Localization and Base Change Techniques in Computational Algebra
Krister Forsman
Dept. of Electrical Engineering, Linkoping University S-581 83 Linkoping, Sweden
email:
krister@isy.liu.se1993-02-04
REGLERTEKNIK
AUTOMATIC CONTROL LINKÖPING
Technical reports from the automatic control group in Linkoping are available by anonymous ftp at the address
130.236.24.1 (joakim.isy.liu.se)This report is contained in the compressed ps-le named
/pub/reports/LiTH-ISY-R-1445.ps.ZLocalization and Base Change Techniques in Computational Algebra
Krister Forsman
Dept. of Electrical Engineering
Linkoping University, S-581 83 Linkoping, Sweden email:
krister@isy.liu.se1993-02-04
Abstract.
This report collects some useful results from commutative algebra and describes how they can be used to decrease the complexity in Grobner base computations. As a byproduct we see how Grobner bases relate to characteristic sets. The algorithms presented are well known but have, as far as the author knows not been collected in one place earlier.
Keywords:
Grobner bases, elimination, commutative algebra, localization, linear algebra, remainders, characteristic sets, zero-dimensional ideals.
1 Introduction and Terminology
The most important methods in elimination theory are resultants, Grobner bases (GB) and characteristic sets (CS).
Resultants are described in most classical textbooks on commutative algebra: 17, 20, 25, 32].
Some references for GB are 3, 10, 14, 15, 16, 29, 33].
Characteristic sets are discussed in e.g. 5, 6, 13, 18, 22, 31, 34].
Grobner bases are occasionally called standard bases, though this terminology does not seem to be historically correct the original reference for standard bases 19], treats a slightly dierent case.
It is the aim of this report to collect some useful results from commutative algebra and describe how they can be used to decrease complexity in GB computations. As a byproduct we will see how GB and CS are related. The reader is supposed to have some minimal background in commutative algebra and Grobner basis theory.
The following terminology and notation is used:
A
=
kX1:::Xn ] where
kis an arbitrary eld.
F
i =
k(
Xn
;i
+1:::Xn )
X1:::Xn
;i ]. This makes
Fd an
n;d-dimensional ring.
Capitals (
X1:::Xn ) are used for free variables, whereas lowercase letters denote vari-
ables subject to relations
xi
2A=a.
The ideal generated by
f1:::fm
2Ais written
hf1:::fm
i.
The leading monomial of a polynomial
pis written lm
p. If
p=
PcX
where
are
N
n -vectors we have
deg
p= max
fc 6= 0
g(1)
and lm
p=
Xdegp .
For an integral domain
Rits eld of fractions is written
Q(
R). For example we have
F
n =
Q(
A).
plex stands for purely lexicographic term-ordering.
The concept of ranking gives us a classication of all plex term-orderings on
A.
Denition 1.1 A ranking of the variables
X1:::Xn is simply a permutation of these symbols. If
aprecedes
bin the permutation we say that
ahas a lower rank than
b, written
a<b
. If
Aand
Bare two sets of variables and any element of
Ahas lower rank than all
elements of
Bwe write
A<B.
2The ranking can be thought of as a reordering of the entries in the exponent vector if
ris an element of the permutation group
Sn then
X
r
(1)>Xr
(2) >:::>Xr
(n
) ) X=
Xr
(1)1 :::Xr
(nn
)(2) Thus the \standard" ranking is
X1 >:::>Xn .
Denition 1.2 We say that
p2Ais regular w.r.t.
Xi if lm
p2kXi ] using plex and ranking
X
i the highest.
2Another word for regular is monic , but be careful not to confuse this with the property of having leading coecient 1. Consider
p
=
XY3;2
Y2+ 5
X3;1 (3)
using plex and the ranking
X <Y. Then lm
p=
XY3, so
pis not regular w.r.t.
Y, but the leading coecient of
pis 1.
Denition 1.3 If
p2Spec
Athen we dene the dimension of
pto be dim
p= trdeg k
Q(
A=p)
2
Note that this denition covers the case
p2Fi too, since we have no requirements on
k.
Denition 1.4 A
d-dimensional ideal
p2Spec
Ais in weak noetherian position if
p\k
Xn
;d
+1:::Xn ] = 0
p
is in noetherian position (n.p.) if it is in weak noetherian position and
A=pis an integral
k
xn
;d
+1:::xn ]-algebra.
2To put an ideal in weak noetherian position one only has to renumber the variables. Any prime ideal can be put in noetherian position by a linear change of variables this is known as Noether's normalization lemma 1, 23, 28].
2 Some Facts about Grobner Bases
We will assume that Grobner bases are minimal, i.e. reduced. With this requirement and the convention that the elements of the GB have leading coecient 1 (the coecient ring is a eld) Grobner bases are unique 16]. We write
GB(
a) for the Grobner base of the ideal
aA
, making an implicit agreement on the term-ordering and that GB are always reduced.
Theorem 2.1 Let
abe an ideal in
A. dim
a= 0 i for all
ithere is a
p 2 GB(
a) such that
pis regular w.r.t.
Xi .
Proof. See 16] or 3, page 209].
2This implies that
A=ais an integral
k-algebra. The theorem is true independently on what term ordering is used. Note that there may still be more than
npolynomials in the GB:
Example 2.1 The set
fX12X1X2X22gis a GB of a zero-dimensional ideal in
kX1X2].
2
For prime ideals the situation is simpler, though:
Theorem 2.2 A plex GB for a prime zero-dimensional ideal in
kX1:::Xn ] has
nele- ments.
Proof. See 16].
2In 16] it is also proved that generically the
n;1 rst elements of the GB are linear in the leading variable. Thus the generic look of a plex-GB for a zero-dimensional prime ideal w.r.t.
Xn
<:::<X1is
8>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
: X
1
;p
1
(
Xn )
X
2
;p
2
(
Xn ) ...
X
n
;1;pn
;1(
Xn )
p
n (
Xn )
(4)
where
8i:
pi
2 kXn ] and deg
pn
>deg
pi for
i= 1
:::n;1. Note that the way this theorem is stated in 16] the GB is not reduced.
An (arbitrary) ideal that has a GB of the type (4) is said to be in generic position or general position. Thus a zero-dimensional ideal is in generic position if for two dierent zeroes (
a1:::an ) and (
b1:::bn ) we have
an
6=
bn . In 16] it is proved that almost all variable changes of the type
X
i
7!Xi
i<n Xn
7!Xn + n
X;11
c
i
Xi (5)
where all
ci
2k, puts a zero-dimensional prime ideal in generic position. In fact, 15] show
that any zero-dimensional radical ideal is put in generic position by a generic morphism of
the type (5). It is clear that with the above denition of genericity
a homogeneous ideal cannot be in generic position unless all elements of the GB are linear.
it depends on the ranking used whether an ideal is in generic position. E.g.
hX1X23iis in generic position, but
hX13X2iis not.
3 Change of Term Ordering Using Linear Algebra
The main disadvantage with Grobner bases is that their computational complexity is very high, in general. It has been showed that the complexity for computing a GB is lower for total degree orderings, e.g. revlex, than for a plex ordering 21, 24, 26, 27].
There is an approach to elimination theory that is also based on Grobner bases, but does not use the plex term ordering. This approach is primarily due to Boege, Gebauer and Kredel 2] it is also discussed in e.g. 3, 15]. The algorithm, let us call it the BGK algorithm, works for the special case when the ideal is zero-dimensional and it gives the univariate polynomial that generates the contraction of
Ito
kXi ] for some
Xi . We call this polynomial the minimal polynomial of
Xi , since we think of
Q(
A=I) as an algebraic extension of
k, even though
Idoes not have to be prime, i.e. maximal. (In fact
Idoes not even have to be zero-dimensional see below.) The idea is rst to compute a Grobner base G w.r.t. a
revlex ordering and then compute remainders
R
d = rem(
Xdi
G ) (6)
for successive
d:s. Each
Rd is considered as an element in the vector space
kX1 :::Xn ] d . Sooner or later these expressions for
Rd will be linearly dependent over
k, since the ideal has dimension zero, and this linear dependency relation gives us the minimal polynomial for
Xi . The reason for using the approach described above is thus that it is typically more ecient than computing an entire plex GB see 2]. The BGK algorithm is part of the Grobner base package in Maple 4], available under the name
finduni.
It turns out that the ideas above can be generalized to the following result: using linear algebra techniques it is possible to nd a Grobner base w.r.t. an arbitrary term-ordering given a GB w.r.t. any other term-ordering. This is known as the Faugere-Gianni-Lazard- Mora algorithm 7].
If the ideal considered is in generic position the minimal polynomial obtained from the BGK algorithm tells us
1. how many zeroes there are, 2. the multiplicity of the zeroes, 3. whether the zeroes are real or not.
These questions can be of practical importance e.g. when dealing with problems in real algebraic geometry, see e.g. 11].
Via Bezout's theorem we get a sucient condition for an ideal to be in generic position:
Theorem 3.1 Let
Ibe a zero-dimensional ideal in
A. Furthermore, let
Nbe the product
of the total degrees of the polynomials of a generating set for the ideal
I(
Ndepends on the
particular set of generators) and
dthe degree of the minimal polynomial for
Xn . If
N=
dthen
Iis in generic position.
Proof. The number of complex zeroes, counting multiplicity, is bounded from above by
Naccording to the ane version of Bezout's theorem 12, page 223]. This means that
d Nwith equality i the ideal has
Nzeroes, all of which dier in the coordinate
Xn , i.e.
Iis in
generic position.
2Theorem 3.1 is often dicult to use directly, since Bezout's theorem only states an inequal- ity. If the original set of generators for the ideal is made (auto-)reduced, chances increase that equality occurs (remember that the number
Ndepends on what generators that are chosen).
It should be noted that the BGK algorithm might terminate even if the ideal in question is not zero-dimensional. The condition
kXi ]
\I 6= 0 is necessary and sucient for termination
a simple example of this is
hXi kXY]. More involved, application oriented examples can be found in 11] or 9, chapter 6].
4 Some Theorems on Localization
An apparent disadvantage with the BGK algorithm is that a (sucient) condition for it to terminate is that the ideal in question be zero-dimensional. However, if the ideal is not zero-dimensional we can in some cases change the coecient eld by localization so that the ideal becomes zero-dimensional and then apply BGK. Before doing this we have to know something about what happens to ideals under localization.
Theorem 4.1 Let
Rbe a commutative ring and
Sa multiplicative system in
R. There is a one-to-one correspondence between the prime ideals of
Rthat do not intersect
Sand the prime ideals of
S;1R(the localization of
Rat
S).
Proof. See 1, proposition 3.11, iv] or 35, chapter IV,
x10, theorem 15].
2. Thus if we consider extension and contraction
R!S;1Rwe have that
8p2
Spec
R:
p\S=
) pec =
p(7) In the applications considered here
Rwill be
Aand
S=
kXn
;d
+1:::Xn ]
nf0
gfor some
d. Thus
S;1R=
Fd . The reasons for extending ideals to
Fd are two:
computing in
Fd is cheaper than computing in
A.
the ideals considered may become zero-dimensional in
S;1Rso that the BGK algorithm can be used.
Theorem 4.2 Let
p 2Spec
Abe
d-dimensional and in weak noetherian position. If
pe is the extension of
pto
Fd then dim
pe = 0 .
Proof.
pe
2Spec
Fd according to theorem 4.1. But dim
p=
dimplies that
x1:::xn
;d
are algebraic over
kxn
;d
+1:::xn ]. Thus
dim
pe = trdeg k
(X
n;d+1:::X
n)Q(
Fd
=pe ) = 0 (8)
2
Many times the ideals involved are not prime, but we can still use the localization tech-
nique as long as we are a little careful. The following theorem explains exactly how the ideals
of
Rrelate to those of
S;1R.
Theorem 4.3 Let
abe an ideal in
Rand consider extension and contraction
R! S;1R. We have
a
ec =
s
2S (
a:
s)
Proof. See 1, proposition 3.11, ii].
25 Characteristic Sets
The reason that there are two apparently dierent methods in elimination theory is that reduction can be made in dierent ways. Buchberger's algorithm uses ordinary polynomial division, whereas Ritt's algorithm for computing CS 31] relies on pseudodivision. For details, see the references mentioned in the introduction. It is sometimes claimed that characteristic sets are more ecient than GB from a computational point of view, but it is questionable if this is still true when the \tricks" described above (sections 3 and 4) are used before applying Buchberger's algorithm.
In fact, in this section we will see that in the case of a prime ideal the dierence between CS and GB is not very large. We will use the following theorem:
Theorem 5.1 Let
pAbe a
d-dimensional prime ideal and G a plex GB for
pw.r.t. the ranking
X1>:::>Xn . If
pis in n.p. then G =
fp1:::pn
;d
gwhere
pi is regular w.r.t.
Xi
for all
i.
Proof. This is proposition 5.9 in 16].
2Furthermore we need a simple fact on reduction:
Lemma 5.1 If
fg2Aand
gis regular w.r.t. its leading variable then
fis reduced w.r.t.
gin the sense of Ritt i it is reduced w.r.t. g using the standard denition in Gr obner base theory.
Proof. Immediate from the denitions.
2Theorem 5.2 Suppose that
pis a
d-dimensional prime ideal of
Ain noetherian position.
If G =
GB(
p) w.r.t. a plex term-ordering then then G is a CS for
p.
Proof. G is a subset of
psatisfying the denition of CS according to theorem 5.1 and
lemma 5.1.
2Theorem 5.3 Suppose that
pis a
d-dimensional prime ideal of
Ain weak noetherian position and that
pe is the extension of
pto
Fd . If G =
GB(
pe ) w.r.t. a plex term-ordering and G
bis obtained from G by clearing denominators, then G
bis a CS for
p.
Proof. Since
pe is zero-dimensional (theorem 4.2) G =
fp1:::pn
;d
gand
pi are regular w.r.t.
Xi . This means that the denominators are polynomials in
kXn
;d
+1:::Xn ]. It is clear that there is no nonzero polynomial
fin
psuch that deg(
fXi )
<deg(
pi
Xi ), so G
bforms a CS for
p.
2Note that G
bis not necessarily a GB for
p:
Example 5.1 Let
p=
hX1X2;X3X12;X4ibe an ideal in
kX1:::X4]. Clearly
pis prime. We have that
F =
GB(
p) =
fX12;X4 X1X2;X3 X1X3;X2X4 X22X4;X32g(9) So
pis in weak noetherian position. If
pe =
F2pthen
G =
GB(
pe ) =
fX1;X4X
3 X
2
X 2
2
; X
2
3
X
4
g
(10)
so that
G
b=
fX1X3;X2X4 X22X4;X32g 6= F
2