LXV.4 (1993)
The greatest prime factor of the integers in an interval
by
Hong-Quan Liu (Harbin)
Introduction. Let P (x) denote the greatest prime factor of Y
x≤n≤x+x1/2
n ,
where x is a large positive number. To estimate the lower bound of P (x) is one of the “greatest prime factor” problems to which the Chebyshev–
Hooley’s machinery can be applied. Chronologically Ramachandra [11], [12], Graham [4], Baker [1], and Jia [7] have contributed to the topic. In partic- ular, Baker [1] proved
P (x) > x
0.7.
Jia [7] improved the above result. But as [7] contained a substantial mistake concerning a multiple exponential sum (which was already picked out by the present author in 1987), the announced estimate P (x) > x
0.71was actually not attainable there.
We prove the following better estimate.
Theorem.
P (x) > x
0.723. We begin with
x
1/2lnx + O(x
1/2) = X
p≤P (x)
N (p) lnp, p primes , where
N (p) = X
x≤pn≤x+x1/2
1 . Thus to prove the Theorem we need to show
(1) X
:= X
p≤x0.723
N (p) lnp < (1 − ε)x
1/2lnx , ε > 0 .
For a suitable number σ, 1/2 < σ < 0.723, we write X
1
= X
p≤xσ
N (p) lnp , X
2
= X
xσ≤p≤x0.723
N (p) lnp . We need to establish an asymptotic formula for P
1
with σ as large as possible, and to estimate P
2
via the sieve.
The proof, as usual, depends mainly on treating some exponential sums, in particular we can prove an asymptotic formula for P
1
with σ = 0.6 − ε. We also make an innovation in the sieve part. To save space we do not analyze the method term by term; the interested reader can find the advantages by himself.
Notations. e(ξ) = exp(2πiξ). [ξ] is the integer part of ξ, and {ξ} = ξ − [ξ], kξk = min({ξ}, 1 − {ξ}).
X
(x1,x2,...)∈D
means summation over all lattice points (x
1, x
2, . . .) inside a domain D.
f (x, y) e ∆g(x, y) means that
f
xiyj(x, y) = g
xiyj(x, y) + O(∆g
xiyj(x, y))
whenever both sides make sense. x ∼ X means X ≤ x < 2X, x ∼ = X means C
1X ≤ x ≤ C
2X for some constants C
1and C
2. The meaning of x = O(X) or x X is as usual. ε, of course, is a sufficiently small number.
1. An asymptotic formula. In this section, we prove Proposition 1. For σ = 0.6 − ε, we have
(2) X
1
= (0.6 − ε)x
1/2lnx + O(x
1/2) . As
N (n) = x
1/2n + ψ x + x
1/2n
− ψ x n
, ψ(ξ) = 1
2 − {ξ} ,
in view of the Prime Number Theorem, to prove (2) it remains to show that (with Λ(n) being the von Mangoldt function)
X
x1/2≤n≤x0.6−ε
Λ(n)
ψ x + x
1/2n
− ψ x n
x
1/2.
We split the range x
1/2≤ n ≤ x
0.6−εinto ranges of the form v ≤ n ≤ v
0, v
0≤ 2v. Then we need to show that
(∗) X
v≤n≤v0
Λ(n)f (n) x
1/2(lnx)
−1, f (n) = ψ x + x
1/2n
− ψ x n
.
In view of [1], (∗) holds for x
1/2v x
13/22−ε, thus we only need to consider the range x
0.59≤ v ≤ x
0.6−ε. We appeal to the following decom- position.
Lemma 1.1. Let 3 ≤ u < w < z ≤ 2v, suppose that z − 1/2 is an integer , and z ≥ 4u
2, v ≥ 32z
2u, w
3≥ 64v. Set
U = max
B
∞
X
m=1
d
3(m)
X
z<n<B v≤mn≤v0
f (mn) ,
W = sup
g
∞
X
m=1
d
4(m)
X
u≤n≤w v≤mn≤v0
g(n)f (mn) ,
where the supremum is taken over all arithmetic functions g(n) such that
|g(n)| ≤ d
3(n). Then
(3) U x
1/2(lnx)
−10and W x
1/2(lnx)
−10imply the estimate
X
v≤n≤v0
Λ(n)f (n) x
1/2(lnx)
−1. P r o o f. See [5].
Choose
u = vx
−1/2+15η, w = x
2v
−3x
−10η, z = [x
1/4−10η] +
12in Lemma 1.1, where η = ε
2, x
0.59v x
0.6−ε. Then it is easy to verify that Lemma 1.1 is applicable. As in [1], by a reduction using the Fourier expansion of the function ψ(ξ), in order to show (3) it suffices to demonstrate, with H = vx
−1/2+η, the following estimates:
(4) X
0<h≤H
X
m∼M
a(m) X
n∼N
e hx
0mn
vx
−ηwhenever v ∼ = M N , N z, |a(m)| ≤ x
η, x ∼ = x
0; and
(5) X
0<h≤H
X
m∼M
a(m) X
n∼N
b(n)e hx
0mn
vx
−ηwhenever v ∼ = M N , u N w, |a(m)|, |b(n)| ≤ x
η, x ∼ = x
0.
Lemma 1.2. (5) holds for v in the (larger ) range v ≤ x
5/8−ε. P r o o f. This is Lemma 9 of [1].
Now it remains to show
Lemma 1. (4) holds.
We cite three lemmata of basic importance.
Lemma 1.3. Let I be a subinterval of [X, 2X], Q be a positive number , and z
n(X ≤ n ≤ 2X) be complex numbers. Then
X
n∈I
z
n2
≤ (1 + XQ
−1) X
|q|≤Q
(1 − |q|Q
−1) X
n,n+q∈I
z
nz
n+q.
P r o o f. This is Weyl’s inequality. Cf. [1].
Lemma 1.4. Let f (x) and g(x) be algebraic functions in the interval [a, b], and
|f
00(x)| ∼ = R
−1, |f
000(x)| (RU )
−1,
|g(x)| ≤ H , |g
0(x)| HU
1−1, U, U
1≥ 1 . Then
X
a≤n≤b
g(n)e(f (n)) = X
α≤u≤β
b
ug(n(u))
pf
00(n(u)) e(f (n(u)) − un(u) + 1/8) + O(H ln(β − α + 2) + H(b − a + R)(U
−1+ U
1−1)) + O(H min(R
1/2, max(1/hαi, 1/hβi))) ,
where [α, β] is the image of [a, b] under the mapping y = f
0(x), n(u) is determined by the equation f
0(n(u)) = u, b
u= 1/2 or 1 according as u is one of α and β or not , and hxi is defined as follows:
hxi = kxk if x is not an integer , β − α if x is an integer.
Moreover , √
f
00> 0 if f
00> 0, √
f
00= ip|f
00| if f
00< 0.
P r o o f. This is Theorem 2.2 of S. H. Min [10].
Lemma 1.5. Let f (x, y) be an algebraic function in the rectangle D
0= {(x, y) | x ∼ X, y ∼ Y }, f (x, y) e ∆Ax
αy
βhold throughout D
0, and D be a subdomain of D
0bounded by O(1) algebraic curves. Suppose that X ≥ Y , N = XY , A > 0, F = AX
αY
β, αβ(α + β − 1)(α + β − 2) 6= 0, 0 < ∆ < ε
0, where ε
0is a small number depending at most on α and β. Then
X
(x,y)∈D
e(f (x, y))
ε,α,β(
6√
F
2N
3+ N
5/6+
10√
∆
4Y
4F
2N
5+
8√
F
−1X
−1N
8+ N F
−1/4+
4√
∆X
−1N
4+ N Y
−1/2)(N F )
η.
P r o o f. This is Lemma 1 of [9].
We also need the following auxiliary lemma.
Lemma 1.6. Let M ≤ N < N
1≤ M
1, and a
n(M ≤ n ≤ M
1) be complex numbers. Then
X
N <n≤N1
a
n≤
∞
R
−∞
K(θ)
X
M <m≤M1
a
me(θm) dθ , with K(θ) = min(M
1− M + 1, (π|θ|)
−1, (πθ)
−2), and
∞
R
−∞
K(θ) dθ ≤ 3 ln(2 + M
1− M ) . P r o o f. This is Lemma 2.2 of [2].
P r o o f o f L e m m a 1. For notational simplicity, we let x
0= x in (4).
For 1 ≤ h ≤ H, let
S(h) = X
m∼M
a(m) X
n∼N
e hx mn
. By Lemma 1.4, after a partial summation, we obtain (6) x
−ηS(h) vN
2hx
1/2X
m∼M
X
u∈I(m)
e(F (u, m)) + v
3hx
1/2+ v
2hx +M , where I(m) is a subinterval of [C
1U, C
2U ], U = hxv
−1N
−1, F (u, m) = C
3(hxum
−1)
1/2(for i ≥ 1, C
idenotes a constant). By Lemma 1.6 we get
x
−ηX
m∼M
X
u∈I(m)
e(F (u, m))
X
m∼M
X
u∼=U
e(G(u, m))
= S
0(h) , say , where G(u, m) = F (u, m) + θu, θ is independent of u and m, and 0 ≤ θ < 1.
Let
Q = min(
8√
hxM
5N
−5, hxM
−1N
−2)x
−η. If Q ≤ 100, then we get, trivially,
vN
2hx
1/2S
0(h) (hxM N
−1Q
−1)
1/2(M N
1/2+
16p
(hx)
7M
3N
−3)x
η(vx
−1/8+ (v
5x)
1/8)x
10ηx
1/2−10η.
If Q > 100, by Cauchy’s inequality, Lemma 1.3, and partial summation, we get, with some Q
1, 1 ≤ Q
1≤ Q, the inequality
(7) |S
0(h)|
2(M U )
2Q
−1+M
3/2U Q
−1X
(u,q)∈D
X
m∼M
m
−1/2e(f (u, q, m))
,
where D = {(u, q) | q ∼ Q
1, u ∼ = U, u + q ∼ = U } and f (u, q, m) =
C
3(hxm
−1)
1/2(u
1/2− (u + q)
1/2) − qθ. We apply Lemma 1.4 to transform
the sum over m to a sum over w, where w ∼ = W = N Q
1M
−1. Then we change the order of summation, and estimate the sum over w trivially, to get, with some w, the estimate
(8) M
3/2U Q
−1X
(u,q)∈D
X
m∼M
m
−1/2e(f (u, q, m))
hxN
−3/2Q
−1/2X
(u,q)∈D1
e(g(u, q))
+ (hx)
2Q
−1/2N
−9/2+ (hx)
2Q
−1N
−5+ (hx)
2M
−1N
−4, where
D
1= D ∩ {(u, q) | 1 ≤ |C
3(2w)
−1(hx)
1/2M
−3/2(u
1/2− (u + q)
1/2)| ≤ √
34} , g(u, q) = C
4(hxw(u
1/2− (u + q)
1/2)
2)
1/3− qθ .
It is easy to verify that
g(u, q) e ∆C
5(hxw)
1/3u
−1/3q
2/3, ∆ = Q
1U
−1.
Choosing X = U , Y = Q
1, F ∼ = N Q
1, ∆ = Q
1U
−1in Lemma 1.5, we find that
(9) hxN
−3/2Q
−1/2X
(u,q)∈D1
e(g(u, q))
( p
6(hx)
9Q
2M
−3N
−13+ p
6(hx)
11Q
2M
−5N
−19+
10p
(hx)
11Q
10M
−1N
−15+ p
8(hx)
15Q
3M
−7N
−27+ p
4(hx)
8QM
−4N
−15+ p
4(hx)
6Q
3M
−2N
−10+ p
(hx)
4M
−2N
−7)x
2η. From (6)–(9), we deduce that
vN
2hx
1/2S
0(h) (
12p
(hx)
3M
3N
5Q
2+
12p
(hx)
5M N
−1Q
2(10)
+
20p
(hx)M
9N
15Q
10+
16p
(hx)
7M N
−3Q
3+ p
8(hx)
4N
−3Q + p
8(hx)
2v
2Q
3+ p
4(hx)
2N
−1+ p
4(hx)
2M
2N
−3Q
−1+ p
hxM N
−1Q
−1)x
10η. As we have already seen,
p hxM N
−1Q
−1x
1/2−20η,
thus
p
4(hx)
2M
2N
−3Q
−1p
4(hx)M N
−2x
8√
v
4x
3N
−6(11)
(v
8x
3)
1/16x
10ηx
1/2−20η. From (9)–(11), we get
vN
2hx
1/2S
0(h) (
48p
(hx)
13M
17N
15+
48p
(hx)
21M
9N
−9+
80p
(hx)
9M
61N
35+
128p
(hx)
59M
23N
−39+
64p
(hx)
33M
5N
−29+
64p
(hx)
19M
31N + p
4(hx)
2N
−1)x
10η+ x
1/2−10η(
96√
v
60x
13N
−4+
96√
v
60x
21N
−36+
160√
v
140x
9N
−52+
256√
v
164x
59N
−124+
128√
v
76x
33N
−68+
128√
v
100x
19N
−60+
4√
v
2xN
−1)x
20η+ x
1/2−10η(
96√
v
60x
12+
160√
v
140x
−4+
128√
v
82x
14+
64√ v
38x
8+
64√
v
50x
2+
16√
v
8x
3)x
30η+ x
1/2−10ηx
1/2−10η. From (6), (7) and the above estimate, we get
S(h) v
3hx
1/2+ v
2hx + vx
−1/4x
20η+ x
1/2−8η, hence
X
1≤h≤H
|S(h)| (v
2x
−3/4)x
30η+ vx
−ηvx
−η. The proof of Lemma 1 is finished.
R e m a r k. The proof of Lemma 6 of [7] is false. On p. 191 we find the equality E
1= E
2, where
E
1=
Q
X
q=1
X
(h,m)∈Pq
X
(h0,m0)∈Pq
ε
hε
h0X
n∼N
e x n
h m − h
0m
0,
E
2= X
h∼J
X
h0∼J
ε
hε
h0X
0≤k≤M hh0(J Q)−1
X
m∼M,m0∼M m0h−mh0=k
X
n∼N
e
kx mm
0n
,
and
P
q=
(h, m) | h ∼ J, m ∼ M, M (q − 1)
J Q ≤ m
h ≤ M q J Q
. But we observe that
0 ≤ k ≤ M hh
0(J Q)
−1and m
0h − mh
0= k
cannot imply that there is a q, 1 ≤ q ≤ Q, such that both (h, m) ∈ P
qand (h
0, m
0) ∈ P
q; this means that there is no one-to-one correspondence between the summation variables of E
1and those of E
2, thus in general E
16= E
2.
2. Multiple exponential sums. Here we give several results on expo- nential sums, thus preparing an application of the sieve.
Lemma 2. For v ∼ = M N , H = vx
−1/2+η, x
0.6vx
εx
2/3,
|a(n)|, |b(m)| ≤ x
η, and vx
−1/2+30ηN (vx
1/2)
1/7x
−30η, we have X
1≤h≤H
X
n∼N
X
m∼M
a(n)b(m)e hx mn
vx
−η. To prove Lemma 2, we need two lemmata.
Lemma 2.1. Let H
1≥ H
10≥ 1, H
2≥ H
20≥ 1, n
1and n
2be positive integers with (n
1, n
2) = 1. Then
ω(n
1, n
2; r) := X
∗h1n1−h2n2=r
1 =
1
R
0
ω(n b
1, n
2; θ)e(θr) dθ and
1
R
0
| ω(n b
1, n
2; θ)| dθ (1 + H
1H
2n
−11n
−12)
1/2(ln(2H
1H
2))
2, where ∗ means the condition H
10≤ h
1≤ H
1, H
20≤ h
2≤ H
2.
P r o o f. This is Lemma 8 of [3].
Lemma 2.2. Suppose α = a/q + θ/q
2, (a, q) = 1, q ≥ 1, |θ| ≤ 1. Then for any β, U > 0, P ≥ 1, we have
P
X
x=1
min
U, 1
kαx + βk
≤ 6 P q + 1
(U + q lnq) . P r o o f. This is Lemma 6 of Chapter 5 of [8].
P r o o f o f L e m m a 2. We assume that x is irrational. Pick an integer j such that M ∼ M
1= 2
j. We denote the triple exponential sum of Lemma 2 by S(M, N ). By Cauchy’s inequality, we have (m ' M
1means M
1≤ m
≤ 4M
1)
x
−2η|S(M, N )|
2M
3/2X
m'M1
m
−1/2H
X
h=1
X
n∼N
a(n)e hx mn
2
= M
3/2X
h,h1,n,n1
a(n
1)a(n) X
m'M1
m
−1/2e h
1n
1− h n
x m
.
The terms with h
1n = hn
1contribute at most O(M
2HN x
η). We classify the remaining terms according to the value of (n, n
1). After a familiar argument, we get
(2.1) x
−3η|S(M, N )|
2M
3/2X
d∼D
X
n,n1∼N1
(n,n1)=1
X
r∼R
ω(n, n
1; r) X
m'M1
e
−rx dmnn
1m
−1/2+ M
2HN ,
for some D, 1 ≤ D ≤ N , N
1= N/d, and some R, 1 ≤ R ≤ HN/D. By Lemma 1.4, the innermost sum is equal to
(2.2) X
α≤u≤β
u
−1/2e
C
6urx dnn
1 1/2C
7+ O
M
−1/2min M
3N
2RDx
1/2, max
1 kαk , 1
kβk
+ O M
3/2N
2RDx x
η+ O(M
−1/2x
η) , where
α = rx
(4M
1)
2dnn
1, β = rx
M
12dnn
1. We consider the sum
S
∗(D, R) = M X
d∼D
X
n,n1∼N1
(n,n1)=1
X
r∼R
ω(n, n
1; r) min M
3N
2RDx
1/2, 1 kβk
;
a similar sum with β being replaced by α can be treated analogously.
We shall prove the following estimate for S
∗(D, R):
(2.3) S
∗(D, R) v
2x
−10ηif N ≤ x
1/4−20η. If D ≥ (M N )
2x
−1−η, then we trivially get
S
∗(D, R) M M
3N
2RDx
1/2D X
|nh1−n1h|≤2R h,h1≤H,n,n1≤2N/D
1
(M N )
5/2H
3/2x
−1/2D
−1x
ηx
1/2(M N )
1/2H
3/2x
2ηv
2x
−1/4+20η. If D ≤ (M N )
2x
−1−η, then we see that
β r = 1
q + θ
q
2, with q = [dnn
1M
12x
−1] ≥ 1 and |θ| ≤ 1 ,
hence by Lemmata 2.1 and 2.2, we have the estimate S
∗(D, R) M D N
2D
21 + DH N
d,n,n
max
1X
r∼R
min M
3N
2RDx
1/2, 1 kβk
M N
2RDx
M
2N
2+ 1 N
2M
3RDx
1/2+ N
2M
2Dx
x
η(vN x
1/4+ v
2x
−1/2N
2+ v
5/2N
1/2x
−1/2+ v
3N x
−1)x
10ηv
2x
−10ηprovided only N ≤ x
1/4−20η. Thus anyhow (2.3) holds. By inserting (2.2) in (2.1), and taking into account (2.3), we get, after changing the order of summation, the following estimate (with U = RDxv
−2):
x
−3η|S(M, N )|
2N M
5/2(RDx)
1/2X
d∼D
X
n,n1∼N1
X
u∼=U
X
r∈I
ω(n, n
1; r)e
C
6urx dnn
1 1/2+ v
2x
−10η,
where I is some subinterval of [R, 2R], which may depend on the variables outside the absolute value symbol. By Lemma 1.6 we get
(2.4) x
−4η|S(M, N )|
2N M
5/2(RDx)
1/2X
d∼D
X
n,n1∼N1
X
u∼=U
X
r∼=R
ω(n, n
1; r)e
C
6urx dnn
1 1/2+ θr
+ v
2x
−10η,
with some θ, which is independent of the variables r, u, n, n
1and d. Now we have arrived at (6.1) of [3], p. 325. The argument in what follows is exactly the same as in [3], and we get (cf. p. 329 of [3]), by the assumption of Lemma 2,
x
−10η|S(M, N )|
4v
2Hx v
1/2H
2N
4+ Hx v
H
3/2N
7/2+ Hx v
1/2N
3H
2M
1/2+ Hx
v
H
4/3M
1/3N
8/3+ (v
2x
−10η)
2v
4x
−20η.
Lemma 2 is proven.
Lemma 3. For v ∼ = M N , H = vx
−1/2+η, x
0.6vx
εx
3/4,
|a(n)|, |b(m)| ≤ x
η, (k, λ) an exponent pair , and
vx
−1/2+10η≤ N ≤ min((v
1−λ+kx
−k+λ/2−1/4)
1/(1−k+λ), x
1/4)x
−20η, we have
X
1≤h≤H
X
m∼M
X
n∼N
a(n)b(m)e hx mn
vx
−η. P r o o f. Note that (2.4) holds provided that
(2.5) v ≤ x
3/4−ε, N ≤ x
1/4−20η. In view of Lemma 2.1, (2.4) and (2.5), we get (2.6) x
−4η|S(M, N )|
2N
3M
3/2(RD)
1/2X
r∼R
e(F (r, d, n, n
1, u))
+ v
2x
−10η, where F (r, d, n, n
1, u) = C
6(urx/(dnn
1))
1/2+ θr, for certain d, n, n
1, u and θ with |θ| ≤ 1. It is easy to see that F
r0(r, d, n, n
1, u) ∼ = xD/(vN ) 1, thus
(2.7) X
r∼R
e(F (r, d, n, n
1, u)) R
λDx vN
k. Lemma 3 follows from (2.6), (2.7) and the fact that R HN .
The last result of the section is
Lemma 4. For v ∼ = M N , H = vx
−1/2+η, x
0.6vx
εx
3/4, |a(n)| ≤ x
ηand N ≤ x
3/8−ε, we have
X
1≤h≤H
X
m∼M
X
n∼N
a(n)e hx mn
vx
−η. To prove Lemma 4 we need again two lemmata.
Lemma 2.3. Let α, β, γ be real constants such that (α − 1)βγ 6= 0. Let M, R, S, x ≥ 1 and let φ
mand ψ
rsbe complex numbers with modulus not exceeding 1. Then
X
m∼M
X
r∼R
X
s∼S
φ
mψ
rse xm
αr
βs
γM
αR
βS
γ(x
1/4M
1/2(RS)
3/4+ M
7/10RS + M (RS)
3/4+ x
−1/4M
11/10RS)(ln(10M RS))
5.
P r o o f. This is Theorem 3 of [3].
Lemma 2.4. Let X ≥ 100, Y ≥ 100, A > 0, f (x, y) = Ax
αy
β, F =
AX
αY
β, α and β being rational numbers (not positive integers). Suppose
F
−2X
4≤ Y
3N
−η, and for a
xand b
ybeing complex numbers with modulus not exceeding 1, define
S(X, Y ) := X
(x,y)∈D
a
xb
ye(f (x, y)) ,
where D is some region contained in the rectangle {(x, y) | x ∼ X, y ∼ Y } such that for a fixed e x, x ∼ X, the intersection D ∩ {( e x, y) | y ∼ Y } has at e most O(1) segments. Then, for W = X
5+ Y
5,
S(X, Y ) (
40√
F
8X
24Y
27W +
40√
F
−4X
28Y
39W +
4√
F
−1X
3Y
5(2.8)
+
20√
F X
13Y
19+
4√
F X
2Y
3+
20√
F
3X
14Y
7+
20√
F
−3X
16Y
13+
10√
F
−3X
6Y
13)(F XY )
η=: E . P r o o f. Put δ = η
2. By Lemma 1.6, we have (with N = F XY )
N
−δ|S(X, Y )| X
x∼X
X
y∼Y
C(y)e(f (x, y)) ,
where C(y) = b
ye(θy) with some real number θ (which is independent of x and y). We choose
Q = (F
−1X
2Y )
2/5≤ Y N
−δ(by assumption).
If Q ≤ N
η, then we trivially get
|S(X, Y )| N
η/2XY Q
−1/2N
η/2(
5√
F X
3Y
4) E . Assume that Q ≥ N
η. By Cauchy’s inequality and Lemma 1.3, we get (2.9) N
−3δ|S(X, Y )|
2(XY )
2Q + XY
Q
X
(y,q)∈D1
C(y)C(y + q) X
x∼X
e(Ax
αt(y, q)) , where D
1= {(y, q) | y, y + q ∼ Y, q ∼ Q
1} for some Q
1, 1 ≤ Q
1≤ Q, and t(y, q) = (y + q)
β− y
β. Applying Lemma 1.4 to the innermost sum, we get (2.10) X
x∼X
e(Ax
αt)
= X
u∈I
C
7|(At)
γu
−1/2−γ|e(C
8(At)
2γu
1−2γ)
+ O
min X
2Y Q
1F
1/2, 1
kg
1(y, q)k + 1 kg
2(y, q)k
+ XY
Q
1F + lnN + R(y, q)
,
where I = (C
9AX
α−1|t|, C
10AX
α−1|t|), γ = 1/(2(1 − α)), g
1= αAX
α−1t,
g
2= αA(2X)
α−1t, and R(y, q) = 0 or O((X
2Y Q
−11F
−1)
1/2) according as
none of the end points of I is an integer or otherwise. For each fixed q, q ∼ Q
1, we consider the sum
Φ = X
y∼Y
min((X
2Y Q
−11F
−1)
1/2, 1/kg(y)k) ,
where g(y) is either g
1(y, q) or g
2(y, q). As both g(y) and g
0(y) are mono- tonic, and g
0(y) ∼ = F Q
1X
−1Y
−2, we can classify y according to the integer part of g(y), and it is easy to get
Φ (1 + F Q
1X
−1Y
−1)((X
2Y Q
−11F
−1)
1/2+ F
−1Q
−11XY
2lnN ) , which contributes to the RHS of (2.9) at most
10√
F
−3X
16Y
13+
5√
F
−3X
6Y
13+
10√
F
3X
14Y
7+
√ XY
2(2.11)
E
2N
−2η,
and similarly for the contribution of R(y, q). From (2.9)–(2.11), after chang- ing the order of summation, and estimating the sum over u trivially, we get, with some u, |u| ∼ = F Q
1X
−1Y
−1, the estimate
(2.12) N
−4δ|S(X, Y )|
2XY Q
1 + F Q XY
A
γXY F Q
1 γ+1/2|S
1| + N
−ηE
2, S
1= S
1(u) = X
(y,q)∈D2
C(y)C(y + q)t
γe(C
8(At)
2γu
1−2γ) , D
2= D
1∩ {(y, q) | (C
10AX
α−1)|t| ≥ u ≥ (C
9AX
α−1)|t|} . If Q
1≤ N
η, then we trivially have
XY Q
1 + F Q XY
A
γXY F Q
1 γ+1/2|S
1|
XY Q
1 + F Q XY
XY
3/2F
−1/2N
1/2 10√
F
−1X
12Y
21+
√
F X
2Y
3√
F
−1X
3Y
5+
10√
F X
13Y
19+
√
F X
2Y
3N
−ηE
2. Assume that Q
1≥ N
η. By Lemma 1.6 we get
(2.13) N
−δ|S
1| X
y∼Y
X
q∼Q1
C(y + q)e(ϕq)t
γe(C
8(At)
2γu
1−2γ)
,
with some real number ϕ, independent of y and q. Assume that t(y, q) > 0
(otherwise we consider −t). Applying Cauchy’s inequality and Lemma 1.3
to (2.13), we get, with Q
2= Q
1/21N
−δ, G = (Q
1Y
β−1)
γ, (2.14) N
−3δ|S
1|
2(Y Q
1G)
2Q
2+ Y Q
1Q
2X
1≤|q1|≤Q2
|S
2(q
1)| ,
S
2(q
1) = X
(y,q)∈D3
C(y + q)C(y + q + q
1)t
1(y, q, q
1)e(t
2(y, q, q
1)) , D
3= {(y, q) | y ∼ Y, q, q + q
1∼ Q
1} ,
t
1(y, q, q
1) = (t(y, q)t(y, q + q
1))
γ, t
2(y, q, q
1) = C
8u A
u
2γ(t(y, q + q
1)
2γ− t(y, q)
2γ) . By writing y + q = z we get
S
2(q
1) = X
(z,q)∈D4
C(z)C(z + q
1)T
1(z, q, q
1)e(T
2(z, q, q
1)) , D
4= {(z, q) | q, q + q
1∼ Q
1, z − q ∼ Y } ,
T
1(z, q, q
1) = (t(z − q, q)t(z − q, q + q
1))
γ, T
2(z, q, q
1) = C
8u A
u
2γ(t(z − q, q + q
1)
2γ− t(z − q, q)
2γ) . Again by Lemma 1.6, we get, with I
1= [0.5Y, 2.5Y ],
N
−δ|S
2(q
1)| X
z∈I1
X
q∼Q1
T
1(z, q, q
1)e(ξq)e(T
2(z, q, q
1)) ,
with some real number ξ, independent of z and q. Applying Cauchy’s in- equality and Weyl’s lemma, we obtain, with Q
3= Q
1N
−δ,
(2.15) N
−3δ|S
2(q
1)|
2(Y Q
1G
2)
2Q
−13+ Y Q
1Q
−13X
1≤|q2|≤Q3
X
q∼Q1
|S
3(q, q
1, q
2)| ,
S
3(q, q
1, q
2) = X
z∈I1
T
3(z, q, q
1, q
2)e(T
4(z, q, q
1, q
2)) , T
3(z, q, q
1, q
2) = T
1(z, q, q
1)T
1(z, q + q
2, q
1) , T
4(z, q, q
1, q
2) = T
2(z, q + q
2, q
1) − T
2(z, q, q
1) . Let
U (z) = T
3(z, q, q
1, q
2) , V (z) = T
4(z, q, q
1, q
2) . It is an easy exercise to verify that, for z ∈ I
1,
U (z) ∼ = G
4, U
0(z) ∼ = G
4Y
−1, V (z) ∼ = |F q
1q
2|Y
−1Q
−11, V
0(z) ∼ = F q
1q
2Y Q
1Y .
As V (z) has nice properties with respect to the variable z, by partial sum- mation and the exponent pair (1/2, 1/2), we obtain
(2.16) S
3(q, q
1, q
2) G
4((|F q
1q
2|Y
−2Q
−11)
1/2Y
1/2+ Y
2Q
1(|F q
1q
2|)
−1) . In view of (2.14) and (2.15), the first term in (2.16) contributes to (2.12) at most
N
η/2XY Q
1 + F Q XY
A
γXY F Q
1 1/2+γG
16q
Y
10Q
171F
21 + F Q XY
16
p
F
−6X
32Y
34Q
−7N
η/2(
20√
F
8X
24Y
27W +
20√
F
−4X
28Y
39W )N
η/2E
2N
−η;
and the second term in (2.16) contributes to (2.12) at most
N
η/2XY Q
1 + F Q XY
A
γXY F Q
1 1/2+γG(Q
71Y
10F
−2)
1/8p
8Q
−5X
16Y
22F
−61 + F Q XY
N
η/2( √
F
−1X
3Y
5+
10√
F X
13Y
19)N
η/2E
2N
−η.
Finally, the term (Y Q
1G
2)
2Q
−13together with the term (Y Q
1G)
2Q
−12con- tributes to (2.12) at most
N
η/2XY Q
1 + F Q XY
A
γXY F Q
1 1/2+γY GQ
3/41X
2Y
5/2F
−1/2Q
−3/41 + F Q XY
N
η/2E
2N
−η. The estimate (2.8) follows from the above observations.
P r o o f o f L e m m a 4. By Lemma 1.4, we obtain (2.17) X
m∼M
e
− hx mn
= X
u∈I
C
11(nu
−3h
−1)
1/4e
C
12hux n
1/2+ O M
3N hx
1/2+ O M
2N hx
+ O(lnx) , where I = (α, β), α = hx/(4M
2n), β = hx/(M
2n). We divide the sum in question into subsums of the form
(2.18) X
h∼H1
X
n∼N
a(n) X
m∼M
e
− hx mn
= S(H
1) , say ,
where H
1≤ H. By substituting (2.17) in (2.18), and exchanging the order of summation, we get, with an application of Lemma 1.6,
S(H
1) M
3N H
1x
1/2X
w∼W
X
n∼N
b(n)e(θn)e
C
12wx n
1/2x
η+ vx
−10η, where W = H
12xv
−1M
−1and θ is some number (independent of the vari- ables).
If H
18≤ x
η(N x
−5v
8), then by Lemma 2.3 with (r, s, m) = (1, w, n), we get the estimate
x
−3η|S(H
1)|
4q
H
15x
2N v
−1+
10q
H
115x
5N
7v
−5+
4q
H
14xN
3+
20q
H
125x
5N
22v
−5v(
80√
N
71x
−35+
8√
N
7x
−3+
160√
N
201x
−85+
32√
N
13x
−9) + vx
−εvx
−ε. If H
18> x
η(N x
−5v
8), then we can apply Lemma 2.4, with (x, y) = (n, w), to get
x
−3η|S(H
1)| M
3N H
1x
1/2(
40q
H
162x
35v
−62N
56+
40q
H
172x
40v
−72N
56+
40q
H
174x
35v
−74N
72+
40q
H
184x
40v
−84N
72+
4q
H
19x
4v
−9N
8+
20q
H
139x
20v
−39N
32+
4q
H
17x
4v
−7N
5+
20q
H
117x
10v
−17N
21+
20q
H
123x
10v
−23N
29+
10q
H
123x
10v
−23N
19)
40q
H
142x
15v
−2N
16+
40q
H
152x
20v
−12N
16+
40q
H
154x
15v
−14N
32+
40q
H
164x
20v
−24N
32+
4q
H
17x
2v
−3N
4+
20q
H
129x
10v
−9N
12+
4q
H
15x
2v
−1N +
20q
H
17v
13N +
20q
H
113v
7N
9+
10q
H
118x
5v
−8N
9vx
−10η.
The proof of Lemma 4 is finished.
3. Sieve methods
3.1. Outline of setting. Let v be a number such that
x
0.6−εv x
0.723.
The sequence A is defined as
A = {n | v ≤ n ≤ ev and there exists an m with x ≤ mn ≤ x + x
1/2} , where e = 2.71828 . . . is the base of the natural logarithms. As usual, S(A, z) denotes the number of elements in A having no prime factors less than z.
For r a positive integer, let
|A
r| = X
n∈A,n≡0 (mod r)
1 . It is easy to see that
|A
r| = x
1/2/r + R(A, r) , R(A, r) = X
v≤rs≤ev
ψ x + x
1/2rs
− ψ x rs
+ O x
1/2v
. Let V (z) = Q
p<z
(1 − 1/p). With the above property of A, we have Lemma 3.1.1. We have
(3.1.1) S(A, z) ≤ x
1/2V (z)
F lnD lnz
+ O(ε)
+ R
+if 2 ≤ z ≤ D , where
R
+= X
(D)
X
r<Dη r|P (Dη)
C
(D)(r, η) X
Di≤pi≤min(z,Di1+η) 1≤i≤t
R(A, rp
1. . . p
t) ,
t 1, P
(D)
is summation over all sequences {D
i}
ti=1with each D
iof the form
D
η2(1+η2)n, n = 0, 1, 2, . . . , such that
D
1≥ . . . ≥ D
t≥ D
η2, D
1. . . D
2sD
2s+13≤ D for all 0 ≤ s ≤ (t − 1)/2 , and
P (D
η) = Y
p<Dη
p , |C
(D)(r, η)| ≤ 1 . Also, for p ∼
12M , 2 ≤ M ≤ D
1/2, we have
(3.1.2) S(A
p, M ) ≥ x
1/2p V (M )
f lnD lnM
+ O(ε)
− X
d<D d|P (M )
λ
dR(A, pd) ,
λ
dbeing some numbers with |λ
d| ≤ D
η. The functions F and f are the well-known functions in the linear sieve.
P r o o f. Both (3.1.1) and (3.1.2) come from [6].
We choose P = vx
−1/2+50η, and x
10εD =
(x
10v
−15)
1/2for x
0.6vx
ε≤ x
11/18, x
3/8for x
11/18< vx
ε≤ x
0.723+ε,
x
50ηQ =
v
−3x
2for x
0.6vx
ε≤ x
27/44, (vx
1/2)
1/7for x
27/44< vx
ε≤ x
67/104, (v
20x
−7)
1/36for x
67/104< vx
ε≤ x
0.665, (v
50x
−21)
1/70for x
0.665< vx
ε≤ x
0.7. Lemma 5. We have
S(A, P ) ≤ x
1/2V (P )
F lnD lnP
+ O(ε)
. P r o o f. By (3.1.1) it suffices to show that
X
p1∼A1
. . . X
pt∼At
X
v0≤p1...ptn≤ev0
ψ
x + x
1/2rp
1. . . p
tn
− ψ
x
rp
1. . . p
tn
x
1/2−0.75η, where
A
1A
2. . . A
t,
A
1. . . A
2sA
32s+1≤ D
1+ηfor 0 ≤ s ≤ (t − 1)/2 , v
0= vr
−1, r < D
η.
Then, by a standard reduction, it is enough to establish
(3.1.3) X
1≤h≤H
X
p1∼A1
. . . X
pt∼At
X
v≤rp1...ptn≤ev
e
hx
0p
1. . . p
trn
vx
−η, where H = vx
−1/2+η, x ∼ = x
0. Our aim is to arrange {r, p
1, . . . , p
t, n} into two subsets, so that we can produce from (3.1.3) an exponential sum of the type of Section 2, and then (3.1.3) will follow from the estimate given there.
We claim that either
(3.1.4) there exists a subset S of {1, 2, . . . , t} with P ≤ Y
i∈S
A
i≤ Qx
−2η:= Q
1, or
(3.1.5) A
1. . . A
t≤ x
3/8−22η:= A
0.
We assume the contrary, that is, neither (3.1.4) nor (3.1.5) is true, and deduce a contradiction. For v ≥ x
11/18−ε, (3.1.5) is obvious. For v <
x
11/18−εwe reason as follows. From A
31≤ D
1+η< Q
31, we have A
1< P . As
A
0< A
1. . . A
tP
2A
3. . . A
t, we have A
3. . . A
t> Q
1. If A
3A
4< P , there
must be a least j such that A
3A
4. . . A
j> P , hence A
3A
4. . . A
j> Q
1; but then
Q
1< A
3A
4. . . A
j(A
3. . . A
j−1)(A
3A
4)
1/2< P
3/2,
a contradiction. If A
3A
4≥ P , then A
3A
4> Q
1, thus A
3> Q
1/21. But now Q
5/21A
1A
2A
33≤ D
1+η,
also a contradiction. The proof of Lemma 5 is finished.
Lemma 6. We have X
P ≤p<Q0
S(A
p, p) ≥ X
P ≤p<Q00
x
1/2V (p) p
f ln(v/p) lnp
+ O(ε)
, where
2Q
00= Q
0=
v
10/37for x
0.6vx
ε≤ x
11/18, Q for x
11/18< vx
ε≤ x
0.7.
P r o o f. Let J be the integer such that 1 ≤ Q
0/(2
JP ) < 2. Then
(3.1.6) X
P ≤p<Q0
S(A
p, p) ≥ X
1≤j≤J
X
p∼2j−1P
S(A
p, 2
jP ) .
By (3.1.2), we have, with D
j= (v/(2
jP ))x
−ε, P
j= 2
jP , (3.1.7) S(A
p, 2
jP ) ≥ x
1/2p V (2
jP )
f lnD
jlnP
j+ O(ε)
− X
d<Dj
λ
dR(A, pd) . We now show that for each fixed j, 1 ≤ j ≤ J , we have
(3.1.8) X
p∼Pj/2,d<Dj
λ
dR(A, pd) x
1/2−η/4.
Again, by the standard argument using the Fourier expansion of ψ(ξ), to prove (3.1.8) it suffices to prove
X
1≤h≤H
X
p∼Pj/2
X
d≤Dj
X
v≤pds≤ev
e hx
0pds
vx
−η,
where H = vx
−1/2+η, x ∼ = x
0. At this stage the condition v ≤ pds ≤ ev can be removed by a familiar lemma (cf. [1]). Thus it is enough to get
(3.1.9) X
1≤h≤H
X
n∼N
X
m∼M
a(n)b(m)e hξ mn
vx
−η,
where N = P
j/2, M N ∼ = v, ξ ∼ = x, |a(n)| ≤ x
η, |b(m)| ≤ x
η. The above
exponential sum is just of the form we considered in Section 2. As P
N = P
j/2 Q for x
0.6vx
ε≤ x
0.7, we find that
(3.1.9) follows from
Lemma 1.2 when x
0.6vx
ε≤ x
27/44; Lemma 2 when x
27/44< vx
ε≤ x
67/104; Lemma 3 with (k, λ) = (2/7, 4/7) when
x
67/104< vx
ε≤ x
0.665;
Lemma 3 with (k, λ) = (11/30, 16/30) when x
0.665< vx
ε≤ x
0.7.
Hence (3.1.8) is true. Using the asymptotic formula for V (·) and the well-known property of f (·), we have, for p ∼ P
j/2 = 2
j−1P ,
(3.1.10)
V (2
jP ) = V (p)
1 + O
1 lnx
, f lnD
jlnP
j= f ln(v/p) lnp
+ O(ε) . Lemma 6 follows from (3.1.6)–(3.1.8) and (3.1.10).
3.2. The contribution of the range (0.6 − ε, 11/18 − ε). We prove Proposition 2. We have
X
21
:= X
x0.6−ε≤p≤x11/18−ε
N (p) lnp < 0.02278x
1/2lnx .
Unless otherwise specified, all symbols have the same meaning as in Subsection 3.1. Now, v satisfies
e
−1x
0.6−ε≤ v ≤ x
11/18−ε. Lemma 3.2.1.
X
p1p2∈A Q0≤p1≤min(p2,Q)
1 = x
1/2lnv
ln 5.4 − 8.1s 4s − 2
+ O(ε)
,
where s = (lnv)/(lnx).
P r o o f. It is clear that X
p1p2∈A Q0≤p1≤min(p2,Q)
1 = X
Q0≤p1≤Q
X
v≤p1p2≤ev
X
x≤p1p2m≤x+x1/2
1 = U + V ,
where
U = X
Q0≤p1≤Q
X
v≤p1p2≤ev
x
1/2p
1p
2,
V = X
Q0≤p1≤Q
X
v≤p1p2≤ev
ψ x + x
1/2p
1p
2− ψ
x p
1p
2.
By the Prime Number Theorem, we easily deduce that U = x
1/2X
Q0≤p1≤Q
1
p
1ln(v/p
1) + O((lnx)
−2)
= x
1/2lnv
ln 5.4 − 8.1s 4s − 2
+ O(ε)
; and the argument in the proof of Lemma 6 gives
V = O(x
1/2(lnx)
−2) . This proves Lemma 3.2.1.
P r o o f o f P r o p o s i t i o n 2. Let A = 0.6 − ε, B = 11/18 − ε, and L = lnx. Then
X
21
≤ X
AL−1≤k≤BL
(k + 1) X
ek≤p≤ek+1
N (p) . For v = e
k, by Buchstab’s identity it is easy to verify that
X
v≤p≤ev
N (p) ≤ S(A, P ) − X
P ≤p<Q0
S(A
p, p) − X
p1p2∈A Q0≤p1≤min(p2,Q)
1 ,
which, in conjunction with Lemmata 5, 6, and 3.2.1, leads to the estimate X
21
≤ x
1/2L(I
1− I
2− I
3+ O(ε)) , where (γ being the Euler constant)
e
γI
1=
B
R
A
s
s − 1/2 F 10 − 15s 2s − 1
ds ,
e
γI
2=
B
R
A g(s)
R
2.7
f (t) dt ds , g(s) = 1/(2s − 1) ,
I
3=
B
R
A
ln 5.4 − 8.1s 4s − 2
ds . The following formulae are well known:
(3.2.1) F (u) = 2e
γu (1 + B(u)) , B(u) =
0 for 0 < u ≤ 3,
u−1
R
2