ALGEBRAIC CURVES

B3b course 2013

Lecturer: Balazs Szendr oi
[email protected]
Notes: Nigel Hitchin
1
These notes trace a path through material which is covered in more detail in the book
for the course which is:
F Kirwan, Complex Algebraic Curves, LMS Student Texts 23, Cambridge 1992, Chap-
ters 26, 26.99
2
Contents
1 Projective spaces 5
1.1 Basic denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Linear subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Projective transformations . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Plane curves 17
2.1 Basic denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Conics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Rational parametrization of the conic . . . . . . . . . . . . . . . . . . 22
3 Intersections of curves 26
3.1 Resultants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Intersection multiplicity . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Cubic curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Bezouts theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 The genus of a curve 39
4.1 Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Maps to P
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 The degree-genus formula . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4 The torus and the cubic . . . . . . . . . . . . . . . . . . . . . . . . . 51
5 The Riemann-Roch theorem 54
5.1 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.2 Canonical divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3 Riemann-Roch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3
5.5 The group law on a cubic . . . . . . . . . . . . . . . . . . . . . . . . 65
4
1 Projective spaces
1.1 Basic denitions
Denition 1 Let V be a vector space. The projective space P(V ) of V is the set of
1-dimensional vector subspaces of V .
Denition 2 If the vector space V has dimension n + 1, then P(V ) is a projective
space of dimension n. A 1-dimensional projective space is called a projective line, and
a 2-dimensional one a projective plane.
For most of the course, the eld F of scalars for our vector spaces will be the complex
numbers C. Our intuition is best served, however, by thinking of the real case. So the
projective space of R
3
is the set of lines through the origin. Each such line intersects
the unit sphere S
2
= {x R
3
:

i
x
2
i
= 1} in two points u, so from this point
of view P(R
3
) is S
2
with antipodal points identied. Since each line intersects the
lower hemisphere, we could equally remove the upper hemisphere and then identify
opposite points on the equatorial sphere.
In the B3a Geometry of Surfaces course this is the way we think of the projective
plane, but it is less appropriate for an algebraic geometry course. Still, it does explain
why we should think of P(R
n+1
) as n-dimensional. In what follows we shall write
P(F
n+1
) as P
n
(usually for F = C) to make this more plain.
A better approach for our purposes is the notion of a representative vector for a point
of P(V ). Any 1-dimensional subspace of V is the set of multiples of a non-zero vector
v V . We then say that v is a representative vector for the point [v] P(V ). Clearly
if = 0 then v is another representative vector so
[v] = [v].
Now suppose we choose a basis {v
0
, . . . , v
n
} for V . The vector v can be written
v =
n

i=0
x
i
v
i
5
and the n +1-tuple (x
0
, . . . , x
n
) provides the coordinates of v V . If v = 0 we write
the corresponding point [v] P(V ) as [v] = [x
0
, x
1
, . . . , x
n
] and these are known as
homogeneous coordinates for a point in P(V ). Again, for = 0
[x
0
, x
1
, . . . , x
n
] = [x
0
, x
1
, . . . , x
n
].
Homogeneous coordinates give us another point of view of projective space. Let
U
0
P(V ) be the subset of points with homogeneous coordinates [x
0
, x
1
, . . . , x
n
]
such that x
0
= 0. (Since if = 0, x
0
= 0 if and only if x
0
= 0, so this is a
well-dened subset, independent of the choice of (x
0
, . . . , x
n
)). Then, in U
0
,
[x
0
, x
1
, . . . , x
n
] = [x
0
, x
0
(x
1
/x
0
), . . . , x
0
(x
n
/x
0
)] = [1, (x
1
/x
0
), . . . , (x
n
/x
0
)].
Thus we can uniquely represent any point in U
0
by one of the form [1, y
1
, . . . , y
n
], so
U
0

= C
n
.
The points we have missed out are those for which x
0
= 0, but these are the 1-
dimensional subspaces of the n-dimensional vector subspace spanned by v
1
, . . . , v
n
,
which is a projective space of one lower dimension. So we can write
P
n
= C
n
P
n1
A large chunk of complex projective n-space is thus our familiar C
n
.
Example: The simplest example of this is the case n = 1. Since a one-dimensional
projective space is a single point (if dimV = 1, V is the only 1-dimensional subspace)
the projective line P
1
= C pt. Since [x
0
, x
1
] maps to x
1
/x
0
C we usually call
this extra point [0, 1] the point . The projective line is what is called in complex
analysis the extended complex plane C {}.
Having said that, there are many dierent copies of C
n
inside P
n
, for we could have
chosen x
i
instead of x
0
, or coordinates with respect to a totally dierent basis. Pro-
jective space should normally be thought of as a homogeneous object, without any
distinguished copy of C
n
inside.
We dened U
0
above as the subset of P
n
where x
0
= 0, so we can similarly dene
subsets U
i
where x
i
= 0. Any point of P
n
lies in one of these sets. We can use them
to make P
n
into a topological space dene a set V to be open if V U
i
is an open
set in C
n
under the map
[x
0
, x
1
, . . . , x
n
] (x
0
/x
i
, . . . , 1, . . . x
n
/x
i
).
6
1.2 Linear subspaces
Denition 3 A linear subspace of the projective space P(V ) is the set of 1-dimensional
vector subspaces of a vector subspace U V .
Note that a linear subspace is a projective space in its own right, the projective space
P(U).
Recall that a 1-dimensional projective space is called a projective line. We have the
following two propositions which show that projective lines behave nicely:
Proposition 1 Through any two distinct points in a projective space there passes a
unique projective line.
Proof: Let P(V ) be the projective space and x, y P(V ) distinct points. Let u, v
be representative vectors. Then u, v are linearly independent for otherwise u = v
and
x = [u] = [v] = [v] = y.
Let U V be the 2-dimensional vector space spanned by u and v, then P(U) P(V )
is a line containing x and y.
Suppose P(U

) is another such line, then u U

and v U

and so the space spanned

by u, v (namely U) is a subspace of U

. But U and U

are 2-dimensional so U = U

and the line is thus unique. 2

Proposition 2 In a projective plane, two distinct projective lines intersect in a
unique point.
Proof: Let the projective plane be P(V ) where dimV = 3. Two lines are dened
by P(U
1
), P(U
2
) where U
1
, U
2
are distinct 2-dimensional subspaces of V . Now from
elementary linear algebra
dimV dim(U
1
+U
2
) = dimU
1
+ dimU
2
dim(U
1
U
2
)
so that
3 2 + 2 dim(U
1
U
2
)
and
dim(U
1
U
2
) 1.
7
But since U
1
and U
2
are 2-dimensional,
dim(U
1
U
2
) 2
with equality if and only if U
1
= U
2
. As the lines are distinct, equality doesnt occur
and so we have the 1-dimensional vector subspace
U
1
U
2
V
which is the required point of intersection in P(V ). 2
Any 2-dimensional subspace of C
3
is dened by a single equation
a
0
x
0
+a
1
x
1
+a
2
x
2
= 0
and if a
1
and a
2
are not both zero, this intersects U
0

= C
2
(the points where x
0
= 0)
where
0 = a
0
+a
1
(x
1
/x
0
) +a
2
(x
2
/x
0
) = a
0
+a
1
y
1
+a
2
y
2
which is an ordinary line in C
2
with coordinates y
1
, y
2
. The projective line has one
extra point on it, where x
0
= 0, i.e. the point [0, a
2
, a
1
]. Conversely, any line in C
2
extends uniquely to a projective line in P
2
.
Two lines in C
2
are parallel if they are of the form
a
0
+a
1
y
1
+a
2
y
2
= 0, b
0
+a
1
y
1
+a
2
y
2
= 0
but then the added point to make them projective lines is the same one: [0, a
2
, a
1
],
so the two lines meet at a single point on the line at innity P
1
.
One of our goals will be to have good theorems about the number of points of inter-
section of two curves, so introducing the projective plane gives us a start: any two
distinct lines meet in a single point.
1.3 Projective transformations
If V, W are vector spaces and T : V W is a linear transformation, then a vector
subspace U V gets mapped to a vector subspace T(U) W. If T has a non-
zero kernel, T(U) may have dimension less than that of U, but if ker T = 0 then
dimT(U) = dimU. In particular, if U is one-dimensional, so is T(U) and so T gives
a well-dened map
: P(V ) P(W).
8
Denition 4 A projective transformation from P(V ) to P(W) is the map dened
by an invertible linear transformation T : V W.
Note that if = 0, then T and T dene the same linear transformation since
[(T)(v)] = [(T(v))] = [T(v)].
The converse is also true: suppose T and T

dene the same projective transformation

. Take a basis {v
0
, . . . , v
n
} for V , then since
([v
i
]) = [T

(v
i
)] = [T(v
i
)]
we have
T

(v
i
) =
i
T(v
i
)
for some non-zero scalars
i
and also
T

(
n

i=0
v
i
) = T(
n

i=0
v
i
)
for some non-zero . But then
n

i=0
T(v
i
) = T(
n

i=0
v
i
) = T

(
n

i=0
v
i
) =
n

i=0

i
T(v
i
).
Since T is invertible, T(v
i
) are linearly independent, so this implies
i
= . Then
T

(v
i
) = T(v
i
) for all basis vectors and hence for all vectors and so
T

= T.
Example: You are, in fact, already familiar with one class of projective transfor-
mations Mobius transformations of the extended complex plane. These are just
projective transformations of the complex projective line P
1
to itself. We describe
points in P
1
by homogeneous coordinates [z
0
, z
1
], and then a projective transforma-
tion is given by
([z
0
, z
1
]) = ([az
0
+bz
1
, cz
0
+dz
1
])
where ad bc = 0. This corresponds to the invertible linear transformation
T =
_
a b
c d
_
.
9
It is convenient to write P
1
= C {} where the point is now the 1-dimensional
space z
1
= 0. Then if z
1
= 0, [z
0
, z
1
] = [z, 1] and
([z, 1]) = [az +b, cz +d]
and if cz +d = 0 we can write
([z, 1]) =
_
az +b
cz +d
, 1
_
which is the usual form of a Mobius transformation, i.e.
z
az +b
cz +d
.
The advantage of projective geometry is that the point = [1, 0] plays no special
role. If cz +d = 0 we can still write
([z, 1]) = [az +b, cz +d] = [az +b, 0] = [1, 0]
and if z = (i.e. [z
0
, z
1
] = [1, 0]) then we have
([1, 0]) = [a, c].
Example: If we view the projective plane P
2
in the same way, we get some less
familiar transformations. Write
P
2
= C
2
P
1
where the projective line at innity is x
0
= 0. A linear transformation T : C
3
C
3
can then be written as the matrix
T =
_
_
d b
1
b
2
c
1
a
11
a
12
c
2
a
21
a
22
_
_
and its action on [1, x, y] can be expressed, with v = (x, y) C
2
, as
v
1
b v +d
(Av +c)
where A is the 2 2 matrix a
ij
and b, c the vectors (b
1
, b
2
), (c
2
, c
2
). These are the
2-dimensional versions of Mobius transformations. Each one can be considered as a
composition of
10
an invertible linear transformation v Av
a translation v v +c
an inversion v v/(b v +d)
Clearly it is easier here to consider projective transformations dened by 3 3 ma-
trices, just ordinary linear algebra.
However, we can see from this viewpoint that the projective transformations that
take C
2
to itself are those with b
1
= b
2
= 0 and then the action on C
2
is
v d
1
(Av +c)
a combination of linear transformation and translation. These are known as ane
transformations, so in a natural way the complement of a line in P
2
is an ane space
( a vector space without a distinguished origin.) The subsets U
i
are called ane
subsets of P
2
.
Example: A more geometric example of a projective transformation is to take two
lines P(U), P(U

) in a projective plane P(V ) and let O P(V ) be a point disjoint

from both. For each point x P(U), the unique line joining O to x intersects P(U

)
in a unique point X = (x). Then
: P(U) P(U

)
is a projective transformation.
To see why, let W be the 1-dimensional subspace of V dened by O P(V ). Then
since O does not lie in P(U

), W U

= 0. This means that

V = W U

.
Now take a U as a representative vector for x. It can be expressed uniquely as
a = w + a

, with w W and a

. The projective line joining O to x is dened

by the 2-dimensional vector subspace of V spanned by w and a and so a

= a w is
a representative vector for (x). In linear algebra terms the map a a

is just the
linear projection map P : V U

restricted to U. It has zero kernel since O does

not lie in P(U), and hence W U = 0. Thus T : U U

is an isomorphism and is
a projective transformation.
If we restrict to the points in R
2
, then this is what this projection from O looks like:
11
O
A linear transformation of a vector space of dimension n is determined by its value on
n linearly independent vectors. A similar statement holds in projective space. The
analogue of linear independence is the following
Denition 5 Let P(V ) be an n-dimensional projective space, then n + 2 points in
P(V ) are said to be in general position if each subset of n+1 points has representative
vectors in V which are linearly independent.
Example: Any two distinct points in a projective line are represented by linearly
independent vectors, so any three distinct points are in general position.
Theorem 3 If X
1
, . . . , X
n+2
are in general position in P(V ) and Y
1
, . . . , Y
n+2
are
in general position in P(W), then there is a unique projective transformation :
P(V ) P(W) such that (X
i
) = Y
i
, 1 i n + 2.
Proof: First choose representative vectors v
1
, . . . , v
n+2
V for the points X
1
, . . . , X
n+2
in P(V ). By general position the rst n +1 vectors are linearly independent, so they
form a basis for V and there are scalars
i
such that
v
n+2
=
n+1

i=1

i
v
i
(1)
If
i
= 0 for some i, then (1) provides a linear relation amongst a subset of n + 1
vectors, which is not possible by the denition of general position, so we deduce that
12

i
= 0 for all i. This means that each
i
v
i
is also a representative vector for X
i
, so
(1) tells us that we could have chosen representative vectors v
i
such that
v
n+2
=
n+1

i=1
v
i
(2)
Moreover, given v
n+2
, these v
i
are unique for
n+1

i=1
v
i
=
n+1

i=1

i
v
i
implies
i
= 1 since v
1
, . . . , v
n+1
are linearly independent.
[Note: This is a very useful idea which can simplify the solution of many problems].
Now do the same for the points Y
1
, . . . Y
n+2
in P(W) and choose representative vectors
such that
w
n+2
=
n+1

i=1
w
i
(3)
Since v
1
, . . . , v
n+1
are linearly independent, they form a basis for V so there is a
unique linear transformation T : V W such that Tv
i
= w
i
for 1 i n+1. Since
w
1
, . . . , w
n+1
are linearly independent, T is invertible. Furthermore, from (2) and (3)
Tv
n+2
=
n+1

i=1
Tv
i
=
n+1

i=1
w
i
= w
n+2
and so T denes a projective transformation such that (X
i
) = Y
i
for all n + 2
vectors v
i
.
To show uniqueness, suppose T

denes another projective transformation

with the
same property. Then T

v
i
=
i
w
i
and

n+2
w
n+2
= T

v
n+2
=
n+1

i=1
T

v
i
=
n+1

i=1

i
w
i
.
But by the uniqueness of the representation (3), we must have
i
/
n+2
= 1, so that
T

v
i
=
n+2
Tv
i
and

= . 2
Examples:
13
1. In P
1
take the three distinct points [0, 1], [1, 1], [1, 0] and any other three distinct
points X
1
, X
2
, X
3
. Then there is a unique projective transformation taking X
1
, X
2
, X
3
to [0, 1], [1, 1], [1, 0]. In the language of complex analysis, we can say that there is a
unique Mobius transformation taking any three distinct points to 0, 1, .
2. In any projective line we could take the three points [0, 1], [1, 1], [1, 0] and then
for X
1
, X
2
, X
3
any permutation of these. Now projective transformations of a space
to itself form a group under composition, so we see that the group of projective
transformations of a line to itself always contains a copy of the symmetric group S
3
.
In fact if we take the scalars to be the eld Z
2
with two elements 0 and 1, the only
points on the projective line are [0, 1], [1, 1], [1, 0], and S
3
is the full group of projective
transformations.
As an example of the use of the notion of general position, here is a classical theorem
called Desargues theorem. In fact, Desargues (1591-1661) is generally regarded as the
founder of projective geometry. The proof we give here uses the method of choosing
representative vectors above.
Theorem 4 (Desargues) Let A, B, C, A

, B

, C

be distinct points in a projective space

P(V ) such that the lines AA

, BB

CC

are distinct and concurrent. Then the three

points of intersection AB A

, BC B

, CA C

are collinear.
Proof: Let P be the common point of intersection of the three lines AA

, BB

, CC

.
Since P, A, A

lie on a projective line and are distinct, they are in general position, so
as in (2) we choose representative vectors p, a, a

such that
p = a +a

.
These are vectors in a 2-dimensional subspace of V . Similarly we have representative
vectors b, b

for B, B

and c, c

for C, C

with
p = b +b

p = c +c

.
It follows that a +a

= b +b

and so
a b = b

= c

and similarly
b c = c

= a

c a = a

= b

.
14
But then
c

+a

+b

= a b +b c +c a = 0
and a

, b

, c

are linearly dependent and lie in a 2-dimensional subspace of V . Hence

the points A

, B

, C

in P(V ) represented by a

, b

, c

are collinear.
Now since c

= a b, c

lies in the 2-dimensional space spanned by a and b, so C

lies on the line AB. Since c

also equals b

, C

lies on the line A

and so c

represents the point AB A

. Repeating for B

and A

we see that these are the

three required collinear points. 2
Desargues theorem is a theorem in projective space which we just proved by linear
algebra linear independence of vectors. However, if we take the projective space
P(V ) to be the real projective plane P
2
(R) and then just look at that part of the
data which lives in R
2
, we get a theorem about perspective triangles in the plane:
Here is an example of the use of projective geometry a higher form of geometry
to prove simply a theorem in R
2
which is less accessible by other means. Another
theorem in the plane for which these methods give a simple proof is Pappus theorem.
Pappus of Alexandria (290-350) was thinking again of plane Euclidean geometry, but
his theorem makes sense in the projective plane since it only discusses collinearity
and not questions about angles and lengths. It means that we can transform the
given conguration by a projective transformation to a form which reduces the proof
to simple linear algebra calculation:
Theorem 5 (Pappus) Let A, B, C and A

, B

, C

be two pairs of collinear triples of

distinct points in a projective plane. Then the three points BC

C, CA

A, AB

B are collinear.
15
Proof: Without loss of generality, we can assume that A, B, C

, B

are in general
position. If not, then two of the three required points coincide, so the conclusion is
trivial. By Theorem 3, we can then assume that
A = [1, 0, 0], B = [0, 1, 0], C

= [0, 0, 1], B

= [1, 1, 1].
The line AB is dened by the 2-dimensional subspace {(x
0
, x
1
, x
2
) F
3
: x
2
= 0}, so
the point C, which lies on this line, is of the form C = [1, c, 0] and c = 0 since A = C.
Similarly the line B

is x
0
= x
1
, so A

= [1, 1, a] with a = 1.
The line BC

is dened by x
0
= 0 and B

C is dened by the span of (1, 1, 1) and

(1, c, 0), so the point BC

C is represented by the linear combination of (1, 1, 1)

and (1, c, 0) for which x
0
= 0, i.e.
(1, 1, 1) (1, c, 0) = (0, 1 c, 1).
The line C

A is given by x
1
= 0, so similarly CA

A is represented by
(1, c, 0) c(1, 1, a) = (1 c, 0, ca).
Finally AB

is given by x
1
= x
2
, so AB

B is
(1, 1, a) + (a 1)(0, 1, 0) = (1, a, a).
But then
(c 1)(1, a, a) + (1 c, 0, ca) +a(0, 1 c, 1) = 0.
Thus the three vectors span a 2-dimensional subspace and so the three points lie on
a projective line. 2
16
2 Plane curves
2.1 Basic denitions
From now on we will mainly work in the projective plane P
2
and use homogeneous
coordinates [x, y, z] instead of [x
0
, x
1
, x
2
]. The equation of a line is
ax +by +cz = 0
and this gives a well-dened subset of P
2
because if we replace (x, y, z) by (x, y, z)
the equation still holds.
Denition 6 A polynomial P(x, y, z) is homogeneous of degree d if
P(x, y, z) =
d
P(x, y, z).
Clearly P(x, y, z) = 0 is a well-dened subset of P
2
.
Denition 7 Let P(x, y, z) be a homogeneous polynomial of degree d > 0 with no
repeated factors, then P(x, y, z) = 0 denes a plane projective curve C in P
2
.
Remark: 1. The subset P(x, y, z) = 0 in P
2
is non-empty because xing y and z
we have a polynomial in x which over C always has roots. This is why we work over
the complex numbers P(x, y, z) = x
2
+ y
2
+ z
2
denes an algebraic curve over R
but it has no real points.
2. The condition of having no repeated factors is to ensure that the polynomial is
uniquely determined (up to a scalar multiple) by the curve C obviously P(x, y, z)
and P
2
(x, y, z) dene the same subset. The fact that with this condition the zeros
determine P follows from Hilberts Nullstellensatz, which is outside the scope of this
course.
Denition 8 The curve C P
2
is said to be irreducible if P has no non-constant
factors other than a scalar multiple of itself. An irreducible plane curve D is said to
be a component of C if its dening polynomial Q divides P.
Remark: Note that this use of the word component is dierent from the topo-
logical one two lines given by linear polynomials P, Q have non-empty intersection
so the reducible curve C dened by PQ = 0 is a connected topological space.
17
Denition 9 The point [a, b, c] P
2
is a singular point of C if P(a, b, c) = 0 and
P
x
(a, b, c) =
P
y
(a, b, c) =
P
z
(a, b, c) = 0.
Remark: For a homogeneous function P, dierentiating P(x, y, z) =
d
P(x, y, z)
with respect to leads via the chain rule directly to Eulers relation:
x
P
x
+y
P
y
+z
P
z
= dP
so that the vanishing of the partial derivatives actually implies the vanishing of P.
Example: 1. The reducible curve C dened by a pair of lines has a unique singular
point the point of intersection since
PQ
x
= P
Q
x
+Q
P
x
and this, and the other partial derivatives, vanish when P = Q = 0.
2. The curve x
2
+ y
2
+ z
2
= 0 has no singularities since the vanishing of the partial
derivatives gives x = y = z = 0 but this does not dene a point in projective space.
If [a, b, c] is not a singular point of C then at least one partial derivative is non-
vanishing. Then we dene:
Denition 10 If p = [a, b, c] is a nonsingular point of C, the tangent line at p is
dened by the equation
x
P
x
(a, b, c) +y
P
y
(a, b, c) +z
P
z
(a, b, c) = 0.
At this point it may be useful to view projective curves in terms of their intersections
with one of the ane open sets U
i
, say U
2
where z = 0. Then identifying U
2
with C
2
via (x/z, y/z) the curve C U
2
is dened by the nonhomogeneous equation
P(x, y, 1) = 0.
Conversely, the subset of C
2
dened by a polynomial p(x, y) = 0 can be extended to
a homogeneous polynomial by
P(x, y, z) = z
d
p(
x
z
,
y
z
)
18
so long as d m + n where x
m
y
n
is the highest power in p. If the line z = 0 is not
a component of C then this gives a one-to-one correspondence, but the extra points
on C {z = 0} have to be analyzed by using one of the other open sets U
0
, U
1
.
Our denition of tangent line makes more sense in this ane open set, where we
might dene the normal direction to p(x, y) = 0 at (a, b) by (p/x, p/y)) and
the tangent would then be the line
(x a)
p
x
(a, b) + (y b)
p
y
(a, b) = 0.
It is easy to check that this line is the intersection with U
2
of the tangent line according
to Denition 10.
Example: The parabola y
2
= 4x in R
2
extends to the curve C of degree 2 in P(R
3
)
dened by the homogeneous polynomial
P(x, y, z) = y
2
4xz.
We can rewrite this as
y
2
(x +z)
2
+ (x z)
2
so its intersection with U
1
, where y = 0 is the curve
(x +z)
2
(x z)
2
= 1
which in coordinates (x+z, xz) is a hyperbola. From the point of view of projective
geometry the curve C intersects the line z = 0 in the single real point [1, 0, 0], and
the line y = 0 in two points [1, 0, 0], [0, 0, 1]. In the ane set x + z = 0, we have
coordinates (y/(x +z), (x z)/(x +z)) and then
y
2
(x +z)
2
+
(x z)
2
(x +z)
2
= 1
which is the equation of a circle. So in projective space all these curves become one
and the same type.
2.2 Conics
A conic is a plane projective curve of degree 2. It is dened by a homogeneous
quadratic form P(x, y, z). This is algebraically the same thing as:
19
Denition 11 A symmetric bilinear form on a vector space V is a map B : V V
F such that
B(v, w) = B(w, v)
B(
1
v
1
+
2
v
2
, w) =
1
B(v
1
, w) +
2
B(v
2
, w)
The form is said to be nondegenerate if B(v, w) = 0 for all w V implies v = 0.
If we take a basis v
1
, . . . , v
n
of V , then v =

i
x
i
v
i
and w =

i
y
i
v
i
so that
B(v, w) =

i,j
B(v
i
, v
j
)x
i
y
j
and so is uniquely determined by the symmetric matrix
ij
= B(v
i
, v
j
). The bilinear
form is nondegenerate if and only if
ij
is nonsingular.
We can add symmetric bilinear forms: (B+C)(v, w) = B(v, w)+C(v, w) and multiply
by a scalar (B)(v, w) = B(v, w) so they form a vector space isomorphic to the space
of symmetric n n matrices which has dimension n(n + 1)/2. If we take a dierent
basis
w
i
=

j
P
ji
v
j
then
B(w
i
, w
j
) = B(

k
P
ki
v
k
,

P
j
v

) =

k,
P
ki
B(v
k
, v

)P
j
so that the matrix
ij
= B(v
i
, v
j
) changes under a change of basis to

= P
T
P.
Over the real or complex numbers we can divide by 2 and then we often speak of the
quadratic form B(v, v) which determines the bilinear form since
B(v +w, v +w) = B(v, v) +B(w, w) + 2B(v, w)
Here we have the basic result:
Theorem 6 Let B be a quadratic form on a complex vector space V of dimension n.
Then there is a basis such that if v =

i
z
i
v
i
B(v, v) =
m

i=1
z
2
i
.
If B is non-degenerate then m = n.
20
Proof: The proof is elementary just completing the square. We note that changing
the basis is equivalent to changing the coecients x
i
of v by an invertible linear
transformation.
First we write down the form in one basis, so that
B(v, v) =

i,j

ij
x
i
x
j
and ask: is there a term
ii
= 0?. If not, then we create one. If the coecient of x
i
x
j
is non-zero, then putting y
i
= (x
i
+x
j
)/2, y
j
= (x
i
x
j
)/2 we have
x
i
x
j
= y
2
i
y
2
j
and so we get a term

ii
= 0.
If there is a term
ii
= 0, then we note that
1

ii
(
i1
x
1
+. . . +
in
x
n
)
2
=
ii
x
2
i
+ 2

k=i

ik
x
k
x
i
+R
where R involves the x
k
with k = i. So if
y
i
=
i1
x
1
+. . . +
in
x
n
then
B(v, v) =
1

ii
y
2
i
+B
1
where B
1
is a quadratic form in the n 1 variables x
k
, k = i.
We now repeat the procedure to nd a basis such that if v has coecients y
1
, . . . , y
n
,
then
B(v, v) =
m

i=1
c
i
y
2
i
.
Over C we can write z
i
=

c
i
y
i
and get a sum of squares. 2
Example: Consider the quadratic form in C
3
B(v, v) = xy +yz +zx.
We put
y
1
= (x +y)/2, y
2
= (x y)/2
21
to get
B(v, v) = y
2
1
y
2
2
+z(2y
1
).
Now complete the square:
B(v, v) = (y
1
+z)
2
y
2
2
z
2
so that with z
1
= y
1
+z, z
2
= iy
2
, z
3
= iz we have z
2
1
+z
2
2
+z
2
3
.
An invertible linear transformation of C
3
denes a projective transformation of P
2
so
Theorem 6 tells us that any conic is equivalent to one of the following by a projective
transformation:
P(x, y, z) = x
2
this is the (double) line x = 0
P(x, y, z) = x
2
+y
2
this is a pair of lines x +iy = 0, x iy = 0
P(x, y, z) = x
2
+y
2
+z
2
a nonsingular conic.
2.3 Rational parametrization of the conic
Theorem 7 Let C be a nonsingular conic in a projective plane P(V ) over the eld
F, and let A be a point on C. Let P(U) P(V ) be a projective line not containing
A. Then there is a bijection
: P(U) C
such that, for X P(U),the points A, X, (X) are collinear.
X
A
(X)
22
Proof: Suppose the conic is dened by the nondegenerate symmetric bilinear form
B. Let a V be a representative vector for A, then B(a, a) = 0 since A lies on the
conic. Let x P(U) be a representative vector for X P(U). Then a and x are
linearly independent since X does not lie on the line P(U). Extend a, x to a basis
a, x, y of V .
Now B restricted to the space spanned by a, x is not identically zero, because if it
were, the matrix of B with respect to this basis would be of the form
_
_
0 0
0 0

_
_
which is singular. So at least one of B(x, x) and B(a, x) is non-zero.
Any point on the line AX is represented by a vector of the form a+x and this lies
on the conic C if
0 = B(a +x, a +x) = 2B(a, x) +
2
B(x, x).
When = 0 we get the point X. The other solution is 2B(a, x) +B(x, x) = 0 i.e.
the point with representative vector
w = B(x, x)a 2B(a, x)x (4)
which is non-zero since the coecients are not both zero.
We dene the map : P(U) C by
(X) = [w]
which has the collinearity property of the statement of the Theorem. If Y C is
distinct from A, then the line AY meets the line P(U) in a unique point, so
1
is
well-dened on this subset. By the denition of in (4), (X) = A if and only if
B(a, x) = 0. Since B is nonsingular f(x) = B(a, x) is a non-zero linear map from V
to F and so denes a line (the tangent to C at A), which hence meets P(U) in one
point. Thus has a well-dened inverse and is therefore a bijection. 2
Example: Consider the case of the conic
x
2
+y
2
z
2
= 0.
Take A = [1, 0, 1] and the line P(U) dened by x = 0. Note that this conic and the
point and line are dened over any eld since the coecients are 0 or 1.
23
A point X P(U) is of the form X = [0, 1, t] or [0, 0, 1] and the map is
([0, 1, t]) = [B((0, 1, t), (0, 1, t))(1, 0, 1) 2B((1, 0, 1), (0, 1, t))(0, 1, t)]
= [1 t
2
, 2t, 1 +t
2
]
or ([0, 0, 1]) = [1, 0, 1].
This has an interesting application if we use the eld of rational numbers F = Q.
Suppose we want to nd all right-angled triangles whose sides are of integer length.
By Pythagoras, we want to nd positive integer solutions to
x
2
+y
2
= z
2
.
But then [x, y, z] is a point on the conic. Conversely, if [x
0
, x
1
, x
2
] lies on the conic,
then multiplying by the least common multiple of the denominators of the rational
numbers x
0
, x
1
, x
2
gives integers such that [x, y, z] is on the conic.
But what we have seen is that any point on the conic is either [1, 0, 1] or of the form
[x, y, z] = [1 t
2
, 2t, 1 +t
2
]
for some rational number t = p/q, so we get all integer solutions by putting
x = q
2
p
2
, y = 2pq, z = q
2
+p
2
.
For example, p = 1, q = 2 gives 3
2
+ 4
2
= 5
2
and p = 2, q = 3 gives 5
2
+ 12
2
= 13
2
.
One other consequence of Theorem 7 is that we can express a point (x, y) on the
general conic
ax
2
+bxy +cy
2
+dx +ey +f = 0
in the form
x =
p(t)
r(t)
, y =
q(t)
r(t)
where p, q and r are quadratic polynomials in t. Writing x, y as rational functions of t
is why the process we have described is sometimes called the rational parametrization
of the conic. It has its uses in integration. We can see, for example, that
_
dx
x +

ax
2
+bx +c
24
can be solved by elementary functions because if y = x +

ax
2
+bx +c then
(y x)
2
ax
2
bx c = 0
and this is the equation of a conic. We can solve it by x = p(t)/r(t), y = q(t)/r(t)
and with this substitution, the integral becomes
_
r

(t)p(t) p

(t)r(t)
q(t)r(t)
dt
and expanding the rational integrand into partial fractions we get rational and loga-
rithmic terms after integration.
25
3 Intersections of curves
3.1 Resultants
We can gain quite a lot of information about curves from Bezouts Theorem which
says roughly that curves C and D of degrees n and m respectively intersect in mn
points. This is clearly not true in general (for example a line which is tangent to a
nonsingular conic meets it in one point, not two). This means that we have to do
some work to dene carefully the intersection multiplicity and the conditions under
which the theorem holds.
The basic idea is to write
P(x, y, z) = a
0
(y, z) +a
1
(y, z)x +. . . +a
n
(y, z)x
n
Q(x, y, z) = b
0
(y, z) +b
1
(y, z)x +. . . +b
m
(y, z)x
m
and forget for the moment the dependence of the coecients on y, z. The condition
that two polynomials in x have a common root gives a polynomial relation on the
a
i
, b
j
and, putting y and z back in, we can nd the number of solutions.
Let
p(x) = a
0
+a
1
x +. . . +a
n
x
n
q(x) = b
0
+b
1
x +. . . +b
m
x
m
be two complex polynomials in x with a
n
= 0. If they have a common factor then
p = r, q = s for polynomials r, s of degree n 1, m1 respectively, and so
sp rq = 0.
This is a linear equation for the n coecients of r and m coecients of s. But a
polynomial of the form sp rq has degree m+n1 and so has m+n coecients,
hence sp rq = 0 is an equation of the form Av = 0 for a vector v C
m+n
and
has a solution if and only if det A = 0. If there is a solution then sp = rq and since
deg r < deg p clearly one of the factors of p must then be a factor of q. The matrix A
appears in the following denition.
26
Denition 12 The resultant R(p, q) of p(x) and q(x) is the (m + n) (m + n)
determinant
det
_
_
_
_
_
_
_
_
_
_
_
_
a
0
a
1
. . . a
n
0 0 . . . 0
0 a
0
a
1
. . . a
n
0 . . . 0
. . . . . . . . . . . . . . . . . . . . . . . .
0 0 . . . 0 a
0
a
1
. . . a
n
b
0
b
1
. . . b
m
0 0 . . . 0
0 b
0
b
1
. . . b
m
0 . . . 0
. . . . . . . . . . . . . . . . . . . . . . . .
0 0 . . . 0 b
0
b
1
. . . b
m
_
_
_
_
_
_
_
_
_
_
_
_
.
Example: A simple example is the condition that a polynomial p has a repeated
root . Then if p(x) = (x)
2
r(x), its derivative is p

(x) = 2(x)r(x)+(x)
2
r

(x)
so p and p

have a common root hence R(p, p

) = 0, which is the discriminant. For

example if p(x) = ax
2
+bx +c, then
R(p, p

) = det
_
_
c b a
b 2a 0
0 b 2a
_
_
= a(4ac b
2
)
which, assuming a = 0, vanishes if and only if b
2
4ac = 0.
The key feature to notice about the resultant is that it is a polynomial in the coe-
cients of p and q which is of degree m in the a
i
and degree n in the b
i
.
Determinants are often of more theoretical than practical value and there is another
way of writing the resultant. Suppose p, q are monic polynomials and let
1
, . . . ,
n
be the roots of p(x) and
1
, . . . ,
m
the roots of q(x). Then

i,j
(
i

j
) =

i
q(
i
) = (1)
mn

j
p(
j
) (5)
clearly vanishes if p and q have a common root. For p = x
2
+bx +c, p

= 2x +b this
gives immediately c b
2
/4.
To see the link with the resultant, note that a
k
is a homogeneous polynomial in
i
of degree n k and b
k
a polynomial in
j
of degree m k. The (i, j) entry in the
determinant is therefore of degree d
ij
= n + i j if 1 i m and degree i j if
m+ 1 i n +m. The determinant of a k k matrix A is a sum

S
k
sgn A
1(1)
A
2(2)
. . . A
k(k)
27
so in our case each sum is of degree
m

1
(n +i (i)) +
m+n

m+1
(i (i)) = mn +
m+n

1
(i (i)) = mn.
So R(p, q) is a homogeneous polynomial in the m+n variables (
1
, . . .
m
) of degree
mn which vanishes precisely when
i
=
j
so (5) must be a scalar multiple of R(p, q),
in fact in general
R(p, q) = a
m
n
b
n
m

i,j
(
i

j
).
With this description it is easy to see that
R(p, qr) = R(p, q)R(p, r) (6)
Now replace p(x), q(x) by P(x, y, z), Q(x, y, z). The resultant is now a polynomial in
y and z and we write it as R
P,Q
(y, z). Since in
P(x, y, z) = a
0
(y, z) +a
1
(y, z)x +. . . +a
n
(y, z)x
n
the coecient a
k
(y, z) is a homogeneous polynomial of degree n k, it follows as we
have just seen that R
P,Q
(y, z) is a homogeneous polynomial of degree mn.
3.2 Applications
Theorem 8 Any two algebraic curves in P
2
intersect in at least one point.
Proof: The coecients a
n
, b
m
are constants which vanish if and only if [1, 0, 0] is a
point on both curves. Without loss of generality assume then that a
n
= 0, then the
resultant R
P,Q
(y, z) is a homogeneous polynomial of degree mn so
R
P,Q
(y, z) = cz
mn
nm

i
(
y
z

i
) = c

y
mn
nm

i
(
z
y

i
)
so there is always a non-zero value (y, z) = (b, c) such that R
P,Q
(b, c) = 0. Then
P(x, b, c) and Q(x, b, c) have a common zero x = a and [a, b, c] P
2
lies in both
curves. 2
Theorem 9 Two algebraic curves C, D in P
2
of degrees n, m respectively intersect in
at most nm points, if they have no common component.
28
Proof: Suppose a set S of nm + 1 distinct points lies in C D. By a projective
transformation assume that [1, 0, 0] is not one of these, nor does it lie on a line joining
any pair of them. Then in particular P(1, 0, 0) = 0 = Q(1, 0, 0) and we can use the
resultant: R
P,Q
(y, z) is a product of linear factors bz cy with (b, c) = 0. Since
R
P,Q
(b, c) = 0 there is an a such that P(a, b, c) = 0 = Q(a, b, c). Conversely if
[a, b, c] S then P(a, b, c) = 0 = Q(a, b, c) and (b, c) = 0 since [1, 0, 0] is not in the
intersection, so bz cy is factor of R
P,Q
(y, z).
The points [a, b, c] and [a

, b, c] cannot both be in S unless a

= a for then [1, 0, 0] =

[a

a, b b, c c] lies on the line joining them, so (b, c) up to a scalar multiple

determines each of the nm+1 points of intersection. But if R
P,Q
(y, z) is not identically
zero, it has at most nm linear factors, which is a contradiction.
If R
P,Q
(y, z) 0 then from the properties of the resultant, P and Q have a common
factor as polynomials in x over the eld of rational functions p(y, z)/q(y, z). But
by Gauss lemma, they have a common polynomial factor, which denes a common
component of the curves C and D. 2
Theorem 10 (i) A nonsingular algebraic curve in P
2
is irreducible.
(ii) An irreducible curve in P
2
has at most a nite number of singular points.
Proof: (i) Let the reducible curve be dened by P(x, y, z)Q(x, y, z) = 0. By
Theorem 8 there is a point [a, b, c] where P(a, b, c) = 0 = Q(a, b, c) and dierentiating
the product PQ at (a, b, c) gives zero.
(ii) Without loss of generality assume [1, 0, 0] does not lie on the curve C, so that the
coecient of x
n
in P(x, y, z) is non-zero. Then P/x is a non-zero homogeneous
polynomial of degree n 1. C is irreducible and P/x has lower degree so there
is no common component, hence by Theorem 9 there are at most n(n 1) points of
intersection. The singular points lie amongst these. 2
Theorem 11 (Pascals mystic hexagram) The pairs of opposite sides of a hexagon
inscribed in an irreducible conic meet in three collinear points.
Proof: Let the successive sides of the hexagon be dened by linear polynomials
L
1
, . . . L
6
, and put P = L
1
L
3
L
5
dening the degree 3 curve C and Q = L
2
L
4
L
6
dening the curve D, then the six vertices of the hexagon lie in C D and also on
the conic.
Let [a, b, c] lie on the conic but not on C D then
Q(a, b, c)P(x, y, z) P(a, b, c)Q(x, y, z)
29
denes a degree 3 curve which meets the conic in the six vertices plus the point [a, b, c],
i.e. in 7 = 2 3 + 1 points. By Theorem 9, since the conic is irreducible we must
have
Q(a, b, c)P(x, y, z) P(a, b, c)Q(x, y, z) = L(x, y, z)R(x, y, z)
where R = 0 is the equation of the conic. Hence the degree one factor L denes a
line which passes through the other 9 6 points of intersection of C D. 2
3.3 Intersection multiplicity
To count properly we need a denition of the multiplicity of the intersection of two
curves. For example if we think of a tangent as the limit of a chord where the two
points of intersection coalesce into one, then we should count that multiplicity as two.
It may also be the case that further points coalesce at the same time, and we need
to be able to count these. Or if we intersect two lines L
1
, L
2
with a third L, then the
two points of intersection with L
1
and L
2
become one if L passes through the singular
point L
1
L
2
.
We use the resultant to make the following (technical) denition:
Denition 13 The intersection multiplicity I
p
(C, D) of curves C and D at p is de-
ned by the following prescription:
I
p
(C, D) = if p lies on a common component of C and D.
30
I
p
(C, D) = 0 if p does not lie on C D
if p CD, but lies in no common component, remove any common components
to get curves C

, D

and choose projective coordinates so that [1, 0, 0] does not lie

on C

, nor on any line joining distinct points of C

, nor on any tangent

line to C

or D

at a point of C

. Then dene I
p
(C, D) at p = [a, b, c] to be
the largest integer k such that (bz cy)
k
divides the resultant R
P

,Q
(y, z).
With this denition, we immediately see:
Proposition 12 Suppose C is nonsingular at p, and let T
p
be the tangent line at p.
Then I
p
(C, T
p
) > 1.
Proof: With the hypotheses of the theorem, P/x = 0 at [a, b, c], so (writing the
partial derivatives as P
x
etc.) the resultant, obtained by substituting from the linear
equation of the tangent, is a multiple of
P(P
x
(p)
1
(P
y
(p)y +P
z
(p)z), y, z).
The multiplicity is greater than one if this has a repeated factor, which is where the
derivative with respect to y (or z) vanishes. But using the chain rule, dierentiating
with respect to y at p gives
(P
1
x
P
y
P
x
+P
y
)(p) = 0.
2
Proposition 13 If p C D is a singular point of C, then I
p
(C, D) > 1.
Proof: We can choose coordinates such that p = [0, 0, 1], so the partial derivatives
of P vanish here, hence when we write
P(x, y, z) = a
0
(y, z) +a
1
(y, z)x +. . . +a
n
(y, z)x
n
the coecient a
0
(y, z) is divisible by y
2
and a
1
(y, z) by y.
Since Q(0, 0, 1) = 0, b
0
(y, z) is divisible by y and so
b
0
(y, z) = b
01
yz
m1
+y
2
c
0
(y, z) b
1
(y, z) = b
10
z
m1
+yc
1
(y, z).
If b
01
= 0 then the rst column of the resultant is divisible by y
2
; if b
01
= 0 then only
y divides the rst column. But taking that factor out and subtracting b
10
/b
01
times
the rst from the second column gives another factor y. So y
2
divides the resultant
and the intersection multiplicity is bigger than one. 2
31
The resultant is a fundamentally useful tool but it has the disadvantage that we have
to select x from (x, y, z) to dene it, and it is not clear that the denition of I
p
(C, D)
is independent of this choice, so it is useful to know that the following properties,
which are clearly independent of choice, actually characterize uniquely I
p
(C, D) (see
Kirwans book for details).
Theorem 14 The intersection multiplicity satises the following properties:
(i) I
p
(C, D) = I
p
(D, C)
(ii) I
p
(C, D) = if p lies on a common component of C and D and is otherwise a
nonnegative integer.
(iii) I
p
(C, D) = 0 if and only if p does not lie in C D.
(iv) Two distinct lines meet at a point with multiplicity one.
(v) If the curve C is the union of components C
1
, C
2
then
I
p
(C, D) = I
p
(C
1
, D) +I
p
(C
2
, D).
(vi) If C and D are dened by P and Q and E is dened by PR +Q, then
I
p
(C, D) = I
p
(C, E).
Proof:
(i) This is clear since R
P,Q
= R
Q,P
from (5).
(ii) This was part of the denition.
(iii) If p does lie in the intersection then the resultant has at least a linear factor, so
I
p
(C, D) 1.
(iv) If the lines meet at p = [0, 0, 1] they are given by ax +by = 0, cx +dy = 0 so the
resultant is (ad bc)y.
(v) This follows from R
P,QR
= R
P,Q
R
P,R
.
(vi) If
R(x, y, z) =
0
(y, z) +
1
(y, z)x +. . . +
nm
(y, z)x
nm
then the resultant R
P,PR+Q
is the determinant of B
ij
where B
ij
is equal to the matrix
A
ij
of R
P,Q
for i m and to
B
ij
+
in

k=im

ink
A
kj
32
for i > m. But this is obtaned from A
ij
by row operations so the determinant is
unchanged. 2
3.4 Cubic curves
Using the notion of multiplicity, we show here how to reduce a nonsingular cubic
curve dened by a homogeneous polynomial P(x, y, z) of degree 3 to a standard
form. The vector space of such polynomials is 10-dimensional so there are 9 free
parameters to specify the cubic curve. A projective transformation is determined by
what it does to four points in general position in the projective plane, which gives
8 parameters, so, unlike the conics, we cant expect to reduce each cubic to a single
equation. There is one remaining degree of freedom, and we shall prove that
Theorem 15 After a projective transformation, the equation for any nonsingular
cubic curve can be put in the form
y
2
z = x(x z)(x z)
where = 0, 1.
To do this we consider inection points, or exes.
Denition 14 A nonsingular point p C is called an inection point if there is a
line L through p with I
p
(C, L) 3. The line is necessarily the tangent to C at p.
Remark: In calculus inection points on the graph y = f(x) are dened to be points
where the second derivative f

(x) vanishes. To see that this is the same we have to

stray from polynomials into holomorphic functions but we shall do this later anyway
when we consider algebraic curves as Riemann surfaces. Suppose that z = 0 then the
curve is given by the equation P(x, y, 1) = 0 in ane coordinates. If it is nonsingular
one of P/x, P/y is non-zero. Suppose it is the latter, then at (x, y) = (a, b) on
the curve, the implicit function theorem tells us that there are neighbourhoods V and
W of a and b in C and a holomorphic function g : V W such that for x V and
y W P(x, y, 1) = 0 if and only if y = g(x).
The tangent line at (a, b) is (x, y) = (a, b) + t(1, g

(a)) so (a, b) is an inection point

if and only if t
3
divides P(a + t, b + tg

(a), 1). Now P(x, g(x), 1) 0, and g(a + t) =

b +tg

(a) +t
2
g

(a)/2 +t
3
h(t). Moreover
P(x, y, 1) =

k,
a
k
x
k
y

33
so

k,
a
k
(a +t)
k
(b +tg

(a) +t
2
g

(a)/2 +t
3
h(t))

0.
But then, expanding the power of ,

k,
a
k
(a +t)
k
(b +tg

(a))

+t
2
g

(a)/2

k,
a
k
(a +t)
k
(b +tg

(a))
1
+t
3
k(t) 0.
Since, near (x, y) = (a, b),
0 =
P
y
=

k,
a
k
x
k
y
1
it follows that t
3
divides the rst term, which is P(a + t, b + tg

(a), 1), only where

g

(a) = 0.
In what follows it is convenient to revert to the notation of (x
0
, x
1
, x
2
) for (x, y, z)
and to set P
i
to be the rst derivatives and P
ij
to be the matrix of second partial
derivatives:
P
ij
=
P
x
i
x
j
.
Proposition 16 A nonsingular point p C is an inection point if and only if
det P
ij
= 0.
Proof: To nd the multiplicity of p with respect to a line we have to take the
resultant of P(x, y, z) and ax+by +cz. But this is just substituting x = (by +cz)/a
into P. It is more convenient to retain the symmetry and consider the line as the set
of points
[a
0
+t
0
, a
1
+t
1
, a
2
+t
2
]
as t varies. Then I
p
(C, L) 3 if and only if t
3
divides
P(a
0
+t
0
, a
1
+t
1
, a
2
+t
2
).
Expanding this, we have
P(a +t) = P(a) +t

i
P
i
(a)
i
+
t
2
2

i,j
P
ij
(a)
i

j
+t
3
R (7)
34
and t
3
divides this if and only if

i
P
i
(a)
i
= 0 =

i,j
P
ij
(a)
i

j
.
The rst equation says that the line is a tangent.
We now use homogeneity the P
i
are homogeneous of degree n 1 so Eulers
relation gives
(n 1)P
i
(a) =

j
P
ij
(a)a
j
.
This always gives

i,j
P
ij
(a)a
i
a
j
= (n 1)

i
P
i
(a)a
i
= n(n 1)P(a) = 0 (8)
and in our case also

i,j
P
ij
(a)a
i

j
= (n 1)

i
P
i
(a)
i
= 0
from the rst equation. Together with the second equation we see that the quadratic
form on the vector space spanned by a and (the subspace of C
3
dening the line L)
vanishes completely. This means that the matrix of the quadratic form with respect
to a basis a, , is of the form
_
_
0 0
0 0

_
_
and so det P
ij
(a) = 0.
Conversely take a basis of a,, where a and dene the tangent at p. From (8)
the matrix is of the form
_
_
0 0
0

_
_
(9)
If

ij
P
ij
a
i

j
= 0 then

i
P
i

i
= 0 by homogeneity, but then [
0
,
1
,
2
] lies on the
tangent so a, , do not form a basis. Hence the determinant det P
ij
vanishes if and
only if the central term

i,j
P
ij
(a)
i

j
= 0.
2
35
Remark: The inection points are the points of CH where H is the curve (called
the Hessian) with equation det P
ij
= 0. From Proposition 8 they exist. In fact they
are nite in number if the degree of P is bigger than one. To see this, suppose there
are innitely many. Then the Hessian and C have a common component. Since C
is nonsingular it must be a component of H and so every point is a ex. But then,
using the description of the curve as a graph y = g(x), we see that g

(x) 0 and
g(x) = +x. But then
P(x, +x, 1) 0
and y x is a factor of P(x, y, 1) which is a contradiction unless P(x, y, 1) is
linear.
Now start the proof of Theorem 15.
Proof: The partial derivatives P
ij
are of degree 3 2 = 1, so det P
ij
= 0 is a cubic
curve H.
Take [0, 1, 0] to be an inection point and z = 0 to be the tangent. Then from (7)
P(t, 1, 0) = t
3
c which means that
P(x, y, z) = kx
3
+a
1
(x, y)z +a
2
(x, y)z
2
+a
3
z
3
.
Since z = 0 is the tangent at [0, 1, 0], P/z = 0 so the coecient of y
2
z is nonvan-
ishing. Moreoever clearly the coecient of y is divisible by z, so the equation can be
put in the form
y
2
z +yz(x +z) +b
1
(x, z) = 0.
Now complete the square to write this as (y (x + z)/2)
2
z b
2
(x, z) = 0 and
apply the projective transformation [x, y, z] [x, x/2 +y z/2, z] to transform
it to y
2
z = b
3
(x, z). Since C is nonsingular it is irreducible, so z does not divide
b
3
(x, z) and therefore the coecient of x
3
is non-zero so we can write
y
2
z = A(x az)(x bz)(x cz)
and then (x, y, z) ((x az)/(b a), y, z) and rescaling y takes it to the form
y
2
z = x(x z)(x z).
If = 0 then [0, 0, 1] is a singularity and if = 1 then [1, 0, 1] is. 2
36
3.5 Bezouts theorem
Now that we have a denition of intersection multiplicity, we can formulate the basic
theorem about the intersection of curves:
Theorem 17 (Bezouts theorem) If C and D are two algebraic curves in P
2
of degrees
n, m with no common component, then

pCD
I
p
(C, D) = mn.
Proof: Using the coordinates in the denition of multiplicity, we express the resul-
tant as a product of linear factors:
R
P,Q
(y, z) =

i
(c
i
z b
i
y)
e
i
where e
1
+. . . +e
k
= mn.
By the arguments in Theorems 8 and 9 each such factor gives a point p
i
C D
with I
p
i
(C, D) = e
i
. 2
When is mn the actual number of intersections? Clearly only when I
p
i
(C, D) = 1.
We need the following:
Proposition 18 The intersection multiplicity I
p
(C, D) is equal to one if and only
if p is a nonsingular point of C and D and the tangent lines to C and D at p are
distinct.
Proof: From Proposition 13, if I
p
(C, D) = 1, p must be a nonsingular point of C
and D. As usual, choose coordinates such that p = [0, 0, 1]. We need to show that the
tangent lines coincide if and only if y
2
divides the resultant R
P,Q
(y, z), or equivalently
that the derivative of R
P,Q
(y, 1) vanishes at y = 0.
Now by assumption [1, 0, 0] does not lie on the tangent line to p for either curve so
P/x(0, 0, 1) = 0 and similarly for Q. The implicit function theorem then tells
us that in a suitable small neighbourhood, the solution x of P(x, y, 1) = 0 is a
holomorphic function of y. In other words, near [0, 0, 1], the roots
1
(y),
1
(y) of P
and Q which coincide when y = 0 are holomorphic functions of y.
Thus
P(x, y, 1) = (x
1
(y))(x, y) Q(x, y, 1) = (x
1
(y))m(x, y)
37
for polynomials , m in x with coecients which are holomorphic functions of y.
Then the resultant R
P,Q
(y, 1) = (
1
(y)
1
(y))S(y) where S(y) is holomorphic.
Dierentiating at y = 0,
R
P,Q
(y, 1)
y
|
y=0
= (

1
(0)

1
(0))S(0). (10)
We shall show next that S(0) = 0.
Since P/x(0, 0, 1) = 0, x = 0 is not a repeated root of P(x, 0, 1) so for i = 1,

i
(0) = 0 and similarly for Q. If
i
(0) =
j
(0) for i, j > 1 then [0, 0, 1] and [
i
(0), 0, 1]
are distinct points in CD and [1, 0, 0] lies on the line joining them which contradicts
our assumptions. Now S(y) is a product of resultants and we see here that there is
no other coincidence of roots than
1
(0) =
1
(0) at y = 0. Thus S(0) = 0.
It follows that the derivative in equation (10) vanishes if and only if

1
(0)

1
(0) = 0.
Now since P(
1
(y), y, 1) 0, dierentiating with respect to y gives
P
x

1
(y) +
P
y
= 0
and at [0, 0, 1] by Eulers identity P/z = nP = 0, so the tangent line to C is
x

1
(0)y = 0 and for D x

1
(0)y = 0. Hence the tangents coincide if and only if

1
(0)

1
(0) = 0, which proves the theorem.
2
38
4 The genus of a curve
4.1 Riemann surfaces
Nonsingular projective algebraic curves provide us with a rich source of examples of
Riemann surfaces, objects dealt with in the Geometry of Surfaces course. This part
of the notes repeats some of the facts in the notes for the earlier course, and applies
them to algebraic curves. Let us recall the notion of an abstract surface: each point
has a neighbourhood U and a homeomorphism
U
from U to an open set V in R
2
. If
two such neighbourhoods U, U

intersect, then

U

1
U
:
U
(U U

)
U
(U U

)
is a homeomorphism from one open set of R
2
to another.
V
U
U
V
If we identify R
2
with the complex numbers C then we can dene:
Denition 15 A Riemann surface is a surface with a class of homeomorphisms
U
such that each map
U

1
U
is a holomorphic homeomorphism.
We call each function
U
a holomorphic coordinate. If we compose with an invertible
holomorphic map f :
U
(U) C then f
U
is another holomorphic coordinate. We
shall be studying properties of Riemann surfaces which do not depend on a particular
choice of coordinate.
Examples:
1. Let X be the complex projective line X = P
1
. Let U = {[z
0
, z
1
] P
1
: z
0
= 0}
with
U
(z) = z
1
/z
0
C. Now take U

= {[z
0
, z
1
] P
1
: z
1
= 0} and dene
39

U
(z) = z
0
/z
1
C. Then

U
(U U

) = C\{0}
and

1
U

(z) = z
1
which is holomorphic.
2. Let
1
,
2
C be two complex numbers which are linearly independent over the
reals, and dene an equivalence relation on C by z
1
z
2
if there are integers m, n such
that z
1
z
2
= m
1
+n
2
. Let X be the set of equivalence classes (with the quotient
topology). A small enough disc V around z C has at most one representative in
each equivalence class, so this gives a local homeomorphism to its projection U in X.
If U and U

intersect, then the two coordinates are related by a map

z z +m
1
+n
2
which is holomorphic.
This surface is topologically described by noting that every z is equivalent to one
inside the closed parallelogram whose vertices are 0,
1
,
2
,
1
+
2
, but that points
on the boundary are identied:
We thus get a torus this way. Another way of describing the points of the torus is as
the quotient group C/ where is the subgroup consisting of complex numbers of
the form m
1
+n
2
.
3. Let C be a nonsingular projective algebraic curve dened by P(x, y, z) = 0. Every
point lies in an ane open set of P
2
which is homeomorphic to C
2
. On z = 0 its
equation is P(x, y, 1) = 0 and if C is nonsingular one of P/x, P/y is non-zero.
Suppose it is the latter, then at (x, y) = (a, b) on the curve, the implicit function
40
theorem tells us that there are neighbourhoods V and W of a and b in C and a
holomorphic function g : V W such that for x V and y W P(x, y, 1) = 0 if
and only if y = g(x). Hence for (x, y) C (V W) the function x has an inverse
x (x, g(x)), and this is a local coordinate for C.
If P/x is non-vanishing we can do the same interchanging the roles of x and y, and
get x = h(y). Where both are non-vanishing
y = g(h(y))
and we have an invertible holomorphic function relating the two local coordinates.
On the ane set y = 0, the equation of the curve is P( x, 1, z) where, when z = 0,
x = x/y, z = 1/y and it is easy to see that the holomorphic coordinates on the
intersection of these two open sets is holomorphic and invertible.
Denition 16 A holomorphic map between Riemann surfaces X and Y is a contin-
uous map f : X Y such that for each holomorphic coordinate
U
on U containing
x on X and
W
dened in a neighbourhood of f(x) on Y , the composition

W
f
1
U
is holomorphic.
Example: Consider a projective algebraic curve C dened by P(x, y, z). If [0, 0, 1] =
C we can dene a map
f : C P
1
by
f([x, y, z]) = [x, y]. (11)
This is well dened because x and y are not simultaneously zero, and will play an
important role in what follows.
4.2 Maps to P
1
Maps from an algebraic curve to the projective line have another interpretation:
Denition 17 A meromorphic function f on a Riemann surface X is a map to
C {} such that for each coordinate neighbourhood f
U
1
is a meromorphic
function on
U
(U) C.
41
Examples: 1. A rational function
f(z) =
p(z)
q(z)
where p and q are polynomials is a meromorphic function on P
1
.
2. Look at the map (11) where f = y/x. In any of our local coordinates this is the
ratio of two holomorphic functions and so is meromorphic. We could also take any
polynomial in f, or ratios of these.
This is an analytic view of meromorphic functions: we can add, multiply them to-
gether and also divide by non-zero functions. These are all meromorphic and form a
eld. On the other hand, any individual meromorphic function on X can be geomet-
rically represented as a map to P
1
: if we remove f
1
(), then f is just a holomorphic
function F with values in C, so on this part of X, we dene the map
x [F(x), 1] P
1
.
If f(x) = , and U is a small enough coordinate neighbourhood of x, then (f
1
U
)
1
is a holomorphic function

F, and the map is dened by x [1,

F(x)] P
1
.
Before proceeding, recall some basic facts about ordinary holomorphic functions (see,
for example, Introduction to Complex Analysis, Second Edition, H. A. Priestley, OUP,
Price: 23.00 (Paperback) ):
A holomorphic function has a convergent power series expansion in a neigh-
bourhood of each point at which it is dened:
f(z) = a
0
+a
1
(z c) +a
2
(z c)
2
+. . .
If f vanishes at c then
f(z) = (z c)
m
(a
0
+a
1
(z c) +. . .)
where a
0
= 0. In particular zeros are isolated.
If f is non-constant it maps open sets to open sets.
If f

(c) = 0 then f has a local holomorphic inverse g (this is very useful for
changing a local coordinate to a more convenient one.)
42
|f| cannot attain a maximum at an interior point of a disc (maximum modulus
principle) unless f is constant.
f : C C preserves angles between dierentiable curves, both in magnitude
and sense.
Remark: One consequence of the maximum modulus principle is that a holomorphic
map f : X C from a compact connected Riemann surface X must be a constant:
by compactness |f| has a maximum at a, so in some coordinate neighbourhood of a
is represented by a holomorphic function F with an interior maximum modulus. It
follows that f(x) = c in a neighbourhood. Now consider the set of all x such that
f(x) = c in a neighbourhood of x. This is open and non-empty, but is also closed
because the zeros of f(x) c are isolated.
Now return to the situation of a holomorphic map f : X P
1
. For each point, this
is described by a locally dened holomorphic function F = f
1
U
.
If the inverse image of a P
1
is innite, then it has a limit point x by compactness
of X. In a holomorphic coordinate z around x with z(x) = 0, f is dened by a
holomorphic function F with a sequence of points z
n
0 for which F(z
n
) a = 0.
But the zeros of a holomorphic function are isolated, so we deduce that f
1
(a) is a
nite set.
Example: Take the map (11). The inverse image of a = [a
0
, a
1
] is the set of points
[a
0
t, a
1
t, z] such that P(a
0
t, a
1
t, z) = 0 which is a homogeneous polynomial in t and
z of degree n and splits into n factors.
By a similar argument the points at which the derivative F

vanishes are nite in num-

ber (check using the chain rule that this condition is independent of the holomorphic
coordinate).
Denition 18 Let f : X Y be a holomorphic map of Riemann surfaces. The point
x X is a ramication point if in local coordinates f is represented by a holomorphic
function F such that F

= 0 at x.
If f is any holomorphic function on C such that f

(0) = 0, we have
f(z) = z
n
(a
0
+a
1
z +. . .)
43
with a
0
= 0. We can expand
(a
0
+a
1
z +. . .)
1/n
= a
1/n
0
(1 +b
1
z +. . .)
in a power series and dene
w = a
1/n
0
z(1 +b
1
z +. . .).
Since w

(0) = 0 we can think of w as a new coordinate and then the map becomes
simply
w w
n
.
So, thinking geometrically of P
1
as a Riemann surface where we are allowed to change
coordinates, a ramication point can be locally put in the form z z
n
. The integer
n is its ramication index. If F

is not zero at z = 0 then clearly the index is 1.

Proposition 19 In the map (11), the ramication points are those points p C at
which the tangent line T
p
at p passes through [0, 0, 1]. The ramication index is the
intersection multiplicity I
p
(C, T
p
).
Proof: Assume that P/y = 0 and use x as a local coordinate to represent the
curve as y = g(x). The map is then g(x)/x if x = 0 and x/g(x) if x = 0. Assume the
rst case, then F

= 0 if and only if xg

(x) g(x) = 0. But P(x, g(x), 1) = 0 so

P
x
+g

(x)
P
y
= 0.
At the ramication point y = g(x) = xg

(x) and g

(x) = 0 so
x
P
x
+y
P
y
= 0
but from Eulers identity this means that P/z = 0. Hence the tangent line passes
through [0, 0, 1].
In coordinates suppose the ramication point is x = c and g(c)/c = a then the tangent
line is y = ax. If the ramication index is n, then locally
g(x) = ax + (x c)
n
h(x)
where h(c) = 0. Then since P(x, g(x), 1) 0 putting x c = s,
P(c +s, ac +as +s
n
k(s), 1) 0.
As in Proposition 16 this means that P(c + s, ac + as, 1) is divisible by s
n
and no
higher power, so the intersection multiplicity with the tangent y = ax is n. 2
44
Remark: The map in (11) has a more geometric character: for each point p C
consider the line joining it to [0, 0, 1]. This denes a map to the space of lines through
[0, 0, 1], but this is a projective line, though not a line in the P
2
in which C lies.
Instead, if we consider the dual space V

of linear maps f : V C, those which

annihilate (0, 0, 1) form a two-dimensional subspace, hence a line in P(V

), the space
of all lines in P(V ) through [0, 0, 1].
[0,0,1]
ramification point
C
4.3 The degree-genus formula
From the Geometry of Surfaces course we have:
Proposition 20 A Riemann surface is orientable.
and hence a nonsingular projective algebraic curve is a compact orientable surface.
We also have:
Proposition 21 A nonsingular projective algebraic curve is connected.
Proof: First consider the special curve x
n
+y
n
z
n
= 0 (sometimes called a Fermat
curve for obvious reasons). The intersection multiplicity of y z = 0 with the curve
45
is clearly n, so [0, 1, 1] is a ramication point of the map f = y/z with multiplicity
n. This means that f
1
(U) is connected for a small neighbourhood U of 1 C. If
there is another connected component C
0
, then 1 = f(C
0
) but then f maps C
0
to
C

= P
1
\ {1} and so f is a constant c. But then the line y cz = 0 divides P but we
have assumed that C is nonsingular. So this curve is denitely connected.
Intuitively, it is clear that if we change the coecients slightly, the number of con-
nected components doesnt change. Below we will give a better justication of this
fact, but lets assume that for the moment. The condition for nonsingularity is the
vanishing of a polynomial in the coecients so if we take curves dene by P and Q,
then tP(x, y, z) + (1 t)Q(x, y, z) for t C will be nonsingular unless a polynomial
in t vanishes at a nite number of points or is identically zero. If the latter we can
replace this path between P and Q by a series of such complex intervals for which
the singular curves are given by the vanishing of a polynomial in t. Either way, we
can avoid a nite number of points in C by a real path joining P to Q, and so have a
path of curves all of which are nonsingular. If we start with P(x, y, z) = x
n
+y
n
z
n
,
then since P = 0 is connected, so is the curve dened by Q. 2
Remark: We give here a bit more detail about one way of seeing that if we vary the
coecients continuously through nonsingular curves, then in fact any two such curves
are homeomorphic, much stronger than just having the same number of connected
components. It involves the rst fundamental form for a three-dimensional object
a dierentiable 3-manifold but it is just a generalization of what you have seen in
the Geometry of Surfaces course.
So suppose P(x, y, z, t) is a homogeneous polynomial whose coecients depend dif-
ferentiably on the real parameter t, and so that for each t the curve P(x, y, z, t) = 0 is
nonsingular. Inside P
2
R we look at the set P(x, y, z, t) = 0. In ane coordinates,
if P
x
(x, y, 1, t) = 0 then by the implicit function theorem with y = u +iv, we have a
locally dened function x(u, v, t) such that
P(x(u, v, t), u +iv, 1, t) 0.
Repeating for the other ane open sets means that the set P(x, y, z, t) = 0 has the
structure of a 3-dimensional dierentiable manifold M, with local coordinates (u, v, t).
Moreover, t is a well-dened smooth function on M and since t is always part of a
coordinate system, its derivative never vanishes. The level set t = c is the algebraic
curve (or real surface in this context), P(x, y, z, c) = 0.
We now want to introduce a Riemannian metric, or rst fundamental form on M.
This means measuring the length of smooth curves on M P
2
R. there are many
46
ways of doing this, the easiest is to embed P
2
R in some Euclidean space R
N
and
just use the usual length of a curve there. Since a point in P
2
is a one-dimensional
subspace of C
3
we can describe it by a 3 3 matrix which acts as 1 in this direction
and 1 on the orthogonal complement with respect to the Hermitian inner product
on C
3
. The space of all 3 3 complex matrices has real dimension 18, so this gives
an embedding P
2
R R
19
.
Using this inner product on tangent vectors to M, we take the one-dimensional orthog-
onal complement to the tangent space of the surface t = c. Since t is a constant on t =
c and its derivative is non-zero, then the derivative of t in this normal direction is non-
zero, so we can nd a family X of tangent vectors to M, normal to t = c, such that the
derivative of t in this direction is 1. Vary c and we get a vector eld on M always point-
ing normal to the level sets of t and with the property that the derivative of t in this di-
rection is 1.
M
t=c X
We now integrate this vector eld for small values of t. Analytically this means solving
a dierential equation. If we think of X as a wind velocity, we want to see where the
surface t = c gets blown to after time s. In local coordinates this is an equation of
the form
du
ds
= a(u, v, t)
dv
ds
= b(u, v, t)
dt
ds
= 1
47
where the last relation holds since the derivative of t in the direction X is one. Since
t = c is compact, the existence theorems for dierential equations say that for s small
we have a solution and this means that (u(0), v(0), t(0)) (u(s), v(s), t(s)) gives a
dieomorphism from t = c to t(s) = c +s. By connectedness this extends for t lying
in a whole interval.
A nonsingular algebraic curve is thus a compact connected oriented surface which
means that C is determined up to homeomorphism by its Euler characteristic, which
is of the form (C) = 2 2g.
Denition 19 The genus of a nonsingular algebraic curve C is the integer g 0
such that (C) = 2 2g.
Example: The projective line P
1
= C {} is homeomorphic to a sphere by
stereographic projection. It has genus zero.
Using the map to P
1
considered above, we shall calculate g in terms of the degree n
of the polynomial P(x, y, z) which denes C, by using the Riemann-Hurwitz formula.
Recall that for a map f : C P
1
there are two types of points: if F

(x) = 0, then
the inverse function theorem tells us that f maps a neighbourhood U
x
of x C
homeomorphically to a neighbourhood V
x
of f(x) P
1
. If F

(x) = 0 then the map

looks like z z
n
which is not a homeomorphism the inverse image of 0 is a single
point but of any other nearby point it is n points. Dene V to be the intersection of
the V
x
as x runs over the nite set of points such that f(x) = a, then f
1
V consists
of a nite number of disjoint open sets, each one of which is mapped to V by a map
of one of the above forms.
V
f
Removing the nite number of images under f of ramication points (this is called
the branch locus of the map) we get a sphere minus a nite number of points. This
is connected. The number of points in the inverse image of a point in this punctured
sphere is integer-valued and continuous, hence constant. It is called the degree d of
the map f.
48
Example: For the map (11) the inverse image of a = [a
0
, a
1
] is the set of points
[a
0
t, a
1
t, z] such that P(a
0
t, a
1
t, z) = 0. This inverse image contains no ramication
points if this has no repeated factors, so the degree of the map is the degree of the
polynomial P(x, y, z).
Now recall from the Geometry of Surfaces course:
Theorem 22 (Riemann-Hurwitz) Let f : X P
1
be a holomorphic map of degree
d on a closed connected Riemann surface X, and suppose it has ramication points
x
1
, . . . , x
n
of multiplicity m
k
. Then
(X) = 2d
n

k=1
(m
k
1)
Proof: The idea is to take a triangulation of the sphere P
1
such that the image of
the ramication points are vertices (see Kirwans book for more details). Now take a
nite subcovering of P
1
by open sets of the form V above where the map f is either
a homeomorphism or of the form z z
m
. Subdivide the triangulation into smaller
triangles such that each one is contained in one of the sets V . Then the inverse images
of the vertices and edges of P
1
form the vertices and edges of a triangulation of X.
If the triangulation of P
1
has V vertices, E edges and F faces, then clearly the
triangulation of X has dE edges and dF faces. It has fewer vertices, though in a
neighbourhood where f is of the form w w
m
the origin gives a single vertex instead
of m of them. For each ramication point of order m
k
we therefore have one vertex
instead of m
k
. The count of vertices is therefore
dV
n

k=1
(m
k
1).
Thus
(X) = d(V E +F)
n

k=1
(m
k
1) = 2d
n

k=1
(m
k
1)
using (P
1
) = 2. 2
Clearly the argument works just the same for a holomorphic map f : X Y and
then
(X) = d(Y )
n

k=1
(m
k
1).
We can now calculate the genus:
49
Theorem 23 Let C be a non-singular projective algebraic curve of degree n. Then
the genus of C is
g =
(n 1)(n 2)
2
.
Proof: Take our familiar map f = y/x. As we saw above, the ramication points
occur where the tangents pass through [0, 0, 1] and are therefore given by the equation
P/z = 0. The multiplicity is bigger than 2 only if I
p
(C, T
p
) > 2, i.e. if p is
an inection point, but there are only nitely many of these, so by a projective
transformation we can assume that [0, 0, 1] does not lie on the tangent to any one of
them. This means that each m
k
in the Riemann-Hurwitz formula is 2, and it remains
to calculate the number of ramication points.
This is the number of points of intersection of P = 0, the curve C of degree n, and
P/z = 0, a curve D which is of degree n1. Since C is nonsingular it is irreducible,
and so C and D can have no common component. We will use Bezouts theorem, so
we need to check that [a] = [a
0
, a
1
, a
2
] C D is a nonsingular point of D and that
the tangent lines are distinct. Now (P
zx
, P
zy
, P
zz
) is not identically zero at [a
0
, a
1
, a
2
]
because this would make the Hessian of C vanish and we know that [a
0
, a
1
, a
2
] is not
an inection point. This shows that D is nonsingular here.
Suppose that the tangents of C and D coincide then (P
zx
, P
zy
, P
zz
) is a multiple of
(P
x
, P
y
, P
z
). As in our discussion of inection points we use the symmetric bilinear
form B dened by the matrix of partial derivatives P
ij
. Then B(a, a) = 0 = B(a, )
where the tangent line joins [a] and []. Put v = (0, 0, 1).
By the Euler identity
a
0
P
zx
+a
1
P
zy
+a
2
P
zz
= (n 1)P
z
= 0
since P
z
(a) = 0. This gives B(a, v) = 0. Moreover since P
zz
(a) = P
z
(a) = 0, we
have B(v, v) = 0.
Since [a] is not an inection point, det B = 0 so from
0 = B(a, a) = B(a, ) = B(a, v)
we deduce v = a +. But then
0 = B(v, v) =
2
B(, )
and, as in Proposition 16, this gives det B = 0 unless = 0. But then [a] = [0, 0, 1]
which we have specically excluded. We conclude that the tangents are distinct and
50
it follows that the conditions for Bezouts theorem hold, so the number of ramication
points is exactly n(n 1).
From the Riemann-Hurwitz formula we obtain
2 2g = 2n n(n 1)
and so
g =
1
2
(n 1)(n 2).
2
Remark: When n = 1, g = 0 and we have already seen that the projective line is
homeomorphic to the sphere. The genus also vanishes when n = 2, a conic, but from
Theorem 7, we saw that this is homeomorphic to the sphere too.
4.4 The torus and the cubic
When n = 3 we get g = 1 which means that a nonsingular cubic surface is homeomor-
phic to a torus. Here is a more concrete realization of that fact. We take the Riemann
surface which is the second example in Examples 4.1. This is the set of equivalence
classes of z C where z
1
z
2
if there are integers m, n such that z
1
z
2
= m
1
+n
2
and is clearly a torus.
As in the Geometry of Surfaces course, dene
(z) =
1
z
2
+

=0
_
1
(z )
2

1

2
_
where the sum is over all non-zero = m
1
+ n
2
, with m, n integers. Since for
2|z| < ||

1
(z )
2

1

10
|z|
||
3
this converges uniformly on compact sets so long as

=0
1
||
3
< .
But m
1
+n
2
is never zero if m, n are real so we have an estimate
|m
1
+n
2
| k

m
2
+n
2
51
so by the integral test we have convergence. Because the sum is essentially over all
equivalence classes, we have
(z +m
1
+n
2
) = (z)
so that this is a meromorphic function on the surface X. It is called the Weierstrass
-function.
We now want to know geometrically what this map from a torus to the projective
line looks like.
Firstly, has degree 2 since (z) = only at z = 0 and there it has multiplicity 2.
The multiplicity of any ramication point cannot be bigger than this because then it
will look like z z
n
and a non-zero point will have at least n inverse images. Thus
the only possible value at the ramication points in this example is m
k
= 2. The
Riemann-Hurwitz formula gives:
0 = 4 n
so there must be exactly 4 ramication points. In fact we can see them directly,
because (z) is an even function, so the derivative vanishes if z = z. Of course
at z = 0, (z) = so we should use the other coordinate on S: 1/ has a zero of
multiplicity 2 at z = 0. To nd the other points recall that is doubly periodic so

vanishes where
z = z +m
1
+n
2
for some integers m, n, and these are the four points
0,
1
/2,
2
/2, (
1
+
2
)/2 :
The geometric Riemann-Hurwitz formula has helped us here in the analysis by show-
ing us that the only zeros of

are the obvious ones. Suppose, by a projective

52
transformation of P
1
, that their images are and the three nite points e
1
, e
2
, e
3
where
e
1
= (
1
/2), e
2
= (
2
/2), e
3
= ((
1
+
2
)/2).
The derivative

(z) vanishes at three points, each with multiplicity 1. At each of

these points has the local form
(z) = e
1
+ (z
1
/2)
2
(a
0
+. . .)
and so
1

(z)
2
((z) e
1
)((z) e
2
)((z) e
3
)
is a well-dened holomorphic function on X away from z = 0. But (z) z
2
near
z = 0, and so

(z) 2z
3
and hence this function is nite at z = 0 with value 1/4.
By the maximum argument, since X is compact, the function is a constant, namely
1/4, and

(z)
2
= 4((z) e
1
)((z) e
2
)((z) e
3
). (12)
This is the equation of a cubic curve C:
y
2
= 4(x e
1
)(x e
2
)(x e
3
).
Now : X P
1
is surjective (otherwise the degree would be zero!), so (z) takes
every value in P
1
. Moreover, since (z) = (z),

(z) =

(z) so for each value

of x there is a z for both values of y.
Therefore z [(z),

(z), 1] denes a homeomorphism from X to C.

53
x
5 The Riemann-Roch theorem
5.1 Divisors
Let f be a meromorphic function on an open set U C. If f(a) = 0 then near a
f(z) = (z a)
m
g(z)
where m > 0 is the multiplicity of the zero and g(z) is holomorphic with g(a) = 0.
So if

f has a zero at a with the same multiplicity then

f
f
=
g(z)
g(z)
which is holomorphic. The multiplicity is independent of the coordinate z.
Similarly near a pole b
f(z) = (z b)
n
h(z)
where h(b) = 0.
Now consider a meromorphic function f on an algebraic curve C. Suppose its zeros are
p
1
, . . . , p
k
with multiplicities m
1
, . . . , m
k
and its poles are q
1
, . . . , q

with multiplicities
n
1
, . . . , n

. If

f is another meromorphic function with exactly the same zeros and poles
and multiplicities, then

f/f is a (non-vanishing) holomorphic function and by the
maximum principle this is a constant. So, up to a constant multiple, f is determined
by the zeros, poles and multiplicities. We write this is
k

i=1
m
k
p
i

i=1
n
i
q
i
.
Denition 20 A divisor D on a curve C is a formal sum
D =

pC
n
p
p
where for each point p C, n
p
is an integer and n
p
= 0 for all but nitely many
points.
The degree of D is dened by
deg D =

pC
n
p
.
54
For a meromorphic function we write its divisor as (f). Not all divisors come from
meromorphic functions. In the rst place deg(f) = 0. This comes from the geometric
interpretation of f as a map f : C P
1
. The number of zeros (counted with
multiplicities) is the corresponding count of the number of points in f
1
(0) which is
the degree of f considered as a map to P
1
. But this is the same number for f
1
()
or any other point.
More importantly, even if the degree of D is zero, there may not be a meromorphic
function with the same divisor. On P
1
it is true, because given any points a, b C,
(z a)/(z b) is a meromorphic function with a simple zero at a and a simple pole
at b, so for any two points p, q P
1
, p q and hence any sum

i
(p
i
q
i
)
is the divisor of a meromorphic function.
However, suppose that on a curve C (or more generally a Riemann surface) we have
a meromorphic function f with a simple zero only at p and a simple pole only at
q. Then in the geometrical viewpoint f : C P
1
, f
1
{0} consists of one point so
the degree of the map is 1 and furthermore there can be no ramication points since
in the neighbourhood of z z
n
a nonzero value has n inverse images. From the
Riemann-Hurwitz formula 2 2g = 2 and so this feature can only occur when g = 0.
A divisor is said to be eective, or positive, if all the n
p
are nonnegative. We can add
and subtract divisors in the obvious way and so we write
D D

if D D

is eective. So eective means D 0. Clearly if D D

, then deg D
deg D

.
Denition 21 A divisor D is said to be a principal divisor if D = (f) for some
meromorphic function f.
Divisors D, D

are said to be linearly equivalent if D D

is a principal divisor.
We write D D

for two linearly equivalent divisors. Since deg(f) = 0 in the divisor

sense, the degree of two linearly equivalent divisors is the same.
We shall be interested in the eective divisors in a given equivalence class. What
does this mean? Consider an eective divisor
k

i=1
m
i
p
i
55
where m
i
> 0. An eective divisor

i
n
i
q
i
is equivalent to this if there is a meromor-
phic function f such that
(f) =

i
m
i
p
i

i
n
i
q
i
.
In other words D is dened by a meromorphic function whose zeros are precisely the
p
i
with multiplicity m
i
. For any divisor we make the following denition:
Denition 22 Let D be a divisor on the curve C. Denote by L(D) the set of mero-
morphic functions f such that (f) +D 0 together with the zero function.
Proposition 24 (i) L(D) is a nite-dimensional vector space.
(ii) If deg D < 0, then L(D) = 0.
(iii) If D D

then dimL(D) = dimL(D

).
(iv) The projective space P(L(D)) is in one-to-one correspondence with the eective
divisors equivalent to D.
Proof: (i) Write D as

i
m
i
p
i

i
n
i
q
i
where the m
i
, n
i
are positive, then we are looking at meromorphic functions f which
have a pole of order m
i
at p
i
and no more poles, and have zeros of order n
i
at q
i
.
In other words, functions f whose zeros cancel the q
i
and whose poles are cancelled
by the p
i
. This set is clearly closed under addition and scalar multiplication, and so
forms a vector space L(D).
At each pole p
i
, in a local coordinate z, we can write
f(z) =
a
m
i
(z z
i
)
m
i
+. . . +
a
1
(z z
i
)
+h(z)
where h is holomorphic and f (a
m
i
, . . . , a
1
) is a linear map from L(D) to C
m
i
.
The intersection of the kernels of this nite number of maps consists of holomorphic
functions which must be constant and hence at most a one-dimensional space, so the
vector space L(D) must be nite-dimensional.
(ii) If (f) +D 0 then 0 deg(f) + deg D = deg D.
(iii) If D = D

+ (g), then f fg denes an isomorphism from L(D) to L(D

).
56
(iv) If f L(D) then by denition (f) +D is eective and linearly equivalent to D.
2
The dimension (D) = dimL(D) is a very subtle thing in general and there is no
simple formula in terms of the genus g and deg D. The Riemann-Roch theorem
however relates a particular pair of these numbers.
5.2 Canonical divisors
The derivative of a meromorphic function is not a function. What do we mean by
this? In one coordinate we write f
1
U
= g(z) and then the derivative with respect
to z is g

(z). But if we change coordinates then

f
1
U

= g(h(u))
where h =
U

1
U

and then the derivative with respect to u is g

(h(u))h

(u) and not

g

(h(u)).
Denition 23 A meromorphic dierential on a Riemann surface is a collection of
meromorphic functions f
U
on
U
(U) such that on
U
(U U

)
f
U
= f
U
(h(u))h

(u)
where h =
U

1
U

.
The derivative df of any meromorphic function is a meromorphic dierential but not
all are of this form.
Remark: 1. The transformation law is more easily described if we write a dierential
as f
U
(z)dz, thinking of dz as the derivative of the locally dened coordinate z.
2. The residue of a dierential at a pole is independent of the local coordinate.
This diers from a function, whose value is invariant. The simplest way to see the
invariance of the residue is to note that the transformation law for dz means that the
contour integral
_

fdz
is well-dened and independent of coordinates. Cauchys residue theorem then gives
the invariance.
57
Two dierentials are dened by local functions f
U
, g
U
satisfying the conditions in
Denition 23. Then
f
U
g
U
=
f
U

g
U

on U U

and this denes a meromorphic function the ratio of two meromorphic

dierentials is a meromorphic function. This means that the divisors of any two
dierentials are linearly equivalent, and this is called the canonical divisor class.
Examples:
1. The function z = z
0
/z
1
on P
1
is a meromorphic function with a simple pole at
[1, 0], because in the coordinate z = z
1
/z
0
near [1, 0] it is 1/ z. Its derivative is the
dierential
dz = d( z
1
) =
1
z
2
d z
which has no zeros but has a double pole. A canonical divisor on P
1
therefore has
degree 2.
2. The Weierstrass -function (z) has a double pole at z = 0. Its derivative has a
triple pole there and vanishes as we have seen at three points
1
/2,
2
/2, (
1
+
2
)/2.
The degree of a canonical divisor here is then 3 3 = 0.
Let be a canonical divisor, then the vector space L() is isomorphic to the space of
holomorphic dierentials. Since deg = 2 for P
1
we see that here there are none.
For the torus deg = 0 which means that any holomorphic dierential must have no
zeros, and so there is at most a one-dimensional space of them. In fact since
d(z +m
1
+n
2
) = dz
the form dz is such a dierential.
More generally:
Proposition 25 On a nonsingular algebraic curve of genus g, the degree of a canon-
ical divisor is 2g 2.
Proof: We use the map f : C P
1
as in the proof of the degree-genus formula
dened by [x, y] P
1
. Let [a, 1] P
1
be a point which is not a branch point and
consider the meromorphic function
g =
y
x ay
.
58
(All we have done is composed f with a projective transformation of P
1
).
Then this meromorphic function has n simple poles so dg has n double poles. But g

vanishes at the ramication points since by choice none of these lie over , and in
the proof of Theorem 23 there are n(n 1) of these. Hence the degree of the divisor
of dg is
n(n 1) 2n = 2g 2
from Theorem 23. 2
The only dierentials we have encountered so far are derivatives of meromorphic
functions, which can only have poles of order 2 or more since they are the derivatives
of (z c)
m
. But on P
1
the dierential dz/z has simple poles at 0 and . If we
tried to integrate this we would get log z which is not meromorphic and not even
single-valued. So dierentials with simple poles do exist, but there is a constraint. In
particular:
Proposition 26 A meromorphic dierential cannot have a single simple pole.
Proof: Suppose for a contradiction that p is the pole of the dierential . It has
non-zero residue and so taking a coordinate neighbourhood of p, and surrounding it
with a small contour , we have
_

= 0.
Now triangulate C such that each triangle lies in a coordinate neighbourhod and p lies
in the interior of one,
0
. By Cauchys theorem the integral of around each triangle

i
, i = 0 is zero and the integrations along adjacent adges of dierent triangles cancel
(this is like the proof of the Gauss-Bonnet Theorem in the Geometry of Surfaces
course.) But then the integral around
0
vanishes which is a contradiction. 2
Remark: It is clear that this argument can be extended to show that the sum of
the residues of a meromorphic dierential is always zero.
5.3 Riemann-Roch
This is the theorem we aim to prove about the dimension (D) of L(D):
Theorem 27 (Riemann-Roch) Let D be any divisor on a nonsingular projective al-
gebraic curve in P
2
and let be a canonical divisor, then
(D) ( D) = deg D + 1 g.
59
The theorem doesnt tell us what the value of (D) is, which can jump up and down
depending on the location of the points, but it tells us the dierence of two such
numbers. This can be used to great eect. For example, if deg D > g 1 the right
hand side is positive so we know that (D) 0. If deg D > 2g2, then deg(D) < 0
so ( D) = 0 and then we have an exact formula (D) = deg D + 1 g.
There is one class of divisors where we can already nd lots of elements in L(D), and
we shall need this for the proof of the theorem. Take a line L in P
2
and consider the
divisor
H =

pLC
I
p
(C, L)p.
By Bezouts theorem the degree of H is n, the degree of the curve C.
If ax +by +cz = 0, a

x +b

y +c

z = 0 are two lines L, L

, then
f =
ax +by +cz
a

x +b

y +c

z
is a meromorphic function, so the divisor of the line L is linearly equivalent to the
divisor of L

.
Any algebraic curve Q(x, y, z) = 0 of degree m denes a divisor in the same way, and
Q(x, y, z)
(ax +by +cz)
m
is a meromorphic function, so the divisor of Q is in the class of mH. Two such
polynomials Q and Q

dene the same function on C if and only if they are divisible

by P, so we can nd easily a large subspace of L(mH) just by counting polynomials.
Writing
Q(x, y, z) = b
0
(y, z) +b
1
(y, z)x +. . . +b
m
(y, z)x
m
the coecient b
i
(y, z) is homogeneous of degree mi and so has mi +1 coecients.
Thus the vector space of all Q has dimension 1 + 2 + (m + 1) = (m + 1)(m + 2)/2.
The subspace of all Q = PR for R of degree m n is then of dimension (m n +
1)(mn + 2)/2 and so
(mH)
1
2
((m+ 1)(m+ 2) (mn + 1)(mn + 2)) = mn
1
2
n(n 3). (13)
The degree of H is the number of points of intersection of C with a line which is n,
so deg(mH) = mn. From the degree-genus formula we see that mn n(n 3)/2 =
deg(mH) + 1 g which is the right hand side of the Riemann-Roch formula.
We start the proof of Theorem 27 with a similar-looking proposition:
60
Proposition 28 If D is any divisor on C then (D) ( D) deg D + 1 g.
Proof: We saw above that (mH) deg(mH)+1g. Moreover if m is large enough
deg( mH) < 0 so ( mH) = 0. Therefore we already have the inequality for
D = mH.
In the general case
D =
k

i=1
m
i
p
i

i=1
n
i
q
i
choose lines a
i
x +b
i
y +c
i
z = 0 that pass through the points p
i
then with m =

i
m
i
the divisor of
(a
1
x +b
1
y +c
1
z)
m
1
(a
2
x +b
2
y +c
2
z)
m
2
. . . (a
k
x +b
k
y +c
k
z)
m
k
is of the form
k

i=1
m
i
p
i
+
N

j=1
r
j
= D +x
1
+x
2
+. . . +x
r
.
Furthermore, this divisor is linearly equivalent to mH and so
(mH) = (D +x
1
+x
2
+. . . +x
r
). (14)
By adding more x
i
we can assume that m is large enough that deg(mH) < 0. We
now need the following lemma:
Lemma 29 For any point p,
0 (D +p) ( D p) (D) +( D) 1
Proof: Firstly f L(D) if and only if (f)+D 0 which clearly implies (f)+D+p
0, so that L(D) L(D +p) and
(D +p) (D).
Suppose
D =
k

i=1
m
i
p
i

i=1
n
i
q
i
.
Take f L(D+p). If p is not one of the p
i
or q
j
then f has at most a simple pole at
p. The condition for f to lie in L(D) is thus a single linear condition, the vanishing
61
of the coecient of (z a)
1
. If p = p
1
say, then in the Laurent expansion around p
we have:
f(z) =
a
m
1
+1
(z z
1
)
m
1
+1
+. . . +a
0
+. . .
and here for f to lie in L(D) is the vanishing of a
m
1
+1
. If p = q
1
then f has a zero of
order at least (n
1
1) at q
1
and to lie in L(D), must have a zero of order n
1
. This is
again one linear condition. In all cases we see that
dimL(D +p) dimL(D) + 1.
Applying this to D and D p, we see that the lemma holds so long as we can
eliminate the case
(D +p) (D) = 1 ( D) ( D p) = 1.
Suppose for a contradiction that this holds. Then there is a meromorphic function f
with (f) +D+p 0 but (f) +D 0, so p is the only negative term in (f) +(D).
Similarly there is g such that (g) + D 0 but (g) + Dp 0 which means
that p does not appear in the divisor (g) + D. Thus in
0 (f) +D +p + (g) + D = (fg) + +p
the positive element p is not cancelled.
But is the divisor of a dierential and this means that fg is a dierential with
a single simple pole at p, which is impossible from Proposition 26.
From the lemma, we have
(D+x
1
+. . .+x
r
)(Dx
1
. . .x
r
) (D+x
1
+. . .+x
r1
)(Dx
1
. . .x
r1
)+1
and repeating we see that
(D +x
1
+. . . +x
r
) ( D x
1
. . . x
r
) (D) ( D) +r
or, using (14),
(mH) ( mH) (D) ( D) +r
or
(D) ( D) (mH) ( mH) r.
So, since we know the inequality for mH,
(D) ( D) deg(mH) + 1 g r = deg D + 1 g
which establishes Proposition 28. 2
62
The Riemann-Roch theorem follows directly from this: use the Proposition for D and
for D then
(D) (K D) deg D + 1 g
and
( D) (D) deg( D) + 1 g
= 2g 2 deg D + 1 g
= deg D +g 1
so equality holds.
5.4 Applications
The rst consequence is another interpretation of the genus g:
Theorem 30 The vector space of holomorphic dierentials on a nonsingular alge-
braic curve has dimension g, the genus of the curve.
Proof: This dimension is (), so take D = 0 and use Riemann-Roch:
(0) () = 1 g.
But L(0) consists of holomorphic functions on C which are just the constants and
hence one-dimensional, so (0) = 1, and so Riemann-Roch gives () = g. 2
We can actually write down these dierentials. First consider the ane part of the
curve given by P(x, y, 1) = 0. Then x is a local coordinate where P/y = 0 so
consider the dierential
=
dx
P/y(x, y, 1)
.
At rst sight this seems to have poles where the denominator vanishes but this is just
where the role of x as a local coordinate breaks down. Since the curve is nonsingular,
at such points P/x = 0 and from the chain rule, on the curve (P/x)dx +
(P/y)dy 0, so that can also be written, using y as a coordinate, as
=
dy
P/x(x, y, 1)
.
This form has no poles and no zeros in the ane part of the curve.
63
Now look at C near z = 0. We have
d(x/z)
P/y(x/z, y/z, 1)
=
d(x/z)
P/y(1, y/x, z/x)(x/z)
n1
=
d(z/x)(x/z)
2
P/y(1, y/x, z/x)(x/z)
n1
and so
=
z
n3
dz
P/y(1, y, z)
and has a zero of order n 3 where z = 0.
This tells us that (n 3)H, and so we can obtain a holomorphic dierential by
writing
Q(x, y, 1)dx
P/y(x, y, 1)
.
for a homogeneous polynomial Q(x, y, z) of degree n3. The dimension of the space of
polynomials of this degree is (n2)(n1)/2 which is g from the degree-genus formula.
Riemann-Roch therefore tells us that every holomorphic dierential is obtained from
a polynomial this way.
Example: For a cubic curve n = 3 and the construction gives, up to a constant
multiple, the unique holomorphic dierential. If the cubic is
y
2
= 4(x e
1
)(x e
2
)(x e
3
)
then this is dx/2y, and if y =

(z), x = (z) as in section 4.4, this is the dierential

dz/2.
Example: When the curve C is a quartic, i.e. n = 4, then H, so the zeros of a
holomorphic dierential are the intersections of C with a line.
If m > (n 3) then deg( mH) < 0 so Riemann-Roch gives
(mH) = mn + 1 g = mn n(n 3)/2.
This is the lower bound we obtained in (13) by explicitly writing down polynomials
so we have a concrete construction also for the divisor class mH.
Another corollary of the theorem provides a link between the complex analysis and
the algebra. We have been considering meromorphic functions a great deal and these
are dened as analytical or geometrical objects. They are in fact all expressed in
terms of polynomials:
64
Theorem 31 Any meromorphic function on a nonsingular curve C is expressible as
a ratio of homogeneous polynomials of the same degree:
f =
Q(x, y, z)
R(x, y, z)
.
Proof: We showed using Riemann-Roch that L(mH) for m n3 is generated by
polynomials, in other words every meromorphic function f with
(f) +mH 0
is of the form Q(x, y, z)/(ax +by +cz)
m
.
If f is any meromorphic function, then taking lines L
i
which pass through the poles
of f, we have, for some m
(f) +H
1
+. . . +H
m
0
and the same argument shows that f is a rational function. 2
5.5 The group law on a cubic
An important property of a cubic curve is that the points on it form an abelian group.
It has a very geometrical description:
Theorem 32 Let C be a nonsingular curve of degree 3 and let e be an inection
point. There is a unique additive group structure on C such that e is the identity
element and p
1
+p
2
+p
3
= 0 if and only if p
1
, p
2
, p
3
are the three points of intersection
(counting multiplicities) of C with a line.
65
Proof: The addition of divisors is commutative and associative. This is also true of
their linear equivalence classes since if p = p

+(f), q = q

+(g) then p+q = p

+q

+(fg).
We noted earlier that p is linearly equivalent to q only if g = 0, so for any curve C
of genus g > 0 the equivalence class of p determines p uniquely. We could also use
Riemann-Roch: if D = p then since deg D = 1 > 0 = deg we have ( D) = 0
and
(D) = 1 + 1 1 = 1.
For the cubic, with g = 1, we take an inection point e and map p [p e] into the
group of equivalence classes of degree zero divisors. From the above, this is injective.
Moroever [e e] = [0] is clearly an identity.
Then p +q maps to [p +q 2e] and we want to show that this is of the form [s e].
The line ax + by + cz = 0 joining p and q (or the tangent at p if p = q) meets the
degree 3 curve in a third point r by Bezouts theorem. Let a

x +b

y +c

z = 0 be the
tangent at e, then the divisor of its intersection with C is 3e since e is an inection
point.
If f = (ax +by +cz)/(a

x +b

y +c

z), the divisor of f is

(f) = p +q +r 3e
which shows that [p +q 2e] = [e r].
Now take q = e in this expression, then [p e] = [e p

] for some p

which we call
the inverse of p.
In general then
[p +q 2e] = [e r] = [r

e]
as required, proving that C is closed under the addition law. 2
In Section 4.4 we showed how a torus C/ could be mapped isomorphically to a cubic
curve by the Weierstrass -function. Since C/ is clearly an abelian group it seems
reasonable to believe that this is the addition law described geometrically above. This
will be so if (1, (u),

(u)), (1, (v),

(v)), (1, (w),

(w)) are linearly dependent if

w = (u +v), so consider
det
_
_
1 (u)

(u)
1 (v)

(v)
1 (u v)

(u v)
_
_
.
Fix v and vary u, then this has a pole at u = 0 and (u) = 1/u
2
+ a(u) where a(u)
is holomorphic. So expanding the determinant gives

1
u
2
(

(u v)

(v))
2
u
3
((u v) (v)) +h(u)
66
where h(u) is holomorphic near u. But
(u v) = (u +v) = (v) +u

(v) +u
2

(v)/2 +u
3

(v)/6 +. . .
so this is

1
u
2
(2

(v) u

(v) u
2

(v)/2)
2
u
3
(u

(v) +u
2

(v)/2 +u
3

(v)/6) +k(u)
where k(u) is holomorphic. But the singular terms cancel so the determinant is nite
near u = 0, and similarly near u = v. But then it is holomorphic everywhere and
hence constant. But when u = v it vanishes, and so it is identically zero.
Remark: The inection points are the points p such that the divisor of a line is 3p,
or in the group law 3[p e] = 0. Given that the group law is addition in C/ these
are the nine points of order 3 in the group.
.
67