Chapter 14
Zeta Functions

14.1 Elliptic Curves over Finite Fields
Let E be an elliptic curve over a п¬Ѓnite п¬Ѓeld Fq . Let

Nn = #E(Fqn )

be the number of points on E over the п¬Ѓeld Fqn . The Z-function of E is
deп¬Ѓned to be
в€ћ
Nn n
ZE (T ) = exp T .
n
n=1

tn /n! is the usual exponential function. The Z-function
Here exp(t) =
encodes certain arithmetic information about E as the coeп¬ѓcients of a gen-
erating function. The presence of the exponential function is justiп¬Ѓed by the
simple form for ZE (T ) in the following result.

PROPOSITION 14.1
Let E be an elliptic curve deп¬Ѓned over Fq , and let #E(Fq ) = q + 1 в€’ a. Then

qT 2 в€’ aT + 1
ZE (T ) = .
(1 в€’ T )(1 в€’ qT )

Factor X 2 в€’ aX + q = (X в€’ О±)(X в€’ ОІ). Theorem 4.12 says that
PROOF

Nn = q n + 1 в€’ О±n в€’ ОІ n .

429

В© 2008 by Taylor & Francis Group, LLC
430 CHAPTER 14 ZETA FUNCTIONS

tn /n, we have
Therefore, using the expansion в€’ log(1 в€’ t) =
в€ћ
Nn n
ZE (T ) = exp T
n
n=1
в€ћ
Tn
(q n + 1 в€’ О±n в€’ ОІ n )
= exp
n
n=1

= exp (в€’ log(1 в€’ qT ) в€’ log(1 в€’ T ) + log(1 в€’ О±T ) + log(1 в€’ ОІT ))

(1 в€’ О±T )(1 в€’ ОІT )
=
(1 в€’ T )(1 в€’ qT )

qT 2 в€’ aT + 1
= .
(1 в€’ T )(1 в€’ qT )

Note that the numerator of ZE (T ) is the characteristic polynomial of the
Frobenius endomorphism, as in Chapter 4, with the coeп¬ѓcients in reverse
order.
A function ZC (T ) can be deп¬Ѓned in a similar way for any curve C over a
п¬Ѓnite п¬Ѓeld, and, more generally, for any variety over a п¬Ѓnite п¬Ѓeld. It is always
a rational function (proved by E. Artin and F. K. Schmidt for curves and by
Dwork for varieties).
The zeta function of E is deп¬Ѓned to be

О¶E (s) = ZE (q в€’s ),

where s is a complex variable. As weвЂ™ll see below, О¶E (s) can be regarded as
an analogue of the classical Riemann zeta function
в€ћ
1
О¶(s) = .
ns
n=1

One of the important properties of the Riemann zeta function is that it sat-
isп¬Ѓes a functional equation relating the values at s and 1 в€’ s:

ПЂ в€’s/2 О“(s/2)О¶(s) = ПЂ в€’(1в€’s)/2 О“((1 в€’ s)/2)О¶(1 в€’ s).

A famous conjecture for О¶(s) is the Riemann Hypothesis, which predicts that
if О¶(s) = 0 with 0 в‰¤ (s) в‰¤ 1 then (s) = 1/2 (there are also the вЂњtrivialвЂќ
zeros at the negative even integers). The elliptic curve zeta function О¶E (s) also
satisп¬Ѓes a functional equation, and the analogue of the Riemann Hypothesis
holds.

В© 2008 by Taylor & Francis Group, LLC
431
SECTION 14.1 ELLIPTIC CURVES OVER FINITE FIELDS

THEOREM 14.2
Let E be an elliptic curve deп¬Ѓned over a п¬Ѓnite п¬Ѓeld.
1. О¶E (s) = О¶E (1 в€’ s)
2. If О¶E (s) = 0, then (s) = 1/2.

PROOF The proof of the п¬Ѓrst statement follows easily from Proposi-
tion 14.1:
q 1в€’2s в€’ aq в€’s + 1
О¶E (s) =
(1 в€’ q в€’s )(1 в€’ q 1в€’s )

1 в€’ aq sв€’1 + q в€’1+2s
=
(q s в€’ 1)(q sв€’1 в€’ 1)

= О¶E (1 в€’ s).

Since the numerator of ZE (T ) is (1 в€’ О±T )(1 в€’ ОІT ), we have

О¶E (s) = 0 в‡ђв‡’ q s = О± or ОІ.

aВ± a2 в€’ 4q
О±, ОІ = .
2
HasseвЂ™s theorem (Theorem 4.2) says that
в€љ
|a| в‰¤ 2 q,

hence a2 в€’ 4q в‰¤ 0. Therefore, О± and ОІ are complex conjugates of each other,
and
в€љ
|О±| = |ОІ| = q.
If q s = О± or ОІ, then
в€љ
= |q s | =
(s)
q q.
Therefore, (s) = 1/2.

There are inп¬Ѓnitely many solutions to q s = О±. However, if s0 is one such
solution, all others are of the form s0 + 2ПЂin/ log q with n в€€ Z. A similar
situation holds for ОІ.
If C is a curve, or a variety, over a п¬Ѓnite п¬Ѓeld, then an analogue of Theo-
rem 14.2 holds. For curves, the functional equation was proved by E. Artin
and F. K. Schmidt, and the Riemann Hypothesis was proved by Weil in the
1940s. In 1949, Weil announced what became known as the Weil conjectures,
which predicted that analogues of Proposition 14.1 and Theorem 14.2 hold for

В© 2008 by Taylor & Francis Group, LLC
432 CHAPTER 14 ZETA FUNCTIONS

varieties over п¬Ѓnite п¬Ѓelds. The functional equation was proved in the 1960s
by M. Artin, Grothendieck, and Verdier, and the analogue of the Riemann
Hypothesis was proved by Deligne in 1973. Much of GrothendieckвЂ™s algebraic
geometry was developed for the purpose of proving these conjectures.
Finally, we show how О¶E (s) can be deп¬Ѓned in a way similar to the Riemann
zeta function. Recall that the Riemann zeta function has the Euler product
expansion

в€’1
1
1в€’ s
О¶(s) =
p
p

when (s) > 1. The product is over the prime numbers. We obtain О¶E (s) if
we replace the primes p by points on E. Consider a point P в€€ E(Fq ). Deп¬Ѓne
deg(P ) to be the smallest n such that P в€€ E(Fqn ). The Frobenius map П†q
acts on P , and it is not diп¬ѓcult to show that the set

SP = {P, П†q (P ), П†2 (P ), . . . , П†nв€’1 (P )}
q q

has exactly n = deg(P ) elements and that П†n (P ) = P . Each of the points in
q
SP also has degree n.

PROPOSITION 14.3
Let E be an elliptic curve over Fq . Then

в€’1
1
1в€’
О¶E (s) = ,
q s deg(P )
SP

where the product is over the points P в€€ E(Fq ), but we take only one point
from each set SP .

PROOF If deg(P ) = m, then P and all the other points in SP have
coordinates in Fqm . Since Fqm вЉ† Fqn if and only if m|n, we see that SP
contributes m points to Nn = #E(Fqn ) if and only if m|n, and otherwise it
contributes no points to Nn . Therefore,

Nn = m.
m|n SP
deg(P )=m

В© 2008 by Taylor & Francis Group, LLC
433
SECTION 14.2 ELLIPTIC CURVES OVER Q

Substituting this into the deп¬Ѓnition of Z(T ), we obtain
в€ћ
Nn n
log Z(T ) = T
n
n=1
в€ћ
1n
= T m
n
n=1 m|n SP
deg(P )=m

в€ћ в€ћ
1
mT mj
= (where mj = n)
mj
j=1 m=1 SP
deg(P )=m

в€ћ
1j deg(P )
= T
j
j=1 SP

=в€’ log(1 в€’ T deg(P ) ).
SP

Let T = q в€’s and exponentiate to obtain the result.

14.2 Elliptic Curves over Q
Let E be an elliptic curve deп¬Ѓned over Q. By changing variables if necessary,
we may assume that E is deп¬Ѓned by y 2 = x3 + Ax + B with A, B в€€ Z. For
a prime p, we can reduce the equation y 2 = x3 + Ax + B mod p. If E mod
p is an elliptic curve, then we say that E has good reduction mod p. This
happens for all but п¬Ѓnitely many primes. For each such p, we have

#E(Fp ) = p + 1 в€’ ap ,

as in Section 14.1. The L-function of E is deп¬Ѓned to be approximately the
Euler product
в€’1
1 в€’ ap pв€’s + p1в€’2s .
good p

This deп¬Ѓnition is good enough for many purposes. However, for completeness,
we say a few words below about what happens at the primes of bad reduction.
The factor 1в€’ap pв€’s +p1в€’2s perhaps seems to be rather artiп¬Ѓcially constructed.
However, it is just the numerator of the zeta function for E mod p, as in
Section 14.1. It might seem more natural to use the whole mod p zeta function,
but the factors arising from the denominator yield the Riemann zeta function
(with a few factors removed) evaluated at s and at s + 1. Since the presence
of the zeta function would complicate matters, the denominators are omitted
in the deп¬Ѓnition of LE (s).

В© 2008 by Taylor & Francis Group, LLC
434 CHAPTER 14 ZETA FUNCTIONS

For the primes where there is bad reduction, the cubic x3 + Ax + B has
multiple roots mod p. If it has a triple root, we say that E has additive
reduction mod p. If it has a double root mod p, it has multiplicative re-
duction. Moreover, if the slopes of the tangent lines at the singular point (see
Theorem 2.31) are in Fp , we say that E has split multiplicative reduction
mod p. Otherwise, it has nonsplit multiplicative reduction.
To treat the primes p = 2 and p = 3, we need to use the general Weierstrass
form for E. For simplicity, we have ignored these primes in the preceding
discussion. However, in the example below, weвЂ™ll include them.
There are many possible equations for E with A, B в€€ Z. We assume that
A, B are chosen so that the reduction properties of E are as good as possible.
In other words, we assume that A and B are chosen so that the cubic has the
largest obtainable number of distinct roots mod p, and the power of p in the
discriminant 4A3 + 27B 2 is as small as possible, for each p. It can be shown
that there is such a choice of A, B. Such an equation is called a minimal
Weierstrass equation for E.

Example 14.1
y 2 = x3 в€’ 270000x + 128250000.
The discriminant of the cubic is в€’28 312 512 11, so E has good reduction except
possibly at 2, 3, 5, 11. The change of variables
x = 25x1 , y = 125y1
transforms the equation into
y1 = x3 в€’ 432x1 + 8208.
2
1

The discriminant of the cubic is в€’28 312 11, so E also has good reduction at
5. This is as far as we can go with the standard Weierstrass model. To treat
2 and 3 we need to allow generalized Weierstrass equations. The change of
variables
x1 = 9x2 в€’ 12, y1 = 27y2
changes the equation to
y2 = x3 в€’ 4x2 + 16.
2
2 2

The discriminant of the cubic is в€’28 11, so E has good reduction at 3. Since
any change of variables can be shown to change the discriminant by a square,
this is the best we can do, except possibly at the prime 2. The change of
variables
x2 = 4x3 , y2 = 8y3 + 4
changes the equation of E to
y3 + y3 = x3 в€’ x2 .
2
3 3

В© 2008 by Taylor & Francis Group, LLC
435
SECTION 14.2 ELLIPTIC CURVES OVER Q

This is nonsingular at 2 (since the partial derivative with respect to y is
2y + 1 в‰Ў 1 в‰Ў 0 (mod 2)). Therefore, E has good reduction at 2. We conclude
that E has good reduction at all primes except p = 11, where it has bad
reduction. The equation y3 +y 3 = x3 в€’x2 is the minimal Weierstrass equation
2
3 3
for E.
LetвЂ™s analyze the situation at 11 more closely. The polynomial in x2 factors
as
x3 в€’ 4x2 + 16 = (x2 + 1)2 (x2 + 5).
2 2

Therefore, E has multiplicative reduction at 11. The method of Section 2.10
shows that the slopes of the tangent lines at the singular point (x2 , y2 ) =
(в€’1, 0) are В±2, which lie in F11 . Therefore, E has split multiplicative reduc-
tion at 11.

We now give the full deп¬Ѓnition of the L-series of E. For a prime p of bad
reduction, deп¬Ѓne
вЋ§
вЋЁ 0 if E has additive reduction at p
1 if E has split multiplicative reduction at p
ap =
в€’1 if E has nonsplit multiplicative reduction at p.

The numbers ap for primes of good reduction are those given above: ap =
p + 1 в€’ #E(Fp ). Then the L-function of E is the Euler product
в€’1
в€’1
1 в€’ ap pв€’s 1 в€’ ap pв€’s + p1в€’2s
LE (s) = .

в€љ
The estimate |ap | < 2 p easily implies that the product converges for (s) >
3/2 (see Exercise 14.3).
Each good factor can be expanded in the form

(1 в€’ ap pв€’s + p1в€’2s )в€’1 = 1 + ap pв€’s + ap2 pв€’2s + В· В· В· ,

where the ap on the left equals the ap on the right (so this is not bad notation)
and

ap2 = a2 в€’ p. (14.1)
p

The product over all p yields an expression
в€ћ
an nв€’s .
LE (s) =
n=1

e
pj j , then
If n = j

an = apej . (14.2)
j
j

В© 2008 by Taylor & Francis Group, LLC
436 CHAPTER 14 ZETA FUNCTIONS

This series for LE (s) converges for (s) > 3/2. It is natural to ask whether
LE (s) has an analytic continuation to all of C and a functional equation, as is
the case with the Riemann zeta function. As weвЂ™ll discuss below, the answer
to these questions is yes. However, the proof is much too deep to be included
in this book (but see Chapter 15 for a discussion of the proof).
To study the analytic properties of LE (s), we introduce a new function.
Let П„ в€€ H, the upper half of the complex plane, as in Chapter 9, and let
q = e2ПЂiП„ . (This is the standard notation; there should be no possibility of
confusion with the q for п¬Ѓnite п¬Ѓelds of Chapter 4.) Deп¬Ѓne
в€ћ
an q n .
fE (П„ ) =
n=1

This is simply a generating function that encodes the number of points on E
mod the various primes. It converges for П„ в€€ H and satisп¬Ѓes some amazing
properties.
Let N be a positive integer and deп¬Ѓne

ab
в€€ SL2 (Z) c в‰Ў 0
О“0 (N ) = (mod N ) .
cd

Then О“0 (N ) is a subgroup of SL2 (Z).
The following result was conjectured by Shimura and has been known by
various names, for example, the Weil conjecture, the Taniyama-Shimura-
Weil conjecture, and the Taniyama-Shimura conjecture. All three
mathematicians played a role in its history.

THEOREM 14.4 (Breuil, Conrad, Diamond, Taylor, Wiles)
Let E be an elliptic curve deп¬Ѓned over Q. There exists an integer N such
that, for all П„ в€€ H,
1.
aП„ + b ab
в€€ О“0 (N )
= (cП„ + d)2 fE (П„ ) for all
fE
cd
cП„ + d
2.
fE (в€’1/(N П„ )) = В±N П„ 2 fE (П„ ).

For a sketch of the proof of this result, see Chapter 15. The theorem (if
we include statements about the behavior at cusps on the real axis) says that
fE (П„ ) is a modular form (in fact, a cusp form; see Section 15.2) of weight
2 and level N . The smallest possible N is called the conductor of E. A
prime p divides this N if and only if E has bad reduction at p. When E has
multiplicative reduction, p divides N only to the п¬Ѓrst power. If E has additive
reduction and p > 3, then p2 is the exact power of p dividing N . The formulas
for p = 2 and 3 are slightly more complicated in this case. See .

В© 2008 by Taylor & Francis Group, LLC
437
SECTION 14.2 ELLIPTIC CURVES OVER Q

The transformation law in (1) can be rewritten as

aП„ + b aП„ + b
fE d = fE (П„ ) dП„
cП„ + d cП„ + d

(this is bad notation: d represents both an integer and the diп¬Ђerentiation
operator; it should be clear which is which). Therefore,

fE (П„ ) dП„

is a diп¬Ђerential that is invariant under the action of О“0 (N ).
Once we have the relation (1), the second relation of the theorem is perhaps
not as surprising. Every function satisfying (1) is a sum of two functions
satisfying (2), one with a plus sign and one with a minus sign (see Exercise
14.2). Therefore (2) says that fE lies in either the plus space or the minus
space.
Taniyama п¬Ѓrst suggested the existence of a result of this form in the 1950s.
Eichler and Shimura then showed that if f is a cusp form (more precisely, a
newform) of weight 2 (and level N for some N ) such that all the coeп¬ѓcients
an are integers, then there is an elliptic curve E with fE = f . This is the
converse of the theorem, but it gave the п¬Ѓrst real evidence that TaniyamaвЂ™s
suggestion was reasonable. In 1967, Weil made precise what the integer N
must be for any given elliptic curve. Since there are only п¬Ѓnitely many modu-
lar forms f of a given level N that could arise from elliptic curves, this meant
that the conjecture (TaniyamaвЂ™s suggestion evolved into a conjecture) could
be investigated numerically. If the conjecture had been false for some explicit
E, it could have been disproved by computing enough coeп¬ѓcients to see that
fE was not on the п¬Ѓnite list of possibilities. Moreover, Weil showed that if
functions like LE (s) (namely LE and its twists) have analytic continuations
and functional equations such as the one given in Corollary 14.5 below, then
fE must be a modular form. Since most people believe that naturally deп¬Ѓned
L-functions should have analytic continuations and functional equations, this
gave the conjecture more credence. Around 1990, Wiles proved that there are
inп¬Ѓnitely many distinct E (that is, with distinct j-invariants) satisfying the
theorem. In 1994, with the help of Taylor, he showed that the theorem is true
for all E such that there is no additive reduction at any prime (but multi-
plicative reduction is allowed). Such curves are called semistable. Finally,
in 2001, Breuil, Conrad, Diamond, and Taylor  proved the full theorem.
LetвЂ™s assume Theorem 14.4 and show that LE (s) analytically continues and
satisп¬Ѓes a functional equation. Recall that the gamma function is deп¬Ѓned
for (s) > 0 by
в€ћ
tsв€’1 eв€’t dt.
О“(s) =
0

Integration by parts yields the relation sО“(s) = О“(s + 1), which yields the
meromorphic continuation of О“(s) to the complex plane, with poles at the

В© 2008 by Taylor & Francis Group, LLC
438 CHAPTER 14 ZETA FUNCTIONS

nonpositive integers. It also yields the relation О“(n) = (n в€’ 1)! for positive
integers n.

COROLLARY 14.5
Let E and N be as in Theorem 14.4. Then
в€љ в€љ
( N /2ПЂ)s О“(s)LE (s) = в€“( N /2ПЂ)2в€’s О“(2 в€’ s)LE (2 в€’ s)

for all s в€€ C (and both sides continue analytically to all of C). The sign here

PROOF Using the deп¬Ѓnition of the gamma function, we have
в€ћ
в€љ в€љ в€ћ
s s
tsв€’1 eв€’t dt
( N /2ПЂ) О“(s)LE (s) = an ( N /2ПЂn)
0
n=1
в€ћ
в€љ
в€ћ
du
(u N )s eв€’2ПЂnu
= an (let t = 2ПЂnu)
u
0
n=1
в€љ
в€ћ
du
(u N )s fE (iu)
=
u
0
в€љ
в€љ в€љ
1/ N в€ћ
du du
(u N )s fE (iu) (u N )s fE (iu) .
+
= в€љ
u u
1/ N
0

(The interchange of summation and integration to obtain the third equality
is justiп¬Ѓed since the sum for f (iu) converges very quickly near в€ћ.) Let be

fE (i/(N u)) = (iu)2 fE (iu) = в€’ u2 fE (iu).

Therefore, let u = 1/N v to obtain
в€љ
в€љ в€љ
1/ N в€ћ
du dv
s
=в€’ (v N )2в€’s fE (iv) .
(u N ) fE (iu) в€љ
u v
1/ N
0

This implies that
в€љ
( N /2ПЂ)s О“(s)LE (s) =
в€љ в€љ
в€ћ в€ћ
du dv
s
в€’ (v N )2в€’s fE (iv) .
в€љ (u N ) fE (iu) в€љ
u v
1/ N 1/ N

Since f (iu) в†’ 0 exponentially as u в†’ в€ћ, it follows easily that both integrals
converge and deп¬Ѓne analytic functions of s. Under s в†’ 2 в€’ s, the right side,
hence the left side, is multiplied by в€’ . This is precisely what the functional
equation claims.

В© 2008 by Taylor & Francis Group, LLC
439
SECTION 14.2 ELLIPTIC CURVES OVER Q

Example 14.2
Let E be the elliptic curve y 2 +y = x3 в€’x2 considered in the previous example.
If we compute the number Np of points on E mod p for various primes, we
obtain, with ap = p + 1 в€’ Np ,

a2 = в€’2, a3 = в€’1, a7 = в€’2,
a5 = 1, a13 = 4, . . .

(except for p = 2, 3, 5, the numbers ap can be calculated using any of the
equations in the previous example). The value

a11 = 1

is speciп¬Ѓed by the formulas for bad primes. We then calculate the coeп¬ѓcients
for composite indices. For example,

a4 = a2 в€’ 2 = 2
a6 = a2 a3 = 2, 2

(see (14.2) and (14.1)). Therefore,

fE (П„ ) = q в€’ 2q 2 в€’ q 3 + 2q 4 + q 5 + 2q 6 в€’ 2q 7 + В· В· В· .

It can be shown that
в€ћ
(1 в€’ q j )2 (1 в€’ q 11j )2
f (П„ ) = q
j=1

is a cusp form of weight 2 and level N = 11. In fact, it is the only such form,
up to scalar multiples. The product for f can be expanded into an inп¬Ѓnite
series
f (П„ ) = q в€’ 2q 2 в€’ q 3 + 2q 4 + q 5 + 2q 6 в€’ 2q 7 + В· В· В· .
It can be shown that f = fE (see ).
The L-series for E satisп¬Ѓes the functional equation
в€љ в€љ
( 11/2ПЂ)s О“(s)LE (s) = +( 11/2ПЂ)2в€’s О“(2 в€’ s)LE (2 в€’ s).

In the early 1960s, Birch and Swinnerton-Dyer performed computer exper-
iments to try to understand the relation between the number of points on
an elliptic curve mod p as p ranges through the primes and the number of
rational points on the curve. Ignoring the fact that the product for LE (s)
doesnвЂ™t converge at s = 1, letвЂ™s substitute s = 1 into the product (weвЂ™ll ignore
в€’1
p в€’ ap + 1 p
в€’1 в€’1
в€’1
1 в€’ ap p +p = = .
p Np
p p p

В© 2008 by Taylor & Francis Group, LLC
440 CHAPTER 14 ZETA FUNCTIONS

If E has a lot of points mod p for many p, then many factors in the product are
small, so we expect that LE (1) might be small. In fact, the data that Birch
and Swinnerton-Dyer obtained led them to make the following conjecture.

CONJECTURE 14.6 (Conjecture of Birch and Swinnerton-Dyer,
Weak Form)
Let E be an elliptic curve deп¬Ѓned over Q. The order of vanishing of LE (s)
at s = 1 is the rank r of E(Q). In other words, if E(Q) torsion вЉ• Zr , then
LE (s) = (s в€’ 1)r g(s), with g(1) = 0, в€ћ.

One consequence of the conjecture is that E(Q) is inп¬Ѓnite if and only if
LE (1) = 0. This statement remains unproved, although there has been some
progress. In 1977, Coates and Wiles showed that if E has complex multipli-
cation and has a point of inп¬Ѓnite order, then LE (1) = 0. The results of Gross
and Zagier on Heegner points (1983) imply that if E is an elliptic curve over
Q such that LE (s) vanishes to order exactly 1 at s = 1, then there is a point
of inп¬Ѓnite order. However, if LE (s) vanishes to order higher than 1, nothing
has been proved, even though there is conjecturally an abundance of points
of inп¬Ѓnite order. This is a common situation in mathematics. It seems that a
solution is often easier to п¬Ѓnd when it is essentially unique than when there
are many choices.
Soon, Conjecture 14.6 was reп¬Ѓned to give not only the order of vanishing,
but also the leading coeп¬ѓcient of the expansion at s = 1. To state the
conjecture, we need to introduce some notation. If P1 , . . . , Pr form a basis for
the free part of E(Q), then

E(Q) = E(Q)torsion вЉ• Z P1 вЉ• В· В· В· вЉ• Z Pr .

Recall the height pairing P, Q deп¬Ѓned in Section 8.5. We can form the r Г— r
matrix Pi , Pj and compute its determinant to obtain what is known as the
elliptic regulator for E. If r = 0, deп¬Ѓne this determinant to equal 1. Let
П‰1 , П‰2 be a basis of a lattice in C that corresponds to E by Theorem 9.21.
We may assume that П‰2 в€€ R, by Exercise 9.5. If E вЉ‚ E(R), let в„¦ = 2П‰2 .
Otherwise, let в„¦ = П‰2 . For each prime p, there are integers cp that we wonвЂ™t
deп¬Ѓne, except to say that if p is a prime of good reduction then cp = 1.
A formula for computing them is given . Finally, recall that is the
(conjecturally п¬Ѓnite) Shafarevich-Tate group of E.

CONJECTURE 14.7 (Conjecture of Birch and Swinnerton-Dyer)
Let E be an elliptic curve deп¬Ѓned over Q. Let r be the rank of E(Q). Then

в„¦ p cp (# E ) det Pi , Pj
r
+ (s в€’ 1)r+1 (br+1 + В· В· В· ).
LE (s) = (s в€’ 1)
#E(Q)2
torsion

В© 2008 by Taylor & Francis Group, LLC
441
SECTION 14.2 ELLIPTIC CURVES OVER Q

This important conjecture combines most of the important information
about E into one equation. When it was п¬Ѓrst made, there were no exam-
ples. As Tate pointed out in 1974 ([116, p. 198]),
This remarkable conjecture relates the behavior of a function L at
a point where it is not at present known to be deп¬Ѓned to the order
which is not known to be п¬Ѓnite!
of a group
In 1986, Rubin gave the п¬Ѓrst examples of curves with п¬Ѓnite , and was able to
compute the exact order of in several examples. Since they were complex
multiplication curves, LE (1) could be computed explicitly by known formulas
(these had been used by Birch and Swinnerton-Dyer in their calculations), and
this allowed the conjecture to be veriп¬Ѓed for these curves. Soon thereafter,
Kolyvagin obtained similar results for elliptic curves satisfying Theorem 14.4
(which was not yet proved) such that LE (s) vanishes to order at most 1 at
s = 1. Therefore, the conjecture is mostly proved (up to small rational factors)
when LE (s) vanishes to order at most one at s = 1. In general, nothing is
known when LE (s) vanishes to higher order. In fact, it is not ruled out (but
most people believe itвЂ™s very unlikely) that LE (s) could vanish at s = 1 to
very high order even though E(Q) has rank 0 or 1.
In 2000, the Clay Mathematics Institute listed the Conjecture of Birch and
Swinnerton-Dyer as one of its million dollar problems. There are surely easier
(but certainly less satisfying) ways to earn a million dollars.
For those who know some algebraic number theory, the conjecture is very
similar to the analytic class number formula. For an imaginary quadratic п¬Ѓeld
K, the zeta function of K satisп¬Ѓes
2ПЂh
О¶K (s) = (s в€’ 1)в€’1 + В·В·В· ,
|d|
w

where h is the class number of K, d is the discriminant of K, and w is the
number of roots of unity in K. Conjecture 14.7 for a curve of rank r = 0
predicts that
в„¦ p cp # E
+ В·В·В· .
LE (s) =
#E(Q)2torsion
The group E can be regarded as the analogue of the ideal class group, the
number в„¦ p cp plays the role of 2ПЂ/ |d|, and #E(Q)torsion is the analogue
of w. Except for the square on the order of the torsion group, the two formulas
for the leading coeп¬ѓcients have very similar forms.
Now letвЂ™s look at real quadratic п¬Ѓelds K. The class number formula says
that
4h log(О·)
О¶K (s) = (s в€’ 1)в€’1 в€љ + В·В·В· ,
2d
where h is the class number of K, d is the discriminant, and О· is the fun-
damental unit. The Conjecture of Birch and Swinnerton-Dyer for a curve of

В© 2008 by Taylor & Francis Group, LLC
442 CHAPTER 14 ZETA FUNCTIONS

rank r = 1, with generator P , predicts that
Л†
в„¦ p cp (# E ) h(P )
LE (s) = (s в€’ 1) + В·В·В· .
#E(Q)2
torsion
в€љ
In this case, в„¦ is the analogue of 4/ d and #E(Q)torsion plays the role of 2,
Л†
which is the number of roots of unity in K. The height h(P ) gives the size of
P . Similarly, log(О·) gives the size of О·.
In general, we can write down a dictionary between elliptic curves and
number п¬Ѓelds:
elliptic curves в†ђв†’ number п¬Ѓelds
points в†ђв†’ units
torsion points в†ђв†’ roots of unity
Shafarevich-Tate group в†ђв†’ ideal class group
This is not an exact dictionary, but it helps to interpret results in one area
in terms of the other. For example, the Dirichlet unit theorem in algebraic
number theory, which describes the group of units in a number п¬Ѓeld, is the
analogue of the Mordell-Weil theorem, which describes the group of rational
points on an elliptic curve. The п¬Ѓniteness of the ideal class group in algebraic
number theory is the analogue of the conjectured п¬Ѓniteness of the Shafarevich-
Tate group.

Exercises
14.1 Let P1 be one-dimensional projective space.
(a) Show that the number of points in P1 (Fq ) is q + 1.
(b) Let Nn = #P1 (Fqn ). Deп¬Ѓne the Z-function for P1 by
в€ћ
Nn n
ZP1 (T ) = exp T .
n
n=1

Show that
1
ZP1 (T ) = .
(1 в€’ T )(1 в€’ qT )
ab
в€€ GL2 (R) with det(M ) > 0. Deп¬Ѓne an action of M
14.2 Let M =
cd
on functions on H by
(f |M )(z) = det(M )(cz + d)в€’2 f (M z),
az+b
where M z = cz+d .

В© 2008 by Taylor & Francis Group, LLC
443
EXERCISES

(a) Show that (f |M1 )|M2 = f |(M1 M2 ).
0 в€’1
. Show that W О“0 (N ) W в€’1 = О“0 (N ).
(b) Let W =
N0
(c) Suppose that f is a function with f |M = f for all M в€€ О“0 (N ). Let
g(z) = (f |W )(z). Show that g|M = g for all M в€€ О“0 (N ). (Hint:
Combine parts (a) and (b).)
(d) Suppose that f is a function with f |M = f for all M в€€ О“0 (N ). Let
f + = 1 (f + f |W ) and f в€’ = 1 (f в€’ f |W ). Show that f + |W = f +
2 2
and f в€’ |W = в€’f в€’ . This gives a decomposition f = f + + f в€’ in
which f is written as a sum of two eigenfunctions for W .
|bn | converges.
14.3 It is well known that a product (1 + bn ) converges if
Use this fact, plus HasseвЂ™s theorem, to show that the Euler product
deп¬Ѓning LE (s) converges for (s) > 3/2.

В© 2008 by Taylor & Francis Group, LLC