XC.1 (1999)
On the number of coprime integer pairs within a circle
by
Wenguang Zhai (Jinan) and Xiaodong Cao (Beijing)
1. Introduction. Let P (x) denote the number of integer pairs within the circle a
2+ b
2≤ x, and E(x) denote the difference P (x) − πx. Then the well-known circle problem is to estimate the upper bound of E(x) and the best result at present is
(1.1) E(x) = O(x
23/73+ε).
See Huxley [6].
Let V (x) denote the number of coprime integer pairs within the circle a
2+ b
2≤ x. It is an exercise to deduce that
(1.2) V (x) = 6
π x + O(x
1/2exp(−c log
3/5x(log log x)
−2/5)),
where c is some absolute constant. The problem of reducing the exponent 1/2 is open. One way to make progress is to assume the Riemann Hypothesis (RH). W. G. Nowak [11] proved that RH implies
(1.3) V (x) = 6
π x + O(x
15/38+ε).
D. Hensley [5] also got a result of this type, but with a larger exponent.
The aim of this paper is to further improve this result. We have the following
Theorem. If RH is true, then
(1.4) V (x) = 6
π x + O(x
11/30+ε).
Notations. e(x) = exp(2πix). m ∼ M means c
1M ≤ m ≤ c
2M for absolute constants c
1and c
2. E(x) always denotes the error term in the circle problem. ε denotes an arbitrary small positive number and may be different at each occurrence.
1991 Mathematics Subject Classification: 11N37, 11P21.
This work is supported by Natural Science Foundation of Shandong Province (Grant No. Q98A02110).
[1]
The authors thank Professor W. G. Nowak for kindly sending reprints of some of his papers.
2. Some preliminary lemmas and results. The following lemmas are needed.
Lemma 1. Suppose 0 < c
1λ
1≤ |f
0(n)| ≤ c
2λ
1and |f
00(n)| ∼ λ
1N
−1for N ≤ n ≤ cN . Then
X
N <n≤cN
e(f (n)) λ
−11+ λ
1/21N
1/2.
P r o o f. If c
2λ
1≤ 1/2, this estimate is contained in the Kuz’min–Landau inequality; otherwise, the estimate follows from the well-known van der Cor- put’s estimate for the second order derivative.
Lemma 2. Suppose a(n) = O(1), 0 < L ≤ M < N ≤ cL, L 1, T ≥ 2.
Then
X
M <n≤N
a(n) = 1 2πi
T
\
−T
X
L<l≤cL
a(l)
l
it· N
it− M
itt dt
+ O
min
1, L
T kM k
+ min
1, L
T kN k
+ O
L log(1 + L) T
. P r o o f. This is the well-known Perron formula.
Lemma 3. Suppose f (n) P and f
0(n) ∆ for n ∼ N. Then X
n∼N
min
D, 1 kf (n)k
(P + 1)(D + ∆
−1) log(2 + ∆
−1).
P r o o f. This is contained in Lemma 2.8 of Kr¨atzel [9].
Lemma 4. Suppose a(n) are any complex numbers and 1 ≤ Q ≤ N. Then
X
N <n≤cN
a(n)
2N
Q X
0≤q≤Q
1 − q
Q
< X
N <n≤cN −q
a(n)a(n + q).
P r o o f. This is Weyl’s inequality.
Lemma 5. Let M ≤ N < N
1≤ M
1and a(n) be any complex numbers.
Then
X
N <n≤N1
a(n) ≤
∞
\
−∞
K(θ)
X
M <n≤M1
a(m)e(mθ)
dθ
with K(θ) = min(M
1− M + 1, (π|θ|)
−1, (π|θ|)
−2) and
∞
\
−∞
K(θ) dθ ≤ 3 log(2 + M
1− M ).
P r o o f. This is Lemma 2.2 of Bombieri and Iwaniec [2].
Lemma 6. Let αβ 6= 0, ∆ > 0, M ≥ 1, N ≥ 1. Let A(M, N, ∆) be the number of quadruples (m, e m, n, e n) such that
|( e m/m)
α− (e n/n)
β| < ∆ with M ≤ m, e m ≤ 2M and N ≤ n, e n ≤ 2N . Then
A(M, N, ∆) M N log 2M N + ∆M
2N
2. P r o o f. This is Lemma 1 of [3].
Lemma 7. Suppose f (x) and g(x) are algebraic functions in [a, b] and
|f
00(x)| ∼ 1
R , |f
000(x)| 1 RU ,
|g(x)| G, |g
0(x)| GU
1−1, U, U
1≥ 1.
Then X
a<n≤b
g(n)e(f (n)) = X
α<u≤β
b
ug(n
u)
p |f
00(n
u)| e(f (n
u) − un
u+ 1/8)
+ O(G log(β − α + 2) + G(b − a + R)(U
−1+ U
1−1)) + O(G min( √
R, 1/hαi) + G min( √
R, 1/hβi)),
where [α, β] is the image of [a, b] under the mapping y = f
0(x), n
uis the solution of the equation f
0(x) = u,
b
u=
1 for α < u < β,
1
2
for u = α ∈ Z or u = β ∈ Z, and the function hti is defined as follows:
hti =
ktk if t is not an integer , β − α otherwise,
where ktk = min
n∈Z{|t − n|}.
P r o o f. This is Theorem 2.2 of Min [10].
Lemma 8. Suppose A
i, B
j, a
iand b
jare all positive numbers. If Q
1and Q
2are real with 0 < Q
1≤ Q
2, then there exists some q such that
Q
1≤ q ≤ Q
2and
X
m i=1A
iq
ai+ X
n j=1B
jq
−bj≤ 2
m+nX
mi=1
X
n j=1(A
bijB
jai)
1/(ai+bj)+ X
m i=1A
iQ
a1i+ X
n j=1B
jQ
−b2 j.
P r o o f. This is Lemma 3 of Srinivasan [12].
Lemma 9. Suppose x is a large positive real number , M, N, U are positive integers such that 1 ≤ N ≤ M
1/4, x
ε≤ M ≤ x
4/15and M
1/6≤ U ≤ M
5/6. Then
S = X
n∼N
a
nX
u∼U
b
uX
v∼M U−1
c
ve
√ nx uv
(2.1)
(x
1/12M
7/12N
11/12+ M
11/12N
1/2) log x,
where a
n, b
u, c
vare coefficients satisfying |a
n| ≤ 1, |b
u| ≤ 1, |c
v| ≤ 1.
P r o o f. We use Heath-Brown’s method [4].
Without loss of generality, suppose M
1/2≤ U ≤ M
5/6; otherwise we change the order of U and V = M U
−1. Suppose 1 ≤ Q ≤ V N is a parameter to be determined. For each q (1 ≤ q ≤ V N ) define
w
q=
(v, n)
v ∼ V, n ∼ N, 2 √
N (q − 1) V Q <
√ n v < 2 √
N q V Q
. Then
S = X
u∼U
b
uX
Q q=1X
(v,n)∈wq
a
nc
ve
√ nx uv
. By Cauchy’s inequality
|S|
2U Q X
u∼U
X
Q q=1X
(v1,n1),(v2,n2)∈wq
a
n1a
n2(2.2)
× c
v1c
v2e
√ x u
√ n
1v
1−
√ n
2v
2U Q X
(∗)
X
u∼U
e
√ x u
√ n
1v
1−
√ n
2v
2, where (∗) denotes the condition
(2.3)
√ n
1v
1−
√ n
2v
2≤ 2 √
N
V Q , v
1, v
2∼ V, n
1, n
2∼ N.
Let λ = √
n
1/v
1− √
n
2/v
2. Then by Lemma 1, the inner sum in (2.2) can be estimated by
(2.4) min
U, U
2√ x|λ|
+ x
1/4|λ|
1/2U
−1/2.
By Lemma 6, the contribution of U (namely |λ| ≤ U x
−1/2) to |S|
2is (2.5) U
2Q(V N log 2V N + U x
−1/2N
3/2V
3) U
2QV N log 2V N.
The contribution of U
2x
−1/2|λ|
−1(|λ| > U x
−1/2) is U
2QV N (log 2V N )
2by a similar argument. The contribution of x
1/4|λ|
1/2U
−1/2is (note |λ|
√ N /(V Q), Q V N )
x
1/4U
1/2N
9/4V
3/2Q
−1/2. Now, taking Q = 1 + [x
1/6U
−1N
5/6V
1/3], we get
(2.6) |S| log
−12V N U V
1/2N
1/2+ x
1/12U
1/2V
2/3N
11/12, whence the lemma follows since M
1/2≤ U ≤ M
5/6.
Lemma 10. Suppose x, M, N satisfy the conditions of Lemma 9. Then T (M, N ) = X
n∼N
a(n) X
m∼M
µ(m)e
√ nx m
(2.7)
(x
1/12M
7/12N
11/12+ M
11/12N
1/2)x
ε, where |a(n)| ≤ 1.
P r o o f. By Heath-Brown’s identity (k = 4), T (M, N ) can be written as O(log
8x) sums of the form
(2.8) T = X
n∼N
a(n) X
m1∼M1
. . . X
m8∼M8
µ(m
1) . . . µ(m
4)e
√
nx m
1. . . m
8, where M M
1. . . M
8M, M
1, M
2, M
3, M
4≤ (2M )
1/4. Some m
imay only take value 1.
To prove the lemma we consider three cases.
Case 1: There is some M
isuch that M
i> M
5/6. It must follow that i ≥ 5; i = 8 for example. We use the exponent pair (1/6, 4/6) to estimate the sum on m
8and estimate other variables trivially, to get (F = √
N xM
−1) T N M M
8−1M
8F +
F M
8 1/6M
82/3(2.9)
x
−1/2M
2N
1/2+ x
1/12M
5/12N
13/12x
1/12M
5/12N
13/12x
1/12M
7/12N
11/12.
Case 2: There is some M
isatisfying M
1/6≤ M
i≤ M
5/6. Let u = m
i, v = Q
j6=i
m
j, U = M
i. Then by Lemma 9 we get
(2.10) x
−εT x
1/12M
7/12N
11/12+ M
11/12N
1/2, where x
εcomes from the divisor argument.
Case 3: All M
isatisfy M
i≤ M
1/6. Without loss of generality, suppose M
1≥ M
2≥ . . . ≥ M
8. Let l be the first positive integer j such that
(2.11) M
1. . . M
j> M
1/6; then l ≥ 2. Thus we have
M
1. . . M
l≤ (M
1. . . M
l−1)M
l≤ M
2/6.
Now take u = m
1. . . m
j, v = m
j+1. . . m
6, U = M
1. . . M
j. By Lemma 9, (2.10) still holds.
Lemma 10 follows from the three cases.
Now we prove the following proposition. It plays an important role in this paper. The idea of the proof has been used by several authors; see Jia [8], Baker and Harman [1], for example.
Proposition 1. Suppose M, N are large positive numbers, A > 0, α, β are rational numbers (not non-negative numbers). Suppose m ∼ M, n ∼ N, F = AM
αN
βN, |a(m)| ≤ 1, |b(n)| ≤ 1. Then
S = X
M <m≤2M
X
N <n≤2N
a(m)b(n)e(Am
αn
β)
(M N
1/2+ F
4/20M
13/20N
15/20+ F
4/23M
15/23N
18/23+ F
1/6M
2/3N
7/9+ F
1/5M
3/5N
4/5+ F
1/10M
4/5N
7/10) log
4F.
P r o o f. Without loss of generality, we suppose β > 0; for β < 0, the proof is the same. By Cauchy’s inequality and Lemma 4 we get
|S|
2M X
M <m≤2M
X
N <n≤2N
b(n)e(Am
αn
β)
2(2.12)
M
2N
2Q + M N Q |Σ
1| with
Σ
1= X
Q q=11 − q
Q
X
N <n≤2N −q
b(n)b(n + q) X
M <m≤2M
e(Am
αg(n, q)),
where Q is a parameter satisfying log N ≤ Q ≤ N log
−1N, g(n, q) =
(n + q)
β− n
β.
We write
Σ
1= X
1≤q≤Q
X
2N −Q<n≤2N −q
X
m
+ X
q≤B
X
N <n≤2N −Q
X
m
(2.13)
+ X
B≤q≤Q
X
N <n≤2N −Q
X
m
= Σ
2+ Σ
3+ Σ
4, where B = max(log N, M N /(c(α, β))F ) and
c(α, β) = 2|αβ| max(2
α−1, 1) max(2
β−1, 1).
By Lemma 1 we have Σ
2M N Q
F + F
1/2Q
5/2N
1/2log N, (2.14)
Σ
3M N
2F + F
1/2N
1/2log
2N.
(2.15)
So we only need to bound Σ
4. Notice that Σ
4can be written as the sum of O(log Q) exponential sums of the form
(2.16) Σ
5= X
Q1<q≤2Q1
c(q) X
N <n≤2N −Q
b(n)b(n+q) X
M <m≤2M
e(Am
αg(n, q)), where B ≤ Q
1≤ Q/2, c(q) = 1 − q/Q.
By Lemma 7 we have
(2.17) X
M <m≤2M
e(Am
αg(n, q))
= c
3X
U1<u≤U2
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α)) + O
log F + M N
F q + min
M N
1/2F
1/2q
1/2, max
1 hU
2i , 1
hU
1i
, where U
1= c
5AM
α−1g, U
2= c
6AM
α−1g, g = g(n, q). By Lemma 3, the contribution of the error term to Σ
5is
N Q
1log F + M N
2F
−1log Q
1+ F
1/2Q
3/21N
−1/2.
It can be easily seen that g(n, q) < βqn
β−1for 0 < β < 1, g(n, q) >
βqn
β−1for β > 1. If βqAM
α−1n
β−1− AM
α−1g > log
−1N, then by the trivial estimate we have (i = 5, 6)
(2.18) X
q∼q1
X
n∼N
X
ciAMα−1g<u≤ciβqAMα−1nβ−1
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)× e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α))
F
1/2Q
5/21N
−1/2log
2N.
Now suppose βqAM
α−1n
β−1− AM
α−1g ≤ log
−1N. Notice that the fact that
[c
iβqAM
α−1n
β−1] − [c
iAM
α−1g] = 1 always implies
kc
iAM
α−1gk ≤ c
iβqAM
α−1n
β−1− c
iAM
α−1g = δ
i. We have
(2.19) X
ciAMα−1g<u≤ciβqAMα−1nβ−1
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)M N
1/2F
1/2Q
1/21X
ciAMα−1g<u≤ciβqAMα−1nβ−1
1
M N
1/2F
1/2Q
1/21min
1, δ
ikc
iAM
α−1gk
M N
1/2F
1/2Q
1/21min
1, F Q
21M N
2· 1 kc
iAM
α−1gk
,
thus by Lemma 3, (2.20) X
q∼q1
X
n∼N
X
ciAMα−1g<u≤ciβqAMα−1nβ−1
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)× e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α))
F
1/2Q
3/21N
−1/2log
2N.
Now it suffices to bound
Σ
6= X
Q1<q≤2Q1
c(q) X
N <n≤2N −Q
b(n)b(n + q) (2.21)
× X
u
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α)), where
c
5βqAM
α−1n
β−1< u ≤ c
6βqAM
α−1n
β−1. By Lemmas 2 and 3 we get (choose T = F
10)
(2.22) Σ
6|Σ
7| + F
1/2Q
3/21N
−1/2log N
with
Σ
7= X
Q1<q≤2Q1
c(q) X
N <n≤2N −Q
b(n)b(n + q) X
u
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)(2.23)
× q
itA
itM
(α−1)itn
(β−1)itu
ite(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α)), where t is a real number independent of the variables and
c
7F Q
1(M N )
−1< u ≤ c
8F Q
1(M N )
−1. It is easy to see that
Σ
7= X
u
A
1/(2(1−α))+itM
(α−1)itu
(1−α/2)/(1−α)+itX
N <n≤2N −Q
b(n)n
(β−1)it(2.24)
× X
Q1<q≤2Q1
j(q)b(n + q)g
1/(2(1−α))× e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α))
X
u
(Ag)
1/(2(1−α))u
(1−α/2)/(1−α)(Q
1N
β−1)
1/(2(1−α))Σ
8(u)
F
1/2Q
1/21N
−1/2Σ
8(u
0) for some u
0with
(2.25) Σ
8(u) = X
N <n≤2N −Q
X
Q1<q≤2Q1
j(q)b(n + q)g
01/(2(1−α))× e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α)) , where j(q) = c(q)q
it, 1 g
0= g(Q
1N
β−1)
−11.
Suppose 10 ≤ R ≤ Q
1log
−1/2N is a parameter to be determined. By Cauchy’s inequality and Lemma 4 we get
Σ
8(u)
2N X
n
X
q
j(q)b(n + q)g
1/(2(1−α))0(2.26)
× e(c
4A
1/(1−α)g
1/(1−α)u
−α/(1−α))
2N
2Q
21R + N Q
1R
X
1≤r≤R
|E
r|, where
E
r= X
N <n≤2N −Q
X
Q1<q≤2Q1−r
j(q)b(n + q)g
01/(2(1−α))(n, q)
× j(q + r)b(n + q + r)g
1/(2(1−α))0(n, q + r)
× e(c
4A
1/(1−α)u
−α/(1−α)(g
1/(2(1−α))(n, q) − g
1/(2(1−α))(n, q + r))).
So it reduces to bound E
rfor fixed r. Making the change of variable n + q = l, we have
E
r= X
Q1<q≤2Q1−r
j(q)j(q + r) X
N +q<l≤2N +q−Q
b(l)b(l + r)g
01/(2(1−α))(l − q, q)
× g
01/(2(1−α))(l − q, q + r)
× e(c
4A
1/(1−α)u
−α/(1−α)(g
1/(2(1−α))(l, q) − g
1/(2(1−α))(l, q + r)))
= X
N +Q1<n≤2N +2Q1−r−Q
b(n)b(n + r)
× X
max(Q1,n−2N +Q)<q≤min(2Q1−r,n−N )
j(q)j(q + r)
× g
01/(2(1−α))(n − q, q)g
01/(2(1−α))(n − q, q + r)
× e(c
4A
1/(1−α)u
−α/(1−α)k(n, q, r)), where
k(n, q, r) = g
1/(2(1−α))(n − q, q) − g
1/(2(1−α))(n − q, q + r).
By Lemma 5 we get E
rlog
−1N X
n∼N
X
q∼Q1
j(q)j(q + r)g
01/(2(1−α))(n − q, q)
× g
1/(2(1−α))0(n − q, q + r)e(c
4A
1/(1−α)u
−α/(1−α)k(n, q, r) + θ
0q) , where θ
0is a real number independent of n and q.
Suppose 10 ≤ T ≤ Q
1log
−1/2N is a parameter to be determined. By Cauchy’s inequality and Lemma 4 we get
(2.27) |E
r|
2log
−2N N
2Q
21T + N Q
1T |D
t(r)|, with
D
t(r) = X
n∼N
X
Q1<q≤2Q1−t
j(q)j(q + r)g
1/(2(1−α))0(n − q, q)
× g
01/(2(1−α))(n − q, q + r)j(q + t + r)j(q + t)
× g
01/(2(1−α))(n − q − t, q + t)g
01/(2(1−α))(n − q − t, q + t + r)
× e(c
4A
1/(1−α)u
−α/(1−α)(k(n, q, r) − k(n, q + t, r)))
X
q∼Q1
X
n∼N
φ(n)e(f (n))
,
where
φ(n) = g
01/(2(1−α))(n − q, q)g
1/(2(1−α))0(n − q, q + r)
× g
01/(2(1−α))(n − q − t, q + t)g
1/(2(1−α))0(n − q − t, q + t + r) and
f (n) = c
4A
1/(1−α)u
−α/(1−α)(k(n, q, r) − k(n, q + t, r)).
It is an easy exercise to verify that φ(n) is monotonic and
|f
0(n)| ∼ F rt
Q
1N
2, |f
00(n)| ∼ F rt
Q
1N
3(N < n ≤ 2N );
thus by Lemma 1 we get (2.28) D
t(r) Q
1Q
1N
2F rt + (F rt)
1/2(Q
1N )
1/2= Q
21N
2F rt + (F rtQ
1)
1/2N
1/2. Inserting (2.28) into (2.27) we get
(2.29) |E
r|
2log
−2N Q
21N
2T + Q
31N
3F rT + Q
3/21(N F T r)
1/2.
Notice that (2.29) is also true for 0 < T < 10; then by Lemma 8 choosing a best T ∈ (0, Q
1log
−1/2N ) we get
(2.30) |E
r|
2log
−2N N
1/2Q
5/61F
1/6r
1/6+ N
2/3Q
1+ N Q
1/21+ N
3/2Q
1F
1/2r
1/2. Inserting (2.30) into (2.26) we have
(2.31) Σ
8(u)
2log
−2N
Q
21N
2R + N
3/2Q
11/61F
1/6R
1/6+ N
5/2Q
21F
1/2R
1/2+ N
5/3Q
21+ N
2Q
3/21. This is also true for 0 < R ≤ 10. Choosing a best R ∈ (0, Q
1log
−1/2N ) via Lemma 8 we get
Σ
8(u) log
−2N N
11/14Q
13/141F
1/14+ N
14/16Q
15/161(2.32)
+ N
5/4Q
3/41F
−1/4+ N
5/6Q
1+ N Q
3/41. Inserting (2.32) into (2.24) we have
Σ
7N
4/14Q
20/141F
8/14+ N
6/16Q
23/161F
8/16+ N
3/4Q
5/41F
1/4(2.33)
+ N
1/3Q
3/21F
1/2+ N
1/2Q
5/41F
1/2.
Combining (2.17), (2.18), (2.20), (2.22) and (2.33) we get Σ
5log
−4F N
4/14Q
20/14F
8/14+ N
6/16Q
23/16F
8/16(2.34)
+ N
3/4Q
5/4F
1/4+ N
1/3Q
3/2F
1/2+ N
1/2Q
5/4F
1/2+ N Q + M N
2F
−1+ F
1/2Q
5/2N
−1/2.
Inserting (2.14), (2.15) and (2.34) into (2.12) we get
|S|
2log
−6F M
2N
2Q + M N
18/14Q
6/14F
8/14(2.35)
+ M N
22/16Q
7/16F
8/16+ M N
7/4Q
1/4F
1/4+ M N
4/3Q
1/2F
1/2+ M N
3/2Q
1/4F
1/2+ M N
2+ M
2N
3Q
−1F
−1+ M N
1/2Q
3/2F
1/2.
Since Q < N F, we have
M N
2+ M N
7/4Q
1/4F
1/4M N
3/2Q
1/4F
1/2, M
2N
3(F Q)
−1M
2N
2Q
−1. So we obtain
|S|
2log
−6F M
2N
2Q + M N
18/14Q
6/14F
8/14(2.36)
+ M N
22/16Q
7/16F
8/16+ M N
4/3Q
1/2F
1/2+ M N
3/2Q
1/4F
1/2+ M N
1/2Q
3/2F
1/2.
Note that (2.36) is trivial for 0 < Q ≤ log N. Now the proposition follows from choosing a best Q ∈ (0, N log
−1N ) via Lemma 8.
3. An expression of the error term. In the rest of this paper, we always use E
P(x) to denote the difference V (x)−
π6x. The aim of this section is to give an expression of E
P(x) subject to RH.
Let y be a parameter, x
ε≤ y ≤ x
1/2−ε, (3.1) f
1(s) = X
n≤y
µ(n)n
−s, f
2(s) = ζ
−1(s) − f
1(s).
Let r(n) be the number of representations of n as a sum of two squares.
Then
V (x) = X
a2+b2≤x (a,b)=1
1 = X
a2+b2≤x
X
m|(a,b)
µ(m) (3.2)
= X
m2(a2+b2)≤x
µ(m) = X
m2k≤x
µ(m)r(k)
= X
m≤y
+ X
m>y
= Σ
1+ Σ
2, say.
Notice that for σ > 1, (3.3)
X
∞ n=1r(n)n
−s= 4ζ(s)L(s, χ),
where χ is the non-principal character mod 4, we have Σ
1= X
m≤y
µ(m)P
x m
2(3.4)
= X
m≤y
µ(m)
Res
s=14ζ(s)L(s, χ) s
x m
2 s+ E
x m
2= Res
s=1(4f
1(2s)ζ(s)L(s, χ)x
ss
−1) + X
m≤y
µ(m)E
x m
2. To treat Σ
2we begin with
f
2(s) = X
m>y
µ(m)m
−sfor σ > 1. Hence
(3.5) 4f
2(2s)ζ(s)L(s, χ) = X
∞ n=1b(n)n
−s(σ > 1), where
b(n) = X
n=m2k, m>y
µ(m)r(k).
By Perron’s formula we have
(3.6) Σ
2= X
n≤x
b(n) = 1 2πi
1+ε+ix
\
21+ε−ix2
g(s) ds + O(x
ε), where
g(s) = 4f
2(2s)ζ(s)L(s, χ)x
ss
−1. By Cauchy’s theorem, we have
(3.7) 1
2πi
1+ε+ix
\
21+ε−ix2
g(s) ds = I
1+ I
2− I
3+ Res
s=1g(s), where
I
1= 1 2πi
1+ε+ix
\
20.5+ε+ix2
g(s) ds, I
2= 1 2πi
0.5+ε+ix
\
20.5+ε−ix2
g(s) ds,
I
3= 1 2πi
1+ε−ix
\
20.5+ε−ix2
g(s) ds.
Since RH is true, it follows that
ζ(s) |t|
ε+ 1, σ ≥ 0.5 + ε,
(3.8)
f
2(2s) y
−1/2(|t|
ε+ 1), σ ≥ 0.5 + ε.
(3.9)
For L(s, χ), we have
(3.10) L(s, χ) (|t| + 1)
(1−σ)/2, 0.5 ≤ σ ≤ 1.
Using (3.8)–(3.10) we get
I
1− I
3y
−1/2(1 + x
ε), (3.11)
I
2y
−1/2x
1/2+ε x\
21
|L(1/2 + ε + it, χ)|
t dt + 1
(3.12)
y
−1/2x
1/2+ε. Combining (3.6)–(3.12) we get
(3.13) Σ
2= Res
s=1(4f
2(2s)ζ(s)L(s, χ)x
ss
−1) + O(y
−1/2x
1/2+ε).
From (3.2), (3.4) and (3.13) we get
V (x) = Res
s=1(4ζ
−1(2s)ζ(s)L(s, χ)x
ss
−1) (3.14)
+ X
m≤y
µ(m)E
x m
2+ O(y
−1/2x
1/2+ε)
= 6
π x + X
m≤y
µ(m)E
x m
2+ O(y
−1/2x
1/2+ε).
Now we obtain the main result of this section.
Proposition 2. If RH is true, then for x
ε≤ y ≤ x
1/2−ε, we have (3.15) E
P(x) = X
m≤y
µ(m)E
x m
2+ O(y
−1/2x
1/2+ε).
4. Proof of the Theorem. Take y = x
4/15in Proposition 2. We only need to estimate the sum
X
m∼M
µ(m)E
x m
2for x
1/10M y. For M x
1/10, we have trivially
(4.1) X
m≤x1/10
µ(m)E
x m
2X
m≤x1/10
x
1/3m
−2/3x
11/30.
By the well-known Voronoi formula of E(t) (see [7], 13.8) we get
(4.2) x
−εX
m∼M
µ(m)E
x m
2X
m∼M
µ(m)x
1/4m
1/2X
n≤x4/15
r(n) n
3/4e
√ nx m
+ x
11/30. It now suffices to show that (1 N Y )
(4.3) S(M, N ) = X
m∼M
µ(m)x
1/4m
1/2X
n∼N
r(n) n
3/4e
√ nx m
x
11/30+ε. We consider three cases.
Case 1: M ≤ N ≤ x
4/15. Let T (M, N ) = X
m∼M
µ(m)M
1/2m
1/2X
n∼N