LXXVII.3 (1996)
On the exceptional set of Goldbach numbers in a short interval
by
Chao Hua Jia (Beijing)
1. Introduction. An even number which can be written as a sum of two primes is called a Goldbach number .
In 1937, I. M. Vinogradov proved the famous three primes theorem. Soon after that, employing Vinogradov’s idea, Hua Loo Keng [3] proved that if B is a sufficiently large positive constant and N is sufficiently large, then the even numbers in (2, N ), except for O(N log
−BN ) values, are all Goldbach numbers.
In 1973, Ramachandra [18] obtained the result in a short interval. He showed that if A = N
ϕ+εwith ϕ =
35, then the even numbers in (N, N + A), except for O(A log
−BN ) values, are all Goldbach numbers. If the estimation for the zero density of Huxley [4] is applied, ϕ =
127can be arrived at. In 1991, Jia Chao Hua [8] employed the sieve method combined with the circle method to get ϕ =
2342.
In 1993, Perelli and Pintz [17] developed a new technique in the circle method to make great progress in solving this problem. They showed ϕ =
367. Mikawa [13] proved ϕ =
487by the sieve method.
In 1994, Jia Chao Hua [10] proved ϕ =
787. Li Hongze [12] improved it to ϕ =
817. Jia Chao Hua [11] got further ϕ =
121. In [10], the new tech- nique of [17], the sieve method and the mean value estimation for Dirichlet polynomials were applied.
In this paper, we shall develop further the technique of [10] to prove the following:
Theorem. Suppose that B is a sufficiently large positive constant, ε is a sufficiently small positive constant, N is sufficiently large and A = N
1087 +12ε. Then the even numbers in (N, N + A), except for O(A log
−BN ) values, are all Goldbach numbers.
Project supported by the Tian Yuan Item in the National Natural Science Foundation of China.
[207]
For A = N
1087 +12ε, we can also get the conclusion of Theorem 2 of [10].
Throughout this paper, we always suppose that B is a sufficiently large positive constant, ε is a sufficiently small positive constant and E = B
2, δ = ε
13. We also suppose that N is sufficiently large and that A = N
1087 +12ε, Y = N
127+ε, Q =
12√
A log
−64BN . Let c, c
1and c
2denote positive constants which have different values at different places. m ∼ M means that there are positive constants c
1and c
2such that c
1M < m ≤ c
2M . We often use M (s, χ) (M may be another capital letter) to denote a Dirichlet polynomial of the form
M (s, χ) = X
m∼M
a(m)χ(m) m
s,
where a(m) is a complex number with a(m) = O(1), and χ is a character mod q.
The author would like to thank Professors Wang Yuan and Pan Cheng- biao for their encouragement.
2. Some preliminary work. Let χ be a character mod q and let χ
0be a principal character. Set E
0= 1 if χ = χ
0, and E
0= 0 if χ 6= χ
0.
We divide the characters mod q into two classes. We call χ a good char- acter if L(s, χ) has no zeros in the region
(1) 1 − 24E log log Y
ε log Y ≤ σ ≤ 1, |t| ≤ 6Y.
Otherwise, we call χ a bad character .
By Siegel’s theorem and the zero free region of the L-function (see pp. 255 and 257 of [15]), we know that L(s, χ) 6= 0 in the region
σ > 1 − c(µ)
max(q
µ, log
45(|t| + 2)) , where µ (> 0) is arbitrary and c(µ) > 0.
Take µ = ε/(500E). If a bad character exists, then
(2) q ≥ log
300EεN.
Moreover, by the estimation for the zero density (see [14] and [5]), X
χ (mod q)
N (σ, T, χ) (qT )
(125+ε)(1−σ)log
14qT,
we know that for any modulus q ≤ Q, the number of bad characters is O(log
64EεN ). Let
I(a, q) =
a
q − log
2EN qY , a
q + log
2EN qY
and let
E
1= [
q≤logEN
[
q (a,q)=1a=1I(a, q), E
2= [1/Q, 1 + 1/Q) − E
1.
Lemma 1. Assume that r (≥ 2) and q are positive integers, d
r(n) = P
n=n1...nr
1 and x
ε< y ≤ x. Then X
x−y<n≤x
d
qr(n) y(log x)
rq−1.
P r o o f. See Theorem 2 of [20].
Let
(3) τ (χ) =
X
q r=1χ(r)e
r q
.
Lemma 2. τ (χ
0) = µ(q). For any character χ mod q, |τ (χ)| ≤ √ q.
P r o o f. See pp. 24 and 28 of [15].
Lemma 3. For any complex numbers a(n),
η
\
−η
X
n
a(n)e(nt)
2dt η
2∞
\
−∞
X
x<n≤x+1/(2η)
a(n)
2dx.
P r o o f. See Lemma 1 of [1].
Lemma 4. Suppose T ≥ 1, χ is a character modq, and P
m∼M
|a(m)|
2M log
cT . Then X
χ (mod q) T
\
−T
X
m∼M
a(m)χ(m) m
12+it2
dt (M + qT ) log
cT.
P r o o f. See Theorem 3 on p. 38 of [15].
3. Mean value estimate (I)
Lemma 5. Assume that Y
δH Y
13516, M H = Y , q ≤ Q, χ is a character modq, M (s, χ) is a Dirichlet polynomial and
H(s, χ) = X
h∼H
Λ(h)χ(h)
h
s.
Let η = qQ/Y , b = 1 + 1/ log N and T
0= log
δ2EY . Then for T
0≤ T ≤ Y , we have
min
2η, 1
T
X
χ (good) 2T
\
T
|M (b + it, χ)H(b + it, χ)|
2dt η
2log
−6EN.
P r o o f. Let s = b + it and let χ be a good character modq. By (1) and the zero free region of the L-function, we know that for |t| ≤ 2Y ,
(4) X
c1H<h≤c2H
Λ(h)χ(h)
h
s= E
0(c
2H)
1−s− (c
1H)
1−s1 − s + O(log
−2Eδ2Y ).
So, for T
0≤ |t| ≤ 2Y ,
(5) H(s, χ) log
−δ2EY.
According to the discussion in [2], there are O(log
2Y ) sets S(V, W ), where S(V, W ) = {t
j(χ) : χ runs through all good characters mod q, j = 1, . . . , J(χ), |t
r(χ) − t
s(χ)| ≥ 1 (r 6= s)}. For t
j(χ) ∈ S(V, W ),
V ≤ M
12|M (b + it
j(χ), χ)| < 2V, W ≤ H
12|H(b + it
j(χ), χ)| < 2W,
where Y
−1≤ M
−12V , Y
−1≤ H
−12W and V M
12, W H
12log
−δ2EY . Thus
(6) X
χ (good) 2T
\
T
|M (b + it, χ)H(b + it, χ)|
2dt V
2W
2Y
−1log
2Y |S(V, W )|, where S(V, W ) is one of the sets with the above properties.
Assume Y
k+11≤ H ≤ Y
1k, where k is a positive integer, k ≥ 8 and kδ 1. Applying the mean value estimate (see Theorem 3 on p. 632 of [16]) and Lemma 1 to M (s, χ) and H
k(s, χ), we have
|S(V, W )| V
−2(M + qT ) log
dY,
|S(V, W )| W
−2k(H
k+ qT ) log
dY,
where d = c/δ
2. Applying the Hal´asz method (see Theorem 6 on p. 650 of [16]) to M (s, χ) and H
k(s, χ), we have
|S(V, W )| (V
−2M + V
−6qT M ) log
dY,
|S(V, W )| (W
−2kH
k+ W
−6kqT H
k) log
dY.
Thus,
V
2W
2|S(V, W )| V
2W
2F log
dY,
where
F = min{V
−2(M + qT ), V
−2M + V
−6qT M, W
−2k(H
k+ qT ),
W
−2kH
k+ W
−6kqT H
k}.
It will be proved that
(7) min
2η, 1
T
V
2W
2F η
2Y log
−2δ2EN.
We consider four cases.
(a) F ≤ 2V
−2M , 2W
−2kH
k. Then
V
2W
2F V
2W
2min{V
−2M, W
−2kH
k}
≤ V
2W
2(V
−2M )
1−2k1(W
−2kH
k)
2k1= V
k1W M
1−2k1H
12Y log
−2δ2EN.
(b) F > 2V
−2M, 2W
−2kH
k. Then
V
2W
2F V
2W
2min{V
−2qT, V
−6qT M, W
−2kqT, W
−6kqT H
k}
≤ V
2W
2(V
−2)
1−2k3(V
−6M )
2k1(W
−2k)
1kqT = qT M
2k1. Since k ≥ 8, we have H ≥ Y
k+11≥ Y
1−k9, M
2k1Y
181and so
min
2η, 1
T
V
2W
2F η
T Y
181qT η
2Y
1−ε. (c) F ≤ 2V
−2M, F > 2W
−2kH
k. Then
V
2W
2F V
2W
2min{V
−2M, W
−2kqT, W
−6kqT H
k}
≤ V
2W
2(V
−2M )
1−3k1(W
−6kqT H
k)
3k1M H
13(qT )
3k1,
since V M
12. As H ≥ Y
k+11≥ (
YQ)
2k1Y
2ε, we have
min
2η, 1
T
V
2W
2F η
2−3k1T
−3k1Y
Y Q
−13k
(qT )
3k1Y
−εη
2Y
1−ε. (d) F > 2V
−2M , F ≤ 2W
−2kH
k. Then
V
2W
2F V
2W
2min{V
−2qT, V
−6qT M, W
−2kH
k}
≤ V
2W
2(V
−2qT )
1−2k3(V
−6qT M )
2k1(W
−2kH
k)
k1= (qT )
1−k1HM
2k1.
If k ≥ 9, then H ≤ Y
1k≤ Y
1−17(k−1)9(2k−1), M Y
17(k−1)9(2k−1), and so min
2η, 1
T
V
2W
2F η
2Y
Y Q
1−1k
Y
−1718(1−1k)η
2Y
1−ε. If k = 8, then Y
19≤ H Y
13516, M Y
119135, and so
min
2η, 1
T
V
2W
2F η
2Y
Y Q
78
Y
−119144η
2Y
1−ε. Combining the above, we obtain (7). Hence, Lemma 5 follows.
Lemma 6. Assume that M Y
1936, M L = Y , q ≤ Q, χ is a character modq, M (s, χ) is a Dirichlet polynomial, and
F (s, χ) = M (s, χ) X
l∼L
χ(l) l
s. Let η = qQ/Y , b = 1 + 1/ log N and T
1= p
L/q. Then for T
1≤ T ≤ Y , we have
min
2η, 1
T
X
χ (mod q) 2T
\
T
|F (b + it, χ)|
2dt η
2Y
−ε.
P r o o f. Perron’s formula yields X
l∼L
χ(l) l
b+it= 1
2πi
4iT
\
−4iT
L(b + it + s, χ) (c
2L)
s− (c
1L)
ss ds
+ O
Y
εT
+ O
1 L
= 1 2πi
12−b+4iT
\
1 2−b−4iT
L(b + it + s, χ) (c
2L)
s− (c
1L)
ss ds
+ O
Y
εT
+ O
1 L
+ O
L
−12(qT )
14+εT
,
X
l∼L
χ(l) l
b+it2
L
−1log T
6T
\
−6T
L
1
2 + iu, χ
2
du
1 + |t − u|
+ O
Y
2εT
2+ O
1 L
2+ O
L
−1(qT )
12+2εT
2.
By Lemma 4 and the mean value estimate for the fourth power of the L-function, we have
X
χ (mod q) 2T
\
T
|F (b + it, χ)|
2dt
L
−1log N X
χ (mod q) 2T
\
T
|M (b + it, χ)|
2dt
6T\
−6T
L
1
2 + iu, χ
2
du
1 + |t − u|
+ O
Y
2εT
2+ 1
L
2+ L
−1(qT )
12+2εT
2X
χ (mod q) 2T
\
T
|M (b + it, χ)|
2dt
L
−1log
2N
X
χ (mod q) 2T
\
T
|M (b + it, χ)|
4dt
12
×
X
χ (mod q) 2T
\
T
dt
6T
\
−6T
|L(
12+ iu, χ)|
41 + |t − u| du
12
+ O
Y
2εT
2+ 1
L
2+ L
−1(qT )
12+2εT
21 + qT M
L
−11 + qT
M
2 12
(qT )
12log
10N + Y
−2ε. Hence,
min
2η, 1
T
X
χ (mod q) 2T
\
T
|F (b + it, χ)|
2dt
min
2η, 1
T
Y
−1(Y
1918+ qT )
12(qT )
12log
10N + η
2Y
−2εη
2Y
−1Y
1918+ Y Q
12
Y Q
12
log
10N + η
2Y
−2εη
2Y
−2ε. Thus Lemma 6 follows.
4. Mean value estimate (II)
Lemma 7. Assume that M HK = Y , q ≤ Q, χ is a character mod q,
M (s, χ), H(s, χ) and K(s, χ) are Dirichlet polynomials and G(s, χ) =
M (s, χ)H(s, χ)K(s, χ). Let η = qQ/Y , b = 1 + 1/ log N , T
0= log
δ2EY .
Assume further that for T
0≤ |t| ≤ 2Y , M (b + it, χ) log
−δ2EY and
H(b + it, χ) log
−δ2EY . Moreover , suppose that M and H satisfy one of the following four conditions:
1) M H Y
142261, Y
1799H, M
29/H Y
919, Y
1757M , H
29/M Y
193, Y
1733M
1211H;
2) M H Y
4781, Y
13516H, M
2919H Y
11489, Y
6881M
2H, H
4/M Y
12, Y
14785M
5849H;
3) M H Y
113198, Y
1790H, M
6H Y
209, Y
1763M , M
18H Y
247, Y
1730M
65H;
4) M H Y
2542, Y
13516H, M
2315H Y
79, Y
1721M
2H, H
3/M Y
187, Y
13568M
1415H, Y
59M H.
Then for T
0≤ T ≤ Y , we have (8) min
2η, 1
T
X
χ (good) 2T
\
T
|G(b + it, χ)|
2dt η
2log
−6EN.
P r o o f. We only show that for T = 1/η = Y /(qQ),
(9) I = X
χ (good) 2T
\
T