LECTURE 24: FROM ZETA TO L-FUNCTIONS

LARRY ROLEN, VANDERBILT UNIVERSITY, FALL 2020

Picture of participants at conference

where the Modularity Theorem was
conjectured, from Shimura’s article
“Yutaka Taniyama and his time:
very personal recollections.”
We have been studying elliptic curves over finite fields via zeta functions. What we
are really after is the structure of elliptic curves over fields like Q. To do this, we can try
to do a local-global principle. Recall that this means that we try to understand points
over Q by understanding it over R and over all p-adic fields Qp . There is a result called
Hensel’s Lemma which allows one to lift “good” solutions to polynomial equations
mod primes to solutions mod powers of primes in infinite sequences (think of this as a
version of Newton’s Method for real functions, but its much more stable). This basically
says that to know the Qp points its enough to know the points over the finite fields Fpr .
The points over R are the easy ones, and the points over the finite fields are exactly
what’s encoded by the congruence zeta functions! So if we believe in the local-global
principle, we would expect that the solutions over R and the data of all congruence zeta
functions for all primes p should tell us something about the Q-points. This is where
the Birch and Swinnerton-Dyer conjecture will come in.
Essentially, there will be one object which encodes all of the congruence zeta functions
at once. To motivate this, recall the Euler product for the Riemann zeta function:
Y 1
ζ(s) = −s
.
p
1 − p
We get something the looks like one of the factors in this product if we plug in T = p−s
in the congruence zeta function at a good prime p not dividing the discriminant of a
curve E:
1 − 2ap p−s + p1−2s
Z(E/Fp ; p−s ) = .
(1 − p−s )(1 − p1−s )
Note that before the term in the numerator was denoted by aE as we were thinking of
the curve as moving but for some fixed prime,
1 but here I want the curve fixed and the
prime moving, so I call it ap to show that dependence.
2 LARRY ROLEN, VANDERBILT UNIVERSITY, FALL 2020

Take the product over the good primes

Y Y 1 − 2ap p−s + p1−2s
Z(E/Fp ; p−s ) = .
(1 − p−s )(1 − p1−s )
p-∆(E) p-∆(E)

The stuff in the bottom essentially gives us a factor ζ(s)ζ(s−1), which is kind of uniform
and doesn’t contain the interesting data about finite field point counts. So we’ll try to
omit it. But more importantly, what about the primes of bad reduction? For those,
we have to define an analogue of the polynomial 1 − 2ap p−s + p1−2s . This choice isn’t
obvious, and the precise reason for it would take us too far afield, so we’ll just state
what it is.
Long Definition. Let E/Q be a rational elliptic curve. If there is bad reduction at a
prime p, we saw before there can be two types of bad behavior. There can be a node
(self-intersection point) or a cusp (sharp point). In the case of a node, we say the
reduction is multiplicative, and if its a cusp, we say its additive. There are two types
of multiplicative reduction: split and non-split. Split means that the two tangent
lines at the node have slopes in Fp .
The conductor of E is the product
Y
N = N (E) := pe p
p

where 

 0 if E has good reduction at p,

1 if E has multiplicative reduction at p,
ep :=

 2 if E has additive reduction at p and p 6= 2, 3,

2 + δp if E has additive reduction at p and p ∈ {2, 3},


where δp is a complicated number that depends on wild ramification (we won’t

define this here). This can be thought of as a refinement of the discriminant. It has the
same primes dividing it, but the exponents of those primes determine the type of bad
reduction. For an elliptic curve E, the local factors are the polynomials in p−s :

−s 1−2s
1 − ap p + p
 p - N,
−s
Lp (E, s) := 1 − ap p p|N, p2 - N,
p2 |N.

1

Here, ap = p + 1 − #E(Fp ) if p is a prime of good reduction, and ap = ±1 if E has

multiplicative reduction depending on if its split or non-split.
The Hasse-Weil L-function of E is the product
Y
L(E, s) := Lp (E, s)−1 .
p
 MODULAR FORMS LECTURE 24 3

That indeed was a long definition! However, the conductor will soon turn out to be
very important to us, and there is no getting around the fact that 2’s and 3’s are nasty
to deal with. Hasse and Weil, after whom this L-function is named, made the following.
Conjecture (Hasse-Weil). The function L(E, s) has an analytic continuation to the
complex plane and satisfies a functional equation.
In the case of the congruent number curves En , this is easier than usual to prove. The
idea is that L(E, s) can be related to the L-function of a modular form. This is again
possible because En has CM. Koblitz’s book continues the calculations we’ve done on
the congruence zeta functions of these and proves that in fact
X
L(En , s) = en (I)(N (I))−s ,
χ
I

where I runs over the non-zero ideals of Z[i], N (I) denotes the “norm” of an ideal, and
χ
en is what’s known as a Hecke character. This implies that
L(En , s) = L(f, s)
where f is a theta function for the Hecke character.
It turns out that something similar holds for all elliptic curves over Q, but its very
difficult to prove.
Theorem (The Modularity Theorem, previously a conjecture of Taniyama–Shimura-Weil
and now a theorem of Wiles and Breuil-Conrad-Diamond-Taylor). Given any elliptic
curve E/Q of conductor N , there is a weight 2, level N Hecke eigenform with
L(E, s) = L(f, s).
That is, there is a modular form in S2 (N ) with
af (p) = p + 1 − #E(Fp )
for all but finitely many primes p.
Getting the level of the modular form right, that is as the conductor, is a very impor-
tant part of this conjecture. We’ve shown that modular forms have an Euler product if
and only if they’re eigenforms, and elliptic curve L-functions are defined as Euler prod-
ucts. So there’s no hope in such a connection if the modular form isn’t an eigenform.
Since the Euler product of the L-function of such a form is of the shape
Y 1
L(f, s) = −s
,
p
1 − ap p + pk−1−2s

its also apparent from the definition of L(E, s) that the only possibility is that f has
weight 2.
This seems like a strange conjecture to believe at first. Initial evidence in small
dimensional spaces gave the first hints (for example, there is no elliptic curve with level
4 LARRY ROLEN, VANDERBILT UNIVERSITY, FALL 2020

less than 11, and there is also no non-zero cusp form in S2 (N ) until N = 11). This was
formulated into a grand conjecture at a famous conference in Japan.
Since modular form L-functions have analytic continuations and functional equations
(in this case under s 7→ 2 − s), the Modularity Theorem implies the following result.
This is still the only known way to prove this.
Theorem. The conjecture of Hasse-Weil is true.
Before it was a theorem, many authors assumed it was true and proved conditional
results based on it. Fortunately, all of those older papers are now rigorous theorems. So
many stronger results can be proven for elliptic curves which are connected to a modular
form. Of course, the main application Wiles had in mind was that it proves Fermat’s
Last Theorem.
Finally, to cap up our story on local-global, I claimed earlier that the L-function,
which “knows” all the local data of the elliptic curve, should know something about the
global data of the curve over Q. This is the content of the (weak) BSD conjecture.
Conjecture (BSD). For any rational elliptic curve, the order of vanishing at the central
critical point s = 1 of the critical strip gives the rank of the elliptic curve. That is,
ords=1 L(E, s) = rk(E/Q).
In particular, L(E, s) = 0 if and only if there are infinitely many rational points on E.
Since L(f, s) = L(E, s) for some modular form f , the L-function of a modular form
near the central critical value determines the rank of the curve. The last step we’ll need
to apply this to the Congruent Number Problem will be that in families of quadratic
twists, the L-values at s = 1 are actually encoded by the Fourier coefficients of a single
modular form of half-integral weight.