You are on page 1of 29

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233052507

Gröbner Bases for Ideals of σ-PBW Extensions

Article  in  Communications in Algebra · January 2011


DOI: 10.1080/00927870903431209

CITATIONS READS
58 272

2 authors:

Claudia Gallego Oswaldo Lezama


National University of Colombia National University of Colombia
8 PUBLICATIONS   78 CITATIONS    41 PUBLICATIONS   247 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Zariski cancellation problem for skew PBW extensions and other noncommutative algebras View project

All content following this page was uploaded by Oswaldo Lezama on 18 August 2014.

The user has requested enhancement of the downloaded file.


Gröbner Bases for Ideals
of σ − P BW Extensions
Claudia Gallego∗ & Oswaldo Lezama†
Grupo de Álgebra Conmutativa Computacional - SAC2
Departamento de Matemáticas
Universidad Nacional de Colombia, Bogotá, COLOMBIA

Abstract
In this paper we introduce the σ − P W B extensions and construct the
theory of Gröbner bases for the left ideals of them. We prove the Hilbert’s
basis theorem and the division algorithm for this more general class of
Poincaré-Birkhoff-Witt extensions. For the particular case of bijective and
quasi-commutative σ − P W B extensions, we implement the Buchberger’s
algorithm for computing Gröbner bases of left ideals.

Key words and phrases. Noncommutative Gröbner bases, P BW exten-


sions, Buchberger’s Algorithm, Noetherian polynomial noncommutative
rings.

2000 Mathematics Subject Classification. Primary: 16Z05. Secondary:


18G15, 18G20.

∗ The first author was sponsored by Vicerrectorı́a Académica UN.


† The second author was partially supported by DIB-UN, project 8003130.

1
2 Claudia Gallego & Oswaldo Lezama

1 σ − P BW extensions
P BW extensions were defined by Bell and Goodearl in 1988 in [1]; let R and
A be rings, we say that A is a P BW (Poincaré-Birkhoff-Witt) extension of R,
denoted by A = Rhx1 , . . . , xn i, if the following conditions hold:

(i) R ⊆ A.

(ii) There exist finite elements x1 , . . . , xn ∈ A such that A is a left R-free


module with basis

M on(A) := {xα = xα αn n
1 · · · xn |α = (α1 , . . . , αn ) ∈ N }.
1

(iii) xi r − rxi ∈ R, for each r ∈ R and 1 ≤ i ≤ n.

(iv) xi xj − xj xi ∈ R + Rx1 + · · · + Rxn , for any 1 ≤ i, j ≤ n.

Many important class of rings and algebras are P BW extensions, for example:

(a) If A = R[t1 , . . . , tn ] is the habitual polynomial ring, so ti r − rti = 0 and


ti tj − tj ti = 0, for any 1 ≤ i, j ≤ n. The R-free basis is M on(A).

(b) Any skew polynomial ring A = R[x; σ, δ] of derivation type, i.e., when
σ = iR , is a P BW extension, in this case xr − rx = δ(r), and xx − xx = 0;
the R-free basis is {xl |l ≥ 0} (see [5] and [8]).

(c) Let R = K[t1 , . . . , tm ], K a field; if A = R[x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] is an


Ore algebra with σi = iR , for 1 ≤ i ≤ n, then A is a P BW extension of
R. In this case xi r = rxi + δi (r), xi xj − xj xi = 0 and δi (r) ∈ R, for any
1 ≤ i, j ≤ n, r ∈ R.

(d) The Weyl algebra An (K) = K[t1 , . . . , tn ][x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ], K a


field, is another important example of P BW extension: in fact, the Weyl
algebra An (K) is an Ore algebra with σi = iK[t1 ,...,tn ] and δi = ∂/∂ti , for
1 ≤ i ≤ n. In this situation, xi p = pxi + ∂p/∂ti , xi xj − xj xi = 0, for any
p ∈ K[t1 , . . . , tn ] and 1 ≤ i, j ≤ n (see [4] and [8]).

(e) Let K be a commutative ring and G a finite dimensional Lie algebra over
K with basis {x1 , . . . , xn }; the universal enveloping algebra of G, U (G), is
a P BW extension of K (see [8]), [9] and [10]). In this case, xi k − kxi = 0
and xi xj − xj xi = [xi , xj ] ∈ G = K + Kx1 + · · · + Kxn , for any k ∈ K
and 1 ≤ i, j ≤ n.

(f) Let K, G, {x1 , . . . , xn } and U (G) be as in the previous example; let R


be a K-algebra containing K. The tensor product A := R ⊗K U (G) of
algebras R and U (G) is a P BW extension of R (see [9]). In this case
{1 ⊗ xα αn
1 · · · xn |αi ≥ 0, 1 ≤ i ≤ n} is a R-free basis of A; moreover,
1

(r ⊗ 1)(1 ⊗ xi ) − (1 ⊗ xi )(r ⊗ 1) = 0 and (1 ⊗ xi )(1 ⊗ xj ) − (1 ⊗ xj )(1 ⊗ xi ) =


1 ⊗ [xi , xj ] ∈ 1 ⊗ G ⊆ R ⊗ 1 + R ⊗ x1 + · · · + R ⊗ xn .
Gröbner Bases for Ideals of σ − P BW Extensions 3

We observe that if in the examples (b) and (c) above, σi 6= iR , then the skew
polynomial ring R[x; σ, δ] and the Ore algebra R[x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] are not
P BW extensions, since for example xr = σ(r)x + δ(r) ∈ R + Rx.

We introduce the following generalization of P BW extensions (compare also


with [2]).
Definition 1. Let R and A be rings, we say that A is a σ − P BW extension
of R (or skew P BW extension), if the following conditions hold:
(i) R ⊆ A.
(ii) There exist finite elements x1 , . . . , xn ∈ A such A is a left R-free module
with basis

M on(A) := {xα = xα αn n
1 · · · xn |α = (α1 , . . . , αn ) ∈ N }.
1

In this case we say also that A is a left polynomial ring over R with
respect to {x1 , . . . , xn } and M on(A) is the set of standard monomials of
A. Moreover, x01 · · · x0n := 1 ∈ M on(A).
(iii) For every 1 ≤ i ≤ n and r ∈ R − {0} there exists ci,r ∈ R − {0} such that

xi r − ci,r xi ∈ R. (1.1)

(iv) For every 1 ≤ i, j ≤ n there exists ci,j ∈ R − {0} such that

xj xi − ci,j xi xj ∈ R + Rx1 + · · · + Rxn . (1.2)

Under these conditions we will write A = σ(R)hx1 , . . . , xn i.


Remark 2. (i) Since that M on(A) is a R-basis for A, the elements ci,r and ci,j
in the above definition are unique.
(ii) If r = 0, then ci,0 = 0. Moreover, in (iv), ci,i = 1.
(iii) Let i < j, by (1.2), there exist cj,i , ci,j ∈ R such that xi xj − cj,i xj xi ∈
R + Rx1 + · · · + Rxn and xj xi − ci,j xi xj ∈ R + Rx1 + · · · + Rxn , but since
M on(A) is a R-basis then 1 = cj,i ci,j , i.e., for every 1 ≤ i < j ≤ n, ci,j has a
left inverse and cj,i has a right inverse.
(iv) Each element f ∈ A − {0} has a unique representation in the form f =
c1 X1 + · · · + ct Xt , with ci ∈ R − {0} and Xi ∈ M on(A), 1 ≤ i ≤ t.
The following proposition justifies the notation that we have introduced for the
skew P BW extensions.
Proposition 3. Let A be a σ−P BW extension of R. Then, for every 1 ≤ i ≤ n,
there exist an injective ring endomorphism σi : R → R and a σi -derivation
δi : R → R such that
xi r = σi (r)xi + δi (r),
4 Claudia Gallego & Oswaldo Lezama

for each r ∈ R.
Proof. For every 1 ≤ i ≤ n and each r ∈ R we have elements ci,r , ri ∈ R such
that xi r = ci,r xi + ri ; since M on(A) is a R-basis of A, ci,r and ri are unique
for r, so we define σi , δi : R → R by σi (r) := ci,r , δi (r) := ri . It is easy
to check that σi is a ring endomorphism and δi is a σi -derivation of R, i.e.,
δi (r + r0 ) = δi (r) + δi (r0 ) and δi (rr0 ) = σi (r)δi (r0 ) + δi (r)r0 , for any r, r0 ∈ R.
Moreover, by the Definition 1 (iii), ci,r 6= 0 for r 6= 0. This means that σi is
injective.
A particular case of σ − P BW extension is when all derivations δi are zero.
Another interesting case is when all σi are bijective. We have the following
definition.
Definition 4. Let A be a σ − P BW extension.
(a) A is quasi-commutative if the conditions (iii) and (iv) in the Definition 1
are replaced by
(iii0 ) For every 1 ≤ i ≤ n and r ∈ R − {0} there exists ci,r ∈ R − {0} such
that
xi r = ci,r xi . (1.3)
(iv 0 ) For every 1 ≤ i, j ≤ n there exists ci,j ∈ R − {0} such that

xj xi = ci,j xi xj . (1.4)

(b) A is bijective if σi is bijective for every 1 ≤ i ≤ n and ci,j is invertible for


any 1 ≤ i < j ≤ n.
Some examples of σ − P BW extensions are the following ones.
Example 5. (i) Any P BW extension is a bijective σ − P BW extension since
in this case ci,r = r for each 1 ≤ i ≤ n and ci,j = 1 for every 1 ≤ i, j ≤ n.
(ii) Any skew polynomial ring R[x; σ, δ], with σ injective, is a σ − P BW ex-
tension; in this case we have R[x; σ, δ] = σ(R)hxi. If additionally δ = 0, then
R[x; σ] is quasi-commutative.
(iii) Any iterated skew polynomial ring R[x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] is a σ −P BW
extension if it satisfies the following conditions:
For 1 ≤ i ≤ n, σi is injective
For every r ∈ R and 1 ≤ i ≤ n, σi (r), δi (r) ∈ R
For i < j, σj (xi ) = cxi + d, with c, d ∈ R, and c has a left inverse.
For i < j, δj (xi ) ∈ R + Rx1 + · · · + Rxi .
Under these conditions we have
R[x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] = σ(R)hx1 , . . . , xn i.
In particular, any Ore algebra K[t1 , . . . , tm ][x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] is a σ −
P BW extension if it satisfies the following condition:
Gröbner Bases for Ideals of σ − P BW Extensions 5

For 1 ≤ i ≤ n, σi is injective
In fact, in Ore algebras for every r ∈ K[t1 , . . . , tn ] and 1 ≤ i ≤ n, σi (r), δi (r) ∈
K[t1 , . . . , tn ], and for i < j, σj (xi ) = xi and δj (xi ) = 0 . Under these conditions
we get
K[t1 , . . . , tm ][x1 ; σ1 , δ1 ] · · · [xn ; σn , δn ] = σ(K[t1 , . . . , tm ])hx1 , . . . , xn i.
(iv) Additive analogue of the Weyl algebra: let K be a field, the K-algebra
An (q1 , . . . , qn ) is generated by x1 , . . . , xn , y1 , . . . , yn and subject to the relations:
xj xi = xi xj , yj yi = yi yj , 1 ≤ i, j ≤ n,
yi xj = xj yi , i 6= j,
yi xi = qi xi yi + 1, 1 ≤ i ≤ n,
where qi ∈ K −{0}. We observe that An (q1 , . . . , qn ) is isomorphic to the iterated
skew polynomial ring K[x1 , . . . , xn ][y1 ; σ1 , δ1 ] · · · [yn ; σn , δn ] over the commuta-
tive polynomial ring K[x1 , . . . , xn ]:
σj (yi ) := yi , δj (yi ) := 0, 1 ≤ i < j ≤ n,
σi (xj ) := xj , δi (xj ) := 0, i 6= j,
σi (xi ) := qi xi , δi (xi ) := 1, 1 ≤ i ≤ n.
Thus, An (q1 , . . . , qn ) satisfies the conditions of (iii) and is bijective; we have
An (q1 , . . . , qn ) = σ(K[x1 , . . . , xn ])hy1 , . . . , yn i.
(v) Multiplicative analogue of the Weyl algebra: let K be a field, the K-algebra
On (λji ) is generated by x1 , . . . , xn and subject to the relations:
xj xi = λji xi xj , 1 ≤ i < j ≤ n,
where λji ∈ K − {0}. We note that On (λji ) is isomorphic to the iterated skew
polynomial ring K[x1 ][x2 ; σ2 ] · · · [xn ; σn ]
σj (xi ) := λji xi , 1 ≤ i < j ≤ n.
Thus, On (λji ) satisfies the conditions of (iii), and hence
On (λji ) = σ(K[x1 ])hx2 , . . . , xn i.
Note that On (λji ) is quasi-commutative and bijective.
(vi) q-Heisenberg algebra: let K be a field , the K-algebra hn (q) is generated
by x1 , . . . , xn , y1 , . . . , yn , z1 , . . . , zn and subject to the relations:
xj xi = xi xj , zj zi = zi zj , yj yi = yi yj , 1 ≤ i, j ≤ n,
zj yi = yi zj , zj xi = xi zj , yj xi = xi yj , i 6= j,
zi yi = qyi zi , zi xi = q −1 xi zi + yi , yi xi = qxi yi , 1 ≤ i ≤ n,
with q ∈ K −{0}. Note that hn (q) is isomorphic to the iterated skew polynomial
ring K[x1 , . . . , xn ][y1 ; σ1 ] · · · [yn ; σn ][z1 ; θ1 , δ1 ] · · · [zn ; θn , δn ] on the commutative
polynomial ring K[x1 , . . . , xn ]:
6 Claudia Gallego & Oswaldo Lezama

θj (zi ) := zi , δj (zi ) := 0, σj (yi ) := yi , 1 ≤ i < j ≤ n,


θj (yi ) := yi , δj (yi ) := 0, θj (xi ) := xi , δj (xi ) := 0, σj (xi ) := xi , i 6= j,
θi (yi ) := qyi , δi (yi ) := 0, θi (xi ) := q −1 xi , δi (xi ) := yi , σi (xi ) := qxi , 1 ≤ i ≤ n,

Since δi (xi ) = yi ∈ / K[x1 , . . . , xn ], then hn (q) is not a σ − P BW extension of


K[x1 , . . . , xn ] (see the example (iii)), however hn (q) is a bijective σ − P BW
extension of K:

hn (q) = σ(K) < x1 , . . . , xn ; y1 , . . . , yn ; z1 , . . . , zn >.

Definition 6. Let A be a σ − P BW extension of R with endomorphisms σi ,


1 ≤ i ≤ n, as in the Proposition 3.

(i) For α = (α1 , . . . , αn ) ∈ Nn , σ α := σ1α1 · · · σnαn , |α| := α1 + · · · + αn . If


β = (β1 , . . . , βn ) ∈ Nn , then α + β := (α1 + β1 , . . . , αn + βn ).

(ii) For X = xα ∈ M on(A), exp(X) := α and deg(X) := |α|.

(iii) Let 0 6= f ∈ A, t(f ) is the finite set of terms that conform f , i.e., if
f = c1 X1 + · · · + ct Xt , with Xi ∈ M on(A) and ci ∈ R − {0}, then
t(f ) := {c1 X1 , . . . , ct Xt }.

(iv) Let f be as in (iii), then deg(f ) := max{deg(Xi )}ti=1 .

The σ − P BW extensions can be characterized in a similar way as was done in


[2] for P BW rings.

Theorem 7. Let A be a left polynomial ring over R w.r.t {x1 , . . . , xn }. A is a


σ − P BW extension of R if and only if the following conditions hold:

(a) For every xα ∈ M on(A) and every 0 6= r ∈ R there exists unique elements
rα := σ α (r) ∈ R − {0} and pα,r ∈ A such that

xα r = rα xα + pα,r , (1.5)

where pα,r = 0 or deg(pα,r ) < |α| if pα,r 6= 0. Moreover, if r is left


invertible, then rα is left invertible.

(b) For every xα , xβ ∈ M on(A) there exist unique elements cα,β ∈ R and
pα,β ∈ A such that
xα xβ = cα,β xα+β + pα,β , (1.6)
where cα,β is left invertible, pα,β = 0 or deg(pα,β ) < |α + β| if pα,β 6= 0.

Proof. ⇒) We divide the proof of (a) in three steps.


Step 1. For each 1 ≤ i ≤ n, 0 6= r ∈ R and for every k ∈ N,

xki r = rk xki + pk,r ,


Gröbner Bases for Ideals of σ − P BW Extensions 7

where rk := σik (r) ∈ R − {0}, pk,r ∈ A and pk,r = 0 or deg(pk,r ) < k. Moreover,
if r is left invertible, then rk is left invertible.
In fact, we will prove this by induction on k: for k = 0 we have r0 = σi0 (r) = r
and p0,r = 0; for k = 1 we have xi r = σi (r)xi + δi (r), so r1 := σi (r) 6= 0 and
p1,r = δi (r), with δi (r) = 0 or deg(δi (r)) = 0 < 1 (if r is left invertible, then
σi (r) is left invertible). By induction we have xk+1 i r = xi xki r = xi (rk xki + pk,r ),
k
where rk = σi (r) ∈ R − {0}, pk,r ∈ A, pk,r = 0 or deg(pk,r ) < k (if r is left
invertible, then rk is left invertible). So, xk+1
i r = (xi rk )xki +xi pk,r = (σi (rk )xi +
k k+1 k
δi (rk ))xi + xi pk,r = σi (rk )xi + δi (rk )xi + xi pk,r . Note that rk+1 := σi (rk ) =
σi (σik (r)) = σik+1 (r) 6= 0 since rk 6= 0 and σi is injective; moreover, pk+1,r :=
δi (rk )xki + xi pk,r = 0 or deg(pk+1,r ) < k + 1 since pk,r = 0 or deg(xi pk,r ) ≤ k <
k + 1 (if rk is left invertible, then σi (rk ) is left invertible).
Step 2. We complete the proof by induction on the number of variables involved
in xα . For one variable only, the proof is the content of the step 1. Then,
αn−1 αn−1
xα r = xα αn α1 αn α1 αn
1 · · · xn r = x1 · · · xn−1 (xn r) = x1 · · · xn−1 (rαn xn + pαn ,r ), with
1

αn
rαn = σn (r) 6= 0 and pαn ,r ∈ A, pαn ,r = 0 or deg(pαn ,r ) < αn (if r is left
invertible, then rαn is left invertible). So by induction
α α
xα r = (xα n−1 αn α1 n−1
1 · · · xn−1 rαn )xn + x1 · · · xn−1 pαn ,r
1

α1 αn−1 αn α1 αn−1
= (rα x1 · · · xn−1 + qα,rαn )xn + x1 · · · xn−1 pαn ,r
α1 αn−1 αn
= rα x1 · · · xn−1 xn + pα,r
= rα xα + pα,r ,
α α
where rα = σ1α1 · · · σn−1
n−1
(rαn ) = (σ1α1 · · · σn−1 σn )(r) = σ α (r) 6= 0 (by induc-
n−1 αn

α1 αn−1
tion and since rαn 6= 0), qα,rαn ∈ A and pα,r := qα,rαn xα n +x1 · · · xn−1 pαn ,r ∈
n

A (if rαn is left invertible, then rα is left invertible); note that pα,r = 0 or
deg(pα,r ) < |α| = α1 + · · · + αn since pαn ,r = 0 or deg(pαn ,r ) < αn , and,
qα,rαn = 0 or deg(qα,rαn ) < α1 + · · · + αn−1 .
Step 3. Since M on(A) is R-basis for A, then rα and pα,r are unique.
Now we will consider the proof of (b). We divide the proof also in three steps.
Step 3.1. We will prove first that for i < j and k, m ≥ 0
xkj xm m k
i = ck,m xi xj + pk,m ,

with ck,m ∈ R left invertible, pk,m ∈ A, pk,m = 0 or deg(pk,m ) < k + m. For


this we will use double induction, on k and on m. For k = 0 we have c0,m := 1,
p0,m := 0.
k = 1: For i < j and m ≥ 0 we will prove by induction on m that
xj xm m
i = c1,m xi xj + p1,m ,

with c1,m ∈ R left invertible, p1,m ∈ A, p1,m = 0 or deg(p1,m ) < 1 + m. For m =


0, c1,0 = 1, p1,0 = 0. Let m = 1, then xj xi = ci,j xi xj + p1,1 , with ci,j ∈ R left
invertible (see Definition 2 (iv)), p1,1 ∈ A, p1,1 = 0 or deg(p1,1 ) ≤ 1 < 1+1. Now
we use the induction hypothesis, so xj xm+1 i = xj xm m
i xi = (c1,m xi xj + p1,m )xi ,
with c1,m ∈ R left invertible, p1,m ∈ A, p1,m = 0 or deg(p1,m ) < 1 + m.
Then, xj xm+1
i = c1,m xm m
i xj xi + p1,m xi = c1,m xi (ci,j xi xj + p1,1 ) + p1,m xi =
8 Claudia Gallego & Oswaldo Lezama

c1,m xm m m m
i ci,j xi xj + c1,m xi p1,1 + p1,m xi = c1,m (rm xi + pm,ci,j )xi xj + c1,m xi p1,1 +
p1,m xi , where rm ∈ R is left invertible since ci,j is left invertible (part (a));
moreover pm,ci,j ∈ A, pm,ci,j = 0 or deg(pm,ci,j ) < m. Hence, xj xm+1 i =
c1,m+1 xm+1
i x j + p 1,m+1 , with c 1,m+1 := c r
1,m m ∈ R left invertible, p 1,m+1 :=
c1,m pm,ci,j xi xj +c1,m xm i p1,1 +p1,m xi ∈ A, p1,m+1 = 0 or deg(p1,m+1 ) ≤ m+1 <
m + 2. This complete the proof for k = 1.
k + 1: xk+1 j xm k m m k
i = xj xj xi = xj (ck,m xi xj + pk,m ), with ck,m ∈ R left in-
vertible, pk,m ∈ A, pk,m = 0 or deg(pk,m ) < k + m. Thus, xk+1 j xmi =
m k m k
(xj ck,m )xi xj + xj pk,m = (r1 xj + p1,ck,m )xi xj + xj pk,m , with r1 ∈ R left
invertible, p1,ck,m = 0 or deg(p1,ck,m ) < 1; then, xk+1 j xm
i = r1 xj xm k
i xj +
m k m k m k
p1,ck,m xi xj + xj pk,m = r1 (c1,m xi xj + p1,m )xj + p1,ck,m xi xj + xi pk,m , by
induction c1,m ∈ R is left invertible, p1,m ∈ A, p1,m = 0 or deg(p1,m ) < 1 + m,
hence xk+1j xmi = ck+1,m xi xj
m k+1
+ pk+1,m , with ck+1,m := r1 c1,m ∈ R left
invertible, pk+1,m := r1 p1,m xj + p1,ck,m xm
k k
i xj + xj pk,m ∈ A, pk+1,m = 0 or
deg(pk+1,m ) ≤ k + m < k + 1 + m. This complete the step 1.
Step 3.2. The proof is by induction on the number of variables involved in xα
or xβ . If xα and xβ include only one variable, then we apply the step 1. So by
induction we assume that (1.6) is true when the number of variables of xα or
xβ is ≤ n − 1.
Then,
αn β1 αn β1 β2
xα xβ =xα βn α1 βn
1 · · · xn x1 · · · xn = (x1 · · · xn x1 )x2 · · · xn
1

=(c1 xα
1
1 +β1 α2
x2 · · · xαn β2 βn
n + p1 )(x2 · · · xn ),
c1 ∈ R left invertible,
p1 ∈ A, p1 = 0 or deg(p1 ) < α1 + · · · + αn + β1 ≤ |α + β|
=c1 (x1α1 +β1 xα αn β2 βn β2 βn
2 · · · xn )(x2 · · · xn ) + p1 x2 · · · xn
2

=c1 (c2 xα
1
1 +β1 α2 +β2
x2 · · · xnαn +βn + p2 ) + p1 xβ2 2 · · · xβnn ,
c2 ∈ R left invertible, p2 ∈ A, p2 = 0 or
deg(p2 ) < α1 + β1 + α2 + · · · + αn + β2 + · · · + βn = |α + β|
=cα,β xα+β + pα,β ,

with cα,β := c1 c2 ∈ R, left invertible, pα,β := c1 p2 + p1 xβ2 2 · · · xβnn ∈ A, pα,β = 0


or deg(pα,β ) < |α + β|.
Step 3.3. Since M on(A) is R-basis for A, then cα,r and pα,r are unique.
⇐) The condition (iii) of the Definition 1 is a particular case of (1.5), and the
condition (iv) is a particular case of (1.6), thus (a) and (b) implies that A is a
σ − P BW extension.
Remark 8. (i) A left inverse of cα,β will be denoted by c0α,β . We observe that
if α = 0 or β = 0, then cα,β = 1 and hence c0α,β = 1.
(ii) Let θ, γ, β ∈ Nn and c ∈ R, then we it is easy to check the following identities:
σ θ (cγ,β )cθ,γ+β = cθ,γ cθ+γ,β ,
σ θ (σ γ (c))cθ,γ = cθ,γ σ θ+γ (c).
Gröbner Bases for Ideals of σ − P BW Extensions 9

(iii) We observe if A is a σ − P BW extension quasi-commutative, then from the


proof of Theorem 7 we conclude that pα,r = 0 and pα,β = 0, for every 0 6= r ∈ R
and every α, β ∈ Nn .
(iv) From the proof of Theorem 7 we get also that if A is a bijective σ − P BW
extension, then cα,β is invertible for any α, β ∈ Nn .
We conclude this section with a natural consequence of Theorem 7.
Corollary 9. If R has no zero divisors, then A = σ(R) < x1 , . . . , xn > has no
zero divisors.

2 Hilbert basis theorem for σ − P BW extensions


It is well known ([8]) that if R is a left Noetherian ring, then the P BW extension
A = R < x1 , . . . , xn > is also a left Noetherian ring (see also [6]). We will extend
this result to σ − P BW extensions.
Theorem 10. If R is a left Noetherian ring then A = σ(R) < x1 , . . . , xn > is
also a left Noetherian ring.
Proof. The proof is based in the following central fact: if A is a filtered ring
and its corresponding graded ring G(A) is left Noetherian, then A is also a left
Noetherian ring (see [3] and [9]).
We note that A is a Z-filtered ring,

0,
 if m ≤ −1,
Fm := R, if m = 0,

{f ∈ A| deg(f ) ≤ m}, if m ≥ 1.

Associated to this filtration we define the graded ring G(A) by


M
G(A) := Fm /Fm−1
m≥0

with product given by

Fn /Fn−1 × Fm /Fm−1 → Fn+m /Fn+m−1


(a + Fn−1 , b + Fm−1 ) 7→ ab + Fn+m−1 .

We will prove next that

G(A) ∼
= R[x1 , . . . , xn ]/I0 ,

where R[x1 , . . . , xn ] is the habitual polynomial ring and I0 is the two-side ideal
hxi r − ci,r xi , xj xi − ci,j xi xj | r ∈ R, 1 ≤ i, j ≤ ni (we will use the same notation
for the variables of A and R[x1 , . . . , xn ]). Let a ∈ G(A), then a = ⊕m≥0 am ,
where am = am + Fm−1 , am ∈ Fm and am 6= 0 only for a finite subset of
integers m; without loss of generality we can assume that if am 6= 0 then am is
10 Claudia Gallego & Oswaldo Lezama

an homogeneous polynomial of degree m presented in the standard form, i.e.,


am = α∈Nn cα xα , where |α| = m for such α with cα 6= 0. Then, we define
P

φ : G(A) → R[x1 , . . . , xn ]/I0


X\
⊕m≥0 am 7→ am ,
am 6=0

where fb := f + I0 , for f ∈ R[x1 , . . . , xn ]. It is easy to check that φ is a well


defined surjective homomorphism of abelian groups. Moreover, φ is injective.
In fact, let a = ⊕am ∈ G(A) such that φ(a) = b 0, then we can represent a
as a = ai1 + ai2 + · · · + ais , with aij ∈ Fij \ Fij −1 for 1 ≤ j ≤ s. Thus,
Ps
0 = φ(a) = (ai1 + ai2 + · · · + ais ) + I0 , and hence
b
j=1 aij ∈ I0 . Since
each aij is homogeneous of degree ij and I0 is an homogeneous ideal (gen-
erated by homogeneous polynomials), aij ∈ I0 for 1 ≤ j ≤ s, so there exist
f1j , hj1 , . . . , fkjj , hjkj ∈ R[x1 , . . . , xn ] \ {0} and q1j , . . . , qkj j ∈ {xi r − ci,r xi , xj xi −
ci,j xi xj | r ∈ R, 1 ≤ i, j ≤ n} such that aij = f1j q1j hj1 + · · · + fkjj qkj j hjkj . On
the other hand, given f ∈ R[x1 , . . . , xn ] \ {0}, there exists m ∈ N such that
f ∈ Fm \ Fm−1 , so f [xj xi − ci,j xi xj ] ∈ Fm+2 , but f [xj xi − ci,j xi xj ] = f q, where
q ∈ R + Rx1 + · · · + Rxn , hence f [xj xi − ci,j xi xj ] = f q = 0 ∈ Fm+2 /Fm+1 . In
a similar way we have f [xi r − ci,r xi ] ∈ Fm+1 , but xi r − ci,r xi = δi (r) ∈ R, then
f [xi r − ci,r xi ] = f δi (r) = 0 ∈ Fm+1 /Fm . From this we conclude that aij = 0
for 1 ≤ j ≤ s, and hence a = 0. φ is multiplicative: following the ideas in the
proof of Theorem 7 we get that

xα r − σ α (r)xα ∈ I0 for all α ∈ Nn and r ∈ R, (2.1)


α β α+β n
x x − cα,β x ∈ I0 for all α, β ∈ N . (2.2)

Let aP= ⊕m≥0 am and b = ⊕m≥0 bm in G(A), then ab = ⊕k≥0 ck , where


ck = i+j=k ai bj = a0 bk + a1 bk−1 + · · · + ak b0 is in Fk /Fk−1 and ai , bj are
homogeneous polynomials of degree i and j respectively. Since φ is additive
we only need to establish that φ(ab) = φ(a)φ(b), with a and b homogeneous
polynomials of degree l and m respectively. Let

a = cα1 xα1 + cα2 xα2 + · · · + cαr xαr


b = dβ1 xβ1 + dβ2 xβ2 + · · · + dβs xβs

with |αi | = l y |βj | = m for 1 ≤ i ≤ r and 1 ≤ j ≤ s. Then, ab ∈ Fl+m and

ab =cα1 σ α1 (dβ1 )cα1 ,β1 xα1 +β1 + cα1 σ α1 (dβ2 )cα1 ,β2 xα1 +β2 + · · · +
cα1 σ α1 (dβs )cα1 ,βs xα1 +βs + · · · + cαr σ αr (dβ1 )cαr ,β1 xαr +β1 +
cαr σ αr (bβ2 )cαr ,β2 xαr +β2 + · · · + cαr σ αr (bβs )cαr ,βs xαr +βs + q,
Gröbner Bases for Ideals of σ − P BW Extensions 11

where q ∈ A with q = 0 or deg(q) < l + m. From (2.1) and (2.2) we obtain

φ(ab) =(cα1 xα1 dβ1 xβ1 + I0 ) + · · · + (cα1 xα1 dβs xβs + I0 ) + · · · +


(cαr xαr dβ1 xβ1 + I0 ) + · · · + (cαr xαr bβs xβs + I0 )
=(cα1 xα1 + · · · + cαr xαr + I0 )(dβ1 xβ1 + · · · + dβs xβs + I0 )
=φ(a)φ(b).

Finally, we observe that φ(1) = b


1, and the proof is over.

3 Monomial orders in σ − P BW extensions


Let A = σ(R) < x1 , . . . , xn > be a σ − P BW extension of R and let  be a
total order defined on M on(A). If xα  xβ but xα 6= xβ we will write xα  xβ .
Let f 6= 0 be a polynomial of A, if
f = c1 X1 + · · · + ct Xt ,
with ci ∈ R − {0} and X1  · · ·  Xt are the monomials of f , then lm(f ) := X1
is the leading monomial of f , lc(f ) := c1 is the leading coefficient of f and
lt(f ) := c1 X1 is the leading term of f . If f = 0, we define lm(0) := 0, lc(0) :=
0, lt(0) := 0, and we set X  0 for any X ∈ M on(A). Thus, we extend  to
M on(A) ∪ {0}.
Definition 11. Let  be a total order on M on(A), we say that  is a monomial
order on M on(A) if the following conditions hold:
(i) For every xβ , xα , xγ , xλ ∈ M on(A)

xβ  xα ⇒ lm(xγ xβ xλ )  lm(xγ xα xλ ).

(ii) xα  1, for every xα ∈ M on(A).


(iii)  is degree compatible, i.e., |β| ≥ |α| ⇒ xβ  xα .
Monomial orders are also called admissible orders. The condition (iii) of the
previous definition is needed in the proof of the following proposition, and this
one will be used in the Division Algorithm (Theorem 21).
Proposition 12. Every monomial order on M on(A) is a well order. Thus,
there are not infinite decreasing chains in M on(A).
Proof. Suppose we have a monomial order  on M on(A) that is not a well
order. This means that we have an infinite sequence of monomials
X1  X2  X3  · · ·
and since  is degree compatible, then we have the infinite sequence
deg(X1 ) > deg(X2 ) > deg(X3 ) > · · · ,
12 Claudia Gallego & Oswaldo Lezama

but this is impossible since deg(X1 ) is finite.


From now on we will assume that M on(A) is endowed with some monomial
order.
Definition 13. Let xα , xβ ∈ M on(A), we say that xα divides xβ , denoted by
xα |xβ , if there exists xγ , xλ ∈ M on(A) such that xβ = lm(xγ xα xλ ).
Proposition 14. Let xα , xβ ∈ M on(A) and f, g ∈ A − {0}. Then,
(a) lm(xα g) = lm(xα lm(g)) = xα+exp(lm(g)) . In particular,

lm(lm(f )lm(g)) = xexp(lm(f ))+exp(lm(g))

and
lm(xα xβ ) = xα+β . (3.1)

(b) The following conditions are equivalent:


(i) xα |xβ .
(ii) There exists a unique xθ ∈ M on(A) such that xβ = lm(xθ xα ) = xθ+α
and hence β = θ + α.
(iii) There exists a unique xθ ∈ M on(A) such that xβ = lm(xα xθ ) = xα+θ
and hence β = α + θ.
(iv) βi ≥ αi for 1 ≤ i ≤ n.
Proof. (a) Let g = cxβ + p, with 0 6= c ∈ R, p = 0 or deg(p) < |β|. Then,
xα g = xα (cxβ + p)= (xα c)xβ + xα p = (cα xα + q)xβ + xα p, with 0 6= cα ∈ R,
q ∈ A, q = 0 or deg(q) < |α|. So, xα g = cα xα xβ + qxβ + xα p= cα (cα,β xα+β +
t) + qxβ + xα p, with cα,β ∈ R left invertible, t ∈ A, t = 0 or deg(t) < |α + β|.
Thus, xα g = c0 xα+β +p0 , with c0 := cα cα,β ∈ R −{0}, p0 := cα t+qxβ +xα p ∈ A,
p0 = 0 or deg(p0 ) < |α + β|. This implies that lm(xα g) = xα+β , but xα lm(g) =
xα xβ = cα,β xα+β + t, so lm(xα lm(g)) = xα+β .
(b) This part is a direct consequence of (3.1).
Remark 15. We note that a least common multiple of monomials of M on(A)
there exists: in fact, let xα , xβ ∈ M on(A), then lcm(xα , xβ ) = xγ ∈ M on(A),
where γ = (γ1 , . . . , γn ) with γi := max{αi , βi } for each 1 ≤ i ≤ n.

4 Reduction in σ − P BW extensions
Some natural computational conditions on R will be assumed in the rest of this
paper (see [7]).
Definition 16. A ring R is left Gröbner soluble (LGS) if the following condi-
tions hold:
(i) R is left Noetherian.
Gröbner Bases for Ideals of σ − P BW Extensions 13

(ii) Given a, r1 , . . . , rm ∈ R there exists an algorithm which decides whether a


is in the left ideal Rx1 + · · · + Rxm , and if so, finds b1 , . . . , bm ∈ R such
that a = b1 r1 + · · · + bm rm .

(iii) Given r1 , . . . , rm ∈ R there exists an algorithm which finds a finite set of


generators of the left R-module

SyzR [r1 · · · rm ] := {(b1 , . . . , bm ) ∈ Rm |b1 r1 + · · · + bm rm = 0}.

Remark 17. The three above conditions imposed to R are needed in order to
guarantee a Gröbner theory in the rings of coefficients, in particular, to have
an effective solution of the membership problem in R (see (ii) in the Definition
18 below). From now on we will assume that A = σ(R) < x1 , . . . , xn > is a
σ − P BW extension of R, where R is a LGS ring and M on(A) is endowed with
some monomial order.

Definition 18. Let F be a finite set of non-zero elements of A, and let f, h ∈ A,


F
we say that f reduces to h by F in one step, denoted f −−→ h, if there exist
elements f1 , . . . , ft ∈ F and r1 , . . . , rt ∈ R such that

(i) lm(fi )|lm(f ), 1 ≤ i ≤ t, i.e., there exists xαi ∈ M on(A) such that
lm(f ) = lm(xαi lm(fi )) = xαi +exp(lm(fi )) .

(ii) lc(f ) = r1 σ α1 (lc(f1 ))cα1 ,f1 + · · · + rt σ αt (lc(ft ))cαt ,ft , where cαi ,fi are de-
fined as in the Theorem 7, i.e., cαi ,fi := cαi ,exp(lm(fi )) .
Pt
(iii) h = f − i=1 ri xαi fi .
F
We say that f reduces to h by F , denoted f −−→+ h, if there exist h1 , . . . , ht−1 ∈
A such that
F F F F F
f −−−−→ h1 −−−−→ h2 −−−−→ · · · −−−−→ ht−1 −−−−→ h.
f is reduced (also called minimal) w.r.t F if f = 0 or there is no one step
reduction of f by F . Otherwise, we will say that f is reducible w.r.t F . If
F
f −−→+ h and h is reduced w.r.t F , then we say that h is a remainder for f
w.r.t F .

Remark 19. (i) By Theorem 7, the coefficients cαi ,fi in the previous definition
are unique and satisfy

xαi lm(fi ) = cαi ,fi xαi +exp(lm(fi )) + pαi ,fi ,

where pαi ,fi = 0 or deg(pαi ,fi ) < |αi + exp(lm(fi ))|, 1 ≤ i ≤ t.


(ii) lm(f )  lm(h) and f − h ∈ hF i, where hF i is the left ideal of A generated
by F .
(iii) The remainder of f is not unique.
F
(iv) By definition we will assume that 0 −
→ 0.
14 Claudia Gallego & Oswaldo Lezama

Using the reduction relation and the identities in Remark 8 (ii), it is easy to
prove the following interesting properties.
Proposition 20. Let A be a σ-P BW extension such that cα,β is invertible for
each α, β ∈ Nn . Let f, h ∈ A, θ ∈ Nn and F = {f1 , . . . , ft } be a finite set of
non-zero polynomials of A. Then,
F
(i) If f −−→ h, then there exists p ∈ A with p = 0 or lm(xθ f )  lm(p)
F
such that xθ f + p −−→ xθ h. In particular, if A is quasi-commutative, then
p = 0.
F
(ii) If f −−→+ h and p ∈ A is such that p = 0 or lm(h)  lm(p), then
F
f + p −−→+ h + p.
F
(iii) If f −−→+ h, then there exists p ∈ A with p = 0 or lm(xθ f )  lm(p) such
F
that xθ f + p −−→+ xθ h. If A is quasi-commutative, then p = 0.
F
(iv) If f −−→+ 0, then there exists p ∈ A with p = 0 or lm(xθ f )  lm(p) such
F
that xθ f + p −−→+ 0. If A is quasi-commutative, then p = 0.
Theorem 21. Let F = {f1 , . . . , ft } be a finite set of non-zero polynomials of A
and f ∈ A, then the Division Algorithm below produces polynomials q1 , . . . , qt , h
F
∈ A, with h reduced w.r.t F , such that f −−→+ h and
f = q1 f1 + · · · + qt ft + h,
with
lm(f ) = max{lm(lm(q1 )lm(f1 )), . . . , lm(lm(qt )lm(ft )), lm(h)}.
Gröbner Bases for Ideals of σ − P BW Extensions 15

Division Algorithm in A = σ(R) < x1 , . . . , xn >

INPUT: f, f1 , . . . , ft ∈ A with fj 6= 0 (1 ≤ j ≤ t)
OUTPUT: q1 , . . . , qt , h ∈ A with f = q1 f1 + · · · + qt ft + h and h is
reduced w.r.t {f1 , . . . , ft } and
lm(f ) = max{lm(lm(q1 )lm(f1 )), . . . , lm(lm(qt )lm(ft )), lm(h)}

INITIALIZATION: q1 := 0, q2 := 0, . . . , qt := 0, h := f
WHILE h 6= 0 and there exists j such that lm(fj ) divides lm(h) DO
Calculate J := {j | lm(fj ) divides lm(h)}
FOR j ∈ J DO
Calculate αj ∈ Nn such that αj +exp(lm(fj )) =
exp(lm(h))
IF the equation lc(h) = j∈J rj σ αj (lc(fj ))cαj ,fj is solu-
P
ble, where cαj ,fj are defined as in the Theorem 7 THEN
Calculate one solution (rj )j∈J
h := h − j∈J rj xαj fj
P

FOR j ∈ J DO
qj := qj + rj xαj
ELSE
Stop

Proof. We first note that the Division Algorithm is the iteration of the reduction
process. If f is reduced with respect to F := {f1 , . . . , ft }, then h = f, q1 =
· · · = qt = 0 and lm(f ) = lm(h). If f is not reduced, i.e., then we make the
F
first reduction, f −−→ h1 , with f = j∈J1 rj1 xαj fj + h1 , with J1 := {j | lm(fj )
P
divides lm(f )} and rj1 ∈ R. If h1 is reduced with respect to F , then the cycle
WHILE ends and we have that qj = rj1 xαj for j ∈ J1 and qj = 0 for j ∈ / J1 .
Moreover, lm(f )  lm(h1 ) and lm(f ) = lm(lm(qj )lm(fj )) for j ∈ J1 such that
rj1 6= 0, hence, lm(f ) = max1≤j≤t {lm(lm(qj )lm(fj )), lm(h1 )}. If h1 is not
F
P so we make the second reduction with respect to F , h1 −−→ h2 , with
reduced,
h1 = j∈J2 rj2 xαj fj + h2 , J2 := {j | lm(fj ) divides lm(h1 )} and rj2 ∈ R. We
have
f = j∈J1 rj1 xαj fj + j∈J2 rj2 xαj fj + h2
P P

If h2 is reduced with respect to F the procedure ends and we get that qj = qj for
j∈ / J2 and qj = qj +rj2 xαj for j ∈ J2 . We know that lm(f )  lm(h1 )  lm(h2 ),
this implies that the algorithm produces polynomials qj with monomials or-
dered according to the monomial order fixed, and again we have lm(f ) =
16 Claudia Gallego & Oswaldo Lezama

max1≤j≤t {lm(lm(qj )lm(fj )), lm(h2 )}. We can continue this way and the al-
gorithm ends since M on(A) is well ordered (Proposition 12).

The following example illustrates the above procedure.

Example 22. In A2 (2, 2), with K := Q, we consider the order deglex with
y1  y2 ; let f1 := x21 x2 y1 y2 , f2 := x2 y1 , f3 := x1 y2 and f := x1 x22 y12 y2 + x21 x2 y1 .
For f we will find q1 , q2 , q3 and h, following the previous algorithm. We will use
the relations of Example 5 (iv); in particular, since yi yj = yj yi , then cα,β = 1 for
every α, β ∈ M on(A2 (2, 2)). For j = 1, 2, 3, we will note αj := (αj1 , αj2 ) ∈ N2 .
Step 1. We start with h := f , q1 := 0, q2 := 0, q3 := 0; since lm(fj )|lm(h) for
j = 1, 2, 3, we compute αj such that αj + exp(lm(fj )) = exp(lm(h)), and also,
σ αj (lc(fj )):

(α11 , α12 ) + (1, 1) = (2, 1) ⇒ α11 = 1, α12 = 0,


σ α1 (lc(f1 )) = σ1 σ20 (lc(f1 )) = σ1 (x21 x2 ) = 4x21 x2 ,
(α21 , α22 ) + (1, 0) = (2, 1) ⇒ α21 = 1, α22 = 1,
σ α2 (lc(f2 )) = σ1 σ2 (x2 ) = 2x2 ,
(α31 , α32 ) + (0, 1) = (2, 1) ⇒ α31 = 2, α32 = 0,
σ α3 (lc(f3 )) = σ12 σ20 (x1 ) = 4x1 .

Now we solve the equation

lc(h) = x1 x22 = r1 4x21 x2 + r2 2x2 + r3 4x1 ⇒ r1 = 1, r2 = 12 x1 x2 , r3 = −x1 x2 ,

and with the relations defining A2 (2, 2) we compute

h =h − (r1 y α1 f1 + r2 y α2 f2 + r3 y α3 f3 )
1
=h − y1 f1 − x1 x2 y1 y2 f2 + x1 x2 y12 f3
2
1
=h − y1 x1 x2 y1 y2 − x1 x2 y1 y2 x2 y1 + x1 x2 y12 x1 y2
2
2
1 2 2
= − x1 x2 y1 + x1 x2 y1 .
2
We compute also

q1 = y1 , q2 = 12 x1 x2 y1 y2 , q3 = −x1 x2 y12 .

Step 2. lm(h) = y12 , lc(h) = − 21 x1 x2 ; since lm(f2 )|lm(h), we compute α2 such


that α2 + exp(lm(f2 )) = exp(lm(h)), and also, σ α2 (lc(f2 )):

(α21 , α22 ) + (1, 0) = (2, 0) ⇒ α21 = 1, α22 = 0,


σ α2 (lc(f2 )) = σ1 σ20 (x2 ) = x2 ,
Gröbner Bases for Ideals of σ − P BW Extensions 17

and from this we get


1 1
− x1 x2 = r2 x2 ⇒ r2 = − x1 ,
2 2
1 1
h = h − r2 y f2 = h + x1 y1 f2 = h + x1 y1 x2 y1 = x21 x2 y1 ,
α2
2 2
1 1
q1 = y1 , q2 = x1 x2 y1 y2 − x1 y1 , q3 = −x1 x2 y12 .
2 2
Step 3. lm(h) = y1 , lc(h) = x21 x2 ; in this case we have

(α21 , α22 ) + (1, 0) = (1, 0) ⇒ α21 = 0, α22 = 0,


σ α2 (lc(f2 )) = σ10 σ20 (x2 ) = x2 ,
x21 x2 = r2 x2 ⇒ r2 = x21 ,
h = h − r2 y α2 f2 = h − x21 f2 = h − x21 x2 y1 = 0,
1 1
q1 = y1 , q2 = x1 x2 y1 y2 − x1 y1 + x21 , q3 = −x1 x2 y12 .
2 2
We can check that f = q1 f1 + q2 f2 + q3 f3 , i.e.,

f = y1 f1 + ( 12 x1 x2 y1 y2 − 12 x1 y1 + x21 )f2 + (−x1 x2 y12 )f3 ;

we observe that

max{lm(lm(q1 )lm(f1 )), lm(lm(q2 )lm(f2 )), lm(lm(q3 )lm(f3 ))}


= max{y12 y2 , y12 y2 , y12 y2 } = y12 y2 = lm(f ).

5 Gröbner bases for σ − P BW extensions


Our next purpose is to define Gröbner bases for the left ideals of A and to prove
a theorem of characterization adapted from Theorem 3.3.4 in [10].

Definition 23. Let I 6= 0 be a left ideal of A and let G be a non empty finite
subset of non-zero polynomials of I, we say that G is a Gröbner basis for I if
each element 0 6= f ∈ I is reducible w.r.t G.

We will say that {0} is a Gröbner basis for I = 0.

Theorem 24. Let I 6= 0 be a left ideal of A and let G be a finite subset of


non-zero polynomials of I. Then the following conditions are equivalent:

(i) G is a Gröbner basis for I.

(ii) For any polynomial f ∈ A,

G
f ∈ I if and only if f −−→+ 0.
18 Claudia Gallego & Oswaldo Lezama

(iii) For any 0 6= f ∈ I there exist g1 , . . . , gt ∈ G such that lm(gj )|lm(f ), 1 ≤


j ≤ t, (i.e., there exist αj ∈ Nn such that αj + exp(lm(gj )) = exp(lm(f )))
and

lc(f ) ∈ hσ α1 (lc(g1 ))cα1 ,g1 , . . . , σ αt (lc(gt ))cαt ,gt i.

(iv) For α ∈ Nn , let hα, Ii be the left ideal of R defined by

hα, Ii := hlc(f )|f ∈ I, lm(f ) = xα i.

Then, hα, Ii = J, with

J := hσ β (lc(g))cβ,g |g ∈ G, with lm(xβ lm(g)) = xα i.

G
Proof. (i)⇒ (ii): Let f ∈ I, if f = 0, then by definition f −−→+ 0. If f 6= 0,
G
then there exists h1 ∈ A such that f −−→ h1 , with lm(f )  lm(h1 ) and f − h1 ∈
hGi ⊆ I, hence h1 ∈ I; if h1 = 0, so we end. If h1 6= 0, then we can repeat this
G
reasoning for h1 , and since M on(A) is well ordered, we get that f −−→+ 0.
G
Conversely, if f −−→+ 0, then by the Theorem 21, there exist g1 , . . . , gt ∈ G and
q1 , . . . , qt ∈ A such that f = q1 g1 + · · · + qt gt , i.e., f ∈ I.
(ii)⇒ (i): evident.
(i)⇔ (iii): this is a direct consequence of the Definition 18.
(iii)⇒ (iv) Since R is a left Noetherian ring, there exist r1 , . . . , rs ∈ R, f1 , . . . , fl
∈ I such that hα, Ii = hr1 , . . . , rs i, lm(fi ) = xα , 1 ≤ i ≤ l, with hr1 , . . . , rs i ⊆
hlc(f1 ), . . . , lc(fl )i, so hlc(f1 ), . . . , lc(fl )i = hα, Ii. Let r ∈ hα, Ii, there exist
a1 , . . . , al ∈ R such that r = a1 lc(f1 )+· · ·+al lc(fl ); by (iii), for each i there exist
g1i , . . . , gti i ∈ G and bji ∈ R such that lc(fi ) = b1i σ α1i (lc(g1i ))cα1i ,g1i + · · · +
bti i σ αti i (lc(gti i ))cαti i ,gti i , with xα = lm(fi ) = lm(xαji lm(gji )), thus hα, Ii ⊆ J.
Conversely, if r ∈ J, then r = b1 σ β1 (lc(g1 ))cβ1 ,g1 + · · · + bt σ βt (lc(gt ))cβt ,gt , with
bi ∈ R, βi ∈ Nn , gi ∈ G such that βi + exp(lm(gi )) = α for any 1 ≤ i ≤ t; note
that xβi gi ∈ I, lm(xβi gi ) = xα , lc(xβi gi ) = σ βi (lc(gi ))cβi ,gi , for 1 ≤ i ≤ t, and
r = b1 lc(xβ1 g1 ) + · · · + bt lc(xβt gt ), i.e., r ∈ hα, Ii.
(iv)⇒ (iii): let 0 6= f ∈ I and let α = exp(lm(f )), then lc(f ) ∈ hα, Ii; by
(iv) lc(f ) = b1 σ β1 (lc(g1 ))cβ1 ,g1 + · · · + bt σ βt (lc(gt ))cβt ,gt , with bi ∈ R, βi ∈ Nn ,
gi ∈ G such that βi + exp(lm(gi )) = α for any 1 ≤ i ≤ t. From this we conclude
that lc(f ) ∈ hσ β1 (lc(g1 ))cβ1 ,g1 , . . . , σ βt (lc(gt ))cβt ,gt i.

From this theorem we get the following direct standard consequences.

Corollary 25. Let I 6= 0 be a left ideal of A. Then,

(i) If G is a Gröbner basis for I, then I = hGi.


G
(ii) Let G be a Gröbner basis for I, if f ∈ I and f −−→+ h, with h reduced,
then h = 0.
Gröbner Bases for Ideals of σ − P BW Extensions 19

6 Constructing Gröbner bases: part I


If A is a quasi-commutative bijective σ − P BW extension, then we will prove
in the present section that every left ideal I of A has a Gröbner basis, and also,
we will construct the Buchberger’s algorithm for computing such bases.

We start fixing some notation and proving a preliminary general result for ar-
bitrary σ − P BW extensions.
Definition 26. Let F := {g1 , . . . , gs } ⊆ A, XF the least common multiple
of {lm(g1 ), . . . , lm(gs )}, θ ∈ Nn , βi := exp(lm(gi )) and γi ∈ Nn such that
γi + βi = exp(XF ), 1 ≤ i ≤ s. BF,θ will denote a finite set of generators of
SF,θ := SyzR [σ γ1 +θ (lc(g1 ))cγ1 +θ,β1 · · · σ γs +θ (lc(gs ))cγs +θ,βs )].
For θ = 0 := (0, . . . , 0), SF,θ will be denoted by SF and BF,θ by BF .
Theorem 27. Let I 6= 0 be a left ideal of A and let G be a finite subset of
non-zero generators of I. Then the following conditions are equivalent:
(i) G is a Gröbner basis of I.
(ii) For all F := {g1 , . . . , gs } ⊆ G, θ ∈ Nn and (b1 , . . . , bs ) ∈ BF,θ ,
Ps γi +θ G
i=1 bi x gi −−→+ 0.

In particular, if G is a Gröbner basis of I then for all F := {g1 , . . . , gs } ⊆ G


and (b1 , . . . , bs ) ∈ BF ,
Ps γi G
i=1 bi x gi −−→+ 0.
Ps
Proof. (i) ⇒ (ii): We observe that f := i=1 bi xγi +θ gi ∈ I, so by the Theorem
G
24 f −−→+ 0.
(ii) ⇒ (i): The proof of this part is an adaptation of the proof of Theorem
3.3.4 in [10]. Let 0 6= f ∈ I, we will prove that the condition (iii) of Theorem 24
holds. Let G := {g1 , . . . , gt }, then there exist f1 , . . . , ft ∈ A such that f = f1 g1 +
· · · + ft gt , we can choose {fi }ti=1 such that xα0 := max{lm(lm(fi )lm(gi ))}ti=1
is minimal. Let lm(fi ) := xαi and lm(gi ) = xβi , 1 ≤ i ≤ t; let F := {gi ∈
G|lm(lm(fi )lm(gi )) = xα0 }, renumbering the elements of G we can assume
that F = {g1 , . . . , gs }. We will consider two possible cases:
Case 1 : lm(f ) = xα0 . Then lm(gi )|lm(f ) for 1 ≤ i ≤ s and
lc(f ) = lc(f1 )σ α1 (lc(g1 ))cα1 ,β1 + · · · + lc(fs )σ αs (lc(gs ))cαs ,βs ,
i.e., the condition (iii) of Theorem 24 holds.
Case 2 : xα0  lm(f ). We will prove that this produces a contradiction. For
every 1 ≤ i ≤ s, xα0 = lm(lm(fi )lm(gi )), then lm(gi )|xα0 and hence XF |xα0 ,
so there exists θ ∈ Nn such that α0 = exp(XF ) + θ. On the other hand,
γi + βi = exp(XF ) and αi + βi = α0 , so αi = γi + θ for every 1 ≤ i ≤ s. We
have
20 Claudia Gallego & Oswaldo Lezama

lc(f1 )σ γ1 +θ (lc(g1 ))cγ1 +θ,β1 + · · · + lc(fs )σ γs +θ (lc(gs ))cγs +θ,βs = 0;


let
BF,θ := {b 1 , . . . , b k } := {(b11 , . . . , b1s ), . . . , (bk1 , . . . , bks )}
be a set of generators of S := SF,θ ; since (lc(f1 ), . . . , lc(fs )) ∈ S, there exist
r1 , . . . , rk ∈ R such that
(lc(f1 ), . . . , lc(fs )) = (r1 b11 + · · · + rk bk1 , . . . , r1 b1s + · · · + rk bks ).
Hence for 1 ≤ i ≤ s we have lc(fi ) = r1 b1i + · · · + rk bki and
t
X s
X t
X
f= fi gi = fi gi + fi gi
i=1 i=1 i=s+1
s
X t
X
= (fi − lc(fi )lm(fi ) + lc(fi )lm(fi ))gi + fi gi
i=1 i=s+1
s
X k
X s X
X k t
X
= (fi − rj bji lm(fi ))gi + rj bji lm(fi )gi + fi gi .
i=1 j=1 i=1 j=1 i=s+1

We consider the three parts of the previous sum: for the last part and by the
definition of F , xα0  max{lm(lm(fi )lm(gi ))}ti=s+1 ; for the first part we note
Pk
that xα0  lm(lm(fi − j=1 rj bji lm(fi ))lm(gi )) for every 1 ≤ i ≤ s. Now we
will consider the second part:
Ps Pk Pk Ps αi
i=1 j=1 rj bji lm(fi )gi = j=1 rj ( i=1 bji x gi ) =
Pk Ps γi +θ
j=1 rj ( i=1 bji x gi );
Ps G
from the hypothesis, i=1 bji xγi +θ gi −−→+ 0, and by Theorem 21, there exist
Ps Pt Ps
q1 , . . . , qt ∈ A such that i=1 bji xαi gi = i=1 qi gi , with lm( i=1 αi
Psbji x gαii) =
t α0
max{lm(lm(qi )lm(gi ))}i=1 , but (bj1 , . . . , bjs ) ∈ S, so x  lm( i=1 bji x gi )
and hence xα0  lm(lm(qi )lm(gi )) for every 1 ≤ i ≤ t. Thus, considering the
three parts of f , we get a contradiction with the way we choose f1 , . . . , ft .
Taking θ = 0 we obtain the particular case.
Remark 28. For P BW extensions, the proof of (i)⇔(ii) in the Theorem 27
involves the proof of Theorem 3.3.4 in [10], since in that case σi = idR for all i
and cα,β = 1 for all α, β ∈ Nn .
With the Theorem 27 we prove next the main result of this section.
Theorem 29. Let A be a quasi-commutative bijective σ − P BW extension. Let
I 6= 0 be a left ideal of A and let G be a finite subset of non-zero generators of
I. Then the following conditions are equivalent:
(i) G is a Gröbner basis of I.
(ii) For all F := {g1 , . . . , gs } ⊆ G and (b1 , . . . , bs ) ∈ BF ,
Gröbner Bases for Ideals of σ − P BW Extensions 21

Ps γi G
i=1 bi x gi −−→+ 0.

Proof. (i)⇒(ii) is valid for any σ − P BW extension (Theorem 27).


(ii)⇒(i) We divide the proof of this part in two steps.
Step 1. Let θ ∈ Nn and F := {g1 , . . . , gs } ⊆ G; we define a new R-module
structure over Rs by

(b1 , . . . , bs ) + (c1 , . . . , cs ) := (b1 + c1 , . . . , bs + cs ),


r ? (b1 , . . . , bs ) := σ θ (r)(b1 , . . . , bs ) = (σ θ (r)b1 , . . . , σ θ (r)bs ).

It is easy to prove that the above operations convert Rs into a R-module denoted
by (Rs )? ; let
?
Ps
SF,θ := {(b1 , . . . , bs ) ∈ (Rs )? | i=1 bi σ θ+γi (lc(gi ))cθ+γi ,βi = 0},
?
it is clear that SF,θ is a R-submodule of (Rs )? . Note that (Rs )? is finitely
generated: (b1 , . . . , bs ) = b1 e 1 + · · · + bs e s = σ θ (b01 )e 1 + · · · + σ θ (b0s )e s (since σ θ
is surjective), so (b1 , . . . , bs ) = b01 ? e 1 + · · · + b0s ? e s . Since R is a left Noetherian
ring, then (Rs )? is a left Noetherian R-module, and consequently, SF,θ ?
is finitely
generated. We observe that if θ = 0, then the R-module structure of (Rs )? is
the habitual, i.e., (Rs )? = Rs , and also SF? = SF .
The function defined by
? α
SF,θ −
→ SF
b := (b1 , . . . , bs ) 7→ b 0 := (b01 , . . . , b0s )

where b0i ∈ R is such that bi c−1 θ 0


θ,γi = σ (bi ), 1 ≤ i ≤P s, is a R-isomorphism. In
0 s θ+γi
fact, we first need Ps to observe that b ∈ S F : from i=1 bi σ (lc(gi ))cθ+γi ,βi
θ 0 θ+γi −1
= 0 we write i=1 σ (bi )cθ,γi σ (lc(gi ))cθ,γi cθ,γi cθ+γi ,βi = 0; multiplying by
c−1 −1 −1
θ,exp(XF ) and replacing cθ,exp(XF ) by cθ,γi +βi (since γi +βi = exp(XF ) for every
Ps
1 ≤ i ≤ s) we get i=1 σ θ (b0i )cθ,γi σ θ+γi (lc(gi ))c−1 c i cθ+γi ,βi c−1 θ,γi +βi = 0; we
Ps θ,γθi θ,γ 0 θ γi θ
can use the identities of the Remark Ps 8, so i=1 σ (b i )σ (σ (lc(g i )))σ (cγi ,βi )
θ 0 γi 0
= 0, and since σ is injective then i=1 bi σ (lc(gi ))cγi ,βi = 0, i.e., b ∈ SF . Of
course α is additive and injective; for r ∈ R, α(r?b) = α(σ θ (r)b1 , . . . , σ θ (r)bs ) =
(rb01 , . . . , rb0s ) since σ θ (r)bi c−1 θ 0
γi ,θ = σ (rbi ) for all i. Then, α(r ?b) = rα(b). Only
rest to prove that α is surjective, for this we can P repeat the above ideas but in the
s
reverseP order: let b 0 = (b01 , . . . , b0s ) ∈ SF , then i=1Pb0i σ γi (lc(gi ))cγi ,βi = 0 and
s θ 0 s
hence i=1 σ (bi )σ (σ (lc(gi )))σ (cγi ,βi ) = 0, so i=1 bi σ θ+γi (lc(gi ))cθ+γi ,βi
θ γi θ
θ 0 ?
= 0, with bi := σ (bi )cθ,γi . From this we conclude that b := (b1 , . . . , bs ) ∈ SF,θ
0
and also α(b) = b .
If BF,θ := {b 1 , . . . , b k } is a system of generators of SF,θ , then BF,θ is also a
? ?
system of generators of SF,θ : in fact, b := (b1 , . . . , bs ) ∈ SF,θ if and only if
Ps θ+γi
b ∈ SF,θ since i=1 bi σ (lc(gi ))cθ+γi ,βi = 0, hence b = r1 b 1 + · · · + rk b k =
? ∼
σ θ (r10 )b 1 + · · · + σ θ (rk0 )b k = r10 ? b 1 + · · · + rk0 ? b k . Since SF,θ =SF , then α(BF,θ )
is a system of generators of SF .
22 Claudia Gallego & Oswaldo Lezama

Thus, we have proved that if A is a bijective σ − P BW extension, θ ∈ Nn ,


F := {g1 , . . . , gs } ⊆ G and BF,θ is a system of generators of SF,θ , then α(BF,θ )
is a system of generators of SF .
Step 2. Now we will prove (ii)⇒(i) for quasi-commutative bijective σ − P BW
extensions. For this we will proof the condition (ii) of Theorem 27. If b :=
(b1 , . . . , bs ) ∈ BF,θ , then α(b) := b 0 = (b01 , . . . , b0s ) ∈ BF , and by the hypothesis
Ps 0 γi G
i=1 bi x gi − −→+ 0. From Proposition 20 (iv) we have
s
G
X
xθ ( b0i xγi gi ) −−→+ 0
i=1
s
G
X
σ θ (b0i )cθ,γi xθ+γi gi −−→+ 0
i=1
s
G
X
bi xθ+γi gi −−→+ 0.
i=1

Corollary 30. Let A be a quasi-commutative bijective σ −P BW extension. Let


F = {f1 , . . . , fs } be a set of non-zero polynomials of A. The algorithm below
produces a Gröbner basis for the left ideal < f1 , . . . , fs > (P (X) denotes the set
of subsets of the set X):

Gröbner Basis Algorithm for


Quasi-Commutative Bijective σ − P BW Extensions

INPUT: F := {f1 , . . . , fs } ⊆ A, fi 6= 0, 1 ≤ i ≤ s
OUTPUT: G = {g1 , . . . , gt } a Gröbner basis for < F >
INITIALIZATION: G := ∅, G0 := F
WHILE G0 6= G DO
D := P (G0 ) − P (G)
G := G0
FOR each S := {gi1 , . . . , gik } ∈ D DO
Compute BS
FOR each b = (b1 , . . . , bk ) ∈ BS DO
Pk γj G0
Reduce j=1 bj x gij −−→+ r, with r
reduced with respect to G0 and γj defined
as in the Definition 26
IF r 6= 0 THEN
G0 := G0 ∪ {r}
Gröbner Bases for Ideals of σ − P BW Extensions 23

From Theorem 10 and the previous corollary we get the following direct conclu-
sion.
Corollary 31. Let A be a quasi-commutative bijective σ − P BW extension.
Then each left ideal of A has a Gröbner basis.
A quick and easy illustration of the above algorithm is presented in the following
example.
Example 32. In O3 (2, 2, 2) = σ(R) < x2 , x3 >, with R := Q[x1 ] and
x2 x1 = 2x1 x2 , x3 x1 = 2x1 x3 , x3 x2 = 2x2 x3 ,
we consider the order deglex with x2  x3 ; let f1 := x2 + x1 and f2 := x3 + 1.
Using the previous algorithm we will construct a Gröbner basis for the left
ideal I :=< f1 , f2 >. We observe that in this example the endomorphisms
σ1 , σ2 : R → R coincide, and are defined by k 7→ k, x1 7→ 2x1 , for any k ∈ Q.
We start with G := ∅ and G0 := {f1 , f2 }.
Step 1. Since G0 6= G we have D = {S1 , S2 , S1,2 }, with
S1 = {f1 }, S2 = {f2 }, S1,2 = {f1 , f2 }.
We make G = G0 .
For S1 we compute BS1 , a system of generators of SyzR [σ γ1 (lc(f1 ))cγ1 ,β1 ]. Sim-
plifying the notation, we will write BS1 = SyzR [σ γ1 (lc(f1 ))cγ1 ,β1 ]. We have
XS1 = lcm{lm(f1 )} = lm(f1 ), β1 = exp(lm(f1 )) = (1, 0), so γ1 = (0, 0) and
cγ1 ,β1 = 1. Thus, BS1 = SyzR [1] = 0, and hence, we do not add a new polyno-
mial to G0 . For S2 we have the same situation. In general, since R has not zero
divisors, if S has only one element, then BS = 0 and no new element is added.
For S1,2 we compute BS1,2 = SyzR [σ γ1 (lc(f1 ))cγ1 ,β1 , σ γ2 (lc(f2 ))cγ2 ,β2 ]: XS1,2 =
lcm{x2 , x3 } = x2 x3 , so γ1 = (0, 1) and x3 x2 = 2x2 x3 , thus cγ1 ,β1 = 2; in a
similar way γ2 = (1, 0) and cγ2 ,β2 = 1. Then, BS1,2 = SyzR [σ10 σ21 (1)2, σ11 σ20 (1)1]
= SyzR [2, 1] = {(1, −2)} and hence
1xγ1 f1 − 2xγ2 f2 = x3 (x2 + x1 ) − 2x2 (x3 + 1) = −2x2 − 2x1 x3 ;
but −2x2 − 2x1 x3 can be reduced to the reduced polynomial f3 := 4x1 , so we
make G0 := {f1 , f2 , f3 }.
Step 2. Since G0 6= G we compute D = P (G0 ) − P (G) and we make G = G0 . In
D we only need to consider three subsets
S1,3 = {f1 , f3 }, S2,3 = {f2 , f3 }, S1,2,3 = {f1 , f2 , f3 }.
For S1,3 we have BS1,3 = {(−8x1 , 1)} and with this we get the polynomial −8x21
that can be reduced to 0 by f3 . For S2,3 we have a similar situation.
For S1,2,3 we found BS1,2,3 = {(1, −2, 0), (0, −16x1 , 1)}; for the first generator
we get −2x2 + 2x1 x3 and this polynomial can be reduced by f1 to 2x1 x3 + 2x1 ,
and this one can be reduce to 0 by f2 . For the second generator we get −16x1 x2
that can be reduced to 16x21 by f1 , but 16x21 is reduced to 0 by f3 .
Thus, G = {f1 , f2 , f3 } is a Gröbner basis of I.
24 Claudia Gallego & Oswaldo Lezama

7 Constructing Gröbner bases: part II


The proof of Corollary 31 was based in the Theorem 29, and in the proof of this
theorem we used that σ θ is surjective (and hence bijective) for all θ ∈ Nn , and
also that all constants cα,β are invertible, i.e., we assumed that A is bijective.
Moreover, we used the Proposition 20 (iv), i.e., that A is quasi-commutative.
However, we think that the Corollary 31 is true in general. In this section we
present some advances in this direction. We introduce the absorbent ideals of A
and we prove that each absorbent ideal of any bijective σ − P BW extension has
a Gröbner basis. We start with a general preliminary result, valid for arbitrary
σ − P BW extensions, which associates to a given σ − P BW extension A a
quasi-commutative extension, defined with the same constants of A.

Proposition 33. Let A be a σ − P BW extension of R. Then, there exists


a quasi-commutative σ − P BW extension Aσ of R in n variables z1 , . . . , zn
defined by

zi r = ci,r zi , zj zi = ci,j zi zj , 1 ≤ i, j ≤ n,

where ci,r , ci,j are the same constants that define A.

Proof. We consider n variables z1 , . . . , zn and the set of standard monomials


M := {z1α1 · · · znαn |αi ∈ Nn , 1 ≤ i ≤ n}, let Aσ be the free R-module with basis
M (i.e., A and Aσ are isomorphic as R-modules), we define the product in Aσ
by the distributive law and the rules
rz α sz β := rσ α (s)cα,β z α+β ,
where the σ’s and the constants c’s are the same that define A. The identities
of Remark 8 show that this product is associative, moreover note that R ⊆ Aσ
since for r ∈ R, r = rz10 · · · zn0 . Thus, Aσ is a quasi-commutative σ − P BW
extension of R, and also, each element f σ of Aσ corresponds to a unique element
f ∈ A, replacing the variables x’s by the variables z’s; the total order assumed
in M is z α  z β if and only if xα  xβ , so each element f σ of Aσ can be
represented in a unique way as at the beginning of Section 3, in particular, for
f σ ∈ Aσ , lm(f σ ), lc(f σ ) and lt(f σ ) are well defined, indeed, if lt(f ) = lc(f )xα ,
so lt(f σ ) = lc(f )z α .

Definition 34. Let A be a σ − P BW extension and I a left ideal of A; we say


that I is absorbent if I satisfies the following condition:

For any f ∈ I, if cxα ∈ t(f ), then cxα ∈ I.

The previous condition on I is equivalent to say that for any f ∈ I, lt(f ) ∈ I.


A and 0 are the trivial absorbent ideals of A. Note that any absorbent ideal of
A is finitely generated by terms, and if A is quasi-commutative, of course, the
converse is true.

Now we are able to present the main result of this section.


Gröbner Bases for Ideals of σ − P BW Extensions 25

Theorem 35. Let A be a bijective σ − P BW extension. Then, each absorbent


ideal of A has a Gröbner basis consisting of a finite set of terms.
Proof. Let I be an absorbent ideal of A; if I = 0, by definition, {0} is a Gröbner
basis of I. Let I 6= 0; if I = A, then a Gröbner of I is {1}. Let I 6= 0, A;
we consider the associated quasi-commutative bijective σ − P BW extension Aσ
(Proposition 33). Let I σ be the left ideal of Aσ generated by the set {f σ |f ∈ I},
then 0 6= I σ has a Gröbner basis Gσ := {g1σ , . . . , gtσ }; let 0 6= f ∈ I, then
0 6= f σ ∈ I σ is reducible with respect to Gσ , i.e., there exists g1σ , . . . , gsσ ∈ Gσ
such that lm(giσ )|lm(f σ ) for 1 ≤ i ≤ s, and there exist r1 , . . . , rs ∈ R such that
lc(f σ ) = r1 σ α1 (lc(g1σ ))cα1 ,g1σ +· · ·+rt σ αt (lc(gsσ ))cαt ,gsσ , with αi +exp(lm(giσ )) =
exp(lm(f σ )), but since lc(f σ ) = lc(f ), lm(giσ ) = lm(gi ), lc(giσ ) = lc(gi ) and the
σ’s and the constants c’s of Aσ are the same that define A, then f is reducible
with respect to G := {g1 , . . . , gt }. If we show that G ⊆ I, then G is a Gröbner
basis of I, and hence, the proof is complete. For each geσ , 1 ≤ e ≤ t, we
have in Aσ the decomposition geσ = hσ1 f1σ + · · · + hσk fkσ , with hσj ∈ Aσ and
fj ∈ I, 1 ≤ j ≤ k; then in A we have that q := h1 f1 + · · · + hk fk ∈ I, let
(j)
(j) (j) (j) µm (j) (j) (j) ν (j)
hj := c1 xµ1 + · · · + cmj x and fj := d1 xν1 + · · · + dpj x pj , but since I
j
Pk Pmj Ppj (j) µ(j) (j) ν (j)
is absorbent, then each term of q = j=1 l=1 w=1 cl x
l dw x w belongs
to I, hence
Pk Pmj Ppj (j) µ(j) (j) (j) (j)
h := j=1 l=1 l (dw )c (j) (j) xµl +νw ∈ I,
w=1 cl σ µ ,ν w
l

but geσ σ
= h , so ge = h, and hence, ge ∈ I for any 1 ≤ e ≤ t.
Finally, since I is absorbent, lt(ge ) ∈ I and each element 0 6= f ∈ I can be
reduced w.r.t {lt(g1 ), . . . , lt(gt )}, i.e., {lt(g1 ), . . . , lt(gt )} is a Gröbner basis of
I.
We conclude this section with some interesting properties of absorbent ideals.
Proposition 36. Let A be a σ − P BW extension of R and I a left ideal of A.
T
(i) If {Ik }k∈κ is a family of absorbent ideals of A, then k∈κ Ik is absorbent,
and \
sa(I) := J
I⊆J
J absorbent

is the least absorbent ideal of A containing I.


(ii) Let t(I) be the set of terms of all elements of I, then

I ⊆R < t(I) >⊆< t(I) >,

where R < t(I) > is the left R-submodule of A generated by t(I) and
< t(I) > is the left ideal of A generated by t(I).
(iii) I is absorbent ⇔ I =R < t(I) > ⇔ I =< t(I) >.
(iv) Let
26 Claudia Gallego & Oswaldo Lezama

I1 :=< t(I) >, Ik :=< t(Ik−1 ) >, k ≥ 2.

Then, I ⊆ I1 ⊆ I2 ⊆ · · · and there exists r ≥ 1 such that Ir = Ir+1 = · · · .

(v) sa(I) = Ir .
P
(vi) If {Ik }k∈κ is a family of absorbent ideals of A, then k∈κ Ik is absorbent,
and X
ia(I) := J
J⊆I
J absorbent

is the geatest absorbent ideal of A contained in I.

(vii) ia(I) ⊆ I ⊆ sa(I). I is absorbent if and only if ia(I) = I = sa(I).

Proof. (i) and (ii) are evident.


(iii): Suppose that I is absorbent and let cxα ∈ t(I), with cxα ∈ t(f ) for some
f ∈ I, then cxα ∈ I and R < t(I) >⊆ I, and by (ii), R < t(I) >= I. Conversely,
suppose that R < t(I) >= I and let cxα ∈ t(f ) for some f ∈ I, then cxα ∈ I
and hence I is absorbent.
Suppose again that I is absorbent and let f ∈< t(I) >, then f = h1 c1 xα1 +
· · · + hl cl xαl , with hi ∈ A and ci xαi ∈ t(I), 1 ≤ i ≤ l, so ci xαi ∈ I and hence
f ∈ I; from (ii), I =< t(I) >. Conversely, if < t(I) >= I, then R < t(I) >= I
and hence I is absorbent.
Since A is left Noetherian, then (iv) is true.
(v): Since Ir is absorbent and contains I, then sa(I) ⊆ Ir ; let J be an absorbent
ideal of A such that I ⊆ J; then t(I) ⊆ J, I1 ⊆ J, t(I1 ) ⊆ J, I2 ⊆ J, . . . , Ir ⊆ J,
and hence, Ir ⊆ sa(I).
(vi): Let cxα ∈ t(f ), f ∈ k∈κ Ik , then f = fk1 + · · · + fkr , with fki ∈ Iki ,
P
ki ∈ κ, Jki absorbent, so cxα = c1 xα + · · · + cP α α
r x , where ci x ∈ t(fki ), but
P since
α α
Iki is absorbent, then ci x ∈ Iki , so cx ∈ k∈κ Ik . This prove that k∈κ Ik
is absorbent. The second assertion is trivially true.
(vii) is trivial.

Example 37. The absorbent ideals ia(I) and sa(I) associated to any left ideal
I of A can be easy illustrated in the classical commutative polynomial ring A :=
Q[x, y, z], with deglex order and x > y > z; we will consider some particular
situations:

(i) For I =< x + y, z >,

0 ( ia(I) =< z >( I (< x, y, z >= sa(I) ( A.

(ii) If I =< x2 + y, 2y 2 + z 2 >,

0 = ia(I) ( I (< y, x2 , z 2 >= sa(I) (< x, y, z >( A.

(iii) For I =< x2 + y + 2, z >,


Gröbner Bases for Ideals of σ − P BW Extensions 27

0 (< z 2 >( ia(I) =< z >( I (< 2, y, x2 , z >= sa(I) = A.

Note that changing Q by Z10 , then I ( sa(I) =< 2, y, x2 , z 2 >( A.


(iv) For I =< x2 + 1, 2y 2 + z 2 >,

0 = ia(I) ( I (< 1, x2 , y 2 , z 2 >= sa(I) = A.

Remark 38. We conjecture that each left ideal of any bijective σ − P BW


extension has a Gröbner basis, and we hope that the properties listed in the
above proposition will be enough to prove this.

Acknowledgements

The authors are grateful to the editors and the referee for valuable suggestions
and corrections.

References
[1] Bell, A. and Goodearl, K., Uniform rank over differential operator
rings and Poincaré-Birkhoff-Witt extensons, Pacific Journal of Mathematics,
131(1), 1988, 13-37.

[2] Bueso, J., Gómez-Torrecillas, J. and Lobillo, F.J., Homological com-


putations in PBW modules, Algebras and Representation Theory, 4, 2001,
201-218.

[3] Bueso, J., Gómez-Torrecillas, J. and Verschoren, A., Algorithmic


Methods in Non-Commutative Algebra: Applications to Quantum Groups,
Kluwer, 2003.

[4] Chyzak, F., Quadrat, A. and Robertz, D., Effective algorithms for
parametrizing linear control systems over Ore algebras, INRIA, Rapport de
recherche n◦ 5181, 2004.

[5] Cohn, P., Free Rings and their Relations, Academic Press, 1985.

[6] Gateva-Ivanova, T., On the Noetherianity of some associative finitely pre-


sented algebras, J. Algebra 138, no. 1, 1991, 13-35.

[7] Lezama, O., Gröbner bases for modules over Noetherian polynomial com-
mutative rings, Georgian Mathematical Journal, 15, 2008, 121-137.

[8] Li, H, Noncommutative Gröbner Bases and Filtered-Graded Transfer, Lec-


ture Notes in Mathematics, Vol. 1795, Springer, 2002.

[9] McConnell, J. and Robson, J., Non-commutative Noetherian Rings,


Graduate Studies in Mathematics, AMS, 2000.
View publication stats
28 Claudia Gallego & Oswaldo Lezama

[10] Zhang, Y., Algorithms for Noncommutative Differential Operators, Ph.D


Tesis, University of Western Ontairo, London, Ontairo, 2004.

Departamento de Matemáticas
Universidad Nacional de Colombia
Bogotá, Colombia
e-mail : jolezamas@unal.edu.co

You might also like