Notes: Introduction to Elliptic Curves

and Modular Forms

Roberto Villaflor

Along this course our main references will be [LR11, Mov19, Sil09, MDG16, DS05, BvdGHZ08].

1 Introduction
In this introductory lecture we will see how elliptic curves and modular forms appear natu-
rally when considering some elementary number theory questions. We will leave the formal
definitions for the upcoming lectures.

1.1 Elliptic curves appearing in nature

Consider the following elementary problem:
Question 1. Are there three consecutive integers whose product is a perfect square?
A first answer would be
0 ⋅ 1 ⋅ 2 = 02 .
Are there non-trivial examples? We want solutions to the equation

y 2 = x(x + 1)(x + 2), x, y ∈ Z, y ≠ 0.

Answer. No, in fact if x is odd, then

gcd(x, x + 1) = gcd(x + 1, x + 2) = gcd(x, x + 2) = 1

and so
x = ±a2 , x + 1 = ±b2 , x + 2 = ±c2 ,
which implies x = −1 and so y = 0. In case x = 2z is even, then our equation turns

y 2 = 4z(z + 1)(2z + 1).

Again since
gcd(z, z + 1) = gcd(z + 1, 2z + 1) = gcd(z, 2z + 1) = 1
we conclude that
z = ±a2 , z + 1 = ±b2 , 2z + 1 = ±c2 ,
and so z = −1, 0, thus y = 0.
Question 2. Are there non-trivial rational solutions? i.e. x, y ∈ Q with y ≠ 0?
Answer. No, but much harder!

1
Question 3. Are there three integers that differ by 5, i.e. x, x + 5 and x + 10, whose product
is a perfect square?

In this case we are looking for solutions of the equation

y 2 = x(x + 5)(x + 10), x, y ∈ Z.

As in the previous equation there are trivial solutions with y = 0, but in this case there are
also non-trivial solutions, e.g.

(−9) ⋅ (−4) ⋅ 1 = 62 , 40 ⋅ 45 ⋅ 50 = 3002 .

Moreover there are non-trivial rational solutions also

5 25 45 75 2 −50 −5 40 100 2
⋅ ⋅ =( ) , ( )⋅( )⋅ =( ) .
4 4 4 4 9 9 9 27
In fact, there are infinitely many rational solutions! For instance
5 −50 961 7200 −12005 −16810 −27910089
x = −9, 40, , , , , , , ,...
4 9 144 961 1681 2401 5094049
Definition 1.1. We say an integer n ≥ 1 is a congruent number if there exists a right
triangle with sides of rational length and whose area equals n.

Question 4. Which natural numbers are congruent?

For example 6 = 3⋅4 2 2 2

2 is congruent since 3 + 4 = 5 . Fibonacci proved in 1225 that 5 is
also congruent, in fact
3
2 ⋅ 20
3 3 2 20 2 41 2
5= and ( ) + ( ) = ( ) .
2 2 3 6
Proposition 1.1. The number n ≥ 1 is congruent if and only if the equation

y 2 = x3 − n2 x, x, y ∈ Q

has a non-trivial solution (i.e. with y ≠ 0).

Proof Let
ab
Cn ∶= {(a, b, c) ∈ Q3 ∶ a2 + b2 = c2 , = n},
2
En ∶= {(x, y) ∈ Q2 ∶ y 2 = x3 − n2 x, y ≠ 0}.
nb 2n 2
The map f ∶ Cn → En given by f (a, b, c) ∶= ( c−a , c−a ) has inverse the map g ∶ En → Cn given
2 −n2 2 +n2
by g(x, y) ∶= ( x y , 2nx
y ,
x
y ).

Remark 1.1. For example, y 2 = x3 − 25x has solution (−4, 6).

2
Theorem 1.2 (Stephens 1975). If Birch-Swinnerton-Dyer conjecture is true, then every
natural number n ≡ 5, 6, 7 (mod 8) is a congruent number.

In particular 157 ≡ 5 (mod 8) should be a congruent number. This fact was proved by
Zagier in 1989 who exhibited the corresponding right triangle with hypotenuse
2244035177043369699245575130906674863160948472041
c= .
8912332268928859588025535178967163570016480830
Another very famous and celebrated problem related to elliptic curves is Fermat Last
Theorem.

Theorem 1.3 (Wiles 1995). Let n ≥ 3, the Fermat equation

xn + y n = z n , x, y, z ∈ Z,

has no non-trivial solutions (i.e. with xyz ≠ 0).

It is elementary to show FLT for n = 4, and to reduce FLT to the cases n = p ≥ 3

prime. The key idea behind the proof of FLT which connects it with elliptic curves was an
observation due to Frey, who conjectured that if there exist a, b, c ∈ Z such that

ap + b p = c p , abc ≠ 0,

then Frey’s elliptic curve

y 2 = x(x − ap )(x − bp )
should not satisfy a conjecture of Taniyama and Shimura. This fact was proved by Rib-
bet, while the Taniyama-Shimura conjecture was proved by Wiles in a case that contained
Frey’s curve. Latter the Taniyama-Shimura conjecture was completed by Wiles, Taylor,
Breuil, Conrad and Diamond, which is nowadays known as the Arithmetic Modularity
Theorem.

1.2 Modular forms appearing in nature

Whenever we want to understand a certain sequence of numbers

a0 , a1 , a2 , . . .

often is very useful to study its associated generating function

f = ∑ an q n .
n≥0

One instance of this is when {an } satisfies a recurrence relation. For example the Fibonacci
sequence F0 = F1 = 1,
Fn+2 = Fn+1 + Fn ∀n ≥ 0.

3
Computing its generating function
f = ∑ Fn q n
n≥0

we see that
f = 1 + q + q 2 ∑ Fn+2 q n = 1 + q + q 2 ∑ (Fn+1 + Fn )q n
n≥0 n≥0

= 1 + q + q 2 ⋅ f + q ∑ Fn+1 q n+1 = 1 + q + q 2 ⋅ f + q ⋅ (f − 1).

n≥0

Thus
1 1
f= 2
=
1−q−q (1 − αq)(1 − βq)
for α, β ∈ C such that αβ =√−1 and α + β = 1. i.e. α, β are the roots of x2 = x + 1, the so
called golden numbers 1±2 5 . Rewriting
1 α β α β
f= ( − )= ∑ (αq)n − ∑ (βq)n
α − β 1 − αq 1 − βq α − β n≥0 α − β n≥0

we conclude that
αn+1 − β n+1
Fn = ∀n ≥ 0.
α−β
In this way we see that relating the generating function with other power series we can
obtain relations among the coefficients. This idea has been exploited for centuries to study
sequences of numbers appearing naturally in mathematics. For example consider the Euler’s
partition function

p(n) ∶= # ways of writing n as a sum, i.e. n = a1 + ⋯ + ak , 1 ≤ a1 ≤ ⋯ ≤ ak .

Defining p(0) ∶= 1 we can compute its generating function

1
∑ p(n)q n = ∏(1 + q i + q 2i + q 3i + ⋯) = ∏ .
n≥0 i≥1 i≥1 1 − qi

Nowadays there is no closed formula for p(n), but we can compute a recurrence formula for
them. In order to explain this consider the q-exponential function
xn xn
Eq (x) ∶= 1 + ∑ 2 n
= ∑ 2 n
.
n≥1 (q − 1)(q − 1)⋯(q − 1) n≥0 (q − 1)(q − 1)⋯(q − 1)

We can compute it as follows

xn ( q qn−1 )
n

x x x
Eq (x) − Eq ( ) = ∑ n
= Eq ( )
q n≥1 (q − 1)⋯(q − 1) q q

then
x x x
Eq (x) = (1 + ) Eq ( ) = ⋯ = ∏ (1 + n ) for ∣q∣ > 1.
q q n≥1 q

4
 1
Replacing q by q we get
q ( 2 ) xn
n+1

∑ n)
= ∏(1 + q n x)
n≥0 (1 − q)⋯(1 − q n≥1
and so
q( 2 ) q(2)
n+1 n
n qx ⎛ ⎞
∏(1 + q x) = 1 + + x + ∑ xn n
+ n−1
n≥0 1−q n≥2 ⎝ (1 − q)⋯(1 − q ) (1 − q)⋯(1 − q ) ⎠

q ( 2 ) xn
n
n
∏ (1 + q x) = ∑ n
for ∣q∣ < 1. (1)
n≥0 n≥0 (1 − q)⋯(1 − q )

Theorem 1.4 (Jacobi triple product identity 1829). For ∣q∣ < 1 and x ≠ 0
q 2n+1 2
∏(1 − q 2n+2 )(1 + q 2n+1 x) (1 + ) = ∑ q n xn .
n≥0 x n∈Z

Proof In (1) replace q by q 2 and x by qx

2
2n+1 q n xn n2 n ∏j≥0
(1 − q 2n+2j+2 )
∏(1 + q x) = ∑ 2 2n
= ∑ q x
n≥0 n≥0 (1 − q )⋯(1 − q ) n≥0 ∏j≥0 (1 − q 2j+2 )
= ∏(1 − q 2j+2 )−1 ∑ q n xn ∏(1 − q 2n+2j+2 ).
2

j≥0 n≥0 j≥0

In (1) replace q by q 2 and x by −q 2n+2

(−1)m q m +2mn+m
2
2n+2j+2
∏(1 − q )= ∑ 2 2m )
.
j≥0 m≥0 (1 − q )⋯(1 − q

Replacing in the previous equality we get

(−1)m q m +2mn+m
2
2n+1 2j+2 −1 n2 n
∏(1 + q x) = ∏(1 − q ) ∑q x ∑ 2 2m )
n≥0 j≥0 n∈Z m≥0 (1 − q )⋯(1 − q
m
2j+2 −1
⎛ ( xq ) ⎞
( ∑ q (m+n) xm+n ) .
2
= ∏(1 − q ) ∑ 2 2m
j≥0 ⎝m≥0 (q − 1)⋯(q − 1)) ⎠ m+n∈Z
We obtain
q 2
∏(1 − q 2n+2 )(1 + q 2n+1 x) = Eq2 ( ) ⋅ ∑ q n xn .
n≥0 x n∈Z
Finally since
n
q q3 ( xq ) −q q
Eq2 ( ) − Eq2 ( ) = ∑ 2 2n
(1 − q 2n ) = Eq2 ( )
x x n≥1 (q − 1)⋯(q − 1) x x
then
q 1 q3 1
Eq2 ( ) = q
Eq 2 ( )=⋯= .
x (1 + x ) x q 2n+1
∏n≥0 (1 + x )

5
Corollary 1.5 (Euler’s pentagonal number theorem).
k(3k−1)
∏(1 − q n ) = ∑ (−1)k q 2 .
n≥1 k∈Z
3 −1
Proof Replace q by q 2 and x by −q 2 in JTPI
3n2 n
−2
∏(1 − q 3n+3 )(1 − q 3n+1 )(1 − q 3n+2 ) = ∑ (−1)n q 2 .
n≥0 n∈Z

In particular, since ∑n≥0 p(n)q n = ∏n≥1 (1 − q n )−1 we get the identity

k(3k−1)
( ∑ p(n)q n ) ( ∑ (−1)k q 2 )=1
n≥0 k∈Z

which is a recursion formula for computing p(n).

Whenever we are dealing with sequences {an } of natural numbers, sometimes it is also
useful to consider power series of the form

∑ q an .
n≥0

This is particularly useful when we want to count the number of sums we can produce with
the elements of the sequence {an }. In fact, if an ≠ 0 for all n ≥ 0 (and there is no number
repeating infinitely in the sequence), taking
f ∶= 1 + ∑ q an
n≥0

then
f k = 1 + ∑ c(n)q n
n≥1

with c(n) = # ways of writing n = ai1 + ⋯ + aik considering ordering. This idea was used by
Jacobi applied to the classical Θ-function
2
Θ(q) ∶= ∑ q n
n∈Z

to obtain the following theorem.

Theorem 1.6 (Jacobi 2-squares theorem). Let n ≥ 1 be a natural number, and
S2 (n) ∶= #{(a, b) ∈ Z2 ∶ a2 + b2 = n}
then
S2 (n) = 4(d1 (n) − d3 (n))
where
di (n) ∶= # divisors of n with d ≡ i (mod 4).

6
 Using the quadratic reciprocity law and Jacobi symbols we can rewrite this as
−1
S2 (n) = 4 ∑ ( ),
d∣n,2∤d d

Where ( −1
d−1
d
) = (−1) 2 . In general we can define

Sk (n) ∶= #{(a1 , . . . , ak ) ∈ Zk ∶ a21 + ⋯ + a2k = n}.

We have the following formulas

S4 (n) = 8 ∑ d,
d∣n,4∤d

−1 2α+4 −1
S6 (n) = (( )2 − 4) ∑ ( ) d2 , for n = 2α ⋅ m,
m d∣m d

⎛ ⎞
S8 (n) = 16 ⋅ (−1)n ∑ d3 − 2 ∑ d3 ,
⎝ d∣n d∣n,2∤d ⎠
and in general it is possible to compute Sk (n) for k even. The reason behind this fact is that

Θk = ∑ Sk (n)q n
n≥0

for k ≥ 2 is a weight k2 modular form for the group Γ1 (4). Modular forms are a very special
class of q-series, and they are strongly related to elliptic curves. One of the most non-
trivial relations among these two concepts is the Arithmetic Modularity Theorem mentioned
previously. Keeping the notions of elliptic curves and modular forms as black-boxes we will
state this theorem as follows.

Theorem 1.7 (Wiles, Taylor, Breuil, Conrad & Diamond). Let E be any elliptic curve
defined over Q. For every prime number p define

Np ∶= #E(Fp ) = # points of E modulo p.

Then there exists a modular form

f = ∑ fn q n
n≥1

such that
Np = p − fp
for all but finitely many primes.

7
2 Algebraic Varieties
2.1 Diophantine equations
A Diophantine equation is an equation given by a polynomial

f (x1 , . . . , xn ) = 0, f ∈ Z[x1 , . . . , xn ], deg f = d.

These equations arise naturally when considering number theoretic questions, e.g. y 2 =
x3 − n2 x related to the congruent number problem. Given a Diophantine equation, can we
determine if it has any integer or rational solution? can we find any of such solutions? can
we find all of them? The first question was reformulated by Hilbert as follows.

Problem 1 (Hilbert’s 10th problem). Devise an algorithm to determine for any given
Diophantine equation if it is solvable over the integers.

Theorem 2.1 (Matiyasevich, Putnam & Robinson 1970). Such an algorithm cannot exist.

However, we can still say something if we restrict ourselves to certain families of equations.

Example 2.1. For polynomials in one variable (n = 1)

f (x) = a0 xd + a1 xd−1 + ⋯ + ad−1 x + ad = 0, ai ∈ Z.

p
We can find all integer solutions. In fact since every rational root of f of the form q ∈Q
with gcd(p, q) = 1 must satisfy

a0 pd + a1 pd−1 q + ⋯ + ad−1 pq d−1 + ad q d = 0,

we conclude that p∣ad and q∣a0 .

Example 2.2. In the case f is a linear polynomial (d = 1) we can also find all the solutions.
Let us explain it for the case n = 2. Consider the linear equation

ax + by = d, a, b, d ∈ Z, ab ≠ 0.

It is trivial to solve the equation over Q, and over Z we see using Bézout’s identity that we
can solve it if and only if gcd(a, b)∣d.

Example 2.3. In the case of conics (n = 2, d = 2) we have

C∶ ax2 + bxy + cy 2 + dx + ey = f, a, b, c, d, e ∈ Z.

Legendre’s criterion determines whether there are rational solutions for this conic. In fact,
it has solutions over Z if and only if it has solutions modulo pn for all n ≥ 1 and all primes
p. This is an instance of Hasse principle also called the Local-to-Global principle.
In practice this method reduces us to verify only a finite amount of primes and powers

8
depending only on the coefficients of the conic C. After we have found a rational point
(if any) (x0 , y0 ) ∈ C(Q) we can use the stereographic projection to find all its rational
points.
The problem of finding the solutions over Z translates into solving Pell’s equation

x2 − Dy 2 = 1, x, y ∈ Z.

This equation can be solved using continued fractions, see [MMST13].

Example 2.4. The case of cubic equations (n = 2, d = 3) is much harder

a3 + bx2 y + cxy 2 + dy 3 + ex2 + f xy + gy 2 + hx + iy + j = 0, a, b, c, d, e, f, g, h, i, j ∈ Z.

These kind of equation can have no rational solution, only 1 rational solution, a finite number
of rational solutions, or an infinite number of rational solutions. All algorithms to solve such
equations are conjectural. These kind of equations with at least one rational point (and
being non-singular) are elliptic curves.
Remark 2.1. For higher degree n = 2, d ≥ 4, the corresponding curve C has typically genus
g ≥ 2, while elliptic curves have genus g = 1. The genus is a topological invariant related
to the amount of holes of the complex curve C(C) (which topologically corresponds to a
compact orientable surface). A celebrated theorem of Faltings (1983) states that for g ≥ 2,
C only has a finite amount of Q-rational points.

2.2 Rudiments of algebraic geometry

In order to state our definition of elliptic curve rigorously we will recall the basic language
and results of algebraic curves. For references see [Ful08, Sha94, Sil09, Har13]. We will
always work over a perfect field K (i.e. every algebraic extension is separable, e.g. Q, R,
C, Qp , Fq and their algebraic extensions), and we denote by K its algebraic closure.
Definition 2.1. The affine n-space over K is
n
An = An (K) ∶= K .

The K-rational points of An are

An (K) ∶= K n = {P ∈ An ∶ σ(P ) = P , ∀σ ∈ Gal(K/K)}.

Let K[x1 , . . . , xn ] be the polynomial ring in n variables, and I ⊆ K[x1 , . . . , xn ] be an ideal.

We associate an algebraic subset

V = VI ∶= {P ∈ An ∶ f (P ) = 0 , ∀f ∈ I},

and to V we associate an ideal

I(V ) ∶= {f ∈ K[x1 , . . . , xn ] ∶ f (P ) = 0 , ∀P ∈ V }.

9
 √
Remark 2.2. Hilbert’s Nullstellensatz says that I(V ) = I.
Definition 2.2. We say that an algebraic subset V ⊆ An is defined over K, denoted V /K,
if I(V ) can be generated by polynomials in K[x1 , . . . , xn ], i.e.

I(V ) = (I(V ) ∩ K[x1 , . . . , xn ]) ⋅ K[x1 , . . . , xn ].

In such a case the set of K-rational points of V is

V (K) ∶= V ∩ An (K) = {P ∈ V ∶ σ(P ) = P , ∀σ ∈ Gal(K/K)}.

Definition 2.3. We will say that an algebraic set V ⊆ An is an (affine) algebraic variety
if I(V ) ⊆ K[x1 , . . . , xn ] is a prime ideal.
Remark 2.3. If we denote

I(V /K) ∶= I(V ) ∩ K[x1 , . . . , xn ]

it is not true V /K is an algebraic variety if I(V /K) ⊆ K[x1 , . . . , xn ] is a prime ideal. For
example
V = {(x, y) ∈ A2 ∶ x2 = 2y 2 }
is defined over Q and I(V /Q) = ⟨x√2 − 2y 2 ⟩ ⊆√
Q[x, y] is prime, but I(V ) = ⟨x2 − 2y 2 ⟩ ⊆ Q[x, y]
is not prime since x2 − 2y 2 = (x − 2y)(x + 2y).
Definition 2.4. Let V /K be an affine algebraic variety, its affine coordinate ring is
K[x1 , . . . , xn ]
K[V ] ∶= .
I(V /K)
Its function field is
K(V ) ∶= Frac K[V ].
Similarly K[V ] and K(V ) are defined.
Remark 2.4. We can identify K[V ] with the ring of polynomial maps f ∶ V → K, since two
polynomial maps are equal if and only if their difference belongs to I(V /K).
Remark 2.5. Given any σ ∈ Gal(K/K) and any f ∈ K[x1 , . . . , xn ] we denote by f σ ∈
K[x1 , . . . , xn ] the polynomial obtained after applying σ to all the coefficients of f . Clearly
under this action I(V ) is invariant and so we have an induced action of Gal(K/K) on K[V ]
and K(V ) fixing K[V ] and K(V ) respectively. We also have

σ(f (P )) = f σ (σ(P )).

Definition 2.5. Let V ⊆ An be an affine algebraic variety. The dimension of V is

dim V ∶= trdeg(K(V )/K).

In particular dim An = n, and if V = {f (x1 , . . . , xn ) = 0} is a hypersurface then dim V = n−1.

10
Definition 2.6. Let V be an affine algebraic variety, P ∈ V and I(V ) = ⟨f1 , . . . , fm ⟩ ⊆
K[x1 , . . . , xn ]. We say V is non-singular at P if

∂fi
rank ( (P )) = n − dim V.
∂xj 1≤i≤m
1≤j≤n

We say V is smooth if it is non-singular at all its points.

Example 2.5. If V = {f (x1 , . . . , xn ) = 0} is a hypersurface, then it is smooth at P ∈ V if
and only if
∂f
(P ) ≠ 0 for some i = 1, . . . , n.
∂xi
For instance V1 = {y 2 = x3 + x} is smooth while V2 = {y 2 = x3 + x2 } is singular at (0, 0).
Definition 2.7. Let V be an affine variety. For every P ∈ V let

mP ∶= I(P ) ⋅ K[V ] = {f ∈ K[V ] ∶ f (P ) = 0}.

This is a maximal ideal since K[V ]/mP ≃ K.

Proposition 2.2. A point P ∈ V is non-singular if and only if

dimK mP /m2P = dim V.

Definition 2.8. The local ring of V at P is

f
K[V ]P ∶= K[V ]mP = { ∈ K(V ) ∶ g(P ) ≠ 0} .
g

This is a local ring (i.e. it only has one maximal ideal) with maximal ideal mP ⋅ K[V ]P .
Along this course we will mostly work with curves given by one affine equation inside A2 ,
but we will be interested also in their “points at infinity", i.e. we will always be considering
their projective closures. In order to introduce projective varieties let us recall the projective
space.
Definition 2.9. The projective n-space over K is
n+1
Pn = Pn (K) ∶= (K ∖ {(0, . . . , 0)})/ ∼

where
×
(x0 . . . , xn ) ∼ (y0 , . . . , yn ) ⇐⇒ ∃λ ∈ K : (x0 , . . . , xn ) = λ ⋅ (y0 , . . . , yn ).
×
In other words, it is the quotient of An+1 ∖ {0} by the action of K given by scalar multipli-
cation. For every (x0 , . . . , xn ) ∈ An+1 ∖ {0} we denote its equivalence class by
×
(x0 ∶ ⋯ ∶ xn ) ∶= K ⋅ (x0 , . . . , xn ) ∈ Pn .

11
The set of K-rational points in Pn is

Pn (K) ∶= {(x0 ∶ ⋯ ∶ xn ) ∈ Pn ∶ (x0 , . . . , xn ) ∈ An+1 (K) ∖ {0}}.

Given P = (x0 ∶ ⋯ ∶ xn ) ∈ Pn , the minimal field of definition of P is

K(P ) ∶= K(x0 /xi , . . . , xn /xi ) for some xi ≠ 0.

Remark 2.6. The Galois group Gal(K/K) acts on Pn by

σ ⋅ (x0 ∶ ⋯ ∶ xn ) ∶= (σ(x0 ) ∶ ⋯ ∶ σ(xn )).

Under this action we have

{σ∈Gal(K/K)∶σ(P )=P }
Pn (K) = {P ∈ Pn ∶ σ(P ) = P , ∀σ ∈ Gal(K/K)} and K(P ) = K .

Definition 2.10. A polynomial f ∈ K[x0 , . . . , xn ] is homogeneous of degree d if

f (λx0 , . . . , λxn ) = λd f (x0 , . . . , xn ) ∀λ ∈ K.

An ideal I ⊆ K[x0 , . . . , xn ] is homogeneous if it is generated by homogeneous polynomials.

To every homogeneous ideal I ⊆ K[x0 , . . . , xn ] we associate a projective algebraic subset

V = VI ∶= {P ∈ Pn ∶ f (P ) = 0 , ∀f ∈ I homogeneous}

and to V we associate its homogeneous ideal

I(V ) ∶= ⟨{f ∈ K[x0 , . . . , xn ] homogeneous ∶ f (P ) = 0 , ∀P ∈ V }⟩.

√
Remark 2.7. Again by Hilbert’s Nullstellensatz we have I(V ) = I.

Definition 2.11. We say a projective subset V is defined over K, denoted V /K, if I(V )
is generated by homogeneous polynomials in K[x0 , . . . , xn ], i.e. if we denote

I(V /K) ∶= I(V ) ∩ K[x0 , . . . , xn ]

then V is defined over K if and only if

I(V ) = I(V /K) ⋅ K[x0 , . . . , xn ].

In such a case the set of K-rational points of V is

V (K) ∶= V ∩ Pn (K) = {P ∈ V ∶ σ(P ) = P , ∀σ ∈ Gal(K/K)}.

Definition 2.12. We say a projective subset V is a projective variety if I(V ) is a prime

ideal of K[x0 , . . . , xn ].

12
Remark 2.8. It is clear that Pn contains many copies of An , e.g. for each 0 ≤ i ≤ n there is
an inclusion
φi ∶ An ↪ Pn
(y1 , . . . , yn ) ↦ (y1 ∶ ⋯ ∶ yi−1 ∶ 1 ∶ yi ∶ ⋯ ∶ yn ).
If we let Hi denote the hyperplane of Pn given by Hi ∶= {xi = 0} and

Ui ∶= {(x0 ∶ ⋯ ∶ xn ) ∈ Pn ∶ xi ≠ 0} = Pn ∖ Hi ,

there is a natural bijection

φ−1
i ∶ Ui → A
n

x0 xi−1 xi+1 xn
(x0 ∶ ⋯ ∶ xn ) ↦ ( , . . . , , ,..., ).
xi xi xi xi
Now, let V be a projective variety and fix some inclusion φi ∶ An ↪ Pn , then

V ∩ An = φ−1 n
i (V ) = {y ∈ A ∶ f (y1 , . . . , yi−1 , 1, yi , . . . , yn ) = 0 , ∀f ∈ I(V )}

is an affine algebraic variety given by

I(V ∩ An ) = {f (y1 , . . . , yi−1 , 1, yi , . . . , yn ) ∶ f (x0 , . . . , xn ) ∈ I(V )},

i.e. is generated by the dehomogenization of the polynomials in I(V ) with respect to xi .

This process can be reversed. For any f ∈ K[y1 , . . . , yn ] its homogenization with respect
to xi is
x0 xi−1 xi+1 xn
f ∗ (x0 , . . . , xn ) ∶= xdi f ( , . . . , , , . . . , ) , deg f = d.
xi xi xi xi
Definition 2.13. Let V ⊆ An be an affine algebraic subset, and consider V as a subset of
Pn via φi . The projective closure of V is

V ∶= {P ∈ Pn ∶ f ∗ (P ) = 0 , ∀f ∈ I(V )}.

It satisfies I(V ) = ⟨{f ∗ ∶ f ∈ I(V )}⟩.

Remark 2.9. If V is an affine variety, V = V ∩An . If V is a projective variety and V ∩An ≠ ∅,

then V = V ∩ An . This correspondence takes varieties defined over K to varieties defined
over K.

Definition 2.14. If V ⊆ An is an affine variety, the points P ∈ V ∖ V are called points

at infinity. In particular for V = An the points at infinity belong to the hyperplane at
infinity Pn−1 n n n n
∞ ∶= P ∖ A = {xi = 0} (if we see A ↪ P via φi ).

Remark 2.10. Since Pn is covered by the affine charts φi ∶ An ↪ Pn , we can these charts
to define the dimension of a projective variety V ⊆ Pn as

dim V ∶= dim V ∩ An for V ∩ An ≠ ∅.

13
We can also define the function field of V as

K(V ) ∶= K(V ∩ An ).

We say P ∈ V is non-singular if it is non-singular for some affine chart with P ∈ V ∩ An .

The local ring of V at P is

K[V ]P ∶= K[V ∩ An ]P for V ∩ An ≠ ∅.

Definition 2.15. Let V be a projective variety and P ∈ V . We say f ∈ K(V ) is regular at

P if f ∈ K[V ]P .
Remark 2.11. We can describe K(Pn ) as
f
K(Pn ) = { ∶ f, g ∈ K[x0 , . . . , xn ] homogeneous with deg f = deg g} .
g
Similarly for V ⊆ Pn a projective variety
f
K(V ) = { ∈ K(Pn ) ∶ g ∉ I(V )} / ∼
g
where
f1 f2
∼ ⇐⇒ f1 g2 − f2 g1 ∈ I(V ).
g2 g2
Definition 2.16. Let V1 ⊆ Pm and V2 ⊆ Pn be projective varieties. A rational map f ∶ V1 ⇢
V2 is a map of the form
f = (f0 ∶ ⋯ ∶ fn )
where f0 , . . . , fn ∈ K(V1 ) are such that for every P ∈ V1 where all fi are regular,

(f0 (P ) ∶ ⋯ ∶ fn (P )) ∈ V2 .

We say f is defined over K if f0 , . . . , fn ∈ K(V1 ).

Remark 2.12. A rational map f ∶ V1 ⇢ V2 is not necessarily well defined at every P ∈ V1 ,
but sometimes it can be evaluated at some P ∈ V1 where some fi is not regular. This can be
done by replacing (f0 ∶ ⋯ ∶ fn ) by some (f0 ⋅ g ∶ ⋯ ∶ fn ⋅ g) for some appropriate g ∈ K(V1 ).
Whenever we can evaluate f at P ∈ V1 we say f is regular at P . A rational map regular at
every P ∈ V1 is called a morphism and is denotes f ∶ V1 → V2 . Every f ∈ K(V1 ) corresponds
to a rational map f ∶ V1 ⇢ P1 given by (f ∶ 1), and conversely given f = (f0 ∶ f1 ) ∶ V1 ⇢ P1 it
corresponds to some ff10 ∈ K(V1 ). For this reason we call the elements of K(V ) as rational
functions.
Definition 2.17. Let V1 and V2 be two projective varieties. We say they are isomorphic if
there exist morphisms f ∶ V1 → V2 and g ∶ V2 → V1 mutually inverse. If V1 /K, V2 /K and both
morphisms are defined over K, we say V1 and V2 are isomorphic over K and in such a
case V1 (K) ≃ V2 (K).

14
3 Algebraic Curves
3.1 Ramification theory
Definition 3.1. We say a projective variety V ⊆ Pn is a curve if dim V = 1.

Proposition 3.1. Let C be a curve and P ∈ C be non-singular, then K[C]P is a discrete

valuation ring, i.e. m ∶= mP ⋅ K[C]P is principal and every ideal J ⊆ K[C]P is a power of
m.

Definition 3.2. Let C be a curve and P ∈ C be non-singular. If t ∈ K[C]P is such that

m = ⟨t⟩, we say t is a local parameter or uniformizer of C at P .

Example 3.1. The curve P1 is smooth. For every point P = (x0 ∶ y0 ) ∈ P1 we have one
coordinate different from assume. Suppose y0 ≠ 0, then an affine chart containing P is

A1 = P1 ∖ {y = 0}

and its coordinate ring is K[A1 ] = K[ xy ]. The maximal ideal is

x x0 x
mP = ⟨ − ⟩ ⊆ K [ ] .
y y0 y

And so a local parameter of P1 at P is t = xy − xy00 . Note that if x0 ≠ 0 we also can take the
local parameter u = xy − xy00 , and that ut = − xx 1
yy0 is a unity in the local ring K[P ]P .
0

Example 3.2. Let C ⊆ P2 be a curve given by f (x, y, z) = 0. If P = (x0 ∶ y0 ∶ z0 ) ∈ C is

non-singular, then some coordinate is non-zero, e.g. z0 ≠ 0. If we take the affine chart

A2 = P1 ∖ {z = 0}.

The coordinate ring of A2 can be taken as K[A2 ] = K[s, t] with s = x

z and t = yz . Thus

K[s, t]
K[A2 ∩ C] = .
⟨f (s, t, 1)⟩

The maximal ideal at P is

mP = ⟨s − s0 , t − t0 ⟩ ⊆ K[A ∩ C]

for s0 = xz00 and t0 = yz00 . Since f (s0 , t0 , 1) = 0, if we take the Taylor expansion of f (s, t, 1)
around (s0 , t0 ) we can write

∂f ∂f
f (s, t, 1) = (s − s0 ) ( (s0 , t0 , 1) + g(s, t)) + (t − t0 ) ( (s0 , t0 , 1) + h(s, t))
∂s ∂t

15
with g(s, t), h(s, t) ∈ mP . Since P is a non-singular at least one of the partial derivatives is
non-zero, e.g. ∂f
∂t (s0 , t0 , 1) ≠ 0. This implies that

∂f
u ∶= (s0 , t0 , 1) + h(s, t) ∈ K[C]×P
∂t
and so in K[C]P
∂f
(s0 , t0 , 1) + g(s, t)) u−1 .
t − t0 = −(s − s0 ) (
∂s
In consequence a local parameter for C at P is
x x0
λ = s − s0 = − .
z z0
Remark 3.1. In general, for any curve C ⊆ Pn and any non-singular point P = (a0 ∶ ⋯ ∶
an ) ∈ C, a local parameter for C at P can be taken of the form
x i ai
t= −
x j aj

for some suitable i, j ∈ {0, . . . , n}. Also for any affine curve C ′ ⊆ An and any non-singular
point P ′ = (a1 , . . . , an ), we can find a local parameter of the form

t ′ = x i − ai

for some i ∈ {1, . . . , n}.

Proposition 3.2. Let C/K be a curve and P ∈ C(K) be non-singular. Then there exists a
local parameter t ∈ K(C). For any such local parameter K(C) is a finite separable extension
of K(t).

Remark 3.2. If t is a uniformizer, every non-zero f ∈ K[C]P is of the form f = u ⋅ tk for

some u ∈ K[C]×P and k ≥ 0. The valuation at P is

ordP ∶ K[C]P → Z≥0 ∪ {∞}

given by
ordP (f ) ∶= sup{r ∈ Z ∶ f ∈ mr } = k.
Defining
ordP (f /g) ∶= ordP (f ) − ordP (g)
we can extend the valuation to

ordP ∶ K(C) → Z ∪ {∞}.

Note that an rational function f ∈ K(C) is regular at P if and only if ordP (f ) ≥ 0, and it
vanishes at P if and only if ordP (f ) > 0.

16
Definition 3.3. Let f ∈ K(C) and P ∈ C non-singular. We say P is a zero of order k if
ordP (f ) = k > 0. We say P is a pole of order k if ordP (f ) = −k < 0.
Proposition 3.3. Let C be a smooth curve and f ∈ K(C)× , then f has a finite amount of
zeros and poles, and
∑ ordP (f ) = 0.
P ∈C
×
Furthermore if f has no poles then f ∈ K .
Proposition 3.4. Let C be a curve, V ⊆ Pn a variety, P ∈ C be non-singular and φ ∶ C ⇢ V
a rational map. Then φ is regular at P . In particular if C is smooth, φ is a morphism.
Theorem 3.5. Let φ ∶ C1 → C2 be a morphism of curves, then φ is surjective or constant. If
both curves are defined over K and φ is non-constant defined over K, then the composition
with φ induces an injection of K-extensions

φ∗ ∶ K(C2 ) ↪ K(C1 )

and K(C1 )/K(C2 ) is a finite extension of fields. Conversely every finite extension K(C2 ) ↪
K(C1 ) of K-extensions comes from a unique non-constant morphism φ ∶ C1 → C2 defined
over K.
Definition 3.4. Let ϕ ∶ C1 → C2 be a non-constant morphism of curves defined over K, we
define the degree of φ as
deg φ ∶= [K(C1 ) ∶ K(C2 )].
If φ is constant we define deg φ ∶= 0. We say φ is separable, inseparable or purely
inseparable if K(C1 )/K(C2 ) is, and we denote by degs φ and degi φ the corresponding
separable and inseparable degrees.
Definition 3.5. Let φ ∶ C1 → C2 be a non-constant morphism of smooth curves and let
P ∈ C1 . The ramification index of φ at P is

eφ (P ) ∶= ordP (φ∗ t)

for t ∈ K(C2 ) a uniformizer of φ(P ). We say P is a ramification point of φ if eφ (P ) > 1,

otherwise we say φ is unramified at P . We say φ is unramified if it has no ramification
points.
Proposition 3.6. Let φ ∶ C1 → C2 be a non-constant map of smooth curves, then for every
Q ∈ C2
∑ eφ (P ) = deg φ.
P ∈φ−1 (Q)

For all but finitely many Q ∈ C2 we have #φ−1 (Q) = degs φ. If ψ ∶ C2 → C3 is another
non-constant map of smooth curves, then

eψ○φ (P ) = eφ (P ) ⋅ eψ (φ(P )).

17
Remark 3.3. It follows from the previous proposition that a map of smooth curves is an
isomorphism if and only if it has degree 1.
Example 3.3. Consider the curve
C = {(x, y) ∈ A2 ∶ y 2 = x3 + x}.
Consider the morphism
φ ∶ C → P1
(x, y) ↦ (y ∶ 1).
Extending this morphism to C ⊆ P2 via the chart (x, y) ∈ A2 ↦ (x ∶ y ∶ 1) ∈ P2 we obtain
φ̄ ∶ C → P1
(x ∶ y ∶ z) ↦ (y ∶ z).
Computing the equation of C = {(x ∶ y ∶ z) ∈ P2 ∶ x3 + xz 2 − y 2 z = 0} it is easy to check that it
is smooth. We claim deg φ̄ = 3. In fact, we just need to note that
K(P1 ) = K(A1 ) = K(y),
and
K(C) = K(C) = K(x, y) where x3 + x = y 2 .
We claim φ̄ has only one ramification point at the point at infinity of C, i.e. at P = (0 ∶ 1 ∶ 0).
Let us compute its ramification index. Since φ̄(P ) = Q = (1 ∶ 0) we have to pull-back a local
parameter of P1 at Q. Since
z z
mQ = ⟨ ⟩ ⊆ K [ ]
y y
z
the local parameter is t = y . In order to pull-it back we have to represent t as a rational map
t ∶ P1 → P1
t(y ∶ z) = (z ∶ y).
Composing it with φ̄ we get that
z
φ̄∗ t = ∈ K(C).
y
Finally to compute
eφ̄ (P ) = ordP (φ̄∗ t)
we have to compare φ̄∗ t with a local parameter of C at P . Since
x z
mP = ⟨ , ⟩ ⊆ K[C ′ ]
y y

where C ′ = C ∩ A2 for the affine chart containing P which is given by K[A2 ] = K [ xy , yz ] an so

K [ xy , yz ]
′
K[C ] = 3 2
.
⟨( xy ) + ( xy ) ( yz ) − yz ⟩

18
Since
x z
u ∶= 1 − ( ) ( ) ∈ K[C]×P
y y
we can write
z x 3 −1
=( ) u
y y
and so x
y is a local parameter for C at P and φ̄∗ t = z
y has order 3 at P , thus

eφ̄ (P ) = 3 = deg φ̄.

Proposition 3.7. Let K be a field of characteristic p > 0 and q = pr . Let C be a curve

defined over K. Let C (q) be the image of the natural q-th Frobenius map

φ ∶ C → C (q)

(x0 ∶ ⋯ ∶ xn ) ↦ (xq0 ∶ ⋯ ∶ xqn ).

The ideal I(C (q) ) = ⟨{f (q) ∶ f ∈ I(C)}⟩ where f (q) is obtained by raising all coefficients of f
to the q-th power. The function field K(C (q) ) = K(C)q . The q-th Frobenius map is purely
inseparable of degree q. Furthermore, every map ψ ∶ C1 → C2 of smooth curves over K
factors as
φ (q) λ
C1 Ð
→ C1 Ð → C2
where q = degi ψ, φ is the q-th Frobenius and λ is separable.

3.2 Divisors
In order to give a rigorous definition of the genus of a curve and for further applications we
will introduce Riemann-Roch theorem.
Definition 3.6. Let C be a curve. The divisor group of C, denoted Div(C), is the free
abelian group generated by the points of C. A divisor D ∈ Div(C) is a formal sum

D = ∑ nP ⋅ P, nP ∈ Z
P ∈C

and nP = 0 for all but a finite number of points P ∈ C. The degree of D is

deg D ∶= ∑ nP ∈ Z.
P ∈C

The degree 0 divisors form a subgroup Div0 (C) ⊆ Div(C). If C is defined over K, the
Gal(K/K) acts on Div(C) (and on Div0 (C)) by

Dσ ∶= ∑ nP ⋅ σ(P ), ∀σ ∈ Gal(K/K).
P ∈C

We say D is defined over K if Dσ = D for all σ ∈ Gal(K/K). We denote DivK (C) (and
Div0K (C)) the group of divisors defined over K.

19
Example 3.4. In general there can be divisors D ∈ DivK (C) given by a sum of points which
are not K-rational. For example D = (i ∶ 1) + (−i ∶ 1) ∈ DivR (P1 ).

Remark 3.4. Assume C is smooth and let f ∈ K(C)× . We can associate to f the divisor

div(f ) ∶= ∑ ordP (f ) ⋅ P ∈ Div0 (C).

P ∈C

If σ ∈ Gal(K/K), (div(f ))σ = div(f σ ). In particular if f ∈ K(C) then div(f ) ∈ Div0K (C).
The map
div ∶ K(C)× → Div(C)
×
is a group homomorphism. And div(f ) = 0 if and only if f ∈ K .

Definition 3.7. A divisor D ∈ Div(C) is principal if D = div(f ) for some f ∈ K(C)× . Two
divisors D1 , D2 ∈ Div(C) are linearly equivalent, denoted D1 ∼ D2 , if D1 − D2 is principal.
The divisor class group (or Picard group) of C is

Div(C)
Pic(C) ∶= .
div(K(C)× )

We denote PicK (C) ∶= {D ∈ Pic(C) ∶ Dσ = D , ∀σ ∈ Gal(K/K)}. We denote

Div0 (C)
Pic0 (C) ∶= ⊆ Pic(C)
div(K(C)× )

and Pic0K (C) ∶= PicK (C) ∩ Pic0 (C).

Definition 3.8. Let φ ∶ C1 → C2 e a non-constant morphism of smooth curves. We define

φ∗ ∶ Div(C2 ) → Div(C1 )

Q↦ ∑ eφ (P ) ⋅ P
P ∈φ−1 (Q)

and
φ∗ ∶ Div(C1 ) → Div(C2 )
P ↦ φ(P ).

Proposition 3.8. The following properties are true:

(a) deg(φ∗ D) = deg φ ⋅ deg D ∀D ∈ Div(C2 ).

(b) φ∗ div(f ) = div(φ∗ f ) ∀f ∈ K(C2 )× .

(c) deg(φ∗ D) = deg D ∀D ∈ Div(C1 ).

(d) φ∗ ○ φ∗ = deg φ ⋅ idDiv(C2 ) .

20
 (e) If ψ ∶ C2 → C3 then (ψ ○ φ)∗ = φ∗ ○ ψ ∗ and (ψ ○ φ)∗ = ψ∗ ○ φ∗ .
Remark 3.5. In particular φ∗ and φ∗ take degree 0 divisors to degree 0 divisors and principal
divisors to principal divisors, thus they induce maps between the Picard groups.
Definition 3.9. Let C be a curve. The space of (meromorphic) differential forms on C,
denoted ΩC , is the K(C)-vector space generated by the symbols of the form df for f ∈ K(C),
subject to the relations:
(i) d(f + g) = df + dg ∀f, g ∈ K(C).
(ii) d(f ⋅ g) = f ⋅ dg + g ⋅ df ∀f, g ∈ K(C).
(iii) df = 0 ∀f ∈ K.
Remark 3.6. Let φ ∶ C1 → C2 be a non-constant morphism. The associated function field
map φ∗ ∶ K(C2 ) → K(C1 ) induces a pull-back map on differentials

φ∗ ∶ ΩC2 → ΩC1

∑ fi dxi ↦ ∑(φ∗ fi )d(φ∗ xi ).

i i

Proposition 3.9. Let C be a curve.

(a) ΩC is a 1-dimensional K(C)-vector space.
(b) Let x ∈ K(C), then dx is a K(C)-basis of ΩC if and only if K(C)/K(x) is finite
separable.
(c) Let φ ∶ C1 → C2 be a non-constant morphism, then φ is separable if and only if
φ∗ ∶ ΩC2 → ΩC1 is non-zero.
Proposition 3.10. Let C be a curve, P ∈ C be non-singular and t ∈ K(C) be a local
parameter of C at P .
(a) For every ω ∈ ΩC there exists a unique g ∈ K(C) such that

ω = g ⋅ dt.

We denote g =∶ ω/dt.
(b) Let f ∈ K(C) be regular at P , then df /dt is also regular at P .
(c) Let ω ∈ ΩC wth ω ≠ 0. The quantity ordP (ω/dt) depends only on ω and P but not on
t. We call this value the order of ω at P and denote it ordP (ω).
(d) Let x, f ∈ K(C) with x(P ) = 0 and let p = char K. Then

= ordP (f ) + ordP (x) − 1 if p = 0 or p ∤ ordP (x),

ordP (f dx) {
≥ ordP (f ) + ordP (x) otherwise.

21
 (e) Let ω ∈ ΩC with ω ≠ 0, then ordP (ω) = 0 for all but finitely many P ∈ C.
Definition 3.10. Let ω ∈ ΩC . The divisor associated to ω is

div(ω) ∶= ∑ ordP (ω) ⋅ P ∈ Div(C).

P ∈C

The differential ω is regular (or holomorphic) if ordP (ω) ≥ 0 for all P ∈ C, and it is
non-vanishing if ordP (ω) ≤ 0 for all P ∈ C. The canonical divisor class of C is

KC ∼ div(ω)

for any non-zero ω ∈ ΩC .

Definition 3.11. Let C be a curve. We say a divisor D ∈ Div(C) is effective, denoted
D ≥ 0, if
D = ∑ nP ⋅ P , nP ≥ 0 ∀P ∈ C.
P ∈C
We denote D1 ≥ D2 if D1 − D2 is effective.
Definition 3.12. Let D ∈ Div(C). The Riemann-Roch space associated to D is

L(D) ∶= {f ∈ K(C)× ∶ div(f ) ≥ −D} ∪ {0}.

This is a finite dimensional K-vector space and we denote its dimension by `(D).
Proposition 3.11. Let D ∈ Div(C).
(a) If deg D < 0, then L(D) = 0.
(b) If D1 ∼ D2 , then L(D1 ) ≃ L(D2 ).
(c) We have L(KC ) ≃ {ω ∈ ΩC ∶ ω is regular}.
Theorem 3.12 (Riemann-Roch theorem). Let C be a smooth curve and KC be a canon-
ical divisor on C. There is an integer g ≥ 0, called the genus of C, such that for every divisor
D ∈ Div(C)
`(D) − `(KC − D) = deg D − g + 1.
Corollary 3.13. `(KC ) = g, deg KC = 2g − 2 and `(D) = deg D − g + 1 for deg D > 2g − 2.
Proposition 3.14. Let C/K be a smooth curve and D ∈ DivK (C), then L(D) has a basis
of elements in K(C).
Theorem 3.15 (Hurwitz formula). Let φ ∶ C1 → C2 be a non-constant separable map of
smooth curves of genus g1 and g2 respectively. Then

2g1 − 2 ≥ (deg φ)(2g2 − 2) + ∑ (eφ (P ) − 1).

P ∈C1

With equality if and only if char K = 0 or char K = p > 0 and p ∤ eφ (P ) for all P ∈ C1 .

22
4 Elliptic Curves
Definition 4.1. An elliptic curve is a pair (E, O) where E is a smooth curve of genus
1 and O ∈ E. We just denote it E with O being understood. We say E is defined over
K, denoted E/K, if E is defined over K as an algebraic variety and O ∈ E(K). We will be
mainly concerned with the case K = Q.
Theorem 4.1. Every elliptic curve E/K with char K ≠ 2, 3 is isomorphic to a curve of the
form
C = {(x, y) ∈ A2 ∶ y 2 = 4x3 − g2 x − g3 } for some g2 , g3 ∈ K
with ∆ ∶= g23 − 27g32 ≠ 0. Under this isomorphism O ∈ E(K) goes to the point at infinity of
C. We say C is a Weierstrass normal form of E.
Proof If n > 0 = 2g − 2 we know

`(n ⋅ O) = n − g + 1 = n.

Thus there exist x ∈ L(2 ⋅ O) ∖ L(O) and y ∈ L(3 ⋅ O) ∖ L(2 ⋅ O), i.e. if P ∈ P1 is the point at
infinity P = (1 ∶ 0) then x∗ P = 2 ⋅ O and y ∗ P = 3 ⋅ O. Then

deg x = 2 and deg y = 3

in other words
[K(E) ∶ K(x)] = 2 and [K(E) ∶ K(y)] = 3.
Since [K(E) ∶ K(x, y)] divides both numbers we see that

K(E) = K(x, y).

On the other hand we have

1, x, x2 , x3 , y, y 2 , xy ∈ L(6 ⋅ O).

Since `(6 ⋅ O) = 6 it follows that there exists a K-linear relation of the form

y 2 = αx3 + a2 x2 + a4 x + a6 − a1 xy − a3 y,

but clearly α = 1 since y 2 − αx3 ∈ L(5 ⋅ O). Since char K ≠ 2 we can replace
1
y ↦ (y − a1 x − a3 )
2
and obtain the equation
y 2 = 4x3 + b2 x2 + b4 x + b6 .
Finally since char K ≠ 2, 3 we can replace
b2
x↦x−
12
23
to get the equation
y 2 = 4x3 − g2 x − g3 .
Writing
4x3 − g2 x − g3 = 4(x − r1 )(x − r2 )(x − r3 ) , r1 , r2 , r3 ∈ K
we have
−g2 g3
, r1 r2 r3 = .
r1 + r2 + r3 = 0 , r1 r2 + r2 r3 + r3 r1 =
4 4
We claim these roots are different. In fact, if r1 = r2 then
g2 −g3
r3 = −2r1 , 3r12 = , 2r13 = ,
4 4
and so r1 = 0 or r1 = −3g3 /2g2 . In any case r1 ∈ K, and since
2
y
( ) = 4(x + 2r1 )
x − r1
y
it follows that K(E) = K(x, y) = K ( x−r 1
) ≃ K(P1 ), i.e. E ≃ P1 →←.
Therefore the roots are all different and so

∆ ∶= 16(r1 − r2 )2 (r2 − r3 )2 (r3 − r1 )2 = 16(r12 + r22 − 2r1 r2 )(r1 + 2r2 )2 (2r1 + r2 )2

= 16(4((r1 + r2 )2 − r1 r2 )3 − 27r12 r22 (r1 + r2 )2 ) = g23 − 27g32 ≠ 0.

Thus we have shown that E ≃ C and O is sent to the pole of x and y, i.e. to a point at
infinity of C. Since

C = {(x ∶ y ∶ z) ∈ P2 ∶ y 2 z = 4x3 − g2 xz 2 − g3 z 3 }

the only point at infinity is P = (0 ∶ 1 ∶ 0), i.e. O goes to P .

Remark 4.1. Conversely, every curve of the form

C = {(x, y) ∈ A2 ∶ y 2 = 4x3 − g2 x − g3 }

for g2 , g3 ∈ K with g23 − 27g32 ≠ 0 is an elliptic curve with O = (0 ∶ 1 ∶ 0). In fact, since

g23 − 27g32 = 16(r1 − r2 )2 (r2 − r3 )2 (r3 − r1 )2 ≠ 0

it follows that P (x) = 4x3 − g2 x − g3 has different roots in K. If at some (x0 , y0 ) ∈ C

∂f ∂f
(x0 , y0 ) = P ′ (x0 ) = 0 and (x0 , y0 ) = −2y0 = 0
∂x ∂y

it follows that y0 = 0 and P ′ (x0 ) = 0. But since f (x0 , y0 ) = y02 − P (x0 ) = 0 it follows that
P (x0 ) = 0, i.e. x0 is a multiple root of P (x) →←, thus C is smooth.

24
 Furthermore, looking at the affine chart around the point at infinity {y ≠ 0} we have

C = {(x, z) ∈ A2 ∶ z = 4x3 − g2 xz 2 − g3 z 3 }

which is non-singular at (x, z) = (0, 0) since ∂f

∂z (0, 0) = −1 ≠ 0. In conclusion C is smooth.
Clearly O ∈ C(K) and so we just need to compute the genus of C. Consider the map

x ∶ C → P1

which extends to
x
∶ C → P1 .
z
Since [K(x, y) ∶ K(x)] = 2, we get that deg xz = 2. Then P = (x0 ∶ y0 ∶ z0 ) ∈ C is a ramification
point if and only if the equation

y 2 z0 = 4x30 − g2 x0 z02 − g3 z03

has only one double root y = y0 . Thus in case z0 ≠ 0 this implies

x0 3 x0
4( ) − g2 ( ) − g3 = 0 and y0 = 0,
z0 z0
i.e. P = (r1 ∶ 0 ∶ 1), (r2 ∶ 0 ∶ 1), (r3 ∶ 0 ∶ 1). The case z0 = 0 implies x0 = 0 and so again there is
only one possible value for (x0 ∶ y0 ∶ z0 ) = O. Thus there are 4 ramification points of index 2
and so by Hurwitz formula (char K ≠ 2)

2g − 2 = 2 ⋅ −2 + 4 = 0

in other words g = 1.
Theorem 4.2. Let char K ≠ 2, 3. Given two elliptic curves in Weiertrass form

E1 ∶ y 2 = 4x3 − g2 x − g3 , g23 − 27g32 ≠ 0,

E2 ∶ ỹ 2 = 4x̃3 − g̃2 x̃ − g̃3 , g̃23 − 27g̃32 ≠ 0,

they are isomorphic if and only if ∃µ ∈ K × such that g̃2 = µ4 g2 and g̃3 = µ6 g3 .
Proof (⇒) Let f ∶ (E1 , O1 ) → (E2 , O2 ) be the isomorphism over K. Since ordO2 (x̃) = −2
and ordO2 (ỹ) = −3 we have that ordO1 (f ∗ x̃) = −2 and ordO1 (f ∗ ỹ) = −3. Since

`(d ⋅ O1 ) = d ∀d > 0,

it follows that L(2 ⋅ O1 ) = Kx ⊕ K and L(3 ⋅ O1 ) = Ky ⊕ Kx ⊕ K and so

f ∗ x̃ = ax + b for a, b ∈ K , a ≠ 0,

f ∗ ỹ = cy + dx + e for c, d, e ∈ K , c ≠ 0.

25
Since ỹ 2 = 4x̃3 − g̃2 x̃ − g̃3 , we have (f ∗ ỹ)2 = 4(f ∗ x̃)3 − g̃2 (f ∗ x̃) − g˜3 , i.e.

(cy + dx + e)2 = 4(ax + b)3 − g̃2 (ax + b) − g̃3 .

Dividing by c2 and comparing with y 2 = 4x3 − g2 x − g3 we see that b = d = e = 0, a3 = c2 ,

g2 = ag̃2 /c2 and g3 = g̃3 /c2 . Let µ ∶= c/a, then

µ2 = a , µ3 = c , g̃2 = µ4 g2 , g̃3 = µ6 g3 .

(⇐) Let f ∶ E1 → E2 be given by f = (µ2 x, µ3 y). This is the desired isomorphism.

Remark 4.2. In particular if E1 ≃ E2 then g23 ⋅ g̃32 = g̃23 ⋅ g32 and so

g23 (g̃23 − 27g̃32 ) = g̃23 (g23 − 27g32 ).

Definition 4.2. Let E be an elliptic curve in Weierstrass form

y 2 = 4x3 − g2 x − g3 , g2 , g3 ∈ K , g23 − 27g32 ≠ 0.

The modular invariant is

g23 g3
j(E) ∶= = 3 2 2.
∆ g2 − 27g3
This is an invariant of elliptic curves, i.e. it is invariant under isomorphism.
Remark 4.3. For every j ∈ K we can describe all ismorphism classes of elliptic curves in
Weiestrass normal form with modular invariant j. For j = 0, i.e. g2 = 0 we have

{isomorphism classes} ←→ K × /(K × )6

E z→ g3 .
For j = 1, i.e. g3 = 0 we have

{isomorphism classes} ←→ K × /(K × )4

E z→ g2 .
For j ≠ 0, 1, i.e. g2 g3 ≠ 0 we have

{isomorphism classes} ←→ K × /(K × )2

E z→ g3 /g2 ,
since if g̃3 /g̃2 = µ2 g3 /g2 and g̃32 /g̃23 = (j − 1)/27j = g32 /g23 , then

g̃23 g̃3 2 g23 2 g3 2

g̃2 = 2
( ) = 2 (µ ) = µ4 g2 ,
g̃3 g̃2 g3 g2

g̃23 g̃3 3 g23 2 g3 3

g̃3 = 2 ( ) = 2 (µ ) = µ6 g3 .
g̃3 g̃2 g3 g2

26
Corollary 4.3. If K × = (K × )2 = (K × )3 , for instance if K = K, then we have a correspondence

j ∶ {isomorphism classes of elliptic curves over K} ←→ A1 (K).

Example 4.1. For K = R, there are exactly two non-isomorphic elliptic curves with modular
invariant j. In fact, for j = 0 they are given by g3 = ±1, for j = 1 the are given by g2 = ±1,
and for j ≠ 0, 1 they are given by g3 /g2 = ±1. For instance we can see this topologically for

E1 = {y 2 = 4x3 − x} and E2 = {y 2 = 4x3 + x}

where E1 has two connected components (two circles) while E2 has only one (a circle). These
are essentially the two possible topologies underlying every pair of non-isomorphic elliptic
curves over R with the same modular invariant.

Remark 4.4. In the case of a singular Weierstrass equation we have two possible topological
spaces, e.g.
C1 = {y 2 = 4x3 − x2 } and C2 = {y 2 = 4x3 }
corresponding to the node (in the first case) where P (x) = 4x− g2 x − g3 has one root of
multiplicity 2, and the cusp (in the second case) where P (x) has only one root of multiplicity
3, i.e. P (x) = 4x3 .

Definition 4.3. We say a singular Weierstrass equation

y 2 = 4x3 − g2 x − g3 , g2 , g3 ∈ K , g23 − 27g32 = 0,

has a node if g2 ≠ 0. In such a case the node is (r, 0) where r is the repeated root of
P (x) ∶= 4x3 − g2 x − g3 . We say it has a cusp if g2 = 0, i.e. P (x) = 4x3 . In this case the cusp
corresponds to the point (0, 0).

Proposition 4.4. If a curve given by a Weiestrass equation

C = {(x, y) ∈ A2 ∶ y 2 = 4x3 − g2 x − g3 }

is singular, then there exists a rational map φ ∶ C ⇢ P1 of degree 1.

Proof If P = (r, 0) is the singularity of C, take

φ ∶ C ⇢ P1

(x, y) ↦ (x − r ∶ y)
2
which induces φ∗ ∶ K(P1 ) = K ( x−r x−r
y ) ↪ K(C) = K(x, y) for 1 = 4 ( y ) (x − s), i.e. x =
y2
s + 41 ( x−r ) ∈ K(P1 ) and so K(C) = K(P1 ).

27
Remark 4.5. Given a Weierstrass equation

y 2 = 4x3 − g2 x − g3 , g2 , g3 ∈ K.

Write 4x3 − g2 x − g3 = 4P (x) for char K ≠ 2. If

P (x) = (x − r1 )(x − r2 )(x − r3 ) , r1 , r2 , r3 ∈ K,

then replacing x by x(r2 − r1 ) + r1 (assume r1 ≠ r2 i.e. that the equation has not a cusp) we
get
r3 − r1
y 2 = 4(r2 − r1 )2 x(x − 1)(x − λ) for λ = .
r2 − r1
Replacing y by ηy where η 2 = 4(r1 − r2 )2 we get the Legendre form

y 2 = x(x − 1)(x − λ) for λ ∈ K.

Proposition 4.5. For char K ≠ 2, every elliptic curve E/K is isomorphic over K to an
elliptic curve in Legendre form for some λ ∈ K ∖{0, 1}. Its modular invariant is for char K ≠ 3
4(λ2 − λ + 1)3
j(E) = .
27λ2 (λ − 1)2
ϕ
The map λ ∈ K ∖ {0, 1} Ð → j(Eλ ) ∈ K is surjective and 6 to 1 except above j = 0 and j = 1
where it is 2 to 1 and 3 to 1 respectively.
Proof By our proof of the Weierstrass equation we can write

E = {(x, y) ∈ A2 ∶ y 2 = 4x3 + b2 x2 + b4 x + b6 }.

Writing 4x3 + b2 x2 + b4 x + b6 = 4P (x) = 4(x − r1 )(x − r2 )(x − r3 ) and repeating the procedure
above we get the Legendre form. Let us compute its modular invariant
g23 −64(r1 r2 + r2 r3 + r3 r1 )3 r3 − r1
j(E) = = for r1 + r2 + r3 = 0 and λ = .
∆ 16(r1 − r2 )2 (r2 − r3 )2 (r3 − r1 )2 r2 − r1
Thus
r12 +r22 +r32 −(r1 r2 +r2 r3 +r3 r1 ) 3
4( (r2 −r1 )2 ) 4(λ2 − λ + 1)3
j(E) = = .
27λ2 (λ − 1)2 27λ2 (λ − 1)2
Finally, it is not hard to check that for any λ ∈ K ∖ {0, 1}
1
ϕ(1 − λ) = ϕ(λ) = ϕ ( ) .
λ
Thus ϕ−1 (j) is invariant under the maps λ ↦ 1 − λ and λ ↦ λ1 . Therefore if ϕ(λ) = j then
1 1 λ−1 λ
{λ, , 1 − λ, , , } ⊆ ϕ−1 (j).
λ 1−λ λ λ−1

28
It is elementary to check that for λ ≠ 2, −1, 12 , α, β for (x − α)(x − β) = x2 − x + 1, the above 6
numbers are different, and so they are the 6 roots of the equation

3(λ2 − λ + 1)3 = 27λ2 (λ − 1)2 j.

On the other hand for j = 0, the previous equation has only 2 roots α and β. While for j = 1
the equation has the 3 roots 12 , −1, 2.
Remark 4.6. In general, if we want to work with char K = 2 or 3, every elliptic curve has
Weierstrass normal form of the form

y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , a1 , . . . , a6 ∈ K.

In this general case we still can compute the discriminant ∆ in terms of the coefficients. John
Tate showed that in these characteristics the right modular invariant should correspond to
27 ⋅ 33 ⋅ j.
Example 4.2. Let K = K and char K ≠ 2, 3. Consider the curve

C = {(x, y) ∈ A2 ∶ y 2 = (x − e1 )(x − e2 )(x − e3 )(x − e4 )}

with e1 , e2 , e3 , e4 ∈ K different. This curve has only one singularity at infinity (0 ∶ 1 ∶ 0).
Considering the Hurwitz formula applied to the rational map x ∶ C → P1 we can see that
g = 1, and so it is birational to an elliptic curve. Let us transform e4 to the point at infinity,
i.e. replace x by x1 + e4
1 1 1 1
y 2 = ( + e4 − e1 ) ( + e4 − e2 ) ( + e4 − e3 ) .
x x x x
In order to solve the singularity replace y by ya/x2 where a2 = (e4 − e3 )(e4 − e2 )(e4 − e1 ), then
1 1 1
y 2 = (x + ) (x + ) (x + ).
e4 − e1 e4 − e2 e4 − e3
x 1 y (e4 −e3 )(e4 −e2 )
Normalizing 2 of the 3 roots replace x by b2 + e3 −e4
and y by b3 for b2 = e3 −e2 we get

(e1 − e3 )(e2 − e4 )
y 2 = x(x − 1)(x − λ) for λ = ,
(e1 − e4 )(e2 − e3 )
i.e. λ is the cross ratio of e1 , e2 , e3 , e4 .
Definition 4.4. Let E = {y 2 = 4x3 − g2 x − g3 } be an elliptic curve in Weierstrass form. The
invariant differential is
dx
ω ∶= ∈ ΩE .
y
Proposition 4.6. The invariant differential is holomorphic and non-vanishing, i.e.

div(ω) = 0.

29
Proof For P = (x0 , y0 ) ∈ E, if y0 ≠ 0 then x − x0 is a local parameter at P and y1 ∈ K[E]×P ,
and so ordP (ω) = 0. For y0 = 0, then P (x0 ) = 0 and P ′ (x0 ) ≠ 0, thus y is a local parameter
at P and P ′2(x) ∈ K[E]×P . Since
2ydy = P ′ (x)dx
then
2dy
ω=
P ′ (x)
and so ordP (ω) = 0. Finally we have to compute the order of ω at the point at infinity
O. Since the sum of all orders equal the degree of ω which is 2g − 2 = 0 we conclude that
ordO (ω) = 0.

5 The Mordell-Weil Group

In this lecture we will introduce the group structure of the rational points of an elliptic curve.
This group structure is the key to construct more rational points from a given set of rational
points, and it is a peculiarity of genus one curves. Before going to the group structure, let
us explore the problem of constructing rational points on an elliptic curve.
Example 5.1. Let E = {(x, y) ∈ A2 ∶ y 2 = x3 − 25x} be defined over Q. We know that
P = (5, 0) ∈ E(Q).
If we try to consider a stereographic projection to a line L defined over Q, we see that for
every P ′ ∈ L(Q) the line P P ′ is defined over Q and so the other two intersection points
Q, R ∈ P P ′ ∩ E
are the roots of a quadratic equation defined over Q. But in general, this is not enough to
conclude that Q or R are Q-rational points. In fact, what we conclude is the following
Q ∈ E(Q) ⇐⇒ R ∈ E(Q).
In this way, given two rational points P, Q ∈ E(Q) we can produce a third rational point
R ∈ E(Q). If we try to apply this method starting with the 3 trivial solutions
(−5, 0), (0, 0), (5, 0) ∈ E(Q)
we see that we cannot produce any other rational point. The same happens if we add to this
set the point at infinity O = (0 ∶ 1 ∶ 0). Suppose we have found now a non-trivial solution,
for instance
Q = (−4, 6) ∈ E(Q).
Considering the line L = P Q, with P = (5, 0) we can construct another rational point. Let
us find this point. We can write
L = {(x, y) ∈ A2 ∶ 3y = −2x + 10},

30
and so the intersection with E is given by
2 10 2
(− x + ) = x3 − 25x
3 3
which is equivalent to

0 = 9x3 − 4x2 − 185x − 100 = (x − 5)(x + 4)(9x + 5).

Thus we obtain R = (− 95 , 100

24 ) ∈ E(Q). We can find more points combining this method with
the fact that
(x, y) ∈ E(Q) ⇐⇒ (x, −y) ∈ E(Q).
We will see that the composition of these two methods is essentially the group law in E(Q).
Remark 5.1. Note that we can also consider in the previous example the case P = Q. In
this case L corresponds to the tangent line to E at Q = (−4, 6), i.e.

L = {(x, y) ∈ A2 ∶ 12y = 23x + 164}.

The points of L ∩ E satisfy

23 41 2 1681
0 = −( x + ) + x3 − 25x = (x + 4)2 (x − )
12 3 144
and so we get
1681 62279
( , ) ∈ E(Q).
144 1728
Proposition 5.1. Let E/K be an elliptic curve, then the map
AE
E(K) Ð→ Pic0K (E)

P ↦P −O
is bijective.
Proof Let P, Q ∈ E(K) such that P − O ∼ Q − O, i.e. ∃f ∈ K(E)× such that

div(f ) = P − Q.
∼
If P ≠ Q then f ∈ L(Q)∖K and so f ∶ E Ð → P1 which is absurd since E has genus 1. This shows
the injectivity. For the surjectivity let D ∈ Div0K (E), then since deg(D + O) = 1 > 0 = 2g − 2
it follows that
`(D + O) = 1 − g + 1 = 1.
Let h ∈ L(D + O) ∖ {0} and let D′ ∶= D + O + div(h) ≥ 0. Since D′ is effective of degree 1,
D′ = P for some P ∈ E(K). Thus
D ∼ P − O.

31
Definition 5.1. Let E/K be an elliptic curve. The Mordell-Weil group of E over K is
∼
→ Pic0K (E), i.e.
the set E(K) together with the group law induced by the map AE ∶ E(K) Ð
for P, Q ∈ E(K)
AE (P + Q) ∶= AE (P ) + AE (Q).
Remark 5.2. By the definition of AE we see that AE (O) = O − O = 0, and so the zero of
E(K) corresponds to the fixed point O ∈ E(K). Given P, Q ∈ E(K) we want to determine
P + Q = R ∈ E(K). We know it is characterized as the unique point such that

P −O+Q−O ∼R−O

in other words
P + Q ∼ R + O.
In the case
E = {(x ∶ y ∶ z) ∈ P2 ∶ F (x, y, z) = 0} for some deg F = 3.
Then if we consider the line P Q = {L(x, y, z) = ax + by + cz = 0} joining P and Q, it will
intersect E again at some point R′ ∈ E(K). Then

P + Q + R′ ∼ R + O + R′ ,

and so there exists some degree one homogeneous polynomial L′ (x, y, z) = a′ x + b′ y + c′ z such
that
L
div ( ′ ) = P + Q + R − (R + O + R′ ).
L
Therefore R, O and R are the three intersection points of the line {L′ = 0}∩E. In conclusion
′

R is obtained as the third intersection point of the line OR′ with E.

Example 5.2. If E is given in Weierstrass form

E = {(x, y) ∈ A2 ∶ y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 } with ai ∈ K,

then O = (0 ∶ 1 ∶ 0) ∈ E(K) is the point at infinity. Given P = (x1 , y1 ) and Q = (x2 , y2 )

different, then R′ = (x3 , y3 ) is the third point of P Q ∩ E, i.e. it satisfies

(mx + n)2 + a1 x(mx + n) + a3 (mx + n) = x3 + a2 x2 + a4 x + a6

y2 −y1 x2 y1 −x1 y2
where m = x2 −x1 and n = x2 −x1 . This equation can be factored as

(x − x1 )(x − x2 )(x − (m2 + a1 m − a2 − x1 − x2 )) = 0

and so
R′ = (x′3 , y3′ ) = (m2 + a1 m − a2 − x1 − x2 , mx′3 + n).
The line joining R′ and O is R′ O = {x = x′3 z} and so

R = (x′3 , −y3′ − a1 x′3 − a3 ).

32
Corollary 5.2. Let E/K be an elliptic curve in Weierstrass form

E = {(x, y) ∈ A2 ∶ y 2 = 4x3 − g2 x − g3 }

and P = (x1 , y1 ), Q = (x2 , y2 ) ∈ E(K) ∖ {O} different. Then (for char K ≠ 2)

1 y2 − y1 2 y2 − y1
P +Q=( ( ) − x1 − x2 , y 1 + ( ) (x3 − x1 )) and − P = (x1 , −y1 ).
4 x2 − x1 x2 − x1
In particular + ∶ E × E → E and − ∶ E → E are morphisms of algebraic varieties over K.
Corollary 5.3 (Homogeneity). Let E/K be an elliptic curve. Then for any two points
P, Q ∈ E(K) there exists an automorphism α ∶ E → E defined over K such that α(P ) = Q.
Proof Take α(x) ∶= x + Q − P defined on E(K). Since P, Q ∈ E(K), this morphism (and
its inverse) is also defined over K.
Corollary 5.4. Let E/K be an elliptic curve and P ∈ E(K), then
⎧
⎪ {µ ∈ K ∶ µ6 = 1} if j(E) = 0,
⎪
⎪
{α ∈ Aut(E/K) ∶ α(P ) = P } ≃ ⎨ {µ ∈ K ∶ µ4 = 1} if j(E) = 1,
⎪
⎪
⎪
⎩ {±1} otherwise.
Proof By the homogeneity we can assume P = O, and so we are looking for the automor-
phisms fixing O. By the proof of the theorem characterizing when two elliptic curves in
Weierstrass normal form are isomorphic, we know α = (µ2 x, µ3 y) with µ ∈ K × and g2 = µ4 g2 ,
g3 = µ6 g3 .

5.1 Elliptic curves over Q

Example 5.3. Let E = {(x, y) ∈ A2 ∶ y 2 = x3 + 1} be defined over Q, and

P = (2, 3) ∈ E(Q).

Let us compute the subgroup generated by P . In order to find 2P we have to write down
the tangent line to E at P which is

L = {(x, y) ∈ A2 ∶ y = 2x − 1}.

The intersection with E is given by

0 = x3 + 1 − (2x − 1)2 = (x − 2)2 x

and so R′ = (0, −1) thus

2P = (0, 1).
In order to find 3P we consider the line joining P and 2P which is

L = {(x, y) ∈ A2 ∶ y = x + 1}.

33
The intersection of this line with E is given by

0 = x3 + 1 − (x + 1)2 = (x + 1)x(x − 2)

and so R′ = (−1, 0), thus

3P = (−1, 0).
In order to find 4P we consider the tangent line to E at 2P which is

L = {(x, y) ∈ A2 ∶ y = 1}.

This line intersects E only at 2P and so R′ = (0, 1), thus

4P = (0, −1).

Let us find 5P joining the points 2P and 3P , i.e. take the line

L = {(x, y) ∈ A2 ∶ y = x + 1}.

We already know the third point of intersection is R′ = (2, 3) and so

5P = (2, −3).

Finally to compute 6P take the tangent line to E at 3P , which is

L = {(x, y) ∈ A2 ∶ x = −1}.

This line intersects E two times at 3P and the third intersection point is O = (0 ∶ 1 ∶ 0) the
point at infinity. Thus R′ = O and so
6P = O.
In other words Z ⋅ P ≃ Z/6Z.

Theorem 5.5 (Mordell-Weil theorem). Let E/Q be an elliptic curve, then

E(Q) is a finitely generated abelian group.

Remark 5.3. By the structure theorem of finitely generated abelian groups we know there
exists a (non-canonical) decomposition

E(Q) = E(Q)tors ⊕ ZRE ,

where RE ∶= rank E(Q) is the rank of the elliptic curve and

E(Q)tors ∶= {P ∈ E(Q) ∶ n ⋅ P = 0 , for some n ∈ Z>0 }

is the torsion subgroup of E(Q).

34
Definition 5.2. We say P ∈ E(Q)tors is a torsion point of order n if n⋅P = 0 and m⋅P ≠ 0
for 1 ≤ m < n. We denote
E[n] ∶= {P ∈ E(Q) ∶ n ⋅ P = 0}
the subgroup of n-torsion points.
Remark 5.4. The Mordell-Weil theorem holds in general for any elliptic curve E/K defined
over a number field K. One of the main ingredients in the proof of Mordell-Weil theorem is
the following.
Theorem 5.6 (Weak Mordell-Weil theorem). Let E/Q be an elliptic curve, then for all
m ≥ 2,
E(Q)/mE(Q) is a finite group.
Example 5.4. In what follows we show some examples of Mordell-Weil groups without
doing the computations.
1. The elliptic curve
E1 ∶= {(x, y) ∈ A2 ∶ y 2 = x3 + 6}
has no other rational point but O, i.e.

E1 (Q) ≃ 0.

2. The curve
E2 ∶= {(x, y) ∈ A2 ∶ y 2 = x3 + 1}
has only 6 rational points and so

E2 (Q) ≃ Z/6Z.

3. The curve
E3 (Q) ∶= {(x, y) ∈ A2 ∶ y 2 = x3 − 2}
does not have any torsion point but O, but P = (3, 5) ∈ E3 (Q). In fact, P is a generator
of the Mordell-Weil group and so

E3 (Q) ≃ Z.

4. The elliptic curve

E4 ∶= {(x, y) ∈ A2 ∶ y 2 = x3 + 7105x2 + 1327104x}

has both torsion and infinite order points. In fact

E4 (Q) ≃ Z/4Z ⊕ Z3 .

The torsion part is generated by T = (1152, 111744), while the free part is generated
by P1 = (−6912, 6912), P2 = (−5832, 188568) and P3 = (−5400, 206280).

35
 5. The elliptic curve associated to the congruent number problem
En ∶= {(x, y) ∈ A2 ∶ y 2 = x3 − n2 x}
has three trivial rational points (−n, 0), (0, 0), (n, 0). It is easy to check that these
three points have order 2, in fact
En (Q)tors = {(−n, 0), (0, 0), (n, 0), O} ≃ Z/2Z ⊕ Z/2Z.
Thus the congruent number problem can be translated into determining when En (Q)
has positive rank.
Remark 5.5. In the previous examples we have seen several different possibilities for the
Mordell-Weil group. Then it is natural to ask ourselves if all finitely generated are realized
as the Mordell-Weil group of some elliptic curve E(Q). Ogg conjectured that there are only
a finite amount of finite abelian groups realized as E(Q)tors and gave a complete list. This
conjecture was later proved by Mazur.
Theorem 5.7 (Mazur’s theorem). Let E/Q be an elliptic curve, then E(Q)tors is isomor-
phic to exactly one of the following groups:
Z/N Z for 1 ≤ N ≤ 10 or N = 12, or
Z/2Z ⊕ Z/2M Z for 1 ≤ M ≤ 4.
Corollary 5.8. Let E/Q be an elliptic curve. If P ∈ E(Q) is not in E[n] for some n ≤ 12,
then it has infinite order.
Example 5.5. The elliptic curve given by
y 2 + 43xy − 210y = x3 − 210x2
has torsion part isomorphic to Z/12Z, generated by (0, 210). On the other hand, the elliptic
curve given by
y 2 + 17xy − 120y = x3 − 60x2
has torsion part isomorphic to Z/2Z⊕Z/8Z, generated by the points (30, −90) and (−40, 400).
Remark 5.6. In fact for any elliptic curve we can effectively find the torsion subgroup of
the Mordell-Weil group. In order to do this note first that every elliptic curve over Q admits
a normal form given by
y 2 = x3 + Ax + B , A, B ∈ Z.
In fact, writing the Weierstrass normal form
y 2 = 4x3 − g2 x − g3 , g2 , g3 ∈ Q,
replacing y by 2y we get a equation of the form
p r
y 2 = x3 + x + , p, q, r, s ∈ Z.
q s
Replacing y by y/(qs)3 and x by x/(qs)2 we get the desired equation
y 2 = x3 + Ax + B , A = pq 3 s4 , B = rp6 s5 .

36
Theorem 5.9 (Nagell-Lutz theorem). Let E/Q be an elliptic curve given by

E = {(x, y) ∈ A2 ∶ y 2 = x3 + Ax + B}

for A, B ∈ Z. Then every torsion point P ≠ O satisfies:

(i) The coordinates of P are integers, i.e. x(P ), y(P ) ∈ Z.

(ii) If P is a point of order n ≥ 3, then

y(P )2 ∣4A3 + 27B 2 .

(iii) If P has order 2, then y(P ) = 0 and so x(P )3 + Ax(P ) + B = 0.

Theorem 5.10 (Siegel’s theorem). Let E/Q be an elliptic curve given by

E = {(x, y) ∈ A2 ∶ y 2 = x3 + Ax + B}

with A, B ∈ Z. Then E has a finite number of integral points.

Remark 5.7. In spite of Siegel’s theorem, there is no effective way to compute the integral
points, nor to bound the amount of them. Another hard problem is computing the rank of
an elliptic curve. Up to the date it is not known if there is an upper bound for the possible
values of RE . The largest known is RE = 28 found by Elkies (see Dujella’s website).

6 Elliptic Curves over Local Fields

6.1 Local fields
In this lecture we will study elliptic curves over local fields. Let us recall the basic definitions
and examples of local fields.
Definition 6.1. Let K be a field. A discrete valuation of K is a surjective map

v ∶ K× → Z

such that
(i) v(f ⋅ g) = v(f ) + v(g),

(ii) v(f + g) ≥ min{v(f ), v(g)}.

We extend the valuation to K by defining v(0) ∶= ∞.
Proposition 6.1. Let v be a discrete valuation on K and let f1 , . . . , fk ∈ K. If v(f1 ) < v(fi ),
∀i = 2, . . . , k, then
v(f1 + ⋯ + fk ) = v(f1 ).

37
Proof By (ii) we know v(f1 +⋯+fk ) ≥ v(f1 ). On the other since v is a group homomorphism
v(1) = 0 and so 2v(−1) = v(1) = 0, then v(−1) = 0. Thus v(−fi ) = v(fi ). Suppose v(f1 + ⋯ +
fk ) > v(f1 ). Then v(f1 ) = v(f1 +⋯+fk −f2 −⋯−fk ) ≥ min{v(f1 +⋯+fk ), v(−f2 ), ⋯, v(−fk )} >
v(f1 ). This is a contradiction.

Definition 6.2. Given a discrete valuation v on a field K, we define its ring of integers

Rv ∶= {f ∈ K ∶ v(f ) ≥ 0}.

This ring is a discrete valuation ring, i.e. it is local with maximal ideal

mv ∶= {f ∈ K ∶ v(f ) > 0}

which is principal, and any generator of it is called a uniformizer or local parameter

π ∈ mv such that
⟨π⟩ = mv .
We define the residue field of K at v as
Rv
kv ∶= .
mv
Remark 6.1. Note that
Rv× = {f ∈ K ∶ v(f ) = 0},
and so π ∈ K is a local parameter if and only if v(π) = 1. Furthermore, every f ∈ K × is of
the form f = u ⋅ π k for u ∈ Rv× and k = v(f ) ∈ Z.

Example 6.1. Our main examples are the following:

1. For every prime number p. The field Q has a p-adic valuation

a
vp (pk ) ∶= k
b
for a, b ∈ Z not divisible by p. In this example
a
Rvp = ZpZ = { ∈ Q ∶ p ∤ b},
b
a
mvp = p ⋅ ZpZ = { ∈ Q ∶ p ∣ a , p ∤ b},
b
kvp ≃ Fp .

2. Let C be a smooth curve and P ∈ C be any non-singular point. Then the function
order at P
v = ordP ∶ K(C)× → Z

38
 is a discrete valuation on K(C), with
f
Rv = K[C]P = { ∶ g(P ) ≠ 0},
g
f
mv = mP = { ∶ f (P ) = 0 , g(P ) ≠ 0},
g
kv = K.
3. Let K be any field and consider K((t)) be the field of formal Laurent power series
of the form
∑ an tn for some m ∈ Z.
n≥m
This field has a t-adic valuation
vt ( ∑ an tn ) ∶= min{n ∶ an ≠ 0}.
n≥m

In this case
Rvt = K[[t]],
mvt = t ⋅ K[[t]],
kv = K.
Definition 6.3. Let v be a discrete valuation on a field K. The norm induced by v is
∣f ∣v ∶= 2−v(f ) .
This is a field norm on K. We say (K, v) is a complete field, if ∣ ⋅ ∣v induces an structure
of complete metric space on K. We say K is a local field if it is a complete field whose
residue field is finite.
Example 6.2. Our main example of local field is Qp the field of p-adic numbers. They
are the completion of Q with respect to the p-adic norm ∣ ⋅ ∣p . Thus every element of Qp
corresponds to a Laurent power series of the form
∑ an p n , with an ∈ {0, 1, . . . , p − 1}.
n≥m

Its ring of integers is denoted

Zp ∶= Rvp = { ∑ an pn ∈ Qp }.
n≥0

Note that we have a natural inclusion

Q ↪ Qp
given by taking any rational number pk ⋅ ab to c ⋅ pk , where c ∈ {0, 1, . . . , p − 1} is such that
b ⋅ c ≡ a (mod p). Note also that the residue field is also
kvp = Fp .
Our second main example is K((t)) for K a finite field.

39
Theorem 6.2 (Hensel’s lemma). Let (K, v) be a complete field, and F ∈ Rv [x] be a
polynomial. Suppose there is an integer n ≥ 1 and some a ∈ Rv such that

F (a) ∈ mnv and F ′ (a) ∈ Rv× .

Then there exists a unique b ∈ Rv such that

F (b) = 0 and b − a ∈ mnv .

6.2 Reduction method

Let us turn now to the study of elliptic curves E/K for K a local field. We will keep the
notation K for a local field, complete with respect to a discrete valuation v. We will denote
by
Rv = {x ∈ K ∶ v(x) ≥ 0}
its ring of integers, by
mv = {x ∈ K ∶ v(x) > 0}
its maximal ideal, and by
kv = Rv /mv
its residue field. We will commonly denote by π ∈ mv a uniformizer or local parameter for
R, i.e. ⟨π⟩ = mv or equivalently v(π) = 1.

Remark 6.2. Recall that for any elliptic curve any field E/K, it admits a Weierstrass
representation of the form

y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 . (2)

If we replace x by x/u2 and y by y/u3 we obtain a new Weierstrass equation where each ai
is replaced by ui ai . Thus if v(u) is large enough we can assume all v(ai ) ≥ 0, i.e. that all
ai ∈ Rv . In particular since ∆ is a polynomial in terms of the coefficients ai we can assume
∆ ∈ Rv .

Definition 6.4. Let E/K be an elliptic curve. A Weierstrass equation for E is called a min-
imal Weierstrass equation for E at v if v(∆) is minimized among all Weierstrass equations
with ai ∈ Rv . This minimal value is called the valuation of the minimal discriminant
of E at v.

Remark 6.3. We will focus on the case

y 2 = x3 + Ax + B with A, B ∈ Rv .

In this case we take

∆ ∶= 16(r1 − r2 )2 (r2 − r3 )2 (r3 − r1 )2 = −16(4A3 + 27B 2 )

40
for x3 + Ax + B = (x − r1 )(x − r2 )(x − r3 ) and r1 , r2 , r3 ∈ K. If we translate Theorem 4.2 to
this context we have that every isomorphism between two of such elliptic curves is given by
a map of the form
α = (µ2 x, µ3 y) , for some µ ∈ K × .
In particular α transforms ∆ into
∆′ = µ−12 ∆.
This fact holds in general for any isomorphism between two Weierstrass equations as in (2).
Thus v(∆) can be changed only for multiples of 12, so we conclude that
v(∆) < 12 Ô⇒ the equation is minimal.
Similarly since A′ = µ−4 A and B ′ = µ−6 B, we have
v(A) < 4 Ô⇒ the equation is minimal,
v(B) < 6 Ô⇒ the equation is minimal.
Corollary 6.3. Every elliptic curve E/K has a minimal Weierstrass equation. Such minimal
equation is unique up to a change of coordinates α = (µ2 x, µ3 y) for some µ ∈ Rv× .
Definition 6.5. Let E/K be an elliptic curve with a minimal Weierstrass equation of the
form
y 2 = x3 + Ax + B , A, B ∈ Rv .
Denoting for the quotient map reduction modulo π by
r ∈ Rv ↦ r̃ ∈ kv = Rv /mv ,
we define Ẽ/kv the reduction of E modulo π as the curve given by the minimal equation
modulo π
y 2 = x3 + Ãx + B̃.
Remark 6.4. Note that by Corollary 6.3 the reduction Ẽ is unique up to isomorphism, i.e.
up to a change of coordinates of the form α̃ = (µ̃2 x, µ̃3 y) for µ̃ ∈ kv× .
Definition 6.6. Let P ∈ E(K). Writing P = (x0 ∶ y0 ∶ z0 ) in homogeneous coordinates, let
k ∶= min{v(x0 ), v(y0 ), v(z0 )} ∈ Z.
Then we can represent P = (x1 ∶ y1 ∶ z1 ) where x1 = x0 ⋅ π −k , y1 = y0 ⋅ π −k and z1 = z0 ⋅ π −k , i.e.
all the coordinates are in Rv and at least one is in Rv× . Thus the reduced point
P̃ ∶= (x̃1 ∶ ỹ1 ∶ z̃1 ) ∈ Ẽ(kv ).
This defines a reduction map
E(K) → Ẽ(kv )
which in fact comes from a reduction map in the projective plane
P2 (K) → P2 (kv ).

41
Remark 6.5. In general the reduction Ẽ/kv is not always an elliptic curve. In fact it may
happen that it becomes singular at some (non infinity) point. This happens if and only if
˜ = 0, i.e. when ∆ ∈ mv . In spite of that, the set of non-singular kv -rational points
∆
Ẽns (kv ) has a natural group structure defined in the same way as the usual Mordell-Weil
group, i.e. for P̃ , Q̃ ∈ Ẽns (kv ) their sum corresponds to

P̃ ⊕ Q̃ ∶= R̃ ∈ Ẽns (kv )

where R̃ is the third intersection point of ÕR̃′ ∩ Ẽns and R̃′ ∈ Ẽns (kv ) is the third intersection
point of P̃ Q̃ ∩ Ẽns .

Definition 6.7. We define the set of points with non-singular reduction as

E0 (K) ∶= {P ∈ E(K) ∶ P̃ ∈ Ẽns (kv )}.

We define the kernel of reduction

E1 (K) ∶= {P ∈ E(K) ∶ P̃ = Õ}.

Proposition 6.4. There is a short exact sequence of abelian groups

0 → E1 (K) → E0 (K) → Ẽns (kv ) → 0

where the right-hand map is reduction modulo π.

Proof Let us first show that E0 (K) is a subgroup of E(K). In fact, given P, Q ∈ E0 (K) let
R′ ∈ E(K) be the third intersection point of P Q ∩ E. Since the intersection of a line with a
cubic curve correspond to exactly 3 points counted with multiplicity, it follows that if R̃′ is
different from P̃ and Q̃, then P̃ Q̃ intersects Ẽ with multiplicity one at R̃′ , i.e. Ẽ is smooth
at R̃′ and P̃ Q̃ is transverse to Ẽ at R̃′ . In case R̃′ is equal to P̃ or Q̃, then R̃′ ∈ Ẽns (kv ). In
any case R′ ∈ E0 (K). Since clearly O ∈ E0 (K), it follows by the same argument that

R = P + Q ∈ E0 (K).

Now we just need to show that the restriction map is an epimorphism of groups. Let us
show first it is surjective. Let f (x, y) ∈ Rv [x, y] be a minimal equation of E/K, and let
f˜(x, y) ∈ kv [x, y] be its reduction modulo π. Let P̃ = (α̃, β̃) ∈ Ẽns (kv ) i.e. such that

∂ f˜ ∂ f˜
f˜(α̃, β̃) = 0 and, (α̃, β̃) ≠ 0 or (α̃, β̃) ≠ 0.
∂x ∂y
∂ f˜
Suppose ∂y (α̃, β̃) ≠ 0. Take any x0 ∈ Rv such that x̃0 = α̃, and consider the equation

f (x0 , y) = 0.

42
When reduced modulo π this equation has β̃ as a simple root, and so by Hensel’s lemma, there
exists some y0 ∈ Rv such that f (x0 , y0 ) = 0 and ỹ0 = β̃. In other words P = (x0 , y0 ) ∈ E(K)
reduces modulo π to P̃ ∈ Ẽns (kv ).
In order to finish the proof we are left to show that the reduction map is a group homo-
morphism. Keeping in mind that the rule for constructing the addition of two points in both
groups is the same, and that for every line L defined over K, there exist a, b, c ∈ Rv with at
least one in Rv× such that L = {ax + by + cz = 0}. Thus we can choose the coefficients of the
equations of the lines involved in the process in such a way that the reduction modulo π at
each step of the algorithm gives us the addition algorithm for the reduced points.

Remark 6.6. It follows from the previous proposition that if L is any line joining two points
P, Q ∈ E0 (K) such that P̃ = Q̃, then L̃ is tangent to Ẽ at P̃ = Q̃ ∈ Ẽns (kv ).

Theorem 6.5. Let p = char kv . Then every torsion point P ∈ E1 (K) has order a power of
p. Furthermore if the order of P is pn then

x(P ) v(p)
v( )≤ n .
y(P ) p − pn−1

Remark 6.7. We will skip the proof the theorem above to see some applications.

Proposition 6.6. Let E/K be an elliptic curve, and let m ≥ 1 be an integer relatively prime
to char kv . Then

(a) The subgroup E1 (K) has no non-trivial points of order m.

(b) Assume further that the reduced curve Ẽ/kv is non-singular. Then the reduction map

E(K)[m] → Ẽ(kv )

is injective.

Proof Item (a) follows directly from Theorem 6.5. For (b) consider any P ∈ E(K)[m] such
that P̃ = Õ. Since Ẽ is non-singular, E(K) = E0 (K) and so P is in the kernel of reduction,
i.e. P ∈ E1 (K). Therefore the order of P is a divisor of m and a power of p = char kv , thus
P = O.

Remark 6.8. Using the above proposition we can get information about the torsion points
of E(K) for instance when K is a number field. In fact, since for any discrete valuation v,
we can extended the number field to its completion K ↪ Kv and so if the reduction Ẽ/kv is
non-singular then
E(K)[m] ↪ E(Kv )[m] ↪ Ẽ(kv )
for m relatively prime to char kv . In particular this shows that E(K)tors is always finite.

43
Example 6.3. Let E/Q be given by

y 2 + y = x3 − x + 1.

The discriminant of E is ∆ = −13 ⋅ 47, and so it is non-singular modulo 2. Let us determine

Ẽ(F2 ). Let P̃ = (x̃0 , ỹ0 ) ∈ Ẽ(F2 ), since x̃0 ∈ F2 we have two options, if x̃0 = 0 then

ỹ02 + ỹ0 = ỹ0 (ỹ0 + 1) ≡ 1 (mod 2),

which is impossible. Thus x̃0 = 1 but again we get to the same impossibility. Therefore
Ẽ(F2 ) = {Õ}. By Proposition 6.6 we conclude that E(Q) has no torsion point of odd order
m > 1, thus every torsion point has to be of order a power of 2. On the other hand if
P = (x0 , y0 ) ∈ E(Q)[2], then consider the tangent line to E at P

L = {(x, y) ∈ A2 ∶ (2y0 + 1)y = (3x20 − 1)x + y02 − 2x20 + 1}.

Since 2P = O we have that 2y0 + 1 = 0 and so 4x30 − 4x0 + 5 = 0. But this implies x0 ∉ Q.
Therefore E(Q) has no points of order 2, and so

E(Q)tors = {O}.

Example 6.4. Let E/Q be given by

y 2 = x3 + 3.

It has discriminant ∆ = −24 ⋅ 35 and so Ẽ is non-singular modulo p ≥ 5. We check that

Ẽ(F5 ) = 6 and Ẽ(F7 ) = 13.

Thus if P ∈ E(Q)[p] for some prime p ≠ 7, then the subgroup generated by P is a

Z/pZ ↪ Ẽ(F7 ) ≃ Z/13Z,

i.e. p = 13. But since 13 ≠ 5, then the subgroup generated by P is a

Z/13Z ↪ Ẽ(F5 )

which is impossible. Finally if p = 7, again the subgroup generated by P is some

Z/7Z ↪ Ẽ(F5 )

which is impossible. In conclusion

E(Q)tors = {O}.

44
Example 6.5. Consider E/Q given by
y 2 = x3 + x.
It has discriminant ∆ = −64 and the point (0, 0) has order 2. Computing
Ẽ(F3 ) = {O, (0, 0), (2, 1), (2, 2)} ≃ Z/4Z,
Ẽ(F5 ) = {O, (0, 0), (2, 0), (3, 0)} ≃ Z/2Z ⊕ Z/2Z,
we conclude that
E(Q)tors = {O, (0, 0)} ≃ Z/2Z.
Theorem 6.7. Assume char K = 0 and p = char kv > 0. Let E/K be an elliptic curve given
by any Weierstrass equation (not necessarily minimal)
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , ai ∈ Rv .
Let P ∈ E(K) be a point of order m ≥ 2, then
(a) If m is not a power of p, then x(P ), y(P ) ∈ Rv .
(b) If m = pn then
v(p)
π 2r x(P ), π 3r y(P ) ∈ Rv with r = ⌊ ⌋.
pn
− pn−1
Proof If x(P ) ∈ Rv we are done, and so assume v(x(P )) < 0. Changing coordinates to
(x′ , y ′ ) in order to obtain a minimal equation we get
v(x(P )) ≥ v(x′ (P )) and v(y(P )) ≥ v(y ′ (P )).
Thus it is enough to prove the theorem for a minimal equation. Since v(x(P )) < 0 we see
from the Weierstrass equation that
3v(x(P )) = 2v(y(P )) = −6s
for some integer s ≥ 1. Further the point P ∈ E1 (K) and by Theorem 6.5 we conclude (a).
If the order of P is pn , then by Theorem 6.5 we conclude
x(P ) v(p)
s = v( )≤ n .
y(P ) p − pn−1
Since π 2s x(P ), π 3s y(P ) ∈ Rv we conclude (b).
Remark 6.9. Let E/Q be an elliptic curve given by a Weierstrass equation having coef-
ficients in Z, and let P ∈ E(Q) be of order m. If m is not a power of a prime p then
x(P ), y(P ) ∈ Z. And if m = pn , then for v the p-adic valuation we get
v(p) 1
⌊ ⌋=⌊ n ⌋=0
pn
−p n−1 p − pn−1
for (p, n) ≠ (2, 1). Thus x(P ), y(P ) ∈ Z for every torsion point of order at least 3.

45
7 Formal Group of an Elliptic Curve
Our main purpose in this lecture is to prove Theorem 6.5. In order to do this we will relate
the group E1 (K) for K a local field, with a formal group associated to any such elliptic
curve. Let us start by introducing formal groups.

7.1 Formal groups

Definition 7.1. Let R be a ring. A (one-parameter commutative) formal group F
over R is a formal power series F (X, Y ) ∈ R[[X, Y ]] with the following properties:
(a) F (X, Y ) = X + Y + XY ⋅ H(X, Y ), for some H(X, Y ) ∈ R[[X, Y ]].
(b) F (X, F (Y, Z)) = F (F (X, Y ), Z).
(c) F (X, Y ) = F (Y, X).
(d) There exists a unique power series i(T ) ∈ R[[T ]] such that F (T, i(T )) = 0.
We call F (X, Y ) the formal group law of F.
Definition 7.2. Let (F, F ) and (G, G) be two formal groups defined over R. A homomor-
phism from F to G defined over R is a power series f ∈ R[[T ]] with no constant term
such that
f (F (X, Y )) = G(f (X), f (Y )).
The formal groups F and G are isomorphic over R if there are homomorphisms f ∶ F → G
and g ∶ G → F defined over R such that

f (g(T )) = g(f (T )) = T.

Example 7.1. The formal additive group Ĝa is defined by

F (X, Y ) = X + Y.

Example 7.2. The formal multiplicative group Ĝm is defined by

F (X, Y ) = X + Y + XY = (1 + X)(1 + Y ) − 1.

Example 7.3. Let (F, F ) be a formal group. We define homomorphisms

[m] ∶ F → F

inductively for m ∈ Z by

[0](T ) ∶= 0 , [m + 1](T ) ∶= F ([m](T ), T ) , [m − 1](T ) ∶= F ([m](T ), i(T )).

One can check by induction that [m] is a homomorphism. This map is called the multipli-
cation by m map.

46
Lemma 7.1. Let a ∈ R× and let f (T ) ∈ R[[T ]] be a power series of the form
f (T ) = aT + (higher order terms).
Then there is a unique power series g(T ) ∈ R[[T ]] such that
f (g(T )) = T.
This series also satisfies g(f (T )) = T .
Proof We construct a sequence of polynomials gn (T ) ∈ R[T ] such that
f (gn (T )) ≡ T (mod T n+1 ) and gn+1 (T ) ≡ gn (T ) (mod T n+1 ).
The second property implies that the limit is a power series g(T ) ∈ R[[T ]], and by the first
property f (g(T )) ≡ f (gn (T )) ≡ T (mod T n+1 ) for all n, in other words f (g(T )) = T .
We will construct the polynomials gn (T ) inductively, beginning with
g1 (T ) ∶= a−1 T.
Suppose now that gn−1 (T ) has been constructed with the desired properties. We are looking
for a polynomial of the form
gn (T ) ∶= gn−1 (T ) + λT n
for some λ ∈ R such that
f (gn (T )) ≡ T (mod T n+1 ).
Since
f (gn (T )) = f (gn−1 (T ) + λT n ) ≡ f (gn−1 (T )) + aλT n (mod T n+1 ),
it follows by induction hypothesis that
f (gn (T )) ≡ T + bT n + aλT n (mod T n+1 )
for some b ∈ R. Thus it is enough to take λ ∶= −ba−1 to complete the induction.
In order to prove that
g(f (T )) = T
note that by what we have proved there exists some h(T ) ∈ R[[T ]] such that
g(h(T )) = T.
Then
h(T ) = f (g(h(T ))) = f (T ).
Finally, to prove the uniqueness of g(T ) consider some G(T ) ∈ R[[T ]] such that
f (G(T )) = T.
Then
G(T ) = g(f (G(T ))) = g(T ).

47
Proposition 7.2. Let F be a formal group over R and let m ∈ Z, then
(a) [m](T ) = mT + (higher order terms).

(b) If m ∈ R× , then [m] ∶ F → F is an isomorphism.

Proof (a) For m ≥ 0 is direct from the definition. For m negative, it also follows from the
definition once we have noted that since

0 = F (X, i(X)) = X + i(X) + X ⋅ i(X) ⋅ H(X, i(X)),

then necessarily
i(X) = −X + (higher order terms).
(b) Applying the lemma above to [m] we get some g(T ) ∈ R[[T ]] such that [m](g(T )) = T
and g([m](T )) = T . We just have to show that g is a homomorphism, and this follows from
the fact [m] is a homomorphism. In fact

g(F (X, Y )) = g(F ([m](g(X)), [m](g(Y )))) = g([m](F (g(X), g(Y )))) = F (g(X), g(Y )).

Remark 7.1. In order to understand better the structure of the multiplication by m map
we have to introduce differential forms on formal groups.
Definition 7.3. Let F/R be a formal group over a ring R. An invariant differential of
F/R is a differential
ω(T ) = P (T )dT ∈ R[[T ]]dT
such that
∂F
ω ○ F (T, S) ∶= P (F (T, S)) (T, S)dT = ω(T ).
∂X
We say ω(T ) is normalized if P (T ) has constant term equal to 1.
Example 7.4. On the formal additive group Ĝa , ω(T ) = dT is invariant normalized.
Example 7.5. On the formal multiplicative group Ĝm ,
dT
ω(T ) = = (1 − T + T 2 − T 3 + ⋯)dT
1+T
is invariant normalized.
Proposition 7.3. Let F/R be a formal group. There exists a unique normalized invariant
differential on F/R. It is given by
∂F
ω(T ) = (0, T )−1 dT.
∂X
Every invariant differential is of the form aω(T ) for some a ∈ R.

48
Proof Suppose P (T )dT is an invariant differential, then
∂F
P (F (0, T )) (0, T )dT = P (0)dT.
∂X
Letting P (0) = a ∈ R and noting that F (0, T ) = T we get that
∂F
P (T ) = a ⋅ (0, T )−1 .
∂X
On the other hand, to see that this is effectively an invariant differential, note that differen-
tiating the equality
F (U, F (T, S)) = F (F (U, T ), S)
with respect to U we get
∂F ∂F ∂F
(U, F (T, S)) = (F (U, T ), S) (U, T ).
∂X ∂X ∂X
Evaluating at U = 0 we get
∂F ∂F ∂F
(0, F (T, S)) = (T, S) (0, T )
∂X ∂X ∂X
and so
∂F ∂F ∂F
ω ○ F (T, S) = a ⋅ (0, F (T, S))−1 (T, S)dT = a ⋅ (0, T )−1 dT = ω(T ).
∂X ∂X ∂X

Corollary 7.4. Let F/R and G/R be formal groups with normalized differentials ωF and
ωG . Let f ∶ F → G be a homomorphism. Then
ωG ○ f = f ′ (0) ⋅ ωF .
Proof We just need to show that ωG ○ f is an invariant differential. In fact
(ωG ○ f ) ○ F (T, S) = ωG ○ (f ○ F (T, S)) = ωG ○ G(f (T ), f (S)) = ωG ○ f.
Writing ωG = P (T )dT we see that
ωG ○ f (T ) = P (f (T )) ⋅ f ′ (T )dT
and so it has constant term f ′ (0).
Corollary 7.5. Let F/R be a formal group and let p ∈ Z be a prime. Then there are power
series f (T ), g(T ) ∈ R[[T ]] with f (0) = g(0) = 0 such that
[p](T ) = pf (T ) + g(T p ).
Proof Let ω(T ) be the normalized invariant differential of F. Since by Proposition 7.2
[p]′ (0) = p, then
pω(T ) = (ω ○ [p])(T ) = P ([p](T )) ⋅ [p]′ (T )dT.
Since the series P ([p](T )) is invertible in R[[T ]] (because it has constant term 1), then
[p]′ (T ) ∈ p ⋅ R[[T ]].
Therefore every term of the power series expansion of [p](T ) is of the form an T n for p∣an or
p∣n.

49
7.2 Group associated to a formal group
Formal groups are merely a group operation without an underlying group. However, in the
case R is the ring of integers of a complete field (K, v), we can evaluate each variable of
F (X, Y ) at the points of mv . Thus F (X, Y ) induces a group structure on mv .

Definition 7.4. We say that R is a complete local ring if it is the ring of integers of a
complete field.

Definition 7.5. Let R be a complete local ring with maximal ideal m, and let F be a formal
group over R. The group associated to F/R, denoted F(m), is the set m with the group
operation
x ⊕F y ∶= F (x, y) ∈ m.
This is a group, with zero the element 0 ∈ m, and inverse given by

⊖F x = i(x) ∈ m.

For n ≥ 1 we define the subgroup

F(mn ) ∶= {x ∈ F(m) ∶ x ∈ mn }.

Example 7.6. The group associated to the formal additive group is

Ĝa (m) = (m, +).

While the group associated to the formal multiplicative group is

Ĝm (m) ≃ (1 + m, ⋅)

x ↦ 1 + x.

Remark 7.2. Every homomorphism of formal groups f ∶ F → G induces a group homomor-

phism f ∶ F(m) → G(m). In fact

f (x ⊕F y) = f (F (x, y)) = G(f (x), f (y)) = f (x) ⊕G f (y).

In particular every isomorphism of formal groups induces an isomorphism their associated

groups.

Example 7.7. For example, the multiplication by m map induces a map [m] ∶ F(m) → F(m)
corresponding to
[m](x) = [m − 1](x) ⊕F x = ⋯ = x ⊕F ⋯ ⊕F x
where the sum is taken m times. In other words

[m](x) = m ⋅ x.

50
Proposition 7.6. Let R be a complete local ring with maximal ideal m and F be a formal
group defined over R.

(a) For each n ≥ 1, the identity map

F(mn ) mn
→
F(mn+1 ) mn+1

is an isomorphism of groups.

(b) Let p be the characteristic of the residue field k = R/m. Then every element of finite
order of F(m) has order a power of p.

Proof (a) We just need to show the identity is a homomorphism. This follows since

x ⊕F y = F (x, y) ≡ x + y (mod mn+1 ).

(b) It is enough to show there is no point of order m relatively prime to p. And this is clear
since if p ∤ m, then m ∈ k × and so m ∈ R× . Thus by Proposition 7.2 the multiplication by m
map is an isomorphism.

Theorem 7.7. Let R be a complete local ring relative to the complete field (K, v), and let
p = char kv . Let F/R be a formal group over R and suppose x ∈ F(m) has order pn for some
n ≥ 1. Then
v(p)
v(x) ≤ n .
p − pn−1
Proof If char R ≠ 0 or p = 0, then v(p) = ∞ and the result is trivial. Thus we can assume
char R = 0 and p > 0.It follows from Corollary 7.5 that

[p](T ) = pf (T ) + g(T p ),

and by Proposition 7.2 it follows that f (T ) = T + (higher order terms). Let us prove the
theorem by induction on n. For n = 1 assume the order of x ∈ F(m) is p, then [p](x) = px = 0.
Thus
0 = pf (x) + g(xp ).
Since the linear term of f (T ) is T , then

v(px) = v(−pf (x)) = v(g(xp )) ≥ v(xp ) = pv(x),

and so
v(p)
v(x) ≤ .
p−1
Now, assume the result is true for n and consider x ∈ F(m) of order pn+1 . Then

v([p](x)) = v(pf (x) + g(xp )) ≥ min{v(px), v(xp )}.

51
Since [p](x) has order pn , we have by induction hypothesis that
v(p)
≥ v([p](x)) ≥ min{v(px), v(xp )}.
pn− pn−1
Since v(x) > 0, then v(px) > v(p)/(pn − pn−1 ) and so
v(p)
≥ v(xp ) = pv(x).
pn − pn−1

7.3 Formal group of an elliptic curve

Finally we will introduce the formal group associated to an elliptic curve. In order to do this
we will recall that every elliptic curve E/K defined over a local field (K, v) has a Weierstrass
normal form
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , ai ∈ Rv .
Considering the affine chart around the point at infinity (z, w) = ( xy , y1 ) we get the following
representation
f (z, w) = w + a1 zw + a3 w2 − z 3 − a2 z 2 w − a4 zw2 − a6 w3 = 0.
∂f
Thus ∂w (z, 0) = 1 for every z ∈ Rv . By Hensel’s lemma, for every z ∈ mv there exists a unique
3
w ∈ mv such that
f (z, w) = 0.
In fact, if we look at f as a polynomial on w with coefficients in the ring Z[a1 , a2 , a3 , a4 , a6 ][[z]],
we see that
w(z) = z 3 (1 + A1 z + A2 z 2 + ⋯) , A1 , A2 , . . . ∈ Z[a1 , a2 , a3 , a4 , a6 ].
In particular we can also compute Laurent series expansion for
1
x(z) =2
(1 + B1 z + B2 z 2 + ⋯) ∈ Z[a1 , a2 , a3 , a4 , a5 ]((z)),
z
1
y(z) = 3 (1 + C1 z + C2 z 2 + ⋯) ∈ Z[a1 , a2 , a3 , a4 , a6 ]((z)).
z
Recalling the formula for the inverse of a rational point on an elliptic curve in Weierstrass
form we define
x(z)
i(z) ∶= ∈ Z[a1 , a2 , a3 , a4 , a6 ][[z]].
−y(z) − a1 x(z) − a3
And furthermore we can compute the formal power series corresponding to the sum of two
rational points also. In fact, given z1 , z2 ∈ mv , we let w1 = w(z1 ) and w2 = w(z2 ). In the
(z, w)-plane the line joining (z1 , w1 ) and (z2 , w2 ) has slope
∞
w2 − w1 z n+3 − z1n+3
λ = λ(z1 , z2 ) = = z12 + z1 z2 + z22 + ∑ An ( 2 ) ∈ Z[a1 , a2 , a3 , a4 , a6 ][[z1 , z2 ]].
z2 − z1 n=1 z2 − z1

52
Letting
ν = ν(z1 , z2 ) = w1 − λz1 ∈ Z[a1 , a2 , a3 , a4 , a6 ][[z1 , z2 ]]
we see that the line connecting (z1 , w1 ) with (z2 , w2 ) corresponds to

w = λz − ν.

Thus
0 = f (z, λz − ν) = (−1 − λa2 − λ2 a4 − λ3 a6 )(z − z1 )(z − z2 )(z − z3 )
where
λa1 + λ2 a3 + νa2 + 2λνa4 + 3λ2 νa6
z3 (z1 , z2 ) = −z1 − z2 + ∈ Z[a1 , a2 , a3 , a4 , a6 ][[z1 , z2 ]].
1 + λa2 + λ2 a4 + λ3 a6
Therefore we can define the formal group law of the elliptic curve E as

F (z1 , z2 ) ∶= i(z3 (z1 , z2 )) ∈ Z[a1 , a2 , a3 , a4 , a6 ][[z1 , z2 ]].

It follows from the definition, and the fact that K-rational points on E form a group, that
this is a formal group, called the formal group of E and is denoted Ê. It is clear that it
comes with a natural group homomorphism

Ê(mv ) → E1 (K)

z ↦ (x(z), y(z))
which is injective. In fact it is an isomorphism, with inverse map

E1 (K) → Ê(mv )
x
(x, y) ↦ .
y
To see that this map is well-defined just note that since (x, y) reduces modulo π to the point
at infinity, then v(x) < 0 and v(y) < 0. But then by the Weierstrass equation

2v(y) = 3v(x) = −6k , k ≥ 1,

and so xy ∈ mv . Since this map is also injective, and is a left inverse of the above map, we
conclude it is the inverse map. Finally, we are in position to prove Theorem 6.5.

Proof of Theorem 6.5 Since E1 (K) ≃ Ê(mv ) every torsion point has order a power of p by
)
Proposition 7.6. And if P ∈ E1 (K) is a point of order pn , then it corresponds to x(P
y(P ) ∈ Ê(mv )
with the same order. By Theorem 7.7

x(P ) v(p)
v( )≤ n .
y(P ) p − pn−1

53
8 Mordell-Weil Theorem
In this lecture we will prove Mordell-Weil theorem. In order to do this we will recall first
some results from Algebraic Number Theory. For a reference see [Neu13, Chapter II]. .

8.1 Some facts from algebraic number theory

Definition 8.1. Let K be a number field. We denote by SK the set of all discrete valuations
of K. For any finite extension L/K and every v ∈ SK there exists a finite amount of extensions
w ∈ SL such that
w ∶ K × → ew/v ⋅ Z ⊆ Z
satisfies w∣K × = ew/v ⋅ v. The integer ew/v ≥ 1 is called the ramification index of L/K at w.
We say L/K is unramified at w if ew/v = 1. We say it is unramified at v if ew/v = 1 for all
extensions w/v. For S ⊆ SK we say L/K is unramified at S if it is unramified at all v ∈ S.

Proposition 8.1. Let L/K be an extension of number fields and let S ⊆ SK . If L1 , L2 ⊆ L

are two subextensions of K, both unramified over S, then L1 L2 /K is unramified over S.

Definition 8.2. Let L/K be a finite Galois extension of number fields, let v ∈ SK and w ∈ SL
an extension. The inertia group of w/v is

Iw/v ∶= {σ ∈ Gal(L/K) ∶ w ○ σ = w , and σ(a) ≡ a (mod mw ) ∀a ∈ Rw }.

Proposition 8.2. A finite Galois extension of number fields L/K is unramified at v ∈ SK if

and only if L is fixed by Iw/v for all w ∈ SL extending v.

Proposition 8.3. Let K be a number field. Let S ⊆ SK be a finite set of discrete valuations,
and let m ≥ 2 be an integer. Let L/K be the maximal abelian extension of K having exponent
m that is unramified outside S. Then L/K is a finite extension.

Proof See [Sil09, Chapter VIII, Proposition 1.6].

8.2 Proof of Mordell-Weil theorem

Theorem 8.4 (Weak Mordell-Weil theorem). Let K be a number field, E/K be an
elliptic curve and m ≥ 2 be any integer. Then

E(K)/mE(K)

is finite.

Proposition 8.5. Let K be a number field, L/K be a finite Galois extension and m ≥ 2 be
any integer. If E(L)/mE(L) is finite then E(K)/mE(K) is also finite.

54
Proof We have a natural map
Φ ∶ E(K)/mE(K) → E(L)/mE(L)
with kernel
E(K) ∩ mE(L)
ker Φ = .
mE(K)
Note that for any P ∈ ker Φ, there exists some QP ∈ E(L) such that
P = m ⋅ QP .
Thus we can define the map of sets
λP ∶ Gal(L/K) → E[m]
σ ↦ σ(QP ) − QP .
This map is well defined since
m ⋅ (σ(QP ) − QP ) = σ(m ⋅ QP ) − m ⋅ QP = σ(P ) − P = O.
We claim the map
ker Φ → Map(Gal(L/K), E[m])
P ↦ λP
is injective. In fact, if λP = λP ′ then
σ(QP ) − QP = σ(QP ′ ) − QP ′ , ∀σ ∈ Gal(L/K).
In other words QP − QP ′ ∈ E(K). It follows that
P − P ′ = m ⋅ (QP − QP ′ ) ∈ mE(K).
Therefore ker Φ is finite, and so if E(L)/mE(L) is finite, then E(K)/mE(K) also is.
Remark 8.1. The above proposition shows that in order to prove the Weak Mordell-Weil
Theorem it is enough to reduce ourselves to the case
E[m] ⊆ E(K). (3)
In fact, this assumption can be done after extending K to the normal closure of the field
∏P ∈E[m] K(P ), which is certainly a finite Galois extension of K. From now on we will assume
K satisfies (3).
Remark 8.2. The next step in our proof is to introduce the Kummer pairing, which will
reduce the finiteness of E(K)/mE(K) to the finiteness of a certain Galois extension L/K.
In order to introduce this paring note that since E[m] is a finite set, the multiplication by
m map
m∶E→E
is non-constant (because E(K) contains infinitely many points). In particular, since it is a
morphism, it is surjective.

55
Definition 8.3. The Kummer pairing

κ ∶ E(K) × Gal(K/K) → E[m]

is defined as follows. Let P ∈ E(K) and choose any Q ∈ E(K) such that

P =m⋅Q

then
κ(P, σ) ∶= σ(Q) − Q.
Proposition 8.6. The Kummer pairing is well-defined and induces a perfect pairing

E(K)/mE(K) × Gal(L/K) → E[m]

where L = ∏m⋅Q∈E(K) K(Q).

Proof It is clear that κ(P, σ) ∈ E[m]. We have to verify it is independent of the choice of
Q. In fact if R ∈ E[m] ⊆ E(K) then

σ(Q + R) − Q − R = σ(Q) + R − Q − R = σ(Q) − Q.

It is clearly linear on P . To see the linearity on σ take

κ(P, στ ) = σ(τ (Q)) − Q = σ(τ (Q) − Q) + σ(Q) − Q = σ(κ(P, τ )) + κ(P, σ).

Since κ(P, σ) ∈ E[m] ⊆ E(K) the linearity follows. Let us compute the left kernel. Consider
P ∈ E(K) such that
σ(Q) = Q for all σ ∈ Gal(K/K).
Then Q ∈ E(K) and so P = m ⋅ Q ∈ mE(K). For the right kernel let σ ∈ Gal(K/K) such that
σ(Q) = Q for all Q ∈ E(K) such that m ⋅ Q ∈ E(K). Then σ ∈ Gal(K/L) by the definition of
L. This shows the right kernel of the Kummer pairing is Gal(K/L) the L/K is normal and
so Galois. Therefore
Gal(K/K)/Gal(K/L) ≃ Gal(L/K)
and the result follows.
Definition 8.4. Let K be a number field, E/K be an elliptic curve and v be a discrete
valuation on K. We say that E has good reduction at v if Ẽ/kv is non-singular. Otherwise
we E has bad reduction at v.
Remark 8.3. Note that if E/Q is any elliptic curve written in some Weierstrass normal
form
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , a1 , a2 , a3 , a4 , a6 ∈ Z.
Then for all but finitely many primes vp (∆) = 0, and so the above equation is a minimal
model for those primes and furthermore E has good reduction at them. This shows that for
K = Q there are only a finite number of primes of bad reduction. The same happens for K
any number field.

56
Proposition 8.7. Let
L= ∏ K(Q).
m⋅Q∈E(K)

(a) The extension L/K is abelian and has exponent m, i.e. Gal(L/K) is an abelian group
and the order of every σ ∈ Gal(L/K) is a divisor of m.

(b) Let
S ∶= {v ∈ SK ∶ E has bad reduction at v} ∪ {v ∈ SK ∶ v(m) ≠ 0}.
Then L/K is unramified at SK ∖ S.
Proof (a) It follows immediately from the perfect paring that Gal(L/K) is abelian, and in
fact we have an injection

Gal(L/K) ↪ hom(E(K)/mE(K), E[m]).

Thus it also has exponent m.

(b) Let v ∈ SK ∖ S, let Q ∈ E(K) such that m ⋅ Q ∈ E(K). By Proposition 8.1 it is enough to
show that K(Q)/K is unramified at v. Let w ∈ SK(Q) be an extension of v. Since v ∉ S, E
has good reduction at v and so it also has good reduction at w, since we can take the same
Weierstrass equation. Let σ ∈ Iw/v , by definition σ acts trivially on Ẽ(kw ), then
̃
σ(Q) ̃ − Q̃ = σ(Q̃) − Q̃ = Õ,
− Q = σ(Q)

i.e. σ(Q) − Q ∈ E1 (K). On the other hand since m ⋅ Q ∈ E(K),

m ⋅ (σ(Q) − Q) = σ(m ⋅ Q) − m ⋅ Q = O.

But since v(m) = 0 it follows that m ∈ Rv× and so m̃ ∈ kv× , thus p ∤ m. Therefore by
Proposition 6.6 we conclude σ(Q) − Q = O and so K(Q) is fixed by Iw/v . The result follows
by Proposition 8.2.

Now we are in position to prove the first ingredient of Mordell-Weil theorem.

Proof of Weak Mordell-Weil Theorem The perfect pairing induced by the Kummer
pairing reduces us to show that L/K is a finite extension. In order to show this we note that
by Proposition 8.7 L/K is contained in the maximal abelian extension K having exponent
m that is unramified outside S. By Proposition 8.3 this extension is finite.

The second ingredient is the descent theorem that we prove next.

Theorem 8.8 (Descent theorem). Let A be an abelian group. Suppose that there exists
a height function
h∶A→R
with the following properties:

57
 (i) For every Q ∈ A there exists a constant C1 , depending on A and Q, such that
h(P + Q) ≤ 2h(P ) + C1 for all P ∈ A.

(ii) There are an integer m ≥ 2 and a constant C2 , depending on A, such that

h(mP ) ≥ m2 h(P ) − C2 for all P ∈ A.

(iii) For every constant C3 , the set

{P ∈ A ∶ h(P ) ≤ C3 }
is finite.
Suppose further that for the integer m of (ii), the quotient group A/mA is finite. Then A is
finitely generated.
Proof Choose elements Q1 , . . . , Qr ∈ A that represent all the cosets of A/mA. Let P ∈ A,
then
P = mP1 + Qi1 , for some 1 ≤ i1 ≤ r.
Repeating this process we can produce a sequence of points P1 , P2 , . . . such that
Pn−1 = mPn + Qin .
It follows that
1 1 1
h(Pj ) ≤ 2
(h(mPj ) + C2 ) = 2 (h(Pj−1 − Qij ) + C2 ) ≤ 2 (2h(Pj−1 ) + C1′ + C2 )
m m m
where C1′ is the maximum of the constants C1 associated to the points −Q1 , . . . , −Qr . Using
the above inequality several times we get
2 n 1 2 2 2 2 n−1 ′ 1 1 ′
h(Pn ) ≤ ( ) h(P )+ (1 + + ( ) + ⋯ + ( ) ) (C 1 +C 2 ) < h(P )+ (C +C2 )
m2 m2 m2 m2 m2 2n 2 1
and so for n >> 1
1
h(Pn ) ≤ 1 + (C1′ + C2 ).
2
Since n
P = mn Pn + ∑ mj−1 Qij ,
j=1

it follows that the finite set

1
{Q1 , . . . , Qr } ∪ {Q ∈ A ∶ h(Q) ≤ 1 + (C1′ + C2 )}
2
generates A.

In order to apply the descent theorem to E(Q) and conclude Mordell-Weil theorem, we
just need to prove the existence of the adequate height function.

58
 p
Definition 8.5. Let t = q ∈ Q with gcd(p, q) = 1. The height of t is

H(t) ∶= max{∣p∣, ∣q∣}.

Let E/Q be an elliptic curve. The (logarithmic) height on E(Q), relative to a given
Weierstrass equation, is the function

hx ∶ E(Q) → R≥0

given by
log H(x(P )) if P ≠ O,
hx (P ) = {
0 if P = O.

Proposition 8.9. Let E/Q be an elliptic curve given by a Weierstrass equation

y 2 = x3 + Ax + B , A, B ∈ Z.

Then the logarithmic height function satisfies the hypothesis of the descent theorem for
m = 2. In consequence E(Q) is finitely generated.

Proof (i) The result is clear for P0 = O, and considering C1 > max{hx (P0 ), hx (2P0 )} we also
get the result for P ∈ {O, ±P0 }. In the other cases we write

a b a0 b 0
P = (x, y) = ( , ) and P0 = (x0 , y0 ) = ( , ),
d2 d3 d20 d30

minimizing the numerator and denominator of the fractions (without changing their forms).
We know that
y − y0 2 (xx0 + A)(x + x0 ) + 2B − 2yy0
x(P + P0 ) = ( ) − x − x0 =
x − x0 (x − x0 )2

(aa0 + Ad2 d20 )(ad20 + a0 d2 ) + 2Bd4 d40 − 2bdb0 d0

=
(ad20 − a0 d2 )2
and so
H(x(P + P0 )) ≤ C1′ max{∣a∣2 , ∣d∣4 , ∣bd∣}.
Furthermore, noting that if δ = gcd(a, d2 ), since

b2 = a3 + Aad4 + Bd6 ,

it follows that δ 3 ∣b2 and so δ 4 ∣(bd)2 . Therefore δ 2 divides the numerator and denominator of
the expression for x(P + P0 ) above, and so we get the better inequality

max{∣a∣2 , ∣d∣4 , ∣bd∣}

H(x(P + P0 )) ≤ C1′ .
δ2

59
On the other hand, again by the equation b2 = a3 + Aad4 + Bd6 we see that

∣b∣ ≤ C1′′ max{∣a∣ 2 , ∣d∣3 }

3

and so replacing on the previous inequality we get

max{∣a∣2 , ∣d∣4 }
H(x(P + P0 )) ≤ C1 = C1 H(x(P ))2 .
δ2
(ii) Choosing C2 ≥ 4 max{hx (Q) ∶ Q ∈ E(Q)[2]} we get the result for all Q ∈ E(Q)[2]. For
P ∈ E(Q) ∖ E(Q)[2] we write P = (x, y) and so
x4 − 2Ax2 − 8Bx + A2
x(2P ) = .
4x3 + 4Ax + 4B
a
Writing x = b we get
F (a, b)
x(2P ) =
G(a, b)
for F (X, Y ) = X 4 − 2AX 2 Y 2 − 8BXY 3 + A2 Y 4 and G(X, Y ) = 4X 3 Y + 4AXY 3 + 4BY 4 .
Consider the following identities

f1 (X, Y )F (X, Y ) − g1 (X, Y )G(X, Y ) = 4∆Y 7 ,

f2 (X, Y )F (X, Y ) − g2 (X, Y )G(X, Y ) = 4∆X 7 ,

where
f1 (X, Y ) = 12X 2 Y + 16AY 3 ,
g1 (X, Y ) = 3X 3 − 5AXY 2 − 27BY 3 ,
f2 (X, Y ) = 4∆X 3 − 4A2 BX 2 Y + 4A(3A3 + 22B 2 )XY 2 + 12B(A3 + 8B 2 )Y 3 ,
g2 (X, Y ) = A2 BX 2 + A(5A3 + 32B 2 )X 2 Y + 2B(13A3 + 96B 2 )XY 2 − 3A2 (A3 + 8B 2 )Y 3 .
Let δ ∶= gcd(F (a, b), G(a, b)). The by the previous identities we see that

δ∣4∆ and so ∣δ∣ ≤ ∣4∆∣.

Thus
max{∣F (a, b)∣, ∣G(a, b)∣}
H(x(2P )) ≥ .
∣4∆∣
On the other hand the same identities give us

∣4∆b7 ∣ ≤ 2 max{∣f1 (a, b)∣, ∣g1 (a, b)∣} max{∣F (a, b)∣, ∣G(a, b)∣},

∣4∆a7 ∣ ≤ 2 max{∣f2 (a, b)∣, ∣g2 (a, b)∣} max{∣F (a, b)∣, ∣G(a, b)∣}.
Since the polynomials f1 , g1 , f2 , g3 are homogeneous of degree 3 it follows that

max{∣f1 (a, b)∣, ∣g1 (a, b)∣, ∣f2 (a, b)∣, ∣g2 (a, b)∣} ≤ C max{∣a∣3 , ∣b∣3 }.

60
Therefore
max{∣4∆a7 ∣, ∣4∆b7 ∣} ≤ 2C max{∣a∣3 , ∣b∣3 } max{∣F (a, b)∣, ∣G(a, b)∣}.
And consequently
max{∣F (a, b)∣, ∣G(a, b)∣}
≥ (2C)−1 max{∣a∣4 , ∣b∣4 } = (2C)−1 H(x(P ))4 .
∣4∆∣
(iii) Finally {P ∈ E(Q) ∶ H(x(P )) ≤ C} is finite since there are only a finite number of
possible values for x(P ), and for each of these values there are at most two possible values
for y(P ).

9 Endomorphisms of Elliptic Curves

In this lecture we will study the endomorphism ring of an elliptic curve.

9.1 Isogenies
Definition 9.1. Let E1 and E2 be two elliptic curves. An isogeny from E1 to E2 is a
morphism (only as algebraic varieties)
φ ∶ E1 → E2 such that φ(O) = O.
Remark 9.1. Since an isogeny is a morphism between smooth curves, there is only one
constant isogeny φ = [0] corresponding to the multiplication by 0 map in the case E1 = E2 ,
and all the others are finite maps. Clearly the multiplication by m maps [m] ∶ E → E are
examples of isogenies.
Definition 9.2. Two elliptic curves are called isogenous if there exist two non-constant
isogenies φ ∶ E1 → E2 and ψ ∶ E2 → E1 .
Remark 9.2. Given two elliptic curves E1 and E2 , the set of morphisms (as algebraic
varieties)
hom(E1 , E2 )
can be turned into a group with the operation
(φ + ψ)(P ) ∶= φ(P ) + ψ(P ) , ∀P ∈ E1 ,
where the sum at the right hand side is the group law of E2 (K). If E1 = E2 , then we can
also compose isogenies, thus
End(E) ∶= hom(E, E)
becomes a ring with respect to the composition and sum. This ring is not necessarily
commutative. Note that it not obvious at all that the distributive law holds, since we
do not know if every isogeny is a group homomorphism. But we will see soon that this is
always the case.

61
Proposition 9.1. For every pair of elliptic curves hom(E1 , E2 ) is a torsion-free abelian
group. Furthermore for any elliptic curve E, End(E) is a characteristic zero integral domain.
Proof Let φ ∈ hom(E1 , E2 ). Suppose [m] ○ φ = [0] for some m ≠ 0, then deg[m] ⋅ deg φ =
deg[0]. Since [m] is not constant, deg[m] ≠ 0 and so deg φ = 0 i.e. φ = [0]. This shows
that End(E) has characteristic zero. Now let φ, ψ ∈ End(E) such that φ ○ ψ = [0]. Then
deg φ ⋅ deg ψ = 0 and so φ = [0] or ψ = [0].
Remark 9.3. If char K = 0, then we have an injective homomorphism
[⋅] ∶ Z → End(E).
And commonly this is an isomorphism, i.e. End(E) ≃ Z for almost all elliptic curves over K.
Those elliptic curves where this is not the case are called complex multiplication (CM)
elliptic curves.
Example 9.1. Assume char K ≠ 2 and let i ∈ K be a primitive fourth root of unity, i.e.
i2 = −1. Consider the elliptic curve E/K given by the equation
y 2 = x3 − x.
This elliptic curve is of CM type since the map
[i](x, y) ∶= (−x, iy)
is an endomorphism fixing the point at infinity O. Note that [i] is defined over K if and only
if i ∈ K, thus even when E is defined over K, End(E) can be strictly bigger than EndK (E).
Noting that
[i] ○ [i](x, y) = [i](−x, iy) = (x, −y) = −(x, y)
we get that [i]2 = [−1] and so if char K = 0 we have an inclusion
[⋅] ∶ Z[i] ↪ End(E)
which is in fact an isomorphism End(E) ≃ Z[i]. Thus all the automorphisms are
Aut(E) = {±id, [±i]}.
Example 9.2. Let K be a field of characteritic p > 0, let q = pr , and let E/K be an
elliptic curve given by a Weierstrass equation. Recall that E (q) /K is defined by raising each
coefficient of the equation of E to the q-th power, and the Frobenius morphism is
φq ∶ E → E (q)
(x, y) ↦ (xq , y q ).
Since clearly ∆(E (q) ) = ∆(E)q , then E (q) is also an elliptic curve. In the case K = Fq is the
finite field with q elements, then the q-th power map is the identity on K (but not on K)
and so E (q) = E. Thus in this case the Frobenius map is an endomorphism of E, called the
Frobenius endomorphism. Note that the set of fixed points by φq is exactly the set of
rational points E(Fq ).

62
Definition 9.3. Let E be an elliptic curve and Q ∈ E. We define the translation by Q
map
τQ ∶ E → E
P ↦ P + Q.
This map is an isomorphim with inverse τ−Q , and is never an isogeny unless it is the identity
map.
Remark 9.4. Given any morphism

F ∶ E1 → E2

we can write
F = τF (O) ○ φ
with φ = τ−F (O) ○ F ∶ E1 → E2 and isogeny.
Theorem 9.2. Let φ ∶ E1 → E2 be an isogeny, then

φ(P + Q) = φ(P ) + φ(Q) , ∀P, Q ∈ E1 .

Proof If φ = [0] the result is clear. Otherwise φ is a finite map and so it induces a group
homomorphism
φ∗ ∶ Pic0 (E1 ) → Pic0 (E2 ).
Since we have isomorphisms
∼
→ Pic0 (E1 ),
f1 ∶ E1 Ð
P ↦ [P ] − [O]
∼
→ Pic0 (E2 ),
f2 ∶ E2 Ð
P ↦ [P ] − [O]
we just have to note that

f2 (φ(P )) = [φ(P )] − [O] = φ∗ [P ] − φ∗ [O] = φ∗ (f1 (P ))

for all P ∈ E1 . In particular

f2 (φ(P + Q)) = φ∗ (f1 (P + Q)) = φ∗ ([P + Q] − [O])

= φ∗ ([P ] − [O]) + φ∗ ([Q] − [O]) + φ∗ [O]

= φ∗ (f1 (P )) + φ∗ (f1 (Q)) + [O]
= f2 (φ(P )) + f2 (φ(Q)) + [O]
= [φ(P )] − [O] + [φ(Q)] − [O] + [O]
= f2 (φ(P ) + φ(Q)).
Since f2 is a bijection we conclude the theorem.

63
Corollary 9.3. Let φ ∶ E1 → E2 be a non-zero isogeny, then

ker φ = φ−1 (O)

is a finite subgroup of E1 .

Proof Since φ is a group homomorphism, ker φ is a subgroup, and since #φ−1 (O) ≤ deg φ
it is finite.

Theorem 9.4. Let φ ∶ E1 → E2 be a non-zero isogeny.

(a) For every Q ∈ E2 ,

#φ−1 (Q) = degs φ.
Further, for every P ∈ E1 ,
eφ (P ) = degi φ.

(b) The map

ker φ → Aut(K(E1 )/φ∗ K(E2 ))
T ↦ τT∗
is an isomorphism.

(c) Suppose that φ is separable. Then φ is unramified,

# ker φ = deg φ,

and K(E1 ) is a Galois extension of φ∗ K(E2 ).

Proof (a) We know by Proposition 3.6 that for almost all Q ∈ E2

#φ−1 (Q) = degs φ.

Let Q′ ∈ E2 be any point, then choose some R ∈ E1 such that φ(R) = Q′ − Q. Since φ is a
homomorphism, there is a 1 to 1 correspondence

φ−1 (Q) ←→ φ−1 (Q′ )

P ↦ P + R.
Thus all fiber have the same size equal to degs φ. Also by Proposition 3.6 it follows that

degs φ ⋅ eφ (P ) = ∑ eφ (P ) = deg φ = degs φ ⋅ degi φ.

P ∈φ−1 (Q)

(b) If T ∈ ker φ and f ∈ K(E2 ) then

τT∗ (φ∗ (f )) = (φ ○ τT )∗ (f ) = φ∗ (f )

64
since φ(P + T ) = φ(P ) for all P ∈ E1 . This shows that τT∗ as an automorphism of K(E1 )
fixes φ∗ K(E2 ). Since
τT∗+T ′ = (τT ′ ○ τT )∗ = τT∗ ○ τT∗′
the map is a group homomorphism. From (a) we see that

ker φ = degs φ ≥ #Aut(K(E1 )/φ∗ K(E2 )).

Therefore it is enough to show that the map is injective to conclude the isomorphism. Let
T ∈ ker φ, then τT∗ fixes K(E1 ). This means that every rational function on E1 is T -periodic.
Considering the rational function x (corresponding to the x-coordinate) we see that it has
only one pole at O, thus it cannot be T -periodic unless T = O.

(c) If φ is separable, then degi φ = 1 and so by (a) φ is unramified and #φ−1 (Q) = deg φ for
all Q ∈ E2 . In particular
# ker φ = #φ−1 (O) = deg φ
and by (b) we conclude

#Aut(K(E1 )/φ∗ K(E2 )) = deg φ = [K(E1 ) ∶ φ∗ K(E2 )]

thus K(E1 )/φ∗ K(E2 ) is Galois.

Corollary 9.5. Let φ ∶ E1 → E2 and ψ ∶ E1 → E3 be two isogenies and assume φ is separable.
If
ker φ ⊆ ker ψ,
then there exists a unique isogeny
λ ∶ E2 → E3
such that ψ = λ ○ φ.
Proof Since φ is separable K(E1 )/φ∗ K(E2 ) is Galois. The inclusion ker φ ⊆ ker ψ implies
that ψ ∗ K(E3 ) is fixed by Gal(K(E1 )/φ∗ K(E2 )) and so

ψ ∗ K(E3 ) ⊆ φ∗ K(E2 ) ⊆ K(E1 ).

Since for smooth curves having extensions of function fields is equivalent to have finite
morphisms of the corresponding curve, this implies that there exists a unique morphism

λ ∶ E2 → E3

such that λ ○ φ = ψ. In consequence λ(O) = λ(φ(O)) = ψ(O) = O and so λ is an isogeny.

Proposition 9.6. Let E be an elliptic curve and let Φ be a finite subgroup of E(K). There
are a unique elliptic curve E ′ and a separable isogeny

φ ∶ E → E′

such that ker φ = Φ.

65
Proof Let
K(E)Φ ∶= {f ∈ K(E) ∶ τT∗ (f ) = f ∀T ∈ Φ}.
Then K(E)/K(E)Φ is a finite Galois extension with Galois group Φ. It follows that K(E)Φ
has transcendence degree 1 over K, and so there is a unique smooth curve C defined over K
and a finite morphism φ ∶ E → C such that

φ∗ K(C) = K(E)Φ .

Let us show that φ is unramified. Let P ∈ E and T ∈ Φ. Then for every f ∈ K(C),

f (φ(P + T )) = (τT∗ (φ∗ (f )))(P ) = (φ∗ (f ))(P ) = f (φ(P )).

Therefore
φ(P + T ) = φ(P ) , ∀P ∈ E, ∀T ∈ Φ.
Now let Q ∈ C and let P ∈ E be any point such that φ(P ) = Q, then

{P + T ∶ T ∈ Φ} ⊆ φ−1 (Q).

On the other hand, since

#φ−1 (Q) ≤ degs φ ≤ deg φ = #Φ
it follows that the above inequality is an equality and so φ is separable and unramified over
Q. Finally applying Hurwitz formula we get

2 − 2 = deg φ(2g(C) − 2),

and since deg φ ≠ 0 it follows that g(C) = 1 and so C is an elliptic curve and φ is the desired
isogeny if we let φ(O) to be the fixed zero point of C.

9.2 The invariant differential

Definition 9.4. Let E/K be an elliptic curve given by a Weierstrass equation

y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 .

The invariant differential is

dx
ω= ∈ ΩE .
2y + a1 x + a3

It has no zero nor poles, i.e. div(ω) = 0.

Proposition 9.7. The invariant differential is invariant under translations, i.e.

τQ∗ ω = ω , ∀Q ∈ E.

66
Proof Since ΩE is a one dimensional K(E)-vector space, we know

τQ∗ ω = aQ ⋅ ω,

for some aQ ∈ K(E). Since div(ω) = 0, then div(τQ∗ ω) = τQ∗ (div(ω)) = 0, thus

0 = div(τQ∗ ω) = div(aQ ⋅ ω) = div(aQ ) + div(ω) = div(aQ ).

×
Therefore aQ ∈ K . Considering an explicit computation of aQ via the equality

dx(P + Q) dx(P )
= aQ ⋅
2y(P + Q) + a1 x(P + Q) + a3 2y(P ) + a1 x(P ) + a3

we see that aQ is a rational function on the coefficients of Q, and so the function f (Q) ∶= aQ
is a rational function, i.e. f ∈ K(E). Since as a map f ∶ E → P1 it is not surjective (since it
never reaches (0 ∶ 1) nor (1 ∶ 0)), we conclude it is constant, thus

aQ = aO = 1.

Theorem 9.8. Let E1 and E2 e two elliptic curves, let φ, ψ ∶ E1 → E2 be two isogenies and
let ω be the invariant differential of E2 . Then

(φ + ψ)∗ ω = φ∗ ω + ψ ∗ ω.

Proof The result is clear if some of the isogenies is zero. If φ + ψ = [0], then

φ∗ = (−φ)∗ = φ∗ ○ [−1]∗ .

Thus it suffices to check that [−1]∗ ω = −ω. And this is clear since

dx(−P ) dx(P ) dx(P )

= =− .
2y(−P ) + a1 x(−P ) + a3 2(−y(P ) − a1 x(P ) − a3 ) + a1 x(P ) + a3 2y(P ) + a1 x(P ) + a3

Let us assume now that φ, ψ and φ + ψ are non-zero. Consider the maps

π1 ∶ E × E → E

(P, Q) ↦ P
π2 ∶ E × E → E
(P, Q) ↦ Q
g ∶E ×E →E
(P, Q) ↦ P + Q.

67
We claim that
g ∗ ω = π1∗ ω + π2∗ ω.
Using this fact we see that

(φ + ψ)∗ ω = (g ○ (φ, ψ))∗ ω = (φ, ψ)∗ (g ∗ ω) = (φ, ψ)∗ (π1∗ ω + π2∗ ω) = φ∗ ω + ψ ∗ ω.

In order to prove our claim, note that by tying an explicit computation of g ∗ ω we see

g ∗ ω = f1 ⋅ π1∗ ω + f2 ⋅ π2∗ ω,

for some f1 , f2 ∈ K(E ×E). Restricting this equality to the curve E ×{Q} for any fixed Q ∈ E
we get that
ω = τQ∗ ω = g ∗ ω∣E×{Q} = f1 (⋅, Q)ω
and so f1 (P, Q) = 1 for any (P, Q) ∈ E × E, i.e. f1 = 1. Analogously f2 = 1 and we are done.

Corollary 9.9. Let ω be an invariant differential of an elliptic curve E, and let m ∈ Z, then

[m]∗ ω = mω.

Corollary 9.10. Let E/K be an elliptic curve and let m ∈ Z such that m ≠ 0 in K. Then
the multiplication by m map is a finite separable endomorphism.
Corollary 9.11. Let E be an elliptic curve defined over a finite field Fq of characteristic p,
let φ ∶ E → E be the q-th Frobenius endomorphism and let m, n ∈ Z. Then the map

m + nφ ∶ E → E

is separable if and only if p ∤ m. In particular 1 − φ is separable.

Proof We use the fact that ψ = m + nφ is separable if and only if ψ ∗ ω ≠ 0. Since φ is purely
inseparable we know φ∗ ω = 0, this can also be computed directly
dx d(xq ) xq−1 dx
φ∗ ( )= q = q = 0.
2y + a1 x + a3 2y + a1 xq + a3 2y q + a1 xq + a3
Therefore
ψ ∗ ω = [m]∗ ω + [n]∗ (φ∗ ω) = mω
which is non-zero if and only if p ∤ m.
Corollary 9.12. Let E/K be an elliptic curve and let ω be a non-zero invariant differential.
Consider the map
End(E) → K
φ ↦ aφ
where φ∗ ω =∶ aφ ω.

68
 (a) The map φ ↦ aφ is a ring homomorphism.
(b) The kernel of this map corresponds to the inseparable endomorphisms.
(c) If char K = 0, then End(E) is commutative.
Proof (a) We know φ∗ ω = aφ ⋅ ω for some aφ ∈ K(E), and since div(φ∗ ω) = φ∗ (div(ω)) = 0
×
it follows that div(aφ ) = 0 and so aφ ∈ K . Furthermore

aφ+ψ ω = (φ + ψ)∗ ω = φ∗ ω + ψ ∗ ω = aφ ω + aψ ω = (aφ + aψ )ω.

And also
aφ○ψ ω = ψ ∗ φ∗ ω = aψ aφ ω.
(b) We have aφ = 0 if and only if φ∗ ω = 0 which is equivalent to φ be inseparable.

(c) If char K = 0, then all morphisms are separable and so by (b) the kernel of the map
φ ↦ aφ is trivial. This induces an injection End(E) ⊆ K and so End(E) is commutative.

10 Elliptic Curves over Finite Fields

In this lecture we will see some results about elliptic curves over finite fields. Before doing
that we will see some more properties of isogenies.

10.1 The dual isogeny

Theorem 10.1. Let φ ∶ E1 → E2 be a non-constant isogeny of degree m.
(a) There exists a unique isogeny

φ̂ ∶ E2 → E1 such that φ̂ ○ φ = [m].

(b) As a group homomorphism, φ̂ equals the composition

f φ∗ g
→ Div0 (E2 ) Ð→ Div0 (E1 ) Ð
E2 Ð → E1

where f (Q) = Q − O and g(∑ki=1 ni ⋅ Pi ) = ∑ki=1 ni ⋅ Pi (where the last sum is in the group
of the elliptic curve E1 ).
Proof (a) For the uniqueness note that if φ̂′ satisfies φ̂′ ○ φ = [m] then

(φ̂ − φ̂′ ) ○ φ = 0 and so φ̂ − φ̂′ = 0.

For the existence let us show first some particular cases. If φ is separable then since deg φ = m,
it follows from Theorem 9.4 that
# ker φ = m,

69
and so ker φ ⊆ E[m] = ker[m]. It follows from Corollary 9.5 that there exists a unique
isogeny φ̂ ∶ E2 → E1 such that φ̂ ○ φ = [m]. Let us consider now the case where φ is the p-th
Frobenius map for some prime p. Thus deg φ = [p]. Looking at the [p] map and using the
fact char K = p, then
[p]∗ ω = pω = 0
for ω the invariant differential of E2 . Therefore [p] is not separable and so we can decompose

[p] = ψ ○ φn

for some n ≥ 1. It follows that φ̂ = ψ ○ φn−1 . Finally, for the general case, if char K = 0 then
φ is separable, and if char K = p > 0 then φ is a composition of a separable isogeny and
a q-th Frobenius for q = pn , which in turn is a composition of a p-th Forobenius n times.
Therefore we just need to show that if φ ∶ E1 → E2 and ψ ∶ E2 → E3 are two isogenies of
degrees deg φ = m and deg ψ = n, and there exist φ̂ and ψ̂, then there exists ψ ̂○ φ. In fact

φ̂ ○ ψ̂ ○ (ψ ○ φ) = φ̂ ○ [n] ○ φ = [n]φ̂ ○ φ = [n] ○ [m] = [nm]

and so
̂
ψ ○ φ = φ̂ ○ ψ̂.
(b) Let Q ∈ E2 , then

g(φ∗ (f (Q))) = g(φ∗ (Q − O)) = ∑ eφ (P ) ⋅ P − ∑ eφ (T ) ⋅ T.

P ∈φ−1 (Q) T ∈φ−1 (O)

By Theorem 9.4 we know eφ (P ) = degi φ for all P ∈ E1 and so

⎛ ⎞
g(φ∗ (f (Q))) = degi φ ∑ P− ∑ T .
⎝P ∈φ−1 (Q) T ∈φ−1 (O) ⎠

Fixing any P ∈ φ−1 (Q) we see that φ−1 (Q) = P + φ−1 (O) and so since #φ−1 (O) = degs φ then

⎛ ⎞
g(φ∗ (f (Q))) = degi φ ∑ (P + T ) − ∑ T = degi φ ⋅ degs φ ⋅ P = deg φ ⋅ P.
⎝T ∈φ−1 (O) T ∈φ−1 (O) ⎠

On the other hand

φ̂(Q) = φ̂(φ(P )) = deg φ ⋅ P.

Definition 10.1. For any isogeny φ ∶ E1 → E2 we say that φ̂ ∶ E2 → E1 is the dual isogeny.
ˆ ∶= [0].
We set [0]

Theorem 10.2. Let φ ∶ E1 → E2 be an isogeny.

(a) Let m = deg φ, then φ̂ ○ φ = [m] on E1 and φ ○ φ̂ = [m] on E2 .

70
 ̂
(b) Let λ ∶ E2 → E3 be another isogeny, then λ ○ φ = φ̂ ○ λ̂.
(c) Let ψ ∶ E1 → E2 be another isogeny, then φ̂
+ ψ = φ̂ + ψ̂.
̂ = [m] and so deg[m] = m2 .
(d) For any m ∈ Z, [m]
(e) deg φ̂ = deg φ.
ˆ
(f) φ̂ = φ.
Proof (a) If φ = [0] the result is trivial. If it is not constant, then
(φ ○ φ̂) ○ φ = φ ○ (φ̂ ○ φ) = φ ○ [m] = [m] ○ φ
implies φ ○ φ̂ = [m]. We already proved item (b) in the proof of the previous theorem. The
̂ = [1] and [−1]
first part of (d) follows directly from (c) after noting that [1] ̂ = [−1]. The
second part is a consequence
̂ ○ [m] = [m] ○ [m] = [m2 ].
[deg[m]] = [m]
For (e) note that by (a) deg φ̂ ⋅ deg φ = deg[m] = m2 and so deg φ̂ = m = deg φ. (f) Follows
from (a) and (e). Thus we are left to prove (c). Writing explicitly the Weierstrass equations
of E1 and E2
E1 = {y12 + a1 x1 y1 + a3 y1 = x31 + a2 x21 + a4 x1 + a6 },
E2 = {y22 + b1 x2 y2 + b3 y2 = x32 + b2 x22 + b4 x2 + b6 },
we have
φ = (φ1 (x1 , y1 ), φ2 (x1 , y1 )).
Looking at φ1 (x1 , y1 ), φ2 (x1 , y1 ) ∈ K(E1 ) = K(x1 , y1 ) we see they satisfy the Weierstrass
equation of E2 and so
φ(x1 , y1 ) = (φ1 (x1 , y1 ), φ2 (x1 , y1 )) ∈ E2 (K(x1 , y1 )).
The same holds for ψ and φ + ψ. Consider the divisor
0
D = (φ + ψ)(x1 , y1 ) − φ(x1 , y1 ) − ψ(x1 , y1 ) + O ∈ DivK(x 1 ,y1 )
(E2 ).

Since this divisor has sum O in the group law of E2 , it follows that it is zero in the Picard
group, i.e. it is a principal divisor. Thus there exists some
g(x1 , y1 , x2 , y2 )
f (x1 , y1 , x2 , y2 ) = ∈ K(x1 , x2 )(E2 ) = K(x1 , y1 , x2 , y2 )
h(x1 , y1 , x2 , y2 )
such that D = div(f ), where f is seen as a function of (x2 , y2 ) with coefficients in K(E1 ) =
K(x1 , y1 ) and g, h ∈ K[x1 , y1 , x2 , y2 ]. If we look now at f as a function of (x1 , y1 ) with coeffi-
cients in K(E2 ) = K(x2 , y2 ) it follows that g vanishes at the points (x1 , y1 ) = (f1 (x2 , y2 ), f2 (x2 , g2 ))
such that
g(f1 (x2 , y2 ), f2 (x2 , y2 ), x2 , y2 ) = 0 ∈ K(x2 , y2 ).

71
Since as a function of x1 and y1 , the only zero of g is of the form

(x1 , y1 , (φ + ψ)(x1 , y1 ))

we see that
(φ + ψ)(f1 (x2 , y2 ), f2 (x2 , y2 )) = (x2 , y2 ).
Certainly such f1 , f2 ∈ K(x2 , y2 ) always exist (they are not necessarily unique, but they are
a finite amount). We see that as a function of (x1 , y1 )

div(f ) = (φ + ψ)∗ ((x2 , y2 )) − φ∗ ((x2 , y2 )) − ψ ∗ ((x2 , y2 )) + nO ∈ Div0 (E1 )

K(x2 ,y2 )

where n is the order f at the point at infinity. Therefore

+ ψ(x2 , y2 ) − φ̂(x2 , y2 ) − ψ̂(x2 , y2 ) = −(φ + ψ)∗ O + φ∗ O + ψ ∗ O − nO ∈ E1 (K(x2 , y2 ))

φ̂

and so it is independent of x2 and y2 , i.e.

φ̂
+ ψ(x2 , y2 ) − φ̂(x2 , y2 ) − ψ̂(x2 , y2 ) ∈ E1 (K)

is constant for any value of (x2 , y2 ). Putting (x2 , y2 ) = O we conclude that

φ̂
+ ψ(x2 , y2 ) − φ̂(x2 , y2 ) − ψ̂(x2 , y2 ) = O

for all (x2 , y2 ) and so

φ̂
+ ψ = φ̂ + ψ̂.

Definition 10.2. Let A be an abelian group. A function

d∶A→R

is a quadratic form if it satisfies the following conditions:

(i) d(a) = d(−a) for all a ∈ A.

(ii) The pairing

A×A→R
(a, b) ↦ d(a + b) − d(a) − d(b),
is bilinear.

A quadratic form d is positive definite if it further satisfies:

(iii) d(a) ≥ 0 for all a ∈ A.

(iv) d(a) = 0 if and only if a = 0.

72
Remark 10.1. Every positive definite quadratic form d induces a inner product
d(a + b) − d(a) − d(b)
⟨a, b⟩ ∶=
2
which satisfies the Cauchy-Schwarz inequality
√
∣⟨a, b⟩∣ ≤ d(a)d(b).

Corollary 10.3. Let E1 and E2 be elliptic curves. The degree map

deg ∶ hom(E1 , E2 ) → Z

is a positive definite quadratic form.

Proof The non-trivial statement is the bilinearity of

2⟨φ, ψ⟩ = deg(φ + ψ) − deg φ − deg ψ.

But this follows noting that

[⟨φ, ψ⟩] = (φ̂

+ ψ) ○ (φ + ψ) − φ̂ ○ φ − ψ̂ ○ ψ = φ̂ ○ ψ + ψ̂ ○ φ.

Corollary 10.4. Let E be an elliptic curve and let m ∈ Z with m ≠ 0.

(a) If m ≠ 0 in K, i.e. if either char K = 0 or char K = p > 0 and p ∤ m, then

E[m] ≃ Z/mZ ⊕ Z/mZ.

(b) If char K = p > 0 then one of the following is true

(i) E[pn ] = {O} for all n ≥ 1.

(ii) E[pn ] = Z/pn Z for all n ≥ 1.

Proof (a) Our assumption means that [m] is separable and so

#E[m] = deg[m] = m2 .

In particular for every d∣m

#E[d] = d2 .
Writing m = pα1 1 ⋯pαk k we have that

E[pαi i ] ≃ Z/pαi i Z ⊕ Z/pαi i Z

α
and since E[pαi i ] ∩ E[pj j ] = {O} for i ≠ j. It follows that

Z/mZ ⊕ Z/mZ ≃ ⊕ki=1 E[pαi i ] ⊆ E[m].

73
Since both sides have the same size, they must be equal.
(b) Let φ be the p-th Frobenius morphism, then
#E[pn ] = # ker[pn ] = degs [pn ] = (degs (φ̂ ○ φ))n = (degs φ̂)n .
Since deg φ̂ = deg φ = p we have two possibilities. If degs φ̂ = 1, then
E[pn ] = {O} for all n ≥ 1.
Otherwise, degs φ̂ = p and so
#E[pn ] = pn for all n ≥ 1 and so E[pn ] ≃ Z/pn Z.

10.2 The number of Fq -points of an elliptic curve

Now we will see some results concerning the number of rational points of elliptic curves
defined over finite fields.
Let E/Fq be an elliptic curve defined over the finite field Fq , we want to estimate #E(Fq ).
Writing a Weierstrass equation for E
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , a i ∈ Fq ,
we see that for each value of x ∈ Fq we have at most 2 values of y ∈ Fq satisfying the above
equation. Thus (counting the point at infinity) we get
#E(Fq ) ≤ 2q + 1.
However, since a “randomly chosen" quadratic equation over Fq has 21 chances of being
solvable, we expect that the correct order of magnitude should be q instead of 2q.
Theorem 10.5 (Hasse). Let E/Fq be an elliptic curve defined over a finite field. Then
√
∣#E(Fq ) − q − 1∣ ≤ 2 q.
Proof Let φ ∶ E → E be the q-th Frobenius endomorphism. Since
P ∈ E(Fq ) if and only if φ(P ) = P,
then
E(Fq ) = ker(1 − φ).
By Corollary 9.11 the map 1 − φ is separable and so
#E(Fq ) = deg(1 − φ).
Since deg is a positive definite quadratic form, it follows by Cauchy-Schwarz inequality that
√
2∣⟨[1], φ⟩∣ = ∣#E(Fq ) − 1 − q∣ ≤ 2 q.

74
Remark 10.2. In general for any algebraic variety V /Fq we can ask ourselves about the
number of rational points #V (Fq ). In order to study these numbers Andre Weil introduced a
generating function, the so called zeta function, and made a series of remarkable conjectures
the so called Weil conjectures. These conjectures are nowadays theorems proved by Weil,
Dwork, Artin, Grothendieck and Deligne. Let us prove these conjectures in the case of
elliptic curves.

Definition 10.3. Let V /Fq be an algebraic variety defined over Fq . The zeta function of
V /Fq is the power series
∞
Tn
Z(V /Fq ; T ) ∶= exp ( ∑ #V (Fqn ) ⋅ ).
n=1 n

Where for any power series F (T ) ∈ Q[[T ]] with no constant term

∞
(F (T ))n
exp(F (T )) = ∑ ∈ Q[[T ]].
n=0 n!

Remark 10.3. Note that we can recover

1 dn
#V (Fqn ) = log Z(V /Fq ; T )∣T =0 .
(n − 1)! dT n

Example 10.1. Let V = Pm . Since #Am+1 (Fqn ) = (q n )m+1 and #F×qn = q n − 1, it follows that

q n(m+1) − 1
#V (Fqn ) = = 1 + q n + q 2n + ⋯ + q mn
qn − 1
and so ∞ m m
Tn
log Z(V /Fq ; T ) = ∑ ∑ q nk = − ∑ log(1 − q k T ).
n=1 k=0 n k=0

Thus
1
Z(V /Fq ; T ) = .
(1 − T )(1 − qT )⋯(1 − q m T )
Theorem 10.6 (Weil conjectures). Let V /Fq be a smooth projective variety of dimension
n.

(a) Then Z(V /Fq ; T ) ∈ Q(T ).

(b) There is an integer e (called the Euler characteristic of V ) such that the following
functional equation holds
1 ne
Z (V /Fq ; ) = ±q 2 T e Z(V /Fq ; T ).
qnT

75
 (c) The zeta function factors as
P1 (T )⋯P2n−1 (T )
Z(V /Fq ; T ) =
P0 (T ) ⋅ P2 (T )⋯P2n (T )
where each Pi (T ) ∈ Z[T ], with
P0 (T ) = 1 − T and P2n (T ) = 1 − q n T,
and such that for every 0 ≤ i ≤ 2n, the polynomial Pi (T ) factors over C as
bi
1
Pi (T ) = ∏(1 − αij T ) with ∣αij ∣ = q 2 .
j=1

Where bi is the i-th Betti number of V .

Remark 10.4. The (c) part of Weil conjectures is called the Riemann hypothesis. To
motivate this and the name of the zeta function we do the change of variables T = q −s . This
gives us a function ζV /Fq (s) with zeros on the points such that
αij q −s = 1
1
and so ∣q s ∣ = q 2 , i.e. Re(s) = 12 . In the case V = E is an elliptic curve, the functional equation
becomes
ζE/Fq (s) = ζE/Fq (1 − s),
as the one of the usual Riemann zeta function. Before proving the Weil conjectures we have
to introduce some preliminary results.
Definition 10.4. Let E be an elliptic curve and let ` ∈ Z be a prime number. The (`-adic)
Tate module of E is the group
T` (E) = lim E[`n ],
←Ð
where the inverse limit is taken with respect the maps
[`] ∶ E[`n+1 ] → E[`n ].
Since each E[`n ] is a Z/`n Z-module, it follows that T` (E) is a Z` -module.
Remark 10.5. To see that the ring of p-adic integers is the inverse limit of the system
Z/`n+1 Z → Z/`n Z
a ↦ a (mod `n ).
Note that the cone map from Z` to the directed system is given by
Z` → Z/`n+1 Z
∞ n
∑ ai ⋅ `i ↦ ∑ ai ⋅ `i ,
i=0 i=0
where ai ∈ {0, 1, . . . , ` − 1}.

76
Proposition 10.7. As a Z` -module the Tate module has the following structure
(a) T` (E) ≃ Z` ⊕ Z` if ` ≠ char K.
(b) T` (E) ≃ 0 or Z` if ` = char K.
Proof It is direct from Corollary 10.4.
Remark 10.6. Every isogeny φ ∶ E1 → E2 induces a Z` -linear map of Tate modules

φ` ∶ T` (E1 ) → T` (E2 ).

This map is very useful to study isogenies. We are mainly interested in the following property
of this map.
Proposition 10.8. Let φ ∈ End(E), then

det(φ` ) = deg φ and tr(φ` ) = 1 + deg φ − deg(1 − φ).

In particular both are in Z and do not depend on the choice of `.

Proof See [Sil09, Chapter III, Proposition 8.6]
Theorem 10.9. Let E/Fq be an elliptic curve, φ be the q-th Frobenius and

a = q + 1 − #E(Fq ).

If α, β ∈ C are the roots of T 2 − aT + q, then they are complex conjugated and for every n ≥ 1

#E(Fqn ) = q n + 1 − αn − β n .

Proof Since

det(T − φ` ) = T 2 − tr(φ` )T + det(φ` ) = T 2 − (1 + q − #E(Fq ))T + deg φ = T 2 − aT + q

we see that α and β are the eigenvalues of φ` . The characteristic polynomial T 2 − aT + q in

non-negative since for every m
n ∈Q

m det(m − nφ` ) deg(m − nφ)

det ( − φ` ) = = ≥ 0.
n n2 n2
Thus it has two complex conjugated roots or a double real root. In any case ᾱ = β. Similarly
for n ≥ 1 the eigenvalues of φn` are αn and β n and so

det(T − φn` ) = (T − αn )(T − β n ).

Therefore

#E(Fqn ) = deg(1 − φn ) = det(1 − φn` ) = 1 − tr(φn` ) + det(φn` ) = 1 − αn − β n + q n .

77
Proof of the Weill conjectures for elliptic curves We just compute
∞
#E(Fqn )T n ∞ (1 + q n − αn − β n )T n
log Z(E/Fq ; T ) = ∑ =∑
n=1 n n=1 n

= − log(1 − T ) − log(1 − qT ) + log(1 − αT ) + log(1 − βT )

and so
(1 − αT )(1 − βT )
Z(E/Fq ; T ) =
(1 − T )(1 − qT )
√
with ∣α∣ = ∣β∣ = q, since they are conjugated and αβ = q. It is elementary to check that it
verifies the functional equation.

11 Methods for Computing the Rank

The computation of the rank of an elliptic curve is an unsolved problem nowadays and an
active area of research. In this lecture we will explain some methods used to compute it for
elliptic curves defined over Q.

11.1 Linear independence of rational points

Theorem 11.1. Let E/Q be an elliptic curve, and hx be the logarithmic height associated
to some Weierstrass equation of E. Then there exists some constant C > 0 such that for all
P, Q ∈ E(Q)
∣hx (P + Q) + hx (P − Q) − 2hx (P ) − 2hx (Q)∣ ≤ C.

Proof See [Sil09, Chapter VIII, Theorem 6.2].

Corollary 11.2. Let E/Q be an elliptic curve given by a Weierstrass equation, and let hx
be the corresponding logarithmic height on E(Q). Then there exists some C > 0 such that

∣hx (2 ⋅ P ) − 4hx (P )∣ ≤ C , ∀P ∈ E(Q).

Proof Apply Theorem 11.1 to P = Q.

Definition 11.1. Let E/Q be an elliptic curve and let hx be the logarithmic height with
respect to some Weierstrass equation of E. The canonical height is

1 hx (2n ⋅ P )
ĥ(P ) ∶= lim , P ∈ E(Q).
2 n→∞ 4n
Proposition 11.3. The height always converges to some non-negative real number.

78
Proof We show that the sequence is Cauchy. Take n ≥ m ≥ 0, then by Corollary 11.2 we
have
n−1
∣4−n hx (2n ⋅ P ) − 4−m hx (2m ⋅ P )∣ = ∣ ∑ 4−i−1 hx (2i+1 ⋅ P ) − 4−i hx (2i ⋅ P )∣
i=m
n−1
≤ ∑ 4−i−1 ∣hx (2i+1 ⋅ P ) − 4hx (2i ⋅ P )∣ ≤ 4−m C.
i=m

Theorem 11.4 (Néron-Tate). Let E/Q be an elliptic curve.

(a) For all P, Q ∈ E(Q) we have

ĥ(P + Q) + ĥ(P − Q) = 2ĥ(P ) + 2ĥ(Q).

(b) For all P ∈ E(Q) and all m ∈ Z,

ĥ(m ⋅ P ) = m2 ĥ(P ).

(c) The canonical height is a quadratic form, i.e. ĥ is even and the Néron-Tate pairing

⟨⋅, ⋅⟩ ∶ E(Q) × E(Q) → R

1
⟨R, Q⟩ ∶= (ĥ(P + Q) − ĥ(P ) − ĥ(Q))
2
is bilinear.

(d) Let P ∈ E(Q). Then ĥ(P ) ≥ 0 and

ĥ(P ) = 0 if and only if P is a torsion point.

(e) There exists some constant C > 0 such that

∣2ĥ(P ) − hx (P )∣ ≤ C ∀P ∈ E(Q).

Proof (a) By Theorem 11.1 there exists C > 0 such that

∣hx (2n ⋅ (P + Q)) − hx (2n ⋅ (P − Q)) − 2hx (2n ⋅ P ) − 2hx (2n ⋅ Q)∣ ≤ C , ∀n ≥ 0.

Dividing by 2 ⋅ 4n and taking n → ∞ we conclude the result.

(b) Since ĥ(−P ) = ĥ(P ) is even (because hx is also even), we can assume m ∈ Z is non-
negative. Furthermore the equality is trivial for m = 0, 1. By induction and item (a) we
have

ĥ(m ⋅ P ) = 2ĥ((m − 1) ⋅ P ) + 2ĥ(P ) − ĥ((m − 2) ⋅ P ) = (2(m − 1)2 + 2 − (m − 2)2 )ĥ(P ) = m2 ĥ(P ).

79
(c) This is a standard fact which follows from the parallelogram law obtained in item (a).

(e) It follows from the proof of Proposition 11.3 that there exists some C > 0 such that for
all n ≥ m ≥ 0
∣4−n hx (2n ⋅ P ) − 4−m hx (2m ⋅ P )∣ ≤ 4−m C.
Taking m = 0 and n → ∞ we get the desired result.

(d) Since hx (P ) ≥ 0 it follows that ĥ(P ) ≥ 0. If P is a torsion point then {2n ⋅ P ∶ n ≥ 0}

is finite and so 4−n hx (2n ⋅ P ) → 0. Conversely if ĥ(P ) = 0, then by (b) ĥ(m ⋅ P ) = 0 for all
m ∈ Z. By (e) we get
hx (m ⋅ P ) = ∣2ĥ(m ⋅ P ) − hx (m ⋅ P )∣ ≤ C
for all m ∈ Z. And so the set {m ⋅ P ∶ m ∈ Z} is finite, i.e. P is a torsion point.
Remark 11.1. In spite the canonical height cannot be exactly computed, item (e) allows us
to control the approximation error of replacing ĥ(P ) by 12 hx (24n ⋅P ) . This control of the error
n

is what allows us to work correctly with canonical heights computationally. In particular we

can effectively decide whether a rational point, or a combination of rational points, is not a
torsion point using item (d).
Definition 11.2. Let E/Q be an elliptic curve. We say that P1 , . . . , Pk ∈ E(Q) are linearly
dependent over Z if there exist n1 , . . . , nk ∈ Z such that
n1 ⋅ P1 + ⋯ + nk ⋅ Pk ∈ E(Q)tors .
Otherwise we say they are linearly independent over Z. The elliptic height matrix
associated to P1 , . . . , Pk is the matrix given by all Néron-Tate pairings

⎛⟨P1 , P1 ⟩ ⋯ ⟨P1 , Pk ⟩⎞
H=⎜ ⋮ ⋱ ⋮ ⎟.
⎝⟨Pk , P1 ⟩ ⋯ ⟨Pk , Pk ⟩⎠

The determinant of H is called the elliptic regulator of P1 , . . . , Pk .

Example 11.1. Let E/Q be the elliptic curve given by
y 2 = x3 + 2308x2 + 665858x.
And consider the rational points
332929 −215405063
P = (−1681, 25543) , Q = (−338, 26) , R = ( , ).
36 216
If we ask ourselves about the linear dependence of these three points we can compute (an
approximate value of) the elliptic regulator of them. The elliptic height matrix is
7.397 . . . −3.601 . . . 3.795 . . .⎞
1⎛
H = ⎜−3.601 . . . 6.263 . . . 2.661 . . .⎟
2⎝
3.795 . . . 2.661 . . . 6.457 . . .⎠

80
whose determinant is very close to 0 (3.368 ⋅ 10−27 approx). Thus we suspect there exists
some relation between P , Q and R. Computing the kernel of H, we get that (1, 1, −1) is
approximately in the kernel, and so P + Q − R should be a torsion point. In fact

P + Q − R = (0, 0) ∈ E[2],

and so they are linearly dependent over Z. On the other hand if we consider
577 332929
S=( , )
16 64
and let H′ be the elliptic height matrix of P , Q and S, we get that det(H′ ) = 101.87727 . . .
and therefore they are linearly independent over Z.

11.2 The effective descent method

In order to compute effectively the rank of an elliptic curve we need some method to bound
the rank, and to find generators. Following the proof of Mordell-Weil theorem it is enough to
bound the rank of the Z/2Z-module E(Q)/2E(Q). This can be done following the effective
method of descent, which we illustrate here.

We will focus on the case of an elliptic curve E/Q with 4 distinct rational points of
2-torsion. These correspond to the elliptic curves of the form

y 2 = (x − e1 )(x − e2 )(x − e3 ), e1 , e2 , e3 ∈ Z different.

If we want to find a solution of the above equation (x0 , y0 ) ∈ E(Q), i.e.

y02 = (x0 − e1 )(x0 − e2 )(x0 − e3 ),

then each term x0 − ei must be almost a square, in other words there are square-free integers
a, b, c ∈ Z (assuming y0 ≠ 0) such that

x0 − e1 = au2 , x0 − e2 = bv 2 , x0 − e3 = cw2 ,
√
for some u, v, w ∈ Q. Therefore abc is a perfect square and y0 = ±uvw abc.
Example 11.2. Let E/Q be given by

y 2 = x3 − 556x + 3120 = (x − 6)(x − 20)(x + 26).

Consider the rational points

−247 −5733
P = (−8, 84) , Q = (24, 60) , S = P + Q = ( , ).
16 64
Note that if we compute the above square-free integers for each point we get

x(P ) − 6 = −14 , x(P ) − 20 = −7 ⋅ 22 , x(P ) + 26 = 2 ⋅ 32 ,

81
 x(Q) − 6 = 2 ⋅ 32 , x(Q) − 20 = 1 ⋅ 22 , x(Q) + 26 = 2 ⋅ 52 ,
7 2 9 2 13 2
x(S) − 6 = −7 ⋅ ( ) , x(S) − 20 = −7 ⋅ ( ) , x(S) + 26 = 1 ⋅ ( ) .
4 4 4
Thus, we see that
aP ⋅ aQ = −7 ⋅ 22 = aS ⋅ 22 ,
bP ⋅ bQ = −7 = bS ,
cP ⋅ cQ = 22 = cS ⋅ 22 .
This suggests that modulo square factors, the numbers a, b and c correspond to group
homomorphisms from E(Q) to Q× /(Q× )2 .

Theorem 11.5. The map

Q× Q× Q×
δ ∶ E(Q) → × ×
(Q× )2 (Q× )2 (Q× )2

given by
⎧
⎪ (x0 − e1 , x0 − e2 , x0 − e3 ) if y0 ≠ 0,
⎪
⎪
⎪
⎪
⎪
⎪ ((e1 − e2 )(e1 − e3 ), e1 − e2 , e1 − e3 ) if x0 = e1 ,
⎪
δ(P ) = ⎨ (e2 − e1 , (e2 − e1 )(e2 − e3 ), e2 − e3 ) if x0 = e2 ,
⎪
⎪
⎪
⎪
⎪ (e3 − e1 , e3 − e2 , (e3 − e1 )(e3 − e2 )) if x0 = e3 ,
⎪
⎪
⎩ (1, 1, 1)
⎪ if P = O,
is a group homomorphism with kernel 2E(Q).

Proof Note first that δ(P ) = δ(x0 , y0 ) = δ(x0 , −y0 ) = δ(−P ), and δ(P )2 = (1, 1, 1) and so
δ(P )−1 = δ(P ). Thus in order to show that it is a group homomorphism it is enough to show
that δ(P ) ⋅ δ(Q) ⋅ δ(R) = (1, 1, 1) for P = (x0 , y0 ), Q = (x1 , y1 ) and R = (x2 , y2 ) in the same
line. Suppose they belong to the line

L = {(x, y) ∶ y = ax + b}.

Replacing this equation in the equation of E we get

p(x) = (ax + b)2 − (x − e1 )(x − e2 )(x − e3 ) = −(x − x0 )(x − x1 )(x − x2 ).

Evaluating at ei we get

(aei + b)2 = (x0 − ei )(x1 − ei )(x2 − ei )

and so
Q×
δ(P ) ⋅ δ(Q) ⋅ δ(R) = ((ae1 + b)2 , (ae2 + b)2 , (ae3 + b)2 ) = (1, 1, 1) ∈ .
(Q× )2

82
The other cases are left to the reader. In order to determine the kernel, let Q = (x1 , y1 ) ∈ ker δ
with y1 ≠ 0 (the other cases are left to the reader), then

x1 − e1 = t21 , x1 − e2 = t22 , x1 − e3 = t23 .

Setting
t1 (e2 + e3 ) t2 (e1 + e3 ) t3 (e1 + e2 )
α ∶= + + ,
(e1 − e2 )(e1 − e3 ) (e2 − e1 )(e2 − e3 ) (e1 − e3 )(e2 − e3 )
t1 t2 t3
β ∶= + + ,
(e1 − e2 )(e1 − e3 ) (e2 − e1 )(e2 − e3 ) (e1 − e3 )(e2 − e3 )
the point P = ( −α 1
β , β ) satisfies Q = 2 ⋅ P .

Proposition 11.6. Let P ∈ E(Q) be such that δ(P ) = (a, b, c) with a, b, c ∈ Z square-free.
Then every prime divisor of abc is a prime divisor of (e1 − e2 )(e2 − e3 )(e3 − e1 ) or equivalently
a prime divisor of ∆(E) = 16(e1 − e2 )2 (e2 − e3 )2 (e3 − e1 )2 .
Proof To see the equivalence note that (e1 − e2 )(e2 − e3 )(e3 − e1 ) is always even. Let p∣abc,
suppose without loss of generality that p∣a. Let pk be the exact power of p appearing in

x0 − e1 = au2 ,

thus k must be odd. Suppose k < 0, then p∣k∣ divides exactly the denominator of x0 . Then
p∣k∣ divides exactly the denominator of x0 − e2 and x0 − e3 , therefore p3∣k∣ divides exactly the
denominator of
(x0 − e1 )(x0 − e2 )(x0 − e3 ) = y02
which is absurd. Therefore k > 0. Since y02 = abc(uvw)2 it follows that p∣bc. Suppose p∣b,
then p divides e2 − e1 = (x0 − e1 ) − (x0 − e2 ) = au2 − bv 2 .
Q×
Corollary 11.7. Let π ∶ Z ∖ {0} → (Q× )2 be the natural projection map, and let

Γ′ ∶= π({n ∈ Z ∖ {0} ∶ n is square-free and if p∣n then p∣∆}).

Then the image of δ is inside

Q×
Γ∆ ∶= {(δ1 , δ2 , δ3 ) ∈ Γ′ × Γ′ × Γ′ ∶ δ1 ⋅ δ2 ⋅ δ3 = 1 ∈ }.
(Q× )2

Remark 11.2. As a consequence of this corollary we get another proof of the Weak Mordell-
Weil theorem for E(Q)/2E(Q) where E(Q) has four point of 2-torsion.
Corollary 11.8. The rank of E(Q) is bounded by

RE ≤ 2ν(∆),

where ν(m) ∶= # prime divisors of m.

83
Proof Clearly #Γ′ = 2ν(∆)+1 then #Γ∆ = #Γ′ × Γ′ = 22ν(∆)+2 and so

#(E(Q)/2E(Q)) = 2RE +2 ≤ 22ν(∆)+2 .

Example 11.3. Consider E/Q given by

y 2 = x(x − 34)(x + 34).

The discriminant is ∆ = 212 ⋅ 176 , thus RE ≤ 4. In fact RE = 2.

Remark 11.3. In general, the bound RE ≤ 2ν(∆) is not so sharp. However the method we
followed to get this bound yields a strategy to find generators of E(Q)/2E(Q) as follows.
We have an embedding
δ ∶ E(Q)/2E(Q) ↪ Γ∆ .
So we want to determine which point of Γ∆ actually belong to the image of δ, and it is
not the image of a torsion point. If (δ1 , δ2 , δ3 ) ∈ Γ∆ is such a point, then there exists some
P = (x0 , y0 ) ∈ E(Q) such that

y02 = (x0 − e1 )(x0 − e2 )(x0 − e3 ),

x0 − e1 = δ1 u2 ,
x 0 − e 2 = δ2 v 2 ,
x0 − e3 = δ3 w2 .
If we replace the third equation in the second and fourth we get

e1 − e2 = δ2 v 2 − δ1 u2 ,

e3 − e2 = δ2 v 2 − δ3 w2 .
Recalling that δ3 = δ1 ⋅ δ2 ⋅ λ2 , performing the change of variables (u, v, w) ↦ (X, Y, Zλ ) we
obtain a system
e −e = δ2 Y 2 − δ1 X 2 ,
C(δ1 , δ2 ) ∶ { 1 2
e3 − e2 = δ2 Y 2 − δ1 δ2 Z 2 .
This space is a intersection of two conics and it may have rational point or not. If (δ1 , δ2 , δ3 )
is in the image of δ, then C(δ1 , δ2 ) must have a rational point. Conversely if (X0 , Y0 , Z0 ) ∈
C(δ1 , δ2 )(Q) then
P = (e1 + δ1 X02 , δ1 δ2 X0 Y0 Z0 ) ∈ E(Q)
such that δ(P ) = (δ1 , δ2 , δ3 ). The spaces C(δ1 , δ2 ) are the so called homogeneous spaces
and can be used to compute the Mordell-Weil group.

84
Example 11.4. Consider the elliptic curve E/Q given by

y 2 = x(x − 34)(x + 34).

Since the only prime divisors of ∆ are 2 and 17, then

Γ′ = {±1, ±2, ±17, ±34}.

Set e1 = 0, e2 = −34 and e3 = 34. Therefore, the homogeneous spaces are of the form

δ2 Y 2 − δ1 X 2 = 34,
C(δ1 , δ2 ) ∶ {
δ2 Y 2 − δ1 δ2 Z 2 = 68,

with δ1 , δ2 ∈ Γ′ . Let us analyze these 64 cases.

1. (δ1 , δ2 , δ3 ) = (1, 1, 1): This point always is in the image of δ, since it corresponds to
δ(O) = (1, 1, 1).
2. (δ1 , δ2 ) = (−1, 34), (−34, 2), (34, 17): These correspond to the three points of order 2,
T1 = (0, 0), T2 = (−34, 0) and T3 = (34, 0) = T1 + T2 respectively.
3. δ2 < 0: In any case, depending on the sign of δ1 either the first or the second equation
of C(δ1 , δ2 ) cannot have solution in R because the left hand side becomes negative.
4. (δ1 , δ2 ) = (−1, 2): The homogeneous space C(−1, 2) has the rational point (4, 3, 5) cor-
responding to the point P = (−16, −120) which is of infinite order.
5. (δ1 , δ2 ) = (1, 17), (34, 1), (−34, 34): They correspond to the points P + T1 , P + T2 and
P + T3 respectively.
6. (δ1 , δ2 ) = (−2, 2): The space C(−2, 2) has the rational point (1, 4, 3) corresponding to
Q = (−2, −48) which is also of infinite order.
7. (δ1 , δ2 ) = (2, 17), (17, 1), (−17, 34), (2, 1), (−2, 34), (−17, 2), (17, 17): They correspond to
the points Q + T1 , Q + T2 , Q + T3 , P + Q, P + Q + T1 , P + Q + T2 , P + Q + T3 respectively.
8. (δ1 , δ2 ) = (1, 2): We have to solve the system C(1, 2) which is equivalent to solve over
the integers the system
2Y 2 − X 2 = 34W 2 ,
{
Y 2 − Z 2 = 34W 2 ,
with gcd(X, Y, Z, W ) = 1. In order to see that this system has no solution, we will
analyze it modulo 8. Note that the quadratic residues modulo 8 are 0, 1, 4. If W 2 ≡ 0
(mod 8), then X 2 ≡ 2Y 2 (mod 8) and so Y 2 ≡ Z 2 ≡ 4 (mod 8) and X 2 ≡ 0 (mod 8).
Thus 2∣ gcd(X, Y, Z, W ) which is a contradiction. If W 2 ≡ 1 (mod 8), then Y 2 ≡ Z 2 + 2
(mod 8) which is impossible. Finally if W 2 ≡ 4 (mod 8), then X 2 ≡ 2Y 2 ≡ 2Z 2 (mod 8)
and so Y 2 ≡ Z 2 ≡ 4 (mod 8) and X 2 ≡ 0 (mod 8). Therefore 2∣ gcd(X, Y, Z, W ) which
is again a contradiction.

85
 9. (δ1 , δ2 ) = (−1, 17), (−34, 1), (34, 34), (−1, 1), (1, 34), (34, 2), (−34, 17), (−2, 1), (2, 34),
(17, 2), (−17, 17), (2, 2), (−2, 17), (−17, 1), (17, 34): Since all these cases are obtained as
the product of a point already in the image of δ and the point (1, 2, 2) which is not in
the image of δ, then none of these cases belong to the image of δ.
In conclusion #(E(Q)/2E(Q)) = 16 = 24 = 2RE +2 , and so RE = 2 and the free part of E(Q)
is generated by P = (−16, −120) and Q = (−2, −48).
Example 11.5. Consider now the elliptic curve E/Q given by

y 2 = x(x − 82)(x + 82).

Set e1 = 0, e2 = −82, e3 = 82. The only divisors of ∆ are 2 and 41, hence

Γ′ = {±1, ±2, ±41, ±82}.

Let us analyze the homogeneous spaces

δ2 Y 2 − δ1 X 2 = 82,
C(δ1 , δ2 ) ∶ {
δ2 Y 2 − δ1 δ2 Z 2 = 164.

1. (δ1 , δ2 ) = (1, 1), (−1, 82), (−82, 2), (82, 41): These correspond to the 2-torsion points O,
T1 = (0, 0), T2 = (−82, 0) and T3 = (82, 0).

2. δ2 < 0: As before this case has no solution in R.

3. (δ1 , δ2 ) = (1, 2): As before this case has no solution. In fact looking at the system
modulo 8 we get the same system as before.

4. (δ1 , δ2 ) = (−1, 41), (−82, 1), (82, 82): These cases are not in the image of δ since they
are products of elements already in the image by (1, 2, 2) which is not in the image.

5. (δ1 , δ2 ) = (−1, 2): The homogeneous space is

2Y 2 + X 2 = 82,
C(−1, 2) ∶ {
Y 2 + Z 2 = 82.

But it turns out that this system has solution over R and also modulo pn for all prime
p and n ≥ 1 (i.e. it has solution over Qp ). Nevertheless it still has no solution over Q
(this is hard to prove), but we cannot verify this as we did in the previous cases. This
is an example of a variety where the Local-to-Global principle fails. This is the main
difficulty that can arise when we apply the descent method, causing its failure.
Remark 11.4. The obstruction of the Local-to-Global principle in the homogeneous spaces
can be studied by means of the Shafarevich-Tate group X2 (E/Q) (see [Sil09, Chapter
X]). Studying these groups it is possible to improve the bound of the rank given in Corollary
11.8 as follows.

86
Proposition 11.9. Let E/Q be an elliptic curve given by

y 2 = x3 + Ax2 + Bx.

Then
RE ≤ ν(A2 − 4B) + ν(B) − 1.

Proof See [ALRP10, Proposition 1.1].

12 Lattices and Complex Tori

In this lecture we will study elliptic curves over C. These curves have a very rich structure
since beside of being algebraic varieties, they are also Riemann surfaces, i.e. complex man-
ifolds of dimension one. We will study the analytic structure of elliptic curves which will
turn out being complex tori.

Definition 12.1. A lattice Λ ⊆ C is an additive discrete subgroup such that

Λ ⊗ R = C.

In other words it is a subgroup of the form

Λ = ω1 Z + ω2 Z = {mw1 + nw2 ∈ C ∶ m, n ∈ Z}

for some ω1 , ω2 ∈ C linearly independent over R, i.e. Im(ω1 /ω2 ) ≠ 0. We say that ω1 and ω2
are generators of Λ (the generators are certainly not unique). The complex torus induced
by Λ is the Riemann surface C/Λ, which topologically corresponds to a torus T2 = S1 × S1 .

Remark 12.1. In spite that all complex tori are topologically the same, there are infinitely
many different complex tori modulo biholomorphism, i.e. there are infinitely many different
analytic structures on T2 .

Definition 12.2. Let Λ be a lattice. An elliptic function relative to Λ is a meromorphic

function f (z) on C such that

f (z + ω) = f (z) , ∀z ∈ C , ∀ω ∈ Λ.

We denote the field of elliptic functions by C(Λ). Note that it correspond exactly to the
field of meromorphic functions defined over the complex torus C/Λ.

Definition 12.3. Let Λ be a lattice. A fundamental parallelogram for Λ is a set of the

form
D = a + ω1 ⋅ [0, 1) + ω2 ⋅ [0, 1) = {a + t1 ω1 + t2 ω2 ∶ 0 ≤ t1 , t2 < 1},
for some a ∈ C and for ω1 and ω2 generators of Λ.

87
Proposition 12.1. Let Λ be a lattice and D be a fundamental parallelogram. Then A(D)
the area of D is independent of the choice of the fundamental parallelogram. We denote
A(Λ) ∶= A(D) and call it the area of Λ.
Proof Let D′ = a′ + ω1′ ⋅ [0, 1) + ω2′ ⋅ [0, 1) be another fundamental domain. Then since ω1′ and
ω2′ are generators of Λ, there exists a matrix M ∈ GL2 (Z) such that takes each ωi ∈ R ⊕ i ⋅ R
to ωi′ ∈ R ⊕ i ⋅ R. Furthermore its inverse also has coefficients in Z and so det(M ) = ±1.
Therefore
A(D′ ) = A(M ⋅ (D − a) + a′ ) = A(M ⋅ D) = ∣ det(M )∣ ⋅ A(D) = A(D).

Proposition 12.2. A holomorphic elliptic function is constant.

Proof Let f ∈ C(Λ) be a holomorphic elliptic function and let D be a fundamental paral-
lelogram. Since f (C) = f (D) and D is compact, f is bounded. Thus by Liouville’s theorem
it is constant.
Theorem 12.3. Let f ∈ C(Λ), then
(a) ∑ω∈C/Λ resω (f ) = 0.
(b) ∑ω∈C/Λ ordω (f ) = 0.
(c) ∑ω∈C/Λ ordω (f ) ⋅ ω ∈ Λ.
Proof (a) Consider D a fundamental parallelogram. Then by the residues theorem

2πi ⋅ ∑ resω (f ) = 2πi ⋅ ∑ resω (f ) = ∫ f (z)dz

ω∈C/Λ ω∈D ∂D

a+ω1 a+ω1 +ω2 a+ω2 a

=∫ f (z)dz + ∫ f (z)dz + ∫ f (z)dz + ∫ f (z)dz
a a+ω1 a+ω1 +ω2 a+ω2
a+ω1 a+ω2 a a
=∫ f (z)dz + ∫ f (z)dz + ∫ f (z)dz + ∫ f (z)dz = 0.
a a a+ω1 a+ω2
(b) Since f is elliptic, f ′ is also elliptic, and so f ′ /f ∈ C(Λ) is also elliptic. Apply (a) to f ′ /f .
(c) By the residues theorem
zf ′ (z)dz
2πi ⋅ ∑ ordω (f ) ⋅ ω = ∫
ω∈C/Λ ∂D f (z)
a+ω1 zf ′ (z)dz a+ω1 +ω2 zf ′ (z)dz a+ω2 zf ′ (z)dz a zf ′ (z)dz
=∫ +∫ +∫ +∫
a f (z) a+ω1 f (z) a+ω1 +ω2 f (z) a+ω2 f (z)
a+ω1 zf ′ (z)dz a+ω2 (z + ω )f ′ (z)dz
1
a (z + ω )f ′ (z)dz
2
a zf ′ (z)dz
=∫ +∫ +∫ +∫
a f (z) a f (z) a+ω1 f (z) a+ω2 f (z)
a+ω2 f ′ (z)dz a f ′ (z)dz
= ω1 ∫ + ω2 ∫ = 2πi ⋅ (nω1 + mω2 ).
a f (z) a+ω1 f (z)

88
Definition 12.4. The degree of an elliptic function is the number of poles counted with
order in a fundamental parallelogram.
Corollary 12.4. Every non-constant elliptic function has degree at least 2.
Proof If f ∈ C(Λ) has degree 1, then by (a) of Theorem 12.3 it has residue zero at its
unique simple pole, which is absurd.
Definition 12.5. Let Λ be a lattice. The divisors group Div(C/Λ) is the free abelian
group generated by the points of the torus C/Λ, i.e. are formal sums of the form

D = ∑ nω ⋅ ω , with nω = 0 for all but finitely many ω ∈ C/Λ.

ω∈C/Λ

We say deg(D) ∶= ∑ω∈C/Λ nω ∈ Z is the degree of D. We denote Div0 (C/Λ) the subgroup of
degree zero divisors. For any f ∈ C(Λ)× we define the divisor of f by

div(f ) ∶= ∑ ordω (f ) ⋅ ω ∈ Div0 (C/Λ).

ω∈C/Λ

We say div(f ) is a principal divisor. The Picard group is defined by

Div(C/Λ) 0 Div0 (C/Λ)

Pic(C/Λ) ∶= , Pic (C/Λ) ∶= .
div(C(Λ)× ) div(C(Λ)× )

The map div ∶ C(Λ)× → Div0 (C/Λ) is clearly a group homomorphism. We also have the
natural summation map
sum ∶ Div(C/Λ) → C/Λ
D = ∑ nω ⋅ ω ↦ ∑ nω ⋅ ω.
ω∈C/Λ ω∈C/Λ

Theorem 12.5. We have the exact sequence

div sum
0 → C× → C(Λ)× Ð→ Div0 (C/Λ) ÐÐ→ C/Λ → 0.

In other words C/Λ ≃ Pic0 (C/Λ).

Before proving Theorem 12.5 we will construct some explicit elliptic functions.
Definition 12.6. Let Λ be a lattice. The Weierstrass ℘ function relative to Λ is defined
by the series
1 1 1
℘(z) ∶= 2 + ∑ ( 2
− 2).
z ω∈Λ∖{0} (z − ω) ω
The Eisenstein series of weight 2k for Λ is the series
1
G2k (Λ) ∶= ∑ 2k
.
ω∈Λ∖{0} ω

89
Theorem 12.6. Let Λ be a lattice.

(a) The Eisenstein series is absolutely convergent for all k > 1.

(b) The series defining Weierstrass ℘ function converges absolutely and uniformly on com-
pact subsets of C ∖ Λ. The series defines a meromorphic function on C having poles of
order 2 at each point of Λ and no other poles.

(c) The Weierstrass ℘ function is an even elliptic function.

Proof (a) Let R > 0 be any positive real number. Consider the set

S ∶= {ω ∈ Λ ∶ ∣ω∣ ≤ R},

and let D = ω1 ⋅ [0, 1) + ω2 ⋅ [0, 1) be a fundamental parallelogram with vertex at 0. Let d > 0
be the size of the biggest diagonal of D, then it is clear that

B(0; R − d) ⊆ S + D ⊆ B(0; R + d)

and so taking area

π(R − d)2 ≤ #S ⋅ A(Λ) ≤ π(R + d)2 .
Thus
πR2 2πd πd2
∣#S − ∣≤ R+ < C(Λ) ⋅ R
A(Λ) A(Λ) A(Λ)
for some constant C > 0 depending on Λ and for R ≥ 1. In consequence for R ≥ 1

#{ω ∈ Λ ∶ R < ∣ω∣ ≤ R + 1}

π(R + 1)2 πR2 π(2R + 1)

= #{ω ∈ Λ ∶ ∣ω∣ ≤ R + 1} − − #{ω ∈ Λ ∶ ∣ω∣ ≤ R} + −
A(Λ) A(Λ) A(Λ)
π
< (C − ) (2R + 1) ≤ C ′ ⋅ R.
A(Λ)
It follows that
∞
1 #{ω ∈ Λ ∶ N < ∣ω∣ ≤ N + 1} ∞ C ′
∑ 2k
≤ ∑ ≤ ∑ 2k−1 < ∞.
ω∈Λ,∣ω∣≥1 ∣ω∣ N =1 N 2k N =1 N

(b) For ∣ω∣ > 2∣z∣ we have

1 1 z(2ω − z) ∣z∣(2∣ω∣ + ∣z∣) 10∣z∣

∣ 2
− 2∣ = ∣ 2 2
∣≤ 2 ≤ .
(z − ω) ω ω (z − ω) ∣ω ∣(∣ω∣ − ∣z∣)2 ∣ω∣3

(c) It is clear by the definition of ℘ that it is even and elliptic, since −Λ = Λ and Λ + ω = Λ
for any ω ∈ Λ.

90
Theorem 12.7. Let Λ be a lattice, then

C(Λ) = C(℘, ℘′ ),

i.e. every elliptic function is a rational combination of ℘ and ℘′ .

Proof Let f ∈ C(Λ). Writing

f (z) + f (−z) f (z) − f (−z)

f (z) = + ,
2 2
we reduce ourselves to odd and even elliptic functions. Further if f (z) is odd, then f (z)℘′ (z)
is even and so we are reduced to the case f (z) is an even elliptic function. Assume f is even,
derivating the equality f (z) = f (−z) we get

f (i) (z) = (−1)i f (i) (−z).

Thus for any 2w ∈ Λ we have

f (i) (w) = f (i) (−w) = (−1)i f (i) (w).

And so f (i) (w) = 0 for all i odd, thus the ordw (f ) is even for all 2w ∈ Λ. Let D = ω1 ⋅ [0, 1) +
ω2 ⋅ [0, 1) be a fundamental parallelogram and H = ω1 ⋅ [0, 1) + ω2 ⋅ [0, 21 ) be its half. Then
(since f is even)
div(f ) = ∑ nω ⋅ (ω + (−ω)),
ω∈H
ordω (f )
for nω = ordω (f ) if 2ω ∉ Λ, and nω = 2 if 2ω ∈ Λ. Consider the function

g(z) ∶= ∏ (℘(z) − ℘(ω))nω .

ω∈H∖{0}

Since div(℘(z) − ℘(ω)) = ω + (−ω) − 2 ⋅ (0), it follows that

div(g) = n ⋅ (0) + ∑ nω ⋅ (ω + (−ω))

w∈H∖{0}

for some n ∈ Z. Since deg(div(g)) = deg(div(f )) = 0, it follows that n = 21 ord0 (f ) and so

div(g) = div(f ). In other words f = C ⋅ g ∈ C(℘, ℘′ ).
Definition 12.7. Let Λ be a lattice. The Weierstrass σ function relative to Λ is
z z + 1 ( z )2
σ(z) ∶= z ∏ (1 − ) eω 2 ω .
ω∈Λ∖{0} ω

Lemma 12.8. (a) The infinite product for σ(z) defines an entire function, with simple
zeros at Λ and no other zeros.
d2
(b) dz 2 log σ(z) = −℘(z) , for all z ∈ C ∖ Λ.

91
 (c) For every ω ∈ Λ there are constants a, b ∈ C depending on ω such that
σ(z + ω) = eaz+b σ(z) , for all z ∈ C.
Proof (a) Let z0 ∈ C ∖ Λ. Since a branch of the logarithm is given (locally around z0 ) by
∞
z 1 z n
log (1 − ) = ∑ ( ) ,
ω n=1 n ω

it follows that
z z 1 z 2
f (z) = ∑ log (1 − )− − ( )
ω∈Λ∖{0} ω ω 2 ω
is absolutely convergent and uniformly convergent on compacts (we can bound the absolute
sum by some constant plus ∣z∣3 ∑ω∈Λ∖{0} ∣ω∣−3 ). The result follows noting that σ(z) = zef (z)
around z0 . In the case z0 ∈ Λ, the same argument works for the above sum over ω ∈ Λ∖{0, z0 }
and so we get the result noting that σ(z) = z (1 − zz0 ) ef (z) locally around z0 . This shows that
σ is entire with only simple zeros at Λ.

(b) Just take logarithm around z0 ∈ C ∖ Λ and derive.

(c) By (b) if follows that the second derivative of log σ(z) is elliptic, and so
log σ(z + ω) = log σ(z) + az + b.

Proposition 12.9. Let n1 , . . . , nr ∈ Z and z1 , . . . , zr ∈ C such that

r r
∑ ni = 0 and ∑ ni ⋅ zi ∈ Λ.
i=1 i=1

Then there exists an elliptic function f ∈ C(Λ) such that

r
div(f ) = ∑ ni ⋅ zi ∈ Div0 (C/Λ).
i=1
r
Proof Let λ ∶= ∑i=1 ni ⋅ zi ∈ Λ. Replacing n1 , . . . , nr by n1 , . . . , nr , 1, −1 and z1 , . . . , zr by
z1 , . . . , zr , 0, λ we may assume that ∑ri=1 ni ⋅ zi = 0. Then
r
f (z) ∶= ∏ σ(z − zi )ni
i=1

has the right orders of zeros and poles. Furthermore for any ω ∈ Λ
r
f (z + ω)
= ∏ e(a(z−zi )+b)ni = e(az+b) ∑i=1 ni −a ∑i=1 zi ⋅ni = 1.
r i

f (z) i=1

Proof of Theorem 12.5 We have to show that Pic0 (C/Λ) ≃ C/Λ. For any P ∈ C/Λ we
have P = sum(P − (0)). And if D ∈ Div0 (C/Λ) with sum(D) = 0. It follows by the above
proposition that D = div(f ) for some f ∈ C(Λ)× .

92
Theorem 12.10. (a) The Laurent series for ℘(z) around z = 0 is
∞
1
℘(z) = 2
+ ∑ (2k + 1)G2k+2 z 2k .
z k=1

(b) For all z ∈ C ∖ Λ the Weierstrass ℘ function and its derivative satisfy the relation

℘′ (z)2 = 4℘(z)3 − 60G4 ℘(z) − 140G6 .

Proof (a) For all ∣z∣ < ∣ω∣ we have

∞
1 1 1 1 zn
− = ( − 1) = ∑ (n + 1) .
(z − ω)2 ω 2 ω 2 (1 − z/ω)2 n=1 ω n+2

(b) Write the principal part of the Laurent series

℘′ (z)2 = 4z −6 − 24G4 z −2 − 80G6 + ⋯,

℘(z)3 = z −6 + 9G4 z −2 + 15G6 + ⋯,

℘(z) = z −2 + 3G4 z 2 + ⋯.
It follows that
f (z) = ℘′ (z)2 = 4℘(z)3 − 60G4 ℘(z) − 140G6
is holomorphic with f (0) = 0, thus it is constant equal to 0.

Remark 12.2. It is standard notation to set

g2 ∶= 60G4 (Λ) , and g3 ∶= 140G6 (Λ).

Thus we have the relation

℘′ (z)2 = 4℘(z)3 − g2 ℘(z) − g3 .

Proposition 12.11. Let Λ be a lattice.

(a) The polynomial

f (x) = x3 − 4g2 x − g3
has distinct roots, and so its discriminant

∆(Λ) = g23 − 27g32

is not zero.

93
 (b) Let E/C be the elliptic curve given by

y 2 = x3 − 4g2 x − g3 .

The map
φ ∶ C/Λ → E
z ↦ (℘(z), ℘′ (z))
is a complex analytic isomorphism of complex Lie groups, i.e. it is an isomorphism of
groups which is also a biholomorphism of Riemann surfaces.
Proof (a) Let {ω1 , ω2 } be a basis of Λ and let ω3 = ω1 + ω2 . Since ℘′ (z) is an odd elliptic
function we see that
ωi −ωi ωi
℘′ ( ) = −℘′ ( ) = −℘′ ( ) = 0.
2 2 2
Thus f (x) vanishes at x = ℘(ωi /2). We just have to show these three values are different.
This follows since ℘(z) − ℘(ωi /2) has only one zero in any fundamental parallelogram, which
is the double zero attained at z = ωi /2.

(b) The image of φ is contained in E. Since it is holomorphic and non-constant it is surjective.

To see the injectivity suppose φ(z1 ) = φ(z2 ). Assume first that 2z1 ∉ Λ. Then the function
℘(z)−℘(z1 ) is elliptic of degree 2 vanishing at z1 , −z1 and z2 . Thus two of them are congruent
modulo Λ, i.e. z2 ≡ ±z1 (mod Λ). Then ℘′ (z2 ) = ℘′ (±z1 ) = ℘′ (z1 ) and so z2 ≡ z1 (mod Λ).
On the other hand, if 2z1 ∈ Λ it follows that ℘(z) − ℘(z1 ) has only one zero at z1 of degree
2, thus z2 = z1 .
Finally we have to check it is a group homomorphism. Let z1 , z2 ∈ C and consider some
f ∈ C(Λ) such that
div(f ) = (z1 + z2 ) − (z1 ) − (z2 ) + (0).
Writing f (z) = F (℘(z), ℘′ (z)) for some F (X, Y ) ∈ C(X, Y ). Evaluating F at the coordinates
of E, F (x, y) ∈ C(E) = C(x, y) it follows that

div(F (x, y)) = (φ(z1 + z2 )) − (φ(z1 )) − (φ(z2 )) + (φ(0)).

Since φ(0) = O it follows that

φ(z1 + z2 ) = φ(z1 ) + φ(z2 ).

Theorem 12.12. Let Λ1 and Λ2 be two lattices and suppose α ∈ C is such that

αΛ1 ⊆ Λ2 .

Thus we have an induced map

φα ∶ C/Λ1 → C/Λ2
z ↦ αz (mod Λ2 ).

94
 (a) The association

α ∈ {α ∈ C ∶ αΛ1 ⊆ Λ2 } ↦ φα ∈ {φ ∶ C/Λ1 → C/Λ2 holomorphic with φ(0) = 0}

is a bijection.

(b) Let E1 and E2 be the elliptic curves associated to the lattices Λ1 and Λ2 . The natural
inclusion

{isogenies φ ∶ E1 → E2 } → {φ ∶ C/Λ1 → C/Λ2 holomorphic with φ(0) = 0}

is a bijection.

Proof (a) If φα = φβ , then

αz ≡ βz (mod Λ2 ) for all z ∈ C.

Hence the map z ↦ (α − β)z sends C to Λ2 and so it is constant, i.e. α = β.

Let φ ∶ C/Λ1 → C/Λ2 with φ(0) = 0. Lifting φ to the universal coverings f ∶ C → C with
f (0) = 0 it follows that

f (z + ω) ≡ f (z) (mod Λ2 ) for all ω ∈ Λ1 and all z ∈ C.

It follows that f (z + ω) − f (z) is constant and so

f ′ (z + ω) = f ′ (z) for all ω ∈ Λ1 and all z ∈ C.

Thus f ′ (z) is elliptic and holomorphic, then it is constant, i.e. f (z) = αz for some α ∈ C.

(b) By (a) consider φα for α ∈ C× with αΛ1 ⊆ Λ2 . The induced map on Weierstrass equations
is
E1 → E2
(℘(z, Λ1 ), ℘′ (z, Λ1 )) ↦ (℘(αz, Λ2 ), ℘′ (αz, Λ2 ))
so we have to show it is a rational map. Since for any ω ∈ Λ1

℘(α(z + ω), Λ2 ) = ℘(αz, Λ2 ),

℘′ (α(z + ω), Λ2 ) = ℘′ (αz, Λ2 ).

It follows that they are in C(Λ1 ) = C(℘(z, Λ1 ), ℘′ (z, Λ1 )).

Corollary 12.13. Let E1 /C and E2 /C be two elliptic curves corresponding to lattices Λ1

and Λ2 respectively. Then E1 and E2 are isomorphic over C if and only if Λ1 and Λ2 are
homothetic, i.e. there exists some α ∈ C× such that αΛ1 = Λ2 .

95
13 Modular Forms
In this lecture we will begin our study of modular forms. We have already encountered
several examples of these function in our study of elliptic curves. But we have not seen them
as functions, only as numbers associated to elliptic curves. In order to see them as functions
we have to look at the moduli space of elliptic curves.

13.1 Modular group and the upper-half plane

One of the goals of this lecture is to show that every elliptic curve E/C comes from a lattice.
This will imply that the space of all elliptic curves over C can be parametrized by the space
of all lattices Λ ⊆ C. In order to parametrize the isomorphism classes of elliptic curves,
it follows from Corollary 12.13 that we have to parametrize the space of lattices modulo
homotheties. In particular, given any lattice Λ = ω1 Z + ω2 Z we can find some α = ω1−1 ∈ C×
such that Λ is homothetic to
ω2
Λτ = Z + τ Z , for τ = ∉ R.
ω1
Replacing τ by −τ if necessary, we can assume further that Im τ > 0.
Definition 13.1. The upper-half plane is
H ∶= {τ ∈ C ∶ Im τ > 0}.
Thus H parametrizes all isomorphism classes of elliptic curves. However, this is not a 1-1
correspondence. In fact, given τ1 , τ2 ∈ H
Eτ1 ≃ Eτ2 if and only if Λτ1 = αΛτ2 for some α ∈ C× .
This condition is equivalent to have a, b, c, d ∈ Z such that α = cτ1 + d and ατ2 = aτ1 + b, i.e.
aτ1 + b
τ2 = ,
cτ1 + d
and conversely we must have some a′ , b′ , c′ , d′ ∈ Z such that α−1 = c′ τ2 + d′ and
a′ τ 2 + b ′
τ1 = .
c′ τ2 + d′
Hence these numbers have to satisfy
aτ +b
a′ τ2 + b′ a′ ( cτ11+d ) + b′ (aa′ + b′ c)τ1 + a′ b + b′ d
τ1 = = 1 +b
= .
c′ τ2 + d′ c′ ( aτ
cτ1 +d
) + d ′ (ac ′ + cd′ )τ + bc′ + dd′
1

This relation is equivalent to have

a′ b ′ a b aa′ + b′ c a′ b + b′ d 1 0
( ′ ′) ( )=( ′ )=( ).
c d c d ac + cd′ bc′ + dd′ 0 1

96
In fact, if x = aa′ + b′ c, y = a′ b + b′ d, z = ac′ + cd′ and w = bc′ + dd′ , then
(ac′ + cd′ )τ1 + bc′ + dd′
1 = αα−1 = (cτ1 + d)(c′ τ2 + d′ ) = (cτ1 + d) ( ) = zτ1 + w.
cτ1 + d
Since τ1 ∈ H it follows that z = 0 and w = 1. Since
xτ1 + y
xτ1 + y = = τ1
zτ1 + w
it follows that y = 0 and x = 1. Hence the moduli of elliptic curves over C is given by H
modulo the following group action.
Definition 13.2. The full modular group is
a b
SL2 (Z) ∶= {( ) ∶ a, b, c, d ∈ Z , ad − bc = 1} .
c d
Lemma 13.1. Let τ ∈ C and a, b, c, d ∈ R with cτ + d ≠ 0. Then
aτ + b (ad − bc) Im τ
Im ( )= .
cτ + d ∣cτ + d∣2
Proof
aτ + b 1 aτ + b aτ + b 1 (ad − bc)τ + (bc − ad)τ̄ (ad − bc) Im τ
Im ( ) = [( )−( )] = = .
cτ + d 2i cτ + d cτ + d 2i ∣cτ + d∣2 ∣cτ + d∣2

Proposition 13.2. The full modular group acts on H by fractional linear transforma-
tions
a b aτ + b
( ) ⋅ τ ∶= .
c d cτ + d
And furthermore given τ1 , τ2 ∈ H
Eτ1 ≃ Eτ2 if and only if A ⋅ τ1 = τ2 for some A ∈ SL2 (Z).
Proof We already saw the group action in the previous discussion. And we saw that if
a b
Eτ! ≃ Eτ2 then there exists some A = ( ) with a, b, c, d ∈ Z and with A−1 also with integer
c d
coefficients, i.e. det(A) = ad − bc = ±1. By Lemma 13.1 we have
(ad − bc) Im τ1
Im τ2 =
∣cτ1 + d∣2
and so ad − bc = 1. Conversely if A ∈ SL2 (Z) and A ⋅ τ1 = τ2 , then taking α = cτ1 + d it follows
that
α(Z + τ2 Z) = Z + τ1 Z
and so Eτ1 ≃ Eτ2 .

97
Theorem 13.3. The full modular group SL2 (Z) is generated by

0 −1 1 1
S=( ) , and T = ( ).
1 0 0 1

Proof Let Γ ∶= ⟨S, T ⟩ ⊆ SL2 (Z). Suppose SL2 (Z) ∖ Γ is not empty. Note that every
a b
A=( ) ∈ SL2 (Z) ∖ Γ must have c ≠ 0, otherwise ad = 1 and so a = d = ±1 hence
c d

1 n
A = ±( ) = S 3±1 T n ∈ Γ.
0 1

b −a
Further we must also have d ≠ 0 since AS = ( ). Consider A ∈ SL2 (Z) ∖ Γ with
d −c
m(A) ∶= min{∣c∣, ∣d∣} = ∣c∣ minimal. Noting that

a b′
AT n = ( )
c nc + d

it follows by the division algorithm that there exists some n ∈ Z such that
∣c∣
∣nc + d∣ ≤ < ∣c∣.
2
Thus m(AT n ) < m(A) and so AT n ∈ Γ. But this implies A ∈ Γ which is a contradiction.
Remark 13.1. We want to show that every elliptic curve E/C is isomorphic to the elliptic
curve Eτ associated to some lattice Λτ for τ ∈ H. Using the modular invariant, it is enough
to show that the map
j∶H→C
τ ↦ j(Eτ )
is surjective. Since j is an isomorphism invariant for elliptic curves, it follows that as a
function on H, j is invariant under the action of SL2 (Z), i.e.
aτ + b a b
j( ) = j(τ ) , ∀τ ∈ H , ( ) ∈ SL2 (Z).
cτ + d c d

Definition 13.3. Let k ∈ Z. A meromorphic function f ∶ H → C is weakly modular of

weight k if
aτ + b a b
f( ) = (cτ + d)k f (τ ) , ∀τ ∈ H , ( ) ∈ SL2 (Z).
cτ + d c d
Remark 13.2. It follows from Theorem 13.3 that a meromorphic function f ∶ H → C is
weakly modular of weight k if and only if
−1
f( ) = τ k f (τ ) and f (τ + 1) = f (τ ) ∀τ ∈ H.
τ

98
Example 13.1. Examples of weakly modular functions of weight 2k ≥ 4 are the Eisenstein
series
1
G2k (τ ) ∶= ∑ 2k
.
(m,n)∈Z2 ∖{(0,0)} (mτ + n)

We have already seen that these series converge absolutely at every point τ ∈ H. This follows
from the inequality
∞
1 C ′ (τ )
∑ 2k
≤ ∑ 2k−1
.
∣mτ +n∣≥1 ∣mτ + n∣ N =1 N

But the continuity of C ′ (τ ) on τ ∈ H implies that the Eisenstein series also converge uniformly
on compact subsets of H. Hence G2k are holomorphic functions on τ ∈ H. As a consequence
g23 (τ ) 603 G4 (τ )3
j(τ ) = =
g23 (τ ) − 27g32 (τ ) 603 G4 (τ )3 − 27 ⋅ 1402 G6 (τ )2
is also holomorphic at H, since

∆(τ ) = 603 G4 (τ )3 − 27 ⋅ 1402 G6 (τ )2 ≠ 0 ∀τ ∈ H.

In particular, j is a weakly modular function of weight 0. The Eisenstein series are weakly
modular of weight 2k. In fact,
1 1
G2k (τ + 1) = ∑ 2k
= ∑ 2k
= G2k (τ ).
(m,n)∈Z2 ∖{(0,0)} (m(τ + 1) + n) (m,n)∈Z2 ∖{(0,0)} (mτ + n)

And also
−1 1 τ 2k
G2k ( )= ∑ 2k
= ∑ 2k
= τ 2k G2k (τ ).
τ −1
(m,n)∈Z ∖{(0,0)} (m ( ) + n)
2 (m,n)∈Z ∖{(0,0)} (nτ − m)
2
τ

In consequence the ∆ function is weakly modular of weight 12.

Remark 13.3. In general, given a weakly modular function f ∶ H → C, the condition

f (τ + 1) = f (τ )

implies that they are Z-periodic functions. Thus it can be factored by the function

τ ∈ H ↦ e2πiτ ∈ D∗ = {z ∈ C ∶ ∣z∣ < 1 , z ≠ 0},

i.e. we can consider f (τ ) = g(q) for q = e2πiτ ∈ D∗ . The meromorphic function g ∶ D∗ → C

admits a Laurent expansion around 0
∞
f (τ ) = ∑ an q n .
n=−∞

The relation
∣q∣ = e−2π Im τ

99
shows that q → 0 if and only if Im τ → +∞. This is why we say that the above Laurent
expansion corresponds to the q-expansion of f at i∞. We say f is holomorphic at i∞
if the Laurent expansion is holomorphic, i.e. if it is of the form
∞
f (τ ) = ∑ an q n .
n=0

This means that f has a Fourier expansion.

Definition 13.4. Let k ∈ Z. A weakly modular function of weight k, f ∶ H → C, is called a
modular form of weight k if it is holomorphic on H and holomorphic at i∞. The space
of all modular forms of weight k is denoted Mk (SL2 (Z)). We say that a modular form of
weight k is a cusp form if it vanishes at i∞, i.e. if a0 = 0 in the Fourier expansion of f .
The subspace of cusp forms is denoted Sk (SL2 (Z)) ⊆ Mk (SL2 (Z)).
Remark 13.4. If f ∈ Mk (SL2 (Z)) and g ∈ Ml (SL2 (Z)), then

f ⋅ g ∈ Mk+l (SL2 (Z)).

And further f ⋅ g ∈ Sk+l (SL2 (Z)) if any of the modular forms is a cusp form. This means that
modular forms form a graded ring

M(SL2 (Z)) ∶= ⊕ Mk (SL2 (Z))

k∈Z

and it has as a maximal ideal, the homogeneous ideal of cusp forms

S(SL2 (Z)) ∶= ⊕ Sk (SL2 (Z)).

k∈Z

Example 13.2. We know that G2k , j and ∆ are weakly modular functions holomorphic on
H. Let us compute their Laurent expansions at i∞. Let us see first the behaviour of G2k at
i∞
1 1 1
G2k (τ ) = ∑ 2k + ∑ ∑ 2k
= 2ζ(2k) + 2 ∑ ∑ 2k
.
n≠0 n m≠0 n∈Z (mτ + n) m≥1 n∈Z (mτ + n)

Considering the function fτ ∶ R → R given by

τ +x
fτ (x) ∶= ∣ ∣
i+x
it follows that fτ (x) > 0 for all x ∈ R and fτ (x) → 1 as x → ±∞. Thus there exists some δ > 0
such that fτ (x) > δ for all x ∈ R. In particular for x = n/m we have that

∣mτ + n∣ > δ∣mi + n∣ ∀m, n ∈ Z , m ≥ 1.

Therefore for any N ≥ 1

1 1 1 1
∑∑ 2k
≤ ∑ 2k
+ k ∑ .
m≥1 n∈Z ∣mτ + n∣ m+∣n∣≤N ∣mτ + n∣ δ m+∣n∣>N ∣mi + n∣2k

100
 1
Since ∑(m,n)∈Z2 ∖{(0,0)} ∣mi+n∣ 2k converges, the second term of the right hand side is small if N

is big enough. While the first term goes to zero as τ goes to i∞. In consequence
G2k (τ ) ÐÐÐ→ 2ζ(2k),
τ →i∞

and so G2k is a modular form of weight 2k.

Theorem 13.4. The Fourier expansion of the Eisenstein series is
2(2πi)2k
G2k (τ ) = 2ζ(2k) + ∑ σ2k−1 (n)q n ,
(2k − 1)! n≥1
where σi (n) ∶= ∑d∣n di , and q = e2πiτ .
Proof Consider the product formula for the sin function
∞
τ2
sin πτ = πτ ∏ (1 − ).
n=1 n2
Taking logarithmic derivative we see that
1 1 1
π cot πτ = +∑( + ).
τ n≥1 τ − n τ + n
On the other hand we can write
∞
1 1 1 i(eiπτ + e−iπτ ) 2πi
+∑( + ) = π cot πτ = π iπτ = iπ − = πi − 2πi ∑ qm.
τ n≥1 τ − n τ + n e − e−iπτ 1−q m=0

Differentiating 2k − 1 times this equality we get

1 (2πi)2k ∞ 2k−1 r
∑ 2k
= ∑r q .
n∈Z (τ + n) (2k − 1)! r=1
It follows that
1 2(2πi)2k ∞ ∞ 2k−1 mr
G2k (τ ) = 2ζ(2k) + 2 ∑ ∑ 2k
= 2ζ(2k) + ∑ ∑r q ,
m≥1 n∈Z (mτ + n) (2k − 1)! m=1 r=1
Taking n = mr we get the desired Fourier expansion.
Remark 13.5. Using the values of the Riemann zeta function
π4 π6
ζ(4) =, ζ(6) =
90 945
we can see that ∆ ∈ S12 (SL2 (Z)) is a cusp form. Thus the modular j invariant is not a
modular form, since its Laurent expansion at i∞ is of the form
1
1728 ⋅ j(τ ) = +⋯
q
From now on we will redefine the modular invariant j as
g23 (τ ) 1
j(τ ) ∶= 1728 ⋅ = +⋯
∆(τ ) q

101
13.2 The uniformization theorem
Finally we are in position to prove the uniformization theorem.

Theorem 13.5. The modular invariant function

j∶H→C

is surjective.

Proof Suppose by contradiction that c ∈ C is not in the image of j. Let γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4

2πi 4πi
be the closed path given by γ1 being the the arc of the unit circle form ω = e 3 to ω 2 = e 3 ,
γ2 being a line from ω 2 to ω 2 + N for some N ∈ N with N > 1, γ3 being the line between
ω 2 + N and ω + N , and γ4 being the line between ω + N and ω. Since j ′ (τ )/(j(τ ) − c) is
holomorphic on H we have
1 j ′ (τ )dτ
= 0.
2πi ∫γ j(τ ) − c
On the other hand since j is Z-periodic, it follows that the integrals over γ2 and γ4 cancel.
Since for τ ∈ γ1
j(τ ) = j(−τ̄ )
it follows that the integral over γ1 is also zero. Thus if j(τ ) = f (q) = 1q + ⋯ for q = e2πiτ , then

1 j ′ (τ )dτ 1 f ′ (q)dq
0= = =1
2πi ∫γ3 j(τ ) − c 2πi ∫δ f (q) − c

for δ = e2πiγ3 being a closed path around q = 0, which is absurd.

Corollary 13.6 (Uniformization theorem). For every elliptic curve E/C there exists
some τ ∈ H such that
E ≃ Eτ .
In consequence, for every pair of complex number g2 , g3 ∈ C such that g23 − 27g32 ≠ 0, there
exists some τ ∈ H such that

60G4 (τ ) = g2 and 140G6 (τ ) = g3 .

Proof Just compute j(E) and consider τ ∈ H such that j(τ ) = j(E).

Proposition 13.7. Let E/C be an elliptic curve with Weierstrass coordinates x and y.

(a) Let α and β be closed paths in E(C) giving a basis of H1 (E, Z). Then the periods

dx dx
ω1 = ∫ and ω2 = ∫
α y β y
are R-linearly independent.

102
 (b) Let the lattice Λ = ω1 Z + ω2 Z. Then the map

F ∶ E(C) → C/Λ
P dx
P ↦∫ (mod Λ),
O y
is a complex analytic isomorphism of Lie groups, inverse to the map (℘, ℘′ ).

Proof (a) By the uniformization theorem, there exists some lattice Λ1 such that

φ = (℘, ℘′ ) ∶ C/Λ1 → E(C)

is a complex analytic isomorphism. Hence φ−1 ○ α and φ−1 ○ β are a basis for H1 (C/Λ1 , Z).
Since the first homology group is naturally isomorphic to Λ1 via the map

γ ∈ H1 (C/Λ1 , Z) ↦ ∫ dz ∈ Λ1 ,
γ

and the invariant differential pulls-back to

dx d℘(z)
φ∗ ( )= ′ = dz,
y ℘ (z)
it follows that the periods

ω1 = ∫ dz and ω2 = ∫ dz
φ−1 ○α φ−1 ○β

are a basis for Λ1 .

(b) The composition F ○ φ gives

(℘(z),℘′ (z)) dx
F (φ(z)) = ∫ .
O y

Since F ∗ (dz) = dx
y and φ∗ ( dx
y ) = dz we see that

(F ○ φ)∗ (dz) = dz.

On the other hand since F ○ φ ∶ C/Λ → C/Λ takes O to O, it must be of the form αz for some
α ∈ C× . Thus α = 1 and so F = φ−1 .

Remark 13.6. We have the natural identification

H1 (E(C), Z) ≃ Λ
dx
γ↦∫ .
γ y

103
Tensoring it by Z/mZ we get the identification

H1 (E(C), Z/mZ) ≃ H1 (E(C), Z) ⊗ Z/mZ ≃ Λ ⊗ Z/mZ ≃ Λ/mΛ.

On the other hand using the isomorphism of groups E(C) ≃ C/Λ we see that
∼
E(C)[m] ≃ (C/Λ)[m] = {z ∈ C ∶ mz ∈ Λ}/Λ Ð→ Λ/mΛ.
⋅m

Hence we have the natural identification

E(C)[m] ≃ H1 (E(C), Z/mZ).

In particular for ` a prime number we have the natural homological interpretation of the
Tate module
T` (E) ≃ H1 (E(C), Z` ).

14 The Algebra of Modular Forms

14.1 Computing dimensions of spaces of modular forms
Modular forms appeared first in mathematics as their q-expansions, usually corresponding
to series with integer coefficients having some enumerative interpretation. One remarkable
instance of this enumerative interpretation of the coefficients of a q-expansion is the case of
the modular j invariant. Once we normalize it
g23 (τ ) 1
j(τ ) = 1728 = + 744 + 196884q + 21493760q 2 + 864299970q 3 + 20245856256q 4 + ⋯
∆(τ ) q
it is possible to show that coefficients of the above Fourier expansion starting by 196884, cor-
respond to some sum of (possible repeated) terms of the sequence rn = 1, 196883, 21296876,
842609326, 18538750076, 19360062527, 293553734298, . . . corresponding to the dimensions
of the irreducible representations of the monster group. This highly non-trivial fact inspired
the Monstrous moonshine conjecture proved by Borcherds in 1992. In general, it is
interesting to determine whether the generating function of a enumerative sequence corre-
sponds to some nice function (e.g. a modular form), and conversely it is also interesting to
determine enumerative interpretations of the coefficients of modular forms. Sometimes even
to determine the rationality of the coefficients is interesting enough from a transcendental
number theory point of view. A simple instance of this phenomenon is the case of Eisenstein
series.
Definition 14.1. For every k ≥ 2 we define the normalized Eisenstein series as
∞
G2k (τ ) (2πi)2k
E2k (τ ) ∶= =1+ ∑ σ2k−1 (n)q n ,
2ζ(2k) ζ(2k) ⋅ (2k − 1)! n=1
for q = e2πiτ .

104
 π4 π6
Remark 14.1. Once we have computed ζ(4) = 90 and ζ(6) = 945 , we can see that E2 , E6 ∈
Q[[q]]. In fact, it turns out that for all k ≥ 2

E2k ∈ Q[[q]].

In other words
ζ(2k) ∈ π 2k ⋅ Q.
This fact can be proved by elementary computations using Bernoulli numbers (see Problem
13), but we will give a different proof only assuming known the values of ζ(4) and ζ(6). It
will be a corollary of the fact that the algebra of modular forms is in fact generated by E4
and E6 . We will prove this in several steps.

Definition 14.2. The fundamental domain for the action of SL2 (Z) on H is
1
F ∶= {x + iy ∈ H ∶ ∣x∣ ≤ , ∣x + iy∣ ≥ 1} .
2
Proposition 14.1. Every orbit of the action of SL2 (Z) on H has a representative in the
fundamental domain F.

Proof Given any τ ∈ H, we claim that there exists some γ = ( ac db ) ∈ SL2 (Z) minimizing the
value ∣cτ + d∣. This is clear since (Z + τ Z) ∩ B(0; 1) is bounded and discrete. Note that since

Im τ
Im γ ⋅ τ =
∣cτ + d∣2

this is equivalent to maximize the value Im γ ⋅ τ . Using the action of T = ( 10 11 ) we can move
γ ⋅ τ to the strip ∣x∣ ≤ 21 . We claim ∣γ ⋅ τ ∣ ≥ 1. In fact, putting ω = γ ⋅ τ , if we suppose ∣ω∣ < 1,
then
Im ω
Im (S ⋅ ω) = > Im ω
∣ω∣2
contradicting our choice of γ. Therefore γ ⋅ τ ∈ F.

Proposition 14.2. For k < 0, Mk (SL2 (Z)) = 0.

Proof Let f ∈ Mk (SL2 (Z)), and write its Fourier expansion

∞
f (τ ) = ∑ an q n .
n=0

Noting that for any ( ac db ) ∈ SL2 (Z)

k k
aτ + b aτ + b 2 Im τ 2 k
∣f ( )∣ (Im ( )) = ∣(cτ + d)k f (τ )∣ ( ) = ∣f (τ )∣(Im τ ) 2
cτ + d cτ + d ∣cτ + d∣2

105
 k
we see that the function ∣f (τ )∣(Im τ ) 2 is SL2 (Z)-invariant. Thus its image correspond to
k
the image of the fundamental domain F. Since ∣f (τ )∣ → ∣a0 ∣ and (Im τ ) 2 → 0 as τ → i∞, it
follows that there exists some C > 0 such that
k
∣f (x + iy)∣ ⋅ y 2 ≤ C ∀x + iy ∈ H.

For any y > 0 and any m ≥ 0 we get

1 ∞ 1
∫0 f (x + iy)e−2πimx dx = ∑ an e−2πny ∫ e2πi(n−m)x dx = am e−2πmy .
n=0 0

Hence
−k
∣am ∣ ≤ Ce2πmy y 2 ÐÐ→ 0,
y→0

therefore am = 0 for all m, and so f = 0.

Remark 14.2. Since S 2 = −Id, it follows that for every f ∈ Mk (SL2 (Z))

f (τ ) = (−1)k f (τ ).

Thus for k odd, Mk (SL2 (Z)) = 0.

Proposition 14.3. For k ∈ {0, 1, 2, 3, 4, 5} we have
⎧
⎪ C if k = 0
⎪
⎪
M2k (SL2 (Z)) = ⎨ 0 if k = 1
⎪
⎪
⎪
⎩ C ⋅ E 2k otherwise.

Proof If k ≠ 1, let f ∈ M2k (SL2 (Z)) and consider E ∈ M2k (SL2 (Z)) to be the constant
function 1 if k = 0, and E = E2k if k ≥ 2. Let a0 ∈ C be the constant term of the Fourier
expansion of f . Since ∆ is a cups form with Fourier expansion

∆(τ ) = (2π)12 q + ⋯,

it follows that
f − a0 E
∈ M2k−12 (SL2 (Z)) = 0
∆
and so f = a0 ⋅ E. For the case k = 1 consider f ∈ M2 (SL2 (Z)), then f (−1/τ ) = τ 2 f (τ ) for all
τ ∈ H. In particular for τ = i we have

f (i) = −f (i) = 0.

Since f 2 ∈ M4 (SL2 (Z)), there exists some c ∈ C such that f 2 = c ⋅ E4 . Thus

∞
0 = f (i)2 = c ⋅ E4 (i) = c ⋅ (1 + 240 ∑ σ3 (n)e−2πn )
n=1

and so c = 0, i.e. f = 0.

106
Proposition 14.4. For every k ≥ 0

⌊k/6⌋ + 1 if k ≡/ 1 (mod 6),

dimC M2k (SL2 (Z)) = {
⌊k/6⌋ if k ≡ 1 (mod 6).

Proof It follows by induction on k. We already proved the cases k = 0, 1, 2, 3, 4, 5. For k ≥ 6

we have the isomorphism
∼
S2k (SL2 (Z)) Ð
→ M2k−12 (SL2 (Z))

f
f↦ .
∆
On the other hand since E2k ∈ M2k (SL2 (Z)) ∖ S2k (SL2 (Z)), it follows that

dimC M2k (SL2 (Z)) = 1 + dimC S2k (SL2 (Z)) = 1 + dimC M2k−12 (SL2 (Z)).

Proposition 14.5. For every k ≥ 0

dimC M2k (SL2 (Z)) = #{(a, b) ∈ (Z≥0 )2 ∶ 4a + 6b = 2k}.

Proof Again by induction, we check the result for k = 0, 1, 2, 3, 4, 5. For k ≥ 6 we have the
bijection

{(a, b) ∈ (Z≥0 )2 ∶ 4a + 6b = 2k} ←→ {(a, b) ∈ (Z≥0 )2 ∶ 4a + 6b = 2k − 12} ∪ {(α, β)}

(a, b) ↦ (a, b − 2)
where (α, β) = ( k−3 k
2 , −1) if k is odd, and (α, β) = ( 2 , −2) if k is even.

Lemma 14.6.
E4 (i) > 0 and E6 (i) = 0.

Proof Since E6 is a modular form of weight 6, we have

E6 (i) = E6 (−1/i) = i6 E6 (i) = −E6 (i) = 0.

On the other hand ∞

E4 (i) = 1 + 240 ∑ σ3 (n)e−2πn > 0.
n=1

Theorem 14.7. The algebra of modular forms is generated by E4 and E6 , i.e.

M(SL2 (Z)) = C[E4 , E6 ].

107
Proof We will prove that for every k ≥ 0

M2k (SL2 (Z)) = ⟨{E4a E6b ∶ a, b ∈ Z≥0 , 4a + 6b = 2k}⟩.

In fact, we just need to check they have the same dimension, i.e. that {E4a E6b ∶ a, b ∈
Z≥0 , 4a + 6b = 2k} are linearly independent. We will prove this by induction on k. Suppose
there exists a non-trivial relation

∑ ca,b E4a E6b = 0.

4a+6b=2k

If the above expression can be factored by E6 , it follows that

∑ ca,b E4a E6b−1 = 0

4a+6b=2k

and so by induction hypothesis all ca,b = 0. Otherwise, k is even and c k ,0 ≠ 0. Evaluating the
2
expression at τ = i it follows by the previous lemma that
k
c k ,0 E4 (i) 2 = 0
2

which is impossible.

Corollary 14.8. If k > 0 and f ∈ M2k (SL2 (Z)) has Fourier coefficients an ∈ Q for all n ≥ 1,
then a0 ∈ Q.

Proof If k = 1, f = 0 and there is nothing to prove. If k ≥ 2, then we can write

∞
f = ∑ an q n = ∑ ca,b E4a E6b .
n=0 4a+6b=2k

It follows that for every automorphism σ ∈ Aut(C) fixing the rational numbers Q,
∞
f σ ∶= ∑ σ(an )q n = ∑ σ(ca,b )E4a E6b .
n=0 4a+6b=2k

Note that we used the fact that E4 and E6 have rational coefficients. It follows that f σ is
also a modular form of weight 2k and so

f − f σ = a0 − σ(a0 ) ∈ M2k (SL2 (Z)),

hence a0 = σ(a0 ) for any automorphism of C/Q. In conclusion a0 ∈ Q.

Corollary 14.9. E2k ∈ Q[[q]] for all k ≥ 4.

E2k
Proof The modular form (2πi)2k /ζ(2k)
has rational Fourier coefficients starting from n ≥ 1.
ζ(2k)
Thus (2πi)2k
∈ Q.

108
14.2 Congruence subgroups
In the first lecture we introduced the Theta function
∞
n2 2
Θ(q) = ∑ q = 1 + 2 ∑ qn .
n∈Z n=1

If we consider it a Fourier series q = e2πiτ , i.e. we define

∞
2τ
θ(τ ) ∶= 1 + 2 ∑ e2πin
n=1

we see it is Z-periodic, and we can ask ourselves about the action of the other elements of
SL2 (Z) on it. It turns out that this function is not weakly modular of any weight, since it
do not behaves as desired with respect to the action of S = ( 01 −1 0 ). Nevertheless it can be
4
shown that θ satisfies
τ
θ4 ( ) = (4τ + 1)2 θ4 (τ ),
4τ + 1
hence it is weakly modular of weight 2 for the subgroup ⟨T, ( 14 01 )⟩ ⊆ SL2 (Z).

Definition 14.3. Let N be a positive integer. The principal congruence subgroup of

level N is
a b a b 1 0
Γ(N ) ∶= {( ) ∈ SL2 (Z) ∶ ( )≡( ) (mod N )} .
c d c d 0 1
We say a subgroup Γ ⊆ SL2 (Z) is a congruence subgroup of level N if

Γ(N ) ⊆ Γ.

The smallest of such N is called the level of Γ.

Remark 14.3. We can see Γ(N ) as the kernel of the homomorphism

SL2 (Z) → SL2 (Z/N Z).

In order to see that this morphism is in fact surjective, take ( ac db ) such that ad − bc ≡ 1 (mod
N ), i.e. ad − bc = 1 + jN . Then gcd(a, b, N ) = 1. We claim there exists some k, l ∈ Z such
that gcd(a + kN, b + lN ) = 1 and so given e, f ∈ Z such that (a + kN )e − (b + lN )f = 1, we get

a + kN b + lN
( ) ∈ SL2 (Z)
c − f (j + kd − lc)N d − e(j + kd − lc)N

as desired. In order to produce k, l ∈ Z such that gcd(a + kN, b + lN ) = 1 write a = a1 a2 where

all prime divisors of a2 divide N and gcd(a1 , N ) = 1. Then we can find r, s ∈ Z such that
ra1 + sN = 1. It is enough to take k = 0 and l = s(1 − a). In fact if p∣ gcd(a, b + s(1 − b)N )
then p∣a1 a2 . If p∣a1 then sN ≡ 1 (mod p) and so b + s(1 − b)N ≡ 1 (mod p) which is absurd.
Hence p∣a2 and so p∣N , then p∣ gcd(a, b) which implies p∣ gcd(a, b, N ) = 1 absurd.

109
 In order to compute the index of Γ(N ) we need to compute the size of SL2 (Z/N Z) ≃
⊕pα ∣∣N SL2 (Z/pα Z). It is easy to see that

#GL2 (Z/pZ) = (p2 − 1)(p2 − p)

since it corresponds to the amount l.i. vectors {v1 , v2 } in F2p (pick v1 ∈ F2p ∖ {(0, 0)} and then
pick v2 ∈ F2p ∖ (Fp ⋅ v1 )). Then #SL2 (Z/pZ) = p(p2 − 1). Since the kernel of the epimorphism
SL2 (Z/pn+1 Z) → SL2 (Z/pn Z) are matrices of the form

1 + t1 pn t 2 pn
( )
t3 pn 1 + t4 pn

with t1 , t2 , t3 , t4 ∈ Z/pZ and t1 + t4 ≡ 0 (mod p), it follows that this kernel has p3 elements and
so #SL2 (Z/pn+1 Z) = p3 #SL2 (Z/pn Z). Therefore

#SL2 (Z/pα Z) = p3α−2 (p2 − 1).

Thus we can compute the index as

1
[SL2 (Z) ∶ Γ(N )] = #SL2 (Z/N Z) = ∏ #SL2 (Z/pα Z) = ∏ p3α−2 (p2 − 1) = N 3 ∏ (1 − ).
pα ∣∣N pα ∣∣N p∣N p2

In consequence all congruence subgroups have finite index.

Definition 14.4. Beside Γ(N ), the most important congruence subgroups are

a b a b ∗ ∗
Γ0 (N ) ∶= {( ) ∈ SL2 (Z) ∶ ( )≡( ) (mod N )} ,
c d c d 0 ∗

a b a b 1 ∗
Γ1 (N ) ∶= {( ) ∈ SL2 (Z) ∶ ( )≡( ) (mod N )} .
c d c d 0 1

Remark 14.4. We have the inclusions

Γ(N ) ⊆ Γ1 (N ) ⊆ Γ0 (N ) ⊆ SL2 (Z).

The map ( ac db ) ∈ Γ1 (N ) ↦ b ∈ Z/N Z is a group homomorphism with kernel Γ(N ), hence

Γ(N ) ◁ Γ1 (N ) and [Γ1 (N ) ∶ Γ(N )] = N.

Similarly the map ( ac db ) ∈ Γ0 (N ) ↦ d ∈ (Z/N Z)× is an epimorphism with kernel Γ1 (N ),

hence
Γ1 (N ) ◁ Γ0 (N ) and [Γ0 (N ) ∶ Γ1 (N )] = ϕ(N ).
It follows that
1
[SL2 (Z) ∶ Γ0 (N )] = N ∏ (1 + ) .
p∣N p

110
Definition 14.5. We will introduce two notations. For every γ = ( ac db ) ∈ SL2 (Z) we intro-
duce the factor of automorphy
j(γ, τ ) ∶= cτ + d.
We define the weight k operator [γ]k on functions f ∶ H → C as
(f [γ]k )(τ ) ∶= j(γ, τ )−k f (γ ⋅ τ ).
Remark 14.5. It follows that f is weakly modular of weight k if it is meromorphic and
f [γ]k = f ∀γ ∈ SL2 (Z).
Lemma 14.10. For all γ, γ ′ ∈ SL2 (Z) and τ ∈ H,
(a) j(γγ ′ , τ ) = j(γ, γ ′ ⋅ τ )j(γ ′ , τ ),
(b) [γγ ′ ]k = [γ ′ ]k ○ [γ]k .
Proof (a) Write γ = ( ac db ) and γ ′ = ( ac′
′ b′ ).
d′ Then
j(γγ ′ , τ ) (a′ c + c′ d)τ + b′ c + dd′ a′ τ + b ′
= = c ( ) + d = j(γ, γ ′ ⋅ τ ).
j(γ ′ , τ ) c′ τ + d′ c′ τ + d ′
(b)
(f [γγ ′ ]k )(τ ) = j(γγ ′ , τ )−k f (γγ ′ ⋅ τ ) = j(γ, γ ′ ⋅ τ )−k j(γ ′ , τ )−k f (γ ⋅ (γ ′ ⋅ τ ))
= j(γ ′ , τ )−k (f [γ]k )(γ ′ ⋅ τ ) = ((f [γ]k )[γ ′ ]k )(τ ).

Definition 14.6. Let Γ ⊆ SL2 (Z) be a congruence subgroup. We say a meromorphic function
f ∶ H → C is weakly modular of weight k if
f [γ]k = f ∀γ ∈ Γ.
Remark 14.6. Since every congruence subgroup contains some matrix of the form
1 h
( )
0 1
for some h > 0 minimal (it can be less than the level of Γ). It follows that every weakly
modular function is hZ-periodic, and so it can be factored as
2πiτ
f (τ ) = g(q) for q = e h ,
and so it has Laurent expansion of the form
∞
f (τ ) = ∑ an q n .
n=−∞

We say f is holomorphic at i∞ if this Laurent series is holomorphic at q = 0, i.e. if f has a

Fourier expansion
∞
f (τ ) = ∑ an q n .
n=0

111
Definition 14.7. Let Γ be a congruence subgroup of SL2 (Z) and let k ∈ Z. A weakly
modular function of weigh k for Γ is called a modular form if it is holomorphic on H and
f [γ]k is holomorphic at i∞ for all γ ∈ SL2 (Z). We say it is a cusp form if a0 = 0. We denote
the spaces of weight k modular forms and cups forms by Mk (Γ) and Sk (Γ) respectively.
They also form a graded ring
M(Γ) ∶= ⊕ Mk (Γ),
k∈Z

with (not necessarily maximal) homogeneous ideal

S(Γ) ∶= ⊕ Sk (Γ).
k∈Z

Remark 14.7. In the same way SL2 (Z) appeared as the natural group determining the
moduli space of elliptic curves over C. We can also interpret the main congruence subgroups
as determining some moduli of enhanced elliptic curves. This means looking not only at the
isomorphisms of elliptic curves, but at isomorphisms preserving some extra structure. As we
preserve more structure, the corresponding congruence subgroup becomes smaller. In fact
Γ0 (N ) will correspond with the moduli space of elliptic curves enhanced with a subgroup of
order N , Γ1 (N ) corresponds with the moduli space of elliptic curves enhanced with a point
of order N , and Γ(N ) corresponds with the moduli space of elliptic curves enhanced with
two points of order N generating E[N ] and with Weil pairing equal to 1.

15 Modular Forms of Higher Level

15.1 Examples of modular forms for congruence subgroups
Example 15.1. In spite there are no modular forms of weight 2 for SL2 (Z), there exists
weight 2 modular forms for other congruence subgroups. This is the case of the weight 2
Eisenstein series. Consider
1
G2 (τ ) ∶= ∑ ∑ 2
m∈Z n∈Z′m (mτ + n)

where Z′m = Z ∖ {0} if m = 0 and Z′m = Z otherwise. This series is not absolutely convergent,
but it is still convergent. In order to show this consider
1 1
am,n (τ ) ∶= −
mτ + n − 1 mτ + n
for (m, n) ≠ (0, 0), (0, 1). Note that
1 −1
− a m,n (τ ) =
(mτ + n)2 (mτ + n)2 (mτ + n − 1)
and the double series
1
∑ ∑
m∈Z n∈Z′m (mτ + n)2 (mτ + n − 1)

112
is absolutely convergent (by a bound similar to the Eisenstein series of higher weight). Fur-
ther for m ≠ 0
∑ am,n (τ ) = 0.
n∈Z

Thus
1 1
G2 (τ ) = 2ζ(2) + ∑ ∑ 2
= 2ζ(2) + ∑ ∑ ( 2
− am,n (τ ))
m≠0 n∈Z (mτ + n) m≠0 n∈Z (mτ + n)

is now absolutely convergent and uniformly convergent in compact subsets of H, hence

determines a holomorphic function on H. Now it is easy to see from the definition of G2 that
it is Z-periodic. Furthermore we can compute
−1 τ2 2 τ2
G2 ( ) = 2ζ(2) + ∑ ∑ 2
= 2ζ(2)(1 + τ ) + ∑ ∑ 2
τ m≠0 n∈Z (−m + nτ ) m≠0 n≠0 (nτ − m)

τ2
= 2ζ(2)τ 2 + ∑ ∑ 2
.
m∈Z n≠0 (nτ − m)

Relabeling we get

−1 1
G2 ( ) = τ 2 (2ζ(2) + ∑ ∑ 2
) = τ 2 (G2 (τ ) + ∑ ∑ am,n (τ )) .
τ n∈Z m≠0 (mτ + n) n∈Z m≠0

This shows that G2 is not a modular form, it is in fact an example of “quasi”-modular form.
Let us compute the error term. Recall the cotangent expansion
1 ∞ 1 1
π cot πτ = +∑( + ).
τ m=1 τ + m τ − m
Then
1 π(n − 1) τ πn τ
∑ ∑ am,n (τ ) = ∑ (π cot ( )− − π cot ( ) + )
n∈Z m≠0 τ n≠0,1 τ n−1 τ n
1 1 1 1
+∑( − )+ ∑ ( − ).
m≠0 mτ − 1 mτ m≠0 mτ mτ + 1
The last two terms add
1 1 −1 1 1 −2 π −2π π
∑( − )= ∑( + )= (π cot ( ) − τ ) = cot ( )+2.
m≠0 mτ − 1 mτ + 1 τ m≠0 (1/τ ) + m (1/τ ) − m τ τ τ τ
While the first sum corresponds to
N
1 π(n − 1) τ πn τ
lim ∑ (π cot ( )− − π cot ( ) + )
τ N →∞ τ n−1 τ n
n=−N, n≠0,1

1 π(N + 1) τ π πN τ
= lim (−π cot ( )+ + 2π cot ( ) − 2τ − π cot ( )+ )
τ N →∞ τ N +1 τ τ N

113
 2π π 2π πN 2π π 2πi
cot ( ) − 2 −
= lim cot ( )= cot ( ) − 2 − .
τ τ τ N →∞ τ τ τ τ
Therefore we get the equation
−1
G2 ( ) = τ 2 G2 (τ ) − 2πiτ.
τ
We can define the normalized Eisenstein series of weight 2 as

G2 (τ ) −1 6τ
E2 (τ ) ∶= , E2 ( ) = τ 2 E2 (τ ) + ,
2ζ(2) τ πi

and the same computation we did for higher weights gives us the Laurent expansion of E2
at i∞ as ∞
E2 (τ ) = 1 − 24 ∑ σ1 (n)q n , for q = e2πiτ .
n=1

For any ( ac db ) ∈ SL2 (Z) we have the identity

aτ + b 6c(cτ + d)
E2 ( ) = (cτ + d)2 E2 (τ ) + .
cτ + d πi
In fact this is true for S, T ∈ SL2 (Z). And so it is enough to verify that the matrices satisfying
the identity are closed under product and inverse. To see this, note that if ( ac db ) satisfies the
−1 d −b ) and so
identity, then ( ac db ) = ( −c a

−2 −1
dτ − b dτ − b 6c dτ − b 6c(−cτ + d)
E2 ( ) = (c ( ) + d) E2 (τ )− (c ( ) + d) = (−cτ +a)2 E2 (τ )− .
−cτ + a −cτ + a πi −cτ + a πi

And given two matrices γ = ( ac db ) , γ ′ = ( ac′

′ b′ )
d′ ∈ SL2 (Z) satisfying the identity we have

6c(c(γ ′ ⋅ τ ) + d)
E2 (γγ ′ ⋅ τ ) = j(γ, γ ′ ⋅ τ )2 E2 (γ ′ ⋅ τ ) +
πi
6c′ (c′ τ + d′ ) 6c(c(γ ′ ⋅ τ ) + d)
= j(γ, γ ′ τ )2 j(γ ′ , τ )2 E2 (τ ) + j(γ, γ ′ ⋅ τ )2 +
πi πi
6((a c + c d)τ + b c + dd ) c (c τ + d ) 6c((a c + c′ d)τ + b′ c + dd′ )
′ ′ ′ ′ 2 ′ ′ ′ ′
= j(γγ ′ , τ )2 E2 (τ ) + +
πi(c′ τ + d′ )2 πi(c′ τ + d′ )
6((a′ c + c′ d)τ + b′ c + dd′ )
= j(γγ ′ , τ )2 E2 (τ ) + ′ ′
(((a′ c + c′ d)τ + b′ c + dd′ )c′ + c)
πi(c τ + d )
6((a′ c + c′ d)τ + b′ c + dd′ ) ′
= j(γγ ′ , τ )2 E2 (τ ) + ′ ′
(a c + c′ d)(c′ τ + d′ )
πi(c τ + d )
6(a′ c + c′ d)((a′ c + c′ d)τ + b′ c + dd′ )
= j(γγ ′ , τ )2 E2 (τ ) + .
πi

114
We can use E2 to define modular forms of weight 2 for Γ0 (N ) as

E2,N (τ ) ∶= E2 (τ ) − N E2 (N τ ).

In fact,
aτ + b aτ + b N aτ + N b
E2,N ( ) = E2 ( ) − N E2 ( )
cτ + d cτ + d cτ + d
6c(cτ + d) a(N τ ) + N b
= (cτ + d)2 E2 (τ ) + − N E2 ( c )
πi N (N τ ) + d

2 6c(cτ + d) 2
6 Nc (cτ + d)
= (cτ + d) E2 (τ ) + − N (cτ + d) E2 (N τ ) − N
πi πi
2 2
= (cτ + d) (E2 (τ ) − N E2 (N τ )) = (cτ + d) E2,N (τ )
for any ( ac db ) ∈ Γ0 (N ). The subtle point is to prove that E2,N [γ]2 is holomorphic at i∞ for
every γ ∈ SL2 (Z). We will assume this for now.

Remark 15.1. The method we used to produce modular forms for Γ0 (N ) from E2 works
in general for any modular form f ∈ M2k (SL2 (Z)). In fact the same proof shows that

fN (τ ) ∶= f (τ ) − N f (N τ ) ∈ M2k (Γ0 (N )).

Despite E2 is not a modular form, we can still use it to derive interesting consequences.
Let us introduce first another famous function.

Definition 15.1. The Dedekind η function is defined for every τ ∈ H by

∞
2πiτ
η(τ ) ∶= e 24 ∏(1 − e2πinτ ).
n=1

Proposition 15.1. For every τ ∈ H

√
−1 τ
η( ) = η(τ ).
τ i
Where the square root is the branch with non-negative real part.

Proof It is not hard to prove that η is a well-defined holomorphic function on H. Taking

the logarithmic derivative we get
∞
η ′ (τ ) πi nq n
= (1 − 24 ∑ n
)
η(τ ) 12 n=1 1 − q

for q = e2πiτ . Since

∞ ∞ ∞ ∞
nq n nm
∑ n
= ∑∑ nq = ∑ σ1 (n)q n
n=1 1 − q n=1 m=1 n=1

115
we get
η ′ (τ ) πiE2 (τ )
= .
η(τ ) 12
Therefore
η ′ (−1/τ ) 1 πiE2 (−1/τ ) 1 πi 2 6τ 1 πiE2 (τ ) 1 η ′ (τ ) 1
⋅ 2= ⋅ 2= (τ E2 (τ ) + ) ⋅ 2 = + = + .
η(−1/τ ) τ 12 τ 12 πi τ 12 2τ η(τ ) 2τ
√
In other words η(−1/τ ) and τ η(τ ) have the same logarithmic derivative, and so
√
η(−1/τ ) = λ τ η(τ ).
√
Evaluating at τ = i it follows that λ = 1i .

Corollary 15.2 (Jacobi). The ∆ function has the following product decomposition
∞
∆(τ ) = (2π)12 q ∏(1 − q n )24 , for q = e2πiτ .
n=1

Proof Proposition 15.1 implies that η 24 ∈ S12 (SL2 (Z)). We know that dimC M12 (SL2 (Z)) =
2 and so it is generated by E12 ∈ M12 (SL2 (Z))∖S12 (SL2 (Z)) and ∆ ∈ S12 (SL2 (Z)). Therefore

S12 (SL2 (Z)) = C ⋅ ∆.

The result follows noting that the Fourier expansion is ∆ = (2π)12 q + ⋯.

Remark 15.2. It follows from the previous corollary that the normalized ∆ function has
integral Fourier coefficients
∞ ∞
∆ n 24
= q ∏ (1 − q ) = ∑ τ (n)q n .
(2π)12 n=1 n=1

The function τ ∶ N → Z collecting there Fourier coefficients is called Ramanujan τ function.

This function was first studied by Ramanujan in 1916, whom observed that this function
satisfies the following remarkable properties (known as Ramanujan conjecture)

(i) τ is a multiplicative function, i.e. τ (m ⋅ n) = τ (m) ⋅ τ (n) for gcd(m, n) = 1,

(ii) τ (pn+1 ) = τ (p)τ (pn ) − p11 τ (pn−1 ) for p prime and n > 0,

(iii) ∣τ (p)∣ ≤ 2p11/2 for all primes.

The first two properties where proved by Mordell in 1917 (and we will prove them using
Hecke operators), while the third one was proved by Deligne in 1974 as a consequence of his
proof of the Weil conjectures.

116
Example 15.2. For examples of modular forms for Γ(N ) consider k ≥ 3 and g ∈ (Z/N Z)2 .
Define the Eisenstein series
′
1
Gk,g (τ ) ∶= ∑ k
(m,n)≡g (mod N ) (mτ + n)

for τ ∈ H. Where ∑′ means we are excluding (m, n) = (0, 0). Since this is a subseries of
an Eisenstein series, which is absolutely convergent and uniformly convergent on compact
subsets of H, it is also absolutely convergent and uniformly convergent in compact subsets
of H. Thus it is holomorphic on H. Moreover, if g = (a1 , a2 ) with a1 ≡/ 0 (mod N ), then using
the same bounds from Example 13.2 we see that

lim Gk,g (τ ) = 0.
τ →i∞

On the other hand, if a1 ≡ 0 (mod N ), then we see similarly that

′
1
lim Gk,g (τ ) = ∑ < ζ(k).
τ →i∞ k
n≡a2 (mod N ) n

In any case, Gk,g is holomorphic at i∞. Now let γ = ( ac db ) ∈ SL2 (Z) then

′ k
aτ + b 1
Gk,g ( ) = (cτ + d)k ∑ ( ) .
cτ + d (m,n)≡g (mod N ) (ma + nc)τ + mb + nd

Since the map (m, n) ∈ Z2 ∖ {(0, 0)} ↦ (m, n)γ ∈ Z2 ∖ {(0, 0)} is a bijection, this is equivalent
to
Gk,g (γ ⋅ τ ) = j(γ, τ )k Gk,gγ (τ )
and so
Gk,g [γ]k = Gk,gγ ∀γ ∈ SL2 (Z).
In consequence all Gk,g [γ]k are holomorphic at i∞. If further γ ∈ Γ(N ), then γ ≡ Id (mod
N ) and so gγ ≡ g (mod N ), thus

Gk,g [γ]k = Gk,g ∀γ ∈ Γ(N ),

i.e. Gk,g ∈ Mk (Γ(N )). In the case a1 ≡ 0 (mod N ), it follows that gγ ≡ g (mod N ) for all
γ ∈ Γ1 (N ) and so Gk,g ∈ Mk (Γ1 (N )).
Remark 15.3. In general for any congruence subgroup Γ of level N we can construct a
modular form of weight k ≥ 3 by taking
r
f ∶= ∑ Gk,gγj ∈ Mk (Γ)
j=1

where γ1 , . . . , γr are right coset representatives for Γ(N ) inside Γ.

117
15.2 The extended upper-half plane
Definition 15.2. Let Γ ⊆ SL2 (Z) be a congruence subgroup. We say that F ⊆ H is a
fundamental domain for Γ if it is a closed subset with connected interior such that

(i) any τ ∈ H is Γ-equivalent to a point in F,

(ii) no two interior points in F are Γ-equivalent,

(iii) ∂F is a finite union of smooth curves.

Proposition 15.3. The set

1
F = {x + iy ∈ H ∶ ∣x∣ ≤ , ∣x + iy∣ ≥ 1}
2
is a fundamental domain for SL2 (Z).

Proof We already proved it satisfies (i), and (iii) is trivial. Let us show (ii). Consider
τ, ω ∈ Int(F) such that ω = γ ⋅ τ for some
√
γ = ( ac db ) ∈√SL2 (Z). We can assume Im τ ≤ Im ω
and so ∣cτ + d∣ ≤ 1. Since τ ∈ F, Im τ ≥ 2 and so ∣c∣ 23 ≤ Im (cτ + d) ≤ 1. Hence c ∈ {0, ±1}.
3

If c = 0, then γ = ±T n for some n ∈ Z and since ω = τ + n ∈ F it follows that n = 0, i.e. ω = τ

and γ = ±Id. If c = ±1, it follows that writing τ = x + iy

∣τ ± d∣2 = (x ± d)2 + y 2 ≤ 1.
√
Since y ≥ 3/2 it follows that ∣x ± d∣ ≤ 1/2. Since ∣x∣ < 1/2 this forces d = 0 and so

a −1 1
ω = γ ⋅ τ = ±( ) ⋅ τ = ± (a − ) .
1 0 τ
1
Since ∣Re (−1/τ )∣ = ∣Re τ ∣/∣τ ∣2 < 2 it follows that a = 0, but ∣ − 1/τ ∣ = 1/∣τ ∣ < 1 contradicting
that ω ∈ F.

Proposition 15.4. Let Γ be a congruence subgroup with −Id ∈ Γ. If g1 , . . . , gr are left coset
representatives of Γ in SL2 (Z), then a fundamental domain for Γ is given by
r
FΓ = ⋃ gi−1 ⋅ F
i=1

if FΓ has connected interior.

Proof For every τ ∈ H we can find some γ ∈ SL2 (Z) such that γ ⋅ τ ∈ F. And we can also
find some γ ′ ∈ Γ such that γ = gi γ ′ . Thus

γ ′ ⋅ τ = gi−1 ⋅ (γ ⋅ τ ) ∈ gi−1 ⋅ F.

118
Now suppose there is some τ ∈ Int(FΓ ) such that γ ⋅ τ ∈ Int(FΓ ) for some γ ∈ Γ. Consider
U ⊆ Int(FΓ ) some open neighbourhood of τ such that γ ⋅ U ⊆ Int(FΓ ). We know there exists
some gi and gj such that
gi ⋅ τ, gj γ ⋅ τ ∈ F.
Let
V ∶= gi ⋅ U ∩ Int(F).
Take any x ∈ V , then
γgi−1 ⋅ x ∈ γgi−1 ⋅ V ⊆ γ ⋅ U ⊆ Int(FΓ ).
Let k be such that γgi−1 ⋅ x ∈ gk−1 ⋅ F. Hence x and gk γgi−1 ⋅ x belong to Int(F) and so

gk γgi−1 = ±Id.

Thus
gi Γ = gk γ(±Id)Γ = gk Γ
and so gi = gk . Therefore γ = ±Id, i.e. γ ⋅ τ = τ .
Example 15.3. Γ0 (2) has index 3 in SL2 (Z) and has coset representatives Id, S and ST .
Therefore a fundamental domain is

FΓ0 (2) = F ∪ S ⋅ F ∪ ST ⋅ F.

Definition 15.3. For every congruence subgroup Γ we define its associated modular curve

Y (Γ) ∶= Γ/H = Γ/FΓ .

We denote by simplicity Y (N ) ∶= Y (Γ(N )), Y0 (N ) ∶= Y (Γ0 (N )) and Y1 (N ) ∶= Y (Γ1 (N )).

Remark 15.4. The modular curve Y (Γ) has a natural structure of Riemann surface. In
terms of these Riemann surfaces we can reinterpret the weak modularity condition as follows:
A meromorphic function f ∶ H → C is weakly modular of weight k with respect to Γ if and
only if
f (γ ⋅ τ )j(γ, τ )−k = f (τ ) ∀τ ∈ H , ∀γ ∈ Γ.
Noting that
d(γ ⋅ τ ) = j(γ, τ )−2 dτ,
we can rephrase the modularity condition for k = 2 as

γ ∗ (f (τ )dτ ) = f (γ ⋅ τ )d(γ ⋅ τ ) = f (τ )dτ.

In other words, this is equivalent as saying that f (τ )dτ if Γ-invariant, and so defines a
meromorphic differential on Y (Γ). In general we can identify the space of weakly modular
forms of weight 2k for Γ with the space of meromorphic differentials of degree k on
Y (Γ)

Ω⊗k (Y (Γ)) ∶= {f (τ )dτ ⊗⋯⊗dτ ∶ f ∶ H → C meromorphic and weakly modular of weight 2k}.

119
We will denote the subspace of holomorphic differentials of degree k by

Ω⊗k
hol (Y (Γ)) ∶= {f (τ )dτ ⊗ ⋯ ⊗ dτ ∶ f ∶ H → C holomorphic and weakly modular of weight 2k}.

In order to get a geometric interpretation of modular forms, we need to consider also the
points at infinity, i.e. we have to compactify Y (Γ).
Definition 15.4. The extended upper-half plane is

H∗ ∶= H ∪ Q ∪ {i∞}.

The points Q ∪ {i∞} are called cusps. This can be made into a topological space by
considering the same fundamental system of neighbourhoods for each τ ∈ H, while for i∞
the fundamental system is given by the sets of the form

Uc ∶= {τ ∈ H ∶ Im τ > c} for c ∈ R.

For every s ∈ Q the fundamental neighbourhood system is given by the sets of the form

{s} ∪ Int(C)

where C is any disc contained in H tangent to the real axis at s.

Remark 15.5. The action of SL2 (Z) on H by fractional linear transformations extends
naturally to H∗ . In fact if c ≠ 0, then we define
a(i∞) + b a
∶= ∈ Q
c(i∞) + d c
and if c = 0 we put
a(i∞) + b
∶= i∞.
c(i∞) + d
Also for any s ∈ Q, if s ≠ −d/c we have
as + b
∈Q
cs + d
and for s = −d/c we put
as + b
= i∞.
cs + d
Definition 15.5. Let Γ be a congruence subgroup. The compact modular curve of Γ is

X(Γ) ∶= Γ/H∗ .

This curve is a compact Riemann surface containing Y (Γ) as an open subset.

Proposition 15.5. The compact modular curve X(Γ) is obtained by adding to Y (Γ) a
finite set of points. We call these points the cusps of X(Γ).

120
Proof Since SL2 (Z) acts transitively on Q ∪ {i∞} and Γ is a subgroup of finite index, it is
clear that the orbit space
Γ/(Q ∪ {i∞})
is finite and contained in the set {gi ⋅ (i∞) ∶ i = 1, . . . , r} where g1 , . . . , gr are right coset
representatives of Γ in SL2 (Z).
Remark 15.6. The Laurent expansion of f [γ]k corresponds to the Laurent expansion of
f around the cusp γ ⋅ (i∞). When f is weakly modular of weight k for some congruence
subgroup Γ, the order of f at a cusp s ∈ Q ∪ {i∞}, i.e. the order of zero or pole of q = 0 in
the q-expansion of f [γ]k for s = γ ⋅ (i∞), is invariant under the action of Γ (since in fact the
q-expansion is invariant under the action of Γ).
Definition 15.6. Let Γ be a congruence subgroup and s ∈ X(Γ) ∖ Y (Γ) be a cusp. The
order at s of a weakly modular function f ∶ H → C with respect to Γ is the order of f at
any point of Γ ⋅ s.
Remark 15.7. We will see that we can identify the space of cusp modular forms of weight
2 for Γ, with the space of holomorphic differentials on X(Γ), i.e.

S2 (Γ) ≃ Ω1hol (X(Γ)).

16 Modular Curves
16.1 Riemann surface structure of modular curves
Before going to the dimension formulas we need to give a more precise description of Riemann
surface structure of X(Γ). Let us start describing the structure of Y (Γ).
Proposition 16.1. Let Γ be a congruence subgroup. If τ ∈ H is such that

{γ ∈ Γ ∶ γ ⋅ τ = τ } ⊆ {±Id}

then the natural projection map π ∶ H → Y (Γ) is a local homeomorphism at τ . Thus a small
enough neighbourhood of τ gives us a chart around π(τ )

π −1 ∶ Vπ(τ ) → Uτ .

Proof It is enough to show that if τn → τ and γn ⋅ τn → τ for some γn ∈ Γ, then γn = ±Id for
some n ∈ N. We claim γn ⋅ τ → τ , in other words that γn ⋅ τn − γn ⋅ τ → 0. Write γn = ( acnn dbnn ),
then
an τ n + b n an τ + b n ∣τn − τ ∣
∣γn ⋅ τn − γn ⋅ τ ∣ = ∣ − ∣= .
cn τn + dn cn τ + dn ∣cn τn + dn ∣∣cn τ + dn ∣
Recalling that Z + τ Z ∩ B(0; 1) is finite and non empty we know there exists some

C = min{∣cτ + d∣ ∶ c, d ∈ Z , (c, d) ≠ (0, 0)}.

121
Similarly there exists

Cn = min{∣cτn + d∣ ∶ c, d ∈ Z , (c, d) ≠ (0, 0)}

and clearly Cn → C. Thus

γn ⋅ τn − γn ⋅ τ → 0
and so
γn ⋅ τ → τ.
Since in a small enough neighbourhood of τ there is no other point of its orbit, we get that
for n ≫ 0, γn ⋅ τ = τ and so γn = ±Id.

Corollary 16.2. Let Γ be a congruence subgroup. For every τ ∈ H there exists some open
neighbourhood Uτ such that

Uτ ∩ γ ⋅ Uτ = ∅ ∀γ ∉ Γτ .

Definition 16.1. Let Γ be a congruence subgroup. For every τ ∈ H we define its isotropy
group as
Γτ ∶= {γ ∈ Γ ∶ γ ⋅ τ = τ }.
We say τ ∈ H is an elliptic point for Γ if Γτ ⊆/ {±Id}. We also say that π(τ ) is an elliptic
point of Y (Γ).

Example 16.1. It is not hard to see that the only elliptic points of Y (1) = Y (SL2 (Z)) are
2πi
π(i) and π(ω) where ω = e 3 . In fact, their isotropy groups are cyclic of order 4 and 6
respectively
SL2 (Z)i = ⟨S⟩,
SL2 (Z)ω = ⟨ST ⟩.

Proposition 16.3. Let Γ be a congruence subgroup. Then Y (Γ) has finitely many elliptic
points, and Γτ is a finite cyclic group for all τ ∈ H.

Proof Let γ1 , . . . , γr be left coset representative of Γ in SL2 (Z). Let τ ∈ H be an elliptic

point for Γ, then Γτ ⊆/ {±Id}. Hence

Γτ ⊆ SL2 (Z)τ ⊆/ {±Id}

and so τ ∈ SL2 (Z) ⋅ i or τ ∈ SL2 (Z) ⋅ ω. Let us suppose that τ = γ ⋅ i for some γ ∈ SL2 (Z).
Then γ = γ ′ γj for some j ∈ {1, . . . , r}. Therefore τ = γ ′ ⋅ (γj ⋅ i), i.e.

π(τ ) ∈ {π(γj ⋅ i) ∶ j = 1, . . . , r}.

Analogously for ω, we conclude that Y (Γ) has only finitely many elliptic points. Since Γτ is
a subgroup of SL2 (Z)τ we conclude all isotropy groups are finite cyclic.

122
Definition 16.2. Let Γ be a congruence subgroup and τ ∈ H. The period of τ is defined as

#Γτ /2 if − Id ∈ Γτ ,
hτ ∶= [{±Id}Γτ ∶ {±Id}] = {
#Γτ if − Id ∉ Γτ .

It follows that τ is an elliptic point if and only if hτ ≠ 1.

Remark 16.1. It is easy to see that for any γ ∈ Γ and any τ ∈ H

Γγ⋅τ = γΓτ γ −1

and so
hγ⋅τ = [{±Id}γΓτ γ −1 ∶ {±Id}] = [{±Id}Γτ = {±Id}] = hτ .
Therefore we can define the period of an elliptic point of Y (Γ) as hπ(τ ) ∶= hτ .
Example 16.2. For SL2 (Z) we have hi = 2 and hω = 3. It follows that for every τ ∈ H and
every congruence subgroup Γ,
hτ ∈ {1, 2, 3}.
Remark 16.2. In order to introduce charts around elliptic points we proceed as follows. Let
τ ∈ H be an elliptic point for Γ of period hτ . Consider Uτ an open neighbourhood satisfying
Corollary 16.2. Consider the map
δτ ∶ Uτ → C
given by the linear fractional transformation which takes τ to 0 and τ to ∞, i.e.
z−τ
δτ (z) ∶= .
z−τ
Since the isotropy group of 0 in the conjugated transformation group is

(δτ {±Id}Γτ δτ−1 )0 /{±Id} = δτ ({±Id}Γτ /{±Id})δτ−1 ,

which is conjugated to the isotropy group of τ , hence it is finite and cyclic. Since the group
of fractional linear transformations fixing 0 and ∞ are rotations of the form z ↦ az, they
must be rotations around the origin of angle h2πτ . Therefore we construct the desired chart
given by
δτ z hτ
ψτ ∶ Uτ ∩ FΓ Ð→ C ÐÐ→ C.
These charts define the Riemann surface structure of Y (Γ). Our next task is to introduce a
Riemann surface structure on X(Γ).
Definition 16.3. Let Γ be a congruence subgroup and s ∈ Q∪{i∞} be a cusp. Let δ ∈ SL2 (Z)
be such that δ ⋅ s = i∞. The width of s is

hs ∶= [SL2 (Z)i∞ ∶ (δ{±Id}Γδ −1 )i∞ ],

where SL2 (Z)i∞ ∶= {γ ∈ SL2 (Z) ∶ γ ⋅ (i∞) = i∞} = ±⟨T ⟩ is the isotropy group of i∞ and for
any subgroup G ⊆ SL2 (Z), Gi∞ ∶= G ∩ SL2 (Z)i∞ .

123
Remark 16.3. It is easy to see that the width is independent f the choice of δ, in fact

{±Id}(δΓδ −1 )i∞ = ±⟨T hs ⟩ and hs = [SL2 (Z)s ∶ {±Id}Γs ]

where Γs ∶= {γ ∈ SL2 (Z) ∶ γ ⋅ s = s} is the isotropy group of s.

Remark 16.4. In order to introduce charts around cusps we proceed as follows. Let s ∈
Q ∪ {i∞} be a cusp of width hs for a congruence subgroup Γ. Take any δ ∈ SL2 (Z) such
that δ ⋅ s = i∞. By our previous remark, we can take the chart as follows. Let N2 ∶= {τ ∈ H ∶
Im τ > 2}, then consider Us ∶= δ −1 (N2 ∪ {i∞}) ∩ FΓ , then the chart is
2πiz
δ e hs
ψs ∶ Us Ð
→ H ÐÐÐ→ C.

With these charts X(Γ) becomes a compact Riemann surface.

Remark 16.5. It turns out that all compact Riemann surfaces can be embedded holomor-
phically inside PN , i.e. they are all projective manifolds. Furthermore, by Chow’s theorem,
this implies that all Riemann surfaces are in fact projective curves over C, i.e. they corre-
spond to the zero locus of homogeneous polynomials in PN . This fact enables us to apply
Serre’s GAGA principle, which roughly means that several algebraically defined concepts
and theorems holding for smooth projective curves over C have an analytic analog for Rie-
mann surfaces and conversely. This is the case, of finite maps, the degree of a map between
Riemann surfaces, the ramification index, divisors, the Picard group, the genus, Hurwitz for-
mula and Riemann-Roch theorem. In the case of the genus, it turns out that it corresponds
to the topological interpretation of genus given by
1
g(X) = rank H 1 (X, Z),
2
and also satisfies the Riemann-Roch theorem. Before moving further let us illustrate some
of these concepts. Given a holomorphic map f ∶ X → Y between compact Riemann surfaces,
it is always constant or surjective. For every y ∈ Y #f −1 (y) ≤ d for some natural d ≥ 1, with
equality attained for all but finitely many y ∈ Y . This number is the degree of f , and the
set of points where the fiber is smaller than the degree is called the branch locus. Outside
the branch locus, the map f defines a covering map with d sheets. For every x ∈ X we can
define the ramification index ex (f ) as the amount of different sheets of the unbranched
cover that pass through x. It follows that

∑ ex (f ) = d ∀y ∈ Y.
x∈f −1 (y)

Theorem 16.4 (Hurwitz formula, analytic version). Let f ∶ X → Y be a non-constant

holomorphic map between compact Riemann surfaces, of degree d. Then

2g(X) − 2 = d(2g(Y ) − 2) + ∑ (ex (f ) − 1).

x∈X

124
Proof Triangulate Y using VY vertices, including all the branch locus, EY edges and FY
faces. Lifting this triangulation by f we obtain a triangulation of X with EX = dEY ,
FX = dFY and VX = dVY − ∑x∈X (ex (f ) − 1). Therefore

χ(X) = 2−2g(X) = VX −EX +FX = d(VY −EY +FY )− ∑ (ex (f )−1) = d(2−2g(Y ))− ∑ (ex (f )−1).
x∈X x∈X

Example 16.3. Using Hurwitz formula we are able to compute the genus of modular curves
as follows. Let Γ1 ⊆ Γ2 be two congruence subgroups. The natural quotient map

f ∶ X(Γ1 ) → X(Γ2 )

is holomorphic. It is not hard to see that for τ ∈ Int(FΓ2 )

deg(f ) = #f −1 (τ ) = Γ1 /(Γ2 ⋅ τ ) = [{±Id}Γ2 ∶ {±Id}Γ1 ].

Now, for any τ ∈ H let h1 and h2 denote hτ with respect to Γ1 and Γ2 respectively. Let us
compute the ramification index of f at π1 (τ ) ∈ X(Γ2 ). Using the charts ψ1 and ψ2 around
π1 (τ ) and π2 (τ ) respectively, we translate f into
ψ1 z h2 /h1 ψ2−1
Uτ ∩ FΓ1 Ð→ C ÐÐÐ→ C ÐÐ→ Vτ ∩ FΓ2

and so
h2
∈ {1, 2, 3}.
eπ1 (τ ) (f ) =
h1
In consequence, π1 (τ ) is a ramification point if and only if τ is an elliptic point for Γ2 but
not for Γ1 . For s ∈ Q ∪ {i∞} we translate f into
ψ1 z h1 /h2 ψ2−1
Uτ ∩ FΓ1 Ð→ C ÐÐÐ→ C ÐÐ→ Vτ ∩ FΓ2

and so
h1
eπ2 (s) (f ) =
∈ {1, 2, 3}.
h2
Thus, the branch locus is always contained in the set of elliptic points and cusps. In the
particular case Γ1 = Γ and Γ2 = SL2 (Z) we get by Hurwitz formula that

2g(X(Γ)) − 2 = −2d + ∑ (ex (f ) − 1)

x∈X(Γ)

= −2d + ∑ (ex (f ) − 1) + ∑ (ex (f ) − 1) + ∑ (ex (f ) − 1).

x∈f −1 (i) x∈f −1 (ω) x∈f −1 (i∞)

If we denote 2 the amount of elliptic points of period 2, 3 the amount of elliptic points of
period 3 and ∞ the amount of cusps of X(Γ) we get

d= ∑ ex (f ) = 2(#f −1 (i) − 2 ) + 2 ,
x∈f −1 (i)

125
 d= ∑ ex (f ) = 3(#f −1 (i) − 3 ) + 3 .
x∈f −1 (ω)

Thus
1 2
2g − 2 = −2d + (#f −1 (i) − 2 ) + 2(#f −1 (ω) − 3 ) + d − ∞ = −2d + (d − 2 ) + (d − 3 ) + d − ∞
2 3
and so
d 2 3 ∞
g(X(Γ)) = 1 + − − − .
12 4 3 2

16.2 Automorphic forms

In order to compute the dimension formulas, we need to interpret the spaces of modular
forms in terms of certain Riemann-Roch spaces. For this we need to introduce the concept
of automorphic forms.
Definition 16.4. Let Γ be a congruence subgroup and let k ∈ Z. A meromorphic function
f ∶ H → C ∪ {∞} = P1 is called an automorphic form of weight k for Γ if f [γ]k = f for all
γ ∈ Γ, and f [γ]k is meromorphic at i∞ for all γ ∈ SL2 (Z). The set of automorphic forms of
weight k for Γ is denoted Ak (Γ).
Example 16.4. The modular j invariant is an automorphic form of wright 0 for SL2 (Z).
In fact
A0 (SL2 (Z)) = C(j).
This is not hard to see, since given any f ∈ A0 (SL2 (Z)) with non-SL2 (Z)-equivalent zeros
z1 , . . . , zk ∈ H, and non-SL2 (Z)-equivalent poles p1 , . . . , pl ∈ H we have

c ⋅ ∏li=1 (j − j(zi ))vzi (f )

f= .
∏h=1 (j − j(ph ))−vph (f )
l

Example 16.5. In general, for any congruence subgroup

A0 (Γ) ≃ C(X(Γ)).

Remark 16.6. Our purpose now is to relate A2k (Γ) to Ω⊗k (X(Γ)). In order to do this
let us associate to any f ∈ A2k (Γ) a meromorphic differential ω ∈ Ω⊗k (X(Γ)) such that
π ∗ (ω) = f (z)(dz)⊗k ∈ Ω⊗k (H). We will do this by charts. Let τ ∈ H with chart ψτ ∶ Uτ → C.
This chart can be factored as ψτ = ρ ○ δτ where δτ = ( 11 −τ hτ −1
−τ ) and ρ(z) = z . Let α = δτ and

λτ ∶= α∗ (f (z)(dz)k ) = f (α(z))(det(α))k j(α, z)−2k (dz)k = (f [α]2k )(z)(dz)k .

Since f (z)(dz)k if Γ-invariant, λτ must be δτ Γδτ−1 -invariant. Recalling that (δτ Γτ δτ−1 )0 is
2πi
generated by the rotation rhτ (z) ∶= e hτ z, we see that λτ must be invariant under the action
of the action of rhτ . Then
rh∗τ (λτ ) = λτ

126
in other words 2πi 2πik
f [α]2k (e hτ z)e hτ (dz)k = f [α]2k (z)(dz)k ,
hence
z k f [α]2k (z) = gτ (z hτ ).
We let
gτ (q)
ωτ (q) ∶= (dq)k
(hτ q)k
to get the desired meromorphic form. In particular we have

hτ ord0 (gτ ) = k + ordτ (f ).

2πiz
Now for a cusp s ∈ Q ∪ {i∞}, the chart is given by ψs = ρ ○ δ where δ ⋅ s = i∞ and ρ(z) = e hs .
2πiz
It follows that for α = δ −1 , f [α]2k has a q-expansion for q = e h , where h > 0 is the least
such that ( 10 h1 ) ∈ δΓδ −1 . Since the width hs is characterized by the equality

{±Id}δΓδ −1 = ±⟨( 10 h1s )⟩

we can have the following possibilities

δΓδ −1 ∈ {±⟨( 10 h1s )⟩, ⟨( 10 h1s )⟩, ⟨− ( 10 h1s )⟩}.

In the first two cases h = hs , while in the third case h = 2hs . Nevertheless, even when h = 2hs ,
since f is a modular form of weight 2k

f (τ + hs ) = f (τ + hs )(−1)2k = f [− ( 10 h1s )]2k (τ ) = f (τ ),

i.e. f is always hs Z-periodic. Therefore we can write

2πiz ords (f ) if h = hs ,
f [α]2k (z) = gs (e hs ), with ord0 (gs ) = { ords (f )
2 if h = 2hs .

The differential form is given by

gs (q)
ωs (q) = (dq)k .
(2πiq/hs )k
Theorem 16.5. Let k ∈ Z and Γ be a congruence subgroup. We have an isomorphism

A2k (Γ) ≃ Ω⊗k (X(Γ)).

Proof The map was defined in the previous remark. It is clearly surjective since every form
ω ∈ Ω⊗k (X(Γ)) pulls-back to a form

π ∗ ω = f (τ )(dτ )k ∈ Ω⊗k (H)

and f is automorphic of weight 2k. It is also clearly injective by definition.

127
17 Dimension Formulas
In view of the last theorem, we now have a natural way to define the order of an automorphic
form at every point of X(Γ). Let us first separate the two possible cases for cusps.
Definition 17.1. Let k ∈ Z and Γ be a congruence subgroup. We say that s ∈ Q ∪ {i∞}
(and also π(s) ∈ X(Γ) ∖ Y (Γ)) is an irregular cusp if

δΓδ −1 = ⟨− ( 10 h1s )⟩

for any δ ∈ SL2 (Z) such that δ ⋅ s = i∞. Otherwise we say s (and π(s)) is a regular cusp.
Remark 17.1. It follows from Theorem 16.5 that for every f ∈ A2k (Γ)
ordτ (f ) 1
ordτ (ω(f )) = + k ( − 1) ,
hτ hτ
ords (f ) − k if s is regular,
ords (ω(f )) = { ords (f )
2 − k if s is irregular.
This suggests how we should define the order of an automorphic form at every point of X(Γ).
Definition 17.2. Let Γ be a congruence subgroup and f ∈ Ak (Γ). We define the order of
f at π(τ ) ∈ Y (Γ) as
ordτ (f )
ordπ(τ ) (f ) ∶= ,
hτ
and the order of f at π(s) ∈ X(Γ) ∖ Y (Γ) as

ords (f ) if s is regular,
ordπ(s) (f ) ∶= { ords (f )
2 if s is irregular.

The Q-divisor associated to f is defined as

div(f ) ∶= ∑ ordP (f ) ⋅ P ∈ QDiv(X(Γ)) ∶= Div(X(Γ)) ⊗ Q.

P ∈X(Γ)

Remark 17.2. Recall that the divisor of any meromorphic differential of degree 1 is a
canonical divisor, and so has degree 2g − 2. Similarly it follows by Riemann-Roch that the
divisor of a meromorphic differential of degree k is k times a canonical divisor. Hence for
any f ∈ A2k (Γ)
2 23
k(2g − 2) = deg(div(ω(f ))) = deg(div(f )) − k ( + + ∞ )
2 3
and so by the genus formula of X(Γ) we get
dk k
deg(div(f )) = = [SL2 (Z) ∶ {±Id}Γ].
6 6
128
Remark 17.3. It follows from the last remark of the previous lecture, that given an auto-
morphic form of weight 0, f ∈ A0 (Γ) it determines a meromorphic function g ∈ C(X(Γ)) =
Ω⊗0 (X(Γ)). This function satisfies for every τ ∈ H
ordτ (f )
ordπ(τ ) (g) = ,
hτ
and for every s ∈ Q ∪ {i∞} irregular
ords (f )
ordπ(s) (g) = .
2
In particular, hτ ∣ordτ (f ) for all τ ∈ H and 2∣ords (f ) for all s ∈ Q ∪ {i∞} irregular.
Remark 17.4. In general div(f ) is not an element of Div(X(Γ)). In the special case f is
an automorphic form of weight 0, then
div(f ) = div(g) for g ∈ C(X(Γ))× ,
hence div(f ) ∈ Div0 (X(Γ)) and furthermore it is a principal divisor, i.e. div(f ) = 0 ∈
Pic0 (X(Γ)).
Definition 17.3. Let X be an algebraic variety and D = ∑ki=1 ri ⋅Pi ∈ QDiv(X) be a Q-divisor.
We define
k
⌊D⌋ ∶= ∑⌊ri ⌋ ⋅ Pi ∈ Div(X).
i=1

Proposition 17.1. For every k ∈ Z and every congruence subgroup Γ

A2k (Γ) ≠ 0.
Proof This can be proved in a more theoretical way using GAGA, noting that regarding
X(Γ) as an algebraic variety, Ω⊗k (X(Γ)) corresponds to the space of global rational sections
of the line bundle OX(Γ) (k ⋅ KX(Γ) ), which is algebraic and so always has rational sections.
We will also provide another proof constructing explicitly some f ∈ A2k (Γ) ∖ {0}. Note first
that since
A(Γ) ∶= ⊕ An (Γ)
n∈Z

has a natural structure of graded ring, i.e. that Ai (Γ) ⋅ Aj (Γ) ⊆ Ai+j (Γ), it is enough to
show that A2 (Γ) ≠ 0. Furthermore since A2 (SL2 (Z)) ⊆ A2 (Γ) we just need to prove it for
Γ = SL2 (Z). Let j ∈ A0 (SL2 (Z)) be the modular invariant. We claim
j ′ ∈ A2 (SL2 (Z)) ∖ {0}.
In fact, it is clearly meromorphic, non-zero, and for every γ ∈ SL2 (Z)
d d
(j ′ [γ]2 )(τ ) = j ′ (γ ⋅ τ )j(γ, τ )−2 = j ′ (γ ⋅ τ )(γ ⋅ τ )′ = (j(γ ⋅ τ )) = (j(τ )) = j ′ (τ ).
dτ dτ

129
Remark 17.5. For every k ∈ Z
C(X(Γ)) ⋅ f if ∃f ∈ Ak (Γ) ∖ {0},
Ak (Γ) ≃ {
0 otherwise.
This is clear since given two f, g ∈ Ak (Γ) ∖ {0}, then f /g ∈ A0 (Γ) ≃ C(X(Γ)).
Proposition 17.2. Let k ∈ Z, Γ be a congruence subgroup and f ∈ Ak (Γ) ∖ {0}. Then
Mk (Γ) ≃ L(⌊div(f )⌋).
Furthermore if Creg and Cirr denote the set of regular and irregular cusps respectively, then
letting C ∶= ∑cr ∈Creg cr + ∑ci ∈Cirr 21 ⋅ ci , then
Sk (Γ) ≃ L(⌊div(f ) − C⌋).
Proof Just note that
Mk (Γ) = {f0 ⋅ f ∶ f0 ∈ A0 (Γ) ∖ {0} and div(f0 ⋅ f ) ≥ 0} ∪ {0}
≃ {g0 ∈ C(X(Γ))× ∶ div(g0 ) + div(f ) ≥ 0} ∪ {0}
= {g0 ∈ C(X(Γ))× ∶ div(g0 ) + ⌊div(f )⌋ ≥ 0} ∪ {0} = L(⌊div(f )⌋).
Similarly
Sk (Γ) = {f0 ⋅ f ∶ f0 ∈ A0 (Γ) ∖ {0} and div(f0 ⋅ f ) ≥ C} ∪ {0}
≃ {g0 ∈ C(X(Γ))× ∶ div(g0 ) + div(f ) ≥ C} ∪ {0}
= {g0 ∈ C(X(Γ))× ∶ div(g0 ) + ⌊div(f ) − C⌋ ≥ 0} ∪ {0} = L(⌊div(f ) − C⌋).

Remark 17.6. Now we are in position to prove the dimension formulas using Riemann-Roch
theorem. Recall that Riemann-Roch says that for every divisor D ∈ Div(X(Γ))
`(D) − `(KX(Γ) − D) = deg D − g + 1,
where KX(Γ) is a canonical divisor, i.e. a divisor linearly equivalent to div(ω) for some mero-
morphic differential form ω ∈ Ω1 (X(Γ)). In particular, if we take D = 0 we get `(KX(Γ) ) = g,
and so taking D = KX(Γ) we get deg(KX(Γ) ) = 2g − 2. Furthermore, since for any divisor D
with `(D) > 0 there exists some f ∈ C(X(Γ))× such that f ∈ L(D), i.e.
div(f ) + D ≥ 0
hence
deg(D) = deg(div(f ) + D) ≥ 0.
In other words for every divisor with deg(D) < 0, we have `(D) = 0. Therefore if D is a
divisor such that deg(D) > 2g − 2, then deg(KX(Γ) − D) < 0 and so
`(D) = deg(D) − g + 1.
We will use all these facts in the proof of the dimension formulas.

130
Theorem 17.3. Let k ∈ Z and Γ be a congruence subgroup. Let g be the genus of X(Γ), 2
be the number of elliptic points of period 2, 3 be the number of elliptic points of period 3
and ∞ be the number of cusps. Then
⎧
⎪
⎪ (2k − 1)(g − 1) + ⌊ k2 ⌋2 + ⌊ 2k
3 ⌋3 + k∞ if k ≥ 1,
⎪
dimC M2k (Γ) = ⎨ 1 if k = 0,
⎪
⎪
⎪
⎩ 0 if k < 0,
and
⎧
⎪
⎪ (2k − 1)(g − 1) + ⌊ k2 ⌋2 + ⌊ 2k
3 ⌋3 + (k − 1)∞ if k ≥ 2,
⎪
dimC S2k (Γ) = ⎨ g if k = 1,
⎪
⎪
⎪
⎩ 0 if k < 1.
Proof Let E2 be the set of elliptic points of X(Γ) of period 2 and E3 be the set of elliptic
points of X(Γ) of period 3. Note first that
k 2k
div(f ) = div(ω(f )) + ∑ ⋅ e2 + ∑ ⋅ e3 + ∑ k ⋅ s,
e2 ∈E2 2 e3 ∈E3 3 s∈X(Γ)∖Y (Γ)

hence
k 2k
⌊div(f )⌋ = div(ω(f )) + ∑ ⌊ ⌋e2 + ∑ ⌊ ⌋e3 + ∑ k ⋅ s.
e2 ∈E2 2 e3 ∈E3 3 s∈X(Γ)∖Y (Γ)

If k ≥ 1
k 2k
deg(⌊div(f )⌋) = k(2g − 2) + ⌊ ⌋2 + ⌊ ⌋3 + k∞ > 2g − 2
2 3
and so by Riemann-Roch theorem
k 2k
`(⌊div(f )⌋) = deg(⌊div(f )⌋) − g + 1 = (2k − 1)(g − 1) + ⌊ ⌋2 + ⌊ ⌋3 + k∞ .
2 3
Similarly for k ≥ 2
k 2k
⌊div(f ) − C⌋ = div(ω(f )) + ∑ ⌊ ⌋e2 + ∑ ⌊ ⌋e3 + ∑ (k − 1) ⋅ s
e2 ∈E2 2 e3 ∈E3 3 s∈X(Γ)∖Y (Γ)

hence
k 2k
deg(⌊div(f ) − C⌋) = k(2g − 2) + ⌊ ⌋2 + ⌊ ⌋3 + (k − 1)∞ > 2g − 2
2 3
and so
k 2k
`(⌊div(f ) − C⌋) = deg(⌊div(f ) − C⌋) − g + 1 = (2k − 1)(g − 1) + ⌊ ⌋2 + ⌊ ⌋3 + (k − 1)∞ .
2 3
For k = 1, ⌊div(f ) − C⌋ is a canonical divisor and so `(⌊div(f ) − C⌋) = g. For k = 0,
⌊div(f )⌋ is a principal divisor and ⌊div(f ) − C⌋ has negative degree, thus `(⌊div(f )⌋) = 1 and
`(⌊div(f ) − C⌋) = 0. Finally for k < 0, using the genus formula for X(Γ) we get
k 2k k
deg(⌊div(f )⌋) ≤ k(2g − 2) + 2 + 3 + k∞ = [SL2 (Z) ∶ {±Id}Γ] < 0
2 3 6
and so `(⌊div(f )⌋) = 0 and `(⌊div(f ) − C⌋) = 0.

131
Remark 17.7. Now we have to find dimension formulas for modular forms of odd weight.
First let us prove that
Mk (Γ) = 0 for k < 0.
This proof works also for k odd. Suppose f ∈ Mk (Γ), then f 12 ⋅ ∆−k ∈ S0 (Γ) = {0} and so
f = 0. Another elementary fact is that for k odd

Mk (Γ) = 0 if − Id ∈ Γ.

In fact, given f ∈ Mk (Γ)

f = f [−Id]k = −f = 0.
Thus we just need to compute the dimension for k odd positive and for −Id ∉ Γ. Since the
unique subgroup of size 2 of ⟨S⟩ and ⟨ST ⟩ is {±Id}, it follows that if −Id ∉ Γ, then X(Γ)
has no elliptic points of period 2.

Theorem 17.4. Let k be a positive odd integer. Let Γ be a congruence subgroup such that
−Id ∉ Γ. Let g be the genus of X(Γ), 3 be the number of elliptic points of period 3, reg
∞ the
number of regular cusps, and irr
∞ the number of irregular cusps. Then
⎧
⎪
⎪ (k − 1)(g − 1) + ⌊ k3 ⌋3 + k2 reg k−1 irr
∞ + 2 ∞ if k ≥ 3,
⎪
dimC Mk (Γ) = ⎨ reg
∞ /2 if k = 1 and reg
∞ > 2g − 2,
⎪
⎪ reg
⎪
⎩ ≥ ∞ /2 otherwise,
reg
(k − 1)(g − 1) + ⌊ k3 ⌋3 + k−2 k−1 irr
2 ∞ + 2 ∞ if k ≥ 3,
dimC Sk (Γ) = { reg
dimC M1 (Γ) − ∞ /2 if k = 1.

Proof Let f ∈ Ak (Γ) ∖ {0}, then f 2 ∈ A2k (Γ). It follows that

1 1 k k
div(f ) = div(f 2 ) = div(ω(f 2 )) + ∑ ⋅ e3 + ∑ ⋅c
2 2 e3 ∈E3 3 c∈X(Γ)∖Y (Γ) 2

where E3 is the set of elliptic points of period 3. In order to compute ⌊div(f )⌋ we have to
analyze the orders of ω(f 2 ). For a point p ∈ Y (Γ) ∖ E3 it is clear that
1
ordp (ω(f 2 )) = ordp (f ) ∈ Z
2
by definition. If e3 ∈ E3 , then
1 1 2 k
orde3 (ω(f 2 )) = (orde3 (f 2 ) − k) = orde3 (f ) −
2 2 3 3

thus 32 orde3 (ω(f 2 )) ∈ Z, hence 12 orde3 (ω(f 2 )) ∈ Z and so

1 k
⌊orde3 (f )⌋ = orde3 (ω(f 2 )) + ⌊ ⌋.
2 3

132
For cr ∈ X(Γ) ∖ Y (Γ) a regular cusp
1 1 k
ordcr (ω(f 2 )) = (ordcr (f 2 ) − k) = ordcr (f ) −
2 2 2
and since ordcr (f ) ∈ Z it follows that
1 k
⌊ordcr (f )⌋ = ordcr (f ) = ordcr (ω(f 2 )) + .
2 2
Finally for ci ∈ X(Γ) ∖ Y (Γ) an irregular cusp
1 1 k
ordci (ω(f 2 )) = (ordci (f 2 ) − k) = ordci (f ) −
2 2 2
but now ordci (f ) is half an integer hence 21 ordci (ω(f 2 )) ∈ Z and so
1 k−1
⌊ordci (f )⌋ = ordci (ω(f 2 )) + .
2 2
Therefore
1 k k k−1
⌊div(f )⌋ = div(ω(f 2 )) + ∑ ⌊ ⌋e3 + ∑ ⋅ cr + ∑ ⋅ ci ,
2 e3 ∈E3 3 cr ∈Creg 2 ci ∈Cirr 2
and
1 k k−2 k−1
⌊div(f ) − C⌋ = div(ω(f 2 )) + ∑ ⌊ ⌋e3 + ∑ ⋅ cr + ∑ ⋅ ci ,
2 e3 ∈E3 3 cr ∈Creg 2 ci ∈Cirr 2
where Creg and Cirr denote the set of regular and irregular cusps respectively. Let us compute
now
1 k k k − 1 irr
deg(⌊div(f )⌋) = k(2g − 2) + ⌊ ⌋3 + reg ∞ +  .
2 3 2 2 ∞
For k ≥ 3
deg(⌊div(f )⌋) > 2g − 2 + reg
∞ + ∞
irr

and so by Riemann-Roch
k k k − 1 irr
`(⌊div(f )⌋) = (k − 1)(g − 1) + ⌊ ⌋3 + reg
∞ +  ,
3 2 2 ∞
k k − 2 reg k − 1 irr
`(⌊div(f ) − C⌋) = (k − 1)(g − 1) + ⌊ ⌋3 +  +  .
3 2 ∞ 2 ∞
Finally for k = 1
deg(⌊div(f )⌋) = g − 1 + reg
∞ /2

then by Riemann-Roch
`(⌊div(f )⌋) = reg reg
∞ /2 + `(KX(Γ) − ⌊div(f )⌋) ≥ ∞ /2

with equality if reg

∞ > 2g − 2. Also KX(Γ) − ⌊div(f )⌋ ∼ ⌊div(f ) − C⌋ hence

`(⌊div(f ) − C⌋) = `(⌊div(f )⌋) − reg

∞ /2.

133
17.1 The Abel-Jacobi map
The only missing fact we need to complete the proof of the dimension formula for odd weight
is to show that Ak (Γ) ≠ {0} for k odd positive and −Id ∉ Γ. In order to prove this we will
use a theorem from Riemann surface theory regarding to the Abel-Jacobi map. This map is
a higher genus analog of the period map defined in Proposition 13.7 for elliptic curves.

Remark 17.8. For any compact Riemann surface X of genus g, we know by Riemann-Roch
that
`(KX ) = g.
On the other hand, taking any meromorphic differential ω ∈ Ω1 (X), we can identify

L(KX ) = {f ∈ C(X) ∶ div(f ⋅ ω) ≥ 0} ∪ {0} ≃ Ω1hol (X)

f z→ f ⋅ ω.
Therefore dimC Ω1hol (X) = g, hence we can take a basis of holomorphic differentials ω1 , . . . , ωg .

Theorem 17.5 (Abel-Jacobi theorem). Let X be a compact Riemann surface of genus

g, with basis of holomorphic differentials given by ω1 , . . . , ωg . Let δ1 , . . . , δ2g be a basis for
H1 (X, Z). Then the periods of X

pi ∶= (∫ ω1 , . . . , ∫ ωg ) ∈ Cg
δi δi

for i = 1, . . . , 2g are R-linearly independent, and so they define a lattice

2g
Λ ∶= ⊕ pi Z ⊆ Cg .
i=1

Furthermore, fixing a point O ∈ X, the map

AJ ∶ X → Cg /Λ
P P
P ↦ (∫ ω1 , . . . , ∫ ωg )
O O

is a well defined holomorphic map, such that for every P, Q ∈ X, AJ(P ) = AJ(Q) if and
only if P ∼ Q. Thus it induces a map

AJ ∶ Pic0 (X) → Cg /Λ

which is in fact an isomorphism of groups.

Proposition 17.6. Let k be a positive odd integer and Γ be a congruence subgroup such
that −Id ∉ Γ. Then
Ak (Γ) ≠ {0}.

134
Proof Let λ be any non-zero element of Ω1 (X(Γ)), e.g. the differential associated to some
non-zero automorphic form of weight 2. Fix any point O ∈ X(Γ). Then

div(λ) − (2g − 2) ⋅ O ∈ Div0 (X(Γ)).

By Abel-Jacobi theorem we know there exists some D ∈ Div0 (X(Γ)) such that

AJ(D) = AJ(div(λ) − (2g − 2) ⋅ O)/2,

and so there exists some f ∈ C(X(Γ))× such that

2D = div(λ) − (2g − 2) ⋅ O + div(f ).

Hence
div(f ⋅ λ) = 2(D + (g − 1) ⋅ O).
Let π ∗ (f ⋅ λ) = f˜(τ )dτ be the differential pulled-back to H. Then by similar computations
as before
2
div(f˜) = div(f ⋅ λ) + ∑ e3 + ∑ cr + ∑ ci .
e3 ∈E3 3 cr ∈Creg ci ∈Cirr

In particular, f̃ has even order at all points in H (since for τ ∈ H elliptic point, ordτ (f̃) =
3ordπ(τ ) (f̃) = 3ordπ(τ ) (f ⋅ λ) + 2). Therefore we can find some meromorphic function f1 such
that f12 = f̃. Since f̃ is weight 2 invariant for Γ, for every γ ∈ Γ

f1 [γ]1 = ±f1 = χ(γ)f1 ,

for some character χ ∶ Γ → {±1}. Let

Γ′ ∶= {γ ∈ Γ ∶ χ(γ) = 1}.

This is a subgroup of Γ of index at most 2. If Γ = Γ′ we have found f1 ∈ A1 (Γ). Otherwise

Γ = Γ′ ∪ αΓ′ where f1 [α]1 = −f1 . Since Γ′ has index 2, it induces 2-1 map

X(Γ′ ) → X(Γ).

This map corresponds to a quadratic extension of function fields C(X(Γ′ ))/C(X(Γ)), i.e.
there exists some f2 ∈ C(X(Γ′ )) ∖ C(X(Γ)), such that f22 ∈ C(X(Γ)). Hence (α∗ (f2 ))2 = f22
and so α∗ f2 = ±f2 . Since f2 ∉ C(X(Γ)) we conclude that α∗ f2 = −f2 and so

(f1 f2 )[α]1 = α∗ f1 ⋅ α∗ f2 ⋅ j(α, τ )−2 = (−f2 ) ⋅ (f1 [α]1 ) = f1 f2 ,

i.e. f1 f2 ∈ A1 (Γ) ∖ {0}.

18 Hecke Operators
In this lecture we will introduce Hecke operators. We will start with Hecke operators of level
1 and the proof of the first two Ramanujan conjectures.

135
18.1 Ramanujan conjectures
In order to define Hecke operators let us denote by Xm the set of 2 × 2 matrices with integer
entries and determinant m ∈ N. The group SL2 (Z) acts on Xm by the left. The following
proposition describes the orbits of this action.

Proposition 18.1. The number of orbits of the action of SL2 (Z) on Xm is σ1 (m), and a
set of representatives is
a b
{( ) ∶ ad = m, 0 ≤ b ≤ d − 1} .
0 d

Proof Given any matrix ( αγ βδ ) ∈ Xm , let a ∶= gcd(α, β), r ∶= α/a and t ∶= γ/a. Since
gcd(r, t) = 1 there exists s, u ∈ Z such that ru − st = 1. Then
−1
r s α β a βu − δs a b
( ) ( )=( )=( ).
t u γ δ 0 −βt + δr 0 d

This gives a representative with ad = m and d > 0. To get a representative with 0 ≤ b < d
note that
1 −k a b a b − kd
( )( )=( )
0 1 0 d 0 d
and so we can find a representative where b is replaced by the remainder of the division of b
by d. To see that these representatives are different, suppose

r s a b a′ b′
( )( )=( ).
t u 0 d 0 d′

Then at = 0, hence t = 0. Also d′ = du and a′ = ar, then m = a′ d′ = adru = mru and so

r = u = ±1. Since a′ , a > 0 it follows that r = u = 1. Finally sd ≤ b + sd = b′ < d′ = d hence s ≤ 0
and since −sd ≤ b′ − sd = b < d it follows that −s ≤ 0, and so s = 0.

Definition 18.1. For any matrix γ ∈ GL+2 (R) we define the weight k operator as

(f [γ]k )(τ ) ∶= f (γ ⋅ τ )j(γ, τ )−k (det(γ)) 2 .

k

Definition 18.2. Let m ≥ 1 and k ≥ 2 be integers. The m-th Hecke operator

Tm ∶ M2k (SL2 (Z)) → M2k (SL2 (Z))

is defined as
d−1
Tm (f ) ∶= mk−1 ∑ ∑ f [( a0 db )]2k , ad = m.
d∣m b=0

Proposition 18.2. For every f ∈ M2k (SL2 (Z)), Tm (f ) is weakly modular of weight 2k.

136
Proof Let γ ∈ SL2 (Z), then
d−1
k−1
Tm (f )[γ]2k = m ∑ ∑ f [( a0 db ) γ]2k
d∣m b=0

Since ( a0 db ) γ ∈ Xm for each term of the sum there exists some γ ′ ∈ SL2 (Z) such that
( a0 db ) γ = γ ′ ( a0
′ b′ )
d′ .

We claim these new representatives are all different, in fact if for some other matrix
( a01 b1 ′ b′ )
d1 ) γ = γ1 ( a0 d′ ,

then
( a01 b1
= γ1 (γ ′ )−1 ( a0 db ) γ
′ b′ )
d1 ) γ = γ1 ( a0 d′

and so
( a01 b1
d1 ) = γ1 (γ ′ )−1 ( a0 db ) .
Therefore by Proposition 18.1 ( a01 b1
d1 ) = ( a0 db ). Since

f [( a0 db ) γ]2k = f [γ ′ ( a0 = (f [γ ′ ]2k )[( a0

′ b′ )] ′ b′ )] ′ b′ )]
d′ 2k d′ 2k = f [( a0 d′ 2k

the result follows.

In order to show that the Hecke operators are well-defined we have to compute the
q-expansions. For this we need the following preliminary fact.
Lemma 18.3. Let n ≥ 0 and d > 0 be integers. Then
d−1
2πinb d if d∣n,
∑e d ={
b=0
0 otherwise.

Proof If d∣n, the sum is d times 1. Otherwise

2πin
d−1
2πinb (e d )d − 1
∑e d = 2πin = 0.
b=0 e d −1

Theorem 18.4. If f ∈ M2k (SL2 (Z)) has q-expansion

∞
f = ∑ λ(n)q n ,
n=0

then
∞ ⎛ mn ⎞
Tm (f ) = ∑ ∑ d2k−1 λ ( 2 ) q n .
n=0 ⎝d∣ gcd(m,n) d ⎠
In particular Hecke operators are well-defined and take S2k (SL2 (Z)) to itself.

137
Proof By definition
d−1 d−1 m
k−1 k−1 dτ+ b −2k k
(Tm (f ))(τ ) = m ∑∑f [( a0 db )]2k (τ ) = m ∑∑f( )d m
d∣m b=0 d∣m b=0 d

1 m 2k d−1 ∞
∑ ( ) ∑ ∑ λ(n)e2πin( d τ +b)/d
m
=
m d∣m d b=0 n=0

1 m 2k ∞ d−1
∑ ( ) ∑ ∑ λ(n)e2πin( d τ +b)/d
m
=
m d∣m d n=0 b=0

m 2k−1 ∞ m
= ∑( ) ∑ λ(rd)e2πir d τ
d∣m d r=0
∞
m 2πirdτ
= ∑ ∑ d2k−1 λ (r )e
r=0 d∣m d
∞
mn 2πinτ
=∑ ∑ d2k−1 λ ( )e .
n=0 d∣ gcd(m,n) d2

Corollary 18.5. For every m ∈ N

Tm (∆) = τ (m) ⋅ ∆.

In consequence
mn
τ (m)τ (n) = ∑ d11 τ ( ).
d∣ gcd(m,n) d2
In particular for gcd(m, n) = 1
τ (m)τ (n) = τ (mn).
Proof Since Tm (∆) ∈ S12 (SL2 (Z)) = C ⋅ ∆, there exists some λm ∈ C such that

Tm (∆) = λm ⋅ ∆.

In order to compute λm we have to compute the first Fourier coefficient of Tm (∆), thus
m
λm = ∑ d11 τ ( ) = τ (m).
d∣ gcd(m,1) d2

Therefore, looking at the n-th term we get

mn
∑ d11 τ ( ) = τ (m) ⋅ τ (n).
d∣ gcd(m,n) d2

138
Corollary 18.6. For p prime and n ∈ N

τ (pn+1 ) = τ (pn )τ (p) − p11 τ (pn−1 ).

Proof Using the previous corollary

pn+1
τ (pn )τ (p) = ∑ d11 τ ( ) = p11 τ (pn−1 ) + τ (pn+1 ).
d∣ gcd(pn ,p) d2

Remark 18.1. The previous properties for the Ramanujan τ function followed essentially by
the fact that dimC S12 (SL2 (Z)) = 1. This fact is also true for cusp forms of weight 16, 18, 20,
∆ ∆
22 and 26. In consequence the Fourier coefficients of the modular forms E4 (2π) 12 , E6 (2π)12 ,
∆ ∆ ∆
E8 (2π) 12 , E10 (2π)12 and E14 (2π)12 are also multiplicative, and satisfy the same relation of the

previous corollary with p2k−1 instead of p11 for each weight 2k = 16, 18, 20, 22, 26 respectively.
In fact, for any cusp form that is an eigenvector for all Hecke operators we will have similar
properties for its Fourier coefficients.

Definition 18.3. A modular form f ∈ M2k (SL2 (Z)) is called a Hecke eigenform it it is
an eigenvector for all Hecke operators.

Theorem 18.7. Hecke operators commute, i.e.

Tn Tm = Tm Tn .

Furthermore, when gcd(m, n) = 1

Tm Tn = Tmn .
And for every prime p and s ∈ N

Tps+1 = Tps Tp − p2k−1 Tps−1 .

Proof Let us show first the second assertion. Let gcd(m, n) = 1 and
∞
f = ∑ λ(n)q n ∈ M2k (SL2 (Z)).
n=0

Then the r-th Fourier coefficient of Tm (Tn (f )) is

mr nmr
∑ d2k−1 c ( 2
) = ∑ d2k−1 ∑ e2k−1 λ ( 2 2 )
d∣ gcd(m,r) d d∣ gcd(m,r) e∣ gcd(n, mr ) de
d2

nmr mnr
= ∑ ∑ (de)2k−1 λ ( 2
)= ∑ h2k−1 λ ( 2 )
d∣ gcd(m,r) e∣ gcd(n,r) (de) h∣ gcd(mn,r) h

139
which is the r-th Fourier coefficient of Tmn (f ). This proves the second assertion, which in
turn reduces the theorem to show that for any prime p and any natural numbers m, n ∈ N

Tpn Tpm = Tpm Tpn .

Once we prove the third assertion, it will follow inductively that for every n ∈ N, Tpn = Pn (Tp )
for some polynomial Pn ∈ Z[x], thus

Tpn Tpm = (Pn ⋅ Pm )(Tp ) = Tpm Tpn

as desired. Hence we just need to show the third assertion, which is the following computation
d−1
ps
Tps (Tp (f )) = ps(k−1) ∑ ∑ (Tp (f ))[( d
b )]
2k
0 d
d∣ps b=0

s pt −1 p−1
s−t
=p s(k−1)
∑∑p k−1
(f [( p0 01 )]2k + ∑ f [( 10 pc )]2k )[( p 0 b
pt
)]2k
t=0 b=0 c=0

s pt −1 s pt −1 p−1
(s+1)(k−1) s+1−t
(s+1)(k−1) ps−t b+cpt
=p ∑∑ f [( p 0 bp pt
)]2k +p ∑ ∑ ∑ f [( 0 pt+1
)]2k
t=0 b=0 t=0 b=0 c=0

s pt−1 −1 s pt+1 −1
(s+1)(k−1) ps−t b
f [( p 0 01 )]2k +p(s+1)(k−1)+1 ∑ ∑ ps+1−t bp (s+1)(k−1)
s+1
=p f [( 0 pt )]2k +p ∑ ∑ f [( 0 pt+1 )]2k
t=1 b=0 t=0 b=0

s pt−1 −1 s−t
2k−1 (s−1)(k−1) p b
= Tps+1 (f ) + p p ∑ ∑ f [( 0 pt−1
)]2k
t=1 b=0

s−1 pt −1
2k−1 (s−1)(k−1) p s−1−t b
= Tps+1 (f ) + p p ∑ ∑ f [( 0 pt
)]2k
t=0 b=0
2k−1
= Tps+1 (f ) + p Tps−1 (f ).

18.2 The Petersson inner product

In order to prove the existence of Hecke eigenforms of any weight we will introduce now an
L2 inner product on the space of cusp forms. For this we need a measure on H invariant
under the action of SL2 (Z).

Proposition 18.8. The measure given by

dxdy
dµ ∶=
y2

140
on H is invariant under the action of SL2 (R) on H by fractional linear transformations. For
every congruence subgroup Γ, every fundamental domain FΓ and every continuous bounded
function ϕ ∶ H → C
∫ ϕ dµ ∈ C. FΓ

If further ϕ is invariant under the action of Γ, then the above integral is independent of the
choice of FΓ .

Proof The invariance of the measure is a tedious but straightforward count. For the second
statement it is enough to show that

∣∫ ϕ dµ∣ ≤ C ∫ dµ < +∞.

F Γ F Γ

Noting that FΓ is a finite union of fundamental domains of SL2 (Z) it is enough to show that
for any γ ∈ SL2 (Z)
∫γ⋅F dµ = ∫F d(µ ○ γ) = ∫F dµ < +∞.
and this is clear since
∞ ∞ dy −1 ∞
1
2 dxdy 2
∫F dµ ≤ ∫ √3 ∫ −1 y 2
= ∫ √
3
y 2
= ∣√ = √ .
y 3
2 2 2 2
3

Finally if ϕ is Γ-invariant, it follows for every γ ∈ Γ that

∫γ⋅F ϕ dµ = ∫F ϕ ○ γ d(µ ○ γ) = ∫F ϕ dµ.

Γ Γ Γ

Definition 18.4. Let Γ be a congruence subgroup with fundamental domain FΓ . The

Petersson inner product on Sk (Γ) is defined as
1
(f, g) ∶= (Im z)k f (z)g(z)dµ(z).
[SL2 (Z) ∶ {±Id}Γ] ∫F

Remark 18.2. The Petersson inner product is well-defined by the previous proposition since
the function (Im z)k f (z)g(z) is bounded and Γ-invariant. To see it is bounded note that
since f is a cusp form we can write

f (z) = e2πiz f̃(z)

thus limIm z→+∞ ∣(Im z)k e2πiz f̃(z)g(z)∣ = 0. It is also clear that the Petersson inner product
is a Hermitian inner product on Sk (Γ).

141
Remark 18.3. Let Γ be a congruence subgroup and let α ∈ GL+2 (Q) with integral entries.
Put Γ′ = α−1 Γα ∩ Γ, then Γ′ is a congruence subgroup since if Γ(N ) ⊆ Γ and d = det(α), then
Γ(dN ) ⊆ Γ′ . Furthermore the map
f ↦ f [α]k
takes Mk (Γ) to Mk (Γ′ ) and Sk (Γ) to Sk (Γ′ ). In fact, this is a consequence of the fact that
every such matrix α can be written as

a b
α=γ( )
0 d

for some γ ∈ SL2 (Z) and a, b, d ∈ Z, a, d > 0. Thus its q-expansion is of the form (where
γ ⋅ s = i∞)
∞ 2πinτ a b
f [α]2k = ( ∑ an e hs ) [( )]
n=0 0 d 2k
a k ∞ 2πin(aτ +b)
= ( ) ∑ an e dhs
d n=0
a k ∞ 2πinb
= ( ) ∑ an e dhs q nac .
d n=0
Theorem 18.9. Let f, g ∈ Sk (Γ) and α ∈ GL+2 (Q) with integral entries, then

(f, g) = (f [α]k , g[α]k ).

Proof Let F ′ be a fundamental domain for Γ′ = α−1 Γα ∩ Γ. Then

1
(f [α]k , g[α]k ) = (Im z)k f [α]k (z)g[α]k (z)dµ(z)
[SL2 (Z) ∶ {±Id}Γ′ ] ∫F ′
1
= ′ ∫ (Im z)k f (z)g(z)dµ(z).
[SL2 (Z) ∶ {±Id}Γ ] α⋅F ′

Let Γ′′ ∶= αΓα−1 ∩ Γ. Since {±Id}Γ′′ = α−1 {±Id}Γ′ α, it has the same index than {±Id}Γ′ .
Also α ⋅ F ′ is a fundamental domain for Γ′′ . Therefore there exist g1 , . . . , gr ∈ Γ such that
r = [{±Id}Γ ∶ {±Id}Γ′′ ] and α ⋅ F ′ = ∪ri=1 gi ⋅ F for some fundamental domain F of Γ. Hence
r
1
(f [α]k , g[α]k ) = ′′ ∑ ∫ (Im z)k f (z)g(z)dµ(z)
[SL2 (Z) ∶ {±Id}Γ ] i=1 gi ⋅F
r
= (Im z)k f (z)g(z)dµ(z) = (f, g).
[SL2 (Z) ∶ {±Id}Γ′′ ] ∫F

Theorem 18.10. For any f, g ∈ S2k (SL2 (Z))

(Tn (f ), g) = (f, Tn (g)).

142
Proof By Theorem 18.7 we can reduce ourselves to the case n = p prime. In this case we
have
p−1
(Tp (f ), g) = pk−1 (f [( p0 10 )]2k + ∑ f [( 10 pc )]2k , g)
c=0
p−1
= pk−1 (f [( p0 p0 )]2k , g[( 10 p0 )]2k ) + pk−1 ∑ (f [( p0 p0 )]2k , g[( p0 −c
1 )]2k )
c=0
p−1
= pk−1 (f, g[( 10 p0 )]2k ) + pk−1 (f, g[( p0 10 )]2k ) + pk−1 ∑ (f, g[( p0 −c
1 )]2k ).
c=1
Since ( 01 −1 0 1
c ) , ( −1 −c ) ∈ SL2 (Z) and
0 −1 p −c 0 1 1 c
( )( )( )=( ).
1 c 0 1 −1 −c 0 p
We get that
(f, g[( p0 −c 1 c
1 )]2k ) = (f, g[( 0 p )]2k )
and so
(Tp (f ), g) = (f, Tp (g)).

Corollary 18.11. The space S2k (SL2 (Z)) has an orthogonal basis of Hecke eigenforms.
Every Hecke eigenform f ∈ S2k (SL2 (Z)) has first Fourier coefficient a1 ≠ 0, and an /a1 is an
eigenvalue of Tn . Furthermore, all the eigenvalues of the Hecke operators are real algebraic
numbers, and every pair of normalized eigenforms are orthogonal.
Proof It follows by a well-known theorem of linear algebra that given a family of commuting
self-adjoint operators on a finite dimensional vector space S2k (SL2 (Z)) with a Hermitian
inner product, then it has an orthogonal basis of eigenvectors for all operators simultaneously.
Since the Hecke operators are self-adjoint their eigenvalues are all real. Furthermore, we know
that S2k (SL2 (Z)) admits a basis of cups forms with integral Fourier coefficients (difference
of two products of powers of E4 and E6 ), hence writing each Hecke operator in this basis
we see that the characteristic polynomial has rational coefficients (note that each Hecke
operator sends a modular form with integral Fourier coefficients to a modular form with
integral Fourier coefficients) and so the eigenvalues are algebraic. Given an eigenforms
f = ∑∞ n
n=1 an q ∈ S2k (SL2 (Z)) it follows that

Tn (f ) = λn ⋅ f
hence looking at the first term of both q-expansions we get
an = λ n ⋅ a1 .
Since f ≠ 0 it follows that a1 ≠ 0 and λn = an /a1 . Finally if f, g ∈ S2k (SL2 (Z)) are two
normalized eigenforms with Fourier coefficients an and bn respectively (i.e. a1 = b1 = 1), then
an (f, g) = (Tn (f ), g) = (f, Tn (g)) = bn (f, g).
Therefore (f, g) = 0 or f = g.

143
19 Dirichlet L-series
19.1 The order of Fourier coefficients of cusp forms
The third part of Ramanujan conjecture was solved by Deligne in 1974 as a consequence of
his proof of Weil conjectures. In this section we will establish a much weaker bound due to
Hecke, that will enable us to associate an L-series to every cusp form of the full modular
group.

Theorem 19.1 (Hecke). Let f ∈ S2k (SL2 (Z)) be a cusp form with Fourier expansion
∞
f (τ ) = ∑ an q n , q = e2πiτ .
n=1

Then an = O(nk ), i.e. there exists some C > 0 such that

∣an ∣ ≤ C ⋅ nk .

Remark 19.1. In particular we get that τ (n) = O(n6 ), which is much weaker than Ramanjan
conjecture
11
∣τ (n)∣ ≤ 2n 2 .

Proof Since f (τ ) = O(∣q∣) = O(e−2π Im τ ) as Im τ → +∞, there exists some c > 0 such that

(Im τ )k ∣f (τ )∣ ≤ c ∀τ ∈ H.

Let τ = x + iy, fix y > 0 and take 0 ≤ x ≤ 1. Then

1 1
−n−1 −n
an = f (τ )q dq = ∫0 f (x + iy)q dx.
2πi ∫∂B(0,e−2πy )
Hence
1 e2πny
∣an ∣ ≤ ∫ ∣f (x + iy)∣e2πny dx ≤ c ⋅ ∀y > 0.
0 yk
1
In particular for y = n we get
∣an ∣ ≤ c ⋅ e2π ⋅ nk .

Corollary 19.2. Let f (τ ) = ∑∞ n

n=1 an q ∈ S2k (SL2 (Z)), then the series

∞
an
Lf (s) ∶= ∑ s
n=1 n

is holomorphic and converges absolutely for Re(s) > k + 1.

144
Proof ∞ ∞
an
∑∣ ∣ ≤ C ∑ nk−Re(s) < +∞.
n=1 ns n=1

Theorem 19.3. Let f ∈ S2k (SL2 (Z)). The function Lf (s) extends analytically to an entire
function and satisfies the functional equation

(2π)−s Γ(s)Lf (s) = (−1)k (2π)−(2k−s) Γ(2k − s)Lf (2k − s),

where Euler’s Γ function is defined as

∞
Γ(s) ∶= ∫ e−t ts−1 dt.
0

Remark 19.2. Recall that Γ(s) is holomorphic and absolutely convergent for Re(s) > 0.
It extends to a meromorphic function on C with simple poles at Z≤0 . And satisfies the
functional equation
Γ(z + 1) = zΓ(z).
In particular, since Γ(1) = 1 it follows that Γ(n + 1) = n! for all n ∈ N.
Proof Since f (τ ) = O(∣q∣) = O(e−2πy ) as y → +∞, it follows that for y0 > 0 fixed
∞
∫y f (iy)y s−1 dy
0

is absolutely convergent ∀s ∈ C. And the same will hold for the derivatives of f . Thus the
above integral defines an entire function on s. Moreover, by modularity
i
f (iy) = i−2k y −2k f ( ) = O(y −2k e y ) as y → 0.
−2π

y
Hence the limit as y0 → 0 of the integral exists and defines an entire function
∞ ∞ ∞ ∞ ∞ ts−1 dt
∫0 f (iy)y s−1 dy = ∑ an ∫ e−2πny y s−1 dy = ∑ an ∫ e−t s−1
= (2π)−s Γ(s)Lf (s).
n=1 0 n=1 0 (2πn) 2πn
We can split the integral on the left as
1 ∞
s−1
∫0 f (iy)y dy + ∫1 f (iy)y s−1 dy.

1
We change variables y ↦ y in the first integral and use the modularity to get
1 ∞ i ∞ −1 ∞
∫0 f (iy)y s−1 dy = ∫ f ( ) y −s−1 dy = ∫ f ( ) y −s−1 dy = ∫ (iy)2k f (iy)y −s−1 dy
1 y 1 iy 1

∞
= (−1)k ∫ f (iy)y 2k−s−1 dy.
1

145
Thus
∞ ∞ dy
(2π)−s Γ(s)Lf (s) = ∫ f (iy)y s−1 dy = ∫ f (iy)(y s + (−1)k y 2k−s )
1 1 y
and so
∞ dy
(−1)k (2π)−(2k−s) Γ(2k−s)Lf (2k−s) = (−1)k ∫ f (iy)(y 2k−s +(−1)k y s ) = (2π)−s Γ(s)Lf (s).
1 y

Definition 19.1. A Dirichlet series is a series of the form

∞
an
∑ s
, an , s ∈ C.
n=1 n

Example 19.1. For an = 1, the associated Dirichlet series is the Riemann zeta function
∞
1
ζ(s) = ∑ s
.
n=1 n

Since ∞ ∞
1 1
∑∣ s
∣ = ∑ Re(s)
< +∞
n=1 n n=1 n

for Re(s) > 1, it follows that for Re(s) > 1

1 1
ζ(s) = ∏ (1 + s
+ 2s + ⋯) = ∏ (1 − p−s )−1 .
p prime p p p prime

Example 19.2. For an = τ (n), the Dirichlet series

∞
τ (n)
L∆ (s) = ∑ s
n=1 n

converges absolutely on Re(s) > 7. Thus there (since τ is multiplicative)

τ (p) τ (p2 )
L∆ (s) = ∏ (1 + + 2s + ⋯) .
p prime ps p

Using the identity τ (pk ) = τ (pk−1 )τ (p) − p11 τ (pk−2 ) for k ≥ 2, we see that
τ (p) p11 ∞ τ (pk ) ∞ τ (pk ) ∞ τ (pk )τ (p) ∞ τ (pk )p11
(1 − + 2s ) ∑ ks = ∑ ks − ∑ + ∑ (k+2)s
ps p k=0 p k=0 p k=0 p(k+1)s k=0 p

τ (p) ∞ τ (pk ) ∞ τ (pk−1 )τ (p) τ (p) ∞ τ (pk−2 )p11

=1+ + ∑ ks − ∑ − s +∑ =1
ps k=2 p k=2 pks p k=2 pks
hence
−1
τ (p)
L∆ (s) = ∏ (1 − + p11−2s ) .
p prime ps

146
Remark 19.3. In general for f (n) multiplicative, its L-series has an Euler product formula
in its absolute convergence domain
∞
f (n) f (p) f (p2 )
L(s) = ∑ s
= ∏ (1 + + 2s + ⋯) .
n=1 n p prime ps p

When f (n) is completely multiplicative (i.e. f (pα1 1 ⋯pαk k ) = f (p1 )α1 ⋯f (pk )αk ) we have further
that
−1
f (p)
L(s) = ∏ (1 − s ) .
p prime p
The analytic continuation and functional equation of Dirichlet series attached to cusp forms
were established using the modularity condition. In order to do this for more general Dirichlet
series we will generalize the modularity condition as follows: Let {an }∞ ∞
n=0 , {bn }n=0 be two
sequence of complex numbers, and suppose they have polynomial growth
an = O(nc ) , bn = O(nc ) for some c > 0.
To each of these sequences we associate for t ≥ 0
∞ ∞
f (t) = ∑ an e−πnt , g(t) = ∑ bn e−πnt .
n=0 n=0

We also associate the following Dirichlet series

∞ ∞
an bn
Lf (s) = ∑ s
, Lg (s) = ∑ s
.
n=1 n n=1 n

Theorem 19.4. With f and g as above, assume that for t > 0

1
f ( ) = wtk g(t)
t
for some w ∈ C and k ∈ R. Then Lf (s) and Lg (s) extend analytically to the entire complex
plane apart from possible simple poles at s = 0 and k and satisfy the functional equation
π −s Γ(s)Lf (s) = wπ −(k−s) Γ(k − s)Lg (k − s).
If a0 = b0 = 0, there are no poles.
Proof By the polynomial growth assumption we see that
∞ ∞ ∞
s dt −πnt dt
∫0 (f (t) − a0 )t t = ∫0 ( ∑ an e ) ts
n=1 t
is an absolutely convergent integral. Moreover, Lf (s) converges absolutely for Re(s) suffi-
ciently large and so in that region the integral can be rewritten as
∞ ∞ ∞ ∞
−πnt s−1 us−1 du
∑ an ∫ e t dt = ∑ an ∫ e−u s
= π −s Γ(s)Lf (s).
n=1 0 n=1 0 (πn)

147
Splitting
∞ 1 ∞
s−1 s−1 s−1
∫0 (f (t) − a0 )t dt = ∫0 (f (t) − a0 )t dt + ∫1 (f (t) − a0 )t dt,
we can compute
1 a0 1 a0 ∞ 1 −s−1
s−1 s−1
∫0 (f (t) − a0 )t dt = − s + ∫0 f (t)t dt = − s + ∫1 f ( t ) t dt

a0 ∞ dt a0 wb0 ∞ dt
=− + w ∫ g(t)tk−s = − − + w ∫ (g(t) − b0 )tk−s
s 1 t s k−s 1 t
and obtain
a0 wb0 ∞ dt ∞ dt
π −s Γ(s)Lf (s) = − − + w ∫ (g(t) − b0 )tk−s + ∫ (f (t) − a0 )ts .
s k−s 1 t 1 t
Similarly, using that g ( 1t ) = w−1 tk f (t) we have
b0 w−1 a0 ∞ dt ∞ dt
π −s Γ(s)Lg (s) = − − + w−1 ∫ (f (t) − a0 )tk−s + ∫ (g(t) − b0 )ts .
s k−s 1 t 1 t
Therefore
b0 w−1 a0 ∞ dt ∞ dt
π −(k−s) Γ(k − s)Lg (k − s) = − − + w−1 ∫ (f (t) − a0 )ts + ∫ (g(t) − b0 )tk−s
k−s s 1 t 1 t
= w−1 π −s Γ(s)Lf (s).

Corollary 19.5. The Riemann zeta function extends analytically to the entire complex
plane except for a simple pole at 1 and satisfies the functional equation
1
π −s Γ(s)ζ(2s) = π −( 2 −s) Γ ( − s) ζ(1 − 2s).
1

2
Proof Since the θ function satisfies for t > 0 that (see Problem 17 and Proposition 15.1)
√5
−1 η 5 ( −1
it
) t η 5 (it) √ it
θ( ) = −1 −2
= √ 2 √ 2 = tθ ( ) ,
2it η 2 ( 2it ) η 2 ( it ) 2t η 2 (2it) 2t η 2 ( it2 ) 2

it follows that for f (t) ∶= θ ( it2 ) we have

1 1
f ( ) = t 2 f (t).
t
Since f (t) = 1 + 2 ∑∞ −πn t , then L (s) = 2ζ(2s) has a possible simple pole at s = 0 and
2
n=1 e f
1/2, and satisfies the functional equation
1 1
π −s Γ(s)Lf (s) = π −( 2 −s) Γ ( − s) Lf ( − s) .
1

2 2

148
19.2 L-series attached to elliptic curves
Now we will explain how to associate to every elliptic curve E/Q an L-series. In order to do
this we have to introduce some notions we have already seen for elliptic curves defined over
local fields.

Definition 19.2. Let E/Q be an elliptic curve. We say that a Weierstrass equation is a
minimal model for E/Q if ∆ ∈ Z and ∣∆∣ is minimal among all Weierstrass equations
defined over Q isomorphic to E/Q.

Definition 19.3. Let E/Q be an elliptic curve given by a minimal model, le p ≥ 2 be a prime
number, and let Ẽ be the reduction of E modulo p. We say E/Q has good reduction modulo
p if Ẽ is smooth. If Ẽ is singular at a point P ∈ E(Fp ) we say E/Q has bad reduction
modulo p and we distinguish two cases:

(1) If Ẽ has a cusp at P , then we say that E has an additive reduction modulo p.

(2) If Ẽ has a node at P , we say that E has a multiplicative reduction modulo p. If the
slopes of the tangent lines are in Fp , the reduction is said to be split multiplicative,
and non-split multiplicative otherwise.

Definition 19.4. Let E/Q be an elliptic curve. We define for every p ≥ 2 prime the local
part at p of the L-series
⎧
⎪
⎪ 1 − ap T + pT 2 if E has good reduction mod p,
⎪
⎪
⎪ 1−T if E has split multiplicative reduction mod p,
Lp (T ) ∶= ⎨
⎪
⎪ 1 + T if E has non-split multiplicative reduction mod p,
⎪
⎪
⎪
⎩ 1 if E has additive reduction mod p,

where ap ∶= p + 1 − #E(Fp ). The Hasse-Weil L-series of E/Q is defined as

L(E, s) ∶= ∏ Lp (p−s )−1 .

p prime

Proposition 19.6. The Hasse-Weil L-series of E/Q is absolutely convergent and anlytic on
Re(s) > 23 .

Proof Since E/Q has only finitely many primes of bad reduction, it is enough to show that
′
ap p −1
L(s) ∶= ∏ (1 − + )
p ps p2s

converges absolutely on Re(s) > 32 , where the product ∏′ means it is taken over the primes
of good reduction. For each prime of good reduction p ≥ 2, let α, β ∈ C be such that

α + β = ap , αβ = p.

149
By the Weil conjectures for elliptic curves (more precisely the Riemann hypothesis) we know
1
∣α∣ = ∣β∣ = p 2 . Then

ap p −1 α −1 β −1 α α2 β β2
(1 − s + 2s ) = (1 − s ) (1 − s ) = (1 + s + 2s + ⋯) (1 + s + 2s + ⋯)
p p p p p p p p
α + β α2 + αβ + β 2 α3 + α2 β + αβ 2 + β 3
=1+ + + +⋯
ps p2s p3s
and so
′ ∞ ′ ∞
αk + αk−1 β + ⋯ + αβ k−1 + β k
∑ (k + 1)(p 2 −Re(s) )k
1
∏∑∣ ks
∣ ≤ ∏
p k=0 p p k=0

′ ∞ 2
1 1
−Re(s)
=∏ ≤ (∑ n 2 ) < +∞
(1 − p 2 −Re(s) )2
1
p n=1
1
for 2 − Re(s) < −1, i.e. Re(s) > 23 .
Remark 19.4. In general, for any prime p
∞
f (pk )
Lp (p−s )−1 = ∑ ks
k=0 p

for some function f . If extend f multiplicatively to N we see that

∞
f (n)
L(E, s) = ∑ s
.
n=1 n

In particular, for a prime of good reduction we have

f (pk+1 ) = f (pk )f (p) − pf (pk−1 ).

This last relation suggests a relation between L(E, s) and an L-series of some Hecke eigen-
form. This is essentially the modularity theorem, which as a consequence implies the analytic
continuation and functional equation of L(E, s).
Definition 19.5. Let E/Q be an elliptic curve. For each prime p we define
⎧
⎪
⎪ 0 if E has good reduction mod p,
⎪
⎪
⎪ 1 if E has multiplicative reduction mod p,
fp ∶= ⎨
⎪
⎪ 2 if E has additive reduction mod p and p ≠ 2, 3,
⎪
⎪
⎩ 2 + δp
⎪ if E has additive reduction mod p = 2 or 3,

where δp is an invariant that describes whether there is wild ramification in the action of the
inertia group at p of Gal(Q/Q) on the Tate module Tp (E). The conductor NE/Q of E/Q
is defined as
NE/Q ∶= ∏ pfp .
p prime

150
Theorem 19.7. The L-series L(E, s) has an analytic continuation to the entire complex
plane, and it satisfies the following functional equation. Define

Λ(E, s) ∶= (NE/Q ) 2 (2π)−s Γ(s)L(E, s),

s

then
Λ(E, s) = wΛ(E, 2 − s)
with w = ±1.

Definition 19.6. The number w = w(E/Q) in the functional equation is called the root
number of E.

Remark 19.5. Theorem 19.7 is a corollary of the modularity theorem as we will see in the
next lecture. Nevertheless it was conjectured before, and it played a fundamental role in
other famous conjecture, the so called Birch-Swinnerton-Dyer conjecture.

Conjecture 19.8 (Birch and Swinnerton-Dyer). Let E/Q be an elliptic curve. Then

(1) L(E, s) has a zero of order RE = rank E(Q) at s = 1. In other words, the Taylor
expansion of L(E, s) at s = 1 is of the form

L(E, s) = c0 (s − 1)RE + c1 (s − 1)RE +1 + c2 (s − 1)RE +2 + ⋯

where c0 ≠ 0.

(2) The coefficient c0 has a concrete expression in terms of the invariants of E/Q. More
precisely
L(E, s) #X ⋅ ΩE ⋅ 2RE ⋅ Reg(E/Q) ⋅ ∏p cp
c0 = lim = ,
s→1 (s − 1)RE (#E(Q)tors )2
where ΩE = ∫E(R) ∣ dx
y ∣ is either the real period or twice the real period of a minimal
model of E, depending on whether E(R) is connected or not. The group X is the
Shafarevich-Tate group of E/Q (its 2-torsion part X2 (E/Q) measures how many
homogeneous spaces of E/Q do not satisfy the Local-to-Global principle). The ellip-
tic regulator is denote Reg(E/Q) and corresponds to the determinant of the height
matrix of E(Q). And cp = #(E(Qp )/E0 (Qp )) where E0 (Qp ) is the set of points in
E(Qp ) whose reduction modulo p is non-singular in E(Fp ), this number is called the
Tamagawa number of E at p.

Conjecture 19.9 (Parity conjecture). The root number of E/Q, denoted w = w(E/Q),
indicates the parity of the rank of E(Q), i.e. w = 1 if and only if RE is even. Equivalently

w = (−1)RE .

151
Remark 19.6. It follows from the functional equation that w determines the parity of the
function Λ(E, 1 − s), and consequently
w = (−1)ords=1 Λ(E,s) = (−1)ords=1 L(E,s) .
Therefore, the parity conjecture is a corollary of the BSD conjecture.
Definition 19.7. We say that ords=1 L(E, s) is the analytic rank of E/Q.
Example 19.3. Let E/Q be given by
y 2 = x3 − 1572 x.
Recall that finding a non-trivial rational point on E/Q is equivalent to show that 157 is a
congruent number. Comparing the values of Λ(s) and Λ(2 − s) we can calculate the root
number w(E/Q) = 1. Thus the parity conjecture suggests that RE ≥ 1 and so 157 should be
a congruent number. Approximating L(E, s) at s = 1 we get that BSD conjecture implies
RE = 1 and
#X ⋅ ΩE ⋅ 2RE ⋅ Reg(E/Q) ⋅ ∏p cp
= 11.4259445007 . . .
(#E(Q)tors )2
If we believe RE = 1 and write P for a generator of the free part, one can show that X is
trivial. We can also estimate ΩE = 0.4185259488 . . ., ∏p cp = c2 ⋅ c157 = 2 ⋅ 4, #E(Q)tors = 4.
However Reg(E/Q) = ⟨P, P ⟩ = ĥ(P ) cannot be calculated since we do not know P . Using
BSD conjecture we get
1
ĥ(P ) = 27.3004446469 . . . ≈ log max{num(x(P )), den(x(P ))}
2
and so
max{num(x(P )), den(x(P ))} ≈ e54.6 ≈ 5.157 ⋅ 1023 .
With the help of homogeneous spaces, and looking for points in the correct height range we
can succeed at finding P , whose coordinates are
166136231668185267540804
x(P ) = − ,
2825630694251145858025
167661624456834335404812111469782006
y(P ) = ,
150201095200135518108761470235125
and the canonical height is precisely 27.3004446469 . . . as predicted by the BSD conjecture.
The following is the strongest evidence in favor of BSD conjecture.
Theorem 19.10 (Gross-Zagier, Kolyvagin). Let E/Q be an elliptic curve of rank RE .
Suppose that the analytic rank is ≤ 1, i.e. ords=1 L(E, S) ≤ 1. Then
(1) The first part of BSD conjecture holds for E/Q, i.e.
ords=1 L(E, s) = RE .

(2) The Shafarevich-Tate group X associated to E/Q is finite.

152
20 Hecke Operators of Higher Level
In the next lecture we will explain the modularity theorem in terms of L-series. Using it we
will prove the analytic continuation of the L-series attached to an elliptic curve. For this
we will reduce ourselves to find a functional equation and analytic continuation of L-series
attached to cusp forms of higher level. This is why we need to generalize our theory of Hecke
operators to higher level.

Let Γ1 and Γ2 be subgroups of GL+2 (Q). Let α ∈ GL+2 (Q). The double coset is

Γ1 αΓ2 = {γ1 αγ2 ∶ γ1 ∈ Γ1 , γ2 ∈ Γ2 }.

Both groups act on it, Γ1 by the left and Γ2 by the right. Thus we can decompose

Γ1 αΓ2 = ⊔ Γ1 αγj .
j

Example 20.1. Let p be a prime number, Γ1 = Γ2 = Γ = SL2 (Z). Let Xp be the set of
matrices with integral entries and determinant p. Then

p 0
Xp = Γ ( ) Γ.
0 1

In fact, if σ ∈ Xp we know
a b
σ = γ1 ( )
0 d
for some γ1 ∈ Γ, ad = p and 0 ≤ b ≤ d − 1. If a = p, d = 1 and b = 0 we are done. Otherwise
a = 1, d = p,
1 b 1 0 1 b
( )=( )( )
0 p 0 p 0 1
and
1 0 0 −1 p 0 0 1 p 0
( )=( )( )( ) ∈ Γ( ) Γ.
0 p 1 0 0 1 −1 0 0 1

Definition 20.1. Fix N ∈ N. Let S + be an additive subgroup of Z (hence S + = M Z for some

M ≥ 0). Let S × be a multiplicative subgroup of (Z/N Z)× . Define

Xn (N, S × , S + ) ∶= {( ac db ) ∈ Xn ∶ N ∣c, a ∈ S × , b ∈ S + }.

Example 20.2. 1. For N = 1, S + = Z, S × = {1}

Xn (1, S × , S + ) = Xn .

2. For n = 1
X1 (N, 1, Z) = Γ1 (N ) = {( ac db ) ∶ N ∣c , a ≡ d ≡ 1 (mod N )}.

153
 3. Γ(N ) = X1 (N, 1, N Z).

4. Γ0 (N ) = X1 (N, (Z/N Z)× , Z).

Remark 20.1. For fixed N , S + , S ×

Xn (N, S × , S + )Xm (N, S × , S + ) ⊆ Xmn (N, S × , S + ).

In general X1 (N, S × , S + ) is a congruence subgroup of SL2 (Z). Observe that

X1 (N, S × , S + )Xn (N, S × , S + ) ⊆ Xn (N, S × , S + ) and Xn (N, S × , S + )X1 (N, S × , S + ) ⊆ Xn (N, S × , S + )

so we have both a left and a right action of X1 (N, S × , S + ) on Xn (N, S × , S + ). Furthermore

the number of orbits of X1 (N, S × , S + ) is finite and so we can write
r
Xn (N, S × , S + ) = ⊔ X1 (N, S × , S + )αi .
i=1

Moreover, multiplying the orbits on the right by any fixed γ ∈ X1 (N, S × , S + ) simply permutes
the orbits r r
⊔ X1 (N, S × , S + )αi γ = ⊔ X1 (N, S × , S + )αi .
i=1 i=1

Definition 20.2. Let N , S + and S × be fixed. For every f ∈ Mk (X1 (N, S × , S + )) we define
the Hecke operator
r
Tn (f ) ∶= n 2 −1 ∑ f [αi ]k ,
k

i=1

where α1 , . . . , αr ∈ Xn (N, S × , S + ) are a set of representatives of the action of X1 (N, S × , S + )

by the left.

Theorem 20.1. Let Γ = X1 (N, S × , S + ) as above. For any k, n ≥ 1, the Hecke operator is a
well-defined linear map from Mk (Γ) to itself and from Sk (Γ) to itself.

Proof Suppose f ∈ Mk (Γ) and γ ∈ Γ. Then for any αi , since f [γαi ]k = f [αi ]k we see that
the map does not depend on the choice of the representatives. Moreover,
r
Tn (f )[γ]k = n 2 −1 ∑ f [αi γ]k = Tn (f )
k

i=1

since γ only permutes the orbits. By Remark 18.3 each f [αi ]k is holomorphic at each cusp
of Q ∪ {i∞} and hence Tn (f ) also is. Thus it takes Mk (Γ) to itself, and by the same remark
it takes cusp forms to cusp forms.

In order to compute explicitly the Hecke operators we have to fix some X1 . We will focus
on the case X1 (N, 1, Z) = Γ1 (N ).

154
Theorem 20.2. For each a ∈ (Z/N Z)× fix σa ∈ SL2 (Z) such that

a−1 0
σa ≡ ( ) (mod N ).
0 a

Let n ∈ N and e = gcd(n, N ), then

de−1 n
b
Xn (N, 1, Z) = ⊔ ⊔ Γ1 (N )σ de
n ( de
0 de
).
d∣ n
e
b=0

Proof Clearly the right hand side is contained in Xn (N, 1, Z). We have to show the union
is disjoint. Suppose
n n
b1 b2
γ1 σ dne ( d1 e ) = γ2 σ dne ( d2 e )
1 0 d1 e 2 0 d2 e
for some γ1 , γ2 ∈ Γ1 (N ). Then
n n −1 n d2
b1 b2 b1 d e −b2 1 ∗
( d1 e ) ( d2 e ) = ( d1 e )( 2 ) = ( d1
d1 ) ∈ SL2 (Z)
0 d1 e 0 d2 e 0 d1 e 0 dn2 e n 0 d2

and so d1 = d2 . Therefore
n
b1 1 j dn1 e b2
( d1 e )=( )( )
0 d1 e 0 1 0 d1 e
and 0 ≤ b1 , b2 ≤ d1 e − 1, hence j = 0 and b1 = b2 . We need to show still that any element
( αγ βδ ) ∈ Xn (N, 1, Z) lies in one of our orbits. Choose coprime integers g, h ∈ Z such that
gα + hγ = 0. Note that

αδ − βγ = n , α ≡ 1 (mod N ) , N ∣γ.

We can find integers x, y ∈ Z such that xh − gy = 1. Set

x y
A=( ) ∈ Γ0 (N ),
g h

then
α β ∗ ∗
A( )=( )
γ δ 0 ∗
of determinant n. We need to show that this has the desired form. Replacing A by ± ( 10 1j ) A
if necessary we may suppose

α β a b
A( )=( ) , a > 0 , 0 ≤ b ≤ d − 1 , ad = n , gcd(a, N ) = 1.
γ δ 0 d

Since S × = 1 we have
a−1 ∗
A−1 ≡ ( ) (mod N )
0 a

155
and so we can find some B ∈ Γ1 (N ) such that

A−1 = Bσa .

Definition 20.3. A Dirichlet character is a group homomorphism

χ ∶ (Z/N Z)× → C×

Remark 20.2. Given any Dirichlet character χ ∶ (Z/N Z)× → C× , let

ψχ ∶ Γ0 (N ) → C×

( ac db ) ↦ χ(d).
It is not hard to see that ψ is a group homomorphism. It is not hard to see also that the
set of Dirichlet characters is a group. For a fixed N = pα1 1 ⋯pαk k > 0 each Dirichlet character
is a product of Dirichlet characters of the form χp ∶ (Z/pα Z)× → C× . Since for p > 2 and for
pα = 2, 4 these groups are cyclic, and since for α ≥ 3, (Z/2α Z)× ≃ Z/2Z ⊕ Z/2α−2 Z, we see that
χ is determined by choosing the values of the generators of χp for each prime p ≥ 2, which
in any case are exactly ϕ(pα ) values. Thus there are exactly ϕ(N ) Dirichlet characters.
Furthermore, these characters satisfy the following orthogonality relations. Let N > 0 fixed
and d ∈ (Z/N Z)× , then for every character ψ

∑ χ(d) = ∑ χψ(d) = ψ(d) ∑ χ(d).

χ χ χ

Hence
ϕ(N ) if d ≡ 1 (mod N ),
∑ χ(d) = {
χ 0 otherwise.
Also for every b ∈ (Z/N Z)×

∑ χ(a) = ∑ χ(ab) = χ(b) ∑ χ(a).

a∈(Z/N Z)× a∈(Z/N Z)× a∈(Z/N Z)×

Thus
ϕ(N ) if χ ≡ 1,
∑ χ(a) = {
a∈(Z/N Z)×
0 otherwise.

Definition 20.4. Let χ ∶ (Z/N Z)× → C× be a Dirichlet character. Define

Mk (Γ0 (N ), χ) ∶= {f ∈ Mk (Γ1 (N )) ∶ f [γ]k = ψχ (γ) ⋅ f ∀γ ∈ Γ0 (N )}.

For an element of this space we call χ the nebentypus. In a similar way we define
Sk (Γ0 (N ), χ).

156
Theorem 20.3. We have the following decompositions

Mk (Γ1 (N )) = ⊕ Mk (Γ0 (N ), χ),

χ

Sk (Γ1 (N )) = ⊕ Sk (Γ0 (N ), χ).

χ

Proof Let us show first that any f ∈ Mk (Γ1 (N )) can be written as a sum of fχ ∈
Mk (Γ0 (N ), χ). Let
1
fχ ∶= ∑ χ(d)f [γd ]k
ϕ(N ) d∈(Z/N Z)×
where γd ∈ Γ0 (N ) is such that the lower right entry of γd is congruent to d modulo N .
This is well-defined since f ∈ Mk (Γ1 (N )). To see that fχ has nebentypus χ pick some
γ = ( pr qs ) ∈ Γ0 (N ), then
1 1
fχ [γ]k = ∑ χ(d)f [γd γ]k = ∑ χ(d)f [γds ]k
ϕ(N ) d∈(Z/N Z)× ϕ(N ) d∈(Z/N Z)×

1
= ∑ χ(ds)χ(s)f [γds ]k = ψχ (γ)fχ .
ϕ(N ) d∈(Z/N Z)×
On the other hand
1
∑ fχ = ∑ ∑ χ(d)f [γd ]k = f [γ1 ]k = f.
χ ϕ(N ) d∈(Z/N Z)× χ

To show that the sum is direct, consider g ∈ Mk (Γ0 (N ), µ), then

⎛ 1 ⎞ g if χ = µ,
gχ = ∑ µχ(d) g = {
⎝ ϕ(N ) d∈(Z/N Z)× ⎠ 0 otherwise.

Remark 20.3. For any f ∈ Mk (Γ0 (N ), χ) we can compute the Hecke operator as

n d−1 n
Tn (f ) = n 2 −1 ∑ χ (
k b
) ∑ f [( de0 )]
de k
d∣ n de b=0
e

where e = gcd(n, N ). Since for gcd(m, N ) ≠ 1 we define χ(m) = 0 we can write this sum as

n d−1 n
Tn (f ) = n 2 −1 ∑ χ ( ) ∑ f [( 0d db )]k .
k

d∣n d b=0

In consequence we see that Tn maps Mk (Γ0 (N ), χ) to itself, since for γ ∈ Γ0 (N )

n d−1 n n d−1 n
Tn (f )[γ]k = n 2 −1 ∑ χ ( ) ∑ f [( 0d db ) γ]k = n 2 −1 ∑ χ ( ) ψχ (γ) ∑ f [γ −1 ( 0d db ) γ]k = ψχ (γ)Tn (f ).
k k

d∣n d b=0 d∣n d b=0

157
We can also compute the Fourier expansion of Tn (f ) in terms of the Fourier expansion of
f = ∑∞
m=0 λ(m)q
m in similar way as we did for the Hecke operators for SL (Z) and get
2

∞ ⎛ mn ⎞
Tn (f ) = ∑ ∑ χ(a)ak−1 λ ( 2 ) q m .
m=0 ⎝a∣ gcd(m,n) a ⎠

In particular if f ∈ Sk (Γ0 (N ), χ) is a Hecke eigenform, i.e. if

Tn (f ) = cn ⋅ f

for all n, then

λ(n) = cn ⋅ λ(1)
and so λ(1) ≠ 0. Furthermore for gcd(m, n) = 1

λ(mn) = cn ⋅ λ(m)

and so if λ(1) = 1, then λ is multiplicative. Moreover

λ(pk+1 ) = λ(p)λ(pk ) − χ(p)pk−1 λ(pk−1 ).

Definition 20.5. For every d ∈ Z+ define the diamond operator

⟨d⟩ ∶ Mk (Γ1 (N )) → Mk (Γ1 (N ))

as follows
f [γd ]k if gcd(d, N ) = 1,
⟨d⟩f ∶= {
0 otherwise,
where γd ∈ Γ0 (N ) is any matrix with lower right entry congruent to d modulo N .

Remark 20.4. As for SL2 (Z), it is routine to check that

Tn Tm = Tnm

for gcd(n, m) = 1. And that

Tpk+1 = Tp Tpk − pk ⟨p⟩Tpk−1 .
Thus the Hecke operators commute. Furthermore, they clearly commute with the diamond
operators (which commute among themselves ⟨d⟩⟨e⟩ = ⟨de⟩). Therefore in order to conclude a
decomposition of Sk (Γ1 (N ), χ) as a direct sum of Hecke eigenforms we just need to compute
the adjoints of Tn with respect to Petersson inner product.

Proposition 20.4. Let gcd(n, N ) = 1, then for f ∈ Sk (Γ0 (N ), χ) and g ∈ Mk (Γ1 (N ))

(Tn (f ), g) = χ(n)(f, Tn (g)).

158
Proof Once we prove it for n = p prime, it follows by induction that it also holds for pk ,
and so using the identity Tmn = Tm Tn for gcd(m, n) = 1 it follows in general. Thus we are
reduced to the prime case. By a similar count as for Hecke operators of level one we see that
p−1
k
−1
(Tp (f ), g) = p 2 (χ(p)f [( p0 10 )]k + ∑ f [( 10 pc )]k , g)
c=0

p−1
= p 2 −1 χ(p)(f [( p0 p0 )]k , g[( 10 p0 )]k ) + p 2 −1 ∑ (f [( p0 p0 )]k , g[( p0 −c
k k
1 )]k )
c=0
p−1
= p 2 −1 χ(p)(f, g[( 10 p0 )]k ) + p 2 −1 (f, g[( p0 10 )]k ) + p 2 −1 ∑ (f, g[( p0 −c
k k k
1 )]k ).
c=1
Let r, s ∈ Z such that rp − sN = 1, then
−1
p −c 1 −c p 0 1 −c 1 s 1 0 p s
( )=( )( )=( )( ) ( )( ).
0 1 0 1 0 1 0 1 N rp 0 p N r
Thus we get
p−1
(Tp (f ), g) = p 2 −1 χ(p)(f, g[( 10 p0 )]k ) + p 2 −1 (f, g[( p0 10 )]k ) + p 2 −1 ∑ (f, g[( 10 p0 ) ( Np rs )]k )
k k k

c=1

p−1
−s
= p 2 −1 χ(p)(f, g[( 10 p0 )]k ) + p 2 −1 (f, g[( p0 10 )]k ) + p 2 −1 ∑ (f [( −N
k k k r 1 c 1 c
p ) ( 0 1 )]k , g[( 0 p )]k ).
c=1
r −s
Since ( −N p ) ∈ Γ0 (N ) with lower right entry congruent to p modulo N , we get
p−1
(Tp (f ), g) = p 2 −1 χ(p)(f, g[( 10 p0 )]k ) + p 2 −1 χ(p)(f, χ(p)g[( p0 01 )]k ) + p 2 −1 ∑ χ(p)(f, g[( 01 pc )]k )
k k k

c=1

= χ(p)(f, Tp (g)).

Corollary 20.5. There is a basis of Sk (Γ0 (N ), χ) consisting of eigenforms for all Tn with
gcd(n, N ) = 1.
−1
Proof Just note that χ(n) 2 Tn is self-adjoint with respect to the Peterssen product.
Remark 20.5. Our previous result only gives us eigenforms for all Tn with gcd(n, N ) = 1,
but we want Hecke eigenforms, i.e. eigenforms for all Tn . Since for primes p∣N each Tp
preserve all eigenspaces for Tn with gcd(n, N ) = 1, we would be done if these eigenspaces
were one dimensional. But unfortunately this is not the case. For instance if f ∈ Sk (Γ0 (N ), χ)
is an eigenform for all Tn with gcd(n, N ) = 1, then f [( p0 01 )]k also is an eigenform with the
same eigenvalues although they are linearly independent in general. In other words, f and
f [( p0 01 )]k have the same Fourier coefficients for gcd(n, N ) = 1, but different Fourier coefficient
for some p∣N . To get rid of this problem we restrict ourselves to a subspace of Sk (Γ1 (N ))
where this phenomenon does not arise.

159
Definition 20.6. We define the subspace of oldforms of level N of Sk (Γ1 (N )) as the
space spanned by the elements

f ∈ Sk (Γ1 (N /p)) ⊆ Sk (Γ1 (N ))

and also the elements of the form

f [( p0 01 )]k ∈ Sk (Γ1 (N ))

for each prime p∣N . We denote this subspace by Sk (Γ1 (N ))old . The subspace of newforms
of level N is its orthogonal complement with respect to Petersson inner product

Sk (Γ1 (N ))new = (Sk (Γ1 (N ))old )⊥ .

Remark 20.6. It can be shown that the decomposition into old and new forms is stable
under all Hecke operators and diamond operators (see [DS05, Proposition 5.6.2]). Thus
Sk (Γ1 (N ))old and Sk (Γ1 (N ))new admit a basis of eigenforms for all Tn with gcd(n, N ) = 1.

Theorem 20.6 (Atkin-Lehner). Let f ∈ Sk (Γ1 (N ))new be a non-zero eigenform for all
Hecke operators Tn and diamond operators ⟨n⟩ for gcd(n, N ) = 1. Then

(a) f is a Hecke eigenform, i.e. it is an eigenform for all Tn , n ∈ N.

(b) If f˜ satisfies the same conditions as f and has the same eigenvalues for all Tn with
gcd(n, N ) = 1, then it is a multiple of f .

Proof See [DS05, Theorem 5.8.2].

Definition 20.7. A form f ∈ Sk (Γ1 (N ))new ∩ Sk (Γ0 (N ), χ) is called a newform if it is a

normalized (i.e. its first Fourier coefficient is a1 = 1) Hecke eigenform, i.e. if it is a normalized
eigenvector for all Tn with gcd(n, N ) = 1.

Remark 20.7. It follows by Atkin-Lehner theorem that Sk (Γ1 (N ))new has as orthogonal
basis the set of newforms, and for each newform f , its Fourier coefficients an correspond to
its eigenvalues for each Hecke operator Tn , i.e.

Tn (f ) = an ⋅ f.

This is possible only after item (b) of Atkin-Lehner theorem, which is known as the multi-
plicity one property.

Proposition 20.7. Let f = ∑∞ n

n=1 an q ∈ Sk (Γ0 (N ), χ). Then f is normalized Hecke eigenform
if and only if its Fourier coefficients satisfy the conditions

(1) a1 = 1,

(2) apr = ap apr−1 − χ(p)pk−1 apr−2 for all p prime and r ≥ 2,

160
 (3) amn = am an for gcd(m, n) = 1.

Proof One implication was already proved. Assume the coefficients of f satisfy the con-
ditions. Then it is normalized by (1). In order to show that it is a Hecke eigenform it is
enough to show that
Tp (f ) = ap ⋅ f
for all primes p. This is equivalent to show that

amp = ap ⋅ am

for p ∤ m, and that

amp + χ(p)pk−1 a mp = ap ⋅ am
for p∣m. The first fact follows from (3), while for the second write m = pk m′ for p ∤ m′ , then

amp = am′ pr+1 = am′ apr+1 = am′ (ap apr − χ(p)pk−1 apr−1 ) = ap am − χ(p)pk−1 a mp .

21 Modularity Theorem and Beyond Modular Forms

Finally we will study the L-functions attached to cusp forms of higher level and state the
modularity theorem as conjectured by Taniyama and Shimura.

21.1 L-series attached to modular forms

As for the case of cusp forms for SL2 (Z), every modular form f ∈ Mk (Γ) has a q-expansion
at i∞ ∞
f = ∑ an q n .
n=0

Its associated L-series is ∞

an
Lf (s) ∶= ∑ s
.
n=1 n

In order to show that this series is holomorphic and absolutely convergent in some region of
the complex plane it is enough to show a polynomial growth of the Fourier coefficients.

Proposition 21.1. Let f = ∑∞ n

n=0 an q ∈ Mk (Γ), then an = O(n
k−1 ). If furthermore f is a
k
cusp form, then an = O(n 2 ).

Proof As before, we can show that (Im τ )k f (τ ) is bounded on H and so the same proof
k
as in the classical case of SL2 (Z) shows that an = O(n 2 ) when f ∈ Sk (Γ). Since

Mk (Γ) = C ⋅ Ek ⊕ Sk (Γ)

161
for some Eisenstein series Ek . Thus it is enough to show that an = Ck ⋅ σk−1 = O(nk−1 ). Since
σk−1 is a multiplicative function, it is enough to note that
p(k−1)(α+1) − 1
σk−1 (pα ) = ≤ 2pα(k−1) .
pk−1 − 1

Corollary 21.2. If f ∈ Mk (Γ) then Lf (s) is holomorphic and absolutely convergent on

Re(s) > k. If f ∈ Sk (Γ), the L-series converges absolutely on Re(s) > k2 + 1.
Theorem 21.3. Let f = ∑∞ n
n=1 an q ∈ Sk (Γ1 (N ), χ). Then f is a normalized Hecke eigenform
if and only if Lf (s) has an Euler product formula
−1
ap pk−1
Lf (s) = ∏ (1 − s + χ(p) 2s ) .
p prime p p

Proof One direction follows from the fact

−1 ∞
ap pk−1 apn
(1 − s + χ(p) 2s ) =1+ ∑ ns
p p n=1 p

which follows from relation apn = ap apn−1 − χ(p)pk−1 apn−2 for n ≥ 2. For the converse, define
recursively bp ∶= ap and bpn ∶= bp bpn−1 −χ(p)pk−1 bpn−2 for n ≥ 2. Extend it to N multiplicatively,
then
−1 ∞ ∞
ap pk−1 bpn bn
Lf (s) = ∏ (1 − s + χ(p) 2s ) = ∏ (1 + ∑ ns ) = ∑ s
p prime p p p prime n=1 p n=1 n

and so bn = an for all n ≥ 1, and the result follows from Propostion 20.7.
Definition 21.1. Let f ∈ Sk (Γ1 (N )), the Mellin transform of f is
∞ dt
g(s) ∶= ∫ f (it)ts .
0 t
Remark 21.1. The growth of the Fourier coefficients imply that the integral
∞ dt
∫y f (it)ts
0 t
converges absolutely and defines an holomorphic function for all s ∈ C, for any fixed y0 > 0.
In order to show that the Mellin transform also converges and is an entire function we have
to introduce another operator.
Definition 21.2. Define the operator

WN ∶ Sk (Γ1 (N )) → Sk (Γ1 (N ))
−1
f ↦ ik f [( N0 0 )]k .

162
Proposition 21.4. The operator WN is idempotent and self-adjoint. Hence we have a
decomposition
Sk (Γ1 (N )) = Sk (Γ1 (N ))+ ⊕ Sk (Γ1 (N ))−
where
Sk (Γ1 (N ))± = {f ∈ Sk (Γ1 (N )) ∶ WN f = ±f }.

Proof It is clear that

WN2 f = (−1)k f [( N0 −1 2
0 ) ]k = (−1)k f [( −N
0
0
−N )]k = (−1)k (−N )−k N k f = f.

Also it is easy to see that

−1 −1 2 0 −1 −1
(WN f, g) = ik (f [( N0 0 )]k , g) = ik (f [( N0 0 ) ]k , g[( N 0 )]k ) = ik ((−1)k f, g[( N0 0 )]k ) = (f, WN g).

Proposition 21.5. The Mellin transform of f ∈ Sk (Γ1 (N )) is entire and satisfies

g(s) = (2π)−s Γ(s)Lf (s)

k
for Re(s) > 2 + 1.

Proof We can reduce ourselves to the case f ∈ Sk (Γ1 (N ))± . In which case
−1
t−k f (
−k
N 2 ) = ±f (it)
N it
and so it also decays exponentially to zero as t → 0 and so the Mellin transform is a well
defined entire function. Since Lf (s) is absolutely convergent for Re(s) > k2 + 1 it follows that
∞ dt ∞ ∞ ∞ ∞ ts−1
∫0 f (it)ts = ∑ an ∫ e−2πnt ts−1 dt = ∑ an ∫ e−t s
dt = (2π)−s Γ(s)Lf (s).
t n=1 0 n=1 0 (2πn)

Theorem 21.6. Let f ∈ Sk (Γ1 (N ))± , then Lf (s) extends analytically to the entire complex
plane and satisfies the following functional equation

N 2 (2π)−s Γ(s)Lf (s) = ±N (2π)−(k−s) Γ(k − s)Lf (k − s).

s k−s
2

Proof We have already seen the analytic continuation so we just have to prove the functional
equation. This follows since

s
∞ it dt
N 2 g(s) = ∫ f ( √ ) ts .
0 N t

163
Splitting the integral we have
1 it dt ∞ i ∞ it dt
∫0 f ( √ ) ts = ∫ f ( √ ) t−s−1 dt = ∫ ±f ( √ ) tk−s
N t 1 t N 1 N t

hence
s
∞ it it dt k−s
N 2 g(s) = ∫ (f ( √ ) ts ± f ( √ ) tk−s ) = ±N 2 g(k − s).
1 N N t

Proposition 21.7. The operator WN commutes with all Hecke operators Tn .

Proof In fact let α1 , . . . , αr ∈ Xn (N, 1, Z) be representatives of the action of Γ1 (N ) on the

left. Then
r r
WN Tn (f ) = ik n 2 −1 ∑ f [αi ( N0 = ik n 2 −1 ∑ f [( N0
k k
−1 −1 0 1 0 −1
0 )]k 0 ) ( −N 0 ) αi ( N 0 )]k
i=1 i=1

r
= n 2 −1 ∑ WN f [( −N
k
0 1 0 −1
0 ) αi ( N 0 )]k .
i=1

Since for every γ = ( ac db ) ∈ Xn (N, 1, Z),

0 1 0 −1 c d 0 −1 dN −c
( )γ ( )=( )( )=( ) = N γ′
−N 0 N 0 −aN −bN N 0 −bN 2 aN

−1
with γ ′ ∈ Xn (N, 1, Z). We just need to show that these new ( −1
0 1/N
0
) αi ( N0 0 ) ∈ Xn (N, 1, Z)
represent different orbits. In fact if

0 1/N 0 −1 0 1/N 0 −1
( ) αi ( )=γ( ) αj ( )
−1 0 N 0 −1 0 N 0

for some γ ∈ Γ1 (N ), then

0 −1 −1 0 1/N
( )γ ( ) = αj αi−1 .
N 0 −1 0

Writing
x y
γ −1 = ( )
z w
we see that
−z
0 −1 −1 0 1/N w N)
( )γ ( )=( ∈ Γ1 (N ).
N 0 −1 0 −yN x

164
Corollary 21.8. Every newform f ∈ Sk (Γ0 (N ), χ) is an eigenvector for WN , i.e. lies in
Sk (Γ1 (N ))± .
Proof Since WN commutes with all Tn , it follows that Wn f is also a Hecke eigenform with
the same eigenvalues. Hence WN f is an scalar multiple of f , i.e. f is an eigenvector for WN .

Theorem 21.9 (Modularity theorem). An L-series L(s) = ∑∞ an

n=1 ns with an ∈ Z is the
Hasse-Weil L-series of an elliptic curve E/Q of conductor N if and only if L(s) = Lf (s) for
some newform f ∈ S2 (Γ0 (N ), χ).
Corollary 21.10. The Hasse-Weil L-series L(E, s) of an elliptic curve E/Q has an analytic
continuation to C and satisfies the functional equation
Λ(E, s) = ±Λ(E, 2 − s)
s
for Λ(E, s) ∶= (NE/Q ) 2 (2π)−s Γ(s)L(E, s).
Proof Just note that by the modularity theorem L(E, s) = Lf (s) for some newform of
S2 (Γ0 (N ), χ) and so f ∈ S2 (Γ1 (N ))± , thus the result follows by Theorem 21.6.

21.2 Quasi-modular forms and beyond

Finally we will close the course with some generalizations of modular forms involving differ-
ential equations. Our starting point is the observation that the derivative of a modular form
is not a modular form, but it is not too far from being it. In fact if
∞
f = ∑ an q n ∈ Mk (Γ)
n=0

∂ 1 ∂
then if we denote D = q ∂q = 2πi ∂τ
∞
′
f = ∑ nan q n
n=0

satisfies for ( ac db ) ∈Γ
aτ + b k
f′ ( ) = (cτ + d)k+2 f ′ (τ ) + c(cτ + d)k+1 f (τ ).
cτ + d 2πi
Definition 21.3. Let Γ ⊆ SL2 (Z) be a congruence subgroup. A quasi-modular form of
weight k and depth ≤ p for Γ is a holomorphic function f ∶ H → C such that
(i) f (x + iy) has subexponential growth as y → +∞ and y → 0,
(ii) For fixed τ ∈ H and variable γ = ( ac db ) ∈ Γ, the function
aτ + b
f [γ]k (τ ) = f ( ) (cτ + d)−k
cτ + d
c
is a polynomial of degree ≤ p in cτ +d .

165
 ̃k (Γ)≤p . The space of weight k quasi-modular
We denote the space of these forms by M
forms for Γ is
̃k (Γ) ∶= ⋃ M
M ̃k (Γ)≤p .
p≥0

The graded algebra of quasi-modular forms for Γ is

̃
M(Γ) ̃k (Γ).
∶= ⊕ M
k∈Z

Remark 21.2. It follows from the definition that every modular form is a quasi-modular
form.
Example 21.1. The Eisenstein series
∞
E2 (τ ) = 1 − 24 ∑ σ1 (n)q n
n=1

is a quasi-modular form of weight 2 and depth ≤ 1 for SL2 (Z), since for γ = ( ac db ) ∈ SL2 (Z)
aτ + b 6 c
E2 ( ) (cτ + d)−2 = E2 (τ ) + ( ).
cτ + d πi cτ + d
Proposition 21.11. The algebra of quasi-modular forms for Γ is closed under derivation.
And the derivation map increases the grading in 2 and the depth in 1
̃k (Γ)≤p → M
D∶M ̃k+2 (Γ)≤p+1 .

̃k (Γ)≤p , τ ∈ H and γ ∈ Γ, then

Proof Let f ∈ M
p
aτ + b c c
f( ) (cτ + d)−k = f0 (τ ) + f1 (τ ) ( ) + ⋯ + fp (τ ) ( ) .
cτ + d cτ + d cτ + d
1 ∂
Applying D = 2πi ∂τ to the above equality we get
aτ + b aτ + b −kc
f′ ( ) (cτ + d)−k−2 + f ( ) (cτ + d)−k−1
cτ + d cτ + d 2πi
2
c f1 (τ ) c
= f0′ (τ ) + f1′ (τ ) ( ) + (f2′ (τ ) − )( ) +
cτ + d 2πi cτ + d
p p+1
fp−1 (τ ) c fp (τ ) c
⋯ + (fp′ (τ ) − (p − 1) )( ) −p ( ) .
2πi cτ + d 2πi cτ + d

Theorem 21.12 (Ramanujan relations). The Eisenstein series E2 , E4 and E6 satisfy the
following differential equations
E22 − E4 E2 E4 − E6 E2 E6 − E42
E2′ = , E4′ = , E6′ = .
12 3 2
166
 E2
Proof It is enough to note that E2′ − 122 ∈ M4 (SL2 (Z)), E4′ − E23E4 ∈ M6 (SL2 (Z)) and E6′ −
E2 E6
2 ∈ M8 (SL2 (Z)) (these are a routine verification). Since these spaces are one dimensional,
we just need to see that their leading coefficient are clearly −1 −1 −1
12 , 3 and 2 respectively, and
the result follows.
Remark 21.3. It follows from Ramanujan relations that the subalgebra C[E2 , E4 , E6 ] ⊆
̃
M(SL2 (Z)) is also closed under derivation. In fact

̃
M(SL2 (Z)) = C[E2 , E4 , E6 ].

̃2 (Γ) ∖ M2 (Γ) such that

In general for every congruence subgroup Γ, there exists some φ ∈ M
̃
M(Γ) = M(Γ)[φ].
Remark 21.4. The notion of quasi-modular forms has proven to be more flexible to gen-
eralizations than modular forms. For instance they can be defined over moduli spaces of
higher dimensional varieties analog to elliptic curves, the so called Calabi-Yau varieties.
Some of these Calabi-Yau quasi-modular forms have been linked in last decades to several
q-expansions coming from theoretical physics (specifically from mirror symmetry) which can-
not be related to the classical quasi-modular forms. One example of such q-expansion is the
Youkawa coupling introduced by Candelas, de la Ossa, Green and Parkes in 1991
∞
qd
Y (τ ) = 5 + ∑ (nd d3 )
d=1 1 − qd
q q2
= 5 + 2875 + 609250 ⋅ 23 +⋯ q = e2πiτ ,
1−q 1 − q2
where the integers
nd = # of degree d rational curves P1 ↪ X in a generic quintic threefold,
i.e. X = {F = 0} ⊆ P4 for a generic F ∈ C[x0 , x1 , x2 , x3 , x4 ] homogeneous polynomial of degree
5. In order to explain briefly these relations let us introduce some more geometric notions.
Definition 21.4. A smooth projective variety X of dimension n is Calabi-Yau if it admits
a nowhere vanishing holomorphic n-form ω ∈ Ωnhol (X).
Example 21.2. 1. The only one dimensional Calabi-Yau varieties are elliptic curves,
since if C is a Calabi-Yau curve, then let ω ∈ Ω1hol (C) without zeros. It follows that
KC ≡ div(ω) = 0
and so 2g − 2 = deg(KC ) = 0, i.e. g = 1.
2. Calabi-Yau sufaces are of two types: the K3 surfaces which are those with
Ω1hol (X) = 0,
and the Abelian surfaces which are biholomorphic to complex tori of the form C2 /Λ
for some lattice Λ ⊆ C2 .

167
 3. In general, all Abelian varieties are Calabi-Yau. These varieties are smooth projec-
tive varieties biholomorphic to X ≃ Cn /Λ for some lattice Λ ⊆ Cn . In particular

dimC Ω1hol (X) = n.

4. Every smooth hypersurface X = {F = 0} ⊆ Pn+1 of degree n + 2 is Calabi-Yau. For n ≥ 2

Ω1hol (X) = 0,

and so these varieties are never Abelian varieties. In particular all smooth quartic
surfaces are K3 surfaces.
Definition 21.5. Let X be a complex manifold of dimension n. We say that a complex
manifold E of dimension n + k is a vector bundle of rank k over X if there exists a
holomorpic map
π∶E→X
such that X = ∪i∈I Ui admits an open covering such that for every i ∈ I there is a biholomorphic
map
∼
EUi ∶= π −1 (Ui ) Ð
→ Ui × Ck
fi

such that for every i, j ∈ I with Ui ∩ Uj ≠ ∅ the map

∼
(Ui ∩ Uj ) × Ck ÐÐÐ−1→ (Ui ∩ Uj ) × Ck
fj ○fi

is of the form
(fj ○ fi−1 )(u, v) = (u, fij (u) ⋅ v)
for some matrix fij (u) ∈ GLk (C) for each u ∈ Ui ∩Uj . A vector bundle is called a line bundle
if it has rank 1. For any vector bundle π ∶ E → X a section of E is a holomorphic map

s∶X →E

such that π ○ s = idX . We denote the vector space of sections of E by Γ(E).

Remark 21.5. In general a vector bundle E → X of rank k should be regarded as a family
of vector spaces of dimension k parametrized by X. Hence all linear algebra constructions
for vector spaces transport naturally to vector bundles, e.g. we can take products, direct
sums, duals, homs, tensor products, wedge products between vector bundles to produce new
ones.
Example 21.3. Every complex manifold X of dimension n has a tangent bundle T X
which is a vector bundle of rank n over X. Hence dimC T X = 2n. The sections of the
tangent bundle are called vector fields. The cotangent bundle T ∗ X the dual bundle of
T X. Its sections are exactly the holomorphic differentials

Γ(T ∗ X) = Ω1hol (X).

168
Example 21.4. For every compact Riemann surface C, and every divisor D ∈ Div(C), the
Riemann-Roch space L(D) corresponds to the space of sections of a line bundle E(D) → C
Γ(E(D)) = L(D).
In particular, if Γ ⊆ SL2 (Z) is any congruence subgroup, the space Mk (Γ) corresponds to
the sections of some line bundle E(D) → X(Γ) for some divisor D = ⌊div(f )⌋ ∈ Div(X(Γ))
with f ∈ Ak (Γ). In resume
Γ(E(⌊div(f )⌋)) = Mk (Γ).
Example 21.5. Movasati realized of a way to obtain the classical algebra of quasi-modular
forms C[E2 , E4 , E6 ] in a geometric way. For this he considered the following family of elliptic
curves: We say that a tuple (C, {α, ω}) is an enhanced elliptic curve if C is an elliptic
1
curve, α ∈ Ω1hol (C) and ω ∈ HdR 1
(C) such that 2πi ∫C α ∧ ω = 1. We say that two enhanced
elliptic curves are isomorphic
(C1 , {α1 , ω1 }) ≃ (C2 , {α2 , ω2 })
if there exists an isomorphism
∼
f ∶ C1 Ð
→ C2
such that
f ∗ α2 = α1 and f ∗ ω2 = ω1 .
We denote by T the moduli space of enhanced elliptic curves. It turns out that this moduli
space can be identified with the space
T = C3 ∖ {∆(t1 , t2 , t3 ) ∶= 27t23 − t32 }.
In fact, for every t = (t1 , t2 , t3 ) ∈ T the corresponding enhanced elliptic curve corresponds to
dx xdx
Xt = {y 2 = 4(x − t1 )3 − t2 (x − t1 ) − t3 } , α= , ω= .
y y
Therefore we have a (universal) family
ϕ ∶ X = {((x ∶ y ∶ 1), (t1 , t2 , t3 )) ∈ P2 × T ∶ y 2 = 4(x − t1 )3 − t2 (x − t1 ) − t3 } → T
of elliptic curves over T , such that
Xt = ϕ−1 (t).
In order to introduce a vector bundle on T we consider the de Rham cohomology bundle
of X → T , i.e. the vector bundle
1 1
HdR (X/T ) ∶= ⊔ HdR (Xt ).
t∈T
1
This is a vector bundle of rank dimC HdR (Xt ) = 2. Movasati’s idea to produce quasi-
modular was to recover the Ramanujan differential equation for the Eisenstein series, for
this he needed not only to consider sections of this vector bundle, but to have a way to
differentiate these sections.

169
Definition 21.6. Let E → X be a vector bundle. A connection is a map

∇ ∶ Γ(E) → Γ(T ∗ X ⊗ E)

such that the Leibniz rule

∇(f s) = df ⊗ s + f ∇s
holds for all f ∶ E → C holomorphic and s ∈ Γ(E). Fixing a vector field v ∈ Γ(T X) the
connection induces a (differential) map

∇v ∶ Γ(E) → Γ(E)

s ↦ ιv (∇(s)).

Example 21.6. Consider any family ϕ ∶ X → T of smooth complex projective varieties such
that it is a locally trivial fibration, i.e. there exists a smooth projective variety F such
that for each simply connected open subset U ⊆ T there exists a diffeomorphism (of C ∞ type)
f
XU ∶= ϕ−1 (U ) Ð
→U ×F

such that pr1 ○ f = ϕ. In particular all fibres Xt of the family are diffeomorphic to F , and
k k
so HdR (Xt ) ≃ HdR (F ). We can consider the de Rham cohomology bundle of the family,
which is a vector bundle on T given by
k k
HdR (X/T ) ∶= ⊔ HdR (Xt ).
t∈T

This bundle always comes with a natural connection called the Gauss-Manin connection.
For every simply connected open set U ⊆ T we can always find a flat frame s1 , . . . , sr ∈
k k
Γ(HdR (X/T )U ) such that ∇(si ) = 0 and each s ∈ Γ(HdR (X/T )U ) can be written as
r
s = ∑ f i si
i=1

k
for some f1 , . . . , fr holomorphic functions from HdR (X/T )U to C. In such a case the Gauss-
Manin connection for every vector field v ∈ Γ(T T ) is simply
r r
∂fi
∇v (s) = ιv (∑ dfi ⊗ si ) = ∑ si .
i=1 i=1 ∂v

Theorem 21.13 (Movasati). Let ϕ ∶ X → T be the universal family of enhanced elliptic

curves. Then there exists a unique vector field R ∈ Γ(T T ) such that the Gauss-Manin
connection is of the form
dx dx
y 0 1 y
∇R ( xdx )=( ) ( xdx ).
y
0 0 y

170
This vector field corresponds to
t2 ∂ ∂ t2 ∂
R = (t21 − ) + (4t1 t2 − 6t3 ) + (6t1 t3 − 3 )
12 ∂t1 ∂t2 3 ∂t3
which in turn corresponds to the differential equations
t2 t23
t′1 = t21 − , t′2 = 4t1 t2 − 6t3 , t′3 = 6t1 t3 − .
12 3
In other words
E2 E4 2E6
t1 = , t2 = , t3 = .
12 12 3 ⋅ 122
Remark 21.6. In general to any family X0 → T0 of Calabi-Yau varieties, Movasati has
attached an enhanced family of Calabi-Yau varieties X → T . Considering the Gauss-
Manin conection on the de Rham cohomology bundle, Movasati and his collaborators have
found similar vector fields R such that
⎛0 1 0 ⋯ 0 0⎞
⎛α1 ⎞ ⎜ ⎜
0 0 Y1 ⋯ 0 0 ⎟ ⎛α1 ⎞
⎟
⎜α ⎟ ⎜ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮ ⎟
⎟⎜ α⎟
∇R ⎜ 2 ⎟ = ⎜ ⎜ 2⎟
⎜ ⋮ ⎟ ⎜ ⎜ 0 0 0 ⋯ Yr−3 0 ⎟ ⎜
⎟ ⋮ ⎟
⎝ αr ⎠ ⎜⎜0 0 0
⎟
⋯ 0 −1⎟ ⎝αr ⎠
⎝0 0 0 ⋯ 0 0⎠
for several families of Calabi-Yau varieties. Once they prove the uniqueness of such a vector
field R, the functions t1 , . . . , tr such that
r
∂
R = ∑ t′i
i=1 ∂ti
form the algebra of Calabi-Yau quasi-modular forms. In the particular case of the one
parameter family of mirror quintic threefold, obtained after resolving the singularities of a
quotient of the so called Dwork family
Xt = {x50 + x51 + x52 + x53 + x54 − 5tx0 x1 x2 x3 x4 = 0},
Movasati has shown that
t33
Y1 =
54 (t51 − t5 )
is exactly the Youkawa coupling.

References
[ALRP10] Julian Aguirre, Álvaro Lozano-Robledo, and Juan Carlos Peral. Elliptic curves
of maximal rank. In Revista Matemática Iberoamericana, proceedings of the
conference “Segundas Jornadas de Teoria de Números, 2010.

171
[BvdGHZ08] Jan Hendrik Bruinier, Gerard van der Geer, Günter Harder, and Don Zagier.
The 1-2-3 of modular forms: lectures at a summer school in Nordfjordeid,
Norway. Springer Science & Business Media, 2008.

[DS05] Fred Diamond and Jerry Shurman. Modular forms, elliptic curves, and modular
curves. In A First Course in Modular Forms. Springer, 2005.

[Ful08] William Fulton. Algebraic curves: An Introduction to Algebraic Geometry.

Addison-Wesley, 2008.

[Har13] Robin Hartshorne. Algebraic geometry, volume 52. Springer Science & Business
Media, 2013.

[LR11] Álvaro Lozano-Robledo. Elliptic curves, modular forms, and their L-functions.
American Mathematical Society Providence, RI, 2011.

[MDG16] M Ram Murty, Michael Dewar, and Hester Graves. Problems in the theory of
modular forms. Springer, 2016.

[MMST13] Fabio Brochero Martinez, Carlos Gustavo Moreira, Nicolau Saldanha, and Ed-
uardo Tengan. Teoria dos números: um passeio com primos e outros números
familiares pelo mundo inteiro. Coleção Projeto Euclides, IMPA, 27, 2013.

[Mov19] Hossein Movasati. A differential introduction to modular forms and elliptic

curves. 2019.

[Neu13] Jürgen Neukirch. Algebraic number theory, volume 322. Springer Science &
Business Media, 2013.

[Sha94] IR Shafarevich. Basic Algebraic Geometry, Vol. 1. Springer-Verlag, 1994.

[Sil09] Joseph H Silverman. The arithmetic of elliptic curves, volume 106. Springer
Science & Business Media, 2009.

172