Enumerative Algebraic Geometry of Conics

Enumerative Algebraic Geometry of Conics
Andrew Bashelor, Amy Ksir, and Will Traves
1. INTRODUCTION. In 1848 Jakob Steiner, professor of geometry at the Univer-

sity of Berlin, posed the following problem [19]: Given five conics in the plane, are
there any conics that are tangent to all five? If so, how many are there? Problems that
ask for the number of geometric objects with given properties are known as enumera-
tive problems in algebraic geometry. The tools developed to solve these problems have
been used in many other situations and reveal deep and beautiful geometric phenom-
ena.
In this expository paper, we describe the solutions to several enumerative problems
involving conics, including Steiner’s problem. The results and techniques presented
here are not new; rather, we use these problems to introduce and demonstrate several
of the key ideas and tools of algebraic geometry. The problems we discuss are the
following: Given p points, l lines, and c conics in the plane, how many conics are
there that contain the given points, are tangent to the given lines, and are tangent to
the given conics? It is not even clear a priori that these questions are well-posed.
The answers may depend on which points, lines, and conics we are given. Nineteenth
and twentieth century geometers struggled to make sense of these questions, to show
that with the proper interpretation they admit clean answers, and to put the subject of
enumerative algebraic geometry on a firm mathematical foundation. Indeed, Hilbert
made this endeavor the subject of his fifteenth challenge problem.
Enumerative problems have a long history: many such problems were posed by
the ancient Greeks. Enumerative geometry is also currently one of the most active
areas of research in algebraic geometry, mainly due to a recent influx of ideas from
string theory. For instance, mirror symmetry and Gromov-Witten theory are two hot
mathematical topics linked to enumerative geometry; both areas developed rapidly
because of their connection to theoretical physics. While we will not discuss these
subjects explicitly in the main part of this paper, many of the ideas and techniques we
introduce are fundamental to these more advanced topics.
In the next section we give basic definitions of what we mean by a “conic,” and
introduce a moduli space of all conics. For each condition imposed on the conics we
are counting, there is a subset of the moduli space consisting of the conics that satisfy
this condition. To find the conics satisfying all of the given conditions, we intersect
the corresponding subsets. If this intersection consists of a finite number of points,
this number is our answer. In section 3 we will carry out this computation in several
examples, each of which leads to some key ideas.
Steiner’s original answer to his problem, 7776, was incorrect. He probably made the
mistake of assuming that the intersection of the five subsets corresponding to the five
given conics was finite. In fact, it is not finite, which we show in section 4. The infinite
component of the intersection consists of double lines, conics whose equation is the
square of a linear equation. If we can remove these, we will be left with a finite number
of points corresponding to ellipses, hyperbolas, and parabolas. The first to successfully
remove the double lines and count the remaining points was the French naval officer
de Jonquieres [18, p. 469], who in 1859 gave the correct answer to Steiner’s problem,
3264. Later, Michel Chasles developed a method for determining the answer 3264 and
solving many other similar problems [19]. In section 4 we introduce the duality of the
October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 701

plane and show how it can be used to remove the double lines in some problems. In
section 5 we use a tool called “blowing up” to remove the double lines in the remaining
problems.
We take a different point of view in section 6, using deformations to look at Steiner’s
problem and give an intuitive description of where the number 3264 comes from. In
section 7 we prove that by removing the double lines, we do indeed get precisely
3264 conics solving Steiner’s problem. In the last section we give some exercises and
suggestions for further reading.
2. A MODULI SPACE OF PLANE CONICS. A plane conic curve is the set of

points (x, y) ∈ R2 that satisfy a degree two polynomial relation,
ax 2 + bx y + cy 2 + d x + ey + f = 0, (1)
where not all of the coefficients are zero. Circles, ellipses, hyperbolas, and parabolas
are common examples of conics. In these examples the polynomial defining the conic
is irreducible and the conic is said to be nondegenerate. If the polynomial defining
the conic factors into a product of linear polynomials, then the conic is just the union
of two lines. Such a conic is said to be degenerate. When the two lines are the same,
or the polynomial defining the curve is a square of a linear polynomial, then the conic
should be thought of as a double line, a line with some additional algebraic structure.
These double lines play a key role in counting problems involving conics.
Figure 1. Parabolas, ellipses, and hyperbolas are conics. So are pairs of lines.
Any conic is completely determined by the coefficients a, b, c, d, e, and f of its

defining equation (1), but not uniquely so; for example, the equations x 2 − y = 0 and
3x 2 − 3y = 0 describe the same curve. If we consider the point (a, b, c, d, e, f ) ∈
R6 as representing the conic ax 2 + bx y + cy 2 + d x + ey + f = 0, we see that for
any nonzero scalar λ, the point (λa, λb, λc, λd, λe, λ f ) represents the same curve.
Therefore any point on the line spanned by the vector a, b, c, d, e, f gives rise to the
same conic. So we have a one-to-one correspondence between lines through the origin
in R6 and the equations defining plane conics (up to scalar multiple).
We’ll use homogeneous coordinates [a : b : c : d : e : f ] to describe the line
spanned by a, b, c, d, e, f . This notation reminds us that the values of the coeffi-
cients a, b, c, d, e, and f are less important than their ratios to one another. What
happens if a, b, c, d, e, and f are all zero? The zero vector does not span a line so
[0 : 0 : 0 : 0 : 0 : 0] is not a valid set of homogeneous coordinates. As well, this set of
parameters does not correspond to a curve since the equation of the associated conic
(1) reduces to 0 = 0 and places no constraints on our points. Therefore this is not a
meaningful set of coefficients to consider.
702
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
The set of lines through the origin in R6 is called the five-dimensional real pro-
jective space and is denoted RP5 . It serves as our moduli space for conics, a space
whose points are in one-to-one correspondence with the set of conics.
Why is RP5 five-dimensional? Well, each point of RP5 is part of an open set which
can be identified with R5 . Given a point in RP5 , one of its homogeneous coordinates
a, . . . , f is not zero. Let us suppose that f = 0. Then the set U f = {[a : b : c : d : e :
f ] : f = 0} can be identified with R5 via

a b c d e a b c d e
[a : b : c : d : e : f ] = : : : : :1 ∼ , , , , ∈ R5 .
f f f f f f f f f f
In this sense, RP5 is a five-dimensional space. More generally, the set of lines through
the origin in Rn+1 forms n-dimensional real projective space, RPn .
2.1. The basic counting strategy. We’ve described the moduli space RP5 for plane
conics. Using certain subsets of this moduli space, we can introduce the basic strategy
to count the conics passing through some fixed points and tangent to some fixed lines
or conics. For each point p we form the subset H p ⊂ RP5 of conics passing through
the point, for each line we form the subset H ⊂ RP5 of conics tangent to , and for
each given nondegenerate conic Q we form the subset HQ of conics tangent to Q. The
points in the intersection of all of these subsets correspond to conics that pass through
all of the points and are tangent to all of the lines and conics. This shift, from counting
conics to counting the number of points in an intersection of certain subsets of RP5 ,
may seem like a sleight of hand but it allows us to use the geometry of RP5 as well as
the geometry of the plane to solve our counting problems.
Let’s examine these subsets H p , H , and HQ in more detail. To be concrete, let’s fix
a point, say p (2, 3). If a conic defined by equation (1) passes through p, then it must
be true that 4a + 6b + 9c + 2d + 3e + f = 0. We see that this is a linear equation
in a, b, c, d, e, and f . The set of points in RP5 satisfying this condition forms a
four-dimensional plane, or a hyperplane in RP5 . Each point on this hyperplane H p
corresponds to a conic passing through p (2, 3). If we chose a different point q ∈ R2
we would get a different hyperplane Hq . Points on the intersection of H p and Hq will
correspond to conics passing through both p and q.
Similarly, if we look at all of the conics tangent to a particular line, for example the
line y = 0, we get a four-dimensional hypersurface H in RP5 . This can be seen by first
finding the intersection of a general conic, ax 2 + bx y + cy 2 + d x + ey + f = 0, and
the line y = 0. The points of intersection have the form (x, 0), where ax 2 + d x + f =
0. Usually we have two different points of intersection and the line y = 0 is a secant
line to the conic. But when the discriminant d 2 − 4a f is zero the two points coincide
and the line y = 0 is tangent to the conic. So the points in RP5 that satisfy the equation
d 2 − 4a f = 0 correspond to the conics that are tangent to the line y = 0. If we started
with a different line in R2 we would get a hypersurface H defined by a different
degree 2 equation.
Lastly, if we look at all of the conics tangent to a particular conic Q, for example
the parabola y = x 2 , we also get a four-dimensional hypersurface in RP5 . To find its
equation, we substitute y = x 2 into the general conic equation to get cx 4 + bx 3 +
(a + e)x 2 + d x + f = 0. The two conics will be tangent when this polynomial has
a multiple root, which again is when the discriminant is zero. The discriminant of a
degree four polynomial has degree six in the coefficients ([6, p. 42]), so HQ is a degree
six hypersurface in RP5 . If we started with a different nondegenerate conic Q, we
would get a hypersurface defined by a different degree 6 equation.

These three hypersurfaces H p , H , and HQ are examples of projective algebraic va-
rieties. A projective algebraic variety is a subset of a projective space RPn consisting
of the common zeros of a collection of homogeneous polynomials—polynomials in
n + 1 variables so that for each polynomial all its terms have the same degree.
If we require a conic to pass through several points and be tangent to several
lines and conics, we will look at the intersection of the corresponding H p ’s, H ’s,
and HQ ’s. When studying intersections of varieties, it is useful to consider each va-
riety’s codimension rather than its dimension. In this case, since H p , H , and HQ
are four-dimensional hypersurfaces in a five-dimensional space, their codimension is
5 − 4 = 1. For most pairs of varieties, the codimension of their intersection is the
sum of their codimensions (of course, this will not be true if the varieties overlap too
much).
Since the parameter space RP5 is five-dimensional, we expect that if we have five
conditions then the intersection of the corresponding five hypersurfaces will have codi-
mension 5 and be a finite collection of points. So it makes sense to ask: How many
conics pass through p points and are tangent to lines and c conics, if p + + c = 5?
In the next section we’ll lay the foundation for answering this question by first consid-
ering the case where no conics are present.
3. SOME BASIC ENUMERATIVE QUESTIONS. In this section we are going

to concentrate on the point-line enumerative problems: Find the number of conics
through p points and tangent to lines if p + = 5.
3.1. Five points. Let’s count the number of conics passing through five points in the
plane. First, each point p imposes a hyperplane condition H p on our conic. Thinking
of RP5 as the set of lines through the origin in R6 , each hyperplane H p corresponds
to a hyperplane (a five-dimensional subspace) of R6 . If these hyperplanes are linearly
independent then their intersection is a one-dimensional subspace of R6 . This line
through the origin in R6 corresponds to a single point in RP5 . In turn, this single point
represents a unique conic passing through all five points. We’ve shown that when the
five points impose independent hyperplane conditions there is a unique conic passing
through all five points.
It turns out that the points impose independent hyperplane conditions precisely
when no four of the points are collinear. This requires a little argument, as follows.
When four or five of the points are collinear, then there are lots of conics that pass
through all of the points (and hence the hyperplanes couldn’t impose independent con-
ditions): just consider conics consisting of pairs of lines, where one line passes through
the four collinear points and the other is any line that passes through the fifth point, as
in Figure 2. So we may restrict our attention to the case where no four of the points lie
on a line.
Figure 2. The case of 4 collinear points.
704
The question is not dependent on our choice of coordinates on the plane, so we
choose coordinates such that three of the points are p1 (0, 0), p2 (1, 0), and p3 (0, 1).
We’ll label the other two points p4 (s, t) and p5 (u, v), as in Figure 3.
p4 (s,t)
p3 (0,1) p5 (u,v)
p1 (0,0)
p2 (1,0)
Figure 3. Choosing coordinates.
The system of equations imposed by these five points has coefficient matrix
⎡ ⎤
0 0 0 0 0 1
⎢1 0 0 1 0 1⎥
⎢ ⎥
M =⎢0 0 1 0 1 1⎥ .
⎣ s 2 st t 2 s t 1⎦
u 2 uv v 2 u v 1
The five constraints fail to impose linearly independent conditions precisely when
this matrix fails to have full rank. The maximal rank of M is 5, so we can detect when
the matrix does not have full rank by checking that all of the 5 × 5-submatrices of M
have determinant zero.1 By deleting one column at a time, we get six polynomials that
are simultaneously zero precisely when M fails to have maximal rank. One of these
(when the last column is deleted) is always zero; the first and fourth are the same, and
the third and fifth are the same, so we end up with three conditions:
tv(s(v − 1) − u(t − 1)) = 0

su(v(s − 1) − t (u − 1)) = 0
(s 2 − s)(v 2 − v) − (u 2 − u)(t 2 − t) = 0.
Some careful case-by-case analysis will show that if all three of these equations hold,
then either four of the given points are collinear or two of the points are coincident. For
example, if the first equation holds, either t = 0, v = 0, or s(v − 1) = u(t − 1). If t =
0, then the second equation says that either s = 0, in which case p4 (s, t) = p1 (0, 0);
s = 1, in which case p4 (s, t) = p2 (1, 0); v = 0, in which case p1 , p2 , p4 , and p5 are
collinear; or u = 0. If t = u = 0, then the third equation requires two points to be
coincident. Going back to the first equation, the case of v = 0 is essentially identical
to the case of t = 0. The last case is where s(v − 1) = u(t − 1), or t−1 s
= v−1
u
. This
says that p3 , p4 , and p5 are collinear. Looking again at equation two in this case, either
s = 0 or u = 0 will cause p1 to be on the same line with p3 , p4 , and p5 ; the last
possibility, that v(s − 1) = t (u − 1), causes p2 to be on this line instead; either way
we have four collinear points.
1 This is a useful characterization of rank: A matrix has rank < d if and only if the determinants of all its
d × d submatrices vanish [1, p. 153, ex. 10].

A more high-tech way (both using a computer and some commutative algebra) to
see this is the following. Using a handy computer algebra program, like Macaulay2,
Singular, or CoCoA, we can check that the ideal generated by the six submatrix deter-
minants in the polynomial ring R[s, t, u, v] has the following primary decomposition:
(t − v, s − u) ∩ (t, s − 1) ∩ (t − 1, s) ∩ (t, s) ∩ (v, u) ∩ (v, u − 1) ∩ (v − 1, u)

∩ (u + v − 1, s + t − 1) ∩ (u, s) ∩ (v, t).
This means that the matrix M fails to have maximal rank precisely when the poly-
nomials generating one of the primary ideals vanish. The first seven of these pairs of
equations just indicate that two of our five points are equal, while the last three pairs
indicate that four of our points are collinear.
The upshot of all this is that the five points impose linearly independent conditions
if and only if no four of the points are collinear.
Theorem 1. Given five points in the plane, no four of which lie on a line, there is
a unique conic passing through the five points. The conic is nondegenerate precisely
when no three of the points are collinear.
Proof. It just remains to prove the last statement. If three of the points are collinear but
no four are collinear, the unique conic passing through all of the points is a degenerate
conic, a pair of lines. This is easy to see: three of the points lie on a line L : G(x, y) =
0, and if : H (x, y) = 0 is the line through the other two points, then G H = 0 is the
equation of the unique conic L ∪ passing through the five points. If no three of the
points are collinear, then the pigeonhole principle tells us that there is no pair of lines
containing all five points. Therefore the unique conic passing through all five points
cannot be degenerate.
3.2. Four points and one line. Now we solve the next problem: How many conics
pass through four given points and are tangent to a given line? In answering this ques-
tion, we’ll see that it is necessary to develop a more expansive view of plane conics
and tangency.
Figure 4. Two conics through four points and tangent to a line.
Each of the four points gives a linear condition, and the reader can check that the
four conditions are independent if and only if the four points are not collinear. The
intersection of the four hyperplanes is a line in RP5 . Recall that the set of conics
tangent to a line in the plane formed a hypersurface H , whose defining equation had
706
degree 2. The intersection of the line with H can be found by plugging the parametric
equation for the line into the equation for H and solving. The resulting equation is
quadratic and so in general we get two solutions. Each solution gives a point in RP5
corresponding to a conic passing through the given four points and tangent to the line
. Therefore in general we expect there to be two such conics as in Figure 4.
Sometimes we need to be a little open-minded to recognize the resulting curves as
conics that satisfy our constraints. For example, consider the conics passing through
the points ( 12 , 2), (2, 12 ), (− 12 , −2), (−2, − 12 ) and tangent to the line y = 0 as in Fig-
ure 5.
Figure 5. A pathological example.
Following the algebraic procedure described above, we get two solutions. One solu-
tion corresponds to the pair of crossed lines (y − 4x)(x − 4y) = 0. This pair of lines
certainly passes through the four given points, but in what sense is it tangent to the
given line? Algebraically, plugging the equation of the line into the equation of the
conic gives a quadratic equation with a double root, which is what we associate with
tangency. Geometrically, the given line intersects the crossed lines at their crossing
point. If the given line were moved slightly, it would intersect each line once in two
different points. This also reminds us of tangency. So we will consider this line to be
tangent to this conic, even though the given line and the crossed lines are not tangent
in the sense of derivatives and do not have the same direction at that point.
The other solution we find is the hyperbola x y − 1 = 0. Again, this hyperbola
passes through the four given points, but is it tangent to the given line? Algebraically,
if we plug y = 0 into x y = 1, we get no solutions at all! However, the line and hyper-
bola approach each other as x → ∞. We would like to consider these to be tangent
as well, and we can do this by adding some points to the plane “at infinity.” One way
to do this is to assume that the plane that we have been working in is part of a two-
dimensional projective space, similar to the five-dimensional projective space RP5 that
parameterizes all conics. The projective plane RP2 is the set of all lines through the
origin in R3 , and has coordinates [X : Y : Z], where [X : Y : Z] = [λX : λY : λZ] for
any nonzero λ. We have been looking at the subset where Z is not zero, so the point
[X : Y : Z] is the same as [ XZ : YZ : 1], and the variables we have been calling x and
y are secretly XZ and YZ . The new points we are adding at infinity are the points with
Z = 0.
How does this help us? If we translate our hyperbola x y − 1 = 0 into these new
coordinates, we get XZ YZ − 1 = 0, or X Y − Z 2 = 0. Now plugging in the (also trans-

lated) equation of our given line, Y = 0, we get Z 2 = 0, which has a double root at
Z = 0. So in a very concrete way the line and the hyperbola are tangent at infinity, or
more specifically at the point [1 : 0 : 0] in the projective plane RP2 .
We obtained our two solutions by solving a quadratic equation with real coefficients,
but such equations often have complex solutions. Indeed, if we move the line so that
it separates one of the four points from the other three then the two conics that solve
our problem both have complex-valued coefficients. We allow our homogeneous coor-
dinates to be complex numbers in order to accommodate such solutions. Our moduli
space for conics becomes the complex projective five space CP5 , the space of one-
dimensional subspaces in C6 . Since these conics are tangent to the line at a point with
complex coordinates, we are also forced to allow the X , Y , and Z coordinates to take
complex values. Thus our solutions are conic curves that live naturally in the complex
projective plane CP2 .
When we first introduced the parameter space RP5 we noted that its points are
in one-to-one correspondence with the equations of the plane conics (up to scalar
multiple). At that time we hid one complication: there can be several equations that
define the same set of points; for example x 2 + y 2 + 1 = 0 and x 2 + y 2 + 3 = 0
both define the empty set. But in CP2 these equations become X 2 + Y 2 + Z 2 = 0
and X 2 + Y 2 + 3Z 2 = 0 and they define different complex curves. Points in CP5 are
in one-to-one correspondence with conic curves in CP2 . This fact follows from the
observation that if we have a plane conic in CP2 defined by a degree two equation,
then any other degree two equation for the conic must be a scalar multiple of the first.
This can be proved using Hilbert’s Nullstellensatz, one of the foundational theorems
in algebraic geometry [5, Sec. 4.1].
From here on, we’ll restrict our attention to points, lines, and conics in the
complex projective plane. We’ll drop the C from our notation and just use Pn to
denote complex n-dimensional projective space. A general conic in P2 will have the
form
a X 2 + bX Y + cY 2 + d X Z + eY Z + f Z 2 = 0,
where the coefficients are allowed to be complex numbers.

To summarize, the solutions to our enumerative problems may include degenerate
conics and points of tangency may occur at complex points or at infinity. In order to
accommodate these issues, we work with the moduli space CP5 of complex conics in
the complex projective plane CP2 .
3.3. Bézout’s Theorem. Using the complex projective plane allows us to discuss in-
tersections of curves consistently. For example, in the projective plane, two lines al-
ways intersect in one point—parallel lines meet at infinity, just like in a perspective
drawing. As another example, consider a circle and a line in the plane. Algebraically,
plugging the line equation into the circle equation gives a degree two polynomial. If
this polynomial has distinct real roots, then the circle and line will intersect in two
points. If the polynomial has a pair of complex conjugate roots, the circle and line
will look as though they miss each other as in Figure 6. But if we allow points in
the plane to have complex coefficients, then they will intersect at these two complex
points. So in general, the number of intersection points between the circle and the line
is two.
This consistent counting in complex projective space is described by Bézout’s the-
orem:
708
Figure 6. Two pictures of a line meeting a circle in two complex points.
Theorem 2 ([24]). If n hypersurfaces of degrees d1 , d2 , . . . , dn in Pn intersect trans-

versally, then the intersection consists of d1 · d2 · · · dn points.
The hypersurfaces X 1 , . . . , X n intersect transversally at a point P ∈ X 1 ∩ · · · ∩

X n when their tangent spaces at P just intersect in the point P alone. The intersection
X 1 ∩ · · · ∩ X n is transverse if it is transverse at each of its points. As an example,
consider again the line and circle. When the line meets the circle in two real points,
then at each point the tangent line to the circle just meets the given line in one point;
so the line and the circle meet transversally. This is also the case when they meet at
two complex points, although it is harder to draw the picture! However, when the line
is tangent to the circle, the two tangent spaces coincide and the intersection is not
transverse. In this case, instead of two intersection points, we only get one. See Figure
7 for more examples.
Figure 7. Nontransverse intersections—left and center—and a transverse intersection, right.
We’ve developed our intuition about Bézout’s theorem in the projective plane P2 ,
but now we want to use it in P5 to answer our enumerative questions. First let’s revisit
the problem of counting the conics that pass through 5 points. We saw that these conics
are in bijection with the points in the intersection of five hypersurfaces H p of degree
1. Since these H p are hyperplanes, they intersect transversally when the hyperplanes
impose linearly independent conditions. We saw that this was true if no four of the
points are collinear. In this case, the five hyperplanes intersect in 15 = 1 point by
Bézout’s theorem.
Let us return to the question, “How many conics pass through four given points and
are tangent to a given line?” Now we are intersecting four degree one hyperplanes H p
and one degree two hypersurface H . If these intersect transversally, Bézout’s theorem
says that they will intersect in 14 · 2 = 2 points, so there will be two such conics. In
this case, it turns out that the intersection is transverse unless three of the points are
collinear or one of the points lies on the given line.
It is possible to generalize Bézout’s theorem to cases where the hypersurfaces do
not meet transversally. If the hypersurfaces X 1 , . . . , X n have degrees d1 , . . . , dn and

the intersection X 1 ∩ · · · ∩ X n consists of finitely many points, then we always get
d1 d2 · · · dn points in the intersection, provided we count the points with their proper
multiplicities. For instance, if a line is tangent to a circle, then the only point of in-
tersection must count with multiplicity 1 · 2 = 2. There are two ways to explain this
result. If we plug the equation for the line into the equation for the circle, we get a
quadratic equation Q = 0, where Q factors as a perfect square. The point of intersec-
tion corresponds to the double root of this quadratic.2 There is also a dynamic way to
compute the multiplicity of the tangent point. As we slide the line across the circle, we
can keep track of the points of intersection. Since two points of intersection collapse
to one point when the line is tangent to the circle, the point of tangency counts for two
points. If we count with multiplicity, any circle meets any line in two (not necessarily
distinct!) points.
In what follows, we will only need two facts about multiplicity: the multiplicity of
the intersection X 1 ∩ · · · ∩ X n at one of its points P is always a positive integer, and
the number is equal to 1 precisely when the X i meet transversally at P. However, if
you know a little commutative algebra, then there is a nice way to assign multiplicity
at a point of intersection of hypersurfaces X i : the multiplicity is just the vector-space
dimension of the quotient of the polynomial ring, localized at the maximal ideal cor-
responding to the point, modulo the polynomials defining the hypersurfaces.
3.4. Three points and two lines. We can use Bézout’s theorem to count the number
of conics passing through three points and tangent to two lines. The answer we expect
is 13 · 22 = 4. In this case, the intersection will again be transverse unless the three
points are collinear or one of the points lies on one of the lines. So far we’ve managed
to fill in a few columns in Table 1.
Table 1. Number of conics through p points and tangent to lines.
Lines 0 1 2 3 4 5
Points p 5 4 3 2 1 0
Conic solutions 1 2 4 ? ? ?
4. EXCESS INTERSECTION AND THE DUALITY OF P2
4.1. Excess intersection and general position. It is tempting to guess that the un-
known entries in Table 1 continue as powers of 2. After all, if the five hypersurfaces
involved in each problem intersect transversally, then this result would follow from
Bézout’s theorem. However, it turns out that for the last three point and line prob-
lems the corresponding hypersurfaces cannot intersect transversally, no matter how
we choose the points and lines!
The reason that these hypersurfaces do not intersect transversally involves the
double line conics. The defining polynomial of a double line factors as a perfect
square. We refer to a conic whose defining polynomial is not a perfect square as a
reduced conic. The reduced conics include nondegenerate conics, like circles and
2 This double point is an example of what algebraic geometers call a scheme. Schemes occur naturally as
limiting objects whenever geometric objects change under deformation; for example, if a pair of crossed lines
pivot about their intersection point the limiting object is a double line.
710
parabolas, as well as degenerate conics consisting of a pair of distinct lines. The only
nonreduced conics are the double line conics.
To see how double lines are connected to the transverse intersection property, let’s
consider conics that pass through one point and are tangent to four lines. Recall that in
the projective plane, any two lines will intersect; in particular, any double line through
the given point will intersect each of the four given lines. These double lines will be
tangent to each of the given lines in the algebraic sense that the equation for the inter-
section has a double root. Thus we get infinitely many double line conics as solutions!
Because we don’t have a finite number of solutions, the five hypersurfaces in P5 must
not intersect transversally. Here we see that the geometry of conics in P2 sheds light
on the geometry of P5 . This phenomenon, in which we expect our intersection to con-
sist of finitely many points but in fact the intersection has higher dimension, is called
excess intersection.
However, there are also a finite number of reduced conics that pass through the
point and are tangent to the four given lines. We would like to ignore the double line
solutions and just count the number of reduced conics that pass through p given points
and are tangent to 5 − p given lines. In general there is a finite solution to this problem.
Moreover, in most cases the reduced conics solving the problem are all nondegenerate.
In each of the remaining enumerative questions we will ask for the number of
reduced conics satisfying the geometric constraints.
In the problems we’ve already solved, none of the solutions can be double lines
because of the constraints we put on the given points and lines—in each case, we did
not allow all of the points in the problem to be collinear. We placed these constraints
to ensure transverse intersection of the corresponding hypersurfaces, which then guar-
anteed that the number of conics was correctly calculated by the formula in Bézout’s
theorem. The constraints ensured that we do not get infinitely many solutions to our
problem, nor do we get solutions with multiplicity. When we have a finite number of
reduced solutions, each appearing with multiplicity one, then we say that the given
points and lines are in general position. In each enumerative problem, the conditions
that constitute general position (for example, that no three points lie on a line) will
be slightly different, but almost all configurations of points and lines will satisfy the
conditions.
4.2. Duality in P2 . The tool that we will use to remove the double lines from our
count is the duality of P2 . Duality allows us to exchange points and lines, and at the
same time it transforms conics into conics. The operation of duality respects inclusion
and tangency. This will allow us to replace the remaining three point-line problems
with the three problems we have already solved.
Consider the set of all lines in P2 . We use a linear equation L : AX + BY + C Z =
0 to describe such a line. Of course, the line doesn’t change if we multiply each of the
coefficients by a nonzero constant, so the line can be represented by a point [A : B : C]
in a projective plane. Thus the set of lines in P2 is called the dual projective plane.
The dual projective plane is just another copy of P2 , but to distinguish the two spaces
we will denote the dual projective plane by P̌2 and use the symbols A, B, and C for its
coordinates.
By definition, a line L in P2 corresponds to a point in P̌2 , which we call Ľ
(“L dual”). We can also define the dual of a point p in P2 as the collection of
lines that pass through p. If p = [X 0 : Y0 : Z 0 ] then this collection is {[A : B : C] :
X 0 A + Y0 B + Z 0 C = 0}. This is a linear equation in the variables A, B, and C, so we
see that the point p naturally corresponds to a line in P̌2 , which we call p̌. Geometri-

cally duality associates lines in P̌2 to points in P2 and vice versa. It is a good exercise
to check algebraically that duality respects inclusion: if p is a point of P2 lying on
the line L in P2 , then p̌ is a line in P̌2 containing the point Ľ. This is illustrated in
Figure 8.
Figure 8. Four lines through one point (left) and their duals (right).
Let’s see what duality does to conics. If Q is a conic curve in P2 , then we define the
dual curve Q̌ in P̌2 to be the collection of all lines tangent to Q. Note that this means
that Q̌ will contain a point Ľ if and only if the corresponding line L in P2 was tangent
to Q.
As an example, consider the conic Q given by the equation X 2 − Y Z = 0. A general
line AX + BY + C Z = 0 meets the curve in two points. If A = 0, we can find these
two points by noting that (AX )2 − A2 Y Z = 0 on the curve and AX = −(BY + C Z)
on the line, so the points of intersection satisfy (BY + C Z)2 − A2 Y Z = 0. Rearrang-
ing gives a homogeneous quadratic equation in the variables Y and Z,
B 2 Y 2 + (2BC − A2 )Y Z + C 2 Z 2 = 0. (2)
As long as the discriminant of this equation is nonzero, its solution consists of two
distinct points [X : Y : Z] ∈ P2 and the line AX + BY + C Z = 0 is the secant line to
the curve Q joining the two points. When the discriminant A2 (A2 − 4BC) of equation
(2) is zero, the line is tangent to the curve. Since we assumed that A = 0, we see that
the line is tangent to the curve when A2 − 4BC = 0. A similar analysis gives the same
equation when B or C is nonzero. So in this example, the equation for the dual curve
Q̌ is A2 − 4BC = 0. Note that Q̌ is a conic in P̌2 .
What is the equation of the dual for a more general conic? If the conic has equa-
tion a X 2 + bX Y + cY 2 + d X Z + eY Z + f Z 2 = 0, then a general line AX + BY +
C Z = 0 with A = 0 meets the conic in two points where
a(BY + C Z)2 − b AY (BY + C Z) + c A2 Y 2

− d(BY + C Z)AZ + e A2 Y Z + f A2 Z 2 = 0.
This is a homogeneous quadratic equation in Y and Z. Just as above, the line is tangent
to the conic when the discriminant of this quadratic equation vanishes. Writing out the
discriminant and using that A = 0, we see that
(e2 − 4c f )A2 + (4b f − 2de)AB + (d 2 − 4a f )B 2 + (4cd − 2be)AC

+ (4ae − 2bd)BC + (b2 − 4ac)C 2 = 0. (3)
712
The same equation would result if we assumed that B or C is nonzero. So this equation
characterizes the lines that are tangent to the conic Q; it is the equation for the dual Q̌.
As in the example, the equation for Q̌ is degree two, so it is again a conic in P̌2 . It can
be checked that when Q is nondegenerate, then Q̌ is also nondegenerate.
We leave it to the reader to check that: (1) the dual to a degenerate conic consisting
of a pair of crossed lines L 1 ∪ L 2 is the double line conic through the two points Ľ 1
and Ľ 2 ; (2) the dual to a double line conic is just P̌2 itself.
Viewing equation (3) in a slightly different way gives the equation for the hyper-
surface H of conics tangent to a line . Let AX + BY + C Z = 0 represent the line
. Fixing A, B, and C, equation (3) gives constraints on the coefficients of the conics
tangent to . This gives an explicit equation for H as a degree 2 hypersurface.
In order to use duality to help us count conics tangent to lines, we need to understand
how tangency transforms under duality. If a line L 1 is tangent to a nondegenerate conic
Q then by definition Ľ 1 is a point on Q̌. If p is a point on Q, is the line p̌ tangent to
the dual conic Q̌? To start, let L 1 be the tangent line to Q at p. Then Ľ 1 is a point lying
on the line p̌. The line p̌ meets the dual conic in two points, Ľ 1 and Ľ 2 , as depicted in
Figure 9. We aim to show that these two points coincide, so that p̌ is a tangent line to
Q̌. Since Ľ 2 ∈ Q̌, L 2 must be a line that is tangent to Q. But Ľ 2 ∈ p̌ so L 2 must also
pass through p. So L 2 is a line through p that is tangent to the conic Q. Since each
tangent line can meet the conic in at most one place, L 2 must be tangent to Q at p and
so L 2 = L 1 . It follows that Ľ 2 = Ľ 1 , so the line p̌ meets the dual conic in a unique
point; the line p̌ must be tangent to Q̌ at this point.
p
L1 L1 Q
L2
Q
p
Figure 9. Tangency under duality (we show that Lˇ2 = Lˇ1 ).
4.3. The rest of the point-line problems. If we have five lines in the plane, then any
nondegenerate conic tangent to them will have a dual conic that passes through the five
dual points. There is only one such dual conic, so there can only be one conic tangent to
all five lines. Any nondegenerate conic tangent to four given lines and passing trough
a given point must have a dual conic that passes through the four dual points and
is tangent to the dual line. If the dual points and dual lines are in general position,
there are 2 such dual conics, so there are 2 conics tangent to four lines and passing
through a point in general position. Finally, the reader should check that there are 4
conics tangent to three lines and passing though a pair of points in general position.
We summarize the solutions to the point-line problems in Table 2.
5. STEINER’S PROBLEM
5.1. Blowing up the Veronese. Now that we have solved all of our point-line prob-
lems, we turn to Steiner’s problem, “How many conics are tangent to five given con-
ics?” We first try to intersect the five hypersurfaces HQ corresponding to the five given

Table 2. Number of conics through p points and tangent to lines.
Lines 0 1 2 3 4 5
Points p 5 4 3 2 1 0
Conic solutions 1 2 4 4 2 1
conics. Bézout’s formula would suggest that the intersection consists of 65 = 7776
points, which was the answer that Steiner gave. However, Bézout’s theorem does not
apply because the five hypersurfaces HQ cannot intersect transversally. Indeed, every
double line conic is tangent to each of the five given conics, so the intersection of the
five hypersurfaces HQ in P5 contains the set of double lines. Unfortunately, we cannot
use duality directly to filter out the double line conics from our answer. For this we
introduce another tool, the blowup. To start, let’s look more closely at the set of double
lines that is causing us so much difficulty.
The set of all points in P5 corresponding to double lines is called the Veronese
surface, V . Recall that a double line is a conic whose defining polynomial is the square
of the defining polynomial of a line:
(AX + BY + C Z)2 = A2 X 2 + 2AB X Y + B 2 Y 2

+ 2AC X Z + 2BCY Z + C 2 Z 2 = 0.
Lines in P2 are parameterized by P̌2 , so the Veronese is indeed a surface (two-

dimensional). It is the image of the injective map from P̌2 into P5 ,
[A : B : C] → [A2 : 2AB : B 2 : 2AC : 2BC : C 2 ].
This map is defined by polynomials, and its domain is all of P̌2 , so it is an example
of what algebraic geometers call a morphism. The image of this morphism, which is
the Veronese surface, is defined by the zeros of some homogeneous polynomials in
the six coordinates on P5 . Because the Veronese is two-dimensional, and it is in the
five-dimensional P5 , we expect that it can be described by three equations. This is true
locally, but if we want to describe the full Veronese we in fact need six equations. The
Veronese surface is an example of a variety which is not a complete intersection. A
little algebra finds us the six equations:
b2 − 4ac = 0 4b f − 2de = 0
d 2 − 4a f = 0 4cd − 2be = 0 (4)
e − 4c f = 0
2
4ae − 2bd = 0.
If we consider an open set where one of these six variables is nonzero, we can use
three of the above equations to solve for three variables and then derive the other
three equations. But this only works on this open set; if we were to choose a different
variable to be nonzero we would need three different equations to derive the others. So
to describe the whole Veronese we need all six equations.
Next we will look at a map which is not a morphism, and try to turn it into one.
Consider the map from P5 to P5 which sends a conic to its dual. Using equation (3) for
the dual conic we found in section 4.2, this map is defined as
714
δ : P5 → P5
[a : b : c : d : e : f ] → [e2 − 4c f : 4b f − 2de : d 2 − 4a f : (5)
4cd − 2be : 4ae − 2bd : b2 − 4ac].
The six polynomials that define δ in (5) are precisely the left sides of the six equa-
tions in (4), and therefore δ is undefined at a point if and only if that point is on the
Veronese—that is, if and only if the point corresponds to a double line. The map is not
a morphism on all of P5 ; however, it is defined by polynomials, so it is what algebraic
geometers call a rational map. A rational map can always be extended to a morphism
by expanding the domain space. In our case, we will take the Veronese surface out of
P5 and replace it with a four-dimensional variety. This is done by looking at the graph
of δ in P5 × P5 and closing it up.
Definition. The blowup of P5 along the Veronese surface, Bl V P5 , is the closure of

the graph of δ in P5 × P5 . The blowing down morphism π : Bl V P5 → P5 is given by
the projection onto the first factor.
Let’s try to write some equations for Bl V P5 . First of all, the graph is a set of points
([a : b : c : d : e : f ], [r : s : t : u : v : w]) in P5 × P5 which must satisfy
λr = e2 − 4c f λu = 4cd − 2be
λs = 4b f − 2de λv = 4ae − 2bd
λt = d 2 − 4a f λw = b2 − 4ac.
Eliminating λ and cross multiplying, we get fifteen equations:
r (4b f − 2de) = s(e2 − 4c f ) s(d 2 − 4a f ) = t (4b f − 2de) t (4ae − 2bd) = v(d 2 − 4a f )

r (d 2 − 4a f ) = t (e2 − 4c f ) s(4cd − 2be) = u(4b f − 2de) t (b2 − 4ac) = w(d 2 − 4a f )
r (4cd − 2be) = u(e2 − 4c f ) s(4ae − 2bd) = v(4b f − 2de) u(4ae − 2bd) = v(4cd − 2be)
r (4ae − 2bd) = v(e − 4c f )
2
s(b − 4ac) = w(4b f − 2de)
2
u(b2 − 4ac) = w(4cd − 2be)
r (b2 − 4ac) = w(e2 − 4c f ) t (4cd − 2be) = u(d 2 − 4a f ) v(b2 − 4ac) = w(4ae − 2bd).
In addition, because the original six equations for the Veronese were not algebraically
independent, the blowup must satisfy eight more equations:
bu + 2ew + 2cv = 0 bv + 2dw + 2au = 0

eu + 2br + 2cs = 0 dv + 2bt + 2as = 0
ds + 2et + 2 f v = 0 4ar − 4ct + du − ev = 0
es + 2dr + 2 f u = 0 4ct − 4 f w + bs − du = 0.
These eight equations are syzygies—linear relations (with polynomial coefficients)

among the six equations (4) defining the Veronese surface. We obtained them using
the syz command in Macaulay2. Syzygies play an important role in understanding
the behavior of systems of equations and their solution sets (see [7] for more details).
Now, let us compare the blowup to P5 . Suppose that a point [a : b : c : d : e : f ] on
P is not on the Veronese surface, so it does not represent a double line conic. Then the
5

duality map is well-defined at this point. In fact we see that the first fifteen equations
completely determine [r : s : t : u : v : w], so π −1 ([a : b : c : d : e : f ]) is a point. In
fact, away from the Veronese surface, P5 and Bl V P5 are isomorphic.
Now suppose that the point [a : b : c : d : e : f ] is on the Veronese surface. In this
case, the first fifteen equations all reduce to 0 = 0. Instead, the last eight equations tell
us what the corresponding points are in the blowup. For example, let us consider the
double line conic X 2 = 0. The corresponding point on the Veronese surface in P5 is
[1 : 0 : 0 : 0 : 0 : 0]. If we let b = c = d = e = f = 0 in the equations on the blowup,
we see that the last eight equations reduce to
2au = 0 2as = 0 4ar = 0.
Since a = 0, this tells us that r , s, and u are forced to be zero, but that t, v, and w
are free. Thus the points in the blowup Bl V P5 that are mapped by the blowing down
morphism to this point are of the form
([1 : 0 : 0 : 0 : 0 : 0], [0 : 0 : t : 0 : v : w]) .
These points define a P2 within the blowup, so π −1 ([a : b : c : d : e : f ]) ∼

= P2 . The
same will be true for any double line that we pick: the corresponding point on the
Veronese in P5 has been replaced by an entire P2 in Bl V P5 .
In essence, in constructing the blowup we have ripped the Veronese out of P5 and
replaced it with something two dimensions larger, a four-dimensional hypersurface.
This hypersurface is called the exceptional divisor of the blowup, and we will call
it E for short. The name “blowing up the Veronese” should make us think not of
explosives, but of inserting a soda straw into P5 right at the Veronese, and blowing in
a bubble of air to stretch it out into a four-dimensional object.
This act of stretching out the Veronese is exactly what we need to pull apart the
excess intersection in Steiner’s problem. Consider a hypersurface Y in P5 that contains
the Veronese. Its preimage π −1 (Y ) in the blowup will then contain the exceptional
divisor. On the other hand, away from V and E, P5 and the blowup are identical. If we
remove V from Y and consider the inverse image π −1 (Y \ V ), this will be isomorphic
to Y \ V . If we take the closure π −1 (Y \ V ), we get a hypersurface in the blowup that
intersects E, but does not contain it (see Figure 10). We call this new hypersurface the
strict transform of Y and denote it by Ỹ .
To solve Steiner’s problem, we will intersect the strict transforms of the hypersur-
faces in P5 . Because the process of constructing the strict transform eliminates the E
components, this will eliminate the excess intersection along the Veronese. We will
show in section 7 that the proper transforms actually do intersect transversally, which
will verify that blowing up eliminates the excess intersection.
5.2. The Chow ring. In P5 we were able to count the number of points in the inter-
section of five hypersurfaces using Bézout’s theorem, but Bézout’s theorem doesn’t
hold on the blowup. This is because on the blowup, the “degree” of a hypersurface is
more complicated—it is not just one number! The extra information is encoded in the
Chow ring of the blowup. In general, for an algebraic variety, the Chow ring is a ring
that describes how its subvarieties intersect. Elements of the Chow ring are classes of
subvarieties that have the same intersection properties. Bézout’s theorem describes the
Chow ring of projective space.
In the case of P5 , Bézout’s theorem says that the degree of a hypersurface is enough
to determine its intersection properties. In particular, all hyperplanes will be in the
716
E
Bl v P 5
˜
Y
v
Y P5
Figure 10. A picture of the blowup.
same class. Let us call that class [H ]; it will be one element of the Chow ring of
P5 . The addition operation in the Chow ring roughly corresponds to the union of two
varieties, so [H ] + [H ] will be the class representing the union of two hyperplanes.
But the union of two hyperplanes is a special case of a degree two hypersurface, and
all degree two hypersurfaces have the same intersection properties, so any degree two
hypersurface is in the class 2[H ]. Similarly, any degree d hypersurface in P5 will be in
the class d[H ].
Multiplication in the Chow ring corresponds to intersection. If two varieties Y1 and
Y2 intersect transversally, then the Chow ring product [Y1 ] · [Y2 ] is defined to be the
class [Y1 ∩ Y2 ]. In P5 , the intersection of five general hyperplanes is just one point,
so [H ]5 is the class of one point. Intersecting five hypersurfaces of general degree
corresponds in the Chow ring to the multiplication
d1 [H ] · d2 [H ] · d3 [H ] · d4 [H ] · d5 [H ] = d1 · d2 · d3 · d4 · d5 [H ]5 ,
which represents d1 · d2 · d3 · d4 · d5 points.

Now let us describe the Chow ring of the blowup Bl V P5 of P5 along the Veronese.
First, consider a hyperplane H in P5 that does not contain the Veronese. Since H does
not contain the Veronese, its strict transform H̃ is equal to its inverse image π −1 (H ).
We will take the class of H̃ to be one generator of the Chow ring of Bl V P5 . If Y is a
hypersurface of degree d in P5 , then [π −1 (Y )] = d[ H̃ ].
The exceptional divisor does not behave like d[ H̃ ] for any d, so it represents a
new class [E] in the Chow ring. Since Bl V P5 and P5 are isomorphic away from the
Veronese, this is the only new generator of the Chow ring. Thus, any hypersurface in
Bl V P5 will be represented by a class m[ H̃ ] + n[E].
Now let Y be a general degree d hypersurface in P5 containing the Veronese. Then
−1
[π (Y )] = [Ỹ ] + n[E] for some n, so we see that
[Ỹ ] = d[ H̃ ] − n[E]. (6)

Next we would like to compute the integer n in equation (6). This integer will represent
“how much” π −1 (Y ) contains E.
We say a function F vanishes to order n along a variety Z if F and all its par-
tial derivatives of order < n vanish everywhere on Z. For instance, F = (x − 3y)2
vanishes to order 2 along the line L : x − 3y = 0 because F, Fx = 2(x − 3y), and
Fy = −6(x − 3y) all vanish along L, while Fx x = 2 does not vanish on L.
Since Y is a hypersurface in P5 , it is defined by one polynomial equation PY = 0. Its
inverse image π −1 (Y ) is defined by PY ◦ π = 0. The integer n appearing in equation
(6) is the order of vanishing of the polynomial PY ◦ π along E. Fortunately it turns out
that this is equal to the order of vanishing of PY along V = π(E) (see [7, p. 106] for a
proof). So the integer n is the largest integer such that PY and all its partial derivatives
of order < n lie in I(V ) (here I(V ) is the ideal of functions vanishing on the Veronese
surface V ; it is generated by the six equations (4)).
Let us find the strict transforms we need to solve our enumerative problems. Given
a point p, there are many double line conics that do not pass through that point, so the
hyperplane H p of conics through p does not contain the Veronese. Thus [ H̃ p ] = [ H̃ ].
The hypersurface H of conics tangent to the line has degree 2. It is easy to see
that the defining equation (3) for H vanishes along V ; a computer algebra system
will verify that its first partial derivatives do not vanish along V . So by (6), [ H̃ ] =
2[ H̃ ] − [E]. The hypersurface HQ of conics tangent to the conic Q has degree 6.
The defining equation for HQ vanishes along V ; its first partial derivatives also vanish
along V but its second derivatives do not. So the strict transform can be written as
[ H̃Q ] = 6[ H̃ ] − 2[E].
5.3. Counting conics. Now we use an intersection computation in the Chow ring
of the blowup to compute the answer to Steiner’s problem! By intersecting the strict
transforms in the blowup, we are throwing away all of the extra double line solutions
that caused us trouble before.
In the last section we showed that
[ H̃ p ] = [ H̃ ],
[ H̃ ] = 2[ H̃ ] − [E], (7)
and [ H̃Q ] = 6[ H̃ ] − 2[E].
Now that we have everything in terms of [ H̃ ] and [E], we could figure out how
to intersect combinations of those and finish our calculations. But an easier way is to
notice that we can instead write everything in terms of [ H̃ p ] and [ H̃ ]. We computed
the intersections of these in sections 3 and 4.3 when we answered questions like “How
many conics pass through 3 given points and are tangent to 2 given lines?” From these
earlier calculations, we learned that in the Chow ring of the blowup,
[ H̃ p ]5 = [ H̃ ]5 = 1,
[H˜ p ] [ H̃ ] = [ H̃ p ][ H̃ ]4 = 2,

4
[H˜ p ] [ H̃ ]2 = [ H̃ p ]2 [ H̃ p ]3 = 4.
3
(Note that on the right-hand side we are abusing notation: 1, 2, and 4 represent mul-
tiples of the Chow ring class of a point.) From (7) we see that
[ H̃Q ] = 6[ H̃ ] − 2[E] = 2[ H̃ p ] + 2[ H̃ ]. (8)
718
The answer to Steiner’s original problem, “How many conics are tangent to five given
conics?”, is
[ H̃Q ]5 = (2[ H̃ p ] + 2[ H̃ ])5

= 32([ H̃ p ]5 + 5[ H̃ p ]4 [ H̃ ] + 10[ H̃ p ]3 [ H̃ ]2
+ 10[ H̃ p ]2 [ H̃ ]3 + 5[ H̃ p ][ H̃ ]4 + [ H̃ ]5 )
= 32(1 + 5(2) + 10(4) + 10(4) + 5(2) + 1)
= 3264.
In a similar way, for each choice of c, , and p satisfying c + + p = 5, we obtain

the answer to the question “How many conics pass through p points and are tangent
to lines and c conics in general position?” Table 3 lists these answers.
Table 3. The number of conics through p points, tangent to lines,

and tangent to 5 − p − conics in general position.
points
0 1 2 3 4 5
0 3264 816 184 36 6 1
1 816 224 56 12 2
2 184 56 16 4
lines
3 36 12 4
4 6 2
5 1
6. VISUALIZING THE 3264 CONICS. In order to supplement our understanding

of Steiner’s problem, we’ll give a geometric construction that produces the 3264 conics
tangent to five special conics. This construction is due to Fulton (see [26] and also
[22]).
Consider five lines L 1 , . . . , L 5 in general position, each containing a marked point
Pi ∈ L i . Instead of looking at a single enumerative problem, we consider simultane-
ously all of the point-line enumerative problems involving some of the lines and some
of the marked points. For example, we can ask for the conics through points P1 and
P4 that are tangent to lines L 2 , L 3 , and L 5 . We know from section 2 that there are four
such conics, as in Figure 11.
More generally, for any two of the points we pick, there are four conics through
those points that are tangent to the other three lines. There are 52 = 10 ways to
choose the two points. If we choose a different number of points, say n, there are

Figure 11. Four conics through two points and tangent to three lines.
5
n
[ H̃ p ]n [ H̃ ]5−n conics passing through n of our five points and tangent to the other
5 − n lines. Summing over all n, we get
[ H̃ p ]5 + 5[ H̃ p ]4 [ H̃ ] + 10[ H̃ p ]3 [ H̃ ]2 + 10[ H̃ p ]2 [ H̃ ]3 + 5[ H̃ p ][ H̃ ]4 + [ H̃ ]5

= ([ H̃ p ] + [ H̃ ])5 = 102
for the total number of conics that satisfy any 5 of the 10 conditions imposed by the
points and lines.
Now we make each of the five lines into a conic, first by thinking of it as a double
line (with a double marked point) and then by deforming the double line into a hyper-
bola, as illustrated in Figure 12.
Figure 12. Left: Holding y constant gives cross sections of the surface x 2 y = y 2 + z 2 that deform to a double
line. Right: Superimposed snapshots of the deformation in the plane.
Let’s look for conics tangent to these five hyperbolas. For each conic that was tan-
gent to L i , there are two conics tangent to the hyperbola. And for each conic that
passed through the point Pi , there are two conics tangent to the hyperbola. In a sense,
this is the geometric meaning of equation (8), [ H̃Q ] = 2[ H̃ p ] + 2[ H̃ ]. The hyperbolas
are pictured in Figure 13.
Thus, deforming each line into a hyperbola doubles the number of conics. Since
there are five lines, the total number of conics when all five are deformed is
25 ([ H̃ p ] + [ H̃ ])5 = 32(102) = 3264.
7. THE STRICT TRANSFORMS MEET TRANSVERSALLY. How can we be

sure that our Chow ring computations give the correct number of reduced conics, when
a similar technique (a naive application of Bézout’s theorem) failed on P5 ? In section
720
Figure 13. Deforming the line L i gives rise to twice as many tangent conics.
5.3 we showed that the intersection of the five hypersurfaces H̃Qi is in the class rep-
resenting 3264 points. We would like to prove that for most sets of five conics, this
intersection is in fact 3264 points of multiplicity one, each corresponding to a reduced
conic. As we did before, we will say that a set of five conics with this property is in
general position. The criteria for five conics to be in general position are quite compli-
cated (including such restrictions as “No three of the conics have a common tangent
line”); a list of such criteria can be found in [3]. However, without going into details
about what constitutes general position for the five conics, we can prove that almost
all sets of five conics are in general position.
One way to formalize the notion of “almost all sets of five conics” is to use the
Zariski topology. Just as small open balls play a key role in analysis, the Zariski
topology is a crucial tool in algebraic geometry. In the Zariski topology, a set is closed
if it is the set of common solutions to a collection of polynomial equations. If you
imagine a hypersurface, you see immediately that Zariski closed sets are very thin. A
Zariski open set is just the complement of a Zariski closed set, so saying that a property
holds on a nonempty Zariski open set means that it holds in almost all situations.3
Theorem 3. The set of all (Q 1 , Q 2 , Q 3 , Q 4 , Q 5 ) ∈ (P5 )5 such that the H̃Qi intersect
transversally in points corresponding to reduced conics is open in the Zariski topology.
Proof. We will prove that the complement is closed in the Zariski topology.
Suppose we have five conics Q 1 , . . . , Q 5 and let
ai x 2 + bi x y + ci y 2 + di x z + ei yz + f i z 2 = 0
be the defining equation for the conic Q i . Let P be a point in the intersection of
the H̃Qi ⊂ Bl V P5 . The blowup is a five-dimensional manifold embedded in the ten-
dimensional space P5 × P5 so at P the blowup has a five-dimensional tangent space
T P (Bl V P5 ) that is a subspace of T P (P5 × P5 ) (we can think of P as lying at the ori-
gin of this space). Each of the five hypersurfaces H̃Qi has a tangent space that is a
hyperplane in T P (Bl V P5 ). To describe these hyperplanes, let’s consider H̃Qi in a chart
near P. To be precise, we have P ∈ Bl V P5 ⊂ P5 × P5 and we can pick charts on both
3 Warning: though the Zariski topology is fundamental for algebraic geometry, it takes some getting used
to. For instance, since the nonempty open sets are so large, it is not Hausdorff. It is a good exercise for the
reader to check that the Zariski open sets form a topology.

factors and intersect with the blowup to get a chart on the blowup. The point P was
described by 12 homogeneous coordinates; after we pass to a chart, it is described by
10 affine coordinates x0 , . . . , x9 . The equation for H̃Qi on this chart is the restriction
of a polynomial gi (x0 , . . . , x9 ) to the blowup; the coefficients of this polynomial gi
are themselves polynomials in the six homogeneous coordinates ai , . . . , f i describing
Q i ∈ P5 . The gradient ∇gi (P) is a vector with ten entries, each of them a polynomial
in 16 variables: x0 , . . . , x9 and ai , . . . , f i . This vector consists of the coefficients of a
linear function that, when restricted to T P (Bl V P5 ), vanishes precisely on the tangent
space to H̃Qi at P. This defines a codimension 1 subspace in T P (Bl V P5 ).
Now we ask: “What conditions do we need to put on our conics Q i so that these
five codimension 1 subspaces intersect in just a point?” We just need the five linear
conditions to be independent. We can check this by forming the Jacobian matrix,
⎛ ⎞
∇g1 (P)
⎜∇g2 (P)⎟
∂gi ⎜ ⎟
J= = ⎜∇g3 (P)⎟ ,
∂xj ⎝∇g (P)⎠
4
∇g5 (P)
and seeing whether it has full rank. The matrix will fail to have full rank precisely
when all its 5 by 5 minors vanish. Each of these minors is the determinant of a 5
by 5 submatrix, so they are polynomials in the entries of the ∇gi (P). Thus, these
are polynomial conditions involving the variables describing the Q i and the variables
describing P; that is, these are polynomial conditions on the product of (P5 )5 with our
chart in Bl V P5 . Since Bl V P5 is covered by charts, we obtain polynomials that cut out
the subset S ⊂ (P5 )5 × Bl V P5 consisting of tuples
{(Q 1 , Q 2 , Q 3 , Q 4 , Q 5 , P) : P ∈ ∩ H̃Qi and the intersection is not transverse at P}.
Because this set is defined by the vanishing of polynomial equations, it is closed in the
Zariski topology.
Now we turn our attention to the points in the intersection that correspond to the
double line conics. The exceptional divisor E = π −1 (V ) ⊂ Bl V P5 is a closed set be-
cause it is the inverse image of the closed set V under the continuous map π. Indeed, if
the Veronese V is obtained as the set of common zeros of polynomials G i then the ex-
ceptional divisor E = π −1 (V ) is the set of common zeros of the polynomials G i ◦ π
on Bl V P5 . It follows that the product (P5 )5 × E is closed in (P5 )5 × Bl V P5 . As the
Zariski closed sets form a topology, the union S = S ∪ [(P5 )5 × E] is also closed.
Now one great fact about projective varieties is that if we have a projection from one
projective variety to another, then the image of a Zariski closed set is closed.4 So the
projection π1 : (P5 )5 × Bl V P5 → (P5 )5 that drops the last factor takes the closed set S
to a closed set. This says that the set of configurations (Q 1 , Q 2 , Q 3 , Q 4 , Q 5 ) ∈ (P5 )5
such that either the H̃Qi fail to intersect transversally or their intersection includes
points on the exceptional divisor is closed in the Zariski topology. Therefore its com-
plement, the set of configurations where the H̃Qi intersect transversally in points cor-
responding to reduced conics, is open in the Zariski topology.
This theorem is not quite enough to guarantee that for most configurations of five
plane conics the intersection of the H̃Qi is transverse and consists of isolated points
4 This is not true for affine spaces though. For example, the image of the parabola x y = 1 projected onto
the x-axis consists of the entire line except for the origin.
722
corresponding to reduced conics. We’ve shown that the set parameterizing such config-
urations is Zariski open, but we have not shown that it is nonempty. This is a common
difficulty in algebraic geometry: it is easy to show that a property holds on an open set
and hard to show that this open set is not empty! To do this, we just need to produce
one example of five conics so that the intersection of the H̃Qi is transverse and con-
sists of points corresponding to reduced conics. We can avoid checking the transverse
condition by just producing an instance of the problem where the solution consists of
precisely 3264 reduced conics; these will all be isolated and count with multiplicity
one since the intersection must be rationally equivalent to 3264 points. Fortunately,
we’ve already produced such an example in section 6.5 So the Zariski open set in The-
orem 3 is not empty and for most configurations of five conics there are 3264 reduced
conics tangent to all five. A precise characterization of the sets of five conics for which
we do not get 3264 reduced conics tangent to all five can be found in [3].
In principle we’d need to repeat this work for each of the other problems stemming
from combinations of five points, lines, and conics. This can be done by appealing to
arguments similar to those given here, or by relying on a general transversality result
due to Kleiman [15, p. 273]. The upshot is that for all our enumerative questions,
almost all configurations give rise to a finite number of reduced solutions and this
number is given in Table 3.
8. CODA. We’ve answered several enumerative questions involving conics; however,

the true value of these problems lies in their connection to interesting mathematics. We
hope that this article whets the reader’s appetite for more algebraic geometry, and with
this in mind, we make a few suggestions for further reading. As well, we’ve always
felt that we understand a subject better after working a few exercises. We include some
fun problems that further develop some of the material we’ve discussed.
Suggestions for further reading. In recent years, enumerative geometry has been
heavily influenced by an influx of ideas from string theory. The major breakthrough
that caused mathematicians to sit up and take notice was a prediction in 1991, using
mirror symmetry, of the number of degree d rational curves on a degree 5 hypersurface
in P4 [4]. This physics computation was not mathematically rigorous, but at the time
these numbers were known to algebraic geometers only for very small d, so it was
amazing to have predictions for all of the numbers at once. The development of the
field of Gromov-Witten theory has put this computation on solid mathematical foot-
ing, as well as leading to many other interesting results. A recent book by Sheldon
Katz [17] provides an introduction to this aspect of enumerative geometry, and we
recommend it very highly.
Duality played a key role in our solutions to enumerative problems involving lines
and points. The theory of duality (and the discriminants that define the dual varieties)
is given extensive treatment in [14], where it is related to toric varieties and systems
of hypergeometric differential equations. Blowing up also played a key role in our
solution to Steiner’s problem. The blowup is commonly used to resolve singularities
in algebraic geometry. Indeed, Hironaka was awarded the Fields medal for showing
that every variety can be desingularized by a sequence of blowups. A brief account of
this theorem can be found in [24, Chap. 7] and an expository proof in [16].
5 Fulton and MacPherson [12] use a dimension count to give a different proof that there must be such an
example. They show that the collection of quintuples of conics for which we get solutions of multiplicity
higher than one is of lower dimension than the set of quintuples of conics itself. So in particular, there is some
configuration of five conics all of whose solutions have multiplicity one (all of the H̃ Q i intersect transversally).

We used techniques from computational algebra and computer algebra systems
throughout the paper. The Macaulay2 book [9] explains how to use a computer al-
gebra system to solve many problems in algebraic geometry. In particular, the reader
is referred to Sottile’s paper [27], which deals directly with enumerative questions.
There are several other ways to solve Steiner’s problem, but all revolve around
removing the double line conics from our count. One way is to generalize Bézout’s
theorem by assigning an intersection multiplicity to each component in the intersec-
tion of our hypersurfaces. Serre [23] showed that the correct intersection multiplicity
for these and more general intersections can be computed using the Tor functor from
commutative algebra. This can be done in low-dimensional examples on a computer
[8].
Another approach to assigning intersection multiplicities is easier to visualize.
To start, if we have n hypersurfaces lying in general position and having degrees
d1 , . . . , dn , then by Bézout’s theorem their intersection consists of d1 · · · dn points.
The reason that we have entire components in the intersection is that our hypersur-
faces are not in general position. However, we can deform our hypersurfaces (changing
each of the coefficients of our hypersurface from a constant to a function of a variable
t, which is equal to our given coefficients when t = 0) so that they are in general
position for t = 0. For nonzero t, we get d = d1 · · · dn points p1 (t), . . . , pd (t), each
a function of the parameter t. If our family deforms nicely (the technical condition
is that the family is flat) then we would expect these d points to approach d limiting
points as t → 0. The number of points that land on each component is the intersection
multiplicity of the component. With this definition, we can determine the contribution
to the Bézout number from each of the higher dimensional components and by sub-
traction compute the number of isolated points (corresponding to conics that are not
counted with multiplicity) in our solution. Katz [17, Chap. 8] gives a nice example of
this process in action.
A final approach to intersection multiplicities involves computations with Chern
classes of vector bundles, leading to the so-called characteristic numbers. See [11,
Sec. 10.4] for an application of these techniques to Steiner’s problem and Fulton’s
lecture notes [10] for a broad overview of intersection theory. A very accessible dis-
cussion of characteristic numbers, together with a history of Steiner’s problem, can
be found in Kleiman’s article [19]. This theory is sufficient to enumerate the conics
tangent to five given plane curves in general position. As in the case of conics, each
curve C gives rise to a class deg(C)[ H̃ p ] + deg(Č)[ H̃ ] and the answer comes from
finding the product of the five classes in Bl V P5 .
There are plenty of other enumerative problems with connections to algebraic ge-
ometry. Schubert calculus deals with enumerative problems involving linear spaces,
rather than conics; for example, “How many lines in P3 meet four other lines in gen-
eral position?” Kleiman and Laksov [21] give a nice introduction to Schubert calculus.
Problem 1. How many lines are simultaneously tangent to two conics in general po-
sition? [Hint: Think of the dual picture.]
Problem 2. To each conic Q : ax 2 + bx y + cy 2 + d x z + eyz + f z 2 = 0 we asso-

ciate the matrix
⎡ ⎤
a b/2 d/2
M Q = ⎣b/2 c e/2 ⎦ ,
d/2 e/2 f

so that if xT = x y z then the equation for the conic is given by xT M Q x = 0.
724
(a) Show that if x̃ = Lx is a linear change of coordinates, then when the quadratic
Q is written in the new variables x̃, it corresponds to the matrix (L −1 )T M Q L −1 .
(b) Show that given any nondegenerate conic Q, there is a linear change of coordi-
nates transforming it to yz = x 2 .
(c) Show that Q is degenerate if and only if det(M Q ) = 0, and Q is a double line
if and only if rank(M Q ) = 1.
(d) Show that if Q is nonsingular, then its dual curve Q̌ corresponds to the matrix
M Q̌ = M Q−1 . In this sense the duality map is a generalization of the inverse op-
eration for matrices. This topic is explored in great detail (for multidimensional
matrices!) in [14].
Problem 3. Verify that if [xi : yi : z i ] (1 ≤ i ≤ 5) are five points in general position,

then the unique conic ax 2 + bx y + cy 2 + d x z + eyz + f z 2 = 0 passing through them
is given by the vanishing of the determinant of the matrix
⎡ 2 ⎤
x1 x1 y1 y12 x1 z 1 y1 z 1 z 12
⎢x 2 x2 y2 y 2 x2 z 2 y2 z 2 z 2 ⎥
⎢ 2 2 2⎥
⎢x 2 x y y 2 x z y z z 2 ⎥
⎢ 3 3 3 3 3 3 3 3⎥
⎢ 2 3
⎥.
⎢x4 x4 y4 y42 x4 z 4 y4 z 4 z 42 ⎥
⎢ 2 ⎥
⎣x5 x5 y5 y52 x5 z 5 y5 z 5 z 52 ⎦
x 2 x y y2 x z yz z 2
Problem 4. (a) Show that the point in P5 corresponding to a conic Q lies on the
hypersurface HQ of all conics tangent to Q. [Hint: Consider the case Q :
x 2 − yz = 0.]
(b) Show that if P is a point on HQ then the entire line joining P and the point in
P5 corresponding to Q is on HQ too. This shows that HQ is a cone over Q.
Problem 5. (a) A circle is a conic passing through the two points [1 : i : 0] and
[1 : −i : 0]. Show that when we homogenize the curve defined by x 2 + y 2 = r 2
we get a circle in P2 .
(b) Show that the parameter space for circles in P2 is a 3-dimensional projective
space.
(c) One approach to count the circles tangent to three general circles is to compute
[ H̃Q ]3 [ H̃ p ]2 = 184 in Bl V (P5 ). Why is this incorrect? Find the correct count.
[Remark: In fact, something very precise can be said if the 3 given circles are mutually
tangent. If r1 , r2 , r3 , and r4 are the radii of the circles, then
2(r12 + r22 + r32 + r42 ) = (r1 + r2 + r3 + r4 )2 .
This is a famous theorem, known to Descartes and extended and immortalized in Sir
Frederick Soddy’s poem, “The Kiss Precise” [25].]
Problem 6. Let γ : P2 × P2 → P5 be the map that sends ([A : B : C], [D : E : F])

to the point representing the conic (Ax + By + C z)(Dx + E y + F z) = 0. We call the
image of γ the degenerate variety because it parameterizes the degenerate conics.
(a) The map ν2,2 : P2 × P2 → P8 given by ([A : B : C], [D : E : F]) → [AD :
B D : C D : AE : B E : C E : AF : B F : C F] is called the Segre map and its

image the Segre variety. Show that the Segre map is an injection and show
that the map γ factors through the Segre map in the sense that there is a map
π : P8 → P5 such that γ = π ◦ ν2,2 .
(b) Show that the map γ is generically 2 to 1. Where does this map ramify? That
is, where does this map fail to be 2 to 1?
(c) Show that the degenerate variety is a degree 3 hypersurface in P5 by a geometric
argument.
(d) Now find the defining equation of the degenerate variety. Hint: use part (c) of
Problem 2.
(e) Show that the exceptional divisor E ⊂ Bl V P5 is isomorphic to the degenerate
variety. In particular, if π2 : P5 × P5 → P5 is the projection onto the second
factor, then π2 | E is an isomorphism onto its image and π2 (E) is the degenerate
variety.
(f) Thus E consists of pairs (Q, L 1 ∪ L 2 ) where Q is a double line conic and
L 1 ∪ L 2 is a crossed line conic. Show that if (Q, L 1 ∪ L 2 ) ∈ E then the line
corresponding to Q is the dual of the point of intersection of L 1 and L 2 .
(g) The two points Ľ 1 and Ľ 2 lie on the double line Q, so we can think of Q as
being a double line with two marked points. While double lines do not have
well-defined duals, if we dualize a double line with two marked points, we get
the pair of crossed lines corresponding to the duals of the marked points. This
interpretation gives a way to define duality on the blowup. Show that with this
interpretation of duality the points on the blowup all have the form (Q, Q̌) and
the duality map just swaps the two coordinates. Because the duality map is
well-defined on the blowup, Bl V P5 is sometimes called the space of complete
conics.
(h) Now suppose that Q is a double line with two marked points, and the two points
come together. What happens in the dual picture?
(i) Using the duality map δ from part (g) show that we get δ( H̃Q ) = H̃Q̌ , [δ( H̃ p )] =
[ H̃ ], and [δ( H̃ )] = [ H̃ p ]. Since Q and Q̌ are both conics, [ H̃Q ] = [ H̃Q̌ ] =
a[ H̃ p ] + b[ H̃ ], say. Applying δ, conclude that a = b. Then show that a =
b = 2 only using the degrees of H p , H , and HQ . This gives another proof of
formula (8), originally due to Clebsch and Lindemann [19, p. 121].
Problem 7. Instead of using the moduli space of conics themselves, one can do enu-
merative calculations with the moduli space of stable maps. This is the point of view
taken in Gromov-Witten theory.
(a) Consider the projective line P1 with homogeneous coordinates [S : T ]. Find
a degree 2 morphism from P1 to P2 whose image is the curve yz = x 2 . (In
this case “degree” means the degree of the polynomial functions defining the
morphism.)
(b) By composing the (inverse of the) change of coordinates in part (b) of Problem
2 with the map in part (a), we see that every nondegenerate conic is the image
of P1 under some degree 2 map. Show that any degree 2 map from P1 to P2 is
either a one-to-one map onto a nondegenerate conic, or a two-to-one map onto
a line with two branch points.
(c) Show that composing a map from P1 to P2 with a linear change of coordinates
on P1 , [S : T ] → [aS + bT : cS + dT ], with ad − bc = 0, does not change
726
the image. Two maps that differ by such an automorphism of P1 are considered
isomorphic.
(d) In general, the moduli space of stable genus 0 maps to P2 consists of (isomor-
phism classes of) maps from a tree of P1 ’s to P2 . In the degree 2 case, the
only possible trees have either one or two “branches.” Thus a stable degree 2
map is either a degree 2 map from P1 to P2 , such as the maps in part (b), or a
map from a pair of intersecting P1 ’s to P2 , where the map has degree 1 on each
“branch.” Explain why the moduli space of stable degree 2 genus 0 maps to P2
is essentially the same as the blowup Bl V P5 , the space of complete conics.
ACKNOWLEDGMENTS. We first learned about this topic from Frank Sottile, whose expository lectures
and writing guided our early investigations. Mike Roth provided help at a crucial time; his suggestions form
the basis for our proof of Theorem 3. Steve Kleiman gave us extensive comments on a first draft of the paper
and rescued us from several errors. Helene Speer’s gracious help improved our figures. This paper grew out of
Midshipman Andrew Bashelor’s Trident project [3], a full-year research project carried out at the U.S. Naval
Academy. We are grateful to the Trident Scholar Committee for their oversight and guidance. In particular
we gratefully acknowledge the Office of Naval Research for partial support of this work in funding document
N0001405WR20153.
REFERENCES
1. M. Artin, Algebra, Prentice Hall, Englewood Cliffs, NJ, 1991.

2. D. Ayala and R. Cavalieri, Counting bitangents with stable maps, Expositiones Mathematicae 24 (2006)
307–336; also available at http://arxiv.org/abs/math.AG/0505139.
3. A. Bashelor, Enumerative Algebraic Geometry: Counting Conics, Trident Scholar Project Report, no.
330, United States Naval Academy, Annapolis, MD, 2005.
4. P. Candelas, X. de la Ossa, P. Green, and L. Parkes, A pair of Calabi-Yau manifolds as an exactly soluble
superconformal theory, Nuclear Phys. B 359 (1991) 21–74.
5. D. Cox, J. Little, and D. O’Shea, Using Algebraic Geometry, Graduate Texts in Mathematics, vol. 185,
Springer-Verlag, New York, 1998.
6. H. Derksen and G. Kemper, Computational Invariant Theory, Encyclopaedia of Mathematical Sciences,
vol. 130, Springer-Verlag, Berlin, 2002.
7. D. Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Texts in Mathe-
matics, vol. 150, Springer-Verlag, New York, 1995.
8. , Projective geometry and homological algebra, in Computations in Algebraic Geometry with
Macaulay 2, D. Eisenbud, D. R. Grayson, and M. Stillman, eds., Algorithms and Computation in Math-
ematics, vol. 8, Springer-Verlag, Berlin, 2002, 17–40.
9. D. Eisenbud, D. R. Grayson, and M. Stillman, eds., Computations in Algebraic Geometry with Macaulay
2, Algorithms and Computation in Mathematics, vol. 8, Springer-Verlag, Berlin, 2002.
10. W. Fulton, Introduction to Intersection Theory in Algebraic Geometry, CBMS Regional Conference
Series in Mathematics, vol. 54, Conference Board of the Mathematical Sciences, Washington, DC, 1984.
11. , Intersection Theory, 2nd ed., Results in Mathematics and Related Areas, 3rd Series, A Series of
Modern Surveys in Mathematics, vol. 2, Springer-Verlag, Berlin, 1998.
12. W. Fulton and R. MacPherson, Defining algebraic intersections, in Algebraic Geometry (Proc. Sympos.,
Univ. Tromsø, Tromsø, 1977), Lecture Notes in Mathematics, vol. 687, Springer-Verlag, Berlin, 1978,
1–30.
13. W. Fulton and R. Pandharipande, Notes on stable maps and quantum cohomology, in Algebraic
geometry—Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, American Mathematical Society, Prov-
idence, RI, 1997, 45–96.
14. I. M. Gelfand, M. M. Kapranov, and A. V. Zelevinsky, Discriminants, Resultants, and Multidimensional
Determinants, Mathematics: Theory & Applications, Birkhäuser, Boston, MA, 1994.
15. R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, vol. 52, Springer-Verlag, Berlin,
1977.
16. H. Hauser, The Hironaka theorem on resolution of singularities (or: A proof we always wanted to under-
stand), Bull. Amer. Math. Soc. (N.S.) 40 (2003) 323–403.
17. S. Katz, Enumerative Geometry and String Theory, American Mathematical Society, Providence, RI,
2006.

18. S. Kleiman, Problem 15: Rigorous foundation of Schubert’s enumerative calculus, in Mathematical De-
velopments Arising from Hilbert Problems, Proc. Symp. Pure Math., vol. 28, American Mathematical
Society, Providence, RI, 1976, 445-482.
19. , Chasles’s enumerative theory of conics: A historical introduction, in Studies in Algebraic Geom-
etry, Mathematical Association of America Studies in Mathematics, vol. 20, Mathematical Association
of America, Washington, DC, 1980, 117–138.
20. , Intersection theory and enumerative geometry: A decade in review, With the collaboration of
Anders Thorup on section 3, in Algebraic Geometry–Bowdoin 1985, Proc. Sympos. Pure Math., vol. 46,
Part 2, American Mathematical Society, Providence, RI, 1987, 321–370.
21. S. L. Kleiman and D. Laksov, Schubert calculus, this M ONTHLY 79 (1972) 1061–1082.
22. F. Ronga, A. Tognoli, and T. Vust, The number of conics tangent to 5 given conics: The real case, Rev.
Mat. Univ. Complut. Madrid 10 (1997) 391–421.
23. J.-P. Serre, Algèbre Locale. Multiplicités, 2nd ed., Lecture Notes in Mathematics, vol. 11, Springer-
Verlag, Berlin, 1965.
24. K. E. Smith, L. Kahanpää, P. Kekäläinen, and W. Traves, An Invitation to Algebraic Geometry, Universi-
text, Springer-Verlag, New York, 2000.
25. F. Soddy, The kiss precise, Nature 137 (1936) 1021.
26. F. Sottile, Enumerative geometry for real varieties, in Algebraic geometry—Santa Cruz 1995, Proc. Sym-
pos. Pure Math., vol. 62, American Mathematical Society, Providence, RI, 1997, 435–447.
27. , From enumerative geometry to solving systems of polynomials equations, in Computations in
Algebraic Geometry with Macaulay 2, D. Eisenbud, D. R. Grayson, and M. Stillman, eds., Algorithms
and Computation in Mathematics, vol. 8, Springer-Verlag, Berlin, 2002, 101–129.
ANDREW BASHELOR received his B.Sc. from the United States Naval Academy in 2005 and an M.Sc. in
Operations Analysis from the Naval Postgraduate School in 2006. He is currently serving as an officer in the
U.S. Navy, stationed in Hawaii.
andrew.bashelor@mac.com
AMY KSIR grew up in Laramie, Wyoming. She received her B.A. from Rice University and her Ph.D. from
the University of Pennsylvania, and currently teaches at the U.S. Naval Academy. Her research interests are in
applications of algebraic geometry to string theory and coding theory; she also enjoys playing the accordion.
Department of Mathematics, U.S. Naval Academy, Annapolis, MD, 21402
ksir@usna.edu
WILL TRAVES received his B.Sc.H. from Queen’s University and his Ph.D. from the University of Toronto.
He teaches at the U.S. Naval Academy, where he helps coach the women’s rugby team.
Department of Mathematics, U.S. Naval Academy, Annapolis, MD, 21402
traves@usna.edu
728

Enumerative Algebraic Geometry of Conics

Uploaded by

Copyright:

Available Formats

Enumerative Algebraic Geometry of Conics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Enumerative Algebraic Geometry of Conics

Uploaded by

Copyright:

Available Formats

Enumerative Algebraic Geometry of Conics

Andrew Bashelor, Amy Ksir, and Will Traves

1. INTRODUCTION. In 1848 Jakob Steiner, professor of geometry at the Univer-

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 701

2. A MODULI SPACE OF PLANE CONICS. A plane conic curve is the set of

Any conic is completely determined by the coefficients a, b, c, d, e, and f of its

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 703

3. SOME BASIC ENUMERATIVE QUESTIONS. In this section we are going

Figure 2. The case of 4 collinear points.

Figure 3. Choosing coordinates.

tv(s(v − 1) − u(t − 1)) = 0

d × d submatrices vanish [1, p. 153, ex. 10].

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 705

(t − v, s − u) ∩ (t, s − 1) ∩ (t − 1, s) ∩ (t, s) ∩ (v, u) ∩ (v, u − 1) ∩ (v − 1, u)

Figure 4. Two conics through four points and tangent to a line.

Figure 5. A pathological example.

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 707

where the coefficients are allowed to be complex numbers.

Theorem 2 ([24]). If n hypersurfaces of degrees d1 , d2 , . . . , dn in Pn intersect trans-

The hypersurfaces X 1 , . . . , X n intersect transversally at a point P ∈ X 1 ∩ · · · ∩

Figure 7. Nontransverse intersections—left and center—and a transverse intersection, right.

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 709

Table 1. Number of conics through p points and tangent to lines.

4. EXCESS INTERSECTION AND THE DUALITY OF P2

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 711

a(BY + C Z)2 − b AY (BY + C Z) + c A2 Y 2

(e2 − 4c f )A2 + (4b f − 2de)AB + (d 2 − 4a f )B 2 + (4cd − 2be)AC

Figure 9. Tangency under duality (we show that Lˇ2 = Lˇ1 ).

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 713

(AX + BY + C Z)2 = A2 X 2 + 2AB X Y + B 2 Y 2

Lines in P2 are parameterized by P̌2 , so the Veronese is indeed a surface (two-

[A : B : C] → [A2 : 2AB : B 2 : 2AC : 2BC : C 2 ].

Definition. The blowup of P5 along the Veronese surface, Bl V P5 , is the closure of

Eliminating λ and cross multiplying, we get fifteen equations:

r (4b f − 2de) = s(e2 − 4c f ) s(d 2 − 4a f ) = t (4b f − 2de) t (4ae − 2bd) = v(d 2 − 4a f )

bu + 2ew + 2cv = 0 bv + 2dw + 2au = 0

These eight equations are syzygies—linear relations (with polynomial coefficients)

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 715

2au = 0 2as = 0 4ar = 0.

([1 : 0 : 0 : 0 : 0 : 0], [0 : 0 : t : 0 : v : w]) .

These points define a P2 within the blowup, so π −1 ([a : b : c : d : e : f ]) ∼

Figure 10. A picture of the blowup.

which represents d1 · d2 · d3 · d4 · d5 points.

[Ỹ ] = d[ H̃ ] − n[E]. (6)

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 717

[H˜ p ] [ H̃ ] = [ H̃ p ][ H̃ ]4 = 2,

[ H̃Q ] = 6[ H̃ ] − 2[E] = 2[ H̃ p ] + 2[ H̃ ]. (8)

[ H̃Q ]5 = (2[ H̃ p ] + 2[ H̃ ])5

In a similar way, for each choice of c, , and p satisfying c + + p = 5, we obtain

Table 3. The number of conics through p points, tangent to lines,

0 3264 816 184 36 6 1

6. VISUALIZING THE 3264 CONICS. In order to supplement our understanding

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 719

[ H̃ p ]5 + 5[ H̃ p ]4 [ H̃ ] + 10[ H̃ p ]3 [ H̃ ]2 + 10[ H̃ p ]2 [ H̃ ]3 + 5[ H̃ p ][ H̃ ]4 + [ H̃ ]5

25 ([ H̃ p ] + [ H̃ ])5 = 32(102) = 3264.

7. THE STRICT TRANSFORMS MEET TRANSVERSALLY. How can we be

October 2008] ENUMERATIVE ALGEBRAIC GEOMETRY OF CONICS 721

{(Q 1 , Q 2 , Q 3 , Q 4 , Q 5 , P) : P ∈ ∩ H̃Qi and the intersection is not transverse at P}.

8. CODA. We’ve answered several enumerative questions involving conics; however,