Discrete Mean Estimates and The Landau-Siegel Zero

Discrete mean estimates and the Landau-Siegel zero
arXiv:2211.02515v1 [math.NT] 4 Nov 2022
Yitang Zhang
Abstract
Let χ be a real primitive character to the modulus D. It is proved that
L(1, χ) ≫ (log D)−2022
where the implied constant is absolute and effectively computable.

In the proof, the lower bound for L(1, χ) is first related to the distribution of zeros of
a family of Dirichlet L-functions in a certain region, and some results on the gaps between
consecutive zeros are derived. Then, by evaluating certain discrete means of the large sieve
type, a contradiction can be obtained if L(1, χ) is too small.
Table of Contents
1. Introduction
2. Notation and outline of the proof
3. The set Ψ1
4. Zeros of L(s, ψ)L(s, χψ) in Ω
5. Some analytic lemmas
6. Approximate formula for L(s, ψ)
7. Mean value formula I
8. Evaluation of Ξ11
10. Proof of Proposition 2.4
1
13. Approximation to Ξ14
14. Mean value formula II
15. Evaluation of Φ1
Appendix A. Some Euler products
Appendix B. Some arithmetic sums
References
1. Introduction
Let χ be a real primitive character to the modulus D. It is known that the Dirichlet
L-function L(s, χ) has at most one real and simple zero ρ̃ satisfying
1 − ρ̃ < c0 (log D)−1
where c0 > 0 is an absolute constant. Such a zero is called the Landau-Siegel zero. The
typical methods to determine zero-free regions for Dirichlet L-functions are unable to
eliminate the Landau-Siegel zero for an intrinsic reason.
The non-vanishing of L(s, χ) near s = 1 is closely related to the lower bound for the
value of L(s, χ) at s = 1. The well-known Siegel theorem [19] asserts that, for any ε > 0,
there exists a positive number C1 (ε) such that
L(1, χ) > C1 (ε)D −ε .
This implies, for any ε > 0, that there exists a positive number C2 (ε) such that
L(σ, χ) 6= 0 if σ > 1 − C2 (ε)D −ε .
However, the process of Siegel’s proof essentially involves certain assumptions, such as the
existence of a real primitive character χ̂ (mod D̂) with L(s, χ̂) vanishing on the interval
[1 − ε, 1) (see [8]), and the constant C1 (ε) relies on the unknown modulus D̂ necessarily .
This makes the result of Siegel’s theorem entirely ineffective.
It is known (see [9]) that the non-existence of the Landau-Siegel zero implies
L(1, χ) ≫ (log D)−1 . (1.1)
2
In the case χ(−1) = −1, Goldfeld [10] and Gross and Zagier [12] proved that
L(1, χ) ≫ D −1/2 (log D)1−ε
for any ε > 0, where the implied constant is effectively computable. This result surpasses
the trivial bound L(1, χ) ≫ D −1/2 and solves Gauss’ class number problem for imaginary
quadratic fields. Granville and Stark [11] proved that the uniform abc conjecture for
number fields implies (1.1) when χ(−1) = −1.
The main result of this paper is
Theorem 1 If χ is a real primitive character to the modulus D, then
L(1, χ) > c1 (log D)−2022
where c1 > 0 is an absolute, effectively computable constant.

As a direct consequence of Theorem 1 we have
Theorem 2 If χ is a real primitive character to the modulus D, then
L(σ, χ) 6= 0
for
σ > 1 − c2 (log D)−2024
where c2 > 0 is an absolute, effectively computable constant.
It is possible to replace the exponent −2022 in Theorem 1 by a larger (negative)
value if the current arguments are modified, but we will not discuss it in this paper. On
the other hand, it seems that the lower bound (1.1) can not be achieved by the present
methods.
Acknowledgements. The basic ideas of the present proof were initially formed during
my visit to the Institute for Advanced Study in the spring of 2014. I thank the Institute
for Advanced Study for providing me with excellent conditions. I also thank Professor
Peter Sarnak for his encouragement.
2. Notation and outline of the proof

Notation and conventions.
Throughout, t, u, v, z and σ denote real variables; x and y denote positive real
variables; s and w denote the complex variables s = σ + it and w = u + iv; d, h, j, k, l, m,
n and r denote natural numbers, while p and q denote primes; the arithmetic functions
3
µ(n) and ϕ(n) are defined as usual, while τk (n) denotes the k-fold divisor function. Let ∗
denote the Dirichlet convolution of arithmetic functions. Let e(s) = exp{2πis}. We write
X X
and
n (n,k)=1
for ∞ ∞
X X
and
n=1 n=1
(n,k)=1
R
respectively. Let (z) denote an integration from z − i∞ to z + i∞.
We let χ denote a real primitive character to the modulus D with D greater than a
sufficiently large and effectively computable number. Write
L = log D. (2.1)
Let c denote a positive, effectively computable and absolute constant, not necessarily
the same at each occurrence. All implied constants, unless specified, are absolute and
effectively computable. In this paper, an error term is usually given in the form that is
small enough for our purpose, although shaper estimates are possible.
For any Dirichlet character θ to the modulus k, let τ (θ) denote the Gauss sum
X
τ (θ) = θ(a)e(a/k)
a( mod k)
(thus we shall not write τ for τ2 , the ordinary divisor function). The expressions
X X′ X∗
, and
θ( mod k) θ( mod k) θ( mod k)
denote, respectively, a sum over all θ(mod k), a sum over all non-principal θ(mod k), and
a sum over all primitive θ(mod k). Let ψk0 denote the principle character (mod k).
In case θ(mod k) is a primitive character, the functional equation for L(s, θ) is
L(s, θ) = Z(s, θ)L(1 − s, θ̄), (2.2)
where −1
1−s

s−1/2 −s s
Z(s, θ) = τ (θ)π k Γ Γ
2 2
if θ(−1) = 1, and
−1
2−s

s−1/2 −s 1+s
Z(s, θ) = −iτ (θ)π k Γ Γ
2 2
4
if θ(−1) = −1. When t is large, it is convenient to use an asymptotic expression for
Z(s, θ) as follows. Let
ϑ(s) = 2(2π)s−1 Γ(1 − s) sin(πs/2) (2.3)
(this function is usually written as χ(s) in the literature). It is known that
−1
1−s

s−1/2 s
π Γ Γ = ϑ(s)
2 2
and −1
2−s

s−1/2 1+s
π Γ Γ = ϑ(s) cot(πs/2).
2 2
Assume t > 1. Then
i cot(πs/2) = 1 + O(e−πt).
Thus we have uniformly
Z(s, θ) = θ(−1)τ (θ)k −s ϑ(s)(1 + O(e−πt)), (2.4)
and
Z(s, θ)−1 = τ (θ̄)k s−1 ϑ(1 − s)(1 + O(e−πt)), (2.5)
the O(e−πt ) term being identically zero in case θ(−1) = 1.
Outline of the proof: Initial steps
Our aim is to derive a contradiction from the following
Assumption (A)
L(1, χ) < L−2022 .
The underlying ideas are inspired by the work of Goldfeld [10], Iwaniec and Sarnak
[15] and Conrey, Ghosh and Gonek [4]. The paper [10] exhibits a relationship between the
lower bound for L(1, χ) and the order of zeros of the function LE (s)LE (s, χ) at the central
point, where E is an elliptic curve; the paper [15] exhibits a relationship between the lower
bound for L(1, χ) and the non-vanishing of central values of a family of automorphic L-
functions; the paper [4] has definite influence on the idea of reducing the problem to
evaluating certain discrete means that is employed in the present work (the reader is also
referred to [3]).
The initial part of this paper consists in relating the lower bound for L(1, χ) to the
distribution of zeros of the function L(s, ψ)L(s, ψχ), with ψ belonging to a large set Ψ of
primitive characters, in a domain Ω. Thus we introduce
P = exp{L9 }. (2.6)
5
For notational simplicity we write
p∼P for P < p < P (1 + L−68 ).
Let Ψ be the set of all primitive characters ψ(mod p) with p ∼ P , and let

Ω = s : |ℜ(s − s0 )| < 1/2, |ℑ(s − s0 )| < L1 + 2 (2.7)
where
1
L1 = L405 , + 2πit0 with t0 = L519 .
s0 = (2.8)
2
Throughout, we will assume that p ∼ P and ψ is a primitive characters (mod p), i.e.,
ψ ∈ Ψ. Note that X
P := p = (1 + o(1))P 2L−77 . (2.9)
p∼P
In Section 3 we shall introduce a subset Ψ1 of Ψ and prove

Proposition 2.1. Let Ψ2 be the complement of Ψ1 in Ψ. If (A) holds, then
X
1 ≪ PL−739 .
ψ∈Ψ2
For ψ(mod p) ∈ Ψ, the average gap between consecutive zeros of L(s, ψ)L(s, ψχ) in
Ω is
2π
≃ = α(1 + O(αL)).
log(Dp2 t20 )
where
π
α= . (2.10)
log P
In Section 4 we shall prove
Proposition 2.2. Suppose that ψ ∈ Ψ1 . Then the following hold.
(i). All the zeros of L(s, ψ)L(s, ψχ) in Ω lie on the critical line.
(ii). All the zeros of L(s, ψ)L(s, ψχ) in Ω are simple.
(iii). The gap between any consecutive zeros of L(s, ψ)L(s, ψχ) in Ω is of the form
α + O(α2L).
It should be stressed that the set Ψ1 is formally defined. In fact, Proposition 2.2
and some other results for ψ ∈ Ψ1 are unconditionally derived from the definition of Ψ1 .
6
However, one is even unable to determine whether Ψ1 is empty or not without assuming
(A).
The results of Proposition 2.1 and 2.2 are not surprising and they go back to Heath-
Brown [15]. One may believe that the assertions (i) and (ii) of Proposition 2.2 hold for
all characters ψ in Ψ. On the other hand, however, in comparison with some conjectures
on the vertical distribution of zeros of the Riemann zeta function (see [2] and [17]), one
may claim that the gap assertion (iii) of Proposition 2.2 fails to hold for most of ψ in Ψ.
Our proof thus consists in deriving a contradiction from the gap assertion.
Assume ψ ∈ Ψ. Recall that the functional equation for L(s, ψ) is given by (2.2) with
θ = ψ. On the upper-half plane, since Z(s, ψ) is analytic and non-vanishing, there is an
analytic function Y (s, ψ) such that
Y (s, ψ)2 = Z(s, ψ)−1.
Let
M(s, ψ) = Y (s, ψ)L(s, ψ).
Note that M(s, ψ) is defined for t > 0 only, and, for each ψ, there are two choices of
M(s, ψ) up to ±, but the expressions
M(s1 , ψ)
M(s1 , ψ)M(s2 , ψ) and
M(s2 , ψ)
are independent of these choices. The functional equation (2.2) gives
M(s, ψ) = Y (s, ψ)−1 L(1 − s, ψ̄).
Since |Y (1/2 + it, ψ)| = 1, it follows that
|M(1/2 + it, ψ)| = |L(1/2 + it, ψ)|,
M(1/2 + it, ψ) ∈ R, (2.11)

and, consequently,
iM ′ (1/2 + it, ψ) ∈ R. (2.12)
The gap assertion (iii) of Proposition 2.2 can be restated as follows. If ψ ∈ Ψ1 and
(1/2 + iγ, 1/2 + iγ ′ ) is a pair of consecutive zeros of L(s, ψ)L(s, χψ) in Ω with γ ′ > γ,
then for some (large) constant c′ > 0,
|γ ′ − γ − α| < c′ α2 L.
Write
β1 = iα(1 − 5c′ αL), β2 = 2iα(1 + c′ αL), β3 = 3iα(1 − c′ αL). (2.13)
7
In what follows we assume ψ ∈ Ψ1 . Let Z(ψ) denote the set of zeros of L(s, ψ) in the
region
|σ − 1/2| < 1/2, |t − 2πt0 | < L1 (2.14)
(this is slightly smaller than Ω). Assume ρ ∈ Z(ψ). Note that
M ′ (ρ, ψ) = Y (ρ, ψ)L′ (ρ, ψ) 6= 0.
Write
−iM(ρ + β1 , ψ)M(ρ + β2 , ψ)M(ρ + β3 , ψ)
C ∗ (ρ, ψ) = .
M ′ (ρ, ψ)
By (2.11) and (2.12) we have
C ∗ (ρ, ψ) ∈ R.
It will be proved that

Lemma 2.3. Suppose ψ ∈ Ψ1 and ρ ∈ Z(ψ). Then
C ∗ (ρ, ψ) ≥ 0.
We introduce the smooth weight

√
(s − s0 )2

π
ω(s) = exp 2
with L2 = L400 , (2.15)
L2 4L2
which is positive for σ = 1/2. Lemma 2.3 implies that

X X
C ∗ (ρ, ψ)|H(ρ, ψ)|2 ω(ρ) ≥ 0. (2.16)
ψ∈Ψ1 ρ∈Z(ψ)
for all functions H(s, ψ) defined on Ω. Thus, in order to derive a contradiction from (A),
it suffices to find certain functions H(s, ψ) and show that , under Assumption (A), (2.16)
fails to hold.
A heuristic argument
It is natural to consider the function H(s, ψ) which is of the form
X f (log n/ log P ′ )µψ(n)
ns
n<P
′
for some P ′ < P , where f is a polynomial with f (1) = 0. It should be stressed, on

assuming (A), that if H(s, ψ) is of the above form with P ′ slightly smaller than P in the
logarithmic scale, the sum in (2.16) can be evaluated. This is analogous to the results of
Conrey, Iwaniec and Soundararajan [5] and [6]. However, it seems that such a choice of
8
H(s, ψ) is not good enough for our purpose. On the other hand, the left side of (2.16)
could be “close” to zero for certain non-zero functions H(s, ψ). For example, one may
consider the choice that H(s, ψ) is a linear combination of L(s + βj , χψ), 1 ≤ j ≤ 3. Thus,
inspired by the approximate formula for L(s, χψ), some new forms of H(s, ψ) could be
introduced. However, our goal is not to show that (2.16) fails to hold for certain H(s, ψ)
directly, instead, a variant of the argument will be adopted.
The crucial part of the proof is to construct functions H1 (s, ψ), H2 (s, ψ), J1 (s, ψ) and
J2 (s, ψ), each of which is of the form
X f (log n/ log P )χψ(n)
ns
n<P
′
with log P ′ / log P close to 1/2, and show that the left side of (2.16) is ”small” if
H(s, ψ) = H1 (s, ψ) + Z(s, χψ)H2 (s, ψ) or H(s, ψ) = J1 (s, ψ) + Z(s, χψ)J2 (s, ψ)
for σ = 1/2. On the other hand, the sum

X X
Ξ∗1 := C ∗ (ρ, ψ) H1 (ρ, ψ)J1 (ρ, ψ) + H2 (ρ, ψ)J2 (ρ, ψ) ω(ρ)

(2.17)
ψ∈Ψ1 ρ∈Z(ψ)
is not too small in absolute value.

For ψ ∈ Ψ1 and ρ ∈ Z̃(ψ), since
H1 (ρ, ψ)J1 (ρ, ψ) + H2 (ρ, ψ)J2 (ρ, ψ)

= H1 (ρ, ψ) + Z(ρ, χψ)H2 (ρ, ψ) J1 (ρ, ψ) − Z(ρ, χψ)J1 (ρ, ψ) − J2 (ρ, ψ) H2 (ρ, ψ)
and
Z(ρ, χψ)J1 (ρ, ψ) − J2 (ρ, ψ) = J1 (ρ, ψ) − Z(ρ, χψ)J2 (ρ, ψ),
it follows that

H1 (ρ, ψ)J1 (ρ, ψ) + H2 (ρ, ψ)J2 (ρ, ψ)

≤H1 (ρ, ψ) + Z(ρ, χψ)H2 (ρ, ψ)|J1 (ρ, ψ)| + J1 (ρ, ψ) − Z(ρ, χψ)J2 (ρ, ψ)|H2 (ρ, ψ)|.
This yields, by Lemma 2.3,

|Ξ∗1 | ≤ Ξ∗2 + Ξ∗3 (2.18)
where
X X
Ξ∗2 = C ∗ (ρ, ψ)H1 (ρ, ψ) + Z(ρ, χψ)H2 (ρ, ψ)|J1 (ρ, ψ)|ω(ρ),

(2.19)
ψ∈Ψ1 ρ∈Z(ψ)
X X
Ξ∗3 = C ∗ (ρ, ψ)J1 (ρ, ψ) − Z(ρ, χψ)J2 (ρ, ψ)|H2 (ρ, ψ)|ω(ρ).

(2.20)
ψ∈Ψ1 ρ∈Z(ψ)
9
Discrete mean estimates
We introduce some parameters and functions as follows. Let
P1 = P 0.504 , P2 = P 0.5 T −10 , P3 = P 0.498 , (2.21)

3iα 5iα
β6 =, β7 = . (2.22)
2 2
(The definition of β4 and β5 will be given in Section 8.) Write
X χψ(n) β6
log n P1
H11 (s, ψ) = s
1− , (2.23)
n<P
n log P1 n
1
X χψ(n) β7
log n P2
H12 (s, ψ) = s
1− , (2.24)
n<P
n log P 2 n
2
X χψ(n) β6
log n P3
H13 (s, ψ) = s
1− (2.25)
n<P
n log P3 n
3
(these sums are written as H1j (s, ψ) rather than H1j (s, χψ)). We also introduce the
numerical constants
ι2 = 0.94977 − 1.38995i, ι3 = −1.00635 − 0.22789i, ι4 = −0.68738 + 1.60688i. (2.26)
We define
H1 (s, ψ) = H11 (s, ψ) + ι2 H12 (s, ψ), H2 (s, ψ) = ῑ3 H13 (s, ψ) + ῑ4 H12 (s, ψ). (2.27)
Let f˜(z) be given by


500(z − 0.5)
 if 0.5 ≤ z ≤ 0.502
f˜(z) = 500(0.504 − z) if 0.502 ≤ z ≤ 0.504 (2.28)

0 otherwise.

We define
X χψ(n) log n
J1 (s, ψ) = s
f˜ , (2.29)
n
n log P
X χψ(n) log n
log Dt0
J2 (s, ψ) = f˜ + 0.004 − α̃ with α̃ = . (2.30)
ns log P log P
n
Let
6 ′ 2
Y q
a= L (1, χ) . (2.31)
π2 q+1
q|D
10
It can be shown that the assumption (A) implies
a ≫ 1.
(see Lemma 5.7).

Proposition 2.4. Assume that (A) holds. Then
|Ξ∗1 | > 5aP.
Ξ∗2 < 2aP.
Ξ∗3 = o(aP).
Under Assumption (A), a contradiction is immediately derived from (2.18), Proposi-

tion 2.4, 2.5 and 2.6, This proves Theorem 1.
It should be remarked that Proposition 2.5 follows from the inequalities (on assuming
(A))
X X 2
C ∗ (ρ, ψ)H1 (ρ, ψ) + Z(ρ, χψ)H2 (ρ, ψ) ω(ρ) < 0.001aP,

(2.32)
ψ∈Ψ1 ρ∈Z(ψ)
and X X
C ∗ (ρ, ψ)|J1 (ρ, ψ)|2 ω(ρ) < 3000aP, (2.33)
ψ∈Ψ1 ρ∈Z(ψ)
by Cauchy’s inequality (and Lemma 2.3). The proof of (2.32) involves some numerical
calculations.
We conclude this section by proving Lemma 2.3.
Proof of Lemma 2.3. By (2.) we see that M(ρ + iv, ψ) 6= 0 if |β2 | ≤ v ≤ |β3 |. This
implies, by the mean-value theorem, that
M(ρ + β2 , ψ)M(ρ + β3 , ψ) > 0,
since M(1/2 + it, ψ) is a real-valued continuous function in t. Similarly we have
M(ρ + β1 , ψ)
>0
M(ρ + iv, ψ)
11
if 0 < v ≤ |β1 |. Since
M(ρ + β1 , ψ) vM(ρ + β1 , ψ)
′
= lim+ ,
iM (ρ, ψ) v→0 M(ρ + iv, ψ)
it follows that
M(ρ + β1 , ψ)
≥ 0.
iM ′ (ρ, ψ)
This completes the proof. ✷
Remark. It is implied in the proof of Lemma 2.3 that

L(ρ + β1 , ψ) M(ρ + β1 , ψ)
L′ (ρ, ψ) = −i M ′ (ρ, ψ) (2.34)

for ψ ∈ Ψ1 and ρ ∈ Z̃(ψ).
3. The set Ψ1
Let ν(n) and υ(n) be given by
X ν(n) X υ(n)
ζ(s)L(s, χ) = , ζ(s)−1L(s, χ)−1 = , σ > 1,
n
ns n
ns
respectively. It is easy to see that
|υ(n)| ≤ ν(n) ≤ τ2 (n). (3.1)
Lemma 3.1. Assume (A) holds. Then

X ν(n)2
≪ L−2011 . (3.2)
n
D 4 <n≤P 2
Proof. Let
X ν(n)2
φ(s) = ζ(s)−2L(s, χ)−2
n
ns
which has the Euler product representation
Y
φ(s) = φp (s) (σ > 1).
p
For σ ≥ σ0 > 0, by checking the cases χ(p) = ±1 and χ(p) = 0 respectively, it can be
seen that
φp (s) = 1 + O(p−2σ ) if p ∤ D,
12
and
φp (s) = (1 − p−s )(1 + O(p−2σ )) if p|D,
the implied constant depending on σ0 . Thus φ(s) is analytic for σ > 1/2 and it satisfies
Y
φ(s) ≪ |1 − p−s | (3.3)
p|D
for σ ≥ σ1 > 1/2, the implied constant depending on σ1 . The left side of (3.2) is
X ν(n)2
exp{−n/P 2 } − exp{−n/D 4 }

≪
n
n
1
Z
= φ(1 + s)ζ(1 + s)2 L(1 + s, χ)2 (P 2s − D 4s )Γ(s) ds.
2πi (1)
Moving the line of integration to the left appropriately and applying standard estimates,
we see that the right side is equal to the residue of the integrand plus an acceptable error
O(D −c ). The residue at s = 0 can be written as
1
Z
φ(1 + s)ζ(1 + s)2 L(1 + s, χ)2 (P 2s − D 4s )Γ(s) ds
2πi |s|=α∗
with α∗ = L−2024 . If |s| = α∗ , then φ(1 + s) = O(1) by (3.3),
L(1 + s, χ) ≪ L−2022
by (A) and standard estimates, and
ζ(1 + s) ≪ L2024 , (P 2s − D 4s )Γ(s) ≪ L9 .
It follows that the residue at s = 0 is ≪ L−2011 , whence the result follows. ✷

Lemma 3.2. Assume (A) holds. Then we have
X ν(n)2 τ2 (n)2
≪ L−2007 .
n
D 4 <n≤D 8
Proof. As the situation is analogous to Lemma 3.1 we give a sketch only. It can be
verified that the function
X ν(n)2 τ2 (n)2
φ∗ (s) = ζ(s)−8L(s, χ)−8
n
ns
13
is analytic for σ > 1/2 and it satisfies
Y
φ∗ (s) ≪ |1 − p−4s |
p|D
for σ ≥ σ1 > 1/2, the implies constant depending on σ1 . Also, one can verify that
Z
φ∗ (1 + s)ζ(1 + s)8 L(1 + s, χ)8 (D 8s − D 4s )Γ(s) ds ≪ L−2007 .
|s|=α∗

Lemma 3.3. For any s and any complex numbers c(n) we have
X X c(n)ψ(n) 2 X |c(n)|2

≪P

ψ∈Ψ n≤P
ns
n≤P
n2σ
and
X X c(n)ψ(n) 2 X |c(n)|2

≪ P2 .
ns n2σ
ψ∈Ψ 2 n≤P 2 n≤P
Proof. The first assertion follows by the orthogonality relation; the second assertion
follows by the large sieve inequality. ✷
Let
X ν(n)ψ(n) X υ(n)ψ(n)
F (s, ψ) = , G(s, ψ) = .
4
ns 4
ns
n≤D n≤D
By (3.1) we may write

X ν20 (n)ψ(n) X υ20 (n)ψ(n)
F (s, ψ)20 = , G(s, ψ)20 =
80
ns 80
ns
n≤D n≤D
with |ν20 n| ≤ τ40 (n), |υ20 n| ≤ τ40 (n). Write

X ν20 (n)ψ(n) X υ20 (n)ψ(n)
X1 (x, ψ) = , X2 (x, ψ) = .
n≤x
ns0 n≤x
ns0
By Cauchy’s inequality and the first assertion of Lemma 3.3 we obtain

Z D80 2
X
80 80 |X1 (x, ψ)| + |X2 (x, ψ)|
|X1 (D , ψ)| + |X2 (D , ψ)| + dx ≪ PL1602 .
1 x
ψ∈Ψ
Thus we conclude
14
Lemma 3.4. The inequality
D 80
|X1 (x, ψ)| + |X2 (x, ψ)|
Z
80 80
|X1 (D , ψ)| + |X2 (D , ψ)| + dx < L1171 (3.4)
1 x
holds for all but at most O(PL−740 ) characters ψ in Ψ.

Write
X ν(n)ψ(n)
X3 (x, ψ) = for x > D 4 .
4
ns0
D <n≤x
Assume that (A) holds. By Cauchy’s inequality, the second assertion of Lemma 3.2 and
Lemma 3.1,
P2 2
X |X3 (x, ψ)|
Z
2
|X3 (P , ψ)| + dx ≪ P 2 L−1993 ≪ PL−1909 .
ψ∈Ψ D4 x
Thus we conclude
Lemma 3.5 Assume that (A) holds. The inequality
P2
|X3 (x, ψ)|
Z
2
|X3 (P , ψ)| + dx < L−585 (3.5)
D4 x

Let X
ς(n) = ν(l)υ(m).
n=lm
l,m≤D 4
We have ς(n) = 0 unless n = 1 or D 4 < n ≤ D 8 . Write

X ς(n)ψ(n)
X4 (x, ψ) =
4
ns0
D <n≤x
for x > D 4 . It is direct to verify that

X
|ς(n)| ≤ ν(l)|υ(m)| ≤ ν(n)τ2 (n).
n=lm
Assume that (A) holds. By Cauchy’s inequality, the first assertion of Lemma 3.2 and
Lemma 3.1,
Z D8 2
X |X4 (x, ψ)|
8
|X4 (D , ψ)| + dx ≪ PL−2005 .
ψ∈Ψ D 4 x
15
Thus we conclude
Lemma 3.6. Assume that (A) holds. The inequality
D8
|X4 (x, ψ)|
Z
8
|X4 (D , ψ)| + dx < L−633 (3.6)
D4 x

We are now in a position to give the definition of Ψ1 : Let Ψ1 be the subset of Ψ such
that ψ ∈ Ψ1 if and only if the inequalities (3.4), (3.5) and (3.6) simultaneously hold.
Proposition 2.1 follows from Lemma 3.4, 3.5 and 3.6 immediately.
4. Zeros of L(s, ψ)L(s, χψ) in Ω

In this section we prove Proposition 2.2. We henceforth assume that ψ(mod p) ∈ Ψ1 .
This assumption will not be repeated in the statements of Lemma 4.1-4.8.
We begin by proving some consequences of the inequalities (3.4)-(3.6).
Lemma 4.1. Let
Ω1 = s : 1/2 − (100L)−1 log L < σ < 1 + (100L)−1 log L,

|t − 2πt0 | < L1 + 5 .
If s ∈ Ω1 , then
|F (s, ψ)| + |G(s, ψ)| ≪ L79 .
Proof. By the Stieltjes integral we may write

Z D 80
F (s, ψ) 20
=1+ xs0 −s d{X1 (x, ψ)}.
1
For s ∈ Ω1 and 1 ≤ x ≤ D 80 we have

s0 −s
d s −s
|x | ≪ L, ≪ x−1 L406 .
dx x
0

Hence, by partial integration,

D 80
|X1 (x, ψ)|
Z
20 406 80
|F (s, ψ)| ≪1+L |X1 (D , ψ)| + dx .
1 x
For G(s, ψ) an entirely analogous bound is valid. The result now follows by (3.4). ✷
16
Lemma 4.2. If s ∈ Ω1 , then
F (s, ψ)G(s, ψ) = 1 + O(L−227 ).
Proof. We have
D8
ς(n)ψ(n)
X Z
F (s, ψ)G(s, ψ) − 1 = = xs0 −s d{X4 (x, ψ)}.
ns D4
D 4 <n≤D 8
Thus, similar to (4.2), by partial integration we obtain

D8
|X4 (x, ψ)|
Z
406 8
F (s, ψ)G(s, ψ) − 1 ≪ L |X4 (D , ψ)| + dx ,
D4 x
the right side being O(L−227 ) by (3.6). ✷
Lemma 4.3. Let
1/2 − L−1 < σ < 1 + L−1 ,

Ω2 = s : |t − 2πt0 | < L1 + 4 .
F′
(s, ψ) = O(L).
F
Proof. Assume |w| ≤ (200L)−1 log L, so that s + w ∈ Ω1 . By Lemma 4.1 and 4.2,
L−88 ≪ |F (s + w, ψ)| ≪ L88 .
Thus the logarithm

F (s + w, ψ)
l(s, w) := log ,
F (s, ψ)
which vanishes at w = 0, is analytic in w, and it satisfies
ℜ{l(s, w)} ≪ log L.
Since
F′ ∂
(s, ψ) = l(s, w)|w=0,
F ∂w
the result follows by Lemma 4 of [15, Chapter 2]. ✷
We proceed to establish an approximate formula for L(s, ψ)L(s, χψ). For this purpose
we first introduce a weight g(x) that will find application at various places. Let
1 xw ω1 (w)
Z
g(x) = dw (c > 0)
2πi (c) w
17
with
ω1 (w) = exp{w 2 /(4L30 )}.
We may write Z x
1
Z
w−1
g(x) = y dy ω1 (w) dw.
2πi (c) 0
Since
1 L15
Z
exp{(log y)w + w 2 /(4L30 )} dw = √ exp{−L30 (log y)2},
2πi (c) π
it follows, by changing the order of integration, that
L15 x dy
Z
g(x) = √ exp{−L30 (log y)2 } .
π 0 y
This yields. by substituting t = L15 log y,

L15 log x
1
Z
g(x) = √ exp{−t2 } dt. (4.1)
π −∞
Thus the function g(x) is increasing and it satisfies 0 < g(x) < 1. Further we have
g(x) = 1 + O(exp{−L30 log2 x}) if x ≥ 1 (4.2)
and
g(x) = O(exp{−L30 log2 x}) if x ≤ 1. (4.3)
Note that χψ is a primitive character (modDp). Write
Z̃(s, ψ) = Z(s, ψ)Z(s, χψ),
so that
L(s, ψ)L(s, χψ) = Z̃(s, ψ)L(1 − s, ψ̄)L(1 − s, χψ̄). (4.4)
Recall that s0 = 1/2 + 2πit0 . Assume that
|ℜ(s − s0 )| < 100, |ℑ(s − s0 )| < L1 + 3.
By (2.4) with θ = ψ and θ = χψ we have
Z̃(s, ψ) = χ(−1)τ (ψ)τ (χψ)(Dp2 )−s ϑ(s)2 (1 + O(e−πt )).
This yields, by Stirling’s formula,
|Z̃(s, ψ)| = (Dp2 t20 )1/2−σ (1 + o(1)) (4.5)
18
and
Z̃ ′
(s, ψ) = −2 log P + O(L). (4.6)
Z̃
Lemma 4.4. Let

Ω3 = s : 1/2 − α < σ < 1 + α, |t − 2πt0 | < L1 + 3 .
L(s, ψ)L(s, χψ) = F (s, ψ) + Z̃(s, ψ)F (1 − s, ψ̄) + O(L−179 ).
Proof. By the residue theorem,

Z
1 ω1 (w) dw
Z
L(s, ψ)L(s, χψ) = − L(s + w, ψ)L(s + w, χψ)P (9/5)w .
2πi (1) (−σ−1/2) w
By (4.2) and (4.3),
1 ω1 (w) dw
Z
L(s + w, ψ)L(s + w, χψ)P (9/5)w
2πi (1) w
X ν(n)ψ(n) P 9/5

= F (s, ψ) + g + O(ε)
4 2
ns n
D <n<P
where
ε = exp{−cL10 }.
By (3.5) and partial summation, the second sum on the right side above is
P2 Z P2
P 9/5 s0 −s |X3 (x, ψ)|
Z
= g x 2
dX3 (x, ψ) ≪ L1 |X3 (P , ψ)| + dx ≪ L−180 .
D4 x D 4 x
On the other hand, by the functional equation, for u = −σ − 1/2,

X ν(n)ψ̄(n)
L(s + w, ψ)L(s + w, χψ) = Z̃(s + w, ψ) .
n
n1−s−w
This sum is split into three sums according to n ≤ D 4 , D 4 < n ≤ P 2 and P 2 < n. The
proof is therefore reduced to showing that
1 ω1 (w) dw
Z
Z̃(s + w, ψ)F (1 − s − w, ψ̄)P (9/5)w
2πi (−σ−1/2) w (4.7)
−c
= −Z̃(s, ψ)F (1 − s, ψ̄) + O(P ),
19

1 ν(n)ψ̄(n) ω1 (w) dw
Z X
Z̃(s + w, ψ) 1−s−w
P (9/5)w ≪ L−179 (4.8)
2πi (−σ−1/2) n w
D 4 <n≤P 2
and X
1 ν(n)ψ̄(n) ω1 (w) dw
Z
Z̃(s + w, ψ) 1−s−w
P (9/5)w ≪ P −c . (4.9)
2πi (−σ−1/2) 2
n w
n>P
To prove (4.7) we move the contour of integration to the vertical segments

(
w = 10 + iv with |v| < L20 ,
w = −σ − 1/2 + iv with |v| ≥ L20 ,
and to the two connecting horizontal segments
w = u ± iL20 with − σ − 1/2 ≤ u ≤ 10.
By a trivial bound for ω1 (w), (4.5) and the residue theorem we obtain (4.7).
To prove (4.8) we move the contour of integration to the vertical segments
(
w = −α + iv with |v| < L20 ,
w = −σ − 1/2 + iv with |v| ≥ L20 ,
w = u ± iL20 with − σ − 1/2 ≤ u ≤ −α.
By a trivial bound for ω1 (w) and (4.5) we see that the left side of (4.8) is
Z L20 X
1−2σ
ν(n)ψ(n) dv
≪P +ε
−L20

4 2
ns∗ +iv α + iv
D <n≤P
with s∗ = 1 + α − s̄. By partial integration,

X ν(n)ψ(n) Z P 2 Z P2
|X3 (x, ψ)|

s0 −s∗ −iv 2σ−1 2
s∗ +iv
= x dX3 (x, ψ) ≪ P L1 |X3 (P , ψ)|+ dx
4 2
n D 4 D 4 x
D <n≤P
for |v| ≤ L20 . From this and (3.5) we obtain (4.8).

The estimate (4.9) follows by moving the contour of integration to the vertical segments
(
w = −σ − 10 + iv with |v| < L20 ,
w = −σ − 1/2 + iv with |v| ≥ L20 ,
w = u ± iL20 with − σ − 10 ≤ u ≤ −σ − 1/2,
20
and applying (4.5) and trivial bounds for ω1 (w) and the involved sum. ✷
In order to prove Proposition 2.2, it is appropriate to deal with the function
L(s, ψ)L(s, χψ)
A(s, ψ) = .
F (s, ψ)
By Lemma 4.2, A(s, ψ) is analytic and it has the same zeros as L(s, ψ)L(s, ψχ) in Ω1 .
Further, for s ∈ Ω1 , we have
F (s, ψ)−1 ≪ L79
by Lemma 4.1 and 4.2. This together with Lemma 4.4 implies that
A(s, ψ) = 1 + B(s, ψ) + O(L−100), (4.10)
for s ∈ Ω3 , where
F (1 − s, ψ̄)
B(s, ψ) = Z̃(s, ψ) .
F (s, ψ)
The proof of Proposition 2.2 is reduced to proving three lemmas as follows.
Lemma 4.5. If
1
+ α2 < σ < 1, |t − 2πt0 | < L1 + 2,
2
then
A(s, ψ) 6= 0.
Proof. We discuss in two cases.

Case 1. 1/2 + L−1 ≤ σ < 1.
By Lemma 4.2 and trivial estimation,
F (1 − s, ψ̄)
≪ Dc .
F (s, ψ)
Hence, by (4.5),
B(s, ψ) ≪ P 1/2−σ .
The result now follows by (4.10).
Case 2. 1/2 + α2 ≤ σ < 1/2 + L−1 .
Assume 1/2 ≤ σ ′ ≤ σ. Then both σ ′ + it and 1 − (σ ′ + it) lie in Ω2 . Hence, by Lemma
4.3 and (4.6),
B′ ′ Z̃ ′ F′ F′
(σ + it, ψ) = (σ ′ + it, ψ) − (σ ′ + it, ψ) − (1 − σ ′ − it, ψ̄)
B Z̃ F F (4.11)
= − 2 log P + O(L).
21
Since |B(1/2 + it, ψ)| = 1, it follows that
Z σ ′
B ′

log |B(s, ψ)| = ℜ (σ + it) dσ < (1/2 − σ) log P,
1/2 B
Hence, by (4.10),
|A(s, ψ)| > 1 − P 1/2−σ + O(L−100 ) ≫ α. ✷
Lemma 4.6. Suppose ρ = β + iγ is a zero of A(s, ψ) satisfying

1 1
≤ β < + α2 , |γ − 2πt0 | < L1 + 2.
2 2
Then β = 1/2, A′ (ρ, ψ) 6= 0 and A(1/2 + iγ + w, ψ) 6= 0 if 0 < |w| < α(1 − c′ αL) where
c′ is a sufficiently constant.
Proof. It suffices to show that the function A(1/2 + iγ + w, ψ) has exactly one zero
inside the circle |w| = α(1 − c′ αL), counted with multiplicity. By the Rouché theorem,
this can be reduced to proving that
|A(1/2 + iγ + w, ψ) − (1 − P −2w )| < |1 − P −2w | for |w| = α(1 − c′ αL),
since the function 1 − P −2w has exactly one zero inside this circle which is at w = 0. In
fact, we can prove that
A(1/2 + iγ + w, ψ) − (1 − P −2w ) ≪ αL (4.12)
if |w| < 2α, the implied constant being independent of c′ , and
|1 − P −2w | > 6c′ αL (4.13)
if |w| = α(1 − c′ αL).

Assume |w| < 2α. By (4.10) we have
A(1/2 + iγ + w, ψ) − (1 − P −2w ) = B(1/2 + iγ + w, ψ) + P −2w + O(L−100 ).
Noting that both s and 1 − s are in Ω2 if s lies on the segment connecting ρ and
1/2 + iγ + w, by (4.11) we have
1/2+iγ+w
B(1/2 + iγ + w, ψ) B′
Z

= exp (s, ψ) ds = exp − 2w log P + O(αL)
B(ρ, ψ) ρ B
−2w
=P + O(αL),
22
Since B(ρ, ψ) = −1 + O(L−100) by (4.10), the estimate (4.12) follows.
Now assume |w| = α(1 − c′ αL). Write
w = α(1 − c′ αL)(cos θ + i sin θ).
Then
|P −2w | = exp − 2π(1 − c′ αL) cos θ , ℑ{P −2w } = −|P −2w | sin 2π(1 − c′ αL) sin θ ;

if cos θ > c′ αL, then |P −2w | < 1 − 6c′ αL;

if cos θ < −c′ αL, then |P −2w | > 1 + 6c′ αL;
if | cos θ| ≤ c′ αL, then |ℑ{P −2w }| > 6c′ αL,
since p
1 − (c′ αL)2 ≤ | sin θ| ≤ 1,
so that
| sin 2π(1 − c′ αL) sin θ | = 2πc′ αL(1 + o(1)).

In either case (4.16) holds. ✷

Lemma 4.5 and 4.6 together imply the assertions (i) and (ii) of Proposition 2.2. It is
also proved that the gap between any distinct zeros of A(s, ψ) in Ω is > α(1 − c′ αL). To
complete the proof of the gap assertion (iii), it now suffices to prove
Lemma 4.7. Suppose ρ = 1/2 + iγ is a zero of A(s, ψ) satisfying |γ − 2πt0 | < L1 + 2.
Then the function A(ρ + w, ψ) has exactly three zeros inside the circle |w| = α(1 + c′ αL),
counted with multiplicity.
Proof. In a way similar to the proof of Lemma 4.6, it is direct to verify that
|A(ρ + w, ψ) − (1 − P −2w )| < |1 − P −2w |
if |w| = α(1 + c′ αL). Hence, the functions A(ρ + w, ψ) and 1 − P −2w have the same
number of zeros inside this circle, while the later has exactly three zeros inside the same
circle which are at w = 0, w = iα and w = −iα. ✷
We conclude this section by giving a result which is implied in the proof of Proposition
2.2.
Lemma 4.8. Assume that ρ is a zero of L(s, ψ)L(s, χψ) in Ω. Then we have
Z̃(ρ, ψ)−1 = −G(ρ, ψ)F (1 − ρ, ψ̄) + O(L−100 ).
Proof. It follows from Lemma 4.4 that
F (ρ, ψ) + Z̃(ρ, ψ)F (1 − ρ, ψ̄) ≪ L−179 .
The result follows by multiplying both sides by Z̃(ρ, ψ)−1 G(ρ, ψ) and applying Lemma
4.1. ✷
23
5. Some analytic lemmas
Lemma 5.1. Suppose ψ(mod p) ∈ Ψ, |σ − 1/2| ≤ α, |t − 2πt0 | < L1 + 2, u = 0 and
|v| < L20 . Then
Z(s + w, ψ) − Z(s, ψ)(pt0 )−w
≪ L−114 (5.1)
w
Z(s + w, ψ) − Z(s, ψ)(P t0 )−w
≪ L−68 (5.2)
w
Z(s + w, χψ) − Z(s, χψ)(Dpt0 )−w
≪ L−114 (5.3)
w
and
Z(s + w, χψ) − Z(s, χψ)(DP t0)−w
≪ L−68 (5.4)
w
Proof. We have
Z w ′
Z(s + w, ψ) Z ′ ′
= exp (s + w , ψ) dw .
Z(s, ψ) 0 Z
Assume |w ′ | ≤ |w|. By (2.6) and the Stirling formula,
Z′ ϑ′
(s + w ′ , ψ) = − log p + (s + w ′ ) + O(ε) = − log(pt/2π) + O(1/t0 ).
Z ϑ
Hence
Z(s + w, ψ) = Z(s, ψ)(pt/2π)−w + O(|w|/t0).
This yields (5.1) and (5.2) since
pt/2π = pt0 (1 + O(L−114 )), pt/2π = P t0 (1 + O(L−68)).
The proofs of (5.3) and (5.4) are similar. ✷

Lemma 5.2. Let ψ and s be as in Lemma 5.1. Then
Y (s + β1 , ψ)Y (s + β2 , ψ)Y (s + β3 , ψ)
= (pt0 )β3 Z(s, ψ)−1(1 + O(L−123 )).
Y (s, ψ)
Proof. The left side is

Y Y (s + βj , ψ)
−1
Z(s, ψ) .
1≤j≤3
Y (s, ψ)
24
By (2.6) and the Stirling formula, for |w| < 5α,
Y′ 1 Z′ 1
(s + w, ψ) = − (s + w, ψ) = log(pt0 ) + O(L−114 ).
Y 2Z 2
Hence, for 1 ≤ j ≤ 3,
βj
Y′
Z
Y (s + βj , ψ)
= exp (s + w, ψ) dw = (pt0 )βj /2 (1 + O(L−123 )).
Y (s, ψ) 0 Y
The result now follows since

β1 + β2 + β3
= β3 . ✷
2
Recall that ϑ(s) and ω(s) are given by (2.3) and (2.15) respectively. It is known that
ϑ(1 − s) = 2(2π)−s Γ(s) cos(πs/2).
For t > 1 we have

ϑ(1 − s) = ϑ∗ (1 − s) 1 + O(e−πt)

(5.5)
where
ϑ∗ (1 − s) = (2π)−s Γ(s)e(−s/4).
Let
1
Z
∆1 (x) = x−s ϑ∗ (1 − s)ω(s) ds (5.6)
2πi (3/2)
and
∆(x) = ∆1 (x)e(x). (5.7)
Note that √ 2
π(x − t0 )

π
ω(1/2 + 2πix) = exp − > 0.
L2 L2
Lemma 5.3. If x ≤ t1.02

0 , then

∆(x) = ω(1/2 + 2πix) 1 + O(α) + O(ε); (5.8)
if x > t1.02
0 , then
∆(x) ≪ exp{−(10−2 L2 log x)2 } + exp{−x0.99 /L2 }. (5.9)
25
Proof. By the Mellin transform (see [1], Lemma 2) we have
Z ∞
∆1 (x) = exp{(s0 − 1) log y − L22 log2 y − 2πixy} dy
0
where the logarithm vanishes at y = 1. This yields, by substituting y = eu ,

Z ∞
∆(x) = exp{s0 u − L22 u2 − 2πix(eu − 1)} du. (5.10)
−∞
First assume x ≤ t1.02

0 . We may write
exp{s0 w − L22 w 2 − 2πix(ew − 1)} = exp{2πi(t0 − x)w − L22 w 2 }f ∗ (x, w)
with
f ∗ (x, w) = exp{(1/2)w − 2πix(ew − 1 − w)}.
By the relation
Z ∞
exp{2πi(t0 − x)u − L22 u2 } du = ω(1/2 + 2πix)
−∞
and Cauchy’s theorem, the proof of (5.8) is reduced to showing that

Z
exp{2πi(t0 − x)w − L + 22 w 2 } f ∗ (x, w) − 1 dw ≪ αω(1/2 + 2πix) + ε

(5.11)
Lj
for 1 ≤ j ≤ 5, where Lj denote the segments
L1 = (−∞, −u∗ ], L2 = [−u∗ , −u∗ + iv ∗ ], L3 = [−u∗ + iv ∗ , u∗ + iv ∗ ],
L4 = {u∗ + iv ∗ , u∗ ], L5 = [u∗ , ∞)
with
π(t0 − x)
u∗ = L−1 5
2 L , v∗ = .
L22
If w ∈ L1 ∪ L5 , then
exp{2πi(t0 − x)w − L22 w 2 } f ∗ (x, w) − 1 ≪ exp{−L22 u2 }(1 + eu/2 ).

Thus the left side of (5.11) is trivially O(ε) if j = 1, 5. By simple estimates,

Z
exp{2πi(t0 − x)w − L2 w 2 } dw ≪ ω(1/2 + 2πix) + ε

2
Lj
if j = 2, 3, 4. Since t3.06 4
0 /L2 ≪ α, we have
f ∗ (x, w) − 1 ≪ |w| + x|w|2 ≪ α
26
for w ∈ L2 ∪ L3 ∪ L4 . These estimates together imply (5.11).
Now assume x > t1.020 . By Cauchy’s theorem, the proof of (5.9) is reduced to showing
that
Z
exp{s0 w − L22 w 2 − 2πix(ew − 1)} dw ≪ exp{−(10−2 L2 log x)2 } + exp{−x0.99 /L2 }
L′j
(5.12)
where L′j , 1 ≤ j ≤ 3, denote the segments
L′1 = {−∞, −10−2 log x], L′2 = [−10−2 log x, −10−2 log x − i/L2 ],
L′3 = [−10−2 log x − i/L2 , ∞ − i/L2 ).

We have
u
1 xe sin v
ℜ{s0 w − L22 w 2 w 2 2 2
− 2πix(e − 1)} = u − L2 (u − v ) + 2πv − t0 . (5.13)
2 v
If w ∈ L′1 , then the right side of (5.13) is

1
≤ u − (10−2 L2 log x)2 .
2
This yields (5.12) with j = 1. For w ∈ L′2 ∪ L′3 we have
xeu ≥ x0.99 > 2t0 and − 1/L2 ≤ v ≤ 0,
so that
xeu sin v

2π − t0 > x0.99 .
v
If w ∈ L′2 , then (L2 u)2 = (10−2L2 log x)2 , so the right side of (5.13) is
≤ −(10−2 L2 log x)2 + O(1).
This yields (5.12) with j = 2 . If w ∈ L′3 , then v = −1/L2 , and the right side of (5.13) is
1
< 1 + u − (L2 u)2 − x0.99 /L2 .
2
This yields (5.12) with j = 3 . ✷
As a consequence of Lemma 5.3, the Mellin transform

Z ∞
δ(s) := ∆(x)xs−1 dx (5.14)
0
is analytic for σ > 0.
27
Lemma 5.4. (i). If 1/2 ≤ σ ≤ 2, then
δ(s) ≪ Lc |s|−2 .
(ii). If |s − 1| < 10α, then
δ(s) = 1 + O(α log L).
Proof. (i). Using partial integration twice we obtain

Z ∞
1
δ(s) = ∆′′ (x)xs+1 ds.
s(s + 1) 0
By (5.10) we have
Z ∞
′′ 2
∆ (x) = −4π (eu − 1)2 exp{s0 u − B 2 u2 − 2πix(eu − 1)} du.
−∞
Thus some upper bounds for ∆′′ (x) analogous to Lemma 5.3 can be obtained, and (i)
follows.
(ii). Assume |s − 1| < 10α. By Lemma 5.3, on the right side of (5.14), the integral on
the part |x − t0 | ≥ L1 contributes O(ε). For |x − t0 | < L1 we have
xs−1 = 1 + O(α log L).
Since Z ∞
ω(1/2 + 2πix) dx = 1 + O(ε),
0
(ii) follows. ✷
Throughout the rest of this paper we assume that (A) holds. This assumption will not
be repeated in the statements of the lemmas and propositions in the sequel.
The next two lemmas are weaker forms of the Deuring-Heillbronn Phenomenon.
Lemma 5.5. The function L(s, χ) has a simple real zero ρ̃ such that
1 − ρ̃ = O(L−2022 ) (5.15)
and L(s, χ) has no other zeros in the region
σ > 1 − 2L−1 , |t| < 2D.
28
Lemma 5.6. For any primitive character θ(mod r) with r < T and θ 6= χ,
X
θ(p)p1+it ≪ P exp{−L9/2 }
p∼P
if |t| ≤ D.
Lemma 5.7. We have

D
L′ (1, χ) ≫ .
ϕ(D)
Proof. The right side of the equality
1 D 4s ω1 (s) X ν(n) D 4
Z
ζ(1 + s)L(1 + s, χ) ds = g .
2πi (1) s n
n n
is X1 D
≫ ≫ ,
n ϕ(D)
n|D
since ν(n) = 1 if n|D, while the left side is, by moving the line of integration to the left
appropriately and applying (A), equal to L′ (1, χ) + o(1). ✷
Lemma 5.8. If
α ≤ |s − 1| ≤ 10α,
then
L(s, χ) = L′ (1, χ)(s − 1) + O(α2 )
where
α2 = L−15 .
Proof. This follows from the relation
Z s
′ ′′
L(s, χ) = L(1, χ) + L (1, χ)(s − 1) + (s − w)L (w, χ) dw,
1
′′
(A) and a simple bound for L (w, χ). ✷
Lemma 5.9. Suppose ψ ∈ Ψ1 , |σ − 1/2| ≤ α, |t − 2πt0 | ≤ L1 + 10 and |s − ρ| ≫ α
for any zero ρ of L(s, ψ). Then
L(s + β1 , ψ)
≪ log P.
L(s, ψ)
Proof. It is known that
L′ ′ X 1
(s , ψ) = ′
+ O(1/α) (5.16)
L s −ρ
|ρ−s |<1
′
29
for |ℜ{s′ } − 1/2| ≤ α and |ℑ{s′ − 2πt0 }| < L1 + 10, where ρ runs through the zeros of
L(s′ , ψ).
We can assume L(s + β1 , ψ) 6= 0. Suppose σ ≥ 1/2. By (5.16),
′
L
−ℜ (u + it + β1 , ψ) < O(1/α) for σ < u < 1/2 + α,
L
so that
|L(s + β1 , ψ)|
log < O(1).
|L(1/2 + α + it + β1 , ψ)|
By (9.1) and the condition |s − ρ| ≫ α for any ρ,
′
L
ℜ (u + it, ψ) = O(1/α) for σ < u < 1/2 + α,
L
so that
|L(1/2 + α + it, ψ)|
log < O(1).
|L(s, ψ)|
Further, by (5.16) and Proposition 2.2 (iii),
′
L
(1/2 + α + it′ , ψ) < 1
X 1 log log P
L α + O(1/α) < + O(1/α)
k α
k<1/al
for t < t′ < t + |β1 |, so that

|L(1/2 + α + it + β1 , ψ)|
log < log log P + O(1).
|L(1/2 + α + it, ψ)|
Combining theses estimates we obtain the result. In the case σ < 1/2 the proof is
analogous. ✷
6. Approximate formula for L(s, ψ)

Write (
g(y) if y > 1/2,
g ∗ (y) =
0 otherwise.
Let
P4 = P T −2t0 with T = exp{L1.1}.
Lemma 6.1. Suppose ψ(mod p) ∈ Ψ, |σ − 1/2| < 2α and |t − 2πt0 | < L1 + 2. Then
L(s, ψ) = K(s, ψ) + Z(s, ψ)N(1 − s, ψ̄) + O(E1 (s, ψ)),
30
where
X ψ(n) P4 ∗
K(s, ψ) = s
g
n
n n
X ψ̄(n) T 2
N(1 − s, ψ̄) = 1−s
g∗
n
n n
and where
L20

X ψ(n)
Z
−68
E1 (s, ψ) = L ω (iv)dv + ε.
s+iv 1
−L 20

3
n
n<T
Proof. By (4.3) we have

1 P w ω1 (w) dw
Z
L(s + w, ψ) 4 = K(s, ψ) + O(ε).
2πi (1) w
The left side above is, by moving the line of integration to u = −1, equal to
1 P w ω1 (w) dw
Z
L(s, ψ) + L(s + w, ψ) 4 .
2πi (−1) w
It therefore suffices to show that

1 P w ω1 (w) dw
Z
L(s + w, ψ) 4 = −Z(s, ψ)N(1 − s, ψ̄) + O(E1 (s, ψ)). (6.1)
2πi (−1) w
For u = −1 we have, by the functional equation (2.2) with θ = ψ,

X X ψ̄(n)
ψ̄(n)
L(s + w, ψ) = Z(s + w, ψ) +
3
n1−s−w 3
n1−s−w
n<T n≥T
We first show that

ψ̄(n) P4w ω1 (w) dw
Z X
Z(s + w, ψ) ≪ ε. (6.2)
(−1) 3
n1−s−w w
n≥T
We move the contour of integration in (6.2) to the vertical segments
u = −1, |v| > L20
and
u = −L9 , |v| ≤ L20
with the horizontal connecting segments
−L9 ≤ u ≤ −1 |v| = L20 .
31
By a trivial bound for ω1 (w) and simple estimates, , the integrals on the segments u = −1,
|v| > L20 contribute ≪ ε. On the other hand, in the case −L9 ≤ u ≤ −1, |v| ≤ L20 , by
(4.) we have
Z(s + w, ψ)P4w ≪ (2T 2 )−u
and
X ψ̄(n)
1−s−w
≪ T 3u+1/2 ,
3
n
n≥T
whence (6.2) follows. The proof of (6.1) is therefore reduced to showing that
−1+iL20
X
1
Z
Z(s + w, ψ)
2πi −1−iL20 3
n1−s−w w (6.3)
n<T
= − Z(s, ψ)N(1 − s, ψ̄) + O(E1 (s, ψ)).
Note that P t0 /P4 = T 2 . The left side of (6.3) is equal to I ′ + I ′′ with

−1+iL20
ψ̄(n) T −2w ω1 (w) dw
X
1
Z
′
I = Z(s, ψ) · , (6.4)
2πi −1−iL20 3
n1−s−w w
n<T

X
1
Z
′′ −w

I = Z(s + w, ψ) − Z(s, ψ)(P t0 ) . (6.5)
2πi (−1) 2
n1−s−w w
n<T
20
Replacing the segment u = −1, |v| ≤ L by u = −1 and using the change of variable
w → −w, we obtain
I ′ = −Z(s, ψ)N(1 − s, ψ̄) + O(ε).
On the other hand, moving the segment u = −1, |v| ≤ L20 to u = 0, |v| ≤ L20 and
applying Lemma 5.1, we find the the right side of (6.5) is ≪ E1 (s, ψ). ✷
7. Mean-value formula I
Let N (d) denote the set of positive integers such that h ∈ N (d) if and only if every
prime factor of h divides d (note that 1 ∈ N (d) for every d and N (1) = {1}). Assume
1 ≤ j ≤ 3 in what follows. Write
X κ(dh)
κ̃(d; m, s) = ,
hs
h∈N (d)
(h,m)=1
32
and
Y (1 − q −s−β1 )(1 − q −s−β2 )(1 − q −s−β3 )
λ(m, s) = .
1 − q −s
q|m
For notational simplicity we write
κ̃(d; m, 1 − βj ) = κ̃0j (d; m), λ(m, 1 − βj ) = λ0j (m).
Let
X κ̃0j (d1 ; drk)µ(k)k 1−βj
ξ0j (n; d, r) = λ̃0j (n, dr)
n=d k
ϕ(k)
1
(k,r)=1
with Y
λ̃0j (n, dr) = λ0j (q),
q|n
(q,dr)=1
For ψ(mod p) ∈ Ψ write

L(s + β1 , ψ)L(s + β2 , ψ)L(s + β3 , ψ)
C(s, ψ) = −i(pt0 )β3 Z(s, ψ)−1 . (7.1)
L(s, ψ)
Let a = {a(n)} denote a sequence of complex numbers satisfying
a(n) ≪ 1, a(n) = 0 if n ≥ P T −2. (7.2)
Write
ā = {ā(n)}, ā(n) = a(n),
X a(n)ψ(n)) X a(n)ψ̄(n)
A(a; s, ψ) = , A(a; 1 − s, ψ̄) = .
n
ns n
n1−s
Let J (z) denote the segment [s0 + z − iL1 , s0 + z + iL1 ].
The goal of this section is to prove
Proposition 7.1. Assume a1 and a2 satisfy (7.2). We have
X 1 Z
Θ1 (a1 , a2 ) : = C(s, ψ)A(a1 ; s, ψ)A(a2 ; 1 − s, ψ̄)ω(s) ds
ψ∈Ψ1
2πi J (1)

1 1 3
= S1 (a1 , a2 ) + 2S2 (a1 , a2 ) + S3 (a1 , a2 ) P
α 2 2
+ O(E(a1 , a2 )) + o(P),
where X
P= p,
p∼P
33
X X |µ(r)|λ0j (dr) X a1 (drm) X a2 (drn)ξ0j (n; d, r)
Sj (a1 , a2 ) = ,
d r
drϕ(r) m
m1−βj n
n
and X
E(a1 , a2 ) = PL2 |Sj (a1 , a2 )|.
1≤j≤3
In this and the next two sections we assume that 1 ≤ j ≤ 3.

Proof of Proposition 7.1: Initial steps
For notational simplicity we write Θ1 for Θ1 (a1 , a2 ) . Note that for any ψ ∈ Ψ, C(s, ψ)
is analytic if σ > 1 and |t − 2πt0 | ≤ L1 . We need to show that the sum over Ψ1 in the
expression for Θ1 can be extend to the sum over Ψ with an acceptable error. Namely we
prove
XZ
C(s, ψ)A(a1 ; s, ψ)A(a2 ; 1 − s, ψ̄)ω(s) ds = o(P). (7.3)
ψ∈Ψ2 J (1)
Here Proposition 2.1 is crucial.

Let κ(n) be given by
X κ(n) ζ(s + β1 )ζ(s + β2 )ζ(s + β3 )
= ,
n
ns ζ(s)
and regard a1 as an arithmetic function. For σ > 1

L(s + β1 , ψ)L(s + β2 , ψ)L(s + β3 , ψ) X (κ ∗ a1 )(m)ψ(m)
A(a1 ; s, ψ) = .
L(s, ψ) m
ms
Thus, for ψ(mod p) ∈ Ψ and s ∈ J (1),

X (κ ∗ a1 )(m)ψ(m)
C(s, ψ)A(a1 ; s, ψ) = −i(pt0 )β3 Z(s, ψ)−1 .
m
ms
The sum over m is split into two sums according to m < P 2 and m ≥ P 2 . To handle the
second one we appeal to the trivial bounds
X (κ ∗ a1 )(m)ψ(m)
s
≪ P 2(1−σ) Lc , A(a2 ; 1 − s, ψ̄) ≪ (P T −2)σ
2
m
m≥P
for σ ≥ 3/2, and

Z(s, ψ)−1 ≪ (pt0 )σ−1/2
for 3/2 ≤ σ ≤ L9 , |t − 2πt0 | ≤ L1 . Thus, moving the segment to J (L9 ) and applying the
simple estimate Z
|ω(s) ds| ≪ 1, |z| ≤ L9 , (7.4)
J (z)
34
we obtain
(κ ∗ a1 )(m)ψ(m)
Z X
−1
Z(s, ψ) s
A(a2 ; 1 − s, ψ̄)ω(s) ds ≪ ε.
J (1) 2
m
m≥P
To handle the sum over m < P 2 we move the path of integration to J (0). Hence
Z
˜

C(s, ψ)A(a1 ; s, ψ)A(a2 ; 1 − s, ψ̄)(s, ψ)ω(s) ds
J (1)
X
(κ ∗ a )(m)ψ(m)
Z
1
≤
s
|A(a2 ; 1 − s, ψ̄)ω(s) ds| + O(ε).
J (0)

2
m
m<P
By (7.4), the proof of (7.3) is reduced to showing that

X X (κ ∗ a1 )(m)ψ(m)
|A(a2 ; 1 − s, ψ̄)| = o(P), σ = 1/2. (7.5)
m s
ψ∈Ψ 2 2
m<P
Note that (κ ∗ a1 )(m) = O(τ5 (m)). The left side above is, by Cauchy’s inequality,
X X 2 1/2 1/4 X 1/4
(κ ∗ a1 )(m)ψ(m) X
4
≤
s
|A(a2 ; 1 − s, ψ̄)| 1

ψ∈Ψ m<P 2
m
ψ∈Ψ ψ∈Ψ2
X τ5 (m) 2
1/2 X τ2 (m) 2
1/4 X 1/4
≪ P2 P2 1 .
2
m 2
m ψ∈Ψ
m<P m<P 2
This yields (7.5) by Proposition 2.1 and (2.9).

By (7.3) we may write
X X∗
Θ1 = −i (pt0 )β3 I1 (ψ) + o(P) (7.6)
p∼P ψ( mod p)
where
(κ ∗ a1 )(m)ψ(m)
X
1
Z
−1
I1 (ψ) = Z(s, ψ) A(a2 ; 1 − s, ψ̄)ω(s) ds.
2πi J (1) m
ms
Assume ψ(mod p) ∈ Ψ. By (2.4), in the integral I1 (ψ), the factor Z(s, ψ)−1 can be
replaced by
τ (ψ̄)ps−1 ϑ∗ (1 − s),
and then, by a trivial bound for ϑ∗ (1 − s)ω(s) on σ = 3/2, the segment J (1) can be
replaced by the line σ = 3/2, with negligible errors. Thus, integration term-by–term
gives
τ (ψ̄) X X (κ ∗ a1 )(m)a2 (n)ψ(m)ψ̄(n)

m
I1 (ψ) = ∆1 + O(ε).
p m n n pn
35
This yields
1 X X (κ ∗ a1 )(m)a2 (n)

X∗ m X∗
I1 (ψ) = ∆1 τ (ψ̄)ψ(m)ψ̄(n) + O(ε).
p m n n pn
ψ( mod p) ψ( mod p)
Note that (n, p) = 1 if n < P T −2 . On substituting d = (m, n), m = dl, n = dk, we find
that the main term on the right side above is
1 X 1 X X (κ ∗ a1 )(dl)a2 (dk)
X
∗ l
τ (ψ̄)ψ(l)ψ̄(k) ∆1 .
p d k pk
d l (k,l)=1 ψ( mod p)
Since τ (ψp0 ) = −1, for (kl, p) = 1 we have

X∗ lk̄
τ (ψ̄)ψ(l)ψ̄(k) = pe + O(1)
p
ψ( mod p)
with k̄k ≡ 1(mod p). Hence, by Lemma 5.4,

X∗ X1 X X (κ ∗ a1 )(dl)b(dk) lk̄ l
I(ψ) = e ∆1 + O(P T −c). (7.7)
d
d k p pk
ψ( mod p) (l,p)=1 (k,l)=1
By trivial estimation, this remains valid if the constraint (l, p) = 1 is removed. Further,
by the relation
k̄ p̄ 1
≡− + (mod 1)
p k pk
we have
lk̄ l lp̄ l
e ∆1 =e − ∆
p pk k pk
Thus the right side of (7.7) is
X 1 X X (κ ∗ a1 )(dl)b(dk) lp̄ l
e − ∆ + O(P T −c). (7.8)
d k k pk
d l (k,l)=1
For (l, k) = 1 we have

−lp̄

µ(k) 1 X′
e = + τ (θ̄)θ(l)θ̄(−p).
k ϕ(k) ϕ(k)
θ( mod k)
Inserting this into (7.8) we deduce that

X∗
I(ψ) = T11 (p) + T12 (p) + O(P T −c) (7.9)
ψ( mod p)
36
where
X 1 X X (κ ∗ a1 )(dl)a2 (dk)µ(k) l
T11 (p) = ∆
d
d l
kϕ(k) pk
(k,l)=1
and
X 1 X X (κ ∗ a1 )(dl)a2 (dk) X′
l
T12 (p) = τ (θ̄)θ(l)θ̄(−p) ∆ .
d
d l
kϕ(k) pk
(k,l)=1 θ( mod k)
Note that (pt0 )β3 = −1 + O(α1 ). By (7.6) and (7.9), the proof of Proposition 7.1 is
reduced to showing that

p 1 3
iT11 (p) = S1 (a1 , a2 ) + 2S2 (a1 , a2 ) + S3 (a1 , a2 )
α 2 2
X (7.10)
2
+ O PL |Sj (a1 , a2 )| + o(P )
1≤j≤3
for p ∼ P , and X
pβ3 T12 (p) = o(P). (7.11)
p∼P
Proof of Proposition 7.1: The error term

In this subsection we prove (7.11).
Changing the order of summation gives
X X 1 X a2 (dk) X′ X
pβ3 T12 (p) = τ (θ̄) (κ ∗ a1 )(dl)θ(l)
p∼P d
d k kϕ(k) l
θ( mod k)
(7.12)
X
β3 l
× p θ̄(−p) ∆ .
p∼P
pk
If k = hr, r > 1 and θ(mod k) is induced by a primitive character θ∗ (mod r), then
|τ (θ̄)| ≤ r 1/2 . Thus, the inner sum over the non-principal θ(mod k) above is
X X∗
≪ r 1/2 |S(r, h, d; θ)|,
hr=k θ( mod r)
r>1
where
X X
β3 l
S(r, h, d; θ) = (κ ∗ a1 )(dl)θ(l) p θ̄(p)∆ .
p∼P
phr
(l,h)=1
Inserting this into (7.12) we obtain

X X′ 1 X∗
pβ3 T12 (p) ≪ √ |S(r, h, d; θ)|, (7.13)
p∼P
dhϕ(hr) r
{d,h,r} θ( mod r)
37
X′
where denotes a sum over the triples {d, h, r} satisfying dhr < P1 and r > 1.
{d,h,r}
By Lemma 5.3, for dhr < P1 , the terms in S(r, h, d; θ) with l > P 2 make a negligible
error. Also, by the Mellin transform and Lemma 5.4 (i),
Z ∞ X
X
β3 l c −1
1+it+β3
dt
p θ̄(p)∆ ≪ L hrl p θ̄(p)
1 + t2 (7.14)
p∼P
phr −∞

p∼P
for l ≤ P 2 .
Assume 1 < r < D and θ is a primitive character (mod r). By Lemma 5.6, the right
side of (7.14) is
hr 2 −c
≪ P D .
l
Hence
S(r, h, d; θ) ≪ τ5 (d)hrP 2D −c .
Thus, on the right side of (7.13), the total contribution from the terms with 1 < r < D is
O(P 2D −c ). The proof of (7.11) is therefore reduced to showing that
1 X X∗
|S(r, h, d; θ)| ≪ τ5 (d)hP 2 D −c (7.15)
R3/2 R≤r<2R θ( mod r)
for
dh < P1 and D ≤ R < (dh)−1 P1
which are henceforth assumed .
Let I(y) denote the interval
[(1/3)P t0 y, 4P t0 y].
By Lemma 5.1, for R ≤ r < 2R, the terms with l ∈ / I(Rh) in S(r, h, d; θ) make a negligible
contribution. Thus, it suffices to prove (7.15) with S(r, h, d; θ) replaced by

∗
X X
β3 l
S (R, r, h, d; θ) := (κ ∗ a1 )(dl)θ(l) θ̄(p)p ∆ .
p≃P
phr
l∈I(Rh)
(l,h)=1
By the Mellin transform and Lemma 5.4 (i),

Z ∞ X X
∗ c
−1−it
1+it+β
dt
S (R, r, h, d; θ) ≪ L
(κ ∗ a1 )(dl)θ(l)l
θ̄(p)p 3
1 + t2 .

−∞ l∈I(Rh) p≃P
(l,h)=1
38
For σ = 1, by the large sieve inequality we have
X X∗ X 2
(κ ∗ a1 )(dl)θ(l)l ≪ τ5 (d)2 Lc (R2 + P Rht0 )(P Rht0 )−1 ≪ τ5 (d)2 Lc
−s

R≤r<2R θ( mod r) l∈I(Rh)
(l,h)=1
and 2
X X∗ X
s ≪ (R2 + P )P 3.

θ̄(p)p
R≤r<2R θ( mod r) p≃P
It follows by Cauchy’s inequality that

1 X X∗
|S ∗ (R, r, h, d; θ)| ≪ τ5 (d)hLc (R1/2 P 3/2 + R−1/2 P 2 ) ≪ τ5 (d)hP 2D −c .
R3/2 R≤r<2R θ( mod r)
This yields (7.15).

Proof of Proposition 7.1: The main term
In this subsection we prove (7.10).
Assume p ∼ P . We may write
X 1 X a2 (dk)µ(k) X
l
T11 (p) = (κ ∗ a1 )(dl)∆ . (7.16)
d
d k
kϕ(k) pk
(l,k)=1
Since X X X
(κ ∗ a1 )(dl) = κ(m1 )a1 (m2 ) = κ(m1 )a1 (m2 ),
m1 m2 =dl d=d1 d2 m1 m2 =dl
(m1 ,d)=d1
it follows, by substituting m1 = d1 l1 , that

X X
(κ ∗ a1 )(dl) = κ(d1 l1 )a1 (d2 l2 ). (7.17)
d=d1 d2 l=l1 l2
(l1 ,d2 )=1
Hence,

X l X X X l1 l 2
(κ ∗ a1 )(dl)∆ = a1 (d2 l2 ) κ(d1 l1 )∆ .
pk d=d d
pk
(l,k)=1 1 2 (l2 ,k)=1 (l1 ,kd2 )=1

X X 1 X a2 (d1 d2 k)µ(k)
X X l1 l2
T11 (p) = a1 (d2 l2 ) κ(d1 l1 )∆ . (7.18)
d d
d1 d2 k kϕ(k) pk
1 2 (l2 ,k)=1 (l1 ,d2 k)=1
39
The innermost sum is, by the Mellin transform, equal to
X s
1 κ(d1 l) pk
Z
s
δ(s) ds (7.19)
2πi (3/2) l l2
(l,d2 k)=1
(here we have rewritten l for l1 ). In view of (7.2), we can assume that
d2 l2 < P T −2 , d1 d2 k < P T −2.
Every l with (l, d2 k) = 1 can be uniquely written as l = hr such that h ∈ N (d1),

(h, d2 k) = 1)and (r, d1 d2 k) = 1. Hence,
X κ(d1 l) ζ(s + β1 )ζ(s + β2 )ζ(s + β3 )
= κ̃(d 1 ; d 2 k, s)λ(d 1 d 2 k, s) ,
ls ζ(s)
(l,kd2 )=1
By the simple bounds

Y c Y c
|κ̃(d1 ; d2 k, s)| ≤ τ5 (d1 ) 1 + q σ ,
|λ(m, s)| ≤ 1 + q σ

q|d1 q|m
for σ > 9/10, we can move the contour of integration in (7.19) to the vertical segments
s = 1 + α + it with |t| ≥ D,
s = 1 − L−1 + it with |t| ≤ D
s = σ ± iD with 1 − L−1 ≤ σ ≤ 1 + α.
This yields, by Lemma 5.2 (i) and standard estimates, that the integral (7.18) is equal
to the sum of the residues of the integrand at s = 1 − βj , 1 ≤ j ≤ 3, plus an acceptable
error. Namely we have
1−βj
X ll2 X pk pkε1
κ(d1 l)∆ = Rj κ̃0j (d1 ; d2 k)λ0j (d1 d2 k) +O ,
pk 1≤j≤3
l2 l2
(l,kd2 )=1
where Rj is the residue of the function
ζ(s + β1 )ζ(s + β2 )ζ(s + β3 )

δ(s)
ζ(s)
at s = 1 − βj , and where
ε1 = exp{−cL1/10 }.
40
This yields

X X l1 l 2
a1 (d2 l2 ) κ(d1 l1 )∆
pk
(l2 ,k)=1 (l1 ,kd2 )=1
X X X a1 (d2 l)
= Rj (pk)1−βj κ̃0j (d1 ; d2 k)λ0j (d1 d2 k) + O(P kε1).
1≤j≤3 d=d1 d2
l1−βj
(l,k)=1
(here we have rewritten l for l2 ). Inserting this into (7.18) and rearranging the terms we
obtain X
T11 (p) = Rj p1−βj Sj∗ (a1 , a2 ) + O(P ε1) (7.20)
1≤j≤3
where
X X X a2 (d1 dk)κ̃0j (d1 , dk)λ0j (d1 dk)µ(k) X a1 (dl)
Sj∗ (a1 , a2 ) = βj 1−βj
.
d d k
d 1 dϕ(k)k l
1 (l,k)=1
By the Möbius inversion,

X X X a2 (d1 dk)κ̃0j (d1 ; dk)λ0j (d1 dk)µ(k)
Sj∗ (a1 , a2 ) =
d d1 k
d1 dϕ(k)k βj
X a1 (dl) X
× 1−βj
µ(r) .
l
l
r|(k,l)
This yields, by substituting k = rk1 , l = rm, n = d1 k1 and changing the order of

summation,
X X |µ(r)| X a1 (drm) X a2 (drn)λ0j (drn) X κ̃0j (d1 , drk1)µ(k1 )k 1−βj

Sj∗ (a1 , a2 ) = 1−β
1
.
r d
drϕ(r) m m j
n
n n=d k
ϕ(k1 )
1 1
(k1 ,r)=1
Since
λ0j (drn) = λ0j (dr)λ̃0j (n, dr),
it follows that
X κ̃0j (d1 , drk1 )µ(k1)k 1−βj
1
λ0j (drn) = λ0j (dr)ξ0j (n; d, r).
n=d k
ϕ(k 1 )
1 1
(k1 ,r)=1
Hence
Sj∗ (a1 , a2 ) = Sj (a1 , a2 ). (7.21)
41
On the other hand, by Lemma 5.2 (ii) and direct calculation we have
1
iR1 p−β1 = + O(L),
2α
2
iR2 p−β2 =
+ O(L),
α
3
iR3 p−β3 = + O(L).
2α
Combining these with (7.20) and (7.21) we obtain (7.10), and complete the proof of
Proposition 7.1. ✷
We first prove a general result as follows.
Lemma 8.1. For any a1 and a2 satisfying (7.2),
X X
C ∗ (ρ, ψ)A(a1 ; ρ, ψ)A(a2 , 1 − ρ, ψ̄)ω(ρ) = Θ1 (a1 , a2 ) + Θ1 (ā2 , ā1 ) + o(P).
ψ∈Ψ1 ρ∈Z̃(ψ)
Proof. Write
˜ ψ) = −i M(s + β1 , ψ)M(s + β2 , ψ)M(s + β3 , ψ) .

C(s,
M(s, ψ)
Assume ψ ∈ Ψ1 . By Proposition 2.2, we can choose a rectangle R with vertices at
s0 ± α + iL±
1
such that L± 1 = ±L1 + O(α) and such that the set of zeros of L(s, ψ) inside R is exactly
Z̃(ψ). Further, without loss of generality, we can assume that |s − ρ| ≫ α if s ∈ R and ρ
is a zero of L(s, ψ). By Lemma 5.9, the residue theorem and a simple bound for ω(s),
X
C ∗ (ρ, ψ)A(a1 ; ρ, ψ)A(a2 , 1 − ρ, ψ̄)ω(ρ)
ρ∈Z̃(ψ)
1 (8.1)
Z
= ˜ ψ)A(a1 ; s, ψ)A(a2 , 1 − s, ψ̄)ω(s) ds
C(s,
2πi R
= I˜1+ (a1 , a2 ; ψ) − I˜1− (a1 , a2 ; ψ) + O(ε)
where
1
Z
I˜1± (a1 , a2 ; ψ) = ˜ ψ)A(a1 ; s, ψ)A(a2 , 1 − s, ψ̄)ω(s) ds.
C(s,
2πi J (±α)
42
Write s′ = 1 − s̄. If
s = α + s0 + iv ∈ J (α).
then
s′ = −α + s0 + iv ∈ J (−α).
By (2.11) and analytic continuation, for s ∈ J (α) we have
˜ ψ) = −C(s
C(s, ˜ ′ , ψ)
and
A(a1 ; s, ψ)A(a2 , 1 − s, ψ̄)ω(s) = A(ā2 ; s′ , ψ)A(ā1 , 1 − s′ , ψ̄)ω(s′).
These together imply that
−I˜1− (a1 , a2 ; ψ) = I˜1+ (ā2 , ā1 ; ψ).
Hence, by (8.6),
C ∗ (ρ, ψ)A(a1 ; ρ, ψ)A(a2 , 1 − ρ, ψ̄)ω(ρ) = I˜1+ (a1 , a2 ; ψ) + I˜1+ (ā2 , ā1 ; ψ) + O(ε).
X
ρ∈Z̃(ψ)
Write
1
Z
I1+ (a1 , a2 ; ψ) = C(s, ψ)A(a1 ; s, ψ)A(a2 , 1 − s, ψ̄)ω(s) ds.
2πi J (α)
By Lemma 5.2 and 5.9,

Z
I˜1+ (a1 , a2 ; ψ)−I1+ (a1 , a2 ; ψ) ≪ L−114

L(s+β2 , ψ)L(s+β3 , ψ)A(a1 ; s, ψ)A(a2 , 1−s, ψ̄)ω(s) ds.

J (α)
By Cauchy’s inequality, Lemma 6.1, and the second assertion of Lemma 3.3, for s ∈ J (α),
X
|L(s + β2 , ψ)L(s + β3 , ψ)|2 ≪ P 2 L36 ,
ψ∈Ψ1
X
|A(a1 ; s, ψ)A(a2 , 1 − s, ψ̄)|2 ≪ P 2 L36 .
ψ∈Ψ1
These estimates together with (7.4) and (2.9) imply
I˜1+ (a1 , a2 ; ψ) − I1+ (a1 , a2 ; ψ) = o(P).

X
ψ∈Ψ1
On the other hand, moving the segment J (α) to J (1) gives

X
I1+ (a1 , a2 ; ψ) = Θ1 (a1 , a2 ) + O(ε).
ψ∈Ψ1
43
Hence
I˜1+ (a1 , a2 ; ψ) = Θ1 (a1 , a2 ) + o(P).
X
ψ∈Ψ1
For the sum of I˜1+ (ā2 , ā1 ; ψ) a similar result holds. This completes the proof. ✷
Recall that H1 (s, ψ) and H2 (s, ψ) are given by (2.27). Write
X X 2
C ∗ (ρ, ψ)H1 (ρ, ψ) + Z(ρ, χψ)H2 (s, ψ) ω(ρ).

Ξ1 =
ψ∈Ψ1 ρ∈Z(ψ)
Since |Z(s, χψ)| = 1 if σ = 1/2, it follows that
Ξ1 = Ξ11 + Ξ12 + 2ℜ{Ξ13 } (8.2)
where X X
Ξ11 = C ∗ (ρ, ψ)|H1(ρ, ψ)|2 ω(ρ), (8.3)
ψ∈Ψ1 ρ∈Z(ψ)
X X
Ξ12 = C ∗ (ρ, ψ)|H2(ρ, ψ)|2 ω(ρ), (8.4)
ψ∈Ψ1 ρ∈Z(ψ)
X X
Ξ13 = C ∗ (ρ, ψ)Z(ρ, χψ)−1 H1 (ρ, ψ)H2 (ρ, ψ)ω(ρ). (8.5)
ψ∈Ψ1 ρ∈Z(ψ)
We may write
X κj (n)χψ(n)
H1j (s, ψ) = (8.6)
n
ns
where (
(1 − log n/ log P1 )(P1 /n)β6 if n < P1 ,
κ1 (n) =
0 otherwise,
(
κ2 (n) =
0 otherwise,
(
κ3 (n) =
0 otherwise.
In this section we evaluate Ξ11 . By Lemma 8.1,
Ξ11 = 2ℜ{Θ1 (a11 , a21 )} + o(P) (8.7)
with
a11 (n) = χ(n) κ1 (n) + ι2 κ2 (n) , a21 (n) = a11 (n). (8.8)
44
By Proposition 7.1, our goal is reduced to evaluating the sum

X X |χ(d)||µχ(r)|λ0j (dr) X χ(m) κ1 (drm) + ι2 κ2 (drm)
Sj (a11 , a21 ) =
d r
drϕ(r) m
m1−βj

χ(n) κ̄1 (drn) + ῑ2 κ̄2 (drn) ξ0j (n; d, r)
X
× .
n
n
Write
β4 = β1 , β5 = β2 ,
so that
{βj , βj+1 , βj+2 } = {β1 , β2 , β3 }.
Lemma 8.2. Suppose T < x < P . Then for µ = 6, 7,

X χ(m) x βµ x
1−β
log = L′ (1, χ)Fjµ(x) + O(L−6 )
m<x
m j m m
where
Fjµ (x) = 1 + (βµ − βj ) log x xβµ .

Proof. The sum is equal to

1 xs ds
Z
L(1 − βj + s, χ) .
2πi (1) (s − βµ )2
We move the contour of integration to the vertical segments
s = α + it with |t| ≥ D,
s = −L−1 + it with |t| ≤ D
s = σ + ±iD, with − L−1 ≤ σ ≤ α.
It follows by Lemma 5.6 that

X χ(m) x βµ x 1
Z
xs ds
1−βj
log = L(1 − βj + s, χ) 2
+ O(ε1 )
m<x
m m m 2πi |s|=5α (s − βµ )
1 xs (s − βj )
Z
′
=L (1, χ) · ds + O(L−6).
2πi |s|=5α (s − βµ )2
The result now follows by direct calculation. ✷
45
The sum involving ξ0j (n; d, r) is more involved. We need the following lemma that
will be proved in Appendix A.
Lemma 8.3. Suppose dr 9/10. Further, if |s − 1| ≤ 5α, then
Uj (d, r; s) = Π(d, r) + O(L−8)
where
Y 1 Y 1 − q −1 − χ(q)q −1
Π(d, r) = .
1 − χ(q)q −1 1 − q −1
q|dr (q,r)=1
q|d
Lemma 8.4. Suppose dr < P T −2 and T < x < P . Then for µ = 6, 7,

X χ(n)ξ0j (n; d, r) x −βµ x
log = L′ (1, χ)Π(d, r)Gjµ(x) + O(L−6 )
n<x
n n n
where
βj+1 βj+2 (βj+1 − βµ )(βj+2 − βµ )

βj+1 βj+2
Gjµ (x) = 2
+ 1− 2
− log x x−βµ .
βµ βµ βµ
Proof. The sum is equal to

xs ds
Z X
1 χ(n)ξ0j (n; d, r)
1+s 2
. (8.9)
2πi (1) n
n (s + βµ )
The contour of integration is moved in the same way as in the proof of Lemma 8.2. Recall
that
X χ(n)ξ0j (n; d, r) L(1 + s + βj+1 , χ)L(1 + s + βj+2 , χ)
1+s
= Uj (d, r, 1 + s).
n
n L(1 + s, χ)
By Lemma 5.5 and 5.6, for |s| = 5α we have

L(1 + s + βj+1 , χ)L(1 + s + βj+2 , χ) (s + βj+1)(s + βj+2 )
= L′ (1, χ) + O(L−15 ).
L(1 + s, χ) s
46
Combining these results with Lemma 8.3, we find that the integral (8.9) is equal to
1 (s + βj+1 )(s + βj+2) xs ds

Z
′
L (1, χ)Π(d, r) · + O(L−6 ).
2πi |s|=5α s (s + βµ )2
The result now follows by direct calculation. ✷

By Lemma 8.2 with x = P1 /dr and x = P2 /dr respectively we have
X χ(m)κ1 (drm) L′ (1, χ)
= Fj6(P1 /dr) + O(L−15)
m
m1−βj log P1
if dr < P1 /T , and
X χ(m)κ2 (drm) L′ (1, χ)
= Fj7(P2 /dr) + O(L−15)
m
m1−βj log P2
if dr < P2 /T . On the other hand, by Lemma 8.3 with x = P1 /dr and x = P2 /dr
respectively we have
X χ(n)κ̄1 (drn)ξ0j (n; d, r) L′ (1, χ)Π(d, r)
= Gj6 (P1 /dr) + O(L−15 )
n
n log P1
if dr < P1 /T , and
= Gj7 (P2 /dr) + O(L−15 )
n
n log P2
if dr < P2 /T . Gathering these results together we conclude, by simple approximation,

that
X |µ(r)χ(dr)|
Sj (a11 , a21 ) =L′ (1, χ)2 λ0j (dr)Π(d, r)
dr<P2
drϕ(r)
Fj6(P1 /dr) Fj7 (P2 /dr) Gj6 (P1 /dr) Gj7 (P2 /dr)

× + ι2 + ῑ2
log P1 log P2 log P1 log P2
X |µ(r)χ(dr)| Fj6(P1 /dr)Gj6(P1 /dr)
+ L′ (1, χ)2 λ0j (dr)Π(d, r) 2
+ o(α).
P ≤dr<P
drϕ(r) (log P 1)
2 1
It can be shown, by verifying the case n = q k , that

X |µ(r)| Y Y
(1 − q −1 − χ(q)q −1 ) = (1 − χ(q)q −1 ),
n=dr
r
(q,r)=1 q|n
q|d
47
so that
X |µ(r)| n
Π(d, r) = . (8.10)
n=dr
ϕ(r) ϕ(n)
It follows, by substituting n = dr, that
X |χ(n)|λ0j (n)
Sj (a11 , a21 ) =L′ (1, χ)2
n<P
ϕ(n)
2
Fj6 (P1 /n) Fj7(P2 /n) Gj6 (P1 /n) Gj7 (P2 /n)

× + ι2 + ῑ2
log P1 log P2 log P1 log P2
X |χ(n)|λ0j (n) Fj6 (P1 /n)Gj6 (P1 /n)
+ L′ (1, χ)2 2
+ o(α).
P ≤n<P
ϕ(n) (log P 1 )
2 1
Since for n < P ,

Y ϕ(n)2
λ0j (n) = (1 − q −1 )2 (1 + O(α log q/q)) = + O(α1 )
n2
q|n
and, for x < P ,

X |χ(n)|ϕ(n) ϕ(D)
Y
6
Y
q

−2
= (1−q ) log x+O(log L) = 2 log x+O(log L),
n<x
n2 D π q+1
(q,D)=1 q|D
(see [T, 1.2.12]), it follows by partial integration that

Z P2
Fj6 (P1 /x) Fj7 (P2 /x) Gj6 (P1 /x) Gj7 (P2 /x) dx

Sj (a11 , a21 ) =a + ι2 + ῑ2
1 log P1 log P2 log P1 log P2 x
Z P1 (8.11)
a Fj6 (P1 /x)Gj6 (P1 /x)
+ dx + o(α).
(log P1 )2 P2 x
This yields, by the change of variable x → P1 /x or x → P2 /x,

Z P1 Z P2
a Fj6 (x)Gj6 (x) a|ι2 |2 Fj7(x)Gj7 (x)
Sj (a11 , a21 ) = 2
dx + 2
dx
(log P1 ) 1 x (log P2 ) 1 x
Z P2
aι2 Fj7(x)Gj6 (P 0.004 T 10 x)
+ dx (8.12)
(log P1 )(log P2 ) 1 x
Z P2
aῑ2 Fj6(P 0.004 T 10 x)Gj7 (x)
+ dx + o(α),
since P1 /P2 = P 0.004 T 10 . By direct calculation, for 0 ≤ z ≤ 1 we have
Fj6 (P z ) = fj6(z) + O(L−8), Gj6 (P z ) = gj6 (z) + O(L−8),
48
Fj7(P z ) = fj7 (z) + O(L−8), Gj7 (P z ) = gj7 (z) + O(L−8)
where

πi (3πi/2)z 8 5 πi
f16 (z) = 1 + z e , g16 (z) = + − − z e(−3πi/2)z , (8.13)
2 3 3 2

πi (3πi/2)z 4 1 πi
f26 (z) = 1 − z e , g26 (z) = + − + z e(−3πi/2)z , (8.14)
2 3 3 2

3πi (3πi/2)z 8 1 πi
f36 (z) = 1 − z e , g36 (z) = + + z e(−3πi/2)z . (8.15)
2 9 9 6

3πi (5πi/2)z 24 1 πi
f17 (z) = 1 + z e , g17 (z) = + + z e(−5πi/2)z , (8.16)
2 25 25 10

πi (5πi/2)z 12 13 3πi
f27 (z) = 1 + z e , g27 (z) = + + z e(−5πi/2)z , (8.17)
2 25 25 10

πi (5πi/2)z 8 17 3πi
f37 (z) = 1 − z e , g37 (z) = + − z e(−5πi/2)z . (8.18)
2 25 25 10
Substituting x = P z we obtain
Z P1 Z 0.504
1 Fj6 (x)Gj6 (x) 1
dx = fj6 (z)gj6 (z) dz + o(α),
(log P1 )2 1 x (0.504)2 log P 0
Z P2 Z 0.5
1 Fj7 (x)Gj7 (x) 1
dx = fj7(z)gj7 (z) dz + o(α),
(log P2 )2 1 x (0.5)2 log P 0
Z P2
1 Fj7(x)Gj6 (P 0.004 T 10 x)
dx
Z 0.5
1
= fj7(z)gj6 (z + 0.004) dz + o(α),
(0.5)(0.504) log P 0
Z P2
1 Fj6(P 0.004 T 10 x)Gj7 (x))
dx
Z 0.5
1
= fj6(z + 0.004)gj7(z) dz + o(α).
(0.5)(0.504) log P 0
Inserting these into (8.13) we obtain
1 2 3
S3 (a11 , a21 ) = a b11 + ι2 b21 + ῑ2 b12 + |ι2 |2 b22 + o(1)

S1 (a11 , a21 ) + S2 (a11 , a21 ) +
2α α 2α
where
0.504
1 1 3
Z
b11 = f16 (z)g16 (z) + 2f26 (z)g26 (z) + f36 (z)g36 (z) dz, (8.19)
0.5042 π 0 2 2
49
0.5
1 1 3
Z
0.52 π 0 2 2
1
b21 =
(0.504)(0.5)π
Z 0.5
1 3
× f17 (z)g16 (z + 0.004) + 2f27 (z)g26 (z + 0.004) + f37 (z)g36 (z + 0.004) dz.
0 2 2
(8.21)
1
b12 =
(0.504)(0.5)π
Z 0.5
1 3
× f16 (z + 0.004)g17 (z) + 2f26 (z + 0.004)g27 (z) + f36 (z + 0.004)g37 (z) dz.
0 2 2
(8.22)
It follows by Proposition 7.1 that
Θ1 (a11 , a21 ) = b11 + ι2 b21 + ι2 b12 + |ι2 |2 b22 aP + o(P).

This yields, by (8.7),

Ξ11 = c1 aP + o(P) (8.23)
where
c1 = c11 + ι2 c21 + ι2 c12 + |ι2 |2 c22
with
c11 = b11 + b11 ,
c22 = b22 + b22 ,
c12 = b12 + b21 ,
c21 = c12 .
Let ǫ be a complex number satisfying |ǫ| < 10−5 , not necessarily the same in each occur-
rence. Numerical calculation shows that
c11 = 3.61226 + ǫ/2,
c22 = 1.32215 + ǫ/2,

√
c12 = −0.45757 − 0.18179i + ǫ/ 2.
It follows that
c1 < 6.9955. (8.24)
50
In this section we evaluate Ξ12 . By Lemma 8.1,
Ξ12 = 2ℜ{Θ1 (a12 , a22 )} + o(P) (9.1)
with with
a12 (n) = χ(n) ῑ3 κ3 (n) + ῑ4 κ2 (n) , a22 (n) = a21 (n). (9.2)
Hence

X |χ(d)||µχ(r)|λ0j (dr) X χ(m) ῑ3 κ3 (drm) + ῑ4 κ2 (drm)
Sj (a12 , a22 ) =
d
drϕ(r) m
m1−βj

χ(n) ι3 κ̄3 (drn) + ι4 κ̄2 (drn) ξ0j (n; d, r)
X
× .
n
n
By Lemma 8.2 and 8.4, for dr < P3 /T ,

X χ(m)κ3 (drm) L′ (1, χ)
= Fj6 (P3 /dr) + O(L−6 )
m
m1−βj log P3
and
= Gj6 (P3 /dr) + O(L−6).
n
n log P3
Thus, in a way similar to the proof of (8.12), we deduce that
X Z P3 Fj6(P3 /x) Fj7 (P2 /x)

Gj6 (P3 /x) Gj7 (P2 /x) dx

Sj (a12 , a22 ) =a ῑ3 + ῑ4 ι3 + ι4
r<D 1
log P3 log P2 log P3 log P2 x
P2
|ι4 |2 a dx
Z
+ F j7 (P 2 /x)G j7 (P 2 /x) + o(α).
(log P2 )2 P3 x
It follows, by the change of variables x → P2 /x or x → P3 /x, that
Z P3 Z P2
a|ι3 |2 Fj6(x)Gj6 (x) a|ι4 |2 Fj7 (x)Gj7 (x)
Sj (a12 , a22 ) = 2
dx + 2
dx
(log P3 ) 1 x (log P2 ) 1 x
Z P3
aῑ3 ι4 Fj6 (x)Gj7 (P 0.002 T −10 x)
+ dx
Z P3
aῑ4 ι3 Fj7 (P 0.002 T −10 x)Gj6 (x)
+ dx + o(α),
since P2 /P3 = P 0.002 T −10 . This yields, by substituting x = P z ,
1 2 3
S1 (a12 , a22 )+ S2 (a12 , a22 )+ S3 (a12 , a22 ) = a |ι3 |2 b33 +ι3 ῑ4 b34 +ι4 ι¯3 b43 +|ι4 |2 b44 +o(1)

2α α 2α
51
where
0.498
1 1 3
Z
0.4982 π 0 2 2
b44 = b22 , (9.4)

0.498
1 1 3
Z
b34= f17 (z+0.002)g16 (z)+2f27 (z+0.002)g26 (z)+ f37 (z+0.002)g36 (z) dz,
(0.504)(0.498)π 0 2 2
(9.5)
Z 0.498
1 1 3
b43 = f16 (z)g17 (z+0.002)+2f26(z)g27 (z+0.002)+ f36 (z)g37 (z+0.002) dz.
(0.504)(0.498)π 0 2 2
(9.6)
It follows by Proposition 7.1 that
Θ1 (a12 , a22 ) = |ι3 |2 b33 + ι3 ῑ4 b34 + ι4 ῑ3 b43 + |ι4 |2 b44 aP + o(P)

This yields
Ξ12 = c2 aP + o(P) (9.7)
where
c2 = |ι3 |2 c33 + ι3 ῑ4 c34 + ι4 ῑ3 c43 + |ι4 |2 c44
with
c33 = b33 + b33 ,
c44 = c22 ,
c34 = b34 + b43 ,
c43 = c34 .
Numerical calculation shows that
c33 = 3.69507 + ǫ/2,

√
c34 = −0.4526 + 0.19474i + ǫ/ 2.
It follows that
c2 < 6.9955. (9.8)
52
The goal of this section is to prove Proposition 2.4. We continue to assume 1 ≤ j ≤ 3.
Note that f˜(z) ∈ R. If σ = 1/2, then
X χψ̄(n) log n
J1 (s, ψ) = 1−s
f˜ = J1 (1 − s, ψ̄),
n
n log P
X χψ̄(n) log n
J2 (s, ψ) = ˜
f + 0.004 − α̃ = J2 (1 − s, ψ̄).
n
n1−s log P
By Lemma 8.1 we have
X X
C ∗ (ρ, ψ)H1 (ρ, ψ)J1 (ρ, ψ)ω(ρ) = Θ1 (a11 , a13 ) + Θ1 (a13 , a21 ) + o(P)
ψ∈Ψ1 ρ∈Z(ψ)
with a11 (n) and a21 (n) given by (8.8), and with

˜ log n
a13 (n) = χ(n)f ,
log P
and
X X
C ∗ (ρ, ψ)H2 (ρ, ψ)J2 (ρ, ψ)ω(ρ) = Θ1 (a14 , a22 ) + Θ1 (a12 , a14 ) + o(P)
ψ∈Ψ1 ρ∈Z(ψ)
with a12 (n) and a22 (n) given by (9.2), and with

log n
a14 (n) = χ(n)f˜ + 0.004 − α̃ ,
log P
Hence
Ξ∗1 = Θ1 (a11 , a13 ) + Θ1 (a13 , a21 ) + Θ1 (a14 , a22 ) + Θ1 (a12 , a14 ) + o(P). (10.1)
Lemma 10.1. Write

X χ(m)f˜ log(ym)/ log P

V1j (y) = 1−βj
.
m
m
If 1 ≤ y ≤ P 0.5 /T , then
V1j (y) ≪ T −c ; (10.2)
if P 0.5 < y ≤ P 0.502 /T , then
500L′ (1, χ)
− 1 − βj log(y/P 0.5) + O(L−15);

V1j (y) = (10.3)
log P
53
if P 0.502 < y ≤ P 0.504 /T , then
500L′ (1, χ)
1 − βj log(P 0.504 /y) + O(L−15 );

V1j (y) = (10.4)
log P
if
y ∈ (P 0.5 /T, P 0.5 ] ∪ (P 0.502 /T, P 0.502 ] ∪ (P 0.504 /T, P 0.504 ),
then
V1j (y) ≪ L−7 . (10.5)
Proof. The inequalities (10.2) follow by the Polya-Vinogradov inequality and partial
summation. Write
P1′ = P 0.504 , P2′ = P 0.502 , P3′ = P 0.5.
Then
500 1 (P1′ )s − 2(P2′ )s + (P3′ )s ds
Z
f˜ log y/ log P =

· . (10.6)
log P 2πi (1) ys s2
In the case P 0.5 < y ≤ P 0.502 /T ,
500 1 (P1′ )s − 2(P2′ )s ds

Z
V1j (y) = · L(1 − βj + s, χ) . (10.7)
log P 2πi (1) ys s2
by (10.6). The contour of integration is moved in the same way as in the proof of lemma
8.1. Hence, by Lemma 5.8, the right side above is equal to
500L′ (1, χ) 1 (P1′ )s − 2(P2′ )s (s − βj )ds

Z
· + o(L−15 ).
log P 2πi |s|=5α ys s2
The yields (11.3) since the residue of the function
(P1′ )s − 2(P2′ )s s − βj
ys s2
at s = 0 is equal to −1 − βj log(y/P 0.5).

Similarly, in the case P 0.502 < y ≤ P 0.504 /T we have
′ s
500 1 P ds
Z
V1j (y) = · L(1 − βj + s, χ) 1 .
log P 2πi (1) y s2
This yields (10.4) in a way similar to the proof of (10.3).

In the case y ∈ (P 0.5/T, P 0.5] ∪ (P 0.504 /T, P 0.504 ) and m < T , we have
f˜ log(ym)/ log P ≪ α1 ,

54
so (10.5) follows by the Polya-Vinogradov inequality and partial summation; in the case
y ∈ (P 0.502 /T, P 0.502 ], we use (10.7) and a simple estimate for the integral on the new
segments to obtain (10.5). ✷
Lemma 10.2. Write
X χ(n)f˜ log(drn)/ log P ξ0j (n; d, r)

V2j (d, r) = .
n
n
If dr ≤ P 0.5 /T , then
L′ (1, χ)Π(d, r)
V2j (d, r) = βj+1 βj+2 log P + O(L−15 ); (10.8)
500
if P 0.5 < dr ≤ P 0.502 /T , then
500L′ (1, χ)Π(d, r)

− 1 + Y1j (dr) + O(L−15 )

V2j (d, r) = (10.9)
log P
with
P 0.504 0.502

0.5 βj+1βj+2 2 2 P
Y1j (y) = (βj+1 + βj+2) log(y/P )+ log − 2 log ;
2 y y
if P 0.502 < dr ≤ P 0.504 /T , then
500L′ (1, χ)Π(d, r)

1 + Y2j (dr) + O(L−15 )

V2j (d, r) = (10.10)
log P
with
βj+1 βj+2
Y2j (y) = (βj+1 + βj+2 ) log(P 0.504 /y) + log2 (P 0.504 /y);
2
if
dr ∈ (P 0.5 /T, P 0.5 ] ∪ (P 0.502 /T, P 0.502 ] ∪ (P 0.504 /T, P 0.504 ),
then
V2j (d, r) ≪ L−7 . (10.11)
Proof. First assume dr ≤ P 0.5 /T . In view of (8.9) and (10.6), we have
500 L(1 + βj+1 + s, χ)L(1 + βj+2 + s, χ)U0j (s; d, r)

Z
V2j (d, r) = ·
log P (1) L(1 + s, χ)
′ s ′ s ′ s
(P ) − 2(P2 ) + (P3 ) ds
× 1 .
(dr)s s2
55
The contour of integration is moved in the same way as the proof of Lemma 8.1. Thus,
by Lemma 5.8 and 8.2,
500L′ (1, χ)Π(d, r) (βj+1 + s)(βj+2 + s) (P1′ )s − 2(P2′ )s + (P3′ )s ds
Z
V2j (d, r) = · s 2
+ O(L−15 ).
log P |s|=10α s (dr) s
This yields (10.8) since the function

(P1′ )s − 2(P2′ )s + (P3′ )s
s2
is analytic and equal to (log P/500)2 at s = 0.
Similarly, in the case P 0.5 /T < dr ≤ P 0.502 we have
500L′ (1, χ)Π(d, r) 1 (βj+1 + s)(βj+2 + s) (P1′ )s − 2(P2′ )s ds
Z
V2j (d, r) = · s 2
+ O(L−15);
log P 2πi |s|=10α s (dr) s
in the case P 0.502 < dr ≤ P 0.504 /T we have

s
500L′ (1, χ)Π(d, r) 1 (βj+1 + s)(βj+2 + s) P1′ ds
Z
V2j (d, r) = · + O(L−15 ).
log P 2πi |s|=10α s dr s2
These yield (10.9) and (10.10). The proof of (10.11) is similar to that of (10.5) in the case
y ∈ (P 0.502 /T, P 0.502 ]. ✷
By direct calculation, for 0 ≤ z ≤ 1 we have
βj log P z−0.5 = πij(z − 0.5) + O(L−8 ), βj log P 0.504−z = πij(0.504 − z) + O(L−8),
Y1j (P z ) = y1j (z) + O(L−8 ), Y2j (P z ) = y2j (z) + O(L−8),

where
y11 (z) = 5πi(z − 0.5) − 3π 2 (0.504 − z)2 − 2(0.502 − z)2 ,

y21 (z) = 5πi(0.504 − z) − 3π 2 (0.504 − z)2 ,

3π 2
(0.504 − z)2 − 2(0.502 − z)2 ,

y12 (z) = 4πi(z − 0.5) −
2
3π 2
y22 (z) = 4πi(0.504 − z) − (0.504 − z)2 ,
2
y13 (z) = 3πi(z − 0.5) − π (0.504 − z)2 − 2(0.502 − z)2 ,
2

y23 (z) = 3πi(0.504 − z) − π 2 (0.504 − z)2 .

Remark. Similar to (10.6),
s
(P1′ )s − 2(P2′ )s + (P3′ )s ds

500 1 Dt0
Z
f˜ log y/ log P + 0.004 − α̃ =

· .
log P 2πi (1) P 0.004 ys s2
56
Thus, with simple modification, Lemma 10.1 and 10.2 apply to the sums
X χ(m)f˜ log(ym)/ log P + 0.004 − α̃

m
m1−βj
and
X χ(n)f˜ log(drn)/ log P + 0.004 − α̃ ξ0j (n; d, r)

.
n
n
Evaluation of Θ1 (a11 , a13 ).

We have

X X |χ(d)||µχ(r)|λ0j (dr) X χ(m) κ1 (drm) + ι2 κ2 (drm)
Sj (a11 , a13 ) =
r d
drϕ(r) m
m1−βj
χ(n)f˜ log(drn)/ log P ξ0j (n; d, r)
X
× .
n
n
The right side is split into three sums according to
dr < P 0.5 , P 0.5 ≤ dr < P 0.502 , P 0.502 ≤ dr < P 0.504 .
By Lemma 10.2 and the results in Section 8, the sum over dr < P 0.5 is equal to
L′ (1, χ)2 X |χ(n)|λ0j (n) Fj6 (P 0.504 /n) Fj7(P 0.5 /n)

βj+1 βj+2 + ι2 + o(α)
500 0.5
ϕ(n) 0.504 0.5
n<P
Z 0.5
aβj+1 βj+2 P Fj6 (P 0.504 /x) Fj7(P 0.5 /x) dx

= + ι2 + o(α)
500 1 0.504 0.5 x
aβj+1 βj+2 log P 0.5 fj6(0.004 + z)

fj7(z)
Z
= + ι2 dz + o(α).
500 0 0.504 0.5
Similarly, the sum over P 0.5 ≤ dr < P 0.502 is equal to
500L′ (1, χ)2 X |χ(n)|λ0j (n) 0.504

F j6 (P /n) − 1 + Y 1j (n) + o(α)
0.504 log2 P P 0.5 ≤n<P 0.502 ϕ(n)
Z 0.502
500a
= fj6(0.504 − z) − 1 + y1j (z) dz + o(α);
0.504 log P 0.5
the sum over P 0.502 ≤ dr < P 0.504 is equal to
500L′ (1, χ)2 X |χ(n)|λ0j (n)
Fj6 (P 0.504 /n) 1 + Y2j (n) + o(α)

2
0.504 log P P 0.502 ≤n<P 0.504 ϕ(n)
Z 0.504
500a
= fj6(0.504 − z) 1 + y2j (z) dz + o(α).
0.504 log P 0.502
57
Noting that
βj+1βj+2 log P = −(11 − 6j + j 2 )πα + o(α)
and gathering the above results together we conclude
1
Sj (a11 , a13 ) = d3j a + o(1)
α
with
(11 − 6j + j 2 )π 0.5

1 ι2
Z
d3j = − fj6 (0.004 + z) + fj7 (z) dz
500 0 0.504 0.5
Z 0.002
500
+ fj6 (z) − fj6(0.002 + z) dz
0.504π 0
Z 0.502 Z 0.504
500
+ fj6(0.504 − z)y1j (z) dz + fj6 (0.504 − z)y2j (z) dz .
0.504π 0.5 0.502
Hence, by Proposition 7.1,

1 3
Θ1 (a11 , a13 ) = d31 + 2d32 + d33 aP + o(P). (10.12)
2 2

We have
X X |χ(d)||µχ(r)|λ0j (dr) X χ(m)f˜ log(drm)/ log P

Sj (a13 , a21 ) =
r
drϕ(r) m
m1−βj
d

χ(n) κ̄1 (drn) + ῑ2 κ̄2 (drn) ξ0j (n; d, r)
X
× .
n
n
The right side is split into three sums in the same way as in the last subsection. By
Lemma 10.1 and the results in Section 8, the sum over dr < P 0.5 is o(α); the sum over
P 0.5 ≤ dr < P 0.502 is equal to
500L′ (1, χ)2 X |χ(n)|λ0j (n) 0.5
Gj6 (P 0.504 /dr) + o(α)

− 1 − βj log(y/P )
0.504 log2 P P 0.5 ≤n<P 0.502 ϕ(n)
Z 0.502
500a
= − 1 − πij(z − 0.5) gj6 (0.504 − z) dz + o(α);
0.504 log P 0.5
the sum over P 0.502 < dr ≤ P 0.504 is equal to
500L′ (1, χ)2 X |χ(n)|λ0j (n)
1 − βj log(P 0.504 /n) Gj6 (P 0.504 /n) + o(α)

2
0.504 log P P 0.502 ≤n<P 0.504 ϕ(n)
Z 0.504
500a
= 1 − πij(0.504 − z) gj6 (0.504 − z) dz + o(α),
0.504 log P 0.502
58
Gathering these results together we conclude
1
Sj (a13 , a21 ) = d4j a + o(1)
α
with Z 0.002
500
d4j = gj6 (z) − gj6 (0.002 + z) dz
0.504π 0
500ij 0.002
Z

− gj6 (0.002 + z)(0.002 − z) + gj6 (z)z dz.
0.504 0

1 3
Θ1 (a13 , a21 ) = d41 + 2d42 + d43 aP + o(P). (10.13)
2 2

We have
X X |χ(d)||µχ(r)|λ0j (dr) X χ(m)f˜ log(drm)/ log P + 0.004 − α̃

Sj (a14 , a22 ) =
r d
drϕ(r) m
m1−βj

X
× .
n
n
dr < P 0.496 , P 0.496 ≤ dr < P 0.498 , P 0.498 ≤ dr < P 0.5 .
By a result similar to Lemma 10.1 and the results in Section 8, the sum over dr < P 0.496
is o(α); the sum over P 0.496 ≤ dr < P 0.498 is equal to
500L′ (1, χ)2 X |χ(n)|λ0j (n) −0.496

− 1 − βj log(P n)
log2 P P 0.496 ≤n<P 0.498
ϕ(n)

ι3 0.498 ι4 0.5
× Gj6 (P /n) + Gj7 (P /n) + o(α)
0.498 0.5
500a 0.498
Z

= − 1 − πij(z − 0.496)
log P 0.496

ι3 ι4
× gj6 (0.498 − z) + gj7 (0.5 − z) dz + o(α);
0.498 0.5
59
500L′(1, χ)2 ι4 X |χ(n)|λ0j (n)
1 − βj log(P 0.5 /n) Gj7 (P 0.5/n) + o(α)

2
log P 0.5 0.498 ϕ(n)
P ≤n<P 0.5
1000ι4 a 0.5
Z

= 1 − πij(0.5 − z) gj7 (0.5 − z) dz + o(α).
log P 0.498
Gathering the above results together we conclude
1
Sj (a14 , a22 ) = (d′5j + d5j )a + o(1)
α
with Z 0.002
′ 500ι3
d5j = − gj6 (z) dz,
0.498π 0
1000ι4 0.002
Z

d5j = gj7 (z) − gj7 (0.002 + z) dz
π 0
Z 0.002
500ijι3
− (0.002 − z)gj6 (z) dz
0.498 0
Z 0.002

−1000ijι4 (0.002 − z)gj7 (0.002 + z) + zgj7 (z) dz.
0

1 ′ ′ 3 ′
Θ1 (a14 , a22 ) = (d + d51 ) + 2(d52 + d52 ) + (d53 + d53 ) aP + o(P). (10.14)
2 51 2
We have

X X |χ(d)||µχ(r)|λ0j (dr) X χ(m) ῑ3 κ3 (drm) + ῑ4 κ2 (drm)
Sj (a12 , a14 ) =
r
drϕ(r) m
m1−βj
d
χ(n)f˜ log(drn)/ log P + 0.004 − α̃ ξ0j (n; d, r)

X
× .
n
n
The right side is split into three sums in the same way as in the last subsection. By a
result similar to Lemma 10.2 and the results in Section 8, the sum over dr < P 0.496 is
equal to
L′ (1, χ)2 X |χ(n)|λ0j (n) ῑ3 Fj6 (P 0.498 /n) ῑ4 Fj7(P 0.5 /n)
βj+1 βj+2 + + o(α)
500 ϕ(n) 0.498 0.5
n<P 0.496
Z 0.496
aβj+1 βj+2 P ῑ3 Fj6(P 0.498 /x) ῑ4 Fj7(P 0.5 /x) dx

= + + o(α)
500 1 0.498 0.5 x
aβj+1 βj+2 log P 0.496 ῑ3 fj6(0.002 + z) ῑ4 fj7 (0.004 + z)
Z
= + dz + o(α);
500 0 0.498 0.5
60
500L′ (1, χ)2 |χ(n)|λ0j (n) ῑ3 Fj6(P 0.498 /n) ῑ4 Fj7 (P 0.5 /n)
X
+
log2 P p 0.496 ≤n<P 0.498
ϕ(n) 0.498 0.5
× − 1 + Y1j (P 0.004 n) + o(α)

500a 0.002 ῑ3 fj6(z) ῑ4 fj7(0.002 + z)

Z

= + − 1 + y1j (0.502 − z) dz + o(α);
log P 0 0.498 0.5
1000ι¯4 L′ (1, χ)2 X |χ(n)|λ0j (n)
Fj7 (P 0.5/n) 1 + Y2j (P 0.004 n) + o(α)

2
log P ϕ(n)
p0.498 ≤n<P 0.5
1000ι¯4 a 0.002
Z

= fj7(z) 1 + y2j (0.504 − z) dz + o(α).
log P 0
Gathering the above results together we conclude
1
Sj (a12 , a14 ) = (d′6j + d6j )a + o(1)
α
with Z 0.002
500ῑ3
d′6j =− fj6(z) dz, (10.15)
0.498π 0
(11 − 6j + j 2 )π 0.496 ῑ3 fj6(0.002 + z) ῑ4 fj7(0.004 + z)
Z
d6j = − + dz
500 0 0.498 0.5
1000ῑ4 0.002
Z

+ fj7 (z) − fj7(0.002 + z) dz
π 0
Z 0.002
500 ῑ3 fj6(z) ῑ4 fj7(0.002 + z)
+ + y1j (0.502 − z) dz
π 0 0.498 0.5
1000ῑ4 0.002
Z
+ fj7 (z)y2j (0.504 − z) dz.
π 0

1 ′ ′ 3 ′
Θ1 (a12 , a14 ) = (d + d61 ) + 2(d62 + d62 ) + d63 + d63 ) aP + o(P). (10.16)
2 61 2
It follows from (10.1) and (10.12)-(10.16) that
Ξ∗1 = (d′ + d)aP + o(P) (10.17)
where
1 ′ 3
d′ = d51 + d′61 + 2 d′52 + d′62 + d′53 + d′63 ,

2 2
61
1 3
d= d31 + d51 + d41 + d61 + 2 d32 + d52 + d42 + d62 + d33 + d53 + d43 + d63
2 2
For µ = 6, 7,
fjµ(0) = gjµ (0) = 1.
Hence
8ι3
d′ ≃ − .
0.498π
We only need a crude lower bound for the real part of d′ . By direct calculation we have
8
ℜ{d′ } > − ℜ{ι3 } − 0.04 > 5.1.
0.498π
The contribution from d is minor since if z is close to 1/2, then for µ = 1, 2,
yµj (z) ≃ 0.
Thus, by direct calculation we have the crude bound
|ℜ{d}| < 0.1.
Combining these bounds with (10.16) we complete the proof of Proposition 2.4. ✷
Remark. Numerical calculation actually shows that
ℜ{d} > 0.

By the result of Section 9,
Ξ12 ≪ aP.
Hence, by Cauchy’s inequality, the proof of Proposition 2.6 is reduced to showing that
X X
C ∗ (ρ, ψ)|J1 (ρ, ψ) − Z(ρ, χψ)J2 (1 − ρ, ψ̄)|2 ω(ρ) = o(aP). (11.1)
Let g̃1 (y) and g̃2 (y) be given by

Z 0.502 z Z 0.504 z
P P
g̃1 (y) = −500 g dz + 500 g dz,
0.5 y 0.502 y
Z 0.498 z Z 0.5 z
P Dt0 P Dt0
g̃2 (y) = −500 g dz + 500 g dz.
0.496 y 0.498 y
62
Write
X χψ(n)g̃µ (n)
J˜µ (s, ψ) = , µ = 1, 2,
n
ns
and
η± = exp{±L−10 }.
By (5) and (5), for σ = 1/2, the terms with n ≤ P 0.5 η− or n ≥ P 0.504 η+ in J˜1 (s, ψ)
contribute ≪ ε.
Lemma 11.1. If
y ∈ [P 0.5 η+ , P 0.502 η− ] ∪ [P 0.502 η+ , P 0.504 η− ],
then
log y
f˜ − g̃1 (y) ≪ ε; (11.2)
log P
if
y ∈ (P 0.5 η− , , P 0.5 η+ ) ∪ (P 0.502 η− , P 0.502 η+ ) ∪ (P 0.504 η− , P 0.504 η+ ),
then
˜ log y
f − g̃1 (y) ≪ L−10 . (11.3)
log P
Proof. Since
−L log x ∞
1 1
Z Z
2
g(1/x) = √ exp(−t ) dt = √ exp(−t2 ) dt,
π −∞ π L log x
it follows that
g(x) + g(1/x) = 1. (11.4)
Write
log y
u= .
log P
First assume T 0.502 η+ ≤ y ≤ P 0.503 . By (11.4),
Z 2u−0.502 z
P
g dz = u − 0.502;
0.502 y
by (6.2),
0.504
Pz
Z
g dz = 1.006 − 2u + O(ε);
2u−0.502 y
by (6.3),
0.502
Pz
Z
g dz ≪ ε.
0.5 y
63
These together imply (11.2).
Now assume P 0.503 ≤ y ≤ P 0.504 η− . By (11.4),
Z 0.504 z
P
g dz = 0.504 − u;
2u−0.504 y
by (6.2),
2u−0.504
Pz
Z
g dz = 2u − 1.006 + O(ε);
0.502 y
by (6.3),
0.502
Pz
Z
g dz ≪ ε.
0.5 y
These together imply (11.2). The proof of (11.2) with y ∈ [P 0.5 η+ , P 0.502 η− ] is similar.
We give the proof of (11.3) with P 0.502 η− < y < P 0.502 η+ only; the other cases are
similar. We have
Z 0.504 z Z 0.502 z
P 1 −10 P
g dz = + O(L ), g dz = O(L−10),
0.502 y 500 0.5 y
by (5.2) and (5.3), and
log y
f˜ = 1 + O(L−10). ✷
log P
It follows from Lemma 11.1 that
X χψ(n)
J1 (s, ψ) − J˜1 (s, ψ) = f˜(log n/ log P ) − g̃1 (n) + O(ε)

s
(11.5)
n∈I
n
1
for σ = 1/2, where
I1 = (P 0.5 η− , P 0.5 η+ ) ∪ (P 0.502 η− , P 0.502 η+ ) ∪ (P 0.504 η− , P 0.504 η+ ).
By (11.3), (8.25) and (8.26),
2
X
χψ(n)
f˜(log n/ log P ) − g̃1 (n) ω(ρ) = o(aP).
X X
∗

C (ρ, ψ) ρ
n∈I
n
ψ∈Ψ1 ρ∈Z̃(ψ) 1
Hence 2
C ∗ (ρ, ψ)J1 (ρ, ψ) − J˜1 (ρ, ψ) ω(ρ) = o(aP).
X X
Similarly, 2
C ∗ (ρ, ψ)J2 (ρ, ψ) − J˜2 (ρ, ψ) ω(ρ) = o(aP).
X X
64
The proof of (11.1) is therefore reduced to showing that
2
C ∗ (ρ, ψ)J˜1 (ρ, ψ) − Z(ρ, χψ)J˜2 (1 − ρ, ψ̄) ω(ρ) = o(aP).
X X
(11.6)
Lemma 11.2. Suppose σ = 1/2 and |t − 2πt0 | < L1 . Then
J˜1 (s, ψ) = Z(s, χψ)J˜2 (1 − s, ψ̄) + O(E2 (s, ψ))
where
L20

X χψ(n)
Z
−68
E(s, ψ) = L ω (iv) dv.
s+iv 1
−L 20

n<P
n
1
Proof. Suppose 0.5 ≤ z ≤ 0.504. In a way similar to the proof of Lemma 6.1, it can
be deduced that
X χψ(n) P z X χψ̄(n) P 1−z Dt0
s
g = L(s, χψ) − Z(s, χψ) 1−s
g + O(E2 (s, ψ)).
n
n n n
n n
Hence
Z 0.502 X
χψ(n) P z

g dz
0.5 n
ns n
0.5
χψ̄(n) P z Dt0
X
1
Z
= L(s, χψ) − Z(s, χψ) g dz + O(E2(s, ψ)),
500 0.498 n
n1−s n
and
0.504 X
χψ(n) P z
Z
g dz
0.502 n
ns n
0.498
χψ̄(n) P z Dt0
X
1
Z
= L(s, χψ) − Z(s, χψ) g dz + O(E2 (s, ψ)).
500 0.496 n
n1−s n

By Lemma 11.2, the proof of (11.6) is reduced to showing that
X X
C ∗ (ρ, ψ)E2 (ρ, ψ)2 ω(ρ) = o(aP).
This follows by (8.25), (8.26) and simple estimates.
65
The sum H11 (s, ψ) is split into
X X
H11 (s, ψ) = + = H14 (s, ψ) + H15 (s, ψ), say. (12.1)
n<P 1/2 P 1/2 ≤n<P 1
We can write
H1 (s, ψ)H2 (s, ψ) = B(s, ψ) + H15 (s, ψ)H2 (s, ψ)
with
B(s, ψ) = (H14 (s, ψ) + ι2 H12 (s, ψ))H2 (s, ψ). (12.2)
Accordingly we have
Ξ13 = Ξ14 + Ξ15 (12.3)
with X X
Ξ14 = C ∗ (ρ, ψ)Z(ρ, χψ)−1 B(ρ, ψ)ω(ρ), (12.4)
X X
Ξ15 = C ∗ (ρ, ψ)Z(ρ, χψ)−1 H15 (ρ, ψ)H2 (ρ, ψ)ω(ρ). (12.5)
The goal of this section is to evaluate Ξ15 .

Write β6 Z 0.504 z 0.5
1 P1 P P
κ12 (y) = g −g dz
0.504 y 0.5 y y
and
X κ12 (n)χψ(n)
H̃15 (s, ψ) = .
ns
P 0.5 η− <n<P1 η+
We first claim that

X X
C ∗ (ρ, ψ)|H15 (ρ, ψ) − H̃15 (ρ, ψ)|2 ω(ρ) = o(aP). (12.6)
By the argument in the last section, for P 0.5 η+ < n < P1 η− ,

Z 0.504 z
P log P1 − log n
g dz = + O(ε),
0.5 y log P
so that
κ12 (n) − κ1 (n) ≪ ε.
66
Also we have
(
L−10 if P1 η− ≤ n < P1 η+ ,
κ12 (n) − κ1 (n) ≪
1 if P 0.5 η− < n ≤ P 0.5 η+ .
These bounds together with (8.25) and (8.26) imply (12.6). We briefly describe the argu-
ment as follows. Let t(y) denote the characteristic function of the interval (P 0.5 η− , P 0.5η+ ].
If d ≤ P 0.5 (η+ − η− ), then
X t(dl)
≪ L−10 .
l
l
If d > P 0.5 (η+ − η− ), then there exists at most one l such that t(dl) = 1. Hence
X X
X 1 X t(dl) 1 t(dl)
≤ 2
≪ L−10 .
0.5
d l
l l
l d
d
d>P (η+ −η− )
Assume σ = 1/2, |t − 2πt0 | < L1 and 0.5 ≤ z ≤ 0.504. Using the proof of Lemma 11.2
with s + β6 in place of s we deduce that
Z(s + β6 , χψ)P1β6 X χψ̄(n)

Z 0.5 0.5 z
P Dt0 P Dt0
H̃15 (s, ψ) = g −g dz
0.504 n
n1−s−β6 0.496 n n
+ O(E(s + β6 , ψ)).
Write
P1′′ = P 0.496 Dt0 , P2′′ = P 0.5 Dt0 .
By (4) and (4), the terms with n ≤ P1′′ η− or n ≥ P2′′ η+ above contribute ≪ ε. If
P1′′ η+ < n < P2′′η− ,
then, in a way similar to the proof of (10.),

Z 0.5 0.5 z
P Dt0 P Dt0 log n
g −g dz = − α̃ − 0.496 + O(ε).
0.496 n n log P
Thus, in a way similar to the proof of Proposition 2.6, by the above discussion we deduce
that
X X
C ∗ (ρ, ψ)|Z(ρ, χψ)−1 H̃15 (ρ, ψ) − H̄16 (1 − s, ψ̄)|2 ω(ρ) = o(aP). (12.8)
where
1 X χψ̄(n) ′′ β6
′′

H̄16 (1 − s, ψ̄) = n/P 1 log n/P 1 .
log P1 n1−s
P1′′ <n<P2′′
67
By (12.7) and (12.8) we obtain
X X
Ξ15 = C ∗ (ρ, ψ)H̄16 (1 − ρ, ψ̄)H2 (ρ, ψ)ω(ρ) + o(P).
This implies, by Lemma 8.1,
Ξ15 = Θ1 (a12 , a25 ) + Θ1 (a15 , a22 ) + o(P) (12.9)
where
a15 (n) = χ(n)κ13 (n)
with (
(log P1 )−1 (n/P1′′)−β6 log(n/P1′′ ) if P1′′ < n < P2′′ ,
κ13 (n) =
0 otherwise,
a25 (n) = a15 (n).

Assume 1 ≤ j ≤ 3 in what follows.
Lemma 12.1. We have
(
X χ(l)κ13 (dl) T −c if d ≤ P1′′ /T,
≪
l
l1−βj α1 if P1′′ /T < d ≤ P1′′ ,
and
X χ(l)κ13 (dl) L′ (1, χ)
− 1 + (2β6 − βj ) log(d/P1′′ ) + ǫ1j (d) , |ǫ1j (d)| < 10−5 ,

=
l
l1−βj log P1
if P1′′ < d < P2 ,.

Proof. Note that κ13 (n) ≪ α1 if P1′′/T ≤ n ≤ P1′′T . In the case d ≤ P1′′ the results
follow by the Polya-Vinogradov inequality and partial summation. On the other hand, in
the case P1′′ < d < P2 the sum becomes
(d/P1′′)−β6 log(d/P1′′) X χ(l) (d/P1′′ )−β6 X χ(l) log l

1+β6 −βj
+ 1+β6 −βj
+ O(T −c )
log P1 ′′
l log P 1 ′′
l
l<P2 /d l<P2 /d
′
L (1, χ)(d/P1′′)−β6
− 1 + (β6 − βj ) log(d/P1′′) + O(L−15 ).

=
log P1
Since P2′′ /d > T 10 and (d/P1′′)−β6 = 1 − β6 log(d/P1′′ ) + ..., the result follows by Lemma
5.8 and a simple estimates. ✷
68
Lemma 12.2. If dr < P1′′ /T , then
X χ(l)κ̄13 (drl)ξj (l; d, r)
= b∗ L′ (1, χ)Π(d, r)(log P )βj+1βj+2 + O (12.10)
l
l
with
0.004
1
Z
∗
b = ze3πiz/2 dz.
0.504 0
If P1′′ /T ≤n≤ P1′′, then
≪ α1 ; (12.11)
l
l
Proof. If P1′′ < n < P2′′, then

1 ∂
κ̄13 (n) = − (n/P1′′ )β6 −w |w=0.
log P1 ∂w
Hence
1 ∂ X χ(l)ξj (l; d, r)
=− (dr/P1′′)β6 −w |w=0 .
l
l log P1 ∂w l1−β6 +w
P1′′ /dr<l<P2′′ /dr
Suppose dr ≤ P1′′ /T . By (4) and (4), for |w| = α we have

′′ ′′
X χ(l)ξj (l; d, r) X χ(l)ξj (l; d, r) P2 P1
1−β +w
= 1−β +w
g −g + O(L−15 )
l 6
l
l 6 drl drl
P1′′ /dr<l<P2′′ /dr
Z X
1 χ(l)ξj (l; d, r) ′′ ′′ s ω1 (s) ds
s
+ O(L−15 ).

= 1−β +w+s
(P 2 /dr) − (P 1 /dr)
2πi (1) l
l 6 s
In a way similar to the proof of Lemma 8.4, by lemma 8.2 and 5.8, we find that the right
side above is equal to
1 (s + w − β6 + βj+1 )(s + w − β6 + βj+2 ) (P2′′/dr)s − (P1′′ /dr)s
Z
′
L (1, χ)Π(d, r) · ds
2πi |s|=5α s + w − β6 s
+ O(L−15 )
(P2′′ /dr)β6−w − (P1′′ /dr)β6−w
=L′ (1, χ)Π(d, r)βj+1βj+2 + O(L−15).
β6 − w
It follows by Cauchy’ integral formula that
Z P 0.004
∂ ′′ β6 −w
X χ(l)ξj (l; d, r) ∂ β6 −w−1
(dr/P1 ) 1−β6 +w
|w=0 = y dy |w=0+O(L−6)
∂w l ∂w 1
′′ ′′
P1 /dr<l<P2 /dr
69
The main term on the right side is, by substituting y = P z , equal to
Z 0.004
2
−(log P ) ze3πiz/2 dz.
0
Gathering these results together we obtain (12.10). The proof of (12.11) is similar to that
of .
Lemma 12.3. If P1′′ < dr < P2 , then
X χ(l)κ̄13 (drl)ξj (l; d, r) L′ (1, χ)Π(d, r)
− 1 + (−2β6 + βj+1 + βj+2 ) log(dr/P1′′) + ǫ2j (dr) ,

=
l log P1
l
|ǫ2j (dr)| < 10−5 .
Proof. The left side is equal to

X χ(l)ξj (l; d, r)
1 ∂ ′′ β6 −w
− (dr/P1 ) |w=0 .
log P1 ∂w ′′
l1−β6 +w
l<P2 /dr
Assume |w| = α. In a way similar to the proof of Lemma 12.1, we deduce that
X χ(l)ξj (l; d, r)
1−β6 +w
= L′ (1, χ)Π(d, r)
l
l<P2′′ /dr
1 (s + w − β6 + βj+1 )(s + w − β6 + βj+2 ) (P2′′/dr)s

Z
×· ds + O(L−15).
2πi |s|=5α s + w − β6 s
By direct calculation,
1 (s + w − β6 + βj+1 )(s + w − β6 + βj+2) (P2′′ /dr)s
Z
ds
2πi |s|=5α s + w − β6 s
Z P ′′ /dr
= w − β6 + βj+1 + βj+2 + βj+1 βj+2 y β6−w−1 dy,
1
and the derivative of

Z P2′′ /dr
(dr/P1′′)β6 −w w − β6 + βj+1 + βj+2 + βj+1 βj+2 y β6 −w−1
dy
1
at w = 0 is equal to
Z P2′′ /dr
′′ β6 β6 −1
(dr/P1 ) 1 − βj+1 βj+2 y log y dy
1
Z P2′′ /dr
− (dr/P1′′)β6 log(dr/P1′′) − β6 + βj+1 + βj+2 + βj+1βj+2 y β6 −1
dy .
1
70
This can be written as the form
1 − (−2β6 + βj+1 + βj+2 ) log(dr/P1′′) − ǫ2j (dr),
since (dr/P1′′)β6 = 1 + β6 log(dr/P1′′) + .... ✷

We have

X X |χ(d)||µχ(r)|λ0j (dr) X χ(m) ῑ3 κ3 (drm) + ῑ4 κ2 (drm)
Sj (a12 , a25 ) =
r d
drϕ(r) m
m1−βj
X
χ(n)κ̄13 (n)ξ0j (n; d, r)
× .
n
n
The sum is split into three sums according to dr ≤ P1′′ /T , P1′′ /T < dr ≤ P1′′ and P1′′ <
dr < P2 . By lemma 8.2, 8.3 and 12.1,the sum over P1′′/T < dr ≤ P1′′ contributes o(α);
the sum over dr ≤ P1′′ /T is equal to
X |χ(n)|λ0j (n) ῑ3 Fj6(P 0.498 /n) ῑ4 Fj7 (P 0.5 /n)
′ 2 ∗
L (1, χ) b βj+1βj+2 + + o(α)
0.496
ϕ(n) 0.498 0.5
n<P
Z 0.496
ῑ3 fj6(0.498 − z) ῑ4 fj7(0.5 − z)

∗
=ab (log P )βj+1βj+2 + dz + o(α).
0 0.498 0.5
(12.12)
′′
By lemma 8.2, 8.3 and 12.3, the sum over P1 < dr < P2 is equal to
L′ (1, χ)2 ῑ3 X |χ(n)|λ0j (n)

Fj6 (P 0.498 /n)Wj (n)
(0.504)(0.498) log2 P P 0.496 <n<P 0.498
ϕ(n)
′ 2
L (1, χ) ῑ4 X |χ(n)|λ0j (n)
+ 2 Fj7 (P 0.5/n)Wj (n) + o(α)
(0.504)(0.5) log P P 0.496 <n<P 0.5 ϕ(n)
Z 0.498 (12.13)
aῑ3
= fj6(0.498 − z)Wj (P z ) dz
(0.504)(0.498) log P 0.496
Z 0.5
aῑ4
+ fj7 (0.5 − z)Wj (P z ) dz + o(α).
(0.504)(0.5) log P 0.496
where
Wj (n) = −1 + (−2β6 + βj+1 + βj+2 ) log(n/P1′′ ) + ǫ2j (n).
For 0 ≤ z ≤ 0.002, a good approximation to fj6(z) is
1 + πi(3 − j)z ≃ 1 + (2β6 − βj )(log P )z.
71
Since
4β6 − β1 − β2 − β3 ≃ 0,
it follows by simple calculation that
Z 0.498
fj6 (0.498 − z)Wj (P z ) dz = −0.002 + ǫ/10.
0.496
For 0 ≤ z ≤ 0.004, a good approximation to fj7(z) is
1 + πi(5 − j)z ≃ 1 + (2β7 − βj )(log P )z.
Thus, for 0.496 ≤ z ≤ 0.5, the function fj7 (0.5 − z)Wj (P z ) can be well approximated by
−1 − (2β7 − βj )(log P )(0.5 − z) − (2β6 − βj+1 − βj+2 )(log P )(z − 0.496).
Since
(2β6 + 2β7 − β1 − β2 − β3 ) log P ≃ 2πi,
it follows that
0.5
πi
Z
fj6 (0.5 − z)Wj (P z ) dz = −0.004 − + ǫ/10.
0.496 2502
inserting these results into (12.13), we find that the sum over P1′′ < dr < P2 is equal to

a 0.002ῑ3 2πiῑ4
− − 0.008ῑ4 − + ǫ/4 . (12.14)
0.504 log P 0.498 2502
It follows from (12.12) and (12.14) that

1 2 3
S1 (a12 , a25 ) + S2 (a12 , a25 ) + S3 (a12 , a25 ) = a(e∗1 + e∗2 + ǫ/2) (12.15)
2α α 2α
where
e∗1 = −πb∗ (3e∗11 + 6e∗12 + 3e∗13 )
with
0.496
ῑ3 fj6 (0.498 − z) ῑ4 fj7(0.5 − z)
Z
e∗1j = + dz,
0 0.498 0.5
and
4 0.002ῑ3 2πiῑ4
e∗2 = − − 0.008ῑ4 − .
0.504π 0.498 2502
72
We have
X X |χ(d)||µχ(r)|λ0j (dr) X χ(m)κ13 (m)
Sj (a15 , a22 ) =
r d
drϕ(r) m
m1−βj

X
× .
n
n
The sum is split into two sums according to dr ≤ P1′′ and P1′′ < dr < P2 . By lemma 8.2,
8.3 and 12.1, the first sum contributes o(α); the second sum is equal to
Z 0.496
∗

ab ῑ3 fj 6(0.498 − z) + ῑ4 fj 6(0.5 − z) dz + o(α).
0
By lemma 8.2, 8.3 and 12.3, the sum over P1′′ < dr < P2 is equal to
L′ (1, χ)2 ι3 X |χ(n)|λ0j (n)

Gj6 (P 0.498 /n)Wj∗ (n)
(0.504)(0.498) log2 P P 0.496 <n<P 0.498
ϕ(n)
′ 2
L (1, χ) ι4 X |χ(n)|λ0j (n)
+ 2 Gj7 (P 0.5 /n)Wj∗ (n) + o(α)
(0.504)(0.5) log P P 0.496 <n<P 0.5 ϕ(n)
Z 0.498
aι3
= gj6 (0.498 − z)Wj∗ (P z ) dz
(0.504)(0.498) log P 0.496
Z 0.5
aι4
+ gj7 (0.5 − z)Wj∗ (P z ) dz + o(α).
(0.504)(0.5) log P 0.496
where
W ∗ (n) = −1 + (2β6 − βj ) log(n/P1′′ ) + ǫ1j (n).
For 0 ≤ z ≤ 0.004 and µ = 6, 7, the function gjµ (z) is well-approximated by
1 − (2βµ − βj+1 − βj+2 )(log P )z.
Thus, in a way similar to the evaluation of Θ1 (a12 , a25 ), we deduce that

1 2 3
S1 (a15 , a23 ) + S2 (a15 , a22 ) + S3 (a15 , a22 ) = a(e∗2 + ǫ/2) (12.16).
2α α 2α
Finally, by (12.9), (12.15) and (12.16) we conclude
Ξ15 = (e∗1 + 2e∗2 + ǫ)aP. (12.17)
73
13. Approximation to Ξ14
In this section we establish an approximation to Ξ14 .
Assume that ψ ∈ Ψ1 and ρ ∈ Z(ψ). By Lemma 5.2 and (2.2),
L(ρ + β1 , ψ)L(ρ + β2 , ψ)L(ρ + β3 , ψ)
C ∗ (ρ, ψ) = −iZ(ρ + β3 , ψ)−1 (1 + O(1/t0))
L′ (ρ, ψ)
(13.1)
L(ρ + β1 , ψ)L(ρ + β2 , ψ)L(1 − ρ − β3 , ψ̄) −123
= −i (1 + O(L )).
L′ (ρ, ψ)
By Lemma 6.1,
L(1−ρ−β3 , ψ̄) = L(ρ + β3 , ψ) = K(1−ρ−β3 , ψ̄)+Z(ρ+β3 , ψ)−1 N(ρ+β3 , ψ)+O(E1(ρ+β3 , ψ)),
and, by Lemma 5.1,
Z(ρ + β3 , ψ)−1 = Z(ρ + β2 , ψ)−1 (pt0 )β1 + O(L−123 ).
Hence
L(ρ + β2 , ψ)L(1 − ρ − β3 , ψ̄)
=L(ρ + β2 , ψ)K(1 − ρ − β3 , ψ̄) + (pt0 )β1 N(ρ + β3 , ψ)L(1 − ρ − β2 , ψ̄) (13.2)
+ O(|L(ρ + β2 , ψ)N(ρ + β3 , ψ)L−123 ) + O(|L(ρ + β2 , ψ)|E1 (ρ + β3 , ψ)).
By Lemma 6.1 and 5.1,
(pt0 )β1 L(1 − ρ − β2 , ψ̄) =(pt0 )β1 K(1 − ρ − β2 , ψ̄) + (pt0 )β3 Z(ρ, ψ)−1 N(ρ + β2 , ψ)
+ O(N(ρ + β2 , ψ)L−123 ) + O(E1 (ρ + β2 , ψ)).
We insert this into (13.2) and then insert the result into (13.1). Thus we obtain
L(ρ + β1 , ψ)L(ρ + β2 , ψ)
C ∗ (ρ, ψ) = − i K(1 − ρ − β3 , ψ̄)
L′ (ρ, ψ)
L(ρ + β1 , ψ)
− i(pt0 )β1 N(ρ + β3 , ψ)K(1 − ρ − β2 , ψ̄)
L′ (ρ, ψ) (13.3)
L(ρ + β1 , ψ)
− i(pt0 )β3 Z(ρ, ψ)−1 N(ρ + β2 , ψ)N(ρ + β3 , ψ)
L′ (ρ, ψ)
+ O(E1∗ (ρ, ψ))
where

L(ρ + β1 , ψ)L(ρ + β2 , ψ)
E1∗ (ρ, ψ) |L(ρ + β3 , ψ)|t−1 −123

= 0 + |N(ρ + β3 , ψ)|L + E1 (ρ + β3 , ψ)

′
L (ρ, ψ)

L(ρ + β1 , ψ)N(ρ + β3 , ψ)
|N(ρ + β2 , ψ)|L−123 + E1 (ρ + β2 , ψ) .

+
′
L (ρ, ψ)
74
Recall that B(s, ψ) is given by (12.2). Multiplying (13.3) by Z(ρ, χψ)−1 B(ρ, ψ) and
applying Lemma 4.8 we obtain
C ∗ (ρ, ψ)Z(ρ, χψ)−1 B(ρ, ψ) = −i K1∗ (ρ, ψ)+(pt0 )β1 K2∗ (ρ, ψ)−(pt0 )β3 K3∗ (ρ, ψ) +O(E1∗ (ρ, ψ)|B(ρ, ψ)|)

where
L(ρ + β1 , ψ)L(ρ + β2 , ψ)
K1∗ (ρ, ψ) = Z(ρ, χψ)−1 B(ρ, ψ)K(1 − ρ − β3 , ψ̄), (13.4)
L′ (ρ, ψ)
L(ρ + β1 , ψ)
K2∗ (ρ, ψ) = Z(ρ, χψ)−1 B(ρ, ψ)N(ρ + β3 , ψ)K(1 − ρ − β2 , ψ̄), (13.5)
L′ (ρ, ψ)
L(ρ + β1 , ψ)
K3∗ (ρ, ψ) = B(ρ, ψ)G(ρ, ψ)N(ρ + β2 , ψ)N(ρ + β3 , ψ)F (1 − ρ, ψ̄), (13.6)
L′ (ρ, ψ)

Ξ14 = −i Φ1 + Φ2 − Φ3 + O(E) (13.7)
where X X
Φ1 = K1∗ (ρ, ψ)ω(ρ), (13.8)
ψ∈Ψ1 ρ∈Z(ψ)
X X
Φ2 = (pψ t0 )β1 K2∗ (ρ, ψ)ω(ρ), (13.9)
ψ∈Ψ1 ρ∈Z(ψ)
X X
Φ3 = (pψ t0 )β3 K3∗ (ρ, ψ)ω(ρ), (13.10)
ψ∈Ψ1 ρ∈Z(ψ)
and X X
E= E1∗ (ρ, ψ)|B(ρ, ψ)|ω(ρ).
ψ∈Ψ1 ρ∈Z(ψ)
Combining (2.34), Cauchy’s inequality, Proposition 7.1, Lemma 5.9, 6.1 and 3.3, we can
verify that
E = o(P). (13.11)
For example, by (2.34),

X X L(ρ + β1 , ψ)
2
L′ (ρ, ψ) |L(ρ + β2 , ψ)| ω(ρ)

ψ∈Ψ1 ρ∈Z(ψ)
Z
1 X M(s + β1 , ψ)
Z
=− − L(s + β2 , ψ)L(1 − s − β2 , ψ̄)ω(s) ds + O(ε),
2π J (α) J (−α) M(s, ψ)
ψ∈Ψ1
the right side being estimated via Lemma 5.9, 6.1 and 3.3.
75
14. Mean-value formula II
Recall that we always assume ψ is a primitive character (mod p), p ∼ P . Sometimes
we write pψ for the modulus p.
Let k∗ = {κ∗ (m)} and a∗ = {a∗ (n)} denote sequences of complex numbers satisfying
κ∗ (m) ≪ τ5 (m), (14.1)
a∗ (n) ≪ 1, a∗ (n) = 0 if n > 2P4 . (14.2)

For β ∈ C write
X (pψ t0 )β Z X ∗ X ∗
∗ ∗ −1 κ (m)ψ(m) a (n)ψ̄(n)
Θ2 (β, k , a ) = Z(s, χψ) s 1−s
ω(s) ds.
ψ∈Ψ
2πi J (1) m
m n
n
1
The goal of this section is to prove

Proposition 14.1. Suppose |β| < 5α. Then
1 X X 1 X µχ(k)a∗ (dk) X
l

∗ ∗ β ∗
Θ2 (β, k , a ) = (pt0 ) χ(l)κ (dl)∆ + o(P).
ϕ(D) p∼P d
d k
kϕ(k) Dpk
(l,k)=1
Proof. We prove this proposition with β = 0 only, as the general case is almost
identical. Write Θ2 for Θ2 (0, k∗ , a∗ ). Similar to the proof of Proposition 7.1, in the
expression for Θ2 , we can extend the sum over Ψ1 to the sum over Ψ, with acceptable
errors. Namely we have
I˜2 (ψ) + o(P)
X
Θ2 = (14.3)
ψ∈Ψ
where
κ∗ (m)ψ(m) a∗ (n)ψ̄(n)
X X
1
Z
I˜2 (ψ) = Z(s, χψ) −1
ω(s) ds.
2πi J (1) m
ms n
n1−s
Assume ψ( mod p) ∈ Ψ. We use (2.5) with θ = χψ and then replace the segment J (1) by
the vertical line σ = 3/2 with a negligible error. Thus, by integration term by term,
τ (χψ̄) X X κ∗ (m)a∗ (n)ψ(m)ψ̄(n) m

I˜2 (ψ) = ∆1 + O(ε).
Dp m n n Dpn
For (mn, p) = 1 we have

X mDn
τ (χψ̄)ψ(m)ψ̄(n) = τ (χ)χ(p)(p − 1)e .
p
ψ( mod p)
76
√
Note that (n, p) = 1 if n < 2P4 and |τ (χψp0 )| = D. Hence, substituting d = (m, n),
m = dl, n = dk and inverting the order of summation we obtain
τ (χ)χ(p) X X κ∗ (m)a∗ (n)

X∗
˜ m mDn
I2 (ψ) = ∆1 e + O(P T −c)
D m n
n Dpn p
ψ( mod p)
τ (χ)χ(p) X 1 X a∗ (dk) X ∗

l lDk
= κ (dl)∆1 e + O(P T −c).
D d
d k
k Dpk p
(l,k)=1
(here we have replaced the constraint (l, pk) = 1 by (l, k) = 1 with an acceptable error,
and used a bound for |∆1 (x)| = |∆(x)| given by Lemma 5.3). Since
Dk p̄ 1
≡− + (mod 1),
p Dk Dpk
it follows that
l lDk l lp̄
∆1 e =∆ e −
Dpk p Dpk Dk
with p̄p ≡ 1(mod Dk). Hence
X∗ τ (χ)χ(p) X 1 X a∗ (dk) X
l

l p̄

I˜2 (ψ) = ∗
κ (dl)∆ e −
D d k Dpk Dk (14.4)
ψ( mod p) d k (l,k)=1
+ O(P T −c).
Assume (k, p) = 1. If (l, k) = 1 and (l, D) = D1 , then (k, D1 ) = 1. Thus, substituting

(l, D) = D1 and l = D1 l1 , we find that the innermost sum in (14.4) is equal to

X X l1 l1 p̄
(κ1 ∗ b)(D1 dl1 )∆ e − .
D=D D
D2 pk D2 k
1 2 (l1 ,D2 k)=1
(D1 ,k)=1
Inserting this into (14.4) and rearranging the terms we conclude

X∗ τ (χ)χ(p) X
I˜2 (ψ) = S(D1 , D2 ; p) + O(P T −c),
D D=D D
ψ( mod p) 1 2
where
a∗ (dk)
X1
X X
∗ l lp̄
S(D1 , D2 ; p) = κ (D1 dl)∆ e −
d k D2 pk D2 k
d (k,D1 )=1 (l,D2 k)=1
77
(here we have rewritten l for l1 ). Since χ(−1)τ (χ)2 = D, the proof of Proposition 14.1 is
now reduced to showing that
X
χ(p)S(1, D; p)
p∼P
χ(−1)τ (χ) X 1 X µχ(k)a∗ (dk) X

∗ l (14.5)
= χ(l)κ (dl)∆
ϕ(D) d
d k kϕ(k) Dpk
(l,k)=1
2 1/2−c
+ O(P D )
and, for D = D1 D2 , D1 > 1,

X
χ(p)S(D1 , D2 ; p) ≪ P 2 D 1/2−c . (14.6)
p∼P
We first prove (14.5). Assume k < 2P4 and (l, Dk) = 1. Then

lp̄ 1 X
e − = τ (θ̄)θ(−l)θ̄(p).
Dk ϕ(Dk)
θ( mod Dk)
Hence
X 1 X a∗ (dk) X X
l

∗
S(1, D; p) = τ (θ̄)θ̄(p) κ (dl)θ(−l)∆ (14.7)
d
d k kϕ(Dk) l
Dpk
θ( mod Dk)
(here we have remove the constraint (l, Dk) = 1 as it is superfluous).

0
If θ is the principal character ψDk , then τ (θ̄) = µ(Dk). The total contribution from
0
the θ = ψDk term to the right side above is, by Lemma 5.3,
X 1 X X |(κ1 ∗ b)(dl)a∗ (dk)| l

≪ ∆ ≪ P Lc
d
d k l
ϕ(Dk)k Dpk
which is admissible for (14.5). Now, let θk1 be the character (mod Dk) induced by the
primitive character χ(mod D) which is real. Then
τ (θk1 )θk1 (p)θk1 (−l) = τ (χ)µχ(k)χ(−pl).
Hence
X 1 X X κ∗ (dl)a∗ (dk) l
∆ τ (θk1 )θk1 (p)θk1 (−l)
d
d k
ϕ(Dk)k Dpk
(l,Dk)=1
χ(−p)τ (χ) X 1 X µχ(k)a∗ (dk) X

∗ l
= χ(l)κ (dl)∆ .
ϕ(D) d
d k kϕ(k) Dpk
(l,k)=1
78
By the above discussion, to complete the proof of (14.5), it now suffices to show that the
0
total contribution from the terms with θ 6= ψDk and θ 6= θk1 in (14.7) to the left side of
(14.5) is O(P 2D −c ). Namely we need to prove
X 1 X |a∗ (dk)| X′

X ∗ X l
|τ (θ̄)|
κ (dl)θ(l) χθ̄(p)∆
d
d k
ϕ(Dk)k l p∼P
Dpk
θ( mod Dk) (14.8)
1
θ6=θk
2 −c
≪P D .
√
If θ( mod Dk) is induced by a primitive character θ∗ ( mod r), then r|Dk and |τ (θ̄)| ≤ r.
For r|Dk we have
Dk D
≡ 0 mod .
r (D, r)
Thus, substituting Dk = hr, we see that the left side of (14.8) is

X1 X X D X∗ X X l
∗
≪ √ κ (dl)θ(l) χθ̄(p)∆ .
d
d 1<r<2DP
ϕ(hr)h r
p∼P
phr
4 h<P/r θ( mod r) (l,h)=1
h≡0(D/(D,r)) θ6=χ
The range for r is divided into two parts according to 1 < r < D 3 and D 3 ≤ r < 2DP4 . In
a way similar to the proof of Proposition 7.1, for 1 < r < D 3 we use the Mellin transform,
Lemma 5.4 (i) and Lemma 5.6; for D 3 ≤ r < 2DP4 we use the Mellin transform, Lemma
5.4 (i) and the large sieve inequality. Thus we obtain (14.5).
The proof of (14.6) is analogous. In the case D = D1 D2 , D1 > 1, the major difference
from the proof of (14.5) is that the constraint (k, D1 ) = 1 implies D ∤ D2 k, so that any
non-principal character θ( mod D2 k) can not be induced by the real character χ( mod D).
Thus, when the left side of (14.6) is treated in much the same way as (14.5), the main
terms involved in the proof of (14.5) do not appear. This yields (14.6). ✷
Recall that Φ1 is given by (13.8). In view of (12.2), B(s, ψ) can be written as
X b(n)χψ(n)
B(s, ψ) = (15.1)
n
ns
with
b(n) ≪ τ2 (n), b(n) = 0 if n > P T −2 η+ . (15.2)
Write
L(s + β1 , ψ)L(s + β2 , ψ)
K1 (s, ψ) = Z(s, χψ)−1 B(s, ψ)K(1 − s − β3 , ψ̄).
L(s, ψ)
79
Similar to (8.1), for ψ ∈ Ψ1 we have
X
K1∗ (ρ, ψ)ω(ρ) = I2+ (ψ) − I2− (ψ) + O(ε)
ρ∈Z(ψ)
where
1
Z
I2± (ψ) = K1 (s, ψ)ω(s) ds.
2πi J (±α)
Hence X
I2+ (ψ) − I2− (ψ) + O(ε).

Φ1 = (15.3)
ψ∈Ψ1
First we prove that X

I2− (ψ) = o(P). (15.4)
ψ∈Ψ1
Assume ψ ∈ Ψ1 and s ∈ J (−α). By (2.2) and Lemma 5.1,

L(s + β1 , ψ)L(s + β2 , ψ) L(1 − s − β1 , ψ̄)L(1 − s − β2 , ψ̄)
= (pt0 )−β3 Z(s, ψ) (1+O(L−123)).
L(s, ψ) L(1 − s, ψ̄)
In a way similar to the proof of Lemma 8.1, it can be shown that the total contribution
of the O(L−123 ) term to the left side of (15.4) is o(P). On the other hand, by (2.4), (2.5)
and the relation
τ (χψ) = τ (χ)τ (ψ)ψ(D)χ(p),
we have
Z(s, ψ)
= τ (χ)χ(p)ψ̄(D)D s−1(1 + O(e−πt )).
Z(s, χψ)
For σ < 0 we can write
L(1 − s − β1 , ψ̄)L(1 − s − β2 , ψ̄)K(1 − s − β3 , ψ̄) X k̃(m)ψ̄(m)
=
L(1 − s, ψ̄) m
m1−s
with k̃(m) ≪ τ4 (m). Hence, moving the segment J (−α) to J (−1) with a negligible error,
we find that
X
−β3 1 k̃(m)ψ̄(Dm)
X X Z
−
I2 (ψ) = τ (χ) χ(pψ )(pψ t0 ) · B(s, ψ)ω(s) ds+o(P).
ψ∈Ψ ψ∈Ψ
2πi J (−1) m
(Dm)1−s
1 1
In a way similar to the proof of (7.3), the sum over ψ ∈ Ψ1 can be extended to the sum
over ψ ∈ Ψ, and then the segment J (−1) can be replaced by the line σ = −1/2, with
acceptable errors. Thus the right side above is equal to
X∗ 1 Z X
X
−β3 k̃(m)ψ̄(Dm)
τ (χ) χ(p)(pt0 ) B(s, ψ)ω(s) ds + o(P).
p∼P
2πi (−1/2) m
(Dm)1−s
ψ( mod p)
80
Since s
1 ω(s) ds 1 Dm 0 2 Dm
Z
2
= exp − L2 log
2πi (−1/2) ns (Dm)1−s Dm n n
and, for n < P T −5 and Dm < 2n,
(
X∗ p if n = Dm,
ψ(n)ψ̄(Dm) ≪
ψ( mod p)
1 6 Dm,
if n =
it follows that
X
1 k̃(m)ψ̄(Dm)
X∗ Z
B(s, ψ)ω(s) ds ≪ P D −4/5 .
2πi (−1/2) m
(Dm)1−s
ψ( mod p)
This yields (15.4).

Let κ1 (m) be given by
ζ(s + β1 )ζ(s + β2 ) X κ1 (m)

= s
, σ > 1.
ζ(s) m
m
Regarding b as an arithmetic function, for σ > 1 we have

L(s + β1 , ψ)L(s + β2 , ψ)B(s, ψ) X (κ1 ∗ b)(m)ψ(m)
= s
.
L(s, ψ) m
m
On the other hand, we can write

X g̃3 (n)ψ̄(n)
K(1 − s − β3 , ψ̄) =
n
n1−s
with
g̃3 (y) = y β3 g ∗ (P4 /y).
Hence, moving the segment J (α) to J (1) with a negligible error, we obtain
X
I2+ (ψ) = Θ2 (0, k∗1 , a∗1 ) + O(ε) (15.5)
ψ∈Ψ1
with
κ∗1 (m) = (κ1 ∗ b)(m), a∗1 (n) = g̃3 (n).
It follows by (15.3)-(15.5) and Proposition 14.1 that
X
Φ1 = Φ1 (p) + o(P) (15.6)
p∼P
81
where

1 X µχ(k) X g̃3 (dk) X l
Φ1 (p) = (κ1 ∗ b)(dl)χ(l)∆ . (15.7)
ϕ(D) k kϕ(k) d d Dpk
(l,k)=1
Assume dk < 2P4 . Similar to (7.17),

X X
(κ1 ∗ b)(dl) = b(d2 l2 )κ1 (d1 l1 ).
d=d1 d2 l=l1 l2
(l1 ,d2 )=1
This yields

X l X X X l1 l2
(κ1 ∗ b)(dl)χ(l)∆ = b(d2 l2 )χ(l2 ) κ1 (d1 l1 )χ(l1 )∆ .
Dpk d=d d
Dpk
(l,k)=1 1 2 (l2 ,k)=1 (l1 ,d2 k)=1
The innermost sum above is, by the Mellin transform, equal to

Z X s
1 κ1 (d1 l1 )χ(l1 ) Dpk
δ(s) ds. (15.8)
2πi (2) l1s l2
(l1 ,d2 k)=1
Every l1 can be uniquely written as l1 = hl such that h ∈ N (d1 ) and (l, d1 ) = 1, so that
κ1 (d1 l1 ) = κ1 (d1 h)κ1 (l). Hence, for σ > 1,
X κ1 (d1 l1 )χ(l1 ) X κ1 (l)χ(l)
=κ̃1 (d1 ; d2 k, s)
l1s ls
(l1 ,d2 k)=1 (l,d1 d2 k)=1
L(s + β1 , χ)L(s + β2 , χ)
=κ̃1 (d1 ; d2 k, s)λ1 (d1 d2 k, s)
L(s, χ)
where
X κ1 (d1 h)χ(h)
κ̃1 (d1 ; r, s) = (15.9)
hs
h∈N (d1 )
(h,r)=1
and
Y 1 − χ(q)q −s−β1 1 − χ(q)q −s−β2

λ1 (n, s) = . (15.10)
1 − χ(q)q −s
q|n
For notational simplicity we shall write κ̃1 (d; r) and λ1 (n) for κ̃1 (d; r, 1) and λ1 (n, 1)
respectively. Note that Dpk/l2 > T if l2 < P T −2. In a way similar to the treatment
of (7.19), by Lemma 5.5, we see that the expression (15.8) is equal to the residue of the
integrand at s = ρ̃ plus an acceptable error. Further, by (5.15), in the expression for this
residue, we can replace ρ̃ by 1 with an acceptable error. Thus we have
100
X l1 l2 ∗ Dpk α τ3 (d1 )Dpk
κ1 (d1 l1 )χ(l1 )∆ = R1 κ̃1 (d1 ; d2 k)λ1 (d1 d2 k) + O ,
Dpk l2 l2
(l1 ,d2 k)=1
82
where
L(1 + β1 , χ)L(1 + β2 , χ)
R∗1 = δ(1).
L′ (1, χ)
This yields

X l X X b(d2 l2 )χ(l2 )
(κ1 ∗ b)(dl)χ(l)∆ =R∗1 Dpk κ̃1 (d1 ; d2 k)λ1 (dk)
Dpk d=d1 d2
l2
(l,k)=1 (l2 ,k)=1
+ O α50 τ3 (d1 )Dpk .

Hence
X g̃3 (dk) X
l
(κ1 ∗ b)(dl)χ(l)∆
d
d Dpk
(l,k)=1
X X g̃3 (d1 d2 k) X b(d2 l2 )χ(l2 )
=R∗1 Dpk κ̃1 (d1 ; d2 k)λ1 (d1 d2 k)
d1 d2
d1 d2 l2
(l2 ,k)=1
+ O α30 Dpk

Inserting this into (15.7) and rewriting d, l and m for d2 , l2 and d1 respectively, we obtain
R∗1 Dp X X b(dl)χ(l)
Φ1 (p) = D1 (d, l) + o(P ) (15.11)
ϕ(D) d l dl
where
X µχ(k) X κ̃1 (m; dk)λ1 (mdk)g̃3 (mdk)
D1 (d, l) = .
ϕ(k) m m
(k,l)=1
On substituting n = mk we can write

X λ1 (dn)g̃3 (dn)
D1 (d, l) = ξ1 (n; d, l) (15.12)
n
n
with
X µχ(k)k
ξ1 (n; d, l) = κ̃1 (m; dk). (15.13)
n=mk
ϕ(k)
(k,l)=1
It can be verified, for given d and l, that ξ1 (n; d, l) is a multiplicative function of n. On

the other hand, we have
λ1 (dn) = λ1 (d)λ̃1 (n, d)
with Y
λ̃1 (n, d) = λ1 (q).
q|n
(q,d)=1
83
Hence
X λ̃1 (n, d)ξ1 (n; d, l)g̃3 (dn)
D1 (d, l) = λ1 (d) .
n
n
On the right side above, we can replace the factor g̃3 (dn) by (dn)β3 g(P4 /(dn)) with a
negligible error. Since
P4s+β3 ω1 (s + β3 )

P4 1
Z
β3
y g = ds,
y 2πi (1) y s s + β3
it follows that
λ̃1 (n, d)ξ1(n; d, l) P4s+β3 ω1 (s + β3 )

X
1
Z
D1 (d, l) = λ1 (d) · ds + O(ε). (15.14)
2πi (1) n
n1+s ds s + β3
Note that λ̃1 (q r , d) = λ̃1 (q, d) for any r. If σ > 1, then

X λ̃1 (n, d)ξ1 (n; d, l) Y X ξ1 (q r ; d, l)
= 1 + λ̃1 (q, d) .
n
ns q r
q rs
If (q, dl) = 1, then

X κ1 (qh)χ(h) χ(q)q
ξ1 (q; d, l) = − = q −β1 + q −β2 − 1 − χ(q) + O(1/q),
h q−1
h∈N (q)
so that
(1 − q −s−β1 )(1 − q −s−β2 ) X ξ1 (q r ; d, l)

1 + λ̃1 (q, d) = 1 + O(q −19/10 )
(1 − q −s )(1 − χ(q)q −s ) r
q rs
for σ > 9/10. In case (q, dl) > 1 and σ > 9/10, the left side above is trivially
1 + O(q −9/10 ).
It follows that the function

ζ(s)L(s, χ) X λ̃1 (n, d)ξ1 (n; d, l)
M1 (d, l; s) :=
ζ(s + β1 )ζ(s + β2 ) n ns
is analytic and it satisfies

Y c

M1 (d, l; s) ≪ 1+
q 9/10
q|dl
84
for σ > 9/10. The right side of (15.14) can be rewritten as
1 ζ(1 + s + β1 )ζ(1 + s + β2 )M1 (d, l; 1 + s) P4s+β3 ω1 (s + β3 )

Z
λ1 (d) · ds + O(ε).
2πi (1) ζ(1 + s)L(1 + s, χ) ds s + β3
Assume dl T . In a way similar to the proof
of Lemma 8.4, we deduce that
X
D1 (d, l) = λ1 (d) R1j dβj M1 (d, l; 1 − βj ) + O(ε1) (15.15)
j≤3
where R1j , 1 ≤ j ≤ 3, is the residue of the function
ζ(1 + s + β1 )ζ(1 + s + β2 ) P4s+β3 ω1 (s + β3 )

(15.16)
ζ(1 + s)L(1 + s, χ) s + β3
at s = −βj (the function (15.16) also has a simple pole at s = ρ̃ − 1, while the residue
at this point can be regarded as an acceptable error). Inserting this into (15.11) and
substituting n = dl we obtain
R∗1 Dp X
Φ1 (p) = R1j S1j + o(p) (15.17)
ϕ(D) j≤3
where
X b(n) X
S1j = λ1 (d)dβj χ(l)M1 (d, l; 1 − βj ).
n
n
n=dl
If (q, dl) = 1, then λ̃1 (q, d) = λ̃1 (q, 1) and ξ1 (q r ; d, l) = ξ1 (q r ; 1, 1) for any r. Hence
M1 (d, l; s) Y
X ξ1 (q r ; d, l) X ξ1 (q r ; 1, 1) −1
= 1 + λ̃1 (q, d) 1 + λ̃1 (q, 1) . (15.18)
M1 (1, 1; s) r
q rs r
q rs
q|dl
We can rewrite
X b(n)̟1j (n)
S1j = M1 (1, 1; 1 − βj ) (15.19)
n
n
with
X M1 (d, l; 1 − βj )
̟1j (n) = λ1 (d)dβj χ(l) .
n=dl
M1 (1, 1; 1 − βj )
In view of (15.18), we see that ̟1j (n) is a multiplicative function.
Let Y
Q= q.
q<D 4
85
Every n can be uniquely written as n = n1 n2 with n1 ∈ N (Q) and (n2 , Q) = 1. Hence
X b(n)̟1j (n) X ̟1j (n1 ) X b(n1 n)̟1j (n)
= . (15.20)
n
n n1 n
n1 ∈N (Q) (n,Q)=1
Using the the Rankin trick (see [14, Section 13.2], for example), we can impose the
constraint n1 < T to the right side with an acceptable error O(ε1 ). In addition, by
(15.18), for n < P T −3 , (n, Q) = 1 and dl = n,
M1 (d, l; s)
λ1 (d) = 1 + O(D −c ),
M1 (1, 1; s)
so that
̟1j (n) = χ(n)̺∗j (n) + O(τ2 (n)D −c ) (15.21)
where X
̺∗j (n) = dβj χ(d).
d|n
The following lemma will be proved in Appendix B.

Lemma 15.1. Suppose n1 ∈ N (Q) and n1 < T . Then
X b(n1 n)χ(n)̺∗j (n)
= χ(n1 )τ2 (n1 )ej + O α1 τ2 (n1 )
n
(n,Q)=1
where
ej = (e1j + ι2 e2j )(ῑ3 e3j + ῑ4 e2j )
with
2j 8j 8j
e2j = 1 − + exp{5πi/4} − ,
5 25πi 25πi

2j j j
e3j = 1− + exp{0.747πi} −
3 1.1205πi 1.1205πi
and
e1j = e′1j − e′′1j ,

′ 2j j j
e1j = 1 − + exp{0.756πi} − ,
3 1.134πi 1.134πi
Z 0.004
j
e′′1j

= exp{(3/2)(0.504 − z)πi} − exp{(3/4)πi} dz.
0.756 0
86
By (15.19)-(15.21) and Lemma 15.1 we obtain
X χ(n)τ2 (n)̟1j (n)
S1j = ej M1 (1, 1; 1 − βj ) + O(1/L). (15.22)
n
n∈N (Q)
n<T
This remains valid if the constraint n ∈ N (Q) on the right side is removed, since, for
D4 < q < T ,
̺∗j (q) = ̺j (q) + O(ν(q)) ≪ α1 + ν(q).
This yields, by (15.21),
X χ(n)τ2 (n)̟1j (n)
≪ 1/L.
n
n∈N
/ (Q)
n<T
To apply (15.22) we need two lemmas which will be proved in Appendix A.

Lemma 15.2. If |s − 1| < 5α, then
Y 1 − χ(q)q −2
M1 (1, 1; s) = + O(α1).
1 − q −2
(q,D)=1
Lemma 15.3. For σ ≥ 9/10 the function
1 X χ(n)τ2 (n)̟1j (n)

U1j (s) :=
ζ(s)2L(s, χ)2 n ns
is analytic and bounded. Further we have
ϕ(D)2 Y (1 − q −2 )2
U2j (1) = + O(α1 ).
D2 1 − χ(q)q −2
(q,D)=1
By (4.2) and (4.3),

X χ(n)τ2 (n)̟1j (n) X χ(n)τ2 (n)̟1j (n) T
= g + O(α1 )
n<T
n n
n n
1 T s ω1 (s) ds
Z
= ζ(1 + s)2 L(1 + s, χ)2 U2j (1 + s) + O(α1 ).
2πi (1) s
By Lemma 15.3, we can move the contour of integration in the same way as in the proof
of Lemma 8.4 to obtain
X χ(n)τ2 (n)̟1j (n)
= L′ (1, χ)2 U2j (1) + O(α1).
n<T
n
87
This together with Lemma 15.2 and 15.3 yields
X χ(n)τ2 (n)̟1j (n) ϕ(D)
M1 (1, 1; 1 − βj ) =a + O(1/L3),
n D
n∈N (Q)
n<T
since
ϕ(D) Y 6 Y q
(1 − q −2 ) = 2 .
D π q+1
(q,D)=1 q|D
It follows by (15.22) that

ϕ(D)
S1j = ej a + O(1/L3) (15.23)
D
By Lemma 5.8,
R∗1 = β1 β2 L′ (1, χ) + O(1/L24 )
and
β −β
P4 3 j
R1j = ′
+ O(1/L3).
(βj+1 − βj )(βj+2 − βj )L (1, χ)
Hence, by direct calculation,
R∗1 R11 = 1 + O(1/L),
R∗1 R12 = 2 + O(1/L),
R∗1 R13 = 1 + O(1/L).
Combining these relations with (15.23) , (15.17) and (15.6) we conclude

Φ1 = e1 + 2e2 + e3 aP + o(P) (15.24)
Recall that Φ2 is given by (13.9). Write
L(s + β1 , ψ)
K2 (s, ψ) = Z(s, χψ)−1 B(s, ψ)N(s + β3 , ψ)K(1 − s − β2 , ψ̄).
L(s, ψ)
Similar to (15.3), X
(pψ t0 )β1 I3+ (ψ) − I3− (ψ) + O(ε)

Φ2 =
ψ∈Ψ1
where
1
Z
I3± (ψ) = K2 (s, ψ)ω(s) ds.
2πi J (±α)
88
Note that the length of the sum B(s, ψ)N(s + β3 , ψ) is ≤ P T −1. For ψ ∈ Ψ1 , moving the
segment J (−α) to J (−L9 ) we obtain, by simple estimation,
I3− (ψ) ≪ ε.
Hence X
Φ2 = (pψ t0 )β1 I3+ (ψ) + O(ε).
ψ∈Ψ1
In a way similar to the proof of (15.6), we can move the segment J (α) to J (1) and then
extend the sum over Ψ1 to that over Ψ with an acceptable error. Hence
X X∗ 1 Z
β1
Φ2 = (pt0 ) K2 (s, ψ)ω(s) ds + o(P). (16.1)
p∼P
2πi J (1)
ψ( mod p)
Let κ2 (n) and b1 (n) be given by

X κ2 (n) ζ(s + β1 )
= , σ > 1,
n
ns ζ(s)
X b1 (n)ψ(n)
= B(s, ψ)N(s + β3 , ψ).
n
ns
Note that
X g̃2 (n)ψ̄(n)
K(1 − s − β2 , ψ̄) =
n
n1−s
with
g̃2 (y) = y β2 g ∗ (P4 /y).
We can rewrite (16.1) as
Φ2 = Θ2 (β1 , k∗2 , a∗2 ) + o(p)
with κ∗2 = κ2 ∗ b1 and a∗2 = g̃2 . It follows by Proposition 14.1 that
X
Φ2 = (pt0 )β1 Φ2 (p) + o(P) (16.2)
p∼P
where

1 X µχ(k) X g̃2 (dk) X l
Φ2 (p) = (κ2 ∗ b1 )(dl)χ(l)∆ . (16.3)
ϕ(D) k kϕ(k) d d Dpk
(l,k)=1
Similar to (7.17),

X l
(κ2 ∗ b1 )(dl)χ(l)∆
Dpk
(l,k)=1

X X X l1 l 2
= b1 (d2 l2 )χ(l2 ) κ2 (d1 l1 )χ(l1 )∆ .
d=d1 d2 (l2 ,k)=1
Dpk
(l1 ,d2 k)=1
89
The innermost sum is, by the Mellin transform, equal to
Z X s
1 κ2 (d1 l1 )χ(l1 ) Dpk
δ(s) ds. (16.4)
2πi (2) l1s l2
(l1 ,d2 k)=1
For σ > 1,
X κ2 (d1 l1 )χ(l1 ) L(s + β1 , χ)
s
= κ̃2 (d1 ; d2 k, s)λ2 (d1 d2 k, s)
l1 L(s, χ)
(l1 ,d2 k)=1
where
X κ2 (d1 h)χ(h)
κ̃2 (d1 ; r, s) =
hs
h∈N (d1 )
(h,r)=1
and
Y 1 − χ(q)q −s−β1
λ2 (n, s) = .
1 − χ(q)q −s
q|n
For notational simplicity we shall write κ̃2 (m; r) and λ2 (n) for κ̃2 (m; r, 1) and λ2 (n, 1)
respectively. Note that Dpk/l2 > T if l2 < P22 . In a way similar to the treatment of
(7.19), we see that the expression (16.4) is equal to the residue of the integrand at s = ρ̃
plus an acceptable error. Further, by (5.15), we can replace ρ̃ by 1 in the expression for
this residue, with an acceptable error. Thus we have
100
X l1 l2 ∗ Dpk α τ2 (d1 )Dpk
κ2 (d1 l1 )χ(l1 )∆ = R2 κ̃2 (d1 ; d2 k)λ2 (d1 d2 k) + O
Dpk l2 l2
(l1 ,d2 k)=1
where
L(1 + β1 , χ)
R∗2 = δ(1).
L′ (1, χ)
This yields

X l X X b1 (d2 l2 )χ(l2 )
(κ2 ∗ b1 )(dl)χ(l)∆ =R∗2 Dpk κ̃2 (d1 ; d2 k)λ2 (dk)
Dpk d=d1 d2
l2
(l,k)=1 (l2 ,k)=1
+ O α50 τ3 (d)Dpk .

Hence
X g̃2 (dk) X
l
(κ2 ∗ b1 )(dl)χ(l)∆
d Dpk
d (l,k)=1
X X g̃2 (d1 d2 k) X b1 (d2 l2 )χ(l2 )
=R∗2 Dpk κ̃2 (d1 ; d2 k)λ2 (d1 d2 k)
d d
d1 d2 l2
1 2 (l2 ,k)=1
20
+ O(α Dpk).
90
Inserting this into (16.3) and rewriting d, l and m for d2 , l2 and d1 respectively, we obtain
R∗2 Dp X X b1 (dl)χ(l)
Φ2 (p) = D2 (d, l) + o(1) (16.5)
ϕ(D) d l dl
where
X µχ(k) X κ̃2 (m; dk)λ2 (mdk)g̃2 (mdk)
D2 (d, l) = .
ϕ(k) m m
(k,l)=1
On substituting n = mk we can write

X λ2 (dn)g̃2 (dn)
D2 (d, l) = ξ2 (n; d, l) (16.6)
n
n
with
X µχ(k)k
ξ2 (n; d, l) = κ̃2 (m; dk). (16.7)
n=mk
ϕ(k)
(k,l)=1
It can be verified, for given d and l, that ξ2 (n; d, l) is a multiplicative function of n, and
X λ̃2 (n, d)ξ2 (n; d, l)g̃2(dn)
D2 (d, l) = λ2 (d)
n
n
with Y
λ̃2 (n, d) = λ2 (q). (16.8)
q|n
(q,d)=1
Hence, similar to (15.14),
λ̃2 (n, d)ξ2 (n; d, l) P4s+β2 ω1 (s + β2 )

X
1
Z
D2 (d, l) = λ2 (d) · ds + O(ε). (16.9)
2πi (1) n
n1+s ds s + β2
It can be verified, for σ > 9/10, that the function
ζ(s)L(s, χ) X λ̃2 (n, d)ξ2 (n; d, l)

M2 (d, l; s) :=
ζ(s + β1 ) n ns
is analytic and it satisfies

Y c

M2 (d, l; s) ≪ 1+ .
q 9/10
q|dl
91
Thus we can rewrite D2 (d, l) as
1 ζ(1 + s + β1 )M2 (d, l; 1 + s) P4s+β2 ω1 (s + β2 )

Z
λ2 (d) · ds + O(ε).
2πi (1) ζ(1 + s)L(1 + s, χ) ds s + β2
Assume dl < P22 , and (dl, D) = 1. Note that P4 /d > T . In a way similar to the proof of
(15.15), we deduce that
X
D2 (d, l) = λ2 (d) R2j dβj M2 (d, l; 1 − βj ) + O(ε1) (16.10)
j=1,2
where R2j , j = 1, 2, is the residue of the function
ζ(1 + s + β1 ) P4s+β2 ω1 (s + β2 )
(16.11)
ζ(1 + s)L(1 + s, χ) s + β2
at s = −βj . Inserting this into (16,5) and substituting n = dl we obtain
R∗2 Dp X
Φ2 (p) = R2j S2j + o(p) (16.12)
ϕ(D) j=1,2
where
X b1 (n) X
S2j = λ2 (d)dβj χ(l)M2 (d, l; 1 − βj ).
n
n n=dl
Let M∗2 (s) be given by
M∗2 (s) = M2 (1, 1; s) if χ(2) 6= 1,
1 − q −s−β1 X ξ2 (q r ; 1, 1)
Y
∗
M2 (s) = 2 −s )(1 − χ(q)q −s )
1 + λ̃2 (q, 1) rs
if χ(2) = 1.
q>2
(1 − q r
q
The following result will be proved in Appendix A.

Lemma 16.1. Suppose dl 2
1 − χ(q)q −1 q−1
92
As a direct consequence of Lemma 16.1 we have |M∗2 (s)| ≫ 1 if |s − 1| < 5α. Thus
we can write
X b1 (n)̟2j (n)
S2j = M∗2 (1 − βj ) (16.13)
n
n
with
X M2 (d, l; 1 − βj )
̟2j (n) = λ2 (d)dβj χ(l) .
M∗2 (1 − βj )
n=dl
Note that ̟2j (n) is multiplicative. In a way similar to the proof of (15.20) we have
X b1 (n)̟2j (n) X ̟2j (n1 ) X b1 (n1 n)̟2j (n)
= + O(D −c).
n
n n1 n
n1 ∈N (Q) (n,Q)=1
n1 <T
Assume n1 ∈ N (Q) and n1 < T . For (n, Q) = 1 we have

2
−β3 ∗ T
X X
b1 (n1 n) = (l1 l) g b(m1 m).
n =l m n=lm
l1 l
1 1 1
Hence
X b1 (n1 n)̟2j (n) X X ̟2j (l) T 2 X b(m1 m)̟2j (m)
= β
g∗ + O(D −c ),
n n =l m
l(l1 l) 3 l1 l m
(n,Q)=1 1 1 1 (l,Q)=1 (m,Q)=1
since the terms with (l, m) > 1 above contribute ≪ D −c . Further, for (m, Q) = 1 and
m < P we have trivially
̟2j (m) = χ(m)̺∗j (m) + O τ2 (m)D −c .

Hence, for m1 < T ,

X b(m1 m)̟2j (m)
= ej χ(m1 )τ2 (m1 ) + O(D −c)
m
(m,Q)=1
by Lemma 15.1. It follows that

X b1 (n1 n)̟2j (n) X X ̟2j (l) T 2
= ej χ(m1 )τ2 (m1 ) β3
g∗ + O(D −c ). (16.14)
n n =l m
l(l 1 l) l 1 l
(n,Q)=1 1 1 1 (l,Q)=1
On the other hand, if 1 < l < T 5 and (l, Q) = 1, then for any q|l,
̺2j (l) ≪ α1 + O(ν(q)).
93
Thus the innermost sum in (16.14) is equal to 1 + O(1/L7). It follows that
X b1 (n1 n)̟2j (n) X
= ej χ(m1 )τ2 (m1 ) + O(1/L4).
n
(n,Q)=1 m1 |n1
Since 1 ∗ χτ2 = ν ∗ χ (here 1 denotes the arithmetic function identically equal to 1), it
follows that
X b1 (n)̟2j (n) X ̟2j (n1 )(ν ∗ χ)(n1 )
= ej + O(1/L). (16.15)
n
n n1
n1 ∈N (Q)
n1 <T
On the right side above, the constraint n1 ∈ N (Q) can be removed with an acceptable
error.
The following lemma will be proved in Appendix A.
Lemma 16.2. The function
1 X ̟2j (n)(ν ∗ χ)(n)
U2j (s) =
ζ(s)3L(s, χ)3 n ns
is analytic and bounded for σ > 9/10. Further we have
6 ϕ(D) Y q
U2j (1) = + O(1/L4).
π 2 Dp q+1
q|D
We have
X ̟2j (n)(ν ∗ χ)(n) X ̟2j (n)(ν ∗ χ)(n) T
= g + O(1/L10)
n<T
n n
n n
1 ds
Z
= ζ(1 + s)3 L(1 + s, χ)3 U2j (1 + s)T s ω1 (s) + O(1/L10).
2πi (1) s
The contour of integration is moved in the same way as in the proof of Lemma 8.4. Thus
the right side above is, by Lemma 16.2, equal to
a ϕ(D) ′
L′ (1, χ)3 U2j (1) + O(1/L4) = L (1, χ) + O(1/L4).
p D
Hence, by (16.15),
X b1 (n)̟2j (n) aej ϕ(D) ′
= L (1, χ) + O(1/L4).
n
n p D
94
Inserting this into (16.13) and applying Lemma 16.1 we obtain
ϕ(D) ′
S2j = aej L (1, χ) + O(1/L4). (16.16)
D
On the other hand, by Lemma 5.8 and direct calculation,
1
R∗2 = β1 + O(1/L10), R2j = − + O(L6), j = 1, 2,
β1 L′ (1, χ)
so that
1
R∗2 R2j = − + O(L−1 L′ (1, χ)−1 ).
L′ (1, χ)
This together with (16.16) and (16.12) yields
Φ2 (p) = −(e1 + e2 )ap + o(p).
Since (pt0 )β1 = −1 + O(α1 ), it follows by (16.2) that
Φ2 = (e1 + e2 )aP + o(P). (16.17)
Recall that Φ3 is given by (13.10). Write
L(s + β1 , ψ)
K3 (s, ψ) = B(s, ψ)G(s, ψ)N(s + β2 , ψ)N(s + β3 , ψ)F (1 − s, ψ̄)
L(s, ψ)
and
1
Z
I4± (ψ) = K3 (s, ψ)ω(s) ds.
2πi J (±α)
Similar to the treatment of Φ2 , we have

X
(pψ t0 )β3 I4+ (ψ) − I4− (ψ) + O(ε).

Φ3 = (17.1)
ψ∈Ψ1
To treat the sum of I4+ (ψ) we move the segment J (α) to J (1), and then extend the
sum over Ψ1 to the sum over Ψ with an acceptable error. Hence
X X
(pψ t0 )β3 I4+ (ψ) = (pt0 )β3 Φ+
3 (p) + o(P) (17.2)
ψ∈Ψ1 p∼P
where
1
X∗ Z
Φ+
3 (p) = K3 (s, ψ) ω(s) ds.
ψ mod p
2πi J (1)
95
For σ > 1 we can write
L(s + β1 , ψ) X ν ∗ (m)ψ(m)
B(s, ψ)G(s, ψ)N(s + β2 , ψ)N(s + β3 , ψ) = .
L(s, ψ) m
ms
Thus, replacing the segment J (1) by the line σ = 3/2 and integrating term by term give
X X ν ∗ (m)ν(n) n s0 X∗
+ 2 2 n
Φ3 (p) = ψ(m)ψ̄(n) exp − L2 log + O(ε).
m 4
n m ψ mod p
m
n<D
By trivial estimation, the contribution from the terms with m 6= n above is o(p). Hence
X ν ∗ (n)ν(n)
Φ+
3 (p) = p + o(p). (17.3)
4
n
n<D
Note that for n < D 4

X
ν ∗ (n) =

κ2 (n1 )χ(n2 n3 ) κ1 (n2 ) + ι2 κ2 (n2 ) ῑ3 κ3 (n3 ) + ι¯4 κ4 (n3 ) υ(n4 )g̃2 (n5 )g̃3 (n6 ).
n=n1 ...n6
A detailed analysis shows that (17.2) remains valid if, in the expression for ν ∗ (n), the
factor κ2 (n1 )is replaced by (1 ∗ µ)(n1 ), the factor

χ(n2 n3 ) κ1 (n2 ) + ι2 κ2 (n2 ) ῑ3 κ3 (n3 ) + ι¯4 κ4 (n3 )
is replaced by e0 χ(n2 n3 ) with

e0 = κ1 (1) + ι2 κ2 (1) ῑ3 κ3 (1) + ι¯4 κ4 (1) , (17.4)
and the factors g̃2 (n5 ) and g̃3 (n6 ) are replaced by 1 respectively. Since
1 ∗ µ ∗ χ ∗ χ ∗ υ ∗ 1 ∗ 1 = ν,
it follows that
X ν(n)2
Φ+
3 (p) = e0 p + o(p). (17.5)
4
n
n<D
The following lemma will be proved in Appendix B.

Lemma 17.1. We have
X ν(n)2
= a + o(1)
4
n
n<D
β3
Since (pt0 ) = −1 + O(α1), it follows by (17.5), (17.2) and Lemma 17.1 that
X
(pψ t0 )β3 I3+ (ψ) = −e0 aP + o(P). (17.6)
ψ∈Ψ1
96
To treat the sum involving I4− (ψ) on J (−α) we first apply Lemma 5.1 to obtain
L(s + β1 , ψ) L(1 − s − β1 , ψ̄)

= (pt0 )−β1 (1 + O(α6))
L(s, ψ) L(1 − s, ψ̄)
for ψ ∈ Ψ1 and s ∈ J (−α). Thus, in a way similar to the proof of (15.4),

1
X X Z
β3 −
(pψ t0 ) I3 (ψ) = β2
(pψ t0 ) · K3∗ (s, ψ)ω(s) ds + o(P)
ψ∈Ψ ψ∈Ψ
2πi J (−α)
1 1
where
L(1 − s − β1 , ψ̄)
K3∗ (ψ) = F (1 − s, ψ̄)B(s, ψ)G(s, ψ)N(s + β2 , ψ)N(s + β3 , ψ).
L1 − s, ψ̄)
Moving the segment J (−α) to J (−1) and then extend the sum over Ψ1 to then sum over
Ψ we obtain X X
(pψ t0 )β3 I3− (ψ) = (pt0 )β2 Φ−
3 (p) + o(P) (17.7)
ψ∈Ψ1 p∼P
where
1
X∗ Z
Φ−
3 (p) = K3∗ (s, ψ) ω(s) ds.
ψ mod p
2πi J (−1)
Write
X ν ∗ (l)ψ(l)
1
G(s, ψ)N(s + β2 , ψ)N(s + β3 , ψ) = ,
l
ls
so that
X (b ∗ ν ∗ )(n)ψ(n)
1
B(s, ψ)G(s, ψ)N(s + β2 , ψ)N(s + β3 , ψ) = .
n
ns
On the other hand, for σ < 0,
L(1 − s − β1 , ψ̄) X κ̄2 (m)ψ̄(m)

=
L(1 − s, ψ̄) m
m1−s
with κ̄2 (m) = κ2 (m), so that
L(1 − s − β1 , ψ̄) X ̺∗ (n)ψ̄(n)

F (1 − s, ψ̄) =
L(1 − s, ψ̄) n
n1−s
with X
̺∗ (n) = ν(l)κ̄2 (m).
n=lm
l<D 4
97
Note that the arithmetic function (b ∗ ν1∗ )(n) is supported on n < P T −2. In a way similar
to the proof of (17.3) we deduce that
X (b ∗ ν ∗ )(n)̺∗ (n)
Φ−
3 (p) = p
1
+ o(p). (17.8)
n
n
We have
X (b ∗ ν ∗ )(n)̺∗ (n) X ν(l) X (b ∗ ν ∗ )(lm)κ̄2 (m)
1 1
= .
n
n 4
l m m
l<D
The inner sum is equal to

XX X b(l1 m1 )ν1∗ (l2 m2 )κ̄2 (m1 m2 )
.
l=l1 l2 m1 (m2 ,l1 )=1
m1 m2
Note that ν1∗ (n) is supported on n < T 5 . On the right side above, we can drop the terms
with m2 > 1 or (m1 , Q) > 1 with an acceptable error. Hence
X (b ∗ ν ∗ )(n)̺∗ (n) X ν(l) X X b(l1 m1 )ν1∗ (l2 )κ̄2 (m1 )
1
= + o(1).
n
n l m1
4
l<D l=l1 l2 (m1 ,Q)=1
If m1 is square-free, then
κ̄2 (m1 ) = µ(m1 )̺1 (m1 ).
Hence, by Lemma 15.1 (see (15.)),
X b(l1 m1 )κ̄2 (m1 )
= e1 χ(l1 )τ2 (l1 ) + O
m1
(m1 ,Q)=1
It follows that
X (b ∗ ν ∗ )(n)̺∗ (n) X ν(l) X
1
= e1 χ(l1 )τ2 (l1 )ν1∗ (l2 ) + o(1).
n
n 4
l l=l l
l<D 1 2
In a way similar to the proof of (17.5), by Lemma 17.1, we find that the right side is equal
to
X ν(l)2
e1 + o(1) = e1 a + o(1).
4
l
l<D
Inserting this into (17.8) gives

Φ−
3 (p) = e1 ap + o(p).
Since (pt0 )β2 = 1 + O(α1), it follows by (17.7) that

X
(pψ t0 )β3 I3− (ψ) = e1 aP + o(P). (17.9)
ψ∈Ψ1
98
Finally, from (17.1), (17.6) and (17.9) we conclude
Φ3 = −(e0 + e1 )aP + o(P). (17.10)

By the discussion at the end of Section 2, it suffices to prove (2.32) and (2.33).
Proof of (2.32).
By (12.3), (12,17), (13.7), (15.24), (16.17) and (17.10),
Ξ13 = c3 aP (18.1)
with
c3 = −i(3e1 + 3e2 + e3 + e0 ) + e∗1 + 2e∗2 + ǫ.
In view of (15.), we can write
ej = e′j − e′′j , 1≤j≤3
with
e′j = (e′1j + ι2 e2j )(ῑ3 e3j + ῑ4 e2j )
e′′j = e′′1j (ῑ3 e3j + ῑ4 e2j ).
By calculation (there is a theoretical interpretation),
i(3e′′1 + 3e′′2 + e′′3 ) + e∗1 = ǫ.
Hence
c3 = −i(3e′1 + 3e′2 + e′3 + e0 ) + 2e∗2 + 2ǫ.
Direct calculation shows that
ℜ{c3 } < −6.9951 (18.2)
It follows from (8.24), (9.8) and (18.2) that
c1 + c2 + 2ℜ{c3} < 0.001.
This with together (8.23), (9.7) and (18.1) yields (2.32).

Proof of (2.33).
99
By Lemma 8.1,
X X
|J1 (ρ, ψ)|2 ω(ρ) = 2ℜ{Θ1 (a23 , a23 } + o(P). (18.3)
ψ∈Ψ1 ρ∈Z(ψ)
We have
X X |χ(d)||µχ(r)|λ0j (dr) X χ(m)f˜ log(drm)/ log P

Sj (a23 , a23 ) =
r d
drϕ(r) m
m1−βj
χ(n)f˜ log(drn)/ log P ξ0j (n; d, r)
X
× .
n
n
dr < P 0.5 , P 0.5 ≤ dr < P 0.502 , P 0.502 ≤ dr < P 0.504 .
By the discussion in Section 8 and 10, the sum over dr < P 0.5 contributes o(α); the sum
over P 0.5 ≤ dr < P 0.502 is equal to
2
500L′ (1, χ) |χ(n)|λ0j (n)
X
− 1 − βj log(n/P 0.5 ) − 1 + Y1j (n) + o(α);

log P 0.5 0.502
ϕ(n)
P ≤n<P

2
500L′ (1, χ)
X |χ(n)|λ0j (n)
1 − βj log(P 0.504 /n) 1 + Y1j (n) + o(α).

log P 0.502 0.504
ϕ(n)
P ≤n<P
The factors βj log(n/P 0.5 ), βj log(P 0.504 /n) and Y1j (n) in the above sums make minor
contribution. If we disregard these factors, the major contribution to Sj (a23 , a23 ) will be
2
500L′ (1, χ) |χ(n)|λ0j (n)
X 1000a
= + o(α).
log P ϕ(n) log P
P 0.5 ≤n<P 0.504
Thus we have the crude bound

1100a
ℜ{Sj (a23 , a23 )} < ,
log P
so that
1 2 3 4400a
ℜ S1 (a23 , a23 ) + S2 (a23 , a23 ) + S3 (a23 , a23 ) < .
2α α 2α π
This yields (2.33) by Lemma 8.1 and Proposition 7.1. ✷
100
Appendix A. Some Euler Products
This appendix is devoted to proving Lemma 8.3, 15.2, 15.3, 16.1 and 16.2. For nota-
tional simplicity we shall write
u = q −1 , v = χ(q).
Proof of Lemma 8.3. Note that
λ(km, s) = λ(k, s)λ̃(m, k; s)
with
Y (1 − q −s−β1 )(1 − q −s−β2 )(1 − q −s−β3 )
λ̃(m, k; s) = .
1 − q −s
q|m
(q,k)=1
For given d and h, the function λ̃(m, h, d; 1 − βj )ξj (m; d, h) is multiplicative in m, and
λ̃(q r , k; s) = λ̃(q, k; s) for any r. Thus, for σ > 1,
Y
Uj (d, h; s) = λ(dh, 1 − βj ) Tj (d, 1, s; q)
q
with
(1 − χ(q)q −s−βj+1 )(1 − χ(q)q −s−βj+2 ) X χ(q r )ξj (q r ; d, h)

Tj (d, h, s; q) = 1+λ̃(q, dh; 1−βj ) .
1 − χ(q)q −s r
q rs
Since
λ̃(q; d, 1 − βj )χ(q)ξj (q r ; d, h) = χ(q)(κ(q) − q −βj ) + O(q −1)
= χ(q)(q −βj+1 + q −βj+2 − 1) + O(q −1),
we have Tj (d, h, s; q) = 1 + O(q −9/5 ) for σ > 9/10, so Ujµ (d, s) is analytic in this region,
and Y
Uj (d, h; s) = λ(dh, 1 − βj ) Tj (d, h1 , s; q) + O(D −c ).
q<D
Since
ϕ(dh)2
λ(dh, 1 − βj ) = + O(α1 )
(dh1 )2
and Tj (d, h, s; q) = 1 if q|D, the proof is reduced to showing that
Tj (d, h, s; q) = 1 + O(α log q/q) if (q, dh1 ) = 1, (A.1)
101
1
Tj (d, h, s; q) = + O(α log q/q) if q|h1 , (A.2)
1 − vu
and
1 − u − vu
Tj (d, h, s; q) = + O(α log q/q) if q|d (q, h1 ) = 1, (A.3)
(1 − vu)(1 − u)
provided
|s − 1| < 5α, q<D and (q, D) = 1,
which are henceforth assumed. We discuss in three cases.
Case 1. (q, dh) = 1.
We have
κ(q r−1 )q 1−βj
ξj (q r , dh) = κ̃(q r ; dh, 1 − βj ) − .
q−1
because κ̃(q r−1 ; dq, 1 − βj ) = κ(q r−1 ). Note that
X X
κ(q r ) = q µ1 β1 +µ2 β2 +µ3 β3 − q µ1 β1 +µ2 β2 +µ3 β3 = τ2 (q r )(1 + O(αr log q))
µ1 +µ2 +µ3 =r µ1 +µ2 +µ3 =r−1
where µ1 , µ2 and µ3 run through non-negative integers. Hence

∞
X r 1
ξj (q r , dh) = (r + η + 1)uη − + O(αr log q/q) = + O(αr log q/q).
η=0
1−u (1 − u)2
It follows that
X χ(q r )ξj (q r ; d, h) 1

1

= − 1 + O(α log q/q).
r
q rs (1 − u)2 1 − vu
This together with the relations
λ̃(q, dh; 1 − βj ) = (1 − u)2 + O(α log q/q)
and
(1 − χ(q)q −s−βj+1 )(1 − χ(q)q −s−βj+2 )
= 1 − vu + O(α log q/q),
1 − χ(q)q −s
yields (A.1).
Case 2. q|h.
We have
ξj (q r , dh) = κ(q r ) = r + 1 + O(αr log q/q),
102
so that
X χ(q r )ξj (q r ; d, h) 1
= − 1 + O(α log q/q).
r
q rs (1 − vu)2
This yields (A.2) since λ̃(q, dh; 1 − βj ) = 1.

Case 3. q|d, (q, h1 ) = 1.
We also have λ̃(q, dh; 1 − βj ) = 1 and
κ(q r−1 )q r
ξj (q r , dh) = κ(q r ) − =r+1− + O(αr log q/q),
q−1 1−u
so that
X χ(q r )ξj (q r ; d, h) 1 vu
= −1− + O(α log q/q).
r
q rs (1 − vu) 2 (1 − u)(1 − vu)2
This yields (A.3). ✷

To prove Lemma 15.2 and 15.3 we need several prerequisite results. In what follows
assume dl < P22 , (dl, D) = 1 and |s − 1| < 5α. We have
Y (1 − q −s−β1 )(1 − q −s−β2 ) X ξ1 (q r ; d, l)
M1 (d, l; s) = 1 + λ̃1 (q, d)
q<D
(1 − q −s )(1 − χ(q)q −s ) r
q rs (A.4)
−c

× 1 + O(D ) .
Note that
κ̃1 (q r−1 , dq) = κ1 (q r−1 ),
so that (
χ(q)q
κ̃1 (q r , d) − q−1 1
κ (q r−1 ) if (q, l) = 1
ξ1 (q r ; d, l) = (A.5)
κ̃1 (q r , d) if q|l
for any q, d, l and r.
Proof of Lemma 15.2. By (A.4), (A.5) and the relations
(1 − q −s−β1 )(1 − q −s−β2 ) 1 − q −1

r α log q
λ̃1 (q , 1) = λ1 (q), = +O ,
(1 − q −s )(1 − χ(q)q −s ) 1 − χ(q)q −1 q
it suffices to show that
X 1 1 − χ(q)q −1 1 − χ(q)q −2

r χ(q)q r−1
1 + λ1 (q) κ̃1 (q , 1) − κ1 (q ) =
r
q rs q−1 1 − q −1 1 − q −2
(A.6)
α log q
+O
q
103
if q < D, (q, D) = 1, and
X 1 χ(q)q

1

α log q

r r−1
1 + λ1 (q) κ̃1 (q , 1) − κ1 (q ) = +O (A.7)
r
q rs q−1 1 − q −1 q
if q|D.
Assume q < D. Since
X X
κ1 (q r ) = q −µ1 β1 −µ2 β2 − q −µ1 β1 −µ2 β2 ,
µ1 +µ2 =r µ1 +µ2 =r−1
where µ1 and µ2 run through non-negative integers, it follows that
κ1 (q r ) = 1 + O(αr log q)
if q r < P . Hence
X κ1 (q r h)χ(h) X χ(h) 1
r
κ̃1 (q , 1) = = + O(αr log q) = + O(αr log q).
h h 1 − vu
h∈N (q) h∈N (q)
It follows that
X 1 χ(q)q

u

1 v

r r−1
κ̃1 (q , 1) − − κ1 (q ) = − + O(α log q/q).
r
q rs q−1 1 − u 1 − vu 1 − u
This together with the relation
λ1 (q) = 1 − vu + O(α log q/q)
yields
X 1 uv(1 − vu)

r χ(q)q r−1 1
1 + λ1 (q) rs
κ̃1 (q , 1) − κ1 (q ) = − 2
+ O(α log q/q).
r
q q − 1 1 − u (1 − u)
It is direct to verify that
1 uv(1 − vu) 1 − χ(q)q −1 1 − χ(q)q −2

− =
1−u (1 − u)2 1 − q −1 1 − q −2
if χ(q) = ±1, and

1 uv(1 − vu) 1
− =
1−u (1 − u) 2 1 − q −1
if χ(q) = 0, whence (A.6) and (A.7) follow. ✷
104
Proof of Lemma 15.3. We give a sketch only, as the situation is similar to Lemma
15.2. it can be verified that the factor U1j (q, s) in the Euler product representation
Y
U1j (s) = U1j (q, s)
q
satisfies U1j (q, s) = 1 + O(q −2σ ) if (q, D) = 1. Further, in the case q ≤ D we have
U1j (q, 1) = (1 − q −1 )2 + O(α1 /q) if q|D,
and
(1 − q −2 )2
U1j (q, 1) = + O(α1 /q) if (q, D) = 1.
1 − χ(q)q −2
Proof of Lemma 16.1. For any q, r, d and l we have
(
κ̃2 (q r ; d) − χ(q)q
q−1 2
κ (q r−1 ) if (q, l) = 1
ξ2 (q r ; d, l) =
κ̃2 (q r ; d) if q|l,
|κ2 (q r )| = |q −β1 − 1| ≪ α log q, λ̃2 (q, d) = 1 + O(α log q/q).

Hence
X ξ2 (q r ; d, l) χ(q)
1 + λ̃2 (q, d) = 1− + O(α log q/q) if (q, l) = 1
r
q rs q−1
and
X ξ2 (q r ; d, l)
1 + λ̃2 (q, d) = 1 + O(α log q/q) if q|l.
r
q rs
On the other hand we have
1 − q −s−β1 1
−s −s
= + O(α log q/q).
(1 − q )(1 − χ(q)q ) 1 − χ(q)q −1
It follows that

Y 1 χ(q) Y 1
M2 (d, l; s) = −1
1− + O(α1).
q<D
1 − χ(q)q q − 1 q<D 1 − χ(q)q −1
(q,l)=1 q|l
It is direct to verify that in either case the assertion holds. ✷

Proof of Lemma 16.2. We give a sketch only. If dl < D, (dl, D) = 1 and |s − 1| ≤ 5α,
then
M2 (d, l; s)
= Π2 (l) + O(α1 τ2 (dl))
M∗2 (s)
105
with −1
Y χ(q)
Π2 (l) = 1− if χ(2) 6= 1,
q−1
q|l
Y −1
χ(q)
Π2 (l) = 1− if χ(2) = 1, 2|l,
q−1
q|l
q>2
Π2 (l) = 0 if χ(2) = (l, 2) = 1.

The assertion follows by discussing the cases χ(2) 6= 1 and χ(2) = 1 respectively. ✷
Appendix B. Some arithmetic sums

Proof of Lemma 15.1. Put
X
̺j (n) = µ(d)dβj .
d|n
First we claim that

X |̺j (n) − ̺∗j (n)|
≪ L−8 . (B.1)
n1
Hence X
̺j (n) − ̺∗j (n) ≪ ν(h)τ2 (n/h).
h|n
h>1
If h > 1 and (h, Q) = 1, then h > D 4 . Hence, by substituting n = hm,

X |̺j (n) − ̺∗j (n)| X ν(h) X τ2 (m) X ν(h)
≪ ≪ (log P )2 .
nD m<P/h D <h1
106
Note that ̺j (q r ) = ̺j (q) for any r, and ̺j (q) ≪ α log q if q < P . Thus the left side above
is X
|̺j (q)| Y X log q
1 + |̺j (q)|/q + O(1/q 2) ≪ α

≤ .
4
q q<P 4
q
q<D q<D
This yields (B.2).

By (B.1) and (B.2), for µ = 2, 3,
X κµ (l1 l)̺∗j (l) X κµ (l1 l)̺j (l)
= + O(L−8 ).
l l
l
(l,Q)=1
We proceed to prove theassertion with µ = 2. Since

X ̺j (l) ζ(s)
=
l
ls ζ(s − βj )
for σ > 1 and

1 (P2 /m)s
Z
κ2 (m) = ds,
2πi (1) (log P2 )(s − β7 )2
it follows that
X κ2 (l1 l)̺j (l) 1 ζ(1 + s) (P2 /l1 )s
Z
= ds.
l
l 2πi (1) ζ(1 + s − βj ) (log P2 )(s − β7 )2
In a way similar to the proof of Lemma 8.1, we see that the right side is equal to the sum
of the residues of the integrand at s = 0 and s = β7 plus an acceptable error. By direct
calculation, the residue at s = 0 is
βj 8j
− =− + O(α1);
β72 log P2 25πi
the residue at s = β7 is, by the Cauchy integral formula,
β β7
ζ(1 + β7 ) log(P2 /l1 ) P2 7 1 P2 d ζ(1 + s)
+ |s=β7
ζ(1 + β7 − βj ) log P2 l1 log P2 l1 ds ζ(1 + s − βj )

2j 8j
= 1− + exp{5πi/4} + O(α1 ).
5 25πi
These together complete the proof in case µ = 2. In case µ = 3 the proof can be obtained
with β6 and P3 in place of β7 and P2 respectively. The same argument also gives
X κ1 (l1 l)̺j (l) 2j j

j
= 1− + exp{0.756πi} − + O(α1).
l 3 1.134πi 1.134πi
l
107
For µ = 1 the proof is therefore reduced to showing that
X κ1 (l1 l)̺j (l) Z 0.004
j
= exp{(3/2)(0.504 − z)πi} − exp{(3/4)πi} dz + O(α1).
12
l 0.756 0
l>P /l1
(B.3)
By (4.2) and (4.3), the left side of (B.3) is equal to
X ̺j (l) P1 β6 1
Z 0.504 z
P
0.5
P
g −g dz + O(α1 ).
l
l l1 l 0.504 0.5 l1 l l1 l
By a change of variable, for 0.5 ≤ z ≤ 0.504,

β6 z 0.5
P1 P P 1 ω1 (s − β6 ) ds
Z
g −g = P β6 (0.504−z) P zs − P 0.004β6 P 0.5s .
y y y 2πi (1) y s (s − β6 )
Hence, in a way similar to the proof of, we find that the left side of (B.3) is
Z 0.504
ζ(1 + s) ω1 (s − β6 ) ds

1 1
Z
β6 (0.504−z) zs 0.004β6 0.5s
P P −P P dz + O
0.504 0.5 2πi (1) ζ(1 + s − βj ) l1s (s − β6 )
1 βj 0.504 β6 (0.504−z)
Z
− P 0.004β6 dz + O(α1 ). ✷

= P
0.504 β6 0.5
Proof of Lemma 17.1. By Lemma 3.1,
X ν(n)2 X ν(n)2 T
= g + o(1).
4
n n
n n
n<D
The sum on the right side is equal to
ν(n)2 T s ω1 (s)
Z X
1
1+s
ds.
2πi (1) n
n s
Assume σ > 1. We have

X ν(n)2 Y X ν(pr )2
= 1+ .
n
ns p r
prs
If χ(p) = 1, then (see [19, (1.2.10)])

X ν(pr )2
1+ = (1 − p−s )−4 (1 − p−2s ) = (1 − p−s )−2 (1 − χ(p)p−s )−2 (1 − p−2s );
r
prs
108
if χ(p) = −1, then
X ν(pr )2 X 1
1+ =1+ = (1 − p−2s )−1 = (1 − p−s )−2 (1 − χ(p)p−s )−2 (1 − p−2s );
r
prs r
p 2rs
if χ(p) = 0, then
X ν(pr )2 X 1
1+ =1+ = (1 − p−s )−1 = (1 − p−s )−2 (1 − χ(p)p−s )−2 (1 − p−s ).
r
prs r
p rs
Hence
X ν(n)2 Y Y
= ζ(s)2 L(s, χ)2 (1 − p−2s ) (1 − p−s ).
n
ns
(p,D)=1 p|D
In a way similar to the proof of, by (A) and simple estimate, we find that the integral
(14) is equal to the residue of the function
Y Y T s ω1 (s)
ζ(1 + s)2 L(1 + s, χ)2 (1 − p−2(1+s) ) (1 − p−(1+s) )
s
(p,D)=1 p|D
at s = 0, plus an acceptable error O, which is equal to

Y Y
L′ (1, χ)2 (1 − p−2 ) (1 − p−1 )(1 − p−2 )−1 + o(1) = a + o(1). ✷
p p|D
References
[1] R. Balasubramanian, J. B. Conrey and D. R. Heath-Brown, Asymptotic mean square

of the product of the Riemann zeta-function and a Dirichlet polynomial, J. reine
angew. Math. 357(1985), 161-181.
[2] J. B. Conrey, A. Ghosh and S. M. Gonek, A note on gaps between zeros of the zeta
function, Bull. London Math. Soc. 16(1984), 421-424.
[3] J. B. Conrey, A. Ghosh and S. M. Gonek, Mean values of the Riemann zeta-function
with application to the distribution of zeros, Number theory, trace formulas and
discrete groups (Academic Press, Boston, 1989), 185-199.
[4] J. B. Conrey, A. Ghosh and S. M. Gonek, Simple zeros of the Riemann zeta-function,
Proc. London Math. Soc. (3)76(1998), 497-522.
109
[5] J. B. Conrey, H. Iwaniec and K. Soundararajan, Asymptotic large sieve, Preprint
2011, arxiv 1105.1176
[6] J. B. Conrey, H. Iwaniec and K. Soundararajan, Critical zeros of Dirichlet L-

functions, Preprint 2011, arxiv 1105.1177
[7 ] H. Davenport, Multiplicative Number Theory, 3rd. ed. (revised by H. L. Mont-

gomery), Springer-Verlag, New York, 2000
[8] D. Goldfeld, A simple proof of Siegel’s theorem, Proc. Nat. Acad. Sci. USA 71(1974),
1055.
[9] D. Goldfeld, An asymptotic formula relating the Siegel zero and the class number of
quadratic fields, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 2(1975), 611-615.
[10] D. Goldfeld, The class number of quadratic fields and the conjectures of Birch and
Swinnerton-Dyer, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 3(1976), 624-663.
[11] A. Granville and H. M. Stark, ABC implies no “Siegel zeros” for L-functions of
characters with negative discriminant, Invent. Math. 139(2000), 509-523.
[12] B. Gross and D. Zagier, Heegner points and derivatives of L-series, Invent. Math.,
84(1986), 225-320.
[13] D. R. Heath-Brown, Small class numbers and the pair correlation of zeros, Conference
in Celebration of the Centenary of the the Proof of the Prime Number Theorem (A
Symposium on the Riemann Hypothesis), Seattle, 1996.
[14] H. Iwaniec and E. Kowalski, Analytic number theory, Coll. Pulb. vol. 53, Amer.
Math. Soc. 2004
[15] H. Iwaniec and P. Sarnak, The non-vanishing of central values of automorphic L-

functions and Landau-Siegel zeros, Israel J. Math., 120(2000), 155-177.
[16] A. Karatsuba, Basic Analytic Number Theory, (translated from Russian by M.

Nathanson), Springer-Verlag, New York, 1993.
[17] H. L. Montgomery, The pair correlation of zeros of the zeta-function, Analytic number
theory, Proc. Symp. Pure Math. Vol 24, 181-193, (Amer. Math. Soc., Providence, RI.
1973)
[18] C. L. Siegel, Über die Classenzahl quadratischer Zuhlkörper, Acta Arith. 1(1936),
83-86.
[19] E. C. Titchmarsh, The theory of the Riemann zeta-function, 2nd. ed. (revised by D.
R. Heath-Brown), Oxford Univ. Press, Oxford, 1986
110
Yitang Zhang
Department of Mathematics
University of California at Santa Barbara
Santa Barbara, CA 93106
E-mail: yitang.zhang@math.ucsb.edu
111

Discrete Mean Estimates and The Landau-Siegel Zero

Uploaded by

Copyright:

Available Formats

Discrete Mean Estimates and The Landau-Siegel Zero

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Discrete Mean Estimates and The Landau-Siegel Zero

Uploaded by

Copyright:

Available Formats

Discrete mean estimates and the Landau-Siegel zero

arXiv:2211.02515v1 [math.NT] 4 Nov 2022

L(1, χ) ≫ (log D)−2022

where the implied constant is absolute and effectively computable.

1 − ρ̃ < c0 (log D)−1

L(1, χ) > C1 (ε)D −ε .

L(σ, χ) 6= 0 if σ > 1 − C2 (ε)D −ε .

L(1, χ) ≫ (log D)−1 . (1.1)

L(1, χ) ≫ D −1/2 (log D)1−ε

L(1, χ) > c1 (log D)−2022

where c1 > 0 is an absolute, eﬀectively computable constant.

2. Notation and outline of the proof

L(s, θ) = Z(s, θ)L(1 − s, θ̄), (2.2)

Z(s, θ) = θ(−1)τ (θ)k −s ϑ(s)(1 + O(e−πt)), (2.4)

p∼P for P < p < P (1 + L−68 ).

In Section 3 we shall introduce a subset Ψ1 of Ψ and prove

Y (s, ψ)2 = Z(s, ψ)−1.

are independent of these choices. The functional equation (2.2) gives

M(s, ψ) = Y (s, ψ)−1 L(1 − s, ψ̄).

Since |Y (1/2 + it, ψ)| = 1, it follows that

|M(1/2 + it, ψ)| = |L(1/2 + it, ψ)|,

M(1/2 + it, ψ) ∈ R, (2.11)

β1 = iα(1 − 5c′ αL), β2 = 2iα(1 + c′ αL), β3 = 3iα(1 − c′ αL). (2.13)

M ′ (ρ, ψ) = Y (ρ, ψ)L′ (ρ, ψ) 6= 0.

It will be proved that

We introduce the smooth weight

which is positive for σ = 1/2. Lemma 2.3 implies that

for some P ′ < P , where f is a polynomial with f (1) = 0. It should be stressed, on

for σ = 1/2. On the other hand, the sum

is not too small in absolute value.

H1 (ρ, ψ)J1 (ρ, ψ) + H2 (ρ, ψ)J2 (ρ, ψ)

This yields, by Lemma 2.3,

P1 = P 0.504 , P2 = P 0.5 T −10 , P3 = P 0.498 , (2.21)

ι2 = 0.94977 − 1.38995i, ι3 = −1.00635 − 0.22789i, ι4 = −0.68738 + 1.60688i. (2.26)

Let f˜(z) be given by

(see Lemma 5.7).

|Ξ∗1 | > 5aP.

Proposition 2.5. Assume that (A) holds. Then

Ξ∗2 < 2aP.

Proposition 2.6. Assume that (A) holds. Then

Under Assumption (A), a contradiction is immediately derived from (2.18), Proposi-

M(ρ + β2 , ψ)M(ρ + β3 , ψ) > 0,

since M(1/2 + it, ψ) is a real-valued continuous function in t. Similarly we have

for ψ ∈ Ψ1 and ρ ∈ Z̃(ψ).

respectively. It is easy to see that

|υ(n)| ≤ ν(n) ≤ τ2 (n). (3.1)

Lemma 3.1. Assume (A) holds. Then

with α∗ = L−2024 . If |s| = α∗ , then φ(1 + s) = O(1) by (3.3),

by (A) and standard estimates, and

ζ(1 + s) ≪ L2024 , (P 2s − D 4s )Γ(s) ≪ L9 .

It follows that the residue at s = 0 is ≪ L−2011 , whence the result follows. ✷

This completes the proof. ✷

By (3.1) we may write

with |ν20 n| ≤ τ40 (n), |υ20 n| ≤ τ40 (n). Write

By Cauchy’s inequality and the first assertion of Lemma 3.3 we obtain

holds for all but at most O(PL−740 ) characters ψ in Ψ.

holds for all but at most O(PL−739 ) characters ψ in Ψ.

We have ς(n) = 0 unless n = 1 or D 4 < n ≤ D 8 . Write

for x > D 4 . It is direct to verify that