A PROOF ON HYPOTHESIS OF DIRICHLET

DIVISOR PROBLEM
arXiv:1105.6155v1 [math.NT] 31 May 2011

MO, Guoduan
Dept.of Appl.Math.,Tongji Univ., Shanghai, China, 200092

Abstract
1
We have proved that △(X) = X 4 +ε(X) , ε(X) = c
log log X .

Key Words: Dirichlet Divisor Problem, 1/4-Hypothesis.

2000 Mathematics Subject Classification: 11N37

1 RESULTS AND THOUGHT OF PROOF

Let X
D(X) = d(n), (1.1)
n≤X

where d(n) denotes the number of divisors of n as usual, let

△(X) = D(X) − X log X − (2γ − 1)X, (1.2)

θ∗
and let θ = inf{θ∗ : △(X) ≪ X }. Then the following bounds for θ have been
obtained:
1
1. θ ≤ 2
= 0.50000 · · · Dirichlet 1849.

1
2. θ ≤ 3
= 0.33333 · · · Voronoi 1903

33
3. θ ≤ 100
= 0.33000 · · · Vander Corput[6] 1922

27
4. θ ≤ 82
= 0.32926 · · · Van der Corput[7] 1928

15
5. θ ≤ 46
= 0.32608 · · · Chih,T. T.[8](1950), Richert[9] 1953

13
6. θ ≤ 40
= 0.32500 · · · Yin Wenlen[10] 1959

1
 12
7. θ≤ 37
= 0.32432 · · · Kolesnik[11] 1969

346
8. θ≤ 1067
= 0.32427 · · · Kolesnik[12] 1973
35
9. θ≤ 108
= 0.32407 · · · Kolesnik[13] 1982

139
10. θ ≤ 429
= 0.32400 · · · Kolesnik[14] 1985

7
11. θ ≤ 22
= 0.31818 · · · Iwniec and Mozzochi[15] 1988

131
12. θ ≤ 416
= 0.31490 · · · M.N.Huxley[16] 2003
The hypothesis value is θ = 1/4. Moreover, in 1916, Hardy [15] proved that θ ≥ 1/4.
In this paper we are going to prove the hypothesis is true. For this purpose, we will
prove the following theorem:

Theorem Let X1 < X2 be large positive number, X1 ≤ X ≤ X2 and

L = log X2 , X1 ≍ X2 . (1.3)

Then
1
△(X) = X 4 Λ(X) + δ(X), (1.4)
where δ(X) ≪ X ε , ε > 0 arbitrary small, and
4K
X X X
− 54
√ √ am0 3π
Λ(X) = 2
C 1 (a) d(n)n cos(4π n( X + √ ) − )
(m0 + 1) a=1 2 2 X 4
X L 2
n≤ V

√ 4K+2
X X X 7 √ √ am0 5π
+ C2 (a) d(n)n− 4 cos(4π n( X + √ ) − ), (1.5)
m0 + 1 a=1 X L2 2 X 4
n≤ 2
V

with
C0 L
K=[ ], C0 ≥ 200, (1.6)
log L
C1 (a), C2 (a) being independent of X, and

L p
C1 (a), C2 (a) ≪ exp(O( )), m0 = 2[ X2 L−2 ]. (1.7)
log L

Clearly, L, K and m0 are independent of X by (1.3) and (1.6). The following

Corollary is obtained immediately by this Theorem.

Corollary
1 1
△(X) ≪ X 4 +ε(X) , 0 < ε(X) = O( ). (1.8)
log log X

2
Proof: In fact,
∞ ∞
L X
− 45
X 7 L 1
Λ(X) ≪ exp(O( ))( d(n)n + d(n)n− 4 ) ≪ exp(O( )) ≪ X (O( log L )) ,
log L n=1 n=1
log L

where the last inequality is given by L ≍ log X1 ≍ log X. ✷

Clearly, the hypothesis is true by this Corollary.

To prove Theorem 1.1, suppose that

U = [X], (1.9)

√
A = C0 UL, C0 ≥ 200, (1.10)

√ k+j
B= U + √ + η, (1.11)
2 U
p
k = 0, 1, · · · , m ≪ m0 = [ X2 L−2 ], (1.12)
√
j = 0, ±1, · · · , ±j0 , j0 ≤ V L, (1.13)

1
V = X2ε0 , ε0 = , (1.14)
log L
C0 L
η = η1 + · · · + ηK , K = [ ], (1.15)
log L
k1
ηλ = √ , λ = 1, 2, · · · , K, (1.16)
U
p
k1 = 0, 1, · · · , m1 ≪ m0 = [ X2 L−2 ], (1.17)

1 3
ξ1 = √ , ξ2 = √ , (1.18)
16 U 16 U
√
5 L
ZA
2 θ2 1
X
S(B) = A e−πA dθ d(n)n− 4 . (1.19)
√
√
ξ1 < n−B−θ<ξ2
− 5 AL

Clearly,
(B + θ + ξ2 )2 − (B + θ + ξ1 )2 = (ξ2 − ξ1 )(2B + 2θ + ξ1 + ξ2 )
1 √
= √ (2 U + O(1))
8 U
1
< (1.20)
3
3
and
√ k+j
{(B + θ + ξ)2 } = {( U + √ + η + θ + ξ)2 }
2 U
√ √ √ k+j
= {U + k + j + 2 U η + 2 U ξ + 2 Uθ + ( √ + η + θ + ξ)2 }
2 U
√ √ −1
= {2 U ξ + 2 U θ + O(L )}.
√
By (1.15) and (1.16) we know that 2 U η is an integer, thus
√
√ √ 1 5 L 1 10 1 1 3
2 U(ξ + θ) ≥ 2 U ( √ − )= − ≥ − =
16 U A 8 C0 8 20 40

and √
√ √ 3 5 L 3 10 3 1 17
2 U (ξ + θ) ≤ 2 U ( √ + )= + ≤ + = ,
16 U A 8 C0 8 20 40
therefore,
1 9
< {(B + θ + ξ)2 } < (1.21)
20 20
√ √
1 3
for 16√ U
≤ ξ ≤ 16√ U
, − 5 AL ≤ θ ≤ 5 AL . It follows from (1.20) and (1.21) that there
doesn’t exist any integer in the interval

[(B + θ + ξ1 )2 , (B + θ + ξ2 )2 ].

Hence X X
= = 0, (1.22)
√
ξ1 < n−B−θ≤ξ2 (B+θ+ξ1 )2 <n≤(B+θ+ξ2 )2

and by (1.19),
S(B) = 0. (1.23)
Denote k k
η1 = √1 ηK = √1
f (η) |
ηK
= f (η1 + η2 + · · · + ηK ) | η1 =0
U
··· | ηK =0
U
. (1.24)
By (1.23),
S(B) | ηK
= 0. (1.25)
Therefore,
2m0 2m0 m
1 −πj 2
|
X X X X
Ω= √ e V S(B) ηK
= 0. (1.26)
V √ √
k1 =m0 +1 m=m0 +1 k=0
− V L≤j≤ V L

On the other hand, in the following we are going to prove that

(−1)K+1(m0 + 1)2 1
Ω= 1 (△(X) − X 4 Λ(X) + O(X ε )). (1.27)
4X 4

4
This and (1.26) lead to (1.4).

For convenience in the proof we are going to use the following definition.

Definition 1.1 If
k k
η1 = √1 ηK = √1
f (η) |ηK
= f (η1 + η2 + · · · + ηK ) | η1 =0
U
··· | ηK =0
U
≪ Y,

then we write
f (η) = Oη (Y ), or f (η) ≪η Y. (1.28)
Moreover, we denote

| |f (η |
X X
f (η) | ηK
|= ··· 1 + η2 + · · · + ηK ) .
k k
η1 =0, √1 η1 =0, √K
U U

Clearly,
f (η) |
ηK
≤ f (η)| | ηK
|. (1.29)
Particularly, if f (η) ≪ M, then

f (η) |ηK
≤ f (η)| |ηK
| ≪ M2K . (1.30)

For arbitrary small ε > 0, or ε = ε(U) = O(1/ log L), we are going to use expressions,
such as
U 2ε ≪ U ε , (1.31)
without explanation.

2 SOME LEMMAS (I)

Let X
S′ = ′
d(n)f (n),
a≤n≤b

where 0 < a < b < ∞, f (x) ∈ C 2 [a, b), ′ is that if a or b is an integer, then 21 d(a)n(a)
P
or 21 d(b)n(b) is to be counted instead of d(a)n(a) or d(b)n(b), then (for example, (3.2),
(3.15) in [2])
X
S′ = ′
d(n)f (n)
a≤n≤b
Z b ∞ Z a
−5
X
= (log x + 2γ)f (x)dx + d(n) f (x)(Θ(nx) + O((nx) 4 ))dx, (2.1)
a n=1 b

5
where
√ √ 1 √
Θ(nx) = 2(nx)−1/4 (cos(4π nx + π/4) − √ cos(4π nx + 3π/4)). (2.2)
32π nx
P
In the following we are going to evaluate general sum dλ (n)f (n) in Γ− function
with more accurate O-term, and indicate the results in the finite term. Denote
Z h0 Z h0
1 X
Sλ = λ−1 ··· dh1 · · · dhλ−1 dλ (n)f (n), (2.3)
h0 0 0 1
a+h<n λ ≤b+h

where X
dλ (n) = 1, λ ≥ 2, (2.4)
n1 ···nλ =n
a
f (x) ∈ C λ [aλ , (b + (λ − 1)h0 )λ ], 0 < h0 < , (2.5)
2

h = h1 + · · · + hλ−1 .

Lemma 2.1 Let

λ  
X λ
β0 = ((λ − 1)h0 )j bλ−j ,
j
j=1

if h0 is sufficient small such that

β0 < min(1 − {aλ }, 1 − {bλ }),

then X
Sλ = dλ (n)f (n). (2.6)
1
a<n λ ≤b

In fact, in this case, there are not any integer in (aλ , (a + h)λ ] and (bλ , (b + h)λ ],
hence, X X
= 0, = 0.
aλ <n≤(a+h)λ bλ <n≤(b+h)λ

Lemma 2.2 For 0.9 < a < b < ∞,

X dλ (n) Z h0 Z h0 Z (b+h)λ
Sλ = Sλl + λ−1
··· dh1 · · · dhλ−1 f (x)Θλ (nx)dx + O(ε1 ) (2.7)
n≤N
h0 0 0 (a+h)λ

holds for any ε1 > 0 as N ≥ N(ε1 ), where

Z h0 Z h0 Z (b+h)λ
1
Sλl = λ−1 ··· dh1 · · · dhλ−1 f (n)pλ (x)dx, (2.8)
h0 0 0 (a+h)λ

6
with
1 1
Z
pλ (x) = xs ς λ (s + 1)ds = (log x)λ−1 + C1 (log x)λ−2 + · · · + Cλ ,
2πi |S|=1/2 (λ − 1)!

αλ (λ) −α
2 1 1
X γα (nx) λ 1 (λ − 1)π απ
Θλ (nx) = √ (nx)− 2 + 2λ α
cos(2nλ(nx) λ + + )
λ α=0
(2πλ) 4 2
−1 1 α +1
+ 2λ − λλ
+O(nx) 2 (2.9)

and
(λ) (λ) 1 − λ2
γ0 = 1, γ1 = ,
12
for any natural number αλ .

(In this paper we need the case λ = 2 only.)

Lemma 2.3
h0
1
Z X
S = S2 = dh d(n)f (n)
h0 0 √
a+h< n≤b+h

X d(n) Z h0 Z (b+h)2
= Sl + dh f (x)Θ(nx)dx + O(ε1) (2.10)
n≤N
h0 0 (a+h)2

holds for any ε1 > 0 as N ≥ N(ε1 ), where

0.9 < a < b < ∞, 0 < h0 < a/2, f (x) ∈ C 2 [a, b + h0 ],

Z (b+h)2
1 h0
Z
Sl = dh f (x)(log x + 2γ)dx, (2.11)
h0 0 (a+h)2
α0 −α
√ − 14
X γα (nx) 2 √ π απ −3 α0
Θ(nx) = 2(nx) cos(4π nx + + ) + O((nx) 4 − 2 ), (2.12)
α=0
(4π)α 4 2
with
1 9 75 3675 59535 2401245
γ0 = 1, γ1 = − , γ2 = , γ3 = − , γ4 = 15 , γ5 = − 18 , γ6 = ···
8 128 1024 2 2 222
(2.13)
for any natural number α0 .

(in this paper we need not the concrete number values of γj , j ≥ 1, we see γ0 , γ1 be
consistent for known results in (2.2).)
A corollary of Lemma 2.3 is the following Lemma 2.4

7
Lemma 2.4 Let
h2
1
Z X
S= g(h)dh d(n)f (n),
h0 h1 √
a+h< n≤b+h

where a + h1 > 0.9, h1 < h2 , 0 < a < b < ∞, g(h) ∈ C 2 [h1 , h2 ], f (x) ∈ C 2 [(a + h1 )2 , (b +
Rh
h2 )2 ], h0 = h12 g(h)dh > 0, then

Z h2 Z (b+h)2
1
S = g(h)dh f (x)(log x + 2γ)dx
h0 h1 (a+h)2
X d(n) Z h2 Z (b+h)2
+ g(h)dh f (x)Θ(nx)dx + O(ε1)
n≤N
h0 h 1 (a+h)2

holds for any ε1 > 0 as N ≥ N(ε1 ), where Θ(nx) is (2.12).

Clearly, Lemma 2.2 has also similar corollary.

Proof of Lemma 2.2: Let X

Dλ (x) = dλ (n), (2.14)
n≤x

then

X Z (b+h)λ
dλ (n)f (n) = f (x)dDλ (x)
(a+h)λ
(a+h)λ <n≤(b+h)λ

x=(b+h)λ Z (b+h)λ
= f (x)Dλ (x) | x=(a+h)λ
− f ′ (x)Dλ (x)dx. (2.15)
(a+h)λ

the last integral is

Z (b+h)λ Z (b+h)λ Z x
′ ′
− f (x)Dλ (x)dx = − f (x)( Dλ (y)dy)′xdx
(a+h)λ (a+h)λ 0

Z x x=(b+h)λ Z (b+h)λ Z x
= −f (x) ′
Dλ (y)dy |x=(a+h)λ
+ ′′
f (x)( Dλ (y)dy)dx
0 (a+h)λ 0
λ
−f ′ (x) σ0 +iT xs+1 ς λ (s)ds x=(b+h)
Z
=
2πi s(s + 1) x=(a+h)λ
|
σ0 −iT
Z σ0 +iT Z (b+h)λ ′′
1 f (x)xs+1 ς λ (s)
+ ds dx + δ, (2.16)
2πi σ0 −iT (a+h)λ s(s + 1)

where σ0 = 1.1 and

8
 Z σ0 −iT Z σ0 +i∞ x=(b+h)λ
1 f ′ (x)xs+1 ς λ (s)
δ = (
2πi σ0 −i∞
ds + ds)(−
s(s + 1)
| x=(a+h)λ
σ0 +iT
Z (b+h)λ ′′
f (x)xs+1 ς λ (s)
+ dx)
(a+h)λ s(s + 1)
Z ∞
dt 1
≪ 2
≪ .
T t T

By (2.16),
(b+h)λ (b+h)λ σ0 +iT
1 xs ς λ (s)ds 1
Z Z Z
′ ′
− f Dλ (x)dx = − f + O( ). (2.17)
(a+h)λ 2πi (a+h)λ σ0 −iT s T
1
Suppose e is the set of hλ−1 and e satisfies |xλ + hλ−1 − m λ | < µ, where m is an
natural number, xλ = x + h1 + · · · + hλ−2 fixed, E = (0, h0 ) \ e. We have
Z
f ((xλ + hλ−1 )λ )Dλ ((xλ + hλ−1 )λ )dhλ−1 ≪ µ,
e
Z
f ((xλ + hλ−1 )λ )Dλ ((xλ + hλ−1 )λ )dhλ−1
E

σ0 +iT
f ((xλ + hλ−1 )λ ) (xλ + hλ−1 )λs ς λ (s)
Z Z
= dhλ−1 ds + δE ,
E 2πi σ0 −iT s
where

Z σ0 −iT Z σ0 +i∞
f ((xλ + hλ−1 )λ ) (xλ + hλ−1 )λs ς λ (s)ds
Z
δE = dhλ−1 ( + )
E 2πi σ0 −i∞ σ0 +iT s
∞ σ0 −iT σ0 +i∞
f ((xλ + hλ−1 )λ ) 1 xλ + hλ−1 λs
Z X Z Z
= dhλ−1 dλ (n)( + ) ( 1 ) ds
E 2πi n=1 σ0 −i∞ σ0 +iT s nλ
∞
1 (xλ + hλ−1 )λ σ0 1
Z X
≪ dhλ−1 dλ (n)( )
T E n=1
n | log +h1λ−1 |
x λ

nλ
1 1
Z X
≪ dhλ−1 (1 + 1 )
T E 1 |n λ − x − h
λ λ−1 |
2
3
≤|n λ −xλ −hλ−1 |≤ 23

1 dhλ
X Z
≪ (1 + 1 )
T 1 E |n λ − xλ − hλ−1 |
1
12
≤|n λ −xλ |≤3

1 1
≪ log
T µ

9
for 0 < h0 < a/2, 0 < µ < 1/2, and the constant in ′′ O ′′ is independent of T . Further-
more,
Z σ0 +iT
(xλ + hλ−1 )λs ς λ (s)
Z Z
λ
f ((xλ + hλ−1 ) )dhλ−1 dS ≪ log T dhλ−1 ≪ µ log T.
e σ0 −iT s e

Hence, taking µ = T1 , we have

Z h0 Z h0 x=(b+h)λ
··· dh1 · · · dhλ−1 f (x)Dλ (x) | x=(a+h)λ
=
0 0

h0 h0 σ0 +iT x=(b+h)λ
1 f (x)xs ς(s)ds log T
Z Z Z
= ··· dh1 · · · dhλ−1 f (x)
2πi s
| x=(a+h)λ
+ O(
T
). (2.18)
0 0 σ0 −iT

It follows from (2.3), (2.15), (2.17) and (2.18) that

h0 h0 σ0 +iT (b+h)λ
1 1 ς λ (s)ds
Z Z Z Z
Sλ = ··· dh1 · · · dhλ−1 (− f ′ (x)xs dx
hλ−1
0 0 0 2πi σ0 −iT s (a+h)λ
x=(b+h)λ
log T
+f (x)xs x=(a+h)λ ) + O( | T
)
Z h0 Z h0 Z (b+h)λ Z σ0 +iT
1 1 log T
= λ−1 ··· dh1 · · · dhλ−1 f (x)dx xs−1 ς λ (s)ds + O( )
h0 0 0 2πi (a+h)λ σ0 −iT T
Z h0 Z h0 Z (b+h)λ Z 1−σ0 +iT
1 1
= λ−1 ··· dh1 · · · dhλ−1 f (x)dx xs−1 ς λ (s)ds
h0 0 0 2πi (a+h)λ 1−σ0 −iT
+Sλl + δ+ + δ− + O(ε1) (2.19)

holds for any ε1 > 0, as T large, where h = h1 + · · · + hλ−1 , λ ≥ 2, Sλl is (2.8) and
σ0 ±iT
1
Z
δ± = ∓ ς λ (s)Iδ (s)ds,
2πihλ−1
0 1−σ0 ±iT

Z h0 Z h0 Z bλ
Iδ (s) = ··· dh1 · · · dhλ−1 f (x + h)(x + h)s−1 dx. (2.20)
0 0 aλ
1
Using the integrations by parts and (2.5), we see that Iδ ≪ Tλ
. It is known that (for
example, 2.1.8, 2.1.3 in [1])

ς(1 − s)
ς(s) = , χ(s)χ(1 − s) = 1, (2.21)
χ(1 − s)
1 πs t t 1
= 2(2π)−s Γ(s) cos = ( )σ−1/2 eit log 2πe (1 + O( )), (2.22)
χ(s) 2 2π t
t > 1, s = σ + it,
1
ς(s) ≪ |t| 6 , σ ≥ 1/2, |t| > 1. (2.23)

10
Hence
Z 1 Z σ0
1 2
δ± ≪ ( + )|ς λ (σ ± iT )|dσ
T λ 1−σ0 1
2
Z σ0
1 λ 1 −λ
≪ λ
(1 + T λ(σ−1/2) )T 6 dσ ≪ T λ( 6 +σ0 −3/2) ≪ T 5 (2.24)
T 1
2

for σ0 = 1.1. Putting s = 1 − s′ in the integral (2.19), noticing (2.21 ) and (2.24),
h0 h0 (b+h)λ σ0 +iT
1 1 f (x)dx x−s ς λ (s)ds
Z Z Z Z
Sλ = ··· dh1 · · · dhλ−1
hλ−1
0 0 0 (a+h)λ 2πi
σ0 −iT 2πi χλ (s)
+Sλl + O(ε1 )
X dλ (n) Z h0 Z h0 Z (b+h)λ
1 f (x)dx σ0 +iT (nx)−s ds
Z
= ··· dh1 · · · dhλ−1
n≤N
hλ−1
0 0 0 2πi (a+h)λ 2πi σ0 −iT χλ (s)
+Sλδ + δN + O(ε1), (2.25)

where
h0 h0 (b+h)λ T
(nχ)−σ0 dt
X Z Z Z Z
δN ≪ dλ (n) ··· dh1 · · · dhλ−1 |f (x)|
n>N 0 0 (a+h)λ −T |χλ (σ0 + it)|
X Z T
≪ dλ (n)n−1.1 (|t| + 1)λ(1.1−1/2) dt ≪ ε1 ) (2.26)
n>N −T

holds for any fixed T as N sufficient large.

Now estimate the integral
Z σ0 +iT1
jλ (s)n−s ds
δλ (n) = , (2.27)
σ0 +iT χλ (s)
Z h0 Z h0 Z (b+h)λ
jλ (s) = ··· dh1 · · · dhλ−1 f (x)x−s dx (2.28)
0 0 (a+h)λ

for σ0 = 1.1, T1 → ∞, n ≤ N. Using the integrations by parts it is easy to obtain that

jλ (s) ≪ t1λ . Thus, by (2.22),
Z T1 Z T1
−1.1 λ(1.1−1/2)−λ −1.1
δλ (n) ≤ n t dt ≤ n t−0.4λ dt ≤ n−1.1 T 1−0.4λ ≤ T −0.2 n−1.1
T T
(2.29)

11
for λ ≥ 3. If λ = 2, then
Z h0 Z (b+h)2
j2 (s) = dh f (x)x−s dx
0 (a+h)2
Z h0 x=b
2
=
−2s + 2 0
dhf ((x + h)2 )(x + h)−2s+2 x=a |
Z h0 Z b
2
− dh (f ((x + h)2 ))′ (x + h)−2s+2 dx
−2s + 2 0 a
x=b h=h0
c1
= 2 f ((x + h)2 )(x + h)−2s+3 x=a h=0
t
| |
c2 h 0 x=b
Z
+ 2
t o
2 ′
dh(f ((x + h) )) (x + h) −2s+3
x=a
|
Z h0
c3
+ 2 dh(f ((x + h)2 ))′′ (x + h)−2S+3 dx + O(t−3). (2.30)
t o
Consider the integral
Z T1 Z T1
(n(x + h))−it dt t
−0.8 2it log eM −0.8
I2n (T, T1 ) = 2 χ2 (1.1 + it)
= t e n dt + O(T ), (2.31)
T t T

where the last equation is given by (2.22), σ0 = 1.1, and

p
Mn = 2π n(x + h). (2.32)
√
If Mn > T − T , then

Z Z
t
I2n (T, T1 ) = ( + )t−0.8 e2it log eMn dt + O(T −0.8)
√ √
|t−Mn |≤ Mn ,T ≤t≤T1 |t−Mn |> Mn ,T ≤t≤T1

T1−0.8 Mn−0.8 Mn−0.8

≪ Mn−0.8+1/2 + T1
+ δT (Mn ) + √
Mn + Mn
+ √
Mn − Mn
| log M n
| | log Mn
| | log Mn
|
t−0.8−1 t−0.8−1
Z
+ ( + )dt.
| log Mtn | | log Mtn |2
|t−Mn |>Mn ,T ≤t≤T1

−0.8
√
For any fixed Mn , as T1 → ∞, the second√ term is ≪ T1 ≪ T −0.8 . If Mn ≤ T + T,
then δT (Mn ) is the first term. If Mn > T + T , then

T −0.8
δT (Mn ) ≪ ≪ T −0.8+1/2 .
| log MTn |
Thus,

12
 Z T1
−0.8
I2n (T, T1 ) ≪ T + Mn−0.8+1/2 + t−0.8−1 dt
T
Mn Mn2
Z
+ √
( + )dt
Mn /2≤|t−Mn |≤2Mn |t − Mn | |t − Mn |2
≪ Mn−0.8+1/2 + T −0.8 ≪ T −0.3 ,
√ √
for Mn > T − T . If Mn ≤ T − T , then it is easy to show I2n (T, T1 ) ≪ T −0.3 . It
follows from (2.27), (2.30) and the definition of I2n (T, T1 ) that δ2 (n) ≪ n−1.1 T −0.3. This
and (2.29) give
δλ (n) ≪ n−1.1 T −0.2 ≪ ε1 n−1.1 , λ ≥ 2. (2.33)
In the same way,
Z σ0 −iT
jλ (s)n−S dS
≪ n−1.1 T −0.2 ≪ ε1 n−1.1 , λ ≥ 2, σ0 = 1.1. (2.34)
σ0 −iT1 χλ (s)
Therefore
X dλ (n) Z h0 Z h0 Z (b+h)λ Z σ0 +iT1 Z σ0 −iT
(nx)−s ds
lim ··· dh1 · · · dhλ−1 ( + ) λ ≪ ε1 .
T1 →∞
n≤N
hλ−1
0 0 0 (a+h)λ σ0 +iT σ0 −iT1 χ (s)
(2.35)
By (2.25), (2.26) and (2.35),
Z h0 Z h0 Z (b+h)λ
X dλ (n) 1 f (x)dx σ0 +iT (nx)−s ds
Z
Sλ = lim
λ−1 T →∞
··· dh1 · · · dhλ−1
n≤N
h 0 0 0 2πi (a+h)λ 2πi σ0 −iT χλ (s)
+Sλl + O(ε1). (2.36)
If s = −1 + it, |t| > 1, then by (2.22),
1 −3
| | ≪ |t| 2 . (2.37)
χ(s)
Let c± be the broken line from σ0 ± iT to −1 ± iT and again to −1 ± i∞, it is easy to
know that
Z h0 Z h0 Z (b+h)λ
(nx)−s ds
Z Z
··· dh1 · · · dhλ−1 f (x)dx( − ) λ → 0(T → ∞) (2.38)
0 0 (a+h)λ c+ c− χ (s)
uniformly for n ≤ N. Therefore, by (2.36), (2.38),
X dλ (n) Z h0 Z h0 Z (b+h)λ
Sλ = ··· dh1 · · · dhλ−1 f (x)Jλn (x)dx + Sλl + O(ε1 ) (2.39)
n≤N
hλ−1
0 0 0 (a+h)λ

holds for any ε1 > 0 as N ≥ N(ε1 ), where Sλl is (2.8) and

1 (nx)−s ds
Z
Jλn (x) = , (2.40)
2πi c χλ (s)

13
the path c is an infinite broken line joining the points −1 − i∞, −1 − iT, αλ + 5/4 −
iT, αλ + 5/4 + iT, −1 + iT, −1 + i∞.
It is know that (for example, (24), P.360 in [3])
√
Γ(s) = 2πe(s−1/2) log s−s+v(s) , (2.41)

where ∞ ∞
({x} − 1/2)dx ϕ(x)dx
Z Z
v(s) = = (2.42)
0 x+s 0 (x + s)2
for s 6= 0, −1, −2 · · · ,
x ∞
1 − cos 2mπx
Z X
ϕ(x) = − ({y} − 1/2)dy = .
0 m=1
2π 2 m2

So that

∞ ∞
1 (1 − cos 2mπx)dx
X Z
v(s) =
m=1
2π m2
2
0 (s + x)2
∞ ∞ Z ∞
1X 1 X 1 (cos 2mπx)dx
= 2 2
− 2 2
s m=1 2π m m=1
2π m 0 (s + x)2
∞ Z ∞
c1 X 4 (sin 2mπx)dx
= − 3
s m=1 (2πm) 0 (s + x)3
c1 c3 c2j +1 1
= + 3 + · · · + 2j00 +1 + O( 2j0 +2 ) (2.43)
s s s s
for large s 6= −m, where
∞
X 1 ς(2)
c1 = = = 1/12, (2.44)
m=1
2π 2 m2 2π 2
∞
X 4 ς(4)
c3 = − 4
= − 4
= −1/360, (2.45)
m=1
(2πm) 4π
c5 = 1/1260, (2.46)
c7 = −1/1680, (2.47)
···

By (2.41) and the first equality of (2.22),

1 λ−1
= 2λ (2π)−s− 2 ×
χλ ( λs + 1/2 − 1
2λ
)
π s 1 √ s 1 s 1 s 1
× cosλ ( ( + 1/2 − ))( 2π)λ e(s−1/2) log( λ +1/2− 2λ )−λ( λ +1/2− 2λ )+λv( λ +1/2− 2λ )
2 λ 2λ
π s 1 √
= (2π)−s (2 cos ( + 1/2 − ))λ 2πe(s−1/2) log s−s+v(s)−(s−1/2) log λ+µλ (s)
2 λ 2λ
14
 √ π s 1
= λ(2πλ)−s Γ(s)eµλ (s) (2 cos ( + 1/2 − ))λ , (2.48)
2 λ 2λ
where
λ−1 λ−1
µλ (s) = (s − 1/2) log(1 + )−
2s 2
s 1
+λv( + 1/2 − ) − v(s)
λ 2λ
1 − λ2 bλ2
= + 2 +··· (2.49)
24s s
and
(λ)
1 − λ2 β2
eµλ (s) = 1 + + 2 +···
24s s
αXλ +1 (λ)
γα
= + Rλ (s). (2.50)
α=0
(s − 1)(s − 2) · · · (s − α)

The first term of the above sum is 1 and

1
Rλ (s) = O(( )αλ +2 ). (2.51)
s − (αλ + 1)

Moreover,
π s 1 πs (λ − 1)π
(2 cos ( + 1/2 − ))λ = 2 cos( + ) + ρλ (s), (2.52)
2 λ 2λ 2 4
where
π (λ − 2)s
ρλ (s) = cλ−2 cos ( + θλ−2 ) + · · · + ρ1λ (s), (2.53)
2 λ

c10λ , λ = 2λ1
ρ1λ (s) = πs
c11λ cos( 2λ + θ1λ ), λ = 2λ1 + 1
′
Therefore, putting s = sλ + 1/2 − 1
2λ
in the integral (2.40) and moving the integral line
return to the original c, we have
αX
λ +1 (λ) 1
γα (2πλ(nx) λ )−s Γ(s)
Z
−1/2 1
− 21 + 2λ
Jnλ (x) = 2λ (nx) ×
α=0
2πi c (s − 1)(s − 2) · · · (s − α)
πs (λ − 1)
× cos( + )ds + Jnλδ (x), (2.54)
2 4

15
where
1 1
λ−1/2 (nx)− 2 + 2λ
Z
1
Jnλδ (x) = Γ(s)(2πλ(nx) λ )−s
2πi c
πs (λ − 1)
×(eµλ (s) ρλ (s) + 2Rλ (s) cos( + ))ds
2 4
1 1 Z
λ−1/2 (nx)− 2 + 2λ αλ +5/4+i∞
=
2πi α +5/4−i∞
Zλ ∞
1 1 αλ +5/4
≪ (nx)− 2 + 2λ − λ |Γ(σ1 + it)|
−∞
π λ−1
×(|ρλ (σ1 + it)| + |Rλ (σ1 + it) cos( (σ1 + it) + )|)dt,
2 4
σ1 = αλ + 5/4.

Since
π
|Γ(σ1 + it) cos( (σ1 + it)(1 − ν0 ) + θ)|
2

(
|t|σ1 −1/2 , ν0 = 0
≪ π|t|ν
σ1 −1/2 − 2 0
|t| e , 0 < ν0 ≤ 1
|t|σ1 −1/2 , ν0 = 0

≪
e−|t|ν0 , 0 < ν0 ≤ 1

for |t| ≥ 1, then by (2.51) and (2.53),

αλ +5/4
Z ∞
1 |t|
−1/2+ 2λ −
Jnλδ (x) ≪ (nx) λ ((|t| + 1)−5/4 + e− λ )dt
−∞
1 αλ +5/4
−1/2+ 2λ −
≪ (nx) λ . (2.55)
Γ(s)
Finally, by (2.54) and (2.55), using (s−1)(s−2)···(s−α)
= Γ(s − α), we obtain that

αX
λ +1 (λ) −α
γα (nx) λ
Z
1 1
−1/2 −1/2+ 2λ
Jnλ (x) = 2λ (nx) (2πλ(nx) λ )−s Γ(s)
α=0
2πi(2πλ)α c

πs (λ − 1)π πα 1 αλ +5/4
× cos( + + )ds + O((nx)−1/2+ 2λ − λ ) (2.56)
2 4 2
It is known that Mellin’s formula:
1 πs
Z
x−s Γ(s) cos( + θ)ds = cos(x + θ), (2.57)
2πi c 2

16
hence
αX
λ +1 (λ) −α
−1/2 1
−1/2+ 2λ γα (nx) λ
Jnλ (x) = 2λ (nx)
α=0
(2πλ)α
1 (λ − 1)π πα 1 αλ +5/4
× cos(2πλ(nx) + λ + ) + O((nx)−1/2+ 2λ − λ )
4 2
αλ (λ)
X γα (nx) λ −α
1
= 2λ−1/2 (nx)−1/2+ 2λ
α=0
(2πλ)α
1 (λ − 1)π πα 1 αλ +1
× cos(2πλ(nx) λ + + ) + O((nx)−1/2+ 2λ − λ ). (2.58)
4 2
By (2.39 ) and (2.58) we obtain (2.9). The proof of Lemma 2.2 is complete.

Lemma 2.3 is the case λ = 2 of Lemma 2.2. To evaluate concretely the first seventh
coefficients γ0 , · · · , γ7 in (2.13), it is sufficient to take λ = 2 in (2.49). We have
1 s
µ(s) = µ2 (s) = (s − 1/2) log(1 + ) − 1/2 + 2v( + 1/4) − v(s)
2s 2
1 1 1 5 1 61
= − − 2
+ 3
+ 4
− 5
− +···
8s 16s 192s 128s 640s 768s6
and the calculation gives
α0
µ2 (s)
X γα
e =1+ + R2 (s),
α=1
(s − 1)(s − 2) · · · (s − α)

where γ1 , · · · , γ6 are (2.13) , R2 (s) satisfies the corresponding (2.51). Proof of Lemma
2.4 is all the same.
It is known that (for example,(3.17) in [3])

x1/4 X √
∆(x) = √ d(n)n−3/4 cos(4π nx − π/4) + O(x1/2+ε N −1/2 ) + O(xε ) (2.59)
π 2 n≤N

for any ε > 0. When x is not an integer, (see (15.24), [3]),

∞
x1/4 X √
∆(x) = √ d(n)n−3/4 (cos(4π nx − π/4)
π 2 n=1
3 √
+ (nx)−1/2 cos(4π nx + π/4)) + O(x−3/4 ). (2.60)
8(4π)

Lemma 2.5 As x ≥ 0.9, 0 < h0 ≪ x, N ≥ N(ε1 ),

h0 Z h0
1 1 X
Z
2
∆((x + h) )dh = d(n) φ(n, (x + h)2 )dh + O(ε1 ) (2.61)
h0 0 h0 n≤N 0

17
holds for any ε1 > 0, where
α0 √
x1/4 n−3/4 X βα cos(4π nx − π/4 + απ/2)
φ(n, x) = √ √
2π α=0 (4π nx)α
α0 +1
+ O(x1/4 n−3/4 (nx)− 2 ) (2.62)
with β0 = 1, β1 = 3/8, · · · , βα0 = · · · for any integer number α0 ≥ 0.
Proof: In fact , as α0 = 0, 1, (2.60) leads to (2.61); for the general α0 ≥ 0, we have
1 (x + h)2s ς 2 (s)
Z
2
∆((x + h) ) = ds,
2πi C s
where the path C is from 1.1 − i∞ to 1.1 − i to 1/2 − i to 1/2 + i to 1.1 + i and to
1.1 + i∞. The proof is the same as Lemma 2.3. ✷

A corollary of Lemma 2.5 is:

Lemma 2.6
Z h2
f (h)∆((x + h)2 )dh
h1
X Z h2
= d(n) f (h)φ(n, (x + h)2 )dh + O(ε1) (2.63)
n≤N h1

holds for any ε1 > 0 as N ≥ N(ε1 ), where f (x) ∈ C 2 [h1 , h2 ], x + h1 ≥ 2, h1 < h2 ≪ x,

φ(n, x) is (2.62).

In the same way, like the proof of Lemma 2.2, for 0.9 < a ≤ x ≤ b < ∞,
Z h0 Z h0
1
··· dh1 · · · dhλ−1 ∆λ ((x + h)λ )
hλ−1
0 0 0
Z h0 Z h0
1 X
= λ−1 dλ (n) ··· dh1 · · · dhλ−1 φλ (n, (x + h)λ )
h0 n≤N 0 0

holds for any ε1 > 0 as N ≥ N(ε1 ), where

h = h1 + · · · + hλ−1 , 0 < h0 < a/2,
1 1 1 1 αλ (λ) 1
x 2 − 2λ n− 2 − 2λ X βα cos(2πλ(nx) λ + (λ − 3)π/4 + απ/2)
φλ (n, x) = √
λπ α=0
(2λπ(nx)1/λ )α
1 1 1 1 αλ +1
+O(x 2 − 2λ n− 2 − 2λ (nx)− λ ), (2.64)
(λ)
β0 = 1,
(λ) λ − 1 1 − λ2
β1 = + ,
2 24
···
for any positive integer number αλ .

18
3 SOME LEMMAS (II)
Lemma 3.1 (see(1.17), [2]). For n ≥ 3,

log n 1
d(n) ≤ exp(O( )) = nε(n) , ε(n) = O( ) (3.1)
log log n log log n

The following Lemma 3.2 is a generalization of Theorem B.1 in [3], also the proof
belong to it.

Lemma 3.2 Let

∞
1 X − π(j+b)2
ψ(V, b, m0 , x) = √ e V e(jx)(j + b)m0 (3.2)
V j=−∞

where ReV > 0, b is real, x = x1 + ix2 , m0 is a negative integer, then

∞
2
X
ψ(V, b, m0 , x) = e−πV (j+x) e(−b(x + j))pm0 (x, j) (3.3)
j=−∞

where Z ∞
2
√
pm0 (x, j) = e−πθ (V i(x + j) + V θ)m0 dθ
−∞
m
[ 20 ]
X  m0  (2ν − 1)!!V ν
= (V i(x + j))m0 −2ν , (3.4)
2ν (2π) ν
ν=0

(−1)!! = 1.

Proof: Let
π(z+b)2
1 e− e(zx)(z + b)m0 dz
Z
V
IN = √
V CN e(z) − 1
where CN is the rectangle with vertices at N + 1/2 ± i, −(N + 1/2) ± i, N is a positive
integer, then
N
1 X − π(j+b)2
IN = √ e V e(jx)(j + b)m0 → ψ(V, b, m0 , x) (N → ∞)
V j=−N

On the other hand, since

±(N +1/2)+i 2 1
e−π(z+b) e(zx)(z + b)m0 dz e−ε0 N dy
Z Z
≪ → 0 (N → ∞), ε0 > 0,
±(N +1/2)−i e(z) − 1 −1 1 + e−2πy

19
then
π(z+b)2
∞−i ∞+i
1 e− e(zx)(z + b)m0 dz
Z Z
V
ψ = √ ( + )
V −∞−i −∞+i e(z) − 1
−1 ∞−i
1
Z
X π(z+b)2
− V
= √ e e(z(x + j))(z + b)m0 dz
V j=−∞ −∞−i
∞ Z ∞+i
1 X π(z+b)2
+√ e− V e(z(x + j))(z + b)m0 dz.
V j=0 −∞+i

Putting z = −b + z ′ , and moving the integral line to (−∞, ∞), we have

∞
X
ψ= e(−b(x + j))Ij ,
j=−∞

where
Z ∞
1 πz 2
Ij = √ e− V e(z(x + j))z m0 dz
V −∞
1 −πv(x+j)2 ∞ − π(z−iV (x+j))2 m0
Z
= √ e e V z dz
V −∞
2
= e−πV (x+j) pm0 (x, j),

pm0 (x, j) is (3.4). The proof is complete. ✷

Lemma 3.3 Let f (x) ∈ C 2 (a − ε, b], ε > 0

X
σ= f (n), (3.5)
a≤n≤b

Mν = max |f (ν) (x)|, ν = 1, 2, (3.6)

a≤n≤b

then
b
M1 + (b − a)M2
X Z
σ= f (x)e(nx)dx + σ0 (a) + σ0 (b) + O( ) (3.7)
−N ≤n≤N a N

for any natural N, where σ0 (X), X = a or b, such that the following condition: If X is
an integer, then
f (X)
σ0 (X) = ; (3.8)
2
if X is not an integer, then

1/{X} + 1/(1 − {X})

σ0 (X) ≪ . (3.9)
N

20
Proof: We have
b b
1
X Z Z
σ = lim f (n) = lim f (x)d(x + − {x}) = f (x)dx + I1 , (3.10)
ε→0+
a−ε≤n≤b
ε→0+ a−ε 2 a

where
b b
1 1
Z
I1 = lim
ε→0+ a−ε
f (x)d( − {x}) = lim f (x)( − {x})
2 ε→0+ 2
| a−ε
+ I2 , (3.11)

with
b
1
Z
I2 = − f ′ (x)( − {x})dx
a 2
Z b Z x
1
= − f ′ (x)( ( − {y})dy)′dx
a 0 2
Z x b Z b Z x
1 1
′
= −f (x) ( − {y})dy a +
2
′′
f (x)( ( − {y})dy)dx.
2
|
0 a 0

Using
x ∞
1 X 1 − cos 2nπx
Z
( − {y})dy = , (3.12)
0 2 n=1
2π 2 n2
we have
N b b N
1 − cos 2nπx 1 − cos 2nπx M1 + (b − a)M2
Z
|
X X
′ ′′
I2 = −f (x) + f (x) dx + O( )
n=1
2π 2 n2 a
a n=1
2π 2 n2 N

N Z b N b
f (x) sin 2nπx M1 + (b − a)M2
|
X X
=2 f (x) cos 2nπxdx − + O( ). (3.13)
n=1 a n=1
πn a N
Since 2 cos 2πx = e(x) + e(−x), then (3.7) is obtained by (3.10), ( 3.11) and (3.13),
where
N
1 X sin 2nπa
σ0 (a) = −f (a)( lim ( − {a − ε}) − ), (3.14)
ε→0+ 2 πn
n=1
N
1 X sin 2nπb
σ0 (b) = f (b)( − {b} − ). (3.15)
2 n=1
πn

Let X = a or b, clearly, if X is an integer, then σ0 (X) = 12 f (X). If {X} = 12 , then

21
σ0 (X) is vanishing. Suppose 0 < {X} < 12 . We have
N ∞
1 X sin 2nπX X sin 2nπ{X}
− {X} − =
2 n=1
πn n=N +1
πn
1 N1 1 X
Z
= lim d sin 2nπ{X}
N1 →∞ π N +1 u
n≤u
Z ∞
1 X 1 X
≪ ( sin 2nπ{X} + 2
( sin 2nπ{X}))du
N n≤N +1 N +1 u n≤u

1 1 1
≪ ( + ).
N {X} 1 − {X}
1
The case 2
< {X} < 1 is the same. The proof is complete. ✷

Lemma 3.4 Let

X √
H = H(X, α, β, a, b) = d(n)n−α cos(4π nX + β), (3.16)
a<n≤b

where 0.9 < a < b ≤ X(log X)C , C ≥ 0; α, β are real, X is large positive number, then

H ≪ (a1/4−α + b1/4−α )X 1/4+ε , (3.17)

where
1
ε = ε(X) = O( ). (3.18)
log log X
1
Proof: Let h0 = 4
and

1′
Z h0 X √
H = dh d(n)n−α cos(4π nX + β), (3.19)
h0 0 √ √ √
a+h< n≤ b+h

then from Lemma 3.1

′ 1
Z h0 X X √
H−H = dh( − )d(n)n−α cos(4π nX + β)
h0 0 √ √ √ √ √ √
a< n≤ b a+h< n≤ b+h
h0
1
Z X X
≪ ( + )d(n)n−α
h0 0 √ √ √ √
| n− a|≤h | n− b|≤h
1 1
≪ a 2 −α+ε + b 2 −α+ε ,

where ε = ε(X) is (3.18). So that

1
Z h0 X √ 1 1
H= dh d(n)n−α cos(4π nX + β) + O(a 2 −α+ε + b 2 −α+ε ) (3.20)
h0 0 √ √ √
a+h< n≤ b+h

22
 √
Taking f (n) = n−α cos(4π nX + β) in Lemma 2.3, we have
1 1
H = Hl + H∆N + O(ε1 ) + O(a 2 −α+ε + b 2 −α+ε ) (3.21)

holds for any ε1 > 0 as N large, where

√
h0 ( b+h)2 √
1
Z Z
Hl = dh √
u−α (log u + 2γ) cos(4π uX + β)du
h0 0 ( a+h)2
h0 Z √b+h √
2
Z
= dh √
u1−2α (log u2 + 2γ) cos(4π Xu + β)du
h0 0 a+h

1 1 √
≪ (a 2 −α+ε + b 2 −α+ε )/ X, (3.22)
√
X d(n) Z h0 Z ( b+h)2 √
H∆N = dh √ u−α cos(4π uX + β)Θ(nu)du
n≤N
h0 0 ( a+h)2

1
√ X − 41
X γν
=2 2 d(n)n √ jnν + δ∆N , (3.23)
n≤N ν=0
(4π n)ν
√
h0 ( b+h)2
1 X u−α−1
Z Z
1
δ∆N ≪ d(n)n− 4 dh √
du ≪ a−α + b−α , (3.24)
h0 n≤N 0 ( a+h)2 n
√
1
Z h0 Z b+h
1 √ √ π πν
jnν = dh √
u 2 −2α−ν cos 4π( Xu + β) cos(4π nu + + )du
h0 0 a+h 4 2
+ −
= jnν + jnν , (3.25)
√
± 1
Z h0 Z b+h
1 √ √
jnν = dh √
u 2 −2α−ν cos(4π( n ± X)u + θ± )du
2h0 0 a+h
h0 1 √ √ √
1 (u + h) 2 −2α−ν sin(4π( n ± X)(u + h) + θ± ) b
Z
=
2h0
√ √
4π( n ± X)
| √
a
0

Z h0 Z √b √
c dh − 12 −2α−ν
√
− √ √ √
(u + h) sin(4π( n ± X)(u + h) + θ± )du (3.26)
2h0 0 n± X a
√ √
for n 6= X. If n > 2X, then integrating by parts for h, we have
1 1 1 1
± a 4 −α + b 4 −α a 4 −α + b 4 −α
jnν ≪ √ √ ≪ . (3.27)
( n ± X)2 n

Therefore, it follows from (3.23), (3.24) and (3.27) that

1
√ X − 14
X γν + − 1 1
H∆N =2 2 d(n)n √ ν (jnν + jnν ) + O(a 4 −α + b 4 −α ). (3.28)
n≤2X ν=0
(4π n)

23
If |n − X| ≤ 2, then by (3.25),
√
Z b+1
1 3 3
±
jnν ≪ √
u 2 −2α du ≪ a 4 −α + b 4 −α . (3.29)
a

If |n − X| > 2, then by (3.26),

± 1 1 1
jnν ≪ √ √ (a 4 −α + b 4 −α ). (3.30)
| n − X|
Hence, by (3.27), (3.28), (3.29) and Lemma 3.1 we obtain
1
X d(n)n− 4 X 1
H∆N ≪ √ √ (a1/4−α + b1/4−α ) + d(n)n− 4 (a3/4−α + b3/4−α )
n≤2X,|n−X|>2
| n − X| |n−X|≤2

X 1 a1/4−α + b1/4−α
≪ d(n)n− 4 √
X
n≤ 21 X

1
X 1 1
+(a1/4−α + b1/4−α )X 4 +ε(X) + (a3/4−α + b3/4−α )X − 4 +ε(X)
|n − X|
X/2<n≤2X,|n−X|>2
1
≪ (a1/4−α + b1/4−α )X 4 +ε(X) (3.31)
C
for a < b ≪ (log X) . (3.20), (3.21), (3.22) and (3.31) implicate (3.17). The proof is
complete. ✷
Lemma 3.5 Suppose a, b, α, β, X be as Lemma 3.4,
X √
H= d(n)n−α f (n) cos(4π nX + β), f (u) ≪ M,
a<n≤b

M
f ′ (u) ≪ √ , u ∈ [a, b + 1], (3.32)
Xu
then
1
H ≪ M(a1/4−α + b1/4−α )X 4 +ε(X) . (3.33)
Proof: In fact, by Lemma 3.4,
Z b X √
H = f (u)d d(n)n−α f (n) cos(4π nX + β)
a a<n≤u
X √
≪ |f (b)|| d(n)n−α f (n) cos(4π nX + β)|
a<n≤b
Z b X √
+ |f ′(u)|| d(n)n−α f (n) cos(4π nX + β)|du
a a<n≤u
b
du
Z
1
1/4−α 1/4−α +ε(X)
≪ M(a +b )X 4 (1 + √ )
a Xu
1
1/4−α 1/4−α +ε(X)
≪ M(a +b )X 4 .

24
The proof is complete. ✷

Let
η = η1 + · · · + ηK ,
define η1 =ν0 ηK =ν0
f (η) |ηk
= f (η1 + · · · + ηK ) | η1 =0
··· | ηK =0
(3.34)
Also we define η1 =ν0
f (η1 )| |η1 =0
| = |f (ν0 )| + |f (0)|
generally,
|
X X
f (η)| ηK
|= ··· |f (η)| (3.35)
η1 =0,ν0 ηK =0,ν0

It is obvious that if f (η) ≪ M, then

f (η) | ηK
≪ f (η)| |ηK
| ≪ M2K ; (3.36)

if f (k) (η) ≪ M, 0 ≤ k ≤ K, then

Z ν0 Z ν0 ηk+1 =ν0 ηK =ν0
f (η) | ηK
= ( ··· f (k) (η)dη1 · · · dηK ) η =0 · · · η =0
k+1 K
| |
0
Z 0ν0 Z ν0
≪ M2K−k ··· dη1 · · · dηK = M2K−k ν0k . (3.37)
0 0

Lemma 3.6 Suppose f (z) is analytic for |z| ≤ √1 , and |f (z)| ≤ M, then
L
η1 =ν0 ηK =ν0
f (η) |
ηk
= f (η1 + · · · + ηK ) | η1 =0
··· | ηK =0
≪ MU −99 (3.38)
C0 L
for large U, L ≍ log U, 0 < ν0 ≤ L−2 , K = [ log L
], C0 ≥ 200.
Proof: Clearly,
1
0 ≤ η1 + · · · + ηK ≤ ν0 K ≪ . (3.39)
L log L
Moreover,
Z ν0 Z ν0
f (η) |ηK
= ··· f (k) (η)dη1 · · · dηK
Z0 ν0 Z0 ν0
K! f (z)dz
Z
= ··· dη1 · · · dηK
0 0 2πi |z|= √1 (z − η)K+1
L

M|dz|
Z
≪ K!ν0K 1 1 K+1
|z|= √1 ( √L − L log L )
L
√
≪ K!L−2K M(2 L)K ≤ M(2KL−3/2 )K
L3/2
= Me−K log 2K

200L
≪ Mexp(− (log L1/2 + O(log log L)))
log L
≪ MU −99 ,

25
for L ≍ log U. The proof is complete. ✷

4 SOME LEMMAS (III)

Lemma 4.1 Let √
√ − 4π(
√
n+λ( U +η+ν))2
Gλ ( n, η) = e A2 , (4.1)
√
Wλ (η) = e(−λ( U + η + ν)2 ), (4.2)
2 √ √
X
− 14
√ X γα (−i)α e(−2 n( U + η + ν) − 1/8)
Sλ (η) = Wλ (η) d(n)n Gλ ( n, η) √ √ √ ,
√ √ α=0 (4π n( n + U + η + ν))α
n≤ U
(4.3)
X
− 41
√ √ √
Sλ∗ (η) = Wλ (η) d(n)n Gλ ( n, η)e(−2 n( U + η + ν) − 1/8), (4.4)
√ √ √
U < n≤(λ0 −λ) U

where ν = O(L−1 ) is real, A is (1.10), η is (1.15),

1 ≤ λ0 ≪ L, 0 ≤ λ ≤ λ0 − 1. (4.5)

Then
λX
0 −1
1
S0∗ (η) = Sλ (η) + O(U −3/4+ε ), ε = O( ). (4.6)
λ=1
log L
(note that Oη is (1.28).)

Proof: Let
1
U = [X], 0 < h ≤ h0 = √ , (4.7)
U L2
then √
0 < ( U + h)2 − U < L−3/2 , (4.8)
√ 2
{ U } = {U} = 0,
√
Hence there doesn’t exist any integer in the interval (U, ( U + h)2 ], that is
X
= 0.
√ √ √
U < n≤ U +h

In the same way, X

= 0.
√ √ √
(λ0 −λ) U < n≤(λ0 −λ) U +h

Noticing e(n) = 1, we have

e(−1/8)Wλ (η)
Sλ∗ (η) =
h0

26
 h0 √ √ √
Z X
× dh d(n)n−1/4 Gλ ( n, η)e(n − 2 n( U + η + ν)). (4.9)
0 √ √ √
U +h< n≤(λ0 −λ) U +h

By Lemma 2.3
√
X e(−1/8)Wλ (η)d(n) Z h0 Z ((λ0 −λ) U+h)2
Sλ∗ (η) = dh √
n≤N
h 0 0 ( U+h)2
√ √ √
x−1/4 Gλ ( n, η)Θ(nx)e(x − 2 x( U + η + ν))dx
+δλ (η) + O(ε1 ), (4.10)

holds for any ε1 > 0, as N ≥ N(ε1 ), where Θ(nx) is (2.12),

√
h0 ((λ0 −λ) U +h)2 √
e(−1/8)Wλ (η)
Z Z
δλ (η) = dh √ x−1/4 Gλ ( n, η)
h0 0 ( U +h)2
√ √
×e(x − 2 x( U + η + ν))(log x + 2γ)dx
√
h0 (λ0 −λ) U +h √
2e(−1/8)Wλ (η)
Z Z
= dh √ x1/2 Gλ ( n, η)
h0 0 U +h
√ √
× e(x2 − 2 x( U + η + ν))(log x2 + 2γ)dx. (4.11)
Moving the integral line of the inner integral we obtain
Z √U +h+z0 Z (λ0 −λ)√U +h+z0 Z (λ0 −λ)√U +h
2e(−1/8)Wλ (η) h0
Z
δλ (η) = dh( √ + √ + √
)
h0 0 U +h U +h+z0 (λ0 −λ) U+h+z0
1 √ √
×z 2 Gλ ( n, η)(log z 2 + 2γ)e(z 2 − 2z( U + η + ν))dz

2e(−1/8)Wλ (η) h0
Z
= dh(I1h (η) + I2h (η) + I3h (η)), say (4.12)
h0 0
where
1
z0 = Le( ). (4.13)
8
Since
√
4π(z+λ( U +η+ν))2
−
Gλ (z, η) = e A2

∞
√
2θ(z + λ( U + η + ν))
Z
−πθ 2
= e e( )dθ
−∞ A
Z L √
−πθ 2 2θλ( U + ν) 2(z + λη)θ 2
= e e( )e( )dθ + O(e−L ), (4.14)
−L A A
then by (4.12),
L
√
2θλ( U + ν)
Z
−πθ 2 2
Wλ (η)I1h (η) = e e( )J1θ (η) + O(e−L ), (4.15)
−L A

27
where
√
Z U +h+z0 √ 2(z + λη)θ
J1θ (η) = Wλ (η) √
fθ (z)e(z 2 − 2z( U + η + ν) + )dz
U +h A
Z z0 √ √ √ √ θ
= fθ ( U + h+ z)e(( U + h+ z)2 + 2( U + h+ z)(− U −ν + ))ω(z, η)dz, (4.16)
0 A
fθ (z) = z 1/2 (log z 2 + 2γ), (4.17)
√ λθ
ω(z, η) = e(2η(− U − h − z + ))Wλ (η)
A
√ λθ √
= e(2η(− U − h − z + ) − λ( U + η + ν)2 )
A
√
= e(−λ( U + ν)2 )gzθ (η), (4.18)
√ λθ
gzθ (η) = e(−2(λ + 1)η U + 2η(−h − z − νλ + ) − λη 2 )
A
λθ
= e(2η(−h − z + ) − λη 2 ). (4.19)
A
The last equality is given by
√ √
e(−2(λ + 1)η U) = e(η U ) = 1 (4.20)
√
where η = η1 + · · · + ηK , ηj = 0 or k1 / U , j = 1, 2 · · · , K. Thus,
λθ 2
√
gzθ (w) ≪ e4π|w|(| A |+|h|+|z|+λ|w| ) ≪ e8π L
(4.21)
√
for |w| ≤ √1 , |z| ≤ L, λ ≪ L, |θ| ≪ L, h = ( √U1 L ), A = C0 UL. By Lemma 3.6, we
L
have √
gzθ (η)|ηK ≪ e8π L
U −99 ≪ U −98
(4.22)
√
for ν0 = k1 / U ≪ L−2 . It follows from (4.15), (4.16), (4.17), (4.18), (4.19) and (4.22)
that
Wλ (η)I1h (η)|η ≪ U −90 . (4.23)
In the same way,
Wλ (η)I3h (η)|η ≪ U −90 . (4.24)
By (4.13),
√
Z (λ0 −λ) U +h+z0 √ √
Wλ (η)I2h (η) = Wλ (η) √
fθ (z)Gλ ( n, η)e(z 2 − 2z( U + η + ν))dz
U +h+z0
√
Z (λ0 −λ) U +h √ √
≪ √
UL|e((x + z0 )2 − 2(z0 + x)( U + η + ν)))|dx
U +h
√
√ Z (λ0 −λ) U +h √
= UL √
|e(z02 + 2z0 (x − ( U + η + ν)))|dx
U +h
√
√ Z (λ0 −λ) U +h
2 −2
√ √
= UL √
e−2πL 2π(x−( U+η+ν))
U +h

28
 √ 2
≪ ULe−L = Oη (U −90 ) (4.25)
for z0 = Le( 1z ). It follows from (4.12), (4.23), (4.24) and (4.25) that

δλ (η) = Oη (U −90 ). (4.26)

Taking α0 = 2 in (2.12), we have
2
√ √ X γα √ √ π απ √
Θ(nx) = 2( nx)−1/2 α
( nx) −α
cos(4π nx+ + )+O(( nx)−7/2 ). (4.27)
α=0
(4π) 4 2

By (4.26) and (4.27) and putting x = x21 in the integral (4.10), we have
3
Sλ∗ (η) = &+ −
λ (η) + &λ (η) + O(ε1 ) + Oη (U
−2
), (4.28)
where
2
1 1 X √ − 41
X ±i
&±
= 2e(− ± )
λ (η) d(n)n γα ( √ )α jnα±
(η), (4.29)
8 8 n≤N α=0
4π n
Z h0 Z (λ0 −λ)√U +h √ 2
√ √
± Wλ (η) G λ ( n, η)e(x + 2x(± n − ( U + η + ν)))dx
jnα (η) = dh √ α
.
h0 0 U +h x
(4.30)
As n > n0 > 2U, the integration by parts gives
±
jnα (η) =

h0 √ √
Wλ (η) Gλ (x + h, η)e((x + h)2 + 2(x + h)(± n − U − η − ν)) (λ0 −λ)√U
Z
= √ √ | √U
h0 0 (x + h)α 4πi(x + h ± n − U − η − ν)
√ √
Wλ (η) h0 Gλ (x + h, η) ′ e((x + h)2 + 2(x + h)(± n − U − η − ν))dx
Z
− ( ) √ √
4πih0 0 (x + h)α x (x + h ± n − U − η − ν)
U
≪ ,
n
for h0 = √1 , where the last inequality is given by integrating by parts for h. Since
UL2

X UL
d(n)n−5/4 U ≪ 1/4
,
n>n0 n0

then taking n0 = U 20 , by (4.29) we obtain

2
√ 1 1 X − 14
X ±i 3
&±
λ (η) = 2e(− ± ) d(n)n γα ( √ )α jnα
±
(η) + Oη (U − 2 ). (4.31)
8 8 20 α=0
4π n
n≤U
√ √ −
As n > (λ0 − λ − 1) U , moving the integral of jnα (η), we obtain
Z h0 Z √U +h−z0 Z (λ0 −λ)√U +h−z0 Z (λ0 −λ)√U +h
− 1
jnα (η) = dh( √ + √ + √
h0 0 U+h U +h−z0 (λ0 −λ) U +h−z0

29
 √ √
e(z 2 − 2z( U + n + ν))f (z, η)dz
×
zα
Z h0
1 − − −
= dh(I1nh (η) + I2nh (η) + I3nh (η)), say (4.32)
h0 0
where z0 = Le( 18 ) and
f (z, η) = Wλ (η)Gλ (z, η)e(−2zη). (4.33)
For z = x − ρe( 81 ), we have
√ √ √ 2
√ √ √ 2
|e(z 2 − 2 n( U + n + ν)z)| = e−2πρ −2 2πρ( U + n+ν−x) ≪ e−2πρ (4.34)
√ √ √
holds for x ≤ (λ0√− λ) U + h, n > (λ0 − λ − 1) U , h, ν √ = O(L−1), 0 ≤ ρ ≤ L. As
− −
z ∈ I1nh (η), z = U + h − ρe( 8 ); as z ∈ I3nh (η), z = (λ0 − λ) U + h − ρe( 81 ). Repeating
1

the proof of (4.26), we have

f (z, η) = Oη (U −90 ), z ∈ I1nh

− −
, I3nh . (4.35)

It follows from (4.34), (4.35 ) and z −α ≤ 1 that

− −
I1nh (η) + I3nh (η) = Oη (U −90 ). (4.36)
−
As z ∈ I2nh (η), z = x − Le( 81 ). Clearly, f (z, η) ≪ 1, z = O(L). Hence, by (4.33) and the
−
definition (4.32) of I2nh (η) we have
√
Z (λ0 −λ) U+h
− 2 2
I2nh (η) ≪ √
e−2πL dx ≪ e−L ≪η U −90 . (4.37)
U+h

Moreover, (4.32), (4.36) and (4.37) deduce that

−
jnα (η) = Oη (U −90 ) (4.38)
√ √ +
for n > (λ0 − λ − 1) U . Moving the integral line of jnα (η), we obtain
√ √ √
h0 U +h+z0 (λ0 −λ) U +h+z0 (λ0 −λ) U +h
1
Z Z Z Z
+
jnα (η) = dh( √
+ √
+ √
)
h0 0 U +h U +h+z0 (λ0 −λ) U +h+z0
√ √
e(z 2 + 2(− U + n − ν)z)f (z, η)dz
×
zα
1 h0
Z
+ + +
= dh(I1nh (η) + I2nh (η) + I3nh (η)), say (4.39)
h0 0
If z = x + ρe( 81 ), then
√ √ 1 √ √
|e(z 2 + 2( U + n − ν)z)| = |e(iρ2 + 2ρe( )(− U + n − ν + x))|
√
8√ √
−2πρ2 −2 2ρ(− U + n−ν+x)
= e

30
 2
≪ e−2πρ (4.40)
√
for x ≥ U + h, n ≥ 1, and ν, h = O(L−1 ), 0 ≤ ρ ≤ L.
Similar to the proof of (4.38), we have
+
jnα (η) = Oη (U −90 ) (4.41)

for n ≥ 1. It follows from (4.28), (4.31), (4.38) and (4.41) that

2
√ 1 X
− 14
X −i α − 3
Sλ∗ (η) = 2e(− ) d(n)n γα ( √ ) jnα (η) + δ(η) + Oη (U − 2 ),
4 √ √
α=0
4π n
n≤(λ0 −λ−1) U
(4.42)
where
1 3
X
δ(η) ≪η d(n)n− 4 U −90 ≪η U − 2 . (4.43)
n≤U 20
−
Moving the inner integral line of jnα (η), we have
√ √ √
h0 U +h−z0 xn −z0 xn +z0 (λ0 −λ) U +h+z0 (λ0 −λ) U +h
1
Z Z Z Z Z Z
−
jnα (η) = dh( √ + √ + + + √ )
h0 0 U +h U +h−z0 xn −z0 xn +z0 (λ0 −λ) U+h+z0

Wλ (η)e((z − xn )2 − x2n )dz

×
zα
Z h0
1 − − + +
= dh(I−1nh (η) + I−2nh (η) + I0 (η) + I1nh (η) + I2nh (η)), say, (4.44)
h0 0
√ √
for n ≤ (λ0 − λ − 1) U, where z0 = Le( 18 ),
√ √
xn = U + n + ν + η. (4.45)

√
xn + z0 (λ0 − λ) U + h + z0
✲

√ π
U +h 4
√
xn (λ0 − λ) U + h

√ ✲
U + h − z0 xn − z0
Fig.4.1

Repeating the proof of (4.38) and (4.41), we have

− − + +
I−2nh (η) + I−1nh (η) + I1nh (η) + I2nh (η) ≪η U −90 . (4.46)

31
Hence
h0
1
Z
−
jnα (η) = I0 (η)dh + Oη (U −90 ) = I0 (η) + Oη (U −90 )
h0 0
Z xn +z0
2 Gλ (z, η)e((z − xn )2 )dz
= Wλ (η)e(−xn ) + Oη (U −90 )
xn −z0 zα
2
1 L
Gλ (xn + ρe( 18 ), η)e−2πρ dρ
Z
= Wλ (η)e(−x2n + ) + Oη (U −90 ). (4.47)
8 −L (xn + ρe( 81 ))α
By (4.1) and (4.2),
√
1 4π(xn +ρe( 1
8 )+λ( U +η+ν))2
Gλ (xn + ρe( ), η) = e− A2
8
√ ρ4
= Gλ+1 ( n, η)(1 + C1 (n)ρ + C2 (n)ρ2 + C3 (n)ρ3 + O( 2 )), (4.48)
U
1 1
C1 (n) ≤ √ , C3 (n) ≪ √ 3 ,
U U
1 1
C2 (n) ≪ , C2′ (x) ≪ √
U U xU
and
√ √ √
Wλ (η)e(−x2n ) = e(−(λ + 1)( U + η + ν)2 − 2 n( U + η + ν) − n)
√ √
= Wλ+1 (η)e(−2 n( U + η + ν))

for e(−n) = 1. Therefore,

√ √ √
− e( 18 )Wλ+1 (η)Gλ+1 ( n, η)e(−2 n( U + η + ν))
jnα (η) = √ √
( U + n + ν + η)α
Z L
ρ4 α|ρ|
× e(−2πρ2 )(1 + C1 (n)ρ + C2 (n)ρ2 + C3 (n)ρ3 + O( 2 ))(1 + O( √ ))
−L U ( U )α
√ √ √
e( 18 )Wλ+1 (η)Gλ+1 ( n, η)e(−2 n( U + η + ν)) 1 α
= √ √ √ (1 + C20 (n) + O( 2 ) + √ )
2( U + n + ν + η) α U ( U )α
(4.49)
where
1 ′ 1
C20 (n) ≪ , C20 (x) ≪ √ . (4.50)
U U xU
It follows from (4.42) and (4.49) that
1 X 1 √
Sλ∗ (η) = e(− )Wλ+1 (η) d(n)n− 4 Gλ+1 ( n, η)
8 √ √
n≤(λ0 −λ−1) U

32
 2
X (−i)α √ √ 3
× γα √ √ e(−2 n( U + η + ν)) + δ1 + δ2 + Oη (U − 2 ), (4.51)
α=0 (4π n( U + η + ν))α
where
2
X
− 14
X 1 1 α
δ1 ≪ d(n)n √ √ ( 2+ √ ))
√ √
α=0 ( n + U) U
α ( U)α
n≤(λ0 −λ−1) U

1
≪ U −3/4 ≪η U −3/4+ε , 0 < ε = O( ), (4.52)
log L

1 X √ 1 √
δ2 = e(− )Wλ+1 (η) C20 ( n)d(n)n− 4 Gλ+1 ( n, η)
8 √ √
n≤(λ0 −λ−1) U
2
X (−i)α √ √
× γα √ √ √ e(−2 n( U + η + ν))
α=0 (4π n( U + n + η + ν))α
1
= e(− )Wλ+1 (η)
8
X √ 1 √ √
× C20 ( n)d(n)n− 4 e(−2 n( U + η + ν)) + O(U −3/4 ). (4.53)
√ √
n≤(λ0 −λ−1) U

By (4.51) and Lemma 3.4,

1
δ2 ≪ U −3/4+ε ≪η U −3/4+ε , 0 < ε = O( ). (4.54)
log L
The last inequality has been used the explain (1.32). Thus, by (4.51), if λ = λ0 − 2,
then
∗
Sλ−2 (η) = Sλ0 −1 (η) + δλ0 −1 (η) + Oη (U −3/4+ε ); (4.55)
if 0 ≤ λ ≤ λ0 − 1, then
Sλ∗ (η) = Sλ+1 (η) + Sλ+1
∗
(η) + δλ+1 (η) + Oη (U −3/4+ε ), (4.56)
where 0 < ε = O( log1 L ) and
δλ+1 (η) =
2 √ √
1 X −i X
− 41 − α e(−2 n( U + η + ν))
= e(− )Wλ+1 (η) γ α ( )α d(n)n 2 √ √ .
8 α=1
4π √ √ √ ( U + n + ν + η)α
U < n≤(λ0 −λ−1) U
(4.57)
1
Taking f (n) = √ √
( U + n+ν+η)α
in Lemma 3.5, we know that the inner sum is ≪
−3/4+ε
U , α ≥ 1, that is
1
δλ+1 (η) ≪ U −3/4+ε , 0 < ε = O( ). (4.58)
log L
Finally, (4.6) follows from (4.55), (4.56) and (4.58) and the proof of Lemma 4.1 is
complete. ✷

33
Lemma 4.2 Let
X 1 4πn √ √ 1
S 1 (η) = d(n)n− 4 e− A2 e(−2 n( U + η + ν) − )
2
√ √ √ 8
U
2
< n≤λ0 U

X 1 4πn √ √ 1
= S0∗ (η) + d(n)n− 4 e− A2 e(−2 n( U + η + ν) − ), (4.59)
√ √ √ 8
U
2
< n≤ U

then
λ0 λ0
X X 1
S 1 (η) = Sλ+ (η) + Sλ− (η) + Oη (U −3/4+ε ), 0 < ε = O( ), (4.60)
2
λ=1 λ=1
log L

where S0∗ (η), η, ν are as Lemma 4.1, and

X 1 √
Sλ+ (η) = Wλ (η) d(n)n− 4 Gλ ( n, η)
√ √
n≤ 2U

2 √ √
(−i)α e(−2 n( U + η + ν) − 81 )
X
× γα √ √ √ , (4.61)
α=0 (4π n( U + n + η + ν))α
X 1 √
Sλ− (η) = Wλ (η) d(n)n− 4 Gλ (− n, η)
√ √
n≤ 2U

2 √ √
X (i)α e(2 n( U + η + ν) + 81 )
× γα √ √ √ α
. (4.62)
α=0 (4π n( U − n + η + ν))

Proof: Sinc
X 1 √
Sλ (η) = Wλ (η) d(n)n− 4 Gλ ( n, η)
√ √
n≤ U
2 √ √
X (−i)α e(−2 n( U + η + ν) − 18 )
× γα √ √ √
α=0
(4π n( U + n + η + ν))α
X X
= Wλ (η)( + ),
√ √ √ √ √
n≤ 2U 2
U
< n≤ U

then by Lemma 4.1,

λ0
X λX
0 −1

S 1 (η) = Sλ+ (η) + Hλ (η) + Oη (U −3/4+ε ), (4.63)

2
λ=1 λ=0

34
 2
where Sλ+0 (η) ≪ e−L is used, and
X 1 √
Hλ (η) = Wλ (η) d(n)n− 4 Gλ ( n, η)
√ √ √
U
2
< n≤ U
2 √ √
X (−i)α e(−2 n( U + η + ν) − 18 )
× γα √ √ √
α=0 (4π n( U + n + η + ν))α
X 1 √ √ √ 1
= Wλ (η) d(n)n− 4 Gλ ( n, η)e(−2 n( U + η + ν) − )
√ √ √ 8
U
2
< n≤ U

+ O(U −3/4+ε ), (4.64)

the O-term is given by the proof of (4.58).
The following proof is similar to Lemma 4.1. Since
√ √ √
U 2 3 U 2 U 2 1
0 ≤ {( ) } ≤ ,( + h) − ) ≤ 3/2
2 4 2 2 L
√ √
for 0 < h ≤ √U1L2 , then we have the corresponding (4.9). If replace U by U /2 in
(4.18), we have also √
U
e(±2η( )) = 1 (4.65)
2
√
for η = η1 + · · · + ηK , ηj = 0 or k1 / U. Similarly, we obtain that
2
√ 1 X 1
X i
Hλ (η) = 2e( ) d(n)n− 4 γα ( )α jnα
+
(η) + Oη (U −3/2 ), (4.66)
8 √ √ α=0
4π
U
n≤ 2
√ √
+
(different from (4.42), present main term is jnα , because x ∈ ( 2U , U + h)) where
Z h0 Z √U +h √ √
+ W λ (η) Gλ (x, η)e(x2 + 2x( n − U − η − ν))dx
jnα (η) = dh √
h0 0 2
U
+h xα
Z √U +h
Wλ (η)e(−x2n ) h0 Gλ (x, η)e((x − xn )2 )dx
Z
= dh √ , (4.67)
h0 0 2
U
+h xα
√ √
xn = U + η + ν − n. (4.68)
Repeating the proof of (4.51) we see that
2
1 X 1
X i 1
Hλ (η) = e( )Wλ (η) d(n)n− 4 e(−x2n )Gλ (xn , η) γ α ( )α √
8 √ √
α=0
4π (4π nxn )α
U
n≤ 2

+Oη (U −3/4+ε ). (4.69)

Since √ √ √
− 4π( U +η+ν− n+λ( U +η+ν))2 √
Gλ (xn , η) = e A2 = Gλ+1 (− n, η) (4.70)

35
 √ √ √
Wλ (η)e(−x2n ) = e(−(λ + 1)( U + η + ν)2 + 2 n( U + η + ν) + n)
√ √
= Wλ+1 (η)e(2 n( U + η + ν)), (4.71)
then by (4.69),
−
Hλ (η) = Sλ+1 (η) + O(U −3/4+ε ), (4.72)
where ε = O( log1 L ) and Sλ− (η) is (4.62). Thus, we obtain (4.60) immediately by (4.63)
and (4.72). The proof of Lemma 4.2 is complete. ✷

Lemma 4.3 Let

λ0 = b0 = [C02 L], C0 ≥ 200, (4.73)
1
√
where η is (1.15). If 8
≤ {2 U ν} ≤ 83 , then
X 1 4πn √ √ 1
ReS 1 (η) = −Re d(n)n− 4 e− A2 e(−2 n( U + η + ν) − )
2
√ √ 8
n≤ 2U

2
X −i
× γα ( √ √ √ )α + Oη (U −3/4+ε ), (4.74)
α=0
4π n( n + U + η + ν)
where S 1 (η) is (4.59), ε = O( log1 L ).
2
√
Proof: By the definitions (4.1) and (4.2) of Gλ ( n, η) and Wλ (η),
√ 4πn
G0 ( n, η) = e− A2 , (4.75)
W0 (η) = 1. (4.76)
By Lemma 4.2,
b0
X 1
S 1 (η) = (Sλ+ (η) + Sλ− (η)) + Oη (U −3/4+ε ), ε = O( ). (4.77)
2 log L
λ=1

Since
1 1 √ 1−α
√ √ √ α = √ √ √ α + ( n) δα (n), (4.78)
( n( U + η + ν − n)) ( n( U + η + ν + n))
where
1 1 1 α
δα (n) = √ ( √ √ α− √ √ α ) ≪ √ α+1 ,(4.79)
n ( U + η + ν − n) ( U + η + ν + n) ( U)
α
δα′ (x) ≪ √ √ , (4.80)
xU ( U )α+1
moreover, δα (n) = 0 for α = 0. By Lemma 3.5 ,
2
X i α X 1 √ √ √ √
γα ( ) d(n)n− 4 ( n)1−α δα (n)Gλ (− n, η)e(2 n( U + η + ν)) ≪ U −3/4+ε .
α=1
4π √ √
U
n≤ 2
(4.81)

36
This and (4.62) give
2
X
− 41
√ X √ √ 1
Sλ− (η) = Wλ (η) d(n)n Gλ (− n, η) γα e(2 n( U + η + ν) + )
√ √
α=0
8
n≤ 2U

i 1
×( √ √ √ )α + O(U −3/4+ε ), ε = O( ). (4.82)
4π n( n + U + η + ν) log L

It follows from (4.77), (4.82) and (4.61) that

2
X X 1 1
S 1 (η) + S 1 (η) = d(n)n− 4 γα √ √ √
2 2
√ √ (4π n( n + U + η + ν))α
n≤ U α=0
√ √ 1
×((−i)α e(−2 n( U + η + ν) − )N(n)
8
√ √ 1
+(−i)α e(−2 n( U + η + ν) − )N(n)) + Oη (U −3/4+ε ),(4.83)
8
where
b0 b0
X √ X √
N(n) = Wλ (η)Gλ ( n, η) + Wλ (η)Gλ (− n, η). (4.84)
λ=1 λ=1

Putting λ = −λ′ in the second sum, note

√
W−λ (η) = e(λ( U + η + ν)) = Wλ (η), (4.85)
√ √
√ − 4π(− n−λ( 2 U +η+ν))
2 √
G−λ (− n, η) = e A = Gλ ( n, η), (4.86)
we have
b0
X √
N(n) = Wλ (η)Gλ ( n, η)
λ=−b0 ,λ6=0
b0
√ X √ √
4π( n+λ( U +η+ν))2 √
= −W0 (η)G0 ( n, η) + e− A2 e(−λ( U + η + ν)2 )
λ=−b0
∞
4πn X −
π(λ+bn )2 √
= −e− A2 + e V1
e(−λ( U + η + ν)2 ) + O(U −10 ),
λ=−∞

where √
A2 n
V1 = √ , bn = √ . (4.87)
( U + η + ν)2 U +η+ν
Using
√ √ √
e(−λ( U + η + ν)2 ) = e(−λ(U + 2η U + 2ν U + (η + ν)2 ))
√
= e(−λ({2ν U } + (η + ν)2 )),

37
 √
for U = [U], η = k1 / U , where k1 is an integer, we have
∞
− 4πn
X − Vπ (λ+bn )2
√
N(n) = −e 2
A + e 1 e(−λ({2ν U } + (η + ν)2 )) + O(U −10 ).
λ=−∞

By Lemma 3.2,
∞
p X √ 2
√ 4πn
N(n) = V1 e−πV1 (λ−{2ν U }−(η+ν) ) e(bn ({2ν U })+(η +ν)2 −λ)−e− A2 +O(U −10 ).
λ=−∞

1
√
Since 8
≤ {2ν U } ≤ 83 , (η + ν)2 = O(L−1 ) and

A2 2002 L
V1 = √ = C02 L(1 + o(1)) ≥ ,
( U + η + ν)2 2
then
4πn √ πC02L

N(n) = −e− A2 + O( Le− 200 )

4πn
= −e− A2 + O(U −10 ). (4.88)
By (4.83) and (4.88) we obtain (4.74). The proof is complete. ✷

5 EVALUATION of S(B)
Now we evaluate S(B) in (1.19), i.e.,
√
Z 5 L
A 2 θ2
X −1
S(B) = A √ e−πA dθ d(n)n 4 ,
− 5 AL √
ξ1 < n−B−θ≤ξ2

where A, B, ξ1 , ξ2 , · · · are (1.10)-(1.18).

By Lemma 2.4 and taking α0 = 2 in (1.12),
√
5 L
Z
A
Z (B+ξ2 +θ)2
1
−πA2 θ 2
X
S(B) = d(n)A √ e dθ x− 4 Θ(nx)dx + Sl (B) + O(ε1)
n≤N − 5 AL (B+ξ1 +θ)2
√
2 5 L
√ X 1
X γα A
Z
A 2 θ2
= 2 2 d(n)n− 4 √ √ e−πA jnα (θ)dθ
n≤N α=0
(4π n)α − 5 AL

− 23
+Sl (B) + O(U ) + O(ε1) (5.1)
for any ε1 > 0 as N ≥ N(ε1 ), where
√
cos(4π nx + π4 + απ
Z B+ξ2 +θ
2
)dx
jnα (θ) =
B+ξ1 +θ xα
Z ξ2 √
e(2 n(B + x + θ))dx
= Re e(α/4 + 1/8) , (5.2)
ξ1 (B + x + θ)α

38
 √
5 L
Z
A
Z (B+ξ2 +θ)2
−1
−πA2 θ 2
Sl (B) = A √ e dθ x 4 (log x + 2γ)dx. (5.3)
− 5 AL (B+ξ1 +θ)2

The inner integral of Sl (B) is

Z ξ2 +θ 10
1
X
2
2 (B + x) 2 (log(B + x) + 2γ)dx = Ck η k + O(U −9 ).
ξ1 +θ k=0
√ k+j
Let B ′ = B − η = U+ √
2 U
(see 1.11). Then by (3.37),

Sl (B) = Sl (B ′ + η) = Oη (U −9 ). (5.4)
By (5.2), we have
√ x=ξ2
e(2 n(B + x + θ))
jnα (θ) = Re e(α/4 + 1/8)( √
4πi n(B + x + θ)α
| x=ξ1
Z ξ2
1 αdx
+O( (1 + α+1
))).
n ξ1 (B + x + θ)

The integration by parts for θ gives

Z 5√L
A
−πA2 θ 2 A
A √ e jnα (θ)dθ ≪ .
− 5 AL n
Therefore,
√
2 5 L
γα A
Z
A
− 41 2 θ2 5
X X X
d(n)n √ √ e−πA jnα (θ)dθ ≪ d(n)n− 4 A ≪ U −4 . (5.5)
α=0
(4π n)α − 5 AL
U 20 <n≤N U 20 <n

By (5.1), (5.4) and (5.5) we obtain

√
2 5 L
√ X −1
X γα A
Z
A 2 θ2 3
S(B) = 2 2 d(n)n 4 √ √ e−πA jnα (θ)dθ + Oη (U − 2 ), (ε1 → 0+ ).
α=0
(4π n)α − 5 AL
n≤U 20

√ √ (5.6)
If n> C02 L U, then
√
Z 5 L
A 2 θ2
√ e−πA jnα (θ)dθ
− 5 AL
√
ξ2 5 L √
√ e(2 nθ)dθ
Z Z
A
−πθ 2
= Re e(α/4 + 1/8) e(2 n(B + x))dxA √ e . (5.7)
ξ1 − 5 AL (B + x + θ)α
Moving the inner integral line of (5.7), we have
√ √ √
− 5 AL + i√ 5 L
+ i√ 5 L √
e(2 nw)dw
Z Z Z
C0 U A C0 U A
−πA2 w 2
A( √ + √ + √ )e = I1 + I2 + I3 , say.
− 5 AL − 5 AL + i√ 5 L
A
+ i√ (B + x + w)α
C0 U C0 U
(5.8)

39
 √
As w ∈ I1 , w = − 5 AL + iv, 0 ≤ v ≤ 1√
C0 U
, and
√ √ √
|e(2 nw)| = |e(2i nv)| = e−4π nv . (5.9)

Hence
1√
√
Z √
C0 U 2 (− 5 L +iv)2
I1 ≪ A |e−πA A |e−4π nv
dv
0
Z 1√
25L 1√
C0 U −πC02 LU ( 2 U L − C4 U )
≪ A e C0 0 dv
0
A
≪ √ e−20πL ≪ U −20 .
U
i
In the same way, I3 ≪ U −20 . When w ∈ I2 , w = u + √
C0 U
. By (5.9),
√
√ −4π n
√ √ √
|e(2 nw)| = e C0 U
≪ e−4πC0 L ≪ U −40 , for n > C02 L U.

Hence
I2 ≪ U −20 .
Therefore, √
Z 5 L
A 2 θ2
√ e−πA jnα (θ)dθ ≪ U −20 (5.10)
− 5 AL
√ √
for n > C02 L U . If b0 is (4.74), i.e.

b0 = [C02 L], (5.11)

then
√
2 5 L
γα A
Z
X −1
X A 2 θ2
d(n)n 4 √ √ e−πA jnα (θ)dθ
√ √ α=0
(4π n)α − 5 AL
(b0 +1/2) U < n≤U 10

−1
X
≪ d(n)n 4 U −20 ≪ U −3 .
√ √
(b0 +1/2) U < n≤U 10

This and (5.6) give

√
2 5 L
√ X −1
X γα A
Z
A 2 θ2
S(B) = 2 2 d(n)n 4 √ √ e−πA jnα (θ)dθ + Oη (U −1 ),
√ √
α=0
(4π n)α − 5 AL
n≤(b0 +1/2) U
(5.12)
where jnα (θ) is (5.2). By (5.12) and (5.2),

√ Z ξ2
S(B) = 2 2Re (T0 (B + ξ) + T1 (B + ξ))dξ + Oη (U −1 ) (5.13)
ξ1

40
where ξ1 , ξ2 are (1.18),

T0 (B + ξ) =
√
2 α Z 5 AL
√
X −1
X γα Ai −πA2 θ 2 e(2 n(B + ξ + θ))dθ
= e(1/8) d(n)n 4 √ e , (5.14)
√ √ α=0
(4π n)α − 5√AL (B + ξ + θ)α
n≤ U /2

T1 (B + ξ) =
√
5 L
√
Z
X −1 A 2 θ2
= e(1/8) d(n)n 4 A √ e−πA e(2 n(B + ξ + θ))dθ
√ √ √ − 5 AL
U /2< n≤(b0 +1/2) U

+δ1 (B + ξ) (5.15)

for e( α4 ) = iα with
2
X γα Aiα
δ1 (B + ξ) = e(1/8) √
α=1
(4π n)α
√
5 L −1 √
d(n)n 4 e(2 n(B + ξ + θ))dθ
Z
A
−πA2 θ 2
X
× √ e · √ . (5.16)
− 5 AL √ √ √ ( n(B + x + θ))α
U /2< n≤(b0 +1/2) U

Using Lemma 3.4 for the inner sum, we have

δ1 (B + ξ) ≪ U −1+ε . (5.17)

The inner integral of (5.15) is

√ √
Z ∞
2 2 2 −4πn 2
A e−πA θ e(2 n( U + ξ + θ))dθ + O(e−L ) = e A2 + O(e−L ).
−∞

Hence
X −1 −4πn √
T1 (B + ξ) = e(1/8) d(n)n 4 e A2 e(2 n(B + ξ)) + Oη (U −1+ε ). (5.18)
√ √ √
U /2< n≤(b0 +1/2) U

Therefore,
√ Z ξ2
S(B) = 2 2Re (T0 (B + ξ) + T1 (B + ξ))dξ + Oη (U −1+ε ), (5.19)
ξ1

where T0 (B + ξ) is (5.14), T1 (B + ξ) is (5.18), ξ1 , ξ2 are (1.18).

41
6 EVALUATION of T1(B + ξ)
We are going to evaluate (5.18). For 0 < h ≤ h0 = 1/U, U = [X]
√
U U √ 1
0<( + h)2 − = Uh + h2 ≤ ,
2 4 L
U 3
0≤{ }≤ .
4 4
√ √
Hence there doesn’t exist any integer in the interval (( 2U )2 , ( 2U + h)2 ]. So that
X
= 0, for 0 < h ≤ h0 = 1/U. (6.1)
√ √ √
U /2< n≤ U /2+h

In the same way, X

= 0. (6.2)
√ √ √
(b0 +1/2) U < n≤(b0 +1/2) U +h

It follows from (5.18) that

e(1/8) h0 √
Z X −1 −4πn
T1 (B+ξ) = dh d(n)n 4 e A2 e(2 n(B+ξ))+Oη (U −1+ε ),
h0 0 √ √ √
U /2< n≤(b0 +1/2) U +h
(6.3)
where h0 = 1/U. By Lemma 2.3
√
h0 ((b0 +1/2) U+h)2 √
e(1/8)
Z Z
X −1 −4πx
T1 (B + ξ) = d(n) dh √
x 4 e A2 e(2 x(B + ξ))Θ(nx)dx
n≤N
h0 0 ( U /2+h)2

+T1l (B + ξ) + O(ε1) (6.4)

for any ε1 > 0, as N ≥ N(ε1 ), where Θ(nx) is (2.12) and

√
h0 ((b0 +1/2) U +h)2 √
e(1/8)
Z Z
−1 −4πx
T1l (B + ξ) = dh √
x 4 e A2 e(2 x(B + ξ))(log x + 2γ)dx
h0 0 ( U /2+h)2
√
h0 (b0 +1/2) U +h
2e(1/8)
Z Z
1 −4πx2
= dh √ x2 e A2 e(2x(B + ξ))(log x2 + 2γ)dx. (6.5)
h0 0 U /2+h

Changing the inner integral line of (6.5), we have

2e(1/8)
T1l (B + ξ) =
h0
√ √ √
Z h0 Z U /2+h+i Z (b0 +1/2) U +h+i Z (b0 +1/2) U +h
× dh( √
+ √
+ √
)e(2z(B + ξ))fl (z)dz
0 U /2+h U /2+h+i (b0 +1/2) U +h+i
h0
2e(1/8)
Z
= dh(I1h (B + ξ) + I2h (B + ξ) + I3h (B + ξ)), say (6.6)
h0 0

42
where
−4πz 2
fl (z) = z 1/2 e A (log z 2 + 2γ). (6.7)
Furthermore,
Z i √ √
I1h (B + ξ) = e(2(B + ξ)( U /2 + h + z))fl ( U /2 + h + z)dz
0
Z i √ √ √
= e(2(B ′ + ξ)( U/2 + h + z))fl ( U /2 + h + z)e(η( U + 2h + 2z))dz,
0
√
where B = B ′ + η. By (1.14) and (1.15), e(η U ) = 1, hence, by Lemma 3.6
√
e(η( U + 2h + 2z)) = e(η(2h + 2z)) = Oη (U −99 ).
Therefore,
−99
Z 1 √
I1h (B + ξ) ≪η U |f ( U /2 + h + iρ)e(2(B ′ + ξ)iρ)|dρ ≪η U −20 .
0
In the same way,
I3h (B + ξ) ≪η U −20 .
When z ∈ I2h (B + ξ), z = x + i. Thus,
√ √
|e(2z(B + ξ))fl (z)| ≪ U L|e(2(B + ξ)i)| = ULe−4π(B+ξ) .
Hence √
√ Z (b0 +1/2) U +h
I2h (B + ξ) ≪ ULe−4π(B+ξ) √ dx ≪ U −21 .
U /2+h

Therefore, by (6.6),
T1l (B + ξ) = Oη (U −19 ). (6.8)
Putting x = x21 in the integral (6.4), taking α0 = 2 in (2.12), noticing 2 cos 2πx =
e(x) + e(−x), e( α4 ) = iα , by (6.4) and (6.8) we obtain
T1 (B + ξ) = T1+ (B + ξ) + T1− (B + ξ) + Oη (U −19 ) + O(ε1 ), (6.9)
where
2
√ X X γα (±i)α ±
T1± (B + ξ) = 2 d(n)n−1/4
√ j (B + ξ), (6.10)
n≤N α=0
(4π n)α nα
√ −4πx2 √
h0 (b0 +1/2) U +h
e(1/8 ± 1/8) e e(2x(B + ξ ± n))dx
Z Z
A2
±
jnα (B + ξ) = dh √
. (6.11)
h0 0 U /2+h xα
√ √
If n > 2 U, then
Z h0 −4π(x+h)2 √ √
x=(b0 +1/2) U
e(1/8 ± 1/8) e A2 e(2(x + h)(B + ξ ± n))
±
jnα (B + ξ) =
4πih0
dh √
(B + ξ ± n)(x + h)α
| √
x= U /2
0
1
+O( √ )
(B + ξ ± n)2
1 U
≪ √ 2 ≪ (6.12)
h0 (B + ξ ± n) n

43
Since X U
d(n)n−1/4−α/2 ≪ U −3
n
U 20 <n≤N

for α ≥ 0, then by (6.10),

2
√ X X γα (±i)α ±
T1± (B + ξ) = 2 d(n)n−1/4 √ α jnα (B + ξ) + O(U −3 ). (6.13)
√
α=0
(4π n)
n≤U 10

Let z = x + iy, we see that

√ √ √
|e(2(B + ξ ± n))| = e(2iy(B + ξ ± n)) = e−4πy(B+ξ± n)
. (6.14)

Using the path (6.6), repeating the proof of (6.8), we obtain

+
jnα (B + ξ) = Oη (U −19 ), for n ≥ 1. (6.15)
√ √
In the same way, if n ≤ U − L2 , then the right side of (6.14) is
√ 2
e−4πy(B+ξ− n)
≪ e−4πyL , y ≥ 0.

The path (6.6) gives also

−
√ √
jnα (B + ξ) = Oη (U −19 ), for n≤ U − L2 . (6.16)
√ √
If n> U + L2 , then moving the inner integral line of jnα
−
(B + ξ) to
√ √ √
Z U /2+h−i Z (b0 +1/2) U +h−i Z (b0 +1/2) U +h

√
+ √
+ √
,
U /2+h U /2+h−i (b0 +1/2) U +h−i

noticing √ √
|e(2(B + ξ − n)z)| = e−4π|y(B+ξ− n)|

for z = x + iy, y ≤ 0, we have also

−
√ √
jnα (B + ξ) ≪η U −19 , for n≤ U − L2 . (6.17)

It follows from (6.9), (6.13), (6.15), (6.16) and (6.17) that

2
√ X
−1/4
X γα (−i)α −
T1 (B + ξ) = 2 d(n)n √ α jnα (B + ξ)
√ √
α=0
(4π n)
−L2 < n− U≤L2

+ δ + Oη (U −2 ) (ε1 → 0+ ), (6.18)
−
where jnα (B + ξ) is (6.11) and
2
X
−1/4
X U −19
δ ≪η d(n)n √ α ≪η U −2 . (6.19)
√
α=0
( n)
n≤U 10

44
 −
Next, we evaluate jnα (B + ξ), α = 1, 2. Using the integration by parts, It is easy to
know that
− 1
jnα (B + ξ) ≪ √ √ ,
( U )α |B + ξ − n|
or,
√
h0 (b0 +1/2) U +h
1 dx
Z Z
−
jnα (B + ξ) ≪ dh √
√
h0 0 U /2+h ( U )α
√
b0 U √
≪ √ ≪ L( U)1−α .
( U )α
Hence
− L 1
jnα (B + ξ) ≪ min( √ , √ √ ).
( U )1−α ( U )α |B + ξ − n|
Moreover,
2
X γα (−i)α X
d(n)n−1/4−α/2 jnα
−
(B + ξ)
α=1
(4π)α √ √
−L2 < n− U≤L2

X X d(n)n−3/4
≪ L d(n)n−3/4 +
√ √ √ √ |(B + ξ)2 − n|
|B+ξ− n|≤ √1 |B+ξ− n|≥ √1 ,| n− U|≤L2
U U

3 1
≪ U − 4 +ε , 0 < ε = O( ) (6.20)
log L
This and (6.18), (6.19), (6.11) give
√ X 3
T1 (B + ξ) = 2 d(n)n−1/4 jn−0 (B + ξ) + δ + Oη (U − 4 +ε )
√ √
−L2 < n− U ≤L2

√ Z h √
(b0 +1/2) U +h
2 √
0
Z
X −4πx2
−1/4
= dh d(n)n √
e A2 e(2x(B + ξ − n))dx
h0 0 √ √ U /2+h
−L2 < n− U≤L2

3 1 1
+ Oη (U − 4 +ε ), h0 = , 0 < ε = O( ) . (6.21)
U log L
By Lemma 3.4,
X √
d(n)n−1/4 e(−2 nx) ≪ x1/4+ε ≪ U 1/4+ε .
√ √
−L2 < n− U ≤L2

Hence
Z √U /2 Z (b0 +1/2)√U +h
1 h0 √
Z
−4πx2 X
dh( √ + √ )e A2 d(n)n−1/4 e(2x(B + ξ − n))
h0 0 U /2+h (b0 +1/2) U √ √
−L2 < n− U ≤L2

45
 h0
U 1/4+ε 1
Z
3
≪ hdh ≪ U − 4 +ε , 0 < ε = O( ). (6.22)
h0 0 log L
It follows from (6.21) and (6.22) that

T1 (B + ξ)
√
√ X Z (b0 +1/2) U −4π(x)2 √ 3
= 2 d(n)n−1/4 √ e A2 e(2x(B + ξ − n))dx + Oη (U − 4 +ε ),
√ √ U /2
−L2 < n− U ≤L2
(6.23)
where b0 = [C02 L], 0<ε= O( log1 L ).

m
P
7 EVALUATION of SUM T1(B + ξ)
k=0
Let
√ j
B1 = U + √ + η + ξ, (7.1)
2 U
then by (1.10),
k
B + ξ = B1 + √ . (7.2)
2 U
Moreover, denote
m m
X X k
Q1 (B1 ) = T1 (B + ξ) = T1 (B1 + √ ), (7.3)
k=0 k=0
2 U

where T1 (B) is (6.27). In this section we are going to prove that

m
X k
Re(Q1 (B1 )) = Re T1 (B1 + √ )
k=0
2 U
√
U X
= −√ d(n)n−1/4 + Re δQ1 (B1 ), (7.4)
2 0<√n−B ≤ √m
1
2 U

1
where δQ1 (B1 ) ≪ U 4 +ε , 0 < ε = O( log1 L ) = O( log log
1
U
).

Proof of (7.4): By (6.23), (7.1), (7.2 ) and (7.3),

√
√ X Z (b0 +1/2) U
4πx2 √
Q1 (B1 ) = 2 d(n)n −1/4
√ e− A2 e(2(B1 − n)x)ρ(x)dx
√ √ U
−L2 < n− U ≤L2 2

1
+Oη (U − 4 +ε ) (7.5)

46
 √
for m ≪ U L−2 , where 0 < ε = O( log1 L ) and
m
X xk
ρ(x) = e( √ ). (7.6)
k=0
U
By Lemma 3.3,
U Z m 10
1 1 xm −9
X x
ρ(x) = + e( √ ) + O(U ) + e(( √ − b)u)du
2 2 U 10 0 U b=−U
−9
= ρ0 (x) + ρ1 (x) + ρ2 (x) + O(U ), (7.7)
where
b0 Z m
X x
ρ0 (x) = e(( √ − b)u)du, (7.8)
b=1 0 U
1 1 xm 1 X xν
ρ1 (x) = + e( √ ) = e( √ ), (7.9)
2 2 U 2 ν=o,m U
X Z m x e( √xuU ) u=m
|
X
′ ′
ρ2 (x) = e(( √ − b)u)du = (7.10)
U 2πi( √x − b) u=0
b 0 b U
P′ −10
for e(−bν) = 1, ν = 0, m, and is for U ≤ b ≤ 0, or b0 + 1 ≤ b ≤ U 10 . It follows
from (7.5) and (7.7) that
Q1 (B1 ) = Q10 (B1 ) + Q11 (B1 ) + Q12 (B1 ) + Oη (U −1 ), (7.11)
where √ X
Q1j (B1 ) = 2 d(n)n−1/4 Jj (n), j = 0, 1, 2, (7.12)
√ √
−L2 < n− U ≤L2
√
Z 2
Z Z
− 4πz2
Jj (n) = e A e(2(B1 − n)z)ρj (z)dz = = , (7.13)
+ −
C CA CA
√ √ √ √
C is from 2U to (b0 + 1/2) U , CA± is the broken line joining the points U
, U
± iA/2,
√ √ 2 2
(b0 + 1/2) U ± iA/2 and (b0 + 1/2) U .

CA+

✻
√
s❄
√
2
U s (b0 + 21 ) U
✻
❄

CA−

Fig. 7.1

47
 By (7.10) and (7.13),
4πz 2 √
X Z e− A2 e(2(B1 + u
√ − n)z)dz u=m
J2 (n) = ′ 2 U
2πi( √zU − b)
| u=0
. (7.14)
b C
√
U
√
It is obvious that the integrand is analytic for 2
≤ Re z ≤ (b0 + 1/2) U , as U −10 ≤
√ √
b ≤ 0, or b0 + 1 ≤ b ≤ U 10 . For z = x + iy, 2U ≤ x ≤ (b0 + 1/2) U , −A/2 ≤ y ≤ A/2,
we have
4πz 2 4π(x2 −y 2 ) 4πx2
|e− A2 | = e− A2 ≪ e− A2 , (7.15)
u √ u √
−4πy(B1 + √ − n)
|e(2(B1 + √ − n)z)| = e 2 U . (7.16)
2 U
√
If n ≤ B1 + 2√uU , then moving the integral line C to CA+ , we see that
√ √ √
U
Z Z Z
2
+ iA
2
Z (b0 +1/2) U + iA
2
Z (b0 +1/2) U
= = ( √ + √ + √ )
+ U U
C CA 2 2
+ iA
2
(b0 +1/2) U + iA
2
4πz 2 √
e− A2 e(2(B1 + u
√
2 U
− n)z)dz
×
2πi( √zU − b)
= I1 + I2 + I3 , say (7.17)
Clearly, for b ≤ 0 or b ≥ b0 + 1 and
1 1
≪ , z ∈ CA± , (7.18)
√z −b |b| + 1
U

we have
A u √
−4πy(B1 + √ − n)
e dy
Z
2 2 U
I1 ≪
0 |b| + 1
1 1
≪ u √ = u √ .
(|b| + 1)(B1 + 2√U − n) (|b| + 1)|B1 + √
2 U
− n|
In the same way
1
I3 ≪ u √ .
(|b| + 1)|B1 + √
2 U
− n|
By (7.15)-(7.18),
√
Z (b0 +1/2)√U
1 u
−2πA(B1 + √ − n) 4πx2
I2 ≪ e 2 U
√ e− A2 dx
|b| + 1 2
U

A u √
−2πA(B1 + √ − n)
≪ e 2 U
|b| + 1
1 u √ −2πA|B1 + √u −√n|
= √ (A|B 1 + √ − n|e 2 U )
(|b| + 1)|B1 + 2√uU − n| 2 U
1
≪ √ .
(|b| + 1)|B1 + 2√uU − n|

48
Hence the integral (7.17) is

1
Z
≪ u √ (7.19)
C (|b| + 1)|B1 + √
2 U
− n|
√ √
for n ≤ B1 + 2√uU . If n > B1 + u
√
2 U
, then moving the path C to CA− , we have also
(7.19). Hence, by (7.14) we obtain
X 1 X
′ 1
J2 (n) ≪ u √
u=0,m
|B1 + √
2 U
− n| b
(|b| + 1)
X 1
≪ L u √ . (7.20)
u=0,m
|B1 + √
2 U
− n|

In the same way,

X 1 X 1
J1 ≪ ν √ = u √ . (7.21)
ν=0,1
|B1 + √
2 U
− n| u=0,m
|B1 + √
2 U
− n|

Therefore, it follows from (7.12),(7.20), (7.21) and Lemma 3.1 that

X X Ld(n)n−1/4
Q11 (B1 ) + Q12 (B1 ) ≪ √
√
u=0,m −L2 <√n− U ≤L2
|B1 + 2√uU − n|
1
X X 1
≪ U 4 +ε u
√ √
u=0,m −L2 U <n−U ≤L2 U
|n − (B1 + √
2 U
)2 |
1
X 1 1
≪ U 4 +ε (1 + ( u 2
+ u , )), (7.22)
u=0,m
{(B1 + 2 U ) } 1 − {(B1 +
√ √
2 U
)2 }

where ε = O((log L)−1 ). Moreover,

u √ j+u
{(B1 + √ )2 } = {( U + √ + η + ξ)2 }
2 U 2 U
√ j+u
= {U + j + u + 2 U(η + ξ) + ( √ + η + ξ)2 }
2 U
√ −1
= {2 U ξ + O(L )}
√ √ √
for u = 0, m, 2 Uη = 2k1 , j, u ≪ U L−2 , η ≪ L−1 . Hence by (1.18): 81 ≤ 2 U ξ ≤ 83 ,
we have
1 u 4
≤ {(B1 + √ )2 } ≤ . (7.23)
9 2 U 9
Furthermore, by (7.22) and (7.23), we have
1
Q11 (B1 ) + Q12 (B1 ) = δQ11 ≪ U 4 +ε . (7.24)

49
This and (7.11) give

Q1 (B1 ) = Q10 (B1 ) + δQ11

√ X
= 2 d(n)n−1/4 J0 (n) + δQ11
√ √
−L2 < n− U ≤L2
b0 Z
√ X m X
−1/4
Z
= 2 du d(n)n fnb (z, u)dz + δQ11 , (7.25)
0 √ √ C
b=1 −L2 < n− U ≤L2

√
U
√ 1
where C is from 2
to (b0 + 12 ) U , δQ11 ≪ U 4 +ε and
4πz 2 √ z
fnb (z, u) = e− A2 e(2(B1 − n)z + ( √ − b)u). (7.26)
U
Clearly,
Z m X Z m X X
du = du( + )
0 √ √ 0 √ √ u u √ √
−L2 < n− U ≤L2 U −L2 < n≤B1 + √ B1 + √ < n≤ U +L2
2 U 2 U

X Z m X Z M2 (n)
= + , (7.27)
√ √ m M1 (n) √ √ 0
U −L2 < n≤B1 + √ B1 < n≤ U +L2
2 U

where √ √
M1 (n) = max(0, 2 U ( n − B1 )), (7.28)
√ √
M2 (n) = min(m, 2 U ( n − B1 )). (7.29)
Using Z Z Z
fnb (z, u)dz = = ,
+ −
C CA CA

where CA± be as Fig 7.1, by (7.25) and (7.27) we have

b0
√ X
Q1 (B1 ) = 2 (w + (b) + w − (b)) + δQ11 , (7.30)
b=1

where
1
δQ11 = O(U 4 +ε ), ε = O((log L)−1 ),

X Z Z m
+ −1/4
w (b) = d(n)n dz fnb (z, u)du
√ +
√ m CA M1 (n)
U −L2 < n≤B1 + √
2 U

4πz 2 √
Z e− A2 e(2(B1 + u
√ − n)z − bu) u=m
|
X 2 U
−1/4
= d(n)n .(7.31)
√ √ m
+
CA 2πi( √zU − b) u=M1 (n)
U−L2 < n≤B1 + √
2 U

50
In the same way,
4πz 2 √
Z e− A2 e(2(B1 + u
√ − n)z − bu) u=M2 (n)
|
X 2 U
w − (b) = d(n)n−1/4 (7.32)
√ √ −
CA 2πi( √zU − b) u=0
B1 < n≤ U +L2

For a function gn (u), by (7.28) we have

u=m
|
X
gn (u) u=M1 (n)
√ √ m
U −L2 < n≤B1 + √
2 U
u=m u=m
| |
X X
= gn (u) u=0
+ gn (u) √ √ √
u=2 U ( n− U )
√ √ √ m
U −L2 < n≤B1 B1 < n≤B1 + √
2 U
u=m √ √
|
X X
= gn (u) u=0
− gn (2 U ( n − B1 )) (7.33)
√ √ u
√ m
U −L2 < n≤B1 + √ 0< n−B1 ≤ √
2 U 2 U

Likewise ,
u=M2 (n)
|
X
gn (u) u=0
=
√ √
B1 < n≤ U +L2

u=m √ √
|
X X
= gn (u) u=0
+ gn (2 U ( n − B1 )) (7.34)
u √ √ √ m
B1 + √ < n≤ U +L2 0< n−B1 ≤ √
2 U 2 U

Noticing e(−ub) = 1 for u = 0, m, and

u √
e(2(B1 + √ − n)z − bu)
2 U
| √ √
u=2 U ( n−B1 )
=

√ √ √ √ √ j
= e(−2b U( n − B1 )) = e(−2b nU + 2b U ( U + √ + η + ξ))
2 U
√ √
= e(−2b nU + 2b U ξ)
√
for η = k1 / U , we know that (7.30)-(7.34) deduce
b0
√ X
Q1 (B1 ) = 2 (w1+ (b) + w1− (b) + w0 (b)) + δQ11 , (7.35)
b=1

where
4πz 2 √
Z e− A2 e(2(B1 + u
√ − n)z) u=m
|
X 2 U
w1+ (b) = d(n)n −1/4
dz (7.36)
√ √ u
+
CA 2πi( √zU − b) u=0
U −L2 < n≤B1 + √
2 U

51
 4πz 2 √
Z e− A2 e(2(B1 + u
√ − n)z) u=m
|
X 2 U
w1− (b) = d(n)n−1/4 dz (7.37)
u √ √ −
CA 2πi( √zU − b) u=0
B1 + √ < n≤ U +L2
2 U

4πz 2 √ √
e− e(−2b nU + 2b U ξ)dz
I
X A2
−1/4
w0 (b) = d(n)n
√ m
−
CA +
−CA 2πi( √zU − b)
0< n−B1 ≤ √
2 U
√ 4πU b2 √ X √
= Ue− A2 e(2b U ξ) d(n)n−1/4 e(−2b nU ) (7.38)
√ m
0< n−B1 ≤ √
2 U

Repeating the proof of (7.24),

b0
X 1 1
δQ10 = (w1+ (b) + w1− (b)) ≪ U 4 +ε , ε = O( ), (7.39)
b=1
log L

we obtain from (7.35)-(7.39) that

b0
√ X 4πU b2 √ X √
Q1 (B1 ) = 2U e− A2 e(2b U ξ) d(n)n−1/4 e(−2b nU ) + δQ1 , (7.40)
√ m
b=1 0< n−B1 ≤ √
2 U

where δQ1 = δQ10 + δQ11 . Therefore, using e((U + n)b) = 1 and

√ √ √ √
e(−2b nU ) = e((U + n)b − 2b nU ) = e(b( n − U )2 ),
we have √
U X
Q1 (B1 ) = √ d(n)n−1/4 N(n) + δQ11 , (7.41)
2 0<
√ m
n−B1 ≤ √
2 U

where
b0
X 4πU b2 √ √ √
N(n) = 2 e− A2 e(b(2 Uξ + ( n − U )2 ))
b=1
∞
X 4πU b2 √ √ √
= 2 e− A2 e(b(2 Uξ + ( n − U )2 )) + O(U −10 ),
b=1

the last equality is given by

2 2
− 4πU2b
2 − 4πb2 b
−10 − C02
e A =e C
0 ≪U e
for b > b0 = [C02 L], C0 ≥ 200. By Lemma 3.2,
∞
X 4πU b2 √ √ √
ReN(n) = −1 + e− A2 e(b(2 Uξ + ( n − U )2 )) + O(U −10 )
b=−∞
∞
A X − πA2 (2√U ξ+(√n−√U )2 −b)2
= −1 + √ e 4U + O(U −10 ).
2 U b=−∞

52
By (7.41), √
√
( n − U )2 = O(L−1 ).
Thus,
1 √ √ √ 3
≤ 2 Uξ + ( n − U )2 ≤
9 7
for
1 √ 3
≤ 2 Uξ ≤ .
8 8
Hence
∞
√ − πC02 L √ X 2 Lb2
C0
ReN(n) = −1 + O( Le 4·11 ) + O( L
2
e− 4 ) + O(U −10 )
b=1
−10
= −1 + O(U ). (7.42)
It follows from (7.41) and (7.42) that
√
U X
Q1 (B1 ) = − √ d(n)n−1/4 + δQ1 , (7.43)
2 0<√n−B ≤ m
√
1
2 U

1 1
where δQ1 = δQ10 + δQ11 ≪ U 4 +ε , 0 < ε ≪ log L
. The proof of (7.4) is complete. 

Note
For other application to be able, let us write down the accurate δQ1 (B1 ), the O-term
of (7.43):
b0
√ X 1
δQ1 (B1 ) = δQ10 + δQ11 = Q11 + Q12 + 2 (w1+ (b) + w1− (b)) + O(U − 4 +ε(U ) ), (7.44)
b=1

where ε(U) ≪ 1/ log log U,

Q1j (B1 )
√ 4πz 2 √
d(n)n−1/4 e−
P R
= 2 C
A2 e(2(B1 − n)z)ρj (z)dz,
√ √
−L2 < n− U ≤L2

j = 1, 2, (7.45)

1 1 zm
ρ1 (z) = + e( √ ),
2 2 U
e( √zuU ) u=m
|
X
′
ρ2 (z) = ,
2πi( √zU − b) u=0
b
Z u=m
|
X
w1+ (b) = d(n)n −1/4
fnb (z, u) u=0
dz, (7.46)
√ +
√ u CA
U −L2 < n≤B1 + √
2 U
Z u=m
|
X
w1− (b) = d(n)n −1/4
fnb (z, u) u=0
dz, (7.47)
√ −
u √ CA
B1 + √ < n≤ U+L2
2 U

53
 √ √ P
where C is from U /2 to (b0 + 1/2) U, ′ is for −U 10 ≤ b ≤ 0 or b0 + 1 ≤ b ≤ U 10 ,
CA± is Fig 7.1,
4πz 2 √
e− A2 e(2(B1 + u
√
2 U
− n)z)
fnb (z, u) = . (7.48)
2πi( √zU − b)

m
P
8 EVALUATION OF SUM T (B + ξ)
k=0
Let
m m
X X k k
Q(B1 ) = T (B + ξ) = (T0 (B1 + √ ) + T1 (B1 + √ ))
k=0 k=0
2 U 2 U
= Q0 (B1 ) + Q1 (B1 ), say (8.1)

where T0 (B1 ) is (5.14), T1 (B1 ) is (5.18) and

√ j √
B1 = U + √ + η + ξ, m ≪ UL−2 . (8.2)
2 U
In this section , we are going to prove that
m
√ X
− 14 − 4πn
Z √
2 U √ π
Re Q(B1 ) = 2 U d(n)n e A2 cos(4π n(B1 + x) + )dx
√ √ 0 4
U
n≤ 2

+Re δQ (B1 ), (8.3)

1
where δQ ≪ U 4 +ε .

(This equality is a key of success of our paper. But if there was no sum of the right
m m
S(B) by k + k ′ and
P P
side, we could replace k and S(B), respectively, and we
k=0 k,k ′ =0
should be succeed, although it would be more numerous.)

Proof of (8.3):
First, we evaluate further Re Q1 (B1 ). By (7.23):
1 u 4
≤ {(B1 + √ )2 } ≤
9 2 U 9
√ √ √
for u = 0, m. For A = C0 UL, C0 ≥ 200, − 5 AL ≤ θ ≤ 5 L
A
and u = 0, m, we have
u u
|(B1 + √ + θ)2 − (B1 + √ )2 |
2 U 2 U
u 5 √ 1
= |θ||2B1 + √ + θ| ≤ √ (2 U + O(1)) < (8.4)
U 200 U 19

54
 u u
Hence it does not run over any integer number from (B1 + √
2 U
)2 to (B1 + √
2 U
+ θ)2
when u = 0, m. So that
√
Z 5 L
A 2 θ2
X
e−πA = 0,
0 √ u
0< n−B1 − √ ≤θ
2 U
√
Z 5 L
A 2 θ2
X
e−πA = 0.
0 √ u
−θ≤ n−B1 − √ ≤0
2 U

Hence √
Z 5 L
A 2 θ2
X
A √ e−πA dθ d(n)n−1/4
− 5 AL √ m
0< n−B1 −θ≤ √
2 U

√
Z 5 L
A 2 θ2
X
= (A √ e−πA dθ) d(n)n−1/4
− 5 AL √ m
0< n−B1 ≤ √
2 U
X
= (1 + O(U −10 )) d(n)n−1/4 , (8.5)
√ m
0< n−B1 ≤ √
2 U

where the last equality is given by

√
5 L
Z
A
Z ∞
−πA2 θ 2
A √ e dθ = A +O(U −10 ) = 1 + O(U −10 ).
− 5 AL −∞

It follows from (7.4) and (8.5) that

√ Z 5√L
UA A
−πA2 θ 2
X
Re Q1 (B1 ) = − √ √ e d(n)n−1/4
2 − 5 AL √
0< n−B −θ≤ m
√
1
2 U

+δQ1 (B1 ) + O(U −1/4+ε(U ) ), (8.6)

where δQ1 (B1 ) ≪ U 1/4+ε(U ) is (7.43). By Lemma 2.4,

√ Z 5√L Z (B1 + √m +θ)2
UA X A
−πA2 θ 2 2 U
Re Q1 (B1 ) = − √ d(n) √ e dθ x−1/4 Θ(nx)dx
2 n≤N 5 L
− A (B1 +θ)2

+&l + δQ1 (B1 ) + O(U −1/4+ε(U ) ) + O(ε1)

√ X 2 Z 5√L
X γ α A A
−πA2 θ 2
= −2 U d(n)n−1/4 √ α √ e jnα (θ)dθ
n≤N α=0
(4π n) − 5 L
A

+&l + δQ1 (B1 ) + O(U −1/4+ε(U ) ) + O(ε1) (8.7)

55
holds for any ε1 > 0 as N ≥ N(ε1 ), where
Z B1 + √m +θ √ π απ
2 U cos(4π nx + 4
+ 2
)
jnα (θ) = dx, (8.8)
B1 +θ xα
√
5 L m
√ Z A
−πA2 θ 2
Z B0 + √
2 U
+θ
&l = &l (η) = − 2U √ e δx (η)dx, (8.9)
− 5 AL B0 +θ

where
√ j
B0 = B1 − η = U + √ + ξ,
2 U
δx (η) = (x + η)1/2 (log(x + η)2 + 2γ).
By Lemma 3.6,
δx (η) ≪η U −99
(the definition of ”≪η ” see (1.28)), we know that

&l = &l (η) ≪η U −90 . (8.10)

Moreover,
√
Z 5 L
A 2 θ2
A √ e−πA jnα (θ)dθ
− 5 AL

√
5 L √ m
sin(4π n(x + θ) + π4 + απ ) x=B1 + 2√U
Z
= A √ e
A
dθ −πA2 θ 2
√
4π n(x + θ)α
2
x=B1
|
− 5 AL
√
Z 5 L Z B1 + √m √
A
−πA2 θ 2 2 U α sin(4π n(x + θ) + π4 + απ
2
)
+A √ e dθ √ α+1
dx
− 5 AL B1 4π n(x + θ)
A
≪ √ , (8.11)
n( U )α
where the last inequality is given by the integration by parts for θ. Hence, using
X A
d(n)n−1/4 ≪ U −1/4 L2 , (8.12)
n
U 3 <n≤N
X A
d(n)n−1/4−α/2 √ ≪ U −1/4 (8.13)
n( U ) α
U/4<n≤N

for α ≥ 1, by (8.7), (8.10), (8.12), (8.13) and putting ε1 → 0+ , we have

√
5 L
√ X Z
A 2 θ2
Re Q1 (B1 ) = −2 U d(n)n−1/4 A √ e−πA jn0 (θ)dθ
U − 5 AL
4
<n≤U 3

+Q10 (B1 ) + Re δQ1 (B1 ) + Oη (U −1/4 ), (8.14)

56
where
√
2 5 L
√ X X γα A
Z
A 2 θ2
Q10 (B1 ) = −2 U d(n)n−1/4 √ √ e−πA jnα (θ)dθ
√ √
α=0
(4π n)α − 5 AL
n≤ 2U
√
2 5 L
√ X X γα A
Z
A 2 θ2
= −2 U d(n)n−1/4 √ √ e−πA dθ ×
√ √
α=0
(4π n)α − 5 AL
n≤ 2U
B1 + m
√ +θ
√ π απ
cos(4π nx + + )
Z
2 U
4 2
× dx. (8.15)
B1 +θ xα

Furthermore,
√
Z 5 L
A 2 θ2
A √ e−πA jn0 (θ)dθ =
− 5 AL

√
m 5 L
Z B1 + √ Z
A √
2 U 2 θ2
= dxRe A √ e−πA e(2 n(x + θ) + 1/8)dθ
B1 − 5 AL
m
Z B1 + √
√
Z ∞ √
2 U 2 θ2
= dxRe e(2 nx + 1/8)A e−πA e(2 nθ)dθ + O(U −10 )
B1 −∞
m
Z B1 + √
√
4πn 2 U
= e− A cos(4π nx + π/4)dx + O(U −10 )
B1
m
√
√
Z
2 U
− 4πn
= e A cos(4π n(B1 + x) + π/4)dx + O(U −10 ). (8.16)
0

4πn √ √
Noticing e− A ≪ U −10 for n > b0 U , it follows from (8.14) and (8.16) that
m
√ Z √
2 U X 1 4πn √
Re Q1 (B1 ) = −2 U ( d(n)n− 4 e− A2 cos(4π n(B1 + x) + π/4))dx
0 √ √ √
U
2
< n≤b0 U

+Q10 (B1 ) + δQ1 (B1 ) + Oη (U −1/4+ε(U ) ), (8.17)

1
where Q10 (B1 ) is (8.15), δQ1 (B1 ) ≪ U 4 +ε is (7.43).
Next we evaluate Re Q0 (B1 ). By (5.14) and (7.2),
m
X
Q0 (B1 ) = T0 (B + ξ)
k=0
√
2 5 L
1 i
Z
X X A 2 θ2
= e( ) d(n)n−1/4 γα ( √ )α A √ e−πA
8 √ √
α=0
4π n − 5 AL
U
n≤ 2
√
×e(2 n(B1 + θ))ρ(n, α, θ)dθ, (8.18)

57
where √
m
X e( √nk
U
)
ρ(n, α, θ) = k
. (8.19)
k=0
(B1 + θ + √
2 U
)α
By Lemma 3.3,
√
Z m e( √nu
U
)du
ρ(n, α, θ) = + ρδ (n, α, θ) + O(U −9 )
0 (B1 + θ + 2√uU )α
Z B1 + √m +θ √
√ e(2 n(x − B1 − θ))dx
2 U
= 2 U
B1 +θ xα
+ρδ (n, α, θ) + O(U −9 ), (8.20)
where
√
X Z m e(( √Un − b)u)du 1
ρδ (n, α, θ) = u +
(B1 +θ+ √ )α 2(B1 + θ)α
−U 10 ≤b≤U 10 ,b6=0 0 2 U
√
e( √nm U
)
+ m α
2(B1 + θ + 2√ U
)
√
e( √nu ) u=m
|
X U
= √
n
10 10
−U ≤b≤U ,b6=0
2πi( √
U
− b)(B1 + θ + u
√
2 U
)α u=0
√
X √α
Z e((
m √n − b)u)du
2 U U
+ √
n u
−U 10 ≤b≤U 10 ,b6=0 0 2πi( √U − b)(B1 +θ+ √
2 U
)α+1
√
1 e( √nm
U
)
+ + m α. (8.21)
2(B1 + θ)α 2(B1 + θ + 2√ U
)
It is easy to prove that in Lemma 3.5 ,
Z 5L 2
A
−πθ 2
X
−1/4
X i √
δQ0 (B1 ) = e dθ d(n)n γα ( √ )α e(2 n(B1 + θ))ρδ (n, α)
− 5L
A √ √
α=0
4π n
n≤ 2U

≪ U 1/4+ε . (8.22)
Then by (8.18), (8.20) and (8.22), we obtain that
√
2 Z 5 L
√
1 X −1/4
X i α
A
Q0 (B1 ) = 2 U e( ) d(n)n γα ( √ ) A √
8 √ √ α=0
4π n − 5 AL
U
n≤ 2
B1 + m
√ +θ √
e(2 nx)
Z
2 U
−πA2 θ 2
×e dx + δQ0 (B1 ) + O(U −1/4+ε(U ) ),
B1 +θ xα
where δQ0 (B1 ) ≪ U 1/4+ε(U ) is (8.22). Hence
Re Q0 (B1 ) = −Q10 (B1 ) + δQ0 (B1 ) + O(U −1/4+ε(U ) ), (8.23)

58
where Q10 (B1 ) is (8.15), δQ0 (B1 ) is (8.22). It follows from (8.17) and (8.23) that

Re Q(B1 ) = Re (Q0 (B1 ) + Q1 (B1 ))

√ Z 2√mU X 1 4πn
= −2 U ( d(n)n− 4 e− A2
0 √ √ √
U
2
< n≤b0 U
√
× cos(4π n(B1 + x) + π/4))dx + δQ (B1 ) + Oη (U −1/4+ε(U ) ). (8.24)
j
Taking ν = √
2 U
+ x in Lemma 4.3, we have ν = O(L−1). By (8.24) and Lemma 4.3 we
obtain

Re Q(B1 ) =

m 2 √
√ Z √
2 U X
− 14 − 4πn
X γα cos(4π n(B1 + x) + π/4 + απ/2)dx
= 2 U d(n)n e A √ √
0 √ √
α=0
(4π)α ( n( n + B1 + x))α
U
n≤ 2

+Re δQ (B1 ) + Oη (U −1/4+ε(U ) )

√ Z √m
X
− 41 − 4πn 2 U √
= 2 U d(n)n e A 2 cos(4π n(B1 + x) + π/4)dx
√ √ 0
U
n≤ 2

+Re δq + Re δQ (B1 ) + Oη (U −1/4+ε(U ) ), (8.25)

where B1 is (8.2) and

δQ (B1 ) = δQ0 (B1 ) + δQ1 (B1 ) ≪ U 1/4+ε ,

√
2 5 L
√ X
− 41
X γα
Z
A 2 θ2
δq = 2 U d(n)n √ A √ e−πA jαn (θ)dθ, (8.26)
√ √
α=1
(4π n)α − 5 AL
U
n≤ 2
m
√ +θ √
cos(4π n(B1 + x) + π/4 + απ/2)dx
Z
2 U
jαn (θ) = √
θ ( n + B1 + x)α
1
≪ √ , for α ≥ 1.
nU
Hence X
δq ≪ d(n)n−5/4 = O(1). (8.27)
√ √
n≤ 2U

By (8.25) and (8.26) we obtain (8.3) and the proof of (8.3) is complete.

Note
Like §7, we need to write down the accurate δQ (B1 ):

δQ (B1 ) = δQ1 (B1 ) + ReδQ0 (B1 ) + δq (B1 ), (8.28)

59
where δQ1 (B1 ) is (7.44), δq (B1 ) is (8.26), δQ0 (B1 ) is (8.22), i.e.
√
2 5 L
√ X
− 14
X γα
Z
A 2 θ2
δq (B1 ) = 2 U d(n)n √ A √ e−πA
√ √
α=1
(4π n)α − 5 AL
U
n≤ 2
m
√ +θ √
cos(4π n(B1 + x) + π/4 + απ/2)dx
Z
2 U
× √
θ ( n + B1 + x)α
√
U X 3 4πn
= d(n)n− 4 e− A2
2π √ √
U
n≤ 2
2 √ u= m
√
γα cos(4π n(B1 + u) − π/4 + απ/2)
| 2 U
X
× √ √
α=1
(4π n( n + B1 + u))α u=0

+O(U −1/4+ε(U ) ), (8.29)

√
2 5 L
i
Z
X 1
X A 2 θ2
δQ0 (B1 ) = d(n)n− 4 γα ( √ )α A √ e−πA dθ
√ √
α=0
4π n − 5 AL
U
n≤ 2
√
×e(2 n(B1 + θ) + 1/8)ρδ (n, α, θ), (8.30)

moreover,
√
e( √nu ) u=m
|
X U
ρδ (n, α, θ) = √
n
10 10
−U ≤b≤U ,b6=0 2πi( √
U
− b)(B1 + θ + u
√
2 U
)α u=0
√
X Z m αe(( √Un − b)u)du
+ √ √
−U 10 ≤b≤U 10 ,b6=0 0 4πi U( √Un − b)(B1 + θ + u
√
2 U
)α+1
√
e( √nm )
1 U
+ + m α
2(B1 + θ)α 2(B1 + θ + √
2 U
)
√
u=me( √nu )
|
X U
= √
n u α u=0
−U 10 ≤b≤U 10 ,b6=0 2πi( U − b)(B1 + 2 U )
√ √
√
1 e( √nm
U
)
+ α
+ m + ρ∗δ (n, α, θ), (8.31)
2(B1 ) 2(B1 + 2√U )α

60
and

ρ∗δ (n, α, θ) =
√
e( √nu ) 1 1 u=m
|
X U
= √ ( u − )
2πi( √Un − b) (B1 + θ + √
2 U
)α (B1 + 2√uU )α u=0
−U 10 ≤b≤U 10 ,b6=0
√
1 1 1 1 e( √nu
U
) 1
+ ( − α) + ( u − )
2 (B1 + θ) α B1 2 (B1 + θ + 2 U )
√ α (B1 + 2√uU )α
√
X Z m αe(( √Un − b)u)du
+ √ √n .
u α+1
10 10
−U ≤b≤U ,b6=0 0 4πi U( √ − b)(B1 + θ + √
U 2 U
)

By Lemma 3.5,
2
X
− 41
X i √
d(n)n γα ( √ )α e(2 n(B1 + θ) + 1/8)ρ∗δ (n, α, θ) ≪ U −1/4+ε(U ) . (8.32)
√ √
α=0
4π n
n≤ 2U

1
Clearly, as α = 1, 2, the sum (8.30) is O(U − 4 +ε(U ) ). Using
5L
√
Z
A 2 θ2 4πn
A e−πA e(2 nθ)dθ = e− A2 + O(U −10 ,
− 5L
A

it follows from (8.30), (8.31) and (8.32) that

Re δQ0 (B1 ) =
√ u
cos(4π n(B1 + √ ) − π4 ) u=m
1 4πn
|
X X 2 U
= d(n)n− 4 e− A2 ( √
√ √
−U 10 ≤b≤U 10 ,b6=0 2π( √Un − b) u=0
n≤ 2U
√ √ m π
cos(4π nB1 + π4 ) cos(4π n(B1 + 2√U ) + 4 ) 1
+ + ) + O(U − 4 +ε(U ) )
2 √
2
n √
cos π √
U √ u π u=m
− 41 − 4πn
|
X U
= d(n)n e A (( 2 √ −
n
√ ) cos(4π n(B1 + √ ) − )
√ √ 2 sin π √U 2π n 2 U 4 u=0
U
n≤ 2

√ √ m π
cos(4π nB1 + π4 ) cos(4π n(B1 + 2√U ) + 4 ) 1
+ + ) + O(U − 4 +ε(U ) ), (8.33)
2 2
where the last equality is given by the following (9.25).

61
9 EVALUATION OF Ω
By (5.13), (8.1) and (8.3),
m m
X √ Z ξ2 X
S(B) = 2 2Re T (B + ξ)dξ + O(U −1 )
k=0 ξ1 k=0
√ Z ξ2
= 2 2 Re Q(B1 )dξ + O(U −1 )
ξ1
m
√ Z ξ2 X
−1/4 − 4πn
Z √
2 U √
= 4 2U dξ d(n)n eA2 cos(4π n(B1 + x) + π/4)dx
ξ1 √ √ 0
U
n≤ 2
Z ξ2
+Re δQ (B1 )dξ + Oη (U −3/4+ε(U ) ), (9.1)
ξ1
√ √
where 0 < ε(U) ≪ 1/ log log U, δQ (B1 ) ≪ U 1/4+ε(U ) , ξ1 = 1/16 U, ξ2 = 3/16 U and

k √ j
B1 = B + ξ − √ = U + √ + η + ξ.
2 U 2 U
Let m
1 X
− πj
2 X
R(η, m) = √ e V S(B), (9.2)
V √ √
k=0
− V L≤j≤ V L

then by (9.1),
m
√ X
−1/4 − 4πn
Z ξ2 Z √
2 U
R(η, m) = 4 2U d(n)n e A2 dξ v(η, ξ, x)dx
√ √ ξ1 0
U
n≤ 2

+Re δR (η, m) + Oη (U −3/4+ε(U ) ), (9.3)

where
δR (η, m) ≪ U −1/4+ε (9.4)
or accurately,
ξ2
1 −2
Z
− πjV
X
δR (η, m) = √ e δQ (B1 )dξ (9.5)
V √ √ ξ1
− V L≤j≤ V L

((8.2) and (9.5) deduce (9.4)), moreover,

1 X πj −2 √ j
v(η, ξ, x) = √ e− V cos(4πn( U + √ + ξ + η + x) + π/4)
V √ √ 2 U
− V L≤j≤ V L

∞ √
√ √ 1 X − πj−2 nj
= Re e(2 n( U + ξ + η + x) + 1/8) √ e V e( √ ) + 0(U −10 ).
V j=−∞ U

62
By Lemma 3.2,
∞
√ √ X √
−πV ( √ n −j)2
v(η, ξ, x) = Re e(2 n( U + ξ + η + x) + 1/8) e U + 0(U −10 )
j=−∞

∞
− πV n √ √ X √
−πv( √ n −j)2
= e U cos(4π n( U + ξ + η + x) + π/4) + O( e U ) + 0(U −10 )
j=−∞,j6=0
− πV n √ √
= e U cos(4π n( U + ξ + η + x) + π/4) + 0(U −10 ) (9.6)
√
for √n ≤ 1/2. Since
U
πV n 2 UL2
e− U ≪ e−L , n > ,
V
then it follows from (9.3) and (9.6) that
√ X V 4
R(η, m) = 4 2U d(n)n−1/4 e−πn( U + A2 )
2
n≤ UVL
m
ξ2 √
√ √
Z Z
2 U
× dξ cos(4π n( U + ξ + η + x) + π/4)dx
ξ1 0

+δR (η, m) + O(U −3/4+ε )

√
2U X V 4
= d(n)n−3/4 e−πn( U + A2 )
π U L2
n≤ V
m
ξ2 √ √ x= √
Z
× cos(4π n( U + ξ + η + x) − π/4)dx |x=0
2 U

ξ1

+δR (η, m) + Oη (U −3/4+ε(U ) ). (9.7)

It is obvious that
k k
η1 = √1 ηK = √1
f (η) |
ηK
|
= f (η1 + · + ηK ) η =0 · · · η =0
1 K
U
| U

K  
X K ak
= (−1)K−a f(√ )
a=0
a U
K  
K
X
K−a K ak1
= (−1) f (0) + (−1) f ( √ ). (9.8)
a=1
a U

By (9.7), (9.8) and definition (1.28) of Oη we have

R(η, m) | ηK
= R0 + Rm + Rmk + δR (η, m) | ηK
+ O(U −3/4+ε(U ) ), (9.9)

63
where 0 < ε(U) ≪ 1/ log log U,
√
K+1 2U V 4
X
R0 = (−1) d(n)n−3/4 e−πn( U + A2 )
π 2
n≤ UVL
ξ2 √ √
Z
× cos(4π n( U + ξ) − π/4)dξ, (9.10)
ξ1
√
(−1)K 2U X V 4
Rm = d(n)n−3/4 e−πn( U + A2 )
π 2
n≤ UVL
ξ2 √ √ m
Z
× cos(4π n( U + ξ + √ ) − π/4)dξ, (9.11)
ξ1 2 U
√ Z ξ2 X K  
2U X −3/4 −πn( V + 42 ) K−a K
Rmk = d(n)n e U A dξ (−1)
π U L2
ξ1 a=1
a
n≤ V
m
√ √ ak x= √

× cos(4π n( U + ξ + x + √ ) − π/4)
U
|x=0
2 U
. (9.12)

Therefore,
2m0 2m0
1 2
− πj
| |
X X X
Ω = √ e V S(B) ηK
= R(η, m) ηK
V √ √
k,m=m0 k,m=m0
− V L≤j≤ V L

= Ω0 + Ω1 + Ω2 + Ωδ + O(U 1/4+ε(U ) ), (9.13)

where
2m0
X
Ω0 = R0 = (m0 + 1)2 R0 , (9.14)
k,m=m0
2m0
X 2m0
X
Ω1 = Rm = (m0 + 1) Rm , (9.15)
k,m=m0 m=m0
2m0
X
Ω2 = Rmk , (9.16)
k,m=m0
2m0
|
X
Ωδ = δR (η, m) ηK
(9.17)
k,m=m0

(we have replaced k1 by k), R0 , Rm , Rmk and δR (η, m) are (9.10), (9.11), (9.12) and (9.5),
respectively. By (9.4),
Ωδ ≪ U 3/4+ε(U ) . (9.18)
We first evaluate Ω1 . By (9.11) and (9.15),
√ X
−3/4 −πn( V + 42 )
Z ξ2
Ω1 = C1 U(m0 + 1) d(n)n e U A w1n (ξ)dξ, (9.19)
√ 2 ξ1
n≤ UVL

64
where C1 = O(1) is real,
2m0
X √ √ m
w1n (ξ) = cos(4π n( U + ξ + √ ) − π/4). (9.20)
m=m0 2 U

By Lemma 3.3,
2m0 √ √ u
Z
w1n (ξ) = cos(4π n( U + ξ + √ ) − π/4)du + wδ′ + wδ′′ + O(U −10 )
m 2 U
√0
U √ √ um0 u=2
= √ cos(4π n( U + ξ + √ ) − 3π/4)
2π n 2 U u=1
|
′ ′′ −10
+wδ + wδ + O(U ) (9.21)

where
2
1X √ √ bm0
wδ′ = cos(4π n( U + ξ + √ ) − π/4), (9.22)
2 2 U
b=1
2m √ √ u
X Z 0

wδ′′ = Re e(2 n( U + ξ + √ ) − bu − 1/8)du,

−U 10 ≤b≤U 10 ,b6=0 m0
2 U
√ √
1 e(2 n( U + ξ + 2√uU ) − 1/8) u=2m0
|
X
= Re √ (9.23)
2πi 10 10 ( √ n − b) u=m0
−U ≤b≤U ,b6=0 U

for e(−bu) = 1, u = m0 , 2m0 . It is easy to test Fourier’s formula:

∞
1 1 X 2α(−1)b cos(bx)
cos(αx) = sin(απ)( + ), (9.24)
π α b=1 α2 − b2

0 < α < 1, −π ≤ x ≤ π.
√ √
Taking x = π, α = n/ U , we have
√
X 1
∞
X 2 √Un −10
√
n
= √
n
+ O(U )
√ −b ( √ )2 − b2
−U 10 ≤b≤U 10 ,b6=0 U b=1 U
√ √
π cos(π √Un ) U
= √ − √
sin(π √Un ) n
2
√ √
π n n
= − √ (1 + ϕ( √ )) + O(U −10 ), (9.25)
3 U U
√ √ 2
where ϕ( √Un ) = C1 Un + C2 ( Un )2 + · · · for n ≤ UVL . By Lemma 3.5,

√ Z ξ2 X V 4
U (m0 + 1) dξ d(n)n−3/4 e−πn( U + A2 ) wδ′′
ξ1 √ 2
n≤ UVL

65
 √ Z ξ2
≪ U (m0 + 1) U −1/2+1/4+ε(U ) dξ ≪ U 1/4+ε(U ) . (9.26)
ξ1

It follows from (9.19), (9.21), (9.22) and (9.26) that

Z ξ2 X V 4
Ω1 = C1 U(m0 + 1) dξ d(n)n−5/4 e−πn( U + A2 )
ξ1 2
n≤ UVL

√ √ um0 u=2
× cos(4π n( U + ξ + √ ) − 3π/4)
2 U
|
u=1
1/4+ε(X)
+Ω1 δ + O(X ), (9.27)

where U = [X],
2 Z
√ X ξ2 X V 4
Ω1δ = C2 U (m0 + 1) d(n)n−3/4 e−πn( U + A2 )
b=1 ξ1 2
n≤ UVL
√ √ bm0
× cos(4π n( U + ξ + √ ) − π/4)dξ. (9.28)
2 U
Furthermore,
1
V 4
Z
−πn( V + 4
) V 4
e U A2 − 1 = −πn( + 2 ) e−θπn( U + A2 ) dθ, (9.29)
U A 0
and
√ √ um0
cos(4π n( U + ξ + √ ) − 3π/4)
2 U
√ √ um0 √ √ √ um0 1 1
= Re e(2 n( X + √ ) − 3/8)e(2 n(ξ + U − X + ( √ − √ )))
2 X 2 U X
√ √ um0
= cos(4π n( X + √ ) − 3π/4)
2 X
√ √ √ um0 1 1 √ √ um0
−4π n(ξ + U − X + ( √ − √ )) cos(4π n( X + √ ) − π/4)
2 U X 2 U
√ √ um0
+Re(e(2 n( X + √ ) − 3/8)δnu (ξ)), (9.30)
2 X
where
√ √ um0 1 1
δnu (ξ) = n(4πi(ξ + U − X + ( √ − √ )))2
2 U X
Z 1Z 1
√ √ √ um0 1 1
× θe(2θθ1 n(ξ + U − X + ( √ − √ )))dθdθ1 . (9.31)
0 0 2 U X
Using (9.29) and Lemma 3.5,
ξ2 √ √ um0
Z
−πn( V 4
X
−5/4 + )
U(m0 + 1) d(n)n (e U A2 − 1) cos(4π n( U + ξ + √ ) − 3π/4)dξ
2 ξ1 2 U
n≤ UVL

66
 ≪ U 1/4+ε(U ) . (9.32)
− 12
Since U = [X], ξ = O(U ), then
√ √ [X] − X {X} 1
ξ+ U− X=ξ+p √ = ξ − √ + O( ), (9.33)
[X] + X 2 X X

um0 1 1 1
( √ − √ )) = O( ), (9.34)
2 U X X
√
for m0 ≍ UL−2 . It follows from (9.31), (9.33), (9.34) and Lemma 3.5 that
ξ2 √ √ um0
X Z
−5/4
U(m0 +1) d(n)n e(2 n( X + √ )−3/8)δnu (ξ)dξ ≪ X 1/4+ε(X) . (9.35)
2 ξ1 2 X
n≤ UVL

Therefore, by (9.27), (9.32), (9.33), (9.34) and (9.35) we obtain that

ξ2 √ √ um0
X Z
−5/4
Ω1 = C1 U(m0 + 1) d(n)n (cos 4π n( X + √ ) − 3π/4)
2 ξ1 2 X
n≤ UVL

√ {X} 1 √ √ um0
+C2 n(ξ − √ + O( )) cos(4π n( X + √ ) − π/4)dξ
2 X X 2 X
1/4+ε(U )
+Ω1δ + O(X )
√ √ √ um0 u=2
|
X
= C0 U (m0 + 1) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4) u=1
U L2
2 X
n≤ V

√ √ um0 u=2
|
X
+(m0 + 1)(C1 + C2 {X}) d(n)n−3/4 cos(4π n( X + √ ) − π/4) u=1
√ 2 2 X
n≤ UVL

+Ω1δ + O(X 1/4+ε(X) ) (9.36)

√ √
for ξ1 = 1/16 U , ξ2 = 3/16 U . In the same way, by (9.28),
2
X X √ √ am0
Ω1δ = (m0 + 1) Ca d(n)n−3/4 cos(4π n( X + √ ) − π/4) + O(X 1/4+ε(X) ).
a=1 √ 2 2 X
n≤ UVL
(9.37)
By Lemma 3.4,
X √ √ um0
d(n)n−5/4 cos(4π n( X + ) − 3π/4) ≪ X −3/4+ε (9.38)
X L2
2X
U L2
V
<n≤ 2V

and √ √
X am0
d(n)n−3/4 cos(4π n( X + ) − π/4) ≪ X −1/4+ε . (9.39)
X L2
2X
U L2
V
<n≤ 2V

67
Therefore, by (9.36), (9.37), (9.38) and (9.39), we may rewrite Ω1 in the form
2
√ X X √ √ am0
Ω1 = (m0 + 1) X Ca d(n)n−5/4 cos(4π n( X + √ ) − 3π/4)
a=1 X2 L2
2 X
n≤ V
2
X
+(m0 + 1) (C2 (a) + C3 (a){X})
a=1

√ √ am0 u=2
|
X
× d(n)n−3/4 cos(4π n( X + √ ) − π/4) u=1
+ O(X 1/4+ε(U ) ) (9.40)
X2 L2
2 X
n≤ V

for X1 ≤ X ≤ X2 , U = [X], where Cj (a) = O(1) is real and independent of X.

Next we evaluate Ω2 in (9.16). By (9.16) and (9.12),
√
2U X V 4
Ω2 = d(n)n−3/4 e−πn( U + A2 )
π 2
n≤ UVL
K  Z ξ2
X
K K m
× (−1) (wna (ξ, √ ) − wna (ξ, 0))dξ
a=1
a ξ1 2 U
= Ω′2 + Ω20 , say, (9.41)

where
2m0
m X √ √ ak m
wna (ξ, √ ) = cos(4π n( U + ξ + √ + √ ) − π/4) (9.42)
2 U k,m=m0
U 2 U
2m0
X √ √ ak
wna (ξ, 0) = cos(4π n( U + ξ + √ ) − π/4)
k,m=m0
U
2m0
X √ √ ak
= (m0 + 1) cos(4π n( U + ξ + √ ) − π/4). (9.43)
k=m0
U

68
Repeating the evaluation of (9.40) we have
√
2U X V 4
Ω20 = − (m0 + 1) d(n)n−3/4 e−πn( U + A2 )
π 2
n≤ UVL
K 2m0
√ √
  Z ξ2 X
X
K K ak
× (−1) cos(4π n( U + ξ + √ ) − π/4)dξ
a=1
a ξ1 k=m U
0
4K
√ X X √ √ am0
= (m0 + 1) X C1 (a) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4)
a=1 X L2
2 X
n≤ 2
V
4K
X
+(m0 + 1) (C2 (a) + C3 (a){X})
a=1
X √ √ am0
× d(n)n−3/4 cos(4π n( X + √ ) − π/4) + O(X 1/4+ε(U ) ), (9.44)
X L2
2 X
n≤ 2V

where Cj (a) is real, and

 
K L
|Cj (a)| ≪ sup ≪ 2K ≪ exp(O( )). (9.45)
0≤a≤K a log(L)

Next we evaluate Ω′2 in (9.41). By (9.42) and Lemma 3.3

m √ √
wna (ξ, √ ) = Re (e(2 n( U + ξ) − 1/8)ρna ), (9.46)
2 U
where
2m0
X √ ak m
ρna = e(2 n( √ + √ )) = I1 I2 + I1 σ2 + I2 σ1 + σ1 σ2 + O(U −9 ), (9.47)
k,m=m0
U 2 U
moreover,
2m0 √ √ √
2 nau U 2 naum0 u=2
Z
I1 = e( √
U
)du =
2πia n
√ e( √
U
|
) u=1 , (9.48)
m0
Z 2m0 √ √ √
nu U num u=2
I2 = e( √ )du =
U 2πi n
√ e( √ 0 )
U
| u=1
,
m0
√
2 2 nam0 u
1 2abm0 e( √ ) u=2
|
X X U
σ1 = e( √ ) + 2a
√
2 U √ n u=1
b=1 −U 10 ≤b≤U 10 ,b6=0 2πi( U − b)
√
2 2 e( 2 nam u √ √
√ 0 )
1 2abm0 π n 2a n u=2
|
X U
= e( √ ) − √ (1 + ϕ( √ )) (9.49)
2 b=1 U 3 2πi U U u=1
√
2 √ nm0 u √ √
1X nbm0 π 2 e( √U ) n n u=2
σ2 =
2 b=1
e( √
U
)−
3 2πi
√ (1 + ϕ( √ ))
U U u=1
|
. (9.50)

69
The last equality is given by (9.25). Since σ1 σ2 ≪ 1, then
√ X V 4
U d(n)n−3/4 e−πn( U + A2 )
√ 2
n≤ UVL

K−a ξ2 √ √
 
K
X Z
K−a
× (−1) σ1 σ2 e(2 n( U + ξ) − 1/8)dξ ≪ X 1/4+ε(X) (9.51)
a=1
a ξ1

By (9.41), (9.46), (9.47) and (9.51), running the process of the evaluation of (9.40), we
obtain
√ X V 4
Ω′2 = C U Re d(n)n−3/4 e−πn( U + A2 )
2
n≤ UVL

K ξ2 √ √
 
K
X Z
K
× (−1) (I1 I2 + I1 σ2 + I2 σ1 ) e(2 n( U + ξ) − 1/8)dξ + O(X 1/4+ε(X) ).
a=1
a ξ1

Thus, we may rewrite in the form:

4K+2
X X √ √ am0
Ω′2 =X C1 (a) d(n)n−7/4 cos(4π n( X+ √ )−π/4)
a=1 X2 L2
2 X
n≤ V

4K+2
√ X X √ √ am0
+ X (C2 (a) + C3 (a){X}) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4)
a=1 X2 L2
2 X
n≤ V

+O(X 1/4+ε(X) ), (9.52)

where Cj (a) satisfies (9.45).
Finally, by (9.13), (9.40), (9.44) and (9.52) we obtain that
Ω = Ω0 + Ω1 + Ω′2 + Ω20 + Ωδ + O(X 1/4+ε(X) )
4K+2
X X √ √ am0
= X C1 (a) d(n)n−7/4 cos(4π n( X + √ ) − π/4)
a=1 X L2
2 X
n≤ 2
V
4K
√ X X √ √ am0
+ X(m0 + 1) C2 (a) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4)
a=1 X2 L2
2 X
n≤ V
√
+ X
4K+2
X X √ √ am0
× (C3 (a) + C4 (a){X}) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4)
a=1 X L2
2 X
n≤ 2
V

+(m0 + 1)
4K
X X √ √ am0
× (C5 (a) + C6 (a){X}) d(n)n−3/4 cos(4π n( X + √ ) − π/4)
a=1 X L2
2 X
n≤ 2
V

+Ω0 + Ωδ + O(X 1/4+ε(U ) ) (9.53)

70
 √
for U = [X], m0 ≍ XL−2 , where Cj (a) is real and independent of X,
L
|Cj (a)| ≪ 2K ≪ exp(O( )), (9.54)
log(L)
Ω0 is (9.14), Ωδ is (9.17) and (9.18).

10 PROOF OF THEOREM
By (9.14) and (9.10),
√
(−1)K+1 2U(m0 + 1)2
Ω0 =
π
ξ2 √ √
Z
−3/4 −πn( V + 42 )
X
× d(n)n e U A cos(4π n( U + ξ) − π/4)dξ. (10.1)
2 ξ1
n≤ UVL

Since V 4 2
e−πn( U + A2 ) ≪ e−L
U L2 X2 L 2 √1 ξ ′ ,
for V
<n≤ V
, then putting ξ = U
we have

2(−1)K+1 (m0 + 1)2 X V 4

Ω0 = √ d(n)n−3/4 e−πn( U + A2 )
2π X L2
n≤ 2
V
3/16 √ √ √
Z
× cos(4π n( U + ξ/ U ) − π/4)dξ + O(U −10 )
1/16

(−1) K+1
(m0 + 1)2 X 1/4 X −3/4
√
= ( √ d(n)n cos(4π nX − π/4)
4X 1/4 2π 2 X2 L
n≤ V

+δ01 (X) + δ02 (X) + O(U −1 )), (10.2)

where

δ01 (X) = C1 X 1/4

3/16 √ √ √
Z
−πn( V 4
X
−3/4 + )
× d(n)n (e U A2 − 1) cos(4π n( U + ξ/ U ) − π/4)dξ, (10.3)
X2 L2
1/16
n≤ V

δ02 (X) = C2 X 1/4

3/16 √ √ √ √
X Z
−3/4
× d(n)n (cos(4π n( U+ξ/ U)−π/4)−cos(4π nX−π/4))dξ,
X2 L2
1/16
n≤ V
(10.4)

71
with C1 , C2 = O(1) be real and independent of X. Denote

X 1/4 X √
δ00 (X) = ∆(X) − √ d(n)n−3/4 cos(4π nX − π/4). (10.5)
π 2 X L2
n≤ 2
V

By (10.2),
(−1)K+1(m0 + 1)2
Ω0 = (∆(X) − δ0 (X) + O(U −1 )), (10.6)
4X 1/4
where
δ0 (X) = δ00 (X) − δ01 (X) − δ02 (X). (10.7)
For X1 ≤ X ≤ X2 , X1 ≍ X2 , V = X2ε0 , ε0 = 1/ log L, (2.59) implicates

δ00 (X) ≪ X ε . (10.8)

Using (9.29) and taking α = −1/4 in Lemma 3.4, we have

δ01 (X) ≪ X ε . (10.9)

Since
√ √ √ √
cos(4π n( U + ξ/ U) −π/4) −cos(4π nX −π/4) =
√ √ √ √ √
= Re e(2 nX − 1/8)(e(2 n( U + ξ/ U − X)) − 1)
√ √ √ √ √ Z 1 √ √ √ √
= nRe e(2 nX − 1/8)4πi( U − X + ξ/ U ) e(2θ n( U − X + ξ/ U ))dθ,
0
then using Lemma 3.4, we have
δ02 (X) ≪ X ε .
Thus,
δ0 (X) = δ00 (X) − δ01 (X) − δ02 (X) ≪ X ε . (10.10)
Moreover, by (9.53) and (10.6),

(−1)K+1 (m0 + 1)2

Ω= (∆(X) − X 1/4 Λ(X) − δ(X) + O(X −1/2+ε(X) )), (10.11)
4X 1/4
where
4K+2
X X X
−7/4
√ √ am0
Λ(X) = 2
C 1 (a) d(n)n cos(4π n( X + √ ) − π/4)
(m0 + 1) a=1 X L2
2 X
n≤ 2
V

√ 4K
X X X √ √ am0
+ C2 (a) d(n)n−5/4 cos(4π n( X + √ ) − 3π/4), (10.12)
m0 + 1 a=1 X L2
2 X
n≤ 2
V

δ(X) = δ0 (X) + δ1 (X) + δ2 (X), (10.13)

72
The three terms in the right side of the last equality are the following:

δ0 (X) is (10.10),

δ1 (X)
4K+2
X 3/4 X X
−5/4
√ √ am0
= 2
(C 3 (a) + C 4 (a){X}) d(n)n cos(4π n( X + √ ) − 3π/4)
(m0 + 1) a=1 2 X
X L2 2
n≤ V

4K
X 1/4 X X √ √ am0
+ (C5 (a) + C6 (a){X}) d(n)n−3/4 cos(4π n( X + √ ) − π/4),
(m0 + 1) a=1 2 X
X L2 2
n≤ V

2m0
X 1/4 X 1/4 X
δ2 (X) = 2
Ωδ = 2
δR (η, m)|ηK ≪ X ε , (10.14)
(m0 + 1) (m0 + 1) k,m=m
0

where Ωδ is (9.17), δR (η, m) is (9.5), Cj (a), δ2 (X) contain new constant C = O(1) to be
independent of X.
Therefore, by (1.26) and (10.11) we obtain that
1
∆(X) = X 1/4 Λ(X) + δ(X) + O(X −1/2+ε(X) ), 0 < ε(X) = O( ). (10.15)
log log X
The proof of Theorem is complete. 

NOTE
For the convenient of the other application afterwards we write down the accurate
δ(X) of (10.15):

X 1/4 X √
δ(X) = ∆(X) − √ d(n)n−3/4 cos(4π nX − π/4)
2π X L2
n≤ 2
V

3/16 √ √ √
Z
−πn( V 4
X
1/4 −3/4 + )
+C1 X d(n)n (e U A2 − 1) cos(4π n( U + ξ/ U) − π/4)dξ
X2 L2
1/16
n≤ V

3/16 √ √ √ √
X Z
1/4 −3/4
+C2 X d(n)n (cos(4π n( U+ξ/ U)−π/4)−cos(4π nX−π/4))dξ
X2 L2
1/16
n≤ V

4K+2
X 3/4 X X
−5/4
√ √ am0
+ 2
(C 3 (a)+C 4 (a){X}) d(n)n cos(4π n( X + √ )−3π/4)
(m0 + 1) a=1 2 2 X
X2 L
n≤ V

4K
X 1/4 X X √ √ am0
+ (C5 (a) + C6 (a){X}) d(n)n−3/4 cos(4π n( X + √ ) − π/4)
(m0 + 1) a=1 2 X
X L2 2
n≤ V

73
 2m0 Z ξ2
X 1/4 X
− πj
2 X
+ √ e V δQ (B1 )dξ|ηK , (10.16)
(m0 + 1)2 V √ √
k,m=m0 ξ1
− V L≤j≤ V L
√
where δQ (B1 ) is (8.28),
√ X1 ≤ X√ ≤ X2 , U √= [X], m0 ≍ X2 L−2 , X1 ≍ X2 ,
L = log X2 , ξ1 = 1/16 U , ξ2 = 3/16 U , A = C0 UL Cj (a) is independent of X, and

L
|Cj (a)| ≪ exp(O( )).
log L
√
We can take V = X2ε0 , ε0 = 1/ log L, m0 = [ X2 L−2 ], such that they are independent
of X.

Acknowledgements. I would like to thank Professor Liang Zhang for pointing out
some errors and checking the paper.

References
[1] E.C.Titchmarsh, (revised by Heatu-Brown D.R.) Theory of The Riemann Zeta-
Function, Second edition, Oxford, 1986.

[2] Ivic′ ,A., The Riemann Zeta-Function, John Wiley & Sons, New York, 1985.

[3] Raymond Ayoub, An Introduction To The Analytic Theory Of Number, American

Mathematical Society, 1963.

[4] Dirichlet,P.G.L. Über die Bestimmung der mittleren werte in der Zahlentheorie,
Abh.Akad.Berlin(werke 2, 49-
66)1849, Math. Abh., 69-83

[5] Voronoi,G., Sur un problem du calcul des functions asymptotiques, J. für math.
126(1903)241-282

[6] Corput,J.G. Van der,Verschärfung der Abschätzung beim Teilerproblem, M. A.

87(1922)39-65

[7] Corput,J.G. Van der, Zum Teilerproblem, M.A. 98(1928)697-716

[8] Chih, T.T. A divisor problem, Acad. Sinica Sci. Record, 3(1950)177-182

[9] Richert,H.E., Verschärfung der Abschätzung beim Dirichletschen Teilerproblem,

Math. Z, 58, (1953)204-
218

[10] Yin Wenlin, Dirichlet Divisor Problem, J. Peking University(Natural Science),

2(1959), 103-126

74
[11] Kolesnik,G., The improvement of the error term in the divisor problem,
Mat.Zametki, 6(1969), 545-
554

[12] Kolesnik,G., On the estimation of certain trigonometric sum, Acta Arith. 25(1973)7-
30

[13] Kolesnik,G., On the order of ζ(1/2 + it) and ∆(R), Pacific J. Math. 82(1982),
107-122

[14] Kolesnik,G., On the method of expanent pairs, Acta Arith. 45(1985), 115-143

[15] Iwaniec,H. and Mozzochi,C.J., On the divisor and circle problem, J.Number Theory
29(1988), no.1, 60-
93

[16] Hardy,G.H., On Dirichlet’s divisor problem,Proc. London Math. Soc. 15(1916),1-25

[17] Matti Jutila, Transformation formule for Dirichlet polynomials, Journal of number
18, 135-
156(1984)

[18] M.N.Huxley, Exponential Sums and Lattice Points III, Proc. London Math. Soc.
(3)87:591-
609(2003)

75