lecture1
lecture1
Examples of L-functions
Roughly speaking, a L-function is a generating series for arithmetic data. Far from
being a rigorous mathematical definition, that is nevertheless the guiding principle underlying
the construction of every L-function, as the following examples show.
Example 1. The most elementary L-function is the Riemann zeta function. This is
defined by the infinite series:
X∞
1
ζ(s) :=
n=1
ns
where s ∈ C must have <[s] > 1 to ensure convergence. This series has a product expansion,
discovered by Euler:
X∞ Y
1 1
s
=
n=1
n p
1 − p−s
which highlights its arithmetic properties. Such product expansion is a common feature of
L-functions in general.
where the sum runs through all the non-zero integral ideals a of K and NK/Q is the ideal norm.
Note that for K = Q we recover the Riemann zeta function, so this is a true generalization
of Example 1. However, for a general number field K it no longer makes sense to form an
infinite series by adding up all the integers, but rather just the integral ideals. A simple
reason for this procedure is that convergence is no longer guaranteed whenever there are
infinitely many units in OK . Moreover, unique factorization of ideals in a number field gives
the product expansion:
X 1 Y 1
s
=
a⊂O
NK/Q (a) p⊂O
1 − NK/Q (p)−s
K K
1
which would not be available if we included every integer in the series defining ζK .
From the product expansion, we deduce that Dedekind’s ζK function is constructed en-
tirely from local data, namely the size of the residue field of each prime ideal p of K. The
class number formula
hK · R K
lim+ (s − 1)ζK (s) = p · 2r1 · (2π)r2 (1)
s→1 wK · |DK |
then asserts that this local data can be patched together to compute global invariants of
K, namely:
From this point of view, L-functions can be seen as an attempt to patch local information
together to yield global information about an arithmetic object.
Example 3. Emil Artin’s very influential idea was to associate an L-function L(ρ, s) to
any continuous representation:
ρ : GQ −→ GLd (C) (2)
of the absolute Galois group GQ := Gal(Q/Q). In many ways, number theory as a whole is
the study of GQ , and the study of such complex Galois representations ρ is a natural starting
point.
How can we attach a L-function to a complex Galois representation such as (2)? The key
idea is to consider Frobenius elements attached to rational primes p. For each prime p,
consider the p-adic completion Qp of Q, and pick an algebraic closure Qp . The Galois group
GQp has a much simpler structure than GQ , and it provides the local data that we would
like to patch together, in the following manner. The fields Qp , Qp have rings of integers Zp
and Zp respectively, on which GQp acts. Reduction modulo p then relates GQp to GFp :
redp
⊇ Zp
/Q
Q p
/ Fp
GQ GQ p GF p
redp
Q / Qp ⊇ Zp Fp
/
2
Now GQp sits inside GQ as the decomposition group Dp of p (this is only defined up to
conjugation in GQ ). Inside Dp we have the inertia group
Ip := {σ ∈ Dp : σ = id}
where by σ we mean the element of GFp obtained by considering the action of σ on Fp /Fp .
We then have an isomorphism:
Dp /Ip ' GFp
and
GFp ' lim (Z/nZ) = Z b
←−
is a profinite group generated by a single element σp : x 7→ xp called the Frobenius element
at p.
Note that the characteristic polynomial of ρ(σp ) only depends on its conjugacy class, and
therefore the product is well-defined.
Whenever ρ(Ip ) = id we say that ρ is unramified at p. In this case, we can take ρ(σp )
to act on all of V . By topological considerations, we can moreover deduce that ρ(σ p ) must
be of finite order in GL(V ) = GLd (C). The eigenvalues of ρ(σp ) are then roots of unity
(p) (p)
ζ1 , . . . , ζd and we obtain the following factorization:
Y 1 1 Y
L(ρ, s) := (p)
·...· (p)
× (‘Euler factors’) .
−s 1 − ζd p−s
p unramified 1 − ζ1 p p ramified
Note that the Euler factors come from lower-dimensional representations. For example, if
V Ip = 0 then the Euler factor at p is just 1.
Artin L-functions generalize the two previous L-functions of Examples 1 and 2. We just
have to pick the right representation ρ:
• ζ(s) = L(ρtriv , s), where ρtriv : GQ → Aut(C) is the trivial representation ρ(σ) = 1.
G
• ζK (s) = L(indGQK ρtriv , s) where ρtriv : GK → Aut(C) is the trivial representation of
GK ⊂ GQ (this will be given as an exercise in Homework 1).
3
Note in particular that if K/Q is Galois, then ζK is the Artin L-function corresponding
to the regular representation of Gal(K/Q).
Example 4. In the same spirit as the the Artin L-functions, where we attach a L-function
to a Galois representation with complex coefficients, we can define L-functions attached to
Galois representations with `-adic coefficients. These are continuous maps:
There is an immediate problem with this definition: we are trying to evaluate the character-
istic polynomial of ρ(σp )|V Ip , which has coefficients in Q` , at a complex number p−s . This is
not possible in general, hence we need to restrict our attention to specific classes of `-adic
representations, such as those where the characteristic polynomial of ρ(σp )|V Ip has rational
coefficients.
Now it turns out that all interesting `-adic representations arise from geometry via `-adic
cohomology, and for those we can always defined a L-function as (4). In particular, let X
be a smooth projective algebraic variety over Q. The `-adic cohomology groups:
(`) i
VX,i := Hét (XQ , Q` )
are Q` -vector spaces with a canonical action of GQ arising from the action of GQ on XQ . We
will not delve into the construction of these representations, but we will use the following
properties that they enjoy:
(`)
• VX,i is unramified (i.e. ρ(Ip ) = id) at all p 6= ` at which X has good reduction (this
means that if we take an integral model for X over Z, its fiber over Fp is smooth).
Then
(p)
|αj | = pi/2
4
Note in particular that by the weight property we deduce that ρ(σp ) has infinite order as
(p)
soon as i 6= 0, otherwise the eigenvalues αj would have absolute value equal to 1. This is
in marked contrast with the complex setting, where ρ(σp ) is always of finite order.
where
ap (E) = p + 1 − #E(Fp )
can be readily computed by computing the number of points on the reduction of E modulo
p (note that the fact that p is unramified implies that E has good reduction there, by the
criterion of Neron-Ogg-Shafarevich).
All the examples of L-functions described so far lead to the notion of motivic L-
functions, i.e. L-functions attached to pieces of motivic cohomology of a variety defined
over Q. These motivic L-functions satisfy an amazing collection of conjectural properties:
5
Analytic continuation and functional equation: The functions L(V, s) converge for
<[s] > i/2 + 1 by the weight property. In fact, we have:
• L(H 1 (E1 ) ⊗ H 1 (E2 ), s) is also known to have analytic continuation, via Rankin’s
method.
ζ(2k) ∈ Q · π 2k , k ∈ Z≥1
ζ(1 − 2k) ∈ Q
and these values have been related to important arithmetic invariants. More generally,
Deligne’s conjecture on ‘critical points’ gives precise expectations on which algebraic
numbers we are supposed to obtain when we evaluate L-functions at prescribed special
values.
p-adic interpolation: Once the special values of an L-function are known and are alge-
braic, we can interpolate them p-adically. The prototypical example of this construction is
the Kubota-Leopoldt p-adic L-function. Let
6
• p-adic Rankin L-functions attached to H 1 (E1 ) ⊗ H 1 (E2 ).
Once such p-adic L-functions are available, one can talk about special values of p-
adic L-functions, and a whole new host of conjectures concerning them (p-adic Birch and
Swinnerton-Dyer conjecture, p-adic Bloch-Beilinson conjecture,...).