Alexandru Gica - Rational Points On Elliptic Curves PDF

Rational Points on Elliptic Curves
Alexandru Gica1
April 8, 2006
1
Notes, LATEXimplementation and additional comments by Mihai Fulger
2
Contents
1 Lecture I 5
1.1 Projective Geometry . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Equivalent definitions . . . . . . . . . . . . . . . . . . 5
1.1.2 The Geometry of the Projective Plane . . . . . . . . . 6
1.2 The Group Law of a Cubic . . . . . . . . . . . . . . . . . . . 7
1.2.1 The choice of origin . . . . . . . . . . . . . . . . . . . 10
1.3 The Weierstrass Normal Form . . . . . . . . . . . . . . . . . . 10
1.3.1 Explicit formulas for the group operation on C(Q) . . 11
1.3.2 The existence of a Weierstrass Normal Form . . . . . 14
2 Nagell-Lutzs Theorem 21
2.1 Discriminants and Resultants . . . . . . . . . . . . . . . . . . 21
2.2 The Nagell-Lutz Theorems . . . . . . . . . . . . . . . . . . . 25
3 Lecture III 31
3.1 Torsion points on rational elliptic curves . . . . . . . . . . . . 31
3.2 Test Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 A Theorem of Gauss 39
5 Mordell-Weils Theorem 43
6 Lecture VI 53
6.1 The Weak Mordell-Weil Theorem . . . . . . . . . . . . . . 53
6.2 Computing the rank of elliptic curves . . . . . . . . . . . . . . 57
7 Lecture VII 63
7.1 Eulers Equation . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.2 Computing the rank of nonsingular elliptic curves . . . . . . . 65
7.2.1 The curves Cp . . . . . . . . . . . . . . . . . . . . . . 67
7.3 Stories and conjectures . . . . . . . . . . . . . . . . . . . . . . 69
7.3.1 Congruent numbers . . . . . . . . . . . . . . . . . . . 69
7.3.2 The Birch and Swinnerton-Dyer Conjecture . . . . . . 70
3
4 CONTENTS
8 Lecture VIII 73
8.1 Algebraic Number Theory Prerequisites . . . . . . . . . . . . 73
8.2 Completing the proof of Mordell-Weils Theorem . . . . . . . 76
8.3 C17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9 Lecture IX 89
9.1 Test Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.1.1 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . 90
10 An unexpectedly hard problem 93
11 Integer points on elliptic curves 101

11.1 Thues Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 101
11.1.1 Proof of Thues Theorem and Diophantine Approxi-
mation . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.2 Ljunggrens Equation - a particular case . . . . . . . . . . . . 111
12 Generators for Elliptic Curves 113
13 Lecture XIII 123

Chapter 1
Lecture I
Definition 1.0.1. Let f Q[X, Y ] be a polynomial of degree 3 i.e. there

exist a, b, c, . . . , j Q such that f (X, Y ) = aX 3 + bX 2 Y + cXY 2 + dY 3 +
eX 2 + f XY + gY 2 + hX + iY + j. The associated set C Q2 defined as
the zero set of f is called an affine rational cubic.
1.1 Projective Geometry

On the subset of the affine space A3R \ {(0, 0, 0)} = R3 \ {(0, 0, 0)}, define
the relation by (a, b, c) (a0 , b0 , c0 ) if and only if there exists t 6= 0
such that t (a, b, c) = (a0 , b0 , c0 ) i.e. a0 = ta, b0 = tb, c0 = tc. It is easy to
see that is an equivalence relation. Denote the equivalence class in
A3R \ {(0, 0, 0)} of the element (a, b, c) by [a : b : c].
Definition 1.1.1. The factor set P2 = A3R \ {(0, 0, 0)}/ is called the
2-dimensional projective plane.
The same construction can be applied to any field, not necessarily R,
therefore we can define the rational or the complex projective plane.
Definition 1.1.2. Similarly, for any field K, we define
PnK = An+1
K \ {(0, 0, . . . , 0)}/
and call it the n-th dimensional projective space over K.

Remark 1.1.3. Let K L be a field extension. Then the canonical map
PnK PnL sending [x0 : . . . : xn ] PnK to [x0 : . . . : xn ] PnL is well defined
and injective.
1.1.1 Equivalent definitions

Remark 1.1.4. PnR can be regarded as the set of lines passing through the
origin of An+1
R .
5
6 CHAPTER 1. LECTURE I
Notice that P0 is the set of lines A1 . But A1 is a line, hence P0 consists

of a single point called the point at infinity of A1 .
Remark 1.1.5. We can regard Pn+1 as An+1 t Pn .
To see this algebraically, notice that Pn+1 = {[x0 : x1 : . . . : xn+1 ]|x0 6=
0} t {[0 : x1 : . . . : xn+1 ]|x1 , . . . , xn+1 are not all zero}. Denote U0 = {[x0 :
x1 : . . . : xn+1 ]|x0 6= 0}. We can construct a map : U0 An+1 by ([x0 :
x1 : . . . : xn+1 ]) = ( xx01 , . . . , xn+1
x0 ), because x0 6= 0. We can also construct a
map : An+1 U0 by ((x1 , x2 , . . . , xn+1 )) = [1 : x1 : x2 : . . . : xn+1 ]. It
is easy to see that and are well defined maps inverse to one another.
Therefore we can identify U0 to An+1 through . Denote V0 = Pn+1 \ U0 .
It is easy to check that the correspondence V0 Pn : [0 : x1 : . . . : xn+1 ]
[x1 : . . . : xn+1 ] is well defined and bijective.
For P0 this construction is obviously impossible so we define P0 by using
the previous remark.
We now give geometric meaning to the algebraic view above. We can
regard U0 as the set of directions in An+2 , passing through the origin, that
cut the hyperplane x0 = 1. Because a line not included in a hyperplane
cuts the hyperplane in at most one point, U0 identifies to An+1 . V0 can be
regarded as the set of directions in An+2 , passing through the origin, which
are parallel to x0 = 1. But this are exactly the lines in the hyperspace x0 = 0
and this hyperspace obviously identifies to An+1 . Therefore V0 is the set of
lines in An+1 passing through the origin, Pn .
V0 is called the hyperplane at infinity of An+1 , or the set of points at
infinity.
1.1.2 The Geometry of the Projective Plane

Let K be an arbitrary field. In this subsection, denote An = AnK and
Pn = Pnk .
Definition 1.1.6. Let An+1 be a hyperplane passing through the origin
of An+1 . The image [] of \ {(0, . . . , 0)} in Pn is called a hyperplane of
the projective space.
If n = 3, [] is called a line in the projective plane.
Remark 1.1.7. If v, w A3 generate a plane , then [] = {[ v +
w]|, K, (, ) 6= (0, 0)}.
Remark 1.1.8. Let F K[X0 , . . . , Xn ] be a homogeneous polynomial of
degree m. Then F (t x0 , . . . , t xn ) = tm F (x0 , . . . , xn ) t K, (x0 , . . . , xn )
An+1 n
K . Let [a0 : . . . : an ] PK . Notice that F (a0 , . . . , an ) = 0 F (ta0 , . . . , tan ) =
0 t 6= 0.
Definition 1.1.9. Under the assumptions of the previous remark, [a0 : a1 :
. . . : an ] PnK is called a zero of F , if F (a0 , . . . , an ) = 0. The remark above
proves that the definition is consistent.
1.2. THE GROUP LAW OF A CUBIC 7
Proposition 1.1.10. Let H be a subset of Pn . Then H is a hyperplane if

and only if H is the zero set of a homogeneous polynomial of degree 1 in
K[X0 , . . . , Xn ].
Proof: If H is a hyperplane, then there exists a hyperspace of An+1
such that H is the image of in Pn . Then is the zero set of a homogeneous
polynomial F of degree 1 in K[X0 , . . . , Xn ]. It is easy to see that H is the
zero set of F in Pn .
Conversely, if H is the zero set in Pn of a homogenous polynomial of
degree 1 in K[X0 , . . . , Xn ], let be the hyperspace in An+1 determined by
F . It is easy to see that the image of in Pn is H, so H is a hyperplane.
Proposition 1.1.11. Any two lines l1 and l2 in P2 intersect.
Proof: Let and be the two planes in A3 , passing through the origin,
that generate l1 and l2. If l1 6= l2, then 6= . We know that any two
distinct planes in A3 that have at least one common point intersect on a
line. The image of this line in P2 is the intersection point of l1 and l2.
1.2 The Group Law of a Cubic

Definition 1.2.1. A rational cubic is the zero set in P2Q of a homogeneous
polynomial of degree 3 in Q[X, Y, Z].
Analogously we can define a real or a complex cubic.
Remark 1.2.2. Let F Q[X, Y, Z] be a homogeneous polynomial of degree
3. Let Q K C be a field extension. The coefficients of F are rational,
but they are also complex numbers, so it makes sense to consider the complex
cubic C defined by F . It also makes sense to consider the cubic C(K) that
is the zero set of F in P2K .
If Q K L C are field extensions, then the canonical injective map
PK , P2L allows us to see C(K) as a subset of C(L).
2
Definition 1.2.3. Let C be the affine rational cubic determined by f (X, Y ) =

aX 3 + bX 2 Y + cXY 2 + dY 3 + eX 2 + f XY + gY 2 + hX + iY + j. Denote
f (X, Y, Z) = aX 3 + bX 2 Y + cXY 2 + dY 3 + eX 2 Z + f XY Z + gY 2 Z +
hXZ 2 + iY Z 2 + jZ 3 . f is called the homogenized polynomial of f . Denote
by C the rational cubic defined by f . In algebraic geometry language, C is
called the projective closure of C.
Definition 1.2.4. Let K be a field and let F K[X, Y, Z] be a homogeneous
polynomial of degree 3. Let C be the associated cubic. Let P = [a : b : c] be
a point on C i.e. F (a, b, c) = 0 and (a, b, c) A3K , not all zero. C is called
F F
nonsingular at P if X (P ), Y (P ), F
Z (P ) 6= (0, 0, 0).
F
If P is a nonsingular point for C, then the line of P2K given by X (P )
F F
X + Y (P ) Y + Z (P ) Z = 0 is called the tangent line to C at P and is
denoted TP C.
C is called nonsingular or smooth if it is nonsingular at every point.
Exercise 1.2.5. Let C be the complex cubic defined by the degree 3 homo-
geneous polynomial F . Then C is smooth if and only if F is irreducible.
Theorem 1.2.6 (Bezout). : Let F1 , F2 C[X, Y, Z] be homogeneous poly-

nomials of degrees m and n respectively. The zero sets of F1 and F2 are
called projective curves of degrees m and n respectively. If F1 and F2 have
no common factors (C[X, Y, Z] is factorial), then C1 and C2 are said to have
no common component, and then C1 C2 is a set of mn points counted with
multiplicities.
Corollary 1.2.7. Two complex cubics with no common component intersect

in exactly 9 points counted with multiplicities.
Theorem 1.2.8 (9 Points Theorem). Let C1 and C2 be two cubics with

no common components. Denote the nine points of intersection of C1 and
C2 by A1 , A2 , . . . , A9 . Assume C is a cubic that contains A1 , . . . , A8 . Then
A9 C.
Proof: The last 3 statements are powerful results of Algebraic Geom-

etry, and their proofs or the complete explanation of the multiplicity of
intersection of two curves in the projective plane exceed the level of this
course.
However, what you can accept is that two projective curves of degrees
m and n intersect in at most mn distinct points.
An intuitive proof can be provided for the 9 Points Theorem if the 9
points are all distinct (algebraic geometry translation: the multiplicity of
intersection of the two cubics at each point of intersection is 1). A cubic
over a field K is given by a degree three homogeneous polynomial. It is
obvious that multiplication of the 10 coefficients of the degree 3 polynomial
by a nonzero constant does not change the zero set of the polynomial. This
gives a bijective correspondence between the set of cubics over K and P9K .
The condition that a point [X : Y : Z] belongs to a cubic given by aX 3 +
bX 2 Y + cXY 2 + dY 3 + eX 2 Z + f XY Z + gY 2 Z + hXZ 2 + iY Z 2 + jZ 3 = 0 is,
by considering a, b, c, d, e, f, g, h, i, j as indeterminates, equivalent to that [a :
b : . . . : j] belongs to a certain hyperplane in P9K . So the set of cubics passing
through the distinct 8 points A1 , . . . , A8 corresponds to the intersection of
8 hyperplane in P9K . This intersection is a line l in P9K . This line passes
though the two distinct points corresponding to the two cubics C1 and C2 .
C contains the 8 points A1 , . . . , A8 if and only if the corresponding point
of C in P9K belongs to l. It can be proved that this happens if and only if
F = F1 + F2 , where F, F1 , F2 are the polynomials corresponding to C, C1
and C2 respectively, and , are in K. F (A9 ) = F1 (A9 ) + F2 (A9 ) = 0
A9 C.
1.2. THE GROUP LAW OF A CUBIC 9
Definition 1.2.9. Let C be a complex nonsingular cubic. Let P, Q C. If

P 6= Q then by Bezout, the line passing through P and Q cuts C in 3 points
with multiplicities. Let P Q be the third point of intersection of the line
P Q with C. It is possible that this point is one of the points P or Q. If
P = Q, then TP C cuts the cubic twice in P . By Bezout it must intersect C
again in a point P P . It is possible that this point is again P .
Remark 1.2.10. is not a group operation on C. To see this, we prove
that this operation has no neutral element. Assume P O = P P C.
Then the line P O cuts C twice in P for all P C. This means that P O
is tangent to C at P for all P C. Let F be the homogeneous polynomial
of degree 3 that defines C and let O = [a : b : c]. We have O TP C if and
F F
only if X (P )a + Y (P )b + F F
Z (P )c = 0. Let G(X, Y, Z) = X (X, Y, Z)a +
F F
Y (X, Y, Z)b+ Z (X, Y, Z)c = 0. G is a degree 2 polynomial which defines a
conic in P 2 . By Bezout we get that gcd(G, F ) 6= 1 or else the intersection of
the zero set of G and C is a finite set which contradicts O TP C P C.
But gcd(G, F ) 6= 1 contradicts the irreducibility of F .
Definition 1.2.11. Let C be a complex nonsingular cubic and fix a point
O C. For arbitrary P, Q C, define P + Q = O (P Q).
Theorem 1.2.12. In the conditions of the definition above, (C, +) is an
abelian group.
Proof: Commutativity of + is implied by the commutativity of
which is obvious.
Let P C. Then P + O = O (P O). The line P O cuts C at O, P and
O P . This line is obviously the same as the line O(O P ) which intersects
C in O, O P and P . Therefore O (O P ) = P . This proves that O is a
neutral element for + on C.
Let Q C, S = O O and Q0 = Q S. Then Q Q0 = S and Q + Q0 =
O S = O, hence Q0 is an inverse for Q.
The most interesting part of the proof is associativity. Let P, Q, R C.
To prove P + (Q + R) = (P + Q) + R, it suffices to prove P (Q + R) =
(P + Q) R. Let S be the intersection of the lines [P, Q + R] and [R, P + Q].
We have to prove S C. Define C1 as the union of the lines (P, Q, P Q),
(QR, O, Q+R) and (P +Q, S, R). Let C2 be the union of the lines (Q, R, Q
R), (O, P Q, P +Q) and (P, S, Q+R). Then C1 and C2 are cubics and their
intersection is the set of 9 points O, P, Q, R, P Q, Q R, P + Q, Q + R, S.
The cubic C passes through O, P, Q, R, P Q, Q R, P + Q, Q + R, thus
through S by the 9 point theorem.
Proposition 1.2.13. If C is a complex nonsingular cubic defined by a poly-
nomial F Q[X, Y, Z] and if C has a rational point O (i.e. a, b, c Q
such that [a : b : c] = O), then + is well defined on C(Q) and (C(Q), +)
is an abelian group.
Proof: To prove that + is well defined it suffices to prove that

P, Q C(Q) P + Q C(Q). It suffices to prove that P, Q C(Q)
P Q C(Q). Basically, this is a consequence of the fact that a degree
3 polynomial in one indeterminate with rational coefficients and 2 rational
roots has all its roots rational.
1.2.1 The choice of origin

In theorem 1.2.12 we have seen that the choice of a point O on a nonsingular
complex cubic defines a group operation + on C such that (C, +) is an
abelian group.
Let O and O0 be points on the nonsingular complex cubic C and let +
and be the abelian group operation they define. Denote G = (C, +)
and G0 = (C, ). For any P C, denote by P and by T P the inverse of
P in G and G0 respectively.
Theorem 1.2.14. G and G0 are isomorphic as abelian groups.
Proof: Let P, Q C be arbitrary points. We prove that P Q =

(P + Q) O0 . This is equivalent to O0 (P Q) = O (P Q) O0
O0 + O0 (P Q) = O (P Q) O (O0 (O0 (P Q))) = O (P Q),
which follows from the easy observation that X (X Y ) = Y X, Y C.
Let f : G G0 , f (P ) = P + O1 . We have f (P + Q) = P + Q + O0 =
(P + O0 ) + (Q + O0 ) O0 = (P + O0 ) (Q + O0 ) = f (P ) f (Q), therefore
f is a group homomorphism. It is easy to check that g : G0 G defined by
g(P ) = P O0 is a homomorphism that is an inverse for f , hence f and g
are isomorphisms.
1.3 The Weierstrass Normal Form

Let F (X, Y ) = Y 2 f (X) where f (X) = X 3 + aX 2 + bX + c and a, b, c Q.
Let C be the affine complex cubic determined by F .
Definition 1.3.1. The affine rational curve C(Q) defined by F (X, Y ) = 0,

is called an elliptic curve.
Remark 1.3.2. The projective closure C(Q) of C(Q) is given by F (X, Y, Z) =

Y 2 Z X 3 aX 2 Z bXZ 2 cZ 3 .
In 1.1.5 we have defined the hyperplane at infinity of PnK for any field K
by working in the first variable X0 . It is clear that nothing changes if we
work with any other variable. In particular, when working in P2 , we will be
working with points at infinity with respect to the last variable, Z.
The points at infinity of C are given by F (X, Y, 0) = 0 X 3 = 0
X = 0. This means that the only point at infinity of C is O = [0 : 1 : 0]
1.3. THE WEIERSTRASS NORMAL FORM 11
because [0 : Y : 0] = [0 : 1 : 0] Y 6= 0. Notice that O is also a point of

C(Q).

O is a nonsingular point for C. This is because F
Y ((0, 1, 0)) = 2 6= 0.
We will usually identify C(Q) and its projective closure C(Q).
F F

Definition 1.3.3. Call a point of P (x, y) of C nonsingular if X (P ), Y (P ) 6=
(0, 0). A point that is not nonsingular is called a singular point.
If P (x, y) is a nonsingular point for C, then the tangent line to C at P ,
F F
TP C is the line given by the equation X (P )(X x) + Y (P )(Y y) = 0.
Proposition 1.3.4. C is smooth if and only if all the complex roots of f

are distinct.
F
Proof: Assume P (x, y) is a singular point for C. Then X (x, y) =
0 F 2
f (x) = 0 and Y (x, y) = 2y = 0 f (x) = y = 0. These imply
f (x) = f 0 (x) = 0, which happens if and only if x is a multiple root of f .
1.3.1 Explicit formulas for the group operation on C(Q)

Assume C is nonsingular i.e. all the complex roots of f are distinct. By
1.2.13 there is a well defined group operation + on C(Q) with neutral
element O, the point at infinity of C, such that (C(Q), +) is an abelian
group.
Lets look more deeply into the structure of C. Let U = {[x : y : 1] P2C }.
Let : U A2C be the map defined by ([x : y : 1]) = (x, y). We know that
is bijective and we can see P2C as the union of (U ) = A2C and a line at
infinity which corresponds to a copy of P1C . We have seen that O = [0 : 1 : 0]
is the only point at infinity of C. It follows easily from the equations of C
and C that (C U ) = C. Let : A2C U be the inverse map of given
by ((x, y)) = [x : y : 1]. It is not hard to prove that if l is a line in A2C ,
then there exists a unique line l P2C such that (l) = l U (the idea is to
homogenize the equation of l with respect to Z). l is called the projective
closure of l.
The point of all these observations was to prove that if we restrict the
geometry of P2C to U , then we find the geometry of the affine plane A2C . We
use this to find affine formulas for the group operation on C for which we
have given only a projective definition.
With this in mind, consider C as a subset of C. We actually have
C = C O because the only point at infinity of C is O. For any line l in
the affine plane, l = l l , where l is the only point at infinity of l. For
ergonomic reasons, we agree to say the O is the point at infinity of C and
l is the point at infinity of l.
Proposition 1.3.5. Let P (x, y) C. Then TP C = TP C.

Proof: The equation of TP C is obtained by homogenizing the equa-

F F
tion of TP C. TP C is given by X (x, y)(X x) + Y (x, y)(Y y) = 0
0
f (x)(X x) + 2y(Y y) = 0. By homogenizing, we find the equation
of TP C, f 0 (x)(X xZ) + 2y(Y yZ) = 0 [1]. The equation of TP C is
F F F 0 2
X (x, y, 1)X + Y (x, y, 1)Y + Z (x, y, 1)Z = 0 f (x)X + 2yY + (y
ax2 2bx 3c)Z = 0 [2]. [1] and [2] are the same thing if and only if
f 0 (x)x 2y 2 = y 2 ax2 2bx 3c f 0 (x)x + ax2 + 2bx + 3c = 3y 2 . Because
P C, y 2 = f (x). So [1]=[2] if and only if (3x2 +2ax+b)x+ax2 +2bx+3c =
3y 2 = 3(x3 + ax2 + bx + c). And the last is obviously true.
Definition 1.3.6. Denote by x : A2C C and by y : A2C C the projections
on the x and y coordinates respectively.
Remark 1.3.7. The point at infinity of C, O = [0 : 1 : 0] is the point at
infinity of the line x = 0.
Remark 1.3.8. If P C, then O P is obtained by taking the line OP and
intersecting it again to C. The line OP and the projective closure of x = 0
intersect in O, so restricted to the affine, they are parallel. Therefore O P
is obtained by taking the other point of intersection of the parallel through
P to x = 0 and C.
Notice that the curve C is symmetric with respect to the x axis. If
P (x, y) C, then x(O P ) = x(P ) and y(O P ) = y(P ). Let P 0 (x, y) =
OP be the symmetric of P with respect to the x-axis. P +P 0 = O(P P 0 ) =
O O. If O O 6= O, then, because O is the only point at infinity of C,
O O = X C. Let X 0 be the symmetric of X with respect to the x-axis.
X 0 C. But O = O + O = O (O O) = O X = X 0 contradicts X 0 C.
Therefore P + P 0 = O P 0 = P .
Let P (x1 , y1 ) and Q(x2 , y2 ) be points on C(Q). We have seen in 1.2.13
that P + Q is an element of C(Q). So either P + Q = O, or P + Q C(Q).
We look, if possible, for formulas for x(P + Q) and y(P + Q).
If x1 = x2 and P 6= Q, then P and Q lie on a parallel to x = 0 and so
P Q = O P + Q = O (P Q) = O O = O.
Assume x1 6= x2 . Then the line P Q is the line of equation y = x + ,
y2
where = xy11 x 2
and = y1 x1 = y2 x2 . The intersection of P Q and
C is described by the equations
( x + )2 = x3 + ax2 + bx + c x3 + (a 2 )x2 + (b 2)x + (c 2 ) = 0

.
y =x+
We know that P and Q are solutions of this system. The third solution is
P Q. By Viete, x(P Q) = 2 ax1 x2 . Plunging this into the equation
of P Q yields y(P Q) = 3 a x1 x2 + . From these we get the
formulas:
x(P + Q) = 2 a x1 x2

. (1.3.1)
y(P + Q) = 3 + a + x1 + x2
If P = Q, then TP C is given by f 0 (x1 )(x x1 ) + 2y1 (y y1 ) = 0. If

y1 = 0, then, because C is smooth, f 0 (x1 ) 6= 0 and TP C is parallel to x = 0,
hence P P = O P + P = O. If y1 6= 0, then TP C is the line of equation
0 (x )
y = x + , where = f 2y 1
1
and = y1 x1 . Just as in the case
x1 6= x2 , we get the formulas:
x(2P ) = 2 a 2x1

. (1.3.2)
y(2P ) = 3 + a + 2x1
We have seen that the inverse of P (x, y) C is P 0 (x, y). This also
holds in C(Q).
Remark 1.3.9. These formulas can be used to give complete proofs for
Theorem 1.2.12 and Proposition 1.2.13 without any prior knowledge of Al-
gebraic Geometry, but only if the affine cubic has the particular form y 2 =
x3 + ax2 + bx + c i.e. the cubic is an elliptic curve.
Example 1.3.10. Let C(Q) be the affine rational cubic of equation y 2 =

x3 43x + 166. Prove that P (3, 8) C(Q). Compute P, 2P, 3P, 4P, 8P .
Compare P to 8P .
Solution: P C(Q) 82 = 33 43 3 + 166 64 = 27 129 + 166.

We apply the formula 1.3.2 to find the coordinates of 2P . We have
0 (3)
= f28 = 3943
16 = 1. = 8 3 = 11. x(2P ) = (1)2 2 3 = 5.
y(2P ) = ( x(2P ) + ) = ((5) + 11) = 16.
f 0 (5) 2 43
We now compute 4P . = 2(16) = 3(5)
32 = 1. = (16)
2
(1)(5) = 21. x(4P ) = (1) 2(5) = 11. y(4P ) = ((1)1121) =
32.
0 (11) 2
For 8P , we have: = f232 = 3116443 = 5. = 32 5 11 = 23.
x(8P ) = 52 2 11 = 3. y(8P ) = (5 3 23) = 8.
We notice that 8P = P 7P = 0 3P = 4P . So x(3P ) = x(4P ) =
x(4P ) = 11 and y(3P ) = y(4P ) = y(4P ) = 32.
Exercise 1.3.11. Let C(Q) be the elliptic curve defined by y 2 = x3 + 17.

Prove that P1 (2, 3), P2 (1, 4), P3 (2, 5), P4 (4, 9), P5 (8, 23) C(Q).
Find integers m, n such that P2 = mP1 + nP3 . Write P4 and P5 in terms
of P1 and P3 .
Prove that P6 = P1 + 2P3 and P7 = 3P1 P3 have integer coordinates
with y(Pi ) > 0, i = 6, 7.
Find a point on C(Q), different from (Pi )i=1,7 , with integer coordinates.
Example 1.3.12. Let C(Q) be the affine rational cubic defined by x3 +y 3 =

with Z . Let C 0 (Q) be the elliptic curve v 2 = u3 4322 . There is
a rational bijective application : C(Q) C 0 (Q) defined by (x, y) =
1
(12 x+y , 36 xy
x+y ).
Proof: Let v = 36 xy 1
x+y and u = 12 x+y . Then solving this system in
36+v 3
3
x, y gives x = 36+v and y = 36v + 36v

u u . (x, y) C(Q) 6u 6u =
v 2 = u3 4322 . This proves (C(Q)) C 0 (Q). It iseasy to check
that : C 0 (Q) C(Q) defined by (u, v) = 36+v 36v
6u , 6u is an inverse
for .
Remark 1.3.13. In the conditions of the example above, can be extended
to a group morphism : C(Q) C 0 (Q) by setting (O) = O0 , where
O([1 : 1 : 0]) and O0 ([0 : 1 : 0]) are the only points at infinity of C(Q) and
C 0 (Q) respectively. The proof that is a group morphism is of geometric
z
nature. ([x : y : z]) = (12 x+y , 36 xy
x+y ) is a projective transformation
0
which sends C to C , O to O and lines to lines.
Example 1.3.14. Find the rational solutions of y 2 + 432 = x3 .
Proof: The previous example proves that there is bijective correspon-
dence between the rational solutions of y 2 = x3 432 and the rational
solutions of u3 + v 3 = 1. This is a well known particular case of Fermats
Last Theorem and thus it has no nontrivial solutions. The trivial solutions
are (u, v) {(1, 0), (0, 1)}. The corresponding solutions of y 2 = x3 432 are
(1, 0) and (0, 1). (1, 0) = (12, 36) and (0, 1) = (12, 36).
1.3.2 The existence of a Weierstrass Normal Form

This is mostly an Algebraic Geometry section. The main goal is to prove
that any rational cubic that admits a flex point with rational coordinates
is projectively equivalent through a rational projective transformation to
an elliptic curve. We have not yet defined what a flex point is, but the
prototype is the point at infinity O[0 : 1 : 0] of an elliptic curve which had
the nice property O O = O. A projective transformation is basically a
linear change of coordinates in P2 , but beware that when restricted to the
affine, it may not look linear at all. A rational projective transformation is
just as usual a transformation with rational coefficients.
Let C P2C be a cubic given by the equation F (X, Y, Z) = 0 with
F C[X, Y, Z] a homogeneous polynomial of degree 3.
Definition 1.3.15. An application A : P2 P2 given by A([x : y : z]) =
[a11 x+a12 y+a13 z : a21 x+a22 y+a23 z : a31 x+a32 y+a33 z] with aij C i, j =
1, 3 is called a projective transformation of P2 if (aij )i,j=1,3 GL3 (C).
Let M = (aij )i,j=1,3 . Then M is a linear automorphism of A3 which

fixes the origin. Therefore M induces a well defined application P2 P2
which coincides with A. If we identify a point P = [x : y : z] in P2 with
the column vector > (x, y, z), then we have A(P ) = [M P ]. Notice that we
need M to be invertible, otherwise [0 : 0 : 0] would be in the image of A and
[0 : 0 : 0] is not an element of P2 .
From now on we identify A with M . A is bijective and its inverse is the

projective transformation given by the matrix M 1 .
Remark 1.3.16. The composition of two projective transformations is a
projective transformation. A projective transformation is bijective and its
inverse is again a projective transformation. The identity of P2 is a pro-
jective transformation and is the identity of a group structure on the set of
projective transformations.
Proposition 1.3.17. Two matrices M , N in GL3 (C) induce the same pro-
jective transformation on P2 if and only if there exists z C such that
M = z N.
Proof: For all P P2 we have [(zN ) P ] = [N (zP )] = [N P ] because
[a : b : c] = [za : zb : zc] for all [a : b : c] P2 and z C . Therefore we have
proved that proportional elements of GL3 (C) induce the same projective
transformation.
If M = (mij )ij and N = (nij )ij induce the same transformation of P2 ,
then [M P ] = [N P ] for all P P2 . This means that M P and N P are
collinear for all P A3 . This is equivalent to P and M 1 N P are collinear
for all P which in turn means that (lij )ij = L = M 1 N and I3 = (ij )ij
induce the same projective transformation. By successively replacing P by
[1 : 0 : 0], [0 : 1 : 0] and [0 : 0 : 1], we find that there exist z1 , z2 , z3 C such
that zj ij = lij i, j = 1, 3 i.e. the columns of I3 and L are proportional.
Therefore L is a diagonal matrix with diagonal entries (zj )j=1,3 . Now replace
P by [0 : 1 : 1], [1 : 0 : 1] and [1 : 1 : 0] successively to find z23 , z13 , z12 C
such that lij + lik = zjk (ij + ik ) for all i, j, k = 1, 3, j < k. Substituting we
have the system:
(zj zjk )ij + (zk zjk )ik = 0, i, j, k = 1, 3, j < k.
Choosing i = j gives zj = zjk for j < k. Choosing i = k gives zk = zjk for

j < k. Therefore zj = zk = zjk for all j < k. This proves z1 = z2 = z3 = z 1
for some z C . So M 1 N = z 1 I3 M = zN .
Definition 1.3.18. The proposition above allows us to identify the set of
projective transformations of P2 with the group GL3 (C)/C , that we will
denote by P GL2 (C). The identification is actually a group isomorphism.
For the factorization above to make sense, notice that C embeds in GL3 (C)
as the central subgroup of matrices of the form {zI3 |z C }. P GL2 (C) is
also called the group of linear automorphisms of P2 .
Remark 1.3.19. A projective transformation of P2 sends lines to lines.
This is because a linear automorphism of A3 sends planes to planes.
Exercise 1.3.20. The action of P GL2 (C) on the set P = {(P, Q, R)
(P2 )3 |P, Q, R are not colinear} is transitive. This means that for all (P, Q, R),
(P 0 , Q0 , R0 ) in P, there exists A P GL2 (C) such that A(P ) = P 0 , A(Q) =

Q0 and A(R) = R0 .
Proposition 1.3.21. For every two lines l1 , l2 in P2 there exists a projective

transformation sending l1 to l2 .
Proof: This is a consequence of the previous exercise, but it also has a

simpler proof. If l1 = l2 then the identity of P2 solves the problem. If l1 6= l2 ,
take P = l1 l2 . There is a rotation in A3 around the line which is the pre-
image of P through the canonical projection : A3 \ {(0, 0, 0)} P2 that
sends the plane which is the pre-image of l1 to the plane corresponding to
l2 . This rotation is an element of GL2 (C), hence it determines a projective
transformation of P2 which sends l1 to l2 .
Proposition 1.3.22. Let l1 and l2 be lines in the projective plane and let
P1 l1 and P2 l2 . There exists a projective transformation sending l1 to
l2 and P1 to P2 .
Proof: By the previous problem there exists a projective transformation

sending l1 to l2 . Using the simple fact that composing two projective trans-
formations we find again a projective transformation, we reduce to the case
l1 = l2 . We must find a linear automorphism of P2 which keeps l1 fixed and
sends P1 to P2 . Through the canonical projection A3 \ {0} P2 we make
this an affine problem which translates to finding a linear automorphism of
A3 which fixes the origin, fixes the plane corresponding to the line l1 and
sends the line corresponding to P1 into the line corresponding to P2 . It is
not hard to find a plane rotation with the properties above and this rotation
then determines a solution to the problem.
So far we have only treated classical topics of projective geometry. The

results above are valid in any field. We will introduce some more algebraic
geometry from now on.
Definition 1.3.23. Let C and C 0 be two projective curves. We say that

C is projectively equivalent to C 0 if there exists P GL2 (C) such that
(C) = C 0 .
Remark 1.3.24. Let C be a complex cubic, P GL2 (C) and let A

GL3 (C) be a representant for . Then the homogeneous polynomial F (X, Y, Z)
gives the equation of C if and only if the homogeneous polynomial F
A1 (X, Y, Z) gives the equation of (C). In particular (C) is a complex
curve of the same degree as C.
I said that the definition of the multiplicity of intersection of two pro-

jective curves at a common point exceeds the level of this course, but I will
introduce the multiplicity of intersection of a line and a curve.
Definition 1.3.25. Let C be a curve in P2 given by the equation F (X, Y, Z) =

0 with F a homogenous polynomial of degree d. Let l be a line in P2 and
assume P is a common point for C and l.
Suppose that l is not a component of C i.e. the linear form defining the
equation of l does not divide F . Let Q L, Q 6= P . The multiplicity of
intersection of l and C at P is by definition the order of the root 0 in the
polynomial f () = F (P + Q). Denote it by i(C, l, P ).
If l C then set i(C, l, P ) = . Throughout, when working with the
intersection multiplicity of a curve and a line at a point, we assume this
multiplicity to be finite i.e. l is not a component of C.
Remark 1.3.26. 0 is in fact a root of f in the equation above because

f (0) = F (P ) = 0 because P C and C is the zero set of F .
The order of a root of a polynomial is obviously upper bounded by the
degree of the polynomial. In our case, i(C, l, P ) deg(f ) = deg(F ) = d.
Since 0 is a root of f , we also have 1 i(C, l, P ). P
In this particular case, Bezouts Theorem states that P Cl i(C, l, P ) =
d.
Let f () = F (P + Q) = F (P ) + R(P, Q) + 2 H(P, Q) for some

df F F
R(P, Q), H(P, Q) C. Then R(P, Q) = d (0) = X (P +0Q)x(Q)+ Y (P +
F
0 Q)y(Q) + Z (P + 0 Q)z(Q) .
F F F
R(P, Q) = (P )x(Q) + (P )y(Q) + (P )z(Q),
X Y Z
where x(Q), y(Q), z(Q) are fixed representatives of Q. Based on this we give
the following definition:
Definition 1.3.27. Let C be a projective curve of degree d and let P C.

Define the tangent space at P to C as the linear subspace TP C of P2 given
F F
by X (P )X + Y (P )Y + F
Z (P )Z = 0.
A line l passing through P is called tangent at P if l TP C.
If i(C, l, P ) = 1, we say that l and C meet transversally at P .
Remark 1.3.28. TP C is either a line or the the whole P2 . It is P2 if and

only if all the partial derivates of F at P vanish.
This remark shows that P is a smooth point of C if and only if TP C is
a line, and in this case the definitions of the tangent space and tangent line
agree.
A point Q in P2 different from P is in TP C if and only if i(C, P Q, P ) 2.
Exercise 1.3.29. Let C be a complex curve of degree d, let l P2 be a line

and let P C l. Prove that i(C, l, P ) = i((C), (l), (P )). In particular
prove that if Q C, then T(Q) ((C)) = (TQ C).
We now introduce the notion of a flex point.

Definition 1.3.30. Let P be a smooth point on a complex curve C of degree

d. P is called a flex point for C if and only if i(C, TP C, P ) > 2.
Remark 1.3.31. If P is a flex point on a complex cubic C, then it is easy

to see that TP C cuts C in P 3 times (multiplicity 3) and cuts C in no other
point. Notice that this implies P P = P .
Remark 1.3.32. Let C(Q) be an elliptic curve given in homogeneous co-

ordinates by Y 2 Z = X 3 + aX 2 Z + bXZ 2 + cZ 3 . By definition, an ellip-
tic curve is smooth. Hence the point O[0 : 1 : 0] is smooth. C is given
by F (X, Y, Z) = Y 2 Z X 3 aX 2 Z bXZ 2 cZ 3 . TO C is given by
Z = 0. Let Q = [1 : 0 : 0]. To compute i(C, TO C, O) we have to com-
pute f () = F ((0, 1, 0) + (1, 0, 0)) = F ((, 1, 0) = 3 . The order of the
root 0 is 3, hence O is a flex point for C.
Remark 1.3.33. It can be proved that a smooth complex curve of degree

d 3 admits at least a flex point.
Actually it can be proved that the flex points of the smooth curve C given
by F (X, Y, Z) = 0 are the intersection points of C and the Hessian of C.
By
definition, the Hessian H(C) is the complex curve given by H(F ) =
F F F
XX XY XZ
F F F
det Y X Y Y Y Z
= 0. H(F ) is a homogeneous polynomial of
F F F
ZX ZY ZZ
degree 3(d2). By Bezout it follows that C and H(C) either have a common
component which must be C or meet in 3d(d 2) points.
The following theorem is the starting point in proving the existence of a

normal Weierstrass form for a special class of cubics:
Theorem 1.3.34. Let C be a projective curve (not necessarily smooth) and

assume that it has a flex point P . Then C is projectively equivalent to a cubic
of equation Y 2 Z F (X, Z) = 0, where F is a homogeneous polynomial of
degree 3.
Proof: By proposition 1.3.22 there exists a projective transformation

sending P to O[0 : 1 : 0] and TP C to the line Z = 0. 1.3.24 says that
the image of C through this projective transformation is again a cubic.
Therefore we can assume P = O and TP C = {Z = 0}. Suppose that the
equation of C is a0 X 3 +X 2 (a1 Y +a2 Z)+X(a3 Y 2 +a4 Y Z +a5 Z 2 )+(a6 Y 3 +
a7 Y 2 Z + a8 Y Z 2 + a9 Z 3 ). [0 : 1 : 0] C implies a6 = 0. Since O is assumed
to be a flex point for C, we must have C {Z = 0} = {[0 : 1 : 0]}. This
says that the equation a0 X 3 + a1 X 2 Y + a3 XY 2 = 0 gives only the solution
[0 : 1 : 0] i.e. it has only solutions of the form (0, y, 0) over the complex
numbers. This happens only if a1 = a3 = 0 and a0 6= 0. Computing
the partials at (0, 1, 0) of the polynomial giving the equation of C we find

( X , Y , Z ) = (0, 0, a7 ). By definition of a flex point, O is smooth therefore
not all the partials at (0, 1, 0) vanish and a7 6= 0. Multiplying the coefficients
of the homogeneous equation defining C by a nonzero constant does not
change its solutions therefore we can assume a7 = 1. The equation of C is
now: Y 2 Z + a4 XY Z + a8 Y Z 2 + (a0 X 3 + a2 X 2 Z + a5 XZ 2 + a9 Z 3 ). It has
the form
Y Z + sXY Z + tY Z 2 + A(X, Z) = Z(Y 2 + sXY + tY Z) + A(X, Z),
for some homogeneous polynomial A of degree 3 in X, Z. We have Y 2 +

sXY + tY Z = Y (Y + sX + tZ) = (Y + 2s X + 2t Z)2 ( 2s X + 2t Z)2 . For Y 0 =
Y + 2s X + 2t Z, we get the equation for C: ZY 02 Z( 2s X + 2t Z)2 +A(X, Z) = 0
which is of the form Y 02 Z F (X, Z) with F a homogeneous polynomial of
degree 3 in X, Z. Since the application [X : Y : Z] [X : Y 0 : Z] induces a
projective transformation, we are done.
Remark 1.3.35. If C in theorem 1.3.34 is a rational cubic i.e. given by

an equation with rational coefficients, and P is a flex point with rational
coordinates, then all the projective transformations used in the proof of the
theorem are rational.
Theorem 1.3.36. If C is a nonsingular complex cubic, then C is pro-

jectively equivalent to a curve of equation Y Z = X 3 + aXZ 2 + bZ 3 with
4a3 + 27b2 6= 0.
Proof: By 1.3.33, a smooth complex cubic always has a flex point.

By theorem 1.3.34, C is projectively equivalent to a curve of equation
3 . a 6= 0 or else C contains the line {Z = 0}
Y 2 Z = aX 3 + bX 2 Z +cXZ 2 + dZ
and is no longer smooth. Replacing X, Y by a1 X, a1 Y is a projective
transformation and the equation of the image C 0 of C through this transfor-
3 = X 3 +b0 X Z +c0 XZ 2 +d0 Z 3 .
mation is Y 2 Z = X 3 + bX 2 Z + acXZ 2 + a2 dZ
The projective transformation defined by X X3Z 3 sends C 0 to the curve
0 3 0 2 0 2
C of equation Y 2 Z = (Xb 27
Z)
+ b0 (Xb9Z) Z + c0 (Xb3Z)Z + d0 Z 3 . The
0
coefficient of X 2 Z is b9 + b0 19 = 0. Therefore the equation of C is of
the form Y 2 Z = X 3 + aXZ 2 + bZ 3 and C is projectively equivalent to C
through the composure of the projective transformations above. The condi-
tion 4a3 + 27b2 6= 0 is, as we will see in the next course, the condition that
the discriminant of f (X) = X 3 + aX + b is nonzero. We will also see that
this is equivalent to the condition that f has distinct roots hence to that C
is nonsingular.
Definition 1.3.37. If C is a smooth complex cubic given by the equation

Y 2 Z = X 3 +aXZ 2 +bZ 3 , then we say that C is in Weierstrass normal form.
Theorem 1.3.36 above states that any smooth complex cubic is projectively
equivalent to a curve in Weierstrass normal form.
Theorem 1.3.38 (Weierstrass Normal Form). Let C be a smooth ratio-

nal complex cubic. Suppose that C admits a flex point P with rational coor-
dinates. Then C is projectively equivalent to an elliptic curve. Moreover we
can assume that the equation of this elliptic curve is Y 2 Z = X 3 +aXZ 2 +bZ 3
with a, b Z and 4a3 + 27b2 6= 0.
Proof: If C is rational and P has rational coordinates, then all the

projective transformations that appear in the proof of theorem 1.3.36 are
rational. Therefore C is projectively equivalent to a complex curve of equa-
tion Y 2 Z = X 3 + a0 XZ 2 + b0 Z 3 with a0 , b0 Q and 4a03 + 27b02 6= 0. There
exist A, B, C Z, C 6= 0 such that a0 = C A
and b0 = B 2
C . [X : Y : Z] [C X :
C 3 Y : Z] defines a projective transformation that sends the complex curve
to the curve of equation Y 2 Z = X 3 + AC 3 XZ 2 + BC 5 Z 3 . Set a = AC 3
and b = BC 5 . Then 4a3 +27b2 = C 9 (4A3 +27B 2 C) = C 12 (4a03 +27b02 ) 6= 0.
Determining whether a rational cubic has a flex point or not is a com-

pletely wild problem. No method effective or not is known, so the condition
that C has a rational flex point is quite restrictive.
We will sometimes say that a smooth rational cubic is in Weierstrass
normal form if it is determined by an equation of the form Y 2 Z = X 3 +
aX 2 Z + bXZ 2 + cZ 3 with a, b, c Z.
Chapter 2
Nagell-Lutzs Theorem
2.1 Discriminants and Resultants

The main goal of this algebraic digression is to present some properties of
the discriminant of a polynomial. We will mainly use it to prove that a
polynomial with coefficients in an arbitrary field k has a multiple root in an
algebraic closure of k if and only if the discriminant of the polynomial is 0.
We will also find some formulas for the discriminant of a polynomial.
Let A be an integral domain, let k be its field of fractions and let be
an algebraic closure for k. Let f A[X] be a monic polynomial. If k was
R or C then we had the notion of a derivate for f and in both cases we had
the formula f 0 (X) = nan X n1 + (n 1)an1 X n2 + . . . + a1 if f (X) was
an X n + an1 X n1 + . . . + a1 X + a0 .
Definition 2.1.1. If f A[X], f (X) = an X n + an1 X n1 + . . . + a1 X + a0

then define the formal derivate of f by f 0 (X) = nan X n1 +(n1)an1 X n2 +
. . . + a1 .
The following properties of the formal derivate are easy exercises:
1. (f + g)0 = f 0 + g 0 for all f, g k[X].
2. (a f )0 = a f 0 for all f k[X] and a k.
3. (f g)0 = f 0 g + f g 0 for all f, g k[X].
4. If k is a field of characteristic 0, then f 0 = 0 f is a constant

polynomial. If k is of characteristic p, then f 0 = 0 g k[X] such
that f (X) = g(X p ).
Definition 2.1.2. Let f, g A[X], f (X) = an X n + . . . + a0 and g(X) =

bm X m + . . . + b0 . The resultant of f and g, is defined as the determinant of
21
22 CHAPTER 2. NAGELL-LUTZS THEOREM
the (n + m) (n + m) matrix:

an an1 . . . a0 0 . . .

0 an . . . a1 a0 0

0 0 an . . . a1 a0
..
.
Res(f, g) = det
bm bm1
.
. . . b0 0 . . .

0 bm . . . b 1 b0 0

0 0 bm . . . b1 b0

..
.
The first m rows are obtained by permuting circularly the 1 (n + m) vector

(an , . . . , an , 0, . . . , 0) and the last n rows are obtained by permuting circularly
the 1 (n + m) vector (bm , . . . , b0 , 0, . . . , 0).
Definition 2.1.3. By definition, the discriminant of a polynomial is
1
f = (1)n(n1)/2 Res(f, f 0 ).
an
The bare definition of the resultant of two polynomials may look a bit
scary, but we will see how it appears naturally in trying to solve the following
problem: Given two polynomials with coefficients in A, when can we say that
they have a common factor? Simple ring theory proves that it is enough to
consider the problem in k, the quotient field of A. The following proposition
proves that the resultant of these polynomials plays a decisive role in solving
this problem. The proof of the proposition sheds some light on how the
definition of the resultant first came to life.
Proposition 2.1.4. If f, g A[X], then f and g have a nonconstant com-
mon divisor if and only if Res(f, g) = 0.
Proof: It is enough to consider the problem in k, the quotient field of
A. Let deg(f ) = n and deg(g) = m. First notice that f and g have a
nonconstant common factor if and only if there exists a nonzero polynomial
h k[X] such that f |h, g|h and deg(h) m + n 1. If f, g have a common
factor d, then f = d f 0 and g = d g 0 for some polynomials f 0 , g 0 k[X].
For h = df 0 g 0 , we have f |h, g|h and deg(h) = deg(d) + deg(f 0 ) + deg(g 0 ) =
n + m deg(d) n + m 1. Conversely let h = f f 0 and h = g g 0 . If
f and g have no common factor then f |h = gg 0 f |g 0 which is impossible
because deg(g 0 ) = deg(h) deg(g) n + m 1 m = n 1. Therefore f
and g have a common factor.
Let Vf be the kvector space of polynomials of degree least of equal
to n + m 1 divisible by f . It is easy to see that a basis for Vf is
given by {f, Xf, X 2 f, . . . , X m1 f }. Similarly define Vg . It has the basis
{g, Xg, X 2 g, . . . , X n1 g}. f and g have a nonconstant common factor if
2.1. DISCRIMINANTS AND RESULTANTS 23
and only if there exists a nonzero polynomial of degree least or equal to

m + n 1 divisible both by f and g. This is equivalent to Vf Vg 6= 0
i.e Vf and Vg are not in direct sum in the vector space V of polynomi-
als of degree least or equal to n + m 1. A basis for V is given by
{1, X, X 2 , . . . , X n+m1 }. Now Vf and Vg are not in direct sum if and only if
{X m1 f, . . . , Xf, f, X n1 g, X n2 g, . . . , Xg, g} are linearly dependent over k
in V . By writing these elements of V in the basis {X n+m1 , . . . , X, 1}, the
condition that they are linearly dependent is equivalent to that the determi-
nant of the resulting matrix of coefficients is zero. The matrix of coefficients
is the same as the matrix of Res(f, g).
Corollary 2.1.5. If f A[X], then f has a multiple root in if and only

if f = 0.
Proof: f has a multiple root in if and only if f and f 0 have a non-

constant common factor in k[X].
We now give some formulas for the resultant and for the discriminant of
a polynomial.
Proposition 2.1.6. If f, g A[X], f (X) = an (X 1 ) . . . (X n ) and

g(X) = bm (X Q . (X m ) with 1 , . . . , n , 1 , . . . , m , then
1 ) . .Q
Res(f, g) = an bm i=1 m
m n n
j=1 (i j ).
Proof: Consider F, G A[X1 , . . . , Xn , Y1 , . . . , Ym ][X] defined by
F (X1 , . . . , Xn , Y1 , . . . , Ym )(X) = an (X X1 ) . . . (X Xn ),
G(X1 , . . . , Xn , Y1 , . . . , Ym )(X) = bm (X Y1 ) . . . (X Ym ).
Clearly f = F (1 , . . . , n , 1 , . . . , m ) and g = G(1 , . . . , n , 1 , . . . , m ) in
[X1 , . . . , Xn , Y1 , . . . , Ym ].
A = A[X1 , . . . , Xn , Y1 , . . . , Ym ] is an integral domain, so Res(F, G) is a
well defined element of A and it is clear from the definitions that
Res(F, G)(1 , . . . , n , 1 , . . . , m ) = Res(f, g).

Qn Qm
Hence it suffices to prove Res(F, G) = am n bm
n
i=1 j=1 (Xi Yj ). The
formalism of a complete proof is tedious, but the idea is simple. For all
i = 1, n, j = 1, m we prove that Xi Yj |Res(F, G). Let Aj be the polynomial
ring obtained from A by removing the variable Yj . Let Fij , Gij and R(F, G)ij
be the polynomials in Aj [X] obtained from F , G and R(F, G) respectively
by replacing Yj with Xi . We have Xi Yj |Res(F, G) Res(F, G)ij = 0.
But clearly in Aj , Res(F, G)ij = Res(Fij , Gij ) and the last is 0 because Fij
and Gij have an obvious common Q factor, namely X Xi . These can be
formalized rigourously to prove ni=1 m
Q
j=1 (X i Yj )|Res(F, G).
Let V = ni=1 m
Q Q
j=1 (Xi Yj ) A. By expanding F and G we see that
the variable Xi appears in the matrix of Res(F, G) only on the first n lines
and on these lines it appears in degree 1, hence the degree of Res(F, G) in
Xi is n for all i = 1, n. Similarly we prove that the degree of Res(F, G) in
each Yj is m. From these we conclude that Res(F, G) = V with A.
Notice that = Res(F, G)(1, . . . , 1, 0 . . . , 0) = Res(an (X 1)n , bm X m ).
Computing Res(an (X 1)n , bm X m ) = am n
n bm is a good exercise.
Qn Qm
Corollary 2.1.7. Res(f, g) = am
n i=1 g(i ) = (1)n+m bnm j=1 f (j ).
Proposition 2.1.8. If f A[X], f (X) = an (X 1 ) . . . (X n ) and

i i = 1, n, then
2
Y
f = a2n2
n
(i j ) .
1i<jn
Proof: Use f 0 (i ) =
Q
j6=i (i j ).
These formulas are often useful for computations and could have just as
well been used to define both the resultant and the discriminant. However
they are not quite useful in proving more theoretical results. For example
dont make clear why Res(f, g) and f are not just elements of but also
of A.
The following is a very important property of the resultant of two poly-
nomials.
Theorem 2.1.9. If f, g A[X], then R(f, g) (f, g) i.e. R(f, g) is in the

ideal generated by f and g i.e. there exist polynomials u, v A[X] such that
uf + vg = Res(f, g).
Proof: Consider R(f, g) as the determinant of a matrix with entries

in A[X]. For all i = 1, n + m 1, add the column i in Res(f, g) multi-
plied by X n+mi to the last column, m + n. The only difference between
the matrix of Res(f, g) and the new matrix, M , is the last column, which is
> (f X m1 , . . . , f X, f, gX n1 , . . . , gX, g). The determinant of M is Res(f, g).
Expanding det M by the last column gives Res(f, g) (f, g).
The following remark will be used intensively throughout the course.
Remark 2.1.10. Let f (X) = X 3 + aX 2 + bX + c be a polynomial with

coefficients in an arbitrary field k. We have:
f = 4a3 c + a2 b2 + 18abc 4b3 27c2 .

2.2. THE NAGELL-LUTZ THEOREMS 25
2.2 Torsion points on elliptic curves, The Nagell-

Lutz Theorem
Let C(Q) be the elliptic curve given by y 2 = x3 +ax2 +bx+c. Assume that C
is nonsingular. This is equivalent to saying that f (X) = X 3 + aX 2 + bX + c
has 3 distinct complex roots. Notice using 2.1.5 that the condition for C
to be smooth is equivalent to f 6= 0, where f = 4a3 c + a2 b2 + 18abc
4b3 27c2 .
Lets see that up to a rational change of coordinates we can assume that
0
a, b, c are integers. There exist a0 , b0 , c0 , d Z, d 6= 0 such that a = ad ,
0 0
b = bd and c = cd . Then y 2 = x3 + ax2 + bx + c (d3 y)2 = (d2 x)3 +
a0 d(d2 x)2 + b0 d3 (d2 x) + c0 d5 . Therefore up to the change of coordinates
(x, y) (x/d2 , y/d3 ) we can assume that a, b, c are integers.
To make writing easier, identify C(Q) with C(Q). Recall that C(Q) has
an abelian group structure with neutral element the only point at infinity
of C, O[0 : 1 : 0].
Definition 2.2.1. A point P on C(Q) is called a torsion point if it is a

torsion element of the group C(Q) i.e. there exists an integer n 6= 0 such
that nP = O. The set of torsion points on C(Q) forms an abelian subgroup
of C(Q) that we will denote M. For P M, denote by o(P ) the least
positive integer n such that nP = O. o(P ) is called the order of P .
The main goal of this section is to prove the Nagell-Lutz Theorem. It

will provide us an effective method for determining the torsion points of
C(Q) by imposing strong necessary conditions on them.
Theorem 2.2.2 (Nagell-Lutz). Let P (x, y) be a torsion point on C(Q).

We assume P 6= O. Then x, y Z. Moreover y = 0 or y|f .
Towards the proof of Nagell-Lutzs Theorem we make the following re-

mark:
Remark 2.2.3. There exist g, h Z[X] such that gf + hf 0 = f . This

follows from 2.1.9 but if you have a strong computational disposition you
can check that the formulas
g(X) = (18b 6a2 )X (4a3 15ab + 27c)

h(X) = (2a2 6b)X 2 + (2a3 7ab + 9c)X + (a2 b + 3ac 4b2 )
provide a solution for the problem.
Proof of 2.2.2: We begin our investigation of M\O with the points of

order 2. We have 2P = O P (x, y) = (P )(x, y) y = y y(P ) = 0.
If y = 0 then y 2 = x3 + ax2 + bx + c x3 + ax2 + bx + c = 0 x Z. An
elementary proof for this is to notice that a rational root a/b with gcd(a, b) =
1 of a polynomial an X n + . . . + a0 with integer coefficients satisfies b|an and

a|a0 . A quick algebraic proof is just knowing that Z is integrally closed. We
have proved o(P ) = 2 y = 0, x Z.
Assume now that y(P ) 6= 0 i.e. P 6= P . Consider the following change
of coordinates (t, s) = ( xy , y1 ) (x, y) = ( st , 1s ). y 2 = x3 + ax2 + bx + c
1 3 2
s2
= st 3 + a st 2 + b st + c s = t3 + at2 s + bts2 + cs3 .
Let C 0 be the cubic defined by s = t3 + at2 s + bts2 + cs3 and let O0 =
(0, 0). In the following short escapade to algebraic geometry I prove that
C(Q) and C 0 (Q) are isomorphic as groups and find explicit formulas for
an isomorphism between them. I prefer the geometric argument to a
mechanical verification that would lead to the same result.
Let : P2 P2 be defined by ([x : y : z]) = [x : z : y]. It is obvious
that is a linear automorphism of P2 . restricts to a bijection : C C 0 .
This is easily seen by considering the equations of the projective closures
C : y 2 z = x3 + ax2 z + bxz 2 + cz 3 and C 0 : su2 = t3 + at2 s + bts2 + cs3 . We
have (O) = ([0 : 1 : 0]) = ([0 : 1 : 0]) = [0 : 0 : 1] = (0, 0) = O0 and
(P (x, y)) = ([x : y : 1]) = ([x : y : 1]) = [x : 1 : y] = ( xy , y1 ) if y 6= 0.
Similarly (P (x, 0)) = [x : 0 : 1]. Because and 1 are linear rational
automorphisms, their restrictions to C(Q) and C 0 (Q) induce bijections that
we will also denote : C(Q) C 0 (Q) : 1 . The simple fact that and
1 are restrictions of rational linear transformations is enough to give a
convincing proof that they are group homomorphisms. Because they are
inverse to each other, they are actually group isomorphisms. Notice that
C 0 is symmetric with respect to the point (0, 0). Let P 0 (x, y) C 0 (Q). Let
Q be the point of coordinates (x, y). I want to prove that P 0 + Q = O0
in C 0 (Q). Because Q is the symmetric of P 0 with respect to O0 , we have
P Q = O0 (recall that for arbitrary points X, Y of a smooth cubic curve,
X Y was defined as the third point of intersection of the line XY with C
if X 6= Y and as the third point of intersection of TX C and C, by imagining
that TX C already cuts C in X twice). To prove P 0 + Q = O0 , it is enough
to prove O0 O0 = O0 . Let O0 O0 = R. This implies R O0 = O0 . On the
other hand, R O0 is the symmetric of R with respect to O0 , hence R = O0 .
We have proved that P 0 = Q.
Back from the escapade. We had P (x, y) a torsion point on C(Q) with
y 6= 0 and o(P ) > 2. We first want to prove that x and y have to be integers.
A
The idea is to write y = B with gcd(A, B) = 1, B 1 and then prove that
for all prime numbers p, we have p 6 |B. This would prove that B = 1, hence
y Z. Because x would be a rational number, root of the polynomial with
integer coefficients X 3 + aX 2 + bX + (c y 2 ), x would have to be an integer.
The hard part is proving that B has no prime divisors.
Definition 2.2.4. Let p be a prime number and let x = m n be a rational
number. If m 6= 0 then there exist a, b N such that pa |m, pa+1 6 |m and
pb |n, pb+1 6 |n. Define vp (x) = a b. Set vp (0) = . If vp (x) = 0, then we
say that x is prime to p. Some of the most important properties of vp are:
1. vp : Q Z is a well defined group homomorphism, called the p-adic

valuation on Q.
2. vp (n) 0 for any integer n Z. Even more, vp (x) 0 for every

prime number p if and only if x Z or x = .
3. vp (x + y) vp (x) if vp (x) = vp (y).
4. vp (x + y) = min{vp (x), vp (y)} if vp (x) 6= vp (y).
Assume there exists a prime number p dividing B. Then there exist

m, n Z with gcd(m, n) = 1, p 6 |mn and there exists an integer r 1 such
that y = npm t u with
r . Also there exist integers u, v, t such that x = p v
gcd(u, v) = 1 and p 6 |uv. From the equation we get
m2 u3 1 u2 1 u1
= + a + b t + c.
n2 p2r v 3 p3t v 2 p2t vp
2 3
u 3t 2
If t 0, then nm2 p2r = v 3 p + a uv2 p2t + b uv pt + c, and it is easy to see
that this implies p|v which is a contradiction.
Therefore t 1 and then
m2 1
2 2r
= 3 3t (u3 + au2 pt v + bup2t v 2 + cv 3 p3t ).
n p v p
Since p 6 |u v, we easily get p 6 |u3 + au2 pt v + bup2t v 2 + cv 3 p3t . From this

2
it follows vp (y 2 ) = vp ( nm 1 3 2 t 2t 2
2 p2r ) = 2r = vp ( v 3 p3t (u + au p v + bup v +
cv 3 p3t )) = vp ( v31p3t ) = 3t, hence 2r = 3t which implies that r = 3 and

t = 2 for some positive integer .
Notice that we have not used that P is a torsion point. We only have
used r 1. We have proved that if P (x, y) C(Q), p is a prime number
such that vp (y) < 0, then vp (y) = 3 and vp (x) = 2 for some N.
This proves immediately that there exist m, n, e, d Z, e, d 6= 0, gcd(m, e) =
gcd(n, ed) = gcd(e, d) = 1 such that y = em3 , x = en2 d . Do not confuse m, n
with the ones used earlier. Let z = nd . Multiplying y 2 = f (x) by e6 yields
z 3 + (ae2 )z 2 + (be4 )z + (ce6 m2 ) = 0. Therefore z is an integer being
a rational that verifies a polynomial equation with integer coefficients and
dominant coefficient 1. Since gcd(n, d) = 1, we can assume d = 1. Dont
forget this argument because it will be used later:
Proposition 2.2.5. Let C(Q) be the smooth rational elliptic curve given by
y 2 = f (x) with f monic in Z[X]. Then for all (x, y) C(Q), there exist
m, n, e Z, e > 0, gcd(m, e) = gcd(n, e) = 1 such that y = em3 and x = en2 .
For arbitrary positive integers , denote
C(p ) = {P (x, y) C(Q)|vp (x) 2, vp (y) 3} {O}.
We want to prove that C(p ) is a subgroup of C(Q). It happens that

it is easier to prove that C 0 (p ) = {P (t, s) C 0 (Q)|vp (t) , vp (s)
3} {O0 } = (C(p )) is a subgroup of C 0 (Q).
First I will prove that C 0 (p ) = (C(p )). Let P (x, y) C(). By what
we have seen, we have vp (x) = 2(+i) and vp (y) = 3(+i) for some i 0
(C(p ) contains no elements with y = 0 because vp (y) 3 < ). Then
(P ) = ( xy , y1 ) = (t, s) and vp (t) = vp (x) vp (y) = + i, vp (t) = vp (y) =
3(+i). Since (O) = O0 we have proved that (C(p )) C 0 (p ). By using
the machinery above with C and C 0 interchanged, we prove 1 (C 0 (p ))
C(p ). So C 0 (p ) = (C(p )).
We now prove that G = C 0 (p ) is a subgroup of C 0 (Q). Let P (t, s)
G. Then (P )(t, s) G so G is stable under taking inverses. Let
P1 (t1 , s1 ), P2 (t2 , s2 ) be two points on C 0 (p ) and let s = t + be the
equation of the line P1 P2 . Assume t1 6= t2 . Then the slope of P1 P2 is
s1
= st22 t 1
and = s1 t1 . We have si = t3i + at2i si + bti s2i + cs3i for
i = 1, 2. By subtracting the two we get s2 s1 = (t32 t31 ) + as2 (t22 t21 ) +
a(s2 s1 )t21 + bt2 (s22 s21 ) + bs21 (t2 t1 ) + c(s32 s31 ). This is the same as
(s2 s1 )(1 at21 bt2 (s1 + s2 ) c(s22 + s1 s2 + s21 )) = (t2 t1 )(t22 + t1 t2 +
t21 + as2 (t2 + t1 ) + bs21 ) .[1] Therefore
t22 + t1 t2 + t21 + as2 (t2 + t1 ) + bs21

= .
1 at21 bt2 (s1 + s2 ) c(s22 + s1 s2 + s21 )
Since vp (ti ) and vp (si ) 3 for i = 1, 2 we easily get that the denomina-
tor of is prime to p, hence vp () = vp (t22 +t1 t2 +t21 +as2 (t2 +t1 )+bs21 ) 2.
Remark 2.2.6. If t1 = t2 = t, then s1 = s2 = s hence P1 = P2 and the

same formula for , the slope of the tangent at P1 to C 0 , holds (just replace
ti and si by t and s respectively).
We first prove t1 = t2 s1 = s2 . From [1] we have (s2 s1 )(1 at21
bt2 (s1 +s2 )c(s22 +s1 s2 +s21 )) = 0. We have seen that vp (1at21 bt2 (s1 +s2 )
c(s22 +s1 s2 +s21 )) = 0 and this implies 1at21 bt2 (s1 +s2 )c(s22 +s1 s2 +s21 ) 6= 0
because vp (0) = . Therefore s1 = s2 . To find , differentiate s =
t3 + at2 s + bts2 + cs3 with respect to t. We get
ds ds ds ds
= 3t2 + 2ats + at2 + bs2 + 2bts + 3cs2
dt dt dt dt
ds 3t2 + 2ats + bs2
= = .
dt 1 at2 2bts 3cs2
Just as in the case t1 6= t2 , vp () 2.
= s1 t1 vp () min{vp (s1 ), vp ( t1 )} 3.
The intersection of P1 P2 and C 0 is given by
s = t3 + at2 s + bts2 + cs3

t+ = t3 +at2 (t+)+bt(t+)2 +c(t+)3 .
s=t+
Let (P1 P2 )(t3 , s3 ) be the third point of intersection of P1 P2 and C 0 . Then

2
we must have t1 + t2 + t3 = +2b+3c
1+a+b2 +c3
. Notice that vp () 2
vp (1 + a + b + c ) = 0 1 + a + b + c3 6= 0.
2 3 2
We have vp (t1 + t2 + t3 ) = vp () + vp (1 + 2b + 3c) 5 + vp (1 + 2b +

3c) 5. Because b, c are integers and vp () 2, vp (1 + 2b + 3c)
min{1, vp (2b), vp (3c) + vp ()} min{1, 0, 0 + 2} = 0. We have P1 + P2 =
P1 P2 , so vp (t1 + t2 + t3 ) 5 vp (t1 + t2 t(P1 + P2 )) 5. Assume
vp (t(P1 +P2 )) < . Then since vp (t1 +t2 ) we get vp (t1 +t2 t(P1 +P2 )) =
vp (t(P1 +P2 )) < which contradicts vp (t1 +t2 t(P1 +P2 )) = vp (t(P1 +P2 ))
5. Hence vp (t(P1 + P2 )) . Notice that p5 |t1 + t2 t(P1 + P2 ) (we
will use this later). s(P1 + P2 ) = t(P1 + P2 ) + vp (s(P1 + P2 )) =
vp (t(P1 + P2 ) + ) min{vp (t(P1 + P2 )), vp ()} 3. This proves that
C 0 (p ) is a subgroup of C 0 (Q). From this it follows easily that C(p ) is a
subgroup of C(Q). maps C(p ) isomorphically onto C 0 (p ).
I prove that the only torsion point of C(p) is O. It suffices to prove
the statement for C 0 (P ) (with O replaced by O0 ). Assume there exists
P (t, s) C 0 (p) such that o(P ) = m > 0. Let vp (t) = 1. Then
vp (s) = 3 (this follows from C 0 (p) = (C(p)) and from 2vp (x) = 3vp (y) for
every (x, y) C(p)).
Assume p 6 |m. We have proved p5 |t1 + t2 t(P1 + P2 ). Using this, a
simple induction proves that t(nP ) nt(P )(mod p5 ) for every every pos-
itive integer n. We have mP = O0 = (0, 0) t(mP ) = 0 p5 |mt(P )
p5 |t(P ). This contradicts = vp (t(P )).
Assume p|m. The abelian group generated by P is isomorphic to Z/mZ
hence there exists an element Q of order p in this subgroup, for example
m
p P . Replacing P by Q we reduce to m = p. Just as before we have
p5 |t(pQ) pt(Q). Since vp (p t(Q)) = + 1 we easily get vp (t(pQ)) =
+ 1. But pQ = O0 t(pQ) = 0 vp (t(pQ)) = and we again have a
contradiction.
In conclusion C(p) and C 0 (p) each have exactly one torsion point. Hence
if P (x, y) M and y 6= 0, then vp (y) 0 for every prime number p. But
this implies that y is an integer and then we know how to prove that x is
an integer.
Let P (x, y) be a torsion point, P 6= O. Then 2P is also a torsion
0 2
point, hence y(2P ) and x(2P ) are integers. x(2P ) = f 2y (x)
a 2x
4y 2 |f 0 (x)2 2y|f 0 (x). But also y 2 = f (x) y|f (x). By remark 2.2.3,
there exist polynomials g, h Z[X] such that f = gf + hf 0 f =
.
g(x)f (x) + h(x)f 0 (x)..y y|f .This completes the proof of the Nagell-Lutz
Theorem.
Theorem 2.2.7 (Nagell-Lutz (strong version)). In the conditions of

the Nagell-Lutz Theorem, y = 0 or y 2 |f .
Proof: Let P (x, y) be a torsion point with o(P ) > 2. We have proved
0 (x)2 f 0 (x)2 4f (x)(a+2x)
that x(P ), y(P ) and x(2P ) = f 4y 2 a2x = 4f (x) are integers.
0 2
Let (x) = f (x) 4(a 2x)f (x). Now x(2P ) Z implies 4f |. To prove
the strong version of Nagell-Lutzs Theorem, we use a stronger version of
Remark 2.2.3, but unfortunately we do not prove it.
Remark 2.2.8. In the context above, there exist g, h Z[X] such that
g + hf = f .
We have y 2 = f (x) y 2 |f (x). Also y 2 = f |4f |. Therefore y 2 |g(x)(x)+

h(x)f (x) = f y 2 |f .
In connection to the problem of studying the torsion points of C(Q) we
have the following result due to Mazur:
Theorem 2.2.9 (Mazur). Let C : y 2 = x3 + ax2 + bx + c be a nonsingular

cubic with a, b, c Q. Let m be the number of torsion points of C(Q). Then
m {1, 2, 3 . . . 10, 12}. Notice that 11 is not a possible value for m.
Example 2.2.10 (Euler). Let C(Q) : y 2 = x3 + 1. Find the torsion points

M of C(Q).
Solution: We obviously have the torsion point O = [0 : 1 : 0]. Let

P (x, y) be a torsion point different from O. Then, by the Nagell-Lutz
Theorem (strong version), we have x, y Z and y = 0 of y 2 |f , where
f (x) = x3 + 1 and f = 4a3 c + a2 b2 + 18abc 4b3 27c2 = 27 for
(a, b, c) = (0, 0, 1). If y = 0, we get the solutions (x, y) = (1, 0). If
y 6= 0, then y 2 | 27 y|3. This gives the solutions (x, y) = (0, 1) and
(x, y) = (2, 3). It is easy to compute 2P = P , where P = (0, 1). Also we
have a point of order 2, namely (1, 0). From this we easily get M ' Z/6Z.
.
Exercise 2.2.11. For an arbitrary prime number p, find the torsion points
of C(Q) : y 2 = x3 + px.
Exercise 2.2.12. Find the torsion points of C(Q) : y 2 + 7xy = x3 + 16x.

Chapter 3
Lecture III
3.1 Torsion points on rational elliptic curves and

elliptic curves mod p
Let C(Q) be the elliptic curve given by y 2 = x3 +ax2 +bx+c. Let p be a prime
number such that p 6 |2f , where f = 4a3 c + a2 b2 + 18abc 4b3 27c2 .
It makes sense to consider the curve over Zp (the prime field of char-
acteristic p), C(Zp ) : y 2 = x3 + ax2 + bx + c = f(x), where a, b, c are the
residue classes of a, b and c modulo p respectively.
Proposition 3.1.1. Let p be an algebraic closure for Zp . Then f has

distinct roots in p if and only if p 6 | 2f .
Remark 3.1.2. Just like working over C we can define what it means for
a curve over p to be nonsingular.
Proposition 3.1.3. In view of the remark above, C(p ) given by y 2 =

x3 + ax2 + bx + c is nonsingular.
Proof: Imitating the complex situation we get that C(p ) is nonsingular

if and only if 0 6= f = f, where f is the mod p reduced of f . Since p 6 |f ,
the conclusion follows.
Proposition 3.1.4. C(Zp ) has a natural group structure with neutral ele-
ment O = [0 : 1 : 0] in C(Zp ) P2Zp . This group structure is induced by the
group structure of C.
Proof: This can be done similarly to 1.2.12 and 1.2.13. Another way
to prove it is to reduce mod p the algebraic formulas in 1.3.1 and 1.3.2 for
P + Q and 2P respectively. If p divides the denominators of the fractions
in these equations, just set P + Q or 2P as [0 : 1 : 0]. Notice that because
p 6 |2f , p 6= 2, 2P is not always the point at infinity of C(Zp ).
31
32 CHAPTER 3. LECTURE III
Remark 3.1.5. In order to ease the notations, we will no longer distinguish

an elliptic curve from its projective closure. The immediate effect is that the
uncomfortable bar will disappear.
Theorem 3.1.6. Let p a prime number as above and let A = {(x, y)

C(Q)|vp (x) 0, vp (y) 0}. There is a natural application A Z2p :
(x, y) (x, y). Define : C(Q) C(Zp ) by:

(x, y), if (x, y) A
(x, y) = .
O, elsewhere
Then is a group homomorphism and |M is injective. Recall that M

is the set of torsion points of C(Q).
The application A Z2p : (x, y) (x, y) is componentwise given by

Q 3 x = ab ab1 Zp , where a, b are coprime integers, a and b are the
mod p residue classes of a and b respectively and b1 is the inverse of b in
Zp . This inverse exists because vp (x) 0 p 6 |b b 6= 0.
Example 3.1.7. Let C(Q) be the elliptic curve given by y 2 = x3 + 3. Find

M.
Proof: f = 35 = 243.
5 6 |235 so M injects in C(Z5 ). By Lagranges Theorem, |M| | |C(Z5 )|.
Because 3 6 |5 1, the function Z5 Z5 : x x3 is bijective. This means
that over Z25 , the equation y 2 = x3 + 3 has 5 solutions (for each y = 0, 4, x
is uniquely defined). The group C(Z5 ) also contains the point at infinity O,
so |C(Z5 )| = 6.
7 6 | 2 35 so M also injects in C(Z7 ) and |M| | |C(Z7 )|. The solutions
over Z7 of y 2 = x3 + 3 are
(1, 2), (1, 5), (2, 2), (2, 5), (3, 3), (3, 4), (4, 2), (4, 5), (5, 3), (5, 4), (6, 3), (6, 4).
Adding the point at infinity O we get |C(Z7 )| = 13.

Therefore |M||gcd(6, 13) = 1 |M| = 1. Since O M, we have proven
that M = {O}.
Exercise 3.1.8. Let (tn )n1 be the sequence of rational numbers defined by
t1 = t2 = t3 = t4 = t5 = 1 and
tn+4 tn+1 + tn+3 tn+2

tn+5 =
tn
for every n 1.
Prove that all the terms of (tn )n1 are in fact integers.
3.1. TORSION POINTS ON RATIONAL ELLIPTIC CURVES 33
Theorem 3.1.9 (Gauss). Let p be a prime number different from 2 and 3

and C(Zp ) be the cubic defined by x3 + y 3 = 1.
Then
p + 1, if p 2(mod3)
|C(Zp )| = ,
p + 1 + A, if p 1(mod3)
where A is uniquely determined by 4p = A2 + 27 B 2 , A, B Z and A
1(mod3).
Theorem 3.1.10 (Hasse-Weil). : Let p be a prime number and let C be

a nonsingular irreducible curve of genus g defined over Zp . Then |C(Zp )| =

p + 1 + with || 2g p.
3.2 Test Paper

1
Exercise 3.2.1. Let C(Q) be the elliptic curve defined by y 2 = x3 4x.

Determine M (up to isomorphism).
Exercise 3.2.2. Let C(Q) : y 2 = x3 + bx, with b Z and p4 6 |b for any

prime number p. Determine M.
Hint: For suitable prime numbers p such that p 3(mod4), prove that
C(Zp ) = p + 1. Use this to prove that |M||4.
Exercise 3.2.3. Let C(Q) : y 2 = x3 + c with c Z and p6 6 |c for any prime

number p. Determine M.
Hint: Prove first that |M||6.
The next page contains the solutions to these problems. If you want to
try to solve them yourself, do not turn the page.
1
Working time 2 hours. This test only counts as extra for the final grade
3.2. TEST PAPER 35
3.2.1 Solutions
Solution to 3.2.1 For y = 0 we get the solutions (2, 0), (0, 0), (2, 0) which
are all elements of order 2 in M. Together with O, M contains at least 4
elements.
f = 44 . 3 6 |2f . The solutions mod 3 of y 2 = x3 4x = x3 x = 0 are
(0, 0), (1, 0), (2, 0). Together with O we find that C(Z3 ) has 4 elements. Like
we did before, we find |M||4. Since M has at least 4 elements, |M| = 4.
M is an 2-torsion abelian group with 4 elements (i.e. 2x = O x M),
therefore M ' Z2 Z2 .
Solution to 3.2.2 f = 4b3 . Let p be a prime number, p 3(mod4)

and p 6 |b. Then p 6 |2f and we can apply 3.1.6.
Since the equation x2 = 1 has no solutions in Zp , using the Legendre
symbol, y 2 = n has solutions in Zp if and only if y 2 = n has no solutions
mod p.
Notice that x3 + bx is an odd function. If x3 + bx 6= 0, exactly one of
x + bx and x3 bx is a square and gives two solutions of the form (x, y)
3
for y 2 = x3 + bx. If x = 0 then we get the only solution (0, 0).

If b is a square mod p, then x2 = b has two distinct solutions u giving
the solutions (u, 0) for y 2 = x3 + bx. We have p 3 solutions for y 2 =
x3 + bx corresponding to the pairs (x, x) with x 6 {u, 0, u}, 2 solutions
for x = u and 1 solution for x = 0. Hence a total of p solutions.
If b is not a square then we have p 1 solutions for pairs (x, x) with
x 6= 0 and 1 solution for x = 0. Again a total of p solutions.
Adding O, |C(Zp )| = p + 1 and by Lagrange, |M||p + 1.
The following is meant to prove that if d = gcd{p + 1|p is prime, p
3(mod4), p 6 |b}, then d = 4. Let P = {p + 1|p is prime, p 3(mod4), p 6
|b}. Assume there exists a prime number dividing d, q 6= 2. The Chinese
Remainder Theorem guarantees the existence of an integer n > b such that
n 3(mod4) and n 1(modq). By Dirichlets Theorem concerning primes
in arithmetic progressions, there exists a prime number p0 in the arithmetic
progression of initial term n and ratio 4q. Since n > b, it is obvious that
p0 6 |n. It is easy to see that p0 is in the set P . But q|d and q 6 |p0 + 1
because q 6= 2 and p0 + 1 n + 1(mod 4q) p0 + 1 1 + 1 6 0(modq).
This is a contradiction. Therefore d is a power of 2. 4|p + 1 p P , so
4|d. Assume for a contradiction that 8|d. Again by Dirichlet, there exists a
prime number q 0 > b with q 3(mod8). Then q 0 P and 8 6 |q 0 + 1. This
again is a contradiction. We have proved that d = 4, hence |M||4.
In M we always have the point (0, 0) of order 2 and the neutral element,
the point at infinity O. Thus M has at least 2 elements.
If M ' Z2 Z2 then M has 3 elements of order 2. The points on C(Q)
of order 2 are the points that have the y coordinate 0. If there are at least
three such points then the polynomial X 3 + bX has three distinct roots in
Q. These roots must be integers and we easily get that b must be a square
different from 0. From the condition p4 6 |b for any prime p, b must be the
square of a square free integer.
If M is cyclic of order 4, let P (x, y) be a generator of this group. Notice
that y 6= 0 because otherwise P would be of order 2. This implies x 6= 0.
Because Z4 has only one element of order 2, namely 2, the only element of
M of order 2 is (0, 0). Therefore 2P = (0, 0) P P = (0, 0). We get the
equations:
2 2
x(P P ) = 3x2y+b 2x = 0

y(P P ) = 3x2 +b x(P P ) + (y 3x2 +b x) = 0
2y 2y

3x2 +b 2 2x = 0

2y
3x2 +b
y 2y x =0
( 2
3x2 +b
2x = 0
2y .
2y 2 = 3x3 + bx
Because P C(Q), we have y 2 = x3 + bx. Together with 2y 2 = 3x3 + bx
this gives x3 = bx. Since x 6= 0, x2 = b y 2 = x3 + bx = 2bx.
So we have arrived to solving
x2 = b

y 2 = 2x3 .
y 2 = 2bx
From the condition p4 6 |b for any prime p, x must be a square free integer.
It is easy to see that y 2 = 2x3 x = 2u2 , y = 4u3 for some u Z. Since x
must be square free, u = 1 b = x2 = 4. Hence M is cyclic of order 4 if
and only if b = 4. A generator for M is then P (2, 4).
A Sherlock Holmes type argument gives that M ' Z2 if and only if b
is not a square and b 6= 4. These in the conditions of the problem of course.
Solution to 3.2.3 f = 27c2 . Let p be a prime number, p 2(mod3)

and p 6 |c. Then p 6 |2f and we can again apply 3.1.6.
Because 3 6 |p1, the application Zp Zp : x x3 is bijective, therefore
for each y Zp , there exists a unique x Zp such that x3 = y 2 c. Together
with O, |C(Zp )| = p + 1.
There is a similar argument to the one in the previous problem which
proves that |M||6. The choices for M are: the trivial group, Z2 , Z3 and Z6 .
The choices for M prove that it has at most one element of order 2. If
this element exists, it is a point on the x-axis. y = 0 x3 = c, hence c is
a cube. The condition p6 6 |c for any prime p, proves that c is the cube of a
square free integer.
3.2. TEST PAPER 37
Assume P (x, y) is an element of order 3 in M. Then y 6= 0 otherwise P

would be of order 2. Then 3P = 0 2P = P P P = P . This gives
the system of equations:
2 2
3x

2y 2x = x
2
2
2 .
3x ( 3x 3x2
2y 2y 2x) + (y 2y x) = y
Notice that the second equation is a consequence of the first, hence the
2 2
system is equivalent to 3x
2y 2x = x x (3x3 4y 2 ) = 0.
If x = 0, then y 2 = c and so c is a square.
If 3x3 = 4y 2 , then x = 12u2 and y = 36u3 for some u Z. The condition
c = y 2 x3 implies u6 |C u = 1 c = 432.
We are now able to characterize M in the conditions of the problem:

Z6 , if c = 1
Z3 , if (c is a square and c 6= 1) or c = 432

M'

Z2 , if c is a cube and c 6= 1
trivial = O, otherwise

Chapter 4
A Theorem of Gauss
Theorem 4.0.4 (Gauss). Let p be a prime number different from 2 and 3

and let C(Zp ) be the cubic defined by x3 + y 3 = 1.
Then
p + 1, if p 2(mod3)
|C(Zp )| = ,
p + 1 + A, if p 1(mod3)
where A is uniquely determined by 4p = A2 + 27 B 2 , A, B Z and A

1(mod3).
Proof: We quickly rid the case p 2(mod3). In this case, the appli-
cation Zp Zp : x x3 is bijective, hence for each y Zp , there is a
unique x such that x3 + y 3 = 1. Together with the unique point at infinity,
|C(Zp )| = p + 1.
Assume p 1(mod3). Recall that we have identified C(Zp ) to its pro-
jective closure C(Zp ). If C(Zp ) is given by x3 + y 3 = 1, then C(Zp ) is given
via homogenization by x3 + y 3 = z 3 . A substitution z z proves that
|C(Zp )| is the number of solutions in P2Zp of x3 + y 3 + z 3 = 0.
(Zp , ) is a cyclic group generated by an element u. let R = {a3 |a Zp }.
Notice that R = {u3i |i Z} and that it is a subgroup of Zp . Let S = uR and
T = u2 R = uS. Because p 1(mod3), R, S and T are disjoint and actually
form a partition of Zp . To see this, let m = p1 m
3 . In Zp , 1, u and u
2m are
distinct elements of order 3. Over Zp , the polynomial equation x3 = 1 has at

most 3 roots, hence {1, um , u2m } is the set of elements of order 3 in Zp . This
set is a subgroup of Zp and it is also the kernel of Zp Zp : x x3 . The
image of this application is R. This proves that R is a subgroup of index
3 in Zp and that S, T are its cosets. In particular R, S and T partition
Zp and |R| = |S| = |T | = m. Moreover, RS = S, ST = R, T R = T ,
rR = R, rS = S, rT = T r R, sR = S, sS = T, sT = R s S and
tR = T, tS = R, tT = S t T .
For arbitrary nonempty subsets X, Y and Z of Zp , denote |XY Z| =
|{(x, y, z) X Y Z|x + y + z = 0}|.
39
40 CHAPTER 4. A THEOREM OF GAUSS
We can now start counting the projective solutions of x3 + y 3 + z 3 = 0. If

xyz 6= 0, then the equation has 27|RRR| solutions in (Zp )3 . This is because
if a R, then the equation x3 = a has 3 distinct solutions in Zp . Therefore
27
there are p1 |RRR| projective solutions of x3 + y 3 + z 3 = 0 with xyz 6= 0.
If x = 0, then y 6= 0 otherwise z = 0 and [0 : 0 : 0] is not a point in P2Zp . For
each y Zp , the solutions to 03 + y 3 + z 3 = 0 are z {y, um y, u2m y},
hence for x = 0, x3 + y 3 + z 3 = 0 has 3(p 1) solutions in (Zp )3 different
from (0, 0, 0). These give 3 projective solutions if x = 0. These solutions are
actually {[0 : 1 : 1], [0 : 1, um ], [0 : 1 : u2m ]}. The same argument holds
for y = 0 and z = 0 and the corresponding sets of solutions are easily seen
27 9
to be disjoint. The conclusion is |C(Zp )| = p1 |RRR| + 9 = m |RRR| + 9.
So computing |C(Zp )| reduces to computing |RRR|.
Remark 4.0.5. It is easily seen that if X, Y, Z1 , Z2 are nonempty subsets
of Zp and Z1 Z2 = , then |XY (Z1 Z2 )| = |XY Z1 | + |XY Z2 |.
We have the partition Zp = {0} t R t S t T . Using the above remark,
|RRZp | = |RR{0}|+|RRR|+|RRS|+|RRT |. For each a, b R, the equation
a + b + x = 0 has a unique solution in Zp , so |RRZp | = |R|2 = m2 . For every
a R, a R is the unique solution to a + x + 0 = 0, hence |RR{0}| = m.
So m2 = m + |RRR| + |RRS| + |RRT |.
Remark 4.0.6. If X, Y, Z are nonempty subsets of Zp and a Zp , then
|(aX)(aY )(aZ)| = |XY Z|. This is because x + y + z = 0, (x, y, z) X
Y Z ax + ay + az = 0, (ax, ay, az) aX aY aZ.
Using this observation, |RRR| = |(uR)(uR)(uR)| = |SSS| = |(uS)(uS)(uS)| =
|T T T |. Similarly |RRS| = |SST | = |T T R| and |RRT | = |SSR| = |T T S|.
Remark 4.0.7. If X, Y, Z are nonempty subsets of Zp , then |XY Z| =
|Y XZ| = |XZY |.
m2 = m + |RRR| + |RRS| + |RRT | = m + |RRR| + |SST | + |T T S| =
m + |RRR| + |SST | + T ST |.
|{0}ST | + |RST | + |SST | + |T ST | = |Zp ST | = m2 . If there exist s S
and t T such that 0 + s + t = 0, then s = t. (1) = (1)3 (1)
R T = T s S T which is a contradiction. Hence |{0}ST | = 0.
m + |RRR| + |SST | + |T ST | = m2 = |RST | + |SST | + |T ST | m +
|RRR| = |RST |
9 9
|C(Zp )| = (|RST | m) + 9 = |RST |.
m m
Let be a complex primitive p-th root of unity. Because p = 1, k is
well defined for every k Zp . Let
X
1 = r
rR
41
X
2 = s
sS
X
3 = t .
tT
1 )(X P
Let F (X) = (X P 2 )(X 3 ). We set to find its coefficients.
i p1 i
1 + 2 + 3 = iZp = ( i=0 ) 1 = 1.
X p1
X
s+t
2 3 = = N (x)x ,
sS x=0
tT
where N (x) = |{(s, t) S T |s + t = x}| = |ST {x}|. We have N (0) =

|ST {0}| = 0. Also N (x) = |ST {x}| = |(rS)(rT ){rx}| S = |ST {rx}| =
N
P (rx) for every r R. We have |ST (xR)| = |ST ( rR {rx})| =
rR N (rx) = |R| N (x) = mN (x). Therefore,

|ST R|, if x R
mN (x) = |ST S|, if x S .
|ST T |, if x T

Let a, b, c N such that |ST R| = ma, |ST S| = mb, |ST T | = mc. Then
|C(Zp )| = 9a and 2 3 = a1 + b2 + c3 .
Similarly, 1 2 = p1 x
P
x=1 N1 (x) , where N1 (x) = |RS{x}|. Also,

|RSR|, if x R
mN1 (x) = |RSS|, if x S ,
|RST |, if x T

and 1 2 = |RSR| |RSS| |RST |

m 1 + m 2 + m 3 = b1 + c2 + a3 . Analogously
we prove 1 3 = c1 + a2 + b3 .
1 2 + 2 3 + 3 1 = (a + b + c)(1 + 2 + 3 ) = a b c =
1 1 1
m (|ST R|+|ST S|+|ST T |) = m (|ST Zp ||ST {0}|) = m (m2 0) = m.
31 2 3 = 1 (2 3 ) + 2 (1 3 ) + 3 (1 2 ) = 1 (a1 + b2 + c3 ) +
2 (c1 + a2 + b3 ) + 3 (b1 + c2 + a3 ) = a(12 + 22 + 32 ) + (b + c)(1 2 +
2 3 + 3 1 ) = a((1 + 2 + 3 )2 2(1 2 + 2 3 + 3 1 )) m(b + c) =
m(2a b c) + a = mk + a, for k = 2a b c. So 1 2 3 = km+a 3 .
We now have F (X) = X 3 + X 2 mX km+a 3 .
k = 2a b c = 3a (a + b + c) = 3a + m |C(Zp )| = 9a = 3(3a) =
3(m + k) = 3m + 3k = p 1 + 3k = p + 1 + (3k 2). Denote 3k 2 = A. We
want to prove that A verifies the conditions of the theorem. Because A Z
and A 1(mod3), all that is left to prove is that there exists B Z such
that 4p = A2 + 27B 2 and that A is unique with this properties.
What follows is an ingenious method to construct B.
Let i = 1 + 3i i = 1, 3. Consider the polynomial G(X) = (X
1 )(X 2 )(X 3 ) and lets compute its coefficients.
42 CHAPTER 4. A THEOREM OF GAUSS
1 + 2 + 3 = 3 + 3(1 + 2 + 3 ) = 0.
1 2 + 2 3 + 3 1 = 3 + 6(1 + 2 + 3 ) + 9(1 2 + 2 3 + 3 1 ) =
3 6 9m = 3(3m + 1) = 3p.
1 2 3 = 1 + 3(1 + 2 + 3 ) + 9(1 2 + 2 3 + 3 1 ) + 271 2 3 =
139m+9(km+a) = 29m+9km+3(3a) = 29m+9km+3(m+k) =
(3k 2) + (9km 6m) = (3k 2)(1 + 3m) = p A.
We have proved G(X) = X 3 3pX pA. Lets compute G , the dis-
criminant of G. Recall the following formulas for the discriminant of a
polynomial:
Proposition 4.0.8. Let f (X) = X 3 + aX 2 + bX + c be a polynomial with

coefficients in an arbitrary field K and let x1 , x2 , x3 be the roots of f in an
algebraic closure of K. Then f = 4a3 c + a2 b2 + 18abc 4b3 27c2 =
((x1 x2 )(x2 x3 )(x3 x1 ))2 and f K.
G = ((1 2 )(1 3 )(2 3 ))2 = 272 ((1 2 )(1 3 )(2 3 ))2 .

(1 2 )(1 3 )(2 3 ) = (12 1 3 1 2 + 2 3 )(2 3 ) =
2 3 (2 3 ) + 1 (1 2 3 )(2 3 ) = 2 3 (2 3 ) +(12 2 12 3
1 22 +1 2 3 1 2 3 +1 32 ) = 2 3 (2 3 )+3 1 (3 1 )+1 2 (1
2 ) = (a1 + b2 + c3 )(2 3 ) + (c1 + a2 + b3 )(2 1 ) + (b1 +
c2 + a3 )(1 2 ) = (12 + 22 + 32 )(b c) + (1 2 + 2 3 + 3 1 )(c b) =
(b c)(12 + 22 + 32 1 2 2 3 3 1 ) = (b c)(1 3(1 2 + 2 3 +
3 1 )) = (b c)(1 + 3m) = (b c)p G = 272 (b c)2 p2 .
On the other hand, G = 4(3p)3 27(pA)2 = 108 p3 27p2 A2 .
By comparing the two formulas we get 108p3 27p2 A2 = 272 (b c)2 p2
4p A2 = 27(b c)2 4p = A2 + 27B 2 , where B = b c.
We now prove the uniqueness of A.
Assume there exists A1 Z with A1 1(mod3) for which there exists
B1 Z such that 4p = A2 + 27B 2 = A21 + 27B12 .
4p(B12 B 2 ) = (A2 + 27B 2 2 2 2 2 2 2
)B1 B (A1 + 27B1 ) = A B1 A1 B =
2 2
p|AB1 + A1 B
(AB1 BA1 )(AB1 +BA1 ) or . Assume that p|AB1 BA1 .
p|AB1 BA1

The other case is treated in perfect analogy.
(4p)2 = 16p2 = (A21 + 27B12 )(A2 + 27B 2 ) = (AA1 + 27BB1 )2 + 27(AB1
2 2
AB1 BA1
BA1 )2 p|AA1 + 27BB1 16 = AA1 +27BB p
1
+ 27 p . If
2 2
AB1 6= BA1 , then 27 AB1 BA p
1
27. This and AA1 +27BB p
1
0
imply 16 27 which is absurd. So AB1 = BA1 A2 B12 = B 2 A21
(4p 27B 2 )B12 = B 2 (4p 27B12 ) 4pB12 27(BB1 )2 = 4pB 2 27(BB1 )2
4pB12 = 4pB 2 B12 = B 2 4p 27B12 = 4p 27B 2 A21 = A2 A1 =
A. Since A A1 1(mod3), we finally have the proof for A = A1 .
Chapter 5
Mordell-Weils Theorem
In this lecture we begin the proof of the Mordell-Weil Theorem. The theorem
states that the abelian group C(Q) associated to an elliptic curve is finitely
generated. Restricted by an elementary proof we will first prove the theorem
in the particular case it has points of order 2. Later, in Lecture VIII, we
will give a complete proof using some facts of Algebraic Number Theory.
Theorem 5.0.9 (Mordell-Weil). Let C(Q) be the elliptic curve defined

by y 2 = x3 + ax2 + bx + c = f (x) with a, b, c Z and f 6= 0. The abelian
group C(Q) is finitely generated.
We begin by introducing some machinery. We first define the height of a

rational point. The general notion of height associated to rational points on
projective varieties or to sheaves is fundamental to Arithmetic Geometry one
of the leading domains in Arithmetics. The definition of height for rational
numbers or points is nevertheless very easy to understand.
Definition 5.0.10. Let q Q. Define H(q) = max{|m|, |n|} if q = m n,

m, n Z, n 6= 0 and gcd(n, m) = 1. Define h(q) = ln H(q). H(q) and h(q)
are both called heights of the rational number q.
Remark 5.0.11. The condition gcd(m, n) = 1 in the definition of H(q)

implies that H(0) = 1 since the only possible writings for 0 in the form m n
0
with gcd(n, m) = 1 are 1 . Since |n| = | n| for any integer n, the height H
is well defined for any rational number. We have h(q) = ln H(q) ln 1 = 0,
therefore h takes only nonnegative real values.
Definition 5.0.12. If P (x, y) is a point in C(Q) different form O = [0 : 1 :

0], the point at infinity of C(Q), then define H(P ) = H(x) and h(P ) = h(x).
Set H(O) = 1 and h(O) = 0.
The main steps of the proof of Mordell-Weils Theorem are given by the
following four lemmas. The first three are elementary and will also be used
in the proof of the general Mordell-Weil Theorem in Lecture VIII. To prove
43
44 CHAPTER 5. MORDELL-WEILS THEOREM
the fourth in elementary manner we will need the additional assumption

that C(Q) has points of order 2.
Lemma 5.0.13 (Lemma 1). If r R, then {P C(Q)|h(P ) < r} is a
finite set.
Proof: h(P (x, y)) = h(x) = ln H(x) < r H(x) < er . If x = m n
with m, n Z, n 6= 0 and gcd(n, m) = 1 then H(x) = max{|m|, |n|}.
H(x) < er |m|, |n| < er . Since m, n range in (er , er ) there are only
a finite number of possibilities to choose them, hence a finite number of
choices for x. Since y is given by y 2 = f (x), there are only a finite number
of choices for P on C(Q).
Remark 5.0.14. Recall that we have proved in 2.2.5 that for any P (x, y)
C(Q) there exist m, n, e Z with e 6= 0 and gcd(m, e) = gcd(n, e) = 1 such
that x = em2 and y = en3 .
Lemma 5.0.15 (Lemma 2). For any P0 C(Q) there exists a constant
c1 depending only on P0 and C such that h(P + P0 ) 2h(P ) + c1 for any
P C(Q).
Proof: If P0 = O then choose c1 = 0.
Assume P0 (x0 , y0 ) 6= O and take P (x, y) C(C), P 6= O, P0 , P0 . We
yy0
have x(P + P0 ) = 2 a x x0 with = xx 0
. Then x(P + P0 ) =
2 2 a(xx )2 (x+x )(xx )2 2 +Cx+D
yy0 (yy )
xx0 axx0 = 0 0
(xx0 )2
0 0
= Ay+Bx
Ex2 +F x+G
, where
A, B, C, D, E, F, G Z depend only on P0 and C. By the preceding remark,
n 2
A +B m4 +C m2 +D Aen+Bm2 +Cme2 +De4
e3
x(P + P0 ) = 2
e e
= Em2 +F me2 +Ge4
. The denominator
E m4 +F m2 +G
e e
and numerator of the last fraction need not be coprime, but we certainly
have H(P + P0 ) max{|Aen + Bm2 + Cme2 + De4 |, |Em2 + F me2 + Ge4 |}.
H(P ) = max{|m|, |e2 |} e2 H(P ) and |m| H(P ). Dont confuse
the integer e that appears here with Eulers constant e = 2.7 . . ..
3
We prove |n| kH(P ) 2 where k depends only on C(Q). We have
2 3 2
y 2 = x3 + ax2 + bx + c ne6 = m e6
+ am e4
+ b em2 + c n2 = m3 + am2 e2 +
4 6 2 3
bme + ce |n| H(P ) (1 + |a| + |b| + |c|). Denote k = 1 + |a| + |b| + |c|.
3
Therefore |n| kH(P ) 2 and k depends only on C.
Now H(P + P0 ) max{H(P )2 (|A|k + |B| + |C| + |D|), H(P )2 (|E| +
|F | + |G|)} = H(P )2 k1 for some k1 depending only on C(Q) and P0 .
From this it easily follows that h(P + P0 ) 2h(P ) + ln k1 . Choosing
c1 = max{h(P0 ), h(2P0 ), h(O), ln k1 } finishes the proof.
Lemma 5.0.16 (Lemma 3). There exists a constant c2 depending only on
C(Q) such that h(2P ) 4h(P ) c2 for all P C(Q).
Proof: Assume P (x, y) is a point on C(Q) of order different from 1 or 2.
0 2 0 2 (3x2 +2ax+b)2 4(a+2x)(x3 +ax2 +bx+c)
We have x(2P ) = f 2y
(x)
a2x = f4f(x) (x) a2x = 4(x3 +ax2 +bx+c)
=
45
(x)
(x) ,
where and are polynomials in x of degrees 4 and 3 respectively. If
and had common roots, then f and f 0 would have had common roots
contradicting the smoothness of C. The coefficients of and are integers
and only depend on C(Q).
To finish the proof of 5.0.16 we need the following lemma:
Lemma 5.0.17. Let , Z[X] be two polynomials with no common com-
plex roots. Let d = max{deg(), deg()}. Then:
1. There exists a nonzero integer R such that gcd(nd ( m d m
n ), n ( n ))|R
for all m, n Z, n 6= 0 and gcd(m, n) = 1.
( m )
2. There exist c2 , c3 R+ such that dh( m
n )c2 h( ( m
n m
) ) dh( n )+c3
n
for rational numbers m m
n such that ( n ) 6= 0.
Assume the above lemma proved. Applied to our setting, it gives the
existence of a constant c02 such that 4h(P ) c02 h(2P ) for every P (x, y)
C(Q) of order different from 1 or 2. Let c2 = max{4h(P ) h(2P )|P
C(Q) is of order 1 or 2}. Choosing c2 = max{c02 , c2 } we have proved 5.0.16.
Proof of 5.0.17: (1) Using that h(q) = h( 1q ) for any nonzero rational
number q, we can assume d = deg() deg() = e. Because and
have no common complex roots, there exist 1 and 1 in Q[X] such that
1 + 1 = 1. There exists a positive integer A such that A1 , A Z[X]
and A depends only on and . Let D = max{deg(1 ), deg(1 )}. Then
m m m m
nd ( ) nD A1 ( ) + nd ( ) nD A1 ( ) = nd+D A.
n n n n
Let k = gcd(nd ( m d m
n ), n ( n )). It is easy to see that the previous equality
implies k|And+D . Let (x) = a0 xd + a1 xd1 + . . . + ad .
k|nd ( m
n ) k|An
d+D1 (nd ( m )) = And+D1 (a md + a md1 n +
n 0 1
. . . + ad nd ). For i > 0, k|And+D k|And+D1+i mdi ai . From this
we conclude k|Aa0 nd+D1 md . Therefore k|gcd(Aa0 nd+D1 md , Ann+D ) =
And+D1 gcd(a0 md , n) = And+D1 gcd(a0 , n)|Aa0 nd+D1 . Repeating this
argument we prove k|Aad+D 0 . A solution to the problem is R = Aa0d+D .
(2) We are only interested in the first inequality i.e. in the existence of
c2 . Proving the existence of c3 is done in analogy to Lemma 2.
nd ( m )
The previous point proves that H( nd ( mn
)
) R1 max{|nd ( m d m
n )|, |n ( n )|}.
n
This is because what is cancelled to make the the denominator and numer-
nd ( m )
ator of nd ( m
n
)
coprime divides R and therefore is less or equal to R.
n
By the well known inequality max{|a|, |b|} 21 (a + b) for positive reals
nd ( m ) 1 m
a, b, we have H( nd ( m
n
)
) d
2R (|n ( n )| + |nd ( m
n )|).
n
varphi( m )
H( n
) |nd ( m d m
|( m m
( m
n
) n )| + |n ( n )| n )| + |( n )|
= k1 > 0
H( mn)
d 2R max{|m|d , |n|d } 2R max{| m d
n | , 1}
for some k1 > 0, for all integers m, n, n 6= 0, gcd(m, n) = 1 and ( m n ) 6= 0.

|(x)|+|(x)|
To prove the existence of k1 , consider the function f (x) = 2R max{|x|d ,1} . f
is a continuous function on R and f (x) 0 for all x R. We have f (x) =
0 |(x)| + |(x)| = 0 (x) = (x) = 0 contradicting the hypothesis
that and have no common complex roots. We have proved f (x) > 0 for
all reals x. For |x| > 1 we have lim|x| f (x) = lim|x| |(x)|+|(x)|
2R|x|d
. This
limit exists and is equal to the sum of the absolute values of the coefficients
of the terms of degree d in and . By the assumption deg() = d, at
least the one in is nonzero, hence the limit is nonzero and greater than
some real l > 0. There exists N N such that f (x) > 2l for |x| > N .
Since f (x) > 0 for x R, f (x) > l0 on [N, N ] for some l0 > 0. Just take
k1 = min{ 2l , l0 }.
( m )
n )
H( ( m )
We have proved H( m
n
)d
k1 > 0. Taking logarithms and setting
n
( m
c2 = ln k1 we find the required inequality: dh( m
n ) c2 h( ( m ) ).
n
n
Lemma 5.0.18 (Lemma 4 Weak Mordell-Weil Theorem).
|C(Q) : 2C(Q)| < .
For the beginning, assume we have proved Lemma 4 and lets see how the
proof of Mordell-Weils Theorem follows. We will use the following descent
argument:
Theorem 5.0.19 (Descent Argument Theorem). Let (G, +) be an

abelian group and let h : G R+ be a function such that:
1. r R the set {g G|h(g) < r} is finite.
2. g0 G there exists c1 > 0 such that h(g + g0 ) 2h(g) + c1 for all

g G.
3. There exists c2 > 0 such that h(2g) 4h(g) c2 for all g G.
4. |G : 2G| < .
Then G is finitely generated.
Proof of Mordell-Weils Theorem: Plunge G = C(Q), h the height

function on C(Q), c1 and c2 the constants given by Lemmas 2 and 3 in the
Descent Argument Theorem. Notice that conditions (1) and (4) in the the-
orem are guaranteed by Lemmas 1 and 4 and conclude that C(Q) is finitely
generated.
Proof of 5.0.19: Let q1 , . . . , qm be a complete system of representants

in G for G/2G. By condition (4), m < .
47
Let p G. By induction, there exists a sequence of elements pk G

such that p0 = p and there exists ik 1, m such that pk1 qik = 2pk for
all k > 0. Just take qik to be the reprezentant of pk1 in G/2G.
A simple induction proves p = qi1 + 2qi2 + . . . + 2s qis+1 + 2s+1 Ps+1 for
all s > 0.
By (2) there exist positive real constants c1,1 , . . . , c1,m such that h(g
qi ) 2h(g)+c1,i for all g G and all i = 1, m. Take c1 = max{c1,i |i = 1, m}.
c1 > 0. Then h(g qi ) 2h(g) + c1 for all g G (5).
Combining (3) and (5) gives 2h(ps1 ) + c1 h(ps1 qis ) = h(2ps )
4h(ps ) c2 2h(ps ) h(ps1 ) + c1 +c 2
2
h(ps ) 43 h(ps1 ) + 14 (c1 + c2
h(ps1 )). (6)
If h(ps1 ) c1 + c2 for some s, then by (6), h(ps ) 34 h(ps1 ) < h(ps1 ).
The inequality is strict otherwise 0 = h(ps1 ) c1 +c2 > 0 is a contradiction.
If h(pt ) c1 +c2 for all t s, then by induction (pt )ts is an infinite sequence
of elements of G with the property h(ps ) > h(ps+1 ) > . . .. This contradicts
the hypothesis (1). Therefore there exists t s such that h(pt ) < c1 + c2 .
For this t, p = qi1 + 2qi2 + . . . + 2t1 qit + 2t pt . So p is a combination of
q1 , q2 , . . . , qm and {r G|h(r) < c1 + c2 }.
We have proved that q1 , . . . , qm and the finite set {r G|h(r) < c1 + c2 }
(this set is finite by (1)) generate G.
Proof of 5.0.18: As I have said, we will first prove the Weak Mordell-
Weil Theorem only in the case C(Q) has a point of order 2. The equation
of C(Q) was given by y 2 = x3 + ax2 + bx + c = f (x) with a, b, c Z
and f 6= 0. We have seen that the points P (x, y) of order 2 of C(Q)
are given by y = 0 f (x) = 0. x Q and f (x) = 0 imply that x is
a rational number, integral over Z, therefore x is an integer. Assume x0
is an integer root of f . Up to the substitution x x x0 we can as-
sume that x0 = 0 c = 0. The condition for non-singularity of C is
0 6= f = 4a3 c + a2 b2 + 18abc 4b3 27c2 = a2 b2 4b3 = b2 (a2 4b).
Define the curve C given by the equation C : y 2 = x3 + ax2 + bx = f(x),

where a = 2a and b = a2 4b. We have f = b2 (a2 4b) = (a2 4b)2 (4a2
4a2 + 16b) = 16b(a2 4b)2 6= 0, hence C(Q) is a smooth elliptic curve.
Remark 5.0.20. Repeating the construction above for C, we have C(Q) :

y 2 = x3 +ax2 +bx, where a = 2a = 4a and b = a2 4b = 4a2 4(a2 4b) =
16b.
The substitutions x x4 and y y8 induce a rational transformation
between C and C. In other words, the projective closures of C and C are
projectively equivalent through a rational transformation. This is because
(x, y) C(Q) y 2 = x3 + ax2 + bx 64y 2 = 64x3 + 64ax2 + 64bx
(8y)2 = (4x)2 + 4a(4x)2 + 16b(4x) (8y)2 = (4x)3 + a(4x)2 + b(4x)
(4x, 8y) C(Q).
Proposition 5.0.21. Let : C C be defined by

( 2 2
( xy 2 , y(xx2b) ), if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
(P ) = .
O[0 : 1 : 0], if P {O, T }
Then is a group homomorphism from C(Q) to C(Q) and ker =

{O, T }.
Let : C C be defined by
(
y 2 y(x2 b)
( 4x 2, 8x2
), if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
(P ) = .
O[0 : 1 : 0], if P {O, T }
Then is a group morphism from C(Q) to C(Q) whose kernel is {O, T }.

Also (P ) = 2P P C(Q) and (P ) = 2P P C(Q).
Proof: It suffices to verify the assertions made for in the proposition

above.
We first prove that is well defined i.e. (P ) C P C. If P
{O, T }, then (P ) = O C. If P (x, y) C, P 6= O, T implies x 6= 0. We
have (P ) C if and only if
2 3 2
y(x2 b) y2 y2 y2

= + a + b
x2 x2 x2 x2
y 2 (x2 b)2 y6 y4 2 y2
= 2a + (a 4b)
x4 x6 x4 x2
If y = 0, the equality is obviously true. If y 6= 0, then the equality is
equivalent to:
(x2 b)2 x2 = y 4 2ay 2 x2 + (a2 4b)x4
(x2 b)2 x2 = (y 2 ax2 )2 4bx4 (x3 bx)2 = (x3 + bx)2 4bx4

which holds by the identity (A B)2 = (A + B)2 4AB. We have proved
that is well defined. It is easy to see that (C(Q)) C(Q).
We prove that is a group homomorphism. First, some particular cases

must be excluded. We have (O) = O. (P + O) = (P ) = (P ) + O =
(P ) + (O).
We wish to prove (P + T ) = (P ) + (T ) = (P ) + O = (P ) for all
P C. For P = O, (O + T ) = (T ) = O = (O). For P = T , (T + T ) =
(O) = O = (T ). I have used that T is a point of order 2 because its
y coordinate is 0. Assume P (x, y) 6= O, T . Then x 6= 0 and x(P + T ) =
49
2
y0
2 a x 0 = a x = xb . y(P + T ) = x(P + T ) = xby2 .
x0
by 2 by b2
( b) 2 2)
In conclusion (P + T ) = ( xb2 , x2 bx22 ) = ( xy 2 , y(bx
x2
) = (P ).
x x2
We have used b 6= 0 as implied by b2 (a2 4b) = f 6= 0.
It is clear from the definition that (P ) = (P ). We now wish to
prove (P1 ) + (P2 ) = (P1 + P2 ) for all P1 , P2 C(Q). This is equivalent
to (P1 ) + (P2 ) (P1 + P2 ) = O (P1 ) + (P2 ) + (P1 P2 ) = O. Thus
it suffices to prove the following problem: Given P1 , P2 , P3 three collinear
points on C(Q), prove that (P1 ) + (P2 ) + (P3 ) = 0. The collinearity of
P1 , P2 , P3 is equivalent to P1 + P2 + P3 = O. By the preceding remarks it is
enough to consider Pj 6= O, T j = 1, 3. Assume P1 , P2 , P3 are 3 distinct.
Let y = x + be the equation of the line l passing through the three
collinear points P1 , P2 and P3 . We prove that y = x + with
b
=

and
2 a + b2
=

is the equation of a line passing through (P1 ), (P2 ) and (P3 ). Notice that
6= 0 otherwise the line l contains T and by Bezout P1 , P2 or P3 is T which
contradicts our choice for the three collinear points. Let P (x, y) = (P1 ).
It is enough to prove y = x + .
b y12 2 a + b2
x + = + =
x21
( b)y12 + x21 ( 2 a + b2 ) (y12 ax21 ) + b(2 x21 y12 ) + 2 x21

= =
x21 x21
(x31 + bx1 ) b(x1 + y1 ) + 2 x21 x31 by1 + x31 x21 y1 by1

= = = y1 .
x21 x21 x21
Therefore is a group morphism.
Checking that ker = {O, T } is a very easy exercise.

Proving that is a well defined group morphism with kernel {O, T } is
done just like for .
We are left with proving (P ) = 2P for all P C(Q) and (P ) = 2P
for all P C(Q). Since their proofs are basically the same, it is enough to
prove (P ) = 2P .
If P {O, T }, (P ) = (O) = O = 2P . Dont forget that T has order
2 in C(Q).
Let P (x, y) C(Q), P 6= O, T i.e. x, y Q and x 6= 0.

y 2 (x2 b)2 y(x2 b)2 4 !
y 2 y(x2 b) x4 x2
( xy 4 a2 + 4b)
(P ) = ( 2 , )= 4 , 4 =
x x2 4 xy 4 8 xy 4
(x2 b)2 x2 (x2 b) y 4

2
, ( 4 a + 4b) =
4y 2 8y 3 x
2
(x b)2 (x2 b)(y 4 + (4b a2 )x4 )

, . (5.0.1)
4y 2 8y 3 x2
We have that (P ) = T if and only if y = 0 if and only if 2P = O. If
2P = 0, then ((P )) = (T ) = O = 2P . Assume now y 6= 0. Then the
equality 5.0.1 is equivalent to
2
(x b)2 (x2 b)(x2 (x2 + ax + b)2 + (4b a2 )x4 )

(P ) = , =
4y 2 8y 3 x2
(x2 b)2 (x2 b)((x2 + ax + b)2 + (4b a2 )x2 )

, =
4y 2 8y 3
2
(x b)2 (x2 b)(x4 + 2ax3 + 6bx2 + 2abx + b2 )

, (5.0.2)
4y 2 8y 3
We now compute x(2P ) and y(2P ).
f 0 (x)2 (3x2 + 2ax + b)2

x(2P ) = 2 a 2x = a 2x = a 2x =
4y 2 4(x3 + ax2 + bx)
(3x2 + 2ax + b)2 4(x3 + ax2 + bx)(2x + a)

=
4y 2
9x4 + 12ax3 + (4a2 + 6b)x2 + 4abx + b2 8x4 12ax3 (4a2 + 8b)x2 4abx
4y 2
x4 2bx2 + b2 (x2 b)2
= = .
4y 2 4y 2
f 0 (x) (x2 b)2

y(2P ) = x(2P ) = x(2P )(yx) = ( x +y) =
2y 4y 2
3x2 + 2ax + b x4 2bx2 + b2 4x(x3 + ax2 + bx)
y =
2y 4y 2
(3x2 + 2ax + b)(x4 2bx2 + b2 4x(x3 + ax2 + bx)) + 8(x3 + ax2 + bx)2
=
8y 3
f 0 (x)((x2 b)2 4xf (x)) + 8f 2 (x) f 0 (x)(x2 b)2 4f (x)(xf 0 (x) 2f (x))
= =
8y 3 8y 3
51
(x2 b)2 f 0 (x) 4f (x)(3x3 + 2ax2 + bx 2x3 2ax2 2bx)

=
8y 3
(x2 b)f 0 (x) 4xf (x)
(x2 b) .
8y 3
To prove 5.0.2 It is enough to prove:
x4 + 2ax3 + 6bx2 + 2abx + b2 = (x2 b)f 0 (x) + 4xf (x).
(x2 b)f 0 (x) 4xf (x) = (x2 b)(3x2 + 2ax + b) 4x(x3 + ax2 + bx) =
3x4 +2ax3 2bx2 2abxb2 4x4 4ax3 4bx2 = x4 2ax3 6bx2 2abxb2 .
Chapter 6
Lecture VI
6.1 The Weak Mordell-Weil Theorem

Let C(Q) be the elliptic curve given by the equation y 2 = f (x) = x3 +ax2 +bx
with a, b Z and b2 (a2 4b) 6= 0. In the condition above we prove:
Theorem 6.1.1 (Weak Mordell-Weil Theorem).
|C(Q) : 2C(Q)| < .
Proof: In the previous lecture we have defined C(Q) : y 2 = f(x) =

x3 + ax2 + bx, with a = 2a and b = a2 4b, and we have proved:
Proposition 6.1.2. Let : C C be defined by

( 2 2
( xy 2 , y(xx2b) ), if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
(P ) = .
O[0 : 1 : 0], if P {O, T }
Then is a group homomorphism from C(Q) to C(Q) and ker =

{O, T }.
Let : C C be defined by
(
y 2 y(x2 b)
( 4x 2, 8x2
), if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
(P ) = .
O[0 : 1 : 0], if P {O, T }
Then is a group morphism from C(Q) to C(Q) whose kernel is {O, T }.

Also (P ) = 2P P C(Q) and (P ) = 2P P C(Q).
The following group theory theorem will provide the finishing argument
in our proof. However matching the conditions in the hypothesis of the
theorem to 6.1.1 will take some time.
Theorem 6.1.3. Let A and B be two abelian groups. Let : A B and

: B A be two group homomorphisms such that
53
54 CHAPTER 6. LECTURE VI
1. = 2 1A and = 2 1B .
2. |B : Im| < and |A : Im| < .
Then |A : 2A| < .
Proof: We prove |A : 2A| |A : Im| |B : Im|.

We have |A : 2A| = |A : (A)| = |A : (B)||(B) : (A)|. The image
of (A) through the canonical projection given by
(B)
B (B)
(A)
is trivial, therefore there exists a surjective group homomorphism B/(A)

(B)/(A) |B : Im| = |B/(A)| |(B)/(A)| = |Im : (A)|.
Therefore |A : 2A| = |A : (B)||(B) : (A)| |A : (B)||B : (A)|.
In 6.1.3 put A = C(Q), B = C(Q) and , the group homomorphisms

given in 6.1.2, to conclude that |C(Q) : 2C(Q)| < if |A : Im| < and
|B : Im| < . To finish the proof of the Weak Mordell-Weil Theorem, it
is enough to prove |A : Im| < . In doing so, we will need the following
lemma:
Lemma 6.1.4. Define

Q
: C(Q) ,
(Q )2
where Q = Q /(Q )2 is the factor group of the multiplicative group of
nonzero rational numbers by the subgroup of squares, by

1, if P = O
(P ) = b, if P = T (0, 0) .
x, for P (x, y) 6= O, T

Then:
1. is a group homomorphism.
2. ker = Im.
3. |Im| 2t+1 , where t is the number of distinct prime factors of b.
Proof: is well defined. This is because b 6= 0 (f = b2 (a2 4b) 6= 0)

and x 6= 0 if P (x, y) 6= O, P .
(1). It is clear from the definition and from x(P ) = x(P ) P C(Q)
that (P ) = (P ). Also (P )2 = 1 P C(Q). Let P, Q C(Q).
Then (P )(Q) = (P + Q) 1 = 1 1 = (P )(P )(Q)(Q) =
(P )(Q)(P + Q).
6.1. THE WEAK MORDELL-WEIL THEOREM 55
For to be a homomorphism is enough to prove that if P, Q, R are

collinear points on C(Q), then (P )(Q)(R) = 1. If all P, Q, R are differ-
ent from O and T , then let y = x + be the equation of the line passing
through them. The condition P, Q, R 6= O, T implies 6= 0. x(P ), x(Q) and
x(R) are the solutions of the system
2
y = x3 + ax2 + bx
x3 + (a 2 )x2 + (b 2)x 2 = 0
y = x +
x(P )x(Q)x(R) = 2 (P )(Q)(R) = x(P

[)x(Q) [ = 2 = 1.
[ x(R)
If one of P , Q or R is O, then without loss of generality we can assume R = O
and the Q = P . (P )(Q)(R) = (P )(P )(O) = (P )(P )1 =
(P )2 = 1. For R = T (and similarly for P and Q), the equation of the line
passing through P, Q and T is y = x. Then x(P ), x(Q) and 0 = x(T ) are
the solutions of the system
2
y = x3 + ax2 + bx
x3 + (a 2 )x2 + bx = 0 x(P )x(Q) = b
y = x
[)x(Q)
(P )(Q)(R) = x(P [ b = x(P
\ )x(Q)b = b b = 1.
We have proved that is a group homomorphism.
(2). ((O)) = (O) = 1 = (O) = ((T )). Let P (x, y) C(Q),
P 6= O, T . If y 6= 0 x 6= 0, then
y 2
x((P )) = ((P )) = 1.
4x2
If y = 0, then P 6= T x 6= 0. Then 0 = y 2 = x3 + ax2 + bx x2 + ax + b =
2 2 4b)
= (2a) 4(a
2
0 (x a2 )2 = a 4
4
b
4 = 4b bb = 4b
b = 1. We have used
that x Q and a2 4b, b 6= 0 (C is nonsingular) to make sure we get elements
of Q.
y = 0 (P ) = T (P ) = b = 1.
We have proved Im ker .
We now prove the reverse inclusion. It is clear that O Im. If
T ker , then b = (T ) = 1 b = d2 for some rational (hence integer)
d. The equation x2 + ax + b = 0 has discriminant a2 4b = 16b = 16d2 ,
hence it has an integer solution x. P (x, 0) is then a point on C(Q) such that
(P ) = T . Therefore
T ker T Im.
Let P (x, y) ker , P 6= O, T . Then x 6= 0 and 1 = (P ) = x x = u2 for
2 integer, u 6=0. We want to prove that P Im i.e.
some rational, hence
y 2 y
(x, y) = (P ) = 4x 2 , (x b) 8x2 for some P C(Q). Let
a y
x1 = 2(u2 + ), y1 = 2ux1 ,
4 u
a y
x2 = 2(u2 ), y2 = 2ux2 .
4 u
We prove that P1 (x1 , y1 ), P2 (x2 , y2 ) C(Q) and (P1 ) = (P2 ) = P
Im.
2x
2 2 x3 +ax2 + a y 2
We have x1 x2 = 4((x a4 )2 yx ) = 4((x + a2 )2 yx ) = 4 4x
4
=
2
a 4b = b 6= 0 x1 , x2 6= 0.
We want to prove yi = x3i + ax2i + bxi i = 1, 2. Since xi 6= 0, this is
2
equivalent to xyii = xi + a + xbi = a + (xi + x1xxi 2 ) = a + (x1 + x2 ) = 4u2
which holds by the definitionof yi . Therefore P1 , P2 C(Q).
y 2 yi
(Pi ) = 4xi2 , (x2i b) 8x 2 .
i i
yi2 (2uxi )2
= = u2 = x.
4x2i 4x2i
yi (x2i b) yi (x2i x1 x2 ) 2ux1 (x21 x1 x2 ) u(x1 x2 )
2 = 2 = 2 = = y.
8xi 8xi 8x1 4
We have also proved (Pi ) = P (x, y). This proves the reverse inclusion
ker Im and we have ker = Im.
(3) By 2.2.5, if P (x, y) C(Q) and P 6= O, then there exist m, n, e Z,
e 6= 0, gcd(m, e) = gcd(n, e) = 1 such that x = em2 and y = en3 .
Assume first P 6= O, T . Then m 6= 0, otherwise P = T , and
m
c
(P ) = = m.
e2
2 3 2
y 2 = x3 + ax2 + bx ne6 = m e6
+ am e4
+ b em2 n2 = m3 + am2 e2 + bme4 =
m(m + ame + be ). Let d = gcd(m, m2 + ame2 + be4 ) = gcd(m, be4 ) =
2 2 4
gcd(m, b). Since m(m2 + ame2 + be4 ) is a square, there exists m1 Z such
that m = d m21 m = d. Let b = pa1 . . . pat be the prime factor
1 t
decomposition of b with p1 , . . . , pt distinct prime numbers and ai 1 i =
1, t. Then d|b d = p11\ . . . pt t with i {0, 1} i = 1, t. It is easy
to see that these give at most 2t+1 possibilities for d, hence also for (P ).

These possibilities include d {1, b}, therefore we need not consider the
cases P {O, T }.
Now Im ' C(Q)/ ker = C(Q)/Im |C(Q) : Im| 2t+1 .
Remark 6.1.5. Similarly to the lemma above we can prove |C(Q) : Im|
2s+1 , where s is the number of distinct prime factors of b = a2 4b.
We have proved |A : Im| < and |B : Im| < and by 6.1.3 we
have the Weak Mordell-Weil Theorem.
The proof of 6.1.3 actually gives the estimation:
|C(Q) : 2C(Q)| 2s+t+2 .
By what we have seen in the previous lecture, we have finished the proof for
the Mordell-Weil Theorem in the case C(Q) has points of order 2.
6.2. COMPUTING THE RANK OF ELLIPTIC CURVES 57
6.2 Computing the rank of elliptic curves

Mordell-Weils Theorem tells that C(Q) is a finitely generated abelian group.
We have proved it just for curves that have points of order 2. The Structure
Theorem for finitely generated abelian groups tells us that C(Q) is isomor-
phic to the direct sum of a free abelian group Zr and a finite group M. By
definition, the rank of C(Q) is r. M is the torsion subgroup of C(Q) and
is characterized sufficiently well by Nagell-Lutzs Theorem. The purpose of
this section is to present a strategy for determining the rank of an elliptic
curve. We will see that the effectiveness of this strategy is given by our
ability to solve some Diophantine equations. We will work in the same hy-
pothesis: C(Q) has points of order 2 i.e. C(Q) : y 2 = f (x) = x3 + ax2 + bx,
a, b Z and b2 (a2 4b) 6= 0. Let C(Q) : y 2 = f(x) = x3 + ax2 + bx for
a = 2a and b = a2 4b. Let G = C(Q) and G0 = C(Q).
By the Structure Theorem for finitely generated abelian groups, M '
Zp1 Zp2 . . . Zpk for some k N, not necessarily distinct prime
1 2 k
numbers pi and integers i N , i = 1, k.
G/2G ' (Z2 )r (Zp1 /2Zp1 ) . . . (Zpk /2Zpk ).

1 1 k k
If p is a prime number, p 6= 2 and 1, then Zp = 2Zp . This is because

2 is a unit in Zp , hence multiplication by it defines an automorphism of
Zp . If p = 2, then 2Z2 is a subgroup of Z2 of index 2. Let
= |{i 1, k|pi = 2}|.
It is easy to see that

|G/2G| = 2r+ .
Let G(2) = {P G|2P = O}. It is easy to see that G(2) is a subgroup of

G. Moreover:
Y
G(2) ' (0 Z)r (2i 1 Z2i ),
i=1,k,pi =2
and |G(2)| = 2 . We have seen that G(2) = {O[0 : 1 : 0]} {P (x, y) G|y =
0}. Since the equation f (x) = 0 has at most 3 integer roots, we find that
|G(2)| 4. G(2) contains at least the elements O and T (0, 0) of G and its
order is a power of 2 least or equal to 4, hence |G(2)| {2, 4}. |G(2)| = 4
if and only if there are 3 points of order 2 i.e. the equation f (x) = 0 has 3
integer solutions i.e. the equation x2 + ax + b = 0 has 2 integer roots. The
equation x2 + ax + b = 0 has 2 integer roots if and only if a2 4b = c2 for
some integer c. Note that b2 (a2 4b) 6= 0 implies that c 6= 0 and that the
roots of f are all distinct.
For : G G0 and : G0 G defined as in 6.1.2, and for : G Q

as defined in 6.1.4, we have:
|G/2G| = |G/(G)| = |G/(G0 )||(G0 )/(G)| = |Im||(G0 )/(G)|.
The kernel of the canonical map
G0 (G0 ) (G0 )/(G)
is the set of elements g 0 G0 such that (g 0 ) (G) (g 0 ) = (g)

for some g G. (g 0 ) = (g) g 0 (g) ker g 0 (G) + ker .
We have proved ker(G0 (G0 )/(G)) (G) + ker . The reverse
inclusion is obvious, hence ker(G0 (G0 )/(G)) = (G) + ker , and by
the Fundamental Theorem of Isomorphism applied several times in many of
its forms:
G0 G0
(G0 ) G0 (G) (G)
' ' (G)+ker
' ker

(G) (G) + ker
(G) ker (G)
(G0 ) G0 ker |Im0 |

| |=| |:| |=
(G) (G) ker (G) s
for s = | ker ker 0 0
(G) | and for : G Q a morphism defined similarly to .
We have proved in 6.1.2 that ker = {O, T }, hence s {1, 2}.
s = 1 ker = ker (G) ker (G). O = (O) is automat-
ically in (G), therefore ker (G) T (G) T ker 0 bb =
0 (T ) = 1 b = a2 4b is a perfect square.
|Im0 |
|G/2G| = |Im|
s
|Im||Im0 | |Im||Im0 |
2r |G(2)| = 2r+ = |G/2G| = 2r = .
s s |G(2)|
if a2 4b is a perfect square

1 4,
We have s |G(2)| = =4
2 2, if a2 4b is not a perfect square
|Im| |Im0 |
2r = .
4
We wish to find Im. Let P (x, y) G. By 2.2.5 there exist m, n, e Z
such that x = em2 , y = en3 , e 6= 0 and gcd(m, e) = gcd(n, e) = 1. The case
e = 0 corresponds to P = O.
P G = C(Q) n2 = m(m2 + ame2 + be4 ).
Let b1 = gcd(b, m) = gcd(m, m2 + ame2 + be4 ), where the sign is taken

such that bm1 is a nonnegative integer. Then there exist m1 N and b2 Z
such that m = b1 m1 , b = b1 b2 and gcd(m1 , b2 ) = 1.
n2 = b1 m1 (b21 m21 + ab1 m1 e2 + b1 b2 e4 ) = b21 m1 (b1 m21 + am1 e2 + b2 e4 )

t2 = m1 (b1 m21 + am1 e2 + b2 e4 ) for an integer t such that n = b1 t. Since
gcd(m1 , b1 m21 +am1 e2 +b2 e4 ) = gcd(m1 , b2 e4 ) = gcd(m1 , e4 ) = 1 and m1 0,
we find that there exist M, N Z such that m1 = M 2 and b1 m21 + am1 e2 +
b2 e4 = N 2 .
gcd(m, e) = 1 gcd(b1 , e) = gcd(M, e) = 1.
gcd(n, e) = 1 gcd(t, e) = 1 gcd(N 2 , e) =
gcd(b1 m21 + am1 e2 + b2 e4 , e) = 1 gcd(N, e) = 1.
1 = gcd(m1 , b1 m21 + am1 e2 + b2 e4 ) = gcd(M 2 , N 2 ) gcd(N, M ) = 1.
gcd(m1 , b2 ) = 1 gcd(M 2 , b2 ) = 1 gcd(b2 , M ) = 1.

If P (x, y) 6= O, T , we have e 6= 0 and m 6= 0 M 6= 0, and:
m 2
= m = b\
1 M = b1 .
c
(P ) =
e2
If M = 0, then P = T (P ) = b. (O) = 1.
Conversely, if there exist M, N, e, b1 , b2 Z such that:
b1 M 4 + aM 2 e2 + b2 e4 = N 2

b1 b2 = b ,
gcd(M, e) = gcd(M, N ) = gcd(e, N ) = gcd(b2 , M ) = gcd(b1 , e) = 1

(6.2.1)
2
then x = b1eM
2 and y = b1 M N
e3
give a point P G and (P ) = b 1 .
Remark 6.2.1. The conclusion is q Im if and only if there exist b1 , b2

Z such that b1 = q, b1 b2 = b and there exists a solution M, N, e to the system
6.2.1.
We have a similar result for Im.
Notice that 1 and b are always in Im as the images of O and T respec-

tively. T corresponds to b1 = b, b2 = 1, M = 0, N = 1 and e = 1. O
corresponds to e = 0, b1 = 1, N = 1 and M = 1.
The remark 6.2.1 reduces determining Im to finding solutions to sys-

tems of type 6.2.1. These are basically Diophantine equations, but we dont
have to solve them, but find one solution subjected to the conditions in
6.2.1, or prove that the finite number of systems with the same b1 dont
have solutions. We will see that this is not always an easy task as many
problems on the subject are still left open.
Example 6.2.2. Lets begin our examples by characterizing C(Q) : y 2 =

x3 x.
Solution: We have seen, for example in 3.2.2, that the torsion group of
C(Q) is isomorphic to Z2 Z2 and as a set is {O, T, (1, 0), (1, 0)}.
0|
We now compute the rank r of C(Q). We know 2r = |Im||Im 4 . We
use 6.2.1 to determine |Im| and |Im0 |.
To find |Im|, we must solve the systems 6.2.1 for all b1 b2 = b = 1
i.e. (b1 , b2 ) {(1, 1), (1, 1)}. The least we can say is |Im| 2. But
{1, 1}
c Im and we get |Im| = 2.
C(Q) : y 2 = x3 + 4x. We apply again 6.2.1 to find |Im0 |. We must
solve systems of type 6.2.1, but the equation is b1 M 4 + b2 e4 = N 2 and
b1 b2 = b = 4. For the cases (b1 , b2 ) {(1, 4), (2, 2), (4, 1)}, the
equation b1 M 4 + b2 e4 = N 2 only has the trivial solution (M, N, e) = (0, 0, 0)
which fails the restriction gcd(M, N ) = 1 in 6.2.1. The remaining cases
b1 = b2 = 2 and (b1 , b2 ) {(1, 4), (4, 1)} are bound to give solutions as they
0|
correspond to 2 or 1 being in Im0 , and 2r = |Im 2 |Im0 | 2. In
conclusion r = 0 and C(Q) is a torsion group isomorphic to Z2 Z2 .
We have proved that the only rational solutions of the equation y 2 =
x3 x are (0, 0), (1, 0).
Example 6.2.3. Our second example is a famous one, Eulers equation.
Find the rational solutions of the equation y 2 = x3 + 1.
In the next lecture we will provide an elementary solution for it. Com-
paring the lengths and depths of the two proofs will show the strength of
Nagell-Lutzs and Mordell-Weils Theorems.
Proof: We will do more than finding the rational solutions. We will
also characterize the abelian group C(Q) : y 2 = x3 + 1. The torsion points
of C(Q) are, as seen for example in 3.2.3, O, (0, 1), (1, 0) and (2, 3).
Because our proof for the Mordell-Weil Theorem and our algorithm for
finding the rank of an elliptic curve are restricted to the assumption T
C(Q), we will first make the substitution x x 1 to find the elliptic curve
that we also denote C(Q) : y 2 = x3 3x2 + 3x. We will prove that the rank
r of C(Q) is 0.
We first find |Im|. We have a = 3 and b = 3. The equations to
solve in the systems 6.2.1, are b1 M 4 3M 2 e2 + b2 e4 = N 2 for b1 b2 = 3.
We have seen that 1, 3 are in Im as the images of O and T respectively.
1
c and 3 c are not as the associated equations only give the trivial solution
(M, N, e) = (0, 0, 0) which contradicts gcd(N, M ) = 1. Therefore |Im| = 2.
For |Im0 | we have a = 6 and b = 3. Again 1, 3
c Im0 . For 3 to be
0 4 2 2 4 2
in Im , the equation 3M +6M e e = N must have a solution satisfying
the conditions in 6.2.1. Reducing mod 3 we find e4 = N 2 (mod3) which
implies 3|e and 3|N , thus contradicting gcd(N, e) = 1. Therefore 3 6 Im0
c 6 Im. We have |Im0 | = 2 and r = 0. It is not hard
and similarly 1
to prove that the rank of the elliptic curve associated to Eulers equation
is also 0 and conclude that the only rational solutions of the equation are
(0, 1), (1, 0) and (2, 3).
Corollary 6.2.4. The only rational solutions of the equation 1 + 2x3 = y 3

are (0, 1) and (1, 1).
t1 3
Proof: Let 2 = 2x t2 = (2xy)3 + 1. From Eulers equation we
t+1 3
2 = y
get (2xy, t) {(0, 1), (0, 1), (1, 0), (2, 3), (2, 3)}.
t = 1 x = 0, y = 1.
t = 1 2x3 = 1 which has no rational solution. t = 0 and 2xy = 1 is

impossible. t = 3 y 3 = 2 which has no rational solution.
t = 3 x = 1, y = 1.
Therefore the only rational solutions to 1 + 2x3 = y 3 are (0, 1) and (1, 1).
Corollary 6.2.5. The only rational solution of the equation 1 + 4x3 = y 3

is (0, 1).
Proof: With the same notations and substitutions as in the previous

corollary, we get t2 = (4xy)3 + 1, so
(4xy, t) {(0, 1), (0, 1), (1, 0), (2, 3), (2, 3)}.
Modulo 4 we see that these possibilities restrict to (4xy, t) {(0, 1), (0, 1)}.
This means 4xy = 0. y 6= 0, otherwise we get a contradiction modulo 2 in
1 + 4x3 = 0, hence x = 0 and y = 1.
Chapter 7
Lecture VII
7.1 Eulers Equation

As an application to Mordell-Weils Theorem and to Nagell-Lutzs Theorem,
we have proved that the only rational solutions of the equation y 2 = x3 + 1
are (x, y) {(0, 1), (1, 0), (2, 3)}. It turns out that this diophantine
equation is quite a famous one. It caries Eulers name as back to 1738,
he was the first to find a proof, even though it turned out he made some
mistakes. Based on Eulers ideas, we will give an elementary proof, free of
the machinery that we have developed earlier.
The disadvantage of the elementary proof is that it is quite difficult and
the ideas are not at all transparent. We will need the following statement
which is apparently not connected to Eulers equation:
Proposition 7.1.1. There are no integers b, c N , gcd(b, c) = 1, b 6= c,

3 6 |c such that bc(c2 3bc + 3b2 ) is a perfect square.
Before proving the proposition, lets see how it applies to:
Theorem 7.1.2 (Euler(1738)). The only rational solutions to y 2 = x3 + 1

are (x, y) {(0, 1), (1, 0), (2, 3)}.
Proof: Assume for a contradiction that there exist x, y Q such that

y2 = x3 + 1 and x 6 {1, 0, 2}.
By 2.2.5 there exist m, n, e Z, e > 0, gcd(m, e) = gcd(n, e) = 1 such
2 3
that x = em2 , y = en3 . Then we have ne6 = m e6
+ 1 n2 = m3 + e6 =
2 2 2 4
(m + e )(m me + e ). We now seek to apply 7.1.1. For this, we must
first define b, c. Let c = m + e2 and b = e2 . Then m = c b and n2 =
c((c b)2 (c b)b + b2 ) = c(c2 2bc + b2 bc + b2 + b2 ) = c(c2 3bc + 3b2 ).
Since b is a perfect square we get that bc(c2 3bc+3b2 ) is a square. Obviously
b > 0. We have c2 3bc + 3b2 = (c 32 b)2 + 34 b2 > 0. Since c(c2 3bc + 3b2 )
is a perfect square we get c 0. If c = 0, then n2 = c(c2 3bc + 3b2 ) =
0 y = 0 x = 1 which contradicts our choice for x. Therefore c > 0.
63
64 CHAPTER 7. LECTURE VII
gcd(b, c) = gcd(m + e2 , e2 ) = gcd(m, e2 ) = 1. If b = c, then m = 0 x = 0

and we again contradict the choice for x. On applying 7.1.1 we find that the
only possibility is 3|c.
Let c = 3d, Then (ne)2 = 9bd(3d2 3bd+b2 ) implies that bd(b2 3bd+3d2 )
is a square. It is clear that b, d N . Also 1 = gcd(b, c) = gcd(3d, b)
gcd(d, b) = 1 and 3 6 |b. Again by 7.1.1 the only possibility left is b = d. But
gcd(b, d) = 1 implies b = d = 1 c = 3 1 = e2 = b and 3 = c = m + e2
m = 2, e = 1 x = 2 which of course contradicts the choice of x.
Since obviously {(0, 1), (1, 0), (2, 3)} are all the solutions to y 2 =
x3 + 1 with x {1, 0, 2}, we have finished solving Eulers equation.
Proof of 7.1.1: The main idea of the proof is Fermats descent ar-
gument. Assume that there exists a solution with the required proper-
ties. Chose one for which b is minimal. We have gcd(b, c2 3bc + 3b2 ) =
gcd(b, c2 ) = 1 and gcd(c, c2 3bc + 3b2 ) = gcd(c, 3b2 ) = 1. Since bc(c2
3bc + 3b2 ) is a square and b, c, c2 3bc + 3b2 are pairwise coprime positive
integers, they must all be squares. Hence there exists k N such that
c2 3bc + 3b2 = k 2 . Let k+c m
b = n with gcd(m, n) = 1, m, n N . Then
m
k = n b c.
2
c2 3bc + 3b2 = ( m 2 2 2 m 2 m
n b c) c 3bc + 3b = n2 b 2 n bc + c
2
2
3(b c)n2 = m2 b 2cmn b(m2 3n2 ) = c(2mn 3n2 ) cb = 2mn3n
m2 3n2
.
m

Lets prove that m2 3n2 > 0 n > 3. This is equivalent to
k+c

b > 3 k > b 3 c. If b 3 c 0 then the inequality is obviously

equivalent to k 2 >
true. Ifb 3 c > 0 then the inequality is (b 3 c)2 =
3b2 2 3bc + c2 c2 3bc + 3b2 > c2 2 3bc + 3b2 3 < 2 3 9 < 12
which holds.
b 2mn3n2 2 2 2 2 2
c = m2 3n2 and m > 3n imply 2mn 3n , m 3n N .
2
Assume 3|m. Then m = 3l cb = 2nln3l2 n2
and gcd(3l, n) = 1. Assume
there exists a prime number p such that p|2ln n2 and p|3l2 n2 . Then
p|n(2l n) and p|(3l2 n2 ) (2ln n2 ) = l(3l 2n). These give four
possibilities:
1. p|gcd(n, l) = 1;
2. p|gcd(n, 3l 2n) = gcd(n, 3l) = 1;
3. p|gcd(2l n, l) = gcd(n, l) = 1;
4. p|gcd(2l n, 3l 2n) p|2(2l n) (3l 2n) = l p|2l (2l n) =

n p|gcd(n, l) = 1.
Since all cases lead to the contradiction p|1, gcd(2ln n2 , 3l2 n2 ) = 1.

2
This, gcd(b, c) = 1, (b, c, 2nl n2 , 3l2 n2 > 0) and cb = 2nln
3l2 n2
imply
7.2. COMPUTING THE RANK OF NONSINGULAR ELLIPTIC CURVES65
b = 2nl n2 = b21

. Recall that b, c are squares. So we have c21 +n2 = 3l2 .
c = 3l2 n2 = c21
It is well known that:
Remark 7.1.3. If p is a prime number p 3(mod4) then p|a2 +b2 p|a, p|b
for all a, b Z.
Using this remark, we have 3|c1 and 3|n which contradicts gcd(3l, n) = 1.
Actually, a descent argument proves that the only integer solutions of c21 +
n2 = 3l2 are c1 = n = l = 0. The conclusion is 3 6 |m. As in the case 3|m, we
b 2 = b = 2mn 3n2
prove gcd(2mn 3n2 , m2 3n2 ) = 1. This implies 1 .
c21 = c = m2 3n2
Let pq = c1n+m such that p, q N , gcd(p, q) = 1 and the sign is taken
such that 3 6 | c1 + m. Lets see that this choices are possible. We have
(m c1 )(m + c1 ) = m2 c21 = m2 c = 3n2 > 0 m c1 > 0, so we can
chose positive p, q. Since 3 6 |m, we cannot simultaneously have 3|m + c1 and
3|m c1 .
Lets see that (q, p) is a solution to the problem with q < b. If we prove
this, then we contradict the minimality of b in the choice of (b, c) and we
solve the problem.
We have (n pq m)2 = (c1 )2 = c = m2 3n2 n( pq )2 2m pq =
m p2 +3q 2 b 2mn3n2
3n np2 2mpq + 3nq 2 = 0 n = 2pq . n2 = n2
= 2m
n 3 =
p2 +3q 2 p2 3pq+3q 2 b1 2
pq 3 = pq pq(p 3pq + 3q ) = (pq n ) pq(p 3pq + 3q 2 )
2 2 2
is a perfect square and n|pqb1 .

p2 +3q 2
p=q m n = 2pq = 2 m = 2n. Since gcd(m, n) = 1, we get n = 1,
m = 2. But then c = 1 and b = 1 contradicting b 6= c. Therefore p 6= q.
We now prove q < b. Assume for a contradiction q b. We have 2mpq =
n(p2 + 3q 2 ) q|n(p2 + 3q 2 ). gcd(p, q) = 1 gcd(q, p2 + 3q 2 ) = 1 q|n.
b = 2mn 3n2 n|b. So q|n|b. This and q b imply q = n = b. Then
p2 +3q 2 2 2
m
n = 2pq m = p +3q
2p 2pm = p2 + 3q 2 p|3q 2 . But our choices for
2
p, q prove gcd(p, 3q) = 1, hence p = 1 m = 3q 2+1 . n = b = 2mn 3n2
2
1 = 2m 3n m = 3n+1 2 .
3n+1
2 = m = 3n 2+1 n = n2 n = 1
b = 1 m = 2 c = 1. We again contradict b 6= c. So (q, p) is also a
solution of the problem with q < b contradicting the minimality of b among
the solutions of the problem.
7.2 Computing the rank of nonsingular elliptic curves

We give more examples of how to compute the rank of an elliptic curve. The
setting is:
C(Q) is the smooth rational elliptic curve given by y 2 = x3 + ax2 + bx
with a, b Z. This amounts to b(a2 4b) 6= 0. Let O = [0 : 1 : 0] be the
point at infinity of C(Q) and let T = (0, 0). Let C(Q) be the smooth rational
elliptic curve given by y 2 = x3 + ax2 + bx with a = 2a and b = a2 4b.
Let : C(Q) Q /(Q )2 be defined by

x, if P = (x, y) 6= (0, 0), O
(P ) = b, if P = T (0, 0) .

1, if P = O
We have proved that is a group morphism. Similarly we define 0 for

0|
C(Q). We have proved that 2r = |Im||Im
4 , where r is the rank of C(Q).
We have seen that finding Im (or Im0 ) amounts to finding solutions
for equations of the form:
N 2 = b1 M 4 + aM 2 e2 + b2 e4
with the numerous but important conditions: b1 b2 = b, 1 = gcd(M, e) =

gcd(M, N ) = gcd(e, N ) = gcd(b2 , M ) = gcd(b1 , e). We say that the equation
above is associated to b1 and we call one of its solutions good if it verifies
the numerous conditions. We have seen that a good solution to the
2 b MN
equation above corresponds to the point P (x, y) = ( b1eM 2 ,
1
e3
) (P = O if
e = 0) on C(Q) such that (P ) = b1 . Conversely, l Im if and only if
there exists b1 like above with b1 = l such that the associated equation has
a good solution. The equations for 0 change accordingly.
Example 7.2.1. Let C(Q) be the smooth rational elliptic curve given by
y 2 = x3 + x. Compute the rank of C(Q).
Solution: C is given by C : y 2 = x3 4x, a = 0, b = 4.
We first compute |Im|. Since b = 1, there are only two cases to consider:
1. b1 = b2 = 1. 1 is always in the image of as (O).
2. b1 = b2 = 1. The associated equation N 2 = M 2 e4 obviously
has only the solution (N, M, e) = (0, 0, 0), which does not verify the
numerous conditions. So 1
c 6 Im.
We have proved that |Im| = 1.

We now compute |Im0 |. b = 4 and it seems like there is a bit more
work to be done. The cases to consider are b1 {4, 2, 1}. Since 4 = 1
in Q /(Q )2 we have only to consider the possibilities b1 {2, 1}. This
means that |Im0 | 4. On the other hand, 4||Im||Im0 | = |Im0 |. These
imply |Im0 | = 4 r = 0.
The rank of C(Q) being 0, C(Q) must be a torsion group. We have seen
in 3.2.2 that the torsion points of C(Q) are T and O. Combining our results
we have proved:
Theorem 7.2.2. The only rational solutions to the equation y 2 = x3 + x
are (x, y) = (0, 0).
7.2. COMPUTING THE RANK OF NONSINGULAR ELLIPTIC CURVES67
Corollary 7.2.3. If N, M, e are rational numbers such that N 2 = M 4 + e4 ,

then e = 0 or M = 0.
2
Proof: If e 6= 0, then x = M e2
and y = Me3N define a point on C(Q) :
y 2 = x3 + x. By the previous theorem, we get (x, y) = (0, 0) M = 0.
As a consequence of this corollary we have Fermats Theorem for expo-
nent 4:
Theorem 7.2.4. All the integer solutions to the equation X 4 + Y 4 = Z 4
satisfy XY Z = 0.
7.2.1 The curves Cp

This subsection is entirely devoted to the special class of elliptic curves
{Cp (Q)|p is a prime number}.
Definition 7.2.5. Let p be a prime number. Cp (Q) is the elliptic curve
given by y 2 = x3 + px.
Remark 7.2.6. If p is a prime number, then Cp is smooth. This is because
Cp : y 2 = f (x) = x3 + ax2 + bx = x3 + px and f = b(a2 4b) = 4p2 6= 0.
The curve Cp is given by Cp : y 2 = x3 4px, a = 0 and b = 4p.
Theorem 7.2.7. If p is a prime number, then rank(Cp (Q)) 2.
Proof: We have b = p, so the elements of Im are given by the solvabil-
ity of the equations we have seen before for b1 {1, p}. b1 {1, p}
dont give points in the image of because the associated equations N 2 =
M 2 pe4 and N 2 = pM 2 e4 only have the trivial solution (0, 0, 0). On
the other hand, 1 = (O) and p = (T (0, 0)). These prove that |Im| = 2.
b = 4p. To find Im0 we must solve equations for
b1 {1, 2, 4, p, 2p, 4p}.
Since 4 = 1 in Q /(Q )2 , we need only consider the 8 cases {1,

c 2,
c p,
c 2p}.
d
0 r |Im||Im0 | 28
Therefore |Im | 8 2 = 4 4 = 4 r 2.
Proposition 7.2.8. If p is a prime number, p 7, 11(mod16), then rank(Cp (Q)) =
0.
Proof: We have seen that |Im| = 2 and we know that 1 and 4pd = p c
0 r |Im||Im0 |
are in Im as the images of O = O and T = T . Since 2 = 4 , we
only have to prove that the associated equations for b1 {1, 2, p, 2p}
dont provide elements in Im0 .
Lets assume we have proved that none of the elements 1, c 2 c is in
Im . Since is a group homomorphism, Im is a group. If p Im0 ,
0 0 0
then 1c = p c p Im0 which is a contradiction. If 2p

d2 = p d Im0 ,
then 2 [2 = p
c = 2p d Im0 which again would be a contradiction.
c 2p
If b1 = 1 the associated equation is N 2 = M 4 + 4pe4 . We have

p|N 2 + M 4 and p 3(mod4) implying that p|gcd(M, N ) = 1 which is a
c 6 Im0 .
contradiction. Similarly we treat the case b1 = 4. Therefore 1
If b1 = 2, the associated equation is N 2 = 2M 4 2pe4 .
If p 11(mod16), then we have N 2 2M 4 (modp). If p 6 |M N, then

2 must be a quadratic residue mod p. But the Legendre symbol, p2 =
p2 1
(1) 8 = 1 if p 3, 5(mod8) which happens since p 11(mod16)
p 3(mod8). But p2 = 1 2 is not a quadratic residue. The other
possibility p|M N is easily seen to contradict gcd(M, N ) = 1.
If p 7(mod16), then N 2 = 2M 4 2pe4 N = 2n for some integer
n. Substituting in the equation and reducing by 2 we get 2n2 = M 4 pe4 .
It is easy to see that being coprime, both M and N must be odd. If x is
an odd number, then x 1, x + 1, x2 + 1 are all even numbers and exactly
one of the numbers x 1 is divisible by 4. The conclusion is 16|x4 1 =
(x 1)(x + 1)(x2 + 1) x4 1(mod16). 2n2 = M 4 pe4 2n2 1 7 =
6(mod16) n2 3(mod16) n2 3(mod8) which is a contradiction
because x2 0, 1, 4(mod8) for all integers x. We have proved 2 6 Im0 .
If b1 = 2, then the associated equation is N 2 = 2M 4 + 2pe4 .
If p 7(mod16), then -2 is not a square mod p and we treat this case
similarly to the previous one.
If p 11(mod16), then like for b1 = 2 we reach to the contradiction
n2 5(mod16).
We have proved that bb1 {1, c are not in Im0 , therefore |Im0 | =
c 2}
2 r = 0.
Proposition 7.2.9. If p is a prime number, p 3, 5, 13, 15(mod16), then
rank(Cp (Q)) 1.
It is conjectured that in these cases, rank(Cp (Q)) = 1.
Proof: We have C : y 2 = x3 4px, b = 4p. We have seen that
|Im| = 2 and |Im0 | 8. All we have to prove is |Im0 | 4. Because
Im0 is a subgroup, all that we actually have to prove is that not all of the
equations associated to b1 {1, 4, 2, p, 4p 2p} give elements in
Im0 . This is because the set above gives 8 elements in Q /(Q )2 .
If p 3, 15(mod16), then p 3(mod4) and then N 2 = M 4 +4pe4 1
is a quadratic residue mod p or p|gcd(M, N ). These contradict p 3(mod4)
c 6 Im0 (the case b1 = 4 is dealt with similarly).
or gcd(M, N ) = 1. So 1
p2 1
p1
If p 5, 13(mod16), then p 5(mod8) and 2 p = (1) 8 + 2 = 1
if p 5, 7(mod8). The equation for b1 = 2 is N 2 = 2M 4 + 2pe4 . Just
as before this implies that -2 is a quadratic residue mod p, or p|gcd(M, N ).
These contradict p 5(mod8) or gcd(M, N ) = 1. Therefore 2 c 6 Im0 .
Example 7.2.10. As an example to the conjecture stated before, lets con-
sider the case p = 13.
7.3. STORIES AND CONJECTURES 69
Solution: If b1 = 4 then we have the solution (N, M, e) = (3, 1, 1),

so 4
c = 1 c Im0 . Since {1, 4, d Im0 and Im0 is a group,
c 13}
0
|Im | 4. This proves r 1. Combining with the previous proposition
we get r = 1.
Example 7.2.11. rank(C73 (Q)) = 2.
Solution: (N, M, e) = (3, 4, 1) provides a solution for b1 = 4. There-
fore 4 c Im0 . (N, M, e) = (12, 1, 1) provides a solution for b1 = 2,
c = 1
c Im0 . We know 1, 292
so 2 [ = 73 d Im0 . It is easy to see that these
four elements generate a group with 8 elements, hence |Im0 | = 8 and then
r = 2.
Remark 7.2.12. It can be proved that rank(Cp (Q)) = 0 for p = 17 or p =
41. An interesting example that dampens our hopes in finding an algorithm
for determining the rank of C(Q), that would be based on solving equations
like we did this section, is given by the equation associated to b1 = 17 if
p = 17. This equation, N 2 = 17M 4 4e4 , has a solution modulo any prime
number q, but has no solution in integers.
7.3 Stories and conjectures about the rank of an

elliptic curve
The stories that follow can be found in A. Wiless description of the Birch
and Swinnerton-Dyer conjecture on Claymaths Web site.
7.3.1 Congruent numbers

A positive integer d is called a congruent number if there exist positive
rational numbers a, b, c that are the side lengths of a right triangle with area
d. A connection of congruent numbers with elliptic curves is given by the
following result:
Proposition 7.3.1. d N is a congruent number if and only if the equation
y 2 = x3 d2 x has integer solutions with y 6= 0.
Proof: Obviously we can reduce to the case when d is square-free. Since
d 6= 0, C(Q) : y 2 = x3 d2 x is a smooth elliptic curve. As an easy con-
sequence of Nagell-Lutzs Theorem it can be proved that the only torsion
points of C(Q) are {O[0 : 1 : 0], (0, 0), (d, 0)} if d is square-free. Therefore
proving the proposition is equivalent to proving that rank(C(Q)) 1. This
reveals the connection of this problem with the theory of elliptic curves.
However we give an elementary proof for it.
2 3 2
2 Let x, y be
rational
2 2 with y 6= 0 and y = x d x. Let a =
number
x d2 2xd x +d
y , b = y and c = y . Then a2 + b2 = c2 , so a, b, c are the sides

2 d2 )
3 2
of a right triangle whose area is 2xd(x
x d x
= d y2 = d.

2y 2
Conversely, assume d is a congruent number and let a, b, c Q+ such

that a2 + b2 = c2 and ab = 2d. Set x = 12 a(a c) and y = 12 a2 (a c). It
is easy to prove that y 6= 0.
a4 (a c)2
y2 =
4
a3 (a c)3 a(a c)d2
x3 d2 x = =
8 2
a3 (a c)3 a(a c)a2 b2 (a c)2 b2
= = a3 (a c) =
8 8
a2 2ac + c2 b2 a2 2ac + a2 a4 (a c)2
= a3 (a c) = a3 (a c) =
8 8 4
So y 2 = x3 d2 x and y 6= 0.
Example 7.3.2. 6 and 5 are congruent numbers.
Solution: For d = 6 we can take a = 3, b = 4 and c = 5.
For d = 5 we can take a = 23 , b = 20 41
3 and c = 6 . History assigns the
proof of 5 being a congruent number to Fibonacci.
The weak form of Birch and Swinnerton-Dyer conjecture you can read
about in the next subsection implies that every positive integer congruent
modulo 8 to 5,6 or 7 is a congruent number.
Theorem 7.3.3 (Tunnel). Let d be an odd, square-free positive integer.
Then d is a congruent number if and only if
|{(x, y, z) Z3 | 2x2 +y 2 +8z 2 = d}| = 2|{(x, y, z) Z3 | 2x2 +y 2 +32z 2 = d}|.
7.3.2 The Birch and Swinnerton-Dyer Conjecture

It is an important problem to determine if there are infinitely many rational
solutions (x, y) to the equation y 2 = x3 + ax + b with a, b Z.
If C(Q) : y 2 = x3 + ax + b is not smooth i.e. f (x) = x3 + ax + b
has multiple complex roots, then it can be proved that C(Q) has infinitely
many rational points. For this, consider a line l in A3 that has an equation
with rational coefficients. Then let be a multiple root of f . It follows
that Q. Consider the point with rational coordinates P (, 0) C(Q).
For every point Q with rational coordinates on l, prove that the line P Q
intersects C again in a point of rational coordinates i.e. in a point of C(Q).
Since there are infinitely many rational points on l, we conclude that C(Q)
is infinite.
Therefore we can consider the following problem: When is C(Q) infinite?
In 1901 Poincare gave the abelian group structure on C(Q). In 1922
Mordell completed a version of the Mordell-Weil Theorem that proved C(Q)
7.3. STORIES AND CONJECTURES 71
is a finitely generated abelian group. Mordells result and Nagell-Lutzs

Theorem proves that C(Q) is infinite if and only if rank(C(Q)) 1.
But how to decide whether the rank of an elliptic curve is greater or
equal to one? No one knows for sure, but an idea is to find a connection
between the rank of the elliptic curve and the number of mod p points of
C(Q) for every prime number p for which this makes sense. Lets make the
terms more clear.
In the beginning of the third Lecture we have proved that if p is a prime
number such that p 6 | 2f , where C(Q) : y 2 = f (x), f = x3 +ax2 +bx+c and
a, b, c Z, then it makes sense to define C(Zp ) which is also a group. In our
case, f = x3 + ax + b. These mean that except for a finite number of prime
numbers, we can define C(Zp ) and give it an abelian group structure. For
these primes p, define N (p) = |C(Zp )|. A result concerning these numbers,
N (p), is:
Theorem 7.3.4 (Hasse-Weil). If p is a prime number such that p 6 | 2f ,

then:

|p + 1 N (p)| 2 p.
Similarly to the definition of Riemanns function

!
X 1 Y 1
(s) = = ,
i=1
ns p
1 p1s
for <e(s) > 1, where the product is taken over all the prime numbers, the
following function was introduced:
1
Y ap 1
L(C, s) = 1 s + 2s1 ,
p p
p6| 2f , p is prime
where ap = p+1Np . It can be proved that L(C, s) converges for <e(s) > 32 .
Hasse conjectured that L(C, s) has a holomorphic extension to C. In 1999,
Taylor, Breuil, Conrad and Diamond proved Hasses Conjecture.
A weak form of The Birch and Swinnerton-Dyer Conjecture states:
Conjecture 7.3.5. C(Q) is infinite if and only if L(C, 1) = 0.
This means that to find out if C(Q) is infinite, it is enough to compute

Q 1
N (p)
L(C, 1) and see if it is 0. Roughly speaking, L(C, 1) = p p =
Q p
p N (p) . The strong version of the conjecture would give even more infor-
mation on C(Q).
Conjecture 7.3.6 (Birch and Swinnerton-Dyer). The order of 1 as a

zero of L(C, s) is r = rank(C(Q)).
Using results of Wiles, Coates (1977) and Zagier, Gross (1983), Koly-
vagin proved in 1990 that for modular elliptic curves the following hold:
L(C, 1) 6= 0 rank(C(Q)) = 0
L(C, 1) = 0 and L0 (C, 1) 6= 0 imply that rank(C(Q)) = 1.

You may not know what a modular elliptic curve is, but in 1994, while
solving Fermats Last Theorem, A. Wiles proved that every elliptic curve is
modular.
Chapter 8
Lecture VIII
In this lecture we give a complete proof to the Mordell-Weil Theorem. This

proof is not elementary and some knowledge in Algebraic Number Theory
is required.
8.1 Algebraic Number Theory Prerequisites

Definition 8.1.1. A number field is an algebraic extension Q K of finite
degree.
Let K be a number filed with [K : Q] = n. K is an algebraic extension

of Q, hence every element of K is a root of a polynomial with rational co-
efficients. By multiplying the coefficients of the polynomial by a convenient
nonzero integer, we can assume that all of them are integers. An element
of K is called integral if it is root to a monic polynomial with integer coef-
ficients.
Proposition 8.1.2. Let A be the set of integral elements of K. Then:
1. A is a ring with quotient field K. A is called the ring of integers of

K. Any element of K can be written as m a
with a A and m Z .
2. A is a free abelian group of rank n. Any basis for A as a free abelian

group is also a basis for K as a vector space over Q.
3. Any nonzero prime ideal of A is maximal.
4. Any nonzero ideal in A decomposes uniquely, up to the order of factors,

as a product of prime ideals.
def
5. If I is a nonzero ideal of A, then N (I) = |A/I| is a finite positive
integer. It is called the norm of the ideal I.
We have N (I J) = N (I) N (J) for all nonzero ideals I, J of A.
73
74 CHAPTER 8. LECTURE VIII
If x K, then we can define the norm N (x) of x over Q as the product

of all the conjugates of x i.e. the product of all the distinct complex
roots of the irreducible polynomial of x over Q. It can be proved that
N (xA) = |N (x)|[K:Q(x)] .
6. If I, J A are nonzero ideals of A, then I J J|I i.e. there exists

an ideal L of A such that J L = I.
7. It can be proved that N (I) I which implies I| N (I) A.
8. If P is a nonzero prime ideal of A, then it is maximal, so A/P is

a field. This means N (P ) = pk for some prime number p and some
k N. Since N (P ) P , pk P and since P is a prime ideal, p P ,
so P | pA.
This can be used to prove that every ideal whose norm is a power of
a prime number p is a product of prime ideals dividing pA and every
ideal whose norm is not a prime power is not prime.
9. If I, J A and I, J 6= (0), we say I J if and only if there exist

a, b A such that aI = bJ. is an equivalence relation on the
set of nonzero ideals of A and the factor set C = {I A|I 6= (0)}/
inherits a group structure from the multiplication of ideals. This means
def
that if we set I J = Id
J, this is a well defined operation on C such
that (C, ) is a group. The neutral element is the class 1 of all principal
ideals of A. We sometimes use the notation I J(mod P rA) to say
that I and J have equal classes in C.
Probably the most famous examples of rings ofintegers are the rings of
integers associated to quadratic extensions Q Q( d) = K with d a square
free integer, d 6= 1. The integral elements of K are:
(
Z[ d], if d 2, 3(mod4)
A= .
Z[ 1+2 d ], if d 1(mod4)
For any two ideals I and J of A, not both 0, there exists the great-
est common divisor gcd(I, J) of I and J defined just like for integers by
min{1 ,1 } min{k ,k }
gcd(I, J) = p1 . . .pk if I = p1 1 . . .pk k and J = p1 1 . . .pk k
with p1 , . . . , pk distinct prime ideals of A and i , i N for all i = 1, k. Since
I I+J and J I+J, I+J|I and I+J|J, hence I+J|gcd(I, J). Conversely
gcd(I, J)|I and gcd(I, J)|J imply gcd(I, J)|I + J, so gcd(I, J) = I + J. In
particular, I and J are coprime if and only if I + J = A. We see that the
ideals of the ring of integers of a number field have similar properties to the
integers in Z.
Theorem 8.1.3 (Dirichlet). C is a finite abelian group.

8.1. ALGEBRAIC NUMBER THEORY PREREQUISITES 75
Proposition 8.1.4. A is factorial if and only if |C| = 1 if and only if A is

a principal ideal domain.
Theorem 8.1.5 (Dirichlet). Denote by U (A) the group of invertible ele-

ments of A. Then U (A) is a finitely generated abelian group.
By the Structure Theorem for Finitely Generated Abelian Groups, U (A) '
Zr W , where W is the torsion part of U (A). W is the set of roots of unity
of K and it can be shown that it is cyclic. We can also determine r. There
exist exactly n distinct field embeddings i : K , C, i = 1, n. Of these
field morphisms, some are real (the image is a subfield of R), and the rest
can be coupled in pairwise complex conjugate homomorphisms. Let s be
the number of real embeddings, and 2t the number of remaining complex
homomorphisms. Then
s + 2t = n
.
s+t1=r
Lets evaluate
the strength of this result by applying it to Pells equation.
Let K = Q( d) with d a square free integer, d 6= 1. Then [K : Q] = 2 and
the two field embeddings of K in C are completely characterized by

1( d) = d
.
2 ( d) = d.
If d < 0, then there are no real embeddings, hence s = 0 t = 1 r =

0 U (A) = W . It can be proven:
{1, i}, if d = 1

W = 1 3
{1, 2 , if d = 3 .
1,

elsewhere
If d > 0, then both 1 and 2 are real, therefore s = 2 t = 0 r = 1

implying that U (A) is of rank 1. Since the only roots of unity in K are 1,
we find
U (A) = {n | U (A), 6= 1},
for some A.
8.2 Completing the proof of Mordell-Weils The-

orem
We are now ready to tackle the complete proof of Mordell-Weils Theorem.
We have seen, for example in the proof of the existence of a Weierstrass
normal form, that any elliptic curve is projectively equivalent to a an elliptic
curve C(Q) given by an equation of the form y 2 = x3 + ax + b = f (x) with
a, b Z and f = 4a3 27b2 6= 0.
Theorem 8.2.1 (Mordell-Weil). Let C(Q) be the elliptic curve given by

y 2 = f (x) = x3 + ax + b with a, b Z and f = 4a3 27b2 6= 0.
Then the abelian group C(Q) is finitely generated.
We have seen that Mordell-Weils Theorem is equivalent to the weak

Mordell-Weil Theorem:
Theorem 8.2.2. |C(Q) : 2C(Q)| < .
We have proved this theorem elementarily, but only in the particular

case C(Q) had at least a point of order 2.
Let 1 , 2 and 3 be the complex roots of f . Because f 6= 0, all s are
distinct. Let
U = U (Q(1 ) Q(2 ) Q(3 )),
where U (A) denotes the multiplicative group of units of a ring A.
Define : C(Q) U/U 2 by

1, if P = O

0 \
(f (1 ), 1 2 , 1 3 ), if P = (1 , 0)

(P ) = \0
(2 1 , f (2 ), 2 3 ), if P = (2 , 0) .

\ 0
(3 1 , 3 2 , f (3 )), if P = (3 , 0)

\
( 1 , 2 , 3 ), if P = (, ), 6= 0
If P = (, ) C(Q) and 6= 0, then (1 )(2 )(3 ) = f () = 2 6=

0 implies that i are nonzero elements of Q(i ) for all i = 1, 3 and (P )
is a well defined element of U/U 2 . If P = (1 , 0) C(Q), then 1 Q and
1 2 and 1 3 are well defined elements of Q(2 ) and Q(3 ) respectively.
They are nonzero because the s are all distinct. Q(1 ) 3 f 0 (1 ) 6= 0 for
otherwise 1 would be a multiple root of f contradicting f 6= 0. We have
proved that (1 , 0) is a well defined element of U/U 2 . The same holds for
1 and 2 . These prove that is well defined.
Lemma 8.2.3. : C(Q) U/U 2 is a group homomorphism.
Proof: (P ) = (P ) for all P C(Q) because x(P ) = x(P ) for

all P 6= O and O = O. (P + Q) = (P )(Q) (P )(Q)(P Q) =
8.2. COMPLETING THE PROOF OF MORDELL-WEILS THEOREM77
(P + Q)(P Q) = (P + Q)2 = 1. So, to prove that is a group

homomorphism, it is enough to prove that if A, B and C are collinear
points of C(Q), then
(A)(B)(C) = 1.
If C = O, then (A)(B)(C) = (A)(A)(O) = (A)2 = 1. Assume
now that all of A, B and C are different from O. Let x(A) = x1 , x(B) = x2
and x(C) = x3 . Let y = x + be the equation of the line passing through
A, B and C. The following identity holds:
f (x) (x + )2 = (x x1 )(x x2 )(x x3 ).
If none of A, B and C is a point of order 2, then (x1 1 )(x2 1 )(x3

1 ) = f (1 )+(1 +)2 = (1 +)2 . The line y = x+ cuts C(Q) in A, B
and C only, so the assumption that A, B and C are not of order 2 guarantees
1 + 6= 0. Similarly we prove (x1 2 )(x2 2 )(x3 2 ) = (2 + )2 6= 0
and (x1 3 )(x2 3 )(x3 3 ) = (3 + )2 6= 0. Now we can say
(A)(B)(C) = ((1 + )2 , (\ 2 2
2 + ) , (3 + ) ) = 1.
If A = (1 , 0), then y = x + y = (x 1 ) and f (x) 2 (x 1 )2 =

(x 1 )(x x2 )(x x3 ). From the last, f 0 (x) 22 (x 1 ) = (x x2 )(x
x3 ) + (x 1 )(x x3 ) + (x 1 )(x x2 ) f 0 (1 ) = (1 x2 )(1 x3 ).
Therefore
(A)(B)(C) = (f 0 (1 )(x2 1 )(x3 \

1 ), (2 + )2 , (3 + )2 ) =
(f 0 (1 )2 , (2\
+ )2 , (3 + )2 ) = 1.
Similarly we treat the cases A = (2 , 0) and A = (3 , 0). These prove that
is a group homomorphism.
C(Q)
Lemma 8.2.4. ker = 2C(Q) and as a simple consequence 2C(Q) ' Im

C(Q)
and 2C(Q) = |Im|.
Proof: Since (2P ) = (P )2 = 1 for all P C(Q) we have 2C(Q)

ker . We prove the reverse inclusion.
We first treat the case f is irreducible over Q. Then we have the isomor-
phisms
Q[X]
i : Q(i ) i = 1, 3
f Q[X]
uniquely determined by i (x) = i , where x denotes the class of X in
Q[X]/f Q[X]. Let ji : Q(j ) Q(i ) be the isomorphism defined by
ji = i 1j for all i, j = 1, 3. Note that ji (j ) = i and ji |Q = 1Q
for all i, j = 1, 3.
U Q(i )
It is obvious that ker ker i , where i : U2
(Q( 2
i) )
denotes the
U Q3 Q(i )
canonical projection via the isomorphism U2
' i=1 (Q(i ) )2 . The reverse
inclusion now follows from i = ji j .
So it suffices to prove that ker 1 = ker 1
1 1 2C(Q), where i :
U (Q[X]/f Q[X]) Q(i )
U (Q[X]/f Q[X])2
(Q(i ) )2 is the isomorphism obtained canonically from i .
1
Denote = 1 1 . Let P ker . If P = O, then P = 2O 2C(Q). Let
P = (, ) with , Q. By the assumption that f is irreducible over Q,
j 6 Q j = 1, 3, hence C(Q) does not have points of order 2. We want to
find Q C(Q) such that 2Q = P .
We have 1 = (P ) = 1 1 \
1 1 (P ) = 1 1 (( 1 , 2 , 3 )) =
1 \
1 ( 1 ) = \ X ^ X = (1 X 2 +^ 2 X + 3 )2 for some 1 , 2 and
3 in Q.
( X 2 +
1
^ X + )( ^
2 3 X ) = (2 X 3 2 X^
1 2 1 + X )=
2 1 3 2 3
((12 22 3 )X (12 + 1 3 )) = e1^

+ 1^ X + f1 for appropriate e1 and f1
in Q. Then (e1 X + f1 ) = ( X) (1 X 2 )2 .
^ 2 ^ ^
If 1 = 0, then ^ X = (2 X ^ + 3 )2 2 = 0 ^ X = 32 , which
is a contradiction. So 1 6= 0 and for suitable e2 , f2 and h in Q, we have
^
(e2 X + f2 )2 = (^ X) (X^ h)2 f |(e2 X + f2 )2 ( X)(X h)2 .
Since (e2 X + f2 ) ( X)(X h)2 is a monic third degree polynomial,
2
f (x) = (e2 X+f2 )2 (X)(Xh)2 f (x)(e2 X+f2 )2 = (X)(Xh)2 .

The last equality shows that the line y = e2 x + f2 cuts C(Q) one point with
the x coordinate and in two points with the x coordinate h. Let Q 3 k =
e2 h+f2 . Then f (h) = k 2 , so Q(h, k) C(Q). If {Q, Q} {y 2 = e2 x+f2 },
then the line y 2 = e2 x + f2 cuts C(Q) in Q, Q and in Q (Q) = O, hence
O = P which contradicts the assumption P 6= O. So y 2 = e2 x + f2 cuts
C(Q) in P or P and twice in Q or Q. So (Q)(Q) = P 2(Q) =
P P 2C(Q).
Lemma 8.2.5. Let {1 , 2 , 3 }. Let A be the ring of integers of Q().
Let f (x) = (x )g(x), g A[X] and deg(g) = 2. For C(Q) 3 P (, ) 6= O
such that = ab , a, b Z, b > 0 and gcd(a, b) = 1, let
a
I(P ) = (a b)A + b2 g( )A.
b
Then I(P ) ranges through a finite number of ideals of A as P varies in
C(Q) \ {O}.
Proof: Notice that since g is a degree 2 polynomial with coefficients in
A, b2 g( ab ) A, hence I(P ) is a well defined ideal of A. g() 6= 0 because f
has no multiple roots.
Let g(x) g() = (x ) h(x), deg(h) = 1, h A[X]. Then
a a a a a
g( ) g() = ( )h( ) b2 g( ) b2 g() = (a b) (bh( ))
b b b b b
8.2. COMPLETING THE PROOF OF MORDELL-WEILS THEOREM79
b2 g() I(P ). We have used that since h is a degree 1 polynomial in A[X],

b h( ab ) A.
Let 2 g(x) x2 g() = (x )h1 (x) with deg(h1 ) = 1 and h1 A[X].
Then
a a2 a a a a
2 g( ) 2 g() = ( )h1 ( ) b2 2 g( )a2 g() = (ab)(bh1 ( ))
b b b b b b
a2 g() I(P ).
a2 g() I(P ) and b2 g() I(P ) imply g() I(P ). So I(P )|g()A
implying that I(P ) is one of the divisors of g()A. Since g()A is a nonzero
ideal, it has only a finite number of divisors, leaving only a finite number of
possibilities for I(P ).
Lemma 8.2.6. If P C(Q) is such that 2P 6= O, then there exists an ideal
D of A such that
(a b)A = I(P ) D2 ,
where a, b, I(P ) are defined as in 8.2.5.
Proof: I(P ) = (a b)A + b2 g( ab )A = gcd((a b)A, b2 g( ab )A). There
exist ideals B and C of A such that (ab)A = B I(P ), b2 g( ab )A = C I(P )
and gcd(B, C) = A B + C = A.
f (x) = (x )g(x) f ( ab ) = ( ab )g( ab ) f ( ab ) b3 A = ((a b)A)
(b g( ab )A) = (B I(P )) (C I(P )). f () = 2 2 b3 A = BC I(P )2 . We
2
have seen in 2.2.5 that b = e2 for some e Z. So ( e3 )2 A = BC I(P )2 .

Since B and C are coprime, they must be squares, hence there exists an
ideal D of A such that (a b)A = I(P ) D2 .
Lemma 8.2.7. There exists a finite number of algebraic integers for which
there exist u and such that a b = u 2 with u U (A) and K =
Q(), as P (, ) varies in C(Q) \ {O} with = ab , a, b Z, b > 0 and
gcd(a, b) = 1.
Proof: Let C1 , . . . , Cs be a complete system of representatives for the
ideal class group of A. Note that by 8.1.3 this group is finite. This and
8.2.5 imply that the set of ideals {Ci2 I(P )|i = 1, s, P C(Q) \ {O}}
is finite. For each principal ideal in this set, choose a generator. We
obtain a finite set of these s. From 8.2.6, there exists an ideal D such
that (a b)A = I(P ) D2 1 = I(P [) D2 in the ideal class group
of A, C(A). Then I(P [) = C 2 for some i = 1, s such that D = Ci , so
i
I(P ) Ci2 = A. Since D = Ci , there exists K such that Ci = D 1 .
Clearly I(P ) D2 2 = A (a b)A = 2 A (a b) = u 2
for some u U (A).

C(Q)
Proof of 8.2.1: By 8.2.4, 2C(Q) ' Im, hence it suffices to prove that
Im is a finite set. Since there are at most 4 points P C(Q) such that
2P = O, we can only investigate (P ) for 2P 6= O and P = (, ) with

= ab , a, b Z, b > 1 and gcd(a, b) = 1.
(P ) = ( 1 , \
2 , 3 ).
From 2.2.5, there exists e N such that b = e2 .
a\ a\ b1
1 = = a\ b1 = \ u 2 = d u,
b e2
for some u U (A), 0 6= K = Q(1 ) and for some belonging to the
finite set in 8.2.7.
By 8.1.5, U (A) is a finitely generated abelian group, hence there exists
r N such that U (A) is generated by u1 , . . . , ur . It is easy to see that every
element of U (A)/U (A)2 is of the form u11 \ . . . urr with i {0, 1} for
2 r
i = 1, r. So |U (A)/U (A) | 2 . Manifestly, there is only a finite number of
possibilities for ab\ 1 . We treat the second and third coordinates similarly
and keeping in mind that
U (Q(1 ) Q(2 ) Q(3 )) Q(1 ) Q(2 ) Q(3 )

' ,
U (Q(1 ) Q(2 ) Q(3 ))2 (Q(1 ) )2 (Q(2 ) )2 (Q(3 ) )2
we conclude that Im is finite.
8.3 C17
We return to 7.2.12 and prove that even though the rank of C17 (Q) : y 2 =
x3 + 17x is 0, one of the equations that we will stumble upon, namely
x2 + 4y 4 = 17e4 , has nontrivial solutions modulo any positive integer m > 1,
but has no nontrivial solution in integers. To simplify the notation, we
will use C(Q) instead of C17 (Q). The procedure is standard. We consider
the associated curve C(Q) : y 2 = x3 68x and the group homomorphisms
: C(Q) Q, 0 : C(Q) Q, defined by

x, if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
(P ) =
17, if P = T ,
1, if P = O

x,
b if P (x, y) 6= O[0 : 1 : 0], T (0, 0)
0
(P ) = 17,
d if P = T ,

1, if P = O
0
where Q = (QQ )2 . It was proved that 2r = |Im||Im
4
|
, where r is the rank of
C(Q). We have proved that for curves of type Cp (Q) with p a prime integer,
|Im| = 2, so all that is left to prove is that |Im0 | = 2.
8.3. C17 81
We have proved that x Im0 if and only if there exist b1 , b2 , M, N, e Z

such that
N 2 = b1 M 4 + b2 e4 ,
b1 = x, b1 b2 = 68 and 1 = gcd(M, e) = gcd(M, N ) = gcd(e, N ) =
gcd(b2 , M ) = gcd(b1 , e). We have seen that a solution to the equation above
2 b MN
corresponds to the point P = ( b1eM2 ,
1
e3
) (P = O if e = 0) on C(Q) such
0
that (P ) = b1 = x.
It is known that 1, 17d Im0 . So we have to prove that for b1
{1, 2, 4, 17, 34, 68}, the equation above has no solutions.
If b1 = 1, we have the equation N 2 = M 4 + 68e4 with some addi-
tional conditions. Lets assume that the equation has a nontrivial solution
(N, M, e) 6= (0, 0, 0). We can assume that gcd(M, N, e) = 1. Reducing mod-
ulo 4 we have N 2 = M 4 (mod4) and since the quadratic residues modulo
4 are 0 and 1, N and M are even numbers. gcd(N, M, e) = 1 implies that e
is odd.
Let N = 2x and M = 2y. The equation is x2 = 4y 4 +17e4 x2 +4y 4 =
17e . If p is a prime number such that p|gcd(x, y), then p|17e4 , hence p|e
4
or p|17. p|e contradicts gcd(2x, 2y, e) = 1, therefore p|17 p = 17. Then

x 2 e 4
+ 4 173 17 = e4 17|e which also contradicts gcd(2x, 2y, e) = 1.

17 17
These prove that gcd(x, y) = 1. Similarly gcd(e, y) = 1. Since e is odd, x is
also odd.
We solve the equation
x2 + 4y 4 = 17e4 , gcd(x, y) = gcd(e, y) = 1, with x and e odd integers,
at the help of Z[] which is an euclidian domain.

(x + 2y 2 )(x 2y 2 ) = 17e4 .
Let d = gcd(x+2y 2 , x2y 2 ). Then d|2x and d|4y 2 , hence d|gcd(4x, 4y 2 ).
Since x and y are coprime in Z, they are also coprime in Z[], so gcd(x, y 2 ) =
1. It follows that d|4. The prime factor decomposition of 4 in Z[] is
4 = (1 + i)4 . If 1 + |d, then 1 + |x2 + 4y 4 = 17e4 which is impossi-
ble since e is odd. Therefore d = 1.
17|x + 2y 2 if and only if x + 2y 2 = 17 z for some z Z[] if and only
if x 2y 2 = 17 z, where z is the complex conjugate of z, if and only if
17|x 2y 2 . Hence 17|x + 2y 2 17|x 2y 2 . In this case, 17|2x and 17|4y 2
which imply 17|x and 17|y. This is a contradiction. Therefore 17 6 |x 2y 2 .
The prime factor decomposition of 17 in Z[] is 17 = (4+)(4) and the
two primes appearing in the decomposition are not associated in divisibility
i.e. they do not generate the same ideal of Z[]. It is easy to obtain x+2y 2 =
(4 )4 or x + 2y 2 = (4 )4 for some = a + b Z[]. ( )4 =
e4 (a2 + b2 )4 = e4 . a and b must have different parities otherwise e would
be even. A simple computation yields 4 = (a4 6a2 b2 + b4 ) + 4ab(a2 b2 ).
x + 2y 2 = (4 )4 x = 4(a4 6a2 b2 + b4 ) 4ab(a2 b2 ) 2|x

which is impossible since x is odd.

x + 2y 2 = (4 )4 2y 2 = 4(a4 6a2 b2 + b4 ) 4ab(a2 b2 )
2|y 2| (a4 + b4 6a2 b2 ) ab(a2 b2 ).
Since a and b have different parities, a4 + b4 6a2 b2 is odd and ab(a2 b2 )
is even thus contradicting the previous divisibility.
Remark 8.3.1. We have proved that the only solution in Z of the equation
x2 + 4y 4 = 17z 4 is (x, y, z) = (0, 0, 0).
We postpone the proof of r = 0 to prove:
Proposition 8.3.2. For all m Z, m > 1, there exist x, y, z Z such that
x2 + 4y 4 17e4 (mod m) and (x, y, e) 6= (0, 0, 0) in Zm .
Proof: The Chinese Remainder Theorem allows us to reduce to the case
m = p for some prime number p and integer > 0. The main tool is the
following lemma:
Lemma 8.3.3. Let f Z[X1 , . . . , Xn ] be a polynomial with integer coeffi-
cients and let p be a prime number. A modulo p zero (x1 , . . . , xn ) of f is
called simple if f (x1 , . . . , xn ) 0(mod p) and one of the partial derivates
f
Xi (x1 , . . . , xn ) is nonzero mod p.
Then for any simple mod p zero (x1 , . . . , xn ) of f and any > 0 there
() () () ()
exists (x1 , . . . , xn ) (Z)n such that f (x1 , . . . , xn ) 0(mod p ).
Assume first that = 1. Then m = p. If p = 2, then take (x, y, e) =
(1, 0, 1). If p = 17, take (x, y, e) = (8, 1, 1). If p 1(mod4), then there exists
z Z such that p|z 2 + 1. Take (x, y, e) = (2z, 1, 0). If p 3(mod4), then
we prove that any square in Zp is also a fourth power in Zp . We know that
if z is not divisible by p then it is a square mod p if and only if z is not
a square mod p. Let t be a square mod p not divisible by p. Then there
exists z Z such that t z 2 (mod p). Then p 6 |z, so z or z is a square
modulo p leading to t is a fourth power in Zp . Take (x, y, e) = (8, 1, z) such
that z 4 22 (mod p).
It is easy to see that if p 6= 2, then the solutions given above are simple
mod p zeros of f (x, y, e) = x2 + 4y 4 17e4 and the proposition follows from
8.3.3.
There is no simple mod 2 zero of f since all the partials are obviously
divisible by 2, so we cannot apply 8.3.3 in this case, but we prove by induc-
tion that we can construct xn Z such that x2n + 4 04 17 14 0(mod 2n )
for any n > 0. Taking (x, y, e) = (xn , 0, 1) would then complete the proof.
Notice that we can take x1 = x2 = x3 = x4 = 1. Assume we have
constructed x2n 17 = 2n yn for some yn Z and n 4. We try to find
t Z such that 2n+1 |(xn + 2n1 t)2 17.
(xn + 2n1 t)2 17 = x2n + 2n xn t + 22n2 t2 17 = 2n (yn + xn t) + 22n2 t2 .
8.3. C17 83
Since n 3, 2n2 n+1, so 2n+1 |(xn +2n1 t)2 17 if and only if 2|yn +xn t.
Such t exists because 2n |x2n 17 xn is odd. Take xn+1 = xn + 2n1 t.
Alternatively we could have used the following lemma for f (x) = x2 17,
n = 3 and k = 1:
Lemma 8.3.4. Let f Z[X] with f 0 its derivate and let p be a prime
number. Let x Z, n, k Z such that 0 2k < n, pn |f (x) and vp (f 0 (x)) =
k i.e. pk |f 0 (x) and pk+1 6 |f 0 (x).
Then there exists y Z of the form x + pnk z such that pn+1 |f (y) and
vp (f 0 (y)) = k.
Lets see what we have worked so hard for. We have proved that the
equation x2 + 4y 4 17z 4 = 0 has no nontrivial solution in Z even though it
has nontrivial solutions in Zm for any m > 1. This means that generally we
cannot hope to have an algorithm, based on reductions by various integers,
that would help us prove that an equation of the form N 2 = b1 M 4 +aM 2 e2 +
b2 e4 has no nontrivial integer solutions. The reason why such an algorithm
was expected until the discovery of a counterexample like the one above was
that such an algorithm exists for quadratic forms:
Theorem 8.3.5 (Minkowski-Hasse). The equation with nonzero integer
coefficients a1 x21 + . . . + an x2n = 0 has nontrivial solutions ((x1 , . . . , xn ) 6=
(0, . . . , 0)) in Z if and only if it has nontrivial solutions in R and in Zm for
any m > 1.
We should now return to C17 (Q). We have proved that the equations
N 2 = M 4 + 68e4 and N 2 = 4M 4 + 17e4 have no nontrivial integer solu-
c 6 Im0 .
tions. This means that 1
If b1 = 2, the equation is N 2 = 2M 4 + 34e4 with 1 = gcd(M, e) =

gcd(M, N ) = gcd(e, N ) = gcd(34, M ) = gcd(2, e). N must be even, so
there exists n Z such that N = 2n and M 4 + 2n2 = 17e4 . M and e have
the same parity hence they are both odd since gcd(M, e) = 1. Reducing
mod 16 we obtain M 4 + 2n2 e4 (mod 16). But 4 {0, 1, 4} mod 16 for
all Z,and since M and e are odd, we obtain 16|2n2 4|n.
2 2
In Z[ 2], 4which is an euclidian domain, we have (M + 2 n)(M
2 n) = 17e .
Let d= gcd(M 2+ 2n, M 2 2n). Then d|2M 2 and d|2 2n imply

2 , 2 2 n) = 2 2. The prime factor decomposition of 2 2
32 M
d|gcd(2
is ( 2) . If 2|d, then d2 |(M 2 + 2 n)(M 2 2 n) = 17e4 2|17e4
contradicting that e is odd. Therefore d = 1.
17|M 2 + 2 n if and only if there exists z Z[ 2] such that M 2 +

2 n = 17z M 2 2 n = 17z, wherez is the complex conjugate
of z,
if and only if 17|M 2 2 n. So 17|M 2 + 2 n 17|M 2 2 n. In this
case we easily find that 17|M and 17|n which contradict gcd(M, N ) = 1.

The prime factor decomposition
of 17
is (3 + 2 2)(3 2 2) and it is
easy to seethat M 2 + 2n = (32 2)(a+b 2)4 for some a, b Z such
that (a + 2 b)4 (a 2 b)4 = e4 . Since e isodd, it follows that a is also
odd. It is an easy 4 = (a4 12a2 b2 + 4b2 ) +
computation to verify (a4+ b 2)
2 2
4ab(a 2b ) 2. It follows that n = 2(a 12a b2 + 4b4 ) 12ab(a2 2b2 ).
2
Modulo 4 we have n 2a4 (mod4) which is impossible because a is odd and

4|n. We have proved that 2 c 6 Im0 .
If b1 = 2, then the equation is N 2 = 2M 4 34e4 and we have the addi-

tional conditions: 1 = gcd(M, e) = gcd(M, N ) = gcd(e, N ) = gcd(34, M ) =
gcd(2, e). gcd(2, e) = gcd(34, M ) = 1 implies that M and e are odd.
N 2 = 2M 4 34e4 N = 2n for some n Z and 2n2 = M 4 17e4 . Re-
ducing modulo 16 we get 2n2 1 17(mod 16) 16|2n2 4|n n = 4m
for some m Z. The equation is 32m2 = M 4 17e4 and the restrictions
are that M and e be odd and gcd(M, e) = 1. It is not hard to prove that
gcd(M, 8m) = gcd(e, 8m) = gcd(34, M ) = gcd(2, e) = gcd(m, 17) = 1.
1+ 17
We consider Z[ 2 ] which as we will prove is a principal ideal domain.

a+ 17b
Its elements are of the form with a, b integers of the same parity.
2
2 + 17e2

M M 2 17e2
32m2 = M 4 17e4 2 2 = 8m2 .
2
2 2 2

Let d = gcd( M +2 17e , M 2 17e ) in Z[ 1+2 17 ]. Then d|M 2 and d| 17

e2 , hence d|gcd( 17 M 2 , 17 e2 ) = 17. 17 is a prime element of

1+ 17
2 M 2 + 17e2 M 2 17e2
Z[ 2 ]. If d = 17, then 17 = d | 2 2 = 8m2 which is
easily seen to be false. Therefore d = 1.
There is the prime factor decomposition 2 = 3+2 17 3+2 17 and the two
primes that appear in the decomposition are not associated in divisibility
i.e. they do not generate the same ideal of Z[ 1+2 17 ].
2

17e2
2 6 |M 2 because M and e are odd integers.

The elements of the group of units U (Z[ 1+2 17 ]) are {(4+ 17)n |n Z}.
We must have
!2
M 2 + 17 e4 n 3 + 17 a + 17 b
= (4 + 17)
2 2 2
such that a and b have the same parity and (a2 17b2 )2 =
for some a, b, n Z
64m . Since M + 17 e2 > 0, we have
2 2
!2
a + 17 b
M 2 + 17 e2 = (4 + 17)n (3 + 17) .
2

Let (4 + 17)r = Ar + 17 Br with Ar , Br Z uniquely defined by the
recurrence:
Ar+1 = 4Ar + 17Br
, r Z
Br+1 = Ar + 4Br
8.3. C17 85
with initial terms A0 = 1 and B0 = 0. Note the recurrence goes both ways.
Ar+1 and Br+1 are obviously uniquely defined by Ar and Br . Conversely,
Ar and Br are uniquely defined by Ar+1 and Br+1 as a consequence of
4 17
det = 1.
1 4
The equation can be rewritten as:

4 (M 2 + 17 e2 ) = (An + 17 Bn ) (3 + 17) ( + 17 )

for = a2 + 17b2 and = 2ab. So 4(M 2 + 17e2 ) = ((3An + 17Bn ) +
(An 3Bn ) 17) ( + 17 ). We
have the same everywhere it appears
in the preceding equality. Since 17 is not a rational number, we find the
system:
4M 2 = (3An + 17Bn ) + 17(An 3Bn )

4e2 = (3An + 17Bn ) + (An 3Bn )

with the same convention on the sign . By adding the two equations
reduced modulo 16 we have
4(M 2 + e2 ) ( + )(3An + Bn + An 3Bn )(mod 16).
If the sign is + , then the equation mod 16 above is equivalent to
M 2 + e2 ( + )(An + Bn )(mod 4). It is easy to prove by induction that
Ar + Br Ar+1 + Br+1 A0 + B0 1(mod 4) for all r Z. + =
a2 + 17b2 + 2ab a2 + b2 + 2ab = (a + b)2 (mod 4). Since a and b have
the same parity, + 0(mod 4), so 4|M 2 + e2 . But this is not possible
because M and e are both odd, hence M 2 + e2 2(mod 4).

If the sign is , then 0 < M 2 + 17 e2 = (4 + 17)n (3 + 17)
2
a+ 17b
2
!2
a 17 b
M2 17 e2 = (4 17)n (3 17) .
2

Since (M 2 + 17e2 )(M 2 17e2 ) = 32m2 > 0, it follows that M 2 17
2
e2 > 0. Then M 2 17 e2 = (4 17)n (3 17) a 217b > 0
n is odd.
4M 2 = (3An +17Bn )+17(An 3Bn ) 4M 2 (3An +Bn )(a2 +
b ) + (An 3Bn )(2ab) (3An + Bn )(a + b)2 + 2ab(4An 4Bn )(mod 16)
2
2
1 M 2 (3An + Bn ) a+b 2 + 2ab(An Bn )(mod 4). Note that a+b2 is
an integer because a and b have the same parity.
We have
Ar+1 Br (mod 4)
, r Z.
Br+1 Ar (mod 4)
It is easy to see that since n is odd, An 0(mod 4) and Bn 1(mod 4).
2
Then 1 a+b 2 2ab(mod 4). If a and b are both odd, then 2ab 2(mod 4),
2
so a+b
2 3(mod 4) which is impossible. Therefore a and b are both even.
Then there exist u, v Z such that a = 2u, b = 2v and (u2 17v 2 )2 = 4m2 .
This means that u and v have the same parity. Just like before, we get the
system:
M 2 = (3An + 17Bn ) (u2 + 17v 2 ) + 17(An 3Bn ) (2uv)

.
e2 = (3An + 17Bn ) (2uv) + (An 3Bn ) (u2 + 17v 2 )
By adding the equations modulo 8 we have
M 2 + e2 2(An + Bn )(u + v)2 0(mod 8).
We have used that u + v is even because u and v have the same parity, hence
4|(u + v)2 . But 8|M 2 + e2 implies that M and e are even which is false.
We have finished proving that 2 6 Im0 .
We know that Im0 is an abelian group containing 1 and 17. d Also
0 0
1, 2 6 Im . Using these and the group structure of Im , it is easy to
c c
d 6 Im0 . Hence |Im0 | = 2 and the rank r of C17 (Q) is 0.
b 34
prove that 17,

Lemma 8.3.6. Z[ 1+2 17 ] is a principal ideal domain.
Towards the proof of the lemma we need the following:
Theorem 8.3.7 (Hasse-Dedekind). Let (A, +, ) be a subring of C. Let

: A N be a function such that:
1. () = 0 = 0;
2. x, y A such that y 6= 0 and y 6 |x there exist u, v A such that

0 < (xu + yv) < (y).
Then A is a principal ideal domain.
Proof: Let I be a nonzero ideal of A. There exists a I such that

0 < (a) (x) for all nonzero elements x I. We will prove that I = aA.
Let x I, x 6= 0. Suppose a 6 |x. Then there exist u, v A such that
0 < (au + xv) < (a). Since a, x I, au + xv is also an element of I.
The condition 0 < (au + xv) implies that au + xv 6= 0. The inequality
(au + xv) < (a) contradicts the choice of a. Therefore a|x for all nonzero
x I. This implies I a A. The reverse inclusion is obvious.
Proof of 8.3.6: To prove this lemma

we
2 use 2the
Hasse-Dedekind
1+ 17 a+b 17 a 17b
Theorem for A = Z[ 2 ] and ( 2 ) = 4 for all a, b Z of
the same parity.
We prove that satisfies the conditions in Hasse-Dedekinds Theorem.
If a and b have the same parity, then a2 b2 17b2 (mod4) 4|a2 17b2 .
8.3. C17 87
Since obviously (x) 0 for all x A, we have proved that (x) N for
all x A. can be extended on Q( 17) by (a + b 17) = |a2 17b2 |.

is multiplicative
on Q( 17) and on A.
2 17b2 | = 0 a2 = 17b2 . It is easy to prove
+ b 17) = 0 2 |a
(a
2
using 17 6 Q that a = 17b a = b = 0. Therefore (x) = 0 x = 0.
Let x, y A such that y 6= 0 and y 6
|x. It is clear that 0 < (xu + yv) <
(y) 0 < ( xy u + v) < 1. xy Q( 17) a, b, c Z, c N such

that gcd(a, b, c) = 1 and xy = a+bc 17 . Because gcd(a, b, c) = 1, there exist
d, e, f Z such that ad + be + cf = 1. Let

u = e + d 17 A.

Then xy u = (a+b 17)(e+d
c
17)
= ae+17db+(ad+be)
c
17
. Let q, r Z be defined
by ae + 17bd = cq + r with 2 r 2c , and let
c

v = q + f 17 A.

x ae+17dbcq+(ad+be+cf ) 17 r+ 17
Then y u+v = c = .
2 c 2
For c 5, we have 0 < ( xy u + v) = r c17 rc
+ 17 14 + 17
25 < 1.

2 c2
If c = 1, then we contradict y 6 |x.
If c = 2, then a and b must have different parities otherwise xy A which

contradicts y 6 |x. Let b 17 and v = q such that a2 17b2 = 2q + 1.
u = a
Then xy u + v = (a+b 17)(ab2
17)2q
= 12 0 < ( xy u + v) = 14 < 1.
If c = 3, then let u = a b 17 and v = q such that a2 17b2 = 3q + r
and r {0, 1, 2}. If r = 0, then 3|a2 17b2 3|a2 + b2 3|gcd(a, b) which
2 2 3q
contradicts gcd(a, b, 3) = 1. Hence r 6= 0. Then xy u + v = a 17b 3 =
r x
3 0 < ( y u + v) < 1.

Assume c = 4. If a and b have different parities, let u = a b 17.
a2 17b2 is an odd number, so there exist q, r Z such that a2 17b2 = 4q+r
and r {1, 3}. Let v = q. Then xy u + v = 4r 0 < ( xy u + v) < 1.
If a and b have the same parity, then they must both be odd because
gcd(a, b, 4) = 1. We have the two cases two consider:

1. If a 3(mod4), then let u = 1 and a = 4k +3.

Let v = k l 17 with
l defined by b = 4l 1. Then xy u+v = 34 17 0 < xy u+v = 12 < 1.
2. If a 1(mod4), let u = 1, a = 4k + 3, b = 4l 1 and v =

k l 17. Then xy u + v = 34 17 0 < ( xy u + v) = 12 < 1.

Lemma 8.3.8. The group of units of A = Z[ 1+2 17 ] is

U (A) = {(4 + 17)n |n Z}.

Proof: It is very easy to see that (4 + 17)n is a unit for every integer
n.
Let now u = a+ 2 17b be a unit in A. a, b Z and have the same parity.
Since u is invertible, there exists v A such that uv = 1. Since is
2 2
multiplicative, we find (u) (v) = 1 (u) = (v) = 1 a 17b 4 = 1.
2
a 17b2

a+b 17 ab 17
Assume first 1 < u < 4 + 17. 1 = 4 = 2 2

ab 17
2 = a+b217 = u1 < 1 1 < b 17a < 1. Since 1 < a+b 17
<

2 2

4 + 17, we get 0 < b 17 < 5 + 17 0 < b < 1 + 5 1717 < 3. We have the
two cases:
1. If b = 1, then |a2 17| = 4 a2 {13, 21} which is not possible if a

is an integer.
2. If b = 2, then |a2 17b2 | = 4 |a2 68| = 4 a2 {64, 72}

2
a = 68 a = 8 u = 4 + 17. The condition 1 < u implies
u = 4 + 17 which contradicts the choice u < 4 + 17.

Assume now that u A is a unit such that 1 < u. 0 < 17 4 < 1 is a
unit. Since
lim u ( 17 4)n = 0,
n

n+1 < 1 u ( 17 4)n . Let
there exists
n N such that u ( 17 4)
v = u ( 17 4)2 . Then v is a unit and v ( 17 4) < 1 v 1 v <
4 + 17. Since we have nseen that there are no units of A in (1, 4 + 17),
v = 1, hence u = (4 + 17) .
Let now u be an arbitrary unit in A, u 6= 1. Then exactly one of the
units u1 of Ais greater than 1. Using the previous case, we conclude
that u = (4 + 17)n for some n Z.
Chapter 9
Lecture IX
9.1 Test Paper

1
Exercise 9.1.1. Compute the rank of the elliptic curve
C(Q) : y 2 = x3 82x.
Exercise 9.1.2. Prove that if x, y and z are integers such that x4 + 2y 2 =

17z 4 , then x = y = z = 0.
Exercise 9.1.3. Consider the rational cubic curve
C(Q) : y(y + x) = x(x 1)(x + 2).
Prove that (2, 2) and (0, 0) belong to C(Q).

Prove that C is smooth and that its unique point at infinity is O[0 : 1 : 0].
In 1.2.12 and 1.2.13, we gave C(Q) an abelian group structure with neutral
element O. For n N , let xn , yn be the rational numbers defined by
(xn , yn ) = n (2, 2) + (0, 0), where the operation + is the one in C(Q).
Compute xn for n 1, 5. It is known that there exists integers an , bn
and tn such that gcd(an , tn ) = gcd(bn , tn ) = 1, tn > 0 and xn = at2n , yn = bt3n .
n n
Prove that
tn tn+5 = tn+4 tn+1 + tn+3 tn+2 n 1.
The next page contains the solutions to these problems.
1
Working time 150 minutes. This test paper only counts as extra for the final exam.
89
90 CHAPTER 9. LECTURE IX
9.1.1 Solutions
Solution to 9.1.1: Let C(Q) : y 2 = x3 + 328x. Let r denote the rank of
the elliptic curve C(Q). We have proved that
|Im| |Im0 |
2r = ,
4

where : C(Q) Q = (QQ )2 and 0 : C(Q) Q are group homomor-
phisms. We have seen that x Im if and only if there exists b1 |82 such
that b1 = x and the equation
82 4
N 2 = b1 M 4 e
b1
has nontrivial solutions i.e. (N, M, e) 6= (0, 0, 0). Similarly, for 0 we have
the equations
328 4
N 2 = b1 M 4 + e .
b1
We first compute Im. The divisors of 82 are {1, 2, 41, 82}.
Therefore Im {1, c 2,
c 41,
d 82}.
d We prove that the previous inclusion
is in fact equality. To do so, we give nontrivial solutions to each of the equa-
tions for b1 {1, 2, 41, 82}. To make computations easier, notice that
if (N, M, e) is a nontrivial solution to the equation N 2 = b1 M 4 82 4
b1 e , then
(N, e, M ) is a nontrivial solution to N 2 = 82 4 82 4 4 82
b1 M 82 e = b1 e b1 M .
4
b1
82
This means that b1 Im
d
b1 Im. We have the solutions:
b1 N M e
1 1 0 1
2 11 1 3 .
41 3 2 1
82 1 1 3
Hence |Im| = 8.
We now compute Im0 . We have b1 < 0 328 b1 < 0. In this case,
2 4 328 4
N = b1 M + b1 e implies N = M = e = 0. Therefore the equations
N 2 = b1 M 4 + 328 4
b1 e give only trivial solutions for b1 < 0.
The positive divisors of 328 are {1, 2, 4, 8, 41, 82, 164, 328}, so Im0
82}.
{1, 2, 41, We prove that this last inclusion is in fact an equality. Just like

for , we need consider only half the cases, because of the pairing b1 , 328 b1 .
b1 b1 N M e
1 1 1 0 1 .
2 8 7 1 1
9.1. TEST PAPER 91
We have proved |Im0 | = 4.

2r = 84
4 = 8 r = 3.
Proof of 9.1.2: The problem was solved while proving that
rankC17 (Q) = 0,
back in Lecture VIII.
Proof of 9.1.3: This problem is the subject of the next lecture.

92 CHAPTER 9. LECTURE IX
Chapter 10
An unexpectedly hard
problem
In Lecture III, the following problem appeared as exercise 3.1.8:

Exercise 10.0.4. Let (tn )n1 be the sequence of rational numbers defined
by t1 = t2 = t3 = t4 = t5 = 1 and
tn+4 tn+1 + tn+3 tn+2
tn+5 =
tn
for every n 1.
Prove that all the terms of (tn )n1 are in fact integers.
This problem, presented by Don Zagier on the fifth day of the St. An-
drews Colloquium in 1996, has a deceivingly elementary text, but it knows
no elementary proof so far. The first hints towards a proof are given by the
third problem, 9.1.3 of the Test Paper in the previous lecture:
Exercise 10.0.5. Consider the rational cubic curve
C(Q) : y(y + x) = x(x 1)(x + 2).
Prove that P (2, 2) and T (0, 0) belong to C(Q).

Prove that C is smooth and that its unique point at infinity is O[0 : 1 : 0].
In 1.2.12 and 1.2.13, we gave C(Q) an abelian group structure with neutral
element O. For n N , let xn , yn be the rational numbers defined by
(xn , yn ) = n (2, 2) + (0, 0), where the operation + is the one in C(Q)
and n P = P
| + P + {z. . . + P} for all P C(Q).
n times
Compute xn for n 1, 5. Prove that there exists integers an , bn and tn
such that gcd(an , tn ) = gcd(bn , tn ) = 1, tn > 0 and xn = at2n , yn = bt3n .
n n
Prove that
tn tn+5 = tn+4 tn+1 + tn+3 tn+2 , n 1.
93
94 CHAPTER 10. AN UNEXPECTEDLY HARD PROBLEM
Proof: P (2, 2) and T (0, 0) belong to the cubic C. Let Pn (xn , yn ) =

n P + T.
The homogenized equation of C(Q) is F (x, y, z) = y(y + x)z x(x
z)(x + 2z) = 0. To prove that C is smooth we have the standard method
of computing the partial derivates of F and proving that at each point of
P2 at least one of them does not vanish. This method is quite tedious but
fortunately we have a more elegant proof. The rational projective transfor-
mation:

[x : y : z] [4x : 8y + 4x : z]
sends O to itself, T to itself, P (2, 2) = [2 : 2 : 1] to [8 : 24 : 1] and C to:
C 0 : y 2 z = x3 + 5x2 z 32xz 2 .
Clearly C is smooth if and only if C 0 is smooth. The affine equation of C 0

has the familiar form y 2 = f (x) = x3 + 5x2 32x. Therefore C 0 is smooth
if and only if C 0 (Q) is a smooth elliptic curve i.e. if and only if f 6= 0.
f = 32(52 + 4 32) 6= 0.
The points at infinity of C are given by F (x, y, 0) = 0 x3 = 0
x = 0. The only projective solution is O[0 : 1 : 0]. The rational projective
transformation induces an isomorphism between C(Q) and C 0 (Q).
To compute xn and yn for n = 1, 5, several methods can be used. We
can use the recurrence Pn+1 = Pn + P , or we can compute (Pn ) and then
apply 1 . The results will be:
n xn yn
1 1 2
2 2 0
3 14 58
4 29 22
27
49 259
5 25 125
The inverse transformation of is given by:
1 x yx
[x : y : z] [ : : z].
4 8
Since (Pn ) C 0 (Q), there exist a0n , b0n , t0n Z such that:
(Pn ) = [a0n t0n : b0n : (t0n )3 ], t0n > 0 and gcd(a0n , t0n ) = gcd(b0n , t0n ) = 1.
0 0
This corresponds to the affine situation (Pn ) = ( (ta0 n)2 , (tb0n)3 ) which follows
n n
from 2.2.5 if we prove that (Pn ) 6= O for all n 1. Assume (Pn ) = O
for some n 1. Then (n P + T ) = O n (8, 24) + (0, 0) = O 2n
(8, 24) + 2 (0, 0) = O in C 0 (Q). In C 0 (Q), we have 2T = 0, so 2n (P ) = O.
If that was so, then (P ) would be a torsion point of C 0 (Q). Since T is also
95
a torsion point and the torsion of an abelian group is a subgroup, we would

have that (Pm ) is a torsion element of C 0 (Q) for all m 1. By Nagell-Lutz,
a torsion point on an elliptic curve has integer coordinates, so (P m ) would
2 22
have integer coordinates for all m 1. But (P4 ) = 9 , 27 = ([6 :
22 : 27]) = [24 : 152 : 27] = 89 , 152 27 which clearly does not have integer
coordinates. Here we also have proved that P is not a torsion point of C(Q).
0 0 0 0 0
Pn = 1 ([a0n t0n : b0n : (t0n )3 ]) = [ an4tn : bn a8 n tn : (t0n )3 ] = [2a0n t0n :
b0n a0n t0n : (2t0n )3 ]. For all n 1, define the triplet (an , bn , tn ) by:
0 0 0 0 0 0
(
(an , bn , tn ) = n ,0 bn 0 a0n t0 n , 2tn ), if an is odd .
(a
an bn an tn 0
4 , 8 , tn , if a0n is even
We will prove that an , bn , tn are integers tn > 0 and gcd(an , tn ) = gcd(bn , tn ) =

1. Since t0n > 0 for all n 1, we get that tn > 1 for all n 1.
If a0n is odd, then clearly an , bn , tn are integers and gcd(a0n , 2t0n ) = 1.
Assume p is a prime number such that p|gcd(bn , tn ) p|b0n a0n t0n and
p|2t0n . Since (Pn ) C 0 (Q), we have (b0n )2 (t0n )3 = (a0n )3 (t0n )3 + 5(a0n )2 (t0n )5
32a0n (t0n )7 (b0n )2 = (a0n )3 + 5(a0n )2 (t0n )2 32a0n (t0n )4 . p|b0n a0n t0n
p|(a0n )3 + 4(a0n )2 (t0n ) 32a0n (t0n )4 . Together with p|2t0n this yields p|(a0n )3
which contradicts gcd(a0n , 2t0n ) = 1. Therefore gcd(b0n a0n t0n , 2t0n ) = 1.
If a0n is even then from (b0n )2 = (a0n )3 + 5(a0n )2 (t0n )2 32a0n (t0n )4 we get
that b0n is also even. We want to prove that an , bn , tn are integers and
gcd(an , tn ) = gcd(bn , tn ) = 1. Let a0n = 2a, b0n = 2b and t0n = t. The
condition gcd(a0n , t0n ) = 1 proves that t = t0n is odd. Then b2 = 2a3 + 5a2 t2
16at4 . To prove that an is an integer, we must prove that 4|a0n 2|a.
Assume a is odd. Modulo 8, we have b2 2a3 + 5a2 t2 . Since a and t
are odd, we get that b is odd, so 1 2a + 5(mod8) 2a 4(mod8)
a 2(mod4) which is impossible if a is odd. Therefore a is even and 4|a0n ,
so an is an integer. To prove that bn is an integer, we must prove that
8|b0n a0n t0n 4|b at. b2 = 2a3 + 5a2 t2 16at4 2|b. Let b = 2b and
a = 2a. Then b2 = 4a3 + 5a2 t2 8at4 4|b2 a2 t2 b and at have the
same parity. So 2|b at 8|b0n a0n t0n bn Z. Similarly to the previous
case we prove gcd(an , tn ) = gcd(bn , tn ) = 1.
We have proved that for all n 1 there exist integers an , bn and tn such
that tn > 0, gcd(an , tn ) = gcd(bn , tn ) = 1 and Pn (xn , yn ) = [an tn : bn : t3n ]
from which it follows that xn = at2n and yn = bt3n .
n n
The only thing left to prove is the recurrence tn+5 tn = tn+4 tn+1 +
tn+3 tn+2 . Lets see that this is the same as proving 10.0.4. The recurrence
is the same, but the initial terms slightly differ. If we look at Pn for n = 1, n,
we get t1 = t2 = 1, t3 = 2, t4 = 3 and t5 = 5. The first five terms in 10.0.4
are t1 = t2 = t3 = t4 = t5 = 1. However t6 = 2, t7 = 3 and t8 = 3. So if
we manage to solve 10.0.5 we also obtain the other sequence, but shifted 3
places to the right which changes nothing on the recurrence or on the terms
of the sequence being integers.
Consider the projective rational transformation given by

[x : y : z] [2z y 3x : 2z + y 2x : 2z x].

The corresponding affine map is (x, y) = 2y3x2x , 2+y2x
2x .
Let C1 : (xy + z 2 )(5z x y) = 6z 3 . We will prove that (C(Q)) =
C1 (Q). For this, let [x : y : z] C(Q), u = 2z y 3x, v = 2z + y 2x
and t = 2z x. We must prove that (uv + t2 )(5t u v) = 6t3 . uv =
4z 2 10xz (y + 3x)(y 2x). u + v = 4z 5x. (uv + t2 )(5t u v) =
(4z 2 10xz y 2 xy + 6x2 + 4z 2 4xz + x2 )(10z 5x 4z + 5x) =
(y 2 xy+7x2 14xz+8z 2 )6z = 6(y(y+x)z+7x2 z14xz 2 +8z 3 ). Now we
use [x : y : z] C(Q). y(y + x)z = x(x z)(x + 2z) (uv + t2 )(5t u v) =
6(x3 x2 z + 2xz 2 + 7x2 z 14xz 2 + 8z 3 ) = 6(x3 + 6x2 z 12xz 2 + 8z 3 ) =
6(2z x)3 = 6t3 .
1
The inverse of is given by [x : y : z] [2x
+ 2y 4z : 2x 4y + 2z :
1 x+y2 x2y+1
x + y 5z] and the affine map is (x, y) = 2 x+y5 , 2 x+y5 .
induces an isomorphism from C(Q) to C1 (Q). The neutral element of
C1 (Q) is (O) = [1 : 1 : 0] and not O like for C and C 0 .
tn tn+3
Lemma 10.0.6. Let vn = tn+1 tn+2 . Then (Pn+2 ) = (vn , vn+1 ) for every
n 1.
Assume we have proved it. Then notice that (u, v) C1 (Q) (v, u)
C1 (Q) because the affine equation of C1 , (xy+1)(5xy) = 6, is symmetric
in x and y. Actually we can prove that if (u, v) C1 (Q), then (u, v), the
inverse of (u, v) in C1 (Q) is (v, u). If uv 6= 0, then (uv + 1)(5 u v) = 6
v 2 + ( u1 + u 5)v + u+1
u = 0. By Vietes relations, the last equation has the
solution v for fixed u if and only if it has also the solution u+1 uv . The same
argument holds for proving that if u 6= 0, then the affine line x = u cuts
C1 (Q) in at most two points. So, for uv 6=0, we have v, v+1 uv C1 (Q)
(v, u) C1 (Q) (u, v) C1 (Q) u, u+1 uv C1 (Q).
Since tn > 0 for all n 1, it is easy to see that vn > 0 for all n 1.
Assuming the lemma, we have (Pn+2 ) = (vn , vn+1 ) C1 (Q) (vn+1 , vn )
vn+1 +1
C1 (Q). Also (vn+1 , vn+2 ) = (Pn+3 ) C1 (Q) and vn+1 , vn vn+1 C1 (Q).
Sincethe affine line x = vn+1 cuts C1 (Q) in at most the two points (vn+1 , vn )
and vn+1 , vvn+1 +1
n vn+1
, we have vn+2 = vn or vn+2 = vvn+1 +1
n vn+1
.
vn+1 +1 1 tn+2 tn+3 tn tn+3
If vn+2 = vn vn+1 , then 1 + vn+1 = vn vn+2 1 + tn+1 tn+4 = tn+1 tn+2
tn+2 tn+5
tn+1 tn+4 + tn+2 tn+3 = tn tn+5 .
tn+3 tn+4
Assume for a contradiction that there exists an n such that vn = vn+2 .
Then (vn , vn+1 ) = (vn+1 , vn+2 ) (Pn+2 ) = (Pn+3 ) (Pn+2 +
Pn+3 ) = O Pn+2 + Pn+3 = O (n + 2)P + T + (n + 3) + T = O
97
(2n + 5)P = O which contradicts that P is not a torsion point of C(Q). We

have used that is a group isomorphism. Therefore we have the recurrence
tn+5 tn = tn+4 tn+1 + tn+3 tn+2 for all n 1. What is left to prove is 10.0.6.
Lemma 10.0.7. If n 1, then xn < 0.
Proof: We prove this by induction on n. We already have checked it for

n = 1, 5. Pn+1 = Pn + (2, 2). Pn 6= P (2, 2) since xn < 0 < 2. The equation
y2
of the line Pn P is x2 = xynn2
2 . When we intersect it again with C we find
4 6yn + 2x2n + 4xn

xn+1 = .
(xn 2)2
xn+1 < 0 4 6yn + 2x2n + 2xn < 0 x2n + 2xn 2 < 3yn yn (yn +
xn ) + x2n = x3n + 2x2n 2xn > 3xn yn which holds if xn 6= yn . If not, then
xn+1 = 0 which implies yn+1 = 0, so Pn+1 = T . This is impossible since P
and Pm are not torsion points of C(Q) for any m.
Lemma 10.0.8. For every n 1, we have an+2 t2n+2 = tn tn+4 .
b b
Proof: Denote an+2 = a, bn+2 = b and tn+2 = t. Pn+2 C(Q)
(
t3 t3
+ ta2 ) = ta2 ta2 1 ta2 + 2 b(b + at) = a(a t2 )(a + 2t2 ). We have
Pn+4 = Pn+2 + 2 (2, 2).
Notice that if Q(x, y) C(Q), then Q = Q O is the second point
of intersection of C and {X = x}. It is not hard to see that this point is
Q(x, y x).
The tangent to C at P (2, 2) is 2x y 2 = 0. It cuts C again at (1, 0),
so 2P = O (1, 0) = (1, 1).
b
y+1 yn+2 +1 +1 b+t3
t3
The equation of the line Pn+2 (1, 1) is x1 = xn+2 1 = a
1 = att3
=
t2
. The x-coordinates of the intersection of Pn+2 (1, 1) and C(Q) are given
by ((x 1) 1)((x 1) + (x 1)) = x(x 1)(x + 2). Since we already
know that this equation has the solutions 1 and xn+2 , we find xn+4 =
2
b+at
(+1)2 t2 (b+at)2 2 2 t2 a(at2 )(a+2t2 )+abt+a2 t2
att3
xn+2 = a = a(at2 )2
= b +2abt+a
a(at2 )2
= a(at2 )2
=
a2 +at2 2t4 +bt+at2 4t2 (at2 )+t(b+t3 )
(at2 )2
=1+ (at2 )2
.
2 2
Similarly, Pn = Pn+2 2(2, 2) = Pn+2 + (1, 0) and xn = 1 + 3t (at (at )tb
2 )2 .
Since xn = at2n , xn+4 = at2n+4 and gcd(an , tn ) = gcd(an+4 , tn+4 ) = 1, the
n n+4
certain thing to say is tn+4 |a t2 and tn |a t2 .
Let p be a prime number such that p|a t2 . Then p|b(b + at) = a(a
t2 )(a + 2t2 ). If p was to divide both b and b + at, then p|at which together
with p|a t2 yields p|gcd(a, t) which contradicts gcd(a, t) = 1. By Lemma
10.0.7, xn+2 < 0 ta2 < 0. In particular, this implies a 6= t2 . If p|b + at,
then since p|a t2 , also p|b + t3 . Let vp (a t2 ) = k i.e. pk |a t2 and
pk+1 6 |a t2 . Then it is easy to prove that pk |b or pk |b + t3 .
If pk |b, then we will prove that p2k |3t2 (a t2 ) tb. If pk |b + t3 , then we

prove that p2k |4t2 (a t2 ) + t(b + t3 ). Lets see how these solve this lemma.
t2n is obtained from (a t2 )2 by clearing the factors it has in common with
3t2 (a t2 ) tb. If p is a prime dividing both a t2 and 3t2 (a t2 ) tb,
then p|tb. If p|t, then also p|a which contradicts gcd(a, t) = 1. Therefore p|b
and in this case vp (a t2 ) = k p2k |3t2 (a t2 ) tb which means nothing
else than p 6 |tn . A similar argument proves that if p divides both a t2 and
4t2 (a t2 ) + t(b + t3 ), then it does not divide tn+4 . But any prime number
dividing a t2 divides exactly one of the numbers 4t2 (a t2 ) + t(b + t3 ) and
3t2 (at2 )tb. So the prime factors of at2 are split at their greatest powers
between tn and tn+4 who both divide a t2 . Hence tn tn+4 = (a t2 ). How
do we decide on the sign then? By the previous lemma, xn+2 < 0 at2n+2 <
n+2
0 a = an+2 < 0. Since tm > 0 m 1, we have (a t2 ) = tn tn+4 1.
If the sign was + , then we would have 0 > a t2 1 which is impossible.
Therefore the sign is and tn tn+4 = a t2 = an+2 t2n+2 .
Assume then that pk |b, so there exist , Z such that a t2 = pk
and b = pk . We know that b(b + at) = a(a t2 )(a + 2t2 ), hence (b + at) =
a(a + 2t2 ) t3 t2 (3t2 )(mod pk ) pk |3t2 t p2k |3t2 (a t2 ) tb.
We have used that gcd(b, t) = 1 and p|b imply p 6 |t. The case pk |b + t3 is
treated similarly.
Lemma 10.0.9. For all n we have:
tn+3 tn 3at b + 2t3

= and
tn+1 2t2 a
tn+4 tn+1 2at + b + 2t3

= ,
tn+3 2t2 a
where (a, b, t) = (an+2 , bn+2 , tn+2 ).
Proof: We first prove that these equalities imply each other, so it

suffices to prove just one of them. By multiplying the two, and keeping in
mind the equation (xy + 1)(5 x y) = 6 of C 0 (Q) = (C(Q)), we get
(3at b + 2t3 )(2at + b + 2t3 )

2 2 1++
tn tn+4 = = t = t ,
(2t2 a)2 5

with (, ) = (x, y) = 2y3x2x , 2+y2x
2x . Since + = 45x
2x , we get
tn tn+4 = t2 (1 x) = a + t2 = an+2 + t2n+2 which is Lemma 10.0.8. Hence
it is enough to prove just one of the assertions of Lemma 10.0.9.
Lets say we want to prove
tn+3 tn 3at b + 2t3

=
tn+1 2t2 a
99
for all n 1. Following the ideas in Lemma 10.0.8 it can be proved that:
2 2 )6t(b2t3 )
(
xn+3 = 2 + 12t (a2t (a2t 2 )2 2t2 a
18t2 (a2t2 )+6t(b+4t3 ) tn+3 tn+1 = gcd(2t2 a,6)

xn+1 = 2 +

(a2t2 )2
( 12t2 (a2t2 )6t(b2t3 ) .
xn+3 = 2 + (a2t 2 )2
2 3atb+2t 3
tn+3 tn = gcd(2t2 a,6)

2 (at2 )tb
xn = 1 + 3t (at

2 )2
The result follows obviously by dividing the two new relations.

a
tn+3 tn 3 3 b3 +2
From Lemma 10.0.9, it follows vn = tn+2 tn+1 = 1t 3atb+2t
2t2 a
= t2
2 a3
t
=
t
2y3x 2+y2x
2x . Similarly vn+1 = 2x .
It follows that (xn+2 , yn+2 ) = (vn , vn+1 )
and so Lemma 10.0.6 is proved. We have seen how this completes the proof
of the problem.
Chapter 11
Integer points on elliptic

curves
So far we have only been interested in the structure of the rational points
C(Q) of an elliptic curve given by C : y 2 = x3 + ax2 + bx + c. But for ages
number theorists have been interested in Diophantine Equations. To honor
their work, we begin the study of integer points on elliptic curves. The
Diophantine Equation to consider is y 2 = x3 + ax2 + bx + c with a, b, c Z.
In the spirit of the course, we denote the set of integer solutions of this
equation by C(Z).
One of the strongest results in connection to this problem is:
Theorem 11.0.10 (Siegel). If C is nonsingular, then the equation y 2 =
x3 + ax2 + bx + c has a finite number of integer solutions i.e. C(Z) is finite.
Notice that Siegels Theorem does not hold if we drop the assumption
on the non-singularity of C. For example, the equation y 2 = x3 has an
infinite number of integer solutions. They can actually pe parameterized as
(x, y) {(t2 , t3 )| t Z}.
For the time being we leave aside the proof of Siegels Theorem and shall
content ourselves to prove another result concerning equations of a more
particular form. This is Thues Theorem proved by Thue around 1909. We
dedicate to it the next section.
11.1 Thues Theorem

Theorem 11.1.1 (Thue). If a, b, c Z , then the equation C(Z) : ax3 +
by 3 = c has a finite number of integer solutions.
In connection to Thues Theorem we have the next two famous equations:
Proposition 11.1.2. If d is a positive integer which is not the cubic power
of another integer then the equation x3 dy 3 = 1 has at most two integer
solutions.
101
102 CHAPTER 11. INTEGER POINTS ON ELLIPTIC CURVES
We will return to this problem in the final lecture.

Another particular case of the equation in Thues Theorem, is Mordells
Equation:
y 2 = x3 k, given that k 6= 0.
We have already treated some particular cases of Mordells Equation in 3.2.3.
11.1.1 Proof of Thues Theorem and Diophantine Approxi-

mation
We want to prove that given a, b, c, Z , the equation ax3 + by 3 = c has a
finite number of integer solutions. By multiplying the equation with a2 we
obtain the equivalent equation (ax)3 + (a2 b)y 3 = (a2 c). Its integer solutions
are in one-to-one correspondence to the integer solutions of the equation
x3 (a2 b)y 2 = (a2 c) for which a|x. We see that if we prove that for
b, c Z , the equation x3 by 3 = c has a finite number of solutions, then
Thues Theorem follows.
If b = k 3 for some k Z, then the equation is (xky)(x2 +kxy +k 2 y 2 ) =
c. Since c 6= 0, solving the equation reduces to solving the simple system of
equations in integers:

x ky = c1
x2 + kxy + k 2 y 2 = c2 .
c1 c2 = c

Lets take an example for this case:

Example 11.1.3. Solve in integers the equation: x3 + y 3 = 1729.
Solution: We have to solve the system:

x + y = c1
x2 xy + y 2 = c2 .
c1 c2 = 1729

We have the prime factor decomposition 1729 = 7 13 19. It is easy to

see that x2 xy + y 2 > 0, so we need only consider the positive can-
didates for c1 and c2 . The solutions given by solving the equations are
(9, 10), (10, 9), (1, 12), (12, 1).
This example was not given at random. 1729 is the first positive integer
that can be written in two distinct ways as a sum of two cubes. It also has an
anecdote attached. It is said that one day Hardy, the famous mathematician,
was visiting his not less famous, hospitalized friend, Ramanujan. To cheer
him up, he told Ramanujan that he was extremely bored by his ride to
the hospital. He was displeased because 1729, the number of the carriage he
used, was completely uninteresting to him. Believe it or not, after a moment
of thought, Ramanujan pointed his friend on the nice property of 1729 we
have described above.
11.1. THUES THEOREM 103
Conjecture 11.1.4. It is an open problem whether for all n 1 there exists

an m Z such that the equation x3 + y 3 = m has exactly n distinct integer
solutions with x y.
We return to the proof of Thues Theorem by studying the equations of

the form x3 by 3 = c with b, c Z and b and integer which is not the
cubic power of another integer. Eventually by multiplying the equation by
1 and replacing x or y by x or y respectively, we can assume that b
and c are positive integers. We can consider only the case when b is not a
cube because we have already seen how to treat the other case.

Let = 3 b. The equation can be rewritten as (xy)(x2 +xy+y 2 2 ) =
c. It is easy to prove x2 +xy +y 2 2 43 2 y 2 . We have at most one solution
for x3 + 03 = c, so at most one solution with y = 0.
c c 4c
With y 6= 0, we have x y = x2 +xy+y 2 2 3 2 2 = 3 2 y 2 . Since
y
4
by the assumption b, c N , c > 0, the right hand side of the previous
inequality is positive, hence so is the left hand side. Therefore

x 4c 1

3 2 y 3 .

y
If we manage to prove that this inequality has a finite number of integer

solutions (x, y), then we are done. This is achieved by Diophantine Ap-
proximation, which will justify in full the second part of the title of this
subsection.
In 1909 Thue actually proved a much stronger version of the theorem we
prove here:
Theorem 11.1.5 (Thue). Let f Z[X, Y ] be a homogeneous polynomial

of degree greater of equal to 3. We assume that f (X, 1) is irreducible in
Q[X]. Then for every k Z, the equation f (x, y) = k has only a finite
number of integer solutions.
To prove this, Thue first proved the following approximation theorem:
Theorem 11.1.6 (Thue). Let f Z[X] be a polynomial, irreducible over

Q[X], of degree d 3 and let be one of its complex roots. Then for all real
> 0 and C > 0 there exists only a finite number of pairs (p, q) Z N
such that pq d +1+
C
.

q2
This theorem is part of the following approximation problem: In the

conditions of the theorem above, what is the smallest function : N R+
for which there are only a finite number of pairs (p, q) Z N such that

p C
q . The history of the advances made in this problem is:

q (d)+
Liouville(1850) (d) = d
T hue(1909) (d) = d2
+1
Siegel(1921) (d) = 2 d
Gelf on, Dyson(1947) (d) = 2d
Roth(1955) (d) = 2
Roth also proved that the bound (d) = 2 cannot be improved. He was
awarded the Fields Prize for his contribution to this problem.
To solve our problem, we will prove the following particularization of the
previous approximation theorem:

Theorem 11.1.7 (Thue). If b N such that = 3 b 6 Q and C is a
positive real
number,
then there is a finite number of pairs (p, q) Z N
p
such that q < qC3 .

The proof of this theorem is made up of a series of technical lemmas.
Lemma 11.1.8 (Siegels Lemma). Let N > M be two positive integers

and let (aij )i=1,M ,j=1,N Z, not all 0. Then the system:

a11 T1 + a12 T2 + . . . + a1N TN = 0
a21 T1 + a22 T2 + . . . + a2N TN = 0

..

.
aM 1 T1 + aM 2 T2 + . . . + aM N TN = 0

has a nontrivial solution such that

M
max{|Ti | | i = 1, N } < 2(4N max{|aij | | i = 1, M , j = 1, N }) N M .
Proof: Let A MM,N (Z) be the matrix whose entries are (aij )i=1,N ,j=1,M .
Let t MN,1 (Z) be the vertical vector whose entries are (Ti )i=1,N . Denote
k t k= max{|Ti | | i = 1, N } and k A k= max{|aij | | i = 1, M , j = 1, N }.
M
We are looking for t such that At = 0 and k t k< 2(4N k A k) N M .
For arbitrary H > 1, let
TH = {t ZN | k t k H}.
By the triangle inequality we see that t TH k At k N k A k H. If we

set
UH = {u ZM | k u k N H k A k},
we have proved that
t TH At UH .
A simple count yields |TH | = (2[H] + 1)N and |UH | = (2[N H k A k] + 1)M ,
where [H] denotes the integer part of H, i.e. the greatest integer least or
equal to H.
Since N > M , there exists H > 0 such that (2[H] + 1)N > (2[N H k A k
] + 1)M . For such H we have |TH | > |UH |. Since A sends TH into UH , there
exist t1 6= t2 in TH such that At1 = At2 . Then for t = t1 t2 we have t 6= 0,
At = 0 and k t k 2H.
To finish the proof of the lemma we must prove that we can choose such
M
H with the additional restriction H < (4N k A k) N M . So we need H > 0
M
with (2[H] + 1)N > (2[N H k A k] + 1)M and H < (4N k A k) N M .
We have (2[H] + 1)N > (2H 1)N H N and (2[N H k A k] + 1)M
(2N H k A k +1)M (3N H k A k)M . Thus for having (2[H] + 1)N >
(2[N H k A k] + 1)M , it suffices H N (3HN k A k)M H (3N k A k
M M
) N M . We take H = (3N k A k) N M to finish the proof of the lemma.
Theorem
11.1.9 (Auxiliary Polynomial). Let b N such that =
3
b 6 Q. Let m, n N such that m + 1 > 2n 2n

3 m 3 i.e. 3 m = 3 .
Then thereP exist P, iQ Z[X],Pnot both 0, both of degree at most m + n,
P (X) = m+n i=0 u i X , Q(X) = m+n i
i=0 vi X , such that F
(k) (, ) = 0 for all
k = 0, n 1 and
max{|ui |, |vj | | i, j = 0, m + n} 2(16b)9(m+n) ,
not 1 k F
where F (X, Y ) = P (X) + Y Q(X) and F (k) (x, y) = k! X k
(x, y).
Proof: First of all we will be looking for a polynomial F (x, y) = m+n i

P
i=0 (ui x +
vi xi y) such that F (k) (, ) = 0 for all k = 0, n 1. It is easy to see that
m+n
X
(k) ik i ik i
F (x, y) = ui x + vi x y =
k k
i=k
m+nk
X
j+k
j j j+k
= uk+j x + vk+j x y .
k k
j=0
n n!
, also denoted by Cnk .

kdenotes the binomial coefficient equal to (nk)!k!
When we substitute x = y = in the previous polynomial equality, we get
m+nk
X
(k) j+k j j+k j+1
F (, ) = uk+j + vk+j =
k k
j=0
m+nk+1
X
i+k i+k1
= uk+i + vk+i1 i .
k k
i=0
We use the convention nk = 0 if k > n. Keeping in mind that 3 = b, we

can write

2 m+nk+1
X i + k
X i+k1
F (k) (, ) = ui+l + vi+l1 bj l .
k k
l=0 0=i=3j+l
The index i in the previous expression is an auxiliary index (no connection

with the name of the theorem we prove) used to shorten the writing of the
formula. The second sum actually runs with j from 0 to m+nkl+13 .
(k)
To have F (, ) = 0 for all k = 0, n 1, it is enough to have
[ m+nkl+1
3 ]
X 3j + l + k 3j + l + k 1
uk+3j+l + vk+3j+l1 = 0
k k
j=0
for all k = 0, n 1 and l = 0, 2. The conditions are also necessary since

{1, , 2 } are linearly independent over Q. We can see this as a linear
system of 3n equations with 2(m + n + 1) numbers to determinate i.e. ur
and vr with r = 0, m + n. Denote M = 3n and N = 2(m + n + 1).
If we prove that N > M , then by Siegels Lemma 11.1.8, it follows that
there exist ui and vj for all i, j = 0, m + n, not all 0, such that
M
max{|ui |, |vj | | i, j = 0, m + n} < 2(4N max{|ars | r = 1, M , s = 1, N }) N M ,
where ars are the coefficients

in the system, all of them having the form
j 3j+l+k j 3j+l+k1
b k or b k with 0 l 2 and 3j + l + k m + n. It is
m+n
easy to see that b 3 2 m+n is greater than any of them. Therefore there
exist ui and vj such that F (k) (, ) = 0 for all k = 0, n 1 and
m+n 3n
max{|ui |, |vj | | i, j = 0, m + n} < 2(4 2(m + n + 1) b 3 2m+n ) 2mn+2 .
To finish the proof, it suffices to show

m+n 3n
2(4 2(m + n + 1) b 3 2m+n ) 2mn+2 2(16 b)9(m+n) .
And dont forget we still had to prove N > M .

N > M 2(m + n + 1) > 3n 2m + 2 > n 2 2n

3 + 2 > n. The last
follows from 2 2n
3 + 2 > 4n
3 > n.
From the inequality x + 1 2x for all x > 0, it follows that 2(4
m+n 3n m+n 3n
2(m + n + 1) b 3 2m+n ) 2mn+2 2(22m+2n+3 b 2 ) 2mn+2 . We have
m+n 3n 3n(2m+2n+3) (m+n)n
2(22m+2n+3 b 2 ) 2mn+2 2(16 b)9(m+n) 2 2mn+2 b 2mn+2 (16
b)9(m+n) . We just have to prove
n (2m + 2n + 3)
< 12(m + n)
2m n + 2
and
(m + n) n
< 9(m + n)
2m n + 2
to be done.
(m+n)n 9m+9
2mn+2 < 9(m + n) n < 18m 9n + 18 n < 4 , which follows
from n < 3(m+1)
2 < 9(m+1)
5 .
4n
2m+2n+3 +2n+3 n(2m+2n+3)
2mn+2 <
3
4n
n
= 10n+9
n . For 2mn+2 < 12(m + n) it suffices to
3
prove 10n + 9 < 12(m + n) 9 < 12m + 2n. By hypothesis, m 3 and the
conclusion follows.
Theorem 11.1.10 (Smallness Theorem). In the conditions of the Aux-

iliary Polynomial Theorem, there exists a real number c1 depending just on
b such that for all x, y R with |x | 1 and for all t = 0, n 1, we have
|F (t) (x, y)| cn1 (|x |nt + |y |).
Proof: The backbone of this proof is Taylors Expansion Formula for F

around (, ). According to this formula, we have
X 1 k+j F
F (x, y) = k j (, ) (x )k (y )j .
k!j! x y
j,k0
Since F (x, y) = P (x) + y Q(x), F (k) (, ) = 0 for all k = 0, n 1, the

not
degrees of F and Q in X are both least or equal to m + n, and F (k) (x, y) =
1 k
F
k! X k (x, y), it is easy to see that
m+n
X m+n
X
F (x, y) = F (k) (, ) (x )k + (y ) Q(k) () (x )k .
k=n k=0
By induction,
m+n m+n
(t)
X
(k) kt k X
(t) kt k
F (x, y) = F (, )(x) +(y) Q ()(x)
t t
k=n k=t
for all t = 0, n 1. By the triangle inequality, for |x | 1, we have:

m+n ! m+n !
(t)
X
(k) k nt
X
(t) k
|F (x, y)| |F (, )| |x| +|y| |Q ()|
t t
| k=n {z } | k=t {z }
A(t) B(t)
for all t = 0, n 1. To complete the proof of the problem it is enough to

prove that we can find c1 > 0 depending only on b = 3 such that A(t) < cn1
and B(t) < cn1 for all t = 0, n 1.
Pm+n
Recall that F (x, y) = i=0 (ui xi + vi xi y). From this it follows
m+n
X
(k) i ik i ik+1
F (, ) = ui + vi
k k
i=0
for all k = n, m + n. From the Auxiliary Polynomial Theorem 11.1.9,

max{|ui |, |vj | | i, j = 0, m + n} 2(16b)9(m+n) . Using this, the triangle
inequality and the obvious inequalities ik+1 < m+n and ki < 2m+n for
i = 0, m + n and k = n, m + n, we get
|F (k) (, )| 2m+n (2(16b)9(m+n) ) m+n (2(m + n + 1)).
From this and again from the triangle inequality, it follows

m+n
X k
A(t) |F (k) (, )|
t
k=n
(m + 1) 2m+n (2m+n (2(16b)9(m+n) ) m+n (2(m + n + 1))))

28 5 28 5
< 242(m+n) b 3 (m+n) (242 3 b 3 3 )n .
140
Let c1 = 270 b 9 . c1 indeed depends only on b. Analogously we prove that
B(t) < cn1 .
Theorem 11.1.11 (Non-Vanishing Theorem). In the conditions of The

Auxiliary Polynomial Theorem 11.1.9, there exists a constant c2 depending
only on b such that for all irreducible fractions with positive
denominators,

p1 p2 c2 n (t) p1 p2
q1 and q2 , there exists 0 t 1 + ln q1 such that F q1 q2 6= 0.
,
Proof: The main ingredient of this proof is what is called the Wronskian
of P and Q, defined as

P (x) Q(x)
W (x) = 0 = P (x)Q0 (x) P 0 (x)Q(x).
P (x) Q0 (x)

We prove that W 6= 0. If Q = 0, then F = P and since F (k) (, ) =

0 for all k = 0, n 1, it follows that (x )n |P (x). Since the minimal
polynomial of is x3 b, we have that (x3 b)n |P (x). By comparing the
degrees of the two, m + n 3n m 2n. But 2n 3m and we
have
0
P (x)
a contradiction. Therefore Q 6= 0. Assume W = 0. Then since Q(x) =
P 0 (x)Q(x)P (x)Q0 (x) P
Q2 (x)
= 0, the rational function Q must be a constant. So
there exists u Q such that P (x) = u Q(x). We then have F (x, y) =
(u + y) Q(x). Just like before, since F (k) (x, y) = (u + y) Q(k) (x), we prove
that (x3 b)n |Q(x) and we find the same contradiction m 2n 3m. We
have proved W 6= 0.

Let T be an integer such that F (t) pq11 , pq22 = 0 for all t = 0, T 1 and

F (T ) pq11 , pq22 6= 0. We use the convention F (0) = F . Such an integer T
exists because F 6= 0. For all t, t0 T 1, we have

P (t) p1 + p2 Q(t) p1 = 0
q1 q2 q1
P (t0 ) p1 + p2 Q(t0 ) p1 = 0
q1 q2 q1

(t) p1 (t0 ) p1 (t0 ) p1 (t) p1
P Q P Q = 0.
q1 q1 q1 q1

By induction, W (r) pq11 is a linear combination of such terms, hence is 0
T 1
for all r = 0, T 2. It follows that x pq11 |W (x). Since p1 and q1
are coprime and since W has integer coefficients, it is not hard to prove
that there exists V Z[X] such that (q1 x p1 )T 1 V (x) = W (x). W 6=
0 V 6= 0. If s and as are the degree and the leading term respectively
of W , then from the previous divisibility relation, q1T 1 |as . Since V 6= 0,
|as | q1T 1 . By passing to logarithms,
ln |as |
T 1 .
ln q1
Pm+n Pm+n P
m+n

W (x) = P Q0 P 0 Q = i
i=0 ui x j=0 j vj x
j1
j=0 j uj x
j1
Pm+n i

i=0 vi x . The leading term of this polynomial expression is, by nota-
tion, the coefficient of the degree s part. This means
X
as = j (ui vj uj vi ).
i+j=s+1
By the Auxiliary Polynomial Theorem 11.1.9, max{|ui |, |vj | | i, j =

0, m + n} 2(16b)9(m+n) = . By the triangle inequality, since in the
previous sum the terms with i or j greater than m + n are 0, |as | 2 2
P m+n 2 m+n [2 (16b)9(m+n) ]2
j=0 j = (m + n)(m + n + 1) 4
5n
ln |as | (m+n) ln 4+ln 4+18(m+n) ln(16b) ln 4+ln 4+45n ln(16b)
3

5
ln 4 + 45 ln(16b) + 1 n.
3
| {z }
c2
c2 n
c2 indeed only depends on b. Also, T 1 ln q1 .
Proof of Thues Diophantine Approximation Theorem 11.1.7:

For clarity, we recall what we want to prove. We want to prove that given

b N such that = 3 b 6 Q and given any real number C > 0, there exist
p
only a finite number of pairs (p, q) Z N such that q < qC3 . It is

easy to see that we can assume that gcd(p, q) = 1 and C > 1.

Assume for a contradiction that there exist infinitely many such pairs.
Then we can find such pairs with arbitrary large q. In particular, there
exists (p1 , q1 ) Z N such that

p1 C
gcd(p1 , q1 ) = 1, < 3 , q1 > e9c2 and q1 > (2c1 C)18 ,

q1 q1
with c1 given by The Smallness Theorem 11.1.10 and c2 given by the Non-
Vanishing Theorem 11.1.11. It is easy to see that we can assume c1 > 1 and
c2 > 1. Take another pair (p2 , q2 ) Z N with

p2 C
gcd(p2 , q2 ) = 1, < 3 and q2 > q165 .

q2 q2
The importance of the dependence of c1 and c2 only on b in 11.1.10 and
11.1.11 is now visible as it intervenes in the choice of q1 . Take

9 ln q2 2n
n= and m = .
8 ln q1 3
We check that with these choices of n and m we are in the conditions of the
Auxiliary Polynomial Theorem. All that we have to prove is that m 3.
ln q2
We have n > 98 ln 9 2n
q1 1 > 8 65 1 > 72 m 3 1 > 47 3.
8(n+1)
9 ln q2 8 8 8n
n= ln q1 n ln q2 < ln q1 (n + 1) q19 q2 < q1 9 .
8 ln q1 9 9
ln q1 c2 n n
q1 > e9c2 c2 < t1+ 1 + n 1,
9 ln q1 9
where t is the one given by The Non-Vanishing Theorem 11.1.11. Since F
has integer
coefficients,
after bringing fractions to a common denominator,
0 6= F (t) pq11 , pq22 = qm+n
z
q
for some z Z , it follows
1 2

p 1 p2 1 1 1
|F (t) , | > = .
q1m+n q2
2n
q1 q2 3
+ 89 (n+1)+n 23n
9
+ 89
q1 q1
p1 p2
We now check that with x = y= q1 ,
and t we are in the condition
q2
of The Smallness Theorem 11.1.10. We have seen that t 6= n 1. By
assumption |x | < qC3 . q1 > (2c1 C)18 > C 18 > 3 C. We have used that
1
c1 and C are greater than 1. Therefore |x | < qC3 < 1 and we can apply
1
The Smallness Theorem 11.1.10 to obtain

nt ! n1 n
p1 p 2 C C C 9 C
|F (t) , | cn1 + 3 cn1 3 + 8n
q1 q2 q13 q2 q1 q13
11.2. LJUNGGRENS EQUATION - A PARTICULAR CASE 111
n
8
n1
C 9 (2c1 C)n q118 1
cn1 2 8
8 < 8 = 47 .
3
n3 3
n3 3
3 18
n3
q1 q1 q1 q1
8
n 9 ln q2
We have used q23 q13 . Which follows from n 8 ln q1 . Therefore

1 (t) p 1 p2 1 8 23 47
47 |F , | 23n + n n3
18
n3 q1 q2 9
+ 89 9 9 18
q1 q1
16 + 46n 47n 54 70 n
which contradicts n > 72.
11.2 Ljunggrens Equation - a particular case

Theorem 11.2.1 (Ljunggren). The integer solutions of the equation
x2 x + 1 = y 3
are (x, y) {(0, 1), (1, 1), (19, 7), (18, 7)}.
It is easy to prove that every element of the set given above is a solution
to the equation. However, proving that these are the only solutions is a
very difficult problem and for now we will content ourselves to solving a
particular case of the problem.
Example 11.2.2 (Balkan Mathematics Olympiad 2005). Find the

integer solutions of the equation p2 p + 1 = k 3 with p being a positive
prime number.
Solution: First of all, notice that we can assume p > 3. This is because
the equations k 3 = 3 = 22 2 + 1 and k 3 = 7 = 32 3 + 1 have no solutions
in integers.
By multiplying by 4, the equation can be rewritten as (2p1)2 +3 = 4k 3 .
Assume p 1(mod 3). Then 3| 2p 1 3| 4k 3 3| k. Reducing modulo
9 in the equation we obtain 3 0(mod 9) which is impossible.
Therefore p 1(mod 9). Then p2 p + 1 = k 3 1 k(mod 3). Let
k = 3s + 1. Then k 3 = 9 (3s3 + 3s2 + s) + 1 9| k 3 1 = p2 p 9| p 1.
We have
p1 k3 1 k 1 k2 + k + 1
p = = .
9 9 3 3
Since p > 1 and k 2 +k +1 > 0, it is easy to prove that k > 1. Since p| p p1
9 ,
k1 k2 +k+1
it follows that p| 3 or p| 3 .
k1
k1 k1 p1 2 +k+1 k2 +k+1
If p| 3 , then p 3 . Also 9 = 3
k 3 3 p1
p
|{z}
N
3k 2 +3k+3. By combining the two inequalities, we have k1 2
3 3k +3k+4
2 k2 +k+1
0 9k + 8k + 13 which is impossible in integers. Therefore p| 3 .
k2 +k+1 p1 k1 k2 +k+1 p1 k1
Set 3 = pt with t N . Then p 9 = 3 3 9 = 3 t.
k2 +k+1 k1 2 k1 k2 +k+1
1(mod k1

We have 3 = 3 3 + 3 3 + 1, hence 3 3 ).
k1 p1 k1 k2 +k+1 k1
3 | 9 p 1(mod 3 ). 1 3 = pt(mod 3 ) t
k1
1(mod 3 ).
k2 +k+1
Assume t > 1. Since t 1(mod k1 3 ), t 3
k+2
3 = p t p k+23 .
k1 p1 2
k +k+1 (3k2)(k+2) 3k2 +4k4
Since t 3 = 9 , p 3k 2. Then 3 3 = 3
0 2k 2 + 3k 5 = (k 1)(2k + 5) 25 k 1 which contradicts k > 1.
2
Therefore t = 1 which implies k +k+1 3 = p and p 1 = 3(k 1) p =
3k2. Substituting yields k +k+1 = 9k6 k 2 8k+7 = 0 k {1, 7}.
2
Since k > 1, k = 7 which implies p = 19.

So the only solution of the equation p2 + p + 1 = k 3 with p prime is
(p, k) = (19, 7).
Chapter 12
Generators for Elliptic

Curves
The thematic of this lecture is obvious from the title. We have seen some
methods for computing the rank of elliptic curves in a few particular cases.
We will illustrate some examples to prove that computing explicit generators
for the abelian group of an elliptic curve is a much harder problem.
We recommend (re)reading the Algebraic Number Theory Prerequisites
section in the beginning of Lecture VIII.
Example 12.0.3. Find a set of generators for the abelian group C(Q) as-
sociated to the elliptic curve y 2 = x3 13.
Proof: The working plan is:
1. There are no torsion points on C(Q) different from O.
2. rank(C(Q)) = 1.
3. (17, 70) or (17, 70) generate C(Q).
We will first prove:
Proposition 12.0.4. The only integer solutions of y 2 = x3 13 are (17, 70).
Proof: (70)2 + 13 = 4913 = 173 , so (17, 70) are indeed solutions to

the given equation. To see that these are all the solutions of the equation,
we will use the ring A = Z[i 13]. Unfortunately, A is not factorial. For this
we prove that 2 is irreducible, but not prime. In A we have a multiplicative
norm
function N : A N which extends
to a group homomorphism N :

Q(i 13) Q+ such that N (a + b i 13) = a2 + 13b2 for all a, b Q, not
both 0. It is easy to prove that x A is a unit if and only if N (x) = 1.
Assume 2 is not irreducible. Then there exist non-units x, y A such that
2 = xy. Then N (2) = N (x)N (y). Since N (2) = 4 and x and y are not
113
114 CHAPTER 12. GENERATORS FOR ELLIPTIC CURVES

units, N (x) = N (y) = 2. Let x = a + b i 13 with a, b Z. Then
N (x) = 2 a2 + 13b2 = 2. But the last equation obviously has no integer
solutions, so we get a contradiction,
therefore
2 is irreducible. We have
2|14 = (1 + i 13)(1 i 13) and 2 6 |1 i 13 in A, hence 2 is not prime.
That A is not factorial is also a consequence of the ideal class group C of A
(see 8.1.2) not being trivial. We will prove that |C| = 2 and that a complete
set of representatives is A (the class of principal ideals of A) and the class
of the maximal ideal P = 2A + (1 + i 13)A.
Assume we have proved these assertionson C. Then the equation y 2 +
3 3 3 3
13 = x is equivalent to (y + i 13)(y i 13) = x (xA) = x A =
(y + i 13)A (y i 13)A. By 8.1.2, every nonzero ideal of A decomposes
uniquely as a product of prime ideals, and, as a consequence, we have proved
the existence of an ideal theoretic greatest common divisorfor two nonzero
ideals. With this in mind, let I = gcd((y + i 13)A, (y i 13)A). Then

I|(y i 13)A (y i 13)A I 2i 13 I I|2i 13A.

2 = (4, 2 + 2i 13, 12 + 2i 13) = (4, 2 +
P = (2, 1 + i 13) P
2i 13, 2i 13) = (4, 2, 2i 13) = 2A. It is not hard to prove that i 13 is
irreducible and prime, hence Q = i 13A is a maximal ideal of A. Therefore
I|P 2 Q.If Q|I, then I Q and since y i 13 it follows
I, 2y Q.
Since i 13 6 | 2 in A, i 13| 2y i 13| y y = i 13 (a + b i 13) for
some a, b Z. We obtain y = 13b. But (13b)2 13 = x3 13| x
132 | (13b)

2 13 which is impossible. So Q 6 | I and I| P 2 . If P | I, then

P 2 | (y+i 13)A(yi 13)A P 2 | (xA)3 2A| (xA)3 (xA)3 2A x
is even. We have y 2 + 13 = x3 and modulo 8 we get y 2 3(mod 8) which is
impossible. Therefore P 6 | I and finally I = A.
The unique decomposition
of the nonzero ideals of A as products of
prime ideals, (y + i 13)A (y i 13)A = (xA) 3 and gcd((y + i 13)A, (y

i 13)A) = A prove that (y + i 13)A = J 3 for some ideal J of A. By going
to classes in C, we get A = J3 = J sincewe assume |C| = 2, thus J is a
principal ideal of 3
A. It follows that y + i 13 is associated to 2a cube2 z of
A. u = a + b i 13 is a unit of A if and only if N (u) = 1 a + 13b = 1.
It is easy to see that the
only solutions are u = 1. Since 1 is also a cube
3
in A, we get that y + i 13 = z for some z A. If z = a + b i 13 for some
a, b Z, then
y = a3 3ab2 13

.
1 = 3a2 b b3 13
The last equation can be rewritten as b(3a2 13b2 ) = 1 b = 1. b = 1
leads to 3a2 13 = 1 which is impossible. Therefore b = 1 and 3a2 13 =
1 a = 2. These lead to the solution (x, y) = (17, 70).
We are left to prove the assertions made on C. To prove them we will
need some more results from Algebraic Number Theory.
115
Theorem 12.0.5 (Minkowski). Let K be an algebraic extension of Q of

degree n and let A be the ring of integers of K. Assume A = Z[u] for some
u A. If J is a nonzero ideal of A, then there exists a nonzero ideal I of A
such that I J(mod P rA) and
t
4 n! p
|A/I| = N (I) |K |,
nn
where K = det(T rK (ei ej ))i,j=1,n , (ei )i is a basis for K over Q such that
A = ni=1 Zei , s + 2t = n and s is the number of real embeddings of K i.e.
field homomorphism K , R. T rK (x) is by definition [K : Q(x)] T r(x),
where T r(x) is the sum of conjugates of x i.e. the sum of all the distinct
complex roots of the irreducible polynomial of x over t Q. p
K is called the discriminant of K over Q. 4 nn!n |K | is called the
Minkowski constant of K.

In our case, we have K = Q(i 13), A = Z[i 13], u = i 13, n = 2,
s = 0, t = 1, e1 = 1, e2 = i 13. With these,

T rK (1) T r K (i 13) 2 0

K = det
= det
= 52.
T rK (i 13) T rK (13) 0 26
By 12.0.5, for every nonzero ideal J of A, there exists a nonzero ideal I

such that I J(mod P rA) and N (I) 4 24 52 = 208 4 N (I)
{1, 2, 3, 4}. So, to determine a complete set of representatives for C it is
enough to look through the non-equivalent ideals of norm least or equal to
4. To determine these ideals we need one more result of Algebraic Number
Theory. This result describes the behavior of prime integers in the rings of
integers of number fields:
Theorem 12.0.6. Let K be a number field such that [K : Q] = n and let A

be its ring of integers. Assume A = Z[u] for some u K. Let f = IrrQ (u)
be the irreducible polynomial of u in Q[X]. Let p be a prime integer. Assume
f = e11 . . . err , where f is the reduced of f modulo p and 1 , . . . , r are
monic polynomials such that when reduced mod p they become irreducible
and pairwise distinct.
Then for all i = 1, r, Pi = pA + i (u)A is a maximal ideal of A of norm
e1
pfi , where
Pr fi = degi . Moreover Pi 6= Pj for all i 6= j, pA = P1 . . . Pr
er
and i=1 ei fi = n.
We are now able to find the ideals of A of norm least or equal to 4.

If N (I) = 1, then |A/I| = 1 I = A.
If N (I) = 4, then |A/I| = 4. So A/I is a ring with 4 elements. It
follows by Lagrange,
that 4 1 = 0 in A/I,
so 42 = 0 4 I I|4A =
(2A +(1 + i 13)A)4 . 2A = (2A + (1 + i 13)A) 4 = N (2A) = N (2A +
(1 + i 13)A)2 N (2A + (1 + i 13)A) = 2. It is not hard to prove that

I = (2A + (1 + i 13)A)2 = 2A. Notice that I A(mod P rA) because I is
principal.
If N (I) = 2, then A/I has 2 elements, hence it is isomorphic to Z/2Z,
so I is a maximal ideal of A. Just like before we prove 2 I, which implies
I|2A. To find the maximal ideals of A dividing 2A we can use 12.0.6. We
have A = Z[i 13]. Let u = i 13. Then f (X) = IrrQ (u) = X + 13. 2
Modulo 2 we have f = (X + 1)2 .From 12.0.6 we obtain 2A = P12 with

P1 = 2A + (u + 1)A = 2A + (1 + i 13)A. Therefore P1 is the only ideal of
A of norm 2. Assume that P1 is principal. Then there exists x P1 such
that P1 = xA. Then by 8.1.2, 2 = N (P1 ) = N (xA) = N (x)[Q(i 13):Q(x)] ,
where N (x) is the norm of x defined in 8.1.2 which is easily seen to coincide
to the norm N we defined on A in the beginning of the proof. It is easy to
prove that x 6 Q, otherwise x Z and it would follow that x2 |2 in A which
is impossible. Therefore
[Q(i 13) : Q(x)] = 1. It follows that 2 = N (x).
Assume x = a + b i 13. Then 2 = N (x) 2 = a2 + 13b2 which is easily
seen to have no integer solutions. We have proved that P1 is not principal.
But P12 = 2A is a principal ideal, so P1 has order 2 in C.
If N (I) = 3, then just like before I is a maximal ideal of A dividing 3A.
Modulo 3 we have f(X) = X 2 + 1 which is irreducible. By 12.0.6, 3A = P2
and N (P2 ) = 9. This proves that A has no ideals of norm 3.
We have proved that the only ideals of A of norm least or equal to 4 are
A, P1 and 2A. Also a complete set of non-equivalent representatives for C is
A and P1 . We conclude C ' Z/2Z as groups. We have also finished proving
that the only integer solutions of y 2 = x3 13 are (17, 70).
We return to the proof of 12.0.3: We prove the first item on the

working plan i.e. that C(Q) has no non-trivial elements of finite order.
This is a simple application of Nagell-Lutzs Theorem. The discriminant of
X 3 13 is = 33 132 . If (x, y) is a non-trivial element of C(Q) of finite
order, then by Nagell-Lutzs x, y Z and y|. But we have seen that the
only integer solutions of y 2 = x3 13 are (17, 70). It is enough to prove
that they are not torsion points. This follows from 70 6 | = 33 132 .
We have proved that C(Q) is torsion-free, hence it is a free abelian group.
Assume we have proved that C(Q) is of rank 1. It follows that there
exists (x, y) C(Q) such that C(Q) = Z (x, y). To prove that (17, 70)
or its opposite (17, 70) = (17, 70) generate C(Q), we use the following
lemma:
Lemma 12.0.7. Let C(Q) : y 2 = x3 + ax2 + bx + c be a nonsingular elliptic
curve with a, b, c Z. Let P C(Q), P 6= O[0 : 1 : 0] and assume that
n P C(Z) \ {O} for some n N . Then P C(Z).
Proof: For a prime number p, denote

a b
Cp = , C(Q)| a, b, t Z, t > 0, gcd(a, t) = gcd(b, t) = 1, p | t {O}.
t 2 t3
117
During the proof of Nagell-Lutzs Theorem, we have proved (Cp , +)

(C(Q), +).
Assume P does not have integer coordinates. Then by 2.2.5, P = ta2 , tb3

for some a, b, t Z, t > 1 and gcd(a, t) = gcd(b, t) = 1. Let p be a prime

number such that p | t. Then P Cp and n P Cp \ {O} which contradicts
n P being a nontrivial point with integer coordinates.
Assume P = (x, y) is a generator for C(Q). Since (17, 70) C(Q)

and C(Q) = Z P , there exists n Z such that n P = (17, 70). From the
previous lemma we get that P C(Z). Since C(Z) = {(17, 70)} {O},
P {(17, 70)}.
To complete the proof of the problem we have to prove rank(C(Q)) = 1.
An algorithm for computing the rank of C(Q) can be deduced from the proof
of the General Mordell-Weil Theorem in Lecture VIII. We have already seen
that C(Q) is a free abelian group. Then C(Q) ' Zr , where r = rank(C(Q))
and C(Q)/2C(Q) ' (Z/2Z)r . It is enough to prove that |C(Q)/2C(Q)| = 2.
2 3 3
The equation of C(Q) is y = x 13= f (x). f (x) = (x 13)(x
2i
13)(x 2 3 13) with = e 3 = 1+i
3
2
3
. Let = 3
13, A = Z[] and
def K
let K = Q(). If we define : C(Q) Q() = (K )2
by
(
1, if P = O
(P ) = ,
, if P = , w , Z, N , gcd(, ) = 1, = t2 , t N
\
then the proofs of 8.2.4 and 8.2.5 show that is a well defined group homo-
morphism whose kernel is 2C(Q). By the fundamental theorem of isomor-
phism it is enough to determine Im. From 8.2.7 we know that there exist
a finite number of algebraic integers A for whichthere exist u U (A)
and K such that = u 2 as P = , w varies in C(Q) \ {O}
with , Z, > 0 and gcd(, ) = 1. Then (P ) = ud . Define

I(P ) = ( )A + 2 g A,

where f (x) = (x )g(x) g(x) = x2 + x + 2 . The s where chosen

as generators for the principal ideals appearing in the set {I(P ) Ci2 } as P
ranges in C(Q) \ {O}, i {1, . . . , s} and {C1 , . . . , Cs } is a complete set of
representatives for the ideal class group C of A. It was proved in 8.2.5 that
there is a finite
number of ideals of A of type I(P ) and all these ideals divide
g()A = 3 169A = 32 A.
3
It is clear that we need more information on C and U (A). For this we

have two lemmas.

Lemma
12.0.8. U (Z[ 3 13]) = {n0 | n Z} with 0 = 94 + 40 3 13 +
17 3 169.
Proof: Define N : K C by N (x) = 1 (x)2 (x)3 (x), where i : K

C are the field homomorphisms uniquely defined by i () = i1 . Notice
that 1 = 1K and 2 = 3 , where for a complex number z, z represents its
complex conjugate. It can be proved that N is a multiplicative function on
K with rational values and N (A) Z. Moreover N (x) = 0 x = 0. It is
easy to prove that u U (A) N (u) = 1. Again a simple computation
proves that if x = a + b + c2 , then N (x) = a3 + 13 b3 + 169 c3 3 13 abc.
N (0 ) = 943 + 13 403 + 169 173 3 13 94 40 17 = 830584 + 832000 +
830297 2492880 = 1, hence 0 is a unit. It follows that every number of
the form n0 is a unit in A.
Conversely, given u a unit in A we want to prove that u = n0 for a
suitable choice of the sign and for some n Z. For this we first prove that
there is no unit of A such that 1 < u < 0 . Assume there is such a unit of
the form u = a + b + c2 with a, b, c Z. We then have
1 (u) = a + b + c2

2 (u) = a + b + c 2 2 .
3 (u) = a + b 2 + c2

1 = |N (u)| = |1 (u)2 (u)3 (u)| = u|2 (u)|2 |2 (u)|2 = u1 < 1. Keeping

in mind that 2 = 3 and 1 + + 2 = 0, we have |3a| = |1 (u) + 2 (u) +
3 (u)| u + 2|2 (u)| < 0 + 2. Therefore
0 + 2
|a| < .
3
|3b| = |1 (u) + 2 2 (u) + 3 (u)| u + 2|2 (u)| < 0 + 2
0 + 2
|b| < .
3
Similarly, from 3c2 = 1 (u) + 2 (u) + 2 3 (u), we prove
0 + 2
|c| < .
32
A simple enough computer program can show that no element of the form
a + b + c2 in between 1 and 0 and with a, b and c subjected to the upper
restrictions is invertible in A.
Assume u is a unit of A. Exactly one of the units of A {u, u1 }
is greater or equal to 1. Without loss of generality, assume 1 u. Since
1 < 0 , limn n0 = 0. Choose n such that 0 < u 0
n1
1 < u n
0 .
It is easy to see that such n exists. We have proved that there is no unit of
A in (1, 0 ). Since 1 < u n n+1
0 0 , we conclude that u = 0 . The other
cases are treated similarly.

Lemma 12.0.9. The ideal class group C of A = Z[ 3 13] is isomorphic to
(Z/3Z, +).
119
Proof:
By Minkowskis Theorem 12.0.5, for every nonzero ideal M of
A = Z[ 3 13], there exists an ideal I of A such that M I(mod P rA) and
t
N (I) 4 nn!n |K |, where n = [K : Q] = [Q() : Q] = 3, t = 1 is half
p
the number of non-real embeddings of K in C and
T rK (1) T rK () T rK (2 )

3 0 0

K = det T rK () T rK (2 ) T rK (3 ) = det 0 0 39 = 27 169.
T rK (2 ) T rK (3 ) T rK (4 ) 0 39 0

Therefore N (I) 4 33!3 27 169 < 20. This means that every class of C
has a representative whose norm is at most 19.
We are now looking for ideals of A of norm least or equal to 19. By
using Theorem 12.0.6, we can find the prime ideals of A of norm at most 19.
We have A = Z[] who is the ring of integers of an algebraic extension of Q
3
of degree 3, K = Q( 13), and f = IrrQ () = X 3 13 so Theorem 12.0.6
3
applies merrily. Remember the notation = 13.
There is only one ideal of norm 1, and this is A.
Since for every ideal I of A, we have N (I) I N (I) A I
I| N (I) A, the ideals of norm 2 can be found among the divisors of 2A.
To decompose 2A as a product of prime ideals, we apply Theorem 12.0.6.
Modulo 2 we have the decomposition in irreducible factors f = X 3 13 =
2
(X 1)(X + X + 1). Therefore 2A = P1 P2 with P1 = 2A + ( 1)A,
N (P1 ) = 2 and P2 = 2A + (2 + + 1)A, N (P2 ) = 4. P1 and P2 are both
maximal ideals of A. Our main goal is to prove that A, P1 and P12 is a
complete set of representatives of C. So far we have P2 P11 (mod P rA).
Modulo 3, f = (X 1)3 , so 3A = P33 with P3 = 3A + ( 1)A and
N (P3 ) = 3. We have N (( 1)A) = |NK ( 1)| = | 1 + 13| = 12. Also
P3 | ( 1)A and P1 | ( 1)A. We conclude ( 1)A = P1 P3 Q for some
prime ideal Q of norm 2. Since P1 is the only such ideal, P12 P3 = ( 1)A
P3 P12 (mod P rA).
As a consequence of 8.1.2, the only ideals of norm 4 are P12 and P2 . For
the same purpose, at first, we will only be looking for the prime ideal of
norm at most 19.
The ideals of norm 5 are divisors of 5A. Modulo 5, f = X 3 13 =
2 2
(X 2)(X + 2X + 4) and X + 2X + 4 is irreducible modulo 5 because it
has no roots modulo 5, or because its discriminant, = 5, is not a square
modulo 5. By 12.0.6, 5A = P4 P5 with P4 = 5A + ( 2)A, N (P4 ) = 5 and
P5 = 5A + (2 + 2 + 4)A, N (P5 ) = 25. P4 and P5 are both maximal. Since
N (P5 ) = 25 > 19, P5 is not of great interest for us. We have N ( 2) =
8 + 13 = 5, so ( 2)| 5 in A, hence P4 = ( 2)A and P4 A(mod P rA).
Since N (( + 3)A) = |N ( + 3)| = |27 + 13| = 40 = 23 5, ( + 3)A = P4 Q,
where Q is an ideal of norm 8. But the only ideals of norm 8 are P13 and
P1 P2 . If Q = P1 P2 = 2A, then 2A| ( + 3)A 2| + 3 which is impossible.
Therefore ( +3)A = P4 P13 A P4 P13 (mod P rA) A P13 (mod P rA).
Modulo 7, f = X 3 + 1 = (X + 1)(X + 2)(X 3), so 7A = P6 P7 P8

with P6 = 7A + ( + 1)A, P7 = 7A + ( + 2)A and P8 = 7A + ( 3)A
are all maximal ideals of norm 7 in A. N (( + 1)A) = N ( + 1) = 14 and
P6 | ( + 1)A imply ( + 1)A = P6 P1 P6 P11 (mod P rA). Similarly
( + 2)A = P7 P3 P7 P31 P12 (mod P rA) and ( 3)A = P8 P1
P8 P11 (mod P rA).
Modulo 11, f = X 3 2 = (X + 4)(X 2 4X + 5) and X 2 4X + 5 is
irreducible modulo 11 because its discriminant modulo 11 is 7 which is not
a square modulo 11. So 11A = P9 P10 with P9 = 11A + ( + 4)A a maximal
ideal of norm 11 and P1 0 a non-interesting maximal ideal of norm 121. We
have P9 | ( + 4)A and P8 | ( + 4)A because P8 = 7A + ( 3)A + 4 =
7 + ( 3) P8 . Since N (( + 4)A) = |N ( + 4)| = 64 + 13 = 77, the only
possibility is ( + 4)A = P8 P9 P9 P81 P1 (mod P rA).
Modulo 13, f = X 3 13A = P11 3 with P
11 = A of norm 11 and
P11 A(mod P rA).
Modulo 17, f = X 3 + 4 = (X 4)(X 2 + 4X + 16) with X 2 + 4X + 16
irreducible modulo 17. Then 17A = P12 P13 with P12 = 17A + ( 4)A,
N (P12 ) = 17 and P13 is a non-interesting maximal ideal of norm 172 > 19.
As before, we can prove ( 4)A = P3 P12 P12 P31 P12 (mod P rA).
Modulo 19, f = X 3 13 is irreducible because 13
is not a cube mod 19,
3
so 19A = P14 is a maximal ideal of norm 19 > 19.
We have proved that all the prime ideals of norm at most 19 are equiva-
lent modulo P rA to a power of P1 and we can easily prove the same result for
every ideal, not necessarily prime, whose norm is least or equal to 19. With
a little help from Minkowskis 12.0.5 we can remove the restriction on the
bound of the ideal from the previous sentence and conclude that P rA is gen-
erated by the class of the ideal P1 . We have proved that P13 A(mod P rA),
so P1 has either order 3 or 1. If P1 A(mod P rA) then P1 is principal, so
there exists x A such that xA = P1 . Let x = a + b + c2 with a, b, c Z.
Then 2 = N (P1 ) = N (xA) = |N (x)| = |a3 + 13b3 + 169c3 39abc|. Modulo
13 we get 2 a3 (mod 13). But the cubes modulo 13 are {0, 1, 8} and
2 is not among them. Therefore the equality 2 = |N (x)| is impossible and
P1 is not principal. We conclude that P rA ' (Z/3Z, +).
Back to our problem, we wanted to describe Im and for this we saw

that we must describe the set of s. With the notations in the proof
of Lemma 12.0.9, all the ideals I(P ) appear among divisors of g()A =
2 . Assume P | I(P ) = ( )A + 2 g for some
3 3 169A = P33 P11 11

P C(Q) with P = , w , , Z, > 0 and gcd(, ) = 1. Then
13 = N (P11 )| N (I(P ))| N (( )A) = |N ( )| = |3 13 3 |.
From 2.2.5, there exist , t Z with t > 0, gcd(, t) = 1 such that w = t3
3 2
and = t2 . Using that P C(Q), we get 3
13 = t6
3 13 3 = 2 .
121
Since 13| |3 13 3 |, 13| 2 13| 13| 13| which contradicts

gcd(, ) = 1.
Assume P3 | I(P ). Just as before we get 3| |3 13 3 | = 2 3| . It
is easy to see that 3| 3| . Using that gcd(, ) = 1, we get that 3 6 |
and 3 6 |. Since 3 13 3 3 (mod 3), 1(mod 3)
or 1(mod 3). If 1(mod 3), then 3 1(mod 9) and then
0 2 = 3 13 3 1 13 = 12(mod 9) which is impossible. Similarly if
1(mod 3) we reach a contradiction.
We have proved that P11 and P3 do not divide I(P ) for any P C(Q) \
{O}. Since I(P )| P33 P11
2 , we obtain I(P ) = A for every P as above. The s
are now selected as generators for the principal ideals in the set {A, P12 , P14 }.
Only A is principal in this set, so there is only one to choose and we choose
it to be 1. It follows that for every choice of P as above, = u 2
for some u U (A) and K . Then (P ) = \ = u[ 2 = u
{1, 0 }. Assume (P ) = 1. c Then = 2 for some K .
Then N ( ) = N ( ) = N ( )2 . But N ( ) = 3 13 3 = 2
2
0. We have reached a contradiction. Similarly (P ) 6= d0 . We have proved

Im {1, 0 }. Therefore |Im| 2 and r 1. If r = rank(C(Q)) = 0,
then C(Q) = {O} which contradicts (17, 70) C(Q).
We have completed the proof for rank(C(Q)) = 1.
Chapter 13
Lecture XIII
We have reserved this lecture for the proof of the problem 11.1.2, related
to Thues Theorem 11.1.1. We will present it in an enriched form that will
provide us a rare example of an effective method of determining the solutions
of a diophantine equation.
Theorem 13.0.10. The equation x3 dy 3 = 1, where d is a positive inte-

ger that is not the cubic power of another integer, has at most two integer
solutions. The equation always has the obvious solution (1, 0).
3
Let A = Z[ d]. Then the units of A are U (A) = {n0 | n Z} for some
0 < 0 < 1, 0 A.
The equation
X 3 dY 3 = 1 has a non-trivial solution if and only if
0 = a + b d for some a, b Z .
3
Remark 13.0.11. The condition d N is not important because we can

reduce to this case anyway by replacing y with y if necessary. The same
substitution proves a strong connections between the equations x3 dy 3 = 1
and x3 + dy 3 = 1.
Before proving this theorem, we give an example to illustrate its strength.
Example 13.0.12. The equation x3 9y 3 = 1 has an obvious solution i.e.

(1, 0) and a visible second solution i.e. (2, 1). By applying 13.0.10, we
obtain that these are the only integer solutions of the equation.

We make some notations and easy remarks first. Let = 3 d. On
K = Q() we define a function N : K C by
N (a + b + c2 ) = a3 + b3 d + c3 d3 3abcd
for all a, b, c Q. It is obvious that N (K) Q and N (A) Z. It is easy to

prove that
N (x) = 1 (x)2 (x)3 (x),
123
124 CHAPTER 13. LECTURE XIII
where : K C are the field homomorphisms uniquely determined by

2i
i () = i1 for all i = 1, 3, where = e 3 . With this remark it follows
immediately that N is a multiplicative function on K and on A. Also, it is
easy to prove that u A is invertible in A if and only if N (u) = 1. Notice
that 1 = 1K and 2 = 3 , where z is the complex conjugate of the complex
number z.
Lemma 13.0.13. Let = P + Q + R2 such that U (A) and || > 1.

Then QR 6= 0.
Proof: Assume QR = 0. Then Q = 0 or R = 0. Eventually by changing

by , we can assume N () = 1.
We treat the case R = 0, the case Q = 0 being treated similarly. We
have 1 = N () = P 3 + dQ3 . It is easy to see that P and Q are nonzero and
cannot have the same sign, hence P Q < 0. From this, we have
1 1 1
= P + Q = < < 1.
P2 2
P Q + Q 2 1++ 2 3
Since || > 1, we obtain < 1. But N () = 1 1 ()2 ()3 () = 1

|2 ()|2 = 1. It follows that is positive which is a contradiction.
Lemma 13.0.14. 1. If there exist x, y Z and n Z such that
(x + y)n = a + b U (A)
for some a, b Z, then |n| 1.
2. If there exist x, y Z and n Z such that
(x + y2 )n = a + b U (A)
for some a, b Z, then n = 0.
Proof: Because the proofs of the two statements are similar, we will only
prove the first. We assume that d 3 and solve the case d = 2 separately
in the last part of the proof.
Assume there exist x, y Z and n Z with |n| 2 and (x + y)n =
a + b U (A). Then obviously x + y U (A). We can eventually replace
x, y by x, y to assume x3 + dy 3 = N (x + y) = 1. From Lemma 13.0.13,
|a + b| 1 and |x + y| 1. Their absolute value cannot be 1 because of
the assumption x, y Z . (x + y)n = a + b, |x + y| < 1 and |a + b| < 1
imply n > 0, so n 2.
We have 1 = N (x + y) = (x + y) |2 (x + y)|2 , so x + y > 0. Since
a+b = (x+y)n , we get a+b > 0. If |x| 1, then dy 3 = 1x3 |dy 3 | 2.
From the assumption d 3, we obtain y = 0. Since x = x + y U (A),
x = 1 and we contradict |x + y| < 1. Therefore |x| 2.
125
We have (x + y)n = An + Bn + Cn 2 for some An , Bn , Cn Z. Since

{1, , 2 } is a basis for K = Q() seen as a vector space over Q, An , Bn and
Cn are uniquely defined. By identifying coefficients in (x + y)n = a + b,
using Newtons binomial expansion formula and n 2, we get Cn = 0
X n
xn3k2 y 3k+2 dk .
3k + 2
03k+2n
Since we have proved that |x| 2, there exists a prime number p dividing
x.
According to the mod 3 residue of n, we have three cases.
If n 2(mod 3), then

n n2 2 n n5 5 n n n2
x y + x y d + ... + y d 3 = 0.
2 5 n
n2
Reducing modulo x we see that x|y n d 3 . But x3 + dy 3 = 1 implies that x
and dy are coprime, so we obtain |x| = 1 which contradicts |x| 2.
If n 1(mod 3), then we can write the condition Cn = 0 as

n n2 n n5 5 n 2 n2 n4
x y+ x y d + ... + x y d 3 = 0.
n2 n5 2
We can divide by x2 which is nonzero for

n n4 n n7 5 n 3 n5 n n2 n4
x y+ x y d+...+ x y + y d 3 = 0.
n2 n5 5 2
Reducing modulo p and keeping in mind that x and dy are coprime (from
n a | n and pa+1 6 | n
x3 + dy 3 = 1), we obtain p| 2 . Let a N such that p 2 2
i.e. 0 < a = vp n2 . We prove that pa+1 | 3k+2 n
3k n3k2
dk for all k 1

x y
i.e. pa+1 divides all the terms different form the last in the above sum. The
n4
last term is n2 y n2 d 3 and it is divisible by pa , but not by pa+1 because

gcd(x, dy) = 1 and p| x. So if we manage to prove what we planned, then

we obtain a contradiction modulo pa+1 in the second combinatorial sum
n n
above. It is enough to prove that pa+1 | 3k+2
3k
x for all k 1. 3k+2 =
n! n(n1) (n2)! 2
(3k+2)!(n3k2)! = 2 (3k)!(n23k)! (3k+1)(3k+2) , so

n n n2 2
x3k = x3k .
3k + 2 2 3k (3k + 1)(3k + 2)
It is enough to prove p| n2 2 3k 3k 3k

3k (3k+1)(3k+2) x . For this, since p | x , we
prove p3k 6 | (3k + 1)(3k + 2). Assume p3k | (3k + 1)(3k + 2). Then p3k | 3k + 1
or p3k | 3k + 2 because gcd(3k + 1, 3k + 2) = 1. But an easy induction proves
p3k > 3k + 2 > 3k + 1 for every prime p 2 and k 1, so we get our long
sought contradiction and we are done with this case.
If 3| n, then like before,

n n3 2 n n1 n2
x y + ... + y d 3 = 0.
n2 1
Reducing modulo p, we obtain p| n. Just like before, for a = vp (n) we prove

n
that pa+1 | 3k+1 x3k for all k 1 and we obtain a contradiction modulo
pa+1 in the combinatorial sum above.
The proof of the lemma is complete if we solve the case d = 2. By 6.2.4,
the only integer solutions to x3 + 2y 3 = 1 are (1, 0) and (1, 1). This time
we prove both parts of the lemma.
If (x + y)n = a + b U (A) = U (Z[ 3 2]) with x, y Z and n Z,
|n| 2, then x + y U (A), so x3 + 2y 3 = 1. We have seen far too many
times how we can assume
x3 + 2y 3 = 1. But x and y are nonzero integers,

so x + y = 1 + 2. Similarly a + b {1, 1 + 3 2}. (1 + 3 2)n
3
{1, 1 + 3 2} contradicts |n| 2.

If (x + y2 )n = a + b U (A) with x, y Z and |n| 1, then x + y2
U (A), so N (x + y2 ) = 1 x3 + 4y 3 = 1. From 6.2.5 we get x = 1
and y = 0 which contradicts x, y Z .
The following two lemmas will help us prove Theorem 13.0.10, but before
dealing with them, we will see how they apply.
Lemma 13.0.15. If there exists U (A) such that 2 = a + b for some

a, b Z, then = 1.
Lemma 13.0.16. If there exists U (A) such that 3 = a + b for some

a, b Z, then = 1.
We prove the following form of 13.0.10:
Theorem 13.0.17. Let d be a positive integer which is not a cube. Let

x, y Z such that x3 + dy 3 = 1. Then there is no other integer solution of
X 3 + dY 3 = 1 with XY 6= 0. Moreover U (A) = {(x + y)n | n Z}.
Proof: Since N (x + y) = x3 + dy 3 = 1, (x + y)|2 (x + y)|2 = 1, so

x + y > 0. By 13.0.13, 0 < x + y < 1. With similar methods to the ones
used in 12.0.8, it can be proved that there exists
0 = max{ U (A)| x + y < 1}
and that
U (A) = {n0 | n Z}.
Since x + y U (A), there exists n Z such that x + y = n0 . Since
x + y and 0 are positive, the sign is actually + . Since x + y and
0 are strictly smaller than 1, n > 0. From 13.0.15 and 13.0.16, n is odd
and not a multiple of 3.
127
Let 0 = P + Q + R2 . We have 1 = N (0 ) = 1 (0 )2 (0 )3 (0 ) =
0 |2 (0 )|2 > 0 N (0 ) = 1
P 3 + dQ3 + d2 R3 3dP QR = 1 (13.0.1)
Let = 2 (0 ) and = 3 (0 ). Then n = x + y and n = x + y 2

because i is a field homomorphism for all i = 1, 3. n0 + n + 2 n =
(1 + + 2 )(x + y) = 0 n + 2 n = n0 U (A).
If n 2(mod 3), then n = 2 and 2n = . U (A) 3 n0 = n +
n = ( 2 )n + ()n
2
( 2 + )(( 2 )n1 ( 2 )n2 () + . . . + ()n1 ) = n0 . (13.0.2)
2 + = 2 2 (0 ) + 3 (0 ) = 2 (P + Q + R 2 2 ) + (P + Q 2 +
R2 ) = P + 2Q R2 A. It is clear that
( 2 )n1 ( 2 )n2 () + . . . + ()n1 Z[, ].
We want to prove that it actually belongs to A. For this it is enough to

prove that it is a real number since A = Z[] = Z[, ] R. But
n1
X
2 n1 2 n2 n1
( ) ( ) ()+. . .+() = (1)k ( 2 2 (0 ))k (3 (0 ))n1k =
k=0
n1
X
= (1)k (2 (0 ))k (2 (0 ))n1k .
k=0
n1
X n1
X
(1)k ( 2 (0 ))k ( 2 (0 ))n1k = (1)k (2 (0 ))k (2 (0 ))n1k =
k=0 k=0
n1
X n1
X
= (1)n1k (2 (0 ))k (2 (0 ))n1k = (1)k (2 (0 ))k (2 (0 ))n1k ,
k=0 k=0
hence ( 2 )n1 ( 2 )n2 () + . . . + ()n1 A. We have used here

that n is odd, so n 1 is even. From the equation 13.0.2 it now follows that
2 + = P + 2Q R2 U (A),
so 1 = N (P +2Q R2 ) = P 3 +8dQ3 d2 R3 6dP QR. If we add this

to the equation 13.0.1 for N (0 ) = 1, we get 9dQ3 9dP QR {0, 2}. Since
3 3
side is a multiple of 9, we get 9dQ 9dP QR = 0 Q =
the left hand
Q=0
P QR or . If Q = 0, then P Q 6= 0 because 0 is invertible and
2
Q = PR
contained in [x + y, 1). (P + R2 )n = x + y and n > 0 now contradict the
second part of Lemma 13.0.14. Therefore Q 6= 0 and Q2 = P R.
We have 1 < 1 2 2 2
0 = = (P + Q + R )(P + Q + R ) =
2
P2 2 2 2 2 2
+ Q + dR P Q P R dRQ = (P dRQ) + (dR P Q) + 2
(Q2 P R)2 = (P 2 dRQ) + (dR2 P Q), which contradicts Lemma

13.0.13.
If n 1(mod 3), then n = and 2n = 2 . Just like before we obtain
+ 2 = P Q + 2R2 U (A), so
P 3 dQ3 + 8d2 R3 6dP QR = 1.
By adding to the equation 13.0.2, we get 9d2 R3 9dP QR {0, 2} dR3 =
P QR. If dR2 = P Q, then 1 < 1 2 2
0 = (P dRQ) + (dR P Q) + (Q
2
2 2 2 2
P R) = (P dRQ) + (Q P R) which contradicts Lemma 13.0.13.
Therefore R = 0. Since n0 = (P + Q)n = x + y, by the first part of
Lemma 13.0.14 and by n > 0, we get n = 1 which means x + y = 0 . We
have proved that U (A) = {(x + y)n | n Z}.
If u + v is another solution of X 3 + dY 3 = 1 with uv 6= 0, then just like
before we prove that u + v = 0 . Therefore u + v = x + y, so (x, y) is the
only solution of the equation X 3 + dY 3 = 1 with x, y Z .
Proof of Theorem 13.0.10: The theorem above proves the hardest

part of Theorem 13.0.10. It proves that if the equation X 3 dY 3 = 1 has a
non-trivial solution (x, y), then 0 = x y and that the non-trivial solution
is unique.
Conversely, if 0 = a + b for some a, b Z , then a3 + db3 = N (0 ) = 1,
so (a, b) is a non-trivial solution to the equation and by applying Theorem
13.0.17 we prove that it is unique.
Proof of Lemma 13.0.15: Assume there exists U (A) \ {1} such

that 2 = u + v for some u, v Z. Let = P + Q + R2 . We can assume
N () = 1 which means
P 3 + dQ2 + d2 R3 3dP QR = 1 (13.0.3)
2 = u + v
2P R + Q2 = 0 (13.0.4)
i.e. the coefficient of 2 in 2 is zero. If gcd(P, R) 6= 1, then there exists a
prime number p dividing both of them. From the equation 13.0.4 it follows
that p| Q (even if p = 2). But then p| in A which is impossible since
is invertible, but N (p) = p3 > 1 implies that p is not invertible. Therefore
gcd(P, R) = 1.
If P R = 0, then 13.0.4 implies Q = 0 and we have the possibilities
(P, Q, R) = (0, 1, 0) or (P, Q, R) = (1, 0, 0). We use here gcd(P, R) = 1.
13.0.3
(P, Q, R) = (0, 1, 0) = d2 = 1 which is impossible because d is not a
13.0.3
cube. (P, Q, R) = (1, 0, 0) = = 1 which contradicts the assumption
U (A) \ {1}. Therefore P R 6= 0.
129
From the equation 13.0.4 we have that there exist a, b Z such that
Q = 2ab and one of the following occurs:
If P = a2 and R = 2b2 , then from 13.0.3, a6 + 8da3 b3 + 8d2 b6 +
12da3 b3 = 1 which yields a contradiction modulo 4. Similarly we prove that
the case P = 2a2 and R = b2 is contradictory.
If P = a2 and R = 2b2 , then a6 + 8da3 b3 8d2 b6 + 12da3 b3 = 1.
Set x = db3 . Then 1 = a6 + 20a3 x 8x2 (3a2 )3 (4x 5a3 )2 = 2.
It is known that the integer solutions of Fermats equation, 2 + x2 = y 3 ,
are (x, y) (5, 3). It follows that a2 = 1 and 4x 5a3 = 5 4x
{0, 10} x = 0 db3 = 0 b = 0 Q = R = 0 and P = 1 which
contradicts 6= 1.
If P = 2a2 and Q = b2 , then 1 = 8a6 + 20da3 b3 + d2 b6 . For x = db3 ,
we have x2 + 20a3 x 8a6 (x + 10a3 )2 108a6 = 1. Let t = x + 10a4 and
s = 3a2 . Then t2 1 = 4s3 t1 t+1 3 t1 t+1
2 2 = s . Since gcd( 2 , 2 ) = 1 we
t1 t+1
have 2 = A and 2 = B for some A, B Z. Then A + 1 = B 3 which
3 3 3
implies (A, B) {(0, 1), (1, 0)} t = 1 and 3a2 = s = 0 a = 0 which

implies P = Q = 0. But we have proved P R 6= 0, so we have found again a
contradiction.
Similar methods can be used for a proof of Lemma 13.0.16. With the
same notations as in the later proof, the equation 13.0.4 is replaced by
P 2 R + P Q2 + dQR2 = 0.
It must be proved that = 1 is the only solution of this equation such that
P 3 + dQ3 + d2 R3 3dP QR = 1 i.e. N () = 1. The proof uses that the only
integer solutions of the equation x3 9y 3 = 1 are (x, y) {(1, 0), (2, 1)}.

Alexandru Gica - Rational Points On Elliptic Curves PDF

Uploaded by

Copyright:

Available Formats

Alexandru Gica - Rational Points On Elliptic Curves PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Alexandru Gica - Rational Points On Elliptic Curves PDF

Uploaded by

Copyright:

Available Formats

Rational Points on Elliptic Curves

10 An unexpectedly hard problem 93

11 Integer points on elliptic curves 101

12 Generators for Elliptic Curves 113

13 Lecture XIII 123

Definition 1.0.1. Let f Q[X, Y ] be a polynomial of degree 3 i.e. there

1.1 Projective Geometry

and call it the n-th dimensional projective space over K.

1.1.1 Equivalent definitions

Notice that P0 is the set of lines A1 . But A1 is a line, hence P0 consists

1.1.2 The Geometry of the Projective Plane

Proposition 1.1.10. Let H be a subset of Pn . Then H is a hyperplane if

1.2 The Group Law of a Cubic

Definition 1.2.3. Let C be the affine rational cubic determined by f (X, Y ) =

C is called nonsingular or smooth if it is nonsingular at every point.

Theorem 1.2.6 (Bezout). : Let F1 , F2 C[X, Y, Z] be homogeneous poly-

Corollary 1.2.7. Two complex cubics with no common component intersect

Theorem 1.2.8 (9 Points Theorem). Let C1 and C2 be two cubics with

Proof: The last 3 statements are powerful results of Algebraic Geom-

Definition 1.2.9. Let C be a complex nonsingular cubic. Let P, Q C. If

Proof: To prove that + is well defined it suffices to prove that

1.2.1 The choice of origin

Theorem 1.2.14. G and G0 are isomorphic as abelian groups.

Proof: Let P, Q C be arbitrary points. We prove that P Q =

1.3 The Weierstrass Normal Form

Definition 1.3.1. The affine rational curve C(Q) defined by F (X, Y ) = 0,

Remark 1.3.2. The projective closure C(Q) of C(Q) is given by F (X, Y, Z) =

because [0 : Y : 0] = [0 : 1 : 0] Y 6= 0. Notice that O is also a point of

Proposition 1.3.4. C is smooth if and only if all the complex roots of f

1.3.1 Explicit formulas for the group operation on C(Q)

Proposition 1.3.5. Let P (x, y) C. Then TP C = TP C.

Proof: The equation of TP C is obtained by homogenizing the equa-

If P = Q, then TP C is given by f 0 (x1 )(x x1 ) + 2y1 (y y1 ) = 0. If

Example 1.3.10. Let C(Q) be the affine rational cubic of equation y 2 =

Solution: P C(Q) 82 = 33 43 3 + 166 64 = 27 129 + 166.

Exercise 1.3.11. Let C(Q) be the elliptic curve defined by y 2 = x3 + 17.

Example 1.3.12. Let C(Q) be the affine rational cubic defined by x3 +y 3 =

1.3.2 The existence of a Weierstrass Normal Form

Let M = (aij )i,j=1,3 . Then M is a linear automorphism of A3 which

From now on we identify A with M . A is bijective and its inverse is the

(zj zjk )ij + (zk zjk )ik = 0, i, j, k = 1, 3, j < k.

Choosing i = j gives zj = zjk for j < k. Choosing i = k gives zk = zjk for

(P 0 , Q0 , R0 ) in P, there exists A P GL2 (C) such that A(P ) = P 0 , A(Q) =

Proposition 1.3.21. For every two lines l1 , l2 in P2 there exists a projective

Proof: This is a consequence of the previous exercise, but it also has a

Proof: By the previous problem there exists a projective transformation

So far we have only treated classical topics of projective geometry. The

Definition 1.3.23. Let C and C 0 be two projective curves. We say that

Remark 1.3.24. Let C be a complex cubic, P GL2 (C) and let A

I said that the definition of the multiplicity of intersection of two pro-

Definition 1.3.25. Let C be a curve in P2 given by the equation F (X, Y, Z) =

Remark 1.3.26. 0 is in fact a root of f in the equation above because

Let f () = F (P + Q) = F (P ) + R(P, Q) + 2 H(P, Q) for some

Definition 1.3.27. Let C be a projective curve of degree d and let P C.

Remark 1.3.28. TP C is either a line or the the whole P2 . It is P2 if and

Exercise 1.3.29. Let C be a complex curve of degree d, let l P2 be a line

We now introduce the notion of a flex point.

Definition 1.3.30. Let P be a smooth point on a complex curve C of degree