An Introduction to the Theory of Resultants

Peter F. Stiller

Introduction
This report is meant to serve as a \working man's" introduction to resultants. It is aimed at
engineers, computer scientists, and applied mathematicians who encounter systems of polynomial
equations in their practice. As such, it is more of a \how to" manual than a theoretical study.
Most of the techniques and results presented here can be found in the classical literature.
Unfortunately, those results are scattered in numerous, sometimes out of print, sources. We have
brought them all together for the practitioner's use. Simple examples illustrate each technique,
and the reader should have little diculty in applying the various methods to his or her particular
problem.
Applicability
Resultants are used to solve systems of polynomial equations, to determine whether or not
solutions exist, or to reduce a given system to one with fewer variables and/or fewer equations.
Input
The typical input will be a system of m equations in n-variables:
f1 (x1 : : : xn ) = 0
..
.
fm(x1 : : : xn) = 0:
Each equation has an associated degree di P 1. Recall that fi(x1 : : : xn ) has degreePdi if all
monomials xe11 xe22 : : : xenn appearing in fi have ni=1 ei
di and at least one monomial has ni=1 ei =
di . Example: f (x1 x2 x3 ) = 3x21 x3 + 4x1 x2 ; x2 + 7x3 ; 1 has degree d = 3. The integers
m n d1 : : : dm are important indicators of the specic resulant that will need to be employed.
Output
There are two essentially dierent cases:
Case 1: m > n (overdetermined) This is the case where we have more equations than unknowns,
and where we generally expect to have no solutions. The resultant will be a system of equations
(one equation when m = n +1) in the symbolic coecients of the fi that has the following property:
when we substitute the specic numerical coecients of the fi , we will get zero in every equation
in the resultant system if and only if the original overdetermined system has a solution.
Case 2: m
n (exact and underdetermined) In this case the number of equations is less than
or equal to the number of variables, and we expect to have solutions. In fact, if we allow complex
solutions and solutions at innity, we are guaranteed to have solutions.
Professor of Mathematics and Computer Science, Texas A&M University, College Station, TX 77843{3363

1
 Of course, only when m = n do we expect a nite number s of solutions. Bezout's Theorem
then provides a count of s = d1 d2 : : : dm solutions (counting complex solutions, solutions at innity,
and counting with appropriate multiplicities). Unfortunately, the possibility also exists (even when
m = n) that there will be an innite number of solutions.
In general, for m
n, the resultant will be the one equation in n ; m + 1 of the variables. In
eect, the resultant makes it possible for us to eliminate m ; 1 of the variables. For example, if we
choose to eliminate xn;m+2 : : : xn , then the resultant R will be a polynomial R(x1 : : : xn;m+1 )
in the remaining variables. If (1 : : : n;m+1 ) is a solution to R = 0, then there will exist values
n;m+2 : : : n such that (1 : : : n;m+1 n;m+2 : : : n ) is a solution to the original system.
(One must be a little careful here. The system should be modied to make it homogeneous with re-
spect to xn;m+2 : : : xn by adding appropriate powers of a variable w. The values n;m+2 : : : n
should be regarded as the coordinates of a point (n;m+2 :    : n : 1) in projective m ; 1 space
Pm;1 . We must allow for the possibility that this point will be at innity, where w = 0. In that
case, a solution to R = 0 would not necessarily give rise to a solution of the original system.)
For example, consider the system of m = 2 equations in n = 3 variables: 4xyz ; 1 = 0 and
y + xz ; 1 = 0. The resultant eliminating z is R(x y) = x(4y2 ; 4y + 1). When x = 0 we will have
R = 0, but clearly our system has no solution when x = 0. However, homogenizing with respect to
z gives the system
4xyz ; w = 0 and (y ; 1)x + xz = 0:
Now when we look at the condition x = 0, we nd that (z w) = (1 : 0) is a solution. This is a
point at innity.
Notice that we also have solutions to R = 0, when x 6= 0 by taking y = 21 . This yields z = 21x .
Geometrically the solution set is a hyperbola in the plane y = 21 in space. The resultant \projects"
that hyperbola to the line y = 12 in the xy-plane, except that (x y) = (0 21 ) is not hit.

z
6

??
??
? -
1
66 2 y
6
66
x

In this context (the underdetermined case) the resultant can be viewed as a projection of the
nominally n ; m dimensional locus of solutions in Rn to an n ; m dimensional locus (hypersurface)
in Rn;m+1 . Note that in our example n = 3, m = 2, and we are projecting the one-dimensional
locus of solutions in R3 to a one-dimensional locus in R2 , where the locus is described by one
equation y ; 21 = 0.

2
Approach in This Paper
We begin with the rst major distinction in methods, namely the one based on the number
of variables n. The case n = 1 of a single variable is discussed in x1. We then move on in x2
to the multivariate case n 2. The nal two sections, x3 and x4, cover various geometric and
combinatorial issues.
Table of Resultants
n m Type of resultant to use Notes

1 2 Determinant of the Sylvester This is what is most commonly

matrix thought of as the resultant.
1 3 Requires a system of equations See the discussion in van der
Waerden ?].
2 m = n + 1 Macaulay resultant This is computed as the quo-
tient of two determinants. It
is a polynomial in the sym-
bolic coecients and is zero if
and only if the system has a
solution.
2 m n + 2 Requires a system of equations See van der Waerden ?].
2 m=n U -resultant or generalized This resultant is designed to
characteristic polynomial nd the nite set of all
solutions to the system of
equations.
2 m<n Macaulay resultant using The result is a single polyno-
m ; 1 variables, while treat- mial in the remaining n ; m +1
ing the other n ; m + 1 variables.
variables as included in the
coecients.
Note: One can also employ the standard Sylvester resultant in the multivariate case, using it
iteratively to successively eliminate variables. For example, with three equations in three unknowns
f (x y z) = 0, g(x y z ) = 0, and h(x y z ) = 0, we can take the resultant of f and g treating z as
the only variable to get R1 (x y). Likewise, we can take the resultant of g and h again treating z as
the only variable to get R2 (x y). Finally, the resultant of R1 and R2 with y as the variable yields
R(x), whose roots can then be found using standard root nding methods.

3
1 Resultants of Polynomials in One Variable
1.1 The Basic Case { Two Polynomials and the Sylvester Matrix
Given two positive integers r s 1 and two polynomials in one variable
f (x) = ar xr +    + a1 x + a0 and g(x) = bs xs +    + b1x + b0
of degree less than or equal to r and s respectively, we dene their resultant Rr s(f g) by Sylvester's
formula:
0 1
a0 a1    ar 0    0
BB 0 a0 a1       ar 0    0 CC
BBB ... CC
CC
BB 0 0    0 a a    a CC
Rr s(f g) = det B BB b0 b1    0 1
bs;1 bs 0    0 C
r
BB 0 b0 b1 CC
BB ..    b s    0 CC
@ . CA
0 0  0 b0 b1    bs
which is the determinant of an r + s by r + s matrix with s rows involving the a's and r rows
involving the b's.
Example 1:
R2 2 (a2 x2 + a1x + a0 b2 x2 + b1 x + b0 )
= a20 b22 + a0 a2 b21 ; a0 a1 b1 b2 + a21 b0 b2 ; a1 a2 b0 b1 + a22 b20 ; 2a0 a2 b0 b2 :
Note that in this example each monomial in the resultant has total degree r + s = 4 and is
bihomogeneous of bidegree (s r) = (2 2) in the a's and b's respectively. This is true in general.
Notice also that the general bihomogeneous polynomial of bidegree (2 2) has 36 terms|any one
of a20 a21 a22 a0 a1 a0 a2 a1 a2 times any one of b20 b21 b22 b0 b1 b0 b2 b1 b2 . However, only seven of these
monomials occur in R2 2 .
Basic Properties of the Resultant Rr s(f g)
Q
1. Relationship to Common Roots: Rr s (f g) = asr brs i j (xi ; yj ) where x1 : : : xr are the
roots of f and y1 : : : ys are the roots of g. (Here we are assuming ar 6= 0 and bs 6= 0.) Thus
Rr s (f g) will be zero if and only if f and g have a root in common.
2. Irreducibility: Rr s(f g) 2 Za0 : : : ar b0 : : : bs ] is irreducible, i.e., the resultant is an
irreducible polynomial with integer (Z) coecients in (r + 1)(s + 1) = rs + s + r + 1 variables.
3. Symmetry: Rr s(f g) = (;1)rs Rs r (g f )
4. Factorization: Rr1 +r2 s(f1 f2 g) = Rr1 s(f1 g)Rr2 s (f2 g).
Example 2: R1 2(a1 x + a0 b2 x2 + b1x + b0 ) = a20 b2 ; a0 a1b1 + a21 b0 . Now consider

R2 2 ((a1 x + a0 )(c1 x + c0 ) b2 x2 + bz x + b0 )
= R2 2 (a1 c1 x2 + (a0 c1 + a1 c0 )x + a0 c0 b2 x2 + b1 x + b0 ):
4
This can be computed using Example 1 or by using property 4 above:
R2 2((a1 x + a0 )(c1 x + c0) b2 x2 + b1 x + b0 )
= R1 2 (a1 x + a0 b2 x2 + b1 x + b0 )R1 2 (c1 x + c0 b2 x2 + b1 x + b0 ):
Thus if the polynomials you are taking the resultant of can be factored, then the resultant can be
factored!
Other Formulas for the Resultant { Method of Bezout:
Suppose f = ar xr +    + a0 and g = bs xs +  + b0 and that a0 = 1, then letting
g(x) = r + r x + r x2 +   
f (x) 0 1 2

we have
0 1
rs    rs+r;1 C
Rr s(f g) = det B
@ rs;1    rs+r;2 A
rs;r+1    rs
which is an r  r determinant. Note that a0 = 1 is not a serious restriction, because as long as
a0 6= 0, we can scale f to make a0 = 1. Also note that the ri 's are polynomial expressions in the
aj 's and bk 's.
Example 3:
f (x) = 2x2 ; 3x + 1
g(x) = 5x2 + x ; 6
g(x) = (;6 + x + 5x2 )(1 + 3x + 7x2 + 15x3 +    ) = ;6 ; 17x ; 34x2 ; 68x3 +   
f (x)
and
 !  !
R2 2 (f g) = det rr2 rr3 = det ; 34 ;68 = 0
;17 ;34
1 2
which is correct because x = 1 is a common root.
Suppose now that f and g have the same degree, so r = s, and let ij ] stand for the expression
ai bj ; bi aj . The Bezout-Cayley formula for the resultant is the r  r determinant
001] 02] 03] 0r]
1
BB02] 03] + 12] 04] + 13] 1r] C
C
BB .. C
Rr r (f g) = det B B03] 04] + 13] 05] + 14] + 23] . C
CC
B@ ... ..
.
..
.
..
. CA
0r] 1r] 2r] r ; 1r]
Example 4:
 !
R2 2 (f g) = det aa0 bb1 ; a1 b0 a0b2 ; a2 b0
0 2 a2 b0 a1 b2 ; a2 b1
;
which can be checked against Example 1.
Various intermediate forms exist that lie between the r  r Bezout-Cayley form and the 2r  2r
Sylvester form.
5
1.2 Discriminants and Resultants
The discriminant (f ) of a polynomial f = ar xr +    + a0 , ar 6= 0, is essentially the resultant of
f and its derivative f 0 . The exact relationship is
(f ) = 1 R (f f 0 )
ar r r;1
which is a homogeneous polynomial of degree 2r ; 2 in the r + 1 variables a0 : : : ar .
Example 5:
(a0 + a1 x + a2 x2 + a3 x3 ) = 27a20 a23 + 4a0 a32 + 4a31 a3 ; a21 a22 ; 18a0 a1 a2 a3 :
Notice that only 5 out of a possible 35 terms (the number of monomials of degree 4 in 4 variables)
actually occur.  !
3 r ; 2
In general (a0 +    + ar x ) could consist of as many as 2r ; 2 = (2(3rr;;2)!2)!r! terms, but in
r

reality it consists of far fewer. For example when r = 5 (i.e., when f is a quintic, the number
of monomials of degree 8(= 2r ; 2) in 6(= r + 1) variables is 1287, but only 59 occur in the
discriminant.
Just as the discriminant can be dened in terms of the resultant, the resultant can be dened
in terms of the discriminant:
(Rr s(f g))2 = (;1)rs (( fg)
f )(g)
when ar 6= 0 and bs 6= 0.
1.3 Finding the Common Roots { Subresultants
Again, suppose we are given two polynomials in a single variable x, say
f (x) = ar xr +    + a1 + a0 and g(x) = bsxs +    + b1 x + b0
of degrees r 1 and s 1 respectively. (We assume that ar 6= 0 and bs 6= 0.) As we saw above, the
resultant Rr s (f g) of f and g will zero if and only if f and g have a common root. Two questions
immediately occur:
Question 1: Suppose Rr s (f g) = 0, so that f and g have at least one common root, can
we determine how many roots they have in common? This is the same as asking for the degree
1
d
min(r s) of the greatest common divisor h(x) of f (x) and g(x).
Question 2: Can we nd the common roots?
Question 1 can be answered using the notion of a subresultant. To understand this idea, we
rst recall the source of the Sylvester matrix whose determinant gives us the resultant Rr s (f g).
To say that f (x) and g(x) have one or more common roots is to say that they have a common
factor h(x) of degree d 1. This means that we can write

f (x) = h(x)f~(x) and g(x) = h(x)~g(x):

It follows that
g~(x)f (x) + (;f~(x))g(x) = 0
6
because
g~(x)f (x) = g~(x)h(x)f~(x) = g(x)f~(x):
Thus when f (x) and g(x) have one or more common roots, there will exist non-zero polynomials
s(x) = ss;1xs;1 +    + s1 x + s0 and t(x) = tr;1xr;1 +    + t1x + t0
of degrees at most s ; 1 0 and r ; 1 0 respectively, with the property that
s(x)f (x) + t(x)g(x) = 0: (1)
Conversely, if two such non-zero polynomials s(x) and t(x) can be found that satisfy (??), then
f (x) and g(x) will have one or more common roots. This is because all r roots of f (x) must then
be roots of t(x)g(x), but the degree of t(x) is at most r ; 1, meaning that at least one root of f (x)
is also a root of g(x).
If we multiply the left-hand side of (??) out and collect terms we get
(ss;1 ar + tr;1 bs )xr+s;1 + (ss;1 ar;1 + ss;2 ar + tr;1 bs;1 + tr;2 bs )xr+s;2 +   
0i 1
X X
i
+ @ ss;j ar+j ;i + tr;k bs+k;i A xr+s;i +   
j =1 k=1
+ (s1 a0 + s0 a1 + t1 b0 + t0 b1 )x + (s0 a0 + t0 b0 )
where all terms in the summation with negative indices are ignored. In order for this to be the
zero polynomial, all of its coecients must be zero. This leads to a system of linear equations for
the r + s variables ss;1 : : : s0 tr;1 : : : t0 . In matrix form this system is
0a ar;1    a1 a0 0 0  0
1
BB 0r ar    a1 a0 0  0 C
CC
BB ... ... CC
BB 0 0 CC
BB ... ..
. ... ... CC
(0 : : : 0) = (ss;1 : : : s0 tr;1 : : : t0 ) B BB 0 0 ar ar;1    a1 a0 C C
BB bs bs;1    b1 b0 0    0 0 C C
BB 0 bs bs;1    b1 b0 0 0C CC
BB . ... ... CA
@ ..
0       bs   b1 b0
and we immediately recognize the r + s by r + s matrix on the right as the Sylvester matrix. In
order for this system to have a non-zero solution (in fact a solution with not all si = 0 and not all
tj = 0), it is necessary and sucient that the determinant of the Sylvester matrix be zero. This of
course is just the resultant Rr s (f g).
Similarly, if f (x) and g(x) have two (or more) roots in common, then we can nd non-zero
polynomials
s(x) = ss;2xs;2 +    + s1 x + s0 and t(x) = tr;2xr;2 +    + t1x + t0
such that
s(x)f (x) + t(x)g(x) = 0
7
and conversely. This leads to a system of r + s ; 1 linear equation
0a ar;1    a0 0    0
1
BB 0r ar       a0    0CC
BB C
C
BB    0 ar ar;1    a0C
CC
(0 : : : 0) = (ss;2 : : : s0 tr;2 : : : t0 ) B BB bs  b1 b0    0 0 C
BB 0 C
B@ bs       b1 b0    0 CCC
A
0       bs       b0
in r + s ; 2 variables. It turns out that a necessary and sucient condition for this system to have
a solution with not all si = 0 and not all tj = 0, is that the rst r + s ; 2 by r + s ; 2 minor
(obtained by deleting the last column) has zero determinant. This matrix is a submatrix of the
Sylvester matrix. In general, if we write the Sylvester matrix as
0 1
ar ar;1    a0 0    0
BB 0 ar       a0    0 CC
BBB C CC
BB 0       ar       a0C C
BB 0    bs         b0 CC
B@   C A
bs       b0    0
the kth subresultant Rr(ks) (f g) is the determinant of the r + s ; 2k by r + s ; 2k submatrix obtained
by deleting the rst k rows, the last k rows, the rst k columns, and the last k columns.
The answer to Question 1 is given by:
Theorem 1 If Rr(0)s (f g) = Rr(1)s (f g) =    = Rr(ks;1)(f g) = 0 but Rr(ks)(r s) 6= 0 for some k 0
then f and g have exactly k roots in common. Here Rr s (f g) is denoted Rr(0)s (f g).
The answer to Question 2 is more subtle. In general, we cannot expect to be able to express
the common roots of f and g (assuming they have a root or roots in common) as rational expres-
sions in the coecients ar : : : a0 bs : : : b0 . For example, if f and g have rational coecients,
i.e., ar : : : a0 bs : : : b0 2 Q, the eld of rational numbers, then any polynomial expression in
the coecients would be a rational number. But polynomials with rational coecients can have
common roots that are not rational.
Example 6: f (x) = 3x4 + x3 +4x2 + x +1 = (x2 +1)(3x2 + x +1) and g(x) = x2 ; 1 = (x2 +1)(x2 ; 1)
both have rational coecients, but the common roots i are not rational numbers.
We can however answer Question 2 in a special case. If Rr s(f g) = 0 and at least one partial
derivative of the resultant computed symbolically
@R : : : @R @R : : : @R (2)
@a0 @ar @b0 @bs
is non-zero when the coecients of f and g are substituted, then f and g have exactly one common
root  and it can be found via the proportions:
 @R @R @R 
r
(1 :  :  :    :  ) = @a (f g) : @ (f g) :    : @ (f g)
2
 @R0 @R
1
@R
r 
s
(1 :  :  :    :  ) = @b (f g) : @b (f g) :    : @b (f g) :
2
0 1 s
8
In particular the common root  can be computed as:

@R @R
= @a1 (f g)
= @b1 (f g)
:
@R @R
@a0 (f g) @b0 (f g)
This result also has a geometric interpretation. The space of all pairs of polynomials (f g) with
the degree of f less than or equal to r and the degree of g less than or equal to s can be identied
with Rr+s+2 having coordinates (ar : : : a0 bs : : : b0 ). The symbolic resultant R is a polynomial
in these variables, and the locus R = 0 in Rr+2+2 is an irreducible hypersurface (of dimension
r + s + 1) consisting of pairs (f g) with a root in common. A point on this hypersurface where at
least one of the partial derivatives (2) is non-zero is a smooth point. At such points we have exactly
one common root. Moreover, that root can be expressed as a quotient of polynomial expressions
in ar : : : a0 bs : : : b0 . We remind the reader that \most" points on the locus R = 0 are smooth
points. Those that are not are called singular points, and they occur in dimension r + s or less.

9
2 Resultant Methods for Systems of Polynomial Equations in Sev-
eral Variables
2.1 Theory
The linear algebra techniques discussed in this section can be used to solve systems of polyno-
mial equations in several variables. If there are only two equations, then the Sylvester technique
(discussed above) can be employed, by treating all but one variable as part of the coecients.
However, when the number of equations exceeds two, the Sylvester approach can be misleading.
For example, taking the equations two at a time using the Sylvester determinant can lead the user
to the conclusion that there is a common solution, when in fact, there are no common solutions for
the system of equations taken as a whole.
What it means to \solve" a given set of polynomial equations depends upon the number of
variables and the number of equations. Assuming the equations are inhomogeneous, let \n" be
the number of variables and \m" be the number of equations. The expected dimensionality of
the set of solutions is n ; m when viewed over the complex numbers. For example, if there are
three equations (m = 3) and ve variables (n = 5), then the space of solutions is expected to have
dimension n ; m = 5 ; 3 = 2. Geometrically, the set of solutions forms a surface. Sometimes,
however, components of excess dimension occur in the set of solutions. These are geometric loci
of higher dimension than the expected dimension. They occur because, in a very loose sense, the
equations have certain dependencies.
Finally a word about homogeneous equations. Recall that a set of polynomial equations is
considered homogeneous if in each equation, all the terms have the same degree. If this is not
the case, even for only one of the equations, the set is regarded as inhomogeneous. For systems
of homogeneous equations, the number n of variables should be taken as one less than the actual
number of variables, when computing expected dimensions. This is because we want to regard the
solutions as lying in an (n ; 1)-dimensional projective space.
2.2 The Macaulay Resultant, the U -Resultant, and the GCP
The Macaulay resultant is the ratio of two determinants formed from the coecients of the given
polynomials in a manner to be described later in this section. If the number of equations exceeds
the number of variables by one (n ; m = ;1), then the Macaulay resultant tests whether or not
a common solution exists. (For systems of homogeneous equations where the number of equations
equals the number of variables, the expected dimension is still ;1, and the Macaulay resultant tests
for a non-trivial common solution, i.e. a solution other than (0 : : : 0).)
If there are as many inhomogeneous equations as unknowns (n ; m = 0), then the equations
can often be solved by adding the U -equation (explained later in this section) to the homogenized
set and forming the Macaulay resultant. The Macaulay resultant is then called the U -resultant.
In some cases, however, there will be a component of excess dimension ( 1) which masks some
or all of the desired solutions. In this case Canny's Generalized Characteristic Polynomial (GCP)
approach is useful (see ?]).
In order to illustrate the various methods, the following system of three polynomial equations
will be used:
f1 = y ; 3 x + 5 = 0
f2 = x2 + y2 ; 5 = 0
f3 = y ; x3 + 3x2 ; 3x + 1 = 0:

10
Here we have three inhomogeneous equations in two variables (n ; m = 2 ; 3 = ;1). The
multiresultant techniques described below can be used to test for the existence of a solution.
Step 1: Homogenization
The equations must rst be homogenized. This is done by adding a third variable, z . Specically
x is replaced by x=z and y is replaced by y=z, and the factors of z are cleared from the denominators.
In the above example this leads to three equations:
f1 = y ; 3x + 5z = 0
f2 = x2 + y2 ; 5z2 = 0
f3 = yz 2 ; x3 + 3x2 z ; 3xz2 + z3 = 0:
This is the homogenized version of the original system.
Step 2: Degree Determination
Each of the multiresultants being considered involves the coecients of various monomials
that appear in the equations. The variables involved in the monomials are the variables that
appear in the homogeneous form of the polynomial equations. For example, the homogeneous
polynomial equations above have the variables x, y, and z . All the monomials in a given equation
are constrained to have the same degree because we have homogenized. The \overall degree" of the
system is determined from the degrees of the individual homogeneous equations by the following
rule:
X
m
d=1+ (di ; 1)
i=1
where
m =the number of equations
di =the degree of the \i'th" equation.
For the homogeneous polynomials above (f1 , f2 , and f3 ) the degrees are:
Equation Degree
f1 d1 = 1
f2 d2 = 2
f3 d 3 = 3:
Therefore,
d = 1 + (1 ; 1) + (2 ; 1) + (3 ; 1) = 4:
Step 3: Matrix Size Determination
Each of the multiresultants to be discussed involves the ratio of two determinants. The numer-
ator is the determinant of a matrix, the formation of which will be discussed in subsequent sections.
The denominator determinant is formed from a submatrix of the numerator matrix.
The number of variables in the inhomogeneous equations is n. Since one additional variable has
to be added to homogenize the equations, the number of variables in the homogeneous equations
is n + 1. The size of the numerator matrix equals the number of monomials in the n + 1 variables
that have overall degree d (discussed in the previous section).
 !
Numerator Matrix Size = n + d : d
11
 For the three polynomial equations (f1 f2 f3 ) we have already calculated that d = 4. Since the
original set of inhomogeneous variables consisted of x and y, we have that n equals 2. Thus for our
example,
 ! !
6! = 15
Numerator Matrix Size = 2 +4 4 = 64 = (2!)(4!)

i.e. it is a 15  15 matrix.
Step 4: Determining \Big" vs. \Small" Exponents
A few of the 15 monomials involving the variables x, y, and z with an overall degree of 4,
include:
yz3 and x2 y 2 :
In the next section, we will discuss whether certain of these monomials are reduced. This will be
determined by whether the exponents are \big" or \small". In this section we discuss how \bigness"
is dened.
Each variable will be associated with a particular equation. For example the rst variable, x,
will be associated with the rst equation, f1 . The second variable, y, will be associated with the
second equation, f2 , etc. The degrees of the associated equations dene \bigness" for the exponents
of that variable. Specically, since d1 (the degree of f1) is 1, if the exponent of x is greater than or
equal to 1, it is considered big. Since d2 = 2, whenever the exponent of y is greater than or equal
to 2, it is considered big. The degree of f3 is 3, therefore, whenever the exponent of z is greater
than or equal to 3, it is considered big.
For example, consider the monomial yz 3 . The exponent of y is 1. This is less than d2 , and is
considered small. The exponent of z is 3. This is equal to d3 , and is therefore big. On the other
hand, consider the monomial x2 y2 . The exponent of x is 2. This is greater than d1 , and is big. The
exponent of y is 2. This is equal to d2 , and is big.
Step 5: Determining the Reduced Monomials
If for a particular monomial of degree d the exponent of only one variable is big, the monomial
is said to be reduced. In the previous step only the monomial yz 3 is reduced. For that monomial
only the exponent of z is big whereas for x2 y2 , both the exponent of x and the exponent of y are
big. Thus the monomial x2 y2 is not reduced.
The Macaulay Resultant
The Macaulay Resultant is the ratio of two determinants. The numerator is the determinant of
a matrix which we will call the A matrix. The denominator is the determinant of a matrix which
we will call the M matrix
detjAj :
R = det jM j
Step 6: Creating the A Matrix:
We have discussed above how the size of the A matrix is determined. In this section we will
show how the matrix entries are obtained.
Each row and column of the matrix should be thought of as being labeled by one of the mono-
mials of degree d. Recall that for f1 , f2 , and f3 in our example there were 15 possible monomials
of degree 4 in x y z , and therefore the A matrix would be 15  15.

12
 There are three rules for determining the elements of the A matrix. After presenting the rules,
the example involving f1 , f2 , and f3 , will be used to illustrate the process. The reader may nd it
helpful to read the example simultaneously with the rules.
Rules for the elements of each column:
(1) Search the monomial labeling that column from left to right for the rst variable with a big
exponent. Such a variable must exist. Call it the marker variable.
(2) Form a new polynomial from the polynomial associated with this marker variable by multi-
plying the associated polynomial by the monomial and dividing by the marker variable raised
to the degree of the associated polynomial.
(3) The coecients of the new polynomial are the elements of the columns. Each coecients
goes in the row labeled by the monomial it multiples. All the other rows get zeroes.
Example 7: Recall that for the system of equations f1 f2 f3 there are 15 monomials of degree 4
that can be formed from x, y, and z . Two of these were considered above, namely yz 3 and x2 y2 .
 For the column labeled by yz3 :
(1) The rst variable with a big exponent is z , so z is the marker variable.
(2) The polynomial associated with z is f3 . Multiply f3 by the monomial yz 3 , and divide
this product by z 3 .
f3(yz 3 ) = (yz 2 ; x3 + 3x2 z ; 3xz2 + z 3 )(yz3 ) = y2 z 2 ; x3 y + 3x2 yz ; 3xyz2 + yz 3 :
z3 z3
(3) The coecient of y2 z 2 is +1. Therefore the element of the row labeled y2 z 2 is +1.
The coecient of x3 y is ;1. Therefore the element of the row labeled x3 y is ;1. The
coecient of x2 yz is +3. Therefore the element of the row labeled x2 yz is +3. The
coecient of xyz 2 is ;3. Therefore the element of the row labeled xyz 2 is ;3. The
coecient of yz 3 is +1. Therefore the element of the row labeled yz 3 is +1. All other
entries in the column are zero.
 For the column labeled by x2y2 :
(1) The rst variable with a big exponent is x, so x is the marker variable.
(2) The polynomial associated with x is f1 . Multiply f1 by the monomial x2 y2 , and divide
this product by x.
f1 (x2 y2 ) = (y ; 3x + 5z )(x2 y2 ) = xy3 ; 3x2 y2 + 5xy2 z:
x x
(3) The coecient of xy3 is +1. Therefore the element of the row labeled xy3 is +1. The
coecient of x2 y2 is ;3. Therefore the element of the row labeled x2 y2 is ;3. The
coecient of xy2 z is +5. Therefore the element of the row labeled xy2 z is +5.
When all the columns are determined, the A matrix in our example takes the form:

13
 A Matrix
x4 x3 x3 x2 x2 x2 x x x x
y y2 y y3 y2 y y4 y3 y2 y
z z z 2 2
z z z 3 z z2 z3 z4
x4 ;3 0 0 0 0 0 0 0 0 0 0 0 0 0 0
x3 y 1 ;3 0 0 0 0 0 0 0 0 0 0 0 ;1 0
x3 z 5 0 ;3 0 0 0 0 0 0 0 0 0 0 0 ;1
x2 y2 0 1 0 ;3 0 0 0 0 0 0 0 1 0 0 0
x2 y z 0 5 1 0 ;3 0 0 0 0 0 0 1 0 3 0
x2 z 2 0 0 5 0 0 ;3 0 0 0 0 0 0 1 0 3
x y3 0 0 0 1 0 0 ;3 0 0 0 0 0 0 0 0
x y2 z 0 0 0 5 1 0 0 ;3 0 0 0 0 0 0 0
x y z 2 0 0 0 0 5 1 0 0 ;3 0 0 0 0 ;3 0
x z3 0 0 0 0 0 5 0 0 0 ;3 0 0 0 0 ;3
y4 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0
y3 z 0 0 0 0 0 0 5 1 0 0 0 1 0 0 0
y2 z2 0 0 0 0 0 0 0 5 1 0 ;5 0 1 1 0
y z 3 0 0 0 0 0 0 0 0 5 1 0 ;5 0 1 1
z4 0 0 0 0 0 0 0 0 0 5 0 0 ;5 0 1
The determinant of the above A matrix is zero. If the determinant of the M matrix is nonzero,
this would imply that the system has a solution.
Step 7: Creating the M Matrix
The denominator of the Macaulay Resultant is the determinant of the M matrix. The M matrix
is a submatrix of the A matrix. It consists of the elements which have row and column monomial
labels which are not reduced. Recall that a monomial is not reduced if it has more than one variable
with a big exponent.
The size of the M matrix equals the size of the A matrix minus D, where
X
m Y
D= dj :
i=1 i6=j
In our example,
D = d2 d3 + d1 d3 + d1 d2 = (2)(3) + (1)(3) + (1)(2) = 11
so that the size of the M matrix is 15 ; 11 = 4. The actual M matrix for f1 , f2 , and f3 is:
M Matrix
x2 y2 xy3 xy2z xz 3
x2y2 ;3 0 0 0
xy3 1 ;3 0 0
xy2z 5 0 ;3 0
xz 3 0 0 0 ;3
The determinant of this M matrix yields a value of 81. Since the determinant of the A matrix
was zero, the Macaulay Resultant is zero, which implies that there is a solution to our system. The
following plot of the three polynomials (f1 , f2 , and f3 ) conrms that there is a common point at
x = 2 and y = 1.
14
 y
6 ..
..
.
..
..
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...
..
...
.
...

.
. ...
..
.
.. ..
. ...
...
. ...
.
..
. ...
.. ....
..
... ....
..
. ..
... ....
.
. .
..............
.......................... ... ..
........
........
....... ... ...
...... ...... .. ...
...... ..... ... ..
.
.. ..... .... ........
... .... ....
... ... ......
..
(2 1)
.......
... ..
.....
... ... ...

-
.... ..... ..
...
......... ....
.
..
.. ...... ... .. .
....... ..

x
..
.
.. ........ .... .
.. ........ .
.
... ....... ... ...
... ....... .
... ....
... ...
. ... . ..
... . . .
... .... ...
. ..
.
.
... . ... ...
..
... .. .. ...
...
.... ... ... ....
..... ... ... .....
...... ..
.
. .... ..........
...... . .
......
....... .. .........
.......... ....
............................................
. .
... ...
...
. ....
. .
... ...
... ...
.. ..
... ..
..
.
. ...
.
.
. .
... ...
... ..
... ...
... ...
..
.
..
..
..
..

Sometimes both the A matrix and the M matrix have zero determinant. This indeterminacy
can often be circumvented if the polynomials are rst written with symbolic coecients. The
determinants of the A and M matrices are obtained, polynomial division is performed, and then
at the end, the symbolic coecients are replaced by their numerical values to check if the resultant
is zero. Since one does not know ahead of time whether or not this \division by zero" condition
will arise, the symbolic coecient approach is the best strategy. It is also often sucient to treat
just a subset of the coecients symbolically|sometimes as few as a single symbolic coecient will
remove the indeterminacy.
The U-Resultant
For problems with as many inhomogeneous equations as variables, the U -resultant can often be
used to solve for the point solutions. The three polynomial equations f1 f2 f3 do not satisfy these
conditions, since there are three equations in two inhomogeneous variables, x and y. However, if
we take just the rst two equations, namely f1 and f2, we would have a system with as many
equations as variables.
The given equations must rst be homogenized. This adds one additional variable. We then
add one additional equation to the system. This equation is called the U -equation. If x and y are
the given variables and z is the homogenizing variable, then the U equation takes the form:

u1 x + u2 y + u3 z = 0:
The Macaulay Resultant, R, is then computed for these m + 1 equations, treating the ui as
symbolic coecients. The result is called the U -resultant. Notice that R will be a polynomial in
the ui 's and the coecients of the original equations.
After R is determined, it is factored into linear factors. For each linear factor there is a point
solution of the original system of equations. The coordinates of each solution are given as ratios of
the coecients of the ui 's. The denominator is always the coecient of the ui associated with the
homogenizing variable. In our example, this is the coecient of u3 . Thus

x = coe. of u1 and y = coe. of u2 :

coe. of u coe. of u 3 3
For example, if a linear factor turned out to be
u1 ; u2 ; u3
15
then the coordinates of the associated solution would be
x= ; +1 = ;1 and y = ;1 = +1:
1 ;1
Example 8: As mentioned above, the polynomial equation system f1 f2 f3 is overdetermined
(n ; m = ;1). However, we can use the U -resultant to solve f1 and f2 for x and y(n ; m = 0). In
this example, we will also demonstrate the symbolic approach alluded to in the previous section.
Recall that the homogenized form of f1 and f2 is:
f1 = y ; 3x + 5z = 0
f2 = x2 + y2 ; 5z2 = 0:
Rewriting these two equations with symbolic coecients and including the U -equation yields:
f1 = a1 x + b1 y + c1z = 0
f2 = a2 x2 + b2 y2 + c2 z2 = 0
U = u1x + uxy + u3 z = 0
where, a1 = ;3, b1 = 1, c1 = 5, a2 = 1, b2 = 1, and c2 = ;5.
The U -resultant is calculated in the same way as the Macaulay resultant, i.e., with the A matrix
and the M matrix, except now we are using symbolic coecients.
A Matrix
x2 xy xz y2 yz z2
x2 a1 0 0 a2 0 0
xy b1 a1 0 0 u1 0
xz c1 0 a1 0 0 u1
y2 0 b1 0 b2 u2 0
yz ; c1 b1 0 u3 u2
z2 0 0 c1 c2 0 u3
The corresponding M matrix is a single element, namely a1 .
The determinant of M is divided into the determinant of A to obtain the U -resultant. Finally,
the symbolic coecients are replaced by their numeric equivalents. (This could have been done
from the outset, unless a1 had been zero.) The result is
10(u1 ; 2u2 + u3 )(2u1 + u2 + u3 ):
This yields two solutions:
Solution #1:
coe. of u1 = +1 = ;1 and y = coe. of u2 = ;2 = ;2:
x = coe. of u3 ;1 coe. of u3 +1
Solution #2:

x = coe. of u1 +2 coe. of u2 +1
coe. of u3 = +1 = +2 and y = coe. of u3 = +1 = +1:

16
We remark that the U -resultant will be identically zero and give no information, if the set of common
solutions contains a component of excess dimension one or more. Moreover, this component may
be at innity where the homogenizing variable is zero.
The GCP Approach
The Generalized Characteristic Polynomial (GCP) approach ?] avoids the problem of compo-
nents of excess dimension in the set of solutions. It can be used together with the U -resultant which
was discussed above. If the U -resultant leads to an indeterminant (0=0) form even when symbolic
coecients are used, an \excess" solution exists. The GCP takes the form
detjA ; sI j evaluated at s = 0 after division
R = det jM ; sI j
where A and M are the matrices dened above, s is a perturbation parameter, and I is the identity
matrix.
One way to carry out the above operation is the following:
(1) Set up the A matrix (as described above). Subtract s along the diagonal. Evaluate the
determinant. Retain the coecient of the lowest surviving power of s.
(2) Repeat (1) for the M matrix.
(3) Divide the result of (1) by the result (2).
All of these multiresultant techniques have one thing in common. They require that there
be one more equation than variable, n ; m = ;1. If there are as many equations as variables
n ; m = 0, the the U equation is added and the eective situation is again n ; m = ;1. If there
are more variables than equations (n ; m is a positive integer), then enough of these variables must
be regarded as parameters in the coecients, so that eectively n ; m = ;1. Geometrically this
amounts to projecting the locus of solutions to a hypersurface in a lower dimensional space. Finally,
if the number of equations exceeds the number of variables by more than one (n ; m
;2), then
some technique other than the above multiresultant techniques (e.g., a system of multiresultants)
must be employed to determine if a solution exists.
2.3 The Jacobian Method of Salmon
Consider a system of m polynomial equations in n variables
f1 (x1 : : : xn ) = 0
.. (3)
.
fm(x1 : : : xn) = 0:
For example, take a system of three quadratic equations in three variables x y z :
0 = ak x2 + bk y2 + ck z 2 + dk xy + ek xz + fk yz + gk x + hk y + ik z + jk k = 1 2 3:
Geometrically, this amounts to 3 quadratic surfaces intersecting in 3-space.
From Bezout's Theorem we expect this system to have 8 solutions (counting complex solutions,
solutions at innity, and counting each solution the appropriate number of times, namely its multi-
plicity). Of course, it is possible in a degenerate situation for the above system to have an innite
number of solutions. This can occur if all the surfaces have a common (surface) component (i.e.,
the polynomials have a common factor), or if they intersect in a common curve.
17
 We can add to our system of equations any equation of the form
fj (x1 : : : xn )xe11 : : : xenn :
In fact, any equation of the form
X
m
gj (x1 : : : xn)fj (x1 : : : xn )
j =1
for arbitrary polynomials gj (x1 : : : xn ), can be added without changing the set of common solu-
tions.
Our goal is to convert problem (??) into a system of linear equations where the methods of linear
algebra can be applied. To accomplish this, we begin by writing each of our quadratic equations as
0 = ak x2 + bk y2 + dk xy + (ek z + gk )x + (fk z + hk )y + (ck z 2 + ik z + jk ) k=1 2 3
where we are thinking of them as quadratic equations in two variables x and y with coecients
that vary with z . We have six monomials in x and y, namely
x2 y2 xy x y 1
which we would like to treat as independent variables, giving us a system of linear equations.
Unfortunately, this would be a system of only 3 equations in the \six variables", and we would
prefer six equations. If we try to increase the number of equations by multiplying the existing
three by various monomials, we invariably end up introducing new monomials in x and y|in eect
increasing the number of \variables". Sometimes this method can be made to work (see Roth ?]
for examples).
The trick, at least for the case of three quadratic equations in three variables, goes back to a
result of Salmon.
Theorem 2 (Salmon ?, pg. 88]): If we have a system of m homogeneous equations in m variables
f1(x1 : : : xm ) = 0
.. (4)
.
fm (x1 : : : xm ) = 0
(note: m = n), then any non-trivial common solution is also a solution of the Jacobian polynomial
0 @f1=@x1    @f1=@xm 1
J (x1 : : : xm ) = det B
@ ... .. C
. A:
@fm=@x1    @fm =@xm
Moreover, if the fi all have the same degree, then any non-trivial common solution to the system
(??) is also a solution of all the polynomials

@J (x : : : x ) i = 1 : : : m:
@xi 1 m

18
 Proof. We do the case of three variables. Thus we consider a system of 3 homogeneous equations
f (x y w) = 0
g (x y w ) = 0
h(x y w) = 0
in the variables x, y, and w. Recall that for homogeneous polynomials we have the identities
x @f +
@x @y y @f + w @f = (degf )f
@w
@g @g @g = (degg)g
x @x + y @y + w @w
x @h +
@x @y y @h + w @h = (degh)h
@w
Consider this as a system of equations
0 @f @f @f 1 0 1 0 1
BB @x@y @w C BxC B(degf )f C
@g@g @g C
@ @x@y @w A @y A = @ (degg)g A : (5)
@h @h @h
@x @y @w z (degh)h
0 1
u 1 v1 w1
Let B
@u2 v2 w2CA be the matrix of cofactors of the Jacobian, so
u3 v3 w3
@g @h ; @h @g
u1 = @y @w @y @w
@f @h + @h @f
v1 = ; @y @w @y @w
@f @g @g @f
w1 = @y @w ; @y @w
etc. Multiplying both sides of (??) by this matrix yields
Jx = u1(degf )f + v1 (degg)g + w1 (degh)h
Jy = u2(degf )f + v2 (degg)g + w2 (degh)h
Jw = u3(degf )f + v3 (degg)g + w3 (degh)h:
Now if (x : y : w) is a non-trivial solution to f , g, and h, then Jx = 0, Jy = 0, and Jw = 0. But at
least one of x, y, and w is non-zero (otherwise the solution is trivial) so J = 0. Algebraically this
means that J is in the radical of the ideal generated by f , g, and h.
Now if degf = degg = deg h = d, we can dierentiate Jx again with respect to x, y, and w, to
get

@J @u 1 @v 1 @w 1
 @f @g @h 
J + x @x = df @x + dg @x + dh @x + d @x u1 + @x v1 + @x w1 :
and

19
 x @J = df @u1 + dg @v1 + dh @w1 + d  @f u + @g v + @h w 
@y @y @y @y @y 1 @y 1 @y 1
and
@J = df @u1 + dg @v1 + dh @w1 + d  @f u + @g v + @h w 
x @w @w @w @w @w 1 @w 1 @w 1
But remember

@f u + @g v + @h w = J
@x 1 @x 1 @x 1
@f u + @g v + @h w = 0
@y 1 @y 1 @y 1
@f u + @g v + @h w = 0
@w 1 @w 1 @w 1
so that

J + x @J
@x = df @u1 + dg @v1 + dh @w1 + dJ
@x @x @x
and

x @J
@y = df @u1 + dg @v1 + dh @w1
@y @y @y
@J @u @v @w
x @w = df @w1 + dg @w1 + dh @w1 :
Now when we have a common solution, i.e., when f , g, and h are all zero, then so is J by what we
have already showed. Thus
x @J
@x = 0
x @J
@y = 0
@J = 0:
x @w
Likewise dierentiating Jy and Jw yields:
y@J=@x = 0
y@J=@y = 0
y@J=@w = 0
w@J=@x = 0
w@J=@y = 0
w@J=@w = 0:

20
From these nine expressions and the fact that (x : y : w) 6= (0 : 0 : 0), we conclude
@J=@x = 0 @J=@y = 0 and @J=@w = 0
as required. Again, algebraically, this says that @J=@x, @J=@y, and @J=@w are in the radical of the
ideal generated by f , g, and h.
We now apply this to our system

ak x2 + bk y2 + dk xy + (ek z + gk )xw + (fk a + hk )yw + (ck z 2 + ik z + jk )w2 k = 1 2 3 (6)

after making it homogeneous by inserting the variable w as appropriate. Keep in mind that z is
treated as a parameter in the coecients, not as one of our variables.
Since each entry in the 3  3 Jacobian matrix of our system is homogeneous linear in x y w,
it is easy to see that J is homogeneous cubic in x y w and that therefore @J @J @J
@x , @y , and @w are
homogeneous quadratic. Adding these three quadratic equations to (??) yields the following system:
0 a b1 d1(e1 z + g1 ) (f1 z + h1 ) (c1 z 2 + i1 z + j1 )1 0 x2 1 001
1
B
B a 2 b 2 d2(e2 z + g2 ) (f2 z + h2 ) (c2 z 2 + i2 z + j2 )C
C BB y2 CC BB0CC
B
B a3 b3 d3(e3 z + g3 ) (f3 z + h3 ) (c3 z 2 + i3 z + j3 )C
CC BBB xy CCC = BBB0CCC
B
B3A ( z ) D ( z ) 2 B (z ) 2
2C ( z ) 2
F (z ) 3
E (z ) CC BBxwCC BB0CC (7)
B
@ B (z) 3G(z) 2D(z) 2
F (z ) 2
2H (z ) 3
I (z ) A @ywA @0A
2 2 2
C (z ) H (z ) F (z ) 2E (z ) 3 3
2I (z ) 4
3J (z ) w2 0
where

J =A(z)x3 + B (z)x2 y + C (z2 )x2 w + D(z )xy2 + E (z3 )xw2

+ F (z 2 )xyw + G(z )y3 + H (z 2 )y2 w + I (z 3 )yw2 + J (z 4 )w4
and the exponents on z indicate the degree of the polynomial in z (so I (z 3 ) is a cubic expression
in z ).
In order for the system (??) to have a solution the determinant of the 6  6 matrix must be
zero. This determinant is an eighth degree polynomial in z whose roots are the z -coordinates of
the eight sought after points (x y z ) that are the common solutions of our original three quadrics.
Once such a z is found, we can set w = 1 and solve (??) for x and y.
While of limited use, this trick does allow one to generate additional equations that can be
added to the system without creating new monomials in the expressions.
Geometrically, we can explain why J vanishes if f , g, and h do. Each of f = 0, g = 0, and
h = 0 can be viewed as a curve in the projective plane P2 (using the fact that f , g, and h are
homogeneous in x y w). Each row of the Jacobian matrix gives the coecients of the tangent line
to this curve. If we have a point p common to all three curves:
. ...
.
.
. ...
.
.

`3
. ....
.
.
.
. ....
.
. .
.
. ....
. .
.
. ... .....
.
.
. ......
...... .....

c3
.
. ......
...
.
.
. .. .... .......
... .
. ......

p
.
............. .
. ...... .......
............. .
.
. ...... .....................
............. .. ...... ................
............. ..
. ...... ..............
.............
.............
.
.
. ...
............................
.............. . ...............
.
.
. ...
.
..
...
................................................................... . . ..
.......................... .......................
................. .......
....... . . ................... ...........................
..... . ....
...... .
.
....
....... ..............
......... . ........ ............

`2
.
. . ....... ..............
.
....
..............
. .....
.
. ....... .............
... .
. . ...... .............
.......... . ...
. ......

c2
..... .
. .
....... ......
. . ....
.
.
......
......
... . .
....
...... ..... . ....
.
. .
....
.
...... .
...
.
.
. ...
...... .
. ...
...... ... .
. ...
... .
.
.
.
...
....
.

`1 c1
.
.
. ...
. .
.
.
.
.
.
.
.

21
then the three tangent lines at that point cannot be \independent" because we are in dimension
2. Thus one line is a linear combination of the other two, i.e., one row of the Jacobian matrix is a
linear combination of the other two, making the Jacobian determinant J equal to zero.
Example 9: Consider the system of equations
x2 + y2 ; 2 = 0
x2 + y2 + z 2 ; 3 = 0
x2 ; y2 = 0:
The set of common solutions is easily seen to consist of the eight points (x y z ) = (1 1 1).
We begin by homogenizing the system with respect to x and y:
1x2 + 1y2 + 0xy + 0xw + 0yw + (;2)w2 = 0
1x2 + 1y2 + 0xy + 0xw + 0yw + (z 2 ; 3)w2 = 0
1x2 + (;1)y2 + 0xy + 0xw + 0yw + 0w2 = 0:
The Jacobian is
0 1
2x 2y ; 4w
J = det B
@2x 2y (2z2 ; 6)wCA = 16xyw(z2 ; 1)
2x ;2y 0
and
@J = 16yw(z2 ; 1)
@x
@J = 16xw(z2 ; 1)
@y
@J = 16xy(z2 ; 1):
@w
Adding these last three equations to our original three yields the following \linearized" system:
01 1 0 0 0 ;2 1 0 x2 1 001
BB1 1 0 0 0 z2 ; 3C
CC BBB y2 CCC BBB0CCC
BB1 ;1 0 0 0 0 C BB xy CC = BB0CC :
BB0 16(z ; 1) 0 C C
C BBxwCC BB0CC
0 0 0 2
B@0 0 0 16(z 2 ; 1) 0 0 A @yw A @0A
0 0 16(z 2 ; 1) 0 0 0 w2 0
The determinant of the 6  6 matrix is 8192(z ; 1) which is of degree eight as expected. The
2 4
roots of this polynomial in z are z = 1. Both cases lead to the system (after setting w = 1):
01 1 0 0 0 ;21 0x2 1 001
BB1 1 0 0 0 ;2CCC BBB y2 CCC BBB0CCC
BB1 ;1 0 0 0 0C BxyC B0C
BB00 0 0 0 0C CC BBB x CCC = BBB0CCC
B@0
0 0 0 0 0 A @ y A @0A
0 0 0 0 0 0 1 0
22
This must now be solved for x and y yielding x = 1 and y = 1.
This example points up one defect of the method. While z is eliminated, we are left in general
with a system of six quadratic equations in x and y that must be solved. This system may not
have a solution, or may not have a solution with w = 1. In the later case we may get a solution
with w = 0, but that is a solution at innity.

23
3 Geometric Aspects of Elimination Theory and the Theory of
Equations
3.1 Intersections of Curves in the Plane
A curve C in the plane of degree d 1 is described by the zeros of a single polynomial f (x y) of
degree d. If f (x y) factors, say as
f (x y) = g(x y)h(x y)
where the degree of g is d1 1, and the degree of h is d2 1, then C can be written as the union of
two curves C1 and C2 of degrees d1 and d2 respectively. We can continue factoring until f (x y) is
written as a product of irreducible (non-factorable) polynomials each of degree 1. C is then just
the union of a nite number of irreducible curves. For that reason, we often assume that f (x y) is
irreducible to begin with, and when that is the case C will be called an irreducible plane curve.
Let (x0 y0 ) be the coordinates of a point P on C , i.e., let (x0 y0 ) be a solution to f (x y) = 0.
We can make a change of coordinates by letting
x0 = (x ; x0) and y0 = (y ; y0 )
which will put P at the origin. Recall that the multivariate Taylor expansion of f (x y) around
(x0 y0 ) yields
f (x y) = f (x0 y0) + @f
@x ( x0 y 0 )(x ; x0 ) + @f (x y )(y ; y ) + higher order terms:
@y 0 0 0

In these new coordinates C is given by the equation

0 = f (x0 + x0 y0 + y0 ) = @f
@x ( x 0 y 0 ) x0
+ @f (x y )y0 + higher order terms in x0 and y0
@y 0 0
and the point P has coordinates (x0 y0 ) = (0 0).
In what follows, we will assume that this has already been done so that P = (0 0) and
f (x y) = f1(x y) + f2 (x y) +    + fd(x y)
where fi(x y) is a homogeneous polynomial of degree i. In particular f1 (x y) = @f
@x (0 0)x +
@f (0 0)y.
@y
Denition 1 We say that P 2 C is a singular point if f1 is identically zero. Otherwise P is a
nonsingular (or smooth) point of C and the line f1 (x y) = 0 is called the tangent line to C to P .
Note that in the original coordinates, where P has coordinates (x0 y0 ), the tangent line is given
by
0 = @f
@x ( x 0 y0 )( x ; x0 ) + @f (x y )(y ; y ):
@y 0 0 0

Denition 2 A singular point P 2 C is called an m-fold singular point on C (double point, triple
point, etc.) if f1 = f2 =    = fm;1 = 0 but fm is not zero.

24
 Since fm is homogeneous in two variables, it can be completely factored (over C , the complex
numbers) into linear factors
Y
m
fm = (i x ; vi y):
i=1
Denition 3 If the lines ix ; viy = 0 are all distinct, we say that P is an ordinary m-fold point.
An ordinary double point is also called a node.
Example 10: Consider the point (0 0) on y2 + x2 ; x3 = 0.
(2 2)
.
...
..
...
....
...
..
...
...
......
.
...
...
....
...
....
....
.
..
...
..
...
...

(1 0)
....
..
.
..
..

(0 0)
...
..
..
...
...
...
...
...
....
....
....
....
....
....
...
...
...
...

(2 ;2)
...
...
...
...
...
.

The real solutions to this equation are picture above. Notice that (0 0) appears to be an isolated
point. This an artifact of working over R over C it is not isolated. In this example f (x y) =
y2 + x2 ; x3 and at (0 0) we have:
f1 = 0 f2 = y2 + x2 and f3 = ;x3 :
We see that (0 0) is a singular point because f1 = 0, and that it is in fact a double point because
f2 6= 0. Moreover f2 = y2 + x2 = (y + ix)(y ; ix) so that f2 factors into two distinct linear factors
which means that P is an ordinary double point.
Example 11: y2 ; x2 ; x3 = 0
....
...
...
...
...
.
.....
...
................................................... ....
.................. ....
.........
......... ....
......
......... .....

(;1 0)
........ ....
..... ....... ....
. ...... .......
...
.
........
..
... ...... .........
... ...... .....
.... ....... .....
.......
.......... ..
............. .....
....................... . .
................... ....
....
............................ ....

(0 0)
....
....
....
...
....
...
...
..

This is the more typical picture of an ordinary double point at (0 0).

Denition 4 A double point is called a cusp if f2 = (ax + by)2 and the intersection multiplicity of
the line ax + by = 0 with the curve is 3 (see below).
Example 12: y2 ; x3 = 0

25
 ...
...
..
...
..
.
.
...
...
...
...
....
..
.....
......
......
.......
......
........
. .
..........

(0 0)
.....
. ..
...........
...........
.............................
............
...........
..........
........
........
.......
.......
......
.....
.....
....
....
....
....
...
...
...
...
...
...
...
...
.

This curve has a cusp at (0 0).

Example 13: The singularity pictured below is a double point called a tacnode:
6
......
........................................
y .................
...........................
....
..........
....... ........... ....
.. ..... ....... ...
... .... ...... ...
.... ....
. . ..... ...
...
. .... .
. .. ... ..... ...
...
... ... ... ...
... .. .
.... ... .. ..
...
... .. ...

-
.
. . . ..
..
..... .....
. . ..
.
.
... .. ..
.... ... .
.
.
... ... ..
..

x
.... ..
... .
.
.
.
.. .. . .
.
.
... . . .
... .......
. ..
.
... .. . ..
..
... ... ... ..
.
... ... ... ...
.... ... ..... ...
... .... ... ..
.
... ... ... .
... ... .... ..
... .... ..... ...
... ..... ...... ..
.... ...... ....... ...
......... ............ ...
.....
............................................ .........................................
..

We will be interested in solving systems like

f (x y ) = 0
g (x y ) = 0:
Let's begin by considering the case where g(x y) = ax + by + c. This means that we are intersecting
the curve C given by f (x y) = 0 and the line L given by g(x y) = 0. If b 6= 0, we can solve for y:
y = ; ab x ; cb :
If we substitute this into f :
a a  
0 = f x ; b x ; c = h(x)
the result is a polynomial h(x) in the one variable x, of degree at most d = degree f . Notice that
h could be identically zero if f (x y) factors into ax + by + c times another polynomial. We will
assume that this is not the case. Then we have at most d intersection points whose x coordinates
can be found by solving h(x) = 0.
Let's consider the following special case. Namely, let P be a smooth point on a curve C (we
assume P = (0 0) by changing coordinates) given by f (x y) = 0. We have that
g(x y) = @f
@x (0 0) x + @f (0 0)y = 0
@y
is the tangent line L to C at P . After another change of coordinates, we can assume that g(x y) =
y = 0 is the tangent line to C at (0 0). We dene the intersection multiplicity of P 2 L \ C to be
the order of vanishing of f (x 0) at x=0. This amounts to restricting the polynomial f (x y) to the
x-axis which is of course the tangent line given by y = 0.
26
Denition 5 If this order of vanishing is 3, then P is called an ordinary ex.
Example 14: y ; x3 = 0 has an ordinary !ex at (0 0).
y
6 .
.
...
..
...
...
.
...
...

..
...
...
.
...
.
...
...
..
...
...
.
..
...
....
...

-
....
..
.......
.
...
.......
.........

x
..............
..............................
..........
..
..
.........
.....
......
....
...
....
.
....
..
...
..
...
...
.
...
..
..
...
..
...
..
.
.
.
...
...
...
...
...

Generally a tangent line to a curve at a nonsingular point meets the curve with intersection multi-
plicity 2. If a tangent line has higher than expected order of contact at P , then P is called a !ex
of the curve.
Now suppose we have two curves C1 given by f (x y) = 0 of degree d1 1 and C2 given by
g(x y) = 0 of degree d2 1. We will assume f g have no common factor so that C1 \ C2 consists
of a nite set of points. Let P be one of those points and change coordinates to that P = (0 0).
Denition 6 If P is a nonsingular point on both C1 and C2 , and the two tangent lines are distinct,
then we dene the intersection multiplicity IP (C1 C2 ) of P 2 C1 \ C2 to be 1. The picture is:
......... ...
....
C1
.........
.........
......... ....
......... ..
......... ...
......... ....
.
.........
......... ...
........ ..
........ ...
........
........ ...
...... ...
.
.
...... ..

P
........
....
...........
... ...........
.. ......
... .....
...
. .....
.....
.
.. ....
..
.. ....
. ....
..
.
. ....
..
.
.
....
...
.
.
. ...
.
. .

C2
..
.
.
.
.
.
.
.
.
..
.
.
.
.

If P is singular on either curve or if C1 and C2 have a common tangent line, then IP (C1 C2 ) 2.
The actual denition of IP (C1 C2 ) is a technical one and details can be found in Fulton ?]. In
particular, if P is an mf -fold singularity on C1 and an mg -fold singularity on C2 then IP (C1 C2 )
mf mg with equality if fmf and gmg have no line in common.
The basic theorem in intersection theory is:
X
Bezout's Theorem: IP (C1 C2 ) = d1 d2 :
P 2C1 \C2
In other words, the number of common solutions to f (x y) = 0 and g(x y) = 0 is the product of
their degrees d1 d2 . Counting is done with multiplicity, complex solutions are counted, and solutions
at innity must also be counted.
Example 15: Let f = (X 2 + Y 2)2 + 3X 2 Y ; Y 3 and g = (X 2 + Y 2 )3 ; 4X 2 Y 2 . The two curves
C1 dened by f = 0 and C2 dened by g = 0 intersect at P = (0 0). Fulton ?] shows that the
intersection multiplicity IP (C1 C2 ) = 14.
27
 Finally, we want to introduce a fundamental invariant of a smooth projective curve C . If C is
given by a homogeneous polynomial F (Z0 Z1 Z2 ) = 0 of degree d 1, and C is smooth, i.e., C
has no singular points including at 1, then we dene the genus gC of C to be
gC = (d ; 1)(2 d ; 2) :
The genus is actually a topological invariant of C viewed as a one-dimensional complex manifold.
Notice that for d = 1 2 3 4 5 : : : we have g = 0 0 1 3 6 : : : , so that certain values of g, notably
4, do not occur as the genus of a smooth plane curve.
There is also a notion of genus for singular projective plane curves. The formula involves sub-
tracting a correction term from (d;1)(2 d;2) depending on the nature of the singularities, specically
X mP (mP ; 1)
g = (d ; 1)(2 d ; 2) ; 2
P
where the sum is taken over all singular points P on the curve and over all singular points P that
arise during successive steps in the resolution process (these are known as innitely near singular
points). The value mP is the multiplicity of the singularity, i.e. P is an mP -fold singular point.
This g is actually the genus of the curve C^ obtained by resolving the singularities of C .
Example 16: A curve of degree four with two ordinary double points has g = 1. This case is
particularly relevant to an example we have in mind, that illustrates some geometric artifacts that
can be introduced when using resultants (see below).
We conclude this section by mentioning Harnack's theorem on real plane curves. A smooth real
plane curve consists of a number of ovals (see Harris ?, pp. 247{248]). Some cases look like those
pictured below:
..................................
......... .......
....... ......
.
................................................... ..... .
...... ... ..
.... ....................................... ......
... ... ......... ........
..... . .........
... ....
.........
... .....
.... ... ... ...
........ .... .. ..... ... .....
.............................................. ... . ... ..
... ... .. ..
.
.. ..
. .. ..
..
..
.
..
..
.
.
.
..
.. ..
.
... .... . ...
... .. ...
. ...
.
.
................................................... ... .... ... ..
.
................................... ...... ... ... . .
.... .....
.
........ ... ... ... ..... .............
.... ..... ..... . ..........
... .... .... .. . .
..............................................................
... ... ........ .... ..
...
. ... ............................................... ...
...... ....

(b)
.. .. ........ ......
.. .. ................ .......................
..... ..
.. .....

(d)
... ...
... ..
... .
.
... ...
.
... ...
.... .
...... ....
.............. .....................
......
....................................... ................ ...........
.......... .... ... ... ........
.
.

(a)
......
...... .... ... . .... ...
... ... ... .... .... ...
..
. . ... .... ... ...
.
. ... ...
. ...
. ..... ...
...
..
.
... .. ..
. ... ... ....
... ................. ... ...
. ... ... .
... .
..... ... .
.
. . .
..
. .. .....
.. .... .. ... ...
. .
.
..
.
. .
.
.
.. ... . .. .. . ... ...
.. . . ... ...
.. .....
.
... ...... .. .... ... . . .
.. ..................... ... ... ... ... ..
.
... ... ............ ..............
.... .. ...................... .. . . .............
... ..
. ..
... ....................... ..
.
.
... .
. ...... ...
... ........ ......
....
....... ........ ................ .......................
............ ... .....
...............................

(c) (e)

Note that we are looking at things projectively, so in the ordinary plane R2 some of the ovals may
go o to innity.
Theorem
(d;1)(d;2)
3 The maximum number of ovals a smooth plane curve of degree d can have is M0 =
2 + 1.

28
3.2 Intersections of Surfaces in Space
In this section, we examine the intersection of two surfaces in space. This is critical for an example
we have in mind. As before, we will frequently need to work over the complex numbers, and we will
also want to \compactify" our curves and surfaces by working in complex projective three-space
P3C .
Since the intersection of two surfaces S1 and S2 , given by f (x y z ) = 0 and g(x y z ) = 0
respectively, is a union of curves C1    Cr , it will be useful to say a few words about curves in
space before discussing surface intersections. (Note that in the projective setting S1 and S2 would
be given by the homogeneous polynomials F (X Y Z W ) = 0 and G(X Y Z W ) = 0 obtained by
\homogenizing" f (x y z ) and g(x y z ).)
Example 17: The surfaces given by xz ; y2 = 0 and y ; x2 = 0 intersect in two curves. One,
C1 , is the z-axis dened by x = 0, y = 0, and the other, C2, is the so-called \twisted cubic" given
parametrically by x = t, y = t2 , and z = t3 . For this same example in P3C , we would consider the
homogeneous quadratic equations XZ ; Y 2 = 0 and Y W ; X 2 = 0. At innity (W = 0) the only
solution is (X : Y : Z : W ) = (0 : 0 : 1 : 0). So the two curves C1 and C2 join at two points|the
origin (0 0 0) and this one point at innity.

.
..
... ...
...
.. ... ...
... ... ...
.. . ... ...
.. ... ...
... ... ..
.. ... ..
... ... ..
..
...
.
...
... ..
.
.. .......................................... .......................................
.

C2
... ......... . .................
... ... .
.......... .
.. .
... .......... .
... .
..
.
.
.. ..... ..
.... ..
.. .
...... ..... .
.
.
..
. . ..
...
.
........ ...
... .
... .
... .... ... ..
................................ .... .... ... ...
.... ... .. ... ... ...
...
. .... ... ... ...
... ..
.... ...... ... ... .
...
.......... ............ ..... .
...
... ...
............ ... ...
... ...
... .
. ... ...
... .. ... ...
... ... ... ......
... ... ... ....
... ... ......
...
...
.... ......
..... ..... ...
... ...... ...... ....
... ........... .......
... .......................... ...
...
... ...
... ...
... ...
... ...
... ...
... ...

C1
... .
...
...
...

C1 C2

Curves in Space
Every smooth irreducible curve C in space has two fundamental invariants, its degree d and its
genus g. The degree is dened to be the number of points in C \ H for a general plane H in three
space. For certain planes, C \ H will contain fewer points (or perhaps C will be contained in H ),
but for most planes, C \ H will consist of d points. The genus g is harder to dene. For smooth
projective curves C P3C it is the usual topological genus. What we want however is a second
notion of genus, called the arithmetic genus pa (C ), which equals g for smooth curves, but which
can be dened for any curve. One important aspect of pa (C ) is that it is a birational invariant and
so remains unchanged if the space curve C is appropriately projected into a plane.
Below is a table which illustrates possible combinations of d and g that can occur for a curve
in space:

29
 12 .
..
.
...
= exists
11
.
.. ..

10 ...
..
.
....
= does not exist
9 .
..
...
.
..
...
.
....
= don't know yet
8
.
.. ...

7 ..
..
..
..

.
...
.
...
..
.......
... ... ... ... ... ... ... ... ... ... ... ... ..
g =
1
2 (d ; 1)( d ; 2)
.
.. plane curve
6
.
........
g ..
..
..
....
....
.......
..

5 ..
.. ... .
........

;d
.. ...

4 1 d2
. ...

..
..
.
...
....
.......
..
...... ..... ..... ..... ..... ..... ..... .....
g =
4 +1
3 ...
..
..
......
....
... .
..
..
.

2 d;3
... ..
... ...... .......
...
..
.......... g =
1
.
. ..
.. ......... ......... ......... .....
.. ....
.
... .......
..
.. . .......

0 ... . ... .......

........ ..... .. .......
d
1 2 3 4 5 6 7 8 9 10

The most important example occurs when a curve C is the complete intersection of two surfaces
of degrees a and b. In that case, we have:
Proposition 1 If C is the complete intersection of two surfaces of degrees a and b respectively,
then the arithmetic genus of C , pa (C ) = 21 ab(a + b ; 4) + 1 and the degree of C is ab.
Example 18: Consider the intersection of the cone x2 +y2 ;z2 = 0 with the cylinder (x;2)2 +y2;1 =
0. These two surfaces intersect transversally because the Jacobian matrix
 !
2y ;2z 2x
2(x ; 2) 2y 0
has rank two at any point of the intersection. (To get all three 22 minors to have zero determinant,
one would need either y = 0, z = 0, or y = 0, x = 2, and it is easy to see that no point
on the intersection has these values.) Viewing things in P3C , where we must use the equations
X 2 + Y 2 ; Z 2 = 0 and (X ; 2W )2 + Y 2 ; W 2 = 0, we nd that the surfaces intersect in two points
at 1, namely (X : Y : Z : W ) = (1 : i : 0 : 0). The intersection is also transverse at these
! equations 1 + y2 ; z2 = 0 and (1 ; 2w)2 + y2 ; w2 = 0 to compute the Jacobian
points. (Use the
2i 0 0 which clearly has rank 2.) Thus the intersection is a smooth curve C of degree
2i 0 ;4
4 and arithmetic genus pa (C ) = 1. Because C is smooth, the topological genus g is also 1. The
topological picture is:
................................................................................................................................................
............................ ...................
................... ...............
............. ...........
....
............... ..........
.........
..
............. .......
.. ......... ......
.....
...
...... ......
....
.
...... ....
....
...... ...
. ...
.... ..... . ...
...... .... .....................................
.......... ... .... ...... ....................................... ...
... ............... ........ ............ ..........
. .
. ...... .. ..... . .... ...... ....
...........
. ..... ........ ..... ...
......
........ ...
. ..... ......
... ... ................................................................. .. ...
.
. . ... ...... ...... . .
.
. ................................ ............... ......... .
.... ......... ............... ...
.... . .. .......................................................................................... .... ....
.....
... ... ..... ....
. .
. ......
.
... ...... ... ... .. ....
.
. .. ... .
.. .. .. ...
... . ....... ... . ....
...... .
... ....... ...... ... .... ... .....
.. ....... ....... ....... ....... ....... . ...
... . ....... ....... ....... .... ...
... ..
.... ....
....
..... .
. ....
.
.
...... .....
....... ......
...... .......
........ ........
......... .........
..........
............. .
.....
..............
.................. .....
....................... .................
..................................... .......................
...........................................................................................................

where the two ovals are the real points. The two points at 1 are also shown.

30
 There is an analogue of Bezout's Theorem for the intersection of two surfaces S1 and S2 of
degrees a and b respectively. It says that
X
ab = IC (S1 S2 )degC
C
where the sum is taken over all irreducible curves C that occur in the intersection and IC (S1 S2 )
is the intersection multiplicity of S1 and S2 along C . For example, we previously considered two
quadrics xz ; y2 = 0 and y ; x2 = 0 intersecting in a line C1 which has degree 1 and the twisted
cubic C2 which has degree 3. The above formula gives 2  2 = 3 + 1. It follows that IC1 (S1 S2 ) and
IC2 (S1 S2 ) are both 1 and that no other curves appear in the intersection.
Curves on Quadric Surfaces
In the examples above, the curves involved all lie on quadric surfaces. In this section, we will
brie!y consider curves on smooth quadrics. As described in ?], any smooth quadric surface is
projectively equivalent to the one dened by Z0 Z3 ; Z1 Z2 = 0 in P3C . This surface, known as the
Serge surface is the image of the map
P1  P1 ;! P3
(X0 : X1 ) (Y0 : Y1 ) 7;! (X0 Y0 : X0 Y1 : X1 Y0 : X1 Y1 ):
To describe a curve on this surface, one species a bihomogeneous polynomial F (X0 X1 Y0 Y1 )
of bidegree (r s). This means that F should be homogeneous of degree r in the X variables and
homogeneous of degree s in the Y variables. The degree of the curve dened by an F of bidegree
(a b) is a + b.
Example 19: F (X0 X1 Y0 Y1 ) = X02 Y1 ; X12 Y0 has bidegree (2 1).
Example 20: Suppose F (X0 X1 Y0 Y1 ) has bidegree (m m ; 1). The resulting curve C on P1  P1
can be described as a determinantal variety in P3 . Namely, C P3 is just the set of 2  3 matrices
of the form
 !
Z0 Z1 G(Z0 Z1 Z2 Z3 )
Z2 Z3 H (Z0 Z1 Z2 Z3 )
that have rank
1. Here G and H are homogeneous polynomials of degree m ; 1. For example,
using F (X0 X1 Y0 Y1 ) = X02 Y1 ; X12 Y0 as above, the matrices in question take the form
 !
Z0 Z1 Z2
Z2 Z3 Z1
Those of rank one are precisely the points on the quadric surface that lie on the curve cut out by
F.
Plane Projections of Space Curves
Linear projection is a technique for reducing the number of variables in a problem. In general
a linear projection is dened as follows: Select a linear subspace Ld Pn of dimension d and a
second linear subspace L~ n;d;1 disjoint from Ld . The projection ' is a mapping
' : Pn ; Ld ;! L~ n;d;1 
= Pn;d;1 :
Specically, for p 2 Pn ; Ld consider the linear space $p of dimension d + 1 spanned by Ld and
p, and dene '(p) to be the point of L~ n;d;1 which is the intersection of $p with L~ n;d;1 . This is
called the projection from Ld .
31
Example 21: In space, P3 , let L be the point (zero dimensional linear subspace) x : y : t : w] =
1 : 0 : 0 : 0], and let L~ be the plane, P2 , dened by x = 0. The projection ' is just the map
' 2
P3 ; f1 : 0 : 0 : 0]g ;! P
x : y : t : w] 7;! y : t : w]:
Now if X is a subvariety of projective space Pn which doesn't meet Ld , we can consider its image
'(X ) Pn;d;1 under the projection. X is usually described by a system of (homogeneous)
polynomial equations f1 (X0 : : : Xn ) = 0 : : : fs (X0 : : : Xn ) = 0. The standard tool to go from
this system of equations to a system of equations in n ; d variables which describe '(X ) is the
resultant!
Example 22: Suppose C is a curve in space described by two equations:
ar (y t)xr + ar;1 (y t)xr;1 +    + a0 (y t) = 0
and
bs(y t)xs + bs;1 (y t)xs;1 +    + b0 (y t) = 0:
If we project C to the y t-plane, we get a curve '(C ) described by one equation, R(y t) = 0, in
two variables. R(y t) is computed via the resultant:
0 1
   00
BBar (0y t) ar;1 (y t)
ar (y t) ar;1(y t) 
a0 (y t)
a0 (y t) 0    0 C
CC
BB CC
BB 0   ar (y t)   a0 (y t)C
R(y t) = det B BBbs(y t) bs;1 (y t)  00 C
CC
BB 0 bs(y t)  00 CCC
BB ..
@ . 0 C A
0 0  bs (y t)   b0 (y t)
which is the determinant of an r + s by r + s matrix.
Now suppose C P3 is a smooth irreducible curve of degree d and arithmetic genus g. We
can project from any point p 62 C to a plane also disjoint from p. The result of this projection is
an irreducible plane curve '(C ) also of degree d and arithmetic genus g. The curve '(C ) however
may be singular. In fact, if g 6= (d;1)(2 d;2) then '(C ) will be forced to have singularities. We have
the following:
Theorem 4 The general linear projection of a smooth irreducible curve C P3 from a point
p 2 P3 , p 62 C is a plane curve having at worst only ordinary double points (nodes) for singularities.
From our formula for the arithmetic genus above, we see that the number of nodes must be
(d;1)(d;2); g.
2
Example 23: If a non-singular curve C is the intersection of two quadrics in space, it will have
degree four and arithmetic genus one. The general projection will be a plane curve '(C ) of degree
four with two ordinary double points.
Away from these double points the map from C to '(C ) is one-to-one. The picture is:

32
 ....
.............
........

..
.
.
....
....

...
.......
.....
p
....
.....
......
......
.....
...
..
..
.
...
......
......

C q
.....
. ...
.........
........
.........
..........
...............

.
...

P3
...
....
..
..
....
.....
........
...........
............
.............
..........
.........................................................

Note that for projections in special directions things might become bad.
Detailed Example
We conclude this section with a lengthy example. For the most part we will work in ordinary
three space, as opposed to projective three space. Thus our equations will not be homogeneous in
four variables, but rather just polynomials in three variables. To analyze behavior at innity it will
be necessary to homogenize the equations.
Our example starts with the curve C dened by the intersection of two conics:
x2 + (y ; 1)2 ; t2 = 0
  t 2  t 2 (8)
x; 4; 2 + y; ; 1 = 0: 2
In R3 with coordinates x y t, the rst equation represents a cone with vertex at (0 1 0)

t
6
..................................................................................................
............................................................... .....................
...............................................
.................................................................. .............................
. ............................................................................
.
.
.
.
.
.
.
.
.
.

(0 1 0)
y
-

x
... ... ... ... ... ... ... ... ... ... ... ... ...
... ... ... ... ... ... ... ... ...
........................... .
.......................................... ....................
............................................................................................

and the second equation represents a slanted cylinder:

33
 ..............................................................
......................
...........................................................

..... .... ..... .... ..... ... .

................. .
...........................................................

Another interpretation is that the rst equation represents an expanding circle in the plane and
that the second equation represents a moving circle. Pictured below is the situation at time t = 2:

y6
.......................................................
............. ..........
......... .......
........ .......
....
. ....... ......
..... ......
..... .....
.......
.
.....
.....
..
..... ....
....
...... ....
... ...
.
. ...
.
.... ...
... ......
... ................ .......................
. ... ....... ......
... ... ...... .....
. ... ..... .....
.... .. ......
. ....
.. ... .. ....
.. ... ..... ...
... .. ... ...
.. .. .. ..
.. .
...... . ...
.
. ..

(0 1) (2 1)
.. . .
.
. .....
.. ..
..
. ...
.... . ..
..

(3 1)
.. ...
.
. ... ..
... ...
.
. . ...
.
.
.. ...... ..
... .
. ..
... .. ... ...
.
... ... ... ..
... ... ..... ...

-x
... ... .... ....
... .
.... .....
..
. ..
.......
... ......
...
. ....... ......
... .......... .......
... .. ..................................
... ...

(3 0)
... ...
...
.... .
....
..
.... ....
.... ....
..... ....
..... .....
..... .....
......
...... . .
.......
.
....... ......
........ .......
.......... .......
.............. .........
..................................................

Notice that x = 2, y = 1, t = 2 is the point of \rst contact".

Projectively, we would need to work with the homogenized versions of the equations (??):
x2 + (y ; w)2 ; t2 = 0
  2  3
2 x ; 4w ; t + y ; t ; w 2 = 0 :
2
The points at innity in P3 correspond to solutions with w = 0. There are exactly four such points
in P3 , namely
 3 i 3 i 
(x : y : z : t) =(;i : 1 : 0 : 0) (i : 1 : 0 : 0) ;4 + 4 : 4 + 4 : 1 : 0
 3 i 3 i 
or ; 4 ; 4 : 4 ; 4 : 1 : 0 :
With a little work one can show that C P3 is a smooth irreducible curve (as usual we are assuming
that we are working over the complex numbers C ).
The real solutions form two ovals in R3 :

34
 ....................... .
..........
... ...
... ..
..
.. ...
.. ....................................
................................................................................. .............................................
.................................. . .. .
........................................................ .. ... ...............
........................................................................................................................
. .
.. .
.
.
.
.................
.
..... ... ...
... ...
... .
... ......
....... ......
.. ..... ...
.. ..
.
... ...
... ...
... ..
.
..
... ...
... ...
... ...
... ..
.
... ..
..
... ..
...........................
.
... ..
... ...
... ...
... ...
... ..
.
... ...
. ...
... ..
... ..
. ...
... ...
... .
... ...............
...... .......
.... . .....
.. .
.
....
.. ..
.. ......
.. .. . .
............. ..
.. .
.
.
. ..
... ... ... ... ... ... ... ... ... .... ... .... ... ... ... ...
................. .. ... ... ... .. .. .................... ..
............................. .
. ..............
............................................................................... .............
.. ..
.. ..
.. .
.
... .
..... ...........
.......

Now we will project C into the y t-plane, the plane dened by x = 0. This is a linear projection
' from the point (1 : 0 : 0 : 0) in P3 to the plane x = 0. Note that (1 : 0 : 0 : 0) 62 C so that the
projection ' is well-dened on all of C . To nd the equation for '(C ) P2 , we write our equations
as polynomials in x:
x2 + (y2 ; t2 ) = 0
4x2 + (4t ; 32)x + ((t ; 8)2 + (2y ; t)2 ; 4) = 0:
We then eliminate x by taking the resultant:
0 1
1 0 y2 ; t2 0
B0 1 2 2C
B@4 4t ; 32 0? y ;0 t CCA ? = ((t ; 8)2 + (2y ; t)2 ; 4):
det B
0 4 4t ; 32 ?
The result (after dividing by 4) is
g(y t) = (8t2 ; 80t + 272)y2 + (;12t3 + 48t2 ; 48t ; 288)y + (5t4 + 16t3 ; 20t2 ; 512t + 1040)
(9)
and the projection '(C ) is described by g(y t) = 0, or in projective space P2 with homogeneous
coordinates y t w by:
0 =8t2 y2 ; 80ty2 w + 272y2 w2 ; 12t3 y + 48t2 yw ; 48tyw2
(10)
; 288yw3 + 5t4 + 16t3 w ; 20t2 w2 ; 512tw3 + 1040w4 :
This is a plane curve of degree 4. We expect this degree because the original C P3 had degree
4, being the intersection of two quadrics. Now a smooth plane curve of degree 4 has arithmetic
genus 3 while our curve has arithmetic genus 1. This means our curve has some singularities. In
general, we expect it to have two ordinary double points (nodes), but it could have cusps instead,
or it could have a single tacnode. These singularities occur as a result of the projection:

35
 ...
....

........ .
........ ........
...
.......
p
.
.......
.....
......
......
....
........
..
.
.....
p
p
........ ...... .........
........ . ....
........ ...... . . .
..... .......... .
.
.
..........
. . .
.
.. ..
. ........
. ..... . . ... .. ...
.......... ........
. ...... . . .... . .. ... .. .........
.

C
............ ...
... .. .... . .... . .. . . .......
. .... ................ .......

q
.. .......... ... .. .. .. . . ..
... ........... .. . .. .
. .

C
.. ........ .. .. .. . .
.... . ......... . . ..
.. .. ... .. ......
....... .. .. .. . . .... ........ .. .. .. .. .
.. .. ..... . .......... .. .. . .
.
. .........
.. . .. .. .. .. . . .. ..
..
.. . .... ..
..........
............. .. .. .. . . ..... .
. .. . .. .. . . .. ..
........ .. .. .. .. .
.. . ... .. .. ............. . . ........ . .. ..
. ............ . ......... .. .. .. . .
.. ....................... . .. . ............ . ..
.........
. . .. ..

C q
... . . ........ . ... . .
.. ... . . . ........ .. . .. .......
.
..
.........
. .. . .. . . . ..
.. .... . . .. ....... .. .. .. .. .
........ .. . . . . . . .
...... .. .. .. ........
.. ......... .. .. .. .. . . . .

q
.. .. .. .. .. .. .
..
..
......... . . ..
........ ......................................... .. .. .
..... .. . . . . . . .
.
. . .. .... .. . ..
...... ... . .. . . ..... ........ . . . .
. . . ..
.. ..
... . .. .. .. .. . ..... . . . .
. . . ..
. .
.. .. ...... .. .. .. .. .. .......... .
. . ..
. .
.. .
.. .. . .. . .
.. .... . . .. .. . ..... . . ..
.. . .. ...... .. .. ..........
. . .. .. .. .. . .
.

P3
. . . . . .. . . .
. . . .. ..
.. .
. .. .. .. . .. . . .
. .
. ..
.. .
. .. .. . .
. .. .. .. . .
.
.. . .. . .. . .
.
. . ..
.. . . . .. . . .. .. .. . .
.
..
.. . .. .. .. .. . . . . . ..
. . . . . .. .. .. . .
.
.
.. . .. .. .. .. . . . .. . . ..
.
. .
.
. .
. .. .. . . ..
.. .. .. .. .. .. .. . .. .. . . . ..
. . . .. . .. . . .
. .. . . .. . . . .. .. .. .. ..
.. . ................................................................. .. . .. .
. ...
.. . ........................ .. .. .....
. .. .
. ..
... .... .
.
. .
. ..
. .....
.
.
. .. .. ...... ...
.... .. .
. .
. .
. ..............
... ....
. ....... ..
.. .........................
.
. .
. . ..
. . . . . .
... ... . . .
...
.......... .....
....... ..
.. . .
. . ..... .. .......
....... .
. .
. .. ..

'(C )
... .... .. .
...... ..... .
.
. .... .. .... .....
...............
. ........ . . ...
. .. ..... .
.....
. .
. . ..
.................................................................................... .........
.......... .. . .. ...... .
....... .
. . ... .. . . . .
.........
........ ..
. ................ . ..... ..
.. ............. . . . ...... .
....................
.................... . ....... .. ... ..
..........
. . .. .......... ............ ......
. . ... .
. .........
.................................................. ..
. ............... ................................................... .............

'(C ) '(C )
. .
.... ...........
........
.
. ........... ..
....
. ...........
....... .......
........
.. ..
...
...
...

We will see below that '(C ) does in fact have two nodes.
What do the real points on '(C ) in R2 look like. We can view
0 = (8t2 ; 80t + 272)y2 + (;12t3 + 48t2 ; 48t ; 288)y + (5t4 + 16t3 ; 20t3 ; 512t + 1040)
as a quadratic equation in y of the form
Ay2 + By + C = 0:
The discriminant is B 2 ; 4AC , which equals
;16t6 ; 64t5 + 3776t4 ; 5120t3 ; 200704t2 + 917504t ; 1048576:
The roots of the discriminant are ;16, ;9:40312, 2, 3.40312, and a double root at 8. This double
root is a bit curious. It is an isolated point at (y t) = (13 8).

13 6
y
(13,8)

1
-t
;16 ;9:40312 2 3.40312 8

We see the two ovals we expect. When t = 2, we have (y t) = (1 2) being the point of \rst
contact" discussed above. The other times t = ;16, ;9:40312, and 3.40312 also correspond to
when the two circles are tangent.

36
 One can easily show that (y t) = (13 8) is an ordinary double point. We change coordinates to
move this point to the origin by setting
t = r + 8 and y = s + 13
in g(y t). This leads to
12(43r2 ; 28rs + 12s2 ) + higher order terms:
Since no linear term is present (13 8) is a singular point, and because 43r2 ; 28rs + 12s2 factors
into two distinct (complex) linear factors, p point. Two points on C project
p it is an ordinary double
to (y t) = (13 8) they are (x y t) = (4 ;5 13 8) and (;4 ;5 13 8).
C must have another singularity, which in this case occurs at innity. Setting  3 wi = 0 in (??)
gives
3 i us the points
 of C at innity. They are ( y : t : w ) = (1 : 0 : 0), 4 + 4 : 1 : 0 , and
4 ; 4 : 1 : 0 . It is clear that (x : y : t : w) = (i : 1 : 0 : 0) and (;i : 1 : 0 : 0) on C both project
to (y : t : w) = (1 : 0 : 0).
To see what '(C ) looks like near (1 : 0 : 0), we set y = 1 in (??) to get:
8t2 + 272w2 ; 80tw + higher order terms:
Again, 8t2 +272w2 ; 80tw factors into two distinct complex lines showing that (1 : 0 : 0) is a second
node on '(C ).
Apart from these two points (y : t : w) = (13 : 8 : 1) and (1 : 0 : 0) on '(C ) and the four points
above them on C , the projection from C to '(C ) can be shown to be one-to-one. Each smooth
point on '(C ) has exactly one point on C projecting to it. This concludes our example. It is an
excellent example of the kinds of geometric, algebraic, and numerical phenomena that can occur
when we use resultants to eliminate variables.

37
4 Complexity and Computational Issues
4.1 Resultants and Combinatorial Methods
In this section we brie!y address certain complexity and computational issues related largely to
the number of terms that can appear in the classical Sylvester resultant.
Terms and Coecients in the Resultant
We begin by attempting to understand what monomials and what coecients can occur in a
resultant. As easy examples show (see section 1 above), the number of monomials that actually
occur is far less than the maximum (the total number of monomials of the appropriate degree and
homogeneity).
Let f (x) = am xm +    + a1 x + a0 and g(x) = bn xn +    + b1 x + b0 and write
X
Rm n(f g) = cpq apbq
pq
where
p = (p0 : : : pm ) q = (q0 : : : qn )
p p p p
a = a00 a11 : : : amm bq = bq00 bq11 : : : bqnn
P P
and p 2 Zm+ +1 with mi=0 pi = n and q 2 Zn++1 with nj=0 qj = m. Here Z+ is the set of non-negative
integers. P
We denote by m (n) the set of all p = (p0 : : : pm ) 2 Zm+ +1 with mi=0 pi = n.
Example 24:

2 (2) = f(2 0 0) (0 2 0) (0 0 2) (1 1 0) (0 1 1) (1 0 1)g:

Note that cpq = 0 unless p 2 m (n) and q 2 n(m).

Now given q 2 n(m) dene a symmetric polynomial Mq (x1 : : : xm ) as the sum of all mono-
mials x1 1 x2 2 : : : xmm such that q0 of the exponents i are equal to n, q1 of the exponents are equal
to n ; 1, etc., down to qn of the exponents equaling 0.
Example 25: If m = n = 2 and q = (q0 q1 q2 ) = (1 0 1), then
Mq (x1 x2 ) = x21 x02 + x01 x22 = x21 + x22
because we must have all monomials with one exponent equal to 2 and one equal to 0.
Example 26: If m = n = 2 and q = (q0 q1 q2 ) = (0 2 0), then
Mq (x1 x2 ) = x11 x12 = x1 x2
because we must have two exponents equal to 1.
The Mq (x1 : : : xm ) are clearly symmetric polynomials P and so can be written in terms of the
elementary symmetric functions ei (x1 : : : xm ) = 1j1 <<jim xj1 : : : xji , for i = 0 : : : m.
Example 27: m = 2, e0 = 1, e1 = x1 + x2 , and e2 = x1x2 .
The next result tells how to compute the coecients cpq , p 2 m (n), q 2 n (m) in the resultant
Rm n(f g).
38
Theorem
P 5 For p = (p0 : : : pm) with pi 0 and Ppmi=0 pi = nq and q = (q0 : : : qn) with qj 0
and nj=0 qj = m, the coe cient cpq of the monomial a00 : : : apmm b00 : :P: bqnn in the resultant Rm n (f g)
where f ( x) = a x m +    + a0 and g(x) = bn xn +    + b0 is (;1) m i=0 ipi times the coe cient of
Q m e (x
m
: : : xm )pi in the expansion of Mq (x1 : : : xm ) in terms of the elementary symmetric
i=1 i 1
functions ei (x1 : : : xm ).
Example 28: m = 2, n = 2, p = (p0 p1 p2 ) = (1 0 1) and q = (q0 q1 q2 ) = (1 0 1). We
are looking for the coecient of a0 a2 b0 b2 in the resultant of two quadratic polynomials. (Note
a0a2 b0b2 = a10 a01 a12 b10 b01b12 .) Example 25 shows Mq (x1 x2 ) = x21 + x22 and in terms of the elementary
symmetric functions (Example 27),
Mq (x1 x2 ) = x21 + x22 = (x1 + x2 )2 ; 2x1 x2 = e21 ; 2e2 :
The coecient cpq we are looking for is (;1)2 times the coecient of e2 . Thus cpq = ;2 which is
correct (see the last term in Example 1).
Algorithm. This yields an algorithm for nding the coecients of the resultant based on the
algorithm for writing a symmetric polynomial in terms of the elementary symmetric polynomials.
4.2 Partitions, Symmetric Functions, and Newton Polytopes
Partitions play an important role in the representation theory of the symmetric group. They also
play an important role in determining which terms can occur in the resultant.
Denition 7 A partition of length
m is just a sequence of non-negative numbers 1 2
   m which we denote by = ( 1 2 : : : m ). The numbers i are the parts of .
Each partition gives rise to a symmetric polynomial
X
m (x1 : : : xm ) = x1 1 : : : xmm
(1 ::: m )=(1 ::: m )
where runs through all distinct permutations of ( 1 : : : m ).
Example 29: = (2 1)
m(2 1) (x1 x2 ) = x21 x12 + x11x22 = x21 x2 + x1 x22 :
Example 30: Mq (x1 : : : xm ) dened above for q = (q0 : : : qn ), qj 0, Pnj=0 qj = m coincides
with m (x1 : : : xm ) for the partition
: : n} n| ; 1 : {z: : n ; 1} : : : 1| : {z: : 1} 0| : {z: : 0}):
= (n| : {z
q0 times q1 times qn;1 times qn times
P
Notice has nj=0 qj = m parts.
Example 31: The elementary symmetric polynomials ei (x1 : : : xm ) are also of the form m (x1 : : : xm )
: : 1} 0| : {z: : 0} ). Notice has m parts.
for the partition = (1| : {z
i times m;i times
Now the m as runs over all partitions with m parts form a basis for the innite dimensional
vector space of symmetric polynomials in m variables x1 : : : xm . A dierent basis is given by
Y
m
ei (x1 : : : xm )pi :
i=0

39
We encode this by a partition  having pi parts equal to i for i = 0 : : : m
 = (m | ; 1 : {z: : m ; 1} : : : 1| : {z: : 1} 0| : {z: : 0})
| : {z: : m} m
pm times pm;1 times p1 times p0 times
and we write
Y
m
e (x1 : : : xm ) = ei (x1 : : : xm )pi
i=0
for any partition  with parts
m but of any length. Note: we dene e0 (x1 : : : xm ) = 1.
A key observation is that the coecients in the resultant are essentially the entries in the change
of basis matrix between the m and the e . Unfortunately this doesn't yield much information.
However, if we write
X
e = d m

much can be said about the d . If = ( 1 : : : m ) and  = (1 : : : n ) with j
m for every
j , then d is equal to the number of m  n matrices with 0,1 entries having row sums 1 : : : m
and column sums 1 : : : n .
Example 32: m = 2, n = 2, = (2 0),  != (2 0) then d = 0 as no 2  2 matrix can have the
required row and column sums, i.e., ac db , a b c d 2 f0 1g with a + b = 2, c + d = 0, a + c = 2,
and b + d = 0. Notice that
Y2
e = ei (x1 x2 )pi = e12 e01 e10 = e2 = x1 x2
i=0
because  = (2 0) has one part equal to 2, no parts equal to 1, and one part equal to 0. To check
that d = 0, note that
m(1 1) = x1x2
so that e = m(1 1) and as m = m(2 0) doesn't occur in this relationship,
 ! we have d = 0. If
we had chosen  = (1 1) then d would be 1 because the matrix 10 10 has row sums 2,0 and
column sums 1,1, and it is the only such matrix. The d play an important role in combinatorics
and in the representation theory of nite groups.
We must dene two concepts:
Denition 8 If = ( 1 : : : m ) and v = (v1 : : : vm ) are two partitions of length m. We say
that v dominates and write v if v1 +    + vi 1 +    + i for i = 1 : : : m ; 1 and
v1 +    + vm = 1 + : : : + m .
Example 33: (4 1 1) dominates (3 2 1) because 4 3, 4 + 1 3 + 2, and 4 + 1 + 1 = 3 + 2 + 1.
Denition 9 If  = (1 : : : n) is a partition with parts j
m for all j and if  has pi parts
equal to i, then we dene the conjugate partition  to be:
u = (p1 +    + pm p2 +    + pm : : : pm ):

40
Example 34: m = 3,  = (3 2 2 0) then u = (3 3 1) because p0 = 1, p1 = 0, p2 = 2, and p3 = 1.
We can now bound which terms appear in the resultant.
Theorem
P 6 Let p = (p0 : : : pm ) with pi 0 and Pmi=0 pi = n and q = (q0 : : : qn) with qj 0
and nj=0 qj = m and let  be the following partitions:
: : n} n| ; 1 : {z: : n ; 1} : : : |1 : {z: : 1} 0| : {z: : 0})
= (n| : {z
q0 q1 qn;1 qn
| ; 1 : {z: : m ; 1} : : : 1| : {z: : 1}):
| : {z: : m} m
 = (m
pm pm;1 p1
(Note: has length m and  has parts j
m.) Then cpq = 0 and the term ap00 : : : apmmPbq0m0 : : : bqnn
not appear in the resultant unless 
. Moreover if  = then cpq = (;1) i=0 ipi =
does P
(;1) i odd pi .
Theorem 7 The pairs (p q) = (p0 : : : pm q0 : : : qn) 2 Zn++m+2 with  = areRn+m+2
exactly the vertices of the Newton polytope of the resultant. Thus a term occurs in Rn m (f g) only
if it is in the convex hull of this set of points.
Note: Without loss of generality, we can assume  has p0 parts equal to 0, so that  will have
length n.
Example 35:
f (x) = a2 x2 + a1 x + a0 a2 6= 0
g(x) = b2 x2 + b1 x + b0 b2 6= 0
p = (p0 p1 p2 ) and q = (q0 q1 q2) must be in the set f(2 0 0) (0 2 0) (0 0 2) (1 1 0) (0 1 1) (1 0 1)g.
Thus there are at most 36 monomials in the resultant R2 2 (f g). We make a table
 u
(2 0 0) (2 2) (0 0) (0 0)
(0 2 0) (1 1) (1 1) (2 0)
(0 0 2) (0 0) (2 2) (2 2)
(1 1 0) (2 1) (1 0) (1 0)
(0 1 1) (1 0) (2 1) (2 1)
(1 0 1) (2 0) (2 0) (1 1):
We then list the partitions which a given dominates:

(2 2) f(2 2)g
(1 1) f(1 1)g
(0 0) f(0 0)g
(2 1) f(2 1)g
(1 0) f(1 0)g
(2 0) f(2 0)(1 1)g:

41
This leads to seven pairs for which  , six of which have =  . We list the corresponding
(p0 p1 p2 ) and (q0 q1 q2 ). Remember that q goes with and that p goes with , and that p goes
with the a's and q with the b's
(q0 q1 q2 ) (p0 p1 p2 )
(2 0 0) (0 0 2) a22 b20
(0 2 0) (1 0 1) a0 a2 b21
(0 0 2) (2 0 0) a20 b22
(1 1 0) (0 1 1) a1 a2 b0 b1
(0 1 1) (1 1 0) a0 a1 b1 b2
(1 0 1) (0 2 0) a21 b0 b2
(1 0 1) (1 0 1) a0 a2 b0 b2 :
P2
The coecients of the rst six terms are given by (;1) i=0 ipi . Thus the resultant is
R2 2 (f g) = a20 b22 + a21b0 b2 + a22 b20 ; a0 a1b1 b2 ; a1 a2b0 b1 + a0 a2b21 + a0a2 b0 b2
which agrees with Example 1 except for the one unknown coecient .
The Newton Polytope of the Resultant
The section above describes an algorithm for nding the Newton polytope of the resultant.
Recall that the Newton polytope N (f ) of a polynomial in several variables
X
f = f (x1 : : : xm ) = c x1 1 : : : xmm
=(1 ::: n )2Zn+
with only nitely
many c 6=0
is dened to be the convex hull in Rm of the integral lattice points  = (1 : : : m ) 2 Zm+ for
which c 6= 0.
Example 36: f (x1 x2 ) = 1 + x1 + x2 + x21 x22 has N (f ) equal to the convex hull of
f(0 0) (1 0) (0 1) (2 2)g, i.e., N (f ) looks like:

6
2

-
0 1 2

Notice that (1 1) is also in N (f ) but the monomial x1 x2 does not occur in f . Thus knowing the
Newton polytope of a polynomial only bounds what terms can occur.
We can prove a few facts about the Newton polytope Nm n of the resultant Rm n (f g).

42
Theorem 8 The polytope Nm n has dimension m + n ; 1 in Rm+n+2 and lies in the linear space
cut out by
X
m X
n X
m X
n
pi = n qj = m (m ; i)pi + (n ; j )qj = mn:
i=0 j =0 i=0 j =0

Example 37: When m = 2 and n = 2 the rst two equations yield 36 possibilities for (p0 p1 p2 )
and (q0 q1 q2 ). The third condition forces
2p0 + p1 + 2q0 + q1 = 4
reducing us to the following possibilities:
(p0 p1 p2 ) (q0 q1 q2 )
(2 0 0) (0 0 2)
(0 2 0) (0 2 0) or (1 0 1)
(0 0 2) (2 0 0)
(1 1 0) (0 1 1)
(0 1 1) (1 1 0)
(1 0 1) (1 0 1) or (0 2 0)
for a total of 8 possibilities of which only seven actually occur. The pair that does not appear is
(0 2 0) (0 2 0) which corresponds to the monomial a21 b21 .
Example 38: N2 2 has dimension 3. It can be realized in R3 and is pictured below.
Theorem 9 If m n 2 then Nm n has exactly mn + 3 faces.
Example 39: m = 2, n = 2, then N2 2 has 7 faces. N2 2 is pictured below with its 6 vertices. Since
it has 7 faces and 6 vertices it must have 11 edges as #vertices ; #faces + #edges must equal 2.

a20 b22
..... ..
;2a0 a2b0 b2 a22 b20
. ....
...... ...
..... ... ... ........
..... ... ... . ... ....
..... ... .. ... .. ........
..... . ... .. ... ... .....
.
..... ... .. .. .....
..... . ... ... .. ....
..... ... .. . ... .....
..... . ... .. .. ....
..... ... .. . ... .....
.....
..... . ... ... ... .. ........
... .. . .
..... . ... ... .. ...
.....
.....
..... ... ... ... .....

a21b0 b2
..... . ... ... ... ... ... ... . .....
..... ... ... .. .. ... ... .....
..... ... ... ... ... ... ... .. ........
. ... . .
... ........... ........ ... ... ... ..
. ... ... ..... .....
.. ... .. .....
. ... ...
............ . .....
....
... ....

;a0a1b1 b2 ;a1a2b0 b1
.....
........ ..... ..... ........
........ ..... .... .......
........ ...
. ........
........
........
.....
..... .
......
. ..
.. ..
. ........
..... ..... .....
........ ..... .... ........
........ ..... ..... ........
........
..... ..... ........
........
..... ........
........ ..... ........
........
........ ......... ..
...... ..............
........ ..... .. ...
........ .... ........................
.........................
....

a0 a2 b21

Note that the one non-vertex point (1 0 1 1 0 1), which corresponds to a0 a2 b0 b2 , is the midpoint
between (2 0 0 0 0 2) and (0 0 2 2 0 0).
There are additional combinatorial ways to describe the vertices of Nm n but they involve
considerable technicalities.

43
Example 40: As a nal example, we work out the m = n = 3 case. Let f (x) = a3 x3 +a2 x2 +a1 x+a0,
a3 6= 0, and g(x) = b3 x3 + b2 x2 + b1 x + b0, b3 6= 0. The resultant R3 3 (f g) is bihomogeneous of
bidegree (3,3). There could be as many as 400 terms. We list the 20 possibilities for (p0 p1 p2 p3 )
and (q0 q1 q2 q3 ), and the associated partitions   .

 
(3 0 0 0) (3 3 3) (0 0 0) (0 0 0)
(0 3 0 0) (2 2 2) (1 1 1) (3 0 0)
(0 0 3 0) (1 1 1) (2 2 2) (3 3 0)
(0 0 0 3) (0 0 0) (3 3 3) (3 3 3)
(2 1 0 0) (3 3 2) (1 0 0) (1 0 0)
(2 0 1 0) (3 3 1) (2 0 0) (1 1 0)
(2 0 0 1) (3 3 0) (3 0 0) (1 1 1)
(0 2 1 0) (2 2 1) (2 1 1) (3 1 0)
(1 2 0 0) (3 2 2) (1 1 0) (2 0 0)
(0 2 0 1) (2 2 0) (3 1 1) (3 1 1) :
(0 0 2 1) (1 1 0) (3 2 2) (3 3 1)
(0 1 2 0) (2 1 1) (2 2 1) (3 2 0)
(1 0 2 0) (3 1 1) (2 2 0) (2 2 0)
(1 0 0 2) (3 0 0) (3 3 0) (2 2 2)
(0 1 0 2) (2 0 0) (3 3 1) (3 2 2)
(0 0 1 2) (1 0 0) (3 3 2) (3 3 2)
(1 1 1 0) (3 2 1) (2 1 0) (2 1 0)
(1 1 0 1) (3 2 0) (3 1 0) (2 1 1)
(1 0 1 1) (3 1 0) (3 2 0) (2 2 1)
(0 1 1 1) (2 1 0) (3 2 1) (3 2 1)

44
 We now list the partitions for which .
  
(3 3 3) (3 3 3)
(2 2 2) (2 2 2)
(1 1 1) (1 1 1)
(0 0 0) (0 0 0)
(3 3 2) (3 3 2)
(3 3 1) (3 3 1) (3 2 2)
(3 3 0) (3 3 0) (3 2 1) (2 2 2)
(2 2 1) (2 2 1)
(3 2 2) (3 2 2)
(2 2 0) (2 2 0) (2 1 1)
(1 1 0) (1 1 0)
(2 1 1) (2 1 1)
(3 1 1) (3 1 1) (2 2 1)
(3 0 0) (3 0 0) (2 1 0) (1 1 1)
(2 0 0) (2 0 0) (1 1 0)
(1 0 0) (1 0 0)
(3 2 1) (3 2 1) (2 2 2)
(3 2 0) (3 2 0) (3 1 1) (2 2 1)
(3 1 0) (3 1 0) (2 2 0) (2 1 1)
(2 1 0) (2 1 0) (1 1 1)
This leads to 34 terms 20 terms
(q0 q1 q2 q3 ) (p0 p1 p2 p3 )
(3 0 0 0) (0 0 0 3) ;a33b30
(0 3 0 0) (1 0 0 2) +a0 a23 b31
(0 0 3 0) (2 0 0 1) ;a20a3 b32
(0 0 0 3) (3 0 0 0) +a30 b33
(2 1 0 0) (0 0 1 2) +a2 a23 b20 b1
(2 0 1 0) (0 0 2 1) ;a22a3b20 b2
(2 0 0 1) (0 0 3 0) +a32 b20 b3
(0 2 1 0) (1 0 1 1) ;a0a2 a3b21 b2
(1 2 0 0) (0 1 0 2) ;a1a23b0 b21
(0 2 0 1) (1 0 2 0) +a0 a22 b21 b3
(0 0 2 1) (2 0 1 0) +a20 a2 b22 b3
(0 1 2 0) (1 1 0 1) +a0 a1 a3 b1 b2
(1 0 2 0) (0 2 0 1) ;a21a3b0 b22
(1 0 0 2) (0 3 0 0) ;a31 b0b23
(0 1 0 2) (1 2 0 0) +a0 a21 b1 b23
(0 0 1 2) (2 1 0 0) ;a20a1b2 b23
(1 1 1 0) (0 1 1 1) +a1 a2 a3 b0 b1 b2
(1 1 0 1) (0 1 2 0) ;a1 a22b0 b1b3
(1 0 1 1) (0 2 1 0) +a21 a2 b0 b2 b3
(0 1 1 1) (1 1 1 0) ;a0a1a2 b1b2 b3

45
 P
where the sign  is equal to (;1) ipi = (;1)p1 +p3 , and 14 terms
(2 0 1 0) (0 1 0 2) c1 a1a23 b20 b2
(2 0 0 1) (0 1 1 1) c2 a1 a2 a3 b20 b3
(2 0 0 1) (1 0 0 2) c3 a0a23 b20 b3
(0 2 0 1) (1 1 0 1) c4 a0 a1 a3 b21 b3
(1 0 2 0) (1 0 1 1) c5 a0 a2 a3 b0 b22
(1 0 0 2) (1 1 1 0) c6 a0 a1 a2 b0 b23
(1 0 0 2) (2 0 0 1) ;c3 a20 a3 b0 b23
(0 1 0 2) (2 0 1 0) ;c1 a20 a2 b1 b23
(1 1 1 0) (1 0 0 2) ;c6 a0 a23 b0 b1 b2
(1 1 0 1) (0 2 0 1) ;c4 a21 a3 b0 b1 b3
(1 1 0 1) (1 0 1 1) c7 a0 a2 a3 b0 b1 b3
(1 0 1 1) (1 0 2 0) ;c5 a0 a22 b0 b2 b3
(1 0 1 1) (1 1 0 1) ;c7 a0 a1 a3 b0 b2 b3
(0 1 1 1) (2 0 0 1) c2 a20 a3 b1 b2 b3
where there are 7 unknown coecients (due to symmetry).

References
1] J. Canny, Generalized characteristic polynomials, Journal of Symbolic Computation 9 (1990),
241{250.
2] Cox, Little, and O'Shea, Ideals, Varieties, and Algorithms, Undergraduate Texts in Mathe-
matics, Springer-Verlag, 1992.
3] W. Fulton, Algebraic Curves, Benjamin, Inc., 1969.
4] I.M. Gelfand, M.M. Kapranov, and A.V. Zelevinsky, Discriminants, Resultants,and Multidi-
mensional Determinants, Birkh&auser, Boston, 1994.
5] Gleeson and Williams, A Primer on Polynomial Resultants, Naval Air Development Center
Technical Report, 1991.
6] J. Harris, Algebraic Geometry: A First Course, Graduate Texts in Mathematics 133, Springer-
Verlag, 1992.
7] A.P. Morgan, Solving Polynomial Systems Using Continuation for Engineering and Scientic
Problems, Prentice-Hall, Inc., 1987.
8] B. Roth, Computation in kinematics, Computational Kinematics (J. Angeles et al., eds.),
Kluwer Academic Publishers, 1993.
9] G. Salmon, Lessons Introductory to the Modern Higher Algebra, Chelsea Publishing Co., Bronx,
New York, 5th edition.
10] J.P. Serre, A Course in Arithmetic, Graduate Texts in Mathematics 7, Springer-Verlag, 1971.
11] B.L. van der Waerden, Modern Algebra 1 and 2, Frederick Ungar Publishing Co., 1949 and
1950.

46