, x2, x3, x4i with x2
dim(N1) = 1, dim(N2) = 5, dim(N3) = 11, dim(N4) = 24 − 1 = 15.
i = 0. Then
(Note N4 = A.) The general inequality 2.8 gives
i(A)
s(A)
≤
7
p4 .
Let us start afresh. From Corollary 2.5 we have
3
i(A) ≤
(p−δ(qt) − pδ(qt+1))s(Nt) + p−δ(4)s(A).
Xt=1
Omitting the negative terms gives
i(A) ≤
1
pδ(q1) s(N1) +
1
pδ(q2) s(N2) +
1
pδ(q3) s(N3) +
1
pδ(q4) s(N4).
Now δ(q1) = δ(1) = 0 and for t > 1, δ(qt) ≥ δ(2t−1) = 22t−4. So we
have
i(A) ≤ s(1) +
Now we use Lemma 1.1 d):
1
p
s(5) +
1
p4 s(11) +
1
p16 s(15).
s(15) ≥ pδ(15)−δ(1)s(1) = p56s(1);
s(15) ≥ pδ(15)−δ(5)s(5) = p50s(5);
s(15) ≥ pδ(15)−δ(11)s(11) = p26s(11).
i(A)
s(A)
≤ (
1
p56 +
1
p51 +
1
p30 +
1
p16 ) ≤
2
p16 .
So
16
LINDSAY N. CHILDS AND CORNELIUS GREITHER
This is a big improvement over the inequality above that comes from
the general approach.
However, the lower bound λ(A) on the number of ideals of A from
Proposition 3.1 is a polynomial in p of degree 9, while
2
p16 s(15) >
2
p16(cid:20)15
7(cid:21) >
2
p16 p56 = 2p40.
So there remains a large gap between the upper and lower bounds on
i(A).
In general, the gap between the upper and lower bounds for i(A)
arises because the upper bound is based on a lower bound on the sizes
of fibers of the ideal generator map
G : ( subspaces of A) → (ideals of A).
For J an ideal of Nk, not in Nk−1, we showed that G−1(J) ≥ pδ(qk)
where qk is the minimum of the dimensions of principal ideals G(x) for
x in Nk \ Nk−1. But for many nilpotent algebras A and many ideals J
of A, this lower bound greatly underestimates G−1(J). We illustrate
this with two examples.
Example 4.4. Let A be the "triangular" algebra A = hx, yi with
Ae+1 = 0. Here one sees that qu = t(t+1)
for u in Nt \ Nt−1. Let us look
at the case e = 2 in detail.
2
Let A = hx, yi with A3 = 0. Then A has a basis B = (x, y, x2, xy, y2),
and the annihilator N1 = Ann(A) has basis x2, xy, y2. Moreover N2 =
A. So we have d1 = 3 and d2 = 5.
Proposition 4.5. There are 3p2 + 4p + 6 ideals in A.
Proof. The lower bound λ(A) from Proposition 3.1 counts ideals of N1
and ideals of A properly containing N1: that number is
λ(A) = s(t1) + s(t2) − 1
= s(3) + s(2) − 1
= (2p2 + 2p + 4) + (p + 2)
= 2p2 + 3p + 6.
To determine the number of ideals of A we let ¯A = A/N1. This is the
two-dimensional algebra spanned by ¯x and ¯y with zero multiplication.
We classify ideals J ⊂ A by their image ¯J in ¯A. Those with ¯J = 0 are
simply the subspaces of N1, which we've already counted. One easily
BOUNDS ON IDEALS
17
sees that ¯J = ¯A only happens once, for J = A, and since that ideal
contains N1, it is already counted.
There remains the case where ¯J is one-dimensional. There are p + 1
one-dimensional subspaces of ¯A, but by applying suitable automor-
phisms of A it suffices to count ideals with ¯J = Fp ¯x, and multiply
that count by p + 1. All such J contain x2 and xy, so the question is
whether they contain y2. If yes, J is simply the linear span of x and
N1 and has been counted. If no, then J contains an element x + ay2
for a unique scalar a ∈ Fp; this scalar determines the ideal. So there
are p such ideals mapping onto Fp¯x. Thus the count of ideals J with
¯J one-dimensional is p(p + 1). Adding that number to L(A) gives the
result.
(cid:3)
We can write down all of the ideals explicitly and determine their
fibers. The notation (m) denotes "subspace generated by m". In the
list, a, b, d are arbitrary elements of Fp.
A = G(x, y)
J1 = J1(a, d) = G(x + ay + dy2) = (x + ay + dy2, x2 − a2y2, xy + ay2)
J15 = J15(a) = G(x + ay, y2) = (x + ay, x2, xy, y2)
J2 = J2(b) = G(y + bx2) = (y + bx2, xy, y2)
J23 = G(y, x2) = (y, x2, xy, y2)
and finally all subspaces of the ideal N1 = (x2, xy, y2).
To describe the subspaces of A, choose the basis (x, y, x2, xy, y2) of A.
Looking at row vectors of coordinates with respect to that basis yields
a bijection between subspaces of A and row spaces of 5 × 5 matrices
with entries in Fp. Those row spaces are in bijective correspondence
with the set of 5 × 5 reduced row echelon matrices. Those, in turn, can
be categorized by specifying the columns where the pivots occur: the
number of pivots specified defines the dimension of the subspace. Thus
the label (124) denotes the 5 × 5 reduced row echelon matrix
(we omit all rows of zeros), where the five unspecified entries can be ar-
bitrary elements of Fp. Thus there are p5 subspaces of A corresponding
to echelon forms with label (124).
1 0 · 0 ·
0 1 · 1 ·
0 0 0 1 ·
18
LINDSAY N. CHILDS AND CORNELIUS GREITHER
The echelon forms (3), (4), (5), (34), (35), (45), (345) define the
non-zero subspaces of N1. Those subspaces are also ideals of A since
multiplication on N1 = Ann(A) is trivial.
Every echelon form that includes both 1 and 2 defines a subspace of
A that generates the ideal A. It is possible to discuss all other forms
in turn, finding the ideals generated by the corresponding subspaces
and the exact size of the fiber of G. Since this is repetitive and space-
consuming, we only write out what happens for three echelon forms.
(1) has the form (1, a′, b′, c′, d′) and generates J1(a, d) for a = a′ and
d = d′ + b′a′2 − c′a′. So for each (a, d) there are p2 subspaces of type
(1) that generate J1(a, d).
(13) has the form (cid:18)1 a′ 0 c′ d′
0 0 1 e′ f ′(cid:19). If a′ = a, d′ − c′a + e′a2 = d
and f ′ = e′a − a2 , then it generates J1(a, d). In that case, for each
(a, d) there are p2 subspaces of type (13) that generate J1(a, d).
If
a′ = a and f ′ 6= e′a − a2, then the subspace generates J15(a). In that
case the choices for (c′, d′, e′, f ′) yield p3(p − 1) subspaces of type (13)
that generate J15(a).
(14) has the form (cid:18)1 a′
0 0
b′ 0 d′
0 1 f ′(cid:19). If f ′ = a′ = a then the sub-
space generates J1(a, d) for all p choices of b′; otherwise for a′ = a 6= f ′
there are p2(p − 1) choices of (b′, d′, f ′) for subspaces of type (14) that
generate J15(a).
As noted, we omit the (easy) discussion of the remaining forms (15),
(134), (135), (2), (23), (234), (235), (2345), (24), (25), (245).
Adding up the number of subspaces that generate each ideal, we get
the tables below. Here a, b, c are arbitrary elements of Fp.
# of ideals
fiber size
ideals
A
J1(a, d)
J15(a)
J2(b)
J23
1
p2
p
p
1
2p6 + p5 + 2p4 + p3 + p2 + 1
2p2 + p + 1
p4 + p3 + p2 + 1
2p2 + p + 1
p4 + p3 + p2 + 1
1
ideal of N1 2p2 + 2p + 4
The center column sums to the number of ideals of A.
The total number of subspaces of A accounted for by fibers of ideals
of each type is:
BOUNDS ON IDEALS
19
ideals
A
J1(a, c)
J15(a)
J2(b)
J23
ideal of N1
# subspaces
2p6 + p5 + 2p4 + p3 + p2 + 1
2p4 + p3 + p2
p5 + p4 + p3 + p
2p3 + p2 + p
p4 + p3 + p2 + 1
2p2 + 2p + 4
The right column sums to s(5) = the number of subspaces of A.
Let us compare this with our more general results. We have
s(5) = 2p6 + 2p5 + 6p4 + 6p3 + 6p2 + 4p + 6.
Given that i(A) = 3p2 + 4p + 6 by Prop. 4.5, the inequality of Propo-
sition 2.9 comes out as
3p2 + 4p + 6 ≤ 4p5 + 4p4 + 12p3 + 12p2 + 12p + 8 + 12p−1.
This inequality was based on assuming that every ideal J not contained
in N1 has dimension ≥ 2, and so G−1(J) ≥ pδ(2) = p .
In Corollary 2.6, the factor in brackets between ≤ and s(A) evaluates
to 2p−4 + p−2, using d1 = 3, d2 = 5 and q1 = 1, q2 = 3. This assumed
that every ideal J not contained in N1 has dimension ≥ 3, so G−1(J) ≥
pδ(3) = p2. Then the inequality is
3p2 + 4p + 6 ≤ 2p4 + 2p3 + 10p2 + 10p + 18 + r(p),
where r(p) = 12p−4 + 8p−3 + 18p−2 + 16p−1 is always positive but tends
to 0 for p → ∞.
Looking at the actual sizes of the fibers of G in this example, the
inequality G−1(J) ≥ p2 has the correct power of p for principal ideals
J. But the non-principal ideals J23, J15(a) and A that are not contained
in N2 have fibers with cardinalities of order p4, p4 and p6, respectively.
This helps explain why the general upper bound on ideals is loose.
Example 4.6. Let A = hx, y, zi with x2
three variables. Then A4 = 0 (e = 3) and A has a basis
i = 0, the binomial algebra in
(x, y, z, xy, xz, yz, xyz)
with N1 = (xyz), N2 = (xy, xz, yz, xyz). The number of subspaces of
A is
s(7) = 2p12 + 2p11 + 6p10 + 8p9 + terms in p of lower degree.
The number of ideals of A turns out to be
i(A) = 7p2 + 4p + 8.
20
LINDSAY N. CHILDS AND CORNELIUS GREITHER
The lower bound on i(A), the number of ideals of A, is
λ(A) = s(1) + (s(3) − 1) + (s(3) − 1) = 4p2 + 4p + 8.
An upper bound on i(A) can be obtained by using Proposition 4.3,
which says that the dimension of a principal ideal of A not contained
in N2 is at least 4. Then we get
1
p4 )s(A)
i(A) ≤ (
1
2
p · pδ(7)−δ(4) +
1
p4 )s(A)
pδ(7)−δ(1) +
1
p12 +
2
p4 s(A) ∼ 4p8 + 2p7 + . . . .
2
p9 +
= (
≤
To see why this upper bound on i(A) is off by a factor of a constant
times p6, we can determine G−1 for the ideals of A, by methods in
the last example. We omit the details. But we observe first that the
fibers of the 2p2 + 3p + 4 ideals of N2 account in total for s(4) =
p4 + 3p3 + 4p2 + 3p + 5 subspaces of A. So most subspaces of A generate
ideals not contained in N2.
In obtaining our upper bound, we used that for principal ideals of A
not contained in N2, G−1(J) ≥ p4. But in fact, we find that:
• For the p2 + p + 1 principal ideals J of the form G(x + by + cz) with
bc 6= 0 in Fp, the dimension of J ≥ p5, so G−1(J) ≥ pδ(5) = p6, not
p4. Thus this set of ideals is generated by approximately p8 subspaces
of A.
• For the p2 + p + 1 non-principal ideals of the form
G(x + bz, y + cz), G(x + by, z), G(y, z),
each is generated by at least p9 subspaces of A. Thus this set of ideals
is generated by approximately p11 subspaces of A.
• Finally, for the ideal A = G(x, y, z) itself, every subspace of A
whose reduced row echelon form has the form (123 . . .) generates A,
and summing the number of such subspaces yields
G−1(A) ≥ 2p12 + p11 + 2p10 + 2p9 + . . . .
Comparing that to s(7) = s(A) above, it is evident that the weak-
ness in the upper bound we found for i(A) arises from the considerable
underestimation of the size of G−1(J) for non-principal ideals not con-
tained in N2, and, in particular, on the size of G−1(A): G−1(A) is a
polynomial in p of the same degree as s(A).
BOUNDS ON IDEALS
21
This last fact turns out to be true in general. One can show (proof
omitted) that G−1(A) is always a polynomial in p with the same
degree as the polynomial s(A), under the fairly mild assumption that
A as an Fp-algebra is generated by at most dim(A)/2 elements.
From these examples it appears that any substantial tightening of
the upper bound for the ideals of A will require a more nuanced look
at the fibers of non-principal ideals whose Fp-dimension is close to the
dimension of A.
However, the primary objective of this paper has been achieved. Let
L/K be a Galois extension with elementary abelian Galois group an
elementary abelian p group G. If L/K is a H-Hopf Galois extension
of type G corresponding to a commutative nilpotent algebra structure
A on G with Ap = 0, then the upper bound on i(A)/s(A) in section
2, weak as it may be for some examples, still provides the first general
quantitative estimate on how far from surjective is the Galois corre-
spondence for the Hopf Galois structure on L/K.
References
[CS69]
[Ch15]
[Ch16]
[Ch17]
[CDVS06] A. Caranti, F. Dalla Volta, M. Sala, Abelian regular subgroups of the
affine group and radical rings, Publ. Math. Debrecen 69 (2006), 297 --
308.
S. U. Chase, M. E. Sweedler, Hopf Algebras and Galois Theory,
Springer LNM 97 (1969).
L. N. Childs, On abelian Hopf Galois structures and finite commutative
nilpotent rings, New York J. Math. 21 (2015), 205 -- 229.
L. N. Childs, Obtaining abelian Hopf Galois structures from finite com-
mutative nilpotent rings, arxiv: 1604.05269
L. N. Childs, On the Galois correspondence for Hopf Galois structures,
New York J. Math. (2017), 1 -- 10.
T. Crespo, A. Rio, M. Vela, From Galois to Hopf Galois: theory and
practice, Contemp. Math. 649 (2015), 29 -- 46.
T. Crespo, A. Rio, M. Vela, On the Galois correspondence theorem
in separable Hopf Galois theory, Publ. Mat. (Barcelona) 60 (2016),
221 -- 234.
S. C. Featherstonhaugh, A. Caranti, L. N. Childs, Abelian Hopf Galois
structures on prime-power Galois field extensions, Trans. Amer. Math.
Soc. 364 (2012), 3675 -- 3684.
C. Greither, B. Pareigis, Hopf Galois theory for separable field exten-
sions, J. Algebra 106 (1987), 239 -- 258.
[CRV15]
[CRV16]
[FCC12]
[GP87]
E-mail address: [email protected]
E-mail address: [email protected]
|
1605.09132 | 1 | 1605 | 2016-05-30T07:59:27 | S-Noetherian generalized power series rings | [
"math.RA"
] | Let R be a ring with identity, (M;\leq) a commutative positive strictly ordered monoid and w_m an automorphism for each m \in M . The skew generalized power series ring R[[M,w]] is a common generalization of (skew) polynomial rings, (skew) power series rings, (skew) Laurent polynomial rings, (skew) group rings, and Mal'cev Neumann Laurent series rings. If S\subset R is a multiplicative set, then R is called right S-Noetherian, if for each ideal I of R, Is \subseteq J\subseteq I for some s\in S and some finitely generated right ideal J . Unifying and generalizing a number of known results, we study transfers of S-Noetherian property to the ring R[[M,w]]. We also show that the ring R[[M,w]] is left Noetherian if and only if R is left Noetherian and M is finitely generated. Generalizing a result of Anderson and Dumitrescu, we show that,when S\subset R is a-anti-Archimedean multiplicative set with a an automorphism of R, then R is right S-Noetherian if and only if the skew polynomial ring R[x,a] is right S-Noetherian. | math.RA | math |
S -Noetherian generalized power series rings
F. Padashnik, A. Moussavi and H. Mousavi
Department of Pure Mathematics, Faculty of Mathematical Sciences,
Tarbiat Modares University, Tehran, Iran, P.O. Box: 14115-134.1
Abstract
Let R be a ring with identity, (M, ≤) a commutative positive strictly ordered monoid and
ωm an automorphism for each m ∈ M . The skew generalized power series ring R[[M, ω]] is a
common generalization of (skew) polynomial rings, (skew) power series rings, (skew) Laurent
polynomial rings, (skew) group rings, and Mal'cev Neumann Laurent series rings. If S ⊂ R is
a multiplicative set, then R is called right S-Noetherian, if for each ideal I of R, Is ⊆ J ⊆ I
for some s ∈ S and some finitely generated right ideal J. Unifying and generalizing a number
of known results, we study transfers of S-Noetherian property to the ring R[[M, ω]]. We also
show that the ring R[[M, ω]] is left Noetherian if and only if R is left Noetherian and M is
finitely generated. Generalizing a result of Anderson and Dumitrescu, we show that,when
S ⊆ R is a σ-anti-Archimedean multiplicative set with σ an automorphism of R, then R is
right S-Noetherian if and only if the skew polynomial ring R[x; σ] is right S-Noetherian.
Keywords: S-Noetherian ring, skew generalized power series ring; right archimedean ring;
skew Laurent series ring; skew polynomial ring.
Subject Classification: 16P 40; 16D15; 16D40; 16D70; 16S36
1
Introduction
In [3], the
Throughout this paper, R is a ring (not necessary commutative) with identity.
authors introduced the concept of "almost finitely generated" to study Querr´e's characterization
of divisorial ideals in integrally closed polynomial rings. Later, Anderson and Dumitrescu [1]
abstracted this notion to any commutative ring and defined a general concept of Noetherian
rings. They call R an S-Noetherian ring if each ideal of R is S-finite, i.e., for each ideal I of
R, there exist an s ∈ S and a finitely generated ideal J of R such that Is ⊆ J ⊆ I. By [1,
Proposition 2(a)], any integral domain R is (R \ {0})-Noetherian; so an S-Noetherian ring is
not generally Noetherian. Also, M is said to be S-finite if there exist an s ∈ S and a finitely
generated R-submodule F of M such that sM ⊆ F. Also, M is called S-Noetherian if each
submodule of M is S-finite.
In [1], the authors gave a number of S-variants of well-known
results for Noetherian rings: S-versions of Cohens result, the Eakin-Nagata theorem, and the
Hilbert basis theorem under an additional condition. More precisely, in [1, Propositions 9
and 10], the authors showed that, if S is an anti-Archimedean subset of an S-Noetherian ring
R, then the polynomial ring R[X1, · · · , Xn] is also an S-Noetherian ring; and if S is an anti-
Archimedean subset of an S-Noetherian ring R consisting of nonzero divisors, then the power
1Corresponding author. [email protected] and [email protected].
[email protected]
[email protected] .
1
series ring R[[X1, · · · , Xn]] is an S-Noetherian ring. Note that if S is a set of units of R, then
the results above are nothing but the Hilbert basis theorem and a well-known fact that R[[X]]
is Noetherian if R is Noetherian. In [16, Theorem 2.3], Liu generalized this result to the ring
of generalized power series as follows: If S is an anti-Archimedean subset of a ring R consisting
of nonzero divisors and (Γ, ≤) is a positive strictly ordered monoid (defined in Secion 4), then
R[[M, ≤]] is S-Noetherian if and only if R is S-Noetherian and Γ is finitely generated. Note
that this recovers the result for the Noetherian case shown in [6, , Theorem 4.3] when S is
a set of units. Also, the authors in [14] study on transfers of the S-Noetherian property to
the constructions D + (X1, · · · , Xn)E[X1, · · · , Xn] and D + (X1, · · · , Xn)E[[X1, · · · , Xn]] and
Nagata's idealization is studied in [15].
The authors in [8, Theorem 7.7, page(65)] proved that R[M ] is Noetherian if and only if R is
Noetherian and M is finitely generated. Brookfield [6] proved that if (M, ≤) is a commutative
positively ordered monoid, then R[[M, ≤]] is right Noetherian if and only if R is right Noetherian
and M is finitely generated.
Ribenboim [22] and Varadarajan [26], have carried out an extensive study of rings of gener-
alized power series. They investigated conditions under which a ring of generalized power series
R[[M, ≤]] is Noetherian, where R is a commutative ring with identity and (M, ≤) is a strictly
ordered monoid.
In this paper we obtain results pertaining to Noetherian nature of generalized power series
rings. These considerably strengthen earlier results of Ribenboim [22], Varadarajan [26], Brook-
field [6], D. D. Anderson, and T. Dumitrescu[1], D. D. Anderson, B. G. Kang, and M. H. Park
[2] , D. D. Anderson, D. J. Kwak, M. Zafrullah [3] on this topic.
More precisely, we show that, if S is an σ-anti-Archimedean multiplicative subset of an S-
Noetherian ring R with an automorphism σ, then the skew polynomial ring R[x; σ] is also an
S-Noetherian ring; and if (M, ≤) is a commutative positively ordered monoid and ωm is an
automorphism over R for every m ∈ M , then the skew generalized power series ring R[[M, ω]] is
right Noetherian if and only if R is right Noetherian and M is finitely generated. When (M, ≤) is
a commutative positive strictly ordered monoid and ωm is an automorphism for each m ∈ M , we
unify and generalize the above mentioned results, and study transfers of S-Noetherian property
to the skew generalized power series ring R[[M, ω]].
2 S-Noetherian property on skew polynomial rings
If R is a commutative ring and S is a multiplicative subset of R, in [1], the authors proved that
the necessary condition for the ring of fractions RS to be a Noetherian ring is that R be an
S-Noetherian ring. In noncommutative rings, the situation is more complicated. In fact, if S is a
right (resp., left) permutable and right (resp., left) reversible (i.e S is right (resp., left) denomi-
nator set), then R has a ring of fraction RS−1 (resp., S−1R). In this situation, denominator sets
(both left and right denominator sets) act like a multiplicatively closed sets in the commutative
case. Our interest in this note is multiplicatively closed subsets (i.e. denominator subsets) in
noncommutative rings. First we define the notion of S-Noetherian rings for noncommutative
rings.
Definition 2.1. Let R be a ring and S a multiplicative subset of R. An ideal I of R is called
right S-finite (resp., S-principal), if there exists a finitely generated (resp., principal) right ideal
J of R and some s ∈ S such that Is ⊆ J ⊆ I.
2
A ring R is said to be right S-Noetherian (resp., S-PRIR), if each right ideal of R is right
S-finite (resp., S-principal). This definition can be done similarly for left side ideals.
Also, we say that an R-module M is right (or left) S-finite if M s ⊆ F (resp., sM ⊆ F ) for
some s ∈ S and a finitely generated submodule F of M . A module M is called right (or left)
S-Noetherian if each submodule of M is a right (or left) S-finite module.
The author in [1] justified the definition of S-Noetherian for commutative rings by prov-
ing some interesting properties of S-Noetherian ring. For example, they showed that if R is
S-Noetherian, then the ring of fractions RS is Noetherian and they found the conditions for the
reverse of this proposition.
Given rings R, T , an ideal J of T is said to be extended, if there exists an ideal I of R such
that ϕ(I) = J where ϕ : R −→ T is a ring monomorphism. Also, a ring R is von Neumann
regular if for every a ∈ R there exists an x in R such that a = axa. The center of a ring R is
denoted by Cent(R).
Proposition 2.2. Let R be a ring, S ⊆ R a multiplicative set and I a right ideal of R.
1) If R is von Neumman regular, S a denominator set and I ∩ S 6= ∅, then I is right
S-principal.
2) If S ⊆ T are right denominator subsets of R and R is right S-Noetherian(resp., S-PRIR),
then R is right T -Noetherian(resp., T -PRIR).
3) If R is von Neumman regular and S a denominator set, then R is right S-Noetherian
(resp., S-PRIR) if and only if R is right Noetherian (resp., PRIR).
4) If S is a denominator set and R is right S-Noetherian (resp., S-PRIR), then RS−1 is
right Noetherian.
5) If S is central in R, then the conditions 1-4 and those of [1, Proposition 2] follow.
Proof. 1) Let S ⊆ R be a denominator set, R a von Neumman regular ring and I a right ideal
of R. Then for each s ∈ I ∩ S, one can see that Is ⊆ Rs = s 1
s is the inverse of s in
RS−1. It is sufficient to see that 1
s Rs ⊆ R. For each s ∈ S, there exists a ∈ R such that sas = s,
so sa = s 1
s ∈ R and Rs ⊆ R, so
1
s Rs ⊆ R.
s = 1 (in RS−1). Thus sa = 1 and hence a = 1
s . Therefore 1
s Rs, where 1
2) Let S ⊆ T be denominator subsets of R. If R is right S-Noetherian (resp., S-PRIR), then
for each right ideal of R, there exists s ∈ S such that Is ⊆ J ⊆ I for some finitely generated
(resp., principal) right ideal of R. Since s ∈ S, S ⊆ T , s ∈ T which means that R is right
T -Noetherian (resp., T -PRIR).
3) Assume that R is a right Noetherian (resp., PRIR) ring. Each right ideal of R is finitely
generated (resp., principal). So for each s ∈ S, one can see that Is ⊆ I. Hence R is right
S-Noetherian (resp., S-PRIR). On the other hand, assume that R is right S-Noetherian (resp.,
S-PRIR), so there exists s ∈ S such that Is ⊆ J ⊆ I for some finitely generated (resp.,
principal) right ideal of R. Also suppose that sts = s for some t ∈ R. So Is ⊆ I. Also,
It ⊆ I, so Its ⊆ Is = Ists. So Its. 1
s . Hence It ⊆ Ist. Also Is ⊆ I yields that
Ist ⊆ It ⊆ Ist. So Ist = It. Thus Ists = Its which means that Is = Ists = Its. However
Its = I 1
s s = I. So Is = I. Thus I = Is ⊆ J ⊆ I and hence I = J, and since J is a
finitely generated (resp., principal) right ideal of R, so is I.
s ⊆ Ists. 1
s sts = I 1
4) This proof is an inspiration from [4, proposition 3.11 part (i)]. First, we claim that each
s = b ∈ J. So
ideal of RS−1 is extended. Let a right ideal J of ring of fraction RS−1 and x
3
s . s
1 ∈ J. s
1 ⊆ J. So x
1 = x
x
ϕ(x) ∈ ϕ(ϕ−1(J)), so x
is an ideal of RS−1 and s ∈ U (RS−1), so we have
1 ∈ J. Hence, ϕ−1( x
1 . s
1 ∈ ϕ(ϕ−1(J)). So x
1 ) ∈ ϕ−1(J) which means that x ∈ ϕ−1(J). Thus,
s ∈ ϕ(ϕ−1(J)). Note that ϕ(ϕ−1(J))
s = x.s
s = xs
s . s
xs
s
.
1
s
=
x
s
∈ ϕ(ϕ−1(J))
1
s
⊆ ϕ(ϕ−1(J)).
So b = x
for each ideal of RS−1. Thus J = ϕ(ϕ−1(J)) and J is an extended ideal of RS−1.
s ∈ ϕ(ϕ−1(J)) which implies J ⊆ ϕ(ϕ−1(J)). On the other hand, ϕ(ϕ−1(J)) ⊆ J holds
Let a right ideal K of ring of fraction RS−1. Since R is right S-Noetherian there exists s ∈ S
and a finitely generated (resp., principal) right ideal W of R such that ϕ−1(K)s ⊆ W ⊆ ϕ−1(K).
So ϕ(ϕ−1(K)s) ⊆ ϕ(W ) ⊆ ϕ(ϕ−1(K)). We know that ϕ(ϕ−1(K)s) = ϕ(ϕ−1(K))ϕ(s). Also,
ϕ(s) ∈ U (RS−1) and ϕ(ϕ−1(K)) = K. So K ⊆ ϕ(W ) ⊆ K. So K = ϕ(W ). Since W is finitely
generated, ϕ(W ) is finitely generated. So K is finitely generated which means that RS−1 is
right Noetherian.
5) The proof is straightforward by [1, Proposition 2].
Now we generalize a theorem of D.D. Anderson and Tiberiu Dumitrescu [1, Proposition 9],
for commutative polynomial ring R[x], in a more general setting. We show that if R is a right (or
left) S-Noetherian ring with an automorphism σ, then R[x; σ] is a right (or left) S-Noetherian
ring.
In [2] the authors defined the notion of anti-Archimedean multiplication set. Now we intro-
duce the notion of σ-anti-Archimedean multiplication set:
Definition 2.3. Let R be a ring with an automorphism σ and S a multiplicative set. Then R
is called left σ-anti -Archimedean over S, if there exists s ∈ S, such that
( \l≥1,ki≥0
Rσk1(s)σk2(s) · · · σkl(s)(cid:1) ∩ S 6= ∅.
Theorem 2.4. Let R be a ring with an automorphism σ and S ⊆ R a σ-anti-Archimedean
multiplicative set. Then R is right (or left) S-Noetherian if and only if R[x; σ] is right (or left)
S-Noetherian.
Proof. (⇒) We prove the theorem for the right version. The proof of left version is similar.
First, we claim that if D is a finitely generated R-module and R is a right S-Noetherian ring,
then D is a right S-Noetherian module. For this claim, assume that D is a finitely generated
right R-module. So there exists a finitely generated free right R-module F and a surjective
homomorphism π : F −→ D. We show that D is a right S-Noetherian R-module. For this, let
N := π−1(T ), for a submodule T of D. We have N ≃ I1 ⊕ I2 · · · ⊕ Il, for some right ideals Ii of
R, 1 ≤ i ≤ l. Since R is a right S-Noetherian ring, there exists si ∈ S such that Iisi ⊆ Ji for a
finitely generated ideals Ji of R, 1 ≤ i ≤ l. Now take s′ := s1s2 · · · sl ∈ S, we show that N s′ ⊆ K
for a finitely generated R-submodule K of F . One can see that N s1 = I1s1 ⊕ I2s1 ⊕ · · · ⊕ Ils1.
Since Ii is a right ideal of R so we have Iis1 ⊆ Ii for i 6= 1 and I1s1 ⊆ J1, for a finitely
generated right ideal J1 of R. So we have N s1 ⊆ J1 ⊕ I2 ⊕ I3 · · · ⊕ Il. Continuing in this
way, N s1s2 · · · sl ⊆ J1 ⊕ J2 ⊕ · · · ⊕ Jl ≃ K, where Ji is a finitely generated right ideal of
R,1 ≤ i ≤ l, and hence K is a finitely generated R-submodule of F . Thus N s′ ⊆ K and
hence F is a right S-Noetherian R-module. Next, since T = π(N ) and N s′ ⊆ K, we have
π(N s′) = π(N )s′ = T s′ ⊆ π(K). We know that K is finitely generated in F , so π(K) is finitely
4
generated R-submodule of D. Thus, T s′ ⊆ π(K) which means that T is S-finite. Since T is an
arbitrary R-submodule of D, D is a right S-Noetherian module.
Now, we prove that A := R[x; σ] is a right S-Noetherian ring. Let I be right ideal of A and
suppose that
J = {ri ∈ Rri is a leading coefficient of any polynomial in I} ∪ {0}.
It is easy to see that J is a right ideal. Since R is right S-Noetherian, Js ⊆ (a1R+a2R+· · ·+anR)
for some s ∈ S and ai ∈ J. So there exist polynomials fi ∈ I with fi = ai,nixni + · · · + a0,i. Let
d = max{ni}. Assume that T is the set of all polynomials in I with degree less than d. Obviously,
T is a finitely generated right R-submodule of A. So by the first claim, T is right S-Noetherian.
Hence there exist t ∈ S, gi ∈ T for 1 ≤ i ≤ m such that T t ⊆ (g1R + g2R + · · · + gmR). Let
i=1 bixi ∈ I, so bz ∈ J which means that bz ∈ (a1R + a2R + · · · + anR). Thus hσ−z(s)
h(x) =Pz
can be written as follows:
hσ−z(s) = v(1) + w(1) + q(1),
where v(1) ∈ (f1A + f2A + · · · + fnA), w(1) ∈ {f ∈ Ad + 1 ≤ deg(f ) ≤ z − 1} and q(1) ∈
T . Continuing in this way and multiplying σ−z+1(s), σ−z+2(s), · · · , σ−1−d(s) from right side
respectively, so there exists some v ∈ (f1A + f2A + · · · + fnA), w ∈ T such that
hσ−z(s)σ−z+1(s) · · · σ−d−1(s) = v + w.
Assume that si = σ−z+i and multipling t from right side, then hs1s2 · · · sz−dt = vt + wt. But
wt ∈ T t, so wt ∈ (g1R + g2R + · · · + gmR) ⊆ (g1A + g2A + · · · + gmA). Hence,
hs1s2 · · · sz−dt ∈ (f1A + f2A + · · · + fnA + g1A + · · · + gmA).
Since si's and t are independent from the choice of h ∈ I, we have
Is1s2 · · · sz−dt ⊆ (f1A + f2A + · · · + fnA + g1A + · · · + gmA).
Finally, since s1s2 · · · sz−dt ∈ S, the ideal I is S-finite and because I was chosen an arbitrary
right ideal of A, hence A is a right S-Noetherian ring.
(⇐) Let I be a right ideal of R. Suppose that
J = {f ∈ A the leading coefficient of f is in I}.
Then J is a right ideal of A. Since A is right S-Noetherian, there exists s ∈ S such that Js ⊆
K ⊆ J, where K is a finitely generated right ideal of A. Suppose that K = (f1A+f2A+· · ·+flA).
Let r ∈ I, then there exists some f ∈ J such that f s =P aifi. So if ri is the leading coefficient
of fi, 1 ≤ i ≤ l, then rs ∈ (r1R + r2R · · · + rlR). So Is ⊆ (r1R + r2R + · · · + rlR). Also, K ⊆ J,
so each leading coefficient of K is a leading coefficient of J. So (r1R + r2R + · · · + rlR) ⊆ I and
hence I is right S-finite and R is right S-Noetherian.
We have the following generalization of a theorem of D.D. Anderson and Tiberiu Dumitrescu
[1, Proposition 9].
Corollary 2.5. Let R be a (not necessarily commutative) ring and S ⊆ R an anti-Archimedean
multiplicative set. If R is S-Noetherian then so is the polynomial ring R[X1, X2, · · · , Xn].
5
3 Noetherian Skew Generalized Power Series Rings
Throughout this section, (M, ≤) is assumed to be a strictly ordered commutative monoid. The
pair (M, ≤) is called an ordered monoid with order ≤, if for every m, m′, n ∈ M , m ≤ m′ implies
that nm ≤ nm′ and mn ≤ m′n. Also, an ordered monoid (M, ≤) is said to be strictly orderd
if for every m, m′, n ∈ M , m < m′ implies that nm < nm′ and mn < m′n. Let (M, ≤) be
a partially ordered set. The set (M, ≤) is called Artinian if every strictly decreasing sequence
of elements of M stablized, and also (M, ≤) is called narrow if the number of incomparable
elements in every subset of M is finite. Thus, we can conclude that (M, ≤) is Artinian and
narrow if and only if every nonempty subset of M has at least one but only a finite number of
minimal elements.
The author in [24] introduced the ring of generalized power series R[[M ]] for a strictly ordered
monoid M and a ring R consisting of all functions from M to R whose support is Artinian and
narrow with the pointwise addition and the convolution multiplication. There are a lot of
interesting examples of rings in this form (e.g., Elliott and Ribenboim, [7]; Ribenboim,[23]) and
it was extensively studied by many authors, recently.
In [21], the authors defined a "twisted" version of the mentioned construction and study on
ascending chain condition for its principal ideals. Now we recall the construction of the skew
generalized power series ring introduced in [21]. Let R be a ring, (M, ≤) a strictly ordered
monoid, and ω : M → End(R) a monoid homomorphism. For m ∈ M , let ωm denote the image
of m under ω, that is ωm = ω(m). Let A be the set of all functions f : M → R such that
the support supp(f ) = {m ∈ M f (m) 6= 0} is Artinian and narrow. Then for any m ∈ M and
f, g ∈ A the set
χm(f, g) = {(u, v) ∈ supp(f ) × supp(g) : m = uv}
is finite. Thus one can define the product f g : M → R of f, g ∈ A as follows:
f g(m) = X(u,v)∈χm(f,g)
f (u)ωu(g(v)),
(by convention, a sum over the empty set is 0). Now, the set A with pointwise addition and
the defined multiplication is a ring, and called the ring of skew generalized power series with
rmxm, where
coefficients in R and exponents in M . To simplify, take A as a formal series Pm∈M
rm = f (m) ∈ R. This ring can be denoted either by R[[M ≤, ω]] or by R[[M, ω]] (see [18] and
[19]).
For every r ∈ R and m ∈ M we can defined the maps cr, em : M −→ R by
cr(x) =(r
; x = 1
0 ; Otherwise
, em(x) =(1 ; x = m
0 ; Otherwise
(3.1)
where x ∈ M . By way of illustration, cr(x) and em(x) are like r and xm in usual polynomial
ring R[x], respectively.
The following proposition which is proved in [11, Theorem 2.1], can characterize all Artinian
and narrow sets.
6
Proposition 3.1. Let (M, ≤) be an ordered set. Then the following conditions are equivalent
(1) (M, ≤) is Artinian and narrow.
(2) For any sequence (mn)n∈N of elements of M there exist indices n1 < n2 < n3 < · · · such
that mn1 ≤ mn2 ≤ mn3 ≤ · · · .
(3) For any sequence (mn)n∈N of elements of M there exist indices i < j such that mi ≤ mj.
The author in [6] introduced the concept of a lower set. A lower set of L is a subset I ⊆ L
such that x ≤ y ∈ I implies x ∈ I for all x, y ∈ L, (which we denoted by ⇓ L for the set of
lower sets of L ordered by inclusion). In this concept, we can ignore the condition narrow by
lower set, indeed it is proved that if L is a partially ordered set, then ⇓ L is Artinian if and
only if L is Artinian and narrow. He also showed that if α : K −→ L is strictly increasing map
between partially ordered sets, then if L satisfies Artinian (or Noetherian) property, then so is
K. Moreover, if α is surjective and ⇓ K satisfies Artinian (or Noetherian) property, then so
does ⇓ L.
An ordered monoid (M, ≤) is called positively ordered if m ≥ 0 for all m ∈ M .
In this
condition, m (cid:22) m′ implies m ≤ m′ for all m, m′ ∈ M . Now, according to [6, in section 4] we
have
R[[M, ω, ≤]] = {f ∈ R[[M, ω]]
⇓ (supp(f ), ≤)
is Artinian}.
(3.2)
If ⇓ (M, ≤) is Artinian, R[[M, ω, ≤]] = R[[M, ω]]. For instance, ⇓ (F, 4) and ⇓ (Fn, 4) are Ar-
tinian, and so R[[F, ω, 4]] = R[[F, ω]] and R[[Fn, ω, 4]] = R[[Fn, ω]] such that F be a free monoid.
Now we give a generalization of a result [6, Theorem 4.3] of G. Brookfield:
Theorem 3.2. Let R be a ring, (M, ≤) a positive strictly ordered monoid and ωm an automor-
phism of R with ωmωn = ωnωm for each m, n ∈ M . Then R[[M, ω]] is left Noetherian if and
only if R is left Noetherian and M is finitely generated.
Proof. ⇐) In the first place, we claim that if ϕ : (N, ≤) → (M, ≤) is a surjective strict monoid
homomorphism, induces a surjective ring homomorphism ϕ∗ : R[[N, ω, ≤]] → R[[M, ω, ≤]]. Since
ϕ is strict, ϕ−1(x) is antichain in (N, ≤) for all x ∈ M . Thus, if f ∈ R[[N, ω, ≤]] then ϕ−1(x) ∩
supp(f ) is finite and we can define ϕ∗(f ) = f ∗, where f ∗(x) =Px′∈ϕ−1(x) f (x′) for x ∈ M . We
show that ϕ∗ is a ring homomorphism. One can see that
(f g)∗(m) = Xm′∈ϕ−1(m)
(f g)(m′) = Xxy=m Xx′y′=m′
m′∈ϕ−1(m)
(cid:18)f (x′)αx′(g(y′))(cid:19).
(3.3)
On the other hand
(f ∗g∗)(m) =(cid:0)ϕ∗(f )ϕ∗(g)(cid:1)(m) = Xxy=m(cid:18)ϕ∗(f (x))αx(cid:0)ϕ∗(g(y)(cid:1)(cid:19)
f (x′)(cid:19)αx(cid:18) Xy′∈ϕ−1(y)
g(y′))(cid:19)
= Xxy=m(cid:18) Xx′∈ϕ−1(x)
= Xxy=m Xx′∈ϕ−1(x) Xy′∈ϕ−1(y)(cid:18)f (x′)αx′(g(y′))(cid:19).
7
Since ϕ−1 is a homomorphism, ϕ−1(x)ϕ−1(y) = ϕ−1(xy) and so ϕ−1(m) = ϕ−1(x)ϕ−1(y). So
(f ∗g∗)(m) = Xxy=m Xm′=x′y′
m′∈ϕ−1(m)
(cid:18)f (x′)αx′(g(y′))(cid:19).
(3.4)
By equations 3.3 and 3.4 we see that (f g)∗(m) = (f ∗g∗)(m). We have also
(f + g)∗(x) = Xx′∈ϕ−1(x)
= Xx′∈ϕ−1(x)
(f + g)(x′) = Xx′∈ϕ−1(x)
f (x′) + Xx′∈ϕ−1(x)
(f (x′) + g(x′))
g(x′) = f ∗(x) + g∗(x).
Thus ϕ∗ : R[[N, ω, ≤]] → R[[M, ω, ≤]] is a ring homomorphism. Now, we show that ϕ∗ is
surjective. Suppose that f ∈ R[[M, ω, ≤]], where {f (n)}n∈M are the coefficients of f in R. For
every n ∈ M , the set ϕ−1(n) is nonempty and finite, say ϕ−1(n) = {m1, m2, . . . , mk}, where k
and all the mi depends on n. We define the function g ∈ R[[N, ω, ≤]] as follows
g(mj ) =(f (n)
0
; j = 1
; otherwise.
(3.5)
Notice that g is independent of n, since if n 6= n′, then ϕ−1(n) ∩ ϕ−1(n′) = ∅. Also, for each
n ∈ M we have
ϕ∗(g)(n) = Xm∈ϕ−1(n)
g(m) =
k
Xj=1
g(mj ) = g(m1) = f (n).
This means that ϕ∗(g) = f , and hence ϕ∗ is surjective. So we proved the claim. It is well-known
that there is an strict monoid surjection ϕ : (Fn, 4) → (M, 4) for some n ∈ N. Also, the identity
map (M, 4) → (M, ≤) is a surjection. So the composition of these two maps is a surjection
and by [6, Lemma 2.1]. Hence R[[M, w, ≤]] is a homomorphic image of the ring R[[Fn, ω, (cid:22)]].
Since R[[Fn, ω, (cid:22)]] = R[[Fn, ω]] and R[[Fn, ω]] is Noetherian, its projection R[[M, w, ≤]] is also
Noetherian. Moreover, we show that R[[M, ω, ≤]] = R[[M, ω]]. If R[[M, ω, ≤]] is left Noetherian,
then ⇓ (M, 4) is Artinian. By applying [6, Lemma 2.1(2)] to the identity map (M, 4) → (M, ≤),
one can see that ⇓ (M, ≤) is Artinian. Thus R[[M, ω, ≤]] = R[[M, ω]].
⇒) The method of this part is inspired from [6, Theorem 4.3]. The trivial case of M is
obvious. By [6, Lemmas 3.1 and 3.2], M is strict and 4 is a partial order on M .
Suppose T = R[[M, ω, ≤]] is left Noetherian. One can see that M is finitely generated similar
to the proof of [6, Theorem 4.3]. Hence we have to prove that R is Noetherian similar to the
proof of ([25, Theorem 5.2(i)], [26, Theorem 3.1(i)]). Let IT = {f ∈ T ωx(f (y)) ∈ I; x, y ∈ M }.
It is easy to see that IT is a left ideal of T . So for each ideal I of R, there is a correspondent
ideal in T . Also if I ⊂ J ,then IT ⊂ JT . Hence if there exists a nonstabilized ascending chain
in R, then there is one in T . But this is impossible, so R is left Noetherian.
In Theorem 3.2 if we set σ the identity homomorphism then we have:
Corollary 3.3. [6, Theorem 4.3] Let R be a ring and (M, ≤) a positive strictly ordered monoid.
Then R[[M, ≤]] is left Noetherian if and only if R is left Noetherian and M is finitely generated.
8
Finally, we conclude the following result which connects the results of previous sections.
Corollary 3.4. Let R be an S-Noetherian von Neumman regular ring and S a denominator
set. Assume that (M, ≤) is a finitely generated positive strictly ordered monoid and ωm an
automorphism of R with ωmωn = ωnωm for each m, n ∈ M . Then (S−1R)[[M, ω]] is a left
Noetherian ring.
Proof. The ring S−1R is Noetherian by Theorem 2.2. Since (M, ≤) is a positive strictly ordered
monoid and ωm is an automorphism for all m ∈ M , (S−1R)[[M, ω]] is a Noetherian ring by
Theorem 3.2.
4 S-Noetherian property of generalized skew power series rings
Recall that a ring is called right duo (resp., left duo) if all of its right (resp., left) ideals are
two-sided. Also, a right and left duo ring is called a duo ring. We know that if a ring is duo,
then every prime ideal is completely prime. It is known that a power series ring over a duo ring
need not be duo (on either side).
Lemma 4.1. Let R be a duo ring and S ⊂ R a denominator set. If s ∈ S, r ∈ R then there
exists s1 ∈ S such that srs1 = rss1.
Proof. Let s ∈ S and r ∈ R. Since R is duo, there exist s′ ∈ S such that sr = rs′, so
1
s . sr
s = 0S−1R. So
r(s − s′)s1 = 0R. So rss1 = rs′s1 and since rs′ = sr we have srs1 = rss1.
s ) = 0, which means that r(s−s′)
1 . Hence r
s . Thus r
1 (1 − s′
1 = rs′
s = r
1 . s′
1 = 1
s . rs′
In the previous result, it is easy to see that if s ∈ S, r ∈ R, then there exists s1 ∈ S such
that s1sr = s1rs. We will use this point in the proposition below.
Proposition 4.2. Let R be a duo ring, S ⊆ R a denominator set and M an S-finite R-module.
Then M is S-Noetherian if and only if P M is an S-finite submodule, for each S-disjoint prime
ideal P of R.
Proof. The "only if" part is clear. For the converse, assume that P M is S-finite for each P
prime ideal of R with P ∩ S = ∅. Since M is S-finite, wM ⊆ F for some w ∈ S and some finitely
generated submodule F . If M is not S-Noetherian, the set F of all non-S-finite submodules
of M is not empty. So F has a maximal element like N by Zorn's lemma. We claim that
P = [N : M ] := {r ∈ R rM ⊆ N } is a prime ideal of R and is disjoint from S. Suppose to the
contrary that P ∩ S 6= ∅ and s ∈ P ∩ S. Then we have
swN ⊆ swM ⊆ sF ⊆ sM ⊆ N.
So swN ⊆ sF ⊆ N and N becomes S-finite. This contradiction shows that P ∩ S = ∅. Now
suppose that P is not a prime ideal of R. So P is not completely prime. So there exist a, b ∈ R\P
and ab ∈ P . So N + aM is S-finite, hence s(N + aM ) ⊆ (R(n1 + am1) + · · · + R(np + amp)) for
some s ∈ S, ni ∈ N and mi ∈ M . Also [N : a] is S-finite. So t[N : a] ⊆ (Rq1+Rq2+· · ·+Rqk) for
some t ∈ S and qj ∈ [N : a]. Since R is duo and S is a denominator set in R, there exists s′′ ∈ S
such that s′′at = s′′ta by Theorem 4.1. Also s(N + aM ) ⊆ (R(n1 + am1) + · · · + R(np + amp)).
i ∈ R. Since
imi for some r′
Thus sx = P rini + riami. This means that sx = P rini + aP r′
sx,P rini ∈ N , we have P r′
s′′tsx = s′′tX rini + s′′tX ar′
imi =X s′′trini + s′′atX r′
imi ∈ [N : a]. So
imi =X s′′trini + s′′aX cjqj.
9
So s′′tsx =P s′′trini +P c′s′′
j ∈ R. Hence s′′tsx ∈ (Rn1 + · · · + Rnp + Rs′′aq1 +
· · · + Rs′′aqk). So s′′tsN ⊆ (Rn1 + · · · + Rnp + Rs′′aq1 + · · · + Rs′′aqk) ⊆ N . Thus N is
S-finite and this contradicts to the fact that N is maximal in F. Therefore P is a prime ideal
of R. Moreover P = [N : M ] ⊆ [N : F ] ⊆ [N : wM ] = [P : w] = P . Hence [N : F ] = P . Let
j aqj for some c′
F = (Rf1+Rf2+· · ·+Rfk). Since R is a duo ring, P = [N :P Rfi] =T[N : fi]. So P = [N : fi]
for some fi ∈ {f1, f2, · · · , fk}. One can show that tN ⊆ (Rn1 + Rn2 + · · · + Rnl) + P M for some
t ∈ S and ni ∈ N as above or in similar way as that employed in [1, Proposition 4]. Since P M
is S-finite, vP M ⊆ G ⊆ P M ⊆ N for some v ∈ S and a finitely generated submodule G of M .
So
vtN ⊆ v(Rn1 + Rn2 · · · + Rnl) + vP M ⊆ (Rn′
1 + Rn′
2 · · · + Rn′
l) + G ⊆ N
for some n′
i ∈ N . So N becomes S-finite which is a contradiction. So M is S-Noetherian.
Lemma 4.3. Let R be a ring with an endomorphism σ. If R[[x; σ]] is a duo ring, then σ is
surjective.
Proof. Suppose that a ∈ R. Since R[[x; σ]] is a duo ring we have ax = xf such that f =
i=0 σ(fi)xi+1. Now, since ax = xf , σ(fi) = 0 for all i 6= 0
and σ(f0) = a. Thus, for each a ∈ R there exists f0 ∈ R such that a = σ(f0).
P∞
i=0 fixi. So xf = xP∞
i=0 fixi = P∞
Theorem 4.4. Let R be a ring, S ⊆ R a σ-anti-Archimedean denominator set (consisting
nonzero devisors) and σ1, · · · , σn are monomorphisms of R with σiσj = σjσi, for each i, j.
Assume that R[[X1, · · · , Xn; σ1, · · · , σn]] is a duo ring.
If R is S-Noetherian, then the ring
R[[X1, · · · , Xn; σ1, · · · , σn]] is also S-Noetherian.
Proof. We use the method in [1, Proposition 10] employed by Anderson and Dumitrescu. As S
is σ-anti-Archimedean in every ring containing R as a subring, we shall prove the case n = 1,
so we assume that T = R[[x; σ]] is duo and σ is an automorphism of R.
It is enough to
prove that every prime ideal P of T is S-finite. Let π : T → R the R-algebra homomorphism
sending x to zero and P ′ = π(P ). Since R is S-Noetherian, there exists s ∈ S such that
sP ′ ⊆ (Rg1(0)+Rg2(0)+· · ·+Rgk(0)) for some gi ∈ P . If x ∈ P , then P = (T P ′+T x). If gi(x) =
P aixi, then gi(x) =P xiσ−i(ai) ∈ (T P ′ + T x). So sP ⊆ (T P ′ + T x) = (T g1 + · · · + T gk) ⊆ P .
This means that P is S-finite. Let x /∈ P and f ∈ P . So sf (0) =P d0,jgj(0) for some d0,j ∈ R.
So xf1 = sf −P d0,jgj ∈ P for some f1 ∈ T . Considering x /∈ P , f1 ∈ P . So sf1 =P d1,jgj +xf2
for some f2 ∈ T . Hence σ(s)sf = P σ(s)d0,jgj + xP d1,jgj + x2f2. Also f2 ∈ P , since x /∈ P
and sf1 −P d1,jgj ∈ P . In this way, one can see that for each L ≥ 0,
L
L
L
k
Yl=0
(cid:0)
σl(s)(cid:1)f =
Xi=0
xi
Xj=1
(
Yl=i+1
σl(s))di,jgj + xL+1fL+1.
Since S ∩(cid:0)Tl≥1,ij∈N∪{0} σi1(s) · · · σil(s)R(cid:1) 6= ∅, there exists t ∈ R such that
for each ij ∈ N ∪ {0}, k ∈ N. Moreover
t
σi1 (s)···σik (s)
∈ R
tf =Xj Xi (cid:16) tsσ−i(dij)
Ql σl(s) (cid:17)xigj.
So tf = Pj hjgj where hj = Pi
tsσ−i(di,j )
Ql σl(s) xi. So tf ∈ (T g1 + T g2 + · · · + T gk). Hence tP ⊆
(T g1 + T g2 + · · · + T gk). Since gi ∈ P , (T g1 + T g2 + · · · + T gk) ⊆ P . Thus R[[x; σ]] is an
S-Noetherian ring.
10
The following proposition which is proved in [1], is the corollary of the above theorem.
Corollary 4.5. [1, Proposition 10] Let R be a commutative ring and S ⊆ R an anti-Archimedean
multiplicative set of R. If R is S-Noetherian, then so is R[[X1, · · · , Xn]].
A ring R is called strongly regular if every principal right (or left) ideal is generated by a
central idempotent. A ring is said to be left self injective if it is injective as a left module over
itself. Hirano in [12, Theorem 4] shows that if R is a self-injective strongly regular ring, then
R[[x]] is a duo ring.
We have the following generalization of a theorem of D.D. Anderson and Tiberiu Dumitrescu
[1, Proposition 10].
Theorem 4.6. Let R be a duo ring with an automorphism σ and S ⊆ R a σ-anti-Archimedean
denominator set (consisting nonzero devisors). If R is S-Noetherian, then so is the skew power
series ring R[[x; σ]].
Proof. We can prove this theorem in a similar way as in Theorem 4.4. Consider the notations in
the proof of Theorem 4.4. Let x ∈ P . Since σ is bijective, P is S-finite. Let x /∈ P and f ∈ P ,
so xf1 = sf −P d0igi ∈ P . Note that for each h ∈ R[[x; σ]] and I is a left ideal of R[[x; σ]],
xh ∈ I yields that xR[[x; σ]]h ∈ I. So f1 ∈ P . The rest of the proof is similar to what we did
in Theorem 4.4.
The following corollary is a generalization of the case n = 1 in [1, Proposition 10] for the
category of duo rings.
Corollary 4.7. Let R be a duo ring and S ⊆ R an anti-Archimedean denominator set (con-
sisting nonzero devisors) of R. If R is S-Noetherian, then so is the power series ring R[[x]].
Now we extend the last result for the skew generalized power series ring R[[M, ω]].
Theorem 4.8. Let R be a duo ring, (M, ≤) a positive strictly ordered commutative monoid
and ωm a monomorphism of R with ωmωn = ωnωm for each m, n ∈ M . Assume that S ⊂ R
is an ωm-anti-Archimedean denominator set (consisting nonzero devisors) of R and R[[M, ω]]
be a duo ring. Then R[[M, ω]] is left (or right) S-Noetherian if and only if R is left (or right)
S-Noetherian and M is finitely generated.
Proof. (⇐) We use the method of G. Brookfeild employed in [6]. We know that the surjective
homomorphism ϕ : Fn −→ M (where F is a free monoid) induces a projection
ϕ∗ : R[[Fn, (ω, (cid:22))]] −→ R[[M, (ω, ≤)]]
and R[[M, (ω, ≤)]] = R[[M, ω]] by [6, Theorem 4.3]. Moreover, since R[[Fn, (ω, (cid:22))]] is S-
Noetherian, so is R[[M, ω]] by [16, Lemma 2.2] for noncommutative version.
(⇒) Let A := R[[M, ω]] be S-Noetherian. Let {mnn ∈ N} be an infinite sequence in M . Let
I = (Aem1 + Aem2 + · · · ). Since A is S-Noetherian, there exists s ∈ S such that csI ⊆ J ⊆ I
) for some k ∈ N. So
for J finitely generated ideal of A. So csI ⊆ (Aemi1
)(m) =
+ · · · + Aemik
+ Aemi2
) for each m ∈ M , (ftemit
t=0 supp(ftemit
cseml =Pk
Pm′m′′=m ft(m′)ωm′(emit
t=0 ftemit for some l 6= it. So ml ∈Sk
(m′′)). So ml ∈Sk
m1 ∈ M such that m1mit = m for some 0 ≤ t ≤ L. So
t=0(cid:8) supp(ft) + supp(ωm′(emit
(m′′)))(cid:9). There exists
(ftemit
)(m) = ft(m1)ωm1(emit
(mit)) = ft(m1).
11
Thus m1 ∈ supp(ft) and m1mit ∈ supp(ftemit
) for some 0 ≤ t ≤ L. So for each m ∈
), mit (cid:22) ml for
supp(ωm′(emit
some 0 ≤ t ≤ L. Since M is positive strictly ordered monoid, M is finitely generated by [6,
Lemma 3.3].
(m′))), mit (cid:22) m for some 0 ≤ t ≤ L. Since ml ∈ supp(ftemit
Let I be an ideal of R, so AI is an ideal of A. So there exists s ∈ S such that csAI ⊆ J ⊆ AI
for some J finitely generated ideal of A. Set
T = {f (π(f ))f ∈ csAI}.
We claim that T = sI. Let t ∈ T , so t = h(π(h)) and h = csg for some g ∈ AI. So t = sg(π(sg)).
This means that t ∈ sI considering the fact that
I = {f (π(f ))f ∈ AI}.
So T ⊆ sI. Now let i ∈ I, so i ∈ AI. Since si(m) = 0 for m 6= 1, si ∈ T . Thus sI ⊆ T . Hence
sI = T . But sI = T ⊆ J ′ ⊆ I where J ′ = {f (π(f ))f ∈ J}. Let J = (Aj1 + Aj2 + · · · + Ajp).
So it is easy to show that J ′ = (Rj1(π(j1)) + Rj2(π(j2)) + · · · + Rjp(π(jp))). So J ′ is finitely
generated in R. Hence I is S-finite and R is left S-Noetherian.
Recall from [5], that a ring R is right (left) ℵ0-injective provided any homomorphism from a
countably generated right (left) ideal of R into R extends to a right (left) R-module endomor-
phism of R. By an ℵ0-injective ring we mean a right and left ℵ0-injective ring.
Corollary 4.9. Let R be an strongly regular and an ℵ0-injective ring with automorphisms σ1, σ2
such that σ1σ2 = σ2σ1. Assume that S ⊂ R is an σ1, σ2 anti-Archimedean denominator set
(consisting nonzero devisors). If R is left (or right) S-Noetherian, then so is R[[x, y; σ1, σ2]].
Proof. Assume that R is an strongly regular and ℵ0-injective ring. Then by [20], A = R[[x; σ1]]
is duo ring and S-left Noetherian ring. Then A[[y; σ2]] is left S-Noetherian ring.
The following corollary is a generalization of the case n = 2 of in [1, Proposition 10] for the
category of duo rings.
Corollary 4.10. Let R be an strongly regular self-injective ring and S ⊆ R an anti-Archimedean
denominator set (consisting nonzero devisors) of R. If R is left (or right) S-Noetherian, then
so is R[[x, y]].
Corollary 4.11. Let R be a duo ring, S ⊆ R an anti-Archimedean denominator set (consisting
nonzero devisors) of R. Assume that R[[M ]] is a duo ring. Then R[[M ]] is left (or right)
S-Noetherian if and only if R is left (or right) S-Noetherian and M is finite generated.
References
[1] D. D. Anderson, and T. Dumitrescu, S-Noetherian rings, Comm. Algebra, 30 (2002), 4407-
4416.
[2] D. D. Anderson, B. G. Kang, and M. H. Park, Anti-archimedean rings and power series
rings, Comm. Algebra, 26 (1998), 3223-3238.
[3] D. D. Anderson, D. J. Kwak, M. Zafrullah, Agreeable domains, Comm. Algebra (1995)
23:4861-4883.
12
[4] M. Atiyah, M. Francis and I. G. Macdonald, Introduction to commutative algebra, Reading:
Addison-Wesley, 2 (1969).
[5] J. W. Brewer, E. A. Rutter, and J. J. Watkins, Coherence and weak global dimension of
R[[X]] when R is von Neumann regular, J. Algebra 46(1) (1977), 278-289.
[6] G. Brookfield, Noetherian Generalized Power Series Rings, Comm. Algebra, 32:3 (2004),
919-926.
[7] G. A. Elliott, P. Ribenboim, Fields of generalized power series, Arch. Math. 54 (1990),
365-371.
[8] R. Gilmer, Commutative Semigroup Rings, The University of Chicago Press (1984).
[9] K. R. Goodearl, R. B. Warfield, An Introduction to Noncommutative Noetherian Rings,
Cambridge University Press (2004).
[10] E. Hamann, E. Houston, and J. Johnson, Properties of uppers to zero in R [X], Pacific
Journal of Mathematics 135, no. 1 (1988): 65-79.
[11] G. Higman, Ordering by divisibility in abstract algebras, Proc. London Math. Soc. (3) 2
(1952) 326-336.
[12] Y. Hirano, C. Y. Hong, J. Y. Kim, and J. K. Park, On strongly bounded rings and duo
rings, Communications in Algebra 23, no. 6 (1995), 2199-2214.
[13] T. Y. Lam, Lectures on Modules and Rings, Grad. Texts in Math., vol. 139, Springer, New
York, (1998).
[14] J. W. Lim, and O. D. Yeol, S-Noetherian properties of composite ring extensions, Comm.
Algebra, 43 (2015), no. 7, 2820-2829.
[15] J. W. Lim, O. D. Yeol, S-Noetherian properties on amalgamated algebras along an ideal,
J. Pure Appl. Algebra 218 (2014), no. 6, 1075-1080.
[16] Z. Liu, On S-Noetherian rings, Arch. Math.(Brno), 43 (2007), 55-60.
[17] G. Marks, A taxonomy of 2-primal rings, Journal of Algebra, 266, no. 2 (2003), 494-520.
[18] G. Marks, R. Mazurek, M. Ziembowski, A Unified Approach to Various Generalizations of
Armendariz Rings, Bull. Aust. Math. Soc., 81 (2010), 361-397.
[19] R. Mazurek, M. Ziembowski, On von Neumann regular rings of skew generalized power
series, Comm. Algebra, 36 (2008), 1855-1868.
[20] R. Mazurek, M. Ziembowski, Duo, Bezout, and Distributive Rings of Skew Power Series.
Publicacions Matem tiques 53.2 (2009), 257-271.
[21] R. Mazurek and M. Ziembowski, The ascending chain condition for principal left or right
ideals of skew generalized power series rings, J. Algebra, 322 (2009), 983-994.
[22] P. Ribenboim, Noetherian rings of generalized power series, J. Pure Appl. Algebra 79(3),
(1992), 293-312.
13
[23] P. Ribenboim, Some examples of valued fields, J. Algebra, 173 (1995b), 668-678.
[24] P. Ribenboim, Semisimple rings and von Neumann regular rings of generalizedpower series,
J. Algebra, 198 (1997), 327-338.
[25] P. Ribenboim, Rings of generalized power series. II. Units and zero-divisors, J. Algebra,
168(1) (1994), 71-89.
[26] K. Varadarajan, Generalized power series modules, Comm. Algebra, 29(3) (2001b), 1281-
1294.
14
|
1605.09069 | 2 | 1605 | 2016-12-15T11:06:43 | On the center-valued Atiyah conjecture for L2-Betti numbers | [
"math.RA",
"math.KT",
"math.OA"
] | The so-called Atiyah conjecture states that the von Neumann dimensions of the L2-homology modules of free G-CW-complexes belong to a certain set of rational numbers, depending on the finite subgroups of G. In this article we extend this conjecture to a statement for the center-valued dimensions. We show that the conjecture is equivalent to a precise description of the tructure as a semisimple Artinian ring of the division closure D(QG) of Q[G] in the ring of affiliated operators. We prove the conjecture for all groups in Linnell's class C, containing in particular free-by-elementary amenable groups.
The center-valued Atiyah conjecture states that the center-valued L2-Betti numbers of finite free G-CW-complexes are contained in a certain discrete subset of the center of C[G], the one generated as an additive group by the center-valued traces of all projections in C[H], where H runs through the finite subgroups of G.
Finally, we use the approximation theorem of Knebusch for the center-valued $L^2$-Betti numbers to extend the result to many groups which are residually in C, in particular for finite extensions of products of free groups and of pure braid groups. | math.RA | math |
On the center-valued Atiyah conjecture for
L2-Betti numbers
Anselm Knebusch∗
HFT-Stutgart
University of Applied Sciences
Germany
Peter Linnell†
Virginia Tech
Blacksburg
USA
Thomas Schick‡
Mathematisches Institut
Georg-August-Universitat Gottingen
Germany
September 24, 2018
Abstract
The so-called Atiyah conjecture states that the N (G)-dimensions
of the L2-homology modules of finite free G-CW-complexes belong to
a certain set of rational numbers, depending on the finite subgroups
of G. In this article we extend this conjecture to a statement for the
center-valued dimensions. We show that the conjecture is equivalent to
a precise description of the structure as a semisimple Artinian ring of
the division closure D(Q[G]) of Q[G] in the ring of affiliated operators.
∗e-mail: [email protected]
†e-mail: [email protected]
www: http://www.math.vt.edu/people/plinnell/
‡e-mail: [email protected]
www: http://www.uni-math.gwdg.de/schick
partially funded by the Courant Research Center "Higher order structures in Mathematics"
within the German initiative of excellence
1
On the center-valued Atiyah conjecture
2
We prove the conjecture for all groups in Linnell's class C, containing
in particular free-by-elementary amenable groups.
The center-valued Atiyah conjecture states that the center-valued
L2-Betti numbers of finite free G-CW-complexes are contained in a
certain discrete subset of the center of C[G], the one generated as an
additive group by the center-valued traces of all projections in C[H],
where H runs through the finite subgroups of G.
Finally, we use the approximation theorem of Knebusch [15] for the
center-valued L2-Betti numbers to extend the result to many groups
which are residually in C, in particular for finite extensions of products
of free groups and of pure braid groups.
1
Introduction
In [3], Atiyah introduced L2-Betti numbers for manifolds with cocompact
free G-action for a discrete group G (later generalized to finite free G-CW-
complexes). There, he asked [3, p. 72] about the possible values these can
assume. This question was later popularized in precise form as the so-called
"strong Atiyah conjecture". One easily sees that the possible values depend
on G. For a finite subgroup of order n in G, a free cocompact G-manifold with
L2-Betti number 1/n can be constructed. For certain groups G which contain
finite subgroups of arbitrarily large order, with quite some effort manifolds
M with π1(M) = G and with transcendental L2-Betti numbers have been
constructed [4, 12, 26]. In the following, we will therefore concentrate on G
with a bound on the orders of finite subgroups.
The L2-Betti numbers are defined using the L2-chain complex. The chain
groups there are of the form l2(G)d, and the differentials are given by convo-
lution multiplication with a matrix over Z[G]. The strong Atiyah conjecture
for free finite G-CW-complexes is equivalent to the following (with K = Z):
1.1 Definition. Let G be a group with a bound on the orders of finite sub-
groups and let lcm(G) ∈ N (the positive integers) denote the least common
multiple of these orders. Let K ⊂ C be a subring.
We say that G satisfies the strong Atiyah conjecture over K, or K[G] satisfies
the strong Atiyah conjecture if for every n ∈ N and every A ∈ Mn(K[G])
dimG(ker(A)) := trG(prker A) ∈
1
lcm(G)
Z.
On the center-valued Atiyah conjecture
3
Here, as before, we consider A : l2(G)n → l2(G)n as a bounded operator,
acting by left convolution multiplication -- the continuous extension of the
left multiplication action on the group ring to l2(G). trG is the canonical trace
on Mn(N (G)), i.e. the extension (using the matrix trace) of trG : N (G) → C;
a 7→ haδe, δeil2(G), where N (G), the weak closure of C[G] ⊂ B(ℓ2(G)) is the
group von Neumann algebra.
If G contains arbitrarily large finite subgroups, we set lcm(G) := +∞.
A projection P will always be a self adjoint idempotent, so P = P 2 = P ∗,
where ∗ indicates the involution on N (G).
If E is an idempotent, then
E is similar to a projection P and then trG(E) = trG(P ). Also a central
idempotent is always a projection. Note that if G is an infinite group, then
the set {trG(P )}, where P runs through the projectors in Mn(N (G)), n ∈
N consists of all non-negative real numbers. The strong Atiyah conjecture
predicts, on the other hand, that the L2-Betti numbers take values in the
subgroup of R generated by traces of projectors defined already over Q[H]
for the finite subgroups H of G: the projector pH = (Ph∈H h)/ H satisfies
trG(pH) = 1/ H. And by the Chinese remainder theorem, the additive
subgroup of R generated by the H−1 is exactly
We now turn to the center-valued refinements of the above statements. The
center-valued L2-Betti numbers are obtained by replacing the canonical (com-
plex-valued) trace trG by the center-valued trace tru
G (see Definition 2.1), tak-
ing values in the center of N (G). Note that by general theory [14, Chapter 8],
as every finite von Neumann algebra has a unique normalized center-valued
trace, this is a powerful invariant: two finitely generated projective Hilbert
N (G)-modules are isomorphic if and only if their center-valued dimensions
coincide. The center of a ring R will be denoted Z(R).
lcm(G) Z.
1
1.2 Definition. Let G be a group with lcm(G) < ∞, let K be a subring of C,
let F be the field of fractions of K, and assume that F is closed under complex
conjugation. Let LK(G) be the additive subgroup of Z(N (G)) generated by
tru
G(P ) ∈ Z(C[G]) ⊂ Z(N (G)) where P runs through projections P ∈ F [H]
with H ≤ G a finite subgroup.
We say that G satisfies the center-valued Atiyah conjecture over K, or K[G]
satisfies the center-valued conjecture if for every n ∈ N and every A ∈
Mn(K[G]) we have dimu
G(prker A) ∈ LK(G).
G(ker(A)) := tru
Observe that G satisfies the center-valued Atiyah conjecture over K if and
only if G satisfies the center-valued conjecture over its field of fractions F .
On the center-valued Atiyah conjecture
4
Indeed the "only if" is obvious. On the other hand if A ∈ Mn(F [G]), then
("clearing denominators") there exists 0 6= k ∈ K such that kA ∈ Mn(K[G]),
and ker A = ker kA, which verifies the "if" part.
1.3 Proposition. If a group G satisfies the center-vlaued Atiyah conjecture
over K of Definition 1.2, then G also satisfies the (classical) strong Atiyah
conjecture over K of Definition 1.1.
G. We therefore only have to check that trG(a) ∈
Proof. By the universal property of the center-valued trace [14, Chapter 8],
trG = trG ◦ tru
lcm(G) Z for
all a ∈ LK(G). By the definition of LK(G), we just have to show that
trG(P ) ∈ 1
lcm(G) Z for each projector P ∈ F [H], where H ≤ G is an arbitrary
finite subgroup. This is of course well known to be true, it follows e.g. from
the fact that finite groups satisfy the strong Atiyah conjecture over K.
1
1.4 Proposition (compare Corollary 3.7). If lcm(G) < ∞ then LK(G) ⊂
Z(N (G)) is discrete. In particular, the center-valued Atiyah conjecture pre-
dicts a "quantization" of the center-valued L2-Betti numbers.
1.5 Remark. As for the ordinary strong Atiyah conjecture, the center-valued
Atiyah conjecture over Z[G] is equivalent to the statement that the center-
valued L2-Betti numbers for finite free G-CW-complexes take values in LZ(G).
The center-valued L2-Betti numbers have been introduced and used in [21].
The strong Atiyah conjecture has many applications. Most interesting are
those for a torsion-free group G, i.e. if lcm(G) = 1. This is exemplified by
the following surprising result of Linnell [17]. We first recall the notion of
"the" division closure of K[G].
1.6 Definition. Let G be a discrete group and let K ⊂ C be a subring. Let
U(G) denote the ring of unbounded operators on l2(G) affiliated to N (G)
(algebraically, U(G) is the Ore localization of N (G) at the set of all non-
zero-divisors).
Define the division closure D(K[G]) to be the smallest subring of U(G) con-
taining K[G] which is closed under taking inverses in U(G).
1.7 Theorem. Let G be a discrete group with lcm(G) = 1 and let K be a
subring of C. Then K[G] satisfies the strong Atiyah conjecture if and only if
D(K[G]) is a skew field.
On the center-valued Atiyah conjecture
5
The appealing feature of this theorem is that it provides a canonical over-ring,
namely D(K[G]) of K[G] which should be a skew field, provided G is torsion
free. Observe that this implies in particular that K[G] has no non-trivial
zero-divisors. For more information on this, see [23, Remark 4.11].
Part of the motivation for the work at hand was the question of how to
generalize Theorem 1.7 if lcm(G) > 1. It turns out that one expects that
D(K[G]) is semisimple Artinian. In the situation at hand this means that
D(K[G]) is a finite direct sum of matrix rings over skew fields. This is proved
in many cases e.g. in [17].
The present paper gives a very precise (conjectural) description of D(K[G]),
and if it is satisfied we call D(K[G]) Atiyah-expected Artinian: the lattice of
finite subgroups and their K-linear representations give a precise prediction
into which matrix summands D(K[G]) decomposes and the size of the cor-
responding matrices. The precise formula is a bit cumbersome, so we don't
give it here but refer to Definition 3.8.
One of our main theorems is the precise generalization of Theorem 1.7.
1.8 Theorem. Let G be a discrete group with lcm(G) < ∞ and let K be
a subfield of C closed under complex conjugation. Then K[G] satisfies the
center-valued Atiyah conjecture if and only if D(K[G]) is Atiyah-expected
Artinian.
Indeed, we show in Theorem 3.9 that these two properties are also equivalent
to the property that K0(D(K[G])) is generated by the images of K0(K[H])
as H runs over the finite subgroups of G.
1.9 Definition. Given a discrete group G with lcm(G) < ∞, let ∆+(G)
denote the maximal finite normal subgroup, and let ∆(G) denote the finite
conjugacy center, i.e. the set of those elements of G which have only a finite
number of conjugates.
Indeed, by [25, §1], ∆(G) is a normal subgroup of G. Recall that the product
of two normal subgroups is a normal subgroup, therefore, as lcm(G) < ∞,
∆+(G) makes sense. Note that ∆+(G) ⊂ ∆(G), indeed, using [25, Lemma
19.3] it is exactly the subset of all elements of finite order in ∆(G).
In the special case ∆+(G) = {1}, we have that D(K[G]) is Atiyah-expected
Artinian if and only if it is an lcm(G) × lcm(G)-matrix ring over a skew field,
and by Theorem 3.9 this is equivalent to the center-valued Atiyah conjecture
(which in this case is implied by the usual Atiyah conjecture, as the relevant
On the center-valued Atiyah conjecture
6
part of Z(N (G)) is C[∆+(G)]). This special case (and slightly more general
situations) have already been covered in [20], but without the use of the
center-valued trace.
It turns out that the general case requires this more
refined dimension function. However, much of our arguments for Theorem
3.9 follow closely the arguments of [20].
In [20], a variant of the division closure, namely the ring E(K[G]) is intro-
duced and used (compare Definition 2.2). It is closed under adding central
idempotents in U(G) which generated the same submodules as elements al-
ready in the ring. We expect that this actually coincides with D(K[G]).
1.10 Theorem. If lcm(G) < ∞, K is a subfield of C which is closed under
complex conjugation and G satisfies the center-valued Atiyah conjecture, then
E(K[G]) = D(K[G]).
As the second main result of the paper we establish the center-valued Atiyah
conjecture for certain classes of groups (namely almost all for which the
original Atiyah conjecture is known). The algebraic closure of Q will be
denoted Q.
1.11 Theorem. Let K be a subfield of C which is closed under complex
conjugation. The center-valued Atiyah conjecture over K is true for the
following groups G:
(1) all groups G which belong to Linnell's class of groups C of Definition
2.7, in particular all free by elementary amenable groups G.
(2) if K is contained in Q, all elementary amenable extensions of
• pure braid groups
• right-angled Artin groups
• primitive link groups
• virtually cocompact special groups, where a "cocompact special groups"
is a fundamental group of a compact special cube complex -- this
class of groups contains Gromov hyperbolic groups which act co-
compactly and properly on CAT(0) cube complexes, fundamental
groups of compact hyperbolic 3-manifolds with empty or toroidal
boundary, and Coxeter groups without a Euclidean triangle Cox-
eter subgroup,
• or of products of the above.
On the center-valued Atiyah conjecture
7
1.12 Question. Missing in the above list are congruence subgroups of SLn(Z)
and finite extensions thereof. Note that the usual Atiyah conjecture for these
groups, as long as they are torsion free, is proved in [11]. For torsion-free
groups, the center-valued Atiyah conjecture is not stronger than the usual
Atiyah conjecture. However, it would be interesting to generalize the work
of [11] to certain extensions which are not torsion free, and then (or along
the way) to deal with the center-valued Atiyah conjecture for these.
Recall that the center-valued Atiyah conjecture for a group G only makes an
assertion when lcm(G) < ∞. For the proof of (1) of Theorem 1.11 we closely
follow the method of [17], making use of the equivalent algebraic formulations
of the Atiyah conjecture of Theorem 3.9. Indeed, we show that the conjecture
is stable under extensions by torsion-free elementary amenable groups. We
actually show (and use) slightly more refined stability properties.
For (2) of Theorem 1.11 we use the approximation theorem for the center-
valued L2-Betti numbers, [15, Theorem 3.2]. Because of the discreteness of
the possible center-valued L2-Betti numbers, the Atiyah conjecture for a suit-
able sequence of quotients implies the Atiyah conjecture for the group itself.
We follow here the general idea as already applied in [29] and for more general
coefficient rings in [9]. That this idea can be used for the class of groups listed
in (2) was shown for the pure braid groups in [19], for primitive link groups in
[8] and for right-angled Coxeter and Artin groups in [18], and for cocompact
special groups by Schreve in [30] (who uses fundamentally the geometric in-
sights of Haglund-Wise [13], and develops further the methods of [18]). Agol
[1] shows in breakthrough work that Gromov hyperbolic cocompact CAT(0)
cube groups are virtually cocompact special; with Bergeron-Wise' construc-
tion of a cocompact action of a hyperbolic 3-manifold group on a CAT(0)
cube complex [5] this implies that hyperbolic 3-manifold groups are virtually
cocompact special.
2 Preliminaries on rings associated to groups
U(G), D(K[G]) and traces on these
2.1 Definition. Let G be a discrete group. The center-valued-trace is the
uniquely defined C-linear map
tru
G : N (G) → Z(N (G))
On the center-valued Atiyah conjecture
8
such that for a, b ∈ N (G) and c ∈ Z(N (G)), we have
• tru
G(ab) = tru
G(ba);
• tru
G(c) = c;
• tru
G(a) ∈ (Z(N (G)))+ if a ∈ (N (G))+.
The trace can be extended to Md(N (G)) by taking tru
abuse of notation), with trMd(C) the non-normalized trace on Md(C).
If P ∈ Md(N (G)) is a projector with image the (Hilbert N (G)-module) V ,
set dimu
G ⊗ trMd(C) (by
G := tru
G(V ) := tru
G(P ).
That a unique such trace exists is established e.g. in [14, Chapter 8].
Later, we want to apply the trace also for the division closure. Recall that
we have (by definition) the following diagram of inclusions of rings
K[G] −−−→ N (G)
y
y
D(K[G]) −−−→ U(G).
Given a finitely presented K[G]-module M, represented by A ∈ Mk×l(K[G]),
i.e. with exact sequence K[G]l A−→ K[G]k → M → 0, the induced modules
M ⊗K[G] N (G), M ⊗K[G] U(G), M ⊗K[G] D(K[G]) are also finitely presented
with the same presenting matrix A. The standard theory of Hilbert N (G)-
modules gives a center-valued dimension for each finitely presented N (G)-
module, in particular for M ⊗K[G] N (G), and dimu
G(M ⊗K[G] N (G)) = k −
dimu
G(ker(A)) in the above situation (compare [21]). In [27], this dimension
is extended to finitely presented U(G)-modules, of course in such a way that
the value is unchanged if we induce up from N (G) to U(G). More precisely,
[27] describes the extension of dimensions based on arbitrary C-valued traces
on N (G), this implies easily the corresponding extension for dimu
G.
The central idempotent division closure E(K[G])
2.2 Definition. Let R be a subring of the ring S and let C = {e ∈ S e is
a central idempotent of S and eS = rS for some r ∈ R}. Then we define
C(R, S) = X
e∈C
eR,
On the center-valued Atiyah conjecture
9
a subring of S. In the case S = U(G), we write C(R) for C(R, U(G)). For
each ordinal α, define Eα(R, S) as follows:
• E0(R, S) = R;
• Eα+1(R, S) = D(C(Eα(R, S), S), S);
• Eα(R, S) = Sβ<α Eβ(R, S) if α is a limit ordinal.
Then E(R, S) = Sα Eα(R, S). Also in the case R = K[G] where G is a group
and K is a subfield of C, we write E(K[G]) for E(K[G], U(G)).
2.3 Conjecture. Let G be a discrete group and K ⊂ C a subfield. Then
D(K[G]) = E(K[G]), at least if lcm(G) < ∞.
We cite some properties of E(K[G]) from [20] which will be useful later.
Indeed, we generalize from the canonical trace to the center-valued trace,
but the proofs literally also cover this more general situation.
2.4 Lemma. (cf. [20, Lemma 2.4]) The following additive subgroups of
Z(N (G)) coincide:
hdimu
G(xU(G)n) x ∈ Mn(K[G]), n ∈ Ni
= hdimu
G(xU(G)n) x ∈ Mn(E(K[G])), n ∈ Ni
This has as an immediate corollary that E(K[G]) = D(K[G]) if K[G] satisfies
the center-valued Atiyah conjecture:
Proof of Theorem 1.10. Let e ∈ E(K[G]) be a central idempotent of U(G).
Then all the spectral projections of e lie in Z(N (G)), therefore e is affiliated
to Z(N (G)). Being an idempotent, even e ∈ Z(N (G)). Therefore, on the
one hand, tru
G(e) =
dimu
G(e) = e while, on the other hand by Lemma 2.4, tru
G(eU(G)) ∈ LK(G), in particular e ∈ Z(K[∆+]) ⊂ K[∆+].
2.5 Remark. The proof just given didn't need the full force of the center-
valued Atiyah conjecture, only the statement that dimu
G(xU(G)n) ∈ Z(N (G))
is supported only on elements of finite order, i.e. lies in Z(K[∆+]).
On the center-valued Atiyah conjecture
10
Approximation of the center-valued trace
The following is a special case of [15, Theorem 3.2] which will be used in the
next section.
2.6 Theorem. Let G be a discrete group with a sequence G = G0 ≥ G1 ≥ · · ·
of normal subgroups with Ti∈N Gi = {1}.
Let A ∈ Md(Q[G]) and g ∈ ∆(G). Let A[i] ∈ Md(Q[G/Gi]) be the image of
A under the map induced by the projection pri : G → G/Gi.
Assume that all G/Gi satisfy the determinant bound property [9, Definition
3.1], e.g. are elementary amenable (or more generally belong to the class G
of groups introduced in [9, Definition 1.8] and corrected in the errata to [28]
at arXiv:math/9807032, or are sofic, compare [10] and [15, Theorem 4.1]).
Then
lim
i→∞
G ker(A), gil2(G).
hdimu
N (G/Gi)(ker(A[i])), pri(g)il2(G/Gi) = hdimu
Linnell's class C
2.7 Definition. Let C denote the smallest class of groups which
(1) contains all free groups,
(2) is closed under directed unions,
(3) satisfies G ∈ C whenever H ✁ G , H ∈ C and G/H is elementary
amenable.
3 Reformulation of the center-valued Atiyah
conjecture
Let G be a group with lcm(G) < ∞. We shall assume that K is a subfield
of C which is closed under complex conjugation. Many of the arguments
given below don't require this assumption; however if K is a subfield closed
under complex conjugation and e is a central idempotent in K[G], then e
is a projection [6, Lemma 9.2(i)]. Furthermore if G is a finite group and
A ∈ Mn(K[G]), then prker A ∈ Mn(K[G]) (use [6, Proposition 9.3]); it is here
where we are using the property that K is closed under complex conjugation.
Recall that ∆+ is the (finite) normal subgroup consisting of all elements of
finite order and having only finitely many conjugates.
On the center-valued Atiyah conjecture
11
3.1 Lemma. Let K ⊂ C be a subfield which contains all ∆+-th roots of 1,
and let cG denote the number of finite conjugacy classes of elements of finite
order in G, i.e. the dimension of Z(N (G)) ∩ Z(K[∆+]). There is a finite
set of primitive central projections {U 1, . . . , U cG} of Z(N (G))∩Z(K[∆+]) ⊂
Z(K[G]), given by
U i :=
X
k s.t. ∃g∈G:guig−1=uk
uk,
where ui are the primitive central idempotents of the semisimple Artinian
ring K[∆+]. Furthermore ui = ni
∆+ Ps∈G χi(s−1)s, with ni the dimensions
of the irreducible representations (over C) of ∆+ and χi the corresponding
characters (extended by 0 to all of G). Moreover the U j form an orthogonal
basis of the vector space Z(N (G)) ∩ Z(C[∆+]).
Proof. By Maschke's theorem of standard representation theory, the algebra
K[∆+] is semisimple Artinian, compare [16, XVIII, Theorem 1.2]. Therefore
it has finitely many primitive central idempotents ui.
Any algebra automorphism must permute the ui, in particular the conjuga-
tion action of G. An element of Z(K[∆+]) belongs to the center of K[G]
(and then also of N (G)) if and only if it is invariant under conjugation by
It follows immediately that the U i are the primitive cen-
elements of G.
tral idempotents of Z(N (G)) ∩ Z(K[∆+]), and furthermore they form an
orthogonal basis for Z(N (G)) ∩ Z(C[∆+]).
The formula for the ui is also standard, [16, XVIII, Proposition 4.4 and
Theorem 11.4].
3.2 Lemma. Let K be a subfield of C and let L/K be a finite Galois ex-
tension of K with Galois group F . Let G be a finite group, let {e1, . . . , en}
denote the primitive central idempotents of K[G], and let {u1, . . . , um} denote
the primitive central idempotents of L[G]. Then F acts as automorphisms
on L[G] according to the rule θ Pg∈G agg = Pg∈G θ(ag)g for θ ∈ F . The
ui form an orthogonal set and hui, 1i = hθui, 1i for all i. For each i, define
Ni = {j ∈ N eiuj = uj}. Then F acts transitively on {uj j ∈ Ni} and
ei = Pj∈Ni
Proof. This is well-known, and follows from Galois descent. Note that uiej
is a central idempotent in L[G] and ui = uiej + (1 − ej)ui. It follows for all
i, j, either uiej = 0 or uiej = ui, because ui is primitive. It follows easily that
ei = Pj∈Ni
uj. Also F acts on {uj j ∈ Ni}, and the sum of the uj in an
uj.
On the center-valued Atiyah conjecture
12
orbit is fixed by F and is therefore in K[G]. Since ei is primitive, it follows
that this orbit must be the whole of Ni. Finally if e = Pg∈G egg ∈ L[G] is
an idempotent, then e1 ∈ Q (by the character formula of Lemma 3.1) and
we see that hui, 1i = hθui, 1i for all i.
3.3 Lemma. Let K ⊂ C be a subfield, let ω be a primitive ∆+-root of 1
and set L = K(ω). Let F denote the Galois group of L over K, and let
U 1, . . . , U cL[G] be the primitive central projections of Z(N (G)) ∩ Z(L[∆+]) ⊂
Z(L[G]) as described above in Lemma 3.1. There is a finite set of primitive
central projections {P 1, . . . , P CK[G]} of Z(N (G)) ∩ Z(K[∆+]) ⊂ Z(K[G]),
given by
P i :=
X
k s.t. ∃g∈G:gpig−1=pk
pk,
where pi are the primitive central idempotents of the semisimple Artinian
ring K[∆+]. Set Ni = {j ∈ N P iU j = U j}. Then
P i = X
j∈Ni
U j
and F acts transitively on {U j j ∈ Ni}.
Proof. This follows from Lemmas 3.1 and 3.2.
3.4 Lemma. Let H be a finite subgroup of G which contains ∆+. For an
irreducible projection Q ∈ K[H] (in the sense that if Q = Q1 + Q2 with pro-
jections in Q1, Q2 ∈ K[H] satisfying Q1Q2 = 0 then either Q1 = 0 or Q2 = 0)
we have tru
G(Q) ∈ Z(N (G)) ∩ Z(K[∆+]) ⊂ Z(N (G)). More precisely, using
the central projections P i of Lemma 3.3 we have
tru
G(Q) =
dimC(Q · C[H]) · ∆+
H · dimC(P i · C[∆+])
P i =
dimN (G)(Q · l2(G))
dimN (G)(P i · l2(G))
P i
(3.5)
where P i is characterized by the property QP i = Q.
Proof. Let ω be a primitive ∆+-th root of 1, let L = K(ω) and let F denote
the Galois group of L/K. The center-valued trace is obtained by orthogonal
projection from l2(G) to the subspace of l2(∆) spanned by functions which are
constant on G-conjugacy classes, using the standard embedding of N (G) into
l2(G). For Q, which is supported on group elements of finite order, therefore
G(Q) ∈ C[∆+]. Let U 1, . . . , U cG and P 1, . . . , P CK[G] be the primitive central
tru
On the center-valued Atiyah conjecture
13
projections as described in Lemma 3.3. Using the standard inner product on
C[H] we obtain, using that (U 1, . . . , U cG) is an orthogonal basis of Z(N (G))∩
Z(L[H]) = Z(N (G)) ∩ Z(L[∆+])
G(Q) = X
tru
j
hQ, U ji
hU j, U ji
U j.
(3.6)
Moreover, we have for each j that QP j +Q(1−P j) = Q and QP jQ(1−P j) =
0, the latter because P j is central. Since Q is irreducible, we get either
QP j = Q or QP j = 0. If QP i = Q we have QPj∈Ni
U j = Q and QU j = 0
for j /∈ Ni. Also if j1, j2 ∈ Ni, then θ(QU j1) = QU j2 for some θ ∈ F
and we see that hQU j1, 1i = hQU j2, 1i, consequently hQ, U ji is independent
of j for j ∈ Ni. Similarly hU j, U ji is independent of j for j ∈ Ni. Thus
hQ, P ii = NihQ, U ji, hP i, P ii = NihU j, U ji for j ∈ Ni, hence
hQ, U ji
hU j, U ji
=
hQ, P ii
hP i, P ii
.
Substitute this in equation (3.6) together with
hQ, P ji = hQP j, 1i = hQ, 1i =
dimC(Q · C[H])
H
hP j, P ji = hP j, 1i =
dimC(P j · C[∆+])
∆+
.
These formulas follow from the character formula for projections or are di-
rectly obtained as follows: for a projection P ∈ C[E] and a finite group E
we have hP, 1il2(E) = hP h, hil2(E) for all h ∈ E, therefore dimC(P · C[E]) =
tr(P ) = Ph∈EhP h, hi = E · hP, 1i.
Note, finally, that dimC(Q·C[H])
the induction rule for von Neumann dimensions.
= dimN (H)(Q · l2(H)) = dimN (G)(Q · l2(G)) by
H
3.7 Corollary. The additive subgroup LK(G) of Z(N (G)) of Definition 1.2
is discrete.
Proof. Recall that F denotes the relevant subfield of C in the setup of Defi-
nition 1.2, namely F is the field of fractions of K. Given a finite subgroup H
of G and a projection P ∈ F [H], tru
G(P ) is a positive integral linear combina-
tion of tru
G(Qα) where Qα ∈ F [H] are irreducible projections, corresponding
to the decomposition of im(P ) into irreducible F [H]-modules.
On the center-valued Atiyah conjecture
14
It therefore suffices to check that the additive subgroup of Z(N (G)) gen-
erated by tru
G(Q) is discrete, where Q runs through the irreducible projec-
tions in F [H] and H runs through the finite subgroups of G.
Increasing
the field and increasing the finite subgroup has the only potential effect that
an irreducible projection breaks up as a sum of new irreducible projections
and therefore the subgroup generated by their center-valued traces increases.
Therefore we may assume that these subgroups contain ∆+ and that F = C.
By Lemma 3.4, these are all integer multiples of lcm(G)−1P i with the orthog-
onal basis (P 1, . . . , P cG), therefore span a discrete subgroup of Z(N (G)).
3.8 Definition. Assume that G is a discrete group with lcm(G) < ∞ and
that K is a subfield of C which is closed under complex conjugation.
We say that D(K[G]) is Atiyah-expected Artinian if it is a semisimple Ar-
tinian ring such that its primitive central idempotents are the central idem-
potents P 1, . . . , P CK[G] ∈ K[Z(K[∆+])] of Lemma 3.3, and if each direct
summand P jD(K[G])P j is an Lj × Lj matrix ring over a skew field.
Here, Lj is determined as follows: consider all irreducible sub-projections
Qα ∈ K[Hα] of P j (i.e. those satisfying QαP j = Qα), where Hα runs through
G(Qα) = qαP j for
all finite subgroups of G containing ∆+. By Lemma 3.4, tru
some rational number qα. Because there are only finitely many isomorphism
classes of finite subgroups of G, formula (3.5) shows that the collection of
these rational numbers is finite. Lj is the smallest integer such that each qα
is an integer multiple of 1
Lj
. Explicitly,
Lj =
dimC(P j · C[∆+]) lcm(G)
gcd(cid:16)dimC(P j · C[∆+]) lcm(G), dimC(Qα · C[Hα]) lcm(G)
Hα
∆+ α(cid:17)
∈ Z.
Proof. We have to show that the two descriptions of Lj coincide, using
Equation (3.5), i.e. we have to find the smallest common denominator of
all these fractions. We expand the denominators to the common value
lcm(G)·dimC(P j ·C[∆+]), then we have to divide this by the greatest common
divisor of this number and of all the new numerators.
3.9 Theorem. Let G be a discrete group, with lcm(G) < ∞ and let K ⊂ C
be a subfield closed under complex conjugation. The following statements are
equivalent.
(1) D(K[G]) is Atiyah-expected Artinian as in Definition 3.8.
On the center-valued Atiyah conjecture
15
(2) φ : LE≤G : E<∞ K0(K[E]) → K0(D(K[G])) is surjective and D(K[G])
is semisimple Artinian.
(3) φ : LE≤G ; E<∞ G0(K[E]) → G0(D(K[G])) is surjective.
(4) KG satisfies the center-valued Atiyah conjecture.
Recall here that, for a ring R, K0(R) is the Grothendieck group of finitely
generated projective R-modules, whereas G0(R) is the Grothendieck group of
arbitrary finitely generated R-modules.
Proof of Theorem 3.9. (1) =⇒ (2): We use the notation of Definition 3.8.
Using the row projectors of matrix rings, there are projections x1, . . . , xCK[G] ∈
D(K[G]) which represent a Z-basis of the free abelian group K0(D(K[G])),
and [P i] = Li[xi] in K0(D(K[G])). We have to show that each xi is an inte-
ger linear combination of images of elements of K0(K[Hα]) with Hα finite. If
Qα ∈ K[Hα] is an irreducible sub-projection of P i, then φ([Qα]) is a multi-
ple of [xi] in K0(D(K[G])), namely (comparing the center-valued dimensions
which are defined for finitely generated projective D(K[G])-modules by the
discussion of Section 2) φ([Qα]) = qα[P i] if tru
G(Qα) = qαP i. By the Chinese
remainder theorem and the definition of Li as the smallest integers such that
all the qα are integer multiples of L−1
[P i] belongs to the
image of φ.
(2) =⇒ (3): For a semisimple Artinian ring every finitely generated module
is projective, therefore G0 = K0 under the assumptions we make.
(3) =⇒ (4): Let M be a finitely presented K[G]-module with presentation
K[G]l A−→ K[G]n → M → 0, A ∈ Mn×l(K[G]). Then M ⊗K[G] D(K[G])
is finitely generated, therefore by the assumption stably isomorphic to an
integer linear combination L aixiD(K[G]) with xi projectors defined over
finite subgroups E of G -- note that G0(K[E]) = K0(K[E]) for any finite
Inducing further to U(G) and
group E, as K[E] is semisimple Artinian.
using that the dimension function extends to finitely presented U(G)-modules
(which is additive, so that we can leave out the stabilization summands), we
read off that
, also [xi] = L−1
i
i
dimu
G(M) = dimu
G(M aixiU(G)) = X ai dimu
G(xiU(G)) ∈ LK(G)
by definition of LK(G). Finally, by additivity of the von Neumann dimension
dimu
G(ker(A)) = n − dimu
G(M) ∈ LK(G).
On the center-valued Atiyah conjecture
16
(4) =⇒ (1): Here, we follow closely the argument of the proof of [20, Propo-
sition 2.14]. Our assumption implies by Theorem 1.10 that E(K[G]) =
D(K[G]). Because the center-valued Atiyah conjecture implies that the or-
dinary L2-Betti numbers are contained in a finitely generated subgroup of Q
(generated by trG(P j)/Lj), by [20, Theorem 2.7] D(K[G]) is a semisimple
Artinian ring.
The P j are central idempotents in D(K[G]). We have to show that they
are primitive central idempotents, and that each is the sum of exactly Lj
orthogonal sub-idempotents which are themselves irreducible. The structure
theory of rings then implies that each P jD(K[G])P j is simple Artinian and
an Lj × Lj-matrix ring over a skew field.
Fix, as in Definition 3.8, the (finite) collection of sub-projections Qα of P j,
where the Qα are irreducible projections supported on K[Hα] and Hα runs
through the (isomorphism classes of) finite extensions of ∆+(G) inside G.
Then tru
P j with integers nα, and by definition of Lj we have
gcdα(nα) = 1. Set d := lcmα(nα).
Consider now P jU(G)d. Because
G(Qα) = nα
Lj
dimu
G(P jU(G)d) = dP j = dimu
G(QαU(G)Ljd/nα)
by [22, Theorem 9.13(1)] then P jU(G)d ∼= QαU(G)Ljd/nα, so we find Ljd/nα
mutually orthogonal projections in Md(U(G)) corresponding to the copies
of Qα. Because the center-valued trace of each of those equals nα
P j =
Lj
tru
G(Qα), by [7, Exercise 13.15A], there exist Ljd/nα similarities (i.e. self-
adjoint unitaries) ui ∈ U(G) with u1 = 1 such that these projections can be
written as uiP ′
α is the diagonal matrix with first entry Pα and
all other entries 0).
Then, exactly as in the proof of [20, Proposition 2.14] we can replace the
ui by ui ∈ Md(D(K[G])) which are invertible and such that we still have a
direct sum decomposition
αui (where P ′
P jD(K[G])d =
Lj d/nα
M
i=1
uiP ′
αD(K[G])d.
(3.10)
This uses the Kaplansky density theorem, the quantization of the center-
valued trace and [20, Lemma 2.12].
Let us now take a central idempotent ǫ in D(K[G]) which is a sub-projection
of P j (i.e. ǫP j = ǫ). We have to show that ǫ = 0 or ǫ = P j. To do this,
On the center-valued Atiyah conjecture
17
we compute tru
therefore by Equation (3.10)
G(ǫ). Note that all the modules ǫuiP ′
αU(G)d are isomorphic,
d tru
G(ǫ) = dimu
G(ǫU(G)d) =
Ljd
nα
dimu
G(ǫP ′
αU(G)d).
(3.11)
By Lemma 2.4 and the assumption (4), Lj · dimu
G(ǫP ′
multiple of P j. Therefore, rearranging Equation (3.11)
αU(G)d) is an integer
As this holds for all α, even
nα tru
G(ǫ) ∈ ZP j.
ǫ = tru
G(ǫ) = lcmα(nα) tru
G(ǫ) ∈ ZP j.
So we can indeed conclude that P j is a primitive central idempotent and
therefore P jD(K[G]) is an l × l matrix ring over a skew field. It follows that
P jD(K[G])nαd is the direct sum of nαdl copies of an irreducible submod-
ule. On the other hand, P jD(K[G])nαd is the direct sum of Ljd isomorphic
summands for every α. As lcmα(nα) = 1 we conclude that Lj l. On the
other hand, by the assumption (4) and Lemma 2.4, the center-valued dimen-
sion of the irreducible submodule (which is generated by one projector as
P jD(K[G]) is Artinian) is an integer multiple of L−1
j P j and therefore Lj l.
It follows that l = Lj as claimed.
4 Special cases and inheritance properties of
the center-valued Atiyah conjecture
Throughout this section, we assume that K is a subfield of C which is closed
under complex conjugation.
4.1 Lemma. The center-valued Atiyah conjecture is true for finitely gener-
ated virtually free groups.
Proof. This follows from the proof of [17, Proposition 5.1(i) and Lemma
5.2(ii)] (in which C can be replaced by any subfield of C) and Theorem
3.9(2).
4.2 Lemma. If G is a directed union of groups Gi and the center-valued
Atiyah conjecture over K is true for all groups Gi, then it is also true for G.
On the center-valued Atiyah conjecture
18
Proof. By [17, Lemma 5.3], D(K[G]) is the directed union of the D(K[Gi]).
Any matrix A over D(K[G]) is therefore already a matrix over D(K[Gi]) for
some i, with dimu
Gi(ker(A)) ∈ LK(Gi). Composition with the center-valued
trace for G gives (by the induction formula for von Neumann dimensions)
dimu
G(LK(Gi)) ⊂ LK(G).
G(ker(A)) ∈ tru
4.3 Proposition. Assume that we have an extension 1 → H → G π−→ E →
1 where E is elementary amenable and for each finite subgroup F ≤ E,
π−1(F ) ≤ G satisfies the center-valued Atiyah conjecture over K . Then also
K[G] satisfies the center-valued Atiyah conjecture.
Proof. By transfinite induction, the statement is a formal consequence of the
same assertion where E is finitely generated virtually abelian, as explained
e.g. in the proof of [29, Proposition 3.1] or in [17].
If E is finitely generated virtually abelian then in the proof of [17, Lemma
5.3] it is shown that
M
F ≤E finite
G0(D(K[π−1(F )])) → G0(D(K[G]))
is onto, using Moody's induction theorem [24, Theorem 1]. Since by assump-
tion LU ≤π−1(F ) finite G0(K[U]) → G0(D(K[π−1(F )])) is onto for each such F
and the composition of surjective maps is surjective we conclude that
M
F ∈F (G)
G0(K[F ]) → G0(D(K[G]))
is onto and (3) of Theorem 3.9 is established.
4.4 Proposition. Let K be a subfield of Q which is closed under complex
conjugation. Assume that G is a group with a sequence G ≥ G1 ≥ · · · of
normal subgroups such that Ti∈N Gi = {1}. Assume moreover that for each
i ∈ N and each finite subgroup F ≤ G/Gi there is a finite subgroup F ′ ≤ G
which is mapped isomorphically to F by the projection G → G/Gi.
Finally, assume that each G/Gi satisfies the determinant bound conjecture
and the center-valued Atiyah conjecture over K. Then K[G] satisfies the
center-valued Atiyah conjecture.
Proof. As the statement is empty if lcm(G) = ∞, we assume that lcm(G) <
∞. We first show that, if i is large enough, πi induces an isomorphism
On the center-valued Atiyah conjecture
19
πi : ∆+(G) → ∆+(G/Gi). Dropping finitely many terms in the sequence we
can then assume that this is the case for all i ∈ N. To prove the asser-
tion, choose a finite subgroup M of G with maximal order (possible since
lcm(G) < ∞). Note that the product ∆+M is also a finite subgroup, there-
fore by maximality equal to M, consequently ∆+ ≤ M. Then choose finitely
many g1, . . . , gn ∈ G such that ∆+(G) = Tn
k=1 M gk (where M g denotes the
conjugate gMg−1), which is possible by the descending chain condition for
finite sets.
Finally, choose r > 0 such that πr : G → G/Gr is injective when restricted
to Sn
Because πr is surjective, πr(∆+(G)) is a finite normal subgroup of G/Gi and
therefore πr(∆+(G)) ≤ ∆+(G/Gr). On the other hand, πr(M) is a finite sub-
group with maximal order in G/Gr (because πrM is injective and every finite
subgroup of G/Gr is an isomorphic image of a finite subgroup of G), there-
fore ∆+(G/Gr) ≤ πr(M), by normality even ∆+(G/Gr) ≤ Tn
k=1 πr(M)πr(g).
As Tn
k=1 M g = ∆+(G) and by injectivity of πr on Sn
k=1 M gk , which is possible because Ti Gi = {1}.
k=1 M g we finally get
∆+(G/Gr) ≤
n
\
k=1
πr(M)πr(g) = πr(∆+(G)) ≤ ∆+(G/Gr).
This implies the statement for all i ≥ r.
Secondly, given g ∈ G of infinite order, for all sufficiently large i, the restric-
tion of πi to {1, g, g2, . . . , glcm(G)} is injective and therefore, as by assumption
the orders of finite subgroups of G/Gi are bounded by lcm(G), πi(g) also has
infinite order.
Fix now A ∈ Md(K[G]) and denote by Qi the projection onto the kernel of
A[i] := pi(A). Recall that
tru
G(Qi) = dimu
G(ker(A)) = X
g∈G
hdimu
G(ker(A)), gil2(G)g,
G(ker(A)), gi the coefficient of g in dimu
G(Qi).
and we denote by hdimu
correspondingly for tru
The center-valued Atiyah conjecture for K[G/Gi] implies in particular that
tru
G(Qi) is contained in K[∆+(G/Gi)], therefore supported only on elements
of finite order. Consequently, if g ∈ G has infinite order, then htru
0 for sufficiently large i and, by Theorem 2.6, hdimu
implies that dimu
contained in Z(N (G)) ∩ K[∆+(G)].
G(ker(A)), gi = 0. This
G(ker(A)) is supported on elements of finite order, i.e. is
G(ker(A)), and
G(Qi), pri(g)i =
On the center-valued Atiyah conjecture
20
As explained above, we can use πi to identify ∆+(G) and ∆+(G/Gi) and
consider tru
G(Qi) as an element of K[∆+(G)]. By Theorem 2.6, for each
g ∈ ∆+(G),
hdimu
G(ker(A)), gi = lim
i→∞
htru
G(Qi), gi.
Since all the (finitely many) coefficients converge, we even have
lim
i→∞
tru
G(Qi) = dimu
G(ker(A)) ∈ Z(N (G)) ∩ K[∆+(G)].
Because the sets of isomorphism classes of finite subgroups of G/Gi and of G
are identified by πi, we get exactly the same relevant irreducible projections
defined over finite subgroups and the same central idempotents in the formu-
las of Lemma 3.1 and Lemma 3.4 for LK(G) and LK(G/Gi). Consequently, πi
identifies LK(G) and LK(G/Gi). Finally, observe that by assumption about
the Atiyah conjecture for G/Gi we have tru
G(Qi) ∈ LK(G). As the latter is a
discrete subset of Z(N (G)), we finally observe that dimu
G(ker(A)) ∈ LK(G),
i.e. K[G] satisfies the center-valued Atiyah conjecture.
4.5 Theorem. The center-valued Atiyah conjecture is true for all groups
G ∈ C.
Proof. In the proof of [17, Lemma 4.9] it is shown that the assertion fol-
lows (by transfinite induction) directly from Lemma 4.1, Lemma 4.2 and
Proposition 4.3.
4.6 Corollary. Let K be a subfield of Q which is closed under complex con-
jugation. Then the center-valued Atiyah conjecture is true for all elementary
amenable extensions of pure braid groups, of right-angled Artin groups, of
primitive link groups, of cocompact special groups, or of products of such.
Proof. Each of the groups in the list has a sequence of normal subgroups
with trivial intersection and with elementary amenable quotients such that
in addition the condition of Proposition 4.4 is met. This is shown for the
extensions of pure braid groups in [19], for primitive link groups in [8] and
for right-angled Coxeter and Artin groups in [18], and combining [30] with
[18] it also follows for special cocompact groups. Combining Theorem 4.5
and Proposition 4.4, the assertion follows.
On the center-valued Atiyah conjecture
21
References
[1] Ian Agol, The virtual Haken conjecture, Doc. Math. 18 (2013), 1045 -- 1087. With an
appendix by Agol, Daniel Groves, and Jason Manning.
[2] William Arveson, An invitation to C ∗-algebras, Springer-Verlag, New York, 1976.
Graduate Texts in Mathematics, No. 39.
[3] Michael F. Atiyah, Elliptic operators, discrete groups and von Neumann algebras,
Colloque "Analyse et Topologie" en l'Honneur de Henri Cartan (Orsay, 1974), Soc.
Math. France, Paris, 1976, pp. 43 -- 72. Ast´erisque, No. 32-33. MR0420729 (54 #8741)
[4] Tim Austin, Rational group ring elements with kernels having irrational dimension,
Proc. Lond. Math. Soc. (3) 107 (2013), no. 6, 1424 -- 1448, DOI 10.1112/plms/pdt029.
MR3149852
[5] Nicolas and Wise Bergeron Daniel T., A boundary criterion for cubulation, Amer. J.
Math. 134 (2012), no. 3, 843 -- 859, DOI 10.1353/ajm.2012.0020.
[6] Gary F. Birkenmeier, Jae Keol Park, and S. Tariq Rizvi, A theory of hulls for rings
and modules, Ring and module theory, Trends Math., Birkhauser/Springer Basel
AG, Basel, 2010, pp. 27 -- 71, DOI 10.1007/978-3-0346-0007-1-2, (to appear in print).
MR2744041 (2012c:16001)
[7] Sterling K. Berberian, Baer ∗-rings, Springer-Verlag, New York, 1972. Die
Grundlehren der mathematischen Wissenschaften, Band 195. MR0429975 (55 #2983)
[8] Inga Blomer, Peter A. Linnell, and Thomas Schick, Galois cohomology of com-
pleted link groups, Proc. Amer. Math. Soc. 136 (2008), no. 10, 3449 -- 3459, DOI
10.1090/S0002-9939-08-09395-7. MR2415028 (2009i:20059)
[9] J´ozef Dodziuk, Peter Linnell, Varghese Mathai, Thomas Schick, and Stuart Yates,
Approximating L2-invariants and the Atiyah conjecture, Comm. Pure Appl. Math. 56
(2003), no. 7, 839 -- 873, DOI 10.1002/cpa.10076. Dedicated to the memory of Jurgen
K. Moser. MR1990479 (2004g:58040)
[10] G´abor Elek and Endre Szab´o, Hyperlinearity, essentially free actions and L2-
invariants. The sofic property, Math. Ann. 332 (2005), no. 2, 421 -- 441, DOI
10.1007/s00208-005-0640-8. MR2178069 (2007i:43002)
[11] Daniel R. Farkas and Peter A. Linnell, Congruence subgroups and the Atiyah con-
jecture, Groups, rings and algebras, Contemp. Math., vol. 420, Amer. Math. Soc.,
Providence, RI, 2006, pp. 89 -- 102. MR2279234 (2008b:16033)
[12] Lukasz Grabowski, On Turing dynamical systems and the Atiyah problem, Invent.
Math. 198 (2014), no. 1, 27 -- 69, DOI 10.1007/s00222-013-0497-5. MR3260857
[13] Fr´ed´eric and Wise Haglund Daniel T., Special cube complexes, Geom. Funct. Anal.
17 (2008), no. 5, 1551 -- 1620, DOI 10.1007/s00039-007-0629-4.
[14] Richard V. Kadison and John R. Ringrose, Fundamentals of the theory of opera-
tor algebras. Vol. II, Pure and Applied Mathematics, vol. 100, Academic Press Inc.,
Orlando, FL, 1986. Advanced theory. MR859186 (88d:46106)
On the center-valued Atiyah conjecture
22
[15] Anselm Knebusch, Approximation of center-valued Betti numbers, Houston Journal
of Mathematics 37 (2011), 161 -- 179.
[16] Serge Lang, Algebra, 3rd ed., Graduate Texts in Mathematics, vol. 211, Springer-
Verlag, New York, 2002. MR1878556 (2003e:00003)
[17] Peter A. Linnell, Division rings and group von Neumann algebras, Forum Math. 5
(1993), no. 6, 561 -- 576, DOI 10.1515/form.1993.5.561. MR1242889 (94h:20009)
[18] Peter Linnell, Boris Okun, and Thomas Schick, The strong Atiyah conjecture for
right-angled Artin and Coxeter groups, Geom. Dedicata 158 (2012), 261 -- 266, DOI
10.1007/s10711-011-9631-y. MR2922714
[19] Peter Linnell and Thomas Schick, Finite group extensions and the Atiyah conjecture,
J. Amer. Math. Soc. 20 (2007), no. 4, 1003 -- 1051 (electronic), DOI 10.1090/S0894-
0347-07-00561-9. MR2328714 (2008m:58041)
[20]
, The Atiyah conjecture and Artinian rings, Pure and Applied Mathematics
Quarterly 8 (2007), 313 -- 328. arXiv:0711.3328.
[21] Wolfgang Luck, Hilbert modules and modules over finite von Neumann algebras
and applications to L2-invariants, Math. Ann. 309 (1997), no. 2, 247 -- 285, DOI
10.1007/s002080050112. MR1474192 (99d:58169)
[22]
, L2-invariants: theory and applications to geometry and K-theory, Ergebnisse
der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in
Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of
Modern Surveys in Mathematics], vol. 44, Springer-Verlag, Berlin, 2002. MR1926649
(2003m:58033)
[23]
, L2-invariants from the algebraic point of view, Geometric and cohomological
methods in group theory, London Math. Soc. Lecture Note Ser., vol. 358, Cambridge
Univ. Press, Cambridge, 2009, pp. 63 -- 161. MR2605176 (2011f:46090)
[24] John Atwell Moody, Brauer induction for G0 of certain infinite groups, J. Algebra
122 (1989), no. 1, 1 -- 14, DOI 10.1016/0021-8693(89)90235-4. MR994933 (90b:18014)
[25] Donald S. Passman, Infinite group rings, Marcel Dekker Inc., New York, 1971. Pure
and Applied Mathematics, 6. MR0314951 (47 #3500)
[26] Mikael Pichot, Thomas Schick, and Andrzej
Zuk, Closed manifolds with transcen-
dental L2-Betti numbers, J. Lond. Math. Soc. (2) 92 (2015), no. 2, 371 -- 392, DOI
10.1112/jlms/jdv026. MR3404029
[27] Holger Reich, On the K- and L-theory of
the algebra of operators affiliated
to a finite von Neumann algebra, K-Theory 24 (2001), no. 4, 303 -- 326, DOI
10.1023/A:1014078228859. MR1885124 (2003m:46103)
[28] Thomas Schick, L2-determinant class and approximation of L2-Betti numbers, Trans.
Amer. Math. Soc. 353 (2001), no. 8, 3247 -- 3265 (electronic), DOI 10.1090/S0002-
9947-01-02699-X. Erratum at arXiv:math/9807032. MR1828605 (2002f:58056)
On the center-valued Atiyah conjecture
23
[29]
, Integrality of L2-Betti numbers, Math. Ann. 317 (2000), no. 4, 727 -- 750, DOI
10.1007/PL00004421. Erratum in vol. 322, 421 -- 422. MR1777117 (2002k:55009a)
[30] Kevin Schreve, The strong Atiyah conjecture for virtually cocompact special groups,
Math. Ann. 359 (2014), no. 3-4, 629 -- 636, DOI 10.1007/s00208-014-1007-9.
|
1706.01147 | 3 | 1706 | 2018-07-24T13:48:52 | Taylor term does not imply any nontrivial linear one-equality Maltsev condition | [
"math.RA",
"math.LO"
] | It is known that any finite idempotent algebra that satisfies a nontrivial Maltsev condition must satisfy the linear one-equality Maltsev condition (a variant of the term discovered by M. Siggers and refined by K. Kearnes, P. Markovi\'c, and R. McKenzie):
\[
t(r,a,r,e)\approx t(a,r,e,a).
\] We show that if we drop the finiteness assumption, the $k$-ary weak near unanimity equations imply only trivial linear one-equality Maltsev conditions for every $k\geq 3$. From this it follows that there is no nontrivial linear one-equality condition that would hold in all idempotent algebras having Taylor terms.
Miroslav Ol\v{s}\'ak has recently shown that there is a weakest nontrivial strong Maltsev condition for idempotent algebras. Ol\v{s}\'ak has found several such (mutually equivalent) conditions consisting of two or more equations. Our result shows that Ol\v{s}\'ak's equation systems can't be compressed into just one equation. | math.RA | math |
Taylor term does not imply any nontrivial linear
one-equality Maltsev condition
Alexandr Kazda
Abstract. It is known that any finite idempotent algebra that satisfies a nontrivial
Maltsev condition must satisfy the linear one-equality Maltsev condition (a variant
of the term discovered by M. Siggers and refined by K. Kearnes, P. Markovi´c, and
R. McKenzie):
t(r, a, r, e) ≈ t(a, r, e, a).
We show that if we drop the finiteness assumption, the k-ary weak near unanimity
equations imply only trivial linear one-equality Maltsev conditions for every k ≥ 3.
From this it follows that there is no nontrivial linear one-equality condition that would
hold in all idempotent algebras having Taylor terms.
Miroslav Ols´ak has recently shown that there is a weakest nontrivial strong Maltsev
condition for idempotent algebras. Ols´ak has found several such (mutually equivalent)
conditions consisting of two or more equations. Our result shows that Ols´ak's equation
systems can't be compressed into just one equation.
1. Introduction
In this note we show that for every k ≥ 3 the free algebra with a k-ary
weak near unanimity term does not satisfy any nontrivial linear one-equality
Maltsev condition. This is in contrast to the finite case where having a Taylor
term means that the algebra in question has the Siggers term [7]. The original
Siggers term is equivalent [4, Theorem 2.2.] to the single equation form (the
mnemonics for names of variables is due to Ryan O'Donnel):
t(r, a, r, e) ≈ t(a, r, e, a).
Miroslav Ols´ak has recently shown that having a Taylor term is a strong Malt-
sev condition even for infinite (idempotent) algebras [6]. Ols´ak's shortest con-
dition consists of two linear identities and it would be natural to ask if one
can't do better and use only one equation. This paper shows that such an
improvement is impossible.
2. Preliminaries
An algebra A consists of a base set A on which acts a set of basic operations
of A. An operation is a mapping f : An → A where n ∈ N is the arity of f .
This work was supported by European Research Council under the European Union's
Seventh Framework Programme (FP7/2007-2013)/ERC grant agreement no 616160, the
PRIMUS/SCI/12 and UNCE/SCI/022 projects of the Charles University.
2
Alexandr Kazda
Algebra univers.
The clone of operations of A is the smallest set of operations that contains
all basic operations of A as well as projections (that is, operations of the form
πk
i (x1, . . . , xk) = xi) and is closed under composing operations. An algebra is
idempotent if for any operation t of A and any a ∈ A we have t(a, . . . , a) = a.
It is an easy exercise to verify that for A to be idempotent it suffices that just
the basic operations of A are all idempotent.
A term of A is a (syntactically correct) composition of basic operation
symbols of A and variables. An equational identity, or equation, is a statement
of the form "u ≈ v" where u and v are terms and the symbol "≈" stands for
"the left hand side equals the right hand side after any assignment of members
of A to variables." An example of an identity is t(x, . . . , x) ≈ x which says
that the operation t is idempotent.
A variety is a class of algebras sharing the same signature (the same basic
operation symbols and arities of basic operations) that is closed under taking
subalgebras, products and homomorphic images, or equivalently(by Birkhoff's
theorem [3, Theorem 4.41]) a class of algebras defined by a system of equational
identities. The set of all identities that holds in a variety is called the equational
theory of the variety.
A strong Maltsev condition is a finite list of identities involving some op-
eration symbols. An algebra A satisfies a strong Maltsev condition M if for
each k-ary operation symbol in M one can choose a k-ary operation of A so
that when we replace the symbols of M by operations of A, we get a system
of equations that are all true in A. A variety satisfies the condition M if
all algebras in the variety satisfy M . Since we consider only strong Maltsev
conditions in this paper, we will omit the adjective "strong" for brevity.
A Maltsev condition is trivial
if it is satisfied by the algebra P on two
elements 0 and 1 whose set of operations consists of projections only. An
example of a trivial strong Maltsev condition is t(t(x, y, z), y, z) ≈ t(x, x, z);
one can satisfy this condition by choosing t to be the third projection (i.e.
t(a, b, c) = c for all a, b, c).
A Maltsev condition is called linear if its identities don't involve compo-
sitions, i.e. all identities have the form t(x) ≈ s(y) or t(x) ≈ z or z ≈ r
where x, y are tuples of variables, t, s are (possibly equal) operation symbols,
and z, r are variables (we include the third case for completeness only; only a
trivial algebra can satisfy an identity of the form z ≈ r where z, r are distinct
variables).
Having a k-ary weak near unanimity operation (k-wnu) for a fixed k ∈
{2, 3, 4, . . . } is a Maltsev condition that consists of the following k linear iden-
tities for the k-ary operation symbol w:
w(x, x, x, . . . , x, x) ≈ x
w(y, x, x, . . . , x, x) ≈ w(x, y, x, . . . , x, x) ≈ w(x, x, y, . . . , x, x) ≈ · · ·
· · · ≈ w(x, x, x, . . . , y, x) ≈ w(x, x, x, . . . , x, y).
Vol. 00, XX
Taylor term does not imply any nontrivial SLEMC
3
Having a Taylor operation (term) refers to having an operation t satisfying
any linear Maltsev condition of the form
t(x, . . . , x) ≈ x
t(x, ?, ?, . . . , ?) ≈ t(y, ?, ?, . . . , ?)
t(?, x, ?, . . . , ?) ≈ t(?, y, ?, . . . , ?)
...
t(?, ?, ?, . . . , x) ≈ t(?, ?, ?, . . . , y),
where x, y are variables and the question marks stand for some choice of x's
and y's.
It is immediate to see that any operation that is a k-wnu is also a Taylor
term (but not the other way around). Finite idempotent algebras with Taylor
terms are well understood as the following theorem shows.
Theorem 1 (Combining [8], [5], [4], and [2]). Let A be a finite idempotent
algebra. Then the following are equivalent:
(1) A satisfies a nontrivial Maltsev condition,
(2) A has a Taylor term,
(3) A has a k-wnu for some k ∈ N,
(4) A has a k-ary cyclic term for some k ∈ N, where a cyclic term satisfies
the equation
c(x1, x2, . . . , xk) ≈ c(xk, x1, . . . , xk−1).
(5) A satisfies the Maltsev condition (known as a Siggers term)
t(r, a, r, e) ≈ t(a, r, e, a)
where a, e, r are variables.
Note that the cyclic and Siggers term conditions, unlike the other equivalent
conditions involve only one identity (plus idempotency, which we assume from
the start). We will abbreviate single linear equality Maltsev condition as
SLEMC. Siggers term and cyclic term conditions are examples of nontrivial
SLEMCs, while the 3-wnu condition is not a SLEMC.
Our work stems from an attempt to generalize Theorem 1 to infinite idem-
potent algebras. We will show that there is no analogue of the last two points,
i.e. that having a k-wnu for k ≥ 3 does not imply a nontrivial SLEMC. (For
k = 2, we have the SLEMC w(x, y) ≈ w(y, x).)
3. 3-wnu implies only trivial SLEMCs
In this section we show in detail that having a 3-wnu term does not imply
any nontrivial SLEMC. The general case of having a k-wnu differs from the
3-wnu situation only by a slightly more complicated notation. This is why
4
Alexandr Kazda
Algebra univers.
we first give the proof for 3-wnu and then, in the next section, we sketch the
argument for the general case without going into details.
Theorem 2. Let V be the variety of algebras with one ternary basic operation
w and with the equational theory generated by the 3-wnu identities
w(x, x, x) ≈ x
w(x, x, y) ≈ w(x, y, x) ≈ w(y, x, x).
This variety (which is idempotent and has a 3-wnu term) does not satisfy any
nontrivial SLEMC.
The proof of this theorem will occupy the rest of this section. From the
equations, we can see that V is idempotent and that w is a 3-wnu operation,
so the only nonobvious statement is that V does not satisfy any nontrivial
SLEMC.
Since V contains algebras on more than one element (for example {0, 1}
with w(x, y, z) = x + y + z (mod 2)), any candidate for a nontrivial SLEMC
has to have a rather specific shape:
Observation 3. Let A be an idempotent algebra on at least two elements. If
A satisfies a nontrivial SLEMC M , then M has the form
t(x1, . . . , xm) ≈ t(y1, . . . , ym)
where t is an operation symbol and x1, . . . , xm, y1, . . . , ym are variable symbols
such that xi 6= yi for i = 1, 2, . . . , m.
Proof. Assume that A satisfies a nontrivial SLEMC M of the form
r(x1, . . . , xm) ≈ s(y1, . . . , yk)
where r, s are two different operation symbols. Since the condition M is sup-
posed to be nontrivial, the variable sets {x1, . . . , xm} and {y1, . . . , yk} must
be disjoint (for had we xi = yj then we could satisfy M by taking r and s to
be the projections to the i-th and j-th coordinates, respectively). Therefore,
the SLEMC M implies
r(x, . . . , x) ≈ s(y, . . . , y)
where x, y are distinct variable symbols. Since A is idempotent, the operations
of A realizing r and s are idempotent and A satisfies x ≈ y, meaning A = 1.
A similar argument rules out the SLEMC
This leaves only the possibility
t(x1, . . . , xm) ≈ y.
t(x1, . . . , xm) ≈ t(y1, . . . , ym),
where xi 6= yi for all i (were xi = yi, we could satisfy M by taking t to be the
m-ary projection to the i-th coordinate).
(cid:3)
Vol. 00, XX
Taylor term does not imply any nontrivial SLEMC
5
We will now construct an algera in V that satisfies no nontrivial SLEMC.
Two comments before we begin: First, we are actually going to construct the
free countably generated algebra in V . Second, we note for readers familiar
with term rewrite systems (see eg. [1]) that we are implicitly studying the term
rewrite system with the rules w(x, x, x) → x, w(y, x, x), w(x, y, x), w(x, x, y) →
u(x, y) where u(x, y) is a new symbol that stands for w(y, x, x). We opted to
not use the machinery of term rewriting because an elementary argument is
reasonably short and prepares us for calculations with normal forms later on.
Let X be a countable set of variable symbols. Let T be the set of all
possible terms we can get using X and a single ternary operation symbol w
(so for example w(y, w(x, y, z), y) ∈ T ).
We define the set A of "normal forms" of terms of T modulo the 3-wnu
identities as follows: A term t lies in A if either t is a variable from X, or
t = w(a1, a2, a3) where a1, a2, a3 ∈ A and we have a1 6= a2, a3 (for example,
w(w(x, y, z), y, y) lies in A, but w(y, w(x, y, z), y) does not).
Let t ∈ T be a term. It is easy to prove by induction on the number of
occurrences of w in t that we can use the 3-wnu identitites to rewrite t to a
term t′ ∈ A such that t ≈ t′ in V (in fact, the term t′ is unique for a given t;
we omit the proof of this as we will not need it).
Consider the algebra A = (A, wA) with the operation wA defined as follows:
(1) If a1, a2, a3 ∈ A are pairwise different then we let wA(a1, a2, a3) =
w(a1, a2, a3),
(2) if a, b ∈ A are different then we let all three of wA(a, a, b), wA(a, b, a),
wA(b, a, a) to be equal to w(b, a, a), and
(3) if a ∈ A, then wA(a, a, a) = a.
It is easy to verify that A is closed under wA. Observe also that the
operation wA is a 3-wnu operation, so A ∈ V .
Note that in many cases we have wA(a, b, c) = w(a, b, c), but this is not
always true. This is why we distinguish between wA (operation symbol of A)
and w (formal symbol used to describe terms of V ).
Since A ∈ V , to prove Theorem 2 it is enough to show that A satisfies
only trivial SLEMCs. To that end let R be the subalgebra of A2 generated
by {(x, y) : x, y ∈ X, x 6= y}. The following observation shows that to prove
Theorem 2, it is enough to show that R does not intersect the diagonal.
Observation 4. If the variety V satisfies a nontrivial SLEMC, then the re-
lation R defined above intersects the diagonal of A2 (in other words, there is
an r ∈ A such that (r, r) ∈ R).
Proof. Assume that V satisfies the SLEMC
t(y1, . . . , ym) ≈ t(z1, . . . , zm)
where y1, . . . , ym, z1, . . . , zm are variables (without loss of generality) taken
from the set X. Let us denote by tA the term of A we obtain from t by
6
Alexandr Kazda
Algebra univers.
replacing each symbol w by wA. In A we thus have the equality (note that
yi's and zi's are members of A here, not variable symbols).
tA(y1, . . . , ym) = tA(z1, . . . , zm) = r.
for some r ∈ A.
We have (yi, zi) ∈ R for all i and so applying the operation tA to pairs
(cid:3)
(y1, z1), . . . , (ym, zm) ∈ R gets us (r, r) ∈ R.
While we would like to show that R does not intersect the diagonal, idem-
potency prevents us from comfortably doing a proof by induction on term
complexity on R itself. This is why we take a detour through subterms.
Definition 5. We define the relation "to be a subterm" on the set A, denoted
by (cid:22), as the reflexive and transitive closure of the set
Q = {(a, b) : a, b ∈ A, ∃c, d, e ∈ A, b = w(c, d, e), a ∈ {c, d, e}}.
Informally, a (cid:22) b if in the term b we can find a subterm that is identical to
a. Note that (cid:22) is defined using the (syntactic) symbol w. However, it turns
out that (cid:22) behaves well with respect to the operation wA, too:
Lemma 6. The following holds for (cid:22):
(a) If x, y are distinct members of X (i.e. variables), then x 6(cid:22) y.
(b) For all b, c, d ∈ A, we have b (cid:22) wA(b, c, d), wA(d, b, c), wA(c, d, b).
(c) For all a, b, c, d ∈ A such that a (cid:22) b we have a (cid:22) wA(b, c, d), wA(d, b, c),
wA(c, d, b).
(d) If a, b, c, d ∈ A are such that a (cid:22) wA(b, c, d) and a 6(cid:22) b, c, d, then a =
wA(b, c, d).
Proof. (a) Since x 6= y, the only way we could have had x (cid:22) y would be if there
was a chain of k ≥ 2 terms x = t1, t2, . . . , tk = y such that (ti, ti+1) ∈ Q
for i = 1, . . . , k − 1 (where Q is the set from the definition of (cid:22)). From
this we get that for all i = 1, . . . , k − 1 we have ti+1 = w(pi, qi, ri) where
pi, qi, ri are members of A and ti appears at least once in (pi, qi, ri). By
induction on i, it follows that each ti must have at least i − 1 occurrences
of the symbol w, so tk contains at least one symbol w. But the term tk = y
has no w in it, a contradiction.
(b) We will show b (cid:22) wA(b, c, d); the other two subterm relationships are
similar.
Unless b = c = d, we have wA(b, c, d) ∈ {w(b, c, d), w(d, b, c), w(c, d, b)}
and b is a subterm of each of the terms on the right side, so we are done.
In the case b = c = d, we get wA(b, c, d) = b and b (cid:22) b follows from the
reflexivity of (cid:22).
(c) This follows from the transitivity of (cid:22) and the previous point: We have
a (cid:22) b (cid:22) wA(b, c, d), wA(d, b, c), wA(c, d, b).
Vol. 00, XX
Taylor term does not imply any nontrivial SLEMC
7
(d) Were b, c, d all equal, we would have a (cid:22) wA(b, b, b) = b, a contradiction.
with a 6(cid:22) b. Therefore, without loss of generality wA(b, c, d) = w(b, c, d)
(we can reorder b, c, d).
Assume for a contradiction that a 6= wA(b, c, d) = w(b, c, d). Since
the relation a (cid:22) w(b, c, d) is not a consequence of reflexivity, there is a
k ≥ 2 and a chain a = t1 (cid:22) t2 (cid:22) · · · (cid:22) tk−1 (cid:22) tk = w(b, c, d) witnessing
a (cid:22) w(b, c, d) with (ti, ti+1) ∈ Q for all i. But then a (cid:22) tk−1 and tk−1 needs
to be one of b, c, d by the definition of Q, a contradiction with a 6(cid:22) b, c, d.
(cid:3)
Let now S be the following relation on A:
S = {(a, b) ∈ A2 : a 6(cid:22) b ∧ b 6(cid:22) a}.
By part (a) of Lemma 6, the generators of R lie in S and since (cid:22) is reflexive,
S does not intersect the diagonal.
Lemma 7. The relation S is a subuniverse of A2.
Proof. Let us take (a1, b1), (a2, b2), (a3, b3) ∈ S such that (without loss of gen-
erality) wA(a1, a2, a3) (cid:22) wA(b1, b2, b3).
We consider several cases:
(a) Assume that wA(a1, a2, a3) 6= wA(b1, b2, b3). Then by part (d) of Lemma 6,
wA(a1, a2, a3) (which plays the role of a in the Lemma) needs to be a sub-
term of one of b1, b2, b3. Without loss of generality assume wA(a1, a2, a3) (cid:22)
b1. But a1 (cid:22) wA(a1, a2, a3) by part (c) of Lemma 6. We have a1 (cid:22)
wA(a1, a2, a2) (cid:22) b1, which is a contradiction with (a1, b1) ∈ S.
(b) Assume that wA(a1, a2, a3) = wA(b1, b2, b3) and a1 = a2 = a3. Then
b1 (cid:22) wA(b1, b2, b3) = wA(a1, a1, a1) = a1 (where the subterm relationship
follows again by part (c) of Lemma 6) and therefore b1 (cid:22) a1.
The same argument takes care of the case b1 = b2 = b3.
(c) Assume that wA(a1, a2, a3) = wA(b1, b2, b3) and {a1, a2, a3} = {c, d} with
d appearing twice, i.e. wA(a1, a2, a3) = w(c, d, d) = wA(b1, b2, b3)
Since b1, b2, b3 are not all equal, by the definition of wA we must have
{b1, b2, b3} = {c, d} with d appearing twice. Since we have three pairs
(a1, b1), (a2, b2), (a3, b3) and only two appearances of c (one for a's, one
for b's), it follows that there exists an i such that ai = d = bi. However,
(d, d) 6∈ S, a contradiction.
(d) Assume that wA(a1, a2, a3) = wA(b1, b2, b3) and a1, a2, a3 are pairwise
different, i,e. w(a1, a2, a3) = wA(b1, b2, b3). Since b1, b2, b3 are not all
equal, the only way to get equality here is to have ai = bi for all i = 1, 2, 3,
a contradiction with (ai, bi) ∈ S.
(cid:3)
Proof of Theorem 2. By Lemma 7, we get that S is a subuniverse of A2 that
contains all generators of R and thus R ⊆ S. As we have seen above, S is
8
Alexandr Kazda
Algebra univers.
disjoint from the diagonal, so R must be disjoint from the diagonal. Therefore,
by Observation 4, the variety V can't satisfy a nontrivial SLEMC.
(cid:3)
4. k-wnu implies only trivial SLEMCs
Theorem 8. For any k ≥ 3 the k-wnu identities don't imply a nontrivial
SLEMC.
Proof. The proof is very similar to the proof of Theorem 2, so we only sketch
the main points here.
We take V to be the variety defined by the k-wnu equations for a k-wnu
operation w, X a countable set of variables and A the smallest set of terms
made from X and w such that X ⊆ A and w(a1, . . . , ak) ∈ A if and only
if a1, . . . , ak ∈ A and there are at least two distinct indices i, j such that
a1 6= ai, aj.
Again, we consider the algebra A = (A, wA) ∈ V with wA(a1, . . . , ak)
defined as
wA(a1, . . . , ak) =
a1
w(c, d, d, . . . , d)
a1 = a2 = · · · = ak
∃i ∈ {1, . . . , k}, ai = c,
a1 = a2 = · · · = ai−1 = ai+1 = · · ·
· · · = ak = d 6= c
w(a1, a2, . . . , ak) otherwise.
The relation R is again generated in A2 by {(x, y) ∈ X : x 6= y}, while the
subterm relation (cid:22) is defined as the reflexive and transitive closure of
{(a, b) : a, b ∈ A, ∃c1, c2, . . . , ck ∈ A, b = w(c1, . . . , ck), a ∈ {c1, . . . , ck}}.
As before, we show that
S = {(a, b) ∈ A2 : a 6(cid:22) b ∧ b 6(cid:22) a}
is A-invariant and thus prove that R does not intersect the diagonal which
implies that V satisfies no nontrivial SLEMC.
(cid:3)
References
[1] Franz Baader and Tobias Nipkow. Term Rewriting and All That. Cambridge University
Press, New York, 1999.
[2] Libor Barto and Marcin Kozik. Absorbing subalgebras, cyclic terms, and the constraint
satisfaction problem. Logical Methods in Computer Science, 8(1), 2012.
[3] Clifford Bergman. Universal Algebra: Fundamentals and Selected Topics. Chapman &
Hall/CRC Press, Boca Raton and New York and Abingdon, 1st edition, 2011.
[4] Keith Kearnes, Petar Markovi´c, and Ralph McKenzie. Optimal strong Mal'cev
conditions for omitting type 1 in locally finite varieties. Algebra universalis,
72(1):91 -- 100, 2014.
[5] Mikl´os Mar´oti and Ralph McKenzie. Existence theorems for weakly symmetric
operations. Algebra Universalis, 59:463 -- 489, 2008.
[6] Miroslav Ols´ak. The weakest nontrivial idempotent equations. Bulletin of the London
Mathematical Society, 49(6):1028 -- 1047, 2017.
Vol. 00, XX
Taylor term does not imply any nontrivial SLEMC
9
[7] Mark H. Siggers. A strong Mal'cev condition for locally finite varieties omitting the
unary type. Algebra universalis, 64(1):15 -- 20, 2010.
[8] Walter Taylor. Varieties obeying homotopy laws. Canadian Journal of Mathematics,
29:498 -- 527, 1977.
Alexandr Kazda
Department of Algebra, Charles University, Sokolovsk´a 83, 186 75, Prague, Czechia,
ORCID:0000-0002-7338-037X
e-mail : [email protected]
|
1908.08948 | 1 | 1908 | 2019-08-23T18:00:00 | Free Bertini's theorem and applications | [
"math.RA"
] | The simplest version of Bertini's irreducibility theorem states that the generic fiber of a non-composite polynomial function is an irreducible hypersurface. The main result of this paper is its analog for a free algebra: if $f$ is a noncommutative polynomial such that $f-\lambda$ factors for infinitely many scalars $\lambda$, then there exist a noncommutative polynomial $h$ and a nonconstant univariate polynomial $p$ such that $f=p\circ h$. Two applications of free Bertini's theorem for matrix evaluations of noncommutative polynomials are given. An eigenlevel set of $f$ is the set of all matrix tuples $X$ where $f(X)$ attains some given eigenvalue. It is shown that eigenlevel sets of $f$ and $g$ coincide if and only if $fa=ag$ for some nonzero noncommutative polynomial $a$. The second application pertains quasiconvexity and describes polynomials $f$ such that the connected component of $\{X \text{ tuple of symmetric $n\times n$ matrices}: \lambda I\succ f(X) \}$ about the origin is convex for all natural $n$ and $\lambda>0$. It is shown that such a polynomial is either everywhere negative semidefinite or the composition of a univariate and a convex quadratic polynomial. | math.RA | math |
FREE BERTINI'S THEOREM AND APPLICATIONS
JURIJ VOL CI C
Abstract. The simplest version of Bertini's irreducibility theorem states that the
generic fiber of a non-composite polynomial function is an irreducible hypersurface.
The main result of this paper is its analog for a free algebra:
if f is a noncommuta-
tive polynomial such that f − λ factors for infinitely many scalars λ, then there exist
a noncommutative polynomial h and a nonconstant univariate polynomial p such that
f = p ◦ h. Two applications of free Bertini's theorem for matrix evaluations of noncom-
mutative polynomials are given. An eigenlevel set of f is the set of all matrix tuples X
where f (X) attains some given eigenvalue. It is shown that eigenlevel sets of f and g
coincide if and only if f a = ag for some nonzero noncommutative polynomial a. The
second application pertains quasiconvexity and describes polynomials f such that the
connected component of {X tuple of symmetric n × n matrices : λI ≻ f (X)} about the
origin is convex for all natural n and λ > 0.
It is shown that such a polynomial is
either everywhere negative semidefinite or the composition of a univariate and a convex
quadratic polynomial.
1. Introduction
Bertini's irreducibility theorem (see e.g.
[Sha94, Theorem 2.26]) is a fundamental
result with a rich history [Kle98] and omnipresent in algebraic geometry. When applied
to a multivariate polynomial function f over a an algebraically closed field, it states
that the hypersurface {f = λ} is irreducible for all but finitely many values λ unless
f is a composite with a univariate polynomial. This particular case is significant in
its own right in commutative algebra, and has been extensively studied and generalized
[Sch00, BDN09].
In this paper we prove its analog for a free associative algebra and
derive consequences of interest for free analysis [K-VV14] and free real algebraic geometry
[HKM12, BPT13].
Let k be a field and d ∈ N. Let k<x> be the free associative k-algebra in freely
noncommuting variables x = (x1, . . . , xd).
Its elements are called noncommutative
polynomials. We say that f factors in k<x> if f = f1f2 for some nonconstant
f1, f2 ∈ k<x>. Otherwise, f is irreducible over k. A nonconstant f ∈ k<x> is
composite (over k) if there exist h ∈ k<x> and a univariate polynomial p ∈ k[t] such
that deg p > 1 and f = p ◦ h = p(h). Our first main result is the free algebra analog of a
special case of the classical Bertini's (irreducibility) theorem.
Theorem A (Free Bertini's theorem). If f ∈ k<x>\k is not composite, then f − λ is
irreducible over k for all but finitely many λ ∈ k.
Date: August 27, 2019.
2010 Mathematics Subject Classification. Primary 16U30, 13P05; Secondary 47A56, 52A05.
Key words and phrases. Bertini's theorem, free algebra, noncommutative polynomial, composition,
factorization, quasiconvex polynomial.
1
2
J. VOL CI C
See Theorem 3.2 for a more comprehensive statement and proof. Next we apply
Theorem A to matrix evaluations of noncommutative polynomials. Let f ∈ k<x>.
Given X ∈ Mn(k)d let f (X) ∈ Mn(k) be the evaluation of f at X. The eigenlevel set
of f at λ ∈ k is
Lλ(f ) = [
(cid:8)X ∈ Mn(k)d : λ is an eigenvalue of f (X)(cid:9) .
n∈N
In terms of [KV17, HKV18, HKV], eigenlevel sets are free loci of polynomials λ−f , which
have been intensively studied for their implications to domains of noncommutative ra-
tional functions [K-VV09], factorization in a free algebra [HKV18, HKV] and matrix
convexity [BPT13, HKM13, DD-OSS17]. Using Theorem A we derive the following al-
gebraic certificate for inclusion of eigenlevel sets (see Theorem 4.3 for the proof).
Theorem B. Let k be an algebraically closed field of characteristic 0 and f, g ∈ k<x>.
Then eigenlevel sets of f are contained in eigenlevel sets of g if and only if there exist
nonzero a, h ∈ k<x> and p ∈ k[t] such that g = p(h) and f a = ah.
Lastly we turn to noncommutative polynomials describing convex matricial sets.
Let Sn(R) ⊆ Mn(R) denote the subspace of symmetric matrices. In [BM14], a symmetric
f ∈ R<x> with f (0) = 0 is called quasiconvex if for every n ∈ N and positive definite
A ∈ Sn(R), the set
(1.1)
{X ∈ Sn(R)d : A − f (X) is positive semidefinite}
is convex; see [BM14, Subsection 1.1] for the relation with the classical (commutative)
notion of quasiconvexity. Furthermore, in [BM14, Theorem 1.1] the authors showed that
every quasiconvex polynomial is either convex quadratic or minus a sum of hermitian
squares (i.e., −f = Pm h∗
mhm for some hm ∈ R<x>, in which case the set (1.1) equals
Sn(R)d for every A ≻ 0 and n ∈ N).
To relate quasiconvexity more closely to the notion of a free semialgebraic set [HM12,
HKM12] in free real algebraic geometry, we say that a symmetric f ∈ R<x> with
f (0) = 0 is locally quasiconvex if there exists ε > 0 such that the connected component
of
{X ∈ Sn(R)d : λI − f (X) is positive definite}
containing the origin is convex for every n ∈ N and λ ∈ (0, ε).
Theorem C. If f ∈ R<x> is locally quasiconvex, then either −f is a sum of hermitian
squares or f = p(ℓ0 + ℓ2
m) for some p ∈ R[t] and linear ℓ0, . . . , ℓm ∈ R<x>.
1 + · · · + ℓ2
A precise biconditional statement is given in Theorem 5.4 below.
Acknowledgments. The author thanks George Bergman for enlightening correspon-
dence and Igor Klep for valuable suggestions.
2. Preliminaries
We start with reviewing certain notions and technical results from the factorization
theory of P. M. Cohn [Coh06] that will be used throughout the paper. Most of this theory
FREE BERTINI'S THEOREM AND APPLICATIONS
3
is based on the fact that k<x> is a free ideal ring (see e.g.
which will be implicitly used when referring to the existing literature.
[Coh06, Corollary 2.5.2]),
Noncommutative polynomials f1, f2 ∈ k<x> are stably associated [Coh06, Section
0.5] if there exist P1, P2 ∈ GL2(k<x>) such that f2 ⊕ 1 = P1(f1 ⊕ 1)P2. Equivalently
[Coh06, Proposition 0.5.6 and Theorem 2.3.7], there exist g1, g2 ∈ k<x> such that f1, g2
are left coprime, g1, f2 are right coprime, and
(2.1)
f1g1 = g2f2.
Here left (right) coprime refers to the absence of a non-invertible common left (right)
factor. The importance of stable association steams from the fact that factorization of a
noncommutative polynomial into irreducible factors is unique up to stable association of
factors [Coh06, Proposition 3.2.9]. The following finiteness result was first proved by G.
M. Bergman in his doctoral thesis.
Proposition 2.1 (Bergman, [Coh06, Exercise 2.8.8]). Given f ∈ k<x>, there are (up
to a scalar multiple) only finitely many polynomials stably associated to f .
We will also require degree bounds on "witnesses" of stable association in (2.1).
Lemma 2.2. If f1, f2 ∈ k<x> \k are stably associated, then deg f1 = deg f2 and there
exist nonzero g1, g2 ∈ k<x> such that deg gi < deg fi and f1g1 = g2f2.
Proof. Following [Coh06, Section 2.7], continuant polynomials pn ∈ k<y1, . . . , yn> are
recursively defined as
p0 = 1,
p1 = y1,
pn = pn−1yn + pn−2 for n ≥ 2.
By [Coh06, Proposition 2.7.6] there exist α1, α2 ∈ k and a1, . . . , ar ∈ k<x> such that
f1 = α1pr(a1, . . . , ar),
f2 = α2pr(ar, . . . , a1)
and ai are nonconstant for 1 < i < r. If ar = 0, then
pr(a1, . . . , ar) = pr−2(a1, . . . , ar−2),
pr(ar, . . . , a1) = pr−2(ar−2, . . . , a1).
If ar ∈ k \ {0}, then an easy manipulation of the recursive relation for pn yields
pr(a1, . . . , ar) = arpr−1(a1, . . . , ar−1 + 1
ar ),
pr(ar, . . . , a1) = arpr−1(ar−1 + 1
ar
, . . . , a1).
Analogous conclusions hold for a1 ∈ k. Hence there exist β1, β2 ∈ k and nonconstant
b1, . . . , bs ∈ k<x> such that
f1 = β1ps(b1, . . . , bs),
f2 = β2ps(bs, . . . , b1).
Since
ps(b1, . . . , bs)ps−1(bs−1, . . . , b1) = ps−1(b1, . . . , bs−1)ps(bs, . . . , b1)
holds by [Coh06, Lemma 2.7.2] and the degree of a continuant polynomial in nonconstant
arguments equals the sum of degrees of its arguments by the recursive relation,
b1 = 1
β1
ps−1(bs−1, . . . , b1),
b2 = 1
β2
ps−1(b1, . . . , bs−1)
satisfy deg b1 = deg b2 < deg f1 = deg f2.
(cid:3)
4
J. VOL CI C
Remark 2.3. While probably known to the specialists for factorization in free algebras,
Lemma 2.2 implies that checking whether two irreducible polynomials are stably associ-
ated corresponds to solving a (finite) linear system.
Let Ωn = (Ωn
d ) be a tuple of generic n × n matrices whose dn2 entries are
commuting independent variables are viewed as coordinates of the affine space Mn(k)d.
1 , . . . , Ωn
Lemma 2.4 ([HKV, Lemma 2.2]). If f ∈ k<x> is nonconstant, then det f (Ω(n)) is
nonconstant for large enough n ∈ N.
3. Free Bertini's theorem
In this section we prove our first main result (Theorem 3.2). First we show that a
certain linear equation in a free algebra has a unique solution (up to a scalar multiple).
Lemma 3.1. Let f, g ∈ k<x> and assume f is not composite. If nonzero α ∈ k and
b1, b2 ∈ k<x> satisfy
f b1 = b1g,
f b2 = αb2g,
deg b1 = deg b2 < deg f,
then α = 1 and b2 ∈ kb1.
Proof. Since f is not composite, its centralizer in k<x> equals k[f ] by [Ber69, Theorem
5.3]. Therefore its centralizer in k (<x )>, the universal skew field of fractions of k<x>
(see [Coh06, Chapter 7] for more information), equals k(f ) by [Coh06, Theorem 7.9.8
and Proposition 3.2.9]. Since
b−1
1 f b1 = g = α−1b−1
2 f b2,
we have det f (Ω(n)) = α−n det f (Ω(n)) for large enough n by Lemma 2.4, so α = 1 and
b2b−1
1 ∈ k (<x )> commutes with f . Hence there exist univariate coprime polynomials
q1, q2 ∈ k[t] such that b2b−1
1 = q1(f )−1q2(f ), and consequently q1(f )b2 = q2(f )b1. Let
b ∈ k<x> be such that bi = cib for right coprime c1, c2 ∈ k<x>. Then
q1(f )c2 = q2(f )c1,
so q1(f ) and c1 are stably associated. Therefore deg q1(f ) = deg c1 by Lemma 2.2. Since
the degree of q1(f ) is either 0 or at least deg f , and deg c1 ≤ deg b1 < deg f , we conclude
deg c1 = 0. Hence c1, c2 are (nonzero) scalars.
(cid:3)
The proof of free Bertini's theorem is based on Bergman's centralizer theorem [Ber69].
While otherwise inherently different from ours, Stein's proof of (the special case of)
classical Bertini's theorem in two commuting variables [Ste89] also uses "centralizers"
with respect to the Poisson bracket on k[t1, t2].
Theorem 3.2. Let k be the algebraic closure of a field k. The following are equivalent
for f ∈ k<x>\k:
(i) f − λ factors in k<x> for infinitely many λ ∈ k;
(ii) f − λ factors in k<x> for all λ ∈ k;
(iii) the centralizer of f in k<x> is strictly larger than k[f ];
(iv) f is composite over k.
FREE BERTINI'S THEOREM AND APPLICATIONS
5
Proof. Implications (iv)⇒(ii)⇒(i) are clear, and (iii)⇒(iv) is a restatement of [Ber69,
Theorem 5.3]. Thus it suffices to prove (i)⇒(iii).
Let Λ ⊆ k be an infinite set of λ such that f − λ factors in k<x>. For each such λ
there exist nonconstant pλ, qλ ∈ k<x> such that f − λ = pλqλ. Observe that
(3.1)
f pλ = (λ + pλqλ)pλ = pλ(qλpλ + λ)
for all λ ∈ Λ. Since deg pλ < deg f for all λ ∈ Λ, there exists an infinite subset Λ0 ⊆
Λ \ {0} and δ < deg f such that deg pλ = δ for all λ ∈ Λ0. Furthermore, λ + pλqλ, pλ
are left coprime and pλ, qλpλ + λ are right coprime whenever λ 6= 0. Therefore f and
qλpλ + λ are stably associated for every λ ∈ Λ0. By Proposition 2.1, there are (up to
a scalar multiple) only finitely many polynomials stably associated to f . Hence there
exist distinct µ, ν ∈ Λ0 such that qνpν + ν is a scalar multiple of qµpµ + µ. Suppose f
is not composite over k. Then pν is be a scalar multiple of pµ by (3.1) and Lemma 3.1.
However, this is impossible since
Therefore f is composite over k. In particular,
pµqµ − pνqν = ν − µ ∈ k \ {0}.
U := {p ∈ k<x> : deg p < deg f, p(0) = 0, f p − pf = 0} 6= {0}.
But U is a subspace given by equations over k, so U ∩ k<x> 6= {0}. Now (iii) follows
because U ∩ k[f ] = {0} and U is contained in the centralizer of f in k<x>.
(cid:3)
A slightly stronger version holds for homogeneous polynomials.
Corollary 3.3. Let f ∈ k<x> \k be homogeneous. Then f − 1 factors in k<x> if and
only if f = f n
0 for some n > 1 and homogeneous f0 ∈ k<x>.
Proof. If f − 1 factors in k<x>, then f − λdeg f factors in k<x> for every λ ∈ k because
it is up to a linear change of variables equal to f (λx) − λdeg f = λdeg f (f − 1). Therefore
f is composite by Theorem 3.2, and furthermore a power by homogeneity.
(cid:3)
4. Eigenlevel sets
Throughout this section let k be an algebraically closed field of characteristic 0.
Recall the definition of the eigenlevel set of f at λ,
Lλ(f ) = [
(cid:8)X ∈ Mn(k)d : λ is an eigenvalue of f (X)(cid:9) .
n∈N
In the terminology of [HKV18, HKV], Lλ(f ) is the free locus of f − λ. Combined with
existing irreducibility results for free loci of noncommutative polynomials [HKV18, HKV],
free Bertini's theorem becomes a geometric statement about eigenlevel sets.
Corollary 4.1. If f ∈ k<x>\k is not composite, then there exists N ∈ N such that for
all but finitely many λ ∈ k,
(4.1)
(cid:8)X ∈ Mn(k)d : λ is an eigenvalue of f (X)(cid:9)
is a reduced and irreducible hypersurface in Mn(k)g for all n ≥ N.
6
J. VOL CI C
Proof. By Theorem 3.2, there is a cofinite subset Λ of k \ {f (0)} such that f − λ is
irreducible over k for λ ∈ Λ. By [HKV18, Theorem 4.3] for each λ ∈ Λ there exists
Nλ ∈ N such that the hypersurface (4.1) is reduced and irreducible for every n ≥ Nλ.
However, since polynomials f − λ for λ ∈ k only differ in the constant part, it follows by
[HKV18, Remark 3.5 and proof of Lemma 4.2] that one can choose N = Nλ independent
of λ.
(cid:3)
Remark 4.2. Let p1, p2 ∈ k[t]. Then p2 ∈ k[p1] if and only if for every λ1 ∈ k there exists
λ2 ∈ k such that every zero of p1 − λ1 is a zero of p2 − λ2. Indeed, p1 − λ1 has only
simple zeros for infinitely many λ1, in which case {p1 − λ1 = 0} ⊆ {p2 − λ2 = 0} implies
that p1 − λ1 divides p2 − λ2. Then the claim follows from the division algorithm in k[t]
by induction on deg p2.
Theorem 4.3. For f, g ∈ k<x> the following are equivalent:
(i) each eigenlevel set of f is contained in an eigenlevel set of g;
(ii) there exist p ∈ k[t] and nonzero a, h ∈ k<x> such that g = p(h) and f a = ah.
Proof. (ii)⇒(i) By Lemma 2.4,
h(Ωn) = a(Ωn)−1f (Ωn)a(Ωn)
for all large enough n, and thus
det(h(Ωn) − λI) = det(f (Ωn) − λI)
for all λ ∈ k and n ∈ N. Hence Lλ(f ) = Lλ(h) for all λ ∈ k. Since every univariate
polynomial over k factors into linear factors, each eigenlevel set of f is contained in an
eigenlevel set of p(h).
(i)⇒(ii) Assume that f, g are nonconstant. Then f = p1(h1) and g = p2(h2) for some
p1, p2 ∈ k[t] and non-composite h1, h2 ∈ k<x> with h1(0) = 0 = h2(0). By Corollary 4.1
there is a cofinite set Λ ⊆ k such that Lλ(h1) ∩ Mn(k)d and Lλ(h2) ∩ Mn(k)d are reduced
and irreducible hypersurfaces for all λ ∈ Λ and large enough n ∈ N. Since eigenlevel sets
of f are contained in eigenlevel sets of g, there are infinitely many pairs (λ1, λ2) ∈ Λ2
such that Lλ1(h1) = Lλ2(h2). By comparing
det(h1(Ω(n)) − λ1I),
det(h2(Ω(n)) − λ2I)
one can replace h2 with αh2 + β for some α ∈ k \ {0} and β ∈ k (and change p2
accordingly) so that
(4.2)
det(h1(Ω(n)) − λI) = det(h2(Ω(n)) − λI)
for all λ ∈ k and n ∈ N. By [HKV18, Theorem 4.3], h1−λ and h2−λ are stably associated
for all λ ∈ Λ. Let δ = deg h1. By Lemma 2.2 there exist nonzero aλ, bλ ∈ k<x> of degree
less than δ for λ ∈ Λ such that
(4.3)
(h1 − λ)aλ = bλ(h2 − λ).
Since (4.3) is a linear system in (aλ, bλ) with a rational parameter λ, there exist nonzero
A, B ∈ k[t] ⊗ k<x> of degree (with respect to x) less than δ such that
(h1 − t)A = B(h2 − t).
FREE BERTINI'S THEOREM AND APPLICATIONS
7
By looking at the degree of A with respect to t one can find C ∈ k[t] ⊗ k<x> such
that a := A − C(t − h2) ∈ k<x>. Note that a 6= 0 since deg A < δ = deg h2. Letting
b := B − (t − h1)C we obtain
(4.4)
(h1 − t)a = b(h2 − t).
By comparing degrees with respect to t in (4.4) we get b ∈ k<x> and consequently a = b.
For h := p1(h2) we thus have
Finally, since for every λ1 ∈ k there exists λ2 ∈ k such that
f a = p1(h1)a = ap1(h2) = ah.
Lλ1(p1(h2)) = Lλ1(h) = Lλ1(f ) ⊆ Lλ2(g) = Lλ2(p2(h2))
and det h2(Ω(n)) is nonconstant for large n by Lemma 2.4, Remark 4.2 implies p2 = p ◦ p1
for some p ∈ k[t].
(cid:3)
Corollary 4.4. Let f, g ∈ k<x>. Then eigenlevel sets of f and g coincide if and only
if there is a nonzero a ∈ k<x> such that f a = ag.
Proof. If eigenlevel sets of f and g coincide, then f = p1(h1), g = p2(h2) and h1a = ah2
for 0 6= a, h1, h2 ∈ k<x> as in the proof of Theorem 4.3. Furthermore,
Lλ(p1(h1)) = Lλ(p2(h2)) = Lλ(p2(h1))
implies p1 = p2 and therefore f a = ag. For the converse see the proof of (ii)⇒(i) in
Theorem 4.3.
(cid:3)
Example 4.5. Let
f = x1 + x2 + x1x2
2,
g = x1 + x2 + x2
2x1,
a = 1 + x2
1 + x1x2 + x2x1 + x1x2
2x1.
Then f a = ag, so eigenlevel sets of f and g coincide. Note that deg a > deg f . While
f (1 + x2x1) = (1 + x1x2)g
holds, which complies with Lemma 2.2, there is no b ∈ k<x> such that f b = bg and
deg b ≤ deg f .
5. Locally quasiconvex polynomials
On the free R-algebra R<x> there is a unique involution ∗ satisfying x∗
j = xj. A
noncommutative polynomial f ∈ R<x> is symmetric if f ∗ = f . Let Sd = Sn∈N Sn(R)d.
Then f is symmetric if and only if f (X) ∈ S1 for all X ∈ Sd. By A ≻ 0 (resp. A (cid:23) 0)
we denote that A ∈ S1 is positive definite (resp. semidefinite).
Let f ∈ R<x> be symmetric. As in [HM12] (cf. [HKMV]) we define its positivity
domain,
D(f ) = [
Dn(f )
n∈N
where Dn(f ) is the closure of the connected component of
{X ∈ Sn(R)d : f (X) ≻ 0}
8
J. VOL CI C
containing the origin 0d ∈ Sn(R)d. It is known [HM12] that D(f ) is convex (i.e., Dn(f ) is
convex for all n ∈ N) if and only if D(f ) is the solution set of a linear matrix inequality.
We will require the following version of [HKMV, Theorem 1.5].
Proposition 5.1. Let f ∈ R<x> be symmetric and irreducible over C, with f (0) = 0.
If D(1 − f ) is proper and convex, then
(5.1)
f = ℓ0 + ℓ2
1 + · · · + ℓ2
m
for some linear ℓ0, . . . , ℓm ∈ R<x>.
1, . . . , y∗
Proof. Let y = (y1, . . . , yd) and y∗ = (y∗
d) be freely noncommuting variables, and
consider C<y, y∗> with the R-linear involution ∗ sending yj to y∗
j and acting on C as
the complex conjugate. Since f ∈ R<x> is symmetric and irreducible over C, the
noncommutative polynomial f := f (y1 + y∗
d) ∈ C<y, y∗> is hermitian and
irreducible in C<y, y∗>. The positivity domain of 1 − f (see [HKMV]) is the union over
n ∈ N of closures of connected components of
1, . . . , yd + y∗
{(Y, Y ∗) ∈ Mn(C)d × Mn(C)d : I − f (Y1 + Y ∗
1 , . . . , Yd + Y ∗
d ) ≻ 0}
containing the origin. Furthermore, as D(1 − f ) is proper and convex, the standard
embedding of hermitian n × n matrices into symmetric (2n) × (2n) matrices implies that
D(1 − f ) is also proper and convex. Therefore
f = ℓ0 + X
ℓ∗
k
ℓk
k>0
for some linear ℓk ∈ C<y, y∗> by [HKMV, Theorem 1.5]. Note that f = f (x/2, x/2).
Since f is hermitian, ℓ is hermitian, so ℓ0(x/2, x/2) is symmetric. Furthermore,
ℓ∗
k
ℓk = (re ℓk)2 + (im ℓk)2 + i[re ℓk, im ℓk]
for k > 0; since f is symmetric, Pk>0
R<x>. Hence f is of the form (5.1).
ℓk(x/2, x/2)∗ ℓk(x/2, x/2) is a sum of squares in
(cid:3)
Remark 5.2. If f is of the form (5.1), then it is easy to present D(1 − f ) as the solution
set of a linear matrix inequality, so D(1 − f ) is convex.
Lemma 5.3. Let h = ℓ0 + Pk>0 ℓ2
be the coordinates of Rd.
k for some linear ℓk ∈ R<x>, and let t = (t1, . . . , td)
(i) If β > 0, then h + β is a sum of squares in R<x> if and only if h(t) + β is a sum
of squares in R[t].
(ii) If D1(α − h) ⊆ D1(β + h) for some α, β > 0, then β + h is a sum of squares.
Proof. (i) Observe that h+β has a unique representation h+β = v∗Sv, where S ∈ Sd+1(R)
and v∗ = (1, x1, . . . , xd). It is easy to see that h(t) + β is a sum of squares in R[t] if and
only if S (cid:23) 0, which is further equivalent to h + β being a sum of squares in R<x>.
(ii) Since D1(α − h) is convex, we have
h(τ ) ≤ α ⇒ h(τ ) ≥ −β
FREE BERTINI'S THEOREM AND APPLICATIONS
9
for all τ ∈ Rd. That is, an upper bound on h(t) implies a lower bound on h(t), which is
clearly possible only if h(τ ) ≥ −β for all τ ∈ Rd. Since h(t)+β is a quadratic nonnegative
polynomial, it is a sum of squares in R[t]. Now (ii) follows by (i).
(cid:3)
Recall that a symmetric f ∈ R<x> with f (0) = 0 is locally quasiconvex if there
exists ε > 0 such that D(λ − f ) is convex for every λ ∈ (0, ε).
Theorem 5.4. Le f ∈ R<x> be symmetric with f (0) = 0. The following are equivalent:
(i) f is locally quasiconvex;
(ii) D(λ − f ) is convex for every λ > 0;
(iii) −f is a sum of hermitian squares; or
(5.2)
f = p(ℓ0 + ℓ2
1 + · · · + ℓ2
m)
for p ∈ R[t] with p(0) = 0 and linear ℓ0, . . . , ℓm ∈ R<x> satisfying one of the
following:
(a) p(τ ) ≤ 0 for inf Rd(ℓ0 + ℓ2
(b) ℓk = 0 for all k > 0.
1 + · · · + ℓ2
m) < τ < 0,
Proof. (ii)⇒(i) Clear.
(i)⇒(iii) Let ε > 0 be such that D(λ −f ) is convex for every λ ∈ (0, ε). If D(λ −f ) =
Sd for all such λ, then −f (X) is positive semidefinite for every X ∈ Sd, so −f is a sum
of hermitian squares by [Hel02, McC01]. Otherwise we can without loss of generality
assume that D(λ − f ) 6= Sd for λ ∈ (0, ε). If λ − f is irreducible over C for some such λ,
then f is of the form (5.1) by Proposition 5.1, and (a) holds with p = t. If λ −f factors in
R<x> for all λ ∈ (0, ε), then f = p(h) for some p ∈ R[t] and a non-composite h ∈ R<x>
with p(0) = 0 = h(0) by Theorem 3.2. Since f is symmetric, h is also symmetric because
it is unique up to a scalar multiple. Furthermore, −p is not a sum of squares since −f is
not a sum of hermitian squares.
Let us introduce some auxiliary notation. If λ − p attains a negative value on (0, ∞),
let πλ ≥ 0 be such that
(λ − p)[0,πλ] ≥ 0,
∃ε′ > 0 : (λ − p)(πλ,πλ+ε′) < 0.
If λ − p attains a negative value on (−∞, 0), let νλ ≤ 0 be such that
(λ − p)[νλ,0] ≥ 0,
∃ε′ > 0 : (λ − p)(νλ−ε′,νλ) < 0.
Then πλ, νλ are zeros of λ − p and strictly monotone functions in λ, continuous for λ close
to 0.
We distinguish two cases. First suppose that −p is nonnegative on (−∞, 0) or (0, ∞).
By replacing p(t), h with p(−t), −h if necessary, we can assume that −p attains a negative
value on (0, ∞). Then
D(λ − f ) = D(πλ − h)
for all small enough λ > 0. Since h is not composite, πλ − h is irreducible for all but
finitely many λ by Theorem 3.2. Because D(λ − f ) is convex, h is of the form (5.1) by
Proposition 5.1, so (a) holds.
10
J. VOL CI C
Now suppose that −p attains negative values on(−∞, 0) and (0, ∞). Then
(5.3)
D(λ − f ) = D(−νλ + h) ∩ D(πλ − h)
for all small enough λ > 0. Suppose that one the sets D(−νλ + h) and D(πλ − h) is
contained in the other. By replacing p(t), h with p(−t), −h if necessary, we can assume
that D(πλ − h) ⊆ D(−νλ + h). Since h is not composite, πλ − h is irreducible for all but
finitely many λ, so h is of the form (5.1) by convexity of D(λ − f ) and Proposition 5.1.
By Lemma 5.3, −νλ + h is a sum of squares. If µ = − limλ↓0 νλ, then µ + h is nonnegative
on Rd and −p is nonnegative on [−µ, 0], so (a) holds. Finally we are left with the scenario
where the intersection (5.3) is irredundant. Then D(−νλ + h) and D(πλ − h) are both
convex by [HKMV, Corollary 1.2]. By Proposition 5.1 we conclude that h is linear, so
(b) holds.
(iii)⇒(ii) If −f is a sum of hermitian squares, then D(λ − f ) = Sd for every λ > 0.
Otherwise let f be as in (5.2). Then D(λ − f ) equals one of
Sd, D(πλ − h), D(−νλ + h), D(−νλ + h) ∩ D(πλ − h),
depending on the existence of νλ, πλ. Note that D(πλ−h) is always convex by Remark 5.2.
If (b) holds, then D(λ − f ) is convex for λ > 0 since h is linear and intersection of convex
sets is again convex. If (a) holds, then νλ ≤ inf Rd(ℓ0 + Pk>0 ℓ2
k), so D(−νλ + h) = Sd and
D(λ − f ) is convex for λ > 0.
(cid:3)
Remark 5.5. Few comments on the condition (a) in Theorem 5.4 are in order. Let
µ = inf
Rd
(ℓ0 + ℓ2
1 + · · · + ℓ2
m).
Then µ > −∞ if and only if ℓ0 lies in the linear span of ℓ1, . . . , ℓm; more precisely, if
ℓ1, . . . , ℓm are linearly independent and ℓ0 = α1ℓ1 + · · ·+ αmℓm, then −4µ = α2
1 + · · ·+ α2
m.
This follows from considering ℓ0 + ℓ2
m + µ = v∗Sv for v∗ = (1, x1, . . . , xd) and
S (cid:23) 0 as in the proof of Lemma 5.3(i). Furthermore, using an algebraic certificate for
nonnegativity [Mar08, Prop 2.7.3], the condition (a) can also be stated as follows. Let
S ⊂ R[t] be the convex cone of sums of (two) squares. If µ = −∞, then
1 + · · · + ℓ2
sup
(−∞,0]
p = p(0) = 0 ⇐⇒ p ∈ t(S − tS);
and if −∞ < µ ≤ 0, then
max
[µ,0]
p = p(0) = 0 ⇐⇒ p ∈ t(S − tS + (t − µ)(S − tS)).
Remark 5.6. Another aspect of Theorem 5.4 is the following. Proposition 5.1 states that
every irreducible symmetric polynomial with a convex positivity domain is quadratic
(and concave). On the other hand, there is no shortage of reducible symmetric polyno-
mials that contain a factor of degree at least 3 and have convex positivity domain; see
[HKMV, Sections 5 and 6]. However, if the constant term of such a polynomial is slightly
perturbed, then its positivity domain is no longer convex by Theorem 5.4.
FREE BERTINI'S THEOREM AND APPLICATIONS
11
References
[BM14] S. Balasubramanian, S. McCullough: Quasi-convex free polynomials, Proc. Amer. Math. Soc.
142 (2014) 2581 -- 2591. 2
[Ber69] G. M. Bergman: Centralizers in free associative algebras, Trans. Amer. Math. Soc. 137 (1969)
327 -- 344. 4, 5
[BDN09] A. Bodin, P. D`ebes, S. Najib: Irreducibility of hypersurfaces, Comm. Algebra 37 (2009) 1884 --
1900. 1
[BPT13] G. Blekherman, P. A. Parrilo, R. R. Thomas (eds.): Semidefinite optimization and convex
algebraic geometry, MOS-SIAM Ser. Optim. 13, SIAM, Philadelphia, PA, 2013. 1, 2
[Coh06] P. M. Cohn: Free Ideal Rings and Localization in General Rings, New Mathematical Mono-
graphs 3, Cambridge University Press, Cambridge, 2006. 2, 3, 4
[DD-OSS17] K. R. Davidson, A. Dor-On, O. M. Shalit, B. Solel: Dilations, Inclusions of Matrix Convex
Sets, and Completely Positive Maps, Int. Math. Res. Not. IMRN (2017) 4069 -- 4130. 2
[Hel02] J. W. Helton: "Positive noncommutative polynomials are sums of squares", Ann. Math. 156
(2002) 675 -- 694. 9
[HKM13] J. W. Helton, I. Klep, S. McCullough: The matricial relaxation of a linear matrix inequality,
Math. Program. 138 (2013) 401 -- 445. 2
[HKM12] J. W. Helton, I. Klep, and S. McCullough: Free analysis, convexity and LMI domains, Math-
ematical methods in systems, optimization, and control, 195 -- 219, Oper. Theory Adv. Appl. 222,
Birkhauser/Springer Basel AG, Basel, 2012. 1, 2
[HKMV] J. W. Helton, I. Klep, S. McCullough, J. Volcic: Noncommutative polynomials describing
convex sets, preprint arXiv:1808.06669. 7, 8, 10
[HKV18] J.W. Helton, I. Klep, J. Volcic: Geometry of free loci and factorization of noncommutative
polynomials, Adv. Math. 331 (2018) 589 -- 626. 2, 5, 6
[HKV] J.W. Helton, I. Klep, J. Volcic: Factorization of noncommutative polynomials and Nullstel-
lensatze for the free algebra, preprint arXiv:1907.04328. 2, 4, 5
[HM12] J.W. Helton, S. McCullough: Every convex free basic semi-algebraic set has an LMI represen-
tation, Ann. of Math. (2) 176 (2012) 979 -- 1013. 2, 7, 8
[K-VV09] D. S. Kalyuzhnyi-Verbovetskyi, V. Vinnikov: Singularities of rational functions and minimal
factorizations: the noncommutative and the commutative setting, Linear Algebra Appl. 430 (2009)
869 -- 889. 2
[K-VV14] D. S. Kalyuzhnyi-Verbovetskyi, V. Vinnikov: Foundations of free noncommutative function
theory, Mathematical Surveys and Monographs 199, American Mathematical Society, Providence
RI, 2014. 1
[Kle98] S. L. Kleiman: Bertini and his two fundamental theorems, Studies in the history of modern
mathematics III, Rend. Circ. Mat. Palermo (2) Suppl. No. 55 (1998) 9 -- 37. 1
[KV17] I. Klep, J. Volcic: Free loci of matrix pencils and domains of noncommutative rational functions,
Comment. Math. Helv. 92 (2017) 105 -- 130. 2
[Mar08] M. Marshall: Positive polynomials and sums of squares, Mathematical Surveys and Monographs
146, American Mathematical Society, 2008. 10
[McC01] S. McCullough: Factorization of operator-valued polynomials in several non-commuting vari-
ables, Linear Algebra Appl. 326 (2001) 193 -- 203. 9
[Sha94] I. R. Shafarevich: Basic algebraic geometry 1. Varieties in projective space, 2nd edition, Springer-
Verlag, Berlin, 1994. 1
[Sch00] A. Schinzel: Polynomials with special regard to reducibility, Encyclopedia of Mathematics and
its Applications 77, Cambridge University Press, Cambridge, 2000. 1
[Ste89] Y. Stein: The total reducibility order of a polynomial in two variables, Israel J. Math. 68 (1989)
109 -- 122. 4
Jurij Volcic, Department of Mathematics, Texas A&M University
E-mail address: [email protected]
|
1704.03075 | 1 | 1704 | 2017-04-10T22:33:32 | A BV-algebra Structure on Hochschild Cohomology of the Group Ring of Finitely Generated Abelian Groups | [
"math.RA"
] | We study a Batalin-Vilkovisky algebra structure on the Hochschild cohomology of the group ring of finitely generated abelian groups. The Batalin-Vilkovisky algebra structure for finite abelian groups comes from the fact that the group ring of finite groups is a symmetric algebra, and the Batalin-Vilkovisky algebra structure for free abelian groups of finite rank comes from the fact that its group ring is a Calabi-Yau algebra. | math.RA | math |
A BV-algebra Structure on Hochschild
Cohomology of the Group Ring of Finitely
Generated Abelian Groups
Andr´es Angel∗1 and Diego Duarte†1
1Department of Mathematics, Universidad de los Andes, Bogot´a, Colombia,
Carrera 1 No 18A - 12
April 9, 2017
Abstract
We study a Batalin-Vilkovisky algebra structure on the Hochschild cohomology of the
group ring of finitely generated abelian groups. The Batalin-Vilkovisky algebra structure
for finite abelian groups comes from the fact that the group ring of finite groups is a
symmetric algebra, and the Batalin-Vilkovisky algebra structure for free abelian groups
of finite rank comes from the fact that its group ring is a Calabi-Yau algebra.
1 Introduction
The Hochschild (co)homology of associative algebras has been extensively studied since its first
appearance in 1945 with the paper On The Cohomology Groups of an Associative Algebra by
Gerard Hochschild [11]. There is a rich algebraic structure on the Hochschild cohomology of
an associative algebra. It is a graded algebra given by the cup product. In [8], Gerstenhaber
proves that the cup product is commutative, and even more that exist a Lie bracket that endows
HH ∗(A, A) with a structure of Lie algebra. These two structures satisfy some compatibility
conditions that are now known to define a Gerstenhaber algebra.
In [18], Tradler proves that if A is a symmetric algebra up to homotopy then HH ∗(A, A)
is a Batalin-Vilkovisky algebra. In [14], Menichi presents another proof for Tradler's result for
E-mail addresses: [email protected] (A. Angel), [email protected] (D. Duarte).
2010 Mathematics subject classification: 16E40, 16S34, 55U25.
Key words and phrases: Hochschild cohomology, group ring, finitely generated abelian group, Batalin-
Vilkovisky structure.
∗A. Angel is supported in part by the FAPA funds from Vicerrector´ıa de Investigaciones de la Universidad
de los Andes
†D. Duarte is supported by Fondo de Investigaciones de la Facultad de Ciencias de la Universidad de los
Andes. Convocatoria 2017-I para la financiaci´on de proyectos de investigaci´on categor´ıa estudiantes de doctorado
candidatos.
1
symmetric differential graded algebras. These structures play an important role due to its
connection with string topology as can be found in [2], [5], [4], [6], [7], [15], [19] and [20].
Given a symmetric algebra, such as a group ring of a finite group, the Batalin-Vilkovisky
structure depends on the duality isomorphism, by using different symmetric forms we get
different Batalin-Vilkovisky structures with the same underlying Gesternhaber algebra. The
Batalin-Vilkovisky algebra structure on the Hochschild cohomology of cyclic groups of prime
order over Fp was calculated by Yang [21] using the isomorphism between the group ring and
the truncated polynomial ring. However, the symmetric form used on those calculations do
not correspond to the canonical form over group rings. For cyclic groups using the canonical
symmetric form, we get
Theorem. Let R be an integral domain with char(R) ∤ n and A = R[Z/nZ]. Then as a
BV-algebra
HH ∗(A; A) = R[x, z]/(xn − 1, nz)
∆(a) = 0 ∀a ∈ HH ∗(A; A)
where x = 0 and z = 2.
Theorem. Let R be a commutative ring with char(R) = p > 0 and A = R[Z/nZ] with n = mp.
If p 6= 2, or p = 2 and m is even. Then as a BV-algebra
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2)
∆(zkyrxl) = r(l − 1)zkxl−1
If p = 2 and m is odd. Then as a BV-algebra
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2 − xn−2z)
∆(zkyrxl) = r(l − 1)zkxl−1
where x = 0, y = 1 and z = 2.
The aim of this paper is to present a Batalin-Vilkovisky algebra structure on the Hochschild
cohomology of the group ring of finitely generated abelian groups. In order to achieve this goal,
we study the behavior of the Batalin-Vilkovisky structure for tensor products. Over fields in
[13], Le and Zhou prove that the Kunneth formula for Hochschild cohomology is an isomorphism
of Gerstenhaber algebras if at least one of the algebras is finite dimensional, and if the algebras
are symmetric is an isomorphism of Batalin-Vilkovisky algebras. In section 3, we extend their
result for a general class of rings. As a particular case over the integers, we get the following
new result
Theorem. Let A = Z[Z/nZ] and B = Z[Z/mZ] with n = km. Then, as a BV-algebra
HH ∗(A ⊗ B; A ⊗ B) =
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2)
∆(xitjalbrcs) = sxi−1tjalbr((i − 1)b − jka)
2
in all cases except when m is even and k is odd, in which case we get
HH ∗(A ⊗ B; A ⊗ B) =
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2 − m
2 xn−2ab(b + ka))
∆(xitjalbrcs) = sxi−1tjalbr((i − 1)b − jka)
where x = t = 0, a = b = 2 and c = 3.
Notice that the tensor product of the corresponding Hochschild cohomology rings gives a
trivial BV-structure. Nevertheless, the Hochschild cohomology of the tensor product gives a
highly non-trivial BV-structure.
When the algebra is not symmetric but satisfies some sort of Poincar´e duality. Ginzburg
[9] and Menichi [14] prove that HH ∗(A; A) is also a Batalin-Vilkovisky algebra by transferring
the Connes B-operator through the isomorphism between Hochschild homology and Hochshild
cohomology. For the tensor product of two such algebras, we prove that if the algebras satisfy
some finiteness condition on their resolutions (5), there is also an isomorphism of Batalin-
Vilkovisky algebras between the Hochschild cohomology of the tensor product and the tensor
product of their cohomologies. In particular, for free abelian groups of finite rank, we have
Theorem. As BV-algebras,
HH ∗(R[Zn]; R[Zn]) = R[x1, x−1
n ] ⊗ Λ(y1, . . . , yn)
∆(xi1
1 · · · xin
n yr1
1 · · · yrn
n ) =
(−1)
nXk=1
1 , . . . , xn, x−1
r1+···+rk−1 rk(ik − 1)xi1
1 · · · xik−1
k
· · · xin
n yr1
k · · · yrn
n
1 · · ·dyrk
where xi = x−1
i
= 0 and yi = 1 for 1 ≤ i ≤ n.
2 Hochschild (Co)homology
Let A be a R-projective R-algebra with unit and R be a commutative ring. Denote by Aop
the opposite algebra of A and by Ae the enveloping algebra A ⊗ Aop. Recall that any left and
right A-module can be considered as a left, or right, Ae-module. Let M be an Ae-module. The
Hochschild homology of A with coefficients in M is
HH∗(A; M) := T orAe
∗ (A; M)
and the Hochschild cohomology of A with coefficients in M is
HH ∗(A; M) := Ext∗
Ae(A; M)
Besides the additive structure, the Hochschild cohomology HH ∗(A; A) has a graded algebra
structure induced from the cup product defined over cochains by
(f ⌣ g)(a1, . . . , ak+j) = f (a1, . . . , ak)g(ak+1, . . . , ak+j)
(1)
where f ∈ Hom( ¯Ak, A) and g ∈ Hom( ¯Aj, A).
3
Since Hochschild cohomology can be computed by using different resolutions. A more
µ
−→ A be an Ae-projective
general notion of the cup product can be defined as follows. Let P(A)
resolution of A, and let ∆ : P(A) → P(A) ⊗
P(A) be a diagonal approximation map, i.e., an
A
Ae-chain map such that (µ ⊗ µ) ◦ ∆ = µ. If M and N are Ae-modules the Hochschild cup
product is defined by
⌣: HH ∗(A; M) ⊗ HH ∗(A; N) −→ HH ∗(A; M ⊗
A
α ⊗ β 7−→ (−1)αβ(α ⊗
A
N)
β)∆
(2)
Notice that if M = A the cup product endows HH ∗(A; N) with the structure of HH ∗(A; A)-
module
HH ∗(A; A) ⊗ HH ∗(A; N) ⌣ /
/ HH ∗(A; A ⊗
A
N)
and if M = A = N the cup product is a product in HH ∗(A; A)
∼= /
/ HH ∗(A; N)
HH ∗(A; A) ⊗ HH ∗(A; A) ⌣ /
/ HH ∗(A; A ⊗
A
A)
∼= /
/ HH ∗(A; A)
that will coincide with the one defined over the bar resolution.
Remark 1. The diagonal approximation map that recovers the cup product defined on the bar
resolution is given by
∆B(A) : B(A) −→ B(A) ⊗
A
B(A)
a0 ⊗ · · · ⊗ an+1 7−→
nXi=0
a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗
A
1 ⊗ ai+1 ⊗ · · · ⊗ an+1
(3)
Lemma 2.1. Let A be a R-projective R-algebra. Then any Hochschild diagonal approximation
map calculates the cup product in HH ∗(A; A).
Proof. Let P(A)
µ
−→ A be an Ae-projective resolution of A, and let ∆ : P(A) → P(A) ⊗
A
P(A)
be a diagonal approximation map. We only need to prove that P(A) ⊗
A
Ae-projective resolution. Since
P(A)
µ⊗µ
−−→ A is an
(cid:18)P(A) ⊗
A
P(A)(cid:19)n
= Mi+j=n
Pi ⊗
A
Pj
and each Pi is Ae-projective (Pi⊕Q ∼= ⊕Ae), it suffices to show that Ae⊗
hypothesis, A is R-projective and Ae ⊗
A
Ae ∼= Ae ⊗A as Ae-modules then Ae ⊗
A
A
Ae is Ae-projective. By
Ae is Ae-projective.
Now, to see that the complex is acyclic, notice that each Pi is A-projective because A is
R-projective, and H∗(P(A)) ∼= A which is A-free then
T orA
p (Hs(P(A)); Ht(P(A))) = 0
∀p ≥ 1
and
T orA
0 (Hs(P(A)); Ht(P(A))) = Hs(P) ⊗
A
4
Ht(P) =
A ⊗
A
0
A if s = t = 0
otherwise
Applying the Kunneth spectral sequence, we get
H∗(P ⊗A P) ∼= A ⊗
A
A ∼= A
Since P(A)
µ
−→ A and P(A) ⊗
A
P(A)
µ⊗µ
−−→ A are both Ae-projective resolutions of A, by the com-
parison theorem, ∆ : P(A) → P(A) ⊗
A
P(A) exists and it is unique up to homotopy. Therefore,
the usual cup product given by the bar resolution (1) coincides with any other cup product
given by different resolutions and diagonal approximation maps.
Recall that HH ∗(A; A) acts on HH∗(A; A). For n ≥ m, f ∈ Hom( ¯Am, A) and a1 ⊗ · · · ⊗
an ⊗ a ∈ ¯An ⊗ A the action is given by
(a1 ⊗ · · · ⊗ an ⊗ a) · f = (−1)nmam+1 ⊗ · · · ⊗ an ⊗ af (a1 ⊗ · · · ⊗ am)
This action can be calculated over any resolution as follows
Proposition 2.2. Let A be a R-projective R-algebra and ∆ be any diagonal approximation
map. The action of Hochschild cohomology on Hochschild homology is given by
ρ : HHn(A; A) ⊗ HH m(A; A) −→ HHn−m(A; A)
(x ⊗
Ae
a) ⊗ f 7−→ (−1)nm(f ⊗
A
id)∆(x) ⊗
Ae
a
Proof. Notice that f is a cochain iff the map f : P(A) → A is a chain map. Then ρ is well
defined because (f ⊗
id)∆ : P(A) → P(A) is a chain map. Since any approximation map is
A
unique up to homotopy, it is sufficient to prove that the formula coincided with the one given
for the bar resolution. Let 1 ⊗ a1 ⊗ · · · ⊗ an ⊗ 1 ⊗
Ae
a ∈ Bn(A) and f ∈ HomAe(Bm(A), A)
(−1)nm(f ⊗
A
id)∆(1 ⊗ a1 ⊗ · · · ⊗ an ⊗ 1) ⊗
Ae
a
= (−1)nm(f ⊗
A
1 ⊗ a1 ⊗ · · · ⊗ ai ⊗ 1 ⊗
A
id) nXi=0
1 ⊗ ai+1 ⊗ · · · ⊗ an ⊗ 1! ⊗
Ae
a
= (−1)nmf (1 ⊗ a1 ⊗ · · · ⊗ am ⊗ 1) ⊗
A
1 ⊗ am+1 ⊗ · · · ⊗ an ⊗ 1 ⊗
Ae
a
In [8], Gerstenhaber proves that the cup product on Hochschild cohomology is graded com-
mutative and that there exists a Lie bracket that endows HH ∗(A; A) with a structure of Lie
algebra. The Gerstenhaber bracket on HH ∗(A; A) using the bar resolution is defined as follows
{f, g} = f ◦ g − (−1)(f −1)(g−1)g ◦ f
where ◦ is defined by
(f ◦ g)(a1 ⊗ · · · ⊗ ak+j−1) =
(−1)(j−1)(i−1)f (a1 ⊗ · · · ⊗ ai−1 ⊗ g(ai ⊗ · · · ⊗ ai+j−1) ⊗ ai+j ⊗ · · · ⊗ ak+j−1)
kXi=1
The cup product and the bracket satisfy the following compatibility conditions.
5
Definition 2.1. A Gerstenhaber algebra is a graded commutative algebra A with a linear map
{−, −} : Ai ⊗ Aj → Ai+j−1 of degree −1 such that
1. The bracket {−, −} endows A with a structure of graded Lie algebra of degree 1, i.e., for
all a, b and c ∈ A
{a, b} = −(−1)(a+1)(b+1) {b, a}
{a, {b, c}} = {{a, b} , c} + (−1)(a+1)(b+1) {b, {a, c}}
2. The product and the Lie bracket satisfy the Poisson identity, i.e., for all a, b and c ∈ A
{a, bc} = {a, b} c + (−1)(a+1)bb {a, c}
If there is a differential of degree −1 of a Gerstenhaber algebra such that the Gerstenhaber
bracket is the obstruction of the operator to be a graded derivation, then the Gerstenhaber
algebra is called a Batalin-Vilkovisky algebra.
Definition 2.2. A Batalin-Vilkovisky algebra is a Gerstenhaber algebra A with a linear map
of degree −1, ∆ : Ai → Ai−1 such that ∆ ◦ ∆ = 0 and
{a, b} = −(−1)a(∆(ab) − ∆(a)b − (−1)aa∆(b))
for all a and b ∈ A.
The way to construct BV-structures on Hochschild cohomology is by dualizing or transfer-
ring the Connes B-operator.
Definition 2.3. Let A be a unital algebra. The Connes B-operator is a map on Hochschild
homology defined on normalized chains as follows
Bn : ¯An ⊗ A −→ ¯An+1 ⊗ A
Bn(a1 ⊗ · · · ⊗ an ⊗ a) =
nXi=0
The dual of this operator
(−1)inai ⊗ · · · ⊗ an ⊗ a ⊗ a1 ⊗ · · · ⊗ ai−1 ⊗ 1
(4)
B∨ : Hom( ¯A∗+1 ⊗ A, R) → Hom( ¯A∗ ⊗ A, R)
defines by adjunction an operator on Hom( ¯A∗, A∨) ∼= Hom( ¯A∗⊗A, R), where A∨ = Hom(A, R)
When A is a symmetric algebra the non-degenerate bilinear form of A induces a chain complex
isomorphism
Hom( ¯A∗, A∨) ∼= Hom( ¯A∗, A)
which defines a BV-operator, ∆, on the Hochschild cochains.
Definition 2.4. Let A be a finitely generated projective R-algebra. A is called a Frobenius
algebra if there exists an isomorphism of left, or right, A-modules
If the isomorphism is of Ae-modules, A is called a symmetric algebra.
ϕ : A
∼=−→ A∨ = HomR(A, R)
6
Remark 2. Given a Frobenius algebra A, it can be defined a non-degenerate bilinear form,
h·, ·i : A ⊗ A −→ R
ϕ(b)(a) = ϕ(1)(ab)
ϕ(a)(b) = ϕ(1)(ab)
if ϕ is a left isomorphism
if ϕ is a right isomorphism
as follows
ha, bi :=
Notice that the pairing is associative
hab, ci = ϕ(c)(ab) = ϕ(1)(abc) = ϕ(bc)(a) = ha, bci
hab, ci = ϕ(ab)(c) = ϕ(1)(abc) = ϕ(a)(bc) = ha, bci
(ϕ left isomorphism)
(ϕ right isomorphism)
Moreover, if ϕ is a two sided isomorphism the pairing is symmetric
ha, bi = ϕ(a)(b) = ϕ(1)(ab) = ϕ(1)(ba) = ϕ(b)(a) = hb, ai
From now on, an associative nonsingular bilinear form will be called a Frobenius form. As
in the case over fields, Frobenius algebras over commutative rings can be characterized by
Frobenius forms.
Proposition 2.3. A finitely generated projective R-algebra A is Frobenius if and only if there
exists a non-degenerate bilinear form, and it is symmetric if and only if there exists such a form
which is also symmetric.
Example 2.1. Let R be a commutative ring. If G is a finite group then the group ring R [G]
is a symmetric algebra with Frobenius form given by
h·, ·i : R [G] × R [G] −→ R,
hg, hi =
1 if g = h−1
0 otherwise
Notice that the Frobenius form of the group ring R [G] could be defined by using the canonical
augmentation of the group ring,
where
ε : R [G] −→ R,
ha, bi := ε(ab)
εXg∈G
αgg = αe
In the case when A is a symmetric algebra, the BV-operator, ∆, is defined as follows
Proposition 2.4. The operator ∆ : Hom( ¯Am+1, A) → Hom( ¯Am, A) is given by
∆(f )(a1, . . . , am) =
NXj=1
mXi=0
(−1)imh1, f (ai, . . . , an, aj, a1, . . . , ai−1)iaj ∨
where {a1, . . . , aN } is a basis of A and {a1∨, . . . , aN ∨} is the dual basis with respect to the
Frobenius form.
7
In [18], Tradler proves that ∆ induces a BV-structure on HH ∗(A; A), which furthermore
induces the Gerstenhaber structure of HH ∗(A; A).
Theorem 2.5 ([18], [15]). Let A be a symmetric R-algebra. Then HH ∗(A, A) is a BV-algebra
with ∆ given by the dual of the Connes operator.
When the algebra is not symmetric but satisfies some sort of Poincar´e duality. It is posible to
obtain a BV-algebra structure on Hochschild cohomology by transferring the Connes operator.
Theorem 2.6. [[9], [14]] Let a ∈ HHn(A, A) such that
ρa : HH ∗(A; A) −→ HHn−∗(A; A)
b 7−→ ρ(a ⊗ b)
is an isomorphism. If B(a) = 0 then HH ∗(A, A) is a BV-algebra with ∆ given by ∆a := ρ−1
a Bρa.
3 Hochschild (Co)homology for Tensor Products
In [13], Le and Zhou prove the following
Theorem 3.1 ([13] Theorem 3.3). Let R be a field and A and B be two R-algebras such that
one of them is finite dimensional. Then there is an isomorphism of Gerstenhaber algebras
HH ∗(A ⊗ B; A ⊗ B) ∼= HH ∗(A; A) ⊗ HH ∗(B; B)
If furthermore, A and B are finite dimensional symmetric algebras, the above isomorphism
becomes an isomorphism of Batalin-Vilkovisky algebras.
In this section, we extend their result for a general class of rings and present an analogous
for algebras that satisfy some sort of Poincar´e duality.
Proposition 3.2. Let A and B be R-projective R-algebras with R a commutative ring. Suppose
that P(A) → A is an Ae-projective resolution of A and P(B) → B is a Be-projective resolution
of B. Then
P(A ⊗ B) := P(A) ⊗ P(B) −→ A ⊗ B
is an (A ⊗ B)e-projective resolution of A ⊗ B.
Proof. Since
Pn(A ⊗ B) = Mi+j=n
Pi(A) ⊗ Pj(B)
and Ae ⊗ Be ∼= (A ⊗ B)e, P(A) ⊗ P(B) → A ⊗ B is an (A ⊗ B)e-projective complex of A ⊗ B.
It only remains to check that the complex is acyclic. Since H∗(P(A)) ∼= A and H∗(P(B)) ∼= B
which are R-projective. Then
T orR
p (Hs(P(A)); Ht(P(B))) = 0
∀p ≥ 1
and
T orR
0 (Hs(P(A)); Ht(P(B))) = Hs(P(A)) ⊗ Ht(P(B)) =
8
A ⊗ B if s = t = 0
0
otherwise
Applying the Kunneth spectral sequence, we have
H∗(P(A) ⊗ P(B)) ∼= A ⊗ B
Therefore, P(A) ⊗ P(B) → A ⊗ B is an (A ⊗ B)e-projective resolution of A ⊗ B.
Proposition 3.3. The following map is an isomorphism of complexes
τ : (P(A) ⊗
A
P(A)) ⊗ (P(B) ⊗
B
P(B)) −→ P(A ⊗ B) ⊗
A⊗B
P(A ⊗ B)
a1 ⊗
A
a2 ⊗ b1 ⊗
B
b2 7−→ (−1)a2b1a1 ⊗ b1 ⊗
A⊗B
a2 ⊗ b2
Proof. Let a1, a2 ∈ P(A) and b1, b2 ∈ P(B) with a1 = i, a2 = j, b1 = k and b2 = l.
τ δn(a1 ⊗
A
a2 ⊗ b1⊗
B
b2) = τ (∂A
b2
a2) ⊗ b1 ⊗
B
i+j(a1 ⊗
A
+ (−1)i+ja1 ⊗
k+l(b1 ⊗
B
A
dA
a2 + (−1)ia1 ⊗
j (a2)) ⊗ b1 ⊗
B
A
i (a1) ⊗
A
a2 ⊗ ∂B
b2))
= τ ((dA
b2
+ (−1)i+ja1 ⊗
A
a2 ⊗ (dB
b2 + (−1)kb1 ⊗
B
dB
l (b2)))
= (−1)kjdA
i (a1) ⊗ b1 ⊗
A⊗B
k (b1) ⊗
B
a2 ⊗ b2
+ (−1)i+kja1 ⊗ dB
+ (−1)i+k(j−1)a1 ⊗ b1 ⊗
A⊗B
k (b1) ⊗
A⊗B
+ (−1)i+j+k(j+1)a1 ⊗ b1 ⊗
A⊗B
dA
j (a2) ⊗ b2
a2 ⊗ b2
a2 ⊗ dB
l (b2)
= (−1)kj((dA
i (a1) ⊗ b1 + (−1)ia1 ⊗ dB
k (b1)) ⊗
A⊗B
a2 ⊗ b2)
j (a2) ⊗ b2 + (−1)ja2 ⊗ dB
l (b2))
= (−1)kj(d⊗
(dA
+ (−1)kj+i+ka1 ⊗ b1 ⊗
A⊗B
i+k(a1 ⊗ b1) ⊗
A⊗B
d⊗
j+l(a2 ⊗ b2))
a2 ⊗ b2
= (−1)kj∂⊗
+ (−1)i+ka1 ⊗ b1 ⊗
A⊗B
n (a1 ⊗ b1 ⊗
A⊗B
b2)
a2 ⊗ b1 ⊗
B
n τ (a1 ⊗
A
= ∂⊗
a2 ⊗ b2)
Therefore, τ is a map of complexes and it is clear that is an isomorphism in each degree, since
the inverse of τ is τ itself.
Proposition 3.4. Let ∆A : P(A) → P(A) ⊗
A
approximation maps. Then
P(A) and ∆B : P(B) → P(B) ⊗
B
P(B) be diagonal
∆ : P(A ⊗ B) ∆A⊗∆B
−−−−−→ (P(A) ⊗
A
P(A)) ⊗ (P(B) ⊗
B
P(B)) τ−→ P(A ⊗ B) ⊗
A⊗B
P(A ⊗ B)
is a diagonal approximation map for A ⊗ B.
9
Proof. Let a ∈ P(A) and b ∈ P(B) with a = i and b = j
i+j∆i+j(a ⊗ b) = ∂⊗
∂⊗
i+jτ (∆A
i ∆A
= τ (∂A
i−1dA
= τ (∆A
= ∆i+j−1(dA
= ∆i+j−1d⊗
j (b)) = τ δi+j(∆A
i (a) ⊗ ∆B
j (b) + (−1)i∆A
i (a) ⊗ ∆B
j (b) + (−1)i∆A
i (a) ⊗ ∆B
i (a) ⊗ b + (−1)ia ⊗ ∆B
i+j(a ⊗ b)
i (a) ⊗ ∆B
j ∆B
i (a) ⊗ ∂B
i (a) ⊗ ∆B
j−1dB
j (b))
j (b))
j (b))
j−1dB
j (b))
For a = b = 0, we have
((µA ⊗ µB) ⊗
A⊗B
(µA ⊗ µB))τ (∆A
0 ⊗ ∆B
0 )(a ⊗ b) = (µA ⊗ µB))(a ⊗ b)
Theorem 3.5. Let A and B be R-projective R-algebras with R a commutative hereditary ring.
Suppose that P(A) → A is a resolution of A of finitely generated projective Ae-modules and
P(B) → B is a Be-resolution of B such that
Hom(A⊗B)e(P(A ⊗ B), A ⊗ B) ∼= HomAe(P(A), A) ⊗ HomBe(P(B), B)
(5)
Then
HH ∗(A; A) ⊗ HH ∗(B; B) ֒→ HH ∗(A ⊗ B; A ⊗ B)
is an injection of graded algebras.
Proof. By Kunneth theorem, there is an injective map of modules. Let ∆A : P(A) → P(A) ⊗
A
P(A) and ∆B : P(B) → P(B) ⊗
P(B) be diagonal approximation maps. By proposition 3.4,
B
∆ = τ (∆A ⊗ ∆B) is a diagonal approximation map for P(A ⊗ B). Let f, f ′ ∈ HomAe(P(A), A)
and g, g′ ∈ HomBe(P(B), B). Notice that the following diagram commutes
P(A) ⊗ P(B)
∆A⊗∆B
P(A) ⊗
A
(−1)f ′gf ⊗
A
P(A) ⊗ P(B) ⊗
B
P(B)
f ′⊗g⊗
B
g′
A ⊗
A
A ⊗ B ⊗
B
B
τ
τ
/ (P(A ⊗ B)) ⊗
A⊗B
(P(A ⊗ B))
f ⊗g ⊗
f ′⊗g′
A⊗B
/ A ⊗ B ⊗
A⊗B
A ⊗ B
Therefore,
((f ⊗ g) ⌣ (f ′ ⊗ g′)) = (−1)(f +g)(f ′+g′)((f ⊗ g) ⊗
A⊗B
f ′)∆A) ⊗ ((g ⊗
B
= (−1)f ′g+f f ′+gg′((f ⊗
(f ′ ⊗ g′))∆
A
g′)∆B)
= (−1)f ′g(f ⌣A f ′) ⊗ (g ⌣B g′)
10
/
/
Corollary 3.6. Under the same hypothesis as in theorem 3.5, if HH ∗(A; A), or H ∗(B; B), is
R-projective. Then
HH ∗(A; A) ⊗ HH ∗(B; B) ∼= HH ∗(A ⊗ B; A ⊗ B)
as graded algebras.
Proof. By Kunneth Theorem, there is an isomorphim of modules
HH n(A ⊗ B; A ⊗ B) ∼= Mr+s=n
HH r(A; A) ⊗ HH s(B; B)
⊕ Mr+s=n+1
T orR
1 (HH r(A; A), HH s(B; B))
Since HH ∗(A; A), or H ∗(B; B), is R-projective, the proof of Theorem 3.5 extends to an iso-
morphism of graded algebras.
Definition 3.1. A (p, q)-shuffle is a sequence of integers
represented by a permutation σ ∈ Sp+q, such that
[i1 · · · ipj1 · · · jq]
σ(1) = i1 < · · · < ip = σ(p) and σ(p + 1) = j1 < · · · < jq = σ(p + q)
The sign of a (p, q)-shuffle is defined by
σ := {(i, j)1 ≤ i < j ≤ p + q and σ(i) > σ(j)}
The set of (p, q)-shuffles will be denoted by Sp,q.
Definition 3.2. The Alexander-Whitney map AW : B(A ⊗ B) → B(A) ⊗ B(B) is defined as
follows
AWr(1 ⊗ 1 ⊗ a1 ⊗ b1 ⊗ · · · ⊗ ar ⊗ br ⊗ 1 ⊗ 1) =
AW0(a1 ⊗ b1 ⊗ a2 ⊗ b2) = a1 ⊗ a2 ⊗ b1 ⊗ b2
(−1)t(r−t)a1a2 · · · at ⊗ at+1 ⊗ · · · ⊗ ar ⊗ 1 ⊗ 1 ⊗ b1 ⊗ · · · ⊗ bt ⊗ bt+1 · · · br
rXt=0
for r ≥ 1, and by convention for t = 0, a1 · · · at = 1 and for t = r, bt+1 · · · br = 1.
The Eilenberg-Zilber map EZ : B(A) ⊗ B(B) → B(A ⊗ B) is defined as follows
EZ0(a1 ⊗ a2 ⊗ b1 ⊗ b2) = a1 ⊗ b1 ⊗ a2 ⊗ b2
EZr(1 ⊗ a1 ⊗ · · · ⊗ ar−t ⊗ 1 ⊗ 1 ⊗ b1 ⊗ · · · bt ⊗ 1) =
(−1)σ1 ⊗ 1 ⊗ F (xσ−1(1)) ⊗ · · · ⊗ F (xσ−1(r)) ⊗ 1 ⊗ 1
Xσ∈Sr−t,t
for r ≥ 1, where F (a) = a ⊗ 1 and F (b) = 1 ⊗ b.
11
Remark 3. These two maps gives an equivalence of complexes. Moreover,
AW EZ = id and EZAW ≃ id
Proposition 3.7. The induced maps for AW∗ and EZ∗ are
AW n : (A ⊗ B)n ⊗ A ⊗ B −→ Mi+j=n
Ai ⊗ A ⊗ Bj ⊗ B
AW 0 ≡ id
AW n((a1 ⊗ b1 ⊗ · · · ⊗ an ⊗ bn) ⊗ a ⊗ b) =
nXk=0
EZ n : Mi+j=n
(−1)k(n−k)(ak+1 ⊗ · · · ⊗ an ⊗ aa1a2 · · · ak) ⊗ (b1 ⊗ · · · ⊗ bk ⊗ bk+1 · · · bnb)
Ai ⊗ A ⊗ Bj ⊗ B −→ (A ⊗ B)n ⊗ A ⊗ B
EZ 0 ≡ id
EZ n((a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)) =
Xσ∈Sn−t,t
(−1)σ(cid:16)F (xσ−1(1)) ⊗ · · · ⊗ F (xσ−1(n))(cid:17) ⊗ a ⊗ b
The Connes B-operator on the Hochschild homology of the tensor product of two algebras
satisfies the following equation
Proposition 3.8. AW B A⊗B EZ = B A ⊗ id + id ⊗ B B
Proof. Let (a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b) ∈ An−t ⊗ A ⊗ Bt ⊗ B. Applying B A⊗B
we get
n EZ n,
(−1)σ+inFi ⊗ · · · ⊗ Fn ⊗ a ⊗ b ⊗ F1 ⊗ · · · ⊗ Fi−1 ⊗ 1 ⊗ 1
where Fi = F (xσ−1(i)) for 1 ≤ i ≤ n. Reordering the inner sum, we get
nXi=0
Xσ∈Sn−t,t
Xσ (−1)σa ⊗ b ⊗ F1 ⊗ · · · ⊗ Fn ⊗ 1 ⊗ 1
+
(−1)σ+inFi ⊗ · · · ⊗ Fn ⊗ a ⊗ b ⊗ F1 ⊗ · · · ⊗ Fi−1 ⊗ 1 ⊗ 1
+ (−1)σ+(n−t+1)nFn−t+1 ⊗ · · · ⊗ Fn ⊗ a ⊗ b ⊗ F1 ⊗ · · · ⊗ Fn−t ⊗ 1 ⊗ 1
+
(−1)σ+(n−t+i)nFn−t+i ⊗ · · · ⊗ Fn⊗
n−tXi=1
tXi=2
⊗ a ⊗ b ⊗ F1 ⊗ · · · ⊗ Fn−t+i−11 ⊗ 1
12
1
2
3
4
Consider the following permutations
1 · · ·
σi = 1 · · ·
σj = 1 · · ·
j
i − 1
i − 1 i + t
i
n − t
· · · n − t + j − 1
· · · n − t n − t + 1 · · ·
· · ·
· · ·
n
i
n
i + t − 1!
n − t + 1 · · · n − t + j − 1 n − t + j
n − t + j
j − 1
· · ·
1
1 ≤ i ≤ n − t
· · · n
· · · n!
Notice that σi = (n − t − i + 1)t and σi = (n − t)(j − 1). Now, applying AW n+1 to 1 the
only non-zero term arises when σ = σ1 and k = 0
AW n+1( 1 ) = (−1)n−t(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b ⊗ b1 ⊗ · · · ⊗ bt ⊗ 1)
Applying AW n+1 to 2 the only non-zero terms arise when σ = σi for 1 ≤ i ≤ n − t and k = t
2 ≤ j ≤ t
AW n+1( 2 ) =
(−1)i(n−t)(ai ⊗ · · · ⊗ an−t ⊗ a ⊗ a1 ⊗ · · · ⊗ ai−1 ⊗ 1)
n−tXi=1
⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)
Applying AW n+1 to 3 the only non-zero terms arise when σ = σi for 1 ≤ i ≤ n − t, k = t and
k = t + 1
AW n+1( 3 ) =(a ⊗ a1 ⊗ · · · ⊗ an−t ⊗ 1) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)
+ (−1)n(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b ⊗ 1)
Applying AW n+1 to 4 the only non-zero terms arise when σ = σi for 2 ≤ i ≤ t and k = t + 1
AW n+1( 4 ) =
tXi=2
(−1)it+n−t(a1 ⊗ · · · ⊗ an−t ⊗ a)
⊗ (bi ⊗ · · · ⊗ bt ⊗ b ⊗ b1 ⊗ · · · ⊗ bi−1 ⊗ 1)
Applying B A
n−t ⊗ id + (−1)n−tid ⊗ B B
t to (a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b), we get
(−1)i(n−t)(ai ⊗ · · · ⊗ an−t ⊗ a ⊗ a1 ⊗ · · · ⊗ ai−1 ⊗ 1) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)
+ (−1)n(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b ⊗ 1)
=AW n+1( 1 ) + AW n+1( 2 ) + AW n+1( 3 ) + AW n+1( 4 )
13
tXi=0
n−tXi=0
n−tXi=1
tXi=2
+ (−1)n−t
(−1)it(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (bi ⊗ · · · ⊗ bt ⊗ b ⊗ · · · ⊗ bi−1 ⊗ 1)
=(a ⊗ a1 ⊗ · · · ⊗ an−t ⊗ 1) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)
+
(−1)i(n−t)(ai ⊗ · · · ⊗ an−t ⊗ a ⊗ a1 ⊗ · · · ⊗ ai−1 ⊗ 1) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b)
+ (−1)n−t(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b ⊗ b1 ⊗ · · · ⊗ bt ⊗ 1)
+
(−1)it+n−t(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (bi ⊗ · · · ⊗ bt ⊗ b ⊗ b1 ⊗ · · · ⊗ bi−1 ⊗ 1)
Therefore,
AW n+1B
n EZ n((a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b))
A⊗B
=(cid:16)B
A
n−t ⊗ id + (−1)n−tid ⊗ B
B
t (cid:17) ((a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b))
Theorem 3.9. Let A and B be finite dimensional symmetric R-algebras with R a commutative
hereditary ring. Then
HH ∗(A; A) ⊗ HH ∗(B; B) ֒→ HH ∗(A ⊗ B; A ⊗ B)
is an injection of BV-algebras.
Proof. Since both algebras are finite dimensional, we have
Hom(A⊗B)e(B(A ⊗ B), A ⊗ B) ∼= HomAe(B(A), A) ⊗ HomBe(B(B), B)
Therefore, by theorem 3.5 there is an injection of graded algebras
HH ∗(A; A) ⊗ HH ∗(B; B) ֒→ HH ∗(A ⊗ B; A ⊗ B)
By theorem 2.5, the BV-operator is given by the dual of the Connes operator. By dualizing
equation 3.8, we get
EZ ∨∆A⊗BAW ∨ = ∆A ⊗ id + id ⊗ ∆B
on the cochain level, which gives the desire injection on the cohomological level.
Corollary 3.10 ([13] Theorem 3.5). Let A and B be finite dimensional symmetric R-algebras
with R a commutative hereditary ring. If HH ∗(A; A), or H ∗(B; B), is R-projective. Then
HH ∗(A; A) ⊗ HH ∗(B; B) ∼= HH ∗(A ⊗ B; A ⊗ B)
is an isomorphism of BV-algebras.
Next, we study the action of Hochschild cohomology on Hochschild homology of tensor
products
Proposition 3.11. If at least one of the algebras is finite dimensional, the action of HH ∗(A ⊗
B; A ⊗ B) on HH∗(A ⊗ B; A ⊗ B) is given by the tensor product of the actions.
Proof. Let (a1 ⊗ · · · ⊗ an−t ⊗ a) ∈ An−t ⊗ A, (b1 ⊗ · · · ⊗ bt ⊗ b) ∈ ⊗Bt ⊗ B, α ∈ Hom(Am−i, A)
and β ∈ Hom(Bi, B) with n − t ≥ m − i ≥ 0 and n − m ≥ t − i ≥ 0. We claim that
on the (co)chain level, which implies the assertion on the (co)homological level.
AW ρA⊗B(EZ ⊗ AW ∨) = ±(cid:16)ρA ⊗ ρB(cid:17)
(a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b) ⊗ (α ⊗ β)
(−1)t(m−i)+σF1 ⊗ · · · ⊗ Fn ⊗ a ⊗ b! ⊗ AW ∨
m(α ⊗ β)
ρA⊗B
7−−−→
(−1)t(m−i)+σ+nmFm+1 ⊗ · · · ⊗ Fn ⊗ (a ⊗ b) AW ∨
m(α ⊗ β)(F1 ⊗ · · · ⊗ Fm)
∨
m
EZn⊗AW
7−−−−−−−→ Xσ
Xσ
14
Applying AW n−m the only non-zero term arise when k = t−i and σ is the following permutation
· · · m − i m − i + 1
· · · n − t n − t + 1 · · · n − t + i n − t + i + 1 · · · n
m + 1 · · · m + t − i!
1
i + 1 · · · m m + t − i + 1 · · ·
n
1
· · ·
i
Since σ = i(m − i) + t(n − t − m + i), we get
(−1)t(m−i)+(m−i)(n−t)+it(am−i+1 ⊗ · · · ⊗ an−t ⊗ a α(a1 ⊗ · · · ⊗ am−i))
⊗(bi+1 ⊗ · · · ⊗ bt ⊗ b β(b1 ⊗ · · · ⊗ bi))
which is precisely
(−1)t(m−i)(cid:16)ρA ⊗ ρB(cid:17) ((a1 ⊗ · · · ⊗ an−t ⊗ a) ⊗ α ⊗ (b1 ⊗ · · · ⊗ bt ⊗ b) ⊗ β)
The following proposition is a slightly generalization of the previous proposition 3.11
Proposition 3.12. Under the same hypothesis as in theorem 3.5, the action of HH ∗(A ⊗
B; A ⊗ B) on HH∗(A ⊗ B; A ⊗ B) is given by the tensor product of the actions.
Proof. Let ∆A : P(A) → P(A)⊗
A
P(A) and ∆B : P(B) → P(B)⊗
B
P(B) be diagonal approximation
maps. By proposition 3.4, ∆ = τ (∆A ⊗ ∆B) is a diagonal approximation map for P(A ⊗ B).
B, f ∈ HomAe(P(A), A) and g ∈ HomBe(P(B), B).
Let x ⊗ a ∈ P(A) ⊗
Ae
A, y ⊗ b ∈ P(B) ⊗
Be
Notice that the following diagram commutes up to the sign (−1)xg
P(A) ⊗ P(B)
∆A⊗∆B
P(A) ⊗
A
P(A) ⊗ P(B) ⊗
B
P(B)
(−1)xg(f ⊗
A
id)⊗(g⊗
B
id)
P(A) ⊗
A
A ⊗ P(B) ⊗
B
B
τ
τ
/ P(A ⊗ B) ⊗
A⊗B
P(A ⊗ B)
(f ⊗g) ⊗
id
A⊗B
/ P(A ⊗ B) ⊗
A⊗B
A ⊗ B
Therefore,
ρA⊗B(((x ⊗ a) ⊗ (y ⊗ b)) ⊗ (f ⊗ g)) = (−1)f yρA((x ⊗ a) ⊗ f ) ⊗ ρB((y ⊗ b) ⊗ g)
To sum up, we get the following
Theorem 3.13. Let R be a commutative hereditary ring. Let A and B be two R-algebras
satisfying the following hypothesis:
• Suppose that P(A) → A is a resolution of A of finitely generated projective Ae-modules
and P(B) → B is a Be-resolution of B such that
Hom(A⊗B)e(P(A ⊗ B), A ⊗ B) ∼= HomAe(P(A), A) ⊗ HomBe(P(B), B)
15
/
/
• HH ∗(A; A), or H ∗(B; B), is R-projective.
• Let a ∈ HHn(A; A) and b ∈ HHm(B; B) such that
ρa : HH ∗(A; A) −→ HHn−∗(A; A)
ρb : HH ∗(B; B) −→ HHm−∗(B; B)
c 7−→ ρ(a ⊗ c)
c 7−→ ρ(b ⊗ c)
are isomorphisms, and B(a) = 0 = B(b).
Then there is an isomorphism of BV-algebras
HH ∗(A ⊗ B) ∼= HH ∗(A) ⊗ HH ∗(B)
Proof. By proposition 3.12, the action for A ⊗ B is given by the tensor product of the actions.
Therefore,
ρa ⊗ ρb : HH ∗(A ⊗ B; A ⊗ B) → HHn+m−∗(A ⊗ B; A ⊗ B)
is an isomorphism. Then
∆A⊗B = (ρ−1
a ⊗ ρ−1
b )(BA ⊗ id + id ⊗ BB)(ρa ⊗ ρb) = ∆A ⊗ id + id ⊗ ∆B
4 BV-Algebra Structure on HH ∗(Z[Z/nZ])
From now on, we assume that A is R [Z/nZ] ∼= R[σ]/(σn − 1) with R a commutative ring.
Since the Hochschild (co)homology of an associative algebra can be calculated using projective
Ae-resolutions and the bar construction is not convenient to make explicit calculations, we are
going to use the following 2-periodical resolution [10], [12].
Proposition 4.1. The following is a Ae-projective resolution of A
P(A) : · · · → A ⊗ A d2−→ A ⊗ A d1−→ A ⊗ A
µ
−→ A → 0
with
µ(a ⊗ b) = ab
d2k+1(a ⊗ b) = (a ⊗ b)(1 ⊗ σ − σ ⊗ 1)
d2k(a ⊗ b) = (a ⊗ b)
σi ⊗ σn−i−1
n−1Xi=0
Proof. First of all, notice that A⊗A ∼= Ae as Ae-modules, so A2 is Ae-free. From the definition,
it follows that drdr+1 = 0. Now, we are going to define the following A-right maps
· · · s3←− A ⊗ A s2←− A ⊗ A s1←− A ⊗ A s0←− A ← 0
16
s0 : A −→ A2,
s2k+1 : A2 −→ A2,
s2k : A2 −→ A2,
s0(σi) = 1 ⊗ σi
s2k+1(σi ⊗ 1) =
s2k(σi ⊗ 1) =
i−1Xj=0
−
0
1 ⊗ 1
0
σj ⊗ σi−j−1
if i 6= 0
if i = 0
if i = n − 1
otherwise
and by direct calculations, it follows that µs0 = id and dk+1sk+1 + skdk = id for all k ≥ 1.
Therefore, the complex is acyclic.
Tensoring this resolution by A as Ae-modules and using the identification A2 ⊗
Ae
A ∼= A,
((x ⊗ y) ⊗ a 7→ yax), we obtain the complex
· · · → A nσn−1
−−−→ A 0−→ A nσn−1
−−−→ A 0−→ A
(6)
Taking HomAe(−, A) of P(A) and using the identification HomAe(A2, A) ∼= A, (f 7→ f (1 ⊗ 1)),
we obtain the complex
A 0−→ A nσn−1
−−−→ A 0−→ A nσn−1
−−−→ A → · · ·
(7)
Then
HHi(A) =
HH i(A) =
A
A/(nσn−1A)
Ann(nσn−1)
A
Ann(nσn−1)
A/(nσn−1A)
if i = 0
if i = 2k + 1
if i = 2k
if i = 0
if i = 2k + 1
if i = 2k
To calculate the algebraic structures of HH ∗(A; A), we use two chain maps between P(A)
and the normalized bar resolution B(A)
ψ∗ : P(A) → B(A)
ϕ∗ : B(A) → P(A)
which are homotopy equivalences.
The Ae-homomorphisms ψ∗ will be defined by
ψ0 = id : A2 −→ A2
ψr+1 : A2 −→ A ⊗ ¯Ar+1 ⊗ A,
ψr+1(1 ⊗ 1) := srψrdr+1(1 ⊗ 1)
By direct computations, it follows that
ψ2r(1 ⊗ 1) = X0≤i1,...,ir≤n−1
ψ2r+1(1 ⊗ 1) = X0≤i1,...,ir≤n−1
(−1)r1 ⊗ σi1 ⊗ σ ⊗ σi2 ⊗ · · · ⊗ σir ⊗ σ ⊗ σr(n−1)−Pr
(−1)r+11 ⊗ σ ⊗ σi1 ⊗ · · · ⊗ σir ⊗ σ ⊗ σr(n−1)−Pr
k=1
ik
k=1
ik
17
And the Ae-homomorphisms ϕ∗ will be defined by
ϕ0 = id : A2 −→ A2
ϕ1 : A ⊗ ¯A ⊗ A −→ A2, ϕ1(1 ⊗ σi ⊗ 1) := −
i−1Xj=0
ϕ2 : A ⊗ ¯A2 ⊗ A −→ A2, ϕ2(1 ⊗ σi ⊗ σk ⊗ 1) :=
σj ⊗ σi−j−1
−1 ⊗ σi+k−n
0
if i + k ≥ n
otherwise
ϕr : A ⊗ ¯Ar ⊗ A −→ A2 for r > 2
ϕr(1 ⊗ σi1 ⊗ · · · ⊗ σir ⊗ 1) = ϕr−2(1 ⊗ σi1 ⊗ · · · ⊗ σir−2 ⊗ 1) · ϕ2(1 ⊗ σir−1 ⊗ σir ⊗ 1)
By direct computations, it follows that
ϕ2r(1 ⊗ σi1 ⊗ · · · ⊗σi2r ⊗ 1) =
ϕ2(1 ⊗ σi2k−1 ⊗ σi2k ⊗ 1)
0
rYk=1
(−1)r ⊗ σP2r
=
σj ⊗ σP2r+1
i1−1Xj=0
k=1
k=1
ik−rn
if i2k−1 + i2k ≥ n for 1 ≤ k ≤ r
otherwise
ϕ2(1 ⊗ σi2k ⊗ σi2k+1 ⊗ 1)
rYk=1
ik−j−rn−1
if i2k + i2k+1 ≥ n for 1 ≤ k ≤ r
otherwise
ϕ2r+1(1 ⊗ σi1 ⊗ · · · ⊗ σi2r+1 ⊗ 1) = ϕ1(1 ⊗ σi1 ⊗ 1)
=
(−1)r+1
0
Remark 4. These two maps gives an equivalence of complexes. Moreover,
Proposition 4.2. Using the identifications
ϕ∗ψ∗ = id and ψ∗ϕ∗ ≃ id
A ⊗ ¯A∗ ⊗ A ⊗
Ae
A ∼= ¯A∗ ⊗ A and A2 ⊗
Ae
A ∼= A
the induced maps for ψ∗ and ϕ∗ are
¯ψ∗ : A −→ ¯A∗ ⊗ A
¯ϕ∗ : ¯A∗ ⊗ A −→ A
¯ψ2r(a) = X0≤i1,...,ir≤n−1
¯ψ2r+1(a) = X0≤i1,...,ir≤n−1
¯ϕ2r(σi1 ⊗ · · · ⊗ σi2r ⊗ a) =
¯ϕ2r+1(σi1 ⊗ · · · ⊗ σi2r+1 ⊗ a) =
0
where 1 ≤ k ≤ r.
(−1)rσi1 ⊗ σ ⊗ σi2 ⊗ · · · ⊗ σir ⊗ σ ⊗ σr(n−1)−Pr
(−1)r+1σ ⊗ σi1 ⊗ · · · ⊗ σir ⊗ σ ⊗ σr(n−1)−Pr
k=1
k=1
ik a
ika
(−1)rσP2r
k=1
ik−rna if i2k−1 + i2k ≥ n
otherwise
(−1)r+1ai1σP2r+1
k=1
0
ik−rn−1
if i2k + i2k+1 ≥ n
otherwise
18
Proposition 4.3. Using the identifications
HomAe (A ⊗ ¯A∗ ⊗ A, A) ∼= Hom( ¯A∗, A)
and HomAe(A2, A) ∼= A
the induced maps for ψ∗ and ϕ∗ are
k=1
(−1)rf (σi1, σ, σi2, . . . , σir , σ)σr(n−1)−Pr
(−1)r+1f (σ, σi1, . . . , σir , σ)σr(n−1)−Pr
(−1)raσP2r
if i2k−1 + i2k ≥ n
otherwise
ik−rn
k=1
k=1
ik
ik
¯ψ∗
r : A −→ HomR( ¯Ar, A)
¯ϕ∗
¯ψ∗
r : Hom( ¯Ar, A) −→ A
¯ψ∗
2r(f ) = X0≤i1,...,ir≤n−1
2r+1(f ) = X0≤i1,...,ir≤n−1
2r(a)(σi1, . . . , σi2r ) =
2r+1(a)(σi1, . . . , σi2r+1) =
¯ϕ∗
¯ϕ∗
0
(−1)r+1ai1σP2r+1
k=1
0
ik−rn−1
if i2k + i2k+1 ≥ n
otherwise
where 1 ≤ k ≤ r.
4.1 Cup Product and Cohomology Ring
Lemma 4.4. Let R be a commutative ring. Then the cup product on the even Hochschild
cohomology of A = R[σ]/(σn − 1) is induced by multiplication in A.
Proof. Let a ∈ HH 2r(A; A) and b ∈ HH 2s(A; A). Then
¯ϕ∗
2r(a) ⌣ ¯ϕ∗
2s(b) ∈ Hom( ¯A2(r+s), A)
and
( ¯ϕ∗
2r(a) ⌣ ¯ϕ∗
2s(b))(σi1, . . . , σi2(r+s)) = ( ¯ϕ∗
2r(a))(σi1, . . . , σi2r ) · ( ¯ϕ∗
2s(b))(σi2r+1, . . . , σi2(r+s))
=
(−1)r+sabσP2(r+s)
k=1
0
= ( ¯ϕ∗
2(r+s)(ab))(σi1, . . . , σi2(r+s))
ik−(r+s)n
if i2k−1 + i2k ≥ n for 1 ≤ k ≤ r + s
otherwise
Since ¯ψ∗ ¯ϕ∗ = id, then the cup product is induced by multiplication in A.
Lemma 4.5. ⌣: HH i(A; A) ⊗ HH j(A; A) → HH i+j(A; A) is induced by multiplication if i or
j is even, and by the formula
if i and j are odd.
a ⌣ b = −
(n − 1)n
2
abσn−2
19
Proof. Let a ∈ HH 2r+1(A; A) and b ∈ HH 2s(A; A). Then
¯ϕ∗
2r+1(a) ⌣ ¯ϕ∗
2s(b) ∈ Hom( ¯A2(r+s)+1, A)
and
( ¯ϕ∗
2r+1(a) ⌣ ¯ϕ∗
2s(b))(σi1, . . . , σi2(r+s)+1) = ( ¯ϕ∗
2r+1(a))(σi1, . . . , σi2r+1) · ( ¯ϕ∗
2s(b))(σi2r+2, . . . , σi2(r+s)+1)
=
(−1)r+s+1abi1σP2(r+s)+1
k=1
0
= ( ¯ϕ∗
2(r+s)+1(ab))(σi1, . . . , σi2(r+s)+1)
ik−(r+s)n−1
if i2k + i2k+1 ≥ n for 1 ≤ k ≤ r + s
otherwise
Then the cup product is induced by multiplication in A if a or b is even.
Assume now that a ∈ HH 2r+1(A; A) and b ∈ HH 2s+1(A; A). Then
¯ϕ∗
2r+1(a) ⌣ ¯ϕ∗
2s(b) ∈ Hom( ¯A2(r+s+1), A)
and
( ¯ϕ∗
2r+1(a) ⌣ ¯ϕ∗
2s+1(b))(σi1, . . . , σi2(r+s+1)) = ( ¯ϕ∗
2r+1(a))(σi1, . . . , σi2r+1) · ( ¯ϕ∗
2s+1(b))(σi2r+2, . . . , σi2(r+s+1))
(−1)r+sabi1i2r+2σP2(r+s+1)
k=1
0
ik−(r+s)n−2
if i2k + i2k+1 ≥ n for 1 ≤ k ≤ r and
i2k−1 + i2k ≥ n for r + 2 ≤ k ≤ r + s + 1
otherwise
Applying ¯ψ∗
2(r+s+1) to f = ¯ϕ∗
2r+1(a) ⌣ ¯ϕ∗
2s+1(b), we have
=
¯ψ∗
2(r+s+1)(f )
(−1)r+s+1f (σi1, σ, σi2, . . . , σir+s+1, σ)σ(r+s+1)(n−1)−Pr+s+1
k=1
=
X0≤i1,...,ir+s+1≤n−1
n−1Xi1=1
= −
abi1σn−2 = −
(n − 1)n
2
abσn−2
Therefore, if char(R) = p > 0 and n = mp, we have
a ⌣ b =
mabσ2m−2
0
if p = 2
if p 6= 2
From these results, now we can described the cohomology ring.
20
ik
(8)
Theorem 4.6. Let R be a commutative ring and A = R[σ]/(σn − 1). Then
HH 2∗(A; A) = R[x, z]/(xn − 1, nz)
where x ∈ HH 0(A; A) and z ∈ HH 2(A; A).
Proof. Consider x ∈ HH 0(A; A) to be the coset [σ] ∈ A and z ∈ HH 2(A; A) the coset [1] ∈ A.
By lemma 4.4, the cup product for even degrees is induced by multiplication in A. Then x
generates HH 0(A; A), and HH 2(A; A) is generated by z and HH 0(A; A). In higher degrees
HH 2i(A; A) is generated by zi and HH 0(A; A). The relations are given by xn − 1 = 0 and
nxn−1z = 0.
Corollary 4.7. Let R be an integral domain with char(R) ∤ n. Then
HH ∗(A; A) = R[x, z]/(xn − 1, nz)
Proof. Since R is an integral domain with char(R) ∤ n, we have Ann(nσn−1) = 0.
Corollary 4.8. Let R be a commutative ring such that n ∈ R∗. Then
HH ∗(A; A) = R[x]/(xn − 1) = A
Proof. Since n ∈ R∗, we have Ann(nσn−1) = Ann(σn−1) = 0, and xn−1z = 0 implies that
z = 0.
Theorem 4.9. Let R be a commutative ring with char(R) = p > 0 and A = R[σ]/(σn − 1)
with n = mp. If p 6= 2, or p = 2 and m is even. Then
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2)
If p = 2 and m is odd. Then
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2 − xn−2z)
where x ∈ HH 0(A; A), y ∈ HH 1(A; A) and z ∈ HH 2(A; A).
Proof. By theorem 4.6, we know that
HH 2∗(A; A) = R[x, z]/(xn − 1)
Consider y ∈ HH 1(A; A) to be the coset [1] ∈ A. Since cup product of an odd degree cohomol-
ogy class and an even degree cohomology class is induced by multiplication in A, HH 1(A; A)
is generated by y and HH 0(A; A). By (8), for p 6= 2, or p = 2 and m even the cup product in
odd degrees is zero. Therefore, y2 = 0 and we have
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2)
For p = 2 and m odd, y2 is the coset [σn−2] ∈ A then y2 − xn−2z = 0, and
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2 − xn−2z)
Remark 5. These calculations agree with the ones presented in [3] and [12].
21
4.2 BV-Algebra Structure
Theorem 4.10. Let R be an integral domain with char(R) ∤ n and A = R[σ]/(σn − 1). Then
the canonical Frobenius form of the group ring induces a BV-algebra structure on HH ∗(A; A)
given by
HH ∗(A; A) = R[x, z]/(xn − 1, nz)
∆(a) = 0 ∀a ∈ HH ∗(A; A)
Proof. By corollary 4.7, we have HH ∗(A; A) = HH 2∗(A; A). Then ∆(a) = 0 for all a ∈
HH ∗(A; A). However, this can be proved directly from the definition of ∆, and the fact that
in a BV-algebra we have the following equation
∆(abc) = ∆(ab)c + (−1)aa∆(bc) + (−1)(a−1)bb∆(ac)
− ∆(a)bc − (−1)aa∆(b)c − (−1)a+bab∆(c)
(9)
Since the BV-operator is defined over the bar complex, we need the cochains that represent the
generators. The class x is represented by the cochain
¯ϕ∗
0(σ)(1) = σ
and the class z by
¯ϕ∗
2(1)(σi, σk) =
−σi+k−n
0
if i + k ≥ n
otherwise
Now, taking {1, σ, . . . , σn−1} as a basis for A and {1, σn−1, . . . , σ} as the dual basis induced by
the canonical Frobenius form (2.1), we have
∆( ¯ϕ∗
2(1))(σi) =
∆( ¯ϕ∗
2(1) ⌣ ¯ϕ∗
0(σ))(σi) =
∆(x) = 0 by degree.
h1, ¯ϕ∗
2(1)(σk, σi)iσn−k −
n−1Xk=0
∆(z) = 0
h1, ¯ϕ∗
2(σ)(σk, σi)iσn−k −
n−1Xk=0
∆(zx) = 0
h1, ¯z(σi, σk)iσn−k
n−1Xk=0
h1, ¯z¯x(σi, σk)iσn−k
n−1Xk=0
∆(( ¯ϕ∗
2(1))2)(σi, σj, σh) =
+
n−1Xk=0
n−1Xk=0
h1, ¯ϕ∗
4(1)(σk, σi, σj, σh)iσn−k −
h1, ¯ϕ∗
4(1)(σi, σj, σh, σk)iσn−k
h1, ¯ϕ∗
4(1)(σj, σh, σk, σi)iσn−k −
h1, ¯ϕ∗
4(1)(σh, σk, σi, σj)iσn−k
n−1Xk=0
n−1Xk=0
∆(z2) = 0
In the last case, h1, ·i 6= 0 only if k + i + j + h − 2n = n, i.e., k + i + j + h = 3n, but also
k + i, j + h, i + j, h + k ≥ n. Therefore, all the coefficients are zero. Using equation (9) and
induction on powers of x and z, we have ∆(a) = 0 for all a ∈ HH ∗(A; A).
22
Theorem 4.11. Let R be a commutative ring with char(R) = p > 0 and A = R[σ]/(σn − 1)
with n = mp. If p 6= 2, or p = 2 and m is even. Then the canonical Frobenius form of the
group ring induces a BV-algebra structure on HH ∗(A; A) given by
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2)
∆(zkxl) = 0
∆(zkyxl) = (l − 1)zkxl−1
If p = 2 and m is odd. Then as a BV-algebra
HH ∗(A; A) = R[x, y, z]/(xn − 1, y2 − xn−2z)
∆(zkxl) = 0
∆(zkyxl) = (l − 1)zkxl−1
where x ∈ HH 0(A; A), y ∈ HH 1(A; A) and z ∈ HH 2(A; A).
Proof. As in the previous theorem, we need the cochains that represent the generators. The
class y is represented by the cochain
1(1)(σi) = −iσi−1
¯ϕ∗
∆( ¯ϕ∗
1(1))(1) =
h1, ¯ϕ∗
1(1)(σj)iσn−j =
−jh1, σj−1iσn−j
−jh1, σjiσn−j
n−1Xi=0
n−1Xi=0
= − σn−1 = − ¯ϕ∗
0(σn−1)(1)
∆(y) = −xn−1
h1, ¯ϕ∗
1(σ)(σj)iσn−j =
n−1Xj=0
∆(xy) = 0
n−1Xj=0
n−1Xk=0
n−1Xk=0
n−1Xk=0
∆( ¯ϕ∗
0(σ) ⌣ ¯ϕ∗
1(1))(1) =
∆( ¯ϕ∗
2(1) ⌣ ¯ϕ∗
1(1))(σi, σj) =
+
+
h1, ¯ϕ∗
3(1)(σk, σi, σj)iσn−k
h1, ¯ϕ∗
3(1)(σi, σj, σk)iσn−k
h1, ¯ϕ∗
3(1)(σj, σk, σi)iσn−k
If i + j < n, h1, ·i = 0 for all k. When i + j ≥ n, h1, ·i 6= 0 only if k + i + j − 1 − n = n, i.e.,
k = 2n + 1 − (i + j). Therefore,
∆( ¯ϕ∗
3(1))(σi, σj) = (2n + 1 − (i + j))σi+j−n−1 + iσi+j−n−1 + jσi+j−n−1
2(σn−1)(σi, σj)
(char(R) = p and n = mp)
=(2n + 1)σi+j−n−1 = σi+j−n−1 = − ¯ϕ∗
∆(zy) = −zxn−1
23
Using equation (9) and induction on powers of x, y and z, we have
∆(zkxl) = 0
∆(zkyxl) = (l − 1)zkxl−1
Since in a BV-algebra, the Gerstenhaber bracket is defined by the following equation
{a, b} = −(−1)a(∆(ab) − ∆(a)b − (−1)aa∆(b))
(10)
It follows that
Corollary 4.12. Let A = R[σ]/(σn − 1) with R an integral domain and char(R) ∤ n. The
Gerstenhaber bracket on HH ∗(A; A) is given by
{a, b} = 0 ∀a, b ∈ HH ∗(A; A)
Corollary 4.13. Let R be a commutative ring with char(R) = p > 0 and A = R[σ]/(σn − 1)
with n = mp. The Gerstenhaber bracket on HH ∗(A; A) is given by
{zk1xl1, zk2xl2} = 0
{zk1xl1, zk2yxl2} = −l1zk1+k2xl1+l2−1
{zk1yxl1, zk2yxl2} = (l2 − l1)zk1+k2yxl1+l2−1
Remark 6. These calculations agree with the Gerstenhaber bracket presented in [17] and [16].
For the cyclic group of order p prime, Z/pZ, we have that the group ring Fp [Z/pZ] =
Fp[σ]/(σp − 1) is naturally isomorphic, as algebra, to a truncated polynomial ring Fp[x]/(xp).
In [1], the authors transfer the canonical Frobenius form of the group ring to the truncated
polynomial ring and get the following Frobenius form
ε
p−1Xi=0
αixi =
(−1)iαi
p−1Xi=0
Using this Frobenius form, the BV-algebra structure is given by
Theorem 4.14. Let A = Fp[x]/(xp) with p an odd prime. Then the canonical Frobenius form
of the group ring induces a BV-algebra structure on HH ∗(A; A) given by
HH ∗(A; A) = Fp[x, v, t]/(xp, v2)
∆(tlxk) = 0
∆(tkvx2l) = 2ltkx2l−1 +
p−1Xi=2l
∆(tkvx2l+1) = (2l + 1)tkx2l +
(−1)i+1tkxi
p−1Xi=2l+1
(−1)itkxi
where x ∈ HH 0(A; A), v ∈ HH 1(A; A) and t ∈ HH 2(A; A).
24
Corollary 4.15. There is an isomorphism of BV-algebras
φ : HH ∗(Fp[x]/(xp); Fp[x]/(xp))
∼=−→ HH ∗(Fp [Z/pZ] ; Fp [Z/pZ])
Proof. The isomorphism φ is defined as follows
φ(x) = x − 1,
φ(v) = y
and
φ(t) = z
It is clear that φ is a ring isomorphism. To verify that it is an isomorphism of BV-algebras, we
need to check that φ ∆ = ∆φ.
• φ ∆(x) = φ(0) = 0 = ∆(x − 1) = ∆φ(x).
• ∆(v) =Pp−1
i=0 (−1)i+1xi, then
φ ∆(v) =
=
p−1Xi=0
p−1Xk=0
(−1)i+1(x − 1)i =
(−1)k+1
k!xk
p−1Xi=k i
≡ −xp−1
= ∆φ(v)
p−1Xi=0
(−1)k+1 i
iXk=0
k!xk
the equivalence module p is due to
p−1Xi=k i
k! = p
k + 1! ≡ 0
(mod p)
for k + 1 6= 0 or k + 1 6= p.
• φ ∆(t) = φ(0) = 0 = ∆(z) = ∆φ(t).
• φ ∆(t2) = φ(0) = 0 = ∆(z2) = ∆φ(t2).
• φ ∆(tx) = φ(0) = 0 = ∆(zx) − ∆(z) = ∆(z(x − 1)) = ∆φ(tx).
• ∆(tv) =Pp−1
i=0 (−1)i+1txi, then
φ ∆(tv) =
(−1)i+1z(x − 1)i = z
p−1Xi=0
(−1)k+1 i
iXk=0
k!xk
p−1Xi=0
p−1Xk=0
≡ −zxp−1
= ∆φ(tv)
= z
(−1)k+1
k!xk
p−1Xi=k i
25
(mod p)
(mod p)
• ∆(vx) =Pp−1
i=0 (−1)ixi, then
φ ∆(vx) =
=
p−1Xi=0
p−1Xk=0
p−1Xi=0
(−1)k i
iXk=0
k!xk
(−1)i(x − 1)i =
(−1)k
k!xk
p−1Xi=k i
≡ xp−1
= ∆(yx) − ∆(y) = ∆(y(x − 1)) = ∆φ(vx)
(mod p)
Since both are BV-algebras, formula (9) holds and φ ∆ = ∆φ.
And for p = 2,
Theorem 4.16. Let A = F2[x]/(x2). Then the canonical Frobenius form of the group ring
induces a BV-algebra structure on HH ∗(A; A) given by
HH ∗(A; A) = F2[x, v, t]/(x2, v2 − t) ∼= Λ(x) ⊗ F2[v]
∆(vkxl) = k(1 + x)vk−1
where x ∈ HH 0(A; A), v ∈ HH 1(A; A) with x = 0 and v = 1.
Corollary 4.17. There is an isomorphism of BV-algebras
φ : HH ∗(F2[x]/(x2); F2[x]/(x2))
∼=−→ HH ∗(F2 [Z/2Z] ; F2 [Z/2Z])
Proof. For p = 2 and n = 2, we have
HH ∗(F2 [Z/2Z] ; F2 [Z/2Z]) = R[x, y, z]/(x2 − 1, y2 − z) ∼= F2[y] ⊗ F2[x]/(x2 − 1)
and the BV-operator, ∆, is given by
∆(ykx) = 0
∆(yk) = kyk−1x
The isomorphism φ is defined as follows
φ(x) = x − 1
and
φ(v) = y
It is clear that φ is a ring isomorphism. Now,
φ ∆(vkxl) = φ(k(1 + x)vk−1) = kxyk−1 = ∆(yk(x − 1)l) = ∆φ(vkxl)
Therefore, φ is an isomorphism of BV-algebras.
26
5 BV-Algebra Structure for Finite Abelian Groups
Let G be a finite abelian group. Then G can be decomposed as follows
G ∼= Z/pα1
1 Z ⊕ Z/pα2
2 Z ⊕ · · · ⊕ Z/pαk
k Z
with the property that αi ≤ αi+1, if pi = pi+1. Therefore,
R[G] ∼= R[Z/pα1
1 Z] ⊗ R[Z/pα2
2 Z] ⊗ · · · ⊗ R[Z/pαk
k Z]
and by corollary 3.10, we have
Theorem 5.1. Let R be a field and G a finite abelian group. Then as BV-algebras
HH ∗(R[G]; R[G]) ∼= HH ∗(R[Z/pα1
1 Z]; R[Z/pα1
1 Z]) ⊗ · · · ⊗ HH ∗(R[Z/pαk
k Z]; R[Z/pαk
k Z])
where the BV-structure for each factor is given by theorem 4.10 or 4.11.
6 BV-Algebra Structure on HH ∗(Z[Z/nZ] ⊗ Z[Z/mZ])
By theorem 3.5, we have an injection of BV-algebras
HH ∗(Z[Z/nZ]) ⊗ HH ∗(Z[Z/mZ]) ֒→ HH ∗(Z[Z/nZ] ⊗ Z[Z/mZ])
where the BV-operator on the left hand side is trivial. Nevertheless, the BV-operator on the
right hand side is highly non-trivial as follows
Theorem 6.1. Let A = Z[Z/nZ] and B = Z[Z/mZ] with n = km. Then as a BV-algebra,
HH ∗(A ⊗ B; A ⊗ B) ∼=
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2)
∆(xitjalbrcs) = sxi−1tjalbr((i − 1)b − jka)
in all cases except when m is even and k is odd, in which case we get
HH ∗(A ⊗ B; A ⊗ B) ∼=
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2 − m
2 xn−2ab(b + ka))
∆(xitjalbrcs) = sxi−1tjalbr((i − 1)b − jka)
where x, t ∈ HH 0(A ⊗ B; A ⊗ B), a, b ∈ HH 2(A ⊗ B; A ⊗ B) and c ∈ HH 3(A ⊗ B; A ⊗ B).
Proof. By Kunneth Theorem, there is an isomorphim of modules
HH i(A ⊗ B; A ⊗ B) ∼= Mr+s=i
HH r(A; A) ⊗ HH s(B; B)
T orZ
1 (HH r(A; A), HH s(B; B))
M Mr+s=i+1
27
Since,
HH ∗(A; A) = Z[x, a]/(xn − 1, na)
and
HH ∗(B; B) = Z[t, b]/(tm − 1, mb)
where x = t = 0 and a = b = 2. All Tor groups vanish except when r and s are both
1 (A/nxn−1A, B/mtm−1B), we use the following Z-projective
even.
resolution
In order to calculate T orZ
0 → A nxn−1
−−−→ A −→ A/nxn−1A → 0
Applying − ⊗ B/mtm−1B, we get
0 → A ⊗ B/mtm−1B nxn−1⊗id
−−−−−→ A ⊗ B/mtm−1B → 0
T orZ
1 (A/nxn−1A, B/mtm−1B) = Ker(nxn−1 ⊗ id) = Ker(kxn−1 ⊗ m· )
= A ⊗ B/mtm−1B
Thus,
Therefore,
Since
HH i(A ⊗ B) =
jMl=1
A ⊗ B
if i = 0
0
if i = 1
A/nxn−1A ⊗ B ⊕ A ⊗ B/mtm−1B if i = 2j
⊕
A/nxn−1A ⊗ B/mtm−1B
j−1Ml=1
A ⊗ B/mtm−1B
if i = 2j + 1
HH ∗(A; A) ⊗ HH ∗(B; B) ֒→ HH ∗(A ⊗ B; A ⊗ B)
is an injection of BV-algebras, we only need to find a generator for the odd dimensions. Let
c ∈ HH 3(A ⊗ B; A ⊗ B) to be the coset [1 ⊗ 1] ∈ A ⊗ B/mtm−1B. Consider
c = 1 ⊗ 1 − kxn−1 ⊗ t
to be the representative of [1 ⊗ 1] in the total complex which calculate the Tor group. In the
tensor product of the bar resolutions, c is represented by
c = ϕA
∗
1(1) ⊗ ϕB
∗
2(1) − kϕA
∗
2(xn−1) ⊗ ϕB
∗
1(t)
Since HH 2∗(A ⊗ B; A ⊗ B) ∼= HH 2∗(A; A) ⊗ HH 2∗(B; B). Let y ∈ HH 2i(A; A) and z ∈
HH 2j(B; B). Then
y ⊗ z ⌣ c ⇒ (ϕA
= (ϕA
∗
2i(y) ⊗ ϕB
∗
2i(y) ⌣ ϕA
∗
∗
1(1) ⊗ ϕB
2j(z)) ⌣ (ϕA
∗
∗
1(1)) ⊗ (ϕB
2j(z)) ⌣ ϕB
∗
2(1))
∗
2(1) − kϕA
∗
2(xn−1) ⊗ ϕB
∗
1(t))
− k(ϕA
∗
2(xn−1)) ⊗ (ϕB
∗
2i(y) ⌣ ϕA
∗
2(j+1)(z)) − k(ϕA
∗
2i+1(y) ⊗ ϕB
∗
∗
2j(z)) ⌣ ϕB
1(t))
2(i+1)(yxn−1) ⊗ ϕB
∗
∗
2j+1(zt))
= (ϕA
28
Applying ¯ψ∗
A ⊗ ¯ψ∗
B , we have
(y ⊗ z)(1 ⊗ 1 − kxn−1 ⊗ t)
Therefore, HH 2k+3(A ⊗ B; A ⊗ B) is generated by HH 2k(A ⊗ B; A ⊗ B) and c ∈ HH 3(A ⊗
B; A ⊗ B). Now, consider x to be the coset [x ⊗ 1] ∈ HH 0(A; A) ⊗ HH 0(B; B), t to be the
coset [1 ⊗ t] ∈ HH 0(A; A) ⊗ HH 0(B; B), a to be the coset [1 ⊗ 1] ∈ HH 2(A; A) ⊗ HH 0(B; B)
and b to be the coset [1 ⊗ 1] ∈ HH 0(A; A) ⊗ HH 2(B; B). Notice that x, t, a and b generate
HH 2∗(A ⊗ B; A ⊗ B), and satisfy the relations xn − 1 = 0, tm − 1 = 0, nxn−1a = 0 and
mtm−1b = 0. Now,
c2 = (ϕA
∗
1(1) ⊗ ϕB
c2 = ϕA
∗
1(1) ⌣ ϕA
∗
2(xn−1) ⊗ ϕB
∗
2(1) − kϕA
∗
∗
2(1) ⌣ ϕB
1(1) ⊗ ϕB
∗
2(1) + kϕA
∗
1(t)) ⌣ (ϕA
− kϕA
∗
1(1) ⌣ ϕA
∗
2(xn−1) ⊗ ϕB
∗
2(1) ⌣ ϕB
∗
1(t) + k2ϕA
∗
∗
∗
∗
1(1) ⊗ ϕB
2(1) − kϕA
∗
∗
2(xn−1) ⌣ ϕA
1(t) ⌣ ϕB
1(1) ⊗ ϕB
∗
2(xn−1) ⊗ ϕB
2(xn−1) ⊗ ϕB
∗
2(1)
∗
1(t) ⌣ ϕB
2(xn−1) ⌣ ϕA
∗
1(t))
∗
∗
1(t)
c2 = −
c2 = −
(n − 1)n
2
(n − 1)n
2
xn−2 ⊗ 1 −
(m − 1)mk2
2
xn−2 ⊗ 1
xn−2ab2 −
(m − 1)mk2
2
xn−2a2b
Thus,
• If n and m are odd then c2 = 0.
• If n is even and m is odd then k is even and c2 = 0.
• If m is even then n is even and
c2 = −
((km − 1)xn−2ab2 + (m − 1)kxn−2a2b)
km
2
km
xn−2ab(b + ka)
2
c2 =
• If k is even then c2 = 0.
• If k is odd then c2 = m
2 xn−2ab(b + ka).
Also, notice that mc = 0. To sum up, as algebras
HH ∗(A ⊗ B; A ⊗ B) ∼=
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2)
in all cases except when m is even and k is odd, in which case we get
HH ∗(A ⊗ B; A ⊗ B) ∼=
Z[x, t, a, b, c]
(xn − 1, tm − 1, na, mb, mc, c2 − m
2 xn−2ab(b + ka))
It only remains to calculate the BV-operator. Using theorem 3.9, the BV-operator ∆A⊗B
can be calculated using ∆A and ∆B. Using the equations calculated before for ∆A and ∆B on
29
the cochain level (4.11), we have
∆(c) =∆(ϕA
=∆A(ϕA
∗
1(1) ⊗ ϕB
∗
1(1)) ⊗ ϕB
∗
2(xn−1) ⊗ ϕB
1(1) ⊗ ∆B(ϕB
∗
2(1) − kϕA
∗
2(1) − ϕA
∗
1(t) − kϕA
∗
∗
2(xn−1)) ⊗ ϕB
∗
1(t))
∗
2(1))
− k∆A(ϕA
∗
2(xn−1) ⊗ ∆B(ϕB
∗
1(t))
= − ϕA
∗
0(xn−1) ⊗ ϕB
∗
2(1)
∆(c) = −xn−1b
∆(xc) =∆(ϕA
∗
0(x) ⌣ ϕA
∗
0(x) ⌣ ϕA
∗
1(x)) ⊗ ϕB
∗
0(1) ⌣ ϕB
∗
1(1) ⊗ ϕB
2(xn−1) ⊗ ϕB
∗
∗
2(1) − ϕA
∗
2(1)
∗
0(1) ⌣ ϕB
1(x) ⊗ ∆B(ϕB
∗
− kϕA
=∆A(ϕA
∗
1(t))
∗
2(1))
2(1) ⊗ ∆B(ϕB
∗
∗
1(t))
− k∆A(ϕA
∗
2(1)) ⊗ ϕB
∗
1(t) − kϕA
=0
∆(xc) = 0
∆(tc) =∆(ϕA
− kϕA
=∆A(ϕA
∗
0(1) ⌣ ϕA
∗
0(1) ⌣ ϕA
∗
1(1)) ⊗ ϕB
∗
1(1) ⊗ ϕB
2(xn−1) ⊗ ϕB
∗
∗
2(t) − ϕA
∗
∗
0(t) ⌣ ϕB
∗
2(1)
∗
0(t) ⌣ ϕB
1(1) ⊗ ∆B(ϕB
∗
∗
1(t))
∗
2(t))
− k∆A(ϕA
∗
2(xn−1)) ⊗ ϕB
1(t2) − kϕA
∗
2(xn−1) ⊗ ∆B(ϕB
∗
1(t2))
= − ϕA
∗
0(xn−1) ⊗ ϕB
∗
2(t) − kϕA
∗
2(xn−1) ⊗ ϕB
∗
0(t)
∆(tc) = −xn−1t(b + ka)
∆(ac) =∆(ϕA
∗
2(1) ⌣ ϕA
∗
2(1) ⌣ ϕA
∗
3(1)) ⊗ ϕB
− kϕA
=∆A(ϕA
∗
0(1) ⌣ ϕB
∗
1(1) ⊗ ϕB
2(xn−1) ⊗ ϕB
∗
∗
2(1) − ϕA
∗
1(t) − kϕA
∗
2(1)
∗
0(1) ⌣ ϕB
3(1) ⊗ ∆B(ϕB
∗
∗
1(t))
∗
2(1))
− k∆A(ϕA
∗
4(xn−1)) ⊗ ϕB
= − (2n + 1)ϕA
∗
2(xn−1) ⊗ ϕB
∗
2(1)
∆(ac) = −xn−1ab
∗
4(xn−1) ⊗ ∆B(ϕB
∗
1(t))
∆(bc) =∆(ϕA
∗
0(1) ⌣ ϕA
∗
0(1) ⌣ ϕA
∗
1(1)) ⊗ ϕB
− kϕA
=∆A(ϕA
∗
2(1) ⌣ ϕB
∗
1(1) ⊗ ϕB
∗
2(xn−1) ⊗ ϕB
∗
4(1) − ϕA
∗
3(t) − kϕA
∗
2(1)
∗
2(1) ⌣ ϕB
1(1) ⊗ ∆B(ϕB
∗
∗
1(t))
∗
4(1))
− k∆A(ϕA
∗
2(xn−1)) ⊗ ϕB
∗
2(xn−1) ⊗ ∆B(ϕB
∗
3(t))
∗
= − ϕA
0(xn−1) ⊗ ϕB
∆(bc) = −xn−1b2
∗
4(1) + 2kmϕA
∗
2(xn−1) ⊗ ϕB
∗
2(1)
30
Using equation 9 and induction on powers of x, t, a, b and c, we have
∆(xitjalbrcs) = sxi−1tjalbr((i − 1)b − jka)
7 BV-Algebra Structure on HH ∗(R[Zk])
From now on, we assume that A is R[Z] ∼= R[t, t−1] with R a commutative ring.
Proposition 7.1. The following is a Ae-projective resolution of A
P(A) : 0 → A ⊗ A d1−→ A ⊗ A
µ
−→ A → 0
(11)
with µ(a ⊗ b) = ab and d1(a ⊗ b) = (a ⊗ b)(1 ⊗ t − t ⊗ 1).
Proof. From the definition, it follows that µd1 = 0. Now, we are going to define the following
A-right maps
0 ← A ⊗ A s1←− A ⊗ A s0←− A ← 0
s0(a) = 1 ⊗ a
s1(ti ⊗ 1) =
0
tj ⊗ ti−j−1
if i ≥ 1
t−j−1 ⊗ ti+j
if i = 0
if i ≤ −1
−
i−1Xj=0
−i−1Xj=0
By direct calculations, it follows that µs0 = id and dk+1sk+1+skdk = id for all k ≥ 0. Therefore,
the complex is acyclic.
Tensoring this resolution by A as Ae-modules, we obtain the complex
Taking HomAe(−, A) of P(A), we obtain the complex
A 0−→ A → 0
0 → A 0−→ A
Then
HHi(A; A) = HH i(A; A) =
A if i = 0, 1
0
otherwise
To calculate the cup product, we define ∆P(A) : P(A) −→ P(A) ⊗
A
P(A) as follows
∆P(A)0 : A2 −→ A2 ⊗
A
A2
a ⊗ b 7−→ a ⊗ 1 ⊗
A
1 ⊗ b
∆P(A)1 : A2 −→ A2 ⊗A A2 ⊕ A2 ⊗
A
A2
a ⊗ b 7−→ (a ⊗ 1 ⊗
A
1 ⊗ b, a ⊗ 1 ⊗
A
1 ⊗ b)
By direct computations, it follows that ∆P(A) is a diagonal approximation map.
31
(12)
(13)
(14)
Proposition 7.2. As algebras,
where x, x−1 ∈ HH 0(A; A) and y ∈ HH 1(A; A).
HH ∗(R[Z]; R[Z]) ∼= R[x, x−1] ⊗ Λ(y)
Proof. Using the diagonal approximation map (14), it can be checked that the cup product is
given by multiplication in degrees 0 and 1, and 0 in degrees greater than 2. Therefore, taking
x, x−1 ∈ HH 0(A; A) to be t, t−1 ∈ A and y ∈ HH 1(A; A) to be 1 ∈ A, we get the desire
isomorphism of algebras.
From the definition of the action 2.2 and the diagonal map 14 follows that
Lemma 7.3. The action of HH ∗(A; A) on HH1(A; A) is given by
ρ : HH1(A; A) ⊗ HH ∗(A; A) −→ HH1−∗(A; A)
a ⊗ b 7−→ (−1)bab
Let ψ : P(A) → B(A) and ϕ : B(A) → P(A) be the chain maps defined as follows
· · ·
· · ·
/ 0
0
0
0
A2
d1
A2
µ
A
ψ1
ϕ1
ψ0
ϕ0
/ A ⊗ ¯A2 ⊗ A
∂2 /
A ⊗ ¯A ⊗ A
∂1
A2
∂0 /
A
/ 0
/ 0
ψ0 ≡ id
ϕ0 ≡ id
ψ1(1 ⊗ 1) = −1 ⊗ t ⊗ 1
ϕ1(1 ⊗ tk ⊗ 1) =
0
−
k−1Xj=0
−k−1Xj=0
tj ⊗ tk−j−1
if k ≥ 1
t−j−1 ⊗ tk+j
if k = 0
if k ≤ −1
A ∼= A
Proposition 7.4. Using the identifications An+2 ⊗
Ae
the induced maps for ψ∗ and ϕ∗ are
A ∼= An ⊗ A and A2 ⊗
Ae
¯ψ0 ≡ id
¯ψ1 : A → A ⊗ A
a 7→ −a ⊗ t
¯ϕ0 ≡ id
¯ϕ1 : A ⊗ A → A
a ⊗ tk 7→ −katk−1
Using the identifications HomAe(An+2, A) ∼= Hom(An, A) and HomAe(A2, A) ∼= A
the induced maps for ψ∗ and ϕ∗ are
¯ψ∗
0 ≡ id
1 : Hom(A, A) → A
¯ψ∗
f 7→ −f (t)
¯ϕ∗
0 ≡ id
¯ϕ∗
1 : A → Hom(A, A)
a 7→ fa : A → A
tk 7→ −katk−1
32
/
/
/
/
/
/
/
/
/
/
O
O
/
/
O
O
/
O
O
/
The BV-structure on Hochschild cohomology of the group ring of the integers is given by
Theorem 7.5. Let a = utk with u ∈ R× and k ∈ Z. As a BV-algebra,
HH ∗(R[Z]; R[Z]) ∼= R[x, x−1] ⊗ Λ(y)
∆a(xi) = 0
∆a(yxi) = (i + k)xi−1
where x, x−1 ∈ HH 0(A; A) and y ∈ HH 1(A; A).
Proof. Let a ∈ HH1(A; A) ∼= R[Z] and ρa be the map defined as follows
ρa : HH ∗(A; A) −→ HH1−∗(A; A)
b 7−→ ρ(a ⊗ b)
Since the action is given by multiplication, ρa is an isomorphism for any unit a ∈ R[Z]. Even
more, any unit in R[Z] is of the form a = utk with u ∈ R× and k ∈ Z. By theorem 2.6,
HH ∗(A; A) is a BV-algebra and the BV-operator ∆a is given by
∆a : HH ∗(A; A)
/ HH ∗−1(A; A)
ρa
ρ−1
a
HH1−∗(A; A)
B /
/ HH1−(∗−1)(A; A)
By degree reasons ∆a(xi) = 0 and ∆a(yxi) is given by
ti ρa7−→ −uti+k
¯ψ∗
7−→ −uti+k B7−→ −u ⊗ ti+k ¯ϕ1
0
7−→ u(i + k)ti+k−1 ρ−1
a7−−→ (i + k)ti−1
In [15], Menichi calculates the BV-algebra structure of the homology of the free loop space
of S1
Theorem 7.6 ([15] Theorem 10). As a BV-algebra,
H∗(LS1; R) ∼= R[x, x−1] ⊗ Λ(z)
∆(xi) = 0
∆(zxi) = ixi
where x = 0 and z = −1.
This BV-algebra and the BV-algebra of the Hochschild cohomology of the group ring of the
integers are related by
Corollary 7.7. There is an isomorphism of BV-algebras
φ : H∗(LS1; R)
∼=−→ HH ∗(R[Z]; R[Z])
33
/
O
O
Proof. By theorem 7.5, HH ∗(R[Z]; R[Z]) can be endowed with many BV-algebra structures as
units in R[Z]. For the existence of this isomorphism, we are considering the BV-operator given
by the unit a = t−1. Then as a BV-algebra
HH ∗(R[Z]; R[Z]) ∼= R[x, x−1] ⊗ Λ(y)
∆(yrxi) = r(i − 1)xi−1
The isomorphism φ is defined as follows
φ(x) = x
and
φ(z) = yx
It is clear that φ is an isomorphism of graded algebras, and
φ∆(zrxi) = φ(rixi) = rixi = r(i + r − 1)xi+r−1 = ∆(yrxi+r) = ∆φ(zrxi)
Since HH ∗(R[Z]; R[Z]) is R-projective and the resolution P(A) (11) satisfies the conditions
of theorem 3.5. By theorem 3.13, we get
Theorem 7.8. As BV-algebras,
HH ∗(R[Zn]; R[Zn]) = R[x1, x−1
1 , . . . , xn, x−1
r1+···+rk−1 rk(ik − 1)xi1
∆(xi1
1 · · · xin
n yr1
1 · · · yrn
n ) =
(−1)
nXk=1
n ] ⊗ Λ(y1, . . . , yn)
1 · · · xik−1
i
· · · xin
n yr1
k · · · yrn
n
1 · · ·dyrk
where xi = x−1
i
= 0 and yi = 1 for 1 ≤ i ≤ n.
As a corollary, we have
Corollary 7.9. As Gerstenhaber algebras,
HH ∗(R[Zn]; R[Zn]) = R[x1, x−1
1 , . . . , xn, x−1
n ] ⊗ Λ(y1, . . . , yn)
where xi = x−1
i
= 0 and yi = 1 for 1 ≤ i ≤ n. The bracket is generated by
{xr
i , xs
j} = 0,
{yi, yj} = 0,
and
{xr
i , yj} = −rδijxr−1
i
Let G be a finitely generated abelian group. Then G can be decomposed as G ∼= Zn ⊕ H
with H a finite abelian group. Therefore,
R[G] ∼= R[Zn] ⊗ R[H]
By theorem 3.1, there is an isomorphism of Gerstenhaber algebras
HH ∗(R[G]; R[G]) ∼= HH ∗(R[Zn]; R[Zn]) ⊗ HH ∗(R[H]; R[H])
Corollary 7.10. Let G be a finitely generated abelian group. Then as a BV-algebra
HH ∗(R[G]; R[G]) ∼= HH ∗(R[Zn]; R[Zn]) ⊗ HH ∗(R[H]; R[H])
with BV-operator given by
∆ = ∆Zn
⊗ id ± id ⊗ ∆H
where ∆Zn is given by theorem 7.8 and ∆H is the BV-operator for the finite group H.
34
References
[1] A. Angel and D. Duarte, "The BV-Algebra Structure of the Hochschild Cohomology of
the Group Ring of Cyclic Groups of Prime Order," Geometric, Algebraic and Topological
Methods for Quantum Field Theory, pp. 353 -- 372, 2016.
[2] D. Burghelea and Z. Fiedorowicz, "Cyclic Homology and Algebraic K-Theory of Spaces
II," Topology, vol. 25, no. 3, pp. 303 -- 317, 1986.
[3] C. Cibils and A. Solotar,
"Hochschild cohomology algebra of abelian groups,"
[Online]. Available:
1997.
Archiv der Mathematik, vol. 68, no. 1, pp. 17 -- 21,
http://dx.doi.org/10.1007/PL00000389
[4] R. L. Cohen and J. D. S. Jones, "A Homotopy Theoretic Realization of String
[Online]. Available:
Topology," Math. Ann., vol. 324, no. 4, pp. 773 -- 798, 2002.
http://dx.doi.org/10.1007/s00208-002-0362-0
[5] R. L. Cohen, J. D. S. Jones, and J. Yan, "The Loop Homology Algebra of Spheres and
Projective Spaces," in Categorical decomposition techniques in algebraic topology (Isle of
Skye, 2001), ser. Progr. Math. Birkhauser, Basel, 2004, vol. 215, pp. 77 -- 92.
[6] Y. F´elix, L. Menichi, and J. C. Thomas, "Gerstenhaber Duality in Hochschild
[Online].
Cohomology," J. Pure Appl. Algebra, vol. 199, no. 1-3, pp. 43 -- 59, 2005.
Available: http://dx.doi.org/10.1016/j.jpaa.2004.11.004
[7] Y. F´elix and J. C. Thomas, "Rational BV-Algebra in String Topology," Bull. Soc. Math.
France, vol. 136, no. 2, pp. 311 -- 327, 2008.
[8] M. Gerstenhaber, "The Cohomology Structure of an Associative Ring," Ann. of Math.
(2), vol. 78, pp. 267 -- 288, 1963. [Online]. Available: http://dx.doi.org/10.2307/1970343
[9] V. Ginzburg, "Calabi-Yau Algebras," arXiv preprint math/0612139, 2006.
[10] B. A. C. H. Group, "Cyclic Homology of Algebras with one Generator," K-Theory,
vol. 5, no. 1, pp. 51 -- 69, 1991, jorge A. Guccione, Juan Jos´e Guccione, Mar´ıa Julia
Redondo, Andrea Solotar and Orlando E. Villamayor participated in this research.
[Online]. Available: http://dx.doi.org/10.1007/BF00538879
[11] G. Hochschild, "On the Cohomology Groups of an Associative Algebra," Ann. of Math.
(2), vol. 46, pp. 58 -- 67, 1945. [Online]. Available: http://dx.doi.org/10.2307/1969145
[12] T. Holm, "The Hochschild Cohomology Ring of a Modular Group Algebra:
Commutative Case," Comm. Algebra, vol. 24, no. 6, pp. 1957 -- 1969, 1996.
Available: http://dx.doi.org/10.1080/00927879608825682
the
[Online].
[13] J. Le and G. Zhou, "On the Hochschild Cohomology Ring of Tensor Products of
Algebras," J. Pure Appl. Algebra, vol. 218, no. 8, pp. 1463 -- 1477, 2014. [Online]. Available:
http://dx.doi.org/10.1016/j.jpaa.2013.11.029
35
[14] L. Menichi, "Batalin-Vilkovisky Algebra Structures on Hochschild Cohomology," Bull. Soc.
Math. France, vol. 137, no. 2, pp. 277 -- 295, 2009.
[15] -- -- , "String Topology for Spheres," Comment. Math. Helv., vol. 84, no. 1, pp.
135 -- 157, 2009, with an appendix by Gerald Gaudens and Menichi. [Online]. Available:
http://dx.doi.org/10.4171/CMH/155
[16] C. Negron and S. Witherspoon, "An Alternate Approach to the Lie Bracket on Hochschild
Cohomology," Homology Homotopy Appl., vol. 18, no. 1, pp. 265 -- 285, 2016. [Online].
Available: http://dx.doi.org/10.4310/HHA.2016.v18.n1.a14
[17] S. S´anchez-Flores, "The Lie Structure on the Hochschild Cohomology of a Modular
Group Algebra," J. Pure Appl. Algebra, vol. 216, no. 3, pp. 718 -- 733, 2012. [Online].
Available: http://dx.doi.org/10.1016/j.jpaa.2011.08.007
[18] T. Tradler, "The BV Algebra on Hochschild Cohomology Induced by Infinity Inner
Products," Ann. Inst. Fourier (Grenoble), vol. 58, no. 7, pp. 2351 -- 2379, 2008. [Online].
Available: http://aif.cedram.org/item?id=AIF 2008 58 7 2351 0
[19] C. Westerland, "Dyer-lashof Operations in the String Topology of Spheres and
Projective Spaces," Math. Z., vol. 250, no. 3, pp. 711 -- 727, 2005. [Online]. Available:
http://dx.doi.org/10.1007/s00209-005-0778-9
[20] -- -- , "String Homology of Spheres and Projective Spaces," Algebr. Geom. Topol., vol. 7,
pp. 309 -- 325, 2007. [Online]. Available: http://dx.doi.org/10.2140/agt.2007.7.309
[21] T. Yang, "A Batalin-Vilkovisky Algebra Structure on the Hochschild Cohomology of
Truncated Polynomials," Topology Appl., vol. 160, no. 13, pp. 1633 -- 1651, 2013. [Online].
Available: http://dx.doi.org/10.1016/j.topol.2013.06.010
36
|
1705.01039 | 3 | 1705 | 2018-03-31T13:00:21 | Polynomial bound for the nilpotency index of finitely generated nil algebras | [
"math.RA",
"math.AC",
"math.RT"
] | Working over an infinite field of positive characteristic, an upper bound is given for the nilpotency index of a finitely generated nil algebra of bounded nil index $n$ in terms of the maximal degree in a minimal homogenous generating system of the ring of simultaneous conjugation invariants of tuples of $n$ by $n$ matrices. This is deduced from a result of Zubkov. As a consequence, a recent degree bound due to Derksen and Makam for the generators of the ring of matrix invariants yields an upper bound for the nilpotency index of a finitely generated nil algebra that is polynomial in the number of generators and the nil index. Furthermore, a characteristic free treatment is given to Kuzmin's lower bound for the nilpotency index. | math.RA | math |
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF
FINITELY GENERATED NIL ALGEBRAS
M. DOMOKOS
Abstract. Working over an infinite field of positive characteristic, an upper bound is
given for the nilpotency index of a finitely generated nil algebra of bounded nil index n in
terms of the maximal degree in a minimal homogenous generating system of the ring of
simultaneous conjugation invariants of tuples of n by n matrices. This is deduced from a
result of Zubkov. As a consequence, a recent degree bound due to Derksen and Makam for
the generators of the ring of matrix invariants yields an upper bound for the nilpotency
index of a finitely generated nil algebra that is polynomial in the number of generators
and the nil index. Furthermore, a characteristic free treatment is given to Kuzmin's lower
bound for the nilpotency index.
1. Introduction
Throughout this note F stands for an infinite field of positive characteristic. All vector
spaces, tensor products, algebras are taken over F. The results of this paper are valid
in arbitrary characteristic, but they are known in characteristic zero (in fact stronger
statements hold in characteristic zero, see Formanek [10], giving in particular an account
of relevant works of Razmyslov [23] and Procesi [22]).
Write Fm := Fhx1, . . . , xmi for the free associative F-algebra with identity 1 on m gen-
erators x1, . . . , xm, and let F +
m be its ideal generated by x1, . . . , xm (so F +
m is the free non-
unitary associative algebra of rank m). For a positive integer n denote by In,m the ideal
in Fm generated by {an a ∈ F +
m}. A theorem of Kaplansky [14] asserts that if a finitely
generated associative algebra satisfies the polynomial identity xn = 0, then it is nilpotent.
Equivalently, there exists a positive integer d such that for all i1, . . . , id ∈ {1, . . . , m} the
monomial xi1 · · · xid belongs to In,m. Denote by dF(n, m) the minimal such d. In other
2010 Mathematics Subject Classification. Primary: 16R10 Secondary: 16R30, 13A50, 15A72.
Key words and phrases. nil algebra, nilpotent algebra, matrix invariant, degree bound.
This research was partially supported by National Research, Development and Innovation Office, NK-
FIH K 119934.
1
2
M. DOMOKOS
words, dF(n, m) is the minimal positive integer d such that all F-algebras that are gener-
ated by m elements and satisfy the polynomial identity xn = 0 satisfy also the polynomial
identity y1 · · · yd = 0. This is a notable quantity of noncommutative ring theory: Jacobson
[13] reduced the Kurosh problem for finitely generated algebraic algebras of bounded de-
gree to the case of nil algebras of bounded degree. We mention also that proving nilpotency
of nil rings under various conditions is a natural target for ring theorists, see for example
the paper of Guralnick, Small and Zelmanov [11].
The number dF(n, m) is tightly connected with a quantity appearing in commutative
invariant theory defined as follows. Consider the generic matrices
Xr = (xij(r))1≤i.j≤n,
r = 1, . . . , m.
These are elements in the algebra An×n of n×n matrices over the mn2-variable commutative
polynomial algebra A = F[xij(r) 1 ≤ i, j ≤ n, 1 ≤ r ≤ m]. The general linear group
GLn(F) acts on A via F-algebra automorphisms: for g ∈ GLn(F) we have that g · xij(r)
is the (i, j)-entry of the matrix g−1Xrg. Set Rn,m = AGLn(F), the subalgebra of GLn(F)-
invariants. This is the algebra of polynomial invariants under simultaneous conjugation
of m-tuples of n × n matrices. The polynomial ring A is graded in the standard way,
and since the GLn(F)-action preserves the grading, the subalgebra Rn,m is generated by
homogeneous elements. Being the algebra of invariants of a reductive group, Rn,m is finitely
generated by the Hilbert-Nagata theorem (see for example [21]). We write βF(n, m) for
the minimal positive integer d such that the F-algebra Rn,m is generated by elements of
degree at most d. The main result of the present note is the following inequality:
Theorem 1.1. We have the inequality
dF(n, m) ≤ βF(n, m + 1).
Remark 1.2. In the reverse direction it was shown in [6, Theorem 3] that for n ≥ 2 we
have
βF(n, m) ≤ ⌊
n
2
⌋dF(n, m).
Theorem 1.1 is derived from a theorem of Zubkov [24] (for which Lopatin [19] gave
versions and improvements), see Theorem 2.1. Using a result of Ivanyos, Qiao and Sub-
rahmanyam [12], Derksen and Makam [4] found strong bounds on the degrees of invariants
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
3
defining the null-cone of m-tuples of n × n matrices under simultaneous conjugation, and
derived from this the following upper bound on βF(n, m):
Theorem 1.3. (Derksen and Makam [5, Theorem 1.4]) We have the inequality
βF(n, m) ≤ (m + 1)n4.
Given this result Derksen and Makam [5, Conjecture 1.5] conjectured that there exists
an upper bound on dF(n, m) that is polynomial in n and m. Combining Theorem 1.1 and
Theorem 1.3 we obtain the following affirmative answer to this conjecture:
Corollary 1.4. We have the inequality
dF(n, m) ≤ (m + 2)n4.
Remark 1.5. Corollary 1.4 is a drastic improvement of the earlier known general upper
bounds on dF(n, m):
(1) dF(n, m) ≤ n6mn+1 by Belov [1].
6n6mn by Klein [15].
(2) dF(n, m) ≤ 1
(3) dF(n, m) ≤ 218mn12 log3(n)+28 by Belov and Kharitonov [2].
It is easy to see that dF(2, m) ≤ m + 1. We note that for the case n = 3 exact results on
dF(3, m) were obtained by Lopatin [17]. Moreover, Lopatin [18] proved that if char(F) > n
2
then dF(n, m) ≤ n1+log2(3m+2) and dF(n, m) ≤ 22+ n
2 m.
Remark 1.6. When char(F) > n2 +1, we have βF(n, m) ≤ n2. Indeed, the proof presented
by Formanek [9] (following the original arguments of Razmyslov [23] and Procesi [22]) for
the zero characteristic case of the corresponding inequality goes through without essential
changes when chat(F) > n2 + 1. Thus by Theorem 1.1 we get that dF(n, m) ≤ n2 when
char(F) > n2 + 1.
In Section 3 we show that the following lower bound for dF(n, m) due to E. N. Kuzmin
[16] when char(F) = 0 or char(F) > n holds in arbitrary characteristic:
Theorem 1.7. The monomial x2x1x2x2
1 · · · x2xn−1
In particular, for m ≥ 2 we have dF(n, m) ≥ n(n + 1)/2.
1x2x3
is not contained in the ideal In,2.
1
4
M. DOMOKOS
Remark 1.8. It is well known that when 0 < char(F) ≤ n, the element x1x2 · · · xm is not
contained in In,m, see for example [20, 5. Remarks. (I)]. So in this case for m ≥ 2 we have
max{m + 1, n(n + 1)/2} ≤ dF(n, m) ≤ (m + 2)n4.
2. Identities of matrices with forms
The map xi 7→ Xi (i = 1, . . . , m) extends to a unique F-algebra homomorphism ϕ1 :
Fm → An×n. We have ϕ1(1) = I, the n × n identity matrix. Consider the commutative
polynomial algebra
Pn,m = F[sl(a) a ∈ F +
m, l = 1, . . . , n]
generated by the infinitely many commuting indeterminates sl(a). Define the F-algebra
homomorphism
ϕ2 : Pn,m → Rn,m, ϕ2(sl(a)) = σl(ϕ1(a))
where for B ∈ An×n we have
det(tI + B) =
tlσn−l(B),
nXl=0
so σl(B) is the sum of the principal l × l minors of B. A theorem of Donkin [7] asserts
that ϕ2 is surjective onto Rn,m. Combining ϕ1 and ϕ2 we get an F-algebra homomorphism
ϕ : Pn,m ⊗ Fm → An×n,
b ⊗ a 7→ ϕ2(b)ϕ1(a).
The subalgebra Cn,m = ϕ(Pn,m⊗Fm) is called the algebra of matrix concomitants. It can be
interpreted as the algebra of GLn(F)-equivariant polynomial maps (Fn×n)m → Fn×n, where
GLn(F) acts on Fn×n by conjugation and on the space (Fn×n)m of m-tuples of matrices by
simultaneous conjugation. For a ∈ F +
m define an element χn(a) in Pn,m ⊗ Fm as follows:
χn(a) =
nXl=0
(−1)lsl(a) ⊗ an−l
(where s0(a) = 1). We need the following result of Zubkov [24] (see also Lopatin [19,
Theorem 2.4]):
Theorem 2.1. (Zubkov [24]) The ideal ker(ϕ) is generated by
{b ⊗ 1, χn(a) b ∈ ker(ϕ2), a ∈ F +
m}.
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
5
Remark 2.2. The papers [24] and [19] use different commutative polynomial algebras than
our Pn,m, however, it is straightforward that Theorem 2.1 is an immediate consequence
of the versions stated in [24], [19]. We note that [24], [19] give descriptions of the ideal
ker(ϕ2) as well. A self-contained approach to the theorem of Zubkov can be found in the
recent book by De Concini and Procesi [3].
Denote by η : Cn,m → Cn,m/R+
n,mCn,m the natural surjection (ring homomorphism),
where R+
n,m is the sum of the positive degree homogeneous components of Rn,m.
Corollary 2.3. The kernel of η ◦ ϕ1 is the ideal In,m = (an a ∈ F +
m) in Fm.
Proof. We have ker(η ◦ ϕ1) = ker(η ◦ ϕ) ∩ Fm (where we identify Fm with the subalgebra
1 ⊗ Fm in Pn,m ⊗ Fm). The ideal (sl(a) ⊗ 1 a ∈ F +
m, 1 ≤ l ≤ n) is mapped surjectively
onto R+
n,mCn,m by [7]. Therefore we have
ker(η ◦ ϕ) = ϕ−1(R+
n,mCn,m) = ker(ϕ) + (sl(a) ⊗ 1 a ∈ F +
m, 1 ≤ l ≤ n)
(the last equality follows from Theorem 2.1 and the fact that 1 ⊗ an − χn(a) belongs to
= (sl(a) ⊗ 1, 1 ⊗ an a ∈ F +
m, 1 ≤ l ≤ n)
(sl(a) ⊗1 a ∈ F +
m, 1 ≤ l ≤ n)). Obviously the ideal (sl(a) ⊗1, 1 ⊗an a ∈ F +
m, 1 ≤ l ≤ n)
intersects Fm in In,m.
(cid:3)
Remark 2.4. Corollary 2.3 implies that the relatively free algebra Fm/In,m is isomorphic
to Cn,m/R+
n,mCn,m. When char(F) = 0, this statement is due to Procesi [22, Corollary 4.7].
The algebras Rn,m and Cn,m are Zm-graded:
degm(Xi1 · · · Xid) = (α1, . . . , αm) where αk = {j ij = k}
and
degm(σl(Xi1 · · · Xid)) = l · degm(Xi1 · · · Xid).
Proof of Theorem 1.1. Set d = βF(n, m + 1). We have to show that xi1 · · · xid ∈ In,m for all
i1, . . . , id ∈ {1, . . . , m}. Recall that by [7] the algebra Rn,m+1 is generated by the elements
σl(W ), where W is a word in X1, . . . , Xm+1, and l ∈ {1, . . . , n}. The total degree of the
6
M. DOMOKOS
element Tr(Xi1 · · · XidXm+1) ∈ Rn,m+1 is strictly greater than βF(n, m + 1), whence we
have a relation
(1)
Tr(Xi1 · · · XidXm+1) =Xλ∈Λ
aλfλ
where Λ is a finite index set, aλ ∈ F, and each fλ ∈ Rn,m+1 is a product fλ = σl1(W1) · · · σlr (Wr)
with r ≥ 2 and W1, . . . , Wr non-empty words in X1, . . . , Xm+1. The Zm+1-multidegree of
Tr(Xi1 · · · XidXm+1) is
degm+1(Tr(Xi1 · · · XidXm+1)) = (degm(Tr(Xi1 · · · Xid)), 1).
The terms fλ are all Zm+1-homogeneous, whence we may assume that each has the above
Zm+1-degree (since the other possible terms on the right hand side of (1) must cancel each
other). It follows that for each fλ exactly one of its factors σl1(W1), . . . , σlr (Wr) has Zm+1-
degree of the form (α1, . . . , αm, 1), say this is σl1(W1), and the remaining factors have Zm+1-
degree of the form (γ1, . . . , γm, 0). Necessarily we have l1 = 1 and so σl1(W1) = Tr(Xm+1Z)
for some (possibly empty) word Z in X1, . . . , Xm, and W2, . . . , Wr are non-empty words in
X1, . . . , Xm. Set
gλ = σl2(W2) · · · σlr (Wr)Z ∈ Cn,m,
and note that fλ = Tr(gλXm+1). Using linearity of Tr(−) relation (1) can be written as
(2)
Tr(Xm+1(Xi1 · · · Xid −Xλ∈Λ
aλgλ)) = 0 ∈ Rn,m+1.
Substituting Xm+1 7→ Eij (the matrix whose (i, j)-entry is 1 and all other entries are 0) we
get from (2) that the (j, i)-entry of Xi1 · · · Xid −Pλ∈Λ aλgλ is 0. This holds for all (i, j),
thus we have the equality
(3)
Xi1 · · · Xid =Xλ
aλgλ.
The right hand side of (3) is obviously contained in R+
n,mCn,m, therefore it follows from (3)
that the element xi1 · · · xid ∈ Fm belongs to the kernel of η ◦ ϕ1. Thus by Corollary 2.3 we
conclude that xi1 · · · xid ∈ In,m.
(cid:3)
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
7
3. Lower bound
Kuzmin's proof of the case char(F) = 0 or char(F) > n of Theorem 1.7 (it is presented
also in the survey of Drensky in [8]) uses crucially Lemma 3.1 below, relating the complete
linearization of xn, namely
Pn(x1, . . . , xn) = Xπ∈Sym{1,...,n}
xπ(1)xπ(2) · · · xπ(n) ∈ Fn.
Lemma 3.1. If char(F) = 0 or char(F) > n, then In,m is spanned as an F-vector space
by the elements Pn(w1, . . . , wn), where w1, . . . , wn range over all non-empty monomials in
x1, . . . , xm.
Remark 3.2. The assumption on char(F) in Lemma 3.1 is necessary, its statement obvi-
ously fails if 0 < char(F) ≤ n (as it can be easily seen already in the special case m = 1).
Now we modify the arguments of Kuzmin to obtain Theorem 1.7 in a characteristic free
manner. It turns out that although Lemma 3.1 can not be applied, the main combinatorial
ideas of Kuzmin's proof do work.
Consider the free Z-algebra Z = Zhx, yi+ without unity. Write M for the set of non-
empty monomials (words) in x, y. For a positive integer k write Z(k) for the Z-submodule
of Z generated by the w ∈ M whose total degree in y is k − 1. It will be convenient to
use the following notation: for (a1, . . . , ak) ∈ Nk
0 set
[a1, . . . , ak] = xa1yxa2y · · · yxak ∈ M.
The symmetric group Sk = Sym{1, . . . , k} acts on the right linearly on Z(k), extending
linearly the permutation action on Z(k) ∩ M given by
[a1, . . . , ak]π = [aπ(1), . . . , aπ(k)]
for π ∈ Sk.
Let B denote the Z-submodule of Z generated by all the elements [a1, . . . , ak] (k ∈ N)
such that ai ≥ n for some i ∈ {1, . . . , k} or ai = aj for some 1 ≤ i < j ≤ k, and by all
the elements of the form [a1, . . . , ak] + [a1, . . . , ak](ij) where (ij) denotes the transposition
interchanging i and j for 1 ≤ i < j ≤ k. We shall use the following obvious properties of
B:
Lemma 3.3.
(i) The Z-submodule B ∩ Z(k) of Z(k) is Sk-stable.
8
M. DOMOKOS
(ii) We have the inclusions yB ⊂ B, ZyB ⊂ B, By ⊂ B, and ByZ ⊂ B.
(iii) Let k be a positive integer, u1, . . . , uk−1 ∈ M monomials such that ui ∈ yZ ∩ Zy
or ui = y for i = 1, . . . , k − 1. Then B contains the image of the Z-module map on
B ∩ Z(k) given by
[a1, . . . , ak] 7→ xa1u1xa2u2xa3 · · · uk−1xak .
(iv) For any positive integer a, the Z-submodule B of Z is preserved by the derivation
δa on Z defined by δa(x) = xa, δa(y) = 0.
(v) The factor Z/B is a free Z-module freely generated by the images under the natural
surjection Z → Z/B of the monomials
cM = {[a1, . . . , ak] k ∈ N,
0 ≤ a1 < a2 < · · · < ak ≤ n − 1}.
Proof. Statements (i), (ii), (iii), (iv) are immediate consequences of the construction of B.
0 ≤ c1 ≤ · · · ≤ ck, and Z(c1, . . . , ck) stands for the Z-submodule generated by [c1, . . . , ck]π
To prove (v) note that Z =L Z(c1, . . . , ck) where the direct sum is taken over k ∈ N and
as π ranges over Sk. Moreover, B =L B(c1, . . . , ck) where B(c1, . . . , ck) = B∩Z(c1, . . . , ck).
Now Z(c1, . . . , ck) ⊂ B if ci = cj for some i 6= j or if ci ≥ n for some i. It is also clear
that for 0 ≤ a1 < · · · < ak we have Z(a1, . . . , ak) = Z · [a1, . . . , ak] + B(a1, . . . , ak),
so the monomials in cM generate the Z-module Z modulo B. Suppose that some non-
trivial Z-linear combination of the elements in cM belongs to B. The above direct sum
decompositions of Z and B imply then that there exist q, k ∈ N, and 0 ≤ a1 < · · · < ak ≤
n − 1 such that q[a1, . . . , ak] ∈ B(a1, . . . , ak). This means that
(4)
q[a1, . . . , ak] =
εi(wi + wπi
i )
sXi=1
where εi = ±1, wi ∈ Z(a1, . . . , ak) ∩ M and πi ∈ Sk is a transposition for i = 1, . . . , s.
Suppose that s in (4) is minimal possible. Without loss of generality we may assume
that w1 = [a1, . . . , ak] and ε1 = 1. The word wπ1
εi(wi + wπi
2 ) = −(wπ1
have ε2(w2 + wπ2
some summand εi(wi + wπi
1 + wπ1π2
1 must be canceled by w1 or by
i ) with i ≥ 3. It means that the right hand side of (4) has a
i ) with i ≥ 2 on the right hand side of (4), so after a possible renumbering we
1 must be canceled by some summand
). Now the term −wπ1π2
1
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
9
subsum of the form
(5)
(w1 + wπ1
1 ) − (wπ1
1 + wπ1π2
) + (wπ1π2
1 + wπ1π2π3
1
) − + · · · + (−1)r−1(wπ1···πr−1
1
+ wπ1···πr
1
)
1
where wπ1···πr
1
= w1. This latter equality forces that π1 · · · πr is the identity permutation,
so r is even, and then the sum (5) is zero. So all these terms can be omitted from (4). This
contradicts the minimality of s. This shows that q[a1, . . . , ak] is not contained in B.
(cid:3)
Lemma 3.4. Let k be a positive integer, a1 ≤ a2 ≤ · · · ≤ ak ∈ N0, and r ∈ N0 with
a1 + k + r > n. Then
(6)
Xc1+···+ck=rXπ∈Sk
[a1 + cπ(1), . . . , ak + cπ(k)] ∈ B.
Proof. Apply induction on k. In the case k = 1 the element in question in (6) is xa1+r,
which belongs to B by the assumption a1 + 1 + r > n. Suppose next that k > 1, and the
statement of the lemma holds for smaller k. The terms [a1 + d1, . . . , ak + dk] in the sum
(6) can be grouped into three classes:
(A) a1 + d1 < a2
(B) a1 + d1 = a2 + d2
(C) a1 + d1 ≥ a2 and a1 + d1 6= a2 + d2.
The sum of the terms of type (A) is a sum of expressions of the form
(7)
xa1+d1y Xc2+···+ck=r−d1 Xπ∈Sym{2,...,k}
[a2 + cπ(2), . . . , ak + cπ(k)].
Here a2 + (k − 1) + (r − d1) ≥ a1 + k + r > n, hence by the induction hypothesis
Pc2+···+ck=r−d1Pπ∈Sym{2,...,k}[a2 + cπ(2), . . . , ak + cπ(k)] belongs to B. Now by Lemma 3.3 (ii)
we conclude that the element in (7) belongs to B. The terms of type (B) belong to B by
construction of B. Finally, a term [a1 + d1, . . . , ak + dk] of type (C) can be paired off with
the term [a1 + e1, a2 + e2, a3 + d3, . . . , ak + dk] where e1 = a2 − a1 + d2 and e2 = a1 − a2 + d1
(so this is also of type (C)), and the sum of these two terms belongs to B by construction
of B.
(cid:3)
10
M. DOMOKOS
Corollary 3.5. Let k be a positive integer, (a1, . . . , ak) ∈ Nk
0, and r ∈ N0 with r + k > n.
Then
Xc1+···+ck=rXπ∈Sk
[a1 + cπ(1), . . . , ak + cπ(k)] ∈ B.
Proof. Take a permutation ρ ∈ Sk such that aρ(1) ≤ · · · ≤ aρ(k). Applying ρ to the element
in the statement we get
Xc1+···+ck=rXπ∈Sk
[aρ(1) + cπ(1), . . . , aρ(k) + cπ(k)],
which belongs to B ∩ Z(k) by Lemma 3.4. Our statement follows by Lemma 3.3 (i). (cid:3)
Lemma 3.6. Suppose 1 ≤ k ≤ n + 1, w1, . . . , wk−1 ∈ M are monomials having positive
degree in y, and a, b ∈ N0. Then
(8)
xaPn(w1, . . . , wk−1, x, . . . , x)xb ∈ B.
Proof. We have wi = xaiuixbi where ai, bi ∈ N0 and ui ∈ yZ∩Zy or ui = y (i = 1, . . . , k−1).
Then the element in (8) is
Xρ∈Sk−1 (n − k + 1)! Xc1+···+ck=n−k+1Xπ∈Sk
xd1+cπ(1)uρ(1)xd2+cπ(2)uρ(2) · · · xdk−1+cπ(k−1)uρ(k−1)xdk+cπ(k)!
where d1 = a + aρ(1), d2 = aρ(2) + bρ(1), d3 = aρ(3) + bρ(2), dk−1 = aρ(k−1) + bρ(k−2), dk =
bρ(k−1) + b. The summand corresponding to ρ ∈ Sk−1 in the outer sum is contained in B
by Corollary 3.5 and Lemma 3.3 (iii).
(cid:3)
Lemma 3.7. For any w1, . . . , wn ∈ M, w0, wn+1 ∈ M ∪ {1} we have
w0Pn(w1, . . . , wn)wn+1 ∈ B.
Proof. By Lemma 3.3 (ii) it is sufficient to deal with the case w0 = xa, wn+1 = xb. We may
assume that w1, . . . , wk−1 have positive degree in y and wk−1+j = xcj for j = 1, . . . , n−k+1.
If n − k + 1 = 0 or all the cj = 1 then we are done by Lemma 3.6. Suppose next that
n − k + 1 > 0, c1, . . . , cl > 1 with l ≥ 1, and cl+1 = · · · = cn−k+1 = 1. By induction on l we
show that xaPn(w1, . . . , wk−1, xc1, . . . , xcl, x, . . . , x)xb ∈ B. By the induction hypothesis (or
by Lemma 3.6 when l = 1) f = xaPn(w1, . . . , wk−1, xc1, . . . , xcl−1, x, . . . , x)xb ∈ B, hence by
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
11
Lemma 3.3 (iv) δcl(f ) ∈ B. We have
δcl(f ) = axa+cl−1Pn(w1, . . . , wk−1, xc1, . . . , xcl−1, x, . . . , x)xb
+
+
k−1Xi=1
l−1Xj=1
xaPn(w1, . . . , δcl(wi), . . . , wk−1, xc1, . . . , xcl−1, x, . . . , x)xb
cjxaPn(w1, . . . , wk−1, xc1, . . . , xcj+cl−1, . . . , xcl−1, x, . . . , x)xb
+ (n − k − l + 2)xaPn(w1, . . . , wk−1, xc1, . . . , xcl, x, . . . , x)xb
+ bxaPn(w1, . . . , wk−1, xc1, . . . , xcl−1, x, . . . , x)xb+cl−1.
All other terms than (n − k − l + 2)xaPn(w1, . . . , wk−1, xc1, . . . , xcl, x, . . . , x)xb on the right
hand side above belong to B by the induction hypothesis. Taking into account that Z/B
is torsion free by Lemma 3.3 (v) we conclude the desired inclusion
xaPn(w1, . . . , wk−1, xc1, . . . , xcl, x, . . . , x)xb ∈ B.
(cid:3)
For λ = (λ1, . . . , λm) ∈ Nm
0 denote by Pλ(x1, . . . , xm) ∈ Zhx1, . . . , xmi the multihomoge-
neous component of (x1 + · · · + xm)n having Zm-degree λ.
Corollary 3.8. For any m ∈ N, w1, . . . , wm ∈ M, w0, wm+1 ∈ M ∪ {1} and for any
λ ∈ Nm
0 we have that
w0Pλ(w1, . . . , wm)wm+1 ∈ B.
Proof. We have the equality
Pλ(x1, . . . , xm) =
Pn(x1, . . . , x1
, . . . , xm, . . . , xm
).
i=1(λi!)
1
Qm
Therefore the statement follows from Lemma 3.7 by Lemma 3.3 (v).
(cid:3)
λ1
{z
}
λm
{z
}
Proposition 3.9. The ideal In,2 is contained in the subspace F ⊗Z B of Fhx, yi.
Proof. The ideal In,2 is spanned as an F-vector space by elements of the form
w0(c1w1 + · · · + cmwm)nwm+1,
12
M. DOMOKOS
where the wi are monomials in x, y and they have positive total degree for i = 1, . . . , m,
and c1, . . . , cm ∈ F. Since we have the equality
(c1w1 + · · · + cmwm)n =
Xλ∈Nm
0 , λ1+···+λm=n
cλ1
1 · · · cλm
m Pλ(w1, . . . , wm),
our statement follows from Corollary 3.8.
(cid:3)
Proof of Theorem 1.7. By Lemma 3.3 (v) the monomials
{xa1yxa2yxa3 · · · yxak 0 ≤ a1 < a2 < · · · < ak ≤ n − 1}
are linearly independent in F2 = Fhx, yi modulo the subspace F⊗ZB. Since F⊗ZB contains
the ideal In,2 by Proposition 3.9, our statement follows.
(cid:3)
References
[1] A. J. Belov, Some estimations for nilpotence of nill-algebras over a field of an arbitrary character-
istic and height theorem, Comm. Alg. 20, No. 10 (1992), 2919-2922.
[2] A. Ya. Belov, M. I. Kharitonov, Subexponential estimates in Shirshov's theorem on height (Rus-
sian), Mat. Sb. 203 (4) (2012), 81-102.
[3] C. De Concini, C. Procesi, The invariant theory of matrices, University Lecture Series 69, Amer.
Math. Soc., Providence, Rhode Island, 2017.
[4] H. Derksen, V. Makam, Polynomial degree bounds for matrix semi-invariants, Adv. Math. 310
(2017), 44-63.
[5] H. Derksen, V. Makam, Generating invariant rings of quivers in arbitrary characteristic, J. Algebra
489 (2017), 435-445.
[6] M. Domokos, Finite generating system of matrix invariants, Math. Pannonica 13 (2002), 175-181.
[7] S. Donkin, Invariants of several matrices, Inv. Math. 110 (1992), 389-401.
[8] V. Drensky, E. Formanek, Polynomial Identity Rings, Advanced Courses in Mathematics, CRM
Barcelona, Birkhauser Verlag, Basel, 2004.
[9] E. Formanek, Generating the ring of matrix invariants, in Ring Theory - Proceedings, Antwerpen,
1985, F. van Oystaeyen, Editor, Springer-Verlag Lecture Notes in Math. No. 1197, New York,
1986, 73-82.
[10] E. Formanek, The polynomial identities and invariants of n × n matrices, Regional Conference
Series in Mathematics 78, Providence, RI; American Math. Soc., 55 p., 1991.
[11] R. M. Guralnick, L. W. Small, E. Zelmanov, Nil subrings of endomorphism rings of finitely gen-
erated modules over affine PI-rings, J. Algebra 324 (2010), no. 11, 3044-3047.
POLYNOMIAL BOUND FOR THE NILPOTENCY INDEX OF NIL ALGEBRAS
13
[12] G. Ivanyos, Y. Qiao, K. V. Subrahmanyam, Non-commutative Edmonds' problem and matrix
semi-invariants, Comput. Complex. (2017), 1-47.
[13] N. Jacobson, Structure theory for algebraic algebras of bounded degree, Ann. of Math. (2) 46,
(1945). 695-707.
[14] I. Kaplansky, On a problem of Kurosch and Jacobson, Bull. Amer. Math. Soc. 52 (1946), 496-500.
[15] A. A. Klein, Bounds for indices of nilpotency and nility, Arch. Math. (Basel) 74 (2000), 6-10.
[16] E. N. Kuzmin, On the Nagata-Higman theorem (Russian), pp. 101-107, in "Mathematical Struc-
tures, Computational Mathematics, Mathematical Modelling, Proceedings dedicated to the sixti-
eth birthday of academician L. Iliev, Sofia, 1975.
[17] A. Lopatin, Relatively free algebras with the identity x
3 = 0, Comm. Algebra 33 (2005), no. 10,
3583-3605.
[18] A. Lopatin, On the nilpotency degree of the algebra with identity x
n = 0, J. Algebra 371 (2012),
350-366.
[19] A. Lopatin, Matrix identities with forms, J. Pure Appl. Alg. 217 (2013), 2056-2075.
[20] M. Nagata, On the nilpotency of nil-algebras, J. Math. Soc. Japan 4 (1952), 296-301.
[21] P. E. Newstead, Introduction to Moduli Problems and Orbit Spaces, Tata Institute Lecture Notes,
Springer-Verlag, 1978.
[22] C. Procesi, The invariant theory of n × n matrices, Adv. Math. 19 (1976), 306-381.
[23] Y. P. Razmyslov, Trace identities of full matrix algebras over a field of characteristic 0, (Russian),
Izv. Akad. Nauk SSSR Ser. Mat. 38 (1974), 723-756.
[24] A. N. Zubkov, On a generalization of the Procesi-Razmyslov Theorem (Russian), Algebra i Logika
35 (1996), 433-457.
MTA Alfr´ed R´enyi Institute of Mathematics, Re´altanoda utca 13-15, 1053 Budapest,
Hungary
E-mail address: [email protected]
|
1603.04991 | 1 | 1603 | 2016-03-16T08:52:19 | Embedding in factorisable restriction monoids | [
"math.RA"
] | Each restriction semigroup is proved to be embeddable in a factorisable restriction monoid, or, equivalently, in an almost factorisable restriction semigroup. It is also established that each restriction semigroup has a proper cover which is embeddable in a semidirect product of a semilattice by a group. | math.RA | math |
EMBEDDING IN FACTORISABLE RESTRICTION MONOIDS
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
Abstract. Each restriction semigroup is proved to be embeddable in a factorisable re-
striction monoid, or, equivalently, in an almost factorisable restriction semigroup. It is
also established that each restriction semigroup has a proper cover which is embeddable
in a semidirect product of a semilattice by a group.
1. Introduction
Restriction semigroups are non-regular generalisations of inverse semigroups. They are
semigroups equipped with two additional unary operations which satisfy certain identities.
In particular, each inverse semigroup determines a restriction semigroup where the unary
operations assign the idempotents aa−1 and a−1a, respectively, to any element a. The
class of restriction semigroups is just the variety of algebras generated by these restriction
semigroups obtained from inverse semigroups, see [3]. Restriction semigroups (formerly
called weakly E-ample semigroups) have arisen from a number of mathematical perspec-
tives. For a historical overview of restriction semigroups and a detailed introduction to
their fundamental properties the reader is referred to [5].
So far, a number of important results of the rich structure theory of inverse semigroups
have been recast in the broader setting of restriction semigroups. The current paper is a
contribution to this body of work.
In the theory of inverse semigroups, semidirect products of semilattices by groups play
an important role. One of the reasons for this is that every inverse semigroup divides
such a semidirect product (see [8] for the main results in this area and for references). In
fact, something stronger is true, namely that every inverse semigroup can be embedded
into an (idempotent separating) homomorphic image of such a semidirect product. One
way to see this result is through extensions of partial isomorphisms of semilattices: an
inverse semigroup determines partial isomorphisms of its semilattice, and constructing the
corresponding semidirect product in fact includes embedding the semilattice into a bigger
one, and finding a group acting on the bigger semilattice such that the original partial
isomorphisms are restrictions of the automorphisms determined by the group action.
The situation for restriction semigroups is similar, but also different in some crucial re-
spects. In this case groups are replaced by monoids, which gives rise to several problems.
Date: January 18, 2016.
Research partially supported by the Hungarian National Foundation for Scientific Research grants
no. K083219, K104251, PD115705, and by EPSRC grant no. EP/IO32312/1.
Mathematical Subject Classification (2010): 20M10, 20M05.
Key words: Restriction semigroup, weakly E-ample semigroup, (almost) factorisability, semidirect product.
1
2
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
The first result analogous to that formulated above for inverse semigroups was obtained
by the third author in [11]: she has shown that every restriction semigroup can be em-
bedded into a (projection separating) homomorphic image of a so-called W -product of a
semilattice by a monoid, that is, into an almost left factorisable restriction semigroup (cf.
[4]). An alternative proof, based on the idea of extending partial isomorphisms to injective
endomorphisms, has been presented by Kudryavtseva in [7].
In general, the monoid acts in a W -product by injective endomorphisms on the semi-
lattice, and this was the case in the results of [11] and [7] mentioned.
In particular, if
the monoid acts by automorphisms on the semilattice then the W -product becomes a
semidirect product.
In this paper we obtain the strongest possible result in this direction by showing that
the monoid and the semilattice can be chosen such that the monoid acts on the semilattice
by automorphisms, thus demonstrating that any restriction semigroup can be embedded
into a (projection separating) homomorphic image of a semidirect product of a semilattice
by a monoid where the monoid acts on the semilattice by automorphisms. In other words,
we prove that any restriction semigroup can be embedded into an almost factorisable
restriction semigroup, or, equivalently, into a factorisable restriction monoid (cf. [4], [11]).
Furthermore, the construction applied in the proof of this result allows us to show that the
proper cover embeddable in a W -product, which is provided for a restriction semigroup in
[10] and [7], is embeddable also in a semidirect product of a semilattice by a group.
Note that restriction semigroups in general, proper restriction semigroups, almost fac-
torisable restriction semigroups and factorisable restriction monoids are defined in a left-
right dual manner, and semidirect products of semilattices by monoids where monoids act
on semilattices by automorphisms are left-right symmetric. However, this is not the case
with almost left factorisable restriction semigroups or W -products. Therefore the embed-
ding results of the present paper reflect the left-right symmetry of restriction and proper
restriction semigroups, while this was not the case with the foregoing results in [10] and
[7].
In the last section, we strengthen some results in [7] and [6] by proving that each
ultra F -restriction, or, equivalently, each perfect restriction monoid S whose greatest re-
duced factor is a free monoid is embeddable in a semidirect product of a semilattice by a
monoid in such a way that the monoid acts on the semilattice by automorphisms, and all
congruences of S extend to the semidirect product.
2. Preliminaries
In general, mappings and partial mappings are considered as right operands, and so their
products are calculated from the left to the right. This applies, amongst others, to the
group of automorphisms Aut Y and to the Munn semigroup TY of a semilattice Y , and to
the symmetric inverse monoid I(X) on a set X. By writing Autop Y , T op
Y , etc., we indicate
that the (partial) mappings in them are considered left operands, and their products are
calculated from the right to the left.
2.1. Restriction semigroups. [5] A left restriction semigroup is defined to be an alge-
bra of type (2, 1), more precisely, an algebra S = (S; ·, +) where (S; ·) is a semigroup and
EMBEDDING OF RESTRICTION SEMIGROUPS
3
+ is a unary operation such that the following identities are satisfied:
(1)
x+x = x,
x+y+ = y+x+,
(x+y)+ = x+y+,
xy+ = (xy)+x.
A right restriction semigroup is defined dually, that is, it is an algebra S = (S; ·, ∗) satisfying
the duals of the identities (1). Finally, if S = (S; ·, +, ∗) is an algebra of type (2, 1, 1) where
S = (S; ·, +) is a left restriction semigroup, S = (S; ·, ∗) is a right restriction semigroup
and the identities
(2)
(x+)∗ = x+,
(x∗)+ = x∗
hold then it is called a restriction semigroup. Notice that the defining properties of a
restriction semigroup are left-right dual. Therefore in the sequel dual definitions and
statements will not be explicitly formulated.
Among restriction semigroups, the notions of a subalgebra, homomorphism, congruence
and factor algebra are understood in type (2, 1, 1), which is emphasised by using the ex-
pressions (2, 1, 1)-subsemigroup, (2, 1, 1)-morphism, (2, 1, 1)-congruence and (2, 1, 1)-factor
semigroup, respectively. A restriction semigroup with identity element is also called a
restriction monoid.
In particular, each monoid M becomes a restriction monoid by defining a+ = a∗ = 1 for
any a ∈ M. It is easy to see that these restriction semigroups are just those with both
unary operations being constant. Such a restriction semigroup (monoid) is called reduced.
Notice that the submonoids, congruences, etc. of monoids and the (2, 1, 1)-subsemigroups,
(2, 1, 1)-congruences, etc. of the reduced restriction semigroups (monoids) obtained from
them coincide. Therefore we often consider reduced restriction semigroups (monoids) just
as monoids, and vice versa. Similarly, each semilattice Y becomes a restriction semigroup
by defining a+ = a∗ = a for every a ∈ Y . These restriction semigroups are just those
where both unary operations equal the identity map. Clearly, the (2, 1, 1)-subsemigroups,
(2, 1, 1)-congruences, etc. of such a restriction semigroup Y are just the subsemilattices,
congruences, etc. of the semilattice Y . Therefore such a restriction semigroup is simply
considered and called a semilattice.
Let S be any restriction semigroup. By (2), we have {x+ : x ∈ S} = {x∗ : x ∈ S}.
This set is called the set of projections of S, and is denoted by P (S). By (1) and its dual,
P (S) can be seen to be a (2, 1, 1)-subsemigroup in S which is a semilattice. Notice that
a restriction semigroup S is reduced if and only if P (S) is a singleton, and, if this is the
case, then the unique element of P (S) is the identity element of S.
Given a restriction semigroup S, we define a relation ≤ on S such that, for every a, b ∈ S,
a ≤ b
if and only if a = a+b.
It is easy to see that a ≤ b if and only if a = eb for some e ∈ P (S). Observe that the
dual of this relation is the same since a = a+b implies a = b(a+b)∗ = ba∗, and the dual
implication is also valid. The relation ≤ is a compatible partial order on S, and it extends
the natural partial order of the semilattice P (S). It is called the natural partial order on
S.
4
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
We also consider a relation on S, denoted by σS, or simply σ: for any a, b ∈ S, let
a σ b
if and only if
ea = eb
for some
e ∈ P (S).
Again notice that if there exists e ∈ P (S) with ea = eb then there exists also f ∈ P (S) with
af = bf , and conversely. Therefore the relation defined dually to σ coincides with σ. The
relation σ is the least congruence on S = (S; ·) where P (S) is in a congruence class, which
we denote by P (S)σ. Consequently, it is the least (2, 1, 1)-congruence ρ on S = (S; ·, +, ∗)
such that the (2, 1, 1)-factor semigroup S/ρ is reduced. Therefore we call σ the least reduced
(2, 1, 1)-congruence on S. Obviously, P (S)σ is the identity element of S/σ. The reduced
restriction monoid S/σ is often considered just as a monoid S/σ = (S/σ; ·, P (S)σ).
Let S, T be restriction semigroups. We say that S is proper if, for every a, b ∈ S, the
relations a+ = b+ and a σ b together imply a = b and the dual property also holds. It is
easy to see that each (2, 1, 1)-subsemigroup of a proper restriction semigroup is proper. A
(2, 1, 1)-morphism ϕ : T → S is called projection separating if eϕ = f ϕ implies e = f for
every e, f ∈ P (T ). We say that T is a proper cover of S if T is proper and there exists a
projection separating (2, 1, 1)-morphism from T onto S.
2.2. Semidirect products and factorisability. [4], [11] Let T be a monoid and Y =
(Y ; ∧) a semilattice. We say that T acts on Y (on the left) by automorphisms if a monoid
homomorphism α : T → Autop Y, t 7→ αt is given. The element αta (t ∈ T, a ∈ Y ) is
usually denoted by t · a. The fact that T acts on Y by automorphisms is equivalent to
requiring that the mapping αt : Y → Y, a 7→ t · a is bijective for every t ∈ T , and the
following equalities are valid for every a, b ∈ Y and t, u ∈ T :
(3)
t · (a ∧ b) = t · a ∧ t · b,
u · (t · a) = (ut) · a,
1 · a = a.
The semidirect product Y ⋊ T is the algebra of type (2, 1, 1) defined on the set Y × T =
{(t · a, t) : a ∈ Y, t ∈ T } with the following operations:
(t · a, t)(u · b, u) = (t · (a ∧ u · b), tu),
(t · a, t)+ = (t · a, 1) and (t · a, t)∗ = (a, 1).
Defining the multiplication this way may seem awkward, however, note that since T acts
by automorphisms, the multiplication and the + operation are in fact the same as in the
'usual' semidirect product. The reason for this unusual formulation is in the definition of
the operation ∗: since T is not necessarily a group, one cannot use t−1. It is routine to check
that Y ⋊T is a proper restriction semigroup where P (Y ⋊T ) = {(a, 1) : a ∈ Y } is isomorphic
to Y , the relation σ is just the kernel of the second projection Y ⋊ T → T, (t · a, t) 7→ t,
and so (Y ⋊ T )/σ is isomorphic to T . Moreover, the natural partial order is the following
relation: (t · a, t) ≤ (u · b, u) if and only if a ≤ b and t = u. The semidirect product Y ⋊ T
is a (proper restriction) monoid if and only if Y has an identity element. In this case the
fact that T acts by automorphisms implies that t · 1 = 1 for any t ∈ T where 1 is the
identity element of Y . To avoid confusion, we call the attention to the fact that, from now
on, when speaking about a semidirect product of a semilattice by a monoid, it is always
meant to be a restriction semigroup just defined. In particular, the action of the monoid
on the semilattice is taken to be an action by automorphisms.
EMBEDDING OF RESTRICTION SEMIGROUPS
5
t
Note that the semidirect product Y ⋊ T is isomorphic to the reverse semidirect product
T ⋉ Y where the right action of T on Y is given by the monoid homomorphism T →
Aut Y, t 7→ α−1
. By definition (see [4]), the reverse semidirect product T ⋉ Y is, in fact, a
W -product W (T, Y ). Therefore the same is the case with the semidirect product Y ⋊ T .
In this paper, an action of a monoid T on a semilattice Y with automorphisms will
frequently come as a restriction of an action of a group G to its submonoid T . In this case,
we will prefer representing the elements of the semidirect product Y ⋊ T , as it is usual
within the inverse semigroup Y ⋊ G, in the form (a, t) with a ∈ Y, t ∈ T . In this form,
the operations are:
(a, t)(b, u) = (a ∧ t · b, tu),
(a, t)+ = (a, 1) and (a, t)∗ = (t−1 · a, 1),
and the natural partial order is (a, t) ≤ (b, u) if and only if a ≤ b and t = u.
Note that, instead of assuming that a monoid T is embeddable in a group and it acts
on a semilattice Y by automorphisms, we can require without loss of generality that T
is a submonoid in a group G which it generates, G acts on Y , and the action of T is
the restriction of the action of G. For, recall first that if T is embeddable in a group
then the inclusion map κ : T → GT where GT = hT Ξi is given by the defining relations
Ξ = {tuv−1 : t, u, v ∈ T and tu = v in T } is an embedding. Second, if the action of
T on Y is defined by the homomorphism α : T → Autop Y then αtαuα−1
is the identity
v
automorphism for every t, u, v ∈ T with tu = v, and so there exists a homomorphism
β : GT → Autop Y such that α = κβ. Thus β defines an action of GT on Y , and its
restriction to T is just α.
Factorisable restriction monoids and almost factorisable restriction semigroups are in-
troduced in [11] (see also [4]) analogously to factorisable inverse monoids and almost fac-
torisable inverse semigroups, respectively. A restriction monoid F is called factorisable if
F = P (F )U(F ) where U(F ) = {u ∈ F : u+ = u∗ = 1} is the reduced (2, 1, 1)-subsemigroup
of F analogous to the group of units in an inverse monoid. Note that F = P (F )U(F ) if
and only if F = U(F )P (F ), therefore factorisability is a left-right symmetric property. The
notion of an almost factorisable restriction semigroup is defined by means of permissible
subsets but, instead of the definition, it suffices for the purposes of this paper to recall
how they relate to factorisable restriction monoids ([11, Proposition 3.10 and Theorem
3.11]). If F is a factorisable restriction monoid then both F and the (2, 1, 1)-subsemigroup
F \ U(F ) of F are almost factorisable, and conversely, each almost factorisable restriction
semigroup is, up to (2, 1, 1)-ismomorphism, of the latter form. In particular, this implies
that a restriction semigroup is (2, 1, 1)-embeddable in a factorisable restriction monoid if
and only if it is (2, 1, 1)-embeddable in an almost factorisable restriction semigroup.
In the proof of the main result of the paper, we need the following description of these
restriction semigroups by means of semidirect products.
Proposition 2.1. [11, Theorem 3.12] A restriction monoid (semigroup) is factorisable
(almost factorisable) if and only if it is a (2, 1, 1)-morphic image of a semidirect product
of a semilattice with identity (a semilattice) by a monoid.
6
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
Given an almost factorisable restriction semigroup F − in the above form, now we
establish for later use, how to construct a factorisable restriction monoid F such that
F − = F \ U(F ).
Lemma 2.2. Consider an almost factorisable restriction semigroup F − = (Y ⋊T )/ρ, where
T is a monoid acting on the semilattice Y by automorphisms, and ρ is a (2, 1, 1)-congruence
on the restriction semigroup Y ⋊ T . Let Y e be the semilattice obtained from Y by adjoining
an identity element e /∈ Y even if Y has an identity element. Extend the action of T on
Y to Y e by putting t · e = e for every t ∈ T , and consider the relation ρe = ρ ∪ ı on the
semidirect product Y e ⋊ T where ı is the equality relation on U = (Y e ⋊ T ) \ (Y ⋊ T ). Then
ρe is a (2, 1, 1)-congruence on the restriction monoid Y e ⋊ T extending ρ, F = (Y e ⋊ T )/ρe
is a factorisable restriction monoid with U(F ) = {uρe : u ∈ U} = {{(e, t)} : t ∈ T }, and
the mapping F − → F, (a, t)ρ 7→ (a, t)ρe ((a, t) ∈ Y ⋊ T ) is an injective (2, 1, 1)-morphism
with range F \ U(F ).
Proof. Obviously, T acts on Y e by automorphisms, the semidirect product Y e ⋊ T is a
restriction monoid with U(Y e ⋊ T ) = U, and Y ⋊ T is a (2, 1, 1)-subsemigroup and ideal of
Y e ⋊ T . Moreover, it is also clear that the relation ρe is an equivalence whose restriction
to Y ⋊ T is ρ, and that it is compatible with both unary operations. It remains to check
that ρe is compatible with the multiplication. We have to verify that if (a, t) ρ (b, u) in
Y ⋊ T and v ∈ T then both (a, t)(e, v) ρ (b, u)(e, v) and (e, v)(a, t) ρ (e, v)(b, u) hold, that
is, both (a, tv) ρ (b, uv) and (v · a, vt) ρ (v · b, vu) are valid. We verify the first relation, the
second being similar. Since ρ is a (2, 1, 1)-congruence on Y ⋊ T , we have (a ∧ t · c, tv) =
(a, t)(c, v) ρ (b, u)(c, v) = (b∧u · c, uv) for every c ∈ Y . Choosing c to be the unique element
of Y such that t · c = a, we obtain that (a, tv)ρ = (b ∧ u · c, uv)ρ ≤ (b, uv)ρ in F −. Changing
the roles of (a, t) and (b, u) the equality follows, proving the first relation.
(cid:3)
2.3. Congruences on restriction semigroups generated by relations. [11] Given
a set X of variables, by a (restriction) term in X we mean a formal expression built up from
the elements of X by means of the operational symbols -- the binary operational symbol
· and the unary operational symbols + and ∗ -- in finitely many steps. Since the binary
operation · is always interpreted in a semigroup we delete the unnecessary parentheses from
the terms. If S is a fixed restriction semigroup then we introduce a nullary operational
symbol for every element s in S, and, for simplicity, denote it also by s. By a polynomial
of S we mean an expression obtained in a way similar to terms, but from variables and
these nullary operational symbols. A polynomial can also be interpreted in the way that
such nullary operational symbols -- briefly elements of S -- are substituted for certain
variables in a term. So a unary polynomial of S, that is, a polynomial in one variable
x, is of the form t(x, s1, . . . , sk−1) for some term t in k variables and for some elements
s1, . . . , sk−1 ∈ S. To simplify our notation, we denote by S⋆ the set of finite sequences
of elements of S, and we use Greek letters to denote elements of S⋆. The length of the
sequences will be determined by the context where they appear. For example, if t is a
term in k variables and we are talking about t(x, α) for some α ∈ S⋆ then the length of α
is k − 1.
EMBEDDING OF RESTRICTION SEMIGROUPS
7
Let S be a restriction semigroup and let τ ⊆ S × S be a symmetric relation. We denote
by τ # the (2, 1, 1)-congruence on S generated by τ . It is a basic fact of universal algebra
that if s, t ∈ S then s τ # t if and only if there exist a sequence p1(x), p2(x), . . . , pk(x) of
unary polynomials of S and elements c1, d1, . . . , ck, dk ∈ S such that (ci, di) ∈ τ for all
i (1 ≤ i ≤ k) and
s = p1(c1), p1(d1) = p2(c2), . . . , pk(dk) = t.
In general, unary polynomials over a restriction semigroup can be quite complicated. How-
ever, it is shown in [11] that the unary polynomials appearing in the sequence can be chosen
to be simpler, and this will be key to our later arguments.
We define two sequences of terms in variables x, y, z, y0, z0, . . . in the following way: let
t(0)
+ (x, y0, z0) = (y0xz0)+,
t(0)
∗ (x, y0, z0) = (y0xz0)∗,
and, for every i ∈ N, let
t(i)
+ (x, y0, z0, . . . , yi−2, zi−1, yi) = (cid:0)yit(i−1)
∗ (x, y0, z0, . . . , zi−2, yi−1, zi) = (cid:0)t(i−1)
+
(x, y0, z0, . . . , yi−2, zi−1)(cid:1)+,
(x, y0, z0, . . . , zi−2, yi−1)zi(cid:1)∗.
∗
t(i)
For convenience we note that
,
t(1)
t(1)
+ (x, y0, z0, y1) = (cid:0)y1(y0xz0)∗(cid:1)+
∗ (x, y0, z0, z1) = (cid:0)(y0xz0)+z1(cid:1)∗,
t(2)
t(2)
+ (x, y0, z0, z1, y2) = (cid:0)y2(cid:0)(y0xz0)+z1(cid:1)∗(cid:1)+
∗ (x, y0, z0, y1, z2) = (cid:0)(cid:0)y1(y0xz0)∗(cid:1)+z2(cid:1)∗.
∗ , t(i+1)
: i ∈ N0 is even}, T(∗) = {t(i)
+
,
: i ∈ N0 is even},
We define the following sets of terms:
T(+) = {t(i)
+ , t(i+1)
∗
T = {t : t(x, y, z) = yxz or t(x, y, z, y0, z0, . . .) = yuz where u ∈ T(+) ∪ T(∗)}.
Notice that T(+) (T(∗)) contains terms with + (∗) as innermost unary operation. The set
T(+) (T(∗)) differs from the set T+ (T∗) used in [11] since the latter consists of the terms
with + (∗) as outermost unary operation. The reason for this modification (and slightly
awkward notation) is that, in contrast to the proof of the main result of [11], we are going
to distinguish cases depending on the innermost unary operation.
It is shown in [11] that the terms in T are enough to determine a (2, 1, 1)-congruence
generated by a symmetric relation.
Proposition 2.3. [11, Proposition 4.5] If τ is a symmetric relation on a restriction semi-
group S then, for any s, t ∈ S, we have s τ # t if and only if s = t or there exist k ∈ N,
t1, . . . , tk ∈ T, α1, . . . , αk ∈ S ∗ and (c1, d1), . . . , (ck, dk) ∈ τ such that
s = t1(c1, α1), t1(d1, α1) = t2(c2, α2), . . . , tk(dk, αk) = t.
Let S, S ′ be restriction semigroups such that S is a (2, 1, 1)-subsemigroup of S ′, and
let ρ be (2, 1, 1)-congruence on S. A (2, 1, 1)-congruence ρ′ on S ′ is said to extend ρ if
8
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
the restriction of ρ′ to S equals ρ. We say that ρ extends to S ′ if there exists a (2, 1, 1)-
congruence ρ′ on S ′ extending ρ. Notice that ρ extends to S ′ if and only if the (2, 1, 1)-
congruence on S ′ generated by ρ extends ρ.
2.4. Free restriction monoids. [3] A transparent model of the free restriction monoid
on a set is given in [3] as a full subsemigroup in the free inverse monoid on the same set
(cf. [8]).
Let Ω be a non-empty set, and denote by Ω∗ and FG(Ω), respectively, the usual models
of the free monoid and the free group on Ω where Ω∗ consists of all words (i.e., of finite
sequences) over Ω and FG(Ω) consists of all reduced words over Ω ∪ Ω−1. The empty word
is denoted by 1. Obviously, Ω∗ is a subsemigroup of FG(Ω), and 1 is the identity element
in both.
Let C be the Cayley graph of FG(Ω) which is well known to be a tree. Therefore
each finite subset A ⊆ FG(Ω) determines a maximum subgraph of C having A as its set
of vertices. From now on, we identify A with this subgraph. Let X be the set of all
finite connected subgraphs of C (i.e., of all finite subtrees), partially ordered by reverse
inclusion. Then X is a semilattice without an identity. Clearly, Y = {A ∈ X : 1 ∈ A} is a
principal order ideal of X and a subsemilattice with identity element 1 = {1}. Moreover,
(FG(Ω), X , Y) is a McAlister triple where FG(Ω) acts on the partially ordered set X by
multiplication: if g ∈ FG(Ω) and A ∈ X then g · A = {ga : a ∈ A}. It is well known that
the P -semigroup P (FG(Ω), X , Y) is a free inverse monoid on Ω.
It is established in [3] that
FR(Ω) = {(A, a) ∈ P (FG(Ω), X , Y) : a ∈ Ω∗} = {(A, a) ∈ Y × Ω∗ : A ⊇ a · 1}
is a (2, 1, 1)-subsemigroup in P (FG(Ω), X , Y), it is a free restriction monoid on Ω and
FRS(Ω) = FR(Ω) \ {(1, 1)} is a free restriction semigroup on Ω.
Since the free inverse monoid P (FG(Ω), X , Y) is F -inverse, the partially ordered set
X , where the partial order will subsequently be denoted ≤, is a semilattice. Thus Y is
a principal ideal in it with identity element 1, and the action of FG(Ω) on the partially
ordered set X is, in fact, an action on the semilattice X . Since C is a tree, for every finite
subgraph A ⊆ FG(Ω), there exists a minimum subtree A′ containing A. The semilattice
operation ∧ of X can be given by the rule A ∧ B = (A ∪ B)′ (A, B ∈ X ). Thus the
semidirect products X ⋊ FG(Ω) and X ⋊ Ω∗ are defined, and FR(Ω) can be given in this
context as
FR(Ω) = {(A, a) ∈ X ⋊ Ω∗ : A ∈ Y, a ∈ A} = {(A, a) ∈ Y × Ω∗ : A ≤ a · 1}.
Note that the free inverse monoid P (FG(Ω), X , Y) is usually considered as an inverse
subsemigroup of a semidirect product of another semilattice by FG(Ω), namely, of the
semilattice of all finite subgraphs of C with respect to the usual join. In the sequel, we
need the approach in the previous paragraph, as it was the case with the W -product applied
in [10] and [11].
EMBEDDING OF RESTRICTION SEMIGROUPS
9
3. Main result
Our aim is to show that every restriction monoid, and consequently, every restriction
semigroup is (2, 1, 1)-embeddable in a factorisable restriction monoid, or equivalently in
an almost factorisable restriction semigroup. We obtain a (2, 1, 1)-embedding by taking
a free restriction monoid FR(Ω), having a given restriction monoid S as its image via a
(2, 1, 1)-morphism, say, ϕ, and by considering the (2, 1, 1)-embedding of FR(Ω) into the
semidirect product X ⋊Ω∗ described in the previous section where the monoid Ω∗ acts on the
semilattice X by automorphisms. The kernel of ϕ : FR(Ω) → S is a (2, 1, 1)-congruence on
FR(Ω), we consider the (2, 1, 1)-congruence on the semidirect product X ⋊ Ω∗ it generates.
If the restriction of the latter congruence to FR(Ω) equals the kernel, then we obtain
a (2, 1, 1)-embedding of S into a (2, 1, 1)-morphic image of the semidirect product, that
is, into an almost factorisable restriction semigroup. As it turns out, the fact that this
approach works depends only on the fact that the actions of the free monoid Ω∗ and of
the free group FG(Ω) on the semilattice X satisfy a simple condition. This allows us to
slightly generalise our approach.
From now on, let X be a semilattice and let Y be a principal ideal of X . Denote the
identity element of Y by 1. Let T be a monoid embeddable in a group, and suppose that
T acts on X (on the left) by automorphisms. We have seen formerly that in this case,
there exists a group G acting on X such that T is a submonoid of G generating G, and the
action of T is the restriction of the action of G. Let us choose and fix such a group G. In
the sequel we will use boldface letters to represent the elements of the semidirect products
X ⋊ G and X ⋊ T , the corresponding capital letters and overlined letters to denote their
first and second components, respectively -- for example, a = (A, a).
Define the following subset of the semidirect product X ⋊ T :
(4)
R = {(A, a) ∈ Y × T : A ≤ a · 1}.
Note that if (A, a) ∈ R and A = a·B ≤ a·1 where B ∈ X , then as T acts by automorphisms
we have B ≤ 1. It is now easy to check that R is a (2, 1, 1)-subsemigroup in X ⋊ T with
identity element (1, 1), and so it is a proper restriction monoid. Notice that, for any
(A, a) ∈ X ⋊ T there exists a maximum element of R less than or equal to (A, a) which we
denote by (A, a)↓. It is clear that (A, a)↓ = (1 ∧ A ∧ a · 1, a).
Observe that G · Y is the smallest subsemilattice in X containing Y which is invariant
under the action of G. Since, for any g, h ∈ G and A, B ∈ Y, we have g · A ∧ h · B =
g · (A ∧ g−1h · B), and Y is an ideal in X , we see that G · Y is, indeed, a subsemilattice of
X . Since we have R ⊆ Y × T , we can assume without loss of generality that X = G · Y.
A sequence w1, w2, . . . , wn of elements of G is called an alternating sequence in T if,
for every i (1 ≤ i ≤ n), we have wi = tǫi
for some ti ∈ T, ǫi ∈ {1, −1}, and, for every
i
i, j (1 ≤ i, j ≤ n), we have ǫi = ǫj if and only if 2 i − j. A factorisation g = w1w2 · · · wn
of an element g ∈ G is called a nice factorisation in T with respect to 1 if w1, w2, . . . , wn is
an alternating sequence in T such that, for every i (1 ≤ i < n), we have that
wi · 1 ≥ 1 ∧ wi · · · wn · 1.
10
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
We say that the action of the monoid T on the semilattice X (necessarily by automor-
phisms) is nice over the group G with respect to the element 1 ∈ X if every element g ∈ G
has a nice factorisation in T with respect to 1. Note that, in particular, the action of a
group on a semilattice is always nice over the group itself with respect to any element of
the semilattice. We say that X ⋊ T is a nice semidirect product of a semilattice X by a
monoid T if T is embeddable in a group and acts on X by automorphisms, G is a group
having T as a submonoid and acting on X such that the action of T is the restriction of
the action of G, and is nice over G with respect to an element 1 ∈ X such that X = G · Y
for the principal ideal Y of X with identity 1.
Let G be a group acting on a semilattice X , and let ε be a congruence on X . We say
that ε is G-invariant if x ε x′ implies g · x ε g · x′ for every x, x′ ∈ X and g ∈ G.
In
this case, G acts by automorphisms on X /ε by the rule g · (xε) = (g · x)ε. The following
observation is straightforward.
Lemma 3.1. If an action of a monoid T on a semilattice X is nice over a group G with
respect to an element 1 ∈ X and ε is a G-invariant congruence on X then the restriction
of the action of G on X /ε to T is also nice over G with respect to 1ε ∈ X /ε. Consequently,
if X ⋊ T is a nice semidirect product then the same holds for (X /ε) ⋊ T .
Example 3.2. The action of Ω∗ on the semilattice X defined in Subsection 2.4 is nice over
FG(Ω) with respect to 1. It is easy to see that if g ∈ FG(Ω), then the unique reduced
alternating factorisation g = w1w2 · · · wn where w1, w2, . . . , wn ∈ Ω∗ ∪ (Ω∗)−1 is a nice
factorisation of g in Ω∗ with respect to 1. Moreover, we have X = FG(Ω) · Y where Y is a
principal ideal with identity 1, and so X ⋊ Ω∗ is a nice semidirect product.
Example 3.3. Let G and T be the free Abelian group and free commutative semigroup,
respectively, on the set Ω. Then for every g ∈ G, there exist unique elements u, t ∈
T such that g = u−1t. We say that a subset A ⊆ G is min-closed if, for every g =
ωa1
s , h = ωb1
1 · · · ωas
s ∈ A, where ω1, . . . , ωs are pairwise distinct elements of Ω, we
have that ωmin(a1,b1)
∈ A. Let X = {A ⊆ G : A 6= ∅, A is min-closed} and,
for all A, B ∈ X , let A ∧ B be the smallest min-closed subset of G containing A ∪ B. Then
X is a semilattice and G acts on X by multiplication. Furthermore, Y = {A ∈ X : 1 ∈ A}
is a principal ideal of X with identity {1} = 1. Note that, for any g ∈ G, the unique
factorisation g = u−1t is nice in T with respect to 1, because if g = ωa1
then
u−1 = ωmin(a1,0)
∈ {1} ∧ {g}, showing that u−1 · 1 ≥ 1 ∧ g · 1.
· · · ωmin(as,0)
· · · ωmin(as,bs)
s
1 · · · ωas
s
1
s
1 · · · ωbs
1
From now on, additionally to our former assumptions on X , Y, T and G, we suppose
that the action of T on X is nice over the group G with respect to 1.
Lemma 3.4. Let t(i) ∈ T(+) ∪ T(∗) and let α ∈ (X ⋊ T )⋆. Then there exist U, V ∈ X ,
g ∈ G depending on α such that
for every c ∈ R where
t(i)(c, α) = (U ∧ g · C ∧ gc · V, 1)
c = ( C, c) = (cid:26) (C, c)
(c−1 · C, c−1)
if t(i) ∈ T(+),
if t(i) ∈ T(∗).
EMBEDDING OF RESTRICTION SEMIGROUPS
11
Proof. We proceed by induction on i. If i = 0 then t(i) = (y0xz0)+ or t(i) = (y0xz0)∗, and
α = (cid:0)(A, a), (B, b)(cid:1) for some (A, a), (B, b) ∈ X ⋊ T . If t(i) ∈ T(+) then
t(i)(c, α) = (cid:0)(A, a)(C, c)(B, b)(cid:1)+ = (A ∧ a · C ∧ a c · B, 1),
so we can set U = A, V = B and g = a . Similarly, if t(i) ∈ T(∗) then
t(i)(c, α) = (cid:0)(A, a)(C, c)(B, b)(cid:1)∗ = (b
−1
−1
.
so we can set U = b
· B, V = a−1 · A and g = b
−1
c−1a−1 · A ∧ b
−1
c−1 · C ∧ b
−1
· B, 1),
Let i ≥ 1 and t(i) ∈ T(+) ∪ T(∗), and suppose that we have proven the statement for
and only if t(i−1) ∈ T(+). For, T(+) contains terms with + as innermost unary operation.
By the induction hypothesis there exist U ′, V ′ ∈ X and g′ ∈ G such that t(i−1)(c, α′) =
i − 1. Then t(i) = (cid:0)yit(i−1)(cid:1)+ or t(i) = (cid:0)t(i−1)zi(cid:1)∗ for an appropriate t(i−1) ∈ T(+) ∪ T(∗)
and α = (cid:0)α′, (A, a)(cid:1) where α′ ∈ (X ⋊ T )⋆ and (A, a) ∈ X ⋊ T . Note that t(i) ∈ T(+) if
(U ′ ∧ g′ · C ∧ g′c · V ′, 1) for all c ∈ R. If t(i) = (cid:0)yit(i−1)(cid:1)+
so we can set U = A ∧ a · U ′, V = V ′ and g = ag′. Similarly, if t(i) = (cid:0)t(i−1)zi(cid:1)∗
t(i)(c, α) = (cid:0)(U ′ ∧ g′ · C ∧ g′c · V ′, 1)(A, a)(cid:1)∗
t(i)(c, α) = (cid:0)(A, a)(U ′ ∧ g′ · C ∧ g′c · V ′, 1)(cid:1)+ = (A ∧ a · U ′ ∧ ag′ · C ∧ ag′c · V ′, 1),
= (a−1 · U ′ ∧ a−1g′ · C ∧ a−1g′c · V ′ ∧ a−1 · A, 1),
so we can set U = a−1 · U ′ ∧ a−1 · A, V = V ′ and g = a−1g′.
then
then
(cid:3)
The previous lemma showed that for all terms t ∈ T(+) ∪ T(∗), the elements t(c, α)
(α ∈ (X ⋊ T )⋆) are of special form. Our aim is to replace in certain circumstances the
sequence of constants α by β ∈ R⋆ and the term t by a term t′ such that t(c, α)↓ = t′(c, β)
for all c ∈ R. The following lemma constitutes the first step in this direction.
Lemma 3.5. Let U, V ∈ X and g ∈ G. Then there exist t ∈ T(+) and β ∈ R⋆ such that
(1 ∧ U ∧ g · C ∧ gc · V, 1) = t(c, β)
for all c ∈ R. Dually, there exists t ∈ T(∗) and β ∈ R⋆ such that
(1 ∧ U ∧ gc−1 · C ∧ gc−1 · V, 1) = t(c, β)
for all c ∈ R.
Proof. Let g = w1 · · · wn be a nice factorisation. We prove the lemma by induction on n.
If n = 1 then there are four cases to consider (recall the definition of the element ( C, c)
from Lemma 3.4):
Conditions
((1 ∧ U ∧ w1 · C ∧ w1c · V, 1) =
w1 ∈ T, t ∈ T+
(cid:0)(1 ∧ U ∧ w1 · 1, w1)(C, c)(1 ∧ V, 1)(cid:1)+
1 )(cid:1)∗
w1 ∈ T −1, t ∈ T+ (cid:0)((C, c)(1 ∧ V, 1))+ (1 ∧ w−1
(cid:0)(1 ∧ U ∧ w1 · 1, w1)(cid:0)(1 ∧ V, 1)(C, c)(cid:1)∗(cid:1)+
1 )(cid:1)∗
(cid:0)(1 ∧ V, 1)(C, c)(1 ∧ w−1
w1 ∈ T, t ∈ T∗
w1 ∈ T −1, t ∈ T∗
· U ∧ w−1
1
· U ∧ w−1
1
· 1, w−1
· 1, w−1
1
1
12
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
The equalities are easy to check and left to the reader. Note that throughout variations of
the inequalities w1 · 1 ≥ w1 · C and c · 1 ≥ C are used.
Let us suppose that for some n ≥ 2 we have proven the lemma for every g′ ∈ G which
admits a nice factorisation having less than n factors, and let us suppose that g = w1 · · · wn
is a nice factorisation. Then w−1
1 g = w2 · · · wn is also a nice factorisation, so applying the
induction hypothesis for U ′ = w−1
1 g, there exists t′ ∈ T(+) and
β ′ ∈ R⋆ such that
· U, V ′ = V and g′ = w−1
1
(1 ∧ w−1
1
· U ∧ w−1
1 g · C ∧ w−1
1 gc · V, 1) = t′(c, β ′)
for all c ∈ R. Now there are two cases to consider. If w1 ∈ T then w1·1 ≥ 1∧g·1 ≥ 1∧g·C,
and so
1
· U ∧ w−1
= t(c, β)
1 g · C ∧ w−1
1 gc · V, 1)(cid:1)+
(1 ∧ U ∧ g · C ∧ gc · V, 1) = (cid:0)(1 ∧ w1 · 1, w1)(1 ∧ w−1
= (cid:0)(1 ∧ w1 · 1, w1)t′(c, β ′)(cid:1)+
for t = (yit′)+ and β = (cid:0)β ′, (1 ∧ w1 · 1, w1)(cid:1). If w1 ∈ T −1 then by the same fact we have
1 )(cid:1)∗
(1 ∧ U ∧ g · C ∧ gc · V, 1) = (cid:0)(1 ∧ w−1
for t = (t′zi)∗ and β = (cid:0)β ′, (1 ∧ w−1
1 )(cid:1). To finish this part of the proof, one
has to note that the fact that the factorisation g = w1 · · · wn is alternating implies that
the terms defined above are all contained in T(+). This proves the first statement of the
lemma. The induction step for the second statement is proven similarly, utilising the fact
that w1 · 1 ≥ 1 ∧ gc−1 · C.
(cid:3)
1 g · C ∧ w−1
1 )(cid:1)∗
· 1, w−1
= (cid:0)t′(c, β ′)(1 ∧ w−1
1 gc · V, 1)(1 ∧ w−1
1
= t(c, β)
· 1, w−1
· U ∧ w−1
1
· 1, w−1
1
1
Lemma 3.6. Let t ∈ T and let α ∈ (X ⋊ T )⋆. Then there exist t′ ∈ T and β ∈ R⋆ such
that
(5)
for all c ∈ R.
t(c, α)↓ = t′(c, β)
Proof. There are three cases to consider: the first one is if t = yxz. In this case let t′ = t,
and let β = (cid:0)(1 ∧ A ∧ a · 1, a), (1 ∧ B ∧ b · 1, b)(cid:1) if α = (cid:0)(A, a), (B, b)(cid:1). Note that since
c ∈ R, the inequalities ac · 1, a · 1 ≥ a · C imply that
t(c, α)↓ = (1 ∧ A ∧ a · C ∧ ac · B ∧ acb · 1, acb)
= (1 ∧ A ∧ a · 1, a)(C, c)(1 ∧ B ∧ b · 1, b) = t′(c, β).
The second case is where t = yrz for some r ∈ T(+). Then α = (cid:0)α′, (A, a), (B, b)(cid:1) for
some (A, a), (B, b) ∈ X ⋊ T and α′ ∈ (X ⋊ T )⋆. By Lemma 3.4 there exist U, V ∈ X and
g ∈ G such that r(c, α′) = (U ∧ g · C ∧ gc · V, 1) for every c ∈ R. In this case
t(c, α) = (A, a)(U ∧ g · C ∧ gc · V, 1)(B, b) = (A ∧ a · U ∧ ag · C ∧ agc · V ∧ a · B, ab).
By Lemma 3.5, there exist t ∈ T(+) and β ′ ∈ R⋆ such that for all c ∈ R we have
(1 ∧ A ∧ a · U ∧ ag · C ∧ agc · V ∧ a · B, 1) = t(c, β ′).
EMBEDDING OF RESTRICTION SEMIGROUPS
13
So altogether
t(c, α)↓ = (1 ∧ A ∧ a · U ∧ ag · C ∧ agc · V ∧ a · B ∧ ab · 1, ab)
= (1 ∧ A ∧ a · U ∧ ag · C ∧ agc · V ∧ a · B, 1)(1 ∧ ab · 1, ab)
= t(c, β ′)(1 ∧ ab · 1, ab),
so t′ = ytz and β = (cid:0)(1, 1), (1 ∧ ab · 1, ab)), β ′(cid:1) satisfy the requirements of the lemma.
The third case, where t ∈ T(∗), can be dealt with similarly to the previous case.
(cid:3)
Lemma 3.7. Let ρ be a (2, 1, 1)-congruence on R and let ρ# be the (2, 1, 1)-congruence on
X ⋊ T generated by ρ. Then the restriction of ρ# to R equals ρ, i.e., ρ# extends ρ. As a
consequence, R/ρ is (2, 1, 1)-embeddable in (X ⋊ T )/ρ#.
Proof. Let (s, t) ∈ ρ# ∩ (R × R). Then there exists a ρ-sequence
s = t1(c1, α1), t1(d1, α1) = t2(c2, α2), . . . , tn(dn, αn) = t
connecting s and t in X ⋊ T . By Lemma 3.6, for every k (1 ≤ k ≤ n), there exist t′
and βk ∈ R⋆ satisfying (5). Thus for every k (1 ≤ k ≤ n), we have
k ∈ T
tk(ck, αk)↓ = t′
k(ck, βk), t′
k(dk, βk) = t(dk, αk)↓,
showing that the sequence
s = t1(c1, α1)↓ = t′
n(dn, βn) = tn(dn, αn)↓ = t
1(c1, β1), t′
1(d1, β1) = t′
2(c2, β2), . . . , t′
connects s and t within R, proving that (s, t) ∈ ρ.
(cid:3)
Although this lemma easily implies our embedding result referred in the title to, before
formulating it, we show two lemmas in order to prepare our result on proper covers. Note
that in the first lemma the action of T need not be nice.
Lemma 3.8. Let ρ ⊆ σ be a (2, 1, 1)-congruence on the semidirect product X ⋊T . Then, for
every (A, a), (B, b) ∈ X ⋊ T , we have (A, a) ρ (B, b) if and only if a = b and (A, 1) ρ (B, 1).
As a consequence, ρ is generated by its restriction to the projections.
Proof. Let τ = {(cid:0)(A, a), (B, b)(cid:1) ∈ (X ⋊ T ) × (X ⋊ T ) : a = b and (A, 1) ρ (B, 1)}. It is
clear that ρ ⊆ τ . Conversely, let (cid:0)(A, a), (B, b)(cid:1) ∈ τ . Then
(A ∧ B, a) = (A, 1)(B, a) ρ (B, 1)(B, a) = (B, a),
which together with its dual completes the proof.
(cid:3)
Lemma 3.9. Let ρ be a (2, 1, 1)-congruence on R. Denote by ρP the restriction of ρ to P =
P (R) and by τ and τ # the (2, 1, 1)-congruences on R and on X ⋊ T , respectively, generated
by ρP . Then R/τ (2, 1, 1)-embeds into (X /ε)⋊T for an appropriate G-invariant congruence
ε on X , and (X /ε) ⋊ T is a nice semidirect product of (X /ε) by T . Moreover, the (2, 1, 1)-
congruence ρ/τ on R/τ is projection separating, and the corresponding congruence on the
image of R/τ in (X /ε) ⋊ T extends to a (2, 1, 1)-congruence on (X /ε) ⋊ T . Consequently,
R/ρ is (2, 1, 1)-embeddable in a (2, 1, 1)-morphic image of (X /ε) ⋊ T .
14
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
Proof. Notice that τ ⊆ σR and τ # ⊆ σX ⋊T . By definition, ρP ⊆ τ ⊆ ρ, and as the
restriction of ρ to P equals ρP the same is true for τ , thus the (2, 1, 1)-congruence ρ/τ on
R/τ is projection separating. Denote by ρ# the (2, 1, 1)-congruence on X ⋊ T generated
by ρ. Since T acts nicely over G on X , we see by Lemma 3.7 that ρ# and τ # extend ρ
and τ , respectively. Put ε = {(A, B) ∈ X × X : (A, 1) τ # (B, 1)}. Then ε is obviously
a congruence on X , and it is easily seen to be G-invariant. For, let A, B ∈ X with
A ε B and a ∈ T . Then we obtain by Lemma 3.8 that (A, a) τ # (B, a) which implies
(a−1 · A, 1) = (A, a)∗ τ # (B, a)∗ = (a−1 · B, 1). Hence a−1 · A ε a−1 · B follows. Moreover,
we also see that (a · A, 1) = (a · A, a)+ = (cid:0)(a · A, a)(A, 1)(cid:1)+ τ #(cid:0)(a · A, a)(B, 1)(cid:1)+ =
(a(A ∧ B), a)+ = (a(A ∧ B), 1), and so we deduce that a · A ε a · (A ∧ B). Similarly,
a · B ε a · (A ∧ B) also holds, implying the relation a · A ε a · B. Since G is generated by T ,
one can see that g · A ε g · B for every g ∈ G. Thus ε is, indeed, G-invariant, so the action
of G on X induces an action of G on X /ε, and defines semidirect products (X /ε) ⋊ G and
(X /ε) ⋊ T . Lemma 3.1 implies that the latter semidirect product is nice.
By Lemma 3.8 we also have that ι : (X ⋊ T )/τ # → (X /ε) ⋊ T, (A, a)τ # 7→ (Aε, a) is an
isomorphism. Let ρ be the congruence on (X /ε) ⋊ T corresponding via ι to the congruence
ρ#/τ # on (X ⋊T )/τ #. The rest follows by basic universal algebra: since ρ# and τ # extend
ρ and τ , respectively, we obtain that ρ#/τ # extends ρ/τ , thus (2, 1, 1)-embedding R/ρ,
which is isomorphic to (R/τ )/(ρ/τ ), into ((X /ε) ⋊ T )/ρ.
(cid:3)
Now we are ready to formulate the main result of the paper. It strengthens [11, Theorem
4.1] which establishes that every restriction semigroup is (2, 1, 1)-embeddable in an almost
left factorisable restriction semigroup, and [10, Theorem 4.11] which says that every re-
striction semigroup has a proper cover (2, 1, 1)-embeddable in a W -product of a semilattice
by a monoid.
Recall from Subsection 2.2 that a restriction semigroup (monoid) is almost factorisable
(factorisable) if and only if it is a (2, 1, 1)-morphic image of a semidirect product of a
semilattice (semilattice with identity) by a monoid.
Theorem 3.10.
(i) Every restriction semigroup is (2, 1, 1)-embeddable in an almost
factorisable restriction semigroup, or, equivalently, in a factorisable restriction
monoid.
(ii) Every restriction semigroup has a proper cover which is (2, 1, 1)-embeddable in a nice
semidirect product of a semilattice by a monoid, and consequently, in a semidirect
product of a semilattice by a group.
Proof. (i) Let S be a restriction semigroup. By Lemma 2.2, it suffices to embed the
restriction monoid S1 into a (2, 1, 1)-morphic image of a semidirect product of a semilattice
by a monoid. Clearly, S1 is isomorphic to FR(Ω)/ρ for some set Ω and (2, 1, 1)-congruence
ρ on FR(Ω). Here FR(Ω) is just the (2, 1, 1)-subsemigroup R in (4) of the semidirect
product X ⋊ Ω∗ given in Subsection 2.4. Example 3.2 shows that the action of Ω∗ on X
is nice over FG(Ω) with respect to 1, and so Lemma 3.7 implies that FR(Ω)/ρ and its
(2, 1, 1)-isomorphic copy S1 (2, 1, 1)-embed into a (2, 1, 1)-morphic image of X ⋊ Ω∗.
EMBEDDING OF RESTRICTION SEMIGROUPS
15
(ii) Applying Lemma 3.9 for R = FR(Ω) and the (2, 1, 1)-congruence ρ in the previous
paragraph, we obtain that FR(Ω)/τ is (2, 1, 1)-embeddable in a nice semidirect product
X ′ ⋊ Ω∗ where X ′ is a factor semilattice of X over a FG(Ω)-invariant congruence. This
implies that the semidirect product X ′ ⋊ FG(Ω) is also defined, and X ′ ⋊ Ω∗ is a (2, 1, 1)-
subsemigroup in X ′ ⋊ FG(Ω). The restriction monoid FR(Ω)/τ is proper since the semidi-
rect product X ′ ⋊ Ω∗ is. Moreover, Lemma 3.9 also establishes that ρ/τ is a projection
separating (2, 1, 1)-congruence on R/τ . This shows that FR(Ω)/τ is a proper cover of
(FR(Ω)/τ )/(ρ/τ ), and so of FR(Ω)/ρ and S1. This completes the proof if S = S1, that
is, if S is a monoid. In the opposite case, it is routine to check that we can ensure the
ρ-class of the identity element of FR(Ω) is a singleton, and the same property follows for τ .
This implies that the restriction semigroup obtained from FR(Ω)/τ by deleting its identity
element is a proper cover of S with the required properties.
(cid:3)
Note that, similarly to the embedding theorems in [11] and [7], the embedding given
here is not a monoid embedding.
4. Additional remarks
One can see from Lemmas 3.8 and 3.9 (see also the proof of Theorem 3.10(ii)) that if C
is a restriction monoid which is a (2, 1, 1)-factor monoid of a free restriction monoid over
a (2, 1, 1)-congruence contained in σ, then C is (2, 1, 1)-embeddable in a nice semidirect
product. Both Kudryavtseva [7] and Jones [6] have introduced a somewhat more general
class of restriction monoids, called ultra F -restriction monoids in [7] and perfect restriction
monoids in [6]. Moreover, if the greatest reduced factor of such a restriction monoid is a free
monoid then it is proved to be (2, 1, 1)-embeddable in a W -product in [7, Theorem 28]. In
this section we strengthen this result by considering an even more general class of restriction
monoids and showing that their members are (2, 1, 1)-embeddable in a nice semidirect
product in such a way that each of their (2, 1, 1)-congruences extends to the semidirect
product. Later on, we adopt the name 'perfect' from [6] for the class of restriction monoids
mentioned, and we use the notation of [7].
Let T be a monoid and Y a semilattice with identity element. Denote by OY the
inverse submonoid of the symmetric inverse monoid I(Y) consisting of all isomorphisms
between the ideals of Y. Notice that the Munn semigroup TY is an inverse submonoid in
OY. A left partial action of T on Y is defined to be a dual monoid prehomomorphism
α : T → Oop
Y , t 7→ αt, that is, α is required to have the properties that α1 is the identity
automorphism of Y, and αtαu is a restriction of αtu for every t, u ∈ T . For any t ∈ T
and A ∈ Y, we denote the element αtA ∈ Y by t ⋄ A. Since OY is an inverse monoid, the
mapping α′ : T → OY, t 7→ α−1
is a dual monoid prehomomorphism, and so α′ defines a
right partial action of T on Y. For any t ∈ T and A ∈ Y, the element Aα′
t ∈ Y will be
denoted by A ◦ t. Clearly, the partial actions ⋄ and ◦ are reverse to each other, that is, for
every t ∈ T and A ∈ Y, if t ⋄ A is defined then (t ⋄ A) ◦ t is also defined and (t ⋄ A) ◦ t = A,
and similarly, if A ◦ t is defined then t ⋄ (A ◦ t) is also defined and t ⋄ (A ◦ t) = A.
t
Consider the set
M(T, Y) = {(A, t) ∈ Y × T : A ◦ t is defined},
16
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
and define the following operations on it:
(A, t)(B, u) = (t ⋄ ((A ◦ t) ∧ B), tu),
(A, t)+ = (A, 1) and (A, t)∗ = (A ◦ t, 1).
Then M(T, Y) is a proper restriction monoid, and, conversely, each proper restriction
monoid S is isomorphic to M(S/σ, P (S)) where the left partial action of S/σ on P (S) is
induced by S in a natural way, see [1].
A congruence ρ on a semigroup S is called perfect if for every a, b ∈ S we have (aρ)(bρ) =
(ab)ρ, where the left hand side of the equality is the set product of the classes aρ and bρ.
A restriction monoid S is called perfect if σ is a perfect congruence and each σ-class has
a greatest element with respect to the natural partial order. It follows ([6, Theorem 1.1])
that a restriction monoid is perfect if and only if it is isomorphic to a restriction monoid
M(T, Y) where the left partial action of T on Y is a monoid homomorphism from T into
T op
Y .
Remark 4.1.
(i) Note that [7] uses the term 'left partial action' in a more general
sense. However, in a semilattice, the order ideals (resp. principal order ideals) and
the ideals (resp. principal ideals) coincide. Moreover, the order isomorphisms and
the isomorphisms between the ideals of a semilattice coincide. Therefore a left
partial action of a monoid T on a semilattice Y defined above is just what is called
in [7] a 'left partial action of T on Y such that axioms (A),(B),(C) hold'.
(ii) The restriction monoid denoted by W (T, Y) in [6] is the left-right dual of M(T, Y)
defined above, and so it is a much more general construction than what is usually
meant by the notation W (T, Y) (see [2], [4], [7], [10], [11]), in particular in this
paper, called a W -product of a semilattice by a monoid and denoted W (T, Y).
Suppose now that T is a monoid embeddable in a group, X is a semilattice, and T acts on
X by automorphisms. As above, let us choose and fix a group G such that T is a submonoid
in G generating G, G acts on X , and the action of T on X is just the restriction of the action
of G on X . Moreover, consider a principal ideal Y of X , and denote its greatest element by
1. It is routine to check that the restriction of the action of G on X to Y is a left partial
action α : G → T op
Y where the domain and range of αg (g ∈ G) are the principal ideals of
Y with greatest elements g−1 · 1 ∧ 1 and g · 1 ∧ 1, respectively. Therefore the restriction
monoids M(G, Y) and M(T, Y) are defined, and we have M(T, Y) 6 M(G, Y). Taking
into account that all elements of X appearing in M(G, Y) belong to the subsemilattice
G · Y, we can suppose without loss of generality that X = G · Y. In this case, (G, X , Y)
is a McAlister triple, and it is easy to see that M(G, Y) = P (G, X , Y), the P -semigroup
defined by it. This implies that M(G, Y) is an F -inverse monoid, and it is an inverse
subsemigroup in the semidirect poduct X ⋊ G. Furthermore, it also follows that M(T, Y)
is equal to the (2, 1, 1)-subsemigroup R of the semidirect product X ⋊ T defined in (4).
Thus Lemmas 3.7 and 3.9 imply the following statement.
Proposition 4.2. Suppose that a left partial action of a monoid T on a semilattice Y
with identity 1 can be extended to an action · of T on a semilattice X containing Y as a
principal ideal such that the action · is nice over a group G with respect to 1, and we have
X = G · Y. Then the following hold:
EMBEDDING OF RESTRICTION SEMIGROUPS
17
(i) M(T, Y) 6 X ⋊ T, M(G, Y) 6 X ⋊ G where X ⋊ T is a nice semidirect product and
M(G, Y) is an F -inverse monoid;
(ii) every (2, 1, 1)-congruence ρ on M(T, Y) extends to X ⋊ T ;
(iii) for every (2, 1, 1)-congruence ρ of M(T, Y) with ρ ⊆ σM (T,Y), the restriction monoid
M(T, Y)/ρ embeds into a nice semidirect product of a factor semilattice of X by T .
Since any proper restriction monoid S is (2, 1, 1)-isomorphic to the monoid M(S/σ, P (S))
defined by means of the induced left partial action, we have the following consequence.
Corollary 4.3. Let S be a proper restriction monoid such that the induced left partial
action of S/σ on P (S) can be extended to an action · of S/σ on a semilattice X containing
P (S) as a principal ideal in such a way that the action · is nice over a group G with
respect to the identity element of P (S) and X = G · P (S). Then S is (2, 1, 1)-embeddable
in the nice semidirect product X ⋊ (S/σ), which is a (2, 1, 1)-subsemigroup in the inverse
semigroup X ⋊ G, and the (2, 1, 1)-embedding has the following properties:
(i) each (2, 1, 1)-congruence of S extends to X ⋊ (S/σ) (along the (2, 1, 1)-embedding),
and
(ii) for every (2, 1, 1)-congruence ρ of S with ρ ⊆ σ, the restriction monoid S/ρ embeds
into a nice semidirect product of a factor semilattice of X by S/σ.
Now we establish that the perfect restriction monoids M(Ω∗, Y) studied in [7] and [6]
satisfy the assumptions of Proposition 4.2 and Corollary 4.3.
Let Ω be a set, Y a semilattice with identity element 1, and consider a left partial action
of the free monoid Ω∗ on Y which is a monoid homomorphism α : Ω∗ → T op
Y . This defines
a perfect restriction monoid S = M(Ω∗, Y). The free monoid Ω∗ is a submonoid of the
free group FG(Ω) such that FG(Ω) is generated by Ω∗, and each element of FG(Ω) can
be uniquely written as a product of members of an alternating sequence, see Example 3.2.
This allows us to extend α to a left partial action, also denoted by α, of FG(Ω) on Y by
putting αt−1 = α′
t for every t ∈ Ω∗, and then by defining αw1w2···wn = αw1αw2 · · · αwn in
T op
Y for every alternating sequence w1, w2, . . . , wn in Ω∗. Notice that the left partial action
α : FG(Ω) → T op
Y is not a monoid homomorphism any more (cf. [7, arXiv:1402.5849v2, p.
23]). However, if g, h ∈ FG(Ω) such that g, h and gh (the concatenation of g and h) are
reduced words then αgαh = αgh. For every g ∈ FG(Ω) denote the identity element of the
range of αg by Mg. By definition, we clearly have M1 = 1 and, since Mg−1 equals the
identity element of the domain of αg, we have
(6)
Mg = g ⋄ Mg−1
for every g ∈ FG(Ω).
The restriction monoid M(FG(Ω), Y) is easily seen to be an F -inverse monoid, where the
greatest element in the σ-class corresponding to g ∈ FG(Ω) is (Mg, g). Hence M(FG(Ω), Y)
can be represented as a P -semigroup P (FG(Ω), Y, X ) where X is a semilattice and is, up
to isomorphism, uniquely determined. Futhermore, by [9], X can be constructed in the
following manner. The relation χ on the set Y × FG(Ω), defined by (A, g) χ (B, h) if and
only if B = A ◦ g−1h, is a preorder, and so it induces a partial order ≤ on the set X of
all χ-classes where χ is the join of χ and its converse. Denoting the χ-class of the element
18
VICTORIA GOULD, MIKL ´OS HARTMANN, AND M ´ARIA B. SZENDREI
(A, g) ∈ Y × FG(Ω) by [A, g], we can deduce from the proof of [8, Theorem 7.4.5] that
(7)
[A, g] ∧ [B, h] = (cid:2)g−1h ⋄(cid:0)(A ∧ Mg−1h) ◦ g−1h ∧ B(cid:1) , g(cid:3)
for every element [A, g], [B, h] in the semilattice X . Hence X0 = {[A, 1] : A ∈ Y} is
a principal ideal in X , and the mapping Y → X0, A 7→ [A, 1] is an isomorphism. By
identifying X0 with Y along this isomorphism, Y becomes a principal ideal of X . The rule
h · [A, g] = [A, hg] (h ∈ FG(Ω),
[A, g] ∈ X ) defines an action of the group FG(Ω) on X
such that X = FG(Ω)· Y, and the restriction of this action to Y is obviously the left partial
action ⋄.
On the other hand, the action · of FG(Ω) on X naturally restricts to an action, also
denoted by ·, of Ω∗ on X . Now we establish that this action of Ω∗ on X is nice over FG(Ω)
with respect to the element 1. It suffices to show that if g, h ∈ FG(Ω) are reduced words
such that there is no non-empty common prefix of g and h then the inequality 1 ≥ g·1∧h·1
holds in X , or equivalently, g · 1 ∧ h · 1 ∈ Y. Applying (7), (6) and that Mg ∈ Y for every
g ∈ FG(Ω), we obtain that
[1, g] ∧ [1, h] = (cid:2)g−1h ⋄(cid:0)(1 ∧ Mg−1h) ◦ g−1h ∧ 1(cid:1) , g(cid:3)
= (cid:2)g−1h ⋄ (Mg−1h ◦ g−1h), g(cid:3) = [Mg−1h, g] = [g−1h ⋄ Mh−1g, g].
Since g−1h is a reduced word by assumption, we have g−1h ⋄ Mh−1g = αg−1hMh−1g =
αg−1(αhMh−1g), whence it follows that both αhMh−1g = h ⋄ Mh−1g and αg−1(αhMh−1g) =
(h ⋄ Mh−1g) ◦ g are defined. This implies by the definition of the preorder χ that
[1, g] ∧ [1, h] = [(h ⋄ Mh−1g) ◦ g, g] = [h ⋄ Mh−1g, 1] ∈ Y,
thus completing the proof that g · 1 ∧ h · 1 ∈ Y.
Summarising, we have strengthened the results [7, Theorem 28] and [6, Proposition 8.4]
in the following way:
Corollary 4.4. Let S be a perfect restriction monoid such that S/σ is a free monoid on a
set Ω. Then the induced left partial action of S/σ on P (S) can be extended to an action · of
S/σ on a semilattice X contaning P (S) as a principal ideal such that the action · is nice over
the free group FG(Ω) with respect to the identity element of P (S) and X = FG(Ω) · P (S).
Consequently, S is (2, 1, 1)-embeddable in the nice semidirect product X ⋊ Ω∗, which is a
(2, 1, 1)-subsemigroup in the inverse semigroup X ⋊ FG(Ω), and the following are valid:
(i) each (2, 1, 1)-congruence of S extends to X ⋊Ω∗ (along the (2, 1, 1)-embedding), and
(ii) for every (2, 1, 1)-congruence ρ of S with ρ ⊆ σ, the restriction monoid S/ρ embeds
into a nice semidirect product of a factor semilattice of X by Ω∗.
References
[1] C. Cornock, V. Gould, Proper restriction semigroups and partial actions, J. Pure Appl. Algebra 216
(2012) 936 -- 949.
[2] J. Fountain, G. M. S. Gomes, Proper left type-A monoids revisited Glasgow Math. J. 35 (1993)
293 -- 306.
EMBEDDING OF RESTRICTION SEMIGROUPS
19
[3] J. Fountain, G. M. S. Gomes, V. Gould, The free ample monoid, Internat. J. Algebra Comput. 19
(2009) 527 -- 554.
[4] G. M. S. Gomes, M. B. Szendrei, Almost factorizable weakly ample semigroups Comm. Algebra 35
(2007) 3503 -- 3523.
[5] V. Gould, V. Notes on restriction semigroups and related structures; formerly (Weakly) left E-ample
semigroups http://www-users.york.ac.uk /~varg1/gpubs.htm
[6] P. Jones, Almost perfect restriction semigroups, J. Algebra 445 (2016) 193 -- 220.
[7] G. Kudryavtseva, Partial monoid actions and a class of restriction semigroups, J. Algebra 429 (2015)
342 -- 370.
[8] M. V. Lawson, Inverse semigroups: The Theory of Partial Symmetries. World Scientific, Singapore,
1998.
[9] W. D. Munn, A note on E-unitary inverse semigroups, Bull. London Math. Soc. 8 (1976) 71 -- 76.
[10] M. B. Szendrei, Proper covers of restriction semigroups and W -products, Internat. J. Algebra Comput.
22 (2012) 1250024, 16 pp.
[11] M. B. Szendrei, Embedding into almost left factorizable restriction semigroups, Comm. Algebra 41
(2013) 1458 -- 1483.
Department of Mathematics, University of York, Heslington, York, YO10 5DD, UK
E-mail address: [email protected]
Bolyai Institute, University of Szeged, Aradi v´ertan´uk tere 1, Szeged, Hungary, H-6720
E-mail address: [email protected]
Bolyai Institute, University of Szeged, Aradi v´ertan´uk tere 1, Szeged, Hungary, H-6720
E-mail address: [email protected]
|
1601.04775 | 3 | 1601 | 2017-04-25T21:57:36 | Generalized nil-Coxeter algebras, cocommutative algebras, and the PBW property | [
"math.RA",
"math.RT"
] | Poincare-Birkhoff-Witt (PBW) Theorems have attracted significant attention since the work of Drinfeld (1986), Lusztig (1989), and Etingof-Ginzburg (2002) on deformations of skew group algebras $H \ltimes {\rm Sym}(V)$, as well as for other cocommutative Hopf algebras $H$. In this paper we show that such PBW theorems do not require the full Hopf algebra structure, by working in the more general setting of a "cocommutative algebra", which involves a coproduct but not a counit or antipode. Special cases include infinitesimal Hecke algebras, as well as symplectic reflection algebras, rational Cherednik algebras, and more generally, Drinfeld orbifold algebras. In this generality we identify precise conditions that are equivalent to the PBW property, including a Yetter-Drinfeld type compatibility condition and a Jacobi identity. We also characterize the graded deformations that possess the PBW property. In turn, the PBW property helps identify an analogue of symplectic reflections in general cocommutative bialgebras.
Next, we introduce a family of cocommutative algebras outside the traditionally studied settings: generalized nil-Coxeter algebras. These are necessarily not Hopf algebras, in fact, not even (weak) bialgebras. For the corresponding family of deformed smash product algebras, we compute the center as well as abelianization, and classify all simple modules. | math.RA | math |
Contemporary Mathematics
Volume 688, 2017
http://dx.doi.org/10.1090/conm/688/13832
Generalized nil-Coxeter algebras, cocommutative algebras,
and the PBW property
Apoorva Khare
Abstract. Poincar´e -- Birkhoff -- Witt (PBW) Theorems have attracted signifi-
cant attention since the work of Drinfeld (1986), Lusztig (1989), and Etingof --
Ginzburg (2002) on deformations of skew group algebras H ⋉ Sym(V ), as well
as for other cocommutative Hopf algebras H. In this paper we show that such
PBW theorems do not require the full Hopf algebra structure, by working
in the more general setting of a "cocommutative algebra", which involves a
coproduct but not a counit or antipode. Special cases include infinitesimal
Hecke algebras, as well as symplectic reflection algebras, rational Cherednik
algebras, and more generally, Drinfeld orbifold algebras. In this generality we
identify precise conditions that are equivalent to the PBW property, includ-
ing a Yetter -- Drinfeld type compatibility condition and a Jacobi identity. We
also characterize the graded deformations that possess the PBW property. In
turn, the PBW property helps identify an analogue of symplectic reflections
in general cocommutative bialgebras.
Next, we introduce a family of cocommutative algebras outside the tra-
ditionally studied settings: generalized nil-Coxeter algebras. These are neces-
sarily not Hopf algebras, in fact, not even (weak) bialgebras. For the corre-
sponding family of deformed smash product algebras, we compute the center
as well as abelianization, and classify all simple modules.
1. Introduction
In the study of deformation algebras, their structure and representations, one
commonly begins by understanding their connection to the corresponding asso-
ciated graded algebras (which are generally better behaved). Such connections
of course provide desirable "monomial bases", but also additional structural and
representation-theoretic knowledge.
A first step in understanding these connections involves showing that these
filtered algebras satisfy the Poincar´e -- Birkhoff -- Witt (PBW) property, in that they
are isomorphic as vector spaces to their associated graded algebras. Such results
are known as PBW theorems in the literature. The terminology of course origi-
nates with the classical result for the universal enveloping algebra of a Lie algebra.
2010 Mathematics Subject Classification. Primary 16S80; Secondary 16S40, 16T15, 20C08,
20F55.
Key words and phrases. Cocommutative algebra, PBW theorem, graded deformation, Jacobi
identity, symplectic reflection, Coxeter group, generalized nil-Coxeter algebra.
139
c(cid:13)0000 (copyright holder)
140
APOORVA KHARE
However, it has gathered renewed attention over the past few decades owing to
tremendous interest in the study of orbifold algebras and their generalizations,
which we now briefly describe.
In a seminal paper [12], Drinfeld pioneered the study of smash product al-
gebras of the form kG ⋉ Sym(V ), where a group G acts on a k-vector space V .
Drinfeld's results were rediscovered and extended by Etingof and Ginzburg in their
landmark paper [14], which introduced symplectic reflection algebras and furthered
our understanding of rational Cherednik algebras. These algebras serve as "non-
commutative" coordinate rings of the orbifolds V /G; see [32] for a related set-
ting. Subsequently, Etingof, Ginzburg, and Gan replaced the group by algebraic
distributions of a reductive Lie group G. This led to the study of infinitesimal
Hecke algebras in [13] (and several recent papers), where U g acts on Sym(V ), with
g = Lie(G). These families of deformed algebras continue to be popular and im-
portant objects of study, with connections to representation theory, combinatorics,
and mathematical physics.
A common theme underlying all of these settings is that a cocommutative Hopf
algebra H acts on the vector space V and hence on Sym(V ). The aforementioned
families of algebras Hλ,κ are created by deforming two sets of relations:
• The relations V ∧ V 7→ 0 in the smash product algebra H ⋉ Sym(V )
are deformed using an anti-symmetric bilinear form κ : V ∧ V → H, or
more generally, κ : V ∧ V → H ⊕ V . These deformed relations feature in
[12 -- 14], and follow-up works.
• The relations g·v = g(v)g for grouplike elements g with H a group algebra,
were deformed by Lusztig [32] to create graded affine Hecke algebras, using
a bilinear form λ : H ⊗ V → H.
The forms λ, κ define a filtered algebra, and an important question is to char-
acterize those deformations Hλ,κ whose associated graded algebra is isomorphic
to H0,0 = H ⋉ Sym(V ). Such parameters λ, κ are said to correspond to PBW
deformations, and have been studied in the aforementioned works as well as by
Braverman and Gaitsgory [5] among others. More recently, in a series of papers
[35 -- 37], Shepler and Witherspoon have shown PBW theorems in a wide variety of
settings (skew group algebras, Drinfeld orbifold algebras, Drinfeld Hecke algebras,
. . . ), that encompass many of the aforementioned cases. We also point the reader
to the comprehensive survey [38] for more on the subject. This includes the case
of Sym(V ) replaced by a quantum symmetric algebra. Perhaps one of the most
general versions in the literature is the recent work [43] by Walton and Wither-
spoon, in which H is replaced by a Hopf algebra, and Sym(V ) by a Koszul algebra.
For completeness, we also mention work in related flavors:
[19] studies general-
ized Koszul algebras, while [1, 44] analyze deformations of Hopf algebra actions on
"doubled" pairs of module algebras.
We now point out some of the novel features and extensions in the present pa-
per. First, all of the aforementioned settings involve H being a bialgebra -- in fact,
a Hopf algebra. In this paper we isolate the structure required to study the PBW
property, and show that it includes the coproduct but not the counit or antipode.
More precisely, we work in the more general framework of a (cocommutative) alge-
bra with coproduct. This is a strictly weaker setting than that of a bialgebra, as
it also includes examples such as the nil-Coxeter (or nil-Hecke) algebra associated
to a Weyl group, N CW . Recall that these algebras were originally introduced by
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
141
Fomin and Stanley [17] as Demazure operators in the study of Schubert polyno-
mials, though they appear implicitly in previous work [3, 28] on the cohomology of
generalized flag varieties for semisimple and Kac -- Moody groups, respectively; see
also [30]. Nil-Coxeter algebras have subsequently been studied in their own right
[6, 45] as well as in the context of categorification [26, 27], among others.
Nil-Coxeter algebras are necessarily not bialgebras (hence not Hopf algebras).
Thus, deformations over such cocommutative algebras have not been considered to
date in the literature.
Second, we introduce a novel class of Hecke-type algebras, the generalized nil-
Coxeter algebras, which encompass the usual nil-Coxeter algebras. These alge-
bras have not been studied in the literature.
In this paper we will specifically
study deformations over generalized nil-Coxeter algebras. Moreover, our results
are characteristic-free.
An additional novelty of the present work is that in all of the aforementioned
works in the literature, either the bilinear form κV : V ∧ V → V is assumed to
be identically zero, or/and λ : H ⊗ V → H is identically zero. The present pa-
per addresses this gap by working with algebras for which all three parameters
λ, κV , κA = κ − κV are allowed to be nonzero. (All notation is explained in Defini-
tion 2.3 below.)
Organization of the paper. We now outline the contents of the present
paper, which can be thought of as having two parts. In Section 2, we introduce the
general notion of a cocommutative k-algebra A, i.e., an algebra with a multiplicative
coproduct map that is cocommutative (over a unital ground ring k). We next state
and prove one of our main results: a PBW-type theorem for deformations Hλ,κ of
the smash product algebra H0,0 = A ⋉ Sym(V ). Here, A acts on tensor powers of
V via the coproduct, and on the symmetric algebra because of cocommutativity.
In Section 3, we explain the connection between the PBW theorem and defor-
mation theory. Specifically, we identify the graded k[t]-deformations of H0,0 whose
fiber at t = 1 has the PBW property. This extends various results in the litera-
ture; see [35, 37]. The first part of the paper concludes in Section 4, by examining
well-known notions in the Hopf algebra literature in the broader setting of cocom-
mutative algebras. This includes studying the cases where A is a cocommutative
bialgebra or Hopf algebra. We classify the parameters λ, κ for which Hλ,κ has
the same structure, and relate the PBW property to the Yetter -- Drinfeld condi-
tion, a natural compatibility condition that arises in Hopf-theoretic settings. We
also extend the notion of 'symplectic reflections' from groups to all cocommutative
bialgebras.
In the second part of the paper, we study a specific family of cocommutative
algebras that are not yet fully explored in the literature. Thus, in Section 5 we
introduce a family of generalized nil-Coxeter algebras associated to a Coxeter group
W ; these are closely related to Coxeter groups and their generalizations studied by
Coxeter and Shephard -- Todd [9, 10, 34].
Generalized nil-Coxeter algebras are necessarily not bialgebras; thus they fall
strictly outside the Hopf-theoretic setting. In the remainder of the paper, we study
the deformations Hλ,κ over generalized nil-Coxeter algebras. We first study the
Jacobi identity in such algebras Hλ,κ, and classify all Drinfeld-type deformations
H0,κ with the PBW property. In the final section of the paper, we study additional
142
APOORVA KHARE
properties of the algebras Hλ,κ, including computing the center and abelianization,
and classifying simple modules.
2. Cocommutative algebras, smash products, and the PBW theorem
Global assumptions: Throughout this paper, we work over a ground ring k,
which is a unital commutative ring. We also fix a cocommutative k-algebra (A, ∆),
defined below, and a k-free A-module V .
By dim V for a free k-module V , we will mean the (possibly infinite) k-rank of
V . In this paper, all k-modules, including all k-algebras, are assumed to be k-free.
Unless otherwise specified, all (Hopf) algebras, modules, and bases of modules are
with respect to k, and all tensor products are over k.
2.1. Cocommutative algebras and the PBW theorem. We begin by
introducing the main construction of interest and the main result of the first part
of this paper.
Definition 2.1. Suppose A is a unital associative k-algebra.
(1) A is an algebra with coproduct if there exists a k-algebra map ∆ : A →
A⊗k A called the coproduct, such that ∆(1) = 1⊗1 and ∆ is coassociative,
i.e., (∆ ⊗ 1) ◦ ∆ = (1 ⊗ ∆) ◦ ∆ : A → A ⊗ A ⊗ A.
(2) An algebra with coproduct is said to be cocommutative if ∆ = ∆op.
Notice that bialgebras and Hopf algebras (with the usual coproduct) are ex-
amples of algebras with coproduct (with k a field). As pointed out to us by Susan
Montgomery, one could a priori have considered weak bialgebras (these feature
prominently in the theory of fusion categories [15]), but these provide no addi-
tional examples, as explained at the end of [4, §2.1]: since ∆(1) = 1 ⊗ 1 by as-
sumption, a cocommutative algebra is a bi/Hopf-algebra if and only if it is a weak
bi/Hopf-algebra. Additional examples do arise, however, using nil-Coxeter alge-
bras, as explained in Remark 5.3 below. These algebras show that the notion of an
algebra with coproduct is strictly weaker than that of a (weak) bialgebra.
We also remark that every unital k-algebra A is an algebra with coproduct, if
we define ∆L(a) := a⊗1 or ∆R(a) := 1⊗a. (Thus, the definition essentially involves
a choice of coproduct.) However, A need not have a cocommutative coproduct in
general.
Given a ∈ A, write ∆(a) = P a(1) ⊗ a(2) and ∆op(a) = P a(2) ⊗ a(1), in the
usual Sweedler notation. We now use ∆ to first define tensor and symmetric product
A-module algebras, as well as undeformed Drinfeld Hecke algebras. Suppose (A, ∆)
acts on a free k-module V (not necessarily of finite rank), denoted by v 7→ a(v).
Notice that T V := TkV has an augmentation ideal T +V := V · TkV , and this ideal
is an A-module algebra via:
a(v1 ⊗ · · · ⊗ vn) :=X a(1)(v1) ⊗ · · · ⊗ a(n)(vn),
We do not include the case n = 0 here, since A does not have a counit ε.
∀a ∈ A, v1, . . . , vn ∈ V, n > 1.
Definition 2.2. Given a k-algebra A, let Amult denote the left A-module A,
under left multiplication. Now given (A, ∆) and V as above, the smash product of
T V and A, denoted by T V ⋊Amult, is defined to be the k-algebra T (V ⊕Amult), with
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
143
the multiplication relations given by a · a′ := aa′ in A, (v′ ⊗ a′) · (1 ⊗ a) = v′ ⊗ a′a,
and
(1)(v)) ⊗ a′
(2) · a,
∀a, a′ ∈ A, v′ ∈ T V, v ∈ T +V.
(v′ ⊗ a′) · (v ⊗ a) :=X(v′ · a′
We use − ⋊ A rather than A ⋉ − in this paper. Also note that for 1A to
commute with V requires ∆(1) = 1 ⊗ 1 as above. Now denote by ∧2V ⊂ V ⊗k V
the k-span of v ∧ v′ := v ⊗ v′ − v′ ⊗ v; then ∧2V is an A-submodule of T +V
because of the cocommutativity assumption on A, which implies that a(v1 ∧ v2) =
P a(1)(v1) ∧ a(2)(v2). Thus, one can quotient T V ⋊ A by the related two-sided
"A-module ideal", to define:
(2.1)
H0,0(A, V ) = H0,0 = Sym(V ) ⋊ A :=
T V ⋊ A
(T V · ∧2V · T V ) ⋊ A
.
The algebra H0,0(A, V ) will be referred to as the smash product of Sym(V ) and A.
We are now able to introduce deformations of this smash product algebra.
Definition 2.3. Given (A, ∆) and V as above, as well as bilinear forms λ ∈
Homk(V ⊗ A, A) and κ ∈ Homk(∧2V, A ⊕ V ), the deformed smash product algebra
Hλ,κ = Hλ,κ(A, V ) with parameters λ, κ is defined to be the quotient of T (V ⊕ A)
by the multiplication in A and by
(2.2)
vv′ − v′v =: [v, v′] = κ(v, v′),
∀a ∈ A, v, v′ ∈ V.
av −X a(1)(v)a(2) = λ(a, v),
Also define κV ∈ Homk(V ∧ V, V ) and κA ∈ Homk(V ∧ V, A) to be the projections
of κ to V, A respectively.
Observe that λ being trivial is equivalent to the A-action preserving the grading
on Sym(V ). Moreover, we will write Hλ,κ instead of Hλ,κ(A, V ) if A, V are clear
from context.
The deformed smash product algebras Hλ,κ = Hλ,κ(A, V ) encompass a very
large family of deformations considered in the literature, including universal en-
veloping algebras, skew group algebras, Drinfeld orbifold algebras, Drinfeld Hecke
algebras, symplectic reflection algebras, rational Cherednik algebras, degenerate
affine Hecke algebras and graded Hecke algebras, Weyl algebras, infinitesimal Hecke
algebras, and many others. This is an area of research that is the focus of tremen-
dous recent activity; see [11 -- 14, 24, 31, 32, 40 -- 42], and subsequent follow-up works
in the literature.
Remark 2.4. In order to place the work in context, we briefly comment on
how our framework compares to other papers in the PBW literature. The paper
encompasses other works in two aspects: first, the algebra (A, ∆) is strictly weaker
than a bialgebra. Second, the deformation parameters λ, κV , κA can all be nonzero.
At the same time, we impose two restrictions that are present in some papers but
not in others: first, we work with Sym(V ) and not a quantum algebra, nor a
general Koszul algebra (e.g., a PBW algebra). Second, for ease of exposition we
only consider algebras with im(κV ) a subset of V instead of V ⊗ A; this is akin to
the assumption λ ≡ 0 in [35, 43], or κV ≡ 0 in [37].
Notice that the algebras Hλ,κ are filtered, by assigning deg A = 0, deg V = 1.
We say that Hλ,κ has the PBW property if the surjection from H0,0 = Sym(V )⋊
A to the associated graded algebra of Hλ,κ is an isomorphism. Equivalently, the
(2.3)
(1)(v))a′
(2).
equations hold in A and V ⊗ A respectively:
λ(aa′, v) = aλ(a′, v) +X λ(a, a′
aκA(v, v′)− X κA(a(1)(v), a(2)(v′))a(3)
X a(1)(κV (v, v′))a(2)− X κV (a(1)(v), a(2)(v′))a(3)
(2.4)
(2.5)
= λ(λ(a, v), v′) − λ(λ(a, v′), v) − λ(a, κV (v, v′)),
144
APOORVA KHARE
PBW theorem holds for Hλ,κ if for any (totally) ordered k-basis {xi : i ∈ I} of the
free k-module V and {a ∈ J1} of the k-free k-algebra A, the collection
{X · a : X is a word in the xi in non-decreasing order of subscripts, a ∈ J1}
is a k-basis of Hλ,κ. We now state the main result of the first part of the paper,
which is a PBW Theorem for the algebras Hλ,κ.
Theorem 2.5 (PBW Theorem). Suppose (A, ∆) is a k-free cocommutative k-
algebra, and V a k-free A-module. Define Hλ,κ with κ = κV ⊕ κA : V ∧ V → V ⊕ A
as above, and suppose A = k · 1L A′ for a free k-submodule A′ ⊂ A. Then the
(1) Hλ,κ has the PBW property (for a k-basis of V and a k-basis of A con-
following are equivalent:
taining 1).
(2) The natural map : A ⊕ (V ⊗ A) → Hλ,κ is an injection.
(3) λ : A ⊗ V → A and κ : V ∧ V → V ⊕ A satisfy the following conditions:
(a) A-action on V : For all a, a′ ∈ A and v ∈ V , the following equation
holds in A:
(b) A-compatibility of λ, κ: For all a ∈ A and v, v′ ∈ V , the following
= X λ(a, v)(1)(v′)λ(a, v)(2) −X λ(a, v′)(1)(v)λ(a, v′)(2)
+X a(1)(v)λ(a(2), v′) −X a(1)(v′)λ(a(2), v).
(c) Jacobi identities: For all v1, v2, v3 ∈ V , the following cyclic sum
X(cid:8)
vanishes:
[κ(v1, v2), v3] := [κ(v1, v2), v3] + [κ(v2, v3), v1] + [κ(v3, v1), v2] = 0.
More precisely, the following equations hold in A and V ⊗ A respec-
tively (identifying V with V ⊗ 1A ⊂ V ⊗ A):
(2.6)
(2.7)
X(cid:8)
λ(κA(v1, v2), v3) = X(cid:8)
X(cid:8)
κV (κV (v1, v2), v3) = X(cid:8)
κA(v1, κV (v2, v3)),
v1κA(v2, v3) −X(cid:8)
κA(v1, v2)(1)(v3)κA(v1, v2)(2).
As observed by Shepler and Witherspoon in their papers [35] -- [38], their ver-
sions of the PBW theorem, and therefore ours, specialize to the PBW criteria for
the algebras studied by Drinfeld, Etingof -- Ginzburg, Lusztig, as well as in numerous
follow-up papers on these families of algebras (see the remarks following Definition
2.3 for additional references). Thus, Theorem 2.5 unifies several results in the litera-
ture and extends them to arbitrary cocommutative algebras. As a specific example,
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
145
we point the reader to [37, Theorem 3.1] for the analogous result with k a field, A
a group algebra, and κV ≡ 0.
Remark 2.6. Notice that the conditions in part (3) of the theorem always
hold in Hλ,κ. In other words, Equations (2.3) -- (2.7) hold in the image of the space
A⊕(V ⊗A) in Hλ,κ, by considering the equations corresponding to the associativity
of the algebra Hλ,κ:
aa′ · v = a · (a′ · v),
a · (vv′ − v′v) = a · κ(v, v′), X(cid:8)
[κ(v1, v2), v3] = 0.
The assertion of Theorem 2.5 is that the PBW property is equivalent to these
equations holding in A ⊕ (V ⊗ A).
Remark 2.7. It is easy to verify that the Jacobi identities (2.6), (2.7) hold in
A ⊕ (V ⊗ A) if dimk V 6 2, since in that case the left and right hand sides of both
equations vanish.
If moreover dimk V 6 1, then the A-compatibility conditions
(2.4), (2.5) also hold in A ⊕ (V ⊗ A), since κV , κA ≡ 0.
2.2. Proof of the PBW Theorem. We now prove Theorem 2.5 using the
Diamond Lemma [2]. As we work with a general cocommutative algebra (which is
strictly weaker than a cocommutative bialgebra), and moreover, work with possibly
nonzero λ, κV , the proof is written out in some detail. To prove Theorem 2.5, we
will require the unit 1 to be one of our k-basis vectors for A; words involving this
basis vector are to be considered "without" the 1.
Proof of the PBW Theorem 2.5. Clearly, (1) =⇒ (2), and (2) =⇒ (3)
using Remark 2.6. The goal in the remainder of this proof (and this section) is
to show that (3) =⇒ (1). We begin by writing down the relations in Hλ,κ
systematically. Recall that A = k · 1A ⊕ A′; now suppose {aj : j ∈ J} is a k-basis
of the k-submodule A′. Write
(2.8)
J1 := {aj : j ∈ J} ⊔ {1A},
a0 := 1A,
J0 := J ⊔ {0}.
We also fix a total ordering on J1 and correspondingly on J0, with 0 6 j for all
j ∈ J0.
Next, fix a totally ordered k-basis of V , denoted by {xi : i ∈ I}. (Thus, I is
also totally ordered.) We then define various structure constants, with the sums
running over J0 and I, and using Einstein notation throughout. We first define the
structure constants from A and its action on V :
(2.9)
ajak = ul
aj(xk) = sh
jkxh,
∆(aj ) = rkl
j ak ⊗ al.
jkal,
0j = δi,j, si
In particular, ui
structure constants for the maps λ, κ:
(2.10)
j0 = ui
0k = δi,k, and rkl
0 = δk,0δl,0. Next, we define the
κA(xj, xk) = vl
jkal.
It now follows that Hλ,κ is a quotient of T (V ⊕ A), with the defining relations:
(j > k);
κV (xj , xk) = wl
λ(aj , xk) = ql
jkxl,
jkal,
(2.11)
jkal + wh
jkxh
(j > k),
xjxk = xkxj + vl
ajak = ul
ajxk = tmn
jkal,
jk xnam + ql
jkal, where tmn
jk = rlm
j sn
lk.
Thus, the q, r, s, t, u, v, w are all structure constants in k, for all choices of indices.
146
APOORVA KHARE
To show (1), we first write down additional consequences of the structure of
A, V . The following equations encode the associativity, coassociativity, and cocom-
mutativity of A:
(2.12)
lm ∀i, j, m, n;
ijun
i rjm
k
ul
jmun
il = ul
rjl
i rmn
l = rkn
rkl
j = rlk
j
∀i, j, m, n;
∀j, k, l ∈ J0 = J ⊔ {0}.
The next condition is that ∆ is multiplicative, which yields:
ul
jkrmn
l
(am ⊗ an) = ul
= rcd
jk∆(al) = ∆(ajak) = ∆(aj)∆(ak)
j ref
k um
k (ac ⊗ ad)(ae ⊗ af ) = rcd
j ref
ceun
df (am ⊗ an).
Equating coefficients in A ⊗ A, we conclude that
(2.13)
ul
jkrmn
l = rcd
j ref
k um
ceun
df .
Finally, V is an A-module, which yields:
sn
jmsm
kixn = aj (sm
kixm) = aj(ak(xi)) = (ajak)(xi) = um
jkam(xi) = um
jksn
mixn,
whereby we get
(2.14)
sn
jmsm
ki = um
jksn
mi.
We now proceed with the proof, using the terminology of [2]. The reduction
system S consists of the set of algebra relations (2.11). Then expressions in the left
and right hand sides in the equations in (2.11) are what Bergman calls fσ and Wσ,
respectively.
Define X := {aj : j ∈ J} ∪ {xi : i ∈ I}. Then the expressions in the free
semigroup hXi generated by X that are irreducible (i.e., cannot be reduced via the
operations fσ 7→ Wσ via the Equations (2.11)) are precisely the PBW-basis that
was claimed earlier, i.e. words xi1 · · · xil · aj, for j ∈ J0 and i1 6 i2 6 · · · 6 il, all
in I. This also includes the trivial word 1.
Next, define a semigroup partial ordering 6 on X, first on its generators via:
(2.15)
1 < xi < aj,
∀j ∈ J, i ∈ I,
and then extend to a total order on hXi, as follows: words of length m are strictly
smaller than words of length n, whenever m < n; and words of equal lengths are
(totally) ordered lexicographically. It is easy to see that 6 is a semigroup partial
order on hXi, i.e., if a 6 b then waw′ 6 wbw′ for all w, w′ ∈ hXi. Moreover, 6 is
indeed compatible with S, in that each fσ reduces to a linear combination Wσ of
monomials strictly smaller than fσ.
We now recall the descending chain condition, which says that given a monomial
B ∈ hXi, any sequence of reductions applied to B yields an expression that is
irreducible in finitely many steps. Now the following result holds.
Lemma 2.8. The semigroup partial order 6 on hXi satisfies the descending
chain condition.
Proof. We prove a stronger assertion; namely, we produce an explicit upper
bound for the number of reductions successively applicable on a monomial. Given
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
147
a word w = T1 · · · Tn, with Ti ∈ X ∀i, define its misordering index mis(w) to be
o + p + pr + q + r3, where
o = o(w) := #{(i, j) : i < j, Ti, Tj ∈ V, Ti > Tj},
p = p(w) := #{(i, j) : i < j, Ti ∈ A′, Tj ∈ V },
q = q(w) := #{i : Ti ∈ A′},
r = r(w) := #{i : Ti ∈ V } = n − q.
We now claim that each reduction strictly reduces the misordering index of each
resulting monomial; this claim shows the result. As an illustration of the claim, we
present the most involved case: when fσ = xjxk with j > k, and the monomial we
consider via the reduction fσ 7→ Wσ corresponds to al for some l ∈ J. For this new
word w′, notice that q increases by 1, whereas r reduces by 2 (so r > 2), o reduces
by at least 1, and p may increase by at most the number of x to the right of the
new a, which is at most r − 2. So, o + q does not increase, and we now claim that
p + pr + r3 strictly reduces. Indeed, p′ 6 p + r − 2, r′ 6 r − 2, whence:
p′(1 + r′) + (r′)3 6 (p + r − 2)(1 + (r − 2)) + (r − 2)3
6 p(1 + r) + (r − 2) + (r − 2)2 + (r − 2)3
= p(1 + r) + (r − 2)(r2 − 3r + 3) < p(1 + r) + r · r2.
Hence mis(w′) < mis(w) as desired.
(cid:3)
The final item utilized in the proof of the PBW theorem, is the notion of
ambiguities. It is clear that no fσ is a subset of fτ for some σ, τ ∈ S; hence there
are no inclusion ambiguities. In light of Lemma 2.8 and the Diamond Lemma [2,
Theorem 1.2], it suffices to resolve all overlap ambiguities using the given conditions
in (3). We begin by writing down these conditions explicitly using the structure
constants in A. Explicit computations using these constants and Equations (2.3) --
(2.7) yield the following five equations, respectively:
(2.16)
(2.17)
(2.18)
(2.19)
(2.20)
jkuh
vl
wl
il − tmn
jktdc
ij tcd
il − tmn
li = ql
ld = ql
nl = qm
kiuh
ij qh
ij tdc
ul
jkqh
ncuh
mkvl
mkwc
ij tdl
ij qh
vl
jnuh
lm,
jkqh
lj − wl
il,
ij qd
mj + tmc
jl + tmn
ki ql
ikqh
lk − ql
mk − qm
iktdc
jkvh
wm
im,
lk = X(cid:8)(i,j,k)
X(cid:8)(i,j,k)
lk · (xh ⊗ a0) = X(cid:8)(i,j,k)
wl
ij wh
X(cid:8)(i,j,k)
mk − tmc
ik qd
mj,
vm
jk(xi ⊗ am) − X(cid:8)(i,j,k)
vl
ij tdc
lk · (xc ⊗ ad).
We now resolve the overlap ambiguities, which are of four types, and correspond
to the associativity of the algebra Hλ,κ (see Remark 2.6):
aiajak, ajakxi, akxixj (i > j), xixj xk(i > j > k).
Notice that the first type is resolvable because A is an associative algebra. We only
analyse the second type of ambiguity in what follows; the others involve carrying
out similar (and more longwinded) computations, that use the structure constants
of the cocommutative algebra A with coproduct.
148
APOORVA KHARE
To resolve the ambiguity ajakxi, using the above analysis in this proof we
compute:
(aj ak)xi = ul
jkalxi = ul
jktmh
li xham + ul
jkqh
liah = ul
jkrmc
l sh
ci · xham + ul
jkqh
liah.
On the other hand,
aj(akxi) = tf g
ki ajxgaf + ql
kiajal = tf g
= tf g
ki tyh
ki tyh
jg xhayaf + tf g
ki ql
jg ul
yf xhal + tmn
ki ql
jgalaf + ql
jnuh
kiuh
lmah + ql
jlah
kiuh
jlah.
The overlap ambiguity is resolved if these two expressions are shown to be equal.
In light of (2.16), it suffices to show that, after relabelling indices, we have for all
i, j, k, l, h (or h-l):
um
jkrlf
msh
f i = tf g
ki tyh
jg ul
yf .
To see why this holds, begin with the right-hand side, expand using the definition
of t, and then use Equations (2.13), (2.14) above:
tf g
ki tyh
jg ul
yf = rf a
= rf a
= um
k sg
k ryn
jkrlg
ai · ryn
j sh
j ul
yf (ug
m · sh
gi,
ng · ul
nash
yf = rf a
gi) = ryn
k ryn
j ul
j rf a
k ul
yf · (sh
yf ug
na · sh
gi
ngsg
ai)
which is precisely the left-hand side. Thus the ambiguity is resolved.
(cid:3)
3. Characterization via deformation theory
We now explain how PBW deformations can be naturally understood via defor-
mation theory. In this section, suppose k is a field. Given an associative algebra B
and an indeterminate t, a deformation of B over k[t] is an associative k[t]-algebra
(Bt, ∗) that is isomorphic to B[t] as a vector space, such that Bt/tBt is isomorphic
to B as a k-algebra. In particular, we can write the multiplication of two elements
b1, b2 ∈ B ⊗ t0 ⊂ Bt as:
b1 ∗ b2 = b1b2 +Xj>0
µj(b1, b2)tj,
where µj : B ⊗ B → B is k-linear and only finitely many terms are nonzero in the
above sum.
If moreover B is Z>0-graded, then a graded k[t]-deformation of B is a defor-
mation of B over k[t] that is graded with deg t = 1, i.e., each µj : B ⊗ B → B is
homogeneous of degree −j. The map µj is also called the jth multiplication map.
Henceforth in this section we will consider the special case of the Z>0-graded
algebra B := H0,0 = Sym(V ) ⋊ A, with (A, ∆) a cocommutative algebra as above.
Our first goal in this section is to show that the PBW property for the algebras
Hλ,κ has a natural reformulation in terms of graded deformations of H0,0 over k[t].
Such a result was shown in [37, §6] in the special case of A a group algebra, and
further assuming that κV ≡ 0. We now explain how the assumption κV ≡ 0 is
related to that in loc. cit. of requiring V ⊗ V ⊂ ker µ1, by extending the result to
general κV : V ∧ V → V and all cocommutative algebras A.
Theorem 3.1. Suppose k is a field (of arbitrary characteristic), (A, ∆) is co-
commutative, and V an A-module. Consider the following two statements.
(1) Hλ,κ satisfies the PBW Theorem 2.5.
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
149
(2) There exists a graded k[t]-deformation Bt of B := H0,0 = Sym(V ) ⋊ A,
whose multiplication maps µ1, µ2 satisfy (for all v, v′ ∈ V and a ∈ A):
(3.1)
λ(a, v) = µ1(a ⊗ v) −X µ1(a(1)(v) ⊗ a(2)),
κV (v, v′) = µ1(v ⊗ v′) − µ1(v′ ⊗ v),
κA(v, v′) = µ2(v ⊗ v′) − µ2(v′ ⊗ v).
Then (1) =⇒ (2), and the converse holds if dim A, dim V are both finite. Moreover,
if these statements hold then Hλ,κ ∼= Btt=1.
Thus, the structure maps λ, κV , κA in Hλ,κ can be naturally reformulated using
the multiplication maps µ1, µ2 in a graded deformation of H0,0, whenever Hλ,κ has
the PBW property.
Proof. We provide a sketch of the proof as it closely resembles the arguments
for proving [37, Proposition 6.5 and Theorem 6.11]. First suppose (1) holds. Define
(Bt, ∗) to be the associative algebra over k[t] generated by A, V , with the following
relations (for all a ∈ A, v, v′ ∈ V ):
a ∗ v = X a(1)(v) ∗ a(2) + λ(a, v)t,
v ∗ v′ − v′ ∗ v = κV (v, v′)t + κA(v, v′)t2.
This yields a Z>0-graded algebra with deg(t) = deg(V ) = 1 and deg(A) = 0.
Moreover, Bt ∼= H0,0 ⊗k k[t] as vector spaces, since Hλ,κ has the PBW property.
Now verify using the definitions and the relations in the algebra (Bt, ∗), that
κV (v, v′)t+κA(v, v′)t2 = v ∗v′ −v′ ∗v = vv′ +Xj>0
µj(v ⊗v′)tj −v′v −Xj>0
µj(v′ ⊗v)tj.
As this is an equality of polynomials in H0,0[t], we equate the linear and quadratic
terms in t on both sides, to obtain the last two equations in (3.1). The first
equation in (3.1) follows from a similar computation. This shows (2), and moreover,
Btt=1 ∼= Hλ,κ.
Conversely, suppose (2) holds, and dim V, dim A < ∞. Define Ft := Tk[t](V ⊕
A)/(a · a′ − aa′); then we have an algebra map f : Ft → Bt, which sends monomials
x1 · · · xk (with each xi ∈ V ⊕ A) to x1 ∗ · · · ∗ xk. One shows as in [37] that f is
surjective, and the vectors
av−X a(1)(v)a(2)−λ(a, v)t = av−X a(1)(v)a(2)−µ1(a, v)t+X µ1(a(1)(v)⊗a(2))t
and
vv′ − v′v − κV (v, v′)t − κA(v, v′)t2
= vv′ − v′v − µ1(v, v′)t + µ1(v′, v)t − µ2(v, v′)t2 + µ2(v′, v)t2
lie in ker(f ). We use here that a ∗ v = av + µ1(a ⊗ v)t and v ∗ v′ = vv′ + µ1(v ⊗
v′)t + µ2(v ⊗ v′)t2, since deg µj = −j for all j > 0.
This analysis implies that Hλ,κ,t ։ Bt as Z>0-graded k-algebras, where Hλ,κ,t
is the quotient of Ft by the relations
av −X a(1)(v)a(2) − λ(a, v)t,
vv′ − v′v − κV (v, v′)t − κA(v, v′)t2.
Now using that A, V are finite-dimensional, verify that the graded components of
the two algebras satisfy: deg Hλ,κ,t[m] 6 deg Bt[m]. Hence the dimensions agree
150
APOORVA KHARE
for each m, whence Hλ,κ,t ∼= Bt. It follows that Hλ,κ = Hλ,κ,tt=1 ∼= Btt=1 as
filtered algebras. Now as explained at the end of the proof of [37, Theorem 6.11],
Hλ,κ has the PBW property.
(cid:3)
4. The case of bialgebras and Hopf algebras
In this section we study a special case of the general framework above, but now
requiring that A is a cocommutative bialgebra (with counit ε), or Hopf algebra
(with counit ε and antipode S). This is indeed the case in a large number of
prominent and well-studied examples in the literature, as discussed after Definition
2.3.
We begin by observing that the cocommutative algebra structure on A auto-
matically extends to H0,0 = Sym(V ) ⋊ A, setting ∆(v) = v ⊗ 1 + 1 ⊗ v for all v ∈ V .
Akin to the usual Hopf-theoretic setting, we now introduce the following notation.
Definition 4.1. Given a cocommutative algebra (A, ∆), an element a ∈ A is
said to be primitive (respectively, grouplike), if ∆(a) = 1 ⊗ a + a ⊗ 1 (respectively,
∆(a) = a ⊗ a).
We now observe that it is possible to classify when the deformed algebra Hλ,κ
is a cocommutative algebra, a bialgebra, or a Hopf algebra, under the assumption
that A has the same structure and V is primitive.
Proposition 4.2. (A, ∆) and V as above. Fix λ : A ⊗ V → A and κ =
κA ⊕ κV : V ∧ V → A ⊕ V as above.
(1) Then Hλ,κ is a cocommutative algebra with (the image of ) V primitive,
if
(4.1)
∆(λ(a, v)) =X λ(a(1), v) ⊗ a(2) +X a(1) ⊗ λ(a(2), v),
for all v, v′ ∈ V, a ∈ A. The converse is true if Hλ,κ has the PBW
property.
κA(v, v′) is primitive,
(2) Suppose A is a cocommutative bialgebra (with counit ε). Then Hλ,κ is a
cocommutative bialgebra with V primitive, if (4.1) holds and im λ ⊂ ker ε.
The converse is true if Hλ,κ has the PBW property.
(3) Suppose A is a cocommutative Hopf algebra (with counit ε and antipode
S). Then Hλ,κ is a cocommutative Hopf algebra with V primitive, if (4.1)
holds and moreover,
im λ ⊂ ker ε,
The converse is true if Hλ,κ has the PBW property.
S(λ(a, v)) =X λ(S(a(1)), a(2)(v)).
In particular, notice that in all three cases, the structure on A automatically extends
to H0,0 = Sym(V )⋊A, and more generally, to all H0,κ for which im κA is primitive.
Proof. To prove the first part, suppose Hλ,κ has the PBW property. If V is
primitive, then we compute in the algebra Hλ,κ ⊗ Hλ,κ:
∆(λ(a, v)) = ∆(av) −X ∆(a(1)(v)a(2))
= ∆(a)∆(v) −X ∆(a(1)(v))∆(a(2))
= X λ(a(1), v) ⊗ a(2) +X a(1) ⊗ λ(a(2), v),
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
151
and similarly,
∆(κA(v, v′)) − (1 ⊗ κA(v, v′) + κA(v, v′) ⊗ 1)
= ∆(κA(v, v′)) + ∆(κV (v, v′)) − (1 ⊗ κ(v, v′) + κ(v, v′) ⊗ 1)
= ∆([v, v′]) − (1 ⊗ [v, v′] + [v, v′] ⊗ 1) = 0.
Since Hλ,κ has the PBW property, the above equalities in fact hold inside V ⊗ A
and A ⊗ A, which inject into Hλ,κ ⊗ Hλ,κ by Theorem 2.5. To prove the converse,
even when Hλ,κ need not have the PBW property, one uses essentially the same
computations as above (but slightly rearranged).
This proves the first part. For the second part, that ε(im κA) = 0 follows
from its primitivity, and that ε(im λ) = 0 follows from applying ε to the defining
relations. The third part now follows from the following computation (using that
SV = − idV as V is primitive):
S(λ(a, v)) = S(a)S(v) −X S(a(1)(v))S(a(2))
= (−v)S(a) +X S(a(2))(a(1)(v))S(a(3)) +X λ(S(a(2)), a(1)(v)),
and now applying the cocommutativity of A, to cancel the first two expressions. (cid:3)
4.1. Symplectic reflections in bialgebras. Our next goal is to show that
the notion of "symplectic reflections" generalizes to arbitrary cocommutative bialge-
bras. The following result extends to such a setting, its group-theoretic counterparts
in [12, 14].
Proposition 4.3. Suppose k is a field, and (A, ∆, ε) is a cocommutative k-
bialgebra. Suppose κV = 0 and Hλ,κ has the PBW property. Given 0 6= a′ ∈ A,
suppose there exists nonzero a′′ ∈ A and a vector space complement U to ka′′ in A
such that
∆(im κA) ⊂ k(a′ ⊗ a′′) ⊕ (A ⊗ U ),
but ∆(im κA) * A ⊗ U . Then a′ − ε(a′) ∈ Endk V has image with dimension at
most 2.
In other words, if κA is supported on a′ ⊗ a′′, then a′ − ε(a′) is akin to a symplectic
reflection [14]. For instance, for symplectic reflection algebras as in [12, 14], with
A = kW a group ring, if a′ = g ∈ W , then choose U :=Pg′6=g kg′.
Proof. We may assume throughout that a′
6= ε(a′). By choice of a′, there
exist x, y ∈ V such that ∆(κA(x, y)) − r(a′ ⊗ a′′) ∈ A ⊗ U , for some r ∈ k×. We
now claim that for all v ∈ V ,
(a′ − ε(a′))(v) ∈ kvx + kvy, where
vx := (a′ − ε(a′))(x), vy := (a′ − ε(a′))(y).
To show the claim, consider the Jacobi identity (2.7) for v1 = x, v2 = y, v3 = v,
which yields:
X(cid:8) (cid:16)κA(v1, v2)(1) − ε(κA(v1, v2)(1))(cid:17) (v3)κA(v1, v2)(2) = 0.
Denote the summand by f (x, y, v). Now split the term κA(x, y) (and the other
two cyclically permuted such terms) into their a′ ⊗ a′′-components and A ⊗ U -
components. Hence there exist rxy = r, ryv, rvx ∈ k such that by the PBW property,
r(a′ − ε(a′))(v) ⊗ a′′ + ryv(a′ − ε(a′))(x) ⊗ a′′ + rvx(a′ − ε(a′))(y) ⊗ a′′ ∈ V ⊗ U.
152
APOORVA KHARE
This shows that the left-hand side vanishes. The claim now follows by the PBW
property.
(cid:3)
4.2. Yetter -- Drinfeld condition. In the remainder of this section, we work
with Hopf algebras. Assume throughout this subsection that A is a k-free co-
commutative k-Hopf algebra, and V is a k-free A-module. In this case it is easy
to verify that the A-action on T V (respectively, Sym(V )) agrees with the adjoint
action of A: ad a(x) :=P a(1)xS(a(2)) = a(x), for x ∈ T V (respectively, Sym(V )).
Our goal is to show that one of the conditions in Theorem 2.5 required for the
PBW property to hold is equivalent to a compatibility condition called the Yetter --
Drinfeld condition (see e.g. [1, Theorem 3.3]). To state the result, we require some
preliminaries.
Proposition 4.4. Suppose a k-Hopf algebra A acts on a free k-module V , and
a k-algebra B contains A, V .
(1) Then the following relations in B are equivalent for all v ∈ V :
If A is cocommutative, then both of these are also equivalent to:
(a) P a(1)vS(a(2)) = a(v) for all a ∈ A.
(b) av =P a(1)(v)a(2) for all a ∈ A.
(c) va =P a(1)S(a(2))(v) for all a ∈ A.
Now suppose in the remaining parts that the conditions (a),(b) hold.
(2) Suppose A is cocommutative. Then τ : A ⊗ V → V ⊗ A, given by a ⊗ v 7→
P a(1)(v) ⊗ a(2), as well as τ op : V ⊗ A → A ⊗ V , given by v ⊗ a 7→
P a(1) ⊗ S(a(2))(v), are A-module isomorphisms that are inverse to one
(3) Any unital subalgebra M of B that is also an A-submodule (via ad), is an
another.
A-(Hopf ) module algebra under the action
a(m) := ad a(m) =X a(1)mS(a(2)) ∀a ∈ A, m ∈ M.
The proof of the following result is standard and is hence omitted. The result may
be applied to B = Hλ,κ. Note as in [36, §4] that the map τ is an isomorphism of
the Yetter -- Drinfeld modules A ⊗ V and V ⊗ A, called the "braiding".
The following preliminary result can (essentially) be found in [23, Lemma 1.3.3].
To state the result, recall that given a module M over a Hopf k-algebra A, the ε-
weight space Mε is {m ∈ M : a · m = ε(a)m ∀a ∈ A}.
Lemma 4.5. Given a Hopf algebra A and a k-algebra map ϕ : A → B, the
centralizer of ϕ(A) in B is the weight space Bε (where B is an A-module via:
a · b :=P ϕ(a(1))bϕ(S(a(2)))).
Consequently, the deformation H0,κ is commutative if and only if A = Aε under the
adjoint action (equivalently, A is commutative), V = Vε (under the given A-action),
and κ ≡ 0.
We now discuss the Yetter -- Drinfeld condition in detail. In the following re-
sult, τ op : M ⊗ A → A ⊗ M is defined as in Proposition 4.4(2), and Aad, Amult
refer to different A-module structures on A (via the adjoint action, and via left
multiplication respectively).
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
153
Proposition 4.6. Suppose A is a Hopf k-algebra, V, M are k-free A-modules,
and κ ∈ Homk(V ∧ V, M ). Suppose (B, µB, 1B) is an (associative) k-algebra con-
taining A, M , with the additional relations m · a = µB(τ op(m ⊗ a)) in B. The
following are equivalent in B:
(1) κ : V ∧ V → M is A-equivariant, or an A-module map:
a(κ(v, v′)) =X κ(a(1)(v), a(2)(v′)) ∀a ∈ A, v, v′ ∈ V.
(2) κ satisfies the Yetter -- Drinfeld (compatibility) condition, i.e.
τ op(cid:16)X κ(a(1)(v), v′)a(2)(cid:17) =X a(1)κ(v, S(a(2))(v′)) ∀a ∈ A, v, v′ ∈ V.
(3) κ is A-compatible: aκ(v, v′) =X κ(a(1)(v), a(2)(v′))a(3) ∀a, v, v′.
(4) κ satisfies: κ(v, v′)a =X a(1)κ(S(a(2))(v), S(a(3))(v′)) ∀a, v, v′.
If κ also satisfies: κ(a(v), v′) = κ(v, S(a)(v′)) for all v, v′, a, then these
are also equivalent to:
(5) im κ commutes (in B) with all of A.
The proof is a relatively straightforward exercise in computations involving Hopf
algebras, and is hence omitted. We remark that the proof uses Proposition 4.4,
Lemma 4.5 and that A is cocommutative.
To conclude this section, we point out how the Yetter -- Drinfeld condition arises,
as in [1, Theorem 3.3]: in the associative algebra B above, compute v′ · a · v in two
different ways (i.e. using the maps τ, τ op, κ). Then,
X a(1)κ(v, S(a(2))(v′)) =X a(1)(v)a(2)S(a(3))(v′) − v′av =X κ(a(1)(v), v′)a(2),
and this is precisely the Yetter -- Drinfeld condition.
5. Generalized nil-Coxeter algebras and grouplike algebras
In the remainder of this paper, we introduce a class of cocommutative algebras
that incorporates group algebras as well as nil-Coxeter algebras and their general-
izations, which are necessarily not bialgebras or Hopf algebras. We then study the
Jacobi identity (2.7) in detail; this is useful in classifying PBW deformations over
nil-Coxeter algebras.
We begin by setting notation concerning unitary/complex reflection groups.
Definition 5.1. A Coxeter matrix is a symmetric matrix A := (aij )i,j∈I in-
dexed by a finite set I and with integer entries, such that aii = 1 and 2 6 aij 6 ∞
for all i 6= j. Given a Coxeter matrix A, define the corresponding braid group
BW = BW (A) to be the group generated by simple reflections {si : i ∈ I}, satisfying
the braid relations sisjsi · · · = sjsisj · · · for all i 6= j, with precisely aij factors on
either side. Finally, define the Coxeter group W = W (A) to be the quotient of the
braid group by the additional relations s2
i = 1 ∀i. More broadly, given an integer
tuple d with di > 2 ∀i ∈ I, define the corresponding generalized Coxeter group
W (d) to be the quotient of BW (A) by sdi
i = 1 ∀i.
We now introduce the corresponding families of generalized (nil-)Coxeter groups
and algebras. This involves considering the "non-negative part" of the braid group,
i.e., the Artin monoid.
154
APOORVA KHARE
Definition 5.2. Given a Coxeter matrix A, first define the Artin monoid B>0
WA
to be the monoid generated by {Ti : i ∈ I} modulo the braid relations. Now given
an integer vector d = (di)i∈I with each di > 2, define the generalized nil-Coxeter
algebra N CWA (d) as:
(5.1)
N CWA(d) :=
(TiTjTi · · ·
khTi, i ∈ Ii
, T di
= TjTiTj · · ·
i = 0, ∀i 6= j ∈ I)
=
kB>0
WA
i = 0 ∀i)
(T di
.
aij times
aij times
{z
}
{z
}
Remark 5.3. The algebras N CW (d) provide a large family of examples of
cocommutative algebras via ∆(Ti) := Ti ⊗ Ti for all i ∈ I (and extending ∆ by
multiplicativity). Moreover, no algebra N CW (d) can be a (weak) bialgebra under
this coproduct. This is because any counit ε necessarily maps the nilpotent element
Ti to 0; but Ti is grouplike so ε(Ti) = 1.
Generalized nil-Coxeter algebras N CW (d) include the well-studied case (see
the Introduction) of the nil-Coxeter algebra N CW , where di = 2 ∀i. Note that
dim N CW (d) > N CW , as N CW (d) surjects onto N CW . Moreover, if W is fi-
nite, then dim N CW ((2, . . . , 2)) = W < ∞; see e.g.
[22, Chapter 7]. No-
tice that there are other finite-dimensional algebras of the form N CW (d). For
instance, N CA1 (d) ∼= k[T1]/(T d
1 ) is finite-dimensional; hence, so is the algebra
N CAn
1 ((d1, . . . , dn)) with all di > 2. This question is completely resolved in related
work [25], where we characterize the generalized nil-Coxeter algebras N CW (d)
that are finite-dimensional. We show that apart from the usual nil-Coxeter al-
gebras N CW ((2, . . . , 2)), there is precisely one other family of type-A algebras,
N CA((2, . . . , 2, d)) with d > 2, which are finite-dimensional. See [25, Theorems
A,C] for further details.
5.1. Grouplike algebras. We begin by unifying the group algebras kW and
the algebras N CW (d) (as well as other algebras considered in the literature) in the
following way.
Definition 5.4. A grouplike algebra is a unital k-algebra A, together with a
distinguished k-basis {Tm : m ∈ MA} containing the unit 1A, such that the map
∆ : A ⊗ A, Tm 7→ Tm ⊗ Tm is an algebra map.
Remark 5.5. Observe from the definitions that the grouplike elements g :=
cmTm in a grouplike algebra A can all be easily identified. Indeed, if g 6= 0
Pm∈MA
and k is a domain, then
Xm,m′∈MA
cmcm′Tm ⊗ Tm′ = ∆(g) = Xm∈MA
cmTm ⊗ Tm,
from which it follows that the sum is a singleton, with coefficient 1. Thus g = Tm
for some m. As a consequence, it follows that the set {Tm : m ∈ MA}⊔{0} is closed
under multiplication, making it a monoid with both a unit and a zero element. This
is formalized presently.
Notice that every grouplike algebra is a cocommutative algebra with coproduct.
(Henceforth we will suppress the monoid operation ∗ when it is clear from context.)
As we presently show, generalized Coxeter groups and generalized nil-Coxeter alge-
bras are examples of grouplike algebras. First we introduce the following notation.
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
155
Definition 5.6. We work over a unital commutative ring k.
(1) Given a monoid (M, ∗), its monoid algebra, denoted by kM and analogous
to the notion of a group algebra, is a k-algebra that has k-basis M , with
the multiplication in M extended by linearity to all of kM .
(2) A zero/absorbing/annihilating element in a monoid M is an element 0M ∈
M such that 0M ∗ m = m ∗ 0M = 0M for all m ∈ M . Such an element is
necessarily unique in M (and idempotent).
We now present several examples of (cocommutative) grouplike algebras.
(1) Every monoid algebra kM is a grouplike algebra, using Tm := m for all
m. This includes the group algebra of every (generalized) Coxeter group.
(2) Suppose M contains a zero element 0M . Then k0M is a two-sided ideal in
the monoid algebra kM , and so kM/k0M is also a grouplike algebra with
basis {Tm : m ∈ M \ 0M }. The previous example is a special case, since
to each monoid M we can formally attach a zero element 0, to create a
new monoid with zero element 0.
(3) Another special case of the preceding example is a nil-Coxeter algebra
N CW . This corresponds to the monoid W ⊔ {0W }, with Tw ∗ Tw′ := 0W
if ℓ(ww′) > ℓ(w) + ℓ(w′) in W . More generally, define for k ∈ N the ideal
Ik to be the k-span of {Tw : ℓ(w) > k}. Then N CW /Ik is a grouplike
algebra, with distinguished basis {Tw : ℓ(w) < k}.
(4) The generalized nil-Coxeter algebra N CAn
1 ((d1, . . . , dn)), with di > 2 for
all i, is yet another example of the above construction. In this case we
use the monoid
M := {0} ⊔ ×i{1, . . . , di − 1},
i)i if maxi(ei +e′
(5) As a final example, recall the 0-Hecke algebra
i)i equal to (ei +e′
with (ei)i ∗(e′
i −di) < 0, and 0 otherwise.
(5.2)
HW (0) :=
kB>0
W
(T 2
i = Ti ∀i ∈ I)
,
where B>0
W is as in Definition 5.2. This algebra was defined in [33] and has
been extensively studied since; see [16, 20, 39] and the references therein.
We recall from [21] that HW (0) is the monoid algebra of a monoid in
bijection with W . As we presently show, it is also a grouplike algebra
with distinguished basis {Tw : w ∈ W }.
Given the profusion of Coxeter-theoretic examples above, it is desirable to
consider a subclass of grouplike algebras that incorporates them all in a systematic
manner. We now present such a family.
Definition 5.7. Given a Coxeter matrix A and an integer vector d with 2 6
di 6 ∞ ∀i, a generic Hecke algebra is any algebra of the form
(5.3)
EW (d, p) :=
kB>0
W
(T di
i = pi(Ti) ∀i ∈ I)
,
where W = WA, and pi ∈ k[T ] has degree at most di − 1 for i ∈ I.
These algebras are so named after the family of "generic Hecke algebras" stud-
ied in [7, 8]; however, unlike loc. cit., we do not require the pi to be equal when
156
APOORVA KHARE
the corresponding simple reflections are conjugate in W . Note that all generalized
(nil-)Coxeter groups and algebras as in Definition 5.2 are covered by our definition.
Recall that our goal in the present paper is to study cocommutative algebras.
Thus, we now study when generic Hecke algebras provide examples of such algebras.
Proposition 5.8. Suppose k is a domain, W = WA is a Coxeter group, and
d, p are as in Equation (5.3).
(1) The map ∆ : Ti 7→ Ti ⊗ Ti extends to make EW (d, p) a (cocommutative)
grouplike algebra, if for all i ∈ I, pi(T ) is either zero or equals T ei for
some 0 6 ei < di.
(2) EW (d, p) is a bialgebra if for all i ∈ I, pi(T ) = T ei for some 0 6 ei < di.
(3) EW (d, p) is a Hopf algebra if pi(T ) = 1 ∀i ∈ I.
The converse statements are all true if for all i, the vectors 1, Ti, . . . , T di−1
k-linearly independent in EW (d, p).
i
are
Notice that the last condition is not always true. For instance, standard argu-
ments as in [29, Introduction] show that the condition fails to hold in a generalized
Coxeter group W (or kW to be precise) whenever aij is odd, pi = 1 is constant
for all i, and di 6= dj. However, the condition does hold in group algebras, 0-Hecke
algebras, and nil-Coxeter algebras corresponding to Coxeter groups.
Proof. We begin by showing the first three assertions. Suppose for all i that
pi(T ) = 0 or T ei for some 0 6 ei < di. Then it is easily verified that ∆ : Ti 7→ Ti⊗Ti
extends to the tensor algebra over the Ti, hence to the Artin monoid kB>0
W , and
hence to EW (d, p). Similarly one verifies that a counit that sends Ti to 1 for all
i, can be extended to EW (d, p) if pi(T ) = T ei for all i. Finally, an antipode that
sends Ti to T −1
can be extended to EW (d, p).
i = T di−1
The "converse" statements are slightly harder to show. Suppose 1, Ti, . . . , T di−1
are k-linearly independent in EW (d, p). To show (the converse of) (1), notice that
every algebra of the form EW (d, p) is a quotient of kB>0
W , so it suffices to classify
the polynomials pi such that the ideal generated by all T di
i − pi(Ti) is a coideal.
i
i
Define pi(T ) :=Pdi−1
(5.4)
∆(T di
i ) = T di
i ⊗ T di
i =
j=0 pijT j, and compute using the multiplicativity of ∆:
pij pikT j
i ⊗ T k
i ,
di−1
Xj,k=0
∆(pijT j
i ) =
di−1
Xj=0
pijT j
i ⊗ T j
i .
di−1
Xj=0
It follows by the assumptions that each nonzero pi(T ) is a monomial pijT j,
with p2
ij = pij in the domain k. This proves (1). To show (2), it suffices to produce
a counit ε that is compatible with the coproduct. Since Ti is grouplike, it follows
that ε(Ti) must equal 1 for all i. This is indeed compatible with the relations
i = T ei
T di
i = 0
implies ε(Ti) = 0, a contradiction.
, which shows one implication. On the other hand, the relation T di
i
Finally, we show (3). If pi(T ) = 1 for all i then EW (d, p) is a group algebra,
hence a Hopf algebra. Conversely, suppose pi(T ) = T ei for some 0 < ei < di
and i ∈ I. Then from above, the subalgebra generated by Ti is isomorphic to
k[T ]/(T di − T ei), which surjects onto the algebra k[T ]/(T 2 − T ). This is precisely
the 0-Hecke algebra of type A1, in which one knows that T is not invertible, yet T
is grouplike. Thus Ti is not invertible in EW (d, p).
(cid:3)
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
157
Remark 5.9. Let A := EW (d, p). If pi(T ) = 0 ∀i, and M := spank{Ti : i ∈ I},
then AM = M A = AM A =: m is a maximal ideal of A. This is because m is a
quotient of the tensor algebra TkM , by relations that strictly lie in the augmentation
ideal T +
k M .
5.2. The Jacobi identity for grouplike algebras. Having defined group-
like algebras and presented examples of them, we specialize the conditions in the
PBW Theorem 2.5 to such a setting. For instance, if λ, κV are identically zero, and
A is a group algebra kG as in [12, 14], then defining κA(v, v′) :=Pg∈G κg(v, v′)Tg,
we see easily that the A-compatibility of κA is equivalent to the following condition
found in loc. cit.:
κghg−1 (Tg(v), Tg(v′)) = κg(v, v′),
∀g, h ∈ G, v, v′ ∈ V.
Our goal in the remainder of this section is to study the Jacobi identity (2.7)
in the case κV ≡ 0, over a grouplike algebra A.
Standing Assumption 5.5. For the remainder of this section, k is a field and
κV ≡ 0.
We begin by setting notation. Define the fixed point space of a ∈ A and its
codimension:
(5.6)
Fix(a) := {v ∈ V : a(v) = v},
da := codimV Fix(a).
Thus, da = dimk im(idV −a).
Now suppose we have fixed a k-basis {aj : j ∈ J1} of A. Then we will write
(5.7)
κ(x, y) = κA(x, y) =: Xj∈J1
κj(x, y)aj ,
∀x, y ∈ V.
Thus, κj is a skew-symmetric bilinear form on V . We also define Rad(κj) to be
the radical of the bilinear form, Rad(κj) := {v ∈ V : κj(v, V ) ≡ 0}. Specifically,
this notation will be applied to a grouplike algebra A with a distinguished basis
{Tm : m ∈ MA} of grouplike elements; see Remark 5.5. In this setting, we will
write κTm = κm and dTm = dm.
We now characterize the Jacobi identity in this general setting.
Theorem 5.10. Suppose κV ≡ 0.
(1) Suppose A contains a grouplike element Tm and a vector space complement
V0 to kTm, such that ∆(V0) ⊂ V0 ⊗ V0. Extend Tm to any basis of V0.
Now if the Jacobi identity (2.7) holds in Hλ,κ (with κV ≡ 0), then one of
the following conditions holds:
(a) κm ≡ 0.
(b) Tm ≡ idV , i.e. dm = 0.
(c) dm is 1 or 2, and Rad(κm) is a subspace of Fix(Tm), of codimension
2 − dm.
(2) Conversely, if A is a grouplike algebra with distinguished k-basis {Tm :
m ∈ MA} of grouplike elements, and for each m ∈ MA one of the above
three conditions holds, then the Jacobi identity (2.7) holds in Hλ,κ (with
κV ≡ 0).
For completeness, we remark that part (1) extends to arbitrary grouplike alge-
bras a result found in [12, 14] for A a group algebra; see also [18, 37].
158
APOORVA KHARE
Proof. Write out the Jacobi identity (2.7) using the distinguished k-basis of
A, and isolate the Tm-component to get:
X(cid:8)
v1κm(v2, v3) =X(cid:8)
or equivalently, for all x, y, z ∈ V ,
κm(v2, v3)Tm(v1),
(5.8)
κm(y, x)(idV −Tm)(z) = κm(y, z)(idV −Tm)(x) + κm(z, x)(idV −Tm)(y).
Before proving the two parts, we make two observations. First, it follows from
(5.8) that κm ≡ 0 or Rad(κm) ⊂ Fix(Tm). Moreover, if Rad(κm) ⊂ Fix(Tm) has
codimension at most 1, then by the skew-symmetry of κm it is clear that Fix(Tm)
is κm-isotropic.
(1) Suppose the Jacobi identity holds. Assume κm is not identically zero;
thus, choose x, y so that κm(y, x) 6= 0. Then Equation (5.8) implies
that im(idV −Tm) ⊂ kx′ + ky′, where x′
:=
(idV −Tm)(y). (This is similar to the proof of Proposition 4.3.) In partic-
ular, dm = dimk im(idV −Tm) 6 2 if κm 6≡ 0.
:= (idV −Tm)(x) and y′
If dm = 0 then assertion (b) holds, so we may assume now that dm
is 1 or 2. Also notice by Equation (5.8) that Rad(κm) ⊂ Fix(Tm), so it
remains to show that the codimension is 2 − dm.
First suppose dm = 2, whence x′, y′ are linearly independent. We
Indeed, suppose z ∈ Fix(Tm). Then
claim that Rad(κm) ⊃ Fix(Tm).
Equation (5.8) yields:
(5.9)
κm(y, z)x′ + κm(z, x)y′ = 0.
Similarly, replacing x by z′ ∈ ker(idV −Tm) yields: κm(z, z′)y′ = 0. From
this and (5.9), it follows that κm(z, −) kills x, y as well as ker(idV −Tm) =
Fix(Tm). Hence it kills their k-span, which is all of V .
0, v0) 6= 0 for some v0, v′
The final case is when dm = 1. Fix v1 6∈ Fix(Tm); thus V = kv1 ⊕
Fix(Tm). We may assume v1 6∈ Rad(κm). Indeed, if instead κm(v1, V ) =
0, then κm(v′
0 ∈ Fix(Tm), since κm 6≡ 0. Then
κm(v1+v′
0. Proceeding, notice that
κm(v1, v0) 6= 0 for some v0 ∈ Fix(Tm). Now define V0 := {v ∈ Fix(Tm) :
κm(v1, v) = 0}; then Fix(Tm) = kv0 ⊕ V0, and V0 ⊃ Rad(κm) from the
observations following (5.8). Finally, applying (5.8) to z, y ∈ Fix(Tm), x =
v1 shows that Fix(Tm) is κm-isotropic. Hence V0 = Rad(κm).
0, v0) 6= 0, so we can replace v1 by v1+v′
(2) Conversely, suppose A is grouplike with basis {Tm : m ∈ MA} as given.
We are to show that Equation (5.8) holds for all m ∈ MA. Certainly this
holds if κm ≡ 0 or Tm ≡ idV . Thus we assume henceforth that κm 6≡ 0,
and show Equation (5.8) for a fixed m ∈ MA, in the two cases dm = 1, 2.
First suppose dm = 2, and x, y ∈ V are linearly independent modulo
Rad(κm). Notice that κm(v, v′) is nonzero only if v, v′ are independent
modulo Rad(κm), so it suffices to prove (5.8) with x, y as above, whence
z = αx+βy+v for some α, β ∈ k and v ∈ Rad(κm) = Fix(Tm). In this case
it is easily shown that both sides of (5.8) equal κm(y, x) · (idV −Tm)(αx +
βy).
Finally, suppose dm = 1, with V ⊃ Fix(Tm) ⊃ Rad(κm) a chain of
codimension one subspaces. Choose x ∈ V \ Fix(Tm) and y ∈ Fix(Tm) \
Rad(κm); once again, if κm(v, v′) is nonzero we may replace v, v′ by x, y,
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
159
and set z = αx + βy + v for v ∈ Rad(κm). Now both sides of (5.8) are
equal to κm(y, x) · (idV −Tm)(αx).
(cid:3)
Theorem 5.10 is useful in characterizing PBW deformations, via the following
consequence.
Corollary 5.11. Suppose A contains a grouplike and nilpotent element Tm,
and a vector space complement V0 to kTm such that ∆(V0) ⊂ V0 ⊗ V0. If the Jacobi
identity (2.7) holds in Hλ,κ with κV ≡ 0, then either κm ≡ 0 or dimk V = 2.
Proof. Since idV −Tm is invertible, Theorem 5.10(1) implies that either κm ≡
(cid:3)
0, or dm = dimk V and Rad(κm) = Fix(Tm) = 0, whence dm = 2.
We conclude this section by specializing to the case of a generalized nil-Coxeter
algebra A = N CW (d). Recall from Remark 2.7 that the condition dimk V = 2 is
sufficient for the Jacobi identities (2.6), (2.7) to hold for Hλ,κ. The following result
shows that over A = N CW (d) and under the original setting of λ, κV ≡ 0 considered
in [12, 14], either κA is highly constrained, or else the condition dimk V = 2 is also
necessary.
Theorem 5.12. Suppose A = N CW (d) is such that the maximal ideal m gen-
erated by {Ti : i ∈ I} is nilpotent. Given an A-module M , define Prim(M ) := {m ∈
M : Tim = 0 ∀i}.
(1) If dimk V 6 2, then H0,κ has the PBW property if and only if im κV ⊂
Prim(V ) and im κA ⊂ Prim(Amult).
(2) If dimk V > 2, and λ, κV ≡ 0, then H0,κA has the PBW property if and
only if κA ≡ 0.
Thus (using Remark 2.7), if H0,κA satisfies the PBW property for A = N CW (d)
finite-dimensional, then either κA ≡ 0 or dimk V = 2.
We also provide examples of Prim(·) for generalized nil-Coxeter algebras. In-
deed, Prim(Amult) equals kTw◦ if A = N CW is the usual nil-Coxeter algebra over a
1 ((d1, . . . , dn)),
finite Coxeter group W with unique longest element w◦. If A = N CAn
. In both of these cases, the maximal ideal m is in-
deed nilpotent, and hence A satisfies the hypotheses of the above theorem for these
families of generalized nil-Coxeter algebras.
then Prim(Amult) = Qi T di−1
i
Proof. Suppose mn = 0 6= mn−1 for some n ∈ N. Before proving the result,
we consider the following filtration on an A-module V :
(5.10)
V ⊃ mV ⊃ m2V ⊃ · · · ⊃ mnV = 0.
We fix k 6 n − 1 such that mkV = 0 6= mk−1V .
(1) By Remark 2.7, and given that λ ≡ 0, it suffices to characterize the
A-compatibilities (2.4), (2.5), assuming further that dim V = 2. Now
observe that mk−1V ⊂ Prim(V ). Choose v0 ∈ mk−1V , and v1 6∈ kv0;
thus V = kv0 ⊕ kv1. Now notice that κV ∧V is completely determined by
κ(v0, v1), since dim V = 2. Thus, we compute using the A-compatibility
(2.4), for any non-trivial grouplike element 1 6= Tm ∈ N CW (d):
TmκA(v0, v1) = κA(Tm(v0), Tm(v1))Tm = 0.
160
APOORVA KHARE
This equation holds for all non-unital Tm, if and only if κm ≡ 0 for
Tm 6∈ Prim(Amult). Similarly, Equation (2.5) reduces to:
Tm(κV (v0, v1)) = κV (Tm(v0), Tm(v1))Tm = 0,
which holds if and only if κV (v0, v1) ∈ Prim(V ), as claimed.
(2) By Corollary 5.11, we see that κA ≡ κ1, since each non-unital grouplike
element Tm is nilpotent by assumption. Now as above, Equation (2.4)
reduces to:
TmκA(x, y) = κA(Tm(x), Tm(y))Tm,
∀m ∈ MA,
so it follows that κA(x, y) = κA(Tm(x), Tm(y)) for all non-unital Tm and
all x, y ∈ V . Repeated applications of this fact show that κA(x, y) =
κA(T k
m(y)) = 0. Conversely, H0,0 = Sym(V ) ⋊ A has the PBW
property.
m(x), T k
(cid:3)
For completeness, we mention two properties of generalized nil-Coxeter alge-
bras, even though they will not be used in the paper. First, the algebras N CW (d),
and more generally, every generic Hecke algebra EW (d, p), is equipped with an
anti-involution that fixes every generator Ti. This is because the defining relations
are preserved by such a map. Such an anti-involution can be used to construct
an exact contravariant duality functor on a suitable category of A-modules, which
preserves the simple object k = A/m.
Second, as discussed in [26], for all finite Coxeter groups W the nil-Coxeter
algebra is a Frobenius algebra, by defining a trace map to kill all words in the Ti
except for the longest word Tw◦ . The same turns out to hold also for the generalized
nil-Coxeter algebra A := N CAn
1 (d), by defining a trace map to kill all words in the
span
i=1 T di−1
i
i=1 T di−1
i
the space Prim(A) = Prim(Aop), as we note after Theorem 6.7 below.
. Note that these two words Tw◦ and Qn
Ti, except for Qn
6. Deformations over cocommutative algebras with nilpotent maximal
ideals
In this final section, we study the representations of deformed smash product
algebras over nil-Coxeter algebras. We will work in somewhat greater generality.
Standing Assumption 6.1. Henceforth, k is a field, and (A, ∆) is a cocom-
mutative k-algebra with coproduct, with a nilpotent maximal ideal m = AmA 6= 0
that satisfies:
A = m ⊕ k · 1A,
∃ℓA ∈ N : mℓA = 0 6= mℓA−1,
∆(m) ⊂ m ⊗ m.
We will use without further reference the following observations, when required:
• (A, m) is local, since every element in A\m is invertible. From this one can
show that m is the Jacobson radical of A, and ExtA−mod(k, k) ∼= (m2)⊥,
where (m2)⊥ ⊂ m∗.
• The assumption ∆(m) ⊂ m ⊗ m is required if char k > 0. Cocommutative
algebras not satisfying this assumption exist; for instance, consider A :=
(Z/pZ)[T ]/(T p), with p > 0 prime and ∆(T ) = 1⊗T +T ⊗1. However, we
do not need to assume ∆(m) ⊂ m ⊗ m if char k = 0. Indeed, given a ∈ m,
let ∆(a) ∈ c(1 ⊗ 1) ⊕ d(1 ⊗ m) ⊕ e(m ⊗ 1) ⊕ (m ⊗ m), with c, d, e ∈ k×. By
multiplicativity, ∆(a)n = 0 for n ≫ 0, which works out to: c = d = e = 0.
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
161
The prototypical example of an algebra satisfying Assumption 6.1 is the nil-
Coxeter algebra N CW for a finite Coxeter group W . Another example is the gen-
i ). In both cases,
eralized nil-Coxeter algebra N CAn
m is the two-sided augmentation ideal generated by the Ti. We remark for com-
pleteness that in related work [25, Theorem C], we characterize the generalized
nil-Coxeter algebras N CW (d) for which the maximal ideal m is nilpotent. This
property turns out to be equivalent to the finite-dimensionality of N CW (d), which
was discussed following Remark 5.3.
1 ((d1, . . . , dn)) = ⊗n
i=1k[Ti]/(T di
6.1. Simple Hλ,κ-modules. We begin by exploring simple modules over
Hλ,κ. In order to state our results, some notation is required.
Definition 6.1. Suppose A is as in Assumption 6.1, and M is an A-module.
(1) The level of a nonzero vector m ∈ M is the integer k > 0 such that
mkm = 0 6= mk−1m. Define the level of 0M to be 0 for convention. The
level of the module, denoted by ℓM , is the highest level attained in M .
(2) For k > 0, define L6k(M ) to be the set of elements of level at most k.
(3) A vector m ∈ M is primitive if mm = 0. Let Prim(M ) denote all primitive
elements.
The following lemma is easily shown.
Lemma 6.2. Suppose M is any A-module. Then L6k(M ) = kerM mk; in
particular,
Prim(M ) = L61(M ),
M = L6ℓM (M ),
ℓM 6 ℓAmult = ℓA.
Moreover, L6k(M ) is a proper submodule of L6k+1(M ) for all k < ℓM .
We now study Hλ,κ-modules. Our first result aims to classify all simple Hλ,κ-
modules in the case when κV ≡ 0.
Theorem 6.3. Suppose A satisfies Assumption 6.1 and V is an A-module. If
λ satisfies Equation (2.3) in A, then λ(mk, L6k(V )) ⊂ mk for all k > 0. If instead
we assume κV ≡ 0, then the following are equivalent for Hλ,κ:
(1) λ(mk, V ) ⊂ mk for all k > 0, and κA : V ∧ V → m.
(2) λ(m, V ) ⊂ m and κA : V ∧ V → m.
(3) There exists a one-dimensional Hλ,κ-module killed by m.
(4) There is a bijection from simple Hλ,κ-modules to simple Sym(V )-modules,
determined uniquely by restriction from Hλ,κ to the image of V ; moreover,
the inverse map is given by restriction to V and inflation to Hλ,κ, letting
m act trivially.
The condition κA : V ∧ V → m is a natural one in characteristic zero, in the sense
that it is necessary if Hλ,κ has a finite-dimensional module and char k = 0. This
is because if π : Hλ,κ → Endk M is a finite-dimensional representation, then for all
a ∈ m, π(a) is nilpotent, hence has trace zero. It follows that im κA = [V, V ] ⊂ m.
The following result will be useful in proving Theorem 6.3.
Proposition 6.4. Suppose M is an A-module.
(1) M is A-semisimple if and only if mM = 0.
162
APOORVA KHARE
(2) Any finite filtration M = M0 ⊃ M1 ⊃ · · · ⊃ Mk = 0 of A-modules
(such as M ⊃ 0) can be refined to a possibly longer finite filtration, so
that the successive subquotients are A-semisimple modules. In particular,
Prim(M ) 6= 0 if M 6= 0.
(3) Every maximal submodule of a nonzero A-module has codimension one.
Thus a d-dimensional A-module has a flag of A-submodules of length d+1.
(4) Prim(A) ⊂ m.
(5) If M is nonzero, mM is contained in every maximal proper (i.e. codimen-
sion one) submodule of M . In particular, it is a proper submodule of M
if M 6= 0.
Proof.
(1) If mM = 0 then M is clearly A-semisimple. Conversely, if M is A-
semisimple, notice that M = mM ⊕ M1 for some A-semisimple com-
plement M1. But then M1 ∼= M/mM is annihilated by m. Repeat this
construction on mM to produce M2, and so on; this process stops after
finitely many steps as m is nilpotent. But then M is a direct sum of
submodules killed by m.
(2) It suffices to prove the result for the filtration M ⊃ 0. Define Mi := miM
for all i > 0, and M0 := M . Now apply the previous part.
(3) This follows from the previous part.
(4) If a ∈ A \ m, then a is invertible, hence cannot lie in Prim(A).
(5) Suppose M = km0 ⊕ M ′ where M ′ is a proper submodule. Fix a ∈ m
such that am0 = rm0 + m′, with r ∈ k and m′ ∈ M ′. Then one shows by
induction on i that
aim0 = rim0 + (ri−1m′ + ri−2am′ + · · · + ai−1m′)
for all i > 0. In particular, since aℓA ∈ mℓA = 0, hence rℓA m0 ∈ M ′,
whence rℓA = 0. Thus r = 0, and am0 = m′ ∈ M ′ for all a ∈ m, whence
mM ⊂ M ′ as claimed.
(cid:3)
Proof of Theorem 6.3. The first assertion holds because the A-action (2.3)
implies that if mk(v) = 0, then (with a slight abuse of notation)
0 = λ(mℓA−kmk, v) = mℓA−kλ(mk, v) + λ(mℓA−k, mk
(1)(v))mk
(2) = mℓA−kλ(mk, v),
from which it follows that λ(mk, v) ⊂ mk.
We now assume κV ≡ 0, and show that (1) and (2) are equivalent. Clearly
(1) =⇒ (2); conversely, if (2) holds, then we compute for a1, . . . , ak ∈ m, by
induction on k:
λ(a1 · · · ak, v) =a1λ(a2 · · · ak, v) +X λ(a1, ((a2)(1) · · · (ak)(1))(v))(a2)(1) · · · (ak)(1)
⊂m · mk−1 + m · mk−1 = mk.
Next, given (2), we show (4) as follows: if M is a simple Sym(V )-module then
the construction in (4) makes it a simple Hλ,κ-module, as the relations in Hλ,κ
indeed hold in Endk M via (2). On the other hand, given any Hλ,κ-module M , by
Proposition 6.4, kerM m 6= 0. We now claim that if λ(m, V ) ⊂ m and M is a Hλ,κ-
module, then kerM mk is a Hλ,κ-submodule of M . Given the claim, if M is now
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
163
a simple Hλ,κ-module, then 0 6= kerM m is a Hλ,κ-submodule, whence mM = 0,
proving (4).
It remains to show the claim (in order to complete the proof of (2) =⇒ (4)).
Let M ′ = kerM mk; then for a ∈ m and m′ ∈ M ′, we have
mk(am′) ⊂ mkA · m′ = mkm′ = 0,
whence am′ ∈ M ′. Thus M ′ is an A-submodule.
vm′ ∈ M ′ for v ∈ V . But if we have a1, . . . , ak ∈ m, then
It thus remains to show that
k
Yi=1
ai · vm′ =X k
Yi=1
(ai)(1)! (v) ·
k
Yi=1
(ai)(2) · m′ + λ(a1 · · · ak, v)m′,
and this is killed by using Assumption 6.1 and the equivalence of (1) and (2). Hence
vm′ ∈ M ′.
Finally, we show (4) =⇒ (3) =⇒ (2).
If (4) holds, choose any linear
functional µ ∈ V ∗ and consider the simple one-dimensional Sym(V )-module
Mµ := Sym(V )/ Sym(V ) · (im(idV −µ)).
By (4), Mµ yields a one-dimensional simple Hλ,κ-module which is killed by m, and
this shows (3). Next, if (3) holds for M then V acts on M by scalars, i.e., by
µ ∈ V ∗. It follows that im κA = [V, V ] kills M , whence κA : V ∧ V → m. Similarly
if a ∈ m, then λ(a, v) ∈ mV − V m also kills M , whence λ(m, V ) ⊂ m.
(cid:3)
Corollary 6.5. Suppose k is algebraically closed and V is finite-dimensional.
If λ(m, V ) ⊂ m, κV ≡ 0, and κA : V ∧ V → m, then all simple finite-dimensional
Hλ,κA-representations are one-dimensional, and in bijection with V ∗.
6.2. PBW property. Our next goal is to prove a result similar to Theorem
5.12 that classifies the PBW deformations Hλ,κ, but in the more general setting of
cocommutative algebras A satisfying Assumption 6.1. Thus we do not assume the
existence of a grouplike basis as for the nil-Coxeter algebra, and alternate methods
are required. In particular, the following provides a second proof of Theorem 5.12.
Theorem 6.6. Suppose A satisfies Assumption 6.1, and V is an A-module.
(1) Suppose κV ≡ 0. Then the Jacobi identity (2.7) holds in Hλ,κA if and
only if dimk V 6 2 or im κA ⊂ k · 1A.
(2) If dimk V 6 2, then H0,κ has the PBW property if and only if im κV ⊂
Prim(V ) and im κA ⊂ Prim(Amult).
(3) If dimk V > 2, and λ, κV ≡ 0, then H0,κA has the PBW property if and
only if κA ≡ 0.
Proof.
(1) By Remark 2.7, and since κV ≡ 0, it suffices to characterize the Jacobi
identity (2.7) under the additional assumption that dim V > 2. Now write
down the identity:
[κ(v1, v2), v3] = 0,
v1, v2, v3 ∈ V.
X(cid:8)
We may assume without loss of generality that the vi are linearly inde-
pendent in V . Moreover, the κ1-component is killed by commuting with
164
APOORVA KHARE
elements of V . (Here, we work with a distinguished k-basis of m, along
with {1A}.) If we now define γv,v′ := κA(v, v′) − κ1(v, v′) ∈ m, then
X(cid:8) (cid:16)v1γv2,v3 −X (γv2,v3)(1)(v1)(γv2 ,v3)(2)(cid:17) = 0.
Now assume without loss of generality that v1 ∈ L6k+1(V ) \ L6k(V ) for
some k > 0, and v1, v2, v3 ∈ L6k+1(V ). Then (γvp,vq )(1)(vr) ∈ L6k(V )
for all {p, q, r} = {1, 2, 3}. Working modulo L6k(V ), it follows by the
linear independence of the vi that γv2,v3 = 0, and hence an entire sum-
mand in the above cyclic sum vanishes. Repeat the same argument twice
to show all summands are zero, and hence, κA ≡ κ1 on V ∧ V .
(2) This is similar to the proof of Theorem 5.12(1) and is omitted for brevity.
(3) Clearly H0,0 has the PBW property. Conversely, assume H0,κA has the
PBW property. By a previous part, we have im κA ⊂ k · 1A. Suppose
κA 6≡ 0. Then there exists k > 0 such that κA(L6k+1(V ), V ) 6≡ 0 =
κA(L6k(V ), V ). Choose nonzero a ∈ m, and any v0 ∈ L6k+1(V ), v1 ∈ V
such that κA(v0, v1) 6= 0. Then by Theorem 2.5,
0 6= aκA(v0, v1) =X κA(a(1)(v0), a(2)(v1))a(3).
But by assumption a(1)(v0) ∈ L6k(V ), whence the right hand side van-
ishes. This contradiction shows that κA ≡ 0.
(cid:3)
6.3. Center and abelianization. We end the paper by computing the center
and abelianization of the algebra Hλ,κ, i.e., the zeroth Hochschild (co)homology.
Theorem 6.7. Suppose A satisfies Assumption 6.1, V, λ, κ are such that Hλ,κ
has the PBW property, and Prim(A) = Prim(Aop). If λ(m, V ) ⊂ m, then Hλ,κ has
trivial center, i.e., HH 0(Hλ,κ, Hλ,κ) = k.
Akin to the remarks following Assumption 6.1, the condition Prim(A) = Prim(Aop)
is satisfied by all nil-Coxeter algebras N CW for a finite Coxeter group W , as well
1 (d). The condition λ(m, V ) ⊂ m was discussed in detail in Theorem
as by N CAn
6.3.
Proof. We first choose a totally ordered basis of V as follows: via Proposition
6.4, fix the filtration 0 = L60(V ) ⊂ L61(v) ⊂ · · · ⊂ L6ℓV (V ) = V according to
the level; then choose any k-basis Bk of the corresponding vector space complement
of L6k−1(V ) in L6k(V ) for k = 1, . . . , ℓV . Now index Bk by any totally ordered set
Thus, every element of B1 is primitive. Now use the PBW property to write any
Sk, and let S :=Fk Sk be totally ordered via: si < sj if i > j and si ∈ Si, sj ∈ Sj.
vector in Hλ,κ asPI vI aI , where I denotes a word in S whose letters occur in non-
increasing order, aI ∈ A, and vI denotes the corresponding monomial in Fk Bk.
Note that m acts on each vI and yields a linear combination of elements vJ such
that I > J in the lexicographic order on words in S. More precisely, if we define
ℓ(vI ) to be the sum of the levels of the letters in the monomial vI (see Definition
6.1), then m strictly reduces ℓ(vI ).
We now proceed to the proof. Suppose 0 6= z = PI vI aI is central in Hλ,κ,
with the vI linearly independent. We first claim that for each non-empty I, the
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
165
vector aI is primitive in A. Indeed, choosing a ∈ m and writing out az = za yields:
XI (cid:16)X a(1)(vI )a(2) + λ(a, vI )(cid:17) aI =X vI aI a.
Choosing I 6= ∅ such that vI has maximal ℓ-value, it follows from above that
aI a = 0 for all a ∈ m. Hence aI ∈ Prim(Aop) = Prim(A) by assumption. Now say
vI = vik · · · vi1 for some ij ∈ I. We notice by induction on k that avI aI = 0 as
well. Indeed,
avI aI =X a(1)(vik ) · a(2)vik−1 · · · vi1 aI + λ(a, vik ) · vik−1 · · · vi1 aI ,
and both expressions vanish by the induction hypothesis (the base case of k = 1
It follows that avI aI = 0 = vI aI a, where I 6= ∅ is such that ℓ(vI ) is
is easy).
maximal. Now cancel these terms from the above equation and work with I of the
next highest ℓ-value. Repeating the above analysis shows the claim.
Next, let v ∈ Prim(V ) and consider zv = vz in Hλ,κ:
a∅v +XI
vI aI v = va∅ +X vvI aI .
Since aI ∈ Prim(A) ⊂ m (by Proposition 6.4), hence aI v = λ(aI , v) for all non-
empty I. Hence working modulo the filtered degree 6 1 piece and using the PBW
property, aI = 0 if I 6= ∅. In other words, z = a∅ ∈ A. Since A = k · 1 ⊕ m, we may
assume that z ∈ m. Now choose nonzero primitive v ∈ V ; then,
vz = zv =X z(1)(v)z(2) + λ(z, v) = λ(z, v),
whence we get that z = 0 by the PBW property. Hence Z(Hλ,κ) = k · 1 as
claimed.
(cid:3)
Next, we compute the zeroth Hochschild homology.
Theorem 6.8. Suppose λ and κV are identically zero, κA : V ∧ V → m,
If k is an infinite field, then as abelian
and H0,κA satisfies the PBW property.
k-algebras, we have
HH0(Hλ,κ, Hλ,κ) =
Hλ,κ
[Hλ,κ, Hλ,κ]
∼= k · 1 +(cid:16)Sym+(V )M (m/([m, m] + A · (im κA) · A))(cid:17) ,
where the direct sum indicates that the two factors are ideals and hence multiply to
zero.
Proof. The proof is in steps. The first step is to show that [Hλ,κ, Hλ,κ]
contains the image of hV i · m, where given a subspace U ⊂ V, hU i := T V · U · T V
is the two-sided ideal in T V generated by U . More precisely, we show by induction
on k that hL6k(V )i · m ⊂ [Hλ,κ, Hλ,κ]. This is clear for k = 0, and given the result
for k, Assumption 6.1 implies that
a(p) ∈ hL6k(V )i,
∀a ∈ m, p ∈ hL6k+1(V )i.
It follows by the induction hypothesis that
p · a = [p, a] + a · p
= [p, a] +X a(1)(p)a(2) ∈ [Hλ,κ, Hλ,κ] + hL6k(V )im ⊂ [Hλ,κ, Hλ,κ].
166
APOORVA KHARE
Next, fix a total ordering on a basis of V . Given any nonzero sum v of monomial
"ordered" words, since k is an infinite field there exists µ ∈ V ∗ such that µ(v) 6= 0.
Now since λ ≡ 0, it follows by Theorem 6.3 that Hλ,κ has a one-dimensional
representation Mµ killed by m, and on which V acts by µ. Since [Hλ,κ, Hλ,κ]
necessarily kills Mµ, it follows that v has nonzero image in Hλ,κ/[Hλ,κ, Hλ,κ].
Hence V generates the symmetric algebra in Hλ,κ/[Hλ,κ, Hλ,κ].
It remains to consider the image of A inside the abelianization. Note that
im κA = [V, V ] and [m, m] lie in [Hλ,κ, Hλ,κ], and are subspaces of m by assumption.
(That this image and Sym+(V ) are ideals follows from the above analysis.) To
complete the proof, it suffices to show the commutator intersects A in [m, m] + A ·
(im κA) · A. Note Hλ,κ = ALhV i · A by the PBW property. Now [A, A] = [m, m],
while [hV i · A, A] ⊂ hV i · A, which intersects A trivially.
It remains to consider [hV i · A, hV i · A] ∩ A. By the relations in Hλ,κ as well
as the PBW property, the only elements that occur here arise from the relations
[v, v′] = κA(v, v′) ∈ A, and hence the intersection is contained in A · (im κA) · A. We
now show that this containment is an equality, via the claim that aκA(v, v′)a′ ∈
[Hλ,κ, Hλ,κ] for v, v′ ∈ V and a, a′ ∈ A. The claim is obvious if a = a′ = 1.
Otherwise we may assume that at least one of a, a′ lies in m. In this case,
[av, v′a′] = avv′a′ − v′a′av = a[v, v′]a′ + av′va′ − v′a′av
= a[v, v′]a′ +X a(1)(v′)a(2)(v)a(3)a′ − v′X (a′a)(1)(v)(a′a)(2).
Since ∆(m) ⊂ m ⊗ m, it follows that all summands of both sums lie in hV i · m, hence
in [Hλ,κ, Hλ,κ] from above. This proves the claim, and with it, the result.
(cid:3)
Acknowledgments
The author would like to thank Sarah Witherspoon for many stimulating and
informative conversations regarding this paper. The author also thanks Ivan Marin,
Susan Montgomery, and Victor Reiner for useful references and discussions. Finally,
the author is grateful to Chelsea Walton for going through a preliminary draft of
this work and for her helpful suggestions.
References
[1] Y. Bazlov and A. Berenstein, Braided doubles and rational Cherednik algebras, Adv. Math.
220 (2009), no. 5, 1466 -- 1530, DOI 10.1016/j.aim.2008.11.004. MR2493618
[2] G. M. Bergman, The diamond lemma for ring theory, Adv. Math. 29 (1978), no. 2, 178 -- 218,
DOI 10.1016/0001-8708(78)90010-5. MR506890
[3] I. N. Bernsteın, I. M. Gel′fand, and S. I. Gel′fand, Schubert cells, and the cohomology of the
spaces G/P , Russian Math. Surveys 28 (1973), no. 2, 1 -- 26. MR0429933
[4] G. Bohm, F. Nill, and K. Szlach´anyi, Weak Hopf algebras. I. Integral theory and C ∗-structure,
J. Algebra 221 (1999), no. 2, 385 -- 438, DOI 10.1006/jabr.1999.7984. MR1726707
[5] A. Braverman and D. Gaitsgory, Poincar´e -- Birkhoff -- Witt theorem for quadratic algebras of
Koszul type, J. Algebra 181 (1996), no. 2, 315 -- 328, DOI 10.1006/jabr.1996.0122. MR1383469
[6] J. Brichard, The center of the nilCoxeter and 0-Hecke algebras, preprint, available at
arXiv:0811.2590 (2008).
[7] M. Brou´e, G. Malle, and R. Rouquier, On complex reflection groups and their associated
braid groups, Representations of groups (Banff, AB, 1994), CMS Conf. Proc., vol. 16, Amer.
Math. Soc., Providence, RI, 1995, pp. 1 -- 13. MR1357192
[8] M. Brou´e, G. Malle, and R. Rouquier, Complex reflection groups, braid groups, Hecke alge-
bras, J. Reine Angew. Math. 500 (1998), 127 -- 190, DOI 10.1515/crll.1998.064. MR1637497
[9] H. S. M. Coxeter, Discrete groups generated by reflections, Ann. of Math. (2) 35 (1934),
no. 3, 588 -- 621, DOI 10.2307/1968753. MR1503182
NIL-COXETER ALGEBRAS, COCOMMUTATIVE ALGEBRAS, PBW PROPERTY
167
[10] H. S. M. Coxeter, Factor groups of the braid group, Proceedings of the 4th Canadian Math-
ematical Congress (Banff, AB, 1957), University of Toronto Press, 1959, pp. 95 -- 122.
[11] W. Crawley-Boevey and M. P. Holland, Noncommutative deformations of Kleinian sin-
gularities, Duke Math. J. 92 (1998), no. 3, 605 -- 635, DOI 10.1215/S0012-7094-98-09218-3.
MR1620538
[12] V. G. Drinfeld, Degenerate affine Hecke algebras and Yangians, Funct. Anal. Appl. 20 (1986),
no. 1, 58 -- 60, DOI 10.1007/BF01077318. MR831053
[13] P. Etingof, W. L. Gan, and V. Ginzburg, Continuous Hecke algebras, Transform. Groups 10
(2005), no. 3-4, 423 -- 447, DOI 10.1007/s00031-005-0404-2. MR2183119
[14] P. Etingof and V. Ginzburg, Symplectic reflection algebras, Calogero-Moser space, and de-
formed Harish-Chandra homomorphism, Invent. Math. 147 (2002), no. 2, 243 -- 348, DOI
10.1007/s002220100171. MR1881922
[15] P. Etingof, D. Nikshych, and V. Ostrik, On fusion categories, Ann. of Math. (2) 162 (2005),
no. 2, 581 -- 642, DOI 10.4007/annals.2005.162.581. MR2183279
[16] M. Fayers, 0-Hecke algebras of finite Coxeter groups, J. Pure Appl. Algebra 199 (2005),
no. 1-3, 27 -- 41, DOI 10.1016/j.jpaa.2004.12.001. MR2134290
[17] S. Fomin and R. P. Stanley, Schubert polynomials and the nil-Coxeter algebra, Adv. Math.
103 (1994), no. 2, 196 -- 207, DOI 10.1006/aima.1994.1009. MR1265793
[18] S. Griffeth, Towards a combinatorial representation theory for the rational Cherednik al-
gebra of type G(r, p, n), Proc. Edinb. Math. Soc. (2) 53 (2010), no. 2, 419 -- 445, DOI
10.1017/S0013091508000904. MR2653242
[19] J.-W. He, F. Van Oystaeyen, and Y. Zhang, PBW deformations of Koszul algebras over a
nonsemisimple ring, Math. Z. 279 (2015), no. 1-2, 185 -- 210, DOI 10.1007/s00209-014-1362-y.
MR3299848
[20] X. He, A subalgebra of 0-Hecke algebra, J. Algebra 322 (2009), no. 11, 4030 -- 4039, DOI
10.1016/j.jalgebra.2009.04.003. MR2556136
[21] F. Hivert, A. Schilling, and N. M. Thi´ery, Hecke group algebras as quotients of affine
Hecke algebras at level 0, J. Combin. Theory Ser. A 116 (2009), no. 4, 844 -- 863, DOI
10.1016/j.jcta.2008.11.010. MR2513638
[22] J. E. Humphreys, Reflection groups and Coxeter groups, Cambridge Studies in Advanced
Mathematics, vol. 29, Cambridge University Press, Cambridge, 1990. MR1066460
[23] A. Joseph, Quantum groups and their primitive ideals, Ergebnisse der Mathematik und ihrer
Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 29, Springer-Verlag,
Berlin, 1995. MR1315966
[24] A. Khare, Category O over a deformation of the symplectic oscillator algebra, J. Pure Appl.
Algebra 195 (2005), no. 2, 131 -- 166, DOI 10.1016/j.jpaa.2004.06.004. MR2108468
[25] A. Khare, Generalized nil-Coxeter algebras over discrete complex reflection groups, preprint,
available at arXiv:1601.08231 (2016).
[26] M. Khovanov, Nilcoxeter algebras categorify the Weyl algebra, Comm. Algebra 29 (2001),
no. 11, 5033 -- 5052, DOI 10.1081/AGB-100106800. MR1856929
[27] M. Khovanov and A. D. Lauda, A diagrammatic approach to categorification of quan-
tum groups. I, Represent. Theory 13 (2009), 309 -- 347, DOI 10.1090/S1088-4165-09-00346-X.
MR2525917
[28] B. Kostant and S. Kumar, The nil Hecke ring and cohomology of G/P for a Kac -- Moody group
G, Adv. Math. 62 (1986), no. 3, 187 -- 237, DOI 10.1016/0001-8708(86)90101-5. MR866159
[29] D. W. Koster, COMPLEX REFLECTION GROUPS, ProQuest LLC, Ann Arbor, MI, 1975.
Thesis (Ph.D.) -- The University of Wisconsin - Madison. MR2625485
[30] A. Lascoux and M.-P. Schutzenberger, Fonctorialit´e des polynomes de Schubert, Invariant
theory (Denton, TX, 1986), Contemp. Math., vol. 88, Amer. Math. Soc., Providence, RI, 1989,
pp. 585 -- 598, DOI 10.1090/conm/088/1000001 (French, with English summary). MR1000001
[31] I. Losev and A. Tsymbaliuk, Infinitesimal Cherednik algebras as W -algebras, Transform.
Groups 19 (2014), no. 2, 495 -- 526, DOI 10.1007/s00031-014-9261-1. MR3200433
[32] G. Lusztig, Affine Hecke algebras and their graded version, J. Amer. Math. Soc. 2 (1989),
no. 3, 599 -- 635, DOI 10.2307/1990945. MR991016
[33] P. N. Norton, 0-Hecke algebras, J. Austral. Math. Soc. Ser. A 27 (1979), no. 3, 337 -- 357, DOI
10.1017/S1446788700012453. MR532754
[34] G. C. Shephard and J. A. Todd, Finite unitary reflection groups, Canadian J. Math. 6 (1954),
274 -- 304, DOI 10.4153/CJM-1954-028-3. MR0059914
168
APOORVA KHARE
[35] A. V. Shepler and S. Witherspoon, Drinfeld orbifold algebras, Pacific J. Math. 259 (2012),
no. 1, 161 -- 193, DOI 10.2140/pjm.2012.259.161. MR2988488
[36] A. V. Shepler and S. Witherspoon, A Poincar´e -- Birkhoff -- Witt theorem for quadratic al-
gebras with group actions, Trans. Amer. Math. Soc. 366 (2014), no. 12, 6483 -- 6506, DOI
10.1090/S0002-9947-2014-06118-7. MR3267016
[37] A. V. Shepler and S. Witherspoon, PBW deformations of skew group algebras in
positive characteristic, Algebr. Represent. Theory 18 (2015), no. 1, 257 -- 280, DOI
10.1007/s10468-014-9492-9. MR3317849
[38] A. V. Shepler and S. Witherspoon, Poincar´e -- Birkhoff -- Witt Theorems, Commutative Alge-
bra and Noncommutative Algebraic Geometry, Volume I: Expository Articles, Mathematical
Sciences Research Institute Proceedings, vol. 67, Cambridge University Press, 2015, pp. 259 --
290.
[39] V. V. Tewari and S. J. van Willigenburg, Modules of the 0-Hecke algebra and quasisym-
metric Schur functions, Adv. Math. 285 (2015), 1025 -- 1065, DOI 10.1016/j.aim.2015.08.012.
MR3406520
[40] A. Tikaradze, On maximal primitive quotients of infinitesimal Cherednik algebras of gln, J.
Algebra 355 (2012), 171 -- 175, DOI 10.1016/j.jalgebra.2012.01.013. MR2889538
[41] A. Tikaradze and A. Khare, Center and representations of infinitesimal Hecke algebras of sl2,
Comm. Algebra 38 (2010), no. 2, 405 -- 439, DOI 10.1080/00927870903448740. MR2598890
[42] A. Tsymbaliuk, Infinitesimal Hecke algebras of soN , J. Pure Appl. Algebra 219 (2015), no. 6,
2046 -- 2061, DOI 10.1016/j.jpaa.2014.07.022. MR3299718
[43] C. Walton and S. Witherspoon, Poincar´e -- Birkhoff -- Witt deformations of smash product al-
gebras from Hopf actions on Koszul algebras, Algebra Number Theory 8 (2014), no. 7, 1701 --
1731, DOI 10.2140/ant.2014.8.1701. MR3272279
[44] C. Walton and S. Witherspoon, PBW deformations of braided products, preprint, available
at arXiv:1601.02274 (2016).
[45] G. Yang, Nil-Coxeter algebras and nil-Ariki-Koike algebras, Front. Math. China 10 (2015),
no. 6, 1473 -- 1481, DOI 10.1007/s11464-015-0498-3. MR3403207
Departments of Mathematics and Statistics, Stanford University, Stanford, CA -
94305
E-mail address: [email protected]
|
1911.00094 | 1 | 1911 | 2019-10-31T20:33:50 | On PBZ*-lattices | [
"math.RA",
"math.LO"
] | We continue our investigation of paraorthomodular BZ*-lattices (PBZ*-lattices), started in \cite{GLP1+,PBZ2,rgcmfp,pbzsums,pbz5}. We shed further light on the structure of the subvariety lattice of the variety $\mathbb{PBZL}^{\ast }$ of PBZ*-lattices; in particular, we provide axiomatic bases for some of its members. Further, we show that some distributive subvarieties of $\mathbb{PBZL}^{\ast }$ are term-equivalent to well-known varieties of expanded Kleene lattices or of nonclassical modal algebras. By so doing, we somehow help the reader to locate PBZ*-lattices on the atlas of algebraic structures for nonclassical logics. | math.RA | math |
On PBZ∗ -- lattices
Roberto Giuntini1, Claudia Mure¸san2, Francesco Paoli1
1Dept. of Pedagogy, Psychology, Philosophy, University of Cagliari
2Faculty of Mathematics and Computer Science,
University of Bucharest
November 4, 2019
Abstract
We continue our investigation of paraorthomodular BZ*-lattices (PBZ∗
-- lattices), started in [18, 19, 20, 21, 33]. We shed further light on the
structure of the subvariety lattice of the variety PBZL∗ of PBZ∗ -- lattices;
in particular, we provide axiomatic bases for some of its members. Fur-
ther, we show that some distributive subvarieties of PBZL∗ are term-
equivalent to well-known varieties of expanded Kleene lattices or of non-
classical modal algebras. By so doing, we somehow help the reader to
locate PBZ∗ -- lattices on the atlas of algebraic structures for nonclassical
logics.
1 Introduction
One of the core topics within the impressive corpus of Mohammad Ardeshir's
contributions to mathematical logic is the algebraic semantics of nonclassical
logics. In particular, Ardeshir and his collaborators intensively investigated the
relationships between Visser's basic propositional calculus [39] and its algebraic
counterpart, basic algebras, generalisations of Heyting algebras where only the
left-to-right direction of the residuation equivalence x ∧ y ≤ z ⇐⇒ x ≤ y → z
is retained [2, 3, 4]. Also, in a basic algebra A there may be a ∈ A such that
1 → a 6= a. Crucially, the introduction of these structures is not motivated
by abstraction per se: Ardeshir argues that basic algebras can contribute to
a deeper understanding of constructive mathematics, whence they can have a
paramount foundational interest.
The approach that led to the introduction of paraorthomodular BZ*-lattices
(PBZ ∗ -- lattices) [18, 19, 20, 21, 33] is similar. The key motivation for this
particular generalisation of orthomodular lattices, in fact, comes from the foun-
dations of quantum mechanics. Consider the structure
E (H) = (E (H) , ∧s, ∨s,′ ,∼ , O, I) ,
where:
• E (H) is the set of all effects of a given complex separable Hilbert space
H, i.e., positive linear operators of H that are bounded by the identity
operator I;
1
• ∧s and ∨s are the meet and the join, respectively, of the spectral ordering
≤s so defined for all E, F ∈ E (H):
E ≤s F iff ∀λ ∈ R : M F (λ) ≤ M E(λ),
where for any effect E, M E is the unique spectral family [28, Ch. 7]
such that E = R ∞
−∞ λ dM E(λ) (the integral is here meant in the sense of
norm-converging Riemann-Stieltjes sums [38, Ch. 1]);
• O and I are the null and identity operators, respectively;
• E ′ = I − E and E∼ = Pker(E) (the projection onto the kernel of E).
The operations in E (H) are well-defined. The spectral ordering is indeed
a lattice ordering [34, 15] that coincides with the usual ordering of effects in-
duced via the trace functional when both orderings are restricted to the set of
projection operators of the same Hilbert space.
A PBZ∗ -- lattice can be viewed as an abstraction from this concrete physical
model, much in the same way as an orthomodular lattice can be viewed as
an abstraction from a certain structure of projection operators in a complex
separable Hilbert space. The faithfulness of PBZ∗ -- lattices to the physical
model whence they stem is further underscored by the fact that they reproduce
at an abstract level the "collapse" of several notions of sharp physical property
that can be observed in E (H).
Referring the reader to [18] for a more detailed discussion of the previous
issues, we now summarise the discourse of the present paper. In Section 2 we
collect some preliminaries, with the twofold aim of fixing the notation to be used
throughout the article and of making the article itself sufficiently self-contained
-- although we will occasionally need to refer the reader to results included in
the previous papers on the subject. In Section 3 we zoom in on some subvarieties
of the variety PBZL∗ of PBZ∗ -- lattices. First, we axiomatise the subvariety of
PBZL∗ generated by a particular algebra whose role in the context of PBZL∗ is
analogous to the role of the benzene ring in the context of ortholattices. Next,
we prove that the subvariety of PBZL∗ generated by the (unique PBZ∗ -- lattice
over the) 4-element Kleene chain is the unique antiorthomodular cover of the va-
riety generated by the (unique PBZ∗ -- lattice over the) 3-element Kleene chain.
Finally, we put to good use the construction of subdirect products of varieties
of PBZ∗ -- lattices, employing them to characterise some joins of subvarieties of
PBZ∗ -- lattices. Section 4 is devoted to term-equivalence results that establish
connections between distributive varieties of PBZ∗ -- lattices and some known
expansions of Kleene lattices, on the one hand, and nonclassical modal algebras
-- i.e., modal algebras whose nonmodal reducts are generic De Morgan algebras
rather than Boolean algebras -- on the other. We hope that these equivalences
can help readers to make out the whereabouts of PBZ∗ -- lattices in the vast land-
scape of algebraic structures for nonclassical logic, a territory whose exploration
has been decisively aided by the research work of Mohammad Ardeshir.
2 Preliminaries
For further information on the notions recalled in this section, we refer the
reader to [18, 19, 20, 21, 33].
2
We denote by N the set of the natural numbers and by N∗ = N \ {0}. If A is
an algebra, then A will denote its universe. We call trivial algebras the singleton
algebras. For any n ∈ N∗, Dn will denote the n -- element chain, as well as any
bounded lattice-ordered structure having this chain as a bounded lattice reduct.
For any lattice L, we denote by Ld the dual of L. For any bounded lattices L
and M, we denote by L ⊕ M the ordinal sum of L with M, obtained by glueing
together the top element of L and the bottom element of M, thus stacking M
on top of L, and by L ⊕ M the universe of the bounded lattice L ⊕ M; clearly,
the ordinal sum of bounded lattices is associative.
Let V be a variety of algebras of similarity type τ and C a class of alge-
bras with τ -- reducts. We denote by IV(C), HV(C), SV(C) and PV(C) the classes
of the isomorphic images, homomorphic images, subalgebras and direct prod-
ucts of τ -- reducts of members of C, respectively, and by VV(C) = HVSVPV(C)
the subvariety of V generated by the τ -- reducts of the members of C. For any
class operator O and any A ∈ C, the notation OV({A}) will be streamlined to
OV(A). If A is an algebra having a τ -- reduct, n ∈ N and κ1, . . . , κn are constants
over τ , then we denote by ConV(A) the complete lattice of the congruences of
the τ -- reduct of A, as well as the set reduct of this congruence lattice, and by
ConVκ1,...,κn (A) the complete sublattice of ConV(A) consisting of the congru-
ences with singleton classes of κA
If V is
the variety of lattices or that of bounded lattices, then the subscript V will be
eliminated from the previous notations. If C ⊆ V, then we denote by Si(C) the
class of the members of C which are subdirectly irreducible in V. The lattice of
subvarieties of V and its set reduct will be denoted by Subvar(V).
n , as well as its set reduct.
1 , . . . , κA
An involution lattice (in brief, I -- lattice) is an algebra L = (L, ∧, ∨, ·′) of
type (2, 2, 1) such that (L, ∧, ∨) is a lattice and ·′ : L → L is an order -- reversing
operation that satisfies a′′ = a for all a ∈ L. This makes ·′ a dual lattice
automorphism of L, called involution.
A bounded involution lattice (in brief, BI -- lattice) is an algebra L = (L, ∧, ∨,
·′, 0, 1) of type (2, 2, 1, 0, 0) such that (L, ∧, ∨, 0, 1) is a bounded lattice and
(L, ∧, ∨, ·′) is an involution lattice. A distributive bounded involution lattice is
called a De Morgan algebra.
For any BI -- lattice L, we denote by S(L) the set of the sharp elements of L,
that is: S(L) = {x ∈ L : x ∨ x′ = 1}. A BI -- lattice L is called an ortholattice
iff all its elements are sharp, and it is called an orthomodular lattice iff, for all
a, b ∈ L, a ≤ b implies b = (b ∧ a′) ∨ a.
A pseudo -- Kleene algebra is a BI -- lattice L that satisfies a ∧ a′ ≤ b ∨ b′
for all a, b ∈ L. The involution of a pseudo -- Kleene algebra is called Kleene
complement. Distributive pseudo -- Kleene algebras are called Kleene algebras or
Kleene lattices.
Clearly, for any bounded lattice L and any BI -- lattice K, if Kl is the bounded
lattice reduct of K, then the bounded lattice L ⊕ Kl ⊕ Ld becomes a BI -- lattice
with the involution that restricts to the involution of K on K, to a dual lattice
isomorphism from L to Ld on L and to the inverse of this lattice isomorphism
on Ld. This BI -- lattice, which we denote by L ⊕ K ⊕ Ld, is a pseudo -- Kleene
algebra iff K is a pseudo -- Kleene algebra.
We denote by BA, OML, OL, KA, PKA, BI and I the varieties of Boolean
algebras, orthomodular lattices, ortholattices, Kleene algebras, pseudo -- Kleene
algebras, BI -- lattices and I -- lattices, respectively. Note that BA ( OML ( OL (
3
PKA ( BI and BA ( KA ( PKA.
An algebra A having a BI -- lattice reduct is said to be paraorthomodular iff,
for all a, b ∈ A, if a ≤ b and a′ ∧ b = 0, then a = b. Note that orthomodu-
lar lattices are paraorthomodular and that paraorthomodular ortholattices are
orthomodular lattices.
A Brouwer -- Zadeh lattice (in brief, BZ -- lattice) is an algebra L = (L, ∧, ∨, ·′,
·∼, 0, 1) of type (2, 2, 1, 1, 0, 0) such that (L, ∧, ∨, ·′, 0, 1) is a pseudo -- Kleene alge-
bra and ·∼ : L → L is an order -- reversing operation, called Brouwer complement,
that satisfies: a ∧ a∼ = 0 and a ≤ a∼∼ = a∼′ for all a ∈ L. In any BZ -- lattice
L, we denote by (cid:3)a = a′∼ and by ♦a = a∼∼ for all a ∈ L. Note that, in any
BZ -- lattice L, we have, for all a, b ∈ L: a∼∼∼ = a∼ ≤ a′, (a ∨ b)∼ = a∼ ∧ b∼
and (a ∧ b)∼ ≥ a∼ ∨ b∼. The class of BZ-lattices is a variety, hereafter denoted
by BZL.
We consider the following equations over BZL, out of which SDM (the Strong
De Morgan identity) clearly implies (∗), as well as SK, while J0 implies J2:
(∗)
(x ∧ x′)∼ ≈ x∼ ∨ x′∼
SDM (x ∧ y)∼ ≈ x∼ ∨ y∼
SK x ∧ ♦y ≤ (cid:3)x ∨ y
DIST x ∧ (y ∨ z) ≈ (x ∧ y) ∨ (x ∧ z)
J0
J2
(x ∧ y∼) ∨ (x ∧ ♦y) ≈ x
(x ∧ (y ∧ y′)∼) ∨ (x ∧ ♦(y ∧ y′)) ≈ x
A PBZ∗ -- lattice is a paraorthomodular BZ -- lattice that satisfies equation
(∗). In any PBZ∗ -- lattice L,
S(L) = {a∼ : a ∈ L} = {a ∈ L : a∼∼ = a} = {a ∈ L : a′ = a∼}
and S(L) is the universe of the largest orthomodular subalgebra of L, that we
denote by S(L).
We denote by PBZL∗ the variety of PBZ∗ -- lattices; note that paraorthomod-
ularity becomes an equational condition under the BZL axioms and condition
(∗). We also denote by DIST = {L ∈ PBZL∗ : L (cid:15) DIST}. By the above, OML
can be identified with the subvariety {L ∈ PBZL∗ : L (cid:15) x′ ≈ x∼} of PBZL∗,
by endowing each orthomodular lattice, in particular every Boolean algebra,
with a Brouwer complement equalling its Kleene complement. With the same
extended signature, OL becomes the subvariety {L ∈ BZL : L (cid:15) x′ ≈ x∼} of
BZL.
A PBZ∗ -- lattice A with no nontrivial sharp elements, that is with S(A) =
{0, 1}, is called an antiortholattice. A PBZ∗ -- lattice A is an antiortholattice iff
it is endowed with the following Brouwer complement, called the trivial Brouwer
complement : 0∼ = 1 and a∼ = 0 for all a ∈ A \ {0}. Every paraorthomodular
pseudo -- Kleene algebra with no nontrivial sharp elements becomes an antiortho-
lattice when endowed with the trivial Brouwer complement. In particular, any
BZ -- lattice with the 0 meet -- irreducible, and thus any BZ -- chain, is an antiortho-
lattice. Moreover, BZ -- lattices with the 0 meet -- irreducible are exactly the an-
tiortholattices that satisfy SDM. Also, if L is a nontrivial bounded lattice and
K is a pseudo -- Kleene algebra, then the pseudo -- Kleene algebra L ⊕ K ⊕ Ld, en-
dowed with the trivial Brouwer complement, becomes an antiortholattice, that
we will also denote by L ⊕ K ⊕ Ld.
4
Antiortholattices form a proper universal class, denoted by AOL. Clearly,
AOL ∪ OML ( PBZL∗ ( BZL ) OL. Note, also, that OML ∩ VBZL(AOL) =
OML ∩ DIST = BA, hence DIST ( VBZL(AOL). We denote by SDM = {L ∈
PBZL∗ : L (cid:15) SDM} and by SAOL = SDM ∩ VBZL(AOL).
If L is a nontrivial bounded lattice and C is a class of bounded lattices, BI --
lattices or pseudo -- Kleene algebras, then we denote by L ⊕ C ⊕ Ld the following
class of bounded lattices, BI -- lattices or antiortholattices:
L ⊕ C ⊕ Ld = {L ⊕ A ⊕ Ld : A ∈ C}.
3 A Study of Some Subvarieties
Throughout this section, the results cited from [33] will be numbered as in the
third arXived version of this paper.
3.1 The F8 Problem
There is a long and time-honoured tradition that aims at characterising sub-
varieties of varieties of ordered algebras in terms of "forbidden configurations",
harking back to Dedekind's celebrated result to the effect that the distributive
subvariety of the variety of lattices is the one whose members do not contain as
subalgebras M3 or N5, while the modular subvariety is the one whose members
do not contain N5. Other important results in the same vein appear in the
theory of ortholattices. For example, the benzene ring B6:
B6 :
1
✟✟ ❍❍
a′
b′
r
r
r
r
❍❍ ✟✟
a
r
b
r
0
is a forbidden configuration for the orthomodular subvariety of the variety of
ortholattices; more precisely,
OML = {L ∈ OL : B6 /∈ SI(L)} .
Consequently:
Lemma 1 (OML, VBI(B6)) is a splitting pair in Subvar (OL).
In this subsection, we intend to give a first,
limited application of this
method, by means of a forbidden configuration consisting of a "paraorthomod-
ular analogue" of B6: the antiortholattice D2 ⊕ B6 ⊕ D2, hereafter denoted by
F8, along with any of its reducts, for the sake of brevity:
5
1
r
F8 = D2 ⊕ B6 ⊕ D2:
c′
✟✟ ❍❍
a′
b′
r
r
r
r
❍❍ ✟✟
a
r
b
r
c
r
0
Since it has the 0 meet -- irreducible, the antiortholattice F8 satisfies SDM,
thus F8 ∈ SAOL. The question arises naturally as to which subvarieties V
of PBZL∗ are maximal with respect to the property that F8 /∈ SI(V),
i.e.
F8 /∈ SI(A) for any A ∈ V. This problem will be referred to as the "F8
problem". Although we will not give an answer to this question, we provide a
quasiequational characterisation of paraorthomodular bounded involution lat-
tices that do not contain F8 as a bounded involution sublattice and we study
the varieties of PBZ∗ -- lattices that contain the antiortholattice F8.
Clearly, for any L, M ∈ BI, we have: D2 ⊕M⊕D2 ∈ SI(L) iff D2 ⊕M⊕D2 ∈
SBI(L). The right-to-left direction is trivial, while, if D2 ⊕ M ⊕ D2 ∈ SI(L) and
A = M ∪ {0, 1}, then D2 ⊕ M ⊕ D2 ∼=BI A ∈ SBI(L). In particular, for any
A ∈ BZL, we have that F8 ∈ SI(A) iff F8 ∈ SBI(A); also, if F8 ∈ SBZL(A),
then F8 ∈ SBI(A), while, if A is an antiortholattice, then F8 ∈ SBZL(A) iff
F8 ∈ SBI(A).
Observe what follows:
• no distributive PBZ∗ -- lattice can contain B6 or F8 as sublattices, in par-
ticular as sub-involution lattices;
• since B6 is a sub-involution lattice of F8 and B6 is not a sub-involution
lattice of any orthomodular lattice, no orthomodular lattice can contain
F8 as a sub-involution lattice;
• by the above, any subvariety V of PBZL∗ such that V ⊆ DIST ∪ OML
satisfies F8 /∈ SI(V);
• F8 ∈ SAOL, whence any subvariety V of PBZL∗ such that SAOL ⊆ V
satisfies F8 ∈ SI(V).
Let us now consider the following quasiequations in the language of I -- lattices:
q(cid:13) x ≤ y′ & x′ ∧ y′ ≤ x ∧ y ⇒ x = y′
q(cid:13) ′
x′ ∧ (x′ ∧ u)′ ≤ x ∧ (x′ ∧ u) ⇒ u ≤ x′
Note that q(cid:13) is equivalent to q(cid:13) ′.
Lemma 2 If A ∈ I and a, b ∈ A are such that a ≤ b′ and a′ ∧ b′ ≤ a ∧ b, then
a ∧ a′ = b ∧ b′ = a′ ∧ b′ = a ∧ b.
Proof. Let c = a′ ∧ b′. Then c ≤ a ∧ b by the choice of a and b, therefore,
since we also have a ≤ b′ and thus b ≤ a′: a ∧ a′ = a ∧ b′ ∧ a′ = a ∧ c = c;
b ∧ b′ = b ∧ a′ ∧ b′ = b ∧ c = c; a ∧ b = a ∧ b′ ∧ b = a ∧ c = c.
6
Lemma 3 For any A ∈ PBI, we have:
B6 ∈ SI(A) iff F8 ∈ SBI(A).
Proof. The right-to-left direction is trivial. Now assume that B6 ∈ SI(A), with
B6 = {c, a, b, a′, b′, c′} ⊆ A, where c = a ∧ b and a < b′. Assume ex absurdo
that c = 0, so that a′ ∧ b′ = 0. Since A is paraorthomodular, it follows that
a = b′, and we have a contradiction. Therefore c 6= 0, so, if we denote by
L = {0, c, a, b, a′, b′, c′, 1}, then F8 ∼=BI L ∈ SBI(A).
Proposition 4 For any A ∈ I, we have:
A (cid:15) q(cid:13) iff B6 /∈ SI(A).
Proof. For the direct implication, assume that B6 ∈ S(A), with B6 = {c, a, b, a′,
b′, c′} ⊆ A, where c = a ∧ b and a < b′. Then a ≤ b′ and a′ ∧ b′ = a ∧ b ≤ a ∧ b,
but a 6= b′, hence A 2 q(cid:13).
For the converse, assume that A 2 q(cid:13), so that there exist a, b ∈ A with
a′ ∧ b′ ≤ a ∧ b and a < b′, so b < a′. Then, by Lemma 2, if we denote by
c = a′ ∧ b′, then c = a ∧ b = a ∧ a′ = b ∧ b′. Since a < b′, a ∧ b ≤ a′ ∨ b′; were
it the case that a ∧ b = a′ ∨ b′, we would have that a′ ≤ a′ ∨ b′ = a ∧ b ≤ b, a
contradiction. Hence c′ = (a∧b)′ = a′ ∨b′ > a∧b = c. Also, a∨b = (a′ ∧b′)′ = c′,
a ∨ a′ = (a ∧ a′)′ = c′ and b ∨ b′ = (b ∧ b′)′ = c′.
If we had a ≤ b, then
a ≤ b ∧ b′ = c = a ∧ a′ ≤ a, hence c = a ∧ a′ = a < b′ ≤ a′ ∧ b′ = c, and we
have a contradiction again. Similarly, b (cid:2) a. Hence a and b are incomparable.
Were it a ≤ a′, then c = a ∧ a′ = a, which would lead to the same contradiction
as above. On the other hand, if a′ ≤ a, then c = a ∧ a′ = a′ > b ≥ b ∧ b′ = c,
which gives us another contradiction. Hence a and a′ are incomparable and so
are, analogously, b and b′. Therefore, if we denote by L = {c, a, b, a′, b′, c′}, then
B6 ∼=I L ∈ SI(A).
Theorem 5 For any A ∈ PBI, we have:
A (cid:15) q(cid:13) iff F8 /∈ SBI(A).
Proof. By Lemma 3 and Proposition 4.
Example 6 Here is an antiortholattice (in particular, a paraorthomodular BI --
lattice) A such that F8 /∈ SBI(A), but F8 ∈ HBZL(A), in particular F8 ∈
SBI((HBZL(A))) ⊆ SBI(HBI(A)):
1
r
A:
r
c′
❅
❍❍❍❍
❅
e′
r
❏❏
a′
r
b′
r
❅
❏❏
❅
❏
❏
r
d′
❏
❏
❏
❏
d
r
e
❏
❅
❍❍❍❍r
❅
a
r
❏
❅
❏
❅
❏
r
b
c
r
r
0
7
The equivalence relation θ with cosets
{0}, {a}, {c, e}, {b, d}, {b′, d′}, {c′, e′}, {a′}, {1}
belongs to ConBI01(A) ⊂ ConBZL(A) and A/θ ∼= F8, but, as announced above,
F8 /∈ SBI(A).
Corollary 7 q(cid:13) is not an equational condition in PBI or PBZL∗.
Now let us investigate the subvarieties of PBZL∗ that contain F8. We con-
sider the following equation in the language of BZ -- lattices:
D2OL∨ (x ∧ x′)∼ ∨ (y ∧ y′)∼ ∨ x ∨ x′ ≈ (x ∧ x′)∼ ∨ (y ∧ y′)∼ ∨ y ∨ y′
By [20], VBZL(AOL) is axiomatised by J0 relative to PBZL∗. By [33],
VBZL(D2 ⊕ OL ⊕ D2) is axiomatised by D2OL∨ relative to SAOL.
We use the following notation from [33]: for any k, n, p ∈ N and any equation
t ≈ u, where t(x1, . . . , xk, z1, . . . , zp) and u(y1, . . . , yn, z1, . . . , zp) are terms in
the language of BI having the arities k + p, respectively n + p, and p common
variables z1, . . . , zp, we denote by m(t, u) the following (k + n) -- ary term in the
language of BZL:
m(t, u)(x1, . . . , xk, y1, . . . , yn, z1, . . . , zp) =
k
_
i=1
(xi ∧ x′
i)∼ ∨
n
_
j=1
(yj ∧ y′
j)∼ ∨
Note that:
p
(zh ∧ z ′
h)∼ ∨ t(x1, . . . , xk, z1, . . . , zp).
_
h=1
m(u, t)(x1, . . . , xk, y1, . . . , yn, z1, . . . , zp) =
k
_
i=1
(xi ∧ x′
i)∼ ∨
n
_
j=1
(yj ∧ y′
j)∼ ∨
p
(zh ∧ z ′
h)∼ ∨ u(y1, . . . , yn, z1, . . . , zp).
_
h=1
Lemma 8 [33, Corollary 6.14] For any C ⊆ BI and any D ⊆ PKA, VBI(D2⊕C⊕
D2) = VBI(D2 ⊕VBI(C)⊕D2) and VBZL(D2 ⊕D⊕D2) = VBZL(D2 ⊕VBI(D)⊕D2).
Proposition 9 VBI(F8) = VBI(D2 ⊕ VBI(B6) ⊕ D2) and VBZL(F8) = VBZL(D2 ⊕
VBI(B6) ⊕ D2).
Proof. By Lemma 8 and the fact that F8 = D2 ⊕ B6 ⊕ D2.
The following consequence of results from [33] shows that we can obtain
an axiomatisation for VBZL(F8) relative to PBZL∗ from an axiomatisation of
VBI(B6) relative to OL; note that any such axiomatisation can be written with
nonnullary terms over BI, since OL satisfies the equations x ∨ x′ ≈ 1 and
x ∧ x′ ≈ 0.
Corollary 10 {ti ≈ ui : i ∈ I} is an axiomatisation of VBI(B6) relative to
OL such that, for each i ∈ I, the terms ti and ui have nonzero arities iff
{m(ti, ui) ≈ m(ui, ti) : i ∈ I} ∪ {J0, D2OL∨} is an axiomatisation of VBZL(F8)
relative to PBZL∗.
8
Proof. By Proposition 9, the fact that VBI(B6) ⊆ OL and [33, Theorem
6.38.(ii)].
Theorem 11 [33, Theorem 6.25] The operator V 7→ VBZL(D2 ⊕ V ⊕ D2) is a
bounded lattice embedding from the lattice of subvarieties of PKA to the principal
filter generated by VBZL(D3) in the lattice of subvarieties of SAOL.
Corollary 12 (VBZL(D2⊕OML⊕D2), VBZL(F8)) is a splitting pair in the lattice
of subvarieties of OL.
Proof. By Lemma 1, Proposition 9 and Theorem 11.
Proposition 13
• VBI(B6) ( VBI(F8) = VBI(Dn⊕F8⊕Dn) for any n ∈ N∗;
• VBZL(F8) ( VBZL(D2 ⊕ F8 ⊕ D2) = VBZL(Dn ⊕ F8 ⊕ Dn) for any n ∈
N \ {0, 1, 2}.
Proof. By Proposition 9, the fact that VBI(B6) ⊆ OL, while D3 ∈ VBI(F8), and
[33, Corollary 6.23], we get that VBI(B6) ( VBI(F8) = VBI(D2 ⊕ F8 ⊕ D2) and
hence VBI(F8) = VBI(Dn ⊕F8⊕Dn) for any n ∈ N∗. This, Theorem 11 and again
Proposition 9 show that VBZL(F8) ( VBZL(D2 ⊕ F8 ⊕ D2) = VBZL(Dn ⊕ F8 ⊕ Dn)
for any n ∈ N \ {0, 1, 2}.
3.2 Covers in the Lattice of Subvarieties of PBZL∗
In this subsection, we continue the study of the lattice Subvar(PBZL∗) of sub-
varieties of PBZ∗ -- lattices, started in [18, 19, 20, 21, 33]. We begin by recapit-
ulating a few known results.
Lemma 14
(i) [18, Subsection 5.3] BA is the unique atom of Subvar(PBZL∗).
(ii) [18, Theorem 5.4.(2)] BA = OML ∩ VBZL(AOL).
(iii) [7, Corollary 3.6] The unique cover of BA in the ideal (OML] of
Subvar(PBZL∗) is VBZL(MO2).
(iv) [18, Theorem 5.5] For any L ∈ PBZL∗ \ OML, we have D3 ∈ HS(L) ⊆
VBZL((L)), so the unique non -- orthomodular cover of BA in Subvar(PBZL∗)
is VBZL(D3).
By the above, in Subvar(PBZL∗) VBZL(MO2) and VBZL(D3) are the only
covers of BA, and OML ∨ VBZL(D3) is the unique cover of OML.
Lemma 15 [19, Lemma 3.3.(1)] All
VBZL(AOL) belong to AOL.
subdirectly irreducible members of
Lemma 16 [33]
(i) BA = OML ∩ VBZL(AOL) = VBZL(D2) ( VBZL(D3) ( VBZL(D4) (
VBZL(D5).
(ii) Si(VBZL(D3)) = VBZL(D3) ∩ AOL = IBZL({D1, D2, D3}).
9
We now prove the main result of this subsection.
Theorem 17 The only cover of VBZL(D3) in Subvar(PBZL∗) included in
VBZL(AOL) is VBZL(D4).
Proof. For any subvariety W of VBZL(AOL) such that VBZL(D3) ( W, there
exists an A ∈ Si(W) \ Si(VBZL(D3)) = (W ∩ AOL) \ IBZL({D1, D2, D3}) by
Lemma 15 and Lemma 16.(ii), thus A is an antiortholattice with A > 3.
Hence, there exists an a ∈ A \ {0, 1} = A \ SBZL(A) with a 6= a′, so that
0 < a∧a′ < a∨a′ < 1. Therefore {0, a∧a′, a∨a′, 1} is the universe of a subalgebra
of A isomorphic to D4, i.e. D4 ∈ SBZL(A), thus VBZL(D4) ⊆ VBZL(A) ⊆
W. Since VBZL(D3) ( VBZL(D4) by Lemma 16.(i), it follows that VBZL(D4)
is the only cover of VBZL(D3) in Subvar(VBZL(AOL)), which is, of course, a
convex sublattice of Subvar(PBZL∗), thus VBZL(D4) is a cover of VBZL(D3) in
Subvar(PBZL∗).
It remains open to determine whether VBZL(D4) is the only cover of VBZL(D3)
in Subvar(PBZL∗). Recall, also, that VBZL(D5) = SDM ∩ DIST contains all
antiortholattice chains, i.e., all PBZ∗ -- chains.
Example 18 Let us consider the following example of a PBZ∗ -- lattice from
[20]:
r
r
r
g′
H :
f ′
❅
✟✟✟✟✟✟✟✟
❍❍❍❍❍❍❍❍
f ∼= b
❅
❅
d
d′
✟✟✟✟✟✟✟✟
❍❍❍❍❍❍❍❍
❅
e′
❅
c = c′
❍❍❍❍❍❍❍❍
✟✟✟✟✟✟✟✟
❅
❅
e
❅
❅
g
r
r
r
r
r
r
r
b′ = e∼ a′ = d∼
r
1
❅
rr
r
r
❅
f
❅
0
r
g∼= a
Note that:
• H (cid:15) {SDM, SK}, thus OML ∨ VBZL(H) (cid:15) {SDM, SK} since OML (cid:15)
{SDM, SK};
• H 2 J2, thus H /∈ OML ∨ VBZL(AOL) (cid:15) J2, in particular H /∈ OML ∨
VBZL(D3);
• D3 ∈ S(H), hence OML ∨ VBZL(D3) ⊆ OML ∨ VBZL(H), therefore OML ∨
VBZL(D3) ( OML ∨ VBZL(H) by the above;
• since OML∨VBZL(H) (cid:15) SK and D4 2 SK, we have D4 /∈ OML∨VBZL(H),
hence D4 does not belong to every proper supervariety of OML∨VBZL(D3).
H (cid:15) {SDM, SK}, H 2 J2 and OML∨VBZL(AOL) (cid:15) J2, hence H ∈ (SDM ∩
SK) \ (OML ∨ VBZL(AOL)), thus SDM ∩ SK * OML ∨ VBZL(AOL). AOL 2
SDM and AOL 2 SK, thus AOL * SDM and AOL * SK, in particular
OML ∨ VBZL(AOL) * SDM ∩ SK. Therefore SDM ∩ SKOML ∨ VBZL(AOL).
Now let V = VBZL(M O2)∨VBZL(D3) ⊆ SDM∩SK. D3 /∈ OML, thus V * OML.
M O2 /∈ VBZL(AOL), thus V * VBZL(AOL). Finally, V satisfies the modular law,
while both OML and VBZL(AOL) fail it, hence OML * V and VBZL(AOL) * V.
Therefore OMLVVBZL(AOL).
10
We list hereafter a few problems that remain open at the time of writing:
• Is OML ∨ VBZL(D4) a successor of OML ∨ VBZL(D3) in Subvar(PBZL∗)?
Is it its only successor?
• Is Subvar(PBZL∗) strongly atomic? If so, then OML ∨ VBZL(H) includes
a successor of OML ∨ VBZL(D3) which differs from OML ∨ VBZL(D4).
3.3 Subdirect Products and Varieties of PBZ∗ -- lattices
Let V and W be varieties of the same type. Obviously, if V and W are in-
comparable, then there exist A ∈ (V ∨ W) \ V and B ∈ (V ∨ W) \ W, so that
A × B ∈ (V ∨ W) \ (V ∪ W) and thus V ∪ W ( V ∨ W. Recall that the subdirect
product of V and W is the class, denoted by V ×s W, whose members are iso-
morphic images of subdirect products of a member of V and a member of W.
Clearly, V ∪ W ⊆ V ×s W ⊆ V ∨ W, so that
Si(V) ∪ Si(W) = Si(V ∪ W) ⊆ Si(V ×s W) ⊆ Si(V ∨ W).
For any M ∈ Si(V ×s W), M is a subdirect product of an A ∈ V and a
B ∈ W, so that A is trivial, case in which M ∈ Si(W), or B is trivial, case in
which M ∈ Si(V). Thus Si(V ×s W) ⊆ Si(V) ∪ Si(W), hence Si(V ×s W) =
Si(V) ∪ Si(W). Since V ×s W ⊆ V ∨ W, we get that the following equivalence
holds: V ∨ W = V ×s W iff Si(V ∨ W) = Si(V) ∪ Si(W).
Sufficient Maltsev-type conditions for the equivalence V ∨ W = V ×s W to
hold are available in the literature: see [35, 26, 27]. These contributions are all
inspired by the celebrated result by Gratzer, Lakser and P lonka according to
which two independent similar varieties V and W are such that every member
of V ∨ W is isomorphic to the direct product of a member of V and a member
of W [24]. Of course, the notion of independence is of limited use in the context
of PBZ∗ -- lattices, since BA is the unique atom in Subvar(PBZL∗) and thus
there are no two nontrivial disjoint (hence, no two independent) varieties of
PBZ∗ -- lattices. The investigation of subdirect products of varieties of PBZ∗ --
lattices, however, can be carried out with more ad hoc methods, yielding useful
information on joins of specific subvarieties.
If V ∨ W = V ×s W and U is a variety of the same type as V and W, then
(U ∩ V) ×s (U ∩ W) ⊆ (U ∩ V) ∨ (U ∩ W) ⊆ U ∩ (V ∨ W) and
Si(U ∩ (V ∨ W)) = Si(U) ∩ Si(V ∨ W)
= Si(U) ∩ (Si(V) ∪ Si(W))
= (Si(U) ∩ Si(V)) ∪ (Si(U) ∩ Si(W))
= Si(U ∩ V) ∪ Si(U ∩ W)
= Si((U ∩ V) ×s (U ∩ W)),
hence U ∩ (V ∨ W) = (U ∩ V) ∨ (U ∩ W) = (U ∩ V) ×s (U ∩ W). For instance, since
OML ∨ VBZL(AOL) = OML ×s VBZL(AOL) (see Lemma 21 below), it follows
that
SDM ∩ (OML ∨ VBZL(AOL)) = (SDM ∩ OML) ∨ (SDM ∩ VBZL(AOL))
= OML ∨ SAOL = OML ×s SAOL.
As a consequence of the above, if V∨W = V×sW and Subvar(V) and Subvar(W)
are distributive, then Subvar(V ∨ W) is distributive.
11
Problem 19 If V ∨ W = V ×s W, C is a subvariety of V and D is a subvariety
of W, under what conditions does it follow that C ∨ D = C ×s D? Does the
condition that C ∩ D = V ∩ W suffice? A partial answer to this question is given
by Lemma 20 below.
If V ∨ W = V ×s W and U is a subvariety of V, then U ∨ W is a subvariety
of V ∨ W, so that
Si(U ∨ W) = (U ∨ W) ∩ Si(V ∨ W)
= (U ∨ W) ∩ (Si(V) ∪ Si(W))
= ((U ∨ W) ∩ Si(V)) ∪ ((U ∨ W) ∩ Si(W))
= ((U ∨ W) ∩ Si(V)) ∪ Si(W).
Lemma 20 Let V and W be varieties of a similarity type τ , U a subvariety of
V and Γ a set of equations over τ such that V ∨ W = V ×s W, W (cid:15) Γ and
U = {A ∈ V : A (cid:15) Γ}. Then:
• U ∨ W = U ×s W = {A ∈ V ∨ W : A (cid:15) Γ};
• U = V iff U ∨ W = V ∨ W.
Proof. Of course, Si(U) ∪ Si(W) ⊆ Si(U ∨ W). For all A ∈ Si(U ∨ W), we
have: A ∈ Si(V ∨ W) = Si(V) ∪ Si(W) and A (cid:15) Γ, so that either A ∈ Si(W)
or A ∈ Si(V) ⊂ V and A (cid:15) Γ, the latter of which implies that A ∈ Si(V) ∩ U =
Si(U). Therefore Si(U ∨ W) = Si(U) ∪ Si(W), thus U ∨ W = U ×s W. We have
that:
Si({A ∈ V ∨ W : A (cid:15) Γ}) = {A ∈ Si(V ∨ W) : A (cid:15) Γ}
= {A ∈ Si(V) ∪ Si(W) : A (cid:15) Γ}
= {A ∈ Si(V) : A (cid:15) Γ} ∪ Si(W)
= Si({A ∈ V : A (cid:15) Γ}) ∪ Si(W)
= Si(U) ∪ Si(W) = Si(U ∨ W),
hence U ∨ W = {A ∈ V ∨ W : A (cid:15) Γ}.
Trivially, U = V implies U∨W = V∨W. Conversely, if V∨W = U∨W = {A ∈
V ∨ W : A (cid:15) Γ}, then V ∨ W (cid:15) Γ, thus V (cid:15) Γ, hence U = {A ∈ V : A (cid:15) Γ} = V.
Lemma 21 [20] All subdirectly irreducible members of OML ∨ VBZL(AOL) be-
long to OML ∪ AOL, in particular OML ∨ VBZL(AOL) = OML ×s VBZL(AOL).
We can derive from the above the following result from [20]:
Proposition 22
• OML∨SAOL = OML×sSAOL and OML∨VBZL(D3) =
OML ×s VBZL(D3);
• OML ∨ VBZL(D3) ( OML ∨ SAOL ( OML ∨ VBZL(AOL).
Proof. Recall from [19, Corollary 3.3] that VBZL(D3) = {A ∈ VBZL(AOL) :
A (cid:15) {SDM, SK}}. Now apply the fact that OML (cid:15) {SDM, SK} and Lemmas
21 and 20 to obtain first that OML ∨ SAOL = OML ×s SAOL, then that
OML ∨ VBZL(D3) = OML ×s VBZL(D3). Recall that D5 ∈ SAOL \ VBZL(D3),
12
2 ⊕ D2
which is easily noticed from the fact that D5 2 SK. The antiortholattice
D2
2 ∈ VBZL(AOL) \ SAOL. Hence VBZL(D3) ( SAOL ( VBZL(AOL), thus
OML ∨ VBZL(D3) ( OML ∨ SAOL ( OML ∨ VBZL(AOL) by Lemma 20 and the
above.
Let us consider the identities:
W DSDM
DIST ∨∼
(x ∧ (y ∨ z))∼ ≈ (x ∧ y)∼ ∧ (x ∧ z)∼
(x ∨ x∼) ∧ (y ∨ y∼ ∨ z ∨ z∼) ≈
((x ∨ x∼) ∧ (y ∨ y∼)) ∨ ((x ∨ x∼) ∧ (z ∨ z∼))
W DIST ∨∼ ((x ∨ x∼) ∧ (y ∨ y∼ ∨ z ∨ z∼))∼ ≈
(((x ∨ x∼) ∧ (y ∨ y∼)) ∨ ((x ∨ x∼) ∧ (z ∨ z∼)))∼
Note that W DSDM implies W DIST ∨∼ and DIST ∨∼ implies W DIST ∨∼.
Also, recall from [19, 33] that VBZL(D5) = SAOL ∩ DIST.
Proposition 23 VBZL(D5) = SAOL ∩ DIST ( SAOL, DIST ( SAOL ∨ DIST (
VBZL(AOL).
Proof. Observe that the identity W DSDM is satisfied both in SAOL and in
DIST. The antiortholattice on M3 ⊕ M3 fails WDSDM, because, if a, b, c are its
three atoms, then (a ∧ (b ∨ c))∼ = a∼ = 0, yet (a ∧ b)∼ ∧ (a ∧ c)∼ = 0∼ ∧ 0∼ = 1.
Hence M3 ⊕ M3 ∈ AOL \ (SAOL ∨ DIST) ⊆ VBZL(AOL) \ (SAOL ∨ DIST).
The antiortholattice D2 ⊕ M3 ⊕ D2 ∈ SAOL \ DIST, while the antiortholattice
D2
2 ∈ DIST\SAOL, hence SAOL and DIST are incomparable, thus SAOL∩
DIST ( SAOL, DIST ( SAOL ∨ DIST.
2⊕D2
Proposition 24
• OML ∨ DIST = OML ×s DIST and OML ∨ VBZL(D5) =
OML ×s VBZL(D5);
• OML∨VBZL(D3) ( OML∨VBZL(D5) = OML∨(SAOL∩DIST) = (OML∨
SAOL) ∩ (OML ∨ DIST) ( OML ∨ SAOL, OML ∨ DIST ( OML ∨ SAOL ∨
DIST ( OML ∨ VBZL(AOL), in particular the varieties OML ∨ SAOL and
OML ∨ DIST are incomparable.
Proof. Note that OML (cid:15) DIST ∨∼ and that, in AOL, DIST ∨∼ is equivalent
to DIST , that is DIST ∩AOL = {A ∈ AOL : A (cid:15) DIST ∨∼}. The latter, along
with the fact that DIST is a subvariety of VBZL(AOL) and Lemma 15, give us:
Si(DIST) = DIST ∩ Si(VBZL(AOL))
= DIST ∩ Si(AOL)
= Si(DIST ∩ AOL)
= Si({A ∈ AOL : A (cid:15) DIST ∨∼})
= Si({A ∈ VBZL(AOL) : A (cid:15) DIST ∨∼}),
therefore DIST = {A ∈ VBZL(AOL) : A (cid:15) DIST ∨∼}. By Lemmas 21 and
20, it follows that OML ∨ DIST = OML ×s DIST. By the above, OML (cid:15)
{SDM, DIST ∨∼} and VBZL(D5) = SAOL ∩ DIST = {A ∈ VBZL(AOL) : A (cid:15)
{SDM, DIST ∨∼}}, hence OML ∨ VBZL(D5) = OML ×s VBZL(D5) by Lemmas
21 and 20. By the above, Propositions 22 and 23 and again Lemma 20, it follows
that:
OML ∨ VBZL(D3) ( OML ∨ VBZL(D5)
( OML ∨ SAOL, OML ∨ DIST
( OML ∨ VBZL(AOL).
13
By Lemma 20, the above and Proposition 22,
OML ∨ VBZL(D5) = OML ∨ (SAOL ∩ DIST)
= OML ∨ {A ∈ VBZL(AOL) : A (cid:15) {SDM, DIST }}
= {A ∈ OML ∨ VBZL(AOL) : A (cid:15) {SDM, DIST }}
= {A ∈ OML ∨ VBZL(AOL) : A (cid:15) SDM }
∩{A ∈ OML ∨ VBZL(AOL) : A (cid:15) DIST }
= (OML ∨ SAOL) ∩ (OML ∨ DIST),
hence (OML ∨ SAOL) ∩ (OML ∨ DIST) ( OML ∨ SAOL, OML ∨ DIST, so that
OML ∨ SAOL and OML ∨ DIST are incomparable. Therefore
OML ∨ SAOL, OML ∨ DIST ( OML ∨ SAOL ∨ OML ∨ DIST
= OML ∨ SAOL ∨ DIST.
Since OML (cid:15) DIST ∨∼ and SAOL ∨ DIST (cid:15) W DSDM , it follows that OML ∨
SAOL ∨ DIST (cid:15) W DIST ∨∼. Note that, in AOL, W DIST ∨∼ is equivalent to
W DSDM , hence, by the proof of Proposition 23, the antiortholattice M3 ⊕ M3
fails W DIST ∨∼. It follows that M3 ⊕ M3 ∈ AOL \ (OML ∨ SAOL ∨ DIST) ⊆
(OML∨VBZL(AOL))\ (OML∨SAOL∨DIST), therefore OML∨SAOL∨DIST (
OML ∨ VBZL(AOL).
Lemma 25 For any subvariety V of OML∨VBZL(AOL), Si(V) = V∩Si(OML∪
AOL).
Proof. By Lemma 21.
Note that, if a PBZ∗ -- lattice L satisfies the SDM, then 0 is meet -- irreducible
in the join -- subsemilattice T (L) of L, but the converse does not hold.
Lemma 26 Let A be an antiortholattice without SDM and (Ai)i∈I be a non --
empty family of antiortholattices. Then:
• if A ∈ SBZL(Qi∈I Ai), then the family (Ai)i∈I contains no nontrivial
antiortholattice with SDM;
• A ∈ SBZL(Qi∈I Ai) iff A ∈ SBZL(Qi∈I,Ai2SDM Ai).
Proof. The second statement obviously follows from the first. Now assume
that A ∈ SBZL(Qi∈I Ai), let J = {j ∈ I : Aj (cid:15) SDM } and assume ex absurdo
that there exists a k ∈ J such that Ak is nontrivial. We may consider A ⊆
Qi∈I Ai. A is an antiortholattice that fails SDM, in particular a nontrivial
antiortholattice, hence there exist a = (ai)i∈I , b = (bi)i∈I ∈ A \ {0} = D(A) =
D(Qi∈I Ai) = Qi∈I D(Ai) = Qi∈I ((Ai \ {0}) ∪ {1}) such that a ∧ b = 0, so that
ak ∧ bk = 0 and ak, bk ∈ D(Ak) = Ak \ {0}, which contradicts the fact that Ak
satisfies the SDM.
Proposition 27 If V is a subvariety of VBZL(AOL), then: V ∨ SAOL = V ×s
SAOL iff (V ∨ SAOL) ∩ AOL = (V ∪ SAOL) ∩ AOL.
Proof. By the above, V∨SAOL = V×sSAOL iff Si(V∨SAOL) = Si(V∪SAOL).
Since Si(VBZL(AOL)) ⊂ AOL, the right-to-left implication holds. Now assume
that Si(V ∨ SAOL) = Si(V ∪ SAOL), and assume ex absurdo that there exists
14
an L ∈ ((V ∨ SAOL) ∩ AOL) \ (V ∪ SAOL). Then L ∈ V ×s SAOL, hence
L ∈ SBZL(A × B) for some A ∈ V and some B ∈ SAOL, therefore L ∈ SBZL(A ×
Qj∈J Bj) for some family (Bj )j∈J ⊆ SAOL ∩ AOL. Thus L ∈ SBZL(A) by
Lemma 26, so that L ∈ V, a contradiction. Hence (V ∨ SAOL) ∩ AOL ⊆
(V ∪ SAOL) ∩ AOL.
4 Comparison with Other Structures
4.1 Distributive Lattices with Two Unary Operations
Bounded distributive lattices expanded both by a De Morgan complementa-
tion and a unary operation with Stone-like properties have been the object of
rather intensive investigations over the past decades. In particular, Blyth, Fang
and Wang [6] have studied, under the label of quasi-Stone De Morgan algebras,
bounded distributive lattices with two unary operations that make their appro-
priate reducts, at the same time, De Morgan algebras and quasi-Stone algebras
[37, 17, 13]. Quasi-Stone De Morgan algebras that are simultaneously Stone al-
gebras and Kleene algebras are known under the name of Kleene-Stone algebras;
they have been studied in [25] and, more recently, in the already quoted [6]. We
begin this section by showing that the variety of antiortholattices generated by
the algebra D5 coincides with the variety of Kleene-Stone algebras. This fact
explains the similarity of some results independently obtained in [6, 19, 33].
Definition 28 A quasi-Stone algebra is an algebra A = (A, ∧, ∨,∼ , 0, 1) of
type (2, 2, 1, 0, 0) such that (A, ∧, ∨, 0, 1) is a bounded distributive lattice and the
unary operation ∼ satisfies the following conditions for all a, b ∈ A:
QS1
QS2
QS3
0∼ = 1 and 1∼ = 0;
(a ∨ b)∼ = a∼ ∧ b∼;
(a ∧ b∼)∼ = a∼ ∨ b∼∼;
QS4
QS5
a ≤ a∼∼;
a∼ ∨ a∼∼ = 1.
A quasi-Stone algebra A is a Stone algebra if it additionally satisfies SDM.
The following useful lemma contains results to be found in [37] and [6]:
Lemma 29 Let A = (A, ∧, ∨,∼ , 0, 1) be a quasi-Stone algebra. Then:
(i) A satisfies the following conditions for all a, b ∈ A:
QS6
QS7
if a ≤ b, then b∼ ≤ a∼; QS8
a ∧ a∼ = 0;
QS9
a∼∼∼ = a∼;
a ∧ b∼ = 0 iff a ≤ b∼∼.
(ii) The set B (A) = {a∼ : a ∈ A} = {a ∈ A : a = a∼∼} is a Boolean subuni-
verse of A.
Clearly, in case A is a Stone algebra, the condition QS9 can be strengthened
to the pseudocomplementation equivalence:
S1
a ∧ b = 0 iff a ≤ b∼ for all a, b ∈ A.
15
Definition 30 A quasi-Stone De Morgan algebra is an algebra A = (A, ∧, ∨,′ ,
∼, 0, 1) of type (2, 2, 1, 1, 0, 0) such that (A, ∧, ∨,′ , 0, 1) is a De Morgan algebra,
(A, ∧, ∨,∼ , 0, 1) is a quasi-Stone algebra, and a′ ∈ B (A) whenever a ∈ B (A).
If (A, ∧, ∨,′ , 0, 1) is a Kleene algebra and (A, ∧, ∨,∼ , 0, 1) is a (quasi-)Stone
algebra, then A is said to be a Kleene-(quasi-)Stone algebra.
Lemma 31 [6] If A is a quasi-Stone De Morgan algebra, then for all a ∈ A we
have that a∼∼ = a∼′∼′.
Recall from Proposition 23 that the variety generated by the 5-element an-
tiortholattice chain D5 is axiomatised relative to PBZL∗ by the lattice distribu-
tion axiom DIST and the Strong De Morgan law SDM (J0 easily follows from
these assumptions in the context of PBZL∗). We now show that:
Theorem 32 VBZL (D5) coincides with the variety of Kleene-Stone algebras.
Proof. It is readily seen that D5 satisfies all the defining conditions of Kleene-
Stone algebras. Conversely, by the above remark, it will be sufficient to show
that Kleene-Stone algebras satisfy all the axioms of PBZ∗ -- lattices, since they
are clearly distributive as lattices and satisfy SDM by definition. We confine
ourselves to the sole nontrivial items. (i) The condition (∗), (x ∧ x′)∼ = x∼ ∨x′∼
directly follows from SDM. (ii) We show that a∼∼ = a∼′. By QS5, a∼∨a∼∼ = 1,
whence a∼′ ∧ a∼∼′ = 0. By S1, a∼∼′ ≤ a∼′∼, whence, given the fact that
a∼∼ ∈ B (A),
a∼′ ≤(QS4) a∼′∼∼ ≤(QS6) a∼∼′∼ = a∼∼.
From this inequality, QS6 and QS8 we obtain that a∼ = a∼∼∼ ≤ a∼′∼ and thus,
by Lemma 31, a∼∼ = a∼′∼′ ≤ a∼′. The converse inequality follows from S1 and
the fact that a∼ ∈ B (A). (iii) To round up our proof, it will suffice to show
that any Kleene algebra is paraorthomodular. Thus, let a ≤ b and a′ ∧ b = 0.
Then a′ ∧ a ≤ a′ ∧ b = 0, whence a is sharp and thus a ∨ a′ = 1. As a ∧ b = a
and a′ ∧ b = 0, distributivity implies that
a = (a ∧ b) ∨ (a′ ∧ b) = (a ∨ a′) ∧ b = 1 ∧ b = b.
The question as to whether the distributive subvariety DIST of VBZL (AOL)
coincides with the variety of Kleene-quasi-Stone algebras is of a certain interest.
The next Example answers this problem in the negative.
Example 33 The BZ-lattice BZ4 (see [18, Figure 5]) is a Kleene-quasi-Stone
algebra, yet it is not even a member of PBZL∗. In fact, call a and a′ its two
atoms. We have that:
(a ∧ a′)∼ = 0∼ = 1 6= 0 = a∼ ∨ a′∼.
Finally, we prove that the variety generated by the 3-element antiortholattice
chain D3 is a discriminator variety [40].
Proposition 34 VBZL (D3) is a discriminator variety.
16
Proof. Clearly, it suffices to find a ternary term that realises the discriminator
function on D3. Let first
e (x, y) = (x∼ ∧ ♦y) ∨ (y∼ ∧ ♦x) ∨ ((cid:3)x ∧ ((cid:3)y)∼) ∨ ((cid:3)y ∧ ((cid:3)x)∼) .
It is a routinary matter to check that for all a, b ∈ D3, eD3 (a, a) = 0 and
eD3 (a, b) = 1 if a 6= b. It follows that
t (x, y, z) = (e (x, y) ∨ z) ∧ (cid:0)e (x, y)′ ∨ x(cid:1)
realises the discriminator function on D3.
Observe that the algebra D3 fails to be primal, because it has the nontrivial
proper subuniverse {0, 1}. Nonetheless, upon identifying D3 with the set of
rational numbers (cid:8)0, 1
2 , 1(cid:9), the truncated sum operation is definable as follows:
x ⊕ y = min (1, x + y) = (x ∨ ♦y) ∧ (y ∨ ♦x) .
It is easy to check that, upon expanding its signature by this binary operation,
D3 becomes an instance of a De Morgan Brouwer-Zadeh MV-algebra [10, 11]
and, therefore, generates a subvariety of such. The interest of this remark lies in
the fact that the variety of De Morgan Brouwer-Zadeh MV-algebras is known to
be term-equivalent to other well-known varieties of algebras of logic, including
Heyting-Wajsberg algebras, Stonean MV-algebras and MV algebras with Baaz
Delta [9]. In the next section, we will see that VBZL (D3) is term-equivalent to
another well-known variety of algebras of logic.
4.2 Modal Algebras
The standard examples of modal algebras (monadic algebras or interior alge-
bras, to name a few examples) were devised as the algebraic counterparts of
normal modal logics, which are extensions of classical propositional logic --
therefore, they all have a Boolean algebra reduct. There is a thriving litera-
ture, however, on "nonstandard" modal algebras based on generic De Morgan
algebras: see below for the appropriate references. The aim of this section is
to chart this area of research and locate term-equivalent counterparts of some
distributive subvarieties of PBZ∗ -- lattices on this map. We consider algebras
M = (M, ∧, ∨,′ , ♦, 0, 1) of type (2, 2, 1, 1, 0, 0), where (M, ∧, ∨,′ , 0, 1) is a De
Morgan algebra. We assume that ′ binds stronger than ♦, to reduce the number
of parentheses. The following list of identities will be crucial for defining the
varieties that follow; henceforth, (cid:3)x is short for (♦x′)′.
M1 ♦0 ≈ 0
M2 ♦ (x ∨ y) ≈ ♦x ∨ ♦y
M3 x ≤ ♦x
M4 ♦x ≈ ♦♦x
M5 ♦x ∧ (♦x)′ ≈ 0
M6 ♦x ≈ (cid:3)♦x
17
M7 ♦ (x ∧ x′) ≈ ♦x ∧ ♦x′
M8 x′ ∨ ♦x ≈ 1
M9 ♦ (x ∧ y) ≈ ♦x ∧ ♦y
M10 x ∧ x′ ≈ ♦x ∧ x′
Definition 35
(i) A ♦-De Morgan algebra is an algebra M = (M, ∧, ∨,′ , ♦,
0, 1) of type (2, 2, 1, 1, 0, 0), where (M, ∧, ∨,′ , 0, 1) is a De Morgan algebra
and the identities M1 and M2 are satisfied.
(ii) A topological quasi-Boolean algebra is a ♦-De Morgan algebra satisfying
the identities M3 and M4.
(iii) A classical ♦-De Morgan algebra is a topological quasi-Boolean algebra
satisfying the identity M5.
(iv) A monadic De Morgan algebra is a classical ♦-De Morgan algebra satis-
fying the identity M6.
♦-De Morgan algebras and classical ♦-De Morgan algebras were introduced
in dual form by Sergio Celani [13, pp. 253-254]. Topological quasi-Boolean
algebras were first investigated by Banerjee and Chakraborty in the context
of the theory of rough sets [5]. The authors of [36] also introduce, under the
label of topological quasi-Boolean algebras 5, a subvariety of topological quasi-
Boolean algebras that satisfy M6 but not M5. Clearly, topological quasi-Boolean
algebras are meant to be a nonclassical counterpart of interior algebras, while
monadic De Morgan algebras can be viewed as a nonclassical counterpart of
monadic algebras. Condition M5, which is of course trivial once our algebras
have a Boolean nonmodal reduct, is there to restore the Boolean behaviour of
the nonmodal operators, when applied to arguments of the form ♦x. Observe
that all classical ♦-De Morgan algebras satisfy the identity M8 [13, Lemma 2.3].
There are several ways to strengthen the defining conditions of classical ♦-
De Morgan algebras with an eye to obtaining varieties with more interesting
properties.
(i) A possible avenue is to impose on the possibility operator properties that
would determine a collapse of modality when the underlying structures
are Boolean algebras. For example, tetravalent modal algebras [32, 29] are
classical ♦-De Morgan algebras that satisfy M10, although they are usually
presented in a streamlined axiomatisation containing only the axioms for
De Morgan algebras plus M8 and M10. They form a discriminator variety,
generated by a quasiprimal four-element algebra (see item (iv) of the proof
of Theorem 40 below).
(ii) On the other hand, one can enforce what Cattaneo et al.
[8] call a "de-
viant" behaviour of the possibility operator, requesting that it distribute
not only over joins, but over meets as well. Involutive Stone algebras ([14];
cp. also [8], where these structures are called MDS5-algebras), thus, are
classical ♦-De Morgan algebras satisfying M9. It is known that both in-
volutive Stone algebras and tetravalent modal algebras are monadic De
Morgan algebras: see [14] and [16, Proposition 1.2], respectively.
18
We now introduce the modal analogue of distributive PBZ∗ -- lattices.
Definition 36 A weak Lukasiewicz algebra is a classical ♦-De Morgan algebra
M = (M, ∧, ∨,′ , ♦, 0, 1) such that its ♦-free reduct is a Kleene algebra and the
identity M7 is satisfied.
Theorem 37
(i) Every weak Lukasiewicz algebra M is a monadic De Mor-
gan algebra.
(ii) The variety of weak Lukasiewicz algebras is term-equivalent to DIST.
Proof. (i) Let a ∈ M . Using M1, M5, M7 and M4, we have that
0 = ♦0 = ♦(cid:0)♦a ∧ (♦a)′(cid:1) = ♦♦a ∧ ♦(cid:0)(♦a)′(cid:1) = ♦a ∧ ♦(cid:0)(♦a)′(cid:1) .
Thus (♦a)′ ∨ (cid:3)♦a = 1, whence, by M5,
♦a = ♦a ∧ (cid:0)(♦a)′ ∨ (cid:3)♦a(cid:1) = (cid:0)♦a ∧ (♦a)′(cid:1) ∨ (♦a ∧ (cid:3)♦a) = ♦a ∧ (cid:3)♦a.
Consequently, ♦a ≤ (cid:3)♦a. The converse inequality follows from M3.
(ii) Let M = (cid:0)M, ∧, ∨,′ , ♦M, 0, 1(cid:1) be a weak Lukasiewicz algebra. We define
f (M) as the algebra (cid:0)M, ∧, ∨,′ ,∼f (M) , 0, 1(cid:1), where for all a ∈ M , a∼f (M) =
(cid:0)♦Ma(cid:1)′
. Conversely, given a distributive PBZ∗ -- lattice L = (cid:0)L, ∧, ∨,′ ,∼L , 0, 1(cid:1),
we define g (L) as the algebra (cid:0)L, ∧, ∨,′ , ♦g(L), 0, 1(cid:1), where for all a ∈ L, ♦g(L)a =
a∼L∼L. Clearly, f (M) has a Kleene lattice reduct. If a ∈ M , then a ∧ a∼f (M) =
a ∧ (cid:0)♦Ma(cid:1)′
≤ ♦Ma ∧ (cid:0)♦Ma(cid:1)′
= 0, by M3 and M5. Moreover,
a∼f (M)∼f (M) = (cid:16)♦M (cid:0)♦Ma(cid:1)′(cid:17)′
= ♦Ma ≥ a,
by M3 and item (1). For the same reason, a∼f (M)′ = (cid:0)♦Ma(cid:1)′′
a∼f (M)∼f (M). Finally, by M2, whenever a ≤ b,
= ♦Ma =
♦Mb = ♦M (a ∨ b) = ♦Ma ∨ ♦Mb,
i.e. ♦Ma ≤ ♦Mb, whence b∼f (M) = (cid:0)♦Mb(cid:1)′
≤ a∼f (M). In sum, f (M)
is a distributive BZ-lattice. Condition (∗) holds because of M7. Similarly, by
reverse-engineering g (L), it is not hard to show that it is a weak Lukasiewicz
algebra. To round off the proof, observe that for a ∈ L,
≤ (cid:0)♦Ma(cid:1)′
a∼f (g(L)) = (cid:16)♦g(L)a(cid:17)′
♦g(f (M))a = a∼f (M)∼f (M) = (cid:16)♦M (cid:0)♦Ma(cid:1)′(cid:17)′
= a∼L∼L′ = a∼L∼L∼L = a∼L,
= ♦Ma.
Thus, f and g are mutually inverse functions.
Similar term-equivalence results with subvarieties of PBZL∗ are obtained in
[10] and [12] for two special subvarieties of weak Lukasiewicz algebras.
Definition 38
(i) [10, Definition 4.2] A Lukasiewicz algebra is a weak
Lukasiewicz algebra that satisfies the identity M9.
19
(ii) [31] A three-valued Lukasiewicz algebra is a Lukasiewicz algebra that sat-
isfies the identity M10.
Clearly, Lukasiewicz algebras are exactly the involutive Stone algebras whose
♦-free reduct is a Kleene lattice. There is a burgeoning literature on three-valued
Lukasiewicz algebras, see e.g.
[1, 31, 30]. Three-valued Lukasiewicz algebras
can be equivalently characterised as tetravalent modal algebras satisfying M9,
in which case, the Kleene identity follows from the axioms. They are also called
pre-rough algebras in the literature [36].
Theorem 39 [10, Theorems 4.3 and 5.7]
(i) The variety of Lukasiewicz algebras is term-equivalent to VBZL (D5).
(ii) The variety of three-valued Lukasiewicz algebras is term-equivalent to
VBZL (D3).
Taking into account the remarks at the end of last section, it is evident that
VBZL (D5) and VBZL (D3) have repeatedly resurfaced in many different incarna-
tions, with different choices of primitives or with different axiomatisations. We
collect many of the observations made thus far in the following result.
Theorem 40 The strict inclusions and incomparabilities depicted in the fol-
lowing diagram all hold:
♦ -- De Morgan algebras
topological quasi -- Boolean algebras
classical ♦ -- De Morgan algebras
involutive
Stone algebras
❅❅
monadic De Morgan algebras
❵❵❵❵❵❵❵❵❵
✏✏✏✏✏
PPP
three -- valued Lukasiewicz algebras ✦✦✦✦✦✦✦✦
weak Lukasiewicz algebras
Lukasiewicz algebras
tetravalent
modal algebras
Proof. All that remains to be proved is that the inclusions are strict and that
the varieties not connected by upward chains are incomparable.
(i) Consider the algebra D2 as a De Morgan algebra, and let ♦0 = ♦1 = 0.
This algebra is a ♦-De Morgan algebra which is not a topological quasi-
Boolean algebra.
(ii) Consider the algebra D3 as a De Morgan algebra, and let ♦x = x for all
x ∈ D3 = {0, a, 1}. This algebra is a topological quasi-Boolean algebra
which is not a classical ♦-De Morgan algebra. In fact, ♦a ∧ (♦a)′ = a 6= 0.
(iii) Consider the algebra D2
2 as a De Morgan algebra with universe {0, a, a′, 1},
and let ♦x = x for all x ∈ {0, a, 1}, and ♦a′ = 1. This algebra is a topo-
logical quasi-Boolean algebra which is not a monadic De Morgan algebra.
In fact, (cid:3)♦a = 0 6= a = ♦a.
20
(iv) Let B4 be the four-element algebra on {0, a, b, 1} that generates De Mor-
gan algebras, with a = a′ and b = b′. Let ♦0 = 0 and ♦x = 1 for all x 6= 0.
This is a tetravalent modal algebra (actually, it generates this variety),
hence a monadic De Morgan algebra, but not an involutive Stone algebra.
In fact, ♦ (a ∧ b) = 0 6= 1 = ♦a ∧ ♦b. Having two fixpoints for the invo-
lution, it also fails to be a weak Lukasiewicz algebra, hence a Lukasiewicz
algebra or a three-valued Lukasiewicz algebra.
(v) Consider the algebra D2
2 as a De Morgan algebra with universe {0, a, a′, 1},
and let ♦0 = 0, and ♦x = 1 for all x 6= 0. This algebra is a monadic
De Morgan algebra which is not a tetravalent modal algebra.
In fact,
♦a ∧ a′ = a′ 6= 0 = a ∧ a′.
(vi) Consider the ordinal sum D2⊕ B4 ⊕ D2 as a De Morgan algebra with
universe {0, a, b, c, a′, 1}, with b = b′ and c = c′, and let ♦0 = 0, and
♦x = 1 for all x 6= 0. This algebra is an involutive Stone algebra which
is not a weak Lukasiewicz algebra (or a Lukasiewicz algebra) since it has
two fixpoints for the involution.
(vii) Consider the ordinal sum D2
2 ⊕ D2
2 as a De Morgan algebra on {0, a, b, c,
b′, a′, 1}, with c = c′, and let ♦0 = 0, and ♦x = 1 for all x 6= 0. This
is a weak Lukasiewicz algebra which is not an involutive Stone algebra,
for ♦ (a ∧ b) = 0 6= 1 = ♦a ∧ ♦b. A fortiori, it fails to be a Lukasiewicz
algebra.
(viii) Finally, consider the algebra D4 as a De Morgan algebra on {0, a, a′, 1},
and let ♦0 = 0, and ♦x = 1 for all x 6= 0. This is a Lukasiewicz algebra,
hence both an involutive Stone algebra and a weak Lukasiewicz algebra.
However, it fails to be a tetravalent modal algebra (hence a three-valued
Lukasiewicz algebra), for ♦a ∧ a′ = a′ 6= a = a ∧ a′.
Acknowledgements
This work was supported by the research grants "Propriet`a d'ordine nella seman-
tica algebrica delle logiche non classiche", Regione Autonoma della Sardegna,
L. R. 7/2007, n. 7, 2015, CUP: F72F16002920002, and "Theory and appli-
cations of resource sensitive logics", PRIN 2017, Prot. 20173WKCM5, CUP:
F74I19000720001.
The authors thank Davide Fazio for the insightful discussions on the topics
of the present paper.
References
[1] Abad M., Figallo M., "Characterization of three-valued Lukasiewicz alge-
bras", Reports on Mathematical Logic,18, 1985, pp. 47-59.
[2] Alizadeh M., Ardeshir M., "Amalgamation property for the class of basic
algebras and some of its natural subclasses", Archive for Mathematical
Logic, 45, 8, 2006, pp. 913-930.
21
[3] Ardeshir M., Ruitenburg W., "Basic propositional calculus I", Mathemati-
cal Logic Quarterly, 44, 3, 1998, pp. 317-343.
[4] Ardeshir M., Ruitenburg W., "Basic propositional calculus II. Interpola-
tion", Archive for Mathematical Logic, 40, 5, 2001, pp. 349-364.
[5] Banerjee M., Chakraborty M.K., "Rough algebra", Bull. Polish Acad. Sci.
(Math.), 41, 4, 1993, pp. 293 -- 297.
[6] Blyth T.S., Fang J., Wang L., "De Morgan algebras with a quasi-Stone
operator", Studia Logica, 103, 2015, pp. 75-90.
[7] Bruns G., Harding J., "Algebraic aspects of orthomodular lattices", in
B. Coecke et al. (Eds.), Current Research in Operational Quantum Logic,
Springer, Berlin, 2000, pp 37 -- 65.
[8] Cattaneo G., Ciucci D., Dubois D., "Algebraic models of deviant modal
operators based on De Morgan and Kleene lattices", Information Sciences,
181, 2011, pp. 4075 -- 4100.
[9] Cattaneo G., Ciucci D., Giuntini R., Konig M., "Algebraic structures re-
lated to many valued logical systems. Part II: Equivalence among some
widespread structures", Fundamenta Informaticae, 63, 2004, pp. 357 -- 373.
[10] Cattaneo G., Dalla Chiara M. L., Giuntini R., "Some algebraic structures
for many-valued logics", Tatra Mountains Mathematical Publications, 15,
1998, pp. 173 -- 196.
[11] Cattaneo G., Giuntini R., Pilla R., "BZMV and Stonian MV algebras (ap-
plications to fuzzy sets and rough approximations)", Fuzzy Sets and Sys-
tems, 108, 1999, pp. 201 -- 222.
[12] Cattaneo G., Nistic`o G., "Brouwer-Zadeh posets and three-valued
Lukasiewicz posets", Fuzzy Sets and Systems, 33, 2, 1989, pp. 165-190.
[13] Celani S.A., "Classical modal De Morgan algebras", Studia Logica, 98,
2011, pp. 251-266.
[14] Cignoli R., Gallego M.S., "Dualities for some De Morgan algebras with op-
erators and Lukasiewicz algebras", Journal of the Australian Mathematical
Society (Series A), 34, 1983, pp. 377-393.
[15] de Groote H.F., "On a canonical lattice structure on the effect algebra of
a von Neumann algebra", arXiv:math-ph/0410018v2, 2005.
[16] Font J.M., Rius M., "An abstract algebraic logic approach to tetravalent
modal logics", Journal of Symbolic Logic, 65, 2, 2000, pp. 481 -- 518.
[17] Gait`an H., "Priestley duality for quasi-Stone algebras", Studia Logica, 64,
2000, pp. 83-92.
[18] Giuntini R., Ledda A., Paoli F., "A new view of effects in a Hilbert space",
Studia Logica, 104, 2016, pp. 1145-1177.
22
[19] Giuntini R., Ledda A., Paoli F., "On some properties of PBZ∗ -- lattices",
International Journal of Theoretical Physics, 56, 12, 2017, pp 3895 -- 3911.
[20] Giuntini R., Mure¸san C., Paoli F., "PBZ∗ -- lattices: Structure theory and
subvarieties", Reports on Mathematical Logic, forthcoming.
[21] Giuntini R., Mure¸san C., Paoli F., "PBZ∗ -- lattices: Ordinal and horizontal
sums", arXiv:1811.01869v2 [math.RA].
[22] Gratzer G., General Lattice Theory, Birkhauser Akademie -- Verlag, Basel --
Boston -- Berlin, 1978.
[23] Gratzer G., Universal Algebra, Second Edition, Springer Science+Business
Media, LLC, New York, 2008.
[24] Gratzer G., Lakser H., P lonka J., "Joins and direct products of equational
classes", Canadian Mathematical Bulletin, 12, 1969, pp. 741-744.
[25] Guzm`an F., Squier C.C., "Subdirectly irreducible and free Kleene-Stone
algebras", Algebra Universalis, 31, 1994, pp. 266-273.
[26] Kowalski T., Paoli F., "Joins and subdirect products of varieties", Algebra
Universalis, 65, 4, 2011, pp. 371-391.
[27] Kowalski T., Ledda A., Paoli F., "On independent varieties and some re-
lated notions", Algebra Universalis, 70, 2013, pp. 107-136.
[28] Kreyszig E., Introductory Functional Analysis with Applications, Wiley,
New York, 1978.
[29] Loureiro I., Algebras Modais Tetravalentes, Ph.D. Thesis, Facultade de
Ciencias de Lisboa, 1983.
[30] Moisil G.C., "Le alg`ebre de Lukasiewicz", Acta Logica (Bucharest), 6, 1963,
pp. 97-135.
[31] Monteiro L., "Sur la d´efinition des algebres de Lukasiewicz trivalentes",
Bulletin de la Societ´e des Sciences Math´ematiques et Physiques de la
Roumanie, Nouvelle Serie, 7, 1963, pp. 3-12.
[32] Monteiro L., "Axiomes ind´ependants pour les alg´ebres de Lukasiewicz triva-
lentes", Bulletin de la Societ´e des Sciences Math´ematiques et Physiques de
la Roumanie, Nouvelle Serie, 7, 1963, pp. 199-202.
[33] Mure¸san C., A Note on Direct Products, Subreducts and Subvarieties
of PBZ∗ -- lattices, Mathematica Slovaca, forthcoming, arXiv:1904.10093v3
[math.RA].
[34] Olson M.P., "The self-adjoint operators of a von Neumann algebra form a
conditionally complete lattice", Proceedings of the American Mathematical
Society, 28, 1971, pp. 537-544.
[35] P lonka J., "A note on the join and subdirect product of equational classes",
Algebra Universalis, 1, 1971, pp. 163-164.
23
[36] Saha A., Sen J., Chakraborty M.K., "Algebraic structures in the vicinity
of pre-rough algebra and their logics", Information Sciences, 282, 2014, pp.
296 -- 320.
[37] Sankappanavar N.H., Sankappanavar H.P., "Quasi-Stone algebras", Math-
ematical Logic Quarterly, 39, 1993, pp. 255-268.
[38] Stroock D.W., A Concise Introduction to the Theory of Integration (3rd
ed.), Birkhauser, Basel, 1998.
[39] Visser A., "A propositional logic with explicit fixed points", Studia Logica,
40, 1981, pp. 155 -- 175.
[40] Werner H., Discriminator Algebras, Studien zur Algebra und ihre Anwen-
dungen, Band 6, Akademie-Verlag, Berlin, 1978.
24
|
1707.07052 | 1 | 1707 | 2017-07-21T21:44:50 | Well-closed subschemes of noncommutative schemes | [
"math.RA",
"math.CT",
"math.QA"
] | Van den Bergh has defined the blowup of a noncommutative surface at a point lying on a commutative divisor. We study one aspect of the construction, with an eventual aim of defining more general kinds of noncommutative blowups. Our basic object of study is a quasi-scheme X (a Grothendieck category). Given a closed subcategory Z, in order to define a blowup of X along Z one first needs to have a functor F which is an analog of tensoring with the defining ideal of Z. Following Van den Bergh, a closed subcategory Z which has such a functor is called well-closed. We show that well-closedness can be characterized by the existence of certain projective effacements for each object of X, and that the needed functor F has an explicit description in terms of such effacements. As an application, we prove that closed points are well-closed in quite general quasi-schemes. | math.RA | math |
WELL-CLOSED SUBSCHEMES OF NONCOMMUTATIVE SCHEMES
D. ROGALSKI
Abstract. Van den Bergh has defined the blowup of a noncommutative surface at a point lying on a
commutative divisor [VdB]. We study one aspect of the construction, with an eventual aim of defining more
general kinds of noncommutative blowups. Our basic object of study is a quasi-scheme X (a Grothendieck
category). Given a closed subcategory Z, in order to define a blowup of X along Z one first needs to have
a functor FZ which is an analog of tensoring with the defining ideal of Z. Following Van den Bergh, a
closed subcategory Z which has such a functor is called well-closed. We show that well-closedness can be
characterized by the existence of certain projective effacements for each object of X, and that the needed
functor FZ has an explicit description in terms of such effacements. As an application, we prove that closed
points are well-closed in quite general quasi-schemes.
1. Introduction
This paper is the first part of a bigger project to study further the method of noncommutative blowing
up developed by Van den Bergh in the monograph [VdB]. Van den Bergh gives a description of the blowup
of a noncommutative surface at a point lying on a commutative divisor on the surface, and shows that this
construction has many good properties, similar to those of a usual commutative blowup. Our overall aim is
both to make some aspects of Van den Bergh's procedure more explicit, as well as to show that the same
ideas apply in a broader setting, which would allow one to define blowups of more general subschemes of
noncommutative schemes. In this paper, we focus primarily on a categorical notion which is a fundamental
building block for Van den Bergh's blowing up machinery: the well-closedness of a closed subscheme of a
noncommutative scheme. Roughly speaking, this is a condition that one has a right exact functor which is
an analog of "tensoring with the ideal sheaf defining the closed subscheme".
Following [VdB], we take as our main object of study in noncommutative geometry a quasi-scheme, in
other words a Grothendieck category X. A Grothendieck category is an abelian category with exact direct
limits and a generator. We review some background on such categories in Section 2. We often assume
in addition that X is locally noetherian, in other words that X has a set of noetherian generators. Let
k be a field.
Important examples of quasi-schemes include X = Mod- A, the category of right modules
over a noetherian k-algebra A, and X = Qcoh Y , the category of quasi-coherent sheaves on a noetherian
(commutative) k-scheme Y . We are especially interested in applications to noncommutative projective
geometry. The most fundamental kind of quasi-scheme in this setting is X = Qgr-A, the quotient category
of right Z-graded modules over a connected N-graded finitely generated noetherian k-algebra A, modulo
2010 Mathematics Subject Classification. 18E15, 18A40, 14A22.
Key words and phrases. Grothendieck category, noncommutative blowing up, adjoint functors, locally noetherian, closed
subcategory.
The author was partially supported by the NSF grant DMS-1201572 and the NSA grant H98230-15-1-0317.
1
the subcategory of modules which are direct limits of finite-dimensional modules. We call such an X a
noncommutative projective scheme. By a result of Serre, when A is commutative and generated by its
degree 1 elements, then Qgr-A is equivalent to the category of quasi-coherent sheaves on the scheme Proj A.
Many important constructions in commutative algebraic geometry can be understood purely in terms of the
category of quasi-coherent sheaves, and this justifies the study of Grothendieck categories as a replacement
for schemes in the noncommutative case, where constructions involving actual spaces and sheaves on them
are often not available.
For example, it is straightforward to find a reasonable definition of a closed subscheme Z of a quasi-
scheme X. This is the notion of a closed subcategory, namely, a full abelian subcategory Z of X such that
the inclusion functor i : Z → X has both a left and right adjoint. In most cases, this is equivalent to Z
being closed under subquotients, direct sums, and products. As evidence that this is the right definition, the
closed subcategories of an affine quasi-scheme X = Mod- A are precisely the categories Mod- A/I for 2-sided
ideals I of A [Ros, Proposition 6.4.1], and the closed subcategories of a category Qcoh Y of quasi-coherent
sheaves on a quasi-projective k-scheme Y are the categories Qcoh W for closed subschemes W of Y [Sm1,
Theorem 4.1]. Smith works out many basic properties of closed subcategories of quasi-schemes in [Sm1]. In
particular, he defines the analogs of closed points in quasi-schemes X: a closed point is a closed subcategory
Z of X which is equivalent to Mod- D for some division ring D. Assuming that X is locally noetherian,
this has a more intrinsic description as follows. A simple object P in a locally noetherian quasi-scheme X
is called tiny if HomX (M, P ) is a finitely generated EndX (P )-module for all noetherian objects M . When
P is tiny, the subcategory Z consisting of all direct sums of P is a closed point, where Z ≃ Mod- D for
D = EndX (P ). Conversely, all closed points are of this form [Sm1, Theorem 5.5].
Since blowing up is such an important construction in commutative algebraic geometry, it is essential
to have some analog of this in the noncommutative case. Recall that if Y is a commutative scheme with
closed subscheme W defined by a sheaf of ideals I, then the blowup of Y along W is defined to be the
relative Proj of a sheaf of graded algebras, namely BlW Y = Proj(OY ⊕ I ⊕ I 2 ⊕ . . . ). Trying to mimic this
definition for more general quasi-schemes X, one runs into immediate problems. Though we know what a
closed subscheme of X should be, finding replacements for the defining ideal sheaf of Z, what it means to
take the powers of this ideal sheaf in order to define the Rees ring, and what the analog of the relative Proj
construction should be, are major difficulties.
Van den Bergh's elegant solution is to work in a category of functors. Let Z be a closed subcategory of a
quasi-scheme X. While in general there is no object of X that plays the role of an ideal sheaf defining Z, in
nice cases one can find a right exact functor which plays the role of tensoring with the ideal sheaf. In more
detail, there is a left exact functor oZ : X → X which takes an object M to the largest subobject of M in
Z. Because Grothendieck categories have enough injectives, one can show that the category L(X, X) of left
exact functors X → X is an abelian category, and so there is some left exact functor GZ : X → X which fits
into an exact sequence of left exact functors 0 → oZ → oX → GZ → 0, where oX : X → X is the identity
functor. For example, when X = Mod- A for a ring A, then GZ = Hom(I, −), where Z = Mod- A/I. Thus
in general, we think of GZ as "Hom from the ideal defining Z". Now suppose that GZ has a left adjoint
FZ : X → X. For example, if X = Mod- A, then FZ = − ⊗ I. Thus in general we think of FZ (when it
2
exists) as the required analog of "tensoring with the ideal defining Z". Following Van den Bergh, we say
that Z is a well-closed subcategory of X if it is a closed subcategory and the functor GZ defined above has
a left adjoint FZ .
Showing that Z is well-closed is the first step in trying to define a blowup of a quasi-scheme X along a
closed subscheme Z. There are already a lot of interesting and nontrivial questions about this step. Thus
in this paper, our main goal is to study the property of well-closedness of subcategories Z of quasi-schemes
and the properties of the functors FZ, in order to lay groundwork for a more general theory of blowing
up. The two main questions we address in this paper are the following. First, if Z is well-closed, does
the functor FZ have a more explicit description, other than just being the adjoint of GZ? Second, which
closed subcategories Z of quasi-schemes are well-closed? To give more context to the first question, in [VdB],
when Van den Bergh shows that certain categories are well-closed, the argument relies ultimately on Freyd's
adjoint functor theorem, which gives a criterion for when a left exact functor has a left adjoint. The adjoint
functor theorem is very abstract, and it seems difficult to see from its proof what the left adjoint it finds
actually does to objects and morphisms.
We will give an answer to the first question above that works in wide generality. We first show in
Lemma 3.5 below that the left exact functor GZ can be given the following explicit description. If i : Z → X
is the inclusion functor, its respective right and left adjoints i! : X → Z and i∗ : X → Z are given explicitly
as follows: i!(N ) is the unique largest subobject of N which is in Z, and i∗(N ) is the unique largest factor
object of N which is in Z. Now given an object M ∈ X, GZ(M ) = M /i!(M ), where M /M = i!(E(M )/M )
for an injective hull E(M ) of M . Thus GZ first extends M by the largest possible essential extension by
an object in Z, and then mods out by the largest subobject in Z. The action of GZ on morphisms can be
defined similarly; any morphism M → N extends (non-uniquely) to a morphism M → N , which induces a
morphism of the factor objects GZ(M ) → GZ(N ) (which does not depend on the choice of extension).
The left adjoint FZ of GZ , when it exists, can be described in a roughly dual way to the explicit description
of GZ just given, even though X does not have enough projectives in general, much less projective covers.
Given a collection S of objects in X, an S-projective effacement of an object M is an epimorphism π : M → M
satisfying the following lifting property: given any epimorphism f : P → M with kernel in S, there exists
g : M → P such that f g = π. We now give our first main result.
Theorem 1.1. (Theorem 5.1.) Let X be a locally noetherian Grothendieck category, let Z be a closed
subcategory of X, and let S be the collection of all objects in Z which are injective in the category Z. Let
0 → oZ → oX → GZ → 0 be the exact sequence in the category L(X, X) as above. Then the following are
equivalent:
(1) GZ has a left adjoint FZ ; that is, Z is well-closed in X.
(2) The natural map Ext1(M,Qα Nα) → Qα Ext1(M, Nα) is an isomorphism, for all small collections
{Nα} of objects in S and for all M ∈ X.
(3) Every object M in X has an S-projective effacement.
Moreover, when the conditions in the theorem hold, then FZ can be explicitly constructed. On objects,
FZ (M ) = KZ(M ), where M → M is a fixed S-projective effacement of M and KZ is the functor which
takes an object N to its unique smallest subobject N ′ such that N/N ′ ∈ Z. To define the action of FZ on
3
morphisms, given a morphism M → N , it lifts (non-uniquely) to a morphism M → N , which restricts to a
morphism GZ (M ) → GZ (N ) (which does not depend on the choice of lift).
Van den Bergh also defines a stronger condition on a closed subcategory Z called very well-closed, which
is equivalent to Z being well-closed and the category Z having exact direct products [VdB, Corollary 3.4.11].
Only very special closed subcategories should be expected to satisfy this stronger condition, but it is quite
useful when it holds. Our second main result is a characterization of very well-closedness similar to Theo-
rem 1.1.
Theorem 1.2. (Theorem 5.3.) Let X be a locally noetherian Grothendieck category and let Z be a closed
subcategory of X. Then the following are equivalent:
(1) Z is well-closed in X and the category Z has exact direct products; that is, Z is very well-closed in
X.
(2) The natural map Ext1(M,Qα Nα) → Qα Ext1(M, Nα) is an isomorphism, for all small collections
{Nα} of objects in Z and for all M ∈ X.
(3) Every object M in X has a Z-projective effacement.
Our description of the functors FZ using projective effacements is helpful in giving a partial answer to
the second question, concerning which closed subcategories are well-closed, in the important special case of
closed points.
Theorem 1.3. (Theorem 6.6.) Let X be a locally noetherian Grothendieck k-category. Suppose that Z is a
closed point in X, that is, Z is the category of direct sums of a tiny simple object P in X. Suppose further
that dimk Ext1
X (M, P ) < ∞ for all noetherian objects M ∈ X. Then Z is very well-closed in X.
The necessary hypothesis that dimk Ext1
X (M, P ) < ∞ for noetherian objects M is automatic in many cases
of interest, in particular for nice noncommutative projective schemes. For example, if A is a connected
graded noetherian algebra satisfying the Artin-Zhang χ1 condition, A is generated in degree 1 as an algebra,
and M is a point module for A, then P = π(M ) ∈ X = Qgr-A is a tiny simple satisfying the hypothesis
above, so the corresponding closed point Z is very well-closed in X. See Section 6 for more details.
The structure of the paper is as follows. In Section 2 we review the basic theory of Grothendieck categories
and their closed subcategories. In Section 3, we work in a general abelian category and study S-projective
effacements, as well as the dual concept of T -injective effacements, and show how they may be used to define
adjoint pairs of functors. Then in Section 4 we specialize to a Grothendieck category, and prove some results
about projective effacements in that particular setting. Section 5 contains the proofs of Theorem 1.1 and
Theorem 1.2. Finally, in Section 6 we show how the theory of the paper works out in specific examples of
quasi-schemes, especially noncommutative projective schemes, and give the proof of Theorem 1.3.
To conclude the introduction, we briefly describe our work in progress that continues the ideas of this
paper. First, we plan to apply the theory of this paper to define the blowing up of a quasi-scheme along a
well-closed subcategory in general, and study some of its properties. This blowup should generalize both Van
den Bergh's blowups of surfaces at points on commutative divisors, as well as the naive blowups defined by
Keeler, Stafford, and the author. As a corollary, we plan to show carefully that the coordinate rings of Van
den Bergh's del Pezzo surfaces in [VdB] are the same as the subrings of the Sklyanin algebra defined in [Ro].
4
An interesting new case we plan to study is the blowup at a point of the category Qgr-S of a 4-dimensional
Sklyanin algebra S. We also want to study in more detail the structure of closed subcategories of Gr-A and
Qgr-A, with hope that this will allow us to prove that more general closed subcategories of noncommutative
projective schemes Qgr-A are well-closed.
We thank Michel Van den Bergh and Paul Smith for helpful conversations.
acknowledgments
2. Categorical Preliminaries
In this section we review some of the basic category theory we need in this paper. The reader looking
for more extensive background about the foundations of abelian categories may consult [Mac]. Many of the
standard facts about Grothendieck categories can be found in [Ga] and [Po], and basic material on closed
subcategories is given in [Sm1].
We adopt standard set-theoretic foundations for categories X in terms of Grothendieck universes. See,
for example, [Mac, Section I.6]. We assume a large enough universe U (a set of sets), where the sets in U
are called small, and assume that HomX (M, N ) is a small set for any objects M, N ∈ X. Direct sums and
products are indexed only over small sets. The universe U is chosen so that given a union T = Sα∈A Sα,
where the index set A is small and each set Sα is small, then T is small. Also, the power set of a set in U
is again in U . The set of all objects in the category X is not small in general, however. We say that X is
well-powered if the set of subobjects of any given object in X is small. Occasionally we will wish to assume
that X is a k-category for some field k: this means that all Hom-sets HomX (M, N ) are k-vector spaces, and
that composition is k-bilinear.
While our main results are proved in the setting of Grothendieck categories, as we define shortly, some
of our results work in the setting of more general abelian categories X. We will always assume at least that
X is an abelian category satisfying Grothendieck's axioms (AB3) and (AB3*), that is, that X is cocomplete
(coproducts of small-indexed families of objects exist in X) and complete (products of small-indexed families
of objects exist in X), respectively. We use the notation Lα Mα and Qα Mα for the respective coproduct
and product of a small set of objects Mα, and we also typically refer to coproducts as direct sums.
Let Z be a full subcategory of a complete, cocomplete, and well-powered abelian category X, and let
iZ = i∗ : Z → X be the inclusion functor. We say that Z is closed under subquotients if every subobject
and quotient object of an object in Z is also in Z. Following Smith [Sm1, Definition 2.4], we say that
Z is weakly closed if Z is closed under subquotients and iZ has a right adjoint i!
Z = i! : X → Z, and
we say that Z is closed if Z is closed under subquotients and iZ has both a right adjoint i!
Z and a left
Z = i∗ : X → Z. (Van den Bergh uses different terminology in [VdB], referring to weakly closed
adjoint i∗
subcategories as closed (following Gabriel), and closed subcategories as biclosed.) It is elementary to see
that if Z is closed under subquotients, then Z is weakly closed if and only if Z is closed under direct sums,
and Z is closed if and only if Z is closed under both direct sums and products [VdB, Proposition 3.4.3].
Moreover, if Z is weakly closed, then the right adjoint i! : X → Z can be described as follows: given M ∈ X,
the object i!(M ) is the unique largest subobject of X which is in Z, and i! acts on morphisms by restriction.
Similarly, if Z is closed, then the left adjoint i∗ acts on objects as i∗(M ) = M/N , where N is the unique
5
smallest subobject of M such that M/N ∈ Z, and i∗ sends a morphism to the induced morphism of quotient
objects. By adjointness, i! is left exact and i∗ is right exact.
Recall that a small set of objects {Oα} is a set of generators for X if for all objects N, P ∈ X and
morphisms f1, f2 ∈ HomX (N, P ) with f1 6= f2, there exists a generator Oα and a map g : Oα → N such
that f1g 6= f2g. If the set of generators consists of a single object O then it is called a generator for X.
Assuming X is cocomplete, it is standard that {Oα} is a set of generators for X if and only if O = Lα Oα
is a generator. It is also easy to see when X is cocomplete that {Oα} is a set of generators if and only if for
every M ∈ X, there is a epimorphism from some direct sum of copies of objects in the set of generators to M
[Po, Proposition 2.8.2]. Additionally, it is a standard fact that if X has a generator then X is well-powered
[Ga, Proposition I.5].
An abelian category X is a Grothendieck category if it satisfies Grothendieck's axiom (AB5) (that is, it is
cocomplete and direct limits of exact sequences are exact), and X has a generator (or equivalently, a set of
generators). A Grothendieck category X automatically has a number of useful other properties: it satisfies
Grothendieck's axiom (AB4) that direct sums of exact sequences are exact (since this is known to be a general
consequence of (AB5) [Po, Corollary 2.8.9]), it is well-powered (since it has a generator), and it is complete
(Grothendieck's (AB3*) [Po, Corollary 3.7.10]). Also, every object has an injective hull; in particular, the
category has enough injectives [Po, Theorem 3.10.10]. However, X need not have enough projectives, and
products of exact sequences need not be exact, that is, the category need not satisfy Grothendieck's axiom
(AB4*). An important example of a locally noetherian Grothendieck category is the category Qcoh Y of
quasi-coherent sheaves on a quasi-projective scheme Y , which typically does not have enough projectives or
exact direct products. We are especially interested in noncommutative projective schemes, as mentioned in
the introduction; in the last section of the paper we will show how the theory we develop applies to these
and other more specific examples.
As usual, an object in an abelian category is noetherian if it has the ascending chain condition on
subobjects. The category X is locally noetherian if X has a (small) set of noetherian generators. We will
assume in our main theorems later that X is a locally noetherian Grothendieck category. Here are some
other important properties of such categories.
Lemma 2.1. Let X be a locally noetherian Grothendieck category with small set of noetherian generators
{Oα}.
(1) Direct limits and direct sums of injective objects in X are injective, and every injective object is a
direct sum of indecomposable injective objects.
(2) Every indecomposable injective of X is isomorphic to an injective hull of some epimorphic image of
a generator Oα; in particular, the set of isomorphism classes of indecomposable injectives in X is
small.
(3) If Z is a closed subcategory of X with inclusion functor i∗ : Z → X, then Z is a locally noetherian
Grothendieck category in its own right, with small set of noetherian generators {i∗(Oα)}.
Proof. (1). See [Po, Theorem 5.8.7, Theorem 5.8.11].
(2). Suppose that I is indecomposable injective. We may choose a nonzero map f : Oα → I for some
generator Oα. Let 0 6= M ⊆ I be the image of f . Since I is indecomposable injective, I = E(M ) must be
6
an injective hull of M . The last statement follows, since by well-poweredness the set of factor objects of a
given object is small, and a small union of small sets is small.
(3). The category Z is clearly abelian since by definition it is closed under subquotients. Since Z is closed
under direct sums and thus also under direct limits, it is clear that any direct limit in X of objects in Z is
also a direct limit in the category Z. Then since X satisfies (AB5), so does Z. Recall that i∗(Oα) is the
unique largest factor object of Oα in Z. Given an object M in Z and a map f : Oα → M , the map factors
through i∗(Oα), and the statement about generators follows.
(cid:3)
Recall from the introduction that a closed point of a locally noetherian Grothendieck category X is a
closed subcategory Z which is equivalent to Mod- D for some division ring D. We say that the locally
noetherian Grothendieck category X is locally finite if X has a set of generators which are both noetherian
and artinian. A closed point Z is a locally finite Grothendieck category. One way to get more locally finite
categories is to join together closed points with the following construction.
Definition 2.2. Given a list of full subcategories Z1, Z2, . . . , Zn of an abelian category X, the Gabriel
product is the full subcategory Z = Z1 · Z2 · . . . · Zn consisting of objects M with filtrations 0 = Mn+1 ⊆
Mn ⊆ · · · ⊆ M1 = M such that Mi/Mi+1 ∈ Zi for all 1 ≤ i ≤ n.
Lemma 2.3. Let X be complete, cocomplete, and well-powered. If Z1, . . . , Zn are closed subcategories of X,
then Z = Z1 · Z2 · . . . · Zn is also closed.
Proof. See [VdB, Proposition 3.3.6].
(cid:3)
Example 2.4. Let Z1, Z2, . . . , Zn be a list of closed points, possibly with repeats, in the locally noetherian
Grothendieck category X. Then the category Z = Z1 · Z2 · . . . · Zn closed subcategory of X by the previous
If {Oα} is a set of noetherian generators for X, then {i∗(Oα)} is a set of generators for Z, by
lemma.
Lemma 2.1. It is clear that a noetherian object in Z has finite length, so each i∗(Oα) has finite length and
Z is a locally finite category.
Conversely, if Z is a locally finite closed subcategory of a locally noetherian Grothendieck category X,
and Z is generated by finitely many objects of finite length whose composition factors are all tiny simples,
then it is easy to see that Z is a full subcategory of some Gabriel product Z1 · Z2 · . . . · Zn for some closed
points Zi.
In any Grothendieck category X, since X has enough injectives, one has Ext groups Exti(M, N ) for
objects M, N ∈ X and all i ≥ 0, calculated using an injective resolution of N . Since we assume that
Hom sets are small, it is clear that Exti(M, N ) is a small set for all M, N and all i ≥ 0. There is also
the usual correspondence between elements of Ext1(M, N ) and equivalence classes of short exact sequences
0 → N → P → M → 0. In the next result, we study how Ext interacts with directs sums and products in
a Grothendieck category X. Given a small index set I, let QI X be the category consisting of I-tuples of
objects in X, and let Q : QI X → X be the functor given by (Mα)α∈I 7→ Qα∈I Mα. The functor Q is only
left exact in general. Since X has enough injectives, so does QI X, and thus we can define right derived
functors RiQ of the product functor.
Lemma 2.5. Let X be a Grothendieck category.
7
(1) Given M and a family of objects {Nα}α∈I in X, the natural map
Ext1(M,Y Nα)
j
→ Y Ext1(M, Nα)
is an isomorphism if R1 Q(Nα) = 0.
(2) Assume that X is locally noetherian. Given a noetherian object M and a directed system of objects
{Nα}α∈I in X, for each i ≥ 0 the natural map
lim−→ Exti(M, Nα) → Exti(M, lim−→ Nα)
is an isomorphism. Similarly, direct sums pull out of the second coordinate of Ext.
(3) Given a family of objects {Mα}α∈I and N in X, for each i ≥ 0 there is an isomorphism
Exti(M Mα, N ) → Y Exti(Mα, N ).
Proof. (1) Note that the natural map arises as follows: For each β there is a projection πβ : Qα Nα → Nβ
and thus by functoriality of Ext a map fβ : Ext1(M,Q Nα) → Ext1(M, Nβ). By the universal property of
the product, the fβ give a map Ext1(M,Q Nα) → Q Ext1(M, Nα) as required.
There is a Grothendieck spectral sequence
Ep,q
2 = Extp
X (M, Rq Qα Nα) =⇒ Qα Extp+q
X (M, Nα)
associated to the composition of the product functor Q : QI X → X and the functor HomX (M, −) [Roos,
Equation (1.2)]. The associated exact sequence of low degree terms begins
0
/ Ext1(M,Qα Nα)
j
/ Qα Ext1(M, Nα)
/ Hom(M, R1 Q Nα)
/ . . .
(2.6)
We claim that the map j in this sequence is the same as the natural map. For fixed β, apply naturality of
the exact sequence (2.6) to the morphism h : (Nα) → (Mα) in QI X, where Mβ = Nβ, Mα = 0 for α 6= β,
and hβ is the identity map. Since Qα Mα ∼= Nβ, one obtains a diagram
Qα Ext1(M, Nα)
/ Ext1(M,Qα Nα)
. . .
0
j
f
g
0
/ Ext1(M, Nβ)
1
/ Ext1(M, Nβ)
/ . . .
where f is the map on Ext induced by the projection Qα Nα → Nβ, and g is the projection onto the βth
coordinate. The commutation of this diagram for all β implies that the map j is the same as the natural
map as claimed. Now the result follows immediately from (2.6).
α → E1
α → lim−→ E1
(2) For each α, let 0 → Nα → E0
α → . . . be an injective resolution of Nα. Since direct limits
of injectives are injective in X by Lemma 2.1(1), and direct limits are exact in a Grothendieck category,
is an injective resolution of lim−→ Nα. Then Exti(M, lim−→ Nα)
0 → lim−→ Nα → lim−→ E0
is the ith homology of the sequence 0 → Hom(M, lim
α) → . . . . Since M is noe-
−→
therian, the natural map lim−→ Hom(M, Pα) → Hom(M, lim−→ Pα) is an isomorphism for any directed sys-
tem {Pα} [Po, Proposition 3.5.10, Exercise 5.8.2]. Thus we can take the ith homology of the sequence
α) → . . . . This is the same as lim−→ Exti(M, Nα) since direct limits
0 → lim−→ Hom(M, E0
are exact in the category of abelian groups. Thus Exti(M, lim
−→
α) → lim−→ Hom(M, E1
α) → Hom(M, lim
−→
Exti(M, Nα) for all i.
Nα) ∼= lim
−→
α → . . .
E0
E1
8
/
/
/
/
/
/
/
/
/
/
/
/
The proof that Exti(M,L Nα) ∼= L Exti(M, Nα) is analogous, since direct sums are exact, and direct
sums of injectives are also injective by Lemma 2.1(1).
(3) This is an easy argument using the isomorphism Hom(L Mα, N ) → Q Hom(Mα, N ) and the fact that
products are exact in the category of abelian groups.
(cid:3)
3. Defining functors via effacements
In this and the next section we define and study some categorical constructions which form the technical
heart of the paper. While our primary interest is in Grothendieck categories, it is more natural to first present
the results in the context of general abelian categories. Then beginning in Section 4, we will specialize to
the case of Grothendieck categories X and study the more special features of that setting.
Let X be an abelian category which is cocomplete, complete, and well-powered. Let Z be a closed
subcategory of X, and let i∗ : Z → X be the inclusion functor. Recall that this means that i∗ has right
adjoint i! : X → Z, where i!(M ) is the unique largest subobject of X which is in Z, and left adjoint
i∗ : X → Z, where i∗(M ) is the unique largest factor object M/N in X such that M/N ∈ Z. Let
oZ = i∗i∗ : X → X and oZ = i∗i! : X → X. In particular, oX = oX : X → X is notation for the identity
functor on X.
It is standard that if C is any category, then since X is abelian, the category Fun(C, X) of functors
from C to X, with morphisms being natural transformations, is also an abelian category, with "pointwise"
operations. (There is a minor set-theoretic issue that the Hom-sets in Fun(C, X) are not necessarily small,
unless C is a small category, that is, the set of objects in C is small. We will allow functor categories to have
Hom-sets in a larger universe than the one we use for our categories X of main interest.) When C is an
additive category, then the subcategory Add(C, X) consisting of additive functors from C → X is an abelian
subcategory of Fun(C, X), again with pointwise operations. There is a morphism η : oX → oZ in Add(X, X),
where ηM : M → i∗i∗(M ) = M/N is the natural quotient map for each M . Then η is an epimorphism in
Add(X, X), and we let K be its kernel. Thus K is an additive functor such that K(M ) = N , where N
is the unique smallest subobject of M such that M/N ∈ Z, and K acts on morphisms by restriction. Let
R(X, X) be the category of right exact functors from X → X.
In general R(X, X) is not abelian, and
although oX and oZ are right exact, K need not be right exact. More specifically, if 0 → N
f
→ M
g
→ P → 0
is a short exact sequence in X, then it is easy to check that K(f ) remains a monomorphism and K(g) an
epimorphism, but 0 → K(N ) → K(M ) → K(P ) → 0 need not be exact in the middle, having homology
(K(M ) ∩ f (N ))/f (K(N )) there.
Dually, there is a monomorphism ρ : oZ → oX in Add(X, X) where ρM : i∗i!(M ) → M is the natural
inclusion, identifying i!(M ) with the largest subobject of M which is in Z. Let C be the cokernel of ρ
in Add(X, X), so C acts on objects by C(M ) = M/i!(M ). Similarly, the category of left exact functors
L(X, X) is not necessarily abelian, and the functor C preserves monomorphisms and epimorphisms but need
not belong to L(X, X), even though oX and oZ do. Our aim is to define and study "corrected" versions of
the functors K and C which have better exactness properties and in some circumstances form an adjoint
pair.
9
Definition 3.1. Let S be any collection of objects in an abelian category X. An epimorphism π : M → M
is an S-projective effacement of M if given any epimorphism f : P → M with kernel in S, there exists
g : M → P such that f g = π. If every object M in X has an S-projective effacement, then we say that X
has S-projective effacements. Dually, an S-injective effacement of M is an injection ι : M → M such that
given any monomorphism h : M → Q with cokernel in S, there is j : Q → M such that jh = ι; we say that
X has S-injective effacements if every object in X has an S-injective effacement.
In this definition, we have anglicized the terms "effacement projectif/injectif" which we believe were originally
due to Grothendieck. We have also added the dependence on S (usually effacements are defined in the case
S = X).
Note that the kernel of an S-projective effacement of M (or the cokernel of an S-injective effacement) is
not required to be in S, though in the applications it will be convenient to have this additional property.
We have the following easy observation in the important special case that S consists of objects in a closed
subcategory Z of X.
Lemma 3.2. Let Z be a closed subcategory of the abelian category X, and let S be some collection of objects
in Z. If M ∈ X has an S-projective effacement, then M has an S-projective effacement πM : M → M such
that ker πM ∈ Z. Similarly, if M has a S-injective effacement then M has a S-injective effacement with
cokernel in Z.
Proof. Let M ′ → M be an S-projective effacement, and let 0 → L → M ′ → M → 0 be the corresponding
exact sequence. Now i∗(L) = L/J is the largest factor object of L in Z, the sequence
0 → K = L/J → M = M ′/J → M → 0
is exact, and M → M is easily seen to be an S-projective effacement, whose kernel K is now in Z. The
statement for injective effacements is proved dually.
(cid:3)
Given epimorphisms πM : M → M ′ and πN : N → N ′, we say that a morphism f : M → N lifts the
morphism g : M ′ → N ′ if g ◦ πM = πN ◦ f . Dually, given monomorphisms ιM : M ′ → M and ιN : N ′ → N ,
we say that a morphism f : M → N extends the morphism g : M ′ → N ′ if f ◦ ιM = ιN ◦ g. Projective
effacements have a general morphism lifting property (and injective effacements have a morphism extending
property) which we give in the next result.
Lemma 3.3.
(1) Let πM : M → M be an S-projective effacement, and let πN : N ′ → N be any
epimorphism with kernel in S. Given any morphism f : M → N , there is a morphism f ′ : M → N ′
lifting f .
(2) Let ιN : N → N be a T -injective effacement, and let ιM : M → M ′ be any monomorphism with
cokernel in T . Given any morphism g : M → N , there is a morphism g′ : M ′ → N extending g.
10
Proof. We prove only the first statement, since the second is proved dually. We construct the commutative
diagram
πM
M
P
g
h
M
=
M
f
q
0
/ N ′
πN
/ N
/ 0
0
0
/ L
/ L
=
as follows. The bottom row is given, with L = ker πN ∈ S by assumption. The second row is formed by
letting P be the pullback of the maps f and πN ; see [Rot, Lemma 7.29]. The map g exists since πM is an
S-projective effacement. Then hg = f ′ is the required lift of f .
(cid:3)
We now show how to use projective and injective effacements to construct some important functors related
to a closed subcategory of X.
Proposition 3.4. Let X be a cocomplete, complete, and well-powered abelian category. Let Z be a closed
subcategory of X, and let the functors i!, i∗, C, K : X → X be defined as above with respect to Z.
(1) Let S be any collection of objects of Z. For each M ∈ X, assume that there exists an S-projective
effacement πM : M → M with kernel in S, and fix one such. Define F (M ) = K(M ). For any
morphism f : M → N in X, by Lemma 3.3 there exists a morphism f : M → N lifting f . Fix any
such f , and apply K to give F (f ) = K(f ) : F (M ) → F (N ). Then F : X → X is a functor which is
independent up to natural isomorphism of the choices of projective effacements and lifts. There is a
canonical morphism of functors ν : F → oX .
(2) Let T be any collection of objects of Z. For each M ∈ X, assume that there exists an T -injective
effacement ιM : M → M which has cokernel in T , and fix one such. Define G(M ) = C(M ). For
any morphism f : M → N in X, by Lemma 3.3 there exists a morphism f : M → N extending
f . Fix any such f , and apply C to give G(f ) = C(f ) : G(M ) → G(N ). Then G : X → X is a
functor which is independent up to natural isomorphism of the choices of injective effacements and
extensions. There is a canonical morphism of functors µ : oX → G.
(3) Suppose that both parts (1) and (2) apply, and that for each M ∈ X we have i!(M ) ∈ S and
i∗(M ) ∈ T . Then the functors (F, G) form an adjoint pair. In particular, this holds if S and T are
both equal to all objects in Z.
Proof. (1) First we show that F (f ) does not depend on the choice of lift f . Suppose that we have two different
lifts g1, g2 : M → N of f . Then πN ◦ (g1 − g2) = 0, and so in particular Im(g1 − g2) ∼= M / ker(g1 − g2) ∈ Z.
Since F (M ) is the smallest subobject M ′ of M such that M /M ′ ∈ Z, we see that F (M ) ⊆ ker(g1 − g2), and
g1(cid:12)(cid:12)F (M) = g2(cid:12)(cid:12)F (M). Thus K(g1) = K(g2), and so F (f ) is independent of the choice of lift. The independence
of the choice of lifts also easily implies that F is a functor: given f : M → N and h : N → P , we first choose
lifts f : M → N and h : N → P , and then we may choose h ◦ f : M → P as our lift of h ◦ f , from which
follows
F (h ◦ f ) = K(h ◦ f ) = K(h) ◦ K(f ) = F (h) ◦ F (f ),
11
/
/
/
/
/
/
/
/
/
/
/
/
/
since K is a functor. Similarly, choosing the identity map 1M : M → M as our lift of the indentity map
1M : M → M implies that F (1M ) = K(1M ) = 1F (M).
Next we show that the definition of F is independent (up to natural isomorphism) of the arbitrary
choices of projective effacements. Suppose that for each object M we choose an S-projective effacement
M : M ′ → M (also with kernel in S), and use these to define a functor F ′ in the same way. By Lemma 3.3,
π′
there are maps g : M → M ′ and h : M ′ → M which lift the identity map 1M . Then g ◦ h : M ′ → M ′ is a
lift of the identity map 1M ′ ; since the identity map 1M ′ : M ′ → M ′ is also a lift, by the independence of
lifts proved above, K(g) ◦ K(h) = K(g ◦ h) = K(1M ′ ) = 1F ′(M) is the identity. Similarly, K(h) ◦ K(g) =
K(h ◦ g) = K(1M ) = 1F (M). Thus ηM = K(g) : F (M ) → F ′(M ) is an isomorphism for each M . To see that
η is natural, if f : M → N is a morphism, choose lifts f : M → N and f ′ : M ′ → N ′ of f by Lemma 3.3.
Choose g1 : M → M ′ lifting 1M and g2 : N → N ′ lifting 1N as above. Then f ′ ◦ g1 and g2 ◦ f : M → N ′
both lift f . By the independence of lifts,
F ′(f ) ◦ ηM = K(f ′) ◦ K(g1) = K(f ′ ◦ g1) = K(g2 ◦ f ) = K(g2) ◦ K(f ) = ηN ◦ F (f ),
as required. Thus η : F → F ′ is a natural isomorphism of functors.
Finally, for each object M there is a morphism νM : F (M ) → M given by νM = πM(cid:12)(cid:12)K(M ). The maps νM
are natural, since if f : M → N then we have f ◦ πM = πN ◦ f by construction, and K is a functor. Thus
we have a morphism of functors ν : F → oX .
(2) This is completely dual to part (1), so the proof is omitted.
(3) For any fixed objects M, N ∈ X consider the diagram
Hom(F M, N )
φ
Hom(M , N )
ψ
/ Hom(M, GN ),
where the maps φ and ψ are defined as follows. Given h ∈ Hom(M , N ), applying K gives a map K(h) :
F (M ) = K(M ) → K(N ). Use the injection ιN : N → N to identify N with a subobject of N . Since
N /N ∈ Z, we have K(N ) ⊆ N because K(N ) is the smallest such subobject. Thus im(K(h)) ⊆ N and so
K(h) defines a morphism φ(h) : F (M ) → N . Similarly, applying C to h gives C(h) : C(M ) → C(N ) = G(N ).
Since the kernel of πM : M → M is in Z, there is an epimorphism M → C(M ) and C(h) induces a map
ψ(h) : M → G(N ).
There is an epimorphism N → G(N ) whose kernel i!(N ) is in S by hypothesis. Thus by Lemma 3.3, given
g ∈ Hom(M, GN ), there is g ∈ Hom(M , N ) lifting g. This is equivalent to ψ(g) = g. Thus ψ is surjective.
A dual argument using the assumption that i∗(M ) ∈ T shows that φ is surjective.
Now suppose that h ∈ ker(φ). This means that h(F (M )) = 0, and in particular, Im(h) is isomorphic to a
factor object of M /F (M ). Thus Im(h) ∈ Z. Conversely, if Im(h) ∈ Z, then M / ker(h) ∈ Z and so F (M ) ⊆
ker(h) since F (M ) is the smallest such subobject. In conclusion, ker(φ) consists of those h whose image is
in Z. A similar argument shows that ker(ψ) has the same description, so ker(φ) = ker(ψ). Thus there are
induced bijections eφ : Hom(M , N )/(ker φ) → Hom(F M, N ) and eψ : Hom(M , N )/(ker ψ) → Hom(M, GN )
and ηM,N : eψ ◦ eφ−1 : Hom(F M, N ) → Hom(M, GN ) is a bijection.
A diagram chase similar to the arguments already given shows that these isomorphisms ηM,N are natural
in M and N , so (F, G) is an adjoint pair as claimed.
(cid:3)
12
o
o
/
When the category X has enough injectives, the constructions above involving injective effacements can
be described in a much simpler way; dually, if the category has enough projectives then all of the results
above using projective effacements become simplified. We make this precise in the next results.
Recall that L(X, X) denotes the category of left exact functors X → X, with morphisms given by natural
transformations. Suppose that X has enough injectives. Then it is standard that L(X, X) is an abelian
category, in the following way. There is an equivalence of categories γ : Add(Inj(X), X) → L(X, X) where
Inj(X) is the full subcategory of X consisting of injective objects [VdB, Proposition 3.1.1(3)]. Explicitly,
given an additive functor G′ : Inj(X) → X, for each M ∈ X one takes the beginning of an injective resolution
in X and defines G(M ) = ker(G′(E0) → G′(E1)). Also, a morphism f : M → N
induces a morphism G(M ) → G(N ) by taking any lift of f to a morphism of the injective resolutions,
applying G′, and taking the induced map of the kernels. It is straightforward to check that this defines a
M → E0 → E1 → . . .
left exact functor G which is independent up to natural isomorphism of the choices involved. As we noted
in the previous section, Add(Inj(X), X) is automatically abelian by taking objectwise kernels and cokernels.
Thus L(X, X) is also an abelian category via the equivalence γ.
It is easy to see that in the category
L(X, X), kernels are still computed objectwise, but cokernels are not, in general. Analogously, if X has
enough projectives, the the category R(X, X) of right exact functors X → X is abelian, since it is equivalent
to the category Add(Pro(X), X); cokernels in R(X, X) are computed objectwise, but not kernels in general.
Lemma 3.5. Let X be an abelian category which is complete, cocomplete, and well-powered. Let Z be a
closed subcategory of X. Suppose that X has enough injectives; in particular X automatically has Z-injective
effacements, and for each M ∈ X we can fix such an effacement M → M with cokernel in Z, by Lemma 3.2.
Let G = GZ be the functor defined by Proposition 3.4(2) with T = Z, and let µ = µZ : oX → G be the
corresponding morphism. Then there is an exact sequence
in L(X, X), where θ : oZ → oX is the natural morphism.
0 → oZ
θ
→ oX
µ
→ G → 0
Proof. The natural transformation θ is a monomorphism in L(X, X) because θM : oZ(M ) → oX (M ) is a
monomorphism in X for all objects M . Let µ′ : oX → G′ be the cokernel of θ in L(X, X). By the construc-
tions outlined above, a left exact functor is uniquely determined by its restriction to the full subcategory
of injective objects. For an injective I we have G′(I) = coker(i!(I) → I) = I/i!(I). Since I is injective, it
is clear that the identity map 1I : I → I is a Z-projective effacement of I. Thus by the definition of the
functor G, we have G(I) = C(I) = G′(I), where C is defined as in the previous section, and by construction
the morphism µ : oX → G defined in Proposition 3.4(2) is also given on I by the natural map I → I/i!(I).
Thus µ = µ′, and µ is a cokernel of θ as required.
(cid:3)
The previous result shows that if X has enough injectives, to construct the functor GZ for any closed
subcategory Z we can simply take a cokernel of θZ : oZ → oX in L(X, X).
It is not necessarily to use
injective effacements at all.
For completeness, we state without proof the dual result we get when the category X has enough projec-
tives.
13
Lemma 3.6. Let X be an abelian category which is complete, cocomplete, and well-powered. Let Z be a closed
subcategory of X. Suppose that X has enough projectives; in particular X automatically has Z-projective
effacements, and for each M ∈ X we can fix such an effacement M → M with cokernel in Z, by Lemma 3.2.
Let F = FZ be the functor defined by Proposition 3.4(2), and let ν = νZ : F → oX be the corresponding
morphism. Then there is an exact sequence
in R(X, X), where ρ : oX → oZ is the natural morphism.
0 → F ν→ oX
ρ
→ oZ → 0
4. Effacements in Grothendieck categories
We now specialize the theory of the previous section to a Grothendieck category X with a closed sub-
category Z. As usual, let i∗ : Z → X be the inclusion functor and i! : X → Z, i∗ : X → Z its respective
right and left adjoints, and let oZ = i∗i!, oZ = i∗i∗ : X → X. A Grothendieck category X has injective
hulls, and an injective hull M → E(M ) is a T -injective effacement for any collection of objects T . Thus
the constructions using injective effacements in the previous sections can be described in the simple way
provided by Lemma 3.5. In particular, there is a left exact functor GZ defined by Proposition 3.4(2) with
T = Z, and GZ is also the cokernel of the natural map θ : oZ → oX in the abelian category L(X, X) of left
exact functors. It is important to know when GZ has a left adjoint. Since many Grothendieck categories of
interest do not have enough projectives, the idea is to use Proposition 3.4 to construct a left adjoint FZ to
the functor GZ. For this, we need X to have S-projective effacements for a suitable collection of objects S
in Z.
First, we will need to study some more general properties of S-projective effacements. We have the
following alternative characterization of an S-projective effacement of M .
Lemma 4.1. Let X be a Grothendieck category, and let S be any collection of objects in X. The following
are equivalent for an object M ∈ X and a surjection π : M → M :
(i) The morphism π is an S-projective effacement.
(ii) For every N ∈ S, the map π∗ : Ext1
X (M, N ) → Ext1
X (M , N ) induced by π is 0.
Proof. This is an immediate generalization of [Roos, Corollary 1.4], which proves the result in case S = X. (cid:3)
Next, we see that in certain cases, to show that X has S-projective effacements it suffices to check the
definition for only some objects in X or only some objects in S.
Lemma 4.2. Fix a collection S of objects in the Grothendieck category X.
(1) If M ∈ X has an S-projective effacement, then so does any epimorphic image of M .
(2) If {Mα} is a set of objects, each of which has an S-projective effacement πα : Mα → Mα, then the
direct sum ⊕πα : Lα Mα → Lα Mα is an S-projective effacement of Lα Mα.
(3) If {Oα} is a (small) set of generators for X, then X has S-projective effacements if and only if every
Oα has an S-projective effacement.
14
(4) Suppose that X is locally noetherian. Let S′ be a collection of objects such that every object in S is
a direct sum or a direct limit of objects in S′. If π : M → M is a S′-projective effacement for a
noetherian object M ∈ X, then π is also an S-projective effacement.
Proof. (1) Let f : M → N be an epimorphism, and suppose that π : M → M is an S-projective effacement
of M . If p : Q → N is an epimorphism with kernel in S, then Lemma 3.3 shows that there is is a map
ef : M → Q covering f , in other words such that pef = f π. This shows that f π : M → N is an S-projective
effacement of N .
(2) Use Lemma 4.1 and the following commutative diagram, in which the vertical maps are isomorphisms
by Lemma 2.5(3):
Qα Ext1(Mα, N )
Qα Ext1(Mα, N )
∼=
∼=
Ext1(Lα Mα, N )
/ Ext1(Lα Mα, N ).
(3) This follows from (1) and (2), since every object in X is an epimorphic image of a direct sum of
generators.
(4) Consider the following commutative diagram for an arbitrary direct sum L Nα with each Nα ∈ S′:
Lα Ext1(M, Nα)
0
Lα Ext1(M , Nα)
∼=
Ext1(M,Lα Nα)
/ Ext1(M ,Lα Nα).
The top arrow is 0 since M → M is an S′-projective effacement. The left vertical arrow is an isomorphism
since M is a noetherian object, using Lemma 2.5(2).
a directed system {Nα} instead, the same argument works to show that the map Ext1
Ext1(M , lim
−→
of objects in S′, we conclude that M → M is an S-projective effacement by Lemma 4.1.
It follows that the bottom arrow is also 0. Given
X (M, lim−→ Nα) →
Nα) is zero, again using Lemma 2.5(2). Since every object in S is a direct sum or direct limit
(cid:3)
When applying Proposition 3.4 to construct a functor using S-projective effacements, it is necessary to
have effacements whose kernels are actually in S. This is easy when S is equal to all objects in the closed
subcategory Z, by Lemma 3.2. The following lemma gives another important case where we can guarantee
this.
Lemma 4.3. Let Z be a closed subcategory of the Grothendieck category X. Let S be the collection of
objects in Z which are injective in the category Z. If M ∈ X has an S-projective effacement, then there is
an S-projective effacement πM : M → M such that ker πM ∈ S.
Proof. By Lemma 3.2, since M has an S-projective effacement, we can find an S-projective effacement
t : M ′ → M with K = ker t ∈ Z. Now let g : K → E = EZ (K) be an injective hull in the category Z. If P
is the pushout of the maps f : K → M ′ and g, we obtain an exact sequence 0 → E → P → M → 0, giving
the first two rows of the commutative diagram below [Rot, Lemma 7.28].
15
/
/
/
/
/
/
We claim that π : P → M is an S-projective effacement; then we will be done since its kernel E is
apparently in S. Now if 0 → F
r
→ Q
s
→ M → 0 is any short exact sequence with F ∈ S, consider the
following commutative diagram:
0
0
0
/ E
g
/ K
h′
/ F
b
f
r
/ P
i
/ M ′
h
/ Q
π
t
s
/ M
=
/ M
=
/ M
/ 0
/ 0
/ 0.
Here, the map h exists completing the bottom right square since t : M ′ → M is an S-projective effacement,
and the map h′ is induced by h. Since F is in S and so is injective in Z, there is d : E → F such that
dg = h′. Since rdg = rh′ = hf , by the universal property of the pushout there is a morphism j : P → Q
such that h = ji and rd = jb. Then πi = t = sh = sji, and so (π − sj)i = 0. Now by construction of the
pushout, one has Im(i) + Im(b) = P . We have seen that Im(i) ⊆ ker(π − sj). We also have πb = 0 and
sjb = srd = 0, and then (π − sj)b = πb − sjb = 0, so that Im(b) ⊆ ker(π − sj). Thus P ⊆ ker(π − sj) and
π = sj, which shows that π is an S-projective effacement as required.
(cid:3)
Recall that since X has Z-injective effacements, for each M ∈ X we can choose a Z-injective effacement
with cokernel in Z, by Lemma 3.2. Following the proof of that lemma, we see in fact that for each M we
have a canonical (up to isomorphism) Z-injective effacement j : M → M , with M /M = i!(E(M )/M ), where
E(M ) is the injective hull of M . In other words, M is the maximal essential extension of M by an object in
Z. These canonical Z-injective effacements have the following property.
Lemma 4.4. Let Z be a closed subcategory of the Grothendieck category X. For M ∈ X, let j : M → M
be the canonical Z-injective effacement discussed above. Then applying the functor i! yields an injection
i!(j) : i!(M ) → i!(M ), which is an injective hull of i!(M ) in the category Z.
Proof. Let N = i!(M ), in other words the largest subobject of M in Z. Let N ⊆ I be an injective hull in
the category Z. Then N ⊆ I is essential in the category X as well, and so we can choose an injective hull
E(N ) of N in X with N ⊆ I ⊆ E(N ). Since N ⊆ M , we can choose an injective hull E(M ) of M in X so
that E(N ) ⊆ E(M ).
Working in E(M ), I +M/M ∼= I/(I ∩M ) is in Z, and thus I +M ⊆ M since M /M is the largest subobject
of E(M )/M in Z. In particular, I ⊆ M and thus I ⊆ i!(M ). On the other hand, if 0 6= P ⊆ i!(M ), then
P ∩M 6= 0 as M ⊆ M is essential. Since P ∈ Z, 0 6= P ∩M ⊆ i!(M ) and this proves that i!(j) : i!(M ) → i!(M )
is essential (in X or in Z). This forces i!(M ) = I since I is a maximal essential extension in Z.
(cid:3)
Considering the definition, an S-projective effacement of an object M is some object which, speaking
loosely, ought to be formed by sticking objects in S to the bottom of M in all possible nontrivial ways. This
naive description can actually be made formal in some cases, as we see in the proof of the following result.
16
/
/
/
/
/
O
O
/
O
O
/
O
O
/
/
/
/
/
Proposition 4.5. Let X be a locally noetherian Grothendieck category, and let S be a collection of objects
in X with a small subcollection S′ ⊆ S such that every element of S is either a direct sum or a direct limit
of objects in S′.
Fix a noetherian object M ∈ X, and suppose that for any small-indexed family of objects (Nα) with each
Nα ∈ S′, the natural map
X (M,Y
Ext1
Nα) → Y
Ext1
X (M, Nα)
α
α
(4.6)
is an isomorphism. Then M has an S-projective effacement.
Proof. We prove this first in the special case that X is a k-category for a field k.
In this case we can
construct a smaller projective effacement, which is useful in applications. Afterward we indicate the easy
changes necessary to remove the assumption that X is a k-category.
Let M be a noetherian object satisfying (4.6) for all small-indexed families of objects in S′. For each
N ∈ S′, the underlying set of Ext1
small sets and is thus small. For each N ∈ S′, pick a k-basis βN for Ext1
the disjoint union of these bases, which is again a small set. For each v ∈ β, let Nv be a copy of the object
X (M, N ) is a small union of
X (M, N ), and let β = SN ∈S ′ βN be
X (M, N ) is small, so the union SN ∈S ′ Ext1
N such that v ∈ βN .
Now we have an isomorphism
φ : Ext1
X (M, Y
Nv) → Y
Ext1
X (M, Nv)
v∈β
v∈β
(4.7)
by hypothesis. Note that there is a special element of the right hand side given by θ = Qv∈β v. The element
θ′ = φ−1(θ) of the left hand side represents some extension
0 → Y
v∈β
Nv
i→ M π→ M → 0
and we claim that π is a S′-projective effacement of M .
Consider an extension 0 → N → P
Ext1
X (M, N ). We may write ρ = Pm
f
→ M → 0 with N ∈ S′ and the corresponding element ρ ∈
i=1 aivi with ai ∈ k and vi ∈ βN ⊆ β for all i. There is a morphism
j : Y
v∈β
Nv
q
→
mY
i=1
Nvi
∼=
mM
i=1
N
p
→ N
where q is the natural projection map, and p is given by the formula Pm
i=1 ai1N . Since finite direct sums pull
out of the second coordinate of Ext (for instance, by Lemma 2.5(2)), the morphism j induces a corresponding
map of Ext groups
bj : Ext1
X (M, Y
v∈β
Nv) → Ext1
X (M,
mY
i=1
Nvi) ∼=
17
mM
i=1
Ext1
X (M, N )
bp
→ Ext1
X (M, N )
where bp is given by the formula (w1, . . . , wm) 7→ Pm
there is a commutative diagram
i=1 aiwi. By construction, bj(θ′) = ρ. This means that
0
0
/ Qv∈β Nv
i
j
/ N
π
f
M
h
/ P
0
,
M
=
/ M
/ 0
where P is a pushout of i and j [Rot, Formula II, p. 429]. Then h : M → P satisfies f h = π, proving the
claim that π is a S′-projective effacement.
Now by Lemma 4.2(4), since M is noetherian and π : M → M is a S′-projective effacement, it is also an
S-projective effacement.
When X is not necessarily a k-category, essentially the same proof works, with the following changes.
First, one replaces βN with the entire set Ext1
is still small. Again one writes Nv for a copy of the N such that v ∈ Ext1
ρ ∈ Ext1
j : Qv∈β Nv → Nρ = N . The rest of the proof is the same.
X (M, N ), which
X (M, N ). Given an extension
X (M, N ) ⊆ β, one replaces the map j above simply with the projection onto the single ρth coordinate:
(cid:3)
X (M, N ) instead, so that β = SN ∈S ′ Ext1
Remark 4.8. We note that only the single instance (4.7) of the hypothesis (4.6) of the proposition is used
in the proof. Thus in practice one only needs to verify the single equation (4.7) in order for the result to
hold.
5. Well-closed and very well-closed subcategories
We are now ready to prove our main theorems.
Theorem 5.1. Let X be a locally noetherian Grothendieck category, let Z be a closed subcategory of X, and
let S be the collection of all objects in Z which are injective in the category Z. Let 0 → oZ → oX → G → 0
be the exact sequence in L(X, X) of Lemma 3.5.
Then the following are equivalent:
(1) G has a left adjoint F .
(2) The functor G commutes with products of objects in X.
(3) [R1 Q](Nα) = 0 for all small families {Nα} of objects in S.
(4) The natural map Ext1(M,Qα Nα) → Qα Ext1(M, Nα) is an isomorphism, for all small families
{Nα} of objects in S and for all M ∈ X.
(5) The category X has S-projective effacements.
Proof. Van den Bergh studies condition (3) and shows that (1) ⇐⇒ (3) [VdB, Proposition 3.4.7]. Also, the
equivalence (1) ⇐⇒ (2) follows from Freyd's adjoint functor theorem [VdB, Theorem 2.1(1)].
Lemma 2.5(1) shows that (3) =⇒ (4).
Now suppose that condition (4) holds. Let S′ be the set of isomorphism classes of indecomposable
injective objects in the category Z. By Lemma 2.1, Z is also a locally noetherian Grothendieck category,
every injective object in Z is a direct sum of indecomposable injectives in Z, and the set S′ is small. Thus
by Proposition 4.5, if M is a noetherian object in X, then M has an S-projective effacement. Since X has a
18
/
/
/
/
/
/
/
/
/
/
/
set of noetherian generators, it follows that every M ∈ X has an S-projective effacement by Lemma 4.2(3).
So (4) =⇒ (5).
Now assuming (5), then in fact every M ∈ X has an S-projective effacement πM : M → M with
ker πM ∈ S, by Lemma 4.3. Thus Proposition 3.4(1) applies and constructs a functor F . Let T be the class
of all objects in Z. We already have the canonical T -injective effacement ιN : N → N with cokernel in
T , where N /N = i!(E(N )/N ), as described before Lemma 4.4, and Proposition 3.4(2) constructs a functor
which is the same as G up to natural isomorphism, by Lemma 3.5. we have i∗(M ) ∈ T trivially, and
i!(N ) ∈ S holds for all N ∈ X by Lemma 4.4. So Proposition 3.4(3) applies and shows that (F, G) form an
adjoint pair. Thus (5) =⇒ (1).
(cid:3)
Definition 5.2. Let Z be a closed subcategory of a locally noetherian Grothendieck category X. We say
that Z is well-closed in X if any of the equivalent conditions of Theorem 5.1 holds. The definition follows
Van den Bergh, who uses this term for condition (3) in the theorem [VdB, Definition 3.4.6].
We expect well-closedness to be a very general condition. In fact, we do not know any example of a closed
subset of a locally noetherian Grothendieck category that is not well-closed. For some applications it is
useful to study a more special condition on closed subcategories than well-closedness. We have the following
variant of Theorem 5.1.
Theorem 5.3. Let X be a locally noetherian Grothendieck category and let Z be a closed subcategory of X.
Then the following are equivalent:
(1) [R1 Q](Nα) = 0 for all small families {Nα} of objects in Z.
(2) Z is well-closed in X and the category Z has exact direct products.
(3) The natural map Ext1(M,Qα Nα) → Qα Ext1(M, Nα) is an isomorphism, for all small families
{Nα} of objects in Z and for all M ∈ X.
(4) Every object M in X has a Z-projective effacement.
Proof. Van den Bergh proves that (1) ⇐⇒ (2) [VdB, Corollary 3.4.11].
Lemma 2.5(1) shows that (1) =⇒ (3).
Assume now that (3) holds. Let S = Z and let S′ be the collection of isomorphism classes of noetherian
objects in Z. The category Z is itself Grothendieck, and so has a generator O. It is easy to see that any
noetherian object in Z is an epimorphic image of O⊕n for some n. By well-poweredness, the set S′ is a
countable union of small sets and is thus small. Because Z is locally noetherian, every object in Z is a direct
limit of noetherian objects [Po, Proposition 8.6]. Since every object in S is a direct limit of objects in S′,
Proposition 4.5 shows that every noetherian object M in X has an S-projective effacement. Then every
object in X has an S-projective effacement by Lemma 4.2(3). Thus (3) =⇒ (4).
Finally, assume that (4) holds. Then in particular, every M ∈ X has an S-projective effacement, where S
is the collection of objects which are injective in the category Z. Thus Z is well-closed in X by Theorem 5.1.
Suppose that M ∈ Z, and that p : M ′ → M is a Z-projective effacement of M in X. We can assume that
ker p ∈ Z by Lemma 3.2, but apriori, M ′ need not be in Z. Let M = i∗(M ′) = M ′/N be the largest factor
object of M ′ which is in Z. Since M ∈ Z, N ⊆ ker p and so there is an induced epimorphism π : M → M .
19
Suppose that 0 → K → L
f
→ M → 0 is a short exact sequence in the category Z. There is h : M ′ → L such
that f h = p, by the definition of Z-projective effacement. Since L ∈ Z, Im h ∈ Z and so N ⊆ ker h. In other
words, h factors through M . Thus π : M → M is a Z-projective effacement in the category Z. We conclude
that the Grothendieck category Z has Z-projective effacements. By a result of Roos [Roos, Theorem 1.3],
this implies that Z is an (AB4*) category, that is, that Z has exact products. Thus (4) =⇒ (2).
(cid:3)
Remark 5.4. Roos leaves the proof of the just-cited part of [Roos, Theorem 1.3] to the reader. The proof
was not obvious to us, so we indicate here a possible argument that if a Grothendieck category Z has Z-
projective effacements, then it has exact products. Given the Z-projective effacement πM : M → M and a
collection {Nα} of objects in Z, recall the Grothendieck spectral sequence
Ep,q
2 = Extp
Z(M, Rq Qα Nα) =⇒ Qα Extp+q
Z (M, Nα)
which was used in Lemma 2.5, now applied in the category Z, so that Rq Q is the qth right derived functor of
the product functor in Z. By naturality of the exact sequences of low degree terms, applied to the morphism
πM , we get the following commutative diagram:
0
0
/ Ext1
Z(M,Qα Nα)
g0
/ Ext1
Z(M ,Qα Nα)
d1
d′
1
Qα Ext1
Z(M, Nα)
d2
HomZ(M, R1 Q Nα)
g1
g2
/ Qα Ext1
Z (M , Nα)
d′
2
/ HomZ (M , R1Q Nα)
d3
d′
3
Ext2
Z (M,Qα Nα)
g3
/ Ext2
Z(M ,Qα Nα)
2g1 = 0, we see that d2 = 0. Thus the natural map Ext1
Since π : M → M is a Z-projective effacement, g0 = g1 = 0 by Lemma 4.1. By left exactness of Hom, g2 is
Z(M,Qα Nα) →
a monomorphism. Since g2d2 = d′
Qα Ext1
Z (M, Nα) is an isomorphism. Since M was an arbitrary object in Z, this fact also applies to the
object M , which shows that d′
3g2. Now let E be an
injective hull of Qα Nα in Z and consider the short exact sequence 0 → Qα Nα → E → C → 0. Then
Z (M, C) → Ext1(M , C) = 0
Ext2
because π is a Z-projective effacement, we get g3 = 0. Now d′
3g2 = g3d3 = 0, but we already saw that
3g2 is a monomorphism. This forces HomZ(M, R1 Q Nα) = 0. Since this holds for all objects M ∈ Z,
d′
R1 Q Nα = 0. This holds for all small families {Nα} of objects in Z, so Z has exact products.
Z (M, C) for all objects M ; since the natural map Ext1
Z(M,Qα Nα) ∼= Ext1
2 = 0. Thus d′
3 is a monomorphism and hence so is d′
Remark 5.5. Theorem 5.3 also gives an alternative proof of the other (harder) direction of [Roos, Theorem
1.3], that if a Grothendieck category Z has exact direct products, then it has Z-projective effacements, but
only in the special case that Z is locally noetherian. This follows from the proof that (4) =⇒ (2) in
Theorem 5.3, in the case X = Z (note that Z is trivially well-closed in Z).
Definition 5.6. Let Z be a closed subcategory of a locally noetherian Grothendieck category X. We say
that Z is very well-closed in X if any of the equivalent conditions in Theorem 5.3 holds (Van den Bergh uses
this term for condition (1)).
Very well-closedness is clearly a much more special condition than well-closedness. One of its advantages is
that it is stable under Gabriel product.
20
/
/
/
/
/
/
/
/
/
/
/
Lemma 5.7. Let X be a locally noetherian Grothendieck category with closed subcategories Z1, Z2. If Z1
and Z2 are very-well-closed in X, then so is Z3 = Z1 · Z2.
Proof. This is [VdB, Proposition 3.5.12].
It is also easy to see why this is true in terms of projective
effacements, as follows. For each M ∈ X fix a Z1-projective effacement πM : M → M and a Z2-projective
effacement ρM : M✿✿ → M . We can assume that ker πM ∈ Z1 and ker ρM ∈ Z2, by Lemma 3.2. Then
θM = πM ◦ ρM : M
→ M is a Z3 = Z1 · Z2-projective effacement of M , as is easy to check.
✿✿
(cid:3)
There is no obvious reason, on the other hand, for the Gabriel product of well-closed subcategories to be
well-closed, as Van den Bergh has also noted. We do not know a counterexample, however.
One easy way to ensure (very) well-closedness, which often occurs in applications, is the following.
Definition 5.8. Let Z be a closed subcategory of a locally noetherian Grothendieck category X, and let S
be a collection of objects in the category Z. We say that an object M ∈ X is S-projective self-effacing if the
identity map M → M is an S-projective effacement. We say that X has a set of S-projective self-effacing
generators if there is a small set of generators {Oα} for X, where each Oα is S-projective self-effacing.
Proposition 5.9. Let X be a Grothendieck category with closed subcategory Z, and let S be a collection of
objects in the category Z. Let {Oα} be a small set of generators for X.
(1) If Ext1(Oα, N ) = 0 for all Oα and all N ∈ S, then {Oα} is a set of S-self-effacing generators for X.
(2) If {Oα} is a set of S-self-effacing generators for X, then X has S-projective effacements.
Proof. (1) Given any short exact sequence 0 → N → P → Oα → 0 where N ∈ S, the sequence must be
split since Ext1(Oα, N ) = 0. Then it is clear that the identity map Oα → Oα is already an S-projective
effacement, so each Oα is S-projective self-effacing.
(2) In particular, the hypothesis implies that each generator has an S-projective effacement. Then X has
S-projective effacements, by Lemma 4.2(3).
(cid:3)
For example, let X be a locally noetherian Grothendieck category with noetherian generators {Oα}, let Z
be a closed subcategory, and let S be the set of injective objects in the category Z. If Ext1(Oα, E) = 0 for
all E ∈ S and all α, then the previous result implies that Z is well-closed in X. We note that in this case
we not only get that the functor F = FZ exists, but also that it can be described in a way similar to its
description if X were to have enough projectives, as given in Lemma 3.6. Namely, let M ∈ X. Then there
is a "partial resolution"
P1 → P0 → M → 0
where P1, P0 are direct sums of generators. The Pi here are of course not projective, in general. Still,
we may apply F to this sequence to obtain an exact sequence F (P1) → F (P0) → F (M ) → 0, since F
is right exact. Since the Oα are S-projective self-effacing, the same is true for each P , by Lemma 2.5(3).
Thus by the construction of the functor F given in Proposition 3.4, we have F (Pi) = K(Pi), where recall
that K is the functor that sends an object N to its smallest subobject N ′ such that N/N ′ ∈ Z. Thus
K(P1) → K(P0) → F (M ) → 0 is exact. We see that F (M ) may be defined by taking such a resolution
P1 → P0 of M by self-effacing objects, applying K, and taking the cokernel. This is the same description
21
as we would get if X had enough projectives, where we would take a partial projective resolution instead.
However, defining the action of F on morphisms is more awkward, as one lacks a comparison lemma for
these resolutions by self-effacing objects.
6. Examples
We close the paper with some examples of how the theory works out for some important kinds of
Grothendieck categories.
Example 6.1. The prototypical example of a Grothendieck category is the category Mod- R of right modules
over a ring R. The closed subcategories of Mod- R are exactly the subcategories of the form Mod- R/I for
(two-sided) ideals I in R, by a result of Rosenberg [Ros, Proposition 6.4.1]. The category has the generator
R and if R is right noetherian, then Mod- R is locally noetherian.
The category X has enough projectives and exact products, so it is obvious that every closed subcategory
Z = Mod- R/I is very well-closed. In fact G = GZ has the explicit description G = HomR(I, −) and the
functor F = FZ has the explicit description F = − ⊗R I. Also, {R} is a projective generator, so it is
S-projective self-effacing for any collection of objects S.
Example 6.2. A somewhat less trivial example is the category X = Qcoh Y of quasi-coherent sheaves on a
k-scheme Y . For simplicity, suppose that Y is projective over k in this example. Then it is well-known that
X is a Grothendieck category. If L is an ample invertible sheaf on Y , then {L⊗nn ∈ Z} is a set of noetherian
generators for X, so X is certainly locally noetherian. Smith has shown that the closed subcategories of
X are those of the form Z = Qcoh W for closed subschemes W of Y [Sm1, Theorem 4.1]. The category X
has enough injectives, but not enough projectives, in general, and need not have exact products. (We are
unaware of a general result about which schemes have categories of quasi-coherent sheaves with exact direct
products, but P1 already does not.)
If I is the ideal sheaf defining the closed subscheme W then the functor GZ has the explicit description
GZ = Hom OX (I, −), and this has the obvious left adjoint FZ = (−) ⊗OX I. Thus every closed subcategory
Z is well-closed, but need not be very well-closed (since this is equivalent to Z having exact products).
Suppose that E is an injective object in the category Z = Qcoh W . Then E is flasque on W , so it is also
flasque (but not injective in general) when considered as a sheaf on Y . This is enough to conclude that
X (OX , E) = 0 [Ha, Proposition 2.5]. Similarly, Ext1(L⊗n, E) ∼= Ext1(OX , E ⊗ L⊗−n) = 0
H 1(Y, E) = Ext1
since E ⊗ L⊗−n is still flasque. In particular, X has a set of S-projective self-effacing generators, where S is
the class of injective objects in Z.
We assumed projectivity of Y in the previous example for convenience only. Ryo Kanda has proved that for
any locally noetherian scheme Y , the closed subcategories of Qcoh Y are still exactly the categories Qcoh W
for closed subschemes W of Y [Kan, Theorem 11.11].
Example 6.3. let H be a group and let R = Lh∈H Rh be an H-graded k-algebra. Let X = Gr-R be the
category of H-graded right R-modules. Then X is a Grothendieck category. For any h ∈ H and M ∈ X we
have the shifted module M (h) with M (h)g = Mhg. Then {R(h)h ∈ H} is a set of generators for X, and
so if R is graded right noetherian, then X is a locally noetherian category. In fact the R(h) are projective
22
generators, so X has enough projectives. The category X also clearly has exact products. Because of this,
every closed subcategory Z of X must be well-closed, using the characterization in Theorem 5.1(3). Also, Z
clearly inherits the property of having exact direct products, so in fact Z is very well-closed. Since the R(h)
are projective, they are in fact S-projective self-effacing generators for X, for any collection S of objects.
If I is a graded ideal of R, then Z = Gr-R/I is closed in X, and we have the explicit descriptions of the
functors GZ = HomR(I, −) and FZ = − ⊗R I as usual. However, closed subcategories of Gr-R need not
be defined by two-sided ideals in this way. For example, consider any N-graded algebra R and let Z be the
subcategory of Gr-R consisting of Z-graded modules M with M = M≥0, in other words, the subcategory of
nonnegatively graded modules. It is clear that Z is closed under subquotients, direct sums, and products,
so that Z is a closed subcategory. However, Z is clearly not equal to Gr-R/I for a graded ideal I of R, since
any such subcategory defined by an ideal is closed under shift, while Z is not.
Sierra studied the group of autoequivalences of the category of graded modules over the Weyl algebra [Si].
It seems interesting to note that the nontrivial autoequivalences in her work arise as functors FZ as defined
in this paper.
Example 6.4. let k be an algebraically closed field of characteristic 0. Let R be the first Weyl algebra
R = khx, yi/(yx − xy − 1), which is Z-graded with deg x = 1, deg y = −1. The simple objects in Gr-R are
parametrized by the affine line over k with its integer points doubled. More specifically, there is a simple
module Mλ for each λ ∈ k \ {Z} and 2 simple modules X(n), Y (n) for each n ∈ Z [Si, Lemma 4.1]. Sierra
proves that any autoequivalence of Gr-R is determined by its action on the simple modules [Si, Corollary
5.6], and for each n she constructs an interesting autoequivalence F that switches X(n) and Y (n) and fixes
all other simple modules [Si, Proposition 5.7]. This F can be defined as follows:
for each graded rank
one projective module P , HomGr-R(P, X(n) ⊕ Y (n)) = k; thus P surjects onto exactly one of the modules
X(n), Y (n). Let F (P ) be the kernel of this surjection. This action extends to a unique exact right exact
functor F on the whole category, since Gr-A has enough projectives, similarly as in our discussion after
Proposition 3.4.
The graded simple modules M satisfy dimk Mn ≤ 1 for all n ∈ Z and HomGr-R(M, M ) = k, from which
one may easily see that all graded simple modules are tiny in the category Gr-R. Now for fixed n let Z
be the full subcategory of Gr-R consisting of all direct sums of X(n) and Y (n), which is closed since these
simples are tiny. Let FZ be the corresponding functor constructed by Proposition 3.4(1) with S = Z. Then
since every projective is Z-self-effacing, we see that for a rank one projective P , the object FZ (P ) is the
smallest subobject Q of P such P/Q ∈ Z. Since HomGr-R(P, X(n) ⊕ Y (n)) = k, it is easy to see that FZ is
the same as the functor F described above.
Finally, we discuss our main motivating example: noncommutative projective schemes. Let A be an
N-graded k-algebra which is connected (A0 = k), finitely generated, and noetherian. Let Tors-A be the full
subcategory of Gr-A consisting of modules M such that for all m ∈ M , mA≥n = 0 for some n ≥ 0. Then
one may define the quotient category Qgr-A = Gr-A/ Tors-A. This category has the same objects as Gr-A,
but for M ∈ Gr-A we write its image in Qgr-A as πM . The morphisms are given by
HomQgr-A(πM, πN ) = lim
−→
23
HomGr-A(Mα, N/Nβ),
where the limit ranges over those graded submodules Mα ⊆ M such that M/Mα ∈ Tors-A and those graded
submodules Nβ ⊆ N such that Nβ ∈ Tors-A. The functor π : Gr-A → Qgr-A is exact, and has a left adjoint
ω : Qgr-A → Gr-A which can be given explicitly as ω(πM ) = lim
n→∞Lm HomGr-A(A≥n, M (m)). The category
Qgr-A is called a noncommutative projective scheme. For more details, see [AZ].
The algebra A is said to satisfy the χi condition if dimk Extj
A(k, M ) < ∞ for all j ≤ i, for all noetherian
modules M ∈ Gr-A, where k = A/A≥1 is the trivial module. Most well-behaved graded algebras satisfy χi
for all i ≥ 0, in which case we say that A satisfies χ. In this case the category Qgr-A is Ext-finite in the
sense that dimk Exti
Qgr-A(M, N ) < ∞ for all noetherian objects M, N ∈ Qgr-A [AZ, Corollary 7.3(3)].
Example 6.5. Let X = Qgr-A for a connected, finitely generated N-graded noetherian k-algebra A satisfying
χ, and maintain the notation above. Since Gr-A is a Grothendieck category, so is its quotient category
X = Qgr-A [Po, Corollary 4.6.2]. As in the case of categories of quasi-coherent sheaves on commutative
schemes, the category X usually does not have enough projectives or exact products.
Since {A(n)n ∈ Z} generates Gr-A, {πA(n)n ∈ Z} generates Qgr-A, and so X is locally noetherian.
Write O = πA. Artin and Zhang defined a cohomology theory for X as follows: for N ∈ X, let H i(X, N ) =
Exti
(1), so O(n) = πA(n). Then H i(X, N (n)) = Exti
X (O, N ). The shift autoequivalence (1) of Gr-A induces an autoequivalence of Qgr-A we also write as
X (O, N (n)) = Exti
X (O(−n), N ).
Suppose that I is a graded ideal of A. Then Z = Qgr-A/I is a closed subcategory of X = Qgr-A. The
non-trivial proof was given by Smith [Sm2, Theorem 3.2], [Sm3, Theorem 1.2].
In this case, Artin and
Zhang showed that cohomology restricts nicely to such a closed subcategory. Let OZ = π(A/I). Then if
Z (OZ, N ) = Exti
N ∈ Z, we have H i(Z, N ) = Exti
X (OX , N ) = H i(X, N ) [AZ, Theorem 8.3(3)], similarly
[Ha, Lemma 2.10]). Thus if E is an injective object in the category
as in the commutative case (cf.
Z, since Z is closed under the shift autoequivalence (1), E(−n) is also injective in Z, and we will have
Exti
Z (OZ, E(−n)) = 0. Hence {O(n)n ∈ Z} is a set of S-projective
self-effacing generators, for S the class of injective objects in Z. In particular, X has S-projective effacements
X (OX , E(−n)) = Exti
X (O(n), E) = Exti
by Proposition 5.9, and Z is well-closed in X. As for categories of commutative sheaves, Z will not generally
have exact products, and so Z is not generally very well-closed.
However, similarly as for the category Gr-A, closed subcategories Z of Qgr-A need not be defined by
two sided ideals of A, even when Z is a closed point. For example, suppose that A is generated by A1
as a k-algebra. A point module for A is a graded right module M which is generated in degree 0 and
satisfies dimk Mn = 1 for all n ≥ 0. For any point module M , π(M ) is a tiny simple object in Qgr-A [Sm1,
Proposition 5.8], and thus the category of small direct sums of π(M ) is a closed point Z in Qgr-A. But
generally M ∼= A/I for a right but not 2-sided ideal I of A, and so Z is not typically defined by a 2-sided
ideal of A.
To close the paper, we show that our theory applies to prove that closed points, and the locally finite
categories built out of them using the Gabriel product, are very well-closed in quite general circumstances.
Theorem 6.6. X be a locally noetherian Grothendieck k-category.
24
(1) Suppose that Z is a closed point in X, that is, Z is the category of direct sums of a tiny simple
object P in X. Suppose that dimk EndX (P ) < ∞ and that dimk Ext1
objects M ∈ X. Then X has Z-projective effacements; that is, Z is very well-closed in X.
X (M, P ) < ∞ for all noetherian
(2) Let Z1, Z2, . . . Zn be closed points in X (possibly with repeats), each of which satisfies the hypothesis
of (1). Then any closed subcategory Z contained in the Gabriel product Z1·Z2·. . .·Zn has Z-projective
effacements; i.e. Z is very well-closed in X.
Proof. (1) We would like to apply Proposition 4.5 with T = {P }. For this, as noted in Remark 4.8, it is
enough to show that for a noetherian object M , then (4.7) holds, namely
φ : Ext1
X (M, Y
Nv) → Y
Ext1
X (M, Nv)
v∈β
v∈β
(6.7)
is an isomorphism, where β is a k-basis of Ext1
X (M, P ).
By assumption, Ext1
X (M, P ) is finite dimensional over k, so β is finite. Thus (6.7) holds because finite
direct products and direct sums coincide, and Ext1 commutes with direct sums in the second coordinate
when M is noetherian, by Lemma 2.5(2). Thus the proof of Proposition 4.5 goes through to show that
any noetherian object M has a Z-projective effacement, since every object in Z is a direct sum of copies of
objects in T . Then X has Z-projective effacements by Lemma 4.2(3).
(2). This is immediate from part (1) and the fact that being very well-closed is stable under the Gabriel
product (by Lemma 5.7) and under taking subcategories.
(cid:3)
Corollary 6.8. Let A be a connected N-graded finitely generated noetherian k-algebra satisfying χ, and let
X = Qgr-A. Then any closed subcategory of a finite Gabriel product of closed points is very well-closed in
X.
Proof. The χ condition implies that X is Ext-finite, as already noted. Since a simple object P is noetherian,
dimk EndX (P ) < ∞ and dimk Ext1
Theorem 6.6 is satisfied.
X (M, P ) < ∞ holds for all noetherian objects M , so the hypothesis of
(cid:3)
The property of Ext-finiteness should be thought of as a kind of properness assumption for noncommuta-
tive schemes. Thus Theorem 6.6 can be interpreted to say that closed points (or more generally, finite length
subschemes) of proper noncommutative schemes ought to be well-closed. This may be useful for iterating
Van den Bergh's blowing up procedure, which does not seem to preserve projectivity in general.
Remark 6.9. When Theorem 6.6(1) applies, Proposition 4.5 constructs a Z-projective effacement of a
noetherian object M quite explicitly, by adding n = dimk Ext1(M, P ) copies of P to the bottom of M in the
n different ways given by a basis of Ext1
X (M, P ). More exactly, the effacement is given by the exact sequence
0 → Ext1(M, P ) ⊗k P → M → M → 0,
corresponding under the correspondence between Ext1 and extensions to a diagonal element
θ =
nX
i=1
vi ⊗ vi ∈ Ext1(M, P ) ⊗k Ext1(M, P ) ∼= Ext1(M, Ext1(M, P ) ⊗k P )
where {vi} is a k-basis of Ext1(M, P ).
25
References
[AZ] M. Artin and J. J. Zhang, Noncommutative projective schemes, Adv. Math. 109 (1994), no. 2, 228 -- 287. MR 96a:14004
[Ga]
[Ha]
Pierre Gabriel, Des cat´egories ab´eliennes, Bull. Soc. Math. France 90 (1962), 323 -- 448. MR 38 #1144
Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977, Graduate Texts in Mathematics, No. 52.
MR 57 #3116
[Kan] Ryo Kanda, Classification of categorical subspaces of locally Noetherian schemes, Doc. Math. 20 (2015), 1403 -- 1465.
MR 3452186
[Mac]
Saunders MacLane, Categories for the working mathematician, Springer-Verlag, New York-Berlin, 1971, Graduate
Texts in Mathematics, Vol. 5. MR 0354798
[Po]
N. Popescu, Abelian categories with applications to rings and modules, Academic Press, London, 1973, London Math-
ematical Society Monographs, No. 3. MR 49 #5130
[Ro]
D. Rogalski, Blowup subalgebras of the Sklyanin algebra, Adv. Math. 226 (2011), no. 2, 1433 -- 1473. MR 2737790
(2012d:16083)
[Roos] Jan-Erik Roos, Derived functors of inverse limits revisited, J. London Math. Soc. (2) 73 (2006), no. 1, 65 -- 83.
MR 2197371
[Ros] Alexander L. Rosenberg, The spectrum of abelian categories and reconstruction of schemes, Rings, Hopf algebras, and
Brauer groups (Antwerp/Brussels, 1996), Dekker, New York, 1998, pp. 257 -- 274. MR 99d:18011
[Rot]
Joseph J. Rotman, An introduction to homological algebra, second ed., Universitext, Springer, New York, 2009.
MR 2455920 (2009i:18011)
[Si]
Susan J. Sierra, Rings graded equivalent to the Weyl algebra, J. Algebra 321 (2009), no. 2, 495 -- 531. MR 2483278
(2010b:16084)
[Sm1]
S. Paul Smith, Subspaces of non-commutative spaces, Trans. Amer. Math. Soc. 354 (2002), no. 6, 2131 -- 2171.
MR 1885647
[Sm2]
, Maps between non-commutative spaces, Trans. Amer. Math. Soc. 356 (2004), no. 7, 2927 -- 2944 (electronic).
MR MR2052602 (2005f:14004)
[Sm3]
, Corrigendum to "Maps between non-commutative spaces"[ MR2052602], Trans. Amer. Math. Soc. 368 (2016),
no. 11, 8295 -- 8302. MR 3546801
[VdB] Michel Van den Bergh, Blowing up of non-commutative smooth surfaces, Mem. Amer. Math. Soc. 154 (2001), no. 734,
x+140. MR MR1846352 (2002k:16057)
26
|
1510.04024 | 2 | 1510 | 2015-10-19T07:52:19 | Quotients of degenerate Sklyanin algebras | [
"math.RA"
] | In this paper it is shown how the Heisenberg group of order 27 can be used to construct quotients of degenerate Sklyanin algebras. These quotients have properties similar to the classical Sklyanin case in the sense that they have the same Hilbert series, the same character series and a central element of degree 3. Regarding the central element of a 3-dimensional Sklyanin algebra, a better way to view this using Heisenberg-invariants is shown. | math.RA | math |
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
KEVIN DE LAET
Abstract. In this paper it is shown how the Heisenberg group of order 27
can be used to construct quotients of degenerate Sklyanin algebras. These
quotients have properties similar to the classical Sklyanin case in the sense
that they have the same Hilbert series, the same character series and a central
element of degree 3. Regarding the central element of a 3-dimensional Sklyanin
algebra, a better way to view this using Heisenberg-invariants is shown.
Contents
Introduction
1.
1.1. Notation
2. G-algebras
2.1. Character series
3. The finite Heisenberg group of order 27
3.1. 3-dimensional Sklyanin algebras are H3-deformations
4. Central elements and Heisenberg invariants
5. Quotients of non-regular quadratic algebras
5.1. The Hilbert series
5.2. Point modules of Tt
5.3. The central element
5.4. An analogue of the twisted coordinate ring
5.5. Connection with the Clifford algebra
5.6. Representation theory
6. The algebra S[1:0:0]
7. The controlling variety
8. The bad case t = 0, ∞
References
1
2
3
4
4
5
5
7
8
10
11
11
14
15
16
17
18
19
1. Introduction
The 3-dimensional Sklyanin algebras form an important class of noncommutative
graded algebras, as they correspond to the notion of a noncommutative P2 following
Artin, Tate, Van den Bergh and others (see for example [3] and [4]). These algebras
/ P2 and a point τ ∈ E. They form
are parametrized by an elliptic curve E
the largest class of examples of quadratic 3-dimensional AS-regular algebras, that
is, graded algebras of global dimension 3 with relations in degree 2 with excellent
homological properties. These AS-regular algebras can be described as the quotient
1
/
2
KEVIN DE LAET
of Chx, y, zi by the relations
(1.1)
ayz + bzy + cx2,
azx + bxz + cy2,
axy + byx + cz2,
with [a : b : c] ∈ P2 not one of the 12 points of
(1.2)
{[0 : 0 : 1], [0 : 1 : 0], [1 : 0 : 0]} ∪ {[a : b : c] ∈ P2a3 = b3 = c3 = 1}.
It was remarked in many early papers (see for example [2]) about these algebras
that there was a central element of degree 3, which was somewhat mysterious. In
[7], a intrinsic presentation of this central element was found.
It turns out that
the central element gave a connection between the Sklyanin algebra Aτ (E) and the
algebra A−2τ (E) if E is an elliptic curve and τ ∈ E. This was proved using the
concept of superpotentials, as explained in for example [12].
The first purpose of this paper is to give a better statement of this theorem
in terms of Heisenberg invariants and the degree 3 part of the Koszul dual of a
Sklyanin algebra.
1
The second purpose is the study of quotients of the 12 nonregular algebras. In
particular, we show that there is a 1-dimensional family of quotients of each of
these 12 algebras parametrized by C∗ such that the quotients have Hilbert series
(1−t)3 . In addition, these algebras also have a central element of degree 3, fixed
by the Heisenberg group. We also show that for the constructed quotients of the
algebra Chx, y, zi/(x2, y2, z2), the nth roots of unity give quotients that are finite
modules over their center of PI-degree 2n.
1.1. Notation. In this article, we use the following notations:
• V(I) for I ⊂ C[a1, . . . , an] an ideal is the Zariski-closed subset of An or
Pn−1 determined by I, it will be clear from the context if the projective or
affine variety is used.
• D(I) for I an ideal I ⊂ C[a1, . . . , an] is the open subset An \ V(I) or
Pn−1 \ V(I), it will be clear from the context if it is an open subset of affine
space or of projective space. If I = (a), then we write D(a) for D(I).
• Zn = Z/nZ for n ∈ N.
• Grass(m, n) will be the projective variety parametrizing m-dimensional vec-
tor spaces in Cn.
• For an algebra A and elements x, y ∈ A, {x, y} = xy + yx and [x, y] =
xy − yx.
• For V a n-dimensional vector space, we set T (V ) = ⊕∞
k=0V ⊗k, the tensor
algebra over V .
• Every graded algebra A will be positively graded, finitely generated over C
and connected, that is A0 = C.
• The group SLm(p) (respectively PSLm(p)) is the special linear group (re-
spectively projective special linear group) of degree m over the finite field
with p elements.
• For any vector space V , C[V ] = T (V )/(wv − vww, v ∈ V ).
• If A is a connected, finitely generated, positively graded algebra, then the
Hilbert series is defined as HA(t) =P∞
k=0 dim Aktk.
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
3
• Take a reductive group G and 2 finite dimensional representations V, W of
G. Then EmbG(V, W ) is the set of injective linear G-maps from V to W
up to G-isomorphisms of V .
• Modules will always be left modules unless otherwise mentioned.
2. G-algebras
This section is a summary of the general theory developed in [6].
Definition 2.1. Let G be a reductive group. We call a positively graded connected
algebra A, finitely generated in degree 1, a G-algebra if G acts on it by gradation
preserving automorphisms.
This definition implies that there exists a finite dimensional representation V
of G such that T (V )/I ∼= A with I a graded ideal of T (V ), which is itself a G-
subrepresentation of T (V ).
One can make quadratic G-algebras as follows. Let V be a G-representation.
Then V ⊗ V is also a G-representation which decomposes as a summation of simple
representations, say V ⊗ V ∼= ⊕m
i where the Si are distinct simple representa-
tions of G and ai ≥ 0. A G-algebra A is then constructed by taking embeddings of
the Si in V ⊗ V as relations of A.
i=1Sai
One can of course do the same for other degrees by taking relations in T (V )i =
j and take different embeddings of the simple representations of G
j=1Saj
V ⊗i ∼= ⊕m
in V ⊗i as relations.
Definition 2.2. Let A be a G-algebra with corresponding ideal I of T (V ). We
call B a G-deformation of A up to degree k if B is also a quotient of T (V ) such
that ∀1 ≤ i ≤ k : Ai ∼= Bi as G-representations. We will call B a G-deformation if
∀i ∈ N : Ai ∼= Bi as G-representations.
If the relations for A are all of the same degree k, then all G-deformations up
to degree k of A depend on a product of Grassmannians. For example, let A be
a quadratic algebra of which we want to find all G-deformations up to degree 2.
Let I2 = ⊕m
i with Si distinct simple representations
and 0 ≤ ei ≤ ai natural numbers. Then the G-deformations up to degree 2 are
parametrized by EmbG(⊕m
i ⊂ V ⊗ V = ⊕m
i=1Sai
i=1Sei
i=1 Grass(ei, ai).
i=1Sei
i , ⊕m
i=1Sai
In general, the total set of G-deformations up to degree k of a G-algebra A =
i ) =Qm
T (V )/I are determined by a Zariski closed subset of
Zk =
Grass(ei,j, ai,j)
k
Yj=1 YSi simple
with Ij = ⊕Si simpleSei,j
i ⊂ T (V )i = ⊕Si simpleSai,j
i
Definition 2.3. We say that a variety Z parametrizes G-deformations up to de-
/ Zk can be embedded in Zk and the point
gree k of a G-algebra A if Z
corresponding to A in Zk belongs to the image of φ. We say that Z parametrizes
G-deformations of A if Z parametrizes G-deformations up to degree k for some k
and for each point x ∈ Z with corresponding algebra Ax, we have
φ
∀i ∈ N : (Ax)i ∼= Ai as G-representations.
/
4
KEVIN DE LAET
We will show in the next section that the 3-dimensional Sklyanin algebras are
H3-deformations of the polynomial ring C[V ]. We first show a computational way
to decode how a G-algebra decomposes as a G-module.
2.1. Character series. Given a G-algebra A, it is a natural question to ask how
A behaves as a G-module. As G acts as gradation preserving automorphisms, we
have a decomposition
∞
A =
Sek,S
Mk=0 MS simple
with almost all ek,S equal to 0. We will only consider the case that G is finite.
Definition 2.4. Let G be a finite group. The character series for an element g ∈ G
and for a G-algebra A is a formal sum
ChA(g, t) = Xn∈Z
χAn (g)tn.
For example, if g = 1, then we have ChA(1, t) = HA(t), the Hilbert series of A.
As a character of a representation is constant on conjugacy classes, we can represent
the decomposition of A in simple G-representations as a vector of length equal to
the number of conjugacy classes and on the ith place the character series ChA(g, t)
with g ∈ Ci, the ith conjugacy class.
Lemma 2.5. Let V be a simple representation of G and let A be a G-algebra
constructed from T (V ). For every element z of the center, we have that ChA(z, t) =
HA(λt), where z acts on V by multiplication with λ.
Proof. It follows that in degree k the action of z on Ak is given by multiplication
with λk, so the character series for the element z in this case is given by
ChA(z, t) =
∞
Xk=0
λk dim Aktk = HA(λt).
(cid:3)
3. The finite Heisenberg group of order 27
While in previous papers (see [5] and [6]) we needed the finite Heisenberg group
of order p3 for any odd prime p, we will only consider here the special case p = 3.
Definition 3.1. The Heisenberg group of order 27 is the finite group given by
generators and relations
H3 = he1, e2 [e1, e2] central, e3
1 = e3
2 = 1i.
H3 is a central extension of the group Z3 × Z3 with Z3,
(3.1)
1
/ Z3
/ H3
/ Z3 × Z3
/ 1 .
H3 has 9 1-dimensional representations coming from the quotient H3/([e1, e2]) =
Z3 × Z3 and 2 3-dimensional simple representations, corresponding to the primitive
3rd roots of unity. These 2 representations are defined in the following way: let ω
be a primitive 3rd root of unity and let V1 = C3 = Cx0 + Cx1 + Cx2. Then the
action is defined by
(3.2)
e1 · xi = xi−1,
e2 · xi = ωixi.
/
/
/
/
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
5
V ∗ is the representation corresponding to ω2 and will be denoted by V2. We will
use χa,b for the character H3
/ C defined by
χa,b
(3.3)
χa,b(e1) = ωa, χa,b(e2) = ωb.
There is an action of SL2(3) on H3 as group automorphisms by the rule
(cid:20)A B
C D(cid:21) · e1 = eA
1 eC
2 ,(cid:20)A B
C D(cid:21) · e2 = eB
1 eD
2 .
The central element [e1, e2] is fixed by this action. From this it follows that the
induced action of SL2(3) on the simple representations fixes V1 and V2.
3.1. 3-dimensional Sklyanin algebras are H3-deformations. Using the con-
struction of Section 2, we will now show that the 3-dimensional Sklyanin algebras
are H3-deformations of C[V ] with V = V1 as defined above.
Write A = C[V ] = T (V )/(V ∧ V ).
In order to find all H3-deformations up
to degree 2 of A, we need to decompose V ⊗ V in H3-representations. A quick
calculation shows that V ⊗ V = (V ∗)3 and that the H3-generators of V ⊗ V are
given by
(3.4)
x1x2 − x2x1, x1x2 + x2x1, x2
0.
The first generator corresponds to the wedge product V ∧ V . Taking another copy
of V ∗ in V ⊗ V corresponds to an element of Grass(1, 3). Given p = [A : B : C] ∈
Grass(1, 3) = P2, then p determines the quotient Chx0, x1, x2i/(I) with I generated
by the relations
A(x1x2 − x2x1) + B(x1x2 + x2x1) + C(x2
A(x2x0 − x0x2) + B(x2x0 + x0x2) + C(x2
A(x0x1 − x1x0) + B(x0x1 + x1x0) + C(x2
0),
1),
2).
Putting a = A + B, b = B − A and c = C, one gets the familiar relations of the
3-dimensional Sklyanin algebras as in equation 1.1.
In particular, the 3-dimensional Sklyanin algebras have the same Hilbert series
as the polynomial ring in 3 variables. Using the results of [6], we see that the
character series with respect to H3 of any Sklyanin algebra B is the same as the
character series of A. In fact, this is true for all Artin-Schelter regular algebras
parametrized by points of EmbH3 (V ∗, (V ∗)3).
In addition, the SL2(3) (left) action on H3 induces a (right) action on
EmbH3 (V ∗, (V ∗)3) ∼= P2. This projective representation of SL2(3) has the property
that points lying in the same orbit determine isomorphic algebras. In particular,
the only non-regular algebras are those lying in the SL2(3)-orbit of either the point
[1 : 0 : 0] or [0 : 0 : 1]. For more information, see amongst others [1], [5].
4. Central elements and Heisenberg invariants
In [7], it was proved that there was a connection between the superpotential
defining the Sklyanin algebra A−2τ (E) and the central element c3 of degree 3 in
Aτ (E). This connection however can better be explained using Heisenberg invariant
elements of V ⊗ V ⊗ V .
/
6
KEVIN DE LAET
Let p = [a : b : c] and define Sp = Chx, y, zi/(Rp) to be the algebra with relations
ayz + bzy + cx2,
azx + bxz + cy2,
axy + byx + cz2.
Rp =
Let Wp ⊂ V ⊗ V be the vector space spanned by these relations, with V = Cx +
Cy + Cz. Then Wp ⊗ V ∩ V ⊗ Wp is generically a 1-dimensional vector space,
generated by
a(zxy + xyz + yzx) + b(yxz + zyx + xzy) + c(x3 + y3 + z3).
This element is easily seen to be fixed by H3. In turn, any element g ∈ P((V ⊗ V ⊗
V )H3 ) determines quadratic relations by taking the cyclic derivatives δx, δy, δz as
in [7]. For AS-regular algebras, Wp ⊗ V ∩ V ⊗ Wp is 1-dimensional, so on the open
subset P2 \ D with D = Sg∈SL2(3) g · {[1 : 0 : 0], [0 : 0 : 1]}, we have an injective
morphism
P2 \ D
φ
/ P((V ⊗ V ⊗ V )H3 ) = P2
In order for this to extend to EmbH3 (V ∗, (V ∗)3) = P2 and to get an isomorphism,
one should need that (Wp ⊗ V ∩ V ⊗ Wp)H3 is always 1-dimensional. This is
indeed the case. Let χa,b be the 1-dimensional representation of H3 defined by
χa,b(e1) = ωa, χa,b(e2) = ωb.
Theorem 4.1. Let V = V(abc) ⊂ P2
[a:b:c]. Then for each vertex p of V, the
decomposition of Wp⊗V ∩V ⊗Wp in H3-representations is given by χ0,0⊕χ1,0⊕χ2,0.
In particular, (Wp ⊗ V ∩ V ⊗ Wp)H3 is 1-dimensional.
Proof. As these algebras are monomial algebras, it is easy to find a basis of Wp ⊗
V ∩ V ⊗ Wp. We have
• for [0 : 0 : 1], we find Cx3 + Cy3 + Cz3,
• for [1 : 0 : 0], we find Czxy + Cxyz + Cyzx,
• for [0 : 1 : 0], we find Cyxz + Czyx + Cxzy.
In these 3 cases, it is clear that e2 works trivially on these elements and e1 works
by cyclic permutation, leading to the claimed decomposition.
(cid:3)
Now, the other points that correspond to nonregular algebras lie in the SL2(3)-
In order to prove that (Rp ⊗ V ∩ V ⊗ Rp)H3 is indeed
orbit of these 3 points.
1-dimensional, we need to work out what the SL2(3)-orbits are in the set of simple
representations of H3. From the natural action of SL2(3) on Z3 × Z3 we find that
• χ00 is fixed,
• V and V ∗ are fixed because the center is fixed,
• the action is transitive on the set χa,b, (a, b) 6= (0, 0).
Now, the vector spaces of the theorem have one thing in common: the action of e2
is fixed. If one takes the H3-representations Wa,b = χa,b ⊕χ−a,−b, a, b ∈ {0, 1}, then
the center of SL2(3) works trivially on the set {Wa,ba, b ∈ {0, 1}}, so the action is
really a PSL2(3)-action.
From this, we see that the induced action of SL2(3) on the decomposition of
W[1:0:0] ⊗V ∩V ⊗W[1:0:0] or W[0:0:1] ⊗V ∩V ⊗W[0:0:1] sends χ0,0 ⊕W1,0 to χ0,0 ⊕Wa,b
for some a, b ∈ {0, 1}. We have proved
Theorem 4.2. (Wp ⊗ V ∩ V ⊗ Wp)H3 is 1-dimensional for every p ∈ P2.
/
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
7
So φ indeed extends to an isomorphism of P2.
Now, theorem 1 of [7] can be described as
Theorem 4.3. Let A = T (V )/Rp, p = [a : b : c] be a Sklyanin algebra, c3 be the
central element of degree 3 in A and let
(V ⊗ V ⊗ V )H3
π
/ AH3
3
be the natural projection map. Then π−1(Cc3) is a 2-dimensional vector space of
(V ⊗ V ⊗ V )H3 , which corresponds in P((V ⊗ V ⊗ V )H3 ) to the tangent line of the
elliptic curve E at the point p.
Proof. According to [7], the vector space corresponding to the point −2p ∈ P((V ⊗
V ⊗ V )H3 ) is indeed mapped to Cc3. As the vector space
C(a(zxy + xyz + yzx) + b(yxz + zyx + xzy) + c(x3 + y3 + z3))
is the kernel of π, it follows that the vector space generated by p and −2p is indeed
π−1(Cc3). The fact that this is the tangent line to p at E follows as the third point
of intersection of the line through p and −2p is p itself.
(cid:3)
5. Quotients of non-regular quadratic algebras
In the projective plane P2
[a:b:c], there are 12 points where the corresponding alge-
bra is not AS-regular: the SL2(3)-orbit of [1 : 0 : 0] (containing 8 elements) and the
orbit of [0 : 0 : 1] (containing 4 elements). All these algebras have as Hilbert series
1+t
1−2t and are clearly not domains, for a detailed description of these algebras, see
[9] and [11]. However, it seems that these algebras have a 1-dimensional family of
quotients that 'behave' like the 3-dimensional AS-regular algebras in the following
sense:
• the Hilbert series is the same,
• the character series is the same for each element of H3 and
• there exists a central element of degree 3, fixed by the H3-action.
Let us consider the following example: take the Clifford algebra C over C[u0, u1, u2]
with associated quadratic form
u2 u1
0
u2
0
u0
u1 u0
0
.
In terms of generators and relations of C, we have 3 generators x0, x1, x2 with
relations
(x2
1 = x2
0 = x2
[{xi, xi+1}, xi+2] = 0, 0 ≤ i ≤ 2.
2 = 0,
This algebra is a quotient of the algebra S[0:0:1] by 2 elements of degree 3 (adding
2 commutation relations of degree 3 automatically implies the third relation).
Theorem 5.1. The character series of C is the same as the character series of the
polynomial ring in 3 variables.
Proof. Define on C an action of H3 by
e1 · xi = xi−1
e2 · xi = ωixi
e1 · ui = ui−1,
e2 · ui = ω2iui.
/
8
KEVIN DE LAET
C is a free module of rank 8 over C[u0, u1, u2] with basis
{1, x0, x1, x2, x1x2, x2x0, x0x1, x0x1x2}.
It is then easy to compute that under these conditions, C is a graded algebra with
character series equal to the polynomial ring in 3 variables with the standard action
of H3 (see [5] for the AS-regular case, this one is similar).
(cid:3)
In particular, this means that the natural epimorphism
S[0:0:1]
π
/ C
has as kernel an ideal generated by 2 elements of degree 3. The C-vector space
generated by these 2 elements decomposes as H3-representations into χ1,0 ⊕ χ−1,0,
which is the 'H3-surplus' of S[0:0:1] in degree 3.
Considering the representations of C, we have
Theorem 5.2. C is Azumaya over every point of Spec(Z(C)) except over the
trivial ideal (u0, u1, u2).
Proof. C is not Azumaya over a point if and only if the associated quadratic form
after specialization is of rank ≤ 1. Taking the 2 × 2-minors of the quadratic form,
this only happens if u2
(cid:3)
i is mapped to 0 for all 0 ≤ i ≤ 2.
This means that, considering representations, C has more in common with
the Sklyanin algebras associated to points of order 2 than the quantum algebra
C−1[x, y, z], although the last one is AS-regular.
In the next section, we will find a 1-dimensional family of quotients of S[0:0:1]
by degree 3 elements that have the correct character series up to degree 4. we
will show that there is an open subset in this quotient that gives algebras with the
correct character series.
5.1. The Hilbert series. We work for now in the algebra S = S[0:0:1]. V will be
the vector space Cx + Cy + Cz and the action of H3 is defined by
(5.1)
(5.2)
e1 · z = y,
e2 · z = ω2z.
Decomposing the degree 3 part S3 in H3-modules, we find
e1 · y = x,
e2 · y = ωy,
e1 · x = z,
e2 · x = x,
S3 ∼= Xi,j∈Z3
χi,j ⊕ χ0,0 ⊕ χ1,0 ⊕ χ2,0.
In particular, the multiplicity of χ1,0 and χ2,0 is 2 in S3. This means that the
variety parametrizing quotients of S with the right character series up to degree 3
is given by P1 × P1.
Let Ip be the ideal of S generated by
v1 = A1(zxy + ωxyz + ω2yzx) + B1(yxz + ωzyx + ω2xzy),
v2 = A2(zxy + ω2xyz + ωyzx) + B2(yxz + ω2zyx + ωxzy),
with p = ([A1 : B1], [A2 : B2]) ∈ P1 × P1. Let Wp = Cv1 + Cv2. One of the
similarities we want to investigate is whether we can get quotients such that the
Hilbert series is correct, in particular, correct in degree 4. As dim S4 = 24 and
dim C[x, y, z]4 = 15, we need to have that dim(Ip)4 = 9. We have
dim(Ip)4 = dim Wp ⊗ V + dim V ⊗ Wp − dim V ⊗ Wp ∩ Wp ⊗ V.
/
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
9
We don't need to worry about character series up to degree 4 as S4 ∼= V 8. We
need to find Wp such that dim V ⊗ Wp ∩ Wp ⊗ V is 3-dimensional. However, as
this vector space is an H3-representation and isomorphic to V e for some e ∈ N, it
is enough to find Wp such that (V ⊗ Wp ∩ Wp ⊗ V )e2 is 1-dimensional.
Lemma 5.3. S/Ip has the correct Hilbert series up to degree 4 iff p lies on the line
V(A1B2 − A2B1) = ∆ ⊂ P1 × P1.
Proof. The elements fixed by e2 in Wp ⊗V +V ⊗Wp lie in the vector space generated
by
xv1 = A1xzxy + A1ω2xyzx + B1xyxz + B1ωxzyx,
xv2 = A2xzxy + A2ωxyzx + B2xyxz + B2ω2xzyx,
v1x = A1zxyx + A1ωxyzx + B1yxzx + B1ω2xzyx,
v2x = A2zxyx + A2ω2xyzx + B2yxzx + B2ωxzyx.
Then V ⊗ Wp ∩ Wp ⊗ V 6= 0 iff the following matrix has rank ≤ 3:
0 A1ω2 B1
A2ω B2
0
A1
A2
0 A1 A1ω
0 A2 A2ω2
B1ω
0
0 B2ω2
0 B1 B1ω2
0 B2 B2ω
The first 4 × 4-minor is equal to
.
A1B2A1A2(ω2 − ω) − A2B1A1A2(ω2 − ω) = (ω2 − ω)A1A2(A1B2 − A2B1).
If A1 = 0, then we can take B1 = 1 and the 4 × 4-minor given by the columns
(2, 3, 4, 5) becomes
0
0
1
0
A2ω B2
0
1
0 B2
Taking the determinant of this matrix, one gets A2
A2B1 = 0. A similar result is true if we set A2 = 0.
0
0
0
A2 A2ω2
0
.
2ω. So this also implies A1B2 −
If A1B2 − A2B1 = 0, then one immediately checks that
0 A1ω2 B1
0
A2ω B2
A1
A2
0 A1 A1ω
0 A2 A2ω2
0
B1ω
0 B2ω2
0 B1 B1ω2
0 B2 B2ω
= 0.
−A2 A1 −A2 A1(cid:21)
(cid:20)−B2 B1 −B2 B1
As either one of the rows of the first matrix is not 0, we have that dim V ⊗Wp ∩Wp ⊗
V ≥ 3. The only points where this inequality is possibly strict is when either both
A1 and B2 are 0 or both A2 and B1 are 0, this follows from taking the determinants
of
0
A2
A2ω
0 A1 A1ω
0 A2 A2ω2
and
0 B2ω2
B2
0 B1 B1ω2
0 B2 B2ω
.
But if for example A1 = 0, then necessarily A2 = 0, but B2 and A2 can not be 0
at the same time. Similar results hold for the other cases, so we are done.
(cid:3)
10
KEVIN DE LAET
This means that the only points we have to consider lie on the diagonal ∆ ⊂
A for [A : B] ∈ P1 and let It be the
[A1:B1] × P1
P1
ideal in S generated by the elements
[A2:B2]. From now on, we write t = B
(v1)t = A(zxy + ωxyz + ω2yzx) + B(yxz + ωzyx + ω2xzy),
(v2)t = A(zxy + ω2xyz + ωyzx) + B(yxz + ω2zyx + ωxzy),
The next obvious question is whether these algebras parametrized by P1 have
the correct Hilbert series. For C∗ = P1 \ {0, ∞} this is true.
Lemma 5.4. The Clifford algebra C corresponds to taking the quotient for the
value t = −1.
Proof. It is enough to prove that the relation [{x, y}, z] belongs to the vector space
C(v1)−1 + C(v2)−1. This means that the following matrix should have rank 2
ω
ω2
1
1
1
−1
which is indeed true.
1
ω2 −1 −ω −ω2
ω −1 −ω2 −ω
0
−1
0
,
(cid:3)
We will later see that all these algebras can be embedded in a smashed product
C#Z if t 6= 0, ∞, from which it will follow that
Theorem 5.5. Each algebra S/It has the correct Hilbert series if t 6= 0, ∞.
5.2. Point modules of Tt. In [9] the point modules of S were classified. We
can classify the point modules of Tt = S/It using these results. Recall that point
modules of S were determined by the following way: let V = V(XY Z) ⊂ P2 be the
union of 3 lines and let q0, q1, q2 be the intersection points. Then a point module
P of S depends on a point sequence p0p1p2 . . . fulfilling the following requirements
• for every i ∈ N, pi ∈ V,
• if pi 6= qj for any j, then pi+1 is the intersection point not lying on the
same line as pi,
• if pi = qj for some j, then pi+1 is any point on the line opposite of qj.
If we write P = ⊕n∈NCen, then the point pi = [ai : bi : ci] corresponds defining an
action of S on P by
x · ei = aiei+1,
y · ei = biei+1,
z · ei = ciei+1.
Theorem 5.6. The point modules of Tt for t 6= 0, ∞ are parametrized by 6 lines,
call this set W. The isomorphism φ induced by sending a point module P to P [1]
on W is such that φ2 fixes the intersection points and sends each line of W to itself.
Proof. Let P be a point module of Tt with associated point sequence p0p1p2 . . ..
Let pipi+1pi+2 be a subtriple of this sequence.
• Assume that pi is one of the intersection points. Using the Heisenberg
action, we may assume that pi = [1 : 0 : 0]. Then pi+1 = [0 : α : β].
-- Assume that [0 : α : β] 6= [0 : 1 : 0] and [0 : α : β] 6= [0 : 0 : 1]. Then
pi+2 = pi and the relations of S/(It) are trivially fulfilled.
-- Assume that [0 : α : β] = [0 : 1 : 0]. Then pi+2 = [a : 0 : b]. But from
the degree 3 relations it follows that b = 0 and so pi+2 = pi.
-- The case [0 : α : β] = [0 : 0 : 1] is similar to the previous case.
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
11
• Assume that pi is not one of the intersection points. Again using the
Heisenberg-action, we may assume that pi = [0 : α : β], pi+1 = [1 : 0 : 0]
and pi+2 = [0 : γ : δ]. But then it follows from the degree 3 relations that
δα = −tγβ or differently put, φ2 is an isomorphism of V(XY Z) fixing the
intersection points.
(cid:3)
From the proof of this theorem we also notice that if −t is primitive nth root
of unity, then φ2n is the identity. This implies that a point module P of Tt in this
case parametrizes a C∗-family of 2n-dimensional simple representations of Tt.
The point module corresponding to q0q1q0q1 . . . corresponds to the C∗-family
of 2-dimensional simple representations coming from the quotient Tt/(z) =
Chx, yi/(x2, y2). Similar results hold for the Heisenberg-orbit of this point module.
5.3. The central element. One of the many similarities we want for these new
algebras is that there exists a central element of degree 3, fixed by the action of the
Heisenberg group.
Theorem 5.7. For every element t ∈ ∆ ⊂ P1 × P1 there exists a degree 3 central
element in (Tt)3 fixed by H3.
Proof. Using for example MAGMA, one checks that
gt = (zxy + xyz + yzx) + t(yxz + zyx + xzy)
is central in Tt.
(cid:3)
One also checks that gt acts trivially on each point module. These observations
show that the constructed quotients have indeed much in common with the AS-
regular algebras, as each point module of the generic AS-regular algebra is also
annihilated by the unique central element in degree 3.
5.4. An analogue of the twisted coordinate ring. We can now calculate the
Hilbert series of Mt = Tt/(gt).
Theorem 5.8. The Hilbert series of Mt is 1−t3
(1−t)3 except if t = 0, ∞.
Proof. Adding the relation (zxy + xyz + yzx) + t(yxz + zyx + xzy), we can find an
easier basis for the degree 3 relations of Mt. Taking linear combinations, we find
zxy = −tyxz,
xyz = −tzyx,
yzx = −txzy,
for t 6= 0. For t = −1, the corresponding algebra is a quotient of the Clifford algebra,
so in this particular case the Hilbert series is correct. But it then follows that
the Hilbert series is correct for all these cases, as all these algebras are monomial
algebras with the same basis as the quotient of the Clifford algebra.
(cid:3)
Corollary 5.9. gt is not a zerodivisor of Tt.
Proof. This follows directly from Hilbert series considerations and from Theorem
5.5.
(cid:3)
We can find an easy basis for Mt = Tt/(gt) if t 6= 0, ∞.
12
KEVIN DE LAET
Lemma 5.10. A monomial w ∈ Mt is 0 iff any letter is both on an even and an
odd place in w.
Proof. If w is a monomial with not one variable on both an even and an odd place,
then w can never be 0, as the 3 relations preserve the even and odd places of the
variables. So the combinations x2, y2, z2 can never occur in w.
Suppose now that x occurs on both an even and an odd place. There is then a
submonomial of w of the form xzyzy . . . yx or xyzyz . . . zx. But then we can bring
the last x to place 2 in the submonomial, so we get x2 and w becomes 0. Similar
results are true for y and z.
(cid:3)
If we now take a monomial which is not 0 in Mt, then it follows that either
all odd or all even places have the same variable, say x. But then we can 'jump'
over this variable to get 1 variable to the left and the other to the right. Fixing a
lexicographical order x > y > z, we can therefore find a nice basis in the following
way: take 2 variables, say for example x, y, take the basis of C[x, y] an put a z
between each variable. Put z first on the odd places and then the even places.
Apply this procedure 3 times (one for each variable), but remember to discard
6 elements (for xzxz . . ., yzyzyzy . . . and others have been counted twice). For
example, a basis for (Mt)4 is given by
zyzy,
xyxy xyzy
xzxz xzyz
yzyz,
yxyx yxzx zxzx,
zxzy xyxz.
yxyz
In the even cases, we then find a linearly independent set consisting of 3( n
2 +
2 + 1) − 6 = 3n elements, so this is a basis in even degree. In odd degree, we
2 + 1) + 3( n−1
2 + 1) − 6 = 3n, so also in this case the constructed linearly
1) + 3( n
find 3( n+1
independent set is a basis.
The next question is to determine the center of Mt.
Proposition 5.11. Mt has no central elements in odd degree.
Proof. Mt maps surjectively to the algebras Chy, zi/(y2, z2), Chx, zi/(x2, z2) and
Chx, yi/(x2, y2). This means that any central element of odd degree, say w, has to
have all variables in its monomials, as each of these quotients doesn't have central
elements in odd degree, so w has to belong to the kernel of each of these algebra
morphisms. So say that w has as one of its monomials xyxy . . . xzx, with all x
on the odd places. For w to belong to the center, we need to get the first y in
yxyxy . . . xzx to the last place. This is however impossible, as the first y is in a
odd place and the last place is even. Similar results hold for monomials of the form
yxyx . . . zy and zxzx . . . yz. For monomials of the form yx . . . xz with x on every
even place, we have to get the first x in xyxyx . . . xz to the last place, but again
this is impossible as the first place is odd and the last place is even. Similar results
hold for xy . . . yz (y at even place) and yz . . . zx (z at even place). This means that
w does not contain any monomials, so w = 0.
(cid:3)
Theorem 5.12. If −t is a primitive nth root of unity, then the elements (x +
y)2n, (x + z)2n and (y + z)2n generate the center.
Proof. It is clear that these 3 elements belong to the center of Mt, as we have that
(x + y)2x = x(x + y)2, (x + y)2y = y(x + y)2, (x + y)2nz = z(x + y)2n.
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
13
The other claimed elements are central by the fact that the center is stable under
the action of the Heisenberg group.
Consider
the second Veronese subalgebra Pt = M (2)
t
with generators
xy, yx, xz, zx, yz, zy. We then have
(xy)(yx) = 0,
(xy)(xz) = (−t)−1(xz)(xy),
(xy)(zx) = 0,
(xy)(yz) = 0,
(xy)(zy) = (−t)(zy)(xy),
together with Heisenberg orbits (in particular, any word consisting of 3 different
couples is necessarily 0). It then follows that the center of Pt is generated by these
6 monomials of degree 2 to the nth power. Take a degree 2n central element w of
Mt and suppose it contains the monomial (xy)na(xz)nb. We then have
y(xy)na(xz)nb = (yx)nay(xz)nb = (yx)na(zx)nby.
We find that this monomial can only be in w if and only if the monomial
(yx)na(zx)nb also occurs in w, with the same coefficient. But this sum of 2 mono-
mials is equal to
(xy + yx)na(xz + zx)nb.
Similar results hold for the other monomials, so we are done.
(cid:3)
From now on, assume that −t is a primitive nth root of unity. We have (xy +
yx)n(yz + zy)n(zx + xz)n = 0, so the dimension of the center is ≤ 2.
Theorem 5.13. Mt is a finite module over its center.
Proof. Mt is a finite module over Pt, which in turn is a finite module over its center.
The center of Pt is generated by u = (xy)n, u′ = (yx)n, v′ = (xz)n, v = (zx)n, w =
(yz)n, w′ = (zy)n. The center of Mt is then equal to u + u′, v + v′, w + w′. Now,
Z(Pt) is generated as a module over Z(Mt) by the elements 1, u, u′, v, v′, w, w′, as
we have for example
uav′b = (u + u′)(v + v′)(ua−1v′b−1)
so by induction we conclude that Mt is a finite module over its center.
(cid:3)
Corollary 5.14. The dimension of the center is 2.
Proof. We know that the dimension of the center is ≤ 2. If it was < 2, then Mt
would not be a finite module over its center.
(cid:3)
We can now give a description of the simple representations of Mt.
Theorem 5.15. The PI-degree of Mt is 2n.
Proof. As the elements of the center of smallest degree are of degree 2n, it follows
from the Cayley-Hamilton polynomial that the PI-degree of Mt is at least 2n.
If we can now find an open subset of Spec(Z(Mt)) with 2n-dimensional simple
representations, we are done. However, by determining the point modules of Tt
and the observation that each point module is annihilated by gt, we have indeed
found a 2-dimensional family of 2n-dimensional simple representations of Mt, as the
14
KEVIN DE LAET
induced automorphism φ on the point variety of Mt coming from the shift functor
of Proj(Mt) has the property φ2n = Id.
(cid:3)
5.5. Connection with the Clifford algebra. Let λ ∈ C∗, for the moment not
a root of unity. Define an action of the infinite cyclic group Z = C∞ = hσi on the
Clifford algebra C by
σ(x) = λ−1x, σ(y) = y, σ(z) = λz
and take the smash product Q = C#Z, which is graded by the classic grading on
C and deg(σ) = 0. Consider the elements x′ = x#1, y′ = y#σ, z′ = z#σ−1. We
then have
z′x′y′ = (z#σ−1)(x#1)(y#σ) = λ(zxy#1),
x′y′z′ = (x#1)(y#σ)(z#σ−1) = λ(xyz#1),
y′z′x′ = (y#σ)(z#σ−1)(x#1) = λ(yzx#1),
y′x′z′ = (y#σ)(x#1)(z#σ−1) = (yzx#1),
z′y′x′ = (z#σ−1)(y#σ)(x#1) = (zyx#1),
x′z′y′ = (x#1)(z#σ−1)(y#σ) = (xzy#1).
This means that x′, y′, z′ are solutions for the relations of T−λ.
Proposition 5.16. The Hilbert series of the algebra Q generated by x′, y′, z′ is
(1−t)3 and all the relations holding in this algebra come from degree 3
equal to
relations that hold in T−λ.
1
Proof. The algebra C = C#Z is defined by generators and relations by
C = Chx, y, z, t, t−1i/(x2, y2, z2, [{x, y}, z], [{y, z}, x], tx − λ−1xt, ty − yt, tz − λzt).
We give C the gradation determined by deg(x) = deg(y) = deg(z) = 1, deg(t) = 0.
Then a basis in degree k is determined by fixing a monomial basis Wk for C in
degree k and taking all powers of t
{f #tmf ∈ Wk, m ∈ Z}.
Now, write an element f ∈ Wk as f (x, y, z). Then the elements f (x′, y′, z′) are
also linearly independent: let a reps. b, c be the number of times x, resp. y, z is in
f (x, y, z). Then f (x′, y′, z′) is equal to λµf (x, y, z)#tb−c for some µ ∈ Z.
The set {f (x′, y′, z′)f ∈ Wk} forms a basis of Qk: let g(x, y, z)#tr be an element
of Qk with g(x, y, z) a monomial of degree k, g 6= 0. Then necessarily r = b − c
where b is the number of times y occurs in g and c is the number of times z occurs
in g. g can be written as a unique linear combination of elements in Wk, say using
the terms f1, . . . , fi. However, due to the fact that in the defining relations of C the
number of times a variable occurs does not change (unless the monomial becomes
0), this linear combination has the following property: for each fi, the number of
occurrences of x, y, z stays the same. But then g(x, y, z)#tb−c can be written as
a linear combination of fi(x, y, z)#tb−c = λ−µf (x′, y′, z′). So the Hilbert series
indeed stays the same.
The fact that the only relations are of degree 3 follows as the only relations
between monomials can occur if all variables occur and the relations come from
[{x, y}, z] and [{y, z}, x].
(cid:3)
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
15
Corollary 5.17. The algebra T−λ can be embedded in a smash product C#Z.
From this, it follows that Theorem 5.5 is proved.
When λ is a primitive root of unity of order n, we can take the same action of
Zn = hσi on C, take the smash product C#Zn and take the same elements x′, y′, z′
as generators of a subalgebra of C#Zn. These 3 elements then fulfil the relations
of T−λ and again we have found a subalgebra of P isomorphic to T−λ.
In the case that λ is a primitive nth root of unity, we can lift the 3 linearly
independent central elements of degree n of T−λ/(g−λ).
Proposition 5.18. The element (x′z′ + z′x′)n belongs to the center of C#Zn.
Proof. We have
x′z′ + z′x′ = (x#1)(z#σ−1) + (z#σ−1)(x#1) = (xz + λzx)#σ−1.
xz and zx are fixed by σ, so we get
((xz + λzx)#σ−1)n = (xz)n + λn(zx)n#1 = (xz + zx)n#1
which is indeed central.
(cid:3)
This implies that (xz + zx)n belongs to the center of T−λ. Then we can use the
Heisenberg action in T−λ to see that (yz + zy)n and (xy + yx)n are also central in
T−λ. From [10, Lemma 3.6], we now deduce
Theorem 5.19. The center of T−λ with λ a primitive nth root of unity is generated
by 1 element of degree 3 g−λ and 3 linear independent elements of degree 2n, say
u, v, w, with one relation of the form uvw = αg2n for some α ∈ C∗. T−λ is also a
finite module over its center.
Proof. All the conditions of the mentioned lemma are satisfied: g−λ is regular and
the image of C[u, v, w] generates the center of M−λ. Moreover, as we have in M−λ
that the relation in the center is uvw = 0, it follows that the only relation is of the
form uvw = αg2n. α is not 0 as (x′z′ + z′x′)n(x′y′ + y′x′)n(y′z′ + z′y′)n is not 0 in
C#Zn.
(cid:3)
5.6. Representation theory. Fix λ a primitive root of unity of order n, n 6= 1.
Let P = Pλ be the smash product of C with Zn defined in last section and let
Q = Qλ = T−λ be the subalgebra of P coming from a quotient of S.
Theorem 5.20. The center of P is isomorphic to the ring C[a, b, c, d, e]/(ae −
d2, bc − en).
Proof. The center of C is generated by (x + y)2, (y + z)2, (x + z)2, g = zxy − yxz
with one relation of the form (x + y)2(y + z)2(x + z)2 = αg2 for some α ∈ C∗.
The elements in Z(C) fixed by σ are xz + zx, (yz + zy)n, (yx + xy)n, g and
(yz + zy)(yx + xy). But then the relations become
(yz + zy)((xz + zx)(yx + xy)) = αg2,
(xz + zx)n(yx + xy)n = ((xz + zx)(yx + xy))n.
Up to a scalar, these are the relations of the claimed ring.
(cid:3)
Corollary 5.21. The PI-degree of P is 2n.
16
KEVIN DE LAET
Proof. P is a free module of rank n over C. C is a module of rank 4 over its center
Z(C), which in turn is a module of rank n over its ring of invariants Z(C)Zn . The
claim follows.
(cid:3)
Proposition 5.22. The PI-degree of Q is 2n.
Proof. We know that the PI-degree is at least 2n, as we have found simple 2n-
dimensional representations coming from the quotient of Q by the degree 3 cen-
tral element g−λ. As Z(P ) is a finite Z(Q)-module of rank n, every simple 2n-
dimensional representation of P induces a representation of Q and the induced
map Spec(Z(P ))
/ Spec(Z(Q)) is surjective. But then there is an open sub-
set of Spec(Z(Q)) containing simple 2n-dimensional representations. This proves
the claim.
(cid:3)
In fact, by using the (C∗)3-action as gradation preserving algebra automor-
phisms, we find
Proposition 5.23. The Azumaya locus of Spec(Z(Q)) is equal to Spec(Z(Q))
minus 3 lines, the 2 by 2 intersections of the 3 planes V(u), V(v), V(w).
Proof. The (C∗)3-action divides Spec(Z(Q)) into 8 orbits: 1 of dimension 3 (the
largest), 3 of dimension 2, 3 of dimension 1 and 1 of dimension 0. The first orbit is
open and therefore intersects Azu2n(Q), but then this orbit is contained in Azu2n(Q).
The 3 orbits of dimension 2 are the ones coming from simple representations of Mt
of dimension 2n, so these also belong to Azu2n(Q). The orbits corresponding to
lines can not belong to Azu2n(Q), as there are simple 2-dimensional representations
coming from the quotient Q/(z) = Chx, yi/(x2, y2).
(cid:3)
6. The algebra S[1:0:0]
In this case we can be brief: S[1:0:0] is a Zhang-twist of S[0:0:1] by the auto-
morphism x 7→ z 7→ y 7→ x. Recall that if A is a graded algebra with a graded
automorphism φ, then the Zhang twist is defined as the algebra with the same
generators but with multiplication rule a ∗ b = aφi(b) if b ∈ Ai. A Zhang twist of a
graded algebra preserves the Hilbert series.
Proposition 6.1. S[1:0:0] has a 1-dimensional family of quotients parametrized by
C∗ such that these quotients have the right Hilbert series. The relations of degree 3
are defined by
(v1)t = (zyx + ωxzy + ω2yxz) + t(y3 + ωz3 + ω2x3),
(v2)t = (zyx + ω2xzy + ωyxz) + t(y3 + ω2z3 + ωx3).
The central element becomes gt = (zyx + xzy + yxz) + t(y3 + z3 + x3)
Proof. We find that
x ∗ x ∗ x = xzy, y ∗ y ∗ y = yxz, z ∗ z ∗ z = zyx,
z ∗ y ∗ x = zxy, y ∗ x ∗ z = yzx, x ∗ z ∗ y = xyz
and the other 3 monomials become 0. It is then clear that the relations from the
proposition are really the Zhang twists of the relations in S[0:0:1].
If we calculate (zyx + xzy + yxz) + t(y3 + z3 + x3) in S[1:0:0], we find
zxy + xyz + yzx + t(yxz + zyx + xzy),
/
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
17
which is central in the original quotient of S[0:0:1]. As this element is also fixed
under φ, we are done.
(cid:3)
7. The controlling variety
In order to get algebras with the correct Hilbert series up to degree 4, we replaced
the point [0 : 0 : 1] with a P1. One of course hopes that the corresponding variety
parametrizing these algebras is just the blow-up of P2 in this point.
Theorem 7.1. The variety parametrizing H3-deformations up to degree 3 of the
polynomial ring C[x, y, z] with the correct Hilbert series up to degree 4 is the blow-up
of P2 in 12 points.
Proof. We will show that the variety
Z ⊂ EmbH3 (V ∗, (V ∗)3) × EmbH3 (χ2
1,0, χ3
1,0) ∼= P2 × P2
parametrizing nice quotients with the right multiplicity of χ1,0 in degree 3 is the
blow-up of P2 in the H3-orbit of [0 : 0 : 1].
Let R correspond to the relations
Let φ be the subspace of V ⊗ V ⊗ V generated by
ayz + bzy + cx2,
azx + bxz + cy2,
axy + byx + cz2.
(A(x3 + ωy3 + ω2z3) + B(zxy + ωxyz + ω2yzx) + C(yxz + ωzyx + ω2xzy),
D(x3 + ωy3 + ω2z3) + E(zxy + ωxyz + ω2yzx) + F (yxz + ωzyx + ω2xzy)
such that the matrix
(cid:20)A B C
D E F(cid:21)
has rank 2. Decomposing V ⊗ R + R ⊗ V in H3-representations, we find that the
(R, φ) belongs to Z if and only if the following matrix has rank 2
aω2
c
c
aω
A B
D E
M =
.
bω
bω2
C
F
We may assume that c = 1, putting a = 1 or b = 1 gives similar results. Put
a01 = AE − BD, a20 = CD − AF, a12 = BF − CE. M has rank 2 if and only if
a12 + aωa20 + bω2a01 = 0,
a12 + aω2a20 + bωa01 = 0.
From this it follows that aa20 = ba01, which is indeed the equation for the blow-up
of P2 in [a : b : c] = [0 : 0 : 1]. The same result holds for the points [1 : 0 : 0] and
[0 : 1 : 0], call Z1,0 the blow-up of P2 in these 3 points.
We could now do the same for the representation χ2,0 to get Z2,0, which is
also isomorphic to P2 blown-up in 3 points. However, in order to get the right
Hilbert series up to degree 4, we have seen that we need to take the 'diagonal'
∆ ⊂ Z1,0 ×P2 Z2,0, which is of course just the blow-up of P2 in 3 points.
For the other 9 points, we can use the SL2(3)-action.
(cid:3)
18
KEVIN DE LAET
8. The bad case t = 0, ∞
In the blow-up of P2 in 12 points, there are still 24 points where the Hilbert
series of the corresponding algebras explodes. Still, it would be useful to know
what the Hilbert series of these algebras are. As all these algebras are isomorphic
to each other (or isomorphic to a Zhang twist) by courtesy of the SL2(3)-action, it
is enough to calculate the Hilbert series of the algebra
C = Chx, y, zi/(x2, y2, z2, zxy + ωxyz + ω2yzx, zxy + ω2xyz + ωyzx).
Proposition 8.1. The element g0 = zxy + xyz + yzx fulfils the conditions
g0x = xg0 = g0y = yg0 = g0z = zg0 = 0.
Proof. Using the Heisenberg action, it is enough to prove this for x. Then it is an
easy computer calculation with for example MAGMA.
(cid:3)
It is therefore enough to compute the Hilbert series of
A = C/(g0) = Chx, y, zi/(x2, y2, z2, xyz, yzx, zxy).
This algebra has the advantage that 2 words w and w′ can only be equal to each
other if and only if w = w′ = 0. From this it follows that each An has a unique
monomial basis Bn on which e1 acts by a permutation of order 3, without fixed
points. This means that
#{w ∈ Bnw begins with x} = #{w ∈ Bnw begins with y}
= #{w ∈ Bnw begins with z}.
We know from Lemma 5.3 that the Hilbert series starts with the terms
HA(t) = 1 + 3t + 6t2 + 9t3 + 15t4 + . . .
Theorem 8.2. Let f (t) = HA(t) = P∞
n=0 antn be the Hilbert series of HA(t)
with an = dim An. Then the coefficients of f (t) fulfil the recurrence relation an =
2an−1 − an−3 for n ≥ 4.
Proof. For n = 4, we are done by Lemma 5.3. Let n ≥ 5. Let
An−1
fx
/ An
be the linear map defined by w 7→ xw. It then follows that an = 3 dim Im(fx). So
we need to calculate ker(fx). Due to the relations, ker(fx) is the subspace of An−1
spanned by the words beginning with x or yz. The space spanned by the words
beginning with x has dimension an−1
. The space spanned by the words beginning
with yz is the image of the map
3
fyz
An−3
/ An−1 .
The kernel of fyz is the subspace spanned by the monomials in An−3 starting with
x or z. So dim Im(fyz) = an−3 − 2an−3
. This gives
3 = an−3
3
an = 3 dim Im(fx)
= 3(an−1 − dim ker(fx))
= 3(an−1 −(cid:16) an−1
3
= 2an−1 − an−3.
+
an−3
3 (cid:17)
/
/
QUOTIENTS OF DEGENERATE SKLYANIN ALGEBRAS
19
(cid:3)
Corollary 8.3. For n ≥ 3, we have the recurrence relation an = an−1 + an−2
Proof. n = 3 is trivial. By induction, we may assume that an−1 = an−2 + an−3.
But then we get
an = 2an−1 − an−3 = an−1 + (an−2 + an−3) − an−3 = an−1 + an−2.
(cid:3)
Corollary 8.4.
HA(t) =
3
1 − (t + t2)
− 2
Unfortunately, this is not exactly the Fibonacci sequence. However, if we take
B = Ae1 , then we get
with Fn the nth Fibonacci number starting from F0 = 1, F1 = 1.
bn = dim Bn = Fn
References
[1] Tarig Abdelgadir, Shinnosuke Okawa and Kazushi Ueda, Compact moduli of noncommutative
projective planes, http://arxiv.org/abs/1411.7770
[2] Michael Artin and William F. Schelter, Graded algebras of global dimension 3, in Advances
in Mathematics, 66, 2, 171-216, Elsevier (1987)
[3] Michael Artin, John Tate and Michel Van den Bergh, Some algebras associated to automor-
phisms of elliptic curves, in The Grothendieck Festschrift I, Progress in Math. 86, 33-85,
Birkhauser Boston (1990)
[4] Michael Artin, John Tate and Michel Van den Bergh Modules over regular algebras of di-
mension 3, Inventiones mathematicae 106, 1, 335-388, Springer (1991)
[5] Kevin De Laet, Graded Clifford algebras of global dimension p with an action of Hp
http://arxiv.org/abs/1401.7800
[6] Kevin De Laet, Character
series and Sklyanin algebras at points of order 2
http://arxiv.org/abs/1412.7001
[7] Kevin De Laet and Lieven Le Bruyn, The geometry of representations of 3-dimensional
Sklyanin algebras, http://arxiv.org/abs/1405.1158
[8] Naihuan Jing and James J Zhang, On the trace of graded automorphisms, Journal of Algebra
189, 2, 353-376, Elsevier (1997)
[9] S. Paul Smith, Degenerate 3-dimensional Sklyanin algebras are monomial algebras, in Journal
of Algebra, 358, 74-86, Elsevier (2012)
[10] S. Paul Smith, The center of the 3-dimensional and 4-dimensional Sklyanin algebras, in
K-theory, 8, 1, 19-63, Springer (1994)
[11] Chelsea Walton,Degenerate Sklyanin algebras and generalized twisted homogeneous coordi-
nate rings, in Journal of Algebra, 322, 7,2508-2527, Elsevier (2009)
[12] Chelsea Walton, Representation theory of three-dimensional Sklyanin algebras, in Nuclear
Physics B, 860, 1, 167-185, Elsevier (2012)
Department of Mathematics, University of Antwerp, Middelheimlaan 1, B-2020
Antwerp (Belgium), [email protected]
|
1701.06936 | 1 | 1701 | 2017-01-24T15:39:09 | Regularly weakly based modules over right perfect rings and Dedekind domains | [
"math.RA"
] | A weak basis of a module is a generating set of the module minimal with respect to inclusion. A module is said to be regularly weakly based provided that each of its generating sets contain a weak basis. In the paper we study (1) rings over which all modules are regularly weakly based, refining results of Nashier and Nichols, (2) regularly weakly based modules over Dedekind domains. | math.RA | math |
REGULARLY WEAKLY BASED MODULES OVER RIGHT
PERFECT RINGS AND DEDEKIND DOMAINS
MICHAL HRBEK, PAVEL RŮŽIČKA
Abstract. A weak basis of a module is a generating set of the module min-
imal with respect to inclusion. A module is said to be regularly weakly based
provided that each of its generating sets contain a weak basis. In the paper
we study
(1) rings over which all modules are regularly weakly based, refining results
of Nashier and Nichols,
(2) regularly weakly based modules over Dedekind domains.
1. Introduction
By a module we always mean a right module over a ring R. Let M a module
and let X, Y by subsets of M . We say that the set X is weakly independent over
Y if x 6∈ Span((X \ {x}) ∪ Y ) for all x ∈ X. We say shortly that X is weakly
independent in the case of Y = ∅. A generating weakly independent subset of a
module M is called a weak basis of M . A module M is weakly based if it contains a
weak basis. Finally, a module M is called regularly weakly based if any generating
set of M contains a weak basis.
Nashier and Nichols characterized right perfect rings as rings over which every
quasi-cyclic right R-module (i.e. every finitely generated submodule is contained
in a cyclic one) is cyclic (i.e. every submodule is contained in a cyclic one). As
a consequence of this they have got that rings over which all right modules are
regularly weakly based are necessarily right perfect and they raised a question
whether, conversely, all modules over right prefect rings are regularly weakly based.
We refine their result proving that infinitely generated free modules over non-right
perfect rings are not regularly weakly based and we observe that their question
regarding right perfect rings easily reduces to semisimple rings.
The other topic of the paper is a study of regularly weakly based modules over
Dedekind domains. This is motivated by the characterization of weakly based
modules over, first abelian groups [3] and then Dedekind domains [4] done by the
authors. For regularly weakly based modules we will not obtain the full character-
ization, however we reduce the problem to a question of characterizing regularly
weakly based modules over commutative semisimple rings, which indeed is a spe-
cial case of the more general open question regarding right perfect rings introduced
above.
Date: 21/6/2021.
2000 Mathematics Subject Classification. 13C05, 13F05, 16L30.
Key words and phrases. Weak basis, regularly weakly based, Dedekind domains, perfect rings.
The first author is partially supported by the project SVV-2015-260227 of Charles University
in Prague. Both the authors were partially supported by the Grant Agency of the Czech Republic
under the grant no. GACR 14-15479S..
1
2
MICHAL HRBEK, PAVEL RŮŽIČKA
There is a few simple facts regarding regularly weakly based modules which we
will freely use within the paper. Namely, it is clear that a finitely generated module
is regularly weakly based. Also observe that unlike in case of weakly based modules,
a direct summand of a regularly weakly based module is regularly weakly based.
Also the next elementary lemma, in different variations, will be repeatedly used
with no reference. Its proof is left to the reader.
Lemma 1.1. Let R be ring and M a right R-module. Let X, Y, Z be subsets of M .
Suppose that X is weakly independent over Y ∪ Z and Y is weakly independent over
X ∪ Z. Then X ∪ Y is weakly independent over Z.
2. Modules over right perfect rings
We start with a natural task of characterizing rings R such that all right R-
modules are regularly weakly based. We refine the result of [6] that all such a
modules must be right perfect.
In particular, in Lemma 2.2, we prove that an
infinitely generate free module over a non-perfect ring is not regularly weakly based.
Nashier and Nichols suggested conversely, that all modules over right perfect rings
are regularly weakly based. We discuss this question in the final part of this section,
adding an observation that we can factor out the Jacobson radical, and so reduce
the question to semisimple rings.
Lemma 2.1. [6, Proposition 1 and Theorem 2] A ring R is right perfect if and
only if for each sequence (rn n ∈ ω) of elements of R there is n0 ∈ ω such that
for all n ≥ n0 there is j ≥ 1 such that rn+j · · · rn+1R = rn+j · · · rn+1rnR.
Lemma 2.2. Let R be a ring that is not right perfect. Than a free right R-module
is regularly weakly based if and only if it is finitely generated.
Proof. A finitely generated module is regularly weakly based. Thus it suffices to
show that an infinitely generated right free module is not regularly weakly based.
Since a direct summand of a regularly weakly based module is regularly weakly
based, we can restrict ourselves to a countably generated free right R module F =
R(ℵ0).
In order to show that F is not regularly weakly based, fix a free basis
{bn ∈ n ∈ ω} of F . Since R is not right perfect, there is by Lemma 2.1 a sequence
(rn n ∈ ω) of elements of R such that for any n ∈ ω and all j ≥ 1 we have
that rn+j · · · rn+1R ) rn · · · rn+1rnR.
In particular, this implies that all rn are
non-invertible in R. For each n ∈ ω, we define the following elements from F :
xn = bn+1rn
and yn = bn − xn = bn − bn+1rn.
Put Z = {xn, yn n ∈ ω} and Y = {yn n ∈ ω}. Clearly B ⊆ Span(Z), hence
Z generates F . We claim that Z does not contain any weak basis of F . Suppose
otherwise and pick a weak basis W ⊆ Z of F . As rn are non-invertible, Y ⊆ W .
Let n ∈ ω, and suppose xn ∈ W . Observe that then bk and thus also xk = bk −yk
belong to Span(W ) for all k ≤ n. Since W is weakly independent and Y ⊆ W , it
contains at most one xn, that is, W ⊆ Y ∪ {xn} for some n ∈ ω. We claim that
bn+1 /∈ Span(W ). Indeed, otherwise
(2.1)
bn+1 = xns + X
yisi
i∈ω
3
for some s, s0, s1 . . . from R such that all but finitely many si are 0. Substituting
for xn, yi we get that
(2.2)
bn+1 = bn+1rns + X
(bisi − bi+1risi).
i∈ω
From this we get that s0 = · · · = sn = 0, sn+1 = (1 − rns) and sn+1+j =
rn+j · · · rn+1(1 − rns), for all j > 0. Since all but finitely many si equal 0, there
is j > 0 such that sn+1+j = 0. Then we get that rn+j · · · rn+1 = rn+j · · · rn+1rns,
which gives rn+j · · · rn+1R = rn+j · · · rn+1rnR. This contradicts our choice of the
sequence (rn n ∈ ω).
(cid:3)
Recall that a subset I of a ring R is right T-nilpotent provided that for every
sequence a1, a2, . . . there is a positive integer n such that an · · · a1 = 0. A right
ideal J of a ring R is T -nilpotent if and only if M J 6= M for every non-zero right
R-module [1, Lemma 28.3]. By the Theorem of Bass [1, Theorem 28.4] a ring R is
right perfect if and only if its Jacobson radical J is T -nilpotent and the ring R/J
is semisimple.
Lemma 2.3. Let J be a right T -nilpotent right ideal of a ring R, let M be a right
R-module. Then every X ⊆ M lifting a weak basis of M/M J over M J is a weak
basis of M .
Proof. Since X lifts a weak basis of M/M J over M J we have that X is weakly
independent and M = Span(X) + M J. From the second equality we infer that
(M/ Span(X))J = M/ Span(X). Since the ideal J is right T -nilpotent, we conclude
that M/ Span(X) = 0, that is, M = Span(X).
(cid:3)
Proposition 2.4. Let R be a ring.
(1) [6, page 311] If all right R-modules are regularly weakly based, then R is
right perfect.
(2) Let J denote a Jacobson radical of R. If R is right perfect, then all right
R-modules are regularly weakly based if and only if all right modules over
the semisimple ring R/J are regularly weakly based.
Proof. (1) follows readily from Lemma 2.2, while (2) follows from Lemma 2.3. (cid:3)
Proposition 2.4 reduces the characterization of rings over which all modules are
regularly weakly based to a question whether all modules over a semisimple ring
are regularly weakly based. The answer to this seems surprisingly non-trivial.
We conclude the section with a straight consequence of Proposition 2.4.
Corollary 2.5. Every module over a local perfect ring is regularly weakly based.
3. Factoring out a finitely generating submodule
It is easily seen that a module M is weakly based if and only a factor M/K is
weakly based for a finitely generated submodule K of M . The situation became
less apparent when replacing weakly based with a regularly weakly based. We
will apply this fact in the subsequent section. Before we proceed to its proof, we
introduce the following notions (taken from [4]).
Let M, N be modules, let φ : M → N a homomorphism, and let X be a subset
of M . We say that X lifts a subset Y of N via φ provided that φ↾X is a bijection
4
MICHAL HRBEK, PAVEL RŮŽIČKA
onto Y . If N is a quotient module of M , we say shortly that X lifts Y , meaning
that X lifts Y via the canonical projection.
Let M be a module and let X, Y be subsets of M . Let
X Y = {x + Span(Y ) x ∈ X}.
denote the image of the set X in the canonical projection M → M/ Span(Y ).
Lemma 3.1. Let R be a ring, let M be a right R-module and let K be a finitely
generated submodule of M . Then M is regularly weakly based if and only if the
factor module M/K is regularly weakly based.
Proof. First suppose that the module M is regularly weakly based. Let ¯X be a
generating set of M/K, and let X be a subset of M which lifts ¯X, i.e., X K = ¯X.
Then X ∪ K generates M , and since M is regularly weakly based, X ∪ K contains
a weak basis of M , say Y . Since K is finitely generated, there is a finite subset F
of Y such that K ⊆ Span(F ). Put Y0 = Y \ F . Since Y is a weak basis of M , Y K
0
is weakly independent in M/K.
Since Y generates M , the factor-module M/(K + Span(Y0)) is generated by
F K∪Y0. Since finitely generated modules are regularly weakly based, there is F0 ⊆
F that lifts a weak basis of M/(K + Span(Y0)). Since Y0 is weakly independent
over F , K ⊆ Span(F ), and F0 lifts a weak basis of M/(K + Span(Y0)), we conclude
that Y K
is a weak basis of M/K. Since Y0 ∪ F0 ⊆ X ∪ K, we infer that
0 ⊆ ¯X. We have proved that the module M/K is
Y0 ∪ F0 ⊆ X, whence Y K
regularly weakly based.
0 ∪ F K
0
0 ∪ F K
Now suppose that the factor-module M/K is regularly weakly based. Let X be
a generating subset of M . Since K is finitely generated, there is a finite subset F of
X such that K ⊆ Span(F ). The already proved implication gives that M/ Span(F )
is regularly weakly based. Thus we can pick a subset, X0, of X lifting a weak basis
of M/ Span(F ). Observe that F Span(X0) generates the factor-module M/ Span(X0)
and, since a finitely generated module is regularly weakly based, there is F0 ⊆ F
lifting a weak basis of M/ Span(X0). We conclude that X0 ∪ F0 is a weak basis of
M contained in X.
(cid:3)
4. Regularly weakly based modules over Dedekind domains
From now on we restrict ourselves to the case of Dedekind domains. Let R be
a Dedekind domain. We denote by Spec(R) the set of all non-zero prime (and
thus, maximal) ideals of R. An R-module T is torsion if Ann(m) 6= 0 for any
m ∈ T . Recall that any torsion R-module T has a primary decomposition, that
is, T = Lp∈Spec(R) Tp, where Tp = {m ∈ T Ann(m) = pk for some k}. We say
that T is p-primary if T = Tp. Alternatively, the p-primary part Tp correspond
naturally to the localization T ⊗R Rp.
In particular, we can view a p-primary
R-module naturally as a module over the localization Rp.
Let us recall a notion from abelian group theory which will prove useful in what
follows. We say that a submodule B of a p-primary module T is basic if B is a
pure submodule of T , B is isomorphic to a direct sum of cyclic modules, and T /B
is divisible. As all these notions hold the same meaning independent of whether
we view T as an R-module or as an Rp-module, we can use [7, Theorem 9.4] to
infer that any p-primary module has a basic submodule (determined uniquely up
to isomorphism).
Module M is said to be bounded if IM = 0 for some non-zero ideal I.
5
Lemma 4.1. Let R be a Dedekind domain and let T be a torsion R-module. If T
is regularly weakly based, then T is bounded.
Proof. Let T be an unbounded torsion R-module. First suppose that T is p-primary
for some p ∈ Spec(R). We claim that there is a projection from T onto a non-
zero divisible module.
In order to prove this, choose a basic submodule B of T
(existence of which is discussed above). If B ( T , then T /B is nonzero divisible and
T → T /B is the desired projection. If B = T , then T is a direct sum of p-primary
cyclic modules of unbounded annihilators, and hence T contains a submodule S
isomorphic to Ln∈N R/pn.
It is well known that the indecomposable p-primary
divisible R-module can be constructed as a direct limit of the system of inclusions
R/p → R/p2 → R/p3 → · · · , and thus it is a quotient of S. As divisible R-modules
are injective, this projection can be extended to the entire T .
We showed that there is a projection π : T → D, where D is non-zero divisible
module. Denote by K the kernel of π and choose a generating set X ′ of D. Since
D is divisible, there is a subset X of pT lifting X ′ via π. Put Z = X ∪ K and note
that Z generates T . Suppose that W ⊆ Z is a weak basis of T . By [4, Corollary
3.3 and Lemma 5.2], any weak basis of T lifts some basis of T /pT over pT . Hence
W ⊆ K, which is a contradiction to W being a generating set.
Let now T be an unbounded (not necessarily p-primary) torsion R-module. Since
regularly weakly based modules are closed under direct summands, the first part
of this proof implies that Tp is bounded for each p ∈ Spec(R). As T is unbounded,
there must be an infinite subset P of Spec(R) such that Tp 6= 0 for each p ∈ P. If
there is a non-zero divisible subgroup of T , it is a non-weakly based direct summand
of T (see [4, Corollary 3.6]). Thus T is not regularly weakly based. We can thus
assume that T is reduced and apply [8, Theorem 9] to infer that there is a non-zero
cyclic direct summand Cp of Tp for each p ∈ P. Since P is infinite, we can pick a
countable infinite sequence pn, n ∈ ω of pairwise distinct primes from P. It will
suffice to show that Ln∈ω Cpn is not regularly weakly based. Fix a generator xn
of Cpn and put yn = x0 + x1 + · · · + xn for each n ∈ ω.
It follows easily that
Span(ym) ( Span(yn) whenever m < n, and so the generating set {yn n ∈ ω} of
Ln∈ω Cpn does not contain a weak basis.
(cid:3)
Lemma 4.2. Let R be a Dedekind domain, and let p ∈ Spec(R). Every bounded
p-primary R-module is regularly weakly based.
Proof. Let B be a bounded p-primary R-module. Then Bpn = 0 for some positive
integer n and B can be naturally viewed as an R/pn module. Since R is a Dedekind
domain, the factor ring R/pn is local perfect, hence B is regularly weakly based by
Corollary 2.5.
(cid:3)
Before proving the main lemma of the paper, we need the following auxiliary
lemma:
Lemma 4.3. Let R be a Dedekind domain and let N be a torsion-free R-module.
If N is an extension of a free module by a torsion bounded module, then N is
projective.
Proof. Let F be a free submodule of N such that the factor-module B = N/F is
bounded torsion. Enumerate a free basis X = {xα α < λ} of F by an ordinal λ
and put Fβ = Span({xα α < β}) for all β < λ. For each α < λ, let Nα denote
It follows that N = Sα<λ Nα
the smallest pure subgroup of N containing Fα.
6
MICHAL HRBEK, PAVEL RŮŽIČKA
is a filtration of N with Nα+1/Nα torsion-free for each α < λ. Finitely generated
torsion free modules over Dedekind domains are projective, see [9, Theorem 6.3.23],
and therefore it will suffice to prove that all Nα+1/Nα are finitely generated (and
thus, projective). Indeed, then N ≃ Lα<λ Nα+1/Nα and so N is projective.
Put Bα = Nα/Fα for each α < λ. As Fα = Nα ∩ F by the independence of X,
we have the isomorphism Bα = Nα/(Nα ∩ F ) ≃ (Nα + F )/F and so we can view
naturally Bα as a submodule of B. Denote by Q the field of quotients of R. For
each α < λ, we obtain the following commutative diagram:
0 −−−−→ (Nα + Span(xα))/Nα −−−−→ Nα+1/Nα −−−−→ Bα+1/Bα −−−−→ 0
0 −−−−→
≃
y
R
−−−−→
⊆
y
Q
⊆
y
−−−−→ Q/R −−−−→ 0
Both exact sequences in the rows are given by the obvious quotient maps. For the
maps in columns, the left most isomorphism follows from the fact, that Span(xα) ∩
Nα = 0, as Nα is the purification of Xα, and xα 6∈ Xα. The middle inclusion is given
by Nα+1/Nα being torsion-free module of rank 1, and the right-most column map
is induced by the two other ones. It is well known that (Q/R)p is uniserial for each
p ∈ Spec(R). As Bα+1/Bα is bounded, it has only finitely many non-zero p-primary
parts, and as each of them is a bounded submodule of a uniserial module, they are
all finitely generated. Therefore, Bα+1/Bα is finitely generated. We conclude that
Nα+1/Nα is an extension of a cyclic module by a finitely generated module, hence
it is finitely generated. This finishes the proof.
(cid:3)
Lemma 4.4. A regularly weakly based module over a Dedekind domain splits into
a direct sum of a projective module and a bounded torsion module.
Proof. Let M be a regularly weakly based module over a Dedekind domain R.
Let T denote the torsion submodule of M , and let F = M/T be the torsion-free
If F is projective, then M decomposes as T ⊕ F , and both the
quotient of M .
direct summands are regularly weakly based, in particular, T is bounded torsion
by Lemma 4.1.
Suppose now that F is not projective. Then we start with the following claim:
Claim 1. There is an ideal p ∈ Spec(R) and a subset X of M which lifts a basis
of M/M p over M p, such that M/ Span(X) is not regularly weakly based.
Proof of Claim 1. We choose arbitrary p ∈ Spec(R) and a subset X ′ of T lifting
a basis of T /T p. As T is a pure submodule of M , we can extend X ′ to a subset
X of M containing X ′ such that X lifts a basis of M/M p. Put Y = X \ X ′ and
note that Y T lifts a basis of F/F p over F p. By [4, Lemma 7.1], Y T is a linearly
independent subset of F = M/T , hence Span(Y T ) is free. Set D = M/ Span(X).
We claim that D is not regularly weakly based.
If D is torsion, then M/(T + Span(X)) ≃ F/ Span(Y T ) is torsion too. As F is
an extension of Span(Y T ) by M/(T + Span(X)), the latter module is not bounded
by Lemma 4.3, otherwise F would be projective. Hence, D is also an unbounded
torsion module, and, by Lemma 4.1, D is not regularly weakly based as desired.
Finally, suppose that D is not torsion. In this case, choose any element d ∈ D
with Ann(d) = 0, and put D′ = D/dp. Because dR ≃ R, we have that dp (
dR ⊆ D, and thus there is a submodule of D′ isomorphic to R/p, showing that
the p-primary component of D′ is non-zero. Since D = Dp, also D′ = D′
p. As
7
the p-primary component of D′ is a pure submodule of D′, it is divisible by p,
and therefore divisible. Hence, D′ contains a non-zero divisible direct summand,
and thus D′ is not regularly weakly based by [4, Corollary 3.6]. As dp is a finitely
generated submodule of D, Lemma 3.1 shows that D is not regularly weakly based
as desired. This concludes the proof of the claim.
(cid:3)Claim 1
We pick a generating set Y ′ of M/ Span(X) which does not contain any weak
basis. As M/ Span(X) is divisible by p, we can find a subset Y of pM lifting Y ′
over Span(X). Then X ∪ Y is a generating set, which does not contain any weak
basis of M . Indeed, any subset of X ∪ Y generating M must contain the entire X
and Y ′ does not contain any weak basis of M/ Span(X).
(cid:3)
Theorem 4.5. Let R be a Dedekind domain that is not a division ring. Then
a regularly weakly based R-module splits into a direct sum of a finitely generated
projective module and a bounded torsion module.
Proof. Let M be regularly weakly based module over a Dedekind domain R. Then
M = P ⊕ B, where P is projective and B a bounded torsion R-module, by
Lemma 4.4. Since R is not a division ring, it is not perfect, indeed the only perfect
domains are division rings. Applying Lemma 2.2 and the fact that regularly weakly
based modules are closed under direct summands, we conclude that P is finitely
generated.
(cid:3)
Lemma 4.6. Let R be a Dedekind domain, F a finitely generated module, and B
a bounded p-primary module. Then F ⊕ B is regularly weakly based.
Proof. Apply Lemma 3.1 and Lemma 4.2.
(cid:3)
Corollary 4.7. Let R be a discrete valuation ring and M an R-module. Then M
is regularly weakly based if and only if M ≃ F ⊕ B, where F is finitely generated
free module, and B is bounded torsion module.
Corollary 4.8. Let A an abelian group. If A is regularly weakly based, then A ≃
F ⊕ B, where F is finitely generated free and nB = 0 for some positive integer n.
5. Closing remarks
· · · P nk
The remaining question is whether any bounded torsion module over a Dedekind
domain is regularly weakly based. In other words, we ask whether all R/I-modules
are regularly weakly based for any non-zero ideal I of a Dedekind domain R.
Since a non-zero ideals over Dedekind domains are products of prime ideals, I =
P n1
k , where P1, . . . , Pk are distinct prime ideals and n1, . . . , nk are posi-
1
tive integers. The Jacobson radical of the ring R/I corresponds to the ideal
(P1 · · · Pk)/I and it is clearly nilpotent. Applying Lemma 2.3 we can reduce
the question to the case when I is a product of distinct primes.
In this case
R/I = R/(P1 · · · Pk) ≃ (R/P1) × · · · × (R/Pk) is a product of fields, i.e, it is a
commutative semisimple ring. Thus we arrived to a particular case of the question
discussed at the end of Section 2. Let us formulate it as an open problem:
Problem 5.1. Is every module over a semisimple ring regularly weakly based. In
particular, is every module over a product of division rings (fields) regularly weakly
based?
8
MICHAL HRBEK, PAVEL RŮŽIČKA
The class of regularly weakly based modules is not closed under submodules in
general. A counterexample can be obtained as follows. Let R be a commutative Von
Neumann regular ring with infinitely generated socle S (e.g. an infinite product
of fields). The regular module R, being finitely generated, is regularly weakly
based. We show that the R-module S is not. There is a submodule (and thus, a
direct summand) S′ of S of length ℵ0, say S′ ≃ Ln∈ω Sn, with Sn simple for each
n ∈ ω. As R is regular, Sn has a direct complement Mn in R for each n. Choose a
generator xn of Sn for each n ∈ ω and put yn = x0 + x1 + · · · + xn. We claim that
Y = {yn n ∈ ω} is a generating set of S′ which does not contain any weak basis.
As M0 ∩ M1 ∩ · · · ∩ Mn−1 6⊆ Mn, we conclude that Span(yn) ⊆ Span(ym) for each
n ≤ m, and that Span(xn) ⊆ Span(yn) for each n ∈ ω. Hence, Y generates S′, and
as S′ is not finitely generated, Y contains no weak bases of S′.
Problem 5.2. Is the class of regularly weakly based modules always closed under
quotients?
References
[1] F. W. Anderson, K. R. Fuller, Rings and Categories of Modules. 2nd. ed., Springer-
Verlag (1992)
[2] Pavel Růžička, Abelian groups with a minimal generating set. Quaestiones Math. 33(2) (2010)
: 147 -- 153.
[3] Michal Hrbek and Pavel Růžička, Characterization of abelian groups having a minimal gen-
erating set, Quaestiones Math. 37 (2014), to appear.
[4] Michal Hrbek and Pavel Růžička, Weakly based modules over Dedekind domains, J. Algebra
399 (2014) 251 -- 268.
[5] Irving Kaplansky, Infinite Abelian Groups (1954), Academic Press; First Edition edition
(February 11, 1970).
[6] B. Nashiers and W. Nichols, A note on perfect rings, Manuscripta Mathematica 70, 307 -- 310,
1991
[7] A. A. Tuganbaev and P. A. Krylov. Modules over discrete valuation rings, De Gruyter
expositions in mathematics, 2008.
[8] I. Kaplansky, Modules over Dedekind rings and valuation rings, Matematische Annalen,
Volume 188, Number 4 (1970), 270-284.
[9] A.J. Berrick and M.E. Keating, An introduction to Rings and Modules, Cambridge
University Press, 2000.
[10] D. S. Passman, A course in ring theory, AMS Chelsea Publishing, 2004.
[11] B. Nashier and W. Nichols, On Steinitz properties, Arch. Math., Volume 57, 1991, 247-253.
Charles University, Faculty of Mathematics and Physics, Department of Algebra,
Sokolovská 83, 186 75 Prague, Czech Republic
E-mail address: [email protected], [email protected]
|
1307.7635 | 1 | 1307 | 2013-07-29T16:26:50 | Hopf algebras in non-associative Lie theory | [
"math.RA",
"math.QA"
] | We review the developments in the Lie theory for non-associative products from 2000 to date and describe the current understanding of the subject in view of the recent works, many of which use non-associative Hopf algebras as the main tool. | math.RA | math |
HOPF ALGEBRAS IN NON-ASSOCIATIVE LIE THEORY
J. MOSTOVOY, J.M. PEREZ-IZQUIERDO, I.P. SHESTAKOV
Abstract. We review the developments in the Lie theory for non-associative products from 2000 to date
and describe the current understanding of the subject in view of the recent works, many of which use
non-associative Hopf algebras as the main tool.
1. Introduction: Lie groups, Lie algebras and Hopf algebras
Non-associative Lie theory, that is, Lie theory for non-associative products, appeared as a a subject of
its own in the works of Malcev who constructed the tangent structures corresponding to Moufang loops. Its
general development has been slow; nevertheless, by now many of the basic features of the theory have been
understood, with much of the progress happening in the last ten years or so. In the present paper we outline
the non-associative Lie theory in general and review the recent developments.
The history of the subject before 2000 has been summarized in the paper of Sabinin [49]. Probably, the
single most important advance to that date has been the definition of a Sabinin algebra by Mikheev and
Sabinin in 1986 (they use the term hyperalgebra). Sabinin algebras are the structures tangent to general
non-associative unital local products; under some convergency conditions one can recover the product from a
Sabinin algebra. Lie, Malcev, Bol, Lie-Yamaguti algebras and Lie triple systems are all particular instances
of Sabinin algebras.
Classical Lie theory relates three kinds of objects: Lie groups, Lie algebras and cocommutative Hopf
algebras. Until recently, however, Hopf algebras had little, if any, role in the non-associative Lie theory. To
a certain extent, this gap has now been filled, and here we shall give an outline of the theory of local loops
and Sabinin algebras from the point of view of non-associative Hopf algebras. We shall be guided by the
analogy with the theory of Lie groups and Lie algebras, so first we recall the basics of the classical theory.
At the heart of Lie theory lies the equivalence of the following three categories:
• simply-connected finite-dimensional Lie groups;
• finite-dimensional Lie algebras;
• irreducible cocommutative finitely generated Hopf algebras.
Let us describe explicitly the functors that establish this equivalence. In what follows all the vector spaces,
Lie algebras etc will be defined over a field of characteristic zero.
1.1. Lie groups → Lie algebras. There are several ways to produce the Lie algebra of a Lie group. Let
G be a Lie group with the unit element e. The most common construction identifies the tangent space TeG
with the vector space of left-invariant vector fields on G. The commutator of two such fields (considered as
derivations acting on functions) is again left-invariant, and defines the Lie bracket on TeG.
This definition can be stated in different terms. On the tangent bundle T G there exists a canonical flat
connection ∇. It can be defined by the associated parallel transport: for any path between a, b ∈ G it sends
TaG to TbG by the differential of the left multiplication by ba−1. The curvature tensor of ∇ is identically
zero and the torsion tensor T is covariantly constant, that is ∇T = 0 on G. This implies that the torsion
tensor T is defined completely by its value on TeG, which is a linear map
This map coincides, up to sign, with the Lie bracket on TeG.
1.2. Lie algebras → Lie groups. One way to obtain a Lie group from a Lie algebra is via the Baker-
Campbell-Hausdorff formula which expresses the formal power series
TeG ⊗ TeG → TeG.
(1)
log(exp x exp y)
1
in terms of iterated commutators in x and y. For finite-dimensional Lie algebras, the Baker-Campbell-
Hausdorff series always has a non-zero radius of convergence and, hence, defines a local Lie group.
It can be shown that finite-dimensional local Lie groups can always be extended to global Lie groups,
but the proof of this fact is somewhat mysterious and uses the facts that seem to come from outside of
Lie theory, such as the Ado theorem. The Ado theorem claims that every finite-dimensional Lie algebra is
isomorphic to a Lie subalgebra of the algebra of the matrices. Therefore, it integrates to a local Lie group
which is embedded as a local subgroup into the group of invertible matrices. Then it can be shown that
this local subgroup generates a subset in the group of matrices which is an image of a smooth injective Lie
group homomorphism. This last step is far from being trivial and uses in a crucial way the associativity of
the matrix multiplication.
1.3. Lie algebras → Hopf algebras. Each Lie algebra g can be embedded into an associative algebra A
in such a way that the bracket in g is induced by the commutator in A. Among all such embedding one is
universal. The corresponding algebra (called universal enveloping algebra) is constructed as the quotient of
the unital tensor algebra on g by the relations
for all x, y ∈ g and is denoted by U (g). The Lie algebra g is naturally embedded into the tensor algebra on
g and this embedding descends to the embedding
x ⊗ y − y ⊗ x − [x, y]
into the universal enveloping algebra.
The algebra U (g) can be given a coproduct
g → U (g)
δ : U (g) → U (g) ⊗ U (g)
δ(x) = x ⊗ 1 + 1 ⊗ x.
which turns it into a Hopf algebra. It is defined by declaring all the elements of g ⊂ U (g) to be primitive:
Since the coproduct in a Hopf algebra is an algebra homomorphism, this is sufficient to determine δ com-
pletely. According to the Poincar´e-Birkhoff-Witt theorem, as a coalgebra U (g) is isomorphic to the symmetric
algebra k[g]. This latter can be thought of as the universal enveloping algebra of the abelian Lie algebra
which coincides with g as a vector space and whose Lie bracket is identically zero.
It is immediate that as a coalgebra U (g) is cocommutative. Also, by construction, it is generated by its
primitive elements.
1.4. Hopf algebras → Lie algebras. A straightforward calculation shows that the subspace of primitive
elements in a Hopf algebra is always closed under the commutator. In particular, there is a functor that
assigns to each Hopf algebra the Lie algebra of its primitive elements. It is not hard to prove that it gives
an equivalence of the category of cocommutative primitively generated Hopf algebras to that of Lie algebras
(of not necessarily finite dimension).
1.5. Lie groups → Hopf algebras. Let DeG be the space of distributions (that is, linear functionals on
functions) on the Lie group G, supported at the unit e. In other words, DeG is the space spanned by the
delta function supported at e and all of its derivatives. The multiplication on G gives rise to a convolution
product on DeG; the diagonal map G → G × G induces a coproduct on DeG. It can be seen that DeG is a
Hopf algebra; moreover, it is precisely the universal enveloping algebra of the Lie algebra of G.
1.6. Hopf algebras → Lie groups. To recover a Lie group from a Hopf algebra, one uses the fact that any
cocommutative Hopf algebra generated by its primitive elements is a universal enveloping algebra of some
Lie algebra. In particular, by the Poincar´e-Birkhoff-Witt theorem, it can be thought of as the symmetric
algebra k[V ] on some vector space V over the base field k with a modified product
Consider the composition of this product with the projection k[V ] → V onto the space of primitive elements.
The resulting map
µ : k[V ] ⊗ k[V ] → k[V ].
Fµ : k[V ] ⊗ k[V ] → V
2
can be interpreted as follows. Choose a basis e1, . . . , en in V . Then k[V ] is the algebra of the polynomials in
the ei. Each component of the map Fµ is a formal power series in xi and yj, where x1, . . . , xn and y1, . . . , yn
are the coordinates dual to the ei in the first and the second copy of V respectively. It turns out that for any
Hopf algebra such that V is of finite dimension these formal power series all converge in a neighbourhood U
of the origin in V × V and thus give an analytic map
V × V ⊇ U → V.
This map is, in fact, a finite-dimensional local Lie group and these are always locally equivalent to simply-
connected Lie groups.
1.7. The Ado theorem. The fundamental triangle of Lie theory is remarkably robust: it can be generalized
to a variety of situations. For instance, finite-dimensional Lie algebras can be replaced by infinite-dimensional
Lie algebras, Lie algebras in tensor categories other than vector spaces, or, as we shall see, can be substituted
for more general kinds of tangent algebras. There is only one construction that does not always survive into
the more general context.
Note that while recovering a Lie group either from its Lie algebra or from its Hopf algebra of distributions
we first construct a formal Lie group, that is, a collection of power series which a priori may not converge.
In order to see that this formal Lie group comes from an actual Lie group we need the Ado theorem, which
states that every finite-dimensional Lie algebra has a faithful finite-dimensional representation. In particular,
it follows that any finite-dimensional formal Lie group can be obtained from a local subgroup in a matrix
group. A further argument is then needed to see that every such local subgroup extends to an injective
homomorphism of a Lie group to the group of invertible matrices.
This proof breaks down already in the case of infinite-dimensional Lie groups. It also fails in the case of
general finite-dimensional loops. Nevertheless, there are situations (Moufang loops, nilpotent loops) where
each formal multiplication extends to a global loop. In these situations we have an analogue of the Ado
theorem, though we should point out that this theorem alone does not guarantee the globalizability of formal
products.
1.8. Nilpotent groups. Lie algebras and Hopf algebras also appear in the theory of discrete nilpotent
groups.
Let G be a discrete group. Write QG for the group algebra of G over the rationals and let ∆ ⊂ QG be the
augmentation ideal. The ith dimension subgroup DiG (or Di(G, Q)) consists of all those g ∈ G for which
g − 1 ∈ ∆i.
The dimension subgroups form a descending central series which is closely related to the lower central series
γiG defined inductively by γ1G = G and γiG = [γi−1G, G] for i > 1. Namely, DiG consists of all g ∈ G for
which there exists n ≥ 1 with the property that gn ∈ γiG. This latter statement was proved by Jennings
[22].
The graded abelian group
LG =Mi>0
DiG/Di+1G
associated with the dimension series is actually a Lie ring with the Lie bracket induced by the group com-
mutator
Tensored with the rational numbers, this Lie ring becomes a Lie algebra and its universal enveloping algebra
can be explicitly identified as the algebra
[a, b] = a−1b−1ab.
associated with the filtration on QG by the powers of the augmentation ideal.
∆i/∆i+1
Mi≥0
In general, a discrete group G cannot be recovered from the Lie algebra LG⊗Q coming from the dimension
series (equivalently, from the lower central series). However, if G is a finitely generated nilpotent group,
LG ⊗ Q is finite-dimensional and can be integrated to a Lie group into which G embeds as a discrete
subgroup if it is torsion-free.
3
One group for which the dimension series and the lower central series can be described explicitly is the
free group Fn on n generators x1, . . . , xn. Let ZhhX1, . . . , Xnii be the ring of formal power series in n non-
commuting variables. The group of invertible elements in this ring consists of power series which start with
±1. We denote this group by ZhhX1, . . . , Xnii∗. There is a homomorphism
(2)
M : Fn → ZhhX1, . . . , Xnii∗,
xi 7→ 1 + Xi,
which is, in fact, injective. The ith terms of the dimension and of the lower central series of Fn coincide and
consist of those elements which map under M to the power series of the form
1 + terms of degree at least m.
This, together with the injectivity of M, shows that the free group Fn is residually nilpotent.
The non-associative versions of the above constructions and statements are the subject of the present
paper. There are fundamental (and vast) parts of Lie theory which have not been understood yet in the
non-associative context, most notably the representation theory. By no means we want to suggest that there
are inherent obstructions to this, apart form time and effort.
2. Local, infinitesimal and formal loops
The non-associative Lie theory deals with the following three equivalent categories:
• formal loops;
• Sabinin algebras;
• irreducible cocommutative and coassociative non-associative Hopf algebras.
In contrast to the usual finite-dimensional Lie theory, there is no procedure to extend formal loops to global
multiplications on manifolds. This, however, should not be seen as a problem since many non-associative
products are inherently of local nature. The true analogue of a Lie group in the non-associative Lie theory is
not a smooth loop, but, rather, a local loop, or a germ of a local loop. (We shall use the term "infinitesimal
loops" for germs of local loops.) A formal loop produces a local loop whenever the power series that define
it converge in some neighbourhood of the origin; verifying this is a problem from analysis and we shall not
discuss it here.
2.1. Local loops and smooth loops. Let M be a smooth finite-dimensional manifold1. A local multipli-
cation on M (or, more precisely, on U ⊆ M ) is a smooth map
F : U × U → M,
where U ⊆ M is a non-empty open subset. If there exists e ∈ U with the property that
Fe×U = Id(U ) = FU×e,
the local multiplication F is called unital, or a local loop. The point e is referred to as the unit; if the unit
exists, it is necessarily unique. Often, when working with a fixed local loop F , we shall write F (x, y) simply
as x · y or xy.
y\x. They are defined by
For any local loop there exist two local multiplications V × V → M with V ⊆ U , denoted by x/y and
x/y · y = x
and
for all x, y ∈ V , and, for obvious reasons, are called the right and the left divison, respectively. The existence
of both divisions follows from the fact that the right and left multiplication maps
y · y\x = x
and
Ry = FU×y : U → M
Ly = Fy×U : U → M
1we assume that dim M > 0 is part of the definition of a smooth manifold.
4
are close to the inclusion map U ֒→ M when y is close to e. In particular, if y is sufficiently close to e, both
maps Ry and Ly are one-to-one and their images contain a neighbourhood of e. We take V to be the largest
neighbourhood on which both divisions are defined.
In general, there is no reason to expect that the right and the left divisions would be expressed as
multiplications by an inverse of any kind.
A local loop is called a smooth loop if U = V = M in the above definitions. Recall that a discrete loop,
or, simply, a loop, is a set M with a globally defined product for which there exists a unit element, and such
that the left and the right multiplications by a fixed element of M are bijections. In particular, for each
loop there are globally defined right and left divisions. In these terms, a smooth loop can be defined as a
loop which, at the same time, is a smooth manifold, with the multiplication and the divisions being smooth
maps.
2.2. Example: invertible elements in algebras. Call an element a of a unital algebra invertible if both
equations ax = 1 and xa = 1 have a unique solution. The invertible elements of a finite-dimensional algebra
over the real numbers form a local loop.
This local loop is not necessarily a loop. Consider, for instance, the Cayley-Dickson algebras An on R2n
.
When n > 3 there exist pairs of invertible elements in An whose product is not invertible.
2.3. Example: homogeneous spaces. Recall that a smooth manifold M is a homogeneous space for a
Lie group G if G acts transitively on M and the action is smooth. Choose a point e ∈ M . Then we have a
smooth map
which sends g ∈ G to g(e) ∈ M . This map identifies M , as a smooth manifold, with the left coset space
G/Ge, where Ge is the stabilizer of e. Conversely, if G is a Lie group and H is a closed subgroup, the set
G/H of left cosets of G by H is a smooth manifold and a homogenous space for G.
p : G → M,
Let M be a homogeneous space for G and U ⊆ M a neighbourhood of a point e ∈ M . Assume that
we are given a section of p over U , that is, a smooth map i : U → G such that i(e) is the unit in G and
p ◦ i = Id(U ). Then M is a local loop, with the multiplication U × U → M defined as
(x, y) 7→ p (i(x)i(y)) .
There are many important examples of homogeneous spaces, among them spheres, hyperbolic spaces and
When p is actually a homomorphism of Lie groups, that is, when Ge is a normal subgroup in G, this local
loop structure is the same thing as the product on M restricted to U × U .
Grassmannians.
2.4. Infinitesimal loops. Let F and F ′ be two local loops on U ⊆ M and U ′ ⊆ M ′ respectively. A
morphism F → F ′ is a map f : M → M ′ such that f (U ) ⊆ U ′ and such that
F ′(f (x), f (y)) = f (F (x, y))
for all x, y ∈ U . With this notion of a morphism local loops form a category.
If a morphism f from F to F ′ is an open embedding, we shall say that F is a restriction of F ′. In this
situation, the local loops F and F ′ may be thought of as being "locally the same" near the unit. Certainly,
they should not be distinguished by any infinitesimal algebraic structure at the unit, such as those arising
in the classical Lie theory (think of formal groups or Lie algebras). With this in mind, we define local
equivalence as the smallest equivalence relation on the category of local loops under which open embeddings
are equivalences. An equivalence class of local loops is called an infinitesimal loop.
Strictly speaking, it is the infinitesimal loops and not the local loops that are the main subject in the
non-associative Lie theory. However, for the sake of simplicity, we shall speak of local loops rather than
infinitesimal loops, wherever possible. This should not lead to confusion.
2.5. Analytic local loops and formal loops. Examples of local loops are very easy to construct via power
series, since the only condition a multiplication of a local loop has to satisfy is the existence of the unit.
Consider an n-tuple of power series Fi(xj, yk) where 1 ≤ i, j, k ≤ n, and assume that all of them converge in
some neighbourhood of the origin in R2n. The map
(x1, . . . , xn, y1, . . . , yn) 7→ (F1(xj, yk), . . . , Fn(xj , yk))
5
defines a local loop on Rn, with the origin as the unit, if and only if
(3)
and
(4)
Fi(0, . . . , 0, y1, . . . , yn) = yi
Fi(x1, . . . , xn, 0, . . . , 0) = xi
for all i. These conditions mean that the coefficient of xj and yj in Fi is equal to one if i = j and vanishes
otherwise. Moreover, a monomial of degree two or more can have a non-zero coefficient in Fi only if it
contains both the xj and yj ′ for some j, j′. The only further restrictions on the coefficients of the Fi come
from the requirement that the power series Fi converge.
A local loop on an analytic manifold whose multiplication can be written in this form in some coordinate
chart is called analytic. Similarly, an infinitesimal loop is analytic if it has an analytic representative.
The convergence of the power series Fi that specify an analytic loop is often irrelevant. For instance,
a power series with no constant term can be substituted into another power series instead of a variable,
and thus we do not need convergence in order to speak of the algebraic identities, such as associativity or
Moufang identites, satisfied by an analytic loop. In what follows we shall refer to an n-tuple of formal power
series Fi satisfying (3) and (4) as a formal loop.
Let us state this definition in the form that does not make reference to any explicit basis in Rn, nor to
the fact that n is finite. For a vector space V and a field k write k[V ] for the symmetric algebra of V over
k. If a basis eα is chosen for V , this is the commutative algebra of polynomials in the eα. Monomials of the
form
(5)
α1 . . . enk
en1
αk
with α1 < . . . < αk form a basis of k[V ]. If xα is the coordinate corresponding to eα, then any formal power
series F in the xα can be thought of as a linear function on k[V ]. This function sends the monomial (5) to
the coefficient of xn1
αk in the series F .
α1 . . . xnk
Now, an analytic map from V to a k-dimensional vector space W is given by a k-tuple of power series in
the xα, where k = dim W . Given that each power series is just a linear function on k[V ], we can define a
formal map from a vector space V to a vector space W as a linear map k[W ] → V which annihilates the
constants.
A formal multiplication is a formal map from V × V to V . Given that k[V × V ] is canonically isomorphic
to k[V ] ⊗ k[V ], this is the same a linear map
A formal multiplication F is a formal loop if
k[V ] ⊗ k[V ] → V.
F1⊗k[V ] = πV = Fk[V ]⊗1,
where πV : k[V ] → V be the projection of a polynomial onto its linear part.
2.6. The canonical connection and geodesic loops. Let F be a local loop on U ⊆ M and let V ⊆ U
be a neighbourhood on which the left division is well-defined. The canonical connection ∇ on the tangent
bundle T V has virtually the same definition as for the Lie groups (apart from the fact that it is not defined
on the whole of M ): for a, b ∈ G the corresponding parallel transport sends TaG to TbG by the differential
of the smooth map
It is clear that ∇ is flat. There is no reason, however, for the torsion tensor of ∇ to be constant.
It can be seen that each flat connection on a neighbourhood of a point comes from a local loop, called the
geodesic loop of the connection. Indeed, a connection ∇ on a manifold M gives rise to the exponential map
x 7→ b(a\x).
expa : TaM ⊇ Ua → M,
defined on a neighbourhood of the origin in TaM , for each a ∈ M . If Ua is chosen to be small enough, this
map has an inverse, denoted by loga. Fix a point e ∈ M . Then there is a neighbourhood U ⊆ M of e such
that
(6)
a · b = expa (a loge b),
6
is well defined for all a, b ∈ U . Here a loge b stands for the parallel transport of the vector loge b ∈ TeM
to TaM . The operation · defines a local loop on U ⊆ M and it is a straightforward check that ∇ is its
canonical connection. This loop is called the geodesic loop of ∇. In order to be precise, we should consider
the geodesic loop as an infinitesimal rather than local loop since there is a choice involved in the construction
of the neighbourhood U .
It can be shown that the necessary condition for an infinitesimal loop (which is also sufficient in the case
of analytic loops) to be a geodesic loop of a flat connection is that it should satisfy the right alternative
property. Namely, it should be represented by a local loop such that
whenever both sides are defined. In particular, each local loop F gives rise to a right alternative infinitesimal
loop, which is the geodesic loop of the canonical connection of F . One can recover a local loop F from the
corresponding geodesic loop with the help of the additional operation Φ defined by
(a · b) · b = a · (b · b)
(7)
Here · denotes the multiplication in F and × is the product in the geodesic loop. Φ here can be any function
of a and b that satisfies Φ(e, b) = b and Φ(a, e) = e.
The construction of the geodesic loop associated with a flat connection is due to Kikkawa [24] and Sabinin
a · b = a × Φ(a, b).
[47].
There is also a corresponding formal notion of a canonical connection. Given a formal loop F on a vector
space V , the formal canonical connection of F is the restriction of F to the subspace
Among all the formal loops that give rise to the same formal connection there is exactly one right alternative
formal loop.
k[V ] ⊗ V ⊂ k[V ] ⊗ k[V ].
The notion of a geodesic loop is of crucial importance in non-associative Lie theory. Indeed, there are two
constructions of the tangent structure to a local (infinitesimal, formal) loop and both are based on the fact
that there is a flat connection associated with a local loop, see the next section. In fact, in the approach
to the Lie theory taken by Sabinin in [50] the role of non-associative analogues of Lie groups is played not
by loops but by affinely connected manifolds. This involves additional algebraic structures such as odules;
at the moment of the writing of the present paper the theory of odules has not noticeably advanced beyond
the results of [50].
3. Sabinin algebras
Many of the well-known generalizations of Lie algebras involve only one or two operations: Malcev algebras
have one binary bracket, Lie triple systems one ternary bracket, Bol and Lie-Yamaguti algebras one binary
and one ternary bracket.
In contrast, Sabinin algebras which are the most general form of the tangent
structure for loops, have an infinite set of independent operations, and, as a consequence, admit an infinite
number of axiomatic definitions. At the moment there are three different natural constructions for the set
of operations in a Sabinin algebra. Two of these constructions are due to Sabinin and Mikheev and we shall
review them in this section. The third set of operations, which appeared for the first time in [55] in the
study of the primitive elements in non-associative bialgebras, will be considered in the next section. The
complete set of axioms for this third set of operations is presently not known.
Remark. The fact that the definition of Mikheev and Sabinin involved an infinite number of operations
may have contributed to its slow acceptance. For some time Akivis algebras were considered as possible
analogues of Lie algebras. The definition of an Akivis algebra involves two operations: an antisymmetric
binary bracket and a ternary bracket, with only one identity that relates the two operations and generalizes
the Jacobi identity. In spite of the elegance and simplicity of this definition, the category of Akivis algebras
is not equivalent to that of formal loops and, hence, is not suitable as a basis for non-associative Lie theory.
We shall consider the Akivis algebras in more detail in Section 4.7.
Both ways of deriving the structure of a Sabinin algebra from a non-associative product that were proposed
by Mikheev and Sabinin consist of two steps. First, a local loop is replaced by the corresponding geodesic
loop (which, in the analytic case is the same as a right alternative local loop), and then the tangent operations
7
for a geodesic loop are extracted from its canonical connection. This results in two infinite sets of completely
independent operations. Mikheev and Sabinin used the term "hyperalgebra" for the algebraic structure
tangent to a geodesic loop and "hyperalgebra with multioperators" for the general tangent structure. In our
terminology hyperalgebras with multioperators will be called Sabinin algebras, and hyperalgebras (which are
the same as hyperalgebras with the trivial multioperator) will be called flat Sabinin algebras. The adjective
"flat" here is supposed to reflect the fact that these are the tangent structures to general flat connections.
3.1. Flat Sabinin algebras from the torsion tensor of a connection. Let k be a field of characteristic
zero. A flat Sabinin algebra is a vector space V over k together with a set of maps
V ⊗n+2 → V
for all integers n ≥ 0, satisfying the following identities:
X1 ⊗ . . . ⊗ Xn ⊗ Y ⊗ Z 7→ hX1, . . . , Xn; Y, Zi
hX1, . . . , Xn; Y, Zi = −hX1, . . . , Xn; Z, Y i,
hX1, . . . , Xr, A, B, Xr+1, . . . , Xn; Y, Zi − hX1, . . . , Xr, B, A, Xr+1, . . . , Xn; Y, Zi
r
hXα1, . . . , Xαk ,hXαk+1 , . . . , Xαr ; A, Bi, Xr+1, . . . , Xn; Y, Zi = 0,
+
Xk=0Xα
σX,Y,Z hX1, . . . , Xr, X; Y, Zi +
r
Xk=0Xα
hXα1 , . . . , Xαk ;hXαk+1 , . . . , Xαr ; Y, Zi, Xi! = 0.
Here α varies over the set of all bijections {1, . . . r} → {1, . . . r}, i → αi such that α1 < α2 < . . . < αk,
αk+1 < . . . < αr, k = 0, 1, . . . , r, r ≥ 0, and σX,Y,Z denotes the sum over all cyclic permutations of X, Y, Z.
The above definition may seem complex, but it has the following geometric meaning.
Consider an affine connection ∇ at a point e of a manifold M . It is characterized by two tensors: the
curvature tensor and the torsion tensor which are related by the Bianchi identities. If ∇ is flat (as it is in
the case of a canonical connection) the curvature tensor is identically zero. If we assume that ∇ is analytic,
the torsion tensor T is uniquely determined in a neighbourhood of e by its value and the values of all its
covariant derivatives at this point. These derivatives are multilinear operations on TeM . Set
hY, Zi = T(Y, Z)
and
hX1, . . . , Xn; Y, Zi = ∇X1 . . .∇Xn T(Y, Z)
for all n > 0. Then the brackets defined in this way are antisymmetric in Y and Z, since the torsion tensor
is antisymmetric, and satisfy the identities which come from the Bianchi identities and their derivatives, and
from the fact that the curvature is zero. These identities are precisely those that appear in the definition of
a flat Sabinin algebra.
From this construction it is clear that given a flat Sabinin algebra l one can reconstruct, locally, the
corresponding flat connection, and, hence, the infinitesimal loop, provided that the operations of l define
converging power series. The convergence conditions are explicitly stated by Mikheev and Sabinin in [33, 48].
In general, even if the convergence conditions are not satisfied, the geometric reasoning can be applied so as
to produce a formal loop from any flat Sabinin algebra.
3.2. Flat Sabinin algebras from pairs of Lie algebras. An alternative approach to flat Sabinin algebras
is to define them as the algebraic structure that exists on a complement to a subalgebra in a Lie algebra.
Start with a direct sum decomposition of vector spaces
g = h ⊕ l,
(8)
where g is a Lie algebra and h is a subalgebra. Let πl be the linear projection map πl : g → l with h as the
kernel. It will be convenient to introduce a simplified notation for the right-normed iterated bracket in g:
[X1, X2, . . . , Xn] = [X1, [X2, [. . . , Xn] . . .]].
8
Define a sequence of multilinear operations2 on l by setting for each n ≥ 2 and X1, . . . , Xn ∈ l
(X1, . . . , Xn) := πl[X1, . . . , Xn].
There are two kinds of relations satisfied by these operations. The relations of the first kind reflect the fact
that they come from Lie brackets. Namely, the anti-symmetry of the Lie bracket gives
(9)
(X1, X2) + (X2, X1) = 0,
and the Jacobi identity translates into the following identity:
(10)
(X1, X2, X3) + (X2, X3, X1) + (X3, X1, X2) = 0.
The identities of the second kind express the fact that h is a subalgebra and not just a vector subspace. If
A, B ∈ g are arbitrary elements, then
Setting A = [X1, . . . , Xn] and B = [Y1, . . . , Ym] we get
πl[[X1, . . . , Xn], [Y1, . . . , Ym]] + ((X1, . . . , Xn), (Y1, . . . , Ym))
πl[πlA − A, πlB − B] = 0.
Now, the first term in this expression can be rewritten iteratively using the Jacobi identity in terms of the
right-normed brackets so that the last relation takes the form
= ((X1, . . . , Xn), Y1, . . . , Ym) − ((Y1, . . . , Ym), X1, . . . , Xn).
(11) − ((X1, . . . , Xn), (Y1, . . . , Ym)) + ((X1, . . . , Xn), Y1, . . . , Ym) − ((Y1, . . . , Ym), X1, . . . , Xn)
(−1)h(α)(Xα1 , . . . Xαn , Y1, . . . , Ym),
=Xα
where the summation is taken over all permutations α of the set {1, . . . , n} for which there exists s such
that αi < αi+1 for i < s, and αi > αi+1 for i ≥ s, and h(α) = n − s. (Clearly, for any such α we have that
αs = n.)
We can now define a flat Sabinin algebra as a vector space with a sequence of multilinear operations
(X1, . . . , Xn) for all n ≥ 2 satisfying the relations (9)-(11). It can be shown that if l is a flat Sabinin algebra
in this sense, there exists a decomposition (8) which gives rise to the operations on l, [39].
The relation between flat Sabinin algebras in this sense and germs of flat affine connections is quite
straightforward. Given such a connection ∇ in the neighbourhood of a point e on a manifold M , we can
identify the tangent space TeM with the space of all ∇-parallel vector fields. Set g to be the Lie subalgebra
generated by the ∇-parallel vector fields inside the Lie algebra of all vector fields, and h to be the subalgebra
of g consisting of the vector fields that vanish at e. Then we have the direct sum decomposition
g = h ⊕ TeM,
and TeM acquires the structure of a flat Sabinin algebra.
One can, in fact, write down explicit formulae relating the two sets of operations in Sabinin algebras, see
[33, 48].
3.3. Multioperators. Let F be a right alternative local loop F on U ⊆ M and
Φ : U × U → M
a function such that Φ(e, b) = b, Φ(a, e) = e and such that the Jacobian matrix of Φ(a, b) with respect to b
is the identity when b = e. Define a new local loop FΦ by setting
FΦ(a, b) = F (a, Φ(a, b)).
The canonical connection of FΦ coincides with that of F ; in particular, F is the geodesic loop associated
with FΦ. (Strictly speaking, FΦ is defined on a smaller neighbourhood than U . As usual, one can replace
local loops here by infinitesimal loops.)
2Our notation may be potentially confusing where it concerns the trilinear operation (·, ·, ·). Here this is not the associator.
9
If Φ is analytic, one can write
(12)
Φ(a, b) = b + Xm≥1,n≥2
Φm,n(a, . . . , a; b, . . . , b),
where Φm,n(X1, . . . , Xm; Y1, . . . , Yn) is linear in each Xj and Yj and invariant with respect to all the permu-
tations of the Xi and of the Yj:
Φm,n(X1, . . . , Xm; Y1, . . . , Yn) = Φm,n(Xσ(1), . . . , Xσ(m); Yτ (1), . . . , Yτ (n))
(13)
for all σ ∈ Σm, τ ∈ Σn.
of operations
This motivates the following definition. A Sabinin algebra is a flat Sabinin algebra l together with a set
for m ≥ 1, n ≥ 2, satisfying the symmetry conditions (13). Given all that has been said about the flat
Sabinin algebras, it is not hard to show that Sabinin algebras form a category which is equivalent to the
category of all formal loops.
Φm,n : l⊗m+n → l
3.4. Free Sabinin algebras. The free Sabinin algebra on a set of generators S is the universal Sabinin
algebra generated by S. It can be constructed as the vector space spanned by symbols corresponding to
all possible operations h·, . . . ,· ;·,·i and Φm,n(·, . . . ,· ; ·, . . . ,·) and their compositions, whose arguments are
elements of S, with the relations of a Sabinin algebra imposed. Another way to construct the free Sabinin
algebras is by using the techniques described in the next section. Namely, the free Sabinin algebra on a
set of generators S is the set of primitive elements in the free non-associative algebra on the same set of
generators.
Several types of bases in free Lie algebras are known: Hall bases, Lyndon-Shirshov bases [46], right normed
bases of Chibrikov [11]. The construction of Lyndon-Shirshov bases was extended by Chibrikov [12] to free
Sabinin algebras. The dimensions of the graded summands of a free Sabinin algebra on n generators were
calculated in [6].
Free Sabinin algebras enjoy many other properties characteristic of free Lie algebras [12]. In particular, the
variety of Sabinin algebras is Schreier (every subalgebra of a free Sabinin algebra is free). This implies that
all automorphisms of finitely generated free Sabinin algebras are tame, and that the occurrence problem for
free Sabinin algebras is decidable. Also, finitely generated subalgebras of free Sabinin algebras are residually
finite and the word problem is decidable for the variety of Sabinin algebras.
3.5. Further remarks. Just as in the associative case, there exists a Baker-Campbell-Hausdorff formula
for Sabinin algebras. We shall discuss it in the next section.
The two sets of multilinear operations in a Sabinin algebra are completely independent and can be
considered as Taylor expansions of two non-linear operations. Sabinin in [50] takes this approach pointing
out that it can be applied to non-analytic loops, and branding the constructions of [33] and [48] as obsolete.
The book [50] was written before it was discovered that Sabinin algebras are important in the theory of
non-associative bialgebras; in this latter context non-linear operations are not easy to deal with.
4. Non-associative Hopf algebras
It would be fair to say that most of the constructions and results on Sabinin algebras mentioned in
the previous section were the product of a quest for a Lie theory of flat connections. The subject of non-
associative Hopf algebras has been developed in the context of the long-standing problem of finding an
alternative enveloping algebra for an arbitrary Malcev algebra. While the existence of such an alternative
envelope is still an open question, it turned out [41] that Malcev algebras have universal enveloping algebras
which are very similar in their properties to usual cocommutative Hopf algebras. Later, it turned out that
a similar construction can be carried out for Bol algebras [43] and, more generally, all Sabinin algebras [44].
Another motivation for the development of the machinery of non-associative Hopf algebras was the ques-
tion of whether the commutator and the associator are the only primitive operations in a non-associative
bialgebra.
It appeared as a conjecture in [20]; if it were true, it would imply an important role for the
Akivis algebras in non-associative Lie theory. This conjecture was answered in the negative in [55]:
in
non-associative bialgebras, apart from the commutator and the associator, there exists an infinite series
10
of independent primitive operations. These operations, called the Shestakov-Umirbaev operations, are ob-
tained from the associator by means of a procedure resembling linearization; they are closely related to the
associator deviations in the nilpotency theory of loops (see Section 8.1).
The role of the non-associative Hopf algebras in the fundamental questions of Lie theory such as integration
was clarified in [38]. Hopf algebraic techniques are also relevant in other problems, such as the Ado theorem,
which will be discussed in Section 7.4.
4.1. The definition and basic properties. In what follows we shall consider algebras and coalgebras over
a field k of characteristic zero.
A unital bialgebra is a unital algebra which, at the same time, is a counital coalgebra in such a way that
the coproduct is an algebra homomorphism (or, equivalently, such that the product is a coalgebra morphism).
It will be convenient to use Sweedler's notation in which the coproduct is written as
A unital bialgebra H with the coproduct δ and the counit ǫ is called a (non-associative) Hopf algebra if is
endowed with two additional bilinear operations, the right and the left division
δ(x) =X x(1) ⊗ x(2).
such that
and
/ : H × H → H,
\ : H × H → H,
(x, y) 7→ x/y,
(x, y) 7→ x\y,
X(yx(1))/x(2) = ǫ(x)y =X(y/x(1))x(2)
X x(1)\(x(2)y) = ǫ(x)y =X x(1)(x(2)\y).
When the product in a Hopf algebra is associative and the coproduct is coassociative we get the usual notion
of a Hopf algebra. In this case the antipode S can be defined as S(x) = 1/x and, generally, we have
x\y = S(x)y
and x/y = xS(y)
for all x, y.
The subspace of a Hopf algebra generated by the unit is a simple subcoalgebra. If this subspace is the
only simple subcoalgebra, the Hopf algebra is called irreducible. In irreducible Hopf algebras the right and
left division are uniquely determined.
While non-associativity is an essential property of Hopf algebras that appear in Lie theory, the coproduct
in many interesting cases is, actually, coassociative. For instance, the loop algebra kL of a loop L is always
a coassociative Hopf algebra. In what follows by a Hopf algebra we shall understand a cocommutative and
coassociative Hopf algebra, unless explicitly stated otherwise.
4.2. Primitive elements and Shestakov-Umirbaev operations. In any associative bialgebra the set of
primitive elements is closed under the algebra commutator. This can be expressed by saying that the com-
mutator is a primitive operation. Any other primitive operation can be written in terms of the commutators,
in the following sense: if an associative bialgebra H is generated by a set {xi} where each xi is primitive,
then each primitive element in H is a linear combination of iterated commutators in the xi. We express this
by saying that the commutator forms a complete set of primitive operations.
In non-associative bialgebras there are other primitive operations, most notably the algebra associator
(a, b, c) = (ab)c − a(bc).
As discovered by Shestakov and Umirbaev in [55], the set of primitive operations obtained from commutators
and associators is not complete. For instance, in the free non-associative algebra on one primitive generator
x the element
(x2, x, x) − 2x(x, x, x) = (x2x)x − x2x2 − 2x(x2x) + 2x(xx2)
is primitive, yet cannot be obtained from x by taking linear combinations and compositions of commutators
and associators.
11
For m, n ≥ 1 let u = x1, . . . , xm and v = y1, . . . , yn be sequences of primitive elements in a bialgebra
and write u = ((x1x2)··· )xm and v = ((y1y2)··· )yn for the corresponding left-normed products. The
Shestakov-Umirbaev operations p(x1, . . . , xm; y1, . . . , yn; z) are defined inductively by
(u, v, z) =X u(1)v(1) · p(u(2); v(2); z),
where (x, y, z) denotes the associator and z is primitive. Here Sweedler's notation is extended so as to mean
that the sum is taken over all decompositions of the sequences u and v into pairs of subsequences u(1), u(2)
and v(1), v(2); the expressions u(1) and v(1) are the left-normed products of the elements of u(1) and v(1),
respectively.
For instance, the operation which corresponds to m = n = 1 is just the associator. The operations
corresponding to m = 2, n = 1 and m = 1, n = 2 are
and
respectively.
p(x1, x2; y; z) = (x1x2, y, z) − x1(x2, y, z) − x2(x1, y, z)
p(x; y1, y2; z) = (x, y1y2, z) − y1(x, y2, z) − y2(x, y1, z),
When the bialgebra in question is a Hopf algebra, the Shestakov-Umirbaev operations can be written with
the help of the left division:
This form of the definition may be preferable since it expresses each operation directly via the associator.
p(x1, . . . , xm; y1, . . . , yn; z) =X(u(1)v(1))\(u(2), v(2), z).
In [55] it is shown that all Shestakov-Umirbaev operations are primitive, and that together with the
commutator they form a complete set of primitive operations.
In associative algebras the commutator satisfies the Jacobi identity; in particular, the primitive elements
of a bialgebra A form a Lie subalgebra of the commutator algebra of A. In the non-associative case, the Lie
algebras are replaced by Sabinin algebras. Indeed, set
h1; y, zi = hy, zi = −[y, z] = −yz + zy
hx1, . . . , xm; y, zi = hu; y, zi = −p(u, y, z) + p(u, z, y)
ΦSU (x1, . . . , xm; y1, . . . , yn) =
1
1
m!
n! Xτ ∈Σm,σ∈Σn
p(xτ (1), . . . , xτ (m); yσ(1), . . . , yσ(n−1); yσ(n))
with u = ((x1x2)··· )xm, Σm the symmetric group on m letters and m ≥ 1, n ≥ 2. Then the identities of a
Sabinin algebra are satisfied for all xi, yj, z in an arbitrary non-associative algebra, [55].
In particular, we get a functor from non-associative algebras to Sabinin algebras
A 7→ UX(A)
that assigns to an algebra A the Sabinin algebra UX(A) on the same vector space. In the case when A is a
bialgebra, the primitive elements of A form a Sabinin subalgebra
Prim(A) ⊂ UX(A),
and we have a functor
generalizing the corresponding functor from associative bialgebras (or Hopf algebras) to Lie algebras.
A → Prim(A)
4.3. Universal enveloping algebras for Sabinin algebras. It was proved in [44] that each Sabinin
algebra l can be realized as the subspace of primitive elements in a certain non-associative irreducible Hopf
algebra U (l). Moreover, the correspondence
is functorial.
l → U (l)
When l is a Lie algebra, the Hopf algebra U (l) is the usual universal enveloping algebra of l. The properties
of U (l) also closely mirror those of the associative enveloping algebras. There is a version of the Poincar´e-
Birkhoff-Witt theorem which says that U (l), as a coalgebra, is isomorphic to the symmetric algebra k[l].
12
As a corollary, if l is given a basis (ei), the algebra U (l) is additively spanned by the left normed products
(. . . (ei1 ei2 ) . . .)ein with ik ≤ ik+1. As a consequence, Hopf algebras U (l) are right and left Noetherian for
finite-dimensional l and have no zero divisors.
Most importantly, we have the Milnor-Moore theorem which states that each irreducible Hopf algebra is
the universal enveloping algebra of the Sabinin algebra of its primitive elements:
U = U (Prim(U )).
This implies that the universal enveloping algebra functor is an equivalence of the category of Sabinin algebras
with that of irreducible Hopf algebras.
An explicit construction of U (l) can be found in [44] or in Section 4.3 of [38].
It seems plausible that for a flat Sabinin algebra l represented as complement to a Lie subaIgebra h in a
Lie algebra g the algebra U (l) should have an explicit description in terms of U (g) and U (h). At the moment
we only have a description of this kind for Malcev algebras and Lie triple systems (Section 6).
4.4. Loops and Hopf algebras of distributions. A distribution on a manifold M is a linear functional
on the functions on M . A distribution is said to be supported at a point if its value on a function only
depends on the germ of the function at this point. Write DeM be the space of distributions on a manifold
M supported at the point e ∈ M . Under some reasonable conditions, which we shall assume to be satisfied,
DeM as a vector space is spanned by the Dirac delta function at e and all of its (higher) derivatives.
The space DeM has the natural structure of a coalgebra with the coproduct induced by the diagonal map
M → M × M . As a coalgebra, it is isomorphic to the symmetric algebra k[TeM ], and the space of primitive
elements in DeM consists of all the derivatives along some vector in TeM . In particular, it is cocommutative,
coassociative and irreducible, since all these properties hold in k[TeM ].
A local loop structure U × U → M in a neighbourhood of e induces a product
DeU ⊗ DeU → DeM = DeU.
This product is readily seen to be a coalgebra homomorphism, and, hence, DeU = DeM is a Hopf algebra.
If L is a local loop on U , the notation DeL will stand for DeU with this Hopf algebra structure.
In the case when a local loop L is analytic, the product on DeL can be expressed in terms of the
coefficients of the power series that define L. In particular, this makes it possible to define the Hopf algebra
of distributions for any formal loop, and the explicit formulae in the formal case are quite straightforward.
Indeed, any formal map
can be lifted to a unique coalgebra morphism
θ : k[V ] → W
θ′ : k[V ] → k[W ],
which is defined as
∞
θ′(µ) =
In particular, any formal loop
lifts to a product
1
n!
θ(µ(1))··· θ(µ(n)) = ǫ(µ)1 + θ(µ) + ··· .
Xn=0
F : k[V × V ] = k[V ] ⊗ k[V ] → V
which gives k[V ] the structure of a non-associative Hopf algebra. If the series that define F converge, this
Hopf algebra coincides with the Hopf algebra of the distributions DeF .
F ′ : k[V ] ⊗ k[V ] → k[V ],
By the Milnor-Moore and the Poincar´e-Birkhoff-Witt theorems, each irreducible Hopf algebra can be
considered as a product on k[V ] for some vector space V . Taking the primitive part of this product we get
a formal loop:
This establishes an equivalence between the category of formal loops and that of irreducible Hopf algebras.
k[V ] ⊗ k[V ] → k[V ] → Prim(k[V ]) = V.
13
4.5. The two structures of a Sabinin algebra on the tangent space to a loop. Given a local loop
L on a manifold M , there are two ways to obtain the operations of a Sabinin algebra on TeM . The first
way was described by Sabinin and Mikheev: the brackets are the derivatives of the torsion tensor of the
canonical connection of L and the multioperator Φ measures the failure of L to be right alternative. The
second possibility is to use the Sabinin algebra operations, as defined by Shestakov and Umirbaev, on the
space of primitive elements in the Hopf algebra DeL.
It turns out that these two Sabinin algebra structures are almost, though not quite, the same. Namely, the
brackets in both Sabinin algebras coincide, while the Mikheev-Sabinin multioperator Φ is different from the
Shestakov-Umirbaev multioperator ΦSU . At the moment it is not known how to express the Mikheev-Sabinin
multioperator in terms of the Shestakov-Umirbaev primitive operations.
4.6. Exponentials, logarithms and the Baker-Campbell-Hausdorff formula. In the associative case,
the exponential map is the unique, up to a rescaling, power series that sends the primitive elements of a
complete Hopf algebra bijectively onto its group-like elements. Since the group-like elements in a Hopf
algebra form a group, the Baker-Campbell-Hausdorff formula (1) endowes the Lie algebra of the primitive
elements with the structure of a group. This construction can be applied to any Lie algebra so as to obtain
the corresponding formal Lie group.
Complete Hopf algebras can be defined in the non-associative context as well; the exponential and logarith-
mic power series in this situation are series in one non-associative variable. In contrast with the associative
case, however, the exponential and the logarithm are not defined uniquely by the condition that they inter-
change the primitive and the group-like elements. In fact, this condition is satisfied by an infinite-dimensional
affine family of power series all of which can be used to define a functor from Sabinin algebras to formal
loops via a formula of Baker-Campbell-Hausdorff type. For instance, there exists a unique non-associative
exponential series with the additional property that
exp 2x = exp x exp x
and whose linear term is x. The first few terms of the corresponding Baker-Campbell-Hausdorff formula
were calculated explicitly by Gerritzen and Holtkamp in [14].They do not speak of Sabinin algebras and, in
particular, do not discuss whether their formula provides any kind of integration.
A different exponential series was considered in [38]. Replace in the usual exponential series the term xn
with the left-normed product of n copies of a non-associative variable x:
exp x = 1 + x +
1
2!
x2 +
1
3!
x2x +
1
4!
(x2x)x + . . .
The exponential defined in this way is, essentially, the exponential map of the canonical connection on the
loop of formal power series in one non-associative variable and non-zero constant term.
The non-associative logarithm, which is defined as the inverse of the exponential, can be calculated
explicitly and its coefficients involve Bernoulli numbers. Let f : N → R be a sequence of elements of some
ring R, indexed by the natural numbers. It can be extended to an R-valued function defined on all non-
associative monomials in x. Given a non-associative monomial τ define fτ inductively as follows. For τ = x
set fτ = 1. If τ 6= x, there is only one way of writing τ as a product (. . . ((xτ1)τ2) . . .)τk. Set
where τ varies over the set of all non-associative monomials in x, and Bτ and τ ! are the extensions of the
Bernoulli numbers and the factorial, respectively, to the set of non-associative monomials [38].
Rather than use (1), we define the formal power series L(x, y) in two non-associative variables x and y
by the formula
where all the products inside the logarithm are assumed to be left-normed. It can be seen that this series
can be written as an infinite linear combination of primitive operations in x and y. If applied to a Sabinin
L(x, y) = log
Xm,n≥0
1
m!n!
xmyn
,
14
With this notation we have
fτ = fk · fτ1 · . . . · fτk .
log(1 + x) =Xτ
Bτ
τ !
· τ,
algebra l, it gives a formal loop whose tangent algebra has the same brackets as l but whose multioperator
is zero. In particular, it provides formal integration for flat Sabinin algebras [39].
If l is not flat, one defines the Baker-Campbell-Hausdorff series to be
where Φ(x, y) is the power series defined by the multioperator as in (12). This formula provides formal
integration for arbitrary Sabinin algebras.
L(x, Φ(x, y)),
4.7. Akivis algebras and their envelopes. An Akivis algebra is a vector space with one bilinear and one
trilinear operation, denoted by [·,·] and (·,·,·) respectively, such that
and
[x, y] = −[y, x]
[[x, y], z] + [[y, z], x] + [[z, x], y] = (x, y, z) + (y, z, x) + (z, x, y) − (x, z, y) − (z, y, x) − (y, x, z)
for all x, y, z. Any algebra is an Akivis algebra, with [·,·] being the commutator and (·,·,·) the associator. If
A is an algebra, denote by Ak(A) the algebra A considered as an Akivis algebra. Each Sabinin algebra is also
automatically an Akivis algebra since the commutator and the associators are both primitive operations.
It was proved in [53] that for each Akivis algebra a there is a Hopf algebra UAk(a) together with an
injective morphism of Akivis algebras a ֒→ UAk(a) which is universal in the sense that any Akivis algebra
morphism a → Ak(A) lifts to a unique algebra homomorphism UAk(a) → A.
If a is actually a Sabinin algebra, UAk(a) is, in general, different from U (a). In particular, the image of
a in UAk(a) is contained in the subspace of primitive elements, but not every primitive element of UAk(a)
comes from a.
4.8. Dual Hopf algebras. Inverting the arrows in the categorical definition of an associative Hopf algebra
one arrives to the definition of the same structure. In the non-associative case this is no longer true.
A unital bialgebra H with the coproduct δ and the counit ǫ is called a dual Hopf algebra if is endowed
with two additional linear operations H → H ⊗ H, called the right and the left codivision
such that
and
x 7→X x/1/ ⊗ x/2/,
x 7→X x\1\ ⊗ x\2\,
X x/1/(1) ⊗ x/1/(2)x/2/ =X x(1) ⊗ ǫ(x(2)) =X x(1)/1/ ⊗ x(1)/2/x(2)
X x\1\x\2\(1) ⊗ x\2\(2) =X ǫ(x(1)) ⊗ x(2) =X x(1)x(2)\1\ ⊗ x(2)\2\.
The dual of a finite-dimensional Hopf algebra is, rather tautologically, a dual Hopf algebra. In particular,
the space of functions on a finite loop is a dual Hopf algebra, with the codivisions defined as
It is dual to the loop algebra, which is a cocommutative and and coassociative Hopf algebra.
X f/1/(x) ⊗ f/2/(y) = f (x/y)
and X f\1\(x) ⊗ f\2\(y) = f (y\x).
A more general example is the space of functions on an algebraic loop.
An (affine) algebraic loop is an affine variety with the structure of a loop for which the multiplication and
both divisions are morphisms (regular functions). Since the space O(X) of regular functions on any affine
variety X has the property that
O(X × X) = O(X) ⊗ O(X),
it follows that for an affine algebraic loop X the space O(X) is a commutative and associative dual Hopf
algebra, with the codivisions defined in the same fashion as for finite loops. Moreover, over an algebraically
closed field k the category of affine algebraic loops is equivalent to that of finitely generated commutative
and associative dual Hopf algebras. (The details are, essentially, the same as for affine algebraic groups.)
Remark. In principle, one may consider bialgebras with two divisions and two codivisions satisfying the
idenitites that hold in Hopf algebras and in dual Hopf algebras. However, it can be shown that, in general,
there is no way to introduce codivisions on the Hopf algebras that are of main interest here, namely, the
universal enveloping algebras of Sabinin algebras.
15
5. Identities and special classes of loops
5.1. Identities in Hopf algebras coming from identities in loops. Informally speaking, a non-associative
word is an expression formed from several indeterminates by applying the multiplication and the right and
the left divisions. (In particular, a monomial is a non-associative word formed using multiplication only.)
If w is such a word, a local loop on M is said to satisfy the identity w = e if for all the values of the
indeterminates in M for which w is defined, the result of performing all the operations in w is the unit.
More generally, one can speak of the identities of the form w1 = w2 where w1 and w2 are two words. As we
already pointed out, identities also make sense in formal loops.
An identity satisfied by a local (formal) loop can be lifted to an identity in its algebra of distributions. If
we start with a formal loop F on a vector space V , then a non-associative word w in n variables gives rise
to a formal map w : k[V ]⊗n → V which can be lifted to a coalgebra map w′ : k[V ]⊗n → k[V ]. Then the
identity
(14)
holds in F if and only if we have
(15)
w1 = w2
w′
1 = w′
2
with any choice of arguments in DeF . The identity (15) in the algebra of distributions is referred to as the
linearization of the identity (14).
The simplest example is the associativity in a formal loop which is necessary and sufficient for the asso-
ciativity of its algebra of distributions. An analogous statement holds for the commutativity. A less trivial
example is the left Moufang identity
x(y(xz)) = ((xy)x)z.
It translates into the identity
(16)
in the corresponding algebra of distributions. Non-associative Hopf algebras in which (16) is satisfied are
called Moufang-Hopf algebras.
Geodesic loops are right alternative; their algebras of distributions satisfy
X µ(1)(ν(µ(2)η)) =X((µ(1)ν)µ(2))η
(17)
X µ(ν(1)ν(2)) =X(µν(1))ν(2).
5.2. Identities in Sabinin algebras coming from identities in loops. Identities in loops also produce
identities in Sabinin algebras. In principle, there is a straightforward way to find all the identities in a Sabinin
algebra that are implied by a given loop identity. Namely, the Baker-Campbell-Hausdorff formula expresses
the loop product in terms of the operations in the corresponding Sabinin algebra; imposing an identity on a
loop, therefore, imposes an infinite set of identities on its Sabinin algebra. This set can be hugely redundant
and choosing a subset consisting of independent identities is difficult in practice. Nevertheless, this is not
necessarily an impossible task: Kuzmin [26] used this method to show that a Malcev algebra integrates to a
local Moufang loop.
Since loop identities translate into identities in Hopf algebras and in Sabinin algebras one can state various
specializations of the general Lie theory for loops that satisfy a given set of identities. The most fundamental
example of this is the Lie theory for right alternative loops which relates them to flat Sabinin algebras and
Hopf algebras satisfying (17). There are many other such Lie theories for loops with identities; we describe
some of them below.
5.3. Moufang loops and Malcev algebras. Historically, the first generalization of the Lie theory to non-
associative products was the theory of Moufang loops and Malcev algebras. Recall that a Malcev algebra is
a vector space with an antisymmetric bracket that satisfies
where
[J(a, b, c), a] = J(a, b, [a, c])
J(a, b, c) = [[a, b], c] + [[b, c], a] + [[c, a], b].
16
In this setup, we have the following three equivalent categories: simply-connected finite-dimensional Moufang
loops, finite-dimensional Malcev algebras and finitely generated irreducible Moufang-Hopf algebras. The
equivalence of first two categories was established in the works of Malcev [30], Kuzmin [26] and Kerdman
[23], while the third category was introduced only recently in [41].
An important feature of this case is the fact that finite-dimensional Malcev algebras always integrate to
globally defined Moufang loops. While the original proof of this fact [23] is rather inaccessible, the modern
argument uses the theory of groups and algebras with triality and consists in translating the problem into
the context of Lie groups and Lie algebras, see Section 6 for more details. The Ado theorem also holds for
finite-dimensional Malcev algebras [41], see more in Section 7.
Moufang loops provide an example of how one can pass from loop identities to Sabinin algebra identities
via the identities in the bialgebra of distributions. Given a Moufang loop M , the Moufang-Hopf identity
(16), which holds in DeM , implies that the tangent space TeM = PrimDeM belongs to the generalized
alternative nucleus of DeM . For an algebra A the generalized alternative nucleus is defined as
Nalt(A) = {a ∈ A (a, x, y) = −(x, a, y) = (x, y, a) for all x, y ∈ A}.
It is a Malcev algebra with the bracket given by [a, b] = ab − ba. The rest of the Sabinin algebra operations
in TeM can be expressed via the Malcev bracket [44]; as a consequence, the Malcev algebra structure is
sufficient to reconstruct a local Moufang loop.
A long-standing open problem proposed by Kuzmin [26] is whether all Malcev algebras are special, that
is, whether each Malcev algebra can be embedded into the commutator algebra of an alternative algebra.
The formal integration procedure gives an embedding of a Malcev algebra into the generalized alternative
nucleus of its universal enveloping algebra. However, the universal enveloping algebra of a Malcev algebra is
not alternative, in general. One can define the universal alternative enveloping algebra of a Malcev algebra
m as the maximal alternative quotient of U (m). Explicit calculations of both kinds of enveloping algebras
for specific Malcev algebras can be found in [7, 9, 57].
5.4. Bol loops and Bol algebras. The variety of left Bol loops is defined by the left Bol identity
x(y(xz)) = (x(yx))z.
Its linearization is the identity
(18)
A Bol algebra is a vector space with one bilinear antisymmetric bracket and one trilinear bracket that satisfy
the following identities:
X µ(1)(ν(µ(2)η)) =X(µ(1)(νµ(2)))η.
[a, a, b] = 0,
[a, b, c] + [b, c, a] + [c, a, b] = 0,
[x, y, [a, b, c]] = [[x, y, a], b, c] + [a, [x, y, b], c] + [a, b, [x, y, c]]
and
[a, b, [x, y]] = [[a, b, x], y] + [x, [a, b, y]] + [x, y, [a, b]] + [[a, b], [x, y]].
The category of infinitesimal analytic finite-dimensional Bol loops is equivalent to the category of finite-
dimensional left Bol algebras and to the category of irreducible Hopf algebras satisfying (18). While the
correspondence between Bol loops and Bol algebras is well-studied (see [33, 40, 50]), the Hopf algebras in
this context were first considered in [43].
Similarly to the case of Moufang loops, one can arrive to the Bol algebra identities via the distribution
bialgebras. The identity (18) satisfied in the bialgebra of distributions shows that the primitive elements
always belong to the left alternative nucleus of the distribution bialgebra, where the left alternative nucleus
of an algebra A is defined as
Any subspace of LNalt(A) which is closed under the operations
LNalt(A) = {a ∈ A (a, x, y) = −(x, a, y) for all x, y ∈ A}.
[a, b] = ab − ba
17
and
is a (left) Bol algebra. This is the case, in particular, for TeM inside LNalt(DeM ), where M is a Bol loop
[43]. All the other Sabinin algebra operations can be expressed in terms of these two brackets and, thus the
structure of a Bol algebra is sufficient to reconstruct a local left Bol loop.
[a, b, c] = a(bc) − b(ac) − c[a, b]
The speciality problem for Malcev algebras has its generalization to Bol algebras. Namely, one may ask
whether each Bol algebra is contained in a left alternative algebra as a Bol subalgebra. I.Hentzel and L.Peresi
have shown that this is not the case [19]. Using computer-aided computations, they found an identity of
degree eight which is satisfied in the Bol algebra of any left alternative algebra, but not in the free Bol
algebra.
An important property that holds for Malcev algebras but not for general Bol algebras is the Ado theorem:
it was proved in [45] that it fails for Lie triple systems (which are Bol algebras with the trivial binary bracket).
5.5. Nilpotent loops and nilpotent Sabinin algebras. Let us say that a loop L is nilpotent of class n
if the n + 1st term of the commutator-associator filtration (see Section 8.1) of L is trivial:
This is not the standard definition of nilpotency in loops; however, there are strong indications that it is the
correct one [18, 34, 35, 39]. Since the commutator-associator filtration is defined in terms of words (namely,
commutators, associators and associator deviations and their compositions) nilpotent loops of class n form
a variety.
γn+1L = {e}.
The corresponding Sabinin algebras, unsurprisingly, are the nilpotent Sabinin algebras of class n. Define
a bracket of weight k in a Sabinin algebra to be a multilinear operation in k arguments formed by composing
the brackets and the operations Φm,n. Then a Sabinin algebra is nilpotent of class n if all the brackets of
weight greater than n vanish in it. For a flat Sabinin algebra, nilpotency is the same as the nilpotency of
the Lie algebra in which it is a summand according to the construction of Section 3.2, see [39].
The Lie theory of nilpotent loops and nilpotent Sabinin algebras exhibits various features typical of Lie
and Malcev algebras. In particular, a nilpotent Sabinin algebra always integrates to a globally defined simply-
connected nilpotent loop of the same class. This is due to the fact that the Baker-Campbell-Hausdorff series
in this case consists of a finite number of terms and, hence, always converges. Also, the Ado theorem holds,
both for Sabinin algebras and (globally) for simply-connected nilpotent loops (see Section 7.4 for more details
on the Ado theorem in the non-associative setting). While in the case of Malcev algebras one may think that
these good properties are a consequence of Malcev algebras being "close" to Lie algebras, nilpotent Sabinin
algebras may have an arbitrarily big (though finite) number of independent operations.
5.6. Connections with quasigroups. In various contexts in geometry where non-associative algebra
proves to be useful, it is quasigroups rather than loops that appear naturally. While we do not intend
to discuss this subject in detail, there are two points that we should mention here.
One important application of quasigroups is the theory of symmetric spaces as developed by Loos [27];
this is where bialgebras first appeared within non-associative Lie theory. A symmetric space can be defined
algebraically as a manifold M with a product ◦ that satisfies, for all x, y ∈ M , the following identities:
x ◦ x = x,
x ◦ (x ◦ y) = y,
x ◦ (y ◦ z) = (x ◦ y) ◦ (x ◦ z)
and such that the map Sx : y 7→ x ◦ y has the unique fixed point y = x in some neighbourhood of x.
Geometrically, the product x◦ y is the reflection of the point y about the point x. The fact that each e ∈ M
is idempotent is sufficient to define a bialgebra structure on DeM for each e and linearize the identities in
M to obtain identities in DeM . For instance, if M is a symmetric space, in each DeM we have:
The local uniqueness of the fixed point for the multiplication map Se implies that the right multiplication
in DeM by the only group-like element
µ(ην) =X(µ(1)η)(µ(2)ν).
(19)
X µ(1)µ(2) = µ, X µ(1)(µ(2)ν) = ǫ(µ)ν,
e : f 7→ f (e)
18
is bijective. This, together with the identities (19), can be used to deduce algebraically that the tangent
space TeM = Prim(DeM ) is a Lie triple system with the product [a, b, c] = a(bc) − b(ac).
The second point is related to the terminology concerning the identities in loops. Given a local quasigroup,
that is, a locally invertible map M × M → M defined in a neighbourhood of a point (a, b), one can define a
local loop on a neighbourhood of a · b ∈ M by
xy = (x/b) · (a\y),
where ·,\ and / are the product and the two divisions in the quasigroup. This construction associates with
a local quasigroup a family of local loops, and it may be useful to compare the identities that hold in each
local loop. It turns out that the Moufang identity is satisfied at the same time in each of its local loops; in
particular, it is sufficient to verify this identity in one of the local loops in order to conclude that it holds in
all of them. In contrast, the monoassociativity identity
(20)
(xx)x = x(xx)
may be satisfied in one of the local loops but not in the others. A monoassociative quasigroup is defined as a
local quasigroup all of whose local loops satisfy (20). (This definition is of importance in web theory where
monoassociative quasigroups correspond to hexagonal three-webs.) As a consequence, a monoassociative
local loop is sometimes defined as a local loop of a monoassociative quasigroup [1, 8]. This is a much
stronger condition on a loop than just the identity (20).
In particular, each right alternative local loop
satisfies (20); note that flat Sabinin algebras have an infinite number of independent operations. On the
other hand, Sabinin algebras tangent to monoassociative loops have three independent operations: one
bilinear, one trilinear and one quadrilinear [32, 52]. The complete set of relations for these operations is not
known.
6. Constructions involving associative Hopf algebras
Since each Sabinin algebra can be thought of a subspace of a Lie algebra, it is not surprising that non-
associative Hopf algebras can be interpreted in terms of associative Hopf algebras. The construction of the
universal enveloping algebra of a Sabinin algebra given in [44] is based on such an interpretation. In general,
this construction does not produce closed formulae; however, there are important special cases when it can be
significantly simplified. This happens, for instance, for Moufang-Hopf algebras and for universal enveloping
algebras of Bol algebras.
6.1. Groups with triality. One important step in the study of Malcev algebras was to establish its relation
with groups with triality [15, 13, 31]. Each self-map d of the octonion algebra, antisymmetric with respect
to its standard quadratic form, gives rise to two antisymmetric maps d′, d′′ uniquely defined by the relation
d(xy) = d′(x)y + xd′′(y).
This phenomenon is known as the local triality principle and, in general, it is often related in one or another
way with exceptional behaviour such as that of exceptional Lie algebras or Jordan algebras [21]. Behind it,
there is a certian representation of the symmetric group Σ3 on three letters, acting on the Dynkin diagram
of the multiplication Lie algebra of the split octonions, which is a central simple Lie algebra of type D4 [51].
This symmetry lifts to the corresponding group Spin(8) and gives it the structure of a group with triality.
More generally, an abstract group with triality is a group G together with two automorphisms σ, ρ such that
and
σ2 = ρ3 = id,
σρσ = ρ2,
P (g)ρ(P (g))ρ2(P (g)) = 1,
where P (g) = gσ(g)−1. The automorphisms σ and ρ define a representation of Σ3, which is independent on
a particular choice of the generators σ y ρ of orders 2 and 3, respectively. If H ⊆ G is the subgroup fixed by
σ, the quotient G/H is a Moufang loop with the product
g1H ∗ g2H = ρσ(P (g1))g2H.
19
Let g and h be the Lie algebras of G and H, respectively; the group Σ3 acts on g with h being fixed by σ.
The Malcev algebra m tangent to G/H is the complement to h in g consisting of the eigenvectors of σ with
eigenvalue −1. The bracket in m is defined by
This is the construction of a Lie algebra with triality. Each Malcev algebra arises as the −1-eigenspace of σ
in such a Lie algebra.
In order to construct the universal enveloping algebra U (m), embed G/H into G by means of the map
[x, y]m = [ρ2(x) − ρ(x), y].
gH 7→ ρ(σ(P (g))),
M(G) = {P (g) g ∈ G}.
which is a global section of the principal fibration G → G/H. The image of this embedding is the set
The Moufang product on M(G) which comes from G/H can be expressed directly in terms of the product
on G as
see [16]. Linearizing this formula we get that the subcoalgebra
x ∗ y = ρ(x)−1yρ2(x)−1,
where S is the antipode of U (g), with the product
M(U (g)) =nX x(1)σ(S(x(2))) x ∈ U (g)o ,
is isomorphic to U (m), see [3].
x ∗ y =X ρ(S(x(1)))yρ2(S(x(2)))
One can speak of the categories of Lie groups with triality, Lie algebras with triality and Hopf algebras
with triality. The constructions of this paragraph establish their equivalence to the categories of analytic
Moufang loops, Malcev algebras and Moufang-Hopf algebras, respectively.
In this way, for instance, the
global integrability of finite-dimensional Lie algebras implies global integrability of finite-dimensional Malcev
algebras.
6.2. Bruck loops. A similar construction works for Bruck loops. These are left Bol loops that satisfy the
automorphic inverse identity
(xy)−1 = x−1y−1,
where x−1 is shorthand for e/x. The binary operation in the corresponding Bol algebra in this case is trivial,
and the structure of a Bol algebra reduces to that of a Lie triple system. This shows that Bruck loops are
related to symmetric spaces. Indeed, the product
endowes an arbitrary local Bruck loop with the structure of a locally symmetric space. Conversely, given a
point e in a symmetric space the Bruck loop product is recovered by means of the relation
x ◦ y = x(y−1x)
xy ◦ e = x ◦ (e ◦ (y ◦ e)).
While the product of a locally symmetric space can be globalized, this is not true for the corresponding local
loop since the map x 7→ x ◦ e may be only locally, and not globally, bijective.
A symmetric space can be considered as a quotient G/H, where G is a Lie group with an involutive
automorphism σ and H is an open subgroup in the fixed set Gσ. Locally, the quotient G/H can be embedded
into G with the help of the map
which is a local section of the quotient map G → G/H. The image of this map coincides, locally, with the
set
gH 7→ P (g) = gσ(g)−1,
M(G) = {P (g) g ∈ G}.
It can be shown that in a neighbourhood of the unit in M(G) each element has a unique square root. The
product of a local Bruck loop on M(G) can be written then as
x ∗ y = √x y √x
20
for x, y ∈ M(G).
Similarly, a Lie triple system can be considered as the −1-eigenspace of an involutive automorphism in a
Lie algebra g. As a consequence, the universal enveloping algebra of a Lie triple system, viewed as a Sabinin
algebra, can be defined in terms of the Hopf algebra U (g) and the automorphism σ as the coalgebra
with the product
M(U (g)) =nX x(1)σ(S(x(2))) x ∈ U (g)o ,
where r is the map corresponding to the linearization of the square root, see [37].
x ∗ y =X r(x(1))yr(x(2))
Moufang loops and Bruck loops are only two examples of the situation when the algebra of distributions
can be constructed with the help of an associative Hopf algebra. In general, we get such a construction
whenever we have a lifting of a local product in G/H to the product in G by means of a local section.
7. Representation theory
Compared to the representation theory of Lie groups and Lie algebras, the representation theory of loops
and Sabinin algebras is still in its infancy. It is, probably, too early to give a coherent picture of the field;
here we touch only on a few topics.
The theory of loop representations has been put on firm foundations by J.D.H. Smith [56]; the same
approach works for Sabinin algebras and non-associative Hopf algebras.
It rests on the observation of
J. Beck [2] that the concept of a split-null extension has a natural definition in the categorical setting: if X
is an object in a category C, an X-module is an abelian group in the comma category C ↓ X. The objects
of this category are morphisms Y π→ X and the morphisms are commutative diagrams of the form
Y
Z
π
π
X
This categorical description encompasses all reasonable definitions of modules; in particular, that of a
group representation and that of a module over a Lie algebra. It has an advantage of producing immediate
bijections between the representations of a formal group, of its Lie algebra and of the universal enveleping
algebra of the latter. The same is true in the non-associative context: the equivalence of the category of
formal loops to that of Sabinin algebras and to the category of irreducible Hopf algebras identifies that the
corresponding sets of representations. Let us spell out what Beck's definition means in practice.
7.1. Loop representations. Let Loops be the category of all loops and let L be an arbitrary loop. Intro-
duce the following notation:
• L[X] - the free product of L with the free loop on one generator X;
• La and Ra - the operators of left and right multiplication, respectively, by a ∈ L in L[X];
• U(L) - the group of self-maps of L[X] generated by the La and Ra for all a ∈ L;
• U(L)e - the stabilizer of the identity in L[X].
Abelian groups in the category Loops ↓ L are in one-to-one correspondence with the representations of the
group U(L)e. Namely, given a U(L)e-module Ee, define
as a set, endowed with the product
(21)
where
E = Ee × L
(x, a)(y, b) = (ra,b(x) + sa,b(y), ab),
ra,b = R−1
ab RbRa and sa,b = R−1
ab LaRb.
Then the projection of the loop E onto L is an abelian group in Loops ↓ L. The sum, inverse and the
neutral element that define on E → L the structure of an abelian group are induced by the corresponding
21
operations in Ee. Each abelian group in Loops ↓ L arises in this way, with the module Ee defined uniquely
up to an isomorphism.
If L belongs to a variety of loops V, the category Loops in this construction can be replaced by V. The
free loop and the free product in this case should be taken in V; instead of the group U(L)e one obtains a
group U(L; V)e. The abelian groups in V ↓ L are then in correspondence with certain representations of
U(L; V)e. For instance, if V = Groups is the variety of groups, we have
Given a representation
the product (21) takes the form
See [56] for further details.
U(L; Groups)e ≃ L.
ρ : L → Aut(V )
(x, a)(y, b) = (x + ρa(y), ab).
7.2. Hopf and Sabinin algebra representations. Let H be the category of all irreducible Hopf algebras
and H an object in H. Beck's definition translates to this context in the manner similar to the case of loops.
An abelian group A π−→ H in the category H ↓ H is a morphism such that the Hopf algebra A factorizes
as
where k[V ] is the symmetric algebra of a vector space V and the product in A is given by
A ∼= k[V ] ⊗ H,
in terms of the action of some operators ra,b, sa,b on V whose action is extended to k[V ] as
(p ⊗ a)(q ⊗ b) =X(ra(1),b(1) (p))(sa(2),b(2) (q)) ⊗ a(3)b(3)
and, similarly,
The subalgebra k[V ] appears naturally as the equalizer of the coalgebra morphisms π : A → H and
ra,b(pq) =X ra(1),b(1) (p) ra(2),b(2) (q)
sa,b(pq) =X sa(1),b(1) (p) sa(2),b(2) (q).
ǫ1 : A → H
x 7→ ǫ(x)1.
The sum, inverse and the neutral element in A π−→ H are given by
P(p ⊗ a(1)) ⊗ (q ⊗ a(2))
π
+
pq ⊗ a
(p, a)
−
(S(p), a)
a
π
π
π
π
ǫ(p)ǫ(q)a
ǫ(p)a
1 ⊗ a
π
0
a
where S is the antipode in k[V ]. The category H ↓ H has finite products: The product A1⊗H A2 of A1
and A2
π2−→ H is the equalizer of the morphisms π1 ⊗ ǫ and ǫ ⊗ π2.
Usually, in the associative context it is the vector space V that is called an H-module and the abelian
−→ H is not given a special name. In this case the action of ra,b is trivial, that is, ra,b(p) = ǫ(a)ǫ(b)p,
group A
while sa,b(q) = ǫ(b)sa,1(q). If we write a · q = sa,1(q), the formula for the product in A becomes
π
π1−→ H
which is nothing but the smash product of the H-module algebra k[V ] with H, where the H-module algebra
structure on k[V ] is induced by the H-module structure on V .
(p ⊗ a)(q ⊗ b) =X p(a(1) · q) ⊗ a(2)b,
Modules over Sabinin algebras are somewhat easier to describe than modules over loops and Hopf algebras.
Beck's notion of a module for a Sabinin algebra l coincides with the usual idea of a split-null extension. In
other words, a module over l is a Sabinin algebra V ⊕ l such that l is a subalgebra and such that each
multilinear operation vanishes if at least two of its arguments belong to V .
22
7.3. An extended concept of a representation. Sometimes there is rationale for enlarging the class of
the extensions considered in the definition of the modules over a loop (Hopf algebra, Sabinin algebra). For
instance, the requirement that the extension π : E→L lies within the variety M of Moufang loops leads to
conditions on ra,b and sa,b ∈ U(L;M)e; these conditions are expressed as the vanishing of certain elements
of the group algebra of U(L;M)e while acting on π−1(e). These elements may fail to vanish on a tensor
product of U(L;M)e-modules. As a consequence, in contrast to the case of groups, a tensor product of
L-modules in general will not be a L-module.
In the same fashion, a tensor product of Sabinin algebra
representations cannot be expected to be a representation.
The failure of the tensor product of two representations to be a representation is a basic reflection of
non-associativity. In case of the Moufang loops this problem can be overcome by studying abelian groups
E → L such that E is not necessarily a Moufang loop, but requiring only that the image of the zero section
L → E is contained in the set of Moufang elements of E. These are the elements a such that
a(x(ay)) = ((ax)a))y
and ((xa)y)a = x(a(ya))
for all x, y ∈ E; Moufang elements in any loop form a Moufang loop. In the context of Hopf algebras this
corresponds to studying abelian groups A π−→ H such that in the decomposition A ∼= k[V ]⊗ H the subspace
Prim(H) (but not V ) is contained in the generalized alternative nucleus of A. As for Malcev algebras,
representations of this kind correspond to split-null extensions (V ⊕ m, [ , ]) of the Malcev algebra m such
that the bracket [ , ] is anticommutative and the adjoint map
satisfies
(22)
ra : x 7→ ra(x) = [x, a]
[[ra, rb], rc] = −[r[a,b], rc] + r[[a,b],c]+[[a,c],b]+[a,[b,c]].
Not only the resulting representation theory behaves well with respect to tensor products but also among
its other good properties is the existence of faithful finite-dimensional representations for finite-dimensional
Malcev algebras, see [41].
In general, no clear criterion is known which would determine the adequate class of abelian groups E
−→ L
for a given loop L. Note that once such class is chosen, methods of non-associative Lie theory provide us
with the classes of extensions for the corresponding Hopf and Sabinin algebras.
π
7.4. The Ado theorem. The Ado theorem is usually stated as a result about representations. While this
may be appropriate in the context of Lie algebras, in the more general setting of Sabinin algebras it is more
convenient to use a different wording which does not mention representations at all.
Each Sabinin algebra appears as a Sabinin sublagebra of UX(A) for some non-associative algebra A; for
instance, one can take A to be the universal enveloping algebra U (l)). Let us say that a variety of Sabinin
algebras satisfies the Ado theorem if for each finite-dimensional Sabinin algebra l in that variety A can be
chosen to be finite-dimensional. By the universal property of U (l), this is the same as to say that the U (l)
has an ideal of finite codimension which contains no non-trivial primitive elements.
We have mentioned before that, while Lie, Malcev and nilpotent Sabinin algebras satisfy the Ado theorem,
this is not the case for all Sabinin algebras, Lie triple systems being a counterexample. The universal
enveloping algebra of a central simple Lie triple system of finite dimension has no proper right ideals apart
from the augmentation ideal [37, 45]. It would be interesting to find criteria for a variety of loops to satisfy
the Ado theorem.
The Ado theorem can be also considered as a statement about loops rather than their tangent algebras.
Let us say that a local loop Q is locally linear if can be locally embedded as a subloop of of the local loop
of invertible elements in some finite-dimensional algebra A. This is equivalent to saying that the Sabinin
algebra of Q is isomorphic to a subalgebra of UX(A), which is the property described by the Ado theorem.
Since the Ado theorem fails for Lie triple systems, local Bruck loops are not locally linear in general.
We should point out that local linearity is a weaker concept than global linearity. This has nothing
to do with non-associativity since the difference already exists for groups. For Moufang loops there also
are examples which illustrate this phenomenon. For instance, the seven-dimensional projective plane is a
Moufang loop which locally embeds into the invertible elements of the octonions. This local embedding,
however, cannot be globalized [54].
23
8. Discrete loops and Sabinin algebras
The definition of a of a nilpotent loop, and, generally, of the lower central series of a loop, was given by
R.H. Bruck in [10]. Since then, the theory of nilpotent groups has made significant advances and it became
increasingly clear that Bruck's definition does not provide an adequate analogue of the associative lower
central series. Namely, one should expect that
abelian groups;
• the successive quotients of the lower central series for a finitely generated loop are finitely generated
• the graded abelian group associated with the lower central series carries an algebraic structure similar
• there exists a close relation to the dimension series.
to that of a Lie ring;
A lower central series for loops satisfying all the above properties has been defined and studied in [34, 35, 36]
and we review this construction in this section. A closely related definition was also given by F. Lemieux, C.
Moore and D. Th´erien in [28]. Essentially, the argument of [28] consists in defining the dimension series for
the free loops and then pushing it to arbitrary loops using the universal property of the free loops. Recently,
M. Hartl and B. Loiseau defined higher commutators in arbitrary semi-abelian categories which lead in the
case of loops to the same definition as the one discussed here, see [17, 18].
8.1. The commutator-associator series. Let L be a loop. The commutator-associator filtration on L is
defined in terms of commutators, associators and associator deviations. The commutator of two elements
a, b of L is
and the associator of a, b and c is defined by
[a, b] = (ab)/(ba)
(a, b, c) = ((ab)c)/(a(bc)).
There is an infinite number of associator deviations. These are functions Ll+3 → L characterized by a non-
negative number l, called level of the deviation, and l indices α1, . . . , αl with 0 < αi ≤ i + 2. The deviations
of level one are
(a, b, c, d)1 = (ab, c, d)/((a, c, d)(b, c, d)),
(a, b, c, d)2 = (a, bc, d)/((a, b, d)(a, c, d)),
(a, b, c, d)3 = (a, b, cd)/((a, b, c)(a, b, d)).
By definition, the deviation (a1, . . . , al+3)α1,...,αl of level l is equal to
A(aαlaαl+1)/(A(aαl )A(aαl+1)),
where A(x) stands for the deviation (a1, . . . , aαl−1, x, aαl+2, . . . , al+3)α1,...,αl−1 of level l − 1. The associator
is thought of as the associator deviation of level zero.
Now, set γ1L = L and for n > 1 define γnL to be the minimal normal subloop of L containing
• [γpL, γqL] with p + q ≥ n;
• (γpL, γqL, γrL) with p + q + r ≥ n;
• (γp1 L, . . . , γpl+3L)α1,...,αl with p1 + . . . + pl+3 ≥ n.
The subloop γnL is called the nth commutator-associator subloop of L. For a group G the subgroup γnG is
the nth term of the lower central series of G.
The commutator-associator subloops of a loop L are normal in L. Moreover, they are fully invariant,
that is, are preserved by all automorphisms of L. If L is finitely generated, each quotient γiL/γi+1L is a
finitely generated abelian group. The crucial property of the commutator-associator filtration is that for an
arbitrary loop L the commutator, the associator and the associator deviations induce multilinear operations
on the graded abelian group
these operations respect the grading. It can be shown that the algebraic structure given by these multilinear
operations on LγL⊗ Q is precisely that of a Sabinin algebra. This, however, is quite non-trivial and requires
an understanding of the relation between the commuator-associator filtration and the dimension series.
LγL =M γiL/γi+1L;
24
8.2. The dimension series. Let L be a discrete loop and QL its loop algebra over the rational numbers.
Denote by ∆ ⊂ QL the augmentation ideal and by ∆n its nth power, that is, the submodule of QL spanned
over Q by all products of at least n elements of ∆ with any arrangement of the brackets. The loop L can be
thought of a subset of QL and we define the nth dimension subloop of L as
DnL = L ∩ (1 + ∆n).
It can be shown that DnL is, indeed, a series of fully invariant subloops of L. The successive quotients
DiL/Di+1L are torsion-free abelian groups, and it turns out that the commutator, the associator and the
associator deviations respect the filtration by the DnL. The induced operations on the associated graded
group can be identified as follows.
The loop algebra QL is a bialgebra, with the coproduct defined as δ(g) = g ⊗ g for g ∈ L. The associated
graded algebra
is then a cocommutative non-associative primitively generated Hopf algebra. If we set
then LL⊗ Q ⊂ DL and the primitive elements of DL are precisely the elements of LL⊗ Q. Therefore, LL⊗ Q
is a Sabinin algebra. It can be shown that the Shestakov-Umirbaev operations pm,n on DL coincide with
the operations induced by certain associator deviations on LL ⊗ Q. (See [36] for a more precise statement.)
These results can be applied to the study of the commutator-associator filtration. For each n the subloop
DnL contains the commutator-associator subloop γnL. Moreover, the analogue of the Jennings theorem [22]
is true:
that is, DnL consists of all those elements of L whose kth power, for at least one k and at least one
arrangement of the parentheses, lies in γnL. In particular, we have that
∆i/∆i+1
DiL/Di+1L,
DL =Mi≥0
LL =Mi>0
DnL =pγnL,
LL ⊗ Q = LγL ⊗ Q.
Remark. The dimension subloops can be defined with the help of the group algebra with coefficients in any
ring, and the result may depend on this ring. A particularly interesting case is that of the integer coefficients.
For many years it was conjectured that the integer dimension subgroups of a group coincide with its lower
central series. The eventual counterexample published by E. Rips in 1972 was very involved, and while by
now there are more counterexamples to this conjecture, the nature of the difference between the integer
dimension series and the lower central series is still absolutely mysterious. It might be interesting to find
a loop where the difference between the integer dimension series and the lower central series is due to the
non-associative effects.
8.3. The Magnus map and the dimension series of a free loop. Let Z{{X1, . . . , Xn}} be the non-
associative ring of formal power series in n non-commutative and non-associative variables. The invertible
elements in this ring are the power series which start with ±1. They form a loop, which we denote by
Z{{X1, . . . , Xn}}∗. Let F{n} the free loop on n generators x1, . . . , xn. The homomorphism
M : F{n} → Z{{X1, . . . , Xn}}∗,
xi 7→ 1 + Xi,
is the non-associative version of the map (2). As in the associative case, the ith term of the dimension series
of F{n} consists of those elements which are sent by M to the power series of the form
1 + terms of degree at least i.
This, however, is not enough to conclude that the loop F{n} is residually nilpotent since it is not known at
the moment whether M is injective, even in the case of the free loop on one generator.
Magnus map is injective in this case, either.
A similar construction can be performed for the free commutative loop. It not known if the corresponding
The Magnus map identifies the completion of the Sabinin algebra LF{n} ⊗ Q with the primitive elements
in Z{{X1, . . . , Xn}}. In particular, this Sabinin algebra is the free Sabinin algebra on n generators.
25
9. Quantum loops
An analogy with associative Hopf algebras suggests that there might exist deformations of the universal
enveloping algebras for Sabinin algebras which are not necessarily coassociative or cocommutative. So far,
no interesting examples of this kind have been found. Let us indicate, nevertheless, what one may be looking
for.
In view of the fact that the Hopf algebra of distributions supported at the identity of a loop naturally
satisfies the linearizations of the identities satisfied by the loop, one may assume (rightly or wrongly) that
its possible deformations should satisfy the same linearized identities. Let us illustrate this logic with the
variety of Moufang loops.
The left and right Moufang identities can be written as
x y z
✞☎
x y z
✞☎
x y z
✞☎
x y z
✞☎
✝✆
=
✝✆
✝✆
✝✆
and
✝✆
=
✝✆
✝✆
✝✆
✝✆
x(y(xz))
✝✆
((xy)x)z
✝✆
x(z(yz))
✝✆
((xz)y)z
x y
✝✆= xy and
x
✞☎= δ(x)
where
with δ(x) = (x, x). The same diagrams represent the Moufang-Hopf identities satisfied in the corresponding
algebra of distributions, with the difference that in this case δ(x) =P x(1) ⊗ x(2). One may suspect that the
universal enveloping algebra of the central simple exceptional Malcev algebra of traceless octonions M(α, β, γ)
has non-cocommutative deformations that satisfy the Moufang-Hopf identities. Given that Hopf algebras
are, in spirit, self-dual objects, it is natural to require that these deformations satisfy not coassociativity
but, rather, the identities dual to Moufang-Hopf identities. These latter can be represented as
✞☎
✞☎
✞☎
✞☎
✞☎
✞☎
✞☎
✞☎
(left co-Moufang-Hopf)
✞☎
=
✞☎
(right co-Moufang-Hopf)
✞☎
=
✞☎
✝✆
✝✆
✝✆
✝✆
These identities can be written in the form of equations as
and
X x(1)x(2)(2)(1) ⊗ x(2)(1) ⊗ x(2)(2)(2) =X x(1)(1)(1)x(1)(2) ⊗ x(1)(1)(2) ⊗ x(2)
X x(1) ⊗ x(2)(2)(1) ⊗ x(2)(1)x(2)(2)(2) =X x(1)(1)(1) ⊗ x(1)(2) ⊗ x(1)(1)(2)x(2).
They arise in the study of the algebraic seven-dimensional sphere [25].
Surprisingly, U (M(α, β, γ)) turns out to be rigid in the sense that any deformation which satisfies Moufang-
Hopf and co-Moufang-Hopf identities is equivalent to a trivial deformation. Even universal enveloping
algebras of finite-dimensional central simple Lie algebras exhibit certain rigidity if considered as Moufang-
Hopf algebras since all their deformations are necessarily associative and coassociative; in other words, are
quantized enveloping algebras in the usual sense [29, 42].
The rigidity characteristic of non-associativity is still poorly understood and needs an adecuate cohomo-
logical interpretation.
Remark. There are a number of deformations of non-associative algebras that carry the adjective "quantum",
such as the quantum octonions of [4] and of [5]. They do not seem to fit in the framework of hypothetical
quantum loops that we discuss here.
26
Acknowledgments. The authors were supported by the following grants: FAPESP 2012/21938-4 and
CONACyT grant 168093-F (J.M.), FAPESP 2012/22537-3 (J.M.P.-I.), CNPq 3305344/2009-9 and FAPESP
2010/50347-9 (I.P.Sh.), Spanish Government project MTM 2010-18370-C04-03 (all three). The first two
authors would like to thank the Institute of Mathematics and Statistics of the University of Sao Paulo for
hospitality.
References
[1] M. Akivis, V. Goldberg, Algebraic aspects of web geometry. Commentat. Math. Univ. Carol. 41, 205–236 (2000)
[2] J. Beck, Triples, Algebras, and Cohomology, Ph.D. thesis, Columbia University, 1967. Repr. Theory Appl. Categ. 2, 1–59
(2003)
[3] G. Benkart, S. Madariaga, J.M. P´erez Izquierdo, Hopf algebras with triality. Trans. Am. Math. Soc. 365, 1001–1023
(2013)
[4] G. Benkart, J.M. P´erez Izquierdo, A quantum octonion algebra. Trans. Am. Math. Soc. 352, 935–968 (2000)
[5] M.R. Bremner, Quantum octonions. Commun. Algebra 27, 2809–2831 (1999)
[6] M.R. Bremner, I.R. Hentzel, L.A. Peresi, Dimension Formulas for the Free Nonassociative Algebra. Commun. Algebra
33, 4063–4081 (2005)
[7] M.R. Bremner, I.R. Hentzel, L.A. Peresi, H. Usefi, Universal enveloping algebras of the four-dimensional Malcev algebra.
Algebras, representations and applications, American Mathematical Society, Providence, RI. Contemp. Math. 483, 73–89
(2009).
[8] M.R. Bremner, S. Madariaga, Polynomial identities for tangent algebras of monoassociative loops, Commun. Algebra, to
appear.
[9] M.R. Bremner, H. Usefi, Enveloping algebras of the nilpotent Malcev algebra of dimension five. Algebr. Represent. Theory
13, 407-425 (2010)
[10] R.H. Bruck, A survey of binary systems. Springer Verlag, Berlin-Gottingen-Heidelberg (1958)
[11] E. Chibrikov, A right normed basis for free Lie algebras and Lyndon-Shirshov words, J. Algebra 302, 593–612 (2006)
[12] E. Chibrikov, On Free Sabinin Algebras. Commun. Algebra 39, 4014–4035 (2011)
[13] S. Doro, Simple Moufang loops. Math. Proc. Camb. Philos. Soc. 83, 377–392 (1978)
[14] L. Gerritzen, R. Holtkamp, Hopf co-addition for free magma algebras and the non-associative Hausdorff series. J. Algebra
265, 264–284 (2003)
[15] G. Glauberman, On loops of odd order II. J. Algebra 8, 393–414 (1968)
[16] A. Grishkov, A. Zavarnitsine, Groups with triality. J. Algebra Appl. 5, 441–463 (2006)
[17] M. Hartl, B. Loiseau, Internal object actions in homological categories. arXiv:1003.0096 [math.CT].
[18] M. Hartl, T. Van der Linden, The ternary commutator obstruction for internal crossed modules. Adv. Math. 232, 571–607
(2013)
[19] I. Hentzel, L. Peresi, Special identities for Bol algebras, Linear Algebra Appl. 436, 2315–2330 (2012)
[20] K.H. Hofmann, K. Strambach, Topological and analytic loops. In O. Chein, H. Pflugfelder and J.D.H Smith (eds.)
Quasigroups and Loops: Theory and Applications, pp. 205–262. Heldermann Verlag, Berlin (1990)
[21] N. Jacobson, Exceptional Lie Algebras. Marcel-Dekker, New York (1971)
[22] S.A. Jennings, The group ring of a class of infinite nilpotent groups, Can. J. Math. 7, 169–187 (1955)
[23] F.S. Kerdman, Analytic Moufang loops in the large. Algebra Logic 18, 325-347 (1980); translation from Algebra Logika
18, 523-555 (1979)
[24] M. Kikkawa, On local loops in affine manifolds. J. Sci. Hiroshima Univ. Ser. A-I Math. 28, 199–207 (1964)
[25] J. Klim, S. Majid, Hopf quasigroups and the algebraic 7-sphere. J. Algebra 323, 3067–3110 (2010)
[26] E. Kuzmin, On the relation between Mal'tsev algebras and analytic Mufang groups. Algebra Logika 10, 3–22 (1971)
[27] O. Loos, Symmetric Spaces I: General Theory. Benjamin, New York (1969).
[28] F. Lemieux, C. Moore and D. Th´erien, Subtree-counting loops. Quasigroups Relat. Syst. 8, 45–65 (2001)
[29] S. Madariaga, J.M. Perez Izquierdo, Non-existence of coassociative quantized universal enveloping algebras of the traceless
octonions. Commun. Algebra 40, 1009–1018 (2012)
[30] A.I. Malcev, Analytic loops (Russian). Mat. Sb. N.S. 36(78), 569–576 (1955)
[31] P.O. Mikheev, Enveloping groups of Moufang loops. Russ. Math. Surv. 48, 195-196 (1993); translation from Usp. Mat.
Nauk 48, 191-192 (1993).
[32] P.O Mikheev, On a problem of Chern-Akivis-Shelekhov on hexagonal three-webs. Aequationes Math. 51, 1–11 (1996)
[33] P.O. Miheev, L.V. Sabinin, Quasigroups and differential geometry. In O. Chein, H. Pflugfelder and J.D.H Smith (eds.)
Quasigroups and Loops: Theory and Applications, pp. 357–430. Heldermann Verlag, Berlin, 1990.
[34] J. Mostovoy, The Notion of Lower Central Series for Loops. In Sabinin, L., Sbitneva, L., Shestakov, I. (eds.) Non-associative
algebra and its applications. pp. 291–298. Chapman & Hall/CRC-Press, Boca Raton, FL (2006)
[35] J. Mostovoy, Nilpotency and Dimension Filtration for Loops. Commun. Algebra 36, 1565–1579 (2008)
[36] J. Mostovoy, J.M. P´erez Izquierdo, Dimension Filtration on Loops. Isr. J. Math. 158, (2007), 105–118.
[37] J. Mostovoy, J.M. P´erez Izquierdo, Ideals in non-associative universal enveloping algebras of Lie triple systems. Forum
Math. 22, 1–20 (2010)
[38] J. Mostovoy, J.M. P´erez Izquierdo, Formal multiplications, bialgebras of distributions and non-associative Lie theory.
Transform. Groups 15, 625–653 (2010).
27
[39] J. Mostovoy, J.M. P´erez Izquierdo, I.Shestakov. Nilpotent Sabinin algebras, preprint.
[40] P.T. Nagy, K. Strambach, Loops in group theory and Lie theory. De Gruyter Expositions in Mathematics, 35. Walter de
Gruyter & Co., Berlin (2002)
[41] J.M. P´erez Izquierdo, I.P. Shestakov, An envelope for Malcev algebras. J. Algebra 272, 379–393 (2004)
[42] J.M. P´erez Izquierdo, I.P. Shestakov, in preparation.
[43] J.M. P´erez Izquierdo, An envelope for Bol algebras. J. Algebra 284, 480–493 (2005)
[44] J.M. P´erez Izquierdo, Algebras, hyperalgebras, nonassociative bialgebras and loops, Adv. Math. 208, 834–876 (2007)
[45] J.M. P´erez Izquierdo, Right ideals in non-associative universal enveloping algebras of Lie triple systems. J. Lie Theory
18, 375–382 (2008)
[46] C. Reutenauer, Free Lie algebras. Clarendon Press, Oxford (1993).
[47] L.V. Sabinin, Odules as a new approach to a geometry with a connection. Sov. Math., Dokl. 18, 515–518 (1977); translation
from Dokl. Akad. Nauk SSSR 233, 800–803 (1977)
[48] L.V. Sabinin, P.O Mikheev, On the infinitesimal theory of local analytic loops. Sov. Math., Dokl. 36, 545–548 (1988);
translation from Dokl. Akad. Nauk SSSR 297, 801–804 (1987)
[49] L.V. Sabinin, Smooth quasigroups and loops: forty-five years of incredible growth. Commentat. Math. Univ. Carol. 41,
377–400 (2000)
[50] L.V. Sabinin, Smooth Quasigroups and Loops. Mathematics and Its Applications, Vol. 492, Kluwer Academic Publishers,
Dordrecht/Boston/London (1999)
[51] R.D. Schafer, An introduction to nonassociative algebras. Academic Press, New York-London (1966), Dover Publications,
Inc., New York (1995)
[52] A.M. Shelekhov, The G-structures associated to a hexagonal 3-web is closed. J. Geom. 35, 169–176 (1989)
[53] I.P. Shestakov, Every Akivis algebra is linear. Geom. Dedicata 77, 215–223 (1999)
[54] I.P. Shestakov, Moufang loops and alternative algebras. Proc. Am. Math. Soc. 132, 313–316 (2004)
[55] I.P. Shestakov, U.U. Umirbaev, Free Akivis algebras, primitive elements, and hyperalgebras. J. Algebra 250, 533–548
(2002)
[56] J.D.H. Smith, An Introduction to Quasigroups and Their Representations. Chapman & Hall/CRC, Boca Raton, FL
(2007)
[57] M.V. Tvalavadze, M.R. Bremner, Enveloping algebras of solvable Malcev algebras of dimension five. Commun. Algebra
39, 2816–2837 (2011)
28
|
1505.00466 | 1 | 1505 | 2015-05-03T19:32:52 | A Characterization of Modules with Cyclic Socle | [
"math.RA"
] | In 2009, J. Wood [15] proved that Frobenius bimodules have the extension property for symmetrized weight compositions. Later, in [9], it was proved that having a cyclic socle is sufficient for satisfying the property, while the necessity remained an open question. Here, landing in Midway, the necessity is proved, a module alphabet RA has the extension property for symmetrized weight compositions built on AutR(A) is necessarily having a cyclic socle. | math.RA | math |
A Characterization of Modules with Cyclic Socle
Ali Assem
Department of Mathematics
[email protected]
September 24, 2018
Abstract
In 2009, J. Wood [15] proved that Frobenius bimodules have the extension property for sym-
metrized weight compositions. Later, in [9], it was proved that having a cyclic socle is sufficient
for satisfying the property, while the necessity remained an open question.
Here, landing in Midway, the necessity is proved, a module alphabet RA has the extension property
for symmetrized weight compositions built on AutR(A) is necessarily having a cyclic socle.
Note: All rings are finite with unity, and all modules are finite too. This may be re-emphasized in
some statements. The convention for functions is that inputs are to the left.
1 Introduction
A (left) linear code of length n over a module alphabet RA is a (left) submodule C ⊂ An. A has
the extension property (EP) for the weight w if for any n and any two codes C1, C2 ⊂ An, any
isomorphism f : C1 → C2 preserving w extends to a monomial transformation of An.
In 1962,
MacWilliams [6] proved the Hamming weight EP for linear codes over finite fields; in 1996, H. Ward
and J. Wood [11] re-proved this using the linear independence of group characters. This kind of proofs
-- using characters -- led to further generalities. In 1997, J. Wood [12] proved that Frobenius rings have
the EP for symmetrized weight compositions (swc), and in his 1999-paper [13], proved that Frobenius
rings have the property for Hamming weight. Besides, for the last case, a partial converse was proved:
commutative rings satisfying the EP for Hamming weight are necessarily Frobenius.
In 2004, Greferath et al.[7] showed that Frobenius bimodules do have the EP for Hamming weight.
In [2], Dinh and López-Permouth suggested a strategy for proving the full converse. The strategy has
three parts. (1) If a finite ring is not Frobenius, its socle contains a matrix module of a particular
type. (2) Provide a counter-example to the EP in the context of linear codes over this special module.
(3) Show that this counter example over the matrix module pulls back to give a counter example
over the original ring. Finally, in 2008, J. Wood [14] provided the main technical result for carrying
out the strategy, and thereby proving that rings having the EP for Hamming weight are necessarily
Frobenius. The proof was easily adapted in [15] (2009) to prove that a module alphabet RA has the
EP for Hamming weight if and only if A is pseudo-injective with cyclic socle.
On the other lane, in [15], J. Wood proved that Frobenius bimodules have the EP for swc, and in
[9] it was shown that having a cyclic socle is sufficient (Theorem 3.4), while the necessity remained
an open question. Here, the necessity is proved, making use of a new notion, namely, the annihilator
weight, defined in section 4 below.
1
2 Background in Ring Theory
Let R be a finite ring with unity, denote by radR its Jacobson radical, by the Wedderburn-Artin
theorem (and Wedderburn's little theorem) the ring R/radR is semi-simple, and (as rings)
R/radR ∼=
kM
Mµi(Fqi),
i=1
(2.1)
where each qi is a prime power, Fqi denotes a finite field of order qi, and Mµi(Fqi) denotes the ring of
µi × µi matrices over Fqi.
It follows that, as left R-modules,
R(R/radR) ∼=
kM
i=1
µiTi,
(2.2)
where RTi is the pullback to R of the matrix module Mµi (Fqi )Mµi×1(Fqi) via the isomorphism in equa-
tion (2.1). It is known that these Ti's form the complete list, up to isomorphism, of all simple left
R-modules, hence the socle of any R-module A can be expressed as
soc(A) ∼=
kM
i=1
siTi,
where si is the number of copies of Ti inside A.
The next two propositions can be found in [15], page 17.
Proposition 2.1. soc(A) is cyclic if and only if si ≤ µi for i = 1, . . . , k; µi defined as above.
Proposition 2.2. soc(A) is cyclic if and only if A can be embedded into R bR, the character group of R
equipped with the standard module structure.
The next theorem (Theorem 4.1, [14]), by J. Wood, was the key to carry out the strategy of Dinh
and López-Permouth mentioned in the introduction, actually, it displays a thoughtfully constructed
piece-of-art example for the failure of the Hamming weight EP.
Theorem 2.3. Let R = Mm(Fq) and A = Mm×k(Fq). If k > m, there exist linear codes C+, C− ⊂
AN , N =
(1 + qi), and an R-linear isomorphism f : C+ → C− that preserves Hamming weight,
k−1Q
i=1
yet there is no monomial transformation extending f .
If soc(A) is not cyclic, then the previous theorem, applied to a certain submodule of soc(A), gives
counter-examples that pull back to give counter-examples for the original module, as the proof of the
following theorem shows (a detailed proof is found in [15], Theorem 6.4).
Theorem 2.4. (Th. 5.2, [14]). Let R be a finite ring, and let A be a finite left R-module. If there
exists an index i and a multiplicity k > µi so that kTi ⊂ soc(A) ⊂ A, then the extension property for
Hamming weight fails for linear codes over the module A.
2
3 Symmetrized Weight Compositions
Definition 3.1. (Symmetrized Weight Compositions) Let G be a subgroup of the automorphism group
AutR(A) of a finite R-module A. Define an equivalence relation ∼ on A: a ∼ b if a = bτ for some
τ ∈ G. Let A/G denote the orbit space of this relation. The symmetrized weight composition (swc)
built on G is a function
swc : An × A/G → Q defined by,
swc(x, a) = {i : xi ∼ a},
where x = (x1, . . . , xn) ∈ An and a ∈ A/G. Thus, swc counts the number of components in each
orbit.
Definition 3.2. (Monomial Transformation) Let G be a subgroup of AutR(A), a map T is called a
G-monomial transformation of An if there are some σ ∈ Sn and τi ∈ G for i = 1, . . . , n, such that
(x1, . . . , xn)T = (xσ(1)τ1, . . . , xσ(n)τn),
where (x1, . . . , xn) ∈ An.
Definition 3.3. (Extension Property) The alphabet A has the extension property (EP) with respect to
swc if for every n, and any two linear codes C1, C2 ⊂ An, any R-linear isomorphism f : C1 → C2
preserving swc is extends to a G-monomial transformation of An.
In [12], J.A.Wood proved that Frobenius rings do have the extension property with respect to swc.
Later, in [9], it was shown that, more generally, a left R-module A has the extension property with
respect to swc if it can be embedded in the character group bR (given the standard module structure).
If A can be embedded into bR (or
Theorem 3.4. (Th.4.1.3, [8]) Let A be a finite left R-module.
equivalently, soc(A) is cyclic), then A has the extension property with respect to the swc built on any
subgroup G of AutR(A). In particular, this theorem applies to Frobenius bimodules.
4 Annihilator Weight
We now define a new notion (the Midway!) on which we'll depend in the rest of this paper.
Definition 4.1. (Annihilator Weight) On RA, define an equivalence relation ≈ by a ≈ b if Anna =
Annb, where a and b are any two elements in A and Anna = {r ∈ Rra = 0} is the annihilator of a.
Clearly, Anna is a left ideal.
Now, on An we can define the annihilator weight aw that counts the number of components in each
orbit.
Remark: It is easily seen that the EP for Hamming weight implies the EP for swc, and the EP for
aw as well.
Lemma 4.2. Let RA be a pseudo-injective module. Then for any two elements a and b in A, a ≈ b if
and only if a ∼ b (∼ corresponds to the action of the whole group AutR(A)).
3
Proof. If a ∼ b, this means a = bτ for some τ ∈ AutR(A), and consequently Anna = Annb.
Conversely, if a ≈ b, then we have (as left R-modules)
Ra ∼= RR/Anna = RR/Annb
∼= Rb,
with ra 7→ r + Anna 7→ rb. By Proposition 5.1 in [15], since A is pseudo-injective, the isomorphism
Ra → Rb ⊆ A extends to an automorphism of A taking a to b.
Corollary 4.3. If RA is a pseudo-injective module, then the EP with respect to swc built on AutR(A)
is equivalent to the EP with respect to aw.
Theorem 4.4. Let R be a principal ideal ring, RA a pseudo-injective module, and let C be a submod-
ule of An for some n. Then a monomorphism f : C → An preserves Hamming weight if and only if it
preserves swc built on AutR(A).
Proof. The "if" part is direct. For the converse, we'll use that any left ideal I contains an element eI
that doesn't belong to any other left ideal not containing I. Now, if
(c1, c2, . . . , cn)f = (b1, b2, . . . , bn),
(4.1)
choose, from c1, c2, . . . , cn; b1, b2, . . . , bn, a component with a maximal annihilator I. Act on equation
(4.1) by eI, then the only zero places are those of the components in equation (4.1) with annihilator
I, and the preservation of Hamming weight gives the preservation of I-annihilated components. Omit
these components from the list c1, c2, . . . , cn; b1, b2, . . . , bn and choose one with the new maximal,
and repeat. This gives that f preserves aw and hence, by Lemma 4.2, f preserves swc built on
AutR(A).
Corollary 4.5. If RA is a module alphabet, then A has the extension property with respect to swc if
and only if soc(A) is cyclic.
Proof. The "if" part is answered by Theorem 3.4. Now, if soc(A) is not cyclic, then by Proposition
2.1, there is an index i such that si > µi, where siTi ⊂ soc(A) ⊂ A. Recall that Ti is the pullback
to R of the matrix module Mµi (Fqi )Mµi×1(Fqi), so that siTi is the pullback to R of the Mµi(Fqi)-
module B = Mµi×si(Fqi). Theorem 2.3 implies the existence of linear codes C+, C− ⊂ BN , and an
isomorphism f : C+ → C− that preserves Hamming weight, yet f does not extend to a monomial
transformation of BN . But the ring Mµi(Fqi) is a principal ideal ring (in fact, more is true, Theorem
ix.3.7, [10]), besides, B is injective, and then Theorem 4.4 implies that f preserves swc built on
AutMµi (Fqi )(B).
Now, a little notice finishes the work. The isomorphism in equation (2.1) and the projection map-
pings R → R/radR → Mµi(Fqi) allow us to consider the whole situation for C± as R-modules. Since
B pulls back to siTi, we have C± ⊂ (siTi)N ⊂ soc(A)N ⊂ AN, as R-modules. Thus C± are linear
codes over A that are isomorphic through an isomorphism preserving swc built on AutR(siTi). Also,
any automorphism of A restricts to an automorphism of siTi, hence the isomorphism preserves swc
built on AutR(A). However, this isomorphism does not extend to a monomial transformation of AN ,
since, as appears in the proof of Theorem 2.3 (found in [14]), C+ has an identically zero component,
while C− does not.
4
References
[1] H. Q. Dinh and S. R. López-Permouth, On the equivalence of codes over finite rings, Appl. Alge-
bra Engrg. Comm. Comput. 15 (2004), no. 1, 37-50. MR MR2142429 (2006d:94097)
[2] H. Q. Dinh and S. R. López-Permouth, On the equivalence of codes over rings and modules, Finite
Fields Appl. 10 (2004), no. 4, 615-625. MR MR2094161 (2005g:94098).
[3] T. Honold, Characterization of finite Frobenius rings, Arch. Math. (Basel) 76 (2001), no. 6, 406-
415. MR MR1831096 (2002b:16033)
[4] T. Y. Lam, Lectures on modules and rings, Graduate Texts in Mathematics, vol. 189, Springer-
Verlag, New York, 1999. MR MR1653294 (99i:16001)
[5] T. Y. Lam, A First course in noncommutative rings, second ed., Graduate Texts in Mathematics,
vol. 131, Springer-Verlag, New York, 2001. MR MR1838439 (2002c:16001)
[6] F. J. MacWilliams, Combinatorial properties of elementary abelian groups, Ph.D. thesis, Radcliffe
College, Cambridge, Mass., 1962.
[7] M. Greferath, A. Nechaev, and R. Wisbauer, Finite quasi-Frobenius modules and linear codes, J.
Algebra Appl. 3 (2004), no. 3, 247-272. MR MR2096449 (2005g:94099)
[8] N .ElGarem, Codes over Finite Modules, Master Thesis - Cairo University, 2014.
[9] N .ElGarem, N. Megahed, and J.A. Wood, The extension Theorem with respect to Symmetrized
Weight Compositions, 4th international castle meeting on coding theory and applications, 2014.
[10] Thomas W.Hungerford, ALGEBRA, Berlin: Springer-Verlag, 1980.
[11] H. N.Ward and J. A.Wood, Characters and the equivalence of codes, J. Combin. Theory Ser. A
73 (1996), no. 2, 348-352. MR MR1370137 (96i:94028)
[12] J. A. Wood, Extension theorems for linear codes over finite rings, Applied algebra, algebraic
algorithms and error-correcting codes (Toulouse, 1997) (T. Mora and H. Mattson, eds.), Lec-
ture Notes in Comput. Sci., vol. 1255, Springer, Berlin, 1997, pp. 329-340. MR MR1634126
(99h:94062)
[13] J. A. Wood, Duality for modules over finite rings and applications to coding theory, Amer. J.
Math. 121 (1999), no.3, 555-575. MR MR1738408 (2001d:94033)
[14] J. A. Wood, Code equivalence characterizes finite Frobenius rings, Proc. Amer. Math. Soc. 136
(2008), 699-706.
[15] J. A. Wood, Foundations of Linear Codes Defined over Finite Modules: The Extension Theo-
rem and MacWilliams Identities, Codes over Rings, Proceedings of the CIMPA Summer School,
Ankara, Turkey, 18-29 August 2008, (Patrick Solé) Series on Coding Theory and Cryptology, Vol.
6, World Scientific, Singapore, 2009, p. 124-190.
5
|
1808.02875 | 3 | 1808 | 2019-04-15T14:49:28 | Polynomial Equations over Octonion Algebras | [
"math.RA"
] | In this paper we present a complete method for finding the roots of all polynomials of the form $\phi(z)=c_n z^n+c_{n-1} z^{n-1}+\dots+c_1 z+c_0$ over a given octonion division algebra. When $\phi(z)$ is monic we also consider the companion matrix and its left and right eigenvalues and study their relations to the roots of $\phi(z)$, showing that the right eigenvalues form the conjugacy classes of the roots of $\phi(z)$ and the left eigenvalues form a larger set than the roots of $\phi(z)$. | math.RA | math | Polynomial Equations over Octonion Algebras
Department of Computer Science, Tel-Hai Academic College, Upper Galilee, 12208 Israel
Adam Chapman
9
1
0
2
r
p
A
5
1
]
.
A
R
h
t
a
m
[
3
v
5
7
8
2
0
.
8
0
8
1
:
v
i
X
r
a
Abstract
In this paper we present a complete method for finding the roots of all polynomials
of the form φ(z) = cnzn + cn−1zn−1 + · · · + c1z + c0 over a given octonion division
algebra. When φ(z) is monic we also consider the companion matrix and its left
and right eigenvalues and study their relations to the roots of φ(z), showing that
the right eigenvalues form the conjugacy classes of the roots of φ(z) and the left
eigenvalues form a larger set than the roots of φ(z).
Keywords: Polynomial Equations, Division Algebras, Octonion Algebras,
Companion Matrix, Right Eigenvalues, Left Eigenvalues
2010 MSC: primary 17D05; secondary 15A18, 15B33
1. Introduction
The question of finding the roots of a given (monic) standard polynomial φ(z) =
zn + cn−1zn−1 + · · · + c1z + c0 over any quaternion division algebra Q with center F
was fully solved in [3]: the polynomial has an assigned "companion polynomial"
Φ(z) whose degree is 2n and its coefficients live in F, which is also the companion
polynomial of the embedding of the companion matrix Cφ of φ(z) into M2n(K)
where K is an arbitrary maximal subfield of Q. The left eigenvalues of Cφ coincide
with the roots of φ(z), and the right eigenvalues of Cφ coincide with the roots of
Φ(z). The roots of Φ(z) group into (up to n) complete conjugacy classes. For
each such conjugacy class, either the entire class consists of roots of φ(z), or it
contains exactly one root of φ(z). Earlier papers on this subject include [4] (solving
equations over the real quaternion algebra), [1] and [2] (solving quadratic equations
over arbitrary quaternion algebras), and [8] (solving monic quadratic equations
over the real octonion algebra).
The aim of this paper is to extend these results to octonion division algebras. A
part of the motivation comes from recent results in physics that translate physical
Email address: [email protected] (Adam Chapman)
Preprint submitted to Journal of Algebra and its Applications
April 16, 2019
problems to equations over octonions and more general Cayley-Dickson algebras,
see for example [6]. We consider standard polynomials over an octonion division
algebra A. These are polynomials with coefficients appearing only on the left-hand
side of the variable: φ(z) = cnzn + cn−1zn−1 + · · · + c1z + c0 where ci ∈ A. For any
λ ∈ A the substitution of λ in φ(z) is defined to be cnλn + cn−1λn−1 + · · · + c1λ + c0
and is denoted by φ(λ). By a root of a standard polynomial φ(z) over A we mean
an element λ ∈ A satisfying φ(λ) = 0. We denote by R(φ) the set of roots of
φ(z). We define the companion polynomial Φ(z) and the companion matrix in the
same manner as in the quaternionic case. We show that the roots of Φ(z) are the
conjugacy classes of the roots of φ(z). We prove that for each class in R(Φ) either
the entire class is in R(φ) or it contains a unique element form R(φ). We prove that
the right eigenvalues of Cφ are exactly the roots of Φ(z) and also describe its left
eigenvalues. Unlike the quaternionic case, the left eigenvalues of Cφ turn out to be
a larger set than the roots of φ(z) and we provide an example of this phenomenon.
Note that our definition of a standard polynomial places the coefficients on
the left-hand side of the variable, but clearly the same methods can be applied to
solving polynomials with coefficients appearing on the right-hand side. Given a
standard polynomial φ(z) = cnzn + · · · + c1z + c0 over an octonion algebra A, we
define its "mirror" polynomial to be φ(z) = zncn + · · · + zc1 + c0.
2. Octonion Algebras
Given a field F, a quaternion algebra Q over F is a central simple F-algebra of
degree 2 (i.e., dimension 4 over F). When char(F) , 2, it has the structure
Q = Fhi, j : i2 = α, j2 = β, i j = − jii
for some α, β ∈ Q×, and when char(F) = 2, it has the structure
Q = Fhi, j : i2 + i = α, j2 = β, ji + i j = ji
for some α ∈ F and β ∈ F×. The algebra Q is endowed with a symplectic involution
mapping a+bi+c j+di j to a−bi−c j−di j when char(F) , 2 and to a+b+bi+c j+di j
when char(F) = 2.
An Octonion algebra over F is an algebra A of the form A = Q⊕Qℓ where Q is a
quaternion algebra over F, and the multiplication table is given by (p+qℓ)·(r+sℓ) =
pr +ℓ2 ¯sq+(sp+q¯r)ℓ where ¯ stands for the symplectic involution on Q and ℓ2 ∈ F×.
This involution extends to A by the formula p + qℓ = p − qℓ. The octonion algebra
is endowed with a quadratic norm form Norm : A → F defined by Norm(x) = x · x
and a linear trace form Tr : A → F defined by Tr(x) = x + x. Every two elements
in A live inside a quaternion subalgebra, unless char(F) = 2, in which case the two
2
elements can also live inside a purely inseparable bi-quadratic field extension of F
inside A, for example the elements j and ℓ in the construction above. In particular,
the algebra is alternative. The algebra is a division algebra if and only if its norm
form is anisotropic. For further reading on octonion algebras see [7] and [5].
3. The Companion Polynomial
The goal of this section is to give a deterministic algorithm for finding all the
roots of a given standard polynomial over an octonion division algebra.
Remark 3.1. The relation g ∼ g′ ⇔ ∃h ∈ A× : hgh−1 = g′ is an equivalence
relation for elements of a given octonion algebra A.
Proof. It is enough to show that g ∼ g′ if and only if Tr(g) = Tr(g′) and Norm(g) =
Norm(g′). Write T = Tr(g) and N = Norm(g). Both live inside F. Then g2 − T g +
N = 0. Since the octonion algebra is alternative and T, N ∈ F, we can conjugate
this equation by h and obtain (hgh−1)2 − T (hgh−1) + N = 0, which means that the
trace and norm of hgh−1 are T and N, resp. In the opposite direction, suppose g and
g′ have the same trace and norm. If they live inside a quaternion subalgebra then
they are conjugates in that subalgebra, and if the live inside a purely inseparable
bi-quadratic field extension of F then they must be equal.
(cid:3)
Definition 3.2. We define the "companion polynomial" Φ(z) of a given polynomial
φ(z) = cnzn + · · · + c1z + c0 over an octonion algebra A to be
Φ(z) = b2nz2n + · · · + b1z + b0
with the coefficients defined in the following way: for each k ∈ {0, . . . , 2n}, if k is
odd then bk is the sum of all Tr( ¯cic j) with 0 ≤ i < j ≤ n and i + j = k, and if k = 2m
is even then bk is the sum of all Tr(cic j) with 0 ≤ i < j ≤ n and i + j = k plus the
element Norm(cm). (Recall that Tr( ¯cic j) = ¯cic j + ¯c jci and Norm(cm) = cmcm.)
Theorem 3.3. Let φ(z) = cn(zn) + · · · + c1z + c0 be a standard polynomial over an
octonion division algebra A over a field F with companion polynomial Φ(z). Then
R(Φ) ⊇ R(φ).
Proof. By the [7, Lemma 1.3.3], for every z ∈ A and i, j ∈ {0, . . . , n}, Norm(ci)z2i =
ci(ciz2i) and Tr(cic j)zi+ j = ci(c jzi+ j) + c j(cizi+ j). Therefore
Φ(z) =
n
X
i=0
n
ci(
X
j=0
c j(zi+ j)) =
n
X
i=0
ci(φ(z)zi).
Consequently, if φ(λ) = 0 for a certain λ ∈ A, then also Φ(λ) = 0.
(cid:3)
3
All the coefficients of Φ(z) are central, i.e. belong to F (because they are sums
of traces and norms of elements in A). Therefore, the roots of Φ(z) depend only on
their norm and trace, i.e. the set R(Φ) is a union of conjugacy classes.
Theorem 3.4. Given companion polynomial Φ(z) of φ(z), the set R(Φ) is the union
of the conjugacy classes of the elements of R(φ). Each such class is either fully
contained in R(φ) or has exactly one representative there.
Proof. Every z ∈ A with Tr(z) = T and Norm(z) = N satisfies z2 − T z +
N = 0. Therefore, by plugging in z2 = T z − N in φ(z), we obtain φ(z) =
E(N, T )z + G(N, T ) for some polynomials E(N, T ) and G(N, T ) in the central
variables N and T . Write ei(N, T ) and gi(N, T ) for the polynomials satisfying
zi = ei(N, T )z + gi(N, T ). Note that ei(N, T ) and gi(N, T ) are central for any
N, T ∈ F, and therefore we can treat them as central elements in the computations.
Then E(N, T ) = Pn
i=0 ci(φ(z)zi) =
i=0 ci((E(N, T )z + G(N, T ))(ei(N, T )z + gi(N, T ))) = Pn
Pn
i=0 ei(N, T )ci(E(N, T )z2) +
Pn
i=0 ei(N, T )ci(G(N, T )z) + Pn
i=0 gi(N, T )ciG(N, T ) =
E(N, T )E(N, T )z2 + E(N, T )G(N, T )z + G(N, T )E(N, T )z + G(N, T )G(N, T ) =
Norm(E(N, T ))z2 + Tr(E(N, T )G(N, T ))z + Norm(G(N, T )).
i=0 gi(N, T )ci(E(N, T )z) + Pn
i=0 ciei and G(N, T ) = Pn
i=0 cigi. Now, Φ(z) = Pn
Consider an element z0 ∈ R(Φ) with norm N0 and trace T0. If E(N0, T0) = 0
then we have 0 = Φ(z0) = Norm(G(N0, T0)), and so G(N0, T0) = 0. This means
that the equality E(N0, T0)λ + G(N0, T0) = 0 holds for all λ ∈ [z0], and hence the
equality φ(λ) = 0 holds for all λ in the conjugacy class of z0.
Suppose E(N0, T0) , 0. Since an element λ of trace T0 and norm N0 satisfies
φ(λ) = E(N0, T0)λ + G(N0, T0), it is a root of φ(z) if and only if it is the unique
solution to the equation E(N0, T0)λ + G(N0, T0) = 0. What is left then in order to
prove that R(φ) ∩ [z0] = {λ} is to show that the unique solution λ to the equation
E(N0, T0)λ + G(N0, T0) = 0 has indeed trace T0 and norm N0, and indeed, this
λ satisfies Norm(E(N0, T0))λ2 + Tr(E(N0, T0)G(N0, T0))λ + Norm(G(N0, T0)) =
E(N0, T0)(E(N0, T0)λ+G(N0, T0))λ+G(N0, T0)(E(N0, T0)λ+G(N0, T0)) = 0, which
means that λ satisfies the same quadratic characteristic equation over F as z0, and
so λ is in the same conjugacy class as z0, i.e. has norm N0 and trace T0.
(cid:3)
The previous two theorems give a complete algorithm for finding all the roots
of an octonion polnyomial:
Algorithm 3.5. One needs first to solve the equation Φ(z) over the algebraic clo-
sure of F. Each root z0 lives in a field extension K of F. For z0 to be in the same
conjugacy class as an element of R(φ), K must be F-isomorphic to a subfield of A,
and therefore [K : F] is either 2 or 1. If it is, then the conjugacy class of z0 in A is
in R(Φ), and then either E(N0, T0) = 0 and then the entire class of [z0] is in R(φ),
4
or −E(N0, T0)−1G(N0, T0) is the unique representative of [z0] in R(φ) where N0 and
T0 are the norm and trace of z0.
Example 3.6. Consider the real octonion algebra A = O with generators i, j, ℓ, and
the polynomial φ(z) = iz2 + jz + ℓ. The companion polynomial is Φ(z) = z4 + z2 + 1,
and it has roots in the conjugacy classes of {z ∈ A : Norm(z) = 1, Tr(z) = 1}
and {z ∈ A : Norm(z) = 1, Tr(z) = −1}. For Norm(z) = 1, Tr(z) = 1, we have
z2 = z − 1, and so the equation φ(z) = 0 reduces to (i + j)z + ℓ − i = 0, which
means that (i + j)−1(i − ℓ) = 1
2 (1 + i j + iℓ + jℓ) is the unique representative of
its conjugacy class in R(φ). For Norm(z) = 1, Tr(z) = −1, we have z2 = −z − 1,
and so the equation φ(z) = 0 reduces to (−i + j)z + ℓ − i = 0, which means that
(−i + j)−1(i − ℓ) = 1
2 (−1 + i j − iℓ + jℓ) is the unique representative of its conjugacy
class in R(φ).
4. The Companion Matrix and its Left Eigenvalues
Suppose A is an octonion division algebra. Let φ(z) = zn + cn−1zn−1 + · · · + c0
be a monic standard polynomial with coefficients c0, . . . , cn−1 in A. We want to
associate the roots of φ(z) with left and right eigenvalues of the companion matrix,
given by
Cφ =
1
0
0
0
...
0
0
−c0 −c1
. . .
0
1
0
0
. . .
0
. . .
. . . −cn−2 −cn−1 .
1
We define the γ-twist of φ(z) to be the polynomial
φγ(z) = γ−1zn + (γ−1cn−1)zn−1 + · · · + (γ−1c1)z + γ−1c0.
A left (or right) eigenvalue of Cφ is an element λ ∈ A which satisfies Cφv = λv
(Cφv = vλ) for some nonzero column vector v of length n with entries in A. Write
LEV(Cφ) and REV(Cφ) for the sets of left and right eigenvalues of Cφ.
Theorem 4.1. For any standard polynomial φ(z) over A, LEV(Cφ) = Sγ∈A× R(φγ).
Proof. The element λ ∈ A is a left eigenvalue of Cφ if and only if there exists a
nonzero vector
∈ An
v =
v1
...
vn
5
satisfying Cφv = λv. This equality is equivalent to the system
v2 = λv1
...
vn = λvn−1
−c0v1 − · · · − cn−1vn = λvn.
Note that since v , 0, v1 , 0. The first n − 1 equations mean that v is
and the last equation then becomes
c0v1 + c1(λv1) + · · · + cn−1(λn−1v1) + λnv1 = 0.
1
λ
...
λn−1
v1
Write γ = v1. Note that if γ = 0 then v is the zero vector, so we have γ , 0. For
each i ∈ {0, . . . , n − 1}, write c′
i
= γ−1ci, and so the equation becomes
γc′
0γ + (γc′
1)(λγ) + · · · + (γc′
n−1)(λn−1γ) + λnγ = 0.
By the Moufang identity (xy)(zx) = x(yz)x, the former equation becomes
γ(c′
0
+ c′
1λ + · · · + c′
n−1λn−1 + γ−1λn)γ = 0.
Therefore, λ is a root of the twisted polynomial φγ(z). In the opposite direction, it
is clear from the same computation that a root λ of φγ(z) is in LEV(Cφ).
(cid:3)
Theorem 4.2. Let φ(z) be a standard monic polynomial over an octonion algebra
A over a field F, and let E(N, T ) and G(N, T ) be as in Theorem 3.4. Then
1. R(φ) ⊆ LEV(Cφ) ⊆ R(Φ).
2. For every conjugacy class [z0] ∈ R(Φ) of norm N0 and trace T0, if E(N0, T0) =
0 then [z0] ⊆ LEV(Cφ). Otherwise,
[z0] ∩ LEV(Cφ) = {−(E(N0, T0)−1γ)(γ−1G(N0, T0)) : γ ∈ A×}.
Proof. The inclusion R(φ) ⊆ LEV(Cφ) is obvious. For LEV(Cφ) ⊆ R(Φ), it is
enough to notice that the twists φγ(z) have the same companion polynomial as φ(z)
up to division by the norm of γ (using the multiplicativity of the norm form and [7,
Equations (1.3) & (1.4), Section 1.2]).
For each γ, the polynomial φγ(z) satisfies φγ(z) = (γ−1E(N, T ))z +(γ−1G(N, T ))
by a straight-forward computation. Consider a given class [z0] in Φ(z) of norm N0
and trace T0.
If E(N0, T0) = 0 then [z0] ⊆ R(φ), and hence [z0] ⊆ LEV(Cφ).
Suppose E(N0, T0) , 0. Then the unique element of R(φγ) ∩ [z0] is
−(γ−1E(N0, T0))−1(γ−1G(N0, T0)) = −(E(N0, T0)−1γ)(γ−1G(N0, T0)).
(cid:3)
6
Note that unlike the case of quaternion algebras, there is no inclusion LEV(Cφ) ⊆
R(φ), not even in the case of quadratic polynomials.
Example 4.3. Consider the polynomial φ(z) = z2 + iz + 1 + i j. The element λ = j
is not a root of this polynomial. However, λ = j is a root of the twisted polynomial
φℓ(z), and so it belongs to LEV(Cφ).
Note that in this example, j belongs to the quaternion subalgebra generated by
the coefficients, which means that even in the case where all the coefficients belong
to the same quaternion subalgebra Q, there is no guarantee that LEV(Cφ) ∩ Q =
R(φ) ∩ Q. The following proposition describes the set LEV(Cφ) ∩ Q in such cases:
Proposition 4.4. Given a standard monic polynomial φ(z) = zn + cn−1zn−1 + · · · +
c1z + c0 over a division octonion algebra A whose coefficients belong to a quater-
nion subalgebra Q of A, we have LEV(Cφ) ∩ Q = (R(φ) ∪ R( φ)) ∩ Q.
Proof. Every left eigenvalue λ of Cφ satisfies
c0γ + c1(λγ) + · · · + cn−1(λn−1γ) + λnγ = 0
for some γ ∈ A×. Suppose all the coefficients belong to a quaternion subalgebra Q,
and suppose λ ∈ Q as well. Then A decomposes as A = Q ⊕ Qℓ. The element γ
decomposes accordingly as γ = γ0 + γ1ℓ. By a straight-forward computation, we
obtain c0γ + c1(λγ) + · · · + cn−1(λn−1γ) + λnγ = (c0 + c1λ + · · · + cn−1λn−1)γ0 +
(γ1(c0 + λc1 + · · · + λn−1cn−1 + λn))ℓ. Therefore,
(γ0 = 0∨c0 + c1λ + · · · + cn−1λn−1 = 0) ∧(γ1 = 0∨c0 + λc1 + · · · + λn−1cn−1 + λn = 0).
Consequently, LEV(Cφ) ∩ Q ⊆ (R(φ) ∪ R( φ)) ∩ Q. The inclusion in the opposite
direction is proven using the same computation.
(cid:3)
5. Right Eigenvalues of the Companion Matrix
Given a polynomial φ(z) = zn + cn−1zn−1 + · · · + c1z + c0, let φγ(z) denote the
two-sided twisted polynomial φγ(z) = γ−2zn +(γ−1cn−1γ−1)zn−1 +· · ·+(γ−1c1γ−1)z+
γ−1c0γ−1.
Theorem 5.1. The set REV(Cφ) is the union of R(φγ) for all γ ∈ A×.
Proof. The element λ ∈ A is a right eigenvalue of Cφ if and only if there exists a
nonzero vector
∈ An
v =
v1
...
vn
7
satisfying Cφv = vλ. This equality is equivalent to the system
v2 = v1λ
...
vn = vn−1λ
−c0v1 − · · · − cn−1vn = vnλ.
Note that since v , 0, v1 , 0. The first n − 1 equations mean that v is v1
and the last equation then becomes
c0v1 + c1(v1λ) + · · · + cn−1(v1λn−1) + v1λn = 0.
1
λ
...
λn−1
Note that if v1 = 0 then v is the zero vector, so we have v1 , 0. Write γ =
v−1
1 . Multiply the equation from the right by γ−1 and use the Moufang identity
x(y(xz)) = (xyx)z to get
γ−1c0γ−1 + (γ−1c1γ−1)λ + · · · + (γ−1cn−1γ−1)λn−1 + γ−2λn = 0.
Therefore, λ is a root of the twisted polynomial φγ(z). In the opposite direction, it
is clear from the same computation that a root λ of φγ(z) is in REV(Cφ).
(cid:3)
Theorem 5.2. Let φ(z) be a standard monic polynomial over an octonion algebra
A over a field F, and let E(N, T ) and G(N, T ) be as in Theorem 3.4. Then
1. R(φ) ⊆ REV(Cφ) ⊆ R(Φ).
2. For every conjugacy class [z0] ∈ R(Φ(z)) of norm N0 and trace T0, if
E(N0, T0) = 0 then [z0] ⊆ REV(Cφ). Otherwise, [z0] ∩ REV(Cφ) =
{−γ(cid:0)E(N0, T0)−1(G(N0, T0)γ−1)(cid:1) : γ ∈ A×}.
Proof. The inclusion R(φ) ⊆ REV(Cφ) is obvious. For REV(Cφ) ⊆ R(Φ), it is
enough to notice that the twists φγ(z) have the same companion polynomial as φ(z)
up to division by the norm of γ2 (using the multiplicativity of the norm form and
[7, Equations (1.3) & (1.4), Section 1.2]).
For each γ,
the polynomial φγ(z) satisfies φγ(z) = (γ−1E(N, T )γ−1)z +
(γ−1G(N, T )γ−1) by a straight-forward computation. Consider a given class [z0]
in R(Φ) of norm N0 and trace T0.
If E(N0, T0) = 0 then [z0] ⊆ R(φ), and
hence [z0] ⊆ REV(Cφ). Suppose E(N0, T0) , 0. Then the unique element
of R(φγ) ∩ [z0] is −(γ−1E(N0, T0)γ−1)−1(γ−1G(N0, T0)γ−1). By the Moufang
identity (xyx)z = x(y(xz)) we obtain −(γ−1E(N0, T0)γ−1)−1(γ−1G(N0, T0)γ−1) =
−γ(cid:0)E(N0, T0)−1(G(N0, T0)γ−1)).
(cid:3)
8
Remark 5.3. Given an octonion algebra A over a field F, if g, h ∈ A are conjugates
(i.e., have the same trace and norm), then g = δhδ−1 for some δ ∈ A× of Tr(δ) = 0.
Proof. If g , h, take δ = g − h. If g = h, take δ to be any element in 1⊥ ∩ g⊥.
(cid:3)
Theorem 5.4. Let A be an octonion division algebra over a field F, and let e and
g be nonzero elements in A. Then {γ(e(gγ−1)) : γ ∈ A×} = {δ(eg)δ−1 : δ ∈ A×}.
Proof. The left-to-right inclusion follows from the fact that every element of the
form γ(e(gγ−1)) has the same trace and norm as eg: The norm is multiplicative, so
it follows immediately that Norm(γ(e(gγ−1))) = Norm(eg). The trace of γ(e(gγ−1))
is h1, γ(e(gγ−1))i where h , i is the polarization of the norm form. Now,
h1, γ(e(gγ−1))i = hγ, e(gγ−1)i = heγ, gγ−1i = h(eγ)γ−1, gi = he, gi = h1, egi = Tr(eg).
The computation makes use of the well-known identity hxy, zi = hy, xzi = hx, zyi
that can be found in [7, Lemma 1.3.2].
For the opposite inclusion, we note that eg and ge are conjugates, so every
element in the right set can be written as δ(ge)δ−1 for some δ ∈ A× of Tr(δ) = 0.
Set γ = δg, and then δ−1 = gγ−1. Now, δ−1 is a scalar multiple of δ (because
Tr(δ) = 0), and by the Moufang identity (xyx)z = x(y(xz)) we obtain
δ(ge)δ−1 = (δg)(eδ−1) = γ(e(gγ−1)).
(cid:3)
Corollary 5.5. Given a standard monic polynomial φ(z) over an octonion divi-
sion algebra A over a field F with companion polynomial Φ(z), we have R(Φ) =
REV(Cφ).
Proof. By Theorem 5.2, for each conjugacy class [z0] in R(Φ), either the en-
tire conjugacy class is in R(φ) (which happens when E(N0, T0) = 0) and
then it is also in REV(Cφ), or the intersection with REV(Cφ) is of the form
{−γ(cid:0)E(N0, T0)−1(G(N0, T0)γ−1)(cid:1) : γ ∈ A×} (which happens when E(N0, T0) , 0).
By Theorem 5.4, this set is the conjugacy class of −E(N0, T0)−1G(N0, T0), which
is [z0].
(cid:3)
Acknowledgements
The author is indebted to Seidon Alsaody for many illuminating discussions
concerning this paper and other problems related to octonion algebras. The author
was visiting Perimeter Institute for Theoretical Physics in the Summer of 2018,
during which a major part of this project was carried out.
9
References
[1] M. Abrate. Quadratic formulas for generalized quaternions. J. Algebra Appl.,
8(3):289 -- 306, 2009.
[2] A. Chapman. Quaternion quadratic equations in characteristic 2. J. Algebra
Appl., 14(3):1550033, 8, 2015.
[3] A. Chapman and C. Machen. Standard polynomial equations over division
algebras. Adv. Appl. Clifford Algebr., 27(2):1065 -- 1072, 2017.
[4] D. Janovsk´a and G. Opfer. A note on the computation of all zeros of simple
quaternionic polynomials. SIAM J. Numer. Anal., 48(1):244 -- 256, 2010.
[5] M.-A. Knus, A. Merkurjev, M. Rost, and J.-P. Tignol. The book of involu-
tions, volume 44 of American Mathematical Society Colloquium Publications.
American Mathematical Society, Providence, RI, 1998. With a preface in
French by J. Tits.
[6] V. L. Mironov and S. V. Mironov. Sedeonic equations of ideal fluid. J. Math.
Phys., 58(8):083101, 12, 2017.
[7] T. A. Springer and F. D. Veldkamp. Octonions, Jordan algebras and excep-
tional groups. Springer Monographs in Mathematics. Springer-Verlag, Berlin,
2000.
[8] Q.-W. Wang, X. Zhang, and Y. Zhang. Algorithms for finding the roots of
some quadratic octonion equations. Comm. Algebra, 42(8):3267 -- 3282, 2014.
10
|
1602.06094 | 1 | 1602 | 2016-02-19T09:56:54 | Elementary matrix reduction over Bezout duo rings | [
"math.RA"
] | A ring $R$ is an elementary divisor ring if every matrix over $R$ admits a diagonal reduction. We further explore various stable like conditions on a bezout duo-domain under which it is an elementary divisor domain. Many known results are thereby generalized to much wider class of rings. | math.RA | math |
Elementary Matrix Reduction Over B´ezout Duo-domains
Huanyin Chen
Department of Mathematics, Hangzhou Normal University
Hangzhou 310036, China
[email protected]
Marjan Sheibani
Faculty of Mathematics, Statistics and Computer Science
Semnan University, Semnan, Iran
[email protected]
Abstract: A ring R is an elementary divisor ring if every matrix over R admits
a diagonal reduction. If R is an elementary divisor domain, we prove that R is a
B´ezout duo-domain if and only if for any a ∈ R, RaR = R =⇒ ∃ s, t ∈ R such
that sat = 1. We further explore various stable-like conditions on a B´ezout
duo-domain under which it is an elementary divisor domain. Many known
results are thereby generalized to much wider class of rings, e.g. [3, Theorem
3.4.], [5, Theorem 14], [9, Theorem 3.7], [13, Theorem 4.7.1] and [14, Theorem
3].
Keywords: Elementary divisor ring; B´ezout duo-domain; Quasi-duo ring; Lo-
cally stable domain; Gelfand range 1.
MR(2010) Subject Classification: 15A21, 16S99, 16L99, 19B10.
1. Introduction
Throughout this paper, all rings are associative with an identity. A matrix A (not necessarily
square) over a ring R admits diagonal reduction if there exist invertible matrices P and Q
such that P AQ is a diagonal matrix (dij ), for which dii is a full divisor of d(i+1)(i+1) (i.e.,
Rd(i+1)(i+1)R ⊆ diiRT Rdii) for each i. A ring R is called an elementary divisor ring
provided that every matrix over R admits a diagonal reduction. A ring R is B´ezout ring if
every finitely generated right (left) ideal is principal. Clearly, every elementary divisor ring
is a B´ezout ring. An attractive problem is to investigate various conditions under which a
B´ezout ring is an elementary divisor ring.
Commutative elementary divisor domains have been studies by many authors, e.g. [3],[5],
[9-10], and [13-14]. But the structure of such rings in the noncommutative case has not been
2
Huanyin Chen and Marjan Sheibani
sufficiently studied (cf.
[6] and [15]). A ring R is duo if every right (left) ideal of R is a
two-sided ideal. Obviously, every commutative ring is duo, but the converse is not true. In
this paper, we are concern on when a B´ezout duo-domain is an elementary divisor domain.
Here, a ring R is a domain ring if there is no any nonzero zero divisor.
In Section 2, We prove that an elementary divisor domain is a B´ezout duo-domain if and
only if for any a ∈ R, RaR = R =⇒ ∃ s, t ∈ R such that sat = 1. A ring R has stable range
1 if, aR+ bR = R with a, b ∈ R =⇒ ∃y ∈ R such that a+ by ∈ R is invertible. This condition
plays an important rule in algebra K-theory. We refer the reader to [1] for the general theory
of stable range 1. Further, we shall introduce certain stable-like conditions on B´ezout duo-
domains so that they are elementary divisor domains. In Section 3, we prove that every
locally stable B´ezout duo-domain and every Zabavsky duo-domainis are elementary divisor
domains (see Theorem 3.2 and Theorem 3.9). In Section 4, we prove that every B´ezout duo-
domain of Gelfand range 1 is an elementary divisor ring (see Theorem 4.3). Finally, in the
last section, we establish related theorems on B´ezout duo-domains by means of countability
conditions on their maximal ideals. Many known results are thereby generalized to much
wider class of rings, e.g. [3, Theorem 3.4.], [5, Theorem 14], [9, Theorem 3.7], [13, Theorem
4.7.1] and [14, Theorem 3].
We shall use J(R) and U (R) to denote the Jacobson radical of R and the set of all units
in R, respectively. For any a, b ∈ R, ab means that a ∈ bR.
2. B´ezout Duo-domains
A ring R is right (left) quasi-duo if every right (left) maximal ideal of R is an ideal. A
ring R is quasi-duo if it is both right and left quasi-duo. It is open whether there exists a
right quasi-duo ring which is not left quasi-duo [8]. We have
Theorem 2.1. Let R be an elementary divisor ring. Then the following are equivalent:
(1) R is quasi-duo.
(2) R is left (right) quasi-duo ring.
(3) For any a ∈ R, RaR = R =⇒ ∃ s, t ∈ R such that sat = 1.
Proof. (1) ⇒ (2) This is trivial.
(2) ⇒ (3) We may assume that R is right quasi-duo. Suppose that RaR = R with a ∈ R.
Write 1 = r1as1 + · · · + rnasn. Then r1aR + · · · + rnaR = R. In light of [8, Theorem 3.2],
Rr1a + · · · + Rrna = R. This shows that sa = 1 for some s ∈ R, as desired.
(3) ⇒ (1) Suppose that aR + bR = R with a, b ∈ R. Let A = (cid:18) a 0
0 (cid:19). Since R is an
b
elementary divisor ring, there exist U, V ∈ GL2(R) such that
U AV = (cid:18) w 0
v (cid:19) ,
0
where RvR ⊆ wRT Rw.
It follows from aR + bR = R that RwR + RvR = R, and so
RwR = R. Since R satisfies Zabavsky condition, we can find some s, t ∈ R such that
swt = 1. Then we verify that
(cid:18)
s
1 − wts wt (cid:19)(cid:18) 0 w
0 (cid:19)(cid:18)
0
v
sw
1 − tsw t (cid:19) = (cid:18) ∗
∗
0
1
∗ (cid:19) ,
Elementary Matrix Reduction Over B´ezout Duo-domains
3
so we show that
(cid:18)
s
1 − wts wt (cid:19) U AV (cid:18) 0 1
1 0 (cid:19)(cid:18)
0
One easily checks that
sw
1 − tsw t (cid:19)(cid:18) 0
0
1
1
0 (cid:19) = (cid:18) 1
∗
∗
∗ (cid:19) .
s
0
(cid:18)
1 − wts wt (cid:19) = B21(−wt)B12(s − 1)B21(1)B12(wt − 1);
(cid:18)
1 − tsw t (cid:19) = B21(−t)B12(sw − 1)B21(1)B12(t − 1).
sw
0
Here,
Thus,
B12(∗) = (cid:18) 1
0
∗
1 (cid:19) , B21(∗) = (cid:18) 1
∗
0
1 (cid:19) .
(cid:18)
s
1 − wts wt (cid:19) ,(cid:18)
0
sw
1 − tsw t (cid:19) ∈ GL2(R).
0
Therefore we can find some P = (pij ), Q = (qij ) ∈ GL2(R) such that
P AQ = (cid:18) 1
0 u (cid:19) .
0
This implies that (p11a + p12b)q11 = 1, and then Ra + Rb = R. It follows by [8, Theorem
3.2] that R is left quasi-quo ring. Likewise, R is right quasi-duo ring.
✷
Following Zabavsky [13], we say that a ring R satisfies Lam condition if RaR = R =⇒ a ∈
R is invertible.
Corollary 2.2. Let R be an elementary divisor ring. Then the following are equivalent:
(1) R is quasi-duo.
(2) R satisfies Lam condition.
Proof. (1) ⇒ (2) This is obvious.
(2) If RaR = R, by hypothesis, a ∈ U (R), which completes the proof by Theorem 2.1.✷
Recall that a ring R satisfies Dubrovin condition if, for any a ∈ R there exists b ∈ R
such that RaR = bR = Rb. We now extend [13, Theorem 4.7.1] as follows:
Corollary 2.3. Let R be an elementary divisor ring. If RaR = R =⇒ ∃ s, t ∈ R such that
sat = 1, then R satisfies Dubrovin condition.
Proof. Let 0 6= a ∈ R. Since R is an elementary divisor ring, there exist U = (uij), V =
(vij ) ∈ GL2(R) such that
(cid:18) a 0
0 a (cid:19) U = V (cid:18) b
0
0
c (cid:19) ,
where RcR ⊆ bRT Rb. It follows that au12 = v12c and au22 = v22c. Since U ∈ GL2(R),
we see that Ru12 + Ru22 = R. In view of Theorem 2.1, R is quasi-duo. By virtue of [8,
Theorem 3.2], we have u12R + u22R = R. Write
u12x + u22y = 1 for some x, y ∈ R.
4
Then
hence,
But
Thus,
Clearly,
Huanyin Chen and Marjan Sheibani
a = au12x + au22y = v12cx + v22cy ∈ RcR;
RaR ⊆ RcR.
RcR ⊆ bR\ Rb ⊆ RbR ⊆ RaR.
RaR = RcR.
RcR ⊆ bR ⊆ RbR ⊆ RaR = RcR and RcR ⊆ Rb ⊆ RbR ⊆ RaR = RcR.
Therefore RaR = RcR = bR = Rb, as desired.
✷
Corollary 2.4 [13, Theorem 4.7.1]. Let R be an elementary divisor ring. If R satisfies
Lam condition, then R satisfies Dubrovin condition.
Proof. Since R satisfies Lam condition, we see that RaR = R implies a ∈ R is invertible.
In view of Corollary 2.3, R satisfies Dubrovin condition, as asserted.
✷
Theorem 2.5. Let R be an elementary divisor domain. Then the following are equivalent:
(1) R is a B´ezout duo-domain.
(2) R satisfies Lam condition.
(3) For any a ∈ R, RaR = R =⇒ ∃ s, t ∈ R such that sat = 1.
Proof. (1) ⇒ (2) ⇒ (3) These are obvious.
(3) ⇒ (1) In view of Theorem 2.1, R is quasi-duo. In view of Corollary 2.3, R satisfies
✷
Dubrovin condition. Therefore R is a duo ring, in terms of [15, Theorem 1].
An element e ∈ R is infinite if there exist orthogonal idempotents f, g ∈ R such that
e = f + g while eR ∼= f R and g 6= 0. A simple domain is said to be purely infinite if every
nonzero left ideal of R contains an infinite idempotent, As is well known, a ring R is a purely
infinite simple ring if and only if it is not a division ring and for any nonzero a ∈ R there
exist s, t ∈ R such that sat = 1 (cf. [1]).
Corollary 2.6. Every purely infinite simple domain is not a B´ezout domain.
Proof. Let R be a purely infinite ring. Suppose that R is a B´ezout domain. In view of [13,
Theorem 1.2.6], R is an Hermite ring. Thus, every 1 × 2 and 2 × 1 matrices over R admit a
diagonal reduction. Let A = (aij ) ∈ M2(R). There exists Q ∈ GL2(R) such that
AQ = (cid:18) a 0
c (cid:19)
b
for some a, b, c ∈ R. If a = b = c = 0, then AQ = 0 is a diagonal matrix. Otherwise, we
may assume that b 6= 0. Thus, we can find s, t ∈ R such that sbt = 1. Then we check that
(cid:18)
sb
1 − tsb
0
b
t (cid:19)(cid:18) a 0
c (cid:19)(cid:18) s
0
1 − bts
bt
(cid:19) = (cid:18) ∗
1
∗
∗ (cid:19) ,
Elementary Matrix Reduction Over B´ezout Duo-domains
5
so we have
B21(∗)(cid:18) 0 1
1 0 (cid:19)(cid:18) sb 1 − tsb
0
t
(cid:19)(cid:18) a 0
c (cid:19)(cid:18)
b
s
1 − bts
0
bt (cid:19) B21(∗) = (cid:18) 1 0
0 ∗ (cid:19) .
As in the proof of Theorem 2.1, we see that
(cid:18)
s
1 − bts
0
bt (cid:19) ,(cid:18) sb
0
1 − tsb
t
(cid:19) ∈ GL2(R).
Therefore, A is equivalent to a diagonal matrix diag(1, ∗). Hence, R is an elementary divisor
ring. If RaR = R, then a 6= 0, and so sat = 1 for some s, t ∈ R. In light of Theorem 2.1, R
is a quasi-duo ring. Thus, R is Dedekind-finite, i.e., uv = 1 in R =⇒ vu = 1. Let 0 6= x ∈ R.
Then there exists c, d ∈ R such that cxd = 1. Hence, xdc = dcx = 1. This implies that
x ∈ U (R), and thus R is a division ring, a contradiction. Therefore we conclude that R is
not a B´ezout domain.
✷
Lemma 2.7. Let R be a B´ezout duo-domain. Then the following are equivalent:
(1) R is an elementary divisor ring.
(2) aR + bR + cR = R with a, b, c ∈ R =⇒ ∃p, q ∈ R such that (pa + qb)R + qcR = R.
Proof. This is clear, by [15, Lemma 2 and Lemma 3].
✷
Theorem 2.8. Let R be a B´ezout duo-domain. Then the following are equivalent:
(1) R is an elementary divisor ring.
(2) Ra + Rb + Rc = R with a, b, c ∈ R =⇒ ∃r, s, t ∈ R such that srt and ra + sb + tc = 1.
Proof. (1) ⇒ (2) Suppose Ra + Rb + Rc = R with a, b, c ∈ R. Then aR + bR + cR = R, as R
is duo. In view of Lemma 2.7, there exist p, q ∈ R such that (pa + qb)R + qcR = R. Hence,
R(pa + qb) + R(qc) = R. Write 1 = x(pa + qb) + y(qc). Then (xp)a + (xq)b + (yq)c = 1. As
R is a duo-domain, we see that (xp)(yq) ∈ xRq = Rxq, as desired.
(2) ⇒ (1) Let a, b, c ∈ R be such that aR + bR + cR = R. Then Ra + Rb + Rc = R. By
hypothesis, there exist r, s, t ∈ R such that tr ∈ Rs and ta + sb + rc = 1. Since R is a B´ezout
ring, we can find a d ∈ R such that Rd = dR = Rs + Rt = sR + tR, as R is a duo-domain.
Case I. d = 0. Then c ∈ U (R), and so (a + c−1 · b)R + (c−1 · c)R = R.
Case II. d 6= 0. Write s = dq and t = dp for some q, p ∈ R. Then Rdr = Rsr + Rtr ⊆
Rs = sR, and so dRr = Rdr ⊇ sR = dqR. This shows that r ∈ qR = Rq. Therefore
1 = ta + sb + rc = d(pa + qb) + kqc ∈ R(pa + qb) + R(qc) = (pa + qb)R + (qc)R; hence,
(pa + qb)R + qcR = R. In light of Lemma 2.7, R is an elementary divisor ring.
✷
Example 2.9. Let R be the ring of all quaternions. Then R is a noncommutative division
ring. Thus, R is an elementary divisor duo-domain, but it is not commutative.
Theorem 2.10. Let R be a B´ezout duo-domain. Then R is an elementary divisor ring if
and only if R/J(R) is an elementary divisor ring.
Proof. =⇒ This is obvious.
⇐= Suppose that aR + bR + cR = R with a, b, c ∈ R. Then (cid:18) a 0
c (cid:19) ∈ M2(R/J(R))
b
admits a diagonal reduction. Thus, we can find some (pij ), (qij ) ∈ GL2(R/J(R)) such that
(pij )(cid:18) a 0
c (cid:19) (qij ) = diag(u, v),
b
6
Huanyin Chen and Marjan Sheibani
where RvR ⊆ uRT Ru. As R is a duo-domain, we see that uR + vR = R. Hence, u ∈
R/J(R) is right invertible. This implies that u ∈ R is invertible, as R is quasi-duo. Therefore
(p11a + p12b)q11 + p12cq12 = u, and so (p11a + p12b)R + p12cR = R. In light of Lemma 2.7,
R is an elementary divisor domain.
✷
Corollary 2.11. Let R be a B´ezout duo-domain. Then R is an elementary divisor ring if
and only if R[[x1, · · · , xn]] is an elementary divisor ring.
Proof. ⇐= This is obvious, as R is a homomorphic image of R[[x1, · · · , xn]].
=⇒ Let φ : R[[x1, · · · , xn]] → R is defined by φ(f (x1, · · · , xn)) = f (0, · · · , 0).
It
is obvious that I := Ker(φ) ⊆ J(R[[x1, · · · , xn]]) and R[[x1, · · · , xn]]/I ∼= R. Since
R/I/J(R)/I ∼= R/J(R), R/I/J(R)/I is an elementary divisor ring if and only if so is
R/J(R). Since R/I ∼= R, the result follows by Theorem 2.10.
✷
3. Locally Stable domains
We say that an element a of a duo-ring R is stable if, R/aR has stable range 1. A duo
ring R is locally stable if, aR + bR = R with a, b ∈ R =⇒ ∃ y ∈ R such that a + by ∈ R is
stable. For instance, every duo-ring of stable range 1 is locally stable. The purpose of this
section is to investigate matrix diagonal reduction over locally stable B´ezout duo-domain.
Lemma 3.1. Let R be a B´ezout domain. If pR + qR = R with p, q ∈ R, then there exists
a matrix (cid:18) p
∗
q
∗ (cid:19) ∈ GL2(R).
Proof. As every B´ezout domain is a Hermite domain, then there exists some Q ∈ GL2(R)
such that (p, q)Q = (d, 0) for some d ∈ R. We have p, q ∈ dR and also R = pR + qR ⊆ dR
this implies that d is right invertible so there exists some x ∈ R such that dx = 1, now
xdx = x then (xd − 1)x = 0. Since R is a domain, then xd − 1 = 0 that implies xd = 1 and
so d is invertible. Now (p, q)Q(cid:18) d−1
(p, q) = (1, 0)U , this means that (p, q) is the first row of U So U = (cid:18) p
d−1 (cid:19) = (1, 0). Set U −1 = Q(cid:18) d−1
d−1 (cid:19). Then
∗ (cid:19) ∈ GL2(R) as
0
q
∗
0
0
0
required.
✷
Let R be a B´ezout domain. If Rx + Ry = R with x, y ∈ R, similarly, we can find a
matrix (cid:18) x ∗
∗ (cid:19) ∈ GL2(R).
y
Theorem 3.2. Every locally stable B´ezout duo-domain is an elementary divisor ring.
Proof. Let R be a locally stable B´ezout duo-domain. Suppose that aR + bR + cR = R with
a, b, c ∈ R. Then ax + by + cz = 1 for some x, y, z ∈ R. Since R is locally stable, there exists
some s ∈ R such that R/(b + (ax + cz)s)R has stable range 1. Set w = b + (ax + cz)s. Then
aR + cR = 1 in R/wR. Thus, we have
(cid:18) 0
1
1
0 (cid:19)(cid:18) 1
xz
0
1 (cid:19)(cid:18) a 0
c (cid:19)(cid:18) 1
zs 1 (cid:19) = (cid:18) w c
0 (cid:19) .
a
0
b
Elementary Matrix Reduction Over B´ezout Duo-domains
7
In light of [13, Theorem 1.2.6], R is a Hermite ring. Thus, we have a Q ∈ GL2(R) such that
(w, c)Q = (v, 0), and then
(cid:18) w c
l (cid:19) .
Clearly, w ∈ vR; hence, R/vR ∼= R/wR
vR/wR has stable range 1.
0 (cid:19) W = (cid:18) v
a
k
0
One easily checks that vR + kR + lR = R, and so kR + lR = R. Thus, we have a t ∈ R
such that vR + (k + lt)R = R. Since R is duo, we see that Rv + (k + lt)R = R. Write
pv + p′(k + lt) = 1. Then
(cid:18) v
k
0
l (cid:19)(cid:18) 1 0
1 (cid:19) = (cid:18) v
k + lt 1 (cid:19) .
0
t
By virtue of Lemma 3.1, we can find some P, Q ∈ GL2(R) such that
Write P = (cid:18) p
p′
q
q′ (cid:19) and Q−1 = (cid:18) m n
P (cid:18) a 0
b
0
0 d (cid:19) .
c (cid:19) Q = (cid:18) 1
m′ n′ (cid:19). Then pa + qb = m and qc = n. As
mR + nR = R, we see that (pa + qb)R + qcR = R. In light of Lemma 2.7, we complete the
proof.
✷
As an immediate consequence, we extend [9, Theorem 3.7] from commutative case to
duo rings.
Corollary 3.3. Let R be a B´ezout duo-domain. If R/aR has stable range 1 for all nonzero
a ∈ R, then R is an elementary divisor domain.
Proof. Given aR + bR = R with a, b ∈ R, then a 6= 0 or a + b 6= 0. Hence, a + by 6= 0 for
some y ∈ R. By hypothesis,
(a+by)R has stable ring 1, i.e., R is locally stable. Therefore R
is an elementary divisor domain, by Theorem 3.2.
✷
R
Since every B´ezout duo-domain of stable range 1 satisfies the condition in Corollary 3.3,
we derive
Lemma 3.4. Every B´ezout duo-domain of stable range 1 is an elementary divisor domain.
Theorem 3.5. Let R be a B´ezout duo-domain. If aR + bR = R =⇒ ∃ x, y ∈ R such that
xR + yR = R, xy ∈ J(R), ax, by, then R is an elementary divisor ring.
Proof. Suppose that aR + bR = R with a, b ∈ R. Then there exist x, y ∈ R such that
xR + yR = R, xy ∈ J(R), ax, by. Write xr + ys = 1 for some r, s ∈ R. Since R is a
duo-domain, we see that xr ∈ Rx, and then (xr)2 ≡ xr(mod J(R)).
Set g = xr and e = g + gxr(1 − g). Then e ∈ xR ⊆ aR. One easily checks that
1 − e = (cid:0)1 − gxr(cid:1)(1 − g) ≡ 1 − g(mod J(R)). Thus, we have some d ∈ J(R) such that
1 − e − d = 1 − g = ys ∈ yR ⊆ bR. Set f = 1 − e − d. Then e + f = 1 − d ∈ U (R), and so
e(1 − d)−1 + f (1 − d)−1 = 1. This implies that eR + f R = R. One easily checks that ae
and bf . Moreover ef = (e − e2) − ed ∈ J(R). Therefore R/J(R) is a duo exchange ring. In
view of [1, Corollary 1.3.5], R/J(R) has stable range 1, and then so does R. According to
Lemma 3.4, we complete the proof.
✷
8
Huanyin Chen and Marjan Sheibani
Recall that a ring R is feckly clean if for any a ∈ R, there exists e ∈ R such that
a − e ∈ U (R) and eR(1 − e) ⊆ J(R) [2]. We now derive
Corollary 3.6. Every feckly clean B´ezout duo-domain is an elementary divisor ring.
Proof. Let R be a feckly clean B´ezout duo-ring. Suppose that aR + bR = R with a, b ∈ R.
Then ax + by = 1 for some x, y ∈ R. Set c = ax. Then there exist f ∈ R and u ∈ U (R)
such that c = f + u and f R(1 − f ) ∈ J(R). Let g = (1 − f ) + (f − f 2)u−1. Then
g − g2 ≡ f − f 2 ≡ 0(cid:0) mod J(R)(cid:1). Hence, g − g2 ∈ J(R). As f u(1 − f ) ∈ J(R), we see that
f u − f uf ∈ J(R). On the other hand, (R(1 − f )uf R)2 = R(1 − f )u(f R(1 − f ))uf R ⊆ J(R),
and then (1 − f )uf ∈ J(R). This implies that uf = f uf . Therefore, f u − uf ∈ J(R).
Write uf = f u + r for some r ∈ J(R). Then there exists some r′ ∈ J(R) such that
a − g = f + u − 1 + f − (f − f 2)u−1 = (u − 2f u − u2 + f − f 2)(−u−1) = (c − c2)(−u−1) + r′.
Write e = g − r′. Then a − e ∈ (c − c2)R. Write c − e = (c − c2)s. Then e = c(cid:0)1 − (1 − c)cs(cid:1) ∈
cR ⊆ aR, and that 1 − e = (1 − c)(1 + cs) ∈ (1 − c)R ⊆ bR. We easily check that
eR + (1 − e)R = R, e(1 − e) ∈ J(R) and ae, b1 − e. According to Theorem 3.5, R is an
elementary divisor ring.
✷
Example 3.7. Let R := Z(2)T Z(3) = { m
B´ezout duo-domain, and then it is an elementary divisor domain.
n m, n ∈ Z, 2, 3 ∤ n}. Then R is a feckly clean
We say that c ∈ R is feckly adequate if for any a ∈ R there exist some r, s ∈ R such that
(1) c ≡ rs (mod J(R)); (2) rR + aR = R; (3) s′R + aR 6= R for each non-invertible divisor
s′ of s. The following is an generalization of [3, Lemma 3.1] in the duo case.
Lemma 3.8. Every feckly adequate element in B´ezout duo-domains is stable.
Proof. Let R be a B´ezout duo-domain, and let a ∈ R be feckly adequate. Set S = R/aR. Let
b ∈ S. Then there exist r, s ∈ R such that a ≡ rs(mod J(R)), rR+bR = R and s′R+bR 6= R
for any noninvertible divisor s′ of s. Hence, a ≡ rs(cid:0)mod J(S)(cid:1), i.e., rs ∈ J(S). Clearly,
rS + bS = S. Let t ∈ S be a noninvertible divisor of s. As in the proof of [3, Lemma 3.1],
we see that sR + tR 6= R. Since R is a B´ezout ring, we have a noninvertible u ∈ R such that
sR + tR = uR. We infer that u is a noninvertible divisor of s. Hence, uR + bR 6= R. This
proves that uS + bS 6= S; otherwise, there exist x, y, z ∈ R such that ux + by = 1 + az. This
implies that ux + by = 1 + w′z + rsz = 1 + w′z + ucrz for c ∈ R and w′ ∈ J(R), because
a − rs ∈ J(R). Hence, u(x − crz)(1 + w′z)−1 + by(1 + w′z)−1 = 1, a contradiction. Thus
tS + bS 6= S. Therefore, 0 ∈ S is feckly adequate.
Let x ∈ S. by the preceding discussion, we have some r, s ∈ R such that rs ∈ J(S),
where rS + xS = S and s′S + xS 6= S for any noninvertible divisor s′ of s. Similarly,
we have rS + sS = S. Since rS + xS = S, we get rS + sxS = S, as S is duo. Write
rc + sxd = 1 in S. Set e = rc. Then e2 − e = (rc)2 − rc = −(rc)(sxd) ∈ rsS ⊆ J(S). Let
u be an arbitrary invertible element of S. As in the proof of [3, Lemma 3.1], we verify that
(u − ex)S + rsS = S, as sS = Ss. Since rs ∈ J(S), we get u − ex ∈ U (S). This implies
that (cid:0)x − (1 − e)x(cid:1) − u = ex − u ∈ U (S). Furthermore, S has stable range 1. We infer that
x − (1 − e)x ∈ J(S). Clearly, 1 − e = sxd ∈ xR. Therefore x = x(sd)x in S/J(S), and so
S/J(S) is regular.
As R is duo, so is S/J(S). Hence, S has stable range 1, by [1, Corollary 1.3.15]. Therefore
✷
a ∈ R is stable, as required.
A B´ezout domain R is a Zabavsky domain if, aR + bR = R with a, b ∈ R =⇒ ∃y ∈ R
Elementary Matrix Reduction Over B´ezout Duo-domains
9
such that a + by ∈ R is feckly adequate. For instance, every domain of stable range 1 is a
Zabavsky domain. We can derive
Theorem 3.9. Every Zabavsky duo-domain is an elementary divisor ring.
Proof. Suppose that aR + bR = R with a, b ∈ R. Then there exists a y ∈ R such that
a + by ∈ R is adequate. In view of Lemma 3.8, a + by ∈ R is stable. Therefore R is locally
stable. This completes the proof, by Theorem 3.2.
✷
A ring R is called adequate if every nonzero element in R is adequate. We can extend
[13, Theorem 1.2.13] as follows:
Corollary 3.10. Every adequate B´ezout duo-domain is an elementary divisor domain.
Proof. Let R be a adequate B´ezout duo-domain. If aR + bR = R with a, b ∈ R, then a
or a + b 6= 0. Hence, a + by ∈ R is adequate, where y = 0 or 1. This implies that R is a
Zabavsky duo-domain. The result follows by Theorem 3.9.
✷
Theorem 3.11. Let R be a B´ezout duo-domain. If for any a ∈ R either a or 1 − a is
adequate, then R is an elementary divisor domain.
Proof. Given aR + bR = R with a, b ∈ R. Write ax + by = 1 for some x, y ∈ R. Then
a(x − y) + (a + b)y = 1. By hypothesis, either a(1 − x) or 1 − a(1 − x) is adequate. In view
of Lemma 3.8, a(1 − x) or 1 − a(1 − x) is stable. If a(1 − x) ∈ R is stable, then R/a(1 − x)R
has stable range 1. As R/aR ∼= R/a(1−x)
aR/a(1−x)R , we easily check that R/aR has stable range 1.
If (a + b)y ∈ R is stable, similarly, R/(a + b)R has stable range 1. Therefore R/(a + by) has
stable range 1 where y = 0 or 1. Hence, R is locally stable, whence the result by Theorem
3.2.
✷
✷
Corollary 3.12. Let R be a B´ezout duo-domain. If for any a ∈ R either J( R
J( R
(1−a)R ) = 0, then R is an elementary divisor domain.
aR ) = 0 or
Proof. Let a ∈ R. If a = 0, then 1 − a ∈ R is adequate. If a = 1, then a ∈ R is adequate.
We now assume that a 6= 0, 1. By hypothesis, J( R
(1−a)R ) = 0. In view of [11,
Theorem 17], a or 1 − a is sqaure-free. It follows by [11, Proposition 16] that a or 1 − a is
adequate in R. Therefore we complete the proof, by Theorem 3.11.
✷
aR ) = 0 or J( R
A ring R is called homomorphically semiprimitive if every ring homomorphic image
(including R) of R has zero Jacobson radical. Von Neumann regular rings are clearly ho-
momorphically semiprimitive. But the converse is not true in general, for example, let
R = W1[F ] be the first Weyl algebra over a field F of characteristic zero. As an immediate
consequence of Corollary 3.12, we have
Corollary 3.13. Let R be a B´ezout duo-domain. If R is homomorphically semiprimitive,
then R is an elementary divisor domain.
4. Rings of Gelfand Range 1
10
Huanyin Chen and Marjan Sheibani
A ring R is a PM ring if every prime ideal of R contains in only one maximal ideal.
aR is a PM ring. We easily extend [4,
An element a of a B´ezout duo-domain R is PM if, R
Theorem 4.1] to duo-rings.
Lemma 4.1. Let R be a duo-ring. Then the following are equivalent:
(1) R is a PM ring.
(2) a + b = 1 in R implies that (1 + ar)(1 + bs) = 0 for some r, s ∈ R.
Lemma 4.2. Let R be a B´ezout duo-domain, and let a ∈ R. Then the following are
equivalent:
(1) a ∈ R is PM.
(2) If bR + cR = R, then there exist r, s ∈ R such that a = rs, rR + bR = sR + cR = R.
(3) If aR+bR+cR = R, then there exist r, s ∈ R such that a = rs, rR+bR = sR+cR = R.
(1) ⇒ (3) Suppose aR + bR + cR = R. Then b(R/aR) + c(R/aR) = R/aR. By
Proof.
virtue of Lemma 4.1, we can find some p, q ∈ R such that (1 + bp)(1 + cq) = 0 in R/aR. Set
k = 1 + bp and l = 1 + cq. Then kR + bR = lR + cR = R and that kl ∈ aR.
Write kl = ap for some p ∈ R. Since R is a B´ezout duo-domain, we have an r ∈ R such
that kR+aR = rR. Thus, there are some s, t ∈ R such that a = rs, k = rt and sR+tR = R.
Clearly, rR + bR = R. It follows that rtl = rsp, and so tl = sp. Write lu + cv = 1. Then
tlu + ctv = spu + ctv = t. But sR + tR = R, we get sR + cR = R, as required.
(3) ⇒ (2) This is obvious.
(2) ⇒ (1) Suppose b + c = 1 in R/aR. Then b + c + ax = 1 for some x ∈ R. By
hypothesis, there exist r, s ∈ R such that a = rs, rR + bR = sR + (c + ax)R = R. Write
rk + bl = sp + (c + ax)q = 1. Then (1 − bk)(1 − cq) = rskp = 0 in R/aR. Therefore R/aR
is a PM ring, by Lemma 4.1. That is, a ∈ R is PM, as asserted.
✷
A B´ezout duo-ring has Gelfand range 1 if, aR + bR = R with a, b ∈ R implies that there
exists a y ∈ R such that a + by ∈ R is PM. We come now to the main result of this section:
Theorem 4.3. Every B´ezout duo-domain of Gelfand range 1 is an elementary divisor ring.
Proof. Let R be a B´ezout duo-domain of Gelfand range 1. Suppose that aR + bR + cR = R
with a, b, c ∈ R. Write ax + by + cz = 1 with x, y, z ∈ R. Then we can find some k ∈ R such
that w := b + (ax + cz)k ∈ R is PM. Hence,
(cid:18) 1
xk
0
1 (cid:19) A(cid:18) 1
zk 1 (cid:19) = (cid:18) a
w c (cid:19) .
0
0
Clearly, wR + aR + cR = R. In view of Lemma 4.2, there exist r, s ∈ R such that w =
rs, rR + aR = sR + cR = R. Write sp′ + cl = 1. Then rsp′ + rcl = r. Set q′ = rl. Then
wp′ + cq′ = r. Write p′R + q′R = dR. Then p′ = dp, q′ = dq and pR + qR = R. Hence,
(wp + cq)R + aR = R. Clearly, cR + pR = qR + pR = R, we get cqR + pR = R. This implies
that (wp+cq)R+pR = R. Thus, we have apR+(wp+cq)R = R. Write aps+(wp+cq)t = 1.
Then p(as + wt) + q(ct) = 1. Hence,
(cid:18)
s
wp + cq −ap (cid:19)(cid:18) a
w c (cid:19)(cid:18) p
−ct
q as + wt (cid:19) = (cid:18) 1 0
0 ∗ (cid:19) ,
0
t
Elementary Matrix Reduction Over B´ezout Duo-domains
11
where (cid:12)(cid:12)(cid:12)(cid:12)
t
s
wp + cq −ap (cid:12)(cid:12)(cid:12)(cid:12)
= (cid:12)(cid:12)(cid:12)(cid:12)
p
−ct
q as + wt (cid:12)(cid:12)(cid:12)(cid:12)
p, q ∈ R such that (pa + qb)R + qcR = R. This completes the proof, by Lemma 2.7.
✷
= 1. As in the proof of Theorem 3.2, we have
Corollary 4.4 [14, Theorem 3]. Every commutative B´ezout domain of Gelfand range 1
is an elementary divisor ring.
We note that Theorem 4.3 can not be extended to any rings with zero divisor as the
following shows.
Example 4.5. Let R+ be the subset {(x, 0) : x > 0} of the plane S+ = {(x, sinπx) : x ≥ 0}
and let X = S+S R+. If R = C(X − β(X)) where β(X) the largest compact Hausdorff
space generated by X in the sense that any map from X to a compact Hausdorff space
factors through β(X), and C(X − β(X) is the ring of all continuous functions on X − β(X)
In view of [7], C(X − β(X)) is a B´ezout duo-ring. On the other hand, C(X − β(x)) is a
PM ring (cf. [4]). Thus C(X − β(x)) is a B´ezout duo-ring of Gelfand range 1. But it is not
Hermite ring (cf. [7]), and therefore R is not an elementary divisor ring.
Lemma 4.6. Let I be an ideal of a PM ring R. Then R/I is a PM ring.
Proof. Let P be a prime ideal of R. Then P = Q/I where Q is a prime ideal of R. Since
R is a PM ring, there exists a unique maximal M of R such that Q ⊆ M . Hence, P ⊆ M/I
where M/I is a maximal of R/I. The uniqueness is easily checked, and therefore R/I is
PM.
✷
Theorem 4.7. Let R be B´ezout duo-domain. Suppose for any a ∈ R, either a or 1 − a is
PM. Then R is an elementary divisor ring.
Proof. Given aR + bR = R with a, b ∈ R. Write ax + by = 1 for some x, y ∈ R. As in the
proof of Theorem 3.11, we see that a(1 − x) or 1 − a(1 − x) is PM. If a(1 − x) ∈ R is PM,
then R/a(1 − x)R is PM. Since R/aR ∼= R/a(1−x)
aR/a(1−x)R , it follows by Lemma 4.6 that R/aR is
PM. If (a + b)y ∈ R is PM, R/(a + b)R is PM. Hence, R/(a + by) is PM, and therefore R is
an elementary divisor ring, by Theorem 4.3.
✷
Corollary 4.8. Every B´ezout PM duo-domain is an elementary divisor ring.
5. Countability Conditions
Let R be a ring and a ∈ R. We use mspec(a) to denote the set of all maximal ideals which
contain a. We now determine elementary divisor domains by means of the countability of
the sets of related elements.
Lemma 5.1. Let R be a B´ezout domain, and let aR + bR + cR = R with a, b, c ∈ R. If
mspec(a) is at most countable, then there exist p, q ∈ R such that (pa + qb)R + qcR = R.
Proof. In view of [13, Theorem 1.2.6], R is an Hermite ring.
12
Huanyin Chen and Marjan Sheibani
Case I. mspec(a) is empty. Then a ∈ U (R). Hence,
(cid:18) a b
c (cid:19)(cid:18) 1 −a−1b
0
0
1
(cid:19) = (cid:18) a 0
c (cid:19) .
0
Case II. mspec(a) 6= ∅. Set mspec(a) = {M1, M2, M3, · · · }. If b ∈ M1, then c 6∈ M1;
hence, b + c 6∈ M1. By an elementary transformation of column, b can be replaced by the
element b + c. Thus, we may assume that b 6∈ M1. As R is a Hermite ring, there exists
P1 ∈ GL2(R) such that P1A = (cid:18) a1
0
b1
c1 (cid:19). This implies that aR+bR = a1R. If a1 ∈ U (R),
then
is a diagonal matrix.
P1A(cid:18) 1 −a−1
1 b1
1
0
(cid:19) = diag(a1, c1)
Assume that a1 6∈ U (R). Then mspec(a1) 6= ∅. As mspec(a1) ⊆ mspec(a), we may
assume that a1 ∈ M2.
If b1 ∈ M2, then c1 6∈ M2 since a1R + b1R + c1R = R. Hence,
b1 + c1 6∈ M2. By an elementary transformation of row, b1 can be replaced by the element
b1 + c1. Thus, we may assume that b1 6∈ M2. Since R is a Hermite ring, there exists
c2 (cid:19). Clearly, a1R + b1R = a2R. Moreover, we
Q1 ∈ GL2(R) such that P1AQ1 = (cid:18) a2
see that a ∈ M1, a1 ∈ M2\M1, a2 ∈ M3\(cid:0)M1S M2(cid:1). By iteration of this process, there is a
collection of matrices of the form
b2
0
PkAQk = (cid:18) ai
0
∗
∗ (cid:19) .
If there exists some ai ∈ U (R), then A admits a diagonal reduction. Otherwise, we get an
infinite chain of ideals
aR $ a1R $ a2R $ · · · .
∞
Further, each an 6∈ Mn(n ∈ N. Let I =
aiR. Then I 6= R; hence, there exists a maximal
Si=1
ideal J of R such that I ⊆ J ( R. This implies that a ∈ J, and so J ∈ mspec(a). Thus,
J = Mk for some k ∈ N. As ak 6∈ Mk, we deduce that ak 6∈ J, a contradiction. Therefore A
admits a diagonal reduction. As in the proof of Theorem 3.2, there exist p, q ∈ R such that
(pa + qb)R + qcR = R, as required.
✷
We say that an element a of a ring R is a non-P element if, R
aR do not satisfy the property
P , where P is some ring theoretical property. We now derive
Theorem 5.2. Let R be a B´ezout duo-domain. Then R is an elementary divisor ring if
(1) the set of nonstable elements in R is at most countable, or
(2) the set of non-PM elements in R is at most countable.
Proof. (1) If R has stable range 1, then it is an elementary divisor ring, by Lemma 3.4. We
now assume that R has no stable range 1. Suppose that aR + bR + cR = R with a, b, c ∈ R.
If a ∈ R is stable, as in the proof of Theorem 3.2, we can find some p, q ∈ R such that
(pa + qb)R + qcR = R. If a ∈ R is nonstable, then a 6= 0; otherwise, R has stable range
1. Since R is a domain, we see that am 6= an for any distinct m, n ∈ N. If mspec(a) is
uncountable, then {a, a2, · · · , ak, · · · } is uncountable as a subset of mspec(a). This gives a
contradiction. This implies that mspec(a) is at most countable. By virtue of Lemma 5.1,
Elementary Matrix Reduction Over B´ezout Duo-domains
13
there exist p, q ∈ R such that (pa + qb)R + qcR = R. Therefore proving (1), in terms of
Lemma 2.7.
(2) Suppose that aR + bR + cR = R with a, b, c ∈ R. If a ∈ R is stable, similarly to
Theorem 3.2, we have some p, q ∈ R such that (pa + qb)R + qcR = R. If a ∈ R is a non-PM
element, by hypothesis, mspec(a) is at most countable. In light of Lemma 5.1, we can find
p, q ∈ R such that (pa + qb)R + qcR = R. Thus, we obtain the result by Lemma 2.7.
✷
Recall that an ideal of a duo ring R is a maximally nonprincipal ideal if it is maximal in
the set of nonprincipal ideals of R with respect to the ordering by set inclusion.
Lemma 5.3. Let R be a B´ezout domain, and let aR + bR + cR = R with a, b, c ∈ R. If
a does not belong to any maximally nonprincipal ideal, then there exist p, q ∈ R such that
(pa + qb)R + qcR = R.
Proof. Set A = (cid:18) a 0
c (cid:19) ∈ M2(R). Since a does not belong to any maximally nonprincipal
b
ideal, it follows by [7, Theorem 1] that A admits a diagonal reduction. Therefore we complete
the proof, as in the proof of Theorem 3.2.
✷
Theorem 5.4. Let R be a B´ezout duo-domain. Then R is an elementary divisor ring if
(1) every nonstable element in R is contained in a maximal nonprincipal ideal, or
(2) every non-PM element in R is contained in a maximal nonprincipal ideal.
Proof. (1) Suppose that aR + bR + cR = R with a, b, c ∈ R. If a ∈ R is stable, similarly to
Theorem 3.2, there are some p, q ∈ R such that (pa + qb)R + qcR = R. If a ∈ R is nonstable,
by hypothesis, a ∈ R is contained in a maximal nonprincipal ideal. In light of Lemma 5.3,
(pa + qb)R + qcR = R for some p, q ∈ R. It follows by Lemma 2.7 that R is an elementary
divisor ring.
(2) Analogously to Theorem 4.3 and the preceding discussion, we complete the proof, by
✷
Lemma 5.3 and Lemma 2.7.
Corollary 5.5. Let R be a B´ezout duo-domain. Then R is an elementary divisor ring if
(1) every nonlocal element in R is contained in a maximal nonprincipal ideal, or
(2) every nonregular element in R is contained in a maximal nonprincipal ideal.
Proof. This is obvious by Theorem 5.4, as every local duo-ring and every regular duo-ring
are both PM rings.
✷
References
[1] H. Chen, Rings Related Stable Range Conditions, Series in Algebra 11, World Scientific, Hack-
ensack, NJ, 2011.
[2] H. Chen; H. Kose and Y. Kurtulmaz, On feckly clean rings, J. Algebra Appl., 14(2015),
1550046(15 pages). DOI:10.1142/S0219498815500462.
[3] H. Chen and M. Sheibani, Elementary matrix reduction over Zabavsky rings, Bull. Korean
Math. Soc., to appear.
[4] M. Contessa, On pm-rings, Comm. Algebra, 10(1982), 93 -- 108.
14
Huanyin Chen and Marjan Sheibani
[5] O.V. Domsha and I.S. Vasiunyk, Combining local and adequate rings, Book of abstracts of
the International Algebraic Conference, Taras Shevchenko National University of Kyiv, Kyiv,
Ukraine, 2014.
[6] A. Gatalevich and B.V. Zabavsky, Noncommutative elementary divisor domains, J. Math. Sci.,
96(1999), 3013 -- 3016.
[7] L. Gillman and M. Henriksen, Rings of continuous functions in which every finitely generated
ideal is principal, Trans. Amer. Math. Soc., 82(1956), 362 -- 365.
[8] T.Y. Lam and A.S. Dugas, Quasi-duo rings and stable range descent, J. Pure Appl. Algebra,
195(2005), 243 -- 259.
[9] W.W. McGovern, B´ezout rings with almost stable range 1, J. Pure Appl. Algebra, 212(2008),
340 -- 348.
[10] M. Roitman, The Kaplansky condition and rings of almost stable range 1, Proc. Amer. Math.
Soc., 141(2013), 3013 -- 3018.
[11] O.S. Sorokin, Finite homomorphic images of B´ezout duo-domain, Carpathian Math. Publ.,
6(2014), 360 -- 366.
[12] S.H. Sun, Noncommutative rings in which every prime ideal is contained in a unique maximal
ideal, J. Pure Appl. Algebra, 76(1991), 179 -- 192.
[13] B.V. Zabavsky, Diagonal Reduction of Matrices over Rings, Mathematical Studies Monograph
Series, Vol. XVI, VNTL Publisher, 2012.
[14] B.V. Zabavsky, Conditions for stable range of an elementary divisor rings, arXiv:1508.07418v1
[math.RA] 29 Aug 2015.
[15] B.V. Zabavsky and N.Y. Komarnitskii, Distributive elementary divisor domains, Ukrainskii
Math. Zhurnal, 42(1990), 1000 -- 1004.
|
1807.11813 | 1 | 1807 | 2018-07-31T13:40:56 | The Dixmier-Moeglin equivalence, Morita equivalence, and homeomorphism of spectra | [
"math.RA"
] | Let $k$ be a field and let $R$ be a left noetherian $k$-algebra. The algebra $R$ satisfies the Dixmier-Moeglin equivalence if the annihilators of irreducible representations are precisely those prime ideals that are locally closed in the ${\rm Spec}(R)$ and if, moreover, these prime ideals are precisely those whose extended centres are algebraic extensions of the base field. We show that if $R$ and $S$ are two left noetherian $k$-algebras with ${\rm dim}_k(R), {\rm dim}_k(S)<|k|$ then if $R$ and $S$ have homeomorphic spectra then $R$ satisfies the Dixmier-Moeglin equivalence if and only if $S$ does. In particular, the topology of ${\rm Spec}(R)$ can detect the Dixmier-Moeglin equivalence in this case. In addition, we show that if $k$ is uncountable and $R$ is affine noetherian and its prime spectrum is a disjoint union of subspaces that are each homeomorphic to the spectrum of an affine commutative ring then $R$ satisfies the Dixmier-Moeglin equivalence. We show that neither of these results need hold if $k$ is countable and $R$ is infinite-dimensional. Finally, we make the remark that satisfying the Dixmier-Moeglin equivalence is a Morita invariant and finally we show that $R$ and $S$ are left noetherian $k$-algebras that satisfy the Dixmier-Moeglin equivalence then $R\otimes_k S$ does too, provided it is left noetherian and satisfies the Nullstellensatz; and we show that $eRe$ also satisfies the Dixmier-Moeglin equivalence, where $e$ is a nonzero idempotent of $R$. | math.RA | math | THE DIXMIER-MOEGLIN EQUIVALENCE, MORITA EQUIVALENCE, AND
HOMEOMORPHISM OF SPECTRA
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
Abstract. Let k be a field and let R be a left noetherian k-algebra. The algebra R satisfies the
Dixmier-Moeglin equivalence if the annihilators of irreducible representations are precisely those
prime ideals that are locally closed in the Spec(R) and if, moreover, these prime ideals are precisely
those whose extended centres are algebraic extensions of the base field. We show that if R and S
are two left noetherian k-algebras with dimk(R), dimk(S) < k then if R and S have homeomorphic
spectra then R satisfies the Dixmier-Moeglin equivalence if and only if S does. In particular, the
topology of Spec(R) can detect the Dixmier-Moeglin equivalence in this case. In addition, we show
that if k is uncountable and R is affine noetherian and its prime spectrum is a disjoint union
of subspaces that are each homeomorphic to the spectrum of an affine commutative ring then R
satisfies the Dixmier-Moeglin equivalence. We show that neither of these results need hold if k is
countable and R is infinite-dimensional. Finally, we make the remark that satisfying the Dixmier-
Moeglin equivalence is a Morita invariant and finally we show that R and S are left noetherian
k-algebras that satisfy the Dixmier-Moeglin equivalence then R ⊗k S does too, provided it is left
noetherian and satisfies the Nullstellensatz; and we show that eRe also satisfies the Dixmier-Moeglin
equivalence, where e is a nonzero idempotent of R.
8
1
0
2
l
u
J
1
3
]
.
A
R
h
t
a
m
[
1
v
3
1
8
1
1
.
7
0
8
1
:
v
i
X
r
a
1. Introduction
Given a ring R, one of the most valuable methods of gaining information about the structure
of R is via its representation theory; that is, by first understanding the structure of the simple
left R-modules and then using this information to gain insight into the ring itself. Although this
is a highly useful method for many classes of rings, in practice it is often very difficult to do so,
and so often one instead settles for a coarser understanding of the representation theory by instead
understanding the annihilators of simple modules; that is the primitive ideals of R. If the Jacobson
radical of R is zero then R is a subdirect product of rings R/P , where P ranges over the primitive
ideals; and by Jacobson's density theorem, rings of the form R/P are dense subrings of rings of
linear operators, and so many structure-theoretic problems about a ring can still be resolved with
a sufficiently good understanding of the primitive ideals of a ring.
One of the most beautiful results in this direction of characterizing primitive ideals is the work
of Dixmier and Moeglin [Dix77, Moe80], who showed that if L is a finite-dimensional complex Lie
algebra then the primitive ideals of the enveloping algebra U (L) are precisely the prime ideals of
Spec(U (L)) that are locally closed in the Zariski topology. In addition to this, they proved that a
prime ideal P of U (L) is primitive if and only if the Goldie ring of quotients of U (L)/P has the
property that its centre is just the base field of the complex numbers. In general, for a field k and
a left noetherian k-algebra R, prime ideals P for which the centre of the Goldie ring of quotients
2010 Mathematics Subject Classification. 16D60 16A20, 16A32.
Key words and phrases. primitive ideals, Dixmier-Moeglin equivalence, prime spectrum, tensor products, idempo-
tents, Morita equivalence.
The first author was supported by a Discovery Grant from the National Sciences and Engineering Research Council
of Canada.
1
2
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
Q(R/P ) of R/P has the property that its centre is an algebraic extension of k are called rational
prime ideals. Hence Dixmier and Moeglin's result can be regarded as saying that for primes P of
Spec(U (L)) we have the following equivalences:
P locally closed ⇐⇒ P primitive ⇐⇒ P rational.
In their honour, we today say that a left noetherian k-algebra R satisfies the Dixmier-Moeglin
equivalence if we have the equivalence of the three above properties for primes in the spectrum of R.
It is now known that the Dixmier-Moeglin equivalence is a very general phenomenon that holds for
many classes of algebras beyond just enveloping algebras of finite-dimensional Lie algebras. Some
additional examples include affine PI algebras [Von96, 2.6], group algebras of nilpotent-by-finite
groups [Zal71], various quantum algebras [GL00] (and see [BG01, II.8.5]), affine cocommutative
Hopf algebras of finite Gelfand-Kirillov dimension in characteristic zero [BL14], Hopf Ore extensions
of affine commutative Hopf algebras [BSM18], twisted homogeneous coordinate rings of surfaces
[BRS10], Hopf algebras of Gelfand-Kirillov dimension two (under mild homological assumptions)
[GZ10], and even in settings where the noetherian property does not hold and in which one must
suitably modify the rationality property [ABR12, Lor09, Lor08]. Nevertheless, the equivalence is
not universal and there are several finitely generated noetherian counterexamples are now known
[BLLM17, BCM17, Irv79, Lor77].
In this short note, our main result is to show that for a left noetherian k-algebra R with the
property that dimk(R) < k, the poset of prime ideals can detect whether the Dixmier-Moeglin
equivalence holds.
Theorem 1.1. Let k be a field and let R and S be left noetherian k-algebras with dimk(R), dimk(S) <
k and suppose that there is an inclusion-preserving bijection between the poset of prime ideals of R
and the poset of prime ideals of S. Then R satisfies the Dixmier-Moeglin equivalence if and only if S
satisfies the Dixmier-Moeglin equivalence. In particular, if Spec(R) and Spec(S) are homeomorphic
then R satisfies the Dixmier-Moeglin equivalence if and only if S does.
This result is somewhat curious, because it says that for sufficiently large base fields the un-
derlying topology on the prime spectrum "sees" the Dixmier-Moeglin equivalence. This is perhaps
somewhat surprising because the rationality property has no obvious strong connection to the topo-
logical structure of the prime spectrum of a ring. In fact, we are able to give an example, inspired by
a construction of Lorenz [Lor77], to show that if the hypothesis dimk(R) < k does not hold then
one can have an example of algebras with homeomorphic spectra in which the Dixmier-Moeglin
equivalence holds for one but not the other.
We are able to prove a related theorem.
Theorem 1.2. Let k be an uncountable field and let R be a finitely generated left noetherian k-
algebra. Suppose that Spec(R) is a finite disjoint union of locally closed subsets X1, . . . , Xd with
each Xi, when endowed with the subspace topology, homeomorphic to the prime spectrum of an
affine commutative k-algebra. Then R satisfies the Dixmier-Moeglin equivalence.
The relevance of this theorem is seen in the fact that many quantum algebras have prime spectra
of this form [GL00], although generally in these cases work of Goodearl and Letzter [GL00] allows
one to deduce that the Dixmier-Moeglin equivalence holds. The stratification is a key ingredient in
the important work of Goodearl and Letzter in obtaining the Dixmier-Moeglin equivalence for many
classes of quantum algebras, where to obtain the Dixmier-Moeglin equivalence they use additional
information about the stratification coming from the rational action of an affine algebraic group on
the algebra. Theorem 1.2 again shows that for large base fields having an abstract stratification with
parts homeomorphic to affine schemes of finite type over k immediately gives the Dixmier-Moeglin
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
3
equivalence without having any underlying action of an algebraic group. Again, the counterexample
we produce to Theorem 1.1 when k is countable is a finitely generated noetherian algebra whose
prime spectrum is homeomorphic to the spectrum of a polynomial ring in one variable over k and
hence Theorem 1.2 does not hold if one relaxes the hypothesis that k be uncountable.
For the remainder of the paper we consider the Dixmier-Moeglin equivalence in three settings
where there is known to be a strong relationship between prime spectra; namely, Morita equivalence
(where equivalent rings have homeomorphic spectra), rings of the form eRe with e an idempotent,
in which there is a relationship between Spec(eRe) and an open subset of Spec(R), and tensor
products of rings (in which the spectrum shares a strong relationship with the cartesian product of
the spectra of the underlying rings). The first setting we consider is Morita equivalence of algebras,
which is a somewhat stronger property than having homeomorphic spectra. Two rings are Morita
equivalent if they have equivalent categories of left modules (which in turn gives that the categories
of right modules over these rings are equivalent). A ring theoretic property is called a Morita
invariant if whenever a ring has this property then every ring that is Morita equivalent to it also
has this property. Many ring theoretic properties are well known to be Morita invariants, including
being Artinian, noetherian, prime, semiprime, and having finite left or right global dimension
[MR01, Proposition 3.5.10]. We show that satisfying the Dixmier-Moeglin equivalence is a Morita
invariant. Much of this is already known and we stress that this is more of an observation, since it
is well known that there is an inclusion-preserving bijection between the prime spectra of Morita
equivalent rings that preserves primitivity and hence it is not much additional work to obtain that
the Dixmier-Moeglin equivalence is a Morita invariant. A coarser notion of equivalence is derived
equivalence and we do not know whether satisfying the Dixmier-Moeglin equivalence is a derived
invariant.
In addition to results concerning Morita equivalent rings, we prove that if R satisfies the
Dixmier-Moeglin equivalence and e is a nonzero idempotent then eRe satisfies the Dixmier-Moeglin
equivalence -- in this case it is well known that there is a continuous surjection from an open subset
of Spec(R) onto Spec(eRe) (see Remark 3.2) and we give an application to invariant subalgebras
of the form AH, where H is a finite-dimensional semisimple Hopf algebra that acts on an algebra
A. Finally, we show that if R and S are left noetherian k-algebras that both satisfy the Dixmier-
Moeglin equivalence then so does R ⊗k S under a hypothesis on the cardinality of the base field.
That is, the Dixmier-Moeglin equivalence is closed under the process of taking tensor products of
reasonably well behaved algebras. We now make these statements precise.
Theorem 1.3. Let k be a field and let R and S be left noetherian k-algebras. Then we have the
following:
(a) If R and S are Morita equivalent then R satisfies the Dixmier-Moeglin equivalence if and
only if S does;
(b) If R satisfies the Dixmier-Moeglin equivalence and e is a nonzero idempotent then eRe
satisfies the Dixmier-Moeglin equivalence;
(c) If R and S satisfy the Dixmier-Moeglin equivalence and if R ⊗k S is left noetherian and
satisfies the Nullstellensatz then R ⊗k S satisfies the Dixmier-Moeglin equivalence.
The outline of the paper is as follows. In §2 we prove Theorems 1.1 and 1.2 and give an example
to show that the conclusion to the statements of these theorems does not hold if we remove the
hypotheses on the cardinality of the base field. In §3 we prove that satisfying the Dixmier-Moeglin
equivalence is a Morita invariant and prove Theorem 1.3(b) and give an application of this result
to invariant subalgebras. Finally in §4 we prove Theorem 1.3(c).
4
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
2. Homeomorphic spectra and stratification
In this section, we prove Theorems 1.1 and 1.2 and show they do not hold without the hypotheses
on the cardinality of the base field. To do this, we need a few preliminary results. The following
lemma is closely related to results that are well known -- specifically, those concerning ideals in rings
obtained by extending scalars in centrally closed algebras. This lemma is somewhat stronger than
these results but requires a hypothesis that when one extends scalars one obtains a prime ring in
order to work.
Lemma 2.1. Let R be a prime noetherian k-algebra and suppose that (0) is rational. If F is an
extension of k such that Q(R) ⊗k F is a prime ring then every nonzero ideal of R ⊗k F contains
an element of the form r ⊗ 1 with r 6= 0.
Proof. Let I be a nonzero ideal of R ⊗k F and pick a nonzero element x ∈ I with x = Pd
i=1 ai ⊗ λi
with d minimal. We claim that d = 1. To see this, suppose that d > 1. We note that any element
of the form
mX
j=1
(cj ⊗ 1)x(dj ⊗ 1) =
dX
i=1
X
j
cjaidj
⊗ λi
is again in I. By minimality of d, a1 is nonzero; and since the two-sided ideal generated by a1
contains a regular element, by the above remarks, we may assume that a1 is regular. Then for
r ∈ R we have
x(ra1 ⊗ 1) − (a1r ⊗ 1)x =
(aira1 − a1rai) ⊗ λi.
dX
i=2
1 ai commutes with every r ∈ R. Then a−1
1 ai is algebraic over k. Then in Q(R) ⊗k F we may write x = (a1 ⊗ 1)(cid:16)Pd
By minimality of d we have λ2, . . . , λd are linearly independent over k and hence by minimality of d
we then have aira1−a1rai = 0 for all r ∈ R. In particular, taking r = 1 we see that [ai, a1] = 0 for all
i and so a−1
1 ai ∈ Z(Q(R)) and since (0) is rational we have
i=1 zi ⊗ λi(cid:17) with
that a−1
z1, . . . , zd ∈ Z(Q(R)). Let Z0 denote the finite extension of k generated by z1, . . . , zd. Then we have
x = (a1 ⊗ 1)y for some nonzero y ∈ Z0 ⊗k F . Since Q(R) ⊗k F is prime, we see that Z(Q(R)) ⊗k F
In
is an integral domain and since [Z0 : k] < ∞, we see that y is algebraic over F = k ⊗k F .
particular, there is a non-trivial relation ym(1 ⊗ cm) + ym−1(1 ⊗ cm−1) + ··· + (1 ⊗ c0) = 0 with
the ci ∈ F . Furthermore, we may assume c0 is nonzero since Z0 ⊗k F is an integral domain. In
particular, we may assume c0 = 1. Then by construction
mX
j=0
(a1 ⊗ 1)m−j xj(1 ⊗ cj) = 0
and since
mX
j=1
(a1 ⊗ 1)m−jxj(1 ⊗ cj) ∈ I,
we then see that am
1 ⊗ 1 ∈ I. Since a1 is regular, we have am
1 is nonzero and the result follows. (cid:3)
Remark 2.2. We note that this result need not hold if Q(R)⊗k F is not a prime ring. For example,
if R = F = Q(√2) and k = Q. Then R ⊗k F is not an integral domain and hence there is some
nonzero prime ideal P of R ⊗k F . But R is a field, so P ∩ (R ⊗ 1) is necessarily zero.
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
5
The next result is the key part of the proof of Theorem 1.1: it shows that rationality of prime
ideals can be captured purely in terms of the poset of prime ideals when the base field is sufficiently
large compared to the dimension of the algebra.
Lemma 2.3. Let k be a field and let R be a prime Noetherian k-algebra and suppose that dimk(R) <
k. Then the zero ideal is rational in R if and only if there is a set X of cardinality less than k
and a set of nonzero prime ideals {Px : x ∈ X} such that every nonzero prime ideal P of R contains
Px for some x ∈ X.
Proof. Notice that if k ≤ ℵ0 then dimk(R) < ∞ and so R is a prime Artinian k-algebra and hence
R is simple. In this case (0) is a rational, maximal ideal and the claim is vacuously true in this
case. Thus we assume henceforth that k is uncountable.
α,βrγ, where c(γ)
Now let k0 denote the prime subfield of k and let F denote the extension of k0 generated by T .
α,β is zero for all but finitely many
α,β : α, β, γ ∈ J}. Then notice that since for (α, β) ∈ J × J there are only finitely
α,β is nonzero we have injection from T → J × J × N. In particular,
First suppose that (0) is not rational. Then there is some z = ab−1 ∈ Z(Q(R)) that is not
algebraic over k, where a, b ∈ R are regular elements. Let B = {rα : α ∈ J} be a basis for R where
J is an index set with J < k. Since z is not in k, we have a and b are linearly independent over
k and we may assume without loss of generality that a, b ∈ B.
Then for α, β ∈ J, we have rαrβ = Pγ∈J c(γ)
γ ∈ J. Let T = {c(γ)
many values of γ for which c(γ)
T ≤ J2N < k, since k is uncountable and J < k.
Then the cardinality of F is at most max(T,ℵ0).
Notice that if we let R0 denote the F -subalgebra of R generated by {rα : α ∈ J} then by
construction R0 = V := Pα∈J F rα as an F -vector space since by construction any product of
elements in V is again in V since F contains T . Furthermore, by construction we have R0⊗F k ∼= R.
Notice that R0 is prime since R is prime and since R is a free R0-module, we have that there is
an inclusion-preserving injection from the set of left R0-modules to the set of left R-modules by
extension. In particular, R0 is left noetherian since R is. Then Q(R0)⊗F k is a localization of R and
since a, b ∈ R0 we have z ∈ Q(R0). Moreover, since dimF (R0) < k and since F < k and since
k is uncountable, we have R0 < k. Moreover, since every element of Q(R0) can be expressed in
the form sr−1 with s, r ∈ R0 we have that Q(R0) < k. Now for λ ∈ k, let zλ = z ⊗ 1− 1⊗ λ. We
note that by the Amitsur trick that the set of λ ∈ k for which zλ is a unit in Q(R) ⊗F k must have
cardinality strictly less than k. Explicitly, since Q(R0) ⊗F k has dimension strictly less than k if
Y := {λ ∈ k : zλ is a unit in Q(R) ⊗F k}
has cardinality k then there is necessarily a (finite) k-dependence of z−1
denominators in this dependence, we get that z is algebraic over k, which is a contradiction.
λ with λ ∈ Y ; after clearing
Since R is left noetherian, we see that Q(R0) ⊗F k is also left noetherian, as it is a localization
It follows from Jategaonkar's principal ideal theorem [Jat74] that for λ ∈ k \ Y we have
of R.
zλQ(R) ⊗F k is contained in a height one prime ideal Pλ. Then we let Qλ = Pλ ∩ R, which is a
height one prime ideal of R for λ ∈ k \ Y . Now suppose that we have a set X with X < k
and a set of nonzero prime ideals {Jx : x ∈ X} such that every nonzero prime ideal contains some
some Jx. Then since X < k and k \ Y = k we then have that there is some x ∈ X such
that Jx ⊆ Qλ for infinitely many λ ∈ k \ Y . But the Qλ are height one primes of R and hence
L := T Qλ = (0), since by Noether's theorem there is a finite set of primes that are minimal with
respect to containing L and if L is nonzero then since each Qλ is height one it must be minimal
over L. In particular, we get a contradiction. Thus we have shown one direction.
6
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
For the remaining direction, suppose that (0) is rational. Then as above we have that R ∼= R0⊗F k
where k is an extension of F , R0 is a prime left noetherian F -algebra and F,Q(R0) < k. By
Lemma 2.1, since (0) is rational and Q(R0) ⊗F k is a prime ring, every nonzero ideal of R contains
an element of the form a⊗ 1 with a a nonzero element of R0. For nonzero a ∈ R0, we let Xa denote
the set of prime ideals of R that contain a ⊗ 1. Then the nonzero prime ideals of R is equal to
the union of the Xa as a ranges over nonzero elements of R0. Now for each nonzero a ∈ R0 let
Ja = TQ∈Xa Q. Then Ja is a nonzero semiprime ideal of R since a ⊗ 1 ∈ Ja. Thus by Noether's
theorem, Ja is a finite intersection of prime ideals Qa,1 ∩ ··· ∩ Qa,na and every nonzero prime ideal
containing Ja contains Qa,i for some i. We now let U = {Qa,i : a ∈ R0 \ 0, i ∈ {1, . . . , na}}. Then
U is a set of nonzero prime ideals of R of cardinality at most R0 × N < k and by construction,
every nonzero prime ideal of R contains some prime ideal from U . This completes the proof.
(cid:3)
We are now able to give the proof of our main result.
Proof of Theorem 1.1. We may assume that R satisfies the Dixmier-Moeglin equivalence. Since
dimk(S) < k, we have that S satisfies the Nullstellensatz [BG01, II.7.16]. Hence by a result from
the book of Brown and Goodearl [BG01, II.7.15], we have the implications
P locally closed =⇒ P primitive =⇒ P rational
for P ∈ Spec(S). Hence it suffices to prove that a rational prime ideal of S is locally closed. Fix
an inclusion preserving bijection Ψ : Spec(R) → Spec(S) and let Q = Ψ−1(P ). Then the prime
ideals containing Q are precisely the prime ideals {Ψ−1(J) : J ) P}. Since P is rational, we have
by Lemma 2.3 that there is a set X with X < k and a set of prime ideals {Jx : x ∈ X} of prime
ideals of S that properly contain P such that every prime ideal of S that properly contains P must
contain Jx. Then since Ψ is an inclusion-preserving bijection we then see that every prime ideal
that properly contains Q must contain a prime ideal from {Ψ−1(Jx) : x ∈ X}. In particular, by
Lemma 2.3 we see that Q is a rational prime ideal of R. Then since R satisfies the Dixmier-Moeglin
equivalence, we have that Q is locally closed. Thus by Noether's theorem there is a finite set of
prime ideals Q1, . . . , Qd of R that properly contain Q such that every prime ideal that properly
contains Q must contain some Qi. But then every prime ideal that properly contains P must
contain Ψ(Qi) for some i ∈ {1, . . . , d}. But this means that P is locally closed in Spec(S). This
completes the proof.
(cid:3)
We are able to prove Theorem 1.2 now.
Proof of Theorem 1.2. Again since dimk(R) < k we have that R satisfies the Nullstellensatz and
so it suffices to prove that a rational prime ideal is locally closed in Spec(R). Let P be rational.
Then by Lemma 2.3, there is a countable set (possibly finite or empty) of prime ideals Q1, Q2, . . .
that properly contain P such that every prime ideal properly containing P contains Qi for some
i. If {Qi : i ≥ 1} is finite then P is locally closed and we are done. Thus we may assume that the
set {Qi : i ≥ 1} is infinite. We have Spec(R) = X1 ⊔ X2 ⊔ ··· ⊔ Xd with each Xi homeomorphic to
Spec(Ti) for some finitely generated commutative k-algebra Ti. For each j ∈ {1, . . . , d} we let
Lj := \
{i : Qi∈Xj}
Qi.
Then for j ∈ {1, . . . , d} we either have Lj = P or Lj properly contains P in which case there is a
finite set Sj of primes that are minimal with respect to containing Lj. If Lj properly contains P
for j = 1, . . . , d then every prime ideal properly containing P contains some prime ideal from the
finite set Sd
j=1 Sj and hence P is locally closed and we are done. Alternatively, we have Lj = P for
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
7
some j. In this case, P ∈ Xj since Xj is locally closed and we also have that P is not locally closed
in Xj since Lj = P . Thus since Tj satisfies the Dixmier-Moeglin equivalence, we see by Lemma
2.3 that there are uncountably many height one primes in Tj/P (where we now identify P with its
image in Spec(Tj) ∼= Xj) and in particular, there are uncountably many height one prime ideals in
R/P and so P is not a rational prime ideal of R by Lemma 2.3, a contradiction.
(cid:3)
Remark 2.4. We note that in the proof of Theorem 1.2 the only place that Ti is affine commutative
is used is to ensure that the Ti satisfy the Dixmier-Moeglin equivalence and so one can relax the
statement of the theorem to only require that the Ti be k-affine algebras that satisfy the Dixmier-
Moeglin equivalence.
We now show that the conclusions to the statements of Theorems 1.1 and 1.2 do not hold if we
relax the hypotheses on the cardinality of the base field. We note that Lorenz [Lor77] gives an
example of an algebra that does not satisfy the Dixmier-Moeglin equivalence. We are unable to use
his example to produce a counterexample to the statements of Theorems 1.1 and 1.2 when k is a
countable field, but by modifying his construction appropriately we can construct a counterexample.
Thus the following example should be seen as heavily drawing inspiration from the construction of
Lorenz. We suspect, in fact, that the example of Lorenz does not have a prime spectrum that is
homeomorphic to that of an algebra that satisfies the Dixmier-Moeglin equivalence.
Example 2.5. Let k = ¯Q. Then there exists a finitely generated infinite-dimensional prime noe-
therian k-algebra R such that R does not satisfy the Dixmier-Moeglin equivalence and Spec(R) is
homeomorphic to Spec(k[t]). Thus neither Theorem 1.1 nor Theorem 1.2 hold when one removes
the hypotheses on the cardinality of the base field.
To do this we require a few basic results. We begin our construction by taking B = k[x±1, y±1]
and letting σ be the k-algebra automorphism of B given by x 7→ x5y4 and y 7→ xy. We note that
σ is an automorphism since it has inverse given by x 7→ xy−4, y 7→ x−1y5.
Lemma 2.6. The algebra B has no nonzero proper principal σ-invariant ideals.
Proof. Suppose that (f (x, y)) is σ-invariant with f 6= 0. Then σ(f ) = γxpyqf for some nonzero
γ and some integers p and q. We write f = P ci,jxiyj and since f is a non-unit and nonzero, we
have that the set of (i, j) such that ci,j is nonzero has at least two elements. By multiplying f by
a unit we may assume that c0,0 6= 0. Then we have
σ(f ) = X ci,jx5i+jy4i+j = X ci,jγxp+iyq+j.
So now let M : Z2 → Z2 be the Z-linear map given by M (m, n) = (5m + n, 4m + n) and let
Φ : Z2 → Z2 be the Z-affine linear map given by Φ(m, n) = (5m+n−p, 4m+n−q) = M (m, n)−(p, q).
Let T = {(0, 0) = (i1, j1), . . . , (id, jd)} denote the set of pairs (i, j) for which ci,j 6= 0. Then by
construction the orbit of (0, 0) under Φ must be contained in T and hence is finite. But notice
that Φn(0, 0) = (I + M + M 2 + ··· + M n−1)(−p,−q), which is infinite unless p = q = 0. But
now this means that the set of values taken by (M − I) ◦ Φn(0, 0) − (p, q) = M n(−p,−q) must be
finite. But M is invertible and has eigenvalues that are not roots of unity and so this only occurs
if (p, q) = (0, 0). Thus Ψ = M . Now since f is a non-unit there is some (i, j) 6= (0, 0) such that
(i, j) ∈ T . Again, the orbit of (i, j) under Ψ = M must lie in T and hence must be finite. But this
is impossible since (i, j) 6= (0, 0) and M is invertible and has no eigenvalues that are roots of unity.
The result follows.
(cid:3)
Now we take A = B[u±1] and we extend σ to A by declaring that σ(u) = 2u. We now characterize
the σ-prime ideals of A.
8
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
Lemma 2.7. Let I be a nonzero σ-prime ideal of A. Then I = JA, for some σ-prime ideal J of
B having finite codimension in B.
Proof. We first claim that I intersects B non-trivially. To see this, let a = Pm
i=0 biui be a nonzero
element of I with m minimal. By minimality of m we have b0 6= 0. We claim that m = 0. To see
this, observe that σ(a) = Pm
j=0 σ(bi)2iui. So
m−1X
2mσ(bm)a − bmσ(a) =
(2mσ(bm)bj − 2jbmσ(bj))uj ∈ I.
j=0
By minimality of m we see that
2mσ(bm)b0 = bmσ(b0),
and so if we set c = b0/bm then we have σ(c) = 2mc. Now we write c = r/s with r, s relatively
prime elements of B. Then σ(c) = σ(r)/σ(s) and so 2mrσ(s) = σ(r)s. Since B is a UFD and r
and s are coprime we see that σ(s) = u1s and σ(r) = u2r for some units u1, u2 of B. But this
means that (s) and (r) are nonzero σ-invariant ideals of B and by Lemma 2.6 we then see that
s and r are units and so c is a unit in B. But then c = γxpyq and it is straightforward to see
that σ(c) 6= 2mc has no solutions of this form unless m = 0 and γ is constant.
It follows that
J := I ∩ B is nonzero and since I is σ-prime, J is a σ-prime ideal of B. Since J is σ-prime, we
have J = Q∩ σ(Q) ∩ ··· ∩ σn−1(Q) for some nonzero prime ideal Q of B with σn(Q) = Q. If Q has
height one then Q is principal since B is a UFD and hence J is principal. But this cannot occur
by Lemma 2.6. Thus Q has height at least 2 and since B has Krull dimension 2, we see that Q is
maximal and so J is of finite codimension in B as required. Now we claim that I = JA. To see
this, observe that QA is a height two prime ideal of A and if I is not JA then there must be a
σ-periodic prime ideal P strictly containing QA such that I is the intersection of the σ-orbit of P .
But if P properly contains QA then it must have height at least three and hence must be maximal
and so u − γ ∈ P for some nonzero γ by the Nullstellensatz. But σn(u − γ) = 2nu − γ and since
σn(Q) = Q for some n ≥ 1, we then get that (2n − 1)γ ∈ Q, a contradiction.
Proof of Example 2.5. Let S = A[z±1; σ]. Then we claim that A does not satisfy the Dixmier-
Moeglin equivalence and Spec(S) and Spec(k[t]) are homeomorphic. To see this, first notice that if
α and β are roots of unity, then the point (α, β) has finite orbit under σ and hence the intersection
of the prime ideals in the σ-orbit of (x− α, y − β) is a σ-prime ideal of A. In particular, since there
are infinitely many ordered pairs of roots of unity, we see that there is an infinite set of nonzero
σ-prime ideals of A and by Lemma 2.7 they are all maximal σ-prime ideals of A. Notice from the
above that if I is a nonzero σ-prime ideal of A then A/I ∼= (B/Q)[u±1] and since σ has infinite order
on (B/Q)[u±1] and I is a maximal σ-prime ideal of A we see that (A/I)[z±1; σ] is simple. Thus we
have shown that (0) is not locally closed and all nonzero prime ideals of S are maximal. Moreover,
since S is noetherian we have that any infinite set of maximal ideals is Zariski dense. Thus Spec(S)
is homeomorphic to Spec(k[t]). Thus to finish the proof it suffices to show that (0) is rational.
Notice this occurs, if and only if there is a non-constant rational map f : Spec(A) 99K P1 such that
f ◦ σ = f [BG18, Lemma 3.5]. Notice that we can write f as P (x, y, u)/Q(x, y, u) where P, Q ∈ A
are coprime. Then f ◦ σ = f gives that P (x5y4, xy, 2u)/Q(x5y4, xy, 2u) = P (x, y, u)/Q(x, y, u) and
so
(cid:3)
P (x5y4, xy, 2u)Q(x, y, u) = P (x, y, u)Q(x5y4, xy, 2u).
Then since P and Q are coprime and A is a UFD, we have (P (x5y4, xy, 2u)) = (P (x, y, u)) and
(Q(x5y4, xy, 2u)) = (Q(x, y, u)). By considering the degree in u and using the fact that A∗ =
k∗hx±1, y±1, u±1i, we see that σ(P ) = γxpyqP and σ(Q) = δxsytQ with γ, δ ∈ k∗ and p, q, s, t ∈ Z.
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
9
Now we consider P as a polynomial in u and write P = P Piui. Then σ(P ) = γxpyqP gives
σ(Pi)2i = γxpyqPi and so each Pi is a σ-invariant prime ideal of P and hence Pi is either 0 or
a unit in B by Lemma 2.6. But the units of B are of the form k∗xpyq and it is straightforward
to check that there are no unit solutions to σ(Pi)2i = γxpyqPi if i > 0. Hence P = P0 ∈ B and
similarly Q = Q0 ∈ B and from the above remarks we have that P0 and Q0 are units in B and
so f is a unit of B. Thus f = γxpyq for some nonzero γ and some integers p and q. But now
f ◦ σ = γx5p+qy4p+q = γxpyq and so p = q = 0 and so we see f is necessarily constant and thus P
is rational. Thus S does not satisfy the Dixmier-Moeglin equivalence as claimed.
(cid:3)
3. Morita Equivalence and corners
In this section we prove that satisfying the Dixmier-Moeglin equivalence is a Morita invariant.
Large parts of this result were already well known and the result itself is not too difficult, but since
it doesn't appear in the literature and since there are non-trivial consequences of this result, we
find it useful to record this fact. We first make the following well known remark, which shows that
Morita equivalence is a much stronger condition than assuming homeomorphic spectra.
Remark 3.1. Let R and S be Morita equivalent algebras. Then there is a homeomorphism Ψ :
Spec(R) → Spec(S) with the property that Ψ(P ) is primitive if and only if P is primitive.
Proof. This is essentially given by Theorem 3.5.9 (i) of McConnell and Robson [MR01]. Let SMR
be a progenerator. Then the map Ψ from ideals of R to ideals of S given by I 7→ M IM ∗ gives a
semigroup isomorphism between the ideals of R and those of S that induces a bijection from Spec(R)
to Spec(S) that preserves primitivity [MR01, Theorem 3.5.9]. We note that Ψ : Spec(R) → Spec(S)
and Ψ−1 : Spec(S) → Spec(R) are continuous since the collection of prime ideals in R containing
I is mapped to the collection of prime ideals in S that contain Ψ(I) and conversely
Ψ−1({P ∈ Spec(S) : P ⊇ J}) = {Q ∈ Spec(R) : Q ⊇ Ψ−1(J)}.
(cid:3)
We now show that satisfying the Dixmier-Moeglin equivalence is a Morita invariant.
Proof of Theorem 1.3(a). Let SMR be a progenerator and let Ψ : Spec(R) → Spec(S) be the
homeomorphism described in Remark 3.1. Let P be a prime ideal of R. Then since being locally
closed is a topological property we see that P is locally closed in Spec(R) if and only if Ψ(P ) is
locally closed in Spec(S). By Remark 3.1 we have that P is primitive if and only if Ψ(P ) is primitive.
Finally to see P is rational if and only if Ψ(P ) is rational, suppose that P is a rational prime ideal
of R. Then by McConnell and Robson [MR01, Theorem 3.5.9 (ii)] we have R/P and S/Ψ(P ) are
Morita equivalent and hence Q(R/P ) and Q(S/Ψ(P )) are Morita equivalent [MR01, Proposition
3.6.9]. Thus there is a k-algebra isomorphism Z(Q(R/P )) ∼= Z(Q(S/Ψ(P )) (cf. [MR01, Theorem
3.59(iii)]), and so P is rational if and only if Ψ(P ) is rational.
(cid:3)
We note that if R is a ring and e is an idempotent of R then eRe is not in general Morita
equivalent to R; one typically requires that e be full; i.e., that ReR = R. So in general the Dixmier-
Moeglin equivalence being satisfied by R is not equivalent to being satisfied by eRe. For example,
if R is a ring that does satisfy the Dixmier-Moeglin equivalence and S does then T = R × S does
not, but R is of the from eT e for an idempotent of T . On the other hand, we are able to show that
if R satisfies the Dixmier-Moeglin equivalence then eRe must too. To do this, we begin with an
easy remark, which is folklore, although we are unable to find a reference so we give a proof.
10
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
Remark 3.2. Let R be a ring with nonzero idempotent e. Let U denote the open subset of Spec(R)
consisting of prime ideals for which e 6∈ P . Then there is a continuous surjection from U (endowed
with the subspace topology) to Spec(eRe) given by P ∈ U 7→ eP e.
Proof. If P ∈ Spec(R) and e 6∈ P then eP e is a proper ideal of eRe and the fact that P is prime in R
immediately gives that eP e is prime in eRe. Now to get surjectivity of the map let Q ∈ Spec(eRe)
and let P denote the sum of all ideals I such that eIe ⊆ Q. Then this is a non-empty sum since
RQR has the property that eRQRe = Q. Moreover, P is a proper ideal and eP e = Q, since
eP e ⊆ Q by construction and P ⊇ RQR so eP e ⊇ Q. Finally, to check that P is prime, we note
that if J1 and J2 are ideals of R with J1J2 ⊆ P . Then eJ1J2e ⊆ Q and so (eJ1e)(eJ2e) ⊆ Q. But
since Q is prime, we then have either eJ1e ⊆ P or eJ2e ⊆ P and so either J1 or J2 is contained
in P by definition and so P is prime. Finally, to see continuity observe that the preimage of the
set of prime ideals of eRe that contains an ideal I is precisely the prime ideals of U that contain
RIR.
(cid:3)
Proof of Theorem 1.3(b). Let Q be a prime ideal of eRe and let P be a prime ideal of R such that
eRe = P . Then we replace R by R/P and assume that R and eRe are prime and that P and Q
are zero. First, we have that (0) is a primitive ideal of R if and only if (0) is a primitive ideal of
eRe by a result of Lanski, Resco, and Small [LRS79, Theorem 1]. Next observe that if (0) is locally
closed in Spec(R) then the intersection of the nonzero prime ideals of R is a nonzero ideal I. Since
every nonzero prime ideal of eRe is of the form eP e for some nonzero prime ideal P of R we see
that the intersection of the nonzero prime ideals of eRe contains eIe, which is nonzero since R is
prime. Thus (0) is locally closed in Spec(eRe) if (0) is locally closed in Spec(R). Conversely, if (0)
is not locally closed in Spec(R) then
\ P = (0),
where we take the intersection of all nonzero prime ideals P of R. But since eP e is either a nonzero
prime ideal of eRe or is equal to eRe we see that the intersection of the nonzero prime ideals of
eRe is contained in the intersection TP 6=0 eP e = e(TP 6=0 P )e = (0) and so (0) is not locally closed
in Spec(eRe). Finally, (0) is a rational prime of R if and only if the centre of Q(R) is an algebraic
extension of k. But now e becomes a full idempotent in Q(R) since the ideal ReR contains a regular
element and hence Q(R) and eQ(R)e are Morita equivalent and since Q(R) is prime Artinian, so is
eQ(R)e [MR01, Proposition 3.5.10]. In particular, since Q(eRe) is a localization of eQ(R)e we see
that Q(eRe) = eQ(R)e and so (0) is rational in R if and only if (0) is rational in eRe (cf. [MR01,
Theorem 3.59(iii)]). Finally since every prime ideal of eRe is of the form eP e for some prime ideal
of R we see that if R satisfies the Dixmier-Moeglin equivalence then so does eRe.
(cid:3)
Remark 3.3. As noted earlier, the converse of this result does not hold: if R and S are algebras and
R satisfies the Dixmier-Moeglin equivalence and S does not, then T := R × S does not satisfy the
Dixmier-Moeglin equivalence since S is a homomorphic image. On the other hand R = eT e with
e = (1, 0). The proof in the other direction fails because the argument gives no information about
primes P ∈ Spec(R) with e ∈ P . Thus the converse only holds in general if we know e is full; i.e.,
R = ReR. But in this case R and eRe are Morita equivalent and the result is already covered by
Theorem 1.3(a).
We now give an application of this result. We recall that if A is a k-algebra and H is a Hopf
k-algebra, then H acts on A if there is there is a k-bilinear map φ : H × A → A (where we let h · a
denote φ(h, a)) such that for a, b ∈ A we have h · (ab) = Pi(fi · a)(gi · b), where ∆(h) = P fi ⊗ gi,
and we have h· 1 = ǫ(h). Given a Hopf algebra H acting on A, we can then construct the invariant
subalgebra AH = {a ∈ A : h · a = ǫ(h)a for all h ∈ H}.
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
11
Corollary 3.4. Let k be a field, let A be a left noetherian k-algebra of finite Gelfand-Kirillov
dimension, and let H be a finite-dimensional semisimple Hopf algebra that acts on A. If A satisfies
the Dixmier-Moeglin equivalence, then the invariant subalgebra AH satisfies the Dixmier-Moeglin
equivalence.
Proof. Since H is finite-dimensional, the smash-product algebra A#H is a finite free left and right
A-module and hence left noetherian and A#H satisfies the Dixmier-Moeglin equivalence by [Let89].
Now since H is semisimple, the trace map from A to AH is surjective and so [Mon93, Lemma 4.3.4]
gives that e(A#H)e ∼= AH for some nonzero idempotent e ∈ A#H. Theorem 1.3(b) now gives that
AH satisfies the Dixmier-Moeglin equivalence.
(cid:3)
Remark 3.5. We observe that if one follows the proof then we see that the semisimple hypothesis can
be replaced by the weaker condition that the trace map t : A → AH being surjective in Corollary
3.4.
4. Dixmier-Moeglin equivalence and Tensor Products
In this section, we prove that if R and S are algebras satisfying the Dixmier-Moeglin equivalence
then R ⊗k S does too under a hypothesis on the base field. We note that there are some subtleties
that arise in general since a tensor product of prime rings need not be prime, and so we find
obtaining the result without some sort of hypothesis that gives the Nullstellensatz to be difficult.
Proof of Theorem 1.3(c). Since R⊗k S satisfies the Nullstellensatz, it suffices to prove that rational
prime ideals are locally closed.
Let P ∈ Spec(R ⊗k S). Let Q = {a ∈ R : a ⊗ 1 ∈ P}. Then Q is a prime ideal of R since if
a1, a2 ∈ R are such that a1Ra2 ⊆ Q then (a1 ⊗ 1)(R ⊗ S)(a2 ⊗ 1) ⊆ Q ⊗ S ⊆ P and so a1 ⊗ 1 or
a2 ⊗ 1 ∈ P , which then gives a1 or a2 is in Q.
Then (R⊗k S)/P is a homomorphic image of (R/Q)⊗ S and so without loss of generality we may
replace R by R/Q and assume that R is prime and that (R⊗1)∩P = (0). Similarly, we may assume
that S is prime and that (1 ⊗ S) ∩ P = (0). Notice that if a ⊗ b ∈ P with a ∈ R and b ∈ S then for
x ∈ R and y ∈ S we have ax ⊗ yb = (1 ⊗ y)(a ⊗ b)(x ⊗ 1) ∈ P . Hence (a ⊗ 1)(R ⊗ S)(1 ⊗ b) ⊆ P
and thus by the above reduction we have either a = 0 or b = 0.
Now suppose that P is rational. We show that P is locally closed. To see this, let U denote the
regular elements of R and let T denote the regular elements of S. Then (U ⊗ T ) is an Ore set of
elements of R⊗S and by the above remarks the prime P survives in the localization (U ⊗T )−1R⊗S.
In particular Q(R) ⊗k Q(S) is a subalgebra of Q((R ⊗k S)/P ) and Z(Q(R)) ⊗k Z(Q(S)) is a
subalgebra of Q((R ⊗k S)/P ).
In particular, since P is rational, Z(Q(R)) ⊗k Z(Q(S)) is an algebraic extension of k and thus
Z(Q(R)) and Z(Q(S)) are algebraic extensions of k and so (0) is a rational prime of R and (0) is
a rational prime of S. By hypothesis, we then have that (0) is a locally closed ideal of R and (0) is
a locally closed ideal of S.
To see that P is locally closed, we let X denote the set of primes in Spec(R ⊗k S) that properly
contain P . We let X1 denote the subset of primes Q ∈ X such that Q ∩ (R ⊗ 1) 6= (0) and we let
X2 denote the set of primes Q ∈ X such that Q ∩ (1 ⊗ S) 6= (0). Arguing as above, we see that for
Q ∈ X1, Q ∩ (R ⊗ 1) is a nonzero prime ideal of R. Since (0) is locally closed, we then have that
there is some nonzero a ∈ R such that a ⊗ 1 ∈ ∩Q∈X1Q. Similarly, there is some nonzero b ∈ S
such that 1 ⊗ b ∈ ∩Q∈X2Q.
12
JASON P. BELL, XINGTING WANG, AND DANIEL YEE
Finally, consider the primes Q such that Q ∩ (R ⊗ 1) = (0) and Q ∩ (1 ⊗ S) = (0). Then
as before, these primes have trivial intersection with the Ore set U ⊗ T , and so the primes in
X \ (X1 ∪ X2) survive in the localization Q(R)⊗k Q(S)/eP , where eP = (U ⊗ T )−1P . We claim that
eP is maximal. To see this, let I be an ideal of Q(R) ⊗k Q(S) that properly contains eP . Then we
pick x = Pd
i=1 ai ⊗ bi ∈ I \ eP with a1, . . . , ad, b1, . . . , bd nonzero and d minimal. Then d > 1 since
a1 ⊗ b1 is a unit in Q(R) ⊗k Q(S).
1 and assume that a1 = b1 = 1. Then for c ∈ Q(R) we
have [x, c ⊗ 1] = Pd
i=2[a2, c] ⊗ b2 ∈ I and so by minimality of d we see that it is in eP ; similarly,
[x, 1 ⊗ c′] ∈ eP for c′ ∈ Q(S). Thus x is central mod eP . But
Then we may right-multiply by a−1
1 ⊗ b−1
Z(Q(R) ⊗ Q(S)/eP ) ⊆ Z(Q(Q(R) ⊗k Q(S)/eP )) = Z(Q(R ⊗ S)/P ),
which is an algebraic extension of k since P is rational. Hence Z(Q(R) ⊗k Q(S)/eP ) is a finite-
dimensional k-algebra that is a domain and hence it is a field. Thus x is a unit, which gives that
I = Q(R) ⊗k Q(S) and so we obtain the desired result.
Thus X = X1 ∪ X2 and so we see that 0 6= a ⊗ b ∈ TQ∈X Q and so P is locally closed. Thus we
have obtained the result when dimk(R), dimk(S) < k.
(cid:3)
It would be interesting to remove the hypothesis that R ⊗k S satisfies the Nullstellensatz in
the proof of Theorem 1.3(c). We note that a tensor product of noetherian algebras need not be
noetherian; for example, if R = S = k(t1, t2, . . .) then R and S are noetherian but R ⊗k S is not.
Remark 4.1. Throughout we have taken rationality of a prime ideal P of a left noetherian algebra R
to mean that Z(Q(R/P )) is an algebraic extension of the base field. If one instead takes rationality
to mean that Z(Q(R/P )) is a finite extension of the base field then all results go through, essentially
verbatim.
References
[ABR12] G. Abrams, J. P. Bell, and K. M Rangaswamy, The Dixmier-Moeglin equivalence for Leavitt path algebras.
[BG18]
Algebr. Represent. Theory 15 (2012), no. 3, 407 -- 425.
J. P. Bell and D. Ghioca, Periodic subvarieties of semiabelian varieties and annihilators of irreducible
representations. Preprint, available online at ArXiv:1806.11054.
[BLLM17] J. Bell, S. Launois, O. Le´on S´anchez, and R. Moosa, Poisson algebras via model theory and differential-
[BSM18]
[BL14]
[BRS10]
algebraic geometry. J. Eur. Math. Soc. (JEMS) 19 (2017), no. 7, 2019 -- 2049.
J. Bell, Jason; O. Le´on S´anchez, and R. Moosa, D-groups and the Dixmier-Moeglin equivalence. Algebra
Number Theory 12 (2018), no. 2, 343 -- 378.
J. P. Bell and W. H. Leung, The Dixmier-Moeglin equivalence for cocommutative Hopf algebras of finite
Gelfand-Kirillov dimension. Algebr. Represent. Theory 17 (2014), no. 6, 1843 -- 1852.
J. Bell, D. Rogalski, and S. J. Sierra, The Dixmier-Moeglin equivalence for twisted homogeneous coordi-
nate rings. Israel J. Math. 180 (2010), 461 -- 507.
[BCM17] K. Brown, P. A.A.B. Carvalho, and J. Matczuk, Simple modules and their essential extensions for skew
[BG01]
[Dix77]
[GL00]
[GZ10]
[Irv79]
polynomial rings. Preprint, available online at ArXiv:1705.06596.
K. A. Brown and K. R. Goodearl, Lectures on algebraic quantum groups. Advanced Courses in Mathe-
matics. CRM Barcelona. Birkhauser Verlag, Basel, 2002.
J. Dixmier, Id´eaux primitifs dans les alg`ebres enveloppantes. J. Algebra 48 (1977), 96 -- 112.
K. R. Goodearl and E. S. Letzter, The Dixmier-Moeglin equivalence in quantum coordinate rings and
quantized Weyl algebras. Trans. Amer. Math. Soc. 352 (2000), no. 3, 1381 -- 1403.
K. R. Goodearl and J. J. Zhang, Noetherian Hopf algebra domains of Gelfand-Kirillov dimension two. J.
Algebra 324 (2010), no. 11, 3131 -- 3168.
R. S. Irving, Noetherian algebras and nullstellensatz. S´eminaire d'Alg`ebre Paul Dubreil 31 `me ann´ee (Paris,
1977 -- 1978), pp. 80 -- 87, Lecture Notes in Math., 740, Springer, Berlin, 1979.
DIXMIER-MOEGLIN EQUIVALENCE AND MORITA EQUIVALENCE
13
[Jat74]
[LRS79]
[Let89]
[Lor09]
[Lor08]
[Lor77]
[MR01]
[Moe80]
[Mon93]
[Von96]
[Zal71]
A. V. Jategaonkar, Relative Krull dimension and prime ideals in right Noetherian rings. Comm. Algebra
2 (1974), 429 -- 468.
C. Lanski, R. Resco, and L. Small, On the primitivity of prime rings. J. Algebra 59 (1979), no. 2, 395 -- 398.
E. Letzter, Primitive ideals in finite extensions of Noetherian rings. J. London Math. Soc. (2) 39 (1989),
no. 3, 427 -- 435.
M. Lorenz, Algebraic group actions on noncommutative spectra. Transform. Groups 14 (2009), no. 3,
649 -- 675.
M. Lorenz, Group actions and rational ideals. Algebra Number Theory 2 (2008), no. 4, 467 -- 499.
M. Lorenz, Primitive ideals of group algebras of supersoluble groups. Math. Ann. 225 (1977), no. 2,
115 -- 122.
J. McConnell and J. Robson, Noncommutative Noetherian rings. With the cooperation of L. W. Small.
Revised edition. Graduate Studies in Mathematics, 30. American Mathematical Society, Providence, RI,
2001.
C. Moeglin, Id´eaux bilat`eres des alg`ebres enveloppantes. Bull. Soc. Math. France 108 (1980), 143 -- 186.
S. Montgomery, Hopf algebras and their actions on rings. CBMS Regional Conference Series in Mathe-
matics, 82. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the
American Mathematical Society, Providence, RI, 1993.
N. Vonessen, Actions of algebraic groups on the spectrum of rational ideals. J. Algebra 182 (1996), no. 2,
383 -- 400.
A. E. Zalesskiı, The irreducible representations of finitely generated nilpotent groups without torsion.
Mat. Zametki 9 (1971), 199 -- 210.
Department of Pure Mathematics, University of Waterloo, 200 W University Ave. Waterloo, ON
N2L 3G1, Canada
E-mail address: [email protected]
Department of Mathematics, Howard University, 2400 Sixth St. NW, Washington DC, 20059, USA
E-mail address: [email protected]
Department of Mathematics, Bradley University, 1501 W Bradley Ave., Peoria, IL 61625, USA
E-mail address: [email protected]
|
1912.11386 | 1 | 1912 | 2019-12-23T11:56:22 | On general linear groups over exchange rings | [
"math.RA"
] | Let $R$ be an exchange ring. We prove that the relative elementary subgroups $E_n(R,I)$ are normal in the general linear group $GL_n(R)$ if $n\geq 1$ and that the standard commutator formula $E_n(R,I)=[E_n(R),E_n(R,I)]=[E_n(R),C_n(R,I)]$ holds if $n\geq 3$. Moreover, we classify the subgroups of $GL_n(R)$ that are normalised by the elementary subgroup $E_n(R)$ in the case $n\geq 3$. | math.RA | math |
On general linear groups over exchange rings
Raimund Preusser
Abstract
Let R be an exchange ring. We prove that the relative elementary subgroups
En(R, I) are normal in the general linear group GLn(R) if n ≥ 1 and that the
standard commutator formula En(R, I) = [En(R), En(R, I)] = [En(R), Cn(R, I)]
holds if n ≥ 3. Moreover, we classify the subgroups of GLn(R) that are nor-
malised by the elementary subgroup En(R) in the case n ≥ 3.
1
Introduction
Let R be a ring. A left R-module M is said to have the exchange property if for any
module X and decompositions
X = M ′ ⊕ Y = M
Ni
i∈I
where M ′ ∼= M, there exists submodules N ′
i ⊆ Ni for each i such that
X = M ′ ⊕ (M
N ′
i ).
i∈I
R is called an exchange ring if RR has the exchange property. W. Nicholson [5] showed
that R is an exchange ring iff for any x ∈ R there is an idempotent e ∈ xR such that
1 − e ∈ (1 − x)R. It follows that any exchange ring R has property (1) below, see [5,
Proposition 1.11].
If m ∈ N and x1, . . . , xm ∈ R such that x1 + · · · + xm = 1, then there are orthogonal
idempotents ei ∈ xiR (1 ≤ i ≤ m) such that e1 + · · · + em = 1.
(1)
The class of exchange rings is quite large, it contains all semiperfect rings [5],
semiregular rings [5], local rings [5], clean rings [5], purely infinite simple rings [2] and
(not necessarily simple) purely infinite Leavitt path algebras [1, Corollary 3.8.19].
The main results of this paper are the following three theorems, where GLn(R)
denotes the general linear group (of degree n) over R, En(R) the elementary subgroup,
En(R, I) the elementary subgroup of level I and Cn(R, I) the full congruence subgroup
of level I.
2010 Mathematics Subject Classification. 15A24, 20G35.
Keywords and phrases. general linear groups, exchange rings, normal subgroups.
1
Theorem 1. Suppose that R is an exchange ring and I ⊳ R an ideal. Then En(R, I)
is a normal subgroup of GLn(R).
Theorem 2. Suppose that R is an exchange ring and n ≥ 3. Then
En(R, I) = [En(R), En(R, I)] = [En(R), Cn(R, I)]
for any ideal I ⊳ R.
Theorem 3. Suppose that R is an exchange ring and n ≥ 3. Let H be a subgroup of
GLn(R). Then H is normalised by En(R) iff there is an ideal I ⊳ R such that
Moreover, the ideal I is uniquely determined.
En(R, I) ⊆ H ⊆ Cn(R, I).
Similar results for other classes of rings (including the class of commutative rings)
were obtained in [3], [11], [4], [8], [9] and [10].
The rest of the paper is organised as follows. In Section 2 we recall some standard
notation which is used throughout the paper. In Section 3 we recall the definitions
of the general linear group GLn(R) and some important subgroups. In Section 4 we
prove Theorem 1, in Section 5 Theorem 2 and in Section 6 Theorem 3.
2 Notation
N denotes the set of positive integers. If G is a group and g, h ∈ G, we let gh := h−1gh
and [g, h] := ghg−1h−1. If H and K are subgroups of G, we denote by [H, K] the
subgroup of G generated by the set {[h, k] h ∈ H, k ∈ K}. By a ring we mean
an associative ring with 1 6= 0. By an ideal of a ring we mean a twosided ideal. To
denote that I is an ideal of a ring R we write I ⊳ R.
Throughout the paper R denotes a ring and n ∈ N a positive integer. We denote
by Rn the set of all columns u = (u1, . . . , un)t with entries in R and by nR the set
of all rows v = (v1, . . . , vn) with entries in R. Furthermore, the set of all n × n
matrices over R is denoted by Mn(R). The identity matrix in Mn(R) is denoted by e
and the matrix with a one at position (i, j) and zeros elsewhere is denoted by eij. If
σ ∈ Mn(R), we denote the the entry of σ at position (i, j) by σij. We denote the i-th
row of σ by σi∗ and its j-th column by σ∗j. If σ ∈ Mn(R) is invertible, we denote the
entry of σ−1 at position (i, j) by σ′
i∗ and the j-th column
of σ−1 by σ′
ij, the i-th row of σ−1 by σ′
∗j.
3 The general linear group
We shall recall the definitions of the general linear group GLn(R) and its elementary
subgroup En(R). For an ideal I ⊳R we shall recall the definition of the preelementary
subgroup En(I) of level I, the elementary subgroup En(R, I) of level I and the full
congruence subgroup Cn(R, I) of level I.
2
3.1 The general linear group and its elementary subgroup
Definition 4. The group GLn(R) consisting of all invertible elements of Mn(R) is
called the general linear group of degree n over R.
Definition 5. Let x ∈ R and i, j ∈ {1, . . . , n} such that i 6= j. Then the matrix
tij(x) := e + xeij ∈ GLn(R) is called an elementary transvection. The subgroup
En(R) of GLn(R) generated by the elementary transvections is called the elementary
subgroup.
Lemma 6. The relations
tij(x)tij(y) = tij(x + y),
[tij(x), thk(y)] = e and
[tij(x), tjk(y)] = tik(xy)
(R1)
(R2)
(R3)
hold where i 6= k, j 6= h in (R2) and i 6= k in (R3).
Proof. Straightforward computation.
Definition 7. Let i, j ∈ {1, . . . , n} such that i 6= j. Then the matrix pij := e +
eij − eji − eii − ejj = tij(1)tji(−1)tij(1) ∈ En(R) is called a generalised permutation
matrix. It is easy to show that p−1
ij = pji. The subgroup of En(R) generated by the
generalised permutation matrices is denoted by Pn(R).
Lemma 8. Suppose n ≥ 3. Let x ∈ R and i, j, i′, j′ ∈ {1, . . . , n} such that i 6= j and
i′ 6= j′. Then there is a τ ∈ Pn(R) such that tij(x)τ = ti′j′(x).
Proof. Easy exercise.
3.2 Relative subgroups
In this subsection I ⊳ R denotes an ideal.
Definition 9. An elementary transvection tij(x) is called I-elementary if x ∈ I. The
subgroup En(I) of En(R) generated by all I-elementary transvections is called the
preelementary subgroup of level I. Its normal closure En(R, I) in En(R) is called the
elementary subgroup of level I.
Definition 10. Let φ : GLn(R) → GLn(R/I) be the group homomorphism induced
by the canonical map R → R/I. The group Cn(R, I) := φ−1(Center(GLn(R/I)) is
called the full congruence subgroup of level I.
Remark 11. Let σ ∈ GLn(R). Then clearly σ ∈ Cn(R, I) iff σij ∈ I (1 ≤ i 6= j ≤ n)
and aσii − σjja ∈ I (1 ≤ i, j ≤ n, a ∈ R).
3
4 Normality of relative elementary subgroups
Lemma 12. Let I ⊳ R be an ideal. Let u ∈ Rn and v ∈ nR such that vu = 0.
Suppose that v has a zero entry and that all the other entries of v lie in I. Then
e + uv ∈ En(R, I)
Proof. Suppose vj = 0. The proof of [7, Lemma 3] shows that
e + uv = (Y
tji(vi)
tij (−ui)
Q
i6=j
i6=j
)Y
i6=j
tji((uj − 1)vi) ∈ En(R, I).
Recall that a row v ∈ nR is called unimodular if there is a column w ∈ Rn such
that vw = 1.
Proposition 13. Suppose R is an exchange ring and n ≥ 2. Let I ⊳ R be an ideal
and x ∈ I. Let u ∈ Rn be a column and v ∈ nR a unimodular row such that vu = 0.
Then e + uxv ∈ En(R, I)
Proof. Let w ∈ Rn such that vw = 1. Since R has property (1), there are elements
r1, . . . , rn ∈ R and idempotents e1, . . . , en ∈ R such that ei = viwiri (1 ≤ i ≤ n) and
e1 + · · · + en = 1. Let now i ∈ {1, . . . , n} and set τ (i) := e + uxeiv. Choose a j 6= i.
Then
τ (i)tij (−wirivj ) = e + u′v′
where u′ = tij(wirivj)u and v′ = xeivtij(−wirivj). Clearly v′
j = 0. It follows from
Lemma 12 that e + u′v′ ∈ En(R, I) and hence τ (i) ∈ En(R, I). Thus e + uxv =
τ (1) . . . τ (n) ∈ En(R, I).
Proof of Theorem 1. The case n = 1 is trivial, so let us assume that n ≥ 2. Let
tij(x) be an I-elementary transvection and σ ∈ GLn(R). Then tij(x)σ = e + σ′
∗ixσj∗.
By Proposition 13 we have e + σ′
∗ixσj∗ ∈ En(R, I). Thus En(R, I) is normal.
Remark 14. Using the proofs of Lemma 12 and Proposition 13 it is easy to obtain an
explicit expression of a matrix tij(x)σ, where σ ∈ GLn(R), i 6= j, x ∈ R, as a product
of 4n2 − 3n elementary transvections.
5 The standard commutator formula
The following result is well-known, see e.g.
(R3) in Lemma 6.
[7, Lemma 1]. It follows from relation
Lemma 15. Suppose n ≥ 3. Then En(R, I) = [En(R), En(R, I)] for any ideal I ⊳ R.
In particular, the absolute elementary subgroup En(R) is perfect.
Proposition 16. Suppose R is an exchange ring. Then [En(R), Cn(R, I)] ⊆ En(R, I)
for any ideal I ⊳ R.
4
Proof. The case n = 1 is trivial. Suppose now that n ≥ 2 and let σ ∈ Cn(R, I),
1 ≤ i 6= j ≤ n and x ∈ R. We have to show that [tij(x), σ] ∈ En(R, I). Since R has
property (1), there are elements r1, . . . , rn ∈ R and idempotents e1, . . . , en ∈ R such
that ek = σ′
jkσkjrk (1 ≤ k ≤ n) and e1 + · · · + en = 1. Let now k ∈ {1, . . . , n} and set
τ (k) := [tij(xek), σ] = tij(xek)(e − σ∗ixekσ′
j∗).
jl). Then
Case 1 Suppose k 6= j. Choose an l
τ (k)ξ = (tij(xek)ξ)(e + uv) where u = −ξ−1σ∗i and v = xekσ′
j∗ξ. Since k 6= j we have
ek ∈ I and hence tij(xek)ξ ∈ En(R, I). Moreover, vl = 0 and hence e + uv ∈ En(R, I)
by Lemma 12. It follows that τ (k) ∈ En(R, I).
6= k and set ξ := tkl(−σkjrkσ′
Case 2 Suppose k = j. set ξ := Q
l6=j
tjl(−σjjrjσ′
jl). Then
τ (j)ξ = (tij(xej)ξ)(e + uv)
where u = −ξ−1σ∗i and v = xejσ′
j∗ξ. Clearly
tij(xej)ξ = [ξ−1, tij(xej)]tij(xej).
(2)
(3)
Moreover, vl = 0 for any l 6= j. One checks easily that
tij(xej)(e + uv) = (e + zejj)tij(xej(1 + z) − σiixejσ′
jj) Y
tlj(−σlixejσ′
jj)
(4)
l6=i,j
where z = −(σji + P
l6=j
σjjrjσ′
jlσli)xejσ′
jj ∈ I. It follows from equations (2)-(4) that
τ (j)ξ = [ξ−1, tij(xej)](e + zejj)tij(xej(1 + z) − σiixejσ′
jj) Y
tlj(−σlixejσ′
jj).
l6=i,j
Clearly [ξ−1, tij(xej)] and Q
Moreover,
l6=i,j
tlj(−σlixejσ′
jj) lie in En(R, I) (note that ξ ∈ En(R, I)).
xej(1 + z) − σiixejσ′
jj
≡xej − σiixejσ′
jj
≡xej − xejσjjσ′
jj
=xej(1 − σjjσ′
jj)
=xej(X
σjlσ′
lj)
l6=j
≡0 mod I
by Remark 11, and hence tij(xej(1 + z) − σiixejσ′
that e + zejj ∈ En(R, I). Clearly
jj) ∈ En(R, I). It remains to show
z = − (σji + X
l6=j
σjjrjσ′
jlσli)xejσ′
jj
= − (σji − σjjrjσ′
= (σjjrjσ′
jj − 1)
{z
}
jjσji)xejσ′
σjixejσ′
jj
}
{z
b:=
jj
a:=
5
Clearly b ∈ I and ba = 0. Let u′ ∈ Rn be the column that has a at position j and
zeros elsewhere. Furthermore, let v′ ∈ nR be the row that has b at position j and
zeros elsewhere. It follows from Lemma 12 that e + zejj = e + u′v′ ∈ En(R, I). Hence
τ (j) ∈ En(R, I).
We have shown that τ (k) = [tij(xek), σ] ∈ En(R, I) for any 1 ≤ k ≤ n. Since
tij(x) = tij(xe1) . . . tij(xen), it follows that [tij(x), σ] ∈ En(R, I).
Theorem 2 follows immediately from Lemma 15 and Proposition 16.
6 Sandwich classification of En(R)-normal subgroups
In Subsection 6.1 we recall the concept of simultaneous reduction in groups, which
was introduced in [6]. In Subsection 6.2 we apply this idea to get a result for general
linear groups over arbitrary rings, namely Proposition 20. In Subsection 6.3 we use
Proposition 20 to prove Theorem 3.
6.1 Simultaneous reduction in groups
In this subsection G denotes a group.
Definition 17. Let (a1, b1), (a2, b2) ∈ G × G. If there is an g ∈ G such that
a2 = [a−1
1 , g] and b2 = [g, b1],
then we write (a1, b1)
g.
g
−→ (a2, b2). In this case (a1, b1) is called reducible to (a2, b2) by
Definition 18. If (a1, b1), . . . , (an+1, bn+1) ∈ G × G and g1, . . . , gn ∈ G such that
(a1, b1)
g1−→ (a2, b2)
g2−→ . . .
gn−→ (an+1, bn+1),
then we write (a1, b1)
(an+1, bn+1) by g1, . . . , gn.
g1,...,gn−−−−→ (an+1, bn+1). In this case (a1, b1) is called reducible to
Let H be a subgroup of G. If g ∈ G and h ∈ H, then we call gh an H-conjugate
of g.
g1,...,gn−−−−→ (a2, b2) for some
Lemma 19. Let (a1, b1), (a2, b2) ∈ G × G.
g1, . . . , gn ∈ G, then a2b2 is a product of 2n H-conjugates of a1b1 and (a1b1)−1 where
H is the subgroup of G generated by {a1, g1, . . . , gn}.
If (a1, b1)
Proof. Assume that n = 1. Then
a2b2 = [a−1
1 , g1][g1, b1] = (a1b1)g−1
1 a1 · ((a1b1)−1)a1.
Hence a2b2 is a product of two H-conjugates of a1b1 and (a1b1)−1. The general case
follows by induction.
6
6.2 Application to general linear groups
The proposition below generalises [6, Proposition 9(ii),(iii)].
Proposition 20. Suppose that n ≥ 3 and let σ ∈ GLn(R). Suppose that
y
n
X
p=1
σipxp = 0
for some i ∈ {1, . . . , n} and x1, . . . , xn, y ∈ R where xj = 1 for some j ∈ {1, . . . , n}.
Then for any k 6= l and a, b ∈ R, the elementary transvection tkl(ayxib) is a product
of 8 En(R)-conjugates of σ and σ−1.
Proof. Case 1 Suppose i = j. Choose 1 ≤ r, s ≤ n such that i, r and s are pairwise
distinct. Set τ := Q
tpi(xp) and
p6=i
ξ := [tri(−y), σ−1]τ = tri(−y)tri(y)στ
(note that tri(−y) and τ commute). Clearly ξ is a product of 2 En(R)-conjugates of
σ and σ−1. Since tri(y)σ = e + σ′
∗ryσi∗ and (yσi∗τ )i = 0, we get that (tri(y)στ )∗i = e∗i.
One checks easily that
(tri(−y), tri(y)στ )
tis(b),tir (a)
−−−−−−→ (tis(ayb), e).
It follows from Lemma 19 that tis(ayb) is a product of 4 En(R)-conjugates of ξ and
ξ−1. Thus tis(ayb) is a product of 8 En(R)-conjugates of σ and σ−1. The assertion of
the proposition follows now from Lemma 8.
Case 2 Suppose i 6= j. Choose 1 ≤ r ≤ n such that i, j and r are pairwise dis-
tinct. Set τ := Q
tpj(xp) and
p6=j
ξ := [tri(−y), σ−1]τ = trj(−yxi)tri(−y)tri(y)στ
(note that tri(−y)τ = trj(−yxi)tri(−y)). Clearly ξ is a product of 2 En(R)-conjugates
of σ and σ−1. Since tri(y)σ = e + σ′
∗ryσi∗ and (yσi∗τ )j = 0, we get that (tri(y)στ )∗j =
e∗j. One checks easily that
(trj(−yxi)tri(−y), tri(y)στ )
tji(b),tjr (a)
−−−−−−→ (tji(ayxib), e).
It follows from Lemma 19 that tji(ayxib) is a product of 4 En(R)-conjugates of ξ and
ξ−1. Thus tji(ayxib) is a product of 8 En(R)-conjugates of σ and σ−1. The assertion
of the proposition follows now from Lemma 8.
6.3 Proof of Theorem 3
Proposition 21. Suppose that R is an exchange ring and n ≥ 3. Let σ ∈ GLn(R),
i 6= j, k 6= l and a, b, c ∈ R. Then
7
(i) tkl(aσijb) is a product of 16n − 8 En(R)-conjugates of σ and σ−1 and
(ii) tkl(a(cσii − σjjc)b) is a product of 48n − 24 En(R)-conjugates of σ and σ−1.
Proof.
(i) Since R has property (1), there are there are elements r1, . . . , rn ∈ R
pirp (1 ≤ p ≤ n) and
pirpσii) = 0 for any p 6= i. It follows from
and idempotents e1, . . . , en ∈ R such that ep = σipσ′
e1 + · · · + en = 1. Clearly ep(σii − σipσ′
Proposition 20 that
tkl(aepb) is a product of 8 En(R)-conjugates of σ and σ−1 for any p 6= i.
(5)
Moreover, ei(σij − σiiσ′
iiriσij) = 0. It follows from Proposition 20 that
tkl(aeiσ′
iiriσijb) is a product of 8 En(R)-conjugates of σ and σ−1.
(6)
Clearly
tkl(aσijb) = (Y
tkl(aσiiepσ′
iiriσijb)(Y
tkl(aepσijb)).
p
p6=i
It follows from (5) and (6) that tkl(aσijb) is a product of 8n + 8(n − 1) = 16n − 8
En(R)-conjugates of σ and σ−1.
(ii) Clearly the entry of σtji(−c) at position (j, i) equals cσii − σjjc + σji − cσijc.
Applying (i) to σtji(−c) we get that tkl(a(cσii − σjjc + σji − cσijc)b) is a product
of 16n − 8 En(R)-conjugates of σ and σ−1. Applying (i) to σ we get that
tkl(a(cσijc−σji)b) = tkl(acσijcb)tkl(−aσjib) is a product of 2·(16n−8) = 32n−16
En(R)-conjugates of σ and σ−1. It follows that tkl(a(cσii − σjjc)b) = tkl(a(cσii −
σjjc + σji − cσijc)b)tkl(a(cσijc − σji)b) is a product of 48n − 24 En(R)-conjugates
of σ and σ−1.
Proof of Theorem 3. (⇒) Suppose that H is normalised by En(R). Let I ⊳ R be
the ideal generated by the set
{σij, aσii − σjja σ ∈ H, i 6= j, a ∈ R}.
Then H ⊆ Cn(R, I) by Remark 11. It remains to shown that En(R, I) ⊆ H. Since H
is normalised by En(R), it suffices to show that En(I) ⊆ H. But that clearly follows
from Proposition 21.
(⇐) Suppose that En(R, I) ⊆ H ⊆ Cn(R, I) for some ideal I ⊳ R. Then
[En(R), H] ⊆ [En(R), Cn(R, I)] = En(R, I) ⊆ H
by Theorem 2. Thus H is normalised by En(R). In order to show the uniqueness
of I suppose that J ⊳ R is an ideal such that En(R, J) ⊆ H ⊆ Cn(R, J). If x ∈ I,
then t12(x) ∈ En(R, I) ⊆ H ⊆ Cn(R, J). Hence, by Remark 11, x ∈ J. Thus we have
shown that I ⊆ J. The inclusion J ⊆ I follows by symmetry.
8
References
[1] G. Abrams, P. Ara, M. Siles Molina, Leavitt path algebras, Lecture Notes in
Mathematics 2191, Springer, 2017. 1
[2] P. Ara, The exchange property for purely infinite simple rings, Proc. Amer. Math.
Soc. 132 (2004), no. 9, 2543-2547. 1
[3] H. Bass, K-theory and stable algebra, Publ. Math. Inst. Hautes ´Etudes Sci. 22
(1964), 5 -- 60. 2
[4] I.Z. Golubchik, On the general linear group over an associative ring, Uspekhi
Mat. Nauk 28 (1973), no. 3, 179 -- 180 (Russian). 2
[5] W.K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math.
Soc. 229 (1977), 269-278. 1
[6] R. Preusser, Reverse decomposition of unipotents over noncommutative rings I:
general linear groups, arXiv:1912.03536 [math.RA]. 6, 7
[7] A.V. Stepanov, N.A. Vavilov, Decomposition of transvections: a theme with vari-
ations, K-Theory 19 (2000), no. 2, 109 -- 153. 4
[8] L.N. Vaserstein, On the normal subgroups of GLn over a ring, Lecture Notes in
Math. 854 (1981), 454 -- 465. 2
[9] L.N. Vaserstein, Normal subgroups of the general linear groups over Banach al-
gebras, J. Pure Appl. Algebra 41 (1986), 99 -- 112. 2
[10] L.N. Vaserstein, Normal subgroups of the general linear groups over von Neu-
mann regular rings, Proc. Am. Math. Soc 96 (1986), no. 2, 209 -- 214. 2
[11] J.S. Wilson, The normal and subnormal structure of general linear groups, Math.
Proc. Cambridge Philos. Soc. 71 (1972), 163 -- 177. 2
9
|
1507.02324 | 1 | 1507 | 2015-07-08T22:06:56 | Conservative algebras of $2$-dimensional algebras, II | [
"math.RA"
] | In 1990 Kantor defined the conservative algebra $W(n)$ of all algebras (i.e. bilinear maps) on the $n$-dimensional vector space. If $n>1$, then the algebra $W(n)$ does not belong to any well-known class of algebras (such as associative, Lie, Jordan, or Leibniz algebras). We describe automorphisms, one-sided ideals, and idempotents of $W(2).$ Also similar problems are solved for the algebra $W_2$ of all commutative algebras on the 2-dimensional vector space and for the algebra $S_2$ of all commutative algebras with trace zero multiplication on the 2-dimensional vector space. | math.RA | math |
Conservative algebras of 2-dimensional algebras, II
Ivan Kaygorodova,b, Yury Volkovc,d
a Universidade Federal do ABC, CMCC, Santo Andr´e, Brazil.
b Sobolev Institute of Mathematics, Novosibirsk, Russia.
c Universidade de Sao Paulo, IME, Sao Paulo, Brazil.
d Saint Petersburg state university, Saint Petersburg, Russia.
E-mail addresses:
Ivan Kaygorodov ([email protected]),
Yury Volkov (wolf86 [email protected]).
Abstract.
In 1990 Kantor defined the conservative algebra W (n) of all algebras (i.e.
bilinear maps) on the n-dimensional vector space. If n > 1, then the algebra W (n) does not
belong to any well-known class of algebras (such as associative, Lie, Jordan, or Leibniz algebras).
We describe automorphisms, one-sided ideals, and idempotents of W (2). Also similar problems
are solved for the algebra W2 of all commutative algebras on the 2-dimensional vector space
and for the algebra S2 of all commutative algebras with trace zero multiplication on the 2-
dimensional vector space.
Keywords: bilinear maps, conservative algebra, ideal, automorphism, idempotent.
2010 MSC: 15A03; 15A69; 17A30; 17A36
1
Introduction
During this paper F is some fixed field of zero characteristic. Algebras under consideration have
not to have a unit and have not to be associative. A multiplication on a vector space W is a
bilinear mapping W × W → W . We denote by (W, P ) the algebra with underlining space W
and multiplication P . Given a vector space W , a linear mapping A : W → W , and a bilinear
mapping B : W × W → W , we can define a multiplication [A, B] : W × W → W by the formula
[A, B](x, y) = A(B(x, y)) − B(A(x), y) − B(x, A(y))
for x, y ∈ W . For an algebra A with a multiplication P and x ∈ A we denote by LP
x the
operator of left multiplication by x. If the multiplication P is fixed, we write Lx instead of LP
x .
In 1990 Kantor [8] defined the multiplication · on the set of all algebras (i.e. all multiplica-
tions) on the n-dimensional vector space Vn as follows:
A · B = [LA
e , B],
1
where A and B are multiplications and e ∈ Vn is some fixed vector. If n > 1, then the algebra
W (n) does not belong to any well-known class of algebras (such as associative, Lie, Jordan, or
Leibniz algebras). The algebra W (n) turns out to be a conservative algebra (see below).
In 1972 Kantor [3] introduced conservative algebras as a generalization of Jordan algebras.
Namely, an algebra A = (W, P ) is called a conservative algebra if there is a new multiplication
F : W × W → W such that
[LP
F (a,b), P ]
b , [LP
a , P ]] = −[LP
(1)
for all a, b ∈ W . In other words, the following identity holds for all a, b, x, y ∈ W :
b(a(xy) − (ax)y − x(ay)) − a((bx)y) + (a(bx))y + (bx)(ay)
− a(x(by)) + (ax)(by) + x(a(by)) = −F (a, b)(xy) + (F (a, b)x)y + x(F (a, b)y).
(2)
The algebra (W, F ) is called an algebra associated to A.
Let us recall some well-known results about conservative algebras. In [3] Kantor classified
all conservative 2-dimensional algebras and defined the class of terminal algebras as algebras
satisfying some certain identity. He proved that every terminal algebra is a conservative alge-
bra and classified all simple finite-dimensional terminal algebras with left quasi-unit over an
algebraically closed field of characteristic zero [5]. Terminal trilinear operations were studied
in [6], and some questions concerning classification of simple conservative algebras were con-
sidered in [7]. After that, Cantarini and Kac classified simple finite-dimensional (and linearly
compact) super-commutative and super-anticommutative conservative superalgebras and some
generalization of these algebras (also known as "rigid" superalgebras) over an algebraically
closed field of characteristic zero [1].
The algebra W (n) plays a similar role in the theory of conservative algebras as the Lie
algebra of all n × n matrices gln plays in the theory of Lie algebras. Namely, in [4, 8] Kantor
considered the category Sn whose objects are conservative algebras of non-Jacobi dimension n.
It was proven that the algebra W (n) is the universal attracting object in this category, i.e.,
for every M ∈ Sn there exists a canonical homomorphism from M into the algebra W (n). In
particular, all Jordan algebras of dimension n with unity are contained in the algebra W (n).
The same statement also holds for all noncommutative Jordan algebras of dimension n with
unity. Some properties of the product in the algebra W (n) were studied in [11]. The study of
low dimensional conservative algebras was started in [12], where derivations and subalgebras of
codimension 1 of the algebra W (2) and of its principal subalgebras were described.
2
2 Conservative algebra W (2) and its subalgebras
A multiplication on 2-dimensional vector space is defined by a 2 × 2 × 2 matrix. Their classi-
fication was given in many papers (see, for example, [13]). Let consider the space W (2) of all
multiplications on the 2-dimensional space V2 with a basis v1, v2. The definition of the multi-
plication · on the algebra W (2) can be found in Section 1. Namely, we fix the vector v1 ∈ V2
and define
(A · B)(x, y) = A(v1, B(x, y)) − B(A(v1, x), y) − B(x, A(v1, y))
for x, y ∈ V2 and A, B ∈ W (2). The algebra W (2) is conservative [8].
Let consider the multiplications αk
i,j(vt, vl) =
δitδjlvk for all t, l. It is easy to see that {αk
i,ji, j, k = 1, 2} is a basis of the algebra W (2). The
multiplication tabel of W (2) in this basis is given in [12]. In this work we use another basis for
the algebra W (2). Let introduce the notation
i,j (i, j, k = 1, 2) on V2 defined by the formula αk
e1 = α1
e6 = 2α2
11 − α2
22 + α1
12 − α2
12 + α1
21, e2 = α2
21, e7 = α1
11, e3 = α2
12 − α1
22 − α1
21, e8 = α2
21, e4 = α1
12 − α1
12 − α2
21.
22, e5 = 2α1
11 + α2
12 + α2
21,
It is easy to see that the multiplication tabel of W (2) in the basis e1, . . . , e8 is the following.
0
0
e2
e1
3e2
e3
e1 −e1 −3e2
e3
2e1
e2
e3 −2e3 −e1 −3e4
0
e4
e5 −2e1 −3e2 −e3
3e4
e6
3e4
e7
−e3 −2e4
e8
e4
3e4
e3
0
0
0
0
0
2e3
2e3
0
e1
e1
e2
0
e5
−e5
0
e6
0
e6
e6
−e5
0
0
e7
e7
e8
0
0
e8
−e8
0
−e7
0
−2e5 −e6 −e7 −2e8
−e6
−e6
0
0
0
−e6 −e7
e7
e7
0
0
0
The subalgebra generated by the elements e1, . . . , e6 is the conservative (and, moreover, ter-
minal) algebra W2 of commutative 2-dimensional algebras. The subalgebra generated by the
elements e1, . . . , e4 is the conservative (and, moreover, terminal) algebra S2 of all commuta-
tive 2-dimensional algebras with trace zero multiplication [12]. We denote by pk : W (2) → F
(k = 1, . . . , 8) the map which sends
akek to ak.
8
Pk=1
3
3 One-sided ideals
3.1 One-sided ideals in W (2)
Let Annl(W (2)) be the space of W (2) generated by the elements e4, e5 − 2e1 − 3e8, e3 + e6, and
e3 + e7. It is easy to see that xW (2) = 0 for any x ∈ Annl(W (2)).
Lemma 1 If I is a non-trivial right ideal of W (2), then I is a subspace of Annl(W (2)).
Proof. Let x ∈ I. If p2(x) 6= 0, then (xe4)e2 = −p2(x)e1 ∈ I. It follows from the multipli-
cation table of the algebra W (2) that I = W (2). Let now consider the case where p2(x) = 0
for any x ∈ I.
We have (xe3)e2 = −(p1 − p5 − p8)(x)e1 ∈ I. So I = W (2) or p1(x) = p5(x) + p8(x) for any
x ∈ I. We have
xe2 = (−p3 + p6 + p7)(x)e1 + (−3p1 − 3p5 + p8)(x)e2.
Then p8(x) = 3(p1 + p5)(x) and p3(x) = (p6 + p7)(x). So any element x ∈ I has the form
x = p5(x)(e5 − 2e1 − 3e8) + p4(x)e4 + p6(x)(e3 + e6) + p7(x)(e3 + e7) ∈ Annl(W (2)).
The lemma is proved.
Lemma 2 If I is a non-trivial left ideal of W (2), then I is one of the following spaces:
1) Wα,β = Fhe1, e2, e3, e4, αe5 + βe8, −αe6 + βe7i, where α, β ∈ F, (α, β) 6= (0, 0);
2) I1 = Fhe1, e2, e3, e4i;
3) I2 = Fhe5, e6, e7, e8i;
4) wα,β = Fhαe5 + βe8, −αe6 + βe7i, where α, β ∈ F, (α, β) 6= (0, 0).
Proof. Let x =∈ I. We have e2(e2(e1x − x)) = 4p4(x)e1 ∈ I.
1. Assume that there is x ∈ I such that p4(x) 6= 0. Then it is obvious that e1, e2, e3, e4 ∈ I.
In this case either I is the subspace generated by e1, e2, e3, e4 (in this case I = I1) or there
exists a nonzero element w =
8
Pi=5
αiei ∈ I. In the former case we have
w1 = w − e1w = 2α5e5 + 2α8e8 ∈ I,
w2 = e2(e1w + w) = −2α6e5 + 2α7e8 ∈ I,
w3 = e1w + w = 2α6e6 + 2α7e7 ∈ I,
w4 = e2(w − e1w) = −2α5e6 + 2α8e7 ∈ I.
4
Now, if elements w1 and w2 are linear independent, then e5, e6, e7, e8 ∈ I and I = W (2). If the
elements w1 and w2 are linear dependent, then I = Wα,β for some α, β ∈ F.
2. Assume that p4(x) = 0 for any x ∈ I. If pi(x) = 0 for any x ∈ I and i = 1, 2, 3, 4, then
by the previous argumention, we have either I = I2 or I = wα,β for some α, β ∈ F.
Let consider x ∈ I. If p1(x) 6= 0, then we have
e6(e3(e3(e2(e1x − x)))) = −36p1(x)e4 ∈ I
and we have a contradiction with p4(I) = 0.
Suppose that p1(I) = p4(I) = 0. If p2(x) 6= 0 or p3(x) 6= 0 for some x ∈ I, then we have
and we have a contradiction with p1(I) = p4(I) = 0. The lemma is proved.
e3(e1x + x) = 2p2(x)e1 − 6p3(x)e4 ∈ I
3.2 One-sided ideals in W2
Let Annl(W2) be the subspace of W2 generated by the elements e4, and e3 + e6. It is easy to
see that xW2 = 0 for any x ∈ Annl(W2).
Lemma 3 If I is a non-trivial right ideal of W2, then I is a subspace of Annl(W2).
Proof. The proof is analogous to the proof of Lemma 1.
Lemma 4 If I is a non-trivial left ideal of W2, then I is one of the following spaces:
1) Fhe1, e2, e3, e4i;
2) Fhe5, e6i.
Proof. The proof is analogous to the proof of Lemma 2.
3.3 One-sided ideals in S2
Let Annl(S2) be the subspace of S2 generated by the element e4. It is easy to see that xS2 = 0
for any x ∈ Annl(S2).
Lemma 5 If I is a non-trivial right ideal of S2, then I is a subspace of Annl(S2).
Proof. The proof is analogous to the proof of Lemma 1.
Lemma 6 There are no non-trivial left ideals in S2.
Proof. The proof is analogous to the proof of Lemma 2.
5
3.4 Corollaries
Here we want to formulate some corollaries about a decomposition of the algebras under con-
sideration in a sum of one-sided ideals and about their ternary derivations.
Corollary 7 For any left ideal Y1 of the algebra W ∈ {W (2), W2, S2} there is some left ideal
Y2 of W such that W = Y1 + Y2 and Y1 ∩ Y2 = 0.
The definition of a ternary derivation arised in the paper of Perez-Izquierdo and Jimenez-
Gestal [2]. A ternary derivation is a triple of linear mappings (D, F, G) such that D(xy) =
F (x)y + xG(y). A ternary derivation is trivial if D, F , and G are sums of a derivation and an
element of a centroid of an algebra under consideration. Ternary derivations of Jordan algebras
and superalgebras, and n-ary algebras were studied in [9, 10, 14].
Corollary 8 The algebras W (2), W2 and S2 have non-trivial ternary derivations.
Proof. Let A be an algebra from the set {W (2), W2, S2}. By the lemmas above A has a
nonzero left annihilator Annl(A). Then for any nonzero linear map φ such that φ(A) ⊆ Annl(A)
the triple (0, φ, 0) is a non-trivial ternary derivation of A. The corollary is proved.
4 Automorphisms
In this section we discuss the automorphism groups of the algebras W (2), W2 and S2. Firstly
we prove a lemma which provides a series of automorphisms of the algebra W (n) for any n ≥ 2.
Let fix some space V and nonzero element e ∈ V . We denote by W (V, e) the algebra whose
elements are bilinear maps from V × V to V and multiplication on which is defined by the
formula
(A · B)(x, y) = A(e, B(x, y)) − B(A(e, x), y) − B(x, A(e, y))
for A, B ∈ W (V, e), x, y ∈ V . Let introduce the notation GL(V, e) = {f ∈ GL(V ) f (e) = e}.
It is clear that GL(V, e) is a subgroup of GL(V ). If V is an n-dimensional vector space, then
GL(V, e) is isomorphic to (n − 1)-dimensional affine group over F.
Lemma 9 There is a monomorphism of groups Φ : GL(V, e) → Aut(cid:0)W (V, e)(cid:1) defined by the
equality
for f ∈ GL(V, e), A ∈ W (V, e), x, y ∈ V.
(cid:0)Φ(f )(A)(cid:1)(x, y) = f A(f −1(x), f −1(y))
6
W (V, e), x, y ∈ V we have
Proof. Let us prove that Φ(f ) ∈ Aut(cid:0)W (V, e)(cid:1) for any f ∈ GL(V, e). Actually, for A, B ∈
(cid:0)Φ(f )(A) · Φ(f )(B)(cid:1)(x, y) = f A(cid:0)f −1(e), f −1f B(f −1(x), f −1(y))(cid:1)
− f B(cid:0)f −1f A(f −1(e), f −1(x)), f −1(y)(cid:1) − f B(cid:0)f −1(x), f −1f A(f −1(e), f −1(y))(cid:1)
= f(cid:16)A(cid:0)e, B(f −1(x), f −1(y))(cid:1) − B(cid:0)A(e, f −1(x)), f −1(y)(cid:1) − B(cid:0)f −1(x), A(e, f −1(y))(cid:1)(cid:17)
= (cid:0)Φ(f )(A · B)(cid:1)(x, y).
It is easy to check that Φ is a homomorphism of groups. Let f ∈ GL(V, e) be such that f 6= IdV .
Then there is such x ∈ V that f −1(x) 6= x. It is clear that there is such A ∈ W (V, e) that
A(e, x) = e and A(e, f −1(x)) 6= e. Then
Φ(A)(e, x) = f A(f −1(e), f −1(x)) = f A(e, f −1(x)) 6= f (e) = A(e, x).
So Φ(f ) 6= IdW (V,e) and we prove that Φ is a monomorphism. The Lemma is proved.
Any element of the group GL(V2, v1) can be written as a composition SbTa = TabSb (a, b ∈ F,
b 6= 0), where Ta(v2) = v2 + av1 and Sb(v2) = v2
b . We write simply Ta and Sb instead of Φ(Ta)
and Φ(Sb). Direct calculations show that the action of Ta and Sb on W (V2) is defined by the
formulas
Ta(e1) = e1 + 2ae3 + 3a2e4, Ta(e2) = e2 + ae1 + a2e3 + a3e4, Ta(e3) = e3 + 3ae4, Ta(e4) = e4,
Ta(e5) = e5 − ae6, Ta(e6) = e6, Ta(e7) = e7, Ta(e8) = e8 + ae7
and
Sb(e1) = e1, Sb(e2) =
e2
b
, Sb(e3) = be3, Sb(e4) = b2e4,
It follows from these formulas that Ta and Sb induce automorphisms of W2 and S2 which we
denote by Ta and Sb too.
Sb(e5) = e5, Sb(e6) = be6, Sb(e7) = be7, Sb(e8) = e8.
Remark 10 It follows from the proof of [12, Theorem 3] that Le7 and Le8 are derivations of
. On
tk
k! is defined and lies
. So if for any b ∈ F∗ there
the algebra W (2). The map Le7 is nilpotent and it is easy to see that Ta = P∞
the other hand the map Le8 is not nilpotent. But if the element et = P∞
in F∗ for some t ∈ F, then we have an equality Set = P∞
is t ∈ F such that b = et, then the automorphism Sb can be defined using the derivation Le8.
(aLe7 )k
(tLe8 )k
k=0
k=0
k!
k!
k=0
7
Now we are ready to describe the automorphism groups of the algebras W (2), W2 and S2.
induced homomorphisms from GL(V2, v1) to Aut(W2) and Aut(S2) are bijective too. In other
Theorem 11 The homomorphism Φ : GL(V2, v1) → Aut(cid:0)W (2)(cid:1) is bijective. Moreover, the
0 F∗(cid:19) (the
words, Aut(cid:0)W (2)(cid:1) ∼= Aut(W2) ∼= Aut(S2) is isomorphic to the matrix group (cid:18)1 F
one-dimensional affine group over F).
It can be easily checked that all three homomorphisms of groups considered in the
Proof.
theorem are monomorphic. So it is enough to prove that they are epimorphic. Firstly we prove
that the automorphisms Ta (a ∈ F) and Sb (b ∈ F∗) generate the group Aut(S2).
Let us take some f ∈ Aut(S2). Since f sends any right ideal to a right ideal, we have
f (e4) = te4 for some t ∈ F∗. Then we have 0 = f (e3e4) = tf (e3)e4. So f (e3) = be3 + ae4
for some a, b ∈ F. Let us consider the equality −3f (e4) = f (e3)f (e3).
It can be rewritten
in the form −3te4 = −3b2e4. So t = b2. Let us consider g = S 1
f . We know that
T−
g(e3) = e3 and g(e4) = e4 and it remains to prove that g = IdS2. It follows from the equality
e3 = g(e1e3) = g(e1)e3 that g(e1) = e1 + te4 for some t ∈ F. Then we have
a
3b
b
−e1 − te4 = g(−e1) = g(e1e1) = (cid:0)g(e1)(cid:1)2
= −e1 + 3te4.
So t = 0 and we have g(e1) = e1. Using the equality 2e1 = g(e2e3) = g(e2)e3 we obtain that
g(e2) = e2 + te4 for some t ∈ F. Consideration of the equality 0 = (cid:0)g(e2)(cid:1)2 shows that t = 0
and g(e2) = e2. So we prove that Aut(S2) is generated by Ta and Sb (a, b ∈ F, b 6= 0).
Let now prove that Ta (a ∈ F) and Sb (b ∈ F∗) generate the group Aut(W2). Since S2
is the unique 4-dimensional left ideal of W2 by Lemma 4, every automorphism of W2 induces
an automorphism of S2. Let take some f ∈ Aut(W2). Since Aut(S2) is generated by auto-
morphisms of the form Ta and Sb, there are such a ∈ F and b ∈ F∗ that the automorphism
g = SbTaf is identical on S2. It remains to prove that g(e5) = e5 and g(e6) = e6. But e5 and e6
generate the unique 2-dimensional left ideal of W2 and so g(e5) = t1e5 + t2e6, g(e6) = t3e5 + t4e6.
Consideration of the equalities −3e2 = g(e5e2) = g(e5)e2 and e1 = g(e6e2) = g(e6)e2 shows that
t1 = t4 = 1 and t2 = t3 = 0, i.e. g = IdW2.
It remains to prove that Ta (a ∈ F) and Sb (b ∈ F∗) generate the group Aut(cid:0)W (2)(cid:1). By
Lemma 2 the algebra W (2) has two 4-dimensional left ideals: S2 and I = Fhe5, e6, e7, e8i. Note
that W (2) contains 2-dimensional ideals and they all lie in I. So any automorphism of the
algebra W (2) maps S2 to S2 and I to I. Analogously to the case of the algebra W2 it is
enough to prove that any automorphism g ∈ Aut(cid:0)W (2)(cid:1) identical on S2 is identical on I. It
follows from the equalities g(e5) = −e1g(e5) and g(e8) = −e1g(e8) that the space Fhe5, e8i is
invariant under the action of g. Using the equalities −2e1 = g(e5)e1 and 0 = g(e5)e4 we see
8
that g(e5) = e5. Using the equalities 0 = g(e8)e1 and −2e4 = g(e8)e4 we see that g(e8) = e8.
Finally, we have g(e6) = g(e3e5) = e3e5 = e6 and g(e7) = −g(e3e8) = −e3e8 = e7. So the
Theorem is proved.
Corollary 12 Let A and B be two multiplications on the space V2. There is an isomorphism
from the algebra (V2, A) to the algebra (V2, B) which sends v1 to v1 iff there is an automorphism
of the algebra W (2) which sends (V2, A) to (V2, B).
5
Idempotents
Let now describe the idempotents of the algebra W (2). For e ∈ W (2) we denote by O(e) =
{SbTa(e) a, b ∈ F, b 6= 0} the orbit of e under the action of the automorphism group of W (2).
It is clear that e is an idempotent iff all elements of O(e) are idempotents. We denote by ¯F the
set of representatives of cosets of the subgroup (F∗)2 in the group F∗.
Theorem 13 The set of nonzero idempotents of the algebra W (2) equals to the disjoint union
of the following sets:
1. O(e8 + e2 − e1 + c(3e8 + e5 − 2e1)) (c ∈ F);
2. O(−e1 + c(e5 − 2e1) + de8) (c, d ∈ F);
3. O(−e1 − 2e8 + 4e3 + e6 + 3e7 + c(3e8 − e5 + 2e1) + de4) (c, d ∈ F);
4. O(−e1 − 2e8 + c(3e8 − e5 + 2e1) + qe4) (c ∈ F, q ∈ ¯F);
Proof. Direct calculations show that all elements of the set described in the theorem are
idempotent. Let e ∈ W (2) be a nonzero idempotent. Let us prove that e lies in the set
described in the theorem. For c, d ∈ F, q ∈ ¯F let introduce the notation
w1(c) = e8 + e2 − e1 + c(3e8 + e5 − 2e1), w2(c, d) = −e1 + c(e5 − 2e1) + de8,
w3(c, d) = −e1 − 2e8 + 4e3 + e6 + 3e7 + c(3e8 − e5 + 2e1) + de4,
w4(c, d) = −e1 − 2e8 + c(3e8 − e5 + 2e1) + de4.
1. Assume that p2(e) 6= 0. Let us take b = p2(e) and a = − p1(e)+2p5(e)
It is easy
to see that p2SbTa(e) = 1 and (p1 + 2p5)SbTa(e) = −1. So we may suppose that p2(e) = 1
and (p1 + 2p5)(e) = −1 initially. Let us introduce the notation c = p5(e). Then we have
p1(e) = −(2c + 1). It follows from the equality p2(e2) = p2(e) that p8(e) = 3c + 1. Rewriting
the equalities p5(e2) = p5(e) and p8(e2) = p8(e) we obtain that p6(e) = p7(e) = 0. Rewriting
p2(e)
− 1.
9
the equality p1(e2) = p1(e) we obtain p3(e) = −p6(e) − p7(e) = 0. Then we have 0 = p3(e2) =
p2(e)p4(e) = p4(e). So we have
e =
8
Xk=1
pk(e)ek = −(2c + 1)e1 + e2 + ce5 + (3c + 1)e8 = w1(c).
Let now prove that (p1 + 2p5)(e) = −1 if e is a nonzero idempotent such that p2(e) = 0.
Using multiplication table we see that
p1(e2) = −(cid:0)p1(e) + 2p5(e)(cid:1)p1(e), p5(e2) = −(cid:0)p1(e) + 2p5(e)(cid:1)p5(e).
Since e2 = e, we have p1(e) + 2p5(e) = −1 in the case where p1(e) 6= 0 or p5(e) 6= 0. It is enough
to prove that e = 0 if p1(e) = p2(e) = p5(e) = 0. But in this case we obtain consequently using
multiplication table and the equality e2 = e that p8(e) = 0, p6(e) = p7(e) = 0, p3(e) = 0 and
p4(e) = 0, i.e. e = 0.
Also, if p2(e) = 0, we have the equalities
p3(e) = p3(e2) = 2(p6(e) + p7(e) − p3(e))p1(e) − (p8(e) + p5(e) − p1(e))p3(e),
p4(e) = p4(e2) = 3(p6(e) + p7(e) − p3(e))p3(e) − (2p8(e) − 3p1(e))p4(e),
p6(e) = p6(e2) = (p3(e) − p6(e) − p7(e))p5(e) − (p8(e) + p5(e) − p1(e))p6(e),
p7(e) = p7(e2) = (p6(e) + p7(e) − p3(e))p8(e) − (p8(e) + p5(e) − p1(e))p7(e),
which can be rewritten in the form
(p8(e) + p5(e) − p1(e) + 1)p3(e) = 2(p6(e) + p7(e) − p3(e))p1(e),
(2p8(e) − 3p1(e) + 1)p4(e) = 3(p6(e) + p7(e) − p3(e))p3(e),
(p8(e) + p5(e) − p1(e) + 1)p6(e) = (p3(e) − p6(e) − p7(e))p5(e),
(p8(e) + p5(e) − p1(e) + 1)p7(e) = (p6(e) + p7(e) − p3(e))p8(e).
(3)
2. Assume that p2(e) = 0, p8(e) − p5(e) − 2p1(e) 6= 0. Let us take a = − p3(e)−p6(e)−p7(e)
p8(e)−p5(e)−2p1(e).
It is easy to see that (p3 − p6 − p7)Ta(e) = 0. So we may suppose that (p3 − p6 − p7)(e) = 0
initially. Since
2p8(e) − 3p1(e) + 1 = 2(p8(e) + p5(e) − p1(e) + 1) = 2(p8(e) + 3p5(e)) 6= 0,
it follows from (3) that p3(e) = p4(e) = p6(e) = p7(e) = 0. Let us denote p5(e) by c and p8(e)
by d. Then p1(e) = −(2c + 1) and we have e = w2(c, d).
3. Assume that p2(e) = 0, p8(e) − p5(e) − 2p1(e) = 0. If p6(e) + p7(e) − p3(e) 6= 0, then
it follows from (3) that p1(e) = p5(e) = 0. This contradicts the equality p1(e) + 2p5(e) = −1
10
proved above. So we have p3(e) = p6(e) + p7(e). Let us take a = p3(e)−4p6(e)
. It is easy to see
that (4p6 − p3)Ta(e) = 0. So we may suppose that (4p6 − p3)(e) = 0 initially. Let now consider
two cases.
2
3a. Let p6(e) 6= 0. Then we have p6S 1
(e) = 1 and so we may suppose that p6(e) = 1
initially. Then we have p3(e) = 4 and p7(e) = p3(e) − p6(e) = 3. Let us denote p5(e) by −c and
p4(e) by d. We have p1(e) = 2c − 1 and p8(e) = p5(e) + 2p1(e) = 3c − 2. So e = w3(c, d).
p6(e)
3b. Let p6(e) = 0. Then we have p3(e) = p7(e) = 0. If p4(e) = 0, then we have e = w2(c, d)
for c = p5(e) and d = p8(e). Let now p4(e) 6= 0. There are b ∈ F∗ and q ∈ ¯F such that
b2p4(e) = q. If we denote p5(e) by −c, we obtain Sb(e) = w4(c, q).
Let us denote by E the set
{w1(c)}c∈F ∪ {w2(c, d)}c,d∈F ∪ {w3(c, d)}c,d∈F ∪ {w4(c, q)}c∈F,q∈¯F.
f (w) = w (it is clear that the union of sets listed in the theorem is disjoint in this case).
We prove that for any w ∈ E and f ∈ Aut(cid:0)W (2)(cid:1) the condition f (w) ∈ E is equivalent to
b and p2(w) ∈ {0, 1}
for any w ∈ E, we have b = 1. Then we have (p1 + 2p5)(e) = a − 1. Since e ∈ E and p2(e) = 1,
we have (p1 + 2p5)(e) = −1 and so a = 0. We obtain that e = w1(c).
1. Suppose that e = SbTa(w1(c)) ∈ E (a, c ∈ F, b ∈ F∗). Since p2(e) = 1
2. Suppose that e = SbTa(w2(c, d)) ∈ E (a, c, d ∈ F, b ∈ F∗). We have p3(e) = −4abc − 2ab,
p6(e) = −abc. We have p3(w) = 4p6(w) for any w ∈ E. Since e ∈ E, we have −4abc − 2ab =
−4abc, i.e. a = 0. Then e = Sb(w2(c, d)) = w2(c, d).
3. Suppose that e = SbTa(w3(c, d)) ∈ E (a, c, d ∈ F, b ∈ F∗). Analogously to the previous
point we have a = 0. Then p6(e) = b. Since p6(w) ∈ {0, 1} for any w ∈ E, we have b = 1. Then
e = w3(c, d).
4. Suppose that e = SbTa(w4(c, q)) ∈ E (a, c ∈ F, b ∈ F∗, q ∈ ¯F). Analogously to the second
point of the present proof we have a = 0. Then e = −e1 − 2e8 + c(3e8 − e5 + 2e1) + b2qe4. So
e ∈ E iff b2q = 0 or b2q ∈ ¯F. Since b 6= 0 and q 6= 0, b2q ∈ ¯F. But q and b2q lie in the same coset
of the group (F ∗)2 in the group F ∗ and so by definition only one of them lies in ¯F if q 6= b2q.
Since q lies in ¯F, we have b2q = q and so e = −e1 − 2e8 + c(3e8 − e5 + 2e1) + qe4 = w4(c, q).
The theorem is proved.
Let introduce the algebras W1(t), W2(t, s), W3(t, s) and W4(t, u) for s, t ∈ F, u ∈ ¯F by the
following multiplication tables
11
W1(t) :
v1
v1 −v1 + v2
v2
0
v2
tv2
0
W2(t, s):
v1
v1 −v1
v2
sv2
v2
tv2
0
W3(t, s):
v1
−v1
v2
−v2
tv2 − v1
v2 + sv1
v1
v2
W4(t, u):
v1
v2
v1 −v1 −v2
v2
uv1
tv2
It is easy to see that the algebra W1(t) corresponds to the element w1(cid:0) t−2
, t−s
corresponds to the element S 1
the algebra W2(t, s) corresponds to the element w2(cid:0) t+s−2
the element w4(cid:0) 3−t
6 (cid:1) ∈ W (2),
2 (cid:1) ∈ W (2), the algebra W3(t, s)
6 , 36s(cid:1)(cid:1) ∈ W (2), and the algebra W4(t, q) corresponds to
6 , q(cid:1) ∈ W (2) (see the proof of Theorem 13 for notation). We denote by L
the set
6 (cid:0)w3(cid:0) 3−t
6
{W1(t)}t∈F ∪ {W2(t, s)}t,s∈F ∪ {W3(t, s)}t,s∈F ∪ {W4(t, u)}t∈F,u∈¯F.
Corollary 14 Let R be a nonzero two-dimensional algebra with left quasi-unit e. Then there is
an unique algebra W ∈ L such that there is an isomorphism φ : R → W satisfiing the condition
φ(e) = v1.
Proof. Let φ0 : R → V2 be some isomorphism of vector spaces such that φ0(e) = v1. There
is an unique multiplication A on V2 such that φ0 is an isomorphism from R to (V2, A). Then
v1 is a quasi-unit of (V2, A) and so (V2, A) is an idempotent of the algebra W (2). By Theorem
13 and the definition of the set L there are such W ∈ L and automorphism f of W (2) that
f sends the element of W (2) corresponding (V2, A) to the element of W (2) corresponding W .
Then by Corollary 12 there is an isomorphism φ1 : (V2, A) → W such that φ1(v1) = v1. The
map φ = φ1φ0 satisfies to all required conditions. If we have some another algebra W ′ ∈ L and
isomorphism φ′ : R → W ′ such that φ′(e) = v1, then there is an isomorphism φ′φ−1 : W → W ′
which sends v1 to v1. By Corollary 12 the elements of W (2) corresponding to W and W ′ lie in
the same orbit under the action of Aut(cid:0)W (2)(cid:1). But it follows from definition of L and Theorem
13 that W = W ′ in this case.
Acknowledgements: The first author was supported by the Brazilian FAPESP, Proc.
2014/24519-8, the second author was supported by the Brazilian FAPESP, Proc. 2014/19521-
3, all authors were supported by RFBR 15-31-21169. We are grateful for the support.
12
References
[1] N. Cantarini, V. Kac, Classification of linearly compact simple rigid superalgebras, Int.
Math. Res. Not. (IMRN), 17 (2010), 3341–3393.
[2] C. Jim´enez-Gestal, J.M. P´erez-Izquierdo, Ternary derivations of generalized Cayley-Dickson
algebras, Comm. Algebra, 31 (2003), no. 10, 5071–5094.
[3] I. Kantor, Certain generalizations of Jordan algebras (Russian), Trudy Sem. Vektor. Tenzor.
Anal., 16 (1972), 407–499.
[4] I. Kantor, A universal attracting object in the category of conservative algebras (Russian),
Trudy Sem. Vektor. Tenzor. Anal. No. 23 (1988), 45–48.
[5] I. Kantor, On an extension of a class of Jordan algebras, Algebra and Logic, 28 (1989),
no. 2, 117–121.
[6] I. Kantor, Terminal trilinear operations, Algebra and Logic, 28 (1989), no. 1, 25–40.
[7] I. Kantor, Some problems in L-functor theory (Russian), Trudy Inst. Mat. (Novosibirsk),
no. 16 (1989), Issled. po Teor. Kolets i Algebr, 54–75.
[8] I. Kantor, An universal conservative algebra, Siberian Math. J., 31 (1990), no. 3, 388–395.
[9] I. Kaygorodov, On (n + 1)-ary derivations of simple n-ary Malcev algebras, St. Petersburg
Math. J., 25 (2014), no. 4, 575–585.
[10] I. Kaygorodov, (n + 1)-ary Derivations of Semisimple Filippov algebras, Math. Notes, 96
(2014), no. 2, 208–216.
[11] I. Kaygorodov, On the Kantor Product, arXiv:1506.00736
[12] I. Kaygorodov, A. Lopatin, Yu. Popov, Conservative algebras of 2-dimensional algebras,
arXiv:1411.7744
[13] H. Petersson, The classification of two-dimensional nonassociative algebras, Results Math.,
37 (2000), no. 1-2, 120–154.
[14] A. Shestakov, Ternary derivations of Jordan superalgebras, Algebra and Logic, 53 (2014),
no. 4, 323–348.
13
|
1101.1116 | 2 | 1101 | 2012-04-02T18:51:02 | Lower bounds of Growth of Hopf algebras | [
"math.RA"
] | Some lower bounds of GK-dimension of Hopf algebras are given. | math.RA | math | LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
Abstract. Some lower bounds of GK-dimension of Hopf algebras are given.
r
p
A
2
]
.
A
R
h
t
a
m
[
2
v
6
1
1
1
.
1
0
1
1
:
v
i
X
r
a
0. Introduction
A seminal result of Gromov states that a finitely generated group has polynomial
growth, or equivalently, the associated group algebra has finite Gelfand-Kirillov
dimension, if and only if it has a nilpotent subgroup of finite index [Gr]. Group
algebras form a special class of cocommutative Hopf algebras. It is natural to ask
Question 0.1. What are necessary and sufficient conditions on a finitely generated
Hopf algebra H such that its Gelfand-Kirillov dimension is finite?
Let k be a base field and everything be over k. Assume that, for simplicity, k
is algebraically closed of characteristic zero. It is clear that an affine (i.e., finitely
generated) commutative Hopf algebra has a finite GK-dimension (short for Gelfand-
Kirillov dimension) which equals its Krull dimension. If H is cocommutative, by a
classification result [Mo, Corollary 5.6.4 and Theorem 5.6.5], it is isomorphic to a
smash product U (g)#kG for some group G and some Lie algebra g. Consequently,
(I0.1.1)
GKdim H = GKdim kG + dim g,
which solves Question 0.1 in terms of conditions on G and g. Question 0.1 is
also answered for several classes of noncommutative and noncocommutative Hopf
algebras, including quantum groups Uq(g) and Oq(G), see [BG2, GZ1]. The present
paper attempts to study Question 0.1 for a larger class of noncommutative and
noncocommutative Hopf algebras by providing three lower bounds of GK-dimension
in terms of certain invariants of skew primitive elements.
Let H be a Hopf algebra over k. A nonzero element y ∈ H is called (1, g)-
primitive (or generally skew primitive) if ∆(y) = y ⊗ 1 + g ⊗ y and such a g is called
the weight of y and denoted by µ(y). Let G(H) denote the group of group-like
elements in H and let C0 = kG(H). Here is the first lower bound theorem.
Theorem 0.2 (First lower bound theorem). Let D ⊇ C0 be a Hopf subalgebra of
H. Let {yi}w
(a) {yi}w
(b) for all i ≤ j, yiµ(yj) = λij µ(yj)yi for some λij ∈ k×,
(c) for each i, λii is either 1 or not a root of unity.
i=1 be a set of skew primitive elements such that
i=1 is linearly independent in H/D.
Then GKdim H ≥ GKdim D + w.
2000 Mathematics Subject Classification. Primary 16P90, 16W30; Secondary 16A24, 16A55.
Key words and phrases. Hopf algebra, Gelfand-Kirillov dimension, skew primitive, pointed.
1
2
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
In general λij in condition (b) may not exist. If that is the case, we have other
ways of obtaining lower bounds.
Let W denote the set of weights µ(y) for all skew primitive elements y 6∈ C0
and let W√ be the subset of W consisting of weights µ(y) for all y such that yn
is also a skew primitive for some n > 1. (Note that in this paper the term "skew
primitive" means "(1, g)-primitive"). For any subset Φ ⊂ G(H), the subgroup of
G(H) generated by Φ is denoted by hΦi. Here is the second lower bound theorem.
Theorem 0.3 (Second lower bound theorem). Suppose hW \W√ i is abelian. Then
(I0.3.1)
GKdim H ≥ GKdim C0 + #(W \ W√ ).
There are examples such that W = W√ and GKdim H = GKdim C0, but
#(W√ ) is arbitrarily large [Example 2.7]. Therefore W√ has to be removed from
W when we estimate the GK-dimension of H.
Let y be a skew primitive element not in C0. If
(I0.3.2)
µ(y)−1yµ(y) − cy ∈ C0
for some c ∈ k×, then c is called the commutator of y (with its weight) and denoted
by γ(y). By Lemma 1.6, (I0.3.2) is equivalent to
(I0.3.3)
µ(y)−1yµ(y) − cy = τ (µ(y) − 1)
for some τ ∈ k. Define Γ to be the set of γ(y) for all skew primitive elements y 6∈ C0
such that γ(y) exists and let Γ√ be the subset of Γ consisting of those γ(y) which
are roots of unity but not 1. If γ(y) exists, the pair (µ(y), γ(y)) is denoted by ω(y)
and is called the weight commutator of y. When (I0.3.3) holds and if c 6= 1, y can
be replaced by z := y + (c − 1)−1τ (µ(y) − 1), which is a skew primitive element
with ω(z) = ω(y) and satisfies the equation µ(z)−1zµ(z) − γ(z)z = 0.
Define Ω to be the set of ω(y) for all skew primitive elements y 6∈ C0 such that
ω(y) exists and let Ω√ be the subset of Ω consisting of those ω(y) in which γ(y)
is a root of unity but not 1. Theorem 0.3 can be improved a little under the same
hypothesis:
GKdim H ≥ GKdim C0 + #(Ω \ Ω√ ).
Let y be a skew primitive element not in C0 with g = µ(y). Let Tg−1 be the in-
verse conjugation by g, namely, Tg−1 : a → g−1ag. A scalar c is called a commutator
of y of level n if n is the least nonnegative integer such that
(I0.3.4)
(Tg−1 − cIdH )n(y) ∈ C0.
In this case we also write γ(y) = c. Let Z denote the space spanned by the identity
element 1 and all skew primitive elements of H and let Y√ denote the subspace of
Z spanned by those y with commutator of finite level and with γ(y) being a root of
unity but not 1. Here is the third lower bound theorem. Let W× be the subset of
W consisting of weights µ(y) such that the commutator of y (as defined in (I0.3.4))
exists and is either 1 or not a root of unity. Note that W \ W√ ⊆ W× and these
are often equal [Remark 3.9].
Theorem 0.4 (Third lower bound theorem). Suppose hW×i is abelian. Then
(I0.4.1)
GKdim H ≥ GKdim C0 + dim Z/(C0 + Y√ ).
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
3
When H is cocommutative, equality holds in Theorem 0.4, see (I0.1.1). There
are examples such that Z = Y√ + C0 and GKdim H = GKdim C0, but dim Y√ is
arbitrarily large [Examples 2.7 and 3.13]. Therefore it is sensible to consider the
quotient space Z/(C0 + Y√ ) in the above theorem. This is analogous to removing
W√ in Theorem 0.3.
If hW×i is abelian, Theorem 0.4 is a generalization of Theorem 0.3 [Lemma 3.12].
After some analysis, Theorem 0.2 (when D = C0) can be viewed as a consequence
of Theorem 0.4. These lower bounds provide some evidence that the GK-dimension
of H is related to some combinatorial data coming from the skew primitive elements
when H is pointed.
The proof of these lower bounds is based on a version of the Poincar´e-Birkhoff-
Witt (PBW) theorem [Theorem 1.5(b)] which states that under some hypotheses
the set of monomials generated by skew primitive elements is linearly independent
(over the Hopf subalgebra C0). Restricted to the universal enveloping algebra of
a finite dimensional Lie algebra, Theorem 1.5 implies the original PBW theorem.
Theorem 1.5 is in a similar spirit to Kharchenko's quantum analog of the PBW
theorem [Kh]. One of the hypotheses in Theorem 1.5 is (I1.2.3) which assume es-
sentially that the action of the group generated by weights on the space generated
by skew primitive elements is locally finite. When GKdim H is finite, this is a rea-
sonable hypothesis indicated by a result of the third-named author [Zhu, Theorem
1.2] (see also Lemma 2.5).
In general we are far from answering Question 0.1. There are a lot of unsolved
questions concerning the growth of Hopf algebras. The hypotheses in Theorems
0.3 and 0.4 could be superfluous, but we don't know how to remove them at this
moment. When hW i is non-abelian, a possible better lower bound could be obtained
by replacing #(W \ W√ ) in Theorem 0.3 by GKdim khW i, see Lemma 2.6(b) for
details. It is expected that these lower bounds can (or should) be improved and
that possible upper bounds should be found once finer invariants are introduced.
The ultimate goal is to find a formula for the GK-dimension of a Hopf algebra which
is analogous to Bass' theorem [KL, Theorem 11.14] in the group algebra case, and
then eventually to solve Question 0.1.
There are further connections between the growth of Hopf algebras and W and
other invariants defined by skew primitive elements. Let rank denote the torsionfree
rank of an abelian group.
Proposition 0.5. Suppose hW i is abelian and torsionfree. If rankhΓi > rankhW i =
1, then H has exponential growth.
Note that rankhΓ \ Γ√ i = rankhΓi since elements in Γ√ have finite order. The
rank of hW i and hΓi should be related when GKdim H is finite.
Question 0.6. Suppose rankhΓi > rankhW i. Does then H have exponential
growth?
Quite a few families of Hopf algebras of finite GK-dimension have been analyzed
extensively by several authors [AA, AS1, AS2, Br1, Br2, BG1, BG2, BZ, GZ1, GZ2,
LWZ, WuZ1, WuZ2, Zhu] during the last few years. But the classification of such
Hopf algebras is far from complete. These lower bounds are useful for studying
pointed Hopf algebras of low GK-dimension. For example, if GKdim H = 2, then
there are only three possibilities for GKdim C0, #(W \ W√ ), #(Ω \ Ω√ ) and
4
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
dim Z/(C0+Y√ ). This is one of the initial steps in our ongoing project of classifying
pointed Hopf algebra domains of GK-dimension two and three.
Definitions and basic properties of GK-dimension can be found in the first three
chapters of [KL]. Our reference book for Hopf algebras is [Mo].
1. First Lower Bound Theorem
In this section we prove Theorem 0.2. We need some lemmas.
Lemma 1.1. Let D be a Hopf subalgebra of H and 0 6= F ∈ H. Suppose that
(a) L is a subcoalgebra of H containing D,
(b) L is a left D-module via the multiplication, and
(c) there are nonzero-divisors (regular elements) h, g ∈ L such that ∆(F ) − F ⊗
h − g ⊗ F ∈ L ⊗ L.
Define V = {a ∈ D aF ∈ L}. Then V is either 0 or D.
Proof. Suppose V is nonzero and let a be a nonzero element in V . Let C be the
subcoalgebra of D generated by a. There is a k-linear basis
{a1, · · · , av, av+1, · · · , aw}
of C such that C ∩ V is spanned by {a1, · · · , av}. This means that aiF ∈ L for all
i ≤ v and that any nontrivial linear combination of {av+1F, · · · , awF } is not in L.
Write ∆(a) = P1≤i,j≤w ξij ai ⊗ aj for some ξij ∈ k.
For simplicity, we use the symbol ldt1 for any element in L and use ldt2 for any
element in L ⊗ L. By the definition of V , we have aF + ldt1 = 0 for some ldt1 ∈ L
and whence
0 = ∆(aF + ldt1) = ∆(a)∆(F ) + ∆(ldt1)
= (Xi,j
= (Xi,j
= (Xi>v
all j
ξij ai ⊗ aj)(F ⊗ h + g ⊗ F + ldt2) + ldt2
ξij ai ⊗ aj)(F ⊗ h) + (Xi,j
ξij ai ⊗ aj)(F ⊗ h) + (Xj>v
all i
ξij ai ⊗ aj)(g ⊗ F ) + ldt2
ξij ai ⊗ aj)(g ⊗ F ) + ldt2
where the last equation uses the fact aiF ∈ L for all i ≤ v. The above equation
implies that
(Xj>v
all i
ξijai ⊗ aj)(g ⊗ F ) = −(Xi>v
all j
ξij ai ⊗ aj)(F ⊗ h) + ldt2 ∈ H ⊗ L
or equivalently Pw
pendent, we have Pj>v ξij ajF ∈ L for all i. By the definition of {av+1, · · · , aw}, we
i=1(aig)⊗(Pj>v ξij ajF ) ∈ H ⊗L. Since {aig}w
i=1 is linearly inde-
obtain that ξij = 0 for all j > v. Similarly, ξij = 0 for all i > v. Thus ∆(a) ∈ V ⊗ V
and hence V is a subcoalgebra of D. Since V is a subcoalgebra, there is an element
v ∈ V such that ǫ(v) = 1. Then
1F = ǫ(v)F = X S(v1)v2F ∈ L
since v2 ∈ V and S(v1) ∈ D. This shows that 1 ∈ V . Since V is a left ideal of D,
V = D.
(cid:3)
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
5
all i ≤ p and cp+1 < dp+1.
i=1 ci = Pv
i=1 di or Pv
Remark 1.2. Recall that (Nv, +), for every v ≥ 1, is a linearly ordered semigroup
with respect to the following ordering. Define (c1, · · · , cv) < (d1, · · · , dv) if either
i=1 di and there is a p < v such that ci = di for
Pv
i=1 ci < Pv
Let T := kh{xj}j∈J i be the free algebra generated by {xj}j∈J . Given any family
(fj)j∈J where fj ∈ Nv, we can define an Nv-graded structure on T by setting
deg xj = fj for all j ∈ J. Then T = Lw∈Nv Tw. Since Nv is linearly ordered, T
has a canonical Nv-filtration defined by Fw(T ) = Pw′≤w Tw′. Let B be any factor
ring of T . The Nv-filtration on T induces a unique Nv-filtration on B, denoted by
{Fw(B) w ∈ Nv}. We say that an element x ∈ B has filtered multi-degree
deg x := w = (d1, · · · , dv)
and filtered total-degree d = P di if x ∈ Fw(B) \ Pw′<w Fw′(B). Note that the
filtered total-degree induces an N-filtration on B.
In applications, we usually start with an algebra A generated by {y1, · · · , yv}
and G = {gi}i∈I for some index set I. By the discussion in the previous paragraph,
we can define two filtrations (and corresponding filtered degrees) on A such that
if f = yi1yi2 · · · yis ∈ A, then the filtered total y-degree of f is at most s and the
filtered multi-y-degree of f is at most (n1, · · · , nv) where ni is the number of yi
appearing in F , and if g ∈ G, the filtered total-y-degree and the filtered multi-y-
degree of g are both 0.
These two filtrations can be extended to the tensor product A ⊗ A, namely,
simplicity, the words "filtered" and "filtration" might be omitted below.
Fw(A ⊗ A) := Pw′+w′′≤w Fw′ (A) ⊗ Fw′′ (A) for all w ∈ Nv (or w ∈ N). For
Assume that S := {yi}i∈I is a set of skew primitive elements of H where I
is either N or {1, · · · , v} for some positive integer v. Suppose that D is a Hopf
subalgebra of H and that
(I1.2.1) gi := µ(yi) ∈ D for all i ∈ I,
(I1.2.2) S is linearly independent in the space H/D,
(I1.2.3) for each pair i ≤ j, yigj = λij gjyi + bij for some λij ∈ k× and bij ∈ D, and
there is a subalgebra A ⊂ D containing all bij such that yiA ⊂ Ayi + A and
giA ⊂ Agi + A for all i.
In most of the applications D is the coradical C0 of H and the commutators of
the yi exist. When bij = 0 for all i ≤ j, we may take A = k and then (I1.2.3) is
automatic. For every positive integer d, define
Sd := {yd1
1 · · · ydn
n · · · Xs
ds = d}.
The following lemma is known and easy to check by a direct computation.
Lemma 1.3. Suppose (I1.2.1)-(I1.2.3) hold.
(a) For every n,
n
i ) =
∆(yn
Xs=0
(cid:18)n
s(cid:19)λii
for some ass′ ∈ Pt≥0 Agt
i yn−s
gs
i ⊗ ys
i + Xs+s′<n
ass′ ys
i ⊗ ys′
i
i . If bii = 0, then ass′ = 0 for all s, s′.
6
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
(b) Let {y1, y2, · · · , yz} be a finite subset of S. Then, for n1, · · · , nz ≥ 0,
∆(yn1
1 · · · ynz
z ) = Xs1,··· ,sz
+ ldt2
z
(
Yt=1
st(cid:19)λtt
(cid:18)nt
)c(st)gs1
1 · · · gsz
z yn1−s1
1
· · · ynz−sz
z
⊗ ys1
1 · · · ysz
z
where c(st) = Qi<j λsj (ni−si)
f ∈ Pt1,··· ,tz≥0 Agt1
elements of the form f ya1
1 · · · gtz
ij
For α = (n1, · · · , nz, 0, · · · ), define
∈ k×. Here ldt2 is a linear combination of
z ⊗yb1
z with Pi(ai +bi) < Pi ni where
z . If bij = 0 for all i ≤ j, then ldt2 = 0.
1 · · · ybz
1 · · · yaz
(I1.3.1)
Lα = XG
DG
where G runs through elements ym1
1
· · · ymw
w such that (m1, · · · , mw, 0, · · · ) < α.
Lemma 1.4. Retain the notation as above and suppose (I1.2.1)-(I1.2.3) hold. Let
α = (n1, · · · , nz, 0, · · · ) and F = yn1
z . Define
1 · · · ynz
Then V is either 0 or D.
V = {a ∈ D aF ∈ Lα}.
Proof. Let L denote Lα in the proof. First we claim that ∆(L) ⊂ L ⊗ L. It suffices
to show that ∆(G) ∈ L ⊗ L for all G = ym1
w with (m1, · · · , mw, 0, · · · ) < α.
By Lemma 1.3,
· · · ymw
1
∆(G) = G ⊗ 1 + gm1
1
· · · gmw
w ⊗ G + ldt2 ∈ L ⊗ L.
Thus we proved our claim. It is easy to see that the hypotheses in Lemma 1.1(a,b)
hold. For the hypothesis in Lemma 1.1(c), we note that
∆(F ) = F ⊗ 1 + gn1
1 · · · gnz
z ⊗ F + ldt′2
by Lemma 1.3(b), where ldt′2 ∈ L ⊗ L. The assertion follows from Lemma 1.1. (cid:3)
Here is the main result of this section. Recall that gi = µ(yi) for all i.
Theorem 1.5. Assume that (I1.2.1)-(I1.2.3) hold. Let λi denote λii for all i.
(a) Suppose the elements in Sj≥0 Sj are linearly dependent over D (on the left
or on the right). Then there is some z ∈ N such that
(i) λz is a primitive pz-th root of unity for some pz > 1,
(ii) there are ai, bj ∈ k, pj ∈ N such that ypz
(iii) gi = gpz
(iv) gpj
j = gpz
part (ii).
z whenever ai 6= 0 in part (ii), and
z and λj is a primitive pj-th root of unity whenever bj 6= 0 in
z +Pi aiyi +Pj6=z bjypj
j ∈ D,
(b) Suppose λi is either 1 or not a root of unity for every i. Then the elements
in Sj≥0 Sj are linearly independent over D (on the left and on the right).
As a consequence,
GKdim H ≥ GKdim D + #(S).
Proof. (a) Suppose that Sj≥0 Sj is linearly dependent over D on the left. Then
there is an F = yn1
1 · · · ynz
z ∈ Sd for some d ≥ 0 such that
for some 0 6= a ∈ D,
aF ∈ Lα,
(I1.5.1)
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
7
where α = (n1, · · · , nz, 0, · · · ). The definition of Lα is given in (I1.3.1). Choose F
among all (a, F ) satisfying (I1.5.1) so that α is minimal with respect to the linear
order < defined in the beginning of Remark 1.2. For simplicity let L = Lα for the
rest of the proof. Let V = {b ∈ D bF ∈ L}. Then 0 6= a ∈ V . By Lemma
1.4, 1 ∈ V , or equivalently, F ∈ L. So we can write F = ldt1 where ldt1 denotes
any element in L. By the minimality of α, L is a free left D-module with a basis
{ym1
· · · ymw
w (m1, · · · , mw, 0, · · · ) < α}. Note that L ⊗ L is a free D ⊗ D-module
with a basis
1
{ym1
1
· · · ymw
w ⊗ yl1
1 · · · ylw′
w′
(m1, · · · , mw, 0, · · · ), (l1, · · · , lw′, 0, · · · ) < α}.
1
· · · ymw
We define a multi-degree on L such that, for any nonzero a ∈ D, deg(a) = 0
and deg(aym1
w ) = (m1, · · · , mw, 0, · · · ) whenever (m1, · · · , mw, 0, · · · ) < α.
Notice that under this definition L is a graded D-module (but not an algebra),
which can be viewed as a filtered D-module obviously. Extend this multi-grading
naturally to L ⊗ L by adding the multi-degrees of the tensor components.
1 · · · ynz
Recall that F = yn1
z . We may assume nz > 0 (if not, delete yz in the
expression of F ). Following the last paragraph, there is an ldt1 ∈ L such that
F = −ldt1, or equivalently, yn1
z + ldt1 = 0. By the choice of F , any element
in L has multi-degree less than α. Let ldt2 denote any element in L⊗L and let lmt2
denote any element in L ⊗ L with multi-degree less than α. Since the multi-degree
of ldt1 is less than α, ∆(ldt1) is an lmt2 by Lemma 1.3. Then, by Lemma 1.3 again,
we have
(I1.5.2)
1 · · · ynz
0 = ∆(F + ldt1) = ∆(F ) + lmt2
)c(st)gs1
1 · · · gsz
z yn1−s1
1
· · · ynz−sz
z
⊗ ys1
1 · · · ysz
z + lmt2
z
(
Yt=1
= Xs1,··· ,sz
= X(st)6=(0),(nt)
= X(st)6=(0),(nt)
(
(
z
st(cid:19)λt
(cid:18)nt
st(cid:19)λt
(cid:18)nt
Yt=1
1 · · · gnz
st(cid:19)λt
(cid:18)nt
Yt=1
z
+ F ⊗ 1 + gn1
z ⊗ F + lmt2
)c(st)gs1
1 · · · gsz
z yn1−s1
1
)c(st)gs1
1 · · · gsz
z yn1−s1
1
· · · ynz−sz
z
· · · ynz−sz
z
⊗ ys1
1 · · · ysz
z
⊗ ys1
1 · · · ysz
z + lmt2
z
1
⊗ ys1
· · · ynz−sz
1 · · · gnz
z yn1−s1
where lmt2 represents an element in L ⊗ L with multi-degree less than α. The
multi-degree of gn1
equals α for any (st) 6=
(0), (nt). Using the fact that L is a free D-module with basis {ym1
= 0
for all (st) 6= (0), (nt). If nj > 0 for some 1 ≤ j < z, we take (st) = (0, 0, · · · , 0, nz),
= 1, a contradiction. Therefore nj = 0 for all j < z which means
(m1, · · · , mw, 0, · · · ) < α}, we obtain that (Qz
then Qz
If nz = 1, we have yz = Pi<z biyi + c for c, bi ∈ D. Hence Pi<z biyi + c is
(1, gz)-primitive. Then applying ∆ we obtain that
)c(st) = 0 orQz
1 · · · ysz
st(cid:1)λt
that F = ynz
z .
t=1(cid:0)nt
st(cid:1)λt
t=1(cid:0)nt
st(cid:1)λt
t=1(cid:0)nt
· · · ymw
w
1
z
∆(bi) = bi ⊗ 1,
∆(bi)(gi ⊗ 1) = gz ⊗ bi,
∆(c) = c ⊗ 1 + gz ⊗ c.
8
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
These imply that bi ∈ k and gi = gz when bi 6= 0. This contradicts (I1.2.2).
Therefore nz > 1.
By the last two paragraphs, nz > 1 and ni = 0 for all i < z and (cid:0)nz
= 0 for
all 1 ≤ sz ≤ nz − 1. This can only happen when λz is a primitive nz-th root of
unity [GZ2, Lemma 7.5].
sz(cid:1)λz
Next let us re-name nz by pz and write F = ypz
z . Then ypz
c0 = 0 where bi, c0 ∈ D and the Gi are monomials with multi-y-degree less than
(0, · · · , 0, pz, 0, · · · ) (where pz is in the z-th position). Repeating a computation
similar to (I1.5.2) (and the induction on the multi-y-degree of Gi) one can show
that each Gi (when bi 6= 0) is of the form yni
is a skew primitive. If
ni > 1, then λi is a primitive ni-th root of unity. In summary, when λi is not a
root of unity, then ni = 1 and when λi is a primitive pi-th root of unity, then ni is
either 1 or pi. So we have
i and each yni
i
z + Pi biGi +
−ypz
z = Xi
aiyi +Xj6=z
bjypj
j + c
where 0 6= ai, bj ∈ D and c ∈ D. Thus Pi aiyi +Pj bjypj
Since L is a free left D-module, each of the nonzero aiyi, bjypj
primitive. The coproduct computation shows that ai, bj ∈ k and gi = gpz
gpj
j = gpz
z .
z )-primitive.
z )-
z and
j and c is (1, gpz
j + c is (1, gpz
(b) The first assertion is an immediate consequence of part (a). To prove the
second assertion, we take W to be a finite dimensional subspace of D and let S be
a finite set {y1, · · · , yz}. For a subspace V ⊂ H, let V n be the linear span of all
elements v1 · · · vn for vi ∈ V . By the first assertion,
dim (W + k1 +
z
Xi=1
kyi)2n ≥ dim W n(k1 +
kyi)n
z
Xi=1
[d=0
n
≥ (dim W n)#(
Sd) ≥ (dim W n)cnz
for some positive constant c. This implies that GKdim H ≥ GKdim D + #S. If S
is infinite, let S′ be any finite subset of S. Then the above argument shows that
GKdim H ≥ GKdim D + #S′ for any S′. Thus GKdim H = ∞ = GKdim D +
#S.
(cid:3)
Proof of Theorem 0.2. Let S = {y1, · · · , yw}. Then (I1.2.1)-(I1.2.3) follow easily
from (a) and (b). The hypothesis in Theorem 1.5(b) is the same as that in Theorem
0.2(c). Therefore the assertion follows from Theorem 1.5(b).
(cid:3)
The following easy lemma will be used implicitly later.
Lemma 1.6. Let P be the set of all skew primitive elements in a Hopf algebra H
with weight µ. Then P is a k-subspace of H and P ∩ C0 = k(µ − 1).
Proof. It is clear that P is a k-subspace of H. For any element y ∈ P ∩ C0, write
i=1 cigi for some ci ∈ k and gi ∈ G(H). Then the equation ∆(y) = y⊗1+µ⊗y
(cid:3)
forces that y ∈ k(µ − 1).
y = Pn
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
9
2. Second Lower Bound Theorem
In this section we prove Theorem 0.3, which is a consequence of Theorem 0.2. A
stronger version will be proved in the next section. Lemmas presented here are also
needed for the next section, and cannot be omitted even if we skip Theorem 0.3. If
GKdim H = ∞, then Theorem 0.3 is vacuous. So we may assume that GKdim H <
∞. We refer to Section 0 for the definitions of W, Ω, Γ and W√ , Ω√ , Γ√ .
Lemma 2.1. Let y be a skew primitive element not in C0 such that γ(y) is defined.
(a) If γ(y) ∈ Γ \ Γ√ , then yn is not skew primitive for any n > 1.
(b) If γ(y) ∈ Γ√ , then µ(y) ∈ W√ .
Proof. (a) Take S to be the singleton {y} and D = C0. Then (I1.2.1)-(I1.2.3) hold
for A = k[µ(y)±1]. Since γ(y) ∈ Γ \ Γ√ the hypothesis in Theorem 1.5(b) holds.
By Theorem 1.5(b), {yn}n≥0 is linearly independent over C0. Since γ(y) is not a
root of unity, for any n > 1, ∆(yn) 6∈ H ⊗ k + C0 ⊗ H by Lemma 1.3(a). Thus yn
is not a skew primitive. The assertion follows.
(b) Suppose γ(y) ∈ Γ√ . Since γ(y) 6= 1, replacing y by y + α(µ(y) − 1) for a
suitable α ∈ k, we have µ(y)−1yµ(y) = γ(y)y. Since γ(y) is a primitive n-th root of
unity for some n > 1, Lemma 1.3(a) says that ∆(yn) = yn ⊗ 1 + µ(y)n ⊗ yn, which
means that yn is a skew primitive (could be zero). Therefore µ(y) ∈ W√ .
(cid:3)
Let G(H) denote the group of all group-like elements in a Hopf algebra H. Recall
that GKdim H < ∞ by a general assumption in this section.
Lemma 2.2. Let y be a skew primitive element not in C0 and let x = µ(y).
Suppose γ(y) exists. Assume that G0 is a subgroup of G(H) commuting with x. Let
V = k(x − 1) +Pg∈G0 k(g−1yg).
(a) Every z ∈ V is (1, x)-primitive; and ω(z) = ω(y) for all z ∈ V \ k(x − 1).
(b) If γ(y) ∈ Γ \ Γ√ , then dim V ≤ GKdim H − GKdim C0 + 1.
(c) Suppose that V is finite dimensional and that G0 is abelian. Then there is
z ∈ V \ k(x − 1) such that, either
(ci) for every g ∈ G0, g−1zg = λgz for some λg ∈ k×, or
(cii) for every g ∈ G0, g−1zg = z + τg(x − 1) for some τg ∈ k.
(d) If γ(y) is not a root of unity, then µ(y) has infinite order.
Proof. (a) Since gx = xg for all g ∈ G0, g−1yg is a (1, x)-primitive with ω(g−1yg) =
ω(y).
(b) Let S = {g−1
i=1 be a finite subset of V which is linearly independent in
the space V /k(x − 1). Here gi ∈ G0 for all i = 1, · · · , w. For different i, we have
µ(g−1
i ygi}w
i ygi) = x, and
x−1(g−1
i ygi)x = g−1
i
(x−1yx)gi = γ(y)(g−1
i ygi) + τ (x − 1)
where τ is the same as the one in (I0.3.3). Then the hypotheses (I1.2.1)-(I1.2.3)
hold for A = k[x±1] and D = C0. Since λ := γ(y) is either 1 or not a root of unity,
Theorem 1.5(b) says that #S ≤ GKdim H − GKdim C0. Clearly V ∩ C0 = k(x − 1).
Thus
dim V − 1 = dim V /(V ∩ C0) = #S ≤ GKdim H − GKdim C0
since S is a basis of V /(V ∩ C0).
(c) First we may assume x ∈ G0. If not, replace G0 by the subgroup generated
by G0 and x (replacing G0 by this larger subgroup does not enlarge V , because of
10
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
(I0.3.3)). Then V is a G0-module by conjugation action. Since G0 is abelian and k
is algebraically closed, every finite dimensional simple G0-module is 1-dimensional.
Thus V has a 1-dimensional simple G0-submodule kz. If z 6∈ k(x−1), then kz being
a simple G0-module is equivalent to (ci). Otherwise, no element z ∈ V \ k(x − 1)
generates a simple G0-submodule. Hence V has a unique simple G0-submodule
M0 := k(x − 1). Note that g−1(x − 1)g = (x − 1) for all g ∈ G0, so M0 is the trivial
G0-module. Since G0 is commutative and V has only one simple submodule, every
simple sub-quotient of V must be isomorphic to the simple M0. Pick z ∈ V \k(x−1)
so that the submodule M generated by z is 2-dimensional. Then M/k(x− 1) ∼= M0,
which says that g−1zg ≡ z modulo k(x − 1). Hence g−1zg = z + τg(x − 1) for some
τg ∈ k.
(d) Let G0 = hµ(y)i. It follows from the definition that the existence of γ(y)
implies that V is finite dimensional. Applying part (ci) to the cyclic group G0 there
is a skew primitive z ∈ H \ C0 such that
g−1zg = λ(g)z
for all g ∈ G0. It is also clear that λ(µ(y)) = γ(y). Since γ(y) is not a root of
unity, the image of λ : G0 → k× is infinite. Consequently, G0 is infinite and µ(y)
has infinite order.
(cid:3)
Lemma 2.3. Let {zi}w
γ(zi) exists for each i. If the elements ω(z1), · · · , ω(zw) are distinct, then {zi}w
is linearly independent in H/C0.
i=1 be a set of skew primitive elements not in C0 such that
i=1
Proof. Suppose {zi}w
{zj}v
j=1, such that Pv
zi 6∈ C0 for any i. Applying ∆ to the equation Pv
v
v
v
i=1 is linearly dependent in H/C0. Pick a minimal subset, say
j=1 ajzj =: c ∈ C0 for some scalars aj ∈ k×. Thus v > 1 since
j=1 ajzj = c we have
Xj=1
v
j, j′.
ajzj) =
Xj=1
Xj=1
∆(c) = ∆(
ajzj ⊗ 1 +
ajµ(zj) ⊗ zj = c ⊗ 1 +
Xj=1
j=1 µ(zj) ⊗ ajzj ∈ C0 ⊗ C0. By the minimality of v, µ(zj) = µ(zj′ ) for all
Hence Pv
Set x = µ(zj) for all 1 ≤ j ≤ v. Applying the conjugation by x to the equation
j=1 ajzj = c, we obtain Pv
Pv
j=1 τj(x − 1) + x−1cx ∈ C0 for
some τj ∈ k. Using the minimality of v, γ(zj) = γ(zj′ ) for all j, j′. Thus we obtain
a contradiction. The assertion follows.
(cid:3)
j=1 γ(zj)ajzj = −Pv
ajµ(zj) ⊗ zj.
Theorem 2.4. Let {yi}w
i=1 be a set of skew primitive elements not in C0 such that
ω(y1), · · · , ω(yw) are defined and distinct elements in Ω \ Ω√ . If the subgroup G0
generated by {µ(yi)}w
i=1 is abelian, then GKdim H ≥ GKdim C0 + w.
Proof. By Lemma 2.2(a,c), for each i, there is a zi in k(µ(yi) − 1) +Pg∈G0 kg−1yig
but not in C0 such that
and, for every g ∈ G0,
ω(zi) = ω(yi),
g−1zig = λigzi + τig(µ(yi) − 1)
for some λig ∈ k×, τig ∈ k. Let A = kG0 and D = C0. Then (I1.2.1) is clear and
(I1.2.2) follows from Lemma 2.3 for the set {zi}w
i=1. (I1.2.3) is a consequence of
Lemma 2.2(c) as we have seen already. By hypothesis each λi := γ(zi) is either 1
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
11
or not a root of unity. Therefore GKdim H ≥ GKdim C0 + w by applying Theorem
1.5(b) to the set {zi}w
(cid:3)
i=1.
The next lemma is a result of [Zhu]. As before we assume that GKdim H < ∞
which is one of the hypotheses in [Zhu, Theorem 1.2].
g = µ(y). Then there is a skew primitive element z = Pn
Lemma 2.5. [Zhu, Theorem 1.2] Let y be a skew primitive element not in C0 with
i=0 big−iygi ∈ H \ C0,
where bi ∈ k, such that g−1zg = λz + τ (g − 1) for some λ ∈ k× and τ ∈ k. Further,
if λ 6= 1, then there is z′ = z + α(g − 1) for a suitable α ∈ k such that g−1z′g = λz′.
Proof. In [Zhu, Theorem 1.2] H is assumed to be pointed, but the statement is
valid without this hypothesis. The first assertion is equivalent to [Zhu, Theorem
1.2]. If λ 6= 1, take α = (λ − 1)−1τ . Then z′ = z + α(g − 1) is a (1, g)-primitive
element satisfying g−1z′g = λz′.
(cid:3)
Now we are ready to prove Theorem 0.3.
Proof of Theorem 0.3. Pick any finite subset {µ(yi)}w
i=1 of W \ W√ where each yi
is a skew primitive not in C0. By Lemma 2.5, for each i there is a skew primitive
y′i not in C0 such that gi := µ(y′i) = µ(yi) and that γ(y′i) is defined. By Lemma
2.1(b), γ(y′i) is not a root of unity or 1. Hence ω(y′i) ∈ Ω \ Ω√ . The assertion
follows from Theorem 2.4.
(cid:3)
Theorem 2.4 shows in fact that if hW \ W√ i is abelian, then
GKdim H ≥ GKdim C0 + #(Ω \ Ω√ ).
There is also an inequality
GKdim H ≥ GKdim C0 + #(W ′)
for any W ′ ⊂ W \ W√ such that hW ′i is abelian.
Suppose there is a surjective Hopf algebra morphism π : H → C0 such that the
restriction to C0 is the identity. Let A be the subalgebra of H generated by all skew
primitive elements in ker π and let GW be the sub-semigroup of G(H) generated by
µ(y) for all skew primitive elements y ∈ A. We do not assume that GW is abelian.
Lemma 2.6. Suppose there is a surjective Hopf algebra morphism π : H → C0
such that the restriction to C0 is the identity. Let A be defined as above.
(a) H = R#C0 where R is the ring of right coinvariants of π. Then A is a
subalgebra of R and
GKdim H ≥ GKdim R + GKdim C0 ≥ GKdim A + GKdim C0.
(b) Assume that A is a domain. Then GKdim A ≥ GKdim kGW . As a conse-
quence,
GKdim H ≥ GKdim kGW + GKdim C0.
Proof. (a) By [Mo, Theorem 7.2.2], H is isomorphic to a crossed product R#σC0
as algebras and by [Mo, Proposition 7.2.3], σ is trivial. Hence H = R#C0 where R
is the ring of right coinvariants of π. It is clear that every skew primitive element
in ker π is in R. Therefore A ⊂ R.
Since H = R#C0, GKdim H ≥ GKdim R + GKdim C0. The assertion follows by
the fact A ⊂ R.
12
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
(b) Define a map ρ : A → C0 ⊗ H to be the composition (π ⊗ IdH ) ◦ ∆. Since
ρ(y) ∈ kGW ⊗ A for all skew primitive elements y ∈ A and since A is generated by
these y's, the image of ρ is in kGW ⊗A. Consequently, (A, ρ) is a left kGW -comodule
algebra. This means that A is a GW -graded algebra. Let f : A → C0 be the map
sending any nonzero homogeneous element h ∈ A to its degree, for example, sending
y1 · · · yn to µ(y1) · · · µ(yn). Since A is a domain, f is multiplicative.
Pick any finite dimensional space V = k + Pm
i=1 kµ(yi) of kGW where the yi
are skew primitive elements in A, let W = {1} ∪ {yi}w
i=1. Then dim(kW )n ≥
#(f (W n)) ≥ #(f (W ))n ≥ dim V n for all n. Hence GKdim A ≥ GKdim kGW . (cid:3)
The next example shows why we need to remove W√ from W (or remove Γ√
from Γ) in the lower bound theorems.
Example 2.7. Let B be the Hopf algebra B(1, 1, p1, · · · , ps, q) defined in [GZ2,
Construction 1.2]. This is a finitely generated, noetherian, pointed Hopf domain of
GK-dimension 2. By [GZ2, Construction 1.2] B is generated by x, x−1, y1, · · · , ys
where x is a group-like element and yi's are skew primitive elements. Let z =
1 . Then z = ypj
yp1
for all j and it is a central skew primitive element. Let H =
B/(z, xm − 1) where m = Qi pi. Then H is a finite dimensional pointed Hopf
By [GZ2, Construction 1.2], W = W√ = {xmi}s
algebra of GK-dimension 0 and C0 = k[x, x−1]/(xm − 1) has GK-dimension 0.
j
i )}s
i=1 where mi = m/pi, Γ = Γ√ =
i=1 kyi.
i=1, and Z = Y√ + C0 and Y√ = Ps
i }s
i=1, Ω = Ω√ = {(xmi, q−m2
{q−m2
Thus
#(W√ ) = #(Γ√ ) = #(Ω√ ) = dim Y√ = s
which can be arbitrarily large.
3. Third lower bound theorem
The first half of this section concerns some preliminary analysis of Hopf algebras
with exponential growth and the proof of Proposition 0.5. The proof of the third
lower bound theorem is given at the end of the section.
Let (G0, ×) be a multiplicative abelian group and Λ := {λ1, · · · , λv} be a list
of 1-dimensional group representations of G0 for some v > 1. Note that this list
is allowed to have repetitions. When some λi is the trivial representation of G0
(namely, λi(g) = 1 for all g ∈ G0), then we also need a group homomorphism
τi : (G0, ×) → (k, +) (which must be zero if G0 is torsion since char k = 0). When
λi is not trivial, we set τi = 0.
Now pick a list of elements µ := {µ1, · · · , µv} in G0 (again allowing repetitions).
Let K := K(Λ, µ) be the Hopf algebra generated as an algebra by the elements in
the abelian group G0 and a set of skew primitive elements y1, · · · , yv subject to the
relations within G0 and the following additional relations between G0 and {yi}v
s=1,
yig = λi(g)gyi + τi(g)g(µi − 1), for all i and all g ∈ G0.
The coalgebra structure of K is determined by
∆(g) = g ⊗ g,
ǫ(g) = 1,
∆(yi) = yi ⊗ 1 + µi ⊗ yi,
ǫ(yi) = 0,
for all g ∈ G0,
for all i = 1, · · · , v.
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
13
And the antipode of K is determined by
S(g) = g−1,
S(yi) = −µ−1
i yi,
for all g ∈ G0,
for all i = 1, · · · , v.
Let λij = λi(µj) for all i, j and let ΛM be the v × v-matrix (λij ).
By Remark 1.2, the total y-degree and the multi-y-degree are defined for elements
in K. For example, the (filtered) multi-y-degree of gy3y2 is (0, 1, 1, 0, · · · , 0) ∈ Nv.
Let F be a nonzero skew primitive element in K with total y-degree z ≥ 2. Write
F = P ch,(is)hyi1 yi2 · · · yin where h ∈ G0 and 0 6= ch,(is) ∈ k. A term of F means
a nonzero monomial ch,(is)hyi1yi2 · · · yin appearing in F .
Lemma 3.1. Let K := K(Λ, µ) be defined as above.
(a) K has a k-linear basis
{gyi1yi2 · · · yis}
where g ∈ G0, i1, · · · , is ∈ {1, · · · , v}. As a consequence, K contains a free
subalgebra khy1, y2i and has exponential growth.
(b) The coradical of K is kG0.
(c) If F is a skew primitive element of total y-degree z ≥ 2, then for any term
of F with multi-y-degree (N1, · · · , Nv) and Pi Ni = z,
v
Yi=1
(λii)Ni(Ni−1)Yi<j
(λij λji)NiNj = 1.
Proof. (a) The first assertion follows from Bergman's Diamond Lemma [Be, Theo-
rem 1.2]. Consequently, K contains the free algebra of rank 2, khy1, y2i. Therefore
K has exponential growth.
(b) By definition, ∆ is compatible with filtrations defined in Remark 1.2. Hence
∆ is a homomorphism of filtered algebras. So every group-like element must have
total y-degree 0. The assertion follows.
let ldt denote any linear combination of monomials of total y-degree less than z.
(c) Let F = P ch,(is)hyi1 yi2 · · · yin with coefficients ch,(is) 6= 0. For simplicity,
Then we can write F = P ch,(is)hyi1 yi2 · · · yiz + ldt. Since ∆(F ) = F ⊗ 1 + µ(F ) ⊗
Since F is skew primitive, S(F ) = −µ(F )−1F . Since S(yi) = −µ−1
F , h = 1 for terms with total degree z. Pick any term of y-degree z in F , say
c1,(is)yi1 yi2 · · · yiz , and let (N1, · · · , Nv) be its multi-y-degree.
i yi, we have
S(c1,(is)yi1 · · · yiz ) = c1,(is)(−µ−1
iz
= c1,(is)(−1)z Ys>t
yiz ) · · · (−µ−1
i1 yi1 )
λ−1
isit
µ−1yiz · · · yi1 + ldt
where µ = Qz
term of the form c′(is)yiz · · · yi1. The same computation shows that
s=1 µis. Since S(F ) = −µ(F )−1F , µ = µ(F ) and F contains a nonzero
S(c′(is)yiz · · · yi1 ) = c′(is)(−µ−1
i1 yi1 ) · · · (−µ−1
iz
yiz )
= c′(is)(−1)z Ya<b
λ−1
iaib
µ−1yi1 · · · yiz + ldt.
14
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
Comparing the coefficients in the terms µ−1yiz · · · yi1 and µ−1yi1 · · · yiz in the equa-
tion S(F ) = −µ−1F , we have
−c′(is) = c1,(is)(−1)z Ys>t
, −c1,(is) = c′(is)(−1)z Ya<b
Since c1,(is) and c′(is) are nonzero, the above two equations imply
λ−1
isit
λ−1
iaib
.
or
(I3.1.1)
(λisit λit is) = 1
Ys>t
Y{s6=t}⊂{1,2,··· ,z}
(λisit λitis ) = 1.
We know the monomial yi1 · · · yin contains Ni copies of yi for all i = 1, · · · , v. Hence
equation (I3.1.1) is in fact
v
Yi=1
(λii)Ni(Ni−1)Yi<j
(λij λji)NiNj = 1.
There is a slight modification of Lemma 3.1. Suppose λ11 is a primitive p1-th
root of unity for some p1 > 1. Recycle most of the notations before Lemma 3.1. Let
L := L(Λ, µ, p1) be the Hopf algebra generated as an algebra by the abelian group
G0 and y1, · · · , yv subject to the relations within G0 and the following additional
relations between G0 and {yi}v
s=1
(cid:3)
yig = λi(g)gyi + τi(g)g(µi − 1), for all i and all g ∈ G0,
yp1
1 = β(µp1
for some β ∈ k.
1 − 1),
The coalgebra structure of L is determined by
∆(g) = g ⊗ g,
ǫ(g) = 1,
∆(yi) = yi ⊗ 1 + µi ⊗ yi,
ǫ(yi) = 0,
for all g ∈ G0,
for all i = 1, · · · , v.
And the antipode of L is determined by
S(g) = g−1,
S(yi) = −µ−1
i yi,
for all g ∈ G0,
for all i = 1, · · · , v.
Define ΛM := (λij ) = (λi(µj)). The total y-degree and the multi-y-degree are
defined as before.
Lemma 3.2. Let L := L(Λ, µ, p1) be defined as above. Suppose either β = 0 or
λ1(g)p1 = 1 for all g ∈ G0.
(a) L has a k-linear basis
{gyi1yi2 · · · yis}
where g ∈ G0, i1, · · · , is ∈ {1, · · · , v} and there is no u such that iu =
iu+1 = · · · iu+p1−1 = 1. As a consequence, L contains a free subalgebra
khy1y2, y1y2
2i and has exponential growth.
(b) The coradical of L is kG0.
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
15
(c) If F is a skew primitive element of total y-degree z ≥ 2, then for any term
of F with multi-y-degree (N1, · · · , Nv) and Pi Ni = z,
v
Yi=1
(λii)Ni(Ni−1)Yi<j
(λij λji)NiNj = 1.
Proposition 3.3. Let H be a Hopf algebra and y1, y2 be two skew primitive ele-
ments linearly independent in H/C0. Suppose that
(i) there is a group-like element x and d1, d2 ∈ Z such that µ(yi) = xdi for
i = 1, 2, and
(ii) there are two scalars q1, q2 ∈ k× such that yix = qixyi for i = 1, 2.
(a) If x has infinite order and H does not contain a free algebra of rank 2, then
(I3.3.1)
qd1(M1(M1−1))+d2(M1M2)
1
qd2(M2(M2−1))+d1(M1M2)
2
= 1
for some integers M1, M2 ≥ 0 satisfying M1 + M2 ≥ 2.
(b) Assume that one of the following holds:
1 = 1 and q2 is not a root of unity and d1d2 > 0;
1 6= 1 is a root of unity and q2 is not a root of unity and d1d2 > 0;
(1) q1 = q2 is not a root of unity and d1d2 > 0;
(2) qd1
(3) qd1
(4) the group hq1, q2i ⊂ k× is free abelian of rank 2, d1d2 6= 0.
Then H contains a free subalgebra of rank 2. Consequently, H has expo-
nential growth.
Proof. (a) Let µi = xdi and λij = qdj
i, j ∈ {1, 2}.
i
. Then yiµj = λij µjyi and γ(yi) = λii for all
Let H0 be the Hopf subalgebra generated by x, x−1, y1, y2. Let G0 = hgi ∼= Z
and let λi(gn) = qn
for i = 1, 2 and all n. Let Λ = {λ1, λ2} and µ = {g, g}. Then
i
there is a surjective Hopf algebra homomorphism φ : K := K(Λ, µ) → H0 sending
g 7→ x and yi 7→ yi for i = 1, 2, where we choose τi = 0. By Lemma 3.1(a) K
contains a free algebra of rank 2. If H does not contain a free algebra of rank 2,
then K → H0 is not injective. By [Mo, Theorem 5.3.1], there is a nonzero skew
primitive element F ∈ K such that φ(F ) = 0. Since φ is injective on skew primitive
elements of y-degree ≤ 1, F has total y-degree z ≥ 2. By Lemma 3.1(c), for any
term of F with multi-y-degree (M1, M2) and M1 + M2 = z, we have the following,
(λ11)M1(M1−1)(λ22)M2(M2−1)(λ12λ21)M1M2 = 1
or equivalently,
(qd1
1 )M1(M1−1)(qd2
2 )M2(M2−1)(qd2
1 qd1
2 )M1M2 = 1.
This can be simplified to
qd1(M1(M1−1))+d2(M1M2)
1
qd2(M2(M2−1))+d1(M1M2)
2
= 1
which is (I3.3.1).
(b) Assume H does not contain a free algebra of rank 2 and we will obtain a
contradiction. If one of the hypotheses holds, then x has infinite order in G(H) by
Lemma 2.2(d). Therefore we can apply part (a).
In case (b1), (I3.3.1) implies that
d1(M1(M1 − 1)) + d2(M1M2) + d2(M2(M2 − 1)) + d1(M1M2) = 0.
16
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
This is impossible since d1d2 > 0 and M1 + M2 ≥ 2. Therefore H contains a free
algebra of rank 2.
A similar argument works for case (b4).
In case (b2), equation (I3.3.1) implies that (qd2(M2(M2−1))+d1(M1M2)
2
)d1 = 1, or
d2(M2(M2 − 1)) + d1(M1M2) = 0
because q2 is not a root of unity. Since d1d2 > 0, the only solution is M2 = 0 and
1 + ldt for some c ∈ k×. By Lemma 1.3 and the
M1 = z ≥ 2. Thus we have F = cyz
fact that λ11 = qd1
1 = 1, F cannot be skew primitive for any z ≥ 2. So case (b2)
has been taken care of.
It remains to consider case (b3). Suppose qd1
1
6∈ C0, then {yp1
1 + βy2 ∈ C0, then x−1(αyp1
is a primitive p1-th root of unity.
1 , y2} is linearly in-
1 + βy2)x ∈ C0,
1 6= q2. The assertion follows from case
1 − 1) for some β. Replacing
1 = β(µp1
is a skew primitive element. If yp1
1
Then yp1
1
dependent in H/C0. Note that if αyp1
which would imply yp1
(b2) applied to {yp1
K by L in the above argument, (I3.3.1) holds again.
1 , y2 ∈ C0 because qp1
1 ∈ C0, then yp1
1 , y2}. If yp1
Since q1 is a root of unity, we have
(qp
2 )d2(M2(M2−1))+d1(M1M2) = 1
for some p > 1. Since q2 is not a root of unity,
d2(M2(M2 − 1)) + d1(M1M2) = 0.
Since d1d2 > 0, the only solution is M2 = 0 and M1 = z ≥ 2. Now F = cyz
1 + ldt,
where c ∈ k× and z < p1. By Lemma 1.3, F cannot be skew primitive. This is a
contradiction.
(cid:3)
Corollary 3.4. Suppose H has subexponential growth. Let y be a skew primitive
element not in C0 such that γ(y) is defined and is not a root of unity. Let G0 be
a finitely generated abelian subgroup of G(H) containing µ(y) (which has infinite
order automatically). Then V := k(µ(y) − 1) +Pg∈G0 k(g−1yg) is 2-dimensional.
As a consequence, there is a group representation λ : G0 → k such that
for all g ∈ G0, where y′ = y + α(µ(y) − 1) for some α ∈ k.
g−1y′g = λ(g)y′
Proof. Since γ(y) is not a root of unity, we may assume that yµ(y) = γ(y)µ(y)y
after replacing y by y +α(µ(y)−1) for some α ∈ k. Let g ∈ G0. Let y1 = y and y2 =
g−1yg. Then γ(y1) = γ(y2) and it is not a root of unity. By Proposition 3.3(b1), y1
and y2 are not linearly independent in H/C0. The assertion that dim V = 2 follows
by applying Lemma 1.6. The consequence follows from Lemma 2.2(ci).
(cid:3)
Proof of Proposition 0.5. We prove the assertion by contradiction. So we assume
that H has subexponential growth.
Pick a pair of skew primitive elements (y1, y2) such that the subgroup of hΓi
generated by {γ(y1), γ(y2)} has rank 2. Let λii = γ(yi) and gi = µ(yi) for i = 1, 2.
Since hW i is abelian, the subgroup G0 := hg1, g2i is abelian. By Corollary 3.4,
we may further assume that, for i = 1, 2 and j = 1, 2, g−1
j yigj = λij yi for some
λij ∈ k×. Since H does not contain a free subalgebra of rank 2, the proofs of
Proposition 3.3 and (I3.3.1) show that
(I3.4.1)
(λ11)M1(M1−1)(λ22)M2(M2−1)(λ12λ21)M1M2 = 1
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
17
for some non-negative M1, M2 with M1 + M2 ≥ 2.
Since G0 is a finitely generated subgroup of hW i and since hW i is abelian and
torsionfree of rank 1, G0 is isomorphic to Z. Therefore there is an x ∈ G0 such that
g1 = xa and g2 = xb for some nonzero integers a, b. Consequently, ga
1. Thus
the equation g−1
i1 for all i = 1, 2. Then (I3.4.1)
implies that
j yigj = λij yi implies that λa
i2 = λb
2 = gb
(λ11)abM1(M1−1)(λ22)abM2(M2−1)(λ11)b2M1M2(λ22)a2M1M2 = 1
Since the rank of hλ11, λ22i is 2, we have
abM1(M1 − 1) + b2M1M2 = abM2(M2 − 1) + a2M1M2 = 0
or
a(M1 − 1) + bM2 = b(M2 − 1) + aM1 = 0.
Since a, b are nonzero, this means that (M1 − 1)(M2 − 1) − M1M2 = 0. This is
impossible when M1 + M2 ≥ 2, which yields a contradiction.
(cid:3)
The rest of this section is devoted to the proof of Theorem 0.4. The next def-
inition was given in the introduction, but maybe it should be reviewed here. Let
y be a (1, g)-primitive element in a Hopf algebra H. Let Tg−1 denote the inverse
conjugation by g, namely, Tg−1 : a → g−1ag.
Definition 3.5. Let y be a (1, g)-primitive element of H not in C0. A nonzero
scalar λ is called the commutator of y of level n if (Tg−1 − λIdH )n(y) ∈ C0 and
(Tg−1 − λIdH )n−1(y) 6∈ C0. In this case we write γ(y) = λ. When n = 1, γ(y) is
the commutator of y defined as in (I0.3.2) or equivalently in (I0.3.3).
In general the commutator of y may not exist. We also need a generalization of
Definition 3.5. Recall that W× is the subset of W consisting of weights µ(y) such
that the commutator of y is either 1 or not a root of unity. Throughout the rest
of the section let G0 be the subgroup hW×i and suppose that G0 is abelian. A 1-
dimensional representation of G0 is equivalent to a multiplicative map λ : G0 → k×.
Let G∗0 denote the set of 1-dimensional representations of G0, which is also called
the character group of G0.
Definition 3.6. Let y be a skew primitive element in H \ C0 and let λ ∈ G∗0.
We say λ is the generalized commutator of y of level n if there is an n such that
(Tg−1 − λ(g)IdH )n(y) ∈ C0 for all g ∈ G0 and (Tg−1 − λ(g)IdH )n−1(y) 6∈ C0 for
some g ∈ G0. If n = 1, λ is called the generalized commutator of y.
Lemma 3.7. In parts (a) and (c) suppose that GKdim H < ∞.
(a) Every skew primitive element y ∈ H is a linear combination of skew primi-
tives with commutator of finite level and weight µ(y).
(b) Suppose V is a conjugation G0-stable finite dimensional subspace spanned
by skew primitives with their weights in G0. Then every element y ∈ V is a
linear combination of skew primitives with generalized commutator of finite
level.
(c) Every skew primitive y ∈ H with commutator of level 1 such that γ(y) ∈ Γ \
Γ√ is a linear combination of skew primitives with generalized commutator
of finite level. Further, each nonzero summand of the linear combination
has weight commutator ω(y).
18
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
Proof. (a) If y ∈ C0, then y = α(µ(y) − 1) for some α ∈ k [Lemma 1.6] and the
commutator of y has level 0 by definition.
If y 6∈ C0, then, by Lemma 2.5, V := Pn∈Z k(g−nygn) + k(g − 1) is finite
dimensional over k, where g = µ(y). Then Tg−1 acts on V as an invertible linear
map. Pick a basis of V so that the presentation of Tg−1 with respect to the basis is
in the Jordon canonical form. Then each basis element is a skew primitive element
with commutator of finite level. The assertion follows.
(b) Write V = Lm
j=1 Vj where each Vj is spanned by skew primitives with weight
gj for distinct group-like elements g1, · · · , gm ∈ G0. Since G0 is abelian, each Vj is
conjugation G0-stable. Passing from V to Vj we may assume that each element in
V is a skew primitive of weight g.
For every h ∈ G0, Th−1 acts on V as an invertible linear map. It is clear that
V is a finite dimensional kG0-module. Since kG0 is commutative, every finite
dimensional simple G0-module is 1-dimensional and Ext1
kG0 (S, S′) = 0 if S and S′
are distinct simple modules over kG0. Then V is a finite direct sum of submodules
Vi so that the support of each Vi is a single closed point of Spec kG0. This closed
point corresponds to a 1-dimensional G0-representation λi. Fix any i, every element
in Vi has a generalized commutator λi with level no more than dim Vi.
(c) Let V be the vector space spanned by all skew primitive elements z with
commutator of level 1 such that ω(z) = ω(y). Pick any finite set {z1, · · · , zw}
which is linearly independent in V /(V ∩ k(µ(y) − 1)). Then (I1.2.1)-(I1.2.3) hold
for D = C0 and A = k[µ(y)±1]. By Theorem 1.5(b), w ≤ GKdim H. Thus V is
finite dimensional. Clearly, Tg−1 stabilizes V for all g ∈ G0. The first assertion
follows from part (b). The final assertion is clear since every element in V has
weight commutator equal ω(y).
(cid:3)
We need to introduce some conventions. Let µ ∈ W . Define Pµ to be the k-
linear space spanned by all skew primitives y ∈ H with µ(y) = µ. Let γ be a
nonzero scalar and let Pµ,γ,n be the k-linear space spanned by all skew primitives
y ∈ Pµ with commutator γ of level no more than n. Let Pµ,γ,∗ = Pn≥0 Pµ,γ,n,
P∗,∗,n = Pµ,γ Pµ,γ,n and P∗,∗,∗ = Pµ,γ,n Pµ,γ,n.
Given any µ ∈ W and λ ∈ G∗0, let Pµ,λ,n be the k-linear space spanned by all
skew primitives y ∈ H with generalized commutator λ of level no more than n
and µ(y) = µ. Let Pµ,λ,∗ = Pn≥0 Pµ,λ,n, P∗,G∗,n = Pµ,λ Pµ,λ,n and P∗,G∗,∗ =
Pµ,λ,n Pµ,λ,n.
Lemma 3.7(a) says that P∗,∗,∗ contains all skew primitive elements and Lemma
3.7(c) says that Pµ,γ,1 is a subspace of P∗,G∗,∗ when γ ∈ k× and γ is 1 or not a root
of unity.
Lemma 3.8. Retain the above notation. Suppose that H does not contain a free
subalgebra of rank 2.
(a) If Pµ,γ,1 ⊂ k(µ − 1), then Pµ,γ,∗ ⊂ k(µ − 1).
(b) If Pµ,γ,1 6⊂ k(µ − 1) and γ is not a root of unity, then Pµ,γ0,∗ ⊂ k(µ − 1) for
every root of unity γ0.
(c) If γ is not a root of unity, then Pµ,γ,1/k(µ − 1) has dimension at most 1.
(d) Suppose γ1 and γ2 are two distinct scalars neither of which is a root of unity.
If Pµ,γi,1 6⊂ k(µ − 1) for i = 1, 2, then γN
2 = 1 for some positive integers
N, M . Further there is no γ3 ∈ k× \ {γ1, γ2} such that Pµ,γ3,1 6⊂ k(µ − 1).
1 γM
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
19
(e) Suppose that GKdim H < ∞. If γ is not a root of unity, then Pµ,γ,∗ =
Pµ,γ,1 ⊂ P∗,G∗,∗.
Proof. (a) This is clear by induction.
(b) First assume that γ0 is not 1. Pick y2 ∈ Pµ,γ,1 \ C0. If Pµ,γ0,∗ 6⊂ k(µ − 1) for
a root of unity γ0, by part (a), there is a y1 ∈ Pµ,γ0,1 \ C0. Since γ0 and γ are not 1,
after adding a suitable term α(µ(yi) − 1) to yi, the hypothesis in Proposition 3.3(ii)
holds. Then we are in the situation of Proposition 3.3(b2,b3) (where d1 = d2 = 1).
By Proposition 3.3, H contains a free subalgebra of rank 2, a contradiction.
Second assume that γ0 = 1. For the statement we only need to consider the Hopf
subalgebra generated by group-like and skew primitive elements. So we may assume
that H is pointed. Passing to the associated graded Hopf algebra of H with respect
to its coradical filtration, one can assume the hypothesis in Proposition 3.3(ii) holds.
So the above proof works.
(c) This follows from Proposition 3.3(b1).
(d) By the proof of Proposition 3.3, see (I3.3.1),
qd1(M1(M1−1))+d2(M1M2)
1
qd2(M2(M2−1))+d1(M1M2)
2
= 1
where M1 and M2 are nonnegative integers with M1 + M2 ≥ 2. Note that γi = qi
and d1 = d2 = 1 in this case. Let N = M1(M1 − 1) + M1M2 and M = M2(M2 −
1) + M1M2. Then N and M are non-negative and N + M ≥ 2. Since γ1 and γ2
are not roots of unity, both N and M must be positive.
If such a γ3 exists, by part (b) it is not a root of unity. By the above assertion
we have positive integers A, B, C, D such that γA
1 γB
3 = 1 and γC
2 γD
3 = 1. Then
1 = (γN
1 γM
2 )AC = γACN
1
γCAM
2
= γ−BCN−DAM
3
which contradicts the fact γ3 is not a root of unity.
(e) If Pµ,γ,∗ 6= Pµ,γ,1, pick a skew primitive y2 ∈ Pµ,γ,2. This means that
µ−1y2µ − γy2 = y1 ∈ Pµ,γ,1 \ k(µ − 1). Consider the Hopf subalgebra K generated
as an algebra by y2, y1, µ, µ−1. By possibly adding a term β(µ − 1) to y1 for some
β ∈ k and adding a term α(µ − 1) to y2 for some α ∈ k, we can assume that
y1µ = γµy1 and y2µ = γµy2 + µy1.
It remains to show that K contains a free subalgebra of rank 2. Define a filtration
Fi of K inductively as follows:
F0 = C0(K) = k[µ±1],
F1 = F0 + F0y1 = F0y1 + F0,
F2 = F0 + F0y1 + F0y2
1 + F0y2 = F0 + y1F0 + y2
1F0 + y2F0.
Fn = Pn−1
i=1 FiFn−i for all n ≥ 3.
It is easy to check that F is an Hopf algebra filtration and grF K is a Hopf
algebra. In grF K, y1, y2 are linearly independent in grF K/k(µ − 1). Since y1, y2 ∈
Pµ,γ,1(K), by part (c), grF K contains a free subalgebra of rank 2. Therefore
K contains a free subalgebra of rank 2. This yields a contradiction. Therefore
the first assertion (namely, the first equation) follows. Finally, by Lemma 3.7(c),
Pµ,γ,1 ⊂ P∗,G∗,∗.
(cid:3)
Remark 3.9. Suppose GKdim H < ∞. By Lemmas 2.1(b) and 3.7(a), W \ W√ ⊆
W×. In practice it often happens that W \ W√ = W×.
20
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
Now we are ready to prove the Third Lower Bound Theorem. Let Y∗ be the
k-linear vector space spanned by all skew primitive elements y with commutator of
finite level such that γ(y) ∈ Γ \ Γ√ .
Theorem 3.10. Suppose G0 is abelian. Then
GKdim H ≥ GKdim C0 + dim Y∗/(Y∗ ∩ C0).
Proof. Nothing needs to be proved if GKdim H = ∞, so we assume GKdim H < ∞.
Let
Yn = Xµ∈W×,γ∈Γ\Γ√
Pµ,γ,n
and YGn =
X
µ∈W×,λ∈G∗0 ,γ:=λ(µ)∈Γ\Γ√
Pµ,λ,n
for all n ≥ 1. Then Y∗ = Pn Yn. Let YG∗ = Pn YGn. We prove the following claim
by induction:
Claim A:
GKdim H ≥ GKdim C0 + dim YGn/(YGn ∩ C0)
for all n ≥ 1. When n = 1, let {yi} be a basis of YG1/(YG1 ∩ C0) such that each yi
is in Pµi,λi,1 for some µi, λi. Then we have
µ−1
j yiµj = λi(µj)yi + τij (µi − 1)
where λi(µi) = γ(yi) is either 1 or not a root of unity and where τij ∈ k. By
Theorem 1.5 (with D = C0 and A = kG0),
GKdim H ≥ GKdim C0 + #{yi} = GKdim C0 + dim YG1/(YG1 ∩ C0),
which proves Claim A for n = 1.
Now assume that Claim A holds for n. Without loss of generality we may
assume that H is generated as an algebra by C0 and all skew primitive elements of
H. Define a filtration Fi of H as follows:
F0 = C0,
F1 = F0 + F0YG1 = F0 + YG1F0,
F2 = F 2
1 + F0Z where Z is the k-linear span of all skew primitive elements of H,
and
Fm = Pm−1
i=1 FiFm−i for all m ≥ 3.
Then {Fm} is a Hopf algebra filtration of H. Let K be the associated graded Hopf
algebra grF H. Note that if y ∈ Pµ,λ,2 ⊂ YG2(H), then y ∈ F2 and the associated
element in K is gr y ∈ F2/F1 and g−1(gr y)g = λ(g)(gr y) for all g ∈ G0 (or, when
trivially y ∈ F1, we have gr y ∈ F1/F0 or gr y ∈ F0). Thus gr y ∈ YG1(K) (which is
easy to see when y ∈ F1). By induction one sees that
YGn(K) ⊇ {gr y y ∈ YG(n+1)(H)} =: gr YG(n+1)(H)
for all n. Applying the induction hypothesis to K,
GKdim K ≥ GKdim C0 + dim YGn(K)/(YGn(K) ∩ C0).
By [KL, Lemma 6.5], GKdim H ≥ GKdim K. It is clear that
dim YG(n+1)(H)/(YG(n+1)(H) ∩ C0) = dim gr YG(n+1)(H)/(gr YG(n+1)(H) ∩ C0).
Therefore
GKdim H ≥ GKdim K ≥ GKdim C0 + dim YGn(K)/(YGn(K) ∩ C0)
≥ GKdim C0 + dim gr YG(n+1)(H)/(gr YG(n+1)(H) ∩ C0)
= GKdim C0 + dim YG(n+1)(H)/(YG(n+1)(H) ∩ C0)
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
21
which finishes the induction step. Therefore we proved Claim A.
When n goes to infinity, we have
(I3.10.1)
GKdim H ≥ GKdim C0 + dim YG∗(H)/(YG∗(H) ∩ C0).
Next we prove the following claim by induction:
Claim B:
GKdim H ≥ GKdim C0 + dim Yn/(Yn ∩ C0).
When n = 1, this follows from (I3.10.1) since Y1(H) ⊂ YG∗(H) by Lemma 3.7(c).
Note that Y1 is G0-stable. Using an argument similar to the proof of Claim A by
passing to the associated graded Hopf algebra grF H (and replacing YGn by Yn),
one sees that Claim B holds. When n goes to infinity, we have
GKdim H ≥ GKdim C0 + dim Y∗(H)/(Y∗(H) ∩ C0).
(cid:3)
The proof of Theorem 3.10 also shows the following.
Corollary 3.11. Suppose G0 is abelian and GKdim H < ∞. Let µ ∈ W and
suppose that γ ∈ k× is not a root of unity. Then
(a) Pµ,1,∗ is finite dimensional.
(b) Pµ,1,∗ and Pµ,γ,∗ are subspaces of P∗,G∗,∗.
(c) Y∗ = YG∗.
By Lemma 3.8(c,e), Pµ,γ,∗ is finite dimensional.
Proof of Corollary 3.11. (a) By Theorem 3.10, Y∗/(Y∗ ∩ C0) is finite dimensional.
Since Pµ,1,∗ ⊂ Y∗ and Pµ,1,∗ ∩ C0 = k(µ − 1), we have
dim Pµ,1,∗/k(µ − 1) ≤ dim Y∗/(Y∗ ∩ C0) < ∞
which implies that Pµ,1,∗ is finite dimensional.
(b) By Lemma 3.8(e) Pµ,γ,∗ is a subspace of P∗,G∗,∗.
By part (a) Pµ,1,∗ is finite dimensional. It is clear that Pµ,1,∗ is G0-stable. By
Lemma 3.7(b,c) Pµ,1,∗ is a subspace of P∗,G∗,∗.
(c) As a consequence of part (b), Pµ,λ,∗ = Pµ,λ(µ),∗ when λ(µ) is either 1 or not
a root of unity. The assertion follows.
(cid:3)
Theorem 0.4 is an immediate consequence of Theorem 3.10.
Proof of Theorem 0.4. Without loss of generality we assume that GKdim H < ∞.
First we claim that Y∗ ∩ (Y√ + C0) ⊂ C0. Suppose y = z + c is in Y∗ ∩ (Y√ + C0)
where y ∈ Y∗ \ C0 and z ∈ Y√ and c ∈ C0. It is easily reduced to the case when
µ(y) = µ(z) = g and c = α(g − 1) for some α ∈ k. Then y ∈ Y√ , a contradiction.
Therefore the claim holds.
By Lemma 3.7(a), every skew primitive element is a linear combination of skew
primitives with commutator of finite level. This says that Z = Y∗ + Y√ + C0. Since
Y∗ ∩ (Y√ + C0) ⊂ C0, we have
Z/(C0 + Y√ ) ∼= Y∗/Y∗ ∩ (C0 + Y√ ) = Y∗/(Y∗ ∩ C0).
The assertion follows by Theorem 3.10.
(cid:3)
22
D.-G. WANG, J.J. ZHANG AND G. ZHUANG
Another way of proving Theorem 0.3 (with a slightly stronger hypothesis that
hW×i is abelian) is using Theorem 0.4 and the following lemma, which is due to an
anonymous referee.
Lemma 3.12. Suppose GKdim H < ∞. Then
dim Z/(C0 + Y√ ) ≥ #(W \ W√ ).
Proof. For any w ≤ #(W \ W√ ), take distinct elements x1, · · · , xw ∈ W \ W√ , we
need to prove that dim Z/(C0 + Y√ ) ≥ w.
For each i, pick a skew primitive element zi ∈ H \ C0 such that xi = µ(zi).
Lemma 2.5 shows how to get a suitable zi such that γ(zi) is defined. Since µ(zi) =
xi 6∈ W√ , Lemma 2.1(b) says that γ(zi) is either 1 or not a root of unity.
It suffices to show that z1, · · · , zw are linearly independent in Z/(C0 + Y√ ), so
it is enough to show that z1, · · · , zw, y are linearly independent in Z/C0 for any
y ∈ Y√ \ C0. We can arrange y = y1 + · · · + yv for some skew primitive elements
yj 6∈ C0, where yj has a commutator γ(yj) of finite level which is a nontrivial root of
unity, and the pairs (µ(yj ), γ(yj)) are distinct. The pairs (µ(zi), γ(zi)) are already
distinct, and (µ(zi), γ(zi)) 6= (µ(yj), γ(yj)) for all i, j because γ(zi) is either 1 or
not a root of unity.
An improved version of Lemma 2.3 for skew primitive elements with commutators
of finite level says that z1, · · · , zw, y1, · · · , yv are linearly independent in Z/C0.
Consequently, z1, · · · , zw, y are linearly independent in Z/C0 as desired.
(cid:3)
Finally we end with a simple example in which Z = Y√ + C0.
Example 3.13. Let H be generated as an algebra by x and {yi}∞i=1 subject to the
relations
x2 = 1,
y2
i = 0,
yiyj + yjyi = 0,
xyi + yix = 0
for all i, j ∈ N. The coalgebra structure and the antipode of H are determined by
∆(x) = x ⊗ x,
ǫ(x) = 1,
S(x) = x,
∆(yi) = yi ⊗ 1 + x ⊗ yi,
ǫ(yi) = 0,
S(yi) = −xyi = yix
for all i. It is easy to check the following
(a) GKdim H = 0,
(b) Ω = {(x, −1)} = Ω√ ,
(c) Z = Y√ + C0 and dim Y√ = ∞,
(d) Px,−1,∗ = P∗,−1,1 = Y√ .
Acknowledgments. The authors thank Ken Goodearl and Ken Brown for their
valuable comments and thank the referee for his/her careful reading, many com-
ments, corrections, suggestions and Lemma 3.12. A part of this research was done
when J.J. Zhang was visiting Fudan University in the Fall of 2009 and the Spring
of 2010. D.-G. Wang was supported by the National Natural Science Foundation
LOWER BOUNDS OF GROWTH OF HOPF ALGEBRAS
23
of China (No. 10671016 and 11171183) and the Shandong Provincial Natural Sci-
ence Foundation of China (No. ZR2011AM013). J.J. Zhang and G. Zhuang were
supported by the US National Science Foundation (NSF grant No. DMS 0855743).
References
[AA] N. Andruskiewitsch and I.E. Angiono, On Nichols algebras with generic braiding, pp. 47 --
64 in "Modules and comodules" (Porto, Portugal, 2006), edited by T. Brzezi´nski et al.,
Birkhauser, Basel, 2008.
[AS1] N. Andruskiewitsch and H.-J. Schneider, Pointed Hopf Algebras, in: Recent developments
in Hopf algebra Theory, MSRI Publications 43 (2002), 168, Cambridge Univ. Press.
, A characterization of quantum groups, J. reine angew. Math. 577 (2004), 81 -- 104.
[AS2]
[Be] G.M. Bergman, The diamond lemma for ring theory, Adv. in Math. 29 (2) (1978) 178 -- 218.
[Br1] K. A. Brown, Noetherian Hopf algebras. Turkish J. Math. 31 (2007), suppl., 7 -- 23.
[Br2]
, Representation theory of Noetherian Hopf algebras satisfying a polynomial identity,
in Trends in the Representation Theory of Finite Dimensional Algebras (Seattle 1997), (E.L.
Green and B. Huisgen-Zimmermann, eds.), Contemp. Math. 229 (1998), 49 -- 79.
[BG1] K. A. Brown and K. R. Goodearl, Homological aspects of Noetherian PI Hopf algebras and
irreducible modules of maximal dimension, J. Algebra 198 (1997), 240 -- 265.
[BG2]
[BZ] K.A. Brown and J.J. Zhang, Prime regular Hopf algebras of GK-dimension one, Proc. London
, Lectures on Algebraic Quantum Groups, Birkhauser, 2002.
Math. Soc. (3) 101 (2010) 260 -- 302.
[GZ1] K.R. Goodearl and J.J. Zhang, Homological properties of quantized coordinate rings of
semisimple groups, Proc. Lond. Math. Soc. (3) 94 (2007), no. 3, 647 -- 671.
[GZ2]
, Noetherian Hopf algebra domains of Gelfand-Kirillov dimension two, J. of Algebra,
324 (2010), Special Issue in Honor of Susan Montgomery, 3131-3168.
[Gr] M. Gromov, Groups of polynomial growth and expanding maps. Inst. Hautes ´Etudes Sci.
Publ. Math. No. 53 (1981), 53 -- 73.
[Kh] V.K. Kharchenko, A quantum analogue of the Poincar´e-Birkhoff-Witt theorem. (Russian.
Russian summary) Algebra Log. 38 (1999), no. 4, 476 -- 507, 509; translation in Algebra and
Logic 38 (1999), no. 4, 259 -- 276.
[KL] G.R. Krause and T.H. Lenagan, Growth of algebras and Gelfand-Kirillov dimension, Revised
edition. Graduate Studies in Mathematics, 22. AMS, Providence, RI, 2000.
[LWZ] D.-M. Lu, Q.-S. Wu and J.J. Zhang, Homological integral of Hopf algebras, Trans. Amer.
Math. Soc. 359 (2007), no. 10, 4945 -- 4975.
[MR] J. C. McConnell and J. C . Robson, Noncommutative Noetherian Rings, Wiley, Chichester,
1987.
[Mo] S. Montgomery, Hopf Algebras and their Actions on Rings, CBMS Regional Conference
Series in Mathematics, 82, Providence, RI, 1993.
[SSW] L. W. Small, J.T. Stafford and R. B. Warfield Jr., Affine algebras of Gelfand-Kirillov
dimension one are PI, Math. Proc. Cambridge Philos. Soc. 97 (1985), no. 3, 407 -- 414.
[WuZ1] Q.-S. Wu and J.J. Zhang, Noetherian PI Hopf algebras are Gorenstein, Trans. Amer.
Math. Soc. 355 (2003), no. 3, 1043 -- 1066.
[WuZ2]
, Regularity of involutary PI Hopf algebras, J. Algebra 256 (2002),no.2,599-610.
[Zhu] G. Zhuang, Existence of Hopf subalgebras of GK-dimension two, J. Pure Appl. Algebra 215
(2011), 2912-2922.
Wang: School of Mathematical Sciences, Qufu Normal University, Qufu, Shandong
273165, P.R.China
E-mail address: [email protected], [email protected]
Zhang: Department of Mathematics, Box 354350, University of Washington, Seattle,
Washington 98195, USA
E-mail address: [email protected]
Zhuang: Department of Mathematics, Box 354350, University of Washington, Seat-
tle, Washington 98195, USA
E-mail address: [email protected]
|
1209.3681 | 1 | 1209 | 2012-09-17T15:07:05 | Duality for groupoid (co)actions | [
"math.RA"
] | In this paper we present Cohen-Montgomery-type duality theorems for groupoid (co)actions. | math.RA | math |
DUALITY FOR GROUPOID (CO)ACTIONS
DAIANA FL ORES AND ANTONIO PAQUES
Abstract. In this paper we present Cohen-Montgomery-type duality theorems for groupoid
(co)actions.
Key words and phrases: groupoid action, groupoid coaction, Cohen-Montgomery duality.
Mathematics Subject Classification: Primary 16S40, 16W50, 20L05, 22D35.
1. Introduction
Groupoids are usually presented as small categories whose morphisms are invertible. This
notion is a natural extension of the notion of a group. Notice that a group can be seen as a
category with a unique objet.
The notion of a groupoid action that we use in this paper arose from the notion of a partial
groupoid action as introduced in [2], which is a natural extension of the notion of a partial group
action [10]. First, partial ordered groupoid actions on sets were introduced in the literature,
as ordered premorphisms, by N. Gilbert [11]. After, partial ordered groupoid actions on rings
were considered by D. Bagio and the authors [1] as a generalization of partial group actions, as
introduced by M. Dokuchaev and R. Exel in [10]. And in [2] this notion was extended to the
general context of groupoids.
Our purpose is to present a generalization of Cohen-Montgomery duality Theorem for group
actions [9, Theorem 3.2] (resp., group coactions or, equivalently, group grading [9, Theorem 3.5])
to the setting of groupoid actions (resp., groupoid coactions or groupoid grading) (see Theorems
3.7 and 4.5).
This paper is organized as follows. In the next section we give preliminaries about groupoids,
groupoid actions, skew groupoid rings, weak bialgebras, groupoid gradings, groupoid coactions and
weak smash products, these later as introduced by S. Caenepeel and E. De Groot in [7]. In that
section we will be concerned only with the results strictly necessary to construct the appropriate
environment to prove our main theorems, whose proofs we set in the sections 3 (actions) and 4
(coactions).
We deal with the groupoid ring KG and its dual KG∗, K being a commutative ring and G a
finite groupoid. The K-algebras KG and KG∗ are perhaps the first examples of weak bialgebras
that are not bialgebras.
Accordingly [2], an action of a groupoid G on a K-algebra R is a pair β = ({Eg}g∈G, {βg}g∈G),
where for each g ∈ G, Eg = Egg−1 is an ideal of R and βg : Eg−1 → Eg is an isomorphism of rings
satisfying some appropriate conditions (see the subsection 2.3).
If the set G0 of all identities of G is finite then there exists a one to one correspondence between
the structures of left KG-module algebra on R and the actions β of G on R such that each ideal
Ee. In particular, the notion of groupoid action introduced
by Caenepeel and De Groot in [8] is equivalent to this previous one (see Proposition 2.2).
Ee (e ∈ G0) is unital and R =Le∈G0
Given an action β of a finite groupoid G on a K-algebra R we can consider the skew groupoid
ring R ⋆β G [1], which is a G-graded algebra or, equivalently, a left KG∗-module algebra. The
corresponding weak smash product [7] (R⋆β G)#KG∗ is a nonunital K-algebra (see the subsection
2.5).
Given any unital K-algebra A graded by a finite groupoid G or, equivalently, a left KG∗-module
algebra, there exists an action β of G on the weak smash product A#KG∗ (see Proposition 2.5)
1
2
DAIANA FL ORES AND ANTONIO PAQUES
and we can consider the corresponding skew group ring (A#KG∗) ⋆β G, which also is a nonunital
K-algebra.
We show in the section 3 (resp., section 4) that (R ⋆β G)#KG∗ (resp., (A#KG∗) ⋆β G) contains
a unital K-subalgebra that is isomorphic to a finite direct sum of matrix K-algebras. In particular,
if G is a group we recover [9, Theorems 3.2 and 3.5].
In [15] D. Nikshych presented a Blattner-Montgomery-type duality for weak Hopf algebras,
generalizing the well known result for Hopf algebras obtained in [4] and [3]. There exists a natural
KG∗
as constructed by Nikshych (in the case that this makes sense), and it will be explicitly given in
the section 5.
relation between (R ⋆β G)#KG∗ and the double weak smash product (RNKGt
KG)NKG∗
t
Throughout, by ring (or algebra) we mean an associative, not necessarily commutative and not
necessarily unital ring (or algebra).
2. Preliminary results
2.1. Groupoids. The axiomatic version of groupoid that we adopt in this paper was taken from
[12]. A groupoid is a non-empty set G equipped with a partially defined binary operation, that
we will denote by concatenation, for which the usual axioms of a group hold whenever they make
sense, that is:
(i) For every g, h, l ∈ G, g(hl) exists if and only if (gh)l exists and in this case they are equal.
(ii) For every g, h, l ∈ G, g(hl) exists if and only if gh and hl exist.
(iii) For each g ∈ G there exist (unique) elements d(g), r(g) ∈ G such that gd(g) and r(g)g
exist and gd(g) = g = r(g)g.
(iv) For each g ∈ G there exists an element g−1 ∈ G such that d(g) = g−1g and r(g) = gg−1.
We will denote by G2 the subset of the pairs (g, h) ∈ G × G such that the element gh exists.
An element e ∈ G is called an identity of G if e = d(g) = r(g−1), for some g ∈ G. In this case
e is called the domain identity of g and the range identity of g−1. We will denote by G0 the set
of all identities of G and we will denote by Ge the set of all g ∈ G such that d(g) = r(g) = e.
Clearly, Ge is a group, called the isotropy (or principal) group associated to e.
The assertions listed in the following lemma are straightforward from the above definition. Such
assertions will be freely used along this paper.
Lemma 2.1. Let G be a groupoid. Then,
(i) for every g ∈ G, the element g−1 is unique satisfying g−1g = d(g) and gg−1 = r(g),
(ii) for every g ∈ G, d(g−1) = r(g) and r(g−1) = d(g),
(iii) for every g ∈ G, (g−1)−1 = g,
(iv) for every g, h ∈ G, (g, h) ∈ G2 if and only if d(g) = r(h),
(v) for every g, h ∈ G, (h−1, g−1) ∈ G2 if and only if (g, h) ∈ G2 and, in this case, (gh)−1 =
h−1g−1,
(vi) for every (g, h) ∈ G2, d(gh) = d(h) and r(gh) = r(g),
(vii) for every e ∈ G0, d(e) = r(e) = e and e−1 = e,
(viii) for every (g, h) ∈ G2, gh ∈ G0 if and only if g = h−1,
(ix) for every g, h ∈ G, there exists l ∈ G such that g = hl if and only if r(g) = r(h),
(x) for every g, h ∈ G, there exists l ∈ G such that g = lh if and only if d(g) = d(h).
2.2. Weak bialgebras: the finite groupoid algebra and its dual. Hereafter K will denote
a unital commutative ring and unadorned ⊗ will mean ⊗K. Following [6], a weak K-bialgebra H
is a unital K-algebra, with a K-coalgebra structure (∆, ǫ) such that
(i) ∆(xy) = ∆(x)∆(y)
DUALITY FOR GROUPOID (CO)ACTIONS
3
(ii) ∆2(1H ) = (∆(1H ) ⊗ 1H)(1H ⊗ ∆(1H )) = (1H ⊗ ∆(1H ))(∆(1H ) ⊗ 1H)
for all x, y, z ∈ H, where ∆2 = (∆ ⊗ IH ) ◦ ∆ = (IH ⊗ ∆) ◦ ∆ and IH denotes the identity map of
(iii) ε(xyz) =P ε(xy1)ε(y2z) =P ε(xy2)ε(y1z),
H. We use the Sweedler-Heyneman notation for the comultiplication, namely ∆(x) =P x1 ⊗ x2,
If H is a bialgebra, that is, the maps ∆ and ε are homomorphisms of algebras, then the above
axioms (i)-(iii) are trivially satisfied. Here we are concern with the algebra KG of a groupoid G
and its dual in the case that G is finite. Both are weak bialgebras but not bialgebras.
for all x ∈ H.
Given a groupoid G, the groupoid algebra KG is free as a K-module with basis {ug g ∈ G},
its multiplication is given by the rule
uguh =(ugh
0
if d(g) = r(h)
otherwise,
for all g, h ∈ G, its identity element exists and is 1KG =Pe∈G0
and its K-coalgebra structure is given by
∆(ug) = ug ⊗ ug
and ε(ug) = 1K,
ue if and only if G0 is finite [14],
for all g ∈ G.
The dual KG∗ of KG, as a K-module, is free with dual basis {vg g ∈ G}, that is, vg(uh) =
δg,h1K, for all g, h ∈ G. If G is finite, its K-algebra structure is given by
vgvh = δg,hvg
vg = 1KG∗,
and Xg∈G
and its K-coalgebra structure is given by
∆(vg) = Xhl=g
vh ⊗ vl = Xd(h)=d(g)
vgh−1 ⊗ vh
and ε(Xg∈G
agvg) = Xe∈G0
ae.
2.3. Groupoid actions. Let G be a groupoid and R a not necessarily unital ring. Following [2],
an action of G on R is a pair
β = ({Eg}g∈G, {βg}g∈G),
where for each g ∈ G, Eg = Er(g) is an ideal of R, βg : Eg−1 → Eg is an isomorphism of rings,
and the following conditions hold:
(i) βe is the identity map IEe of Ee, for all e ∈ G0,
(ii) βgβh(r) = βgh(r), for all (g, h) ∈ G2 and r ∈ Eh−1 = E(gh)−1 .
In particular, β induces an action of the group Ge on Ee, for every e ∈ G0.
In [8] Caenepeel and De Groot developed a Galois theory for weak bialgebra actions on algebras.
In particular, they considered the situation where the weak bialgebra is a finite groupoid algebra
and a notion of groupoid action was introduced. Actually, this later notion and the one above
defined, under some additional conditions, are equivalent, as we will see in the next proposition.
Following [8, section 4], a KG-module algebra is a unital K-algebra R, with a left KG-module
structure given by · : KG ⊗ R → R, ug ⊗ x 7→ ug · x, such that:
ug · (xy) = (ug · x)(ug · y) and ug · 1R = ur(g) · 1R,
for all x, y ∈ R and g ∈ G.
Proposition 2.2. Let G be a groupoid such that G0 is finite, and R be a unital K-algebra. Then
the following statements are equivalent:
(i) There exists an action β = ({Eg}g∈G, {βg}g∈G) of G on R such that every Ee, e ∈ G0, is
unital and R = Le∈G0
Ee.
(ii) R has an structure of KG-module algebra.
4
DAIANA FL ORES AND ANTONIO PAQUES
Proof. (i)⇒(ii) Let 1g denote the identity element of Eg, for every g ∈ G. It easily follows from the
assumptions that each 1g is a central idempotent of R, 1R =Pe∈G0
Consider now the action of KG over R given by
for all e 6= f .
1e and EeT Ef = 0 = EeEf ,
ug · r = βg(r1g−1 ),
for every g ∈ G and r ∈ R. Such an action induces on R an structure of KG-module. Indeed,
1KG · r = Xe∈G0
ue · r = Xe∈G0
βe(r1e) = Xe∈G0
r1e = r Xe∈G0
1e = r1R = r,
for all r ∈ R. Furthermore, it follows from the items (ii),(iv) and (vi) of Lemma 2.1 that E(gh)−1 =
Eh−1 , Eh = Eg−1 , Eg = Egh, for all (g, h) ∈ G2. Hence,
ug · (uh · r) = βg(βh(r1h−1 )1g−1) = βg(βh(r1h−1 ))
= βgh(r1h−1 ) = βgh(r1(gh)−1 )
= ugh · r = uguh · r,
for all r ∈ R and (g, h) ∈ G2. For (g, h) 6∈ G2 it is trivial to check that ug · (uh · r) = 0 = uguh · r.
So, R is a left KG-module.
Now, since
and
ug · (rs) = βg(rs1g−1 ) = βg(r1g−1 )βg(s1g−1) = (ug · r)(ug · s)
ug · 1R = βg(1R1g−1) = 1g = 1r(g) = βr(g)(1R1r(g)−1) = ur(g) · 1R,
for all g ∈ G and r, s ∈ R, the required follows.
4.1] these elements are central orthogonal idempotents in R, and R =Le∈G0
(ii)⇒(i) Put 1g = ug · 1R and Eg = R1g, for every g ∈ G. So, 1g = 1r(g) and by [8, Proposition
Ee. Clearly, each
Eg, g ∈ G, is an ideal of R and a unital ring. Let βg : Eg−1 → Eg given by βg(r) = ug · r, for every
r ∈ Eg−1 and g ∈ G. It is immediate from the assumptions that βg is a well defined isomorphism
of rings with β−1
g = βg−1 . Furthermore,
βe(r) = Xe′∈G0
βe′ (r1e′ ) = Xe′∈G0
ue′ · r = 1KG · r = r,
for every r ∈ Ee−1 = Ee, and
for every (g, h) ∈ G2 and r ∈ Eh−1 = E(gh)−1 . The proof is complete.
(cid:3)
βg(βh(r)) = ug · (uh · r) = ugh · (r) = βgh(r),
2.4. The skew groupoid ring. Let R, G and β = ({Eg}g∈G, {βg}g∈G) be as in the previous
subsection. Accordingly [1, Section 3], the skew groupoid ring R ⋆β G corresponding to β is defined
as the direct sum
in which the δg's are symbols, with the usual addition, and multiplication determined by the rule
Egδg
R ⋆β G =Mg∈G
(xδg)(yδh) =(xβg(y)δgh
0
if (g, h) ∈ G2
otherwise,
for all g, h ∈ G, x ∈ Eg and y ∈ Eh.
This multiplication is well defined. Indeed, if (g, h) ∈ G2 then d(g) = r(h) (see Lemma 2.1(iv)).
(⋆)
=
So, Eg−1 = Er(g−1) = Ed(g) = Er(h) = Eh, βg(y) makes sense, and xβg(y) ∈ Eg = Er(g)
Er(gh) = Egh, where the equality (⋆) is ensured by Lemma 2.1(vi).
By a routine calculation one easily sees that A = R ⋆β G is associative, and by [1, Proposition
3.3] it is unital if G0 is finite and Ee is unital for all e ∈ G0. In this case the identity element of
1eδe, where 1e denotes the identity element of Ee, for all e ∈ G0.
A is 1A =Pe∈G0
DUALITY FOR GROUPOID (CO)ACTIONS
5
Remark 2.3. Proposition 3.3 in [1] asserts that A = R ⋆β G is unital if and only if G0 finite.
Unfortunately, the existence of the identity element 1A does not necessarily imply that G0 is
finite, as we will see in the next subsection (note that, in particular, A is a G-graded algebra by
construction).
2.5. Groupoid gradings, coations and weak smash products. Let G be a groupoid and A
a not necessarily unital K-algebra. We say that A is a G-graded algebra if there exists a family
{Ag}g∈G of K-submodules of A such that
and
Ag
A =Mg∈G
AgAh(⊆ Agh
0
if (g, h) ∈ G2
otherwise,
for all g, h ∈ G. It easily follows from this definition that each Ae, e ∈ G0, is a subalgebra of A.
Remark 2.4. If A is assumed to be unital, then it follows from [13, Propositions 2.2 and 2.3, and
Remark 2.4] that:
(i) the set J0 = {e ∈ G0 Ae 6= 0} is finite,
(ii) Ag = 0 for all g ∈ G such that either d(g) or r(g) does not belong to J0,
(iii) every Ae, with e ∈ J0, is unital,
1e, with 1e denoting the identity element of Ae, for all e ∈ J0.
(v) for every e ∈ J0 and g ∈ G such that r(g) = e (resp., d(g) = e), 1eag = ag (resp.,
(iv) 1A =Pe∈J0
ag1e = ag), for all ag ∈ Ag.
Recall from [5] and [7] that a right H-comodule algebra A (H being a weak bialgebra) is a
unital K-algebra with a right H-comodule structure given by the coaction ρ : A → A ⊗ H such
that ρ(ab) = ρ(a)ρ(b), for all a, b ∈ A, and (ρ ⊗ IH ) ◦ ρ(1A) = (ρ(1A) ⊗ 1H)(1A ⊗ ∆(1H ))
In all what follows we will assume that A is a unital K-algebra and G is finite.
The existence of a G-grading on A is equivalent to say that A has an structure of a right KG-
comodule algebra (see [8, Proposition 3.1]), with coation given by ρ(a) =Pg∈G ag ⊗ ug, for all
a =Pg∈G ag ∈ A. This is also equivalent to say that A is a KG∗-module algebra, with the action
given by vh · a = ah, for all a ∈ A and h ∈ G.
Hence, we can consider the weak smash product A#KG∗ of A by KG∗ (see [7, section 3]),
which is equal to A ⊗ KG∗ as K-modules and the multiplication is given by the following rule:
(a#vg)(b#vh) =(a(vgh−1 · b)#vh
0
if d(h) = d(g)
otherwise.
A routine calculation easily shows that such a multiplication is associative. Also, A#KG∗ is not
unital. To see this it is enough to verify that any element of the type x = b#vh, with b ∈ Ak and
d(k) 6= r(h), is a right annhilator of A#KG∗. Indeed, in this case gh−1 6= k by Lemma 2.1(x) and
therefore vgh−1 · b = 0 which implies (a#vg)x = 0 for all a ∈ A and g ∈ G.
The element u = 1A#1KG∗ is a preunit of A#KG∗, that is, ux = xu = xu2, for all x ∈ A#KG∗
(see [7, section 3]).
In the sequel we will show that there also exists an action β of the groupoid G on A#KG∗,
which allows us to construct the skew groupoid ring (A#KG∗) ⋆β G.
Put B = A#KG∗ and, for each g ∈ G, let
Eg = Ml,k∈G
d(k)=r(g)
Al#vk
and
βg : Eg−1 → Eg
given by βg(al#vk) = al#vkg−1 .
6
DAIANA FL ORES AND ANTONIO PAQUES
Proposition 2.5. The pair β = ({Eg}g∈G, {βg}g∈G) is an action of G on B, and B = Le∈G0
Ee.
Proof. First, it is clear that Eg = Er(g), for all g ∈ G. Now, taking x = al#vk ∈ Eg and
y = bs#vt ∈ B, we have by definition
xy = (al#vk)(bs#vt) =(al(vkt−1 · bs)#vt
0
if d(k) = d(t)
otherwise,
and vkt−1 · bs 6= 0 if and only if kt−1 = s or, equivalently, t = s−1k. Thus,
xy =(albs#vt
0
if t = s−1k
otherwise,
Since d(t) = d(s−1k) = d(k) = r(g), it follows that xy ∈ Eg. Similarly, we also have yx ∈ Eg.
Hence, Eg is an ideal of B.
It is immediate to check that βg : Eg−1 → Eg is a well defined additive map, βg−1 = β−1
g
for
all g ∈ G, and βe = IEe for all e ∈ G0. From the above we also have
βg(xy) =(albs#vtg−1
0
if t = s−1k
otherwise,
for all x = al#vk and y = bs#vt in Eg−1 . On the other hand, since d(kg−1) = d(g−1) = d(tg−1)
(see Lemma 2.1(vi)), and d(k) = r(g−1) = d(g) we have
βg(x)βg(y) = al(vkg−1(tg−1)−1 · bs)#vtg−1
= al(vkg−1 gt−1 · bs)#vtg−1
= al(v(k(d(g))t−1 · bs)#vtg−1
= al(vkt−1 · bs)#vtg−1
= albs#vtg−1
if t = s−1k and 0 otherwise. So, β is multiplicative.
Finally, notice that, for all (g, h) ∈ G2, dom(βgβh) = βh−1 (EhT Eg−1 ) = βh−1 (Eh) = Eh−1 =
Ed(h) = Ed(gh) = E(gh)−1 = dom(βgh), and
βgβh(x) = βg(βh(al#vk)) = βg(al#vkh−1 )
= al#vk(gh)−1 = βgh(al#vk) = βgh(x),
for all x = al#vk ∈ Eh−1. Hence, β is an action of G on B.
The last assertion is immediate, since B =Pg∈G Er(g) and EeT Pe′ ∈G0
e′6=e
Ee′ = 0.
(cid:3)
The skew groupoid ring B ⋆β G is clearly associative (see the previous subsection). However, it
is not unital, because any element of the type x = (aj#vk)δs, with d(k) = r(s) and r(k) 6= d(j),
is a left annhilator of B ⋆β G.
Indeed, it enough to verify that xEg = 0, for all g ∈ G. Let
y = al#vh ∈ Eg, that is, d(h) = r(g). Clearly xy = 0 if d(s) 6= r(g), and, otherwise, we have
xy = (aj #vk)βs(al#vh)δsg
= (aj #vk)(al#vhs−1 )δsg
= (aj (vksh−1 · al)#vhs−1 )δsg
But, vksh−1 ·al 6= 0 if and only if ksh−1 = l, which implies r(l) = r(k) and aj(vksh−1 ·al) = ajal = 0
because d(j) 6= r(l).
DUALITY FOR GROUPOID (CO)ACTIONS
7
3. Duality for groupoid actions
In this section R will denote a not necessarily unital K-algebra, G a finite groupoid and β =
({Eg}g∈G, {βg}g∈G) an action of G on R such that Ee is unital, for all e ∈ G0.
We are concerned with a Cohen-Montgomery-type duality theorem [9] for groupoid actions.
More precisely, we show that the weak smash product (R ⋆β G)#KG∗ (which is not unital)
contains a unital subalgebra isomorphic to a finite direct sum of matrix K-algebras. In particular,
if G is a group we recover Theorem 3.2 of [9].
Let B = (R ⋆β G)#KG∗. The subalgebra that we are looking for is
B0 = Md(g)=r(g)=r(h)
Egδg#vh,
and, in order to get our main result, the strategy is to obtain a decomposition of B0 into a direct
sum of suitable unital subalgebras satisfying the conditions of the following lemma due to D. S.
Passman (see [16, Lemma 1.6, p. 228]).
Lemma 3.1. Let S be a unital ring, and 1S = e1 + · · · + en be a decomposition of 1S into a sum
of orthogonal idempotents. Let U be a subgroup of the group of the units of S, and assume that U
permutes the set {e1, . . . , en} transitively by conjugation. Then S ≃ Mn(e1Se1).
The next lemmas are the necessary preparation to get our purpose. For each e ∈ G0, put
Se = {g ∈ G d(g) = e} and Te = {g ∈ G r(g) = e}. Notice that Ge = SeT Te, for all e ∈ G0.
Lemma 3.2. The following statements hold:
(i) B0 is a unital subalgebra of B, with identity element w = Pe∈G0
(ii) W = {we := 1eδe# Pg∈Te
(iii) B0 =Le∈G0
Be, where Be = B0we = Pg∈Ge
h∈Te
B0 with identity element we.
1eδe#vg.
g∈Te
vg e ∈ G0} is a set of central orthogonal idempotents of B0.
Egδg#vh is a unital ideal (so, a subalgebra) of
(iv) For each e ∈ G0, We = {we,g := 1eδe#vg g ∈ Te} is a set of noncentral orthogonal
idempotents of Be whose sum is we.
Proof. (i) It is clear that B0 is a K-submodule of B. Given x = agδg#vh and y = blδl#vk in B0
we have
and, consequently,
xy =((agδg)(vhk−1 · blδl)#vk
0
if d(k) = d(h)
otherwise,
xy =((agδg)(blδl)#vk
0
if k = l−1h
otherwise,
Since d(g) = r(h) = d(l−1) = r(l), it follows that
(agδg)(blδl) = agβg(bl)δgl ∈ Eglδgl.
So, xy ∈ B0 because r(gl) = r(g) = r(h) = r(l) = r(k) and d(gl) = d(l) = r(k).
8
Also,
and
DAIANA FL ORES AND ANTONIO PAQUES
xw = Pe∈G0
(agδg#vh)(1eδe# Pl∈Te
vl)
agδg(vhl−1 · 1eδe)#vl.
= Pe∈G0 Pl∈Te
d(l)=d(h)
= (agδg)(1r(h)δr(h))#vh
= (agδg)(1d(g)δd(g))#vh
= agβg(1d(g))δgd(g)#vh
= agδg#vh
= x,
wx = Pe∈G0 Pl∈Te
= Pe∈G0 Pl∈Te
d(l)=d(h)
(1eδe#vl)(agδg#vh)
1eδe(vlh−1 · agδg)#vh
= (1r(g)δr(g))(agδg)#vh
= βr(g)(ag)δr(g)g#vh
= agδg#vh
= x.
vh. Then,
vg and wf = 1f δf # Ph∈Tf
1eδe(vgk−1 · 1f δf )#vk(Xh∈Tf
vh),
1eδe(vgh−1 · 1f δf )#vh
and so
wewf = Xg∈Te Xd(k)=d(g)
Pg∈Te Ph∈Tf
(ii) Let e, f ∈ G0, we = 1eδe# Pg∈Te
wewf =
(1eδe)(1eδe)#vg = 1eδe# Pg∈Te
d(h)=d(g)
0
Pg∈Te
vg = we. Therefore,
wewf =(we
0
if e = f
otherwise.
Noticing that vgh−1 .1f δf 6= 0 if and only if h = g and e = r(g) = f , we have wewe =
if d(h) 6= d(g), ∀g ∈ Te, h ∈ Tf .
It remains to show that we is central in B0. Take x = agδg#vh ∈ B0. Then,
Observe that if r(t) 6= e, for all t ∈ G such that d(t) = d(h), then vt(Pl∈Te
agδg(vhl−1 .1eδe)#vl
vl) = Xd(t)=d(h)
agδg(vht−1 · 1eδe)#vt(Xl∈Te
vl).
vl) = 0. Hence,
xwe = (agδg#vh)(1eδe#Xl∈Te
xwe =
Pl∈Te
d(l)=d(h)
0
Since vhl−1 · 1eδe 6= 0 if and only if l = h, and e = r(h) = d(g), it follows that
agδg(vhl−1 · 1eδe)#vl = (agδg)(1eδe)#vh = agβg(1e)δge#vh = agδg#vh = x.
Xl∈Te
d(l)=d(h)
if d(h) 6= d(l), ∀l ∈ Te.
Hence,
xwe =(x if h ∈ Te
otherwise.
0
DUALITY FOR GROUPOID (CO)ACTIONS
9
By a similar calculation one obtains wex = xwe.
(iii) It easily follows from (ii).
(iv) Take we,g = 1eδe#vg and we,h = 1eδe#vh in We. Then,
we,gwe,h =(1eδe(vgh−1 · 1eδe)#vh
0
if d(h) = d(g)
otherwise.
Since vgh−1 · 1eδe 6= 0 if and only if h = g, and r(g) = e by assumption, it follows that
we,gwe,h =(we,g
0
if h = g
otherwise.
To see that each we,l = 1eδe#vl is not central in Be, take x = agδg#vh ∈ Be. Then, r(g) =
d(g) = r(h) = e and
xwe,l = (agδg#vh)(1eδe#vl) =(agδg(vhl−1 · 1eδe)#vl
0
that easily implies
xwe,l =(x if l = h
0
otherwise.
On the other hand,
we,lx = (1eδe#vl)(agδg#vh) =(1eδe(vlh−1 · agδg)#vh
0
if d(l) = d(h)
otherwise,
if d(l) = d(h)
otherwise
and, consequently,
we,lx =(x
0
if l = gh
otherwise.
The proof is complete.
(cid:3)
vh g ∈ Ge} is a subgroup of the group of the units of Be,
Lemma 3.3. For each e ∈ G0,
(i) Ue = {ug,e := 1gδg# Ph∈Te
(ii) Ue acts on We by conjugation,
(iii) wUe
the action of Ue on We.
Proof. (i) Let ug,e = 1gδg# Ph∈Te
e,h = {we,gh g ∈ Ge} = {we,l l ∈ Sd(h)} is the orbit of the element we,h ∈ We, under
vk be elements of Ue. Then,
ug,eul,e = Xh∈Te
(1gδg#vh)(1lδl#Xk∈Te
vk)
1gδg(vhk−1 · 1lδl)#vk
vh and ul,e = 1lδl# Pk∈Te
= Xh∈Te Xk∈Te
= Xh∈Te
= 1glδgl# Xh∈Te
1gδg1lδl#vl−1 h
vl−1h
d(k)=d(h)
The last equality follows from the fact that 1g = 1r(g) = 1e = 1r(l) = 1l and 1gl = 1r(gl) = 1r(g) =
1g. Since r(l−1h) = r(l−1) = d(l) = e and gl ∈ Ge, it follows that ug,eul,e ∈ Ue.
Finally, taking ug−1,e = 1g−1δg−1 # Pl∈Te
vl ∈ Ue we have
10
DAIANA FL ORES AND ANTONIO PAQUES
vh)(1g−1 δg−1 #Xl∈Te
vl)
1gδg(vhl−1 · 1g−1δg−1 )#vl
d(l)=d(h)
ug,eug−1,e = (1gδg# Xh∈Te
= Xh∈Te Xl∈Te
= Xh∈Te
= Xh∈Te
= 1eδe# Xh∈Te
= 1Be.
vh
1r(g)δr(g)#vh
1gδg1g−1 δg−1 #vh
(since
1r(h) = 1e = 1d(g) = 1g−1 )
By a similar calculation one also gets ug−1,eug,e = 1Be.
(ii) Taking we,l = 1eδe#vl ∈ We and ug,e = 1gδg# Ph∈Te
vh)(1eδe#vl)
vh ∈ Ue we have
ug,ewe,l = (1gδg# Ph∈Te
= Ph∈Te Pd(t)=d(h)
= Ph∈Te
d(h)=d(l)
= 1gδg1eδe#vl
= 1gδg#vl.
1gδg(vht−1 · 1eδe)#vtvl
1gδg(vhl−1 .1eδe)#vl
Hence,
ug,ewe,lug−1,e = (1gδg#vl)(1g−1 δg−1 #Xk∈Te
vk)
1gδg(vlt−1 · 1g−1 δg−1 )#vt(Xk∈Te
vk)
1gδg(vlk−1 · 1g−1δg−1 )#vk
= Xd(t)=d(l)
= Xk∈Te
d(k)=d(l)
= 1gδg1g−1δg−1 #vgl
= 1r(g)δr(g)#vgl
= 1eδe#vgl,
which belongs to We because r(gl) = r(g) = e.
(iii) It easily follows from the proof of (ii).
(cid:3)
It follows from the above that any two elements of We, say we,g and we,h, are in the same
orbit by the action of Ue if and only if d(g) = d(h). Hence, the action of Ue on We in general
is not transitive. Nevertheless, We contains subsets Ωe,h1 , . . . , Ωe,hne and Ue contains subgroups
Ue,h1, . . . , Ue,hne such that each Ue,hi acts transitively on Ωe,hi by conjugation, as we shall see in
the sequel.
Lemma 3.4. For each e ∈ G0, let wUe
e,h1
denote the sum of all elements of the orbit wUe
e,hi
, . . . , wUe
be the distinct orbits of Ue in We, and ωe,hi
e,hne
, for each 1 ≤ i ≤ ne. Then,
(i) d(hi) 6= d(hj), for all i 6= j,
(ii) ωe,hi = 1eδe# Pl∈Te∩Sd(hi)
vl and we = P1≤i≤ne
ωe,hi,
DUALITY FOR GROUPOID (CO)ACTIONS
11
(iii) the elements ωe,hi are central orthogonal idempotents of Be,
(iv) for each 1 ≤ i ≤ ne, Be,hi := Beωe,hi = Lg∈Ge
(v) Be = L1≤i≤ne
subalgebra) of Be with identity element ωe,hi,
Be,hi.
l∈Te T Sd(hi )
Egδg#vl is a unital ideal (hence, a
Proof. (i) It follows from Lemma 3.3(iii).
(ii) Immediate.
(iii) Notice that each ωe,hi is a sum of orthogonal idempotents of We, which are all orthogonal.
So, it is immediate to check that all the ωe,hi, 1 ≤ i ≤ ne, are orthogonal.
It remains to show that such idempotents are central in Be.
Take x = alδl#vk ∈ Be. Then, l ∈ Ge, k ∈ Te and
ωe,hix = (1eδe# Xt∈Te T Sd(hi)
vt)(alδl#vk)
(1eδe#vt)(alδl#vk)
= Xt∈Te T Sd(hi)
= Xt∈Te T Sd(hi) Xd(s)=d(t)
=
Pt∈Te T Sd(hi)
=(alδl#vk
0
0
if d(k) = d(hi)
otherwise,
1eδe(vts−1 · alδl)#vsvk
1eδe(vtk−1 · alδl)#vk
if d(k) = d(hi)
otherwise
since vtk−1 · alδl 6= 0 if and only if t = lk and, in this case, d(t) = d(lk) = d(k) and r(t) = r(lk) =
r(l) = e.
By a similar calculation one obtains xωe,hi = ωe,hix.
(iv) It follows from (iii).
(v) It follows from (ii) and (iii).
(cid:3)
Lemma 3.5. For each e ∈ G0 and 1 ≤ i ≤ ne,
(i) Ωe,hi = {ωe,hi,l := 1eδe#vl l ∈ Te ∩ Sd(hi)} is a set of noncentral orthogonal idempotents
of Be,hi whose sum is 1Be,hi
= ωe,hi,
(ii) the set Ue,hi = {ug,e,hi := 1gδg# Pl∈Te T Sd(hi)
(iii) Ue,hi acts transitively on Ωe,hi by conjugation.
vl g ∈ Ge} is a subgroup of Ue,
Proof. (i) It follows from Lemma 3.2(iv) and Lemma 3.4(ii)(iv).
12
DAIANA FL ORES AND ANTONIO PAQUES
(ii) Take ul,e,hi = 1lδl#
Pk∈Te T Sd(hi)
vk and ut,e,hi = 1tδt# Ps∈Te T Sd(hi )
vk)(1tδt# Xs∈Te T Sd(hi )
ul,e,hiut,e,hi = (1lδl# Xk∈Te T Sd(hi )
vs)
vs in Ue,hi. Then,
(1lδl#vk)(1tδt#vs)
(1lδl)(vks−1 · 1tδt)#vs
= Xk,s∈Te T Sd(hi )
= Xk,s∈Te T Sd(hi )
= Xs∈Te T Sd(hi )
= 1ltδlt# Xs∈Te T Sd(hi )
vs,
(1lδl)(1tδt)#vs
because l ∈ Ge, t ∈ Te and 1l = 1r(l) = 1r(lt) = 1lt, which implies (1lδl)(1tδt) = βl(1t)δlt =
βl(1r(t))δlt = βl(1d(l))δlt = 1lδlt = 1ltδlt. And such a product belongs to Ue,hi because r(lt) =
r(l) = e = d(t) = d(lt).
Finally, it is straightforward to check that ul−1,e,hi = 1l−1δl−1 # Pt∈Te T Sd(hi )
ul,e,hi, and clearly it also belongs to Ue,hi.
vt is the inverse of
(iii) Take ug,e,hi = 1gδg#
vl ∈ Ue,hi and ωe,hi,k = 1eδe#vk ∈ We,hi . By a calcula-
tion identical to that done in the proof of the item (ii) of Lemma 3.3 one gets
Pl∈Te∩S(d(hi))
ug,e,hiωe,hi,kug−1,e,hi = 1eδe#vgk,
and this action is clearly transitive.
(cid:3)
Proposition 3.6. Be,hi ≃ Mne,hi
(Ee) as unital K-algebras, for all e ∈ G0 and 1 ≤ i ≤ ne.
Proof. It follows from Lemmas 3.1 and 3.5 that Be,hi ≃ Mne,hi
(Ce,hi ), where
Ce,hi = (ωe,hi,l1)Be,hi (ωe,hi,l1 ),
with l1 ∈ TeT Sd(hi). It remains to prove that Ce,hi is isomorphic to Ee. Recalling from Lemma
3.4 that any element of Be,hi is of the form x = Pg∈Ge
agδg#vk, we have
k∈Te∩Sd(hi )
agδg(vkl−1
1
· 1eδe)#vl1
k∈Te∩Sd(hi )
xωe,hi,l1 = Xg∈Ge
= Xg∈Ge
= Xg∈Ge
agδg#vl1 .
(agδg)(1eδe)#vl1
DUALITY FOR GROUPOID (CO)ACTIONS
13
and
ωe,hi,l1xωe,hi,l1 = (1eδe#vl1 )(Xg∈Ge
agδg#vl1 )
= Xg∈Ge
= Xg∈Ge
(1eδe)(vl1l−1
1
· 1gδg)#vl1
(1eδe)(ve.agδg)#vl1
= (1eδe)(aeδe)#vl1
= aeδe#vl1 .
Hence, Ce,hi = Eeδe#vl1 = Ee(1A#vl1 ), which is naturally isomorphic to Ee as K-algebras via
the map ae 7→ aeδe#vl1 = ae(1A#vl1 ). Observe that this map is surjective by definition and it is
straightforward to check that it is a homomorphism of K-algebras. Its injectivity follows from the
freeness of 1A#vl1 over A = R ⋆β G.
(cid:3)
Theorem 3.7.
as unital K-algebras.
B0 ≃ Me∈G0
(
neMi=1
Mne,hi
(Ee))
Proof. It follows from Lemmas 3.2(iii) and 3.4(v), and Proposition 3.6.
(cid:3)
4. Duality for groupoid coactions
In all this section G is a finite groupoid and A is a unital G-graded K-algebra. Recall from the
subsection 2.5 that A is a left KG∗-module algebra via the action vk ·Pg∈G ag = ak, for all k ∈ G,
and there exists an action β of G on the corresponding weak smash product B = A#KG∗. Let
C = B ⋆β G be the corresponding skew groupoid ring. Like in the section 3, also here we obtain
a Cohen-Montgomery-type duality theorem, that is, we show that the K-algebra C contains a
unital subalgebra isomorphic to a finite direct sum of matrix K-algebras. In particular, if G is a
group we recover [9, Theorem 3.5].
The steps to get our purpose are similar to those in the previous section. Recall from Proposition
2.5 that
Let
Al#vk)δg.
C =Mg∈G
( Md(k)=r(g)
C0 = Me∈G0
(Mg∈Ge
( Mr(l)=d(l)=r(k)
d(k)=e
Al#vk)δg).
Lemma 4.1. The following statements hold:
(i) C0 is a unital K-algebra with identity element w = Pe∈G0
(ii) W = {we := ( Pf ∈G0
(iii) C0 = Le∈G0
Ce, where Ce = C0we = Lg∈Ge
1f # Ph∈Tf T Se
( Lr(l)=d(l)=r(k)
d(k)=e
of C0, whose sum is w,
gebra) of C0 with identity element we.
( Pf ∈G0
1f # Ph∈Tf T Se
vh)δe,
vh)δe e ∈ G0} is a set of central orthogonal idempotents
Al#vk)δg is a unital ideal (so, subal-
Proof. (i) Clearly, C0 is a K-submodule of C. Now, take x = (al#vk)δg and y = (bs#vt)δh in
C0. Then, r(l) = d(l) = r(k), d(k) = e, r(s) = d(s) = r(t), d(t) = f , g ∈ Ge and h ∈ Gf , with
14
DAIANA FL ORES AND ANTONIO PAQUES
e, f ∈ G0. If e 6= f it is immediate that xy = 0. Otherwise, we have
xy = (al#vk)βg(bs#vt)δgh
= (al#vk)(bs#vgt−1 )δgh
= (al(vk(tg−1).bs)#vtg−1 )δgh
= (albs#vtg−1 )δgh,
= (albs#vs−1 k)δgh
if s = kgt−1 or, equivalently, tg−1 = s−1k. Observing that
- albs ∈ Als, because d(l) = r(k) = r(s),
- d(ls) = d(s), r(ls) = r(l) = r(k) = r(s) = d(s) and r(s−1k) = r(s−1) = d(s),
- d(s−1k) = d(k) = e = r(g) = r(gh), and
- gh ∈ Ge
we conclude that xy ∈ C0.
identity element of C0. Indeed, taking x = (al#vk)δg, with g ∈ Ge′ , for some e′ ∈ G0, r(l) =
d(l) = r(k) and d(k) = e′, we have
( Pf ∈G0
1f # Ph∈Tf T Se
vh)δe is the
vh)δe(al#vk)δg
It remains to prove that w = Pe∈G0
wx = Xe∈G0
= (Xf ∈G0
= (Xf ∈G0
= (Xf ∈G0 Xh∈Tf T Se′
= (Xf ∈G0 Xh∈Tf T Se′
1f # Xh∈Tf T Se
(Xf ∈G0
1f # Xh∈Tf T Se′
1f # Xh∈Tf T Se′
(1f #vh)(al#vk))δg
1f (vhk−1 · al)#vk)δg
vh)δe′ (al#vk)δg
vh)βe′ (al#vk)δg
= (1r(l)al#vk)δg.
= (al#vk)δg
= x
(by Remark 2.4(v))
One easily verifies that xw = x by the same way.
(ii) Let e, e′ ∈ G0. It is clear that wewe′ = 0 if e 6= e′. Otherwise we have
vh)δe][( Pf ′∈G0
vh)( Pf ′∈G0
1f ′# Pl∈Tf ′ T Se′
1f ′# Pl∈Tf ′ T Se
vl)δe
(1f #vh)(1f ′ #vl)δe
vl)δe]
1f (vhl−1 · 1f ′)#vlδe
wewe′ = [( Pf ∈G0
1f # Ph∈Tf T Se
1f # Ph∈Tf T Se
= ( Pf ∈G0
= Pf,f ′∈G0 Ph∈Tf T Se
= Pf,f ′ ∈G0 Ph∈Tf T Se
1f # Ph∈Tf T Se
= Pf ∈G0
f ′ T Se
f ′ T Se
= we
l∈T
l∈T
vhδe
because vhl−1 · 1f ′ 6= 0 if and only if h = l, and, in this case, f = f ′.
It remains to prove that we is central in C0, for all e ∈ G0. Let x = (al#vk)δg ∈ C0, that is,
g ∈ Ge′ , for some e′ ∈ G0, r(l) = d(l) = r(k) and d(k) = e′. Again, it is clear that xwe = 0 if
DUALITY FOR GROUPOID (CO)ACTIONS
15
e 6= e′. Otherwise,
xwe = Pf ∈G0 Ph∈Tf T Se
= Pf ∈G0 Ph∈Tf T Se
= Pf ∈G0 Ph∈Tf T Se
= (al1r(k)#vk)δg
= x
(al#vk)βg(1f #vh)δge
(al#vk)(1f #vhg−1 )δge
(al(vkgh−1 · 1f )#vhg−1 )δg
by Remark 2.4(v). Similarly, one also gets wex = x.
(iii) It is immediate from the above.
(cid:3)
vh)δe e ∈ G0} is a set of central orthogo-
Lemma 4.2. The following statements hold:
nal idempotents of Ce, whose sum is we,
(i) For any f ∈ G0, Wf = {we,f := (1f # Ph∈Tf T Se
(ii) For any e ∈ G0, Ce = Lf ∈G0
unital ideal (so, subalgebra) of Ce, with identity element we,f .
Proof. (i) For any e, f, f ′ ∈ G0 we have
Ce,f where Ce,f = Cewe,f = Lg∈Ge
( Ll∈Gf
k∈Tf T Se
Al#vk)δg is a
vh)δe(1f ′ # Xl∈Tf ′ ∩Se
vh)βe(1f ′ # Xl∈Tf ′ T Se
vl)δe
vl)δe
(1f #vh)(1f ′ #vle)δe
(1f #vh)(1f ′ #vl)δe
(1f (vhl−1 · 1f ′)#vl)δe
vh)δe = we,f
if f ′ = f
otherwise.
we,f we,f ′ = (1f # Xh∈Tf T Se
= (1f # Xh∈Tf T Se
= Xh∈Tf T Se Xl∈Tf ′ T Se
= Xh∈Tf T Se Xl∈Tf ′ T Se
= Xh∈Tf T Se Xl∈Tf ′ T Se
=
(1f # Ph∈Tf T Se
xwe,f = (al#vk)βg(1f # Xh∈Tf T Se
= (al#vk)(1f # Xh∈Tf T Se
= Xh∈Tf T Se
= Xh∈Tf T Se
=((al#vk)δg = x
0
0
(al#vk)(1f #vhg−1 )δg
(al(vkgh−1 · 1f )#vhg−1 )δg
if f = r(k)
otherwise,
vh)δg
vhg−1 )δg
To show that each we,f is central in Ce take x = (al#vk)δg ∈ Ce. So, g ∈ Ge, r(l) = d(l) = r(k),
d(k) = e, and
16
DAIANA FL ORES AND ANTONIO PAQUES
since vkgh−1 .1f 6= 0 if and only if h = kg if and only if f = r(h) = r(k) (Lemma 2.1(ix)). One also
gets we,f x = xwe,f in a similar way.
(ii) It is clear from the above.
(cid:3)
Lemma 4.3. For each e, f ∈ G0,
with sum we,f = 1Ce,f ,
(i) We,f = {we,f,h := (1f #vh)δe h ∈ TfT Se} is a set of noncentral orthogonal idempotents,
(ii) Ue,f = {ue,f,g := (1f # Pl∈Tf T Se
vl)δg g ∈ Ge} is a subgroup of the group of the units of
Ce,f ,
(iii) Ue,f acts transitively on We,f by conjugation.
Proof. (i) Let e, f ∈ G0 and h, l ∈ TfT Se. Then,
we,f,hwe,f,l = (1f #vh)δe(1f #vl)δe
= (1f #vh)βe(1f #vl)δe
= (1f #vh)(1f #vl)δe
= (1f (vhl−1 · 1f )#vl)δe
=((1f #vh)δe = we,f,h
0
if h = l
otherwise.
Clearly, Ph∈Te T Se
(al#vk)δg ∈ Ce,f . So, g ∈ Ge, l ∈ Gf , k ∈ Tf ∩ Se, and
we,f,h = we,f = 1Ce,f . Also, we,f,h is not central in Ce,f . Indeed, let x =
we,f,hx = (1f #vh)δe(al#vk)δg
= (1f #vh)βe(al#vk)δeg
= (1f #vh)(al#vk)δg
= (1f (vhk−1 · al)#vk)δg
=(x
0
if h = lk
otherwise.
On the other hand,
xwe,f,h = (al#vk)δg(1f #vh)δe
= (al#vk)βg(1f #vh)δge
= (al#vk)(1f #vhg−1 )δg
= (al(vkgh−1 · 1f )#vhg−1 )δg
=(x
0
if h = kg
otherwise,
since kgg−1k−1 = kr(g)k−1 = kek−1 = kk−1 = r(k) = f and al1f = al by Remark 2.4(v).
DUALITY FOR GROUPOID (CO)ACTIONS
17
(ii) Let e, f ∈ G0 and g, h ∈ Ge. Then,
ue,f,gue,f,h = (1f # Xl∈Tf T Se
= (1f # Xl∈Tf T Se
= (1f # Xl∈Tf T Se
= (1f # Xl∈Tf T Se
= Xl,t∈Tf T Se
= Xl∈Tf T Se
vk)δh
vk)δgh
vkg−1 )δgh
vl)δg(1f # Xk∈Tf T Se
vl)βg(1f # Xk∈Tf T Se
vl)(1f # Xk∈Tf T Se
vl)(1f # Xt∈Tf T Se
(1f (vlt−1 · 1f )#vt)δgh
(1f #vl)δgh,
vt)δgh
(setting t = kg−1)
and this last term belongs to Ue,f because r(gh) = r(g) = e = d(h) = d(gh) (Lemma 2.1(vi)).
The equality ue,f,g−1 = u−1
e,f,g is straightforward.
vk)δg−1
vl)δg(1f #vh)δe(1f # Xk∈Tf T Se
vl)βg(1f #vh)δg(1f # Xk∈Tf T Se
(1f #vl)(1f #vhg−1 )δg(1f # Xk∈Tf T Se
(1f (vl(hg−1)−1 · 1f )#vhg−1 )δg(1f # Xk∈Tf T Se
vk)δg−1
vk)δg−1
vk)δg−1
(iii) Let e, f ∈ G0, g ∈ Ge and h ∈ TfT Se. Then,
ue,f,gwe,f,hue,f,g−1 = (1f # Xl∈Tf T Se
= (1f # Xl∈Tf T Se
= Xl∈Tf T Se
= Xl∈Tf T Se
= (1f #vhg−1 )δg(1f # Xk∈Tf T Se
= (1f #vhg−1 )βg(1f # Xk∈Tf T Se
= (1f #vhg−1 )(1f # Xk∈Tf T Se
= Xk∈Tf T Se
= Xk∈Tf T Se
= (1f #vhg−1 )δe,
vk)δg−1
vk)δgg−1
vkg−1 )δe
(1f (vhg−1(kg−1)−1 · 1f )#vkg−1 )δe
(1f (vhk−1 · 1f )#vkg−1 )δe
and this last term belongs to We,f since r(gh−1) = r(h) = f and d(gh−1) = r(g) = e. One easily
sees from the above that this action of Ue,f on We,f is transitive.
(cid:3)
Proposition 4.4. Let e, f ∈ G0 and h ∈ TfT Se. Then,
Ce,f ≃ Mne,f (Mg∈Ge
Ag),
as unital K-algebras, where ne,f denotes the cardinality of the orbit of we,f,h.
18
DAIANA FL ORES AND ANTONIO PAQUES
Proof. It follows from Lemmas 4.3 and 3.1 that
where Se,f = we,f,hCe,f we,f,h, for some we,f,h ∈ We,f .
Ce,f ≃ Mne,f (Se,f ),
Now, note that Se,f = Lg∈Ge
xwe,f,h = (al#vk)δg(1f #vh)δe
(Ahgh−1 #vhg−1 )δg. Indeed, for any x = (al#vk)δg ∈ Ce,f we have
= (al#vk)βg(1f #vh)δg
= (al#vk)(1f #vhg−1 )δg
= (al(vkgh−1 · 1f )#vhg−1 )δg
=((al#vhg−1 )δg
0
if kg = h
otherwise,
since l ∈ Gf and al1f = al by Remark 2.4(v). Thus, if kg = h
we,f,hxwe,f,h = (1f #vh)δe(al#vhg−1 )δg
= (1f #vh)βe(al#vhg−1 )δg
= (1f #vh)(al#vhg−1 )δg
= (1f (vhgh−1 · al)#vhg−1 )δg
= (ahgh−1 #vhg−1 )δg,
because vhgh−1 .al 6= 0 if and only if hgh−1 = l, and 1f al = al by Remark 2.4(v).
for all ag ∈ Ag.
Since h ∈ TfT Se, it is routine to check that the map g 7→ hgh−1 from Ge to Gf is a bijection
and induces the isomorphism of K-algebras θh : Lg∈Ge
Finally, the map γ : Lg∈Ge
all g ∈ Ge and ag ∈ Ag, is an isomorphism of K-algebras. Indeed, clearly γ is an isomorphism of
K-modules (induced by θh), and
(Ahgh−1 #vhg−1 )δg given by γ(ag) = (ahgh−1#vhg−1 )δg, for
Ag → Lg∈Ge
Ag → Lg∈Ge
Ahgh−1 given by θh(ag) = ahgh−1 ,
γ(ag)γ(bl) = (ahgh−1 #vhg−1 )δg(bhlh−1 #vhl−1 )δl
= (ahgh−1 #vhg−1 )βg(bhlh−1 #vhl−1 )δgl
= (ahgh−1 #vhg−1 )(bhlh−1 #vhl−1g−1 )δgl
= (ahgh−1 (vhg−1(hl−1g−1)−1 · bhlh−1)#vh(gl)−1 )δgl
= (ahgh−1 (vhgh−1 · bhgh−1 )#vh(gl)−1 )δgl
= (ahgh−1 bhlh−1#vh(gl)−1 )δgl
= (θ(ag)θ(bl)#vh(gl)−1 )δgl
= (θ(agbl)#vh(gl)−1 )δgl
= ((agbl)h(gl)h−1 #vh(gl)−1 )δgl
= γ(agbl),
for all ag ∈ Ag and bl ∈ Al. Therefore, we have from the above that
Se,f = Lg∈Ge
(Ahgh−1 #vhg−1 )δg ≃ Lg∈Ge
Ag
and
Ce,f ≃ Mne,f ( Lg∈Ge
Ag).
as unital K-algebras.
(cid:3)
Theorem 4.5.
as unital K-algebras.
DUALITY FOR GROUPOID (CO)ACTIONS
C0 ≃ Me∈G0
(Mf ∈G0
Mne,f (Mg∈Ge
Ag)),
Proof. It follows by Lemmas 4.1 and 4.2, and Proposition 4.4.
5. Final remarks
19
(cid:3)
R such that every Ee, e ∈ G0, is unital and R =Le∈G0
Let G be a finite groupoid, R a unital K-algebra and β = ({Eg}g∈G, {βg}g∈G) an action of G on
Ee. By Proposition 2.2 R is a KG-module
algebra and we can consider the smash product algebra as in [15], which is constructed as follows:
Given a weak Hopf algebra H, denote by Ht the target counital subalgebra of H defined by
Ht := {h ∈ H εt(h) = h} = εt(H), where εt(h) = ε(11h)12, for every h ∈ H. The algebra H has
a natural structure of a left Ht-module via multiplication, and any H-module algebra X is a right
Ht-module via the antipode S of H, that is, x · z := S(x) · x, for all x ∈ X and z ∈ Ht. Hence, we
can take the K-module X ⊗Ht H, which has an structure of a unital K-algebra induced by the
multiplication (x ⊗ h)(y ⊗ l) = x(h1 · y) ⊗ h2l, for all x, y ∈ X and h, l ∈ H. Its identity element
is 1X ⊗ 1H. Notice that X ⊗Ht H also is a left H ∗-module algebra via h∗ · (x ⊗ l) = x ⊗ (h∗ ⇀ l),
for all h∗ ∈ H ∗, l ∈ H and x ∈ X, so we can also consider the K-algebra (XNHt
Our intent in this section is to present a natural exact sequence of K-algebras relating B =
t H ∗, in the case
(R ⋆β G)#KG∗, as considered in the subsection 2.4, to A = (R ⊗Ht H) ⊗H∗
that H = KG. Observe that KG (resp., KG∗) is a weak Hopf algebra, with antipode given
by S(ug) = ug−1 (resp., S(vg) = vg−1 ). We start with the following proposition. Recall that
Te = {g ∈ G r(g) = e}, for all e ∈ G0.
H ∗.
H)NH∗
t
Proposition 5.1. There exist a unital subalgebra C of B containing B0 as subalgebra, and an
ideal D of B such that B = CL D and BD = DB = 0.
B = Mg,h∈G
Egδg#vh = CM D, where C = Md(g)=r(h)
Proof. Notice that
Egδg#vh and D = Md(g)6=r(h)
Egδg#vh.
Furthermore, B0 =
Egδg#vh is a direct summand of C.
It is a routine calcula-
tion to check that C (resp., B0) is a unital subalgebra of B (resp., C) with identity element
vg (see Lemma 3.2(i)), as well as D is an ideal of B. We saw in the subsection 2.5
Lr(g)=d(g)=r(h)
that BD = 0. It follows by similar arguments that also DB = 0.
(cid:3)
Pe∈G0
1eδe# Pg∈Te
Theorem 5.2. The natural map ϕ : B → A, given by agδg#vh 7→ ag ⊗ ug ⊗ vh, induces the
following exact sequence of K-algebras
0 −→ D −→ B −→ ϕ(C) −→ 0.
In particular, B0 is isomorphic to a subalgebra of A.
To prove this theorem we need first to describe the elements of KGt and KG∗
t .
Lemma 5.3.
KGt = Me∈G0
Kue
and KG∗
t = Xe∈G0
K(Xh∈Te
vh).
20
DAIANA FL ORES AND ANTONIO PAQUES
Proof. An element x = Pg∈G
Xg∈G
λgug ∈ KG satisfies εt(x) = x if and only if
λgug =Xg∈G
=Xg∈G
λgεt(ug) =Xg∈G
λg Xe∈G0
λgε(ur(g)g)ur(g) =Xg∈G
ε(ueug)ue
λgur(g),
if and only if λg = 0, para all g /∈ G0.
An element x = Pg∈G
λgvg ∈ KG∗ satisfies εt(x) = x if and only if
Xg∈G
λgvg =Xg∈G
=Xg∈G
=Xg∈G
= Xg∈G0
λgεt(vg)
d(l)=d(h)
λg(Xh∈G Xl∈G
λg( Xh∈G
λg(Xh∈Tg
r(h)=r(g)
vh)
ε(vhl−1 vg)vl)
ε(vg)vg−1 h)
if and only if λh = λg, for all g ∈ G0 and h ∈ Tg.
(cid:3)
Proof of Theorem 5.2:
It is straightforward to check that the map ϕ : B → A, given by ϕ(agδg#vh) = ag ⊗ ug ⊗ vh,
is a well defined homomorphism of K-algebras. Furthermore, the preunit 1R⋆β G#1KG∗ of B is
taken by ϕ onto the identity element 1R ⊗ 1KG ⊗ 1KG∗ of A. Indeed,
vg) = Pe∈G0 Pg∈G
1e#ue#vg.
Kue (Lemma 5.3), we have
1eδe# Pg∈G
ϕ(1R⋆β G#1KG∗) = ϕ( Pe∈G0
And, on the other hand, since KGt = Le∈G0
uf #Xg∈G
1e# Xf ∈G0
S(uf ).1e#uf #Xg∈G
βf (1e1f )#uf #Xg∈G
1R#1KG#1KG∗ = Xe∈G0
= Xe,f ∈G0
= Xe,f ∈G0
vg
vg = Xe,f ∈G0
vg = Xe,f ∈G0
vg = Xe∈G0
1e.uf #uf #Xg∈G
uf .1e#uf #Xg∈G
1e#ue#Xg∈G
vg,
vg
This implies that the ideal D of B is contained in the kernel of ϕ because (1R⋆β G#1KG∗)D = 0
(Proposition 5.1) and so
0 = ϕ((1R⋆β G#1KG∗ )D) = ϕ(1R⋆β G#1KG∗)ϕ(D) = 1Aϕ(D) = ϕ(D).
From this we also have ϕ(B) = ϕ(C) = Ld(g)=r(h)
For this, take the map ψ : Ld(g)=r(h)
EgNKGt
KugNKG∗
t
Kvh.
To end this proof it is enough to show that the K-algebras ϕ(C) and B/D are isomorphic.
Eg × Kug × Kvh → B/D, given by ψ(ag, ug, vh) = agδg#vh.
Notice that for x = Pe∈G0
DUALITY FOR GROUPOID (CO)ACTIONS
21
λeue ∈ KGt we have
ψ(ag.x, ug, vh) = ψ(S(x).ag, ug, vh) = ψ((Xe∈G0
λeβe(ag1e), ug, vh)
= ψ(Xe∈G0
= ψ(λr(g)ag, ug, vh) = λr(g)agδg#vh
λeue).ag, ug, vh)
and
Hence, ψ is KGt-balanced.
ψ(ag, xug, vh) = ψ(ag, Xe∈G0
λe(Pl∈Te
Furthermore, for x = Pe∈G0
vl) ∈ KG∗
t we also have
ψ(ag, ug.x, vh) = ψ(ag, S(x).ug, vh)
λeueug, vh) = ψ(ag, λr(g)ug, vh) = λr(g)agδg#vh.
vl)).ug, vh)
vl−1 )).ug, vh)
vl−1 .ug), vh)
= ψ(ag, S(Xe∈G0
= ψ(ag, (Xe∈G0
= ψ(ag, Xe∈G0
λe(Xl∈Te
λe(Xl∈Te
λe(Xl∈Te
= ψ(ag, λd(g)ug, vh)
= ψ(ag, λd(g)ug, vh)
and
= λd(g)agδg#vh
ψ(ag, ug, x.vh) = ψ(ag, ug, (Xe∈G0
= ψ(ag, ug, Xe∈G0
λe(Xl∈Te
λe(Xl∈Te
vl))vh)
vlvh))
= ψ(ag, ug, λr(h)vh)
= ψ(ag, ug, λr(h))vh
= λr(h)agδg#vh
= λd(g)agδg#vh,
Thus, ψ also is KG∗
t -balanced. Therefore, ψ induces a K-linear map eψ from ϕ(C) into B/D. It is
immediate to see that this map is the inverse of the K-algebra homomorphism eϕ from B/D onto
ϕ(C) induced by ϕ.
(cid:3)
References
[1] D. Bagio, D. Flores and A. Paques, Partial actions of ordered groupoids on rings, J. Algebra Appl. 9 (2010),
501-517.
[2] D. Bagio and A. Paques, Partial groupoid actions: globalization, Morita theory and Galois theory, Comm.
Algebra 40 (2012), 3658-3678.
[3] M. van den Bergh, A duality theorem for Hopf algebras, Methods in Ring Theory, (Antwerp, 1983), 517-522.
[4] R. Blattner and S. Montgomery, A duality theorem for Hopf module algebras, J. Algebra 95 (1985), 153-172.
[5] G. Bohm, Doi-Hopf modules over weak Hopf algebras, Comm. Algebra 28 (2000), 4687-4698.
[6] G. Bohm, F.Nill and K. Szlach´anyi, Weak Hopf algebra I. Integral theory and C ⋆-structure, J. Algebra 221
(1999), 385-438.
22
DAIANA FL ORES AND ANTONIO PAQUES
[7] S. Caenepeel and E. De Groot, Modules over weak entwining structures, Contemporary Math. 267 (2000),
31-54.
[8] S. Caenepeel and E. De Groot, Galois theory for weak Hopf algebras, Rev. Roumaine Math. Pures Appl. 52
(2007), 151-176.
[9] M. Cohen and S. Montgomery, Group-graded rings, smash products and group actions, Trans. AMS 282 (1984),
237-258.
[10] M. Dokuchaev and R. Exel, Associativity of crossed products by partial actions, enveloping actions and partial
representations, Trans. AMS 357 (2005), 1931-1952.
[11] N.D. Gilbert, Actions and expansions of ordered groupoids, J. Pure Appl. Algebra 198 (2005), 175-195.
[12] M.V. Lawson, Inverse Semigroups. The Theory of Partial Symmetries, World Scientific Pub. Co, London, 1998.
[13] G. Liu and F. Li, On strongly groupoid graded rings and the corresponding Clifford Theorem, Algebra Collo-
quium, 13 (2006), 181-196.
[14] P. Lundstrom, Separable groupoid rings, Comm. Algebra 34 (2006), 3029-3041.
[15] D. Nikshych, A duality theorem for quantum groups, Contemporary Math. 267 (2000), 237-243.
[16] D. S. Passman, The Algebraic Structure of Finite Groups, Wiley-Interscience, NY, 1977.
Departamento de Matem´atica, Universidade Federal de Santa Maria, 97105-900, Santa Maria, RS,
Brazil
E-mail address: [email protected]
Instituto de Matem´atica, Universidade Federal do Rio Grande do Sul, 91509-900, Porto Alegre, RS,
Brazil
E-mail address: [email protected]
|
1911.08302 | 1 | 1911 | 2019-11-15T20:38:46 | Some remarks regarding finite bounded commutative BCK-algebras | [
"math.RA",
"cs.IT",
"cs.IT"
] | In this chapter, starting from some results obtained in the papers [FV; 19], [FHSV; 19], we provide some examples of finite bounded commutative BCK- algebras, using the Wajsberg algebra associated to a bounded commutative BCK- algebra. This method is an alternative to the Iseki's construction, since by Iseki's extension some properties of the obtained algebras are lost. | math.RA | math |
Some remarks regarding finite bounded commutative
BCK-algebras
Cristina Flaut, S´arka Hoskov´a-Mayerov´a and Radu Vasile
Abstract.
In this chapter, starting from some results obtained in the papers
[FV; 19], [FHSV; 19], we provide some examples of finite bounded commutative BCK-
algebras, using the Wajsberg algebra associated to a bounded commutative BCK-
algebra. This method is an alternative to the Iseki's construction, since by Iseki's
extension some properties of the obtained algebras are lost.
Keywords: Bounded commutative BCK-algebras, MV-algebras, Wajsberg al-
gebras.
AMS Classification: 06F35, 06F99.
1. Introduction
BCK-algebras were first introduced in mathematics in 1966 by Y. Imai and
K. Iseki, through the paper [II; 66]. These algebras were presented as a general-
ization of the concept of set-theoretic difference and propositional calculi. The
class of BCK-algebras is a proper subclass of the class of BCI-algebras.
Definition 1.1. An algebra (X, ∗, θ) of type (2, 0) is called a BCI-algebra
if the following conditions are fulfilled:
1) ((x ∗ y) ∗ (x ∗ z)) ∗ (z ∗ y) = θ, for all x, y, z ∈ X;
2) (x ∗ (x ∗ y)) ∗ y = θ, for all x, y ∈ X;
3) x ∗ x = θ, for all x ∈ X;
4) For all x, y, z ∈ X such that x ∗ y = θ, y ∗ x = θ, it results x = y.
If a BCI-algebra X satisfies the following identity:
5) θ ∗ x = θ, for all x ∈ X, then X is called a BCK-algebra.
Definition 1.2. i) A BCK-algebra (X, ∗, θ) is called commutative if
x ∗ (x ∗ y) = y ∗ (y ∗ x),
for all x, y ∈ X and implicative if
x ∗ (y ∗ x) = x,
1
for all x, y ∈ X.
ii) ([Du; 99]) A BCK-algebra (X, ∗, θ) is called positive implicative if and
only if
for all x, y, z ∈ X.
(x ∗ y) ∗ z = (x ∗ z) ∗ (y ∗ z),
The partial order relation on a BCK-algebra is defined such that x ≤ y if
and only if x ∗ y = θ.
If in the BCK-algebra (X, ∗, θ) there is an element 1 such that x ≤ 1, for
all x ∈ X, then the algebra X is called a bounded BCK-algebra. In a bounded
BCK-algebra, we denote 1 ∗ x = x.
If in the bounded BCK-algebra X, an element x ∈ X satisfies the relation
x = x,
then the element x is called an involution.
If (X, ∗, θ) and (Y, ◦, θ) are two BCK-algebras, a map f : X → Y with the
property f (x ∗ y) = f (x) ◦ f (y) , for all x, y ∈ X, is called a BCK-algebras
morphism. If f is a bijective map, then f is an isomorphism of BCK-algebras.
Definition 1.3. 1) Let (X, ∗, θ) be a BCK algebra and Y be a nonempty
subset of X. Therefore, Y is called a subalgebra of the algebra (X, ∗, θ) if and
only if for each x, y ∈ Y , we have x ∗ y ∈ Y . This implies that Y is closed to
the binary multiplication "∗".
It is well known that each BCK-algebra of degree n+1 contains a subalgebra
of degree n.
2) Let (X, ∗, θ) be a BCK algebra and I be a nonempty subset of X. There-
fore, I is called an ideal of the algebra X if and only if for each x, y ∈ X we
have:
i) θ ∈ I;
ii) x ∗ y ∈ I and y ∈ I, then x ∈ I.
Proposition 1.4. ([Me-Ju; 94]) Let (X, ∗, θ) be a BCK algebra and Y be a
subalgebra of the algebra (X, ∗, θ). The following statements are true:
i) θ ∈ Y ;
ii) (Y, ∗, θ) is also a BCK-algebra.(cid:3)
Let X be a BCK-algebra, such that 1 /∈ X. On the set Y = X ∪ {1}, we
define the following multiplication " ◦ ", as follows:
x ◦ y =
x ∗ y, if x, y ∈ X;
θ, if x ∈ X, y = 1;
1, if x = 1 and y ∈ X;
θ, if x = y = 1.
The obtained algebra (Y, ◦, θ) is a bounded BCK-algebra and is obtained by
the so-called Iseki's extension. The algebra (Y, ◦, θ) is called the algebra obtained
from algebra (X, ∗, θ) by Iseki's extension ([Me-Ju; 94], Theorem 3.6).
2
Remark 1.5. ([Me-Ju; 94])
i) The Iseki's extension of a positive implicative BCK-algebra is still a posi-
tive implicative BCK-algebra.
ii) The Iseki's extension of a commutative BCK-algebra, in general, is not
a commutative BCK-algebra.
iii) Let X be a BCK-algebra and Y its Iseki's extension. Therefore X is an
ideal in Y .
iv) If I is an ideal of the BCK-algebra X, x ∈ I and y ≤ x, then y ∈ I.
In the following, we will give some examples of finite bounded commutative
BCK-algebras. In the finite case, it is very useful to have many examples of
such algebras. But, such examples, in general, are not so easy to found. A
method for this purpose can be Iseki's extension. But, from the above, we
remark that the Iseki's extension can't be always used to obtain examples of
finite commutative bounded BCK-algebras with given initial properties, since
the commutativity, or other properties, can be lost. From this reason, we use
other technique to provide examples of such algebras. We use the connections
between finite commutative bounded BCK-algebras and Wajsberg algebras and
the algorithm and examples given in the papers [FHSV; 19] and [FV; 19].
2. Connections between finite bounded commutative BCK-algebras
and Wajsberg algebras
Definition 2.1.
([CHA; 58]) An abelian monoid (X, θ, ⊕) is called MV-
algebra if and only if we have an unary operation "′" such that:
i) (x′)′ = x;
ii) x ⊕ θ′ = θ′;
iii) (x′ ⊕ y)′ ⊕ y = (y ′ ⊕ x)′ ⊕ x, for all x, y ∈ X.([Mu; 07]). We denote it by
(X, ⊕,′ , θ) .
We remark that in an MV-algebra the constant element θ′ is denoted with
1. This is equivalent with
With the above definitions, the following multiplications are also defined:
1 = θ′.
x ⊙ y = (x′ ⊕ y ′)′ ,
x ⊖ y = x ⊙ y ′ = (x′ ⊕ y)′ .
(see ([Mu; 07]))
From [COM; 00], Theorem 1.7.1, for a bounded commutative BCK-algebra
(X, ∗, θ, 1), if we define
x′ = 1 ∗ x,
3
x ⊕ y = 1 ∗ ((1 ∗ x) ∗ y) = (x′ ∗ y)′ , x, y ∈ X,
we obtain that the algebra (X, ⊕,′ , θ) is an MV -algebra, with
x ⊖ y = x ∗ y.
The converse is also true, that means if (X, ⊕, θ,′ ) is an MV-algebra, then
(X, ⊖, θ, 1) is a bounded commutative BCK-algebra.
Definition 2.2.
, 1) of
type (2, 1, 0) is called a Wajsberg algebra ( or W-algebra) if and only if the
following conditions are fulfilled:
([COM; 00], Definition 4.2.1) An algebra (W, ◦,
i) 1 ◦ x = x;
ii) (x ◦ y) ◦ [(y ◦ z) ◦ (x ◦ z)] = 1;
iii) (x ◦ y) ◦ y = (y ◦ x) ◦ x;
iv) (x ◦ y) ◦ (y ◦ x) = 1, for every x, y, z ∈ W .
Remark 2.3. ([COM; 00], Lemma 4.2.2 and Theorem 4.2.5)
i) For the Wajsberg algebra (W, ◦,
, 1), if we define the following multipli-
cations
and
x ⊙ y = (x ◦ y)
x ⊕ y = x ◦ y,
for all x, y ∈ W , we obtain that (W, ⊕, ⊙,
, θ, 1) is an MV-algebra.
ii) Conversely, if (X, ⊕, ⊙,′ , θ, 1) is an MV-algebra, defining on X the oper-
ation
we obtain that (X, ◦,′ , 1) is a Wajsberg algebra.
x ◦ y = x′ ⊕ y,
Remark 2.4. From the above, if (W, ◦,
, 1) is a Wajsberg algebra, then
(W, ⊕, ⊙,
, 0, 1) is an MV-algebra, with
Defining
x ⊖ y = (x ⊕ y) = (x ◦ y).
x ∗ y = (x ◦ y),
(2.1.)
(2.2.)
we have that (W, ∗, θ, 1) is a bounded commutative BCK-algebra.
Using the above remark, starting from some known finite examples of Wa-
jsberg algebras given in the papers [FHSV; 19] and [FV; 19], we can obtain
examples of finite commutative bounded BCK-algebras, using the following al-
gorithm.
The Algorithm
1) Let n be a natural number, n 6= 0 and
n = r1r2...rt, ri ∈ N, 1 < ri < n, i ∈ {1, 2, ..., t},
4
be the decomposition of the number n in factors. The decompositions with the
same terms, but with other order of them in the product, will be counted one
time. The number of all such decompositions will be denoted with πn.
2) There are only πn nonismorphic, as ordered sets, Wajsberg algebras with
n elements. We obtain these algebras as a finite product of totally ordered
Wajsberg algebras (see [BV; 10] and [FV; 19], Theorem 4.8 ).
3) Using Remark 2.4 from above, a commutative bounded BCK-algebra can
be associated to each Wajsberg algebra.
3. Examples of finite commutative bounded BCK-algebras
In the following, we use some examples of Wajsberg algebras given in the
paper [FHSV; 19]. To these algebras, we will associate the corresponding com-
mutative bounded BCK-algebras and we give their subalgebras and ideals.
Example 3.1. Let W = {O ≤ A ≤ B ≤ E} be a totally ordered set. On
W we define the multiplication ◦1 as in the below table, such that (W, ◦1, E) is
a Wajsberg algebra. We have A = B and B = A.
◦1 O A B E
O E E E E
A B E E E
B A B E E
E O A B E
.
(see [FHSV; 19], Example 4.1.1)
Therefore, the associated commutative bounded BCK-algebras (W, ∗1, O) has
multiplication given in the below table:
∗1 O A B E
O O O O O
A A O O O
B B A O O
E E B A O
.
The proper subalgebras of this algebra are: {O, A}, {O, B}, {O, E}, {O, A, B}.
There are no proper ideals in the algebra (W, ∗1, O).
Example 3.2. Let W = {O ≤ A ≤ B ≤ E} be a totally ordered set. On
W we define the multiplication ◦2 as in the below table, such that (W, ◦2, E) is
a Wajsberg algebra.
◦2 O A B E
O E E E E
A B E B E
B A A E E
E O A B E
.
5
Therefore, the associated commutative bounded BCK-algebras (W, ∗2, O)
has multiplication given below:
∗2 O A B E
O O O O O
A A O A O
B B B O O
E E B A O
.
The proper subalgebras of this algebra are: {O, A}, {O, B}, {O, E}, {O, A, B}.
The proper ideals are: {O, A}, {O, B}.
Example 3.3. Let W = {O ≤ A ≤ B ≤ C ≤ D ≤ E} be a totally ordered
set. On W we define a multiplication ◦3 given in the below table, such that
(W, ◦3, E) is a Wajsberg algebra. We have A = D, B = C, C = B, D = A.
◦3 O A B C D E
O E E E E E E
A D E E E E E
B C D E E E E
C B C D E E E
D A B C D E E
E O A B C D E
.
Therefore, the associated commutative bounded BCK-algebras (W, ∗3, O) has
multiplication defined below:
∗3 O A B C D E
O O O O O O O
A A O O O O O
B B A O O O O
C
C B A O O O
D D C B A O O
E E D C B A O
.
The proper subalgebras of this algebra are: {O, A}, {O, B}, {O, C}, {O, D},
{O, E}, {O, A, B}, {O, B, D}, {O, A, B, C}, {O, A, B, C, D} .
This algebra has no proper ideals.
Example 3.4. Let W = {O ≤ A ≤ B ≤ C ≤ D ≤ E} be a totally ordered
set. On W we define a multiplication ◦4 given in the below table, such that
(W, ◦4, E) is a Wajsberg algebra. We have
◦4 O A B C D E
O E E E E E E
A D E E D E E
B C D E C D E
C B B B E E E
D A B B D E E
E O A B C D E
.
6
Therefore, the associated commutative bounded BCK-algebras (W, ∗4, O) has
multiplication given in the following table:
∗4 O A B C D E
O O O O O O O
A A O O A O O
B B A O B A O
C
C C C O O O
D D C C A O O
E E D C B A O
.
The proper subalgebras of this algebra are: {O, A}, {O, B}, {O, C}, {O, D},
{O, E}, {O, A, B}, {O, A, B, C}, {O, A, B, C, D}, {O, A, C}, {O, A, C, D}.
The proper ideals of this algebra are: {O, A, B}, {O, C}.
Example 3.5. Let W = {O ≤ A ≤ B ≤ C ≤ D ≤ E} be a totally ordered
set. On W we define a multiplication ◦5 given in the below table, such that
(W, ◦5, E) is a Wajsberg algebra. We have
◦5 O A B C D E
O E E E E E E
A C E A D D E
B D E E D D E
C A E A E E E
D B A B A E E
E O A B C D E
.
Therefore, the associated commutative bounded BCK-algebras (W, ∗5, O) has
multiplication defined in the following table:
∗5 O A B C D E
O O O O O O O
A A O C B B O
B B O O B B O
C O C O O O
C
D D C D C O O
E E C D A B O
.
The proper subalgebras of this algebra are: {O, A}, {O, B}, {O, C}, {O, D},
{O, E}, {O, B, C}, {O, C, D}, {O, A, B, C}, {O, A, B, C, D}.
All proper ideals are: {O, C, D}, {O, B}.
Example 3.6. Let W = {O ≤ X ≤ Y ≤ Z ≤ T ≤ U ≤ V ≤ E} be a totally
ordered set. On W we define a multiplication ◦6 which can be found in the below
table. We obtain that (W, ◦6, E) is a Wajsberg algebra.We have X = V , Y = U ,
7
Z = T . Therefore the algebra W has the following multiplication table:
Z T
◦6 O X Y
U V E
O E E E E E E E E
X V
E E E E E E E
Y
E E E E E E
V
U
V E E E E E
U
T
Z
U V E E E E
T
Z
T
Y
Z
T
U V E E E
U
U V E E
V X Y
Z T
E O X Y
U V E
Z T
.
From here, we get thet the associated commutative bounded BCK-algebras
(W, ∗6, O) has multiplication given in the below table:
U
V
T
Z
∗6 O X Y
E
O O O O O O O O O
X X O O O O O O O
Y X O O O O O O
Y
Y X O O O O O
Z
Z
Y X O O O O
Z
T
T
U
U
T
Z
Y X O O O
Y X O O
Z
T
V
U
V
E E V
U
T
Z
T X O
.
The proper subalgebras of this algebra are: {O, J}, J ∈ {X, Y, Z, T, U, V, E},
{O, X, Y }, {O, X, Y, Z}, {O, X, Y, Z, T }, {O, X, Y, Z, T, U }, {O, X, Y, Z, T, U, V }.
There are no proper ideals.
Example 3.7. Let W = {O ≤ X ≤ Y ≤ Z ≤ T ≤ U ≤ V ≤ E} be a totally
ordered set. On W we define a multiplication ◦7 given in the below table, such
that (W, ◦7, E) is a Wajsberg algebra.
Z T
◦7 O X Y
U V E
O E E E E E E E E
X V
E V E V E V E
U E E E E E E
U
Y
V E V E V E
U
T
Z
U U E E E E
Z
Z
T
U
Y
Z
T
U T E V E
V X X Z Z U U E E
E O X Y
U V E
Z T
.
Therefore, the associated commutative bounded BCK-algebras (W, ∗7, O) has
8
multiplication defined in the below table:
Z T
U V
∗7 O X Y
E
O O O O O O O O O
X X O X O X O X O
Y
Y
O O O O O O
Y X O X O X O
Z
Y O O O O
T
T
T
Y
Z O X O
U
Y O O
Y
T
V
V
E E V
T
Z
Y X O
Y
Z
T
U
V
Y
Z
T
U
.
The proper subalgebras of this algebra are: {O, J}, J ∈ {X, Y, Z, T, U, V, E},
{O, X, Y }, {O, X, Y, Z}, {O, T, Y }, {O, T, Y, V }, {O, X, Y, Z, T }, {O, X, Y, Z, T, U },
{O, X, Y, Z, T, U, V }.
All proper ideals are: {O, Y, T, V }, {O, X}.
Conclusions.
In this chapter, we provided an algorithm for finding ex-
amples of finite commutative bounded BCK-algebras, using their connections
with Wajsberg algebras. This algorithm allows us to find such examples no
matter the order of the algebra. This thing is very useful, since examples of
such algebras are very rarely encountered in the specialty books.
References
[AAT; 96] Abujabal, H.A.S., Aslam, M., Thaheem, A.B., A representation
of bounded commutative BCK-algebras, Internat. J. Math. & Math. Sci.,
19(4)(1996), 733-736.
[BV; 10] Belohlavek, R., Vilem Vychodil, V., Residuated Lattices of Size
≤ 12, Order, 27(2010), 147-161.
[CHA; 58] Chang, C.C., Algebraic analysis of many-valued logic, Trans.
Amer. Math. Soc. 88(1958), 467-490.
[COM; 00] Cignoli, R. L. O, Ottaviano, I. M. L. D, Mundici, D., Algebraic
foundations of many-valued reasoning, Trends in Logic, Studia Logica Library,
Dordrecht, Kluwer Academic Publishers, 7(2000).
[Du; 99] W.A. Dudek, On embedding Hilbert algebras in BCK-algebras,
Mathematica Moravica, 3(1999), 25-28.
[FHSV; 19] Flaut, C., Hoskov´a-Mayerov´a, S., Saeid, A., B., Vasile, R., Wa-
jsberg algebras of order n, n ≤ 9, https://arxiv.org/pdf/1905.05755.pdf
[FV; 19] Flaut, C., Vasile, R., Wajsberg algebras arising from binary block
codes, https://arxiv.org/pdf/1904.07169.pdf
[II; 66] Imai, Y., Iseki, K., On axiom systems of propositional calculi, Proc.
Japan Academic, 42(1966), 19-22.
9
[JS; 11] Jun, Y. B., Song, S. Z., Codes based on BCK-algebras, Inform.
Sciences., 181(2011), 5102-5109.
[Me-Ju; 94] Meng, J., Jun, Y. B., BCK-algebras, Kyung Moon Sa Co. Seoul,
Korea, 1994.
[Mu; 07] Mundici, D., MV-algebras-a short tutorial, Department of Mathe-
matics Ulisse Dini, University of Florence, 2007.
Cristina FLAUT
Faculty of Mathematics and Computer Science,
Ovidius University of Constant¸a, Romania,
Bd. Mamaia 124, 900527,
http://www.univ-ovidius.ro/math/
e-mail: [email protected]; cristina [email protected]
S´arka Hoskov´a-Mayerov´a
Department of Mathematics and Physics,
University of Defence, Brno, Czech Republic
e-mail: [email protected]
Radu Vasile,
PhD student at Doctoral School of Mathematics,
Ovidius University of Constant¸a, Romania
[email protected]
10
|
1609.07392 | 1 | 1609 | 2016-09-23T15:09:05 | Degenerations of binary Lie and nilpotent Malcev algebras | [
"math.RA"
] | We describe degenerations of four-dimensional binary Lie algebras, and five- and six-dimensional nilpotent Malcev algebras over \mathbb{C}. In particular, we describe all irreducible components of these varieties. | math.RA | math | Degenerations of binary Lie and nilpotent Malcev algebras 1
Ivan Kaygorodova, Yury Popovb, Yury Volkovc,d
a Universidade Federal do ABC, CMCC, Santo Andr´e, Brazil.
b Novosibirsk state university, Novosibirsk, Russia.
c Universidade de Sao Paulo, IME, Sao Paulo, Brazil.
d Saint Petersburg state university, Saint Petersburg, Russia.
E-mail addresses:
Ivan Kaygorodov ([email protected]),
Yury Popov ([email protected]),
Yury Volkov (wolf86 [email protected]).
Abstract. We describe degenerations of four-dimensional binary Lie algebras, and five- and six-dimensional nilpotent Malcev
algebras over C. In particular, we describe all irreducible components of these varieties.
Keywords: Malcev algebra, binary Lie algebra, nilpotent algebra, degeneration, rigid algebra
6
1
0
2
p
e
S
3
2
]
.
A
R
h
t
a
m
[
1
v
2
9
3
7
0
.
9
0
6
1
:
v
i
X
r
a
1. INTRODUCTION
Degenerations of algebras is an interesting subject, which was studied in various papers (see, for example, [1–9, 12–14, 16,
17, 25, 26, 28, 30]). In particular, there are many results concerning degenerations of algebras of low dimensions from some
variety defined by a set of identities. One of important problems in this direction is the description of so-called rigid algebras.
These algebras are of big interest, since the closures of their orbits under the action of generalized linear group form irreducible
components of a variety under consideration (with respect to Zariski topology). For example, the problem of finding rigid
algebras was solved for low-dimensional associative (see [9, 25, 26]), Leibniz (see [1, 8, 28]), and Jordan (see [16, 17]) algebras.
There are significantly less works where the full information about degenerations was found for some variety of algebras. This
problem was solved for four-dimensional Lie algebras in [7], for nilpotent five- and six-dimensional Lie algebras in [13, 30], for
two-dimensional pre-Lie algebras in [2], and for three-dimensional Novikov algebras in [3].
The notions of Malcev and binary Lie (BL for short) algebras were introduced by Malcev in [24]. The structure theory and
some properties of Malcev algebras were studied by Kuzmin and other authors (see, for example, [15, 18–21, 23, 27, 29]). Note
that any Lie algebra is a Malcev algebra and any Malcev algebra is a BL algebra. Note also that any alternative algebra can be
turned to a Malcev algebra by defining a new multiplication [, ] by [x, y] = xy − yx. Any Malcev algebra is a tangent algebra of
a suitable locally analytic Moufang loop (see [21]).
In this paper we give the full information about degenerations of BL algebras of dimension 4 and nilpotent Malcev algebras of
dimensions 5 and 6. More precisely, we construct a graph of primary degenerations. The vertices of this graph are isomorphism
classes of algebras from the variety under consideration. An algebra A degenerates to an algebra B iff there is a path from the
vertex corresponding to A to the vertex corresponding to B. Thus, we obtain a generalization of analogous results of [7, 13, 30]
for Lie algebras. Also we describe rigid algebras and irreducible components for these varieties of algebras.
All spaces in this paper are considered over C, and we write simply dim, Hom and ⊗ instead of dimC, HomC and ⊗C. An
algebra A is a set with a structure of vector space and a binary operation that induces a bilinear map from A × A to A.
2. DEFINITONS AND NOTATION
Given an n-dimensional vector space V , the set Hom(V ⊗ V, V ) = V ∗ ⊗ V ∗ ⊗ V is a vector space of dimension n3. This
. Indeed, let us fix a basis e1, . . . , en of V . Then any µ ∈ Hom(V ⊗ V, V ) is
i,jek. A subset of Hom(V ⊗ V, V ) is called closed if it can
ck
space has the structure of the affine variety Cn3
defined by structure constants ck
i,j ∈ C such that µ(ei ⊗ ej) =
n
Let T be a set of polynomial identities. All algebra structures on V satisfying polynomial identities from T form a Zariski-
closed affine subset of the variety Hom(V ⊗ V, V ). We denote this subset by L(T ). The general linear group GLn(C) operates
on L(T ) by conjugation:
(g ∗ µ)(x ⊗ y) = g(µ(g−1(x) ⊗ g−1(y)))
for x, y ∈ V , µ ∈ L(T ) ⊂ Hom(V ⊗ V, V ) and g ∈ GLn(C). Thus, L(T ) is decomposed into GLn(C)-orbits that correspond
to the isomorphism classes of algebras. Let O(µ) denote the orbit of µ ∈ L(T ) under the action of GLn(C). Correspondingly,
O(µ) is the Zariski closure of O(µ).
Let A and B be two n-dimensional algebras satisfying identities from T . Let µ and λ from L(T ) represent A and B respec-
tively. We say that A degenerates to B and write A → B if λ ∈ O(µ). Note that in this case we have O(λ) ⊂ O(µ). Hence, the
definition of degeneration does not depend on the choice of µ and λ. We write A 6→ B if λ 6∈ O(µ).
1The authors were supported by RFBR 16-31-00004, FAPESP 14/24519-8, FAPESP 14/19521-3 and by R & D 6.38.191.2014 of Saint-Petersburg State
University, "Structure theory, classification, geometry, K-theory and arithmetics of algebraic groups and related structures".
1
be defined by a set of polynomial equations in variables ck
i,j (1 ≤ i, j, k ≤ n).
Pk=1
2
Let A be represented by the structure µ ∈ L(T ). The algebra A is called rigid in L(T ) if O(µ) is an open subsetset of L(T ).
Recall that a subset of a variety is called irreducible if it can't be represented as a union of two non-trivial closed subsets. A
maximal irreducible closed subset of a variety is called an irreducible component. In particular, A is rigid in L(T ) iff O(µ) is an
irreducible component of L(T ). Let Irr(L(T )) and Rig(L(T )) denote the set of irreducible components of L(T ) and the set of
rigid algebras in L(T ) respectively. It is well known that any affine variety can be represented as a finite union of its irreducible
components in a unique way.
Let A be an algebra. For x, y, z ∈ A we define their Jacobian J(x, y, z) by the equality J(x, y, z) = (xy)z + (yz)x + (zx)y.
The algebra A is called a Malcev algebra if it satisfies the identities
The algebra A is called a binary Lie (BL for short) algebra if all its 2-generated subalgebras are Lie algebras. It was shown by
Gainov in [11] that A is a BL algebra iff it satisfies the identities
xy = −yx, J(x, y, xz) = J(x, y, z)x.
xy = −yx, J(x, y, xy) = 0.
It is easy to see that any Lie algebra is a Malcev algebra and any Malcev algebra is a BL algebra. It was shown in [10] that any
three-dimensional BL algebra is a Lie algebra. The classification of four-dimensional BL algebras was obtained in [10, 22]. The
classification of five-dimensional and nilpotent six-dimensional non-Lie Malcev algebras is given in [20].
Let BLn, M aln and Lien denote the varieties of n-dimensional BL, Malcev and Lie algebras respectively, and N BLn,
N M aln and N Lien denote their subvarieties formed by nilpotent algebras.
Define the sets Al by the equalities A1 = A and Al = Al−1A (l > 1). Also define the central series Zl(A) (l > 0) of A in the
following way. We define Z1(A) = Z(A) as the center of A, and, for l > 1, Zl(A) is the full inverse image of Z(A/Zl−1(A))
under the canonical projection from A to A/Zl−1(A).
We collect all the information that we need about the algebras under consideration in Tables 4.1, 5.1 and 6.1. In these tables
in the first column we write the names of the algebras. In the second column we give the multiplication tables in some fixed basis
e1, . . . , en of V . All products of basis elements, which are not described in the table, are zero or can be deduced from one of
the described products and the anticommutativity identity. In the third column we give the dimensions of algebras of derivations.
In the columns named Z(A), A2 and A3 we give the dimensions of the corresponding spaces. In the column named Zl(A) we
give the dimensions of the members of central series of A. Also, in the last column of Table 4.1 we have "Lie" for Lie algebras,
"Malcev" for Malcev non-Lie algebras and "BL" for BL non-Malcev algebras.
The names of four-dimensional Lie algebras are from [5]. The classification of BL non-Lie algebras is taken from [22]. One
of them is called g3(β) here, since for β = 2 we obtain the algebra g3 in the notation of [5]. The remaining BL non-Lie algebra
is called g6.
The names of five- and six-dimensional nilpotent Lie algebras are taken from [30]. The classification of Malcev non-Lie
algebras of corresponding dimensions is deduced from [20]. We give names containing the letter "M " to these algebras. So in
our notation a five- or six-dimensional algebra is Malcev and non-Lie iff it contains a letter "M " in its name, except the algebras
M 1
2 that correspond to the algebras g6,4 and g6,12 respectively in the notation of [30].
6 and M 1
3. METHODS
In the present work we use the methods that were applied for Lie algebras in [7, 13, 14, 30]. First of all, it is well known that
if A → B and A 6∼= B, then dim Der(A) < dim Der(B), where Der(A) is the algebra of derivations of A. We have computed
the dimensions of algebras of derivations and have checked the assertion A → B only for such A and B that dim Der(A) <
dim Der(B). Secondly, it is well known that if A → C and C → B, then A → B. If there is no C such that A → C and
C → B, then the assertion A → B is called a primary degeneration. If dim Der(A) < dim Der(B) and there are no C and D
such that C → A, B → D and C 6→ D, then the assertion A 6→ B is called a primary non-degeneration. It is enough to prove
only primary degenerations and non-degenerations to describe all degenerations in the variety under consideration. It is easy to
see that any algebra degenerates to the algebra with zero multiplication.
Degenerations of four-dimensional and nilpotent five- and six-dimensional Lie algebras were described in [7, 13, 30]. Since
the set L(T ) is closed for any T , a Lie algebra can't degenerate to a non-Lie algebra. So when we want to add Malcev or BL
algebras to Lie algebras we don't have to check the degenerations from Lie algebras to any of the added algebras.
To prove the primary degenerations we construct the families of matrices parametrized by t. Namely, let A and B be two
algebras represented by the structures µ and λ from L(T ) respectively. Let e1, . . . , en be a basis of V , for which λ is defined by
aj
i (t)ej is a basis
structure constants ck
i (t) ∈ C (1 ≤ i, j ≤ n, t ∈ C∗) such that Et
i,j (1 ≤ i, j, k ≤ n). If there exist aj
i =
n
Pj=1
of V for t ∈ C∗ and the structure constants of µ in the basis Et
then A → B. In this case Et
n is called a parametrized basis for A → B.
1, . . . , Et
1, . . . , Et
n are such polynomials ck
i,j(t) ∈ C[t] that ck
i,j(0) = ck
i,j,
Tables 4.2 and 6.2 give parametrized bases for primary degenerations between four-dimensional BL algebras and six-
dimensional nilpotent Malcev algebras respectively. These tables include all primary degenerations of the form A → B, where
A is a non-Lie algebra.
We now describe the methods for proving primary non-degenerations. The main tool for this is the following lemma.
Lemma 1 ( [4,13]). Let B be a Borel subgroup of GLn(C) and R ⊂ L(T ) be a B-stable closed subset. If A → B and A can be
represented by a structure µ ∈ R, then there is a structure λ ∈ R representing B.
3
Since any Borel subgroup of GLn(C) is conjugate to the subgroup of upper triangular matrices, Lemma 1 can be applied
in the following way. Let A and B be two algebras. Let µ, λ be some structures in L(T ) representing A and B respectively.
Suppose that there is a set of equations Q in variables xk
i,j (1 ≤ i, j, k ≤ n) is a solution
of all equations from Q, then xk
i,j (1 ≤ i, j, k ≤ n) is a solution for all equations from Q too in the following cases:
i,j (1 ≤ i, j, k ≤ n) such that if xk
i,j = ck
i,j = ck
i,j = αiαj
(1) if ck
(2) if there are some numbers 1 ≤ u < v ≤ n and some α ∈ C such that
i,j for some αi ∈ C∗ (1 ≤ i ≤ n);
ck
αk
v,u + ck
v,j − cu
i,v − cu
v,u + cv
u,v) + α2ck
v,v,
u,j) − α2cu
v,j,
i,u) − α2cu
i,v,
u,v − cu
u,u) + α2(cv
v,v − cu
v,u − cu
u,v) − α3cu
v,v,
if i, j 6= u and k 6= v,
if i = u, j 6= u and k 6= v,
if i 6= u, j = u and k 6= v,
if i, j 6= u and k = v,
if i = j = u and k 6= v,
if i = u, j 6= u and k = v,
if i 6= u, j = u and k = v,
if i = j = u and k = v.
ck
i,j =
ck
i,j,
u,j + αck
ck
v,j ,
i,u + αck
ck
i,v,
i,j − αcu
cv
i,j ,
u,u + α(ck
ck
cv
u,j + α(cv
i,u + α(cv
cv
cv
u,u + α(cv
Assume that there is a basis f1, . . . , fn of V such that the structure constants of µ in this basis form a solution for all equations
from Q, but there is no basis f1, . . . , fn of V such that the structure constants of λ in it form a solution for all equations from Q.
Then A 6→ B.
We will often use two particular cases of Lemma 1. Firstly, if dim Al < dim Bl for some l > 0, then A 6→ B. Secondly, if
dim Zl(A) > dim Zl(B) for some l > 0, then A 6→ B. In the cases where these two criterions can't be applied, we define R
by some conditions, which can be expressed in terms of a set of equations Q satisfying the property described above, and give a
basis for V , in which the structure constants of µ satisfy all equations from Q. We omit everywhere the verification of the fact
that Q satisfies the required conditions and the verification of the fact that structure constants of λ in any basis do not satisfy
some equation from Q. These verifications can be done by direct calculations.
Another argument for the non-degeneration that we use is the so-called (i, j)-invariant. Given i, j > 0, we call ci,j ∈ C an
(i, j)-invariant for the algebra A if
tr(ad x)i · tr(ad y)j = ci,jtr((ad x)i ◦ (ad y)j)
for all x, y ∈ A. If ci,j is an (i, j)-invariant for A, but at the same time it is not an (i, j)-invariant for B, then A 6→ B.
We give the proof of primary non-degenerations in Tables 4.3 and 6.3, where for each primary non-degeneration we give one
of the arguments mentioned above.
If the number of orbits under the action of GLn(C) on the variety L(T ) is finite, then the graph of primary degenerations
gives the whole picture. In particular, the description of rigid algebras and irreducible components can be easily obtained. But in
this work in some cases the situation is not so good. Then we have to be able to verify a little more complicated assertions. Let
A∗ = {Aα}α∈I be a set of algebras and B be some other algebra. Suppose that Aα is represented by the structure µα (α ∈ I)
and B is represented by the structure λ. Then A∗ → B means λ ∈ Sα∈I
O(µα), and A∗ 6→ B means λ 6∈ Sα∈I
Let A∗, B, µα (α ∈ I) and λ be as above. To prove that A∗ → B we have to construct a family of pairs (f (t), g(t))
parametrized by t, where f (t) ∈ I and g(t) ∈ GLn(C). Namely, let e1, . . . , en be a basis of V , for which λ is defined by
i (t) ∈ C (1 ≤ i, j ≤ n, t ∈ C∗) and f : C∗ → I such that
structure constants ck
n are such polynomials
aj
i (t)ej is a basis of V for t ∈ C∗ and the structure constants of µf (t) in the basis Et
i,j (1 ≤ i, j, k ≤ n). If we construct aj
1, . . . , Et
i =
Et
n
O(µα).
i,j(t) ∈ C[t] that ck
ck
parametrized index for A∗ → B respectively.
i,j(0) = ck
i,j, then A∗ → B.
In this case Et
1, . . . , Et
n and f (t) are called a parametrized basis and a
We now explain how to prove that A∗ 6→ B. First of all, if dim Der(Aα) > dim Der(B) for all α ∈ I, then A∗ 6→ B. One
can use also the following generalization of Lemma 1, whose proof is the same as the proof of Lemma 1.
Lemma 2. Let B be a Borel subgroup of GLn(C) and R ⊂ L(T ) be a B-stable closed subset. If A∗ → B, and for any α ∈ I
the algebra Aα can be represented by a structure µα ∈ R, then there is a structure λ ∈ R representing B.
The following table contains the classification and some invariants of four-dimensional BL algebras. It collects results from
[5, 22].
4. BINARY LIE ALGEBRAS OF DIMENSION 4
Pj=1
4
A
n3 ⊕ C
n4
r2 ⊕ C2
r2 ⊕ r2
sl2 ⊕ C
g1
multiplication table
e1e2 = e3
e1e2 = e3, e1e3 = e4
e1e2 = e2
e1e2 = e2, e3e4 = e4
e1e2 = e2, e1e3 = −e3, e2e3 = e1
e1e2 = e2, e1e3 = e3, e1e4 = e4
g2(β)
e1e2 = e2, e1e3 = e3, e1e4 = e3 + βe4
g3(β)
g4(α, β)
g5(α)
g6
e1e2 = e2, e1e3 = e3,
e1e4 = βe4, e2e3 = e4
e1e2 = e2, e1e3 = e2 + αe3,
e1e4 = e3 + βe4
e1e2 = e2, e1e3 = e2 + αe3,
e1e4 = (α + 1)e4, e2e3 = e4
e1e2 = e3, e3e4 = e3
Der(A) Z(A)
10
7
8
4
4
12
8
7
6
5
7
2
1
2
0
1
0
0
0
0
0
0
A2
1
2
1
2
3
3
3, β 6= 0;
2, β = 0
3
3, α 6= 0 6= β;
2, αβ = 0
3, α 6= 0
2, α = 0
1
type
Lie
Lie
Lie
Lie
Lie
Lie
Lie
Lie, for β = 2,
M alcev, for β = −1
BL, for β 6= −1, 2
Lie
Lie
BL
Table 4.1. Binary Lie algebras of dimension 4.
The algebra g4(α1, β1) is isomorphic to g4(α2, β2) iff the proportions 1 : α1 : β1 and 1 : α2 : β2 coincide after some
permutation. The algebra g5(α) is isomorphic to g5(β) iff αβ = 1 or α = β. Apart from these two exceptions, any two algebras
with different names from Table 4.1 are not isomorphic.
Theorem 3. The graph of primary degenerations for binary Lie algebras of dimension 4 has the following form:
r2 ⊕ r2
β = 0
g4(α, β)
α = β = 0
r2 ⊕ C2
g6
4
5
6
7
8
10
12
16
α = 0
α = −1
sl2 ⊕ C
β = α + 1
g5(α)
α = 1, β = 2
n4
α = 1
n3 ⊕ C
C4
g3(β)
g2(β)
β = 1
g1
Figure I. The graph of primary degenerations for four-dimensional binary Lie algebras.
Proof. Tables 4.2 and 4.3 placed below give the proofs for all primary degenerations and non-degenerations including non-Lie
algebras.
degenerations
g3(β) → g2(β) Et
g6 → r2 ⊕ C2
parametrized bases
1 = e1 + e2, Et
2 = te2, Et
Et
1 = e3, Et
2 = e4, Et
3 = (1 − β)e3 + e4, Et
4 = te2
3 = e1, Et
4 = e3 + e4
Table 4.2. Degenerations of binary Lie algebras of dimension 4.
non-degenerations
g3(β) ❍❍→ r2 ⊕ C2, g1(β 6= 1), g2(γ 6= β)
g6 ❍❍→ g1, g2(β)
cij (g3(β)) = (βi+2)(βj+2)
βi+j+2
arguments
, but cij(r2 ⊕ C2) = 1, cij(g1) = 3 and
cij (g2(γ)) = (γ i+2)(γ j+2)
γ i+j+2
dim (g6)2 < dim (g2(β))2 6 dim (g1)2
Table 4.3. Non-degenerations of binary Lie algebras of dimension 4.
Remark. Gorbatsevich classified all finite-dimensional anticommutative algebras of level 3 in [12]. It follows from Theorem
3 that his classification is not correct. Namely, there are binary Lie algebras of level 3 in dimension 4, which are not included in
the classification of Gorbatsevich.
Corollary 4. Irr(BL4) = {Ci}1≤i≤5, where
5
✷
C1 = O(sl2 ⊕ C) = O(cid:0){sl2 ⊕ C, g5(−1), g4(−1, 0), n4, n3 ⊕ C, C4}(cid:1) ,
C2 = O(r2 ⊕ r2) = O {r2 ⊕ r2, g5(0), g2(0), r2 ⊕ C2, n4, n3 ⊕ C, C4} ∪ [α∈C
C3 = [α∈C
C4 = [α,β∈C
C5 = [β∈C
{g4(α, 0)}! ,
O(g5(α)) = O [α∈C
{g5(α), g4(α, α + 1)} ∪ {g2(0), g2(2), g3(2), n4, n3 ⊕ C, C4}! ,
O(g4(α, β)) = O
[α,β∈C
O(g3(β)) = O
{g3(β), g2(β)} ∪ {g6, r2 ⊕ C2, n3 ⊕ C, g1, C4}
[β∈C
.
{g4(α, β), g2(β)} ∪ {r2 ⊕ C2, n4, n3 ⊕ C, g1, C4}
,
In particular, Rig(BL4) = Rig(Lie4) = {sl2 ⊕ C, r2 ⊕ r2}.
Proof. In view of Theorem 3 and the fact that Lie4 is a closed subset of BL4 it is enough to prove that
g5(∗) 6→ g4(α, β) (g4(α, β) 6∼= g4(γ, γ + 1) for any γ ∈ C), g5(∗) 6→ r2 ⊕ C2, g5(∗) 6→ g2(β) (β 6= 0, 2), g5(∗) 6→ g1,
g4(∗, ∗) 6→ g3(2), g3(∗) → g6, g3(∗) 6→ n4,
where g5(∗) = {g5(α)}α∈C, g4(∗, ∗) = {g4(α, β)}α,β∈C and g3(∗) = {g3(β)}β∈C. Let us define
A = hf1, f2, f3, f4i, hf3, f4i2 = 0, hf2, f3, f4i2 ⊂ hf4i, Ahf2, f3, f4i ⊂ hf2, f3, f4i, Ahf3, f4i ⊂ hf3, f4i,
Ahf4i ⊂ hf4i, c2
1,2 + c3
1,3 = c4
1,4, where fifj =
i,jfk for all 1 6 i, j 6 4
ck
One can take f1 = e1, f2 = e3, f3 = e2 and f4 = e4 and check that g5(α) ∈ R for all α ∈ C.
Let us prove that g2(β) 6∈ R if β 6= 0, 2. Assume that there is some basis fi (1 ≤ i ≤ 4) of V such that the structure constants
i,j of g2(β) in it satisfy all required conditions. Let U = h f2, f3, f4i and L : U → U be the operator of left multiplication by
ck
f1. It follows from the definition of R that the matrix of L in the basis f2, f3, f4 is lower triangular. Hence, c2
1,4 are
eigen values of L. On the other hand, it is easy to see that U = he2, e3, e4i and f1 = ce1 + v for some c ∈ C∗ and v ∈ U . Then
the eigen values of L are c, c and βc. Then we have c = (β + 1)c or βc = 2c, i.e. β = 0 or β = 2.
1,3 and c4
1,2, c3
Analogously one can prove that g4(α, β) 6∈ R if α − β 6= 1, α − β 6= −1 and α + β 6= 1, and r2 ⊕ C2, g1 6∈ R.
Since he2, e3, e4i is an abelian subalgebra of g4(α, β) and there is no three-dimensional abelian subagebra in g3(2), we have
g4(∗, ∗) 6→ g3(2) by Lemma 2. Let now define
A = hf1, f2, f3, f4i, hf2, f3, f4i2 ⊂ hf4i, Ahf2, f3, f4i ⊂ hf2, f3, f4i, Ahf3, f4i ⊂ hf3, f4i, Ahf4i ⊂ hf4i,
c2
1,2 = c3
1,3, c3
1,2 = 0, where fifj =
i,jfk for all 1 6 i, j 6 4
ck
4
Pk=1
One can take fi = ei (1 ≤ i ≤ 4) and check that g3(β) ∈ R for all β ∈ C. On the other hand, it is not hard to check that n4 6∈ R.
Finally, to prove that g3(∗) → g6 it is enough to take the parametrized basis
3 = e4, Et
2 = e3, Et
1 = e2, Et
4 = −te1
Et
and the parametrized index β(t) = 1
t .
Corollary 5. Irr(M al4) = {Ci}1≤i≤4 ∪ {C′
5}, where Ci (1 ≤ i ≤ 4) are the same as in Corollary 4, and
In particular, Rig(M al4) = {sl2 ⊕ C, r2 ⊕ r2, g3(−1)}.
C′
5 = O(g3(−1)) = O(cid:0){g3(−1), g2(−1), n3 ⊕ C, C4}(cid:1) .
Proof. Everything follows from Theorem 3, Corollary 4 and Table 4.1.
✷
✷
R =
A(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
R =
A(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
4
Pk=1
.
.
6
5. DEGENERATIONS OF NILPOTENT MALCEV ALGEBRAS OF DIMENSION 5
For five-dimensional nilpotent Malcev algebras we have the following table, which is constructed using results of [13] and [20].
A
n3 ⊕ C2
n4 ⊕ C
g5,1
g5,2
g5,3
g5,4
g5,5
g5,6
M5
multiplication table
e1e2 = e3
e1e2 = e3, e1e3 = e4
e1e2 = e5, e3e4 = e5
e1e2 = e4, e1e3 = e5
e1e2 = e3, e1e4 = e5, e2e3 = e5
e1e2 = e3, e1e3 = e4, e2e3 = e5
e1e2 = e3, e1e3 = e4, e1e4 = e5
e1e2 = e3, e1e3 = e4, e1e4 = e5, e2e3 = e5
e1e2 = e4, e3e4 = e5
Der(A)
16
12
11
15
10
10
9
8
9
Zl(A) A2 A3
0
2 + 13
1 + 124
1
0
0
1
2
2
2
0
15
25
135
235
1235
1235
135
1
2
1
2
2
3
3
3
2
Theorem 6. The graph of primary degenerations for nilpotent Malcev algebras of dimension 5 has the following form:
Table 5.1. Nilpotent Malcev algebras of dimension 5.
8
9
10
11
12
M5
g5,3
n4 ⊕ C
15
g5,2
16
n3 ⊕ C2
25
C5
g5,6
g5,5
g5,4
g5,1
Figure II. The graph of primary degenerations for five-dimensional nilpotent Malcev algebras.
Proof. It is enough to verify the assertions of the form M5 → A for such A that dim Der(A) < 9. So we have to check that
M5 → g5,3 and M5 6→ g5,4. The parametrized basis formed by Et
4 = t2e3 and
Et
5 = t2e5 gives the required degeneration. The assertion M5 6→ g5,4 follows from the fact that dim (g5,4)2 > dim (M5)2.
3 = te4 + te5, Et
1 = e1 − e4, Et
2 = te2 + te3, Et
Corollary 7. Irr(N M al5) = {C1, C2}, where C1 = O(g5,6) = N Lie5 and C2 = O(M5) = N M al5 \ {g5,6, g5,5, g5,4}. In
particular, Rig(N M al5) = {g5,6, M5}.
Proof. Since there is only finite number of isomorphism classes of five-dimensional nilpotent Malcev algebras, everything
follows from Theorem 6.
✷
✷
6. DEGENERATIONS OF NILPOTENT MALCEV ALGEBRAS OF DIMENSION 6
We use the table of invariants for nilpotent six-dimensional Lie algebras from [30] and classification of nilpotent six-
dimensional Malcev non-Lie algebras from [20] to construct the table containing important invariants for nilpotent six-
5 instead of g5,i ⊕ C and
dimensional Malcev algebras. To simplify the notation we write gi instead of g6,i, and gC
M5 ⊕ C respectively.
i and M C
7
multiplication table
Der(A)
M3
M4
M 0
5
M 1
5
M 0
6
6, ǫ 6= 0
M 0
7
M 1
7
M ǫ
e1e2 = e4, e1e3 = e5, e2e4 = e6, e2e5 = e6, e3e4 = −e6
e1e2 = e4, e1e3 = e5, e1e5 = e6, e3e4 = e6,
e1e2 = e4, e2e4 = e5, e3e4 = e6
e1e2 = e4, e1e3 = e5, e2e4 = e5, e3e4 = e6
e1e2 = e3, e1e3 = e5, e1e5 = e6, e3e4 = e6
e1e2 = e3, e1e3 = e5, e1e5 = e6, e2e4 = ǫe5, e3e4 = e6
e1e2 = e4, e1e4 = e5, e1e5 = e6, e2e3 = e5
e1e2 = e4, e1e4 = e5, e1e5 = e6, e2e3 = e5, e2e4 = e6
Table 6.1. Nilpotent Malcev algebras of dimension 6.
The algebra M ǫ
2 is isomorphic to M ǫ′
2 iff ǫǫ′ = 1 or ǫ = ǫ′. Apart from this exception any two algebras with different names
from Table 6.1 are not isomorphic.
A
g1
g2
g3
g5
g6
g7
g8
g9
g10
g14
g15
g16
g17
g18
g20
g21
g23
g24
gC
1
gC
2
gC
3
gC
4
gC
5
gC
6
n3 ⊕ n3
n4 ⊕ C2
n3 ⊕ C3
1
1
M C
5
M 0,1
M 1,0
M 1,1
1
M 0
2
M −1
2
M ǫ
2, ǫ 6= −1, 0
e1e2 = e3, e1e3 = e4, e1e4 = e6, e2e3 = e6, e2e5 = e6
e1e2 = e3, e1e3 = e4, e1e4 = e6, e2e5 = e6
e1e2 = e3, e1e3 = e6, e4e5 = e6
e1e2 = e3, e1e3 = e4, e1e4 = e5, e1e5 = e6, e2e3 = e5, e2e4 = e6
e1e2 = e3, e1e3 = e4, e1e4 = e5, e2e3 = e5, e2e5 = e6, e3e4 = −e6
e1e2 = e3, e1e3 = e4, e1e4 = e5, e1e5 = e6, e2e3 = e6
e1e2 = e3, e1e3 = e4, e2e5 = e6, e3e4 = −e6
e1e2 = e3, e1e3 = e4, e1e4 = e5, e1e5 = e6
e1e2 = e4, e1e3 = e5, e1e4 = e6, e3e5 = e6
e1e2 = e3, e1e3 = e4, e1e5 = e6, e2e3 = e5, e2e4 = e6
e1e2 = e3, e1e3 = e5, e1e4 = e6, e2e3 = e6
e1e2 = e3, e1e3 = e5, e1e4 = e5, e2e3 = e6
e1e2 = e3, e1e3 = e5, e1e4 = e6
e1e2 = e3, e1e3 = e5, e2e4 = e6
e1e2 = e3, e1e3 = e5, e1e4 = e6, e2e4 = e5
e1e2 = e5, e1e3 = e6, e3e4 = e5
e1e2 = e3, e1e3 = e4, e1e4 = e5, e2e3 = e6
e1e2 = e4, e1e3 = e5, e2e3 = e6
e1e2 = e5, e3e4 = e5
e1e2 = e4, e1e3 = e5
e1e2 = e3, e1e4 = e5, e2e3 = e5
e1e2 = e3, e1e3 = e4, e2e3 = e5
e1e2 = e3, e1e3 = e4, e1e4 = e5
e1e2 = e3, e1e3 = e4, e1e4 = e5, e2e3 = e5
e1e3 = e5, e2e4 = e6
e1e2 = e3, e1e3 = e4
e1e2 = e3
e1e2 = e5, e3e5 = e6
e1e2 = e5, e3e4 = e5, e3e5 = e6
e1e2 = e5, e1e4 = e6, e3e5 = e6
e1e2 = e5, e1e4 = e6, e3e4 = e5, e3e5 = e6
e1e2 = e4, e1e3 = e5, e2e5 = e6
e1e2 = e4, e1e3 = e5, e2e5 = e6, e3e4 = −e6
e1e2 = e4, e1e3 = e5, e2e5 = e6, e3e4 = ǫe6
Zl(A)
1346
1346
146
12346
12346
12346
12346
12346
136
1346
246
246
246
246
246
26
2346
36
1 + 15
1 + 25
1 + 135
1 + 235
1 + 1235
1 + 1235
13 + 13
2 + 124
3 + 13
1 + 135
126
136
126
246
136
136
136
136
246
246
1346
1236
1246
1246
A2 A3
2
3
2
3
2
1
3
4
3
4
3
4
3
2
3
4
1
3
3
4
3
2
2
3
1
3
1
3
3
1
0
2
3
4
0
3
1
0
0
2
1
2
2
3
3
2
2
3
0
2
1
2
1
0
1
2
1
2
1
2
2
1
1
3
1
3
1
3
3
1
1
3
2
3
2
3
3
2
2
3
2
3
3
2
11
12
14
9
8
10
9
11
12
10
13
12
15
13
14
17
11
16
21
19
15
15
13
12
16
17
24
14
13
12
11
12
13
11
11
13
11
10
10
10
11
10
8
Theorem 8. The graph of primary degenerations for nilpotent Malcev algebras of dimension 6 has the form presented in Figure
III.
Proof. Tables 6.2 and 6.3 presented below give the proofs for all primary degenerations and non-degenerations including non-Lie
algebras.
degenerations
M 1
7 → g1
M 1
7 → M 0
7
M 1
7 → M4
7 → M 1,1
M 1
1
M 1
5 → M 0
5
M ǫ
6 → g1
6 → M ǫ
M ǫ
2 (ǫ 6= 0)
6 → M 0
M ǫ
5 (ǫ = 0)
6 → M 1,0
M ǫ
(ǫ = 0)
1
M ǫ
6 → M4 (ǫ = 0)
M ǫ
6 → M3 (ǫ = −1)
M ǫ
2 → g10 (ǫ 6= −1, 0)
M ǫ
M ǫ
M ǫ
2 → g18 (ǫ = 0)
2 → M C
5 (ǫ = 0)
2 → g20 (ǫ = −1)
1
M 0
7 → gC
5
M 0
7 → g3
7 → M 0
M 0
2
7 → M 0,1
M 0
1
M 0
5 → g16
5 → M 0
M 0
2
1 → M 0,1
M 1,1
1 → M 1,0
M 1,1
1
M3 → g10
M3 → M −1
2
M 1,0
1 → g3
M 1,0
1 → M C
5
M 0,1
1 → g3
M 0,1
1 → M C
5
M4 → g20
M4 → g3
M4 → M C
5
M C
5 → gC
3
5 → n3 ⊕ n3
M C
Et
1 = e1 − e2
t − e3
Et
parametrized bases
t , Et
2 = e2, Et
5 = te3 + e4 + e5
3 = e4 + e5
t , Et
6 = e6
1 = e1, Et
Et
1 = e2, Et
2 = te2, Et
2 = −e3, Et
Et
Et
1 = e2, Et
Et
1 = e1, Et
2 = e3
t , Et
2 = e2, Et
3 = e3, Et
3 = −te1, Et
3 = e1, Et
3 = te3, Et
4 = te4, Et
4 = −e5, Et
4 = e4
t , Et
4 = e4, Et
Et
1 = e1 + e4
1 = e2, Et
Et
Et
1 = −e2, Et
1 = te1, Et
Et
1 = e1, Et
Et
2 = e2 + (ǫ+2)e3
2t , Et
2t − e4, Et
4 = e5 + e6
2t , Et
5 = te4, Et
Et
4 = ǫe5
2 = e4
3 = e1, Et
t , Et
t , Et
3 = − e4
t , Et
2 = e1, Et
3 = −e4, Et
2 = e2, Et
4 = e3, Et
4 = e5, Et
t , Et
Et
3 = e3
2 = e2, Et
t − e4
1 = −e2 − e3
2 = te1, Et
2 , Et
5 = te5 − te6
Et
2 = te1, Et
t + e2 − e3, Et
3 = te4, Et
6 = t2e6
2 , Et
3 = (ǫ+1)te3
4 = e4, Et
ǫ
Et
5 = e4 + ǫe5, Et
6 = (ǫ + 1)te6
Et
1 = ǫe1
t , Et
4 = e5,
5 = te5, Et
6 = te6
5 = te4, Et
6 = te6
,
4t2
6 = e6
t
6 = te6
t − (ǫ+2)e6
5 = e5
t , Et
5 = e5, Et
3 = e3 + e5
6 = e6
6 = e6
5 = −e3, Et
6 = e6
5 = e5, Et
t
5 = te3, Et
6 = te6
6 = e6
t , Et
4 = te3 + te5
2 ,
5 = e5
t
t
− e1, Et
4 = (ǫ + 1)te5,
Et
1 = e2 + e3, Et
2 = e1, Et
Et
1 = e1, Et
2 = e3, Et
Et
Et
Et
1 = te1, Et
2 = e3, Et
1 = e1 − e2, Et
2 = e2, Et
1 = e1, Et
t2 − e5
2 = e4
t2 , Et
2 = e1, Et
1 = e2, Et
Et
2 = − e2
Et
1 = te3, Et
2 = −e1 − e3 − e4, Et
1 = e1 + e2, Et
1 = te1, Et
Et
Et
1 = e1, Et
2 = e3 + e4, Et
2 = e2, Et
2 = e2, Et
3 = e3
3 = −e4 − e5, Et
4 = e4
3 = e2, Et
3 = e5, Et
3 = e4, Et
3 = e5
4 = te2 + e4, Et
4 = e5, Et
t − e6
t , Et
t , Et
4 = −e4, Et
t , Et
4 = e4
6 = te5
6 = te4
5 = e5, Et
5 = e6, Et
4 = te3, Et
t , Et
5 = −e6, Et
6 = e6
5 = −e6, Et
6 = te3
5 = e3, Et
4 = e2, Et
6 = e6
5 = e5
t , Et
5 = e5, Et
6 = te6
5 = t2e5, Et
4 = te3 + te4, Et
4 = e5, Et
6 = te6
5 = te4, Et
6 = e6
t
3 = te1, Et
t , Et
3 = te4 − te5, Et
3 = te2, Et
3 = e3, Et
3 = te3, Et
4 = te4, Et
4 = te4, Et
5 = te5, Et
5 = e5, Et
6 = te6
6 = te6
Et
1 = te2, Et
Et
6 = −te6
Et
1 = e1 + e2 + e3, Et
2 = t2e2 + t2e3, Et
3 = te2, Et
4 = t2e4 + t2e5, Et
5 = te4, Et
6 = t2e6
Et
1 = e1, Et
1 = e1 + e3, Et
Et
Et
1 = te3, Et
Et
1 = e1, Et
2 = e4
1 = e1, Et
Et
1 = e1
2 = te2, Et
2 = te2, Et
2 = e2, Et
t2 − e5
t2 , Et
2 = e2, Et
3 = e3, Et
3 = te5, Et
3 = e3, Et
3 = e5
3 = e3, Et
4 = te4, Et
4 = te3, Et
4 = te4, Et
5 = e5, Et
6 = te6
5 = −e4 + e5, Et
5 = e5, Et
6 = e6
5 = e2, Et
6 = e6
5 = e5, Et
6 = te6
6 = e6
t − e6
t , Et
4 = e1, Et
4 = te4, Et
Et
t , Et
Et
1 = e1 + e3, Et
1 = te1, Et
Et
2 = e3, Et
3 = e5
t , Et
4 = e2 + e4
2 = e2, Et
2 = e2
t , Et
2 = te2 + te3, Et
4 = te3, Et
4 = e5, Et
3 = e4, Et
3 = e3, Et
3 = te4 + te5, Et
t
t2 , Et
5 = e6
t2 , Et
5 = e4
t − e5
5 = e4, Et
4 = t2e3, Et
6 = e4
t , Et
6 = e6
5 = t2e5, Et
6 = e6
Et
1 = e1, Et
2 = e3, Et
3 = te2, Et
4 = e4 + e5, Et
5 = te5, Et
6 = e6
Et
1 = e1 − e4, Et
6 = e6
Table 6.2. Degenerations of nilpotent Malcev algebras of dimension 6.
non-degenerations
M 1
7 ❍❍→ g9, g23
M 1
7 ❍❍→ M ǫ
2 (ǫ 6= 0), M 0
5
M 1
5 ❍❍→ (cid:26) M ǫ
1 , M4, g3
M 0,1
5 ❍❍→ gC
5
M 1
6 ❍❍→ g9, g23
(ǫ = 0)
M ǫ
6 ❍❍→ M 0,1
1
M ǫ
(cid:27)
M ǫ
6 ❍❍→ (cid:26) M ǫ′
2 (M ǫ′
M C
2 6∼= M ǫ
2),
5 (ǫ 6= 0)
(cid:27)
M ǫ
2 ❍❍→ gC
5 , gC
4
M ǫ
2 ❍❍→ n3 ⊕ n3(ǫ = −1)
M 0
7 ❍❍→ g10, M 1,0
1 , M4, gC
4
M 1,1
1 ❍❍→ M4, g17, g24
M3 ❍❍→ gC
M 1,0
5 , gC
4
1 ❍❍→ M 0,1
1
M4 ❍❍→ gC
4
arguments
R =(cid:26)A(cid:12)(cid:12)(cid:12)(cid:12)
M 1
dim (g9)2 = dim (g23)2 > dim (M 1
A = hf1, f2, f3, f4, f5, f6i, A2 ⊂ hf4, f5, f6i,
hf3, f4, f5, f6i2 = 0, Ahf3, f4, f5, f6i ⊂ hf5, f6i (cid:27)
7 )2
7 ∈ R (take fi = ei for 1 6 i 6 6), but M ǫ
5 6∈ R
2, M 0
7 ) = dim Z(M 1,0
1 )
= dim Z(M 0,1
1 ) = dim Z(M4) = dim Z(g3) < dim Z(M 1
5 )
dim Z2(gC
5 ) < dim Z2(M 1
5 )
dim (g9)2 = dim (g23)2 > dim (M ǫ
6)2
dim Z2(M 0,1
1 ) < dim Z2(M 0
6 )
A = hf1, f2, f3, f4, f5, f6i, A2 ⊂ hf4, f5, f6i,
hf3, f4, f5, f6i2 = 0,
x(yz) = ǫy(xz)∀x ∈ A, y, z ∈ hf2, f3, f4, f5, f6i
6 ∈ R (take f1 = e1, f2 = e4, f3 = e2, f4 = e3,f5 = e5
and f6 = e6), but M ǫ′
5 6∈ R
2 6∈ R and, if ǫ 6= 0, then M C
5 )3 = dim (gC
dim (gC
A = hf1, f2, f3, f4, f5, f6i, A2 ⊂ hf4, f5, f6i,
4 )3 > dim (M ǫ
2)3
hf3, f4, f5, f6i2 = 0, Ahf4, f5, f6i ⊂ hf6i,
hf2, f3, f4, f5, f6ihf5, f6i = 0,
hf2, f3, f4, f5, f6ihf3, f4, f5, f6i ⊂ hf5, f6i,
1,5 + c6
c5
2,3c6
1,3,
i,jfk for all 1 6 i, j 6 6
where fifj =
ck
1,3 = 0, c5
2,4c4
6
1,4 = c6
2,4c5
2,3c6
Pk=1
2 ∈ R (take f1 = e2, f2 = e3, f3 = e1, f4 = e4,f5 = e5
and f6 = e6), but n3 ⊕ n3 6∈ R
A = hf1, f2, f3, f4, f5, f6i, A2 ⊂ hf4, f5, f6i,
hf2, f3, f4, f5, f6ihf4, f5, f6i = 0
7 ∈ R (take fi = ei for 1 6 i 6 6), but g10, M 1,0
1 , M4, gC
dim (M4)2 = dim (g17)2 = dim (g24)2 > dim (M 1,1
1 )2
4 6∈ R
dim (gC
5 )3 = dim (gC
4 )3 > dim (M3)3
dim Z2(M 0,1
dim (gC
1 ) < dim Z2(M 1,0
1 )
4 )3 > dim (M4)3
R =
M ǫ
A(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
R =(cid:26)A(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)
M −1
A
R =
M 0
9
✷
(cid:27)
2 (ǫ 6= 0), M 0
7 , M 1,0
1 ,
dim Z(M ǫ
2) = dim Z(M 0
Table 6.3. Non-degenerations of nilpotent Malcev algebras of dimension 6.
10
8
9
g6
g5
g8
10
g7
g14
ǫ = 1
M 1
7
M 1
5
M ǫ
6
ǫ = 1
ǫ = 1
ǫ = 0
ǫ = −1
11
g9
g23
g1
M ǫ
2
M 0
7
M 0
5
M
1,1
1
ǫ = 0
M3
ǫ = 0
12
g
C
6
g16
g2
g10
M 0
2
M
1,0
1
13
14
15
16
17
18
19
21
24
36
g
C
5
g15
g18
M
−1
2
M
0,1
1
M4
g20
g3
M
C
5
g17
g
C
4
g
C
3
g21
n4 ⊕ C2
g
C
1
n3 ⊕ n3
g24
g
C
2
n3 ⊕ C3
C6
Figure III. The graph of primary degenerations for six-dimensional nilpotent Malcev algebras.
Corollary 9. N M al6 = {C1, C2}, where C1 = O(g6) = N Lie6 and C2 = Sǫ∈C
In particular, Rig(N M al6) = Rig(N Lie6) = {g6}.
Proof. In view of Theorem 8 it is enough to prove that M ∗
{M ǫ
6 6→ g9, M ∗
5 , where M ∗
6 → M 1
6}ǫ∈C. The first two assertions follow from the fact that dim (g9)2 = dim (g23)2 > dim A2 for any A ∈ M ∗
6 .
To prove that M ∗
7 one can choose the parametrized basis
6 6→ g23, M ∗
7 and M ∗
6 → M 1
6 → M 1
6 =
O(M ǫ
6) = N M al6 \ {g6, g5, g8, g7, g14, g9, g23}
11
1 = e1, Et
Et
and the parametried index ǫ(t) = 1
t .
To prove that M ∗
6 → M 1
5 one can choose the parametrized basis
2 = e2 − e4, Et
3 = te4, Et
4 = e3, Et
5 = e5, Et
6 = e6
Et
1 = e2, Et
2 = te1, Et
3 = e4, Et
4 = −te3, Et
5 = −t2e5, Et
6 = te6
and the parametrized index ǫ(t) = −t2.
Acknowledgements. The authors are grateful to Prof. Dietrich Burde for some constructive comments.
✷
REFERENCES
[1] Albeverio S., Omirov B., Rakhimov I., Varieties of nilpotent complex Leibniz algebras of dimension less than five, Comm. Algebra, 33 (2005), 5, 1575–
1585.
[2] Benes T., Burde D., Degenerations of pre-Lie algebras, J. Math. Phys., 50 (2009), 11, 112102.
[3] Benes T., Burde D., Classification of orbit closures in the variety of three-dimensional Novikov algebras, J. Alg. Appl., 13 (2014), 2, 1350081.
[4] Burde D., Degenerations of nilpotent Lie algebras, J. Lie Theory, 9 (1999), 1, 193–202.
[5] Burde D., Sur les degenerations d'algebres de Lie, https://homepage.univie.ac.at/Dietrich.Burde/papers/burde 15 rapp deg.pdf (2003).
[6] Burde D., Degenerations of 7-dimensional nilpotent Lie algebras, Comm. Algebra, 33 (2005), 4, 1259–1277.
[7] Burde D., Steinhoff C., Classification of orbit closures of 4–dimensional complex Lie algebras, J. Algebra, 214 (1999), 2, 729–739.
[8] Casas J., Khudoyberdiyev A., Ladra M., Omirov B., On the degenerations of solvable Leibniz algebras, Lin. Alg. Appl., 439 (2013), 2, 472–487
[9] Gabriel P., Finite representation type is open, in: Proceedings of the International Conference on Representations of Algebras, Carleton University, Ottawa,
Ontario, 1974, in: Lecture Notes in Math., vol. 488, 1975, pp. 132–155.
[10] Gainov A., Binary Lie algebras of lower ranks (Russian), Algebra i Logika Sem., 2 (1963), 4, 21–40.
[11] Gainov A., Identical relations for binary Lie rings (Russian), Uspehi Mat. Nauk N.S., 12 (1957), 3 (75), 141–146.
[12] Gorbatsevich V., Anticommutative finite-dimensional algebras of the first three levels of complexity, St. Petersburg Math. J., 5 (1994), 3, 505–521.
[13] Grunewald F., OHalloran J., Varieties of nilpotent Lie algebras of dimension less than six, J. Algebra, 112 (1988), 315–325.
[14] Grunewald F., OHalloran J., A Characterization of Orbit Closure and Applications, J. Algebra, 116 (1988), 163–175.
[15] Filippov V., On δ-derivations of prime alternative and Malcev algebras, Algebra and Logic, 39 (2000), 5, 618–625.
[16] Kashuba I., Martin M., Deformations of Jordan algebras of dimension four, J. Algebra, 399 (2014), 277–289.
[17] Kashuba I., Martin M., The variety of three-dimensional real Jordan algebras, J. Alg. Appl., 15 (2016), 8, 1650158.
[18] Kaygorodov I., On (n + 1)-ary derivations of simple n-ary Malcev algebras, St. Petersburg Math. J., 25 (2014), 4, 575–585
[19] Kaygorodov I., Popov Yu., A characterization of nilpotent nonassociative algebras by invertible Leibniz-derivations, J. Algebra, 456 (2016), 323–347.
[20] Kuzmin E., Malcev algebras of dimension five over a field of characteristic zero, Algebra and Logic, 9 (1970), 416–421.
[21] Kuzmin E., The connection between Malcev algebras and analytic Moufang loops, Algebra and Logic, 10 (1971), 3–22.
[22] Kuzmin E., Binary Lie algebras of small dimension, Algebra and Logic, 37 (1998), 3, 181–186.
[23] Kuzmin E., Structure and representations of finite dimensional Malcev algebras, Quasigroups Related Systems, 22 (2014), 1, 97–132.
[24] Malcev A., Analytic loops (Russian), Mat. Sb. N.S., 36, (1955), 569–576.
[25] Mazzola G., The algebraic and geometric classification of associative algebras of dimension five, Manuscripta Math., 27 (1979), 81–101.
[26] Mazzola G., Generic finite schemes and Hochschild cocycles, Comment. Math. Helv. 55 (1980), 267–293.
[27] Pozhidaev A., n-ary Malcev algebras, Algebra and Logic, 40 (2001), 3, 170–182.
[28] Rakhimov I., On the degenerations of finite dimensional nilpotent complex Leibniz algebras, J. Math. Sci. (N.Y.), 136 (2006), 3, 3980–3983.
[29] Sagle A., Malcev algebras, Trans. Amer. Math. Soc., 101 (1961), 426–458.
[30] Seeley C., Degenerations of 6-dimensional nilpotent Lie algebras over C, Comm. Algebra, 18 (1990), 3493–3505.
|
1903.00057 | 1 | 1903 | 2019-02-28T20:29:55 | On simple Lie 2-algebra of toral rank 3 | [
"math.RA"
] | Simple Lie algebras of finite dimension over an algebraically closed field of characteristic 0 or $p> 3$ were recently classified. However, the problem over an algebraically closed field of characteristics 2 or 3 there exist only partial results. The first result on the problem of classification of simple Lie algebra of finite dimension over an algebraically closed field of characteristic 2 is that these algebras have absolute toral rank greater than or equal to 2. In this paper we show that there are not simple Lie 2-algebras with toral rank 3 over an algebraically closed field of characteristic 2 and dimension less or equal to 16. | math.RA | math |
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
Abstract. Simple Lie algebras of finite dimension over an algebraically closed
field of characteristic 0 or p > 3 were recently classified. However, the problem
over an algebraically closed field of characteristics 2 or 3 there exist only partial
results. The first result on the problem of classification of simple Lie algebra
of finite dimension over an algebraically closed field of characteristic 2 is that
these algebras have absolute toral rank greater than or equal to 2.
In this
paper we show that there are not simple Lie 2-algebras with toral rank 3 over
an algebraically closed field of characteristic 2 and dimension less or equal to
16.
Introduction
The simple Lie algebras of finite dimension over an algebraically closed field of
characteristic zero were first classified by Killing (1888) and Cartan (1894). These
algebras fall in four infinite families, An, Bn, Cn and Dn, and five exceptional cases,
E6, E7, E8, G2 and F4 (see [5]).
Simple finite-dimensional Lie algebras over an algebraically closed field of prime
characteristic p > 7 were classified by H. Strade, R. Block and R. L. Wilson in
the middle of years 90 (see [1], [2], [3], [14], [16], [17], [18], [19] and [21]).
In a
series of papers, H. Strade and A. Premet classified the finite-dimensional simple
Lie algebras over an algebraically closed field of characteristic p = 5 and p = 7 in
the beginning of this century (see [7], [8], [9], [10], [11] and [12]). It asserts that
every simple Lie algebra over an algebraically closed field of characteristic p > 3 is
either classical, or of Cartan, or Melikian type.
After the classification of simple Lie algebras, of finite dimension, over a field of
characteristic p > 3, the main problem still open in the category of Lie algebras
of finite dimension is the classification of simple Lie algebras on an algebraically
closed field of characteristic p = 2 and p = 3. In particular for p = 2 many new
phenomena arise (for instances, simple Lie algebras in characteristic zero are not
necessarily simple in characteristic two) and the classification will differ significantly
from those in characteristic 0 and p > 3. The first results for the classification
problem in characteristic 2 were made by S. Skryabin in [6].
In that work, S.
Skryabin proved that all finite dimensional Lie algebra of absolute toral rank 1 over
an algebraically closed field K of characteristic 2 is solvable, or equivalently, all finite
dimensional simple Lie algebra over an algebraically closed field of characteristic 2
has absolute toral rank of at least 2.
Date: February 7, 2019.
2010 Mathematics Subject Classification. 17B50, 17B20, 17B22 .
Key words and phrases. simple Lie 2-algebra, toral rank, Cartan decomposition.
1
2
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
In [13], Section 6, A. Premet and H. Strade present the following problem which
is wide open.
Problem 1. Classify all finite dimensional simple Lie algebras of absolute toral
rank two over an algebraically closed field of characteristics 2 and 3.
Strong results closely related to this problem were obtained by A. Grishkov and
A. Premet in [4] (work in progress). They annouced the following result: All finite
dimensional simple Lie algebra over an algebraically closed field of characteristic 2
of absolute toral rank 2 are classical of dimesion 3, 8, 14 or 26. In particular, all
finite dimensional simple Lie 2-algebra over a field of characteristic 2 of (relative)
toral rank 2 is isomorphic to A2, G2 or D4.
The case when the absolute toral rank or relative toral rank is greater than
or equal to 3 is much more difficult. For these cases, we propose the followings
problems, which are still open:
Problem 2. Classify all finite dimensional simple Lie algebras of absolute toral
rank greater than or equal to 3 over an algebraically closed field of characteristic 2.
Problem 3. Classify all finite dimensional simple Lie 2-algebras of relative toral
rank greater than or equal to 3 over an algebraically closed field of characteristic 2.
In this paper we will give a partial result for Problem 3 when the relative toral
rank is 3. Specifically, we show that there are not simple Lie 2-algebras of dimension
less than or equal to 16 with (relative) toral rank 3 over an algebraically closed field
of characteristic 2.
In the next section we will present some basic definitions and well-known results
that will be used throughout the work. In Section 2, we will prove that there are
no simple Lie 2-algebras with toral rank 3 and of dimension less than or equal to
6. In Section 3, we find the Cartan decomposition of a Lie 2-algebra of dimension
greater than or equal to 7 with respect to a maximal subalgebra of dimension 3. This
decomposition will give us the necessary tools for the classification of these algebras
and will be used for the rest of the work. The study of the Cartan decomposition
of these algebras with respect to a toral subalgebra of dimension 3 when the root
spaces have cardinality less than 7 will allow us to conclude in Section 4 that there
are no simple Lie 2-algebras of toral rank 3 with dimension between 7 and 9. When
the root space is equal to the dual of the toral subalgebra, the dimension of the
algebra is greater than or equal to 10. Using this fact and Theorem 5.4, in which
we classify the algebras in whose decomposition of Cartan the root spaces have
different dimensions, we will prove in the last section that there are no simple Lie
2-algebra of dimension between 10 and 16.
1. Preliminaries
Throughout this paper all the algebras are finite-dimensional and are defined
over a fixed algebraically closed field K of characteristic 2 containing the prime
field F2. We will start presenting some basic definitions and well-known facts.
Definition 1.1. A Lie 2-algebra is a pair (g, [2]), where g is a Lie algebra over K,
and [2] : g → g, a 7→ a[2] is a map (called 2-map) such that:
(1) (a + b)[2] = a[2] + b[2] + [a, b], ∀a, b ∈ g.
(2) (λa)[2] = λ2a[2], ∀λ ∈ K, ∀a ∈ g.
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
3
(3) ad(b[2]) = (ad(b))2, ∀b ∈ g.
If the 2-map [2] exists, it is unique for any Lie 2-algebra (g, [2]) with z(g) = 0.
A Lie 2-algebra (g, [2]) is called simple if g is a simple Lie algebra on K.
Example 1.2. Set
o3(K) := {A = (aij ) ∈ M3(K) : aii = 0 and aij = aji}.
The canonical basis of o3(K) is given by: e1 := e12 + e21, e2 := e13 + e31, and
e3 := e23 + e32. Then o3(K) = Ke1 ⊕ Ke2 ⊕ Ke3, with [e1, e2] = e3,
[e1, e3] =
e2, [e3, e1] = e2, is the only (up to isomorphism) simple Lie algebra of dimension 3
on K. However, it is not a Lie 2-algebra.
Definition 1.3. Let (g, [2]) be a Lie 2-algebra over K. An element x ∈ g is called
semisimple (respectively, 2-nilpotent ) if x lies in the 2-subalgebra of g generated by
x, that is x ∈ span{x, x[2], x[2]2
, ...} (respectively, if x[2]n
= 0 for some n ∈ N).
It is well-known that for any x ∈ g there are unique elements xs and xn in g
such that xs is semisimple, xn is 2-nilpotent, and x = xs + xn with [xs, xn] = 0
(Jordan-Chevalley-Seligman decomposition).
Definition 1.4. A torus in g is an abelian subalgebra t for which the 2-mapping
is one-to-one.
For a torus t there is a basis {t1, ..., tn} such that t[2]
i = ti. The elements satisfying
t = t[2] will be called toroidal elements. A torus t1 of g is called maximal if the
inclusion t1 ⊆ t2, with t2 toral, implies t1 = t2.
Definition 1.5. (H. Strade [15]). The (relative) toral rank of a Lie 2-algebra (g, [2])
is given by
M T (g) := max{dimK(t) : t is a torus in g}.
For instances, the centralizer cg(t) of any maximal torus in g is a Cartan subal-
gebra of g and, conversely, the semisimple elements of any Cartan subalgebra of g
lie in its center and form a maximal torus in g.
Let t be a maximal torus of (g, [2]), h := cg(t), and let V be a finite dimen-
sional (g, [2])-module (this means that ρV (x[2]) = ρV (x)2, ∀ x ∈ g where ρV : g →
EndK(V ) denotes the corresponding 2-representation). Since t is abelian, then
ρV (t) is abelian and consists of semisimple elements. Therefore, V can be decom-
posed into weight spaces respect to t as
V = ⊕
λ∈t∗
Vλ,
where Vλ := {v ∈ V : ρ(t)(v) := v · t = λ(t) v, ∀t ∈ t}.
The subset ∆ := {λ ∈ t∗ : Vλ 6= 0} ⊆ t∗ is called the t-weight of V .
Lemma 1.6. If t is a toral element of t, then λ(t) ∈ F2, for any λ ∈ ∆.
Proof. If v ∈ V \ {0}, then
λ(t)v = v · t = v · t[2] = (v · t) · t = (λ(t)v) · t = λ(t)(v · t) = λ(t)2v.
So, λ(t) ∈ F2.
(cid:3)
4
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
If V = g is the adjoint (g, [2])-module, then ∆ = {λ ∈ t∗ : gλ 6= 0} is nothing
but the set of roots of g respect to t, and
g = h ⊕(cid:18) ⊕
α∈∆
gα(cid:19)
where
gα := {g ∈ g : ad(h)(g) = α(h)g, h ∈ h}
is the root space decomposition.
2. Simple Lie 2-algebra with toral rank 3 and dimK ≤ 6
We will obtain in Proposition 2.1 that there are no simple Lie 2-algebras of
dimK ≤ 6 with toral rank 3. Next section will be spend to find the Cartan decom-
position for dimK ≥ 7, which will be very useful throughout the work.
Suppose that there exists a simple Lie 2-algebra (g, [2]) of dimK(g) ≤ 6.
In
particular, g is a simple Lie algebra on K of dimK(g) ≤ 6 and g is not isomorphic
to the Witt algebra W (1; 1) (otherwise, W (1; 1) is simple over K, which is absurd).
Then, by [20], Theorem 2.2, we have that g is of dimension 3 and so g is isomorphic
to o3(K). However o3(K) is not a Lie 2-algebra. This facts proves that:
Proposition 2.1. There are no simple Lie 2-algebras of dimK ≤ 6 with toral rank
3.
3. Cartan decomposition of a Lie 2-algebra of rank toral 3.
As we said in Section 2, in this section we will suppose that (g, [2]) is a Lie
2-algebra with M T (g) = 3 and we will find its Cartan decomposition. We have g
contains a maximal torus t of dimK(t) = 3. Since ad(t) is abelian and consists of
semisimple elements, g can be decomposed into weights spaces respect to t, that is
g = cg(t) ⊕(cid:18) ⊕
λ∈ t∗\{0}
gλ(cid:19) where gλ = {v ∈ g : [t, v] = λ(t)v, ∀t ∈ t}.
Now, since t is a torus of dimension 3, there is a basis {ti ∈ t : t[2]
i = ti, i = 1, 2, 3}
of t such that λ(ti) ∈ F2, ∀λ ∈ t∗ (Lemma 1.6). Hence, for any λ : t → F2, writing
λ = (λ(t1), λ(t2), λ(t3)), we have
t∗ = {(0, 0, 0), (1, 0, 0), (0, 1, 0), (0, 0, 1), (1, 1, 0), (1, 0, 1), (0, 1, 1), (1, 1, 1)}.
Set α = (1, 0, 0), β = (1, 0, 0), γ = (0, 0, 1) and h = cg(t). Thus {α, β, γ} is
a basis of t∗ and h is a Cartan subalgebra of g. Thus, h = t ⊕ n, where n is a
2-nilpotent subalgebra of g and [t, n] = 0. Therefore:
Theorem 3.1. The decomposition into weights spaces of (g, [2]) with respect to t
is:
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ ⊕ gα+β+γ
or
g = t ⊕ n ⊕(cid:18) ⊕
ξ∈G
gξ(cid:19) ,
where G := hα, β, γi is an elementary abelian group of order 8 and n is a 2-nilpotent
subalgebra of g.
Using the decomposition in Theorem 3.1, we show in the next proposition that
there exist at least three non-zero weight spaces. This fact leads us to conclude
that the dimension of g must be greater than or equal to 6, since the dimension of
h is greater than or equal to 3.
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
5
Proposition 3.2. There are ξ, δ, θ in t∗ linearly independent such that gξ 6= 0,
gδ 6= 0 and gθ 6= 0.
Proof. Suppose the proposition is false. We have two cases:
Case 1: There are ξ, δ and t∗ linearly independent such that gξ 6= 0 and gδ 6= 0.
In this case, we can extend {ξ, δ} to a basis of t∗. Thus there is λ ∈ t∗\{0} such
that {ξ, δ, λ} ⊆ t∗\{0} is a linearly independent set and gλ = 0. By the action of
the linear group GL3(F2) on t∗ we can consider the following change of basis given
by
GL3(F2) × t∗
(A, ξ)
(A, δ)
(A, λ)
ψA−−−→ t∗
7→ A · ξ := α
7→ A · δ := β
7→ A · λ := γ
Thus, we find that g = h ⊕ gα ⊕ gβ ⊕ gα+β. Consequently, up to change of basis, g
can be written as
g = h ⊕ gε ⊕ gµ ⊕ gε+µ,
where {ε, µ, µ + ε} is a linearly dependent set of t∗. Hence, g has total range 2,
which is a contradiction.
Case 2: There is ξ ∈ t∗ such that gξ 6= 0.
We extend {ξ} ⊆ t∗\{0} to a basis of t∗, so for π, κ ∈ t∗ \ {0}, {ξ, π, κ} is a
linearly independent set such that gξ 6= 0 and gπ = gκ = 0. Then, by the action of
GL3(F2) on t∗, we can make a change of basis, that is, for A ∈ GL3(F2) fixed, we
have
GL3(F2) × t∗
(A, ξ)
(A, π)
(A, κ)
ψA−−−→ t∗
7→ A.ξ := α
7→ A.π := β
7→ A.κ := γ.
Therefore, up to change of basis, g can be written as g = h ⊕ gε for some
ε ∈ t∗ \ {0}. In this case, we have g has toral rank 1, which is a contradiction. (cid:3)
It follows from Proposition 3.2 that:
Corollary 3.3. dimK(g) ≥ 6.
Proof. Since dimK(h) ≥ 3 and by Proposition 3.2 there exist θ, δ and ξ in t∗ linearly
independent such that dimK(gξ) ≥ 1, dimK(gδ) ≥ 1 and dimK(gθ) ≥ 1, we have
dimK(g) ≥ 6.
(cid:3)
Remark 3.4. By Theorem 3.1, Proposition 3.2 and Corollary 3.3, from now on we
will assume, without loss of generality, that dimK(g) ≥ 6 and the Cartan decom-
position into weight spaces of g is
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ ⊕ gα+β+γ,
where {α, β, γ} is a basis of t∗ such that gα 6= 0, gβ 6= 0, and gγ 6= 0, with
dimK(gα) ≥ dimK(gβ) ≥ dimK(gγ) ≥ dimK(gξ) > 0,
∀ ξ ∈ t∗\{α, β, α + β}.
Let ∆ := {λ ∈ t∗ : gλ 6= 0} be the root system of g with respect to t. It follows
from Theorem 3.1 that α, β and γ are elements of ∆. The following are all the
possibilities for ∆ depending on its cardinality.
1. If Card(∆) = 3, we have ∆1 = {α, β, γ}.
6
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
2. If Card(∆) = 4, we have the following possibilities for ∆:
a. ∆2 = {α, β, γ, α + β}
b. ∆3 = {α, β, γ, α + γ}
c. ∆4 = {α, β, γ, β + γ}
d. ∆5 = {α, β, γ, α + β + γ}
3. If Card(∆) = 5, we have the following possibilities for ∆:
a. ∆6 = {α, β, γ, α + β, α + γ}
b. ∆7 = {α, β, γ, α + β, β + γ}
c. ∆8 = {α, β, γ, α + β, α + β + γ}
d. ∆9 = {α, β, γ, α + γ, β + γ}
e. ∆10 = {α, β, γ, α + γ, α + β + γ}
f. ∆11 = {α, β, γ, β + γ, α + β + γ}
4. If Card(∆) = 6, we have the following possibilities for ∆:
a. ∆12 = {α, β, γ, α + β, α + γ, β + γ}
b. ∆13 = {α, β, γ, α + β, α + γ, α + β + γ}
c. ∆14 = {α, β, γ, α + β, β + γ, α + β + γ}
d. ∆15 = {α, β, γ, α + γ, β + γ, α + β + γ}
5. If Card(∆) = 7, then ∆0 = t∗\{0}.
For 0 ≤ i ≤ 15, we will denote by (g∆i, [2]) the Lie 2-algebra with its associated
root space ∆i in the previous list.
Using the above facts we can conclude the following theorem:
Theorem 3.5. For each 0 ≤ i ≤ 15, we have the Cartan decompositions of (g∆i , [2])
with respect to h is given by
(3.1)
g∆i = t ⊕ n ⊕(cid:18) ⊕
ξ∈∆i
gξ(cid:19) .
4. Analysis of g = h ⊕ (⊕ξ∈∆i gξ), with Card(∆i) < 7.
In this section we will show that for each ∆i, 1 ≤ i ≤ 15, the Cartan decompo-
gξ), associated with each system of root ∆i, generate some
sition g∆i = h ⊕ ( ⊕
ξ∈∆i
contradictions. These incompatibilities allow us to conclude that there is not a
simple Lie 2-algebra of toral rank 3 with these structures (see Proposition 4.3).
The case ∆0 = t∗\{0} corresponding to i = 0 will be studied in the next section.
From now on we suppose that (g, [2]) is a simple Lie 2-algebra with M T (g) = 3
and dimK(g) ≥ 7 (since there is no simple Lie 2-algebra with M T (g) = 3 and
dimK(g) = 6). Furthermore, we will assume that h is a Cartan subalgebra of g of
maximal rank and, therefore, h ⊕ gξ is solvable for all ξ ∈ G = hα, β, γi , where
G = hα, β, γi is the elementary abelian group of order 8.
In order to prove Proposition 4.3, we show the next lemmas:
Lemma 4.1. If dimK(gξ) = 1 for ξ ∈ ∆ \ {0}, then [n, gξ] = 0.
Proof. Set gξ = span{eξ} and take n ∈ n. Thus
y = ad(n)(eξ) = [n, eξ] ∈ [n, gξ] ⊂ gξ.
Therefore ad(n)(eξ) = λeξ for some λ ∈ K. Since ad(n) is a nilpotent operator, we
have λ = 0 and therefore y = 0, which proves the lemma.
(cid:3)
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
7
Lemma 4.2. If g = h ⊕(cid:18) ⊕
ξ∈∆i
gξ(cid:19), then [gξ, gξ] ⊆ Ker(ξ) for all ξ ∈ ∆.
Proof. Let ξ ∈ ∆ be fixed and set Sξ := h ⊕ gξ which is a soluble subalgebra of
g. Suppose that there is h ∈ [gξ, gξ] such that ξ(h) 6= 0. As gξ ⊆ Sξ, we have
(1),
[gξ, gξ] ⊆ [Sξ, Sξ] = Sξ
(1) is an ideal of Sξ. Now, take v ∈ gξ. Given that ξ(h) 6= 0, we have
since Sξ
v = 1
(1). Thus [h, Sξ] ⊆ [Sξ
(1). Therefore gξ ⊆ Sξ
(1), hence h ∈ Sξ
(1), Sξ] ⊆ Sξ
(1) and so
ξ(h) [v, h] ∈ [h, Sξ] ⊆ Sξ
Next, suppose that h ∈ Sξ
have [h, Sξ] ⊆ Sξ
(m) and hence gξ ⊆ Sξ
h ∈ [gξ, gξ] ⊆ [Sξ
(1), Sξ
(2).
(m) for some m ∈ N. As Sξ
(1)] = Sξ
h ∈ [gξ, gξ] ⊆ [Sξ
(m) (since ξ(t) 6= 0). Thus
(m), Sξ
(m)] = Sξ
(m+1)
(m) is an ideal of Sξ, we
(m+1). Then by induction, h ∈ Sξ
(m) for all m, and since Sξ is
and so h ∈ Sξ
solvable, we have h = 0. This fact implies that ξ(h) = 0, which is a contradiction.
Therefore ξ([gξ, gξ]) = 0.
(cid:3)
Proposition 4.3. There are no simple Lie 2-algebras of finite dimensional of toral
rank 3, with Cartan decomposition g∆i = h⊕( ⊕
gξ), for 1 ≤ i ≤ 15. In particular,
ξ∈∆i
there are no simple Lie 2-algebras g with 7 ≤ dimK(g) ≤ 9 and M T (g) = 3.
Proof. If dimK(gξ) = 1 for all ξ ∈ ∆i, with 1 ≤ i ≤ 15, we have that
0 6= I := n ⊕ ( ⊕
ξ∈∆i
gξ)
is a non-trivial ideal of g, since [n, gξ] = 0 (by Lemma 4.1) and [gξ, gξ] = 0. This
fact is a contradiction, because g is a simple Lie algebra. So, we can assume that
dimK(gξ) ≥ 2 for all ξ ∈ ∆i.
We will suppose, by contradiction, that there exist simple Lie 2-algebras of toral
rank 3, with Cartan decomposition given in (3.1) and we will get a contradiction.
Case 1: ∆1 = {α, β, γ}.
In this case we have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ.
i , eξ
ij ∈ n such that xξ = tξ
First we show that [gξ, gξ] ⊆ n, for all ξ ∈ ∆1. Indeed, let eξ
the basis of gξ. Thus xξ := [eξ
ij ∈ t and nξ
tξ
By Lemma 4.2 we have 0 = ξ(xξ) = ξ(tξ
ij ) = β(tβ
ξ(tξ
for all ξ ∈ ∆1. The simplicity of g, implies that h = Pξ∈∆1[gξ, gξ], so h ⊆ n.
j be elements in
j ] ∈ [gξ, gξ] ⊆ h = n ⊕ t and therefore there are
ij)t3 + nξ
ij.
ij) for all ξ ∈ ∆1. Hence 0 = ξ(xξ) =
ij ∈ n
Therefore h = n and t = 0, which is a contradiction.
ij) = 0 and therefore xξ = nξ
ij + nξ
ij ) = γ(tγ
ij ). Then α(tα
ij )t1 + β(tξ
ij )t2 + γ(tξ
ij) = ξ(tξ
ij = α(tξ
i and eξ
ij + nξ
ij + nξ
Case 2.a: ∆2 = {α, β, γ, α + β}.
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β.
i and eξ
Let eξ
ij ∈ t and nξ
tξ
j be elements of the base of gξ. Then xξ := [eξ
ij ∈ n such that xξ = tξ
ij + nξ
i , eξ
j ] ∈ h and there are
0 = ξ(xξ) = ξ(tξ
ij + nξ
for all ξ ∈ ∆2,
ij. So
ij) = ξ(tξ
ij )
8
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
that is
α(tα
ij ) = β(tβ
ij ) = γ(tγ
ij) = (α + β)(tα+β
ij
) = 0.
On the other hand, by the Jacobi identity and since α + γ /∈ ∆2, we have that
0 = [eγ
k, [eα
i , eα
for all eγ
such that ad(nα
k ∈ gγ. So ad(nα
k, nα
ij] + [eγ
j ]] = [eγ
k, xα] = [eγ
k) = γ(tα
ij )m = 0 and then γ(tα
ij + nα
ij )t1 + β(tα
k, tα
ij )eγ
ij) = 0. Hence,
ij )t2 + γ(tα
ij = α(tα
ij)(eγ
k. Since ad(nα
xα = tα
ij ] = γ(tα
ij )eγ
k + ad(nα
ij )(eγ
k),
ij ) is nilpotent, there is m ∈ N
ij)t3 + nα
ij = β(tα
ij )t2 + nα
ij .
Therefore [gα, gα] ⊆ ht2i ⊕ n. Analogously, by the Jacobi identity and since α +
γ, β + γ, α + β + γ /∈ ∆2, we can prove that α(tγ
) = 0.
Therefore
ij) = γ(tα+β
ij ) = β(tγ
ij ) = γ(tβ
ij
ij )t1 + nβ
ij
xβ = α(tβ
xγ = nγ
ij
xα+β = α(tα+β
ij
)t1 + β(tα+β
ij
)t2 + nα+β
ij
, α(tα+β
ij
) = β(tα+β
ij
).
Thus
[gβ, gβ] ⊆ ht1i ⊕ n,
[gγ, gγ] ⊆ n
and
[gα+β+γ, gα+β+γ] ⊆ ht1 + t2i ⊕ n.
As g is simple, we have h = Pξ∈∆2[gξ, gξ] ⊆ ht1, t2i ⊕ n and then dimK(t) ≤ 2,
which is a contradiction.
Case 2.b: ∆3 = {α, β, γ, α + γ}.
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+γ.
By the same argument of the above case and since α + β, β + γ, and α + β + γ do
not belong to ∆3, we find that:
ij)t3 + nα
ij .
xα = γ(tα
xβ = nβ
ij.
xγ = α(tγ
ij )t1 + nγ
ij.
xα+γ = α(tα+γ
ij
)t1 + γ(tα+γ
ij
)t3 + nα+γ
ij
, α(tα+γ
ij
) = γ(tα+γ
ij
).
Therefore
[gα, gα] ⊆ ht3i ⊕ n
[gβ, gβ] ⊆ n
[gγ, gγ] ⊆ ht1i ⊕ n
[gα+γ, gα+γ] ⊆ ht1 + t3i ⊕ n
Hence h = Pξ∈∆3[gξ, gξ] ⊆ ht1, t3i ⊕ n and then dimK(t) ≤ 2, which is a contra-
diction.
Case 2.c: ∆4 = {α, β, γ, β + γ}.
Thus
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gβ+γ.
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
9
Since α + β, α + γ, α + β + γ /∈ ∆4 and using the same argument of the above cases
we have
xα = nα
ij
xβ = β(tβ
xγ = β(tγ
ij )t3 + nβ
ij
ij )t2 + nγ
ij.
xβ+γ = β(tβ+γ
ij
Therefore
)t2 + γ(tβγ
ij )t3 + nβ+γ
ij
,
β(tβ+γ
ij
) = γ(tβ+γ
ij
).
[gα, gα] ⊆ n
[gβ, gβ] ⊆ ht3i ⊕ n
[gγ, gγ] ⊆ ht2i ⊕ n,
[gβ+γ, gβ+γ] ⊆ ht2 + t3i ⊕ n.
Thus h = Pξ∈∆4[gξ, gξ] ⊆ ht2, t3i ⊕ n and hence dimK(t) ≤ 2. A contradiction.
Case 2.b: ∆5 = {α, β, γ, α + β + γ}.
Thus
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β+γ.
We have α+β, α+γ, β +γ /∈ ∆5. This fact implies that [gα, gα], [gβ, gβ], [gγ, gγ] and
[gα+β+γ, gα+β+γ] are subsets of n. Therefore h ⊆ n and thus t = 0. A contradiction.
Case 3.a: ∆6 = {α, β, γ, α + β, α + γ}.
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ.
We can prove that
ij )t2 + γ(tα
ij)t3 + nα
ij.
xα = β(tα
xβ = nβ
ij.
xγ = nγ
ij.
xα+β = nα+β
xα+γ = nα+γ
ij
.
.
ij
Therefore, [gξ, gξ] ⊆ n for all ξ ∈ ∆6\{α} and [gα, gα] ⊆ ht2 + t3i ⊕ n. So
g = Xξ∈∆6
[gξ, gξ] ⊆ ht2, t3i ⊕ n.
Hence dimK(t) ≤ 2. A contradiction.
Case 3.b: ∆7 = {α, β, γ, α + β, β + γ}.
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gβ+γ.
10
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
We find that:
[gα, gα] ⊆ n.
[gβ, gβ] ⊆ ht1 + t3i ⊕ n.
[gγ, gγ] ⊆ n.
[gα+β, gα+β] ⊆ ht1 + t2i ⊕ n.
[gβ+γ, gβ+γ] ⊆ ht2 + t3i ⊕ n.
Set
t0 := ht1 + t3i ⊕ ht1 + t2i ⊕ ht2 + t3i .
Thus h ⊆ Pξ∈∆7[gξ, gξ] ⊆ t0 ⊕ n. Let's see that this fact implies that t1 /∈ [g, g].
Indeed, if t1 ∈ [g, g] ⊆ Pξ∈∆7[gξ, gξ] ⊆ t0 ⊕ n, then there are δ1, δ2, δ3 in K and
n ∈ n such that
t1 = δ1(t1 + t3) + δ2(t1 + t2) + δ3(t2 + t3) + n,
that is,
(1 + δ1 + δ2)t1 + (δ2 + δ3)t2 + (δ1 + δ3)t3 = n.
Since t ∩ n = 0, we have (1 + δ1 + δ2)t1 + (δ2 + δ3)t2 + (δ1 + δ3)t3 = 0. As t1, t2, t3
are linearly independent, we obtain the following system of equations:
1 + δ1 + δ2 = 0
= 0
δ2 + δ3
δ1 + δ3
= 0.
Solving the system, we have that 0δ1 + 0δ2 = 1, which is absurd. Therefore t1 /∈
[g, g]. However, this fact implies that t 6⊆ [g, g], which is absurd, since t ⊆ [g, g].
Case 3.c: ∆8 = {α, β, γ, α + β, α + β + γ}.
We have
So,
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+β+γ.
[gξ, gξ] ⊆ n ∀ξ ∈ ∆8\{α + β}.
[gα+β, gα+β] ⊆ ht1 + t2, t3i ⊕ n.
Then h ⊆ ht1 + t2, t3i ⊕ n and so dimK(h) ≤ dimK(n) + 2. Hence dimK(t) ≤ 2
(contradiction).
Case 3.d: ∆9 = {α, β, γ, α + γ, β + γ}.
Then
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+γ ⊕ gβ+γ.
[gα, gα] ⊆ n.
[gβ, gβ] ⊆ n.
[gγ, gγ] ⊆ ht1, t2i ⊕ n.
[gα+γ, gα+γ] ⊆ n.
[gβ+γ, gβ+γ] ⊆ n.
Therefore h = Pξ∈∆9[gξ, gξ] ⊆ ht1, t2i ⊕ n and then dimk(t) ≤ 2. A contradiction.
Case 3.e: ∆10 = {α, β, γ, α + γ, α + β + γ}.
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
11
Then
We have,
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+γ ⊕ gα+β+γ.
[gξ, gξ] ⊆ n ∀ξ ∈ ∆10\{α + γ}
[gα+γ, gα+γ] ⊆ ht2, t1 + t3i ⊕ n,
and then dimk(T ) ≤ 2, which is absurd.
Case 3.f : ∆11 = {α, β, γ, β + γ, α + β + γ}.
Then
Therefore
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gβ+γ ⊕ gα+β+γ.
[gξ, gξ] ⊆ n ∀ξ ∈ ∆11\{β + γ}.
[gβ+γ, gβ+γ] ⊆ ht2, t1 + t3i ⊕ n.
Hence h ⊆ ht2, t1 + t3i ⊕ n and so dim(t) ≤ 2. A contradiction.
Case 4.a: ∆12 = {α, β, γ, α + β, α + γ, β + γ}.
Then
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ.
By similar arguments presented in the above cases, we have that
[gα, gα],
[gβ, gβ],
[gγ, gγ],
[gβ+γ, gβ+γ] ⊆ ht2 + t3i ⊕ n.
[gα+γ, gα+γ] ⊆ ht1 + t3i ⊕ n.
[gα+β, gα+β] ⊆ ht1 + t2i ⊕ n.
As in Case 3.b, this fact implies that t1 /∈ [g, g], because if
t1 ∈ [g, g] ⊆ Xξ∈∆12
[gξ, gξ] ⊆ t0 ⊕ n, where
t0 := ht2 + t3i ⊕ ht1 + t2i ⊕ ht1 + t3i ,
then, for some δi ∈ K and n ∈ n,
t1 = δ1(t2 + t3) + δ2(t1 + t2) + δ3(t1 + t3) + n.
Therefore (1 + δ2 + δ3)t1 + (δ1 + δ2)t2 + (δ1 + δ3)t3 = n. Since t ∩ n = 0, we have
(1 + δ2 + δ3)t1 + (δ1 + δ2)t2 + (δ1 + δ3)t3 = n = 0. Given that t1, t2, t3 are linearly
independent, we obtain the following system of equations
1 + δ2 + δ3 = 0
= 0
δ1 + δ2
δ1 + δ3
= 0
which has no solution in K, and this is a contradiction. Therefore t1 /∈ [g, g], that
is, t 6⊆ [g, g], which is absurd, since t ⊆ [g, g].
Case 4.b: ∆13 = {α, β, γ, α + β, α + γ, α + β + γ}.
Then
We have
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gα+β+γ.
[gα, gα], [gα+β+γ, gα+β+γ] ⊆ ht2 + t3i ⊕ n.
[gβ, gβ], [gγ, gγ] ⊆ ht1i ⊕ n
[gα+β, gα+β], [gα+γ, gα+γ] ⊆ ht1 + t2 + t3i ⊕ n.
12
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
By the same argument given in Case 3.b, we have that t2 /∈ [g, g], which is a
contradiction, since t ⊆ [g, g].
Case 4.c: ∆14 = {α, β, γ, α + β, β + γ, α + β + γ}.
We have
Therefore,
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gβ+γ ⊕ gα+β+γ.
[gα, gα] ⊆ ht2i ⊕ n
[gβ, gβ] ⊆ ht1 + t3i ⊕ n
[gγ, gγ] ⊆ ht2i ⊕ n
[gα+β, gα+β] ⊆ ht1 + t2 + t3i ⊕ n
[gβ+γ, gβ+γ] ⊆ ht1 + t2 + t3i ⊕ n
[gα+β+γ, gα+β+γ] ⊆ ht1 + t3i ⊕ n
We prove that t1 /∈ [g, g]. Otherwise,
t1 ∈ [g, g] ⊆ Xξ∈∆14
[gξ, gξ] ⊆ t0 ⊕ n, where
t0 := ht2i ⊕ ht1 + t3i ⊕ ht1 + t2 + t3i .
Then t1 = δ1(t2) + δ2(t1 + t3) + δ3(t1 + t2 + t3) + n for some δi ∈ K and n ∈
n. Thus (1 + δ2 + δ3)t1 + (δ1 + δ2)t2 + (δ1 + δ3)t3 = n. As t ∩ n = 0, then
(1 + δ2 + δ3)t1 + (δ1 + δ2)t2 + (δ1 + δ3)t3 = n = 0. Since t1, t2, t3 are linearly
independent, we have the following system
1 + δ2 + δ3 = 0
= 0
δ1 + δ2
δ1 + δ3
= 0
which has no solution, a contradiction. Hence, t1 /∈ [g, g]. This fact is absurd
because t ⊆ [g, g].
Case 4.d: ∆15 = {α, β, γ, α + γ, β + γ, α + β + γ}, then
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+γ ⊕ gβ+γ ⊕ gα+β+γ
We have
[gα, gα] ⊆ ht3i ⊕ n.
[gβ, gβ] ⊆ ht3i ⊕ n.
[gγ, gγ] ⊆ ht1 + t2i ⊕ n.
[gα+γ, gα+γ] ⊆ ht1 + t2 + t3i ⊕ n.
[Lβ+γ, Lβ+γ] ⊆ ht1 + t2 + t3i ⊕ n.
[gα+β+γ, gα+β+γ] ⊆ ht1 + t2i ⊕ n.
Set t0 = ht3i⊕ht1 + t2i⊕ht1 + t2 + t3i. Thus [g, g] ⊆ Pξ∈∆15 [gξ, gξ] ⊆ t0⊕n. If t1 ∈
[g, g] then there are δ1, δ2, δ3 in K such that t1 = δ1t3+δ2(t1+t2)+δ3(t1+t2+t3)+n.
Therefore we have the following system of equations
1 + δ2 + δ3 = 0
δ2 + δ3
= 0
= 0.
δ1 + δ3
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
13
Solving the system, we have 0δ2 + 0δ3 = 1, which is absurd. Therefore t1 /∈ [g, g],
which is a contradiction because t ⊆ [g, g].
(cid:3)
5. Analysis of g = h ⊕ (⊕ξ∈∆gξ), with ∆ = t∗\{0}.
In this section, we will study the Cartan decomposition of g∆, when Card(∆) =
7, that is, when ∆ = ∆0 = t∗\{0}. We will prove that there not simple Lie 2-algebra
of 10 ≤ dimK(g) ≤ 16 with M T (g) = 3.
From now on, we will suppose that (g, [2]) is a simple Lie 2-algebra of toral rank
3, with dimension greater that or equal to 10, whose Cartan decomposition with
respect to t is:
g = t ⊕ n ⊕ gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ ⊕ gα+β+γ.
Remember that we are considering that
dimK(gα) ≥ dimK(gβ) ≥ dimK(gγ) ≥ dimK(gξ) > 0 for
ξ /∈ ∆\{α, β, α + β}
(see Remark 3.4).
Next theorem will help us to study (classify) those algebras in whose decomposi-
tion of Cartan the root spaces have different dimensions, which will be fundamental
to show our main result (Theorem 5.4).
Theorem 5.1. Take α, β ∈ ∆. If there exists eα ∈ gα such that
e[2]
α := tα + nα,
tα ∈ t\{0}, nα ∈ n and β(tα) 6= 0,
then gβ is isomorphic to gα+β.
Proof. Take eα ∈ gα and let ad(eα) : gβ → gα+β be the adjoint mapping. We prove
that ad(eα) is injective. Take eβ, fβ in gβ such that ad(eα)(eβ) = ad(eα)(fβ). Then
[eα, eβ] = [eα, fβ]. Therefore
[e[2]
α , eβ] = [eα, [eα, eβ]] = [eα, [eα, fβ]] = [eα
[2], fβ].
By hypothesis [tα + nα, eβ] = [tα + nα, fβ], that is,
β(tα)eβ + ad(nα)(eβ) = β(tα)fβ + ad(nα)(fβ)
and so β(tα)(eβ + fβ) = ad(nα)(eβ + fβ). Therefore, ad(tα)(eβ + fβ) = ad(nα)(eβ +
fβ). As ad(nα) is nilpotent, there exists m ∈ N such that ad(nα)m = 0. So
ad(tα)m(eβ +fβ) = 0 and therefore (ad(tα)(eβ +fβ))m = 0. Then ad(tα)(eβ +fβ) =
0 and so β(tα)(eβ +fβ) = 0. As β(tα) 6= 0 we have eβ = fβ and so ad(eα) is injective.
Analogously we can prove that ad(eα) : gα+β → gβ is an injective linear trans-
formation, since (α + β)(tα) = α(tα) + β(tα) = 0 + β(tα) = β(tα) 6= 0.
The rank-nullity formula implies that
dimK(gβ) = dimKIm(ad(eα)) ≤ dimK(gα+β)
and
dimK(gα+β) = dimKIm(ad(eα)) ≤ dimK(gβ).
Therefore dimK(gβ) = dimK(gα+β).
(cid:3)
Notation 5.2. For j ∈ {g, t, n, gα, gβ, gγ, gα+β, gα+γ, gβ+γ, gα+β+γ}, we set
Consider
dimK(j) = d(j).
P = (d(g) : d(t), d(n), d(gα), d(gβ), d(gγ), d(gα+β), d(gα+γ), d(gβ+γ), d(gα+β+γ)).
14
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
Next, fixing the dimension of (g, [2]), we will study its structure when ∆ =
t∗\{0}. We have the followings possibilities:
1. If dimK(g) = 10, we have P = (10 : 3, 0, 1, 1, 1, 1, 1, 1, 1). Thus I = ⊕
ξ∈∆
gξ is an
ideal of g, so g is not simple. A contradiction.
2. If dimK(g) = 11, we have the following cases:
i. P = (11 : 3, 0, 2, 1, 1, 1, 1, 1, 1). In this case, as t ⊆ Pξ∈∆[gξ, gξ] ⊆ [gα, gα],
then the dimK(t) ≤ 1 ( A contradiction).
ii. P = (11 : 3, 1, 1, 1, 1, 1, 1, 1, 1). Then I := n ⊕ ( ⊕
ξ∈∆
gξ) is an ideal of g. A
contradiction.
3. If dimK(g) = 12, we have:
i. P = (12 : 3, 0, 3, 1, 1, 1, 1, 1, 1). If eα ∈ [gα, gα], then there exist δ1, δ2, δ3 ∈
K such that eα = δ1t1 + δ2t2 + δ3t3. Thus
0 = [eα, eα] = δ1α(t1)eα + δ2α(t2)eα + δ3α(t3)eα = δ1eα.
Therefore δ1 = 0 and so eα = δ2t2 + δ3t3, that is, eα ∈ ht2, t3i. Hence
dimK(t) ≤ 2, which is a contradiction.
and so dimK(t) ≤ dimK([gα, gα]) + dimK([gβ, gβ]) ≤ 1 + 1 = 2.
[gα, gα] ⊆ ht2, t3i . Since t ⊆ Pξ∈∆[gξ, gξ] ⊆ [gα, gα] ⊆ ht2, t3i, we have
ii. P = (12 : 3, 0, 2, 2, 1, 1, 1, 1, 1). Then t ⊆ Pξ∈∆[gξ, gξ] ⊆ [gα, gα] + [gβ, gβ]
iii. P = (12 : 3, 1, 2, 1, 1, 1, 1, 1, 1). We have h ⊆ Pξ∈∆[gξ, gξ] = [gα, gα] and
since dimK(gα) = 2, it follows that dimK([gα, gα]) ≤ 1. So, dimK(h) ≤ 1.
A contradiction.
iv. P = (12 : 3, 2, 1, 1, 1, 1, 1, 1, 1). We have I := n ⊕Pξ∈∆ ⊕gξ is an ideal of g.
A contradiction.
4. If dimK(g) = 13, then we have the following structures for g:
i. P = (13 : 3, 0, 4, 1, 1, 1, 1, 1, 1).
ii. P = (13 : 3, 0, 3, 2, 1, 1, 1, 1, 1).
iii. P = (13 : 3, 0, 2, 2, 2, 1, 1, 1, 1).
β ∈ t and so there exist δ1, δ2, δ3 ∈ K such that e[2]
For these three cases we have: let eβ ∈ gβ be an element of the base of gβ.
Then e[2]
β = δ1t1 + δ2t2 +
δ3t3. Since β(e[2]
β ) = 0, we have δ2 = 0. Thus tβ = δ1t1 + δ3t3 for some
[2]
δ1, δ3 ∈ K and so g
β ⊆ ht1, t3i. If we choose δ1 6= 0, then α(tβ ) = δ1 6= 0.
By Theorem 5.1 we have gα ≃ gα+β. However, this is impossible because
dimK(gα) 6= dimK(gα+β) = 1.
iv. P = (13 : 3, 1, 1, 1, 1, 1, 1, 1). Thus I := n ⊕ ( ⊕
ξ∈∆
gξ) is an ideal of g, which is
absurd.
v. P = (13 : 3, 2, 2, 1, 1, 1, 1, 1, 1). We have h ⊆ Pξ∈∆[gξ, gξ] = [gα, gα]. Thus
dimK(h) ≤ 1. A contradiction.
vi. P = (13 : 3, 1, 3, 1, 1, 1, 1, 1, 1). We have
h ⊆ Xξ∈∆
[gξ, gξ] = [gα, gα] ⊆ ht2, t3i
and so dimK(h) ≤ 2. A contradiction.
vii. P = (13 : 3, 1, 2, 2, 1, 1, 1, 1, 1). Thus h ⊆ Pξ∈∆[gξ, gξ] = [gα, gα] + [gβ, gβ].
Hence dimK(h) ≤ 2. A contradiction.
5. If dimK(g) = 14. We have the following possibilities for P:
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
15
i. (14 : 3, 0, 5, 1, 1, 1, 1, 1, 1)
ii. (14 : 3, 0, 4, 2, 1, 1, 1, 1, 1)
iii. (14 : 3, 0, 3, 3, 1, 1, 1, 1, 1)
iv. (14 : 3, 0, 3, 2, 2, 1, 1, 1, 1)
v. (14 : 3, 0, 2, 2, 2, 2, 1, 1, 1)
vi. (14 : 3, 1, 2, 2, 2, 1, 1, 1, 1)
vii. (14 : 3, 1, 3, 2, 1, 1, 1, 1, 1)
viii. (14 : 3, 1, 4, 1, 1, 1, 1, 1, 1)
ix. (14 : 3, 2, 2, 2, 1, 1, 1, 1, 1)
x. (14 : 3, 2, 3, 1, 1, 1, 1, 1, 1)
xi. (14 : 3, 3, 2, 1, 1, 1, 1, 1, 1)
xii. (14 : 3, 4, 1, 1, 1, 1, 1, 1, 1)
For the case xii, we have that I := n ⊕ ( ⊕
ξ∈∆
gξ) is an ideal of g, which is
γ ∈ g
[2]
absurd. Now, let eγ be an element of the base of gγ. As e[2]
γ ⊆ h = t ⊕ n,
then there are tγ ∈ t and nγ ∈ n such that e[2]
γ = tγ + nγ. As tγ ∈ t, there
exist δ1, δ2, δ3 in K, such that tγ = δ1t1 + δ2t2 + δ3t3. On the other hand,
0 = [eγ, e[2]
γ ] = γ(tγ)eγ + [eγ, nγ], then ad(nγ)(eγ) = γ(tγ)eγ. As nγ is nilpotent
then ad(nγ) is nilpotent and therefore γ(tγ) = δ3 = 0. So, e[2]
γ = δ1t1 + δ2t2 + nγ.
Choosing δ3 6= 0 we have tγ 6= 0 and α(tγ ) = δ1 6= 0, then by Theorem 5.1,
gα ≃ gα+γ. This fact shows that the remaining cases are impossible because
dimK(gα) 6= dimK(gα+γ) = 1.
6. If dimK(g) = 15, we have the following cases:
i. (15 : 3, 5, 1, 1, 1, 1, 1, 1, 1)
ii. (15 : 3, 4, 2, 1, 1, 1, 1, 1, 1)
iii. (15 : 3, 3, 2, 2, 1, 1, 1, 1, 1)
iv. (15 : 3, 2, 3, 2, 1, 1, 1, 1, 1)
v. (15 : 3, 2, 2, 2, 2, 1, 1, 1, 1)
vi. (15 : 3, 1, 3, 2, 2, 1, 1, 1, 1)
vii. (15 : 3, 0, 6, 1, 1, 1, 1, 1, 1)
viii. (15 : 3, 0, 5, 2, 1, 1, 1, 1, 1)
ix. (15 : 3, 0, 4, 3, 1, 1, 1, 1, 1)
x. (15 : 3, 0, 3, 3, 2, 1, 1, 1, 1)
xi. (15 : 3, 0, 3, 2, 2, 2, 1, 1, 1)
xii. (15 : 3, 0, 2, 2, 2, 2, 2, 1, 1)
xiii. (15 : 3, 0, 2, 2, 2, 2, 2, 1, 1)
For the case i. we have I := n ⊕ ( ⊕
ξ∈∆
gξ) is an ideal of g. A contradiction.
As in the above cases, let eβ be an element of the basis of gβ. Since e[2]
β ∈
β ⊆ h = t ⊕ n, there exist tβ ∈ t and nβ ∈ n such that e[2]
[2]
β = tβ + nβ. As
g
tβ ∈ t, there are δ1, δ2, δ3 in K, such that tβ = δ1t1 + δ2t2 + δ3t3. On the other
hand, 0 = [eβ, e[2]
β ] = β(tβ)eβ + [eβ, nβ] and thus ad(nβ)(eβ) = β(tβ)eβ. Since
nβ is nilpotent we have ad(nβ) is nilpotent and therefore β(tβ ) = δ2 = 0. So,
e[2]
β = δ1t1 + δ3t3 + nβ. Choosing δ1 6= 0, we have tβ 6= 0 and α(tβ ) = δ1 6= 0.
Thus, by Theorem 5.1, we have gα ≃ α+β. Therefore, the cases from ii. to xii.
are all impossible, because dimK(gα) 6= dimK(gα+β ) = 1.
On the other hand, choosing δ1 6= δ3, we have
(α + γ)(tβ) = (α + γ)(δ1t1 + δ3t3) = δ1 + δ3 6= 0.
Therefore, by Theorem 5.1, we have gα+γ ≃ gα+β+γ. However
2 = dimK(gα+γ) 6= dimK(gα+β+γ) = 1,
then the case xiii. is not possible.
7. If dimK(g) = 16, we have the following cases:
16
CARLOS R. PAYARES GUEVARA AND JEOVANNY DE J. MUENTES ACEVEDO
i. (16 : 3, 6, 1, 1, 1, 1, 1, 1, 1)
ii. (16 : 3, 5, 2, 1, 1, 1, 1, 1, 1)
iii. (16 : 3, 4, 3, 1, 1, 1, 1, 1, 1)
iv. (16 : 3, 4, 2, 2, 1, 1, 1, 1, 1)
v. (16 : 3, 3, 4, 1, 1, 1, 1, 1, 1)
vi. (16 : 3, 3, 3, 2, 1, 1, 1, 1, 1)
vii. (16 : 3, 3, 2, 2, 2, 1, 1, 1, 1)
viii. (16 : 3, 2, 5, 1, 1, 1, 1, 1, 1)
ix. (16 : 3, 2, 4, 2, 1, 1, 1, 1, 1)
x. (16 : 3, 2, 3, 3, 1, 1, 1, 1, 1)
xi. (16 : 3, 2, 3, 2, 2, 1, 1, 1, 1)
xii. (16 : 3, 2, 2, 2, 2, 2, 1, 1, 1)
xiii. (16 : 3, 1, 6, 1, 1, 1, 1, 1, 1)
xiv. (16 : 3, 1, 5, 2, 1, 1, 1, 1, 1)
xv. (16 : 3, 1, 4, 3, 1, 1, 1, 1, 1)
xvi. (16 : 3, 1, 4, 2, 2, 1, 1, 1, 1)
xvii. (16 : 3, 1, 3, 3, 2, 1, 1, 1, 1)
xviii. (16 : 3, 1, 2, 2, 2, 2, 2, 1, 1)
xix. (16 : 3, 0, 7, 1, 1, 1, 1, 1, 1)
xx. (16 : 3, 0, 6, 2, 1, 1, 1, 1, 1)
xxi. (16 : 3, 0, 5, 2, 2, 1, 1, 1, 1)
xxii. (16 : 3, 0, 4, 3, 2, 1, 1, 1, 1)
xxiii. (16 : 3, 0, 4, 2, 2, 2, 1, 1, 1)
xxiv. (16 : 3, 0, 4, 2, 2, 2, 1, 1, 1)
xxv. (16 : 3, 0, 4, 2, 2, 2, 1, 1, 1)
xxvi. (16 : 3, 0, 3, 3, 2, 2, 1, 1, 1)
xxvii. (16 : 3, 0, 3, 2, 2, 2, 2, 1, 1)
xxviii. (16 : 3, 0, 2, 2, 2, 2, 2, 2, 1)
For the case i., we have I := n⊕( ⊕
ξ∈∆
gξ) is an ideal of g, which is a contradiction. Let
β ∈ g
[2]
eβ be an element of the basis of gβ. As e[2]
β ⊆ h = t ⊕ n, then there are tβ ∈ t
and nβ ∈ n such that e[2]
β = tβ + nβ. As tβ ∈ t, there exist δ1, δ2, δ3 in K, such that
tβ = δ1t1 + δ2t2 + δ3t3. On the other hand, 0 = [eβ, e[2]
β ] = β(tβ )eβ + [eβ, nβ] and
then ad(nβ)(eβ) = β(tβ )eβ. Since nβ is nilpotent, ad(nβ) is nilpotent and therefore
β(tβ) = δ2 = 0. So e[2]
β = δ1t1 + δ3t3 + nβ. Choosing δ1 6= 0 we have tβ 6= 0 and
α(tβ) = δ1 6= 0. By Theorem 5.1, we have gα ∼= gα+β. Therefore the cases from i.
to xxvii., except xii. and xix., are impossible, because dimK(gα) 6= dimK(gα+β).
On the other hand, choosing δ3 6= 0, we have γ(tβ) = δ3 6= 0. Thus, by Theorem
5.1, we have gγ ≃ gβ+γ, but 2 = dimK(gγ) 6= dimK(gβ+γ) = 1. Hence the cases
xii. and xix. are not possible.
In the case xxviii. choosing δ1 6= δ3, we have
(α + γ)(tβ) = δ1 + δ3 6= 0. Therefore, by Theorem 5.1 we have gα+γ ≃ gα+β+γ,
which is absurd, since 2 = dimK(gα+γ) 6= dimK(gα+β+γ) = 1 and hence this case
is not possible.
It follows from the above facts that:
Proposition 5.3. There are no simple Lie 2-algebras of dimension between 10 and
16, and with toral rank 3.
We finish this work presenting our main result, which follows from Propositions
2.1, 4.3 and 5.3.
Theorem 5.4. There are no simple Lie 2-algebras of dimension less than or equal
to 16, and toral rank 3.
References
[1] BLOCK, Richard E. The classification problem for simple Lie algebras of characteristic p.
Lie Algebras and Related Topics. Springer, Berlin, Heidelberg, 1982. p. 38-56.
[2] BLOCK, Richard E.; WILSON, Robert Lee. The restricted simple Lie algebras are of classical
or Cartan type. Proceedings of the National Academy of Sciences, 1984, vol. 81, no 16, p.
5271-5274.
[3] BLOCK, Richard E.; WILSON, Robert Lee. Classification of the restricted simple Lie alge-
bras. Journal of Algebra, 1988, vol. 114, no 1, p. 115-259.
[4] GRISHKOV, Alexander; PREMET, Alexander. Simple Lie algebras of absolute toral rank 2
in characteristic 2, preprint.
ON SIMPLE LIE 2-ALGEBRA OF TORAL RANK 3.
17
[5] JACOBSON, Nathan. Lie algebras, Interscience, New York, 1962. Google Scholar, 1979, p.
120.
[6] SKRYABIN, Serge. Toral rank one simple Lie algebras of low characteristics. Journal of
Algebra, 1998, vol. 200, no 2, p. 650-700.
[7] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic:
I.
Sandwich elements. Journal of Algebra, 1997, vol. 189, no 2, p. 419-480.
[8] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic II. Ex-
ceptional roots. Journal of Algebra, 1999, vol. 216, no 1, p. 190-301.
[9] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic: III.
The toral rank 2 case. Journal of Algebra, 2001, vol. 242, no 1, p. 236-337.
[10] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic IV.:
Solvable and classical roots. Journal of Algebra, 2004, vol. 278, no 2, p. 766-833.
[11] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic V. The
non-Melikian case. Journal of Algebra, 2007, vol. 314, no 2, p. 664-692.
[12] PREMET, Alexander; STRADE, Helmut. Simple Lie algebras of small characteristic VI.
Completion of the classification. Journal of Algebra, 2008, vol. 320, no 9, p. 3559-3604.
[13] PREMET, Alexander; STRADE, Helmut. Classification of finite dimensional simple Lie al-
gebras in prime characteristics. Contemporary Mathematics, 2006, vol. 413, p. 185-214.
[14] STRADE, Helmut; WILSON, Robert Lee. Classification of simple Lie algebras over alge-
braically closed fields of prime characteristic. Bulletin of the American Mathematical Society,
1991, vol. 24, no 2, p. 357-362.
[15] STRADE, Helmut. The absolute toral rank of a Lie algebra. Lie algebras, Madison 1987.
Springer, Berlin, Heidelberg, 1989. p. 1-28.
[16] STRADE, Helmut. The classification of the simple modular Lie algebras: I. Determination
of the two-sections. Annals of Mathematics, 1989, vol. 130, no 3, p. 643-677.
[17] STRADE, Helmut. The classification of the simple modular Lie algebras II. The toral struc-
ture. Journal of Algebra, 1992, vol. 151, no 2, p. 425-475.
[18] STRADE, Helmut. The classification of the simple modular Lie algebras: III. Solution of the
classical case. Annals of Mathematics, 1991, p. 577-604.
[19] STRADE, Helmut. The classification of the simple modular Lie algebras: VI. Solving the final
case. Transactions of the American Mathematical Society, 1998, vol. 350, no 7, p. 2553-2628.
[20] STRADE, Helmut. Lie algebras of small dimension. Contemporary Mathematics, 2007, vol.
442, p. 233.
[21] WILSON, Robert Lee. Classification of the restricted simple Lie algebras with toral Cartan
subalgebras. Journal of Algebra, 1983, vol. 83, no 2, p. 531-570.
Facultad de ciencias B´asicas, Universidad Tecnol´ogica de Bolivar, Cartagena de
Indias - Colombia
E-mail address: [email protected]
Facultad de ciencias B´asicas, Universidad Tecnol´ogica de Bolivar, Cartagena de
Indias - Colombia
E-mail address: [email protected]
|
1611.03480 | 1 | 1611 | 2016-11-10T20:36:11 | A note on the order of the antipode of a pointed Hopf algebra | [
"math.RA"
] | Let $k$ be a field and let $H$ denote a pointed Hopf $k$-algebra with antipode $S$. We are interested in determining the order of $S$. Building on the work done by Taft and Wilson $[7]$, we define an invariant for $H$, denoted $m_{H}$, and prove that the value of this invariant is connected to the order of $S$. In the case where $\operatorname{char}k=0$, it is shown that if $S$ has finite order then it is either the identity or has order $2m_{H}$. If in addition $H$ is assumed to be coradically graded, it is shown that the order of $S$ is finite if and only if $m_{H}$ is finite. We also consider the case where $\operatorname{char}k=p>0$, generalising the results of $[7]$ to the infinite-dimensional setting. | math.RA | math | A NOTE ON THE ORDER OF THE ANTIPODE OF A POINTED
HOPF ALGEBRA
P. GILMARTIN
v
o
N
0
1
]
.
A
R
h
t
a
m
[
1
v
0
8
4
3
0
.
1
1
6
1
:
v
i
X
r
a
Abstract. Let k be a field and let H denote a pointed Hopf k-algebra with
antipode S. We are interested in determining the order of S. Building on the
work done by Taft and Wilson in [7], we define an invariant for H, denoted
mH , and prove that the value of this invariant is connected to the order of
S. In the case where char k = 0, it is shown that if S has finite order then
it is either the identity or has order 2mH . If in addition H is assumed to be
coradically graded, it is shown that the order of S is finite if and only if mH is
finite. We also consider the case where char k = p > 0, generalising the results
of [7] to the infinite-dimensional setting.
1. Introduction
In this paper we are interested in determining the order of the antipode of a
pointed Hopf algebra over an arbitrary field k. We suspect that the results of this
paper are well-known to experts, but their proofs are apparently lacking in the
literature.
For a pointed Hopf k-algebra H, we introduce an invariant, denoted mH , which
is, in a sense which we shall make precise, a measure of the extent to which a group-
like element x ∈ H commutes with any h ∈ H such that ∆(h) = h ⊗ x + 1 ⊗ h.
Whilst in general mH can take values in Z≥0 ∪ {∞}, the condition that mH is finite
is valid in a variety of natural settings, for example whenever G(H) is finite or
central in H (see Proposition 2.7). We record our main results below (where part
(3) appears as Theorem 4.1 and part (4) appears as Corollary 5.2). These results
connect the order of the antipode of a pointed Hopf algebra H to the value of mH
in both the cases of zero and positive characteristic. For the relevant definitions
see §2.
Theorem 1.1. Let k be a field. Suppose H is a pointed Hopf k-algebra. Let
{Hn}n≥0 denote the coradical filtration of H and let gr H denote the associated
graded pointed Hopf algebra with respect to the coradical filtration. Let S and S
denote the antipode of H and gr H respectively.
(1) If mH = ∞ then S = ∞.
(2) (Taft, Wilson) If mH < ∞ then (S2mH − id)(Hn) ⊆ Hn−1 for n ≥ 1.
(3) Suppose char k = 0. If S < ∞ then either S = id or
S = 2mH = 2mgr H = S.
2010 Mathematics Subject Classification. Primary 16T05, 16T15; Secondary 17B37,20G42.
Some of these results will form part of the authors PhD thesis at the University of Glasgow,
supported by a Scholarship of the Carnegie Trust. The author would like to thank Ken Brown
for very helpful discussions.
1
2
PAUL GILMARTIN
(4) Suppose char k = p > 0. If mH < ∞ and H = khHni for some n ≥ 0, then
S divides 2mH pl, where l ∈ N is such that pl ≥ n ≥ pl−1.
Theorem 1.1(2), from which everything else quickly follows, has exactly the same
proof as part (1) of the following result of Taft and Wilson from 1974, and so we
credit it to them. Parts (3) and (4) of Theorem 1.1 should be compared to the
analogous results for H finite-dimensional as presented in part (2) of the result
below. It will be clear from the definition that mH divides the exponent of G(H)
whenever the exponent is finite.
Theorem 1.2. (Taft, Wilson, [7]) Let k be a field and let H be a pointed Hopf
k-algebra. Assume that G(H) has finite exponent e.
(1) ([7, Proposition 3, Proposition 4]) For n ≥ 1, (S2e − id)(Hn) ⊆ Hn−1.
(2) ([7, Corollary 6]) Assume that H is finite-dimensional and that H = Hn
for some n ≥ 0. If char k = 0 and S has finite order, then S divides 2e.
If char k = p > 0, S2epm
= Id, where pm ≥ n > pm−1.
Remark 1.3. It was subsequently proved by Radford, [5], that the order of the
antipode of a finite-dimensional Hopf algebra is always finite, allowing us to drop
the assumption that S has finite order in the finite dimensional setting of Theorem
1.2(2).
As noted in Remark 4.6, the converse of Theorem 1.1(1) is in general not true.
However, as an almost immediate consequence of of Theorem 1.1(2), in the case
where H is known to be coradically graded (see Definition 3.8), the condition that
mH < ∞ is equivalent to the condition that S has finite order. That is, we deduce
the following, which appears later as Proposition 3.9.
Proposition 1.4. Let k be a field and let H be a pointed coradically graded Hopf
k-algebra.
(1) S = ∞ if and only if mH = ∞.
(2) If mH < ∞, S = id or S = 2mH .
Remark 1.5.
(1) In positive characteristic, there exist examples of pointed
Hopf algebras where the antipode has order strictly less than the bound
obtained in Theorem 1.1(4). Take, for example, a field k such that char k =
p > 0 and let H be a pointed coradically Hopf graded Hopf k-algebra with
mH < ∞ and H 6= H0 (see Example 2.8 for an explicit example of such a
Hopf algebra). By Proposition 1.4, S = 2mH, which is strictly less than
the bound obtained in Theorem 1.1.
On the other hand, there exist examples of pointed Hopf algebras over
fields of positive characteristic where the bound obtained in Theorem 1.1
(4) is actually attained. In Example 5.4 we give an example, originally due
to Taft and Wilson, [6], of a finite dimensional connected Hopf algebra R
over a field of characteristic p ≥ 3 with R = R2, mR = 1 and an antipode
of order 2p.
(2) Note that for an arbitrary pointed Hopf algebra H, mH will be in general
strictly less than the exponent of G(H) - in Example 2.8 we see that, if q
is a primitive nth root of unity, the pointed coradically Hopf graded Hopf
algebra H = Uq(b+) has the property that mH = n and that G(H) has
infinite exponent.
3
2. Preliminaries
Throughout k will denote an arbitrary field (unless otherwise stated). For a Hopf
k-algebra H the usual notation ∆, ǫ and S will denote the coproduct, counit and
antipode respectively. By the order of the antipode S, which we shall denote by
S, we mean the minimal n such that Sn = id, the identity map of H.
The coradical filtration of a Hopf algebra H is denoted {Hn}∞
n=0, where H0 is
the coradical of H (that is, the sum of its simple subcoalgebras), and we define
inductively, for all n ≥ 1,
Hn := ∆−1(H ⊗ Hi−1 + H0 ⊗ H).
For a Hopf algebra H, an element g ∈ H is said to be group-like if ∆(g) = g ⊗ g.
It is a simple exercise to prove that the set of all group-like elements of H forms
a group, which we shall denote G(H). A Hopf algebra H is said to be pointed if
H0 = kG(H) (or, equivalently, if each simple subcoalgebra is one-dimensional). As
proved in [3, Lemma 5.2.8], for example, if H is pointed then the coradical filtration
{Hn}∞
n=0 is in fact a Hopf algebra filtration of H and hence the associated graded
space with respect to the coradical filtration, which we denote by gr H, inherits the
structure of a pointed Hopf algebra from H.
2.1. Defining mH . The aim of this section is to define, for any pointed Hopf
algebra H, the invariant mH which appears in Theorem 1.1. We begin by recalling
the definition of a skew-primitive element of a pointed Hopf algebra.
Definition 2.1. Suppose H is a pointed Hopf algebra. For any x, y ∈ G(H), define
the space of (x, y)-skew-primitive elements of H,
Px,y(H) := {h ∈ H : ∆(h) = h ⊗ x + y ⊗ h} .
Remark 2.2. As noted in the opening remarks of [3, §5.4], if H is a pointed Hopf
algebra and x, y ∈ G(H), then Px,y(H) ∩ H0 = k(x − y). For each such pair
x, y ∈ G(H), let Px,y(H)′ denote a subspace such that
Px,y(H) = k(x − y) ⊕ Px,y(H)′.
The following result, which appears as stated below as [3, Theorem 5.4.1], but is
originally due to Taft and Wilson, [7], is the crux of the proof of our main result,
Theorem 1.1.
Theorem 2.3. Let H be a pointed Hopf algebra. Then
H1 = kG(H) ⊕ (⊕x,y∈GPx,y(H)′).
We also require the following well-known and easy lemma.
Lemma 2.4. Let H be a pointed Hopf algebra, let x ∈ G(H) and let hxi denote the
subgroup of G(H) generated by x.
(1) Px,1(H) is an hxi-invariant subspace of H, where x acts by conjugation.
(2) Px,1(H) ⊆ ker ǫ.
Proof.
(1) Let h ∈ Px,1(H). Since ∆ is an algebra homomorphism,
∆(xhx−1) = (x ⊗ x)(h ⊗ x + 1 ⊗ h)(x−1 ⊗ x−1)
= xhx−1 ⊗ x + 1 ⊗ xhx−1
hence xhx−1 ∈ Px,1(H).
4
PAUL GILMARTIN
(2) Let h ∈ Px,1(H). By the counit axiom of the coproduct, hǫ(x) + ǫ(h) = h.
Since x is group-like, ǫ(x) = 1. The result follows.
(cid:3)
In light of Lemma 2.4 (1), we make the following definition.
Definition 2.5. Let H be a pointed Hopf algebra. For any x ∈ G(H), define
ax = hxi : Chxi(Px,1(H))
where Chxi(Px,1(H)) denotes the centraliser of Px,1(H) in hxi under the conjugation
action by hxi.
Definition 2.6. For a pointed Hopf algebra H, define
mH := lcm{ax : x ∈ G}.
We record a couple of simple observations about mH.
Proposition 2.7. Let H be a pointed Hopf algebra.
(1) If G(H) is central in H then mH = 1.
(2) Suppose G(H) is finite. Then mH is finite and divides the exponent of the
group G(H).
(3) If H = H0 = kG(H) then mH = 1.
Proof. Parts (1) and (2) are immediate from the way we defined mH. For part (3),
let x ∈ G(H). If H = kG(H), then as noted in Remark 2.2, Px,1(H) = k(x − 1). It
follows that ax = 1 and hence mH = 1.
(cid:3)
For an arbitrary pointed Hopf algebra H, mH can take values in Z≥0 ∪ {∞} and
will be in general strictly less than the exponent of the group G(H), as shown by
the following example.
Example 2.8. Let k be a field and let 0, 1 6= q ∈ k. Set H = Uq(b+), the
quantised enveloping algebra of the positive two dimensional Borel. This is defined
as the algebra generated by the letters E, K and K −1, subject to the relations
KK −1 = 1 = K −1K and
KE = qEK.
Then, as proved in [2, I.3.4], for example, H becomes a pointed Hopf algebra, with
coproduct ∆ : H → H ⊗ H and antipode S : H → H defined on generators as
follows
∆(E) = E ⊗ 1 + K ⊗ E, ∆(K) = K ⊗ K
S(K) = K −1, S(E) = −K −1E.
Set E′ := EK −1 ∈ PK−1,1(H). An elementary calculation shows that S2n(E′) =
q−nE′ and K nE′ = qnE′K n for all n ≥ 1. Since K, K −1 and E′ form a set of
generators of H, it is clear that the value of mH depends only on the action of K
on E′. Thus
(1) If q is an nth primitive root of unity for some 1 ≤ n < ∞, G(H) has infinite
exponent, mH = n and S = 2n.
(2) If q is not a root of unity, then G(H) has infinite exponent, mH = ∞ and
S = ∞.
5
3. Preliminary computations
The main results of this section, Proposition 3.3 and Proposition 3.7, along with
their proofs, are almost identical to [7, Proposition 3, Theorem 5], the only differ-
ence being that, for a pointed Hopf algebra H, we state our results in terms of mH ,
instead of the exponent of G(H), and do not restrict ourselves to stating the result
for H being finite-dimensional only.
The results of this section are valid over any field.
Lemma 3.1. Let H be a pointed Hopf algebra. Let x ∈ G(H). Then, for h ∈
Px,1(H), S(h) = −hx−1.
Proof. Let x ∈ G(H) and choose h ∈ Px,1(H), so that ∆(h) = h ⊗ x + 1 ⊗ h. By
Lemma 2.4(2), ǫ(h) = 0. By the counit axiom of the antipode,
S(h)x + h = 0.
That is, S(h) = −hx−1.
(cid:3)
The following lemma is valid over any field.
Lemma 3.2. Let H be a pointed Hopf algebra. Let x ∈ G(H) and let h ∈ Px,1(H).
(1) For any m ≥ 1, S2m(h) = xmhx−m.
(2) If in addition mH < ∞, S2mH (h) = h.
Proof. Fix h ∈ Px,1(H). By Lemma 3.1, S(h) = −hx−1. This gives
(3.1)
S2(h) = −S(x−1)S(h) = xhx−1.
Proceeding inductively, for any m ≥ 1,
S2m(h) = xmhx−m.
It is then clear from the definition of mH that if mH < ∞, S2mH (h) = h.
(cid:3)
Proposition 3.3. Let H be a pointed Hopf algebra with mH < ∞. Then
(S2mH − Id)(H1) = 0.
Proof. Let h ∈ H1. If h ∈ H0 then S2(h) = h since H0 = kG(H) and S2kG(H) = id.
Thus by Theorem 2.3 and linearity, we can without loss of generality assume that
h ∈ Px,y(H) for some x, y ∈ G. Using the fact that ∆ is an algebra homomorphism,
an elementary calculation then shows that hy−1 ∈ Pxy−1,1(H). Then, using Lemma
3.2(2) and the fact that S2mH is an algebra morphism,
hy−1 = S2mH (hy−1) = S2mH (h)S2mH (y−1) = S2mH (h)y−1,
giving S2mH (h) = h, as required.
(cid:3)
Next we shall prove a sufficient condition for the antipode of a pointed Hopf
algebra to have infinite order. Before we do this, we need the following well-known
lemma.
Lemma 3.4. Let H be a Hopf algebra with S < ∞. Then either S = id or
S = 2k for some k ∈ N.
6
PAUL GILMARTIN
Proof. Suppose S 6= id. If H is commutative, S = 2 by [3, Corollary 1.5.12], so
without loss of generality assume that H is noncommutative. Choose x, y ∈ H such
that xy 6= yx. Let S = m < ∞. Suppose m is odd - write m = 2q + 1 for some
q ≥ 1. Since S is an anti-algebra morphism, so too is S2q+1, hence
xy = S2q+1(xy) = S2q+1(y)S2q+1(x) = yx.
This contradicts the assumption that x and y do not commute, thus S must have
even order.
(cid:3)
Proposition 3.5. Let H be a pointed Hopf algebra with mH = ∞. Then S = ∞.
Proof. Suppose S < ∞. By Lemma 3.4, we can write S = 2t for some t ∈ N.
Then by Lemma 3.2(1), for any x ∈ G(H) and h ∈ Px,1(H),
h = S2t(h) = xthx−t.
If mH = ∞, there exists some y ∈ G(H) and f ∈ Py,1(H) such that ytf y−t 6= f .
Thus it must be that mH < ∞.
(cid:3)
Following the arguments of [7], the following result allows us to extend Proposi-
tion 3.3 to higher terms in the coradical filtration.
Proposition 3.6. Let H be a pointed Hopf algebra, let i ≥ 1 and let ψ : H → H be
a coalgebra homomorphism. Suppose (ψ − id)(Hj ) ⊆ Hj−1 for all 0 ≤ j ≤ i. Then
(ψ − id)(Hi+1) ⊆ Hi.
Proof. This is [7, Proposition 4].
(cid:3)
Proposition 3.7. Let H be a pointed Hopf algebra with mH < ∞. Then, for any
n ≥ 1,
(1) (S2mH − id)(Hn) ⊆ Hn−1.
(2) (S2mH − id)n(Hn) = 0.
Proof. This is immediate from Proposition 3.3, Proposition 3.6 and the fact that
S2mH is a coalgebra morphism.
(cid:3)
3.1. Coradically graded Hopf algebras.
Definition 3.8. Let H be a Hopf algebra. We say a family of subspaces {H(n)}n≥0
is a Hopf algebra grading of H if {H(n)}n≥0 is both an algebra and coalgebra
grading with the additional property that S(H(n)) ⊆ H(n) for all n ≥ 0.
If in
addition we have, for each n ≥ 0, that
Hn =
n
M
i=0
H(i)
we say H is a coradically graded Hopf algebra.
Proposition 3.9. Let H be a pointed coradically graded Hopf algebra. Then
(1) mH = ∞ if and only if S = ∞.
(2) If mH < ∞, S = id or S = 2mH .
7
Proof. If mH = ∞, S = ∞ by Proposition 3.5. For the converse, let H =
L∞
i=0 H(i) be a pointed coradically graded Hopf algebra, so that, for n ≥ 1, Hn =
Ln
i=0 H(i), and let mH < ∞. Fix n ≥ 1. By Proposition 3.7, it follows that
(S2mH − id)(H(n)) ⊆
n−1
M
j=0
H(j − 1).
However, since {H(n)}n is a Hopf grading, (S2mH − id)(H(n)) ⊆ H(n) for each
n ≥ 0. It must therefore be that (S2mH −id)(H(n)) = 0 for n ≥ 0, hence S2mH = id.
To complete the proof, it suffices to prove that for any q < mH , there exists h ∈ H
such that S2q(h) 6= h, since Lemma 3.4 guarantees that the order of the antipode is
always either 1 or divisible by 2. Suppose for a contradiction that there exists some
q < mH such that S = 2q. By Lemma 3.2(1), for any x ∈ G(H), h ∈ Px,1(H),
h = S2q(h) = xqhx−q.
However, since q < mH , by definition there exists some y ∈ G and f ∈ Py,1(H)
such that
a contradiction. This completes the proof.
f 6= yqf y−q,
(cid:3)
Let H be a pointed Hopf algebra. As is mentioned at the beginning of §2, the
associated graded space with respect to the coradical filtration of H, gr H, inherits
the structure of a pointed Hopf algebra from H. It is well-known, and proved in [4,
Proposition 4.4.15], for example, that, with respect to the Hopf structure inherited
from H, gr H becomes a pointed coradically graded Hopf algebra. The following is
thus an immediate corollary of Proposition 3.9.
Corollary 3.10. Let H be a pointed Hopf algebra with mH < ∞. Then mgr H < ∞
and either S = id or Sgr H = 2mgr H .
4. The antipode in characteristic zero
We now consider what happens when we work over a field of characteristic 0.
If
Proposition 4.1. Let H be a pointed Hopf k-algebra. Suppose char k = 0.
mH = ∞ then S = ∞. If mH < ∞ then either S divides 2mH, or there exists
h ∈ H such that the orbit of S on h is infinite. In particular, either S divides
2mH or S = ∞.
Proof. We can, without loss of generality, assume that mH < ∞, since otherwise
Proposition 3.5 guarantees that S = ∞. Suppose S2mH 6= id, and choose n
minimal such that h ∈ Hn and S2mH (h) 6= h. We shall prove that S2l(h) 6= h for
any l ≥ 1. By Proposition 3.7(1), S2mH (h) = h + r for some r ∈ Hn−1. By the
choice of h, we can assume r 6= 0.
Claim 4.2. Retain the above notation. For t ≥ 1, S2mH t(h) = h + tr.
Proof. of Claim 4.2: We proceed by induction on t ≥ 1. The t = 1 case was
Proposition 3.7(1). Fix t ≥ 1. Then, by Proposition 3.7(1),
(4.1)
S2(t+1)mH (h) = S2tmH S2mH (h) = S2tmH (h) + S2tmH (r).
8
PAUL GILMARTIN
By the minimality of n, S2tmH (r) = r. By the inductive hypothesis, equation (4.1)
becomes
S2(t+1)mH (h) = (h + tr) + r = h + (t + 1)r
proving the claim by induction.
(cid:3)
So, S2mH t(h) = h + tr for all t ≥ 1. Since char k = 0, this implies S2mH t(h) 6= h
for all t ≥ 1. Thus S2mH = ∞, and so S = ∞. In particular, if i and j are distinct
integers, then Si(h) 6= Sj(h), since otherwise S2mH (i−j)(h) = h, contradicting the
above. This completes the proof.
(cid:3)
Theorem 4.3. Let H be a pointed Hopf k-algebra, where char k = 0. If S < ∞
then either S = 2mH or S = id.
Proof. Suppose S 6= id. By Proposition 4.1 and Lemma 3.4, it suffices to show that
if S < ∞, then, for any q < mH , there exists h ∈ H such that S2q(h) 6= h.
Suppose for a contradiction that there exists some q < mH such that S = 2q.
By Lemma 3.2 (1), for any x ∈ G, h ∈ Px,1(H),
h = S2mH (h) = xqhx−q.
However, since q < mH , by definition there exists some y ∈ G and f ∈ Py,1(H)
such that
f 6= yqf y−q,
a contradiction. This completes the proof.
(cid:3)
Combining Corollary 3.10 and Theorem 4.3, we prove the following result which
connects the order of the antipode of a pointed Hopf algebra H to the order of
the antipode of the associated graded Hopf algebra, gr H. The proof is essentially
exactly the same as the proof of Proposition 4.1.
Theorem 4.4. Let H be a pointed Hopf k-algebra, where char k = 0. If SH < ∞
then SH = Sgr H .
Proof. Suppose S 6= id. If SH < ∞ then mH < ∞ by Proposition 3.5. Let S
denote the antipode of H, let S denote the antipode of gr H and let l = mgr H . By
Corollary 3.10 then it follows that l < ∞ and S = 2l. Let n ≥ 0, h ∈ Hn and let
h = h + Hn−1 ∈ gr H. Then S
(h) = h and so
2l
S2l(h) = h + r
for some r ∈ Hn−1. Then, by exactly the same proof as the one given for Claim
4.2 above, we get that
S2lt(h) = h + tr
for each t ≥ 1. If r 6= 0 then, since char k = 0, S2l = ∞ and hence S = ∞.
However we assumed S < ∞, so we must have r = 0. Thus we have S2l = id and
so S divides 2l. However, S = 2mH by Theorem 4.3, which implies that mH
divides l. Clearly l = mgr H ≤ mH in general, and so it must be that mgr H = mH .
The result follows.
(cid:3)
Corollary 4.5. Let H be a pointed Hopf k-algebra. Assume char k = 0 . Then
9
(1) If G(H) is central in H, then either S = ∞, S = 2 or S = id.
(2) If H is a connected Hopf algebra (i.e. G(H) = {1}), then either S =
∞, S = 2 or S = id.
Proof. Clearly if G(H) is central in H then mH = 1, so (1) follows from Proposition
4.1. Part (2) is immediate from (1).
(cid:3)
Remark 4.6. Note that in Theorem 4.3 all alternatives can occur, even in the
case where H is connected, where mH = 1 always. It is noted in [1, §3.5] that the
connected Hopf algebra B(λ) introduced by Zhuang in [8, §7] has an antipode of
infinite order.
5. The antipode in positive characteristic
In [7, Corollary 6], a bound is obtained for the order of the antipode of a finite
dimensional Hopf algebra H over a field of positive characteristic in terms of the
exponent of G(H). It turns out that the proof of that result is equally valid if we
drop the assumption that H is finite dimensional, and also that, in light of the
results of §3, the result can be restated in terms of mH , rather than the exponent
of G(H). As such, we again credit the results of this section to Taft and Wilson.
Proposition 5.1. Let H be a pointed Hopf k-algebra with mH < ∞. Let n ≥ 1 and
h ∈ Hn. Choose l ∈ N such that pl ≥ n ≥ pl−1. If char k = p > 0, S2mH pl
(h) = h.
Proof. Let n ≥ 1 and h ∈ Hn. Choose l ∈ N such that pl ≥ n ≥ pl−1. By
Proposition Lemma 3.7(2),
0 = (S2mH − id)pl
(h) = S2mH pl
(h) − h.
where the final equality follows from the binomial theorem in characteristic p > 0,
which works here since S and id commute in Endk(H). The result follows.
(cid:3)
The following corollary is immediate.
Corollary 5.2. Let H be a pointed Hopf k-algebra with mH < ∞. Assume char k =
p > 0.
(1) Suppose that H can be generated as an algebra by Hn for some 0 ≤ n < ∞.
Choose l ∈ N such that pl ≥ n ≥ pl−1. Then S divides 2mH pl.
(2) If H is affine (that is, finitely generated as an algebra), then S divides
2mH pn for some 0 ≤ n < ∞. In particular, S < ∞.
Corollary 5.3. Let H be an affine pointed Hopf k-algebra, where char k = p > 0.
(1) Suppose G(H) is central in H. Then there exists some n ≥ 0 such that S
divides 2pn.
(2) Suppose H is connected. Then there exists some n ≥ 0 such that S divides
2pn.
Proof. Part (1) follows immediately from the Corollary 5.2 and fact that if G(H)
is central in H then mH = 1. Part (2) is a special case of (1).
(cid:3)
As shown by the following example, originally due to Taft and Wilson, [6], the
bound on the order of the antipode obtained in Corollary 5.2 is in general not
attained, even in the finite-dimensional connected case.
10
PAUL GILMARTIN
Example 5.4. Let k be a field with char k = p ≥ 3. Let R be the algebra with
generators X, Y and Z subject to the following relations.
[X, Y ] = X [Y, Z] = −Z,
[X, Z] =
X p = 0,
Y p = Y, Z p = 0.
X 2,
1
2
It is proved in [6] that R is a connected Hopf algebra of vector space dimension p3
with coproduct, counit and antipode defined on generators as follows:
ǫ(X) = 0, ∆(X) = 1 ⊗ X + X ⊗ 1,
ǫ(Y ) = 0, ∆(Y ) = 1 ⊗ Y + Y ⊗ 1,
ǫ(Z) = 0, ∆(Z) = 1 ⊗ Z + X ⊗ Y + Z ⊗ 1,
S(X) = −X, S(Y ) = −Y, S(Z) = −Z + XY.
Since R is connected, mR = 1. Morover, notice that X, Y ∈ R1 and Z ∈ R2, so
R can be generated in at least coradical degree 2. If R was generated in coradical
degree one, it would be cocommutative, since R being connected implies R1 =
k ⊕ P (R), [3, Lemma 5.3.2]. Since p ≥ 3, the bound on the order of the antipode
of R as determined by Corollary 5.2 is therefore 2p. A simple calculation yields the
identity
S2t(Z) = Z − tX
for any t ≥ 1. In particular, S2p(Z) = Z. Since S is an anti-algebra morphism, it
follows that S = 2p.
Example 5.5. For an example of a pointed Hopf k-algebra H over a field of positive
characteristic which has an antipode of infinite order, see Example 2.8: when q is
not a root of unity, the Hopf algebra H = Uq(b+) has an antipode of infinite order
over any field.
We know of no example of a connected Hopf algebra H in positive characteristic
with an antipode of infinite order. This prompts the following question.
Question 5.6. Suppose H is a connected Hopf k-algebra, where char k = p > 0.
Does the antipode of H always have finite order?
By Corollary 5.3(2), an example which gives a negative answer to the above
question would be necessarily non-affine. More generally, we could ask the following.
Question 5.7. Suppose H is a pointed Hopf k-algebra, where char k = p > 0 and
mH < ∞. Does the antipode of H always have finite order?
References
[1] K. A. Brown, P. Gilmartin, Quantum Homogeneous Spaces of Connected Hopf Algebras, J.
Algebra 454 (2016) 400 - 432
[2] K. A. Brown, K.R. Goodearl, Lectures on Algebraic Quantum Groups, Birkhauser, 2002.
[3] S. Montgomery, Hopf Algebras and their Actions on Rings, CBMS Regional Conference Series
in Mathematics 82, Amer. Math. Soc., Providence R1, 1993.
[4] D. Radford, Hopf Algebras, Word Scientific Publishing Co. Pte. Ltd., 2012.
[5] D. Radford, The Order of the Antipode of a Finite Dimensional Hopf Algebra is Finite,
American Journal of Mathematics Vol. 98, No. 2 (Summer, 1976), pp. 333-355
[6] E. J.Taft, R. L. Wilson, Hopf Algebras with Nonsemisimple Antipode, Proc. Amer. Math.
Soc. 49 (1975), 269-276.
[7] E. J. Taft, R. L. Wilson, On Antipodes in Pointed Hopf Algebras, J. Algebra 29 (1974),
27-32.
[8] G. Zhuang, Properties of pointed and connected Hopf algebras of finite Gelfand-Kirillov
dimension, J.London Math. Soc. 87 (2013), 877-898.
School of Mathematics and Statistics, University of Glasgow, Glasgow G12 8QW,
Scotland.
E-mail address: [email protected]
11
|
1808.02108 | 1 | 1808 | 2018-08-06T20:56:00 | Cluster automorphisms and quasi-automorphisms | [
"math.RA",
"math.RT"
] | We study the relation between the cluster automorphisms and the quasi-automorphisms of a cluster algebra $\mathcal{A}$. We proof that under some mild condition, satisfied for example by every skew-symmetric cluster algebra, the quasi-automorphism group of $\mathcal{A}$ is isomorphic to a subgroup of the cluster automorphism group of $\mathcal{A}_{triv}$, and the two groups are isomorphic if $\mathcal{A}$ has principal or universal coefficients; here $\mathcal{A}_{triv}$ is the cluster algebra with trivial coefficients obtained from $\mathcal{A}$ by setting all frozen variables equal to the integer 1.
We also compute the quasi-automorphism group of all finite type and all skew-symmetric affine type cluster algebras, and show in which types it is isomorphic to the cluster automorphism group of $\mathcal{A}_{triv}$. | math.RA | math |
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
WEN CHANG AND RALF SCHIFFLER
Abstract. We study the relation between the cluster automorphisms and the quasi-
automorphisms of a cluster algebra A. We proof that under some mild condition,
satisfied for example by every skew-symmetric cluster algebra, the quasi-automorphism
group of A is isomorphic to a subgroup of the cluster automorphism group of Atriv,
and the two groups are isomorphic if A has principal or universal coefficients; here
Atriv is the cluster algebra with trivial coefficients obtained from A by setting all
frozen variables equal to the integer 1.
We also compute the quasi-automorphism group of all finite type and all skew-
symmetric affine type cluster algebras, and show in which types it is isomorphic to the
cluster automorphism group of Atriv.
Contents
Introduction
1.
2. Preliminaries on cluster algebras
3. Automorphisms, weak automorphisms and quasi-automorphisms
4. Cluster automorphism groups and quasi-automorphism groups
5. Finite type
6. Affine type
7. Rank 2
References
1
3
7
12
16
20
24
25
1. Introduction
There are several notions of automorphisms of cluster algebras in the literature. For
cluster algebras with trivial coefficients, cluster automorphisms were introduced in [2]
as Z-algebra homomorphisms that map a cluster to a cluster and commute with the
mutations at that cluster. In the same paper, the authors gave several characterizations
of cluster automorphisms and computed the cluster automorphism groups for finite and
affine types. This notion of automorphisms was later generalized to cluster algebras with
arbitrary coefficients in several ways as follows.
In [1], the authors defined cluster homomorphisms and constructed a category of
(rooted) cluster algebras, and in [7, 8], the resulting cluster automorphism groups were
studied. In this generalization, the cluster automorphisms are required to map the frozen
variables to frozen variables, and this notion turned out to be too restrictive to include
2010 Mathematics Subject Classification. 13F60.
The first author is supported by the NSF of China (Grant No. 11601295) and by Shaanxi Normal
University.
The second author was supported by the NSF-CAREER grant DMS-1254567 and by the University
of Connecticut.
1
2
WEN CHANG AND RALF SCHIFFLER
important symmetries of cluster algebras, notably the twist map on Grassmannians of
[16].
Fraser introduced the less restrictive notions of quasi-homomorphisms and quasi-
automorphisms in [9]. The definition of a quasi-automorphism does not strictly require
that a cluster is mapped to a cluster, but allows for the flexibility that the image of a
cluster is a cluster up to rescaling each cluster variable by some coefficients. Fraser's
definition is quite subtle, for example the set of all quasi-automorphisms does not form a
group in general, see Example 3.8. This led Fraser to construct the quasi-automorphism
group QAut0(A) by considering certain equivalence classes, the proportionality classes,
of quasi-automorphisms.
At this point the following questions are natural.
(1) How does the quasi-automorphism group relate to the cluster automorphism
group?
(2) Can one compute the quasi-automorphism groups of cluster algebras of finite
and affine types?
In order to state our answers to these questions, it will be convenient to introduce
some notation. Given a cluster algebra A, we denote by Atriv the cluster algebra with
trivial coefficients obtained from A by setting all frozen variables equal to the integer 1.
We denote by QAut0(A) the quasi-automorphism group of A and by Aut+(A) the group
of (direct) cluster automorphisms of A. Let Aut+
0 (A) be the quotient group of Aut+(A)
by the subgroup of all automorphisms that fix each cluster variable, also known as the
subgroup of coefficient permutations.
We say that a cluster algebra A satisfies condition (⋆) if A is skew-symmetric, or A is
of finite type, or the exchange matrix of A is non-degenerate. Each of these conditions
guarantees that the exchange graph of A only depends on the principal part of the
exchange matrix B and not on the coefficient system.
Our main result regarding question (1) is the following.
Theorem 1.1 (Theorem 4.9). Let A be a cluster algebra. There is an isomorphism of
groups
QAut0(A) ∼= Aut+(Atriv)
in each of the following cases.
(1) A has principal coefficients and satisfies condition (⋆).
(2) A has universal coefficients.
(3) A is a surface cluster algebra with boundary coefficients.
In general, without imposing any conditions on the coefficients of the cluster algebra
A, we have the following injective group homomorphisms
Aut+
0 (A) ֒→ QAut0(A)
(⋆)
֒→ Aut+(Atriv),
where the first homomorphism always exists, see Proposition 4.3 and the second exists
if A satisfies condition (⋆), see Proposition 4.6. We give examples of finite type cluster
algebras for which these injective homomorphisms are not surjective. Thus the conditions
on the coefficients in the theorem above are necessary.
We then give an affirmative answer to question (2) and compute the quasi-automorphism
groups of all cluster algebras of finite type as well as for all cluster algebras of skew-
symmetric affine type, see Table 1 below.
In particular, we show the following.
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
3
Dynkin type
An(n > 2)
A1
Bn
Cn
D4
D2n(n > 3)
D2n+1
E6
E7
E8
F4
G2
QAut0(A)
Zn+3
Z2
Zn+1
Zn+1
Z4 × G
Z2n × H
Z2n+1 × Z2
Z14
Z10
Z16
Z7
Z4
Affine type
QAut0(A)
eAp,q(p 6= q)
eAp,p
eD4
eDn−1(n > 6)
eE6
eE7
eE8
Hp,q
Hp,p ⋊ G1
Z × G2
H
Z × G3
Z × G1
Z
Table 1. The quasi-automorphism groups for cluster algebras of finite
type (left), where G is a subgroup of S3 and H is a subgroup of Z2, and
for cluster algebras of affine type (right), where G1 is a subgroup of Z2,
G2 is a subgroup of S4, G3 is a subgroup of S3; the group Hp,q is defined
in equation (6.1), and the group H satisfies Z ⊆ H ⊆ G, where G is given
in equation (6.2).
Theorem 1.2. There is an isomorphism of groups
QAut0(A) ∼= Aut+(Atriv)
in each of the following cases.
(1) A is of finite type, but not type Dn with n even.
(2) A is of affine type eAp,q, with p 6= q, type eDn−1, for all n, or type eE8.
(3) A is of rank 2.
For completeness, we also mention yet another notion of automorphisms introduced
by Saleh in [20] as automorphisms of the ambient field that restrict to a permutation of
the set of all cluster variables. The relation between Saleh's notion and that of cluster
automorphisms above was studied in [3] and led to the open question of unistructurality
of cluster algebras, see also [4].
The paper is organized as follows. After recalling some basic notions about cluster
algebras in section 2, we review the definitions of the different automorphisms in sec-
tion 3. In that section, we also introduce weak cluster automorphisms, which lie between
cluster automorphisms and quasi-automorphisms. In section 4, we study the relations
between the different automorphism groups and prove Theorem 4.9. Sections 5 and 6
are devoted to the computation of the quasi-automorphism groups in finite and affine
types, respectively, and section 7 contains the rank 2 case.
2. Preliminaries on cluster algebras
We recall basic definitions and properties on cluster algebras in this section. Through-
out the paper, we use Z as the set of integers and use the notation [x]+ = max(x, 0) for
x ∈ Z.
4
WEN CHANG AND RALF SCHIFFLER
2.1. Labeled seeds versus unlabeled seeds. A labeled seed of rank n is a triple
Σ = (x, p, B), where
• x = (x1, . . . , xn) is an ordered set with n elements;
• p = (xn+1, . . . , xm) is an ordered set with m − n elements;
• B = (bxj xi) ∈ Mm×n(Z) is a matrix labeled by (x ⊔ p) × x, and it is extended
skew-symmetrizable, that is, there exists a diagonal matrix D with positive in-
teger entries such that DB is skew-symmetric, where B is a submatrix of B
consisting of the first n rows.
The ordered set x is the cluster of the labeled seed Σ. The elements of x are called
the cluster variables of Σ and the elements of p are called the coefficient variables of Σ.
We shall write bji for the element bxjxi in B for brevity. The n × n matrix B is called
the exchange matrix of Σ; its rows are called exchangeable rows, and the remaining rows
of B are called frozen rows. We always assume that both B and B are indecomposable
matrices, and we also assume that n > 1 for convenience.
Two labeled seeds Σ = (x, p, B) and Σ′ = (x′, p′, B′) are said to define the same
unlabeled seed if Σ′ is obtained from Σ by simultaneous relabeling of the ordered sets x
and p and the corresponding relabeling of the rows and columns of B.
In an unlabeled seed the cluster x and the coefficients p are sets that are not ordered.
We shall use the notation x = {x1, . . . , xn}, p = {xn+1, . . . , xm} in an unlabeled seed
and the notation x = (x1, . . . , xn), p = (xn+1, . . . , xm) in a labeled seed.
For the remainder of the paper, our seeds will always be labeled seeds unless specified
otherwise.
2.2. Seed mutation. Given an exchangeable cluster variable xk, one may produce a
new labeled seed by a seed mutation.
Definition 2.1. The labeled seed µk(Σ) = (µk(x), p, µk(B)) obtained by mutation of
Σ in direction k is given by:
• µk(x) = (x \ {xk}) ⊔ {x′
k} where
xkx′
k = Y16j6m
xj
[bjk]+ + Y16j6m
xj
[−bjk]+.
• µk(B) = (b′
ji)m×n ∈ Mm×n(Z) is given by
b′
ji =(cid:26) −bji
bji + [−bjk]+bki + bjk[bki]+
if i = k or j = k ;
otherwise.
(2.1)
(2.2)
It is easy to check that the mutation is an involution, that is µkµk(Σ) = Σ. Note that
in a seed mutation, the set p is not changed.
2.3. Base ring ZP and ambient field F . For a finite set p = {xn+1, . . . , xm}, let P
be the free abelian group (written multiplicatively) generated by the elements of p and
define an addition ⊕ in P by
(2.3)
Yj
xaj
j ⊕Yj
xbj
j =Yj
xmin(aj ,bj)
j
.
Then (P, ⊕, ·) is a semifield called tropical semifield. Let ZP be its group ring and QP
its field of quotients. Let F = QP(x1, x2, . . . , xn) be the field of rational functions in n
independent variables with coefficients in QP. F is called the ambient field of the cluster
algebra.
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
5
2.4. Cluster algebra. Recall that a tree is a graph without cycles. An n-regular tree
Tn is a tree in which each vertex has precisely n neighbors. We label the edges of Tn
by 1, . . . , n in such a way that the n edges emanating from each vertex receive different
labels.
A cluster pattern is an assignment of a labeled seed Σt = (xt, pt, Bt) with rank n to
every vertex t ∈ Tn, so that the seeds assigned to the endpoints of any edge labeled by
k are obtained from each other by the seed mutation in direction k. The elements of Σt
are written as follows:
(2.4)
xt = (x1;t , . . . , xn;t) , pt = (xn+1;t , . . . , xm;t) , Bt = (bt
ij) .
Since the frozen variables do not change under mutation, we have pt = pt′ for any
t, t′ ∈ Tn.
Given a cluster pattern on Tn with initial labeled seed Σ = (x, p, B), denote by
(2.5)
X = [t∈Tn
xt = {xi,t : t ∈ Tn , 1 ≤ i ≤ n} ,
the set of all cluster variables in the cluster pattern. Note that the set X does not change
if we use unlabeled seeds instead of labeled seeds.
The cluster algebra A associated with the cluster pattern (or associated with the seed
Σ) is the ZP-subalgebra of the ambient field F generated by X . Thus A = ZP[X ]. The
elements xi,t ∈ X are called cluster variables, P is the coefficient semifield of A, and the
elements in P are called coefficients of A. The cluster algebra is called skew-symmetric
if its initial exchange matrix B is skew-symmetric. The cluster algebra is said to be
of geometric type if P is the tropical semifield defined in section 2.3. In this paper, all
cluster algebras are of geometric type.
2.5. Iced valued quiver. To the matrix B in a seed Σ = (x, p, B), one can associate
an iced valued quiver Q = Q(B), whose mutable vertices are labeled by cluster variables
in x, whose frozen vertices are labeled by the frozen variables in p, and whose arrows
and values are assigned by B (see Example 2.2, we refer to [14] for details). Then the
principal part B of the matrix B corresponds to the full subquiver Q of Q whose vertices
are the mutable vertices.
Example 2.2. Let B be the following matrix, where B is skew-symmetrizable with
skew-symmetrizing diagonal matrix D = diag{2, 2, 1, 1}.
B =(cid:18) B
B′ (cid:19) =
1
0
0
−1 0 −1
2
0
0
0 −2
0
0
0
0
0
2
0
0 −1
The quiver corresponding to B is as follows, where we frame the frozen vertex.
Q(B) = 1
(2,1)
3
/ 2
/ 4
/ 5
2.6. Special coefficient systems. In this subsection, we recall the definitions of several
special choices for the frozen variables p and the frozen rows of B.
/
o
o
/
/
/
/
6
WEN CHANG AND RALF SCHIFFLER
2.6.1. Trivial coefficients. A cluster algebra is said to have trivial coefficients if its initial
seed is of the form Σ = (x, p, B) with p = ∅ and B = B.
If A is an arbitrary cluster algebra with initial seed Σ = (x, p, B), we denote by Atriv
the cluster algebra defined by the initial seed Σtriv = (x, B). Thus Atriv is obtained from
A by setting all frozen variables to 1. We call Atriv the principal part cluster algebra of
A.
2.6.2. Principal coefficients. A cluster algebra A is said to have principal coefficients if it
has a seed ((x1, . . . , xn), (xn+1, . . . , x2n), B) such that B =(cid:0) B
part of B and I is the n×n identity matrix. Principal coefficients were introduced in [13],
where it is shown that from the Laurent expansion of a cluster variable with principal
coefficients one can easily obtain the Laurent expansion of the 'same' cluster variable
with arbitrary coefficients.
I(cid:1), where B is the principal
2.6.3. Universal coefficients. We recall the definition of universal coefficients from [13].
Definition 2.3. Let A and A be cluster algebras of the same rank n over the coeffi-
cient semifields P and P, respectively, with the respective families of cluster variables
(xi;t)i∈[1,n],t∈Tn and (xi;t)i∈[1,n],t∈Tn. We say that A is obtained from A by a coefficient
specialization if
(1) A and A have the same exchange matrices Bt = Bt at each vertex t ∈ Tn;
(2) there is a homomorphism of multiplicative groups ϕ : P → P that extends to a
(unique) ring homomorphism ϕ : A → A such that ϕ(xi;t) = xi;t for all i and t.
In particular, a coefficient specialization from A to itself is a ring homomorphism
ϕ : A → A that fixes every cluster variable and is an endomorphism of the multiplicative
group P.
Definition 2.4. A cluster algebra A has universal coefficients if every cluster algebra
with the same family of exchange matrices (Bt) is obtained from A by a unique coefficient
specialization.
Universal coefficients were constructed for the cluster algebras of finite type in [13]
and for surface type in [18, 19]. We give an example of finite type A2 which is different
from the one given in [13] in Example 3.16.
2.6.4. Gluing free coefficients. A seed Σ = (x, p, B) is said to be gluing free, if any two
frozen rows of B are different. It is not hard to see that a mutation of a gluing free seed
is still gluing free [7]. We say a cluster algebra is gluing free, if its seeds are gluing free.
The gluing free condition is important in the rest of the paper when we study the
cluster automorphism groups and the quasi-automorphism groups.
2.7. Exchange graph. The exchange graph of a cluster algebra is the n-regular graph
whose vertices are the unlabeled seeds of the cluster algebra and whose edges connect the
unlabeled seeds related by a single mutation. The exchange graph of a cluster algebra
is thus a quotient graph of the n-regular tree. It is conjectured by Fomin-Zelevinsky in
[11] that the exchange graph does not depend on the frozen part of the matrix. This
conjecture is proved for several types of cluster algebras which we list in the following
lemma.
Lemma 2.5. Let A be a cluster algebra with initial seed Σ = (x, p, B). Then the
exchange graph only depends on the exchange matrix B if one of the following three
conditions holds.
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
7
(⋆1) A is of finite type [12];
(⋆2) A is skew-symmetric [6];
(⋆3) B is non-degenerate [15].
We shall say that the cluster algebra satisfies the condition (⋆) if one of the three condi-
tions holds.
3. Automorphisms, weak automorphisms and quasi-automorphisms
In this section, we recall the notions of cluster automorphisms [2, 7] and quasi-
automorphisms [9], and we also introduce the notion of weak automorphisms. We study
the relation between these notions in section 4.
3.1. Quasi-homomorphisms. In this subsection, we recall Fraser's definition of quasi-
homomorphisms from [9]. We first need to recall the definition of y-variables from [13].
Definition 3.1. Let Σ = (x, p, B) be a seed with
(3.1)
x = (x1, . . . , xn), p = (xn+1, . . . , xm), B = (bji).
Define y = (y1, . . . , yn) and y = (y1, . . . , yn) to be the n-tuples given by
(3.2)
yi = Yn+1≤j≤m
xbji
j
and
yi = yi Y1≤j≤n
xbji
j
.
The elements in y are called the Y -variables and the elements in y the hatted Y-variables
of the seed Σ.
Now we recall the proportionality relation between two seeds; we do not use the
original definition but rather an equivalent condition, see [9, Proposition 2.3].
Definition 3.2. (1) Two elements x, y ∈ F are proportional, written x ≍ y, if the
quotient x
y lies in P.
(2) Two clusters xt and xt′ are proportional, written xt ≍ xt′, if xi;t ≍ xi;t′ for all
1 ≤ i ≤ n.
(3) Two labeled seeds Σt = (xt, pt, Bt) and Σt′ = (xt′, pt′ , Bt′) are proportional,
written Σt ≍ Σt′, if for all 1 ≤ i ≤ n
xi;t ≍ xi;t′,
yi;t = yi;t′
and
Bt = Bt′.
Lemma 3.3. Let A be a cluster algebra that satisfies the (⋆) condition in Lemma 2.5,
then
(1) two clusters x and x′ are proportional if and only if they are equal;
(2) two seeds Σ and Σ′ are proportional if and only if they are equal.
Proof. (1) If x and x′ are proportional, we see that π(x) = π(x′) under the coefficient
specialization π from A to Atriv, which maps all the coefficients to 1. On the other hand,
since the cluster algebras we consider are of geometric type, the clusters determine the
seeds [15]. On the other hand, by Lemma 2.5 the algebras A and Atriv have the same
exchange graph, thus we have x = x′.
(2) This follows from the first part of the lemma.
(cid:3)
We are ready for the main definition of this subsection. Again, we do not use the
original definition but rather an equivalent condition, see [9, Proposition 3.2].
8
WEN CHANG AND RALF SCHIFFLER
Definition 3.4. Let A and A be two cluster algebras of the same rank n with co-
efficient semifields P and P and with ambient fields F and F , respectively. A quasi-
homomorphism is a field homomorphism Ψ : F → F such that Ψ(P) ⊂ P and there
exists a seed Σ = (x, p, B) of A and a seed Σ = (x, p, B) of A such that
(3.3)
Ψ(Σ) ≍ Σ,
where Ψ(Σ) = (Ψ(x), Ψ(p), B) is the triple obtained by evaluating Ψ on Σ.
Remark 3.5. It is shown in [9] that if Ψ is a quasi-homomorphism then the condition
that Ψ(Σ′) is proportional to a labeled seed of A actually holds for every seed Σ′ in A.
The following result gives a useful interpretation of quasi-homomorphisms in terms of
matrices.
Lemma 3.6 ([9, Corollary 4.5]). Let A and A be two cluster algebras, and let Σ =
((x1, . . . , xn), (xn+1, . . . , xm), B) be a seed in A and Σ = ((x1, . . . , xn), (xn+1, . . . , xm),B)
a seed in A.
(a) If ψ : A → A is a quasi-homomorphism such that ψ(Σ) ≍ Σ then there are integers
mij such that
(3.4)
Ψ(xj) =
x mij
i
,
mYi=1
for j = 1, . . . , m. Moreover, the m × m integer matrix Mψ = (mij) satisfies
(3.5)
B = MΨB.
(b) There exists a quasi-homomorphism ψ : A → A such that ψ(Σ) ≍ Σ if and only
if the principal parts of B, B agree, and the integer row span of B contains the integer
row span of B.
3.2. Quasi-automorphisms. In this subsection, we recall the definition and properties
of the quasi-automorphism group from [9].
Definition 3.7. Let A and A be two cluster algebras.
(1) Two quasi-homomorphisms Ψ1, Ψ2 from A to A are called proportional if
(3.6)
Ψ1(Σ) ≍ Ψ2(Σ)
for some (hence every) seed Σ of A.
(2) A quasi-homomorphism Ψ from A to A is said to be a quasi-isomorphism if there
is a quasi-homomorphism Φ from A to A such that Φ ◦ Ψ is proportional to the identity
map on A. In this case Ψ and Φ are quasi-inverses of one another.
(3) A quasi-automorphism of A is a quasi-isomorphism from A to itself. Let QAut(A)
denote the set of all quasi-automorphisms of A.
The following example shows that the set QAut(A) is not a group.
Example 3.8. Let A be the cluster algebra of type A2 with initial seed Σ = ((x1, x2),
(x3), B), where the matrix B and its corresponding quiver are as follows
Q(B) = 3
/ 1
/ 2
B =
0
1
−1 0
3
0
/
/
/
/
/
/
B′ =
0
−1
1
0
0 −3
Q(B′) = 1
/ 2
/ 3 .
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
9
and where the framed vertex corresponds to the frozen variable. Mutating this seed in
direction 1 and 2 produces the seed Σ′ = ((x′
3)/x1
and x′
2), (x3), B′), where x′
1 = (x2 + x3
1, x′
2 = (x2 + x3
3 + x1x3
3)/x1x2 and
Define Ψ by
Ψ(x1) = x′
1x−3
3 , Ψ(x2) = x′
2x6
3
and Ψ(x3) = x2
3.
2 ∈ P, and
2 x3
Then Ψ(x1)/x′
1 ∈ P and Ψ(x2)/x′
, x′
so Ψ(Σ) ≍ Σ′ and thus Ψ is a quasi-homomorphism.
Ψ(y1, y2) = Ψ(x−1
3, x1) = (x′−1
2
Define a quasi-inverse Φ by
1) = x1x6
3, Φ(x′
Then ΦΨ(x1)/x1, ΦΨ(x2)/x2 ∈ P and
Φ(x′
2) = x2x−3
3
1x−3
3 ) = (by′
1,by′
2);
and Φ(x3) = x2
3.
ΦΨ(y1, y2) = ΦΨ(x−1
2 x3
3, x1) = Φ(x′
2
−1, x′
1x−3
3 ) = (x−1
2 x3
3, x1) = (y1, y2).
This shows that Φ◦Ψ is proportional to the identity and thus Ψ is a quasi-automorphism.
Indeed, if Ψ had an inverse Ψ−1 then
but this is not an
Note that Ψ is not invertible in QAut(A).
3), which would imply Ψ−1(x3) = x1/2
x3 = Ψ−1Ψ(x3) = Ψ−1(x2
element of P. Thus QAut(A) is not a group.
3
It is shown in [9] that the set of proportionality classes of quasi-automorphisms does
form a group under composition. This leads to the following definition.
Definition 3.9. The quasi-automorphism group QAut0A is the group of proportionality
classes of quasi-automorphisms of A.
Remark 3.10. For a matrix B, let Lat(B) denote the integer lattice generated by the
rows of B. Let A be a cluster algebra, and Σ, Σ′ be seeds in A. Then by Lemma 3.6, Ψ is
a quasi-automorphism of A with Ψ(Σ) ≍ Σ′ if and only if B = B′ and Lat(B) = Lat(B′).
The following lemma will be useful later.
Lemma 3.11. Let B =(cid:0) B
matrix obtained from B after a sequence of mutations ω. If Lat(B) = Lat(B′) for every
matrix M with m = 1, then Lat(B) = Lat(B′) for every matrix M with m > 1.
M(cid:1) be a (n + m) × n matrix and B′ = µω(B) =(cid:16) B ′
M ′(cid:17) be a
Proof. Suppose m ≥ 1 and let β, β′ be any rows in M, M ′ respectively. Then our
assumption implies that β ∈ Lat(B′), and β′ ∈ Lat(B), and thus every row of M lies in
Lat(B′) and every row of M ′ lies in Lat(B). By our assumption, we also have every row
of B lies in Lat(B′) and every row of B′ lies in Lat(B). Therefore Lat(B) = Lat(B′). (cid:3)
3.3. Cluster automorphisms and weak cluster automorphisms. In this subsec-
tion we recall the definition of cluster automorphisms. For cluster algebras with trivial
coefficients this definition is due to [2] and for cluster algebras with arbitrary coefficients
to [7]. We also introduce the notion of weak cluster automorphisms.
Definition 3.12. Let A be a cluster algebra.
(1) A Z-algebra automorphism f : A → A is called a cluster automorphism if there
exists a seed (x, p, B) such that
/
/
/
/
/
/
10
WEN CHANG AND RALF SCHIFFLER
(i) f (x) is a cluster,
(ii) f (p) = p as unordered sets,
(iii) f commutes with mutations, that is, for every x, x′ ∈ x,
(3.7)
f (µx,x(x′)) = µf (x),f (x)(f (x′)).
(2) A Z-algebra automorphism f : A → A is called a weak cluster automorphism if
there exists a seed (x, p, B) such that conditions (i) and (iii) above are satisfied and
(ii') f (p) ⊂ P.
Thus a weak cluster automorphism is allowed to map a frozen variable xn+i to an
arbitrary element of P, while a cluster automorphism must map frozen variables to
frozen variables. The following lemma is obvious.
Lemma 3.13. Every cluster automorphism is a weak cluster automorphism.
(cid:3)
We show that the converse of the lemma does not hold in Example 3.16 below.
It is shown in [2, Proposition 2.4] that the condition (3.7) holds for one cluster if and
only if it holds for every cluster. In our case, the cluster algebra is of geometric type, so
the clusters determine the seeds [15]. Denote by B and B′ the matrices of x and f (x)
respectively. We say B ∼= B′, if B′ is obtained from B by simultaneous relabeling of the
exchangeable rows and corresponding columns and the relabeling of the frozen rows.
If f is a cluster automorphism then B ∼= B′ or B ∼= −B′, and if f is a weak cluster
automorphism then we have the same relation for the principal parts of the matrices,
that is, B ∼= B′ or B ∼= −B′. For cluster automorphism with trivial coefficients, this is
shown in [2, Lemma 2.3] and the same proof generalizes to the other cases.
A cluster automorphism such that B ∼= B′ is called a direct cluster automorphism
and a cluster automorphism such that B ∼= −B′ is an inverse cluster automorphism.
Similarly, a weak cluster automorphism such that B ∼= B′ is called a direct weak cluster
automorphism and a weak cluster automorphism such that B ∼= −B′ is an inverse weak
cluster automorphism.
Definition 3.14. (1) Let Aut(A) be the group of all cluster automorphisms of A, and
let Aut+(A) be the subgroup of Aut(A) of all direct cluster automorphisms.
(2) Let WAut(A) be the group of all weak cluster automorphisms of A, and let
WAut+(A) be the subgroup of WAut(A) of all direct cluster automorphisms.
If two frozen rows in a matrix of a cluster algebra A coincide, then exchanging the
corresponding frozen variables induces a cluster automorphism of A. Note that the
cluster variables and the clusters are not changed under such a cluster automorphism.
Such an automorphism is called a coefficient permutation.
0 (A), WAut0(A) and WAut+
0 (A) the quotient groups of
Aut(A), Aut+(A), WAut(A) and WAut+(A) by the respective subgroup of all coeffi-
cient permutations. If A is a gluing free cluster algebra, then Aut0(A) ∼= Aut(A) and
Aut+
Denote by Aut0(A), Aut+
0 (A) ∼= Aut+(A).
Remark 3.15. By definition, a cluster automorphism f maps a cluster to a cluster,
and induces an automorphism of the exchange graph as well as an automorphism of the
n-regular tree.
Example 3.16 (A week cluster automorphism that is not a cluster automorphism). Let
A be the cluster algebra of type A2 given by the initial seed ((x1, x2), (x3, x4, x5, x6, x7), B),
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
11
where the matrix B and its corresponding quiver are as follows
B =
1
0
−1
0
1
2
1
1
−1
0
0 −1
1 −1
Q(B) =
5
3
1
7
4
/ 2
❃❃❃❃❃❃❃❃❃
6
The framed vertices in the quiver are frozen. Then the five cluster variables of A are
x1, x2, x′
1, x′
2, x′′
1, where
x′
1 =
x2
3x4x7 + x5x2
x1
, x′
2 =
x3x4x1 + x6x7
x2
, x′′
1 =
x2
3x4x6x7 + x3
4x1 + x5x6x2
3x2
x1x2
.
Mutation at x1 produces the seed ((x′
1, x2), (x3, x4, x5, x6, x7), B) with
B′ =
0 −1
0
1
−2
3
−1
2
1
0
0 −1
−1
0
Q(B′) =
5
3
/ 1
7
Define a Z-algebra automorphism τ : A → A by
❃❃❃❃❃❃❃❃❃
@
❃❃❃❃❃❃❃❃❃
❃❃❃❃❃❃❃❃❃
4
2
6
x1 7→ x2,
x7 7→ x2
3x4.
x2 7→ x′
1, x3 7→ x3x4x−1
5 , x4 7→ x−1
3 x−1
4 x2
5, x5 7→ x6, x6 7→ x7,
A direct computation shows that τ is the following permutation of the 5 cluster variables
x1 7→ x2 7→ x′
1 7→ x′′
1 7→ x′
2 7→ x1,
and it is a weak cluster automorphism of order 5. However τ is not a cluster automor-
phism since B′ ≇ ±B.
We define another Z-algebra automorphism σ : A → A by
x1 7→ x2, x2 7→ x1, x3 7→ x−1
x7 7→ x7.
5 x6, x4 7→ x2
5x−1
6 , x5 7→ x3x4, x6 7→ x2
3x4,
Another direct computation shows that σ is the following permutation
x1 ↔ x2,
1 ↔ x′
x′
2,
1 ↔ x′′
x′′
1,
and it is a weak cluster automorphism of order 2. Again, σ is not a cluster automorphism.
Note that τ σ = στ −1, so hτ, σi is the dihedral group D5. On the other hand, we have
WAut(A) = WAut0(A) ⊆ Aut(Atriv) ∼= D5, where the last isomorphism is computed in
[2]. Thus WAut(A) = hτ, σi ∼= D5. On the other hand, Aut(A) = Aut0(A) = {id}.
o
o
/
O
O
/
O
O
O
O
@
o
o
12
WEN CHANG AND RALF SCHIFFLER
4. Cluster automorphism groups and quasi-automorphism groups
In this section, we consider the relations between the cluster automorphism groups
and the quasi-automorphism groups. First, we reveal the relations between these at the
matrix level.
Lemma 4.1. Let A be a cluster algebra of a cluster pattern t 7→ Σt over a coefficient
semifield P. Let Ψ : A → A be a quasi-automorphism with Ψ(Σt) ≍ Σt′. Let MΨ;t be the
matrix in Lemma 3.6 satisfying Bt′ = MΨ;tBt. Then
(4.1)
MΨ;t =(cid:18) I
M1;t M2;t (cid:19) ,
0
where I is the identity matrix of rank n, M1;t is a (m − n) × n integer matrix, and M2;t
is a (m − n) × (m − n) integer matrix. Conversely, each matrix M of this form defines
a quasi-automorphism.
Moreover
(1) Ψ is a cluster automorphism if and only if M1;t = 0 and M2;t is a permutation
matrix for some (and thus every) t ∈ Tn;
(2) Ψ is a weak cluster automorphism if and only if M1;t = 0 and B2,t′ = M2;tB2,t for
In particular, there are injective maps given by inclusions of sets
B2,t(cid:17) and Bt′ =(cid:16) Bt′
B2,t′(cid:17).
Aut+(A) ֒→ WAut+(A) ֒→ QAutA.
every t ∈ Tn, where Bt =(cid:16) Bt
Proof. Recall that Ψ(xj,t) = Qm
(4.2)
Ψ(xj;t)
xj;t′
= Qm
i=1 xmij
xj;t′
i,t′
∈ P.
homomorphism, Ψ(xj;t) ≍ xj;t′ for all 1 ≤ j ≤ n, that is,
i=1 xmij
i,t′ , for all 1 ≤ j ≤ m. Since Ψ is a quasi-
Thus the top left block of the matrix M is the identity matrix and the block M1;t
determines the quotient in (4.2). On the other hand, if n + 1 ≤ j ≤ m, we have
i,t′ ∈ P, since Ψ maps coefficient semifield to itself, whence the top
i=1 xmij
Ψ(xj,t) = Qm
right block is the zero matrix.
(1) Note that M1;t = 0 if and only if Ψ(xt) is a cluster, M2;t is a permutation matrix
if and only if Ψ(p) = p as unordered sets. Moreover, under the above two conditions,
is equivalent to the equality (3.7). Then comparing the definition of
Ψ(yi;t) = yi;t′
quasi-homomorphisms and the definition of cluster automorphisms, we are done.
(2) A weak cluster automorphism maps cluster variables to cluster variables; this is
(cid:3)
equivalent to M1;t = 0 and B2,t′ = M2;tB2,t for every t ∈ Tn.
Remark 4.2. The matrix M2;t is determined by the action of Ψ on the coefficients
semifield P and therefore it is independent of t ∈ Tn. Thus the cases (1) and (2) of the
lemma imply that if Ψ is a cluster automorphism or a weak cluster automorphism then
the matrix MΨ,t does not depend on t ∈ Tn. In these cases, we shall denote the matrix
simply by MΨ. However, this is not true for the general quasi-homomorphism. For a
quasi-automorphism Ψ, it seems interesting to consider the relations between MΨ,t for
different t ∈ Tn. That is, consider the action of the seed mutations on the matrix MΨ,t.
Proposition 4.3. Let A = A(B) be a cluster algebra.
(1) There are injective group homomorphisms
Aut+
0 (A) ֒→ WAut+
0 (A) ֒→ QAut0(A).
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
13
(2) If A satisfies the (⋆) condition then there are injective group homomorphisms
Aut+
0 (A) ֒→ WAut+
0 (A) ֒→ Aut+(Atriv).
Proof. (1) This immediately follows from Lemma 4.1.
(2) The existence of the first homomorphism follows from part (1). To prove the
existence of the second, let Ψ ∈ WAut+(A), Σ = (x, p, B) be a seed in A and denote by
Σ′ = (x′, p, B′) its image under Ψ. Then B ∼= B′. When we specialize all coefficients to
1, we see that Ψ induces a direct cluster automorphism Ψ on Atriv. Moreover, if Ψ is a
coefficient permutation then Ψ is the identity, and therefore the map Ψ 7→ Ψ induces a
group homomorphism WAut+
0 → Aut+(Atriv).
Now assume that Ψ′ ∈ WAut+(A) is another weak automorphism such that Ψ = Ψ′.
Then, obviously, Ψ and Ψ′ induce the same automorphism on the exchange graph of
Atriv. But since A satisfies the (⋆) condition, Lemma 2.5 implies that A and Atriv have
isomorphic exchange graphs. In particular, Ψ − Ψ′ fixes each cluster variable and hence
Ψ = Ψ′ in WAut0(A). Thus the homomorphism Ψ 7→ Ψ is injective.
(cid:3)
In the case where the cluster algebra has universal coefficients, we have the following
result, which will be used in the proof of Theorem 4.9.
Proposition 4.4. Let A be a cluster algebra with universal coefficients. Then there are
injective group homomorphisms
Aut+(Atriv) ֒→ WAut+
0 (A) ֒→ QAut0(A).
Proof. The existence of the second homomorphism follows from Proposition 4.3 (1). To
show the existence of the first, let f ∈ Aut+(Atriv), Σ = (x, B) be a seed in A and
Σ′ = (x′, B′) its image under f . Then B ∼= B′. Choose two lifts Σ = (x, p, B) and
Σ′ = (x′, p, B′) of Σ and Σ′ in A. We consider a second copy of A which we denote
by A′ with the difference that we use Σ as the initial seed for A and Σ′ as the initial
seed for A′. Since both A and A′ have universal coefficients, there exist two coefficient
specializations Ψ : A → A′ and Ψ′ : A′ → A such that Ψ(Σ) = Σ′ and Ψ′(Σ′) = Σ.
Moreover, the composition Ψ′ ◦ Ψ is a coefficient specialization from A to itself that
fixes Σ. By the uniqueness of the coefficient specialization for universal coefficients,
we see that Ψ′ ◦ Ψ is the identity on A. Similarly, Ψ ◦ Ψ′ is the identity too.
In
particular, Ψ is a Z-automorphism of A. Moreover, Ψ(x) = x′ is a cluster and Ψ(p) ∈ P.
Also Ψ commutes with mutations, since it maps all clusters to clusters. Thus Ψ is a
weak cluster automorphism and the mapping f 7→ Ψ defines a group homomorphism
Aut+(Atriv) → WAut+
0 (A). This homomorphism in injective with left inverse given by
the map Ψ → Ψ obtained by specializing all coefficients to 1.
(cid:3)
Example 4.5. One can show that the cluster algebra A in Example 3.16 has universal
coefficients, although this realization of universal coefficients is different from the one
given in [13]. We have seen WAut(A) = WAut0(A) is isomorphic to the dihedral group
D5 and in this case it is also isomorphic to Aut(Atriv), see [2].
Now we consider the relations between the quasi-automorphism group and the cluster
automorphism group.
Proposition 4.6. Let A be a cluster algebra that satisfies the (⋆) condition of Lemma
2.5, then we have injective group homomorphisms
Moreover, if A is gluing free, then Aut+(A) ֒→ QAut0(A).
Aut+
0 (A) ֒→ QAut0(A) ֒→ Aut+(Atriv).
14
WEN CHANG AND RALF SCHIFFLER
Proof. We start by proving the existence of the first homomorphism. By Lemma 4.1,
the inclusion h : Aut+(A) ֒→ QAut(A) is an injective map of sets. Since coefficient
permutations map each seed Σt to itself, and therefore are proportional to the identity
on A, the map h induces a homomorphism of groups h0 : Aut+
0 (A) → QAut0(A). To
show that h0 is injective, suppose we have Ψ ∈ Aut+
0 (A) such that h0(Ψ) = 1QAut0(A).
Thus h0(Ψ) is proportional to the identity, which implies Ψ(xi;t) = pi;txi;t with pi;t ∈ P,
for any 1 ≤ i ≤ n and t ∈ Tn. Since Ψ ∈ Aut+(A), we see that (p1,tx1;t, · · · , pn,txn;t)
is a cluster of A. By our assumption, the (⋆) condition is satisfied, hence Lemma 3.3
implies that the two clusters are the same, and thus pi,t = 1 for all i, t. Therefore Ψ
preserves the cluster variables and thus Ψ is the identity in QAut0(A).
0
To show the existence of the second homomorphism, let Ψ ∈ QAut(A) and MΨ =
(cid:0) I
M1 M2(cid:1) be the matrix described in Lemma 4.1. Specializing all coefficients to 1,
we obtain a quasi-automorphism Ψtriv ∈ QAut(Atriv) with matrix MΨtriv = I. By
Lemma 4.1, we also have Ψtriv ∈ Aut+(Atriv) and the rule Ψ 7→ Ψtriv gives a map
h : QAut(A) → Aut+(Atriv). Moreover Ψ is proportional to the identity on A if and
only if h(Ψ) is the identity on Atriv. Thus h induces an injective homomorphism
QAut0(A) ֒→ Aut+(Atriv).
Finally, the last statement of the lemma follows from the observation that if A is
(cid:3)
gluing free then Aut+
0 (A) ∼= Aut+(A).
In general, the two injective homomorphisms of the previous proposition are not
surjective. We give two examples below; we also refer the reader to Example 6.9 in [9].
Example 4.7. Let A be the cluster algebra of type A3 given by the seed Σ = ((x1, x2, x3),
(x4), B), where the matrix B and its quiver Q(B) are as follows
B =
B′ =
1
0
0
−1 0 −1
0
0
0
1
1
0
1
0
0
−1 0 −1
1
0
0
0 −1
0
Q(B) = 1
/ 2
3
4 .
Q(B′) = 1
/ 2
3
/ 4 .
After a sequence of mutations µ3µ1µ2 on Σ (the mutation order is µ2, µ1 and then µ3),
we obtain a new seed Σ′ = ((x′
3), (x4), B′), where
1, x′
2, x′
Since B ≇ B′ (or equivalently Q(B) ≇ Q(B′)), there is no direct cluster automorphism
on A which maps Σ to Σ′. However, since B = B′ and Lat(B) = Lat(B′), there exists
a quasi-automorphism which maps Σ to a seed proportional to Σ′. In fact the map
f :(xi 7→ x′
x4 7→ x−1
4
i
if i = 1, 2, 3;
induces such a quasi-automorphism. Thus
Aut+
0 (A) ≇ QAut0(A).
/
o
o
o
o
/
o
o
/
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
15
Example 4.8. Let A be a cluster algebra of D4 type with a seed Σ = ((x1, x2, x3, x4), (x5), B),
where
B =
0
−1 0 0
−1 0 0
1
0 0
1 0
0
1 1 −1
0
0
0
0
.
Q(B) =
5
/ 2
3
]❀❀❀❀❀❀❀❀
1
4 .
A permutation of the cluster variables x2 and x3, produces another seed Σ′ = ((x1, x3, x2, x4), (x5), B′),
where
B′ =
0
−1 0 0
−1 0 0
0 0
1
0
0 1
1 1 −1
0
0
0
0
Q(B′) =
5
/ 3
2
]❀❀❀❀❀❀❀❀
1
4 .
The last row of B′ is not contained in the integer row span of B, thus Lat(B) 6=
Lat(B′), and there is no quasi-automorphism that maps Σ to a seed proportional to Σ′.
However, changing cluster variables x2 and x3 induces a direct cluster automorphism on
Atriv which maps Σ to Σ′. Thus
QAut0(A) ≇ Aut+(Atriv).
4.1. Main result. In view of Example 4.8 it is natural to ask under which conditions we
do have an isomorphism between the groups QAut0(A) and Aut+(Atriv). The following
theorem gives three such conditions, showing that the groups are isomorphic for a large
class of cluster algebras.
Theorem 4.9. Let A be a cluster algebra. There is an isomorphism
QAut0(A) ∼= Aut+(Atriv)
in each of the following cases.
(1) A has principal coefficients and satisfies the (⋆) condition of Lemma 2.5.
(2) A has universal coefficients.
(3) A is a surface cluster algebra with boundary coefficients.
Proof. (1) We will show that the injective homomorphism h0 : QAut0(A) ֒→ Aut+(Atriv)
from Proposition 4.6 is surjective. Let Ψ ∈ Aut+(Atriv), let Σtriv be a seed in Atriv and
let Σ′
j = Ψ(xj) the image of the
cluster variable xj ∈ Σ under Ψ. Since A satisfies the condition (⋆), Lemma 2.5 implies
that there are unique seeds Σ and Σ′ in A that specialize to Σtriv and Σ′
triv, respectively,
triv = Ψ(Σtriv) be its image under Ψ. Also denote by x′
respectively. It was shown in [17, Theorem 1.2] that, since A has principal coefficients,
the matrices G and G′ are both integer matrices with determinate ±1. In particular,
they are invertible and we can define the following matrix
when coefficients are sent to 1. Let B =(cid:0) B
M =(cid:18) I
G(cid:1) and B′ =(cid:16) B ′
0 G′G−1 (cid:19) .
0
G′(cid:17) be the matrices of Σ and Σ′
/
]
o
o
/
]
o
o
2nYi=1
eΨ(xj) =
(x′
i)mij = x′
j
(x′
i)mij .
2nYi=n+1
16
WEN CHANG AND RALF SCHIFFLER
According to Lemma 4.1, the matrix M defines a unique quasi-automorphismeΨ mapping
Σ to Σ′ by
Ψ(xj). Hence Ψ is in the image of h0 and thus h0 is surjective.
Now, since the homomorphism h0 maps frozen variables to 1, we have h0(eΨ)(xj) = x′
(2) According to Proposition 4.4 and Proposition 4.6 there are injective homomor-
phisms Aut+(Atriv) ֒→ QAut0(A) and QAut0(A) ֒→ Aut+(Atriv), both of which are
induced by inclusions of sets. Thus QAut0(A) ∼= Aut+(Atriv).
j =
(3) Cluster algebras from surfaces are skew-symmetric. Moreover, using our assump-
tion that the rank of a cluster algebra is at least two, it follows from [7, Proposition 3.10]
that the surface cluster algebra is gluing free. Therefore we can apply Proposition 4.6
which yields injective homomorphisms Aut+
0 (A) ֒→ QAut0(A) ֒→ Aut+(Atriv). On
0 (A) ∼= Aut+(Atriv). Therefore
the other hand, Theorem 3.18 in [7] shows that Aut+
QAut0(A) ∼= Aut+(Atriv).
(cid:3)
Remark 4.10. Part (3) of the theorem has an interpretation in terms of the tagged
mapping class group MG⊲⊳(S, M ) of the surface (S, M ) introduced in [2]. For all surfaces
with the exceptions of a sphere with four punctures, a once-punctured square, or a digon
with one or two punctures, we have Aut+(Atriv) ∼= MG⊲⊳(S, M ) (this was conjectured in
[2] and proved in [5]). Thus we see that, unless (S, M ) is one of these four exceptions, the
quasi-automorphism group of the cluster algebra of (S, M ) with boundary coefficients is
isomorphic to the tagged mapping class group. This result recovers a very special case
of Fraser's Theorem [9, Theorem 7.5].
5. Finite type
In this section we will show that QAut0(A) ∼= Aut+(Atriv) for finite type cluster
algebra A with arbitrary coefficients, except for type Dn with n even.
Let A be a cluster algebra of finite type and let Σ = (x, p, B) be a seed in A such
that the valued quiver Q of B is bipartite. Denote the sources of the quiver Q by
i1, · · · , is, and its sinks by is+1, · · · , in. Following [10], we define τ+ = µis · · · µi1 and
τ− = µin · · · µis+1 the compositions of seed mutations at sources and sinks respectively.
Since mutations at any two sources (sinks) commute with each other, the definition of
τ± is independent of the order of the mutations. Then the group generated by τ± is
a dihedral group which is considered in [10] to prove the periodicity conjecture of Y -
systems. This group is also closely related to the cluster automorphism groups, see for
example [2, 8]. We define τ = τ+τ−. It corresponds to the Auslander-Reiten translation
on the corresponding cluster category
Lemma 5.1. Let A be a cluster algebra of finite type with arbitrary coefficients and
assume that A is not of type D2ℓ. Let Σ = (x, p, B) and Σ′ = (x′, p, B′) be bipartite
seeds of A such that B′ = B. Then Lat(B′) = Lat(B).
Proof. By Corollaries 3.3(1) and 3.6(2) of [8], there exists m ∈ Z such that τ mΣ = Σ′. So
it suffices to show the statement for Σ′ = τ −1Σ. According to Lemma 3.11, it is sufficient
to consider the case where the matrix B has only one frozen row. Denote by α1, · · · , αn, β
the rows of B, and denote by α1, · · · , αn, β′ the rows of B′. Let β = (b1, · · · , bn) and let
β′ = (b′
n). Then we have to show β′ ∈ Lat(B) and β ∈ Lat(B′).
1, · · · , b′
Without loss of generality, we assume the vertex 1 is a source of Q. We will distinguish
several cases.
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
17
Case I: Type An. Suppose first that n is even. We have
B =
0
−1
0
1
0
0 −1
1
. . .
0
. . .
. . .
0
1
. . .
. . .
. . .
. . .
. . .
. . .
. . . −1 0
1
1
0 −1 0
0
.
Let ei = (0, . . . , 0, 1, 0, . . . , 0) the i-th standard basis vector of Zn. Then
ei =(cid:26) −αi+1 + αi+3 − αi+5 · · · ± αn
−αi−1 + αi−3 − αi−5 · · · ± α1
if i is odd;
if i is even.
Therefore Lat(B) = Zn. So Lat(B) = Zn = Lat(B′) and the result is true.
Now suppose that n is odd, say n = 2ℓ + 1 ≥ 3. Then
1
0
0
−1 0 −1
. . .
. . .
1
B =
. . .
. . .
. . .
. . .
. . .
−1 0 −1
0
1
1
.
−b1 + [b2 + [b1]+ + [b3]+]+
−bi − [bi−1]+ − [bi+1]+
−bi + [bi+1 + [bi]+ + [bi+2]+]+
+[bi−1 + [bi−2]+ + [bi]+]+
−bn + [bn−1 + [bn−2]+ + [bn]+]+ i = n.
i = 1;
i even;
i odd and i 6= 1, n;
A direct calculation shows that
b′
i =
So
ℓXk=0
β′ = −β −
[b2k+1]+α2k+1 −
[b2k + [b2k−1]+ + [b2k+1]+]+α2k.
ℓXk=1
Therefore β′ ∈ Lat(B) and β ∈ Lat(B′).
Case II: Type Bn. In this case we assume that β = ( b1
2 , · · · , bn) and β′ = (b′
1, · · · , b′
n).
If n is odd, say n = 2ℓ + 1 ≥ 3, we have
2
0
0
−1 0 −1
. . .
. . .
1
. . .
. . .
. . .
B =
.
. . .
. . .
−1 0 −1
0
1
1
18
Then
So
b′
i =
WEN CHANG AND RALF SCHIFFLER
2 b1 + [b2 + [b1]+ + [b3]+]+
− 1
−bi − [bi−1]+ − [bi+1]+
−bi + [bi+1 + [bi]+ + [bi+2]+]+
+[bi−1 + [bi−2]+ + [bi]+]+
−bn + [bn−1 + [bn−2]+ + [bn]+]+ i = n.
i = 1;
i even;
i odd and i 6= 1, n;
β′ = −β −
[b1]+
2
α1 − ℓXk=1
[b2k+1]+α2k+1 + [b2k + [b2k−1]+ + [b2k+1]+]+α2k! .
If n is even, say n = 2ℓ, we have
2
0
0
−1 0 −1
. . .
. . .
1
B =
. . .
. . .
. . .
. . .
. . . −1
0
1
1
−1 0
.
2 b1 + [b2 + [b1]+ + [b3]+]+
− 1
−bi − [bi−1]+ − [bi+1]+
−bi + [bi+1 + [bi]+ + [bi+2]+]+
+[bi−1 + [bi−2]+ + [bi]+]+
−bn + [bn−1]+
i = 1;
i even;
i odd and i 6= 1, n;
i = n.
Then
So
b′
i =
β′ = −β −
[b1]+
2
α1 − ℓ−1Xk=1
[b2k+1]+α2k+1 + [b2k + [b2k−1]+ + [b2k+1]+]+α2k! .
Case II: Type Cn. This case is similar to the proof of type Bn.
Case III: Type Dn, n = 2ℓ + 1, ℓ > 2. We consider the exchange matrix
0
1 1 1
−1 0 0 0
−1 0 0 0
−1 0 0 0 −1
. . .
. . .
1
. . .
. . .
. . .
B =
.
. . .
. . .
−1 0 −1
0
1
1
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
19
Then we have
b′
i =
−b1 + [[b2]+ + [b1]+]+ + [[b3]+ + [b1]+]+
+[b4 + [b1]+ + [b5]+]+
−bi − [b1]+
−bi − [bi−1]+ − [bi+1]+
−bi + [bi+1 + [bi]+ + [bi+2]+]+
+[bi−1 + [bi−2]+ + [bi]+]+
−bn + [bn−1 + [bn−2]+ + [bn]+]+
i = 1;
i = 2, 3;
i even and i > 4;
i odd and 5 6 i 6 n − 1;
i = n.
So
β′ = − β − [b1]+α1 − [b2 + [b1]+]+α2 − [b3 + [b1]+]+α3
− ℓXk=2
[b2k+1]+α2k+1 +
[b2k + [b2k−1]+ + [b2k+1]+]+α2k! .
ℓXk=2
Case IV: Type E7. We consider the exchange matrix
B =
1 1 1
0
0
0
−1 0 0 0 −1
0
−1 0 0 0
0
0
0 −1 0
−1 0 0 0
0
0
1 0 0
0
0
0 0 1
0
1
0 −1 0
0 0 0
0
0
0
0
0
0
.
Then we have
b1 = −b1 + [b2 + [b1]+[b5]+]+ + [b3 + [b1]+]+ + [b4 + [b1]+ + [b6]+]+
b2 = −(b2 + [b1]+ + [b5]+)
b3 = −(b3 + [b1]+)
b4 = −(b4 + [b1]+ + [b6]+)
b5 = −b5 + [b2 + [b1]+ + [b5]+]+
b6 = −b6 + [b4 + [b1]+ + [b6]+]+ + [b7 + [b6]+]+
b7 = −(b7 + [b6]+)
β′ = − β − [b1]+α1 − [b2 + [b1]+ + [b5]+]+α2 − [b3 + [b1]+]+α3
− [b4 + [b1]+ + [b6]+]+α4 − [b6]+α6 − [b7 + [b6]+]+α7.
So
Case V: For types E6, E8 and F4, the lattice Lat(B) is Z6, Z8 and Z4 respectively,
so the result is true by the same argument as for type An with n even.
Case VI: The type G2 is a special case of Theorem 7.1.
(cid:3)
We are now ready for the main result of this section.
Theorem 5.2. Let A be a cluster algebra of finite type with arbitrary coefficients and
assume that A is not of type D2ℓ. Then
QAut0(A) ∼= Aut+(Atriv).
Moreover, if A is of type D2ℓ then
G ⋊ QAut0(A) ∼= Aut+(Atriv),
20
WEN CHANG AND RALF SCHIFFLER
where G is a subgroup of the symmetric group S3, in type D4, and G is a subgroup of
Z2 = Z/2Z, in types D2ℓ with 2ℓ ≥ 6.
Proof. For the type A1 a direct computation shows that QAut(A) ∼= Z2 ∼= Aut+(A). As-
sume now that n > 1. Proposition 4.6 yields an injective homomorphism QAut0(A) ֒→
Aut+(Atriv). By Lemma 5.1 and Remark 3.10 this is an isomorphism unless A is of type
D2ℓ.
In type D2ℓ a similar calculation shows that the Auslander-Reiten translation τ can
again be lifted as a quasi-automorphism of A. However, there may be additional au-
tomorphisms of Atriv which are not induced by some power of τ m, as we have seen in
Example 4.8. So we have hτ i ⊆ QAut0(A) ⊆ Aut(Atriv). By Table 1 in [2], we know
Aut(Atriv) ∼=(cid:26) Z4 × S3
Zn × Z2
in type D4;
in types Dn with n > 4,
where the cyclic part is generated by the Auslander-Reiten translation τ . This completes
the proof.
(cid:3)
Example 5.3. For the cluster algebra of Example 4.8 we have
QAut0(A) ∼= Z4 × Z2 ∼= hτ i × hσi,
where τ is the Auslander-Reiten translation and σ fixes x1 and x3 and interchanges x2
and x4. On the other hand, Aut+(Atriv) ∼= Z4 × S3, where the symmetric group S3
is given by all permutations of the variables x2, x3, x4. Since the symmetric group is a
semidirect product of Z3 and Z2 we have Z3 ⋊ QAut0(A) ∼= Aut+(Atriv).
As an immediate corollary, we obtain the complete list of all quasi automorphism
groups of finite type cluster algebras.
Corollary 5.4. Let A be a cluster algebra of finite type. Then the quasi-automorphism
group QAut0(A) is listed in Table 1. Note that the group is a cyclic group, except for
type D.
Proof. This follows from Theorem 5.2, Table 1 in [2].
(cid:3)
We close this section with the following result on lattices which is a consequence of
our arguments above.
Corollary 5.5. Let A be a cluster algebra of finite type and let Σ, Σ′ be two seeds such
that the principal parts of the exchange matrices are equal. Then the lattices of the
exchange matrices coincide.
Proof. This follows from Remark 3.10 and Theorem 5.2.
(cid:3)
6. Affine type
In this section we consider the quasi-automorphism groups for a cluster algebra A of
affine type, that is, the cluster algebras of types eAp,q, (p > 1, q > 1), eDn−1, (n > 5), eE6,
eE7 and eE8. We proceed case by case.
Case I: Let A be a cluster algebra of type eAp,q. Let Σ = (ex, p, B) be a seed of A,
where the quiver of the exchange matrix B is as follows:
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
21
2
/ 3
/ · · ·
/ p
1
=④④④④④④④④④
!❈❈❈❈❈❈❈❈❈
$❍❍❍❍❍❍❍❍❍❍
:✈✈✈✈✈✈✈✈✈
p + 1.
p + q
/ p + q − 1
/ · · ·
/ p + 2
1, · · · , x′
The cluster automorphism group of Atriv is computed in [2]. Denote by Σtriv = (x =
(x1, · · · , xp+q), B) the initial seed of Atriv, which corresponds to Σ. Let Σ′
triv = (x′ =
p+q), B′) be a seed obtained from Σtriv by an ordered sequence of mutations
(x′
p+q), B′′) be a seed obtained
at x1, xp+q, xp+q−1, · · · xp+2. Let Σ′′
from Σtriv by an ordered sequence of mutations at x1, x2, · · · xp. Then there are two
cluster automorphisms r1 and r2 of Atriv which map Σtriv to Σ′
triv respectively,
where the actions on the initial cluster variables are as follows:
triv = (x′′ = (x′′
triv and Σ′′
1, · · · , x′′
i+1
xp+q 7→ x′
1.
r1 :(xi 7→ x′
r2 :(x1 7→ x′′
xi 7→ x′′
p+q;
i−1
if 1 6 i 6 p + q − 1;
if 2 6 i 6 p + q.
In fact if p 6= q, then
Aut+(Atriv) ∼= Hp,q = hr1, r2r1r2 = r2r1, rp
(6.1)
see [2, section 3.3]. If p = q, then Aut+(Atriv) ∼= Hp,q ⋊ {1, σ}, where σ is induced by
the following permutation of the initial cluster variables:
1 = rq
2i,
x1 ↔ x1;
xp+1 ↔ xp+1;
xi ↔ xp+q+2−i
if 2 6 i 6 p = q.
σ :
Now we show the r1 can be lifted as a quasi-automorphism of A. We shall consider
only the case where p is even and q is odd, since the other cases are similar. According
β(cid:17) be
to Lemma 3.11, we may also assume that A only has one coefficient. Let B =(cid:16) B
the initial matrix of A, where β = (b1, . . . , bp+q). Let B′ = (cid:16) B
β ′(cid:17) be the matrix of a
labeled cluster which is obtained from the labeled cluster ex′ = µp+2 · · · µp+qµ1(ex) after
specializing the coefficient in ex′ to 1, we get x′. So to show the existence of a lift of r1
in QAut0(A) which mapsex toex′ After a direct calculation, we have
a relabeling 1 7→ p + q 7→ p + q − 1 7→ · · · 7→ 2 7→ 1. Denote the frozen row of B′
p+q). The cluster automorphism r1 maps x to x′. Note that after
by β′ = (b′
1, · · · , b′
b′
1 = b2 + [b1]+
b′
i = bi+1
b′
p = {bp+1}
b′
p+1 = −{bp+2}
b′
p+i = −{bp+i+1} + [{bp+i}]+ if 2 6 i 6 q − 1;
b′
p+q = −b1 + [{bp+q}]+
if 2 6 i 6 p − 1;
/
/
/
$
=
!
/
/
/
:
22
WEN CHANG AND RALF SCHIFFLER
where {bp+i}, 1 6 i 6 q, is inductively defined by
{bp+i} =(bp+q + [b1]+
bp+i + [{bp+i+1}]+ if 1 6 i 6 q − 1.
if i = q;
Finally, we have
β′ = −β + a1α1 + · · · + ap+q−2αp+q−2 + ap+qαa+q,
where the α are the rows of B and
kPi=1
kPi=1
kPi=1
kPi=1
a1 = bp+q + b′
p+q;
a2k = −
(b2i−1 + b′
2i−1) +
if 1 6 k 6 p/2;
(b2i + b′
2i) −
p/2Pi=1
(bp+2i + b′
p+2i) − bp+q − b′
p+q
(q−1)/2Pi=1
a2k+1 = −
(b2i + b′
2i) + bp+q + b′
p+q if 1 6 k 6 p/2;
ap+2k =
(bp+2i−1 + b′
p+2i−1) −
(b2i + b′
2i) +
if 1 6 k 6 (q − 1)/2;
if 1 6 k 6 (q − 1)/2.
ap+2k+1 =
(bp+2i + b′
p+2i) −
(b2i + b′
2i) + bp+q + b′
p+q
p/2Pi=1
p/2Pi=1
(bp+2i + b′
p+2i) + bp+q + b′
p+q
(q−1)/2Pi=1
Similarly, the cluster automorphism r2 can also be lifted to QAut0(A). Therefore
QAut0(A) ∼=
∼= Aut+(Atriv)
QAut0(A) ∼= Hp,q ⋊ G ⊆ Aut+(Atriv)
Hp,q
if p 6= q;
if p = q,
where G is a subgroup of Z2.
where the quiver of the exchange matrix B is
Case II: Let A be a cluster algebra of type eDn−1. Let Σ = (ex, p, B) be a seed of A,
c❍❍❍❍❍❍
z✈✈✈✈✈✈
6♥♥♥♥♥♥
(PPPPPPPP
(n odd),
/ n − 3
n − 1
n − 2
· · ·
/ 4
1
2
3
2
3
c❍❍❍❍❍❍
z✈✈✈✈✈✈
1
/ 4
· · ·
n − 3
/ n − 2
v♥♥♥♥♥♥
hPPPPPPPP
(n even).
n
It was shown in [2] that the direct cluster automorphism group Aut+(Atriv) of Atriv is
isomorphic to
(6.2)
G =*τ, σ, ρ1, ρn(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
ρ2
i = 1, τ ρi = ρiτ (i = 1, n)
ρ1σ = σρn, σρ1 = ρnσ + .
τ σ = στ, σ2 = τ n−3
n
n − 1
c
z
/
o
o
/
o
o
6
(
v
c
z
/
o
o
o
o
/
h
−
(n−5)/2Pk=2
−
(n−4)/2Pk=2
(n−1)/2Pk=1
(n−2)/2Pk=1
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
23
Here τ corresponds to the Auslander-Reiten translation of the cluster category. The
automorphism σ corresponds to a kind of reflection of the cluster category, and ρ1
(respectively ρ2) is induced by the automorphism of the initial quiver which permutes
the vertices 2 and 3 (respectively n − 1 and n) and preserves other vertices. We refer
the reader to section 3.3 [2] for details. Similar to the situation in finite type D2ℓ, the
automorphisms σ, ρ1 and ρ2 do not always lift to QAut0(A). However, we will see that τ
can always be lifted. Note that in this case the τ is still the composition τ+τ−, where τ−
(respectively τ+) is a sequence of mutations at the sinks (respectively sources). Similar
β(cid:17) and show that β′
as the method we used previously, we consider any matrix B =(cid:16) B
β ′(cid:17) .
is an integer linear combination of β and the rows α1, · · · , αn of B, where B′ = τ(cid:16) B
By a direct calculation, we see this from the following equality
β′ = −β − [−b1 + [−b2]+ + [−b3]+ + [−b4]+]+α1 − [−b3]+α3
[−b2k+1 + [−b2k]+ + [−b2k+2]+]+α2k+1 −
[−b2k]+α2k
−[−bn−2 + [−bn−3]+ + [−bn−1]+ + [−bn]+]+αn−2 − [−bn]+αn.
for odd n and the equality
β′ = −β − [−b1 + [−b2]+ + [−b3]+ + [−b4]+]+α1 − [−b3]+α3
[−b2k+1 + [−b2k]+ + [−b2k+2]+]+α2k+1 −
[−b2k]+α2k
−[−bn−1 + [−bn−2]+]+αn−1 − [−bn + [−bn−2]+]+αn.
for even n. Therefore QAut0(A) always contains a subgroup isomorphic to Z which is
generated by τ .
with the exchange matrix B, whose quiver is bipartite:
Case III: Let A be a cluster algebra of type eE6. We begin with a initial seed of A
6
3
1
5
/ 2
4
7.
It was shown in [2] that the direct cluster automorphism group Aut+(Atriv) is isomor-
phic to Z × S3, where the cyclic part is generated by the Auslander-Reiten translation
τ and the S3 part comes from the automorphisms of the quiver. Again, the cluster
automorphisms in S3 may not lift to QAut0(A), while τ can always be lifted. We still
use the notations β, β′ and αi, 1 6 i 6 7. The linear relation between them is
β′ = − β − [−b1 + [−b2]+ + [−b3]+ + [−b4]+]+α1 − [−b2]+α2 − [−b3]+α3
− [−b4]+α4 − [−b5 + [−b2]+]+α5 − [−b6 + [−b3]+]+α6 − [−b7 + [−b4]+]+α7.
Therefore
QAut0(A) ∼= Z × G ⊆ Aut+(Atriv),
where G is a subgroup of S3.
/
o
o
/
/
O
O
o
o
24
WEN CHANG AND RALF SCHIFFLER
Case IV: Let A be a cluster algebra of typeeE7 with the initial principal part quiver
3
6
5
/ 2
1
4
7
/ 8.
It was shown in [2] that the direct cluster automorphism group Aut+(Atriv) is iso-
morphic to Z × Z2, where the infinite cyclic part is generated by the Auslander-Reiten
translation τ and the Z2 part comes from the automorphisms of the quiver. Then
QAut0(A) ∼= Z × G ⊆ Aut+(Atriv),
where G is a subgroup of Z2. The linear relation is as follows:
β′ = − β − [−b1 + [−b2]+ + [−b3]+ + [−b4]+]+α1 − [−b2]+α2 − [−b3]+α3
− [−b4]+α4 − [−b5 + [−b2]+ + [−b6]+]−α5 − [−b6]+α6
− [−b7 + [−b4]+ + [−b8]+]+α7 − [−b8]+α8.
Case V: Let A be a cluster algebra of type eE8 with the initial principal part quiver
3
5
/ 2
1
4
6
/ 7
8
/ 9.
The direct cluster automorphism group Aut+(Atriv) = hτ i ∼= Z. Then we have
The linear relation is as follows:
QAut0(A) ∼= Z ∼= Aut+(Atriv).
β′ = − β − [−b1 + [−b2]+ + [−b3]+ + [−b4]+]+α1 − [−b2]+α2 − [−b3]+α3
− [−b4]+α4 − [−b5 + [−b2]+]+α5 − [−b6 + [−b4]+ − [−b7]+]+α6
− [−b7]+α7 − [−b8 + [−b7]+ + [−b9]+]+α8 − [−b9]+α9.
To summarize, we have the following
Theorem 6.1. Let A be a cluster algebra of affine type. Then the quasi-automorphism
group QAut0(A) is listed in Table 1.
Proof. This follows from the above discussion.
(cid:3)
We end the paper with a discussion of rank two cluster algebras.
Theorem 7.1. Let A be a cluster algebra of rank two, then
7. Rank 2
QAut0(A) ∼= Aut+(Atriv) ∼=
Z5
Z3
Z3
Z4
Z
in type A2;
in type B2;
in type C2;
in type G2;
in all other rank 2 types.
o
o
/
o
o
/
/
O
O
o
o
/
/
o
o
/
/
O
O
o
o
/
o
o
/
B =
0
−t
b1
s
0
b2
=
α1
α2
β
CLUSTER AUTOMORPHISMS AND QUASI-AUTOMORPHISMS
25
Proof. Let Σ = (x, p, B) be a seed. According to Lemma 3.11, we may assume without
loss of generality that B has exactly one frozen row. Then
where we may assume s and t are positive integers. Let Σ′ = (x′, p, B′) be another
2) denote the frozen row of B′. Then Σ′ = τ m(Σ) for some
seed and let β′ = (b′
integer m. If m = 1 then β′ = −β − [−b1 + t[−b2]+]+α1 − [−b2]+α2. If m = −1 then
β′ = −β − [b1]+α1 − [b2 + s[b1]+]+α2. In both cases we have Lat(B) = Lat(B′), and by
induction Lat(B) = Lat(τ m(B)) for any m. Therefore QAut0(A) ∼= Aut+(Atriv).
1, b′
If A is of finite type, that is types A2, B2, C2, and G2, then the group has been
computed in Corollary 5.4. Otherwise, the rank of τ is infinite, so QAut0(A) ∼= hτ i ∼=
Z.
(cid:3)
Acknowledgments. This joint work is done while the first author is visiting the Uni-
versity of Connecticut since December 2017. He gratefully acknowledges the support
of the China Scholarship Council, and the University of Connecticut for providing him
with an excellent working environment.
References
[1] Assem I, Dupont G, Schiffler R, On a category of cluster algebras, J. Pure Appl. Algebra 218 (2014),
no. 3, 553-582. 1
[2] Assem I, Schiffler R, Shramchenko V, Cluster automorphisms, Proc. Lond. Math. Soc. 2012,
104(6):1271-1302. 1, 7, 9, 10, 11, 13, 16, 20, 21, 22, 23, 24
[3] Assem I, Schiffler R, Shramchenko V, Cluster automorphisms and compatibility of cluster variables,
Glasg. Math. J. 56 (2014), no. 3, 705-720. 3
[4] Bazier-Matte V, Unistructurality of cluster algebras of type eA, J. Algebra 464 (2016), 297-315. 3
[5] Bridgeland T, Smith I, Quadratic differentials as stability conditions. (English summary) Publ.
Math. Inst. Hautes ´Etudes Sci. 121 (2015), 155-278. 16
[6] Cerulli Irelli G, Keller B, Labardini-Fragoso D and Plamondon P, Linear independence of cluster
monomials for skew-symmetric cluster algebras, Compos. Math. 149 (2013), no. 10, 1753-1764. 7
[7] Chang W, Zhu B, Cluster automorphism groups of cluster algebras with coefficients, Sci. China
Math. 59 (2016), no. 10, 1919-1936. 1, 6, 7, 9, 16
[8] Chang W, Zhu B, Cluster automorphism groups of cluster algebras of finite type, J. Algebra 447
(2016), 490-515. 1, 16
[9] Fraser C, Quasi-homomorphisms of cluster algebras, Adv. in Appl. Math. 81 (2016), 40-77. 2, 7, 8,
9, 14, 16
[10] Fomin S, Zelevinsky A, Y -systems and generalized associahedra, Ann. of Math. (2), 2003,
158(3):977-1018. 16
[11] Fomin S, Zelevinsky A, Cluster algebras: notes for the CDM-03 conference, Current developments
in mathematics, 2003, 1-34, Int. Press, Somerville, MA, 2003. 6
[12] Fomin S, Zelevinsky A, Cluster algebras. II. Finite type classification, Invent. Math. 2003, 154(1):63-
121. 7
[13] Fomin S, Zelevinsky A, Cluster algebras IV: Coefficients, Compos. Math. 2007, 143:112-164. 6, 7,
13
[14] Keller B, Cluster algebras and derived categoreis, Derived categories in algebraic geometry, 123-183,
EMS Ser. Congr. Rep., Eur. Math. Soc., Zrich, 2012. 5
[15] Gekhtman M, Shapiro M, Vainshtein A, On the properties of the exchange graph of a cluster algebra,
Math. Res. Lett, 15(2), (2008), 321 -- 330. 7, 10
[16] Marsh R, Scott, J. Twists of Plcker coordinates as dimer partition functions, Comm. Math. Phys.
341 (2016), no. 3, 821 -- 884. 2
[17] Nakanishi T, Zelevinsky A, On tropical dualities in cluster algebras. Algebraic groups and quantum
groups, 217-226, Contemp. Math. 565, Amer. Math. Soc. Providence, RI, 2012. 15
26
WEN CHANG AND RALF SCHIFFLER
[18] Reading N, Universal geometric cluster algebras from surfaces, Trans. Amer. Math. Soc. 366 (2014)
6647 -- 6685. 6
[19] Reading N, Universal geometric coefficients for the once-punctured torus, S´em. Lothar. Combin. 71
(2015), paper B71e, 30 pp. 6
[20] Saleh, I. Cluster automorphisms and hyperbolic cluster algebras, Thesis (Ph.D.)Kansas State Uni-
versity. 2012. 103 pp. ISBN: 978-1267-60082-0 3
School of Mathematics and Information Science, Shaanxi Normal University, Xi'an
710062, China.
Department of Mathematics, University of Connecticut, Storrs, CT 06269-1009, USA
E-mail address: [email protected]
Department of Mathematics, University of Connecticut, Storrs, CT 06269-1009, USA
E-mail address: [email protected]
|
1807.01041 | 2 | 1807 | 2018-10-26T08:24:18 | Graded-simple algebras and cocycle twisted loop algebras | [
"math.RA",
"math.RT"
] | The loop algebra construction by Allison, Berman, Faulkner, and Pianzola, describes graded-central-simple algebras with split centroid in terms of central simple algebras graded by a quotient of the original grading group. Here the restriction on the centroid is removed, at the expense of allowing some deformations (cocycle twists) of the loop algebras. | math.RA | math |
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP
ALGEBRAS
ALBERTO ELDUQUE
Abstract. The loop algebra construction by Allison, Berman, Faulkner, and
Pianzola, describes graded-central-simple algebras with split centroid in terms
of central simple algebras graded by a quotient of the original grading group.
Here the restriction on the centroid is removed, at the expense of allowing
some deformations (cocycle twists) of the loop algebras.
1. Introduction
The graded-central-simple algebras (not necessarily associative, nor Lie) with
split centroid were shown in [ABFP08] to be isomorphic to loop algebras of algebras
graded by a quotient group that are central simple as ungraded algebras. This is a
very important reduction, as the graded-central-simple algebras may fail to be nice
as ungraded algebras, for instance, they may fail to be simple or semisimple.
The purpose of this paper is to remove the restriction of the centroid being split,
at the expense of allowing certain deformations of the loop algebra construction.
These deformations will be based on a symmetric 2-cocycle on the grading group
with values in the multiplicative group of the ground field.
Graded-central-simple algebras over the real field have been studied, with a
different approach based on Galois descent from C to R, in [BKpr]. Their results
are subsumed nicely in our more general description.
All the algebras considered will be defined over a ground field F, unless otherwise
stated, and they are just vector spaces over F endowed with a bilinear multiplica-
tion, usually denoted by juxtapostion. No assumption on associativity, dimension
(which can be infinite), or existence of unity, is made.
A is a vector space decomposition Γ : A = Lg∈G
For the basic facts on gradings, the reader may consult [EK13]. Here we will
review some basic definition. Let A be an algebra and G a group. A G-grading on
Ag such that AgAh ⊆ Agh for
any g, h ∈ G. The nonzero elements in Ag are said to be homogeneous of degree g.
The support of Γ is the set {g ∈ G Ag 6= 0}.
Gradings by abelian groups often arise as eigenspace decompositions with respect
to a family of commuting diagonalizable automorphisms. Over an arbitrary field,
a G-grading Γ on A is equivalent to a homomorphism of affine group schemes
ηΓ : GD → AutF(A), where GD is the diagonalizable group scheme represented
by the group algebra FG. In this paper, only gradings by abelian groups will be
considered.
Let Γ : A = Lg∈G
g be two gradings by an abelian
group G. The G-gradings Γ and Γ′ are isomorphic if A and A′ are isomorphic as
G-graded algebras, i.e., if there exists an isomorphism of algebras ϕ : A → A′ such
Ag and Γ′ : A′ = Lg∈G
A′
2010 Mathematics Subject Classification. Primary 16W50, 17B70.
Key words and phrases. Loop algebra, cocycle twist, graded-simple, graded-central.
Supported by grants MTM2017-83506-C2-1-P (AEI/FEDER, UE) and E22 17R (Diputaci´on
General de Arag´on).
1
2
ALBERTO ELDUQUE
that ϕ(Ag) = A′
A′ are graded-isomorphic.
g for all g ∈ G. We will write then A ≃G A′ and say that A and
The paper is structured as follows. Section 2 will review known results on ex-
tensions of abelian groups and their associated 2-symmetric cocycles needed in the
sequel. These symmetric 2-cocycles τ : G × G → F× are used in Section 3 to define
a new multiplication on any G-graded algebra A. The new algebra thus obtained is
denoted by Aτ and it is associative, Lie, ..., if A is so. The basic properties of these
cocycle twists are given. Actually, these cocycle twists are particular instances of
the graded contractions considered in [dMP91], [MP91]. The underlying ideas go
back to [IW53].
Section 4 reviews the general loop algebra construction in [ABFP08] and some
of its main properties. Section 5 contains the main results of the paper, which show
that any graded-central-simple algebra is graded-isomorphic to a cocycle twist of
a loop algebra of a central simple algebra graded by a quotient of the original
grading group (Theorem 5.2). Actually, the graded-isomorphism class of a G-
graded-central-simple algebra is determined by a subgroup H of G, an element in
the symmetric second cohomology group H2
sym(H, F×), and the equivalence class
of a central simple (as ungraded algebra) G/H-graded algebra, under a precise
equivalence relation weaker than graded-isomorphism (Corollary 5.5).
The connection with the approach over the real field by Galois descent in [BKpr]
is explained in Section 6.
2. Ext, extensions, and H2
sym
In this section, several well-known results on extensions of abelian groups will
be reviewed. The reader may consult [HS97, Chapters III and VI].
Unless otherwise stated, multiplicative notation will be used for abelian groups.
2.1. Ext(A, B) and extensions. Given two abelian groups (i.e., Z-modules) A, B,
the abelian group ExtZ(A, B), or simply Ext(A, B), can, and will, be identified with
the set of equivalence classes of extensions of A by B (in the category of abelian
groups, and we will refer to them as abelian extensions), where an extension of A
by B is a short exact sequence 1 → B → E → A → 1, and two extensions
1 −→ B −→ E −→ A −→ 1
and
1 −→ B −→ E ′ −→ A −→ 1
are equivalent if there is a homomorphism ϕ : E → E ′, necessarily bijective, such
that the diagram
1
1
B
B
E
ϕ
E ′
A
A
1
1
is commutative.
For a homomorphism f : A′ → A, the natural homomorphism
f ∗ = Ext(f, B) : Ext(A, B) −→ Ext(A′, B)
is obtained as follows. Let ξ : 1 → B i→ E
let E be the pull-back of p and f :
p
−→ A → 1 be an abelian extension, and
E = {(x, a′) ∈ E × A′ p(x) = f (a′)}.
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP ALGEBRAS
3
Then we obtain a commutative diagram with exact rows:
1
ξ : 1
B
B
j
i
p2
p
E
p1
E
A′
f
A
1
1
where p1(x, a′) = x and p2(x, a′) = a′ are the canonical projections, and j(x) =
(cid:0)i(x), e(cid:1) for any x ∈ B (e denotes the neutral element). The extension of A′ by B
in the first row is denoted by ξf , and the map
(1)
f ∗ : Ext(A, B) −→ Ext(A′, B)
[ξ]
7→ [ξf ]
is well defined and equals Ext(f, B).
2.2. Ext(A, B) ≃ H2
sym(A, B). On the other hand, the set of equivalence classes
of central extensions, in the category of groups, of the group A by the abelian
group B, that is, equivalence classes as above, but of short exact sequences of
p
groups 1 → B i→ E
−→ A → 1 with i(B) central in E, can be identified with the
second cohomology group H2(A, B) = Z2(A, B)/B2(A, B), where
Z2(A, B) = {σ : A × A → B
σ(a1, a2)σ(a1a2, a3) = σ(a1, a2a3)σ(a2, a3) ∀a1, a2, a3 ∈ A}
is the set of 2-cocycles, and
B2(A, B) = {dγ γ : A → B a map}
is the set of 2-coboundaries, where dγ(a1, a2) = γ(a1)γ(a2)γ(a1a2)−1 for any map
γ : A → B and a1, a2 ∈ A.
The element in H2(A, B) that corresponds to the central extension ξ : 1 → B i→
p
E
−→ A → 1 is obtained by fixing a section s : A → E of p (s is not a homomorphism
in general). Then for any a1, a2 ∈ A, the element s(a1)s(a2)s(a1a2)−1 is in ker p =
im i, so there is a unique element σ(a1, a2) ∈ B such that
i(cid:0)σ(a1, a2)(cid:1) = s(a1)s(a2)s(a1a2)−1.
(2)
Then σ : A × A → B is a 2-cocycle whose equivalence class [σ] in H2(A, B) does
not depend on the chosen section s. This equivalence class [σ] is the element in
H2(A, B) that corresponds to the equivalence class of ξ.
Moreover, for A and B abelian, ξ is an abelian extension if and only if σ is
symmetric.
Denote by Z2
sym(A, B) the subgroup of Z2(A, B) of the symmetric 2-cocycles,
that is, 2-cocycles σ such that σ(a1, a2) = σ(a2, a1) for any a1, a2 ∈ A, and note
that B2(A, B) is contained in Z2
sym(A, B). Then, for A and B abelian, Ext(A, B)
can be identified too with the quotient
sym(A, B) = Z2
sym(A, B)/B2(A, B).
H2
p
Given an abelian extension ξ : 1 → B i→ E
−→ A → 1, the 2-cocycle σ ∈
sym(A, B) defined in (2), and a homomorphism of abelian group f : A′ → A, the
sym(A′, B) attached to ξf is just [σ ◦ (f × f )]. In other words, f ∗ in
Z2
element in H2
(1) corresponds to the natural map (also denoted by f ∗):
sym(A′, B)
7→ [σ ◦ (f × f )]
sym(A, B) −→ H2
f ∗ : H2
(3)
[σ]
4
ALBERTO ELDUQUE
2.3. The long exact sequence. Given an abelian group G, a subgroup H, and the
associated quotient group G/H, consider the corresponding short exact sequence
(4)
ζ : 1
H
ι
π
G
G/H
1
where ι is the inclusion map and π the canonical projection. Given any abelian
group F , there is the associated long exact sequence
(5) 1 → Hom(G/H, F ) π∗
−→ Hom(G, F ) ι∗
−→ Hom(H, F )
δ−−→ Ext(G/H, F ) π∗
−→ Ext(G, F ) ι∗
−→ Ext(H, F ) → 1
because Extn(., .) is trivial for n ≥ 2 (see, e.g., [Wei94, Lemma 3.3.1]). The maps
π∗ and ι∗ in the first row are just the precompositions by π and ι, respectively.
The maps π∗ and ι∗ in the second row are given by (1). As for the connecting
homomorphism δ, it is obtained as follows. Given a homomorphism f : H → F , let
E be the push-out of f and ι:
E = F × G/(cid:10)(cid:0)f (x), ι(x)−1(cid:1) x ∈ H(cid:11) .
There appears a commutative diagram with exact rows:
ζ : 1
1
ι
j1
H
f
F
π
p
G
j2
E
G/H
G/H
1
1
where j1 and j2 are the canonical homomorphisms and p([x, g]) = π(g) for any
[x, g] ∈ E ([x, g] denotes the class of the element (x, g) modulo the subgroup
(cid:10)(cid:0)f (x), ι(x)−1(cid:1) x ∈ H(cid:11)). The extension in the second row is denoted by f ζ, and
the map
δ : Hom(H, F ) −→ Ext(G/H, F )
f
7→
[f ζ]
is precisely the connecting homomorphism.
Given a section s : G/H → G, the element in H2
sym(G/H, H) corresponding to
ζ is the class [σ] of the 2-cocycle
(6)
σ : G/H × G/H −→ H
(¯g1, ¯g2) 7→ s(¯g1)s(¯g2)s(¯g1¯g2)−1.
Then s : G/H → E, ¯g 7→ [e, s(¯g)] is a section of p in f ζ, and for ¯g1, ¯g2 ∈ G/H
s(¯g1)s(¯g2)s(¯g1¯g2)−1 = [(e, ι(cid:0)σ(¯g1, ¯g2)(cid:1)] = [f(cid:0)σ(¯g1, ¯g2)(cid:1), e].
Hence the connecting homomorphim δ corresponds to the map (denoted too by δ):
(7)
δ : Hom(H, F ) −→ H2
sym(G/H, F )
f 7→ [f ◦ σ] .
The long exact sequence (5) can then be substituted by
(8) 1 → Hom(G/H, F ) π∗
−→ Hom(G, F ) ι∗
δ−→ H2
sym(G/H, F ) π∗
−→ H2
−→ Hom(H, F )
sym(G, F ) ι∗
−→ H2
sym(H, F ) → 1
with δ in (7) and π∗ and ι∗ in the second row as in (3).
The exactness of (5) has the following consequence that will be critical later on:
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP ALGEBRAS
5
Proposition 2.1. Let G and F be abelian groups, H a subgroup of G, and τ ′ : H ×
H → F a symmetric 2-cocycle. Then there is a symmetric 2-cocycle τ ∈ Z2
sym(G, F )
that extends τ ′ (i.e., τ ′ = τ H×H ).
Proof. The homomorphism ι∗ : H2
is a 2-cocycle τ ∈ Z2
sym(G, F ) → H2
γ : H → F such that τ ′ = (cid:0)τ H×H )(dγ). That is,
sym(H, F ) being surjective, there
sym(G, F ) such that [τ H×H ] = ι∗([τ ]) = [τ ′], so there is a map
τ ′(h1, h2) = τ (h1, h2)γ(h1)γ(h2)γ(h1h2)−1
for any h1, h2 ∈ H. Extend γ to a map γ : G → F (for instance, with γ(g) = e for
any g ∈ G \ H). Then τ = τ (dγ) satisfies τ H×H = τ ′.
(cid:3)
3. Cocycle twists
Definition 3.1. Let G be an abelian group, let A be an algebra over F endowed
Ag, and let τ : G×G → F× be a symmetric 2-cocycle.
with a G-grading: A = Lg∈G
Define a new multiplication on A by the formula
(9)
x ∗ y := τ (g1, g2)xy
for g1, g2 ∈ G, x ∈ Ag1 , y ∈ Ag2 .
The new algebra thus defined will be called the τ -twist of A, and will be denoted
by Aτ .
Remark 3.2. In [MP91] (see also [dMP91]) a more general situation is considered,
where τ : G × G → F is allowed to take the value 0, but still being symmetric
and satisfying the cocycle condition: τ (g1, g2)τ (g1g2, g3) = τ (g1, g2g3)τ (g2, g3) for
g1, g2, g3 ∈ G. The resulting twisted algebra is said to be a graded contraction of A
and the interest in the above mentioned references lies in those τ 's that indeed take
the value 0. In this way one can obtain, for example, solvable Lie algebras as graded
contractions of semisimple Lie algebras, as some nonzero structure constants in the
original algebra may become 0 in the twist.
Example 3.3. Consider the real algebra of complex numbers C, graded by the
cyclic group of order 2: C2 = {e, h}, so Ce = R1, Ch = Ci. Let τ : C2 × C2 → R×
be the symmetric 2-cocycle such that τ (e, e) = τ (e, h) = 1 and τ (h, h) = −1. Then
in the τ -twist Cτ we have
1 ∗ 1 = 1,
1 ∗ i = i ∗ 1 = i,
but
i ∗ i = 1,
so that Cτ is isomorphic to the group algebra RC2 (isomorphic to R × R).
Example 3.4. For any abelian group G and symmetric 2-cocycle τ ∈ Z2
sym(G, F×),
the τ -twist (cid:0)FG(cid:1)τ
of the group algebra FG is denoted traditionally by Fτ G.
This makes sense for not necessarily symmetric 2-cocycles and nonabelian groups.
The algebras Fτ G are called twisted group algebras, and they play a key role in
Schur's theory of projective representations of finite groups. Twisted group alge-
bras are, up to isomorphism, the graded-division algebras with one-dimensional
homogeneous components. Moreover, for τ1, τ2 ∈ Z2(G, F×), Fτ1G ≃G Fτ2G if and
only if [τ1] = [τ2] in H2(G, F×).
Natural examples of twisted group algebras are quaternion algebras over a field
F, char F 6= 2, which can be described, up to isomorphism, as the twisted group
algebras Fτ G, with G = C2 × C2 and τ a not symmetric 2-cocycle.
Some basic properties of cocycle twists are given in the next result.
6
ALBERTO ELDUQUE
Proposition 3.5. Let G be an abelian group and let A = Lg∈G
algebra over F.
Ag be a G-graded
sym(G, F×), (cid:0)Aσ(cid:1)τ
(i) For σ, τ ∈ Z2
(ii) If τ ∈ B2(G, F×), then Aτ ≃G A. More generally, if τ1, τ2 ∈ Z2
= Aστ .
sym(G, F×)
In other words, for
sym(G, F×), the G-graded isomorphism class of Aτ depends only on
sym(G, F×).
and [τ1] = [τ2] in H2
τ ∈ Z2
[τ ] ∈ H2
sym(G, F×), then Aτ1 ≃G Aτ2 .
(iii) If F is an algebraic closure of F and τ ∈ Z2
sym(G, F×), then Aτ ⊗F F ≃G
A ⊗F F. In particular, if A is an asociative, alternative, Lie, linear Jordan,
...., algebra, so is Aτ .
(iv) If A is graded-simple, so is Aτ .
Proof. Part (i) is clear. For (ii), let γ : G → F× be a map such that τ1 = τ2(dγ).
The multiplication in Aτ1 and Aτ2 are x ∗ y = τ1(g1, g2)xy and x ⋆ y = τ2(x, y)xy,
respectively, for g1, g2 ∈ G, x ∈ Ag1 , y ∈ Ag2 . Then the linear automorphism
ϕ ∈ EndF(A) such that ϕ(x) = γ(g)x for g ∈ G and x ∈ Ag preserves the grading
and satisfies:
ϕ(x) ⋆ ϕ(y) = γ(g1)γ(g2)τ2(g1, g2)xy
= γ(g1g2)(dγ)(g1, g2)τ2(g1, g2)xy
= γ(g1g2)τ1(g1, g2)xy
= γ(g1g2)x ∗ y = ϕ(x ∗ y).
For (iii) note that for τ ∈ Z2
sym(G, F×) ⊆ Z2
sym(G, F
×
×
F
is divisible (hence injective as a Z-module), so Ext(G, F
and hence, because of (ii), (cid:0)A ⊗F F(cid:1)τ
≃G A ⊗F F.
), Aτ ⊗F F = (cid:0)A ⊗F F(cid:1)τ
×
×
) ≃ H2
sym(G, F
. But
) = 1,
As for (iv), it is enough to note that the ideal generated by any homogeneous
(cid:3)
element x ∈ Ag is the same in A or in Aτ .
Example 3.6. Let G be an abelian group, H a subgroup, and G = G/H the
corresponding quotient. Let A be a G-graded algebra and let χ ∈ Hom(H, F×) (a
character on H). Finally, let s : G → G be a section of the canonical projection
π : G → G. Then in [ABFP08, Definition 6.3.1], the algebra Aχ is defined on the
same vector space as A, but with new multiplication
x ·χ y = χ(cid:16)s(g1)s(g2)s(g1g2)−1(cid:17)xy
for g1, g2 ∈ G, x ∈ Ag1 , y ∈ Ag2 . We see that Aχ is the (χ ◦ σ)-twist Aχ◦σ of A,
where σ ∈ Z2
sym(G, H) is the symmetric 2-cocycle considered in (6). Moreover, note
that [χ ◦ σ] = δ(χ), where δ is the connecting homomorphism in (7) for F = F×.
Recall that the centroid C(A) of an algebra A is the unital associative subalgebra
of EndF(A) given by
C(A) = {c ∈ EndF(A) c(xy) = c(x)y = xc(y) ∀x, y ∈ A} .
If the algebra A is simple, then C(A) is a field extension of F. The algebra A is said
to be central simple if it is simple and this field extension is trivial: C(A) = F1.
(Here 1 denotes the identity map on A.) Any simple algebra is central simple when
considered as an algebra over its centroid.
Given an abelian group G and a G-graded algebra A = Lg∈G
Ag, consider the
subspaces
C(A)g = {c ∈ C(A) cAg′ ⊆ Agg′ ∀g′ ∈ G}
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP ALGEBRAS
7
for g ∈ G. Their sum Lg∈G C(A)g is direct, but it may fail to be the whole C(A).
The G-graded algebra A is called graded-central if C(A)e = F1.
Also, the G-graded algebra A is said to be graded-simple if A2 = A and A does
not contain any proper graded ideals, and graded-central-simple if it is graded-
simple and graded-central.
If A is graded-simple, then C(A) = Lg∈G C(A)g ([BN06, Proposition 2.16]) and
it is a graded-field, i.e., it is commutative and any nonzero homogeneous element is
invertible. If H is the support of the G-grading on C(A): H = {g ∈ G C(A)g 6= 0},
H is a subgroup of G. Then if K = C(A)e (a field extension of F), C(A) is
isomorphic, as a graded algebra, to a twisted group algebra Kτ H (see Example
3.4), for a symmetric 2-cocycle τ ∈ Z2
sym(H, K×), the G-grading (with support
H) in Kτ H being the natural one. Moreover, A is graded-central-simple when
considered as an algebra over K.
The next results shows the behavior of the centroid of a G-graded-central simple
algebra under cocycle twists.
Proposition 3.7. Let G be an abelian group and B a G-graded-central-simple
algebra. For any τ ∈ Z2
sym(G, F×), C(Bτ ) ≃G C(B)τ .
Proof. For any homogeneous element c ∈ C(B)h, define cτ ∈ EndF(B) by
cτ (x) = τ (h, g)c(x)
for g ∈ G and x ∈ Bg. Denote by ∗ the multiplication in Bτ : x ∗ y = τ (g1, g2)xy,
for g1, g2 ∈ G, x ∈ Bg1 , y ∈ Bg2 . Then, under these assumptions,
cτ (x ∗ y) = τ (g1, g2)cτ (xy)
= τ (g1, g2)τ (h, g1g2)c(xy)
= τ (h, g1)τ (hg1, g2)c(x)y
= τ (hg1, g2)cτ (x)y = cτ (x) ∗ y
and also, because of the symmetry of τ :
cτ (x ∗ y) = τ (g1, g2)τ (h, g1g2)c(xy)
= τ (h, g2)τ (g1, hg2)xc(y) = x ∗ cτ (y).
Thus cτ ∈ C(Bτ )h. Besides, if c1 ∈ C(B)h1 and c2 ∈ C(B)h2 , then for x ∈ Bg we
have:
cτ
1cτ
2 (x) = τ (h1, h2g)τ (h2, g)c1c2(x)
= τ (h1, h2)τ (h1h2, g)c1c2(x)
= τ (h1, h2)(c1c2)τ (x),
1cτ
so cτ
c, is an isomorphism.
2 = τ (h1, h2)(c1c2)τ , and the map C(B)τ → C(Bτ ), c 7→ cτ , for homogeneous
(cid:3)
4. Loop algebras
Given an abelian group G, a subgroup H, the canonical projection π : G → G =
Ag, the loop
G/H, g 7→ π(g) = g, and an algebra A graded by G: A = Lg∈G
algebra Lπ(A) is the G-graded algebra
Lπ(A) = M
g∈G
Ag ⊗ g
which is a subalgebra of the tensor product A⊗F FG, where FG is the group algebra
of G (see [ABFP08, Definition 3.1.1]).
8
ALBERTO ELDUQUE
If A is G-graded-central-simple, then its centroid C(A) is said to be split (see
[ABFP08, Definition 4.3.6]) if C(A) is isomorphic, as a graded algebra, to the (un-
twisted) group algebra: C(A) ≃G FH. This is always the case if F is algebraically
closed by Proposition 3.5.
The main results in [ABFP08] reduce the study of the graded-central-simple
algebras with split centroid to the study of the central simple graded algebras (i.e.,
graded algebras which are central simple as (ungraded) algebras).
Theorem 4.1. (see [ABFP08, Theorem 7.1.1]) Let G be an abelian group, H a
subgroup of G, and π : G → G = G/H the canonical projection.
(1) If A is a central simple algebra graded by G, then the loop algebra Lπ(A)
is a G-graded-central-simple algebra, and the map
FH −→ C(cid:0)Lπ(A)(cid:1)
h 7→ (cid:0)x ⊗ g 7→ x ⊗ hg)
for g ∈ G, x ∈ Aπ(g), is an isomorphism of G-graded algebras. (Hence
C(cid:0)Lπ(A)(cid:1) ≃G FH.)
(2) if B is a G-graded-central-simple algebra with split centroid C(B) ≃G FH,
then there exists a central simple and G-graded algebra A such that B ≃G
Lπ(A).
(3) If A and A′ are central simple and G-graded algebras, then Lπ(A) ≃G
Lπ(A′) if and only if there is a character χ ∈ Hom(H, F×) such that A′ ≃
Aχ = Aχ◦σ (as in Example 3.6).
5. Graded-central-simple algebras
The goal in this section is to remove the restriction on the centroid of a graded-
central-simple algebra to be split, at the expense of allowing cocycle twists of loop
algebras.
For simplicity, given an abelian group G, a subgroup H, a G = G/H-graded al-
gebra algebra A, and a symmetric 2-cocycle τ ∈ Z2
will be denoted by Lτ
π(A) and called a cocycle twisted loop algebra.
sym(G, F×), the τ -twist (cid:0)Lπ(A)(cid:1)τ
Remark 5.1. Lτ
where Fτ G is the twisted loop algebra, as in Example 3.4.
π(A) is the subalgebraLg∈G
Ag ⊗ g of the tensor product A⊗F Fτ G,
The next Theorem is the main result of the paper. It reduces the classification of
graded-central-simple algebras, up to isomorphism, to the classification of central
simple and graded algebras, up to isomorphism.
Theorem 5.2. Let G be an abelian group.
(1) If B is a G-graded-central-simple algebra, then there is a subgroup H of
G, a central simple and G = G/H-graded algebra A, and a symmetric 2-
cocycle τ ∈ Z2
π(A). (As usual, π denotes the
canonical projection G → G.)
sym(G, F×) such that B ≃G Lτ
(2) Conversely, let H be a subgroup of G, A a central simple and G-graded
π(A) is G-graded-central-simple
algebra, and let τ ∈ Z2
π(A)(cid:1) ≃G Fτ ′
and C(cid:0)Lτ
algebra, τi ∈ Z2
projection, i = 1, 2. Then Lτ1
conditions are satisfied:
(3) For i = 1, 2, let Hi be a subgroup of G, Ai a central simple and G/Hi-graded
sym(G, F×). Denote by πi : G → Gi = G/Hi the canonical
π2(A2) if and only if the following
π1(A1) ≃G Lτ2
sym(G, F×). Then Lτ
H, where τ ′ = τ H×H .
• H1 = H2 =: H, so π1 = π2 =: π : G → G = G/H, and
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP ALGEBRAS
9
• there is a 2-cocycle µ ∈ Z2
A2.
sym(G, F×) and Aµ
1 ≃G
H2
sym(G, F×) such that [τ1] = π∗([µ])[τ2] in
Proof. For (1), if B is a G-graded-central-simple algebra, and H is the support of its
centroid, then C(B) is a twisted group algebra: C(B) ≃G Fτ ′
H, for a 2-cocycle τ ′ ∈
Z2
sym(H, F×). Proposition 2.1 shows that there is a 2-cocycle τ ∈ Z2
sym(G, F×) such
that τ H×H = τ ′. Then by Proposition 3.7, C(Bτ −1
=
FH, so C(Bτ −1
) is split. [ABFP08, Theorem 7.1.1.(iii)] shows that there is a central
simple G-graded algebra A such that Bτ −1
)τ ≃G
Lτ
≃G (cid:0)Fτ ′
≃G Lπ(A), and hence B = (cid:0)Bτ −1
) ≃G C(B)τ −1
H(cid:1)τ −1
π(A) by Proposition 3.5.
Now (2) follows since (cid:0)Lτ
π(A)(cid:1)τ −1
= Lπ(A) is G-graded-central-simple, and
π(A) by Propositions 3.5 and 3.7, which also imply the last part
hence so is Lτ
of (2).
Finally, if Lτ1
π1(A1) ≃G Lτ2
π2(A2), then their centroids are graded-isomorphic too,
and hence with equal supports. Thus H1 = H2 (see Theorem 4.1.(1)). By part (2),
Fτ ′
sym(H, F×),
that is, ι∗([τ1]) = ι∗([τ2]), where ι : H ֒→ G is the inclusion. By the exactness of
(8), there is a 2-cocycle ν ∈ Z2
i = τiH×H , i = 1, 2, so that [τ ′
sym(G, F×) such that [τ1] = π∗([ν])[τ2].
1H ≃G Fτ ′
2H, where τ ′
2] in H2
1] = [τ ′
2
π
Lπ(Aν
From Lτ1
π (A2) we get Lτ1τ −1
π (A1) ≃G Lτ2
1 ). Hence Lπ(Aν
(A1) ≃G Lπ(A2). But [τ1τ −1
2 ] =
π(A1) =
1) ≃G Lπ(A2), and we conclude from [ABFP08, Theorem
π∗([ν]) = [ν ◦ (π × π)], so with bν = ν ◦ (π × π) we get Lτ1τ −1
7.1.1.(iii)] that there exists a character χ ∈ Hom(H, F×) such that (cid:0)Aν
A2. Example 3.6 shows that (cid:0)Aν
and [ν(χ ◦ σ)] = [ν]δ(χ). With
µ = ν(χ ◦ σ), which lies in Z2
A2, and since π∗([µ]) =
π∗([ν])π∗δ(χ) = π∗([ν]) by the exactness of (8), we obtain [τ1] = π∗([ν])[τ2] =
π∗([µ])[τ2], as required.
1(cid:1)χ = Aν(χ◦σ)
sym(G, F×), we have Aµ
1(cid:1)χ ≃G
(A1) ≃G Lbν
1 ≃G
π
1
2
The converse is clear, because from Aµ
1 ≃G
A2 we get Lπ(Aµ
π(A1) ≃G Lπ(A2), where bµ = µ ◦ (π × π). But [bµ] = π∗([µ]) = [τ1τ −1
Lbµ
Proposition 3.5 gives Lτ1τ −1
(cid:0)Lτ1τ −1
π (A2).
π(A1) ≃G Lπ(A2), and hence Lτ1
(A1)(cid:1)τ2 ≃G Lτ2
(A1) ≃G Lbµ
π
π
2
2
1 ) ≃G Lπ(A2), so
2 ], so
π (A1) =
(cid:3)
Remark 5.3. There is a great freedom in choosing τ ∈ Z2
sym(G, F×) in the proof
above, the only required condition being that τ H×H should be cohomologous to
τ ′: ι∗([τ ]) = [τ ′] in H2
sym(H, F×).
We can express the above result in a concise way as follows. Given the abelian
group G, consider the set B(G, F) of the isomorphism classes (as G-graded alge-
bras) of G-graded-central-simple algebras, denoting by [B] the class of an algebra
B. Consider too the set A(G, F) consisting of the equivalence classes of triples
(H, [τ ], A), where H is a subgroup of G, [τ ] ∈ H2
sym(G, F×), and A is a central
simple and G/H-graded algebra, the equivalence relation being given by
(H1, [τ1], A1) ∼ (H2, [τ2], A2)
if H1 = H2(=: H), ι∗([τ1]) = ι∗([τ2]), (ι : H ֒→ G being the inclusion map), and if
there is a µ ∈ Z2
sym(G/H, F×) such that [τ1] = π∗([µ])[τ2] and Aµ
1 ≃G/H A2.
Corollary 5.4. The map
is a bijection.
A(G, F) −→ B(G, F)
π(A)]
7→ [Lτ
[(H, [τ ], A)]
10
ALBERTO ELDUQUE
sym(G, F×) is such that ι∗([τ ]) ∈ H2
The inverse is given by [B] 7→ [(H, [τ ], A)], where H is the support of C(B),
[τ ] ∈ H2
sym(H, F×) determines the isomorphism
class of C(B) as an H-graded algebra, and A is a central image (see [ABFP08, §6])
of Bτ −1
(whose centroid is split).
If we want to get rid of the freedom in choosing τ ∈ Z2
sym(G, F×) of the map ι∗ in (8). Then we may consider the set A
sym(G, F×) mentioned in
Remark 5.3, we may fix, for all subgroups H of G, a section ξH : H2
sym(H, F×) →
H2
(G, F) of
triples (H, [τ ′], [A]), where H is a subgroup of G, [τ ′] ∈ H2
sym(H, F×), and [A]
is the equivalence class of a central simple and G/H-graded algebra A, under the
equivalence relation being given by A1 ∼ A2 if there is a character χ ∈ Hom(H, F×)
′
such that (cid:0)A1(cid:1)χ ≃G/H A2.
Corollary 5.5. The map
′
A
(G, F) −→ B(G, F)
(H, [τ ′], [A]) 7→ [Lτ
π(A)]
where τ is any 2-cocycle in Z2
sym(G, F×) such that [τ ] = ξH ([τ ′]), is a bijection.
Proof. The map is clearly surjective. If two pairs (H1, [τ ′
2], [A2])
have the same image [B], then the support of C(B) is H1 = H2(=: H), and its
isomorphism class (as H-graded algebra) is determined by both [τ ′
2] in
H2
π(A2), with [τ ] =
ξH ([τ ′
for some χ ∈ Hom(H, F×) by [ABFP08, Theorem 7.1.1.(iii)]. Thus (H1, [τ ′
(H2, [τ ′
2]), which implies Lπ(A1) ≃G Lπ(A2), and hence (cid:0)A1)χ ≃G/H A2
sym(H, F×). Hence (H1, [τ ′
1], [A1]) and (H2, [τ ′
1], [A1]) =
(cid:3)
1]) = (H2, [τ ′
2]). Then Lτ
π(A1) ≃G Lτ
1]) = ξH ([τ ′
1] and [τ ′
2], [A2]).
6. The real case
If the ground field is the field of real numbers R, the G-graded-central-simple
real algebras have been considered in [BKpr]. The approach there is by Galois
descent from C to R. Hence any such algebra is obtained, up to isomorphism, by
Galois descent from a complex loop algebra, and checked to be isomorphic to a loop
algebra twisted by a character χ : G → S1, where S1 denotes the unit circle in C.
Let us finish this section by showing the connection of these "χ-twisted loop
algebras" over R with the cocycle twisted loop algebras considered here.
Given an abelian group G, the short exact sequence
ξ :
1
R×
C×
z
p
S1
z/¯z
1
induces a long exact sequence
(10)
1
Hom(G, R×)
Hom(G, C×)
Hom(G, S1)
δ
Ext(G, R×)
1
as Ext(G, C×) = 1 because C× is divisible.
Lemma 6.1. For any τ ∈ Z2
that, for any choice of elements zg ∈ C with z2
H2
sym(G, R×) there is a character χ ∈ Hom(G, S1) such
g = χ(g) for all g, [τ ] = [dγ] in
sym(G, R×), where γ is the map G → S1, g 7→ zg.
Note that dγ(g1, g2)2 = χ(g1)χ(g2)χ(g1g2)−1 = 1, so dγ(g1, g2) ∈ {±1} ⊆ R×
and hence dγ ∈ Z2
sym(G, R×), even though γ takes values in S1 ⊆ C×.
GRADED-SIMPLE ALGEBRAS AND COCYCLE TWISTED LOOP ALGEBRAS
11
Proof. The connecting homomorphism δ in (10) works as follows. Given any char-
acter χ : G → S1, let E be the pull-back of χ and p:
E = {(z, g) ∈ C× × G χ(g) = z/¯z},
with its natural projections π1 : E → C× and π2 : E → G. There appears the
commutative diagram with exact rows:
ξχ :
1
R×
ξ :
1
R×
π2
E
π1
C×
p
G
χ
S1
1
1
Then δ(χ) is the class in Ext(G, R×) of ξχ. For any choice of elements zg ∈ C with
z2
g = χ(g), the map G → E, g 7→ (zg, g), is a section of π2, and hence the element
in H2
sym(G, R×) corresponding to the extension ξχ is the class of the cocycle in
Z2
sym(G, R×) given by
(g1, g2) 7→ zg1 zg2z−1
g1g2 .
That is, identifying Ext(G, R×) with H2
The Lemma now follows from the surjectivity of δ.
sym(G, R×), δ(χ) = [dγ] with γ : g 7→ zg.
(cid:3)
Now, given an abelian group G, a subgroup H, and a central simple and G/H-
π(A) defined
graded real algebra A, and given a character χ : G → S1, the algebra Lχ
in [BKpr] is the subalgebra Lg∈G(cid:0)Ag ⊗ g(cid:1) of A ⊗R RdγG (γ as in Lemma 6.1),
and hence it is precisely the dγ-twisted loop algebra Ldγ
Lemma 6.1 and Proposition 3.5 show then that for any τ ∈ Z2
a character χ : G → S1 such that Lτ
π(A) ≃G Lχ
π(A).
π (A) (see Remark 5.1).
sym(G, R×), there is
Remark 6.2. Let B be a central simple complex algebra, and denote by BR the real
algebra obtained by restriction of scalars. That is, BR is just B, but considered
as a real algebra. The results in this paper reduce the study of gradings on BR to
gradings on the complex algebra B and gradings on real forms of B, i.e., central
simple real algebras A such that B is isomorphic to A ⊗R C, as follows.
Let G be an abelian group and let Γ : BR = Lg∈G(BR)g be a G-grading on BR.
The centroid C(B) = C(BR) is C1 (1 denotes here the identity map). Then either:
(1) The grading induced by Γ on C(B) is trivial: C(B) = C(B)e. This means
that each (BR)g is a complex subspace of B, so Γ is actually a G-grading
of the complex algebra B: Γ : B = Lg∈G
Bg.
Each isomorphism class of gradings (BR, Γ) corresponds to one or two iso-
morphism classes of gradings (B, Γ), because the index [Aut(BR) : Aut(B)]
may be 1 or 2.
(2) The grading induced by Γ on C(B) is not trivial. Then there is an element
h ∈ G of order 2 such that C(B)e = R1, C(B)h = Ri (identifying here i
with the map x 7→ ix). Then BR is G-graded-central-simple and Theorem
5.2 (and Corollaries 5.4 and 5.5) applies, so that, with H = hhi, we have
an isomorphism BR ≃G Lτ
π(A) for a central simple and G-graded real alge-
bra A and a symmetric 2-cocycle τ ∈ Hsym(G, R×). Since C× is divisible
H2
sym(G, C×) ≃ Ext(G, C×) is trivial, so there is a map γ : G → C× such
that τ = dγ. Then the map:
Φ : Lτ
π(A) −→ A ⊗R C,
x ⊗ g 7→ x ⊗ γ(g)
for g ∈ G and x ∈ Ag, is easily seen to be an isomorphism of G-graded real
algebras. Thus BR is isomorphic to A⊗R C as real algebras, and composing,
12
ALBERTO ELDUQUE
if necessary, with id ⊗ (complex conjugation), we check that B and A ⊗R C
are isomorphic complex algebras, so that A is a real form of B.
References
[ABFP08] B. Allison, S. Berman, J. Faulkner, and A. Pianzola, Realization of graded-simple
[BKpr]
[BN06]
[EK13]
[HS97]
[IW53]
algebras as loop algebras, Forum Math. 20 (2008), no. 3, 395 -- 432.
Y. Bahturin and M. Kochetov, On nonassociative graded-simple algebras over the field
of real numbers, preprint arXiv:1807.00057.
G. Benkart and E. Neher, The centroid of extended affine and root graded Lie algebras,
J. Pure Appl. Algebra 205 (2006), no. 1, 117 -- 145.
A. Elduque and M. Kochetov, Gradings on simple Lie algebras, Mathematical Surveys
and Monographs, vol. 189, American Mathematical Society, Providence, RI, 2013.
P.J. Hilton and U. Stammbach, A course in homological algebra Second edition. Grad-
uate Texts in Mathematics 4. Springer-Verlag, New York, 1997.
E. Inonu and E.P. Wigner, On the contraction of groups and their representations,
Proc. Nat. Acad. Sci. U.S.A. 39 (1953), 510 -- 524.
[dMP91] M. de Montigny and J. Patera, Discrete and continuous graded contractions of Lie
[MP91]
[Wei94]
algebras and superalgebras J. Phys. A 24 (1991), no. 3, 525 -- 547.
R.V. Moody and J. Patera, Discrete and continuous graded contractions of represen-
tations of Lie algebras, J. Phys. A 24 (1991), no. 10, 2227 -- 2257.
C.A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced
Mathematics 38, Cambridge University Press, Cambridge, 1994.
Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplica-
ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain
E-mail address: [email protected]
|
1804.09457 | 1 | 1804 | 2018-04-25T09:56:12 | On nilpotent generators of the special linear Lie algebra | [
"math.RA"
] | Consider the special linear Lie algebra $\mathfrak{sl}_n(\mathbb {K})$ over an infinite field of characteristic different from $2$. We prove that for any nonzero nilpotent $X$ there exists a nilpotent $Y$ such that the matrices $X$ and $Y$ generate the Lie algebra $\mathfrak{sl}_n(\mathbb {K})$. | math.RA | math | ON NILPOTENT GENERATORS OF THE LIE ALGEBRA sln
ALISA CHISTOPOLSKAYA
8
1
0
2
Abstract. Consider the special linear Lie algebra sln(K) over an infinite field of charac-
teristic different from 2. We prove that for any nonzero nilpotent X there exists a nilpotent
Y such that the matrices X and Y generate the Lie algebra sln(K).
r
p
A
5
2
]
.
A
R
h
t
a
m
[
1
v
7
5
4
9
0
.
4
0
8
1
:
v
i
X
r
a
1. introduction
It is an important problem to find a minimal generating set of a given algebra. This
problem was studied actively for semisimple Lie algebras. In 1951, Kuranishi [4] observed
that any semisimple Lie algebra over a field of characteristic zero can be generated by two
elements. Twenty-five years later, Ionescu [3] proved that for any nonzero element X of a
complex or real simple Lie algebra G there exists an element Y such that the elements X
and Y generate the Lie algebra G. In the same year, Smith [5] proved that every traceless
matrix of order n > 3 is the commutator of two nilpotent matrices. These results imply
that the special linear Lie algebra sln can be generated by three nilpotent matrices. In 2009,
Bois [2] extended Kuranishi's result to algebraically closed fields of characteristic different
from 2 and 3.
In this paper we obtain an analogue of Ionescu's result for nilpotent generators of the
Lie algebra sln.
Theorem 1. Let K be an infinite field of characteristic different from 2. For any nonzero
nilpotent X there exists a nilpotent Y such that the matrices X and Y generate the Lie
algebra sln(K).
Our interest to this subject was motivated by the study of additive group actions on affine
spaces, see [1, Theorem 5.17]. In a forthcoming publication we plan to extend Theorem 1
to arbitrary simple Lie algebras.
The author is grateful to her supervisor Ivan Arzhantsev for posing the problem and
permanent support.
2. main results
Let K be an infinite field with char K 6= 2. A set of elements λ1, . . . , λn (λi ∈ K) is called
consistent if the following conditions hold:
(1) λ1 + . . . + λn = 0;
(2) λi 6= 0 for all i;
(3) λi 6= λj for all i 6= j;
(4) λi − λj = λk − λl only for i = j, k = l or i = k, j = l.
Condition (1) defines (n − 1)-dimensional subspace W ⊆ Kn. Conditions (2)-(4) define
linear inequalities on W whose set of solutions is nonempty. For example, if char K = 0, a
set λi = 2i−1 for i = 1, . . . , n − 1 and λn = 1 − 2n−1 is consistent.
2010 Mathematics Subject Classification. Primary 17B05, 17B22; Secondary 15A04.
Key words and phrases. Nilpotent matrix, commutator, Lie algebra, generators.
1
2
ALISA CHISTOPOLSKAYA
A diagonal matrix A = diag(a11, . . . , ann) is called consistent if a11, . . . , ann is a consis-
tent set.
Lemma 1. Let T be a consistent matrix and A be a matrix with nonzero entries outside
the principal diagonal. Then T and A generate the Lie algebra sln(K).
Proof. Consider the following matrices:
A1 = [T, A], Ai = [T, Ai−1],
i = 2, . . . , n2 − n.
Note that all matrices Ai have zeroes on the principal diagonal and are linearly independent.
Indeed, if we consider the coordinates of these matrices in the basis of n2 − n matrix units,
up to scalar multiplication of columns, we come to a Vandermonde matrix with nonzero
determinant.
Hence, the matrices Ai form a basis of the space of n × n-matrices with zeroes on the
principal diagonal. This means that the subalgebra generated by T and A contains all
matrix units Eij, i 6= j. Since Eii −Ejj = [Eij, Eji], it follows that this subalgebra is sln(K).
(cid:3)
Lemma 2. Let C be a diagonal matrix with pairwise distinct nonzero entries on the prin-
cipal diagonal. Then C can be represented as C = A + B with A and B being nilpotent
matrices, rkA = 1, and all entries of A are nonzero.
Proof. Let V be an n-dimensional vector space over K. Let us fix a basis e1, . . . , en in V
and consider linear operators given by matrices in this basis.
For a non-degenerate matrix C = diag(c11, . . . , cnn), let us consider the following set of
vectors:
v1 = e1 + · · · + en, v2 =
e1
c11
+ · · · +
en
cnn
, . . . , vn =
e1
cn−1
11
+ · · · +
en
cn−1
nn
.
Again using a Vandermonde matrix we obtain that v1, . . . , vn form a basis in V . Moreover,
we have C(vi+1) = vi for all i = 1, . . . , n − 1. Let us define the operator B as follows:
B(v1) = 0, B(vi+1) = vi for i = 1, . . . , n − 1, and let A = C − B.
Let us check that A and B satisfy the conditions of Lemma 2. Indeed, B is nilpotent
and rkA = 1, because A(vi) = 0 for i = 1, . . . , n − 1. Since trA = 0, A is also nilpotent. It
is only left to show that in the basis e1, . . . , en all entries of A are nonzero. All columns of
A are proportional to the column (c11, . . . , cnn), because rkA = 1 and A(e1 + · · · + en) =
c11e1 + · · · + cnnen. Finally, we have A(ei) 6= 0, because the vectors v2, v3, . . . , vn, ei are
linearly independent and v2, v3, . . . , vn is a basis of KerA.
(cid:3)
Lemma 3. For each nilpotent matrix N with rkN = 1 there exists a nilpotent matrix M
such that N and M generate sln(K).
Proof. Let T be a consistent matrix, A and B be matrices from Lemma 2. It follows from
Lemma 1 that A and B generate sln(K). All nilpotent matrices of rank 1 are conjugate,
i.e. for any nilpotent N with rkN = 1 there exists C ∈ GLn(K) such that N = CAC −1.
Moreover, if A and B generate sln(K) and C ∈ GLn(K), then N and CBC −1 also generate
sln(K). This completes the proof.
(cid:3)
Lemma 4. Let V be a finite-dimensional vector space over an arbitrary field K. Then a set
of collections consisting of ordered n linearly independent vectors is open in the Zariski
topology on V n.
ON NILPOTENT GENERATORS OF THE LIE ALGEBRA sln
3
Proof. We fix a basis of V . For every set {v1, . . . , vn vl ∈ V } let us build a matrix
consisting of n rows and dimV columns such that l-th row consists of the coordinates of vl.
It is only left to notice that v1, . . . , vn are linearly independent if and only if there is at
least one nonzero minor of order n.
(cid:3)
Lemma 5. For any given matrix B a set of matrices X such that B and X generate sln(K)
is open in the Zariski topology on sln(K).
Proof. For any given matrix B and a variable X let us consider a set of all matrices that
can be obtained from B and X by means of the Lie bracket [ , ]:
Com(X, B) := {X, B, [X, B], [B, X], [X, [X, B]], [B, [X, B]], [[X, B], X], . . . }
Let us enumerate all the matrices from Com(X, B) in some order independent of X. For ex-
ample, we put Com1(X) = X, Com2(X) = B, Com3(X) = [X, B], Com4(X) = [X, [X, B]],
Com5(X) = [B, [X, B]], . . .
Obviously, a subalgebra of sln(K) generated by X and B is a linear span of elements of
Com(X, B). Let I be an arbitrary set of indices i1, . . . , in2−1 and let MI be a set of matrices
X such that Comi1(X), . . . , Comin
−1(X) are linearly independent. Let us construct a map
2
ϕI : sln(K) −→ (sln(K))n2
−1 by the rule X → (Comi1(X), . . . , Comin
2
−1(X)).
It is defined by polynomials. Let us look at ordered collections consisting of n2 − 1 linearly
independent matrices. According to Lemma 4, a set compiled from all such collections is
open in the Zariski topology on (sln(K))n2
−1. Then the preimage of this set under ϕI is
open. Lemma 5 follows from the fact that a set of matrices X such that B and X generate
sln(K) is a union of all possible MI.
(cid:3)
Lemma 6. For any matrix A = Pn−1
such that A and B generate sln(K).
i=1 aiEi i+1, a1 = 1, there exists a nilpotent matrix B
Proof. Lemma 3 implies that there exists a nilpotent matrix B0 such that E12 and B0
generate sln(K). Consider all matrices of the form
X =
n−1
X
i=1
xiaiEi i+1.
Obviously, matrices X and A are conjugate if xi 6= 0 for all i. According to Lemma 5, there
is a polynomial F (x1, . . . , xn−1) such that the following conditions hold:
(1) F (1, 0, . . . , 0) 6= 0;
(2) F (x1, . . . , xn−1) 6= 0 =⇒ B0 and X generate sln(K).
Since F is a nonzero polynomial, there exists a set of nonzero numbers λ1, . . . , λn−1 such that
F (λ1, . . . , λn−1) 6= 0. It implies that matrices B0 and X0 = Pn−1
i=1 λiaiEi i+1 generate sln(K).
The fact that A and X0 are conjugate completes the proof.
(cid:3)
Proof of Theorem 1. For any non-degenerate nilpotent linear operator X there exists a basis
such that the matrix of X in this basis has the following form:
A =
n−1
X
i=1
aiEi i+1, a1 = 1.
In other words, if X is a nonzero nilpotent matrix, there exists C ∈ GLn(K) such that
A = CXC −1. It follows from Lemma 6 that there exists a nilpotent matrix B such that A
and B generate sln(K). Thus, X and C −1BC generate sln(K).
(cid:3)
4
ALISA CHISTOPOLSKAYA
3. examples and problems
Let us give two examples with specific pairs of nilpotent matrices generating sln(K).
Matrices from the first example generate sln(K) over an infinite field K of arbitrary char-
acteristic except for n = 4 if char K = 2.
Example 1. Let K be an infinite field and λ1, . . . , λn (λi ∈ K) be a set such that following
conditions hold:
(1) λ1 + . . . + λn = 0;
(2) λi+1 6= λi for all i;
(3) λi+1 − λi = λk+1 − λk only for i = k;
(4) λ1 + · · · + λk 6= 0 for all k = 1, . . . , n − 1.
Such sets exist except for n = 4 if char K = 2 (in this case condition (1) implies λ44 − λ33 =
λ22 −λ11). Let us denote by sk the element λ1 +. . .+λk and consider the following matrices:
A =
0 1 0 . . . 0
0 0 1 . . . 0
...
...
0 0 0 . . . 1
0 0 0 . . . 0
. . .
...
...
B =
0
s1
...
0
0
0
. . .
0
. . .
...
. . .
. . . sn−2
. . .
0
0
0
...
0
0
0
...
0
sn−1 0
Let us show that A and B generate sln(K). Indeed, T = [A, B] is a diagonal matrix with
the entries tii = λi. Similarly to the proof of Lemma 1, we obtain that matrices T and A
generate a subalgebra of sln(K) containing Ei i+1 for i = 1, . . . , n − 1. Since [Eik, Ekl] = Eil,
this subalgebra contains all upper nil-triangular matrices. Similarly, T and B generate a
subalgebra containing all lower nil-triangular matrices. Since Eii − Ejj = [Eij, Eji], A and
B generate sln(K).
Example 2. If char K = 0 and n is odd, then the matrices
M =
0 1 0 . . . 0
0 0 1 . . . 0
...
...
0 0 0 . . . 1
0 0 0 . . . 0
. . .
...
...
N =
0 0 0 . . . 0
0 0 0 . . . 0
...
...
0 0 0 . . . 0
1 0 0 . . . 0
. . .
...
...
generate sln(K). Firstly let us consider the case K = C. It is possible to make a consistent
set from different complex n-th roots of unity, since if n is odd, a regular n-gon does not have
any parallel and equal sides/diagonals. Let T be a corresponding consistent matrix. It can
be represented as T = A + B, where A, B are the nilpotent matrices from Lemma 2. Using
notations of Lemma 2, in the basis v1, . . . , vn the operators A and B have matrices N = E1n
and M = Pn−1
i=1 Ei i+1, respectively. Thus, M and N generate sln(K). We conclude that
the set Com(M, N) from Lemma 5 contains n2 − 1 linearly independent matrices. Since
linear independence of matrices does not depend on the ground field, the matrices M and
N generate sln(Q) and sln(K), where K is an extension of the field Q.
Remark 1. If n is even, M and N do not generate sln(K).
ON NILPOTENT GENERATORS OF THE LIE ALGEBRA sln
5
Let us look at the set of matrices
Λ = {A ∈ sln(K) AC −1 + C −1AT = 0}, where C =
0 . . .
0
...
...
. . .
1 . . .
0
−1 0 . . .
(−1)n
...
0
0
.
If n > 3 then Λ is a proper subalgebra of sln(K), and if n is even, we have M, N ∈ Λ.
The proof of Theorem 1 implies that for any n > 1 there exists a number N such that
Theorem 1 holds for sln(K), K > N. It may be interesting to extend Theorem 1 to finite
fields and fields of characteristic 2.
Problem 1. What is the minimal generating set consisting of nilpotent matrices of the Lie
algebra sln over a finite field?
Problem 2. Does Theorem 1 hold for infinite field of characteristic 2?
The following example shows that at least some conditions of Theorem 1 are necessary.
Example 3. Let F2 be the field Z/2Z. Let us show that for K = F2, Theorem 1 does not
hold. We claim that for the matrix E12 ∈ sl3(F2) there does not exist a nilpotent matrix Y
such that E12 and Y generate sl3(F2).
Consider linear operators given by the matrices E12 and Y . If rkY = 1 then KerY and
KerE12 have nonempty intersection, hence the subalgebra generated by E12 and Y is not
sl3(F2). Thus rkY = 2. Since all nilpotent matrices of rank 2 are conjugate in sl3(F2),
we only have to check that there is no A of rank 1 such that A and B = E12 + E23
generate sl3(F2).
Since the first column and the last row of the matrix B are zero, the first column and
the last row of the matrix A are nonzero.
It implies that a31 = 1. There are only 8
such matrices. Two matrices of these eight are persymmetric and we can split other six
matrices into pairs symmetric with respect to the antidiagonal matrices. Let X ′ be a matrix
such that X and X ′ are symmetric with respect to the antidiagonal. Since symmetry and
antisymmetry are the same in F2 and B = B ′, we have [X, B]′ = [X ′, B]. It implies that
if X = X ′ the subalgebra sl3(F2) generated by X and B consists of matrices symmetric
with respect to the antidiagonal. Moreover, if X and B generate sl3(F2), then X ′ and B ′
generate sl3(F2). So it is only left to show that A and B do not generate sl3(F2), where A
is one of the following three matrices:
A1 =
1 1 0
1 1 0
1 1 0
, A2 =
1 0 1
1 0 1
1 0 1
, A3 =
0 0 0
1 0 0
1 0 0
Let us denote by Λ1 the linear span of the matrices
and by Λ2 the linear span of the matrices
1 1 0
1 1 0
1 1 0
0 1 0
0 0 1
0 0 0
1 1 1
0 0 1
0 0 1
1 0 1
1 0 1
0 1 1
1 0 1
1 0 1
1 0 1
0 1 0
0 0 1
0 0 0
1 1 1
1 1 1
0 1 0
1 0 0
0 0 1
0 0 1
0 1 1
0 0 1
0 0 0
6
ALISA CHISTOPOLSKAYA
We have A1, B ∈ Λ1 and A2, A3, B ∈ Λ2, and it is easy to verify directly that Λ1 and Λ2
are subalgebras of sln(K).
References
[1] Ivan Arzhantsev, Karine Kuyumzhiyan, and Mikhail Zaidenberg. Infinite transitivity, finite generation,
and Demazure roots. arXiv:1803.10610v1, 24 pages
[2] Jean-Marie Bois. Generators of simple Lie algebras in arbitrary characteristics. Math. Z. 262 (2009),
no. 4, 715-741
[3] Tudor Ionescu. On the generators of semisimple Lie algebras. Linear Algebra and its Applications 15
(1976), 271-292
[4] Masatake Kuranishi. On everywhere dense imbeddings of free groups in Lie groups. Nagoya Math. J. 2
(1951), 63-71
[5] John Howard Smith. Commutators of nilpotent matrices. Linear Multilinear Algebra 4 (1976), 17-19.
Lomonosov Moscow State University, Faculty of Mechanics and Mathematics, Depart-
ment of Higher Algebra, Leninskie Gory 1, Moscow, 119991 Russia
E-mail address: [email protected]
|
1110.5009 | 1 | 1110 | 2011-10-23T00:28:42 | On locally defined formations of soluble Lie and Leibniz algebras | [
"math.RA"
] | It is well-known that all saturated formations of finite soluble groups are locally defined and, except for the trivial formation, have many different local definitions. I show that for Lie and Leibniz algebras over a field of characteristic 0, the formations of all nilpotent algebras and of all soluble algebras are the only locally defined formations and that the latter has many local definitions. Over a field of non-zero characteristic, a saturated formation of soluble Lie algebras has at most one local definition but a locally defined saturated formation of soluble Leibniz algebras other than that of nilpotent algebras has more than one local definition. | math.RA | math |
ON LOCALLY DEFINED FORMATIONS OF SOLUBLE LIE AND
LEIBNIZ ALGEBRAS
DONALD W. BARNES
Abstract. It is well-known that all saturated formations of finite soluble
groups are locally defined and, except for the trivial formation, have many
different local definitions. I show that for Lie and Leibniz algebras over a field
of characteristic 0, the formations of all nilpotent algebras and of all soluble
algebras are the only locally defined formations and the latter has many local
definitions. Over a field of non-zero characteristic, a saturated formation of
soluble Lie algebras has at most one local definition but a locally defined
saturated formation of soluble Leibniz algebras other than that of nilpotent
algebras has more than one local definition.
1. Introduction
In the theory of finite soluble groups, a formation function is a function f which
assigns to each prime p, a formation f (p) of finite soluble groups. The formation
locally defined by f is the class Loc(f ) of all finite soluble groups such that, for
each chief factor H/K of G with p dividing H/K, we have G/CG(H/K) ∈ f (p).
(Here CG(H/K) denotes the centraliser in G of H/K.) It is a saturated formation.
Apart from the trivial formation which contains only the group of order 1, every
saturated formation of finite soluble groups has many such local definitions. This
theory is set out in Doerk and Hawkes [5].
If K is a formation of soluble Lie algebras over the field F , the formation locally
defined by K is the class of Lie algebras Loc(K) of those Lie algebras such that, for
each chief factor H/K of L, we have L/CL(H/K) ∈ K. It is a saturated formation.
In contrast to the group case, not every saturated formation of soluble Lie algebras
is locally defined. The theory of saturated formations of soluble Lie algebras is set
out in Barnes and Gastineau-Hills in [4]. The analogous theory for Leibniz algebras
is set out in Barnes [3].
For finite soluble groups, the formation function f is called integrated if f (p) ⊆
Loc(f ) for all primes p. It is called full if for all p, every extension of a p-group by
a group in f (p) is in f (p). For Lie and Leibniz algebras, trivially, K ⊆ Loc(K), thus
the analogue of "integrated" always holds. The significance of the condition "full"
is that if N is a normal p-subgroup of a group G and V is an irreducible G-module
over the field of p elements, then N acts trivially on V . There is no condition on an
ideal N of a Lie or Leibniz algebra L that ensures that N must act trivially on any
irreducible L-module. Thus there is no Lie or Leibniz analogue of the condition
"full".
2010 Mathematics Subject Classification. Primary 17B30, Secondary 20D10.
Key words and phrases. Lie algebras, Leibniz algebras, saturated formations, local definition.
1
2
DONALD W. BARNES
2. Lie algebras over a field of characteristic 0
The formation N of nilpotent Lie algebras is locally defined by the formation
{0} containing only the zero algebra. If K is a non-zero formation, then it contains
the formation A of abelian algebras. But if H/K is a chief factor of any soluble
Lie algebra L over a field of characteristic 0, then L/CL(H/K) is abelian. Thus for
any non-zero formation K, we have Loc(K) = S, the formation of all soluble Lie
algebras.
For Lie algebras in contrast to the case for groups, it is convenient to require
that formations be non-empty. If we allow formations to be empty, then the zero
formation {0} = Loc(∅) and this is its only local definition.
We thus have, for a field F of characteristic 0, N and S are the only locally
defined formations, N has the unique local definition N = Loc({0}) while S =
Loc(K) for any non-zero formation K. If F is algebraically closed, then {0}, N and
S are the only saturated formations, but if F is not algebraically closed, then other
saturated formations exist. (See Barnes [2].)
3. Lie algebras over a field of characteristic p 6= 0
Let F be a field of characteristic p 6= 0. Then not all saturated formations of
soluble Lie algebras over F are locally definable, for example, the formation U of
supersoluble Lie algebras is saturated but not locally definable. I show that if the
saturated formation F is locally defined, then the formation K with Loc(K) = F is
uniquely determined by F.
Denote the nil radical of L by N (L). The saturated formation locally defined
by K consists of those algebras L such that L/N (L) ∈ K, algebras which are the
extension of a nilpotent algebra N = N (L) by an algebra in K. In a widely used
terminology, they are nilpotent by K. It makes sense to denote the locally defined
formation by NF. I reverse this construction.
Definition 3.1. Let F be a saturated formation. We define the quotient of F by
N to be
F/N = {L/A L ∈ F, A E L, N (L) ⊆ A}.
Lemma 3.2. Let F be a saturated formation of soluble Lie algebras over any field.
Then F/N is a formation.
Proof. Put K = F/N. By its definition, K is quotient closed. Suppose that Ai E L
and that L/Ai ∈ K, i = 1, 2. We have to prove that L/(A1 ∩ A2) ∈ F. We may
suppose without loss of generality that A1 ∩ A2 = 0.
There exist Xi ∈ F with N (Xi) ⊆ Bi E Xi and Xi/Bi ≃ L/Ai. Then L/Ai ∈ F,
so L ∈ F.
We have epimorphisms φi : Xi/Bi → L/(A1 + A2) and can identify L with
{(y1, y2) ∈ X1/B1 ⊕ X2/B2 φ1(y1) = φ2(y2)}. Let ψi be the composite of the
natural homomorphism Xi/N (Xi) → Xi/Bi with φi and let
X = {(x1, x2) ∈ X1/N (X1) ⊕ X2/N (X2) ψ1(x1) = ψ2(x2)}.
Then the natural homomorphism X1/N (X1)⊕X2/N (X2) → X1/B1 ⊕X2/B2 maps
the subalgebra X onto L. As K is quotient closed, it is sufficient to prove that X ∈ K.
But this is the special case Bi = N (Xi).
The chief factors of Xi are F-central irreducible L-modules on which B1 and B2
act trivially. Let V be the direct sum of the chief factors of X1 and X2. Then the
LOCALLY DEFINED FORMATIONS
3
split extension E of V by L is in F. Since the intersection of the centralisers in Xi
of the chief factors of Xi is Bi, the intersection of the centralisers in E of the chief
factors of E is V and N (E) = V . Thus L ∈ K.
(cid:3)
The above lemma holds over any field, but the following theorem needs the
assumption of non-zero characteristic.
Theorem 3.3. Let F = Loc(K) be a locally defined formation of soluble Lie algebras
over a field F of non-zero characteristic. Then K = F/N.
Proof. Clearly, F/N ⊆ K. Suppose L ∈ K. By Jacobson [7, Theorem VI.2, p. 205],
there exists a faithful, completely reducible L-module V . Let V = ⊕Vi where the
Vi are irreducible. Let Ki be the kernel of the representation of L on Vi. Since
L/Ki ∈ K, the split extension Ei of Vi by L/K)i is in F. The representation of
L/Ki on Vi is faithful, so N (Ei) = Vi, and L/Ki ∈ F/N. But ∩i(Ki) = 0 and it
follows that L ∈ F/N.
(cid:3)
4. Leibniz algebras
For Leibniz algebras, the situation is a little more complicated than for Lie
algebras. A left Leibniz algebra is a linear algebra L whose left multiplication
operators da : L → L given by da(x) = ax for a, x ∈ L, are derivations. The basic
theory of Leibniz algebras is set out in Loday and Pirashvili [6]. The theory of
saturated formations of soluble Leibniz algebras is given in Barnes [3].
The subspace Leib(L) = hx2 x ∈ Li is an abelian ideal of L and L/ Leib(L)
is a Lie algebra. Further, Leib(L)L = 0. If V is an irreducible L-module, then by
a theorem of Loday and Pirashvili [6], L/CL(V ) is a Lie algebra and V is either
symmetric (vx = −xv for x ∈ L and v ∈ V ) or antisymmetric (V L = 0). It follows
that if F = Loc(K) is a locally defined formation of soluble Leibniz algebras, then
F = Loc(LieK) where LieK is the class of all Lie algebras in K.
The characteristic 0 case is very like that of Lie algebras. The saturated for-
mation N of nilpotent Leibniz algebras is locally defined by the zero formation. If
K is a non-zero formation, then it contains the abelian algebras and Loc(K) = S,
the class of all soluble Leibniz algebras. As for Lie algebras, N and S are the only
locally defined formations.
The case of characteristic p 6= 0 is closely tied to the corresponding case for Lie
algebras. By Barnes [3, Corollary 3.17], if F is a saturated formation of soluble
Leibniz algebras, then LieF is a saturated formation of soluble Lie algebras and we
have a one to one correspondence F ↔ LieF. If F = Loc(K), then both F and LieF are
locally defined in the corresponding categories by LieK. We thus have the following
theorem.
Theorem 4.1. Let F be a locally defined formation of soluble Leibniz algebras over
a field of charactistic p 6= 0. Then there exists a unique formation K0 of soluble Lie
algebras with F = Loc(K0).
However, except for the case of N = Loc({0}), there are always other formations
K of soluble Leibniz algebras with Loc(K) = Loc(K0).
Lemma 4.2. Let K0 6= {0} be a formation of soluble Lie algebras. Let P1, . . . , Pn
be soluble Leibniz algebras such that Pi/ Leib(Pi) ∈ K0 and let K be the smallest
formation of soluble Leibniz algebras containing K0 and the algebras Pi. Then
LieK = K0 and Loc(K) = Loc(K0).
4
DONALD W. BARNES
Proof. We have to show that, if L ∈ K, then L/ Leib(L) ∈ K0. Since L ∈ K, there
exist ideals N1, . . . Nr of L with ∩iNi = 0 such that, for each i, either L/Ni ∈ K0
or L/Ni ≃ Pj for some j. We have the natural epimorphisms φi : L → L/Ni
and L may be regarded as the subalgebra of ⊕iL/Ni whose elements are the
(φ1(x), . . . , φr(x)) for x ∈ L. But φi maps x2 to φi(x)2, thus maps Leib(L) onto
Leib(L/Ni) and so gives an epimorphism ψi : L/ Leib(L) → (L/Ni)/ Leib(L/Ni).
This expresses L/ Leib(L) as a subdirect sum of the (L/Ni)/ Leib(L/Ni). But for
all i, (L/Ni)/ Leib(L/Ni) ∈ K0. As ψi(L/ Leib(L)) ∈ K0 for all i, L/ Leib(L) ∈ K0
as asserted.
(cid:3)
We can always construct Leibniz algebras P with P/ Leib(P ) ∈ K0. Let L be
any Lie algebra in K0 and let V be any non-trivial left L-module. We make this
into a Leibniz module by defining V L = 0. Let P be the split extension of V
by L. Then P is not a Lie algebra, so not in K0. We have Leib(P ) = V and so
P/ Leib(P ) ≃ L ∈ K0. Thus we can always produce K 6= K0 with Loc(K) = Loc(K0).
References
1. D. W. Barnes, On F-hypercentral modules for Lie algebras, Archiv der Math. 30 (1978), 1 -- 7.
2. D. W. Barnes, Saturated formations of soluble Lie algebras in characteristic 0, Archiv der
Math. 30 (1978), 477 -- 480.
3. D. W. Barnes, Schunck classes of soluble Leibniz algebras, arXiv:1101.3046 (2011).
4. D. W. Barnes and H. M. Gastineau-Hills, On the theory of soluble Lie algebras, Math. Zeitschr.
106 (1968), 343 -- 354.
5. K. Doerk and T. Hawkes, Finite soluble groups, DeGruyter, Berlin-New York 1992.
6. J.-L. Loday and T. Pirashvili, Leibniz representations of Lie algebras, J. Algebra 181 (1996),
414 -- 425.
7. N. Jacobson, Lie algebras Interscience, New York-London (1962).
1 Little Wonga Rd., Cremorne NSW 2090, Australia,
E-mail address: [email protected]
|
1610.05178 | 1 | 1610 | 2016-10-17T15:54:11 | Tensor products of faithful modules | [
"math.RA"
] | If $k$ is a field, $A$ and $B$ $k$-algebras, $M$ a faithful left $A$-module, and $N$ a faithful left $B$-module, we recall the proof that the left $A\otimes_k B$-module $M\otimes_k N$ is again faithful. If $k$ is a general commutative ring, we note some conditions on $A,$ $B,$ $M$ and $N$ that do, and others that do not, imply the same conclusion. Finally, we note a version of the main result that does not involve any algebra structures on $A$ and $B.$ | math.RA | math |
Comments, corrections, and related references welcomed, as always!
TEXed July 5, 2021
TENSOR PRODUCTS OF FAITHFUL MODULES
GEORGE M. BERGMAN
Abstract. If k is a field, A and B k-algebras, M a faithful left A-module, and N a faithful left
B-module, we recall the proof that the left A ⊗k B-module M ⊗k N is again faithful. If k is a general
commutative ring, we note some conditions on A, B, M and N that do, and others that do not, imply the
same conclusion. Finally, we note a version of the main result that does not involve any algebra structures
on A and B.
I needed Theorem 1 below, and eventually found a roundabout proof of it. Ken Goodearl found a simpler
proof, which I simplified further to the argument given here. But it seemed implausible that such a result
would not be classical, and I posted a query [1], to which Benjamin Steinberg responded, noting that Passman
had proved the result in [4, Lemma 1.1]. His proof is virtually identical to one below.
In the mean time, I had made some observations on what is true when the base ring k is not a field, and
on the "irrelevance" of the algebra structures of A and B, and added these to the write-up. So while I no
longer expect to publish this note, I will arXiv it, and keep it online as an unpublished note, to make those
observations available.
(Ironically, I eventually found an easier way to get the result for which I had needed Theorem 1.)
1. The main statement, and a generalization
Except where the contrary is stated, we understand algebras to be associative, but not necessarily unital.
For k a commutative ring, A and B k-algebras, M a left A-module, and N a left B-module, we recall
the natural structure of left A ⊗k B-module on M ⊗k N : An element
(1)
f = P1≤i≤n ai ⊗ bi ∈ A ⊗k B,
where ai ∈ A, bi ∈ B,
acts on decomposable elements u ⊗ v (u ∈ M, v ∈ N ) of M ⊗k N by
(2)
u ⊗ v 7→ Pi aiu ⊗ biv.
Since the right-hand side is bilinear in u and v, this map extends k-linearly to general elements of M ⊗k N.
The resulting action is easily shown to be compatible with the k-algebra structure of A ⊗k B.
Theorem 1. Let k be a field, A and B k-algebras, M a faithful left A-module, and N a faithful left
B-module. Then the left A ⊗k B-module M ⊗k N is also faithful.
Proof (after K. Goodearl, D. Passman). Given nonzero f ∈ A ⊗k B, we wish to show that it has nonzero
action on M ⊗k N. Clearly, we can choose an expression (1) for f such that the bi are k-linearly indepen-
dent. (We could simultaneously make the ai k-linearly independent, but will not need to.) Since we have
assumed (1) nonzero, not all of the ai are zero; so as M is a faithful A-module, we can find u ∈ M such
that not all the aiu ∈ M are zero. Hence there exists a k-linear functional ϕ : M → k such that not all
of the ϕ(aiu) are zero. Since the bi are k-linearly independent, the element Pi ϕ(aiu) bi ∈ B will thus be
nonzero. So as N is a faithful B-module, we can choose v ∈ N such that
(3)
( Pi ϕ(aiu) bi) v 6= 0
in N.
We claim that for the above choices of u and v, if we apply f to u ⊗ v ∈ M ⊗k N, the result, i.e., the
right-hand side of (2), is nonzero. For if we apply to that element the map ϕ ⊗ idN : M ⊗k N → k ⊗k N ∼= N,
we get the nonzero element (3). Thus, as required, f has nonzero action on M ⊗k N.
(cid:3)
2010 Mathematics Subject Classification. Primary: 16D10, 16S99.
Key words and phrases. faithful module; tensor-product module over a tensor product of algebras.
This preprint is readable online at http://math.berkeley.edu/~gbergman/papers/unpub/ .
1
2
GEORGE M. BERGMAN
The above result assumes k a field. Succumbing to the temptation to examine what the method of proof
can be made to give in the absence of that hypothesis, we record
Corollary 2 (to proof of Theorem 1). Let k be a commutative ring, A and B k-algebras, M a faithful
left A-module, and N a faithful left B-module.
Suppose, moreover, that elements of M can be separated by k-module homomorphisms M → k, and that
every finite subset of B belongs to a free k-submodule of B.
Then the left A ⊗k B-module M ⊗k N is again faithful.
Proof. Exactly like the proof of Theorem 1. The added hypothesis on B is what we need to conclude that
any element of A ⊗k B can be written in the form (1) with k-linearly independent bi; the added hypothesis
on M is what we need to construct ϕ. (Of course, the parenthetical comment in the proof of Theorem 1
about also making the ai k-linearly independent does not go over.)
(cid:3)
Warren Dicks (personal communication) pointed out early on another proof of Theorem 1: The actions
of A and B on M and N yield embeddings A → Endk(M ) and B → Endk(N ). Taking k-bases X
and Y of M and N, one can regard the underlying vector spaces of Endk(M ) and Endk(N ) as M X
and N Y . By two applications of [2, II, §3, №7, Cor. 3 to Prop. 7, p.AII.63], or one of [3, Theorem 2], one
concludes that the natural map M X ⊗k N Y → (M ⊗k N )X×Y is an embedding, and deduces that the map
A ⊗k B → End(M ⊗k N ) is an embedding, the desired conclusion.
2. Counterexamples to variant statements
The hypotheses of the above corollary are strikingly asymmetric in the pairs (A, M ) and (B, N ). We
can, of course, get the same conclusion if we interchange the assumptions on these pairs. But what if we try
to use one or the other hypothesis on both pairs; or concentrate both hypotheses on one of them?
It turns out that none of these modified hypotheses guarantees the stated conclusion. Here are three
closely related constructions that give the relevant counterexamples. In all three examples, the k-algebras
A and B are in fact commutative and unital.
Lemma 3. Let C be a commutative principal ideal domain with infinitely many primes ideals, and let P0,
P1 be two disjoint infinite sets of nonzero prime ideals of C. Then for k, A, M, B, N specified in each
of the following three ways, the A-module M and the B-module N are faithful, and A ⊗k B is nonzero, but
M ⊗k N is zero, hence not faithful. In each example, the variant of the hypotheses of Corollary 2 satisfied
by that example is noted at the end of the description.
(i) Let A = B = k = C, and let M = Lp∈P0 k/p and N = Lp∈P1 k/p. In this case, every finite subset
of A or of B (trivially) lies in a free k-submodule of that algebra.
In the remaining two examples, let C + be the commutative C-algebra obtained by adjoining to C an
indeterminate yp for each p ∈ P0 ∪ P1, and imposing the relations p yp = {0} for each such p.
(ii) Let k = C +,
let M be the ideal of k generated by the yp for p ∈ P0,
let N be the ideal of k
generated by the yp for p ∈ P1,
let A = k/N, and let B = k/M. Note that since M N = {0}, we may
regard M as an A-module and N as a B-module. In this case, M and N embed in k, hence k-module
homomorphisms from each of those modules into k separate elements.
(iii) Let k = A = C +, and let M be the ideal of A generated by all the yp. (The partition of our infinite
family of primes into P0 and P1 will not be used here.) On the other hand, let B be the field of fractions
of C, made a k-algebra by first mapping k to C by sending the yp to 0, then mapping C to its field of
fractions; and let N be any nonzero B-vector-space. In this case, every finite subset of A (trivially) lies in
a free k-submodule, and k-module homomorphisms into k clearly separate elements of M.
Proof. The fact that P0 and P1 are infinite sets of primes in the commutative principal ideal domain C
implies that each of those sets has zero intersection, hence that the M and N of (i) are faithful C-modules,
equivalently, are a faithful A-module and a faithful B-module. That their tensor product is zero is clear.
Similar considerations show that in (ii), any element of A or B with nonzero constant term in C acts
nontrivially on M, respectively, N. On the other hand, a nonzero element of A or B with zero constant
term, i.e., a nonzero element u of the ideal M or N, will act nontrivially on the module M, respectively,
N, because for some p in P0, respectively P1, the element u must involve a nonzero polynomial in yp,
which will have nonzero action on yp ∈ M or N as the case may be. (Since yp yp′ = {0} for p 6= p′, every
TENSOR PRODUCTS OF FAITHFUL MODULES
3
element of M or N is a sum of one-variable polynomials in the various yp with zero constant term.) Again,
it is clear that M ⊗k N = {0}. On the other hand, A ⊗k B = (k/M ) ⊗k (k/N ) ∼= k/(M + N ) ∼= C 6= {0}.
In case (iii), M is faithful over A for the same reason as in (ii), while faithfulness of N over B is clear,
as is the condition M ⊗k N = {0}. On the other hand, since A admits a homomorphism into C, we have
A ⊗k B 6= {0}.
(cid:3)
We remark that in the above constructions, the condition that the principal ideal domain k have infinitely
many primes can be weakened to say that it has at least two primes, p0 and p1, by replacing the modules
Lp∈P0 k/p and Lp∈P1 k/p in (i) with Ln>0 k/pn
1 , and making similar adjustments in (ii)
and (iii). (In the last, only one prime is needed.) One just has to be a little more careful in the proofs.
0 and Ln>0 k/pn
In another direction, one may ask:
Question 4 (suggested by A. Ogus, personal communication). If we add to the hypotheses of Corollary 2 the
condition that M be finitely generated as an A-module, and/or that N be finitely generated as a B-module,
can some of the other hypotheses of that corollary be weakened, dropped, or modified (perhaps in some of the
ways that Lemma 3 shows is not possible without such finite generation conditions)?
3. A and B don't have to be algebras
The sharp-eyed reader may have noticed that the proof of Theorem 1 makes no use of the algebra
structures of A and B. This led me to wonder whether the result was actually a special case of a statement
that involved no such structure. As Theorem 5 below shows, the answer is, in a way, yes. But as the second
proof of that theorem shows, one can equally regard Theorem 5 as a special case of Theorem 1.
Given a commutative ring k, and k-modules A, M0 and M1, let us define an action of A on (M0, M1)
to mean a k-linear map A → Homk(M0, M1), which will be written (a, u) 7→ au (a ∈ A, u ∈ M0, au ∈ M1);
and let us call such an action faithful if it is one-to-one as a map A → Hom(M0, M1).
Given two actions, one of a k-module A on a pair (M0, M1) and the other of a k-module B on a pair
(N0, N1), we see that an action of A ⊗k B on (M0 ⊗k N0, M1 ⊗k N1) can be defined just as for algebras,
with each element (1) acting by (2).
Theorem 5. Let k be a field, and suppose we are given an action of a k-vector-space A on a pair (M0, M1)
and an action of a k-vector-space B on a pair (N0, N1).
Then if each of these actions is faithful, so is the induced action of the k-vector-space A ⊗k B on the pair
(M0 ⊗k N0, M1 ⊗k N1).
First proof. Exactly like proof of Theorem 1. (Note that u will be chosen from M0, while ϕ will be a
k-linear functional on M1; and that v will be chosen from N0 to make (3) hold in N1.)
Second proof. Let us make A and B into k-algebras by giving them zero multiplication operations. Then
we can make the vector space M = M0 ⊕ M1 a left A-module using the action a(u0, u1) = (0, a u0), and
similarly make N = N0 ⊕ N1 a B-module. The faithfulness hypotheses on the given actions clearly make
these modules faithful.
Hence by Theorem 1, M ⊗k N is a faithful A ⊗k B-module. Now M ⊗k N is a fourfold direct sum
(M0 ⊗k N0) ⊕ (M0 ⊗k N1) ⊕ (M1 ⊗k N0) ⊕ (M1 ⊗k N1); but the action of A ⊗k B annihilates all summands
but the first, and has image in the last, so its faithfulness means that every nonzero element of A ⊗k B
induces a nonzero map from M0 ⊗k N0 to M1 ⊗k N1, which is the desired conclusion.
(cid:3)
Of course, the analog of Corollary 2 holds for actions as in Theorem 5.
4. Remarks
Composing the proof of Theorem 5 from Theorem 1, and the proof of Theorem 1 from Theorem 5, we see
that the general case of Theorem 1 follows from the zero-multiplication case of the same theorem. This is
striking, since the main interest of the result is for algebras with nonzero multiplication.
One may ask why I made the convention that algebras are associative, if the algebra operations were not
used in the theorem. The answer is that there is no natural definition of a module over a not-necessarily-
associative algebra. (There is a definition of a module over a Lie algebra, based on the motivating relation
between Lie algebras and associative algebras. But there is no natural definition of a Lie or associative
structure on a tensor product of Lie algebras, so the result can't be used in that case.)
4
GEORGE M. BERGMAN
Which of Theorems 1 and 5 is the "nicer" result? I would say that Theorem 5 shows with less distraction
what is going on, while Theorem 1 is likely to be more convenient for applications.
References
[1] George Bergman, http://mathoverflow.net/questions/247332/is-this-result-on-tensor-products-of-faithful-modules-known .
[2] N. Bourbaki, ´El´ements de math´ematique. Alg´ebre, Chapitres 1 ´a 3. Hermann, Paris 1970. MR0274237
[3] K. R. Goodearl, Distributing tensor product over direct product, Pacific J. Math. 43 (1972) 107 -- 110. MR0311714
[4] D. S. Passman, Elementary bialgebra properties of group rings and enveloping rings: an introduction to Hopf algebras,
Comm. Algebra 42 (2014) 2222 -- 2253. MR3169701
Department of Mathematics, University of California, Berkeley, CA 94720-3840, USA
E-mail address: [email protected]
|
1308.5104 | 1 | 1308 | 2013-08-23T11:43:08 | Verma modules for Iwasawa algebras are faithful | [
"math.RA",
"math.NT",
"math.RT"
] | We establish the faithfulness of Verma modules for rational Iwasawa algebras of split semisimple compact $L$-analytic groups. We also prove the algebraic independence of Arens-Michael envelopes over Iwasawa algebras and compute the centre of affinoid enveloping algebras of semisimple $p$-adic Lie algebras. | math.RA | math |
VERMA MODULES FOR IWASAWA ALGEBRAS
ARE FAITHFUL
KONSTANTIN ARDAKOV AND SIMON WADSLEY
To Peter Schneider, on the occasion of his sixtieth birthday
Abstract. We establish the faithfulness of Verma modules for rational Iwa-
sawa algebras of split semisimple compact L-analytic groups. We also prove
the algebraic independence of Arens-Michael envelopes over Iwasawa algebras
and compute the centre of affinoid enveloping algebras of semisimple p-adic
Lie algebras.
1. Introduction
1.1. Prime ideals in Iwasawa algebras. The majority of work so far related to
the study of the prime spectrum of non-commutative Iwasawa algebras has pro-
duced negative results. By this we mean that results in this area have tended to
put constraints on the set of prime ideals for such a ring rather than uncover prime
ideals that were not known before: see for example [33], [34], [6], [4], [2], [8]. This
work continues in that tradition. However, most of these theorems were established
in characteristic p in the first instance with eventual consequences in characteristic
zero; by contrast, our methods here have a definite characteristic zero flavour and
our results do not have immediate implications in positive characteristic.
Suppose that L is a finite extension of Qp and that K is a complete discretely
valued field extension of L. Let G be a compact open subgroup of the group of
L-points of a connected, simply connected, split semisimple affine algebraic group
scheme G defined over OL, and write KG to denote the Iwasawa algebra of con-
tinuous K-valued distributions on G. The annihilator of every simple KG-module
that is finite dimensional over K is a prime ideal of finite codimension in KG, and
moreover every prime ideal with this property will arise in this way. Evidence so
far suggests that non-zero prime ideals in KG that do not arise in this way are very
scarce; indeed we suspect that when the algebraic group scheme G is simple and G
has trivial centre they do not exist. We present further evidence in that direction.
1.2. The main result. A natural place to look for more prime ideals in KG is
as annihilators of simple KG-modules that are not finite dimensional over K. By
standard arguments in ring theory such ideals will always be prime and of infinite
codimension in KG. Thus if our suspicion above is correct then all such annihilators
must be zero. We show that this is the case for a large class of naturally arising
examples. More precisely, in §5.7 below we prove
Theorem A. Let p be an odd very good prime for G and let G be an open subgroup
of G(OL) with trivial centre. Let B+ be a Borel subgroup scheme of G and let
Kθ be a 1-dimensional locally L-analytic K-representation of B+ := G ∩ B+(OL).
Then the induced KG-module KG ⊗KB+ Kθ is faithful.
2010 Mathematics Subject Classification. 16S99, 22E50, 16D25.
1
2
KONSTANTIN ARDAKOV AND SIMON WADSLEY
We refer the reader to [3, §6.8] for a precise definition of what it means for a
prime number p to be a very good prime for G and simply remark here that this
condition is satisfied by any p > 5 if G is not of type A. These "Verma modules"
KG ⊗KB+ Kθ are not always irreducible, but it follows from the work of Orlik
and Strauch [26, Theorem 3.5.2] that generically they are irreducible at least when
L = Qp.
We note that Theorem A refutes the main result in [16] whose proof has been
known to contain a gap for a number of years, and whose statement was already
known to be false for open pro-p subgroups G of SL2(Zp) following the work of
Wei, Zhang and the first author [6], [4], using very different methods to those found
in this paper.
1.3. Two related results. We also prove some other results of independent inter-
est. Write D(H, K) to denote the algebra of L-locally analytic K-valued distribu-
tions on a compact L-analytic group H in the sense of Schneider and Teitelbaum
[31]. There is a natural map from the Iwasawa algebra KH to D(H, K) because
every L-locally analytic function on H is continuous. We may also consider the sub-
algebra D(H, K)1 consisting of those distributions in D(H, K) that are supported
at the identity in the sense of [20]. At the end of §3 we prove the following result,
which is essential to our proof of Theorem A.
Theorem B. The natural map KH ⊗K D(H, K)1 → D(H, K) is an injection.
We note that in the case H = Zp an immediate consequence of Theorem B is
the well-known algebraic independence of the logarithmic series log(1 + T ) over the
Iwasawa algebra OK[[T ]], so Theorem B may be viewed as a (slightly stronger)
non-commutative analogue of this algebraic independence.
In [3, §9.3] we promised a future proof that the centre of the affinoid enveloping
algebra \U (g)n,K is the closure of the image of the centre of U (gK) in \U (g)n,K:
Theorem C. We have Z(cid:16) \U (g)n,K(cid:17) = \U (g)G
n,K.
We provide a proof of Theorem C in §4.4 of this paper, which is much simpler
than that found in [5] for the case n = 0.
1.4. Future work. We believe that this work raises two interesting questions. By
an affinoid highest weight module we mean a module that can be written in the
form \U (g)n,K ⊗U(g) M for some highest weight U (gK)-module M . Recall that an
ideal that arises as the annihilator of a simple module is called primitive.
Question A. Is it the case that every primitive ideal of \U (g)n,K with K-rational
infinitesimal central character is the annihilator of a simple affinoid highest weight
module?
Some evidence pointing towards a positive answer to Question A is provided by
Duflo's main theorem in [14] that states that every primitive ideal of the classical
enveloping algebra U (gK) with K-rational infinitesimal central character is the
annihilator of a highest weight module. In particular to answer yes, it would suffice
to prove that every primitive ideal of \U (g)n,K is controlled by U (gK).
Question B. Is every affinoid highest weight module that is not finite-dimensional
over K faithful as a KG-module?
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
3
Since Verma modules for classical enveloping algebras are generically irreducible,
our Theorem 5.4 below may be viewed as giving evidence towards a positive answer
to Question B. We believe that if we could give positive answers to both these
questions then, in the case L = Qp, we would be able to use the faithful flatness of
D(G, K) over KG due to Schneider and Teitelbaum [32] together with our affinoid
version [3, Theorem D] of Quillen's Lemma to prove that every non-zero prime ideal
of KG is the annihilator of a finite dimensional simple module.
1.5. Acknowledgements. The first author would like James Zhang for the invi-
tation to visit Seattle in 2012, and for his encouragement that led to the eventual
proof of Theorem 5.4. Many of the results in this paper were established during
the Banff 2013 Workshop "Applications of Iwasawa Algebras"; we thank its organ-
isers for the invitation to visit and for the opportunity to speak about this work
there. The second author thanks Homerton College for funding his travel to this
workshop.
2. Generalities on completed group rings
2.1. Module algebras and smash products. Let k be a commutative base ring.
Recall [25, Chapter 4] that if H is a Hopf algebra over k and A is a k-algebra, then
A is a left H-module algebra if there exists an action
such that
H ⊗k A → A,
r ⊗ a 7→ r · a
r · (ab) = (r1 · a)(r2 · b),
r · 1 = ǫ(r)1,
(rs) · a = r · (s · a) and 1 · a = a
for all r, s ∈ H and a, b ∈ A. Here we use the sumless Sweedler notation. There
is a similar notion of right H-module algebra, and the two notions coincide in the
case when H is commutative.
Whenever A is a left H-module algebra, define A#H := A ⊗k H and write a#r for
the tensor a ⊗ r in A#H. Then A#H becomes an associative k-algebra called the
smash product of A with H, with multiplication given by
(a#r)(b#s) = a(r1 · b)#r2s
for all a, b ∈ A and r, s ∈ H.
This smash product contains A and H as k-subalgebras, and A is naturally a left
A#H-module via the rule
(a#r) · b = a(r · b)
for all a, b ∈ A and r ∈ H.
Note that the subset of H-invariants in A, namely
AH := {a ∈ A : r · a = ǫ(r)a for all
r ∈ H}
is always a k-subalgebra of A. We have the following well-known
Lemma. Let H be a Hopf algebra over k and let A be a left H-module algebra.
Then
(a) A is an A#H -- AH -bimodule, and
(b) EndA#H A = (AH )op.
Proof. (a) The left regular representation of A on itself commutes with the right
regular representation, so we have to check that every r ∈ H acts on A by a right
AH -module endomorphism:
r · (ab) = (r · a)b
for all
r ∈ H, a ∈ A and b ∈ AH .
4
KONSTANTIN ARDAKOV AND SIMON WADSLEY
Now r · (ab) = (r1 · a)(r2 · b) = (r1 · a)(ǫ(r2)b) = ((r1ǫ(r2)) · a)b. But r1ǫ(r2) = r by
the counit axiom in H. Therefore r · (ab) = (r · a)b as required.
(b) Let ϕ : A → A be a left A#H-module endomorphism. Since ϕ is H-linear,
r · ϕ(1) = ϕ(r · 1) = ϕ(ǫ(r)1) = ǫ(r) · ϕ(1)
for all
r ∈ H,
which shows that ϕ(1) ∈ AH . Since ϕ is left A-linear,
ϕ(x) = ϕ(x1) = xϕ(1)
for all x ∈ A
and therefore ϕ agrees with right muliplication by ϕ(1) ∈ AH . Hence the anti-
homomorphism AH → EndA#H A which sends r ∈ AH to right multiplication by r
is a bijection.
(cid:3)
2.2. Locally constant functions. Let G be a profinite group. Recall that a
function f : G → k is locally constant if for all g ∈ G there is an open neighbourhood
U of g such that f is constant on U .
Definition. Let C∞ = C∞(G, k) denote the set of all locally constant functions
from G to k.
C∞ becomes a unital commutative k-algebra when equipped with pointwise mul-
tiplication of functions. Moreover it is a Hopf algebra over k, with comultiplication
∆, antipode S and counit ǫ given by the formulas
∆(f )(g, h) = f (gh), S(f )(g) = f (g−1) and ǫ(f ) = f (1)
for all f ∈ C∞ and all g, h ∈ G.
2.3. G-graded algebras. We recall [7, Definition 2.5] for the convenience of the
reader.
Definition. Let G be a profinite group and let A be a k-algebra. We say that A
is G-graded if for each clopen subset U of G there exists a k-submodule AU of A
such that
(i) A = AU1 ⊕ AU2 ⊕ · · · ⊕ AUn if G = U1 ∪ · · · ∪ Un is an open partition of G,
(ii) AU 6 AV if U ⊆ V are clopen subsets of G,
(iii) AU · AV ⊆ AUV if U, V are clopen subsets of G,
(iv) 1 ∈ AU whenever U is an open subgroup of G.
In this situation, [7, Proposition 2.5] asserts that for a profinite group G, a
k-algebra A is G-graded if and only if A is a C∞-module algebra.
For every open subgroup U of G, let U C∞ denote the k-subalgebra of functions
f ∈ C∞ that are constant on the left cosets gU of U in G. Then
C∞ = [U 6oG
U C∞
and U C∞ is even a Hopf subalgebra of C∞ whenever U is normal, because it is
naturally isomorphic to the algebra of k-valued functions on the finite group G/U
in this case.
Proposition. Let G be a profinite group, let U be an open normal subgroup of G
and let A be a G-graded k-algebra. Then the algebra of U C∞-invariants of A is
precisely AU .
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
5
Proof. The Hopf algebra U C∞ is spanned by the characteristic functions δgU of all
the cosets of U in G. By the construction given in the proof of [7, Proposition 2.5],
the action of U C∞ on A and the G-graded structure are related by
AgU = δgU · A for all
g ∈ G.
Now the δgU form a family of commuting idempotents in U C∞, and
ǫ(δgU ) = δgU (1) =(cid:26) 1
0
if
if
g ∈ U
g /∈ U
so that (δgU − ǫ(δgU ))δU = 0 for all g ∈ G. This implies that
On the other hand, let µ ∈ A be U C∞-invariant. Then
AU ⊆ A
U C∞
.
µ = (δU · µ) + (1 − δU ) · µ = (δU · µ) + ǫ(1 − δU )µ = δU · µ ∈ AU
as required.
(cid:3)
Corollary. The algebra of C∞-invariants in A isTU 6oG AU .
2.4. Completed group rings. Let kG denote the completed group ring of G with
coefficients in k:
kG := lim
←−
k[G/U ]
where the inverse limit is taken over all the open normal subgroups U of G.
Lemma. Let G be a profinite group and let S ⊆ k be a multiplicatively closed subset
consisting of non zero-divisors.
(a) kG is a G-graded k-algebra.
(b) The algebra of C∞-invariants in kG is k.
(c) The central localisation S−1kG of kG is a G-graded S−1k-algebra.
(d) The algebra of C∞-invariants in S−1kG is S−1k.
Proof. (a) This follows from [7, Lemma 2.9].
(b) For any open normal subgroup U of G let ǫU : kG ։ k[G/U ] be the natural
surjection. Let x ∈ kG\k; then by the definition of inverse limit, we can find some
open subgroup U of G such that the ǫU (x) /∈ ǫU (k). But ǫU (k) = ǫU (kU ) so x /∈ kU .
HenceTU 6oG kU = k and we may apply Corollary 2.3.
(c) By (a) we can find we family ((kG)U ) (for U ranging over the clopen subsets
of G) of k-submodules of kG satisfying the conditions of the definition of a G-graded
k-algebra. Then the family (S−1(kG)U ) gives S−1kG the structure of a G-graded
k-algebra.
(d) In view of part (b), it is enough to prove that S−1kU ∩ kG = kU for every
open normal subgroup U of G. Suppose that s−1x ∈ kG for some s ∈ S and
x ∈ kU . Let y = δU · (s−1x) ∈ kU and z = (1 − δU ) · s−1x ∈ (1 − δU ) · kG so that
s−1x = y + z. Then x − sy ∈ kU and sz ∈ (1 − δU ) · kG, so
x − sy = sz ∈ δU · kG ∩ (1 − δU ) · kG = 0
and therefore s−1x = y ∈ kU as required.
(cid:3)
6
KONSTANTIN ARDAKOV AND SIMON WADSLEY
3. The multiplication map KG ⊗K D(G, K)1 → D(G, K) is injective
3.1. Compact p-adic analytic groups. Now let G be a compact p-adic analytic
group and let R be a complete discrete valuation ring of characteristic zero with a
residue field k of characteristic p. Fix a uniformiser π ∈ R and let K be the field
of fractions of R. We define the algebra of continuous K-valued distributions of G
to be the central localisation
of the completed group ring RG. In this situation we may naturally form three
smash product algebras following §2.1:
KG := K ⊗R RG.
• Ak = kG#C∞(G, k),
• AR = RG#C∞(G, R), and
• AK = KG#C∞(G, K).
Then AT naturally acts on T G for all T ∈ {K, R, k} by Lemma 2.4.
Theorem. (a) kG is a simple Ak-module.
(b) KG is a simple AK-module.
Proof. (a) Since Ak is generated by kG and C∞(G, k), an Ak-submodule of kG is
just a left ideal I of kG such that
C∞(G, k) · I ⊆ I.
By [7, Definition 2.6], we see that every open subgroup U of G controls I. Hence
I χ, the controller subgroup of I, is trivial. Now [7, Theorem A] is also valid for left
ideals, and in our situation this implies that the left ideal I is generated as a left
ideal by its intersection with the ground field k. Therefore I = 0 or I = kG.
(b) Let I be a AK-submodule of KG. Then I ∩ RG is a AR-submodule of RG
and ((I ∩ RG) + πRG)/πRG is a Ak-submodule of kG. By part (a), we see that
either (I ∩ RG) + πRG = RG or (I ∩ RG) + πRG = πRG. In the first case, the
π-adic completeness of RG implies that I ∩ RG = RG and in the second case,
I ∩ RG ⊆ πRG. Thus in the first case I = KG. In the second case, since I = πI,
an easy induction shows that I ∩ RG ⊆ πnRG for all n > 0 and so I ∩ RG = 0,
therefore I = 0.
(cid:3)
3.2. Theorem. Let G be a compact p-adic analytic group and let KG → D be
a homomorphism of C∞(G, K)-module algebras. Let D1 denote the algebra of
C∞(G, K)-invariants in D. Then the multiplication map
is injective.
KG ⊗K D1 −→ D
Proof. Let α1, · · · , αm ∈ KG be linearly independent over K and let t1, . . . , tm ∈
D1 be given such that
α1t1 + · · · + αmtm = 0
inside D. The AK-module KG is simple by Theorem 3.1(b) and its endomorphism
ring EndAK (KG) seen to be K by Lemma 2.1(b) and Lemma 2.4(d). It follows
that the αi are linearly independent over EndAK (KG), and so using the Jacobson
Density Theorem we can find ξ1, . . . , ξm ∈ AK such that
ξi(αj ) = δij
for all
j = 1, . . . , m.
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
7
Now D is a left KG-module by left multiplication, and this action commutes with
right multiplication by D1. Consequently, D is an AK -- D1-bimodule. Therefore
0 = ξi ·
mXj=1
αjtj =
(ξi · αj)tj =
mXj=1
δijtj = ti
mXj=1
(cid:3)
for all i = 1, . . . , m.
3.3. Locally analytic distribution algebras. Now suppose that L is a finite
extension of Qp that is contained in K and let M be a locally L-analytic manifold.
The space of K-valued L-analytic distributions D(M, K) on M is the strong dual
Can(M, K)′
b of the space Can(M, K) of K-valued locally L-analytic functions on
M -- see [31, §2].
When G is a locally L-analytic group, multiplication in the group G induces a
structure of a unital associative K-algebra on D(G, K) [31, Proposition 2.3]. This
algebra is called the algebra of K-valued locally L-analytic distributions on G.
Lemma. D(G, K) is a G-graded K-algebra, whenever G is compact.
Proof. Since G is a locally L-analytic group, every clopen subset U of G is a locally
L-analytic manifold, so we may set
D(G, K)U := D(U, K).
With this definition, parts (ii) and (iv) of Definition 2.3 are clear. We may assume
that all clopen subsets featuring in the statement of parts (i) and (iii) of the Defi-
nition are finite unions of cosets of a fixed open normal subgroup H of G. For each
g ∈ G let δg ∈ D(G, K) be the Dirac distribution. It was observed in the proof of
[31, Lemma 3.1] that
D(G, K) = Mg∈G/H
δg ∗ D(H, K).
Part (i) follows immediately, and part (iii) follows since D(H, K) is a subalgebra of
G which is stable under conjugation by each δg inside D(G, K).
(cid:3)
Corollary. Let G be a compact L-analytic group. Then
(a) D(G, K) is a C∞-module algebra.
(b) The algebra of invariants under this action is precisely
D(G, K)1 := \H6oG
D(H, K).
(c) The natural map KG ⊗K D(G, K)1 → D(G, K) is injective.
(d) Let gK := K ⊗L L(G). Then the natural map
KG ⊗K U (gK) → D(G, K)
is also injective.
Proof. (a) Apply Lemma 3.3 together with [7, Proposition 2.5].
(b) Apply part (a) together with Corollary 2.3.
(c) This follows from Theorem 3.2.
(d) It was observed in [31, §2] that U (gK) is contained in D(H, K) for every
(cid:3)
open subgroup H of G; therefore U (gK) ⊆ D(G, K)1. Now apply part (c).
8
KONSTANTIN ARDAKOV AND SIMON WADSLEY
We remark that it follows from the work of Kohlhaase [20, Proposition 1.2.8]
that the image of U (gK) is in fact dense in D(G, K)1. It is the hyper-enveloping
algebra or Arens-Michael envelope of U (gK) in the sense of Schmidt [29].
4. Affinoid enveloping algebras and Verma modules
4.1. The adjoint action of G(R) on U (g). Let G be a connected, split semisim-
ple, affine algebraic group scheme over R with Lie algebra g. The Lie algebra g is
a G-module via the adjoint action; see [19, II.1.12(1), I.7.18]. In particular, the
group of R-points G(R) of G acts on g by Lie algebra automorphisms, and there-
fore by functoriality on U (g) by R-algebra automorphisms. This action preserves
the natural PBW-filtration
0 ⊂ F0U (g) ⊂ F1U (g) ⊂ · · ·
on U (g). Let Φ be the root system of G relative to a fixed split maximal torus T,
and let xα : Ga → G and eα = (dxα)(1) ∈ g be the root homomorphism and root
vector corresponding to the root α ∈ Φ, respectively.
Lemma. Let r ∈ R, α ∈ Φ.
(a) For every G-module M , each divided power em
(b) For all b ∈ U (g) there exists i > 1 such that ad(reα)i
α
m! preserves M .
· b = 0.
i!
ad(reα)m
m!
(a) for all a ∈ U (g).
m=0
(c) xα(r) · a =P∞
Proof. (a) We may view M as a Ga-module by restriction via xα. Hence it is a
module over the distribution algebra Dist(Ga) of the additive group Ga, by [19,
I.7.11].
It is known [19, I.7.3, I.7.8] that this distribution algebra has a basis
consisting of the divided powers of the generator of Lie(Ga).
(b) U (g) is a G-module so ad(reα)i
(b) lies in U (g) by part (a). Now [g, Fj U (g)] ⊆
Fj−1U (g) for all j > 0, so if b ∈ Fi−1U (g) for some i > 1 then ad(reα)i(b) = 0.
The result follows because U (g) has no R-torsion.
i!
(c) This follows from the definitions -- see [19, I.2.8(1), I.7.12]. Note that the
(cid:3)
right hand side of the equation makes sense by part (b).
4.2. Deformations and π-adic completions. Recall [3, §3.5] that a deformable
R-algebra is a positively Z-filtered R-algebra A such that F0A is an R-subalgebra
of A and gr A is a flat R-module. A morphism of deformable R-algebras is an
R-linear filtered ring homomorphism. Let A be a deformable R-algebra. Its n-th
deformation is the R-subalgebra
An :=Xi>0
πinFiA ⊆ A.
An becomes a deformable R-algebra when we equip An with the subspace filtration
arising from the given filtration on A, and multiplication by πin on graded pieces
of degree i extends to a natural isomorphism of graded R-algebras
gr A
∼=−→ gr An
by [3, Lemma 3.5]. The assignment A 7→ An is functorial in A. bA := lim
A/πaA will
denote the π-adic completion of A. Recall almost commutative affinoid K-algebras
from [3, §3.8]. Such an algebra B has a double associated graded ring Gr(B); when
←−
B = [An,K := cAn ⊗R K
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
9
for some deformable R-algebra A, [3, Corollary 3.7] tells us that Gr(B) can be
computed as follows:
Gr(B) = Gr( [An,K ) ∼= gr A/π gr A.
We fix the deformation parameter n in what follows.
4.3. Semisimple affinoid enveloping algebras. The enveloping algebra U (g)
has associated graded ring gr U (g) = S(g) and is therefore deformable. For each
n > 0, the semisimple affinoid enveloping algebra
\U (g)n,K
is almost commutative affinoid, and its double associated graded ring is
Gr( \U (g)n,K) ∼= S(gk).
By functoriality, the adjoint action of G(R) on U (g) from §4.1 extends to each
\U (g)n,K. This action preserves the double filtration on \U (g)n,K and induces an
action of G(R) on Gr( \U (g)n,K) ∼= S(gk), which factors through G(k).
Proposition. Let r ∈ R, α ∈ Φ and a ∈ \U (g)n,K. Then the series
ad(reα)m
m!
(a)
Xm>0
converges in \U (g)n,K to xα(r) · a.
Proof. Without loss of generality, we may assume that a ∈ \U (g)n. Let D :=
ad(reα), viewed as a derivation of \U (g)n,K and let N > 0 be given. Then there
exists b ∈ U (g)n such that a ≡ b mod πN \U (g)n. Now Di
i! (b) = 0 for some i > 1
by Lemma 4.1(b). Since U (g)n is also a G-module, Di
i! preserves πN U (g)n and
therefore also πN \U (g)n by Lemma 4.1(a). Thus for all N > 0 there exists i > 1
such that
Di
i!
(a) ∈ πN \U (g)n.
m=0
i! (a) → 0 as i → ∞ inside \U (g)n, and therefore the series P∞
Hence Di
Dm
m! (a)
converges to an element eD(a) of \U (g)n, say. This defines an R-linear endomor-
phism a 7→ eD(a) of \U (g)n, which agrees with the action of xα(r) on its dense
subalgebra U (g)n by Lemma 4.1(c).
(cid:3)
Corollary. (a) Every two-sided ideal I of \U (g)n,K is preserved by G(R).
(b) Every central element z of \U (g)n,K is fixed by G(R).
Proof. Since the ring R is local, the Chevalley group G(R) is generated by elements
of the form xα(r) (where r ∈ R and α ∈ Φ) by a result of Abe [1, Proposition 1.6].
Fix r ∈ R and α ∈ Φ.
(a) ∈ I for all N > 0.
This sequence of elements converges to xα(r) · a by the proposition, so xα(r) · a ∈ I
because I is closed by [21, Corollary I.5.5].
(a) Let a ∈ I. Since I is a two-sided ideal,PN
(b) By the proposition, we have xα(r) · z = P∞
(z) = z because
(cid:3)
[reα, z] = 0 by assumption.
m=0
m!
ad(reα)m
m=0
m!
ad(reα)m
10
KONSTANTIN ARDAKOV AND SIMON WADSLEY
4.4. The centre of \U (g)n,K . We assume from now on that G is simply-connected.
The ring of invariants U (g)G of U (g) under the adjoint action of G is a deformable
R-algebra, and it was shown in [3, Proposition 9.3(a)] that the completion of its
n-th deformation \U (g)G
can now prove Theorem C from the Introduction.
n,K is contained in the centre Z(cid:16) \U (g)n,K(cid:17) of \U (g)n,K. We
Proof. By base-changing to the completion of the maximal unramified extension
of K, we may assume that the residue field k of R is algebraically closed. Let
Theorem. We have Z(cid:16) \U (g)n,K(cid:17) = \U (g)G
z ∈ Z(cid:16)\U (g)n(cid:17). Then z is fixed by the action of G(R) by Corollary 4.3(b), so the
\U (g)n,K is fixed by the induced action of
symbol gr ¯z of the image ¯z of z in gr0
G(R) on Gr( \U (g)n,K) ∼= S(gk), which is just the adjoint action of G(k) on S(gk).
Since the group Gk is reduced and k is algebraically closed, it follows from [19,
Remark I.2.8] that gr ¯z ∈ S(gk)Gk . Therefore
n,K.
Gr( \U (g)G
n,K) ⊆ Gr(Z( \U (g)n,K )) ⊆ S(gk)Gk .
It was shown in the proof of [3, Theorem 6.10] that the identification of Gr( \U (g)n,K)
with S(gk) maps Gr( \U (g)G
n,K ) onto S(gk)Gk . Therefore the two inclusions displayed
above are equalities and the result follows.
(cid:3)
4.5. Affinoid Verma modules. Let T ⊂ B be a split maximal torus in G con-
tained in a Borel subgroup B. We will view the unipotent radical N of B as being
generated by the negative roots of the adjoint action of T on G, and let N+ be the
unipotent radical of the opposite Borel B+ containing T. Let t, b, n, n+ and b+ be
the corresponding Lie algebras, so that we have the root space decomposition
g = n ⊕ t ⊕ n+.
Let λ : πnt → R be an R-linear character. View λ as a character of πnb+
by pulling back along the surjection πnb+ ։ πnt with kernel πnn+, and let Kλ
be the corresponding one-dimensional module over the affinoid enveloping algebra
\U (b+)n,K.
Definition. The affinoid Verma module with highest weight λ is
We will compute the annihilator of this affinoid Verma module in \U (g)n,K. To
cV λ := \U (g)n,K ⊗ \U(b+)n,K
Kλ.
from [3, Definition 3.3] that this is an algebraic subset of the prime spectrum of the
polynomial algebra Gr( \U (g)n,K) = S(gk).
do this, we will first need to understand its characteristic variety Ch(cV λ). Recall
Lemma. Ch(cV λ) = (b+
on cV λ such that
k → k be defined by ¯λ(¯x) = λ(πnx) for any x ∈ b+, and let k¯λ
k )-module. Then there is a natural double filtration
Proof. Let ¯λ : b+
denote the corresponding S(b+
The support of this graded module inside Spec S(gk) is (b+
Gr(cV λ) = S(gk) ⊗S(b+
k )⊥ by definition.
(cid:3)
k ) k¯λ.
k )⊥.
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
11
4.6. The annihilator of an affinoid Verma module. It is well known that the
centre Z(gK) of U (gK) acts on the classical Verma module V λ := U (gK)⊗U(b+
K ) Kλ
by a character χλ : Z(gK) → K; see [12, Proposition 7.4.4]. Since V λ is dense in
cV λ, the action Z(gK) on cV λ also factors through χλ.
Gk-invariant polynomials in S(gk) = O(g∗
In [3, §9.8] we defined the nilpotent cone in g∗ to be the set of zeros N ∗ of
k) with no constant term:
After the next preliminary result, we will be able to compute the annihilator of
N ∗ = V (S(gk)Gk
+ ).
cV λ and thereby prove a precise affinoid analogue of [12, Theorem 8.4.3].
Lemma. Suppose that k is algebraically closed, and p is very good for G. Then
(a) the ideal Gr(ker χλ · \U (g)n,K) equals S(gk)Gk
(b) This ideal is prime.
(c) If G := G(k) then G · (b+
k )⊥ is Zariski dense in N ∗.
+ · S(gk).
Proof. Part (a) follows from [3, Theorem 6.10], and part (b) follows from the proof
of [11, Proposition 3.4.1]. In the proof of [11, Proposition 3.4.1] it was also shown
that under our hypotheses, the natural action map G ×B (b+
k )⊥ → N ∗ induces
an isomorphism O(N ∗) → O(G ×B (b+
k )⊥), and is therefore dominant. Part (c)
follows.
(cid:3)
generated by ker χλ.
Theorem. If p is very good for G, then the annihilator of cV λ inside \U (g)n,K is
Proof. Let Iλ be the annihilator of cV λ inside Un := \U (g)n,K and let Jλ = ker χλ·Un.
Then Jλ ⊆ Iλ by the remarks made at the start of §4.6. By base-changing to
the completion of the maximal unramified extension of K, we may assume that
the residue field k of R is algebraically closed. This allows us to identify the
characteristic varieties of finitely generated Un-modules with their corresponding
sets of k-points.
Since Jλ ⊆ Iλ, the characteristic variety Ch(Un/Iλ) is contained in Ch (Un/Jλ),
which equals N ∗ by part (a) of the lemma.
The two-sided ideal Iλ is G(R)-stable by Corollary 4.3(a), so Ch(Un/Iλ) is stable
under the adjoint action of G := G(k) on g∗
contains Ch(cV λ) which is equal to (b+
k )⊥ by Lemma 4.5. Therefore Ch(Un/Iλ)
k )⊥ which is Zariski dense in N ∗ by part (c) of the lemma. Since
contains G · (b+
Ch(Un/Iλ) is closed, we deduce that Ch(Un/Iλ) = N ∗ = Ch(Un/Jλ). Hence
k. Also Iλ annihilates cV λ, so Ch(Un/Iλ)
Gr(Jλ) ⊆ Gr(Iλ) ⊆pGr(Iλ) =pGr(Jλ) = Gr(Jλ)
because Gr(Jλ) is prime by part (b) of the lemma. The result follows.
(cid:3)
4.7. The action of \U (t)n,K on affinoid Verma modules. In §5 we will need the
following elementary result about the analytic density of certain infinite discrete
subsets in affinoid polydiscs.
Lemma. Let A1, A2, . . . , Aℓ be infinite subsets of R and suppose that an element
f in the Tate algebra Khx1, . . . , xℓi vanishes on A1 × A2 × · · · × Aℓ. Then f = 0.
12
KONSTANTIN ARDAKOV AND SIMON WADSLEY
Proof. Proceed by induction on ℓ. The case when ℓ = 0 is vacuous so we may
assume that ℓ > 1. For every y ∈ Aℓ let gy(x1, . . . , xℓ−1) := f (x1, . . . , xℓ−1, y) ∈
Khx1, . . . , xℓ−1i. Then gy vanishes on A1 × A2 × · · · × Aℓ−1 so by induction, gy = 0
for all y ∈ Aℓ. Therefore xℓ − y divides f for all y ∈ Aℓ.
Now Khx1, . . . , xℓi is a Noetherian unique factorisation domain by [15, Theorem
3.2.1], and as y ranges over Aℓ, the xℓ − y form a collection of infinitely many
distinct irreducible elements of Khx1, . . . , xℓi. Therefore f = 0.
(cid:3)
Now consider the action of \U (t)n,K on the affinoid Verma module cV λ from §4.5.
Let vλ ∈ cV λ be a highest weight vector, let α1, · · · , αm ∈ t∗
K be the positive roots
corresponding to the adjoint action of t on n+ and choose a generator fi ∈ n for the
−αi-root R-submodule of n. Write f β := f β1
m ∈ U (n) for any β ∈ Nm.
It is easy to see that
2 · · · f βm
1 f β2
h · f βvλ = (λ −
mXj=1
βjαj )(h)f βvλ
for all h ∈ t.
Thus f βvλ spans a one-dimensional U (t)K-submodule of U (nK) · vλ ⊂ cV λ, so Kf β
is actually a \U (t)n,K-module where tK acts via the character λ −Pm
In particular, we see that U (nK) · vλ is a locally finite \U (t)n,K-module.
j=1 βjαj ∈ t∗
K.
Proposition. The action of \U (t)n,K on U (nK) · vλ is faithful.
Proof. Let α1, . . . , αℓ be the simple roots, let h1, . . . , hℓ ∈ t be the corresponding
coroots and let ω1, . . . , ωℓ ∈ t∗
K be the corresponding system of fundamental weights,
so that ωi(hj) = δij for all i, j = 1, . . . , ℓ. By [17, §13.1], we may use the Cartan
matrix C = (hαi, αji) associated to the root system of gK to express the simple
roots in terms of the fundamental weights:
αj =
Cjkωk
for all
j = 1, . . . , ℓ.
ℓXk=1
Let C∗ denote the adjugate matrix of C and let d := det C; it then follows that
dωi =
ℓXj=1
C∗
ijαj
for all
i = 1, . . . , ℓ.
All entries of C∗ are known to be non-negative integers; see either [17, §13.2, Table
1] or [22]. Therefore for each µ ∈ Nℓ,
µidωi =
ℓXi=1
ℓXj=1 ℓXi=1
ij! αj
µiC∗
is a linear combination of α1, . . . , αℓ with non-negative integer coefficients. We now
observe that for any µ ∈ Nd, tK acts on the vector
eµ :=
f
ℓYj=1
i=1 µiC ∗
ij
Pℓ
j
vλ ∈ U (nK) · vλ
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
13
via the character
λ −
ℓXj=1 ℓXi=1
ij! αj = λ −
µiC∗
ℓXi=1
µidωi.
Because our group G is simply connected, the coroots hi span t over R and therefore
we may identify the affinoid enveloping algebra \U (t)n,K with the Tate algebra
∼=−→ \U (t)n,K which
Khπnh1, . . . , πnhℓi. Consider the isomorphism Khx1, . . . , xℓi
sends xi to λ(πnhi) − πnhi. Viewing U (nK) · vλ as a Khx1, . . . , xℓi-module via this
isomorphism and remembering that ωi(hj) = δij , we can calculate that
xj · eµ = dπnµjeµ
for all
j = 1, . . . , ℓ
and µ ∈ Nd.
Therefore for every f ∈ Khx1, . . . , xni and µ ∈ Nd we have
f · eµ = f (dπnµ1, . . . , dπnµℓ)eµ.
We may now apply the lemma with each Aj being the infinite subset dπnN of R to
deduce that if f ∈ Khx1, . . . , xℓi kills every eµ, then f = 0. The result follows. (cid:3)
4.8. The Cartan involution. Recall [19, §II.1.4] that for each element w of the
Weyl group W of G we can find a representative w ∈ G(R) of w ∈ W which
normalises T(R). By [19, §II.1.4(4)], these elements permute the root subgroups of
G according to the action of W on the root system Φ. If w0 ∈ W is the longest
element, then it follows that
w0 · b+ = b
and
w0 · b = b+
in the adjoint action of G(R) on g.
Proposition.
w0 · Ann \U(b+)n,K
(cV λ) = Ann \U(b)n,K
(cV λ).
Proof. By Theorem 4.6, I := Ann \U(g)n,K
by the adjoint action of G(R) on \U (g)n,K. Therefore w0 · I = I, so
(cV λ) is generated by ker χλ which is fixed
(cV λ)
(cV λ) = w0 · I ∩ w0 · \U (b+)n,K = I ∩ \U (b)n,K = Ann \U(b)n,K
w0 · Ann \U(b+)n,K
as required.
(cid:3)
5. Faithfulness of affinoid Verma modules over Iwasawa algebras
5.1. L-uniform groups. Throughout §5 we will assume that p is an odd prime,
and that L is a finite extension of Qp contained in K; we have the corresponding
chain of inclusions of discrete valuation rings:
Zp ⊆ OL ⊆ R.
Following Orlik and Strauch [26, Remark 2.2.5(ii)], we say that a uniform pro-p
group G is L-uniform if G is locally L-analytic, and the Lie algebra LG is an OL-
submodule of the L-Lie algebra L(G). The (modified) isomorphism of categories
G 7→ 1
p LG between uniform pro-p groups and finite rank torsion-free Zp-Lie algebras
from [13, Theorem 9.10] induces a one-to-one correspondence between L-uniform
groups and torsion-free OL-Lie algebras of finite rank.
14
KONSTANTIN ARDAKOV AND SIMON WADSLEY
Let G be an L-uniform group. In §5.1, §5.2 and §5.3 we will suspend the notation
from §4 and temporarily use the letter g to denote the R-Lie algebra associated with
G, defined as follows:
g := R ⊗OL
1
p
LG.
This extends [3, Definition 10.2] to arbitrary finite extensions L of Qp. Recall that
the algebra D(G, K) of K-valued locally L-analytic distributions from §3.3 is a
Frechet-Stein algebra by [32, Theorem 5.1], and therefore we have at our disposal
the K-Banach space completions Dr(G, K) for each real number 1/p 6 r < 1. The
abbreviation
\U (g)K := \U (g)0,K
will also be used throughout §5.
5.2. The distribution algebra D1/p(G, K). We begin by recording the following
extension of [5, Theorem 5.1.4] and [30, Proposition 6.10] to arbitrary complete
discrete valuation fields K of mixed characteristic and arbitrary L-uniform groups
in the language of locally analytic distribution algebras.
Lemma. Let g be the R-Lie algebra associated with the L-uniform group G. Then
there is a natural isomorphism of K-Banach algebras
∼=−→ \U (g)K.
D1/p(G, K)
Proof. Suppose first that L = Qp. Let · : K → R be the norm which induces the
topology on K, normalised by p = 1/p. Let g1, . . . , gd be a minimal topological
generating set for G, let bi = gi − 1 ∈ K[G] and write α = α1 + · · · + αd for each
α ∈ Nd. It follows from [32, §4] that the distribution algebra D1/p(G, K) consists
of all formal power series λ =Pα∈Nd λαbα in b1, . . . , bd such that
dα(1/p)α
λ1/p := sup
α∈Nd
is finite. As a consequence of Cohen's Structure Theorem for complete local domains
[23, Corollary 28.P.2], we can find an unramified field extension K ′ of Qp inside K
1
such that K/K ′ is finite. Let R′ be the ring of integers of K ′ and let g′ = R′⊗Zp
p LG.
Then we have a commutative diagram
K ⊗K ′ D1/p(G, K ′)
K ⊗K ′ \U (g′)K ′
D1/p(G, K)
/ \U (g)K
where the vertical maps are induced by multiplication inside D(G, K) and \U (g)K
and the horizontal maps are induced by the inclusion of G into the groups of units
of \U (g′)K ′ and \U (g)K, respectively.
Because K ′ is unramified over Qp, it follows from [3, Theorem 10.4] that the top
arrow is an isomorphism. The arrow on the right is a bijection by [3, Lemma 3.9(c)],
and arrow on the left can be seen to be a bijection from the explicit description
of elements in D1/p(G, K) as power series in the b1, . . . , bd satisfying the particular
convergence condition stated above. The result follows in the case where L = Qp.
/
/
/
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
15
Returning to the general case, let G0 be the uniform pro-p group G viewed as
a locally Qp-analytic group, and let g0 := 1
p R ⊗Zp LG. Then there are natural
surjections of algebras U (g0) ։ U (g) and D1/p(G, K) ։ D1/p(G0, K), and it
follows from [28, Lemma 5.1] that
D1/p(G, K) ∼= U (g) ⊗U(g0) D1/p(G0, K).
Therefore by the first part of this proof we obtain
D1/p(G, K) ∼= U (g) ⊗U(g0)
\U (g0)K.
Now the algebra on the right hand side is isomorphic to \U (g)K by [10, §3.2.3(iii)]
because U (g0) is Noetherian.
(cid:3)
5.3. A general faithfulness result. Let G be an L-uniform group with associated
R-Lie algebra g. The following result will be our main engine for establishing the
faithfulness of modules over the Iwasawa algebra KG.
Theorem. Let N and H be L-uniform subgroups of G with associated R-Lie alge-
bras n and h such that g = n ⊕ h. Let V be a \U (g)K -module and suppose that there
is v ∈ V such that v is a free generator of V as a \U (n)K-module by restriction and
U (nK)v is a faithful, locally finite RH-module again by restriction. Then V is a
faithful KG-module.
Proof. Let r and d be the ranks of n and g as R-modules, respectively. Choose an
R-basis {x1, . . . , xd} for g such that {x1, . . . , xr} is an R-basis for n.
Let l = [L : Qp]; then G, N and H have dimensions dl, rl and (d−r)l respectively
when viewed as uniform pro-p groups. We may choose a minimal topological gen-
erating set {g1, . . . , gdl} for G such that g1, . . . , grl and grl+1, . . . , gdl topologically
generate N and H, respectively. Write bi = gi − 1 ∈ KG for each i = 1, . . . , ld.
Suppose that ζ ∈ AnnRG(V ). It suffices to prove that ζ = 0. We may write
Now given w ∈ U (nK)v, RHw ⊂ U (nK)v is finitely generated over R by assump-
ζ =Pα∈Ndl λαbα with λα ∈ R. Collecting terms together we can then rewrite this
as ζ =Pα∈Nrl bαζα for some ζα ∈ RH.
ζα · w = Xβ∈Nr
tion. Thus there is a natural number t such that we can write
µα
β xβ · v
for some µα
that µα
β is uniformly bounded in α and β. Thus
β ∈ K with µα
β = 0 for β > t and all α. Furthermore, we may assume
But Pα∈Nrl,β∈Nr
0 = ζ · w = Xα∈Nrl
bαζα · w = Xα∈Nrl,β∈Nr
β bαxβ · v.
µα
β bαxβ ∈ \U (n)K and ann \U(n)K
µα
(v) = 0 by assumption, so in fact
Xα∈Nrl,β∈Nr
µα
β bαxβ = 0.
The multiplication map KN ⊗K U (nK) → D(N, K) is injective by Corollary 3.3(d).
Since D(N, K) contained in D1/p(N, K) which is isomorphic to \U (n)K by Lemma
16
KONSTANTIN ARDAKOV AND SIMON WADSLEY
5.2, the multiplication map KN ⊗K U (nK) → \U (n)K is also injective, and so
Xβ∈Nr Xα∈Nrl
µα
β bα ⊗ xβ = 0.
β bα = 0 ∈ KN for each β because the xβ are linearly indepen-
dent over K. It follows that µα
β = 0 for each pair α, β, and hence ζα · w = 0 for each
α. As this last equality is independent of the choice of w ∈ U (nK)v, we deduce
(cid:3)
ThereforePα∈Nrl µα
that ζα ∈ AnnRH (U (n)v) = 0 for each α and hence ζ =Pα∈Nrl bαζα = 0.
5.4. Congruence kernels. We now assume that the L-uniform group G and the
R-algebraic group G from §4.1 satisfy the following conditions:
• G is simply connected,
• the Lie algebra g of G and the Lie algebra LG of G satisfy png = 1
p R⊗OL LG
for some integer n > 0,
• p is a very good prime for G in the sense of [3, §6.8].
For example, for every integer n > 0, G could be the congruence kernel
G = ker(G(OL) → G(OL/pn+1OL)).
As in §4.5, we let t, b, n and b+ denote the Lie algebras of T, B, N and B+ of G
respectively and note that because these groups are defined over OL we can find
L-uniform subgroups T, B, N and B+ whose respective associated R-Lie algebras
are pnt, pnb, pnn and pnb+.
Theorem 5.3.
R-linear character λ : pnt → R.
K-module; this implies that it is also a locally
Since RT is a subring of \U (pnt)K, Proposition 4.7 implies that the action of RT
Theorem. The affinoid Verma module cV λ is faithful as a KG-module for every
Proof. Since pnn is a complement to pnb+ in png, cV λ is a free \U (pnn)K -module
of rank 1 generated by the highest weight vector v ∈ cV λ. The dense submodule
U (nK) · v of cV λ is locally finite as a b+
finite RB+-submodule of cV λ. In particular, it is a locally finite RT -module.
on U (nK) · v is faithful. Since pnb = pnn ⊕ pnt, cV λ is faithful as a RB-module by
Proposition 4.8 now implies that cV λ is also faithful as a RB+-module, so its
Theorem 5.3 again gives that cV λ is faithful as a KG-module, as required.
submodule U (nK) · v is also faithful over RB+. Since png = pnn ⊕ pnb+, invoking
(cid:3)
5.5. Verma modules for congruence kernels. For each locally L-analytic char-
acter θ : B+ → 1 + pR, the contragredient of the natural 1-dimensional representa-
tion given by θ induces a continuous 1-dimensional D(B+, K)-module Kθ via [31,
Corollary 3.4]. We may view Kθ as a KB+-module by restriction recovering the
original 1-dimensional representation. That is b · x = θ(b)x for b ∈ B+ and x ∈ Kθ.
K) → D(B+, K) we may view Kθ
as a representation λ of the K-Lie algebra b+
K. We can compute that for b ∈ B+,
log b ∈ pn+1b+ acts by log θ(b) ∈ pR. Thus λ may be viewed as an R-linear
character pnb+ → R.
By instead restricting along the inclusion U (b+
Conversely, each R-linear map λ : pnb+ → R induces a locally L-analytic homo-
morphism θ : B+ → 1 + pR via the rule
b 7→ exp(λ(log b))
for all
b ∈ B+.
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
17
Definition. Let θ : B+ → 1 + pR be a locally L-analytic group homomorphism.
The Verma module for the Iwasawa algebra KG with highest weight θ is
M θ = KG ⊗KB+ Kθ
where Kθ is the one-dimensional KB+-module K with B+-action given by θ.
Lemma. Let λ : pnb+ → R be the R-linear character of pnb+ corresponding to a
locally L-analytic group homomorphism θ : B+ → 1 + pR. Then the KG-submodule
of cV λ generated by the highest weight vector vλ is naturally isomorphic to M θ.
Proof. By construction, B+ acts on vλ ∈ cV λ via θ. Sending the highest weight
vector mθ ∈ M θ to vλ therefore induces a KG-module homomorphism M θ −→ cV λ,
which fits into the following commutative diagram:
KN
\U (pnn)K
M θ
Corollary. M θ is a faithful KG-module.
Here the vertical arrows are bijections that send x ∈ KN to xmθ and y ∈ \U (pnn)K
to yvλ, respectively. The top arrow is the natural inclusion of KN into \U (pnn)K,
(cid:3)
/ cV λ.
so M θ −→ cV λ is injective. The result follows.
of M θ is dense in cV λ. Now if an element of KG kills M θ, then it must annihilate
all of cV λ by continuity, and is therefore zero by Theorem 5.4.
5.6. Finite normal subgroups. Before we can give a proof of Theorem A, we
will need to understand better the finite normal subgroups of open subgroups of
G(OL).
Proof. The commutative diagram in the proof of the lemma shows that the image
(cid:3)
Proposition. Let G be an open subgroup of G(OL) and let F be a finite normal
subgroup of G. Then F is central in G(OL).
Proof. Choose a torsion-free open subgroup N of G which is normal in G(OL), for
example a congruence kernel of G(OL). Then [N, F ] 6 N ∩ F = 1 because N and
F are both normal in G and because N is torsion-free, so F centralises N . Because
OL is local, G(OL) is generated as an abstract group by the elements of the form
xα(r) from §4 for α ∈ Φ and r ∈ OL by [1, Proposition 1.6], so it will be sufficient
to show that F commutes with each xα(r).
Fix g ∈ F , α ∈ Φ and r ∈ OL, let x := xα(r) and note that g commutes with
xt where t is the index of N in G(OL). Choose a faithful algebraic representation
ρ : G(L) → GLm(L) for some m > 1, let u1 = ρ(x) and u2 = ρ(gxg−1). Then u1
and u2 are unipotent by [18, Theorem 15.4(c)] and ut
2 because g commutes
with xt. Now log and exp give well-defined bijections between unipotent matrices
in GLm(L) and nilpotent matrices in Mm(L), so
1 = ut
Therefore x = gxg−1 because ρ is faithful.
(cid:3)
u1 = exp(cid:18) 1
t
log(ut
1)(cid:19) = exp(cid:18) 1
t
log(ut
2)(cid:19) = u2.
/
/
/
18
KONSTANTIN ARDAKOV AND SIMON WADSLEY
5.7. Proof of Theorem A. By continuity, we can find an open subgroup H of
B+ which is mapped into 1 + pR by θ. Choose n large enough so that G contains an
open normal L-uniform subgroup Gn with associated R-Lie algebra png and such
that H contains an open L-uniform subgroup B+
n with associated R-Lie algebra
pnb+. Let θn be the restriction of θ to B+
n . Then since B+
n = B+ ∩ Gn,
M θn = KGn ⊗KB+
n
Kθn ⊆ KG ⊗KB+ Kθ.
Writing I := AnnKG(KG ⊗KB+ Kθ) and applying Corollary 5.5 gives
I ∩ KGn ⊆ AnnKGn M θn = 0.
Let F be a finite normal subgroup of G. Then F is central in G(OL) by Proposition
5.6, so F is also central in G. But G has trivial centre by assumption so F is trivial.
Therefore KG is a prime ring by [9, Theorem A].
Next, KG is a crossed product of KGn with the finite group G/Gn. By [24,
Theorem 2.1.15], S := KGn\{0} is an Ore set in the Noetherian domain KGn. It
is stable under conjugation by G, so by [27, Lemma 37.7] it is also an Ore set in
the larger ring KG and there is a crossed product decomposition
S−1KG = (S−1KGn) ∗ (G/Gn).
Here S−1KGn is the quotient division ring of fractions of KGn; thus S−1KG is an
Artinian ring because the group G/Gn is finite. Every regular element in KG stays
regular in KGn and therefore becomes invertible in S−1KG by [24, Proposition
3.1.1]; hence S−1KG is the classical Artinian ring of quotients of KG.
Since I ∩ KGn = 0, the intersection I ∩ S is empty and therefore the two-sided
ideal S−1I of S−1KG is proper. Now KG is prime, so S−1KG is a prime Artinian
ring and is therefore simple. Therefore S−1I = 0 and I = 0.
(cid:3)
References
[1] Eiichi Abe. Chevalley groups over local rings. Tohoku Math. J. (2), 21:474 -- 494, 1969.
[2] K. Ardakov and S. J. Wadsley. Γ-invariant ideals in Iwasawa algebras. J. Pure Appl. Algebra,
213(9):1852 -- 1864, 2009.
[3] K. Ardakov and S. J. Wadsley. On irreducible representations of compact p-adic analytic
groups. Annals of Mathematics, 178:453 -- 557, 2013.
[4] K. Ardakov, F. Wei, and J. J. Zhang. Reflexive ideals in Iwasawa algebras. Adv. Math.,
218(3):865 -- 901, 2008.
[5] Konstantin Ardakov. Krull dimension of Iwasawa algebras and some related topics. PhD
thesis, University of Cambridge, 2004.
[6] Konstantin Ardakov. Prime ideals in noncommutative Iwasawa algebras. Math. Proc. Cam-
bridge Philos. Soc., 141(2):197 -- 203, 2006.
[7] Konstantin Ardakov. The controller subgroup of one-sided ideals in completed group rings. In
New trends in noncommutative algebra, volume 562 of Contemp. Math., pages 11 -- 26. Amer.
Math. Soc., Providence, RI, 2012.
[8] Konstantin Ardakov. Prime ideals in nilpotent Iwasawa algebras. Invent. Math., 190(2):439 --
503, 2012.
[9] Konstantin Ardakov and Kenneth A. Brown. Primeness, semiprimeness and localisation in
Iwasawa algebras. Trans. Amer. Math. Soc., 359(4):1499 -- 1515 (electronic), 2007.
[10] Pierre Berthelot. D-modules arithm´etiques I. op`erateurs diff'erentiels de niveau fini. Ann.
Sci. cole Norm. Sup. (4), 29(2):185 -- 272, 1996.
[11] R. Bezrukanvikov, I. Mirkovic, and D. Rumynin. Localization of modules for a semisimple
Lie algebra in prime characteristic. Ann. of Math., 167(3):945 -- 991, 2008.
[12] J. Dixmier. Enveloping Algebras, volume 11 of Graduate Studies in Mathematics. American
Mathematical Society, english edition, 1996.
VERMA MODULES FOR IWASAWA ALGEBRAS ARE FAITHFUL
19
[13] J. D. Dixon, M. P. F. du Sautoy, A. Mann, and D. Segal. Analytic pro-p groups, volume 61
of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge,
second edition, 1999.
[14] Michel Duflo. Sur la classification des id´eaux primitifs dans l'alg`ebre enveloppante d'une
alg`ebre de Lie semi-simple. Ann. of Math. (2), 105(1):107 -- 120, 1977.
[15] Jean Fresnel and Marius van der Put. Rigid analytic geometry and its applications, volume
218 of Progress in Mathematics. Birkhauser Boston Inc., Boston, MA, 2004.
[16] Michael Harris. The annihilators of p-adic induced modules. J. Algebra, 67(1):68 -- 71, 1980.
[17] James E. Humphreys. Introduction to Lie algebras and representation theory. Springer-
Verlag, New York, 1972. Graduate Texts in Mathematics, Vol. 9.
[18] James E. Humphreys. Linear algebraic groups. Springer-Verlag, New York, 1975. Graduate
Texts in Mathematics, No. 21.
[19] Jens Carsten Jantzen. Representations of algebraic groups, volume 107 of Mathematical Sur-
veys and Monographs. American Mathematical Society, Providence, RI, second edition, 2003.
[20] Jan Kohlhaase. Invariant distributions on p-adic analytic groups. Duke Math. J., 137(1):19 --
62, 2007.
[21] H. Li and F. Van Oystaeyen. Zariskian filtrations. Kluwer Academic Publishers, 1996.
[22] George Lusztig and Jacques Tits. The inverse of a Cartan matrix. An. Univ. Timi¸soara Ser.
S¸tiint¸. Mat., 30(1):17 -- 23, 1992.
[23] Hideyuki Matsumura. Commutative algebra, volume 56 of Mathematics Lecture Note Series.
Benjamin/Cummings Publishing Co., Inc., Reading, Mass., second edition, 1980.
[24] J. C. McConnell and J. C. Robson. Noncommutative Noetherian rings, volume 30 of Graduate
Studies in Mathematics. American Mathematical Society, Providence, RI, revised edition,
2001. With the cooperation of L. W. Small.
[25] Susan 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, 1993.
[26] Sascha Orlik and Matthias Strauch. On the irreducibility of locally analytic principal series
representations. Represent. Theory, 14:713 -- 746, 2010.
[27] Donald S. Passman. Infinite crossed products, volume 135 of Pure and Applied Mathematics.
Academic Press Inc., Boston, MA, 1989.
[28] Tobias Schmidt. Auslander regularity of p-adic distribution algebras. Represent. Theory,
12:37 -- 57, 2008.
[29] Tobias Schmidt. Stable flatness of nonarchimedean hyperenveloping algebras. J. Algebra,
323(3):757 -- 765, 2010.
[30] Tobias Schmidt. On locally analytic Beilinson -- Bernstein localization and the canonical di-
mension. Mathematische Zeitschrift, page 141, 2013.
[31] Peter Schneider and Jeremy Teitelbaum. Locally analytic distributions and p-adic represen-
tation theory, with applications to GL2. J. Amer. Math. Soc., 15(2):443 -- 468 (electronic),
2002.
[32] Peter Schneider and Jeremy Teitelbaum. Algebras of p-adic distributions and admissible
representations. Invent. Math., 153(1):145 -- 196, 2003.
[33] Otmar Venjakob. A non-commutative Weierstrass preparation theorem and applications to
Iwasawa theory. J. Reine Angew. Math., 559:153 -- 191, 2003. With an appendix by Denis
Vogel.
[34] Simon Wadsley. A Bernstein-type inequality for localizations of Iwasawa algebras of Heisen-
berg pro-p groups. Q. J. Math., 58(2):265 -- 277, 2007.
Mathematical Institute, University of Oxford, Oxford OX2 6GG
Homerton College, Cambridge, CB2 8PQ
|
1606.02085 | 1 | 1606 | 2016-06-07T10:23:14 | The Ziegler spectrum and Ringel's quilt of the A-infinity plane singularity | [
"math.RA",
"math.AC"
] | We describe the Cohen-Macaulay part of the Ziegler spectrum and calculate Ringel's quilt of the category of finitely generated Cohen--Macaulay modules over the A-infinity plane singularity. | math.RA | math |
THE ZIEGLER SPECTRUM AND RINGEL'S QUILT OF THE A-INFINITY
PLANE SINGULARITY
GENA PUNINSKI
Abstract. We describe the Cohen -- Macaulay part of the Ziegler spectrum and calculate Ringel's
quilt of the category of finitely generated Cohen -- Macaulay modules over the A-infinity plane
singularity.
1. Introduction
The original objective of this paper was to describe the Cohen -- Macaulay part of the Ziegler
spectrum over the A∞ plane singularity R. With a decent knowledge of finitely generated points
in this topological space and Model Theory of Modules this proved to be a task of reasonable
complexity. Surprisingly we found that all non-finitely generated points in this part of the Ziegler
spectrum are natural ones: the integral closure eR of R, its quotient ring Q and a generic module
G.
This classification was achieved by the so-called 'interval method':
instead of classifying
points of this space, i.e.
indecomposable pure injective modules, we calculate some intervals
in the lattice of finitely generated subfunctors of the functor Hom(R, −) from the category of
finitely generated Cohen -- Macaulay modules to abelian groups.
If the lattice structure of such interval is known, then one can use Ziegler's result that each in-
decomposable pure injective module opening this interval is uniquely determined by a nontrivial
filter defined by its realization. This approach resembles the method (hence the name) used by
Gelfand -- Ponomarev [8], and later by Ringel [17], to classify indecomposable finite dimensional
modules over certain classes of finite dimensional algebras.
As for now we have to rely on classification of finitely generated Cohen -- Macaulay modules to
describe the infinite part of the Ziegler spectrum. Furthermore it could be a nontrivial task to
recover an indecomposable pure injective module from the pp-type it realizes, but this is quite
straightforward in the example of our interest.
We completely describe the topology of the Cohen -- Macaulay part of the Ziegler spectrum of
R, in particular prove that the Cantor -- Bendixson rank of this space equals 2; and the same is the
value of the m-dimension of the lattice of pp-formulae of the theory of Cohen -- Macaulay modules.
From this point of view the category of finitely generated Cohen -- Macaulay modules over R
resembles the category of finite dimensional modules over tame hereditary finite dimensional
algebras.
2000 Mathematics Subject Classification. 13C14 (primary), 13L05, 16D50.
Key words and phrases. Cohen -- Macaulay module, Ziegler spectrum, Ringel's quilt, infinite radical.
1
2
However some peculiarities were also observed. For instance one finitely generated Cohen --
Macaulay module has Cantor -- Bendixson rank 1, in particular is non-isolated. This module is
at the end of an almost split sequence in the category of all (finitely generated or not) Cohen --
Macaulay modules whose source is an infinitely generated pure injective module. Furthermore,
contrary to the case of finite dimensional algebras, none of indecomposable infinitely generated
pure injective Cohen -- Macaulay modules is a direct summand of a direct product of finitely
generated ones.
Some non-finitely generated points of the Ziegler spectrum are direct limits of finitely gen-
erated Cohen -- Macaulay modules (along a ray of irreducible morphism), so we use these points
to glue the two components of the Auslander -- Reiten quiver of R obtaining a 2-dimensional
surface we call the Ringel quilt of R. The additional devices used in this knitting procedure are
irreducible morphisms and almost split sequences with infinitely generated terms.
This gluing procedure was implemented by Ringel [19] when investigating modules over do-
mestic string algebras and gives a new insight on the global geometric structure of the category
of finite dimensional modules.
We will show that over the A∞ plane singularity the category of finitely generated Cohen --
Macaulay modules can be neatly lodged on the Mobius stripe. For instance morphisms in this
category amount to easily controlled walks on this surface. Using this representation we prove
that the nilpotency index of the radical of this category equals ω + 2.
We will show that the problem of classifying points of the Ziegler spectrum of higher dimen-
sional A∞ singularities Rd, d ≥ 2 is wild. However there is a good chance to calculate the
closure, in the Ziegler spectrum, of the set of finitely generated Cohen -- Macaulay points; but
this is rather a task for future.
Many statements in this paper admit obvious generalizations, bur we decided to suppress
writing them down. A general philosophy is that when one starts developing a new theory a
carefully calculated example creates a better guide than general statements.
The projected audience of this paper is twofold: the experts in commutative algebra who are
interested in learning methods of model theory of modules; and also people from model theory
of modules who are keen to step on the well fertilized turf with a new exploring tool. It is quite
difficult to make this text easily accessible to both groups, and our exposition is clearly biased:
we will be quite meticulous in explaining basics of the theory of Cohen -- Macaulay modules, but
more sketchy on items in model theory of modules. For those the reader is referred to Mike
Prest's book [14] which is not so short, but contains almost all references we need.
The author is indebted to Ivo Herzog for useful comments on preliminary versions of the
paper.
2. Basics
Let F be an algebraically closed field of characteristic not equal to 2. In fact - see [6, Rem.
1.2.22] - the results are likely to be true also when characteristic is equal to 2, but it is difficult
to find proper references. Let R = F [[x, y]]/(x2) be the factor of the power series ring by the
3
ideal generated by x2, the so-called A∞ plane singularity - see [5]. It is known from this paper
(see also [13, Thm. 14.16]) that R has a countable Cohen -- Macaulay representation type, i.e.
only countably many indecomposable finitely generated Cohen -- Macaulay modules.
We need few easily verified facts on the structure of R. Clearly R is a commutative complete
local noetherian ring whose maximal ideal m is generated by x and y. From x2 = 0 and
R/xR ∼= F [[y]] we conclude that xR ⊂ m are the only prime ideals of R, in particular R has
Krull dimension 1.
Recall that a (commutative noetherian local) ring S is said to be Gorenstein if the regular
module SS has a finite injective dimension; for instance R is Gorenstein. Namely note that y
is a non-zero divisor in R and R/yR ∼= F [[x]]/(x2). Because the socle of this ring is simple, [7,
Prop. 21.5] yields that R/yR is 0-dimensional Gorenstein. Since R is local, we conclude by [4,
3.1.19 b)].
Furthermore R is clearly uniform, so is its quotient ring Q. It is easily checked that it suffices
to invert y (or any nonzero divisor) to get Q. Namely s ∈ R is a non-zero divisor iff s /∈ xR,
hence s = f (y) + xg(y), where f 6= 0. Multiplying by a unit in F [[y]] we may assume that
s = yn + xg, hence s−1 = (yn − xg)y−2n.
Since R is Gorenstein, it follows that Q is an indecomposable injective module. Furthermore
Q/R is the injective envelope of the simple module F , hence the minimal injective cogenerator
in the category of R-modules. In particular 0 → R → Q → Q/R → 0 is the minimal injective
resolution of R, therefore R has injective dimension 1.
Recall that eR denotes the integral closure of R in Q. The following is also straightforward.
Remark 2.1. eR is a non-noetherian valuation ring of Krull dimension 1 with the following
chain of principal ideals:
eR ⊃ y eR ⊃ y2 eR ⊃ . . . ⊃ xy−1 eR ⊃ x eR ⊃ xy eR ⊃ . . . .
In particular the ideal ∩nyn eR = ∪m∈Z xymZ is nilpotent and non-principal. Also eR is not
finitely generated as an R-module.
Proof. It is easily seen that eR = R + xQ, where the inclusion ⊇ is obvious, since every element
of xQ is nilpotent.
As above (up to a multiplicative unit) every element from eR\xQ is of the form r = yn +xg(y).
Furthermore r = yn(1 + xy−ng), where the last factor is a unit in eR. Thus the principal ideal
generated by r equals yn eR.
Similarly every cyclic eR-submodule of xQ is generated by xym for some integer m.
Recall, see [23, p. 1], a definition of maximal Cohen -- Macaulay modules. Let S be a commu-
(cid:3)
tative noetherian local ring of Krull dimension d whose residue field is F . An S-module N is
said to be maximal Cohen -- Macaulay, CM for short, if Exti(F, N ) = 0 for all 0 ≤ i < d. We take
this as a definition even when N is not finitely generated.
In our case this boils down to the following.
4
Remark 2.2. An R-module N is Cohen-Macaulay if and only if N has no y-torsion: ny = 0
for some n ∈ N implies n = 0.
Proof. Because R is 1-dimensional, N is CM iff Hom(F, N ) = 0, i.e. nx = ny = 0 for some
n ∈ N implies n = 0. Since x2 = 0, this is clearly equivalent to the absence of y-torsion.
(cid:3)
3. Finitely generated CM-modules
In this section we will recall the classification of indecomposable finitely generated CM-
modules over R. By the above description we are in the framework of Bass' ubiquity paper
[3], see also comments in [13, Thm. 4.18]. For instance every indecomposable finitely generated
CM-module over R is isomorphic to an ideal of R (equivalently to a finitely generated module
between R and eR). One can be even more precise.
Let In = (x, yn), n ≥ 0 be the ideal of R generated by x and yn, in particular I0 = R and
I1 = m. Further we set I∞ = xR. The following is well known - see [23, Exam. 6.5] or [21].
Fact 3.1. Each indecomposable finitely generated CM-module over R is isomorphic to In for
some n ≥ 0, or to I∞.
For future use we represent these modules by the following diagrams.
x
1
•
⑧⑧⑧⑧⑧
•
❄❄❄❄❄
y
y
x
❄❄❄❄❄
•
⑧⑧⑧⑧⑧
•
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
R
I2
x
•
❄❄❄❄❄
y
•
❄❄❄❄❄
y
x
y2
•
⑧⑧⑧⑧⑧
•
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
I1
I∞
x
•
❄❄❄❄❄
y
•
x
y
•
⑧⑧⑧⑧⑧
❄❄❄❄❄
•
y
x
y
❄❄❄❄❄
•
⑧⑧⑧⑧⑧
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
x
•
❄❄❄❄❄
y
•
❄❄❄❄❄
y
• ...
Furthermore the following is the Auslander -- Reiten quiver of the category of finitely generated
CM-modules.
R
/ I1
I2
. . .
I∞f
Here the (left to right) irreducible morphisms In → In+1 are given by multiplication by y;
and the (right to left) irreducible morphisms In+1 → In are inclusions. Also the arrow I∞ → I∞
is given by multiplication by y.
For instance each In, 1 ≤ n < ∞ is the source and the sink of the following AR-sequence in
the category of finitely generated CM-modules.
/
/
/
o
o
/
/
o
o
o
o
f
5
y(cid:17)
(cid:16) 1
−−−→ In−1 ⊕ In+1
0 → In
(y,−1)
−−−−→ In → 0 .
Here, because we consider right modules, their morphisms act on the left.
Being Gorenstein, R is a projective and injective object in the category of finitely generated
CM-modules, therefore no AR-sequence starts or ends in R.
These AR-sequences and irreducible morphisms are clearly visible on the above diagrams. For
instance the irreducible map I1
y
−→ I2 amounts to dividing x ∈ I1 by y.
I1
x
•
❄❄❄❄❄
y
•
x
y
•
⑧⑧⑧⑧⑧
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
=⇒
I2
x
◦
y
•
❄❄❄❄❄
y
•
x
y2
•
⑧⑧⑧⑧⑧
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
Similarly the inclusion I1 ⊂ R is shown by completing the square on the following diagram.
1
◦
y
x
•
❄❄❄❄❄
y
x
•
⑧⑧⑧⑧⑧
•
❄❄❄❄❄
y
y
x
•
❄❄❄❄❄
⑧⑧⑧⑧⑧ ...
• ...
It follows from the above description that the AR-quiver of R has two components, - the
first contains all 'finite' modules In; and the second consists of I∞. We show, on Figure 1, the
extended version of this quiver which contains both components, but also includes commutativity
relations between irreducible maps.
y
I∞
•
❄❄❄❄❄❄
...
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧ ...
•I4
y
y
•I3
I∞
•
...
...
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧ ...
...
•I3
y
•I1
y
•R
...
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧
•R ...
•R
y
•I1
y
...
•I2
⑧⑧⑧⑧⑧⑧
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧
❄❄❄❄❄❄
•I2
...
•I2
y
.
.
.
.
.
.
y
I∞
•
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧ ...
Figure 1.
6
Here the copies of R form the left (vertical) boarder of this diagram. Furthermore the line
consisting of copies of I∞ goes in southeastern direction and forms another boarder. Note that
starting with a copy of R and moving in the northeastern direction along the coray R ← I1 ←
I2 ← . . . of irreducible maps we obtain I∞ as the inverse limit (i.e. the intersection) of the
corresponding inverse system.
4. Free realizations of pp-formulae and patterns
We will briefly recall some notions of model theory of modules. In the first part of this section
S will denote an arbitrary commutative ring. For more explanations the reader is referred to
[14, Ch. 1 -- 4].
A positive-primitive formula ϕ(v) in one free variable v is an existentially quantified formula
∃ v (vA = v¯b), where v = (v1, . . . , vk) is a tuple of bounded variables, A is a k × n matrix over
S and b is a row of elements of S of length n. If M is an S-module and m ∈ M we say that M
satisfies ϕ(m), written M = ϕ(m), if there exists a tuple m = (m1, . . . , mk) of elements from
M such that mA = m¯b.
This pp-formula claims the decidability, on m in M , of the system of linear equations given
.
=
by the matrix (Ab). For instance if r ∈ S then vr = 0 is the annihilator formula, and r v
∃ w (wr = v) is the divisibility formula.
Since S is commutative, the set ϕ(M ) = {m ∈ M M = ϕ(m)} is a submodule of M . For
instance (r v)(M ) = M r, and (vr = 0)(M ) is the kernel of right multiplication by r.
If ϕ and ψ are pp-formulae we say the ϕ implies ψ, written ϕ → ψ or ϕ ≤ ψ, if ϕ(M ) ⊆ ψ(M )
for any module M . These formulas are said to be equivalent if both ϕ ≤ ψ and ψ ≤ ϕ
hold. The equivalence classes of pp-formulae form a modular lattice L of pp-formulae over
S. The meet in this lattice is the conjunction of pp-formulae, and the sum is the pp-formula
(ϕ + ψ)(v)
.
= ∃ w ϕ(v) ∧ ψ(v − w).
Each pp-formula ϕ has a finitely generated free realization, i.e. a finitely generated module
M pointed at an element m such that 1) M = ϕ(m) and 2) if M = ψ(m) for some pp-formulae
ψ then ϕ ≤ ψ. For instance the pointed module (S, r) is a free realization of the divisibility
formula r v, and the module S/rS pointed at 1 is a free realization of the annihilator formula
vr = 0.
Suppose (M, m) is a free realization of a pp-formula ϕ and (N, n) is a free realization of a
pp-formula ψ. Then ψ ≤ ϕ iff there exists a pointed morphism, i.e. a morphism f : M → N of
S-modules such that f (m) = n.
Algebraically pp-formulae in one free variable can be defined as finitely generated subfunctors
of the functor Hom(S, −) from the category of S-modules to the category of abelian groups.
For instance the annihilator formula vr = 0 corresponds to the functor Hom(S/rS, −), and the
divisibility formula r v gives rise to the functor G such that G(M ) = M r for each module M .
Recall (see [14, Sect. 3.4]) that a class of S-modules D is said to be definable if it is closed
with respect to direct products, direct limits and pure submodules. In fact this class consists of
models of a uniquely determined first order theory T .
7
The notion of the lattice of pp-formulae can be relativized to any definable class. Namely
for pp-formulas ϕ, ψ we set ϕ ≤T ψ if ϕ(M ) ⊆ ψ(M ) for any module in D. The corresponding
lattice of pp-formulae LT will be the factor of the lattice L of all pp-formulae. For instance, if
DM is the smallest definable class containing a module M , then we obtain the lattice LM of
pp-definable submodules of M .
The class of CM-modules over R is clearly definable, hence we obtain the theory TCM of
Cohen-Macaulay modules over R, and could relativize the above notions to this theory. For
instance LCM will denote the lattice of pp-formulae in this theory, which is a factor of the
lattice of pp-formulae by a certain congruence relation. In fact it is not difficult to choose a
canonical representative in each equivalence class.
Namely let ϕ be a pp-formula freely realized by a pair m ∈ M , where M is a finitely generated
R-module. If M is a factor of M by its y-torsion submodule, it is a CM-module. Let m denote
the image of m in M and let a pp-formula ϕCM generate the pp-type of m in M . Clearly we
have the implication ϕCM → ϕ, and both formulas are equivalent in TCM .
Thus the lattice LCM of pp-formulae in TCM can be defined as a lattice of pointed finitely
generated CM-modules. Namely if m ∈ M represents a pp-formula ϕ and n ∈ N gives ψ, then
the pair (m, n) ∈ M ⊕ N represents ϕ + ψ. Their conjunction ϕ ∧ ψ is given by the factor of the
following pushout module (K, k) by its y-torsion submodule.
(R, 1)
(M, m)
(N, n)
/❴❴
(K, k)
π
/ (K, k)
We will often identify pp-formulas in LCM with their free realizations in finitely generated
CM-modules and refer to them as CM-formulae.
Now we consider the so-called patterns of pointed finitely generated CM-modules over R.
The term is borrowed from Ringel [18, Sect. 3] where he calculates the hammock functors: the
traces Hom(N, −) of simple regular modules N in indecomposable finite dimensional modules
over tame hereditary finite dimensional algebras.
A pointed CM-module is a pair (M, m), where M is a CM-module and m is an element of M :
we will often write this as m ∈ M . This pair can be also thought of as a morphism R → M
sending 1 to m. A pointed morphism (or just morphism) of pointed modules (M, m) and (N, n)
is a morphism f : M → N of R-modules such that f (m) = n.
Let (M, m) be an indecomposable finitely generated pointed CM-module. We will introduce
a CM-pattern (or just pattern) P of (M, m), which is a partially ordered set, as follows. The
elements of P will be equivalence classes of pointed morphisms from (M, m) to indecomposable
finitely generated CM-modules.
Let f : (M, m) → (N, n) and g : (M, m) → (K, k) be pointed morphisms in P . We set f ≥ g
if there exists a morphism h : N → K such that h(n) = k. Clearly this relation is reflexive and
transitive. To make it anti-symmetric we factor P by the equivalence relation f ∼ g if f ≥ g
and g ≥ f .
/
/
✤
✤
✤
/
/
8
x∈I2
x∈I∞
•
❉❉❉❉❉❉❉
xy∈I∞
•
❉❉❉❉❉❉❉
xy2∈I∞
•
...
•
...
❉❉❉❉❉❉❉
xy∈I3...
❉❉❉❉❉❉❉
④④④④④④④
xy2∈I4...
❉❉❉❉❉❉❉
...
④④④④④④④
...
④④④④④④④
...
xy2∈I3
•
•
•
xy∈I2
•
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
...
xy2∈I2
•
•
xy∈I1
•
x∈I1
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
...
xy2∈I1
•
x∈R
•
xy∈R
•
xy2∈R
•
Figure 2.
Usually the structure of P is quite involved. The following example of a pattern will be of
crucial importance for this paper.
Proposition 4.1. Figure 2 shows the pattern of the pointed module x ∈ I∞.
Observe that this diagram is just a filter in the AR-quiver of R - see Figure 1.
Proof. Every nonzero morphism f from I∞ to an indecomposable finitely generated CM-module
N sends x, up to a multiplicative unit in F [[y]], to xyi, hence f is given by multiplication by yi.
Clearly multiplication by this unit does not change the equivalence class of f .
The remaining part is by inspection. For instance xy ∈ I2 ≥ xy ∈ I1 by inclusion, and
(cid:3)
xy ∈ I2 ≥ xy2 ∈ I3 via multiplication by y.
Thus on the diagram we see the traces of the functor Hom(I∞, −) on indecomposable finitely
generated CM-modules. The pattern of another pointed module x ∈ R consists of points on
Figure 2 located below (and including) this module.
If ϕ and ψ are pp-formulae then [ϕ ∧ ψ, ϕ] denotes the corresponding interval in the lattice
of pp-formulae (or some relativized version of it). We will often refer to this interval by saying
'ϕ over ψ'. For instance the pair of pp-formulas ψ < ϕ is said to be minimal in a theory T if
the interval [ψ, ϕ]T is simple.
Lemma 4.2. The interval [v = 0, vx = 0] in the lattice of CM formulae over R is distributive.
Proof. By Remark 5.1 and similarly to [15, Prop. 4.4] it suffices to prove that, when evaluated
on any indecomposable finitely generated CM module M , this interval is a chain. But this is
obvious, because every cyclic submodule of (vx = 0)(M ) is generated by xyi for some i, and
such submodules are linearly ordered by inclusion.
(cid:3)
Now we are in a position to show that the diagram on Figure 2 describes the above interval.
9
Proposition 4.3. 1) Every CM-formula in the interval [v = 0, vx = 0] is equivalent to a finite
sum of marked on Figure 2 formulas.
2) This sum is free in the following sense. If ϕi and ψj are finite sets of marked formulae,
then Pi ϕi implies Pj ψj if and only if for every i there exists j such that ϕi ≤ ψj.
3) The intersection of marked formulae equals the marked intersection, but this is not true
for sums.
Proof. 1) Because x ∈ I∞ is a free realization of the annihilator formula vx = 0, every CM-
formula ψ below vx = 0 corresponds to a pointed morphism from x ∈ I∞ to a finitely generated
CM-module N . Decomposing N into a direct sum of indecomposables we conclude that ψ is
equivalent to a finite sum of CM formulae ψj given by pointed morphisms from I∞ to indecom-
posable finitely generated CM-modules.
By the description of such morphisms (x goes to xyi) we derive that ψj is equivalent to a
marked formula on the diagram.
2) Clearly it suffices to consider the implication ϕ → ψ1 + · · · + ψm for CM-formulas on the
diagram. Let the pair (N, n) on the diagram represents ϕ. Evaluating on N , by uniseriality,
we conclude that ϕ(N ) ⊆ ψj(N ) for some j. But then ϕ implies ψj by the definition of a free
realization.
3) Let ϕ and ψ be incomparable CM-formulae marked on the diagram. By what we have
proved the CM-formula ϕ ∧ ψ is equivalent to a sum of formulae on the diagram which lie below
ϕ and ψ. But from the diagram we see that there is a unique largest element in this set, hence
it must be equal to the conjunction.
(cid:3)
Now the CM-formulas in the interval [v = 0, vx = 0] are easily visualized. For instance, the
interval x ∈ I∞ over xy ∈ I∞ in LCM is the following chain of order type ω + 1.
• x∈I∞
...
• (x∈I1)+ (xy∈I∞)
❉❉❉❉❉❉❉
• (x∈R)+ (xy∈I∞)
❇❇❇❇❇❇
• xy∈I∞
In particular the pair (x ∈ R) + (xy ∈ I∞) over xy ∈ I∞ is minimal in TCM .
Similarly the interval x ∈ I1 over x ∈ R in LCM is the following chain of order type 1 + ω∗.
• x∈I1
⑦⑦⑦⑦⑦⑦
• (x∈R)+ (xy∈I2)
⑥⑥⑥⑥⑥⑥⑥
• (x∈R)+ (xy2∈I3)
...
•x∈R
10
Furthermore the pair x ∈ I1 over (x ∈ R) + (xy ∈ I2) is minimal in TCM and corresponds
to the left almost split map I1
in TCM and corresponds to the irreducible map R
CM-modules.
y(cid:17)
(cid:16) 1
−−−→ R ⊕ I2. Also the pair x ∈ R over xy ∈ I1 is minimal
y
−→ I1 in the category of finitely generated
Note that the whole lattice of CM-formulae over R is more complex and involves rather
an unintelligent picture; for instance it is not distributive. Namely otherwise R would have a
distributive lattice of ideals. Since R is local it would yield that R is a valuation domain, a
contradiction.
5. The Ziegler spectrum
First we recall few more definitions from model theory of modules. A submodule M of a
module N is said to be pure if, for every m ∈ M and a pp-formula ϕ, from N = ϕ(m) it follows
that N = ϕ(n); and this definition is obviously extended to embeddings of modules.
A module M is said to be pure injective, if it injective with respect to pure monomorphisms
of modules. An equivalent requirement is that M splits any pure embedding. For instance
each injective module is pure injective. Furthermore by Ringel (see [14, L. 4.2.8]) each module
which is linearly compact over its endomorphism ring is pure injective. Because R is a complete
noetherian ring, it is linearly compact, hence pure injective. Since linearly compactness respects
extensions, each finitely generated R-module is also pure injective.
The Ziegler spectrum of a ring R, ZgR, is a topological space whose points are isomorphism
types of indecomposable pure injective modules. The topology on this space is given by pairs
(ϕ/ψ) of pp-formulae. Here the open set (ϕ/ψ) consists of points M such that there exists
m ∈ M satisfying ϕ but not ψ. In this case we will say that M opens the corresponding interval
[ϕ ∧ ψ, ϕ].
It follows from Ziegler (see [14, Thm. 5.1.22]) that (ϕ/ψ) is a quasi-compact open set, in
particular ZgR is a quasi-compact space, which is often non-Hausdorff. More properties of
Ziegler spectrum can be found in [14, Ch. 5]
Since the class of CM-modules is definable, one could talk about the theory TCM of CM-
modules, therefore about the closed subset of ZgR consisting of indecomposable pure injective
CM-modules, with the induced topology. We will call this topological space the CM-part of the
Ziegler spectrum, ZCMR; and investigating this space is the main objective of this paper. For
instance each basic open set (ϕ/ψ)T is compact so as the whole space ZCM.
As we have already noticed every indecomposable finitely generated CM-module over R is a
point in ZCM. Furthermore, since each CM-module over R is a union of its finitely generated
submodule we obtain the following.
Remark 5.1. The finitely generated points are dense in ZCM.
We add more points to this list.
Remark 5.2. eR, Q and the Laurent power series field G = F ((y)) are points in ZCM.
Here are the shapes of these modules.
11
...
eR
•
❁❁❁❁❁
y
•
❁❁❁❁❁
y
...
G
Q
...
...
•x
✂✂✂✂✂
•
❁❁❁❁❁
y
y
❁❁❁❁❁
•x
✂✂✂✂✂
•
❁❁❁❁❁
y
y
❁❁❁❁❁
•x
✂✂✂✂✂ ...
• ...
• y
❁❁❁❁❁
• y
❁❁❁❁❁
• ...
1
•x
✂✂✂✂✂
•
❁❁❁❁❁
y
y
❁❁❁❁❁
•x
✂✂✂✂✂ ...
• ...
Proof. Clearly all these modules are CM.
Because Q is injective and uniform it is indecomposable. Also G = F ((y)) is annihilated by x
and is injective and indecomposable as an R/xR = F [[y]]-module, hence pure injective over R.
Being uniform, the module eR is indecomposable. We check that it is linearly compact, hence
pure injective. Consider the following filtration of eR considered as a submodule of Q: 0 ⊂ eRx ⊂
Qx ⊂ eR. Here the module eRx ∼= F [[y]] is linearly compact so as eR/Qx ∼= F [[y]]. Further the
same is true for Qx/ eRx which is the Prufer F [[y]]-module. Since linearly compactness respects
extensions, we obtain the desired.
(cid:3)
Note that the filtration on eR is given by pp-formulae, for instance Qx = (vx = 0)( eR).
Suppose that M is a pure injective module and m ∈ M . The set of all pp-formulae ϕ(v) such
that m ∈ ϕ(M ) is said to be a pp-type of m in M , written ppM (m). This set of formulae can be
also described as a filter in the lattice of pp-formulae, i.e. a subset of L which is upward closed
and closed with respect to finite conjunctions; hence as a filter of finitely generated subfunctors
of the functor Hom(R, −). If M is pure injective indecomposable and m is nonzero, then p is
said to be indecomposable.
We will assign to a pp-type p its positive part p+ consisting of pp-formulas in p, and its negative
part p− which consists of pp-formulae not in p. If ψ < ϕ are pp-formulae, then p defines a cut
on the interval [ψ, ϕ] whose upper part is p+ ∩ [ψ, ϕ], and lower part is the intersection of p−
with this interval. If ϕ ∈ p+ and ψ ∈ p−, i.e. the corresponding cut is nontrivial, then we say
that p opens this interval.
It follows from Ziegler that an indecomposable pure injective module M is uniquely determined
by the pp-type of any its nonzero element. Furthermore, see [14, Thm. 5.1.24], to recover M it
suffices to know any arbitrary small piece of local information on p, i.e. a nontrivial cut defined
by p on any interval of the lattice of pp-formulae. Also, see [24, Thm. 4.4], there is a useful
criterion for checking when p is indecomposable. Of course all these relativizes to any theory of
R-modules, in particular to the theory of CM-modules.
We claim that the modules mentioned in Remark 5.2 are the only remaining points in ZCM.
Theorem 5.3. The modules eR, Q and G are the only non-finitely generated points in the
Cohen-Macaulay part of the Ziegler spectrum of R.
Proof. Let N be an indecomposable pure injective CM-module over R. If N is annihilated by x,
it is an indecomposable pure injective R/xR ∼= F [[y]]-module without y-torsion. The description
of the Ziegler spectrum of this valuation domain is well known - see [14, Sect. 5.2.1]. It follows
12
x∈I2
•
xy∈I2
•
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
...
xy2∈I2
•
•
xy∈I1
•
x∈I1
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
❉❉❉❉❉❉❉
④④④④④④④
...
xy2∈I1
•
x∈R
•
xy∈R
•
xy2∈R
•
•
...
❉❉❉❉❉❉❉
xy∈I3...
❉❉❉❉❉❉❉
④④④④④④④
xy2∈I4...
❉❉❉❉❉❉❉
...
④④④④④④④
...
④④④④④④④
...
xy2∈I3
•
•
• x∈eR
• xy∈eR
x∈I∞
•
❉❉❉❉❉❉❉
xy∈I∞
•
❉❉❉❉❉❉❉
xy2∈I∞
•
...
• 1∈G
• xy−2∈eR
• xy−1∈eR
• xy2∈eR
• x∈Q
Figure 3.
that either N ∼= F [[y]] (the adic point), hence N ∼= I∞; or N ∼= F ((y)) (the generic point),
therefore N ∼= G.
Thus we may assume that N x 6= 0. Choose 0 6= n ∈ N x and let p be the pp-type of n in
N . In particular p is uniquely determined by its CM-part, and the formula vx = 0 (even x v!)
belongs to p. It follows that p defines a nontrivial cut on the interval [v = 0, vx = 0] on Figure
2. Furthermore the isomorphism type of N is uniquely determined by this cut.
By Ziegler's criterion for indecomposability and distributivity (see Lemma 4.2) we conclude
that for any CM formulas ϕ, ψ ∈ p− marked on the diagram, their sum ϕ + ψ is also not in p.
Thus the indecomposable pp-type p is uniquely determined by a filter of formulas on the diagram,
i.e. by an upward closed and closed with respect to conjunctions set of marked formulae (recall
that the sums are free, hence hidden on the diagram).
Then either p is finitely generated (i.e. the corresponding filter is principal), hence is realized
in an indecomposable finitely generated CM-module, or is generated by a line going in the
southeastern direction on the diagram, or includes all nonzero CM formulae from the interval.
Comparing pp-types we obtain the possibilities shown on Figure 3.
For instance the (indecomposable) pp-type p defined by the ray (R, x)
y
−→
. . . coincides with the pp-type of x ∈ eR, therefore eR is the pure injective envelope of p. Similarly
y
we will obtain eR as the direct limit of the parallel ray of irreducible morphisms I1
−→ . . .
starting with x ∈ I1, where xy−1 ∈ eR realizes the corresponding pp-type.
y
−→ (I2, xy2)
y
−→ (I1, xy)
y
−→ I2
13
Further G is the direct limit of the ray of irreducible morphisms I∞
y
−→ R
realizes p. Also Q is the direct limit of the directed system R
y
−→ . . . , where 1 ∈ G
y
−→ I∞
y
−→ . . . heading downwards,
where the pp-type of x ∈ Q equals p and is critical over zero.
(cid:3)
To complete a description of ZCM it remains to describe the topology, i.e. to give a basis of
open sets to each point of this space. We will describe this open basis as a collection of basic open
sets (ϕ/ψ) containing this point. Evaluating we will also give a subset of the Ziegler spectrum
corresponding to each open set in this basis. In fact all this can be read off the diagram.
Namely it follows from Ziegler [24, Thm. 4.9] that, if ϕ ∈ p+, ψ ∈ p− for pp-formulae ψ < ϕ
and an indecomposable pp-type p, then the basis of open sets for the pure injective envelope of
p can be chosen among pairs (ϕ′/ψ′), where ψ ≤ ψ′ < ϕ′ ≤ ϕ and ψ′ ∈ p−, ϕ′ ∈ p+.
Remark 5.4. Every finitely generated point In, n < ∞ is isolated by a minimal pair in ZCM.
Proof. By inspection of Figure 2 using AR-sequences.
For instance I1 is isolated by the minimal pair x ∈ I1 over (x ∈ R) + (xy ∈ I2) coming from
the corresponding left almost split map. Also R is isolated by the minimal pair x ∈ R over
xy ∈ I1 given by an irreducible morphism (multiplication by y).
(cid:3)
For I∞ one should work harder.
Lemma 5.5. Let Om denote the (cofinite) set of finite points In, n ≥ m. The basis of open sets
for I∞ is given by sets Om ∪ I∞, m = 1, 2, . . . . For instance I∞ is not isolated.
Proof. Since x ∈ I∞ satisfies the formula vx = 0, the basis of open sets for I∞ can be chosen
it is visible on Figure 2. For instance x ∈ I∞ over
within the interval [x = 0, vx = 0], i.e.
(xy ∈ I∞) + (x ∈ Im) is such a basis.
By evaluating we see that the open set defined by this pair is Om+1 ∪ I∞.
(cid:3)
Recall (see [14, p. 254]) that an indecomposable pp-type p is said to be neg-isolated if there
a pp-formula ψ ∈ p− such that p is maximal among pp-types containing ψ in its negative part.
If M is an indecomposable pure injective module realizing a neg-isolated pp-type p, then M is
called neg-isolated; this notion does not depend on the choice of p. For instance from the above
description it follows that I∞ is not neg-isolated.
Now we are in a position to describe isolated points.
Corollary 5.6. The isolated points in ZCM of R are exactly modules In, 0 ≤ n < ∞.
Proof. Since G is the direct limit of copies of I∞ (see Figure 3) and the space ZCM is compact,
this module cannot be isolated. By a similar reason neither eR nor Q is isolated. As we have
already seen I∞ is also non-isolated. It remains to apply Remark 5.4.
(cid:3)
Here we meet the first peculiarity. For finite dimensional algebras the existence of AR-
sequences implies (see [14, Cor. 5.3.37]) that isolated points in the Ziegler spectrum are exactly
indecomposable finite dimensional modules. The original feeling was that these finite dimen-
sional points should correspond to finitely generated CM-points in out setting. However this is
not the case, because I∞ is finitely generated but not isolated.
14
Now we deal with with eR.
Lemma 5.7. Let Om be as in Lemma 5.5. The basis of open sets for eR is given by sets Om ∪ eR,
m = 1, 2, . . . .
Proof. Consider eR as a pointed module x ∈ eR, i.e. as the pure injective envelope of the pp-type
p of x ∈ eR. From Figure 3 we conclude that a basis of open sets for this point is given by
the line going from x ∈ R in the southeastern direction, hence by pairs xyn ∈ In over xy ∈ R,
n = 1, 2, . . . . In particular this point is neg-isolated.
Evaluating this pair on ZCM we obtain the open set On ∪ eR, as desired.
Recall that the Cantor -- Bendixson analysis on a compact topological space runs by a con-
(cid:3)
secutive removal its isolated points - see [14, Sect. 5.3] how it applies to the Ziegler spectrum.
This way one obtains an ordinal indexed descending chain of closed subspaces T (λ). If T (λ) 6= ∅
and T (λ+1) = ∅ for some ordinal λ then we say that the Cantor -- Bendixson rank, CB-rank, of T
equals λ. In this case each point t ∈ T is assigned its CB-rank, being the smallest µ such that
t ∈ T (µ) \ T (µ+1).
Thus on the first step of the CB-analysis of ZCM we remove all isolated points In, n < ∞.
The next level is described in the following lemma.
Lemma 5.8. I∞ and eR are the only points in ZCM of Cantor -- Bendixson rank 1.
Proof. It follows from Lemmas 5.5, 5.7 that both points have CB-rank 1. Furthermore G is the
direct limit of copies of I∞, hence is in the closure of this point, in particular G has CB-rank at
least 2.
Similarly Q is the direct limit of copies of eR, therefore its CB-rank exceeds 1.
The next level of the Cantor -- Bendixson analysis is the last.
(cid:3)
Lemma 5.9. The point G has CB-rank 2. Furthermore a basis of open sets for G is obtained
from ZCM by removing Q and finitely many finite points In, n ≤ m.
Proof. Choosing 1 ∈ G, from Figure 3 we conclude that a basis of open sets for G is given by
pairs xyn ∈ I∞ over x ∈ Im, m, n = 0, 1, . . . .
The remaining part is by evaluation. For instance the finite point Ik belongs to the above
open set iff k > m + n; and I∞, eR belong to any of those open sets.
We have already seen that CB(G) ≥ 2. Since the pair x ∈ I∞ over x ∈ R separates G from
(cid:3)
Q we conclude that CB(G) = 2.
To complete the Cantor -- Bendixson analysis it remains to deal with Q.
Lemma 5.10. The point Q has CB-rank 2. Furthermore a basis of open sets for Q is obtained
from ZCM by removing G and I∞.
Proof. Pointing Q at x from Figure 3 we see that a basis of open sets for Q is given by pp-pairs
xyn ∈ R over x = 0, hence this point is critical over zero in terminology of [9].
15
The remaining part is by evaluation, in particular xy ∈ R over x = 0 separates Q from G and
I∞. On the other hand none element in this basis separates Q from any finite point Im, hence
Q is in the closure of each Im.
(cid:3)
As a result we conclude on the value of CB-rank.
Theorem 5.11. The Cantor -- Bendixson rank of the Cohen -- Macaulay part of the Ziegler spec-
trum of R equals 2.
From the above description of the topology it follows that the only closed points of ZCM are
endofinite points G and Q, i.e. points of maximal CB-rank.
Recall (see [11, Exerc. 7.10]) that over a finite dimensional algebra each pure injective module
is isomorphic to a direct summand of a direct product of finite dimensional modules. Here all
looks differently.
Lemma 5.12. None of the points eR, Q and G is isomorphic to a direct summand of a direct
product of finitely generated CM-modules.
Proof. We will give a proof for eR, and the remaining cases are similar.
From Figure 3 we conclude that the pp-type p of x ∈ eR is generated by formulas xyn ∈ In,
n = 1, 2, . . . . Suppose that eR is realized as a direct summand of a direct product of finitely
generated CM-modules Mi, i ∈ I such that the element m = (mi)i∈I realizes p. If the CM-
formula ϕi generates the pp-type of mi in Mi, then ϕi will be below each pp-formula in p. Looking
at Figure 3 we see that ϕi is equivalent to the zero formula, hence m = 0, a contradiction. (cid:3)
6. Krull -- Gabriel dimension and m-dimension
In this section we will calculate the m-dimension of the lattice LCM of CM-formulae over R.
This is the same as the Krull -- Gabriel dimension of the definable category of Cohen -- Macaulay
R-modules (see [14, Sect. 13.2.2]), but to avoid introducing many definitions we will not use
this notion.
Suppose that L is a lattice with top and bottom. Following [14, Sect. 7.2] by induction on
ordinals we define a sequence of congruence relations ∼λ, hence factor lattices Lλ = L/ ∼λ.
Namely let ∼0 be the trivial relation, hence L0 = L. On limit steps λ we define ∼λ= ∪µ<λ ∼µ.
Further, for a non-limit ordinal λ + 1, let the congruence relation on Lλ collapses intervals of
finite length, and let ∼λ+1 be the preimage of this relation in L. If there is an ordinal λ such
that the lattice Lλ is nontrivial (i.e. contains at least two elements) and Lλ+1 is a trivial lattice
than we define the m-dimension of L to be equal to λ.
We will be interested in the case when L is the lattice LCM of CM-formulae over R. For
instance on the first step of the m-dimension analysis the simple interval x ∈ R over xy ∈ I1
on Figure 3 collapses in L1. In general, the m-dimension analysis runs parallel to the Cantor-
Bendixson analysis. Namely (see [14, Sect. 5.3.2]) if, at each step of the Cantor-Bendixson
analysis, each point is isolated by a minimal pair, then the lattice Lλ coincides with the lattice
of pp-formulae of the CB derivative T (λ)
CM of the theory of CM-modules.
16
From the analysis of the previous section it follows that this is the case for our singularity R,
therefore we obtain the following.
Proposition 6.1. The m-dimension of the lattice of CM-formulae over R equals 2.
Recall (see [14, Sect. 6.1]) that a ring of definable scalars of a module M consists of biendo-
morphisms of M defined by pp-formulas ϕ(v, w) in two free variables, where the first variable
describes the domain, and the second sets the image of this map. In our case these objects are
easily calculated.
Proposition 6.2. 1) The ring of definable scalars of the module In coincides with the overring
Rn of R generated (over R) by xy−n.
2) The ring of definable scalars of I∞ is the ring F [[y]].
3) The ring of definable scalars of eR equals eR.
4) The ring of definable scalars of G is the field F ((y)).
5) The ring of definable scalars of Q equals Q.
Note that in each case 1) -- 5) the module is isomorphic to its ring of definable scalars.
Proof. We will consider only the case 1), the remaining cases are by inspection.
Because R is Gorenstein, it has a unique minimal overring which is easily checked to coincide
with R1. Now looking at the diagram for I1 from Section 3 we see that multiplication by xy−1
defines a biendomorphism of I1: each element of this module when multiplied by x is uniquely
divisible by y.
Furthermore, because I1 = m is isomorphic to R1 (via the map x 7→ xy−1 and y 7→ 1), we
conclude that the ring of definable scalars of I1 coincides with R1.
The ring R1 is again Gorenstein, hence has a unique minimal overring R2. Continuing this
(cid:3)
way we obtain the desired.
Recall that the lattice LCM of all CM-formulas over R is too large to be completely described.
Its first derivative is manageable.
Proposition 6.3. The diagram on Figure 4 represents the lattice L1 of pp-formulas of the first
derivative T ′ of the theory of Cohen -- Macaulay modules over R. For instance this lattice is
distributive.
Proof. It follows from the previous section that T ′ is the closure, in the Ziegler spectrum, of
points I∞ and eR of CB-rank 1. Since I∞ is definable in eR it follows that T ′ coincides with the
theory of CM-modules defined over eR. Since eR is a valuation ring, this lattice is easily calculated
- see [16, Ch. 11] for a more general setting.
(cid:3)
On the next level of CB-analysis we obtain a finite length lattice.
Proposition 6.4. The following diagram represents the lattice L2 of pp-formulae of the second
derivative T ′′ of the theory TCM of Cohen -- Macaulay modules over R.
17
Q
G
Q
• v=v
• vx=0
• xv
• v=0
On this diagram we match simple intervals with points of the Ziegler spectrum they isolate.
Proof. From the previous section it follows that T ′′ is the closure of points Q and G of CB-rank
2. Because y acts as an automorphism on both modules, we conclude that y−1 is a definable
scalar of this theory. Therefore T ′′ is just the theory of S = F ((y))[[x]]/(x2)-modules. The last
ring is uniserial of length 2, hence its lattice of pp-formulas is well known.
(cid:3)
7. Ringel's quilt
This combinatorial object was introduced by Ringel [19] (see also [20]), under the name of
the Auslander -- Reiten quilt, when investigating the module category of domestic string (finite
dimensional) algebras. It serves as a tool for sewing components of the AR-quiver for better
understanding of morphism between modules from different AR-components. The devices used
in implementing this construction are infinitely generated indecomposable pure injective modules
and irreducible morphisms between them.
We will proceed in the same vein. First we need some supply of irreducible maps and almost
split sequences.
Lemma 7.1. The map Q x−→ G is irreducible in the category of CM-modules.
v=v
•
•
•
⑤⑤⑤⑤⑤
❇❇❇❇❇
⑤⑤⑤⑤⑤ .
•
•
.
yv
y2v
❇❇❇❇❇
⑤⑤⑤⑤⑤ .
•
.
.
.
.
⑤⑤⑤⑤⑤
•
•
vx=0
⑤⑤⑤⑤⑤
.
.
•
.
.
.
.
.
.
• xy−1v
• xv
• xyv
.
.
.
• v=0
Figure 4.
18
Proof. Suppose that f : Q → N and g : N → G, where N is a CM-module, are morphisms
whose composition is the above map. If f is a monomorphism it splits, because Q is injective.
Otherwise f increases the pp-type of x ∈ Q, hence (see Figure 3) annihilates this element.
Because N is CM, it follows that f (Qx) = 0, hence ker f = Qx, otherwise g fail to exists.
Since Q/Qx ∼= G via 1 7→ x, we conclude that f induces the map ¯f : G → N such that the
(cid:3)
composition g ¯f : G → G sends x to x, hence is an isomorphism. It follows that g splits.
Note that this irreducible epimorphism is included into a short exact sequence 0 → G → Q
x−→
G → 0. However this sequence is far from being almost split: for instance the natural inclusion
G ⊂ eR cannot be factored through Q.
Recall that a module M over a CM-ring S is said to be free on the punctured spectrum, if its
localization MP is free over the localized ring SP for each non-maximal prime ideal P . Further
S is an isolated singularity if SP is a regular ring for each non-maximal prime ideal P .
By Auslander's result (see [13, Thm. 13.8]) if M is a f.g. CM-module free on the punctured
spectrum, then M is the sink of an AR-sequence in the category of finitely generated CM-
modules. For R this applies to finite modules In, and it is easily seen that the corresponding
AR-sequences retain their defining properties in the category of all CM-modules.
The module I∞ is not free on the punctured spectrum: if P = xR, then I∞ localized at P is
the generic module G which is not free over the ring RP = F ((y))[[x]]/x2. It follows from [13,
Prop. 13.3] that I∞ cannot be a sink of an AR-sequence in the category of finitely generated
CM-modules.
However, it follows from another result of Auslander [2, Thm. 5] (see also [9]) that I∞ is the
sink of an AR-sequence in the category of all R-modules whose source is indecomposable and
pure injective. Using an approximation of this module in the category of all CM-modules we
found the following AR-sequence, now in the category of CM-modules. We suppress calculations,
because the result is easily verified.
Proposition 7.2. The following is an AR-sequence in the category of Cohen -- Macaulay modules
over R.
0 → eR
f = ( y
x )
−−−−−→ eR ⊕ I∞
(x,−y)
−−−−→ I∞ → 0 .
Proof. The exactness of this sequence is easily shown by inspection. Since the endomorphism
ring of I∞ is local, by [1, Prop. 4.4] it suffices to check that f is left almost split.
Suppose that h : eR → N is a non-split monomorphism, where N is a CM-module.
In
particular for n = h(x) we obtain nx = 0.
eR ⊕ I∞
✇
f
✇
u
✇
eR
h
{✇
N
/
/
{
19
✿
☎
✿
☎
✿
✿
☎
☎
✿
✿
G
•
☎
✿
✿
✿
✿
✿
☎
☎
✿
☎
☎
☎
☎
x
☎
☎
x
☎
I∞...
/.-,
()*+•
o❴ ❴
y
y
y
y
y
I2
•
✿
✿
✿
I∞
•
o❴ ❴ ❴ ❴ ❴
✿
✿
✿
I∞
•
...
❂❂❂❂❂❂
...
I2...
/.-,
()*+•
✿✿✿✿✿✿✿
❅❅❅❅❅❅❅
⑦⑦⑦⑦⑦⑦
I3...
I1...
/.-,
()*+•
•
❂❂❂❂❂❂❂
❅❅❅❅❅❅❅
...
⑦⑦⑦⑦⑦⑦⑦
⑦⑦⑦⑦⑦⑦
I4...
/.-,
()*+•
•
...
❅❅❅❅❅❅R
❅❅❅❅❅❅❅
⑦⑦⑦⑦⑦⑦⑦
✁✁✁✁✁✁✁
...
•eR
I3
•
...
❅❅❅❅❅❅
⑦⑦⑦⑦⑦⑦⑦
☎☎☎☎☎☎☎
⑦⑦⑦⑦⑦⑦⑦
•eR
I2
/.-,
()*+•
...
❅❅❅❅❅❅❅
✁✁✁✁✁✁
⑦⑦⑦⑦⑦⑦⑦
()*+•eR
/.-,
}⑤⑤⑤⑤⑤⑤⑤ ...
}⑤⑤⑤⑤⑤⑤
•eR
❴❴❴❴❴❴
I1
/.-,
()*+•
•R
☎
☎
☎
☎
☎
☎
☎
☎
☎
☎
❴❴
☎
☎
I1
•
y
y
y
y
y
y
y
•R .
.
.
.
.
.
Q
•
Figure 5.
Observe that f (x) = (xy, 0) whose pp-type p in eR (see Figure 3) is generated by the formulas
xyn ∈ In−1, n = 1, . . . .
Since eR is pure injective and indecomposable by [14, Prop. 4.3.45] it follows that h increases
the pp-type of x ∈ eR, hence (see Figure 3 again) the pp-type q of h(x) in N contains the formula
xy ∈ R. But then q contains any conjunction (xy ∈ R) ∧ (xy ∈ In) = (xyn ∈ In−1), n = 2, . . . ,
therefore p ⊆ q.
Since N is pure injective, there exists a morphism u : eR → N sending xy to h(x), therefore
h − f u : eR → N satisfies (h − f u)(x) = 0. Replacing h by h − f u we may assume that h(x) = 0.
Since N is CM, it follows that h(Qx) = 0, hence h factors through eR
x−→ I∞, as desired.
(cid:3)
The following corollary is straightforward.
Corollary 7.3. The following morphisms are irreducible in the category of all Cohen -- Macaulay
R-modules.
1) eR
y
−→ eR;
2) eR x−→ I∞;
and
3) I∞
y
−→ I∞.
Proof. Either by projecting the above AR-sequence, or directly.
(cid:3)
Now we are in a position to draw - see Figure 5 - Ringel's quilt of the category of CM-modules,
and the procedure is very similar to what Ringel did in [19].
Namely first we change the metric to include the AR-quiver of R (see Figure 3) into a finite
region - it will be contained in the ambient triangle erected vertically on the vertex Q. Now
}
}
o
o
20
compactify this triangle by adjusting the copies of indecomposable pure injective CM-modules
to its boarder. Namely add copies of eR as direct limits along rays of irreducible maps. We
also add the irreducible maps eR
−→ eR, which are the direct limits of inclusion maps In+1 ⊂ In
in the AR-quiver; and adjust Q as the direct limit of this ray between the eR (going in the
y
−→ I∞ and
southwestern direction). Further add G as a limit of the ray of irreducible maps I∞
adjust remaining irreducible morphisms eR
x−→ I∞ and Q
x−→ G.
y
We use irreducible morphisms to connect the points on the boundary, hence give an orientation
to sides of the triangle to glue them.
◦
R
I∞
•
%❑❑❑❑❑❑❑❑
xqqqqqqqq G
eR
•
Q
We see that topologically Ringel's quilt of R lives on the Mobius stripe. A small deficiency is
the empty point on the boundary of the triangle which can be easily amended: adjust the zero
module to this point.
From Ringel's quilt (see Figure 5) we get a better understanding of morphisms between finitely
generated CM-modules. For instance, starting from R one can move along the southeastern line,
then live the AR-component in eR and enter the other components through I∞. After that one
could reenter the first component by embedding I∞ in R, - the morphism that heads in the
southwestern direction. This way we have completed the revolution.
Observe that x ∈ eR is divisible by any power of y. Thus every morphism from eR to a finitely
generated CM-module annihilates the submodule Qx of eR, therefore factors through I∞. Thus
I∞ is a point where eR enters the category of finitely generated CM-modules.
There is one misleading point concerning Figure 5. The choice where the morphism eR → I∞
enters the second component is completely arbitrary, but the construction becomes rigid as soon
as such choice has been made. Thus saying 'a revolution' is a bit ambiguous, the corresponding
walk is better represented by the following diagram.
◦
•
I∞
•
r
❚❱❳❨
■■■■■■■■■■■■
ttttttttttttt
%❑
• eR
❡ ❢ ❤ ❥
%▲
•
❑
❑
•
yr
•R
❑R
❑
•
▲R
eR
■
✉
It may be easier to grasp such walks in the following representations of Ringel's quilt - see
Fugure 6, where we dashed the fundamental domain. Thus it is obtained as a quotient of the
vertical strip on which a cyclic group acts by a glide-reflection. Again by adding the zero module
%
x
y
%
%
l
l
✤
we will get rid of small triangular deficiencies on this diagram, but one should add maps G → 0
and 0 → I∞ which are not irreducible.
21
8. The infinite radical
In this section we will calculate the nilpotency index of the radical of the category of finitely
generated CM-modules over R.
Define the radical, written rad, of this category as a set of all morphisms f : M → N
between finitely generated CM-modules such that for every indecomposable finitely generated
CM-module K, any composition K → M
f
−→ N → K is not invertible in the local ring End(K).
Clearly rad is a 2-sided ideal in this category.
Decomposing M and N as direct sums of indecomposable modules we represent f as a finite
matrix fij whose entries are morphisms between indecomposable modules. Clearly f ∈ rad
iff each fij ∈ rad. Thus when describing the radical is usually suffices to consider morphisms
between indecomposable modules.
Furthermore, because each indecomposable finitely generated CM-module is pure injective,
from [14, Prop. 4.3.45] it follows that a map f : M → N between such indecomposables belong
to the radical iff it increases the pp-type of one (equivalently any) nonzero element of M .
Next we will show that there are enough irreducible morphisms in our category.
Lemma 8.1. rad is generated by irreducible morphisms.
Proof. It suffices to prove that every morphism f : M → N between indecomposable finitely
generated CM-modules which lies in the radical belongs to the ideal generated by irreducibles.
If M is a finite point In, n ≥ 1, then this follows from the fact that In is the source of a left
almost split map. Suppose that M = I0 = R and look at at f (x). From Figure 3 we conclude
that f factors through the irreducible morphism R
y
−→ I1.
It remains to consider the case M = I∞. Observing the same figure we conclude that either
y
−→ I∞, or f has In as a target. In the latter case
(cid:3)
f factors through the irreducible morphism I∞
f factors through the irreducible inclusion In+1 ⊂ In.
We define the infinite powers of the radical, radλ, by a transfinite induction starting from
rad1 = rad. If λ is a limit ordinal, then set radλ = ∩µ<λ radµ; for instance the infinite radical
radω equals ∩n radn. Finally if λ = µ + m is a successor ordinal, then define radλ = (radµ)m+1,
for instance radω+1 = (radω)2. The least λ such that radλ = 0 (if exists) is called the nilpotency
index of the radical. For instance, the nilpotency index equals ω + 1 if there exists a nonzero
morphism in the infinite radical, but for any two such morphisms their composition equals zero.
There is a heuristic (see [22]) intuition how to use the m-dimension of the lattice L of pp-
formulae to guess the nilpotency index of the radical. Namely suppose that the m-dimension of
L equals β + 1, hence Lβ is the last nontrivial lattice in the m-dimension analysis. If the length
of Lβ equals n, then one could expect ωβ + n − 1 as the value of the nilpotency index. Since in
our case the m-dimension equals 2 and n = 3, we should get ω + 2, which is confirmed below.
The following remark is straightforward.
❇
I2
•
❇
❇
y
❇
❆❆❆❆❆❆❆❆
~⑥⑥⑥⑥⑥⑥⑥⑥
I2
/.-,
()*+•
...
y
x
•
~⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆❆eR
~⑥⑥⑥⑥⑥⑥⑥⑥
I∞
•
...
...
⑤
⑤
⑤
y
y
~⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆❆eR
x
22
⑧
❇
⑧
•
R
❇
❇
R
/.-,
()*+•
❇
y
y
#❋❋❋❋❋❋❋❋
}③③③③③③③
❆❆❆❆❆❆❆
~⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆❆
~⑥⑥⑥⑥⑥⑥⑥⑥
R
•
...
R
•
y
❇
❇
I1
•
❇
I1
/.-,
()*+•
❇
y
❇
!❉❉❉❉❉❉❉❉
~⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆
~⑥⑥⑥⑥⑥⑥⑥⑥
I1
•
...
y
❈
⑧
❉
•
...
!❉❉❉❉❉❉❉❉
❈
x
❈
❈
Q
•
❈
y
❋
❊
❊
❊
③
③
③
❋
❋
❋
❋
⑤
⑤
⑤
•
③
x
❊
❊
I2
•
G
•
...
❈
⑤
{①①①①①①①①
#❋❋❋❋❋❋❋❋
I∞
•
...
#❋❋❋❋❋❋❋❋
{①①①①①①①①
#❋❋❋❋❋❋❋❋
{①①①①①①①①
I1
•
...
{①①①①①①①①
#❋❋❋❋❋❋❋❋eR
I1
•
•
x
y
I∞
•
...
❊
❊
❊
❋
③
R
•
③
❋
❊
I1
•
❊
⑧
•
R
③
③
I∞
•
...
❋
#❋❋❋❋❋❋❋❋
{①①①①①①①①
#❋❋❋❋❋❋❋❋
{①①①①①①①①
#❋❋❋❋❋❋❋❋
{①①①①①①①①
R
•
...
R
•
Q
•
❉
•
...
#❋❋❋❋❋❋❋❋
❉
❉
x
❉
G
•
❉
①
①
①
①
❇
❇
I3
•
...
❇
❇
❇
❇
y
x
/.-,
()*+•
~⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆eR
~⑥⑥⑥⑥⑥⑥⑥⑥
I∞
•
...
⑤
⑤
⑤
❇
x
❇
❇
❇
•
❆❆❆❆❆❆❆❆eR
~⑥⑥⑥⑥⑥⑥⑥⑥
I∞
/.-,
()*+•
...
⑤
⑤
⑤
⑤
y
❇
❇
I∞
•
...
⑤
⑤
❇
⑤
✿
③
•
③
③
R
/.-,
()*+•
⑤
y
y
y
y
③
③
③
③
③
③
⑤
⑤
⑤
③
y
③
y
I1
•
③
③
y
③
I2
•
③
③
I1
/.-,
()*+•
③
③
I3
•
...
~⑥⑥⑥⑥⑥⑥⑥⑥R
❆❆❆❆❆❆❆
~⑥⑥⑥⑥⑥⑥⑥⑥
!❉❉❉❉❉❉❉❉
}③③③③③③③③
}③③③③③③③
#❋❋❋❋❋❋❋❋
#❋❋❋❋❋❋❋❋
{①①①①①①①①
#❋❋❋❋❋❋❋❋eR
I2
•
...
#❋❋❋❋❋❋❋❋
{①①①①①①①①
eR
•
#❋❋❋❋❋❋❋❋...
#❋❋❋❋❋❋❋❋
{①①①①①①①①
eR
R
•
•
...
#❋❋❋❋❋❋❋❋
#❋❋❋❋❋❋❋❋
{①①①①①①①①
eR
•
#❋❋❋❋❋❋❋❋...
...
{①①①①①①①①
I∞
•
...
...
{①①①①①①①①
❋
I∞
•
❋
I1
•
R
•
③
③
❋
❋
❋
❋
❋
❋
❋
❊
❊
❊
x
x
x
x
y
y
y
Q
•
③
✿
③
G
•
❋
③
③
③
❋
❋
❊
❊
x
❊
❊
I∞
•
#❋❋❋❋❋❋❋❋
{①①①①①①①①
①
①
y
①
①
❊
❊
I3
•
...
❊
#❋❋❋❋❋❋❋❋
{①①①①①①①①
I2
•
...
{①①①①①①①①eR
#❋❋❋❋❋❋❋❋eR
{①①①①①①①①
•
①
①
x
y
y
I∞
•
①
①
①
❊
❊
❊
•
y
{①①①①①①①①eR
#❋❋❋❋❋❋❋❋
{①①①①①①①①
I∞
•
①
①
x
y
①
①
①
❊
①
❊
①
Figure 6.
#
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
}
!
~
~
~
~
~
~
~
~
~
~
~
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
~
~
!
}
!
}
#
{
#
{
#
#
{
#
#
{
#
#
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
✤
#
{
#
#
{
#
{
#
#
{
{
{
#
{
#
{
#
{
{
#
{
{
#
{
{
#
{
#
23
Remark 8.2. Suppose that f : M → N is a morphism of indecomposable finitely generated
CM-modules. If f ∈ radω and mx = 0 for m ∈ M then f (m) = 0.
Proof. Because mx = 0 the pp-type of m in M is visible on Figure 2. Since f belongs to the
infinite radical, for each k, it can be written as a sum of products f1 · . . . · fk of radical maps
between indecomposable finitely generated CM-modules. Since each fi lower the pp-type of an
element on the figure, we derive the desired.
(cid:3)
In fact the above condition characterizes maps in the infinite radical, but we prefer to be even
more precise when describing such morphism between indecomposable modules.
Lemma 8.3. 1) Each morphism In → I∞ and I∞ → In belongs to radω.
2) A morphism h : Im → In belongs to radω iff it factors through I∞.
3) Each nonzero endomorphism of I∞ does not belong to radω.
Proof. 1) Let f : In → I∞. If In
f1 + yf2 where f1 : In−1 → I∞ and f2 : In+1 → I∞, hence we could proceed by induction.
y(cid:17)
(cid:16) 1
−−−→ In−1 ⊕ In+1 is the left almost split map, then f factors as
Furthermore each map g : I∞ → Im is given (up to a multiplicative unit) by multiplication
by yk, therefore it factors as I∞
yk
−→ Im+1 ⊂ Im, and proceed by induction.
2) Suppose that h belongs to the infinite radical. Recall that Im is generated by x and ym.
From Remark 8.2 it follows that f (xR) = 0. Since Im/xR ∼= I∞ we obtain the desired.
The converse is easy, say, because f factors through the infinitely generated module.
3) We may assume that a nonzero endomorphism f of I∞ is given by multiplication by yk.
(cid:3)
For m = x ∈ I∞ we have mx = 0 but f (m) 6= 0, hence f /∈ radω by Corollary 8.2.
Now we are in a position to prove the main result of this section.
Theorem 8.4. The nilpotency index of the radical of the category of finitely generated Cohen --
Macaulay R-modules equals ω + 2.
x−→ I∞ and let g be the inclusion
Proof. One half of this theorem is easy. Namely let f : R
I∞ ⊂ R. Then both maps belong to radω by Lemma 8.3. From gf 6= 0 we conclude that
radω+1 = (radω)2 6= 0.
To prove the converse it suffices to show that any composition M
h−→ D of maps in
radω between indecomposable finite generated CM-modules is zero. By Lemma 8.3 this is the
case when I∞ occurs among these modules at least twice.
f
−→ N
g
−→ K
Suppose that I∞ occurs exactly once. If M = I∞ then N = Im, K = In, hence the morphism
N → K factors through I∞.
f
I∞
/N
✼
u
✼
g
✼
✂
I∞
K
A✂
v
Because f ∈ radω, by Lemma 8.2 we conclude that uf = 0, hence the composite map is zero.
Furthermore the proof is similar when I∞ is positioned as N, K or D.
/
/
/
A
24
Finally if I∞ does not occur in the above sequence then we conclude from the following
diagram.
f
M
✿
✿
✆
I∞
N
B✆
✾
g
✾
✆
I∞
K
B✆
✾
D
B✆
h
✾
✆
I∞
Namely the composite morphism from the first to the last copy of I∞ factors through g, hence
(cid:3)
belongs to radω, therefore equals zero by Lemma 8.3.
9. Heigher dimensions
Recall that the A∞ singularity of dimension d ≥ 1 is defined (see [13, p. 253]) as the
hypersurface Rd = F [[x0, . . . , xd]]/(x2
d). A remarkable result, the so-called Knorrer's
periodicity (see [13, Sect. 8.3]), claims that the stable categories of finitely generated CM-
1 + · · · + x2
modules over these rings are equivalent, for even, and separately for odd d's.
Suppose that one would like to describe the CM-part of the Ziegler spectrum over these rings.
We will show that even for d = 2 this is a problem of enormous complexity. Namely after
changing variables we see that R2 is isomorphic to the ring F [[x, y, z]/(xz). Furthermore the
CM-condition means Hom(F, M ) = 0 and Ext1(F, M ) = 0.
We will consider a very particular case of this problem. Namely let T denote the closed subset
of ZCM over R2 consisting of modules annihilated by z. It is easily seen that those are exactly
the Cohen -- Macaulay modules over the 2-dimensional regular ring S = F [[x, y]]. Further SS is
the unique indecomposable finitely generated CM-module over S.
The following remark refutes a conjecture made in [10, Conj. 6.9].
Remark 9.1. Finitely generated points are not dense in the Cohen -- Macaulay part of the Ziegler
spectrum of F [[x, y]].
Proof. If E is the injective envelope of S/(x + y)S then this module is clearly CM. It opens the
interval [v = 0, v(x + y) = 0], but this interval is closed on S, hence E is not in the closure of
finitely generated points.
(cid:3)
Furthermore there is a little hope to classify all points of ZCM over S. Namely let P = (x3−y2)
be the prime ideal and let S′ be the localization SP . Since x, y /∈ P these elements are invertible
in S′, hence act as automorphisms on any S′-module; in particular every finite generated S′-
module is an (infinitely generated) CM-module over S. However S′ admits as a factor ring the
Drozd ring F [[x, y]]/(x2, xy2, y3), and this 5-dimensional algebra (see [12, p. 346]) is wild.
Thus one should put additional restrictions on non-finitely generated CM-modules over S to
classify them. One natural condition would be the Hochster-like M 6= M m. However we do not
know the answer to the following question.
Question 9.2. Does there exist a non-finitely generated indecomposable pure injective Cohen --
Macaulay module M over the ring F [[x, y]] such that M 6= M m?
/
/
/
/
/
/
B
B
B
25
It seems to be more natural to force finitely generated points to be dense, therefore consider
the closure, in the Ziegler spectrum, of finitely generated CM-modules.
Conjecture 9.3. Let Rd, d ≥ 1 be the d-dimensional A∞-singularity over an algebraically
closed field F and let TCM denote the theory of finitely generated Cohen -- Macaulay modules over
this ring. Then the Cantor -- Bendixson rank of TCM equals 2d, and the same is the value of
the Krull -- Gabriel dimension of the definable category generated by finitely generated Cohen --
Macaulay Rd-modules.
The intuition behind this conjecture is the following. When d increases, the stable part of
this category remains 'the same', but for the projective part of this category we should add an
extra 2 for each addition dimension.
References
[1] M. Auslander, Functors and morphisms determined by objects, pp. 1 -- 112 in: Representation Theory of
Algebras, Lecture Notes Pure Appl. Math., Vol. 37, 1978.
[2] M. Auslander, A survey on existence theorems for almost split sequences, pp. 81 -- 89 in: Representations of
Algebras, London Math. Soc. Lecture Notes, Vol. 116, Cambridge University Press, 1986.
[3] H. Bass, On the ubiquity of Gorenstein rings, Math. Z., 82 (1963), 8 -- 28.
[4] W. Brungs, J. Herzog, Cohen -- Macaulay Rings, revised edition, Cambridge Studies in Advanced Mathematics,
Vol. 39, Cambridge University Press, 2005.
[5] R.O. Buchweitz, G.M. Greuel, F.O. Schreyer, Cohen -- Macaulay modules over hypersurface singularities II,
Invent. Math., 88 (1987), 165 -- 182.
[6] I. Burban, W. Gnedin, Cohen -- Macaulay modules over some non-reduced curve singularities, arXiv:
1301.3305v1.
[7] D. Eisenbud, Commutative Algebra with a View Towards Algebraic Geometry, Graduate Texts in Mathe-
matics, Vol. 150, Springer, 1994.
[8] I. M. Gelfand, V. A. Ponomarev, Indecomposable representations of the Lorenz group, Russian Math. Surveys
23 (1968), 1 -- 58.
[9] I. Herzog, The Auslander -- Reiten translate, pp. 153 -- 165 in: Contemporary Mathematics, Vol. 130, 1992.
[10] I. Herzog, V. Puninskaya, The model theory of divisible modules over a domain, Fund. Applied Math., 2(2)
(1996), 563 -- 594.
[11] C.U. Jensen, H. Lenzing, Model-theoretic Algebra with Particular Emphasis on Fields, Rings, Modules,
Algebra, Logic and Applications, Vol. 2, Gordon and Breach, 1989.
[12] L. Klingler, L.S. Levy, Representation type of commutative rings I: Local wildness, Pacific J. Math., 200
(2001), 345 -- 386.
[13] G.J. Leuschke, R. Wiegandt, Cohen -- Macalay Representations, AMS Mathematicas Surveys and Monographs,
Vol. 181, 2012.
[14] M. Prest, Purity, Spectra and Localization, Encyclopedia of Mathematics and its Applications, Vol. 121,
Cambridge Univesity Press, 2009.
[15] M. Prest, G. Puninski, One-directed indecomposable pure injective modules over string algebras, Colloq.
Math., 101 (2004), 89 -- 112.
[16] G. Puninski, Serial Rings, Kluwer, 2001.
[17] C.M. Ringel, The indecomposable representation of the 2-dihedral groups, Math. Ann., 214 (1975), 19 -- 34.
[18] C.M. Ringel, Tame algebas (On algorithm of solving vector space problems. II), pp. 137 -- 287 in: Proc. Ottawa
Conference in Representation Theory, Lecture Notes in Math., Vol. 831, 1980.
26
[19] C.M. Ringel, Infinite length modules. Some examples as introduction, pp. 1 -- 73 in: Infinite Length Modules,
eds. H. Krause and C.M. Ringel, Birhauser, 2000.
[20] C.M. Ringel, The minimal representation-infinite algebras which are special biserial, pp. 501 -- 560 in: Repre-
sentations of algebras and related topics, EMS Ser. Congr. Rep., Eur. Math. Soc., Zurich, 2011.
[21] F.O. Schreyer, Finite and countable CM-representation type, pp. 9 -- 34 in: Singularities, representation of
algebras, and vector bundles, Lecture Notes in Math., Vol. 1273, Springer, 1987.
[22] J. Schroer, On the infinite radical of the module category, Proc. London Math. Soc., 81 (2000), 651 -- 674.
[23] Y. Yoshino, Cohen -- Macaulay Modules over Cohen -- Macaulay Rings, London Math. Soc. Lecture Note Series,
Vol. 146, Cambridge University Press, 1990.
[24] M. Ziegler, Model theory of modules, Ann. Pure Appl. Logic, 26 (1984), 149 -- 213.
(G. Puninski) Belarusian State University, Faculty of Mechanics and Mathematics, av. Neza-
lezhnosti 4, Minsk 220030, Belarus
E-mail address: [email protected]
|
1704.01322 | 1 | 1704 | 2017-04-05T09:27:46 | The structures on the universal enveloping algebras of differential graded Poisson Hopf algebras | [
"math.RA"
] | In this paper, the so-called differential graded (DG for short) Poisson Hopf algebra is introduced, which can be considered as a natural extension of Poisson Hopf algebras in the differential graded setting. The structures on the universal enveloping algebras of differential graded Poisson Hopf algebras are discussed. | math.RA | math |
THE STRUCTURES ON THE UNIVERSAL ENVELOPING ALGEBRAS OF
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
Abstract. In this paper, the so-called differential graded (DG for short) Poisson Hopf algebra is intro-
duced, which can be considered as a natural extension of Poisson Hopf algebras in the differential graded
setting. The structures on the universal enveloping algebras of differential graded Poisson Hopf algebras
are discussed.
1. Introduction
Poisson algebras appear naturally in Hamiltonian mechanics, and play a central role in the study of
Poisson geometry and quantum groups. With the development of Poisson algebras in the past decades,
many important generalizations have been obtained in both commutative and noncommutative settings:
Poisson PI algebras [10], graded Poisson algebras [3], double Poisson algebras [20], Quiver Poisson
algebras [22], noncommutative Leibniz-Poisson algebras [1], Left-right noncommutative Poisson alge-
bras [2] and differential graded Poisson algebras [8], etc. One of most interesting features in this area is
the Poisson universal enveloping algebra, which was first introduced by Oh [12] in order to describe the
category of Poisson modules. Since then, Poisson universal enveloping algebras have been studied in a
series of papers [15, 19, 21]. In particular, the third author and the fourth author of the present paper
studied the universal enveloping algebras of Poisson Ore-extensions and DG Poisson algebra [7, 8].
We know that, the notion of Poisson Hopf algebras originally arises from Poisson geometry and
quantum groups. For instance, the coordinate ring of a Poisson algebraic group is a Poisson Hopf
algebra [5]. Recently, Poisson Hopf algebras are studied by many authors from different perspectives
[6, 13, 14, 17]. In [13], Oh developed the theory of universal enveloping algebras for Poisson algebras,
and then proved that the universal enveloping algebra Ae of a Poisson Hopf algebra A is a Hopf algebra.
In addition, the last two authors of this paper studied the most basic properties for universal enveloping
algebras of Poisson Hopf algebras [6], which would help us to understand Hopf algebras in general.
Motivated by the notion of differential graded Poisson algebra, our aim in this paper is to study Pois-
son Hopf algebras and their universal enveloping algebras in the DG setting. Roughly speaking, a DG
Poisson Hopf algebra is a DG Poisson algebra together with a Hopf structure satisfying certain compati-
ble conditions; see Definition 3.1. As for universal enveloping algebras, there are many equivalent ways
to define the universal enveloping algebra of a DG Poisson algebra, among which we choose to use the
universal property; see Definition 2.3.
Note that the universal enveloping algebra Ae of a Poisson bialgebra A is a bialgebra, as an extension,
we prove the following result:
Theorem 1.1. If (A, u, η, ∆, ε, {·, ·}, d) is a DG Poisson bialgebra, then
2010 Mathematics Subject Classification. 16E45, 16S10, 17B35, 17B63.
Key words and phrases. differential graded Poisson algebras, differential graded Hopf algebras, differential graded Poisson
Hopf algebras, universal enveloping algebras.
This work was supported by National Natural Science Foundation of China (Grant No.11571316) and Natural Science Foun-
dation of Zhejiang Province (Grant No. LY16A010003).
*corresponding author.
1
2
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
is a DG bialgebra such that
(Ae, ue, ηe, ∆e, εe, de)
∆em = (m ⊗ m)∆
εem = ε
∆eh = (m ⊗ h + h ⊗ m)∆
εeh = 0.
It should be noted that the universal enveloping algebra Ae of a DG Poisson Hopf algebra A is just
a DG bialgebra. But in some special cases, Ae can be endowed with the Hopf structure, such that Ae
becomes a DG Hopf algebra:
Theorem 1.2. Let (A, u, η, ∆, ε, S , {·, ·}, d) be a DG Poisson Hopf algebra. Suppose that Aeop is the DG
opposite algebra of Ae. Then:
(1) There exist a DG algebra homomorphism S e : Ae → Aeop such that
S em = mS ,
S eh = hS .
(2) {S (a(1)), a(2)} = 0, where ∆(a) = a(1) ⊗ a(2) for all a ∈ A, if and only if
(Ae, ue, ηe, ∆e, εe, S e, de)
is a DG Hopf algebra.
The paper is organized as follows. In Section 2, we briefly review some basic concepts and results
related to DG algebras, DG Poisson algebras, DG Hopf algebras and universal enveloping algebras
of DG Poisson algebras. Section 3 is devoted to the study of some definitions and properties of DG
Poisson Hopf algebras and universal enveloping algebras. In particular, we show that there is a natural
DG bialgebra structure on the universal enveloping algebra Ae for a DG Poisson bialgebra A.
Throughout the whole paper, Z denotes the set of integers, k denotes a base field of characteristic zero
unless otherwise stated, all (graded) algebras are assumed to have an identity and all (graded) modules
are assumed to be unitary. We always take the grading to be Z-graded.
2. Preliminaries
In this section, we will recall some definitions and results of DG Poisson algebras, universal envelop-
ing algebras and DG Hopf algebras.
2.1. DG Poisson algebras. By a graded algebra A we mean a Z-graded algebra (A, u, η), where u :
A ⊗ A → A and η : k → A are called the multiplication and unit of A, respectively. For convenience, we
shall write u(a ⊗ b) as ab, ∀a, b ∈ A, whenever this does not cause confusion. A DG algebra is a graded
algebra with a k-linear homogeneous map d : A → A of degree 1, which is also a graded derivation. Let
A, B be two DG algebras and f : A → B be a graded algebra map of degree zero. Then f is called a DG
algebra map if f commutes with the differentials.
Definition 2.1. Let (A, ·) be a graded k-algebra. If there is a k-linear map
{·, ·} : A ⊗ A → A
of degree p such that
(i) (A, {·, ·}) is a graded Lie algebra. That is to say, we have
(ia) {a, b} = −(−1)(a+p)(b+p){b, a};
(ib) {a, {b, c}} = {{a, b}, c} + (−1)(a+p)(b+p){b, {a, c}},
(ii) (graded commutativity): a · b = (−1)abb · a;
(iii) (biderivation property): {a, b · c} = {a, b} · c + (−1)(a+p)bb · {a, c},
for any homogeneous elements a, b, c ∈ A, then A is called a graded Poisson algebra [3]. If in addition,
there is a k-linear homogeneous map d : A → A of degree 1 such that d2 = 0 and
(iv) d({a, b}) = {d(a), b} + (−1)(a+p){a, d(b)};
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
3
(v) d(a · b) = d(a) · b + (−1)aa · d(b),
for any homogeneous elements a, b ∈ A, then A is called a DG Poisson algebra, which is usually denoted
by (A, ·, {·, ·}, d), or simply by (A, {·, ·}, d) or A if no confusions arise.
Lemma 2.2. [8] Let (A, ·, {·, ·}A, dA) and (B, ∗, {·, ·}B, dB) be any two DG Poisson algebras with Poisson
brackets of degree p. Then (A ⊗ B, ⋆, {·, ·}, d) is a DG Poisson algebra, where
(a ⊗ b)⋆(a′ ⊗ b′) := (−1)a′b(a · a′) ⊗ (b ∗ b′),
d(a ⊗ b) := dA(a) ⊗ b + (−1)aa ⊗ dB(b),
{a ⊗ b, a′ ⊗ b′} := (−1)(a′+p)b{a, a′}A ⊗ (b ∗ b′) + (−1)(b+p)a′(a · a′) ⊗ {b, b′}B,
for any homogeneous elements a, a′ ∈ A and b, b′ ∈ B.
For DG Poisson algebras A and B, a graded algebra homomorphism φ : A → B is said to be a DG
Poisson algebra homomorphism if φ ◦ dA = dB ◦ φ and φ({a, b}A) = {φ(a), φ(b)}B for all homogeneous
elements a, b ∈ A. We denote by DG(P)A the category of DG (Poisson) algebras whose morphism
space consists of DG (Poisson) algebras homomorphism.
For a DG algebra B, assume throughout the paper that, BP will be the DG Poisson algebra B with
the standard graded Lie bracket [a, b] = ab − (−1)(a+p)(b+p)ba for all homogeneous elements a, b ∈ B,
where p is the degree of the Poisson bracket for B.
Let us review the definition of a universal enveloping algebra of A ∈ DGPA.
Definition 2.3. For a DG Poisson algebra A, a quadruple (Ae, α, β, ∂), where Ae ∈ DGA, α : (A, d) →
(Ae, ∂) is a DG algebra map and β : (A, {·, ·}, d) → (Ae
P, [·, ·], ∂) is a DG Lie algebra map such that
α({a, b}) = β(a)α(b) − (−1)(a+p)bα(b)β(a),
β(ab) = α(a)β(b) + (−1)abα(b)β(a),
for any homogeneous elements a, b ∈ A and p is the degree of the Poisson bracket for A, is called the
universal enveloping algebra of A if (Ae, α, β, ∂) satisfies the following: if (D, δ) is a DG algebra, f is a
DG algebra map from (A, d) into (D, δ) and g is a DG Lie algebra map from (A, {·, ·}, d) into (DP, [·, ·], δ)
such that
f ({a, b}) = g(a) f (b) − (−1)(a+p)b f (b)g(a),
g(ab) = f (a)g(b) + (−1)ab f (b)g(a),
for all a, b ∈ A, then there exists a unique DG algebra map φ : (Ae, ∂) → (D, δ), such that φα = f and
φβ = g.
For every DG Poisson algebra A, there exists a unique universal enveloping algebra Ae up to isomor-
phic, that Ae is generated by m(A) and h(A) by [8] and that h(1) = 0. Note that Ae is a DG algebra, such
that a k-vector space M is a DG Poisson A-module if and only if M is a DG Ae-module.
As an application of the "universal property" of universal enveloping algebras, we have
Lemma 2.4. [8] For any A, B ∈ DGPA, suppose that (Ae, mA, hA) and (Be, mB, hB) are the universal
enveloping algebra of A and B, respectively. Then (Ae ⊗ Be, mA ⊗ mB, mA ⊗ hB + (−1)pBhA ⊗ mB) is the
universal enveloping algebra for A ⊗ B, where p is the degree of the Poisson bracket for both A and B.
We use (−1)pBhA ⊗ mB to mean that
((−1)pBhA ⊗ mB)(a ⊗ b) = (−1)pbhA(a) ⊗ mB(b)
for any a ∈ A, b ∈ B.
Lemma 2.5. Let (Ae, mA, hA) and (Be, mB, hB) be universal enveloping algebras for DG Poisson alge-
bras A and B respectively. Assume that p is the degree of the Poisson bracket for both A and B. If
4
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
φ : A → B is a DG Poisson homomorphism, then there exists a unique DG algebra homomorphism
φe : Ae → Be such that φemA = mBφ and φehA = hBφ.
A
mA, hA
Ae
φ
φe
/ B
mB, hB
/ Be
Proof. Note that mB is a DG algebra map and hB is a DG Lie algebra map such that
mB({b, b′}) = hB(b)mB(b′) − (−1)(b+p)b′mB(b′)hB(b),
hB(bb′) = mB(b)hB(b′) + (−1)bb′mB(b′)hB(b).
mBφ(aa′) = mB(φ(a)φ(a′)) = mBφ(a)mBφ(a′),
hBφ({a, a′}) = hB({φ(a), φ(a′)}) = [hBφ(a), hBφ(a′)],
mBφ({a, a′}) = mB({φ(a), φ(a′)}) = hBφ(a)mBφ(a′) − (−1)(a+p)a′mBφ(a′)hBφ(a)
We have
and
hBφ(aa′) = hB(φ(a)φ(a′)) = mBφ(a)hBφ(a′) + (−1)aa′mBφ(a′)hBφ(a),
since φ : A → B is a DG Poisson homomorphism. Further, it is easy to see
mBφ(dA(a)) = mBdB(φ(a)) = dBemBφ(a)
and
hBφ(dA(a)) = hBdB(φ(a)) = dBehBφ(a).
Thus there exists a unique DG algebra homomorphism φe from Ae into Be such that φemA = mBφ and
φehA = hBφ.
(cid:3)
2.2. DG Hopf algebras. In this subsection, we recall some definitions and properties of graded Hopf
algebras and DG Hopf algebras.
In the remainder of the paper, the twisting map T : V ⊗ W → W ⊗ V will frequently appear. It is
defined for homogeneous elements v ∈ V and w ∈ W by
T (v ⊗ w) = (−1)vww ⊗ v
and extends to all elements of V and W through linearity.
A graded coalgebra C over k is a Z-graded vector space with the graded vector space homomorphisms
of degree 0 ∆ : C → C ⊗ C and ε : C → k such that the following diagrams
C
∆
∆
/ C ⊗ C
I⊗∆
C ⊗ C
∆⊗I
/ C ⊗ C ⊗ C
C ⊗ C
C ⊗ k
I⊗ ε
z✉✉✉✉✉✉✉✉✉
d■■■■■■■■■■
(cid:27)
C
∆
ε⊗ I
$■■■■■■■■■
:✉✉✉✉✉✉✉✉✉✉
(cid:27)
k ⊗ C
are commutative, where I : C → C is the identity homomorphism, ∆ and ε are called the comultipli-
cation and counit of C, respectively. Note that the commutativity of the first diagram is equivalent to
/
/
/
z
$
/
O
O
d
:
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
5
(∆ ⊗ I)∆ = (I ⊗ ∆)∆, called the coassociativity of ∆. Further, C is graded cocommutative if ∆ = T ∆,
where T : C ⊗ C → C ⊗ C is the twisting morphism.
For any c ∈ C, we write ∆(c) = c(1) ⊗ c(2), which is basically the Sweedler's notation with the
summation sign Σ omitted. In this notation, the comultiplication and counit property may be expressed
as
(∆ ⊗ I)∆(c) = (I ⊗ ∆)∆(c) = c(1) ⊗ c(2) ⊗ c(3),
c = ε(c(1))c(2) = c(1)ε(c(2)),
for any homogeneous elements c ∈ C, respectively.
A homomorphism of graded coalgebras f : A → B is a homomorphism of graded vector spaces such
that the diagrams
A
f
B
∆A
/ A ⊗ A
f ⊗ f
k
∆B
/ B ⊗ B
εA
^❃❃❃❃❃❃❃
εB
f
A
B
are commutative.
Let H be a graded algebra with multiplication u and unit η, and at the same time a graded coalgebra
with comultiplication ∆ and counit ε. Then H is called a graded bialgebra provided that one of the
following equivalent conditions are satisfied:
(i) ∆ and ε are graded algebra homomorphisms;
(ii) u and η are graded coalgebra homomorphisms.
Further, if H admits a graded vector space homomorphism S : H → H of degree 0, which satisfies the
following defining relation
u(I ⊗ S )∆ = u(S ⊗ I)∆ = ηε,
then H is called a graded Hopf algebra, and S is called the antipode of H.
The antipode S has the following properties [4, 9, 16].
Lemma 2.6. Let H be a graded Hopf algebra and S its antipode; then
(1) S · u = u · T · (S ⊗ S ),
(2) S · η = η,
(3) ε · S = ε,
(4) T · (S ⊗ S ) · ∆ = ∆ · S ,
(5) if H is graded commutative or graded cocommutative, then S · S = I,
where I : H → H is the identity morphism and T : H ⊗ H → H ⊗ H is the twisting morphism.
Definition 2.7. [18] Let H be a graded k-vector space. If there is a k-linear homogeneous map d : H →
H of degree 1 such that d2 = 0 and
(i) (H, u, η, d) is a DG algebra. That is to say, we have
(ia) (H, u, η) is a graded algebra;
(ib) d is a (algebra) derivation of degree 1, that means,
d(ab) = d(a)b + (−1)aad(b),
for any homogeneous elements a, b ∈ H.
(ii) (H, ∆, ε, d) is a DG coalgebra. That is, we have
(iia) (H, ∆, ε) is a graded coalgebra;
/
/
^
6
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
(iib) d is a coderivation of degree 1, that means, εd = 0 and
∆d = (d ⊗ I + T (d ⊗ I)T )∆.
(iii) (H, u, η, ∆, ε) is a graded bialgebra.
Then H is called a DG bialgebra. Moreover, if (H, u, η, ∆, ε, S ) is a graded Hopf algebra, then H is
called a DG Hopf algebra, which is usually denoted by (H, u, η, ∆, ε, S , d).
3. DG Poisson Hopf algebras and universal enveloping algebras
Now we will introduce the notion of DG Poisson Hopf algebras and prove that the universal envelop-
ing algebra Ae of a DG Poisson bialgebra A is a DG bialgebra.
Definition 3.1. Let A be a graded k-vector space. If there is a k-linear map
{·, ·} : A ⊗ A → A
of degree p such that:
(i) (A, u, η, {·, ·}) is a graded Poisson algebra;
(ii) (A, u, η, ∆, ε) is a graded bialgebra;
(iii) ∆({a, b}A) = {∆(a), ∆(b)}A⊗A for all a, b ∈ A, where the Poisson bracket {·, ·}A⊗A on A ⊗ A is
defined by
{a ⊗ a′, b ⊗ b′}A⊗A = (−1)(b+p)a′{a, b} ⊗ a′b′ + (−1)(a′+p)bab ⊗ {a′, b′}
for any homogeneous elements a, b, a′, b′ ∈ A.
Then A is called a graded Poisson bialgebra.
d : A → A of degree 1 such that d2 = 0 and
If in addition, there is a k-linear homogeneous map
(iva) d({a, b}) = {d(a), b} + (−1)(a+p){a, d(b)};
(ivb) d(a · b) = d(a) · b + (−1)aa · d(b);
(ivc) εd = 0 and ∆d(a) = d(a(1)) ⊗ a(2) + (−1)a(1)a(1) ⊗ d(a(2)), where ∆(a) = a(1) ⊗ a(2),
for any homogeneous elements a, b ∈ A, then (A, u, η, ∆, ε, {·, ·}, d) is called a DG Poisson bialgebra.
Further, if A admits a graded vector space homomorphism S : A → A of degree 0 which satisfies the
following defining relation
Then A is called a DG Poisson Hopf algebra, which is usually denoted by
u(I ⊗ S )∆ = u(S ⊗ I)∆ = ηε.
(A, u, η, ∆, ε, S , {·, ·}, d).
From now on, we would like to use the degree of the Poisson bracket is zero in our definition of DG
Poisson Hopf algebra, but the result obtained in this paper are also true for DG Poisson Hopf algebra of
degree p with some expected signs, where p ∈ Z is the degree of the Poisson bracket.
As a generalization of Lemma 2.2, we have
Proposition 3.2. Let (A, uA, ηA, ∆A, εA, S A, {·, ·}A, dA) and (B, uB, ηB, ∆B, εB, S B, {·, ·}B, dB) be any two
DG Poisson Hopf algebras with Poisson brackets of degree 0. Then
is a DG Poisson Hopf algebra, where
(A ⊗ B, u, η, ∆, ε, S , {·, ·}, d)
u((a ⊗ b) ⊗ (a′ ⊗ b′)) := (−1)a′baa′ ⊗ bb′,
∆(a ⊗ b) := (−1)a(2)b(1)a(1) ⊗ b(1) ⊗ a(2) ⊗ b(2),
d(a ⊗ b) := dA(a) ⊗ b + (−1)aa ⊗ dB(b),
{a ⊗ b, a′ ⊗ b′} := (−1)a′b({a, a′}A ⊗ bb′ + aa′ ⊗ {b, b′}B),
S (a ⊗ b) := S A(a) ⊗ S B(b),
η(1k) := 1A ⊗ 1B,
ǫ(a ⊗ b) := ǫA(a)ǫB(b),
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
7
for any homogeneous elements a, a′ ∈ A and b, b′ ∈ B. As usual, we write ∆(a) = a(1) ⊗ a(2) and
∆(b) = b(1) ⊗ b(2).
Proof. It is verified routinely that (A ⊗ B, u, η, ∆, ε, S , {·, ·}, d) is a DG Poisson Hopf algebra.
(cid:3)
Now we give some examples of DG Poisson Hopf algebras.
Example 3.3. Let (A, u, η, ∆, ε, S , {·, ·}, d) be any DG Poisson Hopf algebra with Poisson brackets of
degree 0. Then
(Aop, uop, η, ∆op, ε, S , {·, ·}op, d)
is also a DG Poisson Hopf algebra, where
uop(a ⊗ b) = (−1)abb · a = a · b = u(a ⊗ b),
∆op = T ∆,
{a, b}op = (−1)ab{b, a} = −{a, b},
for any homogeneous elements a, b ∈ A, and T : A ⊗ A → A ⊗ A is the twisting morphism.
From the constructions of opposite algebra and tensor product of DG Poisson Hopf algebras, we have
the following observation:
Proposition 3.4. Let A and B be any two DG Poisson Hopf algebras. Then
(A ⊗ B)op = Aop ⊗ Bop.
In [6], there are a lot of examples about Poisson Hopf algebras, which can be considered as DG
Poisson Hopf algebras concentrated in degree 0 with trivial differential. For instance, we have the
following example.
Example 3.5. Assume that the base field k has characteristic p > 2. Let
B = k[x, y, z]/(xp, yp, zp)
be the restricted symmetric algebra with three variables. Then B becomes a Poisson Hopf algebra via
∆(x) = x ⊗ 1 + 1 ⊗ x,
∆(y) = y ⊗ 1 + 1 ⊗ y,
∆(z) = z ⊗ 1 + 1 ⊗ z − 2x ⊗ y,
S (x) = −x,
S (y) = −y,
S (z) = −z − 2xy,
{x, y} = y,
{y, z} = y2,
{x, z} = z.
In the above example, one should view it as a Poisson version of the Hopf algebra in [11]. Similarly,
we can obtain other Poisson Hopf algebras from connected Hopf algebras in [11].
Example 3.6. Let (L, dL) be a finite dimensional DG Lie algebra over k with standard graded Lie bracket
[·, ·] and let (S (L), u, η) be the graded symmetric algebra of L, i.e.,
S (L) :=
T (L)
(a ⊗ b − (−1)abb ⊗ a)
,
for any a, b ∈ L, where u is a multiplication and η is a unit. The differential dL of L can be extended
to the graded symmetric algebra S (L) such that S (L) becomes a graded commutative DG algebra and
denote by d the differential in S (L). Here, we consider the total grading on S (L) coming from the
grading of L. Moreover, the Lie bracket on L also can be extended to a Poisson bracket on S (L) by
graded Jacobi identity such that
{a, b}S (L) := [a, b],
8
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
for any a, b ∈ L. Hence, the graded symmetric algebra S (L) over L has a natural DG Poisson algebra
structure.
Now, we introduce the graded Hopf structure for S (L). Denote by ∆ the familiar comultiplication
∆ : S (L) → S (L) ⊗ S (L) such that
∆(a) = a ⊗ 1 + 1 ⊗ a, ∀a ∈ L,
∆(uv) = ∆(u)∆(v),
∀u, v ∈ S (L)
and let the morphism ε : S (L) → k be defined by
ε(1) = 1,
ε(a) = 0,
∀a ∈ L,
ε(uv) = ε(u)ε(v),
∀u, v ∈ S (L),
then S (L) constitutes a Z-graded bialgebra with comultiplication ∆ and counit ε. Fix a k-basis x1, x2, ..., xn
of L. Note that S (L) is the graded commutative polynomial ring k[x1, x2, ..., xn], we can prove the for-
mula ∆({a, b}) = {∆(a), ∆(b)} by using the induction on degree of homogeneous elements of S (L). In
fact, we have
∆({xi, x j}S (L))
=∆(xi x j − (−1)xix j x j xi)
=∆(xi)∆(x j) − (−1)xix j∆(x j)∆(xi)
=(xi ⊗ 1 + 1 ⊗ xi)(x j ⊗ 1 + 1 ⊗ x j) − (−1)xix j(x j ⊗ 1 + 1 ⊗ x j)(xi ⊗ 1 + 1 ⊗ xi)
=xi x j ⊗ 1 + xi ⊗ x j + (−1)xix j(x j ⊗ xi) + 1 ⊗ xi x j
− (−1)xix j(x j xi ⊗ 1 + x j ⊗ xi + (−1)xix j(xi ⊗ x j) + 1 ⊗ x j xi)
=xi x j ⊗ 1 + 1 ⊗ xi x j − (−1)xix j(x jxi ⊗ 1) − (−1)xix j(1 ⊗ x j xi)
and
{∆(xi), ∆(x j)}
={xi ⊗ 1 + 1 ⊗ xi, x j ⊗ 1 + 1 ⊗ x j}
={xi, x j} ⊗ 1 + 1 ⊗ {xi, x j}
=xi x j ⊗ 1 − (−1)xix j(x j xi ⊗ 1) + 1 ⊗ xi x j − (−1)xix j(1 ⊗ x j xi),
and so we have ∆({xi, x j}) = {∆(xi), ∆(x j)}. For any monomials a, b, c ∈ S (L), assume that ∆({a, c}) =
{∆(a), ∆(c)} and ∆({b, c}) = {∆(b), ∆(c)}. Now we have
∆({ab, c}) = ∆(a{b, c} + (−1)abb{a, c})
= ∆(a)∆({b, c}) + (−1)ab∆(b)∆({a, c})
= ∆(a){∆(b), ∆(c)} + (−1)ab∆(b){∆(a), ∆(c)}
= {∆(ab), ∆(c)}.
Applying the induction and biderivation property, we have
∆({a, b}) = {∆(a), ∆(b)},
for all a, b ∈ S (L). In fact, it is also verified easily using the induction on degree of homogeneous
elements of S (L) that ∆d(a) = d(a(1)) ⊗ a(2) + (−1)a(1)a(1) ⊗ d(a(2)), εd(a) = 0 and ε({a, b}) = 0 for all
a, b ∈ S (L), where ∆(a) = a(1) ⊗ a(2). Hence (S (L), u, η, ∆, ε, {·, ·}, d) is a DG Poisson bialgebra. In order
to finish the proof, it suffice to prove that
u(S ⊗ I)∆ = u(I ⊗ S )∆ = ηε,
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
9
where the k-linear homogeneous map S : S (L) → S (L) of degree 0 is defined by
S (1) = 1,
S (a) = −a,
∀a ∈ L,
S (uv) = (−1)uvS (v)S (u)
for homogeneous elements of S (L) which extends to all S (L) by linearity. Then we have
u(S ⊗ I)∆(xi) = u(S ⊗ I)(xi ⊗ 1 + 1 ⊗ xi) = S (xi) + xi = −xi + xi = 0 = ηε(xi)
and
u(I ⊗ S )∆(xi) = u(I ⊗ S )(xi ⊗ 1 + 1 ⊗ xi) = xi + S (xi) = xi − xi = 0 = ηε(xi).
Now it suffices to prove the formula is true for ab provided that it is true for any homogeneous elements
a, b ∈ S (L). Thus, assume that we have the following two equations:
for any homogeneous elements a, b ∈ S (L). Then we have
u(S ⊗ I)∆(a) = ηε(a) and u(S ⊗ I)∆(b) = ηε(b),
u(S ⊗ I)∆(ab)
=u(S ⊗ I)(∆(a)∆(b)) = u(S ⊗ I)((−1)a(2)b(1)a(1)b(1) ⊗ a(2)b(2))
=(−1)a(2)b(1)S (a(1)b(1))a(2)b(2) = (−1)a(2)b(1)+a(1)b(1)S (b(1))S (a(1))a(2)b(2)
=S (a(1))a(2)S (b(1))b(2) = ηε(a)ηε(b) = ηε(ab).
Similarly, we can prove the formula u(I ⊗ S )∆ = ηε is true, which complete the proof.
Lemma 3.7. Retain the notations of Example 3.6, we have that
(3.1)
{S (a(1)), a(2)} = 0,
where ∆(a) = a(1) ⊗ a(2) for all a ∈ S (L).
Proof. We proceed the proof using induction on the length of a ∈ S (L). Observe that formula (1) is true
on any a ∈ L since
{S (a(1)), a(2)} = {S (a), 1} + {S (1), a} = {−a, 1} + {1, a} = 0,
where a ∈ L and ∆(a) = a ⊗ 1 + 1 ⊗ a.
In order to prove that formula (1) is true for all a ∈ S (L). It suffices to prove the formula (1) is true
for ab provided that it is true for any elements a, b ∈ S (L). Thus, assume that we have the following two
equations:
{S (a(1)), a(2)} = 0,
{S (b(1)), b(2)} = 0,
for any elements a, b ∈ S (L). Note that
∆(ab) = ∆(a)∆(b) = (a(1) ⊗ a(2))(b(1) ⊗ b(2)) = (−1)a(2)b(1)a(1)b(1) ⊗ a(2)b(2).
10
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
We have
{S ((ab)(1)), (ab)(2)}
=(−1)a(2)b(1){S (a(1)b(1)), a(2)b(2)}
=(−1)a(2)b(1){(−1)a(1)b(1)S (b(1))S (a(1)), a(2)b(2)}
=(−1)a(1)a(2)b(1)[{S (b(1))S (a(1)), a(2)}b(2) + (−1)a(1)b(1)a(2)a(2){S (b(1))S (a(1)), b(2)}]
=(−1)a(1)a(2)b(1)[S (b(1)){S (a(1)), a(2)}b(2) + (−1)a(1)a(2){S (b(1)), a(2)}S (a(1))b(2)]
+ (−1)a(1)a(2)b(1)[a(2)S (b(1)){S (a(1)), b(2)} + (−1)a(1)b(2)a(2){S (b(1)), b(2)}S (a(1))]
=(−1)a(1)a(2)b(1)[(−1)a(2)b(1){S (b(1)), a(2)}S (a(1))b(2) + a(2)S (b(1)){S (a(1)), b(2)}].
From the structure of DG Poisson Hopf algebra S (L), we have
0 = {ε(a)1A, b} = {S (a(1))a(2), b} = S (a(1)){a(2), b} + (−1)ba(2){S (a(1)), b}a(2)
and
0 = {a, ε(b)1A} = {a, S (b(1))b(2)} = {a, S (b(1))}b(2) + (−1)ab(1)S (b(1)){a, b(2)},
which imply that
and
Thus
S (a(1)){a(2), b} = −(−1)ba(2){S (a(1)), b}a(2)
{a, S (b(1))}b(2) = −(−1)ab(1)S (b(1)){a, b(2)}.
(−1)a(1)a(2)b(1)[(−1)a(2)b(1){S (b(1)), a(2)}S (a(1))b(2) + a(2)S (b(1)){S (a(1)), b(2)}] = 0
by using the above two formulas. Hence we have the conclusion.
(cid:3)
Lemma 3.8. If (A, u, η, ∆, ε, S , {·, ·}, d) is a DG Poisson Hopf algebra, then dS = S d, S ({a, b}) =
(−1)ab{S (b), S (a)} and the counit ε is a DG Poisson homomorphism.
Proof. Firstly, since {a⊗a′, b⊗b′}A⊗A = (−1)a′b({a, b}⊗a′b′+ab⊗{a′, b′}) and ∆({a, b}) = {∆(a), ∆(b)}A⊗A,
for all a, b, a′, b′ ∈ A, we have
{a, b} = u(ε ⊗ I)∆({a, b})
= u(ε ⊗ I)((−1)b(1)a(2)({a(1), b(1)} ⊗ a(2)b(2) + a(1)b(1) ⊗ {a(2), b(2)}))
= (−1)b(1)a(2)(ε({a(1), b(1)})a(2)b(2) + ε(a(1)b(1)){a(2), b(2)})
= (−1)b(1)a(2)(ε({a(1), b(1)})a(2)b(2) + {a, b}),
for all a, b ∈ A, and so we have
ε({a, b}) = (−1)b(1)a(2)ε(ε({a(1), b(1)})a(2)b(2) + {a, b})
= (−1)b(1)a(2)(ε({a(1), b(1)})ε(a(2)b(2)) + ε({a, b}))
= (−1)b(1)a(2)(ε({a, b}) + ε({a, b})),
hence ε({a, b}) = 0. Note that εd = 0, thus the counit ε is a DG Poisson homomorphism.
Secondly, since
0 = {ε(a)1A, b} = {S (a(1))a(2), b} = S (a(1)){a(2), b} + (−1)ba(2){S (a(1)), b}a(2)
and
0 = {a, ε(b)1A} = {a, S (b(1))b(2)} = {a, S (b(1))}b(2) + (−1)ab(1)S (b(1)){a, b(2)},
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
11
we have
and
Note that
S (a(1)){a(2), b} = −(−1)ba(2){S (a(1)), b}a(2)
{a, S (b(1))}b(2) = −(−1)ab(1)S (b(1)){a, b(2)}.
∆({a, b}) = {∆(a), ∆(b)} = (−1)b(1)a(2)({a(1), b(1)} ⊗ a(2)b(2) + a(1)b(1) ⊗ {a(2), b(2)}),
since A is a DG Poisson Hopf algebra. We have
0 = ε({a, b})1A = S ({a, b}(1)){a, b}(2) = (−1)b(1)a(2)(S ({a(1), b(1)})a(2)b(2) + S (a(1)b(1)){a(2), b(2)}).
Thus
S ({a(1), b(1)})a(2)b(2)
= − S (a(1)b(1)){a(2), b(2)} = −(−1)a(1)b(1)S (b(1))S (a(1)){a(2), b(2)}
=(−1)a(1)b(1)+b(2)a(2)S (b(1)){S (a(1)), b(2)}a(2) = −(−1)b(2)a(2){S (a(1)), S (b(1))}b(2)a(2)
= − {S (a(1)), S (b(1))}a(2)b(2).
Therefore
S ({a, b})
=S ({a(1)ε(a(2)), b(1)ε(b(2))}) = S ({a(1), b(1)})ε(a(2))ε(b(2))
=S ({a(1), b(1)})a(2)S (a(3))b(2)S (b(3)) = −(−1)b(2)a(3){S (a(1)), S (b(1))}a(2)b(2)S (a(3))S (b(3))
= − {S (a(1)), S (b(1))}a(2)S (a(3))b(2)S (b(3)) = −{S (a(1)), S (b(1))}ε(a(2))ε(b(2))
= − {S (a(1)ε(a(2))), S (b(1)ε(b(2)))} = −{S (a), S (b)} = (−1)ab{S (b), S (a)}.
Finally, it only remains to show that dS = S d. Note that
dS (a) =dS (a(1)ε(a(2))) = d(S (a(1))ε(a(2)))
=d(S (a(1))a(2)S (a(3))) = dS (a(1))a(2)S (a(3)) + (−1)a(1)S (a(1))d(a(2)S (a(3)))
=dS (a(1))a(2)S (a(3)) + (−1)a(1)S (a(1))(d(a(2))S (a(3)) + (−1)a(2)a(2)dS (a(3)))
=dS (a) + (−1)a(1)S (a(1))d(a(2))S (a(3)) + (−1)a(1)a(2)dS (a),
we get
Observe that
dS (a) = −(−1)a(2)S (a(1))d(a(2))S (a(3)).
∆d(a) = d(a(1)) ⊗ a(2) + (−1)a(1)a(1) ⊗ d(a(2)),
we can see
thus
Therefore,
0 = εd(a)1A = ηεd(a) = d(a)(1)S (d(a)(2)) = d(a(1))S (a(2)) + (−1)a(1)a(1)S d(a(2)),
d(a(1))S (a(2)) = −(−1)a(1)a(1)S d(a(2)).
dS (a) = −(−1)a(2)S (a(1))d(a(2))S (a(3)) = S (a(1))a(2)S d(a(3)) = S d(a),
12
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
as claimed.
(cid:3)
Lemma 3.9. If (A, u, η, ∆, ε, S , {·, ·}, d) is a DG Poisson Hopf algebra, then for all a ∈ A, {a(1), S (a(2))} =
0 if and only if {S (a(1)), a(2)} = 0, where ∆(a) = a(1) ⊗ a(2).
Proof. By the definition of a DG Poisson Hopf algebra, A is graded commutative graded Hopf algebra,
then we have S · S = I by Lemma 2.6. Note that
{a(1), S (a(2))} = {S · S (a(1)), S (a(2))} = (−1)a(1)a(2)S ({a(2), S (a(1))}) = −S ({S (a(1)), a(2)})
by Lemma 3.8. It is easy to see sufficiency is true. Similarly, we can prove necessity by the same way.
(cid:3)
Therefore, we are done.
Let us consider the bialgebra structure of universal enveloping algebras for DG Poisson bialge-
bras(note: the degree of Poisson bracket is zero). Let (A, ·, {·, ·}, d) be a DG Poisson algebra. Define
a k-linear map {·, ·}1 of degree 0 on A by
{a, b}1 = (−1)ab{b, a} = −{a, b},
for all a, b ∈ A. Then A1 = (A, ·, {·, ·}1, d) is a DG Poisson algebra by Example 3.3. For a DG algebra
(B, ·, d), we denote by Bop = (B, ◦, d) the DG opposite algebra of B, where a ◦ b = (−1)abb · a = a · b.
Proposition 3.10. [8] Let (Ae, m, h) be the universal enveloping algebra for a DG Poisson algebra
(A, ·, {·, ·}, d). Then (Aeop, m, h) is the universal enveloping algebra for A1 = (A, ·, {·, ·}1, d).
Now we are ready to prove Theorem 1.1.
Proof of Theorem 1.1. Since ∆ is a DG Poisson homomorphism and (Ae ⊗ Ae, m ⊗ m, m ⊗ h + h ⊗ m)
is the universal enveloping algebra of a DG Poisson algebra A ⊗ A (see Lemmas 2.2 and 2.4), there
exists a unique DG algebra homomorphism ∆e : Ae → Ae ⊗ Ae such that ∆em = (m ⊗ m)∆ and ∆eh =
(m ⊗ h + h ⊗ m)∆ by Lemma 2.5. Similarly, there exists a DG algebra homomorphism εe from Ae into
k such that εem = ε and εeh = 0, since (k, Ik, 0) is the universal enveloping algebra of the field k with
trivial differential and trivial Poisson bracket.
Now we claim that (Ae, ue, ηe, ∆e, εe, de) is a DG bialgebra. Note that (Ae, ue, ηe, de) is a DG algebra.
Next, we show that (Ae, ∆e, εe, de) is a DG coalgebra. Since (A, ∆, ε, d) is a DG coalgebra, we have
a(11) ⊗ a(12) ⊗ a(2) = a(1) ⊗ a(21) ⊗ a(22), ε(a(1))a(2) = a = a(1)ε(a(2)), εd = 0 and ∆d(a) = d(a(1)) ⊗ a(2) +
(−1)a(1)a(1) ⊗ d(a(2)), for all a ∈ A. Thus the coassociativity is immediate from
(∆e ⊗ I)∆em(a) = (∆e ⊗ I)(m ⊗ m)∆(a) = ∆em(a(1)) ⊗ m(a(2))
= (m ⊗ m)∆(a(1)) ⊗ m(a(2)) = m(a(11)) ⊗ m(a(12)) ⊗ m(a(2)),
(I ⊗ ∆e)∆em(a) = (I ⊗ ∆e)(m ⊗ m)∆(a) = m(a(1)) ⊗ ∆em(a(2))
= m(a(1)) ⊗ (m ⊗ m)∆(a(2)) = m(a(1)) ⊗ m(a(21)) ⊗ m(a(22)),
(∆e ⊗ I)∆eh(a) =(∆e ⊗ I)(m ⊗ h + h ⊗ m)∆(a)
=∆em(a(1)) ⊗ h(a(2)) + ∆eh(a(1)) ⊗ m(a(2))
=(m ⊗ m)∆(a(1)) ⊗ h(a(2)) + (m ⊗ h + h ⊗ m)∆(a(1)) ⊗ m(a(2))
=m(a(11)) ⊗ m(a(12)) ⊗ h(a(2)) + m(a(11)) ⊗ h(a(12)) ⊗ m(a(2))
+ h(a(11)) ⊗ m(a(12)) ⊗ m(a(2))
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
13
and
(I ⊗ ∆e)∆eh(a) =(I ⊗ ∆e)(m ⊗ h + h ⊗ m)∆(a)
=m(a(1)) ⊗ ∆eh(a(2)) + h(a(1)) ⊗ ∆em(a(2))
=m(a(1)) ⊗ (m ⊗ h + h ⊗ m)∆(a(2)) + h(a(1)) ⊗ (m ⊗ m)∆(a(2))
=m(a(1)) ⊗ m(a(21)) ⊗ h(a(22)) + m(a(1)) ⊗ h(a(21)) ⊗ m(a(22))
+ h(a(1)) ⊗ m(a(21)) ⊗ m(a(22)).
Similarly, we can prove the counitary property of A is also true. In order to prove d is a coderivation
of degree 1, first we have εedem(a) = εemd(a) = εd(a) = 0 and εedeh(a) = εehd(a) = 0, and then we
should prove ∆ede = (de ⊗ I + T (de ⊗ I)T )∆e, which follows from
(de ⊗ I + T (de ⊗ I)T )∆em(a) = (de ⊗ I + T (de ⊗ I)T )(m ⊗ m)∆(a)
= dem(a(1)) ⊗ m(a(2)) + (−1)a(1)m(a(1)) ⊗ dem(a(2))
= md(a(1)) ⊗ m(a(2)) + (−1)a(1)m(a(1)) ⊗ md(a(2)),
∆edem(a) =(m ⊗ m)∆d(a)
=(m ⊗ m)(d(a(1)) ⊗ a(2) + (−1)a(1)a(1) ⊗ d(a(2)))
=md(a(1)) ⊗ m(a(2)) + (−1)a(1)m(a(1)) ⊗ md(a(2)),
∆edeh(a) =(m ⊗ h + h ⊗ m)∆d(a)
=(m ⊗ h + h ⊗ m)(d(a(1)) ⊗ a(2) + (−1)a(1)a(1) ⊗ d(a(2)))
=md(a(1)) ⊗ h(a(2)) + hd(a(1)) ⊗ m(a(2))
+ (−1)a(1)m(a(1)) ⊗ hd(a(2)) + (−1)a(1)h(a(1)) ⊗ md(a(2))
and
(de ⊗ I + T (de ⊗ I)T )∆eh(a) =(de ⊗ I + T (de ⊗ I)T )(m ⊗ h + h ⊗ m)∆(a)
=dem(a(1)) ⊗ h(a(2)) + deh(a(1)) ⊗ m(a(2))
+ (−1)a(1)m(a(1)) ⊗ deh(a(2)) + (−1)a(1)h(a(1)) ⊗ dem(a(2))
=md(a(1)) ⊗ h(a(2)) + hd(a(1)) ⊗ m(a(2))
+ (−1)a(1)m(a(1)) ⊗ hd(a(2)) + (−1)a(1)h(a(1)) ⊗ md(a(2)).
Hence (Ae, de, ∆e, εe) is a DG coalgebra. Note that ∆e and εe are DG algebra homomorphisms, which
means (Ae, ue, ηe, ∆e, εe) is a graded bialgebra. Therefore, we finish the proof.
(cid:3)
In general, Ae is just a DG bialgebra. But in some special cases, Ae can be endowed with the Hopf
structure, such that (Ae, ue, ηe, ∆e, εe, S e, de) becomes a DG Hopf algebra. Thus, we give the proof of
Theorem 1.2.
Proof of Theorem 1.2. Since the antipode S is a DG Poisson homomorphism from A into A1 by Lemmas
2.6 and 3.8, there is a DG algebra homomorphism S e : Ae → Aeop such that S em = mS and S eh = hS
by Lemma 2.5 and Proposition 3.10.
As for part (2), since (Ae, ue, ηe, ∆e, εe, de) is a DG bialgebra by Theorem 1.1, it suffices to show that
S e is an antipode for Ae if and only if {S (a(1)), a(2)} = 0, where ∆(a) = a(1) ⊗ a(2) for all a ∈ A. In fact,
we have
ηeεem(a) = ηeε(a) = ε(a)1Ae, ηeεeh(a) = 0,
14
MENTTIAN GUO, XIANGUO HU, JIAFENG L U∗, AND XINGTING WANG
ue(S e ⊗ I)∆em(a) = ue(S e ⊗ I)(m ⊗ m)∆(a)
= S em(a(1))m(a(2)) = mS (a(1))m(a(2))
= m(S (a(1))a(2)) = m(ηε(a)) = ε(a)1Ae
and
ue(S e ⊗ I)∆eh(a) =ue(S e ⊗ I)(m ⊗ h + h ⊗ m)∆(a) = S em(a(1))h(a(2)) + S eh(a(1))m(a(2))
=mS (a(1))h(a(2)) + hS (a(1))m(a(2)) = h(S (a(1))a(2)) + m({S (a1), a2})
=h(ηε(a)) + m({S (a1), a2}) = ε(a)h(1A) + m({S (a1), a2}) = m({S (a1), a2}),
which imply that ue(S e ⊗ I)∆e = ηeεe if and only if {S (a(1)), a(2)} = 0, since m is injective. Note that for
all a ∈ A, we have {a(1), S (a(2))} = 0 iff {S (a(1)), a(2)} = 0 by Lemma 3.9. Similarly, we can prove the
formula ue(I ⊗ S e)∆e = ηeεe if and only if {a(1), S (a(2))} = 0. Therefore, we finish the proof.
(cid:3)
It is not hard to show that the following two corollaries are also true. Namely,
Corollary 3.11. Let (A, uA, ηA, ∆A, εA, S A, {·, ·}A, dA) and (B, uB, ηB, ∆B, εB, S B, {·, ·}B, dB) be any two DG
Poisson Hopf algebras with Poisson brackets of degree 0. Then
(1) (A ⊗ B)e = Ae ⊗ Be is a DG bialgebra.
(2) If {S A(a(1)), a(2)} = 0 and {S B(b(1)), b(2)} = 0, where ∆(a) = a(1) ⊗ a(2), ∆(b) = b(1) ⊗ b(2) for all
a ∈ A and b ∈ B, then (A ⊗ B)e is a DG Hopf algebra.
Corollary 3.12. Let (A, u, η, ∆, ε, S , {·, ·}, d) be a DG Poisson Hopf algebra. Then
(1) (Aop)e = (Ae)op is a DG bialgebra.
(2) (Aop)e is a DG Hopf algebra if and only if {S (a(1)), a(2)} = 0 , where ∆(a) = a(1) ⊗ a(2) for all
a ∈ A.
Example 3.13. Retain the notations of Example 3.5. From Theorem 1.1, we can see that the universal
enveloping algebra Be is a DG bialgebra. But
{S (z1), z2} = {S (z), 1} + {S (1), z} + {S (−2x), y} = 2y , 0,
we can't conclude that Be is a DG Hopf algebra by Theorem 1.2.
Example 3.14. Let (L, dL) be a finite dimensional DG Lie algebra over k with standard graded Lie
bracket [·, ·] and let (S (L), u, η) be the graded symmetric algebra of L, i.e.,
S (L) :=
T (L)
(a ⊗ b − (−1)abb ⊗ a)
,
for any a, b ∈ L, where u is a multiplication and η is a unit. Fix a k-basis x1, x2, · · · , xn of L. Note
that S (L) is the graded commutative polynomial ring k[x1, x2, · · · , xn]. As in Example 3.6, we know
S (L) is a DG Poisson algebra and it also has a DG Hopf structure which is compatible with the Poisson
bracket. That is, S (L) is a DG Poisson Hopf algebra with structure
{a, b}S (L) := [a, b], ∆(a) = a ⊗ 1 + 1 ⊗ a, ε(a) = 0, S (a) = −a,
for all a, b ∈ L. Observe that the universal enveloping algebra (S (L)e, m, h) is the DG algebra generated
by x1, · · · , xn, y1, · · · , yn subject to the relation
xi x j = (−1)xix j x j xi,
yi x j = (−1)xix j x jyi + {xi, x j},
yiy j = (−1)xix jy jyi + ψ({xi, x j}),
DIFFERENTIAL GRADED POISSON HOPF ALGEBRAS
15
for all i, j = 1, · · · , n and, m and h are given by m(xi) = xi and h(xi) = yi, respectively, where ψ : S (L) →
S (L)hyi i = 1, · · · , ni is a k-linear map of degree 0 defined by ψ(xi) = yi for all i = 1, · · · , n. By Lemma
3.7 and Theorem 1.2, the universal enveloping algebra S (L)e is a DG Hopf algebra with Hopf structure
∆e(xi) = xi ⊗ 1 + 1 ⊗ xi
εe(xi) = 0
S e(xi) = −xi
∆e(yi) = yi ⊗ 1 + 1 ⊗ yi
εe(yi) = 0
S e(yi) = −yi,
for all i = 1, · · · , n.
References
1. J. M. Casas and T. Datuashvili, Noncommutative Leibniz-Poisson algebras, Comm. Algebra 34(7) (2006), 2507–2530.
2. J. M. Casas, T. Datuashvili and M. Ladra, Left-right noncommutative Poisson algebras, Cent. Eur. J. Math. 12(1) (2014),
57–78.
3. A. S. Cattaneo, D. Fiorenza and R. Longoni, Graded Poisson algebras, Encyclopedia Math. Phy. (2006), 560–567.
4. M. D. Gould, R. B. Zhang and A. J. Bracken, Quantum double construction for graded Hopf algebras, J. Bull. Aus-
tral. Math. Soc. 47 (1993), 353–375.
5. L. I. Korogodski and Y. S. Soibelman, Algebras of functions on quantum groups, Part I, Mathematical surveys and mono-
graphs, vol. 56, Amer. Math. Soc. Providence, 1998.
6. J.-F. L u, X.-T. Wang and G.-B. Zhuang, Universal enveloping algebras of Poisson Hopf algebras, J. Algebra 426 (2015),
92–136.
7. J.-F. L u, X.-T. Wang and G.-B. Zhuang, Universal enveloping algebras of Poisson Ore-extensions, Proc. Amer. Math. Soc.
143 (2015), 4633–4645.
8. J.-F. L u, X.-T. Wang and G.-B. Zhuang, Universal enveloping algebras of differential graded Poisson algebras,
Sci. China Math. 59 (2016), 849–860.
9. J. M. Milner and J. C. More, On the structure of Hopf algebras, Ann. Math. 81 (1965), 211–264.
10. S. P. Mishchenko, V. M. Petrogradsky and A. Regev, Poisson PI algebras, Trans. Amer. Math. Soc. 359(10) (2007), 4669–
4694.
11. V. Nguyen, L. Wang and X. Wang, Classification of connected Hopf algebras of dimension p3 I, J. Algebra 424 (2015),
473–505.
12. S.-Q. Oh, Poisson enveloping algebras, Comm. Algebra 27 (1999), 2181–2186.
13. S.-Q. Oh, Hopf structure for Poisson enveloping algebras, Beitrage Zur Algebra Und Geometrie. 44(44) (2003), 567–574.
14. S.-Q. Oh, Poisson Hopf algebra related to a twisted quantum group, Comm. Algebra 45(1) (2016), 76–104.
15. S.-Q. Oh, C.-G Park and Y.-Y Shin, A Poincar´e-Birkhoff-Witt theorem for Poisson enveloping algebras, Comm. Algebra
30(10) (2002), 4867–4887.
16. M. E. Sweedler, Hopf algebras, Benjamin, New York, 1969.
17. T. Tanisaki Poisson Hopf algebras associated to quantized enveloping algebras, arXiv: 0802.1590.
18. R. N. Umble, The deformation complex for differential graded Hopf algebras, J. Pure. Appl. Algebra. 106(2) (1996), 199–222.
19. U. Umirbaev, Universal enveloping algebras and universal derivations of Poisson algebras, J. Algebra 354 (2012), 77–94.
20. M. Van den Berger, Double Poisson algebras, Trans. Amer. Math. Soc. 360(11) (2008), 5711–5769.
21. Y.-H. Yang, Y. Yao and Y. Ye, (Quasi-)Poisson enveloping algebras, Acta Math. Sin. (Engl. Ser.) 29(1) (2013), 105–118.
22. Y. Yao, Y. Ye and P. Zhang, Quiver Poisson algebras, J. Algebra 312(2) (2007), 570–589.
Guo: Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang, 321004 P.R. China
E-mail address: [email protected]
Hu: Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang, 321004 P.R. China
E-mail address: [email protected]
L u: Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang 321004, P.R. China
E-mail address: [email protected]
Wang: Department of Mathematics, Temple University, Philadelphia 19122, USA
E-mail address: [email protected]
|
1704.06111 | 3 | 1704 | 2018-01-25T13:42:30 | Modules over axial algebras | [
"math.RA",
"math.GR"
] | We introduce axial representations and modules over axial algebras as new tools to study axial algebras. All known interesting examples of axial algebras fall into this setting, in particular the Griess algebra whose automorphism group is the Monster group. Our results become especially interesting for Matsuo algebras. We vitalize the connection between Matsuo algebras and 3-transposition groups by relating modules over Matsuo algebras with representations of 3-transposition groups. As a by-product, we define, given a Fischer space, a group that can fulfill the role of a universal 3-transposition group. | math.RA | math |
Modules over axial algebras
Tom De Medts∗
Michiel Van Couwenberghe†
July 13, 2021
Abstract
We introduce axial representations and modules over axial algebras as new tools
to study axial algebras. All known interesting examples of axial algebras fall into this
setting, in particular the Griess algebra whose automorphism group is the Monster
group. Our results become especially interesting for Matsuo algebras. We vital-
ize the connection between Matsuo algebras and 3-transposition groups by relating
modules over Matsuo algebras with representations of 3-transposition groups. As a
by-product, we define, given a Fischer space, a group that can fulfill the role of a
universal 3-transposition group.
Keywords. axial algebras, modules, axial representations, Matsuo algebras, 3-transposition
groups, Fischer spaces.
MSC2010. primary: 17A99, 20B25, 20F29, 20C05; secondary: 17B69, 20C34.
1
Introduction
In 1982, Robert Griess constructed a commutative non-associative algebra, called the
Griess algebra, in order to prove the existence of the Monster group [Gri82]. Igor Frenkel,
James Lepowsky and Arne Meurman observed that (a deformation of) this Griess algebra
can be retrieved as a weight-2 component of a vertex operator algebra (VOA) V ♮ [FLM88].
Richards Borcherds then obtained a proof of the moonshine conjectures by studying this
VOA [Bor86].
The weight-2 components of VOAs similar to V ♮ give rise to algebras with properties
like the ones for the Griess algebra. Alexander Ivanov defined Majorana algebras in an
attempt to axiomatize such algebras. The definition of an axial algebra was introduced
by Jonathan Hall, Sergey Shpectorov and Felix Rehren in 2015 by stripping away some
of the requirements in the definition of Majorana algebras in such a way that many
results concerning Majorana algebras still remain true.
∗[email protected]
†[email protected] [PhD Fellow of the Research Foundation - Flanders (Belgium)
(F.W.O.-Vlaanderen)]
1
Since axial algebras have only recently been defined, this new research subject is very
much unexplored and it is not even clear whether the definition of an axial algebra is
yet in its final stage.
In this paper, we provide some new tools to study axial algebras. First, we define
axial representations of a group as a generalization of Majorana representations which
were introduced in [IPS10]. This notion is perhaps more foundational than the definition
of an axial algebra itself, since it describes the connection between axial algebras and
groups, very much like the connection between the Griess algebra and the Monster group.
If an ax-
ial algebra A is an axial representation of a group G, each A-module gives rise to a
U(A)-module where U(A) is a certain central extension of G; see Theorem 3.9 below.
In this sense, our definition of modules over axial algebras reinforces the connection
between axial algebras and groups.
Second, we present a natural definition of modules over axial algebras.
Third, we investigate the theory of modules more thoroughly for Matsuo algebras
over Fischer spaces, an important class of examples of axial algebras. In this case, the
group U(A) will be a universal 3-transposition group related to the given Fischer space,
as we explain in Theorem 4.13. (The well known connection between 3-transposition
groups and Fischer spaces is due to Francis Buekenhout [Bue74].) This enables us to
construct a module for the Matsuo algebra out of every U(A)-module; see Theorem 5.1.
This correspondence between modules over Matsuo algebras and U(A)-modules is not
one-to-one; it turns out that 1-eigenvectors in the modules over the Matsuo algebra play
a special role (see Corollary 5.3) and almost always indicate the presence of a regular
module as submodule, and this is the content of Theorems 5.8 and 5.11.
Acknowledgment. We have benefited from fruitful discussions between Jonathan
Hall and the second author during a visit at Michigan State University. We thank the
anonymous referee of an earlier version of this paper for his suggestions, which improved
the exposition of the results.
2 Axial algebras
This section provides an introduction into the realm of axial algebras. We use the defi-
nition by J. Hall, F. Rehren and S. Shpectorov in [HRS15b]. This definition of an axial
algebra resembles a property satisfied by idempotents in associative and Jordan algebras,
namely the existence of a Peirce decomposition. The right multiplication operator of an
idempotent in a Jordan algebra is diagonalizable and the decomposition into eigenvectors
is compatible with the multiplication, in the sense that the multiplication of eigenvectors
is described by a fusion rule.
Definition 2.1. Let R be a commutative ring with identity. A fusion rule is a pair
(Φ, ⋆) such that Φ ⊆ R and ⋆ : Φ × Φ → 2Φ is a symmetric map.
Definition 2.2. Let R be a commutative ring with identity, A a commutative (not
necessarily associative) R-algebra and e ∈ A an idempotent. For each φ ∈ R we denote
2
⋆
1
0
1
{1}
0
∅
∅
{0}
α
{α}
{α}
α {α} {α} {1, 0}
Table 1: the Jordan fusion rule Φ(α)
the eigenspace of e with eigenvalue φ by
Ae
φ = {a ∈ A ae = φa}.
For each subset Λ ⊆ R, let
with the convention that Ae
axis) for the fusion rule (Φ, ⋆) if
∅ = {0}. We call e a (Φ, ⋆)-diagonalizable idempotent (or an
Ae
φ,
Ae
Λ =Mφ∈Λ
and
Ae
φ
A = Mφ∈Φ
Ae
φ⋆ψ
φ · Ae
ψ ⊆ Ae
for all φ, ψ ∈ Φ. This means that the product of a φ-eigenvector and a ψ-eigenvector is
a sum of χ-eigenvectors where χ runs through φ ⋆ ψ.
Definition 2.3. A (Φ, ⋆)-axial algebra is a pair (A, Ω) where:
(i) A is a commutative (not necessarily associative) R-algebra and,
(ii) Ω ⊂ A is a generating set of (Φ, ⋆)-axes for A.
We will often omit the set Ω in our notation.
Example 2.4. Idempotents of Jordan algebras are Φ(α)-diagonalizable idempotents for
α = 1
2 where Φ(α) denotes the Jordan fusion rule described by Table 1 [Jac68, p. 119].
The important connection between some axial algebras and groups arises from the
special case of a Z/2Z-graded fusion rule.
Definition 2.5.
(i) A fusion rule (Φ, ⋆) is called Z/2Z-graded if Φ can be partitioned
into two subsets Φ+ and Φ− such that
φ ⋆ ψ ⊆ Φ+ whenever φ, ψ ∈ Φ+,
φ ⋆ ψ ⊆ Φ+ whenever φ, ψ ∈ Φ−,
φ ⋆ ψ ⊆ Φ− whenever φ ∈ Φ+ and ψ ∈ Φ−.
3
⋆
1
0
1
4
1
1
∅
1
4
0
∅
0
1
4
1
4
1
4
1
4
1, 0
1
32
1
32
1
32
1
32
1
32
1
32
1
32
1
32
1, 0, 1
4
Table 2: fusion rule of the Griess algebra
(ii) Let (A, Ω) be a (Φ, ⋆)-axial algebra for some Z/2Z-graded fusion rule (Φ, ⋆). We
associate to each (Φ, ⋆)-axis e of A a Miyamoto involution τe ∈ Aut(A) defined by
linearly extending
aτe =( a if a ∈ Ae
−a if a ∈ Ae
Φ+
Φ−
,
.
Because of the Z/2Z-grading of the fusion rule, these maps define automorphisms
of A. Note that, at this point, we allow Ae
to be trivial and hence τe to be trivial.
However, when Ae
6= 0, these automorphisms are indeed involutions.
Φ−
Φ−
(iii) We call the subgroup hτe e ∈ Ωi ≤ Aut(A) the Miyamoto group of the axial
algebra (A, Ω), and we denote it by Miy(A, Ω).
(iv) We say that Ω ⊂ A is Miyamoto-closed when it is invariant under Miy(A, Ω).
Example 2.6.
(i) The Jordan fusion rule Φ(α) from Table 1 is Z/2Z-graded with
Φ+ = {1, 0} and Φ− = {α}.
(ii) The Griess algebra is a 196884-dimensional real axial algebra that satisfies the fu-
sion rule from Table 2 [Iva09, Lemma 8.5.1, p. 209]. This fusion rule is Z/2Z-graded
with Φ+ = {1, 0, 1
32 }. The Miyamoto involutions of the generating
set of axes of the Griess algebra are called Majorana involutions. They generate
the full automorphism group of the Griess algebra which is better known as the
Monster group; see [Iva09, Proposition 8.6.2, p. 210] and [Tit84, p. 497].
4 } and Φ− = { 1
Definition 2.7. Let k be a field. A Frobenius axial algebra is an axial k-algebra equipped
with a bilinear form
h·, ·i : A × A → k
such that hxa, bi = ha, xbi for all a, b, x ∈ A. We call this bilinear form the Frobenius
form.
It is easy to verify that if A is a Frobenius axial algebra A, then for each axis
ψ are perpendicular for distinct φ and ψ [HRS15a,
φ and Aa
a ∈ A, the eigenspaces Aa
Proposition 3].
4
It is already clear from Example 2.6(ii) that there exists an important connection
between axial algebras and groups. This connection comes from the following situation
which is a generalization of a Majorana representation defined in [IPS10].
Definition 2.8. Let G be a group generated by a set D of elements of order at most 2.
An axial representation of (G, D) is an isomorphism ξ : G → Miy(A, Ω), where
(i) (A, Ω) is a (Φ, ⋆)-axial algebra for a Z/2Z-graded fusion rule (Φ, ⋆),
(ii) Ω is Miyamoto-closed,
(iii) ξ(D) = {τe ∈ Aut(A) e ∈ Ω}.
We will often say that (A, Ω) is an axial representation of (G, D) and we will simply
identify G with Miy(A, Ω) and D with the set {τe e ∈ Ω}.
Proposition 2.9 ([Reh15, Lemma 2.4.1]). Let (A, Ω) be an axial representation of
(G, D). For each t ∈ Aut(A) and each (Φ, ⋆)-axis a ∈ A, at is again a (Φ, ⋆)-axis
and (τa)t = τat. In particular, (τx)τy = τxτy for all x, y ∈ Ω.
Remark 2.10.
(i) Of course, every axial algebra (A, Ω) with a Z/2Z-graded fusion
rule such that Ω is Miyamoto-closed, is an axial representation for some (G, D),
namely G = Miy(A, Ω) and D = {τe e ∈ Ω}. The more interesting question
is: which groups G admit an axial representation (for some generating set D of
elements of order at most 2)?
(ii) In Definition 2.8, we explicitly allow the possibility that D contains the trivial
element, or equivalently, that τe is trivial for some e ∈ Ω.
(iii) Let G be a group generated by a set D of elements of order at most 2 which is
invariant under conjugation. If ξ : G → Miy(A, Ω) is an isomorphism satisfying (i)
and (iii) of Definition 2.8, then Ω is not necessarily Miyamoto-closed, but we can
always replace Ω by the possibly larger set of axes
Ω′ := {e ∈ A an axis τe = τf for some f ∈ Ω}
without changing the Miyamoto group. It now follows from Proposition 2.9 that Ω′
is indeed Miyamoto-closed, and hence (A, Ω′) is an axial representation of (G, D).
If (A, Ω) is an axial representation for (G, D), then the map τ : Ω → D : e 7→ τe is
not necessarily a bijection. This motivates the following definition, which we have taken
from [HSS17, Definition 6.8].
Definition 2.11. Let (A, Ω) be an axial representation of (G, D). We call the axial
representation of unique type if the map τ : Ω → D : e 7→ τe is a bijection.
According to [HSS17, Theorem 6.10], "most" axial representations that satisfy the
Jordan fusion rule are of unique type; in particular, those arising from 3-transposition
groups fall in this class (see Proposition 4.10(iii) below). Axial representations of unique
type behave nicer than others; see, for instance, Proposition 3.7 below.
5
Example 2.12. Not every axial representation is of unique type. Indeed, consider the
3-dimensional algebra spanned by vectors 1, u and v and commutative product defined
by u2 = v2 = 1, uv = 0 and such that 1 is the identity. This is a Jordan algebra of
2 + 1
Clifford type generated by the idempotents e1 = 1
2 v, 1 − e1 and 1 − e2.
For the Z/2Z-grading of the Jordan fusion rule Φ( 1
2 ) we get τe1 = τ1−e1 and τe2 = τ1−e2.
This leads to an axial representation of (Z/2Z)2, which is not of unique type.
2 u, e2 = 1
2 + 1
3 Modules over axial algebras
The definition of an axial algebra leads to the following natural definition for modules
over axial algebras, our main object of interest.
Definition 3.1. Let R be a commutative ring with identity and (A, Ω) a (Φ, ⋆)-axial
R-algebra. Let M be an R-module equipped with a (right) R-bilinear action of A,
M × A → M : (m, a) 7→ m · a.
For each e ∈ A and each φ ∈ R, let M e
every nonempty Λ ⊂ R, M e
each e ∈ Ω,
Λ =Lφ∈Λ M e
φ = {m ∈ M m · e = φm}. Define, as usual, for
φ and M e
∅ = {0}. We call M an A-module if, for
(i) there exists a decomposition M =Lφ∈Φ M e
φ and
φ and a ∈ Ae
ψ.
φ⋆ψ whenever m ∈ M e
(ii) m · a ∈ M e
Observe that A is itself an A-module; we refer to it as the regular module for A.
We can extend the definition of Miyamoto involutions and Frobenius forms to arbi-
trary modules over axial algebras.
Definition 3.2. Let A be an axial algebra for a Z/2Z-graded fusion rule and M an
A-module. For every (Φ, ⋆)-axis e ∈ A we define the Miyamoto involution µe ∈ GL(M )
of M :
mµe =( m if m ∈ M e
−m if m ∈ M e
;
.
Φ+
Φ−
Notice that µe is possibly trivial (namely when M e
Φ−
= 0).
Definition 3.3. Let A be an axial k-algebra for a field k. A Frobenius A-module is an
A-module M equipped with a bilinear form, called the Frobenius form,
such that hm · a, ni = hm, n · ai for all m, n ∈ M and a ∈ A.
h·, ·i : M × M → k
The following proposition is reminiscent of Maschke's theorem for linear representa-
tions of finite groups.
6
Proposition 3.4. Let A be a finite dimensional k-algebra and M a Frobenius A-module.
Let N be an A-submodule of M , i.e. a k-subspace for which N · A ⊆ N , such that the
Frobenius form is non-degenerate on N . Then there exists an A-submodule N0 of M
such that M = N ⊕ N0.
Proof. Let N0 = {m ∈ M hm, ni = 0 for all n ∈ N }. For all a ∈ A, n0 ∈ N0 and
n ∈ N , we have hn0 · a, ni = hn0, n · ai = 0 since n · a ∈ N . Hence n0 · a ∈ N0 and
N0 is an A-submodule. Since we require the Frobenius form to be non-degenerate on
N , it follows from the properties of orthogonal complements in finite dimensional vector
spaces that M = N ⊕ N0.
In the remainder of this section, we will study modules over axial representations. In
Theorem 3.9 below, we will show that every module over an axial representation (A, Ω)
of (G, D) leads to a group representation of a central extension U(A, Ω) of G. We start
by presenting the definition of this central extension U(A, Ω) and relating it to G.
Definition 3.5. Let (A, Ω) be an axial representation of (G, D). We define the group
U(A, Ω) as the group with presentation
(cid:28)te for each e ∈ Ω(cid:12)(cid:12)(cid:12)
(te)2 = 1
(tx)ty = txτy
for all e ∈ Ω
for all x, y ∈ Ω(cid:29) .
Note that we need Ω to be Miyamoto-closed in order to define this group.
We prove some useful properties of this group.
Proposition 3.6. Let (A, Ω) be an axial representation of (G, D).
(i) The map τ : U(A, Ω) → G : te 7→ τe is a group epimorphism.
(ii) The group U(A, Ω) is a central extension of G.
(iii) If Ω is finite, then so is U(A, Ω).
Proof.
(i) Consider the map τ : U(A, Ω) → G defined by te 7→ τe for all e ∈ Ω. This
map is a group homomorphism since all relations that define U(A, Ω) hold in G by
Proposition 2.9. Since the elements τe generate G, this map is surjective.
(ii) We prove that the kernel of τ is contained in the center of U(A, Ω). Let u :=
tx1tx2 · · · txn ∈ ker(τ ) where x1, x2, . . . , xn ∈ Ω. Then τx1τx2 · · · τxn = 1 and thus,
for all e ∈ Ω,
tu
e = t
tx1 tx2 ···txn
e
= teτx1 τx2 ···τxn = te.
Thus u commutes with all te and is therefore contained in the center of U(A, Ω).
(iii) We will show that every element of U(A, Ω) can be written as a product of at most
Ω elements te. Since only finitely many such products exist, U(A, Ω) must be
finite.
7
Consider a product of more than Ω elements te. Some element tx will then appear
at least twice and we can use the relation
txte1te2 · · · tentx = (te1te2 · · · ten)tx = t(e1)τx t(e2)τx · · · t(en)τx
to rewrite this element as a product of fewer te's.
For axial representations of unique type, we can say even more.
Proposition 3.7. Suppose (A, Ω) is an axial representation of (G, D) of unique type.
Then the following hold.
(i) G is centerless.
(ii) U(A, Ω) only depends on G and D.
(iii) The kernel of the group homomorphism τ : U(A, Ω) → G : te 7→ τe coincides with
the center of U(A, Ω).
Proof.
(i) Suppose τx1τx2 · · · τxn ∈ Z(G). Then, for all e ∈ Ω,
τe = (τe)τx1 τx2 ···τxn = τeτx1 τx2 ···τxn .
Since (A, Ω) is of unique type, the elements of Z(G) act trivially on Ω and hence
also on A.
(ii) The presentation of Definition 3.5 defining U(A, Ω) can be retrieved from (G, D).
Since (A, Ω) is of unique type, the generators can be identified with the set D and
the relations are the conjugacy relations between elements of D.
(iii) This follows from part (i) and Proposition 3.6(ii).
Remark 3.8. It is possible and might seem more natural to define a group U(G, D) in
a similar way as in Definition 3.5 using the elements of D rather than the idempotents
of an axial representation. This group U(G, D) was already studied by H. Cuypers and
J. Hall in the context of 3-transposition groups [Hal06, Proposition 2.1]. If (A, Ω) is an
axial representation of (G, D) of unique type, then U(G, D) ∼= U(A, Ω). However, for
the axial representation of (Z/2Z)2 from Example 2.12, which is not of unique type, we
have U(G, D) ∼= (Z/2Z)2 while U(A, Ω) ∼= (Z/2Z)4.
We are now able to state one of our main theorems.
Theorem 3.9. Let (A, Ω) be an axial representation of (G, D) and let M be an A-module.
The map defined by
µ : U(A, Ω) → GL(M ) : te 7→ µe
for all e ∈ Ω, is a group homomorphism.
We start by proving the following lemma, the proof of which is inspired by [Reh15,
Lemma 2.4.1].
8
Lemma 3.10. For all m ∈ M , a ∈ A and all x, y ∈ Ω we have:
(i) (m · a)µx = mµx · aτx ,
φ )µx = M yτx
(ii) (M y
φ ,
(iii) (µx)µy = µxτy .
Proof.
Φ+
and a ∈ Ax
(i) Let m ∈ M x
Φ+
, then m·a ∈ M x
Φ+
and therefore (m·a)µx = m·a,
mµx = m and aτx = a. For m ∈ M x
(resp. m ∈ M x
)
Φ+
Φ−
Φ−
we have m·a ∈ MΦ− and hence (m·a)µx = −m·a, mµx = −m (resp. mµx = m) and
. Now m · a ∈ M x
aτx = a (resp. aτx = −a). Finally, let m ∈ M x
Φ+
Φ−
and thus (m · a)µx = m · a, mµx = −m and aτx = −a. Since M = M x
⊕ M x
Φ−
Φ+
and A = Ax
, the desired property follows by linearity of the action of A
on M .
and a ∈ Ax
and a ∈ Ax
and a ∈ Ax
⊕ Ax
Φ+
Φ+
Φ−
Φ−
(ii) Let m ∈ M y
M yτx
M yτx
φ . Therefore (M y
φ = (M yτx
φ. By (i), it follows that mµx · yτx = (m · y)µx = φmµx, i.e. mµx ∈
φ and thus
φ . Similarly, (M yτx
φ )µx ⊆ M yτx τx
φ )µx ⊆ M yτx
= M y
φ
x ⊆ (M y
φ )µx . We find that (M y
φ )µ2
φ , we have mµx = εm with ε = ±1 only depending on whether φ ∈ Φ+
τy = εmµy . By applying µy,
and therefore mµy µx
φ )µx = M yτx
φ .
or φ ∈ Φ−. By (ii), mµy ∈ M xτy
φ
(iii) For m ∈ M x
mµy µx
τy µy = εm = mµx
follows for all m ∈ M x
µxτy .
φ . Since M =L M x
φ , by linearity we conclude that (µx)µy =
Proof of theorem 3.9. By the definition of U(A, Ω) and the fact that all µe have order
at most 2, the theorem follows from Lemma 3.10(iii).
4 Matsuo algebras
In Section 5, we will restrict our attention to the study of modules over Matsuo algebras,
a special type of axial algebras. This section introduces these algebras and explains their
connection with Fischer spaces and 3-transposition groups.
Definition 4.1.
(i) A point-line geometry G is a pair (P, L) where P is a set whose
elements are called points and L is a set of subsets of L. The elements of L are
called lines.
(ii) Two distinct points x and y of a point-line geometry are said to be collinear if
there is a line containing both and we denote this by x ∼ y. We write x ≁ y if x
and y are not collinear. If x ≁ y for all y 6= x, then we call x an isolated point.
(iii) A subspace of point-line geometry (P, L) is a point-line geometry (P ′, L′) such
that:
• P ′ ⊆ P and L ⊆ L′,
9
• if x, y ∈ P ′ and x, y ∈ ℓ for some ℓ ∈ L, then ℓ ∈ L′.
The point-line geometry (P, L) is a subspace of itself. If (P1, L1) and (P2, L2) are
two subspaces of (P, L), then so is (P1 ∩ P2, L1 ∩ L2). Therefore, for any set of
points and any set of lines, there is a smallest subspace containing those points
and lines; we call it the subspace generated by those points and lines.
(iv) Two points x and y of a point-line geometry are called connected if there exist
points x = x0, x1, . . . , xn = y such that xi−1 ∼ xi for all 1 ≤ i ≤ n. This relation
defines an equivalence relation on the set of points of a point-line geometry. The
subspaces generated by its equivalence classes are called the connected components
of the point-line geometry.
(v) An isomorphism between two point-line geometries G = (P, L) and L = (P ′, L′) is
a bijection θ : P → P ′ that induces a bijection between L and L′. We write G ∼= G′
and call G and G′ isomorphic if an isomorphism between them exists.
(vi) Let G be a point-line geometry such that through any two points there is at most
one line and such that each line contains exactly three points. Such a point-line
geometry is called a partial triple system. If x and y are collinear points in a partial
triple system, there is a unique third point on the unique line through x and y and
we will denote it by x ∧ y.
(vii) Let G be a partial triple system. We call G a Fischer space if each subspace
generated by two distinct intersecting lines is isomorphic to the dual affine plane
of order 2 or the affine plane of order 3. A sketch of these two point-line geometries
is given in Figure 1.
Figure 1: the dual affine plane of order 2 and the affine plane of order 3
The main motivation for studying Fischer spaces is their connection with 3-transposition
groups, due to F. Buekenhout [Bue74]; see Proposition 4.5 below.
10
Definition 4.2 ([Asc97]). A 3-transposition group is a pair (G, D) where D is a gener-
ating set of involutions of G closed under conjugation such that the order of the product
of any two elements of D is at most 3.
Example 4.3.
(i) All symmetric groups together with their set of transpositions form
a 3-transposition group.
(ii) The Fischer groups F i22, F i23 and F i24 are 3-transposition groups.
In the connection between Fischer spaces and 3-transposition groups that we will
need, it will be of importance to treat isolated points in Fischer spaces with some care.
Definition 4.4.
(i) Let (P, L) be a Fischer space and let P ′ ⊆ P be the set of isolated
points of (P, L). Then (P \ P ′, L) is clearly a Fischer space without isolated points
that we will denote by (P, L)◦.
(ii) If G = (P, L) is a Fischer space, then we associate with each point x ∈ P an
automorphism τx ∈ Aut(G) defined as
τx : P → P : y 7→(x ∧ y
y
if y ∼ x,
if y ≁ x.
Notice that τx is an involution unless x is an isolated point (in which case τx is triv-
ial). These involutions not only leave the set of points of the Fischer space invariant
but also map collinear points to collinear points; hence they induce automorphisms
of the Fischer space.
Now let D = {τx x ∈ P and τx 6= 1} and G = hDi ≤ Aut(G); then we define
f (G) := (G, D).
(iii) Let (G, D) be a 3-transposition group. Then we write (G, D)◦ for the 3-transposi-
tion group (G/ Z(G), {d Z(G) d ∈ D \ Z(G)}).
(iv) Let (G, D) be a 3-transposition group. Let P = D and let L = {{c, d, cd = dc}
o(cd) = 3}. Then g(G, D) := (P, L) is a point-line geometry.
Proposition 4.5 ([Bue74]). Let G be a Fischer space and let (G, D) be a 3-transposition
group. Then
(i) g(G, D) is a Fischer space,
(ii) f (G) is a 3-transposition group,
(iii) f (g(G, D)) ∼= (G, D)◦,
(iv) g(f (G)) ∼= G◦.
Example 4.6.
(i) The Fischer space corresponding to the 3-transposition group S4
is the dual affine plane of order 2.
(ii) The 3-transposition group related to the affine plane of order 3 is a semidirect
product 32 : 2 where the action is given by inversion.
11
We now present the definition of Matsuo algebras, which will be instances of axial
algebras; see Proposition 4.8 below. When the Matsuo algebra arises from a Fischer
space, its fusion rules will be Z/2Z-graded; see Proposition 4.9 below.
Definition 4.7. Let k be a field with char(k) 6= 2, let α ∈ k \ {0, 1} and let G = (P, L)
be a partial triple system. Define the Matsuo algebra Mα(G) as the k-vector space with
basis P and multiplication defined by linearly extending
x
0
α
2 (x + y − x ∧ y)
if x = y,
if x ≁ y,
if x ∼ y,
xy =
for all x, y ∈ P.
The following proposition gives us a decomposition of a Matsuo algebra as a direct
sum of eigenspaces for any x ∈ P.
Proposition 4.8 ([HRS15a, Theorem 6.2]). For each x ∈ P, the eigenspaces of x in
Mα(G) are
hxi, its 1-eigenspace,
hy + x ∧ y − αx y ∼ si ⊕ hy y ≁ xi, its 0-eigenspace,
hy − x ∧ y y ∼ xi, its α-eigenspace,
and the algebra Mα(G) decomposes as a direct sum of these eigenspaces.
These decompositions satisfy the Jordan fusion rule Φ(α) precisely when G is a
Fischer space:
Proposition 4.9 ([HRS15a, Theorem 6.5]). Let G = (P, L) be a partial triple system,
with corresponding Matsuo algebra A = Mα(G). Then the Matsuo algebra (A, P) is a
Φ(α)-axial algebra if and only if G is a Fischer space.
Since the Jordan fusion rule Φ(α) is Z/2Z-graded with Φ(α)+ = {1, 0} and Φ(α)− =
{α}, we can consider the Miyamoto involutions τx ∈ Aut(Mα(G)) for each x ∈ P. Iso-
lated points of the Fischer space should be treated with some care; they do not pose any
serious difficulties, however, since they would give rise to trivial Miyamoto involutions.
For simplicity, we nevertheless exclude this situation.
Proposition 4.10 ([HRS15a, Theorem 6.4]). Let G = (P, L) be a Fischer space without
isolated points, with corresponding Matsuo algebra A = Mα(G). Then:
(i) The pair (A, P) is an axial representation of the 3-transposition group f (G).
(ii) For each x ∈ P, the Miyamoto involution τx ∈ Aut(A) acts on P as the automor-
phism τx introduced in Definition 4.4(ii).
(iii) The axial representation (A, P) is of unique type.
12
Proof. Statements (i) and (ii) follow from [HRS15a, Theorem 6.4]. We now show (iii).
So let D = {τx x ∈ P} ⊆ Aut(A); we have to show that the map τ : P → D : x 7→ τx
is injective. Suppose τx = τy for some x, y ∈ P; then by (ii), τx and τy induce the same
automorphism of the Fischer space G. Since G has no isolated points, there exists some
point z ∈ P such that z ∼ x. Then zτy = zτx = x ∧ z and therefore both x and y are
the third point on the line through z and x ∧ z.
Remark 4.11. Let G = (P, L) be a Fischer space without isolated points, with cor-
responding Matsuo algebra A = Mα(G). By Proposition 4.10(ii), we can view the
Miyamoto group Miy(A, P) = hτx x ∈ Pi as a subgroup of either Aut(A) or Aut(G),
whichever is the most convenient.
We now start from a 3-transposition group (G, D) and we would like to construct an
axial representation for (G, D). Again, the situation giving rise to isolated points of the
corresponding Fischer space should be treated with some care, and this works out nicely
when we assume that G is centerless.
Proposition 4.12. Let (G, D) be a 3-transposition group with Z(G) = 1. Then (G, D)
has an axial representation (Mα(G), P) for the Fischer space G = g(G, D).
Proof. Since G is centerless, (G, D) = (G, D)◦. The statement now follows from Propo-
sition 4.5(iii) together with Proposition 4.10(i).
The following theorem, which is our main result in this section, shows that the group
U(Mα(G), P) can be interpreted as a universal 3-transposition group for the given Fischer
space G. Recall the definition of U(A) from Definition 3.5 above.
Theorem 4.13. Let G = (P, L) be a Fischer space without isolated points. Consider the
axial representation A = (Mα(G), P) of the 3-transposition group (G, D) := f (G). Then:
(i) (cid:0)U(A), {tx x ∈ P}(cid:1) is a 3-transposition group and g(cid:0)U(A), {tx x ∈ P}(cid:1) ∼= G.
(ii) Let (G′, D′) be a 3-transposition group such that g(G′, D′) ∼= G. Let ϕ : P → D′ be
an isomorphism between G and g(G′, D′). The map defined by
θ : U(A) → G′ : tx 7→ ϕ(x)
is a group epimorphism and G′/ Z(G′) ∼= U(A)/ Z(U(A)) ∼= G.
Proof.
(i) Since all relations that hold in U(A) have even length and all tx have order
at most 2, all tx have order exactly 2. Therefore {tx x ∈ P} is, by definition of
U(A), a generating set of involutions invariant under conjugation. Since G has no
isolated points, the axial representation A is of unique type and all τx are different.
By Proposition 3.6(i), also all tx must be different.
Let x, y ∈ P; then, by Proposition 4.10, either x = y or xτy = x if x ≁ y or xτy =
x ∧ y if x ∼ y. In the first case, txty = 1. In the second case, (txty)2 = txtxτy = 1.
In the third case, yτx = x ∧ y and thus (txty)3 = txτy tyτx = (tx∧y)2 = 1. This
proves that (U(A), {tx x ∈ P}) is indeed a 3-transposition group.
13
Let x, y, z ∈ P; then tx, ty and tz lie on a line in g(U(A), {tx x ∈ P}) if and only
if o(txty) = 3 and (tx)ty = tz. By our previous arguments, this happens if and only
if x ∼ y and z = x ∧ y.
(ii) For each x ∈ P, ϕ(x) ∈ D′ is an involution. Let x, y ∈ P. If x ∼ y, then, by
definition of g(G′, D′), ϕ(x)ϕ(y) = ϕ(x ∧ y) = ϕ(xτy ). If x ≁ y, then ϕ(x)ϕ(y) has
order 2 and ϕ(x)ϕ(y) = ϕ(x) = ϕ(xτy ). This proves that θ is a group epimorphism.
Consider the map θ′ : G′ → G defined by ϕ(x) 7→ τx. Then ker(θ′ ◦ θ) = Z(U(A))
by Proposition 3.7(iii). Because θ is surjective, ker(θ′) ≤ Z(G′) and, since G is
centerless, ker(θ′) = Z(G′). The isomorphism G ∼= U(A)/ Z(U(A)) follows from
Proposition 3.7(iii) or by applying the previous argument to the 3-transposition
group (G′, D′) =(cid:0)U(A), {tx x ∈ P}(cid:1).
Example 4.14. Let G = (P, L) be the affine plane of order 3. Then U(Mα(G), P)
is a semidirect product 32 : S3 which is a central extension of the group 32 : 2 from
Example 4.6(ii) by a cyclic group of order 3.
5 Modules over Matsuo algebras
Theorem 3.9 tells us that we can get a group representation out of a module over an
axial algebra. For Matsuo algebras, the converse will also be true. In Theorem 5.1 we
construct a module over a Matsuo algebra from such a group representation.
Theorem 5.1. Let G = (P, L) be a Fischer space. Let k be a field with char(k) 6= 2
and α ∈ k \ {0, 1}. Consider the Matsuo algebra Mα(G). Let V be a k-vector space and
ρ : U(Mα(G), P) → GL(V ) a group homomorphism. Since ρ(tx)2 = 1 for every x ∈ P
and char(k) 6= 2, we can decompose V as a direct sum of the 1- and (−1)-eigenspace of
ρ(tx). The action defined by linearly extending
v · x =(0
αv
if vρ(tx) = v;
if vρ(tx) = −v,
for each x ∈ P, equips V with the structure of an Mα(G)-module.
We start by mimicking Lemma 3.10(i) with µx replaced by ρ(tx).
Lemma 5.2. For every v ∈ V , a ∈ Mα(G) and x ∈ P, (v · a)ρ(tx) = vρ(tx) · aτx.
Proof. Since Mα(G) is spanned by the elements of P and the action of Mα(G) is linear
by definition, we may assume that a = y ∈ P. Since V is decomposable into a 1- and
(−1)-eigenspace of ρ(tx), it suffices to consider the cases where v is a 1- or (−1)-eigen-
vector of ρ(tx). If vρ(ty ) = v then v · y = 0 and (v · y)ρ(tx) = 0. Moreover
(cid:16)vρ(tx)(cid:17)ρ(tyτx )
= vρ(tx)ρ(txty tx) = vρ(tx)
14
and hence vρ(tx) · yτx = 0. In the second case, vρ(ty ) = −v and therefore v · y = αv and
(v · y)ρ(tx) = αvρ(tx). Now
(cid:16)vρ(tx)(cid:17)ρ(tyτx )
= vρ(tx)ρ(txty tx) = −vρ(tx)
and thus vρ(tx) · yτx = αvρ(tx). In both cases, (v · y)ρ(tx) = vρ(tx) · yτx.
Proof of Theorem 5.1. Let x ∈ P. The vector space V decomposes into a 1-and (−1)-eigenspace
of ρ(tx). By definition of the action of x, V decomposes as V = V x
0 (resp.
V x
α ) is the 1-eigenspace (resp. (−1)-eigenspace) of ρ(tx). It only remains to verify that
the fusion rule is satisfied. Let A = Mα(G). For v ∈ V x
{1,0} (resp. v ∈ V x
α
α) we have vρ(tx) = v (resp. vρ(tx) = −v) and aτx = a (resp. aτx = −a). By
and a ∈ Ax
Lemma 5.2,
0 and a ∈ Ax
α where V x
0 ⊕ V x
(v · a)ρ(tx) = vρ(tx) · aτx = v · a
and thus v · a is a 1-eigenvector of ρ(tx). Therefore v · a belongs to V x
a ∈ Ax
(resp. aτx = a). In both cases, by Lemma 5.2,
0 and
{1,0}), then vρ(tx) = v (resp. vρ(tx) = −v) and aτx = −a
α (resp. v ∈ V x
α and a ∈ Ax
0 . If v ∈ V x
(v · a)ρ(tx) = vρ(tx) · aτx = −v · a.
Therefore v · a is a 1-eigenvector of ρ(tx) and hence v · a ∈ V x
α .
Corollary 5.3. Let G = (P, L) be a Fischer space and let A = (Mα(G), P) be its
Matsuo algebra. There is a bijective correspondence between (isomorphism classes of )
modules M for A such that M x
1 = {0} for all x ∈ P and (isomorphism classes of ) linear
representations of U(A).
Proof. Using Theorem 5.1, we can associate to each linear representation ρ of U(A) an
A-module Mρ with (Mρ)x
1 = {0} for all x. We prove that the resulting map ζ : ρ 7→ Mρ
is a bijection from the set of isomorphism classes of linear representations of U(A) to
the set of isomorphism classes of A-modules with trivial 1-eigenspaces.
To show that ζ is onto, let M be an arbitrary A-module with M x
1 = {0} for all x ∈ P.
By Theorem 3.9, there is an associated linear representation ρ of the group U(A); we
claim that Mρ ∼= M . Indeed, since ρ(tx) = µx and M x
1 = {0}, we see that x · m = 0
(resp. x · m = αm) if and only if mρ(tx) = m (resp. mρ(tx) = −m) for all m ∈ M and
x ∈ P. By definition of Mρ, this proves the claim.
On the other hand, ρ is uniquely determined by its action on the eigenspaces of Mρ,
and hence ζ is also injective.
More generally, given any module M over the Matsuo algebra Mα(G), Theorem 3.9
gives us a representation ρ of the group U(Mα(G), P). The module constructed by
Theorem 5.1 out of ρ resembles M , with the important difference that all 1-eigenvectors
of an axis x ∈ P have become 0-eigenvectors. (The action on the 0- and α-eigenvectors
remains unchanged.) This leaves of course the question what the role of a 1-eigenvector
15
is inside a module. Since a regular module, i.e., the axial algebra as a module over
itself, contains non-trivial 1-eigenspaces, we definitely want to allow the existence of
1-eigenspaces in Definition 3.1. We will now prove that, under certain conditions, this is
the only way a 1-eigenspace can turn up in a module over a Matsuo algebra.
From now on, we would like to restrict to Matsuo algebras over connected Fischer
spaces without isolated points. This is not a serious restriction, as the following propo-
sition allows to generalize results to arbitrary Fischer spaces without isolated points by
looking at their connected components.
Proposition 5.4 ([HRS15a, Theorem 6.2]). Let G be a Fischer space and let {Gi i ∈ I}
be the set of its connected components. Then Mα(Gi) ∼=Li∈I Mα(Gi).
The connectedness of a Fischer space has the following implications on its Matsuo
algebra.
Lemma 5.5. Let G = (P, L) be a Fischer space and let Mα(G) be its Matsuo algebra.
Let D = {τx x ∈ P} and G = hDi ≤ Aut(Mα(G)). The following statements are
equivalent.
(a) The Fischer space G is connected.
(b) The action of G on P is transitive.
(c) The set D of Miyamoto involutions is a conjugacy class of G.
(d) The set {tx x ∈ P} is a conjugacy class of U(Mα(G), P).
Proof. Suppose the Fischer space G is connected. Let x, y ∈ P. Then there exists a path
x, x1, . . . , xn, y from x to y. Now
xτx∧x1 τx1∧x2 ···τxn∧y = y
and therefore G acts transitively on P. Conversely, by Proposition 4.10, all Miyamoto
involutions stabilize the connected components of the Fischer space. Therefore, the
Fischer space is connected when G acts transitively on P. The equivalence between (b)
and (c) follows from Proposition 2.9. By definition of the group U(Mα(G), P), it follows
that (c) and (d) are equivalent.
From now on, let G = (P, L) be a connected Fischer space without isolated points.
Consider the axial representation (Mα(G), P) of the 3-transposition group (G, D) :=
f (G). Let M be an Mα(G)-module and suppose M x
1 6= {0} for some (and hence every)
axis x ∈ P. We will prove in Theorem 5.8 that, under certain conditions, M contains
a submodule that is a quotient of Mα(G) as an Mα(G)-module. First, we will construct
the submodule. We introduce some notation.
Definition 5.6. Let M be an Mα(G)-module and suppose M x
1 6= {0} for each x ∈ P.
16
(i) Let
U := hµx x ∈ Pi ≤ GL(M ).
Denote the group U(Mα(G), P) by T and consider, as in Proposition 3.6(i) and
Theorem 3.9, the group epimorphisms defined by
µ : T → U : tx 7→ µx,
τ : T → G : tx 7→ τx.
(ii) For each x ∈ P, we define a subgroup Ux = µ(CT (tx)) of U , where CT (tx) is the
centraliser of tx in T .
(iii) Fix x ∈ P. Let m ∈ M x
1 with m 6= 0. Let
mµ.
mx = Xµ∈Ux
Because we assume that G is connected, the group T acts transitively on the set
{tx x ∈ P}. Thus for every y ∈ P there exists t ∈ T such that (tx)t = ty and we
define
my = (mx)µ(t).
(⋆)
By Lemma 5.7(iv) below, this definition of my is independent of the choice of t.
Lemma 5.7. Let y ∈ P, t ∈ T and φ ∈ {1, 0, α}. Suppose t′ ∈ T such that (tx)t = (tx)t′
Then the following hold:
.
(i) (ty)t = tyτ (t),
,
φ
φ )µ(t) = M yτ (t)
(ii) (M y
(iii) Uxµ(t) = Uxµ(t′),
(iv) (mx)µ(t) = (mx)µ(t′),
(v) my ∈ M y
1 ,
(vi) (my)µ(t) = myτ (t).
Proof.
(i) This follows by definition of U(Mα(G), P).
(ii) By Lemma 3.10(ii), this follows because µ and τ are group homomorphisms.
(iii) Since CU(Mα(G),P)(tx)t = CU(Mα(G),P)(tx)t′ whenever (tx)t = (tx)t′, also Uxµ(t) =
Uxµ(t′).
(iv) By (iii),
(mx)µ(t) = Xµ∈Uxµ(t)
mµ = Xµ∈Uxµ(t′)
mµ = (mx)µ(t′).
17
(v) Let t ∈ CU(Mα(G),P)(tx). Since tx = (tx)t = txτ (t), it follows from Proposition
1 . Thus mx =
1 . For y ∈ P, let t ∈ U(Mα(G), P) such that (tx)t = txτ (t) = ty and
4.10(iii) that xτ (t) = x. Therefore mµ(t) ∈ (M x
1 )µ(t) = M xτ (t)
mµ ∈ M x
= M x
1
1 )µ(t) = M xτ (t)
1
= M y
1 .
Pµ∈Ux
thus xτ (t) = y. Now, my = mµ(t)
x ∈ (M x
(vi) Because (tx)t′ = ty, xτ (t′) = y. Since (tx)t′t = txτ (t′t) = tyτ (t), it follows, by (⋆),
that myτ (t) = (mx)µ(t′ t) = (my)µ(t).
Theorem 5.8. Consider the setting of Definition 5.6. The subspace hmy y ∈ Pi is a
submodule of M and the map, defined by
Mα(G) → hmy y ∈ Pi : y 7→ my,
is a surjective homomorphism of Mα(G)-modules.
quotient of the regular module for Mα(G).
In particular, hmy y ∈ Pi is a
Proof. It suffices to prove that for idempotents y, z ∈ P
mz · z = mz;
my · z = 0
if y ≁ z;
if y ∼ z;
(my + my∧z − αmz) · z = 0
(my − my∧z) · z = α(my − my∧z) if y ∼ z.
(1)
(2)
(3)
(4)
Statement (1) follows from Lemma 5.7(v). For (2), let z, y ∈ P and y ≁ z. This
implies that z ∈ (Mα(G))y
1 . From the fusion rule M y
1 ·
(Mα(G))y
0 ⊆ {0} we infer my · z = 0. For (3) and (4), let z, y ∈ P and y ∼ z. Since
(my − my∧z)µz = my∧z − my by Lemma 5.7(vi), (4) follows. For (3), note that
0. By Lemma 5.7(v), my ∈ M y
mz · (y + z ∧ y − αz) = 0
since mz ∈ M z
0 ⊆ {0}. Interchanging the
roles of y, z and y ∧ z in this relation and relation (4) gives us the following 5 relations:
1 , y + z ∧ y − αz ∈ (Mα(G))z
1 · (Mα(G))z
0 and M z
my · (z + y ∧ z − αy) = 0,
mz · (y ∧ z + y − αz) = 0,
my∧z · (y + z − αy ∧ z) = 0,
(mz − my∧z) · y − α(mz − my∧z = 0,
(my − mz) · y ∧ z − α(my − mz) = 0.
(5)
(6)
(7)
(8)
(9)
It is easy to verify, using relation (1), that (5) − (6) + (7) + (8) − (9) gives us exactly
(3).
Remark 5.9. We do not know whether it is always possible to choose m ∈ M x
that mx 6= 0.
1 such
18
Theorem 5.11 will tell us when the map from Theorem 5.8 is an isomorphism. The
proof makes use of a Frobenius form for Matsuo algebras defined in Lemma 5.10 below.
The only obstructions are the facts that mx might be zero or that the Frobenius form
might be degenerate.
Lemma 5.10 ([HRS15b, Corollary 7.4]). Let G be a Fischer space. The Matsuo algebra
Mα(G) is a Frobenius axial algebra. When G is connected, its Frobenius form is, up to
scalar, uniquely determined. It is given by
for all x, y ∈ P.
1
α
2
0
if x = y,
if x ∼ y,
if x ≁ y,
hx, yi =
Theorem 5.11. Let G = (P, L) be a connected Fischer space without isolated points.
Let M be an Mα(G)-module, x ∈ P, m ∈ M x
1 \ {0} and define
mµ.
mx = Xµ∈Ux
Let U = hµx x ∈ Pi ≤ GL(M ). If mx 6= 0 and the Frobenius form of Mα(G) is non-
degenerate, then h(mx)µ µ ∈ U i is a regular submodule of Mα(G), i.e., it is isomorphic
to Mα(G) as Mα(G)-module.
Proof. It suffices to prove that the map from Theorem 5.8 is injective. Let ma be the
image of a ∈ Mα(G) under this map. We shall prove that for the stated conditions,
ma 6= 0 when a 6= 0. Since we require G to be connected, for every y ∈ P, there exists a
µ ∈ U such that my = (mx)µ. Because mx 6= 0 and µ ∈ GL(M ), my 6= 0 for all y ∈ P.
Suppose that a ∈ Mα(G) and ma = 0. Let y ∈ P be an arbitrary axis and write
a = a1 + a0 + aα where aφ ∈ (Mα(G))y
φ. Then
ma = ma1 + ma0 + maα = 0.
Since 0 = ma · y = ma·y,
We also have 0 = mµy
a = maτy and therefore
ma1 + αmaα = 0.
ma1 + ma0 − maα = 0.
From these three equations it follows that ma1 = ma0 = maα = 0.
Since the 1-eigenspace of y in Mα(G) is spanned by y, a1 = λy for some λ ∈ k.
Because my 6= 0, we conclude that a1 must be zero. As our choice of y ∈ P was
arbitrary, we infer that for every axis in P the component of a in its 1-eigenspace is zero.
Let h , i be the Frobenius form for Mα(G) as given by Lemma 5.10. Since eigenvectors
corresponding to different eigenvalues are perpendicular to each other, our previous
conclusion is equivalent to
ha, yi = 0 for all y ∈ P.
19
The elements of P span Mα(G) and therefore ha, ni = 0 for all n ∈ Mα(G). Because the
bilinear form h , i is non-degenerate, a = 0.
Remark 5.12.
(i) Let G be a connected Fischer space without isolated points and let
M be a finite dimensional Frobenius Mα(G)-module. Suppose that the Frobenius
form of Mα(G) is non-degenerate. Combining Theorem 5.11, Lemma 5.10 and
Proposition 3.4, we can decompose M as a direct sum of regular modules and a
1 and every x ∈ P.
(ii) Suppose G is a finite connected Fischer space without isolated points. Let A be
its collinearity matrix. The condition that the Frobenius form from Lemma 5.10
mµ = 0 for every m ∈ (M ′)x
module M ′ for which Pµ∈Ux
is non-degenerate can be expressed as det(cid:0)I + α
this can only fail for a finite number of choices for α.
2 A(cid:1) 6= 0. From this, it is clear that
(iii) Since the Miyamoto group of Mα(G) acts transitively on the points of a connected
Fischer space G, the number of lines through a point is a constant d ∈ N if G is
finite. The number of points collinear with any given point is then 2d. The vector
of all ones is therefore a (1 + αd)-eigenvector of I + α
2 A where A is the collinearity
matrix of G. It is easy to verify that
Xx∈P
x! a = (1 + αd)a
If 1 + αd = 0 and hence det(cid:0)I + α
for all a ∈ Mα(G).
is a 1-dimensional Mα(G)-submodule of Mα(G). The quotient module is then an
Mα(G)-module non-isomorphic to Mα(G) with a non-trivial 1-eigenspace for every
axis x ∈ P. The condition that the Frobenius form must be non-degenerate can
therefore not be omitted. Note that if 1 + αd 6= 0, then Mα(G) is a unital algebra
2 A(cid:1) = 0, then hPx∈P xi
with unit (1 + αd)−1(cid:0)Px∈P x(cid:1).
References
[Asc97] M. Aschbacher. 3-Transposition groups, volume 124 of Cambridge tracts in mathemat-
ics. Cambridge University Press, 1997.
[Bor86]
Richard E. Borcherds. Vertex algebras, Kac-Moody algebras, and the Monster. Proc.
Natl. Acad. Sci. USA, 83:3068 -- 3071, 1986.
[Bue74]
Francis Buekenhout. La g´eometrie des groupes de Fischer. Unpublished notes, 1974.
[FLM88]
Igor Frenkel, James Lepowsky, and Arne Meurman. Vertex Operator Algebras and the
Monster, volume 134 of Pure and applied mathematics. Academic Press, Inc., 1988.
[Gri82]
Robert L. Griess. The friendly giant. Invent. math., 69:1 -- 102, 1982.
[Hal06]
Jonathan I. Hall. A characterization of the full wreath product. J. Algebra, 300(2):529 --
554, 2006.
[HRS15a] Jonathan I. Hall, Felix Rehren, and Sergey Shpectorov. Primitive axial algebras of
jordan type. J. Algebra, 437:79 -- 115, 2015.
20
[HRS15b] Jonathan I. Hall, Felix Rehren, and Sergey Shpectorov. Universal axial algebras and
a theorem of Sakuma. J. Algebra, 421:394 -- 424, 2015.
[HSS17]
[IPS10]
[Iva09]
[Jac68]
Jonathan I. Hall, Yoav Segev, and Sergey Shpectorov. Miyamoto involutions in axial
algebras of Jordan type half. Israel Journal of Mathematics, 2017.
Alexander A. Ivanov, Dmitrii V. Pasechnik, and ´Akos Seress. Majorana representa-
tions of the symmetric group of degree 4. J. Algebra, 324:2432 -- 2463, 2010.
Alexander A. Ivanov. The Monster Group and Majorana Involutions, volume 176 of
Cambridge tracts in mathematics. Cambridge University Press, 2009.
Nathan Jacobson. Structure and Representations of Jordan Algebras, volume 39 of
Colloquium Publications. American Mathematical Society, 1968.
[Reh15]
Felix G. Rehren. Axial algebras. PhD thesis, University of Birmingham, 2015.
[Tit84]
Jacques Tits. On R. Griess' "Friendly giant". Invent. math., 78:491 -- 499, 1984.
21
|
1306.1276 | 1 | 1306 | 2013-06-06T01:14:47 | Directional Uncertainty Principle for Quaternion Fourier Transform | [
"math.RA",
"math-ph",
"math-ph",
"quant-ph"
] | This paper derives a new directional uncertainty principle for quaternion valued functions subject to the quaternion Fourier transformation. This can be generalized to establish directional uncertainty principles in Clifford geometric algebras with quaternion subalgebras. We demonstrate this with the example of a directional spacetime algebra function uncertainty principle related to multivector wave packets. | math.RA | math |
Directional Uncertainty Principle for Quater-
nion Fourier Transform
Eckhard M. S. Hitzer
Soli Deo Gloria
Abstract. This paper derives a new directional uncertainty principle for quater-
nion valued functions sub ject to the quaternion Fourier transformation. This
can be generalized to establish directional uncertainty principles in Clifford
geometric algebras with quaternion subalgebras. We demonstrate this with
the example of a directional spacetime algebra function uncertainty principle
related to multivector wave packets.
Mathematics Sub ject Classification (2000). Primary 11R52; Secondary 42A38,
15A66, 83A05, 35L05.
Keywords. Geometric algebra, quaternions, uncertainty, multivector wave pack-
ets, spacetime algebra.
1. Introduction
The Heisenberg uncertainty principle and quaternions are both fundamental for
quantum mechanics, including the spin of elementary particles. The quaternion
Fourier transform (QFT) [1, 2] is used in image and signal processing. It allows to
formulate a component wise uncertainty principle [3, 4, 5].
In Clifford geometric algebras, which generalize real and complex numbers
and quaternions to higher dimensions, the vector differential allows to formulate a
more general directional uncertainty principle [7, 8]. The present paper formulates
a directional uncertainty principle for quaternion functions. As prerequisite for its
proof we further investigate the split of quaternions introduced in [2, 6].
Then we show how the generalization of the QFT to a spacetime algebra
(STA) Fourier transform (SFT) [2] allows us to also generalize the directional
QFT uncertainty principle to STA. There the quaternion split corresponds to the
In memory of our dear friend Hiroshi Matsushita.
2
E. Hitzer
relativistic split of spacetime into time and space. The split of the SFT corresponds
then to analyzing a spacetime multivector function in terms of left and right trav-
elling multivector wave packets [2]. We will see that the energies of these wave
packets determine together with two arbitrary spacetime directions (one space-
time vector and one relativistic wave vector) the resulting uncertainty threshold.
2. Definition and properties of quaternions H
2.1. Basic facts about quaternions
Gauss, Rodrigues and Hamilton’s four-dimensional (4D) quaternion algebra H is
defined over R with three imaginary units:
ij = −j i = k, jk = −kj = i, ki = −ik = j , i2 = j 2 = k2 = ijk = −1.
(2.1)
Every quaternion can be written explicitly as
q = qr + qi i + qj j + qkk ∈ H,
qr , qi , qj , qk ∈ R,
(2.2)
and has a quaternion conjugate (equivalent to reversion in C l+
3,0 )
q = qr − qi i − qj j − qk k.
(2.3)
This leads to the norm of q ∈ H
q = pq q = qq2
j + q2
i + q2
r + q2
k ,
The scalar part of a quaternion is defined as
S c(q) = qr =
1
2
(q + q).
pq = pq .
(2.4)
(2.5)
(2.6)
2.2. The ± split of quaternions
A convenient split [2] of quaternions is defined by
q = q+ + q− ,
q± =
1
2
(q ± iqj ).
Explicitly in real components qr , qi , qj , qk ∈ R using (2.1) we get
q± = {qr ± qk + i(qi ∓ qj )}
1 ± k
2
=
1 ± k
2
{qr ± qk + j (qj ∓ qi )}.
(2.7)
This leads to the following new modulus identity
Lemma 2.1 (Modulus identity). For q ∈ H
q 2 = q− 2 + q+ 2 .
(2.8)
=
2
Directional Uncertainty Principle for QFT
3
Proof. Using (2.7) we get for the right side of (2.8)
1
2
q− 2 + q+ 2 =
[(qr + qk )2 + (qi − qj )2 + (qr − qk )2 + (qi + qj )2 ]
(2.9)
(2.10)
(cid:3)
2
=
1
2
.
because
1 ± k
2
1 ± k
2
1
[2q2
r + 2q2
k + 2q2
i + 2q2
j ] = q 2 ,
2
= (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)i
We can further derive the following useful split product identities.
Lemma 2.2 (Scalar part of mixed split product). Given two quaternions p, q and
applying the ± split we get zero for the scalar part of the mixed products
S c(p− eq+ ) = 0.
S c(p+ eq− ) = 0,
Proof. We only prove the first identity. The second works analogous.
1 + k
1 + k
S c(p+ eq− ) = S c({pr + pk + i(pi − pj )}
{qr − qk − i(qi + qj )})
2
2
= S c({pr + pk + i(pi − pj )}
1
S c({pr + pk + i(pi − pj )}{(qr − qk )k − j (qi + qj )}) = 0,
=
2
where we used
{qr − qk − i(qi + qj )})
(2.11)
(2.12)
k
2
1 + k
2
1 + k
2
=
k
2
.
(2.13)
(cid:3)
3. Overview of Fourier transforms (FT)
We first give a brief overview of some complex and hypercomplex Fourier trans-
forms.
3.1. Complex, Clifford, quaternion, spacetime Fourier transforms
The classical complex Fourier transformation (FT) is defined as
FC{f }(ω ) = ZR
where f ∈ L1 (R, C), and ω , x ∈ R.
Brackx et al. [9] extended the Fourier transform to multivector valued function-
distributions in C l0,n with compact support. A related applied approach for hy-
percomplex Clifford Fourier transformations in C l0,n was followed by Bulow et.
al. [10].
By extending the classical trigonometric exponential function exp(j x ∗ ξ)
(where ∗ denotes the scalar product of x ∈ Rm with ξ ∈ Rm , j the imaginary unit)
f (x) e−iωx dx,
(3.1)
4
E. Hitzer
in [11, 12], McIntosh et. al. generalized the classical Fourier transform. Applied to
a function of m real variables this generalized Fourier transform is holomorphic in
m complex variables and its inverse is monogenic in m + 1 real variables, thereby
effectively extending the function of m real variables to a monogenic function of
m+ 1 real variables (with values in a complex Clifford algebra). This generalization
has significant applications to harmonic analysis, especially to singular integrals
on surfaces in Rm+1 . Based on this approach Kou and Qian obtained a Clif-
ford Payley-Wigner theorem and derived Shannon interpolation of band-limited
functions using the monogenic sinc function [13, and references therein]. The Clif-
ford Payley-Wigner theorem also allows to derive left-entire (left-monogenic in
the whole Rm+1 ) functions from square integrable functions on Rm with compact
support.
The real n-dimensional volume element in = e1e2 . . . en of Gn = C ln,0 over
the field of the reals R has been used in [14, 7, 15, 8] to construct and apply Clifford
Fourier transformations for n = 2, 3 (mod 4) with kernels exp(−inx ∗ ω ), x, ω ∈
Rn . This in has a clear geometric interpretation. Note that i2
n = −1 for n =
2, 3 (mod 4).
For n = 3 the Clifford geometric algebra (GA) FT in G3 (replacing i → i3)
is given by
FG3 {f }(~ω) = ZR3
where f ∈ L1 (R3 , G3 ) and ~ω , ~x ∈ R3 . For n = 2, 3 (mod 4) the GA FT in Gn
(basically replacing i → in ) is
FGn {f }(ω) = ZRn
where f ∈ L1 (Rn , Gn ) and ω , x ∈ Rn .
Ell [1] defined the quaternion Fourier transform (QFT) for application to 2D
linear time-invariant systems of PDEs. Ell’s QFT belongs to the growing family
of Clifford Fourier transformations. But the left and right placement of the ex-
ponential factors in definition (3.4) distinguishes it. Later the QFT was applied
extensively to 2D image processing, including color images [5, 1, 10]. This spurred
research into optimized numerical implementations [16, 17].
The (double sided form of the) QFT in H (replacing i → i, j ) is commonly
defined as
FH{f }(ω) = f (ω ) = ZR2
where f ∈ L1 (R2 , H), d2x = dx1 dx2 and x, ω ∈ R2 .
Ell [1] and others [5, 18] also investigated related commutative hypercomplex
Fourier transforms like in the commutative subalgebra of G4 with subalgebra basis
{1, e12 , e34 , e1234 },
e−ix1ω1 f (x) e−jx2ω2 d2x,
f (x) e−inω ·x dnx,
f (~x) e−i3 ~ω ·~x d3~x,
(3.2)
(3.3)
(3.4)
12 = e2
e2
34 = −1,
e2
1234 = +1 .
(3.5)
Directional Uncertainty Principle for QFT
5
A higher dimensional generalization [2] of the QFT to the spacetime algebra
G3,1 is the spacetime FT (SFT) (replacing the quaternion units by i → et , j → i3)
will be explained and applied in section 7.2.
4. Uncertainty principle
4.1. The Heisenberg uncertainty principle in physics
Photons (quanta of light) scattering off particles to be detected cause minimal
position momentum uncertainty during detection. The uncertainty principle has
played a fundamental role in the development and understanding of quantum
physics. It is also central for information processing [19].
In quantum physics it states e.g. that particle momentum and position cannot
be simultaneously measured with arbitrary precision. The hypercomplex (quater-
nion or more general multivector) function f (x) would represent the spatial part of
a separable wave function and its Fourier transform (QFT, CFT, SFT, ...) F {f }(ω)
the same wave function in momentum space (compare [20, 21, 22]). The variance
in space Rn (in physics often n = 3) would then be calculated as (1 ≤ k ≤ n)
hf (x)(ek · x)2 f (x)i dnx = ZRn
(∆xk )2 = ZRn
where it is customary to set without loss of generality the mean value of ek · x to
zero [22]. The variance in momentum space would be calculated as (1 ≤ l ≤ n)
(2π)n ZRn
1
hF {f }(ω)(el · ω )2 eF {f }(ω)i dnω
(2π)n ZRn
1
(el · ω )2 F {f }(ω)2 dnω .
Again the mean value of el · ω is customarily set to zero, it merely corresponds
to a phase shift [22]. Using our mathematical units, the position-momentum un-
certainty relation of quantum mechanics is then expressed by (compare e.g. with
(4.9) of [21, page 86])
(ek · x)2 f (x)2 dnx,
(∆ωl )2 =
=
1
(4.1)
∆xk∆ωl =
δk,lF,
2
where δk,l is the usual Kronecker symbol. Note that we have not normalized the
squares of the variances by division with F = RRn f (x)2 dnx, therefore the extra
factor F on the right side of (4.1). Further explicit examples from image processing
can be found in [16].
In general in Fourier analysis such conjugate entities correspond to the vari-
ances of a function and its Fourier transform which cannot both be simultaneously
sharply localized (e.g. [19, 23]). Material on the classical uncertainty principle for
the general case of L2 (Rn ) without the additional condition limx→∞ x2 f (x) = 0
can be found in [24] and [25]. Felsberg [16] even notes for two dimensions: In 2D
however, the uncertainty relation is stil l an open problem. In [26] it is stated that
there is no straightforward formulation for the 2D uncertainty relation.
6
E. Hitzer
Let us now briefly define the notion of directional uncertainty principle in
Clifford geometric algebra.
4.2. Directional uncertainty principle in Clifford geometric algebra
From the view point of Clifford geometric algebra an uncertainty principle gives
us information about how the variance of a multivector valued function and the
variance of its Clifford Fourier transform are related. We can shed the restriction
to the parallel (k = l) and orthogonal (k 6= l) cases of (4.1) by looking at the
x ∈ Rn variance in an arbitrary but fixed direction a ∈ Rn and at the ω ∈ Rn
variance in an arbitrary but fixed direction b ∈ Rn . We are now concerned with
Clifford geometric algebra multivector functions f : Rn → Gn , n = 2, 3 (mod 4)
with GA FT F {f }(ω) such that
ZRn
The directional uncertainty principle [7, 8] with arbitrary constant vectors a,
b ∈ Rn means that
(2π)nF ZRn
F ZRn
1
1
(a · x)2 f (x)2 dnx
(b · ω )2 F {f }(ω)2 dnω ≥
Equality (minimal bound) in (4.3) is achieved for optimal Gaussian multivector
functions
f (x)2 dnx = F < ∞.
(a · b)2
4
(4.2)
(4.3)
f (x) = C0 e−α x2
C0 ∈ Gn arbitrary const. multivector, 0 < α ∈ R.
So far there seems to be no directional uncertainty principle for quaternion
functions over R2 sub ject to the QFT. But there already exists a component wise
uncertainty principle.
(4.4)
.
,
f (x)e−iω1 x1 e−j ω2 x2 d2x,
4.3. Component wise uncertainty principle for right sided QFT
For the right sided QFT Fr {f }: R2 → H of f ∈ L1 (R2 ; H) given by [3, 5]
Fr {f }(ω) = ZR2
where x = x1 e1 + x2 e2 , ω = ω1e1 + ω2e2 , and the quaternion exponential prod-
uct e−iω1 x1 e−j ω2 x2 is the quaternion Fourier kernel, it is possible to establish a
component uncertainty principle (k = 1, 2)
4 (cid:26)ZR2
f (x)2 d2x(cid:27)2
k f (x)2 d2x ZR2
ZR2
(2π)2
x2
where f ∈ L2 (R2 ; H) is now a quaternion-valued signal such that both (1 +
xk )f (x) ∈ L2 (R2 ; H) and ∂
f (x) ∈ L2 (R2 ; H). It is further possible to prove [3]
∂xk
that equality holds if and only if f is a Gaussian quaternion function
1−α2 x2
f (x) = C0 e−α1 x2
2 ,
k Fr {f }(ω)2d2ω ≥
ω 2
0 < α1 , α2 ∈ R.
const. C0 ∈ H,
(4.6)
(4.7)
(4.5)
,
Directional Uncertainty Principle for QFT
7
Remark 4.1. Now we want to address the very important question: Is it also
possible to establish a full directional uncertainty principle for the QFT?
In the following we will try to answer this question after investigating some
relevant properties of the double sided QFT, which is slightly different from (4.5).
5. Quaternion Fourier transform (QFT)
From now on we only consider the double sided QFT F {f }: R2 → H of f ∈
L1 (R2 ; H) in the form
F {f }(ω) = f (ω ) = ZR2
Linearity allows to also split up the QFT itself as
e−ix1ω1 f (x) e−jx2ω2 d2x.
(5.1)
F {f }(ω) = F {f− + f+}(ω) = F {f−}(ω ) + F {f+}(ω).
(5.2)
5.1. Simple complex forms for QFT of f±
The QFT of the f± split parts of a quaternion function f ∈ L2 (R2 , H) have simple
complex forms [2]
f± e−j (x2ω2∓x1ω1 )d2x = ZR2
f± = ZR2
We can rewrite this free of coordinates as
f+ = ZR2
f− = ZR2
where the reflection U1ω changes component ω1 → −ω1 . That this applies to ω1
is due to the choice of the kernel in (5.1).
e−i(x1ω1∓x2ω2 ) f±d2x .
f− e−j x·ω d2x,
(5.3)
f+e−j x·(U1ω) d2x,
(5.4)
5.2. Preparations for the full directional QFT uncertainty principle
Now we continue to lay the ground work for the full directional QFT uncertainty
principle. We begin with the following lemma.
Lemma 5.1 (Integration of parts). With the vector differential a · ∇ = a1∂1 + a2∂2 ,
with arbitrary constant a ∈ R2 , g , h ∈ L2 (R2 , H)
g (x)[a ·∇h(x)]d2x = (cid:20)ZR
g (x)h(x)dx(cid:21)a·x=∞
ZR2
−ZR2
a·x=−∞
Remark 5.2. The proof of Lemma 5.1 works very similar to the proof of integration
of parts in [7]. We therefore don’t repeat it here.
[a ·∇g (x)]h(x)d2x. (5.5)
For a quaternion function f and its QFT F {f } itself we also get important
modulus identities.
8
E. Hitzer
Lemma 5.3 (Modulus identities). Due to q 2 = q− 2 + q+ 2 of Lemma 2.1 we get
for f : R2 → H the fol lowing identities
f (x)2 = f−(x)2 + f+(x)2 ,
F {f }(ω)2 = F {f−}(ω )2 + F {f+}(ω )2 .
(5.6)
(5.7)
We further establish formulas for the vector differentials of the QFTs of the
split function parts f− and f+ .
Lemma 5.4 (QFT of vector differentials). Using the split f = f− + f+ we get the
QFTs of the split parts. Let b ∈ R2 be an arbitrary constant vector.
F {b · ∇f− }(ω) = b · ωF {f−}(ω ) j ,
F {b · ∇f+}(ω ) = b · (U1ω ) F {f+}(ω ) j ,
F {(U1b) · ∇f+}(ω) = b · ω F {f+}(ω ) j .
(5.8)
(5.9)
(5.10)
Remark 5.5. The previously explained complex forms of the QFTs make the proof
of Lemma 5.4 very similar to the case n = 2 of [8]. Noncommutativity must be
duly taken into account!
≥
Lemma 5.6 (Schwartz inequality). Two quaternion functions g , h ∈ L2(R2 , H) obey
the fol lowing Schwartz inequality
g (x)2 d2x ZR2
ZR2
h(x)2 d2x
4 (cid:20)ZR2
g (x)h(x) + h(x)g (x)d2x(cid:21)2
1
= (cid:20)ZR2
S c(g (x)h(x))d2x(cid:21)2
.
Remark 5.7. The proof of Lemma 5.6 can be based on the following inequality
(0 < ǫ ∈ R)
ZR2
6. Directional QFT uncertainty principle
[g (x) + ǫh(x)][g (x) + ǫh(x)]∼ d2x ≥ 0.
(5.12)
(5.11)
In this section we state the directional QFT uncertainty principle and prove it
step by step.
Theorem 6.1 (Directional QFT UP). For two arbitrary constant vectors a = a1e1+
a2e2 ∈ R2 , b = b1e1 + b2e2 ∈ R2 (selecting two directions), and f ∈ L2 (R2 , H),
x1/2f ∈ L2 (R2 , H) we obtain
(a·x)2 f (x)2 d2x ZR2
ZR2
+ (cid:3) ,
(cid:2)(a · b)2F 2
− + (a · b′ )2F 2
(6.1)
(b·ω )2 F {f }(ω)2d2ω ≥
(2π)2
4
Directional Uncertainty Principle for QFT
9
with the energies
F± = ZR2
Proof. We now prove the directional QFT uncertainty principle by the following
direct calculation.
b′ = U1b = −b1e1 + b2e2 ∈ R2 .
f±(x)2 d2x,
(6.2)
ZR2
(a · x)2 f (x)2 d2x ZR2
(b · ω )2 F {f }(ω)2d2ω
= (cid:20)ZR2
(a · x)2 f+ (x)2d2x(cid:21)
(a · x)2 f− (x)2d2x + ZR2
±split
(cid:20)ZR2
(b · ω)2 F {f+}(ω )2d2ω(cid:21)
(b · ω )2 F {f−}(ω)2 d2ω + ZR2
= h . . . i (cid:20)ZR2
F {(b′ · ∇)f+}(ω)(−j )2d2ω(cid:21)
F {(b · ∇)f− }(ω)(−j )2d2ω + ZR2
Lem.5.4
= (cid:20)ZR2
(a · x)f+ (x)2d2x(cid:21)
(a · x)f− (x)2 d2x + ZR2
Parsev.
(2π)2 (cid:20)ZR2
(b′ · ∇)f+ (x)2 d2x(cid:21)
(b · ∇)f− (x)2 d2x + ZR2
= (2π)2 (cid:20)ZR2
(a · x)f− (x)2 d2x ZR2
(b · ∇)f− (x)2d2x
(a · x)f+ (x)2 d2x ZR2
+ ZR2
(b′ · ∇)f+ (x)2 d2x
(a · x)f− (x)2 d2x ZR2
+ ZR2
(b′ · ∇)f+ (x)2 d2x
(b · ∇)f− (x)2d2x(cid:21)
(a · x)f+ (x)2 d2x ZR2
+ ZR2
(2π)2 "(cid:26)ZR2
S c (cid:16)a · xf− b · ∇ ef−(cid:17) d2x(cid:27)2
+ (cid:26)ZR2
S c (cid:16)a · xf+ b′ · ∇ ef+(cid:17) d2x(cid:27)2
Schwartz
≤
+ n ZR2
+ n ZR2
d2xo2
d2xo2 i
S c (cid:16)a · xf+ b · ∇ ef−(cid:17)
S c (cid:16)a · xf− b′ · ∇ ef+(cid:17)
{z
}
{z
}
=0 (Lem.2.2)
=0 (Lem.2.2)
b′ · ∇(f+ ef+)d2x(cid:27)2#
= (2π)2 "(cid:26)ZR2
b · ∇(f− ef− )d2x(cid:27)2
+ (cid:26)ZR2
1
1
a · x
a · x
2
2
10
(2π)2
4
Int. by p.
=
E. Hitzer
hn ZR2
+ n ZR2
f− 2d2xo2
b′ · ∇(a · x)
b · ∇(a · x)
}
{z
}
{z
=a·b
=a·b′
+ i .
h (a · b)2 F 2
(2π)2
− + (a · b′ )2 F 2
4
After using the Schwartz inequality of Lemma 5.6 we further used
S c (cid:16)f− b · ∇ ef−(cid:17) =
b · ∇ (cid:16)f− ef−(cid:17) , S c (cid:16)f+ b · ∇ ef+(cid:17) =
b · ∇ (cid:16)f+ ef+(cid:17) .
1
2
=
f+ 2d2xo2 i
(6.3)
1
2
(6.4)
(cid:3)
7. Generalization of directional uncertainty principle to spacetime
The quaternions frequently appear as subalgebras of higher order Clifford geomet-
ric algebras [28]. This is for example the case for the spacetime algebra (STA) [27],
which is of prime importance in physics, and in applications where time matters
as well (motion in time, video sequences, flow fields, ...). The quaternion subal-
gebra allows to introduce generalizations of the QFT to functions in these higher
order Clifford geometric algebras. For example it allows to generalize the QFT to
a spacetime FT [2].
7.1. Spacetime algebra (STA)
The spacetime algebra is the geometric algebra of R3,1 . In R3,1 we can introduce
the following orthonormal basis
2 = e2
1 = e2
t = e2
{et , e1 , e2 , e3}, −e2
3 = 1.
In G (R3,1 ) we thus get three anti-commuting blades that all square to minus one
(7.1)
i3 = e1 e2e3 ,
e2
i2
i2
t = −1,
3 = −1,
st = −1.
The volume-time subalgebra of G (R3,1 ) generated by these blades is indeed iso-
morphic to the quaternion algebra [29].
ist = et e1e2e3 ,
(7.2)
{1, et , i3 , ist} ←→ {1, i, j , k}
(7.3)
This isomorphism allows us now to introduce the ± split to spacetime algebra,
which now turns out to be a very real (physical) spacetime split
f± =
1
2
(f ± et f e∗
t ),
(7.4)
where
t = et i−1
e∗
st = −et ist = −etet i3 = i3 .
The time direction et determines therefore the complimentary 3D space with pseu-
doscalar i3 as well!
(7.5)
Directional Uncertainty Principle for QFT
11
7.2. From the QFT to the spacetime Fourier transform (SFT)
The spacetime Fourier transform maps 16D spacetime algebra functions f : R3,1 →
⋄
f : R3,1 → G3,1 . It is defined in the
G3,1 to 16D spacetime spectrum functions
following way
f (ω ) = ZR3,1
⋄
with
• spacetime vectors x = tet + ~x ∈ R3,1 , ~x = xe1 + ye2 + ze3 ∈ R3
• spacetime volume d4x = dtdxdydz
• spacetime frequency vectors ω = ωtet + ~ω ∈ R3,1 , ~ω = ω1e1 + ω2e2 + ω3e3 ∈
R3
e−et tωt f (x) e−i3 ~x·~ω d4x ,
f → FSF T {f }(ω) =
(7.6)
Remark 7.1. The 3D integration part
Z f (x) e−i3 ~x·~ω d3~x
in (7.6) fully corresponds to the GA FT in G3 , as stated in (3.2), compare [7, 8].
The ± split of the QFT can now, via the isomorphism (7.3) of quaternions to
the volume-time subalgebra of the spacetime algebra, be extended to splitting gen-
eral spacetime algebra multivector functions over R3,1 . This leads to the following
interesting result [2].
f+ e−i3 ( ~x·~ω− tωt )d4 x + ZR3,1
f − = ZR3,1
⋄
This result shows us that the SFT is identical to a sum of right and left propagating
multivector wave packets. We therefore see that these physically important wave
packets arise absolutely naturally from elementary purely algebraic considerations.
f− e−i3 ( ~x·~ω+ tωt )d4 x.
⋄
f + +
⋄
f =
(7.7)
7.3. Directional 4D spacetime uncertainty principle
We now generalize the directional QFT uncertainty principle to spacetime algebra
multivector functions.
Theorem 7.2 (Directional 4D spacetime uncertainty principle). For two arbi-
trary constant spacetime vectors a, b ∈ R3,1 (selecting two directions), and f ∈
L2 (R3,1 , G3,1 ), x1/2f ∈ L2 (R3,1 , G3,1 ) we obtain
(at t − ~a · ~x)2 f (x)2 d4x ZR3,1
ZR3,1
(btωt − ~b · ~ω )2 F {f }(ω)2d4ω
+ i ,
h(at bt − ~a · ~b)2F 2
(2π)4
− + (at bt + ~a · ~b)2F 2
≥
4
with energies of the left and right traveling wave packets:
F± = ZR2
f± (x)2d2x.
(7.8)
12
E. Hitzer
Remark 7.3. The proof is strictly analogous to the proof of the directional QFT
uncertainty principle. In that proof we simply need to replace the integral RR2 by
the integral RR3,1 , the infinitesimal 2D area element d2x by the infinitesimal 4D
volume element d4x, etc. Applying the Parseval theorem in four dimensions leads
to the factor of (2π)4 instead of (2π)2 .
8. Conclusion
We first studied the ± split of quaternions and its effects on the double sided QFT.
Based on this we have established a new directional QFT uncertainty principle.
Via isomorphisms of the quaternion algebra H to Clifford geometric subal-
gebras, the QFT can be generalized to these Clifford geometric algebras. This
generalization to higher dimensional algebras is also possible for the new direc-
tional QFT uncertainty principle. We demonstrated this with the physically most
relevant generalization to spacetime algebra functions.
There the quaternion ± split corresponds to the relativistic split of spacetime
into time and space, e.g. the rest frame of an observer. The split of the SFT
corresponds then to analyzing a spacetime multivector function in terms of left and
right travelling multivector wave packets [2]. The energies of these wave packets
determine together with two arbitrary spacetime directions (one spacetime vector
and one relativistic wave vector) the resulting uncertainty threshold.
Regarding the proliferation of higher dimensional theories in theoretical phys-
ics, and the model of Euclidean space in the socalled conformal geometric algebra
G4,1 , which has recently become of interest for applications in computer graphics,
computer vision and geometric reasoning [30, 31], even higher dimensional gener-
alizations of the directional QFT uncertainty principle may be of interest in the
future.
References
[1] T. A. Ell, Quaternionic-Fourier Transform for Analysis of Two-dimensional Lin-
ear Time-Invariant Partial Differential Systems. in Proceedings of the 32nd IEEE
Conference on Decision and Control, December 15-17, 2 (1993), 1830–1841.
[2] E. Hitzer, Quaternion Fourier Transform on Quaternion Fields and Generalizations,
Adv. in App. Cliff. Alg., 17 (2007), 497–517.
[3] B. Mawardi, E. Hitzer, A. Hayashi, R. Ashino, An Uncertainty Principle for Quater-
nion Fourier Transform. Computer & Mathematics with Applications, 56 (2008),
2398–2410.
[4] B. Mawardi, E. Hitzer, R. Ashino, R. Vaillancourt, Windowed Fourier transform of
two-dimensional quaternionic signals, submitted to Appl. Math. and Computation,
March 2009.
[5] T. Bulow, Hypercomplex Spectral Signal Representation for the Processing and Anal-
ysis of Image. PhD thesis, University of Kiel, 1999.
Directional Uncertainty Principle for QFT
13
[6] S. Georgiev, Real Quaternionic Calculus Handbook, private communication, May
2009.
[7] B. Mawardi and E. Hitzer, Clifford Fourier Transform and Uncertainty Principle for
the Clifford Geometric Algebra C l3,0 . Adv. App. Cliff. Alg., 16(1) (2006), 41–61.
[8] E. Hitzer and B. Mawardi, Clifford Fourier Transform on Multivector Fields and
Uncertainty Principle for Dimensions n = 2 (mod 4) and n = 3 (mod 4). P. Angl`es
(ed.), Adv. App. Cliff. Alg. Vol. 18(3,4) (2008), 715–736.
[9] F. Brackx, R. Delanghe, and F. Sommen, Clifford Analysis, Vol. 76 of Research Notes
in Mathematics, Pitman Advanced Publishing Program, 1982.
[10] T. Bulow, M. Felsberg and G. Sommer, Non-commutative Hypercomplex Fourier
Transforms of Multidimensional Signals, in G. Sommer (ed.), Geom. Comp. with
Cliff. Alg., Theor. Found. and Appl. in Comp. Vision and Robotics, Springer (2001),
187–207.
[11] C. Li, A. McIntosh and T. Qian, Clifford Algebras, Fourier Transform and Singular
Convolution Operators On Lipschitz Surfaces, Revista Matematica Iberoamericana,
10 (3), (1994), 665–695.
[12] A. McIntosh, Clifford Algebras, Fourier Theory, Singular Integrals, and Harmonic
Functions on Lipschitz Domains, chapter 1 of J. Ryan (ed.), Clifford Algebras in
Analysis and Related Topics, CRC Press, Boca Raton, 1996.
[13] T. Qian, Paley-Wiener Theorems and Shannon Sampling in the Clifford Analysis
Setting in R. Ablamowicz (ed.), Clifford Algebras - Applications to Mathematics,
Physics, and Engineering, Birkauser, Basel, (2004), 115–124.
[14] J. Ebling and G. Scheuermann, Clifford Fourier Transform on Vector Fields, IEEE
Transactions on Visualization and Computer Graphics, 11 (4), July/August (2005),
469–479.
[15] E. Hitzer, B. Mawardi, Uncertainty Principle for the Clifford Geometric Algebra
n,0 , n = 3(mod 4) based on Clifford Fourier transform, in Springer SCI book
C l
series Applied and Numerical Harmonic Analysis, 2006, pp. 45–54.
[16] M. Felsberg, Low-Level Image Processing with the Structure Multivector. PhD thesis,
Univ. of Kiel, 2002.
[17] S.C. Pei, J.J. Ding, J.H. Chang, Efficient Implementation of Quat. Fourier Transf.,
Convolution, and Correlation by 2-D Complex FFT, IEEE Trans. on Sig. Proc.
49(11) (2001), 2783–2797.
[18] F. Catoni, R. Cannata, P. Zampeti, An Introduction to Commutative Quaternions
Adv. in App. Cliff. Alg. 16(1) (2006), 1–28.
[19] J. M. Rassias, On The Heisenberg-Weyl Inequality, Jour. of Inequalities in Pure and
Appl. Math., 6 (1), article 11, (2005), 1–18. http://www.primedu.uoa.gr/jrasssias/
[20] C. Doran, A. Lasenby, Geometric Algebra for Physicists, CUP, Cambridge, 2003.
[21] F. Schwabl, Quantenmechanik, 2nd ed., Springer, Berlin, 1990.
[22] H. Weyl, The Theory of Groups and Quantum Mechanics, Dover, New York, 1950.
[23] J. G. Christensen, Uncertainty Principles, Master Thesis, University of Copenhagen,
2003.
[24] S. Mallat, A wavelet tour of signal processing, Academic Press, 2001.
[25] K. Grochenig, Foundations of Time-Frequency Analysis, Birkhauser, 2001.
14
E. Hitzer
[26] G.H. Granlund, H. Knutsson, Signal Processing for Computer Vision, Kluwer, Dor-
drecht, 1995.
[27] D. Hestenes, Space-Time Algebra, Gordon and Breach, 1966.
[28] P. Lounesto, Clifford Algebras and Spinors. 2nd ed., CUP, Cambridge, 2006.
[29] P. Girard, Quaternions, alg`ebre de Clifford et physique relativiste, Presses polytech-
niques, et universitaires romandes, 2004 (French). Quaternions, Clifford Algebras
and Relativistic Physics, Birkhauser, Basel, 2007 (English).
[30] L. Dorst, D. Fontijne and S. Mann, Geometric Algebra for Computer Science:
An Object-oriented Approach to Geometry. Morgan Kaufmann Series in Computer
Graphics, Elsevier, San Francisco, 2007.
[31] H. Li, Invariant algebras and geometric reasoning. World Scientific, Singapore, 2008.
[32] F. Collins, Director of the US National Human Genome Research Institute, in Time
Magazine, 5 Nov. 2006.
Acknowledgment
I do believe in God’s creative power in having brought it all into being in
the first place, I find that studying the natural world is an opportunity
to observe the ma jesty, the elegance, the intricacy of God’s creation. [32]
I thank my family for their constant loving support, as well as B. Mawardi
and S. Buchholz. I further thank the Cognitive Systems group (LS Sommer) in
Kiel for their hospitality.
Eckhard M. S. Hitzer
Department of Applied Physics
University of Fukui
910-8507 Fukui
Japan
e-mail: [email protected]
Submitted: April 22, 2008
Revised: May 9, 2008
|
1903.12105 | 2 | 1903 | 2019-04-18T17:40:28 | Classification of twisted generalized Weyl algebras over polynomial rings | [
"math.RA"
] | Let $R$ be a polynomial ring in $m$ variables over a field of characteristic zero. We classify all rank $n$ twisted generalized Weyl algebras over $R$, up to $\mathbb{Z}^n$-graded isomorphisms, in terms of higher spin 6-vertex configurations. Examples of such algebras include infinite-dimensional primitive quotients of $U(\mathfrak{g})$ where $\mathfrak{g}=\mathfrak{gl}_n$, $\mathfrak{sl}_n$, or $\mathfrak{sp}_{2n}$, algebras related to $U(\widehat{\mathfrak{sl}}_2)$ and a finite W-algebra associated to $\mathfrak{sl}_4$. To accomplish this classification we first show that the problem is equivalent to classifying solutions to the binary and ternary consistency equations. Secondly, we show that the latter problem can be reduced to the case $n=2$, which can be solved using methods from previous work by the authors. As a consequence we obtain the surprising fact that (in the setting of the present paper) the ternary consistency relation follows from the binary consistency relation. | math.RA | math |
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS
OVER POLYNOMIAL RINGS
JONAS T. HARTWIG AND DANIELE ROSSO
Abstract. Let R be a polynomial ring in m variables over a field of characteristic zero.
We classify all rank n twisted generalized Weyl algebras over R, up to Zn-graded iso-
morphisms, in terms of higher spin 6-vertex configurations. Examples of such algebras
include infinite-dimensional primitive quotients of U (g) where g = gln, sln, or sp2n, al-
gebras related to U ( bsl2) and a finite W-algebra associated to sl4. To accomplish this
classification we first show that the problem is equivalent to classifying solutions to the
binary and ternary consistency equations. Secondly, we show that the latter problem can
be reduced to the case n = 2, which can be solved using methods from previous work by
the authors [17], [15]. As a consequence we obtain the surprising fact that (in the setting
of the present paper) the ternary consistency relation follows from the binary consistency
relation.
1. Introduction and motivation
Generalized Weyl algebras (GWAs) form a class of noncommutative rings introduced
by Bavula [1] and, under the name hyperbolic rings, by Rosenberg [26]. They have been
subject to intense research in the last few decades [1, 2, 3, 4] due to their many pleasant
properties. Many rings of interest in ring theory (Weyl algebras, ambiskew polynomial
rings [19], generalized down-up algebras [6]), representation theory (U (g) and Uq(g) where
g = sl2, gl2, or the positive part of sl3) and noncommutative geometry (quantum spheres,
quantum lens spaces [4] and references therein) are examples of GWAs. In addition the
class is closed under taking tensor products, and taking the ring of invariants with respect
to finite order graded automorphisms.
When algebraic structures are given by presentations, the question arises whether two
different presentations give rise to isomorphic algebras. This is known as the isomorphism
problem. In general, answering this question can be very difficult. For example, to deter-
mine whether two presented groups are isomorphic is known to be an NP hard problem.
The isomorphism problem for GWAs has been studied in [2, 5, 9, 28, 30, 21].
Despite the variety of interesting examples of GWAs, there are some algebras that
"should" be GWAs but are not, for example multiparameter quantized Weyl algebras de-
fined by Maltsiniotis [22] and studied in [20]. To remedy this deficiency, as well as including
examples related to higher rank Lie algebras, Mazorchuk and Turowska introduced the class
of twisted generalized Weyl algebras (TGWAs). Their structure and representation theory
Date: April 19, 2019.
1
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 2
have been extensively studied. See for example [24, 25, 23, 27, 12, 13, 16, 7, 8, 17, 15, 14] and
references therein. Besides quantized Weyl algebras, examples include quotients of envelop-
ing algebras of simple Lie algebras by annihilators of completely pointed (i.e. multiplicity-
free) simple weight modules [18].
An additional difficulty arise for TGWAs, as they are not in general free as left modules
over their degree zero subring. In particular, for some choices of input data the relations
will be contradictory, i.e. the algebra will be trivial, consisting of a single element 0 = 1,
see [7].
To show that the algebra is nontrivial, the strategy was to show that the degree zero
subring is isomorphic to the base ring R. This question can be seen as a weak isomor-
phism problem, because we are asking for the structure of the degree zero subring. The
full isomorphism problem for TGWAs has only been studied for special classes such as
multiparameter quantized Weyl algebras [11, 10, 29].
In this paper we solve the graded isomorphism problem for a natural class of TGWAs.
First we show that the graded isomorphism problem is equivalent to classifying solutions
to the binary and ternary consistency equations, up to a natural equivalence that we
define. The latter amounts to the following polynomial problem, which is a higher rank
and multivariate generalization of the problem addressed in [17].
Problem 1.1. Let k be a field of characteristic zero, m a positive integer and let R =
k[u1, u2, . . . , um] be the polynomial algebra over k in m commuting independent inde-
terminates uj. Find all pairs (α, p), where α = (αij) ∈ Mm×n(k) is a matrix and
p = (p1, p2, . . . , pn) ∈ Rn is an n-tuple of polynomials satisfying the following equations:
pi(u − αj
2 )pj(u − αi
2 + αj
2 ) = pi(u + αj
2 ) = pk(u − αi
2 )pj(u + αi
2 + αj
2 ) ∀i 6= j
2 − αj
2 )pk(u + αi
pk(u − αi
2 − αj
2 )pk(u + αi
2 ) ∀i 6= j 6= k 6= i
(1.1a)
(1.1b)
where u = (u1, u2, . . . , um) and αj = (α1j, α2j, . . . , αmj) for j = 1, 2, . . . , n.
In [17], we solved this problem in the case of (m, n) = (1, 2) and k = C.
Secondly, generalizing results from [17], we prove that any solution can be factored into
orbital solutions, i.e. solutions all of whose irreducible factors belong to a given Zn-orbit
in R. (In [17], where (m, n) = (1, 2), orbital solutions were called integral solutions.) Next
we prove that any orbital solution is essentially rank two (see Corollary 4.5). Finally, we
extend the methods of [17, 15] to give a combinatorial classification of all essentially rank
two solutions in terms of vertex configurations on a square lattice.
As a surprising consequence, we deduce that, when R is a polynomial ring and the
automorphisms are given by additive shifts, the binary consistency relation actually implies
the ternary consistency relation. In general, as shown in [7], these relations are independent.
Future directions. It is natural to ask what will happen in finite characteristic. Some
solutions to the consistency equations in finite characteristic were given in [24]. Another
direction would be to study this question when the base algebra is not a polynomial algebra,
for example a Laurent polynomial ring. This case is expected to be related to quantum
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 3
analogs of the algebras from this paper. Lastly we remark that the problem of determining
consistency equations necessary and sufficient for a given TGWA to be consistent is still
open in the case when certain elements ti (see Definition 2.1) are allowed to be zero divisors.
Such TGWAs appear naturally as matrix algebras that are finite-dimensional primitive
quotients of enveloping algebras [18], or as algebras of differential operators with C∞-
function coefficients.
2. On the graded isomorphism problem for twisted generalized Weyl
algebras
2.1. Definition of twisted generalized Weyl algebra. We use slightly different nota-
tion from other papers in the literature. Specifically, our use of σ1/2
comes from the fact
that we want to write the equations (2.1) in a symmetric way. This system of notation is
equivalent to the original one from [24] by simple substitutions, see Section 6.3 for details.
i
Definition 2.1. Let n be a positive integer.
(i) A twisted generalized Weyl datum (TGWD) of rank n is a triple (R, σ, t) where R is
n ) is an n-tuple of commuting ring automorphisms of
a ring, σ = (σ1/2
R, and t = (t1, t2, . . . , tn) is an n-tuple of nonzero elements from the center of R.
, . . . , σ1/2
, σ1/2
1
2
(ii) (R, σ, t) is regular if ti is regular (i.e. not a zero-divisor) in R for all i.
(iii) (R, σ, t) is consistent if the following equations hold:
σ1/2
j
i
(ti) · σ1/2
σ−1/2
j
(tj) = σ−1/2
(tk) = σ1/2
j
σ1/2
i σ1/2
j
(tk) · σ−1/2
i
(ti) · σ−1/2
i
(tj) (i 6= j)
(2.1a)
i σ−1/2
j
(tk) · σ−1/2
i
σ1/2
j
(tk) (i 6= j 6= k 6= i)
(2.1b)
(iv) Let (R, σ, t) be a TGWD of rank n. The associated twisted generalized Weyl construc-
tion (TGWC), C(R, σ, t), is the Zn-graded R-ring generated by 2n indeterminates
X +
1 , X −
1 , . . . , X +
n , X −
n subject to:
i X ∓
i X ∓
(ti)1,
i = σ±1/2
i
j X ±
j = X ∓
i
(r) = σ±1/2
(r)X ±
i
i ) = ±ei,
X ±
X ±
i σ∓1/2
deg(X ±
deg(r1) = 0 (∀r ∈ R),
i
i
X ±
(i 6= j),
(∀r ∈ R),
(2.2a)
(2.2b)
(2.2c)
(2.2d)
(2.2e)
where ⊕n
i=1Zei = Zn.
(v) Let (R, σ, t) be a TGWD of rank n. The associated twisted generalized Weyl algebra
(TGWA), A(R, σ, t), is the quotient ring C(R, σ, t)/I where I is the sum of all graded
ideals J = ⊕d∈ZnJd such that J0 = {0}.
A basic problem for rings given by generators and relations is whether the ring is non-
trivial. The following result, which was the main theorem from [7], gives a sufficient
condition in the case of TGWCs and TGWAs. This motivates the terminology in Definition
2.1(iii).
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 4
Theorem 2.2 ([7]). Let (R, σ, t) be a regular TGWD of rank n, let C = C(R, σ, t) and
A = A(R, σ, t) be the associated TGWC and TGWA respectively, with Zn-gradation C =
Ld∈Zn Cd, A =Ld∈Zn Ad. Then the following are equivalent:
(i) (R, σ, t) is consistent,
(ii) the canonical map R → C0 is an isomorphism,
(iii) the canonical map R → A0 is an isomorphism.
In particular, if (R, σ, t) is regular and consistent, then C and A are both non-trivial rings.
That the canonical maps are both surjective is easy to see. The difficult part is to prove
they are injective. The proof involves the diamond lemma and the relations (2.1). Relation
(2.1a) appeared already in [24].
We will need the following lemma, which is a special case of [12, Lem. 3.2]
Lemma 2.3. Let (R, σ, t) be a TGWD of rank n. Let C = C(R, σ, t) and A = A(R, σ, t)
be the corresponding TGWC and TGWA. Then:
C±ei = C0X ±
A±ei = A0X ±
i = X ±
i = X ±
i C0
i A0
(2.4)
(2.3)
2.2. Graded isomorphism problem. Let A = A(R, σ, t) and A′ = A(R′, σ′, t′) be two
TGWAs of the same rank n. We say that A and A′ are graded isomorphic if there exists a
ring isomorphism ϕ : A → A′ such that ϕ(Ad) ⊆ A′
d for all d ∈ Zn.
Put σi = (σ1/2
)2. In this section we show that, under certain conditions, two TGWAs
are graded isomorphic if and only if their respective TGWDs (R, σ, t) and (R′, σ′, t′) are
equivalent in the following sense.
i
Definition 2.4. Two TGWDs of rank n, (R, σ, t) and (R′, σ′, t′), are equivalent if there
exists a ring isomorphism ψ : R → R′ such that ψ ◦ σi = σ′
i for
all i;
i ◦ ψ and ψ(ti) ∈ (R′)× · t′
Here R× denotes the group of units (invertible elements) in a ring R. To state the result,
we will need the following definitions.
Definition 2.5. Let (R, σ, t) be a TGWD.
(i) (R, σ, t) is commutative if the ring R is commutative;
(ii) (R, σ, t) has scalar units if σi(u) = u for every u ∈ R× and every i.
Proposition 2.6. Let (R, σ, t) and (R′, σ′, t′) be two regular, consistent, commutative,
TGWDs of rank n having scalar units. Let A = A(R, σ, t) and A′ = A(R′, σ′, t′) be the
corresponding TGWAs. Then A and A′ are graded isomorphic if and only if (R, σ, t) is
equivalent to (R′, σ′, t′).
Proof. Suppose that (R, σ, t) and (R′, σ′, t′) are equivalent and let ψ : R → R′ be a ring
isomorphism such that ψ ◦ σi = σ′
i for all i, where ri ∈ R×. Define
−, and Ψ(r) = ψ(r) for all r ∈ R. It is straightforward to
Ψ(X +
verify that Relations (2.2) are preserved by Ψ. Thus Ψ extends to a ring homomorphism
i ◦ ψ and ψ(ti) = ri · t′
i ) = riX ′
i
+, Ψ(X −
i ) = X ′
i
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 5
Ψ : C(R, σ, t) → C(R′, σ′, t′). Clearly Ψ is a graded homomorphism. Replacing ψ by ψ−1
we obtain a graded homomorphism which is the inverse of Ψ. Thus Ψ is an isomorphism of
Zn-graded rings. Since Ψ is graded we have Ψ(I) ⊆ I′ where I and I′ are as in the definition
of TGWA. Thus Ψ induces a graded isomorphism of the quotients A(R, σ, t) → A(R′, σ′, t′).
Conversely, suppose Ψ : A → A′ is a graded isomorphism. That is, Ψ is a ring isomor-
phism such that Ψ(Ad) = (A′)d for every d ∈ Zn. By Theorem 2.2 we can identify A0 ∼= R
and (A′)0 ∼= R′. Thus taking d = 0 we get a ring isomorphism ψ = ΨR : R → R′. To
show that ψ satisfies the required properties we also need to consider the case d = ±ei.
±).
By Lemma 2.3, A±ei = RX ±
± ∈ R′, and
From Ψ(A±ei) = (A′)±ei we conclude that Ψ(X ±
Ψ−1(X ′
i
± (denoting the generators of A′ by X ′
i
i ∈ R. Combining these equalities we get
± for some r′
i
±ei = R′X ′
i
for some r±
i and A′
i ) = r′
i
±) = r±
i X +
i
±X ′
i
X ±
i = Ψ−1(cid:0)Ψ(X ±
i )(cid:1) = ψ−1(r′
i
±)r±
i X ±
i .
Multiplying from the right by X ∓
i and using (2.2a) we get
σ±1/2
i
(ti) = ψ−1(r′
i
±)r±
i σ±1/2
i
(ti)
hence
σ±1/2
i
(ti) · (1 − ψ−1(r′
i
±)r±
i ) = 0
Since ti is regular in R, the same is true for σ±1/2
i
(ti). Hence
ψ−1(r′
i
±)r±
i = 1
i.e. r±
i and r′
i
± are units.
Next, applying Ψ to the relation X ±
i σ∓1/2
i
(r) = σ±1/2
i
(r)X ±
i we obtain:
r′
i
±X ′
i
±ψ(cid:0)σ∓1/2
i
(r)(cid:1) = ψ(cid:0)σ±1/2
i
(r)(cid:1)r′
i
±
±X ′
i
(2.5)
Dividing both sides by the unit r′
i
and (2.2c), we get
± and multiplying from the right by X ′∓
i and using (2.2a)
σ′
i
±1/2(t′
i) · σ′
i
±1 ◦ ψ ◦ σ∓1/2
i
(r) = ψ(σ±1/2
i
±1/2(t′
i)
(2.6)
(r)(cid:1)σ′
i
Since t′
it from both sides. Furthermore since the equation holds for all r ∈ R we obtain
i is central and not a zero-divisor, the same is true for σ′
i
i) and we may cancel
±1/2(t′
σ′
i
±1 ◦ ψ = ψ ◦ σ±1
i
(2.7)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 6
for all i. Therefore
(ti)
±1/2 ◦ ψ ◦ σ±1/2
i
∓1/2 ◦ Ψ(X ±
i X ∓
i )
±X ′
±r′
∓X ′
∓)
i
i
i
±1(r′
±σ′
∓) · σ′
i
i
i
±1(r′
±σ′
i
i
ψ(ti) = σ′
i
= σ′
i
= σ′
i
= σ′
i
= σ′
i
i
∈ (R′)× · t′
i
∓1/2(cid:0)r′
∓1/2(cid:0)r′
∓1/2(cid:0)r′
∓)(cid:1) · t′
i
i
±1/2(t′
i
i)(cid:1)
This proves that ψ gives an equivalence between (R, σ, t) and (R′, σ′, t′).
(cid:3)
Remark 2.7. Notice that, as a consequence, if we have two TGWAs satisfying the hy-
potheses of the theorem, defined respectively over the rings R and R′, that are graded
isomorphic, then the rings are isomorphic, so we can identify R and R′.
Our goal is to classify TGWAs defined over a polynomial ring, up to graded isomorphism,
so we introduce the following definition.
Definition 2.8. Let k be a field of characteristic zero, and R = k[u1, u2, . . . , um]. Let
p = (p1, . . . , pn) (resp. p′ = (p′
n)) be an n-tuple of nonzero monic elements of R,
and α = (αij ) ∈ Mm×n(k) (resp. α′ = (α′
ij) ∈ Mm′×n(k)) and σi ∈ Autk(R) given by
σi(uj) = uj − αji (resp. σ′
ji). We say that the pairs
(α, p) and (α′, p′) are equivalent is there exists an automorphism ψ ∈ Aut(R) such that
ψ ◦ σi = σ′
i ∈ Autk(R′), given by σ′
i ◦ ψ and ψ(pi) ∈ k× · p′
i for all 1 ≤ i ≤ n.
1, . . . , p′
i(uj) = u′
j − α′
Example 2.9. Let g ∈ GLm(k) = Aut(⊕m
i=1kui) be an invertible m × m matrix, then g
gives an automorphism ψg ∈ Aut(R) by acting linearly on the variables. In this case, we
have that if ψg ◦ σi = σ′
n) = (p1 ◦ g, . . . , pn ◦ g). This
shows that there is a GLm(k)-action on the set of pairs (α, p) as defined in Def. 2.8, and
that GLm(k)-orbits are contained in the equivalence classes.
i ◦ ψg, then α′ = gα and p′ = (p′
1, . . . , p′
Example 2.10. Let α = 0, and ψ ∈ Aut(R) be any automorphism of the polynomial
ring, then ψ ◦ σi = σ′
i ◦ ψ is always satisfied, so (0, p) and (0, ψ(p)) are always equivalent
no matter what ψ is. For example, we could choose ψ ∈ Aut(R) defined by ψ(uj) =
uj + qj(uj+1, . . . , um) for some polynomials qj, 1 ≤ j ≤ m.
In particular, this shows
that the equivalence classes of pairs (α, p) are bigger in general than the GLm(k)-orbits
described in Example 2.9.
In the case of polynomial rings R, Proposition 2.6 gives us the following.
Corollary 2.11. Let R, σ, σ′, p, p′ as in Definition 2.8 such that (R, σ, p) and (R, σ′, p′)
are consistent TGWDs.
Then the TGWAs A = A(R, σ, p) and A′ = A(R, σ′, p′) are isomorphic as graded k-
algebras if and only if (α, p) and (α′, p′) are equivalent.
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 7
Proof. This follows directly from Proposition 2.6 with the observation that, given our setup,
the TGWDs (R, σ, p) and (R, σ′, p′) are also regular, commutative and have scalar units.
In addition, the automorphism ψ giving the equivalence has to be k-linear.
(cid:3)
3. Orbital solutions
In this section, unless otherwise stated, k can be any field with char k 6= 2. Equipping
the gradation monoid Nm with an order, such as lexicographical, each nonzero element
of R has a unique leading monomial. We say an element of R is monic if its leading
monomial has coefficient 1. Note that the set of monic elements of R is invariant under
the automorphisms σi. We say that p = (p1, . . . , pn) ∈ Rn is monic if pi is (nonzero and)
monic for each i.
Let Irr(R) denote the set of monic irreducible polynomials in R. The commuting auto-
(a1, a2, . . . , an).p = σa1
morphisms σi induce an action of the group Zn on Irr(R) via
2 ◦ · · · ◦ σan
Since char k 6= 2, σi have square roots in Autk(R) given by σ1/2
Definition 3.1. Let O ∈ Irr(R)/Zn be an orbit of monic irreducible polynomials with
respect to the Zn action defined in (3.1). A monic element p = (p1, p2, . . . , pn) ∈ Rn
is O-orbital if for each i ∈ {1, 2, . . . , n}, every irreducible factor of pi belongs σ1/2
(O) =
(uj) = uj − 1
1 ◦ σa2
2 αji.
n (p)
(3.1)
i
i
(i) p = p(1)p(2) · · · p(k) as an equality in the direct product ring Rn,
(ii) p(i) is a monic solution to (1.1) for each i ∈ {1, 2, . . . , k},
(iii) p(i) is Oi-orbital for each i ∈ {1, 2, . . . , k},
(iv) Oi 6= Oj for all i 6= j.
Proof. Let O ∈ Irr(R)/Zn be any orbit. Let pO = (pO
product of all monic irreducible factors of pi belonging to σ1/2
2 , . . . , pO
1 , pO
n ) ∈ Rn, where pO
is the
i
(O). By convention pO
i = 1 if
there are no such irreducible factors. Then clearly p =Q pO where O ranges over all orbits
in Irr(R)/Zn. Moreover each pO is O-orbital. So it remains to prove that pO is a solution
to (1.1) for each O.
i
Fix i, j ∈ {1, 2, . . . , n}, i 6= j. Let f be any irreducible factor of pO
Without loss of generality suppose that f divides pO
i (u). Hence f belongs to σ1/2
pO
(O). On the other hand f divides pi(u − αj/2), and
hence since p solves (1.1), f divides pi(u + αj/2)pj(u + αi/2). Since f is irreducible, f
divides either pi(u + αj/2) or pj(u + αi/2). If f divides pi(u + αj/2) then σ1/2
(f ) divides
pi(u), hence f divides pO
i (u + αj/2). Similarly in the other case f divides pO
j (u + αi/2).
i σ1/2
j
j
j
i (u−αj /2)pO
i (u − αj/2). Then σ−1/2
j (u−αi/2).
(f ) divides
i
(cid:8)σ1/2
(q) q ∈ O(cid:9).
Theorem 3.2. Let p be any monic solution to (1.1). Then there exists a unique finite
subset
such that
(cid:8)(O1, p(1)), (O2, p(2)), . . . , (Ok, p(k))(cid:9) ⊆ (Irr(R)/Zn) × Rn
(3.2)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 8
This shows that any irreducible factor in the left hand side of (1.1a) for pO divides the
right hand side. Symmetrically the right hand side divides the left hand side. Thus pO
solve (1.1a).
An analogous argument shows that pO also satisfies the relation (1.1b).
(cid:3)
4. Reduction to Rank Two
In this section, unless otherwise stated, k is an arbitrary field of characteristic not two.
Proposition 4.1. Let p be a monic O-orbital solution to (1.1a), where O = Znq0 (q0 ∈
Irr(R)) and let i, j ∈ {1, . . . , n}, i 6= j, such that pi and pj are not constant. Then there
exists a pair of nonnegative integers (rij, sij) 6= (0, 0) such that
σrij
i σsij
j q0 = q0.
(4.1)
i
j
j
j f σ−1/2
j f σ−1/2
j f divides the RHS of the same equation. We have then that σ1/2
pj. Say that σ1/2
Proof. Let f be an irreducible factor of pi. Then σ1/2
j f divides the LHS of (1.1a), which
means that σ1/2
pi
j f σ−1/2
or σ1/2
pi, then σjf pi (in the other case we get that
σ1/2
i σ1/2
j f pj). Iterating the argument, we have a sequence {σrk
j f } of irreducible factors
of pi, and a sequence {σrℓ+1/2
f } of irreducible factors of pj, with rk, sk, rℓ, sℓ
nonnegative integers. Since R is a UFD, the sequence {σrk
: k ≥ 1} will have
repetitions, which implies that there is a pair of nonnegative integers (r, s) 6= (0, 0) such
that σr
j f = f . This argument can be repeated for all irreducible factors of pi and
pj and, by taking the least common multiple of the powers, we have that for each pair
(i, j), i, j ∈ {1, . . . , n}, i 6= j there is a pair of nonnegative integers (rij, sij) 6= (0, 0)
such that σrij
acts as the identity on all the irreducible factors of pi and pj. Since
O = Znq0, with q0 ∈ Irr(R), any irreducible factor qi of pi is in σ1/2
O, so it is of the form
qi = σ1/2
σsℓ+1/2
j
i σsk
j f
i σsk
i σsij
j
i σa1
1 · · · σan
n q0. Then
i σs
i
i
σrij
i σsij
σ1/2
i σa1
j σ1/2
i σa1
1 · · · σan
σrij
i σsij
1 · · · σan
n σrij
i σsij
σrij
i σsij
j qi = qi
n q0 = σ1/2
j q0 = σ1/2
j q0 = q0.
n q0
i σa1
i σa1
1 · · · σan
1 · · · σan
n q0
Proposition 4.2. Suppose char k = 0. Let p be a monic O-orbital solution to (1.1a), with
O = Znq0. Let i 6= j be two indices such that pi = 1 and pj 6= 1, then σiq0 = q0. Viceversa,
let i 6= j be two indices such that σiq0 = q0 and σjq0 6= q0, then pi = 1.
(cid:3)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 9
Proof. If there are two indices such that pi = 1 and pj 6= 1, then by (1.1a) we get
j pi = σ−1/2
σ1/2
i pjσ1/2
σ1/2
i pj = σ−1/2
i
i
pjσ−1/2
j
pi
pj
σipj = pj.
Since pj is non trivial, this implies that σiq0 = q0.
Viceversa, suppose that σiq0 = q0, and that there is an index j such that σjq0 6= q0, then
by (1.1a) we get
pjσ−1/2
j
pi
i
j pi = σ−1/2
j pi = pjσ−1/2
σ1/2
i pjσ1/2
pjσ1/2
σjpi = pi.
j
pi
Since pi is a product of shifts of q0, and σjq0 6= q0, then the only possiblity is that pi = 1,
because char k = 0.
(cid:3)
Lemma 4.3. Suppose char k = 0, and let σ(u) = u − β, that is σ(u1, . . . , um) = (u1 −
β1, . . . , um − βm). For q0 ∈ Irr(R)
σq0 = q0
if and only if
h∇q0, βi ≡ 0.
Here ∇q0 =(cid:16) ∂q0
∂u1
, . . . , ∂q0
∂um(cid:17) is the gradient and
h , i : Rm × km → Rm,
h(γk), (βk)i =
γkβk.
mXk=1
Proof. The irreducible polynomial q0 is invariant under the automorphism σ if and only
if q0(u − β) = q0(u), which is equivalent to q0(u + rβ) = q0(u) for all r ∈ Z. Since q0 is
a polynomial and char k = 0, this is the same as q0(u + tβ) = q0(u) for all t ∈ k, which
is indeed equivalent to h∇q0, βi ≡ 0 (as can be easily seen by considering q0(u + tβ) as a
polynomial function of the variable t).
(cid:3)
Theorem 4.4. Let char k = 0 and let p be a monic O-orbital solution to (1.1a), with
O = Znq0. Then there are at most two distinct indices 1 ≤ i ≤ n such that pi is not
constant and σiq0 6= q0.
Proof. By Proposition 4.1, for each pair of indices i, j such that pi and pj are not constant,
we have (4.1). By Lemma (4.3) we have that
σrij
i σsij
j q0(u) = q0(u − rijαi − sijαj) = q0(u1 − rijα1i − sijα1j, . . . , un − rijαni − sijαnj)
is equal to q0(u) if and only if
h∇q0, rijαi + sijαji = 0
rijh∇q0, αii + sijh∇q0, αji = 0
(4.2)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 10
Now suppose that i 6= j are such that σiq0 6= q0 6= σjq0, i.e., h∇q0, αii 6= 0 6= h∇q0, αji.
Then we have rij, sij > 0 and (4.2) becomes
rijh∇q0, αii + sijh∇q0, αji = 0
rijh∇q0, αii = −sijh∇q0, αji
h∇q0, αii = −
sij
rij
h∇q0, αji.
(4.3)
Now suppose that i, j, k are three distinct indices as in the statement of the Theorem,
then by (4.3), we have negative rational numbers r1, r2 and r3 such that
h∇q0, αii = r1h∇q0, αji
= r1(r2h∇q0, αki)
= r1r2(r3h∇q0, αii)
= (r1r2r3)h∇q0, αii
which is impossible.
(cid:3)
Notice that if there is a single index i such that for all other indices j 6= i we have pj = 1
and σjq0 = q0, then for any value of αi and any choice of pi we trivially get a solution.
As a consequence of Proposition 4.2 and Theorem 4.4, we have the following immediate
consequence.
Corollary 4.5. Suppose that char k = 0. For O = Znq0, any monic O-orbital nontrivial
solutions p of (1.1a) satisfies the following: there are two indices i, j ∈ {1, . . . , n}, i 6= j
such that pi 6= 1 6= pj, σiq0 6= q0 6= σjq0 and for all k ∈ {1, . . . , n} \ {i, j} we have pk = 1
and σkq0 = q0.
We can then also conclude as follows.
Corollary 4.6. For R = k[u1, . . . , um], char k = 0 and σi = u − αi, any monic solution p
to (1.1a) also satisfies (1.1b).
Proof. If p is a solution to (1.1a), then by Theorem 3.2 p = p(1) · · · p(k) with p(i) being a
monic Oi-orbital solution. By Corollary 4.5, each p(i) is either a trivial solution or it is only
nontrivial in two components. It follows immediately that for all 1 ≤ i ≤ k, p(i) trivially
satisfies (1.1b), hence so does p.
(cid:3)
5. Rank two solutions via higher spin 6-vertex configurations
In this section we generalize the rank two solutions obtained in [17] from univariate
complex polynomials to multivariate polynomials. In Theorem 5.2, k can be an arbitrary
field of characteristic not equal to two. To extend the results to higher rank in Corollary
5.3, we need to assume char k = 0. Although the explicit solutions are quite different, the
combinatorics from [17, 15] can still be adopted. Let R = k[u1, u2, . . . , um], and consider
two automorphisms σ1, σ2 of R given by σi(uj) = uj − αji for some m × 2-matrix α = (αij).
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 11
σ2(f )
σ1σ2(f )
σ1/2
2
(f )
σ1/2
1 σ1/2
2
(f )
f
σ1/2
1
(f )
σ1(f )
Figure 1. Part of a square lattice grid with some labels indicated.
1 σl
Let f be a monic irreducible element of R, and let O = {σk
2(f ) k, l ∈ Z} ∈ Irr(R)/Z2
be an orbit of irreducible elements of R with respect to the Z2-action given by σ1, σ2.
We think of elements of O as the midpoints of the faces of a square lattice, see Figure 1.
Applying σi corresponds to taking a unit step in the i:th direction. Thus taking a half-step
in the i:th coordinate direction takes us from the midpoint of a face to the (midpoint of)
an edge with normal vector in the i:th direction. We consider configurations where edges
have been assigned a non-negative multiplicity.
Definition 5.1. Fix O ∈ Irr(R)/Z2. Put Ei = Ei(O) = σ1/2
1 σ1/2
σ1/2
By a higher spin 6-vertex configuration L = (L1, L2) in O we mean a pair of functions
(O). Elements of Ei are edges of O and V12 are the vertices of O.
(O) and V12 = V12(O) =
i
2
Li : Ei → Z≥0 such that for any v ∈ V12:
L1(σ1/2
2
(v)) + L2(σ1/2
1
(v)) = L1(σ−1/2
2
(v)) + L2(σ−1/2
1
(v)).
(5.1)
L is finite if #{e ∈ Ei Li(e) 6= 0} < ∞ for i = 1, 2.
More generally, we can replace the index set {1, 2} in the above definition by an arbitrary
2-subset {i, j} ⊆ {1, 2, . . . , n}, in which case we say that L has index set {i, j}.
The following theorem shows that there is a bijective correspondence between the set of
monic O-orbital solutions p = (p1, p2) to (1.1a) and the set of finite higher spin 6-vertex
configurations L in O.
Theorem 5.2. (a) Given any finite higher spin 6-vertex configuration L in O, define
p1, p2 ∈ R by
pi = Ye∈Ei
Then p = (p1, p2) is a monic O-orbital solution to (1.1a).
f Li(e)
∀i ∈ {1, 2}
(5.2)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 12
(b) Conversely, given any monic O-orbital solution p = (p1, p2) to (1.1a), then there exists
a unique finite cell configuration L such that (5.2) holds.
Proof. Let i ∈ {1, 2}. Since R is a UFD we can uniquely factor pi as follows:
pi = ξi · Yg∈Irr(R)
gmi(g)
for some unit ξi and some non-negative integers mi(g). Since p is O-orbital, each irreducible
factor of pi belongs to Ei = σ1/2(O). In other words mi(g) = 0 if g /∈ Ei and hence we have
pi = ξi · Ye∈Ei
emi(e).
Define Li(e) = mi(e) and let L = (L1, L2). It remains to show that L is a finite higher
spin 6-vertex configuration. Substituting (5.2) into (1.1a) we get
Ye∈Ei
σ1/2
j
(e)Li(e) Ye∈Ej
σ1/2
i σ1/2
j
(e)Lk(e)σ−1/2
i
σ−1/2
j
σ1/2
i
(e)Lj (e) = Ye∈Ei
(e)Lk(e) = Ye∈Ek
Ye∈Ek
σ−1/2
j
(e)Li(e) Ye∈Ej
σ−1/2
i
(e)Lj (e)
(5.3)
σ1/2
i σ−1/2
j
(e)Lk(e)σ−1/2
i
σ1/2
j
(e)Lk(e)
(5.4)
Making appropriate substitutions to write each side as a product over f ∈ O and identifying
powers of f in each side, the claim follows.
(cid:3)
Corollary 5.3. Suppose char k = 0. Let m and n be positive integers. Let R = k[u1, u2, . . . , um],
(αij) be an m × n-matrix, σi(uj) = uj − αji, p = (p1, p2, . . . , pn) be a monic O-orbital solu-
tion to (1.1a). Then there exists two distinct indices i, j ∈ {1, 2, . . . , n} and a finite higher
spin 6-vertex configuration L with index set {i, j} such that
pi = Ye∈Ei
eLi(e),
pj = Ye∈Ej
f Lj(e),
pk = 1 for k /∈ {i, j}.
Proof. Apply Corollary 4.5 and Theorem 5.2.
(cid:3)
We summarize the results of the paper in the following statement.
Main Theorem. Let m and n be positive integers, n ≥ 2. Let k be a field of characteristic
zero. Let R = k[u1, u2, . . . , um] and let (αij) ∈ Mm×n(k).
Let (p1, p2, . . . , pn) ∈ Rn be an n-tuple of monic polynomials satisfying the following
system of equations:
2 ) ∀i 6= j,
where u = (u1, u2, . . . , um) and αi = (α1i, α2i, . . . , αmi). Then there exist
2 ) · pj(u − αi
2 ) · pj(u + αi
2 ) = pi(u + αj
pi(u − αj
(5.5)
(i) a finite set of (distinct) orbits O(1), O(2), . . . , O(d) ∈ Irr(R)/Zn in the set of monic ir-
reducible elements Irr(R) with respect to the action of Zn by k-algebra automorphisms
(ii) for each orbit O(a), a 2-subset {ia, ja} ⊆ {1, 2, . . . , n} such that for all k ∈ {1, 2, . . . , n}\
on R determined by x.uj = uj −Pn
{ia, ja} we have ek.q = q for all q ∈ O(a);
i=1 xiαji for all x =Pn
i=1 xkek ∈ Zn;
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 13
(iii) for each orbit O(a), a finite higher spin 6-vertex configuration L(a), with index set
{ia, ja}, in O(a) such that
k
k · · · p(d)
f L(a)
k (f ),
pk = p(1)
k p(2)
Yf ∈Ek
1,
p(a)
k =
∀k ∈ {1, 2, . . . , n},
if k ∈ {ia, ja},
otherwise.
(5.6a)
(5.6b)
Conversely, given orbits O(a), 2-subsets {ia, ja} and vertex configurations L(a) as in (i) -- (iii),
then (p1, p2, . . . , pn) defined by (5.6), satisfies equations (5.5).
Moreover, the following relations hold (non-vacuously when n ≥ 3):
pk(u − αi
2 ) = pk(u − αi
2 )pk(u + αi
2 )pk(u + αi
2 + αj
2 + αj
2 − αj
2 − αj
2 ) ∀i 6= j 6= k 6= i. (5.7)
where u = (u1, u2, . . . , um) and αi = (α1i, α2i, . . . , αmi).
Proof. By combining Theorem 3.2 and Corollary 5.3. The last claim then follows from
Corollary 4.6.
(cid:3)
5.1. Examples.
Example 5.4. Let (m, n) = (2, 3), α =(cid:2) −1 1 0
0 −1 1(cid:3), R = C[u1, u2] and σi(uj) = uj − αji.
Then
2 , (u1 + 1
2 )(u2 − 1
2 ), u2 + 1
is a solution to the consistency equations (1.1) related to the Lie algebra gl3, see [27]. It
is a monic solution and its factorization into orbital solutions is p = p(1)p(2) where p(i) is
(Z3 · ui)-orbital and given by
p(1) = (u1 − 1
p(2) = (1, u2 − 1
(5.8)
2 , u1 + 1
2 , 1),
2 , u2 + 1
2 ).
p = (p1, p2, p3) =(cid:0)u1 − 1
2(cid:1)
Also note that {i1, j1} = {1, 2} and {i2, j2} = {2, 3}.
Example 5.5. Let (m, n) = (3, 4), R = Q[u1, u2, u3],
f (u1, u2, u3) = (u2 + u3)2 − (u3
1 − u1 + 1),
Notice that we have σ3f = σ4f = f = σ3
1σ2
configuration, with index set {1, 2}, given by
2f . We can then define a higher spin 6-vertex
0
0
1 −3
3 .
2 −1
2 −3
4 −5
−2
α =
1 σ2(f )(cid:9)
(f ), σ3/2
L1(e) =(1,
L2(e) =(1,
1
0, otherwise
if e ∈(cid:8)σ1/2
if e ∈(cid:8)σ1σ1/2
2
0, otherwise
(f ), σ2
1σ3/2
2
(f ), σ3
1σ3/2
2
(f )(cid:9)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 14
1σ3/2
σ2
2
(f )
1σ3/2
σ3
2
(f )
σ3/2
1 σ2(f )
σ1σ1/2
2
(f )
σ1/2
1
(f )
σ1(f )
Figure 2. Diagram depicting a lattice configuration L corresponding to
the solution from Example 5.5. Edges e for which Li(e) = 1 are blue
(solid).
From Figure 2, it is clear that this configuration satisfies (5.1) and hence corresponds to
an orbital solution, which can be written as p = (p1, p2, p3, p4), where
p1 = σ1/2
(f )σ3/2
1
1 σ2(f )
2 α11, u2 − 1
p2 = σ1σ1/2
2
(f ) · σ2
2
2 α21, u3 − 1
2 α11 − α12, u2 − 3
2 α31(cid:1) · f(cid:0)u1 − 3
= f(cid:0)u1 − 1
=(cid:0)(u2 + u3 − 1)2 − ((u1 − 1)3 − u1 + 2)(cid:1) ·(cid:0)(u2 + u3 − 1)2 − ((u1 + 1)3 − u1)(cid:1)
=(cid:0)(u2 + u3 + 3
·(cid:0)(u2 + u3 − 5
2 )(cid:1) ·(cid:0)(u2 + u3 + 1
2 )(cid:1)
1σ3/2
(f ) · σ3
2 )2 − ((u1 + 3
2 )2 − ((u1 − 5
1σ3/2
2 )3 − u1 − 1
2 )3 − u1 + 5
p4 = 1.
p3 = 1,
(f )
2
2 )2 − ((u1 + 1
2 )3 − u1 − 3
2 )(cid:1)
2 α21 − α22, u3 − 3
2 α31 − α32(cid:1)
6. Irreducible factorization of multiquiver solutions
In this section we show how the solutions to the consistency equations obtained in [18]
fit into our classification scheme.
6.1. Input matrix. Let β = (βij) be an m × n-matrix satisfying the following conditions:
βij ∈ Z for all i, j,
βijβik ≤ 0 for all i, j, k with j 6= k.
(6.1)
(6.2)
Condition (6.2) is equivalent to
Every row of β contains at most one positive and at most one negative integer.
(6.3)
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 15
Moreover β can be interpreted as the incidence graph of certain multiquivers, see [18] for
details.
6.2. Solution attached to β. Let R = k[u1, u2, . . . , um] and σ = (σ1, σ2, . . . , σn) be given
by σi(uj) = uj − βji and define p = (p1, p2, . . . , pn) ∈ Rn by
pi(u1, u2, . . . , um) =
pji(uj),
mYj=1
uj(uj + 1) · · · (uj + βji − 1)
1
(uj − 1)(uj − 2) · · · (uj − βji)
if βji > 0,
if βji = 0,
if βji < 0.
(6.4)
(6.5)
where
pji(uj) =
Theorem 6.1 ([18]). σ and p satisfy the TGWA consistency relations from [7]. That is,
σiσj(pipj) = σi(pi)σj(pj) ∀i 6= j,
σiσk(pj)pj = σi(pj)σk(pj) ∀i 6= j 6= k 6= i.
(6.6a)
(6.6b)
6.3. Symmetrized version. We write (6.6) in the symmetric form. Applying σ−1/2
i
σ−1/2
j
i
i
σ1/2
j
(pi) gives
to equation (6.6a) and putting epi = σ1/2
(epi)σ1/2
That is,ep := (ep1,ep2, . . . ,epn) satisfies the symmetrized binary TGWA consistency equation.
(epj)σ−1/2
which shows that ep also satisfies the symmetrized ternary TGWA consistency equation.
In [18], relation (6.6b) was proved separately. But in fact, as we showed in Corollary 4.6,
equation (6.8) actually follows from (6.7), hence (6.6b) follows from (6.6a).
(epj) = σ−1/2
(epj) = σ1/2
Similarly applying σ−1/2
σ1/2
i σ1/2
(epj)
(epj)σ−1/2
to equation (6.6b) gives the equivalent relation
(epi)σ−1/2
(epj)
i σ−1/2
σ−1/2
k
σ−1/2
k
σ1/2
k
σ1/2
j
(6.8)
(6.7)
i
k
k
j
i
i
i
Explicitly, we have
(pi) = pi(cid:0)u1 −
pji(uj −
βji)
1
2
i
epi = σ1/2
mYj=1
=
β1i
2
, u2 −
β2i
2
, . . . , um −
βmi
2 (cid:1)
To write a closed formula we also make an overall shift by 1/2. Put qji(uj) = pji(uj −
2 βji + 1
1
2 ). Then
2 βji + 1
2 )(uj − 1
2 βji + 3
2 ) · · · (uj + 1
2 βji − 1
2 ),
if βji > 0
if βji = 0
if βji < 0
(6.9)
qji(uj) =
(uj − 1
1,
(uj − 1
2 βji + 3
We can summarize these observations as follows.
2 βji + 1
2 )(uj − 1
2 ) · · · (uj + 1
2 βji − 1
2 ),
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 16
Proposition 6.2. Let β be an m × n matrix satisfying conditions (6.1) and (6.2). Put
(6.10)
(6.11)
Define polynomials for (i, j) ∈ J1, nK × J1, mK:
1
2
(βji − 1)
eβji =
then for all j ∈ J1, mK the n-tuple
qji(uj) = (uj −eβji)(uj −eβji + 1) · · · (uj +eβji)
qj = (qj1(uj), qj2(uj), . . . , qjn(uj))
is a solution to the consistency equations (1.1).
Proof. It suffices to observe that if j 6= j′, then
and
(qj1(uj), qj2(uj), . . . , qjn(uj))
(qj ′1(uj ′), qj ′2(uj ′), . . . , qj ′n(uj ′))
belong to different orbits with respect to the action of Zn via σ. More precisely, given
j 6= j′ in J1, mK and i, i′ ∈ J1, nK then there are no integers ks such that some irreducible
factor of σk1
n (qji) divides qj ′i′. This is obvious since the former is a univariate
polynomial in uj and the latter a univariate polynomial in uj ′.
(cid:3)
2 · · · σkn
1 σk2
6.4. Factorization into irreducibles. By [17, Prop. 6.2], for each j ∈ J1, mK the solution
qj factors into a product of gcd(βji1, βji2) solutions, where {i1, i2} = {i ∈ J1, nK βji 6= 0}
(assume without loss of generality that all rows in β are nonzero). So in total the solution
from [18], corresponding to a matrix β, factors into
gcd(βj1, βj2, . . . , βjn)
mYj=1
orbits. Each orbit factor is actually irreducible and corresponds to a (unique) zero area
generalized Dyck path. Explicitly, let γj = gcd(βj1, βj2, . . . , βjn). We have
(qj1(uj), qj2(uj), . . . , qjn(uj)) = (1, . . . , qji1(uj), . . . , qji2(uj), . . . , 1)
i.e. all entries are 1 except possibly 2 places. Then, following [17], we can further factor
(qji1(uj), qji2(uj)) =
(q(k)
ji1
γj −1Yk=0
(uj), q(k)
ji2
(uj))
where q(k)
factors q(k)
ji (uj) is the product of all factors of the form (uj − l) where l ≡ k (mod γj). The
ij (uj) were expressed using zero area generalized Dyck paths in [17].
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 17
References
[1] V.V. Bavula, Generalized Weyl algebras and their representations, Algebra i Analiz 4 Issue 1 (1992)
75 -- 97; English transl. in St Petersburg Math. J. 4 (1993) 71 -- 92.
[2] V.V. Bavula, D.A. Jordan, Isomorphism problems and groups of automorphisms for generalized Weyl
algebras, Trans. Amer. Math. Soc. 353 Issue 2 (2000) 769 -- 794.
[3] G. Benkart, M. Ondrus, Whittaker modules for generalized Weyl algebras, Represent. Theory 13 (2009)
141 -- 164.
[4] T. Brzezi´nski, Noncommutative Differential Geometry of Generalized Weyl Algebras Symmetry, Inte-
grability and Geometry: Methods and Applications SIGMA 12 (2016), 059, 18 pages.
[5] P.A.A.B. Carvalho, S.A. Lopes Automorphisms of Generalized Down-Up Algebras, Communications in
Algebra 37 Issue 5 (2009).
[6] T. Cassidy, B. Shelton, Basic properties of generalized down-up algebras, J. Algebra 279 No. 1 (2004)
402 -- 421.
[7] V. Futorny, J.T. Hartwig, On the consistency of twisted generalized Weyl algebras Proc. Amer. Math.
Soc. 140 (2012) 3349 -- 3363; arXiv:1103.4374.
[8] V. Futorny, J.T. Hartwig, Multiparameter twisted Weyl algebras, J. Algebra 357 (2012) 69 -- 93.
[9] J. Gaddis, Isomorphisms of some quantum spaces, in: Ring Theory and its Applications, Contemporary
Mathematics, Vol. 609, American Mathematical Society, Providence, RI, 2014, pp. 107 -- 116.
[10] J. Gaddis, The isomorphism problem for quantum affine spaces, homogenized quantized Weyl algebras,
and quantum matrix algebras, Journal of Pure and Applied Algebra 221 (2017) 2511 -- 2524.
[11] K.R. Goodearl, J.T. Hartwig, The isomorphism problem for multiparameter quantized Weyl algebras,
Sao Paulo J. Math. Sci. 9 Issue 1 (2015) 53 -- 61.
[12] J.T. Hartwig, Locally finite simple weight modules over twisted generalized Weyl algebras, J. Algebra
303 Issue 1 (2006) 42 -- 76.
[13] J.T. Hartwig, Twisted generalized Weyl algebras, polynomial Cartan matrices and Serre-type relations,
Comm. Algebra 38 Issue 12 (2010) 4375 -- 4389.
[14] J.T. Hartwig, Unitarizability of weight modules over noncommutative Kleinian fiber products, Preprint
arXiv:1702.08168v1.
[15] J.T. Hartwig, Noncommutative singularities and lattice models, J. Algebra 508 (2018) 35 -- 80.
[16] J.T. Hartwig, J. Oinert, Simplicity and maximal commutative subalgebras of twisted generalized Weyl
algebras, J. Algebra 373 (2013) 312 -- 339.
[17] J.T. Hartwig, D. Rosso, Cylindrical Dyck paths and the Mazorchuk-Turowska equation, J. Algebraic
Combin. 44 Issue 1 (2016), 223 -- 247.
[18] J.T. Hartwig, V. Serganova, Twisted Generalized Weyl Algebras and Primitive Quotients of Enveloping
Algebras Algebr. Represent. Theory 19 Issue 2 (2016) 277 -- 313.
[19] D.A. Jordan, Primitivity in skew Laurent polynomial rings and related rings, Math. Z. 213 (1993)
353 -- 371.
[20] D. A. Jordan A simple localization of the quantized Weyl algebra J. Algebra 174 Issue 1 (1995) Pages
267 -- 281.
[21] A. P. Kitchin and S. Launois On the automorphisms of quantum weyl algebras, Journal of Pure and
Applied Algebra 223 (2019) 1514 -- 1530.
[22] G. Maltsiniotis, Groupes quantiques et structures diff´erentielles, C. R. Acad. Sci. Paris S´er. I Math.
311 No. 12 (1990) 831 -- 834.
[23] V. Mazorchuk, M. Ponomarenko, L. Turowska, Some associative algebras related to U (g) and twisted
generalized Weyl algebras, Math. Scand. 92 Issue 1 (2003) 5 -- 30.
[24] V. Mazorchuk, L. Turowska, Simple weight modules over twisted generalized Weyl algebras, Comm.
Algebra 27 Issue 6 (1999) 2613 -- 2625.
[25] V. Mazorchuk, L. Turowska, ∗-Representations of twisted generalized Weyl algebras, Algebras and
Representation Theory 5 (2002) 163 -- 186.
CLASSIFICATION OF TWISTED GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL RINGS 18
[26] A.L. Rosenberg, Noncommutative algebraic geometry and representations of quantized algebras, Kluwer,
1995.
[27] A. Sergeev, Enveloping algebra U (gl(3)) and orthogonal polynomials in several discrete indeterminates
in: "Noncommutative structures in mathematics and physics" Proc. NATO Advanced Research Work-
shop, Kiev 2000, Kluwer (2001) 113 -- 124; arXiv:math/0202182.
[28] M. Surez-Alvarez, Q. Vivas, Automorphisms and isomorphisms of quantum generalized Weyl algebras,
Journal of Algebra, Volume 424, 15 February 2015, Pages 540 -- 552.
[29] X. Tang Automorphisms for some "symmetric" multiparameter quantized Weyl algebras and their
localizations, arXiv:1610.01848v1 [math.RA].
[30] X. Tang, The automorphism groups for a family of generalized Weyl algebras, Journal of Algebra and
Its Applications Vol. 17, No. 08, 1850142 (2018).
Department of Mathematics, Iowa State University, Ames, IA-50011, USA
E-mail address: [email protected]
Department of Mathematics and Actuarial Science, Indiana University Northwest, Gary,
IN-46408, USA
E-mail address: [email protected]
|
1512.06813 | 2 | 1512 | 2018-11-08T20:49:50 | Closed Systems of Invertible Maps | [
"math.RA",
"cs.LO",
"math.LO"
] | We generalise clones, which are sets of functions $f:A^n \rightarrow A$, to sets of mappings $f:A^n \rightarrow A^m$. We formalise this and develop language that we can use to speak about it. We then look at bijective mappings, which have connections to reversible computation, which is important for physical (e.g. quantum computation) as well as engineering (e.g. heat dissipation) reasons. We generalise Toffoli's seminal work on reversible computation to arbitrary arity logics. In particular, we show that some restrictions he found for reversible computation on alphabets of order 2 do not apply for odd order alphabets. For $A$ odd, we can create all invertible mappings from the Toffoli 1- and 2-gates, demonstrating that we can realise all reversible mappings from four generators. We discuss various forms of closure, corresponding to various systems of permitted manipulations. These correspond, amongst other things, to discussions about ancilla bits in quantum computation. | math.RA | math |
Closed Systems of Invertible Maps
TIM BOYKETT1(cid:63)†
Institute for Algebra, Johannes Kepler University Linz, Austria
and Time's Up Research, Linz, Austria
We generalise clones, which are sets of functions f : An → A,
to sets of maps f : An → Am. We formalise this and develop
language that we can use to speak about such maps. In particular
we look at bijective mappings, which model the logical gates of
reversible computation. Reversible computation is important for
physical (e.g. quantum computation) as well as engineering (e.g.
heat dissipation) reasons. We generalise Toffoli's seminal work
on reversible computation to multiple valued logics. In particu-
lar, we show that some restrictions Toffoli found for reversible
computation on alphabets of order 2 do not apply for odd order
alphabets. For A odd, we can create all invertible mappings from
the Toffoli 1- and 2-gates, demonstrating that we can realise all
reversible mappings from four generators. We discuss various
forms of closure, corresponding to various systems of permitted
manipulations. This leads, amongst other things, to discussions
about ancilla bits in quantum computation.
Key words: Toffoli gate, reversible computation, clones, multiple-valued
logic, general algebra, mappings, iterative algebras
1
INTRODUCTION
The goal of this paper is to generalise Toffoli's fundamental results about
reversible computation. There are several motivations for an interest in re-
versible computation, some related to physics and engineering, others of a
(cid:63) email: [email protected] and [email protected]
† Research supported by the Austrian Science Fund (FWF) research grants P24077, P24285
and Y464-N18.
1
more algebraic nature. The engineering perspective notes that destroying
information creates entropy and thus heat, which is bad for circuitry and
wasteful[7]. The physics perspective notes that fundamental physical pro-
cesses such as inelastic collisions and quantum processes are reversible[12].
An algebraic perspective notes that we can say more about groups than about
semigroups owing to the existence of inverses.
Toffoli [16, 17] introduced what later became known as the Toffoli gate,
a simple yet universal basic element for computation with reversible binary
logic. This paper is mostly concerned with the generalisation of this logic to
multiple-valued logics. One main result is the generalisation of Toffoli gates
to larger alphabets and the demonstration that the multiple-valued logics of
odd arity are in some sense more powerful (Section 5).
We use language based upon the function algebras of [13], also known
as clones[15]. Mal'cev [11] introduced the concept of an iterative algebra,
a generalised clone.
If an iterative algebra includes the projections πn
:
i
(x1, . . . , xn) (cid:55)→ xi, then it is a clone. In the next sections we will intro-
duce some similar terminology to deal with arbitrary mappings, in particular
bijections on An. Later we will see that linear term algebras, the "reduct of
iterative algebras" that do not have the ∆ operator, play an important role.
In Section 2 we introduce mappings and operations upon them, to show
that there are several ways to talk about compositional closure. In particular
we show that the algebra of maps can be expressed using a finite signature of
finitary operations (Theorem 2.13). We look at the ways that a given set of
maps can generate a closed system of mappings, then at ways of embedding
other mappings into these closed systems. In Section 3 we talk about realising
one mapping in a multiclone generated by some other mappings.
We show that all invertible mappings on a set of odd order can be gener-
ated by four small Toffoli gates and using a definition of generation that is
important from an engineering perspective, show that these are enough for all
alphabets. In Section 6 look at the various forms of functional closure that
arise in the discussion.
2 FUNCTION TERMINOLOGY
We use N for the set of positive integers. Given a set A, O(A) = {f :
An → A n ∈ N} is the set of all functions on A. We will use the notation
of [10]. The operations of an iterative algebra are τ, ζ, ∆,∇,∗, defined as
follows. Let f be an n-ary function, g an m-ary function. The operations
τ and ζ permute variables, (τ f )(x1, . . . , xn) = f (x2, x1, x3, . . . , xn) and
2
(ζf )(x1, . . . , xn) = f (x2, x3, . . . , xn, x1), with both being the identity on
unary functions. The operation ∆ identifies variables, so (∆f )(x1, . . . , xn−1) =
f (x1, x1, x2, . . . , xn−1), while ∆ is the identity on unary functions. The op-
eration ∇ introduces a dummy variable, (∇f )(x1, x2, . . . , xn+1) = f (x2, . . . , xn+1).
Lastly we have composition, so
(f ∗ g)(x1, . . . , xn+m−1) = f (g(x1, . . . , xm), xm+1, . . . , xn+m−1).
The full iterative algebra is (O(A); τ, ζ, ∆,∇,∗) and an iterative algebra
on A is a subuniverse of this. If an iterative algebra includes the projections
: (x1, . . . , xn) (cid:55)→ xi, then it is a clone on A. The full linear term algebra
πn
i
is (O(A); τ, ζ,∇,∗) and a linear term algebra on A is a subuniverse of this.
We use SA to denote the permutations of a set A, Sn to denote the set
of bijections on {1, . . . , n} and S(N) = ∪n∈NSn. Juxtaposition multiplies
permutations from left to right, so (123)(12) = (1)(23), thus in order to avoid
difficulties when we write permutations as functions, we introduce clarifying
parentheses. Thus if α = (123) and β = (12) then (αβ)(1) = 1αβ =
(1α)β = 1 but α ◦ β(1) = α(β(1)) = (1β)α = 3.
Let A be a nonempty set. For n, m ∈ N, an (n, m)-map is a mapping f :
An → Am. We call n the arity, m the co-arity. We write Mn,m(A) = {f :
n,m Mn,m(A) is the set of all maps on A. Let
f ∈ Mn,m(A), 1 ≤ i ≤ m. We define fi : An → A to be the i-th projection
of f, so we can write f = (f1, . . . , fm). A function is an (n, 1)-map. An
(n, n)-map, with arity equal to co-arity, will be called balanced. Furthermore,
we define Bn,m(A) as the set of all (n, m)-maps that are bijections, so for A
finite n (cid:54)= m implies that Bn,m(A) = ∅ and the bijective maps are balanced.
We will write Bn(A) for the balanced bijections of arity n. Similarly we write
n Bn(A)
For example the map f (x, y) = (x + y, x − y) on an abelian group is a
n,m Bn,m(A) is the set of all bijections on A, B(A) =(cid:83)
An → Am}. Then M (A) =(cid:83)
B(A) =(cid:83)
if A is finite.
balanced (2, 2)-map, it is a bijection if the group is of odd order.
Definition 2.1 Let A be a set. Let f : Am → An , t ≤ n, I = (i1, . . . , it) ∈
{1, . . . , n}t be a tuple without repetitions. Define µ(I, f ) ∈ Mm,t(A) com-
ponentwise as (µ(I, f ))j = fij , for all j ∈ {1, . . . , t}. That is, µ(I, f ) is
obtained from f by projecting the output to the components indexed by the
tuple I.
Let i ∈ {1, . . . , m} and a ∈ A, define κ(i, a, f ) ∈ Mm−1,n(A) by
κ(i, a, f )(x1, . . . , xm−1) = f (x1, . . . , xi−1, a, xi, . . . , xm−1).
3
for all x1, . . . , xm−1 ∈ A. That is, κ(i, a, f ) is obtained from f by substitut-
ing the constant a for the i-th variable.
We extend κ to allow the substitution of constants for several variables si-
multaneously. Let r ≤ m, I ∈ {1, . . . , m}r increasing so that I = (i1, . . . , ir)
with i1 < i2 < ··· < ir and a = (a1, . . . , ar) ∈ Ar, define κ(I, a, f ) ∈
Mm−r,n(A) by the following recursion. Let κ((), (), f ) = f, for nonempty I
define
κ((i1, . . . , ir), (a1, . . . , ar), f ) = κ(i1, a1, κ((i2, . . . , ir), (a2, . . . , ar), f ))
Finally we introduce the following shorthand notation. For I ∈ {1, . . . , n}t
and I(cid:48) ∈ {1, . . . , m}r, tuples as above, and for o ∈ A, define
I(cid:48),I = µ(I, κ(I(cid:48), (o, . . . , o), f ))
f o
I(cid:48),I ∈ Mm−r,t(A)
Then f o
(2),(1)(z) =
For example, take the map f (x, y) = (x+y, x−y) on Z7, then f 5
µ((1), κ((2), (5), f ))(z) = z + 5 is a unary function on Z7.
2.1 Operations
In this section we look at the operations that allow us to combine mappings.
Our main result is that all these combinations can be written as a finite number
of finitary operations.
Definition 2.2 Let A be a set. Let f ∈ Mn,s(A), g ∈ Mm,t(A) be mappings.
Define i1 ∈ B1(A) with i1(x) = x, write in ∈ Bn(A) for the identity
function on An.
We place two mappings next to one another to form a wider mapping.
f ⊕ g : An+m →As+t
(x1, . . . xn+m) (cid:55)→ (f1(x1, . . . , xn), . . . , fs(x1, . . . , xn),
g1(xn+1, . . . , xn+m), . . . , gt(xn+1, . . . , xn+m))
We compose mappings. For k ∈ N with k ≤ n, k ≤ t,
◦k(f, g) : An+m−k →As+t−k
(x1, . . . , xn+m−k) (cid:55)→(f1(g1(x1, . . . , xm), . . . , gk(x1, . . . , xm), xm+1, . . . , xm+n−k),
...
fs(g1(x1, . . . , xm), . . . , gk(x1, . . . , xm), xm+1, . . . , xm+n−k),
gk+1(x1, . . . , xm), . . . , gt(x1, . . . , xm))
4
For k ≥ 2, ◦k is a partial operation.
We identify variables.
∆f : An−1 → As
∆f (x1, . . . xn−1) = f (x1, x1, . . . , xn−1)
and define ∆f = f if f has arity 1.
We introduce dummy variables.
∇f : An+1 → As
∇f (x1, . . . , xn+1) = f (x2, . . . , xn+1)
The following operations allow us to permute the inputs and outputs of a
given mapping.
τ f (x1, . . . , xn) =f (x2, x1, x3, . . . , xn)
ζf (x1, . . . , xn) =f (x2, x3, . . . , xn, x1)
¯τ f (x1, . . . , xn) = (f2(x1, x2, x3, . . . , xn),
f1(x1, x2, x3, . . . , xn),
...
fs(x1, x2, x3, . . . , xn))
¯ζf (x1, . . . , xn) = (f2(x1, x2, x3, . . . , xn),
f3(x1, x2, x3, . . . , xn),
...
fs(x1, x2, x3, . . . , xn),
f1(x1, x2, x3, . . . , xn)).
In the case that f has arity 1, τ f = ζf = f. In the case that f has co-arity
1, ¯τ f = ¯ζf = f.
For example, in the case that n = t, f ◦n g is the composition that we
would expect:
f ◦n g = (f1(g1(x1, . . . , xm), . . . , gn(x1, . . . , xm)),
...
fs(g1(x1, . . . , xm), . . . , gn(x1, . . . , xm)))
5
Note that there are many relations amongst these operations. For instance
let f have co-arity n, then ¯τ f = (τ in) ◦n f and ¯ζf = (ζin) ◦n f. The µ
operator introduced above allows us to say µ((2, . . . , n + 1),∇f ) = f. We
will not examine the axiomatics of these operations in detail. In order to keep
our notation clear, it is important to prove the following result.
Lemma 2.3 The operations ◦k are associative as partial operations, that is,
for all f, g, h ∈ M (A), for all k, l ∈ N,
f ◦k (g ◦l h) = (f ◦k g) ◦l h
(1)
when both sides exist.
xag+ah−l+1, . . . , xaf +ag+ah−(l+k))
if i ≤ cf
if cf < i
Proof: In order to shorten notation, we will write af for the arity of f and cf
for the co-arity of f.
Suppose both sides of equation (1) exist. Then the two sides of (1) have
equal arity and co-arity, that is arity af + ag + ah − (k + l) and co-arity
cf + cg + ch − (k + l). To exist we must have k ≤ min(af , cg + ch − l, cg)
and l ≤ min(ag, ch, af + ag − k). Then l ≤ ch so cg ≤ cg + ch − l
so k ≤ min(af , cg) and similarly l ≤ min(ag, ch). We calculate, for any
1 ≤ i ≤ cf + cg + ch − (k + l).
(f ◦k (g ◦l h))i(x1, . . . , xaf +ag+ah−(k+l))
(g ◦l h)k+i−cf (x1, . . . , xag+ah−l)
fi(g1(h1(x1, . . . , xah ), . . . , hl(x1, . . . , xah), xah+1, . . . , xag+ah−l), . . .
fi((g ◦l h)1(x1, . . . , xag+ah−l), . . . , (g ◦l h)k(x1, . . . , xag+ah−l),
(f ◦k g)i(h1(x1, . . . , xah ), . . . , hl(x1, . . . , xah),
while
((f ◦k g) ◦l h)i(x1, . . . , xaf +ag+ah−(k+l))
if i ≤ cf + cg − k
if i > cf + cg − k
. . . , gk(h1(x1, . . . , xah), . . . , hl(x1, . . . , xah), xah+1, . . . , xag+ah−l),
xag+ah−l+1, . . . , xaf +ag+ah−(l+k))
if i ≤ cf
gk+i−cf (h1(x1, . . . , xah), . . . , hl(x1, . . . , xah), xah+1, . . . , xag+ah−l)
xah+1, . . . , xaf +ag+ah−(k+l))
hl+i−(cf +cg−k)(x1, . . . , xah )
hl+i−(cf +cg−k)(x1, . . . , xah )
if cf < i ≤ cf + cg − k
if cf + cg − k < i
=
=
=
6
=
fi(g1(h1(x1, . . . , xah ), . . . , hl(x1, . . . , xah ), xah+1, . . . , xag+ah−l), . . .
. . . , gk(h1(x1, . . . , xah ), . . . , hl(x1, . . . , xah), xah+1, . . . , xag+ah−l),
xag+ah−l+1, . . . , xaf +ag+ah−(k+l))
if i ≤ cf
gk+i−cf (h1(x1, . . . , xah ), . . . , hl(x1, . . . , xah), xah+1, . . . , xag+ah−l)
if cf < i ≤ cf + cg − k
if i > cf + cg − k
hl+i−(cf +cg−k)(x1, . . . , xah )
(cid:3)
These can be seen to be equal and we are done.
Note that one side of the associativity equation in the previous lemma can
exist without the other existing, for instance if k = l = 2, f, h ∈ M2,3(A)
and g ∈ M2,1(A) then f ◦k (g ◦l h) exists while (f ◦k g) ◦l h does not.
We can look at a general form of mapping composition corresponding to
clones. Remember that to be closed under a 0-ary operation such as i1 means
to include that constant.
Definition 2.4 Let A be a set. A collection of mappings on A is called a
multiclone on A if it is closed under the operations {i1,⊕, ∆,∇, τ, ζ} and
the partial operations {◦i i ∈ N}. Let F ⊆ M (A) be a collection of
mappings. Then we write C∆(F ) for the closure under the operations, i.e.
the multiclone generated by F .
However in order to focus on the realm of bijective mappings, we are
predominately interested in closure without the ∆ and ∇ operations.
Definition 2.5 Let A be a set. A collection of mappings on A is called a
Toffoli algebra on A if it is closed under the operations {i1,⊕, τ, ζ} and the
partial operations {◦i i ∈ N}. Let F ⊆ M (A) be a collection of mappings.
Then we write C(F ) for the closure under the operations, i.e.
the Toffoli
algebra generated by F .
Note that by the comments following Definition 2.2 we can express the ¯τ
and ¯ζ operations in terms of these operations, thus not losing any expressive
power.
We note that there are similarities to the Network Algebra used in e.g.
[2, 14]. We note that the expression Toffoli Algebra has been used in [8] in a
related but distinct form.
These operations reflect what Toffoli [16] calls a combinatorial network,
as well as corresponding to the idea of a read once function as used in in [5].
7
In particular, every output is used at most once as an input to another mapping,
avoiding the duplication of variables or so-called fan-out, where one data
signal is spread to two or more outputs. The mapping φn ∈ M1,n(A) with
φn(a) = (a, . . . , a) is a fan-out mapping. Also, no input or output is thrown
away. As a connection between these concepts, we have the following.
Lemma 2.6 Let F be a Toffoli algebra. Then F is a multiclone iff φ2 ∈ F
and g ∈ F where φ2(x) = (x, x), g(x, y) = y.
Proof: (⇒): We know that i2 = i1 ⊕ i1 ∈ F . We know F is a multiclone, so
∆i2 ∈ F . But ∆i2(x2) = (x2, x2) = φ2(x2) so φ2 ∈ F . Similarly ∇i1 = g.
(⇐): Suppose φ2 ∈ F . Let h ∈ F ∩ Mn,m with n ≥ 2. Then
(h ◦2 φ2)(x1, . . . , xn−1) = h(x1, x1, . . . , xn−1)
= ∆h(x1, . . . xn−1)
so F is closed under ∆.
Let h ∈ F ∩Mn,m. We calculate that h◦1g ∈ Mn+2−1,m+1−1 = Mn+1,m
(h ◦1 g)(x1, . . . , xn+1) = h(g1(x1, x2), x3, . . . , xn+1)
= h(x2, . . . , xn+1)
= ∇h(x1, . . . , xn+1)
so F is also closed under ∇, thus F is a multiclone.
Definition 2.7 A Toffoli algebra is called a reversible Toffoli algebra if each
of its elements element is bijective.
(cid:3)
Note that i1 ⊕ i1 ⊕ . . . ⊕ i1 = in. We will also use ⊕ to concate-
nate tuples, (x1, . . . , xn) ⊕ (y1, . . . , ym) = (x1, . . . , xn, y1, . . . , ym). Let
α ∈ Sn be a permutation. Then let πα be the function πα(x1, . . . , xn) =
(xα−11, . . . , xα−1n).
Example 2.8 The set {πα α ∈ Sn, n ∈ N} is a reversible Toffoli algebra,
in fact it is the unique minimal Toffoli algebra on any set A.
Other multiclones and reversible Toffoli algebras arise from linear algebra.
Example 2.9 Let A be a field. Then the collection of linear mappings on
vector spaces over A form a multiclone. The collection of invertible linear
mappings ∪n∈NGL(n, A) forms a reversible Toffoli algebra, but not a mul-
ticlone. The mappings described by permutation matrices form the smallest
Toffoli subalgebra and as a subalgebra of a reversible Toffoli algebra it is
also reversible.
8
Definition 2.10 Let f, g be mappings, f have arity n, g have co-arity m.
Define f • g = f ◦k g where k = min(m, n).
This is a total operation.
Lemma 2.11 Let f, g be mappings, f have arity n, g have co-arity m. Then
(f ⊕ im−n) • g = (f ⊕ im−n) ◦m g
if n < m
if n = m
if n > m
f • g =
f ◦m g
f • (g ⊕ in−m) = f ◦n (g ⊕ in−m)
Also for any k ∈ N, (f • g) ⊕ ik = (f ⊕ ik) • (g ⊕ ik).
Proof: Let f have co-arity s and g have arity t.
Suppose n < m, f • g = f ◦n g has arity n + t − n = t. We see that
f ⊕ im−n has arity n + m − n = m so (f ⊕ im−n) • g = (f ⊕ im−n) ◦m g
holds. Then
(f • g)(x1, . . . , xt)
= (f ◦n g)(x1, . . . , xt)
= (f1(g1(x1, . . . , xt), . . . , gn(x1, . . . , xt)), . . .
. . . , fs(g1(x1, . . . , xt), . . . , gn(x1, . . . , xt)),
gn+1(x1, . . . , xt), . . . , gm(x1, . . . , xt))
= (f ⊕ im−n)(g1(x1, . . . , xt), . . . , gm(x1, . . . , xt))
= ((f ⊕ im−n) ◦m g)(x1, . . . , xt)
which is what we wanted.
The second claim is immediate from the definition of f • g.
The third claim is proved analogously to the first.
For the fourth claim we start with the case that n = m. We calculate
((f • g) ⊕ ik)(x1, . . . , xt+k)
= (f • g)(x1, . . . , xt) ⊕ (xt+1, . . . , xt+k)
= (f ◦n g)(x1, . . . , xt) ⊕ (xt+1, . . . , xt+k)
= (f1(g1(x1, . . . , xt), . . . , gn(x1, . . . , xt)), . . .
. . . , fs(g1(x1, . . . , xt), . . . , gn(x1, . . . , xt)), xt+1, . . . , xt+k)
= ((f ⊕ ik) ◦n g)(x1, . . . , xt+k)
= ((f ⊕ ik) ◦n+k (g ⊕ ik))(x1, . . . , xt+k)
= ((f ⊕ ik) • (g ⊕ ik))(x1, . . . , xt+k)
9
so the claim holds in this case. Now note that if n < m then f ⊕ im−n has
arity m so (f •g)⊕ik = ((f ⊕im−n)•g)⊕ik = (f ⊕im−n⊕ik)•(g⊕ik) =
(f ⊕ ik) • (g ⊕ ik) so this case is done. Similarly we argue in the case that
(cid:3)
n > m and thus we are done.
Lemma 2.12 The • operation is associative, that is, for all f, g, h ∈ M (A),
f • (g • h) = (f • g) • h.
Proof: In order to simplify notation, we write af for the arity of f and cf for
the co-arity of f. Then f•(g•h) = f◦k(g◦lh) where k = min(af , cg+ch−l)
and l = min(ag, ch). Similarly (f • g) • h = (f ◦¯k g) ◦¯l h where ¯k =
min(af , cg) and ¯l = min(af + ag − ¯k, ch).
Suppose af ≤ cg. Then ¯k = k = af so ¯l = min(ag, ch) = l so by the
associativity of ◦i for all i, Lemma 2.3, we are done.
Similarly if ch ≤ ag, l = ¯l = ch so k = min(af , cg) = ¯k and we are
done.
Suppose that neither of these cases applies, that is, af > cg and ch > ag.
Then
(f • g) • h = (f ◦af (g ⊕ iaf−cg )) • h
f • (g • h) = f • ((g ⊕ ich−ag ) ◦ch h)
(2)
(3)
Let a1 be the arity of f • g = f ◦af (g ⊕ iaf−cg ), that is a1 = ag + af − cg.
Let c2 be the co-arity of g • h = (g ⊕ ich−ag )◦ch h, that is c2 = cg + ch − ag.
Note that a1 − ch = af − c2.
We proceed by cases.
Case a1 < ch so c2 > af . Then
(2) = ((f ◦af (g ⊕ iaf−cg )) ⊕ ich−a1 ) ◦ch h
= ((f ⊕ ich−a1) ◦af +ch−a1 (g ⊕ iaf−cg ⊕ ich−a1 )) ◦ch h
= ((f ⊕ ic2−af ) ◦c2 (g ⊕ ich−ag )) ◦ch h
= (f ⊕ ic2−af ) ◦c2 ((g ⊕ ich−ag ) ◦ch h)
= f • (g • h) = (3)
Case a1 = ch so c2 = af . Then af − cg = ch − ag and
(2) = (f ◦af (g ⊕ iaf−cg )) ◦ch h
= f ◦c2 ((g ⊕ ich−ag ) ◦ch h)
= f • (g • h) = (3)
10
Case a1 > ch so c2 < af . Then
(2) = (f ◦af (g ⊕ iaf−cg )) ◦a1 (h ⊕ ia1−ch )
= f ◦af ((g ⊕ iaf−cg ) ◦a1 (h ⊕ ia1−ch ))
= f ◦af ((g ⊕ iaf−cg ) • (h ⊕ ia1−ch ))
= f ◦af (((g ⊕ ich−ag ) • h) ⊕ iaf−c2)
= f • ((g ⊕ ich−ag ) • h) = (3)
and we are done.
(cid:3)
Theorem 2.13 Let F be a collection of mappings on a set A. Then the fol-
lowing are equivalent:
1. F is a Toffoli algebra.
2. F is closed under {i1,⊕, τ, ζ} ∪ {◦k k ∈ N}.
3. F is closed under {⊕} ∪ {πα α ∈ S(N)} ∪ {◦k k ∈ N}.
4. F is closed under {i1,⊕, τ, ζ,•}.
Proof: The second case is a writing of the definition of the first. We demon-
strate that the others are equivalent by showing that each collection of opera-
tions can be built from the others as terms.
3 ⇒ 2 : In this signature, i1 is πα for α the identity permutation on {1}.
⊕ is the same, as are ◦k. Let f ∈ Mm,n. We see that τ f = f ◦m π(1,2), the
permutation (1, 2) acting upon {1, . . . , m}, and ζf = f ◦m π(m,...,2,1), the
permutation (m, . . . , 2, 1) acting upon {1, . . . , m}.
2 ⇒ 4 : From the definition of •, we know that it can be written in terms
of ◦k for any given arguments.
4 ⇒ 3 : Let α be a permutation on {1, . . . , n}. Let α = α1α2 . . . αk for
αi ∈ {(1, 2), (1, 2, . . . , n)}. Let βi = τ if αi = (1, 2), otherwise βi = ζ.
Then πα = βk . . . β1(i1 ⊕ . . . ⊕ i1) with i1 repeated n times.
Let f, g be mappings, f ∈ Mn,s(A), g ∈ Mm,t(A), k ≤ t, k ≤ n. Then
we claim that
f ◦k g = (f ⊕ it−k) • πα • (g ⊕ in−k)
(4)
with
α =
(cid:18) 1
1
. . .
. . .
k
k
k + 1
n + 1
. . .
. . .
t
t + n − k
t + 1
k + 1
. . .
. . .
11
(cid:19)
t + n − k
n
∈ Sn+t−k.
The left hand side of (4) has arity n + m − k, co-arity s + t − k and is
defined by
f ◦k g(x1, . . . , xn+m−k) =
fi(g1(x1, . . . , xm), . . . , gk(x1, . . . , xm),
xm+1, . . . , xm+n−k)
gk+i−s(x1, . . . , xm)
if i ≤ s
if s < i
On the right hand side of (4), we note that the co-arity of g ⊕ in−k and the
arity of f ⊕ it−k are both t + n − k, the same as the arity and co-arity of πα.
Because of the structure of the permutation α,
gj(x1, . . . , xm)
xm−k+j
gk+j−n(x1, . . . , xm)
if j ≤ k
if k < j ≤ n
if n < j ≤ t + n − k
(πα • (g ⊕ in−k))j(x1, . . . , xm+n−k) =
We can then calculate the right hand side of (4). For j ≤ s, the jth entry is:
fj • πα • (g ⊕ in−k)(x1, . . . , xn+m−k) =
fj(g1(x1, . . . , xm), . . . , gk(x1, . . . , xm), xm+1, . . . , xm+n−k)
while for s < j ≤ s + t − k, the jth entry is the output of the identity map,
so noting that the preimage of n + j − s by α is k + j − s:
((f ⊕ it−k) • πα • (g ⊕ in−k))j(x1, . . . , xn+m−k) =
(πα • (g ⊕ in−k))n+j−s(x1, . . . , xn+m−k) = gk+j−s(x1, . . . , xm)
By comparing entries, we see that the left hand side and the right hand side
(cid:3)
agree at all entries and are thus equal.
The following result enables us to see that the operations of a Toffoli alge-
bra do not destroy reversibility, so we know that B(A) is the largest reversible
Toffoli algebra on A.
Lemma 2.14 Let A be a set. Let F ⊆ B(A). Then C(F ) ⊆ B(A).
Proof: We only need to check that B(A) is a Toffoli algebra. The operations
i1, τ, ζ are invertible. Let f, g ∈ B(A). The inverse of f ⊕ g is f−1 ⊕ g−1. If
the co-arity of g is the same as the arity of f, then (f • g)−1 = g−1 • f−1. If
the co-arity of g is less than the arity of f, then note that f • g = f • (g ⊕ is)
where s is the difference between the arity of f and the co-arity of g. Then
the co-arity of g ⊕ is equals the arity of f and we are done. Similarly if the
arity of f is less that the co-arity of g.
12
so we are done.
Thus all of the operations of a Toffoli algebra map bijections to bijections,
(cid:3)
We know that there is a finite signature for Toffoli algebras, so we can
apply the standard tools and techniques of universal algebra, such as the fol-
lowing.
Corollary 2.15 Let A be a set. Then the set of Toffoli algebras on A, ordered
by inclusion, is an algebraic lattice. The set of reversible Toffoli algebras on
A, ordered by inclusion, is an algebraic lattice.
Proof: The set of Toffoli algebras on A is the set of subuniverses of the algebra
(M (A); i1,⊕, τ, ζ,•). These operations are all finitary, so by [4][Cor 3.3] we
have our result.
Similarly reversible Toffoli algebras are subuniverses of (B(A); i1,⊕, τ, ζ,•)
and we are done.
(cid:3)
We note also that the definition of a multiclone adds only two unary oper-
ations to the definition of a Toffoli algebra, so we obtain the following as a
corollary to Theorem 2.13 and Corollary 2.15.
Corollary 2.16 Let F be a collection of mappings on a set A. Then the
following are equivalent:
1. F is a multiclone.
2. F is closed under {i1,⊕, τ, ζ, ∆,∇} ∪ {◦k k ∈ N}.
3. F is closed under {⊕, ∆,∇} ∪ {πα α ∈ S(N)} ∪ {◦k k ∈ N}.
4. F is closed under {i1,⊕, τ, ζ, ∆,∇,•}.
The set of multiclones, ordered by inclusion, is an algebraic lattice.
3 REALISATION
In this section we look at the ways in which one set of mappings can be found
within another.
Theorem 3.1 Let A be a finite set, g : Am → An, o ∈ A and let r =
max(m, n + (cid:100)logA maxa∈Ang−1(a)(cid:101)). There exists an invertible f ∈
Br(A), such that g = f o
with θ1 = (m + 1, . . . , r), θ2 = (1, . . . , n).
Furthermore max(m, n) ≤ r ≤ n + m and these bounds are achieved.
θ1,θ2
13
Proof: Select o ∈ A arbitrary but fixed.
Suppose that r = n. This occurs iff (cid:100)logA maxa∈Ang−1(a)(cid:101) = 0 and
m ≤ n. Now (cid:100)logA maxa∈Ang−1(a)(cid:101) = 0 iff maxa∈Ang−1(a) = 1
iff g is injective. If m = n then g is already bijective and we are done. If
m < n then define the partial map f on Ar by f (x1, . . . , xm, o, . . . , o) =
g(x1, . . . , xm) for all x1, . . . , xm ∈ A. Then f is a partial injective map on
Ar, thus it can be completed to a bijection f of Ar. Then f o
= κ((m +
1, . . . , n), o, f ) = g and we are done.
Now suppose that r > n. For each a ∈ An, let x(1), . . . , x(k) ∈ Am
be a lexicographical ordering of g−1(a). Let b(1), . . . , b(k) be the first k ele-
ments of Ar−n in lexicographical order. Define f (x(i)
m , o, . . . , o) =
(r−n)) for each i ∈ {1, . . . , k}.
(a1, . . . , an, b(i)
1 , . . . , x(i)
1 , . . . , b(i)
Now f is a partial injective map on Ar, thus it can be completed to a
bijection of Ar.
For any z ∈ Am, let b = g(z1, . . . , zm). Then (z1, . . . , zm) ∈ g−1(b), so
f (z1, . . . , zm, o, . . . , o) = (b1, . . . , bn, c1, . . . , cm) for some c1, . . . , cm ∈ A.
(z) = µ(θ2, κ(θ1, (o, . . . , o), f )(z) = b = g(z), showing
θ1,θ2
Thus f o
= g.
f o
θ1,θ2
θ1,θ2
The average value of g−1(a) is Am
An = Am−n. Thus we know that
(cid:100)logA maxa∈Ang−1(a)(cid:101) ≥ (m − n). In the case that m ≥ n, this value
can be minimised if all g−1(a) are equal to the average value, so g−1(a) =
Am−n, so the minimum value of r is m = max(m, n). In the case that
m < n, this value is minimised if g is injective, so g−1(a) ≤ 1 and thus
(cid:100)logA maxa∈Ang−1(a)(cid:101) = 0 so r = max(m, n+0). The maximum value
is obtained if g is constant, so there is some a ∈ An such that g−1(a) = Am.
In this case, logA maxa∈Ang−1(a) = m and thus r = n + m.
(cid:3)
Corollary 3.2 ( [16] Theorem 4.1) Let A be a finite set. For all g : Am →
An there exists r ≤ n, invertible f : Am+r → Am+r, o ∈ A and θ1 ∈
{1, . . . , m + r}m, θ2 ∈ {1, . . . , m + r}n such that g = f o
Proof: Let r0 be the value of r obtained in the theorem above. There are
two possibilities for r0. If r = m then we take r = 0 < n in this theorem
If r0 = n + (cid:100)logA maxa∈Ang−1(a)(cid:101) then note that
and we are done.
g−1(a) ⊆ Am so g−1(a) ≤ Am. So r0 ≤ n + m so r ≤ n in this theorem
(cid:3)
and we are done.
θ1,θ2
.
3.1 Forms of Realisation
We now look at several ideas about generation, corresponding to different
forms of closure. While the simplest definition talks about the multiclone
14
generated from a set of mappings, Toffoli introduces some interesting ideas
about more general, engineering inspired forms of realisation. The following
is a translation of his definitions to the language we have developed here.
Definition 3.3 Let A be a set, F ⊆ M (A). Then K(F ) is the set of all
functions we can obtain from F by inserting constants using the κ operation.
Hence K(F ) the closure under all possible applications of κ, including none.
S(F ) is the set of all submappings of mappings in F . Then S(F ) is the
closure under all possible applications of µ.
R(F ) is the collection of bijections in F , i.e. R(F ) = F ∩ B(A).
Definition 3.4 Let F be a collection of mappings on a set A. A mapping
g ∈ M (A) is said to be isomorphically realised by F [16, p. 8], or realized
without ancilla bits [18], if g ∈ C(F ).
Let A be a set, F ⊆ M (A) and g ∈ Mm,n(A).
• We say g is realised by F if there exists some f ∈ C(F ) ∩ Ml,k(A)
and a1, . . . , al−m ∈ A such that for all i ∈ {1, . . . , n} and for all
x1, . . . , xm ∈ A:
gi(x1, . . . , xm) = fi(x1, . . . , xm, a1, . . . , al−m)
(5)
Equivalently, we can say g ∈ SKC(F ).
• We say g is realised with no garbage if there exists some f ∈ Ml,n(A)∩
C(F ) satisfying (5), equivalently g ∈ KC(F ). Otherwise g is realised
with garbage.
• We say g is realised with no constants if there exists some f ∈ Mm,k(A)∩
C(F ) satisfying (5), equivalently g ∈ SC(F ). Otherwise g is realised
with constants.
A mapping is isomorphically realised iff it is realised with no constants
and no garbage, that is, that there exists a single function that satisfies both
claims. This should not be confused with the claim that for any F ⊆ M (A),
C(F ) = SC(F ) ∩ KC(F ), which remains open. The inclusion ⊆ is clear
since C(F ) ⊆ SC(F ) and C(F ) ⊆ KC(F ) for any F ⊆ M (A), the other
inlusion is unclear.
However we obtain the following.
Lemma 3.5 Let A be a set, F ⊆ M (A) a collection of balanced maps, then
C(F ) = SC(F ) ∩ KC(F ). In particular, let A be a finite set, F ⊆ B(A),
then C(F ) = SC(F ) ∩ KC(F ).
15
Proof: By the comment above, we only need show that SC(F ) ∩ KC(F ) ⊆
C(F ). Suppose g ∈ SC(F ) ∩ KC(F ) and g ∈ Mm,n(A). Also note that all
the operations of a Toffoli algebra map balanced maps to balanced maps, so
C(F ) consists only of balanced maps.
Since g ∈ KC(F ) we know there exists some h ∈ C(F ) ∩ Ml,n(A)
and a1, . . . , al−m ∈ A such that for all i ∈ {1, . . . , n}, gi(x1, . . . , xm) =
hi(x1, . . . , xm, a1, . . . , al−m). Because h is balanced, l = n, so m ≤ n.
Since g ∈ SC(F ) we know there exists some ¯h ∈ C(F )∩ Mm,k(A) such
that for all i ∈ {1, . . . , n}, gi(x1, . . . , xm) = ¯hi(x1, . . . , xm). Because ¯h is
balanced, m = k, so n ≤ k.
Combining these we see that m = n = k = l so g = h = ¯h and g ∈ C(F )
so we are done.
We know that when A is finite, bijections in B(A) are balanced, so we
(cid:3)
obtain the second result.
This language allows us to rephrase Corollary 3.2 as:
Corollary 3.6 Let A be a finite set. For all g ∈ Mm,n(A), g is realised by
Bm+n(A).
There is a special class of realisations that have constants and garbage, but
in some way do not use them up, the garbage representing the constants.
Definition 3.7 Let A be a set, F ⊆ M (A) and g ∈ Mm,n(A). Then g is
realised with temporary storage by F if there exists some f ∈ C(F ), f ∈
Ml,k(A) and a1, . . . , al−m ∈ A such that n + l − m = k and
∀i ∈ {1, . . . , n} : gi(x1, . . . , xm) = fi(x1, . . . , xm, a1, . . . , al−m)
∀i ∈ {1, . . . , l − m}∀x1, . . . , xm ∈ A :
(6)
fn+i(x1, . . . , xm, a1, . . . , al−m) = ai
(7)
Definition 3.8 Let A be a set, F ⊆ M (A) and g ∈ Mm,n(A). Then g is
realised with strong temporary storage by F if there exists some f ∈ C(F ),
f ∈ Ml,k(A) and a1, . . . , al−m ∈ A such that n + l − m = k and
∀i ∈ {1, . . . , n} : gi(x1, . . . , xm) = fi(x1, . . . , xm, a1, . . . , al−m)
∀i ∈ {1, . . . , l − m}∀x1, . . . , xm ∈ A :
fn+i(x1, . . . , xm, a1, . . . , al−m) = ai
∀b1, . . . , bm ∈ A : (x1, . . . , xl−m) (cid:55)→ (fn+1(b1, . . . , bm, x1, . . . , xl−m), . . .
. . . fk(b1, . . . , bm, x1, . . . , xl−m)) ∈ Bl−m(A)
16
Let F ⊆ M (A) be some mappings on A. Then let T (F ) be the set of
mappings realised with temporary storage by F , TS(F ) the set of mappings
realised with strong temporary storage by F .
These concepts and language are introduced based on Toffoli's work. When
we compose maps, this temporary storage allows us to use the garbage again
in the subsequent function, as we know what it looks like. Thus the extra
inputs and outputs are in some sense not garbage, as they are appropriately
recycled and reduce waste information. In the engineering perspective, this
reduces waste heat in implementations.
Let's look at an example. Let A = Zn, f (x, y) = (2x + y, xy). Then
the map g(x) = 2x is realised with temporary storage by {f}, since g(x) =
f1(x, 0) and f2(x, 0) = 0 for all x. However g is not realised with strong
temporary storage, because y (cid:55)→ f2(0, y) (cid:54)∈ B1(A).
Lemma 3.9 Let A be a finite set, F ⊆ B(A). Then TS(F ) ⊆ T (F ) ⊆ B(A).
Proof: The first inclusion is clear, as the definition of strong temporary storage
is stricter than the definition of temporary storage.
Let g ∈ T (F ). This means there is some f ∈ C(F ), a1, . . . , al−m ∈ A
satisfying the requirements (6) and (7) above. We know that f ∈ B(A)
because F ⊆ B(A). Thus f is balanced, so l = k and thus n = m. By (7), f
fixes {(x1, . . . , xm, a1 . . . , al−m) x1, . . . , xm ∈ A} as a set. Because f is a
bijection, f is a permutation of {(x1, . . . , xm, a1 . . . , al−m) x1, . . . , xm ∈
A}. By the definition of g in (6), g is a permutation of Am, so g ∈ B(A).
Thus T (F ) ⊆ B(A) and we are done.
(cid:3)
4 THE FUNCTIONS OF A REVCLONE
In Toffoli's "Fundamental Theorem," our Corollary 3.2, he shows that all
maps and thus all functions can be realised by the full reversible Toffoli alge-
bra. If we look at clones, this means that the clone of all functions on A can
be realised, i.e. OA ⊆ KS(B(A)).
Let A be a set. Let F ⊆ M (A). We call S1(F ) = {µ((1), f ) f ∈ F} =
{f1 f ∈ F} the function set of F .
Theorem 4.1 Let F be a Toffoli algebra. Then S1(F ) is a linear term alge-
bra with projections.
Proof: The operations τ and ζ act the same in the reversible Toffoli algebra
and the linear term algebra. Let f, g ∈ F , then ∇(f1) = (¯τ (i1 ⊕ f ))1. Also
17
f1 ∗ g1 = (f ◦1 g)1. Thus F being closed as a multiclone means that S1(F )
is closed as a linear term algebra.
Moreover, the constants πα ∈ F mean that all projections are also in
S1(F ), so with (1 i) ∈ Sn, πn
(cid:3)
This also applies for reversible Toffoli algebras. Note that we obtain the ∇
operation "for free" from the other operators. In general we see the following.
i = (π(1 i))1.
Corollary 4.2 Let F be a multiclone. Then S1(F ) is a clone.
Proof: A multiclone is a Toffoli algebra closed under the ∆ and ∇ operations
on multiclones. By Theorem 4.1 we have ∇ operating on S1(F ). For any
f1 ∈ S1(F ), ∆(f1) = (∆f )1 and we have clone closure.
(cid:3)
This justifies the use of the expression "clone" in our expression multi-
Note that the mapping F (cid:55)→ {f1 f ∈ F} is not injective. Let A =
clone.
{0, 1, 2, 3, 4}, C =(cid:83)
n GL(n, Z5) ⊆ B(A) a reversible Toffoli algebra,
(cid:91)
D =
{−1, 1}SGL(n, Z5) = {f ∈ C det f ∈ {−1, 1}}
n
a subclone. D is a proper sub reversible Toffoli algebra of C, by the determi-
5 \ {(0, . . . , 0)} so
nant. Both GL(n, Z5) and D ∩ Bn(A) are transitive on Zn
n{f : An → A
both reversible Toffoli algebras have function sets that are(cid:83)
f (x1, . . . , xn) =(cid:80) aixi, ai ∈ Z5, (a1, . . . , an) (cid:54)= 0}, the clone of nonzero
linear forms on Z5.
Corollary 4.3 Let A be a finite set, let F be a reversible Toffoli algebra. Then
g ∈ S1(F ), g of arity m implies that for all a ∈ A, g−1(a) = A(m−1).
Proof: This is a corollary of Theorem 3.1. Let g = µ((1), f ) for some
f ∈ F , f has arity m. By requiring that the reversible mapping has the
same arity, we require r = m and n = 1 in the statement of the theorem.
Because r = max(m, 1 + (cid:100)logA maxa∈Ang−1(a)(cid:101)) = m, we know that
logA maxa∈Ang−1(a) ≤ m − 1. However the average value of g−1(a)
is Am−1 so the average value must be equal to the maximal value, so all
g−1(a) are equal and we are done.
(cid:3)
5 SOME RESULTS ABOUT REALISATION
Here we collect some results on reversible mappings. Much of this is based
upon section 5 in [16], with the extension to Theorem 5.11.
18
Definition 5.1 Let A be a set. Let α be a permutation of A, o ∈ A some
constant, n ∈ N. For n > 1, define T G(n, α, o) ∈ Bn(A) by
T G(n, α, o)(x1, . . . , xn)i = xi
i < n
(cid:40)
T G(n, α, o)(x1, . . . , xn)n =
α(xn)
xn
if x1 = . . . = xn−1 = o
otherwise
Let T G(1, α, o)(x) = α(x). Then T G(n, α, o) ∈ Bn is an invertible map-
ping, the Toffoli Gate induced by n, α and o.
In [16], this mapping is defined for A = Z2 with α swapping 0 and 1,
o = 1, written as θ(n) = T G(n, (0 1), 1).
Definition 5.2 Let A be a set. A permutation f ∈ Bn is elementary if there
exist distinct x, y ∈ An such that f (x) = y, f (y) = x and all other elements
of An are fixed. An elementary permutation is atomic if x and y only differ in
one entry, i.e. xi = yi for all i except one.
Lemma 5.3 Let A be a set. Let n ∈ N, let o ∈ A. Let f ∈ Bn(A) be
atomic. Then f can be isomorpically realised by {T G(n, α, o) α ∈ SA} ∪
{T G(1, α, o) α ∈ SA}.
Proof: Suppose f exchanges x, y ∈ An which differ at position i. Then
π(i,n)◦n f ◦n π(i,n) is atomic, exchanging two vectors that differ at position n.
Wlog suppose f is of this form. Let αi := (o, xi) ∈ SA be a transposition of
A. Let βi = T G(1, αi, o) for i < n, βn the identity. Then β = β1⊕. . .⊕βn ∈
Bn is an involution, generated by T G(1, αi, o) as a reversible Toffoli algebra.
Note that β(x)i = β(y)i = o for i < n.
Note that f (β(z)) = β(z) unless β(z) = x respectively β(z) = y. But
β(z) = x (resp. β(z) = y) iff zi = o for all i < n and zn = xn (resp.
Zn = yn). So β • f • β fixes all elements of An except (o, . . . , o, xn) and
(o, . . . , o, yn), which it exchanges. Thus β • f • β = T G(n, (xn, yn), o),
so f is in the reversible Toffoli algebra generated by {T G(n, α, o) α ∈
SA} ∪ {T G(1, α, o) α ∈ SA}.
(cid:3)
Theorem 5.4 ([16] Thm 5.1) Let A be a set. Any f ∈ Bn(A) can be iso-
morphically realised by atomic permutations.
Proof: If f is the identity, then f = in and we are done. Suppose f is nontriv-
ial. Any permutation can be written as a product of elementary permutations.
So wlog, let f be elementary, exchanging x, y ∈ An.
19
Define a sequence (a(i) ∈ An 1 ≤ i ≤ n + 1) such that for all k ∈
{1, . . . n + 1}, (a(k))i = yi if i < k, (a(k))i = xi otherwise. Then a(1) = x,
a(n+1) = y. Note that a(i) = a(i+1) or they differ in position i. Thus
fi ∈ Bn(A) exchanging a(i) and a(i+1) is an atomic permutation. Now the
permutation f = f1 • f2 • . . . • fn−1 • fn • fn−1 • . . . • f1 is a product of
(cid:3)
atomic permutations, so we are done.
Corollary 5.5 Let A be a finite set, say A = {1, . . . , k}. Then Bn(A) is
realised by {T G(n, α, 1) α ∈ {(1, 2), (1, . . . , k)}} ∪ {(1, 2), (1, . . . , k)}.
Proof: We remind ourselves that T G(1, α, 1) = α and thus see that the four
gates listed here generate {T G(n, α, 1) α ∈ SA}∪{T G(1, α, 1) α ∈ SA}.
By Lemma 5.3 this is enough to generate all atomic permutations and thus by
(cid:3)
Theorem 5.4, enough to generate Bn(A).
This is one possible correct form of the result hoped for in Conjecture 1
of [18]. The conjecture can be demonstrated to be false using calculations
in GAP [6] for A of order 5 which can be understood as follows, using a
description kindly supplied by Erkko Lehtonen in a private communication.
Conjecture 5.6 (Conjecture 1 of [18]) Let m ≥ 3, n ≥ 2 be integers, A =
Zm. Then Bn(A) is generated by the following permutations, where addition
is modulo m.
• Let (ij) ∈ Sn, σij = π(ij)
• Let i ∈ {1, . . . , n}, for all x1, . . . , xn ∈ A, let νi(x1, . . . , xn) =
(x1, . . . , xi−1, xi + 1, xi+1, . . . , xn)
• For all x1, . . . , xn ∈ A, let
(cid:40)
Υ(x1, . . . , xn) =
(x1 + 1, 0, . . . , 0)
(x1, . . . , xn)
if x2 = ··· = xn = 0
otherwise
The permutation σij is a product of mn−1(m − 1)/2 transpositions. Thus
This can be falsified by considering the parities of the permutations.
σij is odd iff m ≡ 3 mod 4 or (m ≡ 2 mod 4 and n = 2).
Every element of An belongs to a cycle of length m of νi. Thus νi is
a product of mn−1 cycles of length m, each of which is made of m − 1
transpositions, so νi is a product of mn−1(m − 1) transpositions, which is
always even.
The permutation Υ is a cycle of length m, thus is even iff m is odd.
20
Thus for m = 5 or more generally m ≡ 1 mod 4 all of these permuta-
tions are even so they cannot generate Bn(A) for any n.
For m even, n ≥ 2, the conjecture might hold, calculations in GAP for
small values find no contradiction. So we obtain the following conjecture
that a smaller generating set might suffice.
Conjecture 5.7 For A = {1, . . . , m} even, n ≥ 2,
Bn(A) = C({(1, . . . , m), T G(n, (1, . . . , m), 1)})
Below we will see that a small generating set exists for A of odd order.
It is worth noting that the main result in [18] is also subtly wrong for n =
1, as the permutation (0 1) is not generated. The authors use an interesting
property of ternary logic that any permutation of {0, 1, 2} (written in Z3) can
for some n, leading to
be written as
their minimal generating set for n ≥ 2. For this reason it might be that their
result only applies for ternary alphabets.
(cid:16) 0
(cid:16) 0
(cid:17)
n
n+1
n+1+1
(cid:17)
n
n+2
n+2+2
1
2
or
1
2
The following results lets us calculate precisely what the n-ary part of a
reversible Toffoli algebra looks like, given the generators.
Theorem 5.8 Let A be a set. Let F ⊆ B(A), n ∈ N. Let
¯F ={ ¯f f ∈ F, arity f = m < n, ¯f = f ⊕ in−m}
∪ {f ∈ F arity f = n} ∪ {πα α ∈ Sn}.
Then the group ((cid:104) ¯F(cid:105);◦n) is equal to (C(F ) ∩ Bn(A);◦n).
Proof: Both groups are subgroups of (Bn(A);◦n), so we need only prove
they are equal as sets.
elements of ¯F is n so they are all within Bn(A). So this inclusion holds.
We proceed by induction. For the case n = 1, ¯F consists of the permuta-
tions of A within F . Then the subgroup of SA generated by these permuta-
tions is precisely C(F ) ∩ B1(A) = C(F ) ∩ SA. Then we proceed with our
induction. Suppose our claim holds up to n − 1.
(⊆) : Every element ¯f = f ⊕ in−m of ¯F is within C(F ). The arity of all
(⊇) : Let f ∈ C(F ) ∩ Bn(A), so f is either in ¯F or is a term of the form:
1. f = g ⊕ h for some g, h ∈ C(F ), both of smaller arity than n.
2. f = g ◦k h for some g, h ∈ C(F ), both of smaller arity than n, k < n.
3. f = g ◦k h for some g, h ∈ C(F ), exactly one of them in Bn(A), so
the other one has arity k, k < n.
21
4. f = g ◦n h for some g, h ∈ C(F ) ∩ Bn(A).
In case 1, let m be the arity of g, so h has arity n − m. We know from our
induction hypothesis that g is a product g = ¯g1 ◦m . . . ◦m ¯gl of elements of
the form ¯gj = fj ⊕ imj for some fj ∈ F with arity less than m, ¯gj = fj
for some fj ∈ F with arity m, or ¯gj = παj for some permutation αj ∈
Sm. Then let φj = fj ⊕ imj +n−m = fj ⊕ imj ⊕ in−m ∈ ¯F respectively
φj = fj ⊕ in−m ∈ ¯F respectively φj = παj ⊕ in−m = πβj where βj ∈ Sn
is equal to αj on {1, . . . , m} and fixes the elements {m + 1, . . . , n}. Then
φ1 ◦n . . . ◦n φl = g ⊕ in−m, so g ⊕ in−m ∈ (cid:104) ¯F(cid:105).
Similarly we can write h ⊕ im as an element of (cid:104) ¯F(cid:105). Let δ be the permu-
tation of {1, . . . , n} defined by
(cid:18)
δ =
1
(n − m + 1)
. . .
. . .
m
n
(m + 1)
1
. . .
. . .
n
(n − m)
(cid:19)
.
(cid:19)
(cid:18) 1
1
Then
πδ−1 ◦n (h ⊕ im) ◦n πδ ◦n (g ⊕ in−m) = g ⊕ h
Thus we see that f ∈ ((cid:104) ¯F(cid:105);◦n).
In case 2 we proceed similarly. In this case g is m-ary and h is n− m + k-
ary. We use the same techniques as above to show that g⊕ in−m, h⊕ im−k ∈
(cid:104) ¯F(cid:105). Let δ be the permutation of {1, . . . , n} given by
(n − m + k)
(n − m + k + 1)
δ =
. . .
. . .
(k + 1)
(m + 1)
. . .
k
. . .
k
so πδ ∈ ¯F . Then
(g ⊕ in−m) ◦n πδ ◦n (h ⊕ im−k)(x1, . . . , xn)
= (g ⊕ in−m) ◦n πδ(h1(x1, . . . , xn−m+k), . . . , hn−m+k(x1, . . . , xn−m+k),
(k + 1)
. . .
. . .
n
m
n
= (g ⊕ in−m)(h1(x1, . . . , xn−m+k), . . . , hk(x1, . . . , xn−m+k),
xn−m+k+1, . . . , xn)
xn−m+k+1, . . . , xn,
hk+1(x1, . . . , xn−m+k), . . . , hn−m+k(x1, . . . , xn−m+k))
= (g1(h1(x1, . . . , xn−m+k), . . . , hk(x1, . . . , xn−m+k), xn−m+k+1, . . . , xn),
...
gm(h1(x1, . . . , xn−m+k), . . . , hk(x1, . . . , xn−m+k), xn−m+k+1, . . . , xn),
hk+1(x1, . . . , xn−m+k), . . . , hn−m+k(x1, . . . , xn−m+k))
= g ◦k h(x1, . . . , xn)
22
Thus we see that f ∈ ((cid:104) ¯F(cid:105);◦n).
In case 3 we have k < n then one of the terms is of lower arity, so as above
we can write g ⊕ in−k (respectively h ⊕ in−k) as an element of Bn(A) and
have f = (g ⊕ in−k) ◦n h (respectively f = g ◦n (h ⊕ in−k)). Then we have
the next case.
In case 4 we have two terms of arity n but of strictly lower term complex-
ity. We use induction on term complexity. For the initial case, we look at
the trivial terms t ∈ C(F ) ∩ Bn(A). These are either elements of F or the
permutations πα for α ∈ Sn. In both these cases, t ∈ ¯F so t ∈ (cid:104) ¯F(cid:105). Now we
look at f = g ◦n h. Both g and h have lower term complexity than f. Thus
we know that g, h ∈ (cid:104) ¯F(cid:105). Thus f = g ◦n h ∈ (cid:104) ¯F(cid:105) and we are done.
(cid:3)
We can apply this to obtain a generalisation of one of Toffoli's results.
Corollary 5.9 ([16] Thm 5.2) Let A be a finite set of even order, n ∈ N.
Then Bn(A) is not isomorphically realised by {T G(i, α, o) α ∈ SA, i <
n} for any o ∈ A
Proof: For n = 1 the set of generators is empty, so we only have the identity
permutation. For n = 2, we calculate the value of C({T G(i, α, o) α ∈
SA, i < n}) ∩ B2(A) and B2(A). First note that {T G(i, α, o) α ∈
SA, i < n} = {T G(1, α, o) α ∈ SA} so our generating set consists only
of permutations of entries in A. We see that C({T G(i, α, o) α ∈ SA, i <
n})∩ B2(A) thus consists only of bijections that map (a, b) ∈ A2 to (aα, bβ)
or (bβ, aα) for some α, β ∈ SA. Thus C({T G(i, α, o) α ∈ SA, i <
2}) ∩ B2(A) = SA22 = 2(A!)2. On the other hand, B2(A) = (A2)!
which is strictly greater than 2(A!)2 for all A with at least two elements.
We carry on for the other cases. Using the terminology in Theorem 5.8, we
note that T G(i, α, o) = T G(i, α, o)⊕ in−i. The action of τ = T G(i, α, o)⊕
in−i on An, when not identity, is of the form
τ (o, . . . , o, a, a1, . . . , an−i) = (o, . . . , o, α(a), a1, . . . , an−i)
If we write α as a product of transpositions α = α1 . . . αk we can use the
expression above to see that there will be kAn−i transpositions when we
write the action of τ on An. Thus T G(i, α, o) = T G(i, α, o) ⊕ in−i is an
even permutation in Bn(A).
The action of πβ for β ∈ Sn acting on An can also be calculated. If β is a
transposition in Sn, say β = (ij), then πβ acts nontrivially on (a1, . . . , ai, . . . , aj, . . . , an)
with ai (cid:54)= aj. There are A(A−1)
(A)n−2 such tuple pairs. This number
is even when A is even, except in the case n = 2 (when A ≡ 2 mod 4)
which we dealt with above.
2
23
From Theorem 5.8, we know that the arity n part of C({T G(i, α, o)
α ∈ SA, i < n}) is generated by precisely these permutations, which are all
even. Thus for some a, b ∈ A, a (cid:54)= b, T G(n, (a, b), a) is not in the generated
reversible Toffoli algebra and thus Bn(A) is not isomorphically realised by
{T G(i, α, o) α ∈ SA, i < n}.
(cid:3)
The following has been noted using examples in GAP[6].
Conjecture 5.10 Let A be a finite set of even order, n ≥ 3, F = {T G(i, α, o)
α ∈ SA, i < n, o ∈ A}. We claim that C(F ) ∩ Bn is a subgroup of Bn iso-
morphic to the alternating group on An, except for A = 2, n = 3.
Note added in referee process of this manuscript: The conjecture has been
proven in [3].
The restriction shown in Corollary 5.9 does not hold for A odd, as we see
in the following result. Moreover, it shows that we only need use the Toffoli
gates of arity 1 and 2 to obtain all bijections on A.
Theorem 5.11 Let A be a finite set of odd order. Then B(A) is isomorphi-
cally realised by
T2 = {T G(i, α, o) α ∈ SA, i < 3} = SA ∪ {T G(2, α, o) α ∈ SA}
for any o ∈ A.
Proof: Let A = k, wlog A = {1, . . . , k} and o = 1. We proceed by
induction on n, i.e. showing that Bn(A) ⊆ C(T2). Our start is for n = 2,
given by the hypothesis. We assume we have shown the claim up to n. We
first show that we can obtain T G(n + 1, (12), o) from {T G(i, α, o) α ∈
SA, i ≤ n}. Let γ = (n n + 1) ∈ Sn+1.
Define
Σ1 = (T G(n, (1 . . . k)−1, 1) ⊕ i1) • πγ • (T G(n, (12), 1) ⊕ i1)
• πγ • (in−1 ⊕ T G(2, (12), 1))
• (T G(n, (1 . . . k), 1) ⊕ i1) ∈ Bn+1(A)
We calculate.
=
Σ1(x1, . . . , xn+1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
24
•(in−1 ⊕ T G(2, (12), 1))(x1, . . . , xn−1, x(1...k)
if x1 = . . . = xn−1 = 1
•(in−1 ⊕ T G(2, (12), 1))(x1, . . . , xn−1, xn, xn+1)
otherwise
n
, xn+1)
=
=
=
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • πγ
•(T G(n, (12), 1) ⊕ i1)(x1, . . . , xn−1, x(12)
n+1, x(1...k)
n
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • πγ
•(T G(n, (12), 1) ⊕ i1)(x1, . . . , xn−1, xn+1, x(1...k)
n
if x1 = . . . = xn−1 = 1, xn = k
)
)
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • πγ
if x1 = . . . = xn−1 = 1, xn (cid:54)= k
•(T G(n, (12), 1) ⊕ i1)(x1, . . . , xn−1, x(12)
n+1, xn)
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • πγ
•(T G(n, (12), 1) ⊕ i1)(x1, . . . , xn−1, xn+1, xn)
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
•πγ(x1, . . . , xn−1, x(1...k)
n
, x(12)
n+1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
•πγ(x1, . . . , xn−1, x(1...k)
n
, xn+1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
•πγ(x1, . . . , xn−1, xn, x(12)
n+1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1) • . . .
•πγ(x1, . . . , xn−1, xn, xn+1)
if x1 = . . . = xn−1 = 1, xn = k
if x1 = . . . = xn−1 = 1, xn (cid:54)= k
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
(T G(n, (1 . . . k)−1, 1) ⊕ i1)• πγ(x1, . . . , xn−1, x(12)(12)
n+1
, x(1...k)
n
)
(T G(n, (1 . . . k)−1, 1) ⊕ i1)• πγ(x1, . . . , xn−1, x(12)
n+1, x(1...k)
n
)
if x1 = . . . = xn−1 = 1, xn = k
if x1 = . . . = xn−1 = 1, xn (cid:54)= k
(T G(n, (1 . . . k)−1, 1) ⊕ i1)• πγ(x1, . . . , xn−1, x(12)
n+1, xn)
(T G(n, (1 . . . k)−1, 1) ⊕ i1)• πγ(x1, . . . , xn−1, xn+1, xn)
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
25
=
=
(T G(n, (1 . . . k)−1, 1) ⊕ i1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1)
(T G(n, (1 . . . k)−1, 1) ⊕ i1)
(x1, . . . , xn−1, x(1...k)
n
, xn+1)
if x1 = . . . = xn−1 = 1, xn = k
(x1, . . . , xn−1, x(1...k)
n
, x(12)
n+1)
if x1 = . . . = xn−1 = 1, xn (cid:54)= k
(x1, . . . , xn−1, xn, x(12)
n+1)
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
(x1, . . . , xn−1, xn, xn+1)
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
(x1, . . . , xn−1, xn, xn+1)
(x1, . . . , xn−1, xn, x(12)
n+1)
(x1, . . . , xn−1, xn, x(12)
n+1)
(x1, . . . , xn−1, xn, xn+1)
if x1 = . . . = xn−1 = 1, xn = k
if x1 = . . . = xn−1 = 1, xn (cid:54)= k
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
We see that Σ1 is nonidentity iff:
1. Not all of x1 . . . , xn−1 are 1 and xn = 1, then xn+1 (cid:55)→ x(12)
n+1, or
2. x1 = . . . = xn−1 = 1 and xn (cid:54)= k, then xn+1 (cid:55)→ x(12)
n+1.
For m ∈ {2, . . . , k − 1}, let
σm = (T G(n, (1m), 1) ⊕ i1) • (in−1 ⊕ T G(2, (12), 1)) • (T G(n, (1m), 1) ⊕ i1) ∈ Bn+1(A)
We calculate
σm(x1, . . . , xn+1)
(T G(n, (1m), 1) ⊕ i1) • (in−1 ⊕ T G(2, (12), 1))(x1, . . . , xn−1, x(1m)
n
, xn+1)
=
(T G(n, (1m), 1) ⊕ i1) • (in−1 ⊕ T G(2, (12), 1))(x1, . . . , xn−1, xn, xn+1)
if x1 = . . . = xn−1 = 1
if some x1, . . . , xn−1 (cid:54)= 1
26
(T G(n, (1m), 1) ⊕ i1)(x1, . . . , x(1m)
, x(12)
n+1)
(T G(n, (1m), 1) ⊕ i1)(x1, . . . , x(1m)
n
, xn+1)
(T G(n, (1m), 1) ⊕ i1)(x1, . . . , xn, x(12)
n+1)
(T G(n, (1m), 1) ⊕ i1)(x1, . . . , xn, xn+1)
n
if x1 = . . . = xn−1 = 1, xn = m
if x1 = . . . = xn−1 = 1, xn (cid:54)= m
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
=
=
(x1, . . . , xn, x(12)
n+1)
(x1, . . . , xn, xn+1)
(x1, . . . , xn, x(12)
n+1)
(x1, . . . , xn, xn+1)
if x1 = . . . = xn−1 = 1, xn = m
if x1 = . . . = xn−1 = 1, xn (cid:54)= m
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
Once again this function is almost always identity, σm is nonidentity iff:
1. Not all of x1 . . . , xn−1 are 1 and xn = 1, then xn+1 (cid:55)→ x(12)
2. x1 = . . . = xn−1 = 1 and xn = m, then xn+1 (cid:55)→ x(12)
n+1.
Then define
n+1, or
Σ2 = σk−1 • ··· • σ2
We have then that Σ2 is nonidentity iff
1. Not all of x1, . . . , xn−1 are 1 and xn = 1, then xn+1 (cid:55)→ x(12)k−2
=
n+1
n+1 because k and thus k − 2 are odd, or
x(12)
2. x1 = . . . = xn−1 = 1 and xn ∈ {2, . . . , k − 1}, then xn+1 (cid:55)→ x(12)
n+1.
We see that Σ2 • Σ1 is identity unless one of the factors is nonidentity.
There are three cases:
1. Both are nonidentity by their first case. Then not all of x1, . . . , xn−1
are 1 and xn = 1, so
Σ2 • Σ1(x1 . . . xn+1) = Σ2(x1, . . . , x(12)
n+1)
= (x1, . . . , x(12)(12)
n+1
= (x1, . . . , xn+1)
)
so Σ2 • Σ1 is the identity.
27
2. Both are nonidentity by their second case, so x1 = . . . = xn−1 = 1
and xn ∈ {2, . . . , k − 1},
Σ2 • Σ1(x1 . . . xn+1) = Σ2(x1, . . . , x(12)
n+1)
= (x1, . . . , x(12)(12)
n+1
= (x1, . . . , xn+1)
)
so Σ2 • Σ1 is the identity.
3. Only Σ1 is nonidentity by the second case, so x1 = . . . = xn = 1, then
Σ2 • Σ1(x1 . . . xn+1) = Σ2(x1, . . . , x(12)
n+1)
= (x1, . . . , x(12)
n+1)
= (x1, . . . , x(12)
n+1)
so we have the only case that Σ2 • Σ1 is nonidentity.
Thus we see that Σ2 • Σ1 = T G(n + 1, (12), 1).
We now look at T G(n + 1, (1 . . . k), 1). Because the group generated by
(1 . . . k) is cyclic of odd order, the homomorphism x (cid:55)→ x2 of this group
is an automorphism. Thus there exists some β such that β2 = (1 . . . k).
This will also be a k-cycle, write β = (β1 . . . βk). Let k = 2l + 1. Define
α = (β1βk)(β2βk−1) . . . (βlβl+2). Then αβα = β−1, so βαβ−1α = β2 =
(1 . . . k).
Let γ = (n n + 1) ∈ Sn+1. Define
Σ = πγ • (T G(n, α, 1) ⊕ i1) • πγ • (in−1 ⊕ T G(2, β−1, 1)) • πγ
•(T G(n, α, 1) ⊕ i1) • πγ • (in−1 ⊕ T G(2, β, 1)) ∈ Bn+1(A)
We calculate.
Σ(x1, . . . , xn+1)
(cid:40)
(cid:40)
=
=
=
if xn = 1
if xn (cid:54)= 1
πγ • ··· • πγ(x1, . . . , xn, xβ
n+1)
πγ • ··· • πγ(x1, . . . , xn, xn+1)
πγ • ··· • (T G(n, α, 1) ⊕ i1)(x1, . . . , xβ
n+1, xn)
πγ • ··· • (T G(n, α, 1) ⊕ i1)(x1, . . . , xn+1, xn)
πγ • ··· • πγ(x1, . . . , xβα
n+1, xn)
πγ • ··· • πγ(x1, . . . , xβ
n+1, xn)
πγ • ··· • πγ(x1, . . . , xα
n+1, xn)
πγ • ··· • πγ(x1, . . . , xn+1, xn)
if x1 = . . . = xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
if xn = 1
if xn (cid:54)= 1
28
(cid:40)
=
=
=
=
=
=
πγ • ··· • (in−1 ⊕ T G(2, β−1, 1))
πγ • ··· • (in−1 ⊕ T G(2, β−1, 1))
πγ • ··· • (in−1 ⊕ T G(2, β−1, 1))
πγ • ··· • (in−1 ⊕ T G(2, β−1, 1))
)
n+1
πγ • ··· • πγ(x1, . . . , xn, xβαβ−1
n+1
πγ • ··· • πγ(x1, . . . , xn, xββ−1
n+1 )
πγ • ··· • πγ(x1, . . . , xn, xα
n+1)
πγ • ··· • πγ(x1, . . . , xn, xn+1)
πγ • (T G(n, α, 1) ⊕ i1)(x1, . . . , xβαβ−1
πγ • (T G(n, α, 1) ⊕ i1)(x1, . . . , xn+1, xn)
πγ • (T G(n, α, 1) ⊕ i1)(x1, . . . , xα
n+1, xn)
πγ • (T G(n, α, 1) ⊕ i1)(x1, . . . , xn+1, xn)
πγ(x1, . . . , xβαβ−1α
πγ(x1, . . . , xn+1, xn)
πγ(x1, . . . , xαα
n+1, xn)
πγ(x1, . . . , xn+1, xn)
(x1, . . . , xn, xβαβ−1α
n+1
(x1, . . . , xn, xn+1)
(x1, . . . , xn, xn+1)
(x1, . . . , xn, xn+1)
(x1, . . . , xn, x(1...k)
n+1 )
(x1, . . . , xn, xn+1)
, xn)
n+1
n+1)
n+1)
n+1)
(x1, . . . , xn, xβα
if x1 = . . . = xn = 1
(x1, . . . , xn, xβ
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
(x1, . . . , xn, xα
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
(x1, . . . , xn, xn+1)
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
if x1 = . . . = xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
if x1 = . . . = xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
, xn)
if x1 = . . . = xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
)
if x1 = . . . = xn = 1
if some x1, . . . , xn−1 (cid:54)= 1, xn = 1
if x1 = . . . = xn−1 = 1, xn (cid:54)= 1
if some x1, . . . , xn−1 (cid:54)= 1, xn (cid:54)= 1
if x1 = . . . = xn = 1
otherwise
Thus we see that Σ = T G(n + 1, (1, . . . , k), 1).
We have shown that we can realise the two Toffoli gates that generate all
of {T G(n + 1, α, o) α ∈ SA}. Thus by Corollary 5.5, Bn+1(A) is realised
(cid:3)
for all n, so by induction, all of B(A) is realised.
Thus we have a small generating set for the bijections.
29
Corollary 5.12 Let A = {1, . . . , k} be of odd order k. Let α = (1 . . . k), β =
(12) ∈ SA. Then B(A) is realised by {α, β, T G(2, α, 1), T G(2, β, 1)}.
We see that when A is of odd order, T G(i + 1, α, 1) is in the reversible
Toffoli algebra generated by {T G(i, α, 1) α ∈ SA}, which is not the case
for even order A. We see a distinct property looking at closure with tem-
porary storage in the following. Note that T G(n, α, o) = µ((2, 3, . . . , n +
1), κ(1, o, T G(n+1, α, o))) so T G(n, α, o) is realised with strong temporary
storage from T G(n+1, α, o). We see that closure under TS is a stronger form
of generation.
Theorem 5.13 ([16] Thm 5.3) Let A be a set, o ∈ A. For every n, for every
α, T G(n, α, o) can be realised with strong temporary storage by {T G(i, α, o)
i ≤ 3, α ∈ SA}.
Proof: We proceed by induction. Assume that we can construct T G(n −
1, α, o) for all α. Let p ∈ A some non-o element, β = (o p) be an transposi-
tion of A. Define f ∈ Bn+1(A) by
f = (T G(n − 1, β, o) ⊕ i2) • (in−2 ⊕ T G(3, α, o)) • (T G(n − 1, β, o) ⊕ i2)
Then
f (x1, . . . , xn+1)
(T G(n − 1, β, o) ⊕ i2) • (in−2 ⊕ T G(3, α, o))(x1, . . . , xβ
if x1 = ··· = xn−2 = o
n−1, xn, xn+1)
(T G(n − 1, β, o) ⊕ i2) • (in−2 ⊕ T G(3, α, o))(x1, . . . , xn−1, xn, xn+1)
otherwise
=
=
(T G(n − 1, β, o) ⊕ i2)(x1, . . . , xβ
(T G(n − 1, β, o) ⊕ i2)(x1, . . . , xβ
n−1, xn, xα
n+1)
if x1 = ··· = xn−2 = xβ
n−1, xn, xn+1)
n−1 = xn = o
if x1 = ··· = xn−2 = o ∧ (xβ
n−1 (cid:54)= o ∨ xn (cid:54)= o)
(T G(n − 1, β, o) ⊕ i2)(x1, . . . , xn−1, xn, xα
n+1)
(T G(n − 1, β, o) ⊕ i2)(x1, . . . , xn−1, xn, xn+1)
if some x1, . . . , xn−2 (cid:54)= o, xn−1 = xn = o
otherwise
30
=
(x1, . . . , xn−1, xn, xα
n+1)
(x1, . . . , xn−1, xn, xn+1)
(x1, . . . , xn−1, xn, xα
(x1, . . . , xn−1, xn, xn+1) otherwise
n+1)
n−1 = xn = o
if x1 = ··· = xn−2 = xβ
if x1 = ··· = xn−2 = o∧
n−1 (cid:54)= o ∨ xn (cid:54)= o)
(xβ
if some x1, . . . , xn−2 (cid:54)= o, xn−1 = xn = o
Note that fi(x1, . . . , xn+1) = xi for all i ≤ n.
Let g be the reduct
g(x1, . . . , xn) = µ((1, 2, . . . , n − 2, n, n + 1), κ(n − 1, p, f ))
From above we obtain, knowing that pβ = o and p (cid:54)= o
κ(n − 1, p, f )(x1, . . . , xn)
= f (x1, . . . , xn−2, p, xn−1, xn)
(x1, . . . , xn−2, p, xn−1, xα
n)
(x1, . . . , xn−2, p, xn−1, xn)
(x1, . . . , xn−2, p, xn−1, xα
n)
(x1, . . . , xn−2, p, xn−1, xn)
(x1, . . . , xn−2, p, xn−1, xn)
(x1, . . . , xn−2, p, xn−1, xα
n)
(x1, . . . , xn−2, p, xn−1, xn)
(x1, . . . , xn−2, p, xn−1, xα
n)
(x1, . . . , xn−2, p, xn−1, xn)
=
=
=
(cid:40)
Thus
if x1 = ··· = xn−2 = pβ = xn−1 = o
if x1 = ··· = xn−2 = o∧
(pβ (cid:54)= o ∨ xn−1 (cid:54)= o)
if xj (cid:54)= o∃j ≤ n − 2,
p = xn−1 = o [contradiction]
otherwise
if x1 = ··· = xn−2 = xn−1 = o
if x1 = ··· = xn−2 = o ∧ (xn−1 (cid:54)= o)
otherwise
if x1 = ··· = xn−2 = xn−1 = o
otherwise
(cid:40)
g(x1, . . . , xn) = µ((1, 2, . . . , n − 2, n, n + 1), κ(n − 1, p, f ))(x1, . . . , xn)
=
(x1, . . . , xn−2, xn−1, xα
n)
(x1, . . . , xn−2, xn−1, xn)
if x1 = ··· = xn−2 = xn−1 = o
otherwise
= T G(n, α, o)
Note that fn−1(x1, . . . , xn+1) = xn−1, so we have strong temporary stor-
(cid:3)
age, thus completing our claim.
31
Thus we see some essential differences between isomorphic realisation
and realisation with (strong) temporary storage. In the next section we will
further investigate the differences between certain types of realisation and
closure.
6 VARIOUS CLOSURES
The following result shows us how the various class and closure operators that
we have seen above relate and thus we will be able to determine information
about the relationships between various types of closure.
Theorem 6.1 KK = K, SS = S, CC = C, C∆C∆ = C∆, SC = CS,
KC = CK, C∆K = KC∆, C∆S = SC∆ and SK = KS.
Proof: The K and S operators are idempotent as a simple implication of
the definitions. The C and C∆ operators are idempotent because they are
algebraic closure operations.
To show that SC = CS we show that all Toffoli algebra operations com-
mute with s. We use the operation set {⊕, πα,◦k α ∈ S(N), k ∈ N}. We
start with the inclusion SC ⊆ CS. Let f ∈ Mn,m(A), g ∈ Ml,p(A), r ∈ N,
I ∈ {1, . . . , m + p}r without repetitions, α ∈ Sm.
Let J(cid:48) ⊆ {1, . . . , r} such that j ∈ J(cid:48) iff Ij ≤ m. Write J(cid:48) = {j1, . . . , jt}
in ascending order j1 < ··· < jt, then define J = (j1, . . . , jt) ∈ {1, . . . , r}t.
Then let I(cid:48) = (Iji i ∈ {1, . . . , t}) ∈ {1, . . . , m}t. Let {1, . . . , r} \ J =
{¯j1, . . . , ¯ju}. Then let I(cid:48)(cid:48) = (I¯ji − m i ∈ {1, . . . , u}) ∈ {1, . . . , p}u.
Finally define β ∈ Sr with
(cid:40)
β : i (cid:55)→ h such that
h = ji
h = ¯ji−t
if i ≤ t
if i > t
which can be written as
β =
(cid:18) 1
j1
. . .
. . .
t
jt
t + 1
¯j1
. . .
. . .
r
¯ju
(cid:19)
Note that i ≤ t iff Iβ(i) ≤ m and that Ii = I(cid:48)
I(cid:48)(cid:48)
(β−1i)−t + m otherwise. Then we claim that
β−1i if Ii ≤ m, Ii =
µ(I, f ⊕ g) = πβ(µ(I(cid:48), f ) ⊕ µ(I(cid:48)(cid:48), g))
32
The left hand side is
µ(I, f ⊕ g)i =
The right hand side is
(cid:40)
if Ii ≤ m
(f )Ii
(g)Ii−m if Ii > m
πβ(µ(I(cid:48), f ) ⊕ µ(I(cid:48)(cid:48), g))i = (µ(I(cid:48), f ) ⊕ µ(I(cid:48)(cid:48), g))β−1i
(cid:40)
(f )I(cid:48)
(cid:40)
(g)I(cid:48)(cid:48)
=
=
=
µ(I(cid:48), f )β−1i
µ(I(cid:48)(cid:48), g)(β−1i)−t
if β−1i ≤ t
if β−1i > t
if β−1i ≤ t
if β−1i > t
β−1 i
(β−1i)−t
if Ii ≤ m
(f )Ii
(g)Ii−m if Ii > m
which shows our claim.
It is a simple calculation that
µ(I, παf ) = µ(α−1(I), f )
where α−1 acts upon the entries in I, so (α−1I)i = α−1(Ii).
For composition, we use a similar argument to the ⊕ case above. Assume
that I is increasing. Let t be such that It < m and It+1 > m and let I(cid:48) =
(I1, . . . , It). Let I(cid:48)(cid:48) = (1, . . . , k) ⊕ (It+1 − m + k, . . . , Ir − m + k). We
claim that
µ(I, f ◦k g) = µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g)
The left hand side is
µ(I, f ◦k g)i =
(cid:40)
fIi ◦k g
gIi−m+k
The right hand side is
(µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g))i =
=
=
(cid:40)
(cid:40)
(cid:40)
if i ≤ t
if i > t
if i ≤ t
if i > t
◦k g
µ(I(cid:48), f )i ◦k µ(I(cid:48)(cid:48), g)
µ(I(cid:48)(cid:48), g)i−t+k
fI(cid:48)
gI(cid:48)(cid:48)
fIi ◦k g
gIi−m+k
i−t+k
i
if i ≤ t because of the first k entries in I(cid:48)(cid:48)
if i > t
if i ≤ t
if i > t
33
Which is what we wanted. If I is not increasing, then there is a permutation
β ∈ Sr such that πβI is increasing. Then
µ(I, f ◦k g) = πβ−1µ(πβI, f ◦k g) = πβ−1(µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g))
Thus we see that SC ⊆ CS.
For the converse, i.e. CS ⊆ SC, we use similar techniques. Let f ∈
Mn,m(A), g ∈ Ml,p(A), t, u ∈ N, I(cid:48) ∈ {1, . . . , m}t, I(cid:48)(cid:48) ∈ {1, . . . , p}u,
α ∈ St.
Let I = I(cid:48) ⊕ (I(cid:48)(cid:48) + m), i.e. Ii = I(cid:48)
Then it is a simple calculation to see that
i for i ≤ t, Ii = I(cid:48)(cid:48)
i−t + m for i > t.
µ(I(cid:48), f ) ⊕ µ(I(cid:48)(cid:48), g) = µ(I, f ⊕ g)
Let I ∈ {1, . . . , m}t, define I α
i = Iα−1i. Then we see that
παµ(I, f ) = µ(I α, f )
Let β ∈ Sp be defined by β−1(i) = I(cid:48)(cid:48)
i for i ≤ u, with the rest filled in to
make it a permutation. Because I(cid:48)(cid:48) contains no repeats, we know this can be
done. Let k ≤ min(n, u). Then we claim that
µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g) = µ(I(cid:48) ⊕ (m + 1, . . . , m + u − k), f ◦k (πβ ◦p g))
For i ≤ t we have
(µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g))i(x1, . . . , xl+n−k)
= fI(cid:48)
◦k µ(I(cid:48)(cid:48), g)(x1, . . . , xl+n−k)
(gI(cid:48)(cid:48)
(gβ−11(x1, . . . , xl), . . . , gβ−1k(x1, . . . , xl), xl+1, . . . , xl+n−k)
(x1, . . . , xl), xl+1, . . . , xl+n−k)
(x1, . . . , xl), . . . , gI(cid:48)(cid:48)
i
= fI(cid:48)
= (µ(I(cid:48) ⊕ (m + 1, . . . , m + u − k), f ◦k (πβ ◦p g)))i(x1, . . . , xl+n−k)
i
= fI(cid:48)
i
1
k
while for t < i ≤ m + u − k we have
(µ(I(cid:48), f ) ◦k µ(I(cid:48)(cid:48), g))i(x1, . . . , xl+n−k)
= gI(cid:48)(cid:48)
(x1, . . . , xl)
k+i−t
= gβ−1(k+i−t)(x1, . . . , xl)
= (πβ ◦p g)k+(i−t)(x1, . . . , xl)
= (f ◦k (πβ ◦p g))m+i−t(x1, . . . , xl+n−k)
= µ(I(cid:48) ⊕ (m + 1, . . . , m + u − k), f ◦k (πβ ◦p g))i(x1, . . . , xl+n−k)
34
Thus we have that CS ⊆ SC and thus SC = CS.
Similarly we show that KC = CK. We start with KC ⊆ CK. Let
f ∈ Mn,m(A), g ∈ Ml,p(A), i ∈ {1, . . . , n + l}, a ∈ A. It is a simple
application of the definitions to see that
κ(i, a, f ⊕ g) =
κ(i, a, f ) ⊕ g
f ⊕ κ(i − n, a, g)
if i ≤ n
if i > n
Let f ∈ Mn,m(A), α ∈ Sm, i ∈ {1, . . . , n}, a ∈ A. It is a simple calculation
that
κ(i, a, παf ) = πακ(i, a, f )
Let f ∈ Mn,m(A), g ∈ Ml,p(A), k ≤ min(n, p), i ∈ {1, . . . , n + l}, a ∈ A.
Applying the definitions gives us
κ(i, a, f ◦k g) =
f ◦k κ(i, a, g)
κ(i + k − n, a, f ) ◦k g
if i ≤ n
if i > n
For CK ⊆ KC we proceed as follows. Note that we must also include
the nonapplication of κ as a case here. Let f ∈ Mn,m(A), g ∈ Ml,p(A),
i1 ∈ {1, . . . , n}, i2 ∈ {1, . . . , l}, a1, a2 ∈ A. Then we claim that
κ(i1, a1, f ) ⊕ g = κ(i1, a1, f ⊕ g)
f ⊕ κ(i2, a2, g) = κ(i2 + n, a2, f ⊕ g)
κ(i1, a1, f ) ⊕ κ(i2, a2, g) = κ(i1, a1, κ(i2 + n, a2, f ⊕ g))
The first two follow by simple application of the definitions. The third claim
combines these two.
As we saw above, πα commutes with κ.
Let f ∈ Mn,m(A), g ∈ Ml,p(A), k ≤ min(n, p), i1 ∈ {1, . . . , n}, i2 ∈
{1, . . . , l}, a1, a2 ∈ A. Let β ∈ Sn be (i1, i1 + 1, . . . , n). Then we claim that
(cid:40)
(cid:40)
f ◦k κ(i2, a2, g) = κ(i2, a2, f ◦k g)
κ(i1, a1, f ) ◦k g =
(cid:40)
(cid:40)
κ(i1, a1, f ) ◦k κ(i2, a2, g) =
κ(l + n − k, a1, (f ◦n πβ) ◦k g)
if i1 ≤ k
κ(i1 + l − k, a1, f ◦k g)
if i1 > k
κ(i2, a2, κ(l + n − k, a1, (f ◦n πβ) ◦k g))
κ(i2, a2, κ(i1 + l − k, a1, f ◦k g))
if i1 ≤ k
if i1 > k
35
The first claim is direct from the definitions. The second requires some more
work. Let i ∈ {1, . . . , m + p − k}.
(κ(i1, a1, f ) ◦k g)i(x1, . . . , xn+l−k−1)
(cid:40)
=
=
=
=
=
fi(g1(x1, . . . , xl), . . . , gi1−1(. . . ), a1, gi1 (. . . ), . . .
. . . , gk(x1, . . . , xl), xl+1, . . . , xn+l−k−1)
fi(g1(x1, . . . , xl), . . . , gk(x1, . . . , xl), xl+1, . . .
. . . , xi1−k+l−1, a1, xi1+l−k, . . . , xn+l−k−1)
(κ(i1, a1, f )i ◦k g)(x1, . . . , xn+l−k−1)
gi−m+k(x1, . . . , xl)
if i ≤ m
if i > m
κ(l + n − k, a1, (f ◦n πβ) ◦k g)i(x1, . . . , xn+l−k−1)
gi−m+k(x1, . . . , xl)
(f ◦n πβ)i(g1(x1, . . . , xl), . . . , gi1−1(. . . ), gi1 (. . . ), . . .
. . . , gk(x1, . . . , xl), xl+1, . . . , xn+l−k−1, a1)
κ(i1 + l − k, a1, f ◦k g)i(x1, . . . , xn+l−k−1)
gi−m+k(x1, . . . , xl)
(cid:40)
κ(i1 + l − k, a1, f ◦k g)i(x1, . . . , xn+l−k−1)
gi−m+k(x1, . . . , xl)
κ(l + n − k, a1, (f ◦n πβ) ◦k g)i(x1, . . . , xn+l−k−1)
κ(i1 + l − k, a1, f ◦k g)i(x1, . . . , xn+l−k−1)
if i1 ≤ k, i ≤ m
if i1 > k, i ≤ m
i > m
if i1 ≤ k, i ≤ m
if i1 > k, i ≤ m
if i > m
if i1 ≤ k, i ≤ m
if i1 > k, i ≤ m
if i > m
if i1 ≤ k
if i1 > k
which is what we wanted. The third case is the combination of the first two
cases.
So every expression in CK can be written in KC, so we obtain CK =
KC.
We see that SK = KS because it makes no difference whether the inputs
are fed constants and then some outputs are ignored, or some outputs are
ignored and then some inputs are fed constants. To see this formally, let
f ∈ Mn,m(A), 1 ≤ i ≤ n, r ∈ N, I ∈ {1, . . . , m}r and a ∈ A.
µ(I, κ(i, a, f ))(x1, . . . , xn−1) = µ(I, f (x1, . . . , xi−1, a, xi, . . . , xn−1))
= (fj(x1, . . . , xi−1, a, xi, . . . , xn−1) j ∈ I)
= (κ(i, a, fj) j ∈ I)(x1, . . . , xn−1)
= κ(i, a, µ(I, f ))(x1, . . . , xn−1)
36
Lastly, we show that C∆ is well behaved by showing that ∆ and ∇ com-
mute with S and K. Let f ∈ Mn,m(A), r ∈ N, I ∈ {1, . . . , n}r and a ∈ A.
µ(I, ∆f )(x1, . . . , xn−1) = µ(I, f (x1, x1, x2, . . . , xn−1))
= (fj(x1, x1, x2, . . . , xn−1) j ∈ I)
= ∆(fj(x1, x2, . . . , xn−1) j ∈ I)
= ∆µ(I, f )(x1, . . . , xn−1)
µ(I,∇f )(x1, . . . , xn+1) = µ(I, f (x2, . . . , xn+1))
= (fj(x2, . . . , xn+1) j ∈ I)
= ∇µ(I, f )(x1, . . . , xn+1)
So we see that C∆S = SC∆.
Now let f ∈ Mn,m(A), 1 ≤ i ≤ n and a ∈ A.
κ(i, a, ∆f )(x1, . . . , xn−2)
= (∆f )(x1, . . . , xi−1, a, xi, . . . , xn−2)
= f (x1, x1, x2, . . . , xi−1, a, xi, . . . , xn−2)
f (a, a, x1, . . . , xn−2)
f (x1, x1, a, x2, . . . , xn−2)
f (x1, x1, x2, . . . , xi−1, a, xi, . . . xn−2)
if i = 1
if i = 2
if i > 2
(cid:40)
=
=
κ(1, a, κ(1, a, f ))
∆κ(i + 1, a, f )
if i = 1
if i ≥ 2
κ(i, a,∇f )(x1, . . . , xn) = (∇f )(x1, . . . , xi−1, a, xi, . . . , xn)
(cid:40)
(cid:40)
(cid:40)
=
=
=
(∇f )(a, x1, . . . , xn)
(∇f )(x1, x2, . . . , xi−1, a, xi, . . . , xn)
f (x1, . . . , xn)
f (x2, . . . , xi−1, a, xi, . . . , xn)
if i = 1
if i > 1
if i = 1
if i > 1
f (x1, . . . , xn)
∇κ(i − 1, a, f )(x1, . . . , xn)
if i = 1
if i > 1
Thus we have KC∆ ⊆ C∆K. For the converse we proceed as follows. Sup-
pose n = 1, then ∆κ(1, a, f ) = f (a) ∈ KC∆ and we are done. Suppose
37
⇒ ∆κ(i, a, f )(x1, . . . , xn−2)
f (x1, a, x2, . . . , xn−1)
f (x1, x2, . . . , xi−1, a, xi, . . . , xn−1)
f (a, x1, x2, . . . , xn−1)
f (a, x1, x1, x2 . . . , xn−2)
κ(2, a, ∆(f • π(1 2 3)))(x1, . . . , xn−2)
κ(2, a, ∆(f • π(2 3)))(x1, . . . , xn−2)
κ(i − 1, a, ∆f )(x1, . . . , xn−2)
f (x1, a, x1, x2, . . . , xn−2)
f (x1, x1, x2, . . . , xi−2, a, xi−1, . . . , xn−2)
=
=
=
if i = 1
if i = 2
if i > 2
if i = 1
if i = 2
if i > 2
if i = 1
if i = 2
if i > 2
n = 2, then κ(1, a, f ) and κ(2, a, f ) are both unary and thus fixed by ∆ so
we are done. Suppose n ≥ 3, we calculate
κ(i, a, f )(x1, . . . , xn−1)
∇κ(i, a, f )(x1, . . . , xn)
= κ(i, a, f )(x2, . . . , xn)
= f (x2, . . . , xi, a, xi+1, . . . , xn)
= (∇f )(x1, . . . , xi, a, xi+1, . . . , xn)
= κ(i + 1, a,∇f )(x1, . . . , xn)
So we see that KC∆ ⊆ C∆K ⊆ KC∆ so they are equal.
(cid:3)
Let F ⊆ M (A) be a collection of mappings. Then g ∈ SKC(F ) is
equivalent to saying that g is realised by F . We see that we have some inclu-
sions amongst the various closure operations introduced above, obtaining the
inclusion diagram in Figure 1.
We know that many of these inclusions are strict. For A of even order, we
know from Corollary 5.9 and Theorem 5.13 that C(F ) is strictly included in
TS(F ).
On the other hand, for specific classes of F , some closure classes fall
If F ⊆ B(A), then RC(F ) = C(F ), RTS(F ) = TS(F ) and
together.
RT (F ) = T (F ). Thus we obtain the second inclusion diagram, where we
also omit the ∆ operator.
One of the general problems is to look at the ways in which we can define
these various classes as a form of closure via a Galois connection. This has
been preliminarily investigated in [9] for C(F ) and T (F ) from reversible F .
38
KSC∆(F )
KSC(F )
KC∆(F )
SC∆(F )
RKSC(F )
T (F )
KC(F ) SC(F )
RT (F )
TS(F )
·
RTS(F )
C(F )
·
C∆(F )
RC(F )
KSC(F )
RKSC(F )
T (F ) = RT (F )
KC(F )
SC(F )
TS(F ) = RTS(F )
C(F ) = RC(F )
FIGURE 1
At the top we have inclusion of the various closure operations applied to a set of maps
F ⊆ M (A). When we are looking at a set F ⊆ B(A) on a finite A and are not
interested in ∆, we get the inclusions in the bottom diagram. Note that we know that
SC(F ) ∩ KC(F ) = C(F ) in this case.
39
7 CONCLUSION
We have presented a language based upon clone theory in order to discuss
systems of mappings Am → An on a set A. We have then written Toffoli's
model of reversible computation as a theory of mapping composition, finding
that larger alphabets leads to a more complex system than binary reversible
logic.
We see that the ideas can be used more generally. We are now confronted
with the spectrum of questions that arise naturally in clone theory and related
fields of general algebra, such as the size and structure of the lattice of multi-
clones and reversible Toffoli algebras on a given set. It would be of value to
determine a set of exioms for a Toffoli algebra or a reversible Toffoli algebra,
this might imprive the clarity of many of the proofs. The largest challenge
is to determine a suitable combinatorial structure to be used for invariance
type results of the Pol-Inv type for the five natural types of closure. It seems
probable that the results in [1, 9] will be of use to develop such a theory.
8 ACKNOWLEDGMENTS
I would like to thank Erhard Aichinger for many interesting conversations
and suggestions about developing and presenting this paper. Erkko Lehtonen
made many excellent and detailed suggestions that have vastly improved the
paper and for which I am extremely grateful.
REFERENCES
[1] S. Aaronson, D. Grier, and L. Schaeffer.
(2015). The classification of reversible bit
operations. Electronic Colloquium on Computational Complexity, (66).
[2] J.A. Bergstra, C.A. Middleburg, and G. S¸tefanescu.
(2013). Network algebra for syn-
chronous dataflow. https://arxiv.org/pdf/1303.0382.pdf.
[3] T. Boykett, J. Kari, and V. Salo. (2017). Finite generating sets for reversible gate sets under
general conservation laws. Theoretical Computer Science, pages -- . Electronic publication
Feb 2017.
[4] S. Burris and H. P. Sankappanavar. (1981). A course in universal algebra, volume 78 of
Graduate Texts in Mathematics. Springer-Verlag, New York-Berlin.
[5] M. Couceiro and E. Lehtonen. (2012). Galois theory for sets of operations closed under
permutation, cylindrification, and composition. Algebra Universalis, 67(3):273 -- 297.
[6] The GAP Group. (2013). GAP -- Groups, Algorithms, and Programming, Version 4.6.5.
[7] N. Gershenfeld.
(1996). Signal entropy and the thermodynamics of computation.
IBM
Systems Journal, 35(3 & 4).
[8] R. Jensen.
(2014). The logic of reversible structures. Master's thesis, University of
Copenhagen.
40
[9] E. Jer´abek, (September 2014). Answer to classifying reversible gates. Theoretical Com-
puter Science Stack Exchange. http://cstheory.stackexchange.com/questions/25730, Ac-
cessed June 2015.
[10] E. Lehtonen.
(2010). Closed classes of functions, generalized constraints, and clusters.
Algebra Universalis, 63(2-3):203 -- 234.
[11] A. I. Mal(cid:48)tsev.
(1976).
Iterativnye algebry Posta. Novosibirsk. Gos. Univ., Novosi-
birsk.
Seriya "Biblioteka Kafedry Algebry i Matematicheskoi Logiki Novosibirskogo
Universiteta", Vyp. 16. ["Library of the Department of Algebra and Mathematical Logic
of Novosibirsk State University" Series, No. 16].
[12] M. A. Nielsen and I. L. Chuang. (2000). Quantum computation and quantum information.
Cambridge University Press, Cambridge.
[13] R. Poschel and L. A. Kaluznin.
(1979). Funktionen- und Relationenalgebren [Func-
tion and Relation algebras], volume 15 of Mathematische Monographien [Mathemati-
cal Monographs]. VEB Deutscher Verlag der Wissenschaften, Berlin. Ein Kapitel der
diskreten Mathematik. [A chapter in discrete mathematics].
[14] G. S¸tefanescu. (2000). Network algebra. Discrete Mathematics and Theoretical Computer
[15]
Science (London). Springer-Verlag London, Ltd., London.
´A. Szendrei. (1986). Clones in universal algebra, volume 99 of S´eminaire de Math´ematiques
Sup´erieures [Seminar on Higher Mathematics]. Presses de l'Universit´e de Montr´eal, Mon-
treal, QC.
[16] T. Toffoli. (1980). Reversible computing. Technical Report MIT/LCS/TM-151, MIT.
[17] T. Toffoli.
In Automata, languages and programming
(Proc. Seventh Internat. Colloq., Noordwijkerhout, 1980), volume 85 of Lecture Notes in
Comput. Sci., pages 632 -- 644. Springer, Berlin-New York.
(1980). Reversible computing.
[18] G. Yang, X. Song, M. Perkowski, and J. Wu. (2005). Realizing ternary quantum switching
networks without ancilla bits. J. Phys. A, 38(44):9689 -- 9697.
41
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.