Ims All

Elliptic Curves
– An Introduction –
Expanded notes from a mini-workshop held at
Mary Immaculate College

∼ University of Limerick ∼
29 and 30 November 2006
written by
Ciara Daly, Marina Franz, Madeeha Khalid and

Bernd Kreussler
Contents
1. Introduction 1
2. Solving Cubic Equations in Two Variables 7
3. Group law on the cubic curve 29
4. Theta Functions 53
5. Rank two vector bundles on elliptic curves 75
ELLIPTIC CURVES – AN INTRODUCTION
BERND KREUSSLER
The following four articles constitute expanded versions of talks

given during a mini-workshop which took place at Mary Immacu-
late College, Limerick, on the 29th and 30th of November 2006.
The titles of these talks were the following:
(1) Solving Cubic Equations in Two Variables.
(2) Group Law on the Cubic Curve.
(3) Theta Functions.
(4) Rank two vector bundles on elliptic curves.
Elliptic curves are very interesting because their study involves sev-
eral fields of mathematics. The study of elliptic curves has a long
history and still there are many unsolved problems. The goal of the
mini-workshop was to provide an introduction for the non-specialist
to several aspects of elliptic curves.
Elliptic curves reside at the crossroads of arithmetic, geometry and
analysis. This was reflected in the talks as follows: talk (1) dealt with
the arithmetic of elliptic curves whereas in talk (2) elliptic curves
were studied from the point of view of complex algebraic geometry.
The complex analytic side of elliptic curves was touched within talk
(3). After these basics were laid down, talk (4) gave an introduction
to the study of vector bundles on an elliptic curve. This highlighted
the fact that it is not only interesting to study elliptic curves on
their own but also to investigate other geometric objects of interest
constructed on them.
It is clear from the number of pages used that the four articles can
provide only a small bit of the available huge amount of knowledge
and techniques related to elliptic curves. None of the many applica-
tions in physics, engineering and modern communication technology
are discussed. To give the reader a first idea of the subject, a brief
description of included and excluded material is given below.
I would like to thank Pat O’Sullivan for acting as a critical reader
of the first drafts of all the four articles.
1
2 BERND KREUSSLER
Solving Cubic Equations in Two Variables (Bernd Kreussler)

The first article starts with the elementary question of finding all
Pythagorean triples of integers and goes on to apply similar ideas
in order to find integer solutions of equations of degree three in two
variables. The material is illustrated through many explicit exam-
ples. This part is probably suited for interested second-level students
(in fact there was one among the audience for the first talk). The
last section gives a brief overview of the most basic results about the
Mordell-Weil group of a cubic curve.
However, there are many things which are not even mentioned in
this article but which are no less important or fascinating than the
material included. In particular, zeta-functions and L-functions are
not included. As a consequence, the Birch and Swinnerton-Dyer con-
jecture is not formulated even though this is one of the Millennium
Prize Problems. The important method of infinite descent as well as
the Selmer and Tate-Shafarevich groups did not find their way into
the article. The very interesting connection of elliptic curves with
the solution of Fermat’s Last Theorem (through the Frey curve) is
another omission. The growing practical relevance of elliptic curves
in modern cryptography is another issue missing. This list is cer-
tainly not complete. A few books which may help the interested
reader to satisfy his or her thirst for knowledge are [6, 7, 13, 14, 15].
Group Law on the Cubic Curve (Madeeha Khalid)

The aim of the second article is to give an introduction to some basic
concepts from complex algebraic geometry which allow a geometric
understanding of the group structure introduced in the first talk. A
brief introduction is given to complex manifolds, vector bundles on
them and the Picard group (the group of all line bundles). Moreover
the relationship between line bundles and divisors on a curve is ex-
plained, which allows a better understanding of the group structure
introduced in the previous article.
The Weierstraß ℘-function and elliptic integrals are used to ex-
plain how complex analysis enters the picture. As a result, each
cubic curve can also be seen as a complex torus, which comes with
its own group structure. The gem of this article is a sketch of a
proof that this analytically defined group structure coincides with
the one introduced algebraically. This is based on Abel’s Theorem.
The analytic details are provided in the third article.
ELLIPTIC CURVES – AN INTRODUCTION 3
Again, many more things could have been included here. For
example, higher dimensional Abelian Varieties and the Abel-Jacobi
map which naturally emerge in the study of curves of higher genus are
not mentioned. The idea of a scheme over an arbitrary commutative
ring with unity are definitely beyond the scope of this article. To
introduce the ideas of a moduli space and of a universal object would
be a natural next step after the introduction of the Poincaré bundle.
A higher dimensional analogue of an elliptic curve would be a so-
called K3-surface. Their study has much in common with the theory
of elliptic curves but they couldn’t be touched either. There are
many excellent textbooks available, among which are [3, 5, 16].
Theta Functions (Marina Franz)

This article gives a brief introduction to some basics in the mod-
ern theory of elliptic functions. The starting point are theta func-
tions, which are nothing but global sections of line bundles on a
one-dimensional complex torus. Their main properties are investi-
gated from a purely analytic point of view. Moreover, these theta
functions are related to the Weierstraß ℘-function, which can be
considered to be the most basic elliptic function. A proof that this
function satisfies a certain differential equation is given. This equa-
tion shows that a complex torus of dimension one can be embedded
in the projective plane as a cubic curve. A proof of Abel’s Theorem,
which plays a major role in the previous article is also provided.
The same remark applies to this article as to the other two:
there is much more material available than could be included. For
example, an explicit description of the relationship between theta
functions and holomorphic line bundles on elliptic curves is missing.
Moreover, the fascinating theory of elliptic functions is only touched
on. In particular, nothing is said about elliptic integrals. These
arise, for example when the length of an ellipse is to be calculated.
Historically, the study of elliptic integrals motivated the introduction
of elliptic functions by Abel and Jacobi. Weierstraß built the theory
of elliptic functions on the ℘-function, but beforehand Jacobi’s ellip-
tic functions sn(z), cn(z), dn(z) were the main players. Their role in
mathematical applications in engineering are definitely beyond the
scope of this short article. Theta functions are available on higher-
dimensional tori as well, but this is not covered here. Such material
and much more can be found in [12, 1, 10, 9].
4 BERND KREUSSLER
Rank two vector bundles on elliptic curves (Ciara Daly)

In contrast to the three others, this fourth article is not primarily
concerned with the group structure on an elliptic curve. But it is a
direct continuation of these. Vector bundles of rank one and their
sections were studied in the previous two articles. The moduli space
interpretation of the Picard group is already mentioned in the second
article. This article presents the main results about vector bundles
of rank two on an elliptic curve. These go back to a seminal paper of
Atiyah from 1959. This example is used to introduce to the theory
of moduli, which is at the centre of modern algebraic geometry. The
related notion of a stable vector bundle is also introduced.
Of course, there is much more that could be said in this context.
Atiyah studied vector bundles of any rank, not only of rank two, but
this did not find its way into this article. Also, the problems involved
with the notion of stability of vector bundles on higher dimensional
manifolds are not discussed. The theory of moduli of varieties as
opposed to vector bundles is another huge area of algebraic geometry
which is omitted. The relations of algebraic geometry to differential
geometry and to theoretical physics through the theory of moduli
spaces are not mentioned. Another quite recent development was
the introduction of the space of stability conditions by Bridgeland.
To define this invariant it would be necessary to introduce coherent
sheaves and derived categories, so that this development could also
not be covered here. The interested reader will find relevant starting
points in [4, 8, 11, 17, 2].
References
[1] N.I. Akhiezer, Elements of the theory of elliptic functions. Translations of
Mathematical Monographs 79, AMS (1990)
[2] T. Bridgeland, Derived categories of coherent sheaves. Proceedings of the
international congress of mathematicians (ICM), Madrid, Spain, August
22–30, 2006, Volume II, 563–582 (2006)
[3] P. Griffiths, J. Harris, Principles of algebraic geometry. John Wiley & Sons
(1978)
[4] D. Huybrechts, M. Lehn, The geometry of moduli spaces of sheaves. Aspects
of Mathematics E 31, Vieweg (1997)
[5] F. Kirwan, Complex algebraic curves. London Mathematical Society Stu-
dent Texts 23, Cambridge University Press (1992)
[6] A.W. Knapp, Elliptic curves. Mathematical Notes (Princeton) 40, Princeton
University Press (1992)
[7] N. Koblitz, Algebraic Aspects of Cryptography. Springer (2004)
ELLIPTIC CURVES – AN INTRODUCTION 5
[8] J. Le Potier, Lectures on vector bundles. Cambridge Studies in Advanced

Mathematics 54, Cambridge University Press (1997)
[9] H. McKean, V. Moll, Elliptic curves. Function theory, geometry, arithmetic.
Cambridge University Press (1999)
[10] G. Mittag-Leffler, An introduction to the theory of elliptic functions. Annals
of Math. (2) 24, 271–351 (1923)
[11] S. Mukai, An introduction to invariants and moduli. Cambridge Tracts in
Mathematics 81, Cambridge University Press (2003)
[12] D. Mumford, Tata lectures on theta I, II, III. Reprint of the 1991 edition,
Modern Birkhäuser Classics, Birkhäuser (2007)
[13] J.H. Silverman, The arithmetic of elliptic curves. Graduate Texts in Math-
ematics 106, Springer (1986)
[14] J.H. Silverman, Advanced topics in the arithmetic of elliptic curves. Grad-
uate Texts in Mathematics 151, Springer (1994)
[15] J.H. Silverman, J. Tate, Rational points on elliptic curves. Undergraduate
Texts in Mathematics, Springer (1992)
[16] K. Ueno, An introduction to algebraic geometry. Translations of Mathemat-
ical Monographs 166, AMS (1997)
[17] E. Viehweg, Quasi-projective moduli for polarized manifolds. Ergebnisse der
Mathematik und ihrer Grenzgebiete (3. Folge) 30, Springer (1995)
Mary Immaculate College, South Circular Road, Limerick, Ireland

E-mail address: bernd.kreussler@mic.ul.ie
6
SOLVING CUBIC EQUATIONS IN TWO VARIABLES
BERND KREUSSLER
Abstract. After recalling a geometric construction of all

Pythagorean triples of integers, the same idea is applied to
find rational solutions of cubic equations in two variables.
This leads to the definition of the Mordell-Weil group. The fi-
nal section collects some of the basic properties of this group.
1. Pythagoras
The aim of this introductory section is to recall the well-known
geometric construction of all Pythagorean triples of integers. Three
integers a, b, c ∈ Z form a Pythagorean triple, if
a 2 + b 2 = c2 .
Almost everybody knows the Pythagorean triple (3, 4, 5) and many
know (5, 12, 13). However, not everybody has come across (8, 15, 17)
or (20, 21, 29).
Clearly, if n ∈ Z and (a, b, c) is such a triple, (na, nb, nc) will also
be one. In this way, starting with the well known triple (3, 4, 5) we
obtain (6, 8, 10), (−3, −4, −5), (12, 16, 20) etc.
Note that a prime number which divides two of the three inte-
gers in a Pythagorean triple automatically divides the third in the
triple. Therefore, it is enough to find all Pythagorean triples in
which any two of the three integers are co-prime. We shall call such
a Pythagorean triple reduced. Because the only Pythagorean triple
with c = 0 is (a, b, c) = (0, 0, 0), we shall assume in the sequel c 6= 0.
This allows us to introduce the new variables
a b
x= and y = .
c c
Using these coordinates, the search for reduced Pythagorean triples
translates into the problem to find all rational solutions of the equa-
tion
x2 + y 2 = 1.
7
8 BERND KREUSSLER
In other words, we would like to find all points on the unit circle
whose coordinates are rational.
The key observation is that a line which connects two points with
rational coordinates always has a rational slope. Therefore, we shall
look at all lines in the plane which pass through the point (0, −1)
and which have rational slope r ∈ Q.
y = 3x − 1
x2 + y 2 = 1
bc 3 4
5, 5
bc
(0, −1)
Such a line is given by the equation y = rx − 1. Therefore, the x-

coordinates of the two intersection points of this line with the unit
circle satisfy the equation x2 + (rx − 1)2 = 1, which is equivalent to
x (r2 + 1)x − 2r = 0. The solution x = 0 corresponds to the point
(0, −1). The second intersection point has coordinates
2r r2 − 1
x= 2 and y = 2 .
r +1 r +1

2r r 2 −1
The map which sends r ∈ Q to the point r2 +1 , r2 +1 on the unit
circle gives a parametrisation of the set of all rational points on this
curve. This completely solves our problem.
If we wish to derive a complete description of all Pythagorean
triples of integers, we start by writing the slope r as r = uv with co-
prime integers u, v. Using symmetry, we may assume r > 1. More
precisely, switching from r to −r corresponds to a sign change of
x, whereas a sign change of y is achieved by switching from r to
1
r . Thus, we assume u > v > 0 and u, v co-prime. Under these
SOLVING CUBIC EQUATIONS IN TWO VARIABLES 9
u
assumptions, r = v produces the point with coordinates
2r 2uv r2 − 1 u2 − v 2
x= = and y = = .
r2 + 1 u2 + v 2 r2 + 1 u2 + v 2
Now it is not hard to see that each reduced Pythagorean triples in
which a is odd can be written as

u2 − v 2 u2 + v 2
(a, b, c) = uv, ,
2 2
with u > v > 0, both odd and co-prime. Up to interchanging a and
b this gives us all reduced Pythagorean triples, because a and b are
co-prime, hence at least one of these to integers is odd. For small
values of u, v we obtain the following table
u v a b c u v a b c
3 1 3 4 5 7 5 35 12 37
5 1 5 12 13 9 1 9 40 41
5 3 15 8 17 9 3 27 36 45
7 1 7 24 25 9 5 45 28 53
7 3 21 20 29 9 7 63 16 65
2. A cubic example
The aim of this section is to find integer solutions of cubic equa-
tions by using the geometric idea used in the previous section. We
shall explain this method through the following example
b2 c = 4a3 − 4ac2 + c3 .
a
As before, we assume c 6= 0 and introduce new coordinates x = c
and y = cb in which the above equation becomes
y 2 = 4x3 − 4x + 1. (1)
This can be rewritten as
(y − 1)(y + 1) = 4(x + 1)x(x − 1).
In this form it is obvious that we have the following six solutions
(−1, ±1), (0, ±1), (1, ±1).
Question: Are these all the solutions of equation (1)?
It is not hard to produce a sketch of this curve in the real plane.
This can be done through the following step-by-step approach. First,
we draw the graph of the cubic polynomial 4x3 − 4x + 1. The inter-
section points with the x-axis can be found with Cardano’s formula.
10 BERND KREUSSLER
This polynomial has three real roots because its discriminant is posi-
tive. To get the second picture, we remove all points from the graph
which have negative y-coordinate. The next picture is produced
by applying the square root function. Finally, the cubic curve is ob-
tained by adding in the mirror image along the x-axis, because (x, y)
is on this curve if and only if (x, −y) is so.
bc bc bc
bc bc bc
The six marked points in the picture are the points we found before.
If we seek to find more rational points on this curve, we may try to
use lines with rational slope which pass through one of the known
points. This leads to a quadratic equation the solutions of which
correspond to two further intersection points of this line with the
curve given by equation (1).
bc
For example, the line with slope 1 which passes through the point
(0, 1) has the equation y = x+1. The x-coordinates of its intersection
with our curve are the solutions of the equation (x+1)2 = 4x3 −4x+1
or equivalently 4x3 − x2 − 6x = 0. The known solution corresponds
to the factor x of this polynomial. The two new intersection points

correspond to the solutions of the quadratic
√
equation 4x2 −x−6 = 0,
these are the irrational numbers 1±8 97 .
This example shows that we should allow for one new point only.
In other words, we should work with a line connecting two of the
known points.
Example 1. Let us see how this works with P = (0, 1) and Q =

(−1, −1). The line which connects these two points is given by the
equation y = 2x + 1. Substituting this into equation (1) gives 4x3 −
4x2 − 8x = 0. The two points we started with give us two of the
roots of this polynomial, namely x1 = 0 and x2 = −1. Now, it is not
hard to see that 4x3 − 4x2 − 8x = 4x(x + 1)(x − 2). Hence x3 = 2
is the third solution which corresponds to the point (2, 5) on our
curve. We can even produce another new point, because the given
equation does not change when we replace y by −y. This gives the
point S = (2, −5).
bc
bc
Pbc bcR
Q bc bc bc
y = 2x + 1
bc
S
Example 2. We may now continue by using the line through P =

(0, 1) and S = (2, −5). Its equation is y = −3x + 1. Therefore,
we look at 4x3 − (−3x + 1)2 − 4x + 1 which has to be equal to
4x(x − 2)(x − x3 ). Comparing the coefficients of x2 of these two
polynomials leads to the equation −9 = −4(2 + x3 ). This gives
1 1 1
x3 = 4 . The new points we obtain are 4 , ± 4 .
12 BERND KREUSSLER
In general, if we are using a line with slope r ∈ Q which passes

through two points on our curve whose x-coordinates are x1 and
x2 , we obtain the x-coordinate of the third point by comparing the
2
coefficients of x2 as above. The result will be x3 = r4 − x1 − x2 ∈ Q.
We may also use other points from the six found originally.
Example 3. The line connecting S = (2, −5) with R = (1, 1) has

the equation y = −6x + 7. This gives a new point with coordinate
x3 = 9 − 2 − 1 = 6 and y3 = −6x3 + 7 = −29. So we have two new
points (6, −29) and (6, 29) which are not visible in the picture.
Note that we obtained (6, 29) as follows. First we connected P =

(0, 1) and Q = (−1, −1) by a line, whose third point of intersection
with the cubic curve had (2, −5) as its mirror image relative to the
x-axis. Then we connected (2, −5) and R = (1, 1) by a line and
obtained (6, 29) as the mirror image of the third point of intersection.
It is interesting to see what happens if we carry out these steps in
another order. Let us first connect Q = (−1, −1) and R = (1, 1) by
a line, reflect the third point on this line on the x-axis and connect
this point in the second step with P = (0, 1).
Example 4. The line which connects Q = (−1,

1 1
−1) and R = (1, 1)
has the equation y = x. This line has 4 , 4 as its third point of
intersection with the curve given by equation (1). Therefore, we shall
connect its mirror image 14 , − 41 with P = (0, 1). The corresponding
line has the equation y = −5x + 1. The new point produced this
way is (6, −29), the same as we obtained in Example 3.
This coincidence is not an accident. It is in fact a special case

of a theorem from projective geometry which states that a cubic
curve (in projective space) which passes through eight of the nine
intersection points of two other cubics, must also contain the ninth
of these intersection points.
A closer look at examples 3 and 4 suggest that we are dealing here
with a kind of associativity. This can indeed be made precise by the
following definition.
Definition 5. Let P, Q be points on the cubic curve given by equa-

tion (1). We define P + Q to be the mirror image (relative to the
x-axis) of the third point of intersection of the line which connects
P and Q and the cubic curve.
bc
bc P
Q bc
bc
P +Q
In this language, we have shown above (P + Q) + R = P + (Q + R)

where P = (0, 1), Q = (−1, −1) and R = (1, 1). This definition also
extends to give P + P , which is obtained by using the tangent line
to our curve at P .
Example 6. Implicit differentiation reveals that the tangent line to

our curve at P = (0, 1) has slope equal to −2. Therefore, this line is
given by the equation y = −2x + 1. In the same way as before, we
substitute y = −2x + 1 into equation (1) and use the fact that x = 0
will be a double root of the cubic equation so obtained. Then, we
get that the x-coordinate of the new point of intersection is equal to
x = 1. This produces the known points (1, ±1).
This example shows that we actually need to know only one ra-
tional point on our cubic in order to get started. As before, we can
then produce many other points. Using the notation suggested by
Definition 5, we obtain here:
P = (0, 1), 2P = (1, 1), 3P = (−1, −1),

1 1
4P = (2, −5), 5P = , , 6P = (6, 29).
4 4
14 BERND KREUSSLER
b−4P
7P
bc
b bc
P bc2P
−3P
b
5P bc
3P bc b b−2P
b
−7P
bc4P
We shall see in the next section that this is in fact the structure
of an Abelian group in which for each point T , −T is the mirror
image of T with respect to the x-axis. The line which connects an
arbitrary point T on our cubic with its mirror image −T is a vertical
line. Because T + (−T ) = 0, we expect all these lines to go through
the neutral element of this group. Therefore, we shall look for the
neutral element “at infinity”. This can be made more precise with
the aid of the projective plane P2 , introduced in the following section.
3. The complete picture

In order to see all points on our cubic curve we have to return
to the original equation b2 c = 4a3 − 4ac2 + c3 . The key observa-
tion is here that (a, b, c) is a solution of this equation if and only
if (λa, λb, λc) is a solution for all numbers λ. This means that the
solution set is a union of lines which pass through the origin. When
we switched to coordinates (x, y) in the previous two sections, we
agreed that it is sufficient to know one point on each of these lines.
But we missed those lines on which c = 0 due to our division by c. If
we would like to keep these lines as well, we arrive at the idea of the
projective plane. Set theoretically, the projective plane is defined to
be the set of all lines in three-space which pass through the origin.
This leads to the following useful description.
Before we proceed we need to fix our notion of “number”. So far,
we have dealt with rational numbers and integers. But in general
it is much easier and more convenient to work with an algebraically
closed field like the field C of complex numbers. Many things which
will be said below are true for any field K. Therefore, we shall
formulate the next definition for any field K. The reader who is not
familiar with the concept of a field may substitute Q or C for K.
Definition 7. The projective plane P2 (K) over the field K is the set
of all equivalence classes (z0 : z1 : z2 ) of non-zero vectors (z0 , z1 , z2 ) ∈
K3 . Two such vectors (z0 , z1 , z2 ) and (w0 , w1 , w2 ) are equivalent if
and only if there exits a non-zero λ ∈ K such that (z0 , z1 , z2 ) =
λ(w0 , w1 , w2 ). This implies
(z0 : z1 : z2 ) = (λz0 : λz1 : λz2 ) for all λ 6= 0.
The notation (z0 : z1 : z2 ) for the equivalence class of the vector
(z0 , z1 , z2 ) is chosen in order to suggest that we are dealing with the
ratios between the three numbers z0 , z1 and z2 only. A similar con-
struction, of course, can be carried out in any dimension to produce
Pn (K) for all n ≥ 1. The one-dimensional case is particularly easy.
If K = C it leads to the Riemannian sphere. Notations used for the
Riemannian sphere are S 2 = C ∪ ∞ = C and P1 (C), the notation
we are going to use here. Its points are equivalence classes (z0 : z1 )
of non-zero vectors (z0 , z1 ) ∈ C2 . All points in P1 (C) with z0 = 0
are equivalent to ∞ = (0 : 1). Any point with z0 6= 0 is equiva-
lent to (1 : z) where z = zz10 . This gives a bijection between C and
P1 (C) \ {∞}. A neighbourhood of ∞ would be the set of all those
points of P1 (C) which have z1 6= 0. This is again in bijection with
C by using w = zz10 . The relationship between these two patches of
P1 (C) is given by w = z1 . This actually makes P1 (C) into a complex
manifold of dimension one, the simplest compact Riemann surface.
The local structure of P2 (K) can be studied in a similar way.
To this end, we define the three basic open sets which cover P2 (K)
completely
U0 := {(z0 : z1 : z2 ) | z0 6= 0} ⊂ P2 (K)
U1 := {(z0 : z1 : z2 ) | z1 6= 0} ⊂ P2 (K)
U2 := {(z0 : z1 : z2 ) | z2 6= 0} ⊂ P2 (K).
2
Each of these sets is in bijection with
K . For example, the map
U0 → K2 given by (z0 : z1 : z2 ) 7→ zz10 , zz02 has as its inverse the
map K2 → U0 which sends (ξ1 , ξ2 ) to (1 : ξ1 : ξ2 ).
z
Similarly, on U1 we can work with affine coordinates ηj = zj1 ,
j = 0, 2 and on U2 we have ζk = zzk2 , k = 0, 1. The gluing maps
16 BERND KREUSSLER
between these three K2 are given by

1 ζ1 η2 1
ξ1 = = ξ2 = =
η0 ζ0 η0 ζ0
1 ζ0 ξ2 1
η0 = = η2 = =
ξ1 ζ1 ξ1 ζ1
1 η0 ξ1 1
ζ0 = = ζ1 = = .
ξ2 η2 ξ2 η2
If K = C this defines the structure of a two dimensional complex
manifold on P2 (C).
Let us apply this new language to the cubic equation b2 c = 4a3 −
4ac2 + c3 studied in the previous section. As we have seen above,
if we identify (a, b, c) with (z0 , z1 , z2 ) ∈ Q3 , the set of all solutions
of this cubic equation is a well defined subset E(Q) ⊂ P2 (Q). Our
assumption c 6= 0 means that we restricted our attention to the set
U2 . The complement of U2 is the set of all those points which have
z2 = 0. These are the points of the form (z0 : z1 : 0), hence the
complement of U2 in P2 (K) is in bijection with P1 (K). Therefore, we
call
L2 = {(z0 : z1 : 0) | (z0 : z1 ) ∈ P1 (K)} ⊂ P2 (K)
the line at infinity. In a similar way we may define lines at infinity
L0 and L1 which are the complements of U0 and U1 respectively.
In order to see what we missed when restricting to U2 we simply
set c = 0 in our cubic equation. This leaves us with the equation
0 = 4a3 . Therefore, the only point missed is the point O = (0 : 1 :
0) ∈ L2 ⊂ P2 (Q). In order to see how E(Q) looks like around this
point, we restrict our attention to the set U1 . Using the coordinates
(η0 , η2 ) introduced above, E(Q) is described by the equation
η2 = 4η03 − 4η0 η22 + η23 .
The line at infinity L2 intersects U1 at the η0 -axis, given by the
equation η2 = 0. This line is a tangent line to the cubic curve with
a triple contact at the point O = (0, 0). The point O is an inflection
point of our curve.
The main result of the previous section was that we introduced
an “addition” of points in E(Q) by the rule that P + Q + R = O
if and only if the three points P, Q, R are collinear. Therefore, we
need to understand lines in the projective plane. These are given by
linear equations. In general, a line in P2 (K) is the set of all solutions
of an equation of the form
l0 z 0 + l1 z 1 + l2 z 2 = 0
with l0 , l1 , l2 ∈ K but not all three equal to zero. Because λl0 , λl1 , λl2
define the same line in P2 (K) if λ 6= 0, the set of all lines in P2 (K)
is another P2 (K), called the dual projective plane and sometimes
denoted P2 (K)∨ .
Each line in P2 (K) is isomorphic to P1 (K). The three lines at
infinity introduced before are also lines in this sense, because Lj was
given by the equation zj = 0. In particular, the line L2 corresponds
to the point (0 : 0 : 1) ∈ P2 (K)∨ . Any other line, with coefficients
(0 : 0 : 1) 6= (l0 : l1 : l2 ) ∈ P2 (K)∨ intersects U2 in an ordinary line.
The equation of this intersection is
l0 x + l1 y = −l2
where we used x = ac , y = cb instead of ζ0 = zz20 , ζ1 = zz12 as coor-

dinates on U2 . If l1 6= 0, this can be rewritten as y = rx + s with
r = − ll01 and s = − ll12 . If, however, l1 = 0 the equation becomes
l0 x = −l2 and this defines a vertical line which intersects the x-axis
at − ll12 .
On the other hand, the point O = (0 : 1 : 0) is on the line given by
l0 z0 + l1 z1 + l2 z2 = 0 if and only if l1 = 0. Hence, the vertical lines in
U2 correspond precisely to those lines in P2 (K) which pass through
O and are different from L2 . Therefore, the point at infinity O is
the correct choice for the neutral element of the group structure on
E(Q) and reflection at the x-axis corresponds to taking the additive
inverse of a point.
With some background in projective geometry or by other means
it can be shown that the addition of points on E(Q) introduced in
the previous section equips E(Q) with the structure of an Abelian
group. More about projective geometry and a geometric proof can
be found in the article by M. Khalid [11].
Theorem 8. E(Q) is an Abelian group with neutral element O, its

only point at infinity. The group structure is determined by saying
that P + Q + R = O if and only if P, Q and R are on a line in P2 (Q).
This implies that −P is obtained from P by changing the sign of the
y-coordinate.
18 BERND KREUSSLER
Remark 9. This result is true for any field K and any cubic equation
of the form
z12 z2 = z03 + pz0 z22 + qz23 (2)
3 2
with p, q ∈ K satisfying ∆ = −16(4p + 27q ) 6= 0. If the characteris-
tic of K is not equal to two or three (i.e. if 1 + 1 6= 0 and 1 + 1 + 1 6= 0
in K), every regular cubic with a point over K can be given by such
an equation. When working in characteristic zero (e.g. K = Q or
K = C), we can change coordinates so that (2) becomes
z12 z2 = 4z03 − g2 z0 z22 − g3 z23 . (3)
The discriminant of such an equation is ∆ = g23 − 27g32 . A cubic
equation of the form (3) is called a Weierstraß equation, named after
Karl Weierstraß (1815–1897). The coefficient 4 at z03 is used because
it appears in the differential equation of the Weierstraß ℘-function.
(See the article by M. Franz [5].)
The most basic structure result about the group E(Q) was shown
in 1922 by Mordell [17].
Theorem 10 (Mordell). If E is given by (2) with p, q ∈ Q and
4p3 + 27q 2 6= 0 then the Abelian group E(Q) is finitely generated.
Remark 11. Theorem 10 has been generalised by A. Weil to Abelian
varieties of arbitrary dimension over any number field [23]. Therefore
Mordell’s Theorem is also known as the Mordell-Weil Theorem and
the group E(Q) is sometimes called the Mordell-Weil group.
The curve studied in section 2 has E(Q) ∼ = Z with generator
P = (0, 1). The discriminant of this curve is ∆ = 37.
Remark 12. The assumption 4p3 + 27q 2 6= 0 in Mordell’s Theorem
is crucial. If 4p3 + 27q 2 = 0 the cubic polynomial x3 + px + q has
a multiple root and this gives rise to a singular point on the cubic
curve given by (2). This changes the situation completely, because
singular cubics are rational. More explicitly, suppose −4p3 = 27q 2
and p, q ∈ Q \ {0}. A straightforward calculation shows that, under
these assumptions,
2
3 3q 3q
x + px + q = x − x+ .
p 2p

3q
This implies that − 2p , 0 is a singular point of the cubic which
means that each line in P2 (Q) that passes through this point will
have at most one other intersection point with the cubic curve (2).
Just as in the case of the circle in section 1 this produces a bijection
between Q (the set of slopes) and all rational points on a singular
cubic apart from the singular point. This shows that the non-singular
rational points on a singular cubic form a group which is not finitely
generated.
bc
Example 13. Look at the singular cubic z12 z2 = 4z03 − 3z0 z22 +
z23 , which has discriminant ∆ = 33 − 27(−1)2 = 0. On U2 , using
coordinates x, y as before, its equation is
y 2 = 4x3 − 3x + 1.
From 4x3 − 3x + 1 = (x + 1)(2x − 1)2 we see that the singular point
has coordinates 12 , 0 . Any line with rational slope r through this
point will have equation y = r x − 12 , hence the new intersection
2
point will be found by solving r2 (2x − 1)2 = (x + 1)(2x − 1)2 .
Therefore, the coordinates of this point are given by
r2 − 4 r3 − 6r
x= and y= .
4 4
The main difference between this and the non-singular case is that we
cannot find such a closed formula for all rational solutions in the non-
singular case. This is explained by the involvement of transcendental
functions such as theta functions and the Weierstraß ℘-function (see
the article [5]).
20 BERND KREUSSLER
4. Further results
In this section we collect some general results known about the
Mordell-Weil group E(Q). We also discuss several normal forms of
plane cubic curves.
Let us look at a cubic curve given by an equation of the form
y 2 = x3 + px + q with 4p3 + 27q 2 6= 0. (4)
Such an equation is also called a Weierstraß equation or Weierstraß

canonical form. However, as we shall see below, it is not canonical.
To determine this we need to decide whether it is possible that two
curves given by a Weierstraß equation with different (p, q) can be
transformed into each other by a linear transformation of coordi-
nates. Consider the following.
Given two curves y 2 = x3 + px + q and ye2 = x e3 + pex
e + qe with
p, q, pe, qe ∈ Q, the only possible linear transformations of coordinates
with rational coefficients which transform one of these equations into
the other are of the form x e = λ2 x, ye = λ3 y with λ ∈ Q \ {0}. Such a
transform is successful if and only if we have pe = λ4 p and qe = λ6 q.
This can be used, in particular, to obtain integer coefficients p, q ∈ Z.
Therefore, the following result is useful in broader generality than it
first may seem.
Theorem 14 (Siegel, [20, 16, 18]). The equation y 2 = x3 + px + q

with p, q ∈ Z has only finitely many solutions (x, y) ∈ Z2 , provided
that 4p3 + 27q 2 6= 0.
A point P ∈ E(Q) is called a torsion point if there exists a positive

integer n ∈ Z such that nP = O in the additive group E(Q). In
other words, the torsion points of E(Q) are precisely the points of
finite order in the group E(Q). They form the torsion subgroup
E(Q)tor ⊂ E(Q). For example, if the curve is given by a Weierstraß
equation, the two-torsion points in E(Q), i.e. the points P ∈ E(Q)
with 2P = O, are precisely the intersection points of the curve E with
the x-axis (and the point O). The example studied in section 2 did
not have any two-torsion points apart from O, as the cubic equation
4x3 − 4x + 1 = 0 does not have a rational root. The following result
sheds some light on the torsion subgroup more generally.
Theorem 15 (Lutz–Nagell, [13, 19]). All torsion points of y 2 =

x3 +px+q with p, q ∈ Z have integer coordinates (x, y) ∈ Z2 , provided
4p3 + 27q 2 6= 0. Moreover, if (x, y) ∈ E(Q)tor then either y = 0 or

y 2 divides 4p3 + 27q 2 .
Together with Siegel’s Theorem this implies that E(Q)tor is fi-
nite. This, however, is already a consequence of Mordell’s Theorem,
because every finitely generated Abelian group G is isomorphic to
Zr × Z/a1 Z × Z/a2 Z × . . . × Z/as−1 Z × Z/as Z
| {z }
Gtor
with positive integers ai . The number r ≥ 0 is called the rank of this

group. The possibilities for the rank of E(Q) are not yet known, but
it is conjectured that there exist cubic curves for which the rank of
E(Q) is as large as you want. The largest known rank at the moment
seems to be 28, attained by an example found by N. Elkies in 2006.
On the other hand, the torsion subgroup of E(Q) is much better
understood. The main result is the following.
Theorem 16 (Mazur, [14, 15]). If E(Q) is given by the equation
y 2 = x3 + px + q with p, q ∈ Q and 4p3 + 27q 2 6= 0, then its torsion
subgroup E(Q)tor is isomorphic to one of the fifteen groups in the
following list:
Z/nZ, 1 ≤ n ≤ 10, or n = 12,
Z/2Z × Z/2nZ, 1 ≤ n ≤ 4.
All these groups in fact occur as torsion subgroups.
Remark 17. This result is in sharp contrast to the situation over an
algebraically closed field. If K is an algebraically closed field whose
characteristic does not divide the positive integer m, then the m-
torsion subgroup of E(K), which consists of all the elements of E(K)
killed by m, is isomorphic to Z/mZ × Z/mZ. In the case K = C this
will be explained in the article of M. Khalid [11].
Example 18. Because 9 = 8 + 1, it is not so hard to discover that
P = (2, 3) is an element of E(Q), the solution set of the equation
y 2 = x3 + 1. The tangent line at P to this cubic curve has equation
y = 2x−1. If we substitute this into y 2 = x3 +1 we obtain (2x−1)2 =
x3 +1 or equivalently 0 = x3 −4x2 +4x = x(x−2)2 . This means that
this tangent line intersects the cubic at the new point (0, −1), hence
2P = (0, 1). To find 3P , we use the line which connects P = (2, 3)
and 2P = (0, 1). It has the equation y = x+1 and intersects the cubic
at 3P = (−1, 0). This point is on the x-axis, so it is a two-torsion
22 BERND KREUSSLER
point. This implies 6P = O and we obtain 4P = −2P = (0, −1) and

5P = −P = (2, −3). In fact, E(Q) consists of the six points kP ,
k = 0, 1, 2, 3, 4, 5 only, i.e. E(Q) ∼
= Z/6Z. In [21] and [7] examples
of cubic equations which realise all the other E(Q)tor can be found.
The Tate canonical form, see (7) below, is very useful in the study
of cubics whose Mordell-Weil group has torsion. More precisely,
every cubic curve which has at least one point P ∈ E(Q)tor of order
at least four (i.e. P 6= O, 2P 6= O and 3P 6= O) can be brought into
Tate canonical form. For example, if b = 1 and c = d in (7), it can
be shown that the point (0 : 0 : 1) is a point of order four.
A useful method which allows us to gain information about E(Q)
is reduction modulo a prime number p. This means that one studies
solutions of a given cubic equation with coordinates in the finite field
Fp = Z/pZ. These solutions form the group E(Fp ). An interesting
result in this context says that for each prime p > 2 which does not
divide the discriminant ∆, the map which reduces the coordinates of
a torsion point P ∈ E(Q) modulo p embeds E(Q)tor as a subgroup
into E(Fp ). This can be used to determine the group E(Q)tor . More
on the issue of calculating the torsion subgroup of E(Q) can be found
for example in [22], [10] and [6].
Example 19. Let us show that E(Q)tor = {O} for the cubic y 2 =
4x3 − 4x + 1 studied in the previous section. The idea is to calculate
E(F3 ) and E(F5 ) and show that these groups are of co-prime order.
This is sufficient because 3 and 5 do not divide the discriminant
∆ = 37 of this cubic. If we reduce the equation y 2 = 4x3 − 4x + 1
modulo 3 we obtain y 2 = x3 − x+ 1. Because x3 − x = x(x− 1)(x+ 1)
is equal to zero for all x ∈ F3 , we see that (0, ±1), (1, ±1), (2, ±1)
are the only solutions of this equation with coefficients in the finite
field F3 . Therefore, E(F3 ) = {O, (0, ±1), (1, ±1), (2, ±1)} is of order
7. Reducing the equation y 2 = 4x3 − 4x + 1 modulo 5 gives y 2 =
−x3 + x + 1. Its solutions over F5 are (0, ±1), (±1, ±1) and (2, 0).
This means that E(F5 ) is a group of order 8. Because E(Q)tor is
isomorphic to a sub-group of E(F3 ) and of E(F5 ), it must be trivial.
More generally, solutions in all finite fields of fixed characteristic
p can be studied. If the number of solutions for such finite fields
are put together in a kind of generating function, the so-called zeta-
function is obtained. Lack of space forces us to skip the fascinat-
ing theory of zeta-functions of elliptic curves, the Weil conjectures
and their proof by Deligne and, last but not least, the Birch and
Swinnerton-Dyer conjecture which is one of the millennium prize

problems, a solution of which is worth one million US-Dollars (see
http://www.claymath.org/millennium/). A starting point for the
interested reader could be [12], [7] or [21]. We confine ourselves to
look at Weierstraß equations and other canonical forms over more
general fields K for the rest of this article.
There are several ways to proceed. One possibility would be to
introduce the abstract notion of a smooth projective curve of arith-
metic genus 1, defined over the field K. If such a curve has a point
with coordinates in K, it is possible to show that the curve is iso-
morphic to a plane cubic curve which has an inflection point at
O = (0 : 1 : 0). In particular, if K is algebraically closed, such a
point always exists. However, even in the case K = Q an equation
like 3z03 + 4z13 + 5z23 = 0 does not have a single point in P2 (K). Of
course, we shall not proceed along these lines. The interested reader
is referred to standard textbooks on algebraic geometry, such as [8].
We shall assume that we have a cubic equation f (z0 , z1 , z2 ) = 0
which defines a plane cubic curve with at least one point in P2 (K).
Let us first try to see whether any such curve can be described by
a Weierstraß equation, whereby we allow linear transformations of
coordinates only. The key to making progress is to understand in-
flection points. It is not hard to show that a point P ∈ E(K) is an
inflection point if and only if it is on the zero set of the Hessian of
the cubic polynomial f . By definition, the Hessian of f is the deter-
minant of the 3 × 3–matrix formed by the second partial derivatives
of f . This is again a cubic polynomial and Bézout’s Theorem implies
that there are at most 9 inflection points (with coordinates in the
algebraic closure of K). As we have seen earlier, the only point at
infinity O of a curve, which is given by a Weierstraß equation, is an
inflection point. Therefore, a necessary condition for a cubic to be
transformable to a Weierstraß equation is that at least one of the
inflection points is defined over K. Let us assume such a point exists
on our curve. By a linear transformation of coordinates with coeffi-
cients in K we can arrange that this inflection point has coordinates
(0 : 1 : 0) and the tangent line to the curve at this point is the line
at infinity with equation z2 = 0. Under these assumptions and using
coordinates x, y on U2 ⊂ P2 (K), it is clear that the cubic equation is
of the form
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 .
24 BERND KREUSSLER
In order to simplify the left hand side to y 2 we have to complete

the square, which means that y + a1 x+a 2
3
is going to be the new
y-coordinate. This obviously requires that we are able to divide by
2 which is possible in full generality only if the characteristic of K
is not equal to 2. On the other hand, in order to absorb the term
a2 x2 on the right hand side we complete the cube. This is possible
in general only if the characteristic of K is not equal to 3. As a
result we obtain that any non-singular cubic with an inflection point
in P2 (K) can be given by a Weierstraß equation if the characteristic
of K is not equal to 2 or 3. Moreover, if the characteristic of K is
not equal to 2, we can easily switch between (2) and (3), both are
known as the Weierstraß canonical form in the literature.
It seems that the Weierstraß canonical form is the most widely
known one. There are other canonical forms for cubic equations,
each of which has its own advantages. Usually, it is only possible to
achieve such a canonical form under some additional assumptions.
These are the Legendre canonical form (Adrien Marie Legendre,
1752–1833)
z12 z2 = z0 (z0 − z2 )(z0 − λz2 ), (5)
the Hesse canonical form (Ludwig Otto Hesse, 1811–1874)
z03 + z13 + z23 + tz0 z1 z2 = 0 (6)
and the Tate canonical form (John Tate, 1925–)
z12 z2 + bz0 z1 z2 + cz1 z22 = z03 + dz02 z2 . (7)
If the cubic is given by (5) or (7), the only point at infinity will
again be the inflection point O = (0 : 1 : 0). Therefore, we may also
consider
y 2 = x(x − 1)(x − λ)
as the Legendre canonical form and
y 2 + bxy + cy = x3 + dx2
as the Tate canonical form.
The Tate canonical form can be achieved for a cubic curve which
has at least one point of finite order n > 3. So, it is not a general
normal form for all cubics but it is very useful in order to find the
torsion subgroup of E(K).
The Legendre canonical form exhibits our curve as a double cover
of the projective line P1 . This branched double cover is given by the
map which forgets the y-coordinate (or z1 in the projective setting).
The map so defined can be extended to a map which is also defined

at the point O at infinity. It has four branch points, namely O and
the three points given by the roots of the right hand side, these
are (0 : 0 : 1), (1 : 0 : 1) and (λ : 0 : 1). These four points are
precisely the 2-torsion points of E(K), i.e. those points P which
satisfy P + P = O. This shows that only those cubic curves which
have four 2-torsion points with coordinates in the field K can be
transformed into a Legendre canonical form. In particular, if K = C
or any other algebraically closed field of characteristic not equal to
two, this is always possible.
On the other hand, an equation in Hesse normal form does not
have triple contact with the line at infinity. If the field K contains
three cubic roots of unity (e.g. K = C), it has three points at infinity,
namely the solutions of z03 + z13 = 0. These are inflection points of
the cubic. If K = Q, for example, we see only one of them; this
is the point (1 : −1 : 0). This point is available over any field K
and can be taken as the origin for the group structure. Then, the
set of three-torsion points is precisely the set of inflection points of
this cubic. In particular, if K contains three cubic roots of unity,
the cubic contains nine three-torsion points which lie on the three
coordinate lines zi = 0.
The configuration of these nine points was studied by O. Hesse
[9] who found that the nine inflection points lie on 12 lines, each
of which contains three of these points. Each of the nine points is
contained in four of the 12 lines. This set of nine points and 12
lines is now called the Hesse configuration. A recent survey on the
Hesse configuration and an application to the study of examples of
K3-surfaces can be found in [1].
Finally, let us mention that it is not hard to give an explicit for-
mula for the group structure on E(Q) if the curve is given in Weier-
staß canonical form y 2 = x3 + px + q. For example, if P = (x1 , y1 )
and Q = (x2 , y2 ) 6= −P , the point P + Q = (x3 , y3 ) has coordinates
2
y2 − y1
x3 = − x1 − x2
x2 − x1

y2 − y1 y1 x2 − y2 x1
y3 = − x3 − .
x2 − x1 x2 − x1
This can be obtained using precisely the same calculations as in our
examples in section 2. Remarkably, this formula does not depend
26 BERND KREUSSLER
on p, q, which is due to the non-presence of x2 in the Weierstraß

canonical form. However, p, q explicitly appear in the formula which
describes the coordinates of 2P . For each normal form such a formula
can be obtained.
A formula which is impressive because of its beauty and simplicity
is obtained when we start with an equation of the following form
x2 + y 2 = a2 + a2 x2 y 2 .
If P = (x, y) and P ′ = (x′ , y ′ ) are two solutions of this equation, the
group structure on the solution set is given by
′
′ 1 xy + yx′ yy ′ − xx′
P +P = · , .
a 1 + xyx′ y ′ 1 − xyx′ y ′
The point O = (0, a) is easily seen to be the neutral element of this
group. More on this formula can be found in the recent article [3].
Because the given equation is of degree four, it is not clear how this
example fits into the theory explained so far. That the solution set
of this equation indeed forms a Mordell-Weil group can be explained
using projective geometry. The basic idea is to show that, apart
from a small number of points, the curve defined by this equation of
degree four is isomorphic to a plane cubic curve.
The given curve of degree four has two singular points at infinity,
namely (1 : 0 : 0) and (0 : 1 : 0). If a5 6= a the curve has no other
singular point. We are going to show explicitly that a non-singular
version of this curve is the plane cubic given by the equation

2 1 − a2 1 + a2
y =x x+ x+ . (8)
1 + a2 1 − a2
The outline of the construction is the following. We construct a
non-singular version of the degree 4 curve which is embedded in
projective three-space in such a way that a certain projection from a
centre outside this non-singular curve maps it to the original degree
4 curve. We then find another projection whose centre is on this
non-singular curve in three-space and which maps it isomorphically
onto the plane cubic given by equation (8).
More specifically, using coordinates (w : x : y : z) in P3 , we define
the curve Ee in P3 by the two simultaneous quadratic equations
xy − wz = 0
y 2 − a2 w2 + x2 − a2 z 2 = 0.
The projection with centre (1 : 0 : 0 : 0) is the map which sends

a point (w : x : y : z) ∈ P3 to the point (x : y : z) ∈ P2 . This
projection is not defined at the point (1 : 0 : 0 : 0). All other points
on the line in P3 through (1 : 0 : 0 : 0) and (0 : x : y : z) are
sent to the same point (x : y : z) ∈ P2 . Because the line through
(1 : 0 : 0 : 0) and (0 : x : y : z) meets the curve E e precisely when
2 2 2 2 4 2 2 2
z (x + y ) = a z + a x y , the image of this projection is the plane
curve of degree four given by this equation. Moreover, such a line
has more than one intersection point with E e if and only if it passes
trough (0 : 1 : 0 : 0) or (0 : 0 : 1 : 0). Therefore, away from the two
singular points we obtain an isomorphism between E e and the image
curve in P2 .
The point (0 : 0 : −a : 1) is on the curve E. e The projection
3
with centre (0 : 0 : −a : 1) is a map from P \ {(0 : 0 : −a : 1)}
to P2 . It extends to a map which is defined on all of E e and defines
an isomorphism between E e and its image curve in P2 , which can be
given by the cubic equation (8). The point on the curve E e ⊂ P3
which corresponds to the neutral element O = (0, a), is the point
(0 : 0 : a : 1). The second projection sends this point to our usual
neutral element (0 : 1 : 0) ∈ P2 at infinity. We leave the details of
the calculations to the interested reader.
References
[1] M. Artebani, I. Dolgachev, The Hesse pencil of plane cubic curves,
arXiv:math.AG/0611590
[2] J.W.S. Cassels, Diophantine equations with special reference to elliptic
curves. J. Lond. Math. Soc. 41, 193–291 (1966); Corrigenda. Ibid. 42, 183
(1967)
[3] H.M. Edwards, A normal form for elliptic curves. Bull. Amer. Math. Soc.
44, 393–422 (2007)
[4] T. Ekedahl, One semester of elliptic curves. EMS Series of Lectures in Math-
ematics, European Mathematical Society Publishing House. (2006)
[5] M. Franz, Theta Functions, this issue.
[6] I. Garcı́a-Selfa, M. A. Olalla, J. M. Tornero, Computing the rational torsion
of an elliptic curve using Tate normal form. J. Number Theory 96, No. 1,
76–88 (2002)
[7] R.V. Gurjar et al., Elliptic curves. Praveshika Series. New Delhi: Narosa
Publishing House/dist. by the AMS (2006)
[8] R. Hartshorne, Algebraic geometry. Graduate Texts in Mathematics 52,
Springer (1977)
[9] O. Hesse, Über die Wendepuncte der Curven dritter Ordnung. J. Reine
Angew. Math. 28, 97–107 (1844)
28 BERND KREUSSLER
[10] D.H. Husemoller, Elliptic curves. Graduate Texts in Mathematics 111,

Springer (1987)
[11] M. Khalid, Group law on the cubic curve, this issue.
[12] F. Kirwan, Comples Algebraic Curves. London Mathematical Society Study
Texts 23, Cambridge University Press (1992)
[13] E. Lutz, Sur l’équation y 2 = x3 − Ax − B dans les corps p-adiques. J. reine
angew. Math. 177, 238–247 (1937)
[14] B. Mazur, Modular curves and the Eisenstein ideal. Publ. Math., Inst.
Hautes Étud. Sci. 47, 33–186 (1977)
[15] B. Mazur, Rational isogenies of prime degree. (With an appendix by D.
Goldfeld). Invent. Math. 44, 129–162 (1978)
[16] L.J. Mordell, Note on the integer solutions of the equation Ey 2 = Ax3 +
Bx2 + Cx + D.. Messenger (2) 51, 169–171 (1921)
[17] L.J. Mordell, On the rational solutions of the indeterminate equations of
the third and fourth degrees. Cambr. Phil. Soc. Proc. 21, 179–192 (1922)
[18] L.J. Mordell, On the integer solutions of the equation ey 2 = ax3 + bx2 +
cx + d. Lond. M. S. Proc. (2) 21, 415–419 (1923)
[19] T. Nagell, Solution de quelques problèmes dans la théorie arithmétique des
cubiques planes du premier genre. Wid. Akad. Skrifter Oslo 1935, Nr. 1, 25
p (1935)
[20] C.L. Siegel, The integer solutions of the equation y 2 = axn +bxn+1 +· · ·+k.
(Extract from a letter to Prof. L. J. Mordell.) Journal L. M. S. 1, 66–68
(1926)
[21] J.H. Silverman, The arithmetic of elliptic curves. Graduate Texts in Math-
ematics 106, Springer (1986)
[22] J.T. Tate, The arithmetic of elliptic curves. Invent. Math. 23, 179–206 (1974)
[23] A. Weil, Sur un théorème de Mordell. Bulletin Sc. math. (2) 54, 182–191
(1930)
Mary Immaculate College, South Circular Road, Limerick, Ireland

E-mail address: bernd.kreussler@mic.ul.ie
GROUP LAW ON THE CUBIC CURVE
MADEEHA KHALID
Abstract. It is known that the set of rational points on a

cubic curve E forms a group. The same procedure defines a
group law on all points of E with complex coordinates. With
the aid of the Weierstrass ℘-function one can show that E
is isomorphic to a one dimensional complex torus, namely
E = C/Λ where Λ is a rank 2 lattice in C. The additive
group structure of C descends to the quotient C/Λ and so we
get another group structure on E. In fact these two group
structures are the same. A nice proof of this fact follows from
a classical result by Niels Henrik Abel (1802–1829), known as
“Abel’s theorem”. In this article we introduce the notions of
divisors, line bundles, and the Picard group, and then sketch
the isomorphism between the two group structures.
1. Manifolds
Throughout this article we work over C, the field of complex num-
bers. We denote Pn (C) by Pn .
An n dimensional complex manifold M is a topological space
which locally looks like Cn . This means that there exists an open
cover Uα and co-ordinate maps φα : Uα → Cn such that φα φβ−1 :
φβ (Uα ∩ Uβ ) → Cn is holomorphic for all α, β. Similarly a function
f on an open set U ⊂ M is holomorphic if for all α, f ◦ φ−1 α is
n
holomorphic on φα (Uα ∩ U ) ⊂ C . A map f : M → N between two
complex manifolds is holomorphic if it is given in terms of local holo-
morphic co-ordinates on N by holomorphic functions. Open subsets,
products of complex manifolds and suitable quotients of complex
manifolds are also complex manifolds.
The simplest example of a one dimensional complex manifold is
just C itself. Then there is P1 (isomorphic to the Riemann sphere)
This work was supported by the IRCSET Embark Initiative Postdoctoral

Fellowship Scheme.
29
30 MADEEHA KHALID
which we have seen already in [9] Section 3. By P1 and P2 we mean

the same objects as described in [9] Section 3, except that we replace
K by C. Let Λ = {n1 ω1 + n2 ω2 | ni ∈ Z} be a rank two lattice in
C. Then Λ is an additive sub-group of C generated by two complex
numbers ω1 , ω2 which are linearly independent over the real num-
bers. Addition by elements of Λ defines a fixed point free discrete
group action of Λ on C and hence the quotient C/Λ is a complex
manifold. Since R/Z is diffeomorphic to S 1 via the exponential map
r 7→ exp(2πir), C/Λ is diffeomorphic to S 1 × S 1 and is therefore
called the one dimensional complex torus. Although all tori are dif-
feomorphic to each other, they may not be isomorphic as complex
manifolds (see [3]).
C/Λ
The complex torus is a nice example of a one dimensional manifold

which is easy to describe but which also has a very rich geometric
and arithmetic structure. See for example theta-functions in the
article by M. Franz [5], J. Silverman [11] on the arithmetic aspects
of elliptic curves or the survey article by J. B. Bost [2] on construction
of hyperelliptic Riemann surfaces.
A one dimensional complex manifold is called a Riemann sur-
face. Any complex manifold is orientable so Riemann surfaces are
orientable real surfaces. Compact Riemann surfaces are classified by
their genus g which is a topological invariant and is equal to the
number of holes in the surface. A more precise definition is that
the first homology group of a Riemann surface of genus g is a free
abelian group of rank 2g, i.e. H1 (S) ∼ = Z2g .
So P1 has g = 0, the complex torus C/Λ which is diffeomorphic
to S 1 × S 1 has g = 1.
GROUP LAW ON THE CUBIC CURVE 31
Each compact Riemann surface can be embedded holomorphically

into some Pn . In fact we can choose n to be 3. This is like an analogue
of the Whitney embedding theorem which states that any compact n
dimensional real manifold M can be embedded in R2n+1 . A compact
Riemann surface S together with an embedding i : S → Pn is known
as an algebraic curve. In this article however we will often refer to
a compact Riemann surface as a curve without always necessarily
specifying the embedding in Pn .
Examples of two dimensional manifolds include C2 , P2 , and the
two dimensional complex torus given by C2 /Λ, where Λ is now a
rank 4 lattice in C2 . These lead to some simple examples in higher
dimensions such as Cn , Pn , and Cn /Λ where Λ is a rank 2n lattice
in Cn .
Given a manifold M of dimension n, a subset V ⊂ M given locally
(i.e. on open subsets) as the zero set of a single holomorphic function
f is called a hypersurface in M. For example P1 embeds in P2 as
the zero set of the homogeneous linear function z1 = 0. In local
coordinates on U1 it is given by {(ξ1 , ξ2 ) | ξ1 = 0}. Let az0 +bz1 +cz2
be another linear equation. Then there is a matrix T in PGL(3)
such that T (z1 ) = az0 + bz1 + cz2 . Then {z1 = 0} gets mapped
isomorphically to {az0 + bz1 + cz2 = 0}. This shows that the zero
set of any linear homogeneous equation in P2 is isomorphic to P1 .
Next we consider the zero sets of homogeneous equations of degree
2. If the equation is irreducible then this is is isomorphic to a conic
which is again isomorphic to P1 ([9] Section 1).
In general we denote the zero set in P2 of a homogeneous polyno-
mial of degree d by C, (also known as a plane curve) but when d = 3
we denote it by E (also known as a “cubic curve”) for consistency
with the notation in [9].
Suppose the plane curve C = {(z0 : z1 : z2 ) | f (z0 , z1 , z2 ) = 0}, is
given by a (homogeneous) polynomial
X
f (z0 , z1 , z2 ) = aijk z0 i z1 j z2 k
i+j+k=d
of degree d. Then, on open subsets of P2 , the curve C is the zero

set of a single holomorphic function. Recall that P2 = U0 ∪ U1 ∪ U2
where Ui = {zi 6= 0}. Affine coordinates on U0 are ξi = zz0i . Then
f (z0 ,z1 ,z2 )
C ∩ U0 = {(ξ1 , ξ2 ) | F0 (ξ1 , ξ2 ) = 0}, where F0 (ξ1 , ξ2 ) := z0 d
=
32 MADEEHA KHALID
P
aijk ξ1 j ξ2 k . So C ∩ U0 is the zero locus of the holomorphic func-
tion F0 (ξ1 , ξ2 ). The calculations for the other charts U1 and U2 are
similar.
We say that p ∈ C ∩ U0 is a smooth point, if at least one of the
(ξ1 ,ξ2 ) ∂F0 (ξ1 ,ξ2 )
partial derivatives ∂F0∂ξ 1
, ∂ξ2 is not equal to zero at p. We
say C is smooth if every point in C is a smooth point. If C is smooth
then in fact it is a submanifold of P2 . Another example of a smooth
curve is the curve given locally by ξ1 n + ξ2 n = 1. In homogeneous
coordinates it is the zero locus of z1 n + z2 n = z0 n and is known as
the Fermat curve.
There is a nice formula that computes the genus of a smooth plane
curve C of degree d namely g = (d−1)(d−2) 2 . So if C has degree d
where d ≥ 3 then g ≥ 1 and hence C is not isomorphic to P1 . The
genus of the Fermat curve is (n−1)(n−2) 2 so for n = 1 and 2 it is
1
isomorphic to P while for n = 3 it has genus one and is a complex
torus.
The curve C ′ in P2 given by the equation z2 z1 2 − z0 3 + z0 2 z2 = 0
is not smooth as all the partial derivatives vanish at (0 : 0 : 1). We
say (0 : 0 : 1) is a singular point of C ′ .
These notions of smooth points and singular points can also be
extended to higher dimensional manifolds. In fact just as the im-
plicit and inverse function hold in the differentiable case, so do their
analytic versions. For example if V is a hypersurface given locally
as the zero set of a single holomorphic function f and the jacobian
matrix of f has rank 1 everywhere then V is a manifold of dimension
n − 1.
2. Cubic curves and the group law

It is mentioned in [9] that any smooth cubic E in P2 can be written
as the zero set of the Weierstraß equation after an appropriate change
of variables.
E = {(z0 : z1 : z2 ) | z1 2 z2 = 4z0 3 − pz0 z2 2 − qz2 3 }.
Locally on U2 this corresponds to {(x, y) | y 2 = 4x3 − px − q}.
In addition, the set of rational points on E forms a group, see [9]
Theorem 8. In our case, i.e. when E is defined over C, we show
that this defines a group structure on all points of E with complex
coordinates. As before, let O denote the point (0 : 1 : 0). Let P
and Q be any two points on E and consider the line in P2 containing
P and Q. Then, by the same prescription as in [9] Definition 5,

we see that it meets E in a third point R. Now consider the line
containing O and R. It meets E in a third point say R̄. In the local
coordinates (x, y), R̄ is the reflection of R in the x-axis as mentioned
in [9] Definition 5. We define P + Q := R̄.
bc
R
bc P
Q bc
P + Q := R̄
bc
This is exactly the same as in [9] Theorem 8, except now we allow

P and Q to have complex co-ordinates. In this way we get a group
law on all of points of E with complex coordinates.
The choice of O as the zero element of the group E is not unique.
In fact any point on E can be a zero ([10] Chapter 1, Section 2),
however for E in the Weierstraß form this choice of the zero element
simplifies the group law. We state the analogue of [9] Theorem 8
over C.
Theorem 1. Let E be a cubic curve in P2 given by the Weierstraß
equation
E = {(z0 : z1 : z2 ) ∈ P2 | z1 2 z2 = 4z0 3 − g2 z0 z2 2 − g3 z2 3 },
where g2 , g3 are constants. Then there exists a unique group law on
E such that O := (0 : 1 : 0) is the zero element. The group structure
is determined by requiring
P +Q+R =O if and only if P, Q, and R are on a line.
3. Complex torus
In this section we relate E to the one dimensional complex torus
given as the quotient C/Λ of C by a rank 2 lattice Λ in C. Since
34 MADEEHA KHALID
C/Λ is diffeomorphic to S 1 × S 1 it is like a hollow doughnut and so

has genus 1. Recall that by the “genus formula” for smooth plane
curves E also has genus 1. The following theorem shows that despite
its different appearance E is isomorphic to C/Λ.
Theorem 2. Let Λ be a rank two lattice in C. Then the one di-
mensional complex torus C/Λ can be embedded in P2 as a cubic in
Weierstraß form.
We sketch the main ideas of the proof and introduce the notion
of elliptic integrals. For more details see [7] and [3]. Associated to
Λ there is a meromorphic function on C/Λ called the Weierstraß
℘-function (Karl Weierstraß, 1802), defined as follows.
1 X 1 1

℘(z) = 2 + − 2 .
z (z − ω)2 ω
ω∈Λ\{0}
When viewed as a meromorphic function on C, ℘(z) is doubly

periodic with respect to Λ and has poles of order 2 at all the lattice
points. It satisfies the following differential equation.
2
℘(z)′ = 4℘(z)3 − g2 ℘(z) − g3 . (1)
The constants g2 , g3 are related to Λ and are given by
X 1 X 1
g2 = 60 , g 3 = 140 .
ω4 ω6
ω∈Λ\{0} ω∈Λ\{0}
A complete proof of Equation (1) and the derivations of g2 , g3 is

given in [5] Theorem 10.
The map C/Λ → P2 which identifies C/Λ with a cubic curve, is
given as follows:
(
(℘(z) : ℘′ (z) : 1) if [z] 6= [0]
τ ([z]) = (2)
(0 : 1 : 0) if [z] = [0].
Since ℘(z) satisfies the differential equation (1) we see that in the
local co-ordinates (x, y) on U2 , the image of C/Λ via τ is given by
y 2 = 4x3 − g2 x − g3
which is the Weierstraß cubic equation. To justify the definition of
τ ([0]), we observe that ℘(z) has a pole of order 2 and ℘′ (z) has a
pole of order 3 at [0]. So
f (z) g(z)
℘(z) = and ℘′ (z) = ,
z2 z3
for some holomorphic functions f and g such that f (0) 6= 0, and

g(0) 6= 0. Then, for values of z ∈ C close to 0 ∈ C, we have
τ ([z]) = (℘(z) : ℘′ (z) : 1) = (zf (z) : g(z) : z 3 )
and at z = 0 we obtain (zf (z) : g(z) : z 3 ) = (0 : 1 : 0), which is

τ ([0]) = O, the zero element of the group structure on the cubic
curve E. This shows that we get a holomorphic map τ : C/Λ → E,
where E is a cubic curve in Weierstraß form. One way of showing
that it is an isomorphism is via an inverse mapping and this brings
us to the topic of elliptic integrals.
An elliptic integral is an integral of the form
Zx1
R(x, y)dx
x0
where R(x, y) is a rational function, and y 2 is a polynomial in x of

degree 3 or 4 without multiple roots.
They are called elliptic integrals because they first arose in the
context of determining the arc lengths of an ellipse and of other sec-
ond order curves. Early work on such integrals goes back to Wallis,
Bernoulli, MacLaurin, Riccati and D’Alembert. However for a long
time the problem of inverting such integrals was unsolved. It was
found that they cannot be expressed in terms of the known transcen-
dental functions and also that only three types of new transcendents
suffice to express all such integrals.
Building on work of Fagnano (Giulio Carlo Fagnano dei Toschi,
1682–1766), Euler discovered in 1756 an addition formula for such
integrals. In modern language Euler’s formula is an addition formula
for elliptic functions such as the Weierstraß ℘-function. Much later,
in the second half of the 19-th century, Weierstraß showed that in fact
elliptic functions can be characterised by their property of possessing
an algebraic addition theorem.
The mystery surrounding the mathematical nature of elliptic in-
tegrals was only unveiled by the works of Abel and Jacobi, simul-
taneously published in September 1827. The main new idea was to
study the inverse of the function given by an elliptic integral. Nowa-
days, such functions are called elliptic functions. Abel also noted
that while the elliptic integral itself is a highly complicated function
36 MADEEHA KHALID
of the point (x, y), sums of such integrals (known as Abelian sums)
XZ
xi
R(x, y)dx
x0
satisfy simpler relations. We state and use a special case of Abel’s

theorems later (Section 7, Theorem 22).
Liouville identified in 1844 that the property to be doubly periodic
is the crucial one upon which their analytic study should be based.
Jacobi’s theta functions and the Weierstaß ℘-function form now the
fundaments of a modern theory of elliptic functions. Their definition
and basic properties can be found in the article by M. Franz [5].
Even though elliptic integrals are historically older than elliptic
functions, we usually come across elliptic functions first. The reason
being the work of Cauchy in the theory of complex analysis which
has made the latter an easier object of study today than the integrals
themselves. The elliptic integrals appear then as inverses of elliptic
functions.
If y 2 = 4x3 − g2 x − g3 , an elliptic integral (of the first kind) which
is of particular importance, is
ZP
dx
.
y
O
This is to be understood as a contour integral along a path (x(t), y(t))

in C2 which connects the point O with the point P. It is assumed the
this path is completely contained in the curve given by the equation
y 2 = 4x3 − g2 x − g3 which we know is the curve E in terms of
local coordinates (x, y). Since the genus of E is 1, the curve E is
not simply connected and the integral depends on the choice of the
path. However, this dependence is only modulo the periods of dx y .
This means that the value of the elliptic integral changes only by an
additive constant of the form nω1 + mω2 with m, n ∈ Z. Here the
complex numbers ω1 and ω2 are given by
Z Z
dx dx
ω1 = and ω1 = ,
y y
γ1 γ2
where γ1 , γ2 are two closed paths representing a pair of generators

of the fundamental group of E. The numbers ω1 , ω2 are called peri-
ods because the inverse function of this integral (which is the Weier-
straß ℘-function) is doubly periodic with periods ω1 , ω2 . The periods
ω1 , ω2 are linearly independent over R because the paths γ1 , γ2 are
generators of the fundamental group.
So Λ = {nω1 + mω2 | n, m ∈ Z} is a lattice in C and in this way
we reconstruct the lattice Λ we started with. Moreover, if we set
ZP
dx
τ −1 (P ) =
y
O
where O is the point at infinity and P ∈ E any point, we obtain a

well-defined map τ −1 : E → C/Λ which is the inverse of the map
τ defined earlier. This gives an isomorphism of E with C/Λ. The
differential dx
y is actually the familiar differential dz on the torus
C/Λ. That is because x = ℘(z), y = ℘′ (z) so we get τ ∗ dxy = dz, the
integral of which is well defined modulo Λ.
In order for this procedure to work, all we need is that the cubic
curve y 2 = 4x3 − g2 x − g3 be smooth, i.e. g23 − 27g32 6= 0. Since
any smooth cubic curve in P2 is isomorphic to a cubic in Weierstraß
form, it follows that every smooth cubic in P2 is isomorphic to a
complex torus. For more details see [7] Chapter 2.
4. Divisors
In the previous section we saw that a plane cubic curve E is
isomorphic to a complex torus C/Λ. Now C/Λ inherits a group
structure from C and hence induces a group structure on E via the
isomorphism. In Section 2 we defined a group operation on E using
geometry. How do these two compare?
The answer is: they coincide! In the subsequent sections we de-
scribe a proof which weaves together some pretty ideas from algebraic
geometry. To do so we have to first introduce an important notion
in algebraic geometry which is that of a divisor. In the case of a
curve it has a simple description.
Definition 3. Let C be a smooth curve in P2 . A divisor on C is a
formal finite linear combination D = a1 · P1 + · · · + am · Pm of points
Pi ∈ C with integer coefficients ai .
38 MADEEHA KHALID
Divisors can be added or subtracted and hence form a group de-

noted Div(C).
Pm of a divisor D = a1 · P1 + · · · + am · Pm is
Definition 4. The degree
defined to be deg D = i=1 ai and this gives a group homomorphism
deg : Div(C) → Z.
Remark 5. The notion of a divisor extends also to higher dimen-
sional manifolds. In that case a divisor is a linear combination of
subsets given locally by zero sets of irreducible holomorphic func-
tions.
The group Div(C) is very large, even in the one-dimensional case.
Therefore we introduce the sub-group of principal divisors. The ben-
efit is that the factor group of all divisors modulo principal divisors
is finitely generated. This factor group will prove to be useful in Sec-
tion 6 as well. In order to explain the definition of principal divisors,
we need the notion of the order of a function at a point P
Let f be a holomorphic function on an open set U ⊂ C. Let
P ∈ U and let x be the local co-ordinate on U such that P is given
by x − λ for some λ ∈ C. The order of f at P , denoted ordP (f ), is
the largest integer a ∈ Z such that locally
f (x) = (x − λ)a · h(x)
where h is a holomorphic function with h(λ) 6= 0. Since f is holomor-
phic a is non negative. Note that for g, h any holomorphic functions
ordP (gh) = ordP (g) + ordP (h).
We would like to include the cases when ordP (f ) is negative. To
do so we have to include what are known as meromorphic functions.
A function f on C is called a meromorphic function if it can be writ-
ten locally as a ratio hg , where g 6= 0 and h are holomorphic functions
which do not have a common zero. Then, by using a Laurent series
expansion for f at P , we see that ordP (f ) = ordP (g) − ordP (h). So
ordP (f ) is negative if ordP (g) < ordP (h).
Collecting zeros and poles of a global meromorphic function f
gives us a natural way to associated a divisor to it.
Definition 6. Let f be a meromorphic function on C. Then the
divisor of f , called a principal divisor and denoted (f ), is given by
X
(f ) = ordP f · P.
P ∈C
Example 7. Consider P1 , the Riemann sphere with homogeneous

co-ordinates (z0 : z1 ). Then any ratio f = hg , where g and h are
homogeneous polynomials of the same degree d, is a global mero-
2
morphic function. So for instance if f = z1 2z+z 0
0 z1
then (f ) =
2 · P0 − P1 − P2 where P0 = (0 : 1), P1 = (1 : 0), P2 = (1 : −1).
Note that deg(f ) = 0.
Now we do the same thing for curves in P2 . A meromorphic
function f on P2 restricts to a meromorphic function on the curve
C if the denominator in the local expression for f does not vanish
identically on the curve. Its associated divisor (f ) restricts to a
2 2 2
divisor on C. As an example lets take the function f = z0 +zz11 2 +z2
and the line L0 = {z2 = 0}. Then a local computation shows that
(f ) = (1 : i) + (1 : −i) − 2 · (1 : 0).
Given any curve C ⊂ P2 and any divisor D on C, a natural question
to ask is whether D = (f ) for some meromorphic function f on C?
The following example is a partial answer to this question.
Example 8. Consider the line L2 = {(z0 : z1 : z2 )|z2 = 0} in P2 .
(see [9] Section 3) and O the point (0 : 1 : 0). Then D = 2 · O is
a divisor on L2 . If D = (f ) for some meromorphic function on L2 ,
then f has a zero of order 2 at O and is holomorphic and nonzero
everywhere else. Since L2 is isomorphic to P1 there are no non-
constant holomorphic functions on P1 , D 6= (f ) for any f .
In the case of P1 the answer to the above question is very simple.
A divisor D = (f ) if and only if deg D = 0. For a cubic curve E the
answer is not so simple. For instance there exist divisors of degree 0
which are not associated to any meromorphic function. In fact there
are as many such divisors as there are points on E. We discuss this
in more detail in Section 6. See also the article by C. Daly [4].
5. Line Bundles
Divisors are closely tied together to another geometric notion
which is that of a line bundle. A line bundle is a rank 1 holomorphic
vector bundle (Definition 10). In this section we discuss the relations
between line bundles and divisors.
Let us for the moment refer back to Example 8. The homoge-
neous coordinates z0 , z1 of P2 are also natural homogeneous coordi-
nates on L2 , since L2 = {(z0 : z1 : 0) ∈ P2 }. Our aim is to associate
40 MADEEHA KHALID
to a homogeneous polynomial in the ring C[z0 , z1 ] its “divisor of ze-

roes”. For example consider the homogeneous quadratic polynomial
z02 . Since it is a homogeneous polynomial it is invariant under scalar
multiplication and so D := {z02 = 0} is a well defined subset of L2 .
This zero set has a local description. Recall from [9] Section 3 that
L2 is covered by two open charts V0 = {z0 6= 0} and V1 = {z1 6= 0}.
The affine coordinate on V0 is z = zz10 and the affine coordinate on
V1 is w = zz01 . On V0 ∩ V1 we have the identification map z = w1 .
Consider D ∩ V1 = {(z0 : z1 ) ∈ V1 | z02 = 0}. If (z0 : z1 ) ∈ D ∩ V1
then certainly (λz0 : λz1 ) ∈ D ∩ V1 , so we divide by z1 to get that
z2
D ∩ V1 = {( zz01 : 1) | z02 = 0}. In terms of the coordinate on V1 this
1
is just {w ∈ V1 | w2 = 0}.
Similarly D ∩ V0 = {(z0 : z1 ) ∈ V0 | z02 = 0}. In terms of the
z2
coordinate on V0 this corresponds to {(1 : z) | z02 = 1 = 0} which
0
is just the empty set. So, locally D corresponds to the following
subsets
D ∩ V0 = {z ∈ V0 | 1 = 0},
D ∩ V1 = {w ∈ V1 | w2 = 0}.
Set f0 := 1, f1 := w2 , then {(V0 , f0 ), (V1 , f1 )} are local defining
functions for D. Notice that on V0 ∩ V1 , we have w2 = z12 6= 0 and
f0 (z) = f1 (z) · z 2 .
Similarly f1 (w) = f0 (w)·w2 on V0 ∩V1 , so the local defining functions
are related by a nowhere vanishing factor.
Now D ∩ V0 = ∅ and D ∩ V1 is the origin w = 0 counted with
multiplicity 2. The point w = 0 in V1 corresponds to (0 : 1 : 0) on
L2 . Since this occurs with multiplicity 2, D is the divisor 2 · O where
O = (0 : 1 : 0), as before.
The interesting thing is that from these local defining functions
of D we construct a new manifold L called a line bundle. The non-
vanishing factor that relates these local defining functions of D is
known as a transition function. We give one more example before
stating the definitions.
Example 9. Set
L := V0 × C ∪ V1 × C/ ∼
where V0 × C and V1 × C are open charts of L. The equivalence
relation ∼ gives the “patching” condition on the overlap (V0 ∩V1 )×C
1
and is defined as follows. For w = z ∈ V0 ∩ V1 and (w, λ) ∈ V1 × C,
(z, µ) ∈ V0 × C we define
(w, λ) ∼ (z, µ) ⇐⇒ µ = z 2 λ.
Note that we define the “patching” condition using z 2 , the nowhere
vanishing function on V0 ∩ V1 relating the two local descriptions of
D above.
This new manifold L is an example of a line bundle, (see Def. 10
below) and is often denoted O(D). The collection {(f0 , V0 ), (f1 , V1 )}
of local defining functions for D = 2 · O defines a section (see subsec-
tion 5.2) of O(D) and the function z 2 relating these local functions
on the overlap V0 ∩ V1 is a transition function of O(D).
We now give the general definition of a line bundle.
π
Definition 10. Let M be a complex manifold. A line bundle L → M
is a holomorphic vector bundle of rank 1. That is
(1) L is a complex manifold such that for any x ∈ M, π −1 (x) =
Lx is equipped with the structure of a one dimensional com-
plex vector space.
(2) The projection mapping π : L → M is holomorphic.
(3) There is an open cover {Uα } of M and biholomorphic maps,
φα : π −1 (Uα ) → Uα × C, compatible with the projections
onto Uα , such that the restriction to the fibre φα : Lx →
{x} × C is linear for all x ∈ Uα . The pair (φα , Uα ) is called
a trivialisation of L over Uα .
Lx LUα ⊂ L
π
b Uα ⊂ M
x
42 MADEEHA KHALID
Since L is a complex manifold, for any pair of trivialisations φα , φβ

the map gαβ : Uα ∩ Uβ → C∗ given by

−1
φα φβ (x, v) = (x, gαβ (x) · v)
is holomorphic. The maps gαβ are called transition functions of L
with respect to the trivialisations (φα , Uα ), (φβ , Uβ ). They deter-
mine the line bundle L and satisfy the following conditions
(1) gαβ (x) · gβα (x) = 1 for all x ∈ Uα ∩ Uβ ;
(2) gαβ (x) · gβγ (x) · gγα (x) = 1 for all x ∈ Uα ∩ Uβ ∩ Uγ .
Condition (2) is known as the cocycle condition.
Conversely, given an open cover {Uα } of M and holomorphic maps
gαβ : Uα ∩Uβ → C∗ , satisfying the conditions above, we can construct
a line bundle L with transition functions gαβ . Define an equivalence
relation ∼ on the union over all α of Uα × C as follows. For x ∈
Uα ∩ Uβ , (x, λ) ∈ Uβ × C and (x, µ) ∈ Uβ × C set (x, λ) ∼ (x, µ) if
and only if µ = gαβ (x) · λ. Then
[
L= Uα × C/ ∼
α
is a line bundle with transition functions gαβ .
For ease of notation from now on we set LU = π −1 (U )
Given L as above, for any collection of nowhere vanishing holo-
morphic functions fα on Uα we can define alternative trivialisa-
tions φ′α of L over Uα by multiplying the second component of
φα (x) ∈ Uα × C with fα (x). In a more sloppy way we write
φ′α = fα φα . (3)
The transition functions relative to (φ′α , Uα ) are
′ fα
gαβ = gαβ .
fβ
Any other trivialisation of L can be obtained in this way, so we see
′
that the collections {gαβ } and {gαβ } define the same line bundle if
and only if there exist nowhere vanishing holomorphic functions fα
on Uα satisfying (3) above.
Example 11. The simplest example of a line bundle on a manifold
is M × C also known as the trivial bundle OM .
Example 12. The line bundle that we constructed in Example 9
is known as OP1 (2). All line bundles constructed in this way from
a divisor defined by a homogeneous quadratic polynomial on P1 are

isomorphic because for any two such polynomials there is an isomor-
phism of P1 which maps one to the other.
Remark 13. In fact given any n ∈ Z+ all line bundles obtained
from divisors corresponding to a homogeneous polynomial of degree
n on P1 are isomorphic and denoted by OP1 (n).
Example 14. Recall that P1 = (C2 \ {0})/∼ where (z0 , z1 ) ∼
(λz0 , λz1 ) for all λ ∈ C∗ . This means, each line l ⊂ C2 through
the origin corresponds to a point [l] ∈ P1 . Let
L = {((z0 : z1 ), v) ∈ P1 × C2 | v ∈ C · (z0 , z1 )}
= {([l], v) ∈ P1 × C2 | v ∈ l}
and denote projection onto the first factor by π : L → P1 . Then, in
terms of local co-ordinates on U0 and U1 as before, we obtain
LU0 = {(z, (β, βz)) | β ∈ C}
LU1 = {(w, (ηw, η)) | η ∈ C}
with trivialisations
φ0 : LU0 → U0 × C φ1 : LU1 → U1 × C
(z, (β, βz)) 7→ (z, β) (w, (ηw, η)) 7→ (w, η)
The reader can check that the transition function g01 is;
1
g01 (z) = φ0 φ−1
1 = .
z
This vector bundle is also known as the universal bundle on P1 de-
noted OP1 (−1) and is an important example.
A nice property of line bundles is that they can be “pulled back”.
Suppose f : M → N is a holomorphic map of complex manifolds,
and π : L → N is a line bundle on N . Then we define the pull back
bundle f ∗ L by setting (f ∗ L)x = Lf (x) . More precisely,
f ∗ L = {(m, v) | f (m) = π(v)} ⊂ M × L.
If φ : LU → U × C is a trivialisation of L in a neighbourhood U of
f (x), then we obtain a trivialisation
f ∗ φ : (f ∗ L)f −1 (U) → f −1 (U ) × C
which is the composition
Id ×φ pr
(f ∗ L)f −1 (U) ⊂ f −1 (U ) × LU −→ f −1 (U ) × U × C −→ f −1 (U ) × C.
44 MADEEHA KHALID
This gives f ∗ L its manifold structure over the open set f −1 (U ). The
transition functions for f ∗ L are the pull backs f ∗ (gαβ ) := gαβ ◦ f of
the transition functions gαβ of L.
Remark 15. If D is a divisor on N with local defining functions
{(hα , Uα )}, we can pull it back to a divisor f ∗ D on M with local
defining functions {(hα ◦ f, f −1 (Uα ))}. If L = O(D), then f ∗ (L) =
O(f ∗ D).
5.1. Group structure on the set of all line bundles. The ten-
sor product of C with itself, C ⊗ C is C again. Similarly given two
line bundles L1 and L2 with transition functions gαβ and hαβ respec-
tively, we can define the tensor product L1 ⊗ L2 and get a new line
bundle L. The fibres of L are just the tensor product of fibres of L1
and L2 . The transition functions tαβ of L are therefore the product
of the transition functions of L1 and L2 , i.e. for all x ∈ Uα ∩ Uβ
tαβ (x) = gαβ (x)hαβ (x).
This defines a binary operation on the set of line bundles. Tensoring
with the trivial bundle O gives the same bundle back, so it is the
neutral element of the group structure. Associated to each line bun-
dle L with transition functions gαβ , there is another line bundle L∗
−1
whose transition functions are gαβ . It is called the dual bundle of L.
∗
Since L ⊗ L = O, the dual bundle is like the inverse of L. Hence
we get a group structure on the isomorphism classes of line bundles
on M . This group is called the Picard group of M denoted Pic(M ).
In the next section we describe Pic(E) for E a smooth cubic curve
in P2 .
5.2. Sections of a line bundle. A section s of a line bundle L is a
holomorphic map s : M → L such that π◦s = Id. Locally this means
we have an open cover Uα and a collection of holomorphic functions
sα : Uα → C such that
sα (x) = gαβ (x) · sβ (x) ∀ x ∈ Uα ∩ Uβ .
An example of a section is given in Example 9. It may be the case
that a line bundle does not have any holomorphic sections. Lo-
cal holomorphic sections always exist but they may not satisfy the
patching condition on overlaps.
For instance consider the line bundle OP1 (−1) as in Example 14.
Suppose it has a local holomorphic section s1 (w) on U1 where s1 (w)
is a holomorphic function. Then on U0 ∩ U1 it transforms to s0 (z) =
s1 ( 1z ) · z1 which is a meromorphic function on U0 and certainly not

holomorphic. This shows that OP1 (−1) does not have any global
holomorphic sections and therefore we extend our definition to al-
low meromorphic sections of L. A collection of local meromorphic
functions {sα : Uα → C} such that sα = gαβ sβ will be called a mero-
morphic section of L. The section s1 (w) = 1, s0 (z) = z1 is a global
meromorphic section of OP1 (−1) with a simple pole at (1 : 0).
Finally we come to the correspondence between divisors and line
bundles.
5.3. Divisors and line bundles. Let D be a divisor on a curve

C and let {(fα , Uα )} be local defining functions for D. Then the
functions gαβ = ffαβ are holomorphic and non zero on Uα ∩ Uβ . They
also satisfy the cocycle condition on Uα ∩ Uβ ∩ Uγ since
fα fβ fγ
gαβ gβγ gγα = = 1 on Uα ∩ Uβ ∩ Uγ .
fβ fγ fα
So the collection {gαβ } defines a line bundle called the associated
line bundle of D, and denoted O(D), (see also Example 9.)
Conversely since any curve C embeds in some projective space
n
P , given a line bundle L over a curve C, there exists a meromorphic
section s of L (for a proof see [7] Chapter 1, Section 2, the proposition
directly before the Lefschetz theorem on (1, 1) classes.) Consider a
local representation {sα , Uα } of s. Then given any P ∈ C we can
define the order of s at P as
ordP (s) = ordP (sα )
Where α is arbitrary with P ∈ Uα . This does not depend on the
(x)
choice of α, since ssαβ (x) = gαβ (x) ∈ C∗ for all x ∈ Uα ∩ Uβ and so
ordP sα = ordP sβ , if P ∈ Uα ∩ Uβ . We take the divisor (s) of s to be
X
(s) = ordP (s) · P.
P ∈C
If we were to take the line bundle associated to the divisor (s) we

would recover L our original line bundle. So we get a map
Div(C) → Pic(C) (4)
D 7→ O(D) (5)
Remark 16. This correspondence still holds if we replace C by an
algebraic complex manifold M .
46 MADEEHA KHALID
In fact (4) is a group homomorphism. A good exercise is to

check it is well defined. Suppose D1 , D2 are two divisors. We can
choose an open cover fine enough so that they are locally defined
by {fα }, {hα }. Then D1 + D2 has local defining functions {fα hα }.
The corresponding line bundle O(D1 + D2 ) has transition functions
tαβ = ffαβ hhαβ . The line bundles O(D1 ) and O(D2 ) have transition
functions gαβ = ffαβ and qαβ = hhαβ respectively. It is clear that
tαβ = gαβ qαβ , so O(D1 + D2 ) = O(D1 ) ⊗ O(D2 ). In other words
addition of divisors in Div(C) maps to tensor product of line bun-
dles in Pic(C). If D = (f ) for some global meromorphic function
then fα = fβ so gαβ = 1 and hence O(D) is the trivial line bundle.
We say D1 is linearly equivalent to D2 , denoted D1 ∼ D2 , if and
only if there exists a global meromorphic function f on C such that
D1 = D2 + (f ).
Lemma 17. Let C be a curve. Let Div(C) denote the group of
divisors and Pic(C) the group of line bundles on C. Then O(D) is
trivial if and only if D = (f ) for some meromorphic function on C,
i.e. Div(C)/ ∼ ∼ = Pic(C).
Proof. We have seen that (f ) corresponds to the trivial line bundle
so we just need to show that if O(D) is a trivial line bundle then
D = (f ). Let {(fα , Uα )} be local defining functions for D. Then
O(D) trivial implies there exist functions hα : Uα → C∗ such that
fα hα
= gαβ = = 1.
fβ hβ
Hence,
fα gαβ fβ fβ
f= = =
hα gαβ hβ hβ
is a global meromorphic function on C with divisor D.
Let L be a line bundle on C and s a meromorphic section of L. For

the reader familiar with some differential geometry we now mention
a nice relation between the first chern class of O(D) and D. Given
a divisor D on C let ηD denote its Poincare dual in H 2 (C, Z). Let L
be any line bundle. Then L admits a hermitian metric and there is
a unique connection on L compatible with the metric and complex
structure. Let Θ be the curvature form associated to this metric
connection.
Theorem 18. Let L = O(D) be a line bundle. Let Θ be the cur-

vature form associated to a metric connection. Let ηD denote the
2
Poincaré dual of D in HDR (C) and let c1 (L) denote the first Chern
class of L. Then

Θi 2
c1 (L) = = ηD ∈ HDR (C).
2π
For a proof see [7] Chapter 1, Section 1. This implies
Z
i
Θ = hηD , [C]i = deg D.
2π C
Remark 19. All the results of this section also hold when we replace
C by a complex manifold M .
6. Poincaré Bundle
Now we restrict attention to the case of a plane cubic curve E and
ask ourselves the following question. What does the set Pic0 (E) of
all degree zero line bundles on E look like?
Here by degree of a line bundle we mean the degree of its associ-
ated divisor (Definition 4). In the case of P1 up to isomorphism there
is only one line bundle of degree zero and that is the trivial bundle.
However on E there are many non-trivial line bundles having degree
zero as we shall soon see. In fact they form a family parametrised
by E.
First let’s take a point P ∈ E. This is a divisor of degree 1 on
E, and it defines a line bundle O(P ). Now choose another point Q
distinct from P and take the divisor P − Q. This has degree zero
and correspondingly defines a line bundle O(P − Q). One could ask
is O(P − Q) isomorphic to the trivial bundle?
If so then by Lemma 17 there would exist some global meromor-
phic function f on E such that P − Q = (f ). This means that f
has exactly a pole of order 1 at Q and a zero of order 1 at P and no
other poles or zeroes. But then we can define a bijective map
E → P1
x 7→ (f (x) : 1)
Under this mapping Q maps to the point at ∞ = (1 : 0) on P1 . Since
f is meromorphic with exactly one pole and holomorphic elsewhere
it is an isomorphism. But E has genus 1 while P1 has genus 0 so
they cannot be isomorphic. Therefore P − Q ≁ 0, i.e. P , Q are
48 MADEEHA KHALID
inequivalent divisors and define non isomorphic line bundles. The

following theorem says that in fact the family of degree 0 line bundles
on E is itself a manifold.
Theorem 20. Let E be a cubic curve in P2 . Then there is a bijection
E∼
= Pic0 (E).
Proof. We just show that there is an injection from E to Pic0 (E).
For a proof of surjectivity see [6] Chapter 6.
Fix a point in E (doesn’t matter which one) for instance O. Then
given any other point P ∈ E we get a degree 0 line bundle O(P − O).
This defines a map E → Pic0 (E). For P and Q distinct points the
line bundles O(P − O) and O(Q − O) are non isomorphic. Because
if they were isomorphic then by Lemma 17 the divisor P − O would
be linearly equivalent to Q − O which implies P − Q ∼ (f ) for some
meromorphic function f . But as we have already seen, in that case f
defines an isomorphism between E and P1 which is a contradiction.
Hence our map is bijective.
In fact there is a more general theorem.
Theorem 21. Let E be a cubic curve in P2 . Then for all n ∈ Z we
have Picn (E) ∼
= E.
The idea is that if we fix a point O on E then any line bundle L
of degree n, can be mapped to a line bundle of degree zero by taking
the tensor product L ⊗ O(−nO) and vice versa.
There is a special line bundle on E × Pic0 (E) ∼
= E × E called the
Poincaré bundle P. It has the property that P|E×{P } ∼= OE (P − O).
We construct this line bundle as follows. Let (x, y) be the local co-
ordinates on E × E. Consider the subset ∆ = {(x, y)|x = y} called
the diagonal. It is a divisor since it is given by the zero locus of
a single equation. Its associated line bundle O(∆) has the prop-
erty that O(∆)|E×{P } ∼ = OE (P ). The idea is simple, by Remark 15
O(∆)|E×{P } = O(∆|E×{P } )|E×{P } . Let p1 , p2 denote projection of
E × Pic0 (E) ∼= E × E onto the first and second factor respectively.
Consider the line bundle P := O(∆) ⊗ p∗1 O(−O). Then p∗1 O(−O) is
just the line bundle associated to the divisor −({O} × E) in E × E.
P|E×{p} = O(∆) ⊗ p1 ∗ O(−O)|E×{P }
= O(∆)|E×{P } ⊗ O(−(O × E))|E×{P }
∼
= OE (P ) ⊗ OE (−O) = OE (P − O)
The point P in the second factor of E × E represents the line bundle

O(P − O) when viewing E × E as E × Pic0 (E) via the isomorphism
E → Pic0 (E)
P 7 → O(P − O)
So we see that P|E×{P } is isomorphic to the corresponding element
OE (P − O) in Pic0 (E). This is an example of a moduli space and
its universal bundle. The moduli space of degree zero line bundles
on E is isomorphic to E. The universal line bundle on E ×Pic0 (E) is
given by P characterised by the property that for any P ∈ Pic0 (E),
P|E×{P } is a line bundle of degree zero belonging to the isomorphism
class of P ∈ Pic0 (E). For more details about moduli spaces of vector
bundles on elliptic curves see the article by C. Daly[4].
7. Abel’s Theorem; group law revisited

In Section 3 we showed that a cubic curve E is isomorphic to a
complex torus C/Λ. We now have all the pieces to put together a
proof of the fact that the geometric group structure on E is the same
as the group structure on C/Λ.
The main result involved in proving this is the following classical
theorem known as Abel’s theorem [1] (1827).
Theorem 22. Let Λ be a rank two lattice in C, let n1 , . . . , nn and
m1 , . . . , mm be integers and let [a1 ], . . . , [an ] and [b1 ], . . . , [bm ] denote
points in C/Λ.
Then there exists a meromorphic function f : C/Λ → C with
zeroes at [ai ] of order ni and poles at [bj ] of order mj if and only if
n
X m
X n
X m
X
ni = mj and ni [ai ] = mj [bj ] ∈ C/Λ.
i=1 j=1 i=1 j=1
Moreover this function is unique up to a constant factor.

For a proof of Abel’s theorem involving a nice application of theta-
functions see [5] Theorem 7.
Theorem 23. Let E be a smooth cubic curve in P2 . Then E ∼ = C/Λ
for some rank 2 lattice Λ in C. The geometric group structure on
E as defined in Theorem 1 is isomorphic to the group structure on
C/Λ.
50 MADEEHA KHALID
Proof. Consider three points P1 , P2 , P3 on E which lie on a line L.

This is equivalent to saying P1 + P2 + P3 = O. Let [z1 ], [z2 ], [z3 ] be
the unique points on the complex torus C/Λ which are mapped to
P1 , P2 , P3 under the isomorphism τ , i.e. τ ([zi ]) = Pi (see Section 3
for the definition of τ .) If we can show that [z1 ] + [z2 ] + [z3 ] = 0 then
we are done.
Suppose L = {(z0 : z1 : z2 ) ∈ P2 | l0 z0 + l1 z1 + l2 z2 = 0}.
Then since z2 is not identically zero on L the meromorphic function
F = l0 z0 +l1zz21 +l2 z2 on P2 restricts to a meromorphic function f on
E. The divisor of F in P2 has a simple zero along the line L and
a simple pole along the line L2 = {(z0P 2 ) | z2 = 0}. So the
: z1 : zP
divisor of F restricted to E is (f ) = Pi − Qj where {Pi } =
L ∩ E and {Qj } = L2 ∩ E with multiplicities. Implicit differentiation
shows that O is an inflection point of E and hence O is a triple
point of contact of the line L2 and E. So L2 meets E at O with
multiplicity 3. Therefore (f ) = P1 + P2 + P3 − 3O. Now f pulls back
to a meromorphic function τ ∗ f on C/Λ with zeroes of order one each
at [z1 ], [z2 ], [z3 ] and a pole of order three at [0]. By Abel’s theorem
this is the case if and only if [z1 ] + [z2 ] + [z3 ] = 3[0] in C/Λ, i.e. if
the points [z1 ], [z2 ], [z3 ] sum to zero in C/Λ.
This concludes our overview of the group structure on an elliptic

curve E in P2 . For other interesting features of elliptic curves and
moduli spaces of vector bundles on elliptic curves see the article by
C. Daly [4].
In two dimensions the only compact complex manifold that admits
a group structure is a complex torus. However one can consider
families of elliptic curves called elliptic fibrations. The geometry
of these elliptic fibrations is very interesting and has been studied
in detail. In the complex analytic case they have been classified
by Kodaira (see [8]). Recently there has also been much interest
in higher dimensional elliptically fibred manifolds in the context of
mathematical physics.
References
[1] N. H. Abel Recherches sur les fonctions elliptiques J. reine angew. Math.
2, pages 101-181 (1827).
[2] J. B. Bost, An introduction to compact Riemann surfaces, jacobians and
Abelian varieties, From number theory to Physics, M.Waldschmidt et.al
(eds), Springer Verlag (1992).
[3] C. H. Clemens, A scrap book of complex curve theory, American Math-

ematical Society 2002 edition.
[4] C. Daly, Rank two vector bundles on elliptic curves. this issue
[5] M. Franz, Theta functions. this issue
[6] A. Gathmann, Lecture Notes from ”Algebraic geometry”, taught at Uni-
versity of Kaiserslautern, 2002/2003.
[7] P. Griffiths, J. Harris, Principles of algebraic geometry, John Wiley and
Sons Inc 1994 edition.
[8] K. Kodaira, On the structure of compact complex analytic surfaces I,
Am. J. math. 86 (1964) 751-798.
[9] B. Kreussler, Solving cubic equations in two variables. this issue
[10] M. Reid, Undergraduate algebraic geometry, London Mathematical So-
ciety Student texts 12, Cambridge university Press 2001.
[11] J. Silverman The arithmetic of elliptic curves, Springer Verlag 1986.
Department of Mathematics, Institute of Technology Tralee, Clash,

Tralee, Co. Kerry, Ireland
E-mail address: Madeeha.Khalid@staff.ittralee.ie
52
THETA FUNCTIONS
MARINA FRANZ
Abstract. On our analytic way to the group structure of

an elliptic function we meet so called theta functions. These
complex functions are entire and quasi-periodic with respect
to a lattice Λ. In the proof of Abel’s theorem we use their
properties to charcterise all meromorphic functions f : C/Λ →
C. Finally we have a closer look at a very special and in-
teresting Λ-periodic meromorphic function, the Weierstraß
℘-function. This function delivers an analytic way to give a
group structure to an algebraic variety.
1. Introduction
First of all, we want to analyse periodic complex functions f :
C → C with respect to a lattice Λ. So let us fix once and for all
a complex number τ ∈ C with Im τ > 0 and consider the lattice
Λ := Z ⊕ τ Z ⊂ C.

τ V 1 + τ

0 1

Figure 1. The lattice Λ = Z ⊕ τ Z and its funda-

mental parallelogram V = {z = t1 + t2 τ ∈ C : 0 ≤
t1 , t2 < 1}.
Lemma 1. An entire and doubly-periodic complex function is con-

stant.
53
54 MARINA FRANZ
To prove this lemma we need Liouville’s Theorem, which we know

from complex analysis. It states that each entire (i.e. holomorphic,
i.e. complex differentiable, everywhere in C) and bounded complex
function f : C → C is constant.
Proof. The values of a doubly-periodic function are completely deter-

mined by the values on the closure of the fundamental parallelogram
V = {z ∈ C : z = t1 + t2 τ for some 0 ≤ t1 , t2 ≤ 1} which is a com-
pact set. But a continuous function on a compact set is bounded.
Hence our function is entire and bounded. Therefore it is constant
by Liouville’s Theorem.
As we have seen, that entire doubly-periodic functions are not very

intersting (as they are constant), in the following we will consider
entire quasi-periodic functions and use them to prove Abel’s The-
orem which says what meromorphic doubly-periodic functions look
like.
2. Theta Functions and Abel’s Theorem

Definition. The basic theta f unction is defined to be the function
θ : C → C given by
X
θ(z) := θ(τ )(z) := exp(πin2 τ ) exp(2πinz)
n∈Z
Note. The function θ depends on τ . So for each τ ∈ C with Im τ > 0

we get a (not necessarily different) basic theta function. Hence there
is a whole family of basic theta functions {θ(τ )}τ ∈C,Im τ >0 . But here
we assume τ to be fixed, so we have only one basic theta function.
Remark. As the series in the definition above is locally uniformly

unordered convergent (without proof) our basic theta function is an
entire function.
Lemma 2. The basic theta function is quasi-periodic.
Proof. Consider θ(z + λ) for λ ∈ Λ, i.e. λ = pτ + q for p, q ∈ Z.

THETA FUNCTIONS 55
For λ = 1, i.e. for p = 0 and q = 1 we have
def X
θ(z + 1) = exp(πin2 τ ) exp(2πin(z + 1))
n∈Z
X
= exp(πin2 τ + 2πinz + 2πin)
n∈Z
X
= exp(πin2 τ ) exp(2πinz) exp(2πin)
| {z }
n∈Z
=1 for all n∈Z
X
= exp(πin2 τ ) exp(2πinz)
n∈Z
def
= θ(z)
Hence the basic theta function is periodic with respect to the x-

direction.
For λ = τ , i.e. for p = 1 and q = 0 we have
def X
θ(z + τ ) = exp(πin2 τ ) exp(2πin(z + τ ))
n∈Z
X
= exp(πin2 τ + 2πinz + 2πinτ )
n∈Z
if we complete the square and rearrange the summands then
X
= exp πin2 τ + 2πinτ + πiτ − πiτ
n∈Z
+2πinz + 2πiz − 2πiz)
X
= exp(−πiτ − 2πiz) exp(πi(n + 1)2 τ ) exp(2πi(n + 1)z)
n∈Z
if we make a simple index shift m = n + 1 then
X
= exp(−πiτ − 2πiz) exp(πim2 τ ) exp(2πimz)
m∈Z
def
= exp(−πiτ − 2πiz)θ(z)
Hence the basic theta function is not periodic with respect to the
τ -direction as in general exp(−πiτ − 2πiz) 6= 1.
56 MARINA FRANZ
In the general case we obtain

θ(z + λ) = θ(z + pτ + q)
def X
= exp(πin2 τ ) exp(2πin(z + pτ + q))
n∈Z
X
= exp(πin2 τ + 2πinz + 2πinpτ + 2πinq)
n∈Z
if we complete the square and rearrange the summands then
X
= exp πin2 τ + 2πinpτ + πip2 τ − πip2 τ
n∈Z
+2πinz + 2πipz − 2πipz + 2πinq)
= exp(−πip2 τ − 2πipz)
X
· exp(πi(n + p)2 τ ) exp(2πi(n + p)z)
n∈Z
exp(2πinq) ]
| {z }
=1 for all n∈Z
= exp(−πip2 τ − 2πipz)
X
· exp(πi(n + p)2 τ ) exp(2πi(n + p)z)
n∈Z
if we make a simple index shift m = n + p then
X
= exp(−πip2 τ − 2πipz) exp(πim2 τ ) exp(2πimz)
m∈Z
def
= exp(−πip2 τ − 2πipz)θ(z)
Hence the basic theta function θ is quasi-periodic with
θ(z + λ) = θ(z + pτ + q)
= exp(−πip2 τ − 2πipz)θ(z)
for all λ = pτ + q ∈ Λ and z ∈ C.
Definition. We define
e(λ, z) := exp(−πip2 τ − 2πipz)
and call this the automorphy f actor.
Remark. We have e(λ1 + λ2 , z) = e(λ1 , z + λ2 )e(λ2 , z) for all λ1 ,
λ2 ∈ Λ.
THETA FUNCTIONS 57
Let λ1 , λ2 ∈ Λ, i.e. λ1 = p1 τ + q1 and λ2 = p2 τ + q2 for some p1 ,

p2 , q1 , q2 ∈ Z, and thus λ1 + λ2 = (p1 + p2 )τ + (q1 + q2 ) ∈ Λ. Then
e(λ1 + λ2 , z) = e((p1 + p2 )τ + (q1 + q2 ), z)

def
= exp(−πi(p1 + p2 )2 τ − 2πi(p1 + p2 )z))
= exp(−πip21 τ − 2πip1 p2 τ − πip22 τ − 2πip1 z − 2πip2 z)
def
= exp(−πip21 τ − 2πip1 p2 τ − 2πip1 z)e(λ2 , z)
= exp(−πip21 τ − 2πip1 z − 2πip1 p2 τ − 2πip1 q2 )
| {z }
exp(2πip1 q2 )=1
· e(λ2 , z)
= exp(−πip21 τ − 2πip1 (z + λ2 ))e(λ2 , z)
def
= e(λ1 , z + λ2 )e(λ2 , z)
Summary. The basic theta function θ : C → C is entire and quasi-

periodic with automorphy factor e, i.e. we have
θ(z + λ) = e(λ, z)θ(z) = exp(−πip2 τ − 2πipz)θ(z) (1)
for all λ = pτ + q ∈ Λ and all z ∈ C.
Now we want to enlarge our cathegory of theta functions. So far

we have only one (basic) theta function corresponding to the point
0 ∈ C (and each point q ∈ Z ⊂ C). Now, for our fixed τ , we will
define a new theta function for each point in C. Therefore let’s start
with our old theta function and translate z by a fixed ξ, i.e. consider
θ(z + ξ) for ξ = aτ + b for some fixed a, b ∈ R:
θ(z + ξ) = θ(z + aτ + b)
def X
= exp(πin2 τ ) exp(2πin(z + aτ + b))
n∈Z
X
= exp(πin2 τ + 2πinz + 2πinaτ + 2πinb)
n∈Z
If we complete the square and rearrange the summands then we

obtain
58 MARINA FRANZ
X
θ(z + ξ) = exp πin2 τ + 2πinaτ + πia2 τ − πia2 τ
n∈Z
+2πin(z + b) + 2πia(z + b) − 2πia(z + b))
= exp(−πia2 τ − 2πia(z + b))
X
· exp(πi(n + a)2 τ ) exp(2πi(n + a)(z + b))
n∈Z
P
Note that the sum n∈Z exp(πi(n + a)2 τ ) exp(2πi(n + a)(z + b))
looks very similar to the sum in the definition of our basic theta
function above.
Definition. For ξ = aτ +b and a, b ∈ R the modif ied thetaf unction

is defined to be the function θξ : C → C given by
X
θξ (z) := θξ (τ )(z) := exp(πi(n + a)2 τ ) exp(2πi(n + a)(z + b))
n∈Z
and ξ is called theta characteristic.
Note. From the calculation above we obtain a relation between the

basic theta function and the modified theta function with character-
istic ξ = aτ + b for some fixed a, b ∈ R:
X
θξ (z) = exp(πi(n + a)2 τ ) exp(2πi(n + a)(z + b)) (2)
n∈Z
= exp(πia2 τ + 2πia(z + b))θ(z + ξ) (3)
for all z ∈ C.
Remark. As the series in the definition is locally uniformly un-

ordered convergent (without proof) the modified theta functions are
entire functions.
Lemma 3. Modified theta functions are quasi-periodic functions.

THETA FUNCTIONS 59
Proof. Let a, b ∈ R such that ξ = aτ + b is the characteristic of the

modified theta function θξ . Consider θξ (z + λ) for λ = pτ + q ∈ Λ.
(3)
θξ (z + λ) = exp(πia2 τ + 2πia(z + λ + b))θ(z + λ + ξ)
(1)
= exp(πia2 τ + 2πia(z + λ + b))e(λ, z + ξ)θ(z + ξ)
(3)
= exp(πia2 τ + 2πia(z + λ + b))e(λ, z + ξ)
· exp(−πia2 τ − 2πia(z + b))θξ (z)
= exp(2πiaλ) exp(−πip2 τ − 2πip(z + ξ))θξ (z)
= exp(2πiaλ − πip2 τ − 2πip(z + ξ))θξ (z)
Hence the modified theta function θξ is quasi-periodic with
θξ (z + λ) = θaτ +b (z + pτ + q)
= exp(2πiaλ − πip2 τ − 2πip(z + ξ))θξ (z)
for all λ = pτ + q ∈ Λ and z ∈ C.
Definition. Let a, b ∈ R be fixed and let ξ = aτ + b. We define
eξ (λ, z) := exp(2πiaλ − πip2 τ − 2πip(z + ξ))
and call this the automorphy f actor.
Remark. Let a, b ∈ R be fixed and let ξ = aτ + b. We have

eξ (λ1 + λ2 , z) = eξ (λ1 , z + λ2 )eξ (λ2 , z) for all λ1 , λ2 ∈ Λ.
Let λ1 , λ2 ∈ Λ, i.e. λ1 = p1 τ + q1 and λ2 = p2 τ + q2 for some p1 ,
p2 , q1 , q2 ∈ Z, and λ1 + λ2 = (p1 + p2 )τ + (q1 + q2 ) ∈ Λ.
60 MARINA FRANZ
eξ (λ1 + λ2 , z) = eξ ((p1 + p2 )τ + (q1 + q2 ), z)

def
= exp 2πia(λ1 + λ2 ) − πi(p1 + p2 )2 τ
−2πi(p1 + p2 )(z + ξ))
= exp 2πiaλ1 + 2πiaλ2 − πip21 τ − 2πip1 p2 τ − πip22 τ
−2πip1 (z + ξ) − 2πip2 (z + ξ))
def
= exp(2πiaλ1 − πip21 τ − 2πip1 p2 τ − 2πip1 (z + ξ))
eξ (λ2 , z)
= exp 2πiaλ1 − πip21 τ − 2πip1 p2 τ − 2πip1 q2
| {z }
exp(2πip1 q2 )=1
−2πip1 (z + ξ)) eξ (λ2 , z)

= exp(2πiaλ1 − πip21 τ − 2πip1 (z + λ2 + ξ))eξ (λ2 , z)
def
= eξ (λ1 , z + λ2 )eξ (λ2 , z)
Summary. Let ξ = aτ + b with a, b ∈ R fixed. The modified

theta function with characteristic ξ is entire and quasi-periodic with
automorphy factor eξ , i.e. we have
θξ (z + λ) = eξ (λ, z)θξ (z) (4)

= exp(2πiaλ − πip2 τ − 2πip(z + ξ))θξ (z) (5)
for all λ = pτ + q ∈ Λ and all z ∈ C.
Now we want to determine all zeros of all theta functions. There-

fore we consider a special modified theta function, the theta function
with characteristic σ := 21 τ + 21 . In this case the determination of
zeros is very simple because the zeros are easy to describe.
Lemma 4. θσ is an odd function, i.e. θσ (−z) = −θσ (z) for all

z ∈ C.
In particular we have θσ (0) = 0.
THETA FUNCTIONS 61
Proof.
θσ (−z) = θ 12 τ + 21 (−z)
" 2 !
def X 1
= exp πi n + τ
2
n∈Z

1 1
exp 2πi n + −z +
2 2
if we make a simple index shift m = −n − 1 then
" 2 !
X 1
= exp πi −m − τ
2
m∈Z

1 1
exp 2πi −m − −z +
2 2
" !
X 2
1
= exp πi m + τ
2
m∈Z

1 1 1
exp 2πi m + z+ − 2πi m +
2 2 2
" !
X 2
1
= exp πi m + τ
2
m∈Z

1 1
exp 2πi m + z+
2 2
exp(−2πim) exp(−πi) ]
| {z }
=−1 for all m∈Z
def
= −θσ (z)

From complex analysis we know a simple way to count zeros and

poles of a meromorphic function f : C → C:
Z ′
1 f
(z) dz = total number of zeros - total number of poles
2πi γ f
where γ is a piecewise smooth path that runs around each zero and
each pole exactly one time. We will use this integral to determine
all zeros of the theta functions θσ with σ = 12 τ + 12 .
62 MARINA FRANZ
Lemma 5. We have θσ (z) = 0 precisely for all z ∈ Λ and all zeros

are simple zeros.
Proof. Consider the fundamental parallelogram V := {z ∈ C : z =

t1 τ + t2 for some 0 ≤ t1 , t2 < 1}.
Choose w ∈ C such that the border of Vw := w + V contains no
zeros of θσ and 0 ∈ Vw .
Further consider the following paths along the border of Vw :
α : [0, 1] → C; t 7→ w + t
β : [0, 1] → C; t 7→ w + 1 + tτ
γ : [0, 1] → C; t 7→ w + (1 − t) + τ
δ : [0, 1] → C; t 7→ w + (1 − t)τ
w+τ γ w+1+τ

×0
δ
β

-
w α w+1
Figure 2. w ∈ C is chosen such that the border of

the parallelogram Vw = w + V contains no zeros of
f and such that 0 ∈ Vw . The paths α, β, γ and δ
run along the border of Vw .
Note
γ(t) = w + (1 − t) + τ = α(1 − t) + τ
and
δ(t) = w + (1 − t)τ = β(1 − t) − 1
Z
1 θσ′
We want to show that (z) dz = 1.
2πi ∂Vw θσ
THETA FUNCTIONS 63
Therefore we will show
Z Z
1 θσ′ 1 θσ′
(z) dz = 1 − (z) dz
2πi γ θσ 2πi α θσ
and
Z Z
1 θσ′ 1 θσ′
(z) dz = − (z) dz
2πi δ θσ 2πi β θσ
Z Z 1
1 θσ′ 1 θσ′
(z) dz = (γ(t))γ ′ (t) dt
2πi γ θσ 2πi 0 θσ
1 ′Z
1 θσ
= (α(1 − t) + τ )(−1) dt
2πi 0 θσ
Z ′
1 θσ
=− (z + τ ) dz
2πi α θσ
Z ′
1 eσ (τ, z)θσ (z) + eσ (τ, z)θσ′ (z)
=− dz
2πi α eσ (τ, z)θσ (z)
Z ′ Z ′
1 eσ (τ, z) 1 θσ
=− dz − (z) dz
2πi α eσ (τ, z) 2πi α θσ
when we use
1
eσ (τ, z) = exp(2πi τ − πiτ − 2πi(z + σ))
2
then
Z
1 exp′ (−2πi(z + σ))
=− dz
2πi α exp(−2πi(z + σ))
Z ′
1 θσ
− (z) dz
2πi α θσ
Z Z ′
1 1 θσ
=− −2πi dz − (z) dz
2πi α 2πi α θσ
Z ′
1 θσ
=1− (z) dz
2πi α θσ
64 MARINA FRANZ
Z Z 1
1 θσ′ 1 θσ′
(z) dz = (δ(t))δ ′ (t) dt
2πi δ θσ 2πi 0 θσ
Z
1 ′
1 θσ
= (β(1 − t) − 1)(−τ ) dt
2πi 0 θσ
Z ′
1 θσ
=− (z − 1) dz
2πi β θσ
Z ′
1 eσ (−1, z)θσ (z) + eσ (−1, z)θσ′ (z)
=− dz
2πi β eσ (−1, z)θσ (z)
Z ′ Z ′
1 eσ (−1, z) 1 θσ (z)
=− dz − dz
2πi β eσ (−1, z) 2πi β θσ (z)
when we use
1
eσ (−1, z) = exp(−2πi )
2
then
Z Z ′
1 exp′ (−πi) 1 θσ (z)
=− dz − dz
2πi β exp(−πi) 2πi β θσ (z)
Z ′
1 θσ (z)
=− dz
2πi β θσ (z)
Then we have
Z Z ′ Z ′
1 θσ′ 1 θσ 1 θσ
(z) dz = (z) dz + (z) dz
2πi ∂Vw θσ 2πi α θσ 2πi β θσ
Z ′ Z ′
1 θσ 1 θσ
+ (z) dz + (z) dz
2πi γ θσ 2πi δ θσ
=1
As θσ is holomorphic in Vw , i.e. it doesn’t have any poles, we know

that θσ has a single zero. And by Lemma 4 this zero is in 0.
Now consider V w + λ = V w+λ for some λ ∈ Λ. As θσ (z + λ) =
eσ (λ, z)θσ (z) we obtain that θσ has the only zero 0 + λ = λ in V w+λ
and this is a simple zero. But C = ∪λ∈Λ V w+λ . Hence θσ has zeros
exactly in Λ and all zeros are simple.
Corollary 6. Let ξ = aτ + b with a, b ∈ R. We have θξ (z) = 0

precisely for all z ∈ σ − ξ + Λ and all its zeros are simple.
THETA FUNCTIONS 65
Proof. We know θσ (z) = 0 if and only if z ∈ Λ and all the zeros are
simple. Hence
(3)
θξ (z) = 0 ⇔ exp(πia2 τ + 2πia(z + b))θ(z + ξ) = 0
(3)
⇔ exp πia2 τ + 2πia(z + b)
2
1
· exp −πi τ
2

1 1 1 1
−2πi z+ξ− τ − +
2 2 2 2
· θσ (z + ξ − σ) = 0
⇔z+ξ−σ ∈Λ
⇔z ∈σ−ξ+Λ
In particular we have θ(z) = 0 if and only if z ∈ σ + Λ.
So far we have considered entire quasi-periodic functions. Now

we want to use our knowledge about them to see what meromorphic
doubly-periodic functions with given zeros ai and poles bj of given
order ni resp. mj and number n resp. m look like. Futhermore we
will decide whether such a function exists or not and whether it is
unique or not.
Abel’s Theorem 7. There is a meromorphic function on C/Λ with
≤ i ≤ n and
zeros [ai ] of order ni for 1 P Pmpoles [bj ] ofP
order mj for
n n
1 ≤ j ≤ m if and only if i=1 ni = j=1 mj and i=1 ni [ai ] =
Pm
j=1 mj [bj ].
Moreover, such a function is unique up to a constant factor.
Proof. “⇒” Let f : C/Λ → C be a meromorphic function with zeros
[ai ] of order ni and poles [bj ] of order mj . Choose w ∈ C such that
Vw = {w + z ∈ C : z = t1 τ + t2 for some 0 ≤ t1 , t2 < 1} contains
a representative ai resp. bj for every zero resp. pole of f . Further
consider the paths
α : [0, 1] → C; t 7→ w + t
β : [0, 1] → C; t 7→ w + 1 + tτ
γ : [0, 1] → C; t 7→ w + (1 − t) + τ
δ : [0, 1] → C; t 7→ w + (1 − t)τ
66 MARINA FRANZ
along the border of Vw and the paths

αi : [0, 1] → C; t 7→ ai + ri e2πit
βj : [0, 1] → C; t 7→ bj + sj e2πit
around the zeros resp. poles of f where ri resp. sj is chosen small
enough that Di = {z ∈ C : |z − ai | < ri } resp. Dj′ = {z ∈ C :
|z − bj | < sj } contains no other zeros or poles of f .
w+τ γ w+1+τ

×
D a α
0 i0 6 i0
a × i
i1

δ bj× β
1

Dj′ 0bj× βj0
0 6
×
a
- i2
w α w+1
Figure 3. w ∈ C is chosen such that the parallelo-

gram Vw = w + V contains a representative ai resp.
bj for every zero resp. pole of f . The paths α, β,
γ and δ run along the border of Vw , the paths αi0
around the zero ai0 of f and the path βj0 around
the pole bj0 of f .
Pn P
First we show that i=1n i ai − m j=1 mj bj ∈ Λ as follows:
Xn Xm Xn Z Xm Z
1 f′ 1 f′
n i ai − mj b j = z (z) dz + z (z) dz
i=1 j=1 i=1
2πi α i
f j=1
2πi β j
f
Z
1 f′
= z (z) dz ∈ Λ
2πi ∂Vw f
To establish the first equality note that we can write
f (z) = ci (z − ai )ni hi (z)
for a constant ci and with hi (ai ) = 1 around ai and hence
f ′ (z) = ci ni (z − ai )ni −1 hi (z)
THETA FUNCTIONS 67
with hi (ai ) = 1. We obtain

f′ ni hi
z (z) = z (z)
f z − ai h i
hi
with hi (ai ) = 1. Hence we have
Z
1 f′
z (z) dz = ni ai
2πi αi f
by Cauchy’s integral formula for discs. The same holds for the poles
of f .
The second equality is clear since Vw contains a representative for
every zero and pole Zof f in C/Λ.
1 f′
To see, that z (z) dz is an element of Λ, note that
2πi ∂Vw f
Z Z 1
1 f′ 1 f′
z (z) dz = γ(t) (γ(t))γ ′ (t) dt
2πi γ f 2πi 0 f
Z 1
1 f′
= (α(1 − t) + τ ) ((α(1 − t) + τ ))(−1) dt
2πi 0 f
Z 1
1 f′
=− α(1 − t) (α(1 − t)) dt
2πi 0 f
Z 1 ′
1 f
− τ (α(1 − t)) dt
2πi 0 f
Z Z ′
1 f′ 1 f
=− z (z) dz − τ (z) dz
2πi α f 2πi α f
and
Z Z 1
1 f′ 1 f′
z (z) dz = δ(t) (δ(t))δ ′ (t) dt
2πi δ f 2πi 0 f
Z 1
1 f′
= (β(1 − t) − 1) ((β(1 − t) − 1))(−τ ) dt
2πi 0 f
Z 1
1 f′
=− β(1 − t) (β(1 − t))τ dt
2πi 0 f
Z 1 ′
1 f
+ (β(1 − t))τ dt
2πi 0 f
Z Z ′
1 f′ 1 f
=− z (z) dz + (z) dz
2πi β f 2πi β f
68 MARINA FRANZ
hence
Z Z Z
1 f′ 1 f′ 1 f′
z (z) dz = z (z) dz + z (z) dz
2πi ∂Vw f 2πi α f 2πi β f
Z Z
1 f′ 1 f′
+ z (z) dz + z (z) dz
2πi γ f 2πi δ f
Z ′ Z ′
1 f 1 f
= −τ (z) dz + (z) dz ∈ Λ
2πi α f 2πi β f
1
R f′ 1
R f′
since 2πi β f (z) dz, 2πi α f (z) dz ∈ Z.
Secondly we show that
Xn Xm Z
1 f′
ni − mj = (z) dz
i=1 j=1
2πi ∂Vw
f
=0
Again the first equality is clear, since Vw contains a representative
for every zero and pole of f in C/Λ.
The second equality follows from:
Z ′ Z 1 ′
1 f 1 f
(z) dz = (γ(t))γ ′ (t) dt
2πi γ f 2πi 0 f
Z 1 ′
1 f
= (α(1 − t) + τ )(−1) dt
2πi 0 f
Z 1 ′
1 f
=− (α(1 − t)) dt
2πi 0 f
Z ′
1 f
=− (z) dz
2πi α f
and
Z Z 1
1 f′ 1 f′
(z) dz = (δ(t))δ ′ (t) dt
2πi δ f 2πi 0 f
Z 1
1 f′
= (β(1 − t) − 1)(−τ ) dt
2πi 0 f
Z 1 ′
1 f
=− (β(1 − t))τ dt
2πi 0 f
Z ′
1 f
=− (z) dz
2πi β f
THETA FUNCTIONS 69
hence
Z Z Z ′
1 f′ 1 f′ 1 f
(z) dz = (z) dz + (z) dz
2πi ∂Vw f 2πi α f 2πi β f
Z ′ Z ′
1 f 1 f
+ (z) dz + (z) dz
2πi γ f 2πi δ f
=0
“⇐” Now let [ai ], [bj ] ∈PC/Λ and nP i , mj ∈ N for 1P ≤ i ≤ n and

n m n
1 ≤ j ≤ m be such that i=n ni = j=m mj and i=1 ni [ai ] =
Pm
j=1 mj [bj ]. We will contruct a meromorphic function f : C/Λ → C
with zeros [ai ] of order ni and poles [bj ] of order mj .PWe choose
n
representatives
Pm a i , b j ∈ C for [a i ] resp. [b j ] such that i=1 ni ai =
j=1 mj bj and define the function
Qn ni
i=1 θσ (z − ai )
g : C → C; z →
7 Qm mj
j=1 θσ (z − bj )
where θσ is the theta function with characteristic 12 τ + 21 . Obviously

g is a meromorphic function with zeros in ai +Λ of order ni and poles
in bj + Λ of order mj . We have to show that g is doubly-periodic
w.r.t. Λ. Therefore we have to show that g(z + λ) = g(z) for all
λ ∈ Λ. It suffices to show that g(z + 1) = g(z) and g(z + τ ) = g(z).
Qn ni
Qn ni
i=1 θ σ (z + 1 − a i ) i=1 θσ (z − ai )
g(z + 1) = Qm mj
= Qm mj
= g(z)
j=1 θ σ (z + 1 − b j ) j=1 θ σ (z − b j )
and
Qn ni
i=1 θσ (z + τ − ai )
Qm
g(z + τ ) = mj
j=1 θσ (z + τ − bj )
Qn ni
i=1 (eσ (τ, z − ai )θσ (z − ai ))
Qm
= mj
j=1 (eσ (τ, z − bj )θσ (z − bj ))
Qn Qn
i=1 eσ (τ, z − ai )ni i=1 θσ (z − ai )ni
= Qm mj
Qm mj
j=1 eσ (τ, z − bj ) j=1 θσ (z − bj )
Qn ni
i=1 eσ (τ, z − ai )
= Qm mj
· g(z)
j=1 e σ (τ, z − b j )
70 MARINA FRANZ
but
Qn ni
Qn
i=1 e σ (τ, z − a i ) i=1 exp(−2πi(z − ai + σ))ni
Qm mj
= Qm mj
j=1 e σ (τ, z − b j ) j=1 exp(−2πi(z − bj + σ))
Qn ni
Q i=1 exp(−2πi(z + σ))
= m mj
j=1 exp(−2πi(z + σ))
Qn ni
i=1 exp(2πiai )
· Qm mj
j=1 exp(2πibj )
Pn
ni
exp(−2πi(z + σ)) i=1
= Pm
exp(−2πi(z + σ)) j=1 mj
Pn
exp(2πi i=1 ni ai )
· Pm
exp(2πi j=1 mj bj )
=1
So g(z + τ ) = g(z) as well. Hence g is doubly periodic w.r.t. Λ and
the function f : C/Λ → C with f ([z]) = g(z) is well-defined and a
solution.
Now suppose we are given two meromorphic functions f , g :
C/Λ → C with zeros [ai ] of order ni and poles [bj ] of order mj .
Then fg has no zeros or poles. Hence it is constant.
3. Weierstraß ℘-function
Now we want to capitalize on our work above. Therefore we con-
sider a very special periodic function, the Weierstraß ℘-function.
Definition. The W eierstraß ℘ − f unction is defined to be the
function ℘ : C → C given by
1 X 1 1

℘(z) = 2 + − 2
z (z − λ)2 λ
06=λ∈Λ
Proposition 8. (without proof ) ℘ is a Λ-periodic meromorphic func-

tion with poles of order 2 exactly in Λ.
The following lemma gives a connection between the Weierstraß
℘-function and our well known theta function with characteristic
σ = 12 + 21 τ .
Lemma 9. There is a constant c ∈ C such that
′ ′
θ
℘(z) = − σ (z) + c
θσ
THETA FUNCTIONS 71
θ′
Note. The quotient θσσ isn’t doubly-periodic, but the derivative
′ ′
θσ
θσ is doubly-periodic.
′
θσ
To see this consider θσ (z + λ) for some λ = pτ + q ∈ Λ.
′
θσ′ (5) (eσ (λ, z)θσ (z))
(z + λ) =
θσ eσ (λ, z)θσ (z)
eσ (λ, z)θσ (z) + eσ (λ, z)θσ′ (z)
′
=
eσ (λ, z)θσ (z)
′ 2 ′
def exp (πiλ − πip τ − 2πip(z + σ))θσ (z) + eσ (λ, z)θσ (z)
=
eσ (λ, z)θσ (z)
−2πipeσ (λ, z)θσ (z) + eσ (λ, z)θσ′ (z)
=
eσ (λ, z)θσ (z)
θ′
= −2πip + σ (z)
θσ
′
θ
6= σ (z)
θσ
θ′ ′
θσ
as in general p 6= 0. From the equation θσσ (z + λ) = −2πip + θσ (z)
′ ′
θ
above it follows directly that θσσ is doubly-periodic.
Proof. We know that θσ is holomorphic and has its zeros precisely

θ′
in the lattice points λ ∈ Λ. That means that the expansion of θσσ in
a Laurent series around 0 looks like
θσ′ 1
(z) = a−1 + a0 + a1 z + a2 z 2 + a3 z 3 + terms of higher order
θσ z
for some constants ai ∈ C. We can choose a neighborhood U of 0
such that 0 is the only zero of θσ in U . As 0 is a single zero we know
that ′ Z ′
θσ θσ
a−1 = Res0 = (z) dz = 1
θσ α θσ
where α : [0, 1] → C; t 7→ re2πit for some suitable r. We conclude
θσ′ 1
(z) = + a0 + a1 z + a2 z 2 + a3 z 3 + terms of higher order
θσ z
and calculate
′ ′
θσ 1
(z) = − 2 + a1 + 2a2 z + 3a3 z 2 + terms of higher order
θσ z
72 MARINA FRANZ
′
′
θσ
If we add ℘ and θσ then we obtain
′ ′ X
θσ 1 1
℘(z)+ (z) = 2
− 2 + a1 + 2a2 z + 3a3 z 2 + . . .
θσ (z − λ) λ
06=λ∈Λ
′ ′
θ
From this sum we see directly that ℘ + θσσ doesn’t have any poles
′ ′
θ
in U . Hence ℘ + θσσ is holomorphic in a neighborhood of 0 and
thus holomorphic everywhere.
As it is in addition doubly-periodic
′
θ′
(since ℘ is as well as θσσ doubly-periodic) we know from our very
first lemma that it must be constant.
The Weierstraß ℘-function satisfies a number of equations and
differential equation. This feature makes the Weierstraß ℘-function
to be of interest. The most important differential equation that is
satisfied by the Weierstraß ℘-function is the following:
Theorem 10. The Weierstraß ℘ -function satisfies the differential
equation
℘′ (z)2 = c3 ℘(z)3 + c2 ℘(z)2 + c1 ℘(z) + c0
where the constants
X 1 X 1
c3 = 4 , c2 = 0 , c1 = −60 and c 0 = −140
λ4 λ6
06=λ∈Λ 06=λ∈Λ
depend on the lattice Λ.

P
1 1 1
Proof. Consider ℘(z) − = 06=λ∈Λ
z2 − . This function
(z−λ)2 λ2
is holomorphic in a neighborhood of 0. We can expand the sum-
1 1
mands (z−λ) 2 − λ2 :
!
1 1 1 1
2
− 2 = 2 2 − 1
(z − λ) λ λ 1 − λz
 !2 
1
∞
X z n
= 2 − 1
λ n=0
λ

1 z z2 z3 z4
= 2 2 + 3 2 + 4 3 + 5 4 + ...
λ λ λ λ λ
z z2 z3 z4
= 2 3 + 3 4 + 4 5 + 5 6 + ...
λ λ λ λ
THETA FUNCTIONS 73
This sum is absolutly convergent for all z ∈ C with |z| < |λ|; in
particular in a neigborhood of 0. P 1
To simplify the big sum from above we define sn := 06=λ∈Λ λn+2
for n ∈ N. Note that sn = 0 for all odd n ∈ N. We obtain
1
℘(z) = 2 + 2s1 z + 3s2 z 2 + 4s3 z 3 + 5s4 z 4 + . . .
z
1
= 2 + 3s2 z 2 + 5s4 z 4 + 7s6 z 6 . . .
z
which is true in a neigborhood of 0. With the constants
X 1 X 1
c3 = 4 , c2 = 0 , c1 = −60 and c 0 = −140
λ4 λ6
06=λ∈Λ 06=λ∈Λ
we obtain
1 c1 2 c0 4
℘(z) = − z − z + terms of higher order
z2 20 28
hence
2 c1 c0 3
℘′ (z) = − − z − z + terms of higher order
z 3 10 7
4 2c1 1 4c0
℘′ (z)2 = 6 + + + terms of higher order
z 5 z2 7
and
1 3c1 1 3c0
℘(z)3 = − − + terms of higher order
z6 20 z 2 28
Now consider
f (z) := ℘′ (z)2 − c3 ℘(z)3 − c1 ℘(z) − c0
The series of f has only positive powers of z. Hence f is holomorphic
around 0. Hence it is holomorphic everywhere. And as it is doubly-
periodic, it is constant. But the constant part of the series is 47 c0 +
3
4 · 28 c0 − c0 = 0. Hence f = 0.
We will mention one more equation that is satisfied by the Weier-

straß ℘-function:
Remark. Remember that our lattice Λ is generated by 1 and τ.
Hence the set of
zeros of ℘′ is
given by 12 +Λ ∪ τ2 + Λ ∪ 1+τ
2 +Λ .
1 τ 1+τ
Set e1 := ℘ 2 , e2 := ℘ 2 , e3 := ℘ 2 ∈ C. Then we have
(℘′ )2 = 4(℘ − e1 )(℘ − e2 )(℘ − e3 )
74 MARINA FRANZ
and
e1 + e2 + e3 = 0
1
e 1 e 2 + e 1 e 3 + e 2 e 3 = c1
4
1
e 1 e 2 e 3 = − c0
4
where c0 and c1 are the constants from above.
Finally we will see how to use the Weierstraß ℘-function to give
a group structure to an elliptic curve.
Remark. If we consider the elliptic curve
C := {(x, y) ∈ C2 such that y 2 = c3 x3 + c2 x2 + c1 x + c0 }
for the constants
X 1 X 1
c3 = 4 , c2 = 0 , c1 = −60 and c0 = −140
λ4 λ6
06=λ∈Λ 06=λ∈Λ
from the theorem above then we have a bijection

C/Λ \ {0} → C given by z 7→ (℘(z), ℘′ (z))
In particular we can give the variety C the group structure of C/Λ.
This can be extended to an embedding of C/Λ into the projective
plane. For more details see the article of M. Khalid [2].
References
[1] A. Gathmann Algebraic Geometry, Notes for a class taught at the Univer-
sity of Kaiserslautern 2002/2003, available at http://www.mathematik.uni-
kl.de/ gathmann/de/pub.html
[2] M. Khalid Group Law on the Cubic Curve, this issue
[3] D. Mumford Tata Lectures on Theta, Progress in Mathematics Vol 28,
Birkhauser Verlag, 1983
[4] G. Trautmann Complex Analysis II, Notes for a class taught at the Univer-
sity of Kaiserslautern 1996/1997
Marina Franz, Fachbereich Mathematik, Technische Universität Kai-

serslautern, Postfach 3049, 67653 Kaiserslautern, Germany.
E-mail address: franz@mathematik.uni-kl.de
RANK TWO VECTOR BUNDLES ON ELLIPTIC
CURVES
CIARA DALY
Abstract. The aim of this paper is to give an overview of

rank two vector bundles on an elliptic curve. It also aims to
provide an outline of stability of vector bundles to serve as a
motivation for the study of moduli spaces of vector bundles
over elliptic curves. The first section outlines basic defini-
tions and theorems. We will then study vector bundles on
P1 . From here, we go on to classify indecomposable rank
two vector bundles over an elliptic curve. The final section
introduces the notion of stability of vector bundles.
1. Introduction
Nowadays, vector bundles play an inportant role in many areas
of mathematics such as algebraic geometry, algebraic topology and
differential geometry, in the theory of partial differential equations.
The theory of vector bundles and the mathematical formalism
developed over the years, for the study of vector bundle related con-
cepts leads to the clarification or solution of many mathematical
problems. Some of the vector bundle related concepts are general-
isations of well-known classical notions. For instance, the notion of
a section of a vector bundle over a space X is a generalization of a
vector valued function on X.
One of the important problems is the problem of classification
of bundles. The problem of classification of vector bundles over an
elliptic curve (i.e. a nonsingular projective curve of arithmetic genus
one) has been completely solved by Atiyah in [1].
2. Preliminaries
For the purpose of this paper, X will denote a complex manifold,
unless otherwise specified. We will be working over C, the field of
complex numbers, throughout this paper.
75
76 CIARA DALY
Definition 2.1. A complex vector bundle of rank n is a holomorphic

map p : E → X of complex manifolds which satisfy the following
conditions:
(1) For any point x ∈ X, the preimage Ex := p−1 (x) (called a
fibre) has a structure of an n-dimensional C-vector space.
(2) The mapping p is locally trivial, i.e. for any point x ∈ X,
there exists an open neighbourhood Ui containing x and a
biholomorphic map ϕi : p−1 (Ui ) → Ui × Cn such that the
diagram
ϕi
p−1 (Ui ) / U i × Cn
GG w
GG
GG www
p GG
w
G# www pr1
{w
Ui
commutes.
Moreover, ϕi takes the vector space Ex isomorphically onto {x} ×
n
C for each x ∈ Ui ; ϕi is called a trivialisation of E over U . Note
that for any pair of trivialisations ϕi and ϕj , the map
gij : Ui ∩ Vj → GL(n, C)
given by
gij (x) = ϕi ◦ (ϕj |{x}×Cn )−1
is holomorphic; the maps gij are called transition functions for E
relative to the trivialisations ϕi , ϕj . The transition functions of E
necessarily satisfy the identities
gij (x) · gji (x) = I for all x ∈ Ui ∩ Uj
gij (x) · gjk (x) · gki (x) = I for all x ∈ Ui ∩ Uj ∩ Uk .

Conversely, given an open cover {Ui } of X and transition func-
tions gij : Ui ∩ Uj → GL(n, C), for all i, j, then we can define
a vector bundle, E with transition functions gij using the glueing
construction as follows: We glue Ui × Cn F together by taking the
union over all i of Ui × C to get E := (Ui × Cn )/ ∼, where
n
(x, v) ∼ (x, gij (x)(v)), for all x ∈ Ui ∩ Uj , v ∈ Cn .

A vector bundle of rank 1 is called a line bundle (See [7] Section
5 for more details).
RANK TWO VECTOR BUNDLES ON ELLIPTIC CURVES 77
Example 2.2. The simplest example is known as the trivial vec-

tor bundle of rank n, i.e. pr1 : X × Cn → X, where pr1 denotes
projection to the first factor.
The trivial line bundle on X, i.e. X × C → X will be denoted by
OX , (or simply O if it is clear which X we are referring to).
Example 2.3. The set O(−1) ⊂ Pn × Cn+1 that consists of all pairs
(ℓ, z) ∈ Pn ×Cn+1 with z ∈ ℓ forms in a natural way a line bundle over
Pn . To see this, consider the projection p : O(−1)
S → Pn , where p is
n
the projection to the first factor. Let Pn = i=0 Ui be the standard
open covering. A canonical trivialisation of O(−1) over Ui is given
by ϕUi : p−1 (Ui ) ∼
= Ui × C, (ℓ, z) 7→ (ℓ, zi ). The transition functions
gij (ℓ) : C → C are given by w 7→ zzji · w, where ℓ = (z0 : · · · : zn ).
Definition 2.4. Let p : E → X and p′ : E ′ → X be two complex

vector bundles on X. A holomorphic map f : E → E ′ is called a
morphism of vector bundles if the diagram
f
E@ / E′
@@ }
@@ }}}
p @@@ }} ′
~}} p
X
commutes and for each point x ∈ X the map f |Ex : Ex → Ex′ is a
homomorphism of vector spaces.
Let E and E ′ be vector bundles over X with rank r and r′ , re-
spectively and let {Ui } be an open cover of X such that E and E ′ are
trivial over Ui for each i. A morphism f : E → E ′ can be described
locally by holomorphic functions, fi , as follows. For each i, using
trivialisations of E and E ′ , f induces maps
′
U i × Cr → U i × Cr , (x, v) 7→ (x, fi (x)v)
where fi : Ui → Matr′ ×r (C). These holomorphic functions necessar-
ily satisfy
′
fi (x) · gij (x) = gij (x) · fj (x) for all x ∈ Ui ∩ Uj
′
where gij and gij are transition functions of E and E ′ , respectively.
Note that a set of functions {fi } defines an isomorphism of vector
bundles if an only if fi (x) are invertible matrices for all i and x.
78 CIARA DALY
Remark 2.5. It is important to note that this definition of a mor-

phism of vector bundles does not make the category of vector bundles
into an abelian category. For instance, if f : E → E ′ is a morphism
of vector bundles and the rank of f is non-constant then dim(ker(fx ))
jumps and so ker f cannot form a vector bundle. The same would be
true for coker f . On the contrary if dim(ker(fx )) is constant on x,
then both ker f and coker f are vector bundles. This can be shown
locally by a rank argument. It is necessary to use strict morphisms
if an abelian category is required.
Definition 2.6. Let E be a vector bundle on X and let {Ui } be an
open cover of X. If the transition functions of E are gij , then the
dual bundle, E ∗ , of E is given by transition functions
hij (x) := t gij (x)−1 ∀x ∈ Ui ∩ Uj
Definition 2.7. A sequence of morphisms of vector spaces
f g
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
is called a short exact sequence if ker g = im f , and if f is injective
and g is surjective.
Definition 2.8. A sequence of morphisms of vector bundles over X
0 −→ E ′ −→ E −→ E ′′ −→ 0
is an exact sequence of vector bundles if
0 −→ Ex′ −→ Ex −→ Ex′′ −→ 0
is an exact sequence of vector spaces for all x ∈ X. The vector
bundle E ′ is called a subbundle of E, and E ′′ is called a quotient
bundle of E.
We also say that an exact sequence of vector bundles
0 −→ E ′ −→ E −→ E ′′ −→ 0
is an extension of E ′′ by E ′ . In this case E is called an extension of
E ′′ by E ′ .
Definition 2.9. Let U be an open set in X. A holomorphic map
s : U → E is called a holomorphic section of E over U if p ◦ s = idU .
Sections over X are called global sections of E. Global sections can
be added and multiplied with a scalar, so the space of global sections
is in fact a vector space. It will be denoted by H 0 (X, E).
Remark 2.10. Let f : E ′ → E be a morphism of vector bundles. This

induces a linear map of spaces of sections H 0 (f ) : H 0 (E ′ ) → H 0 (E)
by H 0 (f )(s′ ) := f ◦ s′ .
2.1. Cohomology: Given a short exact sequence of vector bundles
over X
f g
0 −−−−→ E ′ −−−−→ E −−−−→ E ′′ −→ 0
we can take global sections to get an exact sequence
H 0 (f ) H 0 (g) (1)
0 −−−−→ H 0 (X, E ′ ) −−−−→ H 0 (X, E) −−−−→ H 0 (X, E ′′ )
in which the last map H 0 (g) : H 0 (X, E) → H 0 (X, E ′′ ) is not in
general surjective. For a counter example to H 0 (g) being surjective
see [4] Chapter 8.
Since we get the exact sequence (1) above, we say that the global
section functor is left exact. This global section sequence extends to
a long cohomology exact sequence. For any vector bundle E on X,
the natural cohomology groups H i (X, E) (also denoted H i (E) if it
is clear which X we are referring to), for all i > 0, can be defined
satisfying the following property. Given a short exact seqence
0 −→ E ′ −→ E −→ E ′′ −→ 0
of vector bundles, there is an induced long exact sequence of coho-
mology groups
0 → H 0 (X, E ′ ) → H 0 (X, E) → H 0 (X, E ′′ ) → H 1 (X, E ′ ) → · · ·
I will not define these cohomology groups in this paper, except to
note that they exist and are very useful in computations. The di-
mension of H i (X, E) will be denoted hi (X, E). In the case of a curve
X the cohomology groups H i (X, E), vanish for all i > 1, where 1 is
the dimension of X, i.e. only the cohomology groups H 0 (X, E) and
H 1 (X, E) are nonzero. (In fact H i (X, E) vanish for all i > dim X
for a more general X than just a curve though we do not need this
fact in this paper. See [4] Chapter 8 for more details).
2.2. Ext groups: If E and E ′ are vector bundles over X, we denote
by HomX (E, E ′ ) (or Hom(E, E ′ ) if it is clear which X we are refer-
ring to) the vector space of vector bundle morphisms. For a fixed
E, Hom(E, ·) is a left exact covariant functor from the category of
vector bundles to the category of vector spaces, i.e. given a short
exact sequence of vector bundles
0 −→ F ′ −→ F −→ F ′′ −→ 0
80 CIARA DALY
we get another exact sequence in which the last map is not surjective
in general
g
0 −−−−→ Hom(E, F ′ ) −−−−→ Hom(E, F ) −−−−→ Hom(E, F ′′ )
(2)
And so in a similar fashion to the way we defined cohomology
groups H i (E), we can define what are called the Ext groups, which
allow us to extend our short exact sequence (2) to a long exact
sequence as follows:
0 −−−−→ Hom(E, F ′ ) −−−−→ Hom(E, F ) −−−−→ Hom(E, F ′′ )
−−−−→ Ext1 (E, F ′ ) −−−−→ Ext1 (E, F ) −−−−→ ···
We say that Exti (E, ·) are the right derived functors of Hom(E, ·). So
in particular we have Ext0 (E, ·) = Hom(E, ·). We have the following
proposition to see the relationship between the cohomology groups
H i and the Ext groups (for a full proof of this proposition see [5]
Proposition 6.3).
Proposition 2.11. For any vector bundle E on a complex manifold
X we have:
Exti (OX , E) ∼
= H i (E) for all i ≥ 0.
Similarly we have:
Exti (E, OX ) ∼
= H i (E ∗ ) for all i ≥ 0.
Proof. Here we will just give a proof of the first statement, where
i = 0. Let f ∈ Hom(O, E), a fibre-wise holomorphic morphism such
that the following diagram commutes
f
X ×C /E.
GG ~
GG ~
GG ~~~
pr1 GG ~ p
G# ~~
~
X
Let s : X → E be a holomorphic section of E, i.e. p ◦ s = idx . Define
two linear maps
α : Hom(O, E) → H 0 (E)
and
β : H 0 (E) → Hom(O, E)
as follows: Define α(f )(x) := f (x, 1) and β(s)(x, λ) := λ · s(x).
We see that β(α(f )) = f as follows:
β(α(f ))(x, λ) = λ · (α(f )(x)) = λ · f (x, 1) = f (x, λ), where the

last equality uses the fact that f is linear on fibres.
Similarly we obtain α(β(s))(x) = β(s)(x, 1) = s(x), so α(β(s)) =
s. Hence α and β are inverses of one another and we get
Hom(O, E) ∼
= H 0 (E).

Remark 2.12. Given a vector bundle E, and given that H 0 (E) 6= 0,

then from Proposition 2.11 we know that Hom(O, E) 6= 0. So we
have f : O → E. We can say that O ⊂ E, though here ‘⊂’ does
not denote a subbundle, but rather a ‘subsheaf’. Vector bundles
can also be described as sheaves (in particular locally free sheaves)
though we have not built up this language in this paper. It suffices
to know that if O ⊂ E as a subsheaf then we can extend this to get
a line subbundle M ⊂ E on X, a smooth curve (See [9] Chapter 10
for more details). This will be useful in proofs later on.
The notion of a degree of a line bundle on a curve was introduced
in [7] (Section 5). We can extend this definition to vector bundles of
arbitrary rank. To do so we must first define the determinant line
bundle.
Definition 2.13. Given a vector bundle E of rank r, it’s deter-
minant line bundle is defined to be the r-th exterior power of E,
denoted:
det E := ∧r E.
where the fibres of X for any x ∈ X are canonically isomorphic to
∧r Ex .
For those of you who are unfamiliar with exterior power, we can
reformulate the definition as follows: Given an open cover {Ui } of
X and a vector bundle E over X with transition functions gij , the
determinant line bundle of E is given by transition functions hij
where
hij (x) := det gij (x) ∈ GL(1, C), for all x ∈ Ui ∩ Uj
This now allows us to define the degree of a vector bundle as
follows.
Definition 2.14. The degree deg E ∈ Z of a vector bundle is the
degree of its determinant line bundle det E.
82 CIARA DALY
Remark 2.15. Recall from [7] that the degree of a line bundle is the
degree of it’s associated divisor and every line bundle can be written
as O(D), with D a divisor on X.
If E lies in an exact sequence of vector bundles on X as follows:
0 → E ′ → E → E ′′ → 0
then there is an isomorphism
det E ′ ⊗ det E ′′ ∼
= det E.
Since det E ′ and det E ′′ are line bundles we get deg E = deg E ′ +
deg E ′′ (as in general deg(L ⊗ L′ ) = deg L + deg L′ , where L and L′
line bundles). In other words, degree is additive on exact sequences.
As a special case, if a vector bundle E = L1 ⊕ L2 is the direct sum
of two line bundles L1 and L2 , then we have
0 → L1 → E → L2 → 0
with deg E = deg L1 + deg L2 .
Remark 2.16. Let E be a vector bundle of rank r over a complex
manifold X. If we tensor E with a line bundle L, then deg(E ⊗ L) =
deg E + r deg L. In particular for E of rank 2 we have deg(E ⊗ L) =
deg E + 2 deg L. To see this let’s look at an example:
Let E = L1 ⊕ L2 for line bundles L1 and L2 . Tensor this with
another line bundle L to get
E ⊗ L = (L1 ⊕ L2 ) ⊗ L = (L1 ⊗ L) ⊕ (L2 ⊗ L)
Now
deg(E ⊗ L) = deg(L1 ⊗ L) + deg(L2 ⊗ L)
= deg L1 + deg L + deg L2 + deg L
= deg L1 + deg L2 + 2 deg L
= deg E + 2 deg L.
In this way, when considering vector bundles of rank 2, it is enough
to consider vector bundles of degree −1 or 0 (or indeed, any even
and odd degree), then by tensoring with a line bundle of appropriate
degree we get all other degrees. We call this the “tensor product
trick”.
Definition 2.17. An exact sequence of vector bundles
0 → E ′ → E → E ′′ → 0
splits if and only if there exists a homomorphism f : E ′′ → E for

f
which the composition E ′′ /E / E ′′ is an isomorphism.
In this case, the map f is called a splitting of the sequence.
Now consider, on any curve, a short exact sequence of vector
bundles
α β
E: 0 −−−−→ M −−−−→ E −−−−→ L −−−−→ 0.
By applying Hom(L, −) to this sequence we get the following mor-
phism:
δ
Hom(L, L) −−−−→ Ext1 (L, M )
Definition 2.18. The image under the coboundary map δ of idL ∈
Hom(L, L), which we will denote by
δ(idL ) ∈ Ext1 (L, M ) ∼

= H 1 (L∗ ⊗ M ),
is called the extension class of E.
By exactness of
ρ δ
Hom(L, E) −−−−→ Hom(L, L) −−−−→ Ext1 (L, M ),
if δ(idL ) = 0, then there exists a homomorphism f : L → E for
which the idL = β ◦ f : L → L, i.e. the sequence E splits. Moreover,
if E splits, there exists f : L → E such that idL = β ◦ f . Because
the composition, ρ ◦ δ is zero, we have δ(idL ) = 0. Hence we have
the following proposition:
Proposition 2.19. The sequence E, i.e.
0 → M → E → L → 0.
splits if and only if Ext1 (L, M ) = 0. In particular, if Ext1 (L, M ) ∼
=
1 ∗
H (L ⊗ M ) = 0, then every exact sequence E splits.
Remark 2.20. For each α ∈ Ext1 (E ′′ , E ′ ) there exists an extension

0 → E ′ → Eα → E ′′ → 0 (3)
with a vector bundle, Eα , in such a way that α is the extension class
of (3). Moreoever, Eα ∼= Eβ if and only if there exists λ ∈ C∗ such
that α = λβ. (See [10] Section 3.4 for more details)
84 CIARA DALY
2.3. Riemann-Roch Formula for curves. We have a very useful

tool, called the Riemann-Roch formula, which tells us a lot about the
cohomology groups of a vector bundle E, H 0 (E) and H 1 (E), once
we know the rank and degree of E. The Riemann-Roch formula is
as follows: If E is a vector bundle of rank r on a curve of genus g,
then:
h0 (E) − h1 (E) = deg E − r(g − 1).
In addition to the Riemann-Roch formula, one of the other major
tools we have in dealing with cohomology is Serre duality. The fol-
lowing proposition outlines Serre duality, though will not be proved
as the proof is too involved for this paper.
Proposition 2.21. (Serre duality) Let X be a smooth projective
curve. Let E be a vector bundle on X. Let KX be a canonical line
bundle on X. Then there are canonical isomorphisms
H 0 (X, E) ∼
= H 1 (X, KX ⊗ E ∗ )∗ .
and
H 1 (X, E) ∼
= H 0 (X, KX ⊗ E ∗ )∗ .
In particular it follows that H 0 (X, E) and H 1 (X, KX ⊗ E ∗ ) have the
same dimension.
While I have not defined KX , the canonical line bundle, for the
purpose of this paper it will suffice to know what KX is in the case
of a curve. This is outlined below:
X = P1 : KX = OP1 (−2), deg(KX ) = −2
X = elliptic curve : KX = OX , deg(KX ) = 0
X = curve of genus g ≥ 2 : deg(KX ) = 2g − 2.
Refer to [5] Section III.7 for more details on Serre duality.
Lemma 2.22. Let L be a line bundle on a curve, C, of genus g.
Then we have the following:
(a) H 0 (L) = 0 if deg L < 0.
(b) H 1 (L) = 0 if deg L > 2g − 2.
(c) L ∼
= O if deg L = 0 and s ∈ H 0 (L), s 6= 0.
The proof of (a) and (c) of the lemma above uses the correspon-
dence between line bundles and divisors (see again [7]) and the fact
that the divisor defined by a nonzero holomorphic section of a line
bundle is always positive. The proof of (b) follows from Serre duality
and part (a).
Lemma 2.23. If E is a vector bundle on a curve C of genus g, then

the degree of its subbundles F ⊂ E is bounded above.
Proof. ([9], Corollary 10.9) Since the global sections functor is left
exact, we get
H 0 (F ) ⊂ H 0 (E)
This implies that h0 (F ) ≤ h0 (E). Now by Riemann-Roch we know
h0 (F ) − h1 (F ) = deg(F ) + rk(F ) · (1 − g).
From this we get
deg(F ) + rk(F ) · (1 − g) + h1 (F ) ≤ h0 (E)
and by rearranging we have
deg(F ) ≤ h0 (E) − rk(F ) · (1 − g) − h1 (F )
Now if g = 1, we see that deg(F ) ≤ h0 (E) − h1 (F ) and since
h1 (F ) ≥ 0, we get deg(F ) ≤ h0 (E).
If g = 0, then deg(F ) ≤ h0 (E)−rk(F )−h1 (F ) and since rk(F ) ≥ 0
and h1 (F ) ≥ 0, we see that deg(F ) ≤ h0 (E).
If g ≥ 2, deg(F ) ≤ h0 (E)+rk(F )·(g −1)−h1 (F ) ≤ h0 (E)+rk(E)·
(g − 1) − h1 (F ) (since rk(F ) ≤ rk(E)). Again, since h1 (F ) ≥ 0, we
get deg(F ) ≤ h0 (E) + rk(E) · (g − 1). Hence we see that in any case
the degree of F ⊂ E is bounded above.
3. Vector bundles on P1
Before we move on to vector bundles on an elliptic curve (i.e. a
curve of genus one), it makes sense to look at vector bundles on a
curve of genus zero (P1 ). Let us now restate Lemma 2.22 in the case
of P1 , where genus g = 0, to see how the cohomology of line bundles
on P1 is particularly simple.
Lemma 3.1. Let L be a line bundle on P1 . Then we have the
following:
(a) H 0 (L) = 0 if deg L ≤ −1.
(b) H 1 (L) = 0 if deg L ≥ −1.
By Riemann-Roch we also have,
h0 (L) − h1 (L) = deg L + 1.
Remark 3.2. We have seen from [7] that for L a line bundle, L∗ is
the inverse of the line bundle L in the Picard group. We have also
seen that deg : Pic X → Z is a homomorphism (where Pic X denotes
the set of line bundles over X) and so we get deg L∗ = − deg L.
86 CIARA DALY
Lemma 3.3. The homomorphism deg : Pic P1 → Z is an isomor-

phism.
Proof. See [4] Lemma 6.2.11.
We have a classification for all vector bundles on P1 as follows:
Lemma 3.4. Every rank 2 vector bundle on P1 is isomorphic to a
direct sum of two line bundles
Proof. ([9], Lemma 10.30) Let E be a rank 2 vector bundle on P1 .
Tensoring with a line bundle if necessary, it is enough to assume
that deg E = 0 or −1. First, by the Riemann-Roch formula we
note that H 0 (E) 6= 0, and so from Remark 2.12 above we get a line
bundle M ⊂ E, and M ∼ = O(D) for some positive divisor D ≥ 0. In
particular, deg M ≥ 0, and denoting the quotient by L := E/M , we
have an exact sequence
0 → M → E → L → 0.
Now deg E = deg L + deg M , hence deg(L∗ ⊗ M ) = deg M − deg L =
− deg E + 2 deg M ≥ − deg E ≥ 0. From Lemma 3.1 (b), we get
H 1 (L∗ ⊗ M ) = 0. By Proposition 2.19, therefore, the sequence
splits.
Grothendieck’s Theorem 3.5. Every vector bundle on P1 is iso-
morphic to a direct sum of line bundles
Proof. ([9] Theorem 10.31) Let E be a vector bundle of rank r on
P1 . Proof is by induction on the rank r ≥ 2 of E, starting with
the previous lemma. Serre’s Theorem ([5] II.5.17) tells us that there
exists a line subbundle in E. Now let M ⊂ E be the line subbundle
whose degree, m = deg M , is maximal among line subbundles of E
(Lemma 2.23). Let F := E/M be a vector bundle of rank r − 1.
Claim: Every line subbundle L ⊂ F has deg L ≤ m.
Now we have a short exact sequence as follows:
0→M →E→F →0
By considering the preimage L̃ ⊂ E of L under the quotient mor-
phism E → F we get a diagram as follows:
0 /M /E /F /0
O O
? ?
0 /M / L̃ /L /0
Clearly L̃ is a rank 2 vector bundle, and we see deg L̃ = m + deg L.

By Lemma 3.4, we know L̃ ∼ = L1 ⊕ L2 for some line bundles L1
and L2 . Now deg(L̃) = deg(L1 ) + deg(L2 ) so one of L1 or L2 must
have deg at least deg(L̃)/2. Let N denote that line subbundle, of
degree at least deg L̃/2. Because N is a subbundle of E, as well as
our choice of M we get m ≥ deg N ≥ (deg L̃)/2 = m+deg 2
L
and the
claim follows easily from this.
By the inductive hypothesis, we know the quotient bundle F is
isomorphic to a direct sum F = L1 ⊕ · · · ⊕ Lr−1 of line bundles and
the claim gives us deg Li ≤ m. Since H 1 (L∗i ⊗ M ) = 0 for each i. It
follows that the exact sequence
r−1
M
0→M →E→ L1 → 0
i=1
splits.
Definition 3.6. A vector bundle,

L E, is called decomposable if it is
isomorphic to the direct sum E1 E2 of two nonzero vector bundles;
otherwise, E is called indecomposable.
By definition of decomposability, every vector bundle is the direct

sum of indecomposable ones. Therefore, it suffices to know the inde-
composable vector bundles on a curve in order to know them all. We
have seen that all vector bundles on rational curves are the direct
sum of line bundles. As well as the notion of an indecomposable
vector bundle, we also have the notion of a simple vector bundle.
Definition 3.7. A vector bundle E is simple if its only endomor-

phisms are scalars, End E = C. Every line bundle is simple.
A simple vector bundle is necessarily indecomposable. To see

this let us start with a decomposable vector bundle E ⊕ F . Consider
f : E⊕F → E⊕F , where f = idE ⊕0F where idE is the identity map
on E and 0F is the zero map on F . Clearly then End(E ⊕ F ) 6= C,
i.e. F is not simple.
Note that the converse is not true, i.e. an indecomposable vector
bundle is not necessarily simple (This can be seen by a counterex-
ample, Example 4.4 below).
88 CIARA DALY
4. Classification of all indecomposable rank two vector

bundles on an elliptic curve C
We are now ready to look at the case of a nonsingular curve
of arithmetic genus one (i.e. an elliptic curve). Atiyah’s paper of
1957 ([1]) provided us with an answer to this case. We have already
seen in Lemma 3.3 that there is exactly one line bundle on P1 for
every degree. In particular Pic0 (P1 ) = {O}, where Pic0 (P1 ) denotes
the set of line bundles of degree 0 on P1 . However it turns out ([7]
Theorem 20) on an elliptic curve, C, that Pic0 (C) is in bijection to
C and so on elliptic curves there are more vector bundles in the sense
that nontrivial extensions appear. For the purpose of this paper we
will be concentrating on rank 2 vector bundles on an elliptic curve.
In this section we will give a classification of all indecomposable rank
2 vector bundles on the elliptic curve C.
First let me return to the Riemann-Roch formula for a vecor bun-
dle E, this time looking at a curve of genus 1, i.e.
h0 (E) − h1 (E) = deg E
Note that every line bundle, L, on C satisfies:
h0 (L) − h1 (L) = deg L
The next lemma follows from the above equation and Lemma 2.22:
Lemma 4.1. Let L be a line bundle on an elliptic curve. Then we
have the following:
(a) H 0 (L) = 0 if deg L < 0.
(b) H 1 (L) = 0 if deg L > 0.
(c) If deg L = 0 and L 6∼= O, then H 0 (L) = H 1 (L) = 0.
Lemma 4.2. If E is an indecomposable vector bundle of rank 2
on a smooth projective curve, X, then every line subbundle L ⊂ E
satisfies
2 deg L ≤ deg E + 2g − 2
Proof. Let M be the quotient line bundle E/L. This gives us the
following short exact sequence:
0→L→E→M →0
which corresponds to an element in Ext1 (M, L) ∼ = H 1 (M ∗ ⊗ L).
Now since E is indecomposable, this sequence cannot split and hence
H 1 (M ∗ ⊗ L) 6= 0 by Proposition 2.19. By Serre duality, this implies
that 0 6= H 0 ((M ∗ ⊗ L)∗ ⊗ KX )∗ = H 0 (M ⊗ L∗ ⊗ KX )∗ . This in turn
implies that deg(M ⊗ L∗ ⊗ KX ) = deg M − deg L + 2g − 2 ≥ 0 from

Lemma 2.22 (a). Now from the short exact sequence above we know
that deg E = deg M + deg L, i.e. deg M = deg E − deg L. Hence we
get
deg E − deg L − deg L + 2g − 2 ≥ 0
From this, we get the inequality in the lemma.
Let E(r, d) denote the set of isomorphism classes of indecompos-
able vector bundles of rank r and degree d over X, an elliptic curve.
Theorem 4.3. (a) There exists a vector bundle Er ∈ E(r, 0), unique
up to isomorphism, with H 0 (Er ) 6= 0. Moreover, we have an exact
sequence:
0 → OX → Er → Er−1 → 0
(b) Let E ∈ E(r, 0), then E ∼= Er ⊗ L, where L is a line bundle of
degree zero, unique up to isomorphism.
Proof. See [1] Theorem 5.
Example 4.4. The bundles Er of Theorem 4.3 are sometimes called
the Atiyah bundles. For r ≥ 2, they are examples of indecomposable
vector bundles which are not simple. Let us prove now that E2 is
not simple.
We know E2 sits in an exact sequence as follows:
f
0 / OX / E2 / OX /0
Applying Hom(−, E2 ) to this sequence, we get

β
0 / Hom(O, E2 ) / Hom(E2 , E2 ) / Hom(O, E2 )
Now Hom(O, E2 ) ∼ = H 0 (E2 ) from Proposition 2.11. From our as-

sumption on E2 , we know H 0 (E2 ) 6= 0, i.e. h0 (E2 ) ≥ 1. Now let
idE2 denote id ∈ Hom(E2 , E2 ). We know, under the morphism β,
that idE2 7→ f ∈ Hom(O, E2 ) 6= 0, i.e. β(idE2 ) = f 6= 0. This
implies β 6= 0. So we get the following short exact sequence:
0 → Hom(O, E2 ) → Hom(E2 , E2 ) → im(β) → 0
Since β 6= 0, we know that dim(im(β)) ≥ 1. Now since dim is addi-
tive on exact sequences, dim(Hom(E2 , E2 )) = dim(Hom(O, E2 )) +
dim(im(β)) ≥ 2. Hence by the definition of a simple vector bundle
(Definition 3.7), we know that E2 is not simple.
90 CIARA DALY
Let us now classify all indecomposable rank 2 vector bundles on

an elliptic curve. We first consider the case of even degree.
Proposition 4.5. On a curve, C, of genus 1 every indecomposable
rank 2 vector bundle, E, of even degree is an extension of the form
0 −→ M −→ E −→ M −→ 0
for some line bundle M on E
Proof. ([9] Proposition 10.48) Using the tensor product trick, it is
enough to consider the case where deg E = 2k. If M1 ∈ Pick (C),
i.e. M1 is a line bundle of degree k, then E ⊗ M1 is of degree 0.
In other words, E ⊗ M1 ∈ E(r, 0). By Theorem 4.3, we know that
there exists M2 ∈ Pic0 (C) such that E ⊗ M1 ∼
= E2 ⊗ M2 , where E2
is the so-called Atiyah bundle from Theorem 4.3. Then E2 sits in a
nonsplit exact sequence as follows
0 → OC → E2 → OC → 0.
If M := M2 ⊗M1∗ , we obtain E ∼= E2 ⊗M and tensoring this sequence
by M , gives a short exact sequence
0 → M → E → M → 0.

The following proposition contains the odd degree case.
Proposition 4.6. On a curve, C, of genus 1, given a line bundle
L of odd degree, there exists, up to isomorphism, a unique indecom-
posable rank 2 vector bundle E with det E ∼= L.
We refer to [9] Proposition 10.47 for the proof.
Theorem 4.7. For each integer n, there is a one-to-one correspon-
dence between the set of isomorphism classes of indecomposable vec-
tor bundles of rank 2 and degree n on the elliptic curve C, and the
set of points on C.
Sketch of correspondence We will denote by Picn (C), the set of
degree n line bundles on C. Recall from [7], Theorem 20, that there is
an isomorphism of manifolds C ∼ = Pic0 (C) and in fact ([7], Theorem
21) there is an isomorphism C ∼= Picn (C) for all n ∈ Z.
Now let E be an indecomposable rank 2 vector bundle of degree
n on C. If n is odd, from Proposition 4.6, we know that there is
a unique indecomposable rank 2 vector bundle E of degree n, with
det E ∼
= L. Hence use Picn (C) ∼= C to obtain the result.
If n is even from Theorem 4.3 we know there exists, L, a line

bundle of degree zero, unique up to isomorphism such that E ⊗ L is
isomorphic to the unique nontrivial extenstion of OC by OC . Since
Pic0 (C) ∼
= C again we obtain the result.
5. Stability
The notion of stability comes from the theory of moduli spaces.
The variety, Pic0 (C) with the Poincaré bundle, of degree 0 line bun-
dles on an elliptic curve (See [7] Section 6) is an example of a mod-
uli space. Loosely described a moduli space is an algebraic variety
which parametrises the set of equivalence classes of some objects.
For example we could consider the moduli space of rank two vector
bundles on an elliptic curve, C. It turns out that the set of isomor-
phism classes of vector bundles of rank 2 and degree d on an elliptic
curve is unbounded (briefly, this means that we can find families of
arbitrarily high dimension which gives us vector bundles of rank 2
and degree d. See [6] Chapter 1), which poses a problem when con-
structing the corresponding moduli space. To overcome this problem
we restrict our study of vector bundles. One form of restriction is to
study ‘stable’ vector bundles. Using stable bundles, one to construct
the moduli space of (stable) vector bundles of rank 2 and degree d
on C.
We will begin by giving a more explicit definition of a subbundle
and quotient vector bundle.
Definition 5.1. Let F and E be vector bundles of rank r and n
respectively, with r ≤ n and F ⊂ E is a submanifold. Then, F
is called a subbundle of E if there exists an open covering {Ui } and
transition functions gij : Ui ∩Uj → GL(r, C) for F and hij : Ui ∩Uj →
GL(n, C) for E such that

gij (x) ∗
hij (x) = .
0 fij (x)
The quotient bundle G = E/F is described by transition functions
fij .
Now we are ready to define the stability of a vector bundle.
Definition 5.2. A vector bundle, E on a curve, is stable (resp.
semi-stable) if every nonzero vector subbundle F ⊂ E satisfies
deg F deg E
< (resp. ≤).
rk F rk E
92 CIARA DALY
(Or equivalently, we can also say that a vector bundle E is stable

(resp. semi-stable) if deg G deg E
rk G > rk E (resp. ≥) for every non-zero
quotient G of E).
From this definition we can see that a vector bundle E of rk 2 is
stable (resp. semi-stable) if every line subbundle F ⊂ E satifies
1
deg F < deg E (resp. ≤).
2
We call the rational number deg E
rk E the slope of E. The picture below
illustrates the reason for this name.
deg
6
r
E

- rank
The definition of stability above is often referred to as slope-stability.
Lemma 5.3. Let E be a vector bundle of rank 2. If deg E is odd,
then stability and semi-stabilty are equivalent.
Proof. Clearly if E is stable, then E is semi-stable. For the other
direction, we assume E is semi-stable with degree n. Now let F ⊂ E
be a nonzero subbundle of E, so deg F ≤ n2 . Since deg F is an integer,
deg F 6= n2 as n is odd. Hence deg F < n2 , i.e. E is stable.
deg E1 deg E2
Lemma 5.4. If E1 and E2 are semi-stable , and rk E1 > rk E2 ,
then Hom(E1 , E2 ) = 0
Proof. Let f : E1 → E2 be a morphism, and let F ⊂ E2 be it’s image.
Since E2 is semi-stable, if F 6= 0, then deg F deg E2
rk F ≤ rk E2 . But E1 is
semi-stable and F is a quotient of E1 , and therefore deg E1 deg F
rk E1 ≤ rk F ,
a contradiction unless F = 0.
5.1. Jordan-Hölder filtrations. Consider a rational number µ and
let C(µ) denote the category of semi-stable vector bundles of slope
µ. This turns out to be an abelian category (See [8] Chapter 5 for
more details). This allows us to define Jordan-Hölder filtrations for
each semi-stable bundle: these filtrations are important in order to
understand the points of the moduli space of stable vector bundles
of rank 2 and degree d (or indeed any fixed rank and degree).
Definition 5.5. Let E be a semi-stable vector bundle of slope µ. A

Jordan-Hölder filtration of E is a filtration of vector subbundles
0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ Ek = E
in C(µ) such that the quotient gri = Ei /Ei−1 is a stable bundle in
C(µ). TheL integer k is called the length of the filtration and the
direct sum i gri is called the associated grading.
Let us now see how to get such a filtration. Consider first if E
is stable, i.e. ∀Ei ⊂ E, µ(Ei ) < µ(E). In this case, the filtration is
clear. Namely
0 ⊂ E1 = E
Now we consider when E is strictly semi-stable, then there exists
subbundles in C(µ) and one of these, E1 must be stable. If not then
we can construct an infinite descending sequence of subbundles in
C(µ) in which the rank strictly decreases but this is impossible as
the rank of nonzero subbundles of E is bounded below by 1. So
E/E1 is also in C(µ) and we can continue the construction until we
obtain our filtration as above.
5.2. Harder-Narasimhan filtrations.
Lemma 5.6. (a) Let d, d′ , r, r′ ∈ Z with r, r′ > 0.
′ ′
d′
(i) If dr > dr′ , then dr > d+d
r+r ′ > r′ .
′ ′ ′ ′
(ii) If dr = d+d d d+d d d
r+r ′ or r ′ = r+r ′ then r = r ′ .
(b) Let 0 → E ′ → E → E ′′ → 0 be a short exact sequence of nonzero
vector bundles on X.
(i) If λ ∈ R such that µ(E ′ ) ≤ λ and µ(E ′′ ) ≤ λ, then µ(E) ≤ λ.
(ii) If µ(E ′ ) = µ(E) or µ(E) = µ(E ′′ ) then µ(E ′ ) = µ(E ′′ ).
(c) If
0 = E0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ En = E
is a filtration by subbundles of E such that µ(Ei /Ei−1 ) ≤ λ for all
i = 1, . . . , n:
(i) then µ(Ei ) ≤ λ for all i = 1, . . . , n. In particular, µ(E) ≤ λ.
(ii) If, for at least one i, we have µ(Ei /Ei−1 ) < λ, then µ(E) < λ.
Proof. (a) The proof of this is a simple calculation.
(b) Because of the fact that rk(E) = rk(E ′ )+rk(E ′′ ) and deg(E) =
deg(E ′ ) + deg(E ′′ ), this follows immediately from (a).
(c) This follows from (b) using exact sequences
0 → Ei−1 → Ei → Ei /Ei−1 → 0
94 CIARA DALY
for all i = 2, . . . n.
Each vector bundle admits a canonical increasing filtration whose

successive quotients are semi-stable. This allows us to classify bun-
dles which are not semi-stable in terms of semi-stable bundles.
Proposition 5.7. Let E be a vector bundle on a curve X. Then E

has an increasing filtration by vector subbundles
0 = E0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ Ek = E
where the quotient gri = Ei /Ei−1 satisfies the following conditions:

(1) the quotient gri is semi-stable;
(2) µ(gri ) > µ(gri+1 ) for i = 1, · · · , k − 1.
Proof. ([8] Proposition 5.4.2) If E is already semi-stable then the
result is trivial. Assume, therefore that E is not semi-stable. We
will prove this by induction on the rank of E. If rk(E) = 1, then
the result is trivial as all line bundles are automatically stable. Now
assume rk(E) ≥ 2. We know, from Lemma 2.23, that the degree
of all subbundles of E is bounded above. On the other hand, sub-
bundles can only have ranks 1, 2, . . . , rk(E) − 1, hence the slope of
the subbundles of E is bounded above. Among all the subbundles
of maximal slope, let E1 be the one of maximal rank. Then E1 is
semi-stable because it has maximal slope. Let E ′ = E/E1 , then we
have the following short exact sequence:
0 → E1 → E → E ′ → 0
where rk(E ′ ) < rk(E).

By inductive assumption E ′ has an increasing filtration satisfying
the conditions of the proposition, i.e.
0 ⊂ F2 ⊂ F3 ⊂ · · · ⊂ Fk = E ′
with
µ(F2 ) > µ(F3 /F2 ) > · · · > µ(Fk /Fk−1 )
and Fj /Fj−1 is semistable for 2 ≤ j ≤ k. In particular, F2 is

semistable.
Let Ej ⊂ E be the preimage of Fj ⊂ E ′ under E → E ′ . This way

we obtain commutative diagrams with exact rows:
0 / E1 / Ej+1 / Fj+1 /0
O O
? ?
0 / E1 / Ej / Fj / 0.
Hence Ej+1 /Ej ∼= Fj+1 /Fj are semistable.

Now we need to prove µ(F2 ) < µ(E1 ), in order to show condi-
tion 2 holds. Since E1 has maximal slope, µ(E2 ) ≤ µ(E1 ). More-
over, since E1 has maximal rank among the subbundles with slope
µ(E1 ), µ(E2 ) < µ(E1 ). From the diagram above we know that
deg(F2 ) = deg(E2 ) − deg(E1 ) and rk(F2 ) = rk(E2 ) − rk(E1 ). So
we know µ(F2 ) = deg(E 2 )−deg(E1 )
rk(E2 )−rk(E1 ) . We can also write this as µ(F2 ) =
rk(E2 )µ(E2 )−rk(E1 )µ(E1 )
rk(E2 )−rk(E1 ) . Then we have
rk(E2 )µ(E2 ) − rk(E1 )µ(E1 ) rk(E2 )µ(E1 ) − rk(E1 )µ(E1 )
<
rk(E2 ) − rk(E1 ) rk(E2 ) − rk(E1 )
i.e. µ(F2 ) < µ(E1 ). Now since E2 /E1 = F2 , we have µ(E1 ) >
µ(E2 /E1 ). We can the repeat the process until we obtain a quotient
E/Ek−1 which is semi-stable.
Lemma 5.8. If
0 = E0 ⊂ E1 ⊂ · · · ⊂ En = E
is a filtration of E satsifying the conditions of Proposition 5.7 above
and E ′ ⊂ E is a nontrivial subbundle of E then µ(E ′ ) ≤ µ(E1 ) and
if µ(E ′ ) = µ(E1 ), then E ′ ⊂ E1 .
Proof. We define a filtration of E ′ by Ei′ := E ′ ∩ Ei for all i =
1, . . . , n. Because Ei′ = Ei ∩ Ei+1
′ ′
we obtain Ei+1 /Ei′ ⊂ Ei+1 /Ei for
i = 1, . . . , n − 1. Now since Ei+1 /Ei is semistable, we have either
′
µ(Ei+1 /Ei′ ) ≤ µ(Ei+1 /Ei ) or Ei+1
′
= Ei′ . Because µ(Ei+1 /Ei ) ≤
µ(E1 ) for i = 1, 2, . . . , n − 1, we obtain from Lemma 5.6 (c) that
µ(E ′ ) ≤ µ(E1 ). Now if i ≥ 1 and Ei+1 ′
6= Ei′ then µ(Ei+1 ′
/Ei′ ) ≤
µ(Ei+1 /Ei ) < µ(E1 ). Hence by Lemma 5.6 (c) again, if µ(E ′ ) =
′
µ(E1 ) we must have Ei+1 = Ei′ for i = 1, 2, . . . , n − 1, i.e. E ′ ⊂
E1 .
Proposition 5.9. This filtration of Proposition 5.7 is unique.
96 CIARA DALY
Proof. ([8] Proposition 5.4.2) Assume (Ei )i=1,...,n and (Fj )j=1,...,m
are two filtrations of E satisfying the conditions of Proposition 5.7
above. Now using the notation of Lemma 5.8 if we let E ′ := F1
we get µ(F1 ) ≤ µ(E1 ). Similarly if we allow E ′ := E1 , we get
µ(E1 ) ≤ µ(F1 ). Clearly then, µ(F1 ) = µ(E1 ).
Lemma 5.8 again implies E1 ⊂ F1 and F1 ⊂ E1 , hence E1 = F1 .
Using E/E1 and F/F1 we can proceed by induction as in the proof
of Proposition 5.7 to conclude that the filtration is unique.
The filtration of Proposition 5.7 is called the Harder-Narasimhan
filtration of E.
6. Conclusion
In conclusion, it is fair to say that the theory of vector bundles is
vast and indeed very interesting. We have seen how vector bundles
on P1 are not very complex, in the sense that they can be written as
a direct sum of line bundles. We have also seen a classification for
indecomposable rank 2 vector bundles on elliptic curves.
One could also go on to study higher rank vector bundles on
elliptic curves or on curves of a higher genus, or even on higher
dimensional complex manifolds. These are all very interesting in
their own right.
In Section 5, we studied slope stability for vector bundles on
curves. As was mentioned, stable bundles are required when con-
structing moduli spaces of vector bundles. For the reader interested
in stability, from here you could go on to study Bridgeland stability
conditions ([2] and [3]) and the space of all stability conditions on a
particular complex manifold (e.g. an elliptic curve).
References
[1] M. F. Atiyah: Vector bundles over an elliptic curve, Proc. London Math.
Soc. (3) 7 (1957) 414-452.
[2] T. Bridgeland: Stability Conditions on Triangulated Categories, preprint
arXiv:math.AG/0212237.
[3] T. Bridgeland: Spaces of Stability Conditions, arXiv:math.AG/0611510.
[4] A. Gathmann: Algebraic Geometry, Notes for a class taught at the Univer-
sity of Kaiserslauten (2002/2003) available at http://www.mathematik.uni-
kl.de/∼gathmann/class/alggeom-2002/main.pdf.
[5] R. Hartshorne: Algebraic Geometry, Graduate Texts in Mathematics,
Springer, (1977).
[6] D. Huybrechts, M. Lehn: The geometry of moduli spaces of sheaves Aspects
of Mathematics, E31. Friedr. Vieweg & Sohn, Braunschweig, (1997).
[7] M. Khalid: Group law on the cubic curve, this issue.
[8] J. Le Potier: Lectures on Vector Bundles Cambridge studies in advanced
mathematics, Cambridge University Press (1997).
[9] S. Mukai: An Introduction to Invariants and Moduli, Cambridge studies in
advanced mathematics, Cambridge University Press, (2003).
[10] C. A. Weibel: An introduction to homological algebra, Cambridge studies
in advanced mathematics, Cambridge University Press, (1994).
Department of Mathematics, Mary Immaculate College, South Cir-

cular Road, Limerick, Co. Cork, Ireland
E-mail address: Ciara.Daly@mic.ul.ie

Ims All

Uploaded by

Copyright:

Available Formats

Ims All

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ims All

Uploaded by

Copyright:

Available Formats

Elliptic Curves

Expanded notes from a mini-workshop held at

Mary Immaculate College

29 and 30 November 2006

Ciara Daly, Marina Franz, Madeeha Khalid and

The following four articles constitute expanded versions of talks

Solving Cubic Equations in Two Variables (Bernd Kreussler)

Group Law on the Cubic Curve (Madeeha Khalid)

Theta Functions (Marina Franz)

Rank two vector bundles on elliptic curves (Ciara Daly)

[8] J. Le Potier, Lectures on vector bundles. Cambridge Studies in Advanced

Mary Immaculate College, South Circular Road, Limerick, Ireland

Abstract. After recalling a geometric construction of all

Such a line is given by the equation y = rx − 1. Therefore, the x-

to the factor x of this polynomial. The two new intersection points

Example 1. Let us see how this works with P = (0, 1) and Q =

Example 2. We may now continue by using the line through P =

In general, if we are using a line with slope r ∈ Q which passes

Example 3. The line connecting S = (2, −5) with R = (1, 1) has

Note that we obtained (6, 29) as follows. First we connected P =

Example 4. The line which connects Q = (−1,

This coincidence is not an accident. It is in fact a special case

Definition 5. Let P, Q be points on the cubic curve given by equa-

In this language, we have shown above (P + Q) + R = P + (Q + R)

Example 6. Implicit differentiation reveals that the tangent line to

P = (0, 1), 2P = (1, 1), 3P = (−1, −1),

3. The complete picture

between these three K2 are given by

of an equation of the form

where we used x = ac , y = cb instead of ζ0 = zz20 , ζ1 = zz12 as coor-

Theorem 8. E(Q) is an Abelian group with neutral element O, its

y 2 = x3 + px + q with 4p3 + 27q 2 6= 0. (4)

Such an equation is also called a Weierstraß equation or Weierstraß

Theorem 14 (Siegel, [20, 16, 18]). The equation y 2 = x3 + px + q

A point P ∈ E(Q) is called a torsion point if there exists a positive

Theorem 15 (Lutz–Nagell, [13, 19]). All torsion points of y 2 =

4p3 + 27q 2 6= 0. Moreover, if (x, y) ∈ E(Q)tor then either y = 0 or

with positive integers ai . The number r ≥ 0 is called the rank of this

point. This implies 6P = O and we obtain 4P = −2P = (0, −1) and

Swinnerton-Dyer conjecture which is one of the millennium prize

In order to simplify the left hand side to y 2 we have to complete

The map so defined can be extended to a map which is also defined

on p, q, which is due to the non-presence of x2 in the Weierstraß

The projection with centre (1 : 0 : 0 : 0) is the map which sends

[10] D.H. Husemoller, Elliptic curves. Graduate Texts in Mathematics 111,

Mary Immaculate College, South Circular Road, Limerick, Ireland

Abstract. It is known that the set of rational points on a

This work was supported by the IRCSET Embark Initiative Postdoctoral

which we have seen already in [9] Section 3. By P1 and P2 we mean

The complex torus is a nice example of a one dimensional manifold

Each compact Riemann surface can be embedded holomorphically

of degree d. Then, on open subsets of P2 , the curve C is the zero

2. Cubic curves and the group law

P and Q. Then, by the same prescription as in [9] Definition 5,

This is exactly the same as in [9] Theorem 8, except now we allow

C/Λ is diffeomorphic to S 1 × S 1 it is like a hollow doughnut and so

When viewed as a meromorphic function on C, ℘(z) is doubly

A complete proof of Equation (1) and the derivations of g2 , g3 is

for some holomorphic functions f and g such that f (0) 6= 0, and