X
X
(n)
(n)
V
V
L(1/2, ) =
n +
n
q
q
n
n
n=1
n=1
where V : R+ R is a smooth function satisfying
1 + O( x)
x<1
V (x) =
3/2 x2
O(x e
)
x1
Proof. Define (s) = (s/2)
L(1/2, ) =
1
(1/2)
1
2i
q s/2
.
We have
1+i
(s + 1/2, )
1i
ds
1
s
2i
1+i
(s + 1/2, )
1i
ds
s
by picking up the single residue inside the contour. Applying the functional
equation in the second integral, and making the change of variable s 7 s, this
is equal to
Z 1+i
Z 1+i
1
ds
ds
1
(s + 1/2, ) +
(s + 1/2, )
,
(1/2) 2i 1i
s
2i 1i
s
Writing the first term as (the second is similar)
Z 1+i s 1 s
X
( 2 + 4 ) q 2 1 ds
(n) 1
2i
ns s
n
( 14 )
1i
n=1
we see that the identity is true if we define
Z 1+i s 1
( 2 + 4 ) s ds
1
V (x) =
x
,
2i 1i
s
( 41 )
so it remains to verify the properties of V .
1
,
Smoothness follows by differentiating under the integral. To prove the estimate for small x, shift the contour left to <(s) = 1, passing a pole at 0 and
another at 1/2. Thus
1
V (x) = 1 + O( x) +
2i
To prove the estimate for large
say. Write
s
( +
2
Let 0 solve
0
2
V (x) =
5
4
1
2i
1+i
1i
( 2s + 14 ) s ds
x
= 1 + O( x).
1
s
( 4 )
s
2
1
s 5
1 ( 2 + 4 ).
+4
0 i
( 2s + 45 ) s
ds
0
5
x
( + )x0 ,
(1/4)
s(s/2 + 1/4)
2
4
since |( + it)| (), as may be seen directly from the integral definition.
By Stirlings approximation, the RHS is bounded by
2
3
0
3
0
5
0
5
exp
+
+
+
log
0 log x + O(1) x 2 ex ,
2
4
2
4
2
4
by using that
0
2
5
4
= x2 .
Remark. We may truncate the sums over n in the approximate functional equa
tion at n q log q with very small error. Indeed, the error is bounded by [sum
dyadically]
n q log q
e
3/4
n2
q
22A e
22A
q
A:2A q log q
Thus we find
|L(1/2, )|
n q log q
p
1
q 1/4 log q,
n
r
r
X
X
(n)
(n)
V
n
+
V
n
w
1w
w
1w
n
q
n
q
n=1
n=1
Proof sketch. We will treat only the even characters. The odd ones may be dealt
with similarly, (the associated smooth function V will be slightly different).
The function
X
d4 (n)(n)
L(s, )4 =
,
<(s) > 1
ns
n=1
satisfies the functional equation
(s, )4 = 4 (1 s, )4 .
Note that
s
q2
s
( )4 L(s, )4 ,
2
2
so that by mimicing the proof of the approximate functional equation for L(s, )
we may write
X (n)d4 (n) 2 n
X (n)d4 (n) 2 n
4
V
+
V
+ tiny error
L(1/2, )4 =
q2
q2
n
n
2+
2+
(s, )4 =
nq
nq
|L(1/2, )|
V
q 2
n
2+
nq
V
,
q2
n
A
2+ A
A+1
A:2 q
2 n<2
Z 2A+1 2
X
X (n)d4 (n)
(n)d4 (n) 2 n
V
=
V
d
q2
q2
n
n
2A
A
A+1
A
2 n<2
X
2A n<2A+1
2 n<y
(n)d4 (n)
V
n
2 A+1
2
q2
2A+1
2A
X
2A n<y
(n)d4 (n) 2 0
V
q2
n
2 y
q2
dy
V
dy
A
q2
q2
n
2
2A n<y
2
Z 2A+1 2 X
Z 2A+1 2 2 2
(n)d4 (n)
0 y
V
dy
dy A
2
q
q2
q2
n
A
2
2
2A n<y
2
Z 2A+1
X
1
(n)d4 (n)
Q
dy
2A A
n
2A
2 n<y
To justify the last line, we need to estimate the integral with V 0 , which is equal
to
Z 2A+1 22
q
V 0 (y)2 dy.
2
2A
q2
Making similar estimates to those in Theorem 22.1 one can show that V 0 (y)
| log y|3
Q
X X
Q
A:2A Q2+
Q2+
X
nQ2+
2A+1
2A
2A+1
2A
Q2 + 2A
2A
1
2A
2
X (n)d4 (n)
dy
n
2A n<y
X
2A n<y
d4 (n)2
dy
n
d4 (n)
Q2+ ,
n
by applying the multiplicative large sieve. Note that we may use d4 (n) n
(Example Sheet 1, Problems 5 and 6), although we could also estimate the sum
over n by (log Q)16 , e.g. by Perron summation.