Algebraic Geometry
Uploaded by
alin444444Algebraic Geometry
Uploaded by
alin444444Lecture Notes
Algebraic Geometry III/IV
Matt Kerr
Contents
Part 1.
Introduction and Motivation
Chapter 1.
Two theorems on conics in the plane
Chapter 2.
Riemann surfaces and algebraic curves
21
Chapter 3.
The normalization theorem
39
Chapter 4.
Lines, conics, and duality
49
Part 2.
General de nitions and results
63
Chapter 5.
Complex manifolds and algebraic varieties
65
Chapter 6.
More on projective algebraic varieties
75
Chapter 7.
Smooth varieties as complex manifolds
87
Chapter 8.
The connectedness of algebraic curves
97
Chapter 9.
Hilbert's nullstellensatz
107
Chapter 10.
Local analytic factorization of polynomials
113
Chapter 11.
Proof of the normalization theorem
121
Chapter 12.
Intersections of curves
131
Chapter 13.
Meromorphic 1-forms on a Riemann surface
143
Chapter 14.
The genus formula
155
Chapter 15.
Some applications of Bzout
167
Part 3.
Cubic curves
173
Chapter 16.
The singular cubic
175
Chapter 17.
Putting a nonsingular cubic in standard form
187
CONTENTS
Chapter 18.
Canonical normalization of the Weierstrass cubic
195
Chapter 19.
Group law on the nonsingular cubic
207
Chapter 20.
Abel's theorem for elliptic curves
219
Chapter 21.
The Poncelet problem
225
Chapter 22.
Periods of families of elliptic curves
241
Chapter 23.
Counting
255
Part 4.
Fp -rational
points on elliptic curves
Curves of higher genus
263
Chapter 24.
The algebraicity of global analytic objects
265
Chapter 25.
The Riemann-Roch Theorem
273
Chapter 26.
Applications of Riemann-Roch, I: special Riemann
surfaces
Chapter 27.
281
Applications of Riemann-Roch, II: general Riemann
surfaces
291
Chapter 28.
Abel's Theorem, part I
303
Chapter 29.
Abel's Theorem, part II
311
Appendix: genera of singular curves
325
Part 1
Introduction and Motivation
CHAPTER 1
Two theorems on conics in the plane
The primary subject of this course is algebraic curves, the simplest
example of which is the solution set of a two-variable polynomial equa-
tion in the plane. What plane?
this course, mainly
C2 ,
R2 , C2 , F2p , Q2 . . .?
For the purposes of
with excursions into the others.
Remark 1.0.1. A perennial point of confusion is whether to call
the complex plane.
This is henceforth forbidden! It is the complex (a ne) line, and a real
C
= R2 . (A complex plane will mean something
2
2
over C, so C will be the complex a ne plane, P the
(a ne) plane via
2-dimensional
complex projective plane, and so forth.
We'll worry about a ne vs.
H := {x + iy | x, y R, y >
plane in C; that terminology is
projective in the next chapter.) Note that
0} C
will denote the upper-half
unavoidable.
The objects which shall concern us, then, will be 1-dimensional over
C (complex algebraic curves), hence 2-dimensional over R (Riemann
surfaces). Our approach will be quite intuitive and visual for the rst
few chapters, to get an idea of what algebraic geometry is before settling
into a more measured approach. My feeling has always been that you
need motivation for introducing formalism, in this case for layering lots
of algebra onto geometry. In this chapter that motivation might consist
of the subtle gaps that open as we try to prove some famous results on
conics from (mostly) linear algebra.
7
1. TWO THEOREMS ON CONICS IN THE PLANE
Example 1.0.2. (a) Consider the set of rational points on the Fer-
mat quartic (degree 4) curve:
{(x, y) Q2 | x4 + y 4 = 1} Q2 .
(1.0.1)
By a case of Fermat's last theorem, (1.0.1) is the empty set, since
4
a 4
+ dc = 1 means (ad)4 + (bc)4 = (bd)4 with ad, bc, bd Z and
b
bd 6= 0.
(b) Next we look at the Fermat cubic (degree 3) curve
{(x, y) C2 | x3 + y 3 = 1} C2 .
(1.0.2)
We will see that (1.0.2) has the structure of a complex
on the left-hand side in
1-torus,
shown
1+
The right-hand side represents the quotient
by the lattice
phism?
any
(for some
It holds topologically (convince yourself of this visually) for
H,
where
:= Z h1, i
C/ of the complex line
H). What is the isomor-
but complex analytically only for the values
a b
c d
!
SL2 (Z) and 0 :=
a0 +b
c0 +d
1+ 3
. We call (1.0.2) an elliptic
2
curve: it isn't an ellipse in any sense, but originally arose in connection
with the arc-length of one.
(c) Finally, take the real degree 2 (a.k.a. quadric or conic) curve
(1.0.3)
{(x, y) R2 | x2 + y 2 = 1} R2 ,
which is of course a circle:
First we shall pursue conics to get a preliminary feel for the interplay
between algebra and geometry. These aren't too hard to visualize: you
know what the real solution sets look like (ellipses, hyperbolas, pairs
1.1. ALGEBRAIC CURVES IN
R2
of lines, etc.) and the complex solutions are all spheres once we add
points at in nity.
1.1. Algebraic curves in
Let
P2n
R2
n in x and y . (If
n
write P .) In an exercise
denote the real polynomials of degree
there is no possibility for confusion, I'll just
below, you are asked to prove that
P2n is a real vector space of dimension
n+2
2
Example 1.1.1. (a) A basis for
dim(P2 ) = 6.
3
(b) For P , the
dimension is 10).
P2
is given by
{1, x, y, xy, x2 , y 2 },
so
basis is
{1, x, y, xy, x2 , y 2 , x3 , x2 y, xy 2 , y 3 }
(and the
For a con guration of distinct points
S = {p1 , . . . , pm } R2 ,
de ne a linear evaluation map
evSn : P n Rm
via
f (p1 )
.
.
.
f 7
f (pm )
Recall from linear algebra the
Rank + Nullity theorem:
dim(image(evS )) + dim(ker(evS )) = dim(P n ).
|
{z
}
|
{z
}
(1.1.1)
rank
nullity
Definition 1.1.2. The con guration of points
evSn
is surjective (onto).
Now if
evSn
is surjective, its rank is
space of polynomials vanishing on
S.
m;
while its kernel is just the
By (1.1.1) we have:
n polynomials vanishing
n+2
dimension
m.
2
Proposition 1.1.3. The space of degree
on a general con guration of
1or
S is called n-general 1
points has
just general or generic if the context is understood
10
1. TWO THEOREMS ON CONICS IN THE PLANE
What does all this have to do with algebraic curves? Well, given a
polynomial in
Pn ,
I can look at its solution set in
2
we have the assignment
P n \P n1
{degree n
R2 .
More precisely,
real a ne algebraic curves}
given by
f 7 Cf := {(x, y) R2 | f (x, y) = 0}.
Notice that
f ker(evS ) Cf S.
(1.1.2)
Proposition 1.1.4. Through ve (2-)general points
in the real plane, there exists a unique conic
Proof.
Q.
6 5 = 1.
OK, OK, so what the proof means is:
of degree
A, B, C, D, E
by Prop.
1.5, the space
polynomials vanishing on a general con guration of
2+2
2
5=1. So given two degree-2 polynomials
f, g P 2 vanishing at all 5 points in S , we must have g = a f (for
some a R). But f = 0 and a f = 0 de ne the same curve Q; and
by (1.1.2) Q contains A, B, C, D, E .
There are two issues with this. First, in order for f to de ne a conic,
we need to know that the 1-dimensional space of solutions doesn't lie in
P 1 (inside P 2 ). But if a linear (degree 1) polynomial h vanishes at all 5
2
2
2
points, so does h ; and then since h, h P are not linearly dependent,
evS2 doesn't have maximal rank, contradicting the genericity condition
on S . (At this point, Proposition 1.6 is completely proved as stated.)
points has dimension
The second problem is more interesting: what exactly does it mean
A, B, C, D, E to be 2-generic? If you think about our little proof, the
existence of Q has nothing to do with this genericity, but the uniqueness has everything to do with it. Moreover, the statement that Q is
unique if S is generic is somewhat circular without knowing (beyond
De nition 5) what the genericity condition is: namely, that no 4 of
for
the points are collinear. It's best to wait until we're doing projective
geometry to prove that, and so we will.
2notation: A\B
denotes set-theoretic exclusion, sometimes also written
A ne just means that the curves are in
R2 .
A B.
1.2. BLAISE PASCAL AND THE MYSTIC HEXAGON
11
Remark 1.1.5. More generally, the abstract ethos surrounding the
word general in algebraic geometry is that you are working in the
complement of nitely many algebraic conditions. The algebraic condition we are avoiding in this section is the vanishing of all
mm
n
minors in a matrix representing evS .
1.2. Blaise Pascal and the mystic hexagon
Similarly, given eight points
A, B, C, D, E, P1 , P2 , P3
in (3-)general position, the vector space of cubic polynomials vanishing
at all of them has dimension
3+2
2
8 = 2.
V , and write (xA yA ), . . . , (xP3 , yP3 ) for the coor3
points. V lies in P , which consists of elements of the
Call this vector space
dinates of given
form
a1 + a2 x + a3 y + + a9 xy 2 + a10 y 3 = fa (x, y).
Asking a polynomial of this form to vanish at our
points yields the
equations
0 =
a1 + a2 xA + a3 yA + + a9 xA yA2 + a10 yA3
.
.
.
i.e.
0 = a1 + a2 xP3 + a3 yP3 + + a9 xP3 yP2 3 + a10 yP3 3
linear constraints on the
10
variables
{aj }
expressing what it
fa to lie in V .
Now let A, B, C, D, E be (2-)general and Q the unique conic through
means for
them:
B
Q
C
A
D
E
In fact, we shall make the stronger assumption that no three of
Q is smooth (an ellipse or hyperbola, not a pair of
We would like to construct (arbitrarily many) points on Q using
are collinear, so that
lines).
A, B, C, D, E
12
1. TWO THEOREMS ON CONICS IN THE PLANE
only a straightedge. Start by drawing secant lines connecting adjacent
points:
B
Q
C
A
D
BC
E
AB
CD
DE
P1
AB DE =: P1 .3 Next, draw any line ` through
through B, C, D, or E , and set ` CD =: P2 .
where we have labelled
that does not pass
C
A
D
BC
E
AB
l
CD
DE
P
1
P
2
(Note that our choice of
constructing.) Next, draw
3Of
q Q we
P1 P2 BC =: P3 .
will determine the point
P1 P2 ,
and label
end up
course, the lines may be parallel. There are two ways to x this: either make
A, B, C, D, E
more general so that none of the lines we intersect are parallel; or
work projectively. Since this chapter is entirely motivational we won't worry about
that level of detail..
1.2. BLAISE PASCAL AND THE MYSTIC HEXAGON
13
C
A
D
BC
E
AB
l
CD
DE
P P
1 2
P
1
Finally, we draw
P
3
P
2
EP3
and set
q := EP3 `.
C
A
D
q
BC
AB
l
CD
EP3
DE
P
1
P
3
P
2
Proposition 1.2.1.
q Q.
To prove this, we shall require:
Lemma 1.2.2.
when
A, B, C, D, E
A, B, C, D, E, P1 , P2 , P3
is in (3-)general position,
are general in the strong sense assumed above.
We won't prove the lemma here. (It is a special case of something
called the Cayley-Bacharach Theorem, which will be easy to prove once
we know a little about residues in algebraic geometry.)
fAB (x, y) = 0 for the (linear)
AB , fQ (x, y) = 0 for the conic Q, and so forth.
Proof. [of Prop.
equation of the line
1.8] We write
Consider the three cubic polynomials
f1 (x, y) := fQ (x, y) fP1 P2 (x, y),
f2 (x, y) := fAB (x, y) fCD (x, y) fEP3 (x, y),
f3 (x, y) := f` (x, y) fBC (x, y) fDE (x, y).
14
1. TWO THEOREMS ON CONICS IN THE PLANE
Their vanishing sets are
C1 := Q P1 P2 ,
C2 := AB CD EP3 ,
C3 := ` BC DE,
each of which contains the set
S := {A, B, C, D, E, P1 , P2 , P3 }.
f1 , f2 , f3 belong to V := ker(evS3 ). Since the dimen2, f1 , f2 , f3 cannot be linearly independent and we have a
So by (1.1.2),
sion of
is
4
nontrivial relation
f1 + f2 + f3 = 0
(1.2.1)
with real coe cients.
Suppose
=0
in (1.2.1). Then
f2 = f3 ,
so that
f2
and
f3
are
proportional hence cut out the same curve:
AB CD EP3 = ` BC DE.
` = AB, CD or EP3 , which implies ` contains A, B, C, D,
E a contradiction to our choice of `!
So 6= 0, and we may rewrite (1.2.1) as
But this means
or
f1 = f2 f3 .
Now since ` and EP3 both contain q , f2 and f3 both vanish at q . By
(1.2.2), f1 also is zero at q , so one of its factors has to be. Therefore q
is contained in Q or P1 P2 .
Suppose q P1 P2 . Then the lines P1 P2 , EP3 , and ` collapse to
the same line (look at the last diagram) and so in particular E `,
again in contradiction to our choise of `.
We conclude that q Q.
(1.2.2)
In the construction described pictorially above,
Proposition 1.9 ultimately just the point where
have proved the beautiful statement:
4not
meets
Q.
Since
` was essentially free, A, B, C, D, E, and q can be thought
6 distinct but otherwise arbitrary points of Q. Consequently, we
our choice of
of as
was in light of
all of
, ,
are zero
1.3. PONCELET'S PORISM
15
Theorem 1.2.3. [B. Pascal, 1639] Intercepts of opposite edges
of a hexagon inscribed in a conic, lie on a line.
When we get to the notion of duality in projective geometry, Theorem 1.10 will dualize for free to:
Corollary 1.2.4. The (three) lines joining opposite vertices of a
hexagon circumscribed about a conic, meet in a single point.
1.3. Poncelet's Porism
According to my dictionary, a porism is a proposition that uncovers
the possibility of nding such conditions as to make a speci c problem
5 The result of Poncelet I'll describe
6
here just had a whole book devoted to it, and in the late 1970's P.
capable of innumerable solutions.
Gri ths (my Ph.D. advisor) and J. Harris devoted two nice articles to
it.
If your local pub had an ellipse-shaped pool table it would even
have practical applications.
C and D be two conics in R2 . They are the vanishing sets
of some fC , fD P2 . For simplicity, assume they are ellipses, with D
contained in the interior of C .
So let
Problem: Does there exist a closed polygon (self-intersections are
OK) inscribed in
and circumscribed about
D?
Solution: Sometimes. But existence of one such polygon implies
that there are in nitely many.
This is clearly a porism.
We shall call such polygons as in the
Problem circuminscribed when a speci c pair
C ,D
is understood.
The precise statement is:
Theorem 1.3.1. [Poncelet, 1822] If the pair
circuminscribed polygon, then for any point on
scribed
n-sided
polygon with one vertex on
C, D has an n-sided
there is a circumin-
C.
Another way of putting this is that circuminscribed polygons can
be rotated continuously: there are beautiful pictures of this at http://
5rather than a complication of the swine
6L. Flatto, Poncelet's Theorem, AMS,
u
2009.
It's on reserve in the library and
aimed at advanced undergraduates, i.e. you. Highly recommended.
16
1. TWO THEOREMS ON CONICS IN THE PLANE
[Link]/intgeo/[Link].
While this
picture
x0
START
etc.
L0
Poncelet
Construction
D
x2
L1
x1
is to those ones as the math building (at virtually any university) is
to Durham cathedral, it allows us to characterize Theorem 1.12 in one
more way:
(1.3.1)
If
xn = x0
for any
x0 ,
then
xn = x 0
for every
x0 .
We will skirt around projective geometry in explaining the idea
here, but can't avoid
solutions to
fC = 0
C.
Henceforth,
and
fD = 0
and
that is,
shall denote all complex
C, D C2 .
Topologi-
cally these are real surfaces (in fact, spheres with one or two missing
points), and are complex-analytically isomorphic to
or
C ,
but it
won't hurt to draw them schematically as real curves on a sheet of
paper. You can think of this as the real solutions standing in for the
complex ones. In general, when we want to see the topology of a complex algebraic curve, we'll draw a surface; when we want to see how
dierent curves intersect or how they lie in space, we'll draw a curve.
Consider the set of pairs
E := {(x, L) C D | x L},
where
is the set of lines tangent to
D.
An involution is a map
which, applied twice, gives the identity. Here are two involutions on
E:
1.3. PONCELET'S PORISM
17
1(x,L):=(x,L)
(1) =id.
x
and
2(x,L):=(x,L)
(2)
2
x
=id.
The composition
:= 2 1
takes
1
2
(x, L) 7
(x0 , L) 7
(x0 , L0 ) =: (x1 , L1 )
(1 , 2 and
are all maps from
to
E ).
More generally,
(xi , Li ) =: (xi+1 , Li+1 )
de nes the
ith
iteration of the Poncelet construction. The construction
starting from some
(x, L)
closes if and only if
n (x, L) = (x, L)
(1.3.2)
for some
n.
C and D aren't tangent
that is, C and D are in general
Thinking of complex points, and assuming
anywhere and don't meet at in nity
position in some sense they meet in exactly four points. This is a
rst taste of Bezout's theorem, which we will prove carefully later. Now
let
x C:
tangent to
x
/ C D, there are exactly two lines containing x and
D; if x C D, there is only one. So we nd that the
if
projection
E
C
(x, L) 7 x
is a two-sheeted branched covering with four branch points:
18
1. TWO THEOREMS ON CONICS IN THE PLANE
(x,L)
(x,L)
E is an elliptic curve, hence isomorphic to
that : E E is a translation
This turns out to mean that
C/lattice.
One deduces
C/ C/
u 7 u + ,
(x, L) is some point u0 . Whether or not n (u0 )
do with whether n , and nothing to do with
and our starting point
u0
has everything to
the choice of
u0 ,
and so (1.3.1) follows.
We will see a more in-depth treatment of this after studying elliptic
curves, including an algebraic criterion for deciding when
a given
C, D,
n =
id. for
and an analysis of elliptical billiards. For now, here are
some examples of Poncelet in action:
Example 1.3.2. (a)
C = {x2 + y 2 = 1}, D = { xa2 +
y2
1a2
n = 4.
C
L0
x1
x0=x4 =(a, 1a2 )
D
L1
x2
(b)
L3
x3
L2
4x
C = {x2 + y 2 = 1}, D = { (1+T
+
)2
4y 2
(1T )2
= 1}, n = 3.
= 1},
EXERCISES
19
1T
2
1+T
2
Exercises
(1)
P2n
n+2
. [For a more challenging
2
is a vector space of dimension
problem, try it with polynomials in
variables instead of two
obviously with a dierent answer.]
(2) Consider the two conics
r2 }.
C = {x2 + y 2 = 1}
and
The corresponding Poncelet elliptic curve
D = {x2 + y 2 =
E is singular (see
De nition 2.9), which means that the problem degenerates and is
solvable by hand using secondary school maths. For which
between
and
1)
does the
th
n,
(say,
iterate of the Poncelet construction
(starting at an arbitrary point on
(Hint: for each
C)
return to the starting point?
there are nitely many; just nd how many, and
the equation they must satisfy.)
CHAPTER 2
Riemann surfaces and algebraic curves
In this chapter we will de ne (complex) algebraic curves (represented by C ),
1 complex
1-manifolds
(represented by M ), and Rie-
mann surfaces, and start to consider under what additional hypotheses
they are equivalent concepts.
2.1. Algebraic curves
Definition 2.1.1. Let
gree
m in x, y (the 2
S2m
denote homogeneous polynomials of de-
2 These are polynomials
stands for 2 variables).
of the form
fm (x, y) =
cjk xj y k ,
j,k0
j+k=m
that is, each term has total degree
More generally,
mials in
Skm
m.
is the space of
S2m is a subset of P2m .
degree-m homogeneous polynoClearly
variables that is, linear combinations
ci1 ,...,ik xi11 xikk
i1 ,...,ik 0
i1 ++ik =m
m. Elements f Skm have the propfm (x1 , x2 , . . . , xk ) = m fm (x1 , x2 , . . . , xk ). (See exercise
of monomials with total degree
erty that
3 below.)
Given a real a ne algebraic curve of degree
C := {0 = f (x, y) = fd (x, y) + fd1 (x, y) + + f0 } R2
1in
the a ne resp. projective plane. Later we will de ne algebraic curves in higher
dimensional projective spaces and products thereof, but the most intrinsic de nition of an algebraic curve as a
1-dimensional reduced scheme (some authors require
this to be irreducible and over an algebraically closed eld as well) is probably
something to learn only once you have a rst course in algebraic geometry under
your belt.
2the
eld of de nition, from which the
cjk
is a vector space over that eld.
21
are taken, will depend on context;
S2m
22
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
with
fd
not identically zero, we would like to count its intersections
with a real line given parametrically by
L : t 7 (t, t)
(where
are real constants). These are just the solutions of
0 = fd (, )td + fd1 (, )td1 + f0 .
(2.1.1)
Naively, we would like to get
points:
#{C L} = d ??
(2.1.2)
Some issues arise . . .
Problem
(a)
Solution
is not algebraically closed!
(b)
solutions at in nity!
(c)
multiple roots!
(d)
might contain
pass to
C2
add a line at in nity to
C2
count intersections with multiplicity
L!
Uh-oh
Each Problem is an obstruction to (2.1.2), and the object of each
Solution is to remove the obstruction.
In a little more depth, (a) says that in spite of the fact that the
fi (, ) R, roots of (2.1.1) can be non-real. So we had better consider
C and L as complex algebraic curves take x, y C in the de nition
of C and t C in the de nition of L. Their dimensions over R then,
of course, double, and C L now contains the points corresponding to
non-real roots of (2.1.1).
1
d
Next, if we plug t = s
into (2.1.1) and multiply by s , then it
becomes
0 = f0 sd + f1 (, )sd1 + + fd (, ).
(2.1.3)
to (2.1.1) is a solution at 0 to (2.1.3), which exists if
and only if fd (, ) = 0. For this to be counted in C L, we must add a
2
2 3
line L to C to get P , the complex projective plane (which we shall
discuss in a moment). This adds a point to L at corresponding to
s = 0, yielding a P1 (projective line), and adds points at in nity to C
A solution at
(yielding a compact projective curve).
3called CP2
or
P2C
in some books
2.1. ALGEBRAIC CURVES
We know that to get
23
d solutions out of a degree d polynomial equa-
tion you have to count a twice repeated root as two solutions. So to
d intersection points you will certainly have to count intersections
of C with L however many times the corresponding root of (2.1.1) is
repeated. The curious case is an m-times repeated root at in nity: via
(2.1.3), this corresponds to fd (, ) = = fdm+1 (, ) = 0. In that
case, (2.1.1) is only in fact of degree d m. One does have to worry
get
about these degenerate cases, but they will look completely natural in
projective coordinates.
Finally, if all
f` (, ) = 0,
then
C L
and we are in trouble
there is no way around (d). We shall have to demand that plane
curves intersect properly (in points only), disallowing this possibility,
in order to make any statement about the number of intersection points.
Example 2.1.2. Given a quintic curve
as in the following picture
L
8
(5,5)
the number of intersection points in
(i,i)
(1,1)
R2
is only
2.
But the number of
complex intersection points, counting multiplicities and intersections
at in nity, is
So .
in nity to
5.
how does one go about adding a line (resp.
(resp.
L)?
First, visualize
as
point) at
24
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
i
1
and think of all arrows as going o to the same point. Adding this point
C {},
resulting in a sphere
gives the 1-point compacti cation
1
circle
unit
0
This is an informal way of thinking of
P1
Definition 2.1.3. Projective space
through the origin
in the following:
Pn
is the set of complex lines
Cn+1 .4 More precisely,
Pn := D
(Cn+1 \{0})
(z0 ,z1 ,...,zn )(z0 ,z1 ,...,zn )
C
consists of nonzero vectors in
Cn+1 ,
modulo the equivalence relation
equating all vectors lying on a complex line.
Elements are written
[z0 : z1 : : zn ].
For
n=1
this yields the projective line
phism
P1 ,
which has the isomor-
P1
C {}
[z0 : z1 ] 7 zz01 =: z
4as will be proved in Chapter 5, one should really think of Pn
(but we haven't de ned these yet).
denotes
(0, 0, . . . , 0).
as a complex manifold
2.1. ALGEBRAIC CURVES
25
given by taking slope (of the line represented by
is well-de ned as
[z0 : z1 ] 7
z1
.
z0
[z0 : z1 ]).
This map
The honest topological
is
picture of
z1
z0
i
1
circle
unit
0
while the schematic real picture is
[0:1]
Next, setting
0
n=2
[1:0]
=
we have the projective plane
P2 ,
and the iso-
morphism
P2
if z0 6=0
[z0 : z1 : z2 ]
(2.1.4)
if z0 =0
expresses how
P2
1
(C C)
P
z1 z2
,
C2
z0 z0
[z1 : z2 ] P1
adds a line at in nity (the
P1 )
to
C2 .
For
P2 ,
the
(rather bad, but standard) schematic picture is
[Link]
L , i.e.
z0=0
z1=0
all 3
P1s
[Link]
[Link]
z2=0
While I'm not going to try to represent
real dimensions on paper,
here is a mostly honest topological depiction of the
3 P1 's:
26
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
z0=0
z1=0
z2=0
Remark 2.1.4. Equation (2.1.4) relates a ne coordinates (on
C2 )
P2 ). Instead of the {zi }, I will frequently
X Y
2
use [Z : X : Y ] for a point in P and (asuming Z 6= 0)
,
=: (x, y)
Z Z
2
for the corresponding point in C .
2
Also, a warning is in order: [0 : 0 : 0] is not a point in P . With
and projective coordinates (on
homogeneous coordinates, some entry must be nonzero.
Returning to our degree
{f (x, y) = 0} C
described above?
C=
P2 as
(and now complex) algebraic curve
, what happens to it as we compactify
to
To treat this, we rst need to introduce the main
object of study of this course.
Since
equation
[Z : X : Y ] = [Z : X : Y ], in order for a
F (Z, X, Y ) = 0 to make sense projectively (i.e.
polynomial
in
P2 ),
we
must have
(2.1.5)
F (Z, X, Y ) = 0 = F (Z, X, Y ) = 0 ( C ).
This condition is guaranteed by homogeneity of
in De nition 2.1).
homogeneity of
(cf.
(In fact, as we shall see later it is equivalent to
F .)
C P2
F S3d .
Definition 2.1.5. A projective algebraic curve
d,
the property
is the zero set of a homogeneous polynomial
of degree
Here, then, is a general procedure for going between a ne and
projective curves:
f (x, y) = 0 7 Z f
(2.1.6)
X Y
,
Z Z
corresponds to taking the projective closure
curve
CC
. Conversely, if the given
=0
C P2
of a given a ne
is already projective (de ned
2.1. ALGEBRAIC CURVES
by
F = 0),
27
then
F (Z, X, Y ) = 0 7 F (1, x, y) = 0
(2.1.7)
restricts
to the a ne curve
C C2 .
Given an a ne curve, taking
closure then restricting gets you back to where you started.
Example 2.1.6. Starting from the homogeneous cubic polynomial
F (Z, X, Y ) = ZXY +3Z 2 Y +4Y 3 , the a nization is f (x, y) = F (1, x, y) =
xy + 3y + 4y 3 . Conversely, if we start from f (x, y) = x3 y y 2 + 2x, the
X Y
4
3
2 2
3
projectivization is F (Z, X, Y ) = Z f ( , ) = X Y Y Z + 2XZ .
Z Z
Take
F, C
deg Fi ),
Q
F = Fi (so that deg F =
set of Fi , we have C = Ci .
to be as in De nition 2.5. If
then writing
Ci
for the zero
Definition 2.1.7. We say that
no proper (deg
1)
is irreducible if and only if
has
homogeneous factors.
Now let's consider our intersection problem (2.1.2) once more, in the
complex projective setting. Referring to the discussion up to Example
m-fold
intersection at in nity, then the degree
of the polynomial in (2.1.1) is
d m. The Fundamental Theorem of
has d m complex roots counted with
2.2, if
and
have an
Algebra then says that (2.1.1)
multiplicity, and we de ne these to be the intersection multiplicities for
and
in
C2
as indicated in our discussion. We have proved a baby
version of Bezout's theorem:
Proposition 2.1.8. Let
L P2
be a (projective) line in
P2 ,
i.e. an
algebraic curve of degree one. A projective algebraic curve of degree
in
not containing
L,
meets
in
points counted with multiplicity.
In proving this result we did a tiny bit of complex analysis on
L, so
were implicitly using its structure as a complex 1-manifold. In general it
is quite useful to be able to do analytic computations on curves, but not
all irreducible algebraic curves are complex manifolds (at least, without
doing something to them called normalization). The obstructions are
called singularities and will be explored in greater depth later.
For
now, we will just give a de nition and a few examples.
Definition 2.1.9. A singularity or singular point of an a ne al-
gebraic (plane) curve
f (x, y) = 0
is a point in
C2
where
f,
f
, and
x
28
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
f
f
f
are all zero that is, a point on the curve where
and
vany
x
y
ish. A singularity of a projective algebraic curve
F (Z, X, Y ) = 0
is a
F
F F
,
, and
are all zero. A curve with one or more
point where F ,
X Y
Z
singular points is called singular ; a curve with none is called smooth.
Example 2.1.10. Here are some local real (schematic) pictures of
plane curve singularities:
cusp
An example of a cusp is the point
on
x3 = y 2 );
[1 : 0 : 0] on X 3 Y 2 Z = 0 (or (0, 0)
XY = 0 has
at [1 : 0 : 0]:
the curve
or normal crossing)
ordinary
triple point
ordinary
double point
an ODP (ordinary double point,
X=0
Y=0
Now, this is just two
P1 's
(namely,
X=0
and
Y = 0)
touching at one
point. A more (though not completely) topologcially honest picture
of this is:
X=0
Y=0
which makes it apparent that an ODP is actually a bi-conical singularity.
2.2. COMPLEX
1-MANIFOLDS
29
Before we pass to the analytic side of our story, there are 2 more
facts about homogeneous polynomials worth a quick mention.
First,
d
d
the map S3 P2 given by F (Z, X, Y ) 7 F (1, x, y) is an isomorphism,
d
d
so dim(S3 ) = dim(P2 ) in particular. Second is the Euler formula
N
X
(2.1.8)
i=0
Zi
F
=dF
Zi
for
d
F SN
+1
which will be used in later chapters (cf. Chapter 6 for a proof ).
2.2. Complex
1-manifolds
Recall from basic point-set topology that a topological space is a set
together with a collection
{UI }I
of open sets containing
X,
the
empty set, and all unions and all nite intersections of its members.
(Here
is some typically huge index set. A base for the topology of
is a sub-collection of the
unions, and
{UI }I
which generates it under taking
is said to be second countable if it has a countable base.)
is called Hausdor if points can be separated: i.e. given
there exist disjoint open sets
and
containing
and
q,
respectively.
In topology, a homeomorphism is a continuous, 1-to-1, open
5 map.
U 3 p that are homeomorphic
n
to R (or equivalently, an open ball in R ) these are called open
neighborhoods of p. If these always exist, we say X is locally homeon
morphic to R . A second countable, Hausdor topological space that
n
is locally homeomorphic to R , is called a real n-manifold.
In the case n = 2, we are going to layer complex analyticity onto
Given a point
p X,
and
we like open sets
this construction:
Definition 2.2.1. A complex
1-manifold
consists of
(i) a connected Hausdor topological space
{U } of M (this is a nite set of
{UI }, such that U = M ); and
(ii) an open cover
from amongst the
5a
map
is open
M;
open sets taken
is open (resp. continuous) if the image (resp. preimage) of any open set
30
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
(iii) mappings
z : U C
that are homeomorphisms onto their
image, such that the transition functions
:= z z1 : z (U ) z (U )
are biholomorphic (i.e., analytic isomorphisms).
The
{z }, { }
are called local coordinates, and the
ing) functions ; the entire collection
atlas. The functions
are key:
transition (or patchis called an analytic
is an complex analytic manifold
because they are complex analytic. If in (iii) we replace
by
Rn
and
require the transition functions to be smooth (i.e., have continuous partial derivatives of all orders), then
dierentiable) real
n-manifold
would have been a smooth (or
instead.
If we think of the (complex analytic) transition functions in De nition 2.11 as maps from
R2
to
R2 ,
then
( |{z}
x ,
y
) = ( u(x, y) , v(x, y) )
|{z}
| {z }
| {z }
real
part
is smooth and
u, v
imag.
part
real
part
imag.
part
satisfy the Cauchy-Riemann equations. These may
be expressed in terms of the Jacobian matrix
ux uy
vx vy
6if U
and
V .
ux vx
vx ux
!
,
U := U U .
goes from V
are distinct open sets in our open cover, we write
Leter we will write
to
for
z (U )
and
for
z (U ),
so that
1-MANIFOLDS
2.2. COMPLEX
which consequently has positive determinant:
is
Example 2.2.2.
of complex
u2x + vx2
(obviously
biholomorphic. Therefore
7
orientable as a real 2-manifold.
cannot equal zero since
orientation, so
31
is
0)
preserves
C, C , H, P1 , and C/ are (the simplest) examples
1-manifolds.
For the rst three, producing an analytic atlas
is trivial (since you only need one
U ),
and we will do this below for
the latter two.
Now assume
nite subcover.
is compact, that is, every open cover has a -
(In fact, since a complex
1-manifold
always admits
a metric, compactness is equivalent to every sequence of points hav-
an = n has no limit in C but
[0 : 1] in P1 , which is compact.)
ing a convergent subsequence. Clearly
an = [1 : n] =
[ n1
Then viewed over
2-manifold.
: 1] does limit to
R, M is an orientable,
compact, connected, smooth
By a theorem in topology, this means that
morphic to a sphere with
handles, and we say
is homeo-
has genus
g:
etc.
g=0
g=2
g=1
g 0 occur for complex manifolds; we'll show this
g = 0, 1. To do complex analysis on M , you can use
It is a fact that all
in a moment for
the local coordinates, but for some purposes it is also convenient to cut
7in
into a simply connected region, e.g.
fact, a matrix of the form
A
B
B
A
is a rotation times a dilation, hence
preserves angles that is to say, under the assumption of the CR-equations,
is
conformal. So a complex 1-manifold is essentially a dierentiable real 2-manifold
with conformal transition functions.
32
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
g=1
g=2
Extrapolating from this, one sees that if you begin with a sphere with
handles, the cut-open version is a polygon with
4g
edges identi ed
in pairs. (The black points on the right-hand side are
all
identi ed.)
8
Using this we can do a quick computation of the Euler characteristic
of
M:
M := faces edges + vertices = 1 2g + 1 = 2 2g.
(2.2.1)
g = 0)
Example 2.2.3. (
nates
[X : Y ].
Let
Consider the open
[0:1]=" "
M := P1 with homogeneous coordicover {U0 , U1 } of M given by:
U0
[1:0}="0"
U01
M
U1
That is,
U0 = M \{[0 : 1]} and U1 = M \{[1 : 0]}.
8traditionally
lation of
M,
For local coordinates,
one would use the numbers of faces, edges, and vertices in a triangu-
but using a polygonal decomposition like this is also OK
2.3. RIEMANN SURFACES
33
we take
z0 :
U0
C
Y
[X : Y ] 7 X
z1 :
U1
C
.
[X : Y ] 7 X
Y
and
Writing
C C.
U01 := U0 U1 ,
z0 (U01 ) = C C and z1 (U01 ) =
(which goes from z0 (U01 ) to z1 (U01 )
we have
The transition function
by de nition) is then
10 : C C
.
u 7 u1
g = 1) Let 1 , 2 C be linearly independent
over R. Then := Z h1 , 2 i = Z1 + Z2 is a lattice, and we set
M := C/. (This means that z, z 0 C give the same point in M if
0
and only if z z .) We endow M with local coordinates on the
Example 2.2.4. (
neighborhoods shown
covering by coordinate
neighborhoods
),
basically by using the coordinate on
(before quotienting by
nd that the transition functions
are all either the idenitity or
translation by some
C
ij
Topologically,
is a torus (cf. the
and
g=1
pictures above), which is evident from performing the identi cations
on the sides of a fundamental region for
C/
as shown.
2.3. Riemann surfaces
Traditionally, a Riemann surface
is a compact complex
1-manifold
obtained as the existence domain of an algebraic (typically multivalued) function over
P1 .
That isn't the de nition I'll use here, but I do
want to explain the concept.
34
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
For example, given distinct complex numbers
tion
i , the algebraic func-
v
u2g+2
uY
F(z) := t
(z i )
i=1
on
can be made single-valued on a complex manifold
(of genus
g,
P1 , cut identical
j = 1, . . . , g + 1. Then
it turns out) constructed as follows: On two copies of
nonintersecting slits from
glue the two copies of
which
2j1
to
2j
for
together on these slits, forming a set on
F becomes single-valued;
nally, endow this set with an analytic
atlas to get a complex manifold M . This manifold has a distinguished
1
morphism M P presenting it as a nite branched cover of the
projective line. We won't do this explicitly here especially endowing
it with an analytic atlas, since that is really a special case of normalizing
an algebraic curve (cf.
3.1).9
Instead, let's visualize what a couple of existence domains for
algebraic functions look like, starting with the Riemann surface of
1
(w =)z 3
over the unit disk. This is some object tting (as {z
= w3 })
into the following picture:
2 i
3
{z=w3}
3
4 i
e3
wdisk
3 2
1
zdisk
To construct it, think about following
z3
around the disk once coun-
terclockwise: when you reach your starting point the function has become
2i
3
times the branch of
more, you get
9it
4i
3
1
3;
z3
you started with; going around once
and one more time gets you back to your original
makes a very instructive exercise though, and the next example gives a hint on
how to do it
2.3. RIEMANN SURFACES
branch.
35
So taking three unit disks, slitting them along the positive
reals, and gluing them as indicated
III
II
"z1/3"
"e 3 z1/3"
"e 3 z1/3"
we get the parking lot
{z=w3}
zdisk
(The green segments are glued but I can't draw in 4 dimensions.) An
easier way to visualize this Riemann surface is this: it's just the
disk. The di culty is in seeing the
w-disk
over the
w-
z -disk.
Next, let's construct an existence domain for
p
(z a)(z b)(z c)
F(z) =
over
P1 .
In a neighborhood of
surface of
(z z0 )
1
2
z0 = a, b, c
this looks like the Riemann
over a disk, which is the same as the construction
we just did except with 2 unit disks instead of 3. Indeed, going once
around
a, b,
or
takes
F 7 F;
and furthermore, because the degree
of the polynomial under the square root is odd, going once around
does the same thing. Since
is equivalent to
going around two points at once gives no change. So taking two
P1 's
36
2. RIEMANN SURFACES AND ALGEBRAIC CURVES
and cutting and pasting them as indicated, we end up with a manifold
of genus
on which
becomes well-de ned:
"open" the
cuts and
join
"+ f "
" f "
8
8
c
a
(In the picture,
and
are called
1-cycles;
there just there to make
the topology clear.) The same construction works if we replace
by
p
(z a)(z b)(z c)(z d),
with
replacing
F(z)
In fact, by a deep result (on existence of nonconstant meromorphic
functions on complex 1-manifolds) any compact complex
1-manifold
is an existence domain of the sort we have just discussed: they are
equivalent objects in the end. The following is motivated by this, and
the desire to keep things simple:
Definition 2.3.1. A Riemann surface is a compact complex
1-
manifold.
Exercises
(1) Take projective closures of
and
L := {x = 0}
in
P2
C := {y 2 = (x 1)(x 2)(x 3)(x 4)}
( nd associated homogeneous equations),
and determine all intersections and their multiplicities (give the
projective coordinates of the points).
What is the sum of multi-
plicities?
Z03 + Z13 + Z23 = Z0 Z1 Z2 .
degree 3 is a scalar, not a
(2) Find the a ne equation associated to
(This equation is homogeneous of
coordinate).
EXERCISES
(3) Let
37
F be a polynomial in 3 variables. Prove that F (z0 , z1 , z2 ) =
= F (z0 , z1 , z2 ) = 0 ( C ) forces F to be homo-
geneous (of some degree). [You will have to assume the following
g (in (z0 , z1 , z2 )), with vanish{f = 0} {g = 0}, then f divides a power of g . This
result: given two polynomials
ing locus
and
is called Study's lemma and will be proved later.]
challenging, show the converse. If you like, do both.
If this is too
CHAPTER 3
The normalization theorem
We state (but do not yet prove) the promised relationship between
algebraic curves and Riemann surfaces, and explain how to work it out
directly for conics. To state the general relationship, however, we need
the notion of meromorphic functions on a Riemann surface, so we will
rst de ne and prove a few results about those.
3.1. Meromorphic functions on a Riemann surface
M
{(U , z )}
Let
be a Riemann surface (De nition 2.15) with analytic atlas
(De nition 2.11), and write
analytic chart
: V U ( M )
V := z (U ) C.
is simply de ned to be the (com-
position) inverse of of the local coordinate
z 1
The local
z .
(I just don't like writing
since in some settings this is easy to confuse with
Definition 3.1.1. A meromorphic (resp.
1
.)
z
holomorphic) function
f K(M ) (resp. O(M )) is a collection of continuous maps f : U
P1 such that
the {f } agree on overlaps (viz., f = f on U ), and
f is a meromorphic (resp. holomorphic) function, in the
sense of complex analysis, for all .
1
U
f
39
40
3. THE NORMALIZATION THEOREM
Remark 3.1.2. (a) One really works with functions of the coordi-
z ,
nate
i.e. the function
f =: g
(mapping
V P1 ),
and then
the compatibility condition reads
g = g .
(3.1.1)
(b)
K(M )
is a eld, since you can multiply, add, and invert (addi-
tively and multiplicatively) meromorphic functions.
Proposition 3.1.3. [Liouville's Theorem]
O(M )
=C
compact
(constant functions).
f O(M ) = f (M ) (P1 \{}) =
C; while on the other, M compact and f continuous = f (M ) is
compact. Applying absolute value gives a compact subset |f (M )|
R0 . This has a maximum element, whch is assumed at some point
p M , and this p lies in some U . Hence, the absolute value of
the holomorphic function g = f attains a maximum on V (at
(p)), and by the maximum modulus principle, g (and thus f ) is
some constant c C.
Let U be any open set of the atlas meeting U . Since f = f = c
on U , and U has accumulation points, f = c on U . One continues
this argument now for any open set meeting U or U , and so forth.
By connectedness of M , this shows f = c on all open sets of the atlas,
hence on all of M .
Proof. On the one hand,
Definition 3.1.4. Let
any
p M, f
f K(M )
be a meromorphic function. For
is locally of the form
z m h(z)
(3.1.2)
m Z, z
p (i.e. z(p) = 0), and
h(z) a local holomorphic function of z with h(0) 6= 0.1 We say that the
order p (f ) of f at p is m.
with
a local coordinate vanishing at
With this bit of language it is easy to compute the meromorphic
function eld for Riemann surfaces of genus
Theorem 3.1.5. (a)
1To
then
be absolutely precise, if
K(P1 )
= C(z) (z
z
and
V.
1.
an indeterminate).
is a local coordinate on
is a holomorphic function on
understood.
U 3 p,
with
V = z(U ),
I'll frequently assume things like this to be
3.1. MEROMORPHIC FUNCTIONS ON A RIEMANN SURFACE
41
:= {m1 1 + m2 2 | mi Z} (1 , 2 C linearly
independent over R) for a lattice, K(C/)
= C(, 0 ) where (u) is
the Weierstrass -function for .
(b) Writing
Proof. (a) Referring to Example 2.13, write
for the two local coordinates.
I am really going
z = z0 and w = z1
to use z as a global
P1 ; the statement we want to prove is that meromorphic
1
functions on P are precisely the rational functions of z .
In one direction, this is easy: if P , Q are polynomials in z (with
P (z)
Q 6= 0), clearly Q(z)
is the restriction to U0 of a meromorphic function
coordinate on
on
P1
(on
U1 ,
it is
1
P(w
)
1 ).
Q( w
)
Conversely, are all meromorphic functions rational?
K(P1 ), p (f ) < 0
Given
2
at nitely many points zi (=
p), and we shall for simP
plicity assume none of these is the point . Let Pi (z) =
zi (f )k<0 ik (z
k
zi ) (sum is over k ) be the principal part of the Laurent expansion of
P
f at zi , and consider G(z) = Pi (z). Then f G O(P1 ) is constant
by Liouville; and since G is rational, we're done.
(b) Next, f K(C/) if and only if f is a doubly-periodic meromorphic function on C: that is, f (u) = f (u + m1 1 + m2 2 ) for all
m1 , m2 Z (also known as an elliptic function). We will see later that
these are generated (rationally) by
X
1
(u) := 2 +
u
1
1
(u )2 2
6=0
and its derivative.
Definition 3.1.6. A morphism (or holomorphic map)
of Riemann surfaces
3 is a collection
{U })
maps (agreeing on the
|z {F 1 (Ui )U }
F : U M
of continuous
such that the composition
is holomorphic for all
tion works more generally for complex
, i.
M M
zi F
(Note that this de ni-
1-manifolds
compactness is
inessential.)
2otherwise
1
1
f have an accumutaion point =
f
identically 0. (Also, note that I am identifying points by the value of the coordinate
z on P1 . If M were not P1 , I would write pi instead of zi .)
compactness
3write {U , z } and {U
i , zi }
4this composition renders z
zeroes of
for the atlases.
i as a function of
this sense)
(and is a local snapshot of
in
42
3. THE NORMALIZATION THEOREM
M
F
Ui
zi
F(U)
Vi
F (U ) U
i
U
z
the composition
(this can ramify!)
p (U )M and q (Ui )M with F (p) = q ,
zi (q) = 0, as shown in the above gure. Assuming
Now suppose we have
z (p) = 0 and
F is nonconstant, then after normalizing the local coordinates,5 we
have z
i (z ) = (z ) for some (unique) Z>0 . One says that f has
rami cation index at p (over q ). If this index is > 1, we say that f
is branched over q (or rami es at p).
= 3, we have already seen this picture in 2.3.
In general, for a holomorphic map of Riemann surfaces : X Y ,
1
for all but nitely many y Y the number |
(y)| is the same, and
this is called the degree of the mapping . (This will be explained in
greater depth in a later chapter.) The branch points of are just the
remaining points of Y . Usually we will just draw a schematic picture
Remark 3.1.7. For
like
Y
and it is understood that the picture is really as in
around the point on the base
cover
moves you between branches of the
X.
Proposition 3.1.8. Let
erally, a complex
1-manifold).
be a Riemann surface (or, more genThe holomorphic maps
cluding the constant map sending all points to
meromorphic functions
5see
2.3 so that going
Exercise 4 below
K(M ).
{},
M P1 ,
ex-
are simply the
3.2. RIEMANN SURFACES PARAMETRIZE ALGEBRAIC CURVES
Proof. Again refer to Example 2.13: given a morphism
P1
43
F :M
z0 F is holomorphic on the
, while z1 F = z0 F1 is
6
of the preimage of 0. Hence, F
(De nition 3.6), by de nition
complement of the preimage of
holomorphic on the complement
is meromorphic and
{F }
de nes a meromorphic function (De nition
3.1). The converse is even more tautological!
Later we will discuss morphisms (holomorphic maps) of complex
manifolds of any dimension. The following is a special case:
[Z0 : Z1 : : Zn ] for (projective) coordin
nates on P . A map from a Riemann surface M to P is called holomorphic if and only if all compositions [Zi : Zj ] are holomorphic
1
as maps to P on the open subsets of M where they are well-de ned.
Definition 3.1.9. Write
(M )
Remark 3.1.10. If the image
does not live in a coordinate
hyperplane, this is the same as saying that composing
coordinate on
and
Zj
with each a ne
gives a meromorphic function.
To see this, rst write
Zi
Mij
for the subsets of
where (under
are not both zero; these are the open subsets in the last
7
de nition. By Prop. 3.1.8, the conditions of Defn. 3.1.9 mean that
Zj
are meromorphic functions on the Mij . We need to show that the
Zi
Z
zj = Z0j extend to meromorphic functions on all of M . First, M is
p
/ M0j (i.e. Zj and Z0
vanish at p), we have a neighborhood U containing p where some other
Zi does not vanish, so that U Mij , Mi0 . Now, on U M0j we can
1
Zj
ZZ0i
write zj =
as a product of functions which are meromorphic
Zi
on all of U , hence showing that zj extends as desired.
covered by the open sets
{Zi 6= 0}.
Hence, for
3.2. Riemann surfaces parametrize algebraic curves
Here is the Normalization Theorem. We will prove part (A) in this
course.
C P2 ,
: M P2
Theorem 3.2.1. (A) Given an irreducible algebraic curve
there exists a Riemann surface
with
6The
holomorphicity of
and a holomorphic map
as its image which is 1-to-1 on
1 (C\sing(C)).
1
z0 F guarantees, in particular, that
poles and not essential singularities.
7Some, but not all, of these M
ij may be empty if
has only
(M ) is contained in a coordinate
hyperplane. We have excluded this degenerate possibility for simplicity.
44
3. THE NORMALIZATION THEOREM
(B) Given a Riemann surface
M,
there exists a holomorphic map
: M P2 such that
(M ) is an irreducible algebraic curve with sing((M )) consisting
of ordinary double points (or empty), and
is 1-to-1 o the preimage of these ordinary double points.
In this sense, irreducible smooth projective algebraic plane curves
are equivalent to, and are isomorphically parametrized by, Riemann
surfaces.
" "
If a curve
is not smooth, then the normalization desingularizes it
(and we shall see this quite explicitly later on). In either case, we say
that
is the normalization of
C.
Let's look brie y at the meaning of (B), which we will not prove
in this course.
For a given Riemann surface (i.e.
compact complex
M , it guarantees a holomorphic map to P2 , with image
(M ) = C =projective closure of {f (x, y) = 0}. Changing coordinates
2
on P if necessary, we may assume that C does not pass through [0 :
0 : 1]. So it makes sense to consider the composition
1-manifold)
2 {[Link]}
[Z:X:Y]
projection
which exhibits
[Z:X]
as a branched cover of
P1
existence domain of the algebraic function
or more precisely, as the
g(x)
obtained by solving
f (x, g(x)) = 0.
So Theorem 3.10(B) contains the statement that every complex 1manifold is an existence domain in the sense of
2.3.
3.3. STEREOGRAPHIC PROJECTION
45
We should also note that any Riemann surface admits a holomorphic embedding
: M , P3 ,
an even nicer result than part (B) above!
3.3. Stereographic projection
As a plausibility check on Theorem 3.10(A), we'd like a recipe for
normalizing conics i.e.
degree-2 (conic) curves
C P2 .
Given a
p C , and any line line ` through p, by Proposition 2.8 ` either
meets C in two points with mutliplicity 1 or in 1 point with multiplicity
point
2. Put dierently, we have either
` C = {p, q}
or
` C = 2p,
i.e.
` = Tp C
is the tangent line to
at
p.
(We will give a systematic treatment of tangent lines below.)
versely, given
and any other point
through them (and it doesn't meet
on
C,
Con-
there is a unique line
anywhere else).
There are two ways to think of why this gives a parametrization of
C . One possibility is to
p to project C onto it:
take a xed line (
=
P1 )
and use lines through
C
p
tan
gen
t
line
This is where the term stereographic projection comes from.
But this auxiliary projective line is super uous, because the family
p already gives a P1 . (Indeed this is close to the original
1
1
de nition of what P is.) We can parametrize this P by the slope of
Z Z
the line with respect to suitable coordinates (usually (x, y) = ( 1 , 2 )).
Z0 Z0
of lines through
The upshot is that we get a 1-1 correspondence between lines through
p and points on C , so that we are in the situation of 3.2 with M
= P1 .
Example 3.3.1. Suppose we wish to nd a parametrization
P1 C
{X 2 + Y 2 = Z 2 } P2 , which in a ne coordinates is
x2 + y 2 = 1. We choose a point on C , say p = (1, 0), and draw lines
y = (x1) through p. (The slope here is , and this should be viewed
1
2
2
as a choice of coordinate on P .) Substituting into x + y = 1 and
of the conic
46
3. THE NORMALIZATION THEOREM
solving for
in terms of
we have
x2 + 2 (x 1)2 = 1
=
(x 1){(1 + 2 )x + (1 2 )} = 0.
=
Ignoring the solution
x=
(2 + 1)x2 22 x + (2 1) = 0
x=1
2 1
,
2 + 1
p), we have
2 1
2
1
=
.
2 + 1
2 + 1
(which corresponds to
y=
Hence, we nd
() =
2 1 2
,
2 + 1 2 + 1
.
One can also do stereographic projection to construct normalizations of singular cubic curves:
p
P
The idea here is to consider lines through the singular point
any such
` already meets C
point (by Proposition 2.8).
twice, it will only hit
p;
since
in one additional
You'll work an example in the exercises
below. This will not work for a smooth cubic.
Exercises
m 7 (x(m), y(m)) (hence an isomorphism
P C ) of the smooth conic curve C that is the projective closure
2
2
of 3x y = 1. (You may work in a ne coordinates.)
Show that for any RS M and meromorphic function (0 6=) f
P
K(M ), one has pM p (f ) = 0. [Hint: Use the residue theorem
(1) Give a parametrization
(2)
from complex analysis.
Cut open the RS as in Chapter 2, and
df
along the boundary.]
integrate
f
(3) Convince yourself that the order
on a RS
p (f )
of a meromorphic function
(De nition 3.1.4) is independent of the choice of local
coordinate.
(4) Prove the following, which was claimed in De nition 3.1.6: Given
M, M 0
Riemann surfaces with a holomorphic map
f : M M0
EXERCISES
47
f (p) = q ). Then there exist (U, z) on M and (V, w) on M 0
satisfying z(p) = 0 = w(q), such that w = z (for some N) is
the local form taken by f near p. [Here for example (U, z) means
an open disk U M with local coordinate z : U C.]
1
2
2
Find a parametrization P C of the singular cubic Y Z X Z +
X 3 = 0 in P2 . (C has an ordinary double point p at [Z : X : Y ] =
[1 : 0 : 0]. Check that this point is indeed a singularity of C .) To
do this, convert to a ne coordinates, substitute in y = mx, and
(and
(5)
solve for the other intersection point's coordinates as a function of
m.
Two points will go to
p = (0, 0).
Picture:
p
p
What are
and
Change coordinates on
(fractional linear
0 and are sent
: P1 C sending
transformation) so that in your new coordinate,
p. Your parametrization should read now
z 7 (x(z), y(z)) with 0, 7 p. This will
to
exercise.
be used in a later
CHAPTER 4
Lines, conics, and duality
To complete our introduction to algebro-geometric concepts on the
level of curves, in this chapter we'll study projective transformations,
tangent lines, and dual curves. Our convention will be to write
Z =
Z0
Z1 for column vectors in C3 (written with respect to the standard
Z2
2
basis e), and [Z] = [Z0 : Z1 : Z2 ] for the corresponding point in P .
4.1. The classi cation of complex conics
The story begins in even lower degree, with lines i.e. degree
algebraic curves. These are subsets of
of the form
L = { t z = 0},
(4.1.1)
where
is a nonzero vector in
C3 .
Note that for
By stereographic projection (cf.
isomorphically parametrized by
Theorem.
3.3),
C , L = L .
lines and smooth conics are
in the sense of the Normalization
(For lines, the projection is done through a point not on
the line; for conics, one chooses any point on the conic.)
However,
not all conics are smooth, and so we will need to classify conics up to
1 The two key non-smooth examples to keep in
projective equivalence.
mind are the pair of lines
{XY = 0} = {X = 0} {Y = 0}
and the double line
{X 2 = 0}.
1There
is a somewhat subtle point here. For smooth curves in general, projective
equivalence is ner (equates fewer curves) than isomorphism as Riemann surfaces.
However, you have to consider curves of degree at least
5 to see this discrepancy.
As
far as conics are concerned, we like projective equivalence simply because it gives
a uniform and algebraic treatment of singular and smooth curves.
49
50
4. LINES, CONICS, AND DUALITY
The rst has two irreducible components (and is hence reducible ), while
the second has one component of multiplicity two (and is said to be
non-reduced ).
To de ne projective equivalence, we introduce the projective general
linear group
GL(n, C)
.
h id. | C i
(We have A B B = A for some C .) Consider the action
2
of P GL(3, C) on P by
a00 a01 a02
T a10 a11 a12 [Z0 : Z1 : Z2 ] =
a20 a21 a22
"
#
a00 Z0 + a01 Z1
a10 Z0 + a11 Z1
a20 Z0 + a21 Z1
:
:
+a02 Z2
+a12 Z2
+a22 Z2
P GL(n, C) :=
or in more compact notation
T (A)[Z] = [A Z].
(We are, consistently with the notation mentioned at the beginning of
the chapter, letting the matrix
A act on Z
viewed as a column vector. )
This action is well-de ned:
Z to 0 (recall [0] is not a point in P2 );
Z = Y , then T (A)[Z] = [AZ] = [A Y ] = [AY ] = [AY ];
A = B , then T (A)[Z] = [BZ] = [BZ] = T (B)[Z].
it sends no nonzero
if
if
Definition 4.1.1. The transformations
T (A) : P2 P2 , A
P GL(3, C), are the projective linear transformations (or projectivities )
2
of P .
2roughly
speaking, reduced means all of its irreducible components are of mul-
tiplcity one. So while
{XY = 0}
is reduced, something like
{XY 3 = 0}
is not.
Obviously this is going a bit beyond the notion of an algebraic curve as a solution
set, since it incorporates multiplicity. To really formalize what such an object is,
we would have to work with scheme theory or algebraic cycles, which I do not want
to do. So unless otherwise stated, in this course an algebraic curve is assumed to
be reduced.
3the
dot indicates matrix multiplication. This will often be omitted.
4.1. THE CLASSIFICATION OF COMPLEX CONICS
Remark 4.1.2. The analogue of projectivities on
P1
51
are simply the
fractional linear transformations:
a b
c d
!!
[Z0 : Z1 ] = [aZ0 + bZ1 : cZ0 + dZ1 ].
So writing z for the point
[1 : z],
!!
a b
c + dz
(z) =
.
a + bz
c d
You probably know from complex analysis that such transformations
preserve the cross-ratio of 4 points. Furthermore, they are the only automorphisms of
P1
(invertible morphisms from
P1 P1 )
as a complex
1-manifold.
C = {F (Z) =
0} of degree d, points in T (A)C are of the form T (A)Z for Z C . These
How do projectivities aect algebraic curves? For a curve
are precisely the solutions of the equation
{F T (A1 )() = 0} (= T (A)C)
(4.1.2)
since then
F (T (A1 )T (A)Z) = F (T (A1 A)Z) = F (Z) = 0.
4
Since (4.1.2) just substitutes linear forms for
for
C,
Z0 , . . . , Zn in the equation
we nd:
Proposition 4.1.3. The images of (smooth resp.
braic curves of degree
singular) alge-
under projectivities, are again (smooth resp.
singular) algebraic curves of degree
d.
So lines are carried to lines, conics to conics, and so on.
eral, if
T (A)C = C
for some
A,
then the curves
C, C
In gen-
are said to be
projectively equivalent.
Proof. To see why smoothness is preserved, write
[] = T (A)[Z]); and suppose (for a contradicF
F
Z C we have Z
(Z) 6= 0 but
(T (A)Z) = 0
0
i
(where we have in mind
tion) that for some
(i).
4i.e.
F () = F (T (A1 ))
homogeneous polynomials of degree
in
Z0 , . . . , Z n
52
4. LINES, CONICS, AND DUALITY
If
F = F T (A1 ),
chain rule
then
F = F T (A),
and by the (multivariable)
X T (A)i F X F
F
=
=
,
ai0
Z0
Z0 i
i
i
i
so that
0 6=
X F
F
(Z) =
(T (A)Z) = 0,
ai0
Z0
i
i
a contradiction.
Next we want to get formulas for the eect of projectivities on lines
and conics.
Given
( A1 )Z ,
so that
t t
L = { t Z = 0},
0 = t A1 Z =
T (A)L = L( t A1 ) .
(4.1.3)
Since
(4.1.2) gives
GL(3, C) acts transitively on C3 , this implies the (relatively triv-
ial)
Proposition 4.1.4. All lines in
P2
are projectively equivalent.
Let
Q = {0 = aZ02 + bZ12 + cZ22 + dZ0 Z1 + eZ0 Z2 + f Z1 Z2 }
be an arbitrary conic. We can rewrite the equation
0=
Z0 Z1 Z2
d
2
d
2
e
2
in terms of a (unique) symmetric
f
2
Z0
t
Z1 =: ZBZ
c
Z2
e
2
f
2
5 matrix
B.
(The expression
called a symetric bilinear form.) Given such a
in
A Z
for
Z,
QB ,
ZBZ
is
(4.1.2) substitutes
yielding
0 = t (A1 Z)BA1 Z = t Z( t A1 BA1 )Z
so that
T (A)QB = Q( t A1 BA1 ) .
M , the transformation B 7 t M BM =:
B, B 0 are cogredient over C. All nonzero
Given an invertible complex matrix
B0
is called a cogredience, and
5i.e.
the transpose
equals
4.2. TANGENT LINES
/C to
1 0 0
0 1 0 ,
0 0 0
symmetric matrices are cogredient
1 0 0
0 0 0 ,
0 0 0
53
one of the form
1 0 0
0 1 0 .
0 0 1
or
We conclude:
P2
Proposition 4.1.5. All conics in
are projectively equivalent to
one of
{X 2 = 0}, {X 2 + Y 2 = 0},
or
{X 2 + Y 2 + Z 2 = 0}.
X 2 + Y 2 = (X + 1Y )(X 1Y ) is a pair of lines,
projectively equivalent to XY = 0.
Notice that
and so
Corollary 4.1.6. (i) All smooth
7 conics are projectively equiva-
lent.
(ii)
QB
is smooth
Proof. (i) Since
det B 6= 0.
{X 2 + Y 2 + Z 2 = 0}
is the only smooth option
in Prop. 4.1.5, by Prop. 4.1.3 all smooth conics must be equivalent to
this hence to each other.
B 7 t M BM multiplies determinant by (det M )2 ,
which is always nonzero (as M GL(3, C)); so projectivities preserve
non-zero-ness of det B .
(ii) Cogredience
4.2. Tangent lines
Let
C = {F (Z0 , Z1 , Z2 ) = 0}
be a projective algebraic curve, and
suppose
R (, ) C
t 7 f (t) = [Z0 (t) : Z1 (t) : Z2 (t)]
is a dierentiable path segment in
C.
Then,
2
X
dZi
F
0 = (F f ) (0) =
(f (0))
(0)
Zi
dt
i=0
0
6This
is Sylvester's theorem. For an easy proof of the real version, see pp. 161-162
of my linear algebra book.
(To get rid of the 1 entries, hence arrive at the
simpler complex version, just multiply the relevant basis vectors by
7or
equivalently, irreducible
1.)
54
4. LINES, CONICS, AND DUALITY
0
(0)
Z
0
0
F
F
F
= Z
(f
(0))
(f
(0))
(f
(0))
Z
1 (0)
Z1
Z2
0
Z20 (0)
= t F (f (0)) f 0 (0),
LF (f (0)) contains
C through f (0).
and so the line
paths in
all tangent vectors
f 0 (0)
to all such
F (f (0)) = 0,
C to be smooth
There is one catch: if the gradient vector
does not de ne a line at all. So we must ask
then it
at
f (0)
for this computation to work.
Tp C to a curve C = {F =
p = [p0 : p1 : p2 ] C is LF (p) .
Definition 4.2.1. The tangent line
0} P
at a smooth point
f(, )
p=f(0)
f (f(0))
TpC =L
C
The next proposition makes the intuitively obvious statement that projectivities respect tangent lines:
L
T (A)C
Proposition 4.2.2. If
is the tangent line to
is the tangent line to
at
at
p,
then
T (A)L
T (A)p.
Proof. We must show
T (A)LF (p) = L(F T (A1 ))(T (A)p) .
Writing
F = F T (A)1 ,
this is equivalent to
T (A)L(F T (A))(p) = L F (T (A)p)
hence to
L t A1 (F T (A))(p) = L F (T (A)p)
or
(F T (A))(p) t A F (T (A)p)
(4.2.1)
where
means up to multiplication by
C .
As you may wish to check
by writing everything out, equality of both sides of (4.2.1) is just an
expression of the chain rule.
4.3. THE DUAL PROJECTIVE PLANE
55
Now, the tangent line to a line (at any point) is the line itself; for
conics the story is less trivial. First we write
F (Z) = t ZBZ,
B
symmetric, and compute the gradient: writing
dard
e0 , e1 , e2
for the stan-
0
0
1
basis vectors 0 , 1 , 0 ,
1
0
0
t
F/Z0
ZBe0 + t e0 BZ
F (Z) = F/Z1 = t ZBe1 + t e1 BZ
t
F/Z2
ZBe2 + t e2 BZ
t
e0 BZ
= 2 t e1 BZ = 2BZ.
t
e2 BZ
t
ZBei = t ( t ei t BZ) = t ( t ei BZ) = t ei BZ
t
symmetric and ei BZ is 11, i.e. a scalar.)
(Here
Proposition 4.2.3. The tangent line to
uses the fact that
QB
at
[p] QB
is
is
LBp .8
4.3. The dual projective plane
V /C, a basis e = {ei } of V and an
T : V V with matrix [T ]e =: M .
Suppose we have a vector space
invertible linear transformation
Recall that the dual of
is the vector space
V := Hom(V, C)
of linear functionals (f
one has the tautological pairing
h,i
V V C
(4.3.1)
given by
: V C);
hf, vi = f (v).
We have a dual basis
e = {ei } ( ei , ej = ij ),
with respect to which one has
hf, vi =
([f ]e )[v]e
{z }
matrix multiplication
Finally, there is a dual transformation
T : V V
de ned by
T f, T v = hf, vi
8here we are treating p as a column vector (which is consistent with earlier notation)
56
4. LINES, CONICS, AND DUALITY
with matrix
[T ]e = t M 1 .
This gives a more conceptual way to look at the story of lines in
above: put
C
= C
V = C3 , [T ]e = A, f V , = [f ]e
P2
and so on. Of course
as vector spaces, but we want to keep them conceptually
separate.
The crucial point is to projectivize
and
V :
writing
3 \{0}
C3 \{0}
C
2
P =
, P =
,
C
C
2.
P2 correspond to points [] P
2
we see that lines in
In fact, since the
notion of duality is de ned by (4.3.1), it is symmetric:
points in
correspond to lines in
2.
P
V = V ,
and so
This entire correspondence is
invariant under projectivities provided one operates simultaneously on
P2
with
T (A)
and
2
P
T ( t A1 ).
with
A bit more formally, then:
2 is the space of lines
P
Definition 4.3.1. The dual projective plane
in
P2 .
Now write
p0
p = p1 ,
p2
= 1
2
for column vectors. Though this is maybe a little awkward, here is how
I want to standardize notation:
p = [p] = [p0 : p1 : p2 ] P2
2,
= [ t ] = [0 : 1 : 2 ] P
in other words, points in
points in
2
P
P2
are thought of as column vectors and
as row vectors. As above the line
L P2
is de ned by
the equation
Z =0
and we say that its dual
solve
= .
L
for
Z0
Z = Z1 ,
Z2
Moreover,
de nes a line
2
Lp P
via
W p=0
solve for
W =
W0 W1 W2
4.3. THE DUAL PROJECTIVE PLANE
and we write
57
p = Lp .
What about the dual of a con guration of
(a) a point
on a line
L()
(important in Poncelet)?
L, L through a point p?
points p, q on a line L?
(b) a pair of lines
(c) a pair of
For the rst one, the equation
t
expresses p
Lp ).
L ;
p=0
but from the above it also expresses
p
L
(i.e.
Repeating this reasoning, we have
(
a) a line
p through
a point
;
L
a pair of points
(b)
(
c) a pair of lines
L on a line p;
L,
.
p, q through a point L
Here is something more interesting to dualize, which is left as an exercise for you.
L, L
L) take
Theorem 4.3.2. [Pappus of Alexandria, c. 300 AD] Let
P2
s = L L. On L (resp.
(3)
(resp. q , q , q ) dierent from s, and set
:= p(i) q (j) p(j) q (i) (where {i, j, k} = {1, 2, 3}).
be two distinct lines, and write
(1)
(2)
(3)
p ,p ,p
(k)
(for k = 1, 2, 3) r
(1)
(2)
(3)
Then r , r , r
are
distinct
(1)
(2)
collinear.
p(3)
p(2)
r(1)
(2)
p(1)
q(3)
(3)
(2)
(1)
58
4. LINES, CONICS, AND DUALITY
p(1) q (2) p(3) q (1) p(2) q (3) is a hexagon inscribed in the
(i)
conic LL, and the {r } are the intercepts of its opposite edges. After
Proof. In fact,
changing by a projectivity (which preserves the gure), this conic is
XY = 0, which is obviously the limit of the smooth conic XY Z 2 = 0
as 0. Since Pascal's theorem implies collinearity of the hexagon
edge intercepts for all 6= 0, this remains true at = 0.
4.4. Dual conics and polar lines
2
C P
Definition 4.4.1. The dual
C = {F = 0} P is the set of
2 | p C}.
= {TpC P
That is, C
of a smooth algebraic curve
(dual points of ) tangent lines to
C.
This is consistent with our de nition for lines. For higher degree
curves, however, the dual is not one point: consider the duality map
2
DC : P2 P
sending
p 7 [ t F (p)].
C = DC (C).
= TpC , then T C = p.
Proposition 4.4.2. (a)
(b) If
is smooth at
Proof. (a) For
p C , Tp C = LF (p) = TpC = [ t F (p)].
So
this is practically a tautology.
(b) Here we jump into a little deep water. It su ces to show that
for any path
() : (, ) C
through
TpC = [ t F (p)] = DC (p),
d t
(0) p = 0.
dt
(4.4.1)
C is the image of C
(, ) C through p. So
Since
by
DC , t (t) = (DC q)(t)
for some
q() :
the left-hand side of (4.4.1) becomes
d
(DC q)(0) p =
dt
(4.4.2)
q00 (0) q10 (0) q20 (0)
9i.e., [Z : Z : Z ] 7
0
1
2
2F
Z02
2F
Z1 Z0
2F
Z2 Z0
F
Z0 (Z0 , Z1 , Z2 )
2F
Z0 Z1
2F
Z12
2F
Z2 Z1
2F
Z0 Z2
2F
Z1 Z2
2F
Z22
F
Z1 (Z0 , Z1 , Z2 )
p0
p1
.
p2
p
F
Z2 (Z0 , Z1 , Z2 )
4.4. DUAL CONICS AND POLAR LINES
The matrix in the middle is the Hessian of
determinant if
DC (p).
is nonsingular at
59
p and has nonvanishing
at
We will return to the Hessian
later in this course.
Now, since each
F
is homogeneous (of degree
Zi
the Euler formula ((2.1.8), with
(d 1).
p)
set equal to
q00 (0) q10 (0) q20 (0)
(d 1).F (q(0))
which is indeed zero by the beginning of
d1, if d = deg(C)),
collapses (4.4.2) to
F
(p)
Z0
F
(p)
Z1
F
(p)
Z2
dq
(0),
dt
4.2.
Remark 4.4.3. One can show that the dual of a smooth algebraic
curve of degree
d2
d is an algebraic curve of degree d(d 1);
moreover, for
this dual is singular, so the duality map cannot be reversed as
de ned.
QB . By the computation
F (Z) = ZBZ ), F (Z) = 2 ZB t ZB and so
DQB (QB ) = [ t ZB | [Z] C]
We consider the dual of the conic
(for
in
4.2
= {[ t ZB] | t ZBZ = 0}.
Making the substitition
= t ZB Z = t B 1 ,
this becomes
2 | t W B 1 B t B 1 W = 0}
B = {[ t ] P
Q
2 | t B 1 = 0}
= { P
where we have used the fact that
is symmetric. This gives part (i)
of:
Proposition 4.4.4. (i)
B = QB1 ,
Q
the dual of a smooth conic is a smooth
= Q . (In particular,
Q
B
B
conic, since det B 6= 0 =
and
det B 1 6= 0.)
(ii) Given
tangent to
p P2 \QB ,
there exist exactly two lines through
and
QB .
Proof. (ii) By Proposition 4.4.2(b), this is dual to the statement:
if the line
2
p P
is not tangent to
B,
Q
then it meets
B
Q
in exactly
points. This last statement then follows from Proposition 2.1.8.
2
60
4. LINES, CONICS, AND DUALITY
QB be a smooth conic and p be a point not
on QB , with Tq QB and Tr QB the two tangent lines to QB containing p.
2
(Here q, r QB .) Then the polar line L(p,QB ) P of p with respect to
QB is the line through q and r.
Definition 4.4.5. Let
p
Tq QB
Tr QB
r
L
p P2 \QB , with polar line L = L(r,QB ) (
2 (of the dual point L
with respect
P2 ). Then the polar line L(L,
Q
B) P
to the dual conic) is p
(the dual line of p). In a picture, where dual
Proposition 4.4.6. Let
objects are the same color:
p
Tq QB
L
Tr QB
q
q
Tq QB
r
Tr QB
(a,b,c)
the de nition of the dual curve, and Proposition 4.4.2(b).
2
2
2
Example 4.4.7. Q = {4Z0 +Z1 +Z2 = 0} B =
1
.
1
Let p = [1 : 2 : 2] (
/ Q) so that the polar line L is given by
Z0
4
0= 1 2 2
1
Z1 = 4Z + 2X + 2Y,
1
Z2
Proof. This is an immediate consequence of the rules
, c),
(a, b
EXERCISES
i.e.
L = L
where
61
= [4 : 2 : 2].
On the dual
2
P
side:
= ; Q
has matrix
41
B 1 =
1
p is the line 0 = W0 + 2W1 +
hence equation 0 =
is
2W2 . On the other hand, the polar line of with respect to Q
1
W0
4
0 = 4 2 2
1
W1 = W0 + 2W1 + 2W2 ,
W2
1
14 Z02
agreeing with
+ Z12
+ Z22 ; and
p.
It is instructive to think about what happens to Pascal and Poncelet
under duality. While the dual of Poncelet is again just Poncelet (but in
2 ), we do nd that if polygons inscribed in C and circumscribed about
P
2 ) inscribed in
D close up after n sides, then so do the polygons (in P
and circumscribed about C .
D
The dual of Pascal, on the other hand, does give a dierent statement:
Proposition 4.4.8. The 3 lines through opposite vertices of a hexagon
circumscribed about a conic, pass through a single point.
Proof is basically the same as for Proposition 4.4.6; I'll let you work
it out.
Exercises
(1) Given a con guration of
4 points p, q, r, s P2 in general position,
i.e. no three of them collinear, show there exists a unique projec-
p 7 [1 : 0 : 0], q 7 [0 : 1 : 0], r 7 [0 : 0 : 1], s 7 [1 :
1 : 1]. [Hint: work with vectors p, q, r, s C3 . You only have to
send p (resp. q, r, s) to a multiple of e0 (resp. e1 , e2 , e0 + e1 + e2 ).]
tivity sending
(2) (a) Give a direct proof of Pappus's theorem.
[Hint: use the last
exercise to rst simplify the coordinates of several of the points.]
(b) State a dual version of Pappus's theorem, and draw a gure.
[Note: it would be better to state the dual version in
of Pappus in
2
P
and then dualize that.]
P2 :
think rst
62
4. LINES, CONICS, AND DUALITY
(3) Show that the equation of the polar line of
has equation
pBZ = 0.
with respect to
QB
Use this to give another (short) proof of
Proposition 4.4.6.
(4) Prove that all automorphisms of
P1
(as a complex manifold) are
fractional linear transformations. Deduce that
[Use material from
3.1.]
Aut(P1 )
= P GL2 (C).
Part 2
General de nitions and results
CHAPTER 5
Complex manifolds and algebraic varieties
In Chapters 2-3 we introduced Riemann surfaces and plane algebraic curves, and stated the Normalization Theorem which produces
a strong relation between them. Here we will introduce the arbitrarydimensional generalizations of these objects. While it is true that an
algebraic variety of dimension
n-manifold, the converse
complex 2-manifolds.
plex
n has a desingularization1 which is a comis false: already there are non-algebraic
However, it is true that any global analytic object (functions or differential forms, for example) on a projective algebraic variety viewed
as a complex manifold, is a algebraic. This is Serre's GAGA (global
analytic
global algebraic) principle. For example, global meromor-
phic functions in this context turn out to be nothing but restrictions
to the algebraic variety of rational functions on the ambient projective
space (elements of
C(z1 , . . . , zn )).
5.1. Complex
n-manifolds
These are the generalization of complex
1-manifolds (or of Riemann
surfaces, if we assume compactness) to higher dimension. Once again,
we begin with a Hausdor topological space
and write
U = U U .
with open cover
This is made into a complex
the additional data of an analytic atlas on
X:
{U }
n-manifold
by
that is, a collection of
2
holomorphic coordinates
'
z : U V Cn ,
1normalization
is no longer the correct term (refers to a weaker process which still
produces a singular object)
2As
usual, an underline means a vector or tuple of some kind; in this case,
(z1 , . . . , zn ).
65
z =
66
5. COMPLEX MANIFOLDS AND ALGEBRAIC VARIETIES
or homeomorphisms between
and an open set in
Cn ,
such that the
transition functions
:= z z 1 : V V
are biholomorphic.
X
U
C
V
(z 1,...,z n)space
V := z (U )
Here
and
(z1 ,...,z n)space
V := z (U )
are open subsets of
Cn ,
and we need to explain what biholomorphic means. First, a function
f : Cn C is holomorphic if and only if it looks locally (about each
P
i1 i2
I
I
in
point) like f (z) =
I=(i1 ,...,in ) aI z where z means z1 z2 zn and
aI C are constants. A holomorphic map Cn Cm is just m of these:
(z1 , . . . zn ) = z 7 (f1 (z), . . . , fm (z)). (Since these de nitions are local,
n
they immediately have meaning when C etc. are replaced by open
sets.) Finally, biholomorphic simply indicates a bijective map (
is bijective by construction) which is holomorphic in each direction.
To generalize the morphisms of Riemann surfaces introduced in
3.1,
we can de ne a morphism
of complex manifolds.
need not be of the same dimension; for X we keep the
'
m
above notation and for Y write Z : U V C . A morphism F
Here
F : X Y
and
is then a collection of continuous functions
the
U )
F : U Y
(agreeing on
such that each composition
Z F z 1 : z F1 (U F (U )) Z (U F (U ))
Pn
5.2.
AS A COMPLEX MANIFOLD
yields a holomorphic map (from a subset of
V Cm ).
If
n = m = 1
67
V Cn
to a subset of
then this reproduces De nition 3.1.6.
Moreover, compositions of morphisms are morphisms.
Basic examples of complex manifolds include (besides Riemann surfaces when
n = 1)
Cartesian products of Riemann surfaces, complex
n-tori
Cn /Z 1 , , 2n
1 , . . . , 2n
projective n-space
(where
Pn :=
are linearly independent over
in
Cn
= R2n ),
and
Cn+1 \{0}
.
h(0 , . . . , n ) (0 , . . . , n ) C i
Demonstrating that
Pn
is a complex
n-manifold
(as we do in the next
section) immediately gives meaning to a morphism of complex manifolds from a Riemann surface to
Pn .
This notion is equivalent to (but
more intrinsic than) De nition 3.1.9, as we shall see.
Pn
5.2.
as a complex manifold
Pn is covered by the open sets Ui := {i 6= 0}, with local coordinates
!
bi
n
0
=
,..., ,...,
: Ui Cn .
zi = (zi1 , . . . , zin ) :=
i
i
i
(Here i replaces ,
Vi = Cn ,
and
c
()
means to omit that term.) We
need to check that the transition functions
ji : Vij Vji
ji tells us how to write the zj = (zj1 , . . . , zjn )
functions of the zi = (zi1 , . . . , zin ) in such a way that
!
!
0
bi
j
n
0
i
bj
n
,..., ,..., ,...,
is sent to
,..., ,..., ,...,
i
i
i
i
j
j
j
j
are holomorphic. Now,
as
(where for convenience we assume
subset where
zij 6= 0.
j > i).
Now,
Vij Cn
is simply the
So the correct transition function is
ji (zi1 , . . . , zin ) = (zj1 (zi1 , . . . , zin ), . . . , zjn (zi1 , . . . , zin ))
3it
might be a good idea to glance back at the picture there (for intuition purposes)
68
5. COMPLEX MANIFOLDS AND ALGEBRAIC VARIETIES
where
zik /zij ,
zjk (zi1 , . . . , zin ) =
zi,k1 /zij ,
1/zij,
(5.2.1)
For
z1 =
0
1
k i, k > j
for i + 1 < k j .
for k = i + 1
for
z0 = 10 and
1
are the two local coordinates, while (5.2.1) becomes z1 (z0 ) =
,
z0
P1 , zi
reduces to
zi (i = 0, 1).
More precisely,
so that we recover Example 2.2.3. Here is a schematic picture of the
P1 :
1 / 0
0 /
local coordinates on
0
For
1 2
,
0 0
z0 = (z01 , z02 ) =
z2 = (z21 , z22 ) = 02 , 21 , with
and
, we have
e.g.
0 2
,
1 1
z1 = (z11 , z12 ) =
,
01
20 (z01 , z02 ) = z102 , zz02
.
Again, the local coordinates can be visualized as follows:
[Link]
1 / 2
0 / 2
/
2
1 / 0
0 /
[Link]
[Link]
So, for instance, the coordinates
plement
U1
2/ 1
1
z1 =
0 2
,
1 1
are de ned on the com-
of the vertical line, and both vanish at
[0 : 1 : 0].
Remark 5.2.1. Whenever you have a local holomorphic coordinate
'
zi : Ui Vi Cn , the inverse mapping i = zi 1 : Vi
Ui X (or just Vi , X ) is called a local analytic chart. In case
X = Pn , i : Cn , Pn is given by
(system) like
i (zi1 , . . . , zin ) = [zi1 : : zii : 1 : zi,i+1 : : zin ]
Cn , (Cn+1 \{0}) Pn .
P1 and P2 :
, and one can visualize this as a map from
Here are pictures of the image of
for
5.2.
0=1
C2
Pn
AS A COMPLEX MANIFOLD
69
U0
U0
0
1
0=1
The next statement says that the notion of holomorphic map from
a Riemann surface to projective space (Defn.
3.1.9) is just a special
case of morphism of complex manifolds. It is enough to consider (as
we do) the situation where the image is not contained in a coordinate
hyperplane, since these are just smaller-dimensional projective spaces
(included into
Pn
by morphisms).
M be a complex 1-manifold, and consider
F : M Pn with F (M ) not contained in any
Proposition 5.2.2. Let
a continuous mapping
{Zi = 0}. The following statements are then equivalent:
(i) F is a morphism of complex manifolds;
(ii) each composition [i F : j F ] is a morphism of complex
1
manifolds to P (on the open subset of M where it is well-de ned);
(iii) each i F gives a meromorphic function on M .
j
Ui := {Zi 6= 0} Pn as above. For each
{i, j} (where i 6= j ), the projections ij : Ui Uj P1 de ned by
[0 : : n ] 7 [i : j ] are morphisms of complex manifolds. So if
F
M Pn is one, then ij F = [i F : j F ] is one too, showing
(i) = (ii). Next, (ii) = (iii) is Remark 3.1.10. Finally, if all
j
the
F give meromorphic functions on all of M , then in particular
i
j
0
\
i
n
zi F = i F, . . . , i F , . . . , i F, . . . , i F is holomorphic on
Proof. First, write
F 1 (Uj )
(or a suitable covering of it by coordinate neighborhoods).
These give the local holomorphic representations of
morphism, proving
(iii) = (i).
We will re ne Proposition 5.2.2 in Chapter
required for a
7
below.
Whilst we are dwelling on the subject of projective space, I would
like to mention (just for
P2 )
a trick for drawing the real solution sets
70
5. COMPLEX MANIFOLDS AND ALGEBRAIC VARIETIES
of homogeneous equations on the page: barycentric coordinates. First
draw 3 points
A(0) , A(1) , A(2)
on a piece of paper:
0
Think of these as vectors
[0 : 1 : 2 ]
is. Now, plot
A(i) R2 ;
as
2
X
(5.2.2)
it doesn't matter where the origin
P2
i=0
j=0 j
A(i) .
To draw an algebraic curve, simply nd all the solutions
with
i R,
and use (5.2.2) to plot them.
Example 5.2.3. (i) The line
completes to
[0 : 1 : 2 ]
2 = 0 .
y=
(ii) The conic
xy = 1
y =
R) projectively
[1 : x : ] in this way gives
(assume
Plotting the points
[1: x :]
completes to
1 2 = (0 )2 ,
and we get
5.3. AFFINE AND PROJECTIVE ALGEBRAIC VARIETIES
71
for its real barycentric plot.
(iii) The cubic curve
0 )(1 0 )
y 2 = x(x + 1)(x 1)
becomes
(2 )2 0 = 1 (1 +
with picture
In fact, this is the precise meaning of the schematic real 1-dimensional
pictures of complex algebraic curves we have been drawing and will
continue to draw we are plotting the real solutions in barycentric
coordinates.
5.3. A ne and projective algebraic varieties
We are going to approach this from a slightly more algebraic angle
than, take the common solution of a bunch of polynomial equations.
Start with the commutative ring
Sn := C[z1 , . . . , zn ]
of polynomials in
variables.
Let
J Sn
be an ideal. The a ne variety associated to
V (J) := {z = (z1 , . . . , zn ) | f (z) = 0 f J} Cn ,
is
72
5. COMPLEX MANIFOLDS AND ALGEBRAIC VARIETIES
which is the vanishing locus of all polynomials in
J.
By a result in
Sn is nitely
generated, that is, of the form (f1 , . . . , fk ); consequently V (J) is simply
of the form f1 (z) = = fk (z) = 0. However, working in terms
algebra known as Hilbert's basis theorem, any ideal in
of ideals does have a payo, in the form of the famous theorem on
zeroes or nullstellensatz :
g Sn
to J .
Theorem 5.3.1. [D. Hilbert, 1893] If
on
V (J),
If
does,
then for some
m N,g
belongs
J = (f ), then this just says if g
then f divides some power of g .
vanishes identically
vanishes (in
Cn )
wherever
Sn+1 = C[Z0 , . . . , Zn ].
d
Its underlying additive group can be viewed as the direct sum d Sn+1 ,
d
where Sn+1 denotes homogeneous polynomials of degree d in n + 1
variables. Hence, any polynomial G can be written uniquely as a nite
P
sum of homogeneous terms
d Gd .
Next we consider the projective case, writing
Definition 5.3.2. An ideal
I Sn+1
is homogeneous if and only
if the condition
GI
Gd I (d)
is satis ed.
The projective variety associated to a homogeneous ideal
I Sn+1 is4
V (I) := {[Z] = [Z0 : : Zn ] | F (Z) = 0 F I} Pn .
A version of the Nullstellensatz suited to this case, which is an immediate consequence of Theorem 3.1, is:
Corollary 5.3.3. Given a homogeneous polynomial
on all of
V (I),
some power of
Remark 5.3.4. If
F1 , . . . , F k
belongs to
vanishing
I.
are homogeneous polynomials (of var-
ious degrees), then
I := (F1 , . . . , Fk ) is a homogeneous ideal;
(I) = {F1 (Z) = = Fk (Z) = 0}.
(ii) V
(i)
4technically,
one should keep track of multiplicities of irreducible components,
rather than just de ning
detail.
and
V (I)
as a set.
For the most part we will suppress this
5.3. AFFINE AND PROJECTIVE ALGEBRAIC VARIETIES
73
As in the case of curves, we want to be able to go between the a ne
and projective settings. To restrict a projective variety to the a ne
world, start with the ring (or algebra) homomorphism
Sn+1 Sn
induced by
F (Z0 , Z1 , . . . , Zn ) 7 F (1, z1 , . . . , zn ).
If we write
for the image of a homogeneous ideal
under this map,
then
V (I ) = V (I) Cn .
To go the other way, recall the space
of degree at most
d.
Pnd = dj=1 Snj
of polynomials
We have a homomorphism of abelian groups (or
vector spaces)
d
d
Pnd
Sn+1
which is de ned by
f (z) 7 (Z0 )d f (Z/Z0 ) .
J Sn an ideal. Given generators {fi }ki=1 (fi of degree di )
for J , so that J = (f1 , . . . , fk ), set J := (d1 (f1 ), . . . , dk (fk )). Then we
Now, take
have
Cn = V (J).
V (J)
So if
n = 3,
then
V (J)
is adding stu in the P
at in nity
{Z0 = 0}
to complete your a ne variety to a projective one, as suggested by the
picture:
hyperplane {Z 0=0} at
in which the black points get added in the process of completing the
curve.
74
5. COMPLEX MANIFOLDS AND ALGEBRAIC VARIETIES
Example 5.3.5. A key example of a ne or projective varieties are
the hypersurfaces cut out of
5
d
Sn+1
and
Cn
or
Pn
by a single equation. Let
X = V (F ) = {F (Z) = 0} Pn
the corresponding projective hypersurface of degree
d.
(Like algebraic
curves, these are called linear, quadric, cubic, quartic, quintic, etc.
d = 1, 2, 3, 4, 5, . . ..) We will de ne dimension rigorously
below, but X is (n 1)-dimensional (and thus of codimension 1).
Now since Sn is a unique factorization domain, we can factor
according as
F =
k
Y
Fimi ,
i=1
uniquely (up to order), where each
Fi
is prime (irreducible). We can
then write unambiguously
X=
k
X
mi Xi ,
i=1
and say
is reduced when all
mi = 1,
and irreducible when
k = 1.
Exercises
(1) Show that each projection
ij : Ui Uj P1
described in the proof
of Prop. 5.2.2 is a morphism of complex manifolds. [This is a quick
one.]
(2) Sketch the real solutions of [the projective closure of]
ai )}
5here
in
, if
a1 < a2 < < a2g+2
we really mean
V (F ).
V ((F )),
{y 2 =
Q2g+2
i=1
(x
are real numbers.
the variety of the ideal
(F ),
but we shorten this to
CHAPTER 6
More on projective algebraic varieties
We warm up with two examples we can get our hands on immediately: linear varieties and quadric hypersurfaces.
Then we launch
into what it means for an algebraic variety to be singular resp. smooth
at a point, and in the latter case introduce its tangent space at that
point (which is a linear variety).
This leads to a careful de nition
of dimension for algebraic varieties. As a sort of appendix I'll give
a long-overdue introduction to plane curve singularities (which were
glossed over in Chapter 2).
6.1. Linear subvarieties of
Pn
We start by generalizing the projectivities of Chapter 4. Recall
that the projective general linear group is de ned as the quotient of
invertible matrices by the scalar action:
P GL(n + 1, C) :=
GL(n + 1, C)
.
+
*
0
..
.
C
This group acts on projective space by the rule
P GL(n + 1, C) Pn Pn
(M, [Z]) 7 [M.Z] =: T (M )[Z].
That is, for each
of
M P GL(n + 1, C), T (M )
gives an automorphism
as a complex manifold. In fact, the projectivities
T (M )
give all
automorphisms (generalizing the last exercise of Chapter 4), but we
won't prove this here.
75
76
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
A system of
linear equations
`10 0 + + `1n n = 0
.
.
.
(6.1.1)
`k0 0 + + `kn n = 0
de nes a linear subspace
V Pn .
Recalling that the rank of a matrix
is its number of linearly independent row (or equivalently, column)
vectors, the matrix
`10
.
.
.
..
.
.
.
`k0
has rank(L)
=: r k .
codim(V
`1n
=: L
`kn
De ning
) := r (equivalently, dim(V ) = n r),
we have the
Proposition 6.1.1. (i) All projective linear subvarieties of the
same (co)dimension are projectively equivalent.
(ii) A linear subvariety of
nr
Pn
of codimension
is isomorphic to
as a complex manifold.
L and V as above, note that if the equations (6.1.1)
are not independent (i.e. k > r ), then without changing V or r , we can
eliminate equations (reducing k ) until they are (and k = r , i.e. L has
maximal rank). Assume this has been done, so that reordering Zi 's if
Proof. Given
necessary,
det
`10
.
.
.
..
`k0
`1,k1
.
.
.
6= 0.
`k,k1
M be the (n + 1) (n + 1) matrix whose rst k rows are given by
L (a k (n + 1) matrix) and last n k + 1 rows by (0 , Ink+1 ), where
0 denotes a (n k + 1) k matrix of zeroes and Im always means an
m m identity matrix.
n
Consider the automorphism T (M ) of P . By de nition of V , [] V
if and only if (matrix multiplication by) L kills . So one should view
T (M ) as taking V to the subspace V0 = {0 = = k1 = 0},
Let
6.2. QUADRIC HYPERSURFACES
which proves (i) since
evidently a
Pnk
77
was arbitrary. This also proves (ii) since
[k : : n ]).
(with homogeneous coordinates
A linear subvariety of codimension
V0
is
is called a hyperplane.
6.2. Quadric hypersurfaces
Recall that a (projective) hypersurface is a subvariety
F (Z) = 0.
2), so that X
out by a single homogeneous equation
in the case where
2
Sn+1
(degree
X Pn
cut
We are interested
is a quadric. The
polynomial can be written
F (Z) = t ZBZ =
Z0
Zn
b00
.
.
.
..
.
.
.
bn0
with
b0n
Z0
.
.
.
bnn
Zn
symmetric. Under a linear change of projective coordinates
Z0
.
.
.
=M
Y0
.
.
.
(M GL(n + 1, C)),
Yn
Zn
we nd
F (Z) =
Y0
Yn
M BM
Y0
.
.
.
=: G(Y ),
Yn
t
where (as in Chapter 4)
M BM
is said to be cogredient to
Lemma 6.2.1. [Sylvester's Theorem
complex
matrices
(n + 1) (n + 1)
matrix
/C]
Any given symmetric
is cogredient to exactly one of the
Mk =
B.
..
1
0
..
0
where
is the number of
1's.
Corollary 6.2.2. A given quadric hypersurface in
Pn
is projec-
tively equivalent to (or transformable by a linear change of coordinates
78
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
into) exactly one of the quadrics
Qk =
( k1
X
)
Yj2 = 0
(k = 1, . . . , n + 1).
j=0
Q1 = {Y02 = 0} is a double hyperplane, Q2 = {Y02 +Y12 =
union of two hyperplanes (each
= Pn1 ), and Qk for k 3 is
Note that
0}
is a
irreducible (equation does not factor). You'll investigate these a tiny
bit further in one of the exercises.
6.3. Singularities, tangent planes, and dimension
We'll need the Euler formula mentioned in
2.1,
so let's prove it
rst:
Lemma 6.3.1. [Euler's formula]
d
=
F Sn+1
Pn
i=0
F
Zi Z
=
i
d.F
(F =)Z0d0 Zndn ,
P
P
P
di = d. We have i Zi Z
(Z0d0 Zndn ) = i Zi Zdii (Z0d0 Zndn ) =
i
P
( i di )Z0d0 Zndn = dZ0d0 Zndn .
Proof. It su ces to check this on monomials
Now, the de nition of smoothness for hypersurfaces is similar to
what we have learned for curves; the general case of varieties cut out
by more than one equation is trickier.
So we'll start, then, with an
a ne hypersurface
V = V (f ) Cn ,
and a point
pV.
f
V is smooth at p z
(p) 6= 0 for some
j
j {1, . . . , n}. Otherwise, p is a singular point (or singularity ) of V .
(ii) If V is smooth at all of its points, V is smooth. Otherwise, V is
Definition 6.3.2. (i)
singular.
(iii) If
V is
(
smooth at
p,
de ne the tangent plane
(z1 (p) + 1 , . . . , zn (p) + n ) |
Tp V :=
n
X
i=1
(Here the
)
f
i
(p) = 0 Cn .
zi
i C.)
So one can think of
coordinates
{i }.
Tp V
as a copy of
Cn1
with origin at
and
It's also worth noting the formal correspondence
6.3. SINGULARITIES, TANGENT PLANES, AND DIMENSION
79
Tp V ) and dierential operators
i i zi . This is not misleading at all, and in fact the
between tangent vectors (points in
at
p,
namely
intrinsic construction of tangent planes (for complex or more generally
dierentiable manifolds) uses local dierential operators.
As for singularities, i.e.
fzn (p) = 0,
f (p) = fz1 (p) = =
points where
we saw examples of those in Example 2.1.10 for curves.
y 2 = x3 x2
(0, 0) since both partials
of y x + x . On the other hand, y = x x is smooth because
this equation together with 0 =
(y 2 x3 + x) = 3x2 + 1 and
x
0 = y
(y 2 x3 + x) = 2y admit no common solution. This is easy to
1
1
see: the points ( , 0) and ( , 0) where both partials vanish, do not
3
3
Here is one more:
is singular at
lie on the curve.
Next, consider a projective hypersurface
V = V (F ) Pn ,
where
is homogeneous and
P V.
F
V is smooth at P Z
(P ) 6= 0 for some
j
j {0, . . . , n}. Otherwise, P is a singular point (or singularity ) of V .
(ii) If V is smooth at all of its points, V is smooth. Otherwise, V is
Definition 6.3.3. (i)
singular.
V
(
(ii) If
TP V :=
is smooth at
P,
de ne the tangent plane (
=
[Z0 (P ) + 0 : . . . : Zn (P ) + n ] |
n
X
i=0
Pn1 )
)
F
(P ) = 0
Zi
Pn .
Now a priori, the de nition of a singular point is one at which
F (P ) = FZ0 (P ) = = FZn (P ) = 0; but by the Euler
X
F
(6.3.1)
Zi (q)
(P ) = deg(F ) F (P )
Z
i
i
and so it is in fact enough to check
formula,
FZ0 (P ) = = FZn (P ) = 0.
In fact,
(6.3.1) also implies (for the projective case only!) the simpli cation
(
(6.3.2)
[0 : : n ] |
TP V =
X
i
F
(P ) = 0
i
Zi
Note that (6.3.2) is really just the solution set of
Chapter 4 (but now in
rather than
).
F (P ) = 0,
as in
80
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
As you might expect, the notions of tangent plane in a ne and
projective cases agree, in the sense that at a point on an a ne
hypersurface the tangent plane of the projective completion is the
completion of the tangent plane:
Tp V (F (1, z1 , . . . , zn )) = T[1:p] V (F )Cn , where
[1 : z1 (p) : : zn (p)].
Proposition 6.3.4.
(P =)[1 : p]
means
Proof. Given
and
Q = [1 : q],
q = (z1 (q), . . . , zn (q)) Cn .
Writing
f (z) = F (1, z)
we want to show
q Tp V (f ) Q TP V (F ).
(6.3.3)
The left-hand (a ne) condition is, writing
zi (q) = zi (p) + i
in De ni-
tion 6.3.2(iii),
n
X
f
(zi (q) zi (p))
(p) = 0.
zi
i=1
This is really
n
X
(Zi (Q) Zi (P ))
i=1
F
(P ) = 0,
Zi
which by Euler becomes
n
X
Zi (Q)
i=1
Since
F (P ) = 0,
F
F
(P ) deg(F ) F (P ) + 1
(P ) = 0.
Zi
Z0
we get
X
F
F
1
(P ) +
Zi (Q)
(P ) = 0,
Z0
Z
i
i=1
which is exactly the right-hand (projective) condition of (6.3.3).
Now let's have a look at singularities and smoothness in the general
projective case. The de nition is complicated, but after this chapter
we won't use it much. Let
V = V (F1 , . . . , Fk ) Pn ,
and
be a point on
V.
6.3. SINGULARITIES, TANGENT PLANES, AND DIMENSION
Definition 6.3.5. (i)
neighborhood
1
such that
W Pn
of
V
p
is smooth at
81
if and only if there exists a
and sub-index set{i1 , . . . , ic }
{1, . . . , k}
V W = V (Fi1 , . . . , Fic ) W , and
Fi1
F
(p) Zin1 (p)
Z0
.
.
..
.
.
(b) rank
.
= c.
.
.
(a)
Fic
(p)
Z0
has codimension
V
(ii) If V
We say
Fic
(p)
Zn
(or dimension
n c) at p.
is smooth at each point p V , then V is smooth
c
(otherwise,
is singular ).
(iii) If
has the same (co)dimension at each smooth point
p V,
V is equidimensional. If moreover that codimension is c, we just
2
say V is a variety of codimension c (dimension n c).
n
(iv) The tangent plane Tp V P to V at a smooth point p is the
solution set of L.p = 0, where
F1
F1
(p)
(p) Z
Z0
n
.
.
..
.
.
L=
.
.
.
.
then
Fk
(p)
Z0
Fk
(p)
Zn
Condition 6.3.5(i)(a) says that locally about
= Fic (Z) = 0,
p, once you set Fi1 (Z) =
the remaining equations are redundant; and roughly
speaking, the condition (b) on rank says that no more (none of the
F i` )
are redundant. In the terminology of
6.1, Tp V
is a linear subva-
riety, and it follows from condition 6.3.5(i)(b) that its codimension is
c.
V
That is, we have really just de ned the (co)dimension of a variety
at a smooth point
p,
to be the (co)dimension of
Tp V
something
we already knew how to de ne.
Finally, the neighborhood
set containing
in the de nition is an analytic open
(such as a ball), but the de nition would also work
if we only permitted algebraic open sets de ned by complements of
(other) subvarieties,
3 known as Zariski open sets. In general, if you
1the matrices in this de nition assume a particular representative (Z (p), . . . , Z (p))
0
n
in
Cn+1
of the projective coordinate
[Z0 (p) : : Zn (p)].
It doesn't matter which
one you take, as long as you are consistent.
2note that the dimension of V
a
m-dimensional variety must
3this
at a singular point is not de ned, so the de nition of
be one that is of dimension
is exactly how Example 6.3.6(iii) is started below
m at all smooth
points.
82
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
want to view an algebraic variety as a complex analytic space (or manifold, if it is smooth), then you must use analytic open sets; on the
other hand, the Zariski open sets introduce a dierent topology on
or
, which is coarser but has the advantage of being algebraic (and
is still Hausdor ). We need both. In brief, when we study varieties analytically, we use the analytic topology; when we want to make heavy
use of the correspondence between varieties and ideals in commutative
rings, we use the Zariski topology.
V be the a ne variety {z12 + z22 + z32 = 0}
P 2
derivatives zi (
j zj ) = 2zi all vanish at p = (0, 0, 0)
Example 6.3.6. (i) Let
C3 .
The partial
and so
is singular there:
Z1 Z3 = 0
3
(ii) Now for a nasty one. Let V P be de ned by
Z2 Z3 = 0
and take p = [1 : 0 : 0 : 0], q = [1 : 0 : 0 : 1], r = [1 : 1 : 0 : 0]:
)
,
r, Z1 6= 0 and so having set Z1 Z3 = 0 (i.e. Z3 = 0), the
second equation Z2 Z3 = 0 is redundant. So the relevant matrix from
6.3.5(i)(b) is
Z0 (Z1 Z3 ) Z1 (Z1 Z3 ) Z2 (Z1 Z3 ) Z3 (Z1 Z3 ) =
r
|
=
,
which
has
rank
1
proving
that
0 Z3 0 Z1 r
0 0 0 1
V has codimension 1 (dimension 2) at r. Locally about q , Z3 6= 0 and
so the equations are eectively Z1 = 0 and Z2 = 0, neither of which is
Locally about
6.4. SINGULARITIES OF PLANE CURVES
83
0 Z3 0 Z1
0 0 Z3 Z2
redundant. The matrix in 6.3.5(i)(b) is now
!
=
q
0 1 0 0
,
0 0 1 0
mension 2 at q .
which does have rank
So
con rming that
is not equidimensional.
Finally, at
equation is redundant but the matrix 6.3.5(i)(b) is
meaning
is singular at
has codi-
neither
!
0 0 0 0
0 0 0 0
p.
(iii) Finally, consider the variety
tions
C P3
de ned by the three equa-
Z0 Z3 Z1 Z2 = 0 (I)
Z12 Z0 Z2 = 0
(II)
2
Z2 Z1 Z3 = 0 (III)
Ui := {Zi 6= 0} of P3 . In U0 , we can divide by Z0 ,
Z12
2
so that (II) becomes Z2 =
. Together with (I), this gives Z2 =
Z0
Z2 Z2 = Z1 Z2 ZZ01 = Z0 Z3 ZZ01 = Z1 Z3 . Consquently, (III) is redundant on
U0 . Now, for any point in U0 C , you can check that the matrix in
6.3.5(i)(b) has rank 2, showing that C is smooth of dimension 1 at all
of those points. To nish, and show that C is a 1-dimensional smooth
variety, carry out a similar analysis in each of U1 , U2 , and U3 (exercise).
and the covering
1 is called a curve,
d 3 a d-fold. So-called
Somewhat unsurprisingly, a variety of dimension
of dimension
a surface, and of dimension
Calabi-Yau threefolds play a central role in mathematical string theory.
6.4. Singularities of plane curves
Consider a curve
C = {F (Z) = 0} P2
F S3d (of degree 3 in Z0 , Z1 , Z2 ).
A point p C is a singularity if and only if FZ0 (p) = FZ1 (p) = FZ2 (p) =
0, and (moving C by a projectivity if necessary) we may assume that
p = [1 : 0 : 0]. To locally analyze C at p, we can pass to a ne coordiZ
Z
nates x = 1 , y = 2 and replace F by
Z0
Z0
de ned by a homogeneous polynomial
f (x, y) = F (1, x, y) =
d
X
m=k
fm (x, y),
84
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
fm S2m for each m, and fk 6= 0. Now 0 = fx (p) = fy (p) = f (p)
(p = (0, 0)) translates to 0 = f1 = f0 , so that k 2. We say that p is
a k -tuple point of C , or a singularity of order k .
where
So far, we have not de ned tangent planes at singular points. Indeed, this can really only be done for curves in general.
0 should be, we think of [x : y] as
1
homogenous coordinates on the P of lines through (0, 0) = p. The
lowest-order homogeneous term fk of f de nes a 0-dimensional vari1
ety p (C) := {fk (x, y) = 0} in this P . For each [x0 : y0 ] p (C),
what the tangent lines to
one should think of
To decide
at
y0
as the slope of a line tangent to some local
x0
irreducible component
4 of the curve
at
p.
C at a
of p (C).
Definition 6.4.1. The tangent lines to a curve
are the lines through
corresponding to points
singularity
fk (x, y) = 0 has k solutions counted with multplicity. If these
are all distinct, i.e. if p (C) is reduced, then we say p is an ordinary
k -tuple point. The most geometric way to think of this is that C has
k distinct tangent lines at p.
Any line through a k -tuple point p other than one of C 's tangent
lines there, meets C with multiplicity k at p: if L is given parametrically
by t 7 (at, bt) (fk (a, b) 6= 0) then the intersection multiplicity is comk
puted as in Chapter 2 by taking the order of f (at, bt) = t fk (a, b) +
at t = 0.
Now,
ab xa y b it
can be useful (for various purposes) to plot the nitely many (a, b) with
ab 6= 0. If you do this when (0, 0) is a k -tuple point and f has degree
n, then these lie in the shaded region
Remark 6.4.2. Given a polynomial
4this
f (x, y) =
will be made precise when we do local normalization
(a,b)Z20
6.4. SINGULARITIES OF PLANE CURVES
85
d
k
This may be useful for one of the exercises below.
There is more to singularities, it turns out, than their order or even
the tangent line con guration re ected by
p (C).
A local analytic clas-
5
si cation of so-called simple singularities of curves has been carried
out.
For the purposes of this classi cation, if
are two curves
p are considered equivalent if
0
2
there are small neighborhoods U, U of (0, 0) in C and a biholomor'
0
0
phism U U carrying p to p and C to C . The dierent classes of
through
p = (0, 0),
C, C 0
their singularities at
simple singularities carry A-D-E labels, which re ect a relation to
other classi cations in mathematics (simple Lie algebras/Dynkin diagrams, rational surface singularities, etc.)
The results are that double points are all equivalent to one of
An : x2 + y n+1 = 0
(n 1),
and (simple) triple points to one of
Dn : y(x2 + y n2 ) = 0
(n 4),
E6 : x3 + y 4 = 0,
E7 : x(x2 + y 3 ) = 0,
E8 : x3 + y 5 = 0.
The ODP's (ordinary double points: two distinct tangents) are all of
type
A1 , as all An2
have only one tangent; amongst the latter, cusps
are the singularities of type
5cf.
A2 .6
OTP's (ordinary triple points) are all
[Barth, Hulek, Peters, and van de Ven, Compact complex surfaces, Springer,
2004] for the de nition. Simple singularities encompass all double (2-tuple) points
and some triple (3-tuple) points, and nothing of higher order.
6refer
back to Chapter 2 for a few pictures (cusps, ODP, OTP)
86
6. MORE ON PROJECTIVE ALGEBRAIC VARIETIES
D4 ; we note that the tangent lines to y(x2 + y 2 ) = 0 have slopes
0, i, i. All Dn5 have two distinct tangents (one with mutliplicity 2)
and the E6,7,8 each have one tangent (of multiplicity 3).
of type
Exercises
Pn de ned by a symmetric
bilinear form B is smooth if and only if det(B) 6= 0.
(ii) Corollary 6.2.2 associates a number k to each projective quadric
n
hypersurface in P . Show that any two are projectively equivalent
if and only if they have the same value of k . [This is easy.]
2
2
3
Show that [the closure in P of] y = 4x + ax + b is smooth unless
a3 + 27b2 = 0.
4
4
4
Find the tangent plane to the complex surface 2x +y +z 4xyz =
0 (in C3 ) at the point p = (1, 1, 1).
Finish the proof in Example 6.3.6(iii) that C is a smooth curve.
2
What form does a degree k projective algebraic curve (in P ) take
if it has a singularity of order k ?
2
2 2
2
3
2
Analyze the singularity of C = {(x + y ) + 3x y y = 0} C
(1) (i) Prove that a quadric hypersurface in
(2)
(3)
(4)
(5)
(6)
at the origin. (What is its order, and type?)
(7) For which values of
are the algebraic curves
singular (in (a) and (b) below)?
F (X, Y, Z) = 0
in
Attempt a sketch of each of
the singular curves, saying where the singularities are located and
what type they are.
F (X, Y, Z) = X 3 + Y 3 + Z 3 + (X + Y + Z)3 ,
3
3
3
(b) F (X, Y, Z) = X + Y + Z + 3XY Z .
(a)
CHAPTER 7
Smooth varieties as complex manifolds
This Chapter starts the long slog toward a proof of part (A) of the
Normalization Theorem 3.2.1. After introducing a bit of the theory of
several complex variables, we'll use the holomorphic implicit function
theorem to put a complex manifold structure on any smooth irreducible
(a ne or projective) algebraic variety:
Theorem 7.0.3. A smooth irreducible algebraic curve
C Pn
is
M
: M , P with C
a Riemann surface. (More precisely, there exists a Riemann surface
and an injective morphism of complex manifolds
as its image.)
This is, of course, the smooth case of Thm. 3.2.1(A). As for going
the other way, from Riemann surfaces to algebraic curves, here is a
statement which is dierent in character from 3.2.1(B):
M with n + 1 linearly indef0 , . . . , fn K(M ), yields an algebraic
Theorem 7.0.4. A Riemann surface
pendent meromorphic functions
curve in
Pn
not contained in any proper linear subvariety.
We won't prove this in full just the existence of a morphism
M Pn
of complex manifolds which is nondegenerate, i.e.
image is not contained in any
n1
whose
. Proving that the image is described
by algebraic equations (hence yields an algebraic curve) is harder.
7.1. Background from several complex variables
On (or C{z1 , . . . , zn }) denote the ring of convergent power series
I aI z in z1 , . . . , zn , or equivalently, holomorphic functions de ned on
n
some neighborhood of 0 C (cf. 5.1).
Let
Proposition 7.1.1. [W. Osgood, 1900] Let
be a function on
0C
which is holomorphic in each zi as the
f
other {zj }j6=i are held xed; that is, z = 0 (i). Then f is in fact a
i
an open neighborhood of
holomorphic function (and so gives an element of
87
On ).
88
7. SMOOTH VARIETIES AS COMPLEX MANIFOLDS
Proof. We will only give the proof for
phic in
z2 ,
we have
f (z1 , z2 ) =
n = 2.
Since
is holomor-
f (z1 , 2 )
d2 ;
2 z1
2 1
z1 , this
using the holomorphicity in
(2 1)2
=
(2 1)2
f (1 , 2 )
d1 d2
(1 z1 )(2 z2 )
f (1 , 2 )d1 d2
1
1 2 1 z11
z2
2
.
Now using the power-series expansion
1
1
zi
i
X zi k
k0
whose uniform convergence allows us to swap integration and summation, we nd
X
f (z1 , z2 ) =
k1 ,k2 0
(2 1)2
f (1 , 2 )d1 d2
1k1 +1 2k2 +1
z1k1 z2k2 .
In order to put a complex manifold structure on a smooth variety,
we will need a way to parametrize zero-loci of holomorphic functions.
This is given by the holomorphic implicit function theorem which here
I will just state and prove in the two variable case.
Proposition 7.1.2. Let
Then there exists
w O1
f O2
f (0, 0) = 0,
such that in a neighborhood of
f (z1 , z2 ) = 0
The upshot of this is that
on
with
z2
f
(0, 0)
z1
6= 0.
(0, 0) in C2 ,
z1 = w(z2 ).
gives a local holomorphic coordinate
{f (z1 , z2 ) = 0}.
Proof. We will assume the
just check that the
implicit function theorem, and
it yields is holomorphic:
0=
=
f (w(z2 ), z2 )
z2
f
f
w
f
w
(w(z2 ), z2 ) +
(w(z2 ), z2 )
+
(w(z2 ), z2 )
.
z2
z1
z2 z1
z2
7.2. SMOOTH NORMALIZATION
Now since
f
z1
f O2 ,
locally. So we nd that
f
z2
w
z2
= 0; moreover, by
= 0, so that w O1 .
89
assumption
f
z1
6= 0
Here is a visual explanation of why the nonvanishing condition on
f /z1
matters:
z2
z2
z1
f (0)=0
z1
f (0)=0
z1
In the left-hand picture, you can write
z1
z1 as a function of z2 (as desired);
on the right-hand side, you cannot.
7.2. Smooth normalization
The more general statement which implies Theorem 7.0.3 is:
Theorem 7.2.1. Given
n-manifold X ;
a system of open neighborhoods {W X} covering Y (with local
holomorphic coordinates z = (z1 , . . . , zn ));
holomorphic functions f1 , . . . , f` O(W ) (for each ) such
1
that Y W = V ({fj }j=1,...,` ) W ; and (also for each )
a closed connected subset
of a compact complex
rank {fj /zk }
= ` [the Jacobian condition ].
j = 1, . . . , `
k = 1, . . . , n
Then Y is an (n `)-dimensional compact complex (n `)-manifold.
We say that
is a codimension-` complex submanifold of
X.
In
fact, Theorem 7.2.1 immediately gives:
Corollary 7.2.2. Any smooth irreducible projective algebraic va-
riety
Y Pn
of dimension
1this condition makes Y
the vanishing locus
is a compact complex
into an analytic subvariety of
f1 = = f` = 0,
d-manifold.
X ; here V (f1 , . . . , f` ) means
just as in the algebraic setting.
90
7. SMOOTH VARIETIES AS COMPLEX MANIFOLDS
Proof. Put
X = Pn and ` = nd.
That
is smooth of dimension
(Defn. 6.3.5) implies the Jacobian condition required in Thm. 7.2.1.
Now we prove the Theorem.
Proof. Re ning the covering if necessary, we can arrange to have
det
(7.2.1)
Write z for
I
z I , z II
(z1 , . . . , z` )
fj
zk
6= 0.
1j,k`
and z
II
for
(z,`+1 , . . . , zn ).
(So
z =
A schematic picture:
z
z
By the condition (7.2.1), and the general holomorphic implicit function
theorem, we have holomorphic functions
subsets of
Cn`
to
C` )
{w }
(mapping from open
such that
Y W = z I = w z II
for each . Hence, the z
give local coordinates
II
which constitute an open cover of Y .
Consider the transition functions for X
on the
{Y W },
'
z , z
I
II
: z (W ) z (W )
7 I z , z
, II z , z
=: z I , z II
I
II
corresponding to change of coordinates on
describing change of coordinates on
Y : z
II
This is 1-to-1 because
7 II w
II
W .
Clearly the functions
Y W are then
, z
=: z II .
z
II
II
is, and holomorphic because
So we have the data of an analytic atlas on
Y.
II
and
are.
7.3. NONDEGENERATE MORPHISMS
91
7.3. Nondegenerate morphisms
The statement related to Theorem 7.0.4 which we shall prove is:
Proposition 7.3.1. Given a Riemann surface
M,
the following
data are equivalent:
(a)
n+1
linearly independent meromorphic functions
fi K(M );
(b) a nondegenerate holomorphic map (morphism of complex manifolds)
: M Pn .
We will need the notion of a meromorphic function on a complex
manifold of any dimension.
Definition 7.3.2. A meromorphic function
plex manifold
X)
{(U , g , h )}
X;
is a collection
{U } is an open cover of
g , h O(U ) (they are
g h = g h on U .
g
2
We write F =
on U .
h
F K(X)
(on a com-
such that
holomorphic functions); and
dim(X) = 1, this coincides with the earlier
De nition 3.1.1 (via g/h); by Prop. 3.1.8 meromorphic functions on
1
Riemann surfaces yield morphisms X P . But this does not generalize: if dim(X) > 1, a meromorphic function on X need not even
1
yield a well-de ned mapping X P .
Remark 7.3.3. For
X = C2 with complex coordinates x, y .
Then F := x/y (one U = X ; g = x, h = y ) de nes a meromorphic
1
function, which is not well-de ned (as a mapping to P ) at (0, 0).
Example 7.3.4. Consider
=2
=1
= 12
x
???
2the
third condition says that
g
h
are de ned! (see Remark 7.3.3)
g
h on overlaps at least, where the quotients
92
7. SMOOTH VARIETIES AS COMPLEX MANIFOLDS
The notion of blowing up in algebraic geometry is motivated (in
part) by the desire to remove such indeterminacies. In the last example,
X = C2
P1 of lines
(mapping down onto X ) on
through the origin, yielding a new space X
the idea would be to replace the origin in
by the
which the meromorphic function becomes well-de ned as a morphism.
Example 7.3.5. The meromorphic functions on
Pn
and its smooth
subvarieties (viewed as complex manifolds) are the rational functions
P (Z)
Z
d
F = Q(Z)
for P, Q Sn+1 . For instance, the a ne coordinates zi = i
Z0
Z
are meromorphic functions (and more generally, zji = i is one).
Zj
Here is how to see at least that rational functions are meromorphic in the sense of De nition 7.3.2. (That meromorphic functions are
rational is more nontrivial.) In
Uj = {Zj 6= 0},
set
gj (zj ) := P (zj0 , . . . , 1 , . . . , zjn ) = P (Z/Zj ) =
j th
entry
hj (Zj ) := Q(zj0 , . . . , 1 , . . . , zjn ) =
j th
entry
then
gj hi =
1
P (Z)
Zjd
1
Q(Z);
Zjd
1 1
P (Z)Q(Z) = gi hj .
Zjd Zid
f : C X from a
Riemann surface to a complex manifold, and let F K(X) be given
by {(g , h , U )}. Assume that f (C)|U {h = 0} is a nite point
1
set, and put W := f
(U ), G := g f , H := h f . Then
f F := {(G , H , W )} (or rather, G/H ) belongs to K(C).
Example 7.3.6. Consider a holomorphic map
The last two examples will be used in the proof of Prop.
7.3.1,
which we now give:
n + 1 meromorphic
points in M which cause a
Proof. The rst issue is how we get from
functions to a morphism to
. The set of
problem is
:= {q M | fi (q) = 0 for
all
i} {q M | fi (q) = for
De ne
f : (M \) Pn
some
i}
7.3. NONDEGENERATE MORPHISMS
93
by
p 7 [f0 (p) : : fn (p)].
q let z be a local holomorphic coordinate with z(q) = 0, then
(f )
write fi (z) = z q i hi (z) (where hi are local holomorphic functions not
vanishing at q ), and put := mini{0,...,n} {q (fi )}. For z 6= 0,
f (z) = z f0 (z) : : z fn (z) ;
Near
none of the entries in this blows up locally, and at least one does not
vanish at
z = 0
(i.e.
at
q ).
Hence,
extends to all of
evident that this extension is still holomorphic as a map
Next, given a morphism
f : M Pn ,
M , and
1
to P .
it is
we want to product an
(n + 1)-tuple of meromorphic functions. Referring to Examples 7.3.5
(for zi ) and 7.3.6 (for f ), simply take fi := f zi and you're done.
Finally, to see that f is degenerate i the {fi } are linearly depenn+1
dent, consider the correspondence between nonzero vectors v C
n
n1
(up to scale) and hyperplanes in P , by taking Pv
to be the projectin+1 v
cation of (C
) . Degeneracy of f occurs i f (M ) Pvn1 for some
v , which is to say (f0 (p), . . . , fn (p)) v for all p M . But this just
P
reads
vi fi (p) = 0 (p), which is a nontrivial linear relation.
We give two examples of nondegenerate projective embeddings of
Riemann surfaces (the rst is actually a series of examples). For these
cases we actually give algebraic equations for the image.
Example 7.3.7. The so-called rational canonical curves are the
images of the nondegenerate morphisms
f : P1 , Pn
given, for each
n N,
by
[Z0 : Z1 ] 7 [Z0n : Z0n1 Z1 : : Z1n ].
(In a ne terms, one can think of this as
z 7 [1 : z : . . . : z n ],
with
7 [0 : : 0 : 1].)
Let's see what this looks like for the rst few values of
map.
for
n = 1, f
sends
[Z0 : Z1 ] 7 [Z0 : Z1 ]
n:
and so is just the identity
94
7. SMOOTH VARIETIES AS COMPLEX MANIFOLDS
for n = 2, we have [Z0 : Z1 ] 7 [Z02 : Z0 Z1 : Z12 ]. If we write
[Y0 : Y1 : Y2 ] for the homogeneous coordinates on P2 , then the image is
2
2
the conic {Y1 Y0 Y2 = 0} P .
for n = 3, [Z0 : Z1 ] 7 [Z03 : Z02 Z1 : Z0 Z12 : Z13 ](= [Y0 : Y1 : Y2 : Y3 ])
has image
V := V (Y0 Y3 Y1 Y2 , Y12 Y0 Y2 , Y22 Y1 Y3 ) P3 .
By exercise 4 from Chapter 6 you know that
Example 7.3.8. Let
1-torus.
is smooth.
a lattice) be a complex
We want to demonstrate that there is a (nondegenerate) mor-
phism from
present
M = C/ ( C
to
P2
with a cubic curve as image. Note that this will
as the normalization of such a cubic curve:
Some of the steps will be exercises.
First, there exists a unique meromorphic function
K(C)
satis-
fying
(u + ) = (u) for every and u C
(u) = u2 + h(u), where h K(C) is holomorphic
borhood of 0, has all its poles in \{0}, and h(0) = 0.
Existence is an exercise. Uniqueness is easy: if
in a neigh-
were another such
Q = ( u ) (Q u ) has no pole at 0 and is periodic, hence has no poles in either. But the only possible zeroes
were in , and so Q is entire. By compactness of a fundamental
region for , any -periodic entire function is bounded hence (by Liouville) constant. Since Q is zero at 0, this constant is zero and
= Q.
function,
EXERCISES
95
is an even function
some a, b C. In each
In the exercises below, you will also show that
((u)
= (u))
and
(0 )2 = 43 + a + b
for
case, you get equality by showing the right-hand side minus the lefthand side has no poles and is zero at some point (as in the uniqueness
argument just described). The upshot is that
f : C/ P2
de ned by
u 7 [1 : (u) : 0 (u)]
for
u 6= 0
and
0 7 [0 : 0 : 1]
parametrizes (or normalizes)
C = {Z0 Z22 = 4Z13 + aZ1 Z02 + bZ03 },
smooth cubic with the a ne equation
y 2 = 4x3 + ax + b.
f (M ) C ,
manifold, and f as
What we have said so far only gives that
the smooth curve
M C,
as a complex
but viewing
a morphism
the open mapping theorem from complex analysis says the
image is open; while on the other hand the image of a compact set by
a continuous map is compact (hence closed in
and closed in
C,
and thus
C ).
So
f (M )
is open
f (M ) = C .
Exercises
f : P1 Pn has the following
property: the image of any collection of k ( n + 1) distinct points
{w1 , . . . , wk } P1 is in general position (spans a Pk1 in Pn ). [Hint:
Vandermonde determinant.] Then, taking k = 2, explain why this
shows f is injective.
Turning to the case n = 3 in Example 7.3.7, (a) actually prove that
V = Image(f ) and (b) that you cannot throw out any of the three
equations de ning V .
2
5
2
Show that the map P P given by [Z0 : Z1 : Z2 ] 7 [(Z0 ) :
Z0 Z1 : Z0 Z2 : (Z1 )2 : Z1 Z2 : (Z2 )2 ] is (a) well-de ned and (b)
(1) Show that the rational canonical map
(2)
(3)
holomorphic (i.e.
a morphism of complex manifolds), then (c)
write (polynomial) equations expressing the image as an algebraic
variety. (For (c) you can just write the equations and not prove it.)
96
7. SMOOTH VARIETIES AS COMPLEX MANIFOLDS
(4) Let
1 , 2 C be two complex numbers which are R-linearly inde-
pendent, and let
= Z1 + Z2 = {n1 1 + n2 2 | n1 , n2 Z}
be the lattice in
that they generate.
(a) Show that the series
1
(u) = 2 +
u
1
1
2
2
(u )
6= 0
is absolutely and uniformly convergent on any compact subset of
the complex
u-plane which does not contain any of the points of .
[Hint: any compact subset is contained inside one of the following
|u| K |u | (). Break the sum into terms
with || 2K , and || > 2K , and use (essentially) the Weierstrass
form:
M-test.]
(b) Verify the pole condition in Example 7.3.8: that all poles are
on
and in a neighborhood
holomorphic and
h(0) = 0.
of
0, (u) = u2 + h(u)
with
[Hint: what do you know about an
absolutely and uniformly convergent series of analytic functions?]
(c) Show that
is a doubly-periodic function; that is, show that
(u + ) = (u)
[Hint:
for every
uC
and every
From (a), you can calculate the derivative
ferentiating each term of the series de ning
(u + ) = (u),
(u).
0 (u)
by dif-
First prove
then integrate.]
(5) Now forget the explicit formula for
(u) just given,
and retain just
K(C) is -periodic with all poles , and
2
locally of the form (u) = u
+ h(u) with h holomorphic (on
some U C containing a fundamental domain) and h(0) = 0.
0
Prove that (a) (u) = (u) [ = h even = h odd] and (b)
(0 (u))2 = 4((u))3 + a(u) + b for some a, b C. [See hint given
these facts: that
in the Example.]
CHAPTER 8
The connectedness of algebraic curves
The main theorem of this chapter will be that the smooth part
C\sing(C)
of an irreducible algebraic curve
(and then, of course, so is
C ).
C P2
For example, in Exercise 5 of Chapter
3, you showed that the complement of the ODP
p = [1 : 0 : 0]
in the
{Y Z X Z + X = 0}, viewed as a complex
isomorphic to C which is certainly connected.
singular cubic curve
1-manifold, is
is path-connected
Just so that there is no confusion, we should say what the situation
is for reducible curves right away and why the result does not generalize.
For plane projective algebraic curves with more than one irreducible
component, say
C = Ci ,
the components
Ci
must intersect (this will
be one consequence of Bezout's theorem later), making
But the complement of the singularities in
connected.
will not be connected, as
these will include all of the intersection points.
We begin by introducing a new, somewhat technically involved, tool
for dealing with singularities, intersections, and projections of curves.
8.1. Resultants and discriminants
Let
be a unique factorization domain (UFD), where we recall
that this is a commutative domain in which each element has a unique
factorization into irreducibles, up to reordering and multiplication by
units. In a UFD, amongst other things, the notion of a greatest com-
2 has meaning. By the Gauss lemma,
mon divisor
In practice we will always take
to be
or
C[x].
D[y]
is also a UFD.
(Note that
C[x]
is a
C[x, y] is not.)
m
m1
Consider f (y) = a0 y + a1 y
+ + am , g(y) = b0 y n + b1 y n1 +
+ bn elements of D[y] with a0 , b0 6= 0.
PID, but
1we will show that the set sing(C) of singular points is always nite
2recall that these are well-de ned up to units (invertible elements); for
C[x] or C[x, y] the units are C , hence the notion of monic gcd
well-de ned).
97
example in
(which is completely
98
8. THE CONNECTEDNESS OF ALGEBRAIC CURVES
Definition 8.1.1. The resultant
the element of
3 of
and
g,
given by the determinant of the
R(f, g), is
(n + m) (n + m)
written
4
Sylvester matrix
M(f,g)
am
0 a0 a1
.. . .
..
.
.
.
0 0
:=
b0 b1
0 b0 b1
.. . .
..
.
.
.
0 0
am
Now writing
a0
a1
a0
bn
bn
R(f, g) = 0
h)
Proof. The gcd (say,
..
.
.
.
..
..
..
bn1
for the eld of fractions of
Proposition 8.1.2.
am1
b0
D,
am
.
0
.
.
0
bn
we have the
gcdK[y] (f, g)
6= 1.5
is nontrivial if and only if
F g = Gf
(8.1.1)
F = A0 y m1 + + Am1 and G = B0 y n1 + + Bn1 in
D[y]. Indeed, if h 6= 1 then put F = f /h and G = g/h. Conversely,
since deg F < deg f and deg G < deg g , and both sides of (8.1.1) factor
into the same irreducibles, f and g have a common factor of degree
> 0.
for some
In turn, (8.1.1) is equivalent to
(8.1.2)
a0 B0 = b0 A0
a1 B0 + a0 B1 = b1 A0 + b0 A1
.
.
.
am Bn1 = bn Am1
n1
{Ai }m1
i=0 , {Bj }j=0 D. To get from
m+n1
coe cients of y
, y m+n2 , . . . , 1.
being satis ed for some
(8.1.2), just take
(8.1.1) to
3also called eliminant, since y is eliminated
4the line in the matrix is just an organizational device it has no meaning
5two further equivalent conditions: (i) deg (gcd
(f, g)) > 0; and, noting
y
K[y]
is a PID, so that the ideal
D[y]
(f, g)K[y] = (gcdK[y] (f, g)),
(ii)
that
(f, g)K[y] 6= (1)K[y] .
8.1. RESULTANTS AND DISCRIMINANTS
99
Now notice that (8.1.2) can be rephrased in matrix multiplication
terms: there exist
{Ai }, {Bj }
such that
B0
.
.
n1
t
M(f,g) .
A0
.
.
Am1
In other words, we have shown
i.e.
det(M(f,g) ) = 0.
Definition 8.1.3.
= 0.
h 6= 1 is the same as ker(t M(f,g) ) 6= {0},
D(f ) := R(f, f 0 ) is the discriminant
of
f.
Here
f
.
denotes the formal derivative
y
Example 8.1.4. If
f C[y],
then
D(f ) C
is a number, and the
criterion
(8.1.3)
D(f )
vanishes
has a multiple root
follows immediately from Prop. 8.1.2. For the a ne curve
z 2 = 4y 3 + ay + b
to be singular, we need two of the roots of the right-hand side to coincide. That is, by (8.1.3), we need
4 0 a
4 0
3
2
0 = R(4y + ay + b, 12y + a) = 12 0 a
12 0
12
a
0 a
b
a b
which after a bit of row-reduction
4
0
0
0
0
0 a
b
0
4 0
a
b
0 2a 3b 0 = 16(4a3 + 12 9b2 )
0 0 2a 3b
0 12
0
a
= 64(a3 + 27b2 ).
100
8. THE CONNECTEDNESS OF ALGEBRAIC CURVES
This recovers the result from Exercise 2 of Chapter 6.
Example 8.1.5. If
f C[x, y],
then
D(f ) C[x]
is a polynomial
and from Prop. 8.1.2 we have:
D(f )
(8.1.4)
vanishes at
The collection of
D(f ),
curve
x0 's
x0 f (x0 , y)
has a multiple root in
y.
where this happens, that is, the set of roots of
is called the discriminant locus for the projection of the a ne
{f (x, y) = 0}
y
onto the
x-line:
f=0
x
discriminant locus
Proposition 8.1.6. An irreducible (reduced) algebraic curve
0} P
{F =
has (if any) nitely many singularities.
f (x, y) = F (1, x, y) has multiple
roots in y for x in the discriminant locus = {(D(f ))(x) = 0} C.
We may assume f has positive degree in y , since otherwise V (f ) is just
Proof. The a ne polynomial
a vertical line.
Since
is irreducible in
C[x, y]
cal vanishing of
would imply that
reduced. So
D(f )
D(f ) is a
nontrivial polynomial, and
(8.1.5)
#{x C | y
such that
It is easy to argue directly that were
},
6or
then
(x )
would divide
y , the identiV (F ), was non-
of positive degree in
V (f ),
hence
is nite:
f (x, y) = fy (x, y) = 0} <
V (f ) to contain a vertical line{x =
(contradicting irreducibility). So by
you can wait for Study's lemma in the next Chapter
8.2. MONODROMY AND CONNECTEDNESS
101
(8.1.5) and Prop. 2.1.8, in fact
#{p C2 | f (p) = fy (p) = 0} < .
V (f ). The only possible
additional singularities of V (F ) are the ( nitely many) points where it
meets the line at .
The set in brackets includes all singularities of
8.2. Monodromy and connectedness
Let
be a region, that is, an open connected subset.
holomorphic function,
when
be a small disk about a point
f O().
on which one is given a
We are interested in the question of
extends to a holomorphic function on all of
doesn't always happen, take
Let
=C
and
To see why this
a small disk about
z = 1:
then f =
only extends to a holomorphic function on C . Even worse,
z
f = log(z) becomes multivalued on C and so (as a holomorphic
function) only extends to
C\R0 .
To give a condition which will ensure the existence of a well-de ned
holomorphic extension, we need the concept of analytic continuation.
from p to q to be the image of a continuous
function P : [0, 1] with P(0) = p and P(1) = q . (Here we are
allowed to pick q = p.) An analytic continuation of f along
De ne a path
,f
an
aly
tic
co
n
tin
ua
tio
of
C
i
consists of
i ,fi
102
8. THE CONNECTEDNESS OF ALGEBRAIC CURVES
into segments {i }N
i=0 ,
a covering of by disks i i (with 0 = ), and
functions fi O(i ) (with f0 = f ) satisfying fi fi+1
i i+1 .
a partition of
If we continue
along two dierent paths from
fN O(N )
p to q
on
and compare the
q)
in each case, these need not agree. In the above example of f = log(z)
on a disk about p = {z = 1}, we can analytically continue f along
any path in C . However, if we take q = p so that the path is closed,
then we do not have fN (p) = f(0) (p): they dier by 2 1 times the
winding number of the path about z = 0, hence the multivaluedness
results, i.e. the last function
(in the neighborhood of
referred to above. This problem only occurs, however, for non-simplyconnected regions:
Proposition 8.2.1. [Riemann Monodromy Principle] Given
a region
which is simply connected, i.e.
1 () = {0}.
f O() can be
starting at p . Then
be a small disk, and assume that
continued along any path
f O()
extending
Let
analytically
there exists
f.
We will frequently use this together with the
Proposition 8.2.2. [Heredity principle] Given
f O()
F (x, y) O(C2 ),
with
F (x, f (x)) = 0.
(8.2.1)
Then the analytic continuation of
along any path
must satisfy
(8.2.1).
fi in the analytic continuation are holomorphic, so is each F (x, fi (x)) (on i ). But F (x, f (x)) 0 on =
0 by assumption, and since f = f0 f1 on 0 1 , we have
F (x, f1 (x)) 0 on 0 1 and therefore (by basic complex analysis) on all of 1 . Simply iterate this argument for i = 1, . . . , N .
Proof. Since
and each
Now given an a ne algebraic curve
degree
n,
it is convenient to write
polynomial in
(8.2.2)
over
C = {f0 (x0 , y) = 0}
with
f0
of
as the vanishing locus of a monic
C[x]:
f (x, y) = y n + a1 (x)y n1 + + an (x) = 0.
8.2. MONODROMY AND CONNECTEDNESS
This is acheived by performing a change of variable
x0 = x + y
103
and
f (x, y) := f0 (x0 , y) = f0 (x + y, y), which has coe cient of
y depending polynomially on ; choose so that this coe cient is 1.
n
(The main point is that in f0 (x0 , y), the y term may be zero, and we
writing
want to remedy that.)
Having put the equation of
in this form, we write
C
C
(x, y) 7 x
for the projection of the curve to the
x-axis.
Writing
D := {D(f )(x) =
0} for the discriminant locus of this projection, by (8.1.4) we have that
1
for x C\D , the bre
(x) consists of n distinct points. For some
xed disk C\D , label these points {y1 (x), . . . , yn (x)}. Notice that
R(f, f
) = D(f )(x) 6= 0 implies that f
6= 0 on {f = 0} 1 (), so
y
y
that the holomorphic IFT (Prop. 7.1.2) gives yi (x) O(). The point
here is that the roots of (8.2.2) in y are algebraic hence multivalued
functions of x, but we can take well-de ned holomorphic branches
of them over . As we shall see, the multivaluedness will intertwine
them outside .
Label the points of D = {p1 , . . . , pK }, and let be the path in
1
P consisting of segments connecting to p1 , p1 to p2 , and so on up
1
to pK . Then the region := (P \) C is simply connected. By
Propositions 8.2.1-2, the {yi (x)} extend to functions in O() which
still satisfy
f (x, yi (x)) = 0.
(8.2.3)
to "i "
= P \
1
= points of D
Analytically continued through
in
C\D,
the
(8.2.3) by the heredity principle, but may swap.
yi
continue to satisfy
104
8. THE CONNECTEDNESS OF ALGEBRAIC CURVES
Example 8.2.3.
through
1
3
3
cyclically
x.
f (x, y) = y 3 x, D = {0}, = R0 . Passing
2 1
3
permutes y1 (x) =
x, y2 (x) = e 3 3 x, y3 (x) =
7
This swapping (or permutation) of the
yi (x) gives rise to an equiv-
alence relation :
yi (x) yj (x) if one may be analytically continued
into the other in C\D . An equivalence class is just all the {y } which
are equivalent to a given yi in this sense.
Proposition 8.2.4. For any equivalence class
ordering if necessary) by
of
formed (re-
y1 (x), . . . , ym (x),
m
Y
(8.2.4)
(y y (x))
=1
C[x, y].
belongs to
Put dierently:
functions on
C\D,
while the
{y (x)}m
=1
are multivalued algebraic
the elementary symmetric polynomials in them are
not multivalued; in fact, they are polynomials!
Corollary 8.2.5.
irreducible
= C\ 1 (D) is connected ( =
connected).
Proof. [assuming Prop. 8.2.4] If
f C[x, y]
doesn't factor, then
by the Proposition there can be only one equivalence class:
{1, . . . , n}.
So the complete set of branches
{yi (x)}
E =
is acted on tran-
D, and one can therefore draw a path on
two points.
sitively by monodromy about
C\ (D)
connecting any
We now prove Proposition 8.2.4, using some theorems from complex
analysis.
In particular, recall that Rouch's theorem asserts that for
f, g O(R) on a simply connected region9
satisfying |f | > |g| on a path , f + g and f have the same number
of zeroes (counted with multiplicity) inside .
two holomorphic functions
7the
transformations of an algebraic structure arising from its transport around
loops (in this case, loops in
about points of
D)
are what is meant by the word
monodromy in general. So the Riemann monodromy principle is really a statement
about the absence of monodromy.
8we use to index E (i.e. 1, . . . , m) and i to index {1, . . . , n}
9the main point is that R should contain the interior of
8.2. MONODROMY AND CONNECTEDNESS
105
Proof. The product (8.2.4) is clearly well-de ned on
monodromy about
simply swaps its factors; hence it is
C\D, since
in O(C\D).
Write
Ym
(8.2.5)
=1
(y y (x)) =
m
X
emj (y1 (x), . . . ym (x))y j
j=0
emj (y1 (x), . . . , ym (x)) =: emj (x) denotes the elementary symmetric polynomials in the {y }. Again, because these are not changed
under monodromy, we have emj (x) O(C\D). Observe that given
D with neighborhood N (a small disk about ), the polynomials
aj (x) from (8.2.2) satisfy
where
x N = |aj (x)| M (j)
for some
M N.
Fixing
x0 N ,
F(y) = y n ,
{yi (x0 )}
so that the
put
aj = aj (x0 )
and
G(y) = y n + a1 y n1 + + an ,
are the roots of
G.
On
= {|y| = M + 1} C,
we have
|F G| = |a1 y n1 + + an | M ((M + 1)n + + 1)
= (M + 1)n 1 < (M + 1)n = |F|.
F and G have the same number
n zeroes (at y = 0!), we nd that
By Rouch,
F=y
has
|yj (x0 )| < M + 1
Consequently the
for all
of zeroes inside
j = 1, . . . , n
ek (x) O(C\D)
and
since
x0 N .
are bounded on
N (C\D) =
N \{}, and so by the Riemann removable singularity theorem extend
across {}. Doing this for each D , we conclude that ek (x) O(C).
j
So the coe cients of the y 's in (8.2.5) are entire functions of x. To
prove that they are polynomials in x, we shall have to consider their
behavior about x = . If we work in the local coordinates x
= x1 ,
y = xy about [0 : 1 : 0] in P2 , then the polynomial (8.2.2) de ning C
becomes
10
x f
10here
to
1 y
,
x x
1
1
n1
n
= y + xa1
y
+ + x an
,
x
x
n
we are essentially taking the projective completion of
U1 P2 .
and restricting that
106
8. THE CONNECTEDNESS OF ALGEBRAIC CURVES
with roots
(8.2.6)
1
.
yi (
x) = xyi
x
N be a small neighborhood of x = 0 and N
:= N \{
x = 0}. By
n
(8.2.6), the monodromy of the {
yi }i=1 about x = 0 stabilizes the subset {
y }m
y1 (
x), . . . , ym (
x)) = xk ek ( x1 ) are well-de ned
=1 , so that the ek (
holomorphic functions on N . Since deg(aj (x)) j , the x
j aj ( x1 ) are
polynomials in x
hence bounded on N . Using Rouch as above, the
m
ek ({
y (
x)}=1 ) are also bounded on N
, and thus extend to holomorphic functions on N .
1
In other words, ek (x) = ek ( ) has a pole at x = of order at most
x
k . Since ek (x) was also holomorphic on C, we have ek K(P1 ). Now
P (x)
K(P1 )
= C(P1 ) and so ek (x) = Q(x) where P, Q are polynomials; since
its only pole is at , Q is a constant. Therefore each ek C[x], and
with (8.2.5) we see that (8.2.4) is a polynomial in C[x, y].
Let
Exercises
(1) Are the real points
11 of a smooth algebraic curve
P2
necessarily
connected?
(2) For what values of
11i.e.
a, b
does
x4 + ax + b
points on the curve which can be written
have a multiple root?
[X0 : X1 : X2 ],
with all
Xi R
CHAPTER 9
Hilbert's nullstellensatz
In something of an algebraic detour, we will now prove Theorem
5.3.1 for a ne hypersurfaces. In the general case, we shall also state
(but not prove) a more precise theorem which lays out the correspondence between a ne algebraic varieties and ideals in commutative
rings.
(This is part of the foundation for scheme theory, which you
can explore further in the books by R. Hartshorne and E. Kunz.)
9.1. Resultants (bis)
8.1 let D be a
UFD with fraction eld K ; and for f = a0 Y + a1 Y
+ + an and
g = b0 Y m + b1 Y m1 + + bm polynomials in D[Y ], de ne R(f, g) :=
det M(f,g) . (In case D is itself a polynomial ring, we will sometimes write
RY (f, g) to make it clear that Y is the variable being eliminated.)
We will need some more results on resultants. As in
R(f, g) = Gf +F g
deg G < deg g , deg F < deg f .
Proposition 9.1.1.
Proof. If
R(f, g) = 0,
n1
for some
F, G D[Y ] with
then we are done by (8.1.1).
Otherwise,
write
(9.1.1)
Y m1 f = a0 Y n+m1 +a1 Y n+m2 +
Y m2 f =
a0 Y n+m2 +
+an Y m1
+an Y m2
.
.
.
f
n1
Y
g
Y n2 g
=
a0 Y n
+
n+m1
n+m2
n1
= b0 Y
+b1 Y
+ +bm Y
=
b0 Y n+m2 +
+bm Y n2
+an
.
.
.
b0 Y m
Working in
K[Y ],
divide by
a0
107
+bm .
m equations, then use elm columns (to the right of
in the rst
ementary row operations to kill the rst
108
9. HILBERT'S NULLSTELLENSATZ
Y j 's
=) apart from the
along the diagonal.
Then, divide out the
equations by the leading coe cients to the right of =, normalizing
those leading coe cients to
row-reduction on
M(f,g)
1.
That is, we have essentially carried out
in the context of the linear system, under the
assumption that its determinant is nonzero.
The new system takes the form
?? = Y n+m1
?? =
Y n+m2
(9.1.2)
.
.
.
..
?? =
where each ?? is a
K -linear
combination of the entries to the left of
= in (9.1.1). In particular, the last row of (9.1.2) is
G0 f + F0 g = 1
deg G0 m 1, deg F0 n 1. Now
multiply by the product of all the elements of D we divided the equations by, clearing denominators of G0 , F0 to give elements G, F of D[Y ].
But this product is in fact one de nition of the determinant of M(f,g) ,
namely R(f,g) . The result follows.
where
G0 , F0 K[Y ]
satisfy
We should mention the formula for the resultant of two polynomials
y,
whose irreducible factors are all linear (or constant) in
although we
will probably neither use nor prove it:
f and g
j (Y yj )
Proposition 9.1.2. If
a0 i (Y xi ), g = b0
Q
n
am
0 b0
i,j (xi yj ).
f =
R(f, g) =
decompose into linear factors
(for
xi , yj D),
then
9.2. Study's lemma
We continue to assume that
n.
Given
at
:
D,
is a UFD with
f D[Y ]
of degree
we have the ring homomorphism given by evaluation
D[Y ]
D
.
G(Y ) 7 G()
Proposition 9.2.1. (i) If
(Y ) | f (Y ).
(ii) f has at most n
f ()(= (f )) = 0,
then
roots in
D.
i.e.
is a root of
f,
9.2. STUDY'S LEMMA
109
Proof. (i) By the division algorithm,
f = q.(Y ) + r
(9.2.1)
where
deg r < deg(Y ) = 1,
i.e.
r D.
Applying
to (9.2.1), we
have
0 = f () = q().0 + r
and thus
r = 0,
so that
(Y r)
divides
f.
(ii) Follows from (i) (and the fact that
have at most
n = deg(f )
D[Y ]
is a UFD) since
can
linear factors.
Now we will specialize to the case
D = C[X];
more generally, the
results of this section will hold with any algebraically closed eld replacing
Let
C and (X1 , . . . , Xn1 ) replacing X .
F D[Y ] = C[X, Y ] = S2 .
V (F ) = C2 ,
of S2 .
Proposition 9.2.2. If
then
F =0
as an element
Proof. Suppose
of
D[Y ], F
C.
Since
vanishes on all of
C2 ,
By Prop. 9.2.1(ii), viewed as an element
has a nite number of roots in
be constants in
that
F 6= 0.
i.e.
C[X].
Some of these may
C such
F (X, )(= (F )) 6= 0 in
is an in nite eld, there exists
is not one of these roots, and then
F (X, ) itself has nitely many roots,
so there exists C such that F (, ) 6= 0. Hence, F is not identically
2
zero on C .
C[X].
Again by Prop. 9.2.1(ii),
Proposition 9.2.3. [Study's Lemma] Given
irreducible and
V (f ) V (g).
Then
divides
f, g S2 ,
with
g.
f be irreV (fi ) V (f )
Remark 9.2.4. Suppose we drop the requirement that
ducible, so that
f =
fimi (fi irreducible in
S2 ).
for each i, and by the Proposition fi |g for each
P
f |g mi , i.e. f divides a power of g .
Then
i.
This implies that
f |0 is trivial, we take g 6= 0. By Prop. 9.2.2, we have
2
which implies V (f ) 6= C hence f 6= 0. We may assume
Proof. Since
V (g) =
6 C,
that f
/ C (since a constant divides anything), and furthermore
degY (f ) 6= 0 (otherwise just swap X and Y ). Writing
f = a0 (X)Y n + a1 (X)Y n1 + + an (X)
/ C[X]
that
110
9. HILBERT'S NULLSTELLENSATZ
> 0 and a0 (X) 6= 0),
g
/ C[X].
(n
claim:1 we can assume that
I make the
claim,
f and g are of degree > 0 in Y , so by
Prop. 9.1.1 (with D = C[X]), RY (f, g) = F g + Gf C[X] for
degY F < degY f , degY G < degY g . Given any C\V (a0 ), since
C is algebraically closed there exists a root C of f (, Y ). From
V (f ) V (g) we see that (, ) V (f (, Y )) V (f (, Y )) C, so
that f (, Y ) and g(, Y ) have a common root for every C\V (a0 ).
It follows that a0 RY (f, g) C[X] evaluates to zero at every C,
hence is zero. As a0 6= 0, we nd RY (f, g) = 0 in C[X]; and then
by Prop. 8.1.2, degY (gcdS (f, g)) > 0. (Alternately, F g = (G)f
2
= f, g have a divisor of nonzero degree in Y .) But f is irreAssuming the
ducible, so divides any nonzero non-unit dividing it; we conclude that
f | gcdS2 (f, g) | g .
g C[X]\{0}. Then there exists
C\V (g.a0 ). Viewed as a function on C2 , g is constant in Y , so
g(, ) 6= 0 C. But since a0 () 6= 0, degY (f (, Y )) > 0; and
then (as C is algebraically closed) C such that f (, ) = 0. By
assumption, V (f ) V (g) and so g(, ) = 0, a contradiction.
To prove the
claim, suppose
9.3. The nullstellensatz
The proof of Study immediately generalizes to
Cn .
This yields a
version of Hilbert's nullstellensatz for hypersurfaces:
V (f ) = V (g) for f, g Sn and . . .
(i) f, g are irreducible, then f = g ( C )
N
M
(ii) f, g are not irreducible, then M, N N such that f |g , g|f .
Equivalently, f and g have the same irreducible factors.
Corollary 9.3.1. If
Proof. (i) Study
= f |g
and
g|f ;
(ii) is by Remark 9.2.4.
The point of this is that, modulo issues with powers, there is a
bijection between hypersurfaces and principal ideals (i.e. polynomials
up to multiplication by constants) in
Sn
which reverses inclusion. That
f and g are reduced (all irreducible factors
1), (f ) (g) f |g V (f ) V (g).
is, provided
multplicity
1at
this point, of course, we can't just swap
and
Y
occur with
9.3. THE NULLSTELLENSATZ
111
To get a more general perspective on this, we introduce a few new
X Cn ,
ideas. First, given a subset
we de ne the ideal of
by
I(X) := {f Sn | f (z) = 0 z S}.
For example, if
Lemma.
I(V (f )) = (f ) by Study's
it is of the form V (J) for
is reduced, we clearly have
A subset
X C
is algebraic if
J Sn . (Indeed, this is just an a ne algebraic variety.)
statement V (I(X)) = X is true (almost a tautology) for algebraic
some ideal
The
subsets.
Given any ideal
J Sn ,
we let rad(J) denote the radical of
which is the ideal comprising all elements of
J.
belongs to
Finally,
Sn
some power of which
A radical ideal is an ideal which equals its own radical.
Sn /J is a domain ( J
ideals in Sn ), and maximal Sn /J
is prime
the monoid of
J,
Theorem 9.3.2. Let
is irreducible in
is a eld.
be an algebraically closed eld.
(i) Every maximal ideal
m Sn
is of the form
(Z1 1 , . . . , Zn
Cn );
(ii) Let J ( Sn be an ideal: then V (J) 6= ;
(iii) For any ideal J Sn , I(V (J)) = rad(J).
n ) = m
(here
Corollary 9.3.3. The correspondence
ideals
subsets
{J Sn }
{X Cn }
V
induces bijections
{radical ideals} {algebraic subsets}
.
{prime ideals} {irred. alg. subsets}
This is clear from the Theorem; to see the last correspondence,
notice that
V (J1 J2 ) = V (J1 ) V (J2 ).
One can push the relation between commutative algebra and a ne
algebraic geometry much further.
For example, the ring of regular
functions on an irreducible a ne variety
V = V (P) (P
is de ned by
C[V ] := Sn /P,
a prime ideal)
112
9. HILBERT'S NULLSTELLENSATZ
and it is easy to see that this embeds (say, for
O(V ).
O(V ) given
smooth) in
Sn to
by restricting polynomial functions to , and so Sn /P is its image.)
C[V ] is sometimes also called the coordinate ring of V . Furthermore,
, the eld of
if V is the a ne part of a smooth projective variety V
) is isomorphic to the fraction eld C(V )
meromorphic functions K(V
(or V ).
of C[V ]. Usually C(V ) is called the function eld of V
(The idea is that
is the kernel of the map from
There is even a way to recover varieties from their coordinate rings;
this is the Spec operation. Very roughly speaking, the a ne story is
this: any nitely generated commutative domain
may be presented
C[z1 , . . . , zN ]/I (where I C[z1 , . . . , zN ] is an ideal), and then you
N
take V (I) C . This gives one realization of Spec(A); of course,
there are many ways of writing A in this form (dierent N , dierent
I , etc.). From the standpoint of scheme theory, Spec(A) is something
as
intrinsic, an a ne scheme which exists in the absence of any particular
embedding in an a ne space
CN .
The best resources on this are the
book by E. Kunz and the classic text by R. Hartshorne.
Exercises
X Cn , V (I(X)) = X; (ii)
J1 , J2 C[Z1 , . . . , Zn ], V (J1 J2 ) = V (J1 )
(1) Prove: (i) that for any algebraic subset
that for any two ideals
V (J2 ).
CHAPTER 10
Local analytic factorization of polynomials
Recall the idea of normalization for an irreducible algebraic curve
C P2 :
there should exist a Riemann surface
phically to
P
nonsingular C
with
as its image.
mapping holomor-
In Chapter 7 we did this for
by using the holomorphic implicit function theorem to
put a complex manifold structure on
sisted, for each
p C,
itself.
This essentially con-
in exhibiting a neighborhood
and a (bi)holomorphic parametrization of
U C.
Np C
Np P2
of
by some open set
(The holomorphicity of the transition functions was then a
consequence.)
Now suppose
has an ordinary double point (ODP) at
recall
that this is a singularity with 2 distinct tangent lines. Denoting disjoint
union by q, one has
Np C '
that is,
U1 q U2
;
0U1 0U2
U1 , U2 ( C) glued together
C , U1 and U2 must be detached:
locally looks like two disks
one point. In order to normalize
113
at
114
10. LOCAL ANALYTIC FACTORIZATION OF POLYNOMIALS
q2
C
q
q2
q1
(irreducible)
p
p
U2
U1
p
(reducible)
topological picture
schematic picture
Our overarching goal is to produce
and
as in this gure. Geomet-
rically it seems clear that the local analytic curve
even though the global curve
is not.
Np C
is reducible,
The rst step, then, will be
to nd an appropriate formalism (in terms of 2-variable power series)
for working with
Np C ,
which one might call analytic localization.
In this setting, the local equation can be uniquely factored. This will
allow us (in the next Chapter) to carry out local normalization that
is, put local coordinates on the irreducible components of
Np C .
Fi-
nally, we will patch these parametrizations together with those of open
subsets of
C\sing(C)
to obtain
C .
p. For con2
venience, replace C for the moment by its a nization in C . From
9.3, we have the coordinate ring R = C[C], and to any point p C
corresponds a maximal ideal in m R (consisting of polynomials vanishing at p). Inverting all primes not contained in m, or localizing R at
m, replaces polynomial functions by rational functions with poles anywhere but p, which roughly corresponds to replacing C by C minus any
set of points not including p. This is quite dierent from intersecting
C with an analytic ball at p, and will not produce a local factorization
of a globally irreducible C . Instead of rational functions, we need conThere are algebraic approaches to localization of
at
vergent power series. The closest construction in algebra is something
10.1. ANALYTIC LOCALIZATION
called completion (or Henselian localization).
115
If you are curious (we
won't get into this), a good reference is the book by D. Eisenbud.
10.1. Analytic localization
{f (x, y) = 0} C2
passing through p = (0, 0). The de ning polynomial f C[x, y] is,
trivially, a convergent power series; so we may consider how f factors
in O2 = C{x, y} (cf. 7.1). In fact, for purposes of examining the intersection of C with a small neighborhood of the origin, we will show that
f may be replaced by an element of C{x}[y] ( O2 ) in a particularly
It will su ce to think of
as an a ne curve
nice form:
Definition 10.1.1. The subset
W C{x}[y] of Weierstrass
poly-
nomials comprises elements of the form
y d + a1 (x)y d1 + + ad1 (x)y + ad (x)
where each
aj (x) C{x}
(d Z0 )
satis es
aj (0) = 0.
Lemma 10.1.2. Let
, > 0
(a)
f O2
with
f / 0
y -axis.
Then
such that:
f 6= 0
on (i){|x|
< , |y| = }
and (ii){x
(b) the number of roots (counted with
with
on the
|y| < ,
is constant in
for
= 0, 0 < |y| < };
multiplicity) of f (x, y) in y
|x| < .
f (0, y) are isolated: otherwise they would
have a limit point, forcing f to be identically zero. We may therefore
choose so that f (0, y) 6= 0 for 0 < |y| . To get (a)(i) from this, just
use continuity and choose su ciently small. The number of roots in
Proof. The zeroes of
1technically,
a submonoid you can multiply (but not add) elements, and it has
the identity element
1;
the notions of irreducible element and uniqueness of
factorization still have meaning.
10.2.2) and has
Since
is inside a UFD (see proof of Thm.
as its sole unit, it does indeed have unique factorization in a very
strong sense. (See the discussion after the proof of Thm. 10.2.2.)
2in
S is some subset of
/ 0 on S as f is
not vanish on S (i.e. f
general, if
the domain of de nition of a function
should read f
not identically zero on
as f does
is zero at no point of
S ,
S)
and f
f,
6= 0
on
one
S
two very dierent
meanings. Henceforth the symbols will be used with no further explanation.
116
10. LOCAL ANALYTIC FACTORIZATION OF POLYNOMIALS
(b) is computed by
2 1
x
which is continuous in
|y|=
fy (x, y)
dy Z,
f (x, y)
and therefore constant.
Lemma 10.1.3. For
as in Lemma 10.1.2, let
{y (x)}=1,...,d
be the
3
roots described in (b). Denote the elementary symmetric polynomials
in them by
ej (x) (=
1 <<j
y1 (x) yj (x)).
Then
w := y d + e1 (x)y d1 + + ed (x)
is a Weierstrass polynomial.
Proof. Note that for each
Clearly then the
ej (x)
, y (0) = 0
from Lemma 10.1.2(a)(i).
are well-de ned and satisfy
ej (0) = 0;
we must
{|x| < }. First we have
X
fy (x, y)
1
yk
dy =
(y (x))k =: k (x),
2 1 |y|= f (x, y)
show that they are holomorphic on
y (x) of the argument is (y (x))k Resy (x) ( ffy ) =
(y (x))k ordy (x) (f (x, )). Here the Newton symmetric polynomials
k (x) span the same vector space over C as the ej (x) (a general fact
from algebra). From the integral expression, the k are evidently holomorphic, and therefore so are the ej .
since the residue at each
Let
U := O2 O2
denote the units, which are just the invertible
convergent power series, or equivalently the convergent power series
with nonzero constant term. (That is, given
g O2 , g U
1
g
O2 .)
f
Lemma 10.1.4. For
such that
uw = f,
Proof. Write
Qd
=1 (y
y (x)),
and
u :=
f
w
uU
|y| }.
as above, there exists a unique
and this holds on all of
V := {|x| <
O(V \{w = 0}).
For xed
and
x, w(x, y) =
ej (x) as coe cients.
|x| < ), w(x, y) and f (x, y) have
as mutliplying this out gives the
Consequently, for each xed
3These
(with
{|x| < } in particular, one should expect
0. So the y (x) are really only well-de ned
of the disk {|x| < } (e.g., deleting the positive
may well be multivalued on
them to be permuted as
goes about
on some simply-connected subset
real numbers gives a slit disk).
10.2. UNIQUENESS OF LOCAL FACTORIZATION
y ). Therefore u 6= 0 on V , and u(x, y)
in y . Now, for any given y0 with |y0 | < ,
1
u(x, y)
u(x, y0 ) =
dy .
2 1 |y|= y y0
the same roots (in
x)
holomorphic
117
is (for each
u(x, y) is holomorphic on a neighborhood of |y| = , this formula
shows u
(x, y0 ) is holomorphic in x. By Osgood's lemma, we have u
O(V ). Since u 6= 0, it has nonzero constant term u(0, 0), and is thus a
unit. Uniqueness is clear since u
w = f and uw = f = (
u u)w = f
= u u = 0.
Since
10.2. Uniqueness of local factorization
The uniqueness of
in the last Lemma was trivial. A slightly less
trivial uniqueness question would be: can we write
as a product of a
unit and a Weierstrass polynomial in two dierent ways i.e., with a
dierent
and
u?
We cannot:
Lemma 10.2.1. Given
decomposition
f = wu
f O2
(with
f / 0
in Lemma 10.1.4 (i.e., into
y -axis), the
w W and u U)
on the
is unique.
u(0, 0) 6= 0, shrinking , (hence V )
if necessary, we have u 6= 0 on V . Thus if f = wu, the zeroes of f and
Q
w are the same. This forces w = (y y (x)) = y d + e1 (x)y d1 + +
ed (x), which makes w (hence u) unique.
Proof. Since any unit
has
Making use of the last two lemmas, we now show that
uniquely (up to units) into irreducibles
fi O2 .
polynomial de ning an irreducible algebraic
then the local piece
{fi = 0}.
CV
f O2 factors
f began its life as a
2
curve C = {f = 0} C ,
If
breaks (uniquely) into irreducible components
Provided there is more than one of them, the
fi
are no longer
polynomials, for that would contradict (global) irreducibility of
Theorem 10.2.2.
O2
is a UFD.
Proof. We must demonstrate that
f1 f`
f O2 factors into irreducibles
uniquely up to order and units.
O1 = C{x} is a UFD: given g O1 , we have a
(f )
decomposition g(x) = x 0
h(x), where h is a unit (convergent
First, note that
unique
C.
118
10. LOCAL ANALYTIC FACTORIZATION OF POLYNOMIALS
h(0) 6= 0) and 0 (f ) Z. The irreducibles in this
case are just the factors of x.
By the Gauss lemma, it follows that C{x}[y] is a UFD.
P
a b
Next, suppose that f (x, y) =
a,b ab x y O2 vanishes identically
P
b
on the y -axis; that is, 0 f (0, y) =
b 0b y . It follows that all 0b = 0
for all b, so that f = x f0 where > 0 and f0 (0, y)
/ 0. We must prove
unique factorization for f0 .
Let f O2 with f (0, y) /
0. Lemmas 10.1.4 and 10.2.1 give
f = uw uniquely. Since w belongs to the UFD C{x}[y], we have a
unique decomposition w = h1 h` into irreducibles hj C{x}[y].
Clearly also hj (0, y) /
0, and so Lemma 10.1.4 applied to each hj
gives uniquely hj = uj wj , with each wj a Weierstrass polynomial irreducible in C{x}[y] (since hj is). This yields w = (u1 w1 ) (u` w` ) =
(u1 u` )w1 w` =: uw , and by Lemma 10.2.1 u must be 1. So far
we have f = uw1 w` .
We do not know yet whether wj is irreducible in O2 . If wj =
0 00
v v (v 0 , v 00 O2 ), then wj (0, y)
/ 0 = the same thing for v 0 , v 00 .
0
0 0
00
00 00
Lemma 10.1.4 applies to yield v = u w and v = u w , so that wj =
(u0 u00 )(w0 w00 ); applying Lemma 10.2.1 yet again gives u0 u00 = 1 =
wj = w0 w00 . But w0 , w00 W C{x}[y], contradicting irreducibility of
wj in C{x}[y].
To see uniqueness, write factorizations f = f1 f` = g1 gk into
irreducibles in O2 ; we may assume f (0, y)
/ 0. Then Lemma 10.1.4
gives fj = uj wj and gi = u
i wi with wj , wi irreducible Weierstrass polynomials. We then have (u1 u` )(w1 w` ) = (
u1 uk )(w1 wk ), so
that by Lemma 10.2.1 u1 u` = u
1 uk and w1 w` = w1 wk .
By uniqueness of factorization in C{x}[y] (and Lemma 10.2.1), the
{wj } and {wi } are the same (up to reordering), and ` = k .
power series with
Note the key statement that comes out of this proof: given
with
f (0, y)
/ 0,
we have
f = uw1 w` ,
(10.2.1)
where
uU
and
wi
are Weierstrass polynomials which are irreducible
(as Weierstrass polynomials, as elements of
of
O2 ).
f O2
C{x}[y],
and as elements
Moreover, this decomposition is completely unique, up to re-
ordering of the
wi .
Finally this also comes out of the proof if
EXERCISES
119
was a Weierstrass polynomial, then u = 1 in (10.2.1), and degy (f ) =
P`
i=1 degy (wi ). This will be useful in working the following problems.
Exercises
f (x, y) = x3 x2 + y 2 is (a) irreducible in C[x, y]
reducible in C{x}[y].
3
2
Show g(x, y) = x y is irreducible in C{x}[y].
(1) Show
(2)
and (b)
CHAPTER 11
Proof of the normalization theorem
The purpose of this chapter is twofold: to nd a method for explicitly parametrizing neighborhoods of singular points on algebraic curves;
and, using this, to completely prove part (A) of Theorem 3.2.1. In fact,
we shall prove a stronger result which contains a uniqueness statement:
Theorem 11.0.3. Let
with
S =
C P2
be an irreducible algebraic curve,
sing(C) its set of singular points. Then there exists a Rie-
mann surface
and morphism (of complex manifolds)
: C P2
such that
=C
(C)
1
(b) #{ (S)} <
1 (S)) (C\S) =: C is injective (hence an isomorphism).
(c) : (C\
(a)
) is called the normalization of C , and is unique in the
(C,
=
0 , 0 ) is another, then there exists a morphism : C
sense that if (C
C 0 such that = 0 .
The pair
9.3) between ideals I
C[x, y], varieties V = V (I), and rings C[V ] = C[x,y]
, normalization
I
means taking the integral closure of C[V ] in C(V ). Taking Spec of
with a morphism to V . This
the result produces an a ne variety V
We remark that in the correspondence (cf.
procedure may be carried out for projective varieties by patching a ne
ones together, and if this is done for curves (V
just
= C ),
then
is really
constructed algebraically. While this is beyond the scope of our
course, it's instructive to look at an example.
Example 11.0.4. If we take
V = {x3 x2 + y 2 = 0} C2 ,
then
the coordinate ring
C[x, y]
x2 + y 2 )
is not integrally closed in its fraction eld C(V ).
2 + (x 1) = 0, while irreducible in C[V ][], is
C[V ] =
(x3
121
That is, the equation
solved by
y
, as
x
122
y2
x2
11. PROOF OF THE NORMALIZATION THEOREM
x2 x3
x2
cubic curve
= 1x
V is
C(V ).
in
A schematic picture of the irreducible
(0,0)
and
(1,0)
y
can be viewed as separating the branches of
x
point
at the singular
(0, 0).
In an exercise above you were asked to carry out the (local) analytic
approach, proving that
in
C{x}[y].
x3 x2 +y 2 is irreducible in C[x, y] but reducible
Here is another such example.
y 4 + x3 x2 (= 0), which
C{x}[y], into the product of
Example 11.0.5. Consider the equation
is irreducible in
C[x, y]
but reducible in
Weierstrass polynomials
(y 2 x 1 x)(y 2 + x 1 x).
x 1 x is
ishing at x = 0.
Here
regarded as a convergent power series (in
The local picture (near
(0, 0))
C{x})
van-
described by this fac-
torization is of two parking lots (topologically, these are just disks)
attached at their centers:
g1
g2
We need a procedure that gives the indicated holomorphic parametrizations of these two branches.
11.1. Overview
Informally, here is the main idea of the proof of Theorem 11.0.3.
Given an irreducible algebraic curve
with singular point
p,
we may
11.2. IRREDUCIBLE LOCAL NORMALIZATION
123
P GL(3, C)-transformation (i.e. a projectivity) of P2 to move
p 7 [1 : 0 : 0]. By another linear transformation of coordinates (cf.
8.2), we can put the a ne equation in the form
use a
(11.1.1)
f = y n + a1 (x)y n1 + + an (x) (= 0) ,
aj (x) C[x].
Now we want to normalize a neighborhood of the singularity
(0, 0).
D(f )(x) is not idenm`
m1
tically zero in C[x]. Hence the local factorization f = f1 f`
into irreducibles in C{x}[y] will have no repeated factors (all mi = 1).
Writing = {|x| < } and W = {|y| < } for su ciently
small , > 0, this corresponds to the decomposition of C W into
C1 C` , where each Ci is homeomorphic to a disk and the union
Since
is irreducible in
C[x][y],
its discriminant
attaches them only at their centres.
More precisely, writing
f = uw1 w`
10.2, the Ci
{wi = 0} of irreducible Weierstrass
polynomials. If we can write down 1-to-1 holomorphic maps
i :
is some other disk related to ) with image i ()
= Ci , and
C2 (
repeat this procedure over all singular points, then the normalization C
as in
are the zero-loci
C = C\sing(C), we have a covering
1
by holomorphic parametrizations = z (from 7.2). Composing the
i with the z whenever Ci U in nonempty, yields holomorphic tran
sition functions. Thinking of C as an abstract complex 1-manifold,
these transition functions indicate how to attach each Ci to C to
. To obtain C (topologically) from
yield a new complex 1-manifold C
this, you simply reattach the centers of the Ci .
can be constructed as follows. On
The rst step indicated in his outline, which we do not yet know
how to do, was the construction of the
{i }.
We shall now do this.
11.2. Irreducible local normalization
w = y k + b1 (x)y k1 + + bk (x) be a Weierstrass polynomial,
irreducible in C{x}[y]. Unless k = 1, the discriminant (D(w))(x) has a
zero at x = 0. Since D(w) is not identically zero, this zero is isolated,
and we can take small enough that x = 0 is its only zero on =
{|x| < }.
Let
124
11. PROOF OF THE NORMALIZATION THEOREM
Q
w = k=1 (y y (x)) which is valid in
the sense of 8.2, but not in C{x}[y]. Namely, the {y (x)} are multival1
ued on , but become well-de ned on minus a slit. (Another, more
algebraic, way to think of this factorization, if 0 < |x0 | < , is as taking
place in C{xx0 }[y].) The multivaluedness is manifested as follows: by
Now, there is a factorization
the heredity principle, going once counterclockwise around the origin
, permutes the roots of w by y (x) 7 y () (x) where Sk (=the
symmetric group on k elements). This permutation must be transitive,
i.e. a k -cycle: otherwise, it splits into a product of (smaller) cycles,
each of which gives rise to an irreducible proper factor of w in C{x}[y],
in
in contradiction to its irreducibility.
Here, then, is how to parametrize the set
Proposition 11.2.1. Let
pick any
{1, . . . , k}.
w W
{w = 0} W :
be irreducible of degree
k,
and
1
k
:= {t C| |t| < },
Then writing
C2
g:
t 7 (tk , y (tk ))
2 with image the local analytic curve
is well-de ned and injective,
C := {(x, y) | w(x, y) = 0, |x| < , |y| < },
and gives a biholomorphism (of complex 1-manifolds)
\{0}
C \{(0, 0)}.
Remark 11.2.2. Here
C \{(0, 0)}
is a complex 1-manifold by the
holomorphic implicit function theorem as in
7.2,
neighborhoods with local holomorphic coordinate
and is covered by
x.
One can regard
the last biholomorphism as giving the transition function between
and (more generally) any open set in
(0 , t)
with holomorphic coordinate
x.
Proof. (of Prop. 11.2.1) Recall that
y (x)
is well-de ned on the
:= \{x R0 }. Analytic continuation of y (x) once
counterclockwise around x = 0 yields y () (x); going around once more
k
gives y 2 () (x), and so on. Since is a k -cycle, () = and going
k
around zero k times returns us to y (x). But t does precisely this when
slit disk
1except at 0, since all y (0) = 0
2The meaning of y (tk ) will be
de ned in proof.
11.3. FINISHING LOCAL NORMALIZATION
goes around
function on
once, and so
y (tk )
125
extends to a well-de ned analytic
= k1
(j) into pie-slices
j=0
(j) := {0 < |t| < and j arg(t) j+1 }. On the interior of each
2
2
j
j+1
slice (that is, where
< arg(t) < 2 ), we can de ne a holomorphic
2
k
k
function by y j () (t ), since t 7 t maps this interior (isomorphically)
to
where y j () (x) is de ned. Extending these functions continu (j) , they patch together (in fact, analytically continue into
ously to
. This
one another) to yield a single holomorphic function y
(tk ) on
by the removable
is bounded exactly as in 8.2, and so extends to O()
A bit more carefully, we subdivide
1
k
singularity theorem.
Let
k := e
. If
(tk1 , y (tk )) = (tk2 , y (tk2 ))
then
t2 = (k )` t1
for some
` Z,
and
y ` () (tk1 ) = y (tk1 ).
{y } are all distinct away from 0, the last equation is impos`
sible unless k|`, which implies (k ) = 1 so that t1 = t2 . This proves
that g is injective.
Since is transitive, g maps surjectively onto C . It gives, Since the
nally, a holomorphic map of Riemann surfaces on the complement of
) and x (for open
0 since in local coordinates t (on
C \{(0, 0)}) we have x = g(t)'s x-coordinate = tk .
subsets covering
11.3. Finishing local normalization
Referring back to
8.1,
for each of the irreducible factors
wj
we now apply Proposition 11.2.1. This yields normalizations
j C
gj :
j
of the irreducible components of the local analytic curve
C W :
of
126
11. PROOF OF THE NORMALIZATION THEOREM
g3
C3
C2
C1
|x|<
Each restriction
=
gj :
(Cj \{(0, 0)}) , C
j
is biholomorphic with respect to local coordinates on
Cj \{(0, 0)}.
covering of
In fact, it takes the form
j and an open
t 7 tk (= x) as
indicated at the end of the last proof. These may be regarded as the
glueing maps that will attach each
holes in
to
thereby plugging the
, which is what we do next.
Before that, we just note that one should carry out the construction
of
of
gj 's
C.
p = (0, 0),
as we have done near
at all the other singular points
11.4. Global normalization (patching)
Suppose for the moment
that
C = C\{(0, 0)}.
(0, 0)
is the only singular point of
Then we put
1
2 ` ,
C := C
g1
where
1
C
g1
g2
g`
means
1
C q
,
1 )
g1 (p) p (p
1
2
C
g1
g2
means
1 q
2
C q
,
1 , q
2 )
g1 (p) p , g2 (q) q (p
and so forth.
C,
so
11.5. EXAMPLES OF LOCAL NORMALIZATION
127
If there are more singularities, then repeat this patching at each
S = sing(C).
P2 with image C , set
get a map : C
c , for c C
(c) :=
.
gj (c), for c
j
point in
To
These two prescriptions are compatible with the patching.
To see that
is compact: given an open cover
{U }
of
C ,
pick
C 0
U(q) containing each q (S). The complement
of these
is isomorphic to a closed subset of C , since is bijective away
in C
1
from
(S). Now a closed subset of C is a closed subset in P2 , P2
0
is compact, and a closed subset of a compact set is compact. So C
0 } has a nite subcover {Ui C 0 }. The {Ui }
is compact and {U C
.
together with the {U(q) } then furnish a nite subcover of C
one
We have now proved all but the uniqueness part of 11.0.3 and it is
time to backtrack and get explicit.
11.5. Examples of local normalization
Example 11.5.1. Assuming
gcd(k, a) = 1,
y k xa
is irreducible in
11.2.1. The
y1 (x) =
C{x}[y],
and we shall apply the procedure of Prop.
(multivalued) roots of
y k xa = 0
If we plug
are
xa , y2 (x) = k xa , . . . , yk (x) = (k )k1 k xa ;
they are well de ned on the slit disk
in
into
y1 (x)
{0 < |x| < , arg(x) (0, 2)}.
and analytically continue, we get
y1 (tk ) = ta .
Hence by de nition
g(t) = (tk , ta ).
We should check that the image of
the statement that
(tk )a = (ta )k .
lies in
y k xa = 0:
this is just
128
11. PROOF OF THE NORMALIZATION THEOREM
Example 11.5.2. Here is a more complicated example where there
is more than one
gj
(as in
5.3):
f = y 8 + y 4 x6 + x3 x 2 y 4 + x5 x2 .
Viewed in
C{x}[y], this is not a Weierstrass polynomial (the coe cient
1 x of y is not zero at x = 0), so we should expect a nontrivial unit
u in (10.2.1). Indeed,
2
f = (y 4 x3 + 1)(y 4 + x3 x2 )
= (y 4 x3 + 1)(y 2 x 1 x)(y 2 + x 1 x),
|
{z
}|
{z
}|
{z
}
u
where
w1
w2
u(0, 0) 6= 0.
is a unit because
w1 , w2 are irreducible Weierstrass polynomials and so we apply
11.2.1 (with k = 2) to normalize their zero-sets.
p
Beginning with w1 , the roots are y11 (x) =
x 1 x and y12 (x) =
p
x 1 x, which are swapped as x goes around 0. So y11 (t2 ) is ob2
tained by substituting t for x and analytically continuing: informally,
p
t2 1 t2 = t 4 1 t2 . So
4
g1 (t) = (t2 , t 1 t2 ).
p
p
For w2 , the roots are y21 (x) = i
x 1 x and y22 (x) = i x 1 x;
Now
and this yields
4
g2 (t) = (t2 , it 1 t2 ).
w2 = 0: one need
p
4
2
2
x(t) 1 x(t) = (it 1 t ) + t2 1 t2 = 0.
Let's check this parametrizes
only write
(y(t))2 +
11.6. Uniqueness
Begin with two normalizations:
CO
?
C \ 1 (S)
PO 2 o
?
C \S
CO 0
?
C 0 \ ( 0 )1 (S)
6
S U
X Z ] _ a d f i k
with
, 0
holomorphic maps of complex manifolds. Essentially what
we have to show is that neighborhoods of the points of
and
1 (S)
(in
C )
( 0 )1 (S) (in C 0 ) are isomorphic in a way which is compatible with
11.6. UNIQUENESS
and
0.
129
Put together with the bottom dotted arrow these isomor-
phisms will yield the desired map
: C C 0
of Riemann surfaces
making the diagram
C ?
?
??
??
??
/ 0
C
0 ~~~
~~
~ ~
C
U P2 be a small
open set containing it. For simplicity assume p = [1 : 0 : 0] and choose
coordinates so that U {|x| < , |y| < } and C U is given by the
vanishing set of a Weierstrass polynomial. Write U := U \{p}.
1
Now pick q
(p); by continuity of , 1 (U ) is open in C . So
there exists an open set V , which we may assume to be connected, with
q V 1 (U ). Since (V \{q}) C U must then be connected,
and C U is homeomorphic to a disjoint union of punctured disks,
maps V \{q} into one of these punctured disks. Consequently, V is
3
mapped into only one (local) irreducible component W of C U . This
To start, let
pS C
be a singular point and
yields the following diagram:
T
d b a _ ] \ Z X
f
+
i g
=
/ W o
{|t|
V OO
g
O O
mm
OX
mmm
m
O O
m
mm
O O
mmm t7tk
'
vmmm
prx
/ ({|x| < } Cx )
C2
(x,y)
in which
prx
composition
on
< k } Ct
and
are morphisms of complex manifolds, so that their
X is evidently a holomorphic (obviously bounded) function
V.
The composition
tion on
V.
is also evidently a bounded, well-de ned func-
By the holomorphic IFT (and holomorphicity of
holomorphic on
V \{q};
X),
it is
hence by the removable singularity theorem,
T O(V ). It is also clear that T(0) = 0. So by the open mapping
theorem, T maps V onto a neighborhood N of 0 in Ct (which we may
assume is a disk). Shrinking U (and thus W ) if necessary, we may
conclude that
the restriction of to a neighborhood of q maps
V onto W . From the diagram, this
is just g T.
3remember
that these components are homeomorphic to disks; take out
is where the punctured disks came from
and that
130
11. PROOF OF THE NORMALIZATION THEOREM
is 1-to-1 o 1 (S), no neighborhood of any other point q0
1 (S) can be sent to W . Repeating the argument above by varying
q , sets up a 1-to-1 correspondence between V 's (i.e. neighborhoods
1
of points in
(S) in C ) and W 's (irreducible local components of
C at points of S ). We can play the same game for the normalization
C 0 , and nd that for a unique q 0 ( 0 )1 (S) we have a neighborhood
V 0 and an isomorphism T0 : V 0 N whose composition with g gives
0 : V 0 W .
0 0
The piece of carrying (V, q) to (V , q ) is now de ned simply by
(T0 )1 T. This is automatically holomorphic, and its composition with
0 is g T = as desired.
Since
Exercises
f (x, y) = y 4 (x + 1)7
4
6
7
of g(x, y) = y x + x
(1) Locally normalize the zero-set of
(2) Locally normalize the zero-set
(1, 0).
at (0, 0).
at
CHAPTER 12
Intersections of curves
Now we come to the applications of normalization, which will occupy this chapter and Chapter 14. You may recall that in Chapter 2
we studied intersections of an plane algebraic curve
with a (projec-
L C were each assigned a multiplicity by
restricting the equation of C under a parametrization of L, and looking
tive) line
L.
The points of
at the multplicities of the roots of the resulting one-variable polynomial. With this de nition, the multiplicities added up to the degree of
the curve (cf. Prop. 2.1.8).
If we had tried to replace
by an arbitrary curve
at that point,
we would have run into the problem of no longer knowing how to locally
parametrize
near the intersection points. Now that we can do this
(Prop. 11.2.1), we can pull the de ning equation of
back under the
parametrization and look at its order of vanishing at the intersection
point. This leads to the general de nition of intersection multiplicity,
and with this in hand that we can nally state (and prove!) Bezout's
theorem in general. In its proof the intersection divisor will make an
appearance, so we begin with a short bit on divisors.
12.1. Divisors on a Riemann surface
Let
be a Riemann surface. The group of divisors on
free abelian group on points of
(
Div(M )
:=
X
nite
is the
M,
)
mi [pi ] mi Z , pi M .
The uncountably many symbols
[pi ]
are the generators of this (very
big) abelian group. Associated to a divisor
D=
a degree
deg(D) :=
131
mi .
mi [pi ] Div(M )
is
132
12. INTERSECTIONS OF CURVES
The resulting group homomorphism
Div(M )
(12.1.1)
deg
Z.
is called the degree map.
The divisor of a (nontrivial) meromorphic function
(f ) :=
p (f ).[p]
is given by
Div(M ),
pM
where
p (f )
is the order of
at
(Defn. 3.1.4). Note that the sum is
actually nite (as required by the de nition of divisor) since at all but
M , p (f ) = 0. Now K(M ) is a multiplicative
Sending f (f ) yields a homomorphism
nitely many points of
abelian group.
()
K(M ) Div(M )
(12.1.2)
of abelian groups, as you will show in an exercise below, which takes
multplication to addition:
(f g) = (f ) + (g), (f 1 ) = (f ).
With these de nitions, the composition of (12.1.2) with (12.1.1)
takes
0.
to
pM
p (f ), which by Exercise 3.2 is zero.
That is,
deg () =
Note that one can de ne meromorphic functions and divisors more
generally on complex 1-manifolds, but it is only in the compact case
(Riemann surfaces) that the divisors of meromorphic functions are always of degree
0.
Example 12.1.1. On
is
the easiest meromorphic function around
Z1
Z0
K(P1 ) . Writing simply 0, for the points [1 : 0], [0 : 1],
divisor is (z) = [0] [], obviously of degree 0.
z=
its
P1 ,
12.2. Intersection multiplicities
For a polynomial in one variable
f (x)
with
f (0) = 0, deg(f )
is
the exponent of the highest degree term, while the order of vanishing
ord0 (f )
:= 0 (f )
is the exponent of the term of lowest degree. Order
(unlike degree) also makes sense for power series in 1 variable.
How does all this generalize to two variables? First, a polynomial
F (x, y) can be written as a sum of homogeneous terms. If this is
F = Fk +Fk+1 + +Fd1 +Fd , then deg(F ) := d (highest homogeneous
degree) while ord(0,0) (F ) := k (lowest homogeneous degree). From 6.4,
k is also the order of singularity of the curve C = {F = 0} at (0, 0), i.e.
the number of tangent lines to C counted with multiplicity. When we
12.2. INTERSECTION MULTIPLICITIES
133
don't want to refer to the polynomial, we will write ord(0,0) C ; remember
this is
when
is smooth at
crossing) there, and so on.
(0, 0), 2
when
has an ODP (normal
Finally, ord(0,0) also makes sense for
2-
variable power series.
Now suppose
V = {f (x, y) = 0}, W = {h(x, y) = 0}
are reduced
a ne algebraic curves that intersect properly i.e. have no common
V W
p V W,
irreducible components. Then
is itself reduced. For
has no repeated components, so
(f h))(p) = fx (p)h(p) + hx (p)f (p) = fx (p).0 + hx (p).0 = 0
x
and similarly
( y
(f h))(p) = 0.
Therefore
V W
sing(V
W ),
and
Prop. 8.1.6 yields
(12.2.1)
#{V W } #{sing(V W )} < .
V
Definition 12.2.1. Assume
tinct), and let
p V W.
Let
Writing the local decomposition of
W are irreducible (and disU C2 be a neighborhood of p.
V into irreducibles (uniquely)
and
V U = V1 + + Vk ,
with local normalizations (again, essentially unique)
gi : Vi ( C2 )
ti 7 (xi (t), yi (t)),
we de ne the (local) intersection multiplicity at
(V W )p :=
k
X
ord0 (h(gi (t))).
i=1
The (global) intersection number is then de ned by
(V W ) :=
(V W )p ,
pV W
in which the sum is nite by (12.2.1).
Remark 12.2.2. (a) If either
or
is smooth, the intersection
number is actually the degree of a divisor,
V W :=
X
pV W
(V W )p [p].
134
12. INTERSECTIONS OF CURVES
This is because we can regard the smooth one (say,
surface and then
V W
V W
Div(W ).
as a formal sum of points of
W)
as a Riemann
Alternatively, you can think of
, known as a zero-cycle
1 on
P2 .
The degree is de ned in the same way as for divisors.
h gi appearing in Defn. 12.2.1 will frequently
be written gi (h) that is, we are pulling the function h back by the
local normalization gi .
(b) The composition
The local intersection multiplicities are well-de ned essentially by
the uniqueness of local normalizations. They also have some reasonable
properties:
Proposition 12.2.3.
(V W )p = (W V )p .
Proposition 12.2.4.
(V W )p
precisely when none of
at
V 's
ordp (V
tangents at
) ordp (W ),
with equality
coincide with the tangents of
p.
We will postpone proof of these results in
12.4 5, since the proofs
get a bit technical.
Example 12.2.5. Here are two pictures of smooth curves meeting
at a point
p:
distinct tangents
(I)
(II)
ordp W = 1 because the curves are smooth. But
in the rst case, (V W ) = 2, while in the second (which has distinct
tangents) (V W ) = 1.
In each case, ordp V
1zero
refers to the fact that we are taking a formal sum of zero-dimensional sub-
varieties (i.e. points) in
P2
12.2. INTERSECTION MULTIPLICITIES
Example 12.2.6. Let
a, b, m, n N
135
with
gcd(n, a) = gcd(m, b) = 1.
Then by Prop. 12.2.4, we should have
{y n = xa } {y m = xb } (0,0) min(n, a) min(m, b).
Let's check this by actually computing the left-hand side. The normalg
n
a
n a
ization of {y = x } is just t 7 (t , t ) by Example 11.5.1. Writing
h = y m xb ,
we have
g (h) = g (y m xb ) = (ta )m (tn )b = tam tbn
and the order of this at
(0, 0)
is the least of
{y n = xa } {y m = xb }
(0,0)
am
and
bn:
= min(am, bn).
This clearly satis es the inequality, and it is easy to cook up an example
where equality doesn't hold:
becomes
with
n = 3 , a = 4, m = 2, b = 9
it
8 6.
P
(V W )p to the more general setting where V = mj Vj
P
W = nk Wk with {Vj } and {Wk } irreducibles, we simply put
X
(V W )p :=
mj nk (Vj Wk ).
To extend
and
j,k
Remark 12.2.7. Here are two other approaches to local intersection
multiplicity which give the same numbers.
(a) The commutative algebra approach makes use of localization.
C(x, y) denotes the fraction eld of C[x, y]. Let p = (a, b)
C . The local ring at p, denoted Op , is the subset of C(x, y) consisting
G1
of rational functions
(here G1 , G2 C[x, y]) with G2 (p) 6= 0. You
G2
can easily check that this is a ring, and it obviously contains C[x, y].
It has a unique maximal ideal mp consisting of functions which vanish
at p.
Now let V = {f = 0}, W = {h = 0} be as above, and assume
p V W . Writing (f, h)p for the ideal in Op generated by f and h,
Recall that
we de ne
(V W )p := dimC (Op /(f, h)p )
2for
instance, the polynomial
y n xa
has order given by the smallest of
and
a.
136
12. INTERSECTIONS OF CURVES
by viewing the quotient
vector space. (Note that from
this de nition,
Op /(f, h)p as a
invariance of (V W )p
under projectivities is imme-
diately clear.)
As a simple example, we know that the intersection
multiplicity at
p = (0, 0)
of
{x = 0}
{y 2 x = 0} should be 2.
basis 1, y . See Chapter 4 of [L.
and
The quotient vector space, indeed, has
Flatto, Poncelet's Theorem] for more on this approach.
(b) For an approach via resultants, it is convenient to work with
= {H = 0}, P =
V = {F = 0},W
(in homogeneous coordinates [Z : X : Y ] on
[P0 : P1 : P2 ] V W
P2 ). Assume that [0 : 0 : 1] neither belongs to (i) C D, nor (ii) any
line joining points of C D , nor (iii) any line tangent to C or D at a
point of C D . Then we may de ne
homogeneous polynomials. Write
)P := ord[P :P ] (RY (F, H)).
(V W
0 1
C[Z, X][Y ] and RY (F, H),
Z and X ; it is in fact homogeneous and of degree deg(F ) deg(H). Its order at [P0 : P1 ] is just the
highest power of (P0 X P1 Z) dividing it.
Here we are thinking of
F, H
as elements of
3
which eliminates Y , is a polynomial in
Justifying this de nition takes a bit of work, but it leads immediately to a proof of Bezout since the intersection multiplicities have to
add up to
deg RY (F, H) = deg V deg W
by construction. This is the
point of view taken in [F. Kirwan, Complex Algebraic Curves].
12.3. Bezout's theorem
We rst do a quick recap of Prop. 2.1.8:
Proposition 12.3.1. Let
curve,
L (
= P1 ) P2
a line
C = {F (Z, X, Y ) = 0} P2 be a degree
not contained in C . Then (L C) = d.
Proof. By a change of coordinates, we may assume
and
[0 : 1 : 0]
/ C.
L = {Y = 0}
Then by the Fundamental Theorem of Algebra,
k
Y
F (Z, X, 0) =
(X i Z)di ,
i=1
Pk
di = d since F
C L = {[1 : i : 0]}ki=1 .
where
3as
i=1
has is homogeneous of degree
d.
Hence
usual you can think about this resultant in terms of a projection onto the
(or rather,
[Z : X]-)
axis.
x-
12.3. BEZOUT'S THEOREM
137
Passing to a ne coordiantes (f =
(x i )di ) and locally normalgi
izing L at (i , 0) by t 7 i + t, we have
We conclude that
(L C)(i ,0) := ord0 (gi f ) = di .
P
(L C) = di = d.
C, E P2 be properly
(C E) = deg C deg E .
Theorem 12.3.2. [E. Bezout, 1779] Let
intersecting projective algebraic curves. Then
k = deg E , and choose lines
L1 , . . . , Lk avoiding the points of C E . Write E = {H(Z, X, Y ) = 0},
Lj = {j (Z, X, Y ) = 0}. Then by Propositions 12.2.3 and 12.3.1,
Proof. Assume
is irreducible. Let
(C Lj ) = (Lj C) = deg C,
and
(C
(kj=1 Lj ))
k
X
=
(C Lj ) = deg C deg E.
j=1
Now by Example 7.3.5, the quotient of two homogeneous polynomials of the same degree gives a meromorphic function on projective
space.
is of degree
and each
:=
is of degree 1, so we may de ne
H
K(P2 ).
1 k
= C ) for the normalization, we have
: C P2 (with (C)
We can compute the divisor of this
Example 7.3.6 K(C).
Writing
by
meromorphic function if we notice that locally about each point of
CE
[resp.
[resp.
Lj ].
C (Lj )],
[resp.
1
] gives a de ning equation for
So by Defn. 12.2.1,
( ) =
pCE
q ( )[q]
qC(Lj )
ordp (
)[p]
pCE
p ( )[p] +
ordq (
qC(Lj )
(C E)p [p]
pCE
1
)[q]
(C (Lj ))q [q].
qC(Lj )
But as divisors of meromorphic functions on Riemann surfaces have
degree
0,
0 = deg(( )) =
X
pCE
(C E)p
(C (Lj ))q
qC(Lj )
138
12. INTERSECTIONS OF CURVES
= (C E) deg C deg E.
C
is reducible, break it into irreducible components and
sum the results!
Finally, if
Remark 12.3.3. In terms of zero-cycles (cf.
Bzout is saying that
C E
has degree
Remark 12.2.2(a)),
deg C deg E .
12.4. Proof of Prop. 12.2.3
We now show the symmetry of intersection numbers. Write
V =
{f = 0}, W = {h = 0}, p V W . For simplicity assume that
p = (0, 0), V and W are irreducible, and the de ning (polynomial)
equations are in the form
f = y m + B1 (x)y m1 + + Bm (x) ,
h = y n + b1 (x)y n1 + + bn (x).
We decompose these according to (10.2.1): viz.,
f = u1 v1 vr ,
h = u2 w1 ws
vj , wk are irreducible Weierstrass polynomials. For the roots
(j)
of vj [resp. wk ] on a slit disk {|x| < , x
/ R>0 } we shall write y (x)
(k)
( = 1, . . . , mj ) [resp. z (x) ( = 1, . . . , nk )]. On the non-slit x-disk
where the
these become multivalued, and we will assume that counterclockwise
analytic continuation sends y 7 y+1 to keep the numbering simple.
(j)
(k)
As in 11.2 the y
(tmj ) [resp. z (tnk )] are well-de ned on a small
t-disk {|t| < 0 },
and we have
(j)
0 mj
y+0 (tmj ) = y(j) ((m
t) )
j
mth
j root of unity mj . (This changes the branch you
(j)
m
start at when arg(t) = 0.) Write gj (t) := (t j , y
(tmj )) and Gk (t) :=
(k)
(tnk , z (tnk )) for the parametrizations of {vj = 0} and {wk = 0}.
for some primitive
We then have the key identity
(12.4.1)
mj
Y
(j)
wk tnk mj , y+0 (tnk mj )
0 =1
4Warning:
you cannot write
(j)
of y
, since this assumes
composition with the
0 mj
(m
t) = (mj )0 mj tmj = tmj
j
inside the argument
(j)
y is well-de ned on an entire disk (whereas only its
mth
j -power
map is!).
12.5. PROOF OF PROP. 12.2.4
139
mj nk n
o
Y
Y
(j)
(k)
y+0 (tnk mj ) z+0 (tnk mj )
=
0 =1 0 =1
nk
Y
(12.4.2)
(k)
vj tnk mj , z+0 (tnk mj ) ,
0 =1
which uses the factorization of each Weierstrass polynomial into a product (for each xed
of a disk by
x)
2/mj
of linear factors. Bearing in mind that rotation
does not change the order of a function at
0,
we
compute
ord0 ((12.4.1))
= nk
mj
X
ord0
(j)
wk (tmj , y+0 (tmj ))
0 =1
= nk mj ord0 wk (tmj , y(j) (tmj ))
Dividing this by
mj nk
= nk mj ord0 (gj wk ).
Pr Ps
and applying
k=1 gives
j=1
!
X
ord0
gj
wk
ord0 (gj h)
= (V W )p .
Similarly
ord0 ((12.4.2))
and dividing out
mj nk
= nk mj ord0 (Gk vj ),
(W V )p .
and summing yields
Q.E.D.
12.5. Proof of prop. 12.2.4
With the same notation as in the last section, we also write out the
irreducible Weierstrass polynomials
(j)
(j)
vj = y mj + amj 1 (x)y mj 1 + + a0 (x).
have
(j)
a0 (x) is the product of the multivalued
P
ord(0,0) vj mj ,
j ord(0,0) vj = ord(0,0) f , and
Note that
ord(0,0) (vj (x, y))
(j)
ord0 (a0 (x)) =
1
=
ord0
mj
mj
Y
roots
(j)
y (x).
1
(j) m
ord0 (a0 (t j ))
mj
!
(j)
y+0 (tmj )
0 =1
= ord0 (
y(j) (tmj )).
We
140
12. INTERSECTIONS OF CURVES
Therefore
(V W )p =
r
X
ord0
h(tmj , y(j) (tmj ))
j=1
r
X
(ord(0,0) h) min
m
ord0 (t j ), ord0 (
y(j) (tmj ))
j=1
(ord(0,0) h)
r
X
min
m
ord0 (t j ), ord(0,0) (vj (x, y))
j=1
= ord(0,0) h
r
X
ord(0,0) vj
j=1
= ord(0,0) h ord(0,0) f
= ordp V ordp W ,
Q.E.D.
Exercises
M be a Riemann surface. Show that the divisor map () :
K(M ) Div(M ) is a homomorphism of (abelian) groups. [Hint:
(1) Let
use local coordinates.]
V = {yx =
0} and W = {y x = 0}. (This will depend on C.)
2
Let C P be an algebraic curve of degree n > 1 and L a (pron
+ 1 singular points of C . (Note: bc
jective) line containing
2
(2) Compute the intersection multiplicity
(3)
(V W )(0,0)
for
is the greatest integer function, which takes the greatest integer
less than a given real number.)
Use Bezout's theorem to prove
C L hence cannot be irreducible. [Hint: prove rst that the
intersection multiplicity of L and C at each singular point through
which L passes, is at least 2.]
2
Let C P be an algebraic curve of degree 4 with 4 singular points.
Using Bezout's theorem and Prop. 12.2.4, prove that C cannot be
irreducible. [Hint: use the Hint from (3) together with a conic Q
through the following 5 points: the 4 singularities of C plus one
more point of C .]
2
A degree d algebraic curve C P can be taken to go through
(d+1)(d+2)
1 distinct points. (This is just because dim(S3d ) =
any
2
that
(4)
(5)
(d+1)(d+2)
.)
2
Prove that if all of these points are taken to lie in a
single curve
of degree
e<
d
2
+ 1,
then
is reducible.
EXERCISES
141
(V W )(0,0) for V = {y 2 x3 =
0} and (a) W = {x3 x2 +y 2 = 0} or (b) W = {(y x)3 4 2xy =
0}.
(6) Compute the intersection multiplicity
CHAPTER 13
Meromorphic 1-forms on a Riemann surface
In the next chapter we will see one more application of the normalization business, via intersection numbers: the degree-genus formula.
As more will be needed for its proof, presently we make a detour to
de ne and study dierential forms (with poles) on manifolds how to
patch them together via local coordinates, how to pull them back under
a morphism, and so forth. Like meromorphic functions, 1-forms have
an associated divisor. In contrast to the function case, the degree of
this divisor is not zero: it tells you the genus of the Riemann surface,
via the so-called Poincar-Hopf theorem.
This result will be key to
proving the Riemann-Hurwitz and genus formulae.
13.1. Dierential 1-forms
These are the expressions you integrate over paths in calculus and
complex analysis. For example, on
R2
= F (x, y)dx + G(x, y)dy
is a
1-form.
Given a dierentiable map
: R2 R2
given by
(u, v) 7 (x(u, v), y(u, v)),
the pullback of
by
is
(13.1.1)
:=
=
F (x(u, v), y(u, v)) d(x(u, v)) + G(x(u, v), y(u, v))d(y(u, v))
y
x
(u, v) + G(x(u, v), y(u, v)) u
(u, v) du
F (x(u, v), y(u, v)) u
+ F (x(u, v), y(u, v)) x
(u, v) + G(x(u, v), y(u, v)) y
(u, v) dv.
v
v
A 0-form is just a function
f (x, y),
and
f := f = f (x(u, v), y(u, v))
143
144
13. MEROMORPHIC 1-FORMS ON A RIEMANN SURFACE
. (13.1.1) is simply the analogue
with . This is exactly what you are
is nothing but precomposing with
for
1-forms
of precomposition
doing when you change variables in an integral.
We want to generalize
to complex
1-manifolds),
1-forms from R2 to real 2-manifolds (and then
which seems to call for a bit of motivation.
2-manifold, f : M R a dieren3
tiable function, and p M a point. If M R , then the notion of
taking partial derivatives of f at p in directions tangent to M makes
immediate sense you just precompose f with a (dierentiable) path
in M having a given tangent at p, and dierentiate with respect to the
Let
be a dierentiable real
variable parametrizing this path.
In abstract dierential topology, one has no embedding in
the dierentiability of
tions
R3 .
Rather,
is arranged by requiring the transition func-
relative to local coordinates on an open cover, to be smooth:
(x ,y )
(x ,y )
(This was discussed at the beginning of
2.2.)
One then de nes the
tangent spaces
Tp M := vector
space of linear dierential operators (at
=R
+
,
x p y p
and tangent bundle
T M := pM Tp M.
p)
13.1. DIFFERENTIAL 1-FORMS
145
1 (p) = Tp M . A
1
global section of T M , that is, is a smooth map : M T M with
= idM , is called a vector eld on M . (Typically one writes ~v ,
with the understanding that ~
v (p) Tp M .)
One has a projection map
: TM M
with
Now integration is dual to dierentiation, so dierentials are dual
, a dual basis (for the dual vector space)
x y
to tangent vectors. For
is
dx , dy :
we write
dx
= 1 , dy
=0,
x
x
dx
= 0 , dy
= 1.
y
y
The co tangent spaces are then
D
E
Tp M
= R dx |p , dy |p .
Global sections of the cotangent bundle
the dierential
1-forms
on
M.
T M = pM Tp M
are then
In local coordinates a dierential
1-form
looks like:
= F (x , y )dx + G (x , y )dy .
(13.1.2)
M given locally by {g : V R} must
= g |V = g |V ,
Just as a function on
g |V
{ }
the
satisfy
are subject to compatibility conditions
|V = |V .
Now since
and
(hence each
is smooth, smoothness of
(i.e. of
in (13.1.2)) is preserved under pullback, and it makes sense to
de ne
A1R (M ) :=
=
1-forms on M
{F , G } in nitely dierentiable.
smooth, real-valued
collections
For a complex
{ }
with
1-manifold,
of smooth real 2-manifold (the
which we recall from is a special kind
are conformal), the labels on the
diagram change:
1to
de ne smoothness one has to put a manifold structure on
do here.
TM,
which I won't
146
13. MEROMORPHIC 1-FORMS ON A RIEMANN SURFACE
C
V
Omitting subscript
indicate
R C ,
's
for the moment, and writing a subscript
to
one has
*
+
+
,
,
TC,p M = C
=C
x p y p
z p z p
D
E
D
E
TC,p
M C = C dx|p , dy|p
z |p ,
= C dz|p , d
where
,
, dz := dx +
:=
1
:=
+
1
z
2 x
y
z
2 x
y
1dy , d
z := dx 1dy . (This makes dz( z ) = 1, dz( z ) = 0,
) = 0, d
z ( z ) = 1 so that the bases are dual.) A smooth section
d
z ( z
of the complexi ed cotangent bundle TC M thus looks locally like
*
F (x, y)dz + G(x, y)d
z
= (F + G)dx + 1(F G)dy,
for
and
G smooth (in nitely dierentiable) complex-valued functions.
The 1-forms we are after are substantially more restricted:
Definition 13.1.1. A holomorphic [resp.
1 (M )
[resp.
= f (z )dz ,
K1 (M )]2
with
meromorphic]
1-form
is a collection of expressions
f : V C holomorphic
[resp. meromorphic],
satisfying
(13.1.3)
2recall
the notation
|V = |V
, .
K(M )
for meromorphic functions; this is short for
one can think of such functions as meromorphic 0-forms.
K0 (M ),
as
13.1. DIFFERENTIAL 1-FORMS
147
Explicitly, (13.1.3) says that
f (z )dz = f ( (z ))d( (z ))
= f ( (z ))0 (z )dz ,
and is thus equivalent to
f (z ) = f ( (z ))0 (z ).
(13.1.4)
1 , 2 K1 (M ), we can consider their quotient as a mero
morphic function 1 K(M ). This is because in local coordinates, one
2
f (z )dz
(z )
can cancel the dz 's viz.,
= fg (z
and the compatibility
g (z )dz
)
Given
condition (13.1.4) implies that such quotients do patch together (the
0 (z )
factors cancel). Conversely, a meromorphic function times a
meromorphic
1-form
gives a new meromorphic
1-form.
M = P1 , let 1 = be arbitrary and 2 = dz .
Z
1
Here z = 1 on P as usual, and dz looks as if it should be not just
Z0
Z
meromorphic but holomorphic. But in the coordinate at w = 0 ,
Z1
. So dz in fact has a pole of order 2 at [0 : 1].
dz becomes d( w1 ) = dw
w2
Now consider F (z) := 1 =
K(P1 )(
= C(z) by Thm. 3.1.5(a));
2
dz
we have then = F (z)dz . Therefore
P (z)
1
1
K (P ) =
dz P, Q C[z] of the same degree .
Q(z)
Example 13.1.2. On
M = C/ a complex 1-torus, write u for the
coordinate on C. Since each transition function sends u 7 u +
0
(for some ), their derivatives are all identically 1. Hence, du
gives a well-de ned global holomorphic 1-form on M (i.e. belongs to
1 (C/)).
So take 1 = arbitrary, 2 = du. The same argument as above,
Example 13.1.3. For
using Thm. 3.1.5(b), gives
K1 (C/)
= {f (u)du | f = -periodic
Example 13.1.4. Let
can represent
df :=
f K(M )
meromorphic function on
C} .
be a meromorphic function. We
as a collection of maps
f : V P1 .
The
1-forms
df
dz are then compatible (via pullback) with the transition
dz
148
13. MEROMORPHIC 1-FORMS ON A RIEMANN SURFACE
functions, as in (13.1.3); hence, they patch together to give a global
meromorphic 1-form
df K1 (M ).
We will refer to this as the dieren-
f.
tial of
K1 (M ) be given by a collection {f (z )dz }; we would like
de ne its order p () at a point p U M . We simply set
Let
to
p () := z (p) (f );
if this is negative
p U
has a pole
at
p.
As a well-de nedness check, suppose
also. Then (using (13.1.4))
p (f ) = p f 0 (z ) = p (f )
must have nonvanishing derivative
point. If has a pole at p U , then its residue is
1
Resp () := Resz (p) (f ) =
f (z )dz
2 1 C (p)
since, as a biholomorphism,
every
where
C (p)
is a small circle (in
V )
about
z (p).
at
The well-de nedness
check boils down to change of variable in the integral.
Let
= {f (z )dz } K1 (M )
be a form, and
= M
be a
3
smooth real closed curve. Then we de ne
:=
f (z )dz ,
where we observe that 1-forms have been set up so that the right-hand
side is independent of choices of local coordinates and the partition of
into local pieces. The following can be viewed as a version of either
Stokes's theorem or Cauchy's theorem.
Proposition 13.1.5. Let
smooth boundary
3A real curve
be a closed region with piecewise
= .
means something 1-dimensional over
of a closed path on the Riemann surface; and
R (not C), so you should think
U are the segments from which
the path is pieced together.
4the
technical term here is
2-chain,
though we won't get into this here
13.1. DIFFERENTIAL 1-FORMS
149
M
Assume that the meromorphic form
containing
is holomorphic on some open set
Then
= 0.
= , but assume that is only
containing (so that may contain poles
Proposition 13.1.6. Again let
holomorphic on an open set
of
).
p()<0
M
(a) Then we have the residue formula
2 1
Resp ().
p
p () < 0
(b) In general for
K1 (M ),
pM
Resp () = 0.
0 be a sum of circular
paths about those p where has poles. Let 0 be the complement
in M of the union of disks containing these {p}, with 0 = 0 . Apply
Prop. 13.1.5 to the pair 0 , 0 .
Applying the residue formula to the case = M , = gives
(b).
Proof. For the residue formula (a), take
Corollary 13.1.7. Consider a nonconstant meromorphic function
f K(M ).
Then
150
13. MEROMORPHIC 1-FORMS ON A RIEMANN SURFACE
(a)
pM
p (f ) = 0,
i.e. the number of zeroes (counted with mult-
plicity) equals to number of poles (counted with multplicity); and
(b)
#{f 1 ()}
(counted with multplicity) is independent of
Proof. (a) is Prop. 13.1.6(b) applied to
f ,
P1 .
df
. Replacing
f
by
and noting that the number of poles doesn't change, by (a) the
number of zeroes can't change either, giving (b).
f , deg(f ), is de ned to be the
Thinking of f as a covering map from
Definition 13.1.8. The degree of
number in Cor.
M P
13.1.7(b).
deg(f )
can be visualized as the number of branches (or
5
sheets).
Remark 13.1.9. We have said nothing about
when
is not a
boundary:
M
Indeed, there is nothing we can say yet this is the study of periods,
which depend on the complex analytic structure of
M.
We will be able
to compute some periods of holomorphic forms on algebraic curves later
in the course.
13.2. Poincar-Hopf theorem
The usual statement of this theorem is that the sum of indices of
any
6 vector eld
~v
on a compact oriented smooth manifold
M ; we'll only worry
M is 2. In that case, the
is equal
to the Euler characteristic
about the case where
the real dimension of
index Indp (~
v ) of
~v
at
p M is the number of counterclockwise rotations done by (the head
of ) ~
v as one goes once counterclockwise on a small circle about p. It
can only be nonzero if ~
v (p) = 0.
5you may wish to refer back to Remark 3.1.7
6technical point: ~v should have only nitely many
zeroes
13.2. POINCAR-HOPF THEOREM
151
I'll give a heuristic proof of the italicized statement, which is probably more illuminating than a formal one. Subdivide a given compact
smooth oriented real 2-manifold
into triangles:
etc.
Then put one marked point on each edge, vertex, and face of the triangulation:
Next draw the following vector eld on each triangle:
These match up to give a global vector eld on
of this
M.
Evidently the index
~v is 1 at the marked points on the edges, and +1 at the marked
152
13. MEROMORPHIC 1-FORMS ON A RIEMANN SURFACE
points on the faces and vertices. Hence,
(13.2.1)
Indp (~
v)
= #F #E + #V = M = 2 2g
pM
where
on
is the genus of
M.
That (13.2.1) holds for any vector eld
is the version of the theorem proved by Poincar. It still holds
if we allow
~v
to have singularities at a nite set of points
(i.e. it is just a section over
indices of
~v
at the
pi
M \{p1 , . . . , pn }),
{p1 , . . . , pn }
provided one adds the
to the sum.
In fact, (13.2.1) even holds if
A1R (M \{p1 , . . . , pn }).
~v
~v
is replaced by a smooth
1-form
M , i.e. a
nonvanishing section of Sym (T M ), to smoothly identify T M with
T M . The corresponding notion of index, if (in local coordinates at p)
takes the form F dx + Gdy , is
1
G
(13.2.2)
Indp :=
d arctan
,
2
F
The idea is to use a metric on
and once again the sum in (13.2.1) must be over all zeroes of
and the
{pi }.
1-manifold, and write K1 (M )
be locally in the form [Link] + [Link] where f, g are complex-valued. To
get in the above setting, we may of course view M as a smooth real
2-manifold, and take the real part of :
Now let
be a compact complex
loc
:= <() = <(f )dx + <(g)dy.
= p (). Of course, in a local
7
holomorphic coordinate z about p with z(p) = 0, we have
loc
z dz = r cos() + 1 sin() (dx + 1dy)
= r cos() 1 sin() dx+r sin() + 1 cos() dy.
Let
be a zero or pole of
and put
So locally we have for the real part
cos()dx + sin()dy,
r
and thus by (13.2.2)
1
Indp () =
2
7up
d[] = = p ()
to multiplication by a locally nonvanishing holomorphic function (which will
not aect index)
EXERCISES
153
p () = 2g 2.
p
We have arrived at the following corollary of (13.2.1), which will henceforth be the meaning of Poincar-Hopf for us:
Theorem 13.2.1. Let
1-form
K1 (M )
on a Riemann surface of genus
(#
|
of zeroes
#
{z
be a nonvanishing meromorphic
g.
Then
of poles) of
counted with multiplicity
= 2g 2.
Remark 13.2.2. Just as for meromorphic functions we can consider
the divisor
() :=
p ()[p]
pM
of a meromorphic 1-form. In this context, the Theorem says that
deg(()) = 2g 2.
Exercises
dx
y
1 (E). (We can
E
talk about holomorphic 1-forms on a smooth algebraic curve now,
(1) Let
E = {y 4x 4x = 0}, =
because they are Riemann surfaces by the smooth normalization
Prop. 7.1.) Consider the complex analytic automorphism
(x, y) 7 (x, iy), and apply this to the 1-form: com
pute the pullback A ().
(a) In Example 13.1.2, dz de nes a meromorphic dierential 1-form
1
1
1
on P . Compute its divisor (dz). Explain why (P ) = {0}. (b)
What is the divisor of du on C/, from Example 13.1.3? Explain
why it is the unique holomorphic 1-form on C/ up to scale.
2
2
Practice with pullbacks: for the map : R R that sends
(x, y) 7 (u(x, y), v(x, y)) := (x2 3xy, y 3 +x), compute where
= udv + vdu. Write it in the form f (x, y)dx + g(x, y)dy .
E
(2)
(3)
A: E
sending
(4) Continuing from Exercise 5 of Chapter 3, compute the pullback of
dx
under
y
y(z)
: P1 C .
[Hint: simply plug in your nal
x(z)
and
from that exercise. After simpli cation, your answer should
be very simple indeed.]
CHAPTER 14
The genus formula
We are ready to prove two formulas for the genus of a Riemann
surface (RS) which are especially useful in algebraic geometry.
For
the rst result (the Riemann-Hurwitz formula), the RS will arise as a
nite branched cover of another RS whose genus is known. The proof
makes essential use of Poincar-Hopf and a rami cation divisor which
we introduce below.
For the second result, which is an application
of the rst (and of the intersection theory from Chap.
12), the RS
will arise as the normalization of an irreducible algebraic curve in
P2
with only ordinary double point (ODP) singularities. (In fact, the 4H
people will, in their reading material, learn how to deal with worse
singularities later.)
This is a very concrete payo for the preceding
hard work: now we can compute the genus of [the desingularization of]
a projective algebraic curve!
14.1. Order and multiplicity for maps of Riemann surfaces
Consider a nonconstant morphism
faces with
f (p) = q .
f : M M0
of Riemann sur-
In Exercise 4 of Chapter 3, the following was
established: there exist
f (U ) V , and
local holomorphic coordinates z : U C and w : V C with
z(p) = 0 = w(q),
neighborhoods
w f = z
U 3 p, V 3 q
with
N. More informally, in
these local coordinates f takes the form (w =)f (z) = z . We write
p (f ) := . This is the rami cation index, and f rami es at p precisely
when it exceeds 1.
0
For any q M , consider the sum
X
d(q) :=
p (f ).
such that
for some unique
pM with f (p)=q
155
156
14. THE GENUS FORMULA
p of index d lies over q , then over a nearby point
q0 , p is replaced by d points with rami cation index 1. This is by virtue
of the local form w = z , as is the fact that the rami cation points are
isolated hence nite in number (M is compact!). Evidently then, d(q)
is constant in q ; we will call this constant d N the degree deg(f ) of
1
the morphism f .
If a rami cation point
Here is a more gentri ed way to de ne this. We can think of a point
q M0
as a divisor
[q] Div(M ),
and pull it back to a divisor on
2
by the formula
f 1 ([q]) :=
p (f )[p]
Div(M ).
f (p)=q
We then put (for any
q M 0,
it doesn't matter)
X
deg(f ) := deg f 1 ([q]) =
p (f ) .
f (p)=q
f : M M 0 , nally, is the rami cation
X
Rf :=
(p (f ) 1) [p] Div(M ).
Associated to
divisor
pM
By the above remarks, the sum is clearly nite.
14.2. Riemann-Hurwitz formula
Again take
d := deg(f ),
f : M M0
to be a nonconstant morphism, write
and put
r := deg(Rf ).
In the following
resp.
Theorem 14.2.1.
g0
will refer to the genus of
resp.
M 0.
r = 2 {g + d dg 0 1}.
Remark 14.2.2. Some alternative ways to write this result are:
g = (g 0 1)d + 2r + 1
(ii) M = deg(f )M 0 deg(Rf )
(i)
These better represent the way you want to think of it: as a formula
1when f
function and take the degree of
can think of it as a morphism of
0.
P1 , you can think of it as a meromorphic
its divisor, deg((f )), which is always 0. Or, you
Riemann surfaces and take deg(f ), which is never
is a nonconstant map from
to
So that extra parenthesis matters!
2This
extends linearly to de ne
f 1 (D)
for any
D Div(M ).
14.2. RIEMANN-HURWITZ FORMULA
for the genus (or Euler characteristic) of
and data about how
Proof. For
z, w
as in
14.1
sits over
p M
so that
with a
f
a
M,
157
if you know that of
M0
M 0.
= p (f ),
we choose local coordinates
z 7 z (= w).
We shall need to assume the existence of a nonzero meromorphic
1-form K1 (M 0 ).
This is obvious if
tion of an algebraic curve in
M0
arises as the normaliza-
, as you can just pull back any non-
constant meromorphic function (say,
Z1 /Z0 )
and take its dierential.
Every Riemann surface arises in this way, but to see that you need
the Riemann-Roch theorem. We proceed with the proof modulo this
detail.
Locally writing
= g(w)dw,
we have
loc
f = g(z a )d(z a ) = a.g(z a )z a1 dz ,
hence
p (f ) = a.0 (g) + (a 1) = p (f ).f (p) () + (p (f ) 1).
In Div(M ) we have therefore
(f ) :=
p (f )[p] =
p (f ).f (p) ()[p] +
pM
|
=
q ()
qM 0
(p (f ) 1) [p]
{z
Rf
p (f )[p] + Rf
f (p)=q
q ()f 1 ([q]) + Rf
qM 0
!
= f 1
q ()[q]
+ Rf
=f
() Div(M 0 ).
1
Now f K (M ),
(()) + Rf ,
where
so Poincar-Hopf on
tells us that
2g 2 = deg ((f ))
which by the computation just done
= deg f 1 (()) + deg Rf
158
14. THE GENUS FORMULA
q ()
p (f ) + r
f (p)=q
{z
deg(f )
X
= deg(f )
q () + r.
| {z }
deg(())
Applying Poincar-Hopf once more (but on
M 0 ),
we get that this
= d(2g 0 2) + r.
So we have shown
2 2g = d(2 2g 0 ) r,
which is the version of R-H
stated in Remark 14.2.2(ii).
We turn to some examples.
C = {y 2 =
Q2m
2
i=1 (xi )} C , and let M be
P2 . The original curve
the normalization of its projective closure C
Example 14.2.3. Let
had a projection map to the
x-axis ((x, y) 7 x),
and this extends to
f : M P1 =: M 0 ,
as depicted below:
ramification points
Clearly
g 0 = 0, d = 2,
and
r=
since
p (f ) 1 = 1
(p (f ) 1) = 2m
at each of the rami cation points. So by Remark
14.2.2(i)
g = (0 1).2 +
2m
+ 1 = m 1.
2
M = M 0 = C/ be a complex 1-torus; as
usual = {m1 1 + m2 2 | m1 , m2 Z}, where 1 , 2 C are inde
pendent over R. Now assume for some C . Then we have
Example 14.2.4. Let
a complex multiplication map
M M 0
14.3. THE GENUS OF A PROJECTIVE ALGEBRAIC CURVE
159
z 7 z
which has
Rf = 0.
You will treat this setting in an exercise below.
14.3. The genus of a projective algebraic curve
C = {F (Z, X, Y ) = 0} P2 be an irreducible algebraic curve
of degree d with S = sing(C) its set of singular points. We assume that
these are all ordinary double points, and that there are exactly |S| =
of these; write S = {p1 , . . . , p }. Of course, = 0 S =
C is smooth.
C its normalization, we shall deduce from
Denoting by : C
Let
Theorem 14.2.1 the formula:
Theorem 14.3.1.
has genus
g=
(d 1)(d 2)
.
2
To get a feel for this before launching into the proof, for
smooth
we have
d = 1 = g = 0,
d = 2 = g = 0,
d = 3 = g = 1,
d = 4 = g = 3,
and so on. For degree 3 with one ODP, we get
g=
(3 1)(3 2)
1 = 0,
2
as we found using stereographic projection. Indeed, we know how to
parametrize all three genus
cases (smooth
d = 1, 2;
singular
d = 3)
by a Riemann sphere.
The rest of this section is devoted to the proof. Begin by choosing
coordinates on
P2
so that
L C consists of d distinct points,
none of the tangents to C at its ODP's are vertical (i.e.
form X = aZ ), and
C does not contain [0 : 0 : 1].
The latter requirement allows us to project from
map
C P1 =: M 0
[0 : 0 : 1]:
of the
that is, the
160
14. THE GENUS FORMULA
given by
[Z : X : Y ] 7 [Z : X],
roughly speaking the projection of
Writing
M := C ,
to the
x-axis,
is well-de ned.
the main idea of the proof of to apply Riemann-
Hurwitz to the composition
f = x : M M 0.
In a picture, where
VT refers to a point with vertical tangent:
C
x
VT
VT
ODP
1
P
Now for
M 0 = P1 , g 0 = 0
so that Thm. 14.2.1 gives
rf = 2(genus(M ) + deg(x) 1) = 2(g + d 1).
(14.3.1)
In particular, the degree of the map
is
because the projection is
done along vertical lines, all but nitely many such lines meet
points by Bezout, and
in
is 1-to-1 o nitely many such points. So we
see that if we can compute the degree of the rami cation divisor
Rf
then we are done.
To do this, let
E := {FY = 0}
where
FY
is the partial derivative. Obviously
by Bzout,
(14.3.2)
(E C) = (d 1)d.
deg(E) = d 1,
and so
14.3. THE GENUS OF A PROJECTIVE ALGEBRAIC CURVE
P0
Denoting by
p the sum over points where
P
and by
j=1 the sum over ODP's, we have
X0
(E C) =
(E C)p +
161
has a vertical tangent,
(E C)pj .
j=1
We will show
Rf =
(14.3.3)
X0
(E C)p [
p]
p
where
p = 1 (p) C .
(Recall that by our choice of coordinates, a
point with vertical tangent cannot be a singular point, and so has a
unique preimage point under normalization.) Taking degrees of both
sides of (14.3.3) gives
rf =
(14.3.4)
X0
(E C)p = (E C)
(E C)pj .
j=1
Further, we will deduce that
(E C)pj = 2 (j);
(14.3.5)
together with (14.3.2) and (14.3.3), this yields
rf = d(d 1) 2.
Now put this together with (14.3.1) to get
2g + 2(d 1) = d(d 1) 2,
2g = (d 2)(d 1) 2,
and divide the last line by
to get Theorem 14.3.1. It remains only to
check (14.3.3) and (14.3.5).
p C E.
By assumption, p
FX (p) 6= 0. By the
holomorphic implicit function theorem, we can parametrize C locally
by writing x = X/Z as an implicit function of y = Y /Z , viz.
If
has a VT at
p, then F (p) = FY (p) = 0;
this implies
3
is a smooth point, so that
0 = F (1, x(y), y).
3if F (p) = 0
X
at
then
FZ (p) = 0
too by the Euler formula, contradicting smoothness
162
14. THE GENUS FORMULA
Now, dierentiating gives
0=
d
F (1, x(y), y) = FX (1, x(y), y) x0 (y) + FY (1, x(y), y).
dy
For the two functions on the right-hand side to sum to zero, they must
have the same order to
y(p):
ordy(p) FY
(1, x(y), y) = ordy(p) x0 (y),
in other words
(E C)p = {ordy(p) x(y) 1}
= {p (x) 1}
= {p(f ) 1}.
As the only rami cation points of
Rf :=
are (
(q (f ) 1)[q] =
of ) vertical tangent points,
X0
(E C)p [
p]
qC
as claimed.
Finally, to see (14.3.5), assume for simplicity (for some
(0, 0).
j ) pj =
The local a ne equation about an ODP is of the form
F (1, x, y) = ax2 + 2bxy + cy 2 + {higher-order
terms}.
To nd the tangent lines, recall that one solves
0 = ax2 + 2bxy + cy 2 =
x y
!
a b
c d
| {z }
x
y
in
P1
(for their slopes). That the solution
consists of two distinct
pj is an ODP) = Q is smooth = det B 6= 0 =
ac b2 6= 0. That there is no vertical tangent = [x : y] = [0 : 1] is
not a solution = c 6= 0. Consider the partial
points (as
FY (1, x, y) = 2bx + 2cy + {higher-order
whose vanishing de nes
about
pj
E;
evidently
terms}
can be locally parametrized
by
b
y = y(x) = x + {higher-order
c
terms}.
14.4. BEYOND STEREOGRAPHIC PROJECTION
To compute its intersection number against
equation of
C,
163
we pull the de ning
back along this parametrization and take the order at
0:
(E C)(0,0) = ord0 (F (1, x, y(x)))
= ord0 ax2 + 2bx y(x) + c(y(x))2 + {higher-order terms}
ac b2 2
= ord0
x + {higher-order terms}
c
= 2,
Q.E.D.
14.4. Beyond stereographic projection
The genus formula is very nice, but needs to pass a smell test: if it
C P2 has genus zero normalization, then we should
parametrize C by the unique genus zero Riemann surface
says that a curve
be able to
P1 .
We know that this can be done for a smooth conic and a nodal
cubic (i.e. a cubic with one ODP); the rst new case predicted by the
formula is that of an irreducible
4 quartic curve (d
= 4)
with 3 ODP's
( = 3):
g=
Let's give this a try.
another curve
2.
If
(4 1)(4 2)
3 = 0.
2
Write {pi }i=0,1,2 for the
D passes through one of these:
then by 12.2.4,
is a line, then it cannot pass through all 3
have
ODP's, and suppose
pi ,
(C D)pi
as then we would
2
X
4 = deg C deg L = (C L)
(C L)pi 6,
i=0
a contradiction. So the ODP's are not collinear, and by a similar argu-
C , then no three of p0 , p1 , p2 , p3
are collinear. We may therefore move C (and the pi ) by a projectivity
2
of P , to have p0 = [1 : 0 : 0], p1 = [0 : 1 : 0], p2 = [0 : 0 : 1],
p3 = [1 : 1 : 1]. (We'll do so for this abstract analysis but not for the
ment if
p3
is any xed smooth point of
concrete example that follows.)
The general conic in
P2
is of the form
aXY + bY Z + cXZ + dX 2 + eY 2 + f Z 2 = 0.
4We
have to say
is irreducible explicitly, because the union of a smooth cubic
and a general line is a quartic with
5the 6
gets replaced by a
ODP's.
in the inequality above
164
14. THE GENUS FORMULA
By substitution, we nd that the general conic through the above four
points is of the form
Q[a:b] = {aXY + bY Z (a + b)XZ = 0}.
[a : b] P1 .
The zero-cycle (cf. Remark 12.2.2(a)) Q[a:b] C has degree 8 by
Bzout, and is of the form 2[p0 ] + 2[p1 ] + 2[p2 ] + [p3 ] + more. This
more can only be one more point q[a:b] with multiplicity one, since
what is already written has degree 7 (and by construction, one doesn't
have negative intersection numbers). Naturally, q could be one of the
pi : if it is p3 , then this would say that Q is tangent to C there. De ne
This is a 1-parameter family parametrized by
a map
: P1 C
by
[a : b] 7 q[a:b] (:= Q[a:b] C {2[p0 ] + 2[p1 ] + 2[p2 ] + [p3 ]}).
P1 P2 (I
that is onto
In fact, this is a morphism of complex manifolds from
won't prove this carefully). Also, since
is irreducible,
essentially follows from the open mapping theorem and compactness of
P1 .
is 1-to-1 o the singular points of C . Take q C
distinct from the pi ; since no three of the pi are collinear, no four of
q, p0 , p1 , p2 , p3 are collinear, so there exists a unique conic Q through
all ve. (The uniqueness when q = p3 then essentially follows from
continuity of .)
We claim that
Example 14.4.1. So what does such a normalization look like?
Take the very concrete quartic curve
C = {X 2 Z 2 + Y 2 Z 2 + 2X 2 Y 2 = 0}.
Irreducibility can be checked by putting the polynomial in a ne form
y 2 (1+2x2 )+x2 and showing it doesn't factor into terms of lower degree
in y . I will let you check that the only singularities are p0 = [1 : 0 : 0],
p1 = [0 : 1 : 0], p2 = [0 : 0 : 1]; pick p3 := [i : 1 : 1] (i = 1). The
general conic through these 4 points is
Q[:] := {XZ + Y Z = i( + )XY }.
EXERCISES
Substituting this into
times the equation of
165
gives
0 = (i( + )XY Y Z)2 + 2 Y 2 Z 2 + 22 X 2 Y 2 =
2 + 2 2
2
2
2
= ( + )Y (Z iX) Z i
X ,
2 + 2
X
in which the last factor gives us the x =
-coordinate of the point
Z
q[:] . The y -coordinate is obtained by substituting into the equation
of Q, and we nd ([ : ]) =
i(2 + 2 2 )(2 + 2 2 ) : (2 + 2 )(2 + 2 2 )
: (2 + 2 )(2 + 2 2 ) .
Or, in a ne coordinates (t
(t) = i
in particular),
1 + t2
1 + t2
,
i
t2 + 2t 1 t2 2t 1
.
Exercises
C , C 0 compact RS with g =
genus(C), g 0 = genus(C 0 ), f : C C 0 nonconstant holomorphic
0
map of degree d. Show that for any d 1, g g . (The covering
(1) Recall the setup of Riemann-Hurwitz:
surface has at least as many handles.)
(2) Let
z=
Z1
(where
Z0
[Z0 : Z1 ]
are the homogeneous coordinates) be
P1 . If a holomorphic map f : P1 P1
n
n1
takes the form f (z) = z + a1 z
+ . . . + an , then
(a) What is deg(f )?
(b) What can you say about the rami cation divisor Rf ? (at least,
the canonical coordinate on
what is its degree?)
(c) Use Riemann-Hurwitz to check your answers.
= {m1 + n2 | m, n Z} is a lattice in
C. (In particular, 1 and 2 are independent over R.) Suppose
that C satis es . [Remark: if
/ Z this places a
strong condition on ; we say , or C , has complex multiplication
(or CM ). ] The multiplication by induces a holomorphic map
: C C , i.e. an automorphism of the RS C .
(a) Show that the rami cation divisor R Div(C) for this map is
(3) Let
C = C/
where
zero.
(b) Prove that the degree of
image lattice
equals the index
[ : ]
of the
166
14. THE GENUS FORMULA
e of the irreducible curve C
C
x2 + x2 y 2 + y 2 = 0 in P2 . (First
(4) Find the genus of the normalization
given by taking the closure of
convert to homogeneous coordinates and check for singularities.
Then apply the genus formula. This is very similar to something
above...)
(5) This problem complements (3) above, but you won't use anything
f : C C of
f : C C (i.e.
from this chapter in doing it. A (holomorphic) map
Riemann surfaces is nothing but an analytic map
an entire function) such that for all
()
i.e.
z1 z2
de nedness
, z C,
f(z + ) f(z) ,
= f(z1 ) f(z2 ) mod (this is just the wellcondition for f ). Show that such a map is necessarily
mod
a ne, i.e. of the form
f(z) = z + .
[If
other than
works, then we are of course in the CM
case described above. So a non-CM complex 1-torus, which is the
z 7 nz + , n Z,
().
generic case, has endomorphisms of the form
and that's all.] Hint: consider
f0 (z),
and use
CHAPTER 15
Some applications of Bzout
We have already put Bzout's theorem to use in proving the genus
formula (and in several interesting exercises at the end of Chapter
12). Now we shall use it to prove a general result on curves through
con gurations of points, which in particular will yield a short (and
rigorous) proof of Pascal's theorem from Chapter 1.
We shall also
deduce some results on cubics will will come in handy in studying the
group law on elliptic curves.
Throughout this Chapter we shall use the following dictionary:
P2
algebraic curve
de ning equation
degree
C
D
E
(homogeneous polynomial)
F S3d
G S3d
H S3e
d
d
e
Recall the theorem we are wanting to apply:
Bzout.
C E
is
0-dimensional (consists of points) = (C E) =
de.
Part of the content of the (equivalent) contrapositive statement is:
tuozB. The number of points
|C E|
exceeds
de = E
have a common component.
From Chapter 9, we have:
Study.
irreducible and
Putting tuozB and
BS.
E C = H
divides
Study together gives:
irreducible and
|C E| > de = H | F .
We'll make use of this statement below.
167
F.
and
168
15. SOME APPLICATIONS OF BZOUT
15.1. Cayley-Bacharach theorem
Let
p1 , . . . , p n P2
be distinct points, and de ne
S d (p1 , . . . , pn ) :=
in
[Z : X : Y ]
vanishing at
E is irreducible
pa+1 , . . . , pn
/ E . Then
Lemma 15.1.1. Suppose
some
a > ed,
while
d)
p1 , . . . , p n .
homogeneous polynomials (of degree
and
p1 , . . . , p a E
for
S d (p1 , . . . , pn ) = H S de (pa+1 , . . . , pn ).
Proof. The inclusion of the RHS into the LHS is easy, since it is
just saying that the product of a polynomial vanishing at the last
points by a polynomial vanishing at the rst
na
points, vanishes at all
of them. So we turn to the reverse inclusion.
S d (p1 , . . . , pn ) is nonzero, take a nonzero element F ; this
de nes a degree d curve C containing p1 , . . . , pn . Clearly p1 , . . . , pa
C E , so |C E| > ed, and by BS, H | F . We can therefore write
F = F0 H with F0 S de . Since F = 0 but H 6= 0 at pa+1 , . . . , pn , F0
de
must vanish at these points. It follows that F0 S
(pa+1 , . . . , pn ) as
desired.
Assuming
Theorem 15.1.2. Let
and assume exactly
d(d e)
d e.
remaining
of degree
ed
|C D| = d2 with d > e,
C D lie on E . Then the
be irreducible,
of the points of
points lie on a (not necessarily irreducible!) curve
[A : B : C] E\{(C D) E}, and set =
F (A, B, C), = G(A, B, C). De ne P := G + F S d ; this vanishes on C D and at [A : B : C]. Label (C D) E =: {p1 , . . . , ped },
[A : B : C] =: ped+1 , and (C D)\{(C D) E} = {ped+2 , . . . , pd2 +1 };
2
set a := ed + 1 and n = d + 1.
d
Since a > ed, Lemma 15.1.1 tells us that S (p1 , . . . , pd2 +1 ) = H
S de (ped+2 , . . . , pd2 +1 ). But then, since P S d (p1 , . . . , pd2 +1 ), we have
P = HP0 for some P0 S de (ped+2 , . . . , pd2 +1 ). This P0 de nes the
required curve.
Proof. Let
Here is the nice application to Pascal:
1it
is enough to check, in applying this, that
points) lie on
can't lie on
E.
E.
ed of the points (not exactly ed of the
ed of these points simply
This is because by Bzout, more than
15.2. INTERSECTIONS OF CUBICS
169
Corollary 15.1.3. The (three) intercepts of opposite sides of a
hexagon inscribed in a conic are collinear.
Proof. Referring to the picture
p5
L2
p6
L1
L6
L3
L4 p 2 L5
p1
p4
p3
q3
q2
C := L1 L3 L5 , D = L2 L4 L6 , and E = Q. Clearly this
means d = 3 and e = 2, and we do indeed see that de = 6 points of
C D = {p1 , . . . , p6 } {q1 , q2 , q3 } lie on E . So the last three points of
C D, which are the intercepts, lie on a curve of degree d e = 1 by
the Theorem.
we put
Remark 15.1.4. If one wanted instead to plug the technical gap in
the proof of Pascal suggested in Chapter 1, part of what one needs is the
statement: if
p1 , . . . , p8 P2 are distinct and in general position
in the
sense that no 4 are collinear and no 7 conconic (lying on an irreducible
conic), then
dim S 3 (p1 , . . . , p8 ) = 2.
This is proved in Reid's book.
15.2. Intersections of cubics
The results of
15.1
dealt with the case where all intersections of
curves have multplicity one (the transversal case), since we required
|C D| = d2 = (C E).
assuming E is smooth and
To deal with the general case, at least
irreducible (so that we may view it as a
Riemann surface), write
C E :=
(E C)p [p]
Div(E).
pEC
If
is irreducible but singular, with an unique ODP
nition gives a divisor
p
/ E C.
C E Div(E)
p,
the same de -
(on the normalization) provided
170
15. SOME APPLICATIONS OF BZOUT
Theorem 15.2.1. Let
(If
C, D, E
E irreducible.
p
/ E C, E D.) Writing
be distinct cubics, with
is singular, assume moreover that
by Bzout
9
X
DE =
[qi ]
Div(E)
i=1
where the
qi
need not be distinct, and assuming
8
X
C E =
[qi ] + [q] Div(E),
i=1
we have
q = q9 .
In the intersection multiplicity one case, the Theorem gives immediately:
Corollary 15.2.2. Let
D E = {q1 , . . . , q9 }
C, D, E
be distinct cubics,
(distinct points) and
then it passes through
irreducible. If
passes through
q1 , . . . , q 8 ,
q9 .
E
Actually this is true without assuming
doesn't share any components with
or
Proof. (of Theorem) First assume
C ),
irreducible (provided
meromorphic function on
P2 .
but we won't prove that.
is smooth. Recall that the
quotient of two homogeneous polynomials say,
D = {G = 0}
F/G
By Example 7.3.6, since
in points, we may pull this back to
yields a
intersects
E:
F
f := K(E) .
G E
Suppose (for a contradiction) that
D = {G = 0},
the divisor of
q 6= q9 .
Since
C = {F = 0}
and
is evidently
(f ) = C E D E = [q] [q9 ] Div(E).
This says that
has one zero (at
q)
and one pole (at
holomorphic map of Riemann surfaces
1.
That is,
EP ,f
q9 );
hence, as a
has mapping degree
is 1-to-1; and since (using the open mapping theorem) its
image must be open and closed (and
P1 is connected), f
is surjective. So
f gives an isomorphism E
= P . Trouble is, this is total rubbish. Since
E is a smooth cubic, its genus is 1 by the genus formula, whereas the
1
genus of P is zero. So they can't be isomorphic for purely topological
1
EXERCISES
171
reasons. This contradication tells us that, indeed, our assumption
q9
q 6=
was wrong, and so they are equal.
E is singular with ODP
: E P2 (of E ) to
P1 . As before,
from E
To extend this argument to the case where
F
p, rst pull back G
along the normalization
. We regard f as a map
f K(E)
assuming q 6= q9 leads to deg(f ) = 1. However, a dierent objection
to deg(f ) = 1 will be required as there is no topological obstruction:
indeed, E
= P1 by the genus formula (a nodal cubic has genus zero
normalization). So argue as follows: since p
/ C, D, we nd that
F
2
K(P ) is well-de ned at p, so its pullback via cannot separate
G
mapping
the two branches of E there. That is, at the two points of E
to p
(under ), f will take the same value. But then, the mapping
degree of f cannot be 1.
The other possibility is that p
is a cusp.2 We may assume p = [1 :
0 : 0] and the equation is of the form x3 = y 2 . Again we need to show
F
, if nonconstant, cannot have mapping degree 1. Let
that f =
G
F (1,x,y)
F (1,0,0)
1
p = (
p), and write R(x, y) := G(1,x,y)
G(1,0,0)
. Then f (t) f (0) =
( R)(t) = R(t2 , t3 ), and
obtain
deg(f ) = deg f 1 ([f (0)])
ord0 (R(t2 , t3 ))
ord(0,0) (R(x, y)) min{ord0 (t2 , t3 )}
{z
}
|
{z
} |
=2
2.
Exercises
F = 0, G = 0, H = 0),
of respective degrees d, d, e with 3 e d. Suppose C and D
2
intersect in d distinct points, and assume that E is smooth (hence
irreducible). Show that if E passes through ed 1 of these, it
passes through ed of them. (Imitate the argument from the proof
(1) Let
C , D,
and
be as above (de ned by
of Theorem 15.2.1.)
2see
the paragraph immediately preceding
16.1
below.
Part 3
Cubic curves
CHAPTER 16
The singular cubic
Recall that a singular cubic curve
D P2
ographic projection through its singular point
is normalized via stere-
p;
that is, we get a nor-
malization morphism
: P1 P2
with image
D.
In particular, all singular cubics have normalization of
genus zero. Moreover, they are all projectively equivalent to one of two
examples.
The nodal cubic. A node is just an ordinary double point. Let
D = {Y 2 Z = X 2 (Z X)};
the a ne equation is
y 2 = x2 (1 x)
and a
schematic picture is
where I have denoted points with real coordinates in blue and points
with only
x-coordinate
real in red. (How these sit inside the full set
of complex points will be pictured below; the dotted stu will connect
up.) By Exercise 5 of Chapter 3, this is parametrized by
: P1 /{0, } D
1D
is for degenerate
175
176
16. THE SINGULAR CUBIC
t 7
The P
/{0, }
4t
t(1 + t)
,
2
(1 t)
(1 t)3
=: (x(t), y(t)).
means the Riemann sphere with the top and bottom
2
points identi ed.
The cuspidal cubic. Take
2
y =x
D = {Y 2 Z = X 3 },
a ne equation
, schematic picture
p
where I have only drawn real points. To do stereographic projection
through the cusp
= x =
(0, 0),
write
y = 1t x
and substitute to get
1 2
x
t2
= x3
1
. Hence we get a normalization
t2
: P1 D
de ned by
1 1
t 7 1 : 2 : 3 .
t t
One of our overarching themes in the next few chapters will be the
study of algebro-geometrically de ned group laws on cubics.
In this
chapter, we focus on the above two singular examples, as the smooth
case is more di cult.
be equivalent (via
For the nodal cubic, the law will turn out to
C = P1 \{0, }; while in
1
addition on C = P \{}. In both
to multiplication on
the cuspidal case, it identi es with
cases, these sets are the preimages under normalization of the smooth
points of
2Note
D,
which is where the group laws will be de ned.
that one can homogenize the formula for
4T1 T0 (T0 T1 ) : T1 T0 (T0 + T1 )],
[0 : 0 : 1].
and then it is
[T0 : T1 ] 7 [(T0 T1 )3 :
clear that (1) = ([1 : 1]) =
by
16. THE SINGULAR CUBIC
177
In the course of studying such laws as well as addition theorems on
these curves, we will pull back rational functions on
P2
(quotients of
homogeneous polynomials of equal degree, or equivalently elements of
C(x, y))
to get meromorphic functions on
P1
(the normalization of our
curve). So in illustrating the simplicity of the group law, hence
Principle 1: Singularities make curves of a given degree more trivial and easier to study,
we will be seeing a concrete example of the following
C P2 an irreducible algebraic
is of the
: C P2 , every f K(C)
Principle 2: Given
normalization
curve with
form
F ,
F C(x, y).
In other words, writing
of
and
gC0 (x, y)
C[C0 ] :=
C0 := C (P2 \{Z = 0})
for the a ne part
for its de ning equation, if we de ne
C[x, y]
,
(gC0 )
C(C) := ( fraction
]
eld of C[C0)
F C(x, y)
F
,
=
F
/ on C
then Principle 2 says that
K(C)
analytic
Since
was projective,
C(C).
algebraic
is compact, and that turns out to be of
fundamental importance: e.g.,
C[C0 ]
is only a subring of, rather than
equal to, the ring of holomorphic functions on (the desingularization
of )
C0 .
Before continuing on, we should address one point:
why should
be an ODP
the only possible singularities of an irreducible cubic
or cusp, and why must it have only one?
First of all, if it had two
L through those two points.
Both intersection multiplicities (of C with L at these two points) would
have to be 2, and so (C L) 4 in violation of Bzout. (See what
a useful theorem this is?) So C can only have one singular point,
and as its equation is of degree 3 that point can only be of order 2
or 3. If it is of order 3, then by the result of Chapter 6 Exercise 5,
singular points, then we could take a line
178
16. THE SINGULAR CUBIC
is a union of
lines, contradicting irreducibility. Finally, the local
equation about a non -ordinary double point of
3. An explicit
local analytic transformation puts this in the form (
x) + (
y )3 = 0. So
2
n
it is a cusp. Alternately, anything which looks like x + y = 0 has
intersection multiplicity n with the line x = 0, again violating Bzout
(in the context of our cubic curve) if n > 3.
form
x + f3 (x, y) = 0,
with
f3
can only be of the
homogeneous of degree
16.1. Warm-up: Functions on a nonsingular conic
Our smooth conic is named
a map
C P
C.
Any
F K(C)
can be viewed as
. Composing this with the stereographically produced
1 =
normalization : P C , yields
=
P1 / C F / < P1 ,
F
that is, a meromorphic function on
t := T1 /T0 ), F
P1 .
Since
K(P1 ) = C(t)
(here
must be of the form
G(T0 , T1 )
g(t)
=
h(t)
H(T0 , T1 )
where
g, h, G, H
same degree.
are polynomials and
are homogeneous of the
By the fundamental theorem of algebra, we can write
this as
for some
G, H
, i , j C.
Q
(T i T0 )mi
Qi 1
,
nj
j (T1 j T0 )
T0N
As
deg G = deg H = N +
mi
nj = 0,
the expression simpli es to
Q
(t i )mi
Q
(= ( F )(t)).
(t j )nj
Note that
(16.1.1)
( F )() 6= 0,
mi =
nj .
16.2. Functions on a nonsingular cubic (nodal case)
Let
F : D P1
be
the restriction to
(16.2.1)
of a rational function on
6 0,
=
p D.
which is well-de ned and
at the singular point
P2
16.2. FUNCTIONS ON A NONSINGULAR CUBIC (NODAL CASE)
P1 D
Since the normalization
0, 7 p
sends
179
but is otherwise
1-to-1, we get
(P1 /{0, })
/
:
P1
F
with
F (0) = F () C
. Henceforth we shall, by abuse of notation,
refer to this composition as
Thinking of
F.
as a meromorphic function on
P1 ,
(16.1.1) applies
and we get
Q
(t i )mi
F (t) = Q
(t j )nj
Furthermore,
with
mi =
nj .
Q mi
= F () = F (0) = Q inj
j
so that
(16.2.2)
relating the
z -coordinates
imi =
j j ,
of the zeroes and poles of
F.
Now introduce the multivalued function
u :=
1
on
dt
= log(t)
t
, which takes well-de ned values in
(16.2.2) in terms of
u:
C/2 1Z.
We can restate
viz.,
p (F ) u(p) 0
mod
2 1Z.
pD
This leads to Abel's theorem for the singular cubic :
P, Z Div(D\
p)
Proposition 16.2.1. Given
eective divisors of
the same degree,
dz
0
z
mod
P = poles
Z = zeroes
3recall
2 1Z
)
of some
as in (16.2.1).
divisors are formal sums of points on a complex manifold with integer coef-
cients. A divisor is eective if none of those coe cients are negative.
180
16. THE SINGULAR CUBIC
P
P
P = nj [j ] and Z = mi [i ] are of the same degree
P
P
P
(d =
nj =
mi ), then we may write Z P = dk=1 ([zk ] [pk ])
Z
P zk
by some choice of paths. Also, in the statement
:=
and
pk
P
Explicitly, if
poles and zeroes are as usual meant with multiplicity.
This is a
rst baby case of a general statement for algebraic curves (Abel's
theorem) connecting integrals of
1-forms
to the question of when a
divisor is the divisor of a meromorphic function.
16.3. Group law on the nodal cubic
4
Fix a normalization
=
: P1 /{0, } D
t 7 (x(t), y(t))
1 7 (1) =: e.
p, q D be arbitrary nonsingular points, and Lpq be the line
through p and q . (If they are the same, then take L to be the tangent line Tp D .) By Bzout, (Lpq D) = 3 and so Lpq meets D in a third
point which we call p q . More precisely, everything is counted with
multiplicity (p q need not be distinct from p or q ) so we really mean
Let
[p q] := Lpq D [p] [q].
Now let
L0
be the line through
pq
and
(or
Te D
if they coincide),
and put
[p + q] := L0 D [p q] [e].
That is,
p+q
is the extra intersection point of this line with
guaranteed by Bzout. Here's a useful picture of the construction:
4when
we need to use homogemeous coordinates,
(t) = [Z(t) : X(t) : Y (t)]
16.4. ADDITION THEOREMS FOR THE NODAL CUBIC
181
p+q
p
e
0,
t(p*q)=1/(t(p)t(q))
t(q)
1=t(e)
t(p)
t(p+q)=t(p)t(q)
0
p*q
Now writing
fL
for the equation of a line
fLpq
F :=
fL0
satis es (16.2.1).
back along
F (t) =
But since
L,
observe that
: D P1
D
In terms of the
t-coordinate
on
P1 ,
i.e.
pulling
we must have:
(t t(p))(t t(q))(t t(p q))
(t t(p))(t t(q))
=
.
(t t(p + q))(t 1)(t t(p q))
(t t(p + q))(t 1)
F (0) = F (), by (16.2.2)
Y
Y
{locations of zeroes} =
{locations
of poles}
= t(p) t(q) = t(p + q) t(e) = t(p + q).
|{z}
1
C = P1 \{0, } with
the one just de ned on D\
p. Alternately, taking log gives
u(p) + u(q) u(p + q) mod 2 1Z,
identifying addition on D\
p with addition in C/2 1Z. This may
This identi es the group law (multiplication) on
be rewritten
(16.3.1)
t(p)
dt
+
t
t(q)
dt
t(p+q)
dt
t
mod
2 1Z.
16.4. Addition theorems for the nodal cubic
Let's unwind the equivalence of group laws in the nodal cubic
example from the beginning of the chapter. Noting that
[0 : 0 : 1],
here is a picture of how the group law works:
e := (1) =
182
16. THE SINGULAR CUBIC
t.
x=cons
p+q
p
p*q
In particular, the
the
y -coordinates
x-coordinates of p + q
are of each other.
and
pq
are the same, while
Just to clarify the topology of the situation, here is what the projection of the normalization of
onto the
x-axis
"schematic" picture
looks like:
topological picture
0
8
xvalues
Px
1
8
1
tvalues
(real axis in bold)
It is a 2-sheeted cover with 2 branch points, with the closed path
indicating the equator (or unit circle
|t| = 1)
on the upper
(i.e.
).
D
Now we get to work. Start by inverting the equivalence
t(p + q):
(t1 ) + (t2 ) = (t1 t2 ).
| {z } | {z } | {z }
p
p+q
t(p)t(q) =
16.4. ADDITION THEOREMS FOR THE NODAL CUBIC
Since
p, q ,
and
pq
183
are collinear by construction,
Z(p) Z(q) Z(p q)
0 = det X(p) X(q) X(p q) .
Y (p) Y (q) Y (p q)
Assuming none of them is
e, Z(p)Z(q)Z(p q)) 6= 0
and we get the
1st addition theorem:
1
1
1
0 = det x(t1 ) x(t2 ) x(t1 t2 ) .
y(t1 ) y(t2 ) y(t1 t2 )
x(t1 t2 ) from x(t1 )
p
y(t) = x(t) 1 x(t).
This allows you to compute
equation of
Next,
dt
, while
t
D
to write
and
x(t2 ),
using the
= d(x(t))
= [use Exercise 4 from Chap.
y(t)
D
dx
= xdx1x ; so (16.3.1) may be expressed
y
dx
y
13]
x(q)
x(p+q)
dx
dx
dx
.
x 1x
x(e)(=) x 1 x
x(e) x 1 x
x(e)
dx
dt
(Note that 2 1 =
=
. Going modulo its integer multi|t|=1 t
y
ples, which is what means here, is necessary not to have the equation's correctness depend upon the choice of paths from to x(p), to
x(q), and to x(p + q).) Solving
1
1
1
det
x(p)
x(q)
x(p + q)
=0
p
p
p
x(p) 1 x(p) x(q) 1 x(q) x(p + q) 1 x(p + q)
for
x(p)
x(p + q)
yields
x(p)x(q)
x(p + q) = p
2 .
p
1 x(p) + 1 x(q)
Forgetting the association with
p, q , p + q D
we get the
2nd addition theorem:
x1
x2
x1x2
dx
dx
dx
( 1x1 + 1x2 )2
x 1x x 1x
x 1x
Note that
dx
x 1x
= log
1x1
1x+1
mod
2 1Z.
by explicit computation of the
184
16. THE SINGULAR CUBIC
integral.
(One way to view this function is
log(t) (= u)
viewed as a
multivalued function of x.) So we have discovered a functional equation
1x1
for log
, which is ugly to check by hand.
1x+1
One aspect of the game we have just played here is:
start with
a natural choice of dierential 1-form on the curve (if possible, one
which is smooth away from any singularities of the curve). In the above,
dx
| . You can think of this as a multivalued 1-form on the
y D
x-axis, and then D is the existence domain of the 1-form over P1x .
this was
Then you integrate this 1-form, which gives a transcendental function
which is multivalued even on
(you have to go to its universal cover
to make it well-de ned), and try to produce a functional equation for
it (as a function of
x).
In the last section we'll summarize this story
for a couple of other curves.
16.5. Other addition theorems (conic, cuspidal cubic)
C = {y 2 + x2 = 1}, parametrized
2t t2 1
t 7 2
,
t + 1 t2 + 1
Consider the example
as in
3.3.
by
P1
via
We compute
2dt
dx
= 2
= 2d(arctan(t)),
y C
t +1
dx
dx
=
= d(arcsin(x)).
y C
1 x2
1
On the universal cover of Px \{1} let = arcsin(x) (starting at x =
0 t = 0 = 0).5 Its role is similar to that of u = log t
above, as the integral of our chosen dierential 1-form on the curve;
takes well-de ned values in C/2Z. Writing x(1 ) =: x1 , x(2 ) =: x2 ,
x(1 + 2 ) =: x12 , the standard trigonometry relations give
q
q
x12 = x1 1 x22 + x2 1 x21 .
5Note: t 7
2t
t2 +1 = x is a degree-2 map (from C to the x-axis) with rami cation
points t = 1 over x = 1. On the complements of these points, we have a 2-to-1
map C C . The universal cover of C is C, and so we have maps C Ct Cx
2 tan 2
sending to t to x. So our setup encodes the relation
= x(t()) = x() =
(tan 2 )2 +1
sin().
EXERCISES
185
The second addition formula for the conic then reads
x1
dx
+
1 x2
x2
dx
1 x2
which is a functional equation for
x12
arcsin.
dx
1 x2
mod
2Z,
More simply put, it is just
the inverse of the trigonometric identity.
Next, look back to the cuspidal example from the beginning of the
chapter. We have
1
t2
1
t3
d
dx
d(x(t))
=
=
y D
y(t)
= 2dt,
while
dx
dx
= 3.
y D x2
dx
(Note that this time, the integral of
| is just 2t
y D
valued on D .) Clearly if t12 = t1 + t2 , then
t1
t2
t12
dt +
dt =
dt
0
1
(=x(t1 ))
t2
1
dx
x
3
2
1
(=x(t2 ))
t2
2
dx
x
3
2
and is not multi-
1
(=x(t1 +t2 ))
t2
12
dx
3
x2
1
So we get a functional equation for , which is unfortunately rather
x
stupid: it says
1
1
21 + 12 =
1
t21
1
t22
1
1
(t1 +t2 )2
12 .
In an exercise below, you will show a less trivial addition theorem for
the cuspidal cubic, to the eect that
P, Q, R (D\
p)
are collinear
t(P ) + t(Q) + t(R) = 0.
Exercises
D = {Y 2 Z = X 3 } P2 and
1
1
1
normalize it as above, with : P D given by t 7 [1 : 2 : 3 ] =
t
t
[Z : X : Y ]. (The singular point is p = [1 : 0 : 0].) Prove directly
1
that the group law given by addition on (P \{})
= C (namely,
t1 , t2 7 t1 + t2 ) corresponds to the following process on (D\{
p}):
0
take the line L through (t1 ) and (t2 ), then a line L through
the third intersection point (t1 ) (t2 ) (of L with D ) and the
neutral point [0 : 0 : 1], and nally locate the third intersection
(1) Consider the cuspidal cubic curve
186
16. THE SINGULAR CUBIC
point of this
L0
with
to get (t1 ) + (t2 ) (also as above, for the
P, Q, R (D\{
p})
t(P ) + t(Q) + t(R) = 0. (Here P, Q, R
nodal cubic). Do this simply by showing that
are collinear if and only if
are distinct.) Hint: use the determinant of the matrix
a3 a 1
3
b b 1 ,
c3 c 1
and rewrite
[1 :
1
t2
1
]
t3
= [t3 : t : 1].
CHAPTER 17
Putting a nonsingular cubic in standard form
An irreducible algebraic curve
genus of its normalization
is
E P2
is an elliptic curve if the
(topologically it looks like a donut).
By the genus formula, all smooth cubic curves are elliptic. In the next
two chapters we will show not only that such a curve is isomorphic to
C/
for some lattice
but will get a description of
which shows
E . This is important, since for two dierent lattices
= Z h, i and 0 = Z h0 , 0 i, the complex 1-tori C/ and C/0
its dependence on
need not be isomorphic as Riemann surfaces. (More precisely, they are
[0 : 0 ] by an integral
1
projectivity, i.e. a transformation of P induced by A P SL2 (Z).)
Even more signi cant is how we do this: by putting E in Weierstrass
isomorphic if and only if
[ : ]
is carried to
form, integrating a holomorphic form on it to get a map to a complex
torus, and showing that the Weierstrass
invert this map.
To put
-function
and its derivative
in this form, a choice of ex is required.
What is that?
17.1. Flexes
C = {F (Z, X, Y ) = 0} P2 be an irreducible algebraic curve of
degree d 3. One way of thinking of the tangent line at a nonsingular
point p C is as the unique line satisfying (C Tp C) 2.
Let
Definition 17.1.1. A smooth point
pC
is called a ex if the
intersection multiplicity
(C Tp C) 3.
TpC
C
C , and can
(see 4.4). Since C
Intuitively these are the in ection points of
be seen to
correspond to cusps of the dual curve C
has nitely
187
188
17. PUTTING A NONSINGULAR CUBIC IN STANDARD FORM
many singularities, this gives one proof that there are nitely many
exes; we will however take a dierent approach.
Denoting partial derivatives by subscript, e.g.
Hessian of
FZX :=
2F
, the
ZX
is the polynomial matrix
FZZ FZX FZY
HessF = FXZ FXX FXY .
FY Z FY X FY Y
Its determinant
H := det(HessF )
3(d 2). Call HC :=
associated to C .
is clearly a homogeneous polynomial of degree
{H(Z, X, Y ) = 0} P
Lemma 17.1.2. Let
the Hessian curve
pC
be a smooth point. Then
p is a ex
p HC .
Proof. Since intersection numbers are invariant under projectiv-
p = [1 : 0 : 0], Tp C = {Y = 0}. In a ne
coordinates, writing f (x, y) := F (1, x, y), this means that the curve
{f (x, y) = 0} C2 contains (0, 0) and is tangent to {y = 0}. So
f (0, 0) = 0 and (fx (0, 0), fy (0, 0)) = (, 0) where 6= 0, so that
ities, we may assume
f (x, y) = y + (ax2 + 2bxy + cy 2 ) + higher-order
Parametrizing
Tp C
by
t 7 (t, 0),
terms.
we have
(C Tp C)p = ord0 (f (t, 0)) = ord0 (at2 + h.o.t.),
which is
(yielding a ex) if and only if
Now the above form of
a = 0.
implies
F (Z, X, Y ) = Y Z d1 + (aX 2 + 2bXY + cY 2 )Z d2 +
so that
0
0 (d 1)
HessF (1, 0, 0) =
0
2a
2b
.
(d 1) 2b
2c
Taking the determinant,
H(p) = det(HessF (p)) = 2(d 1)2 2 a.
This is clearly zero (i.e.
p HC )
if and only if
a = 0.
17.1. FLEXES
Now Bezout guarantees intersections of
189
HC .
and
If
is singular
then these might all be at singular points, so that there might be no
exes (though this isn't typical: see the exercises). On the other hand,
if
is smooth then by Lemma 17.1.2 we do have exes. Re ning this
observation:
Proposition 17.1.3. On a nonsingular curve
there exists at least one and at most
3d(d 2)
of degree
d 3,
exes.
Proof. By Bezout,
(C HC )p = (C HC ) = deg(C) deg(HC ) = d 3(d 2).
pCHC
So the number of points in
C HC
is between
and
3d(d 2),
all
points are smooth points, and we apply Lemma 17.1.2.
HessF is just the multivariable derivative
2
2
(Jacobian matrix) of DC : P P (4.4), the intersections of C and
HC may be viewed as degeneracies of the map DC |C : C C . This is
referred to above.
what gives rise to the cusps in C
Remark 17.1.4. Since
Definition 17.1.5. The multiplicity of a ex
be
pC
is de ned to
(C HC )p .
Now take
C = E
to be a smooth elliptic curve (d
= 3).
Then
in the proof of Lemma 17.1.2, the precise form of the homogeneous
polynomial is
(17.1.1)
F (Z, X, Y ) = Y Z 2 +(aX 2 +2bXY +cY 2 )Z+X 3 +X 2 Y +XY 2 +Y 3 .
Assume
a=0
[1 : 0 : 0]. (Note that must
in order that Y not divide F which would make E
so that we have a ex at
then be nonzero,
reducible hence singular.) Then a short computation gives
2y
2by
2 + 2bx
HessF (1, x, y) =
2by
6x + 2y
2x + 2y + 2b .
2 + 2bx 2x + 2y + 2b 6y + 2x + 2c
Pull this back to
Tp E = {y = 0}
by making the substitution
0
0 2 + 2bt
HessF (1, t, 0) =
0
6t
0
;
2 + 2bt 0 2t + 2c
190
17. PUTTING A NONSINGULAR CUBIC IN STANDARD FORM
this has determinant
H(1, t, 0) = (2 + 2bt)2 6t,
and since
, 6= 0
(Tp E HE )p = ord0 (H(1, t, 0)) = 1.
So
HE
is smooth at
and
Tp E
is not its tangent line.
But then it
E transversely (since they have distinct tangent lines), so
(E HE )p = 1. This computation is valid at any ex of E (after a
intersects
that
projective change of coordinates, of course), and so proves:
Proposition 17.1.6. Any smooth cubic has
exes, each of mul-
tiplicity one.
deg(HE ) = 3(d 2) = 3, Bezout gives us 9 intersection points of HE and E , counted with multiplicity; and we have
demonstrated that the multiplicities are all 1.
Proof. Since
17.2. Weierstrass form
Consider an arbitrary smooth cubic curve
E = {F (Z, X, Y ) = 0} P2 .
In this section we will show that there exists a choice of projective
coordinates putting
uniquely into a convenient form. (Alternately,
you can view this as the existence of a projectivity putting
into this
form, in the same coordinates.)
E has a ex, and rst of all we can choose coordinates so
that this is at [0 : 0 : 1] =: O with TO E = {Z = 0}. To get the general
equation of such a cuve: take (17.1.1), set a = 0 (for a ex), swap Z
and Y , and (without loss of generality since 6= 0) normalize to 1;
We know
this gives
F (Z, X, Y ) = ZY 2 + (2bXZ + cZ 2 )Y + X 3 + X 2 Z + XZ 2 + Z 3 ,
with a ne form
f (x, y) := F (1, x, y) = y 2 + yf2 (x) + f3 (x).
Now the discriminant
Dy (f (x, y)) = Ry (y 2 + yf2 (x) + f3 (x), 2y + f2 (x))
17.2. WEIERSTRASS FORM
191
1 f2
f3
1 f2 f3
= det 2 f2 0 = det
f2 2f3
2
f2
0 2 f2
= f22 + 4f3 = (2bx + c)2 + 4(x3 + x2 + x + )
3 since 6= 0. Roots of (Dy (f ))(x)
correspond to vertical lines x = x0 which are tangent to (the a ne part
of ) E at some point. Bezout tells us that the intersection number there
can only be 2, since deg(E) = 3 and {X = x0 Z} already meets E at O .
is a polynomial in
of degree
Such rst order tangencies mean the roots each have multiplicity one.
Therefore
at
has three vertical tangents (apart from
L = {Z = 0}),
p1 , p2 , p3 .
Lemma 17.2.1. The
{pi }3i=1
are collinear.
Proof. In the picture
T E
p3
p1
q
Lp p 2
1
p2
p to be the third intersection point of Lp1 p2 and E , and q the third
intersection point of LOp with E . Consider the cubic curves C1 = E ,
C2 = LOp1 + LOp2 + LOp3 , and C3 = TO E + 2Lp1 p2 . We have
de ne
C1 C2 = 3O + 2p1 + 2p22 + p + q
and
C1 C3 = 3O + 2p1 + 2p2 + 2p.
15.2, the ratio of the homogeneous polynomials de ning
C2 and C3 gives a degree 1 map E P1 (which is impossible) if p 6= q .
So p = q , and LOp is tangent to E at p. It follows that p is p1 , p2 , or
Arguing as in
192
p3 .
17. PUTTING A NONSINGULAR CUBIC IN STANDARD FORM
p1 doesn't
p1 , p2 , p3 Lp1 p2 .
The rst two are impossible since the tangent to
through
p2
p = p3 .
and vice versa; so
Hence
pass
O to Lp1 p2 (
= P1 ) presents E as
1
a 2 : 1 cover of P branched over p1 , p2 , p3 , and the image TO E Lp1 p2 of
O. Furthermore Lp1 p2 , LOp1 , TO E form a triangle, and so we can choose
0
0
0
0
new projective coordinates X , Y , Z in order that Lp1 p2 = {Y = 0},
LOp1 = {X 0 = 0}, and TO E = {Z 0 = 0}. For simplicity I'll drop the
primes and just write X, Y, Z for this new coordinate system. The
Now stereographic projection from
following picture summarizes what we know:
TOE
LOp
E
p1
Y = 0 ) p1
p2 (resp. p3 ).
where (on
X
at
Z
p2
Lp p
1 2
is at
X
Z
= 0.
Write
(resp.
2 )
for the value of
We would like an equation to correspond to this picture. Now, in
the new coordinate system, the equation of
is still of the form
F (Z, X, Y ) = ZY 2 + (2bXZ + cZ 2 )Y + X 3 + X 2 Z + XZ 2 + Z 3 ,
[0 : 0 : 1] with tangent line Z = 0.
But now (referring to the picture) also [1 : 0 : 0] E , which implies
= 0. Moreover, FY (= 2Y Z + 2bXZ + cZ 2 ) = 0 at p1 = [1 : 0 : 0],
p2 = [1 : 1 : 0], and p3 = [1 : 2 : 0] since the tangents are vertical
there. This yields c = 0, then 2b1 = 2b2 = 0. As the {pi } are distinct
(so i 6= 0), we have , and
because we still have a ex at
F (Z, X, Y ) = Y 2 Z + X(X 2 + XZ + Z 2 )
= Y 2 Z + X(X 1 Z)(X 2 Z).
EXERCISES
193
Now de ne new coordinates by the projective transformation
r
X=
4
1 + 2
X0 +
Z0 , Y = iY0 , Z = Z0 ,
which makes the equation
r
F (Z0 , X0 , Y0 ) = F
4
1 + 2
X0 +
Z0 , iY0 , Z0
= Y02 Z0 + 4X03 g2 X0 Z02 g3 Z03 .
Dropping the subscript
0's and taking the a ne equation,
we have put
in Weierstrass form :
Proposition 17.2.2. (a) Any smooth cubic
E P2
is projectively
1
equivalent to a curve with a ne equation of the form
y 2 = 4x3 g2 x g3 .
(b) This form is unique (given
4
( g2 , g3 )
where
; in particular,
j :=
is an invariant of
E ) up to a change of the form (g2 , g3 ) 7
g23
C
g23 27g32
E.
Proof. We have just seen (a).
equation
To see (b), write the projective
Y Z = 4X g2 XZ g3 Z 3 .
It is not di cult to see that
any linear transformation preserving the form of this equation (up to
rescaling) has the form
3
Z . Taking
2 0
gives exactly the
is unchanged by this
X = X0 , Y = Y0 , Z =
claimed eect on (g2 , g3 ), and j
:=
transformation.
Exercises
(1) Show that the cubic curve
P2
C = {0 = X 3 +Y 3 XY (X +Y +Z)}
has one (ODP) singular point and exactly three collinear exes.
[Hint: start by computing the Hessian, then nd the Hessian curve
and determine its intersections with
1note
that the vanishing of the
roots sum to zero
x2
C .]
term of the right-hand side indicates that its
CHAPTER 18
Canonical normalization of the Weierstrass cubic
This chapter will focus on the precise relationship between Weierstrassform elliptic curves and complex 1-tori (or equivalently, 2-lattices in
C).
E a period lattice
E . These will ulti-
We will begin by associating to a Weierstrass cubic
E ,
and to a (full) lattice
a Weierstrass cubic
mately be shown to be bijections of sets and mutual inverses. The key
-function and its derivative
Abel map u : E C/E . This
step is the inversion of the Weierstrass
P2 )
(embedding a 1-torus in
by the
map is closely related to the elliptic integral
dx
p
,
4x3 g2 x g3
a variant of which will be studied in the exercises.
18.1. Holomorphic forms on an elliptic curve
Let
be a Weierstrass cubic, viz., the projective closure of
f (x, y) := y 2 Q(x) = 0
in
P2 ,
where
Q(x) = (x e1 )(x e2 )(x e3 ) ,
e1 + e2 + e3 = 0.
dx
Claim 18.1.1. :=
1 (E) is nowhere vanishing.
y
E
Remark 18.1.2. This statement perhaps requires clari cation. You
dx
may interpret
y
in either of two equivalent ways:
E
dx
2
(a) any algebraic dierential form (such as
) on C extends to a meroy
2
morphic form on P , and you can think of |E as shorthand for pullback
: E , P2 just to write dx
);
y
Y
X
(b) alternatively, writing x =
and y =
exhibits x and y as meroZ
Z
2
morphic functions on P (and hence, via pullback, on E ), and Example
d( x|E )
13.1.4 tells us that
is a meromorphic 1-form.
y|E
1
Either way, we have K (E); and part of the content of the Claim is
to
(rather than introducing
195
196
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
that
is actually holomorphic:
vanishing statement says that
p () 0 for all p E . The
actually p () = 0 for all p.
Proof. Look at the a ne part
E\O.
Wherever
nowhere
fy 6= 0,
so that
dx
gives a local coordinate,
y
is holomorphic and nonvanishing. We
E
precisely at the three points {(ei , 0)}i=1,2,3 ,
f = 0 and fy = 0
0
where fx = Q (ei ) 6= 0 so that y is a
dx
0
0 = df = 2ydy Q (x)dx so that y
have
local coordinate. On
we have
dy
2 0 ,
Q (x) E
which is evidently nonvanishing and holomorphic in a neighborhood of
each
(ei , 0).
O = [0 : 0 : 1]? By PoincarpE p () = 2g 2 = 0, so that if p () = 0 for all
be no contribution from O either.
What about the ( ex) point at in nity
g = 1 =
p E\O, there can
Hopf,
1 (E)
= C hi.
multiple of .
Corollary 18.1.3.
1-form
on
is a
That is, every holomorphic
0 1 (E), the discussion preceding Example
0
K(E). But since is nowhere vanishing, 0 is
13.1.2 tells us
actually a holomorphic function. Now use Liouville's theorem (O(E)
=
C).
Proof. For any
Amongst the standard topological invariants of a 1-manifold
its rst homology group. An ad hoc de nition is
(
H1 (M, Z) := (
free abelian group generated by
closed piecewise-C
paths on
subgroup generated by
boundaries of nitely triangulable regions
or simply cycles modulo boundaries. From the picture
),
is
18.1. HOLOMORPHIC FORMS ON AN ELLIPTIC CURVE
197
it isn't hard to convince yourself that
H1 (E, Z)
= Z h, i .
That is, for any closed
(with boundary
E , there exists a closed set E
path
such that
= m + n + .
The integers
m, n
are uniquely determined by
+m
+n
=m
+n
One then has
by Cauchy's theorem (Prop. 13.1.5). The values of the integrals
, and we de ne
, C.
over cycles are called the periods of
E := Z
the period lattice
This furnishes an invariant of the complex structure on
E which, unlike
the topological invariant, actually distinguishes elliptic curves which
are non-isomorphic as complex manifolds (or algebraic curves).
Remark 18.1.4. Given a lattice of the form
we have a Weierstrass
Z h1 , 2 i =: C,
P -map
P
C/ P2
u 7 [1 : (u) : 0 (u)]
whose image (by Exercise 5 of Chapter 7) is a Weierstrass cubic! De ne
E := P(C/),
which we henceforth consider to be the range of the map
P.
Obviously
it is of interest to nd out whether all Weierstrass cubics arise in this
fashion (as
1algebraic
E 's).
geometers call this a transcendental invariant
198
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
P is injective. Its composi1
tion with (the x-coordinate projection) x : E P has degree 2 since
has a unique pole on C/ (at 0), which is a double pole. But since
Before moving on we should note that
mapping degrees of Riemann surfaces multiply under composition, and
the degree of
itself is
2,
that of
P : C/ E
must be
1.
18.2. The Abel map
Let
E = {y 2 = 4x3 g2 x g3 } P2
|
{z
}
Q(x)
be a Weierstrass cubic with
dx
y
1 (E).
Integrating this gives
E
a (holomorphic) map of Riemann surfaces
u : E C/E
p
p 7
O
where the integration is over any
map is well-de ned: if
0 , 00
path from
to
p.
This Abel
are two such paths, then their dierence
is closed and so
0 00 = + m + n.
Integrating, we have
= m
00
E .
+n
E has rank 2 (something
2
we haven't yet addressed). Otherwise, C/E is isomorphic to C or C ,
both of which are noncompact. Since E is compact and u is continuous,
its image must be a compact submanifold of C/E . If the latter is
noncompact then the image is therefore a point, meaning u is constant,
clearly false since 6= 0.
Remark 18.2.1. It is now easy to see that
A baby version of Abel's theorem for elliptic curves
3 is then:
Theorem 18.2.2. The Abel map is injective.
=
2
E = Z (rank 1) = C/E = C/2iZ C (by
3The real theorem, which we will deal with later,
divisors.
taking
exp).
has to be phrased in terms of
18.2. THE ABEL MAP
u(p) u(q)
Proof. (Sketch) Suppose
E;
then
mod
for
p 6= q
points of
= u(p) u(q) E .
199
Modifying the path from
to
get
m + n
by
(for some
m, n Z),
we
= 0.
p
Dirichlet's existence theorem (which we won't prove) guarantees the
0 K1 (E) with only simple poles, only at p and q , with
Resp (0 ) = Resq (0 ) = 1. This is true for any two (distinct) points p
and q , and has nothing to do with our assumption (that u(p) = u(q)).
existence of
Now referring to the picture
we have
H1 (E\({p} {q}), Z)
= Z h, , i
(18.2.1)
where
0 = 2i.
Next, normalize
0 ,
putting
0
:= 0
,
which has the same residues as
0 .
Observe that
= 2i,
while
= 0.
Cutting open the above gure along
domain
4In
and
yields the fundamental
4
(the yellow region):
order to accomodate the path from
by an integer factor
M.
to
p,
it may be necessary to dilate
(You can think of this as the fundamental domain of a
200
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
pat
q
O
On the interior of
F, U :=
gives a holomorphic function which is
continuous on the boundary. Now
= U(p) U(q)
0=
p
which by the Residue theorem
Noting that
(resp.
to ), this
1
=
2i
where
6= 0.
U .
F
) is the change in U from
Hence,
1
=
2i
1
=
2i
to (resp.
,
= 0.
By (18.2.1), any closed path on
n + m + ` ; and
` = 2i`. Consequently,
of the form
so
E\({p} {q}) is, up to boundaries,
the integral of over such a path is
F := exp
O
{p} {q}. Let
z (resp. w) be a local coordinate about p (resp. q) with z(p) = 0
dz
(resp. w(q) = 0). We know that the leading term of at p is
, and
z
dw
at q is
. This makes F locally at p (resp. q ) the product of a
w
dz
dw
log z
= z (resp. e w = w1 ), so that
holomorphic function by e z = e
is a well-de ned function on
by
which is holomorphic o
E .) This doesn't
M + M M M .
nite unbranched covering of
replacing
really aect the proof, except for
18.3. ABEL INVERTS WEIERSTRASS
is meromorphic on
201
with divisor
(F ) = [p] [q].
Therefore
deg(F ) = 1,
making
F : E P1
an isomorphism, which is
impossible.
We conclude from this contradiction that
p and q
cannot have been
distinct.
Abel's theorem is usually paired with something called Jacobi inversion, the baby version of which is:
Proposition 18.2.3. The Abel map
is surjective (and thus an
isomorphism).
Proof. This is trivial:
is a closed mapping (since continuous),
and an open mapping (since holomorphic and not constant). The image
is therefore open and closed in
C/E ;
since the latter is connected,
we're done.
Essentially all of the foregoing (with the exception of Remark 18.1.4)
works for any nonsingular cubic. There is a unique holomorphic
1-form
up to scale; it vanishes nowhere; and integrating it from a base point
gives an isomorphism from the cubic to a complex 1-torus. This follows
from the last 2 sections by applying the projective transformation of
Chapter 17 to put the cubic in Weierstrass form (which has just been
slightly more convenient for writing down
).
For the next section,
however, the Weierstrass form will be crucial.
18.3. Abel inverts Weierstrass
We now make the Big Claim that
(18.3.1)
a Weierstrass cubic is always the image
(cf. Remark 18.1.4), and we have
(18.3.2)
u P = idC/
and
(18.3.3)
P u = idE .
of a Weierstrass
P -map
202
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
First we study the case where
is (by assumption) the image of a
P -map.
= Z h1 , 2 i C
Proposition 18.3.1. Let
be a lattice.
The
composition
(
(
=)
=)
C/ E C/E
u
P
is the identity.
Proof. Obviously part of the claim is that
E :=
H1 (E , Z)
= .
(18.3.4)
For
E ,
(x, y)
not a lot is lost by working in a ne
there is only
at
coordinates, since
and we know that corresponds to
u0
on the
complex 1-tori. (Note also that u is used both as the Abel map and
as the coordinate on
Since
C; which one will be clear from the context.)
P(u) = ((u), 0 (u)),
!
dx
0 (u)du
d((u))
P = P
=
= du.
=
y E
0 (u)
0 (u)
P is an isomorphism
H1 (C/, Z)
Moreover, that
image of some
means any cycle
on
is the
so that
=
gives a bijection between
P(u0 )
u(P(u0 )) =
O
P =
=
E
dx
=
y
P (
)
u0 C/,
and
hence (18.3.1). So then taking
P(u0 )
P =
=
P(0)
du
u0
du = u0
0
18.3. ABEL INVERTS WEIERSTRASS
203
proves the Proposition.
Now let
be any Weierstrass cubic.
Proposition 18.3.2. The composition
(
(
=)
=)
E C/E EE P2
P
u
is the identity. In fact,
E = EE
(18.3.5)
exactly as subsets of
P2 .
(x, y) E :
Proof. Here is what the composition looks like, where
(x, y) 7
dx
P
p
7
Q(x)
dx
p
Q(x)
, 0
dx
p
Q(x)
is determined by y . We must show that the right-hand side
recovers (x, y), or equivalently that the inverse (x(u), y(u)) : C/E
E of the Abel map u identi es with ((u), 0 (u)).
Let's start with x, and compare the elliptic functions x(u) and P(u)
on C/E . First I claim that both have double poles at u = 0: you
1
already know that (u) = 2 + higher-order terms. For x, it su ces to
u
check this on E , using
X
x = K(E)
Z E
where the
{Z=0}=TOE
X=0
. . . which is easy:
O (x) = (E {X = 0})O (E {Z = 0})O = 1 3 = 2.
!!
,
204
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
Now
x(u) =
A
u2
+ h.o.t.,
with
A=
x dw 2
Q(w)
2
lim x(u) u = lim x u(x) = lim
u0
x
x
1/ x
lim
!2
p
1/ Q(x)
1/2x
3
2
4x3
= 1.
x Q(x)
= lim
De ne an involution
:EE
by
(x, y) 7 (x, y);
p E,
(p)
p
p
dx
dx
dx
u((p)) =
=
=
= u(p),
y
O=(O) y
O
O y
this xes
O.
For
and so
y(u) = y(u).
x(u) = x(u) ,
x(u) and (u) are both even -periodic
1
functions locally of the form 2 + h.o.t., and so their dierence has no
u
poles and must (by Liouville) be constant: x = c.
x dx
du
Next, dierentiating u =
gives
= y1 , or
dx
y
All told, we now have that
x0 (u) =
dx
= y(u);
du
and then
d
(c) = x0 (u) 0 (u) = y(u) 0 (u).
du
All that is left is to check that c = 0.
The xed points of the involution u 7 u are the 2-torsion
i.e. those u C/E with 2u 0
0=
u3
O
since we must have
(by
u)
u u
u1
P)
E . These are, of course, the images
in E , since u = u. They also
points of (x, y) 7 (x, y) in EE , since
mod
of the xed points of
must map (by
points,
to the xed
EXERCISES
((u), 0 (u)) = ((u), 0 (u)).
205
Writing
y 2 = 4x3 g2 x g3 = 4(x e1 )(x e2 )(x e3 )
for the equation of
E,
O
e2
e1
x (2-torsion
Similarly, if
e3
points)
= e1 , e2 , e3 , .
EE = {y 2 = 4x3 g2 x g3 = 4(x e1 )(x e2 )(x e3 )}
then
(2-torsion
points)
= e1 , e2 , e3 , ;
and clearly
e1 + e2 + e3 = e1 + e2 + e3 = 0.
Since
(u) = x(u) + c,
(u1 ) + (u2 ) + (u3 ) = x(u1 ) + x(u2 ) + x(u3 ) + 3c
which becomes
0 = 0 + 3c
so
c = 0.
We conclude that
(u(x, y)) = x
and
0 (u(x, y)) = y .
Exercises
(1) [Adapted from Silverman-Tate.]
Let
0 < ,
and consider the ellipse
de ned by
y2
x2
+
= 1.
2 2
E is given by the
p
1 k 2 sin2 d
Show that the arc-length of
4
0
integral
206
18. CANONICAL NORMALIZATION OF THE WEIERSTRASS CUBIC
for an appropriate choice of constant
depending on
and
Then prove that this is equal to
4
0
1 k 2 t2
p
dt.
(1 t2 )(1 k 2 t2 )
hence the problem of determining arc-length comes down to evaluating the integral
on the elliptic curve
Remark:
1 k 2 t2
dt
u
0
u2 = (1 t2 )(1 k 2 t2 ).
this curve obviously isn't cubic but is still of genus 1
because of the singularity at in nity. You can think of this as what
you get if instead of writing an elliptic curve as
branching at (i.e.
2:1
over
P1
with
order 2 rami cation over) 3 nite points plus
in nity, you take the branching to be over 4 nite points. So it's
very close to a Weierstrass cubic, even though it's quartic.
The integral in this problem is of course related to the Abel map/abelian
integral above, and is meant to demonstrate why such integrals
(and hence these curves) are called elliptic.
CHAPTER 19
Group law on the nonsingular cubic
It is now high time for the smooth version of Chapter 16: group
laws and addition theorems for elliptic curves. We start by introducing
an algebro-geometrically de ned binary operation on the points of a
cubic curve, and prove it coincides with addition on
C/E
under the
Abel isomorphism. This gives one proof that the operation de nes an
abelian group, and we give another more natural one as well.
From
the fact that it makes Abel's map into a homomorphism we will then
derive functional equations for elliptic integrals.
19.1. De nition of the group law
Let
E P2
be a nonsingular cubic, which we shall not require to
be in Weierstrass form, and x a ex
Step 1: Draw the line
L1
O.
Let
and
q:
through
and
be points of
E.
O
p
L1
p*q
q
By Bezout, there is a third intersection point, which we shall denote
p q,
so that
L1 E = [p] + [q] + [p q].
Note that the three points need not be distinct any two, or all three,
may coincide.
This has the usual interpretation: double-intersection
207
208
19. GROUP LAW ON THE NONSINGULAR CUBIC
means
L1
is tangent to
at that point; triple-intersection means you
have a ex.
Step 2: Draw the line
L2
through
and
p q:
O
p
L2
p*q
p+q
The third intersection point is denoted
p + q,
with the same interpre-
tations as above.
In the special case where
is in Weierstrass form,
{X = x0 Z}. So the map sending p q
: E E taking (x, y) 7 (x, y):
line
to
p+q
L2
is a vertical
is just the involution
E
j
e1
e2
e3
We have therefore constructed a binary operation
EE E
(p, q) 7 p + q
on the (set consisting of the) points of
E.
It is not yet clear that this
de nes a group. It is clear that it is commutative, so that if it de nes
a group, then that group is abelian.
19.2. RELATION TO THE GROUP STRUCTURE ON THE 1-TORUS
209
19.2. Relation to the group structure on the 1-torus
Let
T : P2 P2
be the projective transformation putting
into
O0 := [0 : 0 : 1]). Denote the
0
binary operation de ned on points of E := T (E) (via the method just
0
0
described, with O replacing O ) by + . Since projectivities preserve
0
lines, intersection multiplicities, and so forth, it is clear that T (p) +
T (q) = T (p+q). So to show that + de nes a group law for arbitrary
E , it su ces to check this for Weierstrass cubics.
Hence we may assume E is in Weierstrass form (and O = [0 : 0 : 1]).
Weierstrass form (and taking
to
Take the Abel map
u : E C/E
to be as in Chapter 18, with inverse
E
de nes a group law on
our operation on
C/E .
P.
We know that addition mod
The next result implies not only that
E de nes a group law, it says that u is an isomorphism
of groups.
Theorem 19.2.1. The Abel map respects binary operations. That
is, the equivalent formulas
P(u1 ) + P(u2 ) = P(u1 + u2 )
(19.2.1)
u(p) + u(q) u(p + q)
(19.2.2)
mod
hold.
Proof. We will prove (19.2.2), in a manner reminiscent of the
proof of Theorem 18.2.1.
polynomial de ning
Li ,
Writing
FLi
for the degree-1 homogeneous
consider the meromorphic function
FL2
f :=
K(E) .
FL1 E
Reading its divisor o from the intersection points of the
{Li }
(f ) = [p + q] + [p q] + [O] ([p] + [q] + [p q])
= [p + q] [p] [q] + [O].
So for its pullback
P f = f P
to
C/E ,
(P f ) = [u(p + q)] [u(p)] [u(q)] + [0].
and
E,
210
19. GROUP LAW ON THE NONSINGULAR CUBIC
Cut open
region
C/E
and put
U :=
F:
du
on the resulting fundamental
u(p+q)
u(q)
u(p)
Write
U(p)
for
U(u(p))
(and so on) for simplicity. By the residue theo-
rem,
U(p + q) U(p) U(q) (+0)
1
d(P f )
=
U
2i F
P f
1
1
dlog(P f )
du +
dlog(P f )
du.
=
2i
2i
Since both terms in braces are integers, the whole thing belongs to
du = E .
du,
U(p), U(q), U(p + q) are lifts to C of u(p), u(q), u(p + q),
modulo E we see that u(p + q) u(p) u(q) 0.
Since
going
We will generalize this argument in the next chapter to get Abel's
theorem for
E.
Remark 19.2.2. For an arbitrary smooth cubic
to scale) a unique
(C),
u : C C/(C,) .
C , one still has (up
which gives rise to an Abel isomorphism
(19.2.2) still holds in this setting by more or less
the same proof; this avoids passing through Weierstrass form.
19.3. A more algebro-geometric approach
Returning to the setup of
19.1,
let us suppose that the coe cients
of the homogeneous cubic polynomial de ning
eld
k C.
extension of
of
We shall say
(e.g.
itself, or
belong to some sub-
K C
is a eld
then we can consider the
K -points
is de ned over
C),
E
k.
If
E
E(K) := {[Z : X : Y ] E | Z, X, Y K}.
Proposition 19.3.1.
quently a subgroup of
E.
E(K)
is closed under +, and is conse-
19.3. A MORE ALGEBRO-GEOMETRIC APPROACH
211
p, q E(K)
then the line L1 = Lpq is de ned over K ,
1 =
and so can be parametrized P Lpq over K . (This means that the
Proof. If
formulas expressing
e cients in
[Z : X : Y ] as functions of [T0 : T1 ] P1 involve co-
K .)
So the pullback of the homogeneous polynomial de nP3
3j j
is de ned over K . Now this can be written
j=0 j Z0 Z1 =
ing E
Q3
i=1 (Z1
i Z0 ), assuming without loss of generality that there are no
Z0 factors; and what we know is that the j K . This is no guarantee
that the i K . But in this case we know that two of them say,
1 , 2 correspond to p, q and so must belong to K . Consequently
3 K as well, and its image point p q is also de ned over K . Repeat
the argument for L2 and the claim follows.
Remark 19.3.2. (a) In light of the above de nition, the correct
notation for the set of complex points of
E,
which we have heretofore
denoted simply E , is
E(C).
(b) As a C-module, E(C)
= C/E has rank 1, but as an abelian
group (i.e. Z-module), its rank is in nite. (Consider a bunch of Qlinearly independent complex numbers modulo E there is no bound
on the size of the Q-vector space you can generate in this fashionl.) On
the other hand, for the subgroup E(Q), a famous theorem of Mordell
(1922) asserts that the rank (as an abelian group) is always nite. In
problem (6) below, you will show computationally that the rank of
E(Q)
in one example is at least
Now the Abel map
1.
u is non-algebraic (i.e., transcendental); it should
be seen as providing a link between the complex algebraic and the complex analytic. Such maps, which include multivariate abelian functions,
modular and automorphic forms, are very important in arithmetic algebraic geometry.
While they do not preserve the eld of de nition,
they have nonetheless been essential to the study of things like class
eld theory, the proof of Fermat's last theorem, and so on.
light of the subgroup structures
E(K) E(C),
Still, in
it is a bit sloppy to
prove that + is a group law in a manner that only works over
C.
So we will now give the fully algebraic approach to this proof, by
checking
(19.3.1)
O + p = p ( p E),
212
19. GROUP LAW ON THE NONSINGULAR CUBIC
(p q) + (p + q) = O ( p, q E),
(19.3.2)
+ is
(19.3.3)
Notice that (19.3.2) implies
p O = p,
and
associative.
(p O) + (p + O) = O,
and thus that
so we indeed have inverses. (19.3.1) and (19.3.3) are the
other two group axioms.
To verify (19.3.1), we have the pictorial depiction of the two-step
process for adding
and
p:
O*p
O*p
L2
L1
E
O+p
Step II
Step I
L1 = LOp and L2 = LO,Op = L1 !! Consequently the third
intersection point O + p of L2 with E , is none other than p. (19.3.2)
in which
isn't much harder; again, a picture:
O =(p*q)*(p+q)
L2
p*q
L1
q
p+q
L1 = Lpq, p+q . Since the line through O and p q intersects
(and de nes) p + q , it follows that L1 's third intersection point
where
in
19.3. A MORE ALGEBRO-GEOMETRIC APPROACH
(p q) (p + q)
with
is just
O.
213
Then
L2 := LO,(pq)(p+q) = LO,O = TO E,
and since
again
is a ex ((TO E
E)O = 3),
the third intersection point is
O.
Finally we come to the associativity issue (19.3.3).
Here I won't
break the two steps up into two pictures. Instead, here is a depiction
of
(p + q) + r,
the addition of
where the blue lines compute
and the green ones
to the result:
L1
p+q
L1
r
p
(p+q)*r
p*q
q
p+q
(p+q)+r
L2
L2
...
p + (q + r) looks
p):
l1
l1
r
p
and here is what
for adding result to
like (blue lines for
q + r,
p*(q+r)
q*r
l2
q
p+(q+r)
q+r
l2
We have to show
(p + q) + r = p + (q + r),
or equivalently
(p + q) r = p (q + r).
green
214
19. GROUP LAW ON THE NONSINGULAR CUBIC
Now look at the three cubics
E , C = L1 `2 L01 , and D = `1 L2 `01 ,
with intersections
E C = [O] + [p] + [q] + [r] + [p q] + [p + q] + [q r] + [q + r] + [(p + q) r]
and
E D = [O]+[p]+[q]+[r]+[pq]+[p+q]+[q r]+[q +r]+[p(q +r)].
Argue la
and
15.2:
the ratio of the homogeneous polynomials de ning
induces a meromorphic function on
with divisor
E C E D = [(p + q) r] [p (q + r)],
leading as usual to a contradiction unless these two points are the same.
19.4. Addition theorems
Now assume
E = {y 2 = Q(x)}
(with
Q(x) = 4x3 g2 x g3 )
is in
Weierstrass form; we would like to unwind the statements (19.2.1) and
(19.2.2) that
and
u are group homomorphisms (hence isomorphisms),
to produce something more computationally explicit.
We do this rst for the Weierstrass map.
((u1 ), (u1 )), q = P(u2 ) = ((u2 ), (u2 )),
p = P(u1 ) =
pq
Writing
we have
= (p+q) = (P(u1 )+P(u2 )) = (P(u1 +u2 )) = ((u1 +u2 ), 0 (u1 +u2 ))
= ((u1 + u2 ), 0 (u1 + u2 )).
Now
p, q ,
and
pq
are collinear by construction they lie on
in the group law process for
E.
L1
We may express this in projective
coordinates by saying that
1
1
1
0 = det (u1 ) (u2 ) (u1 + u2 ) .
0 (u1 ) 0 (u2 ) 0 (u1 + u2 )
This is the rst addition theorem, and is the analogue for bi-periodic
functions of the standard trigonometric angle-addition formulas.
really does express
It
(u1 +u2 ) in terms of (u1 ) and (u2 ), since () =
p
Q(()).
Let's actually compute the group law on
ax + b for L1
E.
Start by writing
and substituting this into the equation of
y=
to intersect
19.4. ADDITION THEOREMS
215
them. This gives
0 = 4x3 g2 x g3 (ax + b)2 = 4(x x(p))(x x(q))(x x(p + q))
since
L1
and
meet in
p, q, p q .
(Note that
x(p + q) = x(p q).)1
From expanding these two expressions and comparing coe cients of
x2 ,
one nds that
a2 = 4(x(p) + x(q) + x(p + q));
and since
is the slope of
L1
it is obvious that
a=
y(q) y(p)
.
x(q) x(p)
Therefore we have
2
1 y(q) y(p)
x(p) x(q).
x(p + q) =
4 x(q) x(p)
p dx
x(p) dx
x(q) dx
Now u(p) =
=
, similarly u(q) =
O y
Q(x)
Q(x)
x(p+q) dx
u(p + q) =
. Re-expressing
and
Q(x)
u(p) + u(q) u(p + q)
mod
using all these formulas yields the second addition theorem :
(
)
x2
1 Q(x2 ) Q(x1 ) 2 x2 x1
x1
4
x x
dx
p
+
Q(x)
dx
p
Q(x) E
dx
p
Q(x)
dx
which is a nontrivial functional equation for the elliptic integral
.
Q(x)
The problems below (with the exception of the last one) take place on a
E := {y 2 = 4x3 g2 xg3 } P2 in Weierstrass form,
dx
with base point O = [0 : 0 : 1], holomorphic form =
| 1 (E),
y E
p
and Abel map u : E C/E , u(p) =
(recall this is an isoO
0
morphism), with inverse P(u) = [1 : (u) : (u)]. I have written
nonsingular cubic
1the
x(p) = (u1 ), y(p) = 0 (u1 ); x(q) = (u2 ),
y(q) = (u2 ); x(p + q) = (u1 + u2 ), y(p + q) = 0 (u1 + u2 ).
dictionary we have in mind is:
216
19. GROUP LAW ON THE NONSINGULAR CUBIC
everything in a ne form, which you can convert to projective coordinates if needed.
Exercises
C/E correspond to the x3
intercepts {(ei , 0)}i=1 and the point O .
Assume g2 , g3 Q. Given p, q E(Q) (i.e. the x, y coordinates are
rational), give another proof that p + q E(Q), using the addition
(1) Show that the 2-torsion points on
(2)
theorems.
g2 = 4, g3 = 0.
Consider the complex analytic automorphism A : E E sending
(x, y) 7 (x, iy). In Ch. 13 Exercise (1), you showed that A =
i .
(a) Find A u (i.e. compute u A).
(b) Prove that iE = E . (In fact, E is a square lattice so
(3) For this and the following three problems take
this is a very special elliptic curve!)
(4) Special case of the 2nd addition theorem (or rather, what we did
in
19.4
above doesn't exactly work in the case we'll do here, so
(2u) in terms of
(u), for E as in exercise (3), i.e. with equation y 2 = 4x3 + 4x.
0
0
[Hint: write (u) and then the slope a of E at ((u), (u)), in
terms of (u). Write y = ax + b for the line tangent to E at this
3
2
point. Then factor 4x + 4x (ax + b) into linear factors (what are
you'll have to work it out from scratch): write
the roots?), multiply out both expressions, and compare coe cients
of
x2 .]
(5) Continuing the last problem, show that
point.
(1, 2 2)
is a
4-torsion
Use the CM recalled in exercise (3) to get three more
4-torsion
points. Can you use the group law to nd them all? (If
not, why?)
(6) This problem also depends on (4). Consider a point
of
with
x-coordinate x0 = 2a q , where the fraction is written in
lowest terms, a is an odd natural number and p and q are odd
integers. Show that P is of in nite order (in the group). [Hint:
write (x0 , y0 ) for this point, and let (x1 , y1 ) := 2(x0 , y0 ) under the
group law; if x0 = (u), then x1 = (2u). So rewriting your
rational
EXERCISES
formula from (4) as a formula for
in terms of
x0
and simplifying,
a. Then suppose
the starting point was an N -torsion point for some N and produce
show that
x1
x1
217
is of the same form, but with larger
a contradiction via the pigeonhole principle.]
(7) This problem takes place on an arbitrary nonsingular cubic
P , with base point O = choice of ex in E , holomorphic form
1 (E) with periods generating a lattice E , and Abel map
p
=
u : E C/E , u(p) = O .
Prove that the 3-torsion points on C/ correspond (under u) to
the exes on E .
CHAPTER 20
Abel's theorem for elliptic curves
Given a divisor
D=
ni [pi ] on an elliptic curve E , we can formally
compute the sum in the group law, ending up with a single point on
E.
It seems of interest to ask if anything special is true if this point
is the origin
O.
In fact, assuming
is true precisely if
ni = 0 ,
it will turn out that this
is the divisor of a meromorphic function on the
curve. We begin by describing the statement of Abel's theorem for a
curve of arbitrary genus (which does not have a group law), to place
the statement for genus one in a broader context. Then we prove the
genus-one case, introducing theta functions along the way.
20.1. The Jacobian of an algebraic curve
Let
be a Riemann surface of genus
g.
We will need to accept
some facts in order to state Abel's theorem for
M.
(These will be
returned to in later chapters, along with the proof of Abel.) It turns
out that the space of holomorphic
abelian group of
has rank
2g .
1-cycles
1-forms
has dimension
modulo boundaries (cf.
18.1
g,
whilst the
for de nitions)
In terms of bases,
H1 (M, Z)
= Z h1 , . . . , 2g i ,
1 (M )
= C h1 , . . . , g i .
Remark 20.1.1. A visual explanation of the statement about ho-
mology groups may be the best one:
g+2
g+1
2g
g
Chapter 3 of Gri ths's Introduction to Algebraic Curves gives one
219
220
20. ABEL'S THEOREM FOR ELLIPTIC CURVES
approach to computing the holomorphic forms of
M,
provided one be-
lieves that any Riemann surface is the normalization of an algebraic
curve
in
P2
with only ordinary double point (if any) singularities.
(This statement relies on the existence of nonconstant meromorphic
functions on M , which is nontrivial.)
Since the genus g of M is
(d1)(d2)
(d = deg(C), =# of ODPs), it is enough to show
2
24.1) and furthermore
2
that holomorphic pullbacks of rational 1-forms from P span a space of
d1
dimension
. This is done in Gri ths, and will be discussed a
2
little more in 24.2.
Just to get an idea of how this works, suppose C = {F (Z, X, Y ) =
0} is smooth of degree d, and recall S3m denotes degree-m homogeneous
m+2
polynomials in 3 variables, with dimension
. If G is a homoge2
neous polynomial of degree n, write g(x, y) = G(1, x, y) (and similarly
f (x, y) = F (1, x, y)). Then the meromorphic 1-form on P2 which in
gdx
a ne coordinates takes the form
, restricts to a holomorphic 1-form
fy
1
on C precisely if n = d 3. (This is equivalent to saying deg(g)
(d1)(d2)
d1
=
=
d 3.) Hence,2 1 (C) has dimension (d3)+2
.
2
2
2
that all meromorphic
Anyhow, let
1-forms
are rational (cf.
j H1 (M, Z)
period vector
be a basis element; associated to it is a
j :=
j 1
.
.
.
j
Together these form a
g 2g
Cg .
period matrix
R-linearly independent columns. (This isn't obvious, and will be addressed in 24.2.)
g
Hence their columns generate (over Z) a 2g -lattice M C (
= R2g ).
Recall that if V is a vector space (say, over C) then the dual space
is the space of linear functions V
:= Hom(V, C).
Definition 20.1.2. The Jacobian of
with
is the abelian group
J(M ) :=
1the
(1 (M ))
,
image {H1 (M, Z)}
computation in Gri ths proving this is ugly but straightforward; Poincar
residues facilitate a conceptual and essentially 1-line proof (but at the cost of more
sophisticated machinery).
2putting
o to
24.2
that this formula encompasses all rational holomorphic forms
20.2. THE ABEL-JACOBI MAP
221
where the denominator means the linear functions on
1 (M )
obtained
1 (M ) over 1-cycles. Evaluation of linear functions
basis {1 , . . . , g } induces an isomorphism
by integrating
against the
J(M )
that is, the Jacobian is a complex
Cg
;
M
g -torus.
Lemma 20.1.3. Any morphism of complex manifolds
Cg /M
: P1
is constant.
Cg /M has g independent holomorphic 1-forms: du1 , . . . , dug
g
(where u1 , . . . , ug are just the coordinates on C ). Since (dui )
1 (P1 ) and 1 (P1 ) = {0}, we have
Proof.
0 = (dui ) = d( ui )
locally
ui = ui
i = 1, . . . , g .
which implies
each
(well-de ned only locally) is constant for
20.2. The Abel-Jacobi map
(f ), for some nontrivial mero
morphic function f on M ? Since deg((f )) = 0 for any f K(M ) , it
is clear that D must be of degree 0 i.e. in the kernel of
When is
Div(M ) of the form
deg : Div(M ) Z
X
X
ni [pi ] 7
ni .
So consider a divisor
in
Div
(M ) := ker(deg).
We may write
D=
([qj ] [rj ]) =
j j
|
where means topological boundary and
to
{z
=:
r
j qj
qj .
Definition 20.2.1. The Abel-Jacobi map
AJ :
r
q
Div
(M ) J(M )
}
is a
path from
rj
222
20. ABEL'S THEOREM FOR ELLIPTIC CURVES
sends
D (= )
to
viewed as a functional on
qj
rj
1 (M ).
The rst question that arises is whether this is even well-de ned,
which in this case means independent of the choice of 1-chain (sum
of paths)
To check this, let
meaning that
= D = 0 .
is a 1-cycle hence represents a
Consequently,
belongs to the denominator of
AJ
=
0
( 0 ) = 0,
class in H1 (M, Z).
Then
J(M ).
It's even easier to check that
is a homomorphism (of abelian groups), which is left to you.
Now suppose
D = (f ),
and consider the family of divisors
Dt := f 1 (t) Div(M ),
parametrized by
t P1 .
Then
D = D0 D ,
and the composition
AJ
P1 Div0 (M ) J(M )
sending
t 7 D0 Dt 7 AJ(D0 Dt )
is constant by Lemma 20.1.3, and zero at
t = 0.
Thus
AJ(D) = 0, and
we observe that
AJ
0
factors through Pic (M )
:=
(M )
(K(M ) )
Div
in a well-de ned fashion. (The denominator means divisors of meromorphic functions, and the statement is simply that
Pic
(M )
is called the Picard group of
AJ
kills these.)
M.
Theorem 20.2.2. [Abel's Theorem]
AJ :
Pic
(M ) J(M )
is
an isomorphism.
Leaving aside the surjectivity part of this, the meaning of the wellde nedness + injectivity of this map is that for
(for
D = (f )
some f K(M ) )
D Div0 (M ),
AJ(D) 0
mod
M ,
20.3. DIRECT PROOF OF ABEL FOR GENUS ONE
223
completely answering the question we asked at the outset. (Note that
the forward implication [
] is just well-de nedness, which is com-
pletely proved. What is nontrivial is the injectivity/backward implication, since you actually have to nd some
having
as its divisor.)
1 case,
1
i.e. for M = E (the normalization of ) an elliptic curve. Let (E)
P
0
be nonzero, and consider D Div (E). We can write D =
ni [pi ]
P
with
ni = 0, and
X
X
X pi
X
AJ
ni [pi ] = AJ
ni ([pi ] [O]) =
ni
=
ni u(pi )
Example 20.2.3. We consider what this means in the genus-
O
where
C/E ,
is the Abel map. Here the right-hand sum is taking place in
and we see right away that
AJ
X
ni [pi ] = 0
ni u(pi ) 0.
E
By Abel's theorem (on the left) and the fact that
(C/E , +)
u : (E, +)
is a group-isomorphism (on the right), we have that
(20.2.1)
for
ni [pi ] = (f )
some f K(E)
ni pi = O
in the group law on
E(C).
As above, the forward implication has been proved.
20.3. Direct proof of Abel for genus one
In this section we will deduce a result equivalent to the backward
implication in (20.2.1), recasting it as an existence theorem for elliptic
functions. For simplicity take
= Z h1, i, H
+1
2
(upper half-plane):
1+
P
mj = 0 and
mj uj 0 mod .
P
Then, writing D :=
mj [uj ] Div(C/), there exists g K(C/)
such that (g) = D . (You may think of g as a -periodic meromorphic
function on C.)
Theorem 20.3.1. Suppose
224
20. ABEL'S THEOREM FOR ELLIPTIC CURVES
Proof. Introduce the theta function (on
(u) :=
C)
2 +2nu}
ei{n
nZ
-periodic, it has the properties
(a) (u) = (u)
(b) (u + 1) = (u) [cf. Exercises]
2i( 2 +u)
(c) (u + ) = e
(u). To check this, write ( + u)
X
X
2
2
=
ei{n +2nu+2n } =
ei{(n+1) +2(n+1)u 2u}
While it is not
nZ
nZ
m = n + 1,
X
2
ei(m +2mu)
= ei 2iu
which becomes, reindexing by
mZ
as required.
(d)
has a simple (order
1)
zero at
+1
and no-where else in the
2
fundamental domain bounded by vertices
0, 1, , 1 + .
Now consider
m
Y
+1 j
;
f (u) :=
u uj +
2
j
clearly
f (u + 1) = f (u)
by property (b); but also (using property (c))
!mj
u uj + +1
+
2
u uj + +1
2
Y
mj
1
=
e2i( + 2 +uuj )
f (u + ) Y
=
f (u)
j
j
1
= e2i( + 2 +u) mj e2i mj uj .
P
P
mj = 0 and
mj uj = M + N ,
By asssumption,
pression equals
2iN
so the last ex-
. The function
g(u) := e2iN u f (u)
will therefore satisfy
and the de nition of
(g) =
g(u + ) = g(u) = g(u + 1).
f together with property (d)
So it is
-periodic,
makes it clear that
mj [uj ].
Exercises
(1) Verify property (b) for the theta function above (20.3).
CHAPTER 21
The Poncelet problem
First let's recall the most elementary statement of the porism
from Chapter 1.
One starts with two conics
CR , D R
in
R2 ,
which
for simplicity we can take to be two ellipses cut out by polynomials
fC , fD P 2
with real coe cients:
CR
DR
1.3 whether there exists a closed polygon inscribed in CR
circumscribed about DR . The result stated there, Theorem 1.3.1,
We asked in
and
said that if there is one then there is an in nite family. Our goal in
this chapter is not just to esh out the sketch of proof given there of
this porism, but to actually provide a way of deciding for which pairs
there does exist a circuminscribed polygon.
A slight reformulation of the theorem is this: starting from some
point
hits
x0
CR
on
CR ,
draw a line segment tangent to
DR ,
continue until it
again. Begin again at this new point, by drawing the other line
segment through it and tangent to
CR
DR :
x0
etc.
L0
DR
x2
L1
x1
Iterating this construction, we may ask whether it ever closes up
i.e. returns to its starting point. (We will not care whether the path
crosses itself.) What we will show is that the answer is independent of
the choice of starting point
x0 .
225
226
21. THE PONCELET PROBLEM
21.1. Proof of Theorem 1.3.1
This Theorem has nothing to do with
and
fD
and
being ellipses,
fC
being real polynomials, and so forth it makes sense more
generally for pairs of conics in the complex projective plane
that is the context in which we view it for the proof.
P2 ,
and
Namely, let
C = {FC (Z, X, Y ) = 0}, D = {FD (Z, X, Y ) = 0} be the conics cut
2
out by homogeneous degree-2 polynomials FC , FD S . If the latter
have coe cients in R (not essential for what follows), then the real
points C(R), D(R) make sense and then CR , DR above are just their
intersections with a ne space. Now, these a ne real points need not
meet (as in the above picture), but by Bezout
and
must meet
in four points counted with multiplicity. We will carry out our proof
under the assumption that the multiplicities are all one, i.e.
meet transversely and so
and
|C D| = 4.
Consider the incidence correspondence
E := {(x, L) | x L} C D
2 is the dual curve consisting of lines tangent to D (at any
P
D
point). In 1.3 we de ned pictorially two involutions 1 : E E and
meets C in two points (counted
2 : E E . The idea is that each L D
with multiplicity), and swapping those points gives 1 ; whereas each
x C is in two lines tangent to D (counted with multiplicity), and
swapping those lines gives 2 . Composing involutions gives := 2 1 ,
where
which is no longer an involution and is the complex geometry analogue
of the iteration described just above.
(x0 , L0 ) E ,
If we pick a starting point
then we are interested in whether
n (x0 , L0 ) = (x0 , L0 )
for some
n N.
The projection
: E C(
= P1 )
(x, L) 7 x
has
21.1. PROOF OF THEOREM 1.3.1
mapping degree
2:
there exist two lines
through a general point
227
L, L0
tangent to
xC
C
D
x
4
of
rami cation points (each of order two): namely, the points
xed by the involution
In particular, the rami cation points of
C D,
identify with the points of
since through each of these there is a unique tangent to
(rather than two):
C
D
L
By the Riemann-Hurwitz formula (for ), E = dC r = 224 = 0.
This implies E is elliptic, and so has an Abel map u mapping it isomor1
phically to a 1-torus C/ (where depends on E hence ultimately on
C and D).
1to
de ne the Abel map you also have to choose a holomorphic
aects the scaling of the lattice but not its isomorphism class
1-form
on
E;
this
228
21. THE PONCELET PROBLEM
We could have carried out this same computation using
(sends
points
(x, L) 7 L), whose rami cation points
of 1 and hence identify with bitangents:
(in
E)
:E D
are the xed
C
D
L
There are four of these since
and
are conics in
2
P
hence have
= 4.
|C D|
Now consider an arbitrary involution
nate on
of
u. Any automorphism
u
7 au + b by Ch. 14 Exercise
is denoted
takes the form
C/, where the coordiof C/ (in particular I )
(5), and squaring this
gives
u 7 au + b 7 a(au + b) + b = a2 u + b(a + 1).
If this is to be the identity, we must either have (i)
or (ii)
a = 1
and
bC
a=1
and
b /2,
arbitrary. Case (i) has no xed points as it is
a translation by a 2-torsion point.
By abuse of notation
C/.
Since
and
2 we will think of , , as automorphisms of
1 2
are involutions of
C/
with xed points, they
belong to case (ii):
1 (u) b1 u ,
2 (u) b2 u
(mod ).
Therefore
(u) = 2 (1 (u)) b2 (b1 u) = u + (b2 b1 ),
| {z }
=:
i.e.
is a translation on
C/.
u0 for the image of (x0 , L0 ) under the Abel map. Clearly
(x0 , L0 ) = (x0 , L0 ) i n (u0 ) u0 (mod ). But n (u0 ) = u0 + n ,
which u0 i n 0, i.e. n . We conclude that the Poncelet
Write
2strictly speaking one should write u u1
1
ing to
on
for the involution of
C/ correspond-
21.2. EXPLICIT SOLUTION OF THE PONCELET PROBLEM
construction (starting from
and only if
on
is
n-torsion
(x0 , L0 ))
closes up at the
relative to the lattice. Since
this has nothing to do with the choice of
229
nth iteration if
depends only
(x0 , L0 ).
Q.E.D.
21.2. Explicit solution of the Poncelet problem
The exes are the preferred choices of origin for the group law on
E,
it turns out that the best choice for O is one of the xed points of 2
(the four (x, L) with x C D ). Writing C D = {p1 , p2 , p3 , p }, we
set O := (p , L ) E .
a cubic plane curve. On the incidence-correspondence elliptic curve
Le
p2
p1
e
p3
C
D
(e, Le ) is the rst point in the Poncelet iteration, i.e. (O);
clearly = u((e, Le )), with u : E C/ the usual Abel isomorphism.
n
The question of whether is the identity can be restated in terms of
the (unique) group law on E with origin O :
Here
(21.2.1)
Is
(e, Le )
an
N -torsion
point?
The approach we take to its solution in this section is work of Cayley
as presented in the nice expository article [P. Gri ths and J. Harris, On
Cayley's explicit solution to Poncelet's porism, L'Enseignement Math.
24 (1978), 31-40.].
A family of conics. Consider the collection of conics depending
on
t P1 :
Dt := p P2 | tFC (p) + FD (p) = 0 ,
where
D = C
and
D0 = D.
Each
Dt
passes through
p1 , p2 , p3 , p .
230
21. THE PONCELET PROBLEM
Recall that the equation of a conic may always be written
p.M.p = 0 ,
a symmetric 3x3 matrix;
det M = 0. Write MC , MD for the
matrices corresponding to C, D , so that tMC + MD corresponds to Dt .
Those t for which Dt is singular, are then just the ti in
the conic is singular if and only if
det (tMC + MD ) = (t t1 )(t t2 )(t t3 ).
(21.2.2)
There are three singular conics through the
{pi }i=1,2,3, :
t3
t2
p3
p1
p1
p2
p3
p2
p3
Dt
Dt
For any
t P1 ,
p2
t1
p1
Dt
let
`t := tangent
line (through
`t C =: p + qt
p )
(de nes
to
Dt
qt ).
From the pictures above, we see that
qti = pi
(i = 1, 2, 3),
q = p (double
intersection),
q0 = e.
So stereographic projection through
ization)
p gives
an isomorphism (normal-
P1 C
sending
t 7 qt ,
in particular
ti 7 pi (i = 1, 2, 3),
7 p ,
21.2. EXPLICIT SOLUTION OF THE PONCELET PROBLEM
231
0 7 e.
This makes
existence domain (cf.
3 of
P1 branched at t1 , t2 , t3 , i.e. the
p
2.3) of (21.2.2), which is to say the Riemann
a double-cover
surface
(21.2.3)
s2 = det(tMC + MD ) =: E.
(e, Le ) on E corresponds to a point over t = 0 on E ; call this
. (Moreover, O E corresponds to [0 : 0 : 1] =: O E , as it should.)
The point
Summarizing everything in a picture:
(e,L e )
(p ,L )
P1
t1,t2,t3,t 4
p 1,p 2,p 3,p4 = C D
=(0,s 0)
P1
t gives local
(t 1,0)
holo. coord.
with t( )=0
(t 2,0)
(t 3,0)
Our main question (21.2.1) becomes:
Is
Now
t1 + t2 + t3
N -torsion
E?
on
may not be zero and we are lacking a factor of
4,
so
is not quite in Weierstrass form. But it is easy to see that we have
a normalization
P : C/ E
given by
Clearly this sends
P
ti 0 (u)
u 7 (u) +
,
.
1
3
2 2
0 7 O; de ne u0 C/ to be the
point sent to
The question is now:
Is
3remember, E
u0 N -torsion
was already (via
on
C/?
a double-cover of
branched over
p1 , p2 , p3 , p
232
21. THE PONCELET PROBLEM
Normal elliptic curves and a multiple addition theorem.
Put
uj := u0 + j ,
where
j C.
Abel's theorem implies
Proposition 21.2.1. There exists a
tion
only if
with order-N pole at
-periodic
meromorphic func-
and simple zeroes at
u1 , . . . , u N ,
if and
u1 + + uN 0 (mod ).
What we are really after here is the vector space
of meromorphic
E with at worst an order-N pole at O (and no other poles).
There are N 1 degrees of freedom coming from pushing around the
P
{uj } (while keeping
uj 0) and one degree of freedom from multiplying the function by a constant. So dim V = N ; let {f1 , f2 , . . . , fN }
(with f1 constant) be a basis, and de ne
functions on
1
N : E PN
[w1 ::wN ]
by
([1 : t : s] =:) z 7 [f1 (z) : : fN (z)].
Definition 21.2.2. The image of
mal elliptic curve of degree
f1 , f2 , f3
For
to be
N.
N ,
denoted
(Note that
E3 is
EN ,
is called a nor-
essentially
take
1, t, s.)
uj 0,
there exists a function
on
EN
with zeroes at
N (P(uj )), and order N pole at N (O). Now the hyperplane at inN
nity {w1 = 0} P 1 intersects EN only at N (O) (with multiplicity
N ). If written as the pullback to EN of a rational function, it follows
that F has denominator w1 ; the numerator must
then also be a ho
H(w)
1
. It follows that the
mogeneous linear form H SN , i.e. F =
w1
N (P(uj ))
all lie on
{H = 0} PN 1 ,
EN
and so
coords. of
N (P(uj ))
(21.2.4)
z }| {
0 = det[ fi (P (uj )) ].
|{z}
=:Fi
N (P(uj ))
H(w)
function
w1
Conversely if this is satis ed then the
{H = 0}; one writes down
PN
divisor
j=1 [uj ] N [0], and
the
lie on a hyperplane
and computes its
EN
concludes (by Abel) that
uj 0.
21.2. EXPLICIT SOLUTION OF THE PONCELET PROBLEM
We can push this computation further. Expand
233
Fi (u0 + j ) =
(N 1)
Fi (u0 ) + Fi0 (u0 )j + +
Fi
(u0 ) N 1
+ N
j ( ),
(N 1)! j
then apply multilinearity of the determinant to expand the RHS of
(21.2.4):
h
i
Y
(j1)
0 = const. (k ` )det Fi
(u0 ) +
terms of higher
homog. degree
k>`
Dividing by
uj u0 ),
{j }
in the
` ) and taking the limit as all j 0 (i.e.
k>` (k
all
this becomes
0 = det
(21.2.5)
(j1)
Fi
(u0 )
i = 1, . . . , N
j = 1, . . . , N
The determinant on the RHS of (21.2.5) is called the Wronskian of
P
N P . Notice that in the limit N
j=1 uj 0 becomes N u0 0; so
this last condition is equivalent to (21.2.5)!
Example 21.2.3. Here is what the above calculation (using mul-
tilinearity of the determinant) looks like for
degree higher than
F + F0
1
1 1
F2 + 1 F20
F
1
=
F2
N = 2,
{j }:
0
F1 + 2 F1 F1 + 1 F10
=
F2 + 2 F20 F2 + 1 F20
(2 1 )F10
= (2 1 )
0
(2 1 )F2
ignoring terms of
in the
(2
(2
1 )F10
1 )F20
F1 F10
.
F2 F20
Using the chain rule and again multilinearity of det, one nds
that the vanishing of the Wronskian is independent of the choice of
local coordinate on
E.
So we can replace
by
(and hence
by
f ),
which yields our multiple addition theorem:
u0 is N -torsion in C/ (and
th step) if and only if
the N
h
i
(j1)
det fi
(0) i = 1, . . . , N = 0.
Theorem 21.2.4.
tion closes up at
(21.2.6)
the Poncelet itera-
j = 1, . . . , N
4note:
higher homogeneous degree in the
{j }
means higher than
k>` (k ` )
234
21. THE PONCELET PROBLEM
(j1)
(0) probably requires explanation: rst, we are viewing f locally as a function of t (rather than
st derivative is (total derivof [1 : s : t] =: z on E ), and the (j 1)
ative) with respect to t. The 0 just means t is set to 0 at the end;
this is because we are evaluating at (i.e. u0 ), which has coordinates
[(s, t) =] (s0 , 0).
Remark 21.2.5. The meaning of
Application in the case
fi
odd. Obviously we can't compute
the Wronskian (21.2.6) unless we know the
Take
N = 2m + 1.
Then for
fi .
f1 , . . . , fm+1 ; fm+2 , . . . , f2m
we may
choose
1, t, . . . , tm ; s, st, . . . , stm1 .
These have order of pole at
0, 2, . . . , 2m; 3, 5, . . . , 2m + 1.
dj1 ti1
The determinant in (21.2.6) is then (using that
dtj1
= 0
unless
j = i)
Writing
1
0
0
.
..
.
.
.
0
m!
0
m+1 s
d
dtm+1
. .
.
.
.
.. . . ..
.
m+1
m1 )
d dt(st
m+1
p
s = s(t) = det(tMC + MD )
A0 = s0 ),
.
.
.
0
.
2m
d s
dt2m
.
.
.
d2m (stm1 )
dt2m
..
..
= A0 + A1 t + A2 t2 +
(here
this becomes a nonzero constant times
Am+1 A2m
..
.
..
.
.
.
.
A
Am+1
2
(21.2.7)
(2m + 1)-gon
(21.2.7) vanishes.
We conclude that there is a circuminscribed
a family of such) for the pair
C, D
i
Example 21.2.6. We work out the case
determinant
(21.2.7)
is just
A2 ,
N = 3,
i.e
(and hence
m = 1.
The
so we can get a Poncelet triangle
21.3. ELLIPTIC BILLARDS
A2 = 0.
Writing
Ti =
235
1
, calculate
ti
v
u 3
u Y
p
s = det(tMC + MD ) = t (t ti )
i=1
r
3
Y
t
Ti2 2
Ti
1 =C
t
=C
1 t
t
2
8
i
i=1
i=1
3
Y
A2
1X 2 1
=
T + (T1 T2 + T2 T3 + T1 T3 ) .
C
8 i=1 i
4
=
If
T1 = 1,
solving a quadratic equation we nd
A2 = 0
equation of
(1 + T )2
(1 T )2
, T3 =
for some T
4
4
4
4
2
E reads s = (t1) t
t
.
(1 + T )2
(1 T )2
T2 =
If we take
MD =
4
(1+T )2
4
(1T )2
1 0 0
MC = 0 1 0
0 0 1
0 ,
1
corresponding to
C=
then
= 1
4y 2
4x2
+
=1 ,
(1 + T )2 (1 T )2
D = x2 + y 2 = 1 ,
and indeed
det (tMC + MD ) = t
4
(1 + T )2
t
4
(1 T )2
(1 t) .
This recovers Example 1.3.2(b) from the beginning of the course! It's
easy to draw one triangle, but seems quite nontrivial that you get one
independent of the starting point.
21.3. Elliptic billards
Returning to the real world, let
CR R2
2
be an ellipse with foci
F1
F2 . (CR consists of all points in R , the sum of whose distances from
F1 and F2 is a xed constant.) We imagine that CR is the boundary of
a pool table (frictionless, of course!). A billiard trajectory for CR is a
sequence of pairs (xi , Li )i0 with xi , xi+1 C Li and where Li1 , Li
make equal angles with Txi CR
and
236
21. THE PONCELET PROBLEM
xi
Li1
Li
x i+1
x i1
i.e. one has equality of angles of incidence and re ection.
If
DR
is another conic (ellipse or hyperbola) then a (real) Poncelet
(CR , DR ) is a sequence of pairs (xi , Li )i0
Li tangent to DR .
trajectory for
C Li
and
Theorem 21.3.1. [L. Flatto, 2003] (a) Assume
5
with
CR .
with
DR
xi , xi+1
is confocal
Then the (real) Poncelet trajectories are billiard trajectories
with respect to
CR .
(b) Conversely, any billiard trajectory for
F1 or F2 and not along the minor
CR and some DR confocal with CR .
CR
not passing through
axis, is a Poncelet trajectory for
We will prove only (a); Flatto does (b) in Appendix E of his book
[L. Flatto, Poncelet's Theorem, AMS, 2009] (which is, by the way,
written for undergraduates).
At any rate, the two proofs are very
similar.
Remark 21.3.2. It's worth pointing out right away that given
(x0 , L0 ) (L0
not containing
F1
F2 and not the
with CR and tangent
or
minor axis), there
DR confocal
to L0 . If L0 passes
between F1 and F2 , DR is a hyperbola; otherwise, it's an ellipse. One
p
det(tMC + MD ) obtains infordetermines this DR , and then from
mation (as in 21.2) on whether the Poncelet trajectory closes up. By
is a unique conic
the Theorem, this is also the billiard trajectory! You'll use this to do
a computation in Exercise (2) below. But I should emphasize that if
you change
(x0 , L0 )
change the choice of
5that
is,
F1
and
F2
(i.e. the choice of billiard trajectory), you have to
DR
accordingly.
are the foci of
DR .
If
DR is a hyperbola, this just means that
F1 and F2 must remain constant.
the dierence of distances from its points to
21.3. ELLIPTIC BILLARDS
237
Proof. (of (a)): We begin with a general principle, for a conic
QR
with foci
F1 , F2 .
Given
p0 QR ,
let
L := Tp0 QR
F2
p0
F1
F2
QR
and denote by
F20
the re ection of
|pq|
for the segment and
F2
in
L.
p, q write pq
:= |F1 p0 | + |p0 F2 | and
Given points
for its length. Set
note that by de nition of ellipse,
|F1 q| + |qF2 | =
(q QR ).
p L\{p0 }, |F1 p| + |pF20 | = |F1 p| + |pF2 | visibly exceeds ,
0
that taking p = p0 minimizes |F1 p| + |pF2 |. It follows that
If
F1 p0 p0 F20 = F1 F20 .
(21.3.1)
Now let
meaning
CR , DR be confocal assume that DR is an ellipse.
Apply-
ing the principle that (21.3.1) holds for the above construction, leads
to a picture
F2
1+1
p
2
2
F
1
TpC R
A
F1
F
2
F2
CR
DR
in which the solid black lines are part of a Poncelet iteration and we
must show
1 = 2
(so that it is a billiard trajectory).
re ection in
238
21. THE PONCELET PROBLEM
Tp CR
(dotted black) is denoted by one prime, re ection in solid black
lines by two primes.
F1 A + AF2 = F1 B + BF2 ,
By de nition of ellipse,
which implies
|F100 F2 | = |F1 F200 |.
F100 pF2 and F1 pF200 are rotations
of each other (through p), so that + 21 = + 22 ( = 1 = 2 ).
It is obvious from the picture that 1 + 1 = 2 + 2 , and so we indeed
conclude that 1 = 2 .
From there it is clear that the triangles
Exercises
(1) Consider the pair of conics
C, D
from Exercise (2) of Chapter 1
once more but in the following form: write
1 0 0
MC = 0 1 0 ,
0 0 1
1 0 0
MD = 0 1 0
0 0 r2
and use these to de ne quadratic forms by e.g.
QC (X, Y, Z) =
So
QC = 0
X Y
1
0
0
X
Z 0 1 0 Y = X2 + Y 2 Z2
0 0 1
Z
de nes
and
QD = 0
Working in homogeneous coordinates
D
[V : T : U ],
de nes
P2 .
as conics in
de ne an elliptic
curve by
U 2 V = det (T MC + V MD ) .
In a ne coordinates, this is
u =
U
.
V
u2 = det(t MC + MD ),
where
t=
This is the general prescription for the elliptic curve
arising in the Poncelet construction, exactly as above.
have to show is that in the present situation, with
as given, the elliptic curve is singular.
MC
T
,
V
All you
and
MD
It's practically a one-line
problem, and you may use the a ne setup.
But now you are in
a position to see why the curve being singular should make the
Poncelet problem easier, and even to see why (from the abstract
perspective) the solution involved trig functions.
EXERCISES
(2) Let
a>b>0
239
and
CR =
x2 y 2
+ 2 =1 ;
a2
b
( a2 b2 , 0). We plan to shoot our pool ball vertically
along the line (L0 =) {x = c}, where 0 < c < a (and c 6=
a2 b2 ).
For what value of c does the resulting billiard trajectory yield a
triangle? [Hint: the conics confocal with CR are all of the form
n 2
o
y2
x
+ b2 = 1 . Also: while straightforward, this is not a 1a2
it has foci
line computation!]
CHAPTER 22
Periods of families of elliptic curves
2
E E P are simply elements of the
, where , are 1-cycles generating
The periods of an elliptic curve
period lattice
E = Z
1 (E) is some canonically chosen generator. (That
de ned over Q, then should be the restriction of a rational
2
dierential 1-form on P de ned over Q.) If E is taken to vary with
1
respect to a parameter t P , the periods give interesting multivalH1 (E, Z)
is, if E is
and
ued transcendental functions (e.g. hypergeometric functions) which are
related to modular forms.
In this Chapter we explore (via examples) two dierent approaches
to computing period functions of this sort the Euler integral
method and the Picard-Fuchs method.
The rst of these is just a
way of computing the integral using Laurent polynomials; the second
derives a homogeneous linear ordinary dierential equation satis ed by
the periods, which yields a recurrence relation for their power-series
coe cients. Actually, both methods yield power series at rst but one
can sometimes recognize what functions they are the power series of.
This may sound like complex function theory, but in fact the power
series coe cients (esp. when related to modular forms) can have arithmetic meaning, as we shall see in the next chapter; in the context of
mirror symmetry (one of several interfaces between algebraic geometry
and string theory), power series derived from periods are related to
counting curves on threefolds.
Sections
22.1
and
22.3
will have a bit of overlap with
22.1. Holomorphic 1-forms on a smooth cubic
F S3
18.1.
P2
de ne a smooth curve E = {F (Z0 , Z1 , Z2 )
(31)(32)
= 1, so that E is elliptic.
genus formula g =
2
= 0};
F = Z0 Z1 Z2 t(Z03 + Z13 + Z23 ),
2i
P1 \{0, 13 , 33 , 3 } (3 = e 3 ).
for any
Let
Example 22.1.1.
32
241
by the
242
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
For the a ne forms of the equation we shall use the following notation:
v
u
L1
y
L0
2
P
x
L2
x=
Z1
,
Z0
y=
Z2
, and equation is
Z0
P2 \L2 , the coordinates are u =
g(u, v) := Z13 F (Z0 , Z1 , Z2 );
Z0
,
Z2
v=
Z1
, and equation is
Z2
the third neighborhood is left to you;
P2 \L0 ,
f (x, y) :=
on
the coordinates are
1
F (Z0 , Z1 , Z2 );
Z03
on
P2 \L0 L2 , we have u =
1 x
3
f (x, y) = y g y , y .
on
1
,
y
v =
x
;
y
y =
1
,
u
x =
u
; and
v
on E by
dx
du
dv
dy
=
=
=
=
fy E\(L0 VT)
fx E\(L0 HT)
gv E\(L2 HT)
gu E\(L2 VT)
Now de ne a form
where
the notation
intersects
E\(L0 VT) means E minus those points where
L0 or has a vertical tangent line (similarly, HT
means horizontal tangent);
equality of any two dierentials above is meant in the sense of
where both are de ned;
for example: on
dx
fy
E , f = 0 = 0 = df = fx dx + fy dy =
= 0 and) fx , fy 6= 0;
dy
where (f
fx
the means that the third neighborhood stu is left to
you.
fx and fy
do not simultaneously vanish (E is smooth!), {E\(L0 VT)}{E\(L0
HT)} is all of E\L0 . So the 6 dierent domains of de nition glue to give
(E\L0 ) (E\L1 ) (E\L2 ), which is all of E . Morover, dx
fy
Now consider the domains of the rst two expressions: since
etc.
are all holomorphic where they are de ned.
1 (E).
E\(L0 VT)
We conclude that
22.2. PERIOD OF A FAMILY OF CUBIC CURVES (EULER INTEGRAL METHOD)
243
deg(()) = 2g 2 = 2 2 = 0, and so 's lack
0
1
of poles implies it has no zeroes either. Any other (E) has
0
O(E), and then by Liouville 0 is a constant multiple of . So
1 (E) has dimension 1, and spans it.
By Poincar-Hopf,
22.2. Period of a family of cubic curves (Euler integral
method)
Et
Now consider the Hesse family
of elliptic curves, already given
in Example 22.1.1, with a ne form
f (x, y) = xy t(x3 + y 3 + 1) = 0,
(For the four values of
t C.
excluded in the example,
not an elliptic curve. I won't write
ft
Et
is singular hence
because the subscript is reserved
here for partial derivatives.) An alternate form of the equation, valid
on
C C ,
is
3
x + y3 + 1
1t
= 0,
xy
|
{z
}
=:(x,y)
where
C[x, x1 , y, y 1 ].
holomorphic 1-forms
belongs to the ring of Laurent polynomials
From the last section, we have the family of
dx
1 (Et ).
t :=
fy Et
We can obtain a family of 1-cycles by noticing that
is empty for
|t| <
1
, since
3
|(x, y)| < 3
for
x, y
{|x| = |y| = 1} Et
in the unit circle.
Indeed,
t := {|x| = 1, |y| 1} Et
has this empty set as its boundary
t ; and so we would like to compute
the period
P (t) :=
t
t
as a function of
rank
2,
t,
on the open disk
1
. Now since
3
|t| <
there is a complementary 1-cycle
period
Q(t) :=
t .
t
on
Et ,
H1 (Et , Z)
has
and cooresponding
244
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
Noticing from the homogeneous form of the equation that
E0 = {Z0 Z1 Z2 =
0} is a union of 3 lines (
= P1 ), we can easily visualize what happens to
Et , t , and t as t tends to zero:
Et
E0
Z1 =0
Z0 =0
undergoes
degeneration
(thin black cycles are
pinched to points)
Z2 =0
t 0) to dx
on Z2 = 0, and 0 passes
x
through the poles of this form while 0 traverses the unit circle around
them, we infer that Q(t) as t 0 but P (t) 2i.
We now compute P (t) more precisely, by rst noting that the area
From the fact that
tends (as
integral
|x|=|y|=1
(since
dx dy
=
f (x, y)
dx df
fy f
df = fx dx + fy dy and dx dx = 0)
df (x, y)
1
=
dx
fy (x, y)
|x|=1
|y|=1 f (x, y)
(where inside the parentheses
is a xed constant).
Now thinking
f (x, y) = 0 for |t| small (and x xed with |x| = 1),
y + ay + b = 0 where a = xt is big and b = x3 + 1 is not. This
about the equation
we have
means that two of the roots are big and one is small in particular,
there is exactly one solution
y(x) with modulus less than 1.
Therefore,
by Cauchy's residue theorem, the integral above
2i
=
|x|=1
= 2i
1
fy (x, y(x))
dx
dx
= 2i
fy
|x| = 1
y = y(x)
So
1
P (t) =
2i
|x|=|y|=1
1
=
2i
dx dy
1
=
f (x, y)
2i
X
|x|=|y|=1 n0
t .
t
|x|=|y|=1
dx
x
dy
y
1 t(x, y)
tn n dlogx dlogy
22.2. PERIOD OF A FAMILY OF CUBIC CURVES (EULER INTEGRAL METHOD)
245
where dlogx
dx
and
x
means simply the
nth
power of
(x, y).
Using
Cauchy residue twice this
= 2i
tn n (0, 0)
n0
in which
n (0, 0) =: [n ]0 is the constant term of n = (x2 y 1 +x1 y 2 +
x1 y 1 )n .
We can make this more explicit. Given a product
(x2 y 1 + x1 y 2 + x1 y 1 ) (x2 y 1 + x1 y 2 + x1 y 1 )
|
{z
}
n times
each contribution to the constant term comes from exponents sum-
(0, 0) (i.e. multiplying to x0 y 0 ). But the only combinations of (2, 1), (1, 2), (1, 1) summming to (0, 0) are: m(2, 1) +
m(1, 2) + m(1, 1). Hence, the only possibility for a nozero constant term is to have n = 3m (i.e. 3|n), and the number of ways to
ming to
choose
is
2 1
x y from m factors
x1 y 2 from m factors
1 1
x y from m factors
3m
m,m,m
:=
(3m)!
. That is,
m!m!m!
P (t) = 2i
3m
m0
(3m)!
= 2i 2 F1
(m!)3
where by de nition (writing
1 2
, ; 1; (3t)3
3 3
(a)m := a(a + 1) (a + m 1)
for the
Pochhammer symbol)
2 F1 (a, b; c; u)
:= 1 +
X (a)m (b)m
um
(c)
m!
m
m1
is the Gauss hypergeometric function.
Notice that
P (t)
is really a function of
the symmetry in the family
Et :
Et E3 t
(x, y) 7 (3 x, y).
(3t3 ) = u.
This re ects
246
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
The Gauss hypergeometric function satis es a well-known ODE. In this
P0 (u) = P (t) and Q0 (u) = Q(t))
d2
2
d
u(1 u) 2 + (1 2u)
P0 (u) = 0.
du
du 9
case (writing
Q0 (the
log u
P (u)
other period), which turns out to have a term of the form
2i 0
It turns out that the ODE satis ed by
re ecting the fact that following
P0
must be satis ed by
around
t = 0
yields
t + 3t
in
H1 (Et , Z).
22.3. Cohomology of an elliptic curve
C, with basis {ek }nk=1 .
2
The second tensor power of V , written V V , is the n -dimensional
P
vector space consisting of nite sums
i vi wi (vi , wi V ) subject
to bilinearity (e.g., on the left (v + w) u = v u + w u); it
V2
n
has basis {ek e` }k,`=1 . The second exterior power
V consists of
P
nite sums
vi wi satisfying bilinearity and also v w = w v (so
that v v = 0); it may be viewed as a quotient- or sub-space of V V ,
n
and has basis {ek e` }1k<`n hence dimension
. In particular, if
2
V2
dim V = 2, then dim( V ) = 1; this is essentially the only case we
Let
be a nite-dimensional vector space over
shall use.
Dualizing the homology groups
H1 (E, Z) =
18.1,
from
Z hclosed
Z hboundaries
paths on
Ei
of regions in
Ei
we de ne cohomology groups (with complex coe cients)
by
H 1 (E, C) := Hom (H1 (E, Z), C) (
= C2 ).
Write
A0 (E)
for
functions and
A1 (E)
f dx + gdy , where
A (E). These are objects
latter locally of the form
have the
2-forms
C 1-forms on E , the
z = x + iy ; and nally we
for
locally of the form
i
i
z = Gd
z dz)
Gdx dy (= Gdy dx = Gdz d
2
2
with
smooth, which you may think of as a eld of in nitesimal area
elements.
bundle
1the
V2
In more sophisticated terms, they are C
sections of the
V
2
T E = pE
Tp E . (Refer to 13.1 for notation.)
homology class represented by a 1-cycle
is written
[]
22.3. COHOMOLOGY OF AN ELLIPTIC CURVE
247
The various degrees of forms are connected by exterior dierentiation
d : A0 (E) A1 (E)
sending
F 7 dF := Fx dx + Fy dy
F
F
=
dz +
d
z,
z
z
and
d : A1 (E) A2 (E)
f dx + gdy 7 df dx + dg dy
= (gx fy )dx dy.
The
1st
de Rham cohomology group of
1
HdR
(E, C) :=
ker(d) A1 (E)
=
d(A0 (E))
the class represented by a
1-form
Lemma 22.3.1. The map
[] 7 {[] 7
is then de ned by
closed
exact
is written
C forms
;
C forms
[].
1
: HdR
(E, C) H 1 (E, C)
given by
is well-de ned, and an isomorphism.
is closed ( = 0) and
= d = = 0. If
(d = 0) then
=
=
Proof. First we check well-de nedness: if
exact (
= d ), then ([]) = 0 since
is a boundary ( = ) and is closed
d = 0, so that ([]) is de ned on the
level of homology classes.
(The middle equality in both cases swapping
and
is Stokes's
theorem, a generalization of the fundamental theorem of calculus for
dierential forms.)
is injective, assume ([]) = 0, and let p be a point of
q
E . Then F(q) = p de nes a C function F on E . (The reason F
isn't multivalued is that two paths dier by a cycle , and
=0
by the assumption.) Now = dF by the fundamental theorem of
calculus, and so [] = 0.
Finally, write , for a basis of H1 (E, Z). Using the identi cation
=
1
H (E, C) C2 which evaluates a functional against this basis,
! a nice
way to think about the map is as sending [] 7
C2 .
To see that
2 [] 7
integrating
means the complex-linear functional on homology classes given by
over a representative 1-cycle
248
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
E C/E identi es with du. Rescaling
(by a complex constant)
so that
= 1, the 2-vector becomes
!
!
du
1
([du]) =
=
(where we may assume H), and we
du
Moreover, the Abel map
have the standard picture
C/
coord. u
Noting that
that
([d
u]) =
is surjective since
1(=u)
!
!
d
u
du
=
=
du
d
u
!
!
1
1
2
,
span C .
!
, we conclude
Remark 22.3.2. (a) Note that meromorphic 1-forms on a Riemann
d{f dz} = df dz =
f
we have used
= 0,
z
surface are always closed since locally (f mero.)
f
dz
z
dz
dz dz = dz dz = 0. (Here
which expresses the holomorphicity of f o its poles.) Using the
formula as above (i.e. 7 { 7
}), we can de ne a map
and
same
ker(Res) K1 (E)
H 1 (E, C)
d(K(E))
which also turns out to be an isomorphism.
(Here
ker(Res)
consists
of forms with no residues in particular, with no simple poles. This
doesn't mean they're holomorphic though!)
1-form cannot be d of a smooth function G or meromorphic function f . (Locally the integral of is a
holomorphic function, so in either case G or f would actually have to
be holomorphic, hence by Liouville constant, making zero.) So we
(b) A nonzero holomorphic
have a commuting diagram of injective homomorphisms
(22.3.1)
1 (E)
t
/ ker(d)A1 (E)
exact
NNN
NNN
NNN
NNN =
N'
ker(Res)K1 (E)
/ H 1 (E, C)
exact
22.4. DIFFERENTIATING COHOMOLOGY CLASSES
.In what follows
will all just be denoted
249
, which you should read
take the period vector associated to this 1-form.
22.4. Dierentiating cohomology classes
Given a family
t)
many
{Et }tP1
of elliptic curves (smooth but for nitely
t 1 (Et ),
!
P (t)
(t ) =:
.
Q(t)
with holomorphic forms
write
This assumes a choice (this unfortunately only works locally in
basis
t , t
for
H1 (Et , Z),
so that
P (t) =
t , Q(t) =
t .
t)
of
We can
dierentiate this period vector to obtain
P 0 (t)
Q0 (t)
!
,
t (considering the isomorphisms in (22.3.1)) is of some1
1
1
thing in ker(d) A (Et ) or ker(Res) K (E) (but not (Et )). We
0
will use the latter, and we write t for a family of residue-free mero!
0
P
(t)
0
. The point is that by
morphic 1-forms satisfying (t ) =
Q0 (t)
which for each
dierentiating families of cohomology classes you get a new family of
cohomology classes.
Example 22.4.1. Consider the Legendre family
Et P2
given by
the projective closure of
y 2 = x(x 1)(x t),
with holomorphic 1-form family
dx
t =
1 (Et ).
y Et
We have
(t ) =
where
t , t
P (t)
Q(t)
!
=
dx
t x(x1)(xt)
,
t t
=
dx
t t
t
x(x1)(xt)
are the 1-cycles exhibited in the schmatic picture
250
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
Et
1
"2 sheets over P
t
or the topological picture
" x(x1)(xt) "
"+ x(x1)(xt) "
t
t
which shows the two sheets (each is a
be
with slits from
to
and
2.3).
From the latter picture, it is clear that for t small we may take t to
stationary on its sheet as t moves, and the two pieces of t on the
to
P1
being glued together to give
dierent sheets not to change either.
Et
(cf.
Therefore we may dierentiate
the above integrals under the integral sign (by
P 0 (t)
Q0 (t)
t
=
d
) to obtain
dt
1
dx
2
(xt)
x(x1)(xt)
1
dx
2
t (xt)
x(x1)(xt)
22.5. THE PICARD-FUCHS EQUATION
and
00
P (t)
Q00 (t)
obviously the rst is
t
=
1/2 dx
|
xt y Et
x(x1)(xt)
3
dx
4
t (xt)2
4 dx
2
(xt)
251
x(x1)(xt)
and the second
3/4 dx
|
(xt)2 y Et
, and
so we have
t0
dx
=
,
(x t) y Et
1/2
These both belong to
(t, 0)
t00
dx
=
.
(x t)2 y Et
ker(Res) K1 (Et ),
3/4
since their only poles are at
(orders 2 and 4 resp.) and the sum of the residues of a meromor-
phic form must always be zero.
(t ), (t0 ), and (t00 ) must be linearly dependent in C2 !
[t ], [t0 ], and [t00 ] are linearly dependent in K1 (Et ) modulo
Of course,
Therefore,
d(K(Et )),
i.e. as cohomology classes.
22.5. The Picard-Fuchs equation
Since what is being dierentiated in the last section is really cohomology classes (via the identi cation with
C2 ),
it makes sense to
write
Dt [t ] = [t0 ] ,
Dt2 [t ] = [t00 ].
With this notation, the linear dependence observation above implies
an ODE of the form
A(t)Dt2 + B(t)Dt + C(t) () = 0
|
{z
}
(22.5.1)
=:DPF
satis ed by
[t ]
(as a varying cohomology class) hence by
P (t)
and
Q(t)!
However, to nd
A, B,
and
C,
we have to compute. We start by
dierentiating a meromorphic function
2y
4ydx
2dy
d
=
+
,
(x t)2 Et
(x t)3 Et (x t)2 Et
|
{z
}
K(E)
which using
y 2 = x(x 1)(x t) = dy = 3x 2(1+t)x+t
dx
2y
4y 2
3x2 2(1 + t)x + t
=
+
dx
(x t)3 y
(x t)2 y
Et
becomes
252
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
and using
y 2 = x(x 1)(x t)
again
x2 + (2 2t)x + t
=
dx
(x t)2 y
Et
(x t)2 2tx + t2 + (2 2t)x + t dx
=
(x t)2
y Et
(2 4t)x + t2 + t dx
= t +
(x t)2
y
Et
= = t + 4(1
2t)t0
+ 4t(1 t)t00 .
d of a meromorphic function, hence (has both
1
trivial in H (Et , C). We conclude that (dividing
So this last expression is
0 and) is
by 4 to simplify)
its periods
through
DPF = t(t 1)Dt2 + (2t 1)Dt +
kills
[t ].
1
4
From ODE theory, the associated indicial equation is
C(t)
B(t)
t r + lim
= r2
r(r 1) + lim
t0 A(t)
t0 A(t)
which has a double root, implying one holomorphic solution (unique
up to scale) and one logarithmic solution near
t = 0.
22.6. Computation of a period (Picard-Fuchs method)
First let's compute its limit
t = lim
t0
t0
t
| {z }
lim
dx
p
.
x(x 1)(x t)
P (t)
P1 ) above, this
dx
dx
=
= 2i Res0
x x1
x x1
Referring to the picture of
(on the slit
1
= 2i
= 2,
1
P
P (t) must be the holomorphic solution. Write P (t) = 2 an tn ,
a0 = 1, and apply DPF :
X
0 = DPF
an tn
X
1
=
t(t 1)(n + 2)(n + 1)an+2 + (2t 1)(n + 1)an+1 + an tn
4
n0
and so
22.6. COMPUTATION OF A PERIOD (PICARD-FUCHS METHOD)
where we have shifted indices after dierentiating.
with like powers of
X
n0
t,
253
Collecting terms
this
X
1
an (n + 1)an+1 tn +
[2(n + 1)an+1 (n + 1)(n + 2)an+2 ] tn+1
4
n0
X
(n + 1)(n + 2)an+2 tn+2 .
n0
Shifting indices once more, we have
X 1
n0
an (n + 1)an+1 + 2nan n(n + 1)an+1 + n(n 1)an tn
"
#
2
X
1
=
n+
an (n + 1)2 an+1 tn .
2
n0
Since this power series is zero, we get a recurrence relation for the
coe cients of
P (t):
an+1 =
n + 12
n+1
2
an ,
so that
an
1/2 3/2
1/2 3/2 (1/2 n + 1) 2
1/2 + n 1 2
= a0
=
|{z} 1 2
n
n!
=1
2
1/2
=
and
P (t) = 2
X 1/22
n
n0
Again, the situation as
t0
tn .
looks like
0
t
t
In the next chapter the formula for
rational points on cubics over
Fp .
P (t)
will be connected to counting
254
22. PERIODS OF FAMILIES OF ELLIPTIC CURVES
Exercises
(1) Check that 2 F1
1 2
, ; 1; (3t)3
3 3
m0
t3m
(3m)!
by writing out
(m!)3
the Pochhammer symbols.
(2) Show that the curves
Et := {Z0 Z1 W0 W1 t(Z1 Z0 )2 (W1 W0 )2 =
0} P1Z0 :Z1 P1W0 :W1 are in fact elliptic (except at those nitely
many t which ones? for which they are singular). You could
1
do this by projecting to the rst P and using Riemann-Hurwitz to
compute the genus. (To use R-H in this way, rst nd all vertical
tangents places on the curve where the partials with respect to
W0
and
W1
vanish.)
(3) Writing the family of curves from the last exercise in a ne form,
xy t(x 1)2 (y 1)2 = 0 (or 1 t(x, y) = 0), de ne a family
of loops t H1 (Et , Z) for small t, and a family of holomorphic
1
forms t (Et ), exactly as in the text. Compute the period
P (t) := t t as a power series, using the computation done above
as a model.
in Remark 22.3.2 is well-de ned and an isomorphism.
P
log t
Find Q(t) in the Legendre example by plugging
P (t)+ n1 bn tn
2
(4) Check that
(5)
into the Picard-Fuchs equation.
CHAPTER 23
Counting
Fp -rational
points on elliptic curves
In this nal chapter on elliptic curves, we take a brief dip into something much more arithmetic, counting the number (mod
in
P (Fp )
p) of solutions
to the equations for the Hesse and Legendre cubics from the
last chapter. These cubics still depend on
t,
which is taken to be an
integer now (rather than a complex number) so that we can reduce
p.
modulo
In a truly bizarre twist, the number of points (over
Fp )
in
each case is given by nearly the same power series as the holomorphic
period on the corresponding complex family of elliptic curves from Ch.
22. We brie y explain one abstract way, due to Y. Manin, to understand this connection between arithmetic and transcendental algebraic
geometry.
23.1. Sum formulas
Fp the eld with p elements (i.e. Z/pZ
viewed as a ring). Equality in Fp will generally be denoted by =, not
. (We will use the latter for counting points mod p.)
Let
be an odd prime, and
(p)
Lemma 23.1.1. For
k Z,
(
xk =
xFp (or Fp )
in
0, p 1 - k
1, p 1 | k
Fp .
y Fp , the assignment x 7 yx
groups Fp Fp . Therefore
X
X
X
yk
xk =
(xy)k =
xk .
Proof. Given
phism of additive
(23.1.1)
xFp
1The
xFp
yields an isomor-
xFp
interested reader is urged to study the theory of modular forms(!) and look
at papers by Ahlgren, Ono, Papanikolas et al in particular. Modular forms are, on
the one hand, essentially given by periods on families of (self-products of ) elliptic
curves; on the other hand, they are frequently constructed via point-counts over
nite elds.
255
256
23. COUNTING
Now,
Fp
Fp -RATIONAL
p 1,
is a cyclic (multiplicative) group of order
(Fp )k = {1}
Provided
POINTS ON ELLIPTIC CURVES
p 1 - k,
then, there exists
and so
p 1 | k.
y Fp
with
y k 6= 1.
By (23.1.1),
we have
0 = (y k 1)
xk
xFp
which implies (dividing by
y 1)
X
0=
xk .
xFp
p 1 | k , then xk = 1 for all x Fp
X
X
xk =
1 = p 1 = 1.
On the other hand, if
and so
xFp
xFp
Lemma 23.1.2. For
x Fp ,
p1
X
(
xk =
k=1
Proof. If
0, x 6= 1
.
1, x = 1
x = 1 then the sum is 1| + {z
+ 1} = p1 = 1.
If
x 6= 1
p1 times
then
(1 x)
| {z }
6=0
p1
X
x =
k=1
p1
X
k=1
= xx
p
X
xk
k=2
p
= x(1 xp1 )
which (since
Fp
is cyclic of order
p 1)
= x(1 1) = 0
Lemma 23.1.3. Let
Fp .
F2p
Proof. Since
over, if
Fp
Then
p1
2
1,
(p)
p1
2
and
1.
(p)
p1
p 1, ( 2 )2 = p1 = 1. Morep1
cannot have a 2 = 1 (Fp would then
is cyclic of order
is a generator then we
23.2.
Fp -POINTS
ON THE LEGENDRE ELLIPTIC CURVE
p1
, a contradiction). Hence
2
have order
x 7 x
p1
2
257
yields a surjective
homomorphism of multiplicative groups
Fp {+1, 1},
1
|F | = p1
.
2 p
2
p1
1. As 2 elements of
whose kernel necessarily has order
square, then
exhaust the
p1
2
= p1 =
kernel of and
23.2.
Fp -points
Now if
Fp
= 2
is a
are squares, these
the non-square elements go to
1.
on the Legendre elliptic curve
Consider once again the Legendre family of cubics
Et = {Y 2 Z = X(X Z)(X tZ)},
t Z.
Et (Fp ) Et (Fp ),
p we can look at the
solutions
i.e. with X, Y, Z in Fp resp. its algebraic
.
closure; there is a clear analogy to Et (Q) Et (Q)
We are going to compute the number of points |Et (Fp )| modulo p,
i.e. in Fp . (Computing the number in Z is a much harder problem.)
but this time with
After reducing mod
First we claim that
|Et (Fp )| 1 +
(23.2.1)
(p)
o
Xn
p1
1 + [x(x 1)(x t)] 2 .
xFp
[0 : 0 : 1] at ; the rest
of the curve is described by y = x(x 1)(x t). By Lemma 23.1.3, the
quantity in curly brackets yields (mod p) 2 if x(x1)(xt) is a square,
1 if x(x 1)(x t) = 0, and 0 if x(x 1)(x t) is not a square. This
2
exactly counts pairs (x, y) Fp solving the a ne equation, con rming
The leading 1 on the RHS counts the point
(23.2.1).
P
Fp ) xFp 1 = 0, so the RHS of
X p1
p1
p1
1+
x 2 (x 1) 2 (x t) 2
Now (summing in
(23.2.1) is
xFp
p1
2 p1
X
p1
2 p1
X
p1
p1
p1
2
2
= 1+
x 2
x 2 ` (t)`
xk (1) 2 k
`
k
xFp
`=0
k=0
p1
p1
2
2
p1
p1
X
X
X
p1
p1
`
`
k
k
2
2
2
= 1+
x
x (t)
x (1)
.
`
k
xF
`=0
k=0
X
258
23. COUNTING
Fp -RATIONAL
The sum here can be rewritten
POINTS ON ELLIPTIC CURVES
xFp
xp1 G(x), where G(x) =
P p1
2
m= p1
2
a0 , and so the above
p1
p1
2
2
X
X
p1/2
p1/2
p1
= 1
x` (t)`
xk (1) 2 k
`
k
`=0
k=0
By Lemma 23.1.1, this is just
p1
2
= 1
(t) (1)
p1
`
2
= 1 (1)
p1
2
X
p1/2 2
`=0
= 1 + (1)
2
p1/2
`=0
p+1
2
p+1
2
t`
X 1/2
t`
`
`0
=: P (t).
For the last step, we use the de nition (in
1/2
1/2
:=
Fp )
3/2 (1/2 ` + 1)
,
`!
which is evidently
1p1+2
2
p
2
to be
1
0 when p > ` > p1
(since
p+1
+1 =
2
2
2
p1
p1
2
equals
for 0 `
(and is de ned
2
`
= 0), and
` p). We conclude:
for
P (t) counts (mod p) the Fp -rational points of
Proposition 23.2.1.
Et .
Notice that
(23.2.2)
P (t)
23.3.
is a mod
p
version of the period
Fp -rational
For this section, take
P (t)
from
22.6!
points on the Hesse cubic
to be an odd prime with
p 1
mod
3.
I can't resist doing the same exercise for the other main example
from the last chapter, namely
Et = {XY Z = t(X 3 + Y 3 + Z 3 )}
where again we assume
t Z.
This has a ne form
xy = t(x3 + y 3 + 1)
am x m .
23.3.
Fp -RATIONAL
POINTS ON THE HESSE CUBIC
259
and toric form
1 = t(x2 y 1 + x1 y 2 + x1 y 1 ),
|
{z
}
=:(x,y)
where the Laurent polynomial
(x, y)
is de ned for
(x, y) (Fp )2 .
Using the toric form and Lemma 23.1.2, it is easy to compute the
Fp -points
Et := Et (Fp ) (Fp )2 .
Namely, in
Fp
(i.e. mod
|Et |
p)
X
p1
X
tk ((x, y))k ,
(x,y)(Fp )2 k=1
the point being (besides the Lemma) that
exactly those
(x, y)
on
Et .
t(x, y)
is
(in
Fp )
for
Switching the order of summation this
becomes
(23.3.1)
p1
X
tk
(x, y)k .
(x,y)(Fp )2
k=1
Now by Lemma 23.1.1
xi y j =
xi
xFp
(x,y)(Fp )2
For
yj =
yFp
1, p 1 | i, j
.
0, otherwise
k [1, p 2],
(
((x, y))k = []0 +
Our assumption on
x, y
p1
terms with powers of
not both divisible by
implies that
3 - p 1,
)
.
and so
((x, y))p1 = [p1 ]0 + x2(p1) y (p1) + x(p1) y 2(p1) + x(p1) y (p1)
(
)
terms with powers of x, y
+
not both divisible by p 1
(in particular, there are no
xp1
or
y p1
x(p1) , y (p1) , xp1 y (p1) , x(p1) y p1 ,
terms). So (23.3.1) becomes
p1
X
tk [k ]0 tp1 3.
k=1
Recall from
22.2
that
[k ]0 =
3m
m,m,m
if
k = 3m
(and
if
3 - k ).
260
23. COUNTING
Fp -RATIONAL
POINTS ON ELLIPTIC CURVES
On the other hand, looking along the coordinate axes
Z=0
X = 0, Y = 0,
we get (only) the points
[1 : 1 : 0] , [0 : 1 : 1] , [1 : 0 : 1]
in
Et (Fp ).
and so may assume
only
in
Z=0
For example, along
Y = 1;
(on
Et )
we must have
then the equation is
X, Y 6= 0
X + 1 = 0.
This has
X = 1 as solution: otherwise we would have an element of order
Fp , so 6 | p 1, contradicting our assumption on p.
We conclude that
p1
|Et (Fp )| 3(1 t
(p)
p1
bX
3 c
m=1
again very reminiscent of
P (t)
3m
t3m ,
m, m, m
from before!
23.4. Deep reasons for (23.2.2)
The issue is this:
counting the point at
in
23.2,
why on earth does
|Et (Fp )| 1
(not
appear to solve the Picard-Fuchs equation
p+1
t(t 1)Dt2 + (2t 1)Dt +
where P (t) is the solution
2
() = 0 ? Indeed, P (t) 1 = (1)2 P (t),
from 22.6!! The two computations were
1
4
quite elementary, after all, so maybe there's an elementary explanation
for their equivalence?
Nope.
This is dealt with in [Clemens, A scrapbook of complex
curve theory, pp.65-69] and I'll just give a hint of the avor here. It involves an algebro-geometric version of the Lefschetz trace formula (the
formula from topology for the number of xedmm points of a mapping),
the Riemann-Roch theorem, Serre duality, and abstract sheaf theory.
Not elementary, but we can give a brief summary (with no claim to
total accuracy).
Consider
points, and
Et over Fp , t Fp . Then writing FP for number
f robp for the map [Z : X : Y ] 7 [Z p : X p : Y p ],
|Et (Fp )| = FP f robp : Et (Fp ) Et (Fp ) .
This should make sense to you because as an automorphism of
pth -power
(Frobenius) map xes exactly the elements of
Lefschetz-type theorem, it turns out that this
n
o
= 1 trace f rob |H 1 (Et /Fp ,O)
Fp .
of xed
Fp ,
the
By the
23.4. DEEP REASONS FOR (23.2.2)
261
H 1 is Cech cohomology computed with respect to the Zariski
topology, O is the sheaf of regular functions, and f rob is the action
by pullback (under f rob) on cohomology classes. It is a 1-dimensional
where the
vector space, with generator represented by a certain rational function
h with two simple poles, at q = [1 : 0 : 0] and some other point
p Et (Fp ). More precisely, H 1 (Et /Fp , O) is isomorphic to the space of
rational functions on Et with poles allowed only at p and q modulo the
subspace of rational functions with poles allowed at either p or q (not
both). [That this space is 1-dimensional in the more familiar complex
case is an exercise below.] You should also note that pulling back by
f robp
stabilizes the vector space we have just described, since (as
are taken to be in
f robp .
Et (Fp )
rather than
Et (Fp )) p
and
p, q
are xed under
So the displayed expression at least makes sense.
P
h = y1 + `0 b` y ` about
q ,2 and also expand a generator t 1 (Et /Fp ) (regular dierentials) as
P
[1 + k1 ak (t)y k ]dy . Recall also from the complex case, that residues
h
Next, we expand
in formal power series
of meromorphic functions require, and depend on, a choice of local
coordinate; while residues of meromorphic 1-forms are invariant (i.e.
require no such choice, as they can already be integrated around a loop
without appending a dz ).
take residue by computing
So for functions
Resq (F );
if
with a pole at
q,
we
has no other pole, then (as
residues sum to zero) the residue has to be zero.
f rob above, we have f rob h(=
h f rob) = y1p + `0 b` y `p = h + f + g . (The last equality, in which f
has only a pole at q and g has only a pole at p, is by 1-dimensionality
1
of H (Et , O) and the explicit description we gave of it. In that vector
space, this reads f rob [h] = [h].) Moreover, Resq (ht ) = 1 while
Now, writing
for the trace of
= Resq (ht )+Resq (f t )+Resq (gt ) = Resq ((f rob h)t ) = ap1 (t),
with the last equality obtained by multiplying out the explicit expressions for
f rob h
and
t .
So we end up with
|Et (Fp )| = 1 ap1 (t),
2note
that
gives a local coordinate about
[1 : 0 : 0]
on
Et ; x
does not.
262
23. COUNTING
Fp -RATIONAL
and (like the periods of
tion because
[t ]
does.
POINTS ON ELLIPTIC CURVES
t ) ap1 (t) must satisfy the Picard-Fuchs equaAgain, the regular solution of DPF () = 0 is
unique up to scale, and from there we are essentially done.
Exercises
(1) Check that
Et (Fp )
2
E P
p, q E(C)
(2) Let
is closed under
f robp ,
for
t Fp .
be a smooth cubic over the complex numbers, and
two distinct points.
meromorphic functions on
Let
be the vector space of
with poles only at
and
q,
with sub-
Wp and Wq (the meromorphic functions with poles only at
p and q respectively). Using Abel's theorem, prove that the dimension of V /(Wp + Wq ) is one. [Hint: you will also need to use
1
the fact that (E) has no zeroes, and that the residues of
F V given by Resp (F ) and Resq (F ) must sum to zero (cf.
spaces
Prop. 13.1.6(b)).]
Part 4
Curves of higher genus
CHAPTER 24
The algebraicity of global analytic objects
To kick o the last part of this course, on curves of higher genus,
this Chapter will demonstrate two approaches to the following result:
meromorphic (or holomorphic) functions and forms on normalizations
of algebraic curves, all arise as pullbacks of functions and forms on
projective space constructed from rational functions (quotients of homogeneous polynomials) and their dierentials.
This is a special in-
stance of Serre's GAGA principle (global analytic is global algebraic
in the projective setting), and is proved (in
24.1)
using techniques
from Chapter 8 together with the primitive element theorem.
For holomorphic forms, we would like a more precise result (already
hinted at in Remark 20.1.1) on how to think of the holomorphic forms
on a normalization rationally. It is important at this point to recall
part (B) of the Normalization Theorem 3.2.1, which says that every
Riemann surface can be obtained as the normalization of an algebraic
curve in
P2 ,
even one with only ordinary double point (ODP) singu-
larities. So in the course of analyzing curves with ODP's in
24.2
we
will actually have proved (cf. Prop. 24.2.1(c)) that for any Riemann
surface
of genus
g , dim(1 (M )) = g .
Featuring prominently in this
section is the space of homogeneous polynomials vanishing at the set
of ODP's, which will play a key role in the proof of Riemann-Roch in
the next Chapter.
24.1. Chow's theorem for algebraic curves
Let
C P2
be an irreducible projective algebraic curve of degree
d; applying a projective transformation if necessary we have [0 : 0 :
1]
/ C . Start by normalizing C ; that is, express it as the image of
e (P2 \{[0 : 0 : 1]}). One may evidently produce
a morphism : C
1What I haven't proved here, is that a global analytic subvariety of projective space
is an algebraic subvariety; but I will use this at a couple of points in later chapters
(apologies).
265
266
24. THE ALGEBRAICITY OF GLOBAL ANALYTIC OBJECTS
e, by pulling back the
C
Y
2
x= X
, y =
on P ; and
Z
Z
meromorphic functions on the Riemann surface
rational functions
C(x, y)
under
C(x, y) consists of all
in X, Y, Z of the same degree.)
the eld
(Recall
quotients of homogeneous polynomials
C(x, y)C for the
subring of rational functions whose polar set does not contain C , de ne
e by
the eld of rational functions on C
More precisely, writing
e := C(x, y)C .
C(C)
Next we consider the projection
: P2 \{[0 : 0 : 1]} P1
[Z : X : Y ] 7 [Z : X],
e as a
:= with the normalization presents C
d-sheeted2 branched cover of P1 . Write B ( P1 ) for the branch locus,
1
and for a path containing B with P \ simply connected (cf. 8.2).
whose composition
We have inclusions
e K(C)
C(x) C(C)
(24.1.1)
rat'l
mero.
fcns.
fcns.
where the rst is obtained by noting
C(x) C(x, y)C
and
C(x) =
C(x) C(x, y)C .
Now, one might initially speculate that the right-hand inclusion of
(24.1.1) is proper when
has singularities such as ODP's, since: (a)
of a function would seem to
e, (b) at rst glance the pullback
p, q C
have the same value at p and q , and (c)
e
C
ought to be able to take dierent values
an ODP has 2 preimage points
meromorphic functions on
at distinct points, right? The weak link in this chain of reasoning is
(b), as you can see from the following
C = {Y 2 Z = X 2 (X Z)}
[1 : 0 : 0]. The pullback of xy
Example 24.1.1.
at its ODP
values
and
has tangent lies
to
e
C
Y =
therefore takes
(resp.) at the 2 points lying over the ODP.
The point is that rational functions are not well-de ned at all points
of
P2 , and this can be used to our advantage to get more
functions on
singular curves. So it becomes plausible that the right-hand inclusion
2the
mapping degree
deg(
) = d (= deg(C))
by Bzout
24.1. CHOW'S THEOREM FOR ALGEBRAIC CURVES
267
of (24.1.1) is an equality, and that is exactly what we shall prove in the
rest of the section.
To that end, let
and denote by
e
K(C)
be a nonzero meromorphic function,
the set of poles of
Writing
0 = f (x, y) = y d + a1 (x)y d1 + + ad (x)
(24.1.2)
for the a ne equation of
C,
we have as in
8.2
distinct solutions
{yj (x)}dj=1 to
f (x, ) = 0 over P \, which are interchanged as one
passes through \B. Moreover, by irreducibility of C (hence f ), (24.1.2)
is the minimal polynomial of y , proving that
e :
[C(C)
C(x)] d.
(24.1.3)
x P1 \(
(P )), one can think of (x, yj (x)) as belonge with
ing to C
(x, yj (x)) = x. Consider the elementary symmetric
polynomials (i = 0, . . . , d, with e0 = 1)
For each
ei (x) := ei ((x, y1 (x)), . . . , (x, yd (x))) ,
P1 \(B
(P )). As in 8.2,
the fact that they are bounded away from
(P ) guarantees (by Rie1
mann) their extension to holomorphic functions on P \
(P ). Further,
1
if x0
(P ) has neighborhood x0 P , then for k N su ciently
e. By the
large,
:=
((x x0 )k ) is holomorphic in
1 (x0 ) C
which are well-de ned and holomorphic on
8.2), ei (x) extends holomorphically across x0 ;
x0 )ik ei (x), ei (x) extends meromorphically
this argument at all points of
(P ), we nd that
same argument (from
but since ei (x) = (x
across
x0 .
Repeating
ei K(P1 );
and by Theorem 3.1.5(a)
Next observe that for
0 =
K(P1 )
= C(x).
1
any x P \(
(P ))
d
Y
and
j {1, . . . , d},
((x, yj (x)) (x, yi (x)))
i=1
= (x, yj (x))d e1 (x)(x, yj (x))d1 +e2 (x)(x, yj (x))d2 +(1)d ed (x) ;
3as
in
8.2
we may change projective coordinates if necessary to put the equation
in this form
268
24. THE ALGEBRAICITY OF GLOBAL ANALYTIC OBJECTS
that is, for a dense subset of points
0 =
d
X
e, (p)
pC
satis es the equation
(1)i ei (
(p)) (p)di .
i=0
Therefore the meromorphic function
itself satis es
d
X
0 =
(1)i (
ei ) di ,
(24.1.4)
i=0
with coe cients in
C(x).
Finally, recall the primitive element theorem, which says that an
algebraic eld extension (of degree
ment (of degree
e).
e)
is generated by a single ele-
(A transcendental eld extension, likewise, has
a transcendental element, which satis es no algebraic equation.) Were
e :
[K(C)
C(x)] > d, there would thus
e was arbitrary, (24.1.4)
but as K(C)
be an element of degree
> d;
shows this is not so. hence
e :
[K(C)
C(x)] d.
Putting this together with (24.1.1) and (24.1.3), we see that
e = C(C),
e
K(C)
proving the
Theorem 24.1.2. Every meromorphic function on the normaliza-
tion of an irreducible projective algebraic curve is rational, i.e.
the
pullback of a ratio of homogeneous polynomials.
Corollary 24.1.3. Every meromorphic 1-form on a normalization
is rational (i.e.
f dg
f, g
where
are rational).
e , and let 0 K1 (C)
e
(dx) =: K1 (C)
0
e. Then belongs to K(C)
e ,
be any other meromorphic 1-form on C
hence is rational by Theorem 24.1.2.
Proof. Consider (say)
24.2. Cohomology of a Riemann surface
Let
be a Riemann surface of genus
the 1st homology group
rank
H (M, Z) =
g.
Recall from
20.1
that
closed loops
boundaries is an abelian group of
2g , and de ne M 's 1st cohomology group
to be the
vector space of complex-linear functionals
H 1 (M, C) := Hom(H1 (M, Z), C).
2g -dimensional
24.2. COHOMOLOGY OF A RIEMANN SURFACE
269
22.3 (for elliptic curves) we have the de Rham cohomology
Exactly as in
groups
d
1
(M )
HdR
:=
To any closed
ker{A1 (M ) A2 (M )}
d
image{A0 (M )
1-form
A1 (M )}
we may assign the
C 1-forms
.
exact C 1-forms
functional 7
closed
on
loops. By the rst 2 paragraphs of the proof of Lemma 22.3.1 (which
work for any
M ),
this induces a well-de ned injective map
1
HdR
(M ) , H 1 (M, C).
(24.2.1)
Surjectivity also holds but will require a little more work than for elliptic curves.
Writing
1 (M ) for the space of anti-holomorphic
forms (the com-
plex conjugates of holomorphic ones), we can embed
1
1 (M ) 1 (M ) , HdR
(M )
(24.2.2)
via
(, ) 7 [ + ].
d(1 (M )) = 0 = d(1 (M ))
injectivity, suppose + = df ,
The map (24.2.1) is well-de ned because
(cf.
Remark 22.3.2(a)).
f A0 (M ).
To prove
Then
d(f ) = f d + df = ( + ) =
|{z}
=0
looks locally like a function times dzdz(= 0). Now breaking
M up into triangular regions i with local holomorphic coordinates
zi = xi + 1yi ,
X
X
=
gi dzi gi dzi = 2 1
|gi |2 dxi dyi .
M
i
i
| i
{z
}
since
R0
Since each integral
= 0 gi 0, we have
= 0 0.
M
But using Stokes's theorem and
=
M
M = ,
d(f ) =
M
f = 0
M
270
24. THE ALGEBRAICITY OF GLOBAL ANALYTIC OBJECTS
df = = f
= 0 = f O(M ) = f
z
constant (by Liouville) = = 0. So (24.2.1) is injective.
d3
By now you are quite familiar with the fact that dim(S3
) =
(d1)(d2)
(d3)+2
2
=
. If S is a set of points in P , then the homo2
2
geneous polynomials of degree d 3 vanishing on each of these points
are subject to (possibly dependent) linear conditions. Denoting the
d3
space of such polynomials by S3
(S), we therefore have
which implies
0.
So
(d 1)(d 2)
.
2
dim(S3d3 (S))
(24.2.3)
: M P2 is injective o a
algebraic curve C = {F (Z, X, Y ) = 0}
Now assume
image an
nite point set, with
of degree
d (F S3d ),
having only ODP singularities (as in part (B) of the Normalization
Theorem). Write
for the collection of these ODP's, and note
e.
M =C
By (24.2.3) and the genus formula,
dim(S3d3 (S)) g.
4 we have a map
By Remark 20.1.1,
S3d3 (S) 1 (M )
given by
G 7
g dx
fy
g(x, y) = G(1, x, y) etc. This is necessarily injective: were gdx
to
fy
vanish on C , we would have G 0 on C ; since F is irreducible then
F |G by Study, which is impossible (unless G is trivial) as deg(G) =
d 3 < d = deg(F ).
where
All told, we have a sequence of injective maps of complex vector
spaces
1
(M ) , H 1 (M, C).
S3d3 (S) S3d3 (S) , 1 (M ) 1 (M ) , HdR
Notice that the left-hand side has dimension
side has dimension exactly
2g .
2g
and the right-hand
All the injections are therefore isomor-
phisms and we conclude:
Proposition 24.2.1. For a Riemann surface
malizing an algebraic curve of degree
4see
(of genus
g)
nor-
with ordinary double points
Gri ths's book Introduction to algebraic curves for more details
24.2. COHOMOLOGY OF A RIEMANN SURFACE
271
S = {p1 , . . . , p },
the holomorphic forms are all pullbacks of rational
gdx
forms of the form f (as described above). Moreover, we have:
y
1
1
(a) [de Rham theorem] H (M, C)
(M );
= HdR
1
1
(b) [Hodge decomposition] HdR (M ) = (M ) 1 (M ); and
(d1)(d2)
1
= dim(S3d3 (S)).
(c) dim (M ) = g =
2
We also get an application to the period matrices
20.1.
1 , . . . , 2g
Recall that if
a basis for
(M ),
is a basis for
then
1 1
.
.
.
..
..
1 g
H1 (M, Z)
2g 1
.
.
.
2g
Proposition 24.2.2. Viewed as vectors in
of
are
R-linearly
described in
and
1 , . . . , g
R2g (
= Cg ),
the columns
independent.
Proof. Suppose otherwise, i.e. that there exists a nonzero vector
a R2g
satisfying
0 = a ;
then we have also (by complex conjugating)
0 = a.
That is,
and so the rank of
b C2g
a=0
!
is less than
2g .
But then there is a nonzero
such that
which means explicitly for each
!
= t0 ,
that
g
g
X
X
(
bi i +
bg+1 i ) = 0.
j i=1
| {z }
=:
|i=1 {z
=:
(, ) 1 (M )1 (M ) goes to zero in Hom(H1 (M, Z), C) =
H 1 (M, C). By our sequence of injections above, = = 0. But since
b 6= 0, this contradicts linear independence of 1 , . . . , g in 1 (M ).
We nd then that
272
24. THE ALGEBRAICITY OF GLOBAL ANALYTIC OBJECTS
Exercises
(1) Write a basis for the holomorphic 1-forms on the (smooth) curve
C P2 with
dim(1 (C))?
a ne equation
1 + x6 + y 6 xy 5 = 0.
What is
CHAPTER 25
The Riemann-Roch Theorem
As you know, there are no nonconstant holomorphic functions on a
p but
1
no poles anywhere else? Then you still get nothing, unless M is P (in
1
which case there is (zz(p)) ). This is because for g = genus(M ) 1,
there is a nonzero holomorphic form which doesn't vanish at p. For
P
any meromorphic function f on M , we know that
qM Resq (f ) = 0;
so if f has a simple pole at p, then Resp (f ) 6= 0 and f must have
Riemann surface
M.
What if we allow a simple pole at one point
another pole to cancel this term.
p (but still no
other poles)? Then the answer is more complex; if g = 0 or 1 there are
nonconstant such functions (e.g. the Weierstrass -function), while if
g 2 it can depend on the point p. In general, the vector spaces of
functions f K(M ) with a single pole (at p) with p (f ) k has
dimension max{1, k g + 1}. You are guaranteed to get something
nonconstant as soon as k g + 1 2.
What if we are prepared to allow a double pole at
In the 1850's, Riemann proved a more general inequality which replaces
(and
k)
by multiple points and orders; a decade later, his
student Roch turned this into an exact equality (Theorem 25.2.3 below) incorporating another term related to meromorphic 1-forms.
encompasses the equality
dim( (M )) = g
It
and gives a powerful tool
for studying embeddings of Riemann surfaces into higher dimensional
projective spaces, among other things.
Its statement is in terms of
spaces of functions and forms related to divisors, and we will start
(25.1) by de ning these spaces precisely.
You may prefer this shorter introduction to the topic from a lecture
by Lefschetz: Well, a Riemann surface is a certain kind of Hausdor
space. You know what a Hausdor space is, don't you? Its also compact, ok. I guess it is also a manifold. Surely you know what a manifold
273
274
25. THE RIEMANN-ROCH THEOREM
is. Now let me tell you one nontrivial theorem, the Riemann-Roch Theorem.
25.1. Eective divisors and rational equivalence
P
mp [p] and E = pM np [p]
divisors on M . (Of course, only nitely many mp and np are nonzero.)
If for all p mp np , then we write D E .
Let
D=
be a Riemann surface,
Definition 25.1.1.
pM
D Div(M )
Example 25.1.2. The divisor
is eective
()
D 0.
of a holomorphic
1-form
is
eective. (Why?)
We can use this idea to put constraints on meromorphic functions
D = 3[q] 2[r],
f K(M ) with
divisor (f ) =
pM p (f )[p]. Then imposing the inequality (f )+D 0
forces q (f ) + 3 0 and r (f ) 2 0; that is, f is allowed a pole of
order no worse than 3 at q , and must have a zero of order at least
2 at r. Likewise, if K1 (M ) then () D means has a zero of
order at least 3 at q , and is allowed a pole of order no worse than 2 at
r. The next de nition formalizes this and de nes the quantities which
and forms. For instance, suppose
and
the Riemann-Roch theorem will relate.
D Div(M ),
Definition 25.1.3. For any
L(D) := {f K(M ) | (f ) + D 0} {0},
I(D) := { K1 (M ) | () D} {0}.
(The {0} just means that the zero-function is included, so as to
produce a vector space.) Set
`(D) := dim L(D) ,
i(D) := dim I(D).
The next step is to de ne an equivalence relation on divisors which
is ubiquitous in algebraic geometry.
Definition 25.1.4. Divisors
alent i there exists
f K(M )
1from A Beautiful Mind by S. Nasar
2by Chow's theorem all meromorphic
D, E Div(M )
with
are rationally equiv-
(f ) = D E ;
we write
rat
D E.
functions are rational, hence the terminology
rational equivalence (sometimes also called linear equivalence).
25.1. EFFECTIVE DIVISORS AND RATIONAL EQUIVALENCE
Proposition 25.1.5. If
rat
D E,
275
then
deg(D) = deg(E);
(ii) L(D)
= L(E);
(iii) I(D)
= I(E); and
(iv) `(D) = `(E) and i(D) = i(E).
(i)
Furthermore,
rat
respects the abelian group structure of
Div(M ).
D E = (f ). Now Exercise
(i). Given g L(D),
Proof. By assumption
deg((f )) = 0,
which yields
3.2 says that
(f g) + E = (f ) + (g) + E = (g) + D 0 ;
so
g 7 f g
de nes a map
L(D) L(E),
and
h 7
h
de nes an inverse
f
map. This gives (ii), and (iii) is done in the same way. (iv) obviously
follows from (ii)-(iii). The last statement about
that
rat
is essentially just
(D + (f )) + (E + (g)) = (D + E + (f g)).
Remark 25.1.6. In fact, the Picard group
P ic(M )
of
20.1
is just
the group of equivalence classes
Div(M )
rat
So Proposition 25.1.5(i,iv) can be thought of as saying that
de ne functions from
P ic(M )
to
deg, `,and
Z.
K Div(M ) is just the
divisor of any meromorphic 1-form K (M ). Since any two such are
Definition 25.1.7. A canonical divisor
rationally equivalent (easy exercise), there is a single canonical divisor
class
[K] P ic(M ).
The next (basic) result is sometimes called Brill-Noether reciprocity:
Proposition 25.1.8. Let
D Div(M )
be arbitrary, and
canonical divisor. Then
I(D)
= L(K D),
and so
i(D) = `(K D).
K = (); if (f ) + K D 0, then (f ) = (f ) + K
D K + K = D. So f 7 f maps L(K D) I(D), and 7
gives an inverse.
Proof. Let
276
25. THE RIEMANN-ROCH THEOREM
25.2. Proof and statement
d
projective algebraic curve with ODP singularities S = {p1 , . . . , p }.
e
Let M := C
P2 ((M ) = C ) be its normalization, with 1 (pi ) =
{qi , ri }, and de ne a divisor
Throughout this section we take
to be an irreducible degree
X
E := (S) =
[qi ] + [ri ] Div(M )
1
i=1
of degree
2 .
H P2
Given any line
( H for hyperplane), write
H := 1 (H C) Div(M )
for the intersection divisor (of degree
d).3
m N,
Lemma 25.2.1. For all su ciently large
`(mH E) md 2 g + 1
and
i(mH E) = 0,
where
g=
(d1)(d2)
2
is the genus of
M.
R S31 and F S3d for the de ning
H and C (resp.). Consider the map
Proof. Write
polynomials of
homogeneous
S3m (S) L(mH E)
G
G 7
.
Rm
By Study's lemma,
and so
G
Rm
0
G|C 0
F |G,
ker = F S3md ( S3m (S)).
Therefore, taking dimensions of
L(mH E) im(),
3If H
passes through an ODP
pi ,
then
qi and ri
will both show up in
multiplicities determined by the local intersection multiplicities of
local analytic components of
at
pi .
(See Def. 12.2.1 )
H,
with
with the two
25.2. PROOF AND STATEMENT
277
we nd
`(mH E) dim(im()) = dim
S3m (S)
F S3md (S)
= dim S3m (S) dim S3md .
By Prop. 24.2.1(c), this
(m + 1)(m + 2)
(m d + 1)(m d + 2)
2
2
d(d 3)
= dm
2
(d 1)(d 2)
+1
= dm
2
= dm 2 g + 1.
Lemma 25.2.2. Let
D Div(M ), p M .
Then
0 `(D + [p]) `(D) (i(D + [p]) i(D)) 1.
Proof. First note that
L(D) L(D+[p]) = `(D+[p])`(D)
0.
P
D = qM nq [q],
f K(M ) satisfying
Next, writing
function
(f ) + D + [p] 0
(25.2.1)
If
f, g
an element of
and
so that
f g L(D).
So
:= limxp
f (x)
, we have
g(x)
g) np
`(D + [p]) `(D) 1,
and we conclude
0 `(D + [p]) `(D) 1.
(25.2.2)
Similarly, writing
for a canonical divisor,
0 `(K D) `(K D [p]) 1
or equivalently
(25.2.3)
is a
p (f ) = (np + 1).
are two such functions, then setting
ordp (f
L(D + [p])\L(D)
0 i(D) i(D + [p]) 1.
Altogether,
0 `(D + [p]) `(D) + i(D) i(D + [p]) 2
278
25. THE RIEMANN-ROCH THEOREM
and we just have to show that 2 is impossible.
f satis es (25.2.1),
K (M ) satis es
Suppose (for a contradiction) that
equivalent to 1 in (25.2.2), and
() D
which is
and
p () = np ,
which is equivalent to 1 in (25.2.3). Then
(f ) = (f ) + () [p]
with
p (f ) = p (f ) + p () = 1.
But the sum of residues of a meromorphic form must be zero (Prop.
13.1.6(b)), so
having a single simple pole (and no other poles) is
absurd.
Theorem 25.2.3. [Riemann-Roch] Let
of genus
g, D
a divisor on
M.
be a Riemann surface
Then
`(D) i(D) = deg(D) g + 1.
Proof. By part (B) of the Normalization Theorem 3.2.1, we can
assume we are in the situation described in the beginning of the section,
with
e.
M =C
By Lemma 25.2.1, there exists
m0 Z
such that
m m0 =
`(mH E) i(mH E) md 2 g + 1.
Now for any two lines
lines in
H1 , H2 ,
we have
rat
H1 H2 ;
so if
H1 , . . . , Hm
are
then by Proposition 25.1.5(iv)
`(H1 + + Hm E) i(H1 + + Hm E) md 2 g + 1.
Taking
points
m large enough and lines through (a) all points of S and (b) all
Pm
in D , we can ensure that
i=1 Hi E D is eective, so that
H1 + + Hm E = D + [P1 ] + + [Pk ]
where
k = md 2 deg(D)
(and the
Pj
are points of
M ).
Therefore
we have
` D+
k
X
j=1
!
[Pj ]
k
X
i D+
[Pj ]
j=1
!
k + deg(D) g + 1.
25.2. PROOF AND STATEMENT
279
Repeatedly applying the right-hand inequality of Lemma 25.2.2 gives
k
X
k + `(D) i(D) ` D +
[Pj ]
!
i D+
j=1
k
X
!
[Pj ]
j=1
and we conclude that
`(D) i(D) deg(D) g + 1.
(25.2.4)
Next we show the reverse inequality. Plugging
K D
into (25.2.4),
we have
`(K D) i(K D) deg(K D) g + 1
which becomes (using Brill-Noether reciprocity)
i(D) `(D) 2g 2 deg(D) g + 1 = (deg(D) g + 1),
so that
`(D) i(D) deg(D) g + 1.
Amongst the easy corollaries of this theorem are the Riemann in-
equality
`(D) deg(D) g + 1,
and (by putting
D=0
in the theorem) the formula
dim 1 (M ) = g.
Here is one more simple application:
Proposition 25.2.4. Up to isomorphism,
surface of genus
P1
is the only Riemann
0.
0; then rst of all the above corollary
1
of Riemann-Roch says that dim (M ) = 0. If we take (for some
p M ) D = [p], then I(D) 1 (M ) = {0} = i(D) = 0. So by
Proof. Suppose
has genus
Riemann-Roch,
`(D) = deg(D) g + 1 = 1 0 + 1 = 2.
Now
L(D)
consists of functions with a simple pole allowed at
(and
L(D); and since
dim L(D) = 2 there is also a nonconstant function f L(D), which by
no other poles). The constant function
belongs to
Liouville must have the allowed simple pole. Therefore the mapping
280
25. THE RIEMANN-ROCH THEOREM
degree of
f : M P1
is (cf.
14.1)
deg(f ) = deg(f 1 ([])) = deg([p]) = 1;
that is,
is an isomorphism.
Exercises
(1) Check that any two canonical divisors on a Riemann surface are
rationally equivalent.
D Div(M ), g = genus(M ). Prove that if deg D > 2g 2,
then i(D) = 0. Likewise show that if deg D < 0, then `(D) = 0.
Let M be a genus g Riemann surface, and p M . Using RiemannRoch, nd the smallest value of k for which there must exist f
K(M ) having a pole at p of order no worse than k (i.e. p (f )
k ), and no other poles.
1
Let M have genus g 2. (a) Prove that M has a morphism to P
of degree g + 1. [Hint: use Exercise (3)] (b) Prove that M has
1
a morphism to P of degree g . [Hint: let p M , and look at
i((g 2)[p]). This is a bit harder than (a).]
Assume D > 0. By Exercise (2), if g 1 then i(D) = 0, and
Riemann-Roch becomes `(D) = deg(D) for g = 1 and deg(D) + 1
for g = 0. Prove this directly (a) for M
= P1 and (b) for M
= C/
(1-torus).
(2) Let
(3)
(4)
(5)
CHAPTER 26
Applications of Riemann-Roch, I: special Riemann
surfaces
We now focus our attention on Riemann surfaces with a degree-two
mapping to
P1 ,
starting with the case of genus 1. (The higher genus
cases can be viewed as a generalization of elliptic curves, though there
is no group law.)
The rst section begins with some general claims
which will be more thoroughly investigated in the next Chapter.
26.1. Curves of genus 1
For the proof of Riemann-Roch (Theorem 25.2.3), we needed to use
the (unproved) Normalization Theorem 3.2.1(B). It's actually possible
to argue the other way, from Riemann-Roch to the existence of plane
projective immersions (with ODP singularities) for arbitrary Riemann
surfaces.
When can we do better? The degree-genus formula tells you that
(d1)(d2)
,
only curves of genera 0, 1, 3, 6, 10, 15, . . . (numbers expressible as
2
d N) can ever be embedded as smooth curves in P2 . There is no reason to believe, from this or from the Normalization Theorem, that an
arbitrary Riemann surface of one of these genera can be so embedded. In fact, it isn't true once you get to genera
6, 10, 15, . . ..
That it
works for genus 1 and genus 3 (almost; see next Chapter) is a bit of
a miracle!
So: if you buy that any genus 1 Riemann surface is a complex 1torus and any torus can be Weierstrassed into
isn't surprising.
P2 , the following result
On the other hand, it shows that Riemann-Roch is
powerful and gives us a hint of how we might prove similar results in
higher genus (e.g.,
and
Theorem 26.1.1. Let
3)
M
later.
be a Riemann surface of genus one. There
exists an injective morphism of complex manifolds
image
(M )
a smooth algebraic curve of degree
281
3.
: M , P2
with
282
26. APPLICATIONS OF RIEMANN-ROCH, I: SPECIAL RIEMANN SURFACES
p M,
Proof. Given
we know that
i(2[p]) = 0 = i([p])
by Exer-
cise 25.2, so that Riemann-Roch yields
`(2[p]) = deg(2[p]) g + 1 = 2 1 + 1 = 2,
`([p]) = deg([p]) g + 1 = 1.
In terms of the spaces of meromorphic functions, this says that
L(2[p]) ( L([p]) = L(0) = O(M ),
where
dim L([p]) = 1 means L([p]) consists of constant (or equivalently,
holomorphic) functions. Therefore, we have an element
x L(2[p])\L([p]),
i.e. a meromorphic function with a double pole at
Regard
as a morphism
M P1 .
p and no other poles.
By Riemann-Hurwitz, the
rami cation degree
deg(Rx ) = 2(deg(x) + g 1) = 2(2 + 1 1) = 4,
p (x)
whereas the rami cation indices
all
2.
for a degree two mapping are
Hence, the rami cation divisor is of the form (cf.
14.1
for
notation)
Rx = [p1 ] + [p2 ] + [p3 ] + [p4 ]
ai = x(pi ) P1 . The {ai } are still
distinct points: by the form of Rx , [pi ] must occur with multiplicity
1
1
two in x ([ai ]); and since (deg(x) = 2 = ) deg(x ([ai ])) = 2, the
1
only possibility is x ([ai ]) = 2[pi ].
Now clearly one of the pi , say p4 , has to be p (as x has a double
pole there). So also a4 = , and we have the picture:
with
p1 , p2 , p3 , p4 M
distinct. Set
p1
p2
p3
a1
a2
a3
x
1
In the following,
{pi }
resp.
{ai }
means
i = 1, 2, 3.
26.1. CURVES OF GENUS 1
Next, notice that
x ai
283
is a local coordinate about
ai
P1 .
on
The
pi is simply that there is a local
pi on M such that1 zi2 = x ai . Dif-
meaning of a degree-2 rami cation at
(holomorphic) coordinate about
loc
dx = 2zi dzi .
ferentiating gives
Again using Riemann-Roch, we have
a nonvanishing holomorphic form
1 (M ),
and by the Residue
Theorem
0=
Writing
(locally)
z :=
= dz ;
Resq (x ) = Resp (x ).
qM
p,
we have
+ h(z) (h
holomor-
for a local holomorphic coordinate at
since the residue vanishes,
loc 1
x=
loc
z2
2
1
2
0
term. Taking dierentials, dx =
+
h
(z)
dz . Put
phic) has no
3
z
z
together with the previous local computation, this tells us that dx has
divisor
(dx) =
3
X
!
[pi ]
3[p].
i=1
y0 := dx
K(M ) .
(y0 ) = (dx). If we put
Set
that
In light of the fact that
() = 0,
we have
3
Y
g(x) :=
(x ai ),
i=1
P3
2
then (g(x)) =
i=1 2[pi ]) 6[p] = 2(y0 ) = (y0 ). We
i=1 (x ai ) = (
g(x)
conclude that
has trivial divisor and so is some constant C , and
2
y0
de ne y := y0 C so as to have
P3
y 2 g(x) = 0
on
M.
: M P2 for the morphism
de ned by sending p 7 [0 : 0 : 1] and all other points q 7 [1 :
x(q) : y(q)]. The image (M ) is contained in the projective closure E
2
2
of {y g(x) = 0} (in P ), which is smooth due to distinctness of the
{ai }, and connected due to its irreducibility. By the usual arguments,
(M ) is open and closed in E , hence equals E . At this point we have
Now for the embedding.
1For
Write
2
zi it means
p that x ai = zi hi (zi ) where
h doesn't vanish at 0; and then we can put zi := zi hi (zi ).
2the coe cient of 1 can be achieved by rescaling if needed
z2
an arbitrary choice of local coordinate
284
26. APPLICATIONS OF RIEMANN-ROCH, I: SPECIAL RIEMANN SURFACES
a diagram
C \ {p}
(26.1.1)
E \ {[0 : 0 : 1]}
/ P2 \ {[0
PPP
lll
PPP
lll
l
PPP
|E
l
l
x
PPP
lll
P'
lll
u
l
1
/ /
: 0 : 1]}
([Z : X : Y ]) := [Z : X].
is not injective, there exist
where
If
distinct points
q1 , q2 M \{p}
such
that
(q1 ) = (q2 ) =: Q;
applying
to this gives
x(q1 ) = x(q2 ) = (Q) =: ,
is not or one of the {ai }. Since deg(x) = 2, deg(x1 ([])) =
2 and we must have x1 () = {q1 , q2 }. From the equation for E it is evp
1
g())
ident that deg(|E ) = 2 also, with (|E ) () consisting of (,
p
g()). Clearly one of these points has to be Q. From
and (,
(26.1.1), it is also clear that q1 , q2 are the only points of M that can go
to these points. So whichever is not Q cannot get hit and fails to be
surjective, a contradiction.
in which
26.2. Hyperelliptic curves
Above we used the fact, for a genus one Riemann surface
M,
that
`(2[p]) = 2 > 1 = `([p]) for p M , to construct a degree-two mapping
x : M P1 . Now suppose M has genus 2: how to map it to P1 ? Well,
1
we have a basis {1 , 2 } (M ), and 2 produces a meromorphic
1
function, which does the job. By Poincar-Hopf, deg((1 )) = 2g 2 =
2, and so this map has two simple poles (or one double pole), hence
has degree two.
In terms of homogeneous coordinates, we might write
p 7 [1 (p) : 2 (p)],
loc
i = fi (z)dz in
terms of a local coordinate vanishing at p) simply [f1 (0) : f2 (0)]. If both
fi could simultaneously equal zero we would have a well-de nedness
where the meaning of the right-hand side is (expressing
problem (which could be gotten around by taking a limit), but this
does not happen: we would have to have
i([p]) 2.
By Riemann-Roch
26.2. HYPERELLIPTIC CURVES
285
`([p]) = deg([p]) g + 1 + i([p]) 2, thereby producing an
1
isomorphism M P as in the proof of 25.2.4, contradicting g = 2.
this yields
This discussion hopefully motivates
M is hyperelliptic
1
morphism x : M P .
Definition 26.2.1. A Riemann surface
exists a (nonconstant) degree-two
Clearly, any genus
Now, let
i there
Riemann surface is hyperelliptic.
be hyperelliptic of any genus and consider what the
Riemann-Hurwitz formula has to say when applied to
x:
M = 2P1 rx
X
2 2g = 2 2
(p (x) 1),
pM
where
deg(x) = 2 = p (x) 2.
rami cation points, and this is just
So the sum equals the number of
2g + 2:
Rx = [p1 ] + + [p2g+2 ].
P1 if necessary, we may as1
sume that none of the x(pi ) =: ai are 0 or . Put x ([]) =: [p] + [q]
1
0
0
and x ([0]) =: [p ] + [q ]. We have the picture
x
By composing
with an automorphism of
p1
a2
: M M denotes
M over P1 (cf. Exercise 1).
Lemma 26.2.2. Let
a2g+1 a2g
the involution exchanging the branches
(+1)-
and
(1)-
eigenspaces of
polynomial
m(t) = t 1.
diagonalizable.
= 1.
V =V
J.
Proof. With respect to any basis for
J: V
V into
be a nite-dimensional vector space,
an involution. Then we have a decomposition
the
a1
in which
p2g+1 p2g
P
of
p2
V,J
is a matrix with minimal
This has no repeated roots, and so
Moreover, since
J =
idV , any eigenvalue
is
satis es
286
26. APPLICATIONS OF RIEMANN-ROCH, I: SPECIAL RIEMANN SURFACES
We apply this to the pullback map
: 1 (M ) 1 (M ).
Notice
1 (M )+ = {0} since such forms would be pullbacks of holomorphic
1
1
1
forms from P (cf. Exercise 2). Hence (M ) = (M ) and so
that
=
1 (M ).
Put D = (g + 1)[p] + (g + 1)[q] Div(M ). We have (Exercise 25.2)
i(D) = 0, so that Riemann-Roch gives
for all
`(D) = 2g + 2 g + 1 = g + 3.
Now apply the Lemma again, this time to
that
L(D)+
: L(D) L(D),
noting
contains the linearly independent set
{1, x, x2 , . . . , xg+1 }.
In fact,
LM (D)+ = x LP1 ((g + 1)[]),
|
{z
}
polynomials of
degree g + 1
and so the above set is a basis. Therefore
dim(L(D) ) = `(D) dim(L(D)+ )
= (g + 3) (g + 2) = 1,
y L(D) such
P2g+2
(y) = i=1 [pi ] D.
and there exists a nonzero
Claim 26.2.3.
Proof. Since the
and so
pi
that
y = y .
are rami cation points,
(pi ) = pi
. But then
y(pi ) = ( yi )(pi ) = y((pi )) = y(pi )
P2g+2
y(pi ) = 0. That is, y 1 ([0]) i=1
[pi ], which implies
X
deg((y)) = deg(y 1 ([0])) deg
[pi ] = 2g + 2.
On the other hand,
y L(D) = y 1 ([]) D =
deg((y)) = deg(y 1 ([])) deg(D) = 2g + 2.
So
deg((y)) is forced to equal 2g + 2,
[pi ] and y 1 ([]) = D.
3using
(e.g.
which means also that
y 1 ([0]) =
subscripts to denote which Riemann surface we are considering functions on
LM (D)
just means
L(D))
26.2. HYPERELLIPTIC CURVES
Set
287
2g+2
g(x) :=
(x ai ) K(M ) ,
i=1
Div(M ))
X
X
X
(g(x)) =
((xai )) = 2
[pi ](2g+2)x1 ([]) = 2
[pi ]2D = (y 2 ).
and compute (in
y 2 /g(x) has trivial
that (in K(M ))
But then
we have
divisor, and so is a constant. Rescaling
y,
y 2 g(x) = 0.
By considering the image of
: M P2
given by
m(6= p, q) 7 [1 : x(m) : y(m)]
and
p, q 7 [0 : 0 : 1],
we arrive at:
Theorem 26.2.4. Hyperelliptic Riemann surfaces are precisely the
normalizations of (plane) algebraic curves of the form
2g+2
(
Y 2 Z 2g =
(X ai Z)
P2 .
i=1
A basis of
1 (M )
is given by
j :=
Proof. We just need to show
xj1 dx
,
y
j = 1, . . . , g.
is holomorphic:
(j ) = (j 1)(x) + (dx) (y)
= {(j 1)([p0 ] + [q 0 ]) (j 1)([p] + [q])}
nX
o
+
[pi ] 2([p] + [q])
nX
o
[pi ] (g + 1)([p] + [q])
= (j 1) ([p0 ] + [q 0 ]) + (g j) ([p] + [q]) 0.
4note:
the singular point
[0 : 0 : 1] is not an ODP, so the construction of g
holomor-
phic dierentials that follows shouldn't be compared with the formulas you know
in that case. Also, it should be emphasized that the
ai
are distinct.
288
26. APPLICATIONS OF RIEMANN-ROCH, I: SPECIAL RIEMANN SURFACES
A hyperelliptic curve, by the way, is just an irredicuble projective
curve whose normalization is a hyperelliptic Riemann surface!
The rst two exercises below are ones you could have done long ago,
but ll in (very) small gaps in the proofs above. The same goes for the
third, if you had known the de nition of hyperelliptic! The fourth does
make heavy use of Riemann-Roch.
Exercises
: M M 0 of compact
Riemann surfaces, with corresponding involution de ned as follows: if p M is a rami cation point of , (p) := p; otherwise,
1 ((p)) = {p, p} and (p) := p. Clearly = IdM (i.e. is an
involution) and = . Prove that : M M is a holomorphic map of Riemann surfaces. Since is injective and surjective
(why?), it follows that Aut(M ).
1
Continuing Exercise (1), let (M ) satisfy = . Prove
1
0
that = for some (M ).
2
Suppose that a dth-degree irreducible algebraic curve C P has
a point of multiplicity (d 2). Show that C is hyperelliptic.
Let M be a Riemann surface of genus two. In this problem you
will construct a realization of M as an algebraic curve, dierent to
that produced above. You will need to use that M is hyperelliptic,
1
with x : M P its degree-two mapping and the associated
involution. Take p and q (in contrast to the notation above) xed
non-rami cation points on M with distinct images under x; let
and denote arbitrary points of M .
(a) Prove that `([] + []) = 1 unless () = . [Hint: otherwise
(1) Given a degree
(2)
(3)
(4)
holomorphic map
you get a dierent involution (why?). To see why this is a problem
you might consider the fact that
for all holomorphic
-symmetric.]
(b) For any points , on M , show i([] + [] [p] [q]) = 1 (as
opposed to 2) {, } =
6 {p, q}. [Hint: use (a), and consider
`([] + [] [p] [q]).]
2
(c) Use I([p] [q]) to construct a map : M P . [Hint:
compute i([p][q]).] You will need to check that is well-de ned.
[Hint: compute i([p] [q] + []), using Exercise 25.2.]
forms implies their divisors are
EXERCISES
289
is injective o {p, q}, but that (p) = (q). [Hint:
using part (b), compute i([] + [] [p] [q]).]
(e) Show that there exists a meromorphic form I([p] [q])
with poles at both p and q .
(f ) Explain why the zero-divisor ()0 (the eective part of ())
1
2
is
of the intersection [divisor] of a line in P with C := (M ).
Prove that deg(()0 ) = 4 (easy). Assuming C is an algebraic curve
(this is GAGA), conclude that deg(C) = 4.
(g) Clearly (p) = (q) is a singularity of C . Prove it is the only
(d) Show that
one, and a double point. [Hint: assume otherwise, and produce a
genus zero normalization or similar.]
CHAPTER 27
Applications of Riemann-Roch, II: general
Riemann surfaces
Next we'll use Riemann-Roch to develop two methods for mapping
an arbitrary Riemann surface into a (usually higher-dimensional) projective space, with a nice application to curves of genus three.
The
second approach behaves dierently in the hyperelliptic and nonhyperelliptic cases, so we rst will want to convince ourselves that there are
nonhyperelliptic Riemann surfaces! To see this, we will start with an
heuristic argument for the number of complex parameters governing
Riemann surfaces, and show that the hyperelliptic ones have fewer parameters. But there's much more in this chapter, which should give a
glimpse of how rich the correspondence between algebraic curves and
Riemann surfaces really is.
27.1. Moduli
In algebraic geometry there is the notion of moduli spaces, which
parametrize structures of a prescribed sort modulo some equivalence
relation, such as smooth algebraic curves of degree
equivalence or Riemann surfaces of genus
up to projective
up to isomorphism. A
main point is that these spaces can be given algebraic structure themselves, i.e.
turned into algebraic varieties, in many cases.
Suitably
re ned, the structure of these varieties is one of the hotter topics of
study around.
We shall only be concerned with the notion of moduli as a set of
local parameters (on the moduli space), and will say colloquially that
some structure has a certain number of moduli : e.g. genus
(resp.
0)
Riemann surfaces have one modulus (resp. zero moduli) since they can
all be expressed as
1so-called
C/Z h1, i (resp. P1 ) up to isomorphism.
Underlying
modular curves or modular varieties are a more specialized notion
with an arithmetic avor
291
292
27. APPLICATIONS OF RIEMANN-ROCH, II: GENERAL RIEMANN SURFACES
26.1 that not all Riemann surfaces of genus 6, 10, 15, etc.
2
embedded smoothly in P is a deep calculation of Riemann:
the claim in
can be
Theorem 27.1.1. Riemann surfaces of genus
phism) have
3g 3
[p]) = 0
(up to isomor-
moduli.
Proof. (Sketch) Consider a genus
eective divisor
g2
2g
of degree
for any point
p M.
on
M.
M , and any
i(D) = i(D
Riemann surface
By Exercise 25.2,
So Riemann-Roch says
`(D) = deg(D) g + 1 = g + 1,
`(D [p]) = g.
f L(D) and not in any of the nitely many L(D
[p]) for those p appearing in D; and so f 1 ([]) = D, which means
deg(f ) = deg(f 1 ([])) = 2g . Now Riemann-Hurwitz tells us about
Hence there exists
the rami cation behavior:
M = deg(f ) P1 deg(Rf )
2 2g = 2g 2 rf
rf = 6g 2.
For almost all
the points in
Rf
will have multiplicity one (rami -
cations of order two) and lie over distinct points in
the branch locus
B P1
consists of
6g 2
P1 ,
meaning that
points. We want to use all
this data to compute the number of local deformation parameters of
M.
Look at this in a slightly more formal way, consider the set
S1
(M, f ) where M has genus g and f has degree 2g . This
maps to the set S2 of 2-tuples (M, D) where D 0 of degree 2g (take
D := f 1 ([])). From there you can map to the set S of Riemann
surfaces of genus g , by forgetting D . It's fairly clear that ( xing M ) D
has 2g parameters, making dim(S2 ) dim(S) = 2g . Moreover, given
M and D, there are `(D) = g + 1 choices of parameter for f (to have
D as its poles), meaning dim(S1 ) dim(S2 ) = g + 1. Our argument
of 2-tuples
in the rst paragraph shows that the rst map is surjective (while the
second obviously is), and so
(27.1.1)
dim(S) = {dim(S1 ) g 1} 2g = dim(S1 ) 3g 1.
|
{z
}
dim(S2 )
27.1. MODULI
293
S1 to S3 , the set of (6g 2)-tuples
1
1
of (unordered) points on P , by taking f (Rf ) Div(P ). This map
1
is surjective since given a branch-point set in P you can construct an
On the other hand, you can map
2 and in fact
existence domain for an appropriate algebraic function,
the construction shows that there are only nitely many possibilities
for
M.
Moreover, it shows that a continuous family of degree-2g func-
tions on
with the same branch-point set gives rise to a continuous
M.
family of automorphisms of
But for
g 2 M
has only nitely
many automorphisms. So we see that this map is nite-to-1, and thus
dim(S1 ) = dim(S3 ) = 6g 2.
Plugging this in to (27.1.1), we get the
desired result.
It's much easier to count moduli for hyperelliptic and algebraic
plane curves.
Proposition 27.1.2. Hyperelliptic Riemann surfaces of genus
(up to isomorphism) have
2g 1
moduli.
Proof. They are essentially just the existence domains of the al-
qQ
2g+2
i=1 (z i ), and so are completely determined
2g+2
by the branch locus {i }i=1 . This has 2g + 2 parameters, but we have
1
to account for change of coordinate on P , which is by P GL2 (C), by
gebraic functions
subtracting
dim(P GL2 ) = 3.
Proposition 27.1.3. Smooth algebraic curves of degree
projective equivalence have
d+2
2
dimension
coordinates by
modulo
moduli.
Proof. A curve is determined by a polynomial in
d+2
S3d ,
which has
We have only to account for changing projective
GL3 (C),
which has dimension
9.
(Here
P GL3
is not
what is wanted, as we do want to quotient out the rescalings of the
equation.)
Now we can compare moduli, with two very interesting results.
First, consider the numbers you get for general Riemann surfaces of
1, 2, 3, 4, 5, 6 : the numbers of moduli are 1, 3, 6, 9, 12, 15. For
hyperelliptic ones, we have instead 1, 3, 5, 7, 9, 11. So while all genus
genus
2 Riemann surfaces are hyperelliptic, we have:
2like p(z a)(z b)(z c)(z d) (for g = 1), but more complicated (since g 2);
cf. [Gri ths and Harris], pp. 255-257
294
27. APPLICATIONS OF RIEMANN-ROCH, II: GENERAL RIEMANN SURFACES
Proposition 27.1.4. A general Riemann surface of genus
g 3
is non-hyperelliptic.
So we will need a more general method of realizing Riemann surfaces
as algebraic curves than that discussed in
26.2,
and that is what we
will do in the remainder of this Chapter.
Finally, look at those genera which correspond to nonsingular algebraic curves in
P2 of degree 3, 4, 5, 6, 7, . . .:
namely,
1, 3, 6, 10, 15, and
so on. The Riemann surfaces of these genera have (by Thm. 27.1.1)
moduli
1, 6, 15, 27, 42,etc.
But now look at the smooth algebraic
plane curves of the corresponding degrees (via Prop. 27.1.3): we get
1, 6, 12, 19, 27.
The case of genus
will be treated in
27.3.
Beyond
that, we have:
Proposition 27.1.5. Smooth algebraic plane curves of degree
d5
(d1)(d2)
do not yield all the Riemann surfaces of genus
only a
2
special subset.
27.2. Projective embeddings
Let
degree
M be a Riemann surface of genus g . For any D Div(M ) of
> 2g 2, recall from Exercise 2 of Chap. 25 that i(D) = 0. By
Riemann-Roch, we then have
`(D) = deg(D) g + 1.
This will be used repeatedly in the argument below.
Now x a divisor
D=
pM
np [p] Div(M )
of degree
d 2g + 1.
We will de ne an embedding (injective morphism of complex manifolds)
: M , Pdg .
d g g + 1, this can only give an embedding in P2 for g = 1.)
The support of D , which is the subset of M consisting of the points
aapearing in D (i.e. those p with nonzero np ), is written |D|.
(Since
3It's
very important to understand the argument in this section. Try slimming it
down (I've expressed it in a somewhat bloated manner) and writing it out for a
speci c choice of
and
g (> 1,
say). (Also, if you are stuck on Chap. 26 Exercise
4, some of the steps are similar.)
27.2. PROJECTIVE EMBEDDINGS
295
d > 2g 2 and so `(D) = d g + 1.
basis of L(D), and de ne for p
/ |D|
First o, certainly
{f0 , . . . , fdg }
(p) := [f0 (p) : : fdg (p)].
(27.2.1)
If
for a
Write
p |D|,
this is unsuitable since some functions may blow up (or
all functions may be required to vanish).
coordinate (vanishing at
Therefore if
is a local
to rst order), we put
(p) := [(z np f0 )(p) : : (z np fdg )(p)].
(27.2.2)
q in a neighborhood of p, [(z np f0 )(q) : : (z np fdg )(q)] gives
the same result as [f0 (q) : : fdg (q)], and so we have constructed
an analytic map . . . provided that (27.2.1-2) do not yield [0 : : 0]
For points
at any point. That is the central well-de nedness issue, and we must
check it.
Now for
p, q M ,
each still have degree
notice that
> 2g 2.
D [p], D 2[p]
and
D [p] [q]
Therefore we have
`(D [p]) = d g
and
`(D 2[p]) = d g 1 = `(D [p] [q]),
with the immediate consequences
(27.2.3)
L(D [p]) ( L(D),
(27.2.4)
L(D [p] [q]) ( L(D [p]),
(27.2.5)
L(D 2[p]) ( L(D [p]).
p, q
/ |D|. Then
(27.2.3) says that there exists f L(D) not vanishing at p, meaning
that the {fi (p)} are not all zero; this makes well-de ned on M \|D|.
Next, (27.2.4) gives us g L(D [p])\L(D [p] [q]), a function
vanishing at p but not q , forcing to take dierent values at p and q ;
hence is injective on M \|D|. Finally, (27.2.5) provides h L(D
[p])\L(D 2[p]), i.e. vanishing to exactly rst order at p, so that the
To interpret these, for simplicity rst assume
4Note
p |D|;
np = 0.
that (27.2.2) isn't just a special formula for
general than) (27.2.1) since for
p
/ |D|
we have
it contains (is more
296
27. APPLICATIONS OF RIEMANN-ROCH, II: GENERAL RIEMANN SURFACES
derivative of
hence that of
is nonzero there; together with the
M \|D| is smooth.
In order to extend these statements to all of M , we have to re ne
the argument just a bit. For a general points p, q M , (27.2.3) tells
us that there exists a function f L(D) with p (f ) = np exactly;
(27.2.4) that some g L(D) has p (g) > np but q (g) = nq ; and
(27.2.5) that there exists an h L(D) with p (h) = np + 1 exactly.
injectivity result, this proves that the image of
These give precisely the well-de nedness, injectivity, and smoothness
of image for the map described by (27.2.2). So the image is a compact
complex analytic curve
Pdg ,
which is algebraic by GAGA.
27.3. Canonical maps
Once again we consider a Riemann surface
g 2,
M,
this time of genus
and let
{1 , . . . , g } 1 (M )
be a basis. Instead of choosing a divisor and go through Riemann-Roch
to get a projective embedding from meromorphic functions, why not
just use these? De ne the canonical map
K : M Pg1
by
p 7 [1 (p) : : g (p)].
The meaning of this, as you would expect, is locally writing each
as
fi (z)dz , and taking [f1 (p) : : fg (p)]. This is well-de ned, i.e. the
{i } do not all have a zero at p. Otherwise we would have I([p]) =
I(0) = g hence (by Riemann-Roch) L([p]) = 2, which we know to be
1
false for M not isomorphic to P .
Bottom line:
this looks quite promising, from the standpoint of
getting a convenient projective embedding. Or does it?
Example 27.3.1. For
hyperelliptic, consider the setting of The-
orem 26.2.4; we have
dx x dx
xg1 dx
K (p) =
:
: :
= [1 : x : : xg1 ].
y
y
y
27.3. CANONICAL MAPS
297
Notice that this looks a lot like the rational canonical map
from
Example 7.3.7. In fact, it factors
(27.3.1)
C @
@
@@
@@
@@
P
z<
z
zz
zz
zz f
g1
P1
with
deg(x) = 2,
and so does not give an embedding of
in
Pg1 .
All is not lost: the hyperelliptic case is very special (in a bad way),
and for
g3
we know that there are (lots of ) nonhyperelliptic curves.
Theorem 27.3.2. Let
Riemann surface of genus
(a)
(b)
(c)
K is nondegenerate;5
K (M ) is smooth; and
K is injective M
Proof. (a) Were
Pg1 ,
K be the canonical
g 2. Then
K (M )
map for an arbitrary
is nonhyperelliptic.
contained in a proper linear subspace of
this would produce a linear relation on the
{i }.
But they are
linearly independent by construction, being a basis!
(b) This is clear in the hyperelliptic case, by observing that the derivative of the (injective) rational canonical map is nowhere vanishing.
(We will return to this in the nonhyperelliptic case.)
(c) The implication
is already done (by contrapositive) in
Example 27.3.1.
Now let
be a local coordinate vanishing to rst order at a point
p M , and consider the linear functionals on 1 (M ) (= I(0)) given by
(p)
7
dz
0
7
(p)
dz
.
.
.
(k1)
(p).
dz
If the rst is zero on some given , then I([p]). If the rst and
second are zero, then I(2[p]). If all are zero, then I(k[p]).
Since k linear conditions cut out a subspace of codimension at most k ,
5cf.
the beginning of Chapter 7, and also 7.3.1.
298
27. APPLICATIONS OF RIEMANN-ROCH, II: GENERAL RIEMANN SURFACES
we have
i(k[p]) = dim I(k[p]) g k.
More precisely, we have
i(k[p]) = g k + a
and by Riemann-Roch
`(k[p]) = k g + 1 + i(k[p]) = 1 + a.
k = 1, there can be no redundancies in one linear
condition and so we have a = 0.
Suppose K ([p]) = K ([q]) for p 6= q . Then 7
(p) and 7
dz
1
(q) yield the same functional on (M ), up to a constant multiple;
dz
in particular they vanish on the same 's. So I([p]) = I([p]+[q]), which
yields i([p] + [q]) = i([p]) = g 1 and via R-R
For the special case
`([p] + [q]) = 2 g + 1 + i([p] + [q]) = 2.
Therefore there exists a nonconstant meromorphic function
F L([p]+
[q]). We know that L([p]) and L([q]) have only constant functions
(a = 0 when k = 1), and so F has to have the allowed simple pole at
both of p and q . Thus deg(F) = 2, and so M is hyperelliptic. This
completes the proof of (c).
(b, cont'd.)
have
M is non hyperelliptic.
a = 0 for k = 2. Consequently
Now assume
`(2[p]) = 1,
i.e.
Then we must
i(2[p]) = g 2
i([p]) = g 1),
coordinates at p,
(whilst
local
and we can arrange a basis of
1 (M )
so that in
loc
loc
1 = dz , 2 = zh2 (z)dz ,
loc
j = z 2 hj (z)dz (3 j g).
(Here the
hi (z)
are holomorphic, and
h2
doesn't vanish at
z = 0.)
The
canonical map takes the local form
K (z) = [1 : zh2 (z) : z 2 h3 (z) : : z 2 hg (z)]
with derivative
K (z) = [0 : h2 (z) + zh02 (z) : : : ]
which does not vanish at
p.
This gives the desired smoothness.
27.4. WEIERSTRASS POINTS
Consider a smooth, irreducible algebraic curve
299
and hyperplane
H = {W (Z) = 0}, both in Pn . (Here W is a homogeneous polynomial
in Z1 , . . . , Zn+1 of degree one, with a ne form w .) One can de ne an
intersection divisor C H on C in a way which extends what we have
2
done in P . If C is not necessarily smooth, then the divisor lives on
a normalization M of C and is denoted H ; it is given simply by
P
pM ordp ( w)[p]. We de ne the degree of the curve to be the degree
of this divisor, called a hyperplane section :
deg(C) := deg( H).
Since any two hyperplane sections are rationally equivalent (why?),
any two hyperplane sections have the same degree, making
deg(C)
well-de ned.
In the case at hand,
is
n = g 1. Hyperplane sections are
Pg
we write W (Z) =
i=1 i Zi , then
and
particularly interesting because if
H = (1 1 + + g g )
M!
K (M ) a
is a canonical divisor on
map, and its image
That's why
is called the canonical
canonical curve.
Proposition 27.3.3. Assume
Then the degree of the canonical curve
is nonhyperelliptic of genus
K (M ) P
g1
is
Proof. The assumption is necessary in order that
ize
K (M ).
g.
2g 2.
M
normal-
(In the hyperelliptic case, it is normalized by the ratio-
nal canonical map.)
deg(K) = 2g 2
deg(K (M )) = deg(K H) =
and that's it.
We then compute
by Poincar-Hopf,
And so, we nd that nearly all genus
3 curves have a nice embed-
ding into the projective plane.
Corollary 27.3.4. Every nonhyperelliptic genus
normalization of a smooth quartic curve in
curve is the
27.4. Weierstrass points
We began our discussion of Riemann-Roch with a naive analysis,
for a xed point
on a Riemann surface
M,
of what orders of pole
are possible if we are after a meromorphic function with its only pole
at
p.
To conclude, I will now brie y explain the sense in which this
300
27. APPLICATIONS OF RIEMANN-ROCH, II: GENERAL RIEMANN SURFACES
can depend on the choice of
g>0
and not just the genus
of
M.
Assume
for what follows.
First note that
`(0) = 1
(constant functions), and
`([p]) = 1
by the
argument at the beginning of Chap. 25. By equation (25.2.2), we know
for each
that
0 `((k + 1)[p]) `(k[p]) 1.
(2g 1)[p] exceeds 2g 2,
i((2g 1)[p]) = 0, and so (by
On the other hand, since the degree of
we have (Chap.
25 Exercise 2) that
Riemann-Roch)
`((2g 1)[p]) = (2g 1) g + 1 = g.
More generally, for
k 2g 1,
the fact that
i(k[p]) = 0
yields
`(k[p]) = k g + 1.
`(k[p]) starts (at k = 0) at 1 and works its
way up to g in increments of 0 or 1, as k rises to 2g 1; thereafter it
increases by 1 whenever k does.
The situation with i(k[p]) is dual: it starts at g and works its way
down to 0 in decrements of 0 or 1, as k rises to 2g 1; and then it stays
at 0.
Now it turns out that at all but nitely many points, `(g[p]) = 1;
So the scenario is that
that is, all the increments are postponed as far as possible and the
sequence
`(k[p])
looks like
1, 1, 1, . . . , 1, 2, 3, . . . , g,
and so on.
Those
points where this is not the case are called the Weierstrass points of
M.
The simplest example I am aware of is, for a hyperelliptic curve,
2g + 2 xed points of the involution .
like 1, 1, 2, 2, 3, 3, etc.
the
For these the sequence looks
Exercises
K (p)
near p.
(1) Check that the de nition of
choice of local coordinate
in
27.3
(2) Show that any smooth quartic curve in
genus
3),
P2
is independent of the
is a canonical curve (of
and hence also nonhyperelliptic.
(3) In this exercise you will prove a new Cayley-Bacharach type result
(in
P2 ):
if
C1
and
C2
(of degrees
and
assumed smooth and irreducible) meet in
n respectively, with C1
mn distinct points, and
EXERCISES
C3
301
m+n3) passes through p1 , . . . , pmn1 C1 C2 , then
through the remaining point p := pmn . (We shall write f1 ,
(of degree
it passes
f2 , f3
for the resp. de ning homogeneous polynomials.) Start out
by assuming
C3
does not contain
p,
and follow these steps:
g denote the genus of C1 , and show that (m 3)m = 2g 2
m3
and g = dim(S3
).
f2 h
m3
(b) Let h S3
, set Fh :=
| K(C1 ) , and write (Fh ) =
f3 C 1
[p] + (h) D (this de nes D Div(C1 )). Show that the map
S3m3 L(D)/C (here C =constant functions) given by h 7 Fh is
injective, and use this to put a lower bound on `(D).
(c) Find deg(D), i(D), and obtain a contradiction.
(a) Let
(4) A problem on automorphisms of canonical curves:
(a) Let
: C C
C of a linear automor
1
phism of P
. [Hint: consider the action of on (C).]
(b) Let M be a non hyperelliptic Riemann surface of genus 3 with
an involution . How many xed points does it have? What is the
genus
g.
be an automorphism of a canonical curve of
Prove that
is the restriction to
g1
genus of the quotient Riemann surface? [Hint: consider the canon-
ical embedding and apply (a);
linear automorphism on
(5)
is the restriction of what sort of
P ?]
Using the embedding of 27.2,
try to prove: (a) there exists, for an
arbitrary Riemann surface, an embedding onto a smooth curve in
P3
in
and (b) an immersion onto a curve with only ODP singularities
P2 .
[Hint: use su ciently general projections of the complement
of a linear subspace in
do these curves have?
Pn
onto
P2
and
P3 .]
Also, (c) what degrees
CHAPTER 28
Abel's Theorem, part I
M a Riemann surface of genus
g 1, with closed paths (1-cycles)i giving a basis {[i ]}2g
i=1 for
H1 (M, Z). We have the Jacobian of M , which is the complex g -torus
Recall the setup from Chapter 20:
g
(1 (M ))
= C
.
J(M ) :=
H1 (M, Z)
M
The isomorphism is given by evaluating functionals against a basis
{1 , . . . , g } 1 (M ),
and
is called the period lattice. The Picard
group
Div 0 (M )
(K(M ) )
P ic0 (M ) :=
of degree-0 divisors modulo rational equivalence is the object we want
to understand. To this end, we had shown that the Abel-Jacobi map
AJ : P ic0 (M ) J(M )
D 7
1 D
is a well-de ned homomorphism, where
writing some 1-chain
is that
AJ((f )) = 0
1 D
= D. The
f K(M ) .
with
for any
is just a compact way of
important content of this
By Abel's theorem we will henceforth mean the statement that
AJ
is injective, that is
(28.0.1)
AJ(D) = 0
D = (f ) for
some
f K(M ) ;
while the surjectivity will be known as Jacobi inversion: i.e.,
(28.0.2)
J(M ) [any functional on 1 (M ), up to periods],
divisor D inducing that functional via
().
1 D
given any point in
there exists a
These statements will be proved in Chap.
29.
Our aim here is just
to explain how Abel's theorem relates to Riemann-Roch and develop a
couple technical lemmas to be used in the sequel.
303
304
28. ABEL'S THEOREM, PART I
Before starting let's re ne one aspect of the above picture just a
bit. Intersecting 1-cycles on
or more precisely, intersecting trans-
verse representatives of homology classes gives a perfect, unimodular
pairing
: H1 (M, Z) H1 (M, Z) Z.
There is always a symplectic basis, that is one with the property that
i g+j = ij = g+j i
i j = 0 = i+g j+g
1 i, j g (where ij is the Kronecker delta). This is the situation
2g
pictured in 20.1. We will assume henceforth when writing {i }i=1 that
for
they form a symplectic basis for the rst homology.
We should also remark on what the Picard group is really doing
E , in P ic0 (E) we have [p+q][p][q]+[O]
0, where addition inside the brackets is the group law on E and outside
here. For an elliptic curve
the brackets means adding divisors. What this says is: while as divisors
E ) [p+q] 6= [p]+[q], working
modulo rational equivalence we do have [p + q] + [O] [p] + [q]. So
P ic eectively recovers the group law on E . Now, curves of higher
(i.e. in the free abelian group on points of
genus have no group law on points; but by linearizing points and
working modulo divisors of functions, we get a form of generalization
of the group law in genus
1.
Intriguingly, a more precise form of Jacobi
inversion in the next Chapter will tell us that this may almost be
seen as a group law on unordered
g -tuples
of points on
M.
28.1. From Riemann-Roch to Abel-Jacobi
D be a divisor on M ; we have been interested in the dimensions
vector spaces L(D) and I(D). In the interval
Let
of the
0 deg(D) 2g 2
is where anything of interest lies: outside this range, either
i(D)
(i)
(ii)
1each
`(D)
or
is zero. At the extremes, Abel's theorem will tell us:
`(D) when deg(D) = 0; and
i(D) when deg(D) = 2g 2.
intersection point contributes a
the orientation of
according to the right-hand rule and
given by the complex structure
28.1. FROM RIEMANN-ROCH TO ABEL-JACOBI
In case (i), if there is a meromorphic function
D 0,
305
f K(M )
with
(f ) +
then
deg((f ) + D) = deg((f )) + deg(D) = 0 + 0 = 0
=
(f ) + D = 0
In this event, there can be only one such
(f ) = D = (g)
rat
D 0.
(up to scale), as
(f /g) = 0
f /g
constant.
Together with similar reasoning in case (ii), and assuming Abel, this
argument proves
Proposition 28.1.1. (i) If
deg D = 0,
then
`(D) = 0
or
1;
and
rat
AJ(D) = 0 D 0 `(D) = 1.
(ii) If
deg D = 2g 2,
then
i(D) = 0
or
1;
and
rat
AJ(K D) = 0 D K `(D K) = 1 i(D) = 1.
Another point of contact with the last few chapters comes in the
context of canonical and hyperelliptic curves. First, x
qM
and look
at the mapping
uq : M J(M )
p
1
q.
.
p
7 AJ([p] [q]) =
.
p
g
q
Assuming Abel's theorem, we have (for genus
Proposition 28.1.2. (a)
uq
mod
M .
1)
is injective;
(b) its dierential yields the canonical map; and
(c) if
is hyperelliptic and
is a xed part of
symmetric with respect to the involution
Proof. (a) Assuming
p1 6= p2
and
u 7 u
of
, then uq (M )
J(M ).
uq (p1 ) = u2 (p2 ),
we have
AJ([p1 ] [p2 ]) = 0
Abel
f K(M )
f : M
P1
(=)
contradicting
g 1.
with
(f ) = [p1 ] [p2 ]
has degree one,
is
306
28. ABEL'S THEOREM, PART I
(b) Given
1 (M ),
we can consider
(p) Tp M.
By the funda-
mental theorem of calculus, the dierential
duq (p) : Tp M Tuq (p) J(M )
= Cg
loc
(1 (p), . . . , g (p)). (That is, if i = fi (z)dz , then duq (p)
sends
| 7 (f1 (0), . . . , fg (0)) Cg .) This associates a line in Cg to
z p
g1
each p M ; projectivizing clearly recovers K : M P
.
(c) Using ((x, y)) = (x, y), we have
p
(p)
dx
dx
y
y
q
q=(q)
.
.
.
.
=
uq ((p)) =
.
.
(p) xg1 dx
p xg1 dx
y
q
y
(q)
p
q dx
y
.
= uq (p).
.
=
.
p xg1 dx
q
y
is given by
In fact, in the hyperelliptic case it is clear from (c) that the xed
points of
map to
2-torsion
points of
J(M ).
28.2. Dierential forms of the third kind
There is a classical (and pass) terminology for meromorphic dierential forms on a Riemann surface: rst kind refers to holomorphic
forms; second kind to meromorphic forms with trivial residues (and
hence no simple poles); and third kind to everything else.
In this
section we'll pursue a method for constructing functions with a given
divisor (if possible). The title refers to the essential use we shall make
of meromorphic forms with prescribed (nonzero) residues.
Given
p, q M
i([p] [q]) = g (2) 1 + `([p] [q]) = g + 1 (> g),
|
{z
}
0
so there exists
I([p] [q])\1 (M ).
0 =
By the residue theorem,
Res () + Resq ()
| p
{z
}
both nonzero since poles simple
28.2. DIFFERENTIAL FORMS OF THE THIRD KIND
and so we can normalize
Resp =
so that
,
2 1
Resq =
1
.
2 1
, write () = ()0 () where ()0 , ()
For any meromorphic form
307
are the zero- and polar-divisors.
D Div 0 (M ), there
X
D I
[p]
Lemma 28.2.1. Given
exists
p|D|
such that
() =
[p]
and
Resp =
p|D|
Proof. Write
Resqk k =
2 1
D=
. Then
P
[pk ] [qk ], and
P
add
k =: .
ordp (D)
.
2 1
pick
so that
Respk k =
P
D = nj [Pj ] and D be as in Lemma 28.2.1, and consider
2g
a collection {i }i=1 of closed paths with support |i | M \|D|, such
2g
that their classes {[i ]}i=1 H1 (M, Z) yield a basis.
Next let
Lemma 28.2.2. If
D Z (i),
(28.2.1)
then ( xing
Q M)
f (P ) := exp 2 1
D
yields a well-de ned function
f K(M )
with
(f ) = D.
Cj denote
0
around the Pj . Given two paths Q.P and Q.P ,
X
X
0
Q.P Q.P = +
mj Cj +
`i i
Proof. We must check independence of path. Let
cular paths
where
is a (real-2-dimensional) closed region in
D =
mj
dD =
D =
Cj
0 = 0,
mj nj Z
M \|D|.
Now
cir-
308
28. ABEL'S THEOREM, PART I
ResPk D =
since
n
k , and
2 1
D Z
`i
i
by assumption (28.2.1). So for some
exp 2 1
Q.P D
= exp 2 1 = 1,
exp 2 1
Q.P 0 D
and
For the divisor, let
neighborhood of
Pj
(with
be a holomorphic coordinate de ned in a
z(Pj ) = 0),
loc
=
with
M \|D|.
is well-de ned (and holomorphic) on
and write
n
dz
j
+ h(z)dz
2 1 z
holomorphic. Without loss of generality, we can assume that
lies in the neighborhood, with
(variable). Locally
z(Q) =: z0
f (z) = exp 2 1
( xed) and
D
Q
z
= exp 2 1
where
h(w)dw exp nj
z0
z0
= H(z)
z(P ) =: z
dw
w
exp (nj log z)
exp (nj log z0 )
is holomorphic and nonvanishing in our neighborhood (being
the exponential of something holomorphic). Finally, writing
H(z)
n , the above
z0 j
= H0 (z) z nj .
This makes it clear that
is meromorphic at
Pj
H0 (z) =
with
Pj (f ) = nj .
Doing this for each
j,
we conclude
(f ) =
nj [Pj ] = D.
EXERCISES
In the next chapter we will take
basis for
H1 ,
309
1 , . . . , 2g
to yield a symplectic
i.e.
i g+j = ij
delta)
20.1. It turns out
that the period vectors 1 , . . . , g associated to 1 , . . . , g are actually
2
linearly independent over C, and so the g g matrix they form is
invertible. Applying the inverse matrix to 1 , . . . , g , we may replace
them by {j } satisfying
for
1 i, j g.
(Kronecker
This is the situation pictured in
j = ij .
i
Given
and
as in Lemma 28.2.1, then, we can modify
f
D := D
g
X
D i
so that
i = 1, . . . , g .
(28.2.2)
to
i=1
for
f
D = 0
We will prove that
AJ(D) = 0 =
there exists a further modi cation
Pg
c
f
D :=
D +
j=1 j j
with
so as to a rm condition (28.2.1) (for
c
D Z (i = 1, . . . , 2g),
c
D ).
To attack (28.2.2), we need
the Riemann bilinear relations, our next topic.
The problems below are only loosely related to the material of his
chapter. The second one is rather open ended!
Exercises
(1) This problem, in which you will prove a version of Abel's theorem
for a singular cubic (not its normalization), is only loosely related
C as P1 with 0 identi ed to
dz
1
and coordinate z . We consider (C) to be spanned by
(even
z
1
though it isn't holomorphic) and H1 (C, Z) by S = unit curcle. Divisors must avoid the singularity, and meromorphic functions f
must have f (0) = f () 6= 0, .
to the chapter. Think of the cubic
2to
be proved in
29.1. Since the vectors are
R (which we already have).
independence over
non-real, this doesn't follow from
310
28. ABEL'S THEOREM, PART I
J(C)?
P
P
(b) Compute AJ(D) for D =
ni [zi ],
ni = 0.
(c) Show 0 = AJ((f )) J(C).
(a) What is
(d) Formulate and prove the injectivity statement (Abel's theorem). [Hint: the proof will use what you did in (c), even though
it's a converse, and so needn't be long.]
(2) Let
be a Riemann surface and
a (nonempty) nite set
of points.
(a) De ne divisors on the relative variety
ni [pi ] where no pi
(M, ) to be formal sums
; two of these are rationally equivalent
if their dierence is the divisor of f K(M ) which is 1 on all points
of . Construct an AJ map and Jacobian for (M, ). [Hint: the
1
case M = P , = {0, } should recover the results of exercise
lies in
(1).]
(b) Next consider the complement
Div(M )/Div(),
M \.
We de ne divisors by
and rational equivalence by taking divisors of
meromorphic functions on
(and ignoring any poles/zeroes on
since that information is quaotiented out). Construct an
and Jacobian for
M \.
AJ
map
[Hint: note that there is no such thing as
degree of a divisor, since the points in
eectively have arbitrary
coe cients. Or rather, using these points, you can make the degree
of any divisor zero! This should have some bearing on your choice
of path.]
CHAPTER 29
Abel's Theorem, part II
As suggested at the end of the previous chapter, on any Riemann
M,
surface
1 pairing on the level of homology
we can produce a perfect
h , i : H1 (M, Z) H1 (M, Z) Z
(29.0.3)
2 With respect to the basis
2g ) matrix3
by intersecting 1-cycles.
there, this has (2g
0 Ig
Ig 0
Q=
{j }2g
j=1
described
!
.
We can use this pairing to produce an isomorphism of dual spaces
(29.0.4)
H1 (M, C) = H1 (M, Z) C Hom(H1 (M, Z), C) = H 1 (M, C)
which is a special case of Poincar duality.
Recalling the isomorphisms
1
1 (M ) 1 (M ) HdR
(M, C) H 1 (M, C),
there is also a pairing (the cup-product)
H 1 (M, C) H 1 (M, C) C
1this means that the (bilinear) pairing is described (with respect to an integral basis
of
H1 (C, Z))
2More
, )
by an integrally invertible, i.e. unimodular, matrix.
precisely, any two given homology classes have representative 1-cycles (say,
which intersect transversely.
holomorphic coordinate
z = x + iy ,
p there is a local
v and v to the 1-
At each intersection point
and the tangent vectors
cycles (which have well-de ned directions) can be wedged to produce an element
V2
Tp M (p 6= 0). The intersection is called positively
y
or negatively oriented depending upon the sign of p , and the intersection number
h[], []i is the number of positive intersection points minus the number of negative
v v = p x
ones.
3having
this particular intersection form is the de nition of a symplectic basis of
H1 (M, Z)
311
312
29. ABEL'S THEOREM, PART II
induced on the level of 1-forms by
(, ) 7
.
M
Notice that since two holomorphic forms wedge to zero, this pairing
restricts to zero on
1 (M )1 (M ) (and 1 (M )1 (M )).
Yet another
pairing (the cap-product)
H1 (M, Z) H 1 (M, C) C
is induced by
(, ) 7
The restriction of this pairing to
the period matrix of chapter 20.
H1 (M, Z) 1 (M )
is caputred by
An important fact is that, under
(28.0.2), both of these integration-induced products are nothing but
complex-linear extensions of (28.0.1).
Assuming this compatibility, we can quickly derive the Riemann
bilinear relations as follows.
1 (M ),
If for any closed 1-form
we write
1 (M )
j () :=
,
j
then (28.0.2) identi es
[] =
(29.0.5)
g
X
(j ()[j+g ] j+g ()[j ])
j=1
H 1 (M, C),
1 (M )
in
i.e.
as functionals on homology.
0=
(29.0.6)
M
by writing
g
X
One has for
(j ()j+g () j+g ()j ())
j=1
= h[], []i
and expanding both classes as in
(28.0.3). Similar reasoning together with the local computation
idz d
z = i(dx + idy) (dx idy) = i(2idx dy) = 2dx dy,
leads to
(29.0.7)
= i
0<i
M
g
X
i=1
j ()j+g () j+g ()j () .
29.1. DERIVATION OF THE RIEMANN BILINEAR RELATIONS
313
This is all meant as motivation, though it can be made completely
rigorous.
We'll start the rst section with a more concrete, classical
proof of (29.0.4-5), without the compatibility assumptions on the three
bilinear pairings.
29.1. Derivation of the Riemann Bilinear Relations
We start by cutting
open to get the fundamental domain, a
simply-connected closed region
g+j
q
p
fundamental
domain
g+j
q
p
po
F. (Only a piece of it is shown in
1
of F be xed. Given (M ),
p
u(p) :=
with boundary
in the interior
j
q
the picture.) Let
p0
4 function on
F.
p0
then yields a well-de ned (single-valued) holomorphic
If we take a second holomorphic form
1 (M ),
then
d(u) = = 0.
That is,
is a closed holomorphic form on
that
0 =
by Stokes's theorem.
with the consequence
d(u) =
u
F
Now, the picture above tells us that
is the
composition of paths
1 1
1 1
1 1
2g
g 2g g g+2
2 g+2 2 g+1
1 g+1 1 ,
4to be holomorphic on a closed set means that the function extends to a holomorphic
function on a slightly larger open set (which, in this case, would live on the universal
cover of
M ).
314
29. ABEL'S THEOREM, PART II
written right to left (with inverse meaning the reverse direction). So
the last integral becomes
g
X
=
{
(u(p) u(p )) +
{z }
j |
j=1
and, noting that
p0
q
p0
g
X
g+j
0<i
F
g+j
q
r0
r
q
this
g+j
i
, essentially the same computation yields
!
g
X
.
} =
u(i
) = i
| {z
Replacing
p
q
j=1
(u(r) u(r0 )) }
{z }
j+g |
by
d(u
)
g+j
j+g
j=1
So we have recovered (29.0.4-5).
2g
To reformulate this in matrix terms for any symplectic basis {j }j=1
g
1
th
of H1 (M, Z) and any basis {i }i=1 of (M ), notice that the (k, `)
entry of
2g
t
Q = 1
= g+1
is
g
X
0 Ig
Ig 0
2g 1
.
.
.
2g
.
.
g
2g
(j (k )g+j (` ) j (` )g+j (k ))
j=1
which is zero by (29.0.4); so
Q t = 0.
(29.1.1)
Similarly,
(29.1.2)
in the sense that
1 Q t Pi > 0
x R>0 for any x Cg .
x( 1Qt )
In particular,
the diagonal entries of (29.1.2) are positive real.
5recall
from Chapter 20 that
is the complex
g -vector
with
ith
entry
j (i )
29.1. DERIVATION OF THE RIEMANN BILINEAR RELATIONS
315
Remark 29.1.1. Consider any two symplectic integral bases
{j }
and
= {j0 }
(thought of as row-vectors), so that
0 = A
for some
A SL2g (Z).
Applying the basis
{i }
(viewed as a column-
vector of 1-forms) on the left yields
0 = A.
Furthermore, since both bases are symplectic we have
Q = t
and
Q = t 0 0 = t At A = t AQA;
that is,
A belongs to the symplectic group Sp2g (Z).
It is for this reason
that (29.1.1-2) are compatible with change of symplectic basis: e.g.,
assuming (29.1.1) we have
0 Qt 0 = AQt At = Qt = 0.
Now thinking of the
g 2g
period matrix as two
(29.1.3)
A B
gg
blocks, viz.
we have
Qt =
A B
Ig
Ig
A
t
B
!
= A tB B tA
and
= A t B B t A.
Qt
In these terms, (29.1.1) reads
A tB = B tA
(29.1.4)
while (29.1.2) becomes
(29.1.5)
If
0,
(v Cg ).
has nonzero kernel, then there exists
hence
that
v>0
1t v(At B B t A)
vA = 0
and
v = 0,
A
v Cg
satisfying
contradicting (29.1.5).
28.2.
Av =
It follows
is invertible, and so we have proved the statement on
independence asserted at the end of
C-linear
316
29. ABEL'S THEOREM, PART II
Applying
A1
to the left of
amounts to a change of the basis {i }
6
1
for (M ), viz.
A1 = A1
1
.
.
.
10
2g
.
.
.
2g
g0
g
If we apply it to (28.1.3), then we get
0 := A1 =
{i } is
Z ,
We can therefore always assume that
=
again as claimed in
28.2.
(
Ig
chosen so that
The bilinear relations (29.1.4-5) simplify to
(29.1.6)
Ig A1 B
Z = tZ
,
1(Z Z) > 0
which in particular tell us that the imaginary part
Im(Z) is a positive-
de nite, real symmetric matrix.
29.2. Proof of Abel's Theorem
With the holomorphic basis as normalized above, we can now quickly
establish (28.2.2) and hence (28.0.1). Write
0)
and let
:= f
D
be as
28.2,
D=
ni [Pi ] (with
ni =
so that
(29.2.1)
ResPi () =
and
n
i
(i)
2 1
= 0 (j = 1, . . . , g).
j
For each
k = 1, . . . , g
set
uk (p) :=
k
p0
F, and let be a 1-chain (sum of paths) with = D. Then noting
P
D = ni ([Pi ] [p0 ]), we have
X
X
k =
ni uk (Pi ) = 2 1
Resp (uk )
on
6here
the product of
and
p|D|
is just the integral
29.3. PROOF OF JACOBI INVERSION
317
which by the Residue Theorem
29.1
X
(j (k )g+j () g+j (k )j ()) = g+k ().
| {z }
| {z }
uk =
F
AJ(D) = 0
every k
If
for
jk
2g
X
k =
Using
k = jk
= mk +
mj (j = 1, . . . , 2g )
then there are integers
g
X
k .
mj
j
j=1
Z = tZ
and
such that
(from (29.1.6)), this
mj+g j+g (k ) = mk +
g
X
mj+g k+g (j ).
j=1
j=1
Now
:=
g
X
mj+g j
j=1
is still an element of
I(
p|D| [p]) satisfying (29.2.1). Moreover, for
k {1, . . . , g}
k+g ()
= k+g ()
g
X
mj+g k+g (j )
j=1
g
X
mj+g k+g (j ) = mk Z
j=1
and
k ()
= k ()
| {z }
0
g
X
j=1
mj+g k (j ) = mk+g Z.
| {z }
kj
By Lemma 28.2.2, exp 2 1
now gives a meromorphic function
with (f ) = D .
29.3. Proof of Jacobi Inversion
To show that
AJ
is surjective, we will study the image of a certain
class of (degree zero) divisor on
M,
namely those of the form
[p1 ] + + [pd ] d[q]
given some xed point
qM
and natural number
d.
obviously in 1-to-1 correspondence with unordered
Such divisors are
d-tuples
of points
318
on
29. ABEL'S THEOREM, PART II
M,
in other words with elements of the
dth
symmetric power
d copies
z
}|
{
M
M
Symd M :=
.
(p1 , . . . , pd ) (p(1) , . . . , p(d) )
Sd
In order to be able to use complex analytic techniques we need to put
the structure of a
d-dimensional
complex manifold on this.
d = 2 and consider C instead
2
of a (compact) Riemann surface. The symmetric square Sym C is the
quotient of C C by the involution (z1 , z2 ) 7 (z2 , z1 ). What causes
To get the idea of how this works, take
di culty is the locus consisting of its xed points, i.e.
line.
Take two small open sets in
Sym M ,
the diagonal
one which intersects the
diagonal and one which does not:
z2
U
U
z1
Clearly
(z1 , z2 )
give local holomorphic coordinates on
U .
On
U ,
they are not well-de ned, but their elementary symmetric polynomi-
2 (z1 , z2 ) = z1 z2 are. Moreover, these
functions generate all polynomials in z1 , z2 which are invariant un2
der the involution and hence well-de ned on U Sym C. Taking
(w1 , w2 ) := (z1 + z2 , z1 z2 ) as the holomorphic coordinates there,9 the
transition function is then just (1 , 2 ). To see that this is invert2
ible on U , notice that in U the diagonal is de ned by w1 = 4w2
2
(since z1 = z2 (z1 + z2 ) = 4z1 z2 ). Since U avoids this locus
p
w12 4w2 is well de ned there and we can
(and is simply connected),
als
1 (z1 , z2 ) = z1 + z2
and
7elements are written either p + + p or {p , . . . , p }
1
d
1
d
8Had we started with M itself of dimension > 1, its symmetric
powers would be
singular complex analytic spaces, hence not manifolds. So what happens next is
special for
dim(M ) = 1.
9we could in fact take these as global coordinates, but this situation won't generalize
to
29.3. PROOF OF JACOBI INVERSION
de ne
by
z =
1
z =
2
319
w12 4w2
2
w2 w12 4w2
2
w1 +
More generally, in a neighborhood of
{q1 , . . . , q1 ; . . . ; q` , . . . , q` } Symd M
| {z }
| {z }
k1 times
(where
P`
j=1
kj = d),
k` times
the local coordinate system is given in terms of
holomorphic coordinates
zj
on
qj ,
near each
by
{p1 , . . . , pk1 ; . . . ; pdk` +1 , . . . , pd } 7
| {z }
{z
}
|
all near q1
all near q`
(1 (z1 (p1 ), . . . , z1 (pk1 )) , . . . , k1 (z1 (p1 ), . . . , z1 (pk1 )) ; . . . ;
1 (z` (pdk` +1 ), . . . , z` (pd )) , . . . , k` (z` (pdk` +1 ), . . . , z` (pd ))) .
Inelegant, but it gets the job done.
Now let
be any divisor of degree
on
M,
and consider the map-
ping
D : P(L(D)) Symd M
which sends (for
f L(D))
[f ] 7 (f ) + D.
(f ) + D 0 by de nition, and deg((f ) + D) = deg D = d; so
(f ) + D is of the form [p1 ] + + [pd ]. The map sends to the projective
equivalence class [f ], i.e. f up to a constant multiple, to {p1 , . . . , pd }.)
(Here
Lemma 29.3.1.
is (a) injective and (b) holomorphic.
Definition 29.3.2. The linear system
10
|D| consists of all eective
M rationally equivalent to D. The Lemma evidently realizes
|D| = image(D ) as a subvariety of Symd M isomorphic to P`(D)1 .
divisors on
Proof. (of Lemma)
(f ) + D = (g) + D = (f ) = (g) = (f /g) = 0 = f /g
constant = [f ] = [g].
(b) To show D holomorphic in a neighborhood of [f0 ], augment f0
P`(D)
to a basis {f0 , f1 , . . . , f`(D) } L(D) and write f := f0 +
j=1 j fj
(a)
10the
notation
support of
D,
|D|
is unfortunately standard for both the linear system and the
two completely dierent concepts!
320
29. ABEL'S THEOREM, PART II
`(D)
{j }j=1 are the local holomorphic coordinates (on some small
U L(D)). Let p |D| |(f0 )|, with open neighborhood Np M and
local coordinate z (i.e. ordp (z) = 1). Set k := ordp (f0 ) + ordp (D), and
Wf0 ,p := Symk Np with coordinates 1 (z1 , . . . , zk ), . . . , k (z1 , . . . , zk ).
so that
We must show that the composition
U O(Np ) Wf0 ,p , Ck
7 f z ordp D 7
f z ordp D
1 z(p1 ()), . . . , z(pk ())
..
.
Np
z(p1 ()), . . . , z(pk ())
p1 () + + pk ()
is holomorphic, which in turn boils down to the statement that each
is holomorphic in each
j .
implicit function theorem; for
For
k = 1,
k > 1,
this is the holomorphic
it is this together with Rouch
and the Riemann extension theorem in a manner familiar from previous
chapters.
d divisor D (viewed as an
D = [p1 ] + + [pd ] with
Definition 29.3.3. An eective degree
element of
the
{pj }
Sym M )
is called general
distinct points of
M.
Now look at the Abel-Jacobi mapping
ud : Symd M J(C)
d
X
[p1 ] + + [pd ] 7 AJ
!
[pj ] d[q] ,
j=1
where
q M
is xed.
This is shown to be holomorphic by using
the fundamental theorem of calculus at general
Osgood and Riemann extension theorems.
taking a local lifting of the image of
The next result does not require
Lemma 29.3.4. The ber of
ud
u
D
to
D,
then applying the
(Boundedness is clear by
.)
to be general.
over
ud (D)
is
|D| (
= P`(D)1 ).
u for ud .)
Abel
u1 (u(D)) |D|: u(E) = u(D) = AJ(E D) = 0 = E D
is the divisor of some f K(M )
= (f ) + D = E 0 (since
d
E Sym M ) = f L(D) = E = D (f ) image(D ) = |D|.
Proof. (For simplicity write
29.3. PROOF OF JACOBI INVERSION
u1 (u(D)) |D|:
Given
E |D|,
there exists
321
f L(D)
such that
rat
E = (f ) + D = E D = (f ) 0 = 0 = AJ(E D) =
u(E) = u(D) = E u1 (u(D)).
D = [p1 ]+ +[pd ] is general, then writing zj for local coordinates
about each pj ,
(dud )D : TD Symd M Tu(D) (J(M ))
If
is computed by the matrix
z1
Pd zi
i=1
.
.
.
P
d
i=1
zd
zi
q
z1
1
..
zd
f1 (p1 )
.
.
.
..
f1 (pd )
where
g1
d=1
.
{p1 ,...,pd }
loc
is the canonical map and
. (For
i=1 q
.
.
.
Pd zi
i=1 q g
g
pj ) i = fi (zj )dzj , this
fg (p1 )
^
K (p1 )
.
.
,
.
.
=
.
.
^
fg (pd )
K (pd )
If we write locally (about each
Pd zi
g
^
K (pj ) C
is a lift of
K (pj )
this is just Proposition 28.1.2(b).) From this we see
that
rank
(dud )D = dim (span(K (p1 ), . . . , K (pd ))) + 1,
where span means the projective linear span in
Pg1 .
Taking
d = g,
we now have the following claim:
((dug )D ) = g for D = [p1 ]+ +[pg ] Symg M
d
Zariski open subset of Sym M .
Lemma 29.3.5. rank
generic,
11 i.e. in some
Proof. Choose
all of
g1
p1 , . . . , pg distinct with span (K (p1 ), . . . , K (pg )) =
. This is possible since the canonical map is always nonde-
g
generate by Theorem 27.3.2(a). Consequently rank ((du )D )
this holds more generally for
cause its failure is equivalent to
= g,
in an algebraic open set. This is be-
det(dug ) = 0,
which is an algebraic
condition which will hold on some codimension-one subvariety.
11general
may not be quite enough
subvarieties of
Symg M
and
may have to avoid a larger number of
then just the ones where two or more
pj 's
coincide.
322
29. ABEL'S THEOREM, PART II
Theorem 29.3.6. [Jacobi inversion]
ug
is surjective and gener-
ically injective.
dug
Proof. By Lemma 29.3.5,
tangent spaces. So
Symd M
is generically an isomorphism of
takes an open ball about a general point
Symg M compact,
g
so image(u ) is both a closed analytic subvariety of J(M ) and contains
an open ball, and is therefore all of J(M ) (which is connected).
g
Since at a generic D , du is (in particular) injective, we see that
g 1
g
any such D is an isolated point of (u ) {u (D)}. But the latter is a
projective space by Lemma 29.3.4, and so the only way D is isolated
g 1
g
0
is if (u ) {u (D)} is isomorphic to P , i.e. is just D itself.
to an open ball. But
ug
is continuous and
Finally, to address (28.0.2) head-on, surjectivity of
AJ
follows from
the diagram
AJ
/ J(M )
OO
fMMM
MMM
ug
MM
D 7Dg[q] MM3 S
Div 0 (M )
(29.3.1)
Symg M
29.4. A nal remark on moduli
For any Riemann surface
tic basis of
for
(M )
H1 (M, Z),
(of genus
1)
with given symplec-
we know that there is a unique choice of basis
making the period matrix
over, we know by (29.1.6) that
imaginary part, i.e. belongs to the
of the form
Ig Z
. More-
is symmetric with positive de nite
g th
Siegel upper half space
Hg := {Z Mg (C) | Z = t Z, Im(Z) > 0}.
Note that
H1
The Jacobian
H, the familiar upper half plane.
J(M ) is, of course, the quotient of Cg
is just
by the lattice
. More
generally, let Z be any g g complex matrix such that
Ig Z has
R-linearly independent column vectors. Writing Z for their Z-span,
g
we de ne a complex torus by AZ := C /Z ; any complex g -torus is
given by integral linear combinations of the columns of
isomorphic to one of this form. A major result is the
29.4. A FINAL REMARK ON MODULI
323
Theorem 29.4.1. [Riemann Embedding Theorem]
AZ
is an
abelian variety (i.e., has a holomorphic embedding in projective space)
if and only if
belongs to
(Of course, any
Hg .
R-linearly
independent from
is in the upper
or lower half plane, so every complex 1-torus is algebraic; already for
g =2
this is false.) You can nd (eectively) two proofs in Gri ths
and Harris, one using generalized theta functions and the other using
Kodaira's embedding theorem.
For us, the implications of this theorem are:
(a)
Jacobians of Riemann surfaces of genus
eties of dimension
(b)
g;
Corollary 29.4.2. For
for
g 4,
For
are abelian vari-
and
abelian varieties of dimension
Since Riemann surfaces of genus
have
g(g+1)
moduli.
2
g 2 have 3g3 moduli, we conclude:
g < 4,
all abelian
g -folds
are Jacobians;
most of them are not.
g 4,
then, we have the problem of characterizing the Jaco-
bian locus in the moduli space
Hg /Sp2g (Z),
which is the (very di -
cult) Schottky problem. There are recent results describing this locus
in terms of the vanishing of theta functions.
12
12for example, see the seminar talk [Link]
Appendix: genera of singular curves
Start with an irreducible projective algebraic curve
C P2
of degree
d,
de ned over
C.
We know how to piece the local normal-
izations about singular points together with the smooth part of
construct a Riemann surface
C ,
to
together with a map
: C P2
with image()
= C.
The genus formula of Chapter 14, derived from
a generic stereographic projection and the Riemann-Hurwitz formula,
said that
=
g(C)
if all singularities of
are
(d 1)(d 2)
(if any) are ordinary double points and there
such points.
Here we would like to be able to compute the genus of the normalization of an arbitrary irreducible curve, with singularities of any
order and type. There exist formulas when the singularities are ordinary,
13 but my preference is for methods over formulas, particularly
14
There are
when the methods allow you to treat more general cases.
two methods: the rst one computes the divisor of the pullback of a
P2 to C and applies Poincar-Hopf;
C to a line and applying Riemann-
meromorphic dierential 1-form on
the second is based on projecting
Hurwitz (like in the proof of the genus formula). Rather than trying to
state them formally, I'll use both methods to treat an example which
is su ciently general that you'll be able to adapt the approaches to
any other curve.
13cf. 6.4
for the de nition of an ordinary
k -tuple
the formula is Fulton's book on algebraic curves.
point. One book which proves
14and particularly when you can't remember formulas (welcome to my world)
325
326
APPENDIX: GENERA OF SINGULAR CURVES
So here is the ugly curve we will study: put
F (Z, X, Y ) := X 3 Z 3 + X 6 + Y 5 Z,
and
C := {F (Z, X, Y ) = 0} P2 ,
with a ne form
x3 + x6 + y 5 = 0.
This is a degree
a smooth curve of this degree has genus
10.
6 (i.e.
sextic) curve;
That will not be the answer
here.
One immediately obvious singularity is at
projective coordinates
[Z : X : Y ]).
(0, 0)
(i.e.
[1 : 0 : 0]
in
The lowest-order homogeneous
term (of the a ne equation, in coordinates vanishing at this point) is
x3 .
[1 : 0 : 0] is a triple point, but very de nitely not
point of C . Ugly enough? Well, this turns out to
So
triple
an ordinary
be the only
singularity.
Method I: Poincar-Hopf
Set
dx
y
K1 (C)
(it will actually turn out to be in
although this is inessential for the method).
,
1 (C)
Poincar-Hopf tells us
. So we have to compute
deg(()) = 2g 2, where g = g(C)
P
Where might these {pi } lie in C ? Or rather,
() = mi [pi ] Div(C).
where might the {(pi )} lie on C ? There are four (not necessarily
that
disjoint) possibilities:
(1) on the intersections of
(2) at points where
(3) at singularities of
with the
x-axis,
i.e. in
has a vertical tangent, hence
C,
C {Y = 0};
in C {FY = 0};
i.e. in Sing(C); and
(4) on the line at in nity, i.e. in
C {Z = 0}.
p () at p in one of these sets? For
dx
is zero on the line Y = 0; while for (2)
(1), the denominator of
y
Why might one expect nontrivial
the pullback of
dx
will be zero, since at such a point the curve has no
horizontal variation to rst order. You should always be suspicious
of (3) and (4). Conversely: on the smooth a ne part of
C , dx
and
never blow up, and (1) and (2) are the only ways they can develop a
zero. So (1)-(4) are actually the only places where
can have a zero
or pole.
Now we go through these 4 sets of points for the particular curve
under consideration.
METHOD I: POINCAR-HOPF
327
(1): we look at the a ne equaton and set
y = 0, which yields
1
(Here, 6 = exp(
).)
3
x3 + x6 = 0, hence x = 0, 6 , 6 , or 1.
While (0, 0) is a singular point and will be dealt with below, it is clear
dx
| will behave in the same way near the remaining three points:
that
y C
(1, 0), (6 , 0), and (6 , 0). We look in a neighborhood of (1, 0) on
C . Setting = x + 1 (or x = 1), the equation becomes in (, y):
0 = y 5 + ( 1)3 + ( 1)6 = y 5 3 + {higher-order
terms in
= y 5 3h(),
h(0) 6= 0.15 The local normalization
of C at (, y) = (0, 0) is
p
5
5
5
therefore t 7
t , t 3h(t ) , under which dx
= d
pulls back to
y
y
where
5)
d(t
3h(t5 )
5dt
= t3
5
3h(t5 )
which has a zero of order
at
t = 0.
So we
conclude that
1 [Link] () = 3,
and similarly that
1 [1:6 :0] () = 1 [1:6 :0] () = 3.
(2): for vertical tangents or singularities we will have
0 = FY = 5Y 4 Z,
x-axis or along the line at . The
intersections with the x-axis other than [1 : 0 : 0] were just dealt with.
Any nonsingular intersections with {Z = 0} will be dealt with in step
so that these must occur along the
(4). So vertical tangents are subsumed under the other three categories.
(3): at a singular point we must have
0 = FY ,
0 = FX = 3X 2 Z 3 + 6X 5 = 3X 2 (Z 3 + 2X 3 ),
and
0 = FZ = 3X 3 Z 2 + Y 5 .
Z = 0 or Y = 0. If Z = 0 then the last two equations
imply X = Y = 0, a contradiction. If Y = 0 then the last equation
gives Z = 0 (no!) or X = 0; the latter works, and so [1 : 0 : 0] is
the only singular point. In local coordinates about (x, y) = (0, 0), our
5
3
6
5
3
curve is 0 = y + x + x = y + x h(x) (dierent h(x) from above,
We must have
15Sometimes
(though rarely) one may have to remember more about
types of problems, but not in this example.
in these
328
APPENDIX: GENERA OF SINGULAR CURVES
again
h(0) 6= 0),
which is locally irreducible and has a singularity of
order 3. Undeer the local normalization
back to
d(t5 )
t3 5
h(t5 )
5dt
=t
5
5
h(t )
t 7 (t5 , t3
p
5
h(t5 )),
dx
pulls
y
, and we conclude that
1 [Link] () = 1.
(4):
C {Z = 0}
is the single point
[0 : 0 : 1].
We will need to
switch to a ne coordinates vanishing at this point, namely
v=
x
y
X
(or conversely
Y
y=
1
,
u
x=
u=
1
y
Z
,
Y
v
):
u
y
x
We divide
Z 3 X 3 + X 6 + Y 5 Z = 0 by Y 6 , obtaining
3 3 6
X
X
Z
Z
+
+ =0
Y
Y
Y
Y
v 6 + u3 v 3 + u = 0
which is a locally irreducible Weierstrass polynomial in
with (multi-
valued) roots of the form
s
v (u) :=
u3
(Use the quadratic formula to solve for
stituting in
t6
u6 4u
2
v3,
then take cube root.) Sub-
gives
36
6
18 + t3
+ t 4t
t30 4
3 t
3
6
=
v(t ) =
2
2
s
15 +
t30 4
3 t
=t
2
which is just t times some local analytic H(t) with H(0) 6= 0. So
v
d( u
)
dx
6
the normalization is t 7 (t , t H(t)) and
pulls back to
=
1
y
u
tH(t)
d( 6 )
t
= t6 d H(t)
= (tH 0 (t) 5H(t)) dt, which has neither zero nor
1
t5
s
t6
t18
METHOD II: RIEMANN-HURWITZ
pole at
t = 0.
329
Hence,
1 [Link] () = 0.
Upshot: putting everything together,
() = [ 1 [1 : 0 : 0]]+3[ 1 [1 : 6 : 0]]+3[ 1 [1 : 1 : 0]]+3[ 1 [1 : 6 : 0]]
2g 2 = deg(()) = 1 + 3 + 3 + 3 = 10
g = 6.
Method II: Riemann-Hurwitz
Recall that this dealt with maps of Riemann surfaces
f : M N,
and told us that
M = deg(f ) N rf .
Here
deg(f )
is the mapping degree of
(the number of points in the
N ) and rf is the degree of the rami Rf := pM (p (f ) 1)[p].
2
, N = P1 , =stereographic projection
Now let q P \C , M = C
(P2 \{q}) P1 through q ; and take
preimage of a general point on
16
cation divisor
f : C P1
f := . Usually it is
[0 : 0 : 1] as q . In our case the
[1 : 0 : 0],
these not on C
to be given by
easiest to take
[0 : 1 : 0], or
is [0 : 1 : 0]. So
only one of
16locally
our projection looks like
about any
coodinates
z 7 z p (f )
z resp. w
= w.
p M and its image f (p) N , one has local holomorphic
(with z(p) = 0 resp. w(f (p)) = 0), in which f takes the form
330
APPENDIX: GENERA OF SINGULAR CURVES
P
C
[Link]
(forgets xcoord.)
and the mapping degree is the number of intersection points of
and
P1
{x + x + y = 0} for general y0
= 2 2 0 = 2, so we have
i.e.
deg(f ) = 6.
{y = y0 }
Obviously
2 2g = C = 6 2 rf = 12 rf
1
g = rf 5.
P2
So we will have to compute Rf =
mi [pi ] (or at least rf ) and the rst
issue to resolve is where the (pi ) can lie on C :
=
{F = 0} {FX = 0});
i.e. [1 : 0 : 0] for our
(1) points having horizontal tangents (subset of
(2) singular points
(FX = FY = FZ = 0)
example; and
(3)
L C
(1):
i.e.
[0 : 0 : 1]
in our example.
0 = FX = 3X 2 (Z 3 + 2X 3 )
X = 0,
we
x= X
Z
has solutions other than
which corresponds to the singular point. Namely, writing
6
6
1
3 ,
3 ,
3 . Plugging this into the a ne
2
2
2
52
53
54
5
1
1
5
equation of C yields y =
hence y =
5 ,
5 ,
5 ,
5 ,
5 . These
4
4
4
4
4
4
are independent of the choice amongst the 3 values for x, and so we
get
x3 +
1
2
=0
hence
x =
5 3 = 15 rami cation points. As you may check, the intersections
between F = 0 and FX = 0 at these points are all of rst order, hence
get
correspond to rami cations of order 2 and so make a contribution of
p (f ) 1 = 2 1 = 1 each to rf .
(2): Near (x, y) = (0, 0), the composition
t 7 (t5 , t3 H(t)) 7 t3 H(t)
EXERCISES
has
p (f ) = 3
(3): Near
hence contributes
to
331
rf .
(u, v) = (0, 0),
t 7 (t6 , tH(t)) 7 t6 .
| {z }
u,v
This is because
is supposed to take the
y -coordinate,
which is
1
u
here; but we have to compute the image in a holomorphic coordinate
p = [0 : 0 : 1]. So in fact u is the correct
map indeed has p (f ) = 6 and contributes 5 to rf .
rf = 15 1 + 2 + 5 = 22
vanishing at the image of
variable, and the
Conclusion:
g =
22
5 = 6,
2
con rming the previous computation.
Exercises
(1) Find the genus of the normalization of the curve
a ne equation
2 3
x +y +y x
2 5
y
5
= 0.
C P2
with
Do this in 2 dierent
ways: (a) by using an appropriate projection, computing the degree
of the rami cation divisor, and applying Riemann-Hurwitz; (b)
by computing the divisor of the pull-back of a meromorphic 1form on the normalization and applying Poincare-Hopf (try it with
dy
). [Hint: to do the local normalizations, rst make sure you are
x
dealing with an irreducible Weierstass polynomial
f (x, y) = 0
you may have to change variable, swap coordinates, divide out by
a unit (which can reduce the degree of the equation!), factor into
irreducibles, whatever. If you can't nd a multivalued solution
y(x)
by taking roots, using quadratic equation, and so on, you can always
f (x, y) is an irreducible Weierstrass polynomial
k
k
of degree k in y , then try substituting in t for x: write 0 = f (t , y)
and solve for y as a power-series in t, call this G(t). Then the local
k
normalization is t 7 (t , G(t)).]
use power series. If
You might also like
- Exactly Solvable Schrödinger Equations With Singularities: A Systematic Approach To Solving Complexified Potentials (Part1)100% (1)Exactly Solvable Schrödinger Equations With Singularities: A Systematic Approach To Solving Complexified Potentials (Part1)36 pages
- Some Smarandache Curves Constructed From A Spacelike Salkowski Curve With Timelike Principal NormalNo ratings yetSome Smarandache Curves Constructed From A Spacelike Salkowski Curve With Timelike Principal Normal12 pages
- (Ian W. Knowles, Roger T. Lewis, International Con (BookFi)No ratings yet(Ian W. Knowles, Roger T. Lewis, International Con (BookFi)401 pages
- Quaternion Equations and Quaternion Polynomial MatricesNo ratings yetQuaternion Equations and Quaternion Polynomial Matrices130 pages
- Introduction To Real Analysis Instructor Solution Manual 1nbsped 3030269019 9783030269012 - CompressNo ratings yetIntroduction To Real Analysis Instructor Solution Manual 1nbsped 3030269019 9783030269012 - Compress429 pages
- 01-S - Antoniou - Lie Symmetries, Differential Equations and Applications100% (1)01-S - Antoniou - Lie Symmetries, Differential Equations and Applications1,451 pages
- 1 Applied Asymptotic Analysis (Peter David Miller)No ratings yet1 Applied Asymptotic Analysis (Peter David Miller)489 pages
- (Undergraduate Texts in Mathematics) Stephanie Frank Singer-Linearity, Symmetry, and Prediction in The Hydrogen Atom-Springer (2005) PDF100% (4)(Undergraduate Texts in Mathematics) Stephanie Frank Singer-Linearity, Symmetry, and Prediction in The Hydrogen Atom-Springer (2005) PDF404 pages
- Hatcher A. - Algebraic Topology. Errata (Web Draft, Nov. 2004) PDFNo ratings yetHatcher A. - Algebraic Topology. Errata (Web Draft, Nov. 2004) PDF6 pages
- A "More Topological" Proof of The Tietze-Urysohn Theorem - Brian ScottNo ratings yetA "More Topological" Proof of The Tietze-Urysohn Theorem - Brian Scott3 pages
- Algebra Vol 4. Field Theory (Etc.) (Z-Library)100% (1)Algebra Vol 4. Field Theory (Etc.) (Z-Library)306 pages
- Onishchik, A. L. Lectures On Real Semisimple Lie Algebras and Ther RerpesentationsNo ratings yetOnishchik, A. L. Lectures On Real Semisimple Lie Algebras and Ther Rerpesentations96 pages
- John D. Baum - Elements of Point-Set Topology-Prentice-Hall (1964)No ratings yetJohn D. Baum - Elements of Point-Set Topology-Prentice-Hall (1964)168 pages
- Lectures On Groups of Transformations: J. L. KoszulNo ratings yetLectures On Groups of Transformations: J. L. Koszul85 pages
- Elements of Mathematics - Nicolas Bourbaki - 1989 - Springer-Verlag - 9783540193715 - Anna's Archive100% (1)Elements of Mathematics - Nicolas Bourbaki - 1989 - Springer-Verlag - 9783540193715 - Anna's Archive660 pages
- Lie Algebras by Nathan Jacobson Interscience Tracts in Pure and Applied Mathematics Number 10 John Wiley and Son Inc New York 1962100% (1)Lie Algebras by Nathan Jacobson Interscience Tracts in Pure and Applied Mathematics Number 10 John Wiley and Son Inc New York 19622 pages
- Notes On Mathematical Statistics and Data Analysis 3e by J.A. Rice (2006)No ratings yetNotes On Mathematical Statistics and Data Analysis 3e by J.A. Rice (2006)8 pages
- Group Rings, Crossed Products, and Galois TheoryNo ratings yetGroup Rings, Crossed Products, and Galois Theory77 pages
- (Lecture Notes in Mathematics #15) F. Oort - Commutative Group Schemes-Springer (1966)No ratings yet(Lecture Notes in Mathematics #15) F. Oort - Commutative Group Schemes-Springer (1966)140 pages
- An Introduction To Functional Analysis, Mischa Cotlar, Roberto CignoliNo ratings yetAn Introduction To Functional Analysis, Mischa Cotlar, Roberto Cignoli597 pages
- Modular Functions and Dirichlet Series in Number Theory by Tom M. Apostol (Auth.)No ratings yetModular Functions and Dirichlet Series in Number Theory by Tom M. Apostol (Auth.)216 pages
- Abstract Algebra - An Introduction With Applications (2ed.) by Derek J. S. Robinson 2015No ratings yetAbstract Algebra - An Introduction With Applications (2ed.) by Derek J. S. Robinson 2015347 pages
- Understanding Spectral Sequences in AlgebraNo ratings yetUnderstanding Spectral Sequences in Algebra18 pages
- Entropy and Perpetual Computing InsightsNo ratings yetEntropy and Perpetual Computing Insights9 pages
- Fourier-Mukai and Nahm Transforms in Geometry and Mathematical Physics (Claudio Bartocci, Ugo Bruzzo Etc.)100% (1)Fourier-Mukai and Nahm Transforms in Geometry and Mathematical Physics (Claudio Bartocci, Ugo Bruzzo Etc.)434 pages
- A Short Proof of The Tietze-Urysohn Extension Theorem - Mark MandelkernNo ratings yetA Short Proof of The Tietze-Urysohn Extension Theorem - Mark Mandelkern3 pages
- Lecture Notes in Linear Algebra 1 (80134) : Raz Kupferman Institute of Mathematics The Hebrew UniversityNo ratings yetLecture Notes in Linear Algebra 1 (80134) : Raz Kupferman Institute of Mathematics The Hebrew University342 pages
- Alexandru Gica - Rational Points On Elliptic Curves PDFNo ratings yetAlexandru Gica - Rational Points On Elliptic Curves PDF129 pages
- Wavelet Analysis Based Estimation of Probability Density Function of Wind DataNo ratings yetWavelet Analysis Based Estimation of Probability Density Function of Wind Data12 pages
- Advanced High-School - David B. SurowskiNo ratings yetAdvanced High-School - David B. Surowski435 pages
- Generalizing The Algebra of Physical Quantities: Mark F. SharlowNo ratings yetGeneralizing The Algebra of Physical Quantities: Mark F. Sharlow24 pages
- Hilbert Space Theory Problems & SolutionsNo ratings yetHilbert Space Theory Problems & Solutions91 pages
- Level 2 Remedial Supplementary Materials PackNo ratings yetLevel 2 Remedial Supplementary Materials Pack375 pages
- (BIO) Chapter 15 - Reproduction in PlantsNo ratings yet(BIO) Chapter 15 - Reproduction in Plants12 pages
- A Comparison of Programming Languages in Economics: An UpdateNo ratings yetA Comparison of Programming Languages in Economics: An Update4 pages
- Diversifying Stock Portfolios for Risk ManagementNo ratings yetDiversifying Stock Portfolios for Risk Management8 pages
- Understanding Forward Interpolation TechniquesNo ratings yetUnderstanding Forward Interpolation Techniques4 pages
- 7 Steps To Calculate Measurement Uncertainty - Isobudgets100% (1)7 Steps To Calculate Measurement Uncertainty - Isobudgets17 pages
- Exam Review Techniques for HUMMS StudentsNo ratings yetExam Review Techniques for HUMMS Students5 pages
- Finite Element Method Symmetry: Dr. D. R. PanchagadeNo ratings yetFinite Element Method Symmetry: Dr. D. R. Panchagade13 pages
- Ozonation of Organic Compounds in Water and WastewaterNo ratings yetOzonation of Organic Compounds in Water and Wastewater33 pages
- Leadership Styles As Predictors of Leadership Effectiveness Among Filipino Youth LeadersNo ratings yetLeadership Styles As Predictors of Leadership Effectiveness Among Filipino Youth Leaders16 pages
