Notes On Diophantine Geometry: Rational Points On Curves of Genus Zero

Notes on Diophantine Geometry
Felipe Voloch and students
June 5, 2008
Rational Points on Curves of Genus Zero
An algebraic set over a field K is (the solution set of) a system of equations:

 f1 = 0

X: .. .
 .
fm = 0

with f1 , . . . , fm ∈ K[x1 , . . . , xn ].
Let X be an irreducible algebraic set, i.e., (f1 , . . . , fm ) is a prime ideal.
Next, let R = K[x1 , . . . , xn ]/(f1 , . . . , fm ) and let K(X) denote the field of
fractions of R. Then the dimension of X is the transcendence degree of
K(X) over K:
Tr degK K(X) = dim X.
X is a curve if dim X = 1, that is, if some xi is not in K, and all other xj
are algebraic over K(xi ).
We say that X, Y , irreducible algebraic sets, are birationally isomorphic
if their function fields K(X), K(Y ) are isomorphic as field extensions of K.
Example 1. If n = 2 and m = 1, f (x1 , x2 ) = 0, then this is a curve.
Now, we define the genus g(X) of an irreducible curve X. Over a subfield

of C, the genus can be obtained from the topology of X(C). An algebraic
definition of genus is the dimension of the space of regular 1-forms on a
smooth projective model. If X is a plane curve of degree d, then we have
the following formula for the genus:
1 X
g(X) = (d − 1)(d − 2) − δp ,
2
p singular
1
where, if the singularity is an ordinary singularity with m branches, then
δp = 12 m(m − 1). (So, δp depends on the singularity. There is an algorithm
for computing it in general, but you won’t find it in these notes.)
Background. Singular points are, for example, points (x, y) ∈ C2 with
∂f ∂f
f (x, y) = (x, y) = (x, y) = 0.
∂x ∂y
Example 2. For smooth (projective) plane curves of degree d, the genus is

1
2 (d − 1)(d − 2). In particular
g = 0 ⇐⇒ d = 1, 2
g = 1 ⇐⇒ d = 3
g > 1 ⇐⇒ d ≥ 4.
Definition. An irreducible curve X is parametrizable (also, rational ) over

K if there exist ϕ1 , . . . , ϕn ∈ K(t) (not all constant) such that for all j =
1, . . . m we have fj (ϕ1 , . . . , ϕn ) = 0.
Theorem 1. A curve is parametrizable over K if and only if it has genus

zero and a smooth point with coordinates in K.
Proof. (⇒) If X is parametrizable then K(X) embeds in K(t). This is seen

by considering the map α : K[x1 , . . . , xn ] → K(t) which sends xi 7→ ϕi (t).
Then fi ∈ ker α, so that α induces a map R = K[x1 , . . . , xn ]/(f1 , . . . , fm ) →
K(t). Thus, by Luroth’s theorem, K(X) is purely transcendental over K.
X is therefore birationally isomorphic to a line, and so has genus zero and
many smooth points in K.
(⇐) A sketch of the proof in this direction: Use g(X) = 0 together with
the smooth point P and the Riemann-Roch space L(P ) of functions whose
only pole is a simple pole at P to get that dim L(P ) = 2. Then, we find a
non-constant f ∈ L(P ), where f : X → P1 has only one (simple) pole, so it
has degree 1, making it a birational isomorphism. Thus, f −1 : P1 → X is a
parametrization.
Example 3. Consider the curve X : x2 + y 2 = 1. Then X has a point at

P = (1, 0). By the degree formula for the genus, g(X) = 0. We will give a
parametrization of X working geometrically: Any line through P intersects
2
the curve in exactly one other point (unless it is tangent at P ). The equation
of a line through (1, 0) is given by y = t(x − 1). Substituting this into X for
2 −1
y and solving for x gives: x = tt2 +1 . Then, y = t(x − 1) = t−2t
2 +1 . This is our
parametrization.
Assume K is perfect (e.g. has characteristic zero). By the genus for-

mula and the fact that the genus is non negative, a cubic has at most one
singularity. If a curve has a singular point in a field bigger than K, then
its conjugates are also singular points so the curve must have other singular
points. Therefore, we may conclude that an absolutely irreducible cubic with
a singular point automatically has its singular point with coordinates in the
ground field. The method used in the last example to find a parametrization
works in this instance, too. You take the pencil of lines through the singular
point, and because it is singular, this will give you a parametrization, hence
lots of points. That is we get the following:
Corollary 2. A singular cubic always has lots of points over the ground
field.
Example 4. Consider X : y 2 = x3 + x2 . Then X has a singular point

at the origin P = (0, 0). The equation of a line through (0, 0) is given by
y = tx. Then, we get x = t2 − 1, and so y = t(t2 − 1).
Example 5. But, x2 +y 2 = −1 has no solutions over R. So, a smooth curve

of degree 2 and genus 0, may have no points over Q or even R. However, it
does have points over C.
Example 6. Another
S example is x2 + y 2 = 0. But, this curve is reducible:
X = {x = iy} {x = −iy}.
Theorem 3. Every curve of genus zero over K is birationally isomorphic

to a conic.
Proof. Ww will not give a proof, just remark that the proof uses the negative
of the canonical divisor, which gives us a divisor of degree two on the curve.
3
A conic is an absolutely irreducible curve of degree 2. Given by f (x, y) =
0 where deg f = 2.
If we are looking for rational solutions of an equation f (x, y) = 0 where

deg f = d, we can also look at the homogenous polynomial of degree d in
x, y, z given by z d f (x/z, y/z). We have to be careful with z = 0.
A conic can be further simplified by diagonalization. So we can reduce the
study of conics to equations of the form ax2 + by 2 + cz 2 = 0 where absolute
irreducibility is equivalent to abc 6= 0.
If a, b, c ∈ Q and we want to find out if ax2 + by 2 + cz 2 = 0 has solutions,
we can assume that a, b, c ∈ Z by clearing denominators. We can also
assume that a, b, c have no common factors.
If a = m2 · a0 then ax2 + by 2 + cz 2 = 0 has a solution if and only if
a0 x2 + by 2 + cz 2 = 0 has a solution (replace x by mx). In this way we
can assume that a, b, c are square free.
Suppose there exists a prime p such that p divides both a and b. Then
a = a0 · p and b = b0 · p. If ax2 + by 2 + cz 2 = 0 has a solution then
pa0 x2 + pb0 y 2 + cz 2 = 0
a0 (px)2 + b0 (py)2 + pcz 2 = 0

has a solution. Proceeding in this way, I can eliminate common factors of
any two of a, b, c.
Legendre’s Theorem. Suppose that a, b, c ∈ Z are nonzero, square free,
and pairwise coprime. Then the equation ax2 + by 2 + cz 2 = 0 has a solu-
tion in Z3 −{(0, 0, 0)} if and only if the following two conditions are satisfied.
(i) a, b, c are not all of the same sign.
(ii) For all odd primes p | abc there is a solution to ax2 + by 2 + cz 2 ≡

0 mod p where all of x, y, z 6≡ 0 mod p
Proof. (⇑) If p | abc assume that p | a. We are asuming that ∃ x0 , y0 , z0 ≡6

0 mod p with ax20 + by02 + cz02 ≡ 0 mod p which is equivalent to by02 + cz02 ≡
0 mod p. Let
!2
b z0
u2 = − =
c y0
4
As a polynomial,
ax2 + by 2 + cz 2 ≡ by 2 + cz 2 mod p
≡ c(z 2 − u2 y 2 ) mod p
≡ c(z − uy)(z + uy) mod p
So for every odd prime p | abc there are linear forms Lp and L0p in x, y, z with
ax2 + by 2 + cz 2 ≡ Lp L0p mod p
Also, for p = 2,
ax2 + by 2 + cz 2 ≡ (ax + by + cz)2 mod 2
So there is also L2 and L02 such that
ax2 + by 2 + cz 2 ≡ L2 L02 mod 2
By the Chinese Remainder Theorem, there exist linear forms L and L0 such
that
ax2 + by 2 + cz 2 ≡ LL0 mod abc
Consider integers x0 , y0 , z0 that satisfy the inequalities

p p p
0 ≤ x0 ≤ |bc|, 0 ≤ y0 ≤ |ac|, 0 ≤ z0 ≤ |ab|
Since a, b, c are square free and coprime, we actually have
p p p
0 ≤ x0 < |bc|, 0 ≤ y0 < |ac|, 0 ≤ z0 < |ab|
How many such triples are there? The answer is greater than
! ! !
jp k jp k jp k
1+ |bc| 1+ |ac| 1+ |ab| > |abc|.
5
By the Pigeonhole Principle there exists distinct such triples (x0 , y0 , z0 )
and (x00 , y00 , z00 ) with
L(x0 , y0 , z0 ) ≡ L(x00 , y00 , z00 )mod abc
which implies that
L(x0 − x00 , y0 − y00 , z0 − z00 ) ≡ 0 mod abc
and consequently
a(x0 − x00 )2 + b(y0 − y00 )2 + c(z0 − z00 )2 ≡ 0 mod abc
Let x = x0 − x00 , y = y0 − y00 and z = z0 − z00 . Then

p p p
x< |bc|, y < |ac|, z < |ab|
and
ax2 + by 2 + cz 2 ≡ 0 mod abc
By condition (i) we may suppose without a loss of generality that a < 0, b <
0, and c > 0. This implies that |abc| = abc. As a result
ax2 + by 2 + cz 2 ≤ cz 2 < abc
−(|a|x2 + |b|y 2 ) ≤ ax2 + by 2 + cz 2
−2abc < −(|a|x2 + |b|y 2 )

so that
−2abc < ax2 + by 2 + cz 2 < abc
and
ax2 + by 2 + cz 2 = 0 or − abc
6
If ax2 + by 2 + cz 2 = 0, then we are done. We thus suppose that
ax2 + by 2 + cz 2 = −abc
ax2 + by 2 + c(z 2 + ab) = 0
Since a and b are squarefree and coprime, z 2 + ab 6= 0. The proof of the ⇑

direction is complete by the following polynomial identity.
a(xz +by)2 +b(yz −ax)2 +c(z 2 +ab)2 = (ax2 +by 2 +c(z 2 +ab))(z 2 +ab) = 0
Proof. (⇓) Suppose that ax2 + by 2 + cz 2 = 0 where x, y, z ∈ Z are not

all zero. Let p be an odd prime. If p divides x, y and z then
!2 !2 !2
x y z
a +b + = 0
p p p
We may thus suppose that p does not divide one of x, y, or z. If p divides

x and y but not z then it follows that p2 divides cz 2 and therefore that p2
divides c. c was assumed to be square free. This is a contradiction. So p
can divide at most one of x, y, and z.
Suppose that p divides a and x but not y or z. From ax2 + by 2 + cz 2 = 0
we have
a12 + by 2 + cz 2 ≡ 0 mod p
Suppose that p divides a and y but not x or z. Then
0 = ax2 + by 2 + cz 2 ≡ cz 2 mod p
So p divides c. This is a contradiction since a and c are coprime.
New Topic: P-Adic Numbers

Let p be a prime number. For x ∈ Q, x 6= 0 we can write
!
a
x = pr ·
b
7
with a, b, r ∈ Z and p - ab.
The p−adic norm of x is defined as |x|p = p−r and |0|p = 0.

The p−adic distace from x to y is defined as dp (x, y) = |x − y|p .
Exercise: Prove that dp is a metric.
Qp is the completion of Q with respect to the metric dp .

Zp is the completion of Z with respect to the metric dp .
Exercise: Show that Qp is a field, Q ⊆ Qp and Zp is a ring, Z ⊆ Zp .
Every element of Qp is of the form
∞
X
x = an pn
n=n0
where an ∈ { 1, . . . , p − 1 }, an0 6= 0 and |x|p = p−n0 .

We note the ultrametric inequality. For x, y ∈ Qp ,
|x + y|p ≤ max{ |x|p , |y|p }
Zp is a local ring with maximal ideal (p).

1 ∞
pn ∈ Zp . So 1 − p is a unit.
P
=
1−p n=0
Exercise: Show that for all a ∈ Z, p - a, a is a unit in Zp .
Zp ∼ Z
= n
(pn ) p Z
Exercise: Show that Zp is isomorphic to
Z Z
lim := {(a1 , a2 , . . .) | ai ∈ , ai+1 ≡ ai mod pi }
← pn Z pi Z
Z Z
⊆ × × ···
pZ p2 Z
Theorem. The following are equivalent.
(i) f (x1 , . . . , xn ) ∈ Z[x1 , . . . , xn ] such that ∀ r ∃ solutions to

f (x1 , . . . , xn ) ≡ 0 mod pr , xi ∈ Z.
8
(ii) ∃ solutions to f (x1 , . . . , xn ) = 0 with xi ∈ Zp .
Proof. (i) =⇒ (ii)
(r) (r) (r) (r)
For each r let x1 , . . . , xn ∈ Z with f (x1 , . . . , xn ) ≡ 0 mod pr . Since
(r)
Zp is compact ∃ a convergent subsequence for the xi and the limit satisfies
f (x1 , . . . , xr ) = 0
suitable for generalization.
Theorem 4. A conic defined over Q has a Q-rational point if and only if
it has a Qp -rational point for all p, p prime or p = ∞.
We will see that checking p-adic solubility is easy. Here are some (chrono-
logical) remarks.
Minkowski: A quadric (hypersurface of degree 2) has a point over Q if

and only if it has point over Qp for all p.
Meyer: A quadric of dim ≥ 4 always has points over Qp , if p 6= ∞(clearly

we can always find a quadric with no real points).
Hasse: Generalized Legendre and Minkowski to number fields and func-

tion fields (global fields). He formulated what is now called the Hasse Principle:
Solutions over Qp for all p implies solution over Q. It is not universally true
but it is true for certain families of equations.
Definition. An absolute value on a field K is a map
|·|:K →R
such that
1. ∀x ∈ K, |x| ≥ 0 and |x| = 0 ⇔ x = 0.
2. |xy| = |x||y| ∀x, y ∈ K.
3. |x + y| ≤ |x| + |y| ∀x, y ∈ K.
Example 7. The “usual” absolute value on Q is

x, if x ≥ 0
|x|∞ =
−x, if x < 0
9
| · |p is the p-adic absolute value defined last lecture.
There is also the trivial absolute value :

1, if x 6= 0
| · |0 =
0, if x = 0
Definition. Two absoulte values on K are equivalent if they induce the

same topology.
Theorem 5. (Ostrowski) Every absolute value in Q is equivalent to | · |p for

some p, p prime, p = ∞ or p = 0.
If K/Q is finite extension (i.e. K is a number field), then for each ab-
solute value | · | on Q there is only a finite number of equivalence classes of
absolute values on K that gives | · | when restricted to Q. We denote by
MK the set of equivalence classes of absolute values on K. If v ∈ MK , then,
there exists p=prime,0 or ∞ such that |x|v = |x|p ∀x ∈ Q. We say that v|p
(v divides p).
Product Formula: if x ∈ Q and x 6= 0, then

Y
|x|p = 1
To see this, we write x = ±pα1 1 · · · pαr r , αi ∈ Z, pi prime, and notice that

 1 if p 6= pi , ∞
|x|p = p−αi if p = pi
pα1 1 · · · pαr r if p=∞

The product formula can be generalized to a number field K.
There is a choice of nv ∈ Z, for v ∈ MK , such that ∀x ∈ K, x 6= 0
10
Y
|x|nv v = 1
v
In F(t), where F is any field, there are absolute values corresponding to

each monic irreducible polynomial in F[t] and another absolute value “at
infinity”. Choose c > 1 real number.
For x = a/b, a, b ∈ F[t] and (a, b) = 1, define
|x|∞ = cdeg(a)−deg(b)
1 1
For instance, | |∞ = .
t c
Using valuation to define an absolute value, we have a way of writing
the product formula eliminating the constant:
a
For x = pr ∈ F(t), p ∈ F[t] monic irreducible and p - ab, we set
b
vp (x) = r and |x|p = c−vp (x) deg p and then
Y Y X
|x|p = 1 ⇔ c−vp (x) deg p = 1 ⇔ vp (x) deg p = 0
Notice that the last part is just expressing the fact that for any rational
function we have # zeros=#poles.
The previous discussion was intended to introduce the following result.
Theorem 6. (Hasse-Minkowski) If K is a global field then a quadric defined

over K has K-rational points if and only if it has Kv -rational points for all
v ∈ MK .
The next theorem will be proved later on.
Theorem 7. Let X be an absolutely irreducible algebraic variety over a
global field K. Then X(Kv ) 6= ∅ for all but finitely many v ∈ MK . Moreover,
the exceptional v can be effectively listed.
Theorem 8. f (x1 , · · · , xn ) ∈ Z[x1 , · · · , xn ] non-constant and absolutely ir-
reducible. Then for all but finitely many primes p, and for all r ≥ 0, there
exist solutions to
f (x1 , · · · , xn ) ≡ 0 mod pr
11
Definition. If X is a collection of varieties defined over a global field K,
then we say that X satisfy the Hasse principle if ∀X ∈ X
X(K) 6= ∅ ⇐⇒ X(Kv ) 6= ∅ ∀v ∈ MK
Big question: Which families satisfies the Hasse principle?
We know the following:
• Hasse-Minkowski: Quadrics satisfy the Hasse principle .
• Hooley, Heath Brown: Cubics in ≥ 8 variables satisfy the Hasse prin-

ciple (over Q ).
Question: Which varieties in ≥ 3 variables do not satisfy the Hasse prin-

ciple?
It is known that 3x3 + 4y 3 + 5 = 0 has points in Qp for all p but does not
have a rational point and there is an example with three variables as well.
The case of cubics with 4, 5, 6 and 7 variables is open.
Theorem 9. Let f ∈ Z[x1 , . . . , xn ] be absolutely irreducible. Then for all

but finitely many primes p, there is a solution to the equation f = 0 in Znp .
Remark. The proof given below works for any global field.
The proof will be given in three steps

Step 1: If f ∈ Z[x1 , . . . , xn ] is absolutely irreducible, then f ∈ (Z/pZ)[x1 , . . . , xn ]
is absolutely irreducible for all but finitely many primes p.
Step 2: If f is absolutely irreducible in (Z/pZ)[x1 , . . . , xn ] and p is large
with respect to deg f and n, then the variety defined by f = 0 has a smooth
point in (Z/pZ)n .
Step 3: If the variety defined by f = 0 has a smooth point in (Z/pZ)n , then
it has a point in Znp .
Step 1: If f ∈ Z[x1 , . . . , xn ] is absolutely irreducible, then f ∈ (Z/pZ)[x1 , . . . , xn ]
is absolutely irreducible for all but finitely many primes p ∈ N.
Let d = deg f . Fix d ≥ 2 and n ≥ 1. Let us consider a polynomial of

degree d in n variables as a vector of its coefficients; in this way, we can
identify the set of polynomials of degree d in n variables with the vector
12
space Vn,d of dimension N = n+d

n . Under this identification, we define
the map φk : Vn,k × Vn,d−k → Vn,d , where 1 ≤ k ≤ d − 1, by (g, h) 7→ gh,
where g and h are polynomials in the same n variables with deg g = k,
deg g = n − k, and gh is the product of g and h as polynomials. By the
formula of multiplication of polynomials in terms of their coefficients, the
image of φk is an algebraic set in the N -dimensional affine space Vn,d , so is
the union U of the image of φk over 1 ≤ k ≤ d − 1. Therefore U is the set of
common zeros (c1 , . . . , cN ) of some F1 , . . . , Fr ∈ Z[y1 , . . . , yN ]. This means
that f factors nontrivially if and only if F1 = . . . = Fr = 0 at f ∈ Vn,d , that
is, the coefficients of f satisfies the polynomial equations F1 = . . . = Fr = 0
in N variables. In fact, the polynomial equations can be written explicitly
in some case as the following exercise shows.
Exercise. Let f (x, y) = a1 x2 + a2 xy + a3 y 2 + a4 x + a5 y + a6 be a polyno-
mial of degree 2 over a field whose characteristic is not 2. Then f factors
nontrivially if and only if
a1 a22 a24
 
det  a22 a3 a25  = 0

a4 a5
2 2 a6
By the assumption that f ∈ Z[x1 , . . . , xn ] is absolutely irreducible, there

must be some Fi which does not vanish at f . Since the coefficients of Fi and
f are all in Z, Fi (f ) is a nonzero rational integer. Hence Fi (f ) 6= 0 in Z/pZ
for all but finitely many primes p, which implies that f ∈ (Z/pZ)[x1 , . . . , xn ]
is absolutely irreducible for those prime since the construction of F1 , . . . , Fr
passes naturally through quotients.
Remark. The classical algebraic geometry works only on algebraically closed
fields, hence in the priori the coefficients of F1 , . . . , Fr are in Q̄. However
the elimination theory ensure that we can get F1 , . . . , Fr in Q[y1 , . . . , yN ],
hence in Z[y1 , . . . , yN ] by clearing denominators.
There is a similar statement for smoothness: If the variety X is smooth,
then X mod p is smooth for all but finitely many primes p.
Let X be defined by the polynomial equation f (x1 , . . . , xn ) = 0. We say X
is not smooth if there is c = (c1 , . . . , cn ) such that
∂f
f (c) = (c) = 0, i = 1, . . . , n.
∂xi
For this to be true, there must be a relation among the coefficients of f . For
cubics to be singular, it needs at least one condition for the coefficients; for
13
them to be reducible, it needs at least two conditions for the coefficients.
If X is smooth, primes p with X mod p smooth is called the primes of good
reduction (associated with X).
Now let us get some geometric intuition about good or bad reductions. As
mentioned in the beginning of semester, in order to see the geometry, we
have to work on algebraically closed fields. Thus let F be an algebraically
closed field and R = F [t], K = F (t). Since F is algebraically closed, the
prime ideals in R are of the form (t − c)R with c ∈ F . Thus the set of
prime ideals in R can be identified with F . In this case, F [t] plays the
role of Z, and F plays the role of the set of primes p ∈ N. Now consider
a polynomial f ∈ R[x1 , . . . , xn ], i.e. the coefficients of f are in F [t]. We
write f as ft to indicate the dependence. For any c ∈ F , the element
f ∈ (R/(t − c)R) [x1 , . . . , xn ] is simply fc ; this corresponds the algebraic
set Vc defined by fc = 0 in the n-dimensional affine space. The primes of
good reduction corresponds the points c ∈ F such that Vc is smooth. In this
context, one can show the set of all c ∈ F such that Vc is not smooth is closed
in the Zariski topology of F , hence is finite. Step 1 is the corresponding
statement when F [t] is replaced by Z.
In general, if we do not assume that F is algebraically closed, the primes of
F [t] corresponds the monic irreducible polynomials, which can be identified
with the Galois orbits of elements in the algebraic closure of F .
Step 2: If f is absolutely irreducible in (Z/pZ)[x1 , . . . , xn ] and p is large
with respect to deg f and n, then the variety defined by f = 0 has a smooth
point in (Z/pZ)n .
To show this, we use the Lang-Weil estimate (which we do not prove):
Theorem 10. Given n ≥ 1, d ≥ 1, there are constants C(n, d), C1 (n, d)
such that for any primes q and any absolute irreducible f ∈ Fq [x1 , . . . , xn ]
with degree d,
3
|#{p ∈ Fnq |f (p) = 0} − q n−1 | ≤ C(n, d)q n− 2 + C1 (n, d)
Hyperplane sections reduce to n = 2 which is due to Hasse-Weil. (Func-
tion field analogues of the Riemann Hypothesis)
Fact: C(2, d) = (d − 1)(d − 2)
To do Step 2, we want a solution c ∈ (Z/pZ)n to f = 0 (mod p) with
∂f
∂xi (c) 6= 0 (mod p) for some i.
Lang-Weil gives that the number of solution to f = 0 (mod p) is about pn−1 .
∂f ∂f
Solutions to f = 0 (mod p), ∂x 1
= 0 (mod p) must satisfy Resxn (f, ∂x1
) = 0,
a equation in x1 , . . . , xn−1 ). The latter one has about pn−2 solutions by
Lang-Weil; and given one of its solution x1 , . . . , xn−1 , there are at most d
14
values for xn with f (x1 , . . . , xn ) = 0. Hence we conclude the number of so-
∂f 5
lutions to f = ∂x1
= 0 (mod p) is at most d(pn−2 +C(n−1, d)pn− 2 +C1 (n−
3
1, d). For p large enough, this is less than pn−1 − C(n, d)q n− 2 − C1 (n, d),
which is a lower bound of the number of solution to f = 0 (mod p). There-
fore, for p large enough, there exists a solution to f = 0 (mod p) which is
∂f
not a solutions to ∂x 1
= 0 (mod p). Step 2 is done.
Under the given assumption, we have established the following conse-
quences:
1. f ∈ (Z/pZ)[x1 , . . . , xn ] is absolutely irreducible for all sufficiently large
primes p.
2. For all sufficiently large primes p, the variety defined by f = 0 has a
smooth point in (Z/pZ)n .
The final step is to turn the smooth point in (Z/pZ)n of the variety de-
fined by f = 0 to a p-adic point.
Essentially we want to prove Hensel’s Lemma:
Lemma 1. (Hensel) Suppose f ∈ Z[x] and a ∈ Z is such that f (a) =

0 (mod p) and f 0 (a) 6= 0 (mod p). Then there exists an α ∈ Zp such that
α = a (mod p) and f (α) = 0.
Remark. Note that the assumption f (a) = 0 (mod p) is equivalent to |f (a)|p <
1; the assumption f 0 (a) 6= 0 (mod p) is equivalent to |f 0 (a)|p = 1; the conclu-
sion asserts that there exists an α ∈ Zp such that |α − a|p < 1 and f (α) = 0.
Therefore Hensel’s Lemma is an analogue to the Newton’s method of finding
zeros of differentiable functions on the real line.
First we prove by induction on n that there exist an ∈ Z such that

an = an−1 (mod pn−1 ) for n ≥ 2, and f (an ) = 0 (mod pn ) and f 0 (an ) 6=
0 (mod p) for n ≥ 1. Letting a1 = a, we have the base case for free. As for
inductive steps, we want to find z ∈ Z such that f (an−1 + zpn−1 ) = 0 and
f 0 (an−1 + zpn−1 ) P
6= 0 (mod p), therefore we can put an = an−1 + zpn−1 .
Write f (x) = di=0 ci xi . Observe the equality
f 00 (an−1 )
f (an−1 + zpn−1 ) = f (an−1 ) + f 0 (an−1 )zpn−1 + (zpn−1 )2 + · · · .
2!
We claim f (an−1 + zpn−1 ) ≡ f (an−1 ) + zpn−1 f 0 (an−1 ) (mod pn ) by showing
15
f (k) (c)
k! ∈ Z for any c ∈ Z and 1 ≤ k ≤ d. Note that
d
f (k) (c) X i(i − 1) · · · (i − k + 1) i−k
= ci c ,
k! k!
i=k
and that i(i−1)···(i−k+1) = ki ∈ Z, which proves the claim.

k!
Now I want to find z ∈ Z such that f (an−1 )+zpn−1 f 0 (an−1 ) ≡ 0 (mod pn ).
By inductive hypothesis f (an−1 ) ≡ 0 (mod pn−1 ), we write f (an−1 ) = upn−1
for some u ∈ Z. Therefore we aim to find z ∈ Z so that (u+zf 0 (an−1 ))pn−1 =
0 (mod pn ), i.e. u + zf 0 (an−1 ) = 0 (mod p). By inductive hypothesis
f 0 (an−1 ) 6= 0 (mod p), such z ∈ Z can be therefore found. Since
f (3) (an−1 )
f 0 (an−1 + zpn−1 ) = f 0 (an−1 ) + f 00 (an−1 )zpn−1 + (zpn−1 )2 + · · ·
2!
f (k) (c)
and we have shown that k! ∈ Z for any c ∈ Z and 1 ≤ k ≤ deg f , we
f (k) (an−1 )
conclude that ∈ Z for 2 ≤ k ≤ deg f , and f 0 (an−1 + zpn−1 ) =
(k−1)!
f 0 (an−1 ) 6= 0 (mod p). Therefore we put an = an−1 + zpn−1 and complete
the inductive step.
Now we have a sequence {an }n≥1 of integers with the stated properties.
We claim {an }n≥1 forms a Cauchy sequence in the p-adic norm. To see this,
we apply the strong triangular inequality: for any n > m,
n−1
X
|an −am |p = | (ai+1 −ai )|p ≤ max{|ai+1 −ai |p : m ≤ i ≤ n−1} ≤ p−m → 0
i=m
as m → ∞. So there exists α ∈ Zp such that an → α as n → ∞, which

implies f (an ) → f (α) as n → ∞ since polynomials are continuous. We have
also f (an ) = 0 (mod pn ), which implies f (an ) → 0 as n → ∞, and therefore
f (α) = 0.
Example. f (x) = x2 + 1, p=5
a1 = a = 2
f (2) = 5 = 0 (mod 5)
f 0 (2) = 4 6= 0 (mod 5)
a2 = 2 + 5z2
f (2 + 5z2 ) = 5 + 2 · 2 · 5z2 + 52 z22 = 5(1 + 4z2 ) + 52 z22 = 0 (mod 52 ) ⇒ z2 = 1
a3 = 7 + 52 z3
f (7 + 52 z3 ) = 50 + 2 · 7 · 52 z3 + 54 z32 = 52 (2 + 14z3 ) + 54 z32 = 0 (mod 53 )
⇒ z3 = 2
16
Question: How to write negative rational integer as p-adic numbers?
Answer: For example, we have −1 ≡ pn − 1 = (p − 1)(1 + p + p2 + . . . +
pn−1 ) mod pn . Thus as a p-adic number, −1 = (p − 1)(1 + p + p2 + . . .).
Alternatively, using a bomb to kill an ant, we may apply Hensel’s Lemma
to the polynomial x + 1 in order to write −1 a p-adic number.
∂f
To turn a solution (a1 , . . . , an ) ∈ (Z/pZ)n of f = 0 and ∂xi 6= 0 for some
i into a p-adic one, simply apply Hensel’s Lemma to
f (a1 , . . . , ai−1 , x, ai+1 , . . . , an ) ∈ Z[x].
Then we have an α ∈ Zp such that (a1 , . . . , ai−1 , α, ai+1 , . . . , an ) is a p-adic

solution to f = 0. This completes the proof of Theorem ??.
Examining the proof of Theorem ??, we discuss how large the prime p
needs to be in each step.
In Step 1, we want f ∈ (Z/pZ)[x1 , . . . , xn ] to be absolutely irreducible.
This require that p is so large that does not divide an integer which depends
only on coefficients of f .
In Step 2, we want a smooth point in (Z/pZ)n of the variety defined by
f = 0. This require that p is so large in terms of the degree of f and the
number of its variables that the Lang-Weil estimate gives a smooth solution
of f = 0.
In Step 3, Hensel’s Lemma can be applied for those primes p ∈ N which
a smooth solution in (Z/pZ)n is found in Step 2. This does not require the
primes to be larger further.
Next, we investigate the question: How likely is it that an f ∈ Z[x1 , . . . , xn ]
has zeros in Znp for all primes p?
Theorem 11. For any n ≥ 1, d ≥ 1 and H ≥ 1, let An,d,H be the set

of polynomials in Z[x1 , . . . , xn ] with total degree d whose coefficients are no
bigger than H in their absolute values; let Bn,d,H be the set of polynomials
in An,d,H such that for any prime p they have a zero in Zpn . Then we have
#Bn,d,H
lim >0 if n ≥ 2, d ≥ 2 and (n, d) 6= (2, 2).
H→∞ #An,d,H
We count ”bad” polynomials in each of three steps in the proof of The-

orem ??.
In Step 3, no bad polynomial is there.
In Step 1, for n ≥ 2, d ≥ 2 and (n, d) 6= (2, 2), the subspace of reducible
polynomials has codimension at least 2 in the space of polynomials of total
17
degree d in n variables. So there are at least 2 numbers that need to be
divisible by p in order for a polynomial in Z[x1 , . . . , xn ] to be reducible in
(Z/pZ)[x1 , . . . , xn ]. Thus the probability of a polynomial in Z[x1 , . . . , xn ]
being bad modulo p is no bigger than p12 . Equivalently, the probability of
a polynomial in Z[x1 , . . . , xn ] being good modulo p is no less than 1 − p12 .
Hence the probability of a polynomial in Z[xQ 1 , . . . , xn ] being good, i.e., good
modulo any primes p ∈ N, is no bigger than p (1 − p12 ) = π62 > 0. Of course
this is a heuristic argument.
In Step 2, given fixed n and d, all but finitely many primes p ∈ N are
good for all polynomials in Z[x1 , . . . , xn ] with total degree d. For each q of
finitely many good primes, the probability of a polynomial in Z[x1 , . . . , xn ]
having a smooth zero is at least

∂f 1 1
Prob f (0, . . . , 0) = 0 mod q, (0, . . . , 0) 6= 0 mod q = (1 − ) > 0.
∂x1 q q
Again, this is a heuristic argument.
Conjecture 12. Smooth hypersurfaces of degree d in n variables for d ≤ n

and n ≥ 4 satisfy the Hasse Principle.
Conjecture 13. Let An,d,H be as in Theorem 11, and Bn,d,H0 be the set of
n
polynomials in An,d,H which admits a zero in Q . Then

0
#Bn,d,H  Cn,d > 0 if d ≤ n
lim = ? if d = n + 1
H→∞ #An,d,H
0 if d ≥ n + 2.

No conjecture for d = n + 1. For d ≥ n + 2, this conjecture is supported

by the following heuristic argument:
Lemma 2. If Λ is a lattice in Rn , that is Λ = {a1 λ1 + . . . + an λn | ai ∈ Z},

where the λi are linearly independent, then:
#{λ ∈ Λ : ||λ|| ≤ H} = cH n + O(H n−1 )

where:
Vol(S n )
c= × Something that depends on the norm.
Vol(Rn /Λ)
Thus, letting N = n+d − 1, we have for a ∈ Qn , there exists some

d
constant c(a) > 0 such that
18
#{f : |coeff(f )| ≤ H, f (a) = 0} = c(a)H N + O(H N −1 )
Using this lemma we see that:
X X
#{f : |coeff(f )| ≤ H, f (a) = 0} = c(a)H N + O(H N −1 )
a∈A a∈A
for A finite. If the error terms didn’t matter this would give us:
X
#{f : |coeff(f )| ≤ H, ∃a ∈ A, f (a) = 0} ≤ c(a)H N
a∈A
Now, for d > n + 1,

X
c(a) < ∞
a∈Qn
and
#{f : |coeff(f )| ≤ H} ≈ cH N +1
so:
N
P P
#f with solutions a c(a)H a c(a)
≤ =
#total cH N +1 cH
which goes to 0 as H → ∞.
Let K be a field. We will define n-dimensional projective space over K:
Pn (K) = {(a0 , . . . , an ) ∈ K n+1 : ∃i, ai 6= 0}/ ∼

where the equivalence relation ∼ is defined by identifying a = (a0 , . . . , an )
and b = (b0 , . . . , bn ), denoted a ∼ b if there exists λ ∈ K ∗ such that a = λb.
The class of (a0 , . . . , an ) is denoted (a0 : . . . : an ).
For example, when K = Q by the equivalence we see that we can rep-
resent every element of projective space as a vector of coprime integers in
exactly 2 ways.
Suppose f ∈ K[x0 , . . . , xn ] is homogeneous of degree d, that is to say:
f (λx0 , . . . , λxn ) = λd f (x0 , . . . , xn )

now if a ∼ b we have that f (a) = λd f (b) thus:
19
f (a) = 0 ⇔ f (b) = 0
Thus the set of points in Pn (K) where f (homogeneous) vanishes is well
defined. Thus we can make the following definition:
Definition 14. A projective algebraic set is the set of common zeros of a

collection of homogeneous polynomials.
Let φ : K n ,→ Pn (K) be defined by φ(a1 , . . . , an ) = (1 : a1 : . . . : an ).

If f ∈ K[x1 , . . . , xn ] is any polynomial define:
x1 xn
f = f( , . . . , ) ∈ K[x0 , x1 , . . . , xn ]
x0 x0
which is homogeneous. If X = {f1 = . . . = fm = 0} ⊂ An then we can
define the algebraic set X = {f 1 = . . . = f n = 0} ⊂ Pn called the projective
closure of X such that:
φ(X(K)) = X(K) ∩ φ(K n )
The points of X − φ(X) are called the points at infinity of X.

Consider
f (x, y) = 0
so
x y
f = zdf ( , )
z z
and f = 0 defines a curve in P2 . The solutions to f (x, y, 0) = 0 are
points at infinity.
Example 15. Consider:
y 2 = x3 + ax + b
y 2 z = x3 + axz 2 + bz 3
so the points at infinity correspond to z = 0 which implies x3 = 0 or
x = 0. So (0 : y : 0) ∼ (0 : 1 : 0) is the point at infinity.
A line in P2 (K) is the set of zeros of a linear homogeneous polynomial

L.
20
Theorem 16. Let P, Q ∈ P2 (K) P 6= Q then there exists a unique line L
that contains P and Q. Given L1 and L2 lines in P2 and L1 6= L2 then there
exists a unique point P ∈ L1 ∩ L2 .
Proof. let P = (a0 : a1 : a2 ) and Q = (b0 : b1 : b2 )

then:
 
x0 x1 x2
det a0 a1 a2  = 0
b0 b1 b2
is a line containing P and Q. P P
For the second part notice that the solutions to ai xi = 0 and bi xi =
0 is a 1-dimensional vector space, so this defines a unique point in P2 (K).
If X ⊆ P2 is a cubic then given P, Q ∈ X(K) there is a unique line P Q

through them . Usually P Q meets X(K) at a third point. (Also, usually
the tangent at P also meets X(K) in a new point). This allows us to usually
produce new points from given points. The iteration of this process is known
as the chord-tangent process.
Theorem 17. (Mordell-Weil) Let X/K be a smooth cubic. If K is a global

field then there exists a finite set G ⊆ X(K) such that X(K) can be obtained
from G by the chord-tangent process.
1 Elliptic curves
Definition: If K is a field then an elliptic curve E over K is a smooth
irreducible projective curve of genus 1 with a point P ∈ E(K)
Question: Is it possible to decide if a curve of genus 1 over Q has a point

over Q?
Answer : There is a procedure which conjecturally works.
Proposition: Every elliptic curve over a field of characteristic 6= 2, 3 is

isomorphic to a curve of the type
y 2 = x3 + ax + b
for some a, b ∈ K, 4a3 + 27b2 6= 0 with the given point P mapping to

(0 : 1 : 0) = O.
21
Group law on E:
O is the identity, and if P = (x, y) then −P = (x, −y). By definition we

have ∀P , P + (−P ) = O and P + O = P .
Let P1 = (x1 , y1 ), P2 = (x1 , y1 ) ∈ E(K). If x1 6= x2 then let

y1 − y2
α :=
x1 − x2
otherwise (where P1 = P2 )
3x21 + a
α := .
2y1
We then have P1 + P2 = (x3 , y3 ) where
x3 := α2 − x1 − x2
y3 := −(α(x3 − x1 ) + y1 )
Geometrically the point (x3 , −y3 ) is the third point of intersection of

the line passing through P1 and P2 with the elliptic curve. Substituting the
equation of the line into the equation for E, we get
(α(x − x1 ) + y1 )2 = x3 + ax + b
which becomes
x3 − α2 x2 + · · · = 0
since x1 , x2 , x3 are the roots we get x1 + x2 + x3 = α2 and so forth.
Also, P + Q + R = O iff P, Q, R are collinear.
One can prove associativity of the group law either by brute force com-
putation or by using methods of classical geometry.
Two elliptic curves are isomorphic over an algebraically closed field if they
have the same j-invariant. In other words, E is ismorphic to E 0 iff j(E) =
j(E 0 ), where
1728a3
j(E)= 3 ,
4a + 27b2
22
so for example, changing x to λ−2 x and y to λ−3 y gives us y 2 = x3 + λ4 ax +
λ6 b which is isomorphic to our original curve.
If Λ ⊆ C is a lattice, then C/Λ ∼

= E(C) under z 7→ (℘(z), ℘0 (z)/2), where
1 X 1 1

℘(z) = 2 + − .
z (z + λ)2 λ2
λ∈Λ,λ6=0
Definition: An abelian variety is a smooth projective irreducible variety

together with a (abelian) group law defined by rational functions in the
coordinates.
(Elliptic curves are abelian varieties of dimension 1.)
The Mordell-Weil Theorem:
Theorem 18. (Mordell-Weil): If K is a global field and A/K is an abelian

variety then A(K) is a finitely generated abelian group.
Aside Question: Is the group law unique? No, because in any abelian
group, given g0 the map (g, h) 7→ g + h − g0 is a group law with identity g0 .
But, other than that, yes.
Theorem 19. If f : A → A0 is a regular map of abelian varieties with
f (0) = 0 then f is a group homomorphism.
Aside Question: Can conics have a group structure? Yes, for example
consider xy = 1 with (x1 , y1 )(x2 , y2 ) = (x1 x2 , y1 y2 ). This only works for
affine curves, not projective ones.
Aside Question: Are there other varieties with group laws, or non-abelian
2
ones? Yes, known as group varieties. For example GLn ⊆ An , the set of
n × n matrices with nonzero determinant.
In general, the analogue of the Mordell-Weil does not hold. For example,
if X denotes the affine conic xy = 1 as above, X(Q) = Q× is not finitely
generated. But see the remark below.
Exercise: If y 2 = x3 + x2 (a nodal cubic) then C0 (K) ∼
= K × , and if y 2 = x3
∼
(a cuspidal cubic) then C0 (K) = (K, +), where C0 := C − {(0, 0)}. In other
words, the same formulas define a group law for the given curves (with the
groups mentioned) after we remove the singular point.
Remark: The first curve in the above exercise is an example of a semi-abelian
23
variety. For those, the integer points form a finitely generated group. This
generalizes both the Mordell-Weil Theorem and the Dirichlet Unit Theorem.
Strategy for the proof of the Mordell-Weil Theorem:
Here we assume K is a global field throughout.
Step 1 : First prove the Weak Mordell-Weil Theorem, which states: If m ≥ 2

is an integer then A(K)/mA(K) is finite, where mA(K) := {mP ; P ∈
A(K)}.
Step 2 : Use height functions: There exist h : A(K) → R≥0 so that

(a) ∀c > 0, {P ∈ A(K); h(P ) ≤ c} is finite
(b) h(mP ) = m2 h(P ) + O(1), P ∈ A(K)
(c) ∀P0 ∈ A(K), ∃c ∈ (P0 ) such that h(P + P0 ) ≤ 2h(P ) + c(P0 ).
Remark: For an elliptic curve over Q, h(x, y) = log(max{|p|, |q|}) where

x = p/q with p, q ∈ Z and (p, q) = 1.
We claim Step 1 and Step 2 imply the Mordell-Weil Theorem:
Let P1 , . . . , Pr be representatives for the classes of A(K)/mA(K). Given

Q ∈ A(K), ∃i ∈ {1, . . . , r} and R ∈ A(K) so that Q − Pi = mR, giving
Q = mR + Pi . Let c = max{c(−P1 ), . . . , c(−Pr )} constants from property
(c) of the height functions, then
h(Q − Pi ) = h(mR) = m2 h(R) + c0
for some constant c0 by property (b), and
m2 h(R) + c0 ≤ 2h(Q) + c
by property (c). Hence
2 c − c0
h(R) ≤ h(Q) +
m2 m2
so letting c00 = (c − c0 )/m2 we have
2
h(Q) + c00 ≥ h(R)
m2
24
so, choosing λ, 2/m2 < λ < 1, either h(R) < λh(Q) or

2
h(Q) λ − 2 ≤ c00 .
m
Therefore we have h(R) < λh(Q) for some λ < 1, or h(Q) ≤ c000 where
c000 = c00 /(λ − 2/m2 ). Note that the restriction h(Q) ≤ c000 gives us a finite
set.
Start with an arbitrary Q and put Q0 = R and repeat until we have
h(Q(k) ) ≤ c000 , that is,
Q = mQ0 + Pi
Q0 = mQ00 + Pi0
..
.
Q(k)
where h(Q(k) ) ≤ c000 which will happen eventually since otherwise h(Q(i) ) ≤
λi h(Q) which goes to 0 as i grows.
Since Q = mQ0 + Pi = m(mQ00 + Pi0 ) + Pi = m2 Q00 + mPi0 + Pi , . . . we
have
Xr
Q = mk Q(k) + ai Pi
i=1
therefore {P1 , . . . , Pr } ∪ {P ; h(P ) ≤ c000 }, a finite set, generate A(K).
Lemma 20. Let K be a field of characteristic p, p = 0 or a prime number,

A be an abelian variety over K, m be an integer with m ≥ 2 and p - m.
Then A[m] := {P ∈ A|mP = 0} is a finite set of points defined over a finite
extension of K.
Proof. For any abelian variety, the group law is defined by rational functions
of coordinates, so mP = P + P + · · · + P is a rational function of P , and so
A[m] is an algebraic set. By algebraic geometry, we know A[m] has a finite
number of irreducible components, each defined over a finite extension of
K. Let X be an irreducible component of A[m], take a point Q ∈ X, then
for any P ∈ X, we have m(P − Q) = mP − mQ = 0 − 0 = 0, which implies
X − Q ⊆ A[m], and 0 = Q − Q ∈ X − Q.
Now consider the properties of A[m] near 0 ∈ A. Let t1 , . . . , tn be local
coordinates at 0, then near 0, the group law µ : A×A → A can be considered
as a rational map µ = µ(t1 , . . . , tn , t01 , . . . , t0n ) of the form
µ = (F1 (t1 , . . . , tn , t01 , . . . , t0n ), . . . , Fn (t1 , . . . , tn , t01 , . . . , t0n )).
25
Since 0 + 0 = 0, we have Fi (0, . . . , 0, 0, . . . , 0) = 0 for all 1 ≤ i ≤ n; since t +
0 = 0+t = t, we have Fi (t1 , . . . , tn , 0, . . . , 0) = ti and Fi (0, . . . , 0, t01 , . . . , t0n ) =
t0i , for all 1 ≤ i ≤ n. It follows that Fi (t1 , . . . , tn , t01 , . . . , t0n ) = ti + t0i +
higher order terms (as a power series) and hence, m(t1 , . . . , tn ) = (t1 , . . . , tn )+
· · ·+(t1 , . . . , tn ) = (mt1 +higher order terms, . . . , mtn +higher order terms).
By assumption, charK - m, so m(t1 , . . . , tn ) 6= 0 for (t1 , . . . , tn ) close but
unequal to 0, which implies 0 is an isolated point in A[m].
Combine the two results, we see each irreducible component X must be
a single point, so A[m] is a finite set of points.
Example. Let A be an elliptic curve y 2 = x3 + ax + b, m = 2. By the

addition on elliptic curves, a point (x, y) ∈ A satisfies 2(x, y) = 0 if and
only if y = 0. So
A[2] = {(x, y) ∈ A|y = 0} ∪ {0} = {(x, 0)|x3 + ax + b = 0} ∪ {0}
consists of 4 points, each defined over the splitting field of x3 + ax + b over

K.
Remark: Given this lemma and the Weak Mordell-Weil Theorem, we

can prove Mordell-Weil Theorem by working in the extension of K which
contains coordinates of A[m]. By the lemma, there exists L/K finite such
that A[m] ⊆ A(L) and we get that A(L)/mA(L) is finite. Then using the
height function, we can deduce A(L) is finitely generated, so A(K) ⊆ A(L)
is also finitely generated.
Now suppose K is a field of characteristic p, p = 0 or a prime number,
A is an abelian variety over K, m is an integer with m ≥ 2 and p - m, and
assume A[m] ⊆ A(K). Denote the group H = Hom(Gal(K sep /K), A[m]),
then H has a bijection to the set of pairs (L, λ), where L/K is a finite
Galois extension and λ is a group monomorphism from Gal(L/K) to A[m].
This bijection sends ϕ to ((K sep )ker ϕ , ϕ). Define a map δ from A(K) to
H as follows: let P ∈ A(K), choose Q ∈ A(K sep ) such that mQ = P , let
L = K(Q). We Claim that L/K is Galois, and the map λ : Gal(L/K) →
A[m] : σ 7→ σQ − Q is a group monomorphism. Now (L, λ) ∈ H, and is
defined to be the image of P under δ. Then δ is a well-defined (independent
of the choice of Q) group homomorphism, and ker δ = mA(K). Henceforth,
δ induces a homomorphism from A(K)/mA(K) to H.
Proof of Claim: Firstly, suppose σ ∈ Gal(K sep /K), then m(σQ) =
σ(mQ) = σ(P ) = P , so m(σQ−Q) = P −P = 0, so σQ−Q ∈ A[m] ⊆ A(K),
and so σQ = Q + (σQ − Q) ∈ A(K(Q)) = A(L). This shows L/K is Galois.
26
Secondly, for any σ, τ ∈ Gal(L/K), we have λ(στ ) = στ Q − Q = στ Q −
σQ+σQ−Q = σ(τ Q−Q)+(σQ−Q) = σ(λ(τ ))+λ(σ). As λ(τ ) is an element
in A[m] ⊆ A(K), it is fixed by σ ∈ Gal(L/K), and so λ(στ ) = λ(τ ) + λ(σ).
Besides, suppose λ(σ) = 0, then σ(Q) = Q =⇒ L = K(Q) is fixed by σ =⇒
σ = 1. This shows λ is a group monomorphism.
Our goal today is prove the Weak Mordell-Weil Theorem. Recall our
map δ:
δ : A(K) −→ Hom(Gal(Ksep /K), A[m]) = H
We can also identify H with the (a priori) set
{(L, λ) : L/K Galois , λ : Gal(L/K) ,→ A[m]},
(in fact, this set has a group structure), and we will often go back and forth
between these two interpretations of H. We will also assume A[m] ⊆ A(K).
We defined δ by the following: given P ∈ A(K), let Q be a point on our
abelian variety such that mQ = P . Then we set λ(σ) = σQ − Q and
L = K(Q), where K(Q) is the extension of K gotten by adjoining the
coordinates of Q. We first want to prove some claims from last time.
Claim. δ is independent of the choice of Q.
Proof. Suppose Q, Q0 satisfy mQ = mQ0 = P . Then m(Q − Q0 ) = O, and

by our assumption that A[m] ⊆ A(K), we see that Q − Q0 ∈ A(K). This
immediately tells us that K(Q) = K(Q0 ) = L since Q and Q0 differ by an
element of A(K). Moreover for any σ ∈ Gal(L/K), σ(Q − Q0 ) = Q − Q0 .
Rearranging, we find that for any σ ∈ Gal(L/K), σQ−Q = σQ0 −Q0 . Hence
λ(σ) is independent of Q, as is L.
Claim. λ is an injection.
Proof. λ(σ) = O if and only if σQ = Q if and only if σ when restricted to

L is the identity, i.e. σ is the identity in Gal(L/K).
Claim. δ is a homomorphism.
Proof. Let P = P1 + P2 . Choose Qi such that mQi = Pi for i = 1, 2. Let

Q = Q1 + Q2 , then mQ = P . Let λi (σ) = σQi − Qi for i = 1, 2. Then
(λ1 +λ2 )(σ) = σQ1 +σQ2 −Q1 −Q2 = σ(Q1 +Q2 )−(Q1 +Q2 ) = σQ−Q = λ(σ),
so λ = λ1 + λ2 as desired.
Claim. ker δ = mA(K).
27
Proof. We first show that mA(K) ⊆ ker δ: say P ∈ mA(K), then there
exists a Q ∈ A(K) satisfying mQ = P , so let us choose this Q. Then L = K
so λ is the zero map, so P ∈ ker δ.
Now assume P ∈ ker δ, so λ(σ) = O for all σ. Then σQ = Q for all σ,
hence Q ∈ A(K) and, since mQ = P , we see that P ∈ mA(K).
Using this last claim, we see that δ : A(K)/mA(K) ,→ H.
Example. Let A = E : y 2 = x3 + ax + b be an elliptic curve and m = 2.
We are assuming that E[2] ⊆ E(K), so the points of order 2 (i.e. those
points with y-coordinate 0) are in E(K). Thus our cubic factors over K:
y 2 = (x − e1 )(x − e2 )(x − e3 ), ei ∈ K.
As discussed last time, E[2] = Z/2Z × Z/2Z. We want to know what

extensions L/K are possible. Considering the possible maps Gal(L/K) ,→
Z/2Z × Z/2Z and noting that char(K) 6= 2, we readily see the only possible
field diagram is
nn L PPPPP
nnnnn PPP
nnn PPP
n PPP
√ nnn √ √
F1 = K( d1 ) K( d1 d2 ) F2 = K( d2 )
PPP
PPP nnn
PPP nnnnn
PPP nn
P nnn
K
From this,√the possible λ’s are easily √ determinable.
√ Noting that we can
adjust K( d) by a square (i.e. K( d) = K( r2 d)), we readily see that
H ∼= K × /(K × )2 ⊕ K × /(K × )2 , where (K × )2 is the subgroup of squares in
×
K . Choosing the basis {(e1 , 0), (e2 , 0)} for E[2], for (x, y) 6= (e1 , 0), (e2 , 0),
we see that
δ : (x, y) 7→ (x − e1 , x − e2 ) ∈ E[2].
For (x, y) = (e1 , 0), (e2 , 0), one can readily work out δ by considering what
to do in order to divide by 2.
Remark. In general, if p6 |m, A[m] = (Z/mZ)2 dim A , so H ∼
2 dim A
= (K × /(K × )m ) .
So far we had not assumed any conditions on our field K. We will now
restrict our arguments to a global field K. Let MK be the set of places
(i.e. equivalence classes of absolute values) of K, Kv be the completion of
K with respect to a place v ∈ MK , Ov = {x ∈ Kv : |x|v ≤ 1} ⊆ Kv , and
Mv = {x ∈ Kv : |x|v < 1} the unique maximal ideal in Ov . By clearing
28
denominators, we can view A as being defined over Ov and reduce modulo
Mv to get a variety Av defined over the finite field Ov /Mv . If Av is an
abelian variety, we say A has good reduction at v; otherwise we say A has
bad reduction at v.
We next define a set S ⊂ MK such that S contains the archimedean
places, all places of bad reduction for A, and all places v | m (i.e. |m|v < 1).
We claim without proof that S is finite. This can be shown by extending
techiques from a previous lecture. We will need the following definitions to
prove the Weak Mordell-Weil Theorem:
Definition 21. Let K be a field, L/K a finite extension, and |·| an absolute
value on L. We say that | · | is unramified in L if {|x| : x ∈ K} = {|x| :
x ∈ L}.
Definition 22. A place v ∈ MK is unramified in L if for all | · | of L that
are in the class of v when restricted to K are unramified in L.
√
Example. Let K = Q, L = Q( D) with D a square-free integer. √ Observe
that for odd p, the p-adic absolute value in Q is ramified in Q( D) if and
only if p | D. (We say |·|/Q is ramified in L if there exists an absolute value
in L that restricts to | · | in Q which is not unramified.)
√ If p | D, D = pc
where (c, p) = 1 (since D is square-free). So | D|2p = |D|p = |pc|p = 1/p.
√ √
Thus | D|p = 1/ p 6∈ {|x|p : x ∈ Q} = pZ ∪ {0}. On the other hand if p
√ √
does not divide D, then we have | D|p = 1. Thus√{|x|p : x ∈ Q( D)} =
{|x|p : x ∈ Q}. We also note that 2 ramifies in Q( D) if and only if D is
even or D ≡ 3 (mod 4).
We will make use of the following important theorem:
Theorem 23. If P ∈ A(K), mQ = P . Then K(Q)/K is unramified outside
of S.
Note that S depends on the abelian variety A and the field K. So as you
vary P ∈ A(K), the theorem tells us that the extensions K(Q)/K are all
unramifiied outside the same set S. We will also need the following theorem:
Theorem 24 (Hermite). If K is a global field, S ⊂ MK finite, d ≥ 2 an inte-
ger (with p6 |d if p = charK > 0), then the set {L/K : L unramified outside of S, [L :
K] = d} is finite.
Theorem 25. Theorem 23 and 24 imply the Weak Mordell-Weil Theorem.
29
Proof. By Theorem 23, if P ∈ A(K), K(Q) is unramified outside of S for
any mQ = P . We also know that [K(Q) : K] ≤ #A[m]. Then by Theorem
24, there are only finitely many choices for such K(Q) and only finitely
many λ : Gal(K(Q)/K) ,→ A[m]. Thus the image of δ is finite. Thus
A(K)/mA(K) is finite, giving us the Weak Mordell-Weil Theorem.
Thus we are only left with proving Theorems 1 and 2. We will only do
the case of an elliptic curve over Q and m = 2, leaving the general case to
one of the standard texts.
Proof of Theorem 1. We have
E : y 2 = x3 + ax + b = (x − e1 )(x − e2 )(x − e3 )
where we may assume without loss of generality that ei ∈ Z (performing a

change of coordinates if necessary). As discussed before,
δ(x, y) = (x − e1 , x − e2 ) ∈ Q× /(Q× )2 × Q× /(Q× )2 .
Considering the possible linear change of variables, we observe that a point

(x, y) on E must be of the form (u/z 2 , v/z 3 ) withu, v, z ∈ Z and (u, z) = 1,
(v, z) = 1. Thus
u − ei z 2
x − ei = .
z2
So if p is an odd prime (2 ∈ S since m = 2) which ramifies in Q(Q),
then p | (u − ei z 2 ). (Note that we do not know if u − ei z 2 is square-free,
so we do not have an if and only if in the previous statement.) Plugging
(x, y) = (u/z 2 , v/z 3 ) into the equation for E, we get
v 2 = (u − e1 z 2 )(u − e2 z 2 )(u − e3 z 2 ).
Say p | (u − e1 z 2 ) but p6 |(u − e2 z 2 )(u − e3 z 2 ), then an even power of p divides

√
exactly (u − e1 z 2 ), since p | v 2 . So p does not ramify in Q( x − e1 ).
So the primes that ramify divide two of the factors of the cubic. So
suppose without loss of generality that p | (u − e1 z 2 ) and p | (u − e2 z 2 ).
Then p | (e1 − e2 )z 2 . But if p | z, then p | u, but this cannot happen since
(z, u) = 1. Thus p6 |z, hence p | (e1 − e2 ). So the primes that ramify in Q(Q)
are contained inside the set
 
 Y 
S = {2, ∞} ∪ p : p | (ei − ej ) .
 
i6=j
| {z }
primes of bad reduction
30
Thus Q(Q) is unramified outside of S.
Proof of Theorem 2. We want to count the number of quadratic extensions
Q unramified outside of S = {2, p1 , . . . , pr }. These extensions are those
of √
Q( D) such that D is square-free and for all p ∈ / S, p6 |D. Thus for i = 0, 1,
Y
D=± pi i .
pi ∈S
There are finitely many such D, so there are finitely many extensions Q(D).

The general case of these theorems requires finiteness of class numbers
and Dirichlet’s Unit Theorem, but we will leave it at that.
We now assume that K is a global field, and MK the set of places of K.
Then for v ∈ MK , we let Kv represent the corresponding completion. We
can now look at the induced maps
δv : A(Kv )/mA(Kv ) → HKv
and notice that we obtain a commutative diagram:
δ : A(K)/mA(K) / HK

δv : A(Kv )/mA(Kv ) / HK
v
We now define a new group. Let
Sm := {h ∈ HK | h ∈ Im(δv ) for all v ∈ MK }.
Because of the diagram above, note that elements in A(K)/mA(K) are in

Sm , and in particular, δ induces an injective map 0 → A(K)/mA(K) ,→ Sm .
Theorem 26. Sm is finite.
Proof. While we do not go through the proof of this theorem here, it follows
from much the same argument as the proof of the finiteness of A(K)/mA(K).
Define
Sham = Sm /δ(A(K)).
We want to figure out how to compute A(K). In order to do this, we need to
compute A(K)/mA(K). It turns out that Sm is computable, but it’s hard
31
to determine the image of A(K)/mA(K) in Sm , as it is not clear how to
decide if an element of Sm actually comes from A(K)/mA(K). It turns out
that there is a way around this.
Let l be a prime, l 6= p if charK = p > 0. Then the following commuta-
tive diagram holds:
0 / A(K)/lA(K) / Sl / Shal /0
O O O
0 / A(K)/ln A(K) / Sl n / Shaln /0
We can look at the image of Sln → Sl , which is a subgroup of Sl containing

the image of A(K)/lA(K). (Note: For this to make sense, we need to define
HK , δ, Sln , . . . even when A[ln ] 6⊂ A(K), which can be done using Galois
cohomology).
Descent “algorithm”
We may now use the above to compute the image of A(K)/lA(K) in Sl in
the following manner. Consider two distinct parallel processes.
1. Compute elements of A(K) and map them to Sl . In this manner, we

will build up a larger and larger picture of the image of A(K)/lA(K)
in Sl .
2. Compute Sln and map it to Sl for n = 1, 2, 3, . . .. In this manner, we

will find smaller and smaller subgroups of Sl containing the image of
A(K)/lA(K).
At some point, it is our hope that the subgroups obtained by following these
two processes will be the same. That is, there will come a point where we
can no longer build up our subgroup as in part 1, or restrict the subgroup
further, as we are doing in part 2. At this point, we will have found the
exact image of A(K)/lA(K) in Sl as desired.
Theorem 27. The descent “algorithm” will work if | Shalm | is bounded as
m → ∞.
Consider
Shal∞ := lim Shaln .
←−
n
L
We define the Tate-Shafarevich group by Sha = l Shal∞ .
32
Conjecture 28 (Tate-Shafarevich). Sha is finite.
If this conjecture indeed holds true in all cases, then our algorithm is
always valid. For now, the conjecture is known to be true only for some
elliptic curves over Q of small rank.
Example. Consider the abelian variety A over K = Q given by y 2 = (x −
e1 )(x − e2 )(x − e3 ) where ei ∈ Z. Let m = 2. Then the map δ is given by:
δ : A(Q) → Q× /(Q× )2 ⊕ Q× /(Q× )2

(x, y) 7→ (x − e1 , x − e2 )
Let µ be the restriction of the map δ to the first coordinate. That is,
µ : A(Q) → Q× /(Q× )2
(x, y) 7→ x − e1
We want to know what it means if b ∈ Q× /(Q× )2 is in the image of µ. First

off, we know that b ∈ Im(µ) if and only if there exists some (x, y) ∈ E(Q)
with x − e1 = bu2 for some u ∈ Q× . That is, there exists some solution to
the equation
y 2 = bu2 (bu2 − e1 + e2 )(bu2 − e1 + e3 ).
Letting v = uy , this is the same as the statement that b ∈ Im (µ) if and only
if the equation
v 2 = b(bu2 + e1 − e2 )(bu2 + e1 − e3 ) (∗)
has a solution (u, v) ∈ Q2 . Consider the set S2 . We know b ∈ S2 (i.e.,

b ∈ µ(A(Qp )) for all p) if and only if the above equation (∗) has points in Qp
for all p. In this example, Sha2 measures how much bigger S2 is than Im(µ).
Specifically, Sha2 is the set of equations y 2 = bu2 (bu2 −e1 +e2 )(bu2 −e1 +e3 )
for varying b which have solutions in Qp for all p “modulo” the set of these
equations that have solutions in Q. What this means for us is that
Sha2 = 0 ⇐⇒ the Hasse Principle holds for this particular set of equations.
For every curve C of genus 1 over a field K, one can associate an elliptic
curve E/K called its Jacobian. If K is a global field and C(Kv ) 6= ∅ for all
v ∈ MK , then C can be viewed as an element of Sha(E/K). Sha(E/K) = 0
if and only if the Hasse Principle holds for curves of genus 1 over K with
Jacobian E. If the conjecture of Tate and Shafarevich above holds then
failure of the Hasse Principle is measured by a finite quantity.
33
Suppose C/K has genus 1. If C(Kv ) = ∅ for some v, then C(K) = ∅.
On the other hand, if C(Kv ) 6= ∅ for all v ∈ MK , then [C] ∈ Sha(E/K). It
follows that we have
C(K) 6= ∅ ⇐⇒ [C] = 0.
If Sha(E/K) is finite, then we can verify whether [C] = 0 by a finite compu-
tation. To do this, we use a bilinear pairing (known as the Cassels pairing)
β : Sha × Sha → Q/Z
which is known to be nondegenerate if Sha is finite. From this pairing, we

get [C] = 0 ⇐⇒ β([C], g) = 0 for all g ∈ Sha. Note that this is true only
for curves of genus 1, not for those of higher genus. It does, however, end
up generalizing to “principal homogeneous spaces of abelian varieties.”
Heights
Recall: To complete our proof of the Mordell-Weil Theorem, we need a
function
h : A(K) → [0, ∞)
such that
(i) For all c > 0, {P ∈ A(K) : h(P ) ≤ c} is finite,
(ii) h(mP ) = m2 h(P ) + O(1),
(iii) ∀P0 ∈ A(K), ∃c(P0 ) > 0, h(P + P0 ) ≤ 2h(P ) + c(P0 ).
By definition, an abelian variety is in projective space, so it is natural to

start with heights defined on projective space Pn (K). For the remainder of
these notes, let K denote a global field (i.e. a finite extension of Q or Fq (x))
and let MK be the set of places of K. Then for each v ∈ MK we choose an
absolute value | · |v and number nv such that the product formula holds for
all x ∈ K ∗ . This means that for all x ∈ K ∗ ,
Y
|x|nv v = 1.
v
If a ∈ Pn (K), write a = (a0 : · · · : an ). Then we define

X
h(a) = nv log max {|ai |v }.
0≤i≤n
v
34
Remark: Sometimes the function
Y
H(a) = max {|ai |v }nv
0≤i≤n
v
is used. These heights are related by h(a) = log H(a).
35
Lemma: If a ∼ b, then h(a) = h(b).
Proof: By definition, a ∼ b if there exists some λ ∈ K ∗ such that ai = λbi

for each i. We then observe that
max {|ai |v } = max {|λ|v |bi |v }

0≤i≤n 0≤i≤n
= |λ|v max {|bi |v }.
0≤i≤n
Summing each side over all v ∈ MK , we have

X
h(a) = h(b) + nv log |λ|v .
v
However, the sum on the right vanishes by the product formula:

!
X Y
nv
nv log |λ|v = log |λ|v
v v
= log 1 = 0.
Example: K = Q.
Every point a ∈ Pn (Q) can be represented by a = (a0 : · · · : an ) with

ai ∈ Z, gcd(a0 , . . . , an ) = 1. If p is prime, then we have that, for this
representation,
max {|ai |p } = 1.
0≤i≤n
To see this, note that there must be some ai not divisible by p. For this ai ,
|ai |p = 1. For the rest, we note that each other aj is of the form aj = ps m
with p 6 |m and s ≥ 0, so |aj | = p−s ≤ 1. Hence we see that
h(a) = log max {|ai |∞ }.

0≤i≤n
36
Theorem: If K is a global field, given integers n, c ≥ 1, the set
{a ∈ Pn (K) : h(a) ≤ c}
is finite.
Before we give the proof of this theorem, the following two propositions
are left as exercises:
Proposition 1: Define h(α) = h(1 : α) for α ∈ K. Then, for α1 , . . . , αn ∈

K we have
max {h(αi )} ≤ h((1 : α1 : · · · : αn )) ≤ h(α1 ) + · · · + h(αn ).

0≤i≤n
Proposition 2: For p/q ∈ Q, gcd(p, q) = 1
h(p/q) = max{log |p|∞ , log |q|∞ }.
We also adopt the following notation:
Notation: log+ x := log max{1, x}.
Proof of Theorem: We handle the case where K is a finite extension of

Q (the case for function fields is similar). Using the first proposition, we
may reduce the problem to demonstrating that each set
{α ∈ K : h(α) ≤ c}
is finite. If α ∈ K, let mα (x) ∈ Z[x] be its minimal polynomial. It is then

enough to show a bound for the coefficients of mα (x) if h(α) ≤ c. Since each
finite extension of Q is contained in some Galois extension, we may assume
that K/Q is Galois. Then mα (x) divides the polynomial
Y
(x − σ(α)).
σ∈Gal(K/Q)
37
It is then enough to bound the heights of the coefficients of this polynomial,
i.e. the symmetric functions in {σ(α) : σ ∈ Gal(K/Q)}. Defining
X
β= σi1 (α) · · · σik (α),
i1 ≤···≤ik
we note that X
h(β) = log+ |β|p .
p≤∞
We then estimate each |β|p . If v|p, then
|β|p = |β|v

X
= σi1 (α) · · · σik (α)
i1 ≤···≤ik
v
X
≤ |σi1 (α) · · · σik (α)|v
i1 ≤···≤ik
X
= |σi1 (α)|v · · · |σik (α)|v .
i1 ≤···≤ik
Since the function |σij (·)|v is yet another absolute value restricting to | · |p
on Q, label it | · |vij . Continuing our estimates,
X X
|σi1 (α)|v · · · |σik (α)|v = |α|vi1 · · · |α|vik
i1 ≤···≤ik i1 ≤···≤ik

n
≤ max{1, |α|v1 , . . . , |α|vn }k .
k
By the strong triangle inequality, we don’t need the combinatorial coefficient

in the non-archimedean case. Combining the estimates at each place,
X
h(β) = log+ |β|p
p≤∞

n X
≤ +k log max {1, |α|v1 , . . . , |α|vn |}
k vi ,...,vn |p
p≤∞

n
≤ + kh(α).
k
38
There is a stronger version of the theorem due to Northcott:
Theorem: For fixed integers d, n ≥ 1, the set
{a ∈ Pn (Q) : [Q(a) : Q] ≤ d, h(a) ≤ c}
is finite.
In defining the height h on Pn (Q), it is necessary to make the right choice

of | · |v , nv such that if a ∈ Pn (K) and L/K, then h(a) is the same value
regardless whether it is computed in K or L.
Theorem: Suppose K is a global field, φ0 , . . . , φr ∈ K[x0 , . . . , xn ] are ho-

mogeneous all of degree d. Then, on the set of a ∈ Pn (K) such that some
φi (a) 6= 0, define the function Φ by
Φ : (a0 : · · · : an ) 7→ (φ0 (a) : · · · : φr (a)) ∈ Pr (K).
Then h(Φ(a)) ≤ h(a) + O(1).
Proof: By definition,
X
h(Φ(a)) = nv log max |φi (a)|v .
0≤i≤r
v
If we write φi (a) as
X
φi (a) = cid0 ,...,dn ad00 · · · adnn ,
d1 +···+dn =d
then we have
 
X
log |φi (a)|v ≤ log  |cid0 ,...,dn |v |a0 |dv0 · · · |an |dvn 
d1 +···dn =d

n+d
≤ log kv + d log max {|ai |v } + log
0≤i≤n d
39
Here kv = max |cid0 ,...,dn |v . It is clear that, for all but finitely many v,
log kv = 0. As before, the combinatorial term is not necessary in the non-
archimedean case. The result follows by summing over v.
Remark: If (φ0 , . . . , φr ) = (x0 , . . . , xn )k , then h(Φ(a)) = dh(a) + O(1).
Heights on Abelian Varieties

We want h : A(K) → [0, ∞) (with K a global field) such that
1. {P : h(P ) ≤ c} is finite for all c,
2. h(mP ) = m2 h(P ) + O(1),
3. ∀P0 ∈ A(K), ∃c(P0 ) such that h(P + P0 ) ≤ 2h(P ) + c(P0 ).
We defined a height function h : Pn → [0, ∞). As A ,→ Pn , we get a
height function on A. It satisfies 1 automatically. We proved “h(Φ(a)) ≤
dh(a) + O(1).” This implies 3 and h(mP ) ≤ m2 h(P ) + O(1) (multiplication
by m is a degree m2 polynomial map). We will only show this for elliptic
curves and m = 2.
To get the lower bound on 2 requires some extra geometric properties of
the embedding A ,→ Pn ; namely, the map P 7→ −P is induced by a linear
transformation of Pn .
Remark. Different embeddings A ,→ Pn give different heights.
We will do the proof in detail now for elliptic curves and m = 2:
y 2 = x3 + ax + b,
P0 = (x0 , y0 ), P = (x, y), h ((x, y)) = h ((1 : x : y)). The equation y 2 =
x3 + ax + b implies 3h(x) = 2h(y) + O(1). It is sufficient to work with h(x);
i.e. with the function (x, y) 7→ h(x).
We start with property 3. P + P0 has x-coordinate given by
y − y0 2 (y − y0 )2 − (x + x0 )(x2 − 2xx0 + x20 )

− (x + x0 ) =
x − x0 (x − x0 )2
x3 + ax + b − 2yy0 + y02 − (x3 − x2 x0 − xx20 + x30 )
=
(x − x0 )2
2yy0 − y0 + ax + b + x2 x0 + xx20 − x30 )
2
= .
(x − x0 )2
40
The numerator and denominator have degree 2 in x and degree 1 in y. So
“h (Φ(a)) ≤ dh(a) + O(1)” gives 3.
Now 2. The x-coordinate of 2P is given by
2 2
3x + a (3x2 + a)2 − 8xy 2
=
2y 4y 2
(3x2 + a)2 − 8x(x3 + ax + b) A(x)
= 3
= ,
4(x + ax + b) B(x)
where deg A = 4, deg B = 3; so “h (Φ(a)) ≤ dh(a) + O(1)” gives h(2P ) ≤

4h(P ) + O(1). Now we just need to show h(2P ) ≥ 4h(P ) + O(1)
X
A(x)
h(2P ) = h = nv log (max {|A(x)|v , |B(x)|v })
B(x) v
likewise,
X
h(P ) = h (x) = nv log (max {|x|v , 1}) .
v
Assume |x|v > 1,
|A(x)|v = a0 x4 + a1 x3 + · · · + an v

= x4 v a0 + a1 x−1 + · · · + an x−4 v

≥ |x|4v |a0 |v − |x|−1 a1 + · · · + an x−3

v v
≥ |x|4v (|a0 |v − Cv ) ,
for some constant Cv . Thus, log (|A(x)|v ) ≥ 4 log(|x|v )+Cv0 . Note in the non-
archimedean case, we can take Cv = 1 using the strong triangle inequality.
Also |a0 |v is almost always 1 (that is, all but finitely many are 1). Likewise
for B, we have |B(x)|v ≤ |x|3v + Cv ; so for those v with |x|v > 1, we get
log (max {|A(x)|v , |B(x)|v }) ≥ 4 log(|x|v ) + Cv0 ,
where the Cv0 are almost always 0.

Now look at |x|v ≤ 1. Then log (max {|x|v , 1}) = 0; so we want to
show log (max {|A(x)|v , |B(x)|v }) ≥ Cv0 with Cv0 = 0 for all but finitely
many v. Now B(x) = 4f (x) (where f (x) = x3 + ax + b) and A(x) =
(f 0 (x))1/2 − 2xf (x).
Claim. (A(x), B(x)) = 1.
41
Proof. If p(x)|B(x) then p(x)|f (x). If also p(x)|A(x), then p(x)|f 0 (x). But
(f (x), f 0 (x)) = 1, since f (x) has distinct roots (as we are working on an
elliptic curve).
Then ∃u(x), v(x) ∈ K[x] such that u(x)A(x) + v(x)B(x) = 1. Hence
1 = |u(x)A(x) + v(x)B(x)|
≤ Cv00 max {|A(x)|v , |B(x)|v } Cv00 = 1 for almost all v,
(the non-archimedean case). So max {|A(x)|v , |B(x)|v } ≥ Cv0 .

This gives us our desired height function; so we have proven the Mordell-
Weil Theorem.
Néron-Tate canonical height

The canonical height is a positive-definite quadratic form on A(K) modulo
torsion; ĥ(mP ) = m2 ĥ(P ), and ĥ(P ) = 0 if and only if P is torsion.
Néron’s approach: X
ĥ(P ) = λv (P ).
v
Tate’s approach:
h(2n P )
ĥ(P ) = lim .
n→∞ 4n
We have h(2P ) ∼ 4h(P ), which implies h(2P
4
)
∼ h(P ); thus
h(2n P )

is a Cauchy sequence.
4n
Formally, you get ĥ(2P ) = 4ĥ(P ); one needs to prove it works for all values
of m.
Remark. Now everytime you have a positive-definite quadratic form on a
finitely generated abelian group Γ of rank r means that you can P
embed Γ in
Rr in such a way that the quadratic form is the restriction of x2i . This
leads to many geometric question about the resulting lattice.
Open Problems
1. What is the “shape” of the Lattice? (square, skew, etc.)
42
2. Finding lower bounds for the smallest positive value of ĥ(P ) for P ∈
A(K). (There is a conjecture of Lang in this direction.)
3. Are there elliptic curves E over Q with rank E(Q) of arbitrarily large
values? (The largest known is something like 28.)
4. How often is the rank large? More generally, what is the distribution
of ranks?
5. For elliptic curves, there is a bound C(K) such that # (E(K))tors ≤

C(K) for all E/K. Does the corresponding statement hold for abelian
varieties of fixed dimension? (It is known that C(Q) = 16. Note for
number fields C(K) depends on [K : Q].)
6. Birch and Swinnerton-Dyer Conjecture (to be discussed)
Remark. Some key facts about the proof of C(Q) = 16. Let Y1 (m) be the set
of isomorphism classes of pairs (E, P ), where E is an elliptic curve P ∈ E
is of order m. It turns out that Y1 (m) is an algebraic curve; It has a map
to A1 via the j-invariant which comes from (E, P ) 7→ E. One finds that
Y1 (m)(Q) is finite if the genus of Y1 (m) is at least 2. For m large, the set
is actually empty. Proving C(Q) = 16 is extremely difficult.
Let K be a global field. A/K will be an abelian variety of dimension
n, MK the set of places of K, and S the set of places of bad reduction and
archimedean places.
Suppose v is a non-archimedean place, with corresponding absolute value |·|v
and completion Kv . Let Ov denote the ring of integers {x ∈ Kv : |x|v ≤ 1}
and Mv the unique maximal ideal {x ∈ Kv : |x|v < 1}. Note that Ov /Mv is
a finite field; we will write qv for its cardinality. We can view the equations
defining A/K as equations over Kv ; for all but finitely many v’s, these
equations are in Ov . We can look at the same equations in Ov /Mv .
We say v is of good reduction if we get from A an abelian variety Av over
Ov /Mv .
Theorem (Weil). Fix v ∈
/ S. There exists a polynomial Pv (T ) ∈ Z[T ] such
that
2n
Y
Pv (T ) = (1 − αi T )
i=1
43
1
where the αi ∈ C have |αi | = qv2 , and such that over the finite field Fqvm , for
all m ≥ 1 we have
Y2n
#Av (Fqvm ) = (1 − αim ).
i=1
Proof. This is a consequence of Weil’s proof of the Riemann Hypothesis

for function fields, and is not given here.
Example. Elliptic curves E/K (n = 1). Here

#Ev (Fqv ) = (1 − α1 )(1 − α2 ) = 1 − (α1 + α2 ) + α1 α2 .
Now Pv (T ) = (1 − α1 T )(1 − α2 T ) ∈ Z[T ], because of the above theorem. It
/ R then α2 = α1 , and so α1 α2 = |α1 |2 = qv . If α1 ∈ R,
follows that if α1 ∈
1
then α2 ∈ R and α1 , α2 = ±qv2 . If their signs differed, plugging back into
#Ev (Fqv ) would give 1 − qv , contradicting #Ev (Fqv ) ≥ 0. We therefore
deduce that α1 = α2 , and α1 α2 = qv . Hence #Ev (Fqv ) = 1 − (α1 + α2 ) + qv ,
1
and we note that |α1 + α2 | ≤ |α1 | + |α2 | = 2qv2 .
We can give an interpretation of Pv (T ) as a characteristic polynomial. Let l
be a prime not dividing qv . Then Av [lk ] ⊂ Av (Fqv ) is a Z/lk -module of rank
2n. The Frobenius automorphism φ (which generates Gal(Fqv /Fqv )) acts
on Av [lk ]. Pv (T ) = det(1 − T φ), so essentially, Pv (T ) is the characteristic
polynomial of the Frobenius automorphism.
Definition. L(A/K, s), a function of the complex variable s, is given by
Y Y
L(A/K, s) := Pv (qv−s )−1 (other stuff)
v ∈S
/ v∈S
| {z }
Ψ
where Ψ is entire and nonvanishing for Re(s) ≥ 1.

Exercise. This product converges for Re(s) > 32 .
In the case of elliptic curves over Q, we obtain
Y
L(E/K, s) = (1 − ap p−s + p1−2s )−1
p
1
where 1 − ap + p = #Ep (Fp ), and |ap | ≤ 2p 2 . Although L is strictly only
defined for Re(s) > 23 , we can heuristically consider L(E/Q, 1). This is
Y Y p − ap + 1 Y p
(1 − ap p−1 + p−1 )−1 = ( )−1 = .
p p
p p
#E p (Fp )
44
From the latter expression, if #Ep (Fp ) > p “often enough”, then L(E/Q, 1) =
0.
We now examine whether the existence of many rational points forces the
existence of many points mod p. Empirically, it has been observed that
Y #Ep (Fp )
−→ ∞ ⇔ E(Q) is infinite.
p
p≤x
Indeed, if this is the case, then the product grows proportionally to (log x)r ,
where r = rankZ E(Q).
Conjecture (Birch, Swinnerton-Dyer).

(1) L(A/K, s) has an analytic continuation to C.
(2) ords=1 L(A/K, s) = rankZ A(K)(= r). Q
R|Sha| cv
(3) lims→1 (s − 1)−r L(A/K, s) = |A(K)tor ||Av∈S
∗ (K)
tor |
,
R r
where the regulator R = vol( (A(K)/A(K) tor
) with respect to canonical
height, Sha = Sha(A/K) is the Tate-Shafarevich group, the cv depend on Ψ
above, A(K)tor is the subgroup of A(K) consisting of points of finite order,
and A∗ is the dual abelian variety to A, which we don’t define. For elliptic
curves, A∗ = A.
Remarks. The stronger condition (1 12 ), L(A/K, 2 − s) = ±L(A/K, s) is

known for abelian varieties (over number fields) having “many endomor-
phisms” by Hecke, Deuring, Taniyama-Shimura and others.
Statement (2) implies that the sign in (1 21 ) is equal to (−1)r .
Statements (1) and (1 21 ) are known for elliptic curves over Q from Wiles’
proof of Fermat’s Last Theorem.
The statements (1) and (1 21 ) are known for abelian varieties over func-
tion fields. If K/Fq (t), then L(A/K, s) is a polynomial in q −s (Weil, Dwork,
Grothendieck).
For elliptic curves over Q, ords=1 L(A/Q, s) ≤ 1 implies statement (2) (Ru-
bin, Kolyvagin, Gross-Zagier, ... )
If K is a function field, then rankZ A(K) ≤ ords=1 L(A/K, s), with equality
iff Sha is finite. In this case, (2) implies (3) (Tate, . . . , Kato-Trihan).
We start with a theorem, which was conjectured by Mordell.
45
Theorem 29. If K is a global field and C/K is a curve of genus ≥ 2, then
C(K) is finite except if K is a function field over Fq and C is defined over
Fq .
Remark. In particular, if K is a number field, then C(K) is finite. The
number field case is solved by Faltings in 1983. The function field case was
done by Samuel in the 60’s.
Before we embark on the arithmetic of curves we need to do some
geoemtry. We begin by defining divisors and Jacobians.
Suppose F is algebraically closed and C/F is a smooth irreducible pro-
jective curve. In general C(F ) is not finite.
P
Definition 30. A divisor on C is a formal sum D := P ∈C(F ) nP P where
nP : C(F ) → Z is supported on finitely many points.
0 =
P P 0 0
Given two divisors D = P n P P, D P nP P define D + D :=
0
P
P (nP + nP )P . Denote the set of all divisors on C by Div(C). In other
words, Div(C) is a free abelian group generated by C(F ). P
We also define a map deg : Div(C) → Z by deg(D) := P nP where
P 0
D = P nP P . Set Div (C) := ker(deg).
Recall that the function field F (C) is defined as a field of fractions of
R := F [x1 , . . . , xn ]/(f1 , . . . , fm )
where f1 = · · · = fm = 0 is a system of equations for a non-empty affine

open subset of C. For example, if C is defined by f (x, y) = 0 then R =
F [x, y]/(f (x, y)) and F (C) is a field of fractions of R.
Let h ∈ F (C). Define (h) ∈ Div(C) by
X
(h) := ordP (h)P
P
where ordP (h) is the order of zero or pole of h at P . We have the following
fact.
Proposition 31. deg((h)) = 0 for any h ∈ F (C), that is, any function has
as many zeros as poles with multiplicities.
Verify this for a couple of examples. If F = C, then C is a compact
Riemann surface. Take any function h ∈ F (C), that is, a meromorphic
function on the surface. Then there are only finitely many zeros and poles
46
on C. We can take a closed path γ on C enclosing a disk that does not
contain any zeros or poles of h. By the residue theorem we have that
Z
dh X
0= = ordP (h).
γ h P
A(x)
If C = P1 , then F (C) = F (x). For h = B(x) , we have ord∞ (h) = deg(B) −
deg(A) hence the proposition holds.
Definition 32. Prin(C) := {(h) : h ∈ F (C)}.
Remark. By proposition 1, Prin(C) ⊂ Div0 (C) ⊂ Div(C). Also for any
h1 , h2 ∈ F (C), (h1 h2 ) = (h1 ) + (h2 ).
We are now in position to define the Jacobian of C.
Definition 33. The Jacobian of C is the group Div0 (C)/Prin(C).
Definition 34. We say divisors D, D0 on C are linearly equivalent if D −
D0 ∈ Prin(C) and denote D ∼ D0 if those are equivalent. In other words,
D ∼ D0 if and only if there exists h ∈ F (C) so that D − D0 = (h).
Here are a couple of examples.
0 1 P
Suppose
P D ∈ Div (P ). Then D is a form of α∈F nα [α] + n∞ [∞]. Note
that nα + n∞ = 0. Define a function
Y
h := (x − α)nα .
α∈F
Then (h) = D and this implies that Jac(P1 ) = 0.

Now consider an elliptic curve E : y 2 = x3 + ax + b. We claim that
Jac(E) ∼= E(F ). The isomorphism is given by E(F ) 3 P 7→ P − O ∈ Jac(E)
where O is the point at ∞ on E. This map is well-defined because P − O ∈
Div0 (E). We are going to show first that this is surjective. Later we will
see that this is injective and a homomorphism.
Let P, Q ∈ E and L be the line P Q. Then by definition,
(L) = P + Q + R − 3O
unless the line L is vertical, where R is another intersection point on E by

L. Hence
P + Q − 2O ∼ O − R ∼ R − O
where R = −R in the group law on E. Therefore
P + Q ∼ R + O.
47
This can be used iteratively on a divisor to trade
P two arbitrary
P points
by a pair of points
P one Pof which is O. Let D = n i P i − m i P i where
ni , mi > 0 and ni = mi . Then, applying this procedure gives:
X X
D ∼ Q+ ni − 1 O − R + mi − 1 O = Q − R
Now, Q − R ∼ Q + R − 2O ∼ O − P ∼ P − O. This shows that our map

is surjective.
We begin with letting F be algebraically closed with characteristic not
2 or 3. We saw for an elliptic curve E, E(F ) → Jac E, where P 7→ P − O, is
surjective (and the proof can be extended to show it is a homomorphism).
We now show it is injective. So we need to show that if P 6= O then
P − O 6∼ 0; that is, there is no function h on E with (h) = P − O. Note:
p
F (E) = F (x, y) (degree 2 since y = x3 + ax + b)
2
F (x)
Thus F (E) = {r(x) + ys(x) : r(x), s(x) ∈ F (x)} so we may write h(x, y) =
a(x)+yb(x)
c(x) , with gcd(a, b, c) = 1 (by getting a common denominator and
cancelling common factors).
We claim that if h has no poles in the affine part of E then c is a constant.
To see this, suppose deg(c) ≥ 1. Then ∃α ∈ F with c(α) = 0. Let β satisfy
β 2 = f (α) (where f (x) = x3 + ax + b) so the points (α, ±β) ∈ E. If h has
no pole at (α, ±β), the numerator must vanish; so a(α) ± b(α)β = 0. Since
it must hold for both (α, +β) and (α, −β), we can add both equations to
deduce a(α) = 0, and so b(α)β = 0.
If β 6= 0, then b(α) = 0; implying that (x − α) is a common factor to a,
b, and c, contradiction. If β = 0 then f (α) = 0. So y 2 = f (x) = (x − α)g(x).
We have ord(α,0) (x − α) = 2 and ord(α,0) y = 1, so a(x) = (x − α)r a1 (x) and
ord(α,0) (a(x)) = 2r ≥ 2 and similarly for b. In particular, their orders are
even. If ord(α,0) (h) ≥ 0 then ord(a + by) ≥ ord(c), so min {ord(a), ord(b)} ≥
ord(c). But ord(b) > 0 implies b(α) = 0, contradiction as above. So
ord(b(α)y) = ord(b) + ord(y) = ord(y) = 1. Thus we have 1 ≥ 2, con-
tradiction.
Thus we have h = a + yb for a, b ∈ F [x]. Now the ordO a = −2 deg a and
ordO by = −2 deg b − 3 (the latter follows since y 2 = f (x) the ordO x = −2
and ordO y = −3). Now −1 = ordO h = min {−2 deg a, −2 deg b − 3} for
a, b 6= 0. [If a = 0 or b = 0, we have a term not appearing in the minimum.]
We get a contradiction, and this shows injectivity.
48
Example. Let y 2 = f (x) with deg f (x) = 5 and f has distinct roots. It
is a fact that this gives a curve of genus 2. f (x) − (ax2 + bx + c) has 5
zeroes say α1 , . . . , α5 . Say βi = −(aαi2 + bαi + c) and Pi = (αi , βi ). Then
(h) = P1 + P2 + · · · + P5 − 5P∞ , where h(x, y) = y + (ax2 + bx + c).
Now given P1 , P2 , P3 , we can choose a, b, c so that P1 + P2 + P3 − 3P∞ ∼
2P∞ − (P4 + P5 ) is the relation given by h.
Using these kinds of relations, we can trade three points for two points
and prove that every divisor D, deg D = 0, is linear equivalent to a divisor
of the form P1 + P2 − 2P∞ .
Define C (2) = C × C/S2 (i.e. mod out by switching coordinates); this is
the set of unordered pairs. Hence we have a map Φ : C (2) → Jac(C) given
by {P1 , P2 } 7→ P1 + P2 − 2P∞ , which is surjective.
Unfortunately, it is not injective. The divisor of (x − α) is (α, β) +
(α, −β)−2P∞ (where β 2 = f (α)). Hence all the pairs of the form {(α, β), (α, −β)}
gives O under Φ. It turns out that this is the only source of non-injectivity.
The other points in C (2) uniquely represent a point in Jac(C). [In algebraic
geometry terms, Φ is a birational map and Jac(C) is the blow-down of C (2)
along a curve.]
For curves of genus 2, the Jacobian is a surface. In general the Jacobian
of a curve of genus g is an algebraic variety of dimension g. In fact, Jac(C)
is projective and, since it has a group law, it is an abelian variety. Moreover,
Jac(C) is birational to C (g) := C g /Sg , where {P1 , . . . , Pg } 7→ P1 + · · · + Pg −
gP0 for some fixed P0 . The sources of non-injectivity are more complicated,
but it is a collection of blow-downs.
Example. If F = C and C/C you can choose ω1 , . . . , ωg linear independent
holomorphic differential forms on C with Jac(C) → Cg /L, where L is a
Pg R Pi R Pi
lattice, given by P1 + · · · + Pg − gP0 7→ i=1 P0 ω1 , . . . , P0 ωg ; the lattice
nR R o
is γ ω 1 , . . . , γ ω g : γ ∈ π 1 (C(C)) .
Back to a global field

Suppose K is a global field and C/K is a smooth curve of genus g. Write
F = K. We can view C as a curve over F and get Jac(C) which is an
abelian variety. It turns out that Jac(C) is defined over K. P
We have Jac(C)(K) is a subgroup of Jac(C)(F ). Suppose D = n i Pi
is a divisor and K is perfect. Take σ ∈ Gal(F/K) and Pi ∈ C(F ). Note σ
acts on points by acting on coordinates and gives another
P point in C, since
C is defined over K. Thus σ can act on D by σ(F ) = ni σ(Pi ).
49
Definition. We say D is defined over K if σ(D) ∼ D for all σ ∈ Gal(F/K)
It can be shown that Jac(C)(K) is exactly the set of equivalence classes
of divisors of degree zero defined over K.
Example. For g = 1, Jac(E) = E over F . It turns out Jac(E)(K) = E(K),
since σ(P )−O = σ(P −O) ∼ P −O (that is using the fact that O is rational)
implies σP ∼ P . Which can be shown to imply σ(P ) = P . If this happens
for all σ ∈ Gal(F/K), we have P ∈ E(K).
If C is curve of genus 1 with no rational points, we get Jac(C) = E is
an elliptic curve E/K with E and C isomorphic over F .
Example. Consider g = 2. Let y 2 = f (x), where deg f = 5. Assume P0
is a rational point at ∞. Now Jac(C) = {[P1 + P2 − 2P0 ] : P1 , P2 ∈ C(F )};
note σ(P1 ) + σ(P2 ) − 2P0 = σ(P1 + P2 − 2P0 ) ∼ P1 + P2 − 2P0 “usually”
implies σ(P1 ) + σ(P2 ) = P1 + P2 . If σ(Pi ) = P2 for i = 1, 2 for all σ then
Pi ∈ C(K). Unfortunately this need not always happen. For example, we
could have σ(P1 ) = P2 and σ(P2 ) = P1 , which would occur if P1 ∈ C(L)
where [L : K] = 2 and P2 is the galois conjugate of P1 ; we have
[
Jac(C)(K) = {[P1 + P2 − 2P0 ] : P1 , P2 ∈ C(K)}∪ [P + P − 2P0 ] : P ∈ C(L)
[L:K]=2
where P denotes the conjugate of P . We have a case of this example: If

y 2 = (x2 + 1)(x3 + 2) and P = (i, 0) ∈ C(Q(i)); so (i, 0) + (−i, 0) − 2P0 ∈
Jac(C)(Q).
• Last time:
C/K curve of genus g ≥ 2. We constructed an abelian variety J =
Jac(C) also defined over K and of dimension g.
If P0 ∈ C(K) then we have a map
α : C −→ J
P 7→ P − P0
Remark: We can use any divisor D0 with deg(D0 ) = 1 and get a map
α : C −→ J
P 7→ P − D0
So C(K) ⊆ J(K). Mordell-Weil implies that J(K) is a finitely generated
abelian group when K is a global field. Does it help to understand C(K)?
50
•Today:
Theorem(Chabauty ’38) If K is a number field and rankZ J(K) < g,
then C(K) is finite.
(We use p-adic analysis to prove it)
We can embed K in Qp for some prime p. Let’s study J(Qp ) first.
Lemma There exists a neighborhood of 0 ∈ J(Qp ) which is isomorphic
(as a p-adic analytic group) to Zgp .
More precisely, there are local coordinates t1 , ..., tg near 0 and power
series λ1 , ..., λg in t1 , ..., tg converging in some neighborhood U of 0 such that
P, Q ∈ U, P = (t1 (P ), ..., tg (P )), etc., then λi (t1 (P + Q), ..., tg (P + Q)) =
λi (t1 (P ), ..., tg (P )) + λi (t1 (Q), ..., tg (Q)).
Assume the Lemma for a while.
Proof of Theorem: Let P1 , ..., Pr be generators of the free part of
J(K), r < g. Replace Pi , if necessary, by pm Pi so that without loss of
generality Pi ∈ U . Consider the vectors
(λ1 (P1 ), · · · , λg (P1 ))

.. ..
. .
(λ1 (Pr ), · · · , λg (Pr ))
There exist a1 , ...ag ∈ Zp not all zero such that
X
ai λi (Pj ) = 0, j = 1, ..., r
P
Let λ = ai λi which is an analytic function on U . By construction λ(Pi ) =
0, i = 1, .., r and by the lemma λ is linear in a neighborhood of 0. So
λ(P ) = 0 for any Z-linear combination of the Pi0 s.
If Q ∈ J(K) then pm Q ∈ U for some large m. λ(Q) = p1m λ(pm Q) = 0.
Suppose by contradiction that C(K) is infinite. C(K) ⊆ C(Qp ) is compact
(C is projective). Hence there is an accumulation point P0 . Take P1 ∈
C(K) such that P1 − P0 ∈ U . If P ∈ C(K) is close enough to P0 then
P − P0 , P1 − P0 ∈ U gives P1 − P0 ∈ J(K) ∩ U . Thus λ(P − P1 ) = 0. But λ
is a an analytic function and C is 1-dimensional. So λ can only have finitely
many zeros in C, unless the function ψ : P 7→ λ(P − P1 ) is identically zero.
If ψ 6≡ 0 we get only finitely many P 0 s, contradicting the fact that P0
was an accumulation point of C(K).
If ψ ≡ 0, then λ(P − P1 ) = 0, ∀P ∈ C(K), P − P1 ∈ U . If Q1 , ..., Qg ∈
C(Qp ) are near P1 ,
X
λ(Qi − P1 ) = 0
51
Now Q1 + ... + Qg − gP1 cover an open set of J(Qp ) as Q1 , ..., Qg vary. So
λ ≡ 0, a contradiction.
Idea of Proof of Lemma: Given P ∈ J, translation by P (Q 7→ Q+P )
gives a map τP : J −→ J such that 0 7→ P . The derivative
dτP : T0 J −→ TP J
is an isomorphism. Therefore the dual spaces (T0 J)∗ and (TP J)∗ are also
isomorphic.
Given an element of (T0 J)∗ say v we get a 1-form on J, given by ω =
(dτP )∗ (v) i.e., for each P an element of (TP J)∗ . We have τP∗ ω = ω. The
function Z P
λ : P 7→ ω
0
is linear in P . In fact,
Z P +Q Z P Z P +Q Z P Z Q Z P Z Q
ω= ω+ ω= ω+ τP∗ ω = ω+ ω
0 0 P 0 0 0 0
Since dim(T0 J) = g, we get a g-dimensional set of λ0 s.

Example: C : y 2 = f (x), deg(f ) = 5, g = 2
x2
Holomorphic differentials on C are generated by dx xdx
y , y . If t = y we
can represent x, y as Laurent series in t.
Z P
dx
P∞ y
is a power series in t. Take α : C −→ J. If ω is a differential on J, then α∗ ω
is a differential on C.
Z P Z α(P )
∗
α ω= ω
P∞ α(P∞ )
(integrals on J transfer to integrals on C)

Let K be a field of characteristic p > 0. Let f (x1 , x2 , . . . , xn ) ∈ K[x1 , x2 , . . . , xn ]
given by X
f (x1 , x2 , . . . , xn ) = ai1 ,i2 ,...in xi11 xi22 . . . xinn .
Define
api1 ,i2 ,...in xi11 xi22 . . . xinn .
X
f (p) (x1 , x2 , . . . , xn ) =
52
If X/K is a variety, X is the set of zeros of f1 , f2 , . . . , fm ∈ K[x1 , x2 , . . . , xn ].
(p) (p) (p)
Define X (p) to be the set of common zeroes of f1 , f2 , . . . , fm . There is
a map F : X → X (p) given by (x1 , x2 , . . . , xn ) 7→ (xp1 , xp2 , . . . , xpn ) because
(f (x1 , x2 , . . . , xn ))p = f (p) (xp1 , xp2 , . . . , xpn ). (Follows from the fact that (x +
y)p = xp + y p in characteristic p).
Remark. If X is defined over Fp , then X (p) = X. If X is defined over
2 m
Fpm , then, if we define X (p ) = (X (p) )(p) , etc. then X (p ) = X, and we get
a map F m : X → X via
2) m)
X → X (p) → X (p → · · · → X (p
.
Let A/K be an abelian variety. Then we have the map F : A → A(p) .
We also have the map [p] : A → A defined by P 7→ pP . Fact: There is a
map V : A(p) → A such that [p] = V ◦ F .
Remark. If Φ : X → Y is an onto map of varieties over K, we have an
injection of function fields K(Y ) ,→ K(X). Then the map Φ is separable if
K(X)/K(Y ) is separable.
Definition. An abelian variety A is called ordinary if V is separable.
Theorem 35. If K is a global field of characteristic p > 0 and C/K is a
curve of genus g ≥ 2 such that C is not defined over K p and J = Jac(C) is
ordinary, then C(K) is finite.
Proof. Mordell-Weil implies that J(K) is finitely generated, so J(K)/pJ(K)

is finite. If C(K) = ∅, we are done. If not, we have α : C → J defined by
P 7→ P − P0 . We can assume C ⊆ J. So C(K) = C ∩ (J(K)). If C(K)
is infinite, then there is P1 ∈ J(K) such that C ∩ (P1 + pJ(K)) is infinite.
Define Φ : J (p) → J by Φ(P ) = V (P ) + P1 . Define C 0 = Φ−1 (C), so that
C 0 is a curve on J (p) . Now we have P ∈ P1 + pJ(K) ⇔ there is Q ∈ J(K)
such that P = P1 + pQ = P1 + V (F (Q)) = Φ(F (Q)) ⇔ Φ−1 (P ) ∈ F (J(K)).
But F (J(K)) = J (p) (K p ). So C ∩ (P1 + pJ(K) infinite ⇒ C 0 ∩ J (p) (K p )
is infinite. J (p) is defined over K p . It can be shown that C 0 ∩ J (p) (K p )
being infinite implies C 0 is defined over K p . But using the fact that J is
ordinary, we can show C 0 defined over K p implies C defined over K p , giving
a contradiction.
Remark. The natural hypothesis for the Mordell conjecture is that C is not
defined over Fq . It can be shown that this is equivalent to C not defined over
53
m m
Kp Kp =
T
for some m ≥ 1: If K is a global field of characteristic p,
m≥1
Fq . The proof can be adapted to deal with this more general situation.
What happens over finite fields: Consider C/Fq , with q = pf . Let K
be a global field of characteristic p wth constant field Fq . Suppose P ∈
C(K) − C(Fq ). We can consider F f m (P ) ∈ C(K), and we get infinitely
many points with m = 1, 2, . . ..
Example. C : y 2 = x5 + 1 over F3 . Take K = F3 (t, s) where s2 = t5 + 1.
m m
Then (t3 , s3 ) ∈ C(K).
In place of J(K), we could have used any subgroup Γ ⊆ J(K sep ) with
the property that Γ/pΓ is finite (e.g., one can take Γ to be the group of
prime to p torsion points in J(K sep ). A special case of the Manin-Mumford
conjecture says that C ∩ Jtor is finite.
Another example that one can take is to embed J(K) in J(Kv ) for a
completion Kv of K and take J(K) in J(K). (In the case of number fields,
the Chabauty argument proves that J(K) ∩ C is finite if J(K) ⊆ J(Kv )
and rk(J(K)) < genus(C)). In characteristic p, J(K) ⊆ J(Kv ) is always
“small”, where “big” means that it “contains a neighborhood of 0,” and
“small” means not “big.” In characteristic 0, J(K) ⊆ J(Kv ) is “small” if
v is non-Archimedean and rk(J(K)) < genus(C). Otherwise, it is usually
“big.”
Conjecture. Let K be a global field and A/K an Abelian variety. If X ⊂ A
is a closed algebraic set defined over K, then
Y
X(Kv ) ∩ A(K) = X(K)
v∈MK
Here MK ist the set of placesY

of K and S represents the closure of the set
S in the product topology of A(Kv ) where each A(Kv ) is endowed with
v∈MK
the v-adic metric topology.
Remark. In other words, the conjecture says that if we are given a sequence
Pn ∈ A(K), n = 1, 2..., such that for all v ∈ MK , there exists Qv ∈ X(Kv )
with Pn −→ Qv in the v-adic topology then there exists a sequence {Rn } ⊂
X(K) with Rn −→ Qv in the v-adic topology.
Also notice that if X(K) is finite then X(K) = X(K), so we get a
stronger conclusion, namely: there exists a R ∈ X(K) such that Qv = R,
for all vinMK .
54
We’ll sketch the proof of this conjecture in characteristic p under the
following assumptions:
• X is a curve with Genus(X) ≥ 2, which is not defined over K p , A is an

ordinary abelian variety and the set {P ∈ A(K sep )|∃n ≥ 1, pn P = 0}
is finite.
Recall that in this case the Mordell conjecture was proved by writing
A(K) = ∪ni=1 (pA(K) + Pi )
where Pi , i = 1, 2, ..., n, are coset representatives of A(K)/pA(K). There-

fore
X ∩ A(K) = ∪ni=1 (X ∩ (pA(K) + Pi ))
So we are left to show that each intersection X ∩ (pA(K) + Pi ) is finite.
This is done by first noticing that pA(K) = V (F (A(K))) = V (A(p) (K p )),
where V is the dual of the Frobenius map. Let’s define a map Φ by Φ(P ) =
V (P ) + Pi and assume, by contradiction, that X ∩ (pA(K) + Pi ) is infinite.
So Φ−1 (X) ∩ A(p) (K p ) is infinite. This would imply, after some work, that
Φ−1 (X) is defined over K p . And so X is defined over K p , contradicting our
previous assumption.
Also notice that the same proof would work if we replace A(K) by its
closure in A(Kv ) and work with Kvp instead of Kv , since Kvp ∩ K = K p .
Taking another careful look at this proof, one sees that we’re really proving
that:
(*) There exists a finite set Z ⊂ X such that Pn ∈ A(K) and Pn −→ Pv ∈

X(Kv ) in A(Kv ) then lim Pn ∈ Z.
Namely, there exists a set Z1 such that X ∩ pA(L) ⊂ Z for any L/K
separable extension. Z corresponds to the finitely many cosets of A(K)
in pA(K).
This condition implies that

Y Y
X(Kv ) ∩ A(K) = Z(Kv ) ∩ A(K)
v∈MK v∈MK
which reduces the proof of the conjecture in dimension 1 to proving it in

dimension zero. The case of dimension zero requires an extra argument.
The condition {p ∈ A(K sep )|∃n ≥ 1, pn P = 0} is finite comes in the proof
of the zero dimensional case.
55
Let’s now take a look at the zero characteristic case and assume the
condition rank(A(K)) < dim A of Chabauty’s theorem. In this case we
construct λ : A(Kv ) −→ Kv analytic with A(K) ⊂ λ−1 (0), but X ∩ λ−1 (0)
is finite. Zv = X ∩ λ−1 (0) has the property (*). Unfortunately Zv depends
on v in characteristic zero, so we cannot reduce the 1-dimensional case to
the 0-dimensional one.
Suppose instead that rank(A(K)) < dim A − 1, then you can construct
λ1 , λ2 : A(Kv ) −→ linearly independents, with A(K) ⊂ λ−1 i (0). Define
−1 −1
Zv = X ∩ λ1 ∩ λ2 .
Conjecture (Stoll). In this situation, there is a finite set Z independent of
v with Zv ⊂ Z, for all v
If this is true then the reduction of the 1-dimensional case to the 0-
dimensional case works in zero characteristic.
56

Notes On Diophantine Geometry: Rational Points On Curves of Genus Zero

Uploaded by

Copyright:

Available Formats

Notes On Diophantine Geometry: Rational Points On Curves of Genus Zero

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Notes On Diophantine Geometry: Rational Points On Curves of Genus Zero

Uploaded by

Copyright:

Available Formats

Notes on Diophantine Geometry

Felipe Voloch and students

Rational Points on Curves of Genus Zero

Example 1. If n = 2 and m = 1, f (x1 , x2 ) = 0, then this is a curve.

Now, we define the genus g(X) of an irreducible curve X. Over a subfield

Background. Singular points are, for example, points (x, y) ∈ C2 with

Example 2. For smooth (projective) plane curves of degree d, the genus is

Definition. An irreducible curve X is parametrizable (also, rational ) over

Theorem 1. A curve is parametrizable over K if and only if it has genus

Proof. (⇒) If X is parametrizable then K(X) embeds in K(t). This is seen

Example 3. Consider the curve X : x2 + y 2 = 1. Then X has a point at

Assume K is perfect (e.g. has characteristic zero). By the genus for-

Example 4. Consider X : y 2 = x3 + x2 . Then X has a singular point

Example 5. But, x2 +y 2 = −1 has no solutions over R. So, a smooth curve

Theorem 3. Every curve of genus zero over K is birationally isomorphic

If we are looking for rational solutions of an equation f (x, y) = 0 where

a0 (px)2 + b0 (py)2 + pcz 2 = 0

(i) a, b, c are not all of the same sign.

(ii) For all odd primes p | abc there is a solution to ax2 + by 2 + cz 2 ≡

Proof. (⇑) If p | abc assume that p | a. We are asuming that ∃ x0 , y0 , z0 ≡6

≡ c(z − uy)(z + uy) mod p

ax2 + by 2 + cz 2 ≡ Lp L0p mod p

ax2 + by 2 + cz 2 ≡ (ax + by + cz)2 mod 2

So there is also L2 and L02 such that

ax2 + by 2 + cz 2 ≡ L2 L02 mod 2

ax2 + by 2 + cz 2 ≡ LL0 mod abc

Consider integers x0 , y0 , z0 that satisfy the inequalities

L(x0 , y0 , z0 ) ≡ L(x00 , y00 , z00 )mod abc

which implies that

L(x0 − x00 , y0 − y00 , z0 − z00 ) ≡ 0 mod abc

a(x0 − x00 )2 + b(y0 − y00 )2 + c(z0 − z00 )2 ≡ 0 mod abc

Let x = x0 − x00 , y = y0 − y00 and z = z0 − z00 . Then

ax2 + by 2 + cz 2 ≡ 0 mod abc

ax2 + by 2 + cz 2 ≤ cz 2 < abc

−(|a|x2 + |b|y 2 ) ≤ ax2 + by 2 + cz 2

−2abc < −(|a|x2 + |b|y 2 )

−2abc < ax2 + by 2 + cz 2 < abc

ax2 + by 2 + c(z 2 + ab) = 0

Since a and b are squarefree and coprime, z 2 + ab 6= 0. The proof of the ⇑

Proof. (⇓) Suppose that ax2 + by 2 + cz 2 = 0 where x, y, z ∈ Z are not

We may thus suppose that p does not divide one of x, y, or z. If p divides

Suppose that p divides a and y but not x or z. Then

So p divides c. This is a contradiction since a and c are coprime.

New Topic: P-Adic Numbers

The p−adic norm of x is defined as |x|p = p−r and |0|p = 0.

Qp is the completion of Q with respect to the metric dp .

where an ∈ { 1, . . . , p − 1 }, an0 6= 0 and |x|p = p−n0 .

|x + y|p ≤ max{ |x|p , |y|p }

Zp is a local ring with maximal ideal (p).

(i) f (x1 , . . . , xn ) ∈ Z[x1 , . . . , xn ] such that ∀ r ∃ solutions to

Minkowski: A quadric (hypersurface of degree 2) has a point over Q if

Meyer: A quadric of dim ≥ 4 always has points over Qp , if p 6= ∞(clearly

Hasse: Generalized Legendre and Minkowski to number fields and func-

Definition. An absolute value on a field K is a map

2. |xy| = |x||y| ∀x, y ∈ K.

3. |x + y| ≤ |x| + |y| ∀x, y ∈ K.

Example 7. The “usual” absolute value on Q is

There is also the trivial absolute value :

Definition. Two absoulte values on K are equivalent if they induce the