Artigdoplamencom Angela 2003
Artigdoplamencom Angela 2003
Artigdoplamencom Angela 2003
PLAMEN KOSHLUKOV∗
IMECC, UNICAMP, Cx. P. 6065, 13083-970 Campinas, SP, Brazil
plamen@ime.unicamp.br
ANGELA VALENTI†
Dipartimento di Matematica e Applicazioni
Università di Palermo, Via Archirafi, 34, 90123 Palermo, Italy
avalenti@math.unipa.it
Communicated by A. Ol’shanskii
We consider the algebra Un (K) of n × n upper triangular matrices over an infinite field
K equipped with its usual Zn -grading. We describe a basis of the ideal of the graded
polynomial identities for this algebra.
AMS Mathematics Subject Classification (2000): 16P90, 16R10, 16R20, 16R50, 16W50
0. Introduction
The interest into graded polynomial identities was inspired by their importance for
the structure theory of PI algebras (see for example [5] or [16]). Shortly afterwards
they became an object of independent studies. This in turn was determined by the
various applications that the graded identities can find in PI theory. We mention
some of the most important results concerning graded polynomial identities. Thus,
for example, in [3] and in [10], it was proved that if G is a finite group and A is a
G-graded algebra, then A is PI if and only if its 0-component is PI. In [3, 6–9, 12,
13], several numerical characteristics of T-ideals were transferred to the graded case
and lots of their properties were deduced. In [11], bases of the 2-graded identities
for several important algebras were described. As a corollary it was obtained a
rather elementary proof of the fact that the algebras M11 (G) and G ⊗ G satisfy the
517
December 2, 2003 17:17 WSPC/132-IJAC 00160
same polynomial identities when the base field is of characteristic 0. Here G stands
for the infinite dimensional Grassmann algebra, and M11 (G) is the algebra of the
2 × 2 matrices over G whose main diagonal elements are even (central) elements
of G, and whose other diagonal elements are odd elements of G. In [19], a basis of
the n-graded identities of the algebra of n × n matrices Mn (K) was found when
K is a field of characteristic 0, and in [20], this was extended to the Z-gradings of
Mn (K). Furthermore, in [17], the results of [11] were extended to algebras over an
infinite field of characteristic p 6= 2 (and as a consequence it was obtained a really
elementary proof of the coincidence of the T-ideals of M11 (G) and G ⊗ G). In [2], it
was established that the main result of [19] holds for algebras over an infinite field.
It is worth mentioning that the study of graded identities was one of the key
ingredients in Kemer’s methods for developing the structure theory of PI algebras
and in particular, for resolving positively the Specht problem in characteristic 0.
Graded identities, along with other kinds of “weaker” identities play an essential role
in study of the polynomial identities satisfied by concrete algebras. Some applica-
tions of graded identities satisfied by concrete algebras can be found in [11, 13, 17].
The polynomial identities satisfied by the algebra Un (K) of the n × n upper
triangular matrices are of particular interest. It is well known that for every field
K and every n they are finitely based (as T-ideal) (see for example [21]). Thus,
when the field K is infinite, the T-ideal of Un (K) is generated by the identity
[x1 , x2 ][x3 , x4 ] · · · [x2n−1 , x2n ] where [a, b] = ab − ba is the usual commutator. It is
also well known that the identities of Un (K) are closely related to the problem of
the description of the subvarieties of the variety of associative algebras generated
by M2 (K). Since the latter is still open when charK 6= 0, it is important to obtain
further information about the identities in Un (K). A detailed description of the
2-graded identities satisfied by the algebra U2 (K) when charK = 0 is given in [18].
In this paper we describe a basis of the n-graded polynomial identities for the
algebra Un (K) over an infinite field K and as an application we compute the
asymptotics for the sequence of its graded codimensions.
These results generalize the ones in [18]. We hope they can be useful in describing
the behaviour of the subvarieties of the variety generated by the matrix algebra of
order 2 over an infinite field of characteristic, not 2.
1. Preliminaries
We fix an infinite field K, and consider all algebras (graded and ungraded), vector
spaces etc. over K. Denote by Un (K) the algebra of n×n upper triangular matrices
over K i.e., n × n matrices with zero entries below the main diagonal. The algebra
Un (K) has a natural Zn -grading Un (K) = V0 ⊕ V1 ⊕ · · · ⊕ Vn−1 with
where the eij are the elementary matrices having 1 as (i, j)th entry and 0 elsewhere.
In particular V0 consists of all diagonal matrices. Clearly all Vi are subspaces of
December 2, 2003 17:17 WSPC/132-IJAC 00160
Proof. The identity x01 x02 − x02 x01 ≡ 0 holds in Un (K) since two diagonal ma-
trices commute. Moreover since i + j ≥ n implies Vi Vj = 0, the graded identities
xi1 1 xi2 2 ≡ 0 with i1 + i2 ≥ n, hold for Un (K) as well.
Denote by I the ideal of n-graded identities generated by the identities (1) and
(2), and set J = Idgr (Un (K).
We wish to show that I = J. We start with an obvious observation.
Remark 2.1. For any i1 +i2 +· · ·+ik ≥ n, xi1 1 xi2 2 · · · xik k ≡ 0 is a graded identity
of Un (K).
Proof. We first claim that the monomials (3) are linearly independent modulo
Idgr (Un (K)).
αi ui ∈ Idgr (Un (K)) for some αi ∈ K where the
P
In fact suppose that f =
monomials ui are of the type (3). Since K is an infinite field, graded ideals are
multihomogeneous. Hence we may assume that all ui are multihomogeneous of
the same degree sequence, i.e. the same variables appear in each monomial ui .
Fix one monomial with nonzero coefficient say u1 and let u1 = w0 xk1 i1 w1 · · · wt−1
xkt it wt . We now evaluate f on Un (K) as follows: evaluate each variable appear-
ing in w0 , into e11 , the variable xk1 i1 into e1,k1 +1 , each variable appearing in w1
into ek1 +1,k1 +1 , the variable xk2 i2 into ek1 +1,k1 +k2 +1 , . . . , the variable xkt it into
ek1 +···+km−1 +1,k1 +···+km +1 and each variable in wt into ejj where j = k1 + · · · +
km + 1. It follows that u1 evaluates into e1j and all other monomials of f evaluate
to zero. Thus f 6≡ 0 on Un (K), a contradiction.
Next we claim that, modulo h[x01 , x02 ], xi1 xj2 |i + j ≥ niTn , every element of
K(X) can be written as a linear combination of monomials of type (3). In fact, this
is clear since Tn -ideals are invariant under homogeneous substitutions.
It follows that the monomials (3) are a basis of K(X) modulo Idgr (Un (K) and
that Idgr (Un (K)) = h[x01 , x02 ], xi1 xj2 , i + j ≥ niTn .
December 2, 2003 17:17 WSPC/132-IJAC 00160
We shall work in the algebra Gn instead of the relatively free algebra. We use
some ideas from [2] and [19].
in Gn have the same nonzero entries in the same positions in their first rows. (In
other words, M1 − M2 has zeros in its first row.) Then M1 = M2 .
Proof. We know that the (unique) nonzero entry in the first row occurs in position
(1, r + 1), and the same entry occurs in M2 , at the same position (1, r + 1), for some
r, 0 ≤ r ≤ n − 1. One computes these entries as follows. For M1 , we have that
the (1, r + 1)st entry equals ya1 ,b1 ,c1 ya2 ,b2 ,c2 · · · yak ,bk ,ck where a1 = 1, b1 = i1 + 1,
c1 = j1 , and ct = jt for all t; thus one obtains the recurrence formula
Corollary 3.1. Under the assumptions and in the notation of the previous lemma,
we have m1 − m2 ∈ Idgr (Un (K)).
Using the properties of the relatively free graded algebra we give an alternative
proof of Theorem 2.1.
Proof of Theorem 2.1. Let I be the Tn -ideal generated by the polynomials (1)
and (2), and let J = Idgr (Un (K)). Since I ⊆ J, it suffices to prove that J ⊆ I.
Suppose on the contrary that there exists a multihomogeneous polynomial f ∈ J
and f 6∈ I. We work in the relatively free n-graded algebra K(X)/J = Gn and
choose f ∈ Gn of minimal degree, and among these f , choose one that is expressed
in the form f = α1 m1 + α2 m2 + · · · + αs ms for mt being distinct monomials in Gn ,
all αt 6= 0, αt ∈ K, and s the least possible. Hence s ≥ 1.
Suppose mt = mt (xi1 j1 , xi2 j2 , . . . , xik jk ). Then
s
X
m1 (Yi1 j1 , Yi2 j2 , . . . , Yik jk ) = βz mz (Yi1 j1 , Yi2 j2 , . . . , Yik jk )
z=2
where βz = −αz /α1 6= 0, z = 2, 3, . . . , s.
On the other hand we have that m1 (Yi1 j1 , Yi2 j2 , . . . , Yik jk ) 6= 0 and in the first
row of this matrix there will be some nonzero entry. This nonzero entry appears
on the right hand side as well. Say it comes from the monomial m2 . But then the
monomials m1 and m2 have the same nonzero entry in the same position on their
first rows, and hence m1 − m2 ∈ J, according to Corollary 3.1. But then m1 = m2 ,
and we reduce f to s − 1 monomials which contradicts the choice of s. Hence s = 0
and I = J.
We now prove that the monomials (3) are linearly independent modulo
Pt
Idgr (Un (K)). Suppose that the polynomial i=1 αi mi ∈ Idgr (Un (K)) where mi
are distinct monomials of the type (3), for 0 6= αi ∈ K. Since K is infinite we may
assume that the monomials mi are all multihomogeneous and of the same multide-
gree. One can express m1 in the following way: m1 = − ti=2 βi mi . Now substitute
P
the variables for the respective generic graded matrices; the first row of m1 will
contain some nonzero entry since otherwise m1 would belong to Idgr (Un (K)). The
same nonzero entry must appear in some of the monomials on the right hand side,
say in m2 . Hence m1 − m2 ∈ Idgr (Un (K)), and we reduce our linear combination to
t − 1 terms. Finally we obtain single monomial, but it cannot be a graded identity
for Un (K) since it is of weight < n, and it will contain in its first row some nonzero
entry when evaluated on Gn .
4. Applications
Here we apply the results of the previous section and describe some of the numerical
invariants of the Tn -ideal of Un (K). We start with one of the various kinds of
codimensions one may define for graded identities.
December 2, 2003 17:17 WSPC/132-IJAC 00160
Proposition 4.1. Let m be a fixed positive integer and let x0j1 , x0j2 , . . . , x0js be
fixed. Suppose further that xz1 ,r1 , xz2 ,r2 , . . . , xzg ,rg , g = m − s, are fixed, 1 ≤ zi ≤
n−1 for all i and z1 +z2 +· · ·+zg ≤ n−1. Then the span of all multilinear monomials
on these variables in K(X)/Idgr (Un (K)) is of dimension (m − s)!(m − s + 1)s .
One can consider another kind of graded codimensions (see for example [3]).
We recall the definition of these codimensions that represent a direct generalization
of the ungraded case. Let m be fixed positive integer, and consider the (graded)
variables xij , 0 ≤ i ≤ n, 1 ≤ j ≤ m, in K(X). We consider the multilinear
monomials in these variables that are of the form
xii ,j1 xi2 ,j2 · · · xim ,jm (4)
where {j1 , j2 , . . . , jm } = {1, 2, . . . , m} and 0 ≤ it ≤ n − 1.
In other words, we do not admit repetitions of the second indices in the variables.
These monomials can be obtained by the usual multilinear monomials xj1 xj2 · · · xjm
by assigning all possible weights on the variables. It is straightforward that the
n (n)
span Pm of these monomials in K(X) is of dimension cm = nm m!. Denote by
Pmn
(A) the quotient Pm n
/(Pmn
∩ Idgr (A)) where A is an n-graded algebra, and set
(n) n (n)
cm (A) = dimK Pm (A). Next we compute the codimensions cm = cm (Un (K)).
n
Proof. Let f = g1 y1 g2 y2 · · · gk yk gk+1 ∈ Pm (Un (K)) be a nonzero multilinear
monomial in the variables xij , 0 ≤ i ≤ n−1, 1 ≤ j ≤ m where for i = 1, . . . , q+1, the
gi0 s are monomials (possibly empty) in the variables x0j only, and for r = 1, . . . , k,
December 2, 2003 17:17 WSPC/132-IJAC 00160
n
yr = xar ,br , with 1 ≤ ar ≤ n − 1, 1 ≤ br ≤ m. Since f ∈ Pm (Un (K)), we have to
impose a1 + a2 + · · · + aq ≤ n − 1.
Denote by An−1 (q) the number of q-tuples of positive integers a1 , a2 , . . . , aq such
that a1 + a2 + · · · + aq ≤ n − 1, and as Bn−1 (q) the number of such q-tuples with
a1 + a2 + · · · + aq = n − 1. Then obviously An−1 (q) = Bn−1 (q) + Bn−2 (q) + · · · +
r−1
Bq (q). On the other hand, Br (q) = q−1 is the number of compositions of r into
q parts (see for example [1, p. 54]). Hence, using some elementary transformations
on binomial coefficients, we obtain
n−1 n−1
! !
X X q+r−1 n−1
An−1 (q) = Br (q) = = .
r=q r=q q−1 q
where
! !
m−q
m n−1
C(k, q) = (k + 1) q! = .
q q
Therefore
q
n−1
! !
XX m n−1
cm = q! = ((k + 1)m−q − k m−q )
q=1 k=0
q q
m
! !
X m n−1
= = q!(q + 1)m−q .
q=0
q q
Recall that if f (x) and g(x) are two functions of a natural argument then f (x)
(x)
and g(x) are asymptotically equal, and we write f (x) ' g(x) if limx→∞ fg(x) = 1.
References
1. G. Andrews, The theory of partitions, Ency. Math. Appl., Vol. 2, Addison–Wesley,
1976.
2. S. S. Azevedo, Graded identities for the matrix algebra of order n over an infinite
field, Commun. Algebra 30(12) (2002), 5849–5860.
3. Yu. Bahturin, A. Giambruno and D. Riley, Group graded algebras satisfying a
polynomial identity, Isr. J. Math. 104 (1998), 145–155.
4. Yu. Bahturin, A. Giambruno and M. Zaicev, Codimension growth and graded
identities, Algebra (Moscow, 1998), 57–76, de Gruyter, Berlin, 2000.
5. A. Berele, Magnum PI, Isr. J. Math. 51(1–2) (1985), 13–19.
6. A. Berele, Cocharacters of Z/2Z-graded algebras, Isr. J. Math. 61 (1988), 225–234.
7. A. Berele, Trace identities and Z/2Z-graded invariants, Trans. AMS 309(2) (1988),
581–589.
8. A. Berele, Supertraces and matrices over Grassmann algebras, Adv. Math. 108 (1994),
77–90.
9. A. Berele and L. Rowen, T-ideals and super-Azumaya algebras, J. Algebra 212 (1999),
703–720; Article ID jabr.1998.7633.
10. J. Bergen and M. Cohen, Actions of commutative Hopf algebras, Bull. London
Math. Soc. 18(2) (1986), 159–164.
11. O. M. Di Vincenzo, On the graded identities of M1,1 (E), Isr. J. Math. 80(3) (1992),
323–335.
12. O. M. Di Vincenzo, Cocharacters of G-graded algebras, Commun. Algebra 24(10)
(1996), 323–335.
13. O. M. Di Vincenzo and V. Nardozza, Graded polynomial identities for tensor products
by the Grassmann algebra, Commun. Algebra 31(3) (2002), 1453–1474.
14. A. Giambruno and M. Zaicev, On codimension growth of finitely generated associative
algebras, Adv. Math. 140(2) (1998), 145–155.
15. A. Giambruno and M. Zaicev, Exponential codimension growth of PI algebras: an
exact estimate, Adv. Math. 142(2) (1999), 221–243.
16. A. Kemer, Ideals of identities of associative algebras, in Translations of Mathematics
Monographs, Vol. 87, Amer. Math. Soc., Providence RI, 1991.
17. P. Koshlukov and S. Azevedo, Graded identities for T-prime algebras over fields of
positive characteristic, Isr. J. Math. 128 (2002), 157–176.
December 2, 2003 17:17 WSPC/132-IJAC 00160
18. A. Valenti, On graded identities of upper triangular matrices of size two, J. Pure
Appl. Algebra 172 (2002), 325–335.
19. S. Yu. Vasilovsky, Zn -graded polynomial identities of the full matrix algebra of order
n, Proc. AMS 127(12) (1999), 3517–3524.
20. S. Yu. Vasilovsky, Z-graded polynomial identities of the full matrix algebra, Commun.
Algebra 26(2) (1998), 601–612.
21. S. M. Vovsi, Triangular products of group representations and their applications, in
Progress in Mathematics, Vol. 17, Birkhäuser, Boston, MA, 1981.