MT30512: Polynomials Semester 2, 2006{07

Chapter 1: Polynomials and Ideals

1 Polynomials in n variables

1.1 Polynomials in one variable

A polynomial in the variable x with oeÆ ients in the ring R is an expression
of the form Xn
ai xi = a + a x + : : : + an xn + an xn ;
0 1 1
1
(1)
i=0
where n  0 and the ai are elements of R. If the P notation is used, then x = 1 0

by onvention. Note that x itself is NOT an element of R, but is a formal symbol

whi h helps to make al ulations as natural as possible. If n = 0, the polynomial
is simply an element a 2 R, and is alled a onstant polynomial.
0
The set of all polynomials (1) is a ring under the usual addition and multipli-
ation operations. It is denoted by R[x℄ and is alled the ring of polynomials
in one variable over R. We shall use letters su h as f; g; h or p; q; r to denote
polynomials; if it is ne essary to draw attention to the variable being used, then
we write e.g. f (x), f (y) instead of f . We shall often assume that R is a eld F .
We shall always assume that R is a ommutative ring with 1. You should
be familiar with the meaning of these terms from MT2262.
The most important fa t about divisibility in F [x℄ is the division theorem.
If f 6= 0 (i.e. f is not the zero polynomial, so that some oeÆ ient ai 6= 0), then
we may assume in (1) that an 6= 0; n is then alled the degree of the polynomial
f , and we write n = deg f . Note that the degree of the zero polynomial is not
dened.
Theorem 1.1 (Division Theorem) Let F be a eld and let f; g 2 F [x℄, where
g 6= 0. Then there are unique polynomials q , the quotient, and r, the remain-
der, in F [x℄ su h that (i) f = qg + r, and (ii) either deg r < deg g or r = 0.
We shall see that the division theorem fails for polynomials in two or more
variables. This is one fundamental reason why the algebra of polynomials in more
than one variable is more diÆ ult. One of the obje ts of this ourse is to onvin e
you that it is also more interesting and rewarding! However, before moving on,
we note two more things we an do with one-variable polynomials.
(i) Substitution Given polynomials f (x); g (x) 2 R[x℄, we an repla e every
o urren e of x in f (x) by g(x) to obtain a new polynomial f (g(x)). For
example if f (x) = x + 1 and g(x) = x 2, then f (g(x)) = (x 2) + 1 =
2 2

x 4x + 5 and g (f (x)) = (x + 1) 2 = x 1.
2 2 2

1
(ii) Evaluation Given a polynomial f (x) 2 R[x℄ and an element a 2 R, we
an repla e every o urren e of x in f (x) by a to obtain a new element of
R, whi h we denote by f (a). In this way, the polynomial f (x) denes a
fun tion from R to R. This fun tion is alled a polynomial fun tion.
Although there is a logi al distin tion between polynomials and polynomial fun -
tions, it really does no harm to identify these two on epts. However, you should
be aware that two polynomials are equal if and only if they have identi al o-
eÆ ients, whereas two fun tions are equal if and only if they have the same
graphs. The dieren e an be seen by taking the eld F to be nite; for ex-
ample if p is a prime number and F = Fp, the eld with p elements, then the
polynomials xp and x dene the same fun tion. But if F is an innite eld su h
as Q (the rational numbers), R (the real numbers) or C (the omplex numbers),
two polynomials are equal if and only if the orresponding polynomial fun tions
are equal. Espe ially for F = R, polynomial fun tions give a geometri way to
visualise polynomials via their graphs.
1.2 Polynomials in two variables
When we try to write down a denition like (1) for a general polynomial in two
variables x and y with oeÆ ients in R, we have a diÆ ulty: whi h order do
we write the terms in? For example should the `standard form' of a quadrati
polynomial in x and y be a ; + a ; x + a ; y + a ; x + a ; xy + a ; y , or should
00 10 01 20
2
11 02
2

the term in x ome before the term in y? Noti e that we an olle t terms in
2

either variable, and write this as (a ; + a ; x + a ; x ) + (a ; + a ; x)y + a ; y

00 10 20
2
01 11 02
2

or as (a ; + a ; y + a ; y ) + (a ; + a ; y)x + a ; x .
00 01 02
2
10 11 20
2

We an hide this problem by writing the polynomial in the form

X
ai;j xi y j (2)
i;j 0

where ai;j 2 R and the sum is nite, i.e. ai;j = 0 for all but a nite number of
indexing pairs (i; j ). Although the problem resurfa es when we try to do pra ti al
al ulations, we adopt (2) as the formal denition. We all an expression xiyj a
monomial in x and y , and if ai;j 6= 0 we all ai;j xi y j a term of the polynomial.
The (total) degree of this monomial is i + j , and the (total) degree of a
nonzero polynomial is the maximum of the degrees of its terms. A polynomial is
alled homogeneous of degree d if it has no term of degree 6= d. (Thus the
zero polynomial is `homogeneous of degree d' for all d  0, although its degree is
not dened!)
The set of all polynomials (2) forms a ring R[x; y℄ with the usual addition and
multipli ation operations, alled the ring of polynomials in two variables over
R. By olle ting terms with the same power of x, we an regard a polynomial in
R[x; y ℄ as a polynomial in x with oeÆ ients whi h are polynomials in y , i.e. as

2
an element of S [x℄ where S = R[y℄. Thus we have a natural way to identify the
rings R[x; y℄ and (R[x℄)[y℄. Similarly, we an identify R[x; y℄ and (R[y℄)[x℄.
As for the one variable ase, we shall often use letters su h as f; g; h or p; q; r
to denote polynomials; if it is ne essary to draw attention to the variables being
used, then we write f (x; y) et instead of f . If R = F is a eld, then a polynomial
f in F [x; y ℄ is always divisible by any nonzero onstant polynomial, i.e. any
polynomial of degree 0.
It is easy to see that the division theorem fails for polynomials in two variables.
Suppose, for example, that we try to divide the polynomial x by the polynomial
y . This means we are looking for polynomials q and r su h that x = qy + r and
deg r < 1 or r = 0. Thus r must be a onstant polynomial, and learly there is
then no polynomial q 2 R[x; y℄ su h that x = qy + r.
1.3 Polynomials in n variables
The denitions above are easily extended to the ase of n > 2 variables. A
monomial in n variables x ; : : : ; xn is an expression of the form
1

m = x


xn ; 1
1 n

where  ; : : : n are integers  0. The (total) degree of the monomial m is

1
deg m =  + : : : + n. It is onvenient to adopt a `multi-index' notation for
1
writing monomials, i.e.
x := x


xn 1
1 n

where x = (x ; : : : ; xn) is a `ve tor variable' and  = ( ; : : : ; n) is a `ve tor

1 1
exponent'. We denote the total degree of x by jj =  + : : : + n. 1
Let R be a ommutative ring with 1. A polynomial over R in the variables
x ; : : : ; xn is a nite sum f = f (x ; : : : ; xn ) of terms of the form a x , where
1 1
the oeÆ ient a 2 R. If a 6= 0, we say that the term a x appears in f . We
identify the monomial x


xn with the identity element 1 2 R. The onstant
0 0

term of f is the element a ;:::; 2 R; it may of ourse be zero.

1

(0 0)
In multi-index notation, a polynomial an be written as follows.
f = a x + : : : + a r x r
(1)
(1)
( ) (3)
( )

where the oeÆ ients a ; : : : ; a r are elements of R. The set of all polynomials
(1) ( )
of the form (3) forms a ring R[x ; : : : ; xn℄ with the usual addition and multipli a-
1
tion operations, alled the ring of polynomials in n variables over R. Noti e
that by olle ting powers of one of the variables, xn say, we an identify the rings
R[x ; : : : ; xn ℄ and R[x ; : : : ; xn ℄[xn ℄.
1 1 1
The (total) degree of a polynomial f 6= 0 is the maximum of the degrees of
the terms appearing in f . If all terms appearing in f have the same degree d,
f is alled a homogeneous polynomial of degree d or a form of degree d.
In the ase d = 1 we speak of a linear form, in the ase d = 2 of a quadrati
form, in the ase d = 3 of a ubi form, et .

3
2 Ideals in polynomial rings

2.1 The ideal generated by a nite set

Let R be a ommutative ring with 1 and let a ; : : : ; an 2 R. The ideal generated
1
by a1 ; : : : ; an is the set of all elements of R of the form a = r1 a1 + : : : + rn an ,
where r1; : : : ; rn 2 R. We denote this ideal by ha1; : : : ; ani.
In parti ular, the ideal hai generated by a single element a 2 R is the set of all
multiples of a. In the general ase, elements of the ideal are formed analogously
to linear ombinations of ve tors in a ve tor spa e.
Example 2.1 Let R = F [x; y ℄ where F is any eld. Then hx ; y i is the set of
2

all polynomials f (x; y) of the form x g + yh where g and h are polynomials in

2

x and y . By thinking of f as a sum of monomials, we an see that f 2 hx ; y i 2

if and only if every monomial whi h appears in f is divisible by either x or y. 2

Thus hx ; yi is the set of all polynomials su h that the onstant term and the
2

oeÆ ient of x are both zero.

2.2 Prin ipal ideals
In general, an ideal in a ring R is a non-empty subset of R whi h is losed under
the operations of addition and multipli ation by elements of R. More formally,
I  R is an ideal if and only if
 0 2 I;
 if a; b 2 I then a + b 2 I ;
 if a 2 I and r 2 R then ra 2 I .
It is easy to he k that if a ; : : : ; an 2 R then ha ; : : : ; ani is an ideal. Conversely,
1 1
given an ideal I  R, we an ask if there exist elements a ; : : : ; an 2 R su h that
1
I = ha ; : : : ; an i. Su h a set of elements is alled a generating set or a basis
1
for I . If I is generated by a single element, i.e. if I = hai for some a 2 R, then I
is alled a prin ipal ideal of R.
It is easy to see from Example 2.1 that the ideal I = hx ; yi is not a prin ipal
2

ideal in R = F [x; y℄. For if I = hf i, then the polynomial f would have to divide
both x and y. But the only ommon divisors of x and y are nonzero onstant
2 2

polynomials a 2 R. However a is invertible in R, and hen e hai = R.

Remarkably, the situation is quite dierent in the ase of polynomials in one
variable over a eld.
Theorem 2.2 Let F be a eld. Then every ideal in the polynomial ring F [x℄ is
prin ipal.

4
Proof Let I  F [x℄ be an ideal. If I = f0g then it is prin ipal, so we may
assume that I ontains a nonzero polynomial g of minimum degree. Clearly
hgi  I . We shall prove that I  hgi, so that I = hgi and hen e I is prin ipal.
Thus let f 2 I . By the Division Theorem 1.1, there exist polynomials q; r 2
F [x℄ su h that f = qg + r and deg r < deg g or r = 0. Sin e r = f qg and
f; g 2 I , it follows from the denition of an ideal that r 2 I . By our hoi e of g ,
it is impossible that deg r < deg g, and hen e r = 0. But then f = qg, and so
f 2 hg i. 
2.3 Finitely generated ideals
We see from Theorem 2.2 that the failure of the Division Theorem in the polyno-
mial ring F [x; y℄ in two variables over the eld F is losely related to the existen e
of ideals whi h are not prin ipal. From Example 2.1, we might be tempted to
guess that every ideal in F [x; y℄ is generated by at most two elements. The
following example shows that this is far too optimisti .
Example 2.3 Let F be a eld, let n  1 and let I = hxn ; xn y; : : : ; xy n ; y ni
1 1

be the ideal in F [x; y℄ generated by the set of all monomials of degree n. We shall
prove that I annot be generated by fewer than n + 1 polynomials.
First noti e that the polynomials in I are the polynomials in whi h no term
of degree < n appears. Let f ; : : : ; fm be a generating set for I : we must prove
1
that m  P n +1. Ea h of the n +1 monomials xi y n i an be expressed in the form
xi y n i = mj gi;j fj where gi;j 2 F [x; y ℄. Sin e the fj have no terms of degree
< n, equating terms of degree n gives x y = j ai;j hj , where hj is the degree
=1 Pm
i n i
n homogeneous part of fj and ai;j 2 F is the onstant term of gi;j . However, the
=1

homogeneous polynomials of degree n form a ve tor spa e of dimensionPn +1 over

F , with basis fxn ; xn y; : : : ; xy n ; y ng, and the equations xi y n i = mj ai;j hj
1 1

express ea h of these basis elements as linear ombinations of the h ; : : : ; hm .

=1
1
Hen e m  n + 1, by standard fa ts of linear algebra.
We see from this example that there is no integer N su h that every ideal in F [x; y℄
is generated by some set with at most N elements. Obviously, the same is true
for F [x ; : : : ; xn℄ for every n  2. Thus the algebrai stru ture of a polynomial
1
ring in two or more variables is very dierent from the ase of one variable. Here
are some questions we might ask about polynomials in n variables.
 Do two polynomials have a greatest ommon divisor? (Re all that for n = 1
the answer is yes, and it an be omputed by the Eu lidean Algorithm.)
 Given an ideal I = hf ; : : : fni and a polynomial f , how an we determine
1
whether f is in I or not? (For n = 1, f 2 I if and only if the generator of
the ideal divides f .)

5
  Does every ideal in F [x ; : : : ; xn℄ have a nite generating set?
1

The last question is answered by

Theorem 2.4 (Hilbert's Basis Theorem) Let F be a eld and let I be an ideal
in the polynomial ring P = F [x ; : : : ; xn ℄. Then I is nitely generated, i.e. there
1
exists a positive integer N and a set of polynomialsPf ; : : : ; fN in I su h that every
1
polynomial f 2 I an be written in the form f = N i ri fi for some polynomials
=1
r1 ; : : : ; rN in P .
We shall return to Hilbert's theorem later.
The following example shows that ideals in polynomial rings need not be
nitely generated in general. This example will not be needed in our future
work, and so the details (in luding the pre ise denitions!) are left for you to
think about.
Exer ise 2.5 Let F be a eld and let x ; x ; : : : be an innite sequen e of vari-
1 2
ables. Let P be the polynomial ring F [x ; x ; : : :℄ and let I be the ideal onsisting
1 2
of all polynomials with zero onstant term. Then I is not nitely generated.
An ideal whi h is not nitely generated an always be generated by some set,
for example, the set of all its elements. Here we make pre ise the denition of a
generating set for an ideal.
Let R be a ommutative ring with 1, let I be an ideal in R, and let A be a
subset of R. Then A is a generating set or basis for I (written I = hAi) if and
only if every element a 2 I an be written in the form
a = r a + : : : + rnan ;
1 1 (4)
where a ; : : : ; an 2 A and r ; : : : ; rn 2 R, for some positive integer n.
1 1
The point is that although the set A an be innite, only nitely many ele-
ments of A are involved in (4) for a given a 2 I .
2.4 Quotient Rings
Re all (MT2262) that when I is an ideal in the ommutative ring R, then the
quotient ring R=I is onstru ted as follows. An element of P=I is a oset
f = f + I = ff + g j g 2 I g, and that the ring operations in P=I are dened by
(f + I ) + (f + I ) = f + f + I; i:e: f + f = f + f ;
1 2 1 2 1 2 1 2
(f + I )
(f + I ) = f
f + I; i:e: f
f = f
f :
1 2 1 2 1 2 1 2

This is a good way to onstru t lots of interesting rings, as we shall see later.

6
3 Monomial Ideals and Di kson's Lemma

Monomials are mu h easier to handle than general polynomials. For example,

we an write down the GCD and LCM of two given monomials at sight, e.g.
GCD(x y z; x z ) = x z, LCM(x y z; x z ) = x y z .
3 2 2 2 2 3 2 2 2 3 2 2

Let F be a eld and let P = F [x ; : : : ; xn℄ be the polynomial ring in n variables

1
over F . A monomial ideal in P is an ideal I whi h is generated by a set of
monomials M . We do not assume that M is nite. Examples 2.1 and 2.3 are
examples of monomial ideals in F [x; y℄.
Theorem 3.1 Let I be a monomial ideal in P = F [x ; : : : ; xn ℄, and let f be a
1
polynomial in P . Then f 2 I if and only if every monomial whi h appears in f
is in I .
Proof The `if' part is obvious; we prove the `only if' part. Let f 2 I , and
let a xbe a term in f . Thus f = a x + f , where the monomial x does not
1
appear in f . We must prove that x 2 I .
1
Let S be a set of monomials whi h generates I . Then f is a nite sum
f = g x + : : : + gr x r
1
(1) ( )
(5)
for some polynomials g ; : : : ; gr 2 P and some monomials x ; : : : ; x r 2 S .
1
(1) ( )

Equating terms in (3.1) with exponent ve tor , we obtain

a x = t x + : : : + tr x r
1
(1) ( )
(6)
where ea h ti is either zero or is the term appearing in gi with exponent ve tor
 =  (i). Sin e a 6= 0, equation (6) expresses x as an element of I . 
This result redu es the problem of testing a polynomial for membership of a
monomial ideal I to the problem of testing a monomial. The next result provides
su h a test.
Theorem 3.2 Let S be a set of monomials in P = F [x ; : : : ; xn ℄, and let x be
1
any monomial in P . Then x 2 hS i if and only if x is divisible by some element
of S .
Proof Let I be the set of polynomials f su h that all monomials appearing
in f are divisible by some element of S . Then I is an ideal in P . Sin e every
element of S is in I , we have hS i  I . But if f 2 I , every monomial appearing
in f is in hS i and so f 2 hS i. Hen e I  hS i, and so I = hS i. Taking f = x ,
we have the required statement. 
Theorems 3.1 and 3.2 allow us to des ribe the elements of a monomial ideal
hS i pre isely. For example, hxy; y i  F [x; y℄ onsists of all polynomials whi h
3

have zero onstant term and in whi h the monomials y, y and all xn (n  1) do 2

not appear.
7
3.1 Sums of ideals
Let I and J be ideals in R. Then their sum I + J is dened by
I + J = fa + b j a 2 I; b 2 J g: (7)
This is an ideal in R. If A and B are subsets of R su h that A generates I and
B generates J , then I + J is generated by the union A [ B of these generating
sets. It follows that the sum of two monomial ideals is again a monomial ideal.
Example 3.3 If I = hx y; y i and J = hx ; xy i, then I + J = hx y; y ; x ; xy i.
2 3 4 2 2 3 4 2

Noti e that in general I [J is not an ideal, but if I  J then I [J = J = I +J .

We an think of I + J as the smallest ideal whi h ontains both I and J .
3.2 Produ ts of ideals
Let I and J be ideals in R. Then their produ t IJ is the set of elements r 2 R
of the form n
X
r= ai bi ; where ai 2 I; bi 2 J: (8)
i=1
This is an ideal in R. If A and B are subsets of R su h that A generates I and
B generates J , then IJ is generated by the produ t AB = fai bi j ai 2 A; bi 2 B g
of these generating sets. It follows that the produ t of two monomial ideals is
again a monomial ideal.
Example 3.4 If I = hx y; y i, J = hx ; xy i, then IJ = hx y; x y ; x y ; xy i.
2 3 4 2 6 3 3 4 3 5

Noti e that we an throw out x y from this list of generators, be ause it is di-
4 3

visible by another monomial in the list, x y . The remaining list fx y; x y ; xy g

3 3 6 3 3 5

is a minimal generating set for IJ .

3.3 Interse tions of ideals
The interse tion I \ J of ideals I and J in R is dened to be their interse tion
as sets, i.e. the set of all r 2 R su h that r 2 I and r 2 J . This is an ideal in R.
If x and x are monomials, then the interse tion of the prin ipal ideals hxi
and hx i is the prin ipal ideal hx i where x = LCM(x ; x ). For example,
hx yi \ hxy i = hx y i. This follows immediately from Theorem 3.2. More
2 3 2 3

generally, let I and J be the monomial ideals generated by sets of monomials S 1

and S respe tively. Then by Theorem 3.2 a monomial x is in I \ J if and only
2
if x is divisible by some element x 2 S and also by some element x 2 S .
1 2
Hen e the set fLCM(x; x ) j x 2 S ; x 2 S g is a set of monomial generators
1 2
for I \ J .

8
Example 3.5 If I = hx y; y i and J = hx ; xy i, I \ J = hx y; x y ; x y ; xy i.
2 3 4 2 4 2 2 4 3 3

Noti e that we an throw out the element x y from the list of generators, be ause
4 3

it is divisible by another monomial in the list, namely xy . Thus fx y; x y ; xy g 3 4 2 2 3

is a minimal generating set for I \ J .

3.4 Di kson's Lemma
The next result is a major step towards the proof of the Hilbert Basis Theorem.
Theorem 3.6 Let F be a eld, let S be a set of monomials in P = F [x ; : : : ; xn ℄, 1
and let I = hS i. Then I is generated by a nite subset of the monomials in S .
In parti ular, every monomial ideal in P is nitely generated.
Proof We rst prove that I is generated by some nite set of monomials, whi h
need not belong to S . For this, we use indu tion on n, the number of variables.
The ase n = 1 is trivial, sin e every monomial ideal is of the form hxk i for some
k  0. Thus we assume that the result is true for n 1 variables, and let I
be a monomial ideal in P = F [x ; : : : ; xn℄. We shall write the variable xn as y,
1
so that monomials in P an be written as x ym, where x = (x ; : : : ; xn ) and 1 1
 = ( ; : : : ; n ).
1 1
Given a monomial ideal I in P , let J be the ideal in F [x ; : : : ; xn ℄ generated 1 1
by the monomials x su h that xyk 2 I for some k  0. Then J is a monomial
ideal in F [x ; : : : ; xn ℄. By the indu tion hypothesis, we an hoose a nite set
1 1
of monomials fx ; : : : ; x s g whi h generates J .
(1) ( )

For 1  i  s, by denition of J there is an integer mi su h that x i ym 2 I . ( ) i

Let m = max mi. For 0  k  m 1, let Jk be the ideal in F [x ; : : : ; xn ℄ 1 1

generated by the monomials x su h that xyk 2 I . Then Jk is a monomial ideal
in F [x ; : : : ; xn ℄, so again the indu tion hypothesis allows us to hoose a nite
1 1
set of these monomials fx ; : : : ; x s g whi h generates Jk .
k (1) k( k)

We laim that I is generated by the monomials x i ym, for 1  i  s, together

( )

with the monomials x i yk, for 1  i  sk and 0  k  m 1. This is a nite

k( )

set M of monomials.
Clearly all these monomials are in I . Let xyk be a monomial in I . If k  m,
then sin e x 2 J and J is the monomial ideal generated by fx ; : : : ; x s g, (1) ( )

x is divisible by x i for some i with 1  i  s by Theorem 3.2. Hen e

( )

x y k is divisible by x i y m. On the other hand, if 0  k  m 1 then sin e

( )

x 2 Jk a similar argument proves that x y k is divisible by x i y k for some i k( )

with 1  i  sk . Thus every monomial in I is divisible by some monomial in the

set M , and it follows that I = hM i.
To omplete the proof, we must prove that a nite generating set for I an
be sele ted from any given set of monomials S whi h generate the ideal I . Let
m ; : : : ; mr be a nite set of monomials whi h generates I : su h a set exists by the
1
rst part of the proof. By Theorem 3.2, ea h mi is divisible by some monomial
m0i 2 S , for 1  i  r. Then m0 ; : : : ; m0r lie in S and generate I .
1 
9
4 Orderings on Monomials

We would like to have a division algorithm for polynomials in n variables. That

is, we would like to have an algorithm that has, as inputs, two polynomials f
and g and, as outputs, two polynomials q (the quotient of f by g) and r (the
remainder when f is divided by g). The polynomials q and r should satisfy
(1) f = qg + r, and
(2) r is `smaller' in some sense than g .
For this to work, we must hoose an ordering on monomials. To see why,
onsider the example f = x y + xy , g = x y. Here it is natural to hoose q = y
2 2 2 2

and r = xy . So we would like to regard xy as `smaller' than x y. However, if we

2 2 2

inter hange the variables and divide x y + x y by xy , by symmetry we should

2 2 2 2

regard x y as the remainder and hen e `smaller' than xy . To es ape from this
2 2

ontradi tion, we must rst agree whether to have x y > xy or x y < xy .

2 2 2 2

Denition 4.1 Let F be a eld and let M be the set of all monomials x in
P = F [x1 ; : : : ; xn ℄. A monomial ordering on P is a total ordering < on M
whi h is ompatible with multipli ation, i.e. the following four axioms hold.
 (tri hotomy rule) Given x ; x 2 M , exa tly one of the three statements
x < x , x < x and x = x is true.
 (transitive rule) If x ; x ; x 2 M satisfy x < x and x < x , then
x < x .
 (initialisation rule) If x 2 M and x 6= 1, then 1 < x .
 (multipli ation rule) If x ; x ; x 2 M and x < x , then xx < x x .
Along with the < symbol, we shall use the symbols ; >;  with the obvious
meanings.
We give three examples. In these examples, we shall assume (as part of
the ordering) that the variables are ordered so that x > x >


> xn. Let 1 2
x = x


xn , x = x


xn .
1
1 n
1
1 n

(i) The lexi ographi ordering, Lex

Dene x < x in Lex if and only if 1 = 1 ; : : : ; k 1 = k 1, k < k for
some k = 1; : : : ; n. In other words, reading from left (x1 ) to right (xn),
the rst time the exponents are dierent the exponent in x is smaller.
(ii) The degree lexi ographi ordering, DegLex
Dene x < x in DegLex if and only if jj < j j or if jj = j j and x < x
in Lex. In other words, we use the degree (the sum of the exponents) as
the rst test, and use Lex to order monomials of the same degree.
10
(iii) The degree reverse lexi ographi ordering, DegRevLex
Dene x < x in DegRevLex if and only if jj < j j or if jj = j j and
for some k, n = n; : : : ; k+1 = k+1, k > k . In other words, we use the
degree as the rst test, and order monomials of the same degree as follows:
x < x if and only if when reading from right (xn ) to left (x1 ), the rst
time the exponents are dierent the exponent in x is larger.
These orderings, espe ially DegRevLex, take a bit of getting used to. In
parti ular, it is important to realise that the order of the variables must be
spe ied. For example, if the variables are x and y and we want to use Lex in
F [x; y ℄, we must spe ify whether x < y or y < x. With x < y the Lex order on
monomials looks like
1 < x < x < x <


< y < xy < x y <


< y <



2 3
(9) 2 2

and the DegLex order looks like

1 < x < y < x < xy < y < x < x y < xy < y < x <



2 2
(10)
3 2 2 3 4

If n = 2 in our `standard notation' for n variables, we have x < x , and so 2 1

the DegLex order looks like
1 < x < x < x < x x < x < x < x x < x x < x < x <


(11)
2 1
2
2 1 2
2
1
3
2 1
2
2
2
1 2
3
1
4
2

The dieren e between DegLex and DegRevLex an be seen as follows. Let

x > x > x . Then x x x > x x in DegLex, but x x x < x x in DegRevLex.
1 2 3
2
1 2 3 1
3
2
2
1 2 3 1
3
2

An important feature of all monomial orderings is that there is no innite

stri tly de reasing sequen e of monomials. This is obvious for DegLex and
DegRevLex, sin e there are only nitely many monomials smaller than a given
monomial in these orderings. Che king the property for Lex will help you to
understand this ordering.
Having xed a hoi e of monomial ordering, we an write a given polynomial
f 2 F [x ; : : : xn ℄ in a standard form with respe t to the ordering. This means
1
that we write the terms in f in de reasing order.
Example 4.2 Let n = 5, x > x > x > x > x , and let f = x (1 + 3x x ) +
1 2 3 4 5
3
2 5
x (x x x ). Then the standard forms for f with respe t to the three orderings
1
2 2 2
4
above are tabulated as follows.
1 5 2

Ordering Standard Form

Lex 3x x x + x 3
1 2 5
3
1 x21 x22 x4 + x21 x25
DegLex 3x x x 3
1 2 5 x21 x22 x4 + x21 x25 + x31
DegRevLex x21 x22 x4 + 3x31 x2 x5 + x21 x25 + x31

11
In multi-index notation, the standard form of a typi al nonzero polynomial f 2 P
(with respe t to a given monomial ordering) an be written as
f = a x + : : : + a r x r
(1)
(1)
( )
( )
(12)
where x > : : : > x r and the oeÆ ients a ; : : : ; a r are nonzero elements
(1) ( )
(1) ( )
of F .
We all x the leading monomial of f , LM(f ), a the leading oeÆ-
(1)
(1)
ient of f , LC(f ), and a x the leading term of f , LT(f ).
(1)
(1)

Thus in Example 4.2, if the ordering is Lex or DegLex then LM(f ) = x x x , 3

2 5
LC(f ) = 3 and LT(f ) = 3x x x , and if the ordering is DegRevLex then LM(f ) =
1
3
2 5
x x x , LC(f ) = 1 and LT(f ) = x x x .
1
2 2 2 2
4 4
Note that if f and g are nonzero polynomials, then LT(f
g) = LT(f )
LT(g),
1 2 1 2

and similarly for LM and LC.

To omplete this se tion, we use Di kson's Lemma to prove the important
property of monomial orderings stated above.
Proposition 4.3 Let < be a monomial ordering on the monomials in P =
F [x ; : : : ; xn ℄, and let S be any set of monomials in P . Then S has a minimum
1
element.
Proof Let I = hS i be the ideal generated by S . By Di kson's Lemma 3.6,
there is a nite subset fx ; : : : ; x r g of S whi h generates I . Let x be the
(1) ( )

minimum of these monomials. We laim that x is the minimum element of S .

To see this, let x 2 S , so that x is divisible by some x i , 1  i  r, say ( )

x = x i x . Sin e 1  x , x i  x and hen e x  x .

 ( ) ( )

In parti ular, it follows that there is no innite stri tly de reasing sequen e of
monomials in a monomial ordering, sin e su h a sequen e would have no minimum
element.

12
5 A division algorithm in n variables

In this se tion we will formulate a division algorithm in P = F [x ; : : : ; xn℄. We

1
begin by reviewing the familiar 1-variable ase. Given f and g 6= 0 in F [x℄, the
algorithm produ es a quotient q and remainder r with f = qg + r and deg r <
deg g or r = 0. Thus r = f qg 2 f + I , where I = hgi is the ideal generated by g,
and f + I = ff + h j h 2 I g is the oset of I whi h ontains f . Thus we an think
of the division algorithm as a pro edure for obtaining the `simplest' representative
for that oset ontaining the given polynomial f : note that r + I = f + I , sin e
f r = qg 2 I . We would like to imitate this situation in the n variable ase,
by asso iating to a polynomial f and a nonzero ideal I of P a `remainder' r su h
that f r 2 I . However, the ideal I need not be prin ipal. If I = hg ; : : : ; gr i,
1
then f r 2 I if and only if f an be written in the form
f = q g + : : : + qs gs + r;
1 1 (13)
where the `quotients' qi and `remainder' r are in P .
As an example, if n = 2 and I = hx ; x x ; x i, we would want r to be the
2
1 2
2

polynomial a + bx + x obtained by deleting all terms of degree  2 from f .

1 2

1 2
More generally, if I is a monomial ideal, then we an obtain a uniquely dened
remainder r by deleting from f all terms involving monomials in I . Note that
the qi will not be uniquely determined by (13) when s > 1.
We must rst x a monomial ordering in P . To see the need for this, onsider
the ase f = 2x +3y +4z, g = x + y + z in F [x; y; z℄. With the ordering x > y > z,
the quotient is 2 and the remainder is y + 2z, but with the ordering z > y > x,
the quotient is 4 and the remainder is 2x y. In ea h ase, all monomials whi h
appear in the remainder are lower than the leading monomial in g, and this is a
ondition we wish to use to hara terise the remainder r in (13).
Here is an example to show what is involved. The basi idea is the same as
for division in the ase n = 1, i.e. at ea h step we an el the leading term of f
with respe t to the given monomial order, by multiplying some divisor gi by a
suitable monomial and subtra ting a s alar multiple of this produ t from f .
Example 5.1 Using Lex order in F [x; y ℄ with x > y , we will divide f = xy + 1 2

by g = xy + 1 and g = y + 1. The leading terms xy of g and y of g both

1 2 1 2
divide LT(f ) = xy . We use g sin e it is listed rst. This gives f yg =
2
1 1
(xy + 1) y(xy + 1) = y + 1 = f say. The leading term LT(f ) = y whi h
2
1 1
is not divisible by LT(g ) but is divisible by LT(g ). This gives f ( 1)g =
1 2 1 2
( y + 1) + (y + 1) = 2. Sin e LT(g ) and LT(g ) do not divide 2 we stop: the
1 2
remainder r = 2.
Unfortunately, things do not always go so smoothly as this, as the following
example shows.

13
Example 5.2 Again using Lex order in F [x; y ℄ with x > y , we will divide f =
x2 y + xy 2 + y 2 by g1 = xy 1 and g2 = y 2 1. The rst two steps go as before
(remember that we use g1 when both leading terms divide):
f x(xy 1) = xy + x + y = f
2 2
1
f1 y (xy 1) = x + y + y = f 2
2

The leading term of f is x whi h is not divisible by LT(g ) = xy or LT(g ) = y .

2 1 2
2

However f is not the remainder be ause we an ontinue to `redu e' it using g .

2 2
So the next step is to `move' LT(f ) = x to the remainder: f = r + f where
2 2 1 3
r = x, f = y + y . Now ontinue as before: f 1(y 1) = y + 1 = f . This
1 3
2
3
2
4
time no monomial in f is divisible by LT(g ) or LT(g ), and so both terms in
4 1 2
f go into the remainder: r = r + y , f = y + f where f = 1; r = r + 1,
4 2 1 4 5 5 3 2
f = 1+ f where f = 0. The algorithm stops when the polynomial to be divided
5 6 6
is zero, and the remainder r is the urrent remainder r = x + y +1. The quotients 3
q and q an be re overed from the steps of the algorithm involving g and g
1 2 1 2
respe tively: in this ase we have q = x + y, q = 1. 1 2

Theorem 5.3 (Division Algorithm in P = F [x ; : : : xn ℄) Fix a monomial 1

order in P and let g ; : : : ; gs be a nite list of nonzero polynomials in P . Then
1
every f 2 P an be written in the form f = q g + : : : qs gs + r for some quotients
1 1
q ; : : : ; qs and remainder r in P su h that for 1  i  s
1

 no monomial whi h o urs in r is divisible by LT(gi),

 LM(qi gi)  LM(f ).
Proof We prove the theorem by giving an expli it algorithm for onstru tion
of the quotients q ; : : : ; qs and the remainder r.
1

Input: g ; : : : ; gs; f
1
Output: q ; : : : ; qs; r
1
q := 0; : : : ; qs := 0; r := 0
1
p := f
WHILE p 6= 0 DO
i := 1
div := false
WHILE i  s AND div = false DO
IF LT(gi) divides LT(p) THEN
qi := qi + (LT(p)=LT(gi ))
p := p (LT(p)=LT(gi ))gi
div := true
ELSE
14
 i := i + 1
IF div = false THEN
r := r + LT(p)
p := p LT(p)

The rest of the proof onsists of he king that this algorithm operates orre tly.
The variable p represents the intermediate polynomial still to be divided, and the
algorithm terminates when p = 0. The Boolean variable `div' tells us whether
the leading term of some gi divides the leading term of p.
To prove that the algorithm works, we rst he k that the equation f =
q g + : : : + qs gs + p + r holds at ea h stage. There are two ases, depending
1 1
on whether a `division step' (div = true) or a remainder step (div = false) has
o urred. It follows that f = q g + : : : + qsgs + r when the algorithm stops. Sin e
1 1
the leading monomial of p de reases in the given monomial order at ea h step,
and sin e every stri tly de reasing sequen e in the monomial order is nite, the
algorithm does always stop. You should he k that the polynomials output by
the algorithm have the required properties. 
The next example shows that this algorithm does not have all the properties
we would like.
Example 5.4 We repeat Example 5.2, inter hanging the divisors. That is, using
Lex in F [x; y℄ with x > y, we will divide f = x y + xy + y by g = y 1 and
2 2 2
1
2

g = xy 1.
2
This time the algorithm produ es q = x + 1, q = x, r = 2x + 1. So the
1 2
output depends on the order of the input polynomials g ; : : : ; gs; in parti ular,
1
the remainder is not uniquely spe ied by the properties stated in Theorem 5.3.
The next example shows that this problem an also ae t the question of
whether the remainder is zero.
Example 5.5 Using Lex in F [x; y ℄ with x > y , we will divide f = xy x by 2

g = xy + 1 and g = y 1. The result is q = y , q = 0, r = x y . However,

1 2
2
1 2
if we take g = y 1 and g = xy + 1, we obtain instead q = x, q = 0 and
1
2
2 1 2
r = 0.
Re all that f is in the ideal I = hg ; : : : ; gsi if and only if f = q g + : : : + qs gs
1 1 1
for some hoi e of q ; : : : ; qs. Thus Example 5.5 shows that xy x is in the ideal
1
2

I generated by xy + 1 and y 1. Indeed, though xy x = x(y 1) is in the

2 2 2

smaller ideal generated by y 1 alone, the `bad' hoi e of basis fxy + 1; y 1g

2 2

disguises the fa t that xy x 2 I .

2

To remedy this, we seek `good' generating sets for an ideal I in P . For su h

a set, we want the remainder r to be independent of the order in whi h the
generators are input to the division algorithm, and in parti ular we want to have
r = 0 if and only if f 2 I . We will see that Grobner bases have pre isely these
ni e properties.
15
MT30512 Polynomials Semester 2, 2006{07

Chapter 2: Computing with Polynomials

6 Gr
obner bases

6.1 Leading term ideals

We have seen that leading terms play a entral part in the division pro ess,
and that a hoi e of monomial ordering provides every nonzero polynomial f
with a uniquely dened leading term LT(f ).
Denition 6.1 Let I be a nonzero ideal in F [x ; : : : ; xn ℄ and let LT(I ) be the set
1
of all leading terms of elements of I . Then the ideal hLT(I )i generated by LT(I )
is alled the leading term ideal of I .
Noti e that hLT(I )i is a monomial ideal, sin e LT(f ) = ax where a 6= 0 and
x = LM(f ) is a monomial. Thus we ould equally well have dened hLT(I )i to
be the ideal generated by the leading monomials of elements of I .
Example 6.2 Using any monomial ordering in F [x; y ℄, let I = hx 1; y + 2i.
Then LT(x 1) = x and LT(y + 2) = y, so LT (I ) ertainly ontains hx; yi, the
ideal onsisting of all polynomials with zero onstant term. In fa t we an prove
LT(I ) = hx; yi: if not, then I must ontain a polynomial whose leading term is
onstant, and su h a polynomial is itself onstant. But it is easy to see that I
ontains no non-zero onstant polynomials.
The next example shows that it is not always true that hLT(f ); LT(f )i is 1 2
the leading term ideal of hf ; f i. 1 2

Example 6.3 Let I = hf ; f i, where f = x 2xy , f = x y 2y + x, and

1 2 1
3
2
2 2

use the DegLex order in F [x; y℄ with x > y. Then

x
(x y 2y + x) y
(x 2xy ) = x ;
2 2 3 2

so that x 2 I . Hen e x 2 hLT(I )i. However LT(f ) = x , LT(f ) = x y so the

2 2
1
3
2
2

monomial ideal hLT(f ); LT(f )i = hx ; x yi is stri tly smaller than hLT(I )i.
1 2
3 2

6.2 The Hilbert Basis Theorem

We now ombine the idea of the leading term ideal with Di kson's Lemma to
obtain a proof of the Hilbert Basis Theorem.
Proposition 6.4 Let I be a nonzero ideal in F [x ; : : : ; xn ℄. Then there exists a
1
nite set of polynomials g ; : : : ; gs 2 I su h that hLT(I )i = hLM(g ); : : : ; LM(gs)i.
1 1

16
Proof Sin e hLT(I )i is a monomial ideal, this follows immediately from Di k-
son's Lemma (Theorem 3.6). 
Theorem 6.5 (Hilbert Basis Theorem) Let F be a eld and let I be an ideal
in F [x1 ; : : : ; xn ℄. Then I is nitely generated, i.e. I = hg1 ; : : : ; gsi for some
g1 ; : : : ; gs 2 I .
Proof If I = 0, we take f0g as the generating set. Otherwise we take
g1 ; : : : ; gs as in Proposition 6.4. Sin e ea h gi 2 I , it is lear that hg1; : : : ; gsi  I .
We laim that I  hg1; : : : ; gsi, so that in fa t I = hg1; : : : ; gsi, and so g1; : : : ; gs
is the required nite generating set for I .
Thus let f 2 I . By the division algorithm (Theorem 5.3), we an write f in
the form f = q1g1 + : : : + qsgs + r, where no monomial appearing in the remainder
r is divisible by any of the leading monomials LM(g1 ); : : : ; LM(gs). We shall
prove by ontradi tion that r = 0.
If r 6= 0, then LT(r) 2 hLT(I )i = hLM(g1); : : : ; LM(gs)i. Sin e this is a
monomial ideal, it follows from Theorem 3.2 that LT(r) is divisible by LM(gi)
for some i. This is a ontradi tion. Hen e r = 0, and so f = q1g1 + : : : + qsgs.
Thus f 2 hg1; : : : ; gsi. 
6.3 Noetherian rings
To omplete our dis ussion of the Hilbert Basis Theorem, we shall relate it to the
general stru ture theory of ommutative rings.
Theorem 6.6 (Noetherian rings) Let R be a ommutative ring with 1. Then
the following onditions are equivalent.
(i) Every ideal I in R has a nite generating set.
(ii) There is no innite stri tly in reasing sequen e of ideals in R; more pre isely,
given a sequen e fIn gn1 of ideals in R su h that I1  I2  I3  : : :, there
exists a positive integer n0 su h that In = In0 for all n  n0 .
Condition (ii) is often alled the as ending hain ondition (ACC). A ring
R satisfying either, and hen e both, of these onditions is alled a Noetherian
ring.
Proof (i) ) (ii): Let I  I  I  : : : be an as ending hain. Dene
1 2 3
I= 1 I . It is easy to he k that I is an ideal. By (i), I has a nite generating
S
n=1 n
set f1; : : : ; fs. Choose ni su h that fi 2 In for 1  i  s, and dene n0 = max ni .
Then f1; : : : ; fs are all in In0 . Hen e I  In0 . But In0  I , hen e In0 = I . Hen e
i

In = In0 for all n  n0 .

(ii) ) (i): We prove the ontrapositive, i.e. `not (i)' ) `not (ii)'. Let I be
an ideal whi h is not nitely generated. Dene an as ending hain I1  I2 : : :  In
17
as follows, by indu tion on n. Let r be any element of I , and let I = hr i. Sin e
1 1 1
I is not nitely generated, I 6= I . Let r be any element of I n I , and let
1 2 1
I = hr ; r i. Then I  I . Assume as indu tion hypothesis that a stri tly
2 1 2 1 2
as ending hain I  I : : :  In has been onstru ted, so that Is = hr ; : : : ; rsi
1 2 1
and rs 62 Is for 2  s  n. Sin e I is not nitely generated, I 6= In. Let rn be
1 +1
any element of I n In, and let In = hr ; : : : ; rn i. This ompletes the indu tive
+1 1 +1
onstru tion of the as ending hain fIngn . It is lear from the onstru tion
1
that the ACC (ii) is false for this hain. 

18
7 Gr
obner bases

Denition 7.1 Let I be an ideal in P , and let < be a monomial order. A nite
set of polynomials g1 ; : : : ; gs in I is alled a Gr
obner basis (or standard basis)
of I if and only if
hLT(g ); : : : ; LT(gs)i = hLT(I )i:
1 (14)
Our proof of Theorem 6.5 shows that
 a Grobner basis for I is indeed a generating set for I , and
 every nonzero ideal I in P = F [x ; : : : ; xn℄ has a Grobner basis.
1

However, the proof does not tell us

 how to nd a Grobner basis for a given ideal I , or
 how to he k that a given set of polynomials in I is a Grobner basis.
We shall return to these questions in the next se tion, but one spe ial ase is
worth mentioning now.
Proposition 7.2 If I = hg i is the prin ipal ideal generated by a nonzero g 2 P ,
then fg g is a Grobner basis for I with respe t to any monomial ordering.
Proof Clearly hLT(g )i  hLT(I ). If f 2 I and f 6= 0, then f = gh for some
nonzero h 2 P . Hen e LT(f ) = LT(g)
LT(h) and so LT(f ) 2 hLT(g)i. Hen e
hLT(I )i  hLT(g)i and so hLT(I )i = hLT(g)i. 
7.1 Normal forms and quotient rings
A key property of Grobner bases is that the remainder in the division algorithm
is uniquely determined when we divide by a Grobner basis.
Proposition 7.3 Fix a monomial ordering in P = F [x ; : : : ; xn ℄, let I be an
1
ideal in P , and let f 2 P be a polynomial. Then there is a unique r 2 P su h
that
(i) f = g + r for some g 2 I , and
(ii) no term of r is in hLT(I )i.
The remainder r is sometimes alled the normal form of f with respe t to I .

19
 Proof Let g ; : : : ; gs be a Grobner basis for I . By the division algorithm,
1
Theorem 5.3, we an write f in the form f = q g + : : : + qsgs + r where the
1 1
quotients q ; : : : ; qs and the remainder r are in P . Thus f = g +r where g = q g +
1 1 1
: : : + qs gs 2 I , so that (i) is satised. Theorem 5.3 also tells us that no term of r
is divisible by any of LT(g ); : : : ; LT(gs). Sin e hLT(I )i = hLT(g ); : : : ; LT(gs)i,
1 1
a monomial x is in hLT(I )i if and only if it is divisible by some LT(gi). Hen e
if x 2 hLT(I )i, then x does not divide any term of r.
Finally we must show that if r0 also satises (i) and (ii) then r = r0. By (i),
we have f = g0 + r0 where g0 2 I , so r0 r = g g0 2 I . If r0 r 6= 0, then
LT(r0 r) 2 hLT(I )i. But no monomial in r or r0 is divisible by any monomial
in hLT(I )i. This is a ontradi tion, hen e r = r0. 
The remainder r of f is sometimes written as f . It is parti ularly onvenient
for al ulations in the quotient ring P=I (Se tion 2.4).
There is an important familiar ase of the quotient ring onstru tion, when
F = R and I = hx + 1i. The ring R[x℄=hx + 1i is isomorphi to the eld of
2 2

omplex numbers C, the isomorphism being determined by the orresponden e

x + I ! i. In this ase, the normal form of a polynomial f (x) is the remainder
r(x) = a + bx on division of f (x) by g (x) = x + 1. (Re all that, by Proposition
2

7.2, the generator of a prin ipal ideal is a Grobner basis for the ideal.) Thus
hLT(I )i = hx i. Now the usual method of al ulating with omplex numbers is
2

equivalent to al ulating with these remainders, subje t to the rule that x = 1, 2

i.e. the remainder of f (x) = x is r(x) = 1.

2

The following result shows how to generalise this type of al ulation in quo-
tient rings.
Proposition 7.4 Fix a monomial ordering in P = F [x ; : : : ; xn ℄, and let I be
1
an ideal in P . Then
(i) the orresponden e r ! r + I is a bije tion between the set of all remainders
r of polynomials in P and the set P=I of all osets of I in P ;
(ii) the set of osets whi h orrespond to monomials x 2P su h that x 62
hLT(I )i is a basis for P=I as a ve tor spa e over F ;
(iii) the quotient ring P=I is isomorphi to the ring whose elements are the
remainders of polynomials in P , with the same addition operation as in P
but with the produ t of r1 and r2 taken as the remainder of r1 r2 .
Proof By taking r = 0, Proposition 7.3 shows that f 2 I if and only if
f = 0; and by taking g = 0, Proposition 7.3 shows that f = f if and only if no
term of f is in hLT(I )i. It is also easy to see from Proposition 7.3 again that
f + f = f + f and that the remainder of f
f is f f . The details are left as
1 2 1 2 1 2 1 2
an exer ise. 

20
8 S -polynomials
In this se tion we shall see how to he k that a given set of polynomials in an
ideal I of P is a Grobner basis. This in turn will lead to an algorithm for the
onstru tion of Grobner bases.
Denition 8.1 Let f , g be nonzero polynomials and let
x = LCM(LM(f ); LM(g ))
be the least ommon multiple of their leading monomials. The S-polynomial
S (f; g ) is dened by
x x
S (f; g ) = (15)
LT(f )
f LT(g)
g:
Note that S (f; f ) = 0 and S (g; f ) = S (f; g) for all f; g 2 P .
Example 8.2 Let f = x y + y + x, g = 2xy x , where the ordering is DegLex
2 2 2 2

with x > y. Then LT(f ) = x y, LT(g) = 2xy and x = x y . Thus

2 2 2 2

= xx yy (x y + y + x) 2xxyy (2xy x );
2 2 2 2
S (f; g ) 2
2 2
2
2 2

= 21 x + y + xy:
3 3

The important thing to noti e about this example is that the leading terms of f
and g have an elled, and that although S (f; g) = yf xg 2 I where I = hf; gi, 1

the leading term LT(S (f; g)) = x 62 hx y; xy i, the monomial ideal generated
2
1 3 2 2

by the leading terms of f and g.

2

Sin e S (f; g) 2 I and LT(S (f; g)) 62 hLT(f ); LT(g)i, it follows that f and g
do not form a Grobner basis for the ideal I . It is natural to add h = S (f; g) to
the generating set ff; gg for I , and to ask whether ff; g; hg is a Grobner basis
for I .
We shall see that this method of adding S -polynomials to a given generating
set for an ideal I in P always su eeds in produ ing a Grobner basis for I in a
nite number of steps. Somehow, S -polynomials of pairs of polynomials in the
set fg ; : : : ; gr g a ount for all an ellations of leading terms in forming elements
1
of the ideal hg ; : : : ; gr i. The next result is a key property of S -polynomials.
1

Proposition 8.3 P Let f ; : : : ; fs 2 P be polynomials with the same leading mono-

1
mial xÆ , let g = si i fi be a linear ombination of the fi with oeÆ ients
i 2 F , and suppose that LM(g ) < xÆ . Then g is a linear ombination of the
=1

S-polynomials S (fi ; fi ), 1  i  s 1, and LM(S (fi ; fj )) < xÆ for all pairs

+1
i 6= j .

21
 Proof First we do the spe ial ase where the fi are moni , i.e. LC(fi ) = 1
for 1  i  s. Sin e LT(fi) = LT(fj ) = xÆ in this ase, S (fi; fj ) = fi fj .
Clearly LM(S (fi; fj )) < xÆ sin e the leading terms an el.
Consider the `teles oping' sum
s
X
( + : : : + i)(fi
1 fi+1 );
i=1

where fs = 0. For 1  i  s the oeÆ ient

+1 of fi in this sum is ( + : : : i)
1
( + : : : + i ) = i, so the sum redu es to i ifi = g. The leading oeÆ ient
1 1
Ps

of g is Psi i and so the ondition LM(g) < xÆ is equivalent to Psi i = 0.

=1

If this holds, then the last term in the teles oping sum vanishes, and we have
=1 =1

g = is ( + : : : + i )S (fi ; fi ).
P 1
1 +1
In the general ase, let LC(fi) = ai for 1  i  s, so that g = Psi iai (fi=ai)
=1

and LC(g) = i iai. Sin e fi=ai is moni and S (fi; fj ) = S (fi=ai; fj =aj ), the
Ps =1

general ase follows by applying the spe ial ase to the polynomials fi =ai. 
=1

The following property of S -polynomials will be needed in the next se tion.

Proposition 8.4 Given f; g 2 P and monomials x , x ,
S (x f; x g ) = x S (f; g )
for some monomial x .
Proof Let LT(f ) = ax , LT(g) = bx , a; b 2 F , so that LCM(LM(f ); LM(g)) =
x, where i = max(i; i ). Then LT(x f ) = ax  , LT(x g) = bx  , so that
+ +

LCM(LT(x f ); LT(x g)) = xÆ where Æi = max(i + i; i + i ).

With this notation, we have
x  x 
S (f; g )
= ax f g
 bx
xÆ xÆ
S (x f; x g ) = ( x f )
  (x g )

axÆ  +
bx +

x xÆ
= ax  f
bx
g
= x S (f; g)

where i = Æi i.
We must he k that x really is a monomial, i.e. i  0 for all i. For this,
note that the inequalities i + i  i , i + i  i imply that Æi  i and Æi  i ,
and hen e Æi  i as required.

22
9 Bu hberger's riterion

We next prove Bu hberger's S -polynomial riterion for a set of polynomials to be

a Grobner basis for the ideal that it generates. Re all from the division algorithm
that if the remainder on division of f by a set of polynomials G = fg P; :s: : ; gsg
1
taken in some order is zero, then f an be written in the form f = i qigi
where LM(qi gi)  LM(f ) for 1  i  s.
=1

Theorem 9.1 (Bu hberger) Let g ; : : : ; gs 2 P and let I = hg ; : : : ; gsi. Then

1 1
G = fg ; : : : ; gsg is a Grobner basis for the ideal I if and only if for all i 6= j the
1
remainder of S (gi; gj ) on division by G (taking the elements of G in some order)
is zero.
Proof If G is a Grobner basis for I , then by Proposition 7.3 the remainder
of any element of I with respe t to G is 0. Sin e S (gi; gj ) 2 I , this proves the
ne essity of Bu hberger's riterion.
To prove suÆ ien y, let f 2 I , so that
s
X
f = hi gi (16)
i=1
for some polynomials h ; : : : ; hs 2 P . Among all su h expressions, we hoose one
1
where maxi LM(hi gi) is minimal in the given monomial order. Let us denote this
monomial by xÆ . By (16), LM(f )  xÆ .
If LM(f ) = xÆ , then LM(f ) = LM(higi) = LM(hi )
LM(gi) for some i, so that
LM(gi) divides LM(f ), and hen e LM(f ) 2 hLM(g ); : : : ; LM(gs)i. If this is true
1
for all f 2 I , then we have proved that hLM(I )i = hLM(g ); : : : ; LM(gs)i, i.e. G
1
is a Grobner basis for I , as required.
Thus it remains to prove that LM(f ) = xÆ . We shall assume that LM(f ) < xÆ
and obtain a ontradi tion using Proposition 8.3.
Let us renumber the polynomials g ; : : : ; gs so that the terms in (16) with
1
leading monomial xÆ ome before those with smaller leading monomials, i.e.
LM(hi gi) = xÆ for 1  i  t and LM(higi) < xÆ for i > t. For 1  i  t,
let LT(hi) = ix i where i 2 F , and let fi = x i gi. Then we an write (16) as
( ) ( )

t
X
f = i fi + f0 ; (17)
i=1
where f = i (hi LT(hi ))gi + i t higi. Thus LM(fi) = xÆ forP1  i  t,
0
Pt
=1
Ps
= +1
while LM(f ) < xÆ . It follows from (17) that LM(g) < xÆ , where g = ti i fi.
0
Thus the leading terms of the fi an el when we form the linear ombination
=1

g = ti i fi . By Proposition 8.3 we an write g in the form

P
=1

t 1
X
g= bi S (fi ; fi+1 )
i=1

23
where bi 2 F . Sin e fi = x i gi, Proposition 8.4 shows that S (fi; fi ) =
( )
+1
x i S (gi ; gi ) for some monomial x i . Hen e
( )
+1
( )

t 1
X
g= bi x (i) S (gi; gi+1 ):
i=1

We are now ready to use the hypothesis that all the S -polynomials of pairs of
polynomials in the set G have remainder 0. Thus by the division algorithm
(Theorem 5.3) we an write S (gi; gi ) in the form +1

s
X
S (gi; gi+1 ) = si;j gj
j =1

for some polynomials si;j 2 P su h that LM(si;j gj )  LM(S (gi; gi )) for all i; j . +1
Hen e t s 1 s
X X X
g= bi x i si;j gj = kj gj ;
( )

i=1 j =1 j =1
where kj = Pt 1
i=1 bi x . Now noti e that
(i) s
i;j

LM(kj gj )  max
it
LM(x i si;j gj )
1 1
( )

 max
it
LM(x i S (gi; gi ))
1 1
( )
+1

 max
it
LM(S (fi; fi ))
1 1
+1

< xÆ ;
sin e the terms in xPÆ int fi and fPi s an el in S (fi; fi ).
+1 +1
Now substitute i ifi = j kj gj in (17), and noti e that f = Psi h0igi
=1 =1 0 =1
where LM(h0igi)P<s xÆ for all i. After olle ting terms in g ; : : : ; gs, we get an 1
expression f = i h00i gi where h00i 2 P and LM(h00i gi) < xÆ for 1  i  s. But
this ontradi ts the denition of xÆ . Thus Bu hberger's Criterion is proved. 
=1

24
10 Bu hberger's algorithm

Bu hberger's S -polynomial riterion a tually gives an algorithm for onstru ting

Grobner bases. Given a monomial ordering on P and a nite set of polynomials
f ; : : : ; fr 2 P , we an onstru t a Grobner basis G for the ideal I = hf ; : : : ; fr i
1 1
by adding S -polynomials to the original list of generators f ; : : : ; fr in a systemati
1
way.
Theorem 10.1 (Bu hberger's Algorithm) Let I = hf ; : : : ; ft i be a nonzero
1
ideal in P = F [x ; : : : ; xn ℄ and let < be a monomial ordering in P . Then a
1
Grobner basis G for I an be onstru ted in a nite number of steps by the
following algorithm.

Input: (f1 ; : : : ; ft ), a list of polynomials

Output: G = (g1 ; : : : ; gs), a Grobner basis G for I = hf ; : : : ; fti
1

G := (f1 ; : : : ; ft )
REPEAT
G0 := G
FOR ea h pair fp; q g, p 6= q , in G0 DO
r := remainder of S (p; q ) on division by G0
IF r 6= 0 THEN G := G [ frg
UNTIL G = G0

Proof First note that if p; q 2 I then S (p; q) 2 I . Initially, the list G is the
given list of generators (f ; : : : ; ft ) for I . As the algorithm pro eeds, we add to
1
G all the remainders r of S -polynomials of pairs of elements in the urrent list
G0 on division by polynomials whi h are already in G0 . These remainders must
also be in I , and so, at every stage of the algorithm, G is a list of generators for
the same ideal I .
The algorithm terminates when all these remainders are 0. It follows imme-
diately from Theorem 9.1 that at this point the list G is a Grobner basis for
I.
It remains to prove that the algorithm does terminate for all possible inputs.
Consider the monomial ideal hLT(G)i generated by the leading monomials of
all the elements of G. The new remainder r ontains no monomials whi h are
divisible by the leading terms of elements of G, so in parti ular LT(r) 62 hLT(G)i.
Thus the sequen e of ideals hLT(G)i whi h arise as the algorithm pro eeds is an
as ending hain. By Theorem 6.6, su h a hain must stop after a nite number
of steps. Hen e the algorithm terminates. 
Example 10.2 In F [x; y ℄ with Lex and x < y , let f = xy x, f = y + x ,
1 2
2

I = hf ; f i.
1 2

25
 We start with g = f , g = f , G = (g ; g ). Then S (g ; g ) = (xy x) +
1 1 2 2 1 2 1 2
x( y + x ) = x + x . Sin e LM( x + x ) = x is not divisible by LM(g ) = xy
2 3 3 3
1
or LM(g ) = y, r = S (g ; g ) = x x in this ase.
2 1 2
3

We add r to the basis: g = x x, and REPEAT. This time S (g ; g ) = 3

3
1 2
x + x (of ourse! | this algorithm is not very eÆ ient). But now the remainder
3

r = 0, sin e we have in luded g as a divisor. However, we still have to deal with 3

S (g ; g ) and S (g ; g ).
1 3 2 3
We get S (g ; g ) = x (xy x) y(x x) = x + xy. The leading monomial
1 3
2 3 3

is xy = LM(g ) so this is NOT the remainder. In fa t it is easy to see that

1
S (g ; g ) = g g , so r = 0.
1 3 1 3
We get S (g ; g ) = x ( y + x ) + y(x x) = x xy. Again division is
2 3
3 2 3 5

possible and we nd x xy = ( xy + x) + (x + 1)(x x) = g + (x + 1)g .

5 2 3
1
2
3
So on e again r = 0.
Sin e no new polynomials have been added in this pass through the REPEAT
loop, the algorithm stops. The resulting Grobner basis is fxy x; y + x ; x xg. 2 3

Example 10.3 As in Example 6.3, let I = hf ; f i, where f = x 2xy , f = 1 2 1

3
2
x y 2y + x, and use the DegLex order in F [x; y ℄ with x > y . Then
2 2

S (f ; f ) = x
(x y 2y + x) y
(x 2xy ) = x ;
2 1
2 2 3 2

and we see that the remainder is still x , so, as expe ted from Example 6.3, we 2

must in lude x in the generating set, giving g = x 2xy, g = x y 2y + x,

2
1
3
2
2 2

g =x .
3
2

Now we REPEAT starting with G = (g ; g ; g ). 1 2 3

This time S (g ; g ) = (x 2xy) x(x ) = 2xy, and S (g ; g ) = (x y
1 3
3 2
2 3
2

2y + x) y(x ) = 2y + x. The leading terms are not divisible by any of

2 2 2

LM(g ); LM(g ); LM(g ), so we take g = 2xy, g = 2y + x. One more run

1 2 3 4 5
2

through the REPEAT loop shows that we nally have a Grobner basis
fx 2xy; x y 2y + x; x ; 2xy; 2y + xg:
3 2 2 2 2

The algorithm in its most basi form is not very eÆ ient. In pra ti e it is better
to add the new remainders to G one at a time. Then we only have to he k the
S -polynomials involving the urrent set G and the new remainder to be added
to it. In Example 10.3, this gives the following sequen e of al ulations.
S (g ; g ), getting g :
1 2 3

S (g ; g ), getting g ; S (g ; g ), getting g :
1 3 4 2 3 5

S (g ; g ), S (g ; g ), S (g ; g ), getting no new polynomials:

1 4 2 4 3 4

S (g ; g ), S (g ; g ), S (g ; g ), S (g ; g ), getting no new polynomials:

1 5 2 5 3 5 4 5

DONE!

26
11 Minimal and redu ed Gr
obner bases

A nonzero polynomial ideal I has innitely many Grobner bases, sin e we an add
any element of I to an existing Grobner basis to get a new one, and we an also
multiply any of the polynomials in the Grobner basis by a nonzero s alar. Thus
we are interested in making our Grobner bases as small (and tidy) as possible. In
parti ular, the Grobner basis G output by Bu hberger's algorithm may ontain
`redundant' generators gi whose leading term LT(gi) is ontained in the ideal
generated by the leading terms of the other elements of G. Thus in Example
10.3 we an drop out g and g from G, sin e their leading terms x and x y
1 2
3 2

are divisible by the leading term x of g . This gives a smaller Grobner basis
2
3
fx ; 2xy; 2y + xg.
2 2

Denition 11.1 A Grobner basis G = fg ; : : : ; gsg is minimal if and only if

1

(i) ea h gi is moni , i.e. LC(gi ) = 1 for all i, and

(ii) LT(gi ) does not divide LT(gj ) if i 6= j .
If (ii) is repla ed by the stronger ondition
(iii) LT(gi) does not divide any term of gj if i 6= j ,
we say that G is a redu ed Grobner basis.

Thus in Example 10.3 the Grobner basis fx ; 2xy; 2y + xg leads us easily 2 2

to the minimal Grobner basis fx ; xy; y xg, whi h is in fa t redu ed. In the
2 2 1

same way, the Grobner basis fxy x; y + x ; x xg of Example 10.2 leads to

2
2 3

the redu ed Grobner basis fy x ; x xg. 2 3

We now have the following amazing result.

Theorem 11.2 Every nonzero polynomial ideal I in F [x ; : : : ; xn ℄ has a unique 1
redu ed Grobner basis with respe t to a given monomial ordering.
Proof First we prove existen e of a redu ed basis. It is lear how to
obtain a minimal Grobner basis: starting from any Grobner basis, we rst omit
any gi su h that LM(gi) is divisible by LM(gj ) for some j 6= i, and then divide
ea h remaining gi by LC(gi) to make it moni .
Thus let G = fg ; : : : ; gsg be a minimal Grobner basis for I . The idea is to
1
work through this basis, repla ing ea h element by its remainder with respe t
to the other elements of the ( urrent) basis. We start by repla ing g by h , 1 1
its remainder with respe t to fg ; : : : ; gsg, then repla e g by h , its remainder
2 2 2
with respe t to fh ; g ; : : : ; gsg, and so on until we nally repla e gs by hs, its
1 3
remainder with respe t to fh ; h ; : : : ; hs g. We laim that this results in a
1 2 1
redu ed basis fh ; : : : ; hsg.
1

27
 To see this, noti e that this redu tion pro ess does not hange the leading
terms, i.e. LT(hi) = LT(gi). Hen e we have a minimal Grobner basis at ea h
stage of the redu tion. Finally note that, in the division algorithm, any term
whi h appears in the remainder is not divisible by the leading monomials of any
of the divisors.
Next we prove uniqueness. Thus let G = fg ; : : : ; gsg and H = fh ; : : : ; htg
1 1
be two redu ed Grobner bases for I . We rst show that s = t and that the
leading terms of the two bases are the same, i.e. we an renumber the h's so that
LM(gi) = LM(hi) for all i. Consider g . Sin e g 2 I and H is a Grobner basis
1 1
for I , there is an i su h that LM(hi) divides LM(g ). But sin e hi 2 I and G
1
is a Grobner basis for I , there is a j su h that LM(gj ) divides LM(hi). Hen e
LM(gj ) divides LM(g ). Sin e G is minimal, it follows that j = 1 and hen e
1
LM(hi ) = LM(g ). Renumber H so that hi be omes h . Now onsider g . In
1 1 2
the same way there is an i su h that LM(hi ) divides LM(g ). This annot be h ,
2 1
sin e LM(h ) = LM(g ) and LM(g ) does not divide LM(g ), by minimality of G.
1 1 1 2
Now LM(gj ) divides LM(hi) for some j , and again minimality of G implies that
j = 2. Hen e LM(g ) = LM(hi ), and we renumber Hi as h . It is lear that this
2 2
pro ess an be ontinued until all the elements of G and H are paired o.
Finally we must prove that gi = hi for 1  i  s. Let f = gi hi , and assume
f 6= 0 so that LM(f ) is a monomial. Sin e we have proved that LT(gi ) = LT(hi ),
we have LM(f ) < LM(gi). Sin e f 2 I and G is a Grobner basis for I , LM(gj )
divides LT(f ) for some j . Clearly j 6= i sin e LM(f ) < LM(gi). But LM(gj ) =
LM(hj ) does not divide any term of gi or hi, sin e the Grobner bases G and H
are both redu ed. Hen e LM(gj ) does not divide any term of f , and in parti ular
it does not divide LT(f ). This is a ontradi tion. Hen e f = 0. 
This result gives a way to test whether two sets of polynomials are generators
for the same ideal. We need only ompute a redu ed Grobner basis starting from
ea h set in turn. The ideals will be equal if and only if the redu ed Grobner bases
are the same set.
Many omputer algebra systems (e.g. Maple, Mathemati a) an be used
to ompute Grobner bases. These always return the redu ed Grobner basis with
respe t to a given monomial ordering (ex ept that, to avoid fra tions, the basis
may not be moni ). Thus it is easy to ompare results obtained using dierent
systems. These systems use various implementations of (renements of) Bu h-
berger's algorithm. Be ause of the wide appli ations of Grobner bases in algebra
and geometry, these pa kages are useful to s ientists and engineers as well as to
mathemati ians.
It is worth mentioning that when the given generating set for I onsists of
linear polynomials, the redu ed Grobner basis orresponds to the `redu ed row-
e helon form' of the oeÆ ient matrix of the orresponding system of linear equa-
tions.

28
12 Appli ations of Gr
obner bases

12.1 The ideal membership problem

Problem: Given an ideal I = hf ; : : : ; fs i in P , how an we de ide 1
whether a given polynomial f lies in I?
Here is how.
1. Choose a monomial ordering < in P .
2. Compute a Grobner basis G for I with respe t to <.
3. Find the remainder r of f on division by G.
Then f 2 I if and only if r = 0.
Example 12.1 Determine whether (i) x y x and (ii) x y + y are in the 2 4 6 4 2 2

ideal I of F [x; y℄ generated by xy x and y + x . 2

In Example 10.2 we have already al ulated a Grobner basis of I with respe t

to the Lex order with x < y. We also noted above that G = fg ; g g, where 1 2
g = y x , g = x x is a redu ed Grobner basis in this ase, so we may as
1
2
2
3

well use it. (But the method works with any Grobner basis.)
Using the division algorithm, we nd that
x y x = (x y + x y + x y + x )(y x ) + (x + x )(x x):
2 4 6 2 3 4 2 6 8 2 7 5 3

The remainder r = 0. We on lude that x y x 2 I . 2 4 6

A similar al ulation gives

x y + y = (x y + y + x + x )(y x ) + (x + x + 2x)(x x) + 2x ;
4 2 2 4 6 2 2 5 3 3 2

so that the remainder r = 2x is nonzero. We on lude that x y + y 62 I .

2 4 2 2

Noti e that in both ases we have arranged the terms of the quotients q and 1
q in de reasing Lex order.
2

There is another way to think about the ideal membership problem. If we have a
set of equations f = 0; : : : ; fs = 0 orresponding to the generators of the ideal I ,
1
we an ask whether the equation f = 0 is a onsequen e of these. If f 2 I , then
f = q f + : : : + qs fs = 0 when all fi = 0. In Example 12.1, the equations xy = x
1 1
and x = y imply x y x = (xy)
y x = x
(x ) x = 0. Grobner bases
2 2 4 6 2 2 6 2 2 2 6

provide a way of doing al ulations of this kind systemati ally, so that they an
be done by a omputer.

29
12.2 The elimination problem
Problem: Given a set of polynomial equations f1 = 0, . . . , ft = 0 in
F [x1 ; : : : ; xr ; y1; : : : ; ys℄, how an we eliminate the variables x1 ; : : : ; xr from
these equations so as to obtain equations involving only y1 ; : : : ; ys?
To get an idea of what is involved here, we onsider a simple example.
Example 12.2 Consider the equations x = t , y = t , z = t . We an think of
4 3 2

these as parametri equations of a urve C in R . Elimination of the parameter

3

t between pairs of equations gives equations x = z , y = z , x = y . Geometri-

2 2 3 3 4

ally, ea h of these 3 equations represents a surfa e whi h ontains the urve C .

Noti e also that we an eliminate z from the equations x = z , y = z to get the 2 2 3

equation x = y .
3 4

From the point of view of ideal theory, the elimination problem is equivalent
to looking for polynomials in the ideal hf ; : : : ; ft i involving only the variables
1
y ; : : : ; ys. In Example 12.2, it is easy to he k that x z , y x and x y
1
2 2 3 3 4

are in the ideal I = hx t ; y t ; z t i, e.g. x z = (x t ) (z + t )(z t ),

4 3 2 2 4 2 2

y z = (y + t )(y t ) (z + zt + t )(z t ).
2 3 3 3 2 2 4 2

If we ompute the redu ed Grobner basis of I with respe t to Lex order with
t > x > y > z , we obtain
ft z; ty z ; tz y; x z ; y z g:
2 2 2 2 3

Noti e that the polynomials here whi h do not involve t are x z and y z . 2 2 3

Sin e t was hosen as the largest variable in the ordering, t is eliminated rst in
omputing the Grobner basis.
Theorem 12.3 Let I be an ideal in F [x ; : : : ; xr ; y ; : : : ; ys℄, let G be a Grobner
1 1
basis of I with respe t to Lex order with xi > yj for all i; j , and let G0 = G \
F [y ; : : : ; ys℄. Then G0 is a Grobner basis of I 0 = I \ F [y ; : : : ; ys℄.
1 1

Proof Clearly G0  I 0 , so we need only prove that if f 2 I 0 then LT(f ) is

divisible by LT(g) for some g 2 G0. Sin e f 2 I and G is a Grobner basis for I ,
LT(f ) is divisible by LT(g) for some g 2 G. Sin e f 2 F [y ; : : : ; ys℄, it follows 1
that LT(g) 2 F [y ; : : : ; ys℄. Now sin e we are using Lex order with xi > yj for all
1
i; j , any monomial involving x's is > any monomial involving only y 's. Hen e g
annot have any term involving the x's, i.e. g 2 G0. 

30
12.3 The ideal interse tion problem
Problem: Given ideals I = hg ; : : : ; gk i and J = hh ; : : : ; hl i in F [x ; : : : ; xn ℄,
1 1 1
nd a generating set for the ideal I \ J.
We saw in Se tion 3.3 how to solve this problem in the ase where I and J
are monomial ideals. The elimination theorem (Theorem 12.3) provides a tri k
to solve this problem in general, by introdu ing a dummy variable whi h we then
eliminate.
Theorem 12.4 Given ideals I and J in P = F [x ; : : : ; xn ℄, let t be a new vari-
1
able. Then I \ J = K \ P , where K is the ideal htI; (1 t)J i in P [t℄.
Proof Given f 2 I \ J , by writing f in the form tf + (1 t)f we see that
f 2 tI +(1 t)J . Hen e f 2 K \ P . Conversely, if f 2 K \ P , then we an write
f in the form
Xk l
X
f= tgi pi + (1 t)hj qj ;
i=1 j =1
where I = hg ; : : : ; gk i and J = hh ; : : : ; hl i, and the pi's and qj 's are polynomials
1 1
in P [t℄.
This is a polynomial identity and so it remains true when we spe ialise the
variables to elements of F . In parti ular, setting t = 1 we obtain
k
X
f = gi pi (x1 ; : : : ; xn ; 1) 2 I;
i=1
and t = 0 gives
l
X
f = hj qj (x1 ; : : : ; xn ; 0) 2 J;
j =1
so that f 2 I \ J . 
This method in fa t gives a Grobner basis for I \ J , starting from any gener-
ating sets for I and J .
Example 12.5 Find a Grobner basis for I \ J in F [x; y ℄, where I = hx + y ; xy i 2 2

and J = hx y i.
2 2

The ideal K = htI; (1 t)J i  F [x; y; t℄ is given by

K = ht(x + y ); txy; (1 t)(x y )i:
2 2 2 2

Using Lex order with t > x > y, a Grobner basis for K is

f2tx x + y ; txy; 2ty + x y ; x xy ; x y y g:
2 2 2 2 2 2 3 2 2 3

By the Elimination Theorem 12.3, it follows that fx xy ; x y y g is a Grobner

3 2 2 3

basis for K \ F [x; y℄. By Theorem 12.4, K \ F [x; y℄ = I \ J .

31
MT30512 Polynomials Semester 2, 2006{07

Chapter 3: Symmetri Polynomials

13 Elementary symmetri fun tions

Denition 13.1 A polynomial f 2 P = F [x1 ; : : : ; xn ℄ is symmetri if and only

if f is invariant under any permutation of x1 ; : : : ; xn , i.e. if  : f1; 2; : : : ; ng !
f1; 2; : : : ; ng is any bije tion, then f (x(1) ; : : : ; x(n) ) = f (x1 ; : : : ; xn).
The following examples of symmetri polynomials will be important.
 the rth elementary symmetri fun tion
X
er = xi xi


xi1 2 r

i1 <i2 <:::<ir

is the sum of all produ ts of the n variables x ; : : : ; xn taken r at a time.

1
In parti ular, e = x + : : : + xn is the sum of all the variables, and en =
1 1
x x


xn is their produ t.
1 2

 the rth omplete symmetri fun tion

X
hr = x x


xn
1
1
2
2 n

1 +2 +:::+n =r

is the sum of all monomials of total degree r in the n variables x ; : : : ; xn. 1

 the rth power sum

pr = xr + : : : + xrn
1

is the sum of the rth powers of the variables.

Other symmetri polynomials are easily onstru ted by noting that the sum,
dieren e and produ t of two symmetri polynomials are again symmetri . Thus
the set of all symmetri polynomials in the variables x ; : : : ; xn is a subring S of
1
the polynomial ring P .
Symmetri polynomials arise naturally when studying the roots of a polyno-
mial. For example if the ubi polynomial f (x) = x + ax + bx + has roots 3 2

; ; , then
x + ax + bx + = (x )(x  )(x )
3 2

and equating oeÆ ients we get

 +  + = a;  +  +  = b;  = :

32
Thus the oeÆ ients of f (x) an be regarded as polynomials in the `variables'
; ; . If we permute the roots in any way, e.g. ex hange  and  , we do not
hange the polynomial f (x) and so the polynomials  +  + ,  +  +  ,
 are also un hanged. In general, if the moni polynomial
f (x) = xn + an xn + : : : + a x + a 1
1
1 0

fa tors as the produ t

f (x) = (x x )(x x )


(x xn ); 1 2

then the relations an r = ( 1)r er hold for 1  r  n, and express ea h oeÆ ient
an r of the polynomial f as (plus or minus) a orresponding elementary symmetri
fun tion er of its roots x ; : : : ; xn.
1

Theorem 13.2 (Fundamental Theorem of Symmetri Polynomials) Ev-

ery symmetri polynomial in F [x1 ; : : : ; xn ℄ an be written uniquely as a polynomial
in the elementary symmetri fun tions e1 ; : : : ; en . Thus S = F [e1 ; : : : ; en℄.

Proof We use DegLex order with x > x >


> xn . Let f 2 P be a 1 2

nonzero symmetri polynomial, and let LM(f ) = x = x x


xn . Note that 1 2 n

   


 n ; if i < i for any i, then we an obtain a higher monomial

1 2

1 2 +1
by swit hing the variables xi and xi . Sin e f is symmetri , this monomial will
+1
also be a term of f. 
Now let g = e e


en be a `monomial' in the `variables' e ; e ; : : : ; en.
1 2 n
1 2
Sin e LM(er ) = x x


xr ,
1 2

1 2

LM(g) = x (x x:::)


(x:::


xn) 

1
1 2
2
1
n

= x
1
x 1
1+ 2+ +



xn : n
2
2+ + n n

Thus LM(g) = LM(f ) if and only if r = r + r + : : : + n for 1  r  n, i.e +1

if and only if r = r r for r < n and n = n.
+1
Now let f = f ag, where g is determined as above and a = LC(f ). Then
1
f is again a symmetri polynomial, and if f 6= 0 then LM(f ) < x , sin e the
1 1 1
terms in x an el. If f = 0 we are done, otherwise we repeat the pro edure with
1
f . Sin e there are only a nite number of monomials < x in DegLex order, the
1
pro ess must eventually stop, and we obtain an expression for f as a polynomial
in the elementary symmetri fun tions.
The uniqueness of this expansion follows immediately from what we have
already proved: the `monomials' e = e e


en in the `variables' e ; : : : ; en
   1 2 n
1
all have distin t leading terms when expressed as polynomials in x ; : : : ; xn, and
1 2

1
hen e they are linearly independent over the eld of oeÆ ients F . 

33
14 Families of Symmetri Fun tions

14.1 Monomial symmetri fun tions

The proof of Theorem 13.2 gives an algorithm for omputing a given symmetri
polynomial f as a polynomial in the elementary symmetri fun tions. For pra -
ti al al ulations, it is onvenient to group together all monomials in f with the
same set of exponents, as follows.
Denition 14.1 Let  = ( ; : : : ; n) where  


 n  0. Then the
1 1
monomial symmetri fun tion m is the sum
X
m = x


of all monomials whose list of exponents  = ( ; : : : ; n) is a permutation of

1
 = (1 ; : : : ; n ).
Example 14.2 If the variables are x, y, z, then
m ; ; = x y +x y +x z +x z +y z +y z ;
(3 2 0)
3 2 2 3 3 2 2 3 3 2 2 3

m ; ; = x yz + xy z + xyz ;
(3 1 1)
3 3 3

m ; ; = x y z + x yz + xy z :
(2 2 1)
2 2 2 2 2 2

It is easy to see that m is symmetri , and that every symmetri polynomial an

be uniquely expressed as f = P m ,  2 F . Thus the monomial symmetri
fun tions form a basis for the ring of symmetri fun tions S , regarded as a ve tor
spa e over F . It follows from Theorem 13.2 that the monomials e = e


en 1 n

in the elementary symmetri fun tions form another ve tor spa e basis for S .
1

Example 14.3 Let f = m ; ; = x y + x y + x z + x z + y z + y z . to

(3 2 0)
3 2 2 3 3 2 2 3 3 2 2 3

express f in terms of e ; : : : ; en, we rst note that  = (3; 2; 0) )  = (1; 2; 0),

1
and so e = e e . Expanding e in terms of monomial symmetri fun tions, we
2
1 2
have
e e = (x + y + z )(xy + xz + yz ) = m ; ; + 2m ; ; + 5m ; ; :
2
1 2
2
(3 2 0) (3 1 1) (2 2 1)

Then LM(f e e ) = LM(m ; ; ) = x yz, so the new  = (3; 1; 1), the new
2
1 2 (3 1 1)
3

 = (2; 0; 1), and the new e = e e . Then 2

1 3

e e = (x + y + z )
xyz = m ; ; + 2m ; ; :
2
1 3
2
(3 1 1) (2 2 1)

Noting that m ; ; = x y z + x yz + xy z = e e , we nally obtain

(2 2 1)
2 2 2 2 2 2
2 3

f = e e 2(e e 2e e ) 5e e = e e 2e e e e :
2
1 2
2
1 3 2 3 2 3
2
1 2
2
1 3 2 3

34
14.2 Complete symmetri fun tions
We shall prove that the omplete symmetri fun tions h ; : : : ; hn provide another 1
set of polynomial generators for the ring of symmetri fun tions S . The following
identities provide a re ursive pro edure for expressing ea h hk as a polynomial in
e ; : : : ; en . For this we extend the denition of er by dening e = 1 and er = 0
1 0
for r > n, where n is the number of variables.
Proposition 14.4 For 1  r  n and i  0, let hi be the ith omplete symmetri
fun tion in the variables x1 ; : : : ; xr and let ei be the ith elementary symmetri
fun tion in the variables x1 ; : : : ; xn . Then
X
( 1)ieihj = 0; for k > n r:
i+j =k

In parti ular, when r = n this identity holds for all k > 0.

Proof We introdu e a new variable t, and onsider the generating fun tions
n
X 1
X
E (t) = ( 1) i e ti ;
i H ( t) = hj tj :
i=0 j =0

Note that E (t) is a polynomial in P [t℄, but H (t) is a formal power series in t
(with oeÆ ients in P ).
As we noti ed at the beginning of Se tion 13, the elementary symmetri fun -
tions e ; : : : ; en (with alternating signs) are the oeÆ ients of the polynomial with
1
roots x ; : : : ; xn. Hen e
1

E (t) = (1 x t)(1 x t)


(1 xn t):

1 2

On the other hand

H (t) = 1 + h t + h t + : : :
1 2
2

= (1 + x t + x t + : : :)(1 + x t + x t + : : :)


(1 + xr t + xr t + : : :)
2 2 2 2 2 2

= (1 1x t)
(1 1x t)


(1 1x t) :
1 1 2 2

1 2 r

Hen e E (t)H (t) = (1 xr t)


(1 xn t), a polynomial of degree n r. But if we

+1
expand the produ t E (t)H (t) in powers of t, the oeÆ ient of tk is Pi j k ( 1)ieihj ,
so this sum is zero for k > n r. 
+ =

Theorem 14.5 Every symmetri polynomial in F [x ; : : : ; xn ℄ an be expressed 1

uniquely as a polynomial in the omplete symmetri fun tions h ; : : : ; hn . Thus 1
S = F [h ; : : : ; hn ℄.
1

35
Proof We an solve the system of equations in Proposition 14.4 with r = n to
express ei as a polynomial in h ; : : : ; hi for 1  i  n. Hen e we an express any
1
polynomial in e ; : : : ; en as a polynomial in h ; : : : ; hn. Hen e by Theorem 13.2
1 1
every symmetri polynomial an be written as a polynomial in h ; : : : ; hn. 1
To prove uniqueness, we use the fa t that the monomials e in e ; : : : ; en of 1
total degree d in x ; : : : ; xn form an F -ve tor spa e basis for the homogeneous
1
symmetri polynomials S d of total degree d in x ; : : : ; xn. Sin e we have seen that
1
the monomials h = h


hn in S d form a spanning set, and sin e there are
1 n

the same number of h 's as e 's in S d, the h 's are also a F -ve tor spa e basis.
1


14.3 Power sums and Newton's identities
The rth power sum pr = xr + : : : + xrn was introdu ed in Se tion 13. For some
purposes, it turns out to be useful to express symmetri fun tions as polynomials
1

in the pr 's rather than the er 's or the hr 's. However, sin e denominators arise
in the oeÆ ients, we must work over a eld F of hara teristi zero. First we
obtain relations between the pr 's and the er 's.
Proposition 14.6 (Newton's identities) For r  1,
r
X
rer = ( 1)k 1
pk er k :
k=1

Proof Let P (t) be the formal power series

P (t) = p + p t + : : : + pr tr + : : :
1 2
1

Xn
= (xi + xi t + : : : + xri tr + : : :)
2 1

i=1
n
X
= xi (1 + xi t + : : : + (xi t)r 1
+ : : :)
i=1
n
X xi
= 1 xi t
:
i=1

As in Se tion 13, let E (t) = 1 ePtn+ e t + : : : + ( 1)nentn = (1 x t)(1

1 2
2
1
x t)


(1 xn t). Then log E (t) = i log(1 xi t). Dierentiation with respe t
2
to t gives
=1

E 0 ( t) Xn
xi
E (t)
= 1 x t = P ( t) :
i=1 i
Hen e P (t)E (t) = E 0 (t) = e 2e t + : : : + ( 1)n nen tn . Equating the
1 2
1 1

oeÆ ients of tr in this equation gives Newton's identities: note that these are
1

valid for all r  1 if we let er = 0 for r > n. 

36
Theorem 14.7 If har(F ) = 0 then every symmetri polynomial in F [x1 ; : : : ; xn ℄
an be written uniquely as a polynomial in the power sums p1 ; : : : ; pn , i.e. S =
F [p1 ; : : : ; pn ℄.
Proof We prove by indu tion on r that er is a polynomial in p1 ; : : : ; pr . This
holds for r = 1 sin e e1 = p1 . The indu tive step is given by Newton's identity.
Note that we an divide by r in F , sin e har(F ) = 0. It follows from Theorem
13.2 that every symmetri polynomial is a polynomial in p1; : : : ; pr . Uniqueness
is proved as for the omplete symmetri fun tions (see Theorem 14.5). 
14.4 Determinantal Formulae
The above relations between the elementary, omplete and power sum symmetri
fun tions an be solved so as to express one of these fun tions as an expli it
polynomial in the members of one of the other families. Thus Proposition 14.4 is
equivalent to either of the determinantal formulae
   
 h
1 h h


hn 
2 3
 e
 e e


en 
1 2 3
 1 h h


hn   1 e e


en 
 1 2 1 1 2 1

en =  0 1 h


hn  ; hn =  0 1 e


en  :
 . ... . . . . . . ...   . ... . . . . . . ... 
1 2 1 2

 ..  ..
 0 0 0


h   0 0 0


e 
 
1 1

Similarly, Newton's identities 14.6 an be written

   
 e
 1 2 e
2 3 e
3

ne 
n 
 p
 p 1

2
p n 1 p n


 1 e e


e   1 p


p p 
 1 2 n 1  1 n 2 n  1
 0 1 e


en  ; n!en =  0 2


pn pn  :
pn =   
 . ... . . . . . . ...   . ... . . . . . . ... 
1 2 3 2

 ..  ..
 0 0 0


e   0 0


n 1 p 
 
1 1

The equivalen e with Propositions 14.4 and 14.6 an be seen by using suitable
row or olumn expansions of the determinants and indu tion. There are similar
formulae relating the h's and p's (Examples 7, Question 6.) Although these
expli it formulae are elegant, in pra ti e it is easier to use the re ursive formulae
of Propositions 14.4 and 14.6 dire tly in al ulations.

37
15 Invariants and Coinvariants

In this se tion we onsider the ideal I = he ; : : : ; eni generated by the elementary

1
symmetri polynomials in P = F [x ; : : : ; xn ℄, and the quotient ring P=I . In view
1
of Theorem 13.2, we may equivalently dene I as the ideal generated by all
symmetri polynomials with zero onstant term. There are many polynomials in
I whi h are not symmetri : for example, the polynomial identity
xni e xin + e xin + : : : + ( 1)n en = 0
1
1
2
2
(18)
shows that the nth power xni of every variable xi is in I . This identity follows
from the relation between the oeÆ ients of a polynomial and its roots whi h we
noti ed in Se tion 13.
In lassi al invariant theory, the ring S is alled the ring of invariants of the
symmetri group Sn onsisting of all permutations of the variables x ; : : : ; xn . 1
The ring P=I is alled the ring of oinvariants. By Theorem 14.5, the omplete
symmetri fun tions h ; : : : ; hn also generate I .
1
We have seen that a Grobner basis for I is helpful in understanding the stru -
ture of P=I . Unfortunately, none of the bases e ; : : : ; en, h ; : : : ; hn or p ; : : : ; pn
1 1 1
is a Grobner basis for I for any monomial ordering. This follows from our obser-
vation above that xni 2 I for all variables xi. A Grobner basis for I must therefore
ontain a polynomial gi whose leading term divides xni and is thus a power of xi .
If we order the variables so that x > x > : : : > xn , then LM(er ) = x x


xr
1 2 1 2
and LM(hr ) = LM(pr ) = xr , so these hoi es do not work.
1

Theorem 15.1 For the monomial order DegLex with xn >


> x > x , 2 1

H = fh (x ; : : : ; xn ); h (x ; : : : ; xn ); : : : ; hn (x ; x ); hn (x )g
1 1 2 1 1 1 1 2 1

is a (redu ed) Grobner basis of I .

For example, when n = 3 and x > y > z, this basis is fx + y + z; y + yz + z ; z g. 2 2 3

Proof Note that the leading term of the polynomial hn r (x ; : : : ; xr ) in +1 1

H is xnr r . Thus ea h leading monomial in H is a power of a dierent variable,
+1

so that the leading monomials are all oprime. Thus if we an show that H is
a basis of I , it will follow from our work on Bu hberger's algorithm that all S -
polynomials of pairs of elements of H redu e to zero, and hen e H is a Grobner
basis for S . It is also easy to see that H is redu ed.
Note that hn(x ) = xn is in I by equation (18), and h (x ; : : : ; xn ) = x +
1 1 1 1
: : : + xn is learly in I sin e it is symmetri . To prove that the other elements of
1

H are in I , we use the identity of Proposition 14.4:

X
( 1)i ei(x ; : : : ; xn)hj (x ; : : : ; xr ) = 0 for k > n r:
1 1 (19)
i+j =k

38
 By putting k = n r + 1 in (19), we see that hn r (x ; : : : ; xr ) is an ele-
+1 1
ment of the ideal he ; e ; : : : ; en r i. Hen e hH i  I . Conversely, by solving
1 2 +1
the equations (19) re ursively, we an express en r as an element of the ideal
+1
hh (x ; : : : ; xn); h (x ; : : : ; xn ); : : : ; hn r (x ; : : : ; xr )i. Sin e e ; e ; : : : ; en gen-
1 1 2 1 1 +1 1 1 2
erate I , it follows that I  hH i. Hen e hH i = I . 
Theorem 15.1 allows us to nd a normal form for any polynomial in P in the
ring of oinvariants P=I , and to perform al ulations in this ring. Thus we obtain
a basis for P=I as a ve tor spa e over F , as follows.
Theorem 15.2 (Artin basis for the ring of oinvariants) The osets of the
monomials x = x1 1 : : : xn su h that r  n r form a F -ve tor spa e basis for
n

the ring of oinvariants P=I of the symmetri group in P = F [x1 ; : : : ; xn ℄.

Proof Sin e the leading terms of the elements of the Grobner basis H of
Theorem 15.1 are xn ; xn ; : : : ; xn, a monomial x 2 P is divisible by one of
2

these leading terms if and only if r  n r + 1 for some r. Thus the result
1 1

follows from Theorem 7.4(ii). 

This result shows that the ring of oinvariants has dimension n! as a ve tor
spa e over F . This is equal to the order of the symmetri group Sn. We an
al ulate eÆ iently in P=I by using the elements of our Grobner basis as relations.
For example in the ase n = 3 we have relations x + y + z = 0, x + xy + y = 0, 2 2

x = 0 (where x is the oset x = x + I 2 P=I , and similarly for y and z ).

3

The ring P=I has further interesting stru ture whi h we des ribe brie y. By
onsidering the oinvariants of a xed degree d, we obtain a ve tor spa e on
whi h the symmetri group Sn a ts by permutation of the variables x ; : : : ; xn. 1
These give important examples of matrix representations of the group Sn:
with ea h permutation in Sn we asso iate the matrix whi h des ribes its a tion
on the Artin basis elements. In parti ular, the a tion of Sn on the whole ring P=I
gives the regular representation of Sn. These ideas are developed in group
representation theory.

39
16 Alternating Polynomials

In this se tion we onsider an important family of polynomials whi h are not

symmetri but are nearly so, in the sense that they hange sign when two of the
variables are ex hanged. The key example of su h a polynomial is the Vander-
monde determinant
1 1


1
 
 
 

 x1 x2


xn 


n = . 
x21 x22 :


xn
2 

..



...
. . . ... 



x

n x 1


xn
1 n
2
1 n 1 

Su h polynomials are said to be alternating. Throughout Se tion 16, we

assume that the hara teristi of the eld F is not 2, so that f 6= f
for f 2 P .
Every permutation of x ; : : : ; xn an be obtained by a nite sequen e of ex-
1
hanges of two variables. Hen e every permutation  2 Sn transforms an alter-
nating polynomial f either into f or into f . Noti e that the produ t of two al-
ternating polynomials is a symmetri polynomial, sin e if f and g are transformed
into f and g by some ex hange, then fg is transformed into ( f )( g) = fg.
Thus fg is xed by an ex hange of two variables, and so it is xed by any nite
sequen e of ex hanges. Likewise the produ t gh of a symmetri polynomial g
and an alternating polynomial h is alternating. Thus we an obtain a plentiful
supply of alternating polynomials by multiplying
n by symmetri polynomials.
We next prove that all alternating polynomials are obtained in this way.
We an sometimes nd fa tors of polynomials using the Remainder Theorem.
This theorem is usually found in textbooks only for polynomials in one variable
over a eld, but we need something stronger. We begin by proving a stronger
version of the division theorem, Theorem 1.1.
Theorem 16.1 Let R be a ommutative ring with 1, and let f; g 2 R[x℄, where
the leading oeÆ ient of g is invertible in R. Then there are unique polynomials
q and r in R[x℄ su h that (i) f = qg + r, and (ii) either deg r < deg g or r = 0.
Proof We use indu tion on the degree of f . Let f = nr ar xr , where
P

an 6=P0, so that deg f = n  0. (If f = 0, we simply take q = r = 0.) Let

=0

g= m r br x , where bm 6= 0, so that deg g = m  0. If m > n we simply take

r
q = 0, r = f , so we may assume that m  n.
=0

For the base of indu tion, let n = 0. Then m = 0 and f and g are onstant
polynomials, i.e. nonzero elements of R. We have f = a , g = b and f = qg 0 0
where q = b a 2 R sin e b is invertible in R.
1
0 0
Now let n > 0, and assume as indu tion hypothesis that the result holds for
0

any polynomial f of degree < n. In parti ular we may take f = bm f an xn m g,

1 1

40
sin e learly deg f  n and the oeÆ ient of xn in f is bm an anbm = 0,
1 1
so that in fa t deg f < n. Hen e the indu tion hypothesis applies to f , to
1 1
give f = q g + r where q ; r 2 R[x℄ and deg r < m or r = 0. Hen e bm f =
1 1 1 1 1 1 1
an xn m g + f = (an xn m + q )g + r . Hen e f = qg + r, where q = bm (an xn m + q )
1 1 1
1
1
and r = bm r , and deg r = deg r < m or r = 0, as required.
1
1 1
Uniqueness is proved by the same argument as when R is a eld (exer ise!).
Theorem 16.2 (Remainder Theorem) Let R be a ommutative ring with 1
and let a 2 R. Then a polynomial f (x) 2 R[x℄ is divisible by x a if and only if
f (a) = 0, i.e. if and only if a is a zero (or root) of f (x).
Proof By Theorem 16.1 we an divide an arbitrary polynomial f (x) 2 R[x℄
by the moni polynomial g(x) = x a.
We now use a standard argument: we have f (x) = q(x)(x a) + r where the
remainder r is onstant, i.e r 2 R. Substitute x = a to get r = f (a). Thus f (x)
is divisible by x a i f (a) = 0. 
As a rst appli ation, we obtain the well-known fa torisation of the Vander-
monde determinant
n. If xi = xj , then
n = 0 sin e the determinant has two
equal olumns. By Theorem 16.2 it follows that
n is divisible by xi xj for all
i; j with 1  i < j  n. Hen e
Y

n = (xj xi):
i<j n
1

Proposition 16.3 Every alternating polynomial h is divisible in F [x1 ; : : : ; xn ℄

by the Vandermonde determinant
n .
Proof Sin e
n = i<j (xj xi ), h is divisible by
n if and only if xi xj divides
Q

h for all i 6= j . Sin e h is transformed into h when xi and xj are ex hanged,

h = h when xi = xj . But h = h implies that h = 0, sin e har(F ) 6= 2. By
Theorem 16.2, xi xj is a fa tor of h. 
16.1 The alternating group An
We re all some standard fa ts from group theory. The symmetri group Sn
onsists of all permutations of x ; : : : ; xn. Thus Sn has order n!. A permutation
1
 2 Sn is even if it an be obtained by an even number of ex hanges of two
variables, and odd if an odd number of ex hanges is required. The set of even
permutations is denoted by An. The matrix of the linear transformation xi 7!
 (xi ) has determinant 1 if  is even, 1 if  is odd. Sin e the determinant is a
homomorphism from Sn on to the multipli ative group f1; 1g if n > 1, and An
is the kernel of this homomorphism, An is a subgroup of Sn of order n!. It is 1

alled the alternating group on the n variables x ; : : : ; xn.

2
1

41
 It is easy to show that the ee t of a permutation  2 Sn on an alternating
polynomial f is given by (f ) = f if  is even, (f ) = f if  is odd. Thus
alternating polynomials, as well as symmetri polynomials, are invariant under
the a tion of the alternating group An.
Proposition 16.4 Let f 2 F [x ; : : : ; xn ℄ be invariant under An , i.e.  (f ) = f
1
for all  2 An . Then f an be uniquely expressed as the sum of a symmetri
polynomial and an alternating polynomial.
Proof Clearly a polynomial f of the given form is An-invariant: we prove
the onverse. Let f be invariant under An, and let  be an odd permutation.
Then  2 An, and hen e  (f ) = f . Let g = (f + (f )), h = (f (f )).
2 2 1 1

(Note that we an divide by 2, sin e har(F ) 6= 2.) Then using  (f ) = f we

2 2
2

have (g) = ((f ) + f ) = g, (h) = ((f ) f ) = h.

1 1

Sin e f is An-invariant, so also is (f ), and hen e so are g and h. Now any
2 2

odd permutation an be written as  Æ , where  2 An. Thus  Æ (g) = (g) = g

and  Æ (h) = (h) = h. Hen e g is symmetri and h is alternating, and
f = g + h is in the required form.
To prove uniqueness, suppose that f = g + h where g is symmetri and h
1 1 1 1
is alternating. Then (f ) = (g )+ (h ) = g h . Hen e g = (f + (f )) = g
1 1 1 1 1
1

and h = (f (f )) = h, and so the de omposition of f is unique. 

2
1
1 2

The set A of all An-invariant polynomials is losed under addition and mul-
tipli ation, i.e. it is a subring of F [x ; : : : ; xn℄. Clearly A ontains the Vander-
1
monde determinant
n, and also all symmetri polynomials. Proposition 16.3
now allows us to give a omplete des ription of the ring A.
Theorem 16.5 If har(F ) 6= 2, then
(i) every An -invariant polynomial f 2 F [x ; : : : ; xn ℄ an be uniquely expressed as
1
f = g +
n h, where g and h are symmetri polynomials, and
(ii) the ring A of all An -invariant polynomials is given by
A
= F [e1; : : : ; en;
n℄=h
2n Æ (e1 ; : : : ; en )i
where Æ (e1 ; : : : ; en ) is the polynomial obtained by writing
n 2
in terms of
e1 ; : : : ; en .
Proof Part (i) follows from Propositions 16.4 and 16.3, noting that the
quotient of the `alternating part' of f by
n is a symmetri polynomial h. Part
(ii) follows from (i) if we observe that sin e h
n Æ (e ; : : : ; en )i is a prin ipal
2
1
ideal, every element of the quotient ring has a unique representative of the form
g +
n h, where g and h are polynomials in e ; : : : ; en . (Re all that the moni
1
generator of a prin ipal ideal is the redu ed Grobner basis, for any monomial
ordering.) 
42
16.2 The dis riminant of a polynomial
We an apply these ideas to obtain a ne essary and suÆ ient ondition for a
polynomial in one variable to have equal roots. Thus let f (x) = xn + an xn + 1
1

: : : + a x + a , with ai 2 F and har(F ) 6= 2. Suppose that f (x) = (x x )(x

1 0 1
x )


(x xn ) where the roots x ; x : : : ; xn lie in a suitable extension eld of F .
2 1 2
(This an always be done.) Then f has equal roots if and only if xi = xj for some
i; j with i 6= j . Sin e
n = i<j (xj xi ), f has equal roots if and only if
n = 0,
Q

or equivalently
n = Æ(e ; : : : ; en) = 0. Sin e we may identify the elementary

2
1
symmetri fun tions e ; : : : ; en in the roots x ; : : : ; xn with the oeÆ ients of f
1 1
by er = ( 1)r an r for 1  r  n, this gives the required ondition on f .
In the ase n = 2, for f (x) = ax + bx + we have
= (x x ) = e 4e =
2 2
1 2
2 2
2
( b=a) 4( =a), giving the familiar ondition b 4a = 0 for equal roots. The
1
2 2

polynomial Æ(e ; : : : ; en) (when expressed in terms of the oeÆ ients of f ) is

1
alled the dis riminant of a polynomial f of degree n. Thus the dis riminant
of f (x) = ax + bx + is b 4a .
2 2

The dis riminant of the ubi polynomial x + ax + bx + is a mu h more

3 2

formidable beast:

= a b 4a 4b + 18ab 27 :
2 2 2 3 3 2

This takes a more familiar form if we assume a = 0 (we an always redu e to this
ase by a linear hange of variable). Thus the equation x + bx + = 0 has at least
3

two equal roots if and only if 4b + 27 = 0, or equivalently ( =2) = (b=3) .

3 2 2 3

As an example, b = 3, = 2 gives x 3x + 2 = (x 1) (x + 2).

3 2

16.3 Con lusion

In this ourse we have only s rat hed the surfa e of the vast subje t of symmetri
fun tions, whi h plays an important role in many areas of mathemati s today,
in luding algebrai ombinatori s and group representation theory. There are
also onne tions to applied mathemati s and theoreti al physi s here. A good
textbook for students wishing to know more about this topi is Chapter 1 of
I. G. Ma donald's book Symmetri Fun tions and Hall Polynomials. We have
also hinted at developments in a dierent dire tion, that of the invariant theory
of nite groups. This topi motivated mu h of the development of algebra in the
19th entury, and it has undergone a signi ant revival in re ent years. The book
by Larry Smith, Polynomial Invariants of Finite Groups, is re ommended.

43