Академический Документы
Профессиональный Документы
Культура Документы
YICHAO TIAN
The aim of this short note is to explain Tates thesis [Ta50] on the harmonic analysis on
Ad`eles and Id`eles, the functional equations of Dedekind Zeta functions and Hecke L-series.
For general reference on ad`eles and id`eles, we refer the reader to [We74].
1. Local Theory
1.1. Let k be a local eld of characteristic 0, i.e. R, C or a nite extension of Q
p
. If k is
p-adic, we denote by O k the ring of integers in k, p O the maximal ideal, and p
a uniformizer of O. If a is a fractional ideal of O, we denote by Na Q the norm of a. So
if a O is an ideal, we have Na = [O/a[. Let [ [ : k R
0
be the normalized absolute
value on k, i.e. for x k, we have
[x[ =
_
_
[x[
R
if k = R;
[x[
2
C
if k = C;
N(p)
ord(x)
if k is p-adic and x = u
ord(x)
with u O
.
We denote by k
+
the additive group of k. Consider the unitary character : k
+
C
dened as follows:
(1.1.1) (x) =
_
_
e
2ix
if k = R;
e
2i(x+ x)
if k = C;
e
2i(Tr
k/Qp
(x))
if k is p-adic,
where () means the decimal part of a p-adic number. For any k, we note by
the
additive character x (x) of k
+
. Note that if k is non-archimedean, (x) = 1 if and
only if x d
1
, where d is the dierent of k over Q
p
, i.e.
x d
1
Tr
k/Qp
(xy) Z
p
y O.
Proposition 1.2. The map :
k
+
, where
k
+
denotes the group of unitary characters of k
+
.
Proof. If k = R or C, this is well known in classical Fourier analysis. We assume here k is
non-archimedean.
(1) Its clear that is a homomorphism of groups. We show rst that is continuous
(at 0). If p
m
, then
is trivial on d
1
p
m
. Since the subsets
U
m
=
k
+
[ is trivial on d
1
p
m
contains no subgroup of S
1
. Hence we have (x
n
z) = 1 for all z p
m
, so
x
n
d
1
p
m
.
(3) The image of is dense in k. Let H be the image of , and
H
k be its closure.
Then we have
= x
k k [ (x) = 1, H
= x k [ (x) = 1, k = 0
Hence, we have
H =
k.
(4) The proof of the Proposition will be complete by the Lemma 1.3 below.
Lemma 1.3. Let G be a locally compact topological group, H G be a locally compact
subgroup. Then H is closed in G.
Proof. Let h
n
be a sequence in H that converges to g G. We need to prove that
g H. Let (U
r
)
r0
be a fundamental system of compact neighborhoods of 0. We have
r0
U
r
= 0. Then for any r, there exists an integer N
r
> 0 such that h
n
g +U
r
for all
n N
r
. Up to modifying U
r
, we may assume h
n
h
m
HU
r1
for any n, m N
r
. Note
that H U
r1
is also compact by the local compactness of H. Up to replacing U
r
r0
by a subsequence, we may choose m
r
for each integer r such that
h
m
r+1
+ U
r
H h
mr
+ U
r1
H.
By compactness, the intersection
r1
(h
mr
+ U
r1
H)
must contain an element h H. Its easy to see that h = g, since
r0
U
r
= 0.
1
2
.
Let L
1
(k, C) be the space of complex valued absolutely integrable functions on k. For
f L
1
(k, C), we dene the Fourier transform of f to be
(1.4.1)
f() =
_
k
f(x)(x)dx.
Let o(k) be the space of Schwartz functions on k, i.e.
o(R) = f C
(R) [ n, m N, [x
n
d
m
f
dx
m
[ is bounded;
NOTES ON TATES THESIS 3
we have a similar denition for k = C; and if k is p-adic, o(k) consists of locally constant
and compactly supported functions on k. In all these cases, the space o(k) is dense in
L
1
(k, C).
Proposition 1.5. The map f
f preserves o(k), and we have
, is
(1.6.1)
f(x) = (ax)(Nd)
1
2
(Np)
1
d
1
p
.
In particular, we have
f o(k).
Proof. By denition, we have
f(x) =
_
a+p
(xy)dy = (ax)
_
p
(xy)dy.
The lemma follows immediately from
_
p
(xy)dy =
_
(Nd)
1
2
(Np)
if x d
1
p
0 otherwise.
f(x) =
_
k
f(y)(xy)dy = (Nd)
1
2
(Np)
_
d
1
p
((x + a)y)dy
= (Nd)
1
2
(Np)
(Nd)
1
2
(Np)
ord(d)+
1
a+p
= 1
a+p
.
In the third equality above, we have used (1.6.1) with replaced by ord
(d) and x
replaced by x + a. Now its clear that
f(x) = f(x).
1.7. Now consider the multiplicative group k
, and put
U = x k
[ [x[ = 1.
So U = 1 if k = R, U = S
1
is the group of unit circle if k = C, and U = O
if k is
p-adic. We have
k
/U =
_
R
+
if k = R, C;
Z if k is p-adic.
4 YICHAO TIAN
Recall that a quasi-character of k
is a continuous homomorphism : k
. We say
is a (unitary) character if [(x)[ = 1 for all x k
, and is unramied if [
U
is trivial.
So is unramied if and only if there is s C such that (x) = [x[
s
. Note that such an
s is determined by if k = R or C, and determined up to 2i/log(Np) if k is p-adic.
Lemma 1.8. For any quasi-character of k
such that =
0
[ [
s
.
Proof. For any x k
+
if k = R
or C, and
Z
if k is non-archimedean. We dene
0
as
0
(x) = ([
U
)( x). One checks
easily that the quasi-character /
0
is unramied.
Let be a quasi-character of k
= p
to be d
x = (k)dx/[x[, where
(1.9.1) (k) =
_
1 if k = R, C;
Np
Np1
if k is non-archimedean.
If k is non-archimedean, the factor (k) is justied by the fact that
_
U
dx = (Nd)
1
2
.
Denition 1.10. For f o(k), we put
(f, ) =
_
k
f(x)(x) d
x,
which converges for any quasi-character with () > 0. We call (f, ) the local zeta
function associated with f (in quasi-characters).
Proposition 1.11. For any f, g o(k), we have
(f, )( g, ) = (
f, )(g, ),
where
f, g are Fourier transforms of f and g, and = [ [
1
for any quasi-character
with 0 < () < 1.
Proof.
(f, )( g, ) =
_
k
__
k
f(x) g(xy)[x[d
x
_
(y
1
)[y[d
y
= (k)
_
k
__
k
_
k
f(x)g(z)(xyz)dzdx
_
(y
1
)[y[d
y.
To nish the proof of the Proposition, it suces to note that the expression above is
symmetric for f and g.
NOTES ON TATES THESIS 5
We endow the set of quasi-characters with a structure of complex manifold such that
for any xed quasi-character the map s [ [
s
induces an isomorphism of complex
manifolds from C to a connected component of the set of quasi-characters.
Theorem 1.12. For any f o(k), the function (f, ) can be continued to a meromorphic
function on the space of all quasi-characters. Moreover, it satises the functional equation
(1.12.1) (f, ) = ()(
f, ),
where () is a meromorphic function of independent of f given as follows:
(1) If k = R, then (x) = [x[
s
or (x) = sgn(x)[x[
s
for some s C. We have
([ [
s
) = 2
1s
s
cos(
s
2
)(s), (sgn[ [
s
) = i2
1s
s
sin(
s
2
)(s).
(2) If k = C, then there exists n Z and s C such that =
n
[ [
s
where
n
is the
unitary character
n
(re
i
) = e
in
. We have
(
n
[ [
s
) = i
|n|
(2)
1s
(s +
|n|
2
)
(2)
s
(1 s +
|n|
2
)
.
(3) Assume k is p-adic. If is unramied, then
([ [
s
) = (Nd)
s
1
2
1 (Np)
s1
1 Np
s
.
If =
0
[ [
s
is ramied, where
0
is unitary with
0
() = 1 as in Lemma 1.8,
then one has
(
0
[ [
s
) = N(df
)
s
1
2
0
(
0
)
with
0
(
0
) = N(f
1
2
0
(x)(
x
ord(df)
)
where x runs over a set of representatives of O
/(1 + f
).
Proof. By Proposition 1.11, the function () =
(f,)
(
f, )
is independent of f. This proves
the functional equation (1.12.1). Note that (f, ) is well dened if () > 0, and (
f, )
is well dened if () < 1. Therefore, once we show that () is meromorphic as in the
statement, it will follow from the functional equation (1.12.1) that (f, ) can be continued
to a meromorphic function in . It remains to compute () by choosing special functions
f o(k).
(1) Assume k = R. If = [ [
s
, we choose f = e
x
2
. We have
(f, [ [
s
) =
_
R
e
x
2
[x[
s
d
x = 2
_
+
0
e
x
2
x
s1
dx =
s
2
(
s
2
).
On the other hand,
(1.12.2)
f(y) =
_
R
e
(x
2
+2ixy)
dx = e
y
2
_
R
e
(x+yi)
2
dx.
6 YICHAO TIAN
Using the well-known fact that
_
R
e
(x+yi)
2
dx =
_
R
e
x
2
dx = 1,
we get
f = f. Hence, we have (
f, [ [
1s
) =
1s
2
(
1s
2
) and
([ [
s
) =
s
2
(
s
2
)
1s
2
(
1s
2
)
=
s
(
s
2
)(
s+1
2
)
(
1s
2
)(
1+s
2
)
Now the formula for ([ [
s
) follows from the properties of Gamma functions
(
s
2
)(
s + 1
2
) = 2
1s
(s), (
1 s
2
)(
1 + s
2
) =
sin(
(1+s)
2
)
.
If = sgn[ [
s
, we take f = xe
x
2
. A similar computation shows that
(f, sgn[ [
s
) =
s+1
2
(
s + 1
2
).
Taking derivatives with respect to y in (1.12.2), we get
f = if. So we have
(
f, sgn[ [
1s
) = i
s
2
1
(1
s
2
).
Therefore, we get
(sgn[ [
s
) =
s+1
2
(
s+1
2
)
i
s
2
1
(1
s
2
)
= i
s
(
s+1
2
)(
s
2
)
(1
s
2
)(
s
2
)
= i2
1s
s
sin(
s
2
)(s).
(2) Assume k = C. If = [ [
s
, we take f(z) = e
(z z)
. The local zeta function
associated with f is
(f, [ [
s
) =
_
C
e
z z
(z z)
s
d
z
=
_
2
=0
_
+
r=0
e
r
2
r
2s
2rdrd
r
2
= 4
_
+
0
e
r
2
r
2s1
dr
= 4
_
+
0
t
s
1
2
e
t
dt
2
t
(set t = r
2
)
= 2
1s
(s).
NOTES ON TATES THESIS 7
The Fourier transform of f is
f(z) =
_
C
e
w w
e
2i(zw+ z w)
dw (1.12.3)
= 2
_
+
_
+
e
(u
2
+v
2
)
e
4i(uxvy)
dudv (put z = x + iy, w = u + iv)
= 2e
4(x
2
+y
2
)
_
+
e
(u+2ix)
2
du
_
+
e
(v2iy)
2
dv
= 2f(2z).
Therefore, one has (
f, [ [
1s
) = 2
2s1
(f, [ [
1s
) = 2
2s
s
(1 s), thus
([ [
s
) = (2)
12s
(s)
(1 s)
.
Let n 1 and =
n
[ [
s
. We put f
n
= z
n
e
(z z)
. We compute rst the local zeta
function of f
n
:
(f
n
,
n
[ [
s
) =
_
C
z
n
e
(z z)
n
(z)(z z)
s
d
z (1.12.4)
=
_
2
=0
_
+
r=0
e
r
2
r
2s+n
2rdrd
r
2
= 4
_
+
0
e
r
2
r
2s+n1
dr
= 4
_
+
0
e
t
t
s+
n1
2
dt
2
t
= 2
1(s+
n
2
)
(s +
n
2
).
To nd the Fourier transform of f
n
, we consider the equality (1.12.3)
2e
4(z z)
=
_
C
e
(w w)
e
2i(zw+ z w)
dw.
Regarding z and z as independent variables and applying
n
z
n
, we get
2(2i z)
n
e
4z z
=
_
C
w
n
e
(w w)
e
2i(zw+ z w)
dw,
that is,
f
n
(z) = 2
f
n
(2iz). A similar computation as (1.12.4) shows that
(
f
n
(z), ) = (2
f
n
(2iz),
n
[ [
1s
) = (i)
n
2
2s
s
n
2
(s +
n
2
).
Therefore, we get
(
n
[ [
s
) =
2
1(s+
n
2
)
(s +
n
2
)
(i)
n
2
2s
s
n
2
(s +
n
2
)
= i
n
(2)
12s
(s +
n
2
)
(
n
2
+ 1 s)
.
The formulae for (
n
[ [
s
) can be proved in the same way by choosing f =
f
n
.
8 YICHAO TIAN
(3) Assume k is p-adic. Consider rst the case = [ [
s
. We take f = 1
O
. In the proof
of Proposition 1.5, we have seen that
f = (Nd)
1
2
1
d
1. We have
(f, ) =
_
O{0}
[x[
s
d
x.
As O 0 =
+
n=0
n
O
, it follows that
(f, ) =
+
n=0
(Np)
ns
_
O
x = (Nd)
1
2
1
1 (Np)
s
Similarly, using d
1
0 =
+
n=ord(d)
n
O
, one obtains
(
f, ) = (Nd)
1
2
_
d
1
{0}
[x[
1s
d
x
= (Nd)
1
2
+
n=ord(d)
(Np)
n(s1)
_
O
x
= (Nd)
1
(Np)
ord(d)(1s)
+
n=0
Np
n(s1)
= (Nd)
s
1
1 Np
s1
.
The formula for ([ [
s
) follows immediately.
Now consider the case =
0
[ [
s
with
0
ramied, unitary and
0
() = 1. We take
f(x) = (
x
ord(df)
)1
O
.
The local zeta function of f is
(f, ) =
_
O{0}
(
x
ord(df)
)
0
(x)[x[
s
d
x
=
+
n=0
(Np)
ns
_
O
(
x
n
ord(df)
)
0
(x)d
x
We claim that
(1.12.5)
_
O
(
x
n
ord(df)
)
0
(x)d
x = 0 for n 1.
Consider rst the case n ord
(f
). We have
(
x
n
ord(df)
) = 1 as
x
n
ord(df)
d
1
.
If S is a set of representatives of O
/(1 + f
0
(x)d
x =
_
_
1+f
d
x
_
xS
0
(x) = 0.
NOTES ON TATES THESIS 9
Assume 0 n ord
(f
) 1. For any y 1 + p
n
f
, we have
(
xy
n
ord(df)
) = (
x
n
ord(df)
).
Therefore, if S
n
S denotes a subset of representatives of O
/(1 + p
n
f
), we get
_
O
(
x
n
ord(df)
)
0
(x)d
x =
_
_
1+f
d
x
_
xS
0
(x)(
x
n
ord(df)
)
=
_
_
1+f
d
x
_
xSn
0
(x)(
x
n
ord(df)
)
0
(y),
where y runs over a set of representatives of (1 + p
n
f
)/(1 + f
). Note that
0
(y) =
_
0 if 1 n ord
(f
),
1 if n = 0.
This proves the claim. It follows that
(1.12.6)
(f, ) =
_
_
1+f
d
x
_
xS
0
(x)(
x
ord(df)
) =
0
(1)(
_
1+f
d
x)(Nf
)
1
2
0
(
0
),
where we have used the denition of
0
in the last step. As in the proof of 1.5, the Fourier
transform of f is
f(x) =
_
O
(
y
ord(df)
)(xy)dy
=
_
O
(y(x +
1
ord(df)
))dy
= (Nd)
1
2
1
ord(df)
+d
1
.
We get the local zeta function of
f
(
f, ) = (Nd)
1
2
_
ord(df)
+d
1
[x[
1s
1
0
(x)d
x
= (Nd)
1
2
(Np)
ord(df)(1s)
_
ord(df)
(1+f)
1
0
(x)d
x
Since
1
0
(
ord(df)
(1 + y)) =
0
(1) for any y f
, we get
(
f, ) =
0
(1)(Nd)
1
2
N(df
)
1s
_
_
1+f
d
x
_
.
It thus follows that
(
0
[ [
s
) =
(f,
0
[ [
s
)
(
f,
1
0
[ [
1s
)
= N(df
)
s
1
2
0
(
0
).
10 YICHAO TIAN
Remark 1.13. The number
0
(
0
) in (3) is a generalization of (normalized) Gauss sum.
By the same method as the classical case, we can show that [
0
(
0
)[ = 1. In general, its
an interesting and dicult problem to nd the exact argument of
0
(
0
).
2. Global Theory
Let F be a number eld, O
F
be its ring of integers. Let be the set of all places of
F, and
f
(resp.
v
F
v
consisting of elements x = (x
v
)
v
with x
v
O
v
for
almost all v, and A
F,f
be the ring of nite ad`eles. We choose the Haar measure on A
F
as
dx =
v
dx
v
. It induces a quotient Haar measure on A
F
/F.
Lemma 2.1. Under the notation above, we have
_
A
F
/F
dx = 1.
Proof. By Chinese reminders theorem, we have A
F
= F +
v
f
O
v
v
F
v
. We get
thus an isomorphism
A
F
/F (
v
f
O
v
v
F
v
)/O
F
.
Hence we have
_
A
F
/F
dx =
v
f
_
Ov
dx
v
_
(
v
Fv)/O
F
v
dx
v
=
v
f
(Nd
v
)
1
2
[
F
[
1/2
,
where d
v
denotes the dierent of F
v
and
F
is the discriminant of F. If d denotes the
dierent of F/Q, then the lemma follows easily from the product formula:
[
F
[ = Nd =
v
f
Nd
v
.
For v , let
v
be the additive character of the local eld F
v
dened in (1.1.1).
Its easy to check that =
v
v
is trivial on additive group F, therefore it denes a
character of the quotient A
F
/F. We call it the basic character of A
F
/F (or A
F
). For any
A
F
, let
: A
F
C
is a character of A
F
/F if and only if F, i.e.
: A
F
C
v
=
[
Fv
to the v-th local
NOTES ON TATES THESIS 11
component denes a continuous character of F
v
. By Proposition 1.2, there exists
v
F
v
such that
v
=
v
(
v
). Since
vS
U
v
v/ S
O
v
of 0 such that its image under
. As B(1, 1/2)
contains no non-trivial subgroups of S
1
, we see that for any v / S, we have
v
O
v
. This
shows that = (
v
)
v
A
F
, and
v
Uv
< for n suciently large. By Proposition
1.2, for any nite subset S containing
, we can take (U
v
)
vS
suciently large and
U
v
= O
v
for v / S such that [
n
[
v
< for v S and
n
O
v
for v / S. This means that
n
0 in A
F
.
For the second part, let A
F
be the subgroup such that ()
A
F
consists of all
characters trivial on F. Its clear that F since is trivial on F. To show that = F,
we consider rst the case F = Q. Let . Since A
Q
= Q + (
1
2
,
1
2
]
p
Z
p
, we can
write = b + c, where b Q, c
(1) = () = (b + c) = (c) = e
2ic
.
Hence we have c
(
1
p
n
) = (
1
p
n
(b + c)) = e
2i(
cp
p
n
)
that c
p
p
n
Z
p
, i.e. we have c
p
= 0. This shows = b, and hence = Q. In the general
case, we note that the basic character of A
F
is the composition of that on A
Q
with the
trace map Tr
F/Q
: A
F
A
Q
. The following lemma will conclude the proof.
Lemma 2.3. Let x = (x
v
)
v
A
F
such that Tr
F/Q
(xy) Q A
Q
for all y F. Then
we have x F.
Proof. Let (e
i
)
1id
be a basis of F/Q, and (e
i
)
1id
be the dual basis with respect to
the perfect pairing F F Q given by (x, y) Tr
F/Q
(xy). For any place p of Q,
we have a canonical isomorphism of Q
p
-algebras
F Q
p
v|p
F
v
.
We put x
p
= (x
v
)
v|p
v|p
F
v
. Then we can write x
p
=
d
i=1
a
p,i
e
i
with a
p,i
Q
p
. As
Tr
F/Q
(xe
i
) Q A
Q
for any i, we deduce that a
p,i
Q and its independent of p. This
shows that x F.
Let o(A
F
) be the space of Schwartz functions on A
F
, i.e. the space of nite linear
combinations of functions on A
F
of the form f =
v
f
v
, where f
v
o(F
v
) and f
v
= 1
Ov
for almost all v. For any f o(A
F
), we dene the Fourier transform of f to be
(2.3.1)
f() =
_
A
F
f(x)(x)dx.
12 YICHAO TIAN
Proposition 2.4. (a) The Fourier transform f
f preserves the space o(A
F
), and
f(x) = f(x).
(b) If f =
v
f
v
with f
v
o(F
v
) and f
v
= 1
Ov
for almost all v. Then
f =
v
f
v
, where
f
v
is the local Fourier transform (1.4.1) of f
v
.
(c) For any f o(A), the innite sum
xF
[f(x)[ converges, and we have the Poisson
formulae
(2.4.1)
xF
f(x) =
f().
Proof. Statement (a) is a direct consequence of (b), which in turn follows from the local
computations in the proof of 1.5. Now we start to prove (c). We may assume f =
v
f
v
with f
v
o(F
v
) and f
v
= 1
Ov
for almost all v. Then there exists an open compact
subgroup U A
f
such that Supp(f) U
v
F
v
. Put O
U
= F (U
v
F
v
).
This is a lattice in F. Each individual term in the summation
xF
f(x) is non-zero only
if x O
U
. Write f = f
, where f
=
v
f
f
v
and f
=
v
f
v
. Then there exists
a constant C > 0 such that [f
xF
[f(x)[ =
xO
U
[f(x)[ < C
xO
U
[f
(x)[.
By classical analysis, the sum on the right hand side is convergent. This proves the rst
part of (c). It remains to show Poissons summation formula (2.4.1). Consider the function
g(x) =
yF
f(x + y), which converges for any x A
F
by the rst part of (c). As g(x)
is invariant under translation of F, we regard g(x) as a function on A
F
/F. Its Fourier
transform of g(x) is
g() =
_
A
F
/F
g(x)(x)dx (for F)
=
_
A
F
f(x)(x)dx =
f().
By the Fourier inverse formulae (a), we have
g(x) =
F
g()(x).
The formulae (2.4.1) follows by setting x = 0.
2.5. Let I
F
= A
F
be the multiplicative group of id`eles of F, i.e. the subgroup of
v
F
v
consisting of elements x = (x
v
)
v
with x
v
O
v
for almost all v, and I
1
F
be the subgroup of
I
F
of id`eles with norm 1. The diagonal embedding F
I
1
F
identies F
with a discrete
subgroup of I
1
for the induced restricted product topology on I
1
F
. A fundamental theorem
in the theory of id`eles says that the quotient I
1
F
/F
x =
v
d
x
v
on I
F
, where d
x
v
is the local Haar measure
on F
v
considered in 1.9. We use the same notation for the induced Haar measures on I
1
F
and I
1
F
/F
.
NOTES ON TATES THESIS 13
Proposition 2.6. Under the notation above, we have
Vol(I
1
F
/F
) =
_
I
1
F
/F
x =
2
r
1
(2)
r
2
hR
[
F
[
1/2
w
,
where r
1
(resp. r
2
) is the number of real places (resp. complex places) of F, h is the class
number of F,
F
is the discriminant, R is the regulator, and w denotes the number of
roots of unity in F.
Proof. Note rst that Vol(I
1
F
/F
) is nite, since I
1
F
/F
v
f
p
ordv(xv)
v
be the fractional ideal associated
with x. Then Div induces a short exact sequence
0 (
v
f
O
v
(R
)
r
1
(C
)
r
2
) F
I
F
Cl
F
0,
where Cl
F
denotes the class group of F. Let be the subgroup of (R
)
r
1
(C
)
r
2
with
product of absolute values
r
1
i=1
[x
i
[
r
2
i=1
[z
i
[
C
= 1. The the exact sequence above
induces a similar exact sequence
0 (
v
f
O
v
) F
I
1
F
Cl
F
0.
Therefore, one gets
_
I
1
F
/F
= h
_
(
v
O
v
)/(
v
O
v
)F
x.
Let U
F
denote the group of units of F. We have (
v
O
v
) F
= U
F
, and hence
_
(
v
O
v
)/F
v
O
v
)
= (
v
f
_
O
v
dx
v
)
_
/U
F
d
x =
v
f
Nd
1
2
v
_
/U
F
d
x.
In view of the product formula
v
f
Nd
1
2
= [
F
[
1
2
, to complete the proof, it suces
to prove that
(2.6.1)
_
/U
F
d
x =
2
r
1
(2)
r
2
R
w
.
Consider the map
Log : (R
)
r
1
(C
)
r
2
R
r
1
+r
2
((x
i
)
1ir
1
, (z
j
)
1jr
2
) ((log [x
i
[)
1ir
1
, (log [z
j
[
2
)
1jr
2
).
Let S
1
be the unit circle subgroup of C
r
1
i=1
x
i
+
r
2
j=1
y
j
= 0. Then the map Log induces a short exact sequence
of abelian groups
0 1
r
1
(S
1
)
r
2
Log
V 0.
14 YICHAO TIAN
If
F
denotes the group of roots of unity in F, we have (1
r
1
(S
1
)
r
2
) U
F
=
F
.
Therefore, one obtains
_
/U
F
d
x =
_
_
{1}
r
1(S
1
)
r
2/
F
d
x
_
_
_
V/Log(U
F
)
d
x
_
.
By the denition of the Haar measure on I
F
, the induced measure on (R
)
r
1
(C
)
r
2
is determined as follows. On each copy of R
x =
dx
|x|
,
where dx is the usual Lebesgue measure on R; on each copy of C
z = 2
dx dy
[z[
2
=
d(r
2
)
r
2
d,
where z = x + iy = re
i
. Therefore, the Haar measure on I
F
induces the usual Lebesgue
measure on Log() = V , and the measure
r
2
j=1
d
j
on (S
1
)
r
2
.
1
It follows that
_
{1}
r
1(S
1
)
r
2/
F
d
x =
2
r
1
(2)
r
2
w
.
By the denition of the regulator, we have R =
_
V/Log(U
F
)
dx. Now the formula (2.6.1)
follows immediately. This nished the proof.
2.7. A Hecke character (or Gr ossencharacter) of F is a continuous homomorphism :
I
F
/F
= I
1
F
/F
+
of the norm map [ [ : I
F
/F
+
.
For every Hecke character , we put
0
= [
I
1
F
/F
and denote still by
0
its extension
to I
F
/F
by requiring
0
is trivial on the chosen complement R
+
of I
1
F
/F
. Note that
0
is necessarily unitary since I
1
F
/F
is compact and /
0
is unramied, i.e. =
0
[ [
s
with s C. We put () = 1(s), which is independent of the choice of the splitting. For
v , we put
v
= [
F
v
. The local component
v
is unramied for almost all v.
Denition 2.8. Let f o(A
F
) and be a Hecke character of F. We dene the zeta
function of f at to be
(f, ) =
_
I
F
f(x)(x)d
x.
Lemma 2.9. Let f o(A
F
) and X. Then the zeta function (f, ) converges
absolutely for () > 1.
1
Note that the niteness of Vol(I
1
/F
) implies that
V/Log(U
F
)
d
S
1
unitary. By denition, we have an Euler product
(f, ) =
v
(f
v
,
v
),
where (f
v
,
v
) is the local zeta function dened in 1.10. We have seen in the proof of
Theorem 1.12(3) that
(1
Ov
, [ [
s
v
) = (Nd
v
)
1
2
1
1 Np
s
v
.
Thus there exists a nite subset S of places such that
[(f,
0
[ [
s
)[
vS
[(f
v
,
0,v
[ [
s
v
)[
v/ S
1
1 Np
v
.
Since each (f
v
,
0,v
[ [
s
v
) converges for 1(s) > 0, we are reduced to showing that the
product
v/ S
1
1Np
v
converges absolutely for > 1. If F = Q, this is a well-known
theorem of Euler. In the general case, we have
v/ S
1
1 Np
v|p
1
1 Np
v
(
p
1
1 p
)
[F:Q]
.
) at = 1 and
f(0)Vol(I
1
F
/F
) at = [ [, where
Vol(I
1
F
/F
) =
2
r
1
(2)
r
2
hR
w
_
[
F
[
.
Proof. Let I
1
F
(resp. I
1
F
) be the subset of I
F
with norm 1 (resp. 1). Since I
1
F
=
I
1
F
I
1
F
has Haar measure 0 in I
F
, we have
(f, ) =
_
I
F
f(x)(x)d
x =
_
I
1
F
f(x)(x)d
x +
_
I
1
F
f(x)(x)d
x.
Note that f is well-behaved when [x[ , the rst integral
_
I
1
F
f(x)(x)d
x converges
absolutely for all X, thus denes a holomorphic function on the whole complex
manifold X. For the second integral, we have
_
I
1
F
f(x)(x)d
x =
_
I
1
F
/F
f(x))(x)d
x
16 YICHAO TIAN
by the triviality of on F
x
). It follows from the Poisson formulae (2.4.1) that
f(x) =
1
[x[
f(
x
) +
1
[x[
f(0) f(0).
Therefore, we get
_
I
1
F
/F
f(x)(x)d
x =
_
I
1
F
/F
1
[x[
f(
x
))(x)d
x+
_
I
1
F
/F
(
1
[x[
f(0)f(0))(x)d
x.
Making the change of variable y =
1
x
, the rst term on the right hand side above becomes
_
I
1
F
/F
f(
x
))(x)d
=
_
I
1
F
/F
f(y)) (y)d
y
=
_
I
1
F
f(y) (y)d
y.
We choose a splitting I
F
/F
= I
1
F
/F
+
as in 2.7, and write that =
0
[ [
s
, with
0
: I
1
F
/F
f(0)
[x[
f(0))(x)d
x = (
_
I
1
F
/F
0
(x)d
x)(
_
1
t=0
(
f(0)
t
f(0))t
s1
dt.
We have
_
I
1
F
/F
0
(x)d
x = Vol(I
1
F
/F
0
,1
=
_
0 if
0
is non-trivial;
Vol(I
1
F
/F
) if
0
is trivial.
For the second term, we have
_
1
t=0
(
f(0)
t
f(0))t
s1
dt =
f(0)
s 1
f(0)
s
.
Combining all the computations above, we get
(f, ) =
_
I
1
F
f(x)(x)d
x +
_
I
1
F
f(x) (x)d
x + Vol(I
1
F
/F
)(
f(0)
s 1
f(0)
s
)
0
,1
.
Now its clear that the right hand side of the equation above is invariant with f replaced
by
f and replaced by . Thus (2.10.1) follows immediately. The moreover part follows
from the fact that the rst two integrals above dene holomorphic functions on X.
2.11. We indicate how to apply Tates general theory to recover the classical results on
the Dedekind Zeta function of a number eld. Recall that Dedekinds zeta function is
dened to be
F
(s) =
v
1
1 Np
s
v
=
aO
F
1
(Na)
s
,
which converges absolutely for 1(s) > 1. In the classical theory of Dedekinds zeta func-
tion, we have
NOTES ON TATES THESIS 17
Theorem 2.12. Let F be a number eld with r
1
real places and r
2
complex places. We
put
Z
F
(s) = G
1
(s)
r
1
G
2
(s)
r
2
F
(s),
where G
1
(s) =
s
2
(
s
2
), G
2
(s) = (2)
1s
(s). Then Z
F
(s) is a meromorphic function
in the s-plan, holomorphic except for simple zeros at s = 0 and s = 1, and satises the
functional equation
Z
F
(s) = [
F
[
1
2
s
Z
F
(1 s).
Its residues at s = 0 and s = 1 are respectively
_
[
F
[Vol(I
1
F
/F
) and Vol(I
1
F
/F
).
Proof. We apply Tates theorem 2.10 to = [ [
s
, and f = f
v
with
f
v
=
_
_
e
x
2
v
if v is real;
e
xv xv
if v is complex;
1
Ov
if v is non-archimedean.
By the local computations in 1.12, we have
(f
v
, [ [
s
) =
_
s
2
(
s
2
) if v is real;
2
1s
(s) if v is complex;
(Nd
v
)
1
2
1
1Np
s
if v is non-archimedean.
Therefore, we get
(f, ) =
v
(f
v
, [ [
s
) = 2
r
2
s
[
F
[
1
2
Z
F
(s).
On the other hand, we have
f =
v
f
v
with
f
v
= f
v
if v is real,
f
v
(z) = 2f
v
(2z) if v is
complex, and
f
v
= (Nd
v
)
1
2
1
d
1
v
if v is non-archimedean. The local zeta functions are
(
f
v
, [ [
1s
) =
_
1s
2
(
1s
2
) if v is real;
2
s
(2)
s
(1 s) if v is complex;
(Nd
v
)
s 1
1Np
s1
if v is non-archimedean.
Hence, we obtain
(
f, [ [
s
) =
v
(
f
v
, [ [
s
) = 2
r
2
s
[
F
[
s
Z
F
(1 s).
The functional equation of Z
F
(s) follows immediately from (f, [ [
s
) = (
f, [ [
1s
). The
resides of Z
F
(s) follows from the residues of (f, [ [
s
) and the fact that f(0) = 1 and
f(0) = 2
r
2
[
F
[
1
2
.
References
[Ta50] Tate, John T. (1950), Fourier analysis in number elds, and Heckes zeta-functions, Algebraic
Number Theory (Proc. Instructional Conf., Brighton, 1965), Thompson, Washington, D.C., pp.
305347.
[We74] A. Weil, Basic Number Theory, Third Edition, Springer-Verlag, (1974).