Z-Graded Identities of The Lie Algebra

Journal of Algebra 427 (2015) 226–251
Contents lists available at ScienceDirect
Journal of Algebra
www.elsevier.com/locate/jalgebra
Z-graded identities of the Lie algebra W1

José A. Freitas a,∗ , Plamen Koshlukov b , Alexei Krasilnikov a
a
Departamento de Matemática, Universidade de Brasília,
70910-900 Brasília, DF, Brazil
b
Sergio Buarque de Holanda 651, IMECC, UNICAMP,
13083-859 Campinas, SP, Brazil
a r t i c l e i n f o a b s t r a c t
Article history: Let K be a field of characteristic 0 and let W1 be the Lie

Received 29 May 2014 algebra of the derivations of the polynomial ring K[t]. The
Available online 15 January 2015 algebra W1 admits a natural Z-grading. We describe the
Communicated by E.I. Khukhro
graded identities of W1 for this grading. It turns out that
MSC: all these Z-graded identities are consequences of a collection
16R10 of polynomials of degree 1, 2 and 3 and that they do not
16R50 admit a finite basis. Recall that the “ordinary” (non-graded)
17B01 identities of W1 coincide with the identities of the Lie algebra
17B70 of the vector fields on the line and it is a long-standing open
problem to find a basis for these identities. We hope that our
Keywords: paper might be a step to solving this problem.
Graded identities © 2015 Elsevier Inc. All rights reserved.
Graded Lie algebra
Graded T-ideal
Basis of identities
Infinite basis of graded identities
1. Introduction
The development of the structure theory for the T-ideals in the free associative algebra
was one of the major achievements in the PI theory. This development is mainly asso-
* Corresponding author.
E-mail addresses: jfreitas@mat.unb.br (J.A. Freitas), plamen@ime.unicamp.br (P. Koshlukov),
alexei@unb.br (A. Krasilnikov).
http://dx.doi.org/10.1016/j.jalgebra.2014.12.023
0021-8693/© 2015 Elsevier Inc. All rights reserved.
J.A. Freitas et al. / Journal of Algebra 427 (2015) 226–251 227
ciated with the results obtained by Kemer in the 1980-ies. Details concerning Kemer’s
fundamental contributions to this theory can be found in several monographs [9,7,8].
The key points in Kemer’s theory are the description of the T-ideals in terms of identi-
ties satisfied by certain “good” algebras: finitely generated and finite dimensional ones,
as well as the connections between ordinary and graded identities. Soon after Kemer’s
work graded algebras and graded polynomial identities became objects of independent
interest. The reader may find an extensive list of references concerning recent results
about gradings on algebras and their graded identities in [3].
It is somewhat surprising that graded identities for Lie algebras have not been studied
to that extent. (This is not the case with Lie superalgebras where the proper defining
relations involve gradings.) The group gradings for many finite dimensional simple Lie
algebras over an algebraically closed field K of characteristic = 2 were classified in [2],
see also [4]. However, the only nontrivial instance where we know exactly the graded
identities is that of sl 2 (K) whenever K is an infinite field and char K = 2. The relatively
free algebras corresponding to the graded identities satisfied by sl 2 (K) endowed with each
of the three possible nontrivial gradings were described in [15]. Bases of these graded
identities were exhibited in [10] (it should be mentioned that the results in [10] hold for
K infinite, char K = 2), see a streamlined and simplified version in [11]. Recently, it
was proved in [5] that over a filed K of characteristic 0 the T-ideal of graded identities
of sl 2 (K) has the Specht property. The growth of the variety of graded Lie algebras
generated by sl 2 (K) was found in [6].
In [12] it was proved that over an infinite field K of characteristic 2 the Z2 -graded
identities of the Lie algebra gl 2 (K) do not admit any finite basis. Moreover an example
of a just nonfinitely based variety of Z2 -graded algebras was constructed in [12]. (Recall
that a just nonfinitely based variety is a variety without finite basis of its identities whose
proper subvarieties admit such a finite basis.)
Let K be a fixed field of characteristic 0 and let G be a group. An algebra (not

necessarily associative) A over K is G-graded if A = g∈G Ag , a direct sum of vector
subspaces such that Ag Ah ⊆ Agh for every g, h ∈ G. If G = Z is the additive group of
the integers then A is Z-graded. In this case Ag Ah ⊆ Ag+h .
Now consider the polynomial algebra K[t] in one variable t. The derivations of the
algebra K[t] form a Lie algebra denoted by W1 , W1 = Der(K[t]). It is immediate that
the elements en = tn+1 d/dt, n ≥ −1, form a basis of W1 . The Lie algebra structure on
the vector space W1 is given by the multiplication
[ei , ej ] = (j − i)ei+j . (1)

The algebra W1 is Z-graded, W1 = i∈Z Li where Li = 0 whenever i ≤ −2, and Li is
the span of ti+1 d/dt if i ≥ −1. The algebra W1 is known as the Witt algebra and plays
an important role in the applications.
One denotes by Wn the Lie algebra of derivations of the polynomial algebra
K[x1 , . . . , xn ], n ≥ 1. It is well known that the Lie algebras Wn form one of the se-
228 J.A. Freitas et al. / Journal of Algebra 427 (2015) 226–251
ries of infinite dimensional simple Lie algebras (the so-called algebras of Cartan type).
They satisfy non-trivial identities, see for example [1]. The identities in these algebras
were studied by Mishchenko, see [13], by Razmyslov, see for example [14, Chapter 6],
and by many others (see the bibliography of [14]). It is well known that the algebras Wn
satisfy the standard Lie identity of some degree but do not satisfy any system of Capelli
identities. (Recall that in associative algebras the standard identity implies some system
of Capelli identities.)
From now on we consider Lie algebras graded by Z. Let X i = {xi1 , xi2 , . . .} (i ∈ Z)

be countable infinite disjoint sets, and put X = i∈Z X i . Denote by L(X) the free Lie
algebra freely generated over K by the set X. Sometimes we shall use other letters like
y, z, t, with or without indices, to denote the free generators of L(X). We define a
Z-grading on L(X) as follows. The elements of X i are of degree i, and if m, n ∈ L(X)
are of degrees i and j respectively then [m, n] is of degree j + i. We shall consider the
commutators left-normed, that is [u, v, w] = [[u, v], w]. Then we put L(X)i to be the
span of all elements of degree i in L(X). It is immediate that this is indeed a Z-grading.
A polynomial f (xij11 , . . . , xijkk ) ∈ L(X) is a graded identity of the Z-graded Lie algebra

L = r∈Z Lr if f (a1 , . . . , ak ) = 0 in L for every choice of at ∈ Lit . Sometimes an
expression f (xij11 , . . . , xijkk ) = 0 will be called a graded identity of L as well. The set TZ (L)
of all graded identities of L is a graded T-ideal in L(X). Indeed, it can be easily seen
that it is a graded ideal and it is closed under all Z-graded endomorphisms of L(X).
The set B = {fi | i ∈ I} ⊆ TZ (L) of graded identities of L is a basis for its graded
identities if B generates TZ (L) as a graded T-ideal. If f and g are graded identities for
L then g follows from (or is a consequence of) f if g lies in the graded T-ideal generated
by f . Analogously, f and g are equivalent as graded identities if f follows from g and
g follows from f . In other words, f and g are equivalent if f and g generate the same
graded T-ideal.
Our main result is the following theorem.
Theorem 1. Let K be a field of characteristic 0. The graded identities

i i
x1 , x2 = 0, i ≥ −1 (2)
a b c a c b
α x1 , x2 , x3 − β x1 , x3 , x2 = 0, (3)
xd = 0, d ≤ −2 (4)
where α = (c −a)(b −c −a), β = (b −a)(c −b −a), form a basis for the Z-graded identities
of the Lie algebra W1 over K.
Furthermore we prove that the graded identities of W1 do not admit any finite basis.
To be more precise, we show (see Theorem 21) that the graded identities (4) and most
of (2) and (3) are independent.
Here we recall that the “ordinary” identities of W1 coincide with the identities of the
Lie algebra of the vector fields on the line (if K = R). The variety V1 of Lie algebras
defined by these identities and the relatively free algebras of V1 were studied in various
papers (see [13] and the bibliography therein). It is well-known and easy to check that
the (ordinary) identity

sgn(σ)[y0 , yσ(1) , yσ(2) , yσ(3) , yσ(4) ] (5)
σ∈S4
is satisfied in W1 . On the other hand, it is a long-standing open problem to determine

whether or not all the identities satisfied in W1 are consequences of (5), that is, whether
or not the identity (5) forms a basis for the identities of W1 . We hope that our result
about the Z-graded identities of W1 will contribute to the solution of this problem.
2. Proof of Theorem 1
Our algebras are over a field of characteristic 0. Therefore, every graded T-ideal is
generated by its multilinear elements. Thus, we shall work only with multilinear poly-
nomials.
Let a1 , . . . , an be integers such that a1 , . . . , an ≥ −1 and a1 + · · · + ak ≥ −1 for all k,
2 ≤ k ≤ n. It can be easily deduced from (1) that
[ea1 , . . . , ean ] = αn ea1 +···+an (6)
where αn = (a2 − a1 )(a3 − a1 − a2 ) . . . (an − a1 − · · · − an−1 ). Define ed = 0 if d ≤ −2.

The next lemma shows that (6) remains valid if a1 + · · · + ak ≤ −2 for some k provided
that a1 , . . . , an ≥ −1 (if al ≤ −2 for some l then (6), in general, is false).
Lemma 2. The equality (6) is valid for all a1 , . . . , an ≥ −1 (n ≥ 2).
Proof. Induction on n. If n = 2 then Lemma 2 holds by (1). Suppose that n > 2. Then,
by the inductive hypothesis, for all k < n we have
[ea1 , . . . , eak ] = αk ea1 +···+ak .
Suppose that a1 + · · · + an−1 ≥ −1. By (1), we have
[ea1 , . . . , ean ] = αn−1 [ea1 +···+an−1 , ean ]

= αn−1 (an − a1 − · · · − an−1 )ea1 +···+an = αn ea1 +···+an ,
as required.
Now suppose that a1 + · · · + an−1 ≤ −2. Take the least k, 2 ≤ k < n, such that
a1 + · · · + ak ≤ −2. It can be easily seen that a1 + · · · + ak−1 = −1 and ak = −1. On
one hand, we have
[ea1 , . . . , eak−1 , eak ] = αk−1 [ea1 +···+ak−1 , eak ] = αk−1 [e−1 , e−1 ] = 0
so [ea1 , . . . , ean ] = 0. On the other hand, ak − a1 − · · · − ak−1 = 0 so αn = 0. Thus,
[ea1 , . . . , ean ] = 0 = αn ea1 +···+an ,
as required. 2
Lemma 3. The graded identities (2), (3), and (4) hold for W1 .

Proof. Let W1 = i∈Z Li as above. For each d ≤ −2, the identity (4) is satisfied since
Ld = 0. Also the identities (2) hold since dim Li = 1 when i ≥ −1.
Now we check that the identities (3) hold in W1 . Note that if a ≤ −2 then xa1 = 0
so [xa1 , xb2 , xc3 ] = [xa1 , xc3 , xb2 ] = 0 and (3) holds. Similarly, (3) holds if b ≤ −2 or c ≤ −2.
Suppose a, b, c ≥ −1. Then to check that the identity (3) holds in W1 it suffices to deduce
the equality
α[ea , eb , ec ] − β[ea , ec , eb ] = 0, (7)
where α and β are as in (3). It can be easily seen, using Lemma 2 that [ea , eb , ec ] =
βea+b+c and [ea , ec , eb ] = αea+b+c . It follows that
α[ea , eb , ec ] − β[ea , ec , eb ] = αβea+b+c − βαea+b+c = 0.
Thus the equality (7) holds and so does the identity (3). 2
We denote by I the graded T-ideal generated by the polynomials on the left-hand side
of (2), (3) and (4). By Lemma 3, we have I ⊆ TZ (W1 ). To prove Theorem 1 one has to
prove that I = TZ (W1 ).
Lemma 4. Let m = [xai11 , . . . , xainn ] ∈ L(X) be a multilinear commutator. Suppose that

m ∈ TZ (W1 ). Then m ∈ I.
Proof. If al ≤ −2 for some l, 1 ≤ l ≤ n, then m belongs to the graded T-ideal in

L(X) generated by {xd | d ≤ −2}. Suppose that al ≥ −1 for all l, 1 ≤ l ≤ n. Since
m ∈ TZ (W1 ), we have [ea1 , . . . , ean ] = 0. It follows from Lemma 2 that αn ea1 +···+an = 0
so either αn = 0 or ea1 +···+an = 0. In the former case al − a1 − · · · − al−1 = 0 for
some l, 2 ≤ l ≤ n. It follows that [xai11 , . . . , xaill ] belongs to the graded T-ideal generated
by {[xi1 , xi2 ] | i ≥ −1} and so does m. In the latter case a1 + · · · + an ≤ −2 so m belongs
to the graded T-ideal generated by {xd | d ≤ −2}.
Thus, in all cases m ∈ I, as required. 2
We write xai > xbj if either a > b or a = b and i > j. It is well known (and can be easily
proved by induction) that every multilinear commutator m = [xbj11 , . . . , xbjnn ] ∈ L(X) can
be written as a linear combination of left-normed commutators starting with any fixed

variable xbjll (1 ≤ l ≤ n). Therefore we can suppose without loss of generality that every
multilinear element of L(X) is a linear combination of commutators [xai11 , . . . , xainn ] where
xai11 > xaill for all l > 1.
Now we define 5 types of commutators and, for each type, we define the commutators
of canonical form. Let m = [xai11 , . . . , xainn ] (n ≥ 3) be a multilinear commutator such that
m∈ / I and xai11 > xaill for all l > 1.
We say that m is of type 1 if a1 > 1 and there is l, 2 ≤ l ≤ n, such that 0 < al < a1 .
A commutator m of type 1 is of canonical form if

m = xai11 , xai22 , . . . , xairr , x−1 −1
ir+1 , . . . , xin ,
where a2 ≥ 0, 2 ≤ r ≤ n, xai22 < xai33 < . . . < xairr < xai11 , ir+1 < . . . < in and 0 < al < a1
for some l, 2 ≤ l ≤ r. It might happen that r = n; in this case m does not contain
variables of Z-degree −1.
We say that m is of type 2 if a1 > 1, al ∈ {−1, 0, a1 } for all l > 1 and there is l > 1
such that al = a1 . A commutator m of type 2 is of canonical form if

m = xai1 , x0i2 , . . . , x0ir , x−1 a a −1 −1
ir+1 , xir+2 , . . . , xis , xis+1 , . . . , xin ,
where a = a1 > 1, r ≥ 1, r + 2 ≤ s ≤ n, i2 < . . . < ir , ir+1 < is+1 < . . . < in ,

ir+2 < . . . < is < i1 .
A commutator m is of type 3 if a1 > 1 and al ∈ {−1, 0} for all l > 1; such m is of
canonical form if

m = xai11 , x0i2 , . . . , x0ir , x−1 −1
ir+1 , . . . , xin ,
where 1 ≤ r ≤ n and i2 < · · · < ir , ir+1 < · · · < in .

A commutator m is of type 4 if a1 = 1 and of type 5 if a1 = 0. A commutator of
type 5 is of canonical form if

m = x0i1 , x−1 0 0
i2 , xi3 , . . . , xin ,
where i3 < i4 < · · · < in < i1 .

A commutator m = [x1i1 , xai22 , xai33 ] of type 4 and of length 3 is of canonical form if
⎧
⎨ (0, −1);
⎪
(a2 , a3 ) = (−1, 1) and i3 < i1 ;
⎪
⎩
(0, 0) or (−1, −1), i2 < i3 .
A commutator m of type 4 and of length n ≥ 4 is of canonical form if

m = x1i1 , x0i2 , . . . , x0ir , x−1 −1
1 1 an−2 an−1 an
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−2 , xin−1 , xin .
Here 1 ≤ r ≤ n, i2 < · · · < ir ,

⎧
⎪
⎪ (1, −1, −1) or (1, −1, 1) if r ≤ n − 3 and (n − r) is even;
⎪
⎪
⎪
⎪
⎨ (−1, 1, −1)
⎪ if r ≤ n − 3 and (n − r) is odd;
(an−2 , an−1 , an ) = (0, −1, −1) or (0, −1, 1) if r = n − 2;
⎪
⎪
⎪ (0, 0, −1)
⎪ if r = n − 1;
⎪
⎪
⎪
⎩
(0, 0, 0) if r = n,
and the following hold:
if (an−2 , an−1 , an ) = (1, −1, −1) then ir+1 < ir+3 < · · · < in−3 < in−1 < in , ir+2 <
ir+4 < · · · < in−2 < i1 ;
if (an−2 , an−1 , an ) = (1, −1, 1) then ir+1 < ir+3 < · · · < in−3 < in−1 , ir+2 < ir+4 <
· · · < in−2 < in < i1 ;
if (an−2 , an−1 , an ) = (−1, 1, −1) then ir+1 < ir+3 < · · · < in−2 < in , ir+2 < ir+4 <
· · · < in−1 < i1 ;
if (an−2 , an−1 , an ) = (0, −1, −1) then in−1 < in ;
if (an−2 , an−1 , an ) = (0, −1, 1) then in < i1 .
Note that
i) each multilinear commutator m = [xai11 , . . . , xainn ] (n ≥ 3) such that m ∈

/ I and
xai11 > xaill for all l > 1 is of a certain type i, 1 ≤ i ≤ 5;
ii) for each such commutator m, there is exactly one multilinear commutator m1 of
canonical form on the same set of variables {xai11 , . . . , xainn }.
Indeed, it is straightforward to check that for each m described in i) we have a1 ≥ 0

because otherwise m ∈ I; the assertion i) follows.
Further, one can easily check that for a given set of variables {xai11 , . . . , xainn } there is
at most one multilinear commutator m1 of canonical form on this set of variables. On
the other hand, for each m described in i) there exists a multilinear commutator m1 of
canonical form on the same set of variables. Indeed, this is clear if m is of type 1 or 3: to
get m1 one just permutes the entries xaill in m in a suitable way. If m is of type 2 then to
construct a multilinear commutator m1 of canonical form on the same set of variables
one needs al = −1 for some l > 1; however, this holds because otherwise m ∈ I. If m is
of type 5 then a2 = −1 and al = 0 for l > 2 (otherwise m ∈ I) so a commutator m1 of
canonical form on the same set of variables exists.
Suppose that m is a multilinear commutator described in i) of type 4. Note that if,
for some l, 1 < l < n, the commutator m = [xai11 , . . . , xaill ] is of Z-degree 1 then al+1 = 1
(otherwise m ∈ I), therefore al+1 ≤ 0. Similarly, if m is of Z-degree −1 then al+1 ≥ 0.
Hence, m is of Z-degree 1, 0 or −1 so the difference between the number of entries xaill in
m of Z-degree 1 and the number of entries of Z-degree −1 is at most 1. It follows that
in this case one can construct a multilinear commutator m1 of canonical form of type 4
on the set {xai11 , . . . , xainn } of variables of m.
Thus, for each m described in i) there exists a multilinear commutator m1 of canonical
form on the same set of variables. The assertion ii) follows.
In what follows u ≡ v stands for u ≡ v (mod I), that is, u − v ∈ I.
Proposition 5. Let m = [xai11 , . . . , xainn ] (n ≥ 3) be a multilinear commutator such that

m∈ / I and xai11 > xaill for all l > 1. Then m ≡ μm , where μ ∈ K and m is a commutator
of canonical form.
Proposition 5 will be proved in the next sections.

Now we are in a position to complete the proof of Theorem 1. Let f ∈ TZ (W1 ) be a
multilinear graded identity of W1 of degree n. Suppose, in order to get a contradiction,
that f ∈/ I. It follows from Proposition 5 that

f ≡ α xai11 , xai22 , . . . , xainn
where α ∈ K and [xai11 , xai22 , . . . , xainn ] is a commutator of canonical form. Since f ∈

TZ (W1 ) and I ⊆ TZ (W1 ), we have α [xai11 , xai22 , . . . , xainn ] ∈ TZ (W1 ). If α = 0 then f ∈ I,
an absurd. Suppose that α = 0; then [xai11 , xai22 , . . . , xainn ] ∈ TZ (W1 ) so, by Lemma 4,
[xai11 , xai22 , . . . , xainn ] ∈ I. It follows that f ∈ I, an absurd again.
Thus, we have f ∈ I. It follows that TZ (W1 ) ⊆ I, therefore TZ (W1 ) = I, as required.
This completes the proof of Theorem 1 provided that Proposition 5 is proved.
3. Auxiliary results
In this section we deduce several graded identities that are consequences of the gen-
erators of I. We need these consequences in order to prove Proposition 5.
Using the Jacobi identity one sees immediately that
a1 a2 a3 a3 a2 a1 a2 a3 a1 a2 a1 a3
x1 , x2 , x3 − x3 , x2 , x1 ∈ I, x2 , x3 , x1 − x2 , x1 , x3 ∈ I (8)
whenever a1 = a3 or a2 = a1 + a3 . Furthermore, by (8), for each i ≥ 3

ai−2 ai−1 ai−2 ai−1
xa1 1 , . . . , xi−2 , xi−1 , xai i − xa1 1 , . . . , xi−2 , xai i , xi−1 ∈I
if ai−1 = ai . It follows that, for each n ≥ 3,

a1 a
x1 , xσ(2) , . . . , xaσ(n) − xa1 1 , xa2 , . . . , xan ∈ I (9)
for every permutation σ of {2, 3, . . . , n} because σ can be written as a product of trans-

positions ((i − 1)i), 3 ≤ i ≤ n.
Recall that u ≡ v stands for u ≡ v (mod I), that is, u − v ∈ I.
Lemma 6. We have

xa1 , xb2 , xb3 , xa4 ≡ xa4 , xb2 , xb3 , xa1 . (10)
More generally, for each n ≥ 4,

xa1 , xb2 , . . . , xbn−1 , xan ≡ xan , xbσ(2) , . . . , xbσ(n−1) , xa1 , (11)
where σ is an arbitrary permutation of {2, . . . , n − 1}.
Proof. By the identity (2), we have [xa1 , xb2 , [xb3 , xa4 ]] ≡ 0 and [xa4 , xb2 , [xa1 , xb3 ]] ≡ 0 so
[xa1 , xb2 , xb3 , xa4 ] ≡ [xa1 , xb2 , xa4 , xb3 ] and [xa4 , xb2 , xa1 , xb3 ] ≡ [xa4 , xb2 , xb3 , xa1 ]. On the other hand,
using (8) we have [xa1 , xb2 , xa4 ] ≡ [xa4 , xb2 , xa1 ] so [xa1 , xb2 , xa4 , xb3 ] ≡ [xa4 , xb2 , xa1 , xb3 ]. Thus,
[xa1 , xb2 , xb3 , xa4 ] ≡ [xa4 , xb2 , xb3 , xa1 ], as required.
Now we prove (11). Since, by (9),

xa1 , xb2 , . . . , xbn−1 , xan ≡ xa1 , xbσ(2) , . . . , xbσ(n−1) , xan
for all σ, it suffices to prove that

a b
x1 , x2 , . . . , xbn−1 , xan ≡ xan , xb2 , . . . , xbn−1 , xa1 .
We prove this by induction on n. For n = 4 the statement has already been proved. If
n > 4 then, by the inductive hypothesis, we have

xa1 , xb2 , xb3 , . . . , xbn−2 , xan , xbn−1 ≡ xan , xb2 , xb3 , . . . , xbn−2 , xa1 , xbn−1 (12)
and
a b
x1 , xn−1 , xb3 , . . . , xbn−2 , xan , xb2 ≡ xan , xbn−1 , xb3 , . . . , xbn−2 , xa1 , xb2 . (13)
Moreover

xa1 , xb2 , xb3 , . . . , xbn−2 , xan , xbn−1

= xa1 , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xan + xa1 , xb2 , xb3 , . . . , xbn−2 , xan , xbn−1 ,
a b b
xn , x2 , x3 , . . . , xbn−2 , xa1 , xbn−1

= xan , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xa1 + xan , xb2 , xb3 , . . . , xbn−2 , xa1 , xbn−1
and, by the inductive hypothesis,

xa1 , xb2 , xb3 , . . . , xbn−2 , xan , xbn−1 ≡ xan , xbn−1 , xb3 , . . . , xbn−2 , xa1 , xb2
so (12) can be rewritten as follows:


xa1 , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xan + xan , xbn−1 , xb3 , . . . , xbn−2 , xa1 , xb2

≡ xan , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xa1 + xan , xb2 , xb3 , . . . , xbn−2 , xa1 , xbn−1 . (14)
Similarly, one can deduce from (13), using (9) that

xa1 , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xan + xan , xb2 , xb3 , . . . , xbn−2 , xa1 , xbn−1

≡ xan , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xa1 + xan , xbn−1 , xb3 , . . . , xbn−2 , xa1 , xb2 . (15)
Since, over a field of characteristic = 2, the congruences A+C ≡ B+D and A+D ≡ B+C
imply A ≡ B, it follows from (14) and (15) that
a b b
x1 , x2 , x3 , . . . , xbn−2 , xbn−1 , xan ≡ xan , xb2 , xb3 , . . . , xbn−2 , xbn−1 , xa1 ,
as required. 2
The next statement is a key lemma in the proof of Proposition 5.
Lemma 7. Let n ≥ 0 and let ai ≥ 0, i = 1, 2, . . . , n. Then

x−1 , z1a1 , . . . , znan , y −1 ≡ y −1 , z1a1 , . . . , znan , x−1 .
Proof. If n = 0 then [x−1 , y −1 ] ≡ 0 ≡ [y −1 , x−1 ] because [x−1 , y −1 ] ∈ I. Thus, it suffices

to prove the lemma for n ≥ 1.
First assume [x−1 , z1a1 , . . . , znan , y −1 ] ∈ I. Then [e−1 , ea1 , . . . , ean , e−1 ] = 0. Since
[e−1 , ea1 , . . . , ean , e−1 ] = (a1 + 1)(a2 − a1 + 1) . . . (an − an−1 − · · · − a1 + 1)

× (−1)(a1 + · · · + an )ea1 +···+an −2 ,
we have either a1 + · · · + ai−1 − 1 = ai for some i ≥ 2 or a1 + · · · + an = 0. In both cases

[y −1 , z1a1 , . . . , znan , x−1 ] ∈ I so [x−1 , z1a1 , . . . , znan , y −1 ] ≡ 0 ≡ [y −1 , z1a1 , . . . , znan , x−1 ].
Now suppose that m = [x−1 , z1a1 , . . . , znan , y −1 ] ∈ / I. We induct on n. If n = 1 then the
lemma follows from (8). If n = 2 we have two cases to analyze.
Case 1: a1 = a2 . Here it suffices to apply (10).
Case 2: a1 = a2 . Note that a2 = a1 − 1 because otherwise [x−1 , z1a1 , z2a2 ] ∈ I and
therefore [x−1 , z1a1 , z2a2 , y −1 ] ∈ I, a contradiction.
By (3), we have

α z1a1 , x−1 , z2a2 ≡ β z1a1 , z2a2 , x−1
where α = −(a2 − a1 )(a1 + a2 + 1), β = −(a1 + 1)(a2 − a1 + 1). Similarly,

α z1a1 , y −1 , z2a2 ≡ β z1a1 , z2a2 , y −1 .
Notice that α = 0 and β = 0. We have

−1 a1 a2 −1
x , z1 , z2 , y = − z1a1 , x−1 , z2a2 , y −1 ≡ −βα−1 z1a1 , z2a2 , x−1 , y −1

≡ −βα−1 z1a1 , z2a2 , y −1 , x−1 ≡ −βα−1 αβ −1 z1a1 , y −1 , z2a2 , x−1

= − z1a1 , y −1 , z2a2 , x−1 = y −1 , z1a1 , z2a2 , x−1 ,
as required.
Now we deal with the general case. Suppose that the lemma holds for the com-
mutators of length less than (n + 2). Take m = [x−1 , z1a1 , . . . , znan , y −1 ]. Let M =
[x−1 , z1a1 , . . . , znan , y −1 ] − [y −1 , z1a1 , . . . , znan , x−1 ]. It is clear that to prove the lemma it
suffices to see that M ≡ 0.
Note that
ai−1 ai ai+1 ai+2
m = x−1 , z1a1 , . . . , zi−1 , zi , zi+1 , zi+2 , . . . , znan , y −1

+ x−1 , z1a1 , . . . , zi−1 , . . . , znan , y −1 .
ai−1 ai+1 ai+2
, zi+1 , ziai , zi+2
By the inductive hypothesis, we have

−1 a1 ai−1 ai ai+1 ai+2
x , z1 , . . . , zi−1 , zi , zi+1 , zi+2 , . . . , znan , y −1
ai−1 ai ai+1 ai+2
≡ y −1 , z1a1 , . . . , zi−1 , zi , zi+1 , zi+2 , . . . , znan , x−1
therefore

M ≡ x−1 , z1a1 , . . . , zi−1 , . . . , znan , y −1
ai−1 ai+1 ai+2

− y −1 , z1a1 , . . . , zi−1 , . . . , znan , x−1 .
ai−1 ai+1 ai+2
Note that the transpositions {(i, i+1) | i = 1, . . . , n−1} generate the symmetric group Sn .
Let
aσ(1) aσ(n) aσ(1) aσ(n)
Mσ = x−1 , zσ(1) , . . . , zσ(n) , y −1 − y −1 , zσ(1) , . . . , zσ(n) , x−1 .
It follows that, for each σ ∈ Sn , we have M ≡ Mσ . Hence, to prove that M ≡ 0 it suffices

to show that Mσ ≡ 0 for some permutation σ ∈ Sn .
Now we are in a position to complete the proof of Lemma 7. If a1 = a2 = · · · = an
then, by (11), we are done. Suppose that at least two of the Z-degrees αi are distinct.
Then we may assume without loss of generality that a1 is the largest among a1 , a2 , . . . , an
and a1 > a2 .
Note that, by (3), we have α[x−1 , z1a1 , z2a2 ] ≡ β[x−1 , z2a2 , z1a1 ], where α = (a2 + 1) ×
(a1 − a2 + 1), β = (a1 + 1)(a2 − a1 + 1). Suppose that β = 0. Then a2 = a1 − 1 so
[x−1 , z1a1 , z2a2 ] ≡ 0 and [y −1 , z1a1 , z2a2 ] ≡ 0, hence M ≡ 0, as required.
Now suppose that β = 0. Then

−1 a2 a2 −1 a2 a1
x , z2 , z1 ≡ ρ x−1 , z1a1 , z2a2 , y , z2 , z1 ≡ ρ y −1 , z1a1 , z2a2 , (16)
where ρ = αβ −1 . If ρ = 1 then α = β so a21 − a22 + a1 − a2 = 0, that is, (a1 − a2 )(a1 +

a2 + 1) = 0. This is impossible due to the choice of a1 and a2 , therefore ρ = 1.
Recall that

M = x−1 , z1a1 , z2a2 , z3a3 , . . . , znan , y −1 − y −1 , z1a1 , z2a2 , z3a3 , . . . , znan , x−1 ,

M(12) = x−1 , z2a2 , z1a1 , z3a3 , . . . , znan , y −1 − y −1 , z2a2 , z1a1 , z3a3 , . . . , znan , x−1 .
Take

M = x−1 , z1a1 , z2a2 , z3a3 , . . . , znan , y −1 − y −1 , z1a1 , z2a2 , z3a3 , . . . , znan , x−1 .
By (16), we have ρM ≡ M(12) ; on the other hand, M = M(12) + M . It follows that

(1 − ρ)M ≡ M . By the inductive hypothesis, M ≡ 0. Since ρ = 1, we have M ≡ 0, as
required. The proof of Lemma 7 is complete. 2
Corollary 8. Let n ≥ 3 and let m = [xa1 1 , . . . , xann ] ∈ L(X), m ∈

/ I be a multilinear
commutator. Suppose that as = ar = −1 where 1 ≤ s < r ≤ n. Then

, x−1 −1
as−1 as+1 ar−1 ar+1
m = xa1 1 , . . . , xs−1 an
s , xs+1 , . . . , xr−1 , xr , xr+1 , . . . , xn

, x−1 −1
as−1 as+1 ar−1 ar+1
≡ xa1 1 , . . . , xs−1 an
r , xs+1 , . . . , xr−1 , xs , xr+1 , . . . , xn .
Proof. Note that it suffices to prove the corollary assuming that ai ≥ 0 for all i such
that s < i < r, that is, to prove that we can permute modulo I any two consecutive
entries of Z-degree −1 in m. Indeed, each permutation of entries of Z-degree −1 in m
is a composition of transpositions permuting two consecutive entries. For instance, the
transposition (13) that permutes the first and the third entries of Z-degree −1 in m is
equal to the product (12)(23)(12) of the transpositions (i(i + 1)) (i = 1, 2) permuting
two consecutive entries. Thus, we may assume that ai ≥ 0 for all i such that s < i < r.
Let m = [xa1 1 , . . . , xakk , x−1 , y1b1 , . . . , ytbt , y −1 ] where k = s − 1 ≥ 0, t = r − s − 1 ≥ 0
and bi ≥ 0 for every i, 1 ≤ i ≤ t. Suppose that m ∈ / I.
If k = 0 then m = [x−1 , y1b1 , . . . , ytbt , y −1 ] ≡ [y −1 , y1b1 , . . . , ytbt , x−1 ] by Lemma 7. (In
this case t > 0 because otherwise m ∈ I.)
If k ≥ 1 and m ∈ / I then a1 + · · · + ak ≥ 0. Denote d = [xa1 1 , . . . , xakk ], then m =
[d, x−1 , y11 , . . . , yt t , y −1 ]. By Lemma 7,
b b

m = − x−1 , d, y1b1 , . . . , ytbt , y −1 ≡ − y −1 , d, y1b1 , . . . , ytbt , x−1

= d, y −1 , y1b1 , . . . , ytbt , x−1
so we can permute (modulo I) these two consecutive entries of Z-degree −1 in m . The

result follows. 2
Lemma 9. Let n ≥ 3 and let

, x−1
an−2
m = xai11 , . . . , xin−2 an
in−1 , xin ,
where m ∈
/ I and an ≥ 0. Suppose that an = a1 + · · · + an−2 . Then

, xainn , x−1
an−2
m ≡ μ xai11 , . . . , xin−2 in−1 ,
where μ ∈ K.
Proof. We have by (3) that

an−2 −1 an−2 an
α xai11 , . . . , xin−2 , xin−1 , xainn ≡ β xai11 , . . . , xin−2 , xin , x−1
in−1 ,
where α = −(an − a1 − · · · − an−2 )(1 + a1 + · · · + an−2 + an ) and also β = −(1 + a1 +

· · · + an−2 )(an − a1 − · · · − an−2 + 1). To prove Lemma 9 it suffices to check that α = 0.
, x−1
an−2
Note that [xai11 , . . . , xin−2 in−1 ] ∈
/ I so a1 + · · · + an−2 − 1 ≥ −1, that is, a1 + · · · +
an−2 ≥ 0. Since an ≥ 0, we have 1+(a1 +· · ·+an−2 )+an ≥ 1 so 1+a1 +· · ·+an−2 +an = 0.
Since an − a1 − · · · − an−2 = 0, we have α = 0 and
β a1
, xainn , x−1
an−2
m≡ x , . . . , xin−2 in−1 ,
α i1
as required. 2
Lemma 10. Let n ≥ 4 and let

, x−1 −1
an−3
m = xai11 , . . . , xin−3 an
in−2 , xin−1 , xin ,
where m ∈
/ I and an ≥ 0. Suppose that an = a1 + · · · + an−3 − 1. Then

, xainn , x−1 −1
an−3
m ≡ μ xai11 , . . . , xin−3 in−1 , xin−2
where μ ∈ K.
Proof. Since an = a1 + · · · + an−3 − 1, we have

a1
, x−1 −1
an−3
in−2 , xin , xin−1 ≡ 0.
an
xi1 , . . . , xin−3
It follows that
a1 −1
, x−1 −1 −1
an−3 an−3
xai11 , . . . , xin−3 in−2 , xin−1 , xin ≡ xi1 , . . . , xin−3 , xin−2 , xin−1 , xin
an an
.
By (3),
−1 an−3 −1
, x−1 , xin−1 , xainn , x−1
an−3
α xai11 , . . . , xin−3 an
in−2 , xin−1 , xin ≡ β xai11 , . . . , xin−3 in−2
where α = −(an − 1 − a1 − · · · − an−3 )(a1 + · · · + an−3 + an ), β = −(1 + a1 + · · · +

an−3 )(an − a1 − · · · − an−1 ). Note that
an − 1 − a1 − · · · − an−3 = (an − a1 − · · · − an−3 + 1) − 2 = −2,

a1 + · · · + an−3 + an = (a1 + · · · + an−3 − 1) + an + 1 = 2an + 1 ≥ 1.
It follows that α = 0 and
−1 β a1 an−3 −1
, x−1 , xin−1 , xainn , x−1
an−3
xai11 , . . . , xin−3 an
in−2 , xin−1 , xin ≡ xi1 , . . . , xin−3 in−2 .
α
Since an = a1 + · · · + an−3 − 1, we have
a1
, x−1 −1
an−3
in−1 , xin , xin−2 ≡ 0
an
xi1 , . . . , xin−3
and thus
a1 an−3 −1 a1
, xin−1 , xainn , x−1 −1 −1
an−3
in−2 ≡ − xi1 , . . . , xin−3 , xin , xin−1 , xin−2 .
an
xi1 , . . . , xin−3
It follows that

, xainn , x−1 −1
an−3
m ≡ μ xai11 , . . . , xin−3 in−1 , xin−2
where μ = − α
β
∈ K, as required. 2
4. Proof of Proposition 5
Let m be a commutator of type 5. Then m = [x0i1 , x−1 0 0

i2 , xi3 , . . . , xin ] where i1 > il

for all l, 3 ≤ l ≤ n. Let m be the commutator of canonical form on the variables
x0i1 , x−1 0 0 0 −1
i2 , xi3 , . . . , xin . Then m = [xi1 , xi2 , xj3 , . . . , xjn ] where {j3 , . . . , jn } = {i3 , . . . , in }
0 0
and j3 < · · · < jn . By (9), we have m ≡ m , as required.

Suppose that m is of type 1, 2, 3 or 4. In order to prove Proposition 5 in each of these
cases we induct on n.
Let n = 3; then m = [xai11 , xai22 , xai33 ]. Since m ∈ / I, we have a2 < a1 .
But either m or [xai11 , xai33 , xai22 ] is of canonical form. Indeed, if [xbj11 , xbj22 , xbj33 ] is the
commutator of canonical form such that {xbj11 , xbj22 , xbj33 } = {xai11 , xai22 , xai33 } then xbj11 > xbjll
(l = 2, 3) so xbj11 = xai11 . Note also that, by (3),

(a3 − a1 )(a2 − a1 − a3 ) xai11 , xai22 , xai33 ≡ (a2 − a1 )(a3 − a1 − a2 ) xai11 , xai33 , xai22 .
Suppose (a3 −a1 )(a2 −a1 −a3 ) = 0; then Proposition 5 holds. In fact in this case either m
2 −a1 )(a3 −a1 −a2 )
is of canonical form or m ≡ μm where μ = (a (a3 −a1 )(a2 −a1 −a3 ) ∈ K and the commutator
m = [xai11 , xai33 , xai22 ] is of canonical form.
Now suppose that (a3 − a1 )(a2 − a1 − a3 ) = 0; then m is of canonical form. Indeed,
if a3 − a1 = 0 then m = [xai11 , xai22 , xai31 ] where a2 < a1 . Since m ∈ / I, we have a2 = 0.
It follows that m is a commutator (of type 1, 2 or 4) of canonical form. On the other
hand, if a2 − a1 − a3 = 0 then a3 < 0 so a3 = −1 and a2 = a1 + a3 = a1 − 1. Hence,
m = [xai11 , xia21 −1 , x−1 i3 ]. Since m ∈/ I, we have 2a1 − 1 ≥ 0. It follows that a1 ≥ 1 and m
is a commutator (of type 1 or 4) of canonical form.
Therefore Proposition 5 holds whenever n = 3.
Suppose that it holds for all commutators of length < n. We will prove that it holds
for the commutator m = [xai11 , . . . , xainn ] of length n and type 1, 2, 3 or 4 as well. Let
an−1
m1 = [xai11 , . . . , xin−1 ]. By the inductive hypothesis, we may assume without loss of
generality that the commutator m1 is of canonical form.
Case 1. Suppose that an = −1. In this case either m is of canonical form or m ≡ m
where m is the commutator of canonical form obtained from m by a permutation of
variables of Z-degree −1. In order to prove this assertion we have to consider separately
commutators of the types 1, 2, 3, 4. We do in detail the case when m is of type 2, and
give (detailed) sketches for the remaining three cases.
Let now m be of type 2. Then

m = xai1 , x0i2 , . . . , x0ir , x−1 a a −1 −1 −1
ir+1 , xir+2 , . . . , xis , xis+1 , . . . , xin−1 , xin
where a = a1 > 1, i2 < . . . < ir , 1 ≤ r ≤ n − 3, ir+2 < . . . < is < i1 , r + 2 ≤ s ≤ n − 1,

ir+1 < is+1 < is+2 < . . . < in−1 . If either s = n − 1 and ir+1 < in or s < n − 1 and
in−1 < in then m is of canonical form. On the other hand, if s = n − 1 and ir+1 > in or
s < n − 1 and in−1 > in then, by Corollary 8, m ≡ m where

m = xai1 , x0i2 , . . . , x0ir , x−1 a a −1 −1
jr+1 , xir+2 , . . . , xis , xjs+1 , . . . , xjn ,
{jr+1 , js+1 , js+2 , . . . , jn } = {ir+1 , is+1 , is+2 , . . . , in−1 , in } and jr+1 < js+1 < js+2 <
· · · < jn . Then m is of canonical form and m ≡ m , as required.
Now suppose that m is of type 1, then

m = xai11 , xai22 , . . . , xairr , x−1 −1 −1
ir+1 , . . . , xin−1 , xin
where 2 ≤ r ≤ n − 1, a2 ≥ 0, xai22 < · · · < xairr < xai11 , ir+1 < . . . < in−1 . If either
r = n − 1 or r < n − 1 and in−1 < in then m is of canonical form. If r < n − 1
and in−1 > in then, by (9), m ≡ m where m = [xai11 , xai22 , . . . , xairr , x−1 −1
jr+1 , . . . , xjn ],

{jr+1 , . . . , jn } = {ir+1 , . . . , in−1 , in } and jr+1 < · · · < jn . Therefore m is of canonical
form and m ≡ m , as required.
Let m be of type 3, then m = [xai11 , x0i2 , . . . , x0ir , x−1 −1 −1
ir+1 , . . . , xin−1 , xin ], where 1 ≤
r ≤ n − 1, i2 < · · · < ir , ir+1 < · · · < in−1 . If either r = n − 1 or r < n − 1 and
in−1 < in then m is of canonical form. If r < n − 1 and in−1 > in then, by (9), m ≡ m
where m = [xai11 , x0i2 , . . . , x0ir , x−1 −1
jr+1 , . . . , xjn ], {jr+1 , . . . , jn } = {ir+1 , . . . , in−1 , in } and
jr+1 < · · · < jn . Then m is of canonical form and m ≡ m , as required.
If, finally, m is of type 4 then m = [m1 , x−1 1 a2 an−1
in ] where m1 = [xi1 , xi2 , . . . , xin−1 ] is of
canonical form. Clearly, (an−3 , an−2 , an−1 ) is equal to one of the following: (1, −1, −1),
(−1, 1, −1), (1, −1, 1), (0, −1, −1), (0, −1, 1), (0, 0, −1) or (0, 0, 0). If (an−3 , an−2 , an−1 ) =
(1, −1, −1) then m1 has Z-degree −1 and m = [m1 , x−1 in ] ∈ I, a contradiction. It follows
that (an−3 , an−2 , an−1 ) = (1, −1, −1). Similarly, (an−3 , an−2 , an−1 ) = (0, −1, −1). If
(an−3 , an−2 , an−1 ) = (0, 0, 0) then al = 0 for all l, 2 ≤ l ≤ n − 1 so m is of canonical
form. If (an−3 , an−2 , an−1 ) is equal to (−1, 1, −1) or (0, 0, −1) and in−1 < in then m
is of canonical form. Similarly, m is of canonical form if (an−3 , an−2 , an−1 ) is equal to
(1, −1, 1) or (0, −1, 1) and in−2 < in . Finally, if either (an−3 , an−2 , an−1 ) is equal to
(−1, 1, −1) or (0, 0, −1) and in−1 > in or if (an−3 , an−2 , an−1 ) is equal to (1, −1, 1) or
(0, −1, 1) and in−2 > in then, by Corollary 8, m ≡ m where m is of canonical form, as
required.
Case 2. Suppose that an ≥ 0 and an−1 = −1.
If an = a1 + · · · + an−2 then, by Lemma 9, m ≡ μ1 m where μ1 ∈ K and

m = xai11 , . . . , xin−2 , xainn , x−1
an−2
in−1 .
Let m1 = [xai11 , . . . , xin−2 , xainn ]; by the inductive hypothesis m1 ≡ μ2 m1 where μ2 ∈ K
a n−2
and m1 is a commutator of canonical form. Hence, m ≡ μ2 [m1 , x−1 in−1 ]. By Case 1,
[m1 , x−1
in−1 ] ≡ m
where m
is of canonical form. It follows that m ≡ μ m where μ =
μ1 μ2 ∈ K and m is of canonical form, as required.
Suppose that an = a1 + · · · + an−2 . Note that if in this case an−2 = −1 then it follows
from Lemma 10 that m ≡ μ1 m where μ1 ∈ K and

m = xai11 , . . . , xin−3 , xainn , x−1 −1
an−3
in−1 , xin−2 .
Now using Case 1 and the induction hypothesis as above we prove that m ≡ μ m where
μ ∈ K and m is a commutator of canonical form, as required. Thus, in the rest of Case 2
if we prove that an−2 = −1 then we are done.
Since a1 ≥ an , the equality an = a1 + · · · + an−2 implies
a2 + · · · + an−2 ≤ 0. (17)
, x−1
a
Note that if m1 = [xai11 , . . . , xin−2
n−2
in−1 ] is a commutator of canonical form of type 1
or 2 then (17) implies an−2 = −1. Indeed, if m1 is of type 1 then al > 0 for some l,
2 ≤ l ≤ n − 2. Together with (17) this implies that al < 0 (that is, al = −1) for some l ,
2 ≤ l ≤ n − 2. Then it follows from the definition of the canonical form of a commutator
of type 1 that an−2 = −1, as claimed. If m1 is of type 2 then the proof is similar. In this
case al > 1 for some l, 2 ≤ l ≤ n − 2 so there are l , l , 2 ≤ l < l ≤ n − 2 such that
al = al = −1. Then by the definition of the canonical form of a commutator of type 2,
an−2 = −1, as claimed.
Now suppose m1 is of canonical form of type 3. Then a2 , . . . , an−1 ≤ 0 and it follows
from (17) that either a2 = · · · = an−2 = 0 or a2 + · · · + an−2 < 0. In the former case
an = a1 and m = [xai11 , x0i2 , . . . , x0in−2 , x−1
in−1 , xin ] where i2 < · · · < in−2 and i1 > in .
a1
Thus m is a commutator (of type 2) of canonical form, as required. In the latter case
it is straightforward to check that for a commutator m1 of canonical form of type 3
the condition a2 + · · · + an−2 < 0 implies an−2 = −1. Then by the observation above,
Proposition 5 holds.
Finally, let m1 be of canonical form of type 4. If a2 + · · · + an−2 < 0 then it follows
from the definition of the canonical form of a commutator of type 4 that an−2 = −1.
Hence, as above, in this case Proposition 5 holds. Suppose that a2 + · · · + an−2 = 0.
Then either a2 = · · · = an−2 = 0 or al < 0 (that is, al = −1) for some l, 2 ≤ l ≤ n − 2.
In the former case an = 1 and

m = x1i1 , x0i2 , . . . , x0in−2 , x−1 1
in−1 , xin
where i2 < · · · < in−2 and i1 > in so m is a commutator (of type 4) of canonical form,
as required. In the latter case it follows from the definition of the canonical form of a
commutator of type 4 that an−2 = 1 hence

m1 = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−3 , xin−2 , xin−1
where 1 ≤ r ≤ n − 4, i2 < · · · < ir , ir+1 < ir+3 < . . . < in−1 , ir+2 < ir+4 < . . . <
in−2 < i1 . We have an = a1 + · · · + an−2 = 1 therefore

m = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−3 , xin−2 , xin−1 , xin
where i1 > in because x1i1 > xainn . If in−2 < in then m is of canonical form and we are
done.
Suppose that in−2 > in . Since, for each l,

x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 1 1 −1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xil−2 , xil , xil−1 ≡ 0,
we have that

x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xil−2 , xil−1 , xil
−1 1
≡ x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xil−2 , xil−1 , xil . (18)
Using (18) and (9) we obtain that


m = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1 1
−1 −1
≡ x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 1
1
−1 −1
≡ x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−1 , xin , xin−3 , xin−2

≡ x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−1 , xin , xin−3 , xin−2 .
By the inductive hypothesis,

1 0
xi1 , xi2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−1 , xin , xin−3 ≡ μ m1
where μ ∈ K and m1 is a commutator of canonical form of type 4. Let m = [m1 , x1in−2 ];
then

m = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1 −1 1
ir+1 , xjr+2 , xir+3 , xjr+4 , . . . , xin−3 , xjn−2 , xin−1 , xjn
where {jr+2 , jr+4 , . . . , jn−2 } = {ir+2 , ir+4 , . . . , in−4 , in } and jn = in−2 , jr+2 < jr+4 <
· · · < jn−2 < jn < i1 . It is straightforward to check that m is of canonical form and
m ≡ μ m , as required.
Case 3. Suppose that an−1 , an ≥ 0.
If m1 is of type 1 or of type 3 then m = [m1 , xainn ] = [xai11 , . . . , xainn ] where a2 ≥ 0,
an−1
xai22 < xai33 < · · · < xin−1 < xai11 and xainn < xai11 . Observe that a1 > 1, and as ≥ 0,
an−1
2 ≤ s ≤ n. If xin−1 < xin then m is of canonical form.
an
an−1
Suppose that xin−1 > xainn ; we have
an−2 an−1 an−2 an−1

α xai11 , . . . , xin−2 , xin−1 , xainn ≡ β xai11 , . . . , xin−2 , xainn , xin−1 (19)
where α = (an − a1 − · · · − an−2 )(an−1 − a1 − · · · − an−2 − an ), and also β = (an−1 −

a1 − · · · − an−2 )(an − a1 − · · · − an−2 − an−1 ).
Note that α = 0. Indeed, an−1 ≤ a1 ≤ a1 + · · · + an−2 . On the other hand, since
m∈ / I, we have an−1 = a1 + · · · + an−2 . It follows that an−1 < a1 + · · · + an−2 . Hence,
an−1 −a1 −· · ·−an−2 −an ≤ an−1 −a1 −· · ·−an−2 < 0 so an−1 −a1 −· · ·−an−2 −an = 0.
Also, an ≤ an−1 < a1 + · · · + an−2 so an − a1 − · · · − an−2 = 0. Thus, α = 0, as claimed.
It follows from (19) that m ≡ μ1 m where m = [xai11 , . . . , xin−2
an−2 an−1
, xainn , xin−1 ] and
∈ K. Let m1 = [xai11 , . . . , xin−2 , xainn ]; by the inductive hypothesis, m1 ≡ μ2 m1
β an−2
μ1 = α
where m1 = [xai11 , xbj22 , . . . , xjn−2
bn−2 bn−1
, xjn−1 ] is a commutator of canonical form of type 1 or of
b b a b
type 3, {xbj22 , . . . , xjn−2
n−2 n−1
, xjn−1 } = {xai22 , . . . , xin−2
n−2
, xainn }, xbj22 < · · · < xjn−1
n−1
< xai11 . Hence,
m = [m1 , xin−1 ] ≡ μ2 m where m = [m1 , xin−1 ]. It is straightforward to check that m
a n−1 a n−1
is a commutator of canonical form. Thus, m ≡ μ m where μ = μ1 μ2 ∈ K and m is of

canonical form, as required.
Let m1 be of type 2. Then, by the definition of the canonical form of a commutator
of type 2, we have an−1 = a1 (recall that m1 is of canonical form and an−1 ≥ 0). Hence,

m = xai1 , x0i2 , . . . , x0ir , x−1 a a an
ir+1 , xir+2 , . . . , xin−1 , xin
where a = a1 > 1, 1 ≤ r ≤ n − 3, i2 < . . . < ir , ir+2 < . . . < in−1 < i1 , 0 ≤ an ≤ a. Note
that n − r ≥ 3.
Suppose that an = a. If in−1 < in then m is a commutator of canonical form of type 2.
On the other hand, if in−1 > in then, by (9), m ≡ m where

m = xai1 , x0i2 , . . . , x0ir , x−1 a a a
ir+1 , xjr+2 , . . . , xjn−1 , xjn ,
{jr+2 , . . . , jn } = {ir+2 , . . . , in } and jr+2 < . . . < jn . Then m is of canonical form and
m ≡ m , as required.
Let an = 0. Then, by (3), m ≡ μ1 m , where

m = xai1 , x0i2 , . . . , x0ir , x−1 a a 0 a
ir+1 , xir+2 , . . . , xin−2 , xin , xin−1
and μ1 = (−(n−r−3)a+1)(−(n−r−1)a+1) (n−r−1)a−1

(−(n−r−2)a+1)(−(n−r−3)a+1) = (n−r−2)a−1 . Here −(n − r − 3)a + 1 = 0 because
a > 1 and ((n − r − 1)a − 1), ((n − r − 2)a − 1) > 0 because n − r ≥ 3. Let

m1 = xai1 , x0i2 , . . . , x0ir , x−1 a a 0
ir+1 , xir+2 , . . . , xin−2 , xin .
By the inductive hypothesis, m1 ≡ μ2 m1 where

m1 = xai1 , x0j2 , . . . , x0jr , x0jr+1 , x−1 a a
ir+1 , xir+2 , . . . , xin−2 ,
{j2 , . . . , jr+1 } = {i2 , . . . , ir , in }, j2 < · · · < jr+1 . Then m = [m1 , xain−1 ] is of canonical
form and m ≡ μm where μ = μ1 μ2 ∈ K, as required.
Suppose that 0 < an < a. Then

m = xai1 , x0i2 , . . . , x0ir , x−1 a a b
ir+1 , xir+2 , . . . , xin−1 , xin
where b = an , 0 < b < a. By (3), we have

a
α xai1 , . . . , x0ir , x−1 b 0 −1
ir+1 , . . . , xin−1 , xin ≡ β xi1 , . . . , xir , xir+1 , . . . , xin , xin−1
a b a
where α = (b − (n − r − 2)a + 1)(−(n − r − 3)a − b + 1).

If α = 0 then

m ≡ μ1 xai1 , x0i2 , . . . , x0ir , x−1 a a b a
ir+1 , xir+2 , . . . , xin−2 , xin , xin−1
where μ1 = β/α ∈ K. By the inductive hypothesis, we have

xai1 , x0i2 , . . . , x0ir , x−1 a a b
ir+1 , xir+2 , . . . , xin−2 , xin

≡ μ2 xai1 , x0i2 , . . . , x0ir , xbin , xair+2 , . . . , xain−2 , x−1
ir+1
so

m ≡ μ1 μ2 xai1 , x0i2 , . . . , x0ir , xbin , xair+2 , . . . , xain−2 , x−1 a
ir+1 , xin−1 .
By (3),
a 0
xi1 , xi2 , . . . , x0ir , xbin , xair+2 , . . . , xain−2 , x−1 a
ir+1 , xin−1

≡ μ3 xai1 , x0i2 , . . . , x0ir , xbin , xair+2 , . . . , xain−2 , xain−1 , x−1
ir+1
where
(−1 − (n − r − 2)a − b)(a − (n − r − 2)a − b + 1)

μ3 =
(a − (n − r − 2)a − b)(−1 − (n − r − 1)a − b)
((n − r − 2)a + b + 1)((n − r − 3)a + b − 1)
= .
((n − r − 3)a + b)((n − r − 1)a + b + 1)
Here ((n − r − 2)a + b + 1), ((n − r − 3)a + b), ((n − r − 1)a + b + 1) > 0 because n − r ≥ 3,
a > b > 0 and ((n − r − 3)a + b − 1) = 0 because α = 0. It follows that m ≡ μ m where
μ = μ1 μ2 μ3 and

m = xai1 , x0i2 , . . . , x0ir , xbin , xair+2 , . . . , xain−2 , xain−1 , x−1
ir+1
is a commutator of canonical form, as required.

If α = 0 then either b = (n − r − 2)a − 1 or (n − r − 3)a + b = 1. Since r ≤ n − 3, a > 1
and 0 < b < a, we have b = a − 1, r = n − 3 in the former case and b = 1, r = n − 3 in
the latter case. If b = a − 1 and r = n − 3 then using (2) and (3) we obtain

m = xai1 , x0i2 , . . . , x0in−3 , x−1 a a−1
in−2 , xin−1 , xin
a
≡ xai1 , x0i2 , . . . , x0in−3 , x−1 a−1
−1
≡ μ xai1 , x0i2 , . . . , x0in−3 , xain−1 , xa−1
in , xin−2
−1

≡ −μ xai1 , x0i2 , . . . , x0in−3 , xa−1 a
in , xin−1 , xin−2
−(a+1)a
where μ = (−1−a)(2a−1−a+1) a+1
(a−1)(−1−3a+1) = −(a−1)3a = 3(a−1) . If b = 1 and r = n − 3 then since
(a−1−a)(1−2a+1) (−2)(1−a)
k= (1−a)(a−1−1−a) = (1−a)(−2) = 1 we have

m = xai1 , x0i2 , . . . , x0in−3 , x−1 a
1
1
≡ xai1 , x0i2 , . . . , x0in−3 , x−1 a

≡ xai1 , x0i2 , . . . , x0in−3 , x1in , x−1 a
in−2 , xin−1

≡ − xai1 , x0i2 , . . . , x0in−3 , x1in , xain−1 , x−1
in−2 .
−1
Hence we obtain the commutators [xai1 , x0i2 , . . . , x0in−3 , xa−1 a a 0
in , xin−1 , xin−2 ] and [xi1 , xi2 ,
−1
. . . , x0in−3 , x1in , xain−1 , xin−2 ] which are of canonical form (of type 1). Therefore in both
cases above m ≡ μ m where μ ∈ K and m is a commutator of canonical form, as
required.
Let m1 be of type 4. Then either an−1 = 0 or an−1 = 1. In the former case, by the
definition of the canonical form of a commutator of type 4, we have a2 = · · · = an−2 = 0
thus m1 = [x1i1 , x0i2 , . . . , x0in−1 ]; in the latter case

m1 = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−2 , xin−1
where 1 ≤ r ≤ n − 3, i2 < · · · < ir , ir+1 < ir+3 < · · · < in−2 , ir+2 < ir+4 <
· · · < in−1 < i1 . Note that 0 ≤ an ≤ a1 , and an ∈ {0, 1}; since [m1 , x1in ] ∈ I and
m = [m1 , xainn ] ∈/ I, we have an = 0.
If an−1 = 0 then m = [m1 , x0in ] = [x1i1 , x0i2 , . . . , x0in−1 , x0in ] so, by (9), m ≡ m where

m = [x1i1 , x0j2 , . . . , x0jn−1 , x0jn ], {j2 , . . . , jn } = {i2 , . . . , in } and j2 < j3 < . . . < jn . Hence
m is a commutator of canonical form and m ≡ m , as required.
Suppose that an−1 = 1. By (9) and (18), we have

m = x1i1 , x0i2 , . . . , x0ir , x−1 1 −1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−2 , xin−1 , xin
0
−1 −1 0
≡ x1i1 , x0i2 , . . . , x0ir , x−1 1 1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−2 , xin−1 , xin
−1 −1
≡ x1i1 , x0i2 , . . . , x0ir , x0in , x−1 1 1
1

≡ x1i1 , x0i2 , . . . , x0ir , x0in , x−1 1 −1 1 −1 1
ir+1 , xir+2 , xir+3 , xir+4 , . . . , xin−2 , xin−1 .
By (9), [x1i1 , x0i2 , . . . , x0ir , x0in ] ≡ [x1i1 , x0j2 , . . . , x0jr , x0jr+1 ] where {j2 , . . . , jr+1 } = {i2 , . . . ,
ir , in } and j2 < . . . < jr+1 . It follows that m ≡ m where

m = x1i1 , x0j2 , . . . , x0jr , x0jr+1 , x−1 1 −1 1 −1 1
is of canonical form. This completes the proof of Proposition 5.
5. Independence of graded identities
Denote by fi = [xi1 , xi2 ] ∈ L(X) the graded polynomials from (2). Note that the
identity f−1 = [x−1 −1
1 , x2 ] is a consequence of the identity x
−2
.
Lemma 11. For each i ≥ 0, the graded identity fi is not a consequence of all identities
(3), (4) and all identities fj where j ≥ 0, j = i.
Proof. Let i ≥ 0 and let H = U T (3, K) be the Lie algebra of strictly upper triangular
3 × 3 matrices over K. Define the vector subspaces Hk (k ∈ Z) in H as follows: if i = 0
then H0 = H and Hk = 0 for k = 0; if i > 0 then Hi is the linear span of E12 and E23 ,
H2i is spanned by E13 and Hk = 0 for all k = i, 2i. Here Eij is the matrix that has 1

at the position (i, j) and 0 elsewhere. It is clear that H = k∈Z Hk is a Z-graded Lie
algebra. Since [E12 , E23 ] = E13 = 0, the graded identity [xi1 , xi2 ] is not satisfied in H. On
the other hand, one can easily see that H satisfies all graded identities (3), (4) as well
as all identities fj where j ≥ 0, j = i. The result follows. 2
Corollary 12. The set of polynomials {fi | i ≥ 0} is an independent set of graded identities
in L(X).
Proof. It is a direct consequence of Lemma 11. 2
Lemma 13. For each d ∈ Z, the graded identity xd is not a consequence of all graded
identities (2), (3) and all identities xc where c = d.
Proof. Let d ∈ Z and let H be the 1-dimensional Lie algebra over K. The algebra

H = i∈Z Hi is Z-graded with i-th homogeneous component Hi equal to H if i = d
and 0 otherwise. It is clear that H satisfies the graded identities (2) and (3) as well as
all graded identities xc where c = d but does not satisfy the identity xd . The result
follows. 2
Corollary 14. The set of polynomials {xd | d ≤ −2} is an independent set of graded
identities in L(X).
Proof. It is a direct consequence of Lemma 13. 2
Now let us denote fabc = α[xa1 , xb2 , xc3 ] − β[xa1 , xc3 , xb2 ] the graded identities defined
in (3). Recall that α = (c − a)(b − c − a) and β = (b − a)(c − b − a). First we observe
that whenever c = a we have α = 0 but the second commutator of faba vanishes due to
the graded identity (2), and analogously for b = a. So we suppose that b = a and c = a.
Furthermore, if b = c + a then α = 0 but the second commutator in fabc vanishes also
due to (2). So we may suppose b = a + c and analogously c = a + b. Observe further that
if c = b then one has

fabb xa1 , xb2 , xb3 = −(b − a)a xa1 , xb2 , xb3 − xa1 , xb3 , xb2 = −(b − a)a xa1 , xb2 , xb3
that lies in the graded T-ideal generated by the identities (2). So we may suppose in
addition b = c. Similarly if a = b + c one obtains that
b+c b c b+c c b b c
1 , x2 , x3 = 2bc x1 , x2 , x3 − x1 , x3 , x2
f(b+c)bc xb+c = 2bc xb+c
b c
1 , x2 , x3
is in the graded T-ideal generated by (2). Therefore we shall suppose a = b + c.

The above considerations and Theorem 1 imply the following lemma.
Lemma 15. The polynomials (2), (4) and the polynomials (3) with a, b, c distinct and
a = b + c, b = a + c, c = a + b, form a basis of the Z-graded identities of W1 .
We shall impose further restrictions on the graded identities of the type (3).
Lemma 16. For all a, b, c ≥ −1,

facb xa1 , xc3 , xb2 = fbac xb2 , xa1 , xc3 = −fabc xa1 , xb2 , xc3 .
Proof. Note that facb (xa1 , xc3 , xb2 ) = γ[xa1 , xc3 , xb2 ] − δ[xa1 , xb2 , xc3 ] where γ = (b − a)(c −
b − a) = β, δ = (c − a)(b − c − a) = α. Hence, facb (xa1 , xc3 , xb2 ) = −fabc (xa1 , xb2 , xc3 ), as
required.
Further, by the Jacobi identity,

fbac xb2 , xa1 , xc3 = μ xb2 , xa1 , xc3 − ν xb2 , xc3 , xa1

= −μ xa1 , xb2 , xc3 + ν xc3 , xa1 , xb2 + ν xa1 , xb2 , xc3

= (ν − μ) xa1 , xb2 , xc3 − ν xa1 , xc3 , xb2
where μ = (c −b)(a −b −c), ν = (a −b)(c −b −a). Clearly, ν = −β; it is straightforward to

check that ν − μ = −α. It follows that fbac (xb2 , xa1 , xc3 ) = −α[xa1 , xb2 , xc3 ] + β[xa1 , xc3 , xb2 ] =
−fabc (xa1 , xb2 , xc3 ), as required. 2
Corollary 17. For all a, b, c ≥ −1,

fbca xb2 , xc3 , xa1 = fcab xc3 , xa1 , xb2 = fabc xa1 , xb2 , xc3 ,

fcba xc3 , xb2 , xa1 = −fabc xa1 , xb2 , xc3 .
Thus we can rewrite Lemma 15 as follows:
Lemma 18. The polynomials (2), (4) and the polynomials (3) with a > b > c ≥ −1 and
a = b + c, b = a + c, form a basis of the Z-graded identities of W1 .
Proposition 19. Let a, b, c ∈ Z, a > b > c ≥ −1, a = b + c, b = a + c. Then the graded

identity fabc is not a consequence of all identities (2), (4) and all other identities fa b c
where a > b > c ≥ −1, (a , b , c ) = (a, b, c).
Proof. We shall present a Z-graded Lie algebra L(a,b,c) such that
i) L(a,b,c) satisfies all identities (2), (4) and all identities fa b c where a > b > c ≥ −1,
(a , b , c ) = (a, b, c);
ii) L(a,b,c) does not satisfy the identity fabc .
Clearly the existence of such an algebra L(a,b,c) proves the proposition.

Let L(a,b,c) = U T (4, K) be the Lie algebra of strictly upper triangular 4 × 4 matrices
over K. Let ha = E12 , hb = E23 , hc = E34 . Then L(a,b,c) is a nilpotent Lie algebra of
class 3 and of dimension 6, with a K-basis consisting of ha , hb , hc , ha+b = [ha , hb ] = E13 ,
hb+c = [hb , hc ] = E24 , ha+b+c = [ha , hb , hc ] = E14 .
For each i ∈ Z, let Hi be the vector space generated by the elements hi , hi , hi if such
elements exist and 0 otherwise. Note that Hi = 0 if i ∈ / {a, b, c, a + b, b + c, a + b + c}; in

particular, Hi = 0 if i < −1. It is clear that the algebra L(a,b,c) = i∈Z Hi is Z-graded
with i-th homogeneous component equal to Hi . Since Hi = 0 if i < −1, the algebra
L(a,b,c) satisfies the graded identities xd = 0 for d ≤ −2, that is, the identities (4).
Now we check that the graded Lie algebra L(a,b,c) satisfies the identities (2), that is,
the identities [xi1 , xi2 ] for i ≥ −1. If i ∈ / {a, b, c, a + b, b + c, a + b + c} then Hi = 0 so
L (a,b,c)
satisfies the identity [x1 , x2 ]. If i ∈ {a, b, c, a + b, b + c, a + b + c} and dim Hi = 1
i i
then L(a,b,c) also satisfies the identity [xi1 , xi2 ]. However, neither of the above is the case
for any i if b = 0, c = 0 or b + c = 0 so we have to consider the following particular cases.
Suppose that i = a. Since b = a, c = a and b + c = a, we have hb , hc , hb+c ∈ / Hi .
It follows that Hi is contained in the K-linear span of ha = E12 , ha+b = E13 and
ha+b+c = E14 . Since the latter elements commute, the identity [xi1 , xi2 ] is satisfied in
L(a,b,c) in this case.
Suppose that i = b. Then, similarly, we have ha , hc , ha+b ∈ / Hi so Hi is contained in
the K-linear span of the (commuting) elements hb , hb+c , ha+b+c . Analogously, if i = c
the Hi is contained in the span of hc and hb+c that commute. Hence, the identity [xi1 , xi2 ]
is also satisfied in these cases.
Finally, suppose that i ∈ {a + b, b + c, a + b + c} and i = a, b, c. Then Hi is contained
in the abelian Lie subalgebra spanned by ha+b , hb+c and ha+b+c so the graded identity
[xi1 , xi2 ] is satisfied.
Thus the algebra L(a,b,c) satisfies the graded identities (2).
Let a , b , c ∈ Z, a > b > c ≥ −1 and (a , b , c ) = (a, b, c). We claim that L(a,b,c)

satisfies the graded identities [xa1 , xb2 , xc3 ] and [xa1 , xc3 , xb2 ].
Indeed, if a ∈ / {a, b, c, a + b, b + c, a + b + c} then Ha = 0 so [Ha , Hb , Hc ] =

[Ha , Hc , Hb ] = 0, that is, the graded identities [xa1 , xb2 , xc3 ] and [xa1 , xc3 , xb2 ] are satis-

fied in L(a,b,c) . Similarly, if b or c does not belong to the set {a, b, c, a + b, b + c, a + b + c}

then [xa1 , xb2 , xc3 ] and [xa1 , xc3 , xb2 ] are graded identities in L(a,b,c) .
Now suppose that a , b , c ∈ {a, b, c, a + b, b + c, a + b + c}. Since (a , b , c ) = (a, b, c),
at least one of the elements a , b , c does not belong to the set {a, b, c}. Suppose that
a ∈ / {a, b, c}. Then a ∈ {a + b, b + c, a + b + c} and Ha is contained in the K-linear
span V of ha+b , hb+c and ha+b+c , that is, in the span of the matrices E13 , E24 , E14 .
It follows that [Ha , Hb , Hc ] ⊆ [V, Hb , Hc ] = 0 and similarly [Ha , Hc , Hb ] = 0.
One can check in a similar way that if b or c does not belong to {a, b, c} then

[Ha , Hb , Hc ] = [Ha , Hc , Hb ] = 0, that is, [xa1 , xb2 , xc3 ] and [xa1 , xc3 , xb2 ] are graded
identities in L(a,b,c) .
Thus, if a , b , c ∈ Z, a > b > c ≥ −1 and (a , b , c ) = (a, b, c) then L(a,b,c) sat-

isfies the graded identities [xa1 , xb2 , xc3 ] and [xa1 , xc3 , xb2 ], as claimed. Hence, the Lie

algebra L(a,b,c) satisfies the graded identity fa b c = α[xa1 , xb2 , xc3 ] − β[xa1 , xc3 , xb2 ], as
required.
Now let us compute
fabc (ha , hb , hc ) = (c − a)(b − c − a)[ha , hb , hc ] − (b − a)(c − b − a)[ha , hc , hb ]
= (c − a)(b − c − a)ha+b+c .
Note that (c − a)(b − c − a) = 0 since c < a and b = a + c so fabc is not a graded identity
in L(a,b,c) .
Therefore L(a,b,c) satisfies all graded identities (2), (4) as well as all graded identities
fa b c where a > b > c ≥ −1, (a , b , c ) = (a, b, c) but does not satisfy the identity fabc .
The proof of the proposition is completed. 2
Corollary 20. The set of polynomials {fabc | a > b > c ≥ −1, a = b + c, b = a + c} is an

independent set of graded identities in L(X).
Proof. It is a direct consequence of Proposition 19. 2
Putting together all the above we have the following theorem.
Theorem 21. The graded identities [xi1 , xi2 ] (i ≥ 0), xd (d ≤ −2) and fabc with a > b >
c ≥ −1, a = b + c, b = a + c form a minimal basis for the Z-graded identities of W1 over
a field K of characteristic 0.
In other words, the set of polynomials [xi1 , xi2 ] (i ≥ 0), xd (d ≤ −2) and fabc
(a > b > c ≥ −1, a = b + c, b = a + c) is a minimal generating set for the Z-graded
T-ideal TZ (W1 ).
Corollary 22. The Z-graded identities for the Lie algebra W1 over a field of characteris-
tic 0 do not admit any finite basis.
Remark 23. 1. As we mentioned in the Introduction, the ordinary identities of W1 are

unknown. It is well known and easy to check that W1 satisfies the standard Lie identity s5 .
It is not known even whether the ordinary identities of W1 admit a finite basis.
2. It might be of interest to compute the graded identities of the Lie algebra Wn
for n ≥ 2. One can consider this algebra graded naturally by Zn where we believe the
argument would follow somewhat our argument. It should be much more difficult to
compute the Z-graded identities of Wn when n ≥ 2.
Acknowledgments
The work of José A. Freitas was partially supported by CAPES 9288/2012-06 and
CNPq 480139/2012-1. The work of Plamen Koshlukov was partially supported by CNPq
Grants 304003/2011-5, 480139/2012-1, and by FAPESP Grant 2014/09310-5 and the
work of Alexei Krasilnikov was partially supported by CNPq grants 307328/2012-0 and
480139/2012-1, FAPDF, CAPES, CNPq/FAPDF/PRONEX 2009/00091-0 (193.000.580/
2009) and DPP/UnB. Thanks are due to the referee whose remarks and suggestions
improved considerably the paper.
References
[1] Yu. Bahturin, Identical Relations in Lie Algebras, first English ed. VNU Science Press BV, Utrecht,
1987.
[2] Yu. Bahturin, M. Kochetov, Classification of group gradings on simple Lie algebras of types A, B,
C and D, J. Algebra 324 (2010) 2971–2989.
[3] Yu. Bahturin, M. Zaicev, Graded algebras and graded identities, in: Polynomial Identities and
Combinatorial Methods, in: Lect. Notes Pure Appl. Math., vol. 235, M. Dekker, 2003, pp. 101–139.
[4] A. Elduque, M. Kochetov, Gradings on Simple Lie Algebras, Math. Surveys Monogr., vol. 189,
Amer. Math. Soc., Providence, RI, 2013; Atlantic Association for Research in the Math. Sciences
(AARMS), Halifax, NS.
[5] A. Giambruno, M.S. Souza, Graded polynomial identities and Specht property of the Lie algebra
sl 2 , J. Algebra 389 (2013) 6–22.
[6] A. Giambruno, M.S. Souza, Minimal varieties of graded Lie algebras of exponential growth and the
special Lie algebra sl 2 , J. Pure Appl. Algebra 218 (2014) 1517–1527.
[7] A. Giambruno, M. Zaicev, Polynomial Identities and Asymptotic Methods, Math. Surveys Monogr.,
vol. 122, Amer. Math. Soc., Providence, RI, 2005.
[8] A. Kanel-Belov, L. Rowen, Computational Aspects of Polynomial Identities, Res. Notes Math.,
vol. 9, A. K. Peters, Ltd., Wellesley, MA, 2005.
[9] A. Kemer, Ideals of Identities of Associative Algebras, Transl. Math. Monogr., vol. 87, Amer. Math.
Soc., Providence, RI, 1991.
[10] P. Koshlukov, Graded polynomial identities for the Lie algebra sl 2 (K), Internat. J. Algebra Comput.
18 (2008) 825–836.
[11] P. Koshlukov, A. Krasilnikov, D.D.P. Silva, Graded identities for Lie algebras, in: Contemp. Math.,
vol. 499, Amer. Math. Soc., 2009, pp. 181–188.
[12] P. Koshlukov, A. Krasilnikov, Just nonfinitely based varieties of 2-graded Lie algebras, submitted
for publication.
[13] S.P. Mishchenko, Solvable subvarieties of a variety generated by a Witt algebra, Math. USSR Sb.
64 (1989) 415–426.
[14] Yu. Razmyslov, Identities of Algebras and Their Representations, Transl. Math. Monogr., vol. 138,
Amer. Math. Soc., Providence, RI, 1994.
[15] D.V. Repin, Graded identities of a simple three-dimensional Lie algebra, Vestn. Samar. Gos. Univ.
Estestvennonauchn. Ser. (2) (2004) 5–16, Special Issue (in Russian).

Z-Graded Identities of The Lie Algebra

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Z-Graded Identities of The Lie Algebra

Uploaded by

Copyright:

Available Formats

Journal of Algebra 427 (2015) 226–251

Contents lists available at ScienceDirect

Z-graded identities of the Lie algebra W1

Article history: Let K be a ﬁeld of characteristic 0 and let W1 be the Lie

[ei , ej ] = (j − i)ei+j . (1)

Theorem 1. Let K be a ﬁeld of characteristic 0. The graded identities

is satisﬁed in W1 . On the other hand, it is a long-standing open problem to determine

[ea1 , . . . , ean ] = αn ea1 +···+an (6)

where αn = (a2 − a1 )(a3 − a1 − a2 ) . . . (an − a1 − · · · − an−1 ). Deﬁne ed = 0 if d ≤ −2.

Lemma 2. The equality (6) is valid for all a1 , . . . , an ≥ −1 (n ≥ 2).

[ea1 , . . . , eak ] = αk ea1 +···+ak .

Suppose that a1 + · · · + an−1 ≥ −1. By (1), we have

[ea1 , . . . , ean ] = αn−1 [ea1 +···+an−1 , ean ]

so [ea1 , . . . , ean ] = 0. On the other hand, ak − a1 − · · · − ak−1 = 0 so αn = 0. Thus,

[ea1 , . . . , ean ] = 0 = αn ea1 +···+an ,

α[ea , eb , ec ] − β[ea , ec , eb ] = 0, (7)

α[ea , eb , ec ] − β[ea , ec , eb ] = αβea+b+c − βαea+b+c = 0.

Lemma 4. Let m = [xai11 , . . . , xainn ] ∈ L(X) be a multilinear commutator. Suppose that

Proof. If al ≤ −2 for some l, 1 ≤ l ≤ n, then m belongs to the graded T-ideal in

be written as a linear combination of left-normed commutators starting with any ﬁxed

where a = a1 > 1, r ≥ 1, r + 2 ≤ s ≤ n, i2 < . . . < ir , ir+1 < is+1 < . . . < in ,

where 1 ≤ r ≤ n and i2 < · · · < ir , ir+1 < · · · < in .

where i3 < i4 < · · · < in < i1 .

A commutator m of type 4 and of length n ≥ 4 is of canonical form if

Here 1 ≤ r ≤ n, i2 < · · · < ir ,

and the following hold:

i) each multilinear commutator m = [xai11 , . . . , xainn ] (n ≥ 3) such that m ∈

Indeed, it is straightforward to check that for each m described in i) we have a1 ≥ 0

Proposition 5. Let m = [xai11 , . . . , xainn ] (n ≥ 3) be a multilinear commutator such that

Proposition 5 will be proved in the next sections.

where α ∈ K and [xai11 , xai22 , . . . , xainn ] is a commutator of canonical form. Since f ∈

whenever a1 = a3 or a2 = a1 + a3 . Furthermore, by (8), for each i ≥ 3

if ai−1 = ai . It follows that, for each n ≥ 3,

for every permutation σ of {2, 3, . . . , n} because σ can be written as a product of trans-

More generally, for each n ≥ 4,

where σ is an arbitrary permutation of {2, . . . , n − 1}.

for all σ, it suﬃces to prove that

and, by the inductive hypothesis,

so (12) can be rewritten as follows:

Similarly, one can deduce from (13), using (9) that

The next statement is a key lemma in the proof of Proposition 5.

Lemma 7. Let n ≥ 0 and let ai ≥ 0, i = 1, 2, . . . , n. Then

Proof. If n = 0 then [x−1 , y −1 ] ≡ 0 ≡ [y −1 , x−1 ] because [x−1 , y −1 ] ∈ I. Thus, it suﬃces

[e−1 , ea1 , . . . , ean , e−1 ] = (a1 + 1)(a2 − a1 + 1) . . . (an − an−1 − · · · − a1 + 1)

we have either a1 + · · · + ai−1 − 1 = ai for some i ≥ 2 or a1 + · · · + an = 0. In both cases

where α = −(a2 − a1 )(a1 + a2 + 1), β = −(a1 + 1)(a2 − a1 + 1). Similarly,

Notice that α = 0 and β = 0. We have

By the inductive hypothesis, we have

It follows that, for each σ ∈ Sn , we have M ≡ Mσ . Hence, to prove that M ≡ 0 it suﬃces

Now suppose that β = 0. Then

where ρ = αβ −1 . If ρ = 1 then α = β so a21 − a22 + a1 − a2 = 0, that is, (a1 − a2 )(a1 +

By (16), we have ρM ≡ M(12) ; on the other hand, M = M(12) + M . It follows that

Corollary 8. Let n ≥ 3 and let m = [xa1 1 , . . . , xann ] ∈ L(X), m ∈

so we can permute (modulo I) these two consecutive entries of Z-degree −1 in m . The

Lemma 9. Let n ≥ 3 and let

Proof. We have by (3) that

where α = −(an − a1 − · · · − an−2 )(1 + a1 + · · · + an−2 + an ) and also β = −(1 + a1 +

Lemma 10. Let n ≥ 4 and let

Proof. Since an = a1 + · · · + an−3 − 1, we have

where α = −(an − 1 − a1 − · · · − an−3 )(a1 + · · · + an−3 + an ), β = −(1 + a1 + · · · +

an − 1 − a1 − · · · − an−3 = (an − a1 − · · · − an−3 + 1) − 2 = −2,

It follows that α = 0 and

Let m be a commutator of type 5. Then m = [x0i1 , x−1 0 0

an−2 an−1 an−2 an−1