TATE’S THESIS

BJORN POONEN

These are an approximation of what was covered in lecture in 18.786 Number theory II at
MIT during the first month or so of Spring 2015. This PDF file is divided into sections; the
instructions for viewing the table of contents depend on which PDF viewer you are using.
The syllabus for the semester was
• Tate’s thesis
• Galois cohomology
• An introduction to Galois representations (as much as we have time for)
but these notes cover only Tate’s thesis. We discuss both the number field and global function
field cases, and also the subsequent reformulation by Deligne using L- and -factors.
I thank Pavel Etingof, Sol Friedberg, Kenny Lau, and Jianing Li for some comments on
these notes.

February 3

1. The prototype: analytic continuation and functional equation of the

Riemann zeta function
Recall the Riemann zeta function, defined for Re s > 1 by
Y X
ζ(s) := (1 − p−s )−1 = n−s .
prime p n≥1

Riemann gave two proofs of the following:

Theorem 1.1 (Riemann 1860).

(a) (Analytic continuation) The function ζ(s) extends to a meromorphic function on C,
holomorphic except for a simple pole at s = 1.
(b) (Functional equation) The completed zeta function
s
−s/2
ξ(s) := π Γ ζ(s)
2
defined below

satisfies ξ(s) = ξ(1 − s).

Today we will explain one of these proofs, as presented in [Dei05, Appendix A].
Hecke in 1918–1920 generalized Theorem 1.1 to Dedekind zeta functions ζK (s), and then
quickly realized that his method applied also to Dirichlet L-functions and even more general
1
L-functions. Tate in his 1950 Ph.D. thesis (reprinted in [Tat67]), building on the 1946 Ph.D.
thesis of Margaret Matchett (both of them were students of Emil Artin), gave an adelic
variant of Hecke’s proof. The adelic methods provided further insight and became part of the
inspiration for the Langlands program.
Before proving Theorem 1.1, we need to recall some definitions and theorems from analysis,
which we will assume without proof.
1.1. The gamma function. The definition
Z ∞
dt
Γ(s) := e−t ts
0 t
(convergent for Re s > 0) involves
• an additive character R → C× ,
• a multiplicative character R× → C× , and
• a Haar measure on the group R× .
Properties:
• Γ(s + 1) = s Γ(s). (Proof: Integration by parts.)
• Γ(s) extends to a meromorphic function on C, holomorphic except for simple poles at
0, −1, −2, . . . . (Proof: Use the functional equation for Γ above.)
• Γ(s) has no zeros.
• Γ(n) = (n − 1)! for each n ∈ Z≥1 . (Proof: Induction.)
√ R∞ 2 √
• Γ(1/2) = π. (Proof: Equivalent to −∞ e−x dx = π).
1.2. The Fourier transform. We will consider complex-valued functions on the domain R.

Definition 1.2. Say that a function f : R → C tends to 0 rapidly if for every n ≥ 1 we have
xn f (x) → 0 as |x| → ∞. (Equivalently, for every n we have |f (x)| = O(1/xn ) as |x| → ∞.)

Definition 1.3. Call f : R → C a Schwartz function if for every r ≥ 0 the rth derivative
f (r) exists and tends to 0 rapidly. The Schwartz space S = S (R) is the set of all Schwartz
functions.

Examples of Schwartz functions:

• any C ∞ function of compact support
2
• e−x

Definition 1.4. Given f ∈ S , define the Fourier transform fb by

Z
f (y) :=
b f (x) e−2πixy dx.
R

It turns out that fb ∈ S .

Example 1.5. The function f (x) := e−πx is in S . It turns out that fb = f .

2

2
Fourier inversion formula. If f ∈ S , then
Z
f (x) := fb(y) e2πixy dy.
R
ˆ
In other words, fˆ(x) = f (−x).
Remark 1.6. With more work, one can define fb more generally for f ∈ L2 (R), and the Fourier inversion
formula still holds.

Poisson summation formula. If f ∈ S , then

X X
f (n) = fb(n).
n∈Z n∈Z

The Poisson summation theorem can be generalized to a wider class of functions: see Theorem 3.6.1 in
[Dei05].

1.3. A theta function. For real t > 0, define

2
X
Θ(t) := e−πn t .
n∈Z

Theorem 1.7 (Functional equation for Θ). For every real t > 0,
 
−1/2 1
Θ(t) = t Θ .
t

Proof. If f ∈ S and c 6= 0, then the Fourier transform of f (x/c) is cfb(cy) (integrate by

2
substitution). Apply this to f (x) := e−πx (which satisfies fb = f ) and c := t−1/2 : the Fourier
2 2
transform of ft (x) := e−πtx is fbt (y) = t−1/2 e−π(1/t)y . Apply the Poisson summation formula
to ft :
2 2
X X
e−πn t = t−1/2 e−πn (1/t)
n∈Z n∈Z
 
−1/2 1
Θ(t) = t Θ . 
t
It is the functional equation for Θ(t) that will imply the functional equation for ζ(s).
Remark 1.8. For z in the upper half plane h := {z ∈ C : im z > 0}, define q := e2πiz . Jacobi’s classical theta
function X 2 X 2
θ(z) := q n /2 = eiπn z
n∈Z n∈Z
converges to a holomorphic function on h, and θ(it) = Θ(t). Theorem 1.7 translates into
θ(−1/z) = (z/i)1/2 θ(z),
for the branch of the square root defined using the principal branch of the complex logarithm. This, together
with the obvious identity θ(z + 2) = θ(z) (and a technical growth condition),
! implies !
that θ(z) is a modular
0 −1 1 2
form of weight 1/2 for the subgroup Γ of SL2 (Z) generated by and .
1 0 0 1
3
 2
Remark 1.9. Since e−πn t is an even function of n, we have
2
X
Θ(t) = 1 + 2 e−πn t
n≥1
X 2 Θ(t) − 1
e−πn t = .
n≥1
2

1.4. Riemann’s proof. Look at the contribution to ξ(s) coming from one summand in ζ(s):
Z ∞
−s/2 −s −s/2 −s dx
π Γ(s/2) n = π n e−x xs/2
0 x
Z ∞
2 dt
= e−πn t ts/2 (substitute x = πn2 t).
0 t
Summing over integers n ≥ 1 yields
Z ∞ 
Θ(t) − 1 dt
ξ(s) = ts/2 (1)
0 2 t
if we can justify the interchange of summation and integration. For s ∈ R>1 , this is OK
because everything is nonnegative and the sum on the left converges. In fact, the right hand
side converges absolutely for any complex s with Re s > 1, since changing the imaginary part
of s does not affect |ts/2 |. Thus (1) is valid whenever Re s > 1. If s is a negative real number,
on the other hand, the integral in formula (1) does not converge, because the integrand grows
too quickly as t → 0+ .
The fix will be to split the range of integration (0, ∞) into the good part (1, ∞) and the
problematic part (0, 1), and then to rewrite the problematic part by using the functional
equation of Θ(t).
• The good part contributes
∞  
Θ(t) − 1
Z
dt
I(s) := ts/2 ,
1 2 t
which converges to a holomorphic function on all of C since Θ(t) − 1 = O(e−πt ) as
t → ∞.
• As for the problematic part, the substitution t 7→ 1/t negates dt/t and leads to
Z ∞ !
1
Z 1

−

Θ(t) − 1 dt Θ 1 dt
ts/2 = t
t−s/2
0 2 t 1 2 t
Z ∞ 1/2
 
t Θ(t) − 1 dt
= t−s/2 (by Theorem 1.7)
1 2 t
Z ∞
1 ∞ (1−s)/2 dt 1 ∞ −s/2 dt

Θ(t) − 1
Z Z
(1−s)/2 dt
= t + t − t
1 2 t 2 1 t 2 1 t
1 1
= I(1 − s) − − .
1−s s
4
Putting the two parts back together yields
1 1
ξ(s) = I(s) + I(1 − s) − −
1−s s
for Re s > 1. But the right hand side is meromorphic on C, symmetric with respect to
s 7→ 1−s, and holomorphic except for simple poles at 0 and 1, so it is the desired meromorphic
continuation of ξ(s). Dividing by π −s/2 Γ(s/2), which has simple poles at 0, −2, −4, . . . and
no zeros, shows that ζ(s) has a meromorphic continuation to all of C with a simple pole at
s = 1 and no other poles.

Remark 1.10. As a bonus, we see that ζ(s) has zeros at −2, −4, −6, . . . , and that all other
zeros lie in the “critical strip” {s ∈ C : 0 ≤ Re s ≤ 1} (exercise).

1.5. Generalization. To generalize, we will use a Poisson summation formula in which

Z ⊂ R is replaced by K ⊂ A, where K is a global field and A is its adèle ring. To carry this
out, we must develop integration and Fourier analysis on locally compact abelian groups such
as A.

2. Review of integration
This is not supposed to be a course in analysis. In order to get to the number theory as
quickly as possible, we will state many analysis facts without proof in Sections 2 and 3. Some
references containing the details are [Dei05] and [DE09].

2.1. Measure. Let X be a set, and let M be a collection of subsets of X.

Definition 2.1. Call M a σ-algebra if M is closed under complementation and countable

unions (including finite unions, even empty unions!)

Example 2.2. Let X be a topological space. The collection of Borel sets is the σ-algebra
B = B(X) generated by the open subsets of X.

Fix a σ-algebra M . Elements of M are called measurable sets.

Definition 2.3. A function f : X → C is called measurable if inverse images of measurable

subsets are measurable, i.e., if for every S ∈ B(C), we have f −1 (S) ∈ M .

Remark 2.4. It suffices to check the condition for open disks S ⊂ C. For real-valued f , it
suffices to check the condition for S = (a, ∞) for each a ∈ R.

Definition 2.5. A measure on (X, M ) is a function µ : M → [0, ∞] such that µ( Ai ) =

S
P
µ(Ai ) for any countable (or finite) collection of disjoint measurable sets (Ai ). In the special
case in which M = B, the measure is called a Borel measure.
5
 Think of µ(S) as the “volume” of S.
A set N ⊆ X is called a null set if N is contained in a measure-0 set. It is easy and
convenient to enlarge M so that all null sets are measurable. Call f : X → C a null function
if {x ∈ X : f (x) 6= 0} is a null set.

2.2. Integration. Fix (X, M , µ). Given S ∈ M with µ(S) < ∞, let 1S be the function that
R
is 1 on S and 0 outside S, and define 1S := µ(S). A step function f is a finite C-linear
R
combination of such functions 1S ; define f so that it is linear in f . Also, define the L1 -norm
R
of f by kf k1 := |f | ∈ R≥0 ; this leads to a notion of distance and Cauchy sequence. Call
f : X → C integrable if outside a null set it equals the pointwise limit of some L1 -Cauchy
R R
sequence of step functions (fi ); then define f := limi→∞ fi ∈ C. (It turns out to be
R
well-defined.) We may also write X f dµ to indicate the dependence on X or µ.
If f and g are functions X → [−∞, ∞], then the notation f ≥ g means that f (x) ≥ g(x)
for all x ∈ X. For measurable f ≥ 0, the alternative definition
R R
f := sup g : g is a step function and 0 ≤ g ≤ f .

agrees with the previous one if f is integrable, and yields ∞ if f is not integrable. (If f takes
R
the value ∞ on a non-null set, then f = ∞.)

February 5

For a measurable f : X → C, it turns out that f is integrable if and only if |f | is integrable;

R R
in this case, | f | ≤ |f |.
Here are two important theorems for interchanging limits and integrals:

Monotone convergence theorem. Suppose that (fn ) is a sequence of measurable functions

X → [0, ∞] such that 0 ≤ f1 ≤ f2 ≤ · · · . Let f be the pointwise limit of the fn . Then
R R
fn → f .

Dominated convergence theorem. Let f1 , f2 , . . . and f be measurable functions X → C

such that fn → f pointwise. If there is an integrable function g : X → C such that |fn | ≤ |g|
R R
for all n, then all the fn and f are integrable, and fn → f .

The dominated convergence theorem applies also when the limit limn→∞ fn (x) of a sequence
is replaced by a limit limt→0 f (x, t) of a family of functions depending on a real parameter t.

2.3. Lp -spaces. For p ∈ R≥1 , define

Lp (X) := {measurable f : X → C such that |f |p is integrable}.

R
1/p
Define the Lp -norm of f ∈ Lp (X) by kf kp := |f |p ∈ R≥0 . Strictly speaking, k kp is not
p
a norm on L (X), since there exist nonzero functions f such that kf kp = 0, namely the null
6
functions. But k kp induces a norm on
Lp (X)
Lp (X) :=
{null functions}
making Lp (X) a Banach space. We will often pretend that elements of Lp (X) are functions;
this is OK as far as integration is concerned.
The space L2 (X) is also a Hilbert space, under the inner product
Z
hf, gi := f ḡ.

2.4. Measures and integrals on a locally compact Hausdorff space.

Definition 2.6. A Hausdorff topological space X is called locally compact if every x ∈ X

has a compact neighborhood.

In the rest of this section, X denotes a locally compact Hausdorff space.

Definition 2.7. An outer Radon measure on X is a Borel measure µ : B → [0, ∞] that is

• locally finite: every x ∈ X has an open neighborhood U such that µ(U ) < ∞;
• outer regular: every S ∈ B satisfies µ(S) = inf open U ⊃ S µ(U ); and
• inner regular on open sets: every open U ⊆ X satisfies µ(U ) = supcompact K ⊆ U µ(K).

A function f : X → C has compact support if the closure of {x ∈ X : f (x) 6= 0} is compact.

Define
C(X) := {continuous f : X → C}
Cc (X) := {continuous f : X → C of compact support}.

Definition 2.8. A Radon integral on X is a C-linear map I : Cc (X) → C such that I(f ) ≥ 0
whenever f ≥ 0.

Given an outer Radon measure µ, define

Iµ : Cc (X) −→ C
R
f 7−→ f dµ

Riesz representation theorem. Let X be a locally compact Hausdorff space. Then

∼
{outer Radon measures on X} −→ {Radon integrals on X}
µ 7−→ Iµ
is a bijection.
R
Example 2.9. Let X = Rn . The map sending f ∈ Cc (Rn ) to the Riemann integral Rn f ∈ C
is a Radon integral. Lebesgue measure is defined to be the corresponding outer Radon measure
on Rn .
7
 2.5. Haar measure.

Definition 2.10. A topological group is a topological space G equipped with a group structure
such that multiplication G × G → G and inversion G → G are continuous maps.

Definition 2.11. A Borel measure µ on a topological group G is left-invariant if µ(gS) = µ(S)

for all g ∈ G and S ∈ B. (Right-invariant is defined similarly.)

In the rest of this section, G denotes a locally compact Hausdorff topological group.
Examples:
• R, C, Zp , Qp , A under addition;
• the unit group A× of any of the rings A above;
• GLn (A) for any of these rings A and any n ≥ 1; and
• any group equipped with the discrete topology.

Definition 2.12. A left Haar measure on G is a nonzero left-invariant outer Radon measure
on G.

Theorem 2.13 (Existence and uniqueness of Haar measure). Let G be a locally compact
Hausdorff topological group.
(a) There exists a left Haar measure µ on G (and hence also a corresponding Haar integral).
(b) Every other left Haar measure on G is cµ for some c ∈ R>0 .

Example 2.14. On Rn , Lebesgue measure is a Haar measure.

Example 2.15. On a discrete group, the counting measure assigning measure 1 to each singleton
is a Haar measure.

Warning 2.16. A left Haar measure µ need not be right-invariant. If g ∈ G, then µg (S) :=
µ(Sg) defines a possibly different left Haar measure µg . By Theorem 2.13(b), µg = ∆(g)µ
for some ∆(g) ∈ R>0 . In fact, ∆ : G → R× >0 is a homomorphism, called the modular function
of G. If G is abelian, then left-invariant implies right-invariant, so ∆ ≡ 1. If G is compact,
then ∆ ≡ 1 since ∆(G) is a compact subgroup of R× >0 , and the only such subgroup is {1}.
Thus, if G is abelian or compact, the notions of left Haar!measure and right Haar measure
1 ∗
coincide. (An example with ∆ 6≡ 1 is the subgroup of GL2 (R).)
0 ∗

Remark 2.17. Let µ be a left Haar measure on G. One can show that G is compact if and
only if µ(G) < ∞.

Definition 2.18. Suppose that G is compact. The normalized Haar measure on G is the
unique Haar measure µ such that µ(G) = 1.
8
 3. Duality of locally compact abelian groups
3.1. LCA groups.

Definition 3.1. An LCA group is a locally compact abelian Hausdorff topological group.

LCA groups from a category, with continuous homomorphisms as the morphisms.

Example 3.2. Let T be the unit circle in the complex plane, i.e., the group {z ∈ C : |z| = 1}
under multiplication. Then T ' R/Z as an LCA group.

3.2. Short exact sequences. Call 0 → A → B → C → 0 a short exact sequence of LCA

groups if
• A, B, C are LCA groups,
• the sequence is exact as a sequence of groups without topology, and
• the topologies of A and C are induced by the topology on B (subspace topology and
quotient topology, respectively).
In this case, A must be a closed subgroup of B, since otherwise B/A would not be Hausdorff.
Conversely, given a closed subgroup A of an LCA group B, the sequence 0 → A → B →
B/A → 0 is a short exact sequence of LCA groups.
Let 0 → A → B → C → 0 be a short exact sequence of LCA groups. Given Haar measures
on any two of A, B, C, there exists a unique Haar measure on the third such that the three
Haar measures da, db, dc are compatible in the sense that
Z Z Z
f (b) db = f (ca) da dc (2)
B C A

for every f ∈ Cc (B) (the inner integral on the right is a function of c ∈ B that is constant on
each coset of A, and hence may be viewed as a function on C). In fact, (2) then holds for all
f ∈ L1 (B) [DE09, Theorem 1.5.3].

3.3. Characters.
From now on, G denotes an LCA group.

Definition 3.3. A character of G is a continuous homomorphism χ : G → C× . Let X(G) be

the group of all characters under pointwise multiplication.

Definition 3.4. A unitary character of G is a continuous homomorphism χ : G → T.


Warning 3.5. Some authors call these quasi-characters and characters, respectively.

Proposition 3.6. If G is compact, then every character χ of G is unitary.

Proof. Let |χ| be the character obtained by taking the complex absolute value of each value.
The image of |χ| is a compact subgroup of R× >0 , but the only such subgroup is {1}. 
9
3.4. The Pontryagin dual.

Definition 3.7. The Pontryagin dual of G is the group

b := Homconts (G, T) = {unitary characters of G}.

G

under pointwise multiplication. Equip Gb with the compact-open topology, i.e., the topology
generated by the sets {χ ∈ Gb : χ(K) ⊆ U } for every compact K ⊆ G and open U ⊆ T.

b is another LCA group. Any continuous homomorphism φ : G → H of

It turns out that G
LCA groups induces a continuous homomorphism H b →G b taking χ to χ ◦ φ. In fact, taking
the Pontryagin dual is a contravariant functor from the category of LCA groups to itself.

Theorem 3.8 (Exactness of the Pontryagin dual functor). If

0→A→B→C→0

is a short exact sequence of LCA groups, then so is

0→C
b→B
b→A
b → 0.

Theorem 3.9 (Pontryagin duality). The canonical homomorphism

ˆ
G −→ Ĝ
g 7−→ (χ 7→ χ(g))

is an isomorphism of LCA groups.

There is a continuous bilinear pairing

G×G
b→T

(g, χ) 7→ χ(g).

Pontryagin duality says that the corresponding pairing for G

b is the same with the roles of G
and G
b reversed. If H is a subgroup of G, let

H ⊥ := {χ ∈ G
b : χ(h) = 1 for all h ∈ H}.

Here is a table of some Pontryagin duals (we will prove some of these only later). Let K
be a global field, and let A be its adèle ring.
10
 G Gb
R R
Qp Qp
A A
Z R/Z
Zp Qp /Zp
K A/K
finite finite
discrete compact
discrete torsion profinite
Example 3.10. The self-duality of R is given by the pairing
R×R→T
(x, y) 7→ e2πixy .

In other words, we are claiming that the homomorphism R → R b sending y to the unitary
character χy (x) := e2πixy is an isomorphism. (This will be proved later on.)

3.5. The Fourier transform. Recall: If f ∈ S (R), then

Z
f (y) :=
b f (x) e−2πixy dx.
R

Definition 3.11. If f ∈ L1 (G), define the Fourier transform fb: G

b → C by
Z
fb(χ) := f (g) χ(g) dg.
G

b → C is always continuous.
It turns out that fb: G

February 10
Cancelled due to snow!

February 12

Fourier inversion formula. Let G be an LCA group. Let dg be a Haar measure on G.

Then there exists a unique Haar measure dχ on G, b called the dual measure or Plancherel
measure, such that if f ∈ L1 (G) is such that fb ∈ L1 (G),
b then
Z
f (g) := fb(χ) χ(g) dχ. (3)
G
b

for almost all g ∈ G (i.e., all g outside a null set). If, moreover, f is continuous, then (3)
holds for all g ∈ G.
11
Remark 3.12. Changing the values of f on a null set does not change fb, so it is impossible to
recover f exactly from fb. This explains why in the theorem it is necessary to exclude a null
set, or to assume that f is continuous.
ˆ
Remark 3.13. Equation 3 is equivalent to the identity fˆ(g) = f (−g) for almost all g.
b respectively. If f ∈
Plancherel theorem. Let dg and dχ be dual measures on G and G,
L1 (G) ∩ L2 (G), then kf k2 = kfbk2 .

(The hypothesis f ∈ L1 (G) ensures that fb is defined, and f ∈ L2 (G) ensures that kf k2 is
defined.)

Corollary 3.14. The Fourier transform

L1 (G) ∩ L2 (G) −→ L2 (G)
b

f 7−→ fb

extends to a map L2 (G) → L2 (G),

b and this extended Fourier transform is an isomorphism of
Hilbert spaces.

Proof. The Plancherel theorem implies that the Fourier transform maps L2 -Cauchy sequences
in L1 (G) ∩ L2 (G) to L2 -Cauchy sequences, so the Fourier transform can be extended to the
completion of L1 (G) ∩ L2 (G) with respect to k k2 . That completion is all of L2 (G), since
already the subspace Cc (G) of L1 (G) ∩ L2 (G) is dense in L2 (G) (with respect to k k2 ). (In
fact, Cc (G) is dense in Lp (G) for each p ≥ 1.)
The Plancherel theorem implies that the extended Fourier transform preserves L2 -norms.
The extended Fourier transform on G, b combined with composition-with-negation-on-G,
provides the inverse map. 

Example 3.15. If G is discrete, the counting measure on G is dual to the normalized Haar
measure on the compact group G. b To see this, check the Fourier inversion formula or the
Plancherel theorem for the function f : G → C that is 1 at the identity and 0 everywhere else.

Remark 3.16. Many results of Section 3 can be extended to locally compact Hausdorff
topological groups that are not abelian. But the theory becomes much more difficult, for
the same reason that representation theory of finite nonabelian groups is more difficult than
representation theory of finite abelian groups: instead of unitary characters G → T one
must consider representations of dimension greater than 1 (and even infinite-dimensional
representations, when the nonabelian group is not compact).

4. Local fields
In this section and the next, we follow Tate’s thesis as well as [RV99, Chapter 7].
12
4.1. Characterizations of local fields.

Definition 4.1. A local field is a field F satisfying one of the following equivalent conditions:
(1) F is R or C, or else F is the fraction field of a complete discrete valuation ring with
finite residue field.
(2) F is a finite separable extension of R, Qp , or Fp ((t)) for some prime p. (Moreover,
every finite extension of Fp ((t)) is isomorphic to Fq ((u)) for some power q of p.)
(3) F is the completion of a global field with respect to a nontrivial absolute value.
(4) F is a nondiscrete locally compact topological field.
See [RV99, Theorem 4-12] for a proof of the difficult part of the equivalence, namely that
nondiscrete locally compact topological fields satisfy the other conditions. We won’t use this
part — for us, it serves only as extra motivation for considering these particular fields.

From now on, F denotes a local field.

If F is R or C, then F is called archimedean; other local fields are called nonarchimedean.
In the nonarchimedean case, we use the following notation:

O : the valuation ring

p : the maximal ideal of O
k : the residue field O/p
$ : a uniformizer
p : the characteristic of k
q := #k.

4.2. The normalized absolute value. Let F be a local field. The additive group of F
is an LCA group. Let dx be a Haar measure on F . If a ∈ F × , multiplication-by-a is an
isomorphism F → F , under which dx pulls back to another Haar measure, which must be a
positive multiple of dx:
a
F −→ F
|a| dx ←− dx

for some “stretching factor” |a| ∈ R× >0 . Replacing dx by a positive real multiple does not
change |a|, so |a| is independent of the choice of dx. Then
• |a| = the ordinary absolute value if F = R
• |a| = the square of the ordinary absolute value if F = C
• |a| = #(O/aO)−1 if F is nonarchimedean and a ∈ O.
13

Warning 4.2. In the case F = C, we might sometimes write k k to emphasize that we mean
the square of the ordinary absolute value. The function k k is not literally an absolute value
since the triangle inequality does not quite hold.

A subset S ⊆ F is called bounded if there exists B > 0 such that |x| ≤ B for all x ∈ S.
The notion extends to subsets of F n for any n ≥ 0 by using the sup norm |(x1 , . . . , xn )|sup :=
sup{|xi | : 1 ≤ i ≤ n}.

Theorem 4.3 (Heine–Borel theorem for local fields). Let S be a subset of a local field F .
Then S has compact closure if and only if S is bounded. The same holds for S ⊆ F n for any
n ≥ 0.

The nonarchimedean case will be assigned as homework.

4.3. Additive characters. By an additive character, we mean a nontrivial unitary character

ψ : F → T of the additive group of F . Fix ψ. Given a ∈ F , define ψa (x) := ψ(ax); this is
another unitary character of F .

Theorem 4.4. The map

Ψ : F → Fb
a 7→ ψa
is an isomorphism of LCA groups.

Proof. Checking that Ψ is an injective homomorphism is easy.

In comparing the original topology on F with the topology corresponding under Ψ to the
subspace topology on Ψ(F ), it is enough to consider neighborhoods of 0, because F is a
topological group under both topologies. For the second topology a basis of neighborhoods
of 0 is given by the sets
{a ∈ F : ψa (K) ⊆ U } = {a ∈ F : aK ⊆ ψ −1 U }
for compact K and open U 3 1. To check:
(1) Given compact K and open U 3 1, does there exist δ > 0 such that
|a| < δ =⇒ aK ⊆ ψ −1 U ?
(2) Given  > 0, does there exist a compact K and open U 3 1 such that
|a| <  ⇐= aK ⊆ ψ −1 U ?
The answer to (1) is yes, since K is bounded and ψ −1 U contains an open disk around 0. The
answer to (2) is yes also: choose b ∈ F such that ψ(b) 6= 1, choose U 3 1 so that ψ(b) ∈/U
−1
and hence b ∈
/ ψ U , and choose K to be a closed disk centered at 0 of radius at least |b|/;
14
then aK ⊆ ψ −1 U implies b ∈
/ aK, so |b| > |a| · |b|/, so |a| < . Thus Ψ is a homeomorphism
onto its image.
Since F is locally compact, F is complete, so Ψ(F ) is complete, so Ψ(F ) is closed in Fb.
The remaining claim Ψ(F ) = Fb is equivalent to Ψ(F )⊥ = {0}, since
⊥ : {closed subgroups of Fb} → {closed subgroups of F }
is a order-reversing bijection (homework). We prove Ψ(F )⊥ = {0}: If x ∈ Ψ(F )⊥ , then
ψa (x) = 0 for all a ∈ F , so ψ(ax) = 0 for all a ∈ F , so x = 0. 

Definition 4.5. Suppose that F is nonarchimedean. Then ψ|pm = 1 for some m ∈ Z; choose
the smallest such m. The fractional ideal pm is called the conductor of ψ.

There is a standard ψ on each F :

• If F = R, let ψ(x) := e−2πix . (The minus sign is there so that a global product
formula later on will hold.)
• If F = Qp , let ψ be the composition
Qp ∼ Z[1/p] R
Qp  ← ,→ ' T,
Zp Z Z
n
which is characterized by ψ|Zp = 1 and ψ(1/pn ) = e2πi/p for all n ≥ 1.
• If F = Fp ((t)), define ψ( ai ti ) := e2πia−1 /p (lift a−1 ∈ Fp to Z to make sense of this).
P

• Finally, if F0 is one of the three fields above, and ψ0 is the additive character
chosen for F0 , and F is a finite separable extension of F0 , let ψ be the composition
TrF/F ψ0
F −→0 F0 −→ T.
4.4. Schwartz–Bruhat functions. Recall that a Schwartz function R → C is a C ∞ function
whose derivatives tend to 0 rapidly. The same definition applies to functions on Rn for any
n ≥ 0, using partial derivatives. In particular, it applies to functions on C, viewed as R2 .
If F is nonarchimedean, however, one cannot take the derivative of a function F → C, so
instead of requiring that a function be C ∞ , we require it to be locally constant.

Definition 4.6. A function f : F → C is called a Schwartz–Bruhat function if it is


a Schwartz function if F = R or F = C,
a locally constant function of compact support if F is nonarchimedean.

Let S = S (F ) be the C-vector space of Schwartz–Bruhat functions.

If F is nonarchimedean and f ∈ S , then the support of f is covered by finitely many open

disks Di on which f is constant, and by refining the cover we may assume that the Di are
disjoint; then f is a finite C-linear combination of the characteristic functions 1Di .

February 17
15
 Presidents’ Day holiday.

February 19

4.5. The Fourier transform for local fields.

Definition 4.7. Fix a local field F , a nontrivial additive character ψ on F , and a Haar
measure dx on F . Given f ∈ S , define the Fourier transform fb by
Z
f (y) :=
b f (x) ψ(xy) dx.
F

(Tate originally took the complex conjugate of the additive character, but many references since then have not done so.) It turns
out that fb ∈ S .
∼
Under the isomorphism F → Fb induced by ψ, the measure on Fb dual to dx might not pull
ˆ
back to dx, but in any case fˆ(x) = rf (−x) for some r ∈ R>0 . We could arrange r = 1 by
scaling dx appropriately; then the Fourier inversion formula
Z
f (x) = fb(y) ψ(xy) dy
F

holds for all x. (There is no need to exclude a null set, since Schwartz–Bruhat functions are
continuous.)

Proposition 4.8. The unique Haar measure dx that is self-dual (relative to the standard ψ)
is described explicitly as follows:
• If F = R, then dx is Lebesgue measure.
• If F = C, then dx is twice Lebesgue measure.
• If F is nonarchimedean, then dx is the Haar measure for which O gets measure
(N D)−1/2 , where D is the different of F/F0 , and N D := #(O/D).

Sketch of proof. It suffices to check the Fourier inversion formula for a single nonzero f ∈ S .
ˆ
• For R, take f (x) := e−πx . A calculation shows that fb = f , so fˆ = f .
2

ˆ
• For C, take f (z) := e−2πzz̄ . A calculation shows that fb = f , so fˆ = f .
• For nonarchimedean F , take f = 1O . Then

Z Vol(O), if ψ| = 1,
Oy
f (y) :=
b ψ(xy) dx =
O 0, if ψ|Oy is a nontrivial character.

The first case holds if TrF/F0 (Oy) is contained in the valuation ring O0 of F0 , i.e., if
ˆ
y ∈ D−1 . Thus fb = (N D)−1/2 1D−1 . A similar calculation shows that fˆ = f .
Moreover, in all three cases, f (x) = f (−x), so the Fourier inversion formula is verified. 
16
4.6. Multiplicative characters. Define

{±1}, if F = R,


×
U := {x ∈ F : |x| = 1} = T, if F = C,

O× ,

if F is nonarchimedean,


R× , if F is archimedean,
× × >0
|F | := {|x| : x ∈ F } =
value group q Z if F is nonarchimedean.

We have a short exact sequence of LCA groups

||
1 −→ U −→ F × −→ |F × | −→ 1 (4)

Call χ ∈ X(F × ) unramified if χ|U = 1.

Remark 4.9. Let’s explain the reason for this terminology. Suppose that F is nonarchimedean. A character η
of Gal(F ab /F ) factors as Gal(F ab /F )  Gal(L/F ) ,→ C× for some finite Galois extension L of F . Let I be
the inertia group of Gal(F ab /F ), which maps onto the inertia group IL/F of Gal(L/F ). The following are
equivalent:
• L/F is unramified.
• IL/F = 1.
• η|I = 1.
Therefore, call η unramified if η|I = 1.
Let θ : F × → Gal(F ab /F ) be the local Artin homomorphism, which maps U isomorphically to I. Let χ be
θ η
the composition F × → Gal(F ab /F ) → C× . Then the three conditions above are equivalent also to
• χ|U = 1.

Proposition 4.10. For a character χ : F × → C× , the following are equivalent:

(i) χ is unramified (i.e., χ|U = 1);
(ii) χ factors through |F × |;
(iii) χ = | |s for some s ∈ C.

Proof.
(i)⇔(ii): Use (4).
(iii)⇒(ii): Trivial.
(ii)⇒(iii): We must show that every character χ of |F × | is x 7→ xs for some s ∈ C. If
|F × | = q Z , it is enough to choose s so that q s = χ(q). If |F × | = R×
>0 , we use the following
observations:
• R×
>0 is isomorphic to R;
• R is simply connected, so every character R → C× factors through the universal cover
exp
C  C× ;
17
 • Every continuous homomorphism η : R → C is r 7→ sr for some s ∈ C: namely, η(1)
determines η|Z , which determines η|Q (since C is divisible and torsion-free), which
determines η on R by continuity. 

Corollary 4.11. The unramified characters of F × form a subgroup of X(F × ) isomorphic to

C (if F is archimedean) or C/ (Z · 2πi/ log q) (if F is nonarchimedean).

Thus the subgroup of unramified characters has the structure of a Riemann surface, and
so does each coset. Viewing X(F × ) as the disjoint union of these cosets makes X(F × ) a
Riemann surface with infinitely many connected components.

Proposition 4.12. Every character χ of F × is η | |s for some unitary character η and some
s ∈ C.
χ χ
Proof. Factor χ as |χ| · |χ|. The first factor |χ| is unitary. The second factor |χ| is unram-
ified since χ restricted to the compact group U is automatically unitary, so by Proposi-
tion 4.10(i)⇒(iii) |χ| is | |s for some s ∈ C. 

Corollary 4.13. If χ ∈ X(F × ), then |χ| = | |σ for a uniquely determined σ ∈ R.

Proof. If χ = η | |s with η unitary, then |χ| = | |σ with σ := Re s. 

The number σ in Corollary 4.13 is called the exponent of χ.

Definition 4.14. Suppose that F is nonarchimedean. Then χ|1+pm = 1 for some m ∈ Z≥0
(we interpret 1 + p0 as U = O× ); choose the smallest such m. The ideal pm is called the
conductor of χ. The conductor of χ is the unit ideal if and only if χ is unramified, so the
conductor measures the extent to which χ is ramified.

In the archimedean cases, we can classify the characters of F × in an even more explicit
way:

Corollary 4.15.
(a) Every character of R× has the form χa,s (x) := x−a |x|s for some a ∈ {0, 1} and s ∈ C.
(b) Every character of C× has the form χa,b,s (z) := z −a z̄ −b kzks for some a, b ∈ Z with
min(a, b) = 0 and some s ∈ C. (Here z̄ is the complex conjugate of z.)

Proof. If χ ∈ X(R× ), then χ|U is given by x−a for some a ∈ {0, 1}, so xa χ(x) is unramified,
and hence equals |x|s for some s ∈ C. A similar argument works for C× . 
Remark 4.16. The character x 7→ x−1 of R× is not unitary, but the sign character sgn(x) := x−1 |x| is. The
reason for writing x−a |x|s instead of just the cases | |s and sgn ·| |s is to simplify the definition of local
L-factors later on. A similar comment applies to characters of C× .
18
 In what follows, L-factors and zeta integrals will be not functions of s ∈ C, but functions
of χ ∈ X(F × ).
The Riemann zeta function had a functional equation relating its values at s and 1 − s.
What operation on characters sends | |s to | |1−s ? Answer:

Definition 4.17. If χ ∈ X(F × ), define its twisted dual as χ∨ := χ−1 | |.

4.7. Local L-factors. Each factor in

Y
ζ(s) = (1 − p−s )−1
p

is an example of a local L-factor. If F = Qp and χ = | |s , then p−s = χ($). This partially

motivates the following definition.
If F is nonarchimedean, and χ ∈ X(F × ), then

(1 − χ($))−1 if χ is unramified,
L(χ) :=
1 if χ is ramified.

We need to define L(χ) also in the archimedean cases:

L(χa,s ) := ΓR (s) := π −s/2 Γ(s/2),

L(χa,b,s ) := ΓC (s) := ΓR (s)ΓR (s + 1) = 2(2π)−s Γ(s).

(Recall that π −s/2 Γ(s/2) was the archimedean factor used to complete the Riemann zeta function. The
definition over C can be explained in terms of induced representations when the definition is extended to
Call L(χ) the local L-factor of χ.
higher-dimensional representations of Weil groups.)
In all cases, L(χ) is a meromorphic function of χ ∈ X(F × ) with no zeros.

4.8. Multiplicative Haar measure. Fix a Haar measure d× x on F × . Then d× x = c |x| dx

for
some c ∈ R>0 , by which we mean that the Radon integrals agree: for all f ∈ Cc (F × ),
Z Z
× c
f (x) d x = f (x) dx.
F× F |x|
4.9. Local zeta integrals. Fix d× x. Given f ∈ S and χ ∈ X(F × ), define the local zeta
integral Z
Z(f, χ) := f (x) χ(x) d× x.
F×
We are now ready for the main theorem in the local setting:

Theorem 4.18 (Meromorphic continuation and functional equation of local zeta integrals).
(a) For every f ∈ S , the integral Z(f, χ) converges for χ of exponent σ > 0.
(b) For every f ∈ S , the function Z(f, χ) of χ extends to a meromorphic function on X(F × ).
(c) For every f ∈ S , the meromorphic function Z(f, χ)/L(χ) on X(F × ) is holomorphic.
19
(d) For each component X0 of X(F × ), there exists f ∈ S such that the holomorphic function
Z(f, χ)/L(χ) is nonvanishing on X0 . Moreover, if F is nonarchimedean, X0 = {| |s : s ∈
C}, and f = 1O , then Z(f, χ)/L(χ) is 1 on X0 .
(e) Fix choices of ψ and dx (to define Fourier transforms). There exists a nonvanishing
holomorphic function (χ, ψ, dx) of χ ∈ X(F × ) such that

Z(fb, χ∨ ) Z(f, χ)
∨
= (χ, ψ, dx) (5)
L(χ ) L(χ)
for all f ∈ S . For χ = η | |s for fixed η (i.e., on one component of X(F × )), the function
(χ, ψ, dx) has the form AeBs for some A, B ∈ C. Moreover, if F is nonarchimedean, ψ
R
has conductor p0 , and O dx = 1 (so dx is self-dual relative to ψ), then (| |s , ψ, dx) = 1
for all s ∈ C.

Parts (c) and (d) can be viewed as saying that the function L(χ) is the “greatest common
divisor” of the functions Z(f, χ) as f varies. Part (e) implies that Z(fb, χ∨ ) is related to
Z(f, χ) by a meromorphic factor that is independent of f . In (e), the function (χ, ψ, dx)
is called a local -factor; it is independent of the choice of d× x, but it depends on ψ and dx
since these appear in the definition of fb.

Remark 4.19. In his thesis, Tate had a single function measuring the “multiplicative error” in
the functional equation relating Z(f, χ) and Z(fb, χ∨ ). Only later, in [Del73, §3], did Deligne
separate this function into L-factors and an -factor.

Proof of convergence for σ > 0. We will prove absolute convergence. Since |χ(x)| = |x|σ , the
question is whether F × |f (x)| |x|σ d× x < ∞. Since f ∈ S , the integrand decays rapidly as
R

x → ∞, so the integral over the region |x| > 1 is finite. On the other hand, f is bounded
in the region |x| ≤ 1. Thus it suffices to prove finiteness of I := 0<|x|≤1 |x|σ d× x. Choose
R

a ∈ F × with |a| < 1. For n ∈ Z≥0 , let An be the annulus |a|n+1 < |x| ≤ |a|n , and let
In := An |x|σ d× x, so I = n≥0 In . Since An has compact closure in F × and |x|σ is bounded
R P

on An , we have In < ∞. Also, In = |an |σ I0 , by translation invariance of d× x. Thus n≥0 In

P

is a convergent geometric series if σ > 0. 

This proves part (a) of Theorem 4.18.

February 24

Proof that Z(f, χ) and Z(f, χ)/L(χ) are holomorphic for σ > 0. Write χ = η | |s with η uni-
tary and s ∈ C. For Re s > 0, the derivative of Z(f, χ) = F × f (x) η(x) |x|s d× x with respect
R

to s exists (differentiate under the integral sign, by using absolute convergence). Since L(χ)
has no zeros, Z(f, χ)/L(χ) is holomorphic in this region too. 
20
 We next prove the functional equation for one f for each X0 . Each X0 is {η | |s : s ∈ C}
for some unitary character η ∈ Fc× (determined up to multiplication by | |ir for r ∈ R). A
function of χ = η | |s ∈ X0 may be viewed as a function of s ∈ C.

Lemma 4.20. For each η ∈ Fc× , there exists f ∈ S such that for χ = η | |s ,

(a) The holomorphic function Z(f, χ)/L(χ) on the half-plane Re s > 0 is nonvanishing (and
equal to 1 if F is nonarchimedean, η = 1, and f = 1O ).
(b) The holomorphic function Z(fb, χ∨ )/L(χ∨ ) on the half-plane Re s < 1 is nonvanishing.
(c) There exists a nonvanishing holomorphic function (s) = (χ, ψ, dx) of the form AeBs
such that the functional equation (5) holds on the strip 0 < Re s < 1 (and  = 1 if F is
R
nonarchimedean, ψ has conductor p0 , and O dx = 1).

Proof. Changing ψ or dx changes Z(fb, χ∨ ) by an easy factor, so it suffices to prove this for
one choice of ψ and dx. We may also choose d× x.
Case 1: F = R. Let ψ(x) := e−2πix , let dx be Lebesgue measure, and let d× x = dx/|x|.
By Corollary 4.15(a), it suffices to consider η = 1 and η = sgn.
2
• Suppose that η = 1. Let f (x) := e−πx . Then for Re s > 0,

Z
2
Z(f, χ) = e−πx |x|s d× x
R×
Z ∞
2 dx
=2 e−πx xs
0 x
Z ∞
du
= π −s/2 e−u us/2 (we let u = πx2 )
0 u
= π −s/2 Γ(s/2)
= ΓR (s)
= L(χ).

Also fb = f , so Z(fb, χ∨ ) = L(χ∨ ) for Re s < 1. Set  := 1.

2
• Suppose that η = sgn, i.e., η(x) = x−1 |x|. Choosing f (x) := e−πx would give
2
Z(f, χ) = 0 by symmetry, so instead let f (x) := xe−πx . A calculation similar to that
above leads to  := −i.

Case 2: F = C. Let ψ(z) := e−2πi(z+z̄) , let dx be twice Lebesgue measure, and let
d× x = dx/kxk. By Corollary 4.15(b), it suffices to consider the cases where η is z −a kzka/2
or z̄ −b kzkb/2 for some a, b ∈ Z≥0 . By symmetry (changing the identification of F with C), it
suffices to consider z −a kzka/2 . Let f (z) := π −1 z a e−2πzz̄ . A calculation in polar coordinates
21
leads to

Z(f, χ) = ΓC (s + a/2),
L(χ) = ΓC (s + a/2),
fb(z) = (−iz̄)a e−2πzz̄ ,
χ∨ = z̄ −a kzk1−s+a/2 ,
Z(fb, χ∨ ) = (−i)a ΓC (1 − s + a/2),
L(χ∨ ) = ΓC (1 − s + a/2),
 := (−i)a .

Case 3: F is nonarchimedean.
Choose dx so that O dx = 1; then pk dx = q −k for all k ∈ Z. Choose d× x = dx/|x|; then
R R
× −1
d× x = q −k for each k ≥ 1.
R R
O × d x = 1 − q and 1+pk
The calculations for ramified η will depend on the following properties of Gauss sums:

Proposition 4.21. Given unitary characters ω : O× → T and ψ : O → T, define the Gauss

sum
Z
g(ω, ψ) := ω(x) ψ(x) d× x.
O×

Suppose ω is of conductor pn with n > 0, and ψ is of conductor pm .

(i) If m 6= n, then g(ω, ψ) = 0.
(ii) If m = n, then |g(ω, ψ)|2 = q −m . (Here | | is the usual absolute value on C, so |z|2 = z z̄.)

Proof.
(i) If m > n, then the integral over each coset of 1 + pn is 0 since ω is constant and ψ is a
nontrivial character on pn . If m < n, then the integral over each coset of 1 + pm is 0
since ψ is constant and ω is a nontrivial character on 1 + pm (interpret 1 + pm as O× if
m = 0).
(ii) If m = n > 0, then
Z Z
2 ×
|g(ω, ψ)| = ω(x) ψ(x) d x ω(y) ψ(y) d× y
O× O×
Z Z
= ω(xy −1 ) ψ(x − y) d× x d× y
× ×
ZO ZO
= ω(z) ψ(yz − y) d× y d× z (let x = yz)
O× O×
Z
= ω(z) h(z) d× z
O×
22
 where
Z
h(z) = ψ(yz − y) d× y
O×
Z
= ψ(y(z − 1)) dy (since |y| = 1 on O× )
O×
Z Z
= ψ(y(z − 1)) dy − ψ(y(z − 1)) dy
O p

−1
1 − q if v(z − 1) ≥ m (both integrands are 1),


= −q −1 if v(z − 1) = m − 1 (second integrand is 1),


0 if v(z − 1) < m − 1 (neither integrand is constant),


= 11+pm (z) − q −1 11+pm−1 (z).

Thus
Z Z
× −1
|g(ω, ψ)| = 2
ω(z) d z − q ω(z) d× z = q −m − 0 = q −m . 
1+pm 1+pm−1

Now we return to Case 3 of the zeta integral calculations. Choose ψ : F → T to be of

conductor p0 . Every χ can be written η | |s where η is a unitary character of O× extended to
F × by setting η($) = 1, so assume that η has this form. Let pn be the conductor of η. We
will assume n > 0, and leave the easier case n = 0 as homework. We choose f = 11+pn , since
this makes Z Z
s ×
Z(f, χ) = η(x)|x| d x = dx = q −n 6= 0.
1+pn 1+pn
Next,
Z Z Z
fb(y) = ψ(xy) dx = ψ((1 + z)y) dz = ψ(y) ψ(zy) dz = q −n ψ(y) 1p−n (y),
1+pn pn pn

since ψ|pn y = 1 if and only if y ∈ p−n . For Re s < 1,

Z
∨
Z(fb, χ ) = fb(x) χ∨ (x) d× x
F ×
Z
−n
=q ψ(x) η(x)−1 |x|1−s d× x
p−n
X Z
= q −n ψ(x) η(x)−1 (q −k )1−s d× x
k≥−n pk −pk+1

X Z
−n −k(1−s)
=q q ψ($k z) η($k z)−1 d× z (substitute x = $k z)
k≥−n O×
X
= q −n q n(1−s) g(η −1 , ψ$k )
k≥−n

= q −n q n(1−s) g(η −1 , ψ$−n ),

23
by Proposition 4.21, since the conductor p−k of ψ$k equals the conductor pn of η −1 if and
only if k = −n. Also, since n > 0, the characters χ and χ∨ are ramified, so L(χ) = 1 and
L(χ∨ ) = 1. Hence the functional equation holds with

 := q n(1−s) g(η −1 , ψ$−n ). 

To pass from knowing the functional equation (5) for one f to knowing it for an arbitrary
g ∈ S , Tate used Fubini’s theorem to prove the following, which measures the “difference”
between (5) for f and (5) for g.

Lemma 4.22. Let f, g ∈ S . For χ of exponent σ satisfying 0 < σ < 1, we have

g , χ∨ ) = Z(g, χ) Z(fb, χ∨ ).
Z(f, χ) Z(b

Proof. We may assume that d× x = dx/|x|. By definition,

Z Z Z 
∨ ×
Z(f, χ) Z(b
g, χ ) = f (x) χ(x) d x g(z) ψ(yz) dz χ(y)−1 |y| d× y
F× F×
| F
| {z }
{z } χ∨ (y)
g
b(y)
Z
= f (x) g(z) χ(xy −1 ) ψ(yz) |yz| d× x d× y d× z (Fubini’s theorem)
(F × )3
Z
= f (x) g(z) χ(t−1 ) ψ(txz) |txz| d× x d× t d× z (set y = xt),
(F × )3

which is symmetric in f and g (the dummy variables x and z can be renamed z and x). 

Proof of Theorem 4.18. We already proved (a). Let f and  be as in Lemma 4.20, so
Z(fb, χ∨ ) Z(f, χ)
∨
= (χ, ψ, dx)
L(χ ) L(χ)
for 0 < σ < 1. Multiply by the identity of Lemma 4.22, and cancel Z(f, χ)Z(fb, χ∨ ) to obtain
an identity of meromorphic functions
g , χ∨ )
Z(b Z(g, χ)
∨
= (χ, ψ, dx) ,
L(χ ) L(χ)
again for 0 < σ < 1. The two sides define holomorphic functions in σ < 1 and σ > 0,
g , χ∨ )/L(χ∨ ) on all of X(F × ).
respectively, so they glue to give a holomorphic extension of Z(b
Dividing by (χ, ψ, dx) gives the corresponding extension of Z(g, χ)/L(χ), which proves (c).
Multiplying by the meromorphic function L(χ) yields (b). Finally, (d) and (e) follow from
Lemma 4.20. 

February 26

24
 5. Adèle ring of a global field
5.1. Global fields. A number field is a finite extension of Q. A global function field is a finite
extension of Fp (t) for some prime p, or equivalently is the function field of a geometrically
integral curve over a finite field Fq , where q is a power of some prime p. Notation:

K : global field (number field or global function field)

Kv : the completion of K at v, a local field
Ov : the valuation ring of Kv (set Ov := Kv if v is archimedean)
pv : the maximal ideal of Ov
kv : the residue field Ov /pv
| |v : normalized “absolute value” on Kv
Y0
A := (Kv , Ov ) adèle ring
v
Y0
A× = (Kv× , Ov× ) idèle group.
v
Q
A basis for the topology on A consists of the subsets v Uv , where Uv is open in Kv , and
Uv = Ov for all but finitely many v. A basis for the topology on A× consists of the subsets
× ×
Q
v Uv , where Uv is open in Kv , and Uv = Ov for all but finitely many v. Both the additive
group A and the multiplicative group A× are LCA groups.
Recall that the diagonally embedded K ⊂ A is analogous to Z ⊂ R: for example, K is
discrete and A/K is compact.

5.2. Additive characters. As on a local field, additive characters of A are assumed to be

unitary.

Proposition 5.1. We have an isomorphism

Y0
b −→
A (K \
bv , K v /Ov )
v
ψ 7−→ (ψ|Kv )
Y
ψv ←−[ (ψv ).

In other words, to give an additive character ψ of A is the same as giving a collection (ψv )
such that each ψv is an additive character of Kv , and ψv |Ov = 1 for all but finitely many v.

Proof. This is a formal property that holds for any restricted direct product 0v (Gv , Hv )
Q

where Gv is an LCA group and Hv ≤ Gv is an open compact subgroup for all but finitely
many v. 
Q
• If K is a number field, choose the standard ψv on each Kv , and choose v ψv as the
standard ψ.
25
 • Now suppose that K is the function field of a curve X over Fq . Let ΩX be its sheaf
of 1-forms. Let ΩK be the stalk at the generic point, so ΩK is the 1-dimensional
K-vector space of “meromorphic” 1-forms on X. Define Ωv := ΩK ⊗ Kv . There is a
K
residue map Resv : Ωv → kv , which can be described in terms of Laurent expansions
as follows: if u is a uniformizer at the closed point v, then the residue map is
Resv : Ωv = kv ((u)) du −→ kv
X
ai ui du 7−→ a−1
i∈Z

(this is independent of the choice of uniformizer). A choice of nonzero global mero-

morphic 1-form ω ∈ ΩK gives rise to
 
2πi
ψv (x) := exp Trkv /Fp Resv (xω)
p
for each v. Let κv be the “order of vanishing of ω at v”, i.e., the largest integer such
that ω ∈ pκv v ΩX,v ; then the conductor of ψv is p−κ
Q
v
v
. Let ψv serve as a “standard”
ψ.
For all but finitely many v,
• (number field case) v is nonarchimedean and K/Q is unramified at v
• (function field case) ω generates the OX,v -module ΩX,v (i.e., ω is regular and nonvan-
ishing at v),
in which case ψv has conductor p0v .
From now on, ψ is a standard additive character as above. For a ∈ A, define ψa (x) := ψ(ax).

Corollary 5.2. The map

Ψ: A → A
b

a 7→ ψa
is an isomorphism of LCA groups.
∼ b
Proof. Each ψv gives rise to an isomorphism Ψv : Kv → K v . For all but finitely many v, the
0 ⊥ \
character ψv has conductor pv , so Ov = Ov , and Ψv identifies Ov with K v /Ov . Now Ψ is the
isomorphism
Y0 Q
Ψv Y0
(Kv , Ov ) −→ (K \
bv , K v /Ov ).
v v


Proposition 5.3. We have ψ|K = 1.

Sketch of proof. The number field case is left as an exercise. In the function field case, it
reduces to the fact that for a meromorphic 1-form on a smooth projective integral curve over
an algebraically closed field (e.g., Fq ), the sum of the residues is 0. 
26
Corollary 5.4. The map
[ = K⊥
K → A/K
a 7→ ψa

is an isomorphism.

Proof. We have
(1) K ⊥ is discrete. (Proof: It is the Pontryagin dual of the compact group A/K.)
(2) K ⊥ is a K-subspace of A b ' A. (Proof: If ψa |K = 1 and κ ∈ K, then ψκa |K = 1.)

By (1), K ⊥ /K is a discrete subgroup of the compact group A/K, so K ⊥ /K is finite. By (2),

K ⊥ /K is also a K-vector space. But K is infinite! Thus K ⊥ /K = 0, so K ⊥ = K. 

5.3. The Tamagawa measure. Let dxv be the self-dual measure on Kv with respect to
Q
the standard ψv . The Tamagawa measure dx := dxv on A, is the Haar measure such that
Q R QR
for each basic open set Uv , we have Q Uv dx = Uv
dxv .
The counting measure on K and the measure dx on A induce a Haar measure on A/K
compatible with respect to
0 → K → A → A/K → 0.

Proposition 5.5. The volume of A/K is 1.

Proof. We’ll give the proof in the number field case. Later we’ll explain a calculation-free
way of proving this for any global field.
Let K be a number field. Let D be a fundamental domain for the quotient A/K, i.e., a
`
measurable subset of A such that A is the disjoint union κ∈K (D + κ). For any fundamental
domain D, we have Vol(A/K) = Vol(D). By strong approximation,
Y Y
Kv × Ov −→ A/K
v|∞ v finite
| {z }
KR

is surjective; the kernel is OK (ring of integers) embedded diagonally. Therefore, let D∞ be a

Q
fundamental domain for KR /OK , and let D = D∞ × v finite Ov . Then
Y
Vol(D) = Vol(D∞ ) Vol(Ov )
v finite
Y
= (disc OK )1/2 (N Dv )−1/2
v finite
Q
(the first volume was calculated last semester). Since v finite N Dv = disc OK , we get
Vol(D) = 1. 
27
5.4. The adelic Fourier transform. If for each v we have fv ∈ S (Kv ), and fv = 1Ov for
Q
all but finitely many v, then fv : A → C is defined. A Schwartz–Bruhat function f : A → C
is a finite C-linear combination of such functions. Let S = S (A) be the space of all
Schwartz–Bruhat functions.
Fix the standard ψ and the self-dual measure dx. Given f ∈ S , the Fourier transform fb
is defined as usual, Z
f (y) :=
b f (x) ψ(xy) dx,
A

and it is again in S , since the Fourier transform of fv as above is fbv , where fbv = 1Ov
Q Q

for all v for which Ov⊥ = Ov (all but finitely many).

There is also a Fourier transform for A/K and its Pontryagin dual K. Functions on A/K
will be identified with functions on A that are K-periodic (i.e., f (x + κ) = f (x) for all κ ∈ K).
Given f ∈ L1 (A/K), define fb: K → C by
Z
f (κ) :=
b f (x) ψ(κx) dx.
D

(If we do not take for granted that Vol(D) = 1, then the right side should be multiplied by
1 dx
Vol(D)
since the dual of the counting measure on K is the normalized Haar measure Vol(D) on
1 1
A/K.) For continuous f ∈ L (A/K), if fb ∈ L (K), then the Fourier inversion formula
X
f (x) = fb(k) ψ(κx)
κ∈K

holds (an integral with respect to counting measure is a sum).

5.5. The adelic Poisson summation formula.

Lemma 5.6. If f ∈ S (A), then κ∈K f (x + κ) converges, absolutely and uniformly on

P

compact sets, to a K-periodic function.

Q
Proof. We may assume that f = fv . Every compact subset of A is contained in one of the
Q
form S = Sv with Sv ⊆ Kv compact and Sv = Ov for all but finitely many v, so it suffices
to consider convergence on S. At each nonarchimedean v, the set of κ such that fv (Sv + κ)
is not identically 0 is v-adically bounded, and the bound is |κ|v ≤ 1 for all but finitely many
v. In the function field case, this means that the sum becomes a finite sum. In the number
field case, the sum is effectively a sum over κ in a fractional ideal I of K; since I is a lattice
in v|∞ Kv = K ⊗ R, and since fv ∈ S (Kv ) tends to 0 rapidly, κ∈I f (x + κ) converges
Q P
P
absolutely and uniformly on S. Translating the function κ∈K f (x + κ) by an element of K
just permutes the terms. 

Theorem 5.7 (Poisson summation formula for K ⊂ A). If f ∈ S (A), then

X X
f (κ) = fb(κ).
κ∈K κ∈K
28
 March 3

Proof. The idea is to apply the Fourier inversion formula to the function
X
F (x) := f (x + κ)
κ∈K

on A/K. Let Fb : K → C be the Fourier transform of F . (Although the notation for fb and Fb
look the same, fb is a function on the Pontryagin dual A of A, while Fb is a function on the
Pontryagin dual K of A/K.) For κ ∈ K,
Z X
Fb(κ) = f (x + `) ψ(κx) dx
D `∈K
XZ
= f (x + `) ψ(κx) dx
`∈K D
XZ
= f (z) ψ(κ(z − `)) dz (substitute z = x + `)
`∈K D+`
Z
= f (z) ψ(κz) dz
A

= fb(κ).
The Fourier inversion formula then gives
X
F (x) = Fb(κ) ψ(κx),
κ∈K

which may be rewritten as

X X
f (x + κ) = fb(κ) ψ(κx).
κ∈K κ∈K

Set x = 0 to obtain
X X
f (κ) = fb(κ). 
κ∈K κ∈K

Remark 5.8. If we had not assumed Vol(D) = 1, we would have obtained

X 1 X
f (κ) = fb(κ)
κ∈K
Vol(D) κ∈K

for all f ∈ S (A), but then applying this to fb would yield

X 1 X
fb(κ) = f (κ).
κ∈K
Vol(D) κ∈K

Together, these prove that Vol(D)2 = 1, so Vol(D) = 1!

29
5.6. The Riemann–Roch theorem. Let K be the function field of a curve X over Fq .
Places v of K may be identified with closed points of X.
The divisor group of X is the free abelian group having as basis the set of all places v.
P
We write a divisor D as v dv v with dv ∈ Z and dv = 0 for all but finitely many v. Define
deg D := v dv [kv : Fq ] ∈ Z. Each f ∈ K × gives rise to a divisor v v(f ) v. Similarly, our
P P
P
chosen nonzero ω ∈ ΩK gives rise to a canonical divisor K = v κv v, where κv is the “order
of vanishing of ω at v”.
Q P
Define OA := Ov ⊂ A. Given D = dv v, define
Y
OA (D) := p−d
v
v
⊂ A.
v

(The product is a direct product, not an ideal product!) The Fq -vector space

L(D) := K ∩ OA (D)

consists of rational functions f on X having at worst a pole of order dv at v for each v. Let
`(D) := dimFq L(D).

Example 5.9. We claim that L(0) = Fq , and hence `(0) = 1. If t ∈ Fq , then t ∈ K ∩OA = L(0).
If t ∈ K − Fq , then the 1t -adic valuation on Fq (t) extends to a place v of K for which v(t) < 0,
so t ∈/ L(0).

Define the genus of X (or of K) by g := `(K) ∈ Z≥0 .

Theorem 5.10 (Riemann–Roch theorem). For any divisor D on X,

`(D) − `(K − D) = deg D + 1 − g.

Sketch of proof. First calculate some local and adelic Fourier transforms:
dv −κv /2
1p−d
b
v
v = q 1pdvv −κv (local calculation)
1OA (D) = q deg D−deg K/2 1OA (K−D)
b (product over v).

(An alternative way to get the right powers of q is to compare L2 -norms, by the Plancherel theorem.)
Applying the Poisson summation formula to f = 1OA (D) yields
X X
1 = q deg D−deg K/2 1
x∈L(D) x∈L(K−D)

deg K
`(D) = deg D − + `(K − D)
2
`(D) − `(K − D) = deg D + 1 − h

for some number h independent of D. Setting D = 0 shows that h = g. 

30
5.7. Norm of an idèle. If a = (av ) ∈ A× , multiplication-by-a is an isomorphism A → A,
under which dx corresponds to another Haar measure:
a
A −→ A
|a| dx ←− dx
for some “stretching factor” |a| ∈ R×
Q
>0 ; in fact, then |a| = v |av |v (note that |av |v = 1 for all
but finitely many v).
We can reprove the following result from last semester in a conceptual way:

Product formula. If a ∈ K × , then |a| = 1.

Sneaky proof. Multiplication-by-a is an isomorphism A/K → A/K, but it scales volume by

|a|, so
Vol(A/K) = |a| Vol(A/K).
Since Vol(A/K) is finite and positive, we get |a| = 1! 
Let |A× | be the group {|x| : x ∈ A× }. For t ∈ |A× |, let A× ×
t := {x ∈ A : |x| = t}.

5.8. Idèle-class characters.

“We can do nothing really significant with the idele group until we imbed the multiplicative
group K × in it”. — John Tate, in his thesis.
Not all characters of A× are of interest to us.

Definition 5.11. An idèle-class character or Hecke character or Größencharakter is a continuous

homomorphism χ : A× → C× such that χ|K × = 1.

Idèle-class characters can be identified with characters of the idèle class group A× /K × .

Example 5.12. If s ∈ C, then | |s is an idèle-class character.

Example 5.13. Take K = Q. Let χ be a Dirichlet character, i.e., a homomorphism (Z/N Z)× →
C× for some N ≥ 1. Define the ring Zb := lim Z/N Z ' Q Zp , in which the inverse limit is
←− p
over positive integers N ordered by divisibility (the hat on the Z signifies completion, not
Pontryagin dual). Then the composition
× χ
A× = Q× × R× b× b×
>0 × Z  Z  (Z/N Z) → C
×

is an idèle-class character (which by abuse of terminology might again be called a Dirichlet

character). One can show that every finite-order idèle-class character arises in this way.
Over Q, Dirichlet characters look simpler than the corresponding idèle-class characters, but if one is working
over higher number fields K, then the class group and unit group prevent such an easy decomposition of A×
into three factors. For such K, it is the idèle-theoretic definition of character that is the useful definition, the
one that lets us avoid worrying about class groups and unit groups.
31
 Example 5.14. Let K s be a separable closure of K. Let K ab ⊂ K s be the maximal abelian
extension of K. Let θ : A× /K × → Gal(K ab /K) be the global Artin homomorphism of class
field theory. Any continuous character ρ : Gal(K s /K) → C× factors through Gal(K ab /K),
and then the composition along the bottom row of
Gal(K s /K) ρ

' &
θ
A× / / A× /K × / Gal(K ab /K) / C×
is an idèle-class character.

Warning 5.15. Because θ is not quite an isomorphism, it is not quite true that all idèle-class
characters arise from continuous characters Gal(K s /K) → C× . In fact, since Gal(K s /K) is
profinite and since a sufficiently small open neighborhood of 1 in C× contains no nontrivial
subgroups, any continuous character Gal(K s /K) → C× has finite image, but idèle-class
characters can have infinite image. One can fix this by replacing Gal(K s /K) by a closely
related group called a Weil group WK .

Using the compactness of A× ×

1 /K and the exact sequence
||
1 −→ A× × × × ×
1 /K −→ A /K −→ |A | −→ 1

analogous to (4), we see that every idèle-class character χ is η | |s for some unitary η and some
s ∈ C. Define its exponent σ and twisted dual χ∨ as in the local case, and let χv := χ|Kv× .
Let X be the group of all idèle-class characters χ. Then X has the structure of a Riemann
surface, just as X(F × ) did for a local field F .

5.9. Multiplicative Haar measure.

First attempt: Define d× xv := |xdxv v|v , and define d× x := v d× xv so that Q Uv d× x =
Q R
×
Uv of A× . Problem: For almost all v, we have
QR Q
Uv
d x v for each basic open subset
Uv = Ov× and Ov dxv = 1, so
R
Z Z Z
× ×
d xv = d xv = dxv = 1 − qv−1 ,
Uv Ov× Ov×

but finite v (1 − qv−1 ) diverges to 0 (for example, when K = Q, this is p (1 − p−1 ), whose
Q Q

inverse is the divergent Euler product for 1 + 12 + 31 + · · · ).

Second attempt (the good one): Define
dxv


 , if v is archimedean
×
 |xv |v
d xv :=
dx
(1 − qv−1 )−1 v , if v is nonarchimedean.


|xv |v
Then Ov× d× xv = 1 for all but finitely many v, so we can define d× x := d× xv .
R Q
32
 March 5

If K is a number field, then |A× | = R× >0 , which we give the Haar measure dt/t. If K is
a function field, then there exists q > 1 such that |A× | = q Z , which we give the counting
measure multiplied by log q (the “logarithmic spacing” between powers of q) and denote dt/t
again!
Equip A× ∗
1 with the Haar measure d x compatible with the other two measures d x and
×

dt/t in the short exact sequence

||
1 −→ A× × ×
1 −→ A −→ |A | −→ 1.

For t ∈ |A× |, multiplication by any idèle a of norm t is a homeomorphism A× × ∗

1 → At so d x
gives rise to a measure on A× t (independent of the choice of a).
Finally, the measure on A× × ×
t induces a measure on the orbit space At /K ; any of these
measures will be denoted d∗ x. Let V be the volume of the compact group A× ×
1 /K .

5.10. Global zeta integrals. Given f ∈ S (A) and an idèle-class character χ, define the
global zeta integral Z
Z(f, χ) := f (x) χ(x) d× x; (6)
A×
this is a generalization of the completed Riemann zeta function ξ(s). We are now ready for
the main theorem in the global setting:

Theorem 5.16 (Meromorphic continuation and functional equation of global zeta integrals).
Fix f ∈ S (A).
(a) The integral Z(f, χ) converges for idèle-class characters χ of exponent σ > 1.
(b) The function Z(f, χ) of χ extends to a meromorphic function on X . In fact, it is
holomorphic except for
• a simple pole at | |0 with residue −V f (0), and
• a simple pole at | |1 with residue V fb(0).
(c) We have Z(f, χ) = Z(fb, χ∨ ) as meromorphic functions of χ ∈ X .
Q
Proof of convergence for σ > 1. We may assume that f = fv . We may replace f and χ by
their absolute values, so χ becomes | |σ . Now
Y
Z(f, χ) = Z(fv , | |σv ).
v

For all but finitely many v, the function fv is 1Ov and Z(fv , | |σv ) = (1 − qv−σ )−1 , so we need
to prove convergence of
Y
(1 − qv−σ )−1 .
nonarchimedean v
33
If K is a number field of degree d, this is the Dedekind zeta function ζK evaluated at σ,
which converges for σ > 1 — the argument given last semester was that there are at most d
places of K above each prime p of Q, and each has norm qv at least p, so
!d
Y Y X
(1 − qv−σ )−1 ≤ (1 − p−σ )−d = n−σ < ∞.
nonarchimedean v prime p n≥1

If K is a function field, we can similarly reduce to the case of Fq (t), in which every v except
one corresponds to a maximal ideal of Fq [t], so if Nn denotes the number of monic irreducible
polynomials in Fq [t] of degree n, we need convergence of
Y Y
(1 − qv−σ ) = (1 − q −nσ )−Nn ,
v6=∞ n≥1

or equivalently, of X X
Nn q −nσ ≤ q n q −nσ < ∞
n≥1 n≥1
if σ > 1 (geometric series). 
Slice A× according to norm. Then K × is a discrete co-compact subgroup in the slice A×
1.
For σ > 1, performing the integration in (6) slice-by-slice yields
Z
dt
Z(f, χ) = Zt (f, χ) , (7)
|A|× t
where the integral over the norm t slice, defined for χ of arbitrary exponent, is
Z
Zt (f, χ) := f (x) χ(x) d∗ x.
A×
t

Remark 5.17. Equation (7) in the case χ = | |s is the analogue of the equation
Z ∞ 
Θ(t) − 1 dt
ξ(s) = ts/2
0 2 t
that appeared in Riemann’s proof. In particular, the next lemma shows that Zt (f, χ) has its own functional
equation,
 analogous
 to the functional equation of Θ(t), or more closely, analogous to the functional equation
Θ(t)−1
of 2 ts/2 .

Lemma 5.18. For any idèle-class character χ,

Z Z
∗ ∨
Zt (f, χ) + f (0) χ(x) d x = Z1/t (f , χ ) + f (0)
b b χ∨ (x) d∗ x.
A×
t /K
× A×
1/t
/K ×

Proof. By tiling A× ×
t into translates of a fundamental domain for the action of K , we get
Z X
Zt (f, χ) = f (ax) χ(ax) d∗ x
A×
t /K
×
a∈K ×
Z !
X
= f (ax) χ(x) d∗ x (since χ is 1 on K × ).
A×
t /K
×
a∈K ×
34
It would be better to have a sum over K instead of K × , in order to apply the Poisson
summation formula/Riemann–Roch theorem, so we add an a = 0 term to both sides:

Z Z !
X
Zt (f, χ) + f (0) χ(x) d∗ x = f (ax) χ(x) d∗ x
A×
t /K
× A×
t /K
×
a∈K
Z !
1 Xb
= f (a/x) χ(x) d∗ x
A×
t /K
× |x| a∈K
(by the Riemann–Roch theorem)
Z !
X
= |y| fb(ay) χ(y −1 ) d∗ y (set x = y −1 )
A×
1/t
/K × a∈K
Z !
X
= fb(ay) χ∨ (y) d∗ y
A×
1/t
/K × a∈K
Z
∨
= Z1/t (fb, χ ) + fb(0) χ∨ (y) d∗ y,
A×
1/t
/K ×

by the same argument in reverse. 

Lemma 5.19. For t ∈ |A× | and an idèle-class character χ,


Z V ts if χ = | |s for some s ∈ C,
χ(x) d∗ x =
A×t /K
× 0 otherwise.

Proof. If χ = | |s for some s ∈ C, then the integrand is the constant ts , and Vol(A× ×
t /K ) =
Vol(A× × × ×
1 /K ) = V . Otherwise χ does not factor through A /A1 , so χ is not constant on the
compact group A× × × ×
1 /K , so its integral on any coset of A1 /K is 0. 

Proof of Theorem 5.16. Divide the range of integration in (7) in half:

Z 1 Z ∞
dt dt
Z(f, χ) = Zt (f, χ) + Zt (f, χ) ;
t t
|0 {z } |1 {z }
J(f,χ) I(f,χ)

in the function field case, these integrals are really (log q times) sums over q Z for some q > 1,
and by convention we assign half of the q 0 term to each of I(f, χ) and J(f, χ).
In the region σ > 1, both integrals converge to a holomorphic function. In fact, I(f, χ)
converges to a holomorphic function everywhere and its convergence only gets better as σ
decreases, since the χ(x) in Zt (f, χ) has absolute value tσ , which for t ≥ 1 decreases as σ
decreases.
35
 We handle the problematic half J(f, χ) by using the functional equation for Zt (f, χ) and
flipping the integral with t 7→ 1/t, as in Riemann’s proof: for σ > 1,
Z 1
dt
J(f, χ) = Zt (f, χ)
0 t
Z 1 Z 1  1−s !
dt 1 dt
= Z1/t (fb, χ∨ ) + V fb(0) − V f (0) ts
0 t 0 t t
| {z }
if χ = | |s

(by Lemmas 5.18 and 5.19)

Z∞ Z 1
∨ du  dt
= Zu (fb, χ ) + V fb(0) ts−1 − V f (0) ts
1 u t
|0 {z }
if χ = | |s
Z 1 Z 1
dt dt
= I(fb, χ∨ ) + V fb(0) s−1
t − V f (0) ts .
0 t 0 t
| {z }
if χ = | |s

Adding I(f, χ) to both sides yields

Z 1 Z 1
dt dt
Z(f, χ) = I(f, χ) + I(fb, χ∨ ) + V fb(0) ts−1
− V f (0) ts .
0 t 0 t
| {z }
if χ = | |s

In the region where χ is not of the form | |s , the formula on the right hand side provides
a holomorphic extension of Z(f, χ) that is symmetric with respect to (f, χ) ↔ (fb, χ∨ ).
ˆˆ
(Note that Z(f, χ) = Z(f (−x), χ) = Z(f, χ) by the substitution x 7→ −x, since χ(−x) =
χ(−1)χ(x) = χ(x).)
Now consider the region where χ = | |s . In the number field case, the extra terms contribute
V fb(0) V f (0)
− ,
s−1 s
which is meromorphic on C and symmetric with respect to (f, s) ↔ (fb, 1 − s), with residues
−V f (0) at s = 0 and V fb(0) at s = 1, as claimed. In the function field case, the extra terms
contribute
∞
! ∞
!
1 X 1 X
V fb(0) log q − + (q −n )s−1 − V f (0) log q − + (q −n )s
2 n=0 2 n=0
V log q b 1 + q 1−s 1 + qs
 
= f (0) + f (0) ,
2 1 − q 1−s 1 − qs
which again is meromorphic and symmetric, again with the claimed residues when χ is | |0 or
| |1 , that is, when q s is q 0 or q 1 . 
36
 March 10

5.11. Hecke L-functions. By comparing the local and global functional equations, we will
get functional equations for global L-functions.

Definition 5.20. For each idèle-class character χ, define

Y
(χ) := (χv )
v
Y
L(χ) := L(χv )
v

(when the products converge).

Remark 5.21. In defining (χv ) we are implicitly using the standard ψ and the Tamagawa
measure dx. If we change ψ to some other additive character ψa trivial on K (so a ∈ K × ),
then each (χv ) is multiplied by χv (a) |av |−1
Q
v , and the product is unchanged since v χv (a) =
−1 −1
Q
χ(a) = 1 and v |av |v = |a| = 1. Thus it is reasonable to omit ψ and the canonical dx
from the notation for global -factors.

Theorem 5.22.
(a) The product (χ) converges to a nonvanishing holomorphic function on all of X .
(b) The product L(χ) converges for σ > 1, and extends to a meromorphic function on all of
X , holomorphic everywhere outside | |0 and | |1 .
(c) We have L(χ) = (χ)L(χ∨ ) as meromorphic functions on X .

Proof. We restrict attention to one component X0 of X . The local calculations for Theo-
rem 4.18 showed that
(i) v (χv ) is a nonvanishing holomorphic function on X0 , equal to 1 for all but finitely
many v;
and that we can find fv ∈ S (Kv ) with fv = 1Ov for all but finitely many v such that
(ii) Z(fv , χv )/L(χv ) is a nonvanishing holomorphic function on X0 , equal to 1 for all but
finitely many v; and
(iii) the local functional equation
Z(fbv , χ∨v ) Z(fv , χv )
∨
= (χv )
L(χv ) L(χv )
of Theorem 4.18 holds for all v.
fv ∈ S (A) shows that
Q
Taking the product over v and setting f :=
(i) (χ) is a nonvanishing holomorphic function on X0 ;
(ii) v (Z(fv , χv )/L(χv )) is a nonvanishing holomorphic function on X0 , so the properties
Q

of L(χ) in (b) follow from those of Z(f, χ) in Theorem 5.16; and

37
(iii)
Z(fb, χ∨ ) Z(f, χ)
= (χ) ,
L(χ∨ ) L(χ)
so that dividing by the global functional equation

Z(fb, χ∨ ) = Z(f, χ)

of Theorem 5.16 gives

1 1
∨
= (χ)
L(χ ) L(χ)
L(χ) = (χ) L(χ∨ ). 

To get meromorphic functions of s ∈ C, fix an idèle-class character χ, and define

(s, χ) := (χ| |s )
L(s, χ) := L(χ| |s ).

Functions of the form L(s, χ) are called Hecke L-functions; these include (completed) Dedekind
zeta functions and Dirichlet L-functions as special cases. In these terms, the functional
equation is
L(s, χ) = (s, χ) L(1 − s, χ−1 ).
For homework, you will specialize this to obtain the functional equation of the Dedekind zeta
function.

Remark 5.23. In Warning 5.15, we alluded to the fact that idèle-class characters are in
bijection with continuous characters WK → GL1 (C) = C× of the Weil group WK . One can
generalize the whole theory by considering continuous representations WK → GLn (C) for
larger n. For an introduction to this generalization, see [Tat79].

References
[Dei05] Anton Deitmar, A first course in harmonic analysis, 2nd ed., Universitext, Springer-Verlag, New
York, 2005. MR2121678 (2006a:42001) ↑1, 1.2, 2
[DE09] Anton Deitmar and Siegfried Echterhoff, Principles of harmonic analysis, Universitext, Springer,
New York, 2009. MR2457798 (2010g:43001) ↑2, 3.2
[Del73] P. Deligne, Les constantes des équations fonctionnelles des fonctions L, Modular functions of one
variable, II (Proc. Internat. Summer School, Univ. Antwerp, Antwerp, 1972), Lecture Notes in Math.,
vol. 349, Springer, Berlin, 1973, pp. 501–597 (French). MR0349635 (50 #2128) ↑4.19
[RV99] Dinakar Ramakrishnan and Robert J. Valenza, Fourier analysis on number fields, Graduate Texts in
Mathematics, vol. 186, Springer-Verlag, New York, 1999. MR1680912 (2000d:11002) ↑4, 4.1
[Tat67] J. T. Tate, Fourier analysis in number fields, and Hecke’s zeta-functions, Algebraic Number The-
ory (Proc. Instructional Conf., Brighton, 1965), Thompson, Washington, D.C., 1967, pp. 305–347.
MR0217026 ↑1
38
[Tat79] J. Tate, Number theoretic background, Automorphic forms, representations and L-functions (Proc.
Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Proc. Sympos. Pure Math., XXXIII,
Amer. Math. Soc., Providence, RI, 1979, pp. 3–26. MR546607 (80m:12009) ↑5.23

39