Grobner-Shirshov Bases Theory andShirshov Algorithm
Leonid BokutSobolev Institute of Mathematics, Russia and South China Normal University, China
Yuqun ChenSouth China Normal University, China
Ministry of Education and Science of the Russian Federations
National Research University-Novosibirsk State University (NRU-NSU)
Department of Mechanics and Mathematics
Master education program for teaching foreign student in English
“Modern trends in discrete mathematics and combinatorial optimization”
The development is done in the framework of the Program of NSU as a national Research
University
Grobner-Shirshov bases theory and Shirshov algorithm
Authors: Leonid Bokut and Yuqun Chen
Novosibirsk, 2013
c© 2013 Leonid Bokut and Yuqun Chen
i
Forewords
This small book presents the principals of the theory of Grobner-Shirshov bases for
associative and Lie algebras. The Buchberger and Shirshov algorithms for commutative,
non-commutative and Lie polynomials are given in the introductory Chapter 1 without
proofs. All the proofs presented in the following Chapters 2-6. Chapters 7 and 8 are de-
voted to Grobner-Shirshov bases for pre-Lie algebras and associative dialgebras. Finally,
the last Chapter 9 contains the history of Grobner-Shirshov bases in the works of A.I. Shir-
shov, MSU and Sobolev Institute of mathematics, NSU, 1953-1962, and the influence by
Alexander Gennadievich Kurosh and Anatolii Ivanovich Malcev to these activities. Also
we include some survey of Grobner-Shirshov bases theory for algebras, groups, semigroups
and categories.
The book is written based on the authors’ many years of teaching this subject at
the Novosibirsk State University and the South China Normal University (Guangzhou).
We are grateful to all participants of the algebra seminars at NSU, Sobolev Institute of
Mathematics and SCNU for many helps and support, especially to G.P. Kukin, Yu.N.
Maltsev, I.V. Lvov, V.K. Kharchenko, E.I. Zelmanov, P.S. Kolesnikov, E.S. Chibrikov,
Yongshan Chen, Chanyan Zhong, Yu Li and Qiuhui Mo.
i
Contents
Forewords i
Chapter 1 Introduction 2
§1.1 The Euclidean g.c.d. algorithm . . . . . . . . . . . . . . . . . . . . . . . . 2
§1.2 The Gauss elimination algorithm . . . . . . . . . . . . . . . . . . . . . . . 4
§1.3 Systems of polynomial equations . . . . . . . . . . . . . . . . . . . . . . . 6
§1.4 Examples of systems of noncommutative (NC) polynomial equations . . . 9
§1.5 Systems of Lie polynomials equations . . . . . . . . . . . . . . . . . . . . 13
§1.5.1 Words and Shirshov’s bracketing (arrangement of parentheses) . . 13
§1.5.2 Lie monomials and Lie polynomials . . . . . . . . . . . . . . . . . . 15
§1.5.3 Serre system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
§1.5.4 Lyndon-Shirshov words . . . . . . . . . . . . . . . . . . . . . . . . 17
§1.5.5 Lyndon-Shirshov Lie monomials . . . . . . . . . . . . . . . . . . . 17
§1.5.6 Lie compositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Chapter 2 Free Commutative, Associative and Lie Algebras 21
§2.1 Commutative polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 21
§2.2 Noncommutative polynomials . . . . . . . . . . . . . . . . . . . . . . . . 23
§2.3 Lie polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
§2.3.1 Lazard-Shirshov elimination process . . . . . . . . . . . . . . . . . 26
§2.3.2 Constructing generators for an arbitrary subalgebra in a free Lie
algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
§2.3.3 A proof of Shirshov-Witt Theorem . . . . . . . . . . . . . . . . . . 28
Chapter 3 Grobner Bases for Commutative Algebras 32
§3.1 Buchberger theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
§3.2 Buchberger algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Chapter 4 Grobner-Shirshov Bases for Associative Algebras 37
§4.1 Composition-Diamond lemma for associative algebras . . . . . . . . . . . 37
§4.1.1 PBW-theorem for Lie algebras . . . . . . . . . . . . . . . . . . . . 41
§4.1.2 Normal forms for groups and semigroups . . . . . . . . . . . . . . . 42
§4.2 Composition-Diamond lemma for modules . . . . . . . . . . . . . . . . . . 47
ii
§4.3 Grobner-Shirshov bases for tensor product of free algebras . . . . . . . . . 49
§4.3.1 Composition-Diamond lemma for tensor product . . . . . . . . . . 49
§4.3.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Chapter 5 Grobner-Shirshov Bases for Lie Algebras over a Field 67
§5.1 Lyndon-Shirshov words . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
§5.2 Free Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
§5.3 Composition-Diamond lemma for Lie algebras . . . . . . . . . . . . . . . . 79
§5.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
§5.4.1 Kukin’s construction of a Lie algebra with unsolvable word problem 82
§5.4.2 Grobner-Shirshov basis for the Drinfeld-Kohno Lie algebra Ln . . . 84
Chapter 6 Grobner-Shirshov Bases for Lie Algebras over a Commutative
Algebra 88
§6.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
§6.2 Composition-Diamond lemma for Liek[Y ](X) . . . . . . . . . . . . . . . . 90
§6.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Chapter 7 Grobner-Shirshov Bases for Pre-Lie Algebras 109
§7.1 Composition-Diamond lemma for pre-Lie algebras . . . . . . . . . . . . . 109
§7.2 PBW theorem for pre-Lie algebras . . . . . . . . . . . . . . . . . . . . . . 115
Chapter 8 Grobner-Shirshov Bases for Dialgebras 117
§8.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
§8.2 Composition-Diamond lemma for dialgebras . . . . . . . . . . . . . . . . 118
§8.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter 9 A History and a Survey of Grobner-Shirshov Bases Theory 142
Reference 153
Notation 169
Chapter 1 Introduction
The classical Grobner and Grobner–Shirshov bases theory deals with three types of
polynomials: commutative associative, noncommutative associative, and Lie, providing
algorithms for passing from a set of polynomials of a fixed type to another set (of the same
type) with better properties. Buchberger’s algorithm works for commutative polynomials,
while Shirshov’s algorithms work for noncommutative and Lie polynomials.
Two classical cases of Buchberger’s algorithm have been known for ages: the Eu-
clidean algorithm for polynomials in one variable, and the Gauss Elimination algorithm
for linear (degree 1) polynomials in several variables.
A big advantage of Buchberger’s algorithm is that it is finite. It means that for
any given finite set of commutative polynomials in several variables, the computation
terminates after finitely many steps.
Buchberger’s algorithm is a specialization of Shirshov’s algorithms, which, in con-
trast, are infinite in general; nevertheless, they have been used with great success in both
theoretical and applied research.
§1.1 The Euclidean g.c.d. algorithm
Let k be a field, and let f(x) and g(x) be two polynomials in one variable x over
k. Our goal is to present an algorithm that finds gcd(f(x), g(x)), the greatest common
divisor of f(x) and g(x). It is the well-known Euclidean g.c.d. algorithm based on the
Euclidean division algorithm.
Example 1.1 Take
f(x) = x4 + x2 − x− 1, g(x) = x3 − 2x2 + 1.
Do the following reduction of f(x) modulo g(x):
f(x) 7→ f(x)− xg(x) = 2x3 + x2 − 2x− 1.
Here, we use the leading monomial x3 of g(x) to reduce the leading monomial x4 of
f(x), which is the same as computing the remainder of the division of f(x) by g(x) (the
elimination of the leading monomial of g in f). The other elementary step of the algorithm
is the division of a polynomial by using its leading coefficient. In our case,
2x3 + x2 − 2x− 1 7→ f1(x) = x3 +1
2x2 − x− 1
2.
2
The two steps make up a transformation that preserves the g.c.d., which we indicate by
writing f(x), g(x) ' f1(x), g(x).Let us apply the same steps now to f1(x) and g(x):
f1(x) 7→ f1(x)− g(x) =5
2x2 − x− 3
2
7→ f2(x) = x2 − 2
5x− 3
5.
In order to save space below, we write this as
f1(x) 7→ f1(x)|g(x)=0 7→ f2(x).
Thus,
f1(x), g(x) ' f2(x), g(x).
Actually, instead of the two steps
f, g ' f1, g ' f2, g,
we can reduce f(x) in one step by substituting x3 = 2x2 − 1, or in other words, by using
the formal equation g(x) = 0:
f(x) 7→ x4 + x2 − x− 1|x3=2x2−1 = 2x3 + x2 − 2x− 1
7→ 2x3 + x2 − 2x− 1|x3=2x2−1 = 5x2 − 2x− 3
7→ f2(x) = x2 − 2
5x− 3
5.
Using f2(x) = 0, we reduce g(x):
g(x) 7→ x3 − 2x2 + 1|x2= 25x+ 3
5=
2
5x2 − 1
5x− 1
5
7→ x2 − 1
2x− 1
2|x2= 2
5x+ 3
5= − 1
10x+
1
10
7→ g1(x) = x− 1,
and then we obtain that
f2, g ' f2, g1.
Finally, by using g1(x) = 0, we reduce f2(x) to:
f2(x) 7→ x2 − 2
5x− 3
5|x=1 = 0.
The chain of equivalences yields f, g ' x − 1, which shows that gcd(f(x),
g(x)) = x− 1.
3
Later on, we will define Grobner basis GB of a set of polynomials, we will see that
GBx4 + x2 − x− 1, x3 − 2x2 + 1 = x− 1.
In general,
GBf1(x), . . . , fm(x) = gcdf1(x), . . . , fm(x)
for polynomials fi(x) in one indeterminate.
§1.2 The Gauss elimination algorithm
Let k be a field and x1, . . . , xn indeterminates. We now solve a system of linear
equations a11x1 + · · ·+ a1nxn = b1,
. . . . . .
am1x1 + · · ·+ amnxn = bm,
(1.1)
where aij, bi ∈ k.It is well-known that this system can be successfully solved by using the Gauss
Elimination algorithm.
Example 1.2 Solve the following system of equations:2x1 + 3x2 + x3 = 1,
3x1 + 2x2 − x3 = 2.
Instead of these equations, let us deal with the corresponding polynomials
f = 2x1 + 3x2 + x3 − 1, g = 3x1 + 2x2 − x3 − 2.
First of all, we need to order the variables. The method is to set x1 > x2 > x3; then in
our case, x1 is the leading monomial of both f and g. Dividing f and g by their leading
coefficients:
f 7→ f1 = x1 +3
2x2 +
1
2x3 −
1
2,
g 7→ g1 = x1 +2
3x2 −
1
3x3 −
2
3,
we have f, g ' f1, g1, where the equivalence means that the corresponding linear
systems are equivalent (have the same solution sets over k). Using f1, we reduce g1:
g1 7→ g1 − f1 = −5
6x2 −
5
6x3 −
1
6
7→ g2 = x2 + x3 +1
5
4
to obtain f1, g1 ' f1, g2. We may in fact stop here, but in order to have something
really simple, we can further use g2 to eliminate x2 in f1:
f1 7→ f2 = x1 − x3 −4
5.
Thus our original system is equivalent to the systemx1 + x3 =
4
5,
x2 + x3 = −1
5,
which is in row-echelon form. As we shall see later, GBf, g = f2, g2.
In general, the Gauss elimination algorithm produces for any consistent system (one
that has solutions) an equivalent system in row-echelon form:xi1 +c1,i1+1xi1+1+ + · · ·+ c1nxn = d1
xi2 + c2,i2+1xi2+1+ + · · ·+ c2nxn = d2
. . . . . .
xil + cl,il+1xil+1 + · · ·+ clnxn = dl
(1.2)
where 1 ≤ i1 < i2 < · · · < il ≤ n.
The system (1.1) is inconsistent if and only if after finitely many steps we arrive at
0 = dl+1,
where dl+1 6= 0. In this case, the system (1.1) is equivalent to the system
0 = 1.
In terms of the Grobner bases theory, for all tuples of linear polynomials f1, . . . , fmwe have
GBf1, . . . , fm = h1, . . . , hl,
where the fi corresponds to (1.1), and either the hj corresponds to (1.2) or
h1, . . . , hl = 1.
The elimination of xi2 from h1 using h2, of xi3 from h1, h2 using h3, and so on, is
not absolutely necessary, but it gives a unique (reduced) Grobner basis independent of
the choice of the elementary transformation steps, which therefore depends only on the
ordering of the indeterminates. A different ordering leads to a different (reduced) basis.
5
§1.3 Systems of polynomial equations
What will be the common features of the two cases considered above?
(i) We fix a total ordering on monomials, which determines the leading monomial of
each polynomial. For 1-variable polynomials, it is the degree ordering xn > xm ⇔ n > m;
For multivariable linear polynomials, it is enough to order its indeterminates.
(ii) We reduce a given set of polynomials by 1. dividing a polynomial by its lead-
ing coefficient; 2. eliminating the leading monomial of one polynomial from the other
polynomials of the set.
We would like to do something similar to an arbitrary system of polynomial equations
in several indeterminates. However, we will see that reductions in (ii) are insufficient,
and we need a new operation s(f, g), which is called the s-polynomial operation. An
alternative name for this is the composition (f, g)w of f and g with respect to some
uniquely defined monomial w.
Example 1.3 Solve the following system of polynomial equationsf = x21x2 − 1 = 0,
g = x31 + 7x1x2 + 1 = 0.
(1.3)
Let us fix x1 > x2 as before, and use the deg-lex (degree-lexicographic) ordering of
monomials to compare two monomials first by degree and then lexicographically. For
instance,
x31 > x2
1x2 > x1x2 > 1.
Again we will deal with the polynomials f and g rather than the equations f = 0
and g = 0. We aim to reduce the set f, g as much as possible by passing to equivalent
system of equations.
The leading monomials of f and g in (1.3) are f = x21x2 and g = x3
1 respectively. We
cannot use g = 0 to reduce f , neither can we use f = 0 to reduce g, for neither f nor g
is a submonomial of the other.
The great discovery of Buchberger (and Shirshov for noncommutative polynomials)
was that in such cases the system can be simplified further by a new s-polynomial (com-
position) operation. Take the least common multiple w = lcm(f , g), which in our case is
x31x2 = fx1 = gx2, and consider s(f, g) = (f, g)w = fx1 − gx2.
For our system (1.3), we have
s(f, g) = (x21x2 − 1)x1 − (x3
1 + 7x1x2 + 1)x2 = −7x1x22 − x1 − x2
7→ h = x1x22 +
1
7x1 +
1
7x2,
6
thus
f, g ' f, g, h, f = x21x2, g = x3
1, h = x1x22.
Proceeding with s(f, h), we obtain the following:
w1 = lcm(f , h) = x21x
22 = fx2 = hx1,
s(f, h) = (f, h)w1 = (x21x2 − 1)x2 − (x1x
22 +
1
7x1 +
1
7x2)x1
7→ q = x21 + x1x2 + 7x2,
thus
f, g, h ' f, g, h, q, f = x21x2, g = x3
1, h = x1x22, q = x2
1.
The reduction
g 7→ g|q=0 = −x21x2 + 1 7→ x2
1x2 − 1|f=0 = 0,
shows that we may exclude g from the set; we are then left with
f, h, q, f = x21x2, h = x1x
22, q = x2
1.
Next, we reduce f :
f 7→ f |q=0 = −x1x22 − 7x2
2 − 1
7→ x1x22 + 7x2
2 + 1|h=0
7→ f1 = x22 −
1
49x1 −
1
49x2 +
1
7,
thus
f, h, q ' f1, h, q, f1 = x22, h = x1x
22, q = x2
1.
Now we see that h is also redundant:
h 7→ h|f1=0 7→ x21 + x1x2 + 7x2|q=0 7→ 0,
thus
f1, h, q ' f1, q, f1 = x22, q = x2
1,
and we conclude that (1.3) is equivalent tof1 = x2
2 −1
49x1 −
1
49x2 +
1
7= 0,
q = x21 + x1x2 + 7x2 = 0.
(1.4)
We cannot reduce (1.4) any further: since x22 and x2
1 are coprime, the eliminations are
impossible, while s-polynomial operations only introduce redundancies.
7
The theory of Grobner bases guarantees that (1.4) is the unique simplest system
equivalent to (1.3) as long as we fix the ordering of indeterminants x1 > x2 and use the
deg-lex ordering on monomials.
In terms of Grobner bases, we have GBf, g = f1, q.To solve (1.4) let us change the ordering to lex-deg ordering to compare two mono-
mials first lexicographical and then by degree. For example,
x21 > x1x
22 > x1x2 > x1 > x2
2 > x2 > 1.
Then (1.3) is equivalent to
f2 = x1 + x2 − 49x22 − 7 = 0, q1 = (49x2
2 − x2 + 7)2 + (49x22 − x2 + 7)x2 + 7x2 = 0.
Actually GBf, g = f2, q1 in lex-deg ordering of monomials.
This system can be solved in radicals.
Now we are going to put another problem concerning to the system (1.3). Namely
we address to the following equality, or word problem for (1.3):
Problem. Find an algorithm that for a given polynomial F (x1, x2) would answer whether
F (x1, x2) = 0 modulo f = 0 and g = 0.
In other words, we need to determine the validity of the implication for each F , that
is,
f = 0, g = 0⇒ F = 0. (1.5)
To be more precise, it means that F (x1, x2) = h1(x1, x2)f(x1, x2) + h2(x1, x2)g(x1, x2) for
some polynomials h1, h2.
The desired algorithm is: to reduce F (x1, x2) as much as possible by using f1 = 0
and q = 0 (but not f and g!). If zero results, then (1.5) is true; otherwise, (1.5) is false.
Example 1.4 Keeping the same f and g as in (1.3), let
F = x1x32 + α1x1x
22 + α2x1x2 + α3x1 + α4x2 + α5,
where αi ∈ k. For which values of αi is (1.5) valid? The series of reductions
F 7→ F |f1=0 = F1
7→ F1|q=0 = F2
7→ F2|f1=0 = (α2 −1
7)x1x2 + (α3 −
α1
7− 1
73)x1
+(α4 −α1
7− 1
73)x2 + α5
8
quickly yields the answer: (1.5) is valid if and only if
α2 =1
7, α3 = α4 = α1
7+ 1
73 , α5 = 0,
where α1 ∈ k is arbitrary. Note that we cannot reduce F at all just by using the initial
polynomials f and g!
Anyway, we can formulate the Buchberger’s algorithm as follows:
Algorithm. Given any set of polynomials in several indeterminates, add the s-polynomials
to the set, and reduce polynomials by eliminating their leading monomials, also dividing
them by their leading coefficients. Stop when no further reductions are possible (so in
particular, when all s-polynomials reduce to 0).
The main significance is that we can always solve the equality problem for every finite
system of polynomial equations, by using the Buchberger’s algorithm.
§1.4 Examples of systems of noncommutative (NC)
polynomial equations
Let X = x1, x2, . . . be a set (an alphabet) of letters (indeterminates), X∗ the set
of all words (strings) in X including the empty word 1,
1, x1, x2, x21, x1x2, x2x1, x
22, x
21x2, x1x2x1, x
22x1, . . . .
A linear combination of words is called a NC polynomial. Product and sum of NC
polynomials are defined in a natural way. For example,
(x1 + x2)2 − x2
1 − x22 = x1x2 + x2x1,
(x1 + x2 + x3)3 − (x2 + x3)
3 − (x1 + x3)3 − (x1 + x2)
3 + x31 + x3
2 + x33
= x1x2x3 + x1x3x2 + x2x1x3 + x2x3x1 + x3x1x2 + x3x2x1,
(x1 + · · ·+ xn)n −∑
1≤i≤n
(x1 + · · ·+ xi + · · ·+ xn)n
+∑
1≤i1<i2≤n
(x1 + · · ·+ xi1 + · · ·+ xi2 + · · ·+ xn)n − · · ·+ (−1)n−1∑
i
xni
=∑
(i1,i2,...,in)∈Pn
xi1xi2 · · ·xin ,
9
where Pn is the set of all permutations of (1, 2, . . . , n).
In this section we discuss Shirshov algorithm for a reduction of a system of NC poly-
nomial equations in order to try to solve the word (equality) problem (in general, the word
problem for a finite system of NC polynomial equations is algorithmically unsolvable).
Let X is a well-ordered set. Then we define the deg-lex ordering on X∗: first compare
two words by length (degree) and then by comparing them lexicographically.
We will use a Shirshov’s notion of composition (f, g)w of two NC polynomials relative
to a word w, where w = f b = ag, a, b ∈ X∗, f , g are the maximal words of f, g
correspondingly in deg-lex ordering of words, and deg(w) < deg(f) + deg(g). We call w
a least common multiple of f , g, w = lcm(f , g).
Example 1.5 The quaternion system H over real numbers R (due to W.R. Hamilton
1805-1865) is
x21 + 1 = 0, x2
2 + 1 = 0, x23 + 1 = 0, x1x2 − x3 = 0, x2x3 − x1 = 0,
x3x1 − x2 = 0, x2x1 + x3 = 0, x3x2 + x1 = 0, x1x3 + x2 = 0.
Any NC polynomial f(x1, x2, x3) can be uniquely present mod (H) in a form
f = α0 + α1x1 + α2x2 + α3x3 mod(H), αi ∈ R. (1.6)
Let us order x1 < x2 < x3 and the deg-lex ordering of words in X = x1, x2, x3. There
are compositions relative to following words wi = lcm(f , g), 1 ≤ i ≤ 16, where f, g are
from the list.
w1 = x21x2, w2 = x2
1x3, w3 = x22x3, w4 = x2
2x1, w5 = x23x1, w6 = x2
3x2,
w7 = x1x22, w8 = x1x2x1, w9 = x2x
23, w10 = x2x3x2, w11 = x3x
21,
w12 = x2x21, w13 = x3x
22, w14 = x3x2x3, w15 = x1x
23, w16 = x1x3x1.
Let us check first two compositions (f, g are uniquely defined by w).
(f, g)w1 = (x21 + 1)x2 − x1(x1x2 − x3) = x2 + x1x3 ≡ 0 mod(x1x3 = −x2),
(f, g)w2 = (x21 + 1)x3 − x1(x1x3 + x2) = x3 − x1x2 ≡ 0 mod(x1x2 = x3).
So, all compositions are trivial. By Composition-Diamond lemma (Theorem 4.4) any
presentation (1.6) is unique. It is solved the word problem for H: for any two polyno-
mials f(x1, x2, x3), g(x1, x2, x3) one can decide whether f = g mod(H), i.e., f − g =∑αiaihibi, αi ∈ R, hi ∈ H, ai, bi ∈ X∗.
10
Example 1.6 The Grassmann (exterior) system G(X) of NC polynomial equations (due
to H.G. Grassmann 1809-1877) is
x2i = 0, 1 ≤ i ≤ n, xixj + xjxi = 0, 1 ≤ j < i ≤ n.
Let us order xi < xj for i < j, and deg-lex ordering of words in X = x1, . . . , xn. There
are compositions of NC polynomials from G(X) relative to words w = lcm(f , g), where
f, g are from the list.
wiij = xixixj, i > j, wijj = xixjxj, i > j, wijk = xixjxk, i > j > k.
Let us check the last composition
(f, g)wijk= (xixj + xjxi)xk − xi(xjxk + xkxj) = xjxixk − xixkxj
≡ −xjxkxi + xkxixj mod(xixk = −xkxi)
≡ xkxjxi − xkxjxi mod(xjxk = −xkxj, xixj = −xjxi)
≡ 0.
Again by CD-lemma any polynomial has unique presentation as a linear combinations of
words
xi11 · · ·xin
n , 0 ≤ ij ≤ 1.
Example 1.7 For n = 2 and the deg-lex ordering on words in X = x1, x2 with x1 < x2,
the set
f = x1x22 − x2, g = x2
2x1 − x1
admits only three compositions:
w1 =x1x22x1 = fx1 = x1g, (f, g)w1 =x2x1 − x2
1;
w2 =x1x32x1 = fx2x1 = x1x2g, (f, g)w2 =x2
2x1 − x1x2x1;
w3 =x22x1x
22 = gx2
2 = x22f , (g, f)w3 =x3
2 − x1x22.
Consider now the equality problem for the system
f = x1x22 − x2 = 0, g = x2
2x1 − x1 = 0
applying Shirshov algorithm which means adding compositions and reducing the polyno-
mials as before. The first compositions
(f, g)w1 = h = x2x1 − x21,
(f, g)w2 7→ x22x1 − x1x2x1|g=0, h=0 7→ q = x3
1 − x1,
(g, f)w3 7→ x32 − x1x
22|f=0 7→ r = x3
2 − x2
11
lead to the further compositions
(f, h)x1x22x1
= fx1 − x1x2h
7→ x1x2x21 − x2x1|h=0
7→ x41 − x2x1|q=0 7→ x2
1 − x2x1|h=0 7→ 0,
(h, f)x2x1x22
= hx22 − x2f
7→ x21x
22 − x2
2|f=0 7→ t = x22 − x1x2,
and then to the reductions
f 7→ x1x22 − x2|t=0 = f1 = x2
1x2 − x2,
r 7→ x32 − x2|t=0 7→ x1x
22 − x2|t=0 = r1 = x2
1x2 − x2.
Now g is redundant because
g 7→ x22x1 − x1|t=0 7→ x1x2x1 − x1|h=0 7→ x3
1 − x1|q=0 = 0.
The remaining set of four polynomials
f1 = x21x2 − x2, h = x2x1 − x2
1, t = x22 − x1x2, q = x3
1 − x1
admits only the following trivial compositions:
(f1, h)x21x2x1
= f1x1 − x21h 7→ x4
1 − x2x1|q=0, h=0 = 0,
(h, f1)x2x21x2
= hx1x2 − x2f 7→ x31x2 − x2
2|q=0, t=0 = 0,
(q, f1)x31x2
= qx2 − x1f = 0,
(q, f1)x41x2
= qx1x2 − x21f = 0,
(q, f1)x51x2
= qx21x2 − x3
1f = 0,
(f1, r)x21x3
2= f1x
22 − x2
1r 7→ x32 − x2
1x2|r=0, f1=0 = 0,
(f1, t)x21x2
2= fx2 − x2
1t 7→ x31x2 − x2
2|q=0, t=0 = 0,
(h, q)x2x31
= hx21 − x2q 7→ x4
1 − x2x1|q=0, h=0 = 0,
(r, h)x32x1
= rx1 − x22h 7→ x2
2x21 − x2x1|t=0
7→ x1x2x21 − x2x1|h=0 7→ x4
1 − x21|q=0 = 0,
(t, t)x32
= tx2 − x2t 7→ x1x22 − x2x1x2|t=0,h=0 = 0,
as well as some manifestly trivial compositions (q, q)w. Therefore, we have
GSBx1x22 − x2, x
22 − x1 = x2
1x2 − x2, x2x1 − x21, x
22 − x1x2, x
31 − x1.
12
The Shirshov’s Composition-Diamond Lemma (Theorem 4.4) implies that any polynomial
F (x1, x2) can be uniquely presented mod(f = 0, g = 0) as a linear combination of the five
words 1, x1, x21, x2, x1x2.
Let us call a presentation of NC polynomials by these monomials its canonical pre-
sentation. The Composition-Diamond lemma gives the positive solution to the equality
problem: a polynomial is equal to zero mod(f = 0, g = 0) if and only if its canonical
presentation is zero. For instance, let us check whether Fn = xn2x
n1 −xn
1xn2 is equal to zero
mod(f = 0, g = 0):
F0 = 1− 1 = 0;
F1 7→ x2x1 − x1x2|h=0 = x21 − x1x2 6= 0,
which is the canonical presentation of this polynomial,
F2 7→ x22x
21 − x2
1x22|t=0
7→ x1x2x21 − x2
1x1x2|h=0 7→ x41 − x3
1x2|q=0 = x21 − x1x2,
Fn 7→ · · · 7→ x21 − x1x2.
§1.5 Systems of Lie polynomials equations
In this section we will deal with Shirshov algorithm for the word (equality) problem
for systems of Lie polynomial equations. It would take a lot of combinatorial preparations
connected with Lyndon-Shirshov words and Lyndon-Shirshov Lie monomials. All proofs
will be in Chapter 5.
§1.5.1 Words and Shirshov’s bracketing (arrangement of parentheses)
Let X = x1, x2, . . . be an alphabet with x1 < x2 < . . . , X∗ the set of all words
in X. For simplicity we will use also numbers 1, 2, . . . instead of letters x1, x2, . . . . We
denote e the empty word.
Let us order X∗ using lex-antideg ordering to compare two words first lexicographi-
cally and then by “anti-degree”: the shorter word is larger. For example, 21 < 2 < 333 <
33 < 34 < e.
Algorithm of the bracketing (the arrangement of parentheses) of words. Let w be a
word and xi be the minimal letter in w. If w = (xi)ku, k > 0, u 6= xiv, then go to u,
the longest suffix of w that does not begin with the minimal letter of w. If w does not
begin with xi, then join minimal letters of w to the previous ones, order “new letters” in
lex-antideg and continue this process.
13
For example, w = 11212122311312321, w1 = 11(21)(21)22((31)1)(31)23(21), w2 =
11(21)(21)22((31)1)(31)2(3(21)), w3 = 11(21)(21)22((31)1)((31)2)(3(21)). Here our algo-
rithm stops. New “letters with brackets” are 1, (21), ((31)1), ((31)2), (3(21)) and new
“letters without brackets” are 1, 21, 311, 312, 321.
First information in w3. Any new “letter without parentheses” u is so called Lyndon-
Shirshov (LS) word: u = u1u2 > u2u1 lexicographically for any u1, u2 6= 1.
In our case 1, 21, 2, 311, 312, 321 are LS words.
Second information in w3. Word w is a product of LS words in non-decreasing lex-
antideg ordering, w = 1 · 1 · 21 · 21 · 2 · 2 · 311 · 312 · 321, 1 < 21 < 2 < 311 < 312 < 321.
It is an illustration of the
Theorem 1.8 (Shirshov factorization) Any word is a unique product of LS words in non-
decreasing ordering relative to lex-antideg ordering.
Third information in w3. New “letters with brackets” are non-associative LS mono-
mials: (w) = ((u)(v)) is called a LS monomial if w (the associative support of (w)) is a
LS word, (u), (v) are LS monomials, u > v in lex-antideg, and if (u) = ((u1)(u2)), then
u2 ≤ v in lex-antideg.
In our case, monomials 1, (21), 2, ((31)1), ((31)2), (3(21)) are LS monomials. For
example, w1 = ((31)1), (31) > 1, 1 ≤ 1, w2 = ((31)2), (31) > 2, 1 ≤ 2, w3 =
(3(21)), 3 > (21), 2 > 1.
It is an illustration of the
Theorem 1.9 (Lyndon-Shirshov theorem) There is 1-1 correspondence between LS words
and non-associative LS monomials. It means that for any LS word w one may put one
and only one parentheses (w) such that (w) is a LS monomial.
Shirshov’s Algorithm of bracketing a LS word w with a LS subword u, w = aub (the
special bracketing), w = aub → (w)u = (a(u)b), where (u) is a standard bracketing as
before.
– To put the Shirshov’s brackets (the Shirshov’s arrangement parentheses) on u first
as before, w1 = a(u)b.
– To add a minimal (in lex-antideg ordering) “letter” of w1, including (u), to the
previous ones.
– To order new “letters” in lex-antideg and to continue the same process.
14
Example 1.10 w = 2212112111211 = 2u2111211, u = 21211.
w1 = 2((21)((21)1))2111211,
w2 = 2((21)((21)1))(((21)1)1)((21)1),
w3 = 2(((21)((21)1))(((21)1)1))((21)1),
w4 = 2((((21)((21)1))(((21)1)1))((21)1)),
w5 = (2((((21)((21)1))(((21)1)1))((21)1))).
To better understand the special bracketing algorithm, let us start with LS monomial
(w) = (2(21)[((211)(2111))(211)]). Here brackets . . . are the minimal pair of brackets
that cover u. We change /// in the following way . . . → (u)c, c = 2111211. Then
we use Shirshov factorization for c, (u)c → (u)(2111)(211), and now the left normed
bracketing (u)(2111)(211)→ (((u)(2111))(211)). Then (w)u = w5.
§1.5.2 Lie monomials and Lie polynomials
Let X = x1, x2, . . . be an alphabet with x1 < x2 < . . . . Lie monomials in X are
defined as follows:
(i) xi’s are Lie monomials of the degree 1;
(ii) If u, v are Lie monomials, deg(u) = n, deg(v) = m, then [uv] = uv − vu is a Lie
monomial of degree n+m.
Example 1.11
x1, x2, [x1x2] = −[x2x1] (by the anticommutatvity identity [ab] = −[ba]),
[[x2x1]x1] = x2x1x1 − 2x1x2x1 + x1x1x2,
[x2[x2x1]] = x2x2x1 − 2x2x1x2 + x1x2x2,
[x2[x2x1]][[x2x1]x1]] (by the Leibniz identity [[ab]c] = [[ac]b] + [a[bc]])
= [[x2[[x2x1]x1]][x2x1]] + [x2[[x2x1][[x2x1]x1]]]
= x2x2x1x2x1x1 + · · ·+ x1x1x2x1x2x2 + . . . ,
where x2x2x1x2x1x1 >lex · · · >lex x1x1x2x1x2x2 >lex . . . .
A linear combination of Lie monomials over a field k is called a Lie polynomial. Let
Lie(X) be the linear space of Lie polynomials. It is a subspace of NC polynomials k〈X〉.How to check that a NC polynomial is a Lie polynomial? There are several criterions.
One of the most popular is
15
Friedrichs’ criterion (due to K.O. Friedrichs 1901-1982). Let chk = 0. A NC polyno-
mial f(x1, x2, . . . , xn) is a Lie polynomial iff
[x1y1] = 0, . . . , [xnyn] = 0⇒ f(x1 + y1, . . . , xn + yn) = f(x1, . . . , xn) + f(y1, . . . , yn).
Another interesting and important observation about Lie polynomials is as follows.
Zassehnaus-Jacobson Lie polynomial. Let k be a field of characteristic p > 0. Then
(a+ b)p = ap + bp + Λ(a, b),
where Λ(a, b) is a Lie polynomial of Zassenhaus-Jacobson. For example, p = 2, (a+ b)2 =
a2 + b2 + [ab]; p = 3, (a+ b)3 = a3 + b3 + [a[ab]] + [[ab]b].
In general,
(a+ b)p = ap + bp +∑
1≤i≤p−1
1/isi(a, b),
where si is the coefficient on ti−1 of the polynomial
[(ta+ b)p−1a] =: [[
p−1︷ ︸︸ ︷(ta+ b)[. . . [(ta+ b) a] . . . ]]], ta = at, tb = bt.
§1.5.3 Serre system
Serre system L(An) of Lie polynomial equations of type An (due to J.-P. Serre, born
in 1926) are as follows.
Let k be a field with chk 6= 2.
Let H = h1, . . . , hn, X = x1, . . . , xn, Y = y1, . . . , yn. We order alphabet
H ∪X ∪ Y by hi > xj > yk for any i, j, k and hi > hj, xi > xj, yi > yj for any i > j. Let
An is the Cartan matrix (due to Elie Cartan 1869-1951), i.e., aii = 2, ai,i+1 = −1, ai,i−1 =
−1, aij = 0 if |i− j| > 1.
Then L(An) is the Lie system of equations in H ∪X ∪ Y over k:
[hihj] = 0 (i > j), [xiyi] = hi, [xiyj] = 0 (i 6= j),
[hixj]− aijxj = 0, [hiyj] + aijyj = 0,
[xi[xixi−1]] = 0, [[xixi−1]xi−1] = 0, [xixj] = 0 (i > j + 1),
[yi[yiyi−1]] = 0, [[yiyi−1]yi−1] = 0, [yiyj] = 0 (i > j + 1).
In order to use the Shirshov algorithm for Lie polynomials we need
(i) To define Lyndon-Shirshov linear basis of the linear space of Lie polynomials in
H ∪X ∪ Y with an ordering;
16
(ii) To define Lie composition of Lie polynomials from L(An);
(iii) To add all non-trivial compositions to initial set of equations in order to have a
GS basis GSB(L(An));
(iv) To find all LS monomials [w] such that the word w does not contain maximal
words s for all s ∈ GSB(L(An)).
In fact, we have a CD-lemma for Lie algebras in Chapter 5 and solve all the above
problems. Now we outline the solution.
§1.5.4 Lyndon-Shirshov words
Let us remind that a word w is called a LS word if
w = uv >lex vu for any u, v 6= 1.
For example, x2x2x1x2x1x1 is a LS word, and x2x1x1x2x1, x2x2x1x2x2x1 are not.
Proposition 1.12 A deg-lex maximal word of any Lie monomial (and polynomial) is a
LS word.
For example, The deg-lex maximal word of [[[x2x1][x2x1x1x1]][x2x1x1]] is LS word
x2x1x2x1x1x1x2x1x1. For [[[x2x1][x2x1x1]][x2x1x1]], it is x2x1x2x1x1x2x1x1.
§1.5.5 Lyndon-Shirshov Lie monomials
A Lie monomial [w] is called a Lyndon-Shirshov Lie monomial if
(i) the word w is a LS word;
(ii) if [w] = [[u][v]], then [u], [v] are LS Lie monomials and u v in the lex-antideg
ordering;
(iii) if [w] = [[[u1][u2]][v]], then u2 v in the lex-antideg ordering.
For examples, [xi[xixi−1]], [[xixi−1]xi−1] are LS Lie monomials since xi xi−1 and
xi xixi−1. Also the following Lie monomials are Lyndon-Shirshov’s
[[xi+jxi+j−1 . . . xi−1]xi+j−1], [[xi+j . . . xi][xi+j . . . xixi−1]].
Proposition 1.13 The set of LS Lie monomials is a linear basis of Lie(X), the linear
space of Lie polynomials.
Proposition 1.14 There is 1-1 correspondence between LS Lie monomials [w] and LS
words w.
17
We order LS Lie monomials [w] by the deg-lex ordering of LS words w.
Proposition 1.15 For any LS monomial [w], a maximal in deg-lex ordering word of [w],
as NC polynomial, is equal to w.
Proposition 1.16 For any LS word w there are two equivalent ways to put Lie brackets
in order to get a LS monomial [w]:
(i) To join a minimal letter of w to the previous ones, the ordering of new alphabet
by lex-antideg ordering, and continue the process.
(ii) If [w] = [[u][v]], where v is the longest proper LS suffix of w, then u is a LS word.
For examples,
[xnxn−1 . . . x1] = [xn[xn−1 . . . x1]], or by step by step
xnxn−1 . . . x2x1, xnxn−1 . . . x3[x2x1], xnxn−1 . . . [x3[x2x1]], . . . , [xn[xn−1[...[x3[x2x1]]]]].
[xi+k . . . xi+1xixi+k . . . xi+1xixi−1] = [[xi+k . . . xi+1xi][xi+k . . . xi+1xixi−1]], or
xi+k . . . xi+1xixi+k . . . xi+1xixi−1,
xi+k . . . xi+1xixi+k . . . xi+1[xixi−1],
xi+k . . . xi+1xixi+k . . . [xi+1[xixi−1]],
xi+k . . . xi+1xi[xi+k[. . . [xi+1[xixi−1]] . . . ]],
xi+k . . . [xi+1xi][xi+k[. . . [xi+1[xixi−1]] . . . ]],
. . .
xi+k[. . . [xi+1xi] . . . ][xi+k[. . . [xi+1[xixi−1]] . . . ]],
[xi+k[. . . [xi+1xi] . . . ]][xi+k[. . . [xi+1[xixi−1]] . . . ]],
[[xi+k[. . . [xi+1xi] . . . ]][xi+k[. . . [xi+1[xixi−1]] . . . ]]].
There is an algorithm to present any Lie monomial as a linear combination of LS
monomials by induction on degree:
(i) do [[u][v]]→ −[[v][u]], if [u], [v] are LS monomials and u ≺ v,
(ii) do [[[u1][u2]][v]]→ [[[u1][v]][u2]] + [[u1][[u2][v]]], if u2 v.
For example, [[32]1] = [[31]2] + [3[21]], [[12]2] = −[[21]2] = [2[21]], [[[32]1][[12]2]] =
[[[31]2][2[21]]] + [[3[21]][2[21]]] = [[[31][2[21]]]2] + [[31][2[2[21]]]] + [[3[21]][2[21]]].
18
§1.5.6 Lie compositions
To define Lie composition 〈f, g〉w we need to define an associative composition (f, g)w
then to put special Lie brackets on the result.
Let us define and check some compositions for L(An).
1. w = xi+1xixixi−1. Then (f, g)w = fxi−1 − xi+1g, 〈f, g〉w = [fxi−1] − [xi+1g] =
[[[xi+1xi]xi]xi−1]− [xi+1[xi[xixi−1]]] = A−B.
Here B is LS monomial, A is not. So, first of all we must present A as a linear com-
binations of LS monomials using anticommutativity ([ab] = −[ba]) and Leibniz identity
[[ab]c] = [[ac]b] + [a[bc]]). We know in advance that the maximal LS monomial of A is B
with coefficient +1. We have A = [[[xi+1xi−1]xi]xi]+[[xi+1[xixi−1]]xi]+[[xi+1xi][xixi−1]] =
[[[xi+1xi−1]xi]xi] + [[xi+1[xixi−1]]xi] + [[xi+1[xixi−1]]xi] +B.
The first LS monomial is zero mod(L(An), w). So, after reduction,
〈f, g〉w = 2[[xi+1[xixi−1]]xi] mod(S,w)
and we must add [[xi+1[xixi−1]]xi] = 0 to S (since 2 6= 0 in k).
2. w = xi+1xi+1xixixi−1, (f, g)w = fxixi−1 − xi+1xi+1g, 〈f, g〉w = [f [xixi−1]] −[xi+1[xi+1g]].
The rule of special bracketing. In fxixi−1 a minimal letter is xi−1. Join by bracket
this letter to the previous one, [xixi−1], then joint “new letter” [xixi−1] to f .
Then 〈f, g〉w = [[xi+1[xi+1xi]][xixi−1]]− [xi+1[xi+1[xi[xixi−1]]]] = A−B.
Again, B is a LS monomial and A is not. We have A = [[xi+1[xixi−1]][xi+1xi]] +
[xi+1[[xi+1xi][xixi−1]]] = −A1 + A2, A2 = −A1 +B.
So, [f, g]w = −2A1, and we must add to L(An) equation A1 = [[xi+1xi][xi+1[xixi−1]] =
0.
Continuing Shirshov algorithm, we will have the following set of multiple composi-
tions (see §1.5.3):
[[xi+jxi+j−1 . . . xi−1]xi+j−1] = 0, [[xi+j . . . xi][xi+j . . . xixi−1]] = 0,
[[yi+jyi+j−1 . . . yi−1]yi+j−1] = 0, [[yi+j . . . yi][yi+j . . . yiyi−1]] = 0,
j ≥ 1, i ≥ 2, i+ j ≤ n,
where [z1z2 . . . zm] = [z1[z2 . . . zm]]. A GS basis GSB(L(An)) of the system L(An) consists
of initial equations together with the above relations.
19
The set Irr(GSB(L(An)) is h1, . . . , hn ∪ [xi+j . . . xi], n ≥ i + j ≥ i ≥ 1 ∪[yi+j . . . yi], n ≥ i+ j ≥ i ≥ 1.
By CD-lemma for Lie equations,
L(An)→ F = 0
iff F goes to zero by eliminations of maximal monomials of GSB(L(An)). It is an algo-
rithmically solvable problem since GSB(L(An)) is a finite set.
20
Chapter 2 Free Commutative, Associative and Lie
Algebras
We will deal with the following classes (varieties) of linear algebras over a fixed field
k – commutative algebras Com, associative algebras As, Lie algebras Lie:
Com : xy = yx, (xy)z = x(yz),
As : (xy)z = x(yz),
Lie : x2 = 0 (it follows xy = −yx, the anticommutativity identity),
(xy)z + (yz)x+ (zx)y = 0 (the Jacobi identity).
Examples.
1) Com contains polynomial algebras. Any commutative algebra is a quotient algebra
of a polynomial algebra.
2) As contains non-commutative polynomial algebras. Any associative algebra is a
quotient of a non-commutative polynomial algebra.
3) Lie contains Lie polynomial algebras. Any Lie algebra is a quotient of an algebra
of Lie polynomials. For any associative algebra A the algebra A(−) with the Lie product
[xy] = xy−yx is a Lie algebra. Any Lie algebra (over a field!) is a Lie subalgebra of some
associative algebra (Poincare-Birkhoff-Witt theorem, see below).
§2.1 Commutative polynomials
Let X = x1, x2, . . . , xn, . . . be a finite or countable linearly ordered set, x1 > x2 >
· · · > xn > . . . (variables or letters) (actually X may be any well-ordered set).
Remark The above ordering onX (x1 > x2 > . . . ) is common for the case of commutative
variables and follows the ordering of variables in the Gauss elimination algorithm. For
noncommutative variables one uses the opposite ordering x1 < x2 . . . .
By k[X] one denotes the usual polynomial algebra on X,
k[X] = ∑
i=(i1,...,in)
αixi11 . . . x
inn |n ≥ 0, ij ≥ 0, αi ∈ k.
The algebra k[X] satisfies a following universal property: For any commutative alge-
bra C and any map ε : X → C, there exists a unique algebra homomorphism f : k[X]→ C
such that the following triangle is commutative
21
-
?
C
k[X]X i
ε∃! f
where i is the inclusion map.
A monomial u = xi11 . . . x
inn is called a commutative word of the length |u| (degree
deg(u)). Let [X] be the set of all commutative words in X (including 1). Then [X] is a
linear basis of k[X].
From the universal property it follows that any commutative algebra C is a quotient
of some k[X],
C = k[X]/I,
where I is an ideal of k[X]. Let S be a set of generators of I, i.e., I = Id(S) =
∑αiaisi|ai ∈ [X], si ∈ S, αi ∈ k. Then by definition
C = k[X]/Id(S) = k[X|S],
the commutative algebra with generators X and defining relations si = 0, si ∈ S.
Our main problem is, starting with S, to find a special system Sc of defining relations
of C that is called a Grobner basis of the ideal I, or a GB of the C. Sc can be found
using the Buchberger algorithm.
We will use the following theorem.
Theorem 2.1 (Hilbert Basis Theorem) Let X = x1, . . . , xn be a finite set. Then any
ideal I C k[X] is finitely generated,
I = Id(f1, . . . , fm) = m∑
i=1
gifi|gi ∈ k[X], 1 ≤ i ≤ m, m ≥ 1.
This theorem follows by induction on n, using the following lemma.
Lemma 2.2 Let R be a commutative Noetherian ring with 1 (i.e., any ideal of R is finitely
generated). Then the polynomial ring R[x] (on one variable) is a commutative Noetherian
ring as well.
Proof. Let I C R[x] and In (the nth leading ideal) the set of leading coefficients of
elements in I of degree ≤ n. Then clearly In CR, In ⊆ In+1. Moreover, if I ′ CR[x] with
I ⊆ I ′ and with In = I ′n for all n ≥ 0, then I = I ′. Otherwise, take f ∈ I ′ \ I with least
22
possible degree, say, the degree of f to be m. Then f = axm + f1, 0 6= a ∈ I ′m, where
either f1 = 0 or the degree of f1 less than m. Since a ∈ Im, there exists a g = axm+g1 ∈ I,where either g1 = 0 or the degree of g1 less than m. Thus, f − g = f1 − g1 ∈ I ′ which
implies that f1 = g1 since m is the least. This contradicts f 6∈ I.Let L0 ⊆ L1 ⊆ . . . be an ascending chain of ideals of R[x] and Lin the nth leading
ideal of Li. Note that Lij ⊆ Lkm whenever i ≤ k and j ≤ m. Since R is Noetherian, the
ascending chain Lii|i ≥ 0 of ideals of R stabilizes, say at Ljj. For any n with 0 ≤ n ≤j−1 the ascending chain Lin|i ≥ 0 stabilizes, say at Ltnn. Let m = maxj, t1, . . . , tj−1.Then for all i ≥ m and n ≥ 0, Lin = Lmn. Thus Li = Lm. So R[x] is Noetherian.
The algebra k[X] is called a free commutative-associative algebra over X and k.
§2.2 Noncommutative polynomials
Again, let X = x1, . . . , xn, . . . be a finite or countable linearly ordered set, x1 <
x2 < . . . xn < . . . (actually X may be any well-ordered set).
Following P.M. Cohn, by k〈X〉 one denotes the algebra of noncommutative polyno-
mials on X with coefficients in k,
k〈X〉 = ∑
i1,...,im
αi1...imxi1 . . . xim|m ≥ 0, αi1...im ∈ k.
A monomial u = xi1 . . . xim is called a (associative or noncommutative) word in X, the
length |u| (or degree deg(u)) is m.
Let X∗ be the set of all words in X. Then X∗ is a linear basis of k〈X〉.The algebra k〈X〉 satisfies a following universal property: For any associative algebra
A and any map ε : X → A, there exists a unique algebra homomorphism f : k〈X〉 → A
such that the following triangle is commutative
-
?
A
k〈X〉X i
ε∃! f
It follows that any associative algebra A is a quotient of some k〈X〉,
A = k〈X〉/I,
where I is an ideal of k〈X〉. Let S be a set of generators of I, i.e., I = Id(S) =
23
∑αiaisibi|ai, bi ∈ X∗, si ∈ S, αi ∈ k. Then by definition
A = k〈X〉/Id(S) = k〈X|S〉,
the associative algebra with generators X and defining relations si = 0, si ∈ S.
Our main problem is, starting with S, to find a special system Sc of defining relations
of A that is called a Grobner-Shirshov basis of the ideal I, or a GSB of A. Sc can be
found using the Shirshov algorithm for associative algebras.
The algebra k〈X〉 is refereed to as a free ring (free associative algebra on X and k).
The theory of free rings had been developed in the book
P.M. Cohn, Free rings and their relations, Academic Press, 1974 (second edition
1981).
One of the main properties of k〈X〉 is
Theorem 2.3 (Cohn Theorem) Any left (right) ideal I of k〈X〉 is a free left (right) k〈X〉-module, i.e., there exist S ⊂ I such that any polynomial f ∈ I has a unique presentation
f =∑fisi, si ∈ S, fi ∈ k〈X〉.
It may be proved using GS bases theory for modules (actually S is a minimal GS
basis of I as a left k〈X〉-module, see section §4.2).
Another general property of k〈X〉 is
Theorem 2.4 (Kemer Theorem) Any fully characteristic ideal T of k〈X〉 (i.e., an ideal
that is closed under substitutions xi → fi for any fi ∈ k〈X〉) is finitely generated as a
fully characteristic ideal.
The result is proved in the book: A.R. Kemer, Identities of associative algebras,
AMS, 1990.
§2.3 Lie polynomials
Again, let k be a field, X = x1, x2, . . . a well-ordered set, x1 < x2 < . . . . A Lie
polynomial f in X and k is a noncommutative polynomial in X and k (i.e., f ∈ k〈X〉)which is a linear combination of Lie words in X. A Lie word is defined by induction:
1) xi is a Lie word of the length (degree) 1; 2) If u, v are Lie words of length k and l
respectively, then [uv] = uv − vu is a Lie word of length k + l.
Examples of Lie words are xi, [xixj], [[xixj]xk], [xi[xjxk]], [[xixj[[xkxl]], [[. . . [xi1xi2 ] . . . ]xik ]
(a left normed Lie word), [xi1 [. . . [xik−1xik ] . . . ] (a right normed Lie word).
24
The set of all Lie polynomials in X and k will be denoted by Liek(X) or Lie(X) if
k is fixed. It is a Lie algebra under the Lie product [fg] = fg − gf . Actually it is a Lie
subalgebra of k〈X〉(−) generated by X.
Again, the algebra Lie(X) satisfies a following universal property: For any Lie algebra
L and any map ε : X → L, there exists a unique Lie algebra homomorphism f : Lie(X)→L such that the following triangle is commutative
-
?
L
Lie(X)X i
ε∃! f
It follows from PBW Theorem, see Section §4.1.1, also Theorem 5.28.
Again it follows that any Lie algebra L is a quotient of some Lie(X),
L = Lie(X)/I,
where I is an ideal of Lie(X). Let S be a set of generators of I, i.e., I = Id(S) =
∑αi[aisibi]|ai, bi ∈ X∗, si ∈ S, αi ∈ k. Then by definition
C = Lie(X)/Id(S) = Lie(X|S),
the Lie algebra with generators X and defining relations si = 0, si ∈ S.
Our main problem is, starting with S, to find a special system Sc of defining relations
of L that is called a Grobner-Shirshov basis of the ideal I, or a GSB of L. Sc can be
found using the Shirshov algorithm for Lie algebras.
Liek(X) is called a free Lie algebra on X and k.
It is not difficult to prove that any Lie word is a linear combination of left (right)
normed Lie words. So left (right) normed Lie words is a set of linear generators of Lie(X).
But they are linearly dependant.
A natural question arises: What is a linear basis of Lie(X)?
Up to now there are known several such bases:
The Hall basis (M. Hall, 1949), the Hall-Shirshov series of bases (Shirshov, 1953,
1962), the Lyndon-Shirshov basis (Chen, Fox, Lyndon, 1958, Shirshov, 1958).
Let us remark that Lyndon-Shirshov basis is a particular case of Hall-Shirshov series
of bases.
One of the main property of a free Lie algebra is
Theorem 2.5 (Shirshov-Witt Theorem) Any subalgebra of a free Lie algebra is a free Lie
algebra.
25
In a proof of this theorem, Shirshov (1953) used Lazard-Shirshov elimination: In
Lie(X), elements [xixn1 ] = [. . . [xix1] . . . ]x1], i > 1, n ≥ 0 are free generators of the
subalgebra generated by them.
Lazard-Shirshov elimination is used in GS bases theory for Lie algebras together with
Lyndon-Shirshov associative and Lie words.
We prove Theorem 2.5 in the following sections.
§2.3.1 Lazard-Shirshov elimination process
Let A be an index set, N set of natural number, Y = yα,n|α ∈ A, n ∈ N, and
Lie(Y ) the free Lie algebra generated by Y over a filed k. Let δ : Lie(Y ) → Lie(Y )
be a derivation such that yα,n 7→ yα,n+1, i.e., δ is a linear map such that for any f, g ∈Lie(Y ), δ(fg) = δ(f)g+ fδ(g). Let L1 be the direct sum of the linear spaces Lie(Y ) and
Fδ, denoted by Lie(Y ) ⊕ Fδ. In order to obtain a Lie structure on L1, we define a
multiplication [ ] on Lie(Y ) as usual, and for any u ∈ Lie(Y ), [u δ] = δ(u), [δ u] = −δ(u)and [δ δ] = 0. It is easy to prove that (L1, [ ]) is a Lie algebra.
Let X = xα| α ∈ A and I be the Lie ideal, generated by X, of the free Lie
algebra Lie(X ∪ y) over X ∪ y. Obviously I as a vector space is spanned by those
Lie monomials in X ∪y containing at least one xα, α ∈ A. The ideal I has codimension
1 in Lie(X ∪ y), and y is a linear basis of Lie(X ∪ y) modulo I.
Denote xnα (n ∈ N) the Lie product [[· · · [[xα y]y] · · · ]y︸ ︷︷ ︸
n
] in Lie(X ∪ y).
Lemma 2.6 As a Lie subalgebra of Lie(X∪y), the Lie ideal I is generated by xnα| α ∈
A, n ∈ N.
Proof. Let J be the Lie subalgebra, generated by xnα| α ∈ A, n ∈ N, of Lie(X ∪ y).
It is clear that I contains J . Let u be an arbitrary Lie monomial in X ∪ y containing
at least one xα, α ∈ A. It suffices to show that u belongs to J . Let us use induction
on the length of u, denoted by |u|. If |u| = 1, then u = xα and of course it belongs to
J . Suppose that u = [u1u2]. If u1 and u2 contains xα and xβ, respectively, then by the
induction hypothesis both u1 and u2 belong to J , and thus u ∈ J . Otherwise, we may
assume that u2 does not contain any xα, α ∈ A, because of the anti-commutativity law of
Lie algebras. By the induction hypothesis, u1 belongs to J and without loss of generality
let u1 = [xn1α1xn2
α2· · ·xnt
αt] be an arbitrary Lie monomial in xn
α| α ∈ A, n ∈ N. Therefore,
it follows from the Jacobi identity, that u = [u1u2] = [[xn1α1xn2
α2· · ·xnt
αt]y] belong to J . This
completes our proof.
26
Lemma 2.7 The Lie ideal I as a Lie subalgebra of Lie(X ∪ y) is freely generated by
xnα, α ∈ A, n ∈ N.
Proof. Let ϕ : Lie(Y ) → Lie(X ∪ y) be a Lie homomorphism such that yα,n 7→ xnα.
Let ϕ : L1 = Lie(Y ) ⊕ Fδ → Lie(X ∪ y) be a linear map such that δ 7→ y and
u 7→ ϕ(u) if u ∈ Lie(Y ). It is easy to check that ϕ preserves Lie multiplication and thus
it is a Lie homomorphism. Now, let us define a Lie homomorphism ψ : Lie(X∪y)→ L1
by letting xα 7→ yα,0 and y 7→ δ. Then ϕ is a Lie isomorphism since ϕψ = 1 and ψϕ = 1.
It together with lemma 2.6 follows that I ∼= Lie(Y ). This completes our proof.
§2.3.2 Constructing generators for an arbitrary subalgebra in a free Lie al-
gebra
Let H be a linear basis of monomials of a free Lie algebra Lie(X). Define d : X → Nby setting d(xα) = 1 for all α ∈ A and d([xi1 , . . . , xit ]) = t for [xi1 , . . . , xit ] ∈ H. For
any Lie polynomial f in Lie(X), it can be uniquely presented as f =∑αiui, α ∈ k
and ui ∈ H. Then the degree of f is defined by the highest degree of its monomials
ui, denoted by deg(f). Every nonzero Lie polynomial f may also be uniquely written as
f = f (1) + . . . + f (m) where f (i) is a summation of monomials in f with degree i. Then,
f (m) is called the highest part of f , denoted by f ′.
Let L be a Lie subalgebra of Lie(X), and Vi = f ∈ L| deg(f) ≤ i, and Bi be a
linear basis of Vi, satisfying Bi ⊆ Bi+1, for all i. Clearly, ∪∞i=1Vi = L and B := ∪∞
i=1Bi is
a linear basis of L.
Definition 2.8 The set M = bα of elements of a free Lie algebra Lie(X) is called irre-
ducible iff for every b ∈M , its highest part b′ does not belong to the subalgebra generated
by the highest parts b′α’s of the remaining elements bα ∈M , bα 6= b.
Lemma 2.9 There exists an irreducible generating set for the Lie subalgebra L of a free
Lie algebra Lie(X).
Proof. Let < be a well ordering on B such that
B1 < B2 \B1 < . . . < Bi+1 \Bi < . . . .
LetM = b ∈ B|b /∈ the subalgebra generated by all elements in B which are less than b.Claim 1: M is a generating set of L.
In order to prove that M generates L, it suffices to show that B is a subset of 〈M〉,the subalgebra of Lie(X) generated by M . Firstly, the smallest element of B is of course
27
in 〈M〉. Suppose that b ∈ B, and all elements in B less than b belong to 〈M〉. If b is in
M , then b surely belongs to 〈M〉. If b is not in M , then, by the definition of M and the
induction hypothesis, b belongs to 〈M〉.Claim 2: M is an irreducible set.
Assume that there is an element b ∈ M such that b′ belongs to the subalgebra
generated by the highest parts b′α’s of the remaining elements bα ∈ M , bα 6= b. Then b
can be presented by
b′ =∑
αib′i +∑
βj1...jnj[b′j1 . . . b
′jnj
] (2.1)
where bi, bjnk∈ M\b. The summations on the right-hand side of equation (2.1) may
contain factors of degree greater than the degree of b′. Since rewriting the product of two
Hall words, they will either become zero or will keep the the same degree and the same
content. In view of the linear independence of Hall words, all such factors must cancel
each other and hence we may assume the first summation contains only the elements of
the same degree as b and the remaining elements appearing on the right-hand side of
equation (2.1) have degree strictly less than the degree of b. Without loss of generality,
we may also assume that b > bi, for all i. From the above equation, it follows that the
element
b−∑
αibi +∑
βj1...jnj[bj1 . . . bjnj
]
belongs to Vd(b)−1, which contradicts b ∈M . This completes our proof.
§2.3.3 A proof of Shirshov-Witt Theorem
Definition 2.10 A finite subset T = t1, . . . , tn of a free Lie algebra Lie(X) is said to
be algebraically independent if there does not exist a nonzero Lie polynomial F (y1, . . . , yn)
in a free Lie algebra Lie(y1, . . . , yn) such that F (t1, . . . , tn) = 0 in Lie(X). An infinite
subset S of Lie(X) is said to be algebraically independent if its arbitrary finite subset is
algebraically independent.
It’s well known that a set S of Lie(X) is freely generated a Lie subalgebra iff S is alge-
braically independent.
Claim: every irreducible subset of a free Lie algebra Lie(X) is algebraically indepen-
dent.
Let us assume the contrary. There exists a finite irreducible subset S = b1, b2, . . . , bnof Lie(X) which is algebraically dependent. Then we have the following lemma.
Lemma 2.11 The set S ′ = b′1, b′2, . . . , b′n of highest parts of elements in S is an irre-
ducible and algebraically dependent set.
28
Proof. By the assumption that S is algebraically dependent, there exists a nonzero Lie
polynomial F (y1, . . . , yn) in the free Lie algebra Lie(y1, . . . , yn) such that F (b1, . . . , bn) =
0. Setting d : Y → N, yi 7→ deg(bi). The degree of an Hall word u = [yi1yi2 · · · yit ] in
y1, . . . , yn is defined by d(u) =∑
j d(yij). Then F (y1, . . . , yn) can be presented as
F (y1, . . . , yn) = F1(y1, . . . , yn) + F2(y1, . . . , yn)
where F1(y1, . . . , yn) is a summation of all Lie monomials in F (y1, . . . , yn) with highest
degree. Clearly, F1(y1, . . . , yn) 6= 0. Then
F (b1, . . . , bn) = F1(b1, . . . , bn) + F2(b1, . . . , bn)
= F1(b′1 + r1, . . . , b
′n + rn) + F2(b1, . . . , bn)
= F1(b′1, . . . , b
′n) +G(b1, . . . , bn, r1, . . . , rn)
= 0
where G(b1, . . . , bn, r1, . . . , rn) is a Lie polynomial on b1, . . . , bn, r1, . . . , rn. It follows
that F1(b′1, . . . , b
′n) = 0. Therefore S is algebraically dependent. S ′ is irreducible since S
is irreducible. This completes our proof.
Lemma 2.11 allows us to assume that b1, . . . , bn are homogeneous. We may also
assume that deg(b1) ≤ deg(b2) ≤ . . . ≤ deg(bn).
Lemma 2.12 Suppose that c1 = b1, c2 = b2+v2, . . . , cn = bn+vn, where vi ∈ 〈b1, . . . , bi−1〉and each vi is homogeneous with degree deg(bi). Then c1, c2, . . . , cn is an irreducible and
algebraically dependent set of homogeneous elements.
Proof. Clearly, each cj is homogeneous. Let us assume that c1, c2, . . . , cn is reducible.
Like the argument in the proof of Lemma 2.9, ci can be presented as
ci =∑
αjcj +∑
βk1...ktk[ck1 . . . cktk
]
where ci 6= cj, ckl, l = 1, 2 . . . , tk for all j, k. Without loss of generality, we may assume
that i > j for all j. Then, replacing ct by bt + vt(v1 = 0), t = 1, . . . , n, we have
bi + vi =∑
αj(bj + vj) +∑
βk1...ktk[(bk1 + vk1) · · · (bktk
+ vktk)].
Since vj ∈ 〈b1, . . . , bj−1〉 and deg(b1) ≤ deg(b2) ≤ . . . ≤ deg(bn), we have
bi =∑
αsbs +∑
βt1...tlt[bt1 · · · btlt ],
where bi 6= bs, btj , j = 1, 2 . . . , lt for all s, t. This contradicts that b1, b2, . . . , bn is irre-
ducible.
29
We claim that vi ∈ 〈c1, . . . , ci−1〉, i = 2, . . . , n.
Induction on i. If i = 2, then v2 ∈ 〈c1〉 = 〈b1〉. For an arbitrary vi, we have
vi ∈ 〈b1, . . . , bi−1〉 = 〈c1, c2 − v2, . . . , ci−1 − vi−1〉. By the induction hypothesis, vi ∈〈c1, . . . , ci−1〉.
For an arbitrary Hall word u = [yi1yi2 . . . , yit ] in y1, . . . , yn, we denote nyi(u) the
number of the occurrences of yi in u and we let wt(u) = (nyn(u), . . . , ny1(u)). We write
F1(y1, . . . , yn) as
F1(y1, . . . , yn) = F11(y1, . . . , yn) + F12(y1, . . . , yn)
where F11(y1, . . . , yn) is a summation of all Lie monomials in F (y1, . . . , yn) with highest
weight.
Then
F (b1, . . . , bn) = F11(c1, c2 − v2, . . . , cn − vn) + F12(c1, c2 − v2, . . . , cn − vn)
= F11(c1, . . . , cn) +G1(c1, . . . , cn, v1, . . . , vn)
= F11(c1, . . . , cn) +G2(c1, . . . , cn), (by vi ∈ 〈c1, . . . , ci−1〉),
= 0
where G1 and G2 are Lie polynomials in c1, . . . , cn, v1, . . . , vn.Define F2(y1, . . . , yn) by replacing ci with yi in F11(c1, . . . , cn) +G2(c1, . . . , cn). From
the definition of F11(y1, . . . , yn) and the above discussion, it follows that F2(y1, . . . , yn) 6=0 but F2(c1, . . . , cn) = 0. Therefore c1, c2, . . . , cn is algebraically dependent. This
completes our proof.
Lemma 2.13 There exists an irreducible and algebraically dependent subset U = u1, . . . , unof Lie(X) where each Lie polynomial ui is homogeneous in each letter xα, α ∈ A, and
deg(ui) ≤ deg(bi), i = 1, . . . , n.
Proof. Choose an arbitrary generator xα from among the elements b1, . . . , bn. Then
each bi can be written as
bi = bi0 + bi1 + . . .+ biti
where bij is a Lie polynomial, homogeneous in xα and has xα degree j. The assumption
b1, . . . , bn algebraically dependent may lead to b1t1, . . . , bntn
also algebraically dependent
like the proof of Lemma 2.11 (just by setting d(yi) := the xα degree of bi).
If there exists some i such that biti ∈ 〈b1t1, . . . , bi−1ti−1
〉, then there is a Lie polynomial
h(y1, . . . , yi−1) such that biti = h(b1t1, . . . , bi−1ti−1
) and then set bi := bi − h(b1, . . . , bi−1).
30
By Lemma 2.12, b1, b2, . . . , bn is irreducible and algebraically dependent. After re-
peat this kind of process finite times, we may reach the following situation that biti /∈〈b1t1
, . . . , bi−1ti−1〉 for all i even more biti /∈ 〈b1t1
, . . . , bi−1ti−1, bi+1ti+1
, . . . , bntn〉 for all i.
Let v1 = b1t1, . . . , vn = bntn
. Then the set v1, . . . , vn is irreducible and algebraically
dependent and each vi is homogeneous in xα. Enumerating one by one all generators that
occur in the elements b1, . . . , bn, we finally obtain the desired set U = u1, . . . , un.This completes our proof.
Lemma 2.13 allows us to assume that S = b1, b2, . . . , bn is a finite irreducible and
algebraically dependent subset of Lie(X) and each bi is homogeneous in each generator
xα, α ∈ A. We may also assume that deg(b1) ≤ deg(b2) ≤ . . . ≤ deg(bn). In the follow
argument, we show that this assumption will lead to a contradiction. The ordered n-
tuple (λ1; . . . ;λn) where λi is the degree of bi, will be called the height the set S =
b1, b2, . . . , bn. Let M be the set of all possible heights of S = b1, b2, . . . , bn and we
order it lexicographically. If (λ1; . . . ;λn) = (1; . . . ; 1), then b1, b2, . . . , bn are free generators
and thus b1, b2, . . . , bn is algebraically independent which contradicts the assumption in
the very beginning of this section. Suppose that (λ1; . . . ;λn) > (1; . . . ; 1), then there exists
a generator xβ such that bi 6= xβ for all i, and xβ occurs in bj for some j. Then all elements
bi, i = 1, . . . , n belong to the ideal of Lie(X) generated by X\xβ. By Lemma 2.7, bi
can be regarded as Lie monomials in new free generator xnα = [[· · · [[xα xβ]xβ] · · · ]xβ︸ ︷︷ ︸
n
], α ∈
A, n ∈ N. Now in this free subalgebra of Lie(X), we define the degree function by
deg(xnα) = 1. Since each bi is homogeneous in each generator including xβ, Lie polynomials
bi are homogeneous in new generators xnα and hence the set S = b1, b2, . . . , bn is still
irreducible and algebraically dependent. For each i, the degree of bi in new generators
is less or equal to its degree in original generators xα, α ∈ A and there exists at least
one j such that the degree of bj in new generators is strictly smaller than its degree in
xα, α ∈ A. By Lemma 2.13, we may get an irreducible and algebraically dependent subset
U = u1, . . . , un of L(xnα|α ∈ A, n ∈ N) where each Lie polynomial ui is homogeneous in
each letter xnα, α ∈ A, n ∈ N and the degree of ui in xn
α, α ∈ A, n ∈ N is less or equal to
the degree bi in xnα, α ∈ A, n ∈ N, i = 1, . . . , n. By the induction hypothesis on the set S,
we may get a contradiction to the assumption that there exists a finite irreducible subset
of Lie(X) which is algebraically dependent.
The proof of Theorem 2.5. By Lemma 2.9, L always has an irreducible generating set
M . From the above argument, we know that M is algebraically independent. Therefore
L is freely generated by M . This completes our proof of Theorem 2.5.
31
Chapter 3 Grobner Bases for Commutative
Algebras
§3.1 Buchberger theorem
Let k[X] be a polynomial algebra on X = x1, x2, . . . , xn and a field k, x1 > x2 >
· · · > xn.
Remark 2.1 We choose ordering x1 > x2 > . . . following ordering of variables in the
Gauss elimination algorithm and a tradition of Grobner bases theory for commutative
polynomials. For noncommutative and nonassociative variables we will choose x1 < x2 <
. . . .
Let us define the deg-lex ordering on the set of commutative words [X] to compare
two words first by degree and then lexicographically.
For example, x31 > x2
1x2 > x1x2x3 > x21. In general,
xi11 x
i22 . . . x
inn > xj1
1 xj22 . . . x
jnn iff
∑ik >
∑jk, or∑
ik =∑
jk and (i1, i2, . . . , in) > (j1, j2, . . . , jn) lexicographically.
This ordering is a monomial ordering in the sense
(i) If u > v, then wu > wv for any w ∈ [X],
(ii) [X] is a well-ordered set, i.e., a linearly ordered set with minimal (descending
chain) condition – any chain u1 > u2 > · · · > ur > . . . is finite.
Another example of monomial ordering on [X] is lex-deg ordering to compare two
commutative words lexicographically with a condition that a proper beginning of a word
is less than the word.
Let us fix a monomial ordering on [X]. A polynomial f ∈ k[X] has a maximal
(leading) word f with a leading coefficient αf6= 0,
f = αff +
∑ui<f
αiui, ui ∈ [X], αf, αi ∈ k.
If αf
= 1, f is called a monic polynomial.
It is clear that fg = f g.
Let us define a least common multiple of two words as usual:
lcm(xi11 · · ·xin
n , xj11 · · ·xjn
n ) = xmaxi1,j11 · · ·xmaxin,jn
n .
Then
lcm(u, v) = au = bv, a, b ∈ [X].
32
If lcm(u, v) = uv then it is called a trivial lcm, otherwise a non-trivial lcm. If u = bv
then lcm(u, v) = u is called an inclusion lcm.
Definition 3.1 Let f, g be monic polynomials and w = lcm(f , g) = af = bg. Then
polynomial
(f, g)w = af − bg
is called an s-polynomial (a composition) of f and g.
It is clear that (f, g)w ∈ Id(f, g) and (f, g)w < w. If w is a trivial lcm, then
(f, g)w = gf − fg = (g − g)f − (f − f)g, and the composition is trivial mod(f, g;w) in a
sense of the following definition.
Definition 3.2 Let S ⊂ k[X] be a monic set of polynomials. S is called a Grobner
basis if any composition (f, g)w of polynomials from S is trivial mod(S,w), denoted by
(f, g)w ≡ 0 mod(S,w), i.e.,
(f, g)w =∑
αiaisi, si ∈ S, ai ∈ [X], αi ∈ k, aisi < w for all i.
From the above, it is equivalent to that (f, g)w ≡ 0 mod(S,w) for a non-trivial
w = lcm(f , g).
It is clear the case if (f, g)w goes to zero using the Elimination of Leading Words
(ELW’s) of S (in the leading monomials of polynomials), i.e., transformations h→ h−αas,where s ∈ S and αas is the leading term (a monomial with a coefficient) of h in the
standard presentation of h. Then h− αas < h.
Theorem 3.3 (Buchberger Theorem) Let S ⊂ k[X] be a monic set and < a monomial
ordering on [X]. Then the following conditions are equivalent.
(i) S is a Grobner basis.
(ii) If f ∈ Id(S) then f = as, s ∈ S, a ∈ [X].
(iii) Irr(S) = u ∈ [X]|u 6= as, s ∈ S, a ∈ [X] is a k-linear basis of k[X|S], the commu-
tative algebra with generators X and defining relations S.
Proof. (i)⇒(ii). Let f =∑m
i=1 αiaisi, si ∈ S, ai ∈ [X], αi ∈ k. We may assume
that all similar terms (si = sj, ai = aj) are reduced. Let us choose the leading words
wi = aisi. Let
w1 = w2 = · · · = wl > wl+1 ≥ · · · ≥ wm, l ≥ 1.
33
We will use induction on (w1, l), w1 ≥ f , l ≥ 1 with the lex-ordering of the pairs. If w1 = f
or l = 1, the claim is clear. Otherwise l > 1 and w1 = a1s1 = a2s2. Since w1 is a common
multiple of s1, s2, it must contain w = lcm(s1, s2) as a subword, w1 = blcm(s1, s2) such
that a1s1 = bw|s1→s1 , a2s2 = bw|s2→s2 . Then
a1s1 − a2s2 = b(s1, s2)w ≡ 0 mod(S,w1).
It means that we may rewrite the initial equality for f in a way
f = (α1 + α2)a2s2 + · · ·+ αlalsl + . . . .
If l > 2 or α1 + α2 6= 0, we may apply the induction (since we decrease l). Otherwise we
decrease w1.
(ii)⇒(iii). For any S, the ELW’s of S in all monomials of a polynomial gives rise a
following presentation of an arbitrary polynomial h
h =∑
αiui +∑
βjajsj, (3.1)
where each αi, βj ∈ k, ui ∈ Irr(S), sj ∈ S, ajsj ≤ h. It means that Irr(S) is a set of
linear generators of k[X|S].
If a linear combination of S-irreducible words ui’s belongs to Id(S) then by (ii) some
of ui contains s for some s ∈ S, a contradiction.
(iii)⇒(i). If h = (f, g)w as above then all αi = 0 and (f, g)w ≡ 0 mod(S,w).
Corollary 3.4 A monic set S ⊆ k[X] is a Grobner basis if and only if any composition
of polynomial from S goes to zero by ELW’s of S.
Proof. (⇐) It was proved above. (⇒) Let h = (f, g)w and f, g ∈ S. By ELW’s of S
one can present h as before. By Theorem 3.3 all αi’s are zero. It means that h goes to
zero by ELW’s of S.
§3.2 Buchberger algorithm
Let X = x1, . . . , xn, S = s1, . . . , sm ⊂ k[X] and I = Id(S) the ideal of k[X]
generated by S. If each composition of elements from S is trivial mod(S,w), then S is a
Grobner basis. Otherwise let us join to S a nontrivial composition h = (f, g)w, f, g ∈ S.
Actually we present first h in a form (3.1) and to join to S an S-irreducible part of
h, namely h′ =∑αiui, ui ∈ Irr(S). Let S ⊂ S ′ = S ∪ h′. It is easy to see that
34
Id(s1, . . . , sm) is a proper subset of Id(s1, . . . , sm, s′), since h′ 6∈ Id(s1, . . . , sm). Then we
continue the process and would have a serious of generator sets of Id(S),
S ⊂ S ′ ⊂ S ′′ ⊂ . . . (3.2)
such that the ideals generated by the leading monomials of S, S ′, S ′′, . . . constitute an
ascending chain of ideals of k[X]. By Hilbert basis theorem the chain must be finite, so
is (3.2).
Definition 3.5 A Grobner basis S in k[X] is called reduced if for any s ∈ S, supp(s) ⊆Irr(S\s), where supp(s) = u1, u2, . . . , um if s =
∑mi=1 αiui each 0 6= αi ∈ k, ui ∈ [X].
A Grobner basis S in k[X] minimal, if there are no compositions of inclusion of
polynomials from S, i.e., f 6= ag, a ∈ [X], for any different f, g ∈ S.
Clearly, a reduced Grobner basis is minimal.
Corollary 3.6 For any ideal I ⊂ k[X] and any monomial ordering < on [X] there exists
only one reduced Grobner basis of I.
Proof. Clearly, there exists a minimal Grobner-Shirshov basis S ⊂ k[X] for the ideal
I = Id(S).
For any s ∈ S, by using (3.1), we have s = s′+s′′, where supp(s′) ⊆ Irr(S\s), s′′ ∈Id(S \ s). Since S is a minimal Grobner-Shirshov basis, we have s = s′ for any s ∈ S.
Let S ′ = s′|s ∈ S. It is clear that S ′ ⊆ Id(S) = I. For any f ∈ Id(S), by Theorem
3.3, f = as = as′ for some a ∈ [X], s ∈ S. From this and Theorem 3.3 again, it follows
that S ′ = s′|s ∈ S is a reduced Grobner-Shirshov basis for the ideal I.
Suppose that S, R are two reduced Grobner-Shirshov bases for the ideal I. For any
s ∈ S, by Theorem 3.3, s = ar, r = bs1 for some a, b ∈ [X], r ∈ R, s1 ∈ S and hence
s = abs1. Since s ∈ supp(s) ⊆ Irr(S \ s), we have s = s1. It follows that a = b = 1
and so s = r.
If s 6= r then 0 6= s− r ∈ I = Id(S) = Id(R). By Theorem 3.3, s− r = a1r1 = b1s2
for some a1, b1 ∈ [X] with r1, s2 < s = r. This means that s2 ∈ S \ s and r1 ∈ R \ r.Noting that s− r ∈ supp(s)∪ supp(r), we have either s− r ∈ supp(s) or s− r ∈ supp(r).If s− r ∈ supp(s) then s− r ∈ Irr(S \ s) which contradicts s− r = b1s2; if s− r ∈supp(r) then s− r ∈ Irr(R \ r) which contradicts s− r = a1r1. This shows that s = r
and then S ⊆ R. Similarly, R ⊆ S.
35
Corollary 3.7 For any ideal I = Id(S) ⊂ k[X] the membership problem
for any f(X) ∈ k[X] to answer whether f(X) ∈ Id(S) or not
is algorithmically solvable.
Proof. We may assume that S is a finite Grobner basis of I. Induction on u = f ≥ 1.
If u = 1, it is clear. Otherwise if u does not contain s, s ∈ S, the answer is negative. If
u = as, s ∈ S, a ∈ [X] then f1 = f − αas has less leading monomial and is belonged to
I. By induction we are done.
Corollary 3.8 Gauss elimination algorithm for solving systems of linear equations is a
particular case of Buchbereger algorithm.
Proof. Let
Ax = b, A ∈Matm,n(k), x = (x1, . . . , xn)Tr, b = (b1, . . . , bm)Tr
be a system of linear equations over k. Let fi = aix − bi, 1 ≤ i ≤ m be the correspond-
ing system of linear polynomials. Then Buchberger algorithm for fi gives the Gauss
elimination algorithm for the system, and visa versa.
Theorem 3.9 By Grobner bases of ideals I, J in k[X] one may algorithmically find
Grobner bases of I + J, I ∩ J , and rad(I) := f ∈ k[X]|fm ∈ I for some m ≥ 1.
36
Chapter 4 Grobner-Shirshov Bases for Associative
Algebras
In this chapter, we give a proof of Shirshov CD-lemma for associative algebras.
§4.1 Composition-Diamond lemma for associative algebras
Let k be a field, k〈X〉 the free associative algebra over k generated by X and X∗
the free monoid generated by X, where the empty word is the identity which is denoted
by 1. For a word w ∈ X∗, we denote the length (degree) of w by |w| (deg(w)). Let X∗
be a well-ordered set. Let f ∈ k〈X〉 with the leading word f and f = αf − rf where
0 6= α ∈ k. We call f monic if α = 1.
A well ordering > on X∗ is called monomial if it is compatible with the multiplication
of words, that is, for u, v ∈ X∗, we have
u > v ⇒ w1uw2 > w1vw2, for all w1, w2 ∈ X∗.
A standard example of monomial ordering on X∗ is the deg-lex ordering to compare two
words first by degree and then lexicographically, where X is a well-ordered set.
Let f and g be two monic polynomials in k〈X〉 and < a monomial ordering on X∗.
Then, there are two kinds of compositions:
(i) If w is a word such that w = f b = ag for some a, b ∈ X∗ with |f | + |g| > |w|,then the polynomial (f, g)w = fb − ag is called the intersection composition of f and g
with respect to w.
(ii) If w = f = agb for some a, b ∈ X∗, then the polynomial (f, g)w = f − agb is
called the inclusion composition of f and g with respect to w.
In the composition (f, g)w, w is called an ambiguity, or a lcm(f , g).
Let S ⊂ k〈X〉 such that every s ∈ S is monic. Let h ∈ k〈X〉 and w ∈ X∗. Then h is
called trivial modulo (S,w), denoted by h ≡ 0 mod(S,w), if h =∑αiaisibi, where each
αi ∈ k, ai, bi ∈ X∗, si ∈ S and aisibi < w.
Elements asb, a, b ∈ X∗, s ∈ S are called S-words.
Definition 4.1 A monic set S ⊂ k〈X〉 is called a GS basis in k〈X〉 with respect to the
monomial ordering < if any composition of polynomials in S is trivial modulo S and
corresponding w.
37
A set S is called a minimal GS basis in k〈X〉 if S is a GS basis in k〈X〉 such that
there is no inclusion compositions in S, i.e., for any f, g ∈ S with f 6= g, one has f 6= agb
for any a, b ∈ X∗.
Denote
Irr(S) = u ∈ X∗|u 6= asb, s ∈ S, a, b ∈ X∗.
A GS basis S in k〈X〉 is called reduced if for any s ∈ S, supp(s) ⊆ Irr(S \ s),where supp(s) = u1, u2, . . . , un if s =
∑ni=1 αiui each 0 6= αi ∈ k, ui ∈ X∗.
The following lemma plays a key role for proving CD-lemma for associative algebras.
Lemma 4.2 If S is a GS basis in k〈X〉 and w = a1s1b1 = a2s2b2, where a1, b1, a2, b2 ∈X∗, s1, s2 ∈ S, then a1s1b1 ≡ a2s2b2 mod(S,w).
Proof. There are three cases to consider.
Case 1. Suppose that subwords s1 and s2 of w are disjoint, say, |a2| ≥ |a1| + |s1|.Then, a2 = a1s1c and b1 = cs2b2 for some c ∈ X∗, and so, w1 = a1s1cs2b2. Now,
a1s1b1 − a2s2b2 = a1s1cs2b2 − a1s1cs2b2
= a1s1c(s2 − s2)b2 + a1(s1 − s1)cs2b2.
Since s2 − s2 < s2 and s1 − s1 < s1, we conclude that
a1s1b1 − a2s2b2 =∑
i
αiuis1vi +∑
j
βjujs2vj
for some αi, βj ∈ k, S-words uis1vi and ujs2vj such that uis1vi, uj s2vj < w.
Case 2. Suppose that the subword s1 of w contains s2 as a subword. Then s1 =
as2b, a2 = a1a and b2 = bb1, that is, w = a1as2bb1 for some S-word as2b. We have
a1s1b1 − a2s2b2 = a1s1b1 − a1as2bb1 = a1s1 − as2bb1 = a1(s1, s2)s1b1.
By the triviality of compositions it follows that a1s1b1 ≡ a2s2b2 mod(S,w).
Case 3. s1 and s2 have a nonempty intersection as a subword of w. We may assume
that a2 = a1a, b1 = bb2, w = s1b = as2, |w| < |s1|+ |s2|. Then, similar to the Case 2, we
have a1s1b1 ≡ a2s2b2 mod(S,w).
Remark: A short proof of Lemma 4.2 uses a notation of lcm(u, v) of two words:
lcm(u, v) =
ub = av, a, b ∈ X∗, deg(ub) < deg(u) + deg(v) (an intersection lcm)
avb, a, b ∈ X∗ (an inclusion lcm)
ucv, c ∈ X∗ (a trivial lcm)
38
Note that lcm(u, v) 6= lcm(v, u) in general. Then a general composition (f, g)w, w =
lcm(f , g) of two monic polynomials f, g is defined as follows
(f, g)w = lcm(f , g)|f 7→f − lcm(f , g)|g 7→g.
If w is a trivial lcm, then (f, g)w ≡ 0 mod(f, g;w).
Now, in (i) ⇒ (ii), if w = a1s1b1 = a2s2b2, then up to order of s1, s2, we have
w1 = cwb, c, b ∈ X∗, w = lcm(s1, s2) and a1s1b1 = ws1 7→s1 , a2s2b2 = ws2 7→s2 .
As the result,
a1s1b1 − a2s2b2 = c(s1, s2)wb ≡ 0 mod(S, cwb = w1).
The result (i)⇒ (ii) follows by induction on (w1, l).
Lemma 4.3 Let S ⊂ k〈X〉 be a subset of monic polynomials. Then for any f ∈ k〈X〉,
f =∑ui≤f
αiui +∑
ajsjbj≤f
βjajsjbj
where each αi, βj ∈ k, ui ∈ Irr(S) and ajsjbj an S-word. So, Irr(S) is a set of linear
generators of the algebra f ∈ k〈X|S〉.
Proof. The result follows by induction on f .
Theorem 4.4 (CD-lemma for associative algebras) Let < be a monomial ordering on X∗,
S ⊂ k〈X〉 a monic set and Id(S) the ideal of k〈X〉 generated by S. Then the following
statements are equivalent.
(i) S is a Grobner-Shirshov basis in k〈X〉.
(ii) f ∈ Id(S)⇒ f = asb for some s ∈ S and a, b ∈ X∗.
(iii) Irr(S) = u ∈ X∗|u 6= asb, s ∈ S, a, b ∈ X∗ is a linear basis of the algebra k〈X|S〉.
Proof. (i) ⇒ (ii). Let S be a GS basis and 0 6= f ∈ Id(S). Then, we have
f =∑n
i=1 αiaisibi where each αi ∈ k, ai, bi ∈ X∗, si ∈ S. Let
wi = aisibi, w1 = w2 = · · · = wl > wl+1 ≥ · · · .
We will use the induction on (w1, l) to prove that f = asb for some s ∈ S and a, b ∈ X∗.
If l = 1, then f = a1s1b1 = a1s1b1 and hence the result holds. Assume that l ≥2. Then, we have w1 = a1s1b1 = a2s2b2. It follows from Lemma 4.2 that a1s1b1 ≡
39
a2s2b2 mod(S,w1). Now the result follows from the induction whenever α1 + α2 6= 0, or
l > 2, or both α1 + α2 = 0 and l = 2. This shows (ii).
(ii) ⇒ (iii). By Lemma 4.3, Irr(S) generates k〈X|S〉 as a linear space. Suppose
that∑i
αiui = 0 in k〈X|S〉, where 0 6= αi ∈ k, ui ∈ Irr(S). It means that∑i
αiui ∈ Id(S)
in k〈X〉. Then∑i
αiui = uj ∈ Irr(S) for some j which contradicts (ii).
(iii) ⇒ (i). For any f, g ∈ S , by using Lemma 4.3 and (iii), we have (f, g)w ≡0 mod(S,w). Therefore, S is a GS basis.
Shirshov algorithm. If a monic subset S ⊂ k〈X〉 is not a GS basis then one can add to
S all nontrivial compositions making them monic. Continuing this process repeatedly, we
finally obtain a GS basis Sc that contains S and generates the same ideal, Id(Sc) = Id(S).
Sc is called GS completion of S. Using the reduction algorithm (elimination of leading
words of polynomials), we may get a minimal GS basis Sc.
Let K be a commutative ring with identity 1.
Remark: Both associative and Lie (Theorem 5.39) CD-lemmas are valid if we replace
the base field k by an arbitrary commutative ring K with identity since we assume that
all GS bases consist of monic polynomials. For example, let (L, [ ]) be a Lie algebra over
K and L a free K-module with a well-ordered K-basis ai|i ∈ I. Then with the deg-lex
ordering on ai|i ∈ I∗, its universal enveloping associative algebra UK(L) has usual GS
basis aiaj − ajai =∑αt
ijat|i > j, i, j ∈ I, where αtij ∈ K and [ai, aj] =
∑αt
ijat in L,
and by CD-lemma for associative algebras over K, L ⊂ UK(L) and ai1 · · · ain|i1 ≤ · · · ≤in, n ≥ 0, i1, . . . , in ∈ I is a K-basis of UK(L), see Theorem 4.7.
In fact, by the same reason, all CD-lemmas in this book are valid if we replace the
base field k by an arbitrary commutative ring K with identity. If this is the case, then the
(iii) in a CD-lemma should be: K(X|S) is a free K-module with a K-basis Irr(S). But
in general case, Shirshov algorithm does not work: if S is a monic set, then S ′, a set of
adding all non-trivial compositions to S, is not a monic set in general and the algorithm
may stop without any result.
The following theorem gives a linear basis of the ideal Id(S) if S ⊂ k〈X〉 is a GS
basis.
Theorem 4.5 Suppose that S ⊂ k〈X〉 is a Grobner-Shirshov basis. Then for any u ∈X∗ \ Irr(S), by Lemma 4.3, there exists u ∈ kIrr(S) with u < u (if u 6= 0) such that
40
u− u ∈ Id(S) and the set u− u|u ∈ X∗ \ Irr(S) is a linear basis of the ideal Id(S) of
k〈X〉.
Proof. Suppose that 0 6= f ∈ Id(S). Then by CD-lemma for associative algebras,
f = a1s1b1 = u1 for some s1 ∈ S and a1, b1 ∈ X∗ which implies that f = u1 ∈ X∗\Irr(S).
Let f1 = f−α1(u1−u1), where α1 is the coefficient of the leading term of f and u1 < u1 or
u1 = 0. Then f1 ∈ Id(S) and f1 < f . By induction on f , the set u− u|u ∈ X∗ \ Irr(S)generates Id(S) as a linear space. It is clear that the set u− u|u ∈ X∗ \ Irr(S) is linear
independent.
By a similar proof of Corollary 3.6, we have the following theorem.
Theorem 4.6 Let I be an ideal of k〈X〉 and > a monomial ordering on X∗. Then there
exists uniquely the reduced Grobner-Shirshov basis S for I.
§4.1.1 PBW-theorem for Lie algebras
Let (L, [ ]) be a Lie algebra over a field k with a well-ordered linear basis X = xi|i ∈I. Let S = [xixj] = xi, xj|i > j, i, j ∈ I be the multiplication table of L, where for
any i, j ∈ I, xi, xj = Σtαtijxt, α
tij ∈ k. Then U(L) = k〈X|S(−)〉 is called the universal
enveloping associative algebra of L, where S(−) = xixj − xjxi = xi, xj|i > j, i, j ∈ I.
Theorem 4.7 (PBW Theorem) Let the notation be as above. Then with the deg-lex order-
ing on X∗, S(−) is a Grobner-Shirshov basis in k〈X〉. Then by CD-lemma for associative
algebras, the set Irr(S(−)) consists of the elements
xi1 . . . xin , i1 ≤ · · · ≤ in, i1, . . . , in ∈ I, n ≥ 0
which forms a linear basis of U(L).
The result is valid for a Lie algebra that is a free K-module over a commutative ring
K (see Remark after Theorem 4.4).
Theorem 4.8 (PBW Theorem in a Shirshov form) Let L = Lie(X|S), S ⊂ Lie(X) ⊂k〈X〉, S a Grobner-Shirshov basis in Lie(X) and U(L) = k〈X|S〉. Then S is a GS basis
in k〈X〉 and a linear basis of U(L) consists of words u = u1 · · ·un, where u1 · · · un
in lex-ordering, n ≥ 0, each ui is S-irreducible associative Lyndon-Shirshov word.
41
§4.1.2 Normal forms for groups and semigroups
Let X be a set, S ⊆ X∗ ×X∗, ρ(S) the congruence on X∗ generated by S,
A = sgp〈X|S〉 = X∗/ρ(S)
the quotient semigroup and k(X∗/ρ(S)) the semigroup algebra. We identify the set u =
v|(u, v) ∈ S with S. Then it is easy to see that
σ : k〈X|S〉 → k(X∗/ρ(S)),∑
αi(ui + Id(S)) 7→∑
αi(uiρ(S))
is an algebra isomorphism.
Shirshov completion Sc of S consists of “semigroup relations”, Sc = ui− vi, i ∈ I.Then Irr(Sc) is a linear basis of k〈X|S〉 and so σ(Irr(Sc)) is a linear basis of k(X∗/ρ(S)).
This shows that Irr(Sc) is exactly normal forms of elements of the semigroup sgp〈X|S〉.Thus, in order to find normal forms of the semigroup sgp〈X|S〉, it suffices to find a
GS basis Sc in k〈X|S〉. In particular, let G = gp〈X|S〉 be a group, where S = (ui, vi) ∈F (X)× F (X)|i ∈ I and F (X) is the free group on a set X. Then G has a presentation
as a semigroup
G = sgp〈X ∪X−1|S, xεx−ε = 1, ε = ±1, x ∈ X〉, X ∩X−1 = ∅.
It is clear that in a group G = gp〈X|S〉, if for any x ∈ X there exists an n(x) ∈ Nsuch that the relation xn(x) = 1 is in S, then
G = gp〈X|S〉 = sgp〈X|S〉.
We give some examples.
1. Symmetric group Sn and alternating group An
Let Sn be the symmetric group of the permutations of 1, 2, · · · , n. Then the subset
An of all even permutations in Sn is a normal subgroup of Sn. One calls An the alternating
group of degree n. The following presentations of Sn and An are given in the monograph
of N. Jacobson (see [115], p71):
Sn = gp〈x1, . . . , xn−1 | x2i = 1, xjxi = xixj (j − 1 > i, 1 ≤ i, j ≤ n),
xi+1xixi+1 = xixi+1xi (1 ≤ i ≤ n− 2)〉.
under the isomorphism Sn → Sn, xi 7→ (i, i + 1), i = 1, 2, · · · , n − 1, where (ij) is the
transposition.
42
An = gp〈xi (1 ≤ i ≤ n− 2) ; x31 = 1, (xi−1xi)
3 = x2i = 1 (2 ≤ i ≤ n− 2),
(xixj)2 = 1 (1 ≤ i < j − 1, j ≤ n− 2)〉
under the isomorphism An → An, xi 7→ (12)((i+ 1)(i+ 2)), i = 1, 2, · · · , n− 2.
We will give proofs of these statements.
In Sn, let us define words xij = xixi−1 · · ·xj, where i > j; and xii = 1, xi,i+1 = 1.
Lemma 4.9 Let S be the set of defining relations in Sn together with xi+1jxi+1 =
xixi+1j|1 ≤ j < i, 1 ≤ i, j ≤ n. Then S is a Grobner-Shirshov basis of Sn in deg-
lex ordering.
Theorem 4.10 Let the notations be as above. Then
(i) the set N = x1i1x2i2 · · ·x(n−1)in−1|1 ≤ ij ≤ j + 1, j = 1, 2, · · · , n − 1 is a normal
form for Sn in generators xi = (i, i + 1), i = 1, 2, · · · , n − 1 relative to the deg-lex
ordering;
(ii) the map φ : xi 7→ (i, i+1), i = 1, 2, · · · , n− 1 induces an isomorphism of Sn and Sn.
Proof. By Lemma 4.9 and Theorem 4.4,
Irr(S) = x1i1x2i2 · · ·x(n−1)in−1| 1 ≤ ij ≤ j + 1, j = 1, 2, · · · , n− 1
which gives |Sn| = n!. On the other hand, φ induces a homomorphism φ : Sn → Sn and
|Sn| = n!. So, φ is an isomorphism.
We now give a presentation of the group An as a semigroup:
An = sgp〈x−11 , xi (1 ≤ i ≤ n− 2); x1x
−11 = x−1
1 x1 = x31 = 1, (xi−1xi)
3 = x2i = 1
(2 ≤ i ≤ n− 2), (xixj)2 = 1 (1 ≤ i < j − 1, j ≤ n− 2)〉
Let us order the generators in the following way:
x−11 < x1 < x2 < · · · < xn−2
Let X = x−11 , x1, x2, · · · , xn−2. Now we define the words
xji = xjxj−1...xi,
where j > i > 1 and xj1ε = xjxj−1...x1ε, ε = ±1.
43
Lemma 4.11 For ε = ±1, the set S of the following relations is a Grobner-Shirshov
basis of An in deg-lex ordering.
1) x2ε1 = x−ε
1
2) x2i = 1, (i > 1)
3) xjxi = xixj, (j − 1 > i ≥ 2)
4) xjxε1 = x−ε
1 xj, (j > 2)
5) xjixj = xj−1xji, (j > i ≥ 2)
6) xj1εxj = xj−1xj1−ε, (j > 2)
7) x2x1εx2 = x−ε
1 x2x−ε1
8) xε1x
−ε1 = 1
Now, we can state a presentation of An by generators and defining relations.
Theorem 4.12 Let
An = gp〈xi (1 ≤ i ≤ n− 2) ; x31 = 1, (xi−1xi)
3 = xi2 = 1 (2 ≤ i ≤ n− 2),
(xixj)2 = 1 (1 ≤ i < j − 1, j ≤ n− 2)〉.
Then
(i) every element w of An has a unique representation w = x1j1x2j2 ...xn−2jn−2, where
xtt = xt (t > 1), xii+1 = 1, xi1 = xixi−1 · · ·xε1, 1 ≤ ji ≤ i+ 1, 1 ≤ i ≤ n− 2, ε = ±1
(here we use xi1 instead of xi1ε).
(ii) the map φ : xi 7→ (12)((i+ 1)(i+ 2)), i = 1, 2, · · · , n− 2 induces an isomorphism of
An and An.
Proof. By Lemma 4.11, it is easy to see that every element w of An has a normal
form w = x1j1x2j2 ...xn−2jn−2 ∈ Irr(S), ji ≤ i + 1. Here x1j1 may have 3 possibilities
1, x1, x−11 ; x2j2 4 possibilities, and generally, xiji
i + 2 possibilities, 1 ≤ i ≤ n − 2. So
|An| = n!/2. On the other hand, φ is a homomorphism and |An| = n!/2. Hence φ is an
isomorphism.
2. Chinese monoid
44
The Chinese monoid CH(X,<) over a well-ordered set (X,<) has the following
presentation CH(X) = sgp〈X|S〉, where X = xi|i ∈ I and S consists of the following
relations:
xixjxk = xixkxj = xjxixk for every i > j > k,
xixjxj = xjxixj, xixixj = xixjxi for every i > j.
Theorem 4.13 With the deg-lex ordering on X∗ the following relations (1)-(5) forms a
Grobner-Shirshov basis of the Chinese monoid CH(X).
(1) xixjxk − xjxixk,
(2) xixkxj − xjxixk,
(3) xixjxj − xjxixj,
(4) xixixj − xixjxi,
(5) xixjxixk − xixkxixj,
where xi, xj, xk ∈ X, i > j > k.
Let Λ be the set which consists of words on X of the form un = w1w2 · · ·wn, n ≥ 0,
where
w1 = xt111
w2 = (x2x1)t21xt22
2
w3 = (x3x1)t31(x3x2)
t32xt333
· · ·
wn = (xnx1)tn1(xnx2)
tn2 · · · (xnxn−1)tn(n−1)xtnn
n
with xi ∈ X, x1 < x2 < · · · < xn and all exponents non-negative.
Corollary 4.14 The set Λ is normal forms of elements of the Chinese monoid CH(X).
3. Free inverse semigroup
45
Let S be a semigroup. s ∈ S is called an inverse of t ∈ S if sts = s, tst = t. An
inverse semigroup is a semigroup in which every element t has a unique inverse, denoted
by t−1.
Let X−1 = x−1|x ∈ X with X ∩X−1 = ∅. Denote X ∪X−1 by Y . We define the
formal inverses for elements of Y ∗ by the rules
1−1 = 1, (x−1)−1 = x (x ∈ X),
(y1y2 · · · yn)−1 = y−1n · · · y−1
2 y−11 (y1, y2, · · · , yn ∈ Y ).
It is well known that
FI(X) = sgp〈Y | aa−1a = a, aa−1bb−1 = bb−1aa−1, a, b ∈ Y ∗〉
is the free inverse semigroup (with identity) generated by X.
We give definitions in Y ∗ of formal idempotent, (prime) canonical idempotent and
ordered (prime) canonical idempotent. Let < be a well ordering on Y and then deg-lex
ordering on Y ∗.
(i) The empty word 1 is an idempotent.
(ii) If h is an idempotent and x ∈ Y , then x−1hx is both an idempotent and a prime
idempotent.
(iii) If e1, e2, · · · , em (m > 1) are prime idempotents, then e = e1e2 · · · em is an idempotent.
(iv) An idempotent w ∈ Y ∗ is called canonical if w has no subword of the form x−1exfx−1,
where x ∈ Y , both e and f are idempotents.
(v) A canonical idempotent w ∈ Y ∗ is called ordered if for any subword e = e1e2 · · · em
of w, fir(e1) < fir(e2) < · · · < fir(em), where each ei is an idempotent, m > 2 and
fir(u) is the first letter of u ∈ Y ∗.
Theorem 4.15 Let S be the set of the following two kinds of polynomials in k〈Y 〉:
• ef − fe, where e, f are ordered prime canonical idempotents such that ef > fe;
• x−1e′xf ′x−1−f ′x−1e′, where x ∈ Y, x−1e′x and xf ′x−1 are ordered prime canonical
idempotents.
Then, with deg-lex ordering on Y ∗, S is a Grober-Shirshov basis for free inverse
semigroup sgp〈Y |S〉.
46
Theorem 4.16 Normal forms of elememts of the free inverse semigroup sgp〈Y |S〉 are
u0e1u1 · · · emum ∈ Y ∗,
where m ≥ 0, u1, · · · , um−1 6= 1, u0u1 · · ·um has no subword of form yy−1 for y ∈ Y ,
e1, · · · , em are ordered canonical idempotents, and the first (last, respectively) letters of
ei (1 ≤ i ≤ m) are not equal to the first (last, respectively) letter of ui (ui−1, respectively).
The above normal form (Bokut, Chen, Zhao [41]) is an analogous to “semi-normal”
forms that were given in Poliakova-Schein [172], 2005.
§4.2 Composition-Diamond lemma for modules
Let X, Y be sets and modk〈X〉〈Y 〉 a free left k〈X〉-module with the basis Y . Then
modk〈X〉〈Y 〉 = ⊕y∈Y k〈X〉y is called a “double-free” module. We now define the GS basis
in modk〈X〉〈Y 〉. Suppose that < is a monomial ordering on X∗, < a well ordering on Y
and X∗Y = uy|u ∈ X∗, y ∈ Y . We define an ordering < on X∗Y as follows: for any
w1 = u1y1, w2 = u2y2 ∈ X∗Y ,
w1 < w2 ⇔ u1 < u2 or u1 = u2, y1 < y2
Let S ⊂ modk〈X〉〈Y 〉 with each s ∈ S monic. Then we define the composition in S
only the inclusion composition which means f = ag for some a ∈ X∗, where f, g ∈ S and
(f, g)f = f − ag. If (f, g)f =∑αiaisi, where αi ∈ k, ai ∈ X∗, si ∈ S and aisi < f , then
this composition is called trivial modulo (S, f).
Theorem 4.17 (CD-lemma for modules) Let S ⊂ modk〈X〉〈Y 〉 be a non-empty set with
each s ∈ S monic and < the ordering on X∗Y as before. Then the following statements
are equivalent:
(i) S is a Grobner-Shirshov basis in modk〈X〉〈Y 〉.
(ii) If 0 6= f ∈ k〈X〉S, then f = as for some a ∈ X∗, s ∈ S.
(iii) Irr(S) = w ∈ X∗Y |w 6= as, a ∈ X∗, s ∈ S is a linear basis for the factor
modk〈X〉〈Y |S〉 = modk〈X〉〈Y 〉/k〈X〉S.
Outline of the proof. Let u ∈ X∗Y, u = uXyu, uX ∈ X∗, yu ∈ Y. Define
47
cm(u, v) = aXu = bXv, lcm(u, v) = u = dXv,
where yu = yv.
Up to the ordering of u, v, one has cm(u, v) = c · lcm(u, v).
The composition of two monic f, g ∈ modk〈X〉(Y ) is
(f, g)|lcm(f ,g) = lcm(f , g)|f 7→f − lcm(f , g)|g 7→g.
If a1s1 = a2s2 for monic s1, s2, then a1s1 − a2s2 = c · (s1, s2)lcm(s1,s2).
It gives an analogy of Lemma 4.2 for modules and (i)⇒ (ii) of Theorem 4.17.
Let S ⊂ k〈X〉 and A = k〈X|S〉. Then, for any left A-module AM , we can regard
AM as a k〈X〉-module in a natural way: for any f ∈ k〈X〉, m ∈M , fm := (f+Id(S))m.
We note that AM is an epimorphic image of some free A-module. Now, we assume
that AM = modA〈Y |T 〉 = modA〈Y 〉/AT , where T ⊂ modA〈Y 〉. Let T1 = ∑fiyi ∈
modk〈X〉〈Y 〉|∑
(fi + Id(S))yi ∈ T and R = SX∗Y ∪ T1. Then, as k〈X〉-modules, AM =
modk〈X〉〈Y |R〉.
Assume that S is a GS basis for the left ideal k〈X〉S with no compositions at all
between different elements of S. It means that S is a minimal GS basis of the left ideal
k〈X〉S. Then, k〈X〉S is a free k〈X〉-module with the basis S. This shows the Cohn
Theorem 2.3.
As an application of CD-lemma for modules, we give GS bases for the Verma modules
over a Lie algebras of coefficients of a free Lie conformal algebras. We find a linear basis
for such a module.
Let B be a set of symbols. Let the locality function N : B × B → Z+ be a constant,
i.e. N(a, b) ≡ N for any a, b ∈ B. Let X = b(n)| b ∈ B, n ∈ Z and L = Lie(X|S) be
a Lie algebra over a field k of characteristic 0 generated by X with the relation S, where
S = ∑
s
(−1)s
(N
s
)[b(n− s)a(m+ s)] = 0| a, b ∈ B, m, n ∈ Z.
For any b ∈ B, let b =∑n
b(n)z−n−1 ∈ L[[z, z−1]]. Then, it is well-known that they
generate a free Lie conformal algebra C with data (B, N) (see [189]). Moreover, the
coefficient algebra of C is just L. Let B be a linearly ordered set. Define an ordering on
X in the following way:
a(m) < b(n)⇔ m < n or (m = n and a < b).
48
We use the deg-lex ordering on X∗. Then, it is clear that without loss of generality one
may assume the leading term of each polynomial in S is b(n)a(m) such that
n−m > N or (n−m = N and (b > a or (b = a and N is odd))).
Lemma 4.18 With the deg-lex ordering on X∗, S is a GS basis in Lie(X).
Corollary 4.19 Let U = U(L) be an universal enveloping algebra of L. Then a linear
basis of U consists of monomials
a1(n1)a2(n2) · · · ak(nk), ai ∈ B, ni ∈ Z
such that for any 1 ≤ i < k,
ni − ni+1 ≤
N − 1 if ai > ai+1 or (ai = ai+1 and N is odd)
N otherwise.
Proof. Viewing U as a k〈X〉-module, we have
UU = modU〈X| S(−)〉 = modk〈X〉〈I| S(−)X∗I〉.
Since S is a GS basis in Lie(X), S(−) is a GS basis in k〈X〉 by Theorem 5.37. Therefore,
S(−)X∗I is a GS basis in the free module modk〈X〉〈I〉. Now, the result follows from
Theorem 4.17.
§4.3 Grobner-Shirshov bases for tensor product of free
algebras
Let X and Y be sets and k〈X〉⊗k〈Y 〉 be the tensor product algebra. In this section,
we will give a Composition-Diamond lemma for the algebra k〈X〉 ⊗ k〈Y 〉. We will also
prove a theorem on the pair of algebras (k[X] ⊗ k〈Y 〉, k〈X〉 ⊗ k〈Y 〉) following the spirit
of Eisenbud, Peeva and Sturmfels’ theorem [97] on (k[X], k〈X〉).
§4.3.1 Composition-Diamond lemma for tensor product
Let X and Y be two well-ordered sets, T = yx = xy|x ∈ X, y ∈ Y . With the deg-
lex ordering (y > x for any x ∈ X, y ∈ Y ) on (X ∪ Y )∗, T is clearly a Grobner-Shirshov
basis in k〈X ∪ Y 〉. Then, by Theorem 4.4,
N = X∗Y ∗ = Irr(T ) = u = uXuY |uX ∈ X∗ and uY ∈ Y ∗
49
is a set of normal words of the tensor product
k〈X〉 ⊗ k〈Y 〉 = k〈X ∪ Y | T 〉.
Let kN be a k-linear space spanned by N . For any u = uXuY , v = vXvY ∈ N , we
define the multiplication of the normal words as follows
uv = uXvXuY vY ∈ N.
It is clear that kN is exactly the tensor product algebra k〈X〉 ⊗ k〈Y 〉, that is, kN =
k〈X ∪ Y |T 〉 = k〈X〉 ⊗ k〈Y 〉.Now, we order the set N . For any u = uXuY , v = vXvY ∈ N ,
u > v ⇔ |u| > |v| or (|u| = |v| and (uX > vX or (uX = vX and uY > vY ))),
where |u| = |uX | + |uY | is the length of u. Obviously, > is a monomial ordering on N .
Such an ordering is also called the deg-lex ordering on N = X∗Y ∗. Throughout this
section, we will adopt the above ordering unless it is otherwise stated.
For any polynomial f ∈ k〈X〉 ⊗ k〈Y 〉, f has a unique presentation of the form
f = αf f +∑
αiui,
where f , ui ∈ N, f > ui, αf , αi ∈ k.The proof of the following lemma are straightforward and we hence omit the details.
Lemma 4.20 Let f ∈ k〈X〉 ⊗ k〈Y 〉 be a monic polynomial. Then ufv = ufv for any
u, v ∈ N .
We give here the definition of compositions. Let f and g be two monic polynomials
of k〈X〉 ⊗ k〈Y 〉 and w = wXwY ∈ N . Then we have the following compositions.
1. Inclusion
1.1 X-inclusion only
Suppose that wX = fX = agXb for some a, b ∈ X∗, and fY , gY are disjoint. Then
there are two compositions according to wY = fY cgY and wY = gY cfY for c ∈ Y ∗,
respectively:
(f, g)w1 = fcgY − fY cagb, w1 = fX fY cgY
and
(f, g)w2 = gY cf − agbcfY , w2 = fX gY cfY .
1.2 Y -inclusion only
50
Suppose that wY = fY = cgY d for c, d ∈ Y ∗ and fX , gX are disjoint. Then there are
two compositions according to wX = fXagX and wX = gXafX for a ∈ X∗, respectively:
(f, g)w1 = fagX − fXacgd, w1 = fXagXfY
and
(f, g)w2 = gXaf − cgdafX w2 = gXafXfY .
1.3 X, Y -inclusion
Suppose that wX = fX = agXb for some a, b ∈ X∗ and wY = fY = cgY d for some
c, d ∈ Y ∗. Then
(f, g)w = f − acgbd.
The transformation f 7→ (f, g)w = f − acgbd is said to be the elimination of leading
word (ELW) of g in f .
1.4 X, Y -skew-inclusion
Suppose that wX = fX = agXb for some a, b ∈ X∗ and wY = gY = cfY d for some
c, d ∈ Y ∗. Then
(f, g)w = cfd− agb.
2. Intersection
2.1 X-intersection only
Suppose that wX = fXa = bgX for some a, b ∈ X∗ with |fX | + |gX | > |wX |, and
fY , gY are disjoint. Then there are two compositions according to wY = fY cgY and
wY = gY cfY for c ∈ Y ∗, respectively:
(f, g)w1 = facgY − fY cbg, w1 = wX fY cgY
and
(f, g)w2 = gY cfa− bgcfY , w2 = wX gY cfY .
2.2 Y -intersection only
Suppose that wY = fY c = dgX for some c, d ∈ Y ∗ with |fY | + |gY | > |wY |, and
fX , gX are disjoint. Then there are two compositions according to wX = fXagX and
wX = gXafX for a ∈ X∗, respectively:
(f, g)w1 = fcagX − fXadg, w1 = fXagXwY
and
(f, g)w2 = gXafc− dgafX , w2 = gXafXwY .
2.3 X, Y -intersection
51
If wX = fXa = bgX for some a, b ∈ X∗ and wY = fY c = dgY for some c, d ∈ Y ∗
together with |fX |+ |gX | > |wX | and |fY |+ |gY | > |wY |, then
(f, g)w = fac− bdg.
2.4 X, Y -skew-intersection
If wX = fXa = bgX for some a, b ∈ X∗ and wY = cfY = gY d for some c, d ∈ Y ∗
together with |fX |+ |gX | > |wX | and |fY |+ |gY | > |wY |, then
(f, g)w = cfa− bgd.
3. Both inclusion and intersection
3.1 X-inclusion and Y -intersection
There are two subcases to consider.
If wX = fX = agXb for some a, b ∈ X∗ and wY = fY c = dgY for some c, d ∈ Y ∗ with
|fY |+ |gY | > |wY |, then
(f, g)w = fc− adgb.
If wX = fX = agXb for some a, b ∈ X∗ and wY = cfY = gY d for some c, d ∈ Y ∗ with
|fY |+ |gY | > |wY |, then
(f, g)w = cf − agbd.
3.2 X-intersection and Y -inclusion
There are two subcases to consider.
If wX = fXa = bgX for some a, b ∈ X∗ with |fX |+ |gX | > |wX | and wY = fY = cgY d
for some c, d ∈ Y ∗, then
(f, g)w = fa− bcgd.
If wX = fXa = bgX for some a, b ∈ X∗ with |fX |+ |gX | > |wX | and wY = cfY d = gY
for some c, d ∈ Y ∗, then
(f, g)w = cfad− bg.
From Lemma 4.20, it follows that for any case of compositions
(f, g)w < w.
If Y = ∅, then the compositions of f, g are the same as in k〈X〉.Let S be a monic subset of k〈X〉 ⊗ k〈Y 〉 and f, g ∈ S. A composition (f, g)w is said
to be trivial modulo (S,w), denoted by
(f, g)w ≡ 0 mod(S,w),
52
if (f, g)w =∑
i αiaisibi, where ai, bi ∈ N, si ∈ S, αi ∈ k and aisibi < w for any i.
We call S a Grobner-Shirshov basis in k〈X〉 ⊗ k〈Y 〉 if all compositions of elements
in S are trivial modulo S and corresponding w.
Lemma 4.21 Let S be a Grobner-Shirshov basis in k〈X〉 ⊗ k〈Y 〉 and s1, s2 ∈ S. If
w = a1s1b1 = a2s2b2 for some ai, bi ∈ N, i = 1, 2, then a1s1b1 ≡ a2s2b2 mod(S,w).
Proof. There are four cases to consider.
Case 1 Inclusion
1.1 X-inclusion only
Suppose that wX1 = s1
X = as2Xb, a, b ∈ X∗ and s1
Y , s2Y are disjoint. Then
aX2 = aX
1 a and bX2 = bbX1 . There are two subcases to be considered: wY1 = s1
Y cs2Y and
wY1 = s2
Y cs1Y , where c ∈ Y ∗.
For wY1 = s1
Y cs2Y , we have w1 = sX
1 s1Y cs2
Y , aY2 = aY
1 s1Y c, bY1 = cs2
Y bY2 , w =
a1w1bX1 b
Y2 = a1s1acs2b2 and hence
a1s1b1 − a2s2b2 = a1s1bX1 cs2
Y bY2 − aX1 aa
Y1 s1
Y cs2bbX1 b
Y2
= a1(s1cs2Y − s1
Y cas2b)bX1 b
Y2
= a1(s1, s2)w1bX1 b
Y2
≡ 0 mod(S,w).
For wY1 = s2
Y cs1Y , we have w1 = sX
1 s2Y cs1
Y , aY1 = aY
2 s2Y c, bY2 = cs1
Y bY1 , w =
aX1 a
Y2 w1b1 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 s2
Y cs1b1 − aX1 aa
Y2 s2bb
X1 cs1
Y bY1
= aX1 a
Y2 (s2
Y cs1 − as2bcs1Y )b1
= aX1 a
Y2 (s1, s2)w1b1
≡ 0 mod(S,w).
1.2 Y -inclusion only
This case is similar to 1.1.
1.3 X, Y -inclusion
We may assume that s2 is a subword of s1, i.e., w1 = s1 = acs2bd, a, b ∈ X∗, c, d ∈ Y ∗,
aX2 = aX
1 a, bX2 = bbX1 , aY
2 = aY1 c and bY2 = dbY1 . Thus, a2 = a1ac, b2 = bdb1, w = a1w1b1
53
and hence
a1s1b1 − a2s2b2 = a1s1b1 − a1acs2bdb1
= a1(s1 − acs2bd)b1
= a1(s1, s2)w1b1
≡ 0 mod(S,w).
1.4 X,Y -skew-inclusion
Assume that wX1 = s1
X = as2Xb, a, b ∈ X∗ and wY
1 = s2Y = cs1
Y d, c, d ∈ Y ∗. Then
aX2 = aX
1 a, bX2 = bbX1 , aY
1 = aY2 c and bY1 = dbY2 . Thus, w = aX
1 aY2 w1b
X1 b
Y2 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 cs1b
X1 db
Y2 − aX
1 aaY2 s2bb
X1 b
Y2
= aX1 a
Y2 (cs1d− as2b)b
X1 b
Y2
= aX1 a
Y2 (s1, s2)w1b
X1 b
Y2
≡ 0 mod(S,w).
Case 2 Intersection
2.1 X-intersection only
We may assume that s1X is at the left of s2
X , i.e., wX1 = s1
Xb = as2X , a, b ∈ X∗
and |s1X | + |s2
X | > |wX1 |. Then aX
2 = aX1 a and bX1 = bbX2 . There are two subcases to be
considered: wY1 = s1
Y cs2Y and wY
1 = s2Y cs1
Y , c ∈ Y ∗.
For wY1 = s1
Y cs2Y , i.e., w1 = s1bcs2
Y , we have aY2 = aY
1 s1Y c, bY1 = cs2
Y bY2 , w =
a1s1acs2b2 = a1w1b2 and
a1s1b1 − a2s2b2 = a1s1bbX2 cs2
Y bY2 − aX1 aa
Y1 s2
Y cs2b2
= a1(s1bcs2Y − as2
Y cs2)b2
= a1(s1, s2)w1b2
≡ 0 mod(S,w).
For wY1 = s2
Y cs1Y , i.e., w1 = s2
Y cs1b, we have aY1 = aY
2 s2Y c, bY2 = cs1
Y bY1 , w =
aX1 a
Y2 w1b
X2 b
Y1 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 s2
Y cs1bbX2 b
Y1 − aX
1 aaY2 s2b
X2 cs1
Y bY1
= aX1 a
Y2 (s2
Y cs1b− as2cs1Y )bX2 b
Y1
= aX1 a
Y2 (s1, s2)w1b
X2 b
Y1
≡ 0 mod(S,w).
2.2 Y -intersection only
54
This case is similar to 2.1.
2.3 X,Y -intersection
Assume that wX1 = s1
Xb = as2X , wY
1 = s1Y d = cs2
Y , a, b ∈ X∗, c, d ∈ Y ∗, |s1X | +
|s2X | > |wX
1 | and |s1Y | + |s2
Y | > |wY1 |. Then aX
2 = aX1 a, b
X1 = bbX2 , aY
2 = aY1 c, b
Y1 = dbY2 ,
w = a1w1b2 and hence
a1s1b1 − a2s2b2 = a1s1bbX2 db
Y2 − aX
1 aaY1 cs2b2
= a1(s1bd− acs2)b2
= a1(s1, s2)w1b2
≡ 0 mod(S,w).
2.4 X,Y -skew-intersection
Assume that wX1 = s1
Xb = as2X , wY
1 = cs1Y = s2
Y d, |s1X | + |s2
X | > |wX1 |, |s1
Y | +|s2
Y | > |wY1 |, a, b ∈ X∗, c, d ∈ Y ∗. Then aX
2 = aX1 a, b
X1 = bbX2 , aY
1 = aY2 c, b
Y2 = dbY1 ,
w = aX1 a
Y2 w1b
X2 b
Y1 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 cs1bb
X2 b
Y1 − aX
1 aaY2 s2b
X2 db
Y1
= aX1 a
Y2 (cs1b− as2d)b
X2 b
Y1
= aX1 a
Y2 (s1, s2)w1b
X2 b
Y1
≡ 0 mod(S,w).
Case 3 Both inclusion and intersection
3.1 X-inclusion and Y -intersection
We may assume that wX1 = s1
X = as2Xb, a, b ∈ X∗. Then aX
2 = aX1 a and bX2 = bbX1 .
There two cases to consider: wY1 = s1
Y d = cs2Y and wY
1 = cs1Y = s2
Y d, where c, d ∈ Y ∗,
|s1Y |+ |s2
Y | > |wY1 |.
For wY1 = s1
Y d = cs2Y , we have aY
2 = aY1 c, b
Y1 = dbY2 , w = a1w1b
X1 b
Y2 and hence
a1s1b1 − a2s2b2 = a1s1bX1 db
Y2 − aX
1 aaY1 cs2bb
X1 b
Y2
= a1(s1d− acs2b)bX1 b
Y2
= a1(s1, s2)w1b2
≡ 0 mod(S,w).
For wY1 = cs1
Y = s2Y d, we have aY
1 = aY2 c, b
Y2 = dbY1 , w = aX
1 aY2 w1b
X2 db
Y1 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 cs1b1 − aX
1 aaY2 s2bb
X1 db
Y1
= aX1 a
Y2 (cs1 − as2bd)b1
= aX1 a
Y2 (s1, s2)w1b1
≡ 0 mod(S,w).
55
3.2 X-intersection and Y -inclusion
Assume that wX1 = s1
Xb = as2X , a, b ∈ Y ∗ with |s1
X | + |s2X | > |wX
1 |. Then
aX2 = aX
1 a, bX1 = bbX2 . There are two subcases to be considered: wY
1 = s1Y = cs2
Y d and
s2Y = cs1
Y d, where c, d ∈ Y ∗.
For wY1 = s1
Y = cs2Y d, we have aY
2 = aY1 c, b
Y2 = dbY1 , w = a1w1b
X2 b
Y1 and hence
a1s1b1 − a2s2b2 = a1s1bbX2 b
Y1 − a1acs2b
X2 db
Y1
= a1(s1b− acs2d)bX2 b
Y1
= a1(s1, s2)w1bX2 b
Y1
≡ 0 mod(S,w).
For wY1 = s2
Y = cs1Y d, we have aY
1 = aY2 c, b
Y1 = dbY2 , w = aX
1 aY2 w1b2 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 cs1bb
X2 db
Y2 − aX
1 aaY2 s2b2
= aX1 a
Y2 (cs1bd− as2)b2
= aX1 a
Y2 (s1, s2)w1b2
≡ 0 mod(S,w).
Case 4. s1 and s2 are disjoint
For w = wXwY , by symmetry, there are two subcases to be considered: wY =
aY1 s1
Y cs2Y bY2 and wY = aY
2 s2Y cs1
Y bY1 , where wX = aX1 s1
Xas2XbX2 , a ∈ X∗, aX
2 =
aX1 s1
Xa, bX1 = as2XbX2 and c ∈ Y ∗.
For w = aX1 s1
Xas2XbX2 a
Y1 s1
Y cs2Y bY2 = a1s1acs2b2, we have a2 = a1s1ac, b1 = acs2b2
and hence
a1s1b1 − a2s2b2 = a1s1acs2b2 − a1s1acs2b2
= a1(s1 − s1)acs2b2 − a1s1ac(s2 − s2)b2
≡ 0 mod(S,w).
For w = aX1 s1
Xas2XbX2 a
Y2 s2
Y cs1Y bY1 , we have aY
1 = aY2 s2
Y c, bY2 = cs1Y bY1 and hence
a1s1b1 − a2s2b2 = aX1 a
Y2 s2
Y cs1as2XbX2 b
Y1 − aX
1 s1XaaY
2 s2bX2 cs1
Y bY1
= aX1 a
Y2 (s2
Y cs1as2X − s1
Xas2cs1Y )bX2 b
Y1 .
56
Let s1 =∑n
i=1 αiuX1iu
Y1i and s2 =
∑mj=1 βju
X2ju
Y2j, where α1 = β1 = 1. Then
s2Y cs1as2
X − s1Xas2cs1
Y =n∑
i=2
αiuX1ias2cu
Y1i −
m∑j=2
βiuY2jcs1au
X2j
=n∑
i=2
αiuX1ia(s2 − s2)cu
Y1i +
m∑j=2
βjuY2jc(s1 − s1)au
X2j
+n∑
i=2
αiuX1ias2cu
Y1i −
m∑j=2
βjuY2jcs1au
X2j
≡n∑
i=2
m∑j=2
αiβjuX1iau
X2ju
Y2jcu
Y1i −
m∑j=2
n∑i=2
αiβjuY2jcu
Y1iu
X1iau
X2j
≡ 0 mod(S,w1)
where w1 = s2Y cs1as2
X = s1Xas2cs1
Y . Since w = aX1 a
Y2 w1b
X2 b
Y1 , we have
a1s1b1 − a2s2b2 = aX1 a
Y2 (s2
Y cs1as2X − s1
Xas2cs1Y )bX2 b
Y1
≡ 0 mod(S,w).
This completes the proof.
Lemma 4.22 Let S ⊂ k〈X〉 ⊗ k〈Y 〉 with each s ∈ S monic and Irr(S) = w ∈ N |w 6=asb, a, b ∈ N, s ∈ S. Then for any f ∈ k〈X〉 ⊗ k〈Y 〉,
f =∑
aisibi≤f
αiaisibi +∑uj≤f
βjuj,
where each αi, βj ∈ k, ai, bi ∈ N, si ∈ S and uj ∈ Irr(S).
Proof. Let f =∑i
αiui ∈ k〈X〉 ⊗ k〈Y 〉, where 0 6= αi ∈ k and u1 > u2 > · · · . If
u1 ∈ Irr(S), then let f1 = f − α1u1. If u1 6∈ Irr(S), then there exist some s ∈ S and
a1, b1 ∈ N such that f = a1s1b1. Let f1 = f − α1a1s1b1. In both cases, we have f1 < f .
Then the result follows from induction on f .
By summing up the above lemmas, we arrive at the following theorem:
Theorem 4.23 (CD-lemma for tensor product k〈X〉 ⊗ k〈Y 〉) Let S ⊂ k〈X〉 ⊗ k〈Y 〉 with
each s ∈ S monic and < the ordering on N = X∗Y ∗ as before. Then the following
statements are equivalent.
(i) S is a Grobner-Shirshov basis in k〈X〉 ⊗ k〈Y 〉.
(ii) f ∈ Id(S)⇒ f = asb for some a, b ∈ N, s ∈ S.
57
(iii) Irr(S) = w ∈ N |w 6= asb, a, b ∈ N, s ∈ S is a k-linear basis for the factor algebra
k〈X〉 ⊗ k〈Y 〉/Id(S).
Proof. (i) ⇒ (ii). Suppose that 0 6= f ∈ Id(S). Then f =∑αiaisibi for some
αi ∈ k, ai, bi ∈ N, si ∈ S. Let wi = aisibi and w1 = w2 = · · · = wl > wl+1 ≥ · · · . We will
prove that f = asb for some a, b ∈ N, s ∈ S, by using induction on (w1, l). If l = 1, then
the result is clear. If l > 1, then w1 = a1s1b1 = a2s2b2. Now, by (i) and Lemma 4.21, we
see that a1s1b1 ≡ a2s2b2 mod(S,w1). Thus,
α1a1s1b1 + α2a2s2b2 = (α1 + α2)a1s1b1 + α2(a2s2b2 − a1s1b1)
≡ (α1 + α2)a1s1b1 mod(S,w1).
Now (ii) follows.
(ii)⇒ (iii). For any 0 6= f ∈ k〈X〉 ⊗ k〈Y 〉, by Lemma 4.22, we can express f as
f =∑
αiaisibi +∑
βjuj,
where αi, βj ∈ k, ai, bi ∈ N, si ∈ S and uj ∈ Irr(S). Then Irr(S) generates the factor
algebra. Moreover, if 0 6= h =∑βjuj ∈ Id(S), uj ∈ Irr(S), u1 > u2 > · · · and β1 6= 0,
then u1 = h = asb for some a, b ∈ N, s ∈ S by (ii), a contradiction. This shows that
Irr(S) is a linear basis of the factor algebra.
(iii)⇒ (i). For any f, g ∈ S, we have h = (f, g)w ∈ Id(S). The result is now trivial
if (f, g)w = 0. Assume that (f, g)w 6= 0. Then, by Lemma 4.22 and (iii), we have
h =∑
aisibi≤h
αiaisibi.
Now, by noting that h = (f, g)w < w, we see immediately that (i) holds.
Remark: Theorem 4.23 is valid for any monomial ordering on X∗Y ∗.
Remark: Theorem 4.23 is precisely the Composition-Diamond lemma for associative
algebras (Theorem 4.4) when Y = ∅.
§4.3.2 Applications
Now, we give some applications of Theorem 4.23.
Example 4.24 Suppose that for the deg-lex ordering, S1 and S2 are GS bases in k〈X〉 and
k〈Y 〉 respectively. Then for the deg-lex ordering on X∗Y ∗ as before, S1 ∪S2 is a GS basis
in k〈X ∪Y |T 〉 = k〈X〉⊗k〈Y 〉. It follows that k〈X|S1〉⊗k〈Y |S2〉 = k〈X ∪Y |T ∪S1∪S2〉.
58
Proof. The possible compositions in S1 ∪ S2 are X-including only, X-intersection
only, Y -including only and Y -intersection only. Suppose that f, g ∈ S1 and (f, g)w1 ≡0 mod(S1, w1) in k〈X〉. Then in k〈X〉⊗k〈Y 〉, (f, g)w = (f, g)w1c, where w = w1c for any
c ∈ Y ∗. From this, it follows that each composition in S1 ∪ S2 is trivial modulo S1 ∪ S2.
A special case of Example 4.24 is the following.
Example 4.25 Let X,Y be well-ordered sets, k[X] the free commutative associative al-
gebra generated by X. Then S = xixj = xjxi|xi > xj, xi, xj ∈ X is a GS basis in
k〈X〉⊗k〈Y 〉 with respect to the deg-lex ordering. Therefore, k[X]⊗k〈Y 〉 = k〈X∪Y |T∪S〉.
In [97], a GS basis in k〈X〉 is constructed by lifting a commutative Grobner basis and
adding some commutators. Let X = x1, x2, . . . , xn, [X] the free commutative monoid
generated by X and k[X] the polynomial ring. Let
S1 = hij = xixj − xjxi| i > j ⊂ k〈X〉.
Then, consider the natural map γ : k〈X〉 → k[X] which maps xi to xi and the lexicographic
splitting of γ, which is defined as the k-linear map
δ : k[X]→ k〈X〉, xi1xi2 · · ·xir 7→ xi1xi2 · · ·xir if i1 ≤ i2 · · · ≤ ir.
For any u ∈ [X], we present u = xl11 x
l22 · · ·xln
n , where li ≥ 0.
We use any monomial ordering on [X].
Following [97], we define an ordering on X∗ using the ordering x1 < x2 < · · · < xn
as follows: for any u, v ∈ X∗,
u > v ⇔ γ(u) > γ(v) in [X] or (γ(u) = γ(v) and u >lex v).
It is easy to check that this ordering is monomial on X∗ and δ(s) = δ(s) for any s ∈ k[X].
Moreover, for any v ∈ γ−1(u), v ≥ δ(u).
For any m = xi1xi2 · · ·xir ∈ [X], i1 ≤ i2 · · · ≤ ir, denote the set of all the monomials
u ∈ [xi1+1, · · · , xir−1] by U(m).
The proofs of the following lemmas are straightforward.
Lemma 4.26 Let a, b ∈ X∗, a = δ(γ(a)), b = δ(γ(b)) and s ∈ k[X]. If w = aδ(s)b =
δ(γ(ab)s), then, in k〈X〉,
aδ(s)b ≡ δ(γ(ab)s) mod(S1, w).
59
Proof. Suppose that s = s + s′ and h = aδ(s)b − δ(γ(ab)s). Since aδ(s)b = δ(γ(ab)s),
we have h = aδ(s′)b− δ(γ(ab)s′), and h < w. By noting that γ(aδ(s′)b) = γ(δ(γ(ab)s′)),
h ≡ 0 mod(S1, w).
Lemma 4.27 Let f, g ∈ k[X], g = xi1xi2 · · ·xir (i1 ≤ i2 ≤ · · · ≤ ir) and w = δ(f g).
Then, in k〈X〉,
δ((f − f)g) ≡∑
αiaiδ(uig)bi mod(S1, w)
where αi ∈ k, ai ∈ [x ∈ X|x ≤ xi1 ], bi ∈ [x ∈ X|x ≥ xir ], ui ∈ U(g) and γ(∑αiaiuibi) =
f − f .
Theorem 4.28 Let the orderings on [X] and X∗ be defined as above. If S is a minimal
Grobner basis in k[X], then S ′ = δ(us)|s ∈ S, u ∈ U(s) ∪ S1 is a Grobner-Shirshov
basis in k〈X〉.
Proof. We will show that all the possible compositions of elements in S ′ are trivial. Let
f = δ(us1), g = δ(vs2) and hij = xixj − xjxi ∈ S ′.
(i) f ∧ g
Case 1. f and g have a composition of including, i.e., w = δ(us1) = aδ(vs2)b for
some a, b ∈ X∗ and a = δ(γ(a)), b = δ(γ(b)).
If s1 and s2 have no composition in k[X], i.e., lcm(s1s2) = s1s2, then u = u′s2, γ(ab)v =
u′s1 for some u′ ∈ [X]. By Lemma 4.26 and Lemma 4.27, we have
(f, g)w = δ(us1)− aδ(vs2)b
≡ δ(us1)− δ(γ(ab)vs2)
≡ δ(u′s2s1)− δ(u′s1s2)
≡ δ(u′(s1 − s1)s2)− δ(u′(s2 − s2)s1)
≡ 0 mod(S ′, w).
Since, in k[X], S is a minimal Grobner basis, the possible compositions are only
intersection. If s1 and s2 have composition of intersection in k[X], i.e., (s1, s2)w′ =
a′s1− b′s2, where a′, b′ ∈ [X], w′ = a′s1 = b′s2 and |w′| < |s1|+ |s2|, then w′ is a subword
of γ(w). Hence, we deduce that w = δ(tw′) = δ(ta′s1) = δ(tb′s2) and u = ta′, γ(ab)v = tb′
60
for some t ∈ [X]. Then
(f, g)w = δ(us1)− aδ(vs2)b
≡ δ(us1)− δ(γ(ab)vs2)
≡ δ(ta′s1)− δ(tb′s2)
≡ δ(t(a′s1 − b′s2))
≡ δ(t(s1, s2)w′)
≡ 0 mod(S ′, w)
since t(s1, s2)w′ < tw′ = γ(w).
Case 2. If f and g have a composition of intersection, we may assume that f is on
the left of g, i.e., w = δ(us1)a = bδ(vs2) for some a, b ∈ X∗ and a = δγ(a), b = δγ(b).
Similarly to Case 1, we have to consider whether s1 and s2 have compositions in k[X] or
not? One can check that both cases are trivial mod(S ′, w) by Lemma 4.26 and Lemma
4.27.
(ii) f ∧ hij
By noting that hij = xixj can not be a subword of f = δ(us1) since i > j, only
possible compositions are intersection. Suppose that s1 = xi1 · · ·xirxi, (i1 ≤ i2 ≤ · · · ≤ir ≤ i). Then f = δ(us1) = xi1vxi for some v ∈ k〈X〉, v = δγ(v) and w = δ(us1)xj.
If j ≤ i1, then
(f, hij)w = δ(us1)xj − xi1v(xixj − xjxi)
= δ(u(s1 − s1))xj + xi1vxjxi
≡ xjδ(u(s1 − s1)) + xjxi1vxi
≡ xj(δ(u(s1 − s1)) + δ(us1))
≡ xjδ(us1)
≡ 0 mod(S ′, w).
If j > i1, then uxj ∈ U(s1) and
(f, hij)w = δ(us1)xj − xi1v(xixj − xjxi)
= δ(u(s1 − s1))xj + xi1vxjxi
≡ δ(uxj(s1 − s1)) + δ(xi1vxixj)
≡ δ(uxj(s1 − s1)) + δ(uxj s1)
≡ δ(uxjs1)
≡ 0 mod(S ′, w).
61
Thus, the proof is completed.
Now we extend γ and δ as follows.
γ ⊗ 1 : k〈X〉 ⊗ k〈Y 〉 → k[X]⊗ k〈Y 〉, uXuY 7→ γ(uX)uY ,
δ ⊗ 1 : k[X]⊗ k〈Y 〉 → k〈X〉 ⊗ k〈Y 〉, uXuY 7→ δ(uX)uY .
Any polynomial f ∈ k[X] ⊗ k〈Y 〉 has a presentation f =∑αiu
Xi u
Yi , where αi ∈
k, uXi ∈ [X] and uY
i ∈ Y ∗.
Let the orderings on [X] and Y ∗ be any monomial orderings respectively. We order
the set [X]Y ∗ = u = uXuY |uX ∈ [X], uY ∈ Y ∗ as follows. For any u, v ∈ [X]Y ∗,
u > v ⇔ uY > vY or (uY = vY and uX > vX).
Now, we order X∗Y ∗: for any u, v ∈ X∗Y ∗,
u > v ⇔ γ(uX)uY > γ(vX)vY or (γ(uX)uY = γ(vX)vY and uX >lex vX).
This ordering is clearly a monomial ordering on X∗Y ∗.
The following definitions of compositions and GS bases are taken from [165].
Let f, g be monic polynomials of k[X] ⊗ k〈Y 〉, L the least common multiple of fX
and gX .
1. Inclusion
Let gY be a subword of fY , say, fY = cgY d for some c, d ∈ Y ∗. If fY = gY then
fX ≥ gX and if gY = 1 then we set c = 1. Let w = LfY = LcgY d. We define the
composition
C1(f, g, c)w =L
fXf − L
gXcgd.
2. Overlap
Let a non-empty beginning of gY be a non-empty ending of fY , say, fY = cc0, gY =
c0d, fY d = cgY for some c, d, c0 ∈ Y ∗ and c0 6= 1. Let w = LfY d = LcgY . We define the
composition
C2(f, g, c0)w =L
fXfd− L
gXcg.
3. External
Let c0 ∈ Y ∗ be any associative word (possibly empty). In the case that the greatest
common divisor of fX and gX is non-empty and fY , gY are non-empty, we define the
composition
C3(f, g, c0)w =L
fXfc0g
Y − L
gXfY c0g,
62
where w = LfY c0gY .
Let S be a monic subset of k[X]⊗k〈Y 〉. Then S is called a GS basis (standard basis)
if for any element f ∈ Id(S), f contains s as its subword for some s ∈ S.
It is defined as usual that a composition is trivial modulo S and corresponding w. We
also have that S is a GS basis in k[X]⊗ k〈Y 〉 if and only if all the possible compositions
of its elements are trivial. A GS basis in k[X] ⊗ k〈Y 〉 is called minimal if for any s ∈ Sand all si ∈ S \ s, si is not a subword of s.
Similar to the proof of Theorem 4.28, we have the following theorem.
Theorem 4.29 Let the orderings on [X]Y ∗ and X∗Y ∗ be defined as before. If S is a
minimal GS basis in k[X] ⊗ k〈Y 〉, then S ′ = δ ⊗ 1(us)|s ∈ S, u ∈ U(sX) ∪ S1 is a GS
basis in k〈X〉 ⊗ k〈Y 〉, where S1 = hij = xixj − xjxi| i > j.
Proof. We will show that all the possible compositions of elements in S ′ are trivial.
For s1, s2 ∈ S, let f = δ ⊗ 1(us1), g = δ ⊗ 1(vs2), hij = xixj − xjxi ∈ S ′ and
L = lcm(s1X , s2
X).
1. f ∧ gIn this case, all the possible compositions are related to the ambiguities w’s (in the
following, a, b ∈ X∗, c, d ∈ Y ∗).
1.1 X-inclusion only
wX = δ(us1X) = aδ(vs2
X)b, wY = s1Y cs2
Y or wY = s2Y cs1
Y .
1.2 Y -inclusion only
wX = δ(us1X)aδ(vs2
X) or wX = δ(vs2X)aδ(us1
X), wY = s1Y = cs2
Y d.
1.3 X, Y -inclusion
w = δ ⊗ 1(us1) = acδ ⊗ 1(vs2)bd.
1.4 X, Y -skew-inclusion
wX = δ(us1X) = aδ(vs2
X)b, wY = s2Y = cs1
Y d.
2.1 X-intersection only
wX = δ(us1X)a = bδ(vs2
X), wY = s1Y cs2
Y or wY = s2Y cs1
Y .
2.2 Y -intersection only
wX = δ(us1X)aδ(vs2
X) or wX = δ(us2X)aδ(vs1
X), wY = s1Y c = ds2
Y .
2.3 X, Y -intersection
wX = δ(us1X)a = bδ(vs2
X), wY = s1Y c = ds2
Y .
2.4 X, Y -skew-intersection
wX = δ(us1X)a = bδ(vs2
X), wY = cs1Y = s2
Y d.
3.1 X-inclusion and Y -intersection
63
wX = δ(us1X) = aδ(vs2
X)b, wY = s1Y c = ds2
Y or wY = cs1Y = s2
Y d.
3.2 X-intersection and Y -inclusion
wX = δ(us1X)a = bδ(vs2
X), wY = s1Y = cs2
Y d or wY = s2Y = cs1
Y d.
We only check the cases of 1.1, 1.2 and 1.3. Other cases are similarly checked.
1.1 X-inclusion only
Suppose that wX = δ(us1X) = aδ(vs2
X)b, a, b ∈ X∗ and s1Y , s2
Y are disjoint. There
are two cases to consider: wY = s1Y cs2
Y and wY = s2Y cs1
Y , where c ∈ Y ∗. We will only
prove the first case and the second is similar.
If s1 and s2 have no composition in k[X] ⊗ k〈Y 〉, i.e., lcm(s1, s2) = s1s2, then
u = u′sX2 , γ(ab)v = u′sX
1 for some u′ ∈ [X]. By the proof of Theorem 4.28, we have
(f, g)w = δ ⊗ 1(us1)cs2Y − s1
Y caδ ⊗ 1(vs2)b
≡ δ ⊗ 1(us1γ(cs2Y ))− δ ⊗ 1(γ(s1
Y c)γ(ab)vs2)
≡ δ ⊗ 1(u′s2Xs1cs2
Y )− δ ⊗ 1(s1Y cu′s1
Xs2)
≡ δ ⊗ 1(u′s1cs2)− δ ⊗ 1(u′s1cs2)
≡ δ ⊗ 1(u′(s1 − s1)cs2)− δ ⊗ 1(u′s1c(s2 − s2))
≡ 0 mod(S ′, w).
If s1 and s2 have composition of external (the elements of S have no composition of
inclusion because S is minimal and s1 and s2 have no composition of overlap because sY1
and sY2 are disjoint ) in k[X]⊗ k〈Y 〉, i.e., C3(s1, s2, c)w′ = L
s1X s1γ(cs2
Y )− Ls2
X γ(s1Y c)s2 =
t2s1γ(cs2Y ) − t1γ(s1
Y c)s2 where gcd(s1X , s2
X) = t 6= 1, s1X = tt1, s2
X = tt2 and L =
tt1t2, w′ = Lγ(s1
Y cs2Y ), then w′ is a subword of γ(w). Therefore, we have w = δ⊗1(mw′)
and u = mt2, γ(ab)v = mt1 since ut1 = γ(ab)vt2 and gcd(t1, t2) = 1. Then
(f, g)w = δ ⊗ 1(us1)cs2Y − s1
Y caδ ⊗ 1(vs2)b
≡ δ ⊗ 1(us1γ(cs2Y ))− δ ⊗ 1(γ(s1
Y c)γ(ab)vs2)
≡ δ ⊗ 1(mt2s1γ(cs2Y ))− δ ⊗ 1(mt1γ(s1
Y c)s2)
≡ δ ⊗ 1(mC3(s1, s2, c)w′)
≡ 0 mod(S ′, w)
since mC3(s1, s2, c)w′ < mw′ = γ(w).
1.2 Y -inclusion only
Suppose that wY = s1Y = cs2
Y d, c, d ∈ Y ∗ and δ(us1X), δ(vs2
X) are disjoint. Then
there are two compositions according to wX = δ(us1X)aδ(vs2
X) and wX = δ(vs2X)aδ(us1
X)
64
for a ∈ X∗. We only prove the first.
(f, g)w = δ ⊗ 1(us1)aδ(vs2X)− δ(us1
X)acδ ⊗ 1(vs2)d
≡ δ ⊗ 1(us1γ(a)vs2X − us1
Xγ(a)vγ(c)s2γ(d))
≡ δ ⊗ 1(uγ(a)v(s1s2X − s1
Xγ(c)s2γ(d)))
≡ δ ⊗ 1(uγ(a)vC1(s1, s2, γ(c))w′)
≡ 0 mod(S ′, w),
where w′ = s1X s2
X s1Y = s1
X s2Xγ(c)s2
Y γ(d) and uγ(a)vC1(s1, s2, γ(c))w′ < uγ(a)vw′ =
γ(w).
1.3 X,Y -inclusion
We may assume that g is a subword of f , i.e., w = δ⊗1(us1) = acδ⊗1(vs2)bd, a, b ∈X∗, c, d ∈ Y ∗. Then us1
X = γ(ab)vs2X = mL for some m ∈ [X], us1
Y = γ(c)s2Y γ(d).
(f, g)w = δ ⊗ 1(us1)− acδ ⊗ 1(vs2)bd
≡ δ ⊗ 1(us1 − γ(ac)vs2γ(bd))
≡ δ ⊗ 1(mL
s1Xs1 −m
L
s2Xγ(c)s2γ(d))
≡ δ ⊗ 1(mC1(s1, s2, γ(c))w′)
≡ 0 mod(S ′, w),
where w′ = Lγ(c)s2Y γ(d) and mC1(s1, s2, c)w′ < mw′ = γ(w).
2. f ∧ hij
Similar to the proof of Theorem 4.28, they only have compositions of X-intersection.
Suppose that s1X = xi1 · · ·xirxi, (i1 ≤ i2 ≤ · · · ≤ ir ≤ i). Then f = δ ⊗ 1(us1) =
xi1vxis1Y for some v ∈ k〈X〉, and v = δγ(v) and w = δ ⊗ 1(us1)xj s1
Y .
If j ≤ i1, then
(f, hij)w = δ ⊗ 1(us1)xj − xi1vs1Y (xixj − xjxi)
= δ ⊗ 1(u(s1 − s1))xj + xi1vxjxis1Y
≡ xjδ ⊗ 1(u(s1 − s1)) + xjxi1vxis1Y
≡ xj(δ ⊗ 1(u(s1 − s1)) + δ ⊗ 1(us1))
≡ xjδ ⊗ 1(us1)
≡ 0 mod(S ′, w).
65
If j > i1, then uxj ∈ U(s1) and
(f, hij)w = δ ⊗ 1(us1)xj − xi1vs1Y (xixj − xjxi)
= δ ⊗ 1(u(s1 − s1))xj + xi1vxjxis1Y
≡ δ ⊗ 1(uxj(s1 − s1)) + δ ⊗ 1(xi1vxixj s1Y )
≡ δ ⊗ 1(uxj(s1 − s1)) + δ ⊗ 1(uxj s1)
≡ δ ⊗ 1(uxjs1)
≡ 0 mod(S ′, w).
This completes the proof.
As an application of the above result, we have now constructed a GS basis for the
tensor product k〈X〉⊗k〈Y 〉 by lifting a given GS basis in the tensor product k[X]⊗k〈Y 〉in which k[X] is a commutative algebra.
66
Chapter 5 Grobner-Shirshov Bases for Lie Algebras
over a Field
§5.1 Lyndon-Shirshov words
We start with the Lyndon-Shirshov associative words.
Let X = xi|i ∈ I be a well-ordered set with xi > xp if i > p for any i, p ∈ I. Let
X∗ be the free monoid generated by X. For u = xi1xi2 · · ·xik ∈ X∗, let
xβ = min(u) = minxi1 , xi2 , · · · , xik,
fir(u) = xi1 ,
length of u : |u| = k.
Definition 5.1 Let u = xi1xi2 · · ·xik ∈ X∗. Then u is called Weak-ALSW if fir(u) >
min(u) or |u| = 1, where ALSW means an “associative Lyndon-Shirshov word”.
Let u be a Weak-ALSW, min(u) = xβ and |u| ≥ 2. We define
X ′(u) = xji = xi xβ · · ·xβ︸ ︷︷ ︸
j
|i > β, j ≥ 0.
Note that xji = xi xβ · · ·xβ︸ ︷︷ ︸
j
is just a symbol.
Now, we order X ′(u) by the following way:
xj1i1> xj2
i2⇔ i1 > i2 or (i1 = i2, j2 > j1).
Suppose that u, v are Weak-ALSW’s and min(v) ≥ min(u) = xβ. Then we define
v′u = xm1i1· · ·xmt
itin (X ′(u))∗ ⇔ v = xi1 xβ · · ·xβ︸ ︷︷ ︸
m1
· · ·xit xβ · · ·xβ︸ ︷︷ ︸mt
in X∗,
where xij > xβ, mj ∈ N , 1 ≤ j ≤ t. For the sake of simpler notation, we use u′ instead
of u′u.
Throughout this section and next section, we assume that x1 < x2 < x3 < · · · .
Example 5.2 Let u = x2x1, v = x3x2. Then v′u = x03x
02, v′ = x1
3.
The following lemma is obvious.
67
Lemma 5.3 Let u be a Weak-ALSW, xβ = min(u), u = vw, v, w 6= 1 and w 6= xβw1.
Then u′ = v′uw′u.
Example 5.4 Let v = x3x2x1, w = x2x2 and u = vw = x3x2x1x2x2. Then u′ = x03x
12x
02x
02,
v′u = x03x
12, w
′u = x0
2x02 and u′ = v′uw
′u.
Recall that without specific explanation, we always use the lexicographic ordering
both on (X ′(u))∗ and X∗ (i.e., w > wt if t 6= 1 and zxit1 > zxjt2 if xi > xj).
Lemma 5.5 Let xβ = min(uv), fir(u) 6= xβ and fir(v) 6= xβ. Then u > v ⇔ u′uv > v′uv.
Proof. Assume that u > v. Then there are two cases to consider.
Case 1:
u = xi1 xβ · · ·xβ︸ ︷︷ ︸l1
· · ·xis−1 xβ · · ·xβ︸ ︷︷ ︸ls−1
xis xβ · · ·xβ︸ ︷︷ ︸ls
· · ·
v = xi1 xβ · · ·xβ︸ ︷︷ ︸l1
· · ·xis−1 xβ · · ·xβ︸ ︷︷ ︸ls−1
yz · · · , where xis > y.
(a) If y = xβ, then
u′uv = xl1i1· · ·xls−2
is−2x
ls−1
is−1xls
is· · ·
v′uv = xl1i1· · ·xls−2
is−2x
l′s−1
is−1· · · , where l′s−1 > ls1 .
So, u′uv > v′uv.
(b) If y > xβ, then
u′uv = xl1i1· · ·xls−1
is−1xls
is· · ·
v′uv = xl1i1· · ·xls−1
is−1yn · · · .
So, u′uv > v′uv.
Case 2:
u = xi1 xβ · · ·xβ︸ ︷︷ ︸l1
· · ·xis xβ · · ·xβ︸ ︷︷ ︸ls
v = xi1 xβ · · ·xβ︸ ︷︷ ︸l1
· · ·xis xβ · · ·xβ︸ ︷︷ ︸ls
yz · · · .
(a) If y = xβ, then
u′uv = xl1i1· · ·xls
is
v′uv = xl1i1· · ·xl′s
is· · · , where l′s > ls.
So, u′uv > v′uv.
68
(b) If y > xβ, then v′uv = u′uvyn · · · and so, u′uv > v′uv.
Conversely, assume that u′uv > v′uv. We will prove that u > v. There are also two
cases to consider.
Case 1: u′uv = xl1i1· · ·xls
is, v′uv = xl1
i1· · ·xls
isyn · · · .
Case 2: u′uv = xl1i1· · ·xls−1
is−1xls
is· · · , v′uv = xl1
i1· · ·xls−1
is−1xi′s
ls′ · · · , where xis > xi′s or
(xis = xi′s and l′s > ls).
In both cases, it is clear that u > v.
Definition 5.6 Let u ∈ X∗. Then u is called an ALSW if
(∀v, w ∈ X∗, v, w 6= 1) u = vw ⇒ vw > wv.
Remark: Let u, v ∈ X∗ and the v′u ∈ (X ′(u))∗ be as before. We denote by |v| the length
of v in X∗ and |v′u|X′ the length of v′u in (X ′(u))∗.
Lemma 5.7 Let u be a Weak-ALSW with |u| ≥ 2. Then u is an ALSW in X∗ if and
only if u′ is an ALSW in (X ′(u))∗.
Proof. “ =⇒ ” If |u′|X′ = 1, then u′ is an ALSW. Suppose that |u′|X′ > 1 and
u′ = v′uw′u. Then u = vw. Since u is an ALSW, vw > wv which implies (vw)′u > (wv)′u
by Lemma 5.5. Therefore, by Lemma 5.3, v′uw′u > w′
uv′u and so, u′ is an ALSW.
“ ⇐= ” Let u = vw and xβ = min(u). If fir(w) = xβ, then vw > wv. If
fir(w) 6= xβ, then
u′ = v′uw′u ⇒ v′uw
′u > w′
uv′u ⇒ (vw)′u > (wv)′u ⇒ vw > wv.
Hence, u is an ALSW.
Remark For a Weak-ALSW u, it is clear that |u′|X′ < |u| if |u| > 1. For an ALSW u,
we denote by u′′ = (u′)′ and u(k) = (u′)(k−1) for k > 0 generally. From this, it follows that
Xk(u) = Xk−1(u′).
Lemma 5.8 For u ∈ X∗, u is an ALSW if and only if (∃k ≥ 0), s.t., |u(k)|Xk(u) = 1.
Proof. We apply induction on |u|. If |u| = 1, then there is nothing to do. Assume that
|u| > 1. Since |u′|X′ < |u| and
|u(k)|Xk(u) = |(u′)(k−1)|Xk−1(u′),
by induction and by Lemma 5.7, the result follows.
69
Example 5.9 Let u = x5x4x5x3. Then
u′ = x05x
04x
15, u
′′ = (x05)
1(x15)
0 and u′′′ = ((x05)
1)1.
Therefore, by Lemma 5.8, u is an ALSW.
Lemma 5.10 Let u ∈ X∗. Then u is an ALSW if and only if
(∀v, w ∈ X∗, v, w 6= 1) u = vw ⇒ u > w.
Proof. “ =⇒ ” Induction on |u|. If |u| = 1, then the result clearly holds. Suppose that
|u| ≥ 2, xβ = min(u) and u = vw, v, w 6= 1. If w = xβw1, then u > w. If w 6= xβw1, then
u′ = v′uw′u. Since u′ is an ALSW, by induction, u′ > w′
u. Hence, by Lemma 5.5, u > w.
“⇐= ” Since u = vw > w > wv, we have u is an ALSW.
Lemma 5.11 Suppose that u is an ALSW, xβ = min(u) and |u| > 1. Then uxβ is an
ALSW.
Proof. Follows from Lemma 5.10.
Lemma 5.12 Let u and v be ALSW’s. Then uv is an ALSW if and only if u > v.
Proof. “ =⇒ ” Suppose that uv is an ALSW. Then, by Lemma 5.10, u > uv > v.
“ ⇐= ” We use induction on |uv|. Suppose that u > v. If |uv| = 2 or v = xβ =
min(uv), then the result is obvious. Otherwise, we can get that u′uv > v′uv, where u′uv, v′uv
are ALSW’s. By induction, u′uvv′uv = (uv)′ is an ALSW and so is uv.
Lemma 5.13 For any u ∈ X∗, there exists a unique decomposition u = u1u2 · · ·uk, where
ui is an ALSW, 1 ≤ i ≤ k, and u1 ≤ u2 ≤ · · · ≤ uk.
Proof. To prove the existence, we use induction on |u|. If |u| = 1 then it is trivial.
Let |u| > 1 and xβ = min(u). If u = xβv, then v has the required decomposition and so
does u. Otherwise, u is a Weak-ALSW. Thus, u′ has the decomposition and so does u,
by Lemmas 5.5 and 5.7.
To prove the uniqueness, we let u = u1 · · ·uk = w1 · · ·ws be the decompositions such
that ui, wj are ALSW’s for any i, j; u1 ≤ · · · ≤ uk and w1 ≤ · · · ≤ ws. If u = xβv,
then u1 = w1 = xβ and the result follows from the induction on |u|. Otherwise, u is a
Weak-ALSW and u′ = u′1u · · ·u′ku = w′1u · · ·w′
su are the decompositions of u′. Now, by
induction again, the result follows.
Remark: In Lemma 5.13, the word uk is the longest ALSW end of u.
70
Example 5.14 Let u = x1x1x2x1x2x1x1. Then
u = x1︸︷︷︸u1
x1︸︷︷︸u2
x2x1x2x1x1︸ ︷︷ ︸u3
= u1u2u3
is the decomposition of u.
Lemma 5.15 Let u be an ALSW and |u| ≥ 2. If u = vw, where w is the longest ALSW
proper end of u, then v is an ALSW.
Proof. Suppose that v is not an ALSW. Then, by Lemma 5.13, we can assume that
v = v1v2 · · · vm (m > 1),
where each vi is an ALSW and v1 ≤ v2 ≤ · · · ≤ vm. If vm > w, then vmw is an ALSW
and |vmw| > |w|, a contradiction. If vm ≤ w, then we get another decomposition of u
which contradicts the uniqueness in Lemma 5.13. Thus, v must be an ALSW.
Example 5.16 Let u = x5x4x5x4x3x5x3. Then
u = x5x4︸︷︷︸v
x5x4x5x3︸ ︷︷ ︸w
= vw
and u, v, w are all ALSW’s.
Now, for an ALSW u, we introduce two bracketing ways.
One is up-to-down bracketing which is defined inductively by
[xi] = xi, [u] = [[v][w]],
where u = vw and w is the longest ALSW proper end of u.
Example 5.17 Let u = x2x2x1x1x2x1. Then
u → [[x2x2x1x1][x2x1]]→ [[x2[x2x1x1]][x2x1]]→ [[x2[[x2x1]x1]][x2x1]].
The other is down-to-up bracketing. Let us explain it on a sample word
u = x2x2x1x1x2x1.
Join the minimal letter x1 to the previous letters:
u 7→ x2[x2x1]x1[x2x1].
71
Form a new alphabet of the nonassociative words x2, [x2x1] and x1 ordered lexicograph-
ically, i.e.,
x2 > [x2x1] > x1.
Join the minimal letter x1 to the previous letters:
x2[x2x1]x1[x2x1] 7→ x2[[x2x1]x1][x2x1].
Form a new alphabet
x2 > [x2x1] > [[x2x1]x1].
Join the minimal letter [[x2x1]x1] to the previous letter:
x2[[x2x1]x1][x2x1] 7→ [x2[[x2x1]x1]][x2x1].
Form a new alphabet
[x2[[x2x1]x1]] > [x2x1].
Finally, join the minimal letter [x2x1] to the previous letter:
[x2[[x2x1]x1]][x2x1] 7→ [[x2[[x2x1]x1]][x2x1]] = [u].
Remark: We denote by [ ] the down-to-up bracketing and by [[ ]] the up-to-down brack-
eting. Clearly, if u = xixβ · · ·xβ, then [[u]] = [u].
Lemma 5.18 Let u be an ALSW with xβ = min(u). Then
(i) [[u′]] |xji 7→[[xixβ ]···xβ ]= [[u]].
(ii) [u′] |xji 7→[[xixβ ]···xβ ]= [u].
Proof. Induction on |u′|. If |u′| = 1, then u = xixβ · · ·xβ and the results hold. Assume
that |u′| > 1.
Let u′ = v′uw′u, where w′
u is the longest ALSW proper end of u′. Then u = vw and w
is the longest ALSW proper end of u. By induction, we can get
[[u′]] |xji 7→[[xixβ ]···xβ ]= [[[v′u]][[w
′u]]] |xj
i 7→[[xixβ ]···xβ ]= [[[v]][[w]]] = [[u]].
This shows (i).
Now we prove (ii). Since |u′| > 1, there exists ALSW’s v′u, w′u such that v′u > w′
u and
[u′] = [[v′u][w′u]]. By Lemma 5.5, we have v > w and [u] = [[v][w]]. So
[u′] |xji 7→[[xixβ ]···xβ ]= [[v′u][w
′u]] |xj
i 7→[[xixβ ]···xβ ]= [[v][w]] = [u].
72
Lemma 5.19 [u] = [[u]] for any ALSW u.
Proof. Induction on |u|. If |u| = 1, then u = xi and [xi] = [[xi]] = xi. Assume that
|u| > 1. Since |u′|X′ < |u|, by induction, we can get [u′] = [[u′]]. By Lemma 5.18,
[u] = [u′] |xji 7→[[xixβ ]···xβ ]= [[u′]] |xj
i 7→[[xixβ ]···xβ ]= [[u]].
§5.2 Free Lie algebras
Now we give the definition of a non-associative Lyndon-Shirshov word.
Definition 5.20 Let < be the ordering on X∗ as before and (u) a non-associative word.
Then (u) is called a non-associative Lyndon-Shirshov word, denoted by NLSW, if
(i) u is an ALSW,
(ii) if (u) = ((v)(w)), then both (v) and (w) are NLSW’s,
(iii) in (ii) if (v) = ((v1)(v2)), then v2 ≤ w in X∗.
Remark In Definition 5.20 (ii), v > w by Lemma 5.12.
Theorem 5.21 Let u be an ALSW. Then there exists a unique bracketing way such that
(u) is a NLSW.
Proof. (Existence). Let u be an ALSW. We will prove that up-to-down bracketing is
one of bracketing way such that [[u]] is a NLSW. Induction on |u|. If |u| = 1, then nothing
to do. Suppose that |u| > 1 and u = vw where w is the longest ALSW proper end of u.
Then, [[u]] = [[[v]][[w]]]. By induction, both [[v]] and [[w]] are NLSW’s. Now, we assume
that [[v]] = [[v1]][[v2]] and v2 > w. Then, v2w is an ALSW, a contradiction. So, v2 ≤ w
and hence, [[u]] is a NLSW.
(Uniqueness). Assume that ( ) is a bracketing way such that (u) is a NLSW. We use
induction on |u| to show (u) = [[u]]. If |u| = 1, then (u) = [[u]] clearly. Suppose that
|u| > 1. Since |u′|X′ < |u|, (u′) = [[u′]] by induction. By substitution and Lemma 5.18,
(u′) |xji 7→[[xixβ ]···xβ ]= [[u′]] |xj
i 7→[[xixβ ]···xβ ], i.e., (u) = [[u]].
73
Let X∗∗ be the set of all non-associative words (u) in X. If (u) is a NLSW, then we
denote it by [u].
From now on, let k〈X〉 be the free associative algebra generated by X. We consider
( ) as Lie bracket in k〈X〉, i.e., for any a, b ∈ k〈X〉, (ab) = ab − ba. Denote by Lie(X)
the subLie-algebra of k〈X〉 generated by X.
Given a polynomial f ∈ k〈X〉, it has the leading word f ∈ X∗ according to the above
ordering on X∗ such that
f =∑u∈X∗
f(u)u = αf +∑
αiui,
where f, ui ∈ X∗, f > ui, α, αi, f(u) ∈ k. We call f the leading term of f . Denote the
set u|f(u) 6= 0 by suppf and deg(f) by |f |. f is called monic if α = 1.
Note that if |u| = |v| and u < v, then the lexicographic ordering which we use before
is the same as the degree-lexicographic ordering on X∗.
Theorem 5.22 Let the ordering < be as before. Then, for any (u) ∈ X∗∗, (u) has a
representation:
(u) =∑
αi[ui],
where each αi ∈ k, [ui] is a NLSW and |ui| = |u|. Even more, if (u) = ([v][w]), then
ui > minv, w.
Proof. Induction on |u|. If |u| = 1, then (u) = [u] and the result holds. Suppose that
|u| > 1 and (u) = ((v)(w)). Then, by induction,
(v) =∑
αi[vi] and (w) =∑
βj[wj],
where αi, βj ∈ k, [vi], [wj] are NLSW’s, |vi| = |v| and |wj| = |w|. Without loss of
generality, we may assume that (u) = ([v][w]) with v > w because of ([v][w]) = −([w][v]).
If |v| = 1, then
(u) = ([v][w])
is a NLSW. Suppose that |v| > 1 and [v] = [[v1][v2]].
There are two subcases
(a) If v2 ≤ w, then (u) = (([v1][v2])[w]) is a NLSW.
(b) If v2 > w, then
(u) = (([v1][v2])[w]) = (([v1][w])[v2]) + ([v1]([v2][w])).
74
By induction,
([v1][w]) =∑
γi[ti], ti > minv1, w = w
([v2][w]) =∑
γj′[t′j], t′j > minv2, w = w.
Then,
(u) =∑
γi([ti][v2]) +∑
γj′([v1][t
′j]).
By noting that
minti, v2 and mint′j, v1 > minv, w = w,
the result follows from the inverse induction on minv, w.
Example 5.23 Let (u) = (((x3x2)(x2x1))(x2x1x1)). Then
(u) = (((x3(x2x1))x2)(x2x1x1)) + ((x3(x2(x2x1)))(x2x1x1)),
(((x3(x2x1))x2)(x2x1x1)) = (((x3(x2x1))(x2x1x1))x2) + ((x3(x2x1))(x2(x2x1x1)))
= (((x3(x2x1x1))(x2x1))x2) + ((x3((x2x1)(x2x1x1)))x2)
+((x3(x2x1))(x2(x2x1x1))),
((x3(x2(x2x1)))(x2x1x1)) = ((x3(x2x1x1))(x2(x2x1))) + (x3((x2(x2x1))(x2x1x1)))
= ((x3(x2x1x1))(x2(x2x1))) + (x3((x2(x2x1x1))(x2x1))
+(x3(x2((x2x1)(x2x1x1)))),
and hence,
(u) = (((x3(x2x1x1))(x2x1))x2) + ((x3((x2x1)(x2x1x1)))x2)
+((x3(x2x1))(x2(x2x1x1))) + ((x3(x2x1x1))(x2(x2x1)))
+(x3((x2(x2x1x1))(x2x1)) + (x3(x2((x2x1)(x2x1x1))))
is a linear combination of NLSW’s.
Lemma 5.24 Let [u] be a NLSW. Then [u] = u.
Proof. We use induction on |u|. If |u| = 1, then the result holds immediately. Let
|u| > 1 and [u] = [[v][w]]. Then, by induction, [v] = v and [w] = w. Suppose that
[v] = v +∑vi<v
αivi, [w] = w +∑
wj<w
βjwj,
75
where αi, βj ∈ k, v, vi, w, wj ∈ X∗. It is easy to see that |vi| = |v| and |wj| = |w| for any
i, j. Then,
[u] = [(v +∑vi<v
αivi)(w +∑
wj<w
βjwj)]
= (v +∑vi<v
αivi)(w +∑
wj<w
βjwj)− (w +∑
wj<w
βjwj)(v +∑vi<v
αivi)
= vw +∑
wj<w
βjvwj +∑vi<v
αiviw +∑
αiβjviwj
− wv −∑
wj<w
βjwjv −∑vi<v
αiwvi −∑
βjαiwjvi.
Since
vw > vwj, viw, viwj, wv and wv > wvi, wjv, wjvi,
we have [u] = u.
Remark. By the proof of Lemma 5.24, if we consider [u] as a polynomial in k〈X〉, then
each r ∈ supp([u]) has the same length as u, moreover, cont(r) = cont(u), where, for
example, cont(u) = xi1 , · · · , xit if u = xi1 · · ·xit ∈ X∗.
Lemma 5.25 NLSW’s are k-linear independent.
Proof. Supposek∑
i=1
αi[ui] = 0,
where each αi ∈ k, [ui] is a NLSW and u1 > u2 > · · · > uk. If α1 6= 0, then∑i
αi[ui] =
u1 6= 0, a contradiction. Then, all αi must be 0.
By Theorem 5.22 and Lemma 5.25, we have the following corollary.
Corollary 5.26 NLSW’s is a linear basis of Lie(X).
From Corollary 5.26 and Lemma 5.24, we have
Corollary 5.27 For any f ∈ Lie(X), f is an ALSW.
Theorem 5.28 Lie(X) is the free Lie algebra generated by X.
Proof. Let L be a Lie algebra and g : X −→ L a mapping. Then, we define a mapping
g1 : Lie(X) −→ L; [xi1 · · ·xin ] 7−→ [g(xi1) · · · g(xin)],
76
where [xi1 · · ·xin ] is NLSW. It is easy to check g1 is a unique Lie homomorphism such
that g1i = g.
-
?
L
Lie(X)X i
g∃! g1
The following theorem plays a key role in proving the Composition-Diamond lemma
for Lie algebras (see Theorem 5.39).
Theorem 5.29 (A.I. Shirshov) Let u, v be ALSW’s, u = avb, a, b ∈ X∗. Then
(i) [u] = [a[vc]d], where b = cd, c, d ∈ X∗.
(ii) Let
[u]v = [u]|[vc] 7→[[[v][c1]]···[ck]], (5.1)
where c = c1 · · · ck, cj is an ALSW and c1 ≤ c2 ≤ · · · ≤ ck. Then, in k〈X〉,
[u]v = u.
Moreover,
[u]v = a[v]b+∑
i
αiai[v]bi,
where each αi ∈ k and aivbi < avb.
Proof. (i) Induction on |u|. If |u| = 1, then u = v = xi and the result holds. Assume
that |u| > 1. If v = xi, then [u] = [a[xi]d] and the result holds. Now, we consider the case
of |v| > 1. Let xβ = min(u) and b = xeβ b, where e ≥ 0 and fir(b) 6= xβ. Then
u = avb = avxeβ b = avb,
where v = vxeβ is also an ALSW, by Lemma 5.11. Then, by induction, for u′ = a′uv
′ub
′u,
we have [u′] = [a′u[v′uc
′u]d
′u], b
′u = c′ud
′u. By substitution
xji 7→ [[xixβ] · · ·xβ], we obtain
[u] = [a[vc]d] = [a[vxeβ c]d] = [a[vc]d], where c = xe
β c.
(ii) If c = 1, then [u]v = [u] and the results hold clearly. Otherwise, by Lemma 5.13,
we may assume that
c = xβ · · ·xβcl+1 · · · ck,
77
where each ci is an ALSW and xβ < cl+1 ≤ · · · ≤ ck.
Then
[u]v = [u]|[vxeβ c] 7→[[[v]xβ ]···xβ [cl+1]···[ck]] and [u]v = [u]|[vc] 7→[[[v][cl+1]]···[ck]].
Now, we use induction on |u|. If |u| = 1, then this is a trivial case. Suppose that |u| > 1
and |v| > 1. Then, by (i),
u = avcd, u′ = a′uv′uc
′ud
′u
and by induction,
[u′]v′u = a′u[v′u]c
′ud
′u +
∑i∈I1
αiai′u[v
′u]bi
′u,
where each ai′uv
′ubi
′u < u′. Now, it is easy to check that
[[xixβ] · · ·xβ] =∑m≥0
(−1)m
(j
m
)xm
β xixj−mβ and xm
β xixj−mβ < xix
jβ (m > 0).
Now, by substitution xji 7→ [[xixβ] · · ·xβ], we obtain
[u]v = a[v]cd+∑i∈I2
αiai[v]bi,
where each aivbi < avcd. Also, by substitution [v] 7→ [[[v]xβ] · · ·xβ], we have
[u]v = a[v]xeβ cd+
∑j∈I
βjaj[v]bj = a[v]b+∑j∈I
βjaj[v]bj,
where each ajvbj < avb.
Remark. By the proof of Theorem 5.29, if we consider [u]v as a polynomial in k〈X〉,then for any w ∈ supp([u]v), cont(w) = cont(u).
Definition 5.30 Let S ⊂ Lie(X) with each s ∈ S monic, a, b ∈ X∗ and s ∈ S. If asb
is an ALSW, then we call [asb]s = [asb]s|[s] 7→s a special normal S-word (or special normal
s-word) while [asb]s is called a relative nonassociative Lyndon-Shirshov word, denoted by
RNLSW, where [asb]s is defined by (5.1) (see Theorem 5.29).
A Lie S-word (asb) is called a normal S-word if (asb) = asb.
Clearly, any special normal s-word is a normal s-word.
Corollary 5.31 Let u, v be ALSW’s, f ∈ Lie(X), f = v and u = avb, a, b ∈ X∗. Then,
for the normal f -word [afb]v = [avb]v|[v] 7→f , we have
[afb]v = afb+∑
i
αiaifbi,
where each αi ∈ k, ai, bi ∈ X∗, aif bi < u.
78
§5.3 Composition-Diamond lemma for Lie algebras
In this section, we give the Composition-Diamond lemma for Lie algebras.
Throughout this section, we extend the lexicographic ordering on X∗ mentioned in
section §5.2 to the deg-lex ordering < on X∗.
Lemma 5.32 Let ac, cb be ALSW’s, where a, b, c ∈ X∗ and c 6= 1. Then w = acb is also
an ALSW.
Proof. We use induction on |w| = n. If n = 3, then w = xixjxk is an ALSW, because
ac and cb are ALSW’s implies that xi > xj > xk. In the inductive case n > 3, suppose
min(w) = xβ, b = xeβ b, e ≥ 0, fir(b) 6= xβ and c = cxe
β. Then
w = acb and w′ = a′wc′wb
′w.
It is clear that a′wc′w, c
′wb
′w are ALSW’s. By induction, w′ is an ALSW and so is w.
Definition 5.33 Let f and g be two monic Lie polynomials in Lie(X) ⊂ k〈X〉. Then,
there are two kinds of Lie compositions:
(i) If w = f = agb for some a, b ∈ X∗, then the polynomial 〈f, g〉w = f − [agb]g is called
the composition of inclusion of f and g with respect to w.
(ii) If w is a word such that w = f b = ag for some a, b ∈ X∗ with deg(f)+deg(g) >deg(w),
then the polynomial 〈f, g〉w = [fb]f − [ag]g is called the composition of intersection of
f and g with respect to w.
By Lemma 5.32, in the Definition 5.33 (i) and (ii), w is an ALSW.
Definition 5.34 Let S ⊂ Lie(X) be a nonempty subset, h a Lie polynomial and w ∈X∗. We shall say that h is trivial modulo (S,w), denoted by h ≡Lie 0 mod(S,w), if
h =∑i
αi[aisibi]si, where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi]si
is a RNLSW and
aisibi < w.
Definition 5.35 Let S ⊂ Lie(X) be a nonempty set of monic Lie polynomials. Then S
is called a Grobner-Shirshov set (basis) in Lie(X) if any composition 〈f, g〉w with f, g ∈ Sis trivial modulo (S,w), i.e., 〈f, g〉w ≡Lie 0 mod(S,w).
Lemma 5.36 Let f, g be monic Lie polynomials. Then
〈f, g〉w − (f, g)w ≡ass 0 mod(f, g, w).
79
Proof. If 〈f, g〉w and (f, g)w are compositions of intersection, where w = f b = ag, then,
by Corollary 5.31, we may assume that
〈f, g〉w = [fb]f − [ag]g = fb+∑I1
αiaifbi − ag −∑I2
βjajgbj,
where aif bi, aj gbj < fb = ag = w. It follows that
〈f, g〉w − (f, g)w ≡ass 0 mod(f, g, w).
Similarly, for the case of the compositions of inclusion, we have the same conclusion.
Theorem 5.37 Let S ⊂ Lie(X) ⊂ k〈X〉 be a nonempty set of monic Lie polynomials.
Then S is a Grobner-Shirshov basis in Lie(X) if and only if S is a Grobner-Shirshov
basis in k〈X〉.
Proof. Note that, by the definitions, for any f, g ∈ S, they have composition in Lie(X)
if and only if so do in k〈X〉.Suppose that S is a Grobner-Shirshov basis in Lie(X). Then, for any composition
〈f, g〉w, we have
〈f, g〉w =∑I1
αi[aisibi]si,
where [aisibi]siare RNLSW’s and aisibi < w. By Corollary 5.31,
〈f, g〉w =∑I2
βjcjsjdj,
where each cj sjdj < w. Thus, by Lemma 5.36, we get
(f, g)w ≡ass 0 mod(S,w).
Hence, S is a Grobner-Shirshov basis in k〈X〉.Conversely, assume that S is a Grobner-Shirshov basis in k〈X〉. Then, for any
composition 〈f, g〉w in S, by Lemma 5.36, we obtain
〈f, g〉w ≡ass (f, g)w ≡ass 0 mod(S,w).
Therefore, we can assume by Theorem 4.4 that
〈f, g〉w =∑I1
αiaisibi,
80
where aisibi < w and w > a1s1b1 > a2s2b2 > . . .. By noting that 〈f, g〉w ∈ Lie(X), 〈f, g〉w =
a1s1b1 is an ALSW which shows that [a1s1b1]s1 is a RNLSW. Let h1 = 〈f, g〉w−α1[a1s1b1]s1 .
Clearly, h1 < 〈f, g〉w. Then, by Corollary 5.31, we have
h1 ≡ass 0 mod(S,w).
Now, by induction on 〈f, g〉w, we have
〈f, g〉w =∑I2
αi[cisidi]si,
where each [cisidi]siis a RNLSW and cisidi < w. This proves that S is a Grobner-Shirshov
basis in Lie(X).
Lemma 5.38 Let S ⊂ Lie(X) with each s ∈ S monic. Let
Irr(S) = [u] | [u] is a NLSW, u 6= asb, s ∈ S, a, b ∈ X∗.
Then, for any h ∈ Lie(X), h has a representation:
h =∑
[ui]∈Irr(S), ui≤h
αi[ui] +∑
sj∈S, aj sjbj≤h
βj[ajsjbj]sj.
Proof. We can assume that h =∑i
αi[ui], where each [ui] is a NLSW, 0 6= αi ∈ k and
u1 > u2 > · · · . If [u1] ∈ Irr(S), then let h1 = h − α1[u1]. If [u1] 6∈ Irr(S), then there
exists s ∈ S and a1, b1 ∈ X∗ such that u1 = a1s1b1. Now, let
h1 = h− α1[a1s1b1]s1 ∈ Lie(X).
Hence, in both cases, we have h1 < h. Now, the result follows from induction on h.
Theorem 5.39 (CD-lemma for Lie algebras over a field) Let S ⊂ Lie(X) ⊂ k〈X〉 be
nonempty set of monic Lie polynomials. Let IdLie(S) be the Lie-ideal of Lie(X) generated
by S. Then the following statements are equivalent.
(i) S is a Grobner-Shirshov basis in Lie(X).
(ii) f ∈ IdLie(S) =⇒ f = asb, for some s ∈ S and a, b ∈ X∗.
(ii’) f ∈ IdLie(S) =⇒ f = α1[a1s1b1]s1 +α2[a2s2b2]s2 + · · · , where αi ∈ k and f = a1s1b1 >
a2s2b2 > · · · .
(iii) Irr(S) = [u] | [u] is a NLSW, u 6= asb, s ∈ S, a, b ∈ X∗ is a k-basis for Lie(X|S).
81
Proof. (i) =⇒ (ii). By noting that IdLie(S) ⊆ Idass(S), where Idass(S) is the ideal of
k〈X〉 generated by S, and by using Theorems 5.37 and 4.4, the result follows.
(ii) =⇒ (iii). Suppose that∑
[ui]∈Irr(S)
αi[ui] = 0 in Lie(X|S) with u1 > u2 > · · · ,
that is,∑
[ui]∈Irr(S)
αi[ui] ∈ IdLie(S). Then each αi must be 0. Otherwise, say α1 6= 0. Then,
by (ii), we know that∑i
αi[ui] = u1 which implies that [u1] 6∈ Irr(S), a contradiction.
On the other hand, for any f ∈ Lie(X), by Lemma 5.38, we have
f + IdLie(S) =∑
i
αi([ui] + IdLie(S)).
(iii) =⇒ (i). For any composition 〈f, g〉w with f, g ∈ S, we have 〈f, g〉w ∈ IdLie(S).
Then, by (iii) and by Lemma 5.38,
〈f, g〉w =∑
βj[ajsjbj]sj,
where each βj ∈ k, [ajsjbj]sjis normal S-word and ajsjbj < w. This proves that S is a
Grobner-Shirshov basis in Lie(X).
(ii)⇐⇒ (ii′). This part is clear.
Example 5.40 (Shirshov) For an arbitrary polynomial f ∈ Lie(X), it is clear that fis a Grobner-Shirshov basis in Lie(X). From this it follows that the word problem is
solvable for each one-relator Lie algebra Lie(X|f).
§5.4 Applications
§5.4.1 Kukin’s construction of a Lie algebra with unsolvable word problem
Let P = sgp〈x, y|ui = vi, i ∈ I〉 be a semigroup. Consider the Lie algebra
AP = Lie(x, x, y, y, z|S),
where S consists of the following relations:
1) [xx] = 0, [xy] = 0, [yx] = 0, [yy] = 0,
2) [xz] = −[zx], [yz] = −[zy],
3) bzuic = bzvic, i ∈ I.
Here, bzuc means the left normed bracketing.
In this section, we give a Grobner-Shirshov basis for Lie algebra AP and by using
this result we give another proof for Kukin’s theorem, see Corollary 5.42.
82
Let the ordering x > y > z > x > y and > the deg-lex ordering on x, y, x, y, z∗.Let ρ be the congruence on x, y∗ generated by (ui, vi), i ∈ I. Let
3)′ bzuc = bzvc, (u, v) ∈ ρ with u > v.
Theorem 5.41 With the above notation, the set S1 = 1), 2), 3)′ is a Grobner-Shirshov
basis in Lie(x, y, x, y, z).
Proof. For any u ∈ x, y∗, by induction on |u|, bzuc = zu. All possible compositions
in S1 are intersection of 2) and 3)′, and inclusion of 3)′ and 3)′.
For 2)∧ 3)′, w = xzu, (u, v) ∈ ρ, u > v, f = [xz] + [zx], g = bzuc − bzvc. We have
([xz] + [zx], bzuc − bzvc)w = [fu]f − [xg]g
≡ b([xz] + [zx])uc − [x(bzuc − bzvc)]
≡ bzxuc+ bxzvc ≡ bzxuc − bzxvc ≡ 0 mod(S1, w).
For 3)′ ∧ 3)′, w = zu1 = zu2e, e ∈ x, y∗, (ui, vi) ∈ ρ, ui > vi, i = 1, 2. We have
(bzu1c − bzv1c, bzu2c − bzv2c)w ≡ (bzu1c − bzv1c)− b(bzu2c − bzv2c)ec
≡ bbzv2cec − bzv1c ≡ bzv2ec − bzv1c ≡ 0 mod(S1, w).
Thus, the set S1 = 1), 2), 3)′ is a Grobner-Shirshov basis in Lie(x, y, x, y, z).
Corollary 5.42 (Kukin [138]) Let u, v ∈ x, y∗. Then
u = v in the semigroup P ⇔ bzuc = bzvc in the Lie algebra AP .
Proof. Suppose that u = v in the semigroup P . Without loss of generality, we
may assume that u = au1b, v = av1b for some a, b ∈ x, y∗ and (u1, v1) ∈ ρ. For
any r ∈ x, y, by the relations 1), we have [xr] = 0 and so bzxcc = b[zx]cc =
[bzccx], bzycc = [bzccy] for any c ∈ x, y∗. From this it follows that in AP , bzuc =
bzau1bc = bbzau1cbc = bbzu1←−a cbc = bzu1
←−a bc = bzv1←−a bc = bzav1bc = bzvc, where
←−−−−−−−−xi1xi2 · · ·xin = xinxin−1 · · ·xi1 and xi1xi2 · · ·xin = xi1 xi2 · · · xin , xij ∈ x, y. Moreover,
3)′ holds in AP .
Suppose that bzuc = bzvc in the Lie algebra AP . Then both bzuc and bzvc have
the same normal form in AP . Since S1 is a Grobner-Shirshov basis in AP by Theorem
5.41, both bzuc and bzvc can be reduced to the same normal form of the form bzcc for
some c ∈ x, y∗ only by the relations 3)′. This implies that in P , u = c = v.
By the above corollary, if the semigroup P has the undecidable word problem then
so does the Lie algebra AP .
83
§5.4.2 Grobner-Shirshov basis for the Drinfeld-Kohno Lie algebra Ln
In this section we give a Grobner-Shirshov basis for the Drinfeld-Kohno Lie algebra
Ln.
Definition 5.43 Let n > 2 be an integer. The Drinfeld-Kohno Lie algebra Ln over Z is
defined by generators tij = tji for distinct indices 1 ≤ i, j ≤ n− 1, and relations
tijtkl = 0,
tij(tik + tjk) = 0,
where i, j, k, l are distinct.
Clearly, Ln has a presentation LieZ(T |S), where T = tij| 1 ≤ i < j ≤ n− 1 and S
consists of the following relations
tijtkl = 0, k < i < j, k < l, l 6= i, j (5.2)
tjktij + tiktij = 0, i < j < k (5.3)
tjktik − tiktij = 0, i < j < k (5.4)
Now we order T : tij < tkl if either i < k or i = k and j < l. Let < be the deg-lex
ordering on T ∗.
Theorem 5.44 Let S = (5.2), (5.3), (5.4) be as before, < the deg-lex ordering on T ∗.
Then S is a Grobner-Shirshov basis for Ln.
Proof. We list all the possible ambiguities. Denote (i)∧ (j) the composition of the type
(i) and type (j). For convenience, we denote (5.2), (5.3), (5.4) by (1), (2), (3) respectively.
For (1) ∧ (n), 1 ≤ n ≤ 3, the possible ambiguities w’s are:
(1) ∧ (1) tijtkltmr, (k < i < j, k < l, l 6= i, j, m < k < l, m < r, r 6= k, l),
(1) ∧ (2) tijtkltmk, ( k < i < j, k < l, l 6= i, j, m < k < l),
(1) ∧ (3) tijtkltml, (k < i < j, k < l, l 6= i, j, m < k < l).
For (2) ∧ (n), 1 ≤ n ≤ 3, the possible ambiguities w’s are:
(2) ∧ (1) tjktijtmr, ( m < i < j < k, m < r, r 6= i, j),
(2) ∧ (2) tjktijtmi, ( m < i < j < k),
(2) ∧ (3) tjktijtmj, ( m < i < j < k).
84
For (3) ∧ (n), 1 ≤ n ≤ 3, the possible ambiguities w’s are:
(3) ∧ (1) tjktiktmr, ( m < i < j < k, m < r, r 6= i, k),
(3) ∧ (2) tjktiktmi, ( m < i < j < k),
(3) ∧ (3) tjktiktmk, ( m < i < j < k).
We claim that all compositions are trivial relative to S.
Here, we only prove cases (1)∧ (1), (1)∧ (2), (2)∧ (1), (2)∧ (2), and the other cases
can be proved similarly.
For (1) ∧ (1), let f = tijtkl, g = tkltmr, k < i < j, k < l, l 6= i, j, m < k < l, m <
r, r 6= k, l. Then w = tijtkltmr and
(f, g)w = (tijtkl)tmr − tij(tkltmr)
= (tijtmr)tkl mod(S,w).
There are three subcases to consider: r 6= i, j, r = i, r = j.
Subcase 1. If r 6= i, j, then
(tijtmr)tkl ≡ 0 mod(S,w).
Subcase 2. If r = i, then
(tijtmr)tkl = (tijtmi)tkl
≡ tkl(tmjtmi)
≡ (tkltmj)tmi + tmj(tkltmi)
≡ 0 mod(S,w).
Subcase 3. If r = j, then
(tijtmr)tkl = (tijtmj)tkl
≡ −tkl(tmjtmi)
≡ −(tkltmj)tmi − tmj(tkltmi)
≡ 0 mod(S,w).
For (1)∧ (2), let f = tijtkl, g = tkltmk + tmltmk, k < i < j, k < l, l 6= i, j, m < k <
l. Then w = tijtkltmk and
(f, g)w = (tijtkl)tmk − tij(tkltmk + tmltmk)
= (tijtmk)tkl − tij(tmltmk)
≡ −tij(tmltmk)
≡ −(tijtml)tmk − tml(tijtmk)
≡ 0 mod(S,w).
85
For (2)∧(1), let f = tjktij + tiktij, g = tijtmr, m < i < j < k, m < r, r 6= i, j. Then
w = tjktijtmr and
(f, g)w = (tjktij + tiktij)tmr − tjk(tijtmr)
≡ (tjktmr)tij + (tiktmr)tij mod(S,w).
There are two subcases to consider: r 6= k, r = k.
Subcase 1. If r 6= k, then
(tjktmr)tij + (tiktmr)tij ≡ 0 mod(S,w).
Subcase 2. If r = k, then
(tjktmr)tij + (tiktmr)tij = (tjktmk)tij + (tiktmk)tij
≡ −tij(tmktmj)− tij(tmktmi)
≡ (tijtmj)tmk + (tijtmi)tmk
≡ −tmk(tmjtmi) + tmk(tmjtmi)
≡ 0 mod(S,w).
For (2) ∧ (2), let f = tjktij + tiktij, g = tijtmi + tmjtmi, m < i < j < k. Then
w = tjktijtmi and
(f, g)w = (tjktij + tiktij)tmi − tjk(tijtmi + tmjtmi)
= (tjktmi)tij + (tiktmi)tij + tik(tijtmi)− tjk(tmjtmi)
≡ tij(tmktmi)− tik(tmjtmi)− tjk(tmjtmi)
≡ −(tijtmi)tmk + (tiktmi)tmj − (tjktmj)tmi
≡ −tmk(tmjtmi)− (tmktmi)tmj + (tmktmj)tmi
≡ 0 mod(S,w).
So S is a Grobner-Shirshov basis for Ln.
Let L be a Lie algebra over a commutative ring K, L1 an ideal of L and L2 a
subalgebra of L. Then we call L a semidirect product of L1 and L2 if L = L1 ⊕ L2 as
K-modules.
By Theorems 5.39 and 5.44, we have immediately the following corollaries.
Corollary 5.45 The Drinfeld-Kohno Lie algebra Ln is a free Z-module with a Z-basis
Irr(S) = [tik1tik2 · · · tikm ] | tik1tik2 · · · tikm is an ALSW in T ∗, m ∈ N.
86
Corollary 5.46 Ln is an iterated semidirect product of free Lie algebras.
Proof. Let Ai be the free Lie algebra generated by tij | i < j ≤ n− 1. Clearly,
Ln = A1 ⊕ A2 ⊕ · · · ⊕ An−2
as Z-modules, and from the relations (1), (2), (3), we have
Ai / Ai + Ai+1 + · · · + An−2.
87
Chapter 6 Grobner-Shirshov Bases for Lie Algebras
over a Commutative Algebra
In this chapter, we establish the Composition-Diamond lemma for Lie algebras over
a polynomial algebra, i.e., for “double free” Lie algebras. It provides a Grobner-Shirshov
bases theory for Lie algebras over a commutative algebra.
Let k be a field, K a commutative associative k-algebra with identity, and L a Lie
K-algebra. Let LieK(X) be the free Lie K-algebra generated by a set X. Then, of course,
L can be presented as K-algebra by generators X and some defining relations S,
L = LieK(X|S) = LieK(X)/Id(S).
In order to define a Grobner-Shirshov basis for L, we first present K in a form
K = k[Y |R] = k[Y ]/Id(R),
where k[Y ] is a polynomial algebra over the field k, R ⊂ k[Y ]. Then the Lie K-algebra Lhas the following presentation as a k[Y ]-algebra
L = Liek[Y ](X|S,Rx, x ∈ X)
Now by definition, a Grobner-Shirshov basis for L = LieK(X|S) is Grobner-Shirshov
basis of the ideal Id(S,Rx, x ∈ X) in the “double free” Lie algebra Liek[Y ](X).
As an application of Composition-Diamond lemma (Theorem 6.14), a Grobner-Shirshov
basis of L gives rise to a linear basis of L as a k-algebra.
We give applications of Grobner-Shirshov bases theory for Lie algebras over a com-
mutative algebra K (over a field k) to the Poincare-Birkhoff-Witt theorem. A Lie algebra
over a commutative ring is called special if it is embeddable into an (universal enveloping)
associative algebra. Otherwise it is called non-special. There are known classical exam-
ples by A.I. Shirshov [196] and P. Cartier [67] of Lie algebras over commutative algebras
over GF (2) that are not embeddable into associative algebras. Shirshov and Cartier used
ad hoc methods to prove that some elements of corresponding Lie algebras are not zero
though they are zero in the universal enveloping algebras, i.e., they proved non-speciality
of the examples. Here we find Grobner-Shirshov bases of these Lie algebras and then
use Composition-Diamond lemma (Theorem 6.14) to get the result, i.e., we give a new
conceptual proof.
P.M. Cohn [93] gave the following examples of Lie algebras
Lp = LieK(x1, x2, x3|y3x3 = y2x2 + y1x1)
88
over truncated polynomial algebras
K = k[y1, y2, y3|ypi = 0, 1 ≤ i ≤ 3],
where k is a filed of characteristic p > 0. He conjectured that Lp is non-special Lie algebra
for any p. Lp is called the Cohn’s Lie algebra. Using Theorem 6.14 we will prove that L2,
L3 and L5 are non-special Lie algebras. We present an algorithm that one can check for
any p, whether Cohn’s Lie algebras is non-special.
We give new class of special Lie algebras in terms of defining relations (Theorem ??).
For example, any one relator Lie algebra LieK(X|f) with a k[Y ]-monic relation f over
a commutative algebra K is special (Corollary ??). It gives an extension of the list of
known special Lie algebras (ones with valid PBW Theorems) (see P.-P. Grivel [107]). Let
us give this list:
1. L is a free K-module (G. Birkhoff [11], E. Witt [213]),
2. K is a principal ideal domain (M. Lazard [143, 144]),
3. K is a Dedekind domain (P. Cartier [67]),
4. K is over a field k of characteristic 0 (P.M. Cohn [93]),
5. L is K-module without torsion (P.M. Cohn [93]),
6. 2 is invertible in K and for any x, y, z ∈ L, [x[yz]] = 0 (Y. Nouaze and P. Revoy
[168]),
7. No nonzero element of K annihilates an element of the center of LK (A.I. Shirshov
[196]).
As a last application we prove that every finitely or countably generated Lie algebra
over an arbitrary commutative algebra K can be embedded into a two-generated Lie
algebra over K.
§6.1 Preliminaries
We start with some concepts and results from the literature concerning with the
Grobner-Shirshov bases theory of a free Lie algebra Liek(X) generated by X over a field
k.
Let X = xi|i ∈ I be a well-ordered set with xi > xj if i > j for any i, j ∈ I. Let
X∗ be the free monoid generated by X. For u = xi1xi2 · · ·xim ∈ X∗, let the length of u
be m, denoted by |u| = m.
89
We use two linear orderings on X∗:
(i) (lex-ordering) 1 > t if t 6= 1 and, by induction, if u = xiu′ and v = xjv
′ then
u > v if and only if xi > xj or xi = xj and u′ > v′;
(ii) (deg-lex ordering) u v if |u| > |v|, or |u| = |v| and u > v.
We regard Liek(X) as the Lie subalgebra of the free associative algebra k〈X〉, which
is generated by X under the Lie bracket [u, v] = uv − vu. Given f ∈ k〈X〉, denote by f
the leading word of f with respect to the deg-lex ordering; f is monic if the coefficient of
f is 1.
We denote the set of all NLSW’s on X by NLSW (X).
In fact, NLSW’s may be defined as Hall–Shirshov words relative to lex-deg ordering.
Considering any NLSW [w] as a polynomial in k〈X〉, we have [w] = w (see Lemma
5.24). This implies that if f ∈ Liek(X) ⊂ k〈X〉 then f is an ALSW.
Lemma 6.1 Let w = aub be as in Theorem 5.29. Then [uc] = [u[c1][c2] . . . [cn]], that is
[w] = [a[. . . [u[c1]] . . . [cn]]d].
Lemma 6.2 Suppose that w = aubvc, where w, u, v ∈ ALSW (X). Then there is some
bracketing
[w]u,v = [a[u]b[v]d]
in the word w such that
[w]u,v = w.
More precisely,
[w]u,v =
[a[up]uq[vs]vl] if [w] = [a[up]q[vs]l],
[a[u[c1] · · · [ct]v · · · [cn]]up] if [w] = [a[u[c1] · · · [ct] · · · [cn]]p] with v a subword of ct.
§6.2 Composition-Diamond lemma for Liek[Y ](X)
Let Y = yj|j ∈ J be a well-ordered set and [Y ] = yj1yj2 · · · yjl|yj1 ≤ yj2 ≤ · · · ≤
yjl, l ≥ 0 the free commutative monoid generated by Y . Then [Y ] is a k-linear basis of
the polynomial algebra k[Y ].
Let the set X be a well-ordered set, and let the lex-ordering < and the deg-lex
ordering ≺X on X∗ be defined as before.
Let Liek[Y ](X) be the “double” free Lie algebra, i.e., the free Lie algebra over the
polynomial algebra k[Y ] with generating set X.
90
From now on we regard Liek[Y ](X) ∼= k[Y ] ⊗ Liek(X) as the Lie subalgebra of
k[Y ]〈X〉 ∼= k[Y ]⊗ k〈X〉 the free associative algebra over polynomial algebra k[Y ], which
is generated by X under the Lie bracket [u, v] = uv − vu.Let
TA = u = uY uX |uY ∈ [Y ], uX ∈ ALSW (X)
and
TN = [u] = uY [uX ]|uY ∈ [Y ], [uX ] ∈ NLSW (X).
By Chapter 5, we know that the elements of TA and TN are one-to-one corresponding
to each other.
Remark: For u = uY uX ∈ TA, we still use the notation [u] = uY [uX ] where [uX ] is a
NLSW on X.
Let kTN be the linear space spanned by TN over k. For any [u], [v] ∈ TN , define
[u][v] =∑
αiuY vY [wX
i ]
where αi ∈ k, [wXi ]’s are NLSW’s and [uX ][vX ] =
∑αi[w
Xi ] in Liek(X).
Then k[Y ]⊗ Liek(X) ∼= kTN as k-algebra and TN is a k-basis of k[Y ]⊗ Liek(X).
We define the deg-lex ordering on
[Y ]X∗ = uY uX |uY ∈ [Y ], uX ∈ X∗
as follows: for any u, v ∈ [Y ]X∗,
u v if (uX X vX) or (uX = vX and uY Y vY ),
where Y and X are the deg-lex ordering on [Y ] and X∗ respectively.
Remark: By abuse of the notation, from now on, in a Lie expression like [[u][v]] we will
omit the external brackets, [[u][v]] = [u][v].
Clearly, the ordering is “monomial” in a sense of [u][w] [v][w] whenever wX 6= uX
for any u, v, w ∈ TA.
Considering any [u] ∈ TN as a polynomial in k-algebra k[Y ]〈X〉, we have [u] = u ∈ TA.
For any f ∈ Liek[Y ](X) ⊂ k[Y ]⊗ k〈X〉, one can present f as a k-linear combination
of TN -words, i.e., f =∑αi[ui], where [ui] ∈ TN . With respect to the ordering on
[Y ]X∗, the leading word f of f in k[Y ]〈X〉 is an element of TA. We call f k-monic if the
coefficient of f is 1. On the other hand, f can be presented as k[Y ]-linear combinations
of NLSW (X), i.e., f =∑fi(Y )[uX
i ], where fi(Y ) ∈ k[Y ], [uXi ] ∈ NLSW (X) and
uX1 X uX
2 X . . .. Clearly fX = uX1 and fY = f1(Y ). We call f k[Y ]-monic if the
f1(Y ) = 1. It is easy to see that k[Y ]-monic implies k-monic.
Equipping with the above concepts, we rewrite Lemma 5.24 as follows.
91
Lemma 6.3 (Shirshov) Suppose that w = aub where w, u ∈ TA and a, b ∈ X∗. Then
[w] = [a[uc]d],
where [uc] ∈ TN and b = cd.
Represent c in a form c = c1c2 . . . cn, where c1, . . . , cn ∈ ALSW (X) and c1 ≤ c2 ≤. . . ≤ cn. Then
[w] = [a[u[c1][c2] . . . [cn]]d].
Moreover, the leading word of [w]u = [a[· · · [[[u][c1]][c2]] . . . [cn]]d] is exactly w, i.e.,
[w]u = w.
We still use the notion [w]u as the special bracketing of w relative to u.
Let S ⊂ Liek[Y ](X) and Id(S) be the k[Y ]-ideal of Liek[Y ](X) generated by S. Then
any element of Id(S) is a k[Y ]-linear combination of polynomials of the following form:
(u)s = [c1][c2] · · · [cn]s[d1][d2] · · · [dm], m, n ≥ 0
with some placement of parentheses, where s ∈ S and ci, dj ∈ ALSW (X). We call such
(u)s an s-word (or S-word).
Now, we define two special kinds of S-words.
Definition 6.4 Let S ⊂ Liek[Y ](X) be a k-monic subset, a, b ∈ X∗ and s ∈ S. If
asb ∈ TA, then by Lemma 6.3 we have the special bracketing [asb]s of asb relative to s.
We define [asb]s = [asb]s|[s] 7→s to be a special normal s-word (or special normal S-word).
Definition 6.5 Let S ⊂ Liek[Y ](X) be a k-monic subset and s ∈ S. We define the
normal s-word, denoted by bucs, where u = asb, a, b ∈ X∗ (u is an associative S-word),
inductively.
(i) s is normal of s-length 1;
(ii) If bucs is normal with s-length k and [v] ∈ NLSW (X) such that |v| = l, then [v]bucswhen v > buc
X
s and bucs[v] when v < bucX
s are normal of s-length k + l.
From the definition of the normal s-word, we have the following lemma.
Lemma 6.6 For any normal s-word bucs = (asb), a, b ∈ X∗, we have bucs = asb ∈ TA.
92
Remark: It is clear that for an s-word (u)s = [c1][c2] · · · [cn]s[d1][d2] · · · [dm], (u)s is
normal if and only if (u)s = c1c2 · · · cnsd1d2 · · · dm.
Now we give the definition of compositions.
Definition 6.7 Let f, g be two k-monic polynomials of Liek[Y ](X). Denote the least com-
mon multiple of fY and gY in [Y ] by L = lcm(fY , gY ).
If gX is a subword of fX , i.e., fX = agXb for some a, b ∈ X∗, then the polynomial
C1〈f, g〉w =L
fYf − L
gY[agb]g
is called the inclusion composition of f and g with respect to w, where w = LfX = LagXb.
If a proper prefix of gX is a proper suffix of fX , i.e., fX = aa0, gX = a0b, a, b, a0 6= 1,
then the polynomial
C2〈f, g〉w =L
fY[fb]f −
L
gY[ag]g
is called the intersection composition of f and g with respect to w, where w = LfXb =
LagX .
If the greatest common divisor of fY and gY in [Y ] is not 1, then for any a, b, c ∈ X∗
such that w = LafXbgXc ∈ TA, the polynomial
C3〈f, g〉w =L
fY[afbgXc]f −
L
gY[afXbgc]g
is called the external composition of f and g with respect to w.
If fY 6= 1, then for any special normal f -word [afb]f , a, b ∈ X∗, the polynomial
C4〈f〉w = [afXb][afb]f
is called the multiplication composition of f with respect to w, where w = afXbafb.
Immediately, we have that Ci〈−〉w ≺ w, i ∈ 1, 2, 3, 4.
Remarks
1) When Y = ∅, there are no external and multiplication compositions. This is the case
of Shirshov’s compositions over a field.
2) In the cases of C1 and C2, the corresponding w ∈ TA by the property of ALSW’s, but
in the case of C4, w 6∈ TA.
93
3) For any fixed f, g, there are finitely many compositions C1〈f, g〉w, C2〈f, g〉w, but
infinitely many C3〈f, g〉w, C4〈f〉w.
Definition 6.8 Given a k-monic subset S ⊂ Liek[Y ](X) and w ∈ [Y ]X∗ (not nec-
essary in TA), an element h ∈ Liek[Y ](X) is called trivial modulo (S,w), denoted by
h ≡ 0 mod(S,w), if h can be presented as a k[Y ]-linear combination of special normal
S-words with leading words less than w, i.e., h =∑
i αiβi[aisibi]si, where αi ∈ k, βi ∈ [Y ],
ai, bi ∈ X∗, si ∈ S, and βiaisibi ≺ w.
In general, for p, q ∈ Liek[Y ](X), we write p ≡ q mod(S,w) if p− q ≡ 0 mod(S,w).
S is a Grobner-Shirshov basis in Liek[Y ](X) if all the possible compositions of ele-
ments in S are trivial modulo S and corresponding w.
If a subset S of Liek[Y ](X) is not a Grobner-Shirshov basis then one can add all
nontrivial compositions of polynomials of S to S. Continuing this process repeatedly,
we finally obtain a Grobner-Shirshov basis Sc that contains S. Such a process is called
Shirshov’s algorithm. Sc is called Grobner-Shirshov complement of S.
Lemma 6.9 Let f be a k-monic polynomial in Liek[Y ](X). If fY = 1 or f = gf ′ where
g ∈ k[Y ] and f ′ ∈ Liek(X), then for any special normal f -word [afb]f , a, b ∈ X∗,
(u)f = [afXb][afb]f has a presentation:
(u)f = [afXb][afb]f =∑
buicf(u)f
αiβibuicf
where αi ∈ k, βi ∈ [Y ].
Proof. Case 1. fY = 1, i.e., f = fX . By Lemma 6.3 and since ≺ is monomial, we have
[afb] = [afb]f −∑
βivi≺afb αiβi[vi], where αi ∈ k, βi ∈ [Y ], vi ∈ ALSW (X). Then
(u)f = [afb][afb]f = [afb]f [afb]f +∑
βivi≺afb
αiβi[afb]f [vi] =∑
βivi≺afb
αiβi[afb]f [vi].
The result follows since vi ≺ afb and each [afb]f [vi] is normal.
Case 2. f = gf ′, i.e., fX = f ′. Then we have
(u)f = [af ′b][afb]f = g([af ′b][af ′b]f ′).
The result follows from Case 1.
The following lemma plays a key role.
94
Lemma 6.10 Let S be a k-monic subset of Liek[Y ](X) in which each multiplication com-
position is trivial. Then for any normal s-word bucs = (asb) and w = asb = bucs, where
a, b ∈ X∗, we have
bucs ≡ [asb]s mod(S,w).
Proof. For w = s the lemma is clear.
For w 6= s, since either bucs = (asb) = [a1](a2sb) or bucs = (asb) = (asb1)[b2], there
are two cases to consider.
Let
δ(asb) =
|a1| if (asb) = [a1](a2sb),
s-length of (asb1) if (asb) = (asb1)[b2].
The proof will be proceeding by induction on (w, δ(asb)), where (w′,m′) < (w,m)⇔ w ≺w′ or w = w′, m′ < m (w,w′ ∈ TA,m,m
′ ∈ N).
Case 1. bucs = (asb) = [a1](a2sb), where a1 > a2sXb, a = a1a2 and (a2sb) is normal
s-word. In this case, (w, δ(asb)) = (w, |a1|).Since w = asb = a1a2sb a2sb, by induction, we may assume that (a2sb) = [a2sb]s +∑
αiβi[cisidi]si, where βicisidi ≺ a2sb, a1, a2, ci, di ∈ X∗, si ∈ S, αi ∈ k and βi ∈ [Y ].
Thus,
bucs = (asb) = [a1][a2sb]s +∑
αiβi[a1][cisidi]si.
Consider the term [a1][cisidi]si.
If a1 > cisiXdi, then [a1][cisidi]si
is normal s-word with a1cisidi ≺ w. Note that
βia1cisidi ≺ w, then by induction, βi[a1][cisidi]si≡ 0 mod(S,w).
If a1 < cisiXdi, then [a1][cisidi]si
= −[cisidi]si[a1] and [cisidi]si
[a1] is normal s-word
with βicisidia1 ≺ βia2sba1 ≺ βia1a2sb = w.
If a1 = cisiXdi, then there are two possibilities. For si
Y = 1, by Lemma 6.9 and by
induction on w we have βi[a1][cisidi]si≡ 0 mod(S,w). For si
Y 6= 1, [a1][cisidi]siis the
multiplication composition, then by assumption, it is trivial mod(S,w).
This shows that in any case, βi[a1][cisidi]siis a linear combination of normal s-words
with leading words less than w, i.e., βi[a1][cisidi]si≡ 0 mod(S,w) for all i.
Therefore, we may assume that bucs = (asb) = [a1][a2sb]s and a1 > wX > a2sXb.
If either |a1| = 1 or [a1] = [[a11][a12]] and a12 ≤ a2sXb, then bucs = [a1][a2sb]s is
already a normal s-word, i.e., bucs = [a1][a2sb]s = [a1a2sb]s = [asb]s.
If [a1] = [[a11][a12]] and a12 > a2sXb, then
bucs = [a1][a2sb]s = [[a11][a12]][a2sb]s = [a11][[a12][a2sb]s] + [[a11][a2sb]s][a12].
95
Let us consider the second summand [[a11][a2sb]s][a12]. Then by induction on w and
by noting that [a11][a2sb]s is normal, we may assume that [a11][a2sb]s =∑αiβi[cisidi]si
,
where βicisidi a11a2sb, si ∈ S, αi ∈ k, βi ∈ [Y ], ci, di ∈ X∗. Thus,
[[a11][a2sb]s][a12] =∑
αiβi[cisidi]si[a12],
where a11 > a12 > a2sXb, w = a11a12a2sb.
If a12 < cisiXdi, then [cisidi]si
[a12] is normal with w′ = βicisidia12 βia11a2sba12 ≺w. By induction, βi[cisidi]si
[a12] ≡ 0 mod(S,w).
If a12 > cisiXdi, then [cisidi]si
[a12] = −[a12][cisidi]siand [a12][cisidi]si
is normal with
w′ = βia12cisidi βia12a11a2sb ≺ w. Again we can apply the induction.
If a12 = cisiXdi, then as discussed above, it is either the case in Lemma 6.9 or the
multiplication composition and each is trivial mod(S,w).
These show that [[a11][a2sb]s][a12] ≡ 0 mod(S,w).
Hence,
bucs ≡ [a11][[a12][a2sb]s] mod(S,w).
where a11 > a12 > a2sXb.
Noting that [a11][[a12][a2sb]s] is normal and now (w, δ[a11][[a12][a2sb]s]) = (w, |a11|) <(w, |a1|), the result follows by induction.
Case 2. bucs = (asb) = (asb1)[b2] where asXb1 > b2, b = b1b2 and (asb1) is normal
s-word. In this case, (w, δ(asb)) = (w,m) where m is the s-length of (asb1).
By induction on w, we may assume that
bucs = (asb) = [asb1]s[b2] +∑
αiβi[cisidi]si[b2].
where βicisidi ≺ asb1, si ∈ S, αi ∈ k, βi ∈ [Y ], ci, di ∈ X∗.
Consider the term βi[cisidi]si[b2] for each i.
If b2 < cisiXdi, then [cisidi]si
[b2] is normal s-word with βicisidib2 ≺ w.
If b2 > cisiXdi, then [cisidi]si
[b2] = −[b2][cisidi]siand [b2][cisidi]si
is normal s-word
with βib2cisidi ≺ βib2asb1 ≺ βiasb1b2 = w.
If b2 = cisiXdi, then as above, by Lemma 6.9 and induction on w or by assumption,
βi[cisidi]si[b2] ≡ 0 mod(S,w).
These show that for each i, βi[cisidi]si[b2] ≡ 0 mod(S,w).
Therefore, we may assume that bucs = (asb) = [asb1]s[b2], a, b ∈ X∗, where b = b1b2
and asXb1 > b2.
Noting that for [asb1]s = s or [asb1]s = [a1][a2sb1]s with a2sXb1 ≤ b2 or [asb1]s =
[asb11]s[b12] with b12 ≤ b2, bucs is already normal. Now we consider the remained cases.
96
Case 2.1. Let [asb1]s = [a1][a2sb1]s with a1 > a1a2sXb1 > a2s
Xb1 > b2. Then we have
bucs = [[a1][a2sb1]s][b2] = [[a1][b2]][a2sb1]s + [a1][[a2sb1]s[b2]].
We consider the term [[a1][b2]][a2sb1]s.
By noting that a1 > b2, we may assume that [a1][b2] =∑
uia1b2αi[ui] where αi ∈
k, ui ∈ ALSW (X). We will prove that [ui][a2sb1]s ≡ 0 mod(S,w).
If ui > a2sXb1, then [ui][a2sb1]s is normal s-word with w′ = uia2sb1 a1b2a2sb1 ≺
w = a1a2sb1b2.
If ui < a2sXb1, then [ui][a2sb1]s = −[a2sb1]s[ui] and [a2sb1]s[ui] is normal s-word with
w′ = a2sb1ui a2sb1a1b2 ≺ w, since a1a2sb1 is an ALSW.
If ui = a2sXb1, then as above, by Lemma 6.9 and induction on w or by assumption,
[ui][a2sb1]s ≡ 0 mod(S,w).
This shows that
bucs ≡ [a1][[a2sb1]s[b2]] mod(S,w).
By noting that a1 > a2sXb1 > b2, the result now follows from the Case 1.
Case 2.2. Let [asb1]s = [asb11]s[b12] with asXb11 > asXb11b12 > b12 > b2. Then we
have
bucs = [[asb11]s[b12]][b2] = [[asb11]s[b2]][b12] + [asb11]s[[b12][b2]].
Let us first deal with [[asb11]s[b2]][b12]. Since asb11b2 < asb11b12, we may apply induc-
tion on w and have that
[[asb11]s[b2]][b12] =∑
αiβi[cisidi]si[b12],
where βicisidi asb11b2, w = asb11b12b2.
If b12 < cisiXdi, then [cisidi]si
[b12] is normal s-word with w′ = βicisidib12 ≺ w.
If b12 > cisiXdi, then [cisidi]si
[b12] = −[b12][cisidi]siand [b12][cisidi]si
is a normal
s-word with w′ = βib12cisidi βib12asb11b2 ≺ asb11b12b2 = w.
If b12 = cisiXdi, then as above, by Lemma 6.9 and induction on w or by assumption,
βi[cisidi]si[b12] ≡ 0 mod(S,w).
These show that
bucs ≡ [asb11]s[[b12][b2]] mod(S,w).
Let [b12][b2] = [b12b2] +∑
ui≺a1b2αi[ui] where αi ∈ k, ui ∈ ALSW (X). By noting
that asXb11 > b12b2, we have [asb11]s[ui] ≡ 0 mod(S,w) for any i. Therefore,
bucs ≡ [asb11]s[b12b2] mod(S,w).
97
Noting that [asb11]s[b12b2] is normal and now (w, δ[asb11]s[b12b2]) < (w, δ[asb1]s[b2]), the
result follows by induction.
The proof is complete.
Lemma 6.11 Let S be a k-monic subset of Liek[Y ](X) in which each multiplication com-
position is trivial. Then any element of the k[Y ]-ideal generated by S can be written as a
k[Y ]-linear combination of normal S-words.
Proof. Note that for any h ∈ Id(S), h can be presented by a k[Y ]-linear combination
of S-words of the form
(u)s = [c1][c2] · · · [ck]s[d1][d2] · · · [dl] (6.1)
with some placement of parentheses, where s ∈ S, cj, dj ∈ ALSW (X), k, l ≥ 0. By
Lemma 6.10 it suffices to prove that (6.1) is a linear combination of normal S-words. We
will prove the result by induction on k + l. It is trivial when k + l = 0, i.e., (u)s = s.
Suppose that the result holds for k + l = n. Now let us consider
(u)s = [cn+1]([c1][c2] · · · [ck]s[d1][d2] · · · [dn−k]) = [cn+1](v)s.
By inductive hypothesis, we may assume without loss of generality that (v)s is a normal
s-word, i.e., (v)s = bvcs = (csd) where csd ∈ TA, c, d ∈ X∗. If cn+1 > csXd, then (u)s is
normal. If cn+1 < csXd then (u)s = −bvcs[cn+1] where bvcs[cn+1] is normal. If cn+1 = csXd
then by Lemma 6.10, (u)s = [cn+1](csd) ≡ [cn+1][csd]s. Now the result follows from the
multiplication composition and Lemma 6.9.
Lemma 6.12 Let S be a k-monic subset of Liek[Y ](X) in which each multiplication com-
position is trivial. Then for any normal S-word basbcs = [a1][a2] · · · [ak]bvcs[b1][b2] · · · [bl]with some placement of parentheses, the three following S-words are linear combinations
of normal S-words with the leading words less than asb:
(i) w1 = basbcs|[ai] 7→[c] where c ≺ ai;
(ii) w2 = basbcs|[bj ] 7→[d] where d ≺ bj;
(iii) w3 = basbcs|bvcs 7→bv′cs where bv′cs ≺ bvcs.
Proof. We first prove (iii). For k + l = 1, for example, basbcs = bvcs[b1], it is easy to
see that the result follows from Lemmas 6.11 and 6.9 since either bv′cs[b1] or [b1]bv′cs is
98
normal or w3 is the multiplication composition. Now the result follows by induction on
k + l.
We now prove (i), and (ii) is similar. For k+ l = 1, basbcs = [a1]bvcs and then w1 =
[c]bvcs. Then either bvcs[c] or [c]bvcs is normal or w1 is equivalent to the multiplication
composition with respect to w = bvcX
s bvcs. Again by Lemmas 6.11 and 6.9, the result
holds. For k + l ≥ 2, it follows from (iii).
Let s1, s2 ∈ Liek[Y ](X) be two k-monic polynomials in Liek[Y ](X). If asX1 bs
X2 c ∈
ALSW (X) for some a, b, c ∈ X∗, then by Lemma 6.2, there exits a bracketing way
[asX1 bs
X2 c]sX
1 ,sX2
such that [asX1 bs
X2 c]sX
1 ,sX2
= asX1 bs
X2 c. Denote
[as1bs2c]s1,s2 = sY2 [asX
1 bsX2 c]sX
1 ,sX2|[sX
1 ] 7→s1,
[as1bs2c]s1,s2 = sY1 [asX
1 bsX2 c]sX
1 ,sX2|[sX
2 ] 7→s2,
[as1bs2c]s1,s2 = [asX1 bs
X2 c]sX
1 ,sX2|[sX
1 ] 7→s1,[sX2 ] 7→s2
.
Thus, the leading words of the above three polynomials are as1bs2c = sY1 s
Y2 as
X1 bs
X2 c.
The following lemma is also essential.
Lemma 6.13 Let S be a Grobner-Shirshov basis in Liek[Y ](X). For any s1, s2 ∈ S, β1, β2 ∈[Y ], a1, a2, b1, b2 ∈ X∗ such that w = β1a1s1b1 = β2a2s2b2 ∈ TA, we have
β1[a1s1b1]s1 ≡ β2[a2s2b2]s2 mod(S,w).
Proof. Let L be the least common multiple of sY1 and sY
2 . Then wY = β1sY1 = β2s
Y2 = Lt
for some t ∈ [Y ], wX = a1sX1 b1 = a2s
X2 b2 and
β1[a1s1b1]s1 − β2[a2s2b2]s2 = t(L
sY1
[a1s1b1]s1 −L
sY2
[a2s2b2]s2).
Consider the first case in which sX2 is a subword of b1, i.e., wX = a1s
X1 as
X2 b2 for some
a ∈ X∗ such that b1 = asX2 b2 and a2 = a1s
X1 a. Then
β1[a1s1b1]s1 − β2[a2s2b2]s2
= t(L
sY1
[a1s1asX2 b2]s1 −
L
sY2
[a1sX1 as2b2]s2)
= tC3〈s1, s2〉w′
if L 6= sY1 s
Y2 , where w′ = LwX . Since S is a Grobner-Shirshov basis, C3〈s1, s2〉 ≡
0 mod(S, LwX). The result follows from w = tLwX = tw′.
99
Suppose that L = sY1 s
Y2 . By noting that 1
sY1[a1s1as2b2]s1,s2 and 1
sY2[a1s1as2b2]s1,s2 are
normal, by Lemma 6.10 we have
[a1s1as2b2]s1,s2 ≡ sY2 [a1s1as
X2 b2]s1 mod(S,w′),
[a1s1as2b2]s1,s2 ≡ sY1 [a1s
X1 as2b2]s2 mod(S,w′).
Thus, by Lemma 6.12, we have
β1[a1s1b1]s1 − β2[a2s2b2]s2
= t(sY2 [a1s1as
X2 b2]s1 − sY
1 [a1sX1 as2b2]s2)
= t((sY2 [a1s1as
X2 b2]s1 − [a1s1as2b2]s1,s2) + ([a1s1as2b2]s1,s2 − [a1s1as2b2]s1,s2)
−([a1s1as2b2]s1,s2 − [a1s1as2b2]s1,s2)− (sY1 [a1s
X1 as2b2]s2 − [a1s1as2b2]s1,s2))
= t((sY1 [a1s1as
X2 b2]s1 − [a1s1as2b2]s1,s2) + [a1(s1 − [s1])as2b2]s1,s2
−[a1s1a(s2 − [s2])b2]s1,s2 − (sY1 [a1s
X1 as2b2]s2 − [a1s1as2b2]s1,s2))
≡ 0 mod(S,w).
Second, if sX2 is a subword of sX
1 , i.e., sX1 = asX
2 b for some a, b ∈ X∗, then [a2s2b2]s2 =
[a1as2bb1]s2 . Let w′ = LsX1 . Thus, by noting that [a1[as2b]s2b1] is normal and by Lemmas
6.10 and 6.12,
β1[a1s1b1]s1 − β2[a2s2b2]s2
= t(L
sY1
[a1s1b1]s1 −L
sY2
[a1as2bb1]s2)
= t(L
sY1
[a1s1b1]s1 −L
sY2
[a1s1b1]s1|s1 7→[as2b]s2)− L
sY2
([a1as2bb1]s2 − [a1s1b1]s1|s1 7→[as2b]s2)
= t[a1(L
sY1
s1 −L
sY2
[as2b]s2)b1]−L
sY2
([a1as2bb1]s2 − [aX1 [as2b]s2b1])
= t[a1C1〈s1, s2〉w′b1]−L
sY2
([a1as2bb1]s2 − [a1[as2b]s2b1])
≡ 0 mod(S,w).
One more case is possible: A proper suffix of sX1 is a proper prefix of sX
2 , i.e., sX1 = ab
and sX2 = bc for some a, b, c ∈ X∗ and b 6= 1. Then abc is an ALSW. Let w′ = Labc. Then
by Lemmas 6.10 and 6.12, we have
β1[a1s1b1]s1 − β2[a2s2b2]s2
= t(L
sY1
[a1s1cb2]s1 −L
sY2
[a1as2b2]s2)
= tL
sY1
([a1s1cb2]s1 − [a1[s1c]s1b2])− tL
sY2
([a1as2b2]s2 − [a1[as2]s2b2])
+t([a1C2〈s1, s2〉w′b2]
≡ 0 mod(S,w).
100
The proof is complete.
Theorem 6.14 (CD-lemma for Lie algebras over a commutative algebra) Let S ⊂ Liek[Y ](X)
be a nonempty set of k-monic polynomials and Id(S) be the k[Y ]-ideal of Liek[Y ](X) gen-
erated by S. Then the following statements are equivalent.
(i) S is a Grobner-Shirshov basis in Liek[Y ](X).
(ii) f ∈ Id(S)⇒ f = βasb ∈ TA for some s ∈ S, β ∈ [Y ] and a, b ∈ X∗.
(iii) Irr(S) = [u] | [u] ∈ TN , u 6= βasb, for any s ∈ S, β ∈ [Y ], a, b ∈ X∗ is a k-linear
basis for Liek[Y ](X|S) = Liek[Y ](X)/Id(S).
Proof. (i) ⇒ (ii). Let S be a Grobner-Shirshov basis and 0 6= f ∈ Id(S). Then by
Lemma 6.11 f has an expression f =∑αiβi[aisibi]si
, where αi ∈ k, βi ∈ [Y ], ai, bi ∈X∗, si ∈ S. Denote wi = βi[aisibi]si
, i = 1, 2, . . . . Then wi = βiaisibi. We may assume
without loss of generality that
w1 = w2 = · · · = wl wl+1 wl+2 · · ·
for some l ≥ 1.
The claim of the theorem is obvious if l = 1.
Now suppose that l > 1. Then β1a1s1b1 = w1 = w2 = β2a2s2b2. By Lemma 6.13,
α1β1[a1s1b1]s1 + α2β2[a2s2b2]s2
= (α1 + α2)β1[a1s1b1]s1 + α2(β2[a2s2b2]s2 − β1[a1s1b1]s1)
≡ (α1 + α2)β1[a1s1b1]s1 mod(S,w1).
Therefore, if α1 + α2 6= 0 or l > 2, then the result follows from the induction on l.
For the case α1 + α2 = 0 and l = 2, we use the induction on w1. Now the result follows.
(ii)⇒ (iii). For any f ∈ Liek[Y ](X), we have
f =∑
βi[aisibi]sif
αiβi[aisibi]si+∑
[uj ]f
α′j[uj],
where αi, α′j ∈ k, βi ∈ [Y ], [uj] ∈ Irr(S) and si ∈ S. Therefore, the set Irr(S) generates
the algebra Liek[Y ](X)/Id(S).
On the other hand, suppose that h =∑αi[ui] = 0 in Liek[Y ](X)/Id(S), where
αi ∈ k, [ui] ∈ Irr(S). This means that h ∈ Id(S). Then all αi must be equal to zero.
Otherwise, h = uj for some j which contradicts (ii).
101
(iii)⇒ (i). For any f, g ∈ S, we have
Cτ (f, g)w =∑
βi[aisibi]si≺w
αiβi[aisibi]si+∑
[uj ]≺w
α′j[uj].
For τ = 1, 2, 3, 4, since Cτ (f, g)w ∈ Id(S) and by (iii), we have
Cτ (f, g)w =∑
βi[aisibi]si≺w
αiβi[aisibi]si.
Therefore, S is a Grobner-Shirshov basis.
§6.3 Applications
In this section, all algebras (Lie or associative) are understood to be taken over an
associative and commutative k-algebra K with identity and all associative algebras are
assumed to have identity.
Let L be an arbitrary Lie K-algebra which is presented by generators X and defining
relations S, L = LieK(X|S). Let K have a presentation by generators Y and defining
relations R, K = k[Y |R]. Let Y and X be deg-lex orderings on [Y ] and X∗ respectively.
Let RX = rx|r ∈ R, x ∈ X. Then as k[Y ]-algebras,
L = Liek[Y |R](X|S) ∼= Liek[Y ](X|S,RX).
The Poincare-Birkhoff-Witt theorem cannot be generalized to Lie algebras over an
arbitrary ring. This implies that not any Lie algebra over a commutative algebra has
a faithful representation in an associative algebra over the same commutative algebra.
Following P.M. Cohn, a Lie algebra with the PBW property is said to be “special0 . The
first non-special example was given by A.I. Shirshov ([196], see also [204]), and he also
formulated that if no nonzero element of K annihilates an element of the center of LK ,
then a faithful representation always exists. Another classical non-special example was
given by P. Cartier [67]. In the same paper, he proved that each Lie algebra over Dedekind
domain is special. In both examples the Lie algebras are taken over commutative algebras
over GF(2). Shirshov and Cartier used ad hoc methods to prove that some elements of
corresponding Lie algebras are not zero though they are zero in the universal enveloping
algebras. P.M. Cohn [93] proved that any Lie algebra over Kk, where chk = 0, is special.
Also he claimed that he gave an example of non-special Lie algebra over a truncated
polynomial algebra over a filed of characteristic p > 0. But he did not give a proof.
Here we find Grobner-Shirshov bases of Shirshov and Cartier’s Lie algebras and
then use Theorem 6.14 to get the results and we give proof for P.M. Cohn’s example of
102
characteristics 2, 3 and 5. We present an algorithm that one can check for any p, whether
Cohn’s conjecture is valid.
Note that if L = LieK(X|S), then the universal enveloping algebra of L is UK(L) =
K〈X|S(−)〉 where S(−) is just S but substituting all [u, v] by uv − vu.
Example 6.15 (Shirshov [196]) Let the field k = GF (2) and K = k[Y |R], where
Y = yi, i = 0, 1, 2, 3, R = y0yi = yi (i = 0, 1, 2, 3), yiyj = 0 (i, j 6= 0).
Let L = LieK(X|S1, S2), where X = xi, 1 ≤ i ≤ 13, S1 consists of the following relations
[x2, x1] = x11, [x3, x1] = x13, [x3, x2] = x12,
[x5, x3] = [x6, x2] = [x8, x1] = x10,
[xi, xj] = 0 (for any other i > j),
and S2 consists of the following relations
y0xi = xi (i = 1, 2, . . . , 13),
x4 = y1x1, x5 = y2x1, x5 = y1x2, x6 = y3x1, x6 = y1x3,
x7 = y2x2, x8 = y3x2, x8 = y2x3, x9 = y3x3,
y3x11 = x10, y1x12 = x10, y2x13 = x10,
y1xk = 0 (k = 4, 5, . . . , 11, 13), y2xt = 0 (t = 4, 5, . . . , 12), y3xl = 0 (l = 4, 5, . . . , 10, 12, 13).
Then L is not special.
Proof. L = LieK(X|S1, S2) = Liek[Y ](X|S1, S2, RX). We order Y and X by yi >
yj if i > j and xi > xj if i > j respectively. It is easy to see that for the ordering on [Y ]X∗ as before, S = S1 ∪ S2 ∪ RX ∪ y1x2 = y2x1, y1x3 = y3x1, y2x3 = y3x2 is a
Grobner-Shirshov basis in Liek[Y ](X). Since x10 ∈ Irr(S) and Irr(S) is a k-linear basis
of L by Theorem 6.14, x10 6= 0 in L.
On the other hand, the universal enveloping algebra of L has a presentation:
UK(L) = K〈X|S(−)1 , S2〉 ∼= k[Y ]〈X|S(−)
1 , S2, RX〉,
where S(−)1 is just S1 but substituting all [uv] by uv − vu.
But the Grobner-Shirshov complement (see Mikhalev-Zolotyhk [165]) of S(−)1 ∪ S2 ∪
RX in k[Y ]〈X〉 is
Sc = S(−)1 ∪ S2 ∪RX ∪ y1x2 = y2x1, y1x3 = y3x1, y2x3 = y3x2, x10 = 0.
Thus, L is not special.
103
Example 6.16 (Cartier [67]) Let k = GF (2), K = k[y1, y2, y3|y2i = 0, i = 1, 2, 3] and
L = LieK(X|S), where X = xij, 1 ≤ i ≤ j ≤ 3 and
S = [xii, xjj] = xji (i > j), [xij, xkl] = 0 (otherwise), y3x33 = y2x22 + y1x11.
Then L is not special.
Proof. Let Y = y1, y2, y3. Then
L = LieK(X|S) ∼= Liek[Y ](X|S, y2i xkl = 0 (∀i, k, l)).
Let yi > yj if i > j and xij > xkl if (i, j) >lex (k, l) respectively. It is easy to see
that for the ordering on [Y ]X∗ as before, S ′ = S ∪ y2i xkl = 0 (∀i, k, l) ∪ S1 is a
Grobner-Shirshov basis in Liek[Y ](X), where S1 consists of the following relations
y3x23 = y1x12, y3x13 = y2x12, y2x23 = y1x13, y3y2x22 = y3y1x11,
y3y1x12 = 0, y3y2x12 = 0, y3y2y1x11 = 0, y2y1x13 = 0.
The universal enveloping algebra of L has a presentation:
UK(L) = K〈X|S(−)〉 ∼= k[Y ]〈X|S(−), y2i xkl = 0 (∀i, k, l)〉.
In UK(L), we have
0 = y23x
233 = (y2x22 + y1x11)
2 = y22x
222 + y2
1x211 + y2y1[x22, x11] = y2y1x12.
On the other hand, since y2y1x12 ∈ Irr(S ′), y2y1x12 6= 0 in L. Thus, L is not special.
Cohn conjecture (Cohn [93]) Let K = k[y1, y2, y3|ypi = 0, i = 1, 2, 3] be the algebra
of truncated polynomials over a field k of characteristic p > 0. Let
Lp = LieK(x1, x2, x3 | y3x3 = y2x2 + y1x1).
Then Lp is not special. We call Lp the Cohn’s Lie algebra.
Remark In UK(Lp) we have
0 = (y3x3)p = (y2x2)
p + Λp(y2x2, y1x1) + (y1x1)p = Λp(y2x2, y1x1),
where Λp is a Jacobson-Zassenhaus Lie polynomial. P.M. Cohn conjectured that Λp(y2x2, y1x1) 6=0 in Lp.
Theorem 6.17 Cohn’s Lie algebras L2, L3 and L5 are not special.
104
Proof. Let Y = y1, y2, y3, X = x1, x2, x3 and S = y3x3 = y2x2 + y1x1, ypi xj =
0, 1 ≤ i, j ≤ 3. Then Lp∼= Liek[Y ](X|S) and UK(Lp) ∼= k[Y ]〈X|S〉. Suppose that Sc is
a Grobner-Shirshov complement of S in Liek[Y ](X). Let SX
p ⊂ Lp be the set of all the
elements of Sc whose X-degrees do not exceed p.
First, we consider p = 2 and prove the element Λ2 = [y2x2, y1x1] = y2y1[x2x1] 6= 0 in
L2.
Then by Shirshov’s algorithm we have that SX2 consists of the following relations
y3x3 = y2x2 + y1x1, y2i xj = 0 (1 ≤ i, j ≤ 3), y3y2x2 = y3y1x1, y3y2y1x1 = 0,
y2[x3x2] = y1[x3x1], y3y1[x2x1] = 0, y2y1[x3x1] = 0.
Thus, Λ2 is in the k-linear basis Irr(Sc) of L2.
Now, by the above remark, L2 is not special.
Second, we consider p = 3 and prove the element Λ3 = y22y1[x2x2x1]+y2y
21[x2x1x1] 6= 0
in L3.
Then again by Shirshov’s algorithm, SX3 consists of the following relations
y3x3 = y2x2 + y1x1, y3i xj = 0 (1 ≤ i, j ≤ 3), y2
3y2x2 = y23y1x1, y
23y
22y1x1 = 0,
y2[x3x2] = −y1[x3x1], y23y1[x2x1] = 0, y2
2y1[x3x1] = 0,
y3y22[x2x2x1] = y3y2y1[x2x1x1], y3y
22y1[x2x1x1] = 0, y3y2y1[x2x2x1] = y3y
21[x2x1x1].
Thus, y22y1[x2x2x1], y2y
21[x2x1x1] ∈ Irr(Sc), which implies Λ3 6= 0 in L3.
Third, let p = 5. Again by Shirshov’s algorithm, SX5 consists of the following
relations
1) y3x3 = y2x2 + y1x1,
2) y5i xj = 0, 1 ≤ i, j ≤ 3,
3) y43y2x2 = −y4
3y1x1,
4) y43y
42y1x1 = 0,
5) y2[x3x2] = −y1[x3x1],
6) y43y1[x2x1] = 0,
7) y42y1[x3x1] = 0,
8) y33y
22[x2x2x1] = y3
3y2y1[x2x1x1],
9) y33y
42y1[x2x1x1] = 0,
10) y33y2y1[x2x2x1] = y3
3y21[x2x1x1],
105
11) y1[x3x2x3x1] = 0,
12) y1[x3x1x2x1] = 0,
13) y1[x3x2x2x1] = −y1[x3x2x1x2],
14) y2[x3x1x2x1] = 0,
15) y23y
32[x2x2x2x1] = 2y2
3y22y1[x2x2x1x1]− y2
3y2y21[x2x1x1x1],
16) y33y
32y
21[x2x1x1x1] = 0,
17) y23y
22y1[x2x2x2x1] = 2y2
3y2y21[x2x2x1x1]− y2
3y31[x2x1x1x1],
18) y23y
42y
21[x2x1x1x1] = 0,
19) y23y
42y1[x2x2x1x1] =
1
2y2
3y32y
21[x2x1x1x1],
20) y33y
21[x2x2x1x2x1] = 0,
21) y33y2y1[x2x1x2x1x1] = 0,
22) y33y
21[x2x1x2x1x1] = 0,
23) y33y
22[x2x1x2x1x1] = 0,
24) y23y
22y1[x2x2x1x2x1] = −y2
3y2y21[x2x1x2x1x1],
25) y23y2y
21[x2x2x1x2x1] = −y2
3y31[x2x1x2x1x1],
26) y23y
42y
21[x2x1x2x1x1] = 0,
27) y3y42[x2x2x2x2x1] = 3y3y
32y1[x2x2x2x1x1]− y3y
32y1[x2x2x1x2x1]− 3y3y
22y
21[x2x2x1x1x1]
−2y3y22y
21[x2x1x2x1x1] + y3y2y
31[x2x1x1x1x1],
28) y3y32y1[x2x2x2x2x1] = 3y3y
22y
21[x2x2x2x1x1]− y3y
22y
21[x2x2x1x2x1]− 3y3y2y
31[x2x2x1x1x1]
−2y3y2y31[x2x1x2x1x1] + y3y
41[x2x1x1x1x1],
29) y3y42y
31[x2x1x1x1x1] = 0,
30) y23y
32y
31[x2x1x1x1x1] = 0,
31) y3y42y
21[x2x2x1x1x1] = −2
3y3y
42y
21[x2x1x2x1x1] +
1
3y3y
32y
31[x2x1x1x1x1],
32) y3y42y1[x2x2x2x1x1] =
1
3y3y
42y1[x2x2x1x2x1] + y3y
32y
21[x2x2x1x1x1]
+2
3y3y
32y
21[x2x1x2x1x1]−
1
3y3y
22y
31[x2x1x1x1x1],
33) y32y
21[x3x3x1x3x1] = 0,
34) y32y
21[x3x1x3x1x1] = 0,
35) y33y
22y
31[x2x1x1x1x1] = 0,
36) y23y
32y
21[x2x2x1x1x1] = −2
3y2
3y32y
21[x2x1x2x1x1] +
2
3y2
3y22y
31[x2x1x1x1x1].
Thus, Λ5(y2x2, y1x1) = y42y1[x2x2x2x2x1] ∈ Irr(Sc), which implies Λ5 6= 0 in L5.
106
Remark Note that the Jacobson-Zassenhaus Lie polynomial Λp(y2x2, y1x1) is of X-degree
p. Then Λp(y2x2, y1x1) ∈ Irr(Sc) if and only if Λp(y2x2, y1x1) ∈ Irr(SXp). Since the defin-
ing relation of Lp is homogenous on X, SXp is a finite set. By Shirshov’s algorithm, one
can compute SXp for Lp.
Now we give some examples which are special Lie algebras.
Lemma 6.18 Suppose that f and g are two polynomials in Liek[Y ](X) such that f is
k[Y ]-monic and g = rx, where r ∈ k[Y ] and x ∈ X, is k-monic. Then each inclusion
composition of f and g is trivial modulo f ∪ rX.
Proof. Suppose that f = [axb] for some a, b ∈ X∗, f = f + f ′ and g = rx + r′x. Then
w = raxb and
C1〈f, g〉w = rf − [a[rx]b]rx
= rf ′ − r′[axb]
= rf ′ − r′f
≡ 0 mod(f ∪ rX,w).
Now we give other applications.
Theorem 6.19 Suppose that S is a finite homogeneous subset of Liek(X). Then the word
problem of LieK(X|S) is solvable for any finitely generated commutative k-algebra K.
Proof. Let Sc be a Grobner-Shirshov complement of S in Liek(X). Clearly, Sc consists
of homogeneous elements in Liek(X) since the compositions of homogeneous elements are
homogeneous. Since K is finitely generated commutative k-algebra, we may assume that
K = k[Y |R] with R a finite Grobner-Shirshov basis in k[Y ]. By Lemma 6.18, Sc ∪ RXis a Grobner-Shirshov basis in Liek[Y ](X). For a given f ∈ LieK(X), it is obvious that
after a finite number of steps one can write down all the elements of Sc whose X-degrees
do not exceed the degree of fX . Denote the set of such elements by SfX . Then SfX is a
finite set. By Theorem 6.14, the result follows.
Theorem 6.20 Every finitely or countably generated Lie K-algebra can be embedded into
a two-generated Lie K-algebra, where K is an arbitrary commutative k-algebra.
107
Proof. Let K = k[Y |R] and L = LieK(X|S) where X = xi, i ∈ I and I is a subset
of the set of nature numbers. Without loss of generality, we may assume that with the
ordering on [Y ]X∗ as before, S ∪RX is a Grobner-Shirshov basis in Liek[Y ](X).
Consider the algebra L′ = Liek[Y ](X, a, b|S ′) where S ′ = S∪RX∪Ra, b∪[aabiab]−xi, i ∈ I.
Clearly, L′ is a Lie K-algebra generated by a, b. Thus, in order to prove the theorem,
by using our Theorem 6.14, it suffices to show that with the ordering on [Y ](X∪a, b)∗
as before, where a b xi, xi ∈ X, S ′ is a Grobner-Shirshov basis in Liek[Y ](X, a, b).
It is clear that all the possible compositions of multiplication, intersection and inclu-
sion are trivial. We only check the external compositions of some f ∈ S and ra ∈ Ra:Let w = Lu1f
Xu2au3 where L = L(fY , r) and u1fXu2au3 ∈ ALSW (X, a, b). Then
C3〈f, ra〉w
=L
fY1
[u1fu2au3]f −L
r[u1f
Xu2(ra)u3]
= (L
fY1
[u1fu2au3]f − rL
r[u1f
Xu2au3]fX )− (L
r[u1f
Xu2(ra)u3]− rL
r[u1f
Xu2au3]fX )
= ([u1(L
fY1
f)u2au3]f − [u1(rL
rfX)u2au3]fX )− rL
r([u1f
Xu2au3]− [u1fXu2au3]fX )
≡ [u1C3〈f, rx〉w′u2au3] mod(S ′, w)
for some x occurring in fX and w′ = LfX . Since S ∪ RX is a Grobner-Shirshov basis in
Liek[Y ](X), C3〈f, rx〉w′ ≡ 0 mod(S∪RX,w′). Thus by Lemma 6.12, [u1C3〈f, rx〉w′u2au3] ≡0 mod(S ′, w).
108
Chapter 7 Grobner-Shirshov Bases for Pre-Lie
Algebras
§7.1 Composition-Diamond lemma for pre-Lie algebras
Definition 7.1 An algebra A over a field k is called a pre-Lie (or right-symmetric) algebra
if A satisfies (x, y, z) = (x, z, y) for any x, y, z ∈ A, where (x, y, z) = (xy)z − x(yz).
Let X = xi|i ∈ I be a set, X∗ the set of all associative words u in X, X∗∗ the set
of all non-associative words (u) in X, and |(u)| the length of the word (u).
Let I be a well-ordered set. We order X∗∗ by the induction on the lengths of the
words (u) and (v) in X∗∗:
(i) If |((u)(v))| = 2, then (u) = xi > (v) = xj if and only if i > j.
(ii) If |((u)(v))| > 2, then (u) > (v) if and only if one of the following cases holds:
(a) |(u)| > |(v)|.
(b) If |(u)| = |(v)|, (u) = ((u1)(u2)) and (v) = ((v1)(v2)), then (u1) > (v1) or
((u1) = (v1) and (u2) > (v2)).
It is clear that the ordering < on X∗∗ is well-ordered. This ordering is called deg-lex
(degree-lexicographical) ordering and we use this ordering throughout this paper.
We now cite the definition of good words (cf. [193]) by induction on length:
1) xi is a good word for any xi ∈ X.
Suppose that we define good words of length < n.
2) non-associative word ((v)(w)) is called a good word if
(a) both (v) and (w) are good words,
(b) if (v) = ((v1)(v2)), then (v2) ≤ (w).
We denote (u) by [u], if (u) is a good word.
Let W be the set of all good words in the alphabet X and pre-Lie(X) the free pre-Lie
algebra over a field k generated by X. Then W forms a linear basis of the free pre-Lie
algebra pre-Lie(X), see [193].
109
Every nonzero element f in pre-Lie(X) can be uniquely represented as
f = λ1[w1] + λ2[w2] + · · ·+ λn[wn]
where [wi] ∈ W , 0 6= λi ∈ k for all i and [w1] > [w2] > · · · > [wn]. Denote by f the
leading word [w1] of f . f is called monic if the coefficient of f is 1.
For any (w), (w1) ∈ X∗∗, denote by
(w)R(w1) = ((w)(w1)).
The following results are actually proved in [137].
Lemma 7.2 In X∗∗, every good word [w] ∈ W can be uniquely represented as
[w] = xiR[w1]R[w2] · · ·R[wn]
where [wj] ∈ W for all j and [w1] ≤ [w2] ≤ · · · ≤ [wn].
Lemma 7.3 Let [u] and [v] be arbitrary good words and assume that [u] =
xiR[u1]R[u2] · · ·R[un]. Then, in pre-Lie(X),
[u][v] = xiR[u1] · · ·R[us]R[v]R[us+1] · · ·R[un]
where [u1] ≤ · · · ≤ [us] ≤ [v] < [us+1] ≤ · · · ≤ [un] and s 6 n.
By Lemma 7.3, we have
Corollary 7.4 Let [u], [v] ∈ W and [u] = xiR[u1] · · ·R[um−1]R[um] = [u′][um], where
[u′] = xiR[u1] · · ·R[um−1]. If [um] > [v], then, in pre-Lie(X), [u][v] = ([u′][v])[um] and
[u′]([um][v]), [u′]([v][um]) < [u][v].
Lemma 7.5 Let [u], [v], and [w] be arbitrary good words. If [u] < [v] then [w][u] < [w][v]
and [u][w] < [v][w]. It follows that if f, g ∈ pre-Lie(X), then fg = f g.
Definition 7.6 Let S ⊂ kX∗∗ be a set of monic polynomials, s ∈ S and (u) ∈ X∗∗, where
kX∗∗ is the free non-associative algebra over a field k generated by X. We view S as a
new set of letters with S ∩X = ∅. We define S-word (u)s by induction:
(i) (u)s = s is an S-word of S-length 1.
110
(ii) If (u)s is an S-word with S-length k and [v] is a good word with length l, then
(u)s[v] and [v](u)s
are S-words with S-length k + l.
Definition 7.7 An S-word (u)s is called a normal S-word if (u)s = (asb) is a good word.
We denote (u)s by [u]s if (u)s is a normal S-word. From Lemma 7.5, it follows that
[u]s = [u]s.
Definition 7.8 Let f, g ∈ pre-Lie(X) be monic polynomials, [w] ∈ W , and a, b ∈ X∗.
Then there are two kinds of compositions.
(i) If f = [agb], then
(f, g)f = f − [agb]
is called composition of inclusion.
(ii) If (f [w]) is not good, then
f · [w]
is called composition of right multiplication.
Let S ⊂ pre-Lie(X) be a given nonempty subset. The composition of inclusion (f, g)f
is said to be trivial modulo S if
(f, g)f =∑
i
αi[aisibi]
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi] is normal S-word and [aisibi] < f . If this
is the case, then we write
(f, g)f ≡ 0 mod(S, f).
In general, for any normal word [w] and p, q ∈ pre-Lie(X), we write
p ≡ q mod(S, [w])
which means that p − q =∑αi[aisibi], where each αi ∈ k, ai, bi ∈ X∗, si ∈ S and
[aisibi] < [w].
The composition of right multiplication f · [w] is said to be trivial modulo S if
f · [w] =∑
i
αi[aisibi]
111
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi] is normal S-word and [aisibi] ≤ f · [w]. If
this is the case, then we write
f · [w] ≡ 0 mod(S).
Definition 7.9 Let S ⊂ pre-Lie(X) be a nonempty set of monic polynomials and the
ordering < on X∗∗ be defined as before. Then the set S is called a Grobner-Shirshov basis
in pre-Lie(X) if any composition in S is trivial modulo S.
Lemma 7.10 Let S ⊂ pre-Lie(X) be a set of monic polynomials and (u)s an S-word. If
any right multiplication composition in S is trivial module S, then (u)s has a representa-
tion:
(u)s =∑
i
αi[ui]si
where each αi ∈ k, [ui]siis a normal S-word and [ui]si
≤ (u)s.
Proof. We use induction on (u)s. If (u)s = s, then (u)s = s and the result holds.
Assume that (u)s > s. Then (u)s = (v)s[w] or (u)s = [w](v)s. We consider only the case
(u)s = (v)s[w]. The other case can be similarly proved.
By induction, we may assume that (u)s = [v]s[w]. If [v]s = s, then the result holds
clearly because each right multiplication composition in S is trivial modulo S. Suppose
that [v]s = [v1]s[v2] or [v]s = [v1][v2]s. We consider only the case [v]s = [v1]s[v2]. The
other case can be similarly proved. If [v2] ≤ [w], then (u)s = [v]s[w] is a normal S-word
and we get the result. If [v2] > [w], then
(u)s = ([v1]s[v2])[w] = ([v1]s[w])[v2] + [v1]s([v2][w])− [v1]s([w][v2]).
By induction, [v1]s[w] =∑j
βj[vj]sj, where βj ∈ k, [vj]sj
is a normal S-word, and [vj]sj≤
[v1]s[w]. If [vj]sj= [v1]s[w], then [vj]sj
[v2] is a normal S-word since [vj]sj[v2] = [vj]sj
[v2] =
([v1]s[w])[v2] = ([v1]s[v2])[w] = (u)s by Corollary 7.4. If [vj]sj< [v1]s[w], then [vj]sj
[v2] <
([v1]s[w])[v2] = ([v1]s[w])[v2] = (u)s. By Corollary 7.4 again,
[v1]s([v2][w]), [v1]s([w][v2]) < ([v1]s[w])[v2] = (u)s.
Now the result follows from the induction.
Lemma 7.11 Let [a1s1b1], [a2s2b2] be normal S-words. If S is a Grobner-Shirshov basis
in pre-Lie(X) and [w] = [a1s1b1] = [a2s2b2], then
[a1s1b1] ≡ [a2s2b2] mod(S, [w]).
112
Proof. We have w = a1s1b1 = a2s2b2 as associative words. There are two cases to
consider.
Case 1. Suppose that s1 and s2 are disjoint, say, |a2| ≥ |a1| + |s1|. Then, we can
assume that
a2 = a1s1c and b1 = cs2b2
for some c ∈ X∗. Thus, [w] = [a1s1cs2b2]. Now,
[a1s1b1]− [a2s2b2] = [a1s1cs2b2]− [a1s1cs2b2]
= [a1s1cs2b2]− (a1s1cs2b2) + (a1s1cs2b2)− [a1s1cs2b2]
= (a1s1c(s2 − s2)b2) + (a1(s1 − s1)cs2b2).
Since [s2 − s2] < s2 and [s1 − s1] < s1, and by Lemmas 7.5 and 7.10, we conclude that
[a1s1b1]− [a2s2b2] =∑
i
αi[uisivi]
for some αi ∈ k, normal S-words [uisivi] such that [uis1vi] < [w]. Hence,
[a1s1b1] ≡ [a2s2b2] mod(S, [w]).
Case 2. Suppose that s1 contains s2 as a subword. We assume that
s1 = [as2b], a2 = a1a and b2 = bb1, that is, [w] = [a1[as2b]b1]
for the normal S-word [as2b]. We have
[a1s1b1]− [a2s2b2] = [a1s1b1]− [a1[as2b]b1]
= (a1(s1 − [as2b])b1)
= (a1(s1, s2)s1b1).
Since S is a Grobner-Shirshov basis, (s1, s2)s1 =∑i
αi[cisidi] for some αi ∈ k, normal
S-words [cisidi] with each [cisidi] < s1. By Lemmas 7.5 and 7.10, we have
[a1s1b1]− [a2s2b2] = (a1(s1, s2)s1b1)
=∑
i
αi(a1[cisidi]b1) =∑
j
βj[ajsjbj]
for some βj ∈ k, normal S-words [ajsjbj] with each [aj sjbj] < [w] = [a1s1b1].
Consequently, [a1s1b1] ≡ [a2s2b2] mod(S, [w]).
113
Lemma 7.12 Let S ⊂ pre-Lie(X) be a set of monic polynomials and Irr(S) = [u] ∈W |[u] 6= [asb] a, b ∈ X∗, s ∈ S and [asb] is a normal S-word. Then for any f ∈pre-Lie(X),
f =∑
[ui]≤f
αi[ui] +∑
[aj sjbj ]≤f
βj[ajsjbj],
where each αi, βj ∈ k, [ui] ∈ Irr(S) and [ajsjbj] is a normal S-word.
Proof. Let f =∑i
αi[ui] ∈ pre-Lie(X) where 0 6= αi ∈ k and [u1] > [u2] > · · · . If
[u1] ∈ Irr(S), then let f1 = f − α1[u1]. If [u1] 6∈ Irr(S), then there exist some s ∈ S
and a1, b1 ∈ X∗ such that f = [a1s1b1]. Let f1 = f − α1[a1s1b1]. In both cases, we have
f1 < f . Then the result follows by using induction on f .
Theorem 7.13 (CD-lemma for pre-Lie algebras) Let S ⊂ pre-Lie(X) be a nonempty set
of monic polynomials and the ordering < be defined as before. Let Id(S) be the ideal of
pre-Lie(X) generated by S. Then the following statements are equivalent:
(i) S is a Grobner-Shirshov basis in pre-Lie(X).
(ii) f ∈ Id(S)⇒ f = [asb] for some s ∈ S and a, b ∈ X∗, where [asb] is a normal S-word.
(ii)′ f ∈ Id(S)⇒ f = α1[a1s1b1]+α2[a2s2b2]+ · · · , where αi ∈ k, [a1s1b1] > [a2s2b2] > · · ·and each [aisibi] is a normal S-word.
(iii) Irr(S) = [u] ∈ W |[u] 6= [asb] a, b ∈ X∗, s ∈ S and [asb] is a normal S-word is a
linear basis of the algebra pre-Lie(X|S) = pre-Lie(X)/Id(S).
Proof. (i)⇒ (ii). Let S be a Grobner-Shirshov basis and 0 6= f ∈ Id(S). We can also
assume, by Lemma 7.10, that
f =n∑
i=1
αi[aisibi]
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S and [aisibi] a normal S-word. Let
[wi] = [aisibi], [w1] = [w2] = · · · = [wl] > [wl+1] ≥ · · ·
We will use the induction on l and [w1] to prove that f = [asb] for some s ∈ S and a, b ∈X∗.
If l = 1, then f = [a1s1b1] = [a1s1b1] and hence the result holds. Assume that l ≥ 2.
Then, by Lemma 7.11, we have
[a1s1b1] ≡ [a2s2b2] mod(S, [w1]).
114
Thus, if α1 + α2 6= 0 or l > 2, then the result holds. For the case that α1 + α2 = 0 and
l = 2, we use the induction on [w1]. Hence, the result follows.
(ii)⇒ (ii)′. Assume that (ii) and 0 6= f ∈ Id(S). Let f = α1f + · · · . Then, by (ii),
f = [a1s1b1]. Therefore,
f1 = f − α1[a1s1b1], f1 < f, f1 ∈ Id(S).
Now, by using induction on f , we have (ii)′.
(ii)′ ⇒ (ii). This part is clear.
(ii) ⇒ (iii). Suppose that∑i
αi[ui] = 0 in pre-Lie(X|S), where αi ∈ k, [ui] ∈
Irr(S). It means that∑i
αi[ui] ∈ Id(S) in pre-Lie(X). Then all αi must be equal to zero.
Otherwise, we have∑i
αi[ui] = [uj] ∈ Irr(S) for some j which contradicts (ii).
Now, for any f ∈ pre-Lie(X), by Lemma 7.12, we have
f =∑
[ui]∈Irr(S), [ui]≤f
αi[ui] +∑
[aj sjbj ]≤f
βj[ajsjbj].
Hence, (iii) follows.
(iii) ⇒ (i). Applying Lemma 7.12 to a composition of elements of S, we get by
(iii) that any composition is trivial because any composition belongs to Id(S). So S is a
Grobner-Shirshov basis.
§7.2 PBW theorem for pre-Lie algebras
For any pre-Lie algebra A, the algebra A(−) with new product [xy] = xy−yx is a Lie
algebra. So, for any Lie algebra L, one may define its universal enveloping pre-Lie algebra
Upre-Lie(L) as follows. If L = Lie(X|S), then Upre-Lie(L) = pre-Lie(X|S(−)) which has
the universal enveloping property: there exists a Lie homomorphism i : L→ Upre-Lie(L)
such that for any pre-Lie algebra A and any Lie homomorphism ε : L → A, there exists
a unique pre-Lie homomorphism f : Lie(X) → L such that the following triangle is
commutative
-
?
A
Upre-Lie(L)L i
ε∃! f
In this section, we give a Grobner-Shirshov basis for such an algebra.
115
Theorem 7.14 Let (L , [, ]) be a Lie algebra with a well-ordered basis ei| i ∈ I. Let
[ei, ej] =∑m
αmij em
where αmij ∈ k. We denote
∑m
αmij em by eiej. Let
Upre-Lie(L ) = pre-Lie(eiI | eiej − ejei = eiej, i, j ∈ I)
be the universal enveloping pre-Lie algebra of L . Let
S = fij = eiej − ejei − eiej, i, j ∈ I and i > j.
Then
(i) S is a Grobner-Shirshov basis in pre-Lie(X) where X = eiI .
(ii) (Segal PBW theorem) L can be embedded into the universal enveloping pre-Lie
algebra Upre-Lie(L ) as a vector space.
Proof. (i). It is clear that fij = eiej (i > j). So, there exists a unique kind of
composition fijek (i > j > k). Then, in pre-Lie(X), we have
fijek − fikej + fjkei − eifjk + ejfik − ekfij −∑m
αmjkfim −
∑m
αmij fkm −
∑m
αmikfmj
=(eiej)ek − (ejei)ek − eiejek − (eiek)ej + (ekei)ej + eiekej
+ fjkei − eifjk + ejfik − ekfij −∑m
αmjkfim −
∑m
αmij fkm −
∑m
αmikfmj
=(eiek)ej + ei(ejek)− ei(ekej)− (ejek)ei − ej(eiek) + ej(ekei)− eiejek − (eiek)ej + (ekej)ei
+ ek(eiej)− ek(ejei) + eiekej + fjkei − eifjk + ejfik − ekfij
−∑m
αmjkfim −
∑m
αmij fkm −
∑m
αmikfmj
=− (ejek)ei + (ekej)ei + ei(ejek)− ei(ekej) + ej(ekei)− ej(eiek)− ek(ejei) + ek(eiej)
− eiejek + eiekej + fjkei − eifjk + ejfik − ekfij −∑m
αmjkfim −
∑m
αmij fkm −
∑m
αmikfmj
=− ejekei + eiejek − ejeiek+ ekeiej − eiejek + eiekej
−∑m
αmjkfim −
∑m
αmij fkm −
∑m
αmikfmj
=− ekeiej − eiejek − ejekei
=0 (by Jacobi identity).
By invoking fmn = −fnm, we have fijek ≡ 0 mod(S). Therefore, S is a Grobner-
Shirshov basis for Upre-Lie(L ).
(ii). It follows from Theorem 7.13.
116
Chapter 8 Grobner-Shirshov Bases for Dialgebras
§8.1 Preliminaries
Definition 8.1 Let k be a field. A k-linear space D equipped with two bilinear multipli-
cations ` and a is called a dialgebra, if both ` and a are associative and
a a (b ` c) = a a b a c
(a a b) ` c = a ` b ` c
a ` (b a c) = (a ` b) a c
for any a, b, c ∈ D.
Definition 8.2 Let D be a dialgebra, B ⊂ D. Let us define diwords of D in the set B by
induction:
(i) b = (b), b ∈ B is a diword in B of length |b| = 1.
(ii) (u) is called a diword in B of length |(u)| = n, if (u) = ((v) a (w)) or (u) = ((v) `(w)), where (v), (w) are diwords in B of length k, l respectively and k + l = n.
Proposition 8.3 Let D be a dialgebra and B ⊂ D. Any diword of D in the set B is
equal to a diword in B of the form
(u) = b−m ` · · · ` b−1 ` b0 a b1 a · · · a bn (8.1)
where bi ∈ B, −m ≤ i ≤ n, m ≥ 0, n ≥ 0. Any bracketing of the right side of (8.1) gives
the same result.
Definition 8.4 Let X be a set. A free dialgebra D(X) generated by X over k is defined
in a usual way by the following commutative diagram:
-
?
D
D(X)X i
∀ϕ∃!ϕ∗ (di-homomorphism)
where D is any dialgebra.
117
In [147], a construction of a free dialgebra is given.
Proposition 8.5 Let D(X) be a free dialgebra over k generated by X. Any diword in
D(X) is equal to the unique diword of the form
[u] = x−m ` · · · ` x−1 ` x0 a x1 a · · · a xn , x−m · · ·x−1x0x1 · · ·xn (8.2)
where xi ∈ X, m ≥ 0, n ≥ 0, and x0 is called the center of the normal diword [u]. We
call [u] a normal diword (in X) with the associative word u, u ∈ X∗. Clearly, if [u] = [v],
then u = v. In (8.2). Let [u], [v] be two normal diwords. Then [u] ` [v] is the normal
diword [uv] with the center at the center of [v]. Accordingly, [u] a [v] is the normal diword
[uv] with the center at the center of [u].
Example 8.6
(x−1 ` x0 a x1) ` (y−1 ` y0 a y1) = x−1 ` x0 ` x1 ` y−1 ` y0 a y1,
(x−1 ` x0 a x1) a (y−1 ` y0 a y1) = x−1 ` x0 a x1 a y−1 a y0 a y1.
§8.2 Composition-Diamond lemma for dialgebras
Let X be a well-ordered set, D(X) the free dialgebra over k, X∗ the free monoid
generated by X and [X∗] the set of normal diwords in X. Let us define the deg-lex
ordering on [X∗] in the following way: for any [u], [v] ∈ [X∗],
[u] < [v]⇐⇒ wt([u]) < wt([v]) lexicographicaly,
where
wt([u]) = (n+m+ 1,m, x−m, · · · , x0, . . . , xn)
if [u] = x−m · · ·x−1x0x1 · · ·xn.
Throughout this chapter, we will use this ordering.
It is easy to see that the ordering < is satisfied the following properties:
[u] < [v] =⇒ x ` [u] < x ` [v], [u] a x < [v] a x, for any x ∈ X.
Any polynomial f ∈ D(X) has the form
f =∑
[u]∈[X∗]
f([u])[u] = α[f ] +∑
αi[ui],
118
where [f ], [ui] are normal diwords in X, [f ] > [ui], α, αi, f([u]) ∈ k, α 6= 0. We call [f ]
the leading term of f . Denote suppf by the set [u]|f([u]) 6= 0 and deg(f) by |[f ]|. f is
called monic if α = 1. f is called left (right) normed if f =∑αiuixi (f =
∑αixiui),
where each αi ∈ k, xi ∈ X and ui ∈ X∗.
If [u], [v] are both left normed or both right normed, then it is clear that for any
[w] ∈ [X∗],
[u] < [v] =⇒ [u] ` [w] < [v] ` [w], [w] ` [u] < [w] ` [v],
[u] a [w] < [v] a [w], [w] a [u] < [w] a [v].
Let S ⊂ D(X). By an S-diword g we will mean a diword in X ∪ S with only one
occurrence of s ∈ S. If this is the case and g = (asb) for some a, b ∈ X∗, s ∈ S, we also
call g an s-diword.
From Proposition 8.3 it follows that any s-diword is equal to
[asb] = x−m ` · · · ` x−1 ` x0 a x1 a · · · a xn|xk 7→s (8.3)
where −m ≤ k ≤ n, s ∈ S, xi ∈ X, −m ≤ i ≤ n. To be more precise, [asb] = [asb] if
k = 0; [asb] = [asb1x0b2] if k < 0 and [asb] = [a1x0a2sb] if k > 0. If the center of the
s-diword [asb] is in a, then we denote it by [asb] = [a1x0a2sb]. Similarly, [asb] = [asb1x0b2]
(of course, either ai or bi may be empty).
Definition 8.7 The s-diword (8.3) is called a normal s-diword if one of the following
conditions holds:
(i) k = 0,
(ii) k < 0 and s is left normed,
(iii) k > 0 and s is right normed.
We call a normal s-diword [asb] a left (right) normed s-diword if both s and [asb] are
left (right) normed. In particulary, s is a left (right) normed s-diword if s is left (right)
normed polynomial.
The following lemma follows from the above properties of the ordering <.
Lemma 8.8 For a normal s-diword [asb], the leading term of [asb] is equal to [a[s]b], that
is, [asb] = [a[s]b]. More specifically, if
[asb] = x−m ` · · · ` x−1 ` x0 a x1 a · · · a xn|xk 7→s,
119
then corresponding to k = 0, k < 0, k > 0, respectively, we have
x−m ` · · · ` x−1 ` s a x1 a · · · a xn = x−m ` · · · ` x−1 ` [s] a x1 a · · · a xn,
x−m ` · · · ` s ` · · · ` x0 a · · · a xn = x−m ` · · · ` [s] ` · · · ` x0 a · · · a xn,
x−m ` · · · ` x0 a · · · a s a · · · a xn = x−m ` · · · ` x0 a · · · a [s] a · · · a xn.
Now, we define compositions of polynomials in D(X).
Definition 8.9 Let the ordering < be as before and f, g ∈ D(X) with f, g monic.
1) Composition of left (right) multiplication.
Let f be not a right normed polynomial and x ∈ X. Then x a f is called the
composition of left multiplication. Clearly, x a f is a right normed polynomial (or 0).
Let f be not a left normed polynomial and x ∈ X. Then f ` x is called the
composition of right multiplication. Clearly, f ` x is a left normed polynomial (or 0).
2) Composition of inclusion.
Let
[w] = [f ] = [a[g]b],
where [agb] is a normal g-diword. Then
(f, g)[w] = f − [agb]
is called the composition of inclusion. The transformation f 7→ f − [agb] is called the
elimination of leading diword (ELW) of g in f , and [w] is called the ambiguity of f
and g.
3) Composition of intersection.
Let
[w] = [[f ]b] = [a[g]], |f |+ |g| > |w|,
where [fb] is a normal f -diword and [ag] a normal g-diword. Then
(f, g)[w] = [fb]− [ag]
is called the composition of intersection, and [w] is called the ambiguity of f and g.
120
Remark In the Definition 8.9, for the case of 2) or 3), we have (f, g)[w] < [w]. For the
case of 1), deg(x a f) ≤ deg(f) + 1 and deg(f ` x) ≤ deg(f) + 1.
Definition 8.10 Let the ordering < be as before, S ⊂ D(X) a monic set and f, g ∈ S.
1) Let x a f be a composition of left multiplication. Then x a f is called trivial modulo
S, denoted by x a f ≡ 0 mod(S), if
x a f =∑
αi[aisibi],
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi] right normed si-diword and |[ai[si]bi]| ≤deg(x a f).
Let f ` x be a composition of right multiplication. Then f ` x is called trivial
modulo S, denoted by f ` x ≡ 0 mod(S), if
f ` x =∑
αi[aisibi],
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi] left normed si-diword and |[ai[si]bi]| ≤deg(f ` x).
2) Composition (f, g)[w] of inclusion (intersection) is called trivial modulo (S, [w]), de-
noted by (f, g)[w] ≡ 0 mod(S, [w]), if
(f, g)[w] =∑
αi[aisibi],
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi] normal si-diword, [ai[si]bi] < [w] and
each [aisibi] is right (left) normed si-diword whenever either both f and [agb] or both
[fb] and [ag] are right (left) normed S-diwords.
We call the set S a Grobner-Shirshov basis in D(X) if any composition of polynomials
in S is trivial modulo S (and [w]).
The following lemmas play key role in the proof of Theorem 8.14.
Lemma 8.11 Let S ⊂ D(X) and [asb] an s-diword, s ∈ S. Assume that each composition
of right and left multiplication is trivial modulo S. Then, [asb] has a presentation:
[asb] =∑
αi[aisibi],
where each αi ∈ k, si ∈ S, ai, bi ∈ X∗ and each [aisibi] is normal si-diword.
121
Proof. Following Proposition 8.3, we assume that
[asb] = x−m ` · · · ` x−1 ` x0 a x1 a · · · a xn|xk 7→s.
There are three cases to consider.
Case 1. k = 0. Then [asb] is a normal s-diword.
Case 2. k < 0. Then [asb] = a ` (s ` xk+1) ` b, k < −1 or [asb] = a ` (s ` x0) a b.If s is left normed then [asb] is a normal s-diword. If s is not left normed then for the
composition s ` xk+1 (k < 0) of right multiplication, we have
s ` xk+1 =∑
αi[aisibi],
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S and [aisibi] is left normed si-diword. Then
[asb] =∑
αi(a ` [aisibi] ` b)
or
[asb] =∑
αi(a ` [aisibi] a b)
is a linear combination of normal si-diwords.
Case 3. k > 0 is similar to the Case 2.
Lemma 8.12 Let S ⊂ D(X) and each composition (f, g)[w] in S of inclusion (inter-
section) trivial modulo (S, [w]). Let [a1s1b1] and [a2s2b2] be normal S-diwords such that
[w] = [a1[s1]b1] = [a2[s2]b2], where s1, s2 ∈ S, a1, a2, b1, b2 ∈ X∗. Then,
[a1s1b1] ≡ [a2s2b2] mod(S, [w]),
i.e., [a1s1b1] − [a2s2b2] =∑αi[aisibi], where each αi ∈ k, ai, bi ∈ X∗, si ∈ S, [aisibi]
normal si-diword and [ai[si]bi] < [w].
Proof. In the following, all letters a, b, c with indexis are words and s1, s2, sj ∈ S.
Because a1s1b1 = a2s2b2 as ordinary words, there are three cases to consider.
Case 1. Subwords s1, s2 have empty intersection. Assume, for example, that b1 =
bs2b2 and a2 = a1s1b. Because any normal S-diword may be bracketing in any way, we
have
[a2s2b2]− [a1s1b1] = (a1s1(b(s2 − [s2])b2))− ((a1(s1 − [s1])b)s2b2).
For any [t] ∈ supp(s2 − [s2]), we prove that (a1s1b[t]b2) is a normal s1-diword. There are
five cases to consider.
1.1 [w] = [a1[s1]b[s2]b2];
122
1.2 [w] = [a1˙[s1]b[s2]b2];
1.3 [w] = [a1[s1]b[s2]b2];
1.4 [w] = [a1[s1]b ˙[s2]b2];
1.5 [w] = [a1[s1]b[s2]b2].
For 1.1, since [a1s1b1] and [a2s2b2] are normal S-diwords, both s1 and s2 are right
normed by the definition, in particular, [t] is right normed. It follows that (a1s1b[t]b2) =
[a1s1b[t]b2] is a normal s1-diword.
For 1.2, it is clear that (a1s1b[t]b2) is a normal s1-diword and [t] is right normed.
For 1.3, 1.4 and 1.5, since [a1s1b1] is normal s1-diword, s1 is left normed by the
definition, which implies that (a1s1b[t]b2) is a normal s1-diword. Moreover, [t] is right
normed, if 1.3, and left normed, if 1.5.
Clearly, for all cases, we have [a1s1b[t]b2] = [a1[s1]b[t]b2] < [a1[s1]b[s2]b2] = [w].
Similarly, for any [t] ∈ supp(s1−[s1]), (a1[t]bs2b2) is a normal s2-diword and [a1[t]b[s2]b2] <
[w].
Case 2. Subwords s1 and s2 have non-empty intersection c. Assume, for example,
that b1 = bb2, a2 = a1a, w1 = s1b = as2 = acb.
There are following five cases to consider:
2.1 [w] = [a1[s1]bb2];
2.2 [w] = [a1[s1]bb2];
2.3 [w] = [a1acbb2];
2.4 [w] = [a1acbb2];
2.5 [w] = [a1acbb2].
Then
[a2s2b2]− [a1s1b1] = (a1([as2]− [s1b])b2) = (a1(s1, s2)[w1]b2),
where [w1] = [acb] = [[s1]b] = [a[s2]] is as follows:
2.1 [w1] is right normed;
2.2 [w1] is left normed;
2.3 [w1] = [acb];
2.4 [w1] = [acb];
2.5 [w1] = [acb].
Since each composition (f, g)[w] in S is trivial modulo (S, [w]), there exist βj ∈k, uj, vj ∈ X∗, sj ∈ S such that [s1b] − [as2] =
∑j βj[ujsjvj], where each [ujsjvj] is
normal S-diword and [uj[sj]vj] < [w1] = [acb]. Therefore,
[a2s2b2]− [a1s1b1] =∑
j
βj(a1[ujsjvj]b2).
123
Now, we prove that each (a1[ujsjvj]b2) is normal sj-diword and (a1[ujsjvj]b2) < [w] =
[a1[[s1]b]b2].
For 2.1, since [a1s1bb2] and [a1as2b2] are normal S-diwords, both [s1b] and [as2] are
right normed S-diwords. Then, by definition, each [ujsjvj] is right normed S-diword, and
so each (a1[ujsjvj]b2) = [a1ujsjvjb2] is normal S-diword.
For 2.2, both [s1b] and [as2] must be left normed S-diwords. Then, by definition,
each [ujsjvj] is left normed S-diword, and so each (a1[ujsjvj]b2) = [a1ujsjvj b2] is normal
S-diword.
For 2.3, 2.4 or 2.5, by noting that (a1[ujsjvj]b2) = ((a1) ` [ujsjvj] a (b2)) and [ujsjvj]
is normal S-diword, (a1[ujsjvj]b2) is also normal S-diword.
Now, for all cases, we have [a1ujsjvjb2] = [a1uj[sj]vjb2] < [w] = [a1[acb]b2].
Case 3. One of the subwords s1 and s2 contains another as a subword. Assume, for
example, that b2 = bb1, a2 = a1a, w1 = s1 = as2b.
Again there are following five cases to consider:
2.1 [w] = [a1a[s2]bb1];
2.2 [w] = [a1a[s2]bb1];
2.3 [w] = [a1a[s2]bb1];
2.4 [w] = [a1a ˙[s2]bb1];
2.5 [w] = [a1a[s2]bb1].
Then
[a1s1b1]− [a2s2b2] = (a1(s1 − as2b)b1) = (a1(s1, s2)[w1]b1).
It is similar to the proof of the Case 2 that we have [a1s1b1] ≡ [a2s2b2] mod(S, [w]).
Let S ⊂ D(X). Denote
Irr(S) = u ∈ [X∗]|u 6= [a[s]b], s ∈ S, a, b ∈ X∗, [asb] is normal s-diword.
Lemma 8.13 Let S ⊂ D(X) and h ∈ D(X). Then h has a representation
h =∑I1
αi[ui] +∑I2
βj[ajsjbj]
where [ui] ∈ Irr(S), i ∈ I1, [ajsjbj] normal sj-diwords, sj ∈ S, j ∈ I2 with [a1[s1]b1] >
[a2[s2]b2] > · · · > [an[sn]bn].
Proof. Let h = α1[h] + · · · . We prove the result by induction on [h].
124
If [h] ∈ Irr(S), then take [u1] = [h] and h1 = h−α1[u1]. Clearly, [h1] < [h] or h1 = 0.
If [h] 6∈ Irr(S), then [h] = [a1[s1]b1] with [a1s1b1] a normal s1-diword. Let h1 =
h− β1[a1s1b1]. Then [h1] < [h] or h1 = 0.
The following theorem is the main result.
Theorem 8.14 (Composition-Diamond lemma for dialgebras) Let S ⊂ D(X) be a monic
set and the ordering < as before, Id(S) is the ideal of D(X) generated by S. Then
(i)⇒ (ii)⇔ (ii)′ ⇔ (iii), where
(i) S is a Grobner-Shirshov basis in D(X).
(ii) f ∈ Id(S)⇒ [f ] = [a[s]b] for some s ∈ S, a, b ∈ X∗ and [asb] a normal S-diword.
(ii)′ f ∈ Id(S)⇒ f = α1[a1s1b1]+α2[a2s2b2]+· · ·+αn[ansnbn] with [a1[s1]b1] > [a2[s2]b2] >
· · · > [an[sn]bn], where [aisibi] is normal si-diword, i = 1, 2, · · · , n.
(iii) The set Irr(S) is a linear basis of the dialgebra D(X|S) = D(X)/Id(S) generated
by X with defining relations S.
Proof. (i) ⇒ (ii). Let S be a Grobner-Shirshov basis and 0 6= f ∈ Id(S). We may
assume, by Lemma 8.11, that
f =n∑
i=1
αi[aisibi],
where each αi ∈ k, ai, bi ∈ X∗, si ∈ S and [aisibi] normal S-diword. Let
[wi] = [ai[si]bi], [w1] = [w2] = · · · = [wl] > [wl+1] ≥ · · · , l ≥ 1.
We will use induction on l and [w1] to prove that [f ] = [a[s]b] for some s ∈ S and a, b ∈ X∗.
If l = 1, then [f ] = [a1s1b1] = [a1[s1]b1] and hence the result holds. Assume that l ≥ 2.
Then, by Lemma 8.12, we have [a1s1b1] ≡ [a2s2b2] mod(S, [w1]).
Thus, if α1 + α2 6= 0 or l > 2, then the result follows from induction on l. For the
case α1 + α2 = 0 and l = 2, we use induction on [w1]. Now, the result follows.
(ii)⇒ (ii)′. Assume (ii) and 0 6= f ∈ Id(S). Let f = α1[f ] +∑
[ui]<[f ] αi[ui]. Then,
by (ii), [f ] = [a1[s1]b1], where [a1s1b1] is a normal S-diword. Therefore,
f1 = f − α1[a1s1b1], [f1] < [f ] or f1 = 0, f1 ∈ Id(S).
Now, by using induction on [f ], we have (ii)′.
(ii)′ ⇒ (ii). This part is clear.
125
(ii)⇒ (iii). Assume (ii). Then by Lemma 8.13, Irr(S) spans D(X|S) as k-space.
Suppose that 0 6=∑αi[ui] ∈ Id(S) where [u1] > [u2] > · · · , [ui] ∈ Irr(S). Then by
(ii), [u1] = [a1[s1]b1] where [a1s1b1] is a normal S-diword, a contradiction.
This shows (iii).
(iii) ⇒ (ii). Assume (iii). Let 0 6= f ∈ Id(S). Since the elements in Irr(S) are
linearly independent in D(X|S), by Lemma 8.13, [f ] = [a[s]b], where [asb] is a normal
S-diword. Thus, (ii) follows.
Remark In general, (iii) 6⇒ (i). For example, it is noted that
Irr(S) = xj a xi1 a · · · a xik | j ∈ I, ip ∈ I − I0, 1 ≤ p ≤ k, i1 ≤ · · · ≤ ik, k ≥ 0
is a linear basis of D(X|S) in Theorem 8.17. Let
S1 = xj ` xi − xi a xj + xi, xj, xt a xi0 , i, j, t ∈ I, i0 ∈ I0.
Then Irr(S1) = Irr(S) is a linear basis of D(X|S). But in the proof of Theorem 8.17,
we know that S1 is not a Grobner-Shirshov basis of D(X|S).
§8.3 Applications
In this section, we give Grobner-Shirshov bases for the universal enveloping dialgebra
of a Leibniz algebra (PBW theorem), the bar extension of a dialgebra, the free product
of two dialgebras, and the Clifford dialgebra. By using Theorem 8.14, we obtain some
normal forms for dialgebras mentioned the above.
Definition 8.15 A k-linear space L equipped with bilinear multiplication [, ] is called a
Leibniz algebra if for any a, b, c ∈ L,
[[a, b], c] = [[a, c], b] + [a, [b, c]]
i.e., the Leibniz identity is valid in L.
It is clear that if (D,a,`) is a dialgebra then D(−) = (D, [, ]) is a Leibniz algebra,
where [a, b] = a a b− b ` a for any a, b ∈ D.
If f is a Leibniz polynomial in variables X, then by f (−) we mean a dialgebra poly-
nomial in X obtained from f by transformation [a, b] 7→ a a b− b ` a.
126
Definition 8.16 Let L be a Leibniz algebra. A dialgebra U(L) together with a Leibniz
homomorphism ε : L → U(L) is called the universal enveloping dialgebra for L, if the
following diagram commute:
-
?
∃!f
D
U(L)L ε
∀δ
where D is a dialgebra, δ is a Leibniz homomorphism and f : U(L) → D is a dialgebra
homomorphism such that fε = δ.
An equivalent definition is as follows: Let L = Lei(X|S) is a Leibniz algebra presented
by generators X and definition relations S. Then U(L) = D(X|S(−)) is the dialgebra with
generators X and definition relations S(−) = s(−)|s ∈ S.
Theorem 8.17 Let L be a Leibniz algebra over a field k with the product , . Let L0 be
the subspace of L generated by the set a, a, a, b + b, a | a, b ∈ L. Let xi|i ∈ I0be a basis of L0 and X = xi|i ∈ I a well-ordered basis of L such that I0 ⊆ I. Let
U(L) = D(X|xi a xj − xj ` xi−xi, xj) be the universal enveloping dialgebra for L and
the ordering < on [X∗] as before. Then
(i) D(X|xi a xj − xj ` xi − xi, xj) = D(X|S), where S consists of the following
polynomials:
(a) fji = xj ` xi − xi a xj + xi, xj (i, j ∈ I)
(b) fji`t = xj ` xi ` xt − xi ` xj ` xt + xi, xj ` xt (i, j, t ∈ I, j > i)
(c) hi0`t = xi0 ` xt (i0 ∈ I0, t ∈ I)
(d) ftaji = xt a xj a xi − xt a xi a xj + xt a xi, xj (i, j, t ∈ I, j > i)
(e) htai0 = xt a xi0 (i0 ∈ I0, t ∈ I)
(ii) S is a Grobner-Shirshov basis in D(X).
(iii) The set
xj a xi1 a · · · a xik | j ∈ I, ip ∈ I − I0, 1 ≤ p ≤ k, i1 ≤ · · · ≤ ik, k ≥ 0
is a linear basis of the universal enveloping algebra U(L). In particular, L is a Leibniz
subalgebra of U(L).
127
Proof. (i) By using the following
fji`t = fji ` xt and fji ` xt + fij ` xt = (xi, xj+ xj, xi) ` xt,
we have (b) and (c) are in Id(fji). By symmetry, (d) and (e) are in Id(fji). This shows
(i).
(ii) We will prove that all compositions in S are trivial modulo S (and [w]). For conve-
nience, we extend linearly the functions fji, fji`t, ftaji, hi0`t and htai0 to fjp,q (fp,qi), fji`p,q
and hp,qai0 , etc respectively. For example, if xp, xq =∑αs
pqxs, then
fjp,q = xj ` xp, xq − xp, xq a xj + xp, xq, xj =∑
αspqfjs,
fji`p,q =∑
αspq(xj ` xi ` xs − xi ` xj ` xs + xi, xj ` xs) = fji ` xp, xq,
hp,qai0 =∑
αspqhsai0 .
By using the Leibniz identity,
a, b, c = a, b, c+ a, c, b, (8.4)
we have
a, b, b = 0 and a, b, c+ c, b = 0
for any a, b, c ∈ L. It means that for any i0 ∈ I0, j ∈ I,
xj, xi0 = 0 (8.5)
and by noting that xi0 , xj = xj, xi0+ xi0 , xj, we have
xi0 , xj ∈ L0. (8.6)
This implies that L0 is an ideal of L. Clearly, L/L0 is a Lie algebra.
The formulas (8.4), (8.5) and (8.6) are useful in the sequel.
In S, all the compositions are as follows.
1) Compositions of left or right multiplication.
All possible compositions in S of left multiplication are ones related to (a), (b) and
(c).
By noting that for any s, i, j, t ∈ I, we have
xs a fji = fsaji (j > i),
xs a fji = −fsaij + xs a (xi, xj+ xj, xi) (j < i),
xs a fii = xs a xi, xi,
xs a fji`t = fsaji a xt (j > i) and
xs a hi0`t = hsai0 a xt,
128
it is clear that all cases are trivial modulo S.
By symmetry, all compositions in S of right multiplication are trivial modulo S.
2) Compositions of inclusion and intersection.
We denote, for example, (a ∧ b) the composition of the polynomials of type (a) and
type (b). It is noted that since (b) and (c) are both left normed, we have to prove that
the corresponding compositions of the cases of (b∧b), (b∧c), (c∧c) and (c∧b) must be a
linear combination of left normed S-diwords in which the leading term of each S-diword
is less than w. Symmetrically, we consider the cases for the right normed (d) and (e).
All possible compositions of inclusion and intersection are as follows.
(a ∧ c) [w] = xi0 ` xi (i0 ∈ I0). We have, by (8.5),
(fi0i, hi0`i)[w] = −xi a xi0 + xi, xi0 = −hiai0 ≡ 0 mod(S, [w]).
(a ∧ d) [w] = xj ` xi a xq a xp (q > p). We have
(fji, fiaqp)[w]
= −xi a xj a xq a xp + xi, xj a xq a xp + xj ` xi a xp a xp − xj ` xi a xp, xq
= −xi a fjaqp + fi,jaqp + fji a xp a xq − fji a xp, xq
≡ 0 mod(S, [w]).
(a ∧ e) [w] = xj ` xi a xi0 (i0 ∈ I0). We have
(fji, hiai0)[w] = −xi a xj a xi0 + xi, xj a xi0 = −xi a hjai0 + hi,jai0 ≡ 0 mod(S, [w]).
(b ∧ a) There are two cases to consider: [w] = xj ` xi ` xt and [w] = xj ` xi ` xt ` xp.
For [w] = xj ` xi ` xt (j > i), by (8.4), we have
(fji`t, fit)[w] = −xi ` xj ` xt + xi, xj ` xt + xj ` xt a xi − xj ` xt, xi
= −xi ` fjt + fi,jt + fjt a xi − fjt,i + fit,j − fit a xj + ftaji
≡ 0 mod(S, [w]).
For [w] = xj ` xi ` xt ` xp (j > i), we have
(fji`t, ftp)[w]
= −xi ` xj ` xt ` xp + xi, xj ` xt ` xp + xj ` xi ` xp a xt − xj ` xi ` xp, xt
= −xi ` xj ` ftp + xi, xj ` ftp + fji`p a xt − fji`p,t
≡ 0 mod(S, [w]).
129
(b ∧ b) There are two cases to consider: [w] = xj ` xi ` xt ` xs ` xp and [w] = xj ` xi `xt ` xp.
For [w] = xj ` xi ` xt ` xs ` xp (j > i, t > s), we have
(fji`t, fts`p)[w]
= −xi ` xj ` xt ` xs ` xp + xi, xj ` xt ` xs ` xp + xj ` xi ` xs ` xt ` xp
−xj ` xi ` xs, xt ` xp
= −xi ` xj ` fts`p + xi, xj ` fts`p + fji`s ` xt ` xp − fji`s,t ` xp
≡ 0 mod(S, [w])
since it is a combination of left normed S-diwords in which the leading term of each
S-diword is less than w.
For [w] = xj ` xi ` xt ` xp (j > i > t), suppose that
xi, xj =∑m∈I1
αmijxm + αt
ijxt +∑n∈I2
αnijxn (m < t < n).
Denote
Bt`i,j`p = xt ` xi, xj ` xp − xi, xj ` xt ` xp − xt, xi, xj ` xp.
Then
Bt`i,j`p =∑m∈I1
αmij ftm`p −
∑n∈I2
αnijfnt`p −
∑q∈I0
βqhq`p
is a linear combination of left normed S-diwords of length 2 or 3, where∑q∈I0
βqxq =∑m∈I1
αmij (xt, xm+ xm, xt) + αt
ijxt, xt.
Denote∑l∈I0
γlxl = −(xj, xt, xi+ xt, xi, xj) + (xi, xt, xj+ xt, xj, xi).
Now, by (8.4), we have
(fji`t, fit`p)[w]
= −xi ` xj ` xt ` xp + xi, xj ` xt ` xp + xj ` xt ` xi ` xp − xj ` xt, xi ` xp
= −xi ` fjt`p −Bt`i,j`p + fjt`i ` xp −Bj`t,i`p +∑l∈I0
γlhl`p
+Bi`t,j`p − fit`j ` xp + xt ` fji`p
≡ 0 mod(S, [w])
since it is a combination of left normed S-diwords in which the leading term of each
S-diword is less than w.
130
(b ∧ c) There are three cases to consider: [w] = xj ` xi0 ` xt (i0 ∈ I0), [w] = xj0 ` xi `xt (j0 ∈ I0) and [w] = xj ` xi ` xt0 ` xn (t0 ∈ I0).
Case 1. [w] = xj ` xi0 ` xt (j > i0, i0 ∈ I0). By (8.6), we can assume that
xi0 , xj =∑
l∈I0γlxl. Then, we have
(fji0`t, hi0`t)[w] = −xi0 ` xj ` xt+xi0 , xj ` xt = −hi0`j ` xt+∑l∈I0
γlhl`t ≡ 0 mod(S, [w]).
Case 2. [w] = xj0 ` xi ` xt (j0 > i, j0 ∈ I0). By (8.5), we have
(fj0i`t, hj0`i)[w] = −xi ` xj0 ` xt + xi, xj0 ` xt = −xi ` hj0`t ≡ 0 mod(S, [w]).
Case 3. [w] = xj ` xi ` xt0 ` xn (j > i, t0 ∈ I0). We have
(fji`t0 , ht0`n)[w] = −xi ` xj ` xt0 ` xn + xi, xj ` xt0 ` xn
= (−xi ` xj + xi, xj) ` ht0`n
≡ 0 mod(S, [w]).
(b ∧ d) [w] = xj ` xi ` xt a xq a xp (j > i, q > p). We have
(fji`t, ftaqp)[w]
= −xi ` xj ` xt a xq a xp + xi, xj ` xt a xq a xp
+xj ` xi ` xt a xp a xq − xj ` xi ` xt a xp, xq
= −xi ` xj ` ftaqp + xi, xj ` ftaqp + fji`t a xp a xq − fji`t a xp, xq
≡ 0 mod(S, [w]).
(b ∧ e) [w] = xj ` xi ` xt a xn0 (j > i, n0 ∈ I0). We have
(fji`t, htan0)[w] = −xi ` xj ` xt a xn0 + xi, xj ` xt a xn0
= (−xi ` xj + xi, xj) ` htan0
≡ 0 mod(S, [w]).
(c ∧ a) There are two cases to consider: [w] = xn0 ` xt (n0 ∈ I0) and [w] = xn0 ` xt `xs (n0 ∈ I0).
For [w] = xn0 ` xt (n0 ∈ I0), we have
(hn0`t, fn0t)[w] = xt a xn0 − xt, xn0 = htan0 ≡ 0 mod(S, [w]).
For [w] = xn0 ` xt ` xs (n0 ∈ I0), we have
(hn0`t, fts)[w] = xn0 ` xs a xt − xn0 ` xs, xt = hn0`s a xt − hn0`s,t ≡ 0 mod(S, [w]).
131
(c ∧ b) [w] = xn0 ` xt ` xs ` xp (t > s, n0 ∈ I0). We have
(hn0`t, fts`p)[w] = xn0 ` xs ` xt ` xp − xn0 ` xs, xt ` xp
= hn0`s ` xt ` xp − hn0`s,t ` xp
≡ 0 mod(S, [w]).
(c ∧ c) [w] = xn0 ` xt0 ` xr (n0, t0 ∈ I0). We have
(hn0`t0 , ht0`r)[w] = 0.
(c ∧ d) [w] = xn0 ` xt a xq a xp (q > p, n0 ∈ I0). We have
(hn0`t, ftaqp)[w] = xn0 ` xt a xp a xq − xn0 ` xt a xp, xq
= hn0`t a (xp a xq − xp, xq)
≡ 0 mod(S, [w]).
(c ∧ e) [w] = xn0 ` xt a xs0 (n0, s0 ∈ I0). We have
(hn0`t, htas0)[w] = 0.
Since (d∧ d), (d∧ e), (e∧ d), (e∧ e) are symmetric with (b∧ b), (b∧ c), (c∧ b), (c∧ c)respectively, they have the similar representations. We omit the details.
So, we show that S is a Grobner-Shirshov basis.
(iii) Clearly, the mentioned set is just the set Irr(S). Now, the results follow from
Theorem 8.14.
Recall a Grobner-Shirshov basis S is called minimal if S has no inclusion composition
in S.
Remark Let the notation be in Theorem 8.17. Let Smin consist of the following polyno-
mials:
(a) fji = xj ` xi − xi a xj + xi, xj (i ∈ I, j ∈ I − I0)
(b) fji`t = xj ` xi ` xt − xi ` xj ` xt + xi, xj ` xt (i, j ∈ I − I0, j > i, t ∈ I)
(c) hi0`t = xi0 ` xt (i0 ∈ I0, t ∈ I)
(d) ftaji = xt a xj a xi − xt a xi a xj + xt a xi, xj (i, j ∈ I − I0, j > i, t ∈ I)
(e) htai0 = xt a xi0 (i0 ∈ I0, t ∈ I)
Then Smin is a minimal Grobner-Shirshov basis for D(X|S).
We have the following corollary.
132
Corollary 8.18 Let the notation be as in Theorem 8.17. Then as linear spaces, U(L) is
isomorphic to L⊗U(L/L0), where U(L/L0) is the universal enveloping of the Lie algebra
L/L0.
Proof. Clearly, xj | j ∈ I − I0 is a k-basis of the Lie algebra L/L0. It is well known
that the universal enveloping U(L/L0) of the Lie algebra L/L0 has a k-basis
xi1xi2 . . . xik | i1 ≤ · · · ≤ ik, ip ∈ I − I0, 1 ≤ p ≤ k, k ≥ 0.
By using (iii) in Theorem 8.17, the result follows.
Definition 8.19 Let D be a dialgebra. An element e ∈ D is called a bar unit of D if
e ` x = x a e = x for any x ∈ D.
Theorem 8.20 Each dialgebra has a bar unit extension.
Proof. Let (D,`,a) be an arbitrary dialgebra over a field k and A the ideal of D
generated by the set a a b − a ` b| a, b ∈ D. Let X0 = xi0|i0 ∈ I0 be a k-basis
of A and X = xi|i ∈ I a well-ordered k-basis of D such that I0 ⊆ I. Then D has
a presentation by the multiplication table D = D(X|S), where S = xi ` xj − xi `xj, xi a xj −xi a xj, i, j ∈ I, where xi ` xj and xi a xj are linear combinations
of xt, t ∈ I.Let D1 = D(X ∪ e|S1), where S1 = S ∪ e ` y − y, y a e− y, e a x0, x0 ` e | y ∈
X ∪ e, x0 ∈ X0. Then D1 is a dialgebra with a bar unit e.
Denote
1. fi`j = xi ` xj − xi ` xj,
2. fiaj = xi a xj − xi a xj,
3. ge`y = e ` y − y,
4. gyae = y a e− y,
5. hxi0`e = xi0 ` e,
6. heaxi0= e a xi0 ,
where i, j ∈ I, i0 ∈ I0, y ∈ X ∪ e.We show that xt a xi0 = 0 and xi0 ` xt = 0 for any t ∈ I, i0 ∈ I0.Since xi0 ∈ A, we have xi0 =
∑αi(cifidi), where fi = ai a bi − ai ` bi, αi ∈ k,
ai, bi ∈ D and ci, di ∈ X∗.
133
Since xt a (ci(ai a bi − ai ` bi)di) = 0, we have xt a ciai a bi − ai ` bidi = 0
for each i. Then xt a xi0 = 0.
By symmetry, we have xi0 ` xt = 0.
To prove the theorem, by using our Theorem 8.14, it suffices to prove that with the
ordering on [(X ∪ e)∗] as before, where x < e, x ∈ X, S1 is a Grobner-Shirshov basis
in D(X ∪ e). Now, we show that all compositions in S1 are trivial.
All possible compositions of left and right multiplication are: z a fi`j, z a ge`y,
z a hxi0`e, fiaj ` z, gyae ` z, heaxi0
` z, z ∈ X ∪ e.For z a fi`j, z = xt ∈ X, since (xt a xi) a xj = xt a (xi ` xj), we have xt a xi a
xj = xt a xi ` xj and
xt a fi`j
= xt a xi a xj − xt a xi ` xj
= ftai a xj + ftaiaj − ftai`j + xt a xi a xj − xt a xi ` xj
= ftai a xj + ftaiaj − ftai`j
≡ 0 mod(S1).
For z a fi`j, z = e, let xi a xj − xi ` xj =∑αi0xi0 . Then
e a fi`j = e a xi a xj − e a xi ` xj
= e a (xi a xj − xi a xj) + e a xi a xj − e a xi ` xj
= e a fiaj +∑
αi0heaxi0
≡ 0 mod(S1).
For z a ge`y, we have
z a ge`y = z a e a y − z a y = (z a e− z) a y = gzae a y ≡ 0 mod(S1).
For z a hxi0`e, we have
z a hxi0`e = z a xi0 a e = z a gxi0
ae + z a xi0 .
It is clear that z a xi0 = heaxi0if z = e and z a xi0 = xt a xi0 − xt a xi0 = ftai0 if
z = xt ∈ X, since xt a xi0 = 0. This implies that z a hxi0`e ≡ 0 mod(S1).
Thus we show that all compositions of left multiplication in S1 are trivial modulo S1.
By symmetry, all compositions of right multiplication in S1 are trivial modulo S1.
Now, all possible ambiguities [w] of compositions of intersection in S1 are:
1 ∧ 1, [xixjxt]; 1 ∧ 2, [xixjxt]; 1 ∧ 4, [xixje]; 1 ∧ 5, [xixi0 e].
134
2 ∧ 2, [xixjxt]; 2 ∧ 4, [xixje].
3 ∧ 1, [exixj]; 3 ∧ 2, [exixj]; 3 ∧ 3, [eey]; 3 ∧ 4, [eye]; 3 ∧ 5, [exi0 e]; 3 ∧ 6, [eexi0 ].
4 ∧ 4, [yee]; 4 ∧ 6, [yexi0 ].
5 ∧ 3, [xi0ey]; 5 ∧ 4, [xi0 ee]; 5 ∧ 6, [xi0 exj0 ].
6 ∧ 2, [exi0xj]; 6 ∧ 4, [exi0e].
In the above, all i, j, t ∈ I, i0, j0 ∈ I0 and y ∈ X ∪ e.There is no composition of inclusion in S1.
We will show that all compositions of intersection in S1 are trivial. We check only
the cases of 1 ∧ 2, 1 ∧ 5 and 4 ∧ 6. Others can be similarly proved.
For 1 ∧ 2, [w] = [xixjxt], since (xi ` xj) a xt = xi ` (xj a xt), we have xi ` xj axt = xi ` xj a xt and
(1 ∧ 2)[w] = −xi ` xj a xt + xi ` xj a xt
= −fi`jat + fi`jat − xi ` xj a xt+ xi ` xj a xt
= −fi`jat + fi`jat
≡ 0 mod(S1, [w]).
For 1 ∧ 5, [w] = [xixi0 e], since xi ` xi0 ∈ A, we have xi ` xi0 =∑αj0xj0 and
(1 ∧ 5)[w] = xi ` xi0 ` e =∑
αj0hxj0`e ≡ 0 mod(S1, [w]).
For 4 ∧ 6, [w] = [yexi0 ], we have (4 ∧ 6)[w] = −heaxi0if y = e and (4 ∧ 6)[w] = −ftai0
if y = xt ∈ X since xt a xi0 = 0. Then (4 ∧ 6)[w] ≡ 0 mod(S1, [w]).
Then all the compositions in S1 are trivial.
The proof is complete.
Remark Let the notation be as in the proof of Theorem 8.20. Let D′ = D(X ∪ejJ |S ′)be a dialgebra, where S ′ = S∪ej ` y−y, y a ej−y, ej a x0, x0 ` ej | y ∈ X∪ejJ , x0 ∈X0, j ∈ J. Let J be a well-ordered set. Then with the ordering on [(X ∪ ejJ)∗] as
before, where xi < ej for all i ∈ I, j ∈ J , by a similar proof of Theorem 8.20, S ′ is a
Grobner-Shirshov basis in D(X ∪ ejJ). It follows from Theorem 8.14 that D can be
embedded into the dialgebra D′ while D′ has bar units ejJ .
Definition 8.21 Let D1, D2 be dialgebras over a field k. The dialgebra D1 ∗D2 with two
dialgebra homomorphisms ε1 : D1 → D1 ∗D2, ε2 : D2 → D1 ∗D2 is called the free product
of D1, D2, if the following diagram commute:
135
-
?
@@
@@
@@R
D
∃!f
D1 ∗D2D1 D2ε1 ε2
∀δ1 ∀δ2
where D is a dialgebra, δ1, δ2 are dialgebra homomorphisms and f : D1 ∗ D2 → D is a
dialgebra homomorphism such that fε1 = δ1, fε2 = δ2.
An equivalent definition is as follows: Let Di = D(Xi|Si) be a presentation by genera-
tors and defining relations with X1∩X2 = ∅, i = 1, 2. Then D1∗D2 = D(X1∪X2|S1∪S2).
Let (D1,`,a), (D2,`,a) be two dialgebras over a field k, A1 the ideal of D1 generated
by the set a a b − a ` b| a, b ∈ D1 and A2 the ideal of D2 generated by the set
c a d− c ` d| c, d ∈ D2. Let X0 = xi0|i0 ∈ I0 be a k-basis of A1 and X = xi|i ∈ Ia well-ordered k-basis of D1 such that I0 ⊆ I. Let Y0 = yl0|l0 ∈ J0 be a k-basis of A2
and Y = yl|l ∈ J a well-ordered k-basis of D2 such that J0 ⊆ J . Then D1 and D2 have
multiplication tables:
D1 = D(X|S1), S1 = xi ` xj − xi ` xj, xi a xj − xi a xj, i, j ∈ I,
D2 = D(Y |S2), S2 = yl ` ym − yl ` ym, yl a ym − yl a ym, l,m ∈ J.
The free product D1 ∗D2 of D1 and D2 is
D1 ∗D2 = D(X ∪ Y |S1 ∪ S2).
We order X∪Y by xi < yj for any i ∈ I, j ∈ J . Then we have the following theorem.
Theorem 8.22 (i) S is a Grobner-Shirshov basis of D1 ∗D2 = D(X ∪Y |S1 ∪S2), where
S consists of the following relations:
1. fxi`xj= xi ` xj − xi ` xj, i, j ∈ I,
2. fxiaxj= xi a xj − xi a xj, i, j ∈ I,
3. fyl`ym = yl ` ym − yl ` ym, l,m ∈ J,
4. fylaym = yl a ym − yl a ym, l,m ∈ J,
5. hxi0`yl
= xi0 ` yl, i0 ∈ I0, l ∈ J,
6. hylaxi0= yl a xi0 , i0 ∈ I0, l ∈ J,
7. hyl0`xi
= yl0 ` xi, i ∈ I, l0 ∈ J0,
8. hxiayl0= xi a yl0 , i ∈ I, l0 ∈ J0.
136
(ii) Irr(S), which is a k-linear basis of D1∗D2, consists of all elements z−m · · · z−1z0z1 · · · zn,
where m,n ≥ 0, z0 ∈ X∪Y, zi ∈ (X\X0)∪(Y \Y0),−m ≤ i ≤ n, i 6= 0, neither zj, zj+1 ⊆X nor zj, zj+1 ⊆ Y,−m ≤ j ≤ n− 1.
Proof. By the proof of Theorem 8.20, we have xi a xi0 = 0, xi0 ` xi = 0, yl ayl0 = 0 and yl0 ` yl = 0 for any i ∈ I, i0 ∈ I0, l ∈ J, l0 ∈ J0.
Firstly, we prove that hylaxi0∈ Id(S1 ∪ S2) for any i0 ∈ I0, l ∈ J .
Since yl a (ci(ai a bi−ai ` bi)di) = yl a (ci((ai a bi−ai a bi)− (ai ` bi−ai `bi)di) ∈ Id(S1 ∪S2), we have yl a ciai a bi− ai ` bidi ∈ Id(S1 ∪S2) for all i, l. Then
hylaxi0∈ Id(S1 ∪ S2).
Similarly, we have hxi0`yl, hyl0
`xi, hxiayl0
∈ Id(S1 ∪ S2) for any i ∈ I, i0 ∈ I0, l ∈J, l0 ∈ J0.
Secondly, we will show that all compositions in S are trivial.
All possible compositions of left and right multiplication are: z a fxi`xj, z a
fyl`ym , z a hxi0`yl, z a hyl0
`xi, fxiaxj
` z, fylaym ` z, hylaxi0` z, hxiayl0
` z, where
z ∈ X ∪ Y .
By a similar proof in Theorem 8.20, all compositions of left and right multiplication
mentioned the above are trivial modulo S.
Now, all possible ambiguities [w] of compositions of intersection in S are:
1 ∧ 1, [xixjxt]; 1 ∧ 2, [xixjxt]; 1 ∧ 5, [xixi0 yl]; 1 ∧ 8, [xixjyl0 ].
2 ∧ 2, [xixjxt]; 2 ∧ 8, [xixjyl0 ].
3 ∧ 3, [ylymyt]; 3 ∧ 4, [ylymyt]; 3 ∧ 6, [ylymxi0 ]; 3 ∧ 7, [ymyl0xi].
4 ∧ 4, [ylymyt]; 4 ∧ 6, [ylymxi0 ].
5 ∧ 3, [xi0ylyt]; 5 ∧ 4, [xi0 ylyt]; 5 ∧ 6, [xi0 ylxj0 ]; 5 ∧ 7, [xi0yl0 xt].
6 ∧ 2, [ylxi0xt]; 6 ∧ 8, [ymxi0yl0 ].
7 ∧ 1, [yl0xixj]; 7 ∧ 2, [yl0xixj]; 7 ∧ 5, [yl0xi0 ym]; 7 ∧ 8, [yl0xiym0 ].
8 ∧ 4, [xiyl0yt]; 8 ∧ 6, [xiyl0xi0 ].
There is no composition of inclusion in S.
We will show that all compositions of intersection in S are trivial. We check only the
cases of 1 ∧ 5 and 2 ∧ 8. Others can be similarly proved.
For 1 ∧ 5, [w] = [xixi0 yl], let xi ` xi0 =∑αt0xt0 . Then
(1 ∧ 5)[w] = −xi ` xi0 ` yl = −∑
αt0hxt0`yl≡ 0 mod(S, [w]).
For 2 ∧ 8, [w] = [xixjyl0 ], let xi a xj =∑αtxt. Then
(2 ∧ 8)[w] = −xi a xj a yl0 = −∑
αthxtayl0≡ 0 mod(S, [w]).
137
Then all the compositions in S are trivial. This show (i).
(ii) follows from our Theorem 8.14.
Definition 8.23 Let X = x1, . . . , xn be a set, k a field of characteristic 6= 2 and
(aij)n×n a non-zero symmetric matrix over k. Denote
D(X ∪ e | xi ` xj + xj a xi − 2aije, e ` y − y, y a e− y, xi, xj ∈ X, y ∈ X ∪ e)
by C(n, f). Then C(n, f) is called a Clifford dialgebra.
We order X ∪ e by x1 < · · · < xn < e.
Theorem 8.24 Let the notation be as the above. Then
(i) S is a Grobner-Shirshov basis of Clifford dialgebra C(n, f), where S consists of the
following relations:
1. fxixj= xi ` xj + xj a xi − 2aije,
2. ge`y = e ` y − y,
3. gyae = y a e− y,
4. fyaxixj= y a xi a xj + y a xj a xi − 2aijy, (i > j),
5. fyaxixi= y a xi a xi − aiiy,
6. fxixj`y = xi ` xj ` y + xj ` xi ` y − 2aijy, (i > j),
7. fxixi`y = xi ` xi ` y − aiiy,
8. hxie = xi ` e− e a xi,
where xi, xj ∈ X, y ∈ X ∪ e.
(ii) A k-linear basis of C(n, f) is a set of all elements of the form yxi1 · · ·xik, where
y ∈ X ∪ e, xij ∈ X and i1 < i2 < · · · < ik (k ≥ 0).
Proof. Let S1 = fxixj, ge`y, gyae | xi, xj ∈ X, y ∈ X ∪ e.
Firstly, we will show that fyaxixj, fyaxixi
, fxixj`y, fxixi`y, hxie ∈ Id(S1).
In fact, fyaxixj= y a fxixj
+ 2aijgyae implies fyaxixj, fyaxixi
∈ Id(S1). By symmetry,
we have fxixj`y, fxixi`y ∈ Id(S1).
If there exists t such that ait 6= 0, then
2aithxie = fxixi`xt − xi ` fxi`xt + fxi`xt a xi − fxtaxixi∈ Id(S1).
138
Otherwise, ait = 0 for any t. Since (aij) 6= 0, there exists j 6= i such that ajt 6= 0 for some
t. Then
2ajthxie
= fxixj`xt − xi ` fxj`xt − xj ` fxi`xt + fxi`xt a xj + fxj`xt a xi − fxtaxixj∈ Id(S1).
This shows that hxie ∈ Id(S1).
Secondly, we will show that all compositions in S is trivial.
All possible compositions of left and right multiplication are: z a fxixj, z a ge`y, z a
fxixj`y, z a fxixi`y, z a hxie, fxixj` z, gyae ` z, fyaxixj
` z, fyaxixi` z, hxie ` z, where
z ∈ X ∪ e. We just check the cases of fyaxixj` z and hxie ` z. Others can be similarly
proved.
For fyaxixj` z, we have
fyaxixj` z = y ` xi ` xj ` z + y ` xj ` xi ` z − 2aijy ` z = y ` fxixj`z ≡ 0 mod(S).
For hxie ` z,
hxie ` z = xi ` e ` z − e ` xi ` z = xi ` ge`z − ge`xi` z ≡ 0 mod(S).
Now, all possible ambiguities [w] of compositions of intersection in S are:
1 ∧ 3, [xixje]; 1 ∧ 4, [xixjxmxn] (m > n); 1 ∧ 5, [xixjxnxn].
2 ∧ 1, [exixj]; 2 ∧ 2, [eey]; 2 ∧ 3, [eye]; 2 ∧ 4, [eyxixj] (i > j);
2 ∧ 5, [eyxixi]; 2 ∧ 6, [exixj y] (i > j); 2 ∧ 7, [exixiy]; 2 ∧ 8, [exie].
3 ∧ 3, [yee]; 3 ∧ 4, [yexixj] (i > j); 3 ∧ 5, [yexixi].
4 ∧ 3, [yxixje] (i > j); 4 ∧ 4, [yxixjxmxn] (i > j,m > n), [yxixjxt] (i > j > t);
4 ∧ 5, [yxixjxtxt] (i > j), [yxixjxj] (i > j).
5 ∧ 3, [yxixie]; 5 ∧ 4, [yxixixmxn] (m > n), [yxixixj] (i > j);
5 ∧ 5, [yxixixmxm], [yxixixi].
6 ∧ 1, [xixjxmxn] (i > j); 6 ∧ 2, [xixjey] (i > j); 6 ∧ 3, [xixj ye] (i > j);
6 ∧ 4, [xixj yxmxn] (i > j,m > n); 6 ∧ 5, [xixj yxmxm] (i > j);
6 ∧ 6, [xixjxmxny] (i > j,m > n), [xixjxty] (i > j > t);
6 ∧ 7, [xixjxmxmy] (i > j), [xixjxj y] (i > j); 6 ∧ 8, [xixjxte] (i > j).
7 ∧ 1, [xixixmxn]; 7 ∧ 2, [xixiey]; 7 ∧ 3, [xixiye]; 7 ∧ 4, [xixiyxmxn] (m > n);
7 ∧ 5, [xixiyxmxm]; 7 ∧ 6, [xixixmxny] (m > n), [xixixty] (i > t);
7 ∧ 7, [xixixmxmy], [xixixiy]; 7 ∧ 8, [xixixj e].
8 ∧ 3, [xiee]; 8 ∧ 4, [xiexmxn] (m > n); 8 ∧ 5, [xiexmxm].
139
All possible ambiguities [w] of compositions of inclusion in S are:
6 ∧ 1, [xixjxt] (i > j); 6 ∧ 8, [xixj e] (i > j).
7 ∧ 1, [xixixj]; 7 ∧ 8, [xixie].
We just check the cases of intersection 1 ∧ 4, 4 ∧ 4, 6 ∧ 4, 6 ∧ 8, 8 ∧ 4 and of inclusion
6 ∧ 1, 6 ∧ 8. Others can be similarly proved.
For 1 ∧ 4, [w] = [xixjxmxn] (m > n), we have
(1 ∧ 4)[w]
= xj a xi a xm a xn − 2aije a xm a xn − xi ` xj a xn a xm + 2amnxi ` xj
= xj a fxiaxmxn − 2aijfeaxmxn − fxixja xn a xm + 2amnfxixj
≡ 0 mod(S, [w]).
For 4 ∧ 4, there are two cases to consider: [w1] = [yxixjxmxn] (i > j,m > n) and
[w2] = [yxixjxt] (i > j > t). We have
(4 ∧ 4)[w1]
= y a xj a xi a xm a xn − 2aijy a xm a xn − y a xi a xj a xn a xm + 2amny a xi a xj
= y a xj a fxiaxmxn − 2aijfyaxmxn − fyaxixja xn a xm + 2amnfyaxixj
≡ 0 mod(S, [w1]) and
(4 ∧ 4)[w2]
= y a xj a xi a xt − 2aijy a xt − y a xi a xt a xj + 2ajty a xi
= y a fxjaxixt − fyaxjxt a xi − fyaxixt a xj + y a fxtaxixj
≡ 0 mod(S, [w2]).
For 6 ∧ 4, [w] = [xixj yxmxn] (i > j,m > n), we have
(6 ∧ 4)[w]
= xj ` xi ` y a xm a xn − 2aijy a xm a xn − xi ` xj ` y a xn a xm + 2amnxi ` xj ` y
= xj ` xi ` fyaxmxn − 2aijfyaxmxn − fxixj`y a xn a xm + 2amnfxixj`y
≡ 0 mod(S, [w]).
For 6 ∧ 8, [w] = [xixjxte] (i > j), we have
(6 ∧ 8)[w] = xj ` xi ` xt ` e− 2aijxt ` e+ xi ` xj ` e a xt
= xj ` xi ` hxte − 2aijhxte + fxixj`e a xt
≡ 0 mod(S, [w]).
140
For 8 ∧ 4, [w] = [xiexmxn] (m > n), we have
(8 ∧ 4)[w] = −e a xi a xm a xn − xi ` e a xn a xm + 2amnxi ` e
= −e a fxiaxmxn − hxie a xn a xm + 2amnhxie
≡ 0 mod(S, [w]).
Now, we check the compositions of inclusion 6 ∧ 1 and 6 ∧ 8.
For 6 ∧ 1, [w] = [xixjxt] (i > j), we have
(6 ∧ 1)[w] = xj ` xi ` xt − 2aijxt − xi ` xt a xj + 2ajtxi ` e
= xj ` fxixt − fxixt a xj + 2ajthxie − fxjxt a xi + fxtaxixj+ 2aithxje
≡ 0 mod(S, [w]).
For 6 ∧ 8, [w] = [xixj e] (i > j), we have
(6 ∧ 8)[w] = xj ` xi ` e− 2aije+ xi ` e a xj
= xj ` hxie + hxie a xj + hxje a xi + feaxixj
≡ 0 mod(S, [w]).
Then all the compositions in S are trivial. We have proved (i).
For (ii), since the mentioned set is just the set Irr(S), by Theorem 8.14 the result
holds.
The proof is complete.
Remark In the Theorem 8.24, if the matrix (aij)n×n = 0, then Clifford dialgebra C(n, f)
has a Grobner-Shirshov basis S ′ which consists of the relations 1–7.
141
Chapter 9 A History and a Survey of
Grobner-Shirshov Bases Theory
A.I. Shirshov (also A.I. Sirsov) in his pioneering work ([202], 1962) posted the fol-
lowing fundamental question:
How to find a linear basis of any Lie algebra presented by generators and defining
relations?
He gave an infinite algorithm for a solution of this problem using a new notion
of composition (“s-polynomial” in a later terminology of Buchberger [61, 62]) of two
Lie polynomials and a new notion of completion of any set of Lie polynomials (to add
consequently “non-trivial” compositions, critical pair/completion (cpc-) algorithm in a
later Knuth-Bendix [135] and Buchberger [64, 65] terminology).
Shirshov’s algorithm is as follows. Let S be a subset of Lie polynomials in Lie(X) of
the free algebra k〈X〉 on X over a field k (the algebra of non-commutative polynomials
on X over k) . Let S ′ be a set by adding all non-trivial Lie compositions (“Lie s-
polynomials”) of elements of S to S (S ⊆ S ′). A problem of triviality of a Lie polynomial
modulo a finite (or recursive) set S is algorithmically solvable using Shirshov Lie reduction
algorithm from his previous paper [198], 1958. Then one gets in general an infinite
sequence of Lie multi-compositions S ⊆ S ′ ⊆ S ′′ ⊆ · · · ⊆ S(n) ⊆ . . . . The union Sc of the
sequence has the property that any Lie composition of elements of Sc is trivial relative
to Sc. It is what is now called a Lie GS basis. Then from a new “Composition-Diamond
lemma∗ for Lie algebras” (Lemma 3 in [202]) it follows that the set Irr(Sc) of all Sc-
irreducible (or Sc-reduced) “basic” Lie monomials [u] in X is a linear basis of the Lie
algebra Lie(X|S) generated by X with defining relations S. Here “basic” Lie monomial
means a Lie monomial from a special linear basis of the free Lie algebra Lie(X) ⊂ k〈X〉,so called Lyndon-Shirshov (LS for short) basis (Shirshov [202] and Chen-Fox-Lyndon [69],
see below). A LS monomial [u] is called Sc-irreducible (or Sc-reduced) if u, the associative
support of [u], does not contain word s for any s ∈ S, where s is the maximum (in deg-
lex ordering) word of s as associative polynomial. To be more precise, Shirshov used
his reduction algorithm at each step S, S ′, S ′′, . . . . Then one has a direct system of sets
S → S ′ → S ′′ → . . . and Sc = lim−→S(n) is what is now called a “minimal GS basis”. As
the result, Shirshov algorithm gives a solution of the above problem for Lie algebras.
Shirshov algorithm dealing with the word problem is an infinite algorithm as Knuth-
∗The name “Composition-Diamond lemma” is a combination name of “Neuman Diamond Lemma”
[167], “Shirshov Composition Lemma” [202] and “Bergman Diamond Lemma” [9].
142
Bendix algorithm [135], 1970 that deals with the identity problem for any variety of
universal algebras.† An initial data for Knuth-Bendix algorithm is defining identities of a
variety. The result of the algorithm, if any, is a “Knuth-Bendix basis” of identities of the
variety (not a GS basis of defining relations, say, a Lie algebra).
Shirshov algorithm gives linear basis and algorithmic decidability of the word problem
for one relator Lie algebras [202], (recursive) linear basis for Lie algebras with (finite)
homogeneous defining relations [202], and linear basis for free product of Lie algebras
with known linear basis [201]. Also he proved the Freiheitssatz (the freeness theorem)
for Lie algebras [202] (for any one-relator Lie algebra Lie(X|f), a subalgebra 〈X\xi〉,where xi is involved in f , is a free Lie algebra). Shirshov problem [202] for decidability
of the word problem for Lie algebras was solved negatively in [19]. More generally, it was
proved [19] that some recursively presented Lie algebras with an unsolvable word problem
can be embedded into finitely presented Lie algebras (with the unsolvable word problem).
It is a weak analogy of Higman embedding theorem for groups [113]. A problem [19]
whether an analogy of Higman embedding theorem is valid for Lie algebras is still open.
For associative algebras the similar problem [19] was solved positively by V.Y. Belyaev
[8]. A simple example of a Lie algebra with the unsolvable word problem was given by
G.P. Kukin [138].
Following Shirshov’s paper [202] a similar algorithm is valid also for associative al-
gebras. It is since he fixed that Lie(X) is the subspace of the Lie polynomials in the
free associative algebra k〈X〉. Then to define a Lie composition 〈f, g〉w of two Lie poly-
nomials relative to an associative word w, he defines first the associative composition
(non-commutative “s-polynomial”) (f, g)w = fb − ag, a, b ∈ X∗. Then he puts some
brackets 〈f, g〉w = [fb]f − [ag]g following his special bracketing lemma in [198]. One may
get Sc for any S ⊂ k〈X〉 in the same way as for Lie polynomials and in the same way
as for Lie algebras (“CD-lemma for associative algebras”) to get that Irr(Sc) is a linear
basis of associative algebra k〈X|S〉 generated by X with defining relations S. All proofs
are in the same manner and much easier than in [202].
Even more, the cases of semigroups and groups presented by generators and defining
relations are just special cases of associative algebras via semigroup and group algebras.
All in all, Shirshov algorithm gives linear bases and normal forms of elements of any Lie
algebra, associative algebra, semigroup or group presented by generators and defining
relations! The algorithm does work in many cases (see below), as opposed to for example
an analogous algorithm for non-associative algebras.
†We use the usual algebraic terminology “the word problem”, “ the identity problem”, see, for example,
O.G. Kharlampovich, M.V. Sapir [134].
143
The Grobner bases theory and Buchberger algorithm were initiated by B. Buchberger
(Thesis [61] 1965, paper [62] 1970) for commutative-associative algebras. Buchberger
algorithm is a finite algorithm for finitely generated commutative algebras. It is one of
the most useful and famous algorithm in modern computer science.
The Shirshov paper [202] was in a spirit of A.G. Kurosh (1908-1972) program to
study non-associative (relatively) free algebras and free products of algebras initiated in
Kurosh paper [139], 1947. In this paper he proved non-associative analogies of Nielsen-
Shreier and Kurosh theorems for groups. It took quite a few years to clarify the situation
for Lie algebras in Shirshov’s papers [195], 1953 and [201], 1962 closely connected with
his GS bases theory. It is important to note that quite unexpectedly the Kurosh program
led to the Shirshov’s theory of GS bases for Lie and associative algebras [202].
A step in the Kurosh program was done by his student A.I. Zhukov in his PhD Thesis
[221], 1950 who algorithmically solved the word problem for non-associative algebras. In a
sense it was a beginning GS bases theory for non-associative algebras. The main difference
with the future Shirshov’s approach is that Zhukov did not use any linear ordering of non-
associative monomials. Instead he chooses any monomial of maximal degree as a “leading”
monomial of a polynomial. Also for non-associative algebras there is no “composition of
intersection” (“s-polynomial”). In this sense it cannot be a model for Lie and associative
algebras.‡
A.I. Shirshov, also a student of Kurosh, defended his Candidate of Sciences Thesis
[194] at Moscow University in 1953. His Thesis together with next followed paper [198],
1958 may be viewed as a background of his later method of GS bases. He proved the
free subalgebra theorem for free Lie algebras (now known as Shirshov-Witt theorem,
see also Witt [214], 1956) using the elimination process rediscovered by Lazard [145],
1960. He used the elimination process later [198], 1958 as a general method to prove
properties of regular (LS) words, including an algorithm of (special) bracketing of a LS
word (with a fix LS subword). The last algorithm is of some importance in his GS
bases theory for Lie algebras (particular in a definition of the compositions of two Lie
polynomials). Also Shirshov proved the free subalgebra theorem for (anti-) commutative
non-associative algebras [197], 1954. He used the paper lately in [203], 1962 for GS bases
theory of (commutative, anti-commutative) non-associative algebras. Last but not the
‡After PhD Thesis, 1950, A.I. Zhukov moved to now Keldysh Institute of Applied Mathemat-
ics (Moscow) and then to an atomic organization that is now known as Chelyabinsk-90 (Ural Re-
gion) doing computational mathematics. S.K. Godunov in “Reminiscence about numerical schemes”,
arxiv.org/pdf/0810.0649, 2008, mentioned his name in according with a creation of the famous Godunov
numerical method. So, A.I. Zhukov was a forerunner of two important computational methods!
144
least Shirshov found a series of bases of a free Lie algebra published 10 years later in the
first issue of the new A.I. Malcev’s Journal “Algebra and Logic” [200], 1962.§
LS basis is a particular case of the Shirshov or Hall-Shirshov series of bases (cf.
Reutenauer [185], where this series was called “Hall series”). In the definition of his series,
Shirshov used the Hall inductive procedure (see Ph. Hall [111], 1933, M. Hall [110], 1950):
a non-associative monomial w = ((u)(v)) is a basic monomial if 1) (u), (v) are basic, 2)
(u) > (v), 3) if (u) = ((u1)(u2)), then (u2) ≤ (v). But instead of an ordering by the
degree function (Hall words), he used any linear ordering of non-associative monomials
such that ((u)(v)) > (v). For example, in his Thesis [194], 1953 he used an ordering by the
“content” of monomials (content of, say, a monomial (u) = ((x2x1)((x2x1)x1)) is a vector
(x2, x2, x1, x1, x1)). Actually content u of (u) may be viewed as a commutative-associative
word that equals to u in the free commutative semigroup. Two contents are compared
lexicographically (a proper prefix of a content is greater than the content).
If one uses lexicographical ordering, (u) (v) if u v lexicographically, then one will
get the LS basis.¶ For example, for alphabets x1, x2, x2 x1 one has Lyndon-Shirshov
basic monomials by induction:
x2, x1, [x2x1], [x2[x2x1]] = [x2x2x1], [[x2x1]x1] = [x2x1x1], [x2[x2x2x1]] = [x2x2x2x1],
[x2[x2x1x1]] = [x2x2x1x1], [[x2x1x1]x1] = [x2x1x1x1], [[x2x1][x2x1x1]] = [x2x1x2x1x1] and
so on.
They are exactly all Shirshov regular (LS) Lie monomials and their associative sup-
ports are exactly all Shirshov regular words with 1-1 correspondence between two sets
given by the Shirshov elimination (bracketing) algorithm of (associative) words.
§It must be pointed out that A.I. Malcev (1909-1967) inspired Shirshov’s works very much. Malcev
was an official opponent (referee) of his (second) Doctor of Sciences Dissertation, MSU, 1958. L.A. Bokut
remembers this event at the Science Council Meeting, Chairman A.N. Kolmogorov, and Malcev’s words
“Shirshov’s Dissertation is a brilliant one!”. Malcev and Shirshov worked together at the now Sobolev
Institute of Mathematics (Novosibirsk) since 1959 until Malcev’ unexpected death at 1967, and have been
friends despite the age difference. Malcev was head of the Algebra and Logic department (by the way,
L.A. Bokut is a member of the departement since 1960) and Shirshov was the first deputy-director of the
Institute (director was S.L. Sobolev). In those years, Malcev was interested in the theory of algorithms
of mathematical logic and algorithmic problems of model theory. Because of that, Shirshov had an
additional motivation to work on algorithmic problems for Lie algebras. Both Maltsev and Kurosh were
delighted with the Shirshov’s results of [202]. Malcev successfully presented the paper for an award of
the Presidium of the SB of the RAS (Sobolev and Malcev were only members of the Presidium from
Institute of mathematics at that time).¶Lyndon-Shirshov basis for the alphabet x1, x2 is different from above Shirshov “content” basis starting
with monomials of degree 7.
145
Let us remind that an elementary step of Shirshov elimination algorithm is to joint
the minimal letter of a word to previous ones by the bracketing and to continue this
process with lexicographical ordering of the new alphabet. For example, let x2 x1.
Then one has induction bracketing:
x2x1x2x1x1x2x1x1x1x1x2x1x1, [x2x1][x2x1x1][x2x1x1x1][x2x1x1],
[x2x1][[x2x1x1][x2x1x1x1]][x2x1x1], [[x2x1][[x2x1x1][x2x1x1x1]]][x2x1x1],
[[[x2x1][[x2x1x1][x2x1x1x1]]][x2x1x1]],
x2x1x1x1x2x1x1x2x1x2x2x1, [x2x1x1x1][x2x1x1][x2x1]x2[x2x1],
[x2x1x1x1][x2x1x1][x2x1][x2[x2x1]], x2x1x1x1 ≺ x2x1x1 ≺ x2x1 ≺ x2x2x1.
By the way, the second series of partial bracketing is an illustration of Shirshov
factorization theorem [198], 1958 that any word is a non-decreasing product of LS words
(it is often called by a mistake, see [10], as “Lyndon theorem”.)
Shirshov’s [198] the special bracketing of any LS word with a fix LS subword is as fol-
lows. Let us give an example of special bracketing of LS word w = x2x2x1x1x2x1x1x1 with
LS-subword u = x2x2x1. The Shirshov standard bracketing is [w] = [x2[[[x2x1]x1][x2x1x1x1]]].
The Shirshov special bracketing is [w]u = [[[u]x1][x2x1x1x1]]. In general, if w = aub then
the Shirshov standard bracketing gives [w] = [a[uc]d], where b = cd. Now, c = c1 · · · ct,each ci is LS-word, c1 · · · ct in lex-ordering (Shirshov factorization theorem). Then
one must change bracketing of [uc]:
[w]u = [a[. . . [[u][c1]] . . . [ct]]d]
The main property of [w]u is that [w]u is a monic associative polynomial with the maximal
monomial w, i.e., [w]u = w.
Actually, Shirshov [202], 1962 needs to use “double” relative bracketing of a regular
word with two separated LS subwords. Then he used implicitly the following property:
any LS subword of c = c1 · · · ct as above is a subword of some ci, 1 ≤ i ≤ t.
The Shirshov [198], 1958 definition of regular monomials ia as follows: (w) = ((u)(v))
is a regular monomial iff 1) w is a regular word, 2) (u), (v) are regular monomials (then
automatically u v in lex-ordering), 3) if (u) = ((u1)(u2)), then u2 v.
Once again, if one formally omits all Lie brackets in Shirshov’s paper [202], essentially
the same algorithm and essentially the same CD-lemma (with the same but much simpler
proof) give a linear basis of an associative algebra presented by generators and defining
relations. The differences are the following:
—No need to use LS monomials and LS words (since the set X∗ is a linear basis of
the free associative algebra k〈X〉);—The definition of associative compositions for monic polynomials f, g,
146
(f, g)w = fb− ag, w = f b = ag, deg(w) < deg(f) + deg(g), or (f, g)w = f − agb, w =
agb, w, a, b ∈ X∗, are much simpler than the definition of Lie compositions for monic Lie
polynomials f, g,
〈f, g〉w = [fb]f − [ag]g, w = f b = ag, deg(w) < deg(f) + deg(g), or
〈f, g〉w = f − [agb]g, w = agb, w, a, b, f , g ∈ X∗,
where [fb]f , [ag]g, [agb]g are special Shirshov’s bracketing of LS words w with fixed LS
subwords f , g correspondingly.
—A definition of the elimination of a leading word s of an associative monic polyno-
mial s is straightforward (asb→ a(rs)b, if s = s−rs, a, b ∈ X∗). But for Lie polynomials,
it is much more involved and uses the special Shirshov’s bracketing.
—The main idea of the Shirshov’s proof may be formulated as follows. Let S be a
complete set of monic Lie polynomials. If w = a1s1b1 = a2s2b2, where w, , ai, bi ∈ X∗, w
is a LS word, s1, s2 ∈ S, then Lie monomials [a1s1b1]s1 and [a2s2b2]s2 are equal modulo
the less Lie monomials from the Id(S):
[a1s1b1]s1 = [a2s2b2]s2 +∑i>2
αi[aisibi]si, where each si ∈ S, [aisibi]si
= aisibi < w.
Actually Shirshov proved more general result: if (a1s1b1) = a1s1b1, (a2s2b2) = a2s2b2 and
w = a1s1b1 = a2s2b2 then
(a1s1b1) = (a2s2b2) +∑i>2
αi(aisibi), where each si ∈ S, (aisibi) = aisibi < w.
Later we call a Lie polynomial (asb) as Lie normal S-word if (asb) = asb.
Exactly in this place he used the notion of composition and other notions and prop-
erties mentioned above.
For associative algebras (and for commutative-associative algebras) it is much easy
to prove the analogy of this property: If S is a complete monic set in k〈X〉 (k[X]) and
w = a1s1b1 = a2s2b2, ai, bi ∈ X∗, s1, s2 ∈ S, then
a1s1b1 = a2s2b2 +∑i>2
αiaisibi, where each si ∈ S, aisibi < w.
Summarizing we can say with confidence that the work (Shirshov [202]) implicitly
contains the CD-lemma for associative algebras as a simple exercise that does not require
new ideas. L.A. Bokut can confirm that Shirshov clearly understood this and told to him,
that “the case of associative algebras is the same”. Explicitly the lemma was formulated
147
in Bokut [20], 1976 (with reference to the Shirshov’s paper [202]), Bergman [9], 1978 and
Mora [166], 1986.
Let us stress once again that CD-Lemma for associative algebras can be applied to
any semigroups P = sgp〈X|S〉, in particular to any group, through the semigroup algebra
kP over a field k. The last algebra has the same generators and defining relations as P ,
kP = k〈X|S〉. Any composition of “twonomials” u1 − v1, u2 − v2 is a twonomial u − v.As the result, Shirshov’s algorithm applying to a set of semigroups relations S gives rise a
complete set of semigroup relations Sc. The Sc-irreducible words in X is a set of normal
forms of elements of P .
Before we go any farther, let us give some well known examples of presentations of
algebras, groups and semigroup.
—Grassman algebra k〈X|x2i = 0, xixj + xjxi = 0, i > j〉.
—Clifford algebra k〈X|xixj + xjxi = aij, 1 ≤ i, j ≤ n〉, where (aij) is an n × n
symmetric matrix over k.
—Universal enveloping algebra U(L) = k〈X|xixj − xjxi =∑αk
ijxk, i > j〉, where L
is a Lie algebra with a well ordered linear basis X = xi|i ∈ I, [xixj] =∑αk
ijxk, i > j.
The PBW theorem provides a linear basis of U(L): xi1xi2 · · ·xin | i1 ≤ i2 ≤ · · · ≤in, it ∈ I, t = 1, . . . , n, n ≥ 0. A. Kandri-Rody and V. Weispfenning [123] invented
an important class of “algebras of polynomial type” that included Clifford and universal
enveloping algebras. They created a GS bases theory for this class.
—More generally, let L = Lie(X|S) be a Lie algebra presented by generators and
a Lie GS basis S (in deg-lex ordering). Then the universal enveloping algebra U(L) has
presentation U(L) = k〈X|S(−)〉 , where S(−) is S as the set of associative polynomials.
PBW Theorem in a Shirshov form (see Theorem 4.8) says that S(−) is an associative GS
basis and a linear basis of U(L) consists of words u1u2 · · ·un, where ui are S-reduced LS
words, u1 u2 · · · un (in lex-ordering),‖ see [52, 53].
—Free Lie algebras over a commutative algebra. Ph. Hall, M. Hall, A.I. Shirshov,
R. Lyndon provide different linear bases of the algebras (Hall basis, Hall-Shirshov series
of bases, Lyndon-Shirshov basis). See also R-basis [14] and recent papers [31, 34].
—Lie algebras presented by Chevalley generators and defining relations of types
An, Bn,
Cn, Dn, G2, F3, E6, E7, E8. The Serre theorem provides linear bases and multiplication
tables of the algebras (they are finite dimensional simple Lie algebras over a field k). See
also [45, 46, 47].
‖For any Lie polynomial s ∈ Lie(X) ⊂ k〈X〉, the maximal associative word s of s is a LS word and
the regular LS monomial [s] is the maximal LS monomial of s (both in deg-lex ordering).
148
—Finite Coxeter groups. Groups presented by Coxeter generators and defining rela-
tions of types An, Bn, Dn, G2, F3, E6, E7, E8. Coxeter theorem provides normal forms and
Cayley tables (they are finite groups generated by reflections). See also [54].
—Coxeter groups (for any Coxeter matrix). J. Tits [205] algorithmically solved the
word problem for any Coxeter group. See also [13].
—Affine Kac-Moody algebras [118]. Kac-Gabber theorem provides linear bases of
the algebras under “symmetrizability” condition on the Cartan matrix. See also [174,
175, 176].
—Borcherds-Kac-Moody algebras [57, 58, 59, 118].
—Quantum enveloping algebras (V.G. Drinfeld, M. Jimbo). Lusztig theorem [150]
provides linear “canonical bases” of the algebras. Different approaches were given by
C.M. Ringel [186, 187] and V.K. Kharchenko [129, 130, 131, 132, 133]. See also [51, 84].
—Koszul algebras. Quadratic algebras having a basis of “standard” monomials,
called PBW-algebras, are always Koszul (S. Priddy [179]), but not conversely. In another
terminology PBW-algebras are algebras with quadratic GS bases. See [173].
—Elliptic algebras. Elliptic algebras (B. Feigin, A. Odesskii) are associative algebras
presented by n generators and n(n− 1)/2 homogeneous quadratic relations for which the
dimensions of graded components are the same as for polynomial algebra in n variables.
The first example of such algebras was Sklyanin algebra (1982) generated by x1, x2, x3
with 3 defining relations [x3, x2] = x21, [x2, x1] = x2
3, [x1, x3] = x22. See [171].
—Cluster algebras. A cluster algebra (S. Fomin, A. Zelevinski) is a subalgebra of
a field of rational functions Q(x1, . . . , xn) generated by “cluster” elements. Cluster alge-
bras were invented in order to better understand the Lustig canonical bases of quantum
enveloping algebras.
—Leavitt path algebras. GS bases of these algebras are found in A. Alahmedi et al
[2] and applied by the same authors to determination of the structure of Leavitt path
algebras of polynomial growth in [3].
—Hecke (Iwahory-Hecke) algebras. A deep connection of Hecke algebras (of type
An) and braid groups (and invariants of links) were found in V.F.R. Jones [117].
—Artin braid group Bn. Markov-Artin theorem provides normal form and a semi-
direct structure of the group in Burau generators. Other normal forms of Bn were obtained
by Garside, Birman-Ko-Lee and Adjan-Thurston. See also [21, 22].
—Artin-Tits groups. Normal forms are known in the spherical case, see E. Brieskorn,
K. Saito [60].
—Novikov-Boon type of groups [56, 95, 119, 120, 121, 122, 169] with the unsolvable
word or conjugacy problem. They are groups with “standard bases” (“standard normal
149
forms” or “standard GS bases”), see [15, 16, 75].
—Adjan-Rabin constructions of groups with unsolvable isomorphism and Markov
properties [1, 182]. See also [24].
—Markov-Post semigroups with unsolvable word problem. See also [157].
—Markov construction of semigroups with unsolvable isomorphism and Markov prop-
erties.
—Plactic monoids. A theorem due to Richardson, Schensted, Knuth provides normal
form of elements of the monoids. See also [73].
—Groups of quotients of multiplicative semigroups of some power series rings with
“topological quadratic defining relations” (of the type k〈〈x, y, z, t|xy = zt〉〉) that are
embeddable into groups but not embeddable, in general, into division algebras (an answer
to Malcev problem). “Relative standard normal forms” of the groups were found in [17, 18]
between line of what was later called “relative GS basis” [55].
Up to date, GS bases method invented in particular for the following classes of linear
universal algebras as well as for operads, categories and semirings. Unless otherwise
stated, all linear algebras are considered over a field k. Following the terminology of
Higgins-Kurosh, by an ((differential) associative) Ω-algebra we mean a linear space (an
(differential) associative algebra) with a set of multi-linear operations Ω:
—Associative algebras, Shirshov [202], Bokut [20], Bergman [9],
—Associative algebras over a commutative algebra, A.A. Mikhalev, Zolotykh [165],
—Associative Γ-algebras, where Γ is a group, Bokut, Shum [55],
—Lie algebras, Shirshov [202],
—Lie algebras over a commutative algebra, Bokut, Chen, Chen [28],
—Lie p-algebras over k, chk = p, A.A. Mikhalev [161],
—Lie superalgebras, A.A. Mikhalev [160, 162],
—Metabelian Lie algebras, Chen, Chen [72],
—Quiver (path) algebras, Farkas, Feustel, Green [98],
—Tensor products of associative algebras, Bokut, Chen, Chen [27],
—Associative differential algebras, Chen, Chen, Li [74],
—Associative (n−)conformal algebras over k, chk = 0, Bokut, Fong, Ke [42], Bokut,
Chen, Zhang [40],
—Dialgebras, Bokut, Chen, Liu [35],
—Pre-Lie (Vinberg-Koszul-Gerstenhaber, right symmetric) algebras, Bokut, Chen,
Li [32],
—Associative Rota-Baxter algebras over k, chk = 0, Bokut, Chen, Deng [29],
—L-algebras, Bokut, Chen, Huang [30],
150
—Associative Ω-algebras, Bokut, Chen, Qiu [38],
—Associative differential Ω-algebras, Qiu, Chen [180]
—Ω-algebras, Bokut, Chen, Huang [30],
—Differential Rota-Baxter commutative associative algebras,
—Semirings, Bokut, Chen, Mo [37],
—Modules over an associative algebra, Kang, Lee [124, 125], Chibrikov [88],
—Small categories, Bokut, Chen, Li [33],
—Non-associative algebras, Shirshov [203],
—Non-associative algebras over a commutative algebra, Chen, Jing Li, Zeng [79],
—Commutative non-associative algebras, Shirshov [203]
—Anti-commutative non-associative algebras, Shirshov [203],
—Symmetric operads, Dotsenko, Khoroshkin [96].
At the heart of the GS method for a class of linear algebras, there is a CD-lemma
for a free object of the the class. For above cases, free objects are correspondingly a
free associative algebra k〈X〉, a “double free” associative k[Y ]-algebra k[Y ]〈X〉, a free Lie
algebra Lie(X), a “double free” Lie k[Y ]-algebra Liek[Y ](X). For tensor product of two
associative algebras, one needs to use tensor product of two free algebras, k〈X〉 ⊗ k〈Y 〉.Any semiring can be viewed as a “double semigroup” with two associative “products”
· and . So, CD-lemma for semirings is CD-lemma for a “semiring algebra” kRig(X)
of the free semiring Rig(X). CD-lemma for modules is CD-lemma for a “double free”
module Modk〈Y 〉(X), a free module over a free associative algebra. CD-lemma for small
categories is CD-lemma for a “free partial k-algebra” kC〈X〉 generated by an oriented
graph X (a sequence z1z2 · · · zn, where zi ∈ X, is a partial word in X, iff it is a path;
a partial polynomial is a linear combination of partial words with the same sources and
targets).
All CD-lemmas are essentially the same wording. Let V be a class of linear universal
algebras, V(X) a free algebra from V, N(X) a well-ordered term basis of V(X). A subset
S ∈ V(X) is called a GS basis, if any “composition” of elements of S is “trivial” (goes
to zero by “elimination” of leading terms s, s ∈ S). Then the following conditions are
equivalent:
(i) S is a GS basis.
(ii) If f ∈ Id(S), then the leading term f contains a sub-term s for some s ∈ S.
(iii) A set of S-irreducible terms is a linear basis of the V-algebra V〈X|S〉, generated
by X with defining relations S.
151
In some cases ((n−) conformal algebras, dialgebras), conditions (i) and (ii) are not
equivalent. To be more precise in that cases (i)⇒ (ii)⇔ (iii).
Typical compositions are compositions of intersection and inclusion. Shirshov [202,
203] avoided the inclusion composition. Instead, he suggested that a GS basis must be
reduced (leading words do not contain each other as subwords). In some cases, one needs
to define new compositions, for example, the composition of left (right) multiplication.
Also, sometimes one needs to combine all these compositions. We present here a new
approach to the definition of a composition based on the concept of a least common
multiple lcm(u, v) of two terms u, v.
In some cases (Lie algebras, (n−) conformal algebras) the “leading” term f of a
polynomial f ∈ V(X) does not belong to V(X). For Lie algebras, one has f ∈ k〈X〉, for
((n)-) conformal algebras f belongs to an associative Ω-semigroup.
For almost all CD-lemmas one needs to use a new notion of “normal S-term”. A
term (asb) in X, Ω, s ∈ S, with only one occurrence of s is called a normal S-term if
(asb) = (a(s)b). For any S ⊂ k〈X〉, each S-word (i.e., S-term) is a normal S-word. For
any S ⊂ Lie(X), each Lie S-monomial (Lie S-term) is a linear combination of normal Lie
S-terms (Shirshov [202]).
One of the two key lemmas asserts that if S is complete under compositions of
multiplications then any elements of the ideal generated by S is a linear combination of
normal S-terms. Another key lemma says that if S is a GS basis and the leading words
of two normal S-terms are the same, then these terms are the same modulo lower normal
S-terms. As we mentioned above these results were proved in Shirshov [202] for Lie(X)
(there are no compositions of multiplications for Lie and associative algebras).
In conclusion, we give some information about the works of Shirshov, see more on
this in the book [204]. A.I. Shirshov (1921-1981) was a famous Russian mathematician.
His name associated with notions and results on GS bases, the CD-lemma, the Shirshov-
Witt theorem, the Lazard-Shirshov elimination, the Shirshov height theorem, Lyndon-
Shirshov words, Lyndon-Shirshov basis (in free Lie algebras), Hall-Shirshov series of bases,
the Cohn-Shirshov theorem for Jordan algebras , the Shirshov theorem on the Kurosh
problem, the Shirshov factorization theorem. Shirshov’s ideas were used by his students
Efim Zelmanov for a solution of the restricted Burnside problem and Aleksander Kemer
for a solution of the Specht problem.
152
Reference
[1] Adjan, S.I.: Algorithmic undecibility of certain decision problems of the group theory.
Trudy Moscow Mat. Ob. 6, 231 - 298 (1957).
[2] Alahmadi, A., Alsulami, H., Jain, S.K., Zelmanov, E.: Leavitt path algebras of finite
Gelfand-Kirillov dimension. J. Algebra Appl. 11, No. 6, 1250225-1–1250225-6 (2012)
[3] Alahmadi, A., Alsulami, H., Jain, S.K., Zelmanov, E.: Structure of Leavitt path algebras
of polynomial growth. www.pnas.org/cgi/doi/10.1073/pnas.1311216110
[4] Artin, E.: Theory der Zorf. Abh. Math. Sem. Hamburg Univ. 4, 47-72 (1926)
[5] Artin, E.: Theory of braids. Ann. Math. 48, 101-126 (1947)
[6] Bahturin, Yu.A., Olshanskii, A.: Filtrations and Distortion in Infinite-Dimensional Alge-
bras. J. Algebra 327, 251-291 (2011)
[7] Bahturin, Yu.A., Mikhalev, A.A., Petrogradskij, V.M., Zajtsev, M.V.: Infinite dimen-
sional Lie superalgebras. [B] De Gruyter Expositions in Mathematics. 7. Berlin etc.: W. de
Gruyter. x, 250 p. (1992)
[8] Belyaev, V.Ya.: Subrings of finitely presented associative rings. Algebra Logika 17, 627-638
(1978)
[9] Bergman, G.M.: The diamond lemma for ring theory. Adv. Math. 29, 178-218 (1978)
[10] Berstel, J.D., Perrin, D.: The origins of combinatorics on words. Eur. J. Comb. 28, 996-1022
(2007)
[11] Birkhoff, G.: Representability of Lie algebras and Lie groups by matrices. Ann. of Math.
38(2), 526-532 (1937) (Selected Papers, Birkhauser 1987, 332-338.)
[12] Birman, J., Ko, K.H., Lee, S.J.: A new approach to the word and conjugacy problems for
the braid groups. Adv. Math. 139, 322-353 (1998)
[13] Bjoner, A., Brenti, F.: Combinatorics of Coxter Groups. Graduate Texts in Math. 231,
Springer 2005
[14] Bokut, L.A.: A base of free polynilpotent Lie algebras. Algebra Logika 2, 13-19 (1963)
[15] Bokut, L.A.: On one property of the Boone group. Algebra Logika 5, 5-23 (1966)
153
[16] Bokut, L.A.: On the Novikov groups. Algebra Logika 6, 25-38 (1967)
[17] Bokut, L.A.: Groups of fractions for the multiplicative semigroups of certain rings I-III.
Sibirsk. Mat. Zh. 10, 246-286, 744-799, 800-819 (1969)
[18] Bokut, L.A.: On the Malcev problem. Sibirsk. Mat. Zh. 10, 965-1005 (1969)
[19] Bokut, L.A.: Insolvability of the word problem for Lie algebras, and subalgebras of finitely
presented Lie algebras. Izvestija AN USSR (mathem.) 36, 1173-1219 (1972)
[20] Bokut, L.A.: Imbeddings into simple associative algebras. Algebra Logika 15, 117-142
(1976)
[21] Bokut, L.A.: Grobner-Shirshov bases for braid groups in Artin-Garside generators. J. Sym-
bolic Computation 43, 397-405 (2008)
[22] Bokut, L.A.: Grobner-Shirshov bases for the braid group in the Birman-Ko-Lee generators.
J. Algebra 321, 361-379 (2009)
[23] Bokut, L.A., Chainikov, V.V., Shum, K.P.: Markov and Artin normal form theorem for
braid groups. Commun. Algebra 35, 2105-2115 (2007)
[24] Bokut, L.A., Chainikov, V.V.: Grobner-Shirshov bases of Adjan extension of the Novikov
group. Discrete Mathematics 308, 4916-4930 (2008).
[25] Bokut, L.A., Chen, Y.Q.: Grobner-Shirshov bases for Lie algebras: after A.I. Shirshov.
Southeast Asian Bull. Math. 31, 1057-1076 (2007)
[26] Bokut, L.A., Chen, Y.Q.: Grobner-Shirshov bases: some new results, Advance in Alge-
bra and Combinatorics. Proceedings of the Second International Congress in Algebra and
Combinatorics, Eds. K. P. Shum, E. Zelmanov, Jiping Zhang, Li Shangzhi, World Scientific,
2008, 35-56.
[27] Bokut, L.A., Chen, Y.Q., Chen, Y.S.: Composition-Diamond lemma for tensor product of
free algebras. J. Algebra 323, 2520-2537 (2010)
[28] Bokut, L.A., Chen, Y.Q., Chen, Y.S.: Grobner-Shirshov bases for Lie algebras over a
commutative algebra. J. Algebra 337, 82-102 (2011)
[29] Bokut, L.A., Chen, Y.Q., Deng, X.M.: Grobner-Shirshov bases for Rota-Baxter algebras.
Siberian Math. J. 51, 978-988 (2010)
154
[30] Bokut, L.A., Chen, Y.Q., Huang, J.P.: Grobner-Shirshov bases for L-algebras. Internat. J.
Algebra Comput. 23, 547-571 (2013)
[31] Bokut, L.A., Chen, Y.Q., Li, Y.: Anti-commutative Grobner-Shirshov basis of a free Lie
algebra. Science in China Series A: Mathematics 52, 244-253 (2009)
[32] Bokut, L.A., Chen, Y.Q., Li, Y.: Grobner-Shirshov bases for Vinberg-Koszul-Gerstenhaber
right-symmetric algebras. J. Math. Sci. 166, 603-612 (2010)
[33] Bokut, L.A., Chen, Y.Q., Li, Y.: Grobner-Shirshov bases for categories. Nankai Series in
Pure, Applied Mathematics and Theoretical Physical, Operads and Universal Algebra 9,
1-23 (2012)
[34] Bokut, L.A., Chen, Y.Q., Li, Y.: Lyndon-Shirshov words and anti-commutative algebras.
J. Algebra 378, 173-183 (2013)
[35] Bokut, L.A., Chen, Y.Q., Liu, C.H.: Grobner-Shirshov bases for dialgebras. Internat. J.
Algebra Comput. 20, 391-415 (2010)
[36] Bokut, L.A., Chen, Y.Q., Mo, Q.H.: Grobner-Shirshov bases and embeddings of algebras.
Internat. J. Algebra Comput. 20, 875-900 (2010)
[37] Bokut, L.A., Chen, Y.Q., Mo, Q.H.: Grobner-Shirshov bases for semirings. J. Algebra 385,
47-63 (2013)
[38] Bokut, L.A., Chen, Y.Q., Qiu, J.J.: Grobner-Shirshov bases for associative algebras with
multiple operations and free Rota-Baxter algebras. J. Pure Applied Algebra 214, 89-100
(2010)
[39] Bokut, L.A., Chen, Y.Q., Shum, K.P.: Some new results on Grobner-Shirshov bases. Pro-
ceedings of International Conference on Algebra 2010, Advances in Algebraic Structures,
2012, pp.53-102.
[40] Bokut, L.A., Chen, Y.Q., Zhang, G.L.: Composition-Diamond lemma for associative n-
conformal algebras. arXiv:0903.0892
[41] Bokut, L.A., Chen, Y.Q., Zhao, X.G.: Grobner-Shirshov beses for free inverse semigroups.
Internat. J. Algebra Comput. 19, 129-143 (2009)
[42] Bokut, L.A., Fong, Y., Ke, W.-F.: Composition Diamond lemma for associative conformal
algebras. J. Algebra 272, 739-774 (2004)
155
[43] Bokut, L.A., Fong, Y., Ke, W.-F., Kolesnikov, P.S.: Grobner and Grobner-Shirshov bases in
algebra and conformal algebras. Fundamental and Applied Mathematics 6, 669-706 (2000)
[44] Bokut, L.A., Kang, S.-J., Lee, K.-H., Malcolmson, P.: Grobner-Shirshov bases for Lie
superalgebras and their universal enveloping algebras, J. Algebra 217, 461-495 (1999)
[45] Bokut, L.A., Klein, A.A.: Serre relations and Grobner-Shirshov bases for simple Lie alge-
bras I, II. Internat. J. Algebra Comput. 6, 389-400, 401-412 (1996)
[46] Bokut, L.A., Klein, A.A.: Grobner-Shirshov bases for exceptional Lie algebras. I. J. Pure
Applied Algebra 133, 51-57 (1998)
[47] Bokut, L.A., Klein, A.A.: Grobner-Shirshov bases for exceptional Lie algebras E6, E7, E8.
Algebra and Combinatorics (Hong Kong), 37-46, Springer, Singapore (1999)
[48] Bokut, L.A., Kolesnikov, P.S.: Grobner-Shirshov bases: from their incipiency to the present.
J. Math. Sci. 116, 2894-2916 (2003)
[49] Bokut, L.A., Kolesnikov, P.S.: Grobner-Shirshov bases, conformal algebras and pseudo-
algebras. J. Math. Sci. 131, 5962-6003 (2005)
[50] Bokut, L.A., Kukin, G.P.: Algorithmic and Combinatorial Algebra. Mathematics and its
Applications, Kluwer Academic Publishers Group, Dordrecht (1994)
[51] Bokut, L.A., Malcolmson, P.: Grobner-Shirshov bases for quantum enveloping algebras.
Isr. J. Math. 96, 97-113 (1996)
[52] Bokut, L.A., Malcolmson, P.: Grobner-Shirshov bases for Lie and associative algebras.
Collection of Abstracts, ICAC,97, Hong Kong, 1997, 139-142.
[53] Bokut, L.A., Malcolmson, P.: Grobner-Shirshov bases for relations of a Lie algebra and its
enveloping algebra. Shum, Kar-Ping (ed.) et al., Algebras and combinatorics. Papers from
the international congress, ICAC’97, Hong Kong, August 1997. Singapore: Springer. 47-54
(1999)
[54] Bokut, L.A., Shiao, L.-S.: Grobner-Shirshov bases for Coxeter groups. Commun. Algebra
29, 4305-4319 (2001)
[55] Bokut, L.A., Shum, K.P.: Relative Grobner-Shirshov bases for algebras and groups. St.
Petersbg. Math. J. 19, 867-881 (2008)
[56] Boone, W.W.: The word problem, Ann. Math. 70, 207-265 (1959)
156
[57] Borcherds, R.E.: Vertex algebras, Kac-Moody algebras, and the monster. Proc. Natl. Acad.
Sci. USA 84, 3068-3071 (1986)
[58] Borcherds, R.E.: Generalized Kac-Moody algebras. J. Algebra 115, No.2, 501-512 (1988)
[59] Borcherds, R.E.: The monster Lie algebra. Adv. Math. 83, No.1, 30-47 (1990)
[60] Brieskorn, E., Saito, K.: Artin-Gruppen und Coxeter-Gruppen. Invent. Math. 17, 245-271
(1972)
[61] Buchberger, B.: An algorithm for finding a basis for the residue class ring of a zero-
dimensional polynomial ideal. Ph.D. thesis, University of Innsbruck, Austria (1965)
[62] Buchberger, B.: An algorithmical criteria for the solvability of algebraic systems of equa-
tions. Aequationes Math. 4, 374-383 (1970)
[63] Buchberger, B.: Some properties of Grobner bases for polynomial ideals. ACM SIGSAM
Bulletin 10(4), 19-24 (1976)
[64] Buchberger, B.: History and basic feature of the critical-pair/completion procedure. J.
Symbolic Computation 3, 3-38 (1987)
[65] Buchberger, B., Collins, G.E., Loos, R., Albrecht, R.: Computer algebra, symbolic and al-
gebraic computation, Computing Supplementum, Vol.4, New York: Springer-Verlag (1982)
[66] Cain, A.J., Gray, R., Malheiro, A.: Finite Grobner-Shirshov bases for Plactic algebras and
biautomatic structures for Plactic monoids, arXiv:1205.4885v2
[67] Cartier, P.: Remarques sur le theoreme de Birkhoff-Witt, Annali della Scuola Norm. Sup.
di Pisa serie III vol XII(1958), 1-4.
[68] Cassaigne, J., Espie, M., Krob, D., Novelli, J.C., Hivert, F.: The Chinese Monoid. Internat.
J. Algebra Comput. 11, 301-334 (2001)
[69] Chen, K.-T., Fox, R., Lyndon, R.: Free differential calculus IV: The quotient group of the
lower central series. Ann. Math. 68, 81-95 (1958)
[70] Chen, Y.Q.: Grobner-Shirshov basis for Schreier extensions of groups. Commun. Algebra
36, 1609-1625 (2008)
[71] Chen, Y.Q.: Grobner-Shirshov basis for extensions of algebras. Algebra Colloq. 16 283-292
(2009)
157
[72] Chen, Y.S., Chen, Y.Q.: Grobner-Shirshov bases for matabelian Lie algebras. J. Algebra
358, 143-161 (2012)
[73] Chen, Y.Q., Chen, W.P., Li, J.: Grobner-Shirshov bases for plactic monoids. Preprint.
[74] Chen, Y.Q., Chen, Y.S., Li, Y.: Composition-Diamond lemma for differential algebras. The
Arabian Journal for Science and Engineering 34, 135-145 (2009)
[75] Chen, Y.Q., Chen, W.S., Luo, R.I.: Word problem for Novikov’s and Boone’s group via
Grobner-Shirshov bases. Southeast Asian Bull. Math. 32, 863-877 (2008)
[76] Chen, Y.Q., Chen, Y.S., Zhong, C.Y.: Composition-Diamond lemma for modules.
Czechoslovak Math. J. 60, 59-76 (2010)
[77] Chen,Y.Q., Li,Y.: Some remarks for the Akivis algebras and the Pre-Lie algebras.
Czechoslovak Math. J. 61(136), 707-720 (2011)
[78] Chen, Y.Q., Li, Y., Tang, Q.Y.: Grobner-Shirshov bases for some Lie algebras. Preprint.
[79] Chen,Y.Q., Li, J., Zeng, M.J.: Composition-Diamond Lemma for Non-associative Algebras
over a Commutative Algebra. Southeast Asian Bull. Math., 34, 629 - 638 (2010).
[80] Chen, Y.Q., Mo, Q.H.: Artin-Markov normal form for braid group. Southeast Asian Bull.
Math. 33, 403-419 (2009)
[81] Chen, Y.Q., Mo, Q.H.: Embedding dendriform algebra into its universal enveloping Rota-
Baxter algebra. Proc. Am. Math. Soc. 139, 4207-4216 (2011)
[82] Chen, Y.Q., Mo, Q.H.: Grobner-Shirshov bases for free partially commutative Lie algebras.
Commun. Algebra, to appear.
[83] Chen, Y.Q., Qiu, J.J.: Grobner-Shirshov basis for the Chinese monoid. J. Algebra Appl.
7, 623-628 (2008)
[84] Chen, Y.Q., Shao, H.S., Shum, K.P.: On Rosso-Yamane theorem on PBW basis of Uq(AN ).
CUBO A Mathematical Journal 10, 171-194 (2008)
[85] Chen, Y.Q., Zhong, C.Y.: Grobner-Shirshov basis for HNN extensions of groups and for
the alternative group. Commun. Algebra 36, 94-103 (2008)
[86] Chen, Y.Q., Zhong, C.Y.: Grobner-Shirshov basis for some one-relator groups. Algebra
Colloq. 19, 99-116 (2011)
158
[87] Chen, Y.Q., Zhong, C.Y.: Grobner-Shirshov bases for braid groups in Adyan-Thurston
generators. Algebra Colloq. 20, 309-318 (2013)
[88] Chibrikov, E.S.: On free conformal Lie algebras. Vestn. Novosib. Gos. Univ. Ser. Mat.
Mekh. Inform. 4, No. 1, 65-83 (2004)
[89] Chibrikov, E.S.: A right normed basis for free Lie algebras and Lyndon-Shirshov words. J.
Algebra 302, 593-612 (2006)
[90] Chibrikov, E.S.: The right-normed basis for a free Lie superalgebra and Lyndon-Shirshov
words. Algebra Logika 45, No. 4, 458-483 (2006)
[91] Chibrikov, E.S.: On free Sabinin algebras. Commun. Algebra 39, 4014-4035 (2011)
[92] Chibrikov, E.S.: On some embedding of Lie algebras. J. Algebra Appl. 11, No. 1, 12 p.
(2012)
[93] Cohn, P.M.: A remark on the Birkhoff-Witt theorem. Journal London Math. Soc. 38,
197-203 (1963)
[94] Cohn, P.M.: Universal algebra. Harper’s Series in Modern Mathematics. New York-
Evanston-London: Harper and Row, Publishers xv, 333 p. (1965). Second edition: Reidel,
Dordrecht-Boston (1981)
[95] Collins, D.J.: Representation of Turing reducibility by word and conjugacy problems in
finitely presented groups. Acta Math. 128, 73-90 (1972)
[96] Dotsenko, V., Khoroshkin, A.: Grobner bases for operads. Duke Mathematical Journal
153, 363-396 (2010)
[97] Eisenbud, D., Peeva, I., Sturmfels, B.: Non-commutative Grobner bases for commutative
algebras. Proc. Am. Math. Soc. 126, 687-691 (1998)
[98] Farkas, D.R., Feustel, C., Green, E.I.: Synergy in the theories of Grobner bases and path
algebras 45, 727-739 (1993)
[99] Gao, X., Guo, L., Zheng, S.H.: Constrction of free commutative integro-differential algebras
by the method of Grobner-Shirshov bases. arXiv:1302.0041
[100] Garside, A.F.: The braid group and other groups. Q. J. Math. Oxford 20, 235-254 (1969)
[101] Gelfand, S.I., Manin, Y.I.: Homological Algebra. Springer-Verlag (1999)
159
[102] Gerdt, V.P., Kornyak, V.V.: Lie algebras and superalgebras defined by a finite number of
relations: computer analysis. J. Nonlinear Math. Phys. 2, No.3-4, 367-373 (1995)
[103] Gerdt, V.P., Robuk, V.N., Sever’yanov, V.M.: The construction of finitely represented Lie
algebras. Comput. Math. Math. Phys. 36, No.11, 1493-1505 (1996)
[104] Gerdt, V.P., Kornyak, V.V.: Program for constructing a complete system of relations,
basis elements, and commutator table for finitely presented Lie algebras and superalgebras.
Program. Comput. Softw. 23, No.3, 164-172 (1997)
[105] Golod, E.S.: Standard bases and homology. in: Algebra: Some current trends. Lect. Notes
Math. 1352, 88-95 (1988)
[106] Grauert, H.: uber die Deformationen isolierter Singularitaten analytischer Mengen. Inv.
Math. 15, 171-198 (1972)
[107] Grivel P.-P.: Une histoire du theoreme de Poincare-Birkhoff-Witt. Expositiones Mathe-
maticae 22, 145-184 (2004)
[108] Guo, L., Sit, W., Zhang, R.: Differemtail type operators and Grobner-Shirshov bases. J.
Symolic Comput. (2012)
[109] Gupta, C.K., Umirbaev, U.U.: The occurrence problem for free metanilpotent Lie algebras.
Commun. Algebra 27, No.12, 5857-5876 (1999)
[110] Hall, M.: A basis for free Lie rings and higher commutators in free groups. Proc. Am.
Math. Soc. 3, 575-581 (1950)
[111] Hall, P.: A contribution to the theory of groups of prime power order. Proc. London Math.
Soc. Ser. 36, 29-95 (1933)
[112] Hermann, G.: Die Frage der endlich vielen Schritte in der Theorie der Polynomideale.
Ann. Math. 95, 736-788 (1926)
[113] Higman, G.: Subgroups of finitely presented groups. Proc. Royal Soc. London (Series A)
262, 455-475 (1961)
[114] Hironaka, H.: Resolution of singularities of an algebraic variety over a field of characteristic
zero, I,II. Ann. Math. 79, 109-203; 205-326 (1964)
[115] Jacobson, N.: Basic Algebra I, W. H.Freeman and Company (1985)
160
[116] Janet, M.: Sur les systemes d’Equations aux derivees partielles. J. Math. Pure et Appl. 3,
65-151 (1920)
[117] Jones, V.F.R.: Hecke algebra representations of braid groups and link polynimials. Ann.
Math. 128, 335-388 (1987)
[118] Kac, G.: Infinite dimensional Lie algebras. Cambridge University Press, Cambridge (1990)
[119] Kalorkoti, K.: Decision problems in group theory. Proc. Lond. Math. Soc., III. Ser. 44,
312-332 (1982)
[120] Kalorkoti, K.: Turing degrees and the word and conjugacy problems for finitely presented
groups. Southeast Asian Bull. Math. 30, 855-887 (2006)
[121] Kalorkoti, K.: A finitely presented group with almost solvable conjugacy problem. Asian-
Eur. J. Math. 2, 611-635 (2009)
[122] Kalorkoti, K.: Sufficiency conditions for Bokut’ normal forms. Commun. Algebra 39,
2862-2873 (2011)
[123] Kandri-Rody, A., Weispfenning, V.: Non-commutative Grobner bases in algebras of solv-
able type A. Journal of Symbolic Computation 9, 1-26 (1990)
[124] Kang, S.-J., Lee, K.-H.: Grobner-Shirshov bases for representation theory. J. Korean Math.
Soc. 37, 55-72 (2000)
[125] Kang, S.-J., Lee, K.-H.: Grobner-Shirshov bases for irreducible sln+1-modules. J. Algebra
232, 1-20 (2000)
[126] Kang, S.-J.; Lee, I.-S., Lee, K.-H., Oh, H.: Hecke algebras, Specht modules and Grobner-
Shirshov bases. J. Algebra 252, 258-292 (2002)
[127] Kang, S.-J.; Lee, I.-S., Lee, K.-H., Oh, H.: Representations of Ariki-Koike algebras and
Grobner-Shirshov bases. Proc. Lond. Math. Soc., III. Ser. 89, 54-70 (2004)
[128] Kang, S.-J., Lee, K.-H.: Linear algebraic approach to Grobner-Shirshov basis theory. J.
Algebra 313, 988-1004 (2007)
[129] Kharchenko, V.K.: A quantum analog of the PoincarW-Birkhoff-Witt theorem. Algebra
Logika 38, No. 4, 476-507 (1999)
[130] Kharchenko, V.K.: A combinatorial approach to the quantification of Lie algebras. Pac.
J. Math. 203, 191-233 (2002)
161
[131] Kharchenko, V.K.: Braided version of Shirshov-Witt theorem. J. Algebra 294, 196-225
(2005)
[132] Kharchenko, V.K.: PBW-bases of coideal subalgebras and a freeness theorem. Trans. Am.
Math. Soc. 360, 5121-5143 (2008)
[133] Kharchenko, V.K.: Triangular decomposition of right coideal subalgebras. J. Algebra 324,
3048-3089 (2010)
[134] Kharlampovich, O.G., Sapir, M.V.: Algorithmic problems in varieties. Inter. J. Algebra
Compt. 5, 379-602 (1995)
[135] Knuth, E., Bendix, P.B.: Simple word problems in universal algebras. In: Computational
problems in abstract algebra. Pergamon Press, Oxford 263-297 (1970)
[136] Kolchin, E.R.: Differential algebras and algebraic groups. Academic Press, New York
(1973)
[137] Kozybaev, D., Makar-Limanov, L., Umirbaev, U.: The Freiheitssatz and autoumorphisms
of free right-symmetric algebras. Asian-European Journal of Mathematics 1, 243-254 (2008)
[138] Kukin, G.P.: On the word problem for Lie algebras. Sibirsk. Mat. Zh. 18, 1194-1197 (1977)
[139] Kurosh, A.G.: Nonassociative free algebras and free products of algebras. Mat. Sb. 20,
239-262 (1947)
[140] Lascoux, A., Leclerc, B., Thibon, J.Y.: The plactic monoid. in: Algebraic Combinatorics
on Words. Cambridge University Press, Cambridge (2002)
[141] Latyshev, V.N.: General version of standard bases in linear structures. Bahturin, Yu. (ed.),
Algebra. Berlin: Walter de Gruyter. 215-226 (2000)
[142] Latyshev, V.N.: An improved version of standard bases. Krob, Daniel (ed.) et al., Formal
power series and algebraic combinatorics. Berlin: Springer. 496-505 (2000)
[143] Lazard, M.: Sur les algebres enveloppantes universelles de certaines algebres de Lie. CRAS
Paris 234(1), 788-791 (1952)
[144] Lazard, M.: Sur les algebres enveloppantes de certaines algebres de Lie. Publ. Sci. Univ.
Alger serie A 1, 281-294 (1954)
[145] Lazard, M.: Groupes, anneaux de Lie et probleme de Burnside. Istituto Matematico dell’
Universita di Roma (1960)
162
[146] Lee, D.V.: Grobner-Shirshov bases and normal forms for the Coxeter groups E6 and E7.
Shum, K.P. (ed.) et al., Advances in algebra and combinatorics. Hackensack, NJ: World
Scientific. 243-255 (2008)
[147] Loday J.-L.: Dialgebras, in: Dialgebras and related operads. Lecture Notes in Mathematics
Vol. 1763. Berlin: Springer Verl. 7-66 (2001)
[148] Lothaire, M.: Combinatorics on words. Addison-Wesley Publishing Company, xix, 238 p.
(1983). Second edition: Cambridge University Press, Cambridge (1977)
[149] Lothaire, M.: Algebraic Combinatorics on Words. Cambridge University Press, Cambridge
(2002)
[150] Lusztig, G.: Canonical bases arising from quantized enveloping algebras. J. Amer. Math.
Soc. 3, 447-498(1990)
[151] Lyndon, R.C.: On Burnside’s problem I. Trans. Amer. Math. Soc. 77, 202-215 (1954)
[152] Maclane, S.: Homology. Springer-Verlag (1963)
[153] Macaulay, F.S.: Some properties of enumeration in the theory of modular systems. Proc.
London Math. Soc. 26, 531-555 (1927)
[154] Makar-Limanov, L.: A version of the Poincare-Birkhoff-Witt theorem. Bull. Lond. Math.
Soc. 26, No.3, 273-276 (1994)
[155] Makar-Limanov, L., Umirbaev, U.U.: The Freiheitssatz for Poisson algebras. J. Algebra
328, No. 1, 495-503 (2011)
[156] Markov, A.A.: An introduction to the algebraical theory of braids. Proceedings of the
Steklov Mat. Ins. RAS, vol. 16, 1945.
[157] Markov, A.A.: Impossibility of some algorithms in the theory of some associative system.
Dokl. Akad. Nauk SSSR 55, 587-590 (1947)
[158] Michel, J.: Bases des algebres de Lie et serie de Hausdorff, Semin. P. Dubreil, 27e annee
1973/74, Algebre, Fasc. 1, Expose 6, 9 p. (1975)
[159] Michel, J.: Calculs dans les algebres de Lie libres: la serie de Hausdorff et le probleme de
Burnside, Asterisque 38/39, 139-148 (1976)
[160] Mikhalev, A.A.: A composition lemma and the equality problem for color Lie superalge-
bras. Mosc. Univ. Math. Bull. 44, No.5, 87-90 (1989)
163
[161] Mikhalev, A.A.: The composition lemma for color Lie superalgebras and for Lie p-
superalgebras. Algebra, Proc. Int. Conf. Memory A.I. Mal’cev, Eds. L.A. Bokut, Yu.L.
Ershov, A.I. Kostrikin, Novosibirsk/USSR 1989, Contemp. Math. 131, Pt. 2, 91-104 (1992)
[162] Mikhalev, A.A.: Shirshov composition techniques in Lie superalgebras (noncommutative
Grobner bases). J. Math. Sci., New York 80, No.5, 2153-2160 (1996)
[163] Mikhalev, A.A., Shpilrain, V., Umirbaev, U.U.: On isomorphism of Lie algebras with one
defining relation. Int. J. Algebra Comput. 14, No. 3, 389-393 (2004)
[164] Mikhalev, A.A., Zolotykh, A.A.: Combinatorial aspects of Lie superalgebras. Boca Raton,
FL: CRC Press. viii, 260 p. (1995)
[165] Mikhalev, A.A., Zolotykh, A.A.: Standard Grobner-Shirshov bases of free algebras over
rings, I. Free associative algebras. Internat. J. Algebra Comput. 8, 689-726 (1998)
[166] Mora, F.: Grobner bases for non-commutative polynomial rings. in: Algebraic algorithms
and error-correcting codes. Lect. Notes Comput. Sci. 229, 353-362 (1986)
[167] Newman, M.H.A.: On theories with a combinatorial definition of “equivalence”. Ann.
Math. 43, 223-243 (1942)
[168] Nouaze, Y., Revoy, P.: Un eas particulier du theoreme de Poincare-Birkhoff-Witt. CRAS
Paris serie A, 329-331 (1971)
[169] Novikov, P.S.: On algorithmic undecidability of the word problem in the theory of groups.
Trudy MKat. Inst. Steklov. 44, 1-144 (1955)
[170] Obul, A., Yunus, G.: Grobner-Shirshov basis of quantum group of type E6. J. Algebra
346, 248-265 (2011)
[171] Odesskii, A.: Introduction to the theory of elliptic algebras.
data.imf.au.dk/conferences/FMOA05/
[172] Poliakova, O., Schein, B.M.: A new construction for free inverse semigroups. J. Algebra
288, 20-58 (2005)
[173] Polishchuk, A., Positselski, L.: Quadratic algebras, AMS (2005)
[174] Poroshenko, E.N.: Grobner-Shirshov bases for Kac-Moody algebras A(1)n and B
(1)n . Krob,
Daniel et al. Formal power series and algebraic combinatorics. Berlin Springer 552-563
(2000)
164
[175] Poroshenko, E.N.: Grobner-Shirshov bases for Kac-Moody algebras of types C(1)n and D
(1)n .
Vestn. Novosib. Gos. Univ. Ser. Mat. Mekh. Inform. 2, 58-70 (2002)
[176] Poroshenko, E.N.: Grobner-Shirshov bases for Kac-Moody algebras of type A(1)n . Commun.
Algebra 30, 2617-2637 (2002)
[177] Poroshenko, E.N.: Bases for partially commutative Lie algebras. Algebra Logika 50, 405-
417 (2011)
[178] Post, E.: A variant of a recursively unsolvable problem. Bull. Amer. Math. Soc. 52, 264-268
(1946)
[179] Priddy, S.B.: Koszul resolutions. Trans. Amer. Math. Soc. 152, 39-60 (1970)
[180] Qiu, J.J., Chen, Y.Q: Composition-Diamond lemma for λ-differential associative algebras
with multiple operators. J. Algebra Appl. 9, 223-239 (2010)
[181] Qiu, J.J.: Grobner-Shirshov bases for commutative algebras with multiple operators and
free commutative Rota-Baxter algebras. Asian-European Jour. Math. to appear.
[182] Rabin,M.: Recursice unsolvability of group thepretic problems. Ann. of Math. 67, (1)
172-194, (1958).
[183] Razmyslov, Yu.P.: Identities of algebras and their representations. Providence, RI: AMS
xiii, 318p. (1994)
[184] Ren, Y.H., Obul, A.: Grobner-Shirshov basis of quantum group of type G2. Commun.
Algebra 39, 1510-1518 (2011)
[185] Reutenauer, C.: Free Lie algebras. Oxford University Press, New York (1993)
[186] Ringel, C.M.: Hall algebras and quantum groups. Invent. Math. 101, 583-592 (1990)
[187] Ringel, C.M.: PBW-bases of quantum groups. J. reine angew. Math. 170, 51-88 (1996)
[188] Ritt, J.F. Differential algebras. AMS, New York (1950)
[189] Roitman, M.: On the free conformal and vertex algebras. J. Algebra 217, 496-527 (1999).
[190] Rosso, M.: An analogue of the Poincare-Birkhoff-Witt theorem and the universal R-matrix
of Uq(sl(N + 1)). Commun. Math. Phys. 124, 307-318 (1989)
[191] Schutzenberger, M.P., Sherman, S.: On a formal product over the conjugate classes in a
free group. J. Math. Anal. Appl. 7, 482-488 (1963)
165
[192] Schutzenberger, M.P.: On a factorization of free monoids. Proc. Am. Math. Soc. 16, 21-24
(1965)
[193] Segal, D.: Free left-symmetric algebras and an analogue of the Poincare-Birkhoff-Witt
Theorem. J. Algebra 164, 750-772 (1994)
[194] Shirshov, A.I.: Some problems in the theory of non-associative rings and
algebras, Candidate of Science Thesis, Moscow State University, 1953. see
http://math.nsc.ru/LBRT/a1/ShirshovPhD.djvu
[195] Shirshov, A.I.: Subalgebras of free Lie algebras. Uspekhi Mat. Nauk 8, (3) 173 (1953)
[196] Shirshov, A.I.: On the representation of Lie rings in associative rings. Uspekhi Mat. Nauk
N. S. 8, (5)(57) 173-175 (1953)
[197] Shirshov, A.I.: Subalgebras of free commutative and free anticommutative algebras. Mat.
Sb. 4, (1) 82-88 (1954)
[198] Shirshov, A.I.: On free Lie rings. Mat. Sb. 45, (2) 113-122 (1958)
[199] Shirshov, A.I.: Some problems in the theory of rings that are nearly associative. Uspekhi
Mat. Nauk 13, no. 6 (84) 3-20 (1958)
[200] Shirshov, A.I.: On the bases of a free Lie algebra. Algebra Logika 1, (1) 14-19 (1962)
[201] Shirshov, A.I.: On a hypothesis in the theory of Lie algebras. Sibirsk Mat. Zh. 3, (2)
297-301 (1962)
[202] Shirshov, A.I.: Some algorithmic problem for Lie algebras. Sibirsk. Mat. Zh. 3, (2) 292-296
(1962); English translation in SIGSAM Bull. 33, 3-6 (1999)
[203] Shirshov, A.I.: Some algorithmic problem for ε-algebras. Sibirsk. Mat. Zh. 3, 132-137
(1962)
[204] Selected works of A.I. Shirshov. Eds. Bokut, L.A., Latyshev, V., Shestakov, I., Zelmanov,
E., Bremner, Trs.M., Kochetov, M. Birkhauser, Basel, Boston, Berlin (2009)
[205] Tits, J.: Le probleme des mots dans les groupes de Coxeter. Symposia Math. 1, 175-185
(1968)
[206] Turing, A.M.: The word problem in semi-groups with cancellation. Ann. Math. 52, 191-505
(1950)
166
[207] Ufnarovski, V.A.: Combinatorial and asymptotic methods in algebra, Algebra VI (57)
(1995) 1-196. Springer, Berlin Heidelberg New York.
[208] Ufnarovski, V.A.: Introduction to noncommutative Grobner bases theory. Buchberger,
Bruno (ed.) et al., Grobner bases and applications. Cambridge: Cambridge University
Press. Lond. Math. Soc. Lect. Note Ser. 251, 259-280 (1998)
[209] Umirbaev, U.U.: Equality problem for center-by-metabelian Lie algebras. Algebra Logic
23, 209-219 (1984)
[210] Umirbaev, U.U.: The occurrence problem for Lie algebras. Algebra Logic 32, No.3, 173-181
(1993)
[211] Umirbaev, U.U.: Algorithmic problems in associative algebras. Algebra Logic 32, No.4,
244-255 (1993)
[212] Viennot, G.: Algebras de Lie libres et monoid libres. Lecture Notes in Mathematics, 1978,
No.691.
[213] Witt, E.: Treue Darstellung Liescher Ringe. J. reine u. angew. Math. 177, 152-160 (1937)
[214] Witt, E.: Die Unterringe der freien Lienge Ringe. Math. Z. 64, 195-216 (1956)
[215] Wu, W.-T.: On the decision problem and the mechanization of theorem proving an ele-
mentary geometry. Scientia Sinica 21, 157-179 (1978)
[216] Yamane, I.: A Poincare-Birkhoff-Witt theorem for quantized universal enveloping algebras
of type AN . Publ. RIMS. Kyoto Univ. 25, 503-520 (1989)
[217] Yunus, G., Obul, A.: Grobner-Shirshov basis of quantum group of type D4. Chin. Ann.
Math. 32, B(4) 581-592 (2011)
[218] Yunus, G., Gao, Z.Z., Obul, A.: Grobner-Shirshov bases of quantum groups. Algebra
Colloq. to appear.
[219] Zelmanov, E.: Nil rings and periodic groups. KMS Lecture Notes in Mathematics. Seoul:
Korean Mathematical Society, x, 79 p. (1992)
[220] Zhou, M., Winkler, F.: Grobner bases in difference-differential modules and difference-
differential dimension polynomials. Science in China, Ser. A: Mathematics 51, 1732-1752
(2008)
167
[221] Zhukov, A.I.: Complete systems of defining relations in noassociative algebras. Mat.
Sbornik 69, (27) 267-280 (1950)
[222] Zolotykh, A.A., Mikhalev, A.A.: A complex of algorithms for computations in Lie super-
algebras. Program. Comput. Softw. 23, No.1, 8-16 (1997)
[223] Zolotykh, A.A., Mikhalev, A.A.: Algorithms for construction of standard Grobner-
Shirshov bases of ideals of free algebras over commutative rings. Program. Comput. Softw.
24, No.6, 271-272 (1998)
168
Notation
CD-lemma: Composition-Diamond lemma.
GS basis: Grobner-Shirshov basis.
LS word (basis): Lyndon-Shirshov word (basis).
ALSW(X): the set of all associative Lyndon-Shirshov words in X.
NLSW(X): the set of all non-associative Lyndon-Shirshov words in X.
PBW theorem: Poincare-Birkhoff-Witt theorem.
X∗: the free monoid generated by the set X.
[X]: the free commutative monoid generated by the set X.
X∗∗: the set of all non-associative words (u) in X.
gp〈X|S〉: the group generated by X with defining relations S.
sgp〈X|S〉: the semigroup generated by X with defining relations S.
k: a field.
K: a commutative algebra over k with unit.
k〈X〉: the free associative algebra over k generated by X.
k〈X|S〉: the associative algebra over k with generators X and defining relations S.
Sc: Grobner-Shirshov completion of S.
Id(S): the ideal generated by the set S.
s: the leading term of the polynomial s.
Irr(S): the set of all monomials which do not contain any subwords s, s ∈ S.
k[X]: the polynomial algebra over k generated by X.
Lie(X): the free Lie algebra over k generated by X.
LieK(X): the free Lie algebra over a commutative algebra K generated by X.
169
Top Related