CONSECUTIVE PRIMES IN SHORT INTERVALS

ARTYOM RADOMSKII

Abstract. We obtain a lower bound for

arXiv:2012.12672v1 [math.NT] 23 Dec 2020

#{x/2 < pn ≤ x : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m −pn ≤ y},

where pn is the nth prime.

1. Introduction
Let pn denote the nth prime. We prove the following result.
Theorem 1.1. There are positive absolute constants c and C such that
the following holds. Let ε be a real number with 0 < ε < 1. Then there
is a number c0 (ε) > 0, depending only on ε, such that if x ∈ R, y ∈ R,
m ∈ Z, q ∈ Z, a ∈ Z are such that
c0 (ε) ≤ y ≤ ln x,
1 ≤ m ≤ c · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1,
then
#{x/2 < pn ≤ x : pn ≡ · · · ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y} ≥
 exp(Cm)
y
≥ π(x) .
2q ln x
Theorem 1.1 extends a result of Maynard [5, Theorem 3.3] which
showed the same result but with y = ε ln x.
From Theorem 1.1 we obtain
Corollary 1.1. There is an absolute constant C > 0 such that if m is
a positive integer, x and y are real numbers satisfying exp(Cm) ≤ y ≤
ln x, then
 y exp(Cm)
#{x/2 < pn ≤ x : pn+m − pn ≤ y} ≥ π(x) .
2 ln x
Let us introduce somePnotation. The symbolQ b|a means that b divides
a. For fixed a the sum b|a and the product b|a should be interpreted
as being over all positive divisors of a.
Key words and phrases. Euler’s totient function, sieve methods, the distribution
of prime numbers.
This work was performed at the Steklov International Mathematical Center and
supported by the Ministry of Science and Higher Education of the Russian Feder-
ation (agreement no. 075-15-2019-1614).

1
2 ARTYOM RADOMSKII

We shall use the notation of I. M. Vinogradov. The symbol A ≪ B

means that |A| ≤ cB, where c is a positive absolute constant. P
We reserve the letter p for primes. In particular, the sum p≤K
should be interpreted as being over all prime numbers not exceeding
K.
We also use the following notation:
#A — the number of elements of a finite set A.
N — the set of all positive integers.
Z — the set of all integers.
R — the set of all real numbers.
C — the set of all complex numbers.
P — the set of all prime numbers.
[x] — the integer part of a number x, i. e. [x] is the largest integer n
such that n ≤ x.
{x} — the fractional part of a number x, i. e. {x} = x − [x].
⌈x⌉ — the smallest integer n such that n ≥ x.
Re(s) — the real part of a complex number s.
Im(s) — the imaginary part of a complex number s.
(a1 , . . . , an ) — the greatest common divisor of integers a1 , . . . , an .
[a1 , . . . , an ] — the least common multiple of integers a1 , . . . , an .
ϕ(n) — the Euler totient function:
ϕ(n) = #{1 ≤ m ≤ n : (m, n) = 1}.
µ(n) — the Moebius function, which is defined as follows:
i) µ(1) = 1;
ii) µ(n) = 0, if there is a prime p such that p2 |n;
iii) µ(n) = (−1)s , if n = q1 · · · qs , where q1 < . . . < qs are primes.
Λ(n) — the von Mangoldt function:
(
ln p, if n = pk ;
Λ(n) =
0, if n 6= pk .
P −
(n) — the least prime factor of n (by convention P − (1) = +∞).
n
k
= n!/(k!(n − k)!) — the binomial coefficient.
If (a, b) = 1, a and b are said to be prime to one another or coprime.
The numbers a, b, c, . . . , k are said to be coprime if every two of them
are coprime.
For real numbers x, y we also use (x, y) to denote the open interval
and [x, y] to denote the closed interval. Also by (a1 , . . . , an ) we denote
a vector. The usage of the notation should be clear from the context.
We put X Y
= 0, = 1.
∅ ∅
We define
M = {n ∈ N : µ(n) 6= 0}.
We use the following functions:
 CONSECUTIVE PRIMES IN SHORT INTERVALS 3

Z x
dt
li(x) = ;
2 ln t
Φ(x, z) = #{1 ≤ n ≤ x : P − (n) > z};
X
π(x) = 1;
p≤x
X
θ(x) = ln p;
p≤x
X
ψ(x) = Λ(n);
n≤x
X
π(x; q, a) = 1;
p≤x
p≡a (mod q)
X
ψ(x; q, a) = Λ(n).
n≤x
n≡a (mod q)

Let m be an integer with m > 1 and a be an integer. If (a, m) = 1,

then aϕ(m) ≡ 1 (mod m) (the Fermat–Euler theorem). Let d be the
smallest positive value of γ for which aγ ≡ 1 (mod m). We call d the
order of a (mod m), and say that a belongs to d (mod m).
Let q be a positive integer. We recall that a Dirichlet character
modulo q is a function χ : Z → C such that
1) χ(n + q) = χ(n) for all n ∈ Z (i. e. χ is a periodic function with the
period q);
2) χ(mn) = χ(m)χ(n) for all m, n ∈ Z (i. e. χ is a totally multiplicative
function);
3) χ(1) = 1;
4) χ(n) = 0 for all n ∈ Z such that (n, q) > 1.
By Xq we denote the set of all Dirichlet characters modulo q. We
recall that #Xq = ϕ(q) and that the principal character modulo q is
(
1, if (n, q) = 1;
χ0 (n) :=
0, if (n, q) > 1.
Let χ ∈ Xq . We say that the character χ restricted by (n, q) = 1 has
period q1 if it has the property that χ(m) = χ(n) for all m, n ∈ Z such
that (m, q) = 1, (n, q) = 1 and m ≡ n (mod q1 ). Let c(χ) denote the
conductor of χ, defined to be the least positive integer q1 such that χ
restricted by (n, q) = 1 has period q1 . We say that χ is primitive if
c(χ) = q, and imprimitive precisely if c(χ) < q. By Xq∗ we denote the
set of all primitive characters modulo q. We observe that the principal
character modulo 1 is primitive. On the other hand, any principal
character modulo q > 1 is imprimitive, since its conductor is clearly 1.
4 ARTYOM RADOMSKII

For χ ∈ Xq we put
(
1, if χ is the principal character modulo q;
Eχ0 (χ) :=
0, otherwise;
X
ψ(x, χ) = Λ(n)χ(n);
n≤x
′
ψ (x, χ) = ψ(x, χ) − Eχ0 (χ)x.
A character χ is called real if χ(n) ∈ R for all n ∈ Z. A character χ
is called complex if there is an integer n such that Im(χ(n)) 6= 0.
We say that characters χ1 and χ2 (modulo q1 and modulo q2 respec-
tively) are equal and write χ1 = χ2 if χ1 (n) = χ2 (n) for any integer n.
Otherwise, we say that characters χ1 and χ2 are not equal and write
χ1 6= χ2 .
Let χ be a Dirichlet character modulo q. The corresponding L-
function is defined by series
X∞
χ(n)
L(s, χ) =
n=1
ns
for s ∈ C such that Re(s) > 1. It is well-known, if χ is not the principal
character modulo q, then L(s, χ) can be analytically continued to C. If
χ is the principal character modulo q, then L(s, χ) can be analytically
continued to C \ {1} with a simple pole at s = 1.
We say that two linear functions with integer coefficients L1 (n) =
a1 n + b1 and L2 (n) = a2 n + b2 are equal and write L1 = L2 if a1 = a2
and b1 = b2 . Otherwise, we say that linear functions L1 and L2 are not
equal and write L1 6= L2 .
Let L = {L1 , . . . , Lk } be a set of k linear functions with integer
coefficients:
Li (n) = ai n + bi , i = 1, . . . , k.
For L(n) = an + b, a, b ∈ Z, we define
k
Y
∆L = |a| |abi − bai |.
i=1
We say that L = an + b ∈ L if there is 1 ≤ i ≤ k such that L = Li .
Otherwise, we say that L = an + b ∈ / L.
This paper is organized as follows. In sections 2 – 4 we give necessary
lemmas. In section 5 we prove Theorem 1.1 and Corollary 1.1.
2. Preparatory Lemmas
In this section we give some well-known lemmas which will be used
in the following sections.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 5

Lemma 2.1 (The Fundamental Theorem of Arithmetic; see, for ex-

ample, [7, Chapter 1]). Let n be an integer with n > 1. Then n is a
product of primes and, apart from rearrangement of factors, n can be
expressed as a product of primes in one way only.
Let n be an integer with n > 1. From Lemma 2.1 we have
n = qe1 · · · qel ,
where qei , i = 1, . . . , l, are primes. The primes qei , i = 1, . . . , l, are
not necessarily distinct, nor arranged in any particular order. If we
arrange them in increasing order, associate sets of equal primes into
single factors, we obtain
n = q1α1 · · · qrαr ,
where q1 < . . . < qr are primes and α1 , . . . , αr are positive integers.
We then say that n is expressed in standard form. From Lemma 2.1 it
follows that the standard form of n is unique.
Lemma 2.2 (see, for example, [6, Chapter 1]). Let x be a real number
with x ≥ 2. Then
Y 1 −1
b1 ln x ≤ 1− ≤ b2 ln x,
p≤x
p
Y 1
b3 ln x ≤ 1+ ≤ b4 ln x,
p≤x
p
where b1 , b2 , b3 and b4 are positive absolute constants.
Lemma 2.3 (see, for example, [4, Chapters 1, 2]). The limits limx→∞ ψ(x)/x,
limx→∞ θ(x)/x, limx→∞ π(x)/(x/ ln x), limn→∞ pn /(n ln n) exist and
ψ(x) θ(x) π(x)
lim = lim = lim = 1,
x→+∞ x x→+∞ x x→+∞ x/ ln x
pn
lim = 1.
n→+∞ n ln n

From Lemma 2.3 we obtain

Lemma 2.4. The following holds
b5 x ≤ ψ(x) ≤ b6 x, x ≥ 2; (2.1)
b7 x ≤ θ(x) ≤ b8 x, x ≥ 2;
x x
b9 ≤ π(x) ≤ b10 , x ≥ 2; (2.2)
ln x ln x
b11 n ln(n + 2) ≤ pn ≤ b12 n ln(n + 2), n ≥ 1,
where b5 , b6 , b7 , b8 , b9 , b10 , b11 and b12 are positive absolute constants.
6 ARTYOM RADOMSKII

Lemma 2.5 (see, for example, [7, Chapter 2]). Let n be an integer
with n > 1. Then Y 1
ϕ(n) = n 1− .
p
p|n

From Lemma 2.5 we obtain the following two lemmas.

Lemma 2.6. Let m and n be integers with m ≥ 1 and n ≥ 1. Then
ϕ(mn) ≥ ϕ(m)ϕ(n).
Lemma 2.7. Let m and n be integers with m ≥ 1, n ≥ 1 and (m, n) =
1. Then
ϕ(mn) = ϕ(m)ϕ(n).
Lemma 2.8. Let n be an integer with n ≥ 1. Then
n X µ2 (d)
= . (2.3)
ϕ(n) ϕ(d)
d|n

Proof of Lemma 2.8. For n = 1 the equality (2.3) holds. Let

n > 1. Let n be expressed in standard form
n = q1α1 · · · qrαr ,
where q1 < . . . < qr are prime numbers. Applying Lemmas 2.5 and 2.7,
we have
n Y 1 −1 Y 1 
= 1− = 1+ =
ϕ(n) p p−1
p|n p|n
 1   1 
= 1+ ··· 1+ =
q1 − 1 qr − 1
 1   1 
= 1+ ··· 1+ =
ϕ(q1 ) ϕ(qr )
Xr X 1
=1+ =
s=1 1≤i <...<i ≤r
ϕ(qi1 ) · · · ϕ(qis )
1 s
r
X X 1
=1+ =
s=1 1≤i1 <...<is ≤r
ϕ(qi1 · · · qis )
X 1 X µ2 (d)
= = .
ϕ(d) ϕ(d)
d|n d|n
d∈M
Lemma 2.8 is proved.
Lemma 2.9 (see, for example, [6, Chapter 1]). Let n be an integer
with n ≥ 1. Then
n
ϕ(n) ≥ c ,
ln ln(n + 2)
where c > 0 is an absolute constant.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 7

Lemma 2.10 (see, for example, [1, Chapter 28]). Let x be a real num-
ber with x ≥ 2. Then
X 1
≤ c ln x,
1≤n≤x
ϕ(n)
where c > 0 is an absolute constant.
Lemma 2.11 (see, for example, [2, Chapter 5]). Let n be an integer
with n ≥ 1. Then
X ln p
≤ c ln ln(3n),
p
p|n
where c > 0 is an absolute constant.
Lemma 2.12. Let a, b and c be integers with (a, b)|c. Then the equa-
tion
ax + by = c (2.4)
has a solution in the integers.
Proof of Lemma 2.12. We put d = (a, b). We have c = dl, where
l ∈ Z. It is well-known (see, for example, [7, Chapter 1, Exercise 1]),
the equation
ax + by = d (2.5)
has a solution in the integers. Let x0 ∈ Z and y0 ∈ Z be a solution
of (2.5). Then the integers lx0 and ly0 satisfy (2.4). Lemma 2.12 is
proved.
Lemma 2.13. Let n and k be integers with 1 ≤ k ≤ n. Then
 
n
≥ k −k (n − k)k . (2.6)
k
Proof of Lemma 2.13. For k = n the inequality (2.6) holds. Let
1 ≤ k < n. Then
 
n n! n(n − 1) · · · (n − k + 1) (n − k)k
= = ≥ ≥
k k!(n − k)! k! k!
≥ k −k (n − k)k .
Lemma 2.13 is proved.
Lemma 2.14 (see [3, Chapter 0]). Let x and z be real numbers with
2 ≤ z ≤ x/2. Then
x
Φ(x, z) ≥ c0 ,
ln z
where c0 > 0 is an absolute constant.
8 ARTYOM RADOMSKII

3. Lemmas on Dirichlet characters

In this section we give some well-known lemmas on Dirichlet char-
acters which will be used in the following sections.
Lemma 3.1. Let q be an integer with q > 1 and χ be a Dirichlet
character modulo q. Let n be an integer with (n, q) = 1. Then
 b 
χ(n) = exp 2πi , b ∈ {0, . . . , ϕ(q) − 1}.
ϕ(q)
In particular, |χ(n)| = 1.
Proof of Lemma 3.1. By the Fermat–Euler theorem, we have
nϕ(q) ≡ 1 (mod q).
Hence,
(χ(n))ϕ(q) = 1.
We obtain
 b 
χ(n) = exp 2πi , b ∈ {0, . . . , ϕ(q) − 1}.
ϕ(q)
In particular, |χ(n)| = 1. Lemma 3.1 is proved.
Lemma 3.2. Let q be a positive integer and χ be a real character
modulo q. Let n be an integer with (n, q) = 1. Then χ(n) ∈ {−1, 1}.
Proof of Lemma 3.2. If q = 1, then χ(m) = 1 for any integer m.
Let q be an integer with q > 1 and χ be a real character modulo q. Let
n be an integer with (n, q) = 1. From Lemma 3.1 we have |χ(n)| = 1.
Since χ(n) is real, we see that χ(n) ∈ {−1, 1}. Lemma 3.2 is proved.
Lemma 3.3. Let a, b and n be integers with 1 ≤ a < b, a|b and
(n, a) = 1. Then there is an integer t such that (n + ta, b) = 1.
Proof of Lemma 3.3. If (n, b) = 1, we take t = 0. Let (n, b) > 1.
Then the set
Ω = {p|b : p ∤ a} = 6 ∅.
Let
Ω = {q1 , . . . , qr }, q1 < . . . < qr .
Let 1 ≤ i ≤ r. Since (a, qi ) = 1, the congruence
n + ta ≡ 1 (mod qi )
has a solution, i. e. there is an integer mi such that n + ami ≡
1 (mod qi ). We consider the system

t ≡ m1 (mod q1 )

.. (3.1)
 .

t ≡ mr (mod qr ).
 CONSECUTIVE PRIMES IN SHORT INTERVALS 9

Since the numbers q1 , . . . , qr are coprime, the system has a solution.

Let an integer t0 satisfy the system (3.1). We claim that the number
t0 is desired, i. e. we claim that (n + t0 a, b) = 1. Assume the converse:
(n + t0 a, b) > 1. Then there is a prime p such that p|b and p|(n + t0 a).
If p ∤ a, then p ∈ Ω, i. e. p = qi for some 1 ≤ i ≤ r. But
n + t0 a ≡ 1 (mod qi )
and hence p ∤ (n + t0 a). We obtain a contradiction. Thus this case is
impossible. Hence, p|a. Since p|(n + t0 a), we obtain that p|n. Hence,
(n, a) > 1. This contradicts the assumption of the lemma. Therefore,
the assumption (n + t0 a, b) > 1 is false. Hence, (n + t0 a, b) = 1. Lemma
3.3 is proved.
Lemma 3.4. Let q be an integer with q ≥ 2, let χ ∈ Xq . Suppose that
χ restricted by (n, q) = 1 has period q1 . Then χ restricted by (n, q) = 1
has also period (q, q1 ).
Proof of Lemma 3.4. We put δ = (q, q1 ). Let m and n be integers
such that (m, q) = 1, (n, q) = 1 and m ≡ n (mod δ). We must prove
that χ(m) = χ(n). By Lemma 2.12, there are integers k and l such
that
m + q1 k = n + ql.
We put A = m + q1 k = n + ql. Since (n, q) = 1, we have (n + ql, q) = 1.
Hence, (A, q) = 1. Since χ has period q, we have
χ(A) = χ(n + ql) = χ(n).
Since (A, q) = 1, (m, q) = 1 and A ≡ m (mod q1 ), we have χ(A) =
χ(m). Hence, χ(m) = χ(n). Lemma 3.4 is proved.
Lemma 3.5. Let q ≥ 1 and χ ∈ Xq . Then c(χ) divides q.
Proof of Lemma 3.5. If q = 1, then c(χ) = 1 and the statement
is obvious. Let q ≥ 2. By Lemma 3.4, χ restricted by (n, q) = 1 has
period δ = (c(χ), q). If c(χ) is not a divisor of q, then δ < c(χ), which
contradicts the definition of the conductor. Lemma 3.5 is proved.
Lemma 3.6. Let q ≥ 1 and χ ∈ Xq . Then there exists a unique
Dirichlet character χ1 ∈ Xc(χ) such that
(
χ1 (n), if (n, q) = 1;
χ(n) = (3.2)
0, if (n, q) > 1.
Furthermore, χ1 is primitive.
We say that χ1 induces χ.
Proof of Lemma 3.6. I) Let q = 1. Then c(χ) = 1, #X1 = 1,
χ1 = χ and the statement is obvious.
II) Let q ≥ 2 and χ be a primitive character modulo q. Then c(χ) = q
and we can take χ1 = χ. Let us prove the uniqueness. Suppose that
10 ARTYOM RADOMSKII

there are two different characters χ1 , χ2 ∈ Xq satisfying (3.2). Then for

any n such that (n, q) > 1, we have χ1 (n) = 0 = χ2 (n). For any n such
that (n, q) = 1, we have χ1 (n) = χ(n) = χ2 (n). Hence, χ1 (n) = χ2 (n)
for any integer n, i. e. χ1 = χ2 . We obtain a contradiction.
III) Let q ≥ 2 and χ be an imprimitive character modulo q. Then
1 ≤ c(χ) < q and by Lemma 3.5 we have c(χ)|q. We define χ1 . Let
n ∈ Z. We consider several cases.
1) (n, c(χ)) > 1. Then we put χ1 (n) = 0.
2) (n, c(χ)) = 1. By Lemma 3.3 there is an integer t such that
(n + tc(χ), q) = 1. (3.3)
We put
χ1 (n) = χ(n + tc(χ)).
The choice of t, subject to (3.3), is immaterial, since χ restricted by
(n, q) = 1 has period c(χ). Thus, χ1 (n) is defined for any integer n.
We claim that χ1 is a character modulo c(χ). By construction,
χ1 (n) = 0 for any n ∈ Z such that (n, c(χ)) > 1.
By Lemma 3.3, there is an integer t such that (1 + tc(χ), q) = 1. Since
the choice of such t is immaterial, we take t = 0. We have
χ1 (1) = χ(1) = 1.
Now we prove that
χ1 (n + c(χ)) = χ1 (n) for any n ∈ Z. (3.4)
If (n, c(χ)) > 1, then we have (n + c(χ), c(χ)) > 1. Hence,
χ1 (n + c(χ)) = 0 = χ1 (n).
Let (n, c(χ)) = 1. Then we have (n + c(χ), c(χ)) = 1. By Lemma
3.3, there are integers t1 and t2 such that (n + t1 c(χ), q) = 1 and
(n + c(χ) + t2 c(χ), q) = 1. By construction, we have
χ1 (n) = χ(n + t1 c(χ)),
χ1 (n + c(χ)) = χ(n + c(χ) + t2 c(χ)).
Since χ restricted by (n, q) = 1 has period c(χ), we have χ(n+t1 c(χ)) =
χ(n + c(χ) + t2 c(χ)). Hence, χ1 (n) = χ1 (n + c(χ)) and (3.4) is proved.
Now we prove that
χ1 (mn) = χ1 (m)χ1 (n) for any m, n ∈ Z. (3.5)
If (m, c(χ)) > 1, then we have (mn, c(χ)) > 1. Hence, χ1 (mn) = 0,
χ1 (m) = 0. Therefore (3.5) holds. Similarly, (3.5) holds if (n, c(χ)) > 1.
Let (m, c(χ)) = 1, (n, c(χ)) = 1. Then (mn, c(χ)) = 1. By Lemma
3.3, there are integers t1 , t2 and t3 such that (m + t1 c(χ), q) = 1,
 CONSECUTIVE PRIMES IN SHORT INTERVALS 11

(n + t2 c(χ), q) = 1 and (mn + t3 c(χ), q) = 1. We put m1 = m + t1 c(χ),

n1 = n + t2 c(χ) and u = mn + t3 c(χ). By construction,
χ1 (mn) = χ(u),
χ1 (m) = χ(m1 ),
χ1 (n) = χ(n1 ).
Since χ is a totally multiplicative function, we have
χ1 (m)χ1 (n) = χ(m1 )χ(n1 ) = χ(m1 n1 ).
Since (m1 , q) = 1, (n1 , q) = 1, we have (m1 n1 , q) = 1. It is clear that
m1 n1 ≡ u (mod c(χ)). Since χ restricted by (n, q) = 1 has period c(χ),
we have χ(u) = χ(m1 n1 ). Hence, χ1 (mn) = χ1 (m)χ1 (n) and (3.5) is
proved. Thus, we have proved that χ1 is a character modulo c(χ), i. e.
χ1 ∈ Xc(χ) .
Now we prove that χ1 satisfies (3.2). It suffices to show that
χ1 (n) = χ(n), if (n, q) = 1. (3.6)
Since (n, q) = 1, we have (n, c(χ)) = 1 (see Lemma 3.5). By Lemma
3.3, there is an integer t such that (n + tc(χ), q) = 1. By construction
χ1 (n) = χ(n+tc(χ)). Since (n+tc(χ), q) = 1, (n, q) = 1 and n+tc(χ) ≡
n (mod c(χ)), we have χ(n + tc(χ)) = χ(n). Hence, χ1 (n) = χ(n) and
(3.6) is proved.
Now we prove that χ1 is a primitive character. Suppose that a posi-
tive integer q2 satisfies the property that χ1 restricted by (n, c(χ)) = 1
has period q2 . Let m and n be integers such that (m, q) = 1, (n, q) = 1
and m ≡ n (mod q2 ). By Lemma 3.5, we have (m, c(χ)) = 1, (n, c(χ)) =
1. We have (see (3.6))
χ(m) = χ1 (m) = χ1 (n) = χ(n).
Hence, χ restricted by (n, q) = 1 has period q2 . From the definition of a
conductor it follows that q2 ≥ c(χ). Hence, χ1 is a primitive character.
Now we prove the uniqueness. Suppose that there are two different
characters χ1 , χ2 ∈ Xc(χ) satisfying (3.2). If (n, c(χ)) > 1, then χ1 (n) =
0 = χ2 (n). Let (n, c(χ)) = 1. By Lemma 3.3, there is an integer t such
that (n + tc(χ), q) = 1. Since χ1 and χ2 are periodic functions with
period c(χ), we have
χ1 (n) = χ1 (n + tc(χ)) = χ(n + tc(χ)) = χ2 (n + tc(χ)) = χ2 (n).
Thus, χ1 (n) = χ2 (n) for any n ∈ Z and hence χ1 = χ2 . We obtain a
contradiction. The uniqueness is proved. Lemma 3.6 is proved.
Lemma 3.7. Let q be an integer with q > 1 and let q be expressed in
standard form
q = q1α1 · · · qrαr ,
12 ARTYOM RADOMSKII

where q1 < . . . < qr are primes and α1 , . . . , αr are positive integers. Let
χ be a Dirichlet character modulo q. Then there exist unique characters
χi modulo qiαi , i = 1, . . . , r, such that
χ(n) = χ1 (n) · · · χr (n) for all n. (3.7)
Furthermore, if the character χ is real, then all characters χi , i =
1, . . . , r, are real. If the character χ is primitive, then all characters
χi , i = 1, . . . , r, are primitive.
Proof of Lemma 3.7. For any 1 ≤ i ≤ r we take Ai such that
(
Ai ≡ 1 (mod qiαi ),
α (3.8)
Ai ≡ 0 (mod qj j ) for any 1 ≤ j ≤ r, j 6= i.
Since the moduli of these congruences are coprime, the system has a
solution (see, for example, [7, Chapter 4]). Thus, integers A1 , . . . , Ar
are defined.
Let 1 ≤ i ≤ r and n ∈ Z. We put
X
χi (n) = χ(nAi + Aj ). (3.9)
1≤j≤r
j6=i

We claim that χi is a Dirichlet character modulo qiαi .

1) It is clear that χi : Z → C.
2) Let us show that
χi (n + qiαi ) = χi (n) for all n ∈ Z. (3.10)
Let n ∈ Z. We put
X X
m1 = (n + qiαi )Ai + Aj , m2 = nAi + Aj .
1≤j≤r 1≤j≤r
j6=i j6=i

We have
χi (n + qiαi ) = χ(m1 ),
χi (n) = χ(m2 ),
m1 − m2 = qiαi Ai .
α
The number Ai is divisible by qj j for any 1 ≤ j ≤ r, j 6= i. Hence, the
number m1 − m2 is divisible by q, i. e.
m1 ≡ m2 (mod q).
Therefore χ(m1 ) = χ(m2 ) and, hence, χi (n+qiαi ) = χi (n). Thus, (3.10)
is proved.
3) Let us show that
χi (n) = 0 for all n ∈ Z such that (n, qiαi ) > 1. (3.11)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 13

Let n ∈ Z be such that (n, qiαi ) > 1. We put

X
m = nAi + Aj .
1≤j≤r
j6=i

We have
χi (n) = χ(m).
The number qi divides n and divides Aj for any 1 ≤ j ≤ r, j 6= i.
Hence, the number qi divides m. Therefore (m, q) > 1. We obtain
χ(m) = 0 and, hence, χi (n) = 0. Thus, (3.11) is proved.
4) Let us show that
χi (1) = 1. (3.12)
We put
Xr
m= Aj .
j=1
Then
χi (1) = χ(m).
We see from (3.8) that m ≡ 1 (mod qiαi ), i = 1, . . . , r. The number
m − 1 is divisible by qiαi for any i = 1, . . . , r. Hence, the number m − 1
is divisible by q, i. e.
m ≡ 1 (mod q).
We obtain χ(m) = χ(1) = 1 and, hence, χi (1) = 1. Thus, (3.12) is
proved.
5) Let us show that
χi (mn) = χi (m)χi (n) for all m, n ∈ Z. (3.13)
Let m, n ∈ Z.
i) If (m, qi ) > 1, then (see (3.11))
χi (m) = 0, χi (mn) = 0.
Hence, (3.13) holds in this case.
ii) Similarly, (3.13) holds if (n, qi ) > 1.
iii) Let (m, qi ) = (n, qi ) = 1. We put
X X X
l1 = mAi + Aj , l2 = nAi + Aj , l3 = mnAi + Aj .
1≤j≤r 1≤j≤r 1≤j≤r
j6=i j6=i j6=i

We have
χi (m)χi (n) = χ(l1 )χ(l2 ) = χ(l1 l2 ),
χi (mn) = χ(l3 ).
14 ARTYOM RADOMSKII

Since
X
l3 − l1 l2 = mnAi + Aj − mnA2i −
1≤j≤r
j6=i
X X X 2
−mAi Aj − nAi Aj − Aj ,
1≤j≤r 1≤j≤r 1≤j≤r
j6=i j6=i j6=i

we have (see (3.8))

α
l3 − l1 l2 ≡ 0 (mod qj j ) for any 1 ≤ j ≤ r.
Hence, l3 − l1 l2 is divisible by q, i. e.
l3 ≡ l1 l2 (mod q).
We obtain χ(l3 ) = χ(l1 l2 ) and, hence, χi (mn) = χi (m)χi (n). Thus,
(3.13) is proved. We have proved that χi is a Dirichlet character modulo
qiαi , i = 1, . . . , r.
Now we prove that (3.7) holds. Let n ∈ Z. We put
X
ni = nAi + Aj , i = 1, . . . , r.
1≤j≤r
j6=i

We have
χ1 (n) · · · χr (n) = χ(n1 ) · · · χ(nr ) = χ(n1 · · · nr ).
From (3.8) we obtain
n1 · · · nr ≡ n (mod qsαs ) for any 1 ≤ s ≤ r.
Hence, (n1 · · · nr − n) is divisible by q, i. e.
n1 · · · nr ≡ n (mod q).
Hence, χ(n1 · · · nr ) = χ(n) and (3.7) is proved.
Now we prove the uniqueness. Suppose that
χ(n) = χ
e1 (n) · · · χ
er (n), (3.14)
where χ
ei is a Dirichlet character modulo qiαi ,
i = 1, . . . , r. Let 1 ≤ i ≤ r
and n ∈ Z. We have (see (3.8))
X
nAi + Aj ≡ 1 (mod qsαs ) for any 1 ≤ s ≤ r, s 6= i,
1≤j≤r
j6=i

and X
nAi + Aj ≡ n (mod qiαi ).
1≤j≤r
j6=i
Hence,
X
χ
es (nAi + Aj ) = 1 for any 1 ≤ s ≤ r, s 6= i ,
1≤j≤r
j6=i
 CONSECUTIVE PRIMES IN SHORT INTERVALS 15

and X
χ
ei (nAi + Aj ) = χ
ei (n).
1≤j≤r
j6=i
From (3.14) we obtain
X
χ(nAi + Aj ) = χ
ei (n).
1≤j≤r
j6=i

Hence (see (3.9)), χ ei (n) = χi (n). Since this equation holds for any
n ∈ Z, we have χ ei = χi , i = 1, . . . , r.
We see from (3.9) that if the character χ is real, then all characters
χi , i = 1, . . . , r, are real. We claim that if the character χ is primitive,
then all characters χi , i = 1, . . . , r, are primitive. Assume the converse:
there is 1 ≤ i ≤ r such that the character χi is imprimitive. Then
c(χi ) < qiαi . Since c(χi )|qiαi (see Lemma 3.5), we have
c(χi ) = qiβ , β < αi .
We put Y α
qe = qiβ qj j .
1≤j≤r
j6=i
Let us show that the character χ restricted by (n, q) = 1 has period
qe. Let m and n be integers such that (m, q) = (n, q) = 1 and m ≡ n
(mod qe). Let 1 ≤ j ≤ r, j 6= i. Since
α
m ≡ n (mod qj j ),
we have χj (m) = χj (n). Since (m, qiαi ) = (n, qiαi ) = 1,
m ≡ n (mod qiβ )
and χi restricted by (n, qiαi ) = 1 has period qiβ , we have χi (m) = χi (n).
We obtain
Y Y
χ(m) = χi (m) χj (m) = χi (n) χj (n) = χ(n).
1≤j≤r 1≤j≤r
j6=i j6=i

We have proved that χ restricted by (n, q) = 1 has period qe. But

then c(χ) ≤ qe < q. This contradicts the fact that the character χ is
primitive. Hence, all characters χi , i = 1, . . . , r, are primitive. Lemma
3.7 is proved.
Lemma 3.8. Let q be a positive integer such that there exists a real
primitive character χ modulo q. Then the number q is of the form 2α k,
where α ∈ {0, . . . , 3} and k ≥ 1 is an odd square-free integer.
Proof of Lemma 3.8. Modulo q = 1 there exists a real primitive
character; namely, χ(n) = 1 for all n ∈ Z. The number 1 is of the form
2α k; namely, α = 0 and k = 1.
16 ARTYOM RADOMSKII

Let q be an integer such that q > 1 and there exists a real primitive
character χ modulo q. Suppose that q = pr s, where p ≥ 3 is a prime
number, (p, s) = 1, r ≥ 2. We put qe = pr−1 s. We claim that the
character χ restricted by (n, q) = 1 has period qe. Let m and n be
integers such that (m, q) = (n, q) = 1 and m ≡ n (mod qe). We have
m = n + qet, t ∈ Z, and
pr−1  r−1 
X p
pr−1 pr−1 pr−1 r−1
m = (n + qet) =n + q t)i np −i =
(e
i=1
i
pr−1
X  r−1 
pr−1 i pr−1 −i p
=n + Ai t n , where Ai = q )i .
(e (3.15)
i=1
i
Let 2 ≤ i ≤ pr−1 . Then
 r−1   r−1 
p r−1 i r p
Ai = (p s) = p s p(i−1)r−i si−1 .
i i
It is clear that i − 1 ≥ 1. We claim that
(i − 1)r − i ≥ 0 (3.16)
or, that is equivalent,
i(r − 1) ≥ r.
In fact, since i ≥ 2 and r ≥ 2, we have
i(r − 1) ≥ 2(r − 1) ≥ r.
The inequality (3.16) is proved. Hence, Ai = pr sN, where N ∈ N.
Thus, for any 2 ≤ i ≤ pr−1 we have
Ai ≡ 0 (mod q).
We have
A1 = pr−1 (pr−1 s) = pr spr−2 .
Since r ≥ 2, we obtain
A1 ≡ 0 (mod q).
Hence (see (3.15)),
r−1 r−1
mp ≡ np (mod q).
Using properties of a character, we obtain
r−1 r−1
(χ(m))p = (χ(n))p .
Since (m, q) = (n, q) = 1 and the character χ is real, by Lemma 3.2
we have χ(m), χ(n) ∈ {−1, 1}. Since p ≥ 3 is a prime number and
r ≥ 2 is an integer, we have pr−1 is an odd positive integer. Therefore,
if χ(m) = 1, then χ(n) = 1; if χ(m) = −1, then χ(n) = −1. Thus,
χ(m) = χ(n). We have proved that the character χ restricted by
(n, q) = 1 has period qe. We obtain
c(χ) ≤ qe < q.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 17

This contradicts the fact that χ is a primitive character. Hence, the

number q is of the form 2α k, where α ≥ 0 is an integer and k ≥ 1 is an
odd square-free integer.
We claim that α ≤ 3. Assume the converse: α ≥ 4. Let
k = q1 · · · qr ,
where q1 < . . . < qr are odd primes. By Lemma 3.7, we have
χ(n) = χ1 (n)χ2 (n) · · · χr+1 (n), (3.17)
α
where χ1 is a real primitive character modulo 2 , χi is a real primitive
character modulo qi−1 , i = 2, . . . , r + 1 (if k = 1, then χ2 , . . . , χr+1
are omitted in (3.17)). It is well-known (see, for example, [7, Chapter
6]), if numbers ν and γ run independently through sets {0, 1} and
{0, . . . , 2α−2 − 1} respectively, then (−1)ν 5γ runs through a reduced
system of residues modulo 2α . Hence, for any n such that (n, 2) = 1
there are unique numbers ν(n) ∈ {0, 1} and γ(n) ∈ {0, . . . , 2α−2 − 1}
such that
n ≡ (−1)ν(n) 5γ(n) (mod 2α ). (3.18)
2 2
Since (−1) = 1, we have (χ1 (−1)) = 1. We obtain
χ1 (−1) = (−1)a , a ∈ {0, 1}.
It is well-known (see, for example, [7, Chapter 6]), the number 5 belongs
to 2α−2 (mod 2α ) and, in particular,
α−2
52 ≡ 1 (mod 2α ).
Hence,
α−2
(χ1 (5))2 = 1.
We obtain
 b 
χ1 (5) = exp 2πi α−2 , b ∈ {0, . . . , 2α−2 − 1}.
2
We see from (3.18) that if n is such that (n, 2) = 1, then
 bγ(n) 
χ1 (n) = (−1)aν(n) exp 2πi α−2 . (3.19)
2
We claim that (b, 2) = 1. Assume the converse: (b, 2) > 1. We show
that then χ1 restricted by (n, 2α ) = 1 has period 2α−1 . Let m and n be
integers such that (m, 2α ) = (n, 2α ) = 1 and m ≡ n (mod 2α−1 ). We
have
m ≡ (−1)ν(m) 5γ(m) (mod 2α ),
n ≡ (−1)ν(n) 5γ(n) (mod 2α ).
Since these congruences hold also modulo 2α−1 , we have
(−1)ν(m) 5γ(m) ≡ (−1)ν(n) 5γ(n) (mod 2α−1 ). (3.20)
Since α ≥ 4, we obtain
(−1)ν(m) 5γ(m) ≡ (−1)ν(n) 5γ(n) (mod 4).
18 ARTYOM RADOMSKII

It is clear that
(−1)ν(m) 5γ(m) ≡ (−1)ν(m) (mod 4),
(−1)ν(n) 5γ(n) ≡ (−1)ν(n) (mod 4).
Hence,
(−1)ν(m) ≡ (−1)ν(n) (mod 4).
If ν(m) = 0, then ν(n) = 0; if ν(m) = 1, then ν(n) = 1. Thus,
ν(m) = ν(n). (3.21)
We obtain (see (3.20))
5γ(m) ≡ 5γ(n) (mod 2α−1 ).
Suppose, for the sake of definiteness, that γ(m) ≥ γ(n). We have
5γ(n) (5γ(m)−γ(n) − 1) ≡ 0 (mod 2α−1 ).
Since (5γ(n) , 2α−1 ) = 1, we obtain
5γ(m)−γ(n) − 1 ≡ 0 (mod 2α−1 ).
Hence,
5γ(m)−γ(n) ≡ 1 (mod 2α−1 ).
Since 5 belongs to 2α−3 (mod 2α−1 ), we have (see [7, Chapter 6])
γ(m) − γ(n) ≡ 0 (mod 2α−3 ).
We obtain
γ(m) = γ(n) + 2α−3 t, (3.22)
where t ≥ 0 is an integer. Since (b, 2) > 1, we have
b = 2eb, (3.23)
where eb ≥ 0 is an integer. We obtain (see (3.19), (3.21), (3.22) and
(3.23))
 ebγ(m) 
aν(m)
χ1 (m) = (−1) exp 2πi α−3 =
2
 eb(γ(n) + 2α−3 t) 
= (−1)aν(n) exp 2πi =
2α−3
 ebγ(n) 
= (−1) aν(n)
exp 2πi α−3 exp(2πiebt) =
2
 ebγ(n) 
= (−1)aν(n) exp 2πi α−3 = χ1 (n).
2
We have proved that χ1 restricted by (n, 2α ) = 1 has period 2α−1 .
Hence,
c(χ1 ) ≤ 2α−1 < 2α .
This contradicts the fact that χ1 is a primitive character. Hence,
(b, 2) = 1.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 19

For n = 5 we have ν(5) = 0 and γ(5) = 1. Hence (see (3.19)),

 b   b 
χ1 (5) = exp 2πi α−2 = exp πi α−3 .
2 2
Since α ≥ 4 and (b, 2) = 1, we have
Im(χ1 (5)) 6= 0.
This contradicts the fact that χ1 is a real character. Hence, 0 ≤ α ≤ 3.
Lemma 3.8 is proved.
Lemma 3.9. Let q1 and q2 be positive integers with q1 6= q2 , χ1 be a
primitive character modulo q1 , χ2 be a primitive character modulo q2 .
Then χ1 6= χ2 .
Proof of Lemma 3.9. Assume the converse: χ1 = χ2 . Let m and
n be integers such that (m, q1 ) = (n, q1 ) = 1 and m ≡ n (mod q2 ).
Then
χ1 (m) = χ2 (m) = χ2 (n) = χ1 (n).
Hence, χ1 restricted by (n, q1 ) = 1 has period q2 . Hence, c(χ1 ) ≤ q2 .
Since χ1 is a primitive character modulo q1 , we have c(χ1 ) = q1 . Hence,
q1 ≤ q2 . Similarly, it can be proved that q2 ≤ q1 . Hence, q1 = q2 . We
obtain a contradiction. Hence, χ1 6= χ2 . Lemma 3.9 is proved.
4. Lemmas on ψ(x, χ)
In this section we give some lemmas on ψ(x, χ). Most of these lem-
mas are well known. The proof of Lemma 4.6 is based on ideas of
Maynard (see the proof of Theorem 3.2 in [5]). The proof of Lemma
4.9 follows a standard proof of the Bombieri–Vinogradov Theorem (see,
for example, [1, Chapter 28]).
Lemma 4.1. Let Q and W be integers with Q ≥ 2 and (W, Q) = 1.
Let u be a real number with u ≥ 2. Then
u 1 X
ψ(u; Q, W ) − = χ(W )ψ ′ (u, χ).
ϕ(Q) ϕ(Q) χ∈X
Q

Here the line denotes a complex conjugation.

Proof of Lemma 4.1. We define
(
1, if n ≡ W (mod Q);
IQ,W (n) =
0, otherwise.
Since (see, for example, [1, Chapter 4])
1 X
χ(W )χ(n) = IQ,W (n),
ϕ(Q) χ∈X
Q
20 ARTYOM RADOMSKII

we have
X X
ψ(u; Q, W ) = Λ(n) = Λ(n)IQ,W (n) =
n≤u n≤u
n≡W (mod Q)
X 1 X
= Λ(n) χ(W )χ(n) =
n≤u
ϕ(Q) χ∈X
Q

1 X X 
= χ(W ) Λ(n)χ(n) =
ϕ(Q) χ∈X n≤u
Q

1 X
= χ(W )ψ(u, χ).
ϕ(Q) χ∈X
Q

Let χ0 be the principal character modulo Q. Since (W, Q) = 1, we have

χ0 (W ) = 1. We have
X
χ(W )Eχ0 (χ)u = χ0 (W )u = u.
χ∈XQ

Hence,
u 1 X

ψ(u; Q, W ) − = χ(W ) ψ(u, χ) − Eχ0 (χ)u =
ϕ(Q) ϕ(Q) χ∈X
Q

1 X
= χ(W )ψ ′ (u, χ).
ϕ(Q) χ∈X
Q

Lemma 4.1 is proved.

Lemma 4.2 (see, for example, [1, Chapter 14]). There is a positive
absolute constant a > 0 such that the following holds. If χ is a complex
character modulo q, then L(s, χ) has no zero in the region defined by
(here s = σ + it, σ = Re(s), t = Im(s))
(
1 − a/ ln(q|t|), if |t| ≥ 1,
Ωq : σ>
1 − a/ ln q, if |t| < 1.
If χ is a real nonprincipal character modulo q, the only possible zero of
L(s, χ) in this region is a single (simple) real zero. Furthermore, for
at most one of the real nonprincipal characters χ (mod q) can L(s, χ)
have a zero in the region Ωq .
Remark. It is easy to see that the constant a can be replaced by
any constant a∗ such that 0 < a∗ < a.
Lemma 4.3 (see [1, Chapter 20]). Let χ be a nonprincipal character
modulo q and 2 ≤ T ≤ u. Then
uβ1
ψ(u, χ) = − + R4 (u, T ),
β1
 CONSECUTIVE PRIMES IN SHORT INTERVALS 21

where


|R4 (u, T )| ≤ C u ln2 (qu)exp −a ln u/ ln(qT ) +uT −1 ln2 (qu)+u1/4 ln u .
Here C > 0 is an absolute constant, a > 0 is the absolute constant
in Lemma 4.2. The term −uβ1 /β1 is to be omitted unless χ is a real
character for which L(s, χ) has a zero β1 (which is necessary unique,
real and simple) satisfying
β1 > 1 − a/ ln q.
Lemma 4.4 (Theorem of Page (see, for example, [1, Chapter 14])).
There are absolute constants a1 > 0 and a′1 > 0 such that the following
holds. Let z be a real number with z ≥ 3. Then there is at most one
real primitive χ to a modulus q0 , 3 ≤ q0 ≤ z, for which L(s, χ) has a
real zero β satisfying
β > 1 − a1 / ln z.
If such a character χ exists, then
a′ (ln z)2
q0 ≥ 1 .
(ln ln z)4
Such a modulus q0 is said to be an exceptional modulus in the interval
[3, z].
Lemma 4.5. Let z be a real number with z ≥ 3. If an exceptional
modulus q0 in the interval [3, z] exists, then the number q0 is of the form
2α k, where α ∈ {0, . . . , 3} and k ≥ 1 is an odd square-free integer.
Proof of Lemma 4.5. Let an exceptional modulus q0 in the in-
terval [3, z] exist. In particular, this means that there exists a real
primitive character χ modulo q0 . By Lemma 3.8, the number q0 is of
the form 2α k where α ∈ {0, . . . , 3} and k ≥ 1 is an odd square-free
integer. Lemma 4.5 is proved.
Lemma 4.6. There are positive absolute constants c0 , c1 , γ0 and C
such that the following holds. Let x be a real number√with x ≥ c0 , q0
be an exceptional modulus in the √ interval [3, exp(2c1 ln x)], Q be an
integer with 3 ≤ Q ≤ exp(2c1 ln x) and Q 6= q0 (the last inequality
should be interpreted as follows: if q0 exists, then Q 6= q0 ; if q0 does not
exist, then Q is any integer in the mentioned interval ), χ be a primitive
character modulo Q. Then
√
max √ |ψ(u, χ)| ≤ Cx exp(−3c1 ln x).
2≤u≤x1+γ0 / ln x

Proof of Lemma 4.6. We choose c1 and γ0 later. The number c0

depends on c1 and γ0 ; the number c0 is large enough and x ≥ c0 (c1 , γ0 ).
We put √
z = exp(2c1 ln x).
22 ARTYOM RADOMSKII

We have
z ≥ 3,
if the number c0 (c1 , γ0) is chosen large enough. By Lemma 4.4, there
is at most one real primitive χ to a modulus q0 , 3 ≤ q0 ≤ z, for which
L(s, χ) has a real zero β satisfying
√
β > 1 − a1 / ln z = 1 − a1 /(2c1 ln x). (4.1)
If such a character χ exists, then
√
a′1 (ln z)2 a′1 (2c1 ln x)2 a′1 c21 ln x
q0 ≥ =
4 ≥ , (4.2)
(ln ln z)4 (1/2) ln ln x + ln(2c1 ) (ln ln x)4
if c0 (c1 ,√
γ0) is chosen large enough. Let Q be an integer with 3 ≤ Q ≤
exp(2c1 ln x) and Q 6= q0 , let χ be a primitive character modulo Q.
Since Q > 1, we see that χ is a nonprincipal character. By Lemma 4.3,
if 2 ≤ T ≤ u, then
uβ1
ψ(u, χ) = − + R4 (u, T ), (4.3)
β1
where


|R4 (u, T )| ≤ C u ln2 (Qu)exp −a ln u/ ln(QT ) + uT −1 ln2 (Qu)+

+ u1/4 ln u = C(∆1 + ∆2 + ∆3 ). (4.4)
The term −uβ1 /β1 is to be omitted unless χ is a real character for
which L(s, χ) has a zero β1 (which is necessary unique, real and simple)
satisfying
β1 > 1 − a/ ln Q.
Let √
2 ≤ u ≤ x1+γ0 / ln x .
Let u ≥ c2 (c1 ), where c2 (c1 ) > 0 is a number depending only on c1 . We
choose √
T = exp(4c1 ln u). (4.5)
Then
2 ≤ T ≤ u,
if c2 (c1 ) is chosen large enough.
I) Now we estimate


∆1 = u ln2 (Qu)exp −a ln u/ ln(QT ) .
If c0 (c1 , γ0 ) is chosen large enough, then
γ0
1+ √ ≤ 2. (4.6)
ln x
 CONSECUTIVE PRIMES IN SHORT INTERVALS 23

Hence,
 γ0 
ln u ≤ 1 + √ ln x ≤ 2 ln x, (4.7)
ln x
√ √ √
QT ≤ exp(2c1 ln x + 4c1 ln u) ≤ exp(10c1 ln x),
√
ln(QT ) ≤ 10c1 ln x,
a ln u a ln u
− ≤− √ .
ln(QT ) 10c1 ln x
If c0 (c1 , γ0 ) is chosen large enough, then
√
ln Q ≤ 2c1 ln x ≤ ln x.
Hence,
ln2 (Qu) ≤ 2(ln2 Q + ln2 u) ≤ 10 ln2 x = 10 exp(2 ln ln x). (4.8)
We have  
a ln u
∆1 ≤ 10u exp − √ + 2 ln ln x .
10c1 ln x
Let us consider two cases. √
1) x1/4 ≤ u ≤ x1+γ0 / ln x . Then
ln x  γ0 
≤ ln u ≤ 1 + √ ln x ≤ 2 ln x.
4 ln x
If r
a
0 < c1 ≤ ,
160
then
a
− ≤ −4c1 .
40c1
Hence,
√
a ln u (a/4) ln x a ln x √
− √ ≤− √ =− ≤ −4c1 ln x,
10c1 ln x 10c1 ln x 40c1
a ln u √ √
− √ + 2 ln ln x ≤ −4c1 ln x + 2 ln ln x ≤ −3.5c1 ln x,
10c1 ln x
if c0 (c1 , γ0 ) is chosen large enough. If
0 < γ0 ≤ 0.5c1 ,
then
√ √
∆1 ≤ 10x1+γ0 / ln x
exp(−3.5c1 ln x) =
√ √ √
= 10x exp(−3.5c1 ln x + γ0 ln x) ≤ 10x exp(−3c1 ln x).
24 ARTYOM RADOMSKII

4
2) c2 (c1 ) ≤ u < x1/4 (we may assume that c0 (c1 , γ0 ) > c2 (c1 ) and
c2 (c1 ) ≥ 10). We have
 a ln u 
∆1 ≤ 10u exp − √ + 2 ln ln x ≤
10c1 ln x
≤ 10u exp(2 ln ln x) ≤ 10x1/4 exp(2 ln ln x) ≤
√
≤ 10x exp(−3c1 ln x),
if c0 (c1 , γ0 ) is chosen p
large enough.
Thus, if √0 < c1 < a/160, 0 < γ0 ≤ 0.5c1 , x ≥ c0 (c1 , γ0 ), c2 (c1 ) ≤
u ≤ x1+γ0 / ln x , then
√
∆1 ≤ 10x exp(−3c1 ln x).
II) Now we estimate
∆2 = uT −1 ln2 (Qu).
From (4.5) and (4.8) we obtain
√
∆2 ≤ 10u exp(−4c1 ln u + 2 ln ln x).
Let us consider two √cases.
1) x0.9 ≤ u ≤ x1+γ0 / ln x . Then
 γ0 
0.9 ln x ≤ ln u ≤ 1 + √ ln x ≤ 2 ln x,
ln x
√ √ √
−4c1 ln u ≤ −4c1 0.9 ln x < −3.7c1 ln x.
Since 0 < γ0 ≤ 0.5c1 , we have
√ √
∆2 ≤ 10x1+γ0 / ln x
exp(−3.7c1 ln x + 2 ln ln x) =
√ √
= 10x exp(−3.7c1 ln x + 2 ln ln x + γ0 ln x) ≤
√
≤ 10x exp(−3.2c1 ln x + 2 ln ln x) ≤
√
≤ 10x exp(−3c1 ln x),
if c0 (c1 , γ0 ) is chosen large enough.
2) c2 (c1 ) ≤ u < x0.9 . Then
√
∆2 ≤ 10u exp(−4c1 ln u + 2 ln ln x) ≤
≤ 10u exp(2 ln ln x) ≤ 10x0.9 exp(2 ln ln x) ≤
√
≤ 10x exp(−3c1 ln x),
if c0 (c1 , γ0 ) is chosen p
large enough.
Thus, if √0 < c1 < a/160, 0 < γ0 ≤ 0.5c1 , x ≥ c0 (c1 , γ0 ), c2 (c1 ) ≤
u ≤ x1+γ0 / ln x , then
√
∆2 ≤ 10x exp(−3c1 ln x).
 CONSECUTIVE PRIMES IN SHORT INTERVALS 25

III) Now we estimate

∆3 = u1/4 ln u.
Since (see (4.6) and (4.7))
ln u ≤ 2 ln x,
√
u1/4 ≤ x(1+γ0 / ln x)/4
≤ x1/2 ,
we have √
∆3 ≤ 2x1/2 ln x ≤ x exp(−3c1 ln x),
if c0 (c1 , γ0 ) is chosen large enough. p
Finally, we obtain (see (4.4)): if√ 0 < c1 < a/160, 0 < γ0 ≤ 0.5c1 ,
x ≥ c0 (c1 , γ0 ), c2 (c1 ) ≤ u ≤ x1+γ0 / ln x , then
√
|R4 (u, T )| ≤ 21Cx exp(−3c1 ln x), (4.9)
where C > 0 is an absolute constant.
IV) Now we estimate (see (4.3))
 β 
 u 1
∆4 = − .
β1 
If χ is not a real character for which L(s, χ) has a zero β1 (which is
necessary unique, real and simple) satisfying
β1 > 1 − a/ ln Q,
then the term −uβ1 /β1 is to be omitted, and there is nothing to esti-
mate. Let χ be such a character. Then χ is a real primitive character
modulo Q. Since Q 6= q0 , we have (see Lemma 3.9 and (4.1))
√
β1 ≤ 1 − a1 / ln z = 1 − a1 /(2c1 ln x).
Hence,
√ √

|uβ1 | = uβ1 ≤ u1−a1 /(2c1 ln x)
= u exp −(a1 ln u)/(2c1 ln x) .
By Remark made below Lemma 4.2, we may assume that 0 < a < 1/2.
Since Q ≥ 3, we have
1 1
β1 > 1 − a/ ln Q > 1 − > .
2 ln 3 2
Hence,
1
0< ≤ 2.
β1
We obtain √

∆4 ≤ 2u exp −(a1 ln u)/(2c1 ln x) . (4.10)
Let us consider two cases.
√
1) x1/2 ≤ u ≤ x1+γ0 / ln x . We have (see (4.6))
ln x  γ0 
≤ ln u ≤ 1 + √ ln x ≤ 2 ln x.
2 ln x
26 ARTYOM RADOMSKII

If we take r
min(a, a1 )
0 < c1 < ,
160
then
a1
− ≤ −3.5c1 .
4c1
Hence,
√
a1 ln u (a1 /2) ln x a1 ln x √
− √ ≤− √ =− ≤ −3.5c1 ln x.
2c1 ln x 2c1 ln x 4c1
Since 0 < γ0 ≤ 0.5c1 , we obtain (see (4.10))
√ √
∆4 ≤ 2x1+γ0 / ln x exp(−3.5c1 ln x) =
√ √
= 2x exp(−3.5c1 ln x + γ0 ln x) ≤
√
≤ 2x exp(−3c1 ln x).
2) c2 (c1 ) ≤ u < x1/2 . Then (see (4.10))
√
∆4 ≤ 2u ≤ 2x1/2 ≤ 2x exp(−3c1 ln x),
if c0 (c1 , γ0 ) is chosen large enough. p Putting I–IV together, we obtain
(see (4.3), (4.9)): if 0 < c1 < √ min(a, a1 )/160, 0 < γ0 ≤ 0.5c1 ,
x ≥ c0 (c1 , γ0 ), c2 (c1 ) ≤ u ≤ x1+γ0 / ln x , then
√
|ψ(u, χ)| ≤ (21C + 2)x exp(−3c1 ln x),
where C > 0 is an absolute constant.
There is a number d(c1 ) > 0, depending only on c1 , such that
√
t exp(−3c1 ln t) ≥ 1, if t ≥ d(c1 ).
We may assume that c0 (c1 , γ0) > d(c1 ). Hence, if 2 ≤ u < c2 (c1 ), then
(see (2.1))
X  X
 
|ψ(u, χ)| =  Λ(n)χ(n) ≤ Λ(n) = ψ(u) ≤
n≤u n≤u
√
≤ b6 u ≤ b6 c2 (c1 ) ≤ b6 c2 (c1 )x exp(−3c1 ln x).
p
Thus, if 0 < c1 < min(a, a1 )/160, 0 < γ0 ≤ 0.5c1 , x ≥ c0 (c1 , γ0 ),
then

√
max √ |ψ(u, χ)| ≤ 21C + 2 + b6 c2 (c1 ) x exp(−3c1 ln x),
2≤u≤x1+γ0 / ln x

where C > 0 is an absolute constant. We take

p p
min(a, a1 ) min(a, a1 )
c1 = , γ0 = 0.5c1 = .
16 32
Since a > 0 and a1 > 0 are absolute constants, we see that c1 , γ0 ,
c0 (c1 , γ0) and c2 (c1 ) are positive absolute constants. Lemma 4.6 is
proved.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 27

Lemma 4.7. Let u be a real number with u ≥ 2, Q be an integer with

Q ≥ 2, χ ∈ XQ , χ1 be a primitive character modulo q1 inducing χ.
Then
|ψ ′ (u, χ) − ψ ′ (u, χ1 )| ≤ ln2 (Qu). (4.11)
Proof of Lemma 4.7. From Lemma 3.6 and the definition of the
inducing character, given below Lemma 3.6, we have q1 = c(χ) and,
hence, q1 |Q (see Lemma 3.5). Let us prove (4.11). We consider two
cases.
1) χ is a nonprincipal character modulo Q. We claim that then χ1
is a nonprincipal character modulo q1 . In fact, assume the converse.
Then (
1, if (n, q1 ) = 1;
χ1 (n) =
0, if (n, q1 ) > 1.
Since q1 |Q, we see that if (n, Q) = 1, then (n, q1 ) = 1. We obtain (see
(3.2)) (
1, if (n, Q) = 1;
χ(n) =
0, if (n, Q) > 1.
Hence, χ is the principal character modulo Q. We obtain a contradic-
tion. Thus, χ1 is a nonprincipal character modulo q1 . We have (see
(3.2))
X
ψ ′ (u, χ) = ψ(u, χ) = Λ(n)χ(n) =
n≤u
X X
= Λ(n)χ(n) = Λ(n)χ1 (n);
n≤u n≤u
(n,Q)=1 (n,Q)=1
X X
ψ ′ (u, χ1 ) = ψ(u, χ1) = Λ(n)χ1 (n) = Λ(n)χ1 (n) =
n≤u n≤u
(n,q1 )=1
X X
= Λ(n)χ1 (n) + Λ(n)χ1 (n) =
n≤u n≤u
(n,q1 )=1 (n,q1 )=1
(n,Q)=1 (n,Q)>1
X X
= Λ(n)χ1 (n) + Λ(n)χ1 (n).
n≤u n≤u
(n,Q)=1 (n,q1 )=1
(n,Q)>1
28 ARTYOM RADOMSKII

Hence,
X X
ψ ′ (u, χ1 ) − ψ ′ (u, χ) = Λ(n)χ1 (n) = χ1 (pm ) ln p =
n≤u pm ≤u
(n,q1 )=1 p|Q
(n,Q)>1 (p,q1 )=1
X X
= ln p χ1 (pm ).
p|Q 1≤m≤logp u
(p,q1 )=1

For any prime p we have

 X  ln u ln u
 m 
 χ1 (p ) ≤ logp u = ≤ ≤ 2 ln u.
1≤m≤log u
ln p ln 2
p

We obtain
X X
|ψ ′ (u, χ1 ) − ψ ′ (u, χ)| ≤ 2 ln u ln p ≤ 2 ln u ln p ≤
p|Q p|Q
(p,q1 )=1

≤ 2 ln u ln Q ≤ (ln u + ln Q)2 = ln2 (Qu).

2) χ is the principal character modulo Q. We have
ψ ′ (u, χ) = ψ(u, χ) − u;
(
1, if (n, Q) = 1;
χ(n) =
0, if (n, Q) > 1.
In this case the least period of χ, restricted by (n, Q) = 1, is 1. There-
fore q1 = 1, and χ1 is a primitive character modulo 1, i. e. χ1 (n) = 1
for any integer n. Hence, χ1 is the principal character modulo 1. We
have
ψ ′ (u, χ1) = ψ(u, χ1 ) − u;
ψ ′ (u, χ1) − ψ ′ (u, χ) = ψ(u, χ1) − ψ(u, χ).
Since
X X
ψ(u, χ) = Λ(n)χ(n) = Λ(n),
n≤u n≤u
(n,Q)=1
X X X X
ψ(u, χ1 ) = Λ(n)χ1 (n) = Λ(n) = Λ(n) + Λ(n),
n≤u n≤u n≤u n≤u
(n,Q)=1 (n,Q)>1

we have
X X X X
ψ ′ (u, χ1 ) − ψ ′ (u, χ) = Λ(n) = ln p = ln p 1.
n≤u pm ≤u p|Q 1≤m≤logp u
(n,Q)>1 p|Q
 CONSECUTIVE PRIMES IN SHORT INTERVALS 29

For any prime p we have

X ln u ln u
1 ≤ logp u = ≤ ≤ 2 ln u.
1≤m≤log u
ln p ln 2
p

Hence,
X
|ψ ′ (u, χ1 ) − ψ ′ (u, χ)| ≤ 2 ln u ln p ≤ 2 ln u ln Q ≤ ln2 (Qu).
p|Q

Lemma 4.7 is proved.

Lemma 4.8 (see [1, Chapter 28]). Let Q1 , Q2 and t be real numbers
with 1 ≤ Q1 < Q2 , t ≥ 2. Then
X 1 X  t 
max |ψ(u, χ)| ≤ C ln4 (tQ2 ) +t5/6 ln Q2 +t1/2 Q2 ,
Q <Q≤Q
ϕ(Q) χ∈X ∗ 2≤u≤t Q1
1 2 Q

where C > 0 is an absolute constant.

Lemma 4.9. Let ε and δ be real numbers with 0 < ε < 1 and 0 <
δ < 1/2. Then there is a number c(ε, δ) > 0, depending only on ε
and δ, such that if x ∈ R and q ∈ Z are such that x ≥ c(ε, δ) and
1 ≤ q ≤ (ln x)1−ε , then there is a positive integer B such that
√ B
1 ≤ B ≤ exp(c1 ln x), 1 ≤ ≤ 2, (B, q) = 1
ϕ(B)
and 
X  u  √
max √ max ψ(u; Q, W )−  ≤ c2 x exp(−c3 ln x).
1/2−δ
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
1≤Q≤x (W,Q)=1
(Q,B)=1

Here c1 , γ, c2 and c3 are positive absolute constants.

Proof of Lemma 4.9. Let c0 , c1 , γ0 and C be the positive absolute
constants in Lemma 4.6. We choose γ and c(ε, δ) = c(ε, δ, γ) later; γ is
small enough, and c(ε, δ, γ) is large enough; 0 < γ ≤ γ0 , c(ε, δ, γ) ≥ c0
and x ≥ c(ε,
√ δ, γ). Let q0 be the exceptional modulus in the interval
[3, exp(2c1 ln x)]. If q0 does not exist, then we take B = 1. If q0 exists,
then (see (4.2))
a′ c2 ln x c4 ln x
q0 ≥ 1 1 4 = ,
(ln ln x) (ln ln x)4
where c4 > 0 is an absolute constant. We have q0 ≥ 24, if c(ε, δ, γ) is
chosen large enough. By Lemma 4.5, the number q0 is of the form 2α k,
where α ∈ {0, . . . , 3} and k ≥ 3 is an odd square-free integer. We put
q0 q0 c4 ln x
M1 = α ≥ ≥ .
2 8 8(ln ln x)4
Let τ = (M1 , q), M2 = M1 /τ . Then (M2 , q) = 1. Since
τ ≤ q ≤ (ln x)1−ε ,
30 ARTYOM RADOMSKII

we have
M1 M1 c4 ln x c4 (ln x)ε
M2 = ≥ ≥ = .
τ (ln x)1−ε 8(ln ln x)4 (ln x)1−ε 8(ln ln x)4
The number M2 ≥ 3, if c(ε, δ, γ) is chosen large enough. Hence, M2 ≥ 3
is an odd square-free integer. Furthermore, we have (M2 , q) = 1 and
M2 divides q0 . Let B to be the largest prime divisor of M2 . Hence,
B ≥ 3 is a prime number and B divides q0 . We have (see Lemma 2.5)
B B 1 1 3
= = ≤ = .
ϕ(B) B(1 − 1/B) 1 − 1/B 1 − 1/3 2
√
Thus, 1 ≤ B ≤ exp(2c1 ln x) is an integer, (B, q) = 1,
B
1≤ ≤2
ϕ(B)
and B ≥ 3 is a prime divisor of q0 , if q0 exists. √
Let u be a real number with 2 ≤ u ≤ x1+γ/ ln x , let Q and W be
integers with 2 ≤ Q ≤ x1/2−δ , (Q, B) = 1 and (W, Q) = 1. By Lemma
4.1, we have
u 1 X
ψ(u; Q, W ) − = χ(W )ψ ′ (u, χ).
ϕ(Q) ϕ(Q) χ∈X
Q

Hence, 
 u  1 X ′
ψ(u; Q, W ) − ≤ |ψ (u, χ)|.
ϕ(Q) ϕ(Q) χ∈X
Q

Since the right-hand side of this inequality does not depend on W , we

have 
 u  1 X ′
max ψ(u; Q, W ) − ≤ |ψ (u, χ)|.
W ∈Z: ϕ(Q) ϕ(Q) χ∈X
(W,Q)=1 Q

Let χ ∈ XQ , let χ1 be a primitive character modulo q1 inducing χ.

From Lemma 3.6 and the definition of the inducing character, given
below Lemma 3.6, we have q1 = c(χ), and hence q1 |Q (see Lemma 3.5).
Applying Lemma 4.7 we have
|ψ ′ (u, χ)| ≤ |ψ ′ (u, χ1)| + ln2 (Qu).
Since #XQ = ϕ(Q), we obtain
 u  1 X


max ψ(u; Q, W ) − ≤ |ψ ′ (u, χ1 )| + ln2 (Qu) =
W ∈Z: ϕ(Q) ϕ(Q) χ∈X
(W,Q)=1 Q

1 X ′
= ln2 (Qu) + |ψ (u, χ1 )|.
ϕ(Q) χ∈X
Q

We have
γ
1+ √ ≤ 2, (4.12)
ln x
 CONSECUTIVE PRIMES IN SHORT INTERVALS 31

if c(ε, δ, γ) is chosen large enough. Hence,

 γ 
0 < ln u ≤ 1 + √ ln x ≤ 2 ln x,
ln x
ln2 u ≤ 4 ln2 x,
1 
0 < ln Q ≤ − δ ln x ≤ ln x,
2
ln2 Q ≤ ln2 x,
ln2 (Qu) ≤ 2(ln2 Q + ln2 u) ≤ 10 ln2 x.
We obtain 
 u 
max √ max ψ(u; Q, W ) − ≤
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
(W,Q)=1
1 X
≤ 10 ln2 x + max √ |ψ ′ (u, χ1 )|.
ϕ(Q) χ∈X 2≤u≤x1+γ/ ln x
Q

Hence,
X X
S= AQ = A1 + AQ ≤ A1 +
1≤Q≤x1/2−δ 2≤Q≤x1/2−δ
(Q,B)=1 (Q,B)=1
X  1 X 
2 ′
+ 10 ln x + max √ |ψ (u, χ1 )| ≤
ϕ(Q) χ∈X 2≤u≤x1+γ/ ln x
2≤Q≤x1/2−δ Q
(Q,B)=1
X X 1
≤ 10x1/2−δ ln2 x + A1 + max √ |ψ ′ (u, χ1 )| =
ϕ(Q) 2≤u≤x1+γ/ ln x
2≤Q≤x1/2−δ χ∈XQ
(Q,B)=1

= 10x1/2−δ ln2 x + A1 + S ′ , (4.13)

where 
 u 
AQ := max √ max ψ(u; Q, W ) − .
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
(W,Q)=1

Let us estimate the sum S ′ . Let Q be an integer with 2 ≤ Q ≤ x1/2−δ

and (Q, B) = 1, let χ ∈ XQ , let χ1 be the primitive character modulo
q1 inducing χ. Since q1 |Q, we have 1 ≤ q1 ≤ x1/2−δ and (q1 , B) = 1.
Hence,
X X 1
S′ = max √ |ψ ′ (u, χ1)| ≤
1/2−δ χ∈X
ϕ(Q) 2≤u≤x1+γ/ ln x
2≤Q≤x Q
(Q,B)=1
X X  X 1 
≤ max √ |ψ ′ (u, χ1)| .
∗ 2≤u≤x
1+γ/ ln x ϕ(mq1 )
1≤q1 ≤x1/2−δ χ1 ∈Xq1 1≤m≤x1/2−δ /q1
(q1 ,B)=1
32 ARTYOM RADOMSKII

Applying Lemmas 2.6 and 2.10 we have

X 1 1 X 1
≤ ≤
1/2−δ
ϕ(mq1 ) ϕ(q1 ) ϕ(m)
1≤m≤x /q1 1≤m≤x1/2−δ /q1
1 X 1 1
≤ ≤ C ln x,
ϕ(q1 ) ϕ(m) ϕ(q1 )
1≤m≤x1/2

where C > 0 is an absolute constant. We obtain

X 1 X
S ′ ≤ C ln x max √ |ψ ′ (u, χ1)|.
1/2−δ
ϕ(q1 ) χ ∈X ∗ 2≤u≤x1+γ/ ln x
1≤q1 ≤x 1 q1
(q1 ,B)=1

Replacing q1 by Q and χ1 by χ, we have

X 1 X
S ′ ≤ C ln x max √ |ψ ′ (u, χ)| =
ϕ(Q) ∗ 2≤u≤x
1+γ/ ln x
1/2−δ
1≤Q≤x χ∈XQ
(Q,B)=1
 X X X 
= C ln x RQ + RQ + RQ =
√ √
1≤Q≤ln x ln x<Q≤exp(c1 ln x) exp(c1 ln x)<Q≤x1/2−δ
(Q,B)=1 (Q,B)=1 (Q,B)=1


= C ln x S1′ + S2′ + S3′ , (4.14)
where
1 X
RQ := max √ |ψ ′ (u, χ)|,
ϕ(Q) χ∈X ∗ 2≤u≤x1+γ/ ln x
Q

c1 > 0 is the absolute constant in Lemma 4.6.

I) Now we estimate S1′ . We have
X X
S1′ = RQ ≤ R1 + RQ = R1 + S4′ . (4.15)
1≤Q≤ln x 2≤Q≤ln x
(Q,B)=1

1) Now we estimate R1 . Since #X1∗ = 1, we have

R1 = max √ |ψ ′ (u, χ)|,
2≤u≤x1+γ/ ln x

where χ ∈ X1∗ ,
i. e. χ(n) = 1 for any n ∈ Z. Since χ is the principal
character modulo 1, we have
ψ ′ (u, χ) = ψ(u, χ) − u.
We have
X X
ψ(u, χ) = Λ(n)χ(n) = Λ(n) = ψ(u),
n≤u n≤u
′
ψ (u, χ) = ψ(u) − u.
It is well-known that (see, for example, [1, Chapter 18])
√
|ψ(u) − u| ≤ Cu exp(−c ln u), u ≥ 2, (4.16)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 33

where C > 0 and c > 0 are absolute constants.

Let us consider two √ cases.
i) x1/4 ≤ u ≤ x1+γ/ ln x (we may assume that c(ε, δ, γ) > 16). We have
(see (4.12))
1  γ 
ln x ≤ ln u ≤ 1 + √ ln x ≤ 2 ln x,
4 ln x
√ c√
−c ln u ≤ − ln x.
2
Hence,
√ √ √

|ψ ′ (u, χ)| ≤ Cu exp(−c ln u) ≤ Cx1+γ/ ln x exp −(c/2) ln x =
√
√

= Cx exp (γ − c/2) ln x ≤ Cx exp (−c/4) ln x ,
if 0 < γ ≤ c/4.
ii) 2 ≤ u < x1/4 . Then
√
|ψ ′(u, χ)| ≤ Cu exp(−c ln u) ≤ Cu ≤
√

≤ Cx1/4 ≤ Cx exp (−c/4) ln x ,
if c(ε, δ, γ) is chosen large enough.
We obtain
√

R1 = max √ |ψ ′ (u, χ)| ≤ Cx exp (−c/4) ln x . (4.17)
2≤u≤x1+γ/ ln x

2) Now we estimate
X 1 X
S4′ = max √ |ψ ′ (u, χ)|. (4.18)
2≤Q≤ln x
ϕ(Q) χ∈X ∗ 2≤u≤x1+γ/ ln x
Q

Let Q be an integer with 2 ≤ Q ≤ ln x, and let χ ∈ XQ∗ . Then χ is a

nonprincipal character modulo Q, and hence
ψ ′ (u, χ) = ψ(u, χ).
Let us consider two√cases.
i) x1/4 ≤ u ≤ x1+γ/ ln x . Then (see (4.12))
1  γ 
ln x ≤ ln u ≤ 1 + √ ln x ≤ 2 ln x.
4 ln x
We may assume that c(ε, δ, γ) ≥ e16 . Hence, ln u ≥ (ln x)/4 ≥ 4. We
have
2 ≤ Q ≤ ln x ≤ 4 ln u ≤ ln2 u.
Therefore (see, for example, [1, Chapter 22])
√
|ψ(u, χ)| ≤ Cu exp(−c(2) ln u),
where C > 0 and c(2) > 0 are absolute constants. We have
√ c(2) √
−c(2) ln u ≤ − ln x,
2
34 ARTYOM RADOMSKII
√

|ψ(u, χ)| ≤ Cu exp −(c(2)/2) ln x ≤
√ √

≤ Cx1+γ/ ln x exp −(c(2)/2) ln x =
√

= Cx exp (γ − c(2)/2) ln x ≤
√

≤ Cx exp −(c(2)/4) ln x ,
if 0 < γ ≤ c(2)/4.
ii) 2 ≤ u < x1/4 . We have (see (2.1))
X
ψ(u, χ) = Λ(n)χ(n),
n≤u
X
|ψ(u, χ)| ≤ Λ(n) = ψ(u) ≤ b6 u ≤ b6 x1/4 ≤
n≤u
√

≤ Cx exp −(c(2)/4) ln x ,
if c(ε, δ, γ) is chosen large enough.
Hence,
√

max √ |ψ(u, χ)| ≤ Cx exp −(c(2)/4) ln x .
2≤u≤x1+γ/ ln x

Substituting this estimate into (4.18) and using the fact that #XQ∗ ≤
#XQ = ϕ(Q), we obtain
√

S4′ ≤ Cx exp −(c(2)/4) ln x ln x =
√

= Cx exp −(c(2)/4) ln x + ln ln x ≤
√

≤ Cx exp −(c(2)/8) ln x , (4.19)
if c(ε, δ, γ) is chosen large enough.
Substituting (4.17) and (4.19) into (4.15), we obtain
√
S1′ ≤ Cx exp(−c ln x), (4.20)
where C > 0 and c > 0 are absolute constants.
II) Now we estimate
X 1 X
S3′ = max √ |ψ ′ (u, χ)|.
√ 1/2−δ
ϕ(Q) χ∈X ∗
2≤u≤x 1+γ/ ln x
exp(c1 ln x)<Q≤x Q
(Q,B)=1
√
Let Q be an integer with exp(c1 ln x) < Q ≤ x1/2−δ and (Q, B) = 1,
and let χ ∈ XQ∗ . Since Q > 1, we see that χ is a nonprincipal character
modulo Q. Hence,
ψ ′ (u, χ) = ψ(u, χ).
 CONSECUTIVE PRIMES IN SHORT INTERVALS 35

We obtain
X 1 X
S3′ = max √ |ψ(u, χ)| ≤
√ ϕ(Q) χ∈X ∗ 2≤u≤x1+γ/ ln x
exp(c1 ln x)<Q≤x1/2−δ Q
(Q,B)=1
X 1 X
≤ max √ |ψ(u, χ)|.
√ ϕ(Q) χ∈X ∗ 2≤u≤x1+γ/ ln x
exp(c1 ln x)<Q≤x1/2−δ Q
√
Applying
√
Lemma 4.8 with Q 1 = exp(c 1 ln x), Q2 = x1/2−δ and t =
x1+γ/ ln x , we obtain
√
 √

S3′ ≤ C ln4 x3/2−δ+γ/ ln x x exp (γ − c1 ) ln x +
√ √ 
+ x(5/6)(1+γ/ ln x) ln(x1/2−δ ) + x1−δ+γ/(2 ln x) .
We have
γ 5 γ 
√
≤ δ, 1+ √ ≤ 0.9,
ln x 6 ln x
if c(ε, δ, γ) is chosen large enough. Replacing C, we have
 √

S3′ ≤ C ln4 x x exp (γ − c1 ) ln x + x0.9 ln x + x1−δ/2 .
We have √ c1 √
(γ − c1 ) ln x ≤ − ln x,
2
if 0 < γ ≤ c1 /2. We obtain
√
√

x exp (γ − c1 ) ln x ln4 x ≤ x exp −(c1 /2) ln x + 4 ln ln x ≤
√

≤ x exp −(c1 /4) ln x ,
√

x0.9 ln5 x ≤ x exp −(c1 /4) ln x ,
√

x1−δ/2 ln4 x ≤ x exp −(c1 /4) ln x ,
if c(ε, δ, γ) is chosen large enough. Replacing 3C by C and c1 /4 by c,
we obtain √
S3′ ≤ Cx exp(−c ln x), (4.21)
where C > 0 and c > 0 are absolute constants.
III) Now we estimate
X 1 X
S2′ = max √ |ψ ′ (u, χ)|.
√ ϕ(Q) χ∈X ∗ 2≤u≤x1+γ/ ln x
ln x<Q≤exp(c1 ln x) Q
(Q,B)=1
√
Let Q be an integer with ln x < Q ≤ exp(c1 ln x) and (Q, B) =
1, and let χ ∈ XQ∗ . Since Q > 1, we see that χ is a nonprincipal
character modulo Q, and hence ψ ′ (u, χ) = ψ(u, χ). We√ recall that if
the exceptional modulus q0 in the interval [3, exp(2c1 ln x)] does not
exist, then B = 1; if q0 exists, then B ≥ 3 is a prime divisor of q0 ,
36 ARTYOM RADOMSKII

and hence Q 6= q0 . Since 0 < γ ≤ γ0 and c(ε, δ, γ) ≥ c0 , we see from

Lemma 4.6 that
√
max √ |ψ(u, χ)| ≤ Cx exp(−3c1 ln x).
2≤u≤x1+γ/ ln x

Since #XQ∗ ≤ #XQ = ϕ(Q), we obtain

X √
S2′ ≤ Cx exp(−3c1 ln x) ≤
√
ln x<Q≤exp(c1 ln x)
(Q,B)=1
√ √
≤ Cx exp(−3c1 ln x)exp(c1 ln x) =
√
= Cx exp(−2c1 ln x). (4.22)
From (4.20), (4.21) and (4.22) we obtain
√
e exp(−e
S1′ + S2′ + S3′ ≤ Cx c ln x), (4.23)
where C e > 0 and e c > 0 are absolute constants. Substituting (4.23)
into (4.14) we obtain
√ √

c ln x + ln ln x) ≤ C ′ x exp −(e
S ′ ≤ C ′ x exp(−e c/2) ln x ,
if c(ε, δ, γ) is chosen large enough. Replacing C ′ by C and ec/2 by c, we
obtain √
S ′ ≤ Cx exp(−c ln x), (4.24)
where C > 0 and c > 0 are absolute constants.
IV) We have
√
x1/2−δ ln2 x ≤ x1/2 ln2 x ≤ x exp(−c ln x), (4.25)
if c(ε, δ, γ) is chosen large enough (here c > 0 is the absolute constant
in (4.24)).
V) Now we estimate
 
A1 = max √ maxψ(u; 1, W ) − u.
2≤u≤x1+γ/ ln x W ∈Z

Let W ∈ Z. We have
X X
ψ(u; 1, W ) = Λ(n) = Λ(n) = ψ(u).
n≤u n≤u
n≡W (mod1)

Hence,
A1 = max √ |ψ(u) − u|.
2≤u≤x1+γ/ ln x

Using (4.16) and arguing as in I, 1, i) and I, 1, ii), we obtain

√
A1 ≤ Cx exp(−c ln x), (4.26)
where C > 0 and c > 0 are absolute constants.
Substituting (4.24), (4.25) and (4.26) into (4.13), we obtain
√
S ≤ Cx exp(−c ln x),
 CONSECUTIVE PRIMES IN SHORT INTERVALS 37

where C > 0 and c > 0 are absolute constants. Thus, if γ is sufficiently

small positive absolute constant, x is a real number with x ≥ c(ε, δ, γ),
q is an integer with 1 ≤ q ≤ (ln x)1−ε , then there is an integer B such
that
√ B
1 ≤ B ≤ exp(2c1 ln x), 1 ≤ ≤ 2, (B, q) = 1
ϕ(B)
and 
X  u  √
max √ max ψ(u; Q, W )−  ≤ Cx exp(−c ln x),
1/2−δ
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
1≤Q≤x (W,Q)=1
(Q,B)=1

where c1 , C and c are positive absolute constants. Let us denote 2c1

by c1 , C by c2 , c by c3 . Since γ is an absolute constant, we see that the
positive number c(ε, δ, γ) = c(ε, δ) depends only on ε and δ. Lemma
4.9 is proved.
Lemma 4.10. Let ε and δ be real numbers with 0 < ε < 1 and 0 <
δ < 1/2. Then there is a number c(ε, δ) > 0, depending only on ε
and δ, such that if x ∈ R and q ∈ Z are such that x ≥ c(ε, δ) and
1 ≤ q ≤ (ln x)1−ε , then there is a positive integer B such that
√ B
1 ≤ B ≤ exp(c1 ln x), 1 ≤ ≤ 2, (B, q) = 1
ϕ(B)
and
X  li(u)  √

max √ max π(u; Q, W )−  ≤ c2 x exp(−c3 ln x).
1/2−δ
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
1≤Q≤x (W,Q)=1
(Q,B)=1

Here c1 , γ, c2 and c3 are positive absolute constants.

Proof of Lemma 4.10. We choose e c(ε, δ) later; this number is
large enough. Let e c(ε, δ) ≥ c(ε, δ), where c(ε, δ) is the number in
Lemma 4.9. Let x ∈ R and q ∈ Z be such that x ≥ e c(ε, δ) and
1−ε
1 ≤ q ≤ (ln x) . Then, by Lemma 4.9, there is a positive integer B
such that
√ B
1 ≤ B ≤ exp(c1 ln x), 1 ≤ ≤ 2, (B, q) = 1 (4.27)
ϕ(B)
and
X √
max √ max |R(u; Q, W )| ≤ c2 x exp(−c3 ln x),
2≤u≤x1+γ/ ln x W ∈Z:
1≤Q≤x1/2−δ (W,Q)=1
(Q,B)=1
(4.28)
where
u
R(u; Q, W ) := ψ(u; Q, W ) −
ϕ(Q)
and c1 , γ, c2 and c3 are positive absolute constants.
38 ARTYOM RADOMSKII

We put
li(u)
R1 (u; Q, W ) := π(u; Q, W ) − . (4.29)
ϕ(Q)
Let Q ∈ Z, W ∈ Z and u ∈√Z be such that 1 ≤ Q ≤ x1/2−δ , (Q, B) = 1,
(W, Q) = 1, 3 ≤ u ≤ x1+γ/ ln x . We claim that
X |R(n; Q, W )|
|R1 (u; Q, W )| ≤ C1 u1/2 + |R(u; Q, W )| + 2 , (4.30)
2≤n≤u−1
n ln n
where C1 > 0 is an absolute constant. We define
(
1, if n ≡ W (mod Q),
α(n) =
0, otherwise;
X Λ(n)α(n)
π1 (u; Q, W ) = .
n≤u
ln n
Let us show that
e Q, W ),
π(u; Q, W ) = π1 (u; Q, W ) + R(u; e Q, W )| ≤ Cu1/2 ,
|R(u;
(4.31)
where C > 0 is an absolute constant. Let u ≥ 8. Then
X α(pm ) ln p X X α(pm )
π1 (u; Q, W ) = = =
m
p ≤u
m ln p m
1/m 1≤m≤ln u/ ln 2 p≤u
X X 1 X
= α(p) + α(pm ) = S1 + S2 .
p≤u
m
2≤m≤ln u/ ln 2 1/m
p≤u

We have X
S1 = 1 = π(u; Q, W ),
p≤u
p≡W (mod Q)

X u1/m 1 X u1/m
S2 ≤ = u1/2 + ≤
m 2 m
2≤m≤ln u/ ln 2 3≤m≤ln u/ ln 2

1 1 ln u
≤ u1/2 + u1/3 ≤ u1/2 + u1/3 ln u ≤ C ′ u1/2 ,
2 3 ln 2
where C ′ > 0 is an absolute constant. If 3 ≤ u < 8, then
 X 1 X  1 X 1 X
 m 
 α(p ) ≤ 1+ 1 = C ′′ ≤ C ′′ u1/2 .
m 2 3
2≤m≤ln u/ ln 2 1/m p≤u 1/2 1/3
p≤8 p≤8

Thus, (4.31) is proved.

Since X
ψ(x; Q, W ) = Λ(m)α(m),
m≤x
 CONSECUTIVE PRIMES IN SHORT INTERVALS 39

we have
X ψ(n; Q, W ) − ψ(n − 1; Q, W )
π1 (u; Q, W ) = =
2≤n≤u
ln n
X  1 1  ψ(u; Q, W )
= ψ(n; Q, W ) − + =
2≤n≤u−1
ln n ln(n + 1) ln u
X  n  1 1  u
= + R(n; Q, W ) − + +
2≤n≤u−1
ϕ(Q) ln n ln(n + 1) ϕ(Q) ln u
R(u; Q, W )
+ .
ln u
We have
Z n+1
X n  1 1  X n dt
− = =
2≤n≤u−1
ϕ(Q) ln n ln(n + 1) 2≤n≤u−1
ϕ(Q) n t ln2 t
1 X Z n+1 t − {t} Z
1  u dt
Z u
{t}dt 
= dt = − .
ϕ(Q) 2≤n≤u−1 n t ln2 t ϕ(Q) 2 ln2 t 2
2 t ln t

Since
Z Z u  u Z u
u
dt 1  t  dt
2 = td − =−  + =
2 ln t 2 ln t ln t 2 2 ln t
u 2
=− + + li(u),
ln u ln 2
we obtain
u R(u; Q, W ) u 2
π1 (u; Q, W ) = + − + +
ϕ(Q) ln u ln u ϕ(Q) ln u ϕ(Q) ln 2
Z u X  1 
li(u) 1 {t} 1
+ − dt + R(n; Q, W ) − .
ϕ(Q) ϕ(Q) 2 t ln2 t 2≤n≤u−1
ln n ln(n + 1)
We have (see (4.31))
li(u)
π(u; Q, W ) = + R1 (u; Q, W ),
ϕ(Q)
where
Z u
2 1 {t} e Q, W )+
R1 (u; Q, W ) = − dt + R(u;
ϕ(Q) ln 2 ϕ(Q) 2 t ln2 t
R(u; Q, W ) X  1 1 
+ + R(n; Q, W ) − .
ln u 2≤n≤u−1
ln n ln(n + 1)
40 ARTYOM RADOMSKII

We obtain
2 Z u {t} 
  e Q, W )|+
|R1 (u; Q, W )| ≤ + 2 dt + |R(u;
ln 2 2 t ln t
|R(u; Q, W )| X  1 1 
+ + |R(n; Q, W )| − . (4.32)
ln u 2≤n≤u−1
ln n ln(n + 1)
Since u ≥ 3, we have
|R(u; Q, W )|
≤ |R(u; Q, W )|. (4.33)
ln u
Since
Z u {t}  Z u dt u
  1  1 1 1
 2 dt ≤ 2 = − = − ≤ ,
2 t ln t 2 t ln t ln t 2 ln 2 ln u ln 2
we have (see (4.31))
2 Z u {t}  3  3  1/2
  e 1/2
+ 2 dt + |R(u; Q, W )| ≤ + Cu ≤ C + u .
ln 2 2 t ln t ln 2 ln 2
(4.34)
Let f (x) = − ln−1 x and let n be an integer with n ≥ 2. By the mean
value theorem, there is ξ ∈ (n, n + 1) such that
1 1 1 1
− = f (n + 1) − f (n) = f ′ (ξ) = 2 ≤ . (4.35)
ln n ln(n + 1) ξ ln ξ n ln2 n
Substituting (4.33), (4.34) and (4.35) into (4.32), we obtain (4.30).
Hence,
X
max √ max |R1 (u; Q, W )| ≤
3≤u≤x1+γ/ ln x W ∈Z
1≤Q≤x1/2−δ u∈Z (W,Q)=1
(Q,B)=1
X
≤ max √ max |R(u; Q, W )|+
3≤u≤x1+γ/ ln x W ∈Z
1≤Q≤x1/2−δ u∈Z (W,Q)=1
(Q,B)=1
X
+ max √ max |C1 u1/2 |+
3≤u≤x1+γ/ ln x W ∈Z
1≤Q≤x1/2−δ u∈Z (W,Q)=1
(Q,B)=1
X X |R(n; Q, W )|
+ max √ max =
3≤u≤x1+γ/ ln x W ∈Z
2≤n≤u−1
n ln2 n
1≤Q≤x1/2−δ u∈Z (W,Q)=1
(Q,B)=1

= S1 + S2 + S3 . (4.36)

I) Now we estimate S1 . We have (see (4.28))

X √
S1 ≤ max √ max |R(u; Q, W )| ≤ c2 x exp(−c3 ln x).
2≤u≤x1+γ/ ln x W ∈Z
1≤Q≤x1/2−δ (W,Q)=1
(Q,B)=1
(4.37)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 41

II) Now we estimate S2 . We have

γ
√ ≤ δ,
ln x
if e
c(ε, δ) is chosen large enough. We have
√ √
S2 ≤ C1 x1−δ+γ/(2 ln x)
≤ C1 x1−δ/2 ≤ x exp(−c3 ln x), (4.38)
if e
c(ε, δ) is chosen large enough.
III) Now we estimate S3 . Let Q, W , u and n be integers such
√
that
1/2−δ 1+γ/ ln x
1 ≤ Q ≤ x , (Q, B) = 1, (W, Q) = 1, 3 ≤ u ≤ x and
2 ≤ n ≤ u − 1. Then
|R(n; Q, W )| ≤ max √ max |R(m; Q, V )|.
2≤m≤x1+γ/ ln x V ∈Z
(V,Q)=1

Hence,
X |R(n; Q, W )|
≤ max √ max |R(m; Q, V )|·
2≤n≤u−1
n ln2 n 2≤m≤x1+γ/ ln x V ∈Z
(V,Q)=1
X 1
· 2 ≤ c0 max √ max |R(m; Q, V )|,
2≤n≤u−1
n ln n 2≤m≤x 1+γ/ ln x V ∈Z
(V,Q)=1

where ∞
X 1
c0 := < +∞.
n=2
n ln2 n
We have
X |R(n; Q, W )|
max √ max 2 ≤
3≤u≤x1+γ/ ln x W ∈Z
2≤n≤u−1
n ln n
u∈Z (W,Q)=1

≤ c0 max √ max |R(m; Q, V )|.

2≤m≤x1+γ/ ln x V ∈Z
(V,Q)=1

We obtain (see (4.28))

X
S3 ≤ c 0 max √ max |R(m; Q, V )| ≤
2≤m≤x1+γ/ ln x V ∈Z
1≤Q≤x1/2−δ (V,Q)=1
(Q,B)=1
√
≤ c0 c2 x exp(−c3 ln x). (4.39)
Substituting (4.37), (4.38) and (4.39) into (4.36), we obtain (see (4.29))
X  li(u) 

max √ max π(u; Q, W ) − ≤
1/2−δ 3≤u≤x 1+γ/ ln x W ∈Z ϕ(Q)
1≤Q≤x u∈Z (W,Q)=1
(Q,B)=1
√
≤ c4 x exp(−c3 ln x), (4.40)
where c4 = c2 + 1 + c0 c2 > 0 is an absolute constant.
42 ARTYOM RADOMSKII

Let Q and W be integers such that 1 ≤ Q ≤ x1/2−δ , (Q, B)√= 1 and

(W, Q) = 1, and let u be a real number with 2 ≤ u ≤ x1+γ/ ln x . Let
us consider two cases.
1) 2 ≤ u ≤ 3. Then
|π(u; Q, W )| ≤ π(u) ≤ 2,
 
 li(u) 
 
 ϕ(Q)  ≤ li(u) ≤ li(3),
   
   
π(u; Q, W ) − li(u)  ≤ |π(u; Q, W )| +  li(u)  ≤ 2 + li(3). (4.41)
 ϕ(Q)   ϕ(Q) 
√
2) 3 < u ≤ x1+γ/ ln x . Then
  Z [u]+1 Z 3
 li(u) − li([u])  dt dt
 ≤ ≤ = li(3).
 ϕ(Q)  ln t ln t
[u] 2

Hence,
   
   
π(u; Q, W ) − li(u)  = π([u]; Q, W ) − li(u) − li([u]) + li([u])  ≤
 ϕ(Q)   ϕ(Q) ϕ(Q) ϕ(Q) 
   
 li([u])   li(u) − li([u]) 
≤ π([u]; Q, W ) − +  ≤ li(3)+
ϕ(Q)   ϕ(Q)
 
 li([u]) 
+ π([u]; Q, W ) − . (4.42)
ϕ(Q) 
From (4.41) and (4.42) we obtain
 
 li(u) 
max √ 
max π(u; Q, W ) − ≤
2≤u≤x1+γ/ ln x W ∈Z ϕ(Q) 
(W,Q)=1
 
 li(u) 
≤ max √ 
max π(u; Q, W ) − + 2li(3) + 2. (4.43)
3≤u≤x1+γ/ ln x W ∈Z ϕ(Q) 
u∈Z (W,Q)=1

We have √
x1/2 ≤ x exp(−c3 ln x), (4.44)
if e
c(ε, δ) is chosen large enough. From (4.40), (4.43) and (4.44) we
obtain  
X  li(u) 
max √ max π(u; Q, W ) − ≤
2≤u≤x1+γ/ ln x W ∈Z ϕ(Q) 
1≤Q≤x1/2−δ (W,Q)=1
(Q,B)=1
 
X  li(u) 
≤ max √ max π(u; Q, W ) − +
3≤u≤x1+γ/ ln x W ∈Z ϕ(Q) 
1≤Q≤x1/2−δ u∈Z (W,Q)=1
(Q,B)=1
√
+ (2li(3) + 2)x1/2 ≤ (c4 + 2li(3) + 2)x exp(−c3 ln x). (4.45)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 43

Thus, if x is a real number with x ≥ e c(ε, δ) and q is an integer with

1−ε
1 ≤ q ≤ (ln x) , then there is a positive integer B such that (4.27)
and (4.45) hold. Let us denote e
c(ε, δ) by c(ε, δ) and c4 + 2li(3) + 2 by
c2 . Lemma 4.10 is proved.
5. Proof of Theorem 1.1 and Corollary 1.1
Let us introduce some notation. Let A be a set of integers, P be a
set of primes, L(n) = l1 n + l2 be a linear function with coefficients in
the integers. We define
A(x) = {n ∈ A : x ≤ n < 2x},
A(x; q, a) = {n ∈ A(x) : n ≡ a (mod q)},
L(A) = {L(n), n ∈ A},
PL,A (x) = L(A(x)) ∩ P,
PL,A (x; q, a) = L(A(x; q, a)) ∩ P,
ϕL (q) = ϕ(|l1 |q)/ϕ(|l1 |).
Let L = {L1 , . . . , Lk } be a set of distinct linear functions Li (n) =
ai n + bi , i = 1, . . . , k, with coefficients in the positive integers. We say
such a set is admissible if for every prime p there is an integer np such
Q
that ( ki=1 Li (np ), p) = 1.
We focus on sets which satisfy the following hypothesis, which is
given in terms of (A, L, P, B, x, θ) for L an admissible set of linear
functions, B ∈ N, x a large real number, and 0 < θ < 1.
Hypothesis 1. (A, L, P, B, x, θ). Let k = #L.
(1) A is well-distributed in arithmetic progressions: we have
X  #A(x)  #A(x)

max#A(x; q, a) − ≪ .
θ
a∈Z q (ln x)100k2
1≤q≤x

(2) Primes in L(A) ∩ P are well-distributed in most arithmetic pro-

gressions: for any L ∈ L we have
X  #PL,A (x)  #PL,A (x)

max #PL,A (x; q, a) − ≪ .
θ
a∈Z ϕL (q) (ln x)100k2
1≤q≤x (L(a),q)=1
(q,B)=1

(3) A is not too concentrated in any arithmetic progression: for any

1 ≤ q < xθ we have
#A(x)
max #A(x; q, a) ≪ .
a∈Z q
Maynard proved the following result (see [5, Proposition 6.1]).
Proposition 5.1. Let α and θ be real numbers with α > 0 and 0 < θ <
1. Let A be a set of integers, P be a set of primes, L = {L1 , . . . , Lk }
be an admissible set of k linear functions, and let B, x be integers.
44 ARTYOM RADOMSKII

Let the coefficients Li (n) = ai n + bi ∈ L satisfy 1 ≤ ai , bi ≤ xα for all

1 ≤ i ≤ k, and let k ≤ (ln x)1/5 and 1 ≤ B ≤ xα . Let xθ/10 ≤ R ≤ xθ/3 .
Let ρ, ξ satisfy k(ln ln x)2 / ln x ≤ ρ, ξ ≤ θ/10, and define
S(ξ; D) = {n ∈ N : p|n ⇒ (p > xξ or p|D)}.
Then there is a number C > 0 depending only on α and θ such that
the following holds. If k ≥ C and (A, L, P, B, x, θ) satisfy Hypothesis
1, then there is a choice of nonnegative weights wn = wn (L) satisfying
k
Y Y
wn ≪ (ln R)2k 4 (5.1)
i=1 p|Li (n)
p∤B

such that the following statements hold.

(1) We have
X   1  B k
wn = 1 + O SB (L)#A(x)(ln R)k Ik . (5.2)
(ln x)1/10 ϕ(B)k
n∈A(x)

(2) For L(n) = aL n + bL ∈ L we have

X   1  B k−1 ϕ(aL )
1P (L(n))wn ≥ 1 + O 1/10 k−1
SB (L) ·
(ln x) ϕ(B) aL
n∈A(x)
 Bk 
k+1 k−1
· #PL,A (x)(ln R) Jk + O SB (L)#A(x)(ln R) Ik . (5.3)
ϕ(B)k
/ L and D ≤ xα , if ∆L 6= 0 we have
(3) For L(n) = a0 n + b0 ∈
X ∆L D Bk
1S(ξ;D) (L(n))wn ≪ ξ −1 SB (L)#A(x)·
ϕ(∆L ) ϕ(D) ϕ(B)k
n∈A(x)

· (ln R)k−1 Ik , (5.4)

where
k
Y
∆L = |a0 | |a0 bi − b0 ai |.
i=1
(4) For L ∈ L we have
X X  Bk
1 wn ≪ ρ2 k 4 (ln k)2 SB (L)#A(x)(ln R)k Ik . (5.5)
ϕ(B)k
n∈A(x) p|L(n)
p<xρ
p∤B
 CONSECUTIVE PRIMES IN SHORT INTERVALS 45

Here Ik , Jk are quantities depending only on k, and SB (L) is a

quantity depending only on L, and these satisfy
Y Q
#{1 ≤ n ≤ p : p| ki=1 Li (n)}  1 −k
SB (L) = 1− 1− ≥
p p
p∤B

≥ exp(−ck), (5.6)
Z ∞ Z ∞
Ik = ··· F 2 (t1 , . . . , tk ) dt1 . . . dtk ≫ (2k ln k)−k , (5.7)
0 0
Z ∞ Z ∞ Z ∞ 2 ln k
Jk = ··· F (t1 , . . . , tk ) dtk dt1 . . . dtk−1 ≫ Ik ,
0 0 0 k
(5.8)
for a smooth function F = Fk : Rk → R depending only on k.
Here the implied constants depend only on α, θ, and the implied
constants from Hypothesis 1. The constant c in the inequality (5.6) is
a positive absolute constant.
Proof of Theorem 1.1. First we prove the following
Lemma 5.1. Let k be a positive integer. Let a, q, b1 , . . . , bk be positive
integers with b1 < . . . < bk and (a, q) = 1. Let
Li (n) = qn + a + qbi , i = 1, . . . , k.
Then L = {L1 , . . . , Lk } is an admissible set iff for any prime p such
that p ∤ q there is an integer mp such that mp 6≡ bi (mod p) for all
1 ≤ i ≤ k.
Proof of Lemma 5.1. 1) Let L = {L1 , . . . , Lk } be an admissible
set. Let p be a prime such that p ∤ q. Since L is an admissible set, there
Q
is an integer np such that ( ki=1 Li (np ), p) = 1. Since (q, p) = 1, there is
an integer q ′ such that qq ′ ≡ 1 (mod p). We put mp = −(np + q ′a). Let
i be an integer with 1 ≤ i ≤ k. Since (q ′ , p) = 1 and (Li (np ), p) = 1,
we have (q ′ Li (np ), p) = 1. We have
q ′ Li (np ) ≡ −mp + bi (mod p).
Hence, mp 6≡ bi (mod p).
2) Suppose that for any prime p such that p ∤ q there is an integer
mp such that mp 6≡ bi (mod p) for all 1 ≤ i ≤ k. Let us show that
then L is an admissible set. First we observe that L = {L1 , . . . , Lk } is
the set of distinct linear functions Li (n) = qn + li , i = 1, . . . , k, with
coefficients in the positive integers. Thus,Qwe must prove that for any
prime p there is an integer np such that ( ki=1 Li (np ), p) = 1. Let p be
a prime number. Let us consider two cases.
i) p|q. Since (a, q) = 1, we have (a, p) = 1. Let i be an integer with
1 ≤ i ≤ k. For any integer n we have
Li (n) ≡ a (mod p),
46 ARTYOM RADOMSKII
Q
and hence Li (n) 6≡ 0 (mod p). Therefore ( ki=1 Li (n), p) = 1. Therefore
in this case we may take as np any integer.
ii) p ∤ q. Then (q, p) = 1, and hence there is an integer c such that
qc ≡ a (mod p). (5.9)
By assumption, there is an integer mp such that mp 6≡ bi (mod p) for all
1 ≤ i ≤ k. We put np = −mp − c. Let i be an integer with 1 ≤ i ≤ k.
We have
np + c + bi 6≡ 0 (mod p).
Since (q, p) = 1, we obtain
qnp + qc + qbi 6≡ 0 (mod p).
Using (5.9), we obtain Li (np ) 6≡ 0 (mod p). Hence, (Li (np ), p) = 1.
Q
Since this holds for all 1 ≤ i ≤ k, we have ( ki=1 Li (np ), p) = 1. Lemma
5.1 is proved.
The proof of the following lemma is based on ideas of Maynard (see
the proof of Lemma 8.1 in [5]).
Lemma 5.2. There are positive absolute constants c and C such that
the following holds. Let x and η be real numbers with x ≥ c and
(ln x)−9/10 ≤ η ≤ 1. Let k and a be positive integers. Let b1 , . . . , bk
be integers with 1 ≤ bi ≤ ln x, i = 1, . . . , k. Let L = {L1 , . . . , Lk } be
the set of k linear functions, where
Li (n) = an + bi , i = 1, . . . , k.
For L(n) = an + b, b ∈ Z, we define
k
Y
∆L = ak+1 |bi − b|.
i=1
Then
X ∆L
≤ C ln ln(a + 2) ln(k + 1)η ln x.
1≤b≤η ln x
ϕ(∆L )
L=an+b∈L
/

Proof of Lemma 5.2. Let us consider two cases.

1) Let k > ln ln x. We have
ln ln x ≥ 100,
if c is chosen large enough. Therefore k ≥ 100. Let b to be an integer
such that 1 ≤ b ≤ η ln x and L = an + b ∈ / L. Then ∆L ∈ N. Applying
Lemma 2.9, we have
∆L
≤ c0 ln ln(∆L + 2), (5.10)
ϕ(∆L )
where c0 > 0 is an absolute constant. We have
k
X
ln ∆L = (k + 1) ln a + ln |bi − b|.
i=1
 CONSECUTIVE PRIMES IN SHORT INTERVALS 47

For any 1 ≤ i ≤ k we have

|bi − b| ≤ ln x.
Hence,
ln ∆L ≤ (k + 1) ln a + k ln ln x ≤ 2k ln a + k 2 .
Since
2k ln a ≤ k 2 ln(a + 2),
k 2 ≤ k 2 ln(a + 2),
we have
ln ∆L ≤ 2k 2 ln(a + 2).
We observe that if u and v are real numbers with u ≥ 2 and v ≥ 2,
then
u + v ≤ uv. (5.11)
Applying (5.11), we obtain
ln(∆L + 2) ≤ ln(3∆L ) = ln ∆L + ln 3 ≤
≤ 2k 2 ln(a + 2) + 3 ≤ 6k 2 ln(a + 2).
Applying (5.11) again, we have
ln ln(∆L + 2) ≤ ln 6 + 2 ln k + ln ln(a + 2) ≤
≤ 2 + 2 ln k + 25 ln ln(a + 2) ≤ 4 ln k + 25 ln ln(a + 2) ≤
≤ 100 ln k ln ln(a + 2) ≤ 100 ln(k + 1) ln ln(a + 2).
Substituting this estimate into (5.10), we obtain
∆L
≤ 100c0 ln ln(a + 2) ln(k + 1) = c1 ln ln(a + 2) ln(k + 1),
ϕ(∆L )
where c1 = 100c0 > 0 is an absolute constant. We have
X ∆L X
≤ c1 ln ln(a + 2) ln(k + 1) 1≤
1≤b≤η ln x
ϕ(∆L ) 1≤b≤η ln x
L=an+b∈L
/ L=an+b∈L
/
≤ c1 ln ln(a + 2) ln(k + 1)[η ln x] ≤
≤ c1 ln ln(a + 2) ln(k + 1)η ln x. (5.12)
2) Let 1 ≤ k ≤ ln ln x. For an integer b we define
k
Y
∆(b) := |b − bi |.
i=1

Let b to be an integer such that 1 ≤ b ≤ η ln x and L = an + b ∈

/ L.
Applying Lemmas 2.6 and 2.5, we obtain
∆L ak+1 ∆(b) ak+1 ∆(b) a ∆(b)
= k+1
≤ k+1
= .
ϕ(∆L ) ϕ(a ∆(b)) ϕ(a ) ϕ(∆(b)) ϕ(a) ϕ(∆(b))
48 ARTYOM RADOMSKII

Hence,
X ∆L a X ∆(b) a e
S= ≤ = S. (5.13)
1≤b≤η ln x
ϕ(∆L ) ϕ(a) 1≤b≤η ln x
ϕ(∆(b)) ϕ(a)
L=an+b∈L
/ L=an+b∈L
/

Applying Lemma 2.8, we have

X ∆(b) X X µ2 (d)
Se = = =
1≤b≤η ln x
ϕ(∆(b)) 1≤b≤η ln x
ϕ(d)
d|∆(b)
L=an+b∈L
/ L=an+b∈L
/
X  X µ2 (d) X µ2 (d) 
= + = S1 + S2 . (5.14)
ϕ(d) ϕ(d)
1≤b≤η ln x 1≤d≤η ln x d>η ln x
L=an+b∈L
/ d|∆(b) d|∆(b)

First we estimate the sum S2 . Let b and d be positive integers such

that 1 ≤ b ≤ η ln x, L = an + b ∈ / L, d > η ln x and d|∆(b). We claim
that P
µ2 (d) µ2 (d) p|d ln p
≤ . (5.15)
ϕ(d) ϕ(d) ln(η ln x)
We have
d > η ln x ≥ (ln x)1/10 ≥ 100,
if c is chosen large enough. If µ2 (d) = 0, then the inequality (5.15)
holds. Let µ2 (d) 6= 0. Then d is square-free. Therefore
X
ln p = ln d.
p|d

The inequality (5.15) is equivalent to the inequality

X
ln(η ln x) ≤ ln p = ln d,
p|d

which, obviously, holds. Thus, (5.15) is proved. We have

P
X X µ2 (d) X X µ2 (d) p|d ln p
S2 = ≤ =
1≤b≤η ln x d>η ln x
ϕ(d) 1≤b≤η ln x d>η ln x
ϕ(d) ln(η ln x)
/ d|∆(b)
L=an+b∈L / d|∆(b)
L=an+b∈L
X X ln p X µ2 (d)
= .
ln(η ln x) ϕ(d)
1≤b≤η ln x p|∆(b) d>η ln x
L=an+b∈L
/ d is a multiple of p
d|∆(b)

Let b ∈ N, d ∈ N and p ∈ P be such that 1 ≤ b ≤ η ln x, L = an+b ∈ / L,

p|∆(b), d > η ln x, d is a multiple of p and d|∆(b). Then d = pt, where
t ∈ N, t > (η ln x)/p and t|∆(b). We have (see Lemmas 2.6 and 2.5)
p
ϕ(d) = ϕ(pt) ≥ ϕ(p)ϕ(t) = (p − 1)ϕ(t) ≥ ϕ(t).
2
 CONSECUTIVE PRIMES IN SHORT INTERVALS 49

Hence,
µ2 (d) µ2 (pt) 2µ2 (pt) 2µ2 (t)
= ≤ ≤ .
ϕ(d) ϕ(pt) pϕ(t) pϕ(t)
We obtain (see Lemma 2.8)
X µ2 (d) 2 X µ2 (t) 2 X µ2 (t) 2 ∆(b)
≤ ≤ = .
d>η ln x
ϕ(d) p ϕ(t) p ϕ(t) p ϕ(∆(b))
t>(η ln x)/p t|∆(b)
d is a multiple of p t|∆(b)
d|∆(b)

Hence,
X X ln p 2 ∆(b)
S2 ≤ =
1≤b≤η ln x p|∆(b)
ln(η ln x) p ϕ(∆(b))
L=an+b∈L
/
2 X ∆(b) X ln p
= .
ln(η ln x) 1≤b≤η ln x
ϕ(∆(b)) p
p|∆(b)
L=an+b∈L
/

Since
η ≥ (ln x)−9/10 ,
we have
2 2 20
≤ = .
ln(η ln x) ln((ln x)1/10 ) ln ln x
We obtain
20 X ∆(b) X ln p
S2 ≤ . (5.16)
ln ln x 1≤b≤η ln x
ϕ(∆(b)) p
p|∆(b)
L=an+b∈L
/
Let b to be an integer such that 1 ≤ b ≤ η ln x and L = an + b ∈/ L.
Applying Lemmas 2.9 and 2.11, we have
∆(b)
≤ c2 ln ln(∆(b) + 2) ≤ c2 ln ln(3∆(b)), (5.17)
ϕ(∆(b))
X ln p
≤ c3 ln ln(3∆(b)), (5.18)
p
p|∆(b)

where c2 > 0 and c3 > 0 are absolute constants. We have

k
X
ln ∆(b) = ln |bi − b| ≤ k ln ln x ≤ (ln ln x)2 .
i=1
Hence,
ln ln(3∆(b)) = ln(ln 3 + ln ∆(b)) ≤
≤ ln(ln 3 + (ln ln x)2 ) ≤ 3 ln ln ln x, (5.19)
if c is chosen large enough. From (5.17), (5.18) and (5.19) we obtain
∆(b) X ln p
≤ c2 c3 9(ln ln ln x)2 = c4 (ln ln ln x)2 ,
ϕ(∆(b)) p
p|∆(b)
50 ARTYOM RADOMSKII

where c4 = 9c2 c3 > 0 is an absolute constant. Substituting this esti-

mate into (5.16), we obtain
20c4 (ln ln ln x)2
S2 ≤ η ln x.
ln ln x
We have
20c4 (ln ln ln x)2
≤ 1,
ln ln x
if c is chosen large enough. Hence,
1
S2 ≤ η ln x ≤ ln(k + 1)η ln x ≤ 2 ln(k + 1)η ln x. (5.20)
ln 2
Now we estimate S1 . We have
X X µ2 (d) X X µ2 (d)
S1 = = =
1≤b≤η ln x 1≤d≤η ln x
ϕ(d) 1≤d≤η ln x 1≤b≤η ln x
ϕ(d)
L=an+b∈L
/ d|∆(b) L=an+b∈L/
d|∆(b)
X µ2 (d) X X µ2 (d)
= 1= N0 (d) =
1≤d≤η ln x
ϕ(d) 1≤b≤η ln x 1≤d≤η ln x
ϕ(d)
L=an+b∈L/
d|∆(b)
X 1
= N0 (d). (5.21)
1≤d≤η ln x
ϕ(d)
d∈M

Let d be an integer such that 1 ≤ d ≤ η ln x and d ∈ M. We claim

that
2η ln x Y
N0 (d) ≤ min(p, k). (5.22)
d
p|d
If d = 1, then the inequality is obvious. Let d > 1. We define
R(b) = (b − b1 ) · · · (b − bk ).
Then ∆(b) = |R(b)|. We have
X X
N0 (d) = 1= 1.
1≤b≤η ln x 1≤b≤η ln x
L=an+b∈L/ L=an+b∈L/
d|∆(b) R(b)≡0 (mod d)

Let d be expressed in standard form

d = q1 · · · qr ,
where q1 < . . . < qr are prime numbers. It is well-known (see, for
example, [7, Chapter 4]), the congruence
R(b) ≡ 0 (mod d)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 51

is equivalent to the system of congruences


R(b) ≡ 0 (mod q1 ),

.. (5.23)
 .

R(b) ≡ 0 (mod qr ).
Let 1 ≤ j ≤ r. Let Ωj be the set of numbers of a complete system of
residues modulo qj satisfying the congruence
R(b) ≡ 0 (mod qj ).
Since R(b1 ) = 0, we see that Ωj 6= ∅. Since the leading coefficient of
the polynomial R(b) is 1 and the degree of the polynomial R(b) is k,
we have #Ωj ≤ k (see, for example, [7, Chapter 4])). It is clear that
#Ωj ≤ qj . We obtain
#Ωj ≤ min(qj , k).
The system (5.23) is equivalent to the union of
T = #Ω1 · · · #Ωr
systems 
b ≡ τ1 (mod q1 ),

.. (5.24)
 .

b ≡ τr (mod qr ),
where τ1 ∈ Ω1 , . . . , τr ∈ Ωr . It is well-known (see, for example, [7,
Chapter 4]), the system of congruences (5.24) is equivalent to the con-
gruence
b ≡ x0 (mod d),
where x0 = x0 (τ1 , . . . , τr ). It is also well-known that the numbers
x0 (τ1 , . . . , τr ), τ1 ∈ Ω1 , . . . , τr ∈ Ωr , are incongruent modulo d. Thus,
G
{b ∈ Z : R(b) ≡ 0 (mod d)} = {x0 (τ1 , . . . , τr ) + dt, t ∈ Z}.
τ1 ∈Ω1 ,...,τr ∈Ωr

Let τ1 ∈ Ω1 , . . . , τr ∈ Ωr , x0 = x0 (τ1 , . . . , τr ). We have

   
η ln x − x0 1 − x0
#{t ∈ Z : 1 ≤ x0 + dt ≤ η ln x} = − +1=
d d
 
η ln x − x0 1 − x0 η ln x 1
= − θ1 − + θ2 + 1 = + 1 − θ1 − θ2 − ,
d d d d
where θ1 and θ2 are real numbers with 0 ≤ θ1 < 1 and 0 ≤ θ2 < 1.
Since 1 ≤ d ≤ η ln x, we obtain
η ln x η ln x
#{t ∈ Z : 1 ≤ x0 + dt ≤ η ln x} ≤ +1≤2 .
d d
We obtain
2η ln x 2η ln x Y
N0 (d) ≤ T ≤ min(p, k).
d d
p|d
52 ARTYOM RADOMSKII

The inequality (5.22) is proved.

Substituting (5.22) into (5.21), we obtain
X 1 2η ln x Y
S1 ≤ min(p, k) =
1≤d≤η ln x
ϕ(d) d
p|d
d∈M
Q
X p|d min(p, k)
= 2η ln x = 2η ln xS3 . (5.25)
1≤d≤η ln x
dϕ(d)
d∈M

Let d be an integer with 1 ≤ d ≤ η ln x and d ∈ M. We have (see

Lemmas 2.7 and 2.5)
Y
d= p,
p|d
Y Y
ϕ(d) = ϕ(p) = (p − 1),
p|d p|d
Q Q
p|d min(p, k) p|d min(p, k)
Y 1 Y k
= Q = .
dϕ(d) p|d p(p − 1) p−1 p(p − 1)
p|d p|d
p≤k p>k

Hence,
X Y 1 Y k
S3 = ≤
1≤d≤η ln x p|d
p−1 p(p − 1)
p|d
d∈M p≤k p>k
Y 1  Y k 
≤ 1+ 1+ = AB. (5.26)
p≤k
p − 1 p>k p(p − 1)
We have (see Lemma 2.2)
Y 1  Y  1 
A= 1+ ≤ 1+ =
p−1 p−1
p≤k p≤k+1
Y  1 −1
= 1− ≤ c5 ln(k + 1), (5.27)
p≤k+1
p
where c5 > 0 is an absolute constant.
Now we estimate B. Since ln(1 + u) ≤ u, u > 0, we have
X  k  X k
ln B = ln 1 + ≤ =
p>k
p(p − 1) p>k
p(p − 1)
X 1 X 1
=k ≤k .
p(p − 1) n(n − 1)
p≥k+1 n≥k+1
 CONSECUTIVE PRIMES IN SHORT INTERVALS 53

We define
m 
1
m
X X
1 1
sm = = − =
n=k+1
n(n − 1) n=k+1 n − 1 n
1 1
= − , m ≥ k + 1.
k m
Hence,
X 1 1
= lim sm = .
n≥k+1
n(n − 1) m→+∞ k
We obtain ln B ≤ 1, i. e.
B ≤ e < 3. (5.28)
We have (see (5.26), (5.27) and (5.28))
S3 ≤ c6 ln(k + 1),
where c6 > 0 is an absolute constant. Substituting this estimate into
(5.25), we obtain
S1 ≤ c7 ln(k + 1)η ln x, (5.29)
where c7 > 0 is an absolute constant.
We obtain (see (5.14), (5.20) and (5.29))
Se ≤ (c7 + 2) ln(k + 1)η ln x = c8 ln(k + 1)η ln x,
where c8 = c7 + 2 > 0 is an absolute constant. We obtain (see (5.13)
and Lemma 2.9)
a
S ≤ c8 ln(k + 1)η ln x ≤ c9 ln ln(a + 2) ln(k + 1)η ln x, (5.30)
ϕ(a)
where c9 > 0 is an absolute constant. We put
C = c1 + c9 ,
where c1 is the constant in (5.12). Then C > 0 is an absolute constant
and in both cases, 1 ≤ k ≤ ln ln x and k > ln ln x, we have
X ∆L
≤ C ln ln(a + 2) ln(k + 1)η ln x.
1≤b≤η ln x
ϕ(∆L )
L=an+b∈L
/

Lemma 5.2 is proved.

Lemma 5.3. Let A = N, P = P, α = 1/5, θ = 1/3, let C0 =
C(1/5, 1/3) > 0 be the absolute constant in Proposition 5.1. Let ε be a
real number with 0 < ε < 1. Then there is a number c0 (ε) > 0 such that
the following holds. Let x ∈ N, y ∈ R, q ∈ N be such that x ≥ c0 (ε),
1 ≤ y ≤ ln x, 1 ≤ q ≤ y 1−ε . Then there is a positive integer B such
that
√ B
1 ≤ B ≤ exp(ϑ ln x), 1 ≤ ≤ 2, (B, q) = 1. (5.31)
ϕ(B)
54 ARTYOM RADOMSKII

Furthermore, let
k ∈ N, ρ ∈ R, ξ ∈ R, R ∈ R, η ∈ R, a ∈ Z,
C0 ≤ k ≤ (ln x)1/5 , (5.32)
k(ln ln x)2 1
≤ρ≤ , ξ = ρ, (5.33)
ln x 30
1
R = x1/9 , 0<η≤ , (5.34)
2
1 ≤ a ≤ q, (a, q) = 1. (5.35)
Let L = {L1 , . . . , Lk } be an admissible set of k linear functions, where
Li (n) = qn + a + qbi , i = 1, . . . , k, b1 , . . . , bk are positive integers with
b1 < . . . < bk and qbk ≤ ηy. Then the assumption of Proposition
5.1 holds and there are nonnegative weights wn = wn (L) satisfying the
statement of Proposition 5.1; the implied constants in (5.1) – (5.5) are
positive and absolute.
Here ϑ > 0 is an absolute constant.
Proof of Lemma 5.3. We choose c0 (ε) later; this number is large
enough. We take δ = 0.1 and let c0 (ε) ≥ c(ε, δ) = c(ε, 0.1), where
c(ε, δ) is the quantity in Lemma 4.10. Let x ∈ N, y ∈ R, q ∈ N be such
that x ≥ c0 (ε), 1 ≤ y ≤ ln x, 1 ≤ q ≤ y 1−ε . By Lemma 4.10, there is a
positive integer B such that
√ B
1 ≤ B ≤ exp(c1 ln x), 1 ≤ ≤ 2, (B, q) = 1
ϕ(B)
and
X  li(u)  √

max √ max π(u; Q, W ) −  ≤ c2 x exp(−c3 ln x),
0.4 2≤u≤x
1+γ/ ln x W ∈Z: ϕ(Q)
1≤Q≤x (W,Q)=1
(Q,B)=1
(5.36)
where c1 , γ, c2 and c3 are positive absolute constants. Let (5.32) –
(5.35) hold. Let L = {L1 , . . . , Lk } be an admissible set of k linear func-
tions, where Li (n) = qn + a + qbi , i = 1, . . . , k, b1 , . . . , bk are positive
integers with b1 < . . . < bk and qbk ≤ ηy. Let us show that the assump-
tion of Proposition 5.1 holds. First we show that (A, L, P, B, x, 1/3)
satisfy Hypothesis 1.
I) Let us show that part (2) of Hypothesis 1 holds. Let L(n) =
l1 n + l2 ∈ L. It is clear that
1 ≤ l1 ≤ ln x, 1 ≤ l2 ≤ ln x. (5.37)
Let us show that
X  #PL,A (x)  #PL,A (x)

S := max #PL,A (x; r, b)− ≤ . (5.38)
1/3
b∈Z ϕL (r) (ln x)100k2
1≤r≤x (L(b),r)=1
(r,B)=1
 CONSECUTIVE PRIMES IN SHORT INTERVALS 55

It is not hard to see that


PL,A (x) = l1 x + l2 ≤ p < 2l1 x + l2 : p ≡ l2 (mod l1 ) ,

PL,A (x; r, b) = l1 x + l2 ≤ p < 2l1 x + l2 : p ≡ l1 b + l2 (mod l1 r) ,
and hence
#PL,A (x) = π(2l1 x + l2 − 1; l1 , l2 ) − π(l1 x + l2 − 1; l1 , l2 ), (5.39)
#PL,A (x; r, b) = π(2l1 x + l2 − 1; l1 r, L(b)) − π(l1 x + l2 − 1; l1 r, L(b)).
We obtain 
X 
S= max π(2l1 x + l2 − 1; l1 r, L(b))−
b∈Z
1≤r≤x1/3 (L(b),r)=1
(r,B)=1

−π(l1 x + l2 − 1; l1 r, L(b))−
π(2l1 x + l2 − 1; l1 , l2 ) − π(l1 x + l2 − 1; l1 , l2 ) 
− ≤
ϕ(l1 r)/ϕ(l1 )
≤ S1 + S2 + S3 + S4 , (5.40)
where
X  li(l1 x + l2 − 1) 

S1 = max π(l1 x + l2 − 1; l1 r, L(b)) − ,
b∈Z ϕ(l1 r)
1≤r≤x1/3 (L(b),r)=1
(r,B)=1
X  π(l1 x + l2 − 1; l1 , l2 ) li(l1 x + l2 − 1) 
S2 =  − ,
1/3
ϕ(l1 r)/ϕ(l1 ) ϕ(l1 r)
1≤r≤x
(r,B)=1
X  li(2l1 x + l2 − 1) 

S3 = max π(2l1 x + l2 − 1; l1 r, L(b)) − ,
b∈Z ϕ(l1 r)
1≤r≤x1/3 (L(b),r)=1
(r,B)=1
X  π(2l1 x + l2 − 1; l1 , l2 ) li(2l1 x + l2 − 1) 
S4 =  − .
1/3
ϕ(l1 r)/ϕ(l1 ) ϕ(l1 r)
1≤r≤x
(r,B)=1

Let us show that

(L(b), l1 ) = 1 (5.41)
for any b ∈ Z. Assume the converse: there is an integer b such that
(L(b), l1 ) > 1. Then there is a prime p such that p|l1 and p|L(b). Hence
p|l2 , and we see that p|L(n) for any integer n. Since L ∈ L, we see that
p|L1 (n) · · · Lk (n) for any integer n. But this contradicts the fact that
L = {L1 , . . . , Lk } is an admissible set. Thus, (5.41) is proved.
We observe that since (B, q) = 1 and l1 = q, we have
(B, l1 ) = 1. (5.42)
56 ARTYOM RADOMSKII

Let r be an integer with 1 ≤ r ≤ x1/3 and (r, B) = 1. Applying (5.37),

we have
l1 r ≤ x1/3 ln x ≤ x0.4 ,
l1 x + l2 − 1 ≥ l1 x ≥ x ≥ 2,
√
l1 x + l2 − 1 ≤ 2x ln x ≤ x1+γ/ ln x
,
if c0 (ε) is chosen large enough. Hence, we obtain (see (5.41), (5.42) and
(5.36))
X  li(l1 x + l2 − 1) 

S1 = max π(l1 x + l2 − 1; l1 r, L(b)) − ≤
r:
b∈Z ϕ(l1 r)
(L(b),l1 r)=1
l1 ≤l1 r≤l1 x1/3
(l1 r,B)=1
X  li(u)  √

≤ max √ max π(u; Q, W ) −  ≤ c2 x exp(−c3 ln x).
2≤u≤x1+γ/ ln x W ∈Z: ϕ(Q)
1≤Q≤x0.4 (W,Q)=1
(Q,B)=1
(5.43)
Applying Lemmas 2.6 and 2.10, we obtain
 li(l1 x + l2 − 1)  X 1

S2 = ϕ(l1 )π(l1 x + l2 − 1; l1 , l2 ) −  ≤
ϕ(l1 ) ϕ(l1 r)
1≤r≤x1/3
(r,B)=1
 li(l1 x + l2 − 1)  X 1

≤ π(l1 x + l2 − 1; l1 , l2 ) −  ≤
ϕ(l1 ) 1/3
ϕ(r)
1≤r≤x
 li(l1 x + l2 − 1) 

≤e c ln xπ(l1 x + l2 − 1; l1 , l2 ) − ,
ϕ(l1 )
where e
c > 0 is an absolute constant. Since (see (5.37))
l1 x + l2 − 1 ≥ l1 x ≥ x,
we obtain
1 ≤ l1 ≤ ln x ≤ ln(l1 x + l2 − 1).
Hence, (see, for example, [1, Chapter 22])
 li(l1 x + l2 − 1) 

π(l1 x + l2 − 1; l1 , l2 ) − ≤
ϕ(l1 )
p

≤ C(l1 x + l2 − 1)exp −c ln(l1 x + l2 − 1) , (5.44)
where C and c are positive absolute constants. We have
p
√
exp −c ln(l1 x + l2 − 1) ≤ exp(−c ln x), (5.45)
l1 x + l2 − 1 ≤ x ln x + ln x ≤ 2x ln x. (5.46)
We have √ c√
− c ln x + 2 ln ln x ≤ − ln x, (5.47)
2
 CONSECUTIVE PRIMES IN SHORT INTERVALS 57

if c0 (ε) is chosen large enough. Hence,

√
e exp(−c ln x + 2 ln ln x) ≤
S2 ≤ Cx
√
e exp(−(c/2) ln x),
≤ Cx (5.48)
where Ce = 2e
cC is a positive absolute constant. Similarly, it can be
shown that
√ √
S3 ≤ Cx exp(−c ln x), S4 ≤ Cx exp(−c ln x), (5.49)
where C and c are positive absolute constants. Substituting (5.43),
(5.48) and (5.49) into (5.40), we obtain
X  #PL,A (x)  √

max #PL,A (x; r, b) −  ≤ c4 x exp(−c5 ln x),
1/3
b∈Z ϕL (r)
1≤r≤x (L(b),r)=1
(r,B)=1
(5.50)
where c4 and c5 are positive absolute constants.
Applying (5.44) – (5.47), we have
li(l1 x + l2 − 1)
π(l1 x + l2 − 1; l1 , l2 ) = + R1 ,
ϕ(l1 )
√
|R1 | ≤ Cx exp(−c ln x),
where C and c are positive absolute constants. Similarly, it can be
shown that
li(2l1 x + l2 − 1)
π(2l1 x + l2 − 1; l1 , l2 ) = + R2 ,
ϕ(l1 )
√
|R2 | ≤ Cx exp(−c ln x),
where C and c are positive absolute constants. Hence, we have (see
(5.39))
li(2l1 x + l2 − 1) − li(l1 x + l2 − 1)
#PL,A (x) = + R, (5.51)
ϕ(l1 )
√
|R| ≤ c6 x exp(−c7 ln x), (5.52)
where c6 and c7 are positive absolute constants. We have
2l1 x + l2 − 1 ≤ 2x ln x + ln x ≤ 3x ln x,
ln(2l1 x + l2 − 1) ≤ ln x + ln ln x + ln 3 ≤ 2 ln x,
if c0 (ε) is chosen large enough. Hence,
Z 2l1 x+l2 −1
li(2l1 x + l2 − 1) − li(l1 x + l2 − 1) 1 dt
= ≥
ϕ(l1 ) ϕ(l1 ) l1 x+l2 −1 ln t
l1 x l1 x
≥ ≥ . (5.53)
ϕ(l1 ) ln(2l1 x + l2 − 1) 2ϕ(l1 ) ln x
58 ARTYOM RADOMSKII

Let us show that

l1 x
|R| ≤ . (5.54)
4ϕ(l1 ) ln x
Since l1 /ϕ(l1 ) ≥ 1, we see from (5.52) that it is sufficient to show that
√ x
c6 x exp(−c7 ln x) ≤ .
4 ln x
This inequality holds, if c0 (ε) is chosen large enough. Thus, (5.54) is
proved. From (5.51), (5.53) and (5.54) we obtain
l1 x
#PL,A (x) ≥ . (5.55)
4ϕ(l1 ) ln x
Now we prove (5.38). Since l1 /ϕ(l1 ) ≥ 1, we see from (5.50) and
(5.55) that it suffices to show that
√ x
c4 x exp(−c5 ln x) ≤ . (5.56)
4(ln x)100k2 +1
Taking logarithms, we obtain
√
ln(c4 ) + ln x − c5 ln x ≤ ln x − ln 4 − 100k 2 ln ln x − ln ln x
or, that is equivalent,
√
100k 2 ln ln x ≤ c5 ln x − ln ln x − ln(4c4 ).
Since k ≤ (ln x)1/5 , we have
100k 2 ln ln x ≤ 100(ln x)2/5 ln ln x.
The inequality
√
100(ln x)2/5 ln ln x ≤ c5 ln x − ln ln x − ln(4c4 )
holds, if c0 (ε) is chosen large enough. The inequality (5.56) is proved.
Thus, (5.38) is proved.
II) Let us show that part (1) of Hypothesis 1 holds. We show that
X  #A(x)  #A(x)

S := max#A(x; r, b) − ≤ . (5.57)
1/3
b∈Z r (ln x)100k2
1≤r≤x

Let 1 ≤ r ≤ x1/3 , b ∈ Z. We have

A(x) = {x ≤ n < 2x},
A(x; r, b) = {x ≤ n < 2x : n ≡ b (mod r)}.
Hence,
#A(x) = x; (5.58)
x
#A(x; r, b) = + ρ, |ρ| ≤ 1. (5.59)
r
We obtain 
 #A(x) 
#A(x; r, b) −  = |ρ| ≤ 1.
r
 CONSECUTIVE PRIMES IN SHORT INTERVALS 59

Hence,
S ≤ x1/3 .
Thus, to prove (5.57) it suffices to show that
x
x1/3 ≤
(ln x)100k2
or, that is equivalent,
2
(ln x)100k ≤ x2/3 .
Taking logarithms, we obtain
2
100k 2 ln ln x ≤ ln x.
3
1/5
Since k ≤ (ln x) , we have
100k 2 ln ln x ≤ 100(ln x)2/5 ln ln x.
The inequality
2
100(ln x)2/5 ln ln x ≤ ln x
3
holds, if c0 (ε) is chosen large enough. Thus, (5.57) is proved.
III) Let us show that part (3) of Hypothesis 1 holds. Let us show
that for any integer r with 1 ≤ r < x1/3 we have
#A(x)
max #A(x; r, b) ≤ 2 . (5.60)
b∈Z r
Let 1 ≤ r < x1/3 and b ∈ Z. We may assume that c0 (ε) ≥ 2. Hence,
r ≤ x1/3 ≤ x.
Applying (5.58) and (5.59), we obtain
x x #A(x)
#A(x; r, b) ≤ + 1 ≤ 2 = 2 ,
r r r
and (5.60) is proved. Thus, (A, L, P, B, x, 1/3) satisfy Hypothesis 1.
We have √
exp(c1 ln x) ≤ x1/5 , ln x ≤ x1/5 ,
√
if c0 (ε) is chosen large enough. Since 1 ≤ B ≤ exp(c1 ln x), we obtain
1 ≤ B ≤ x1/5 . Let L = l1 n + l2 ∈ L. Applying (5.37), we have 1 ≤ l1 ≤
x1/5 , 1 ≤ l2 ≤ x1/5 . Thus, the assumption of Proposition 5.1 holds and
there are nonnegative weights wn = wn (L) satisfying the statement
of Proposition 5.1. Since in Proposition 5.1 the implied constants in
(5.1) – (5.5) depend only on α, θ and on the implied constants from
Hypothesis 1, and in our case these constants are absolute (α = 1/5,
θ = 1/3 and see (5.38), (5.57) and (5.60)), we see that in our case the
implied constants in (5.1) – (5.5) are positive and absolute. Finally, let
us denote c1 by ϑ. Lemma 5.3 is proved.
60 ARTYOM RADOMSKII

Lemma 5.4. There are positive absolute constants c and C such that
the following holds. Let ε be a real number with 0 < ε < 1. Then there
is a number c0 (ε) > 0, depending only on ε, such that if x ∈ N, y ∈ R,
m ∈ Z, q ∈ Z, a ∈ Z are such that c0 (ε) ≤ y ≤ ln x, 1 ≤ m ≤ c · ε ln y,
1 ≤ q ≤ y 1−ε and (a, q) = 1, then
#{qx < pn ≤ 2qx − 5q : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y}
 exp(Cm)
y
≥ π(2qx) .
2q ln x
Proof of Lemma 5.4. Let A = N, P = P, α = 1/5, θ = 1/3, let
C0 = C(1/5, 1/3) > 0 be the absolute constant in Proposition 5.1. Let
c0 (ε) be the quantity in Lemma 5.3. We choose c(ε) later; this number
is large enough. Let c(ε) ≥ c0 (ε). Let x ∈ N, y ∈ R, q ∈ Z are such
that
c(ε) ≤ y ≤ ln x, (5.61)
1−ε
1≤q≤y . (5.62)
By Lemma 5.3, there is a positive integer B such that (5.31) holds. We
assume that
Ce0 ≤ k ≤ y ε/14 , (5.63)
where C e0 > 0 is an absolute constant. We choose C e0 later; this number
e
is large enough. We may assume that C0 ≥ C0 . It follows from (5.61)
and (5.63) that k ≤ (ln x)1/5 . Thus, (5.32) holds. Let (5.33) – (5.35)
hold. Let L = {L1 , . . . , Lk } be an admissible set of k linear functions,
where Li (n) = qn + a + qbi , i = 1, . . . , k, b1 , . . . , bk are positive integers
with b1 < . . . < bk and qbk ≤ ηy. Then (see Lemma 5.3) the assumption
of Proposition 5.1 holds and there are nonnegative weights wn = wn (L)
satisfying the statement of Proposition 5.1; the implied constants in
(5.1) – (5.5) are positive and absolute. We write L = L(b) for such a
set given by b1 , . . . , bk .
Let m be a positive integer. We consider
X X X k
S= 1P (Li (n)) − m−
1≤b1 <...<bk n∈A(x) i=1
qbk ≤ηy
L=L(b) admissible
k
X X X 
−k 1−k 1S(ρ;B) (L(n)) wn (L) =
i=1 p|Li (n) 1≤b≤2ηy
p<xρ L=qt+b∈L
/
p∤B
X X
= An (L)wn (L). (5.64)
1≤b1 <...<bk n∈A(x)
qbk ≤ηy
L=L(b) admissible
 CONSECUTIVE PRIMES IN SHORT INTERVALS 61

Let L = {L1 , . . . , Lk } and n be in the range of summation of S such

that An (L) > 0. Then the following statements hold.
1) The number of primes among L1 (n), . . . , Lk (n) is at least m + 1.
2) For any 1 ≤ i ≤ k, Li (n) has not a prime factor p such that p < xρ
and p ∤ B.
3) For any linear function L = qt + b ∈ / L, where b is an integer with
1 ≤ b ≤ 2ηy, L(n) has a prime factor p such that p < xρ and p ∤ B (we
choose ρ so that xρ is not an integer; therefore the statements p ≤ xρ
and p < xρ are equivalent). Since
L(n) > n ≥ x > xρ ,
we see that L(n) is not a prime number.
As a consequence we obtain the following statements.
i) None of n ∈ A(x) can make a positive contribution to S from two dif-
ferent admissible sets (since if n makes a positive contribution for some
admissible set L = {L1 , . . . , Lk }, then the numbers L1 (n), . . . , Lk (n)
are uniquely determined as the integers in [qn + 1, qn + 2ηy] with no
prime factors p such that p < xρ and p ∤ B).
ii) If L = {L1 , . . . , Lk } and n are in the range of summation of S
such that An (L) > 0, then there can be no primes in the interval
[qn + 1, qn + 2ηy] apart from possibly L1 (n), . . . , Lk (n), and so the
primes counted in this way must be consecutive.
Let L = {L1 , . . . , Lk } and n be in the range of summation of S such
that An (L) > 0. Let 1 ≤ i ≤ k. If p|Li (n) and p ∤ B, then p ≥ xρ . Let
Ω = {p : p|Li (n) and p ∤ B}.
We have Y
xρ#Ω ≤ p ≤ Li (n).
p∈Ω
Since
q ≤ y 1−ε ≤ y ≤ ln x,
a + qbi ≤ 2ηy ≤ ln x,
we have
Li (n) = qn + a + qbi ≤ n ln x + ln x ≤ 2x ln x + ln x ≤ x2 ,
if c(ε) is chosen large enough. Hence, ρ#Ω ≤ 2, i. e.
2
#Ω ≤ .
ρ
62 ARTYOM RADOMSKII

We have
Y Y
4= 4 = 4#Ω ≤ 42/ρ = e(2/ρ) ln 4 ≤ e4/ρ ,
p|Li (n) p∈Ω
p∤B
k
Y Y
4 ≤ e(4k)/ρ .
i=1 p|Li (n)
p∤B

Thus, if L = {L1 , . . . , Lk } and n are in the range of summation of S

such that An (L) > 0, then (see (5.1))
wn (L) ≤ C(ln R)2k e(4k)/ρ , (5.65)
where C > 0 is an absolute constant.
Let L = {L1 , . . . , Lk } be in the range of summation of S. We consider
X X k Xk X
e
S(L) = 1P (Li (n)) − m − k 1−
n∈A(x) i=1 i=1 p|Li (n)
p<xρ
p∤B
X 
−k 1S(ρ;B) (L(n)) wn (L) = S1 − S2 − S3 − S4 .
1≤b≤2ηy
L=qt+b∈L
/

e
We are going to obtain a lower bound for S(L). We write wn instead
of wn (L) for the brevity. Let 1 ≤ i ≤ k. Since #A(x) = x, we have
(see (5.3))
X
B k−1 ϕ(q)
1P (Li (n))wn ≥ 1 + o(1) k−1
SB (L) ·
ϕ(B) q
n∈A(x)
 Bk 
k+1 k−1
· #PLi ,A (x)(ln R) Jk + O SB (L)x(ln R) Ik .
ϕ(B)k
 CONSECUTIVE PRIMES IN SHORT INTERVALS 63

Hence,
k
X X k
X X
S1 = 1P (Li (n))wn = 1P (Li (n))wn ≥
n∈A(x) i=1 i=1 n∈A(x)

B k−1 ϕ(q)
≥ 1 + o(1) k−1
SB (L) (ln R)k+1 Jk ·
ϕ(B) q
Xk  Bk 
k−1
· #PLi ,A (x) + O k S
k B
(L)x(ln R) I k =
i=1
ϕ(B)
k

Bk k+1 ϕ(B) ϕ(q) X
= 1 + o(1) S B (L)(ln R) J k #PLi ,A (x)+
ϕ(B)k B q i=1
 Bk 
+o SB (L)x(ln R) Ik = S1′ + S1′′ ,
k
ϕ(B)k
since
k (ln x)1/5
0< ≤ → 0 as x → +∞.
ln R (1/9) ln x
We have shown (see (5.55)), if x ≥ c0 , where c0 > 0 is an absolute
constant, then for any L ∈ L
qx
#PL,A (x) ≥ .
4ϕ(q) ln x
We may assume that c(ε) ≥ c0 . Since ϕ(B)/B ≥ 1/2 (see (5.31)), we
obtain
k
ϕ(B) ϕ(q) X kx kx
#PLi ,A (x) ≥ = .
B q i=1 8 ln x 72 ln R
We have |o(1)| ≤ 1/2 in S1′ , if x ≥ c′ , where c′ > 0 is an absolute
constant. We may assume that c(ε) ≥ c′ . Since (see (5.8))
ln k
Jk ≥ c′′ Ik ,
k
where c′′ > 0 is an absolute constant, we obtain
c′′ B k
S1′ ≥ SB (L)x(ln R)k Ik ln k.
144 ϕ(B)k
We have
′′ c′′ B k k c′′ B k
|S1 | ≤ k
SB (L)x(ln R) Ik ≤ k
SB (L)x(ln R)k Ik ln k,
288 ϕ(B) 288 ϕ(B)
if c(ε) is chosen large enough. We obtain
c′′ B k
S1 ≥ SB (L)x(ln R)k Ik ln k =
288 ϕ(B)k
Bk
=c SB (L)x(ln R)k Ik ln k, (5.66)
ϕ(B)k
64 ARTYOM RADOMSKII

where c > 0 is an absolute constant.

We have (see (5.2))
X Bk
S2 = m wn = m(1 + o(1)) k
SB (L)x(ln R)k Ik ≥
ϕ(B)
n∈A(x)

m Bk
≥ k
SB (L)x(ln R)k Ik , (5.67)
2 ϕ(B)
if c(ε) is chosen large enough. Applying (5.5), we have
X X k X X k  X X

S3 = k wn = k 1 wn ≤
n∈A(x) i=1 p|Li (n) i=1 n∈A(x) p|Li (n)
p<xρ p<xρ
p∤B p∤B

Bk
≤ c2 ρ2 k 6 (ln k)2 SB (L)x(ln R)k Ik ,
ϕ(B)k
where c2 > 0 is an absolute constant. Let c3 > 0 be an absolute
constant such that
1 c3 1
c2 c23 ≤ and 3
≤ for any j ≥ 2.
12 j ln j 30
We choose a number ρ in the interval
h c c3 i
3
, (5.68)
2k 3 ln k k 3 ln k
so that xρ is not an integer. It is clear that ρ ≤ 1/30. Let us show that
the first inequality in (5.33) holds. It suffices to show that
k(ln ln x)2 c3 /2
≤ 3 .
ln x k ln k
This inequality is equivalent to
c3
k 4 ln k(ln ln x)2 ≤ ln x.
2
1/5
Since k ≤ (ln x) , we have
1 c3
k 4 ln k(ln ln x)2 ≤ (ln x)4/5 (ln ln x)3 ≤ ln x,
5 2
if c(ε) is chosen large enough. Thus, inequalities in (5.33) hold. We
have (see (5.67))
c23 6 2 B
k
S3 ≤ c 2 k (ln k) SB (L)x(ln R)k Ik ≤
k 6 (ln k)2 ϕ(B)k
1 Bk k m Bk
≤ S B (L)x(ln R) I k ≤ SB (L)x(ln R)k Ik ≤
12 ϕ(B)k 12 ϕ(B)k
1
≤ S2 . (5.69)
6
 CONSECUTIVE PRIMES IN SHORT INTERVALS 65

Now we estimate
X X
S4 = k 1S(ρ;B) (L(n))wn =
n∈A(x) 1≤b≤2ηy
L=qt+b∈L
/
X X
=k 1S(ρ;B) (L(n))wn .
1≤b≤2ηy n∈A(x)
L=qt+b∈L
/

Let b to be in the range of summation of S4 . Then L = qt + b ∈

/ L and
k
Y
k+1
∆L = q |(a + qbi ) − b| =
6 0.
i=1
1/5
Since 1 ≤ B ≤ x , we have (see (5.4))
X c4 ∆L B Bk
1S(ρ;B) (L(n))wn ≤ k
SB (L)x(ln R)k−1 Ik ,
ρ ϕ(∆L ) ϕ(B) ϕ(B)
n∈A(x)

where c4 > 0 is an absolute constant. Since B/ϕ(B) ≤ 2 and ρ lies in

the interval (5.68), we obtain
X 4c4 3 ∆L Bk
1S(ρ;B) (L(n))wn ≤ k ln k SB (L)x(ln R)k−1 Ik =
c3 ϕ(∆L ) ϕ(B)k
n∈A(x)

∆L Bk
= c5 k 3 ln k SB (L)x(ln R)k−1Ik .
ϕ(∆L ) ϕ(B)k
Hence,
Bk X ∆L
S4 ≤ c5 k 4 ln k SB (L)x(ln R)k−1 Ik . (5.70)
ϕ(B)k 1≤b≤2ηy
ϕ(∆L )
L=qt+b∈L
/

We put
c6 = 36Cc5 , (5.71)
where C > 0 is the absolute constant in Lemma 5.2. We put
1
η= . (5.72)
12c6 k 4 (ln k)2 ln ln(q + 2)
Let us show that
(ln x)−9/10 ≤ 2η ≤ 1. (5.73)
The second inequality in (5.73) is equivalent to the inequality
6c6 k 4 (ln k)2 ln ln(q + 2) ≥ 1.
e0 ≥ 3; therefore ln k ≥ 1. We have
We may assume that C
e4 ≥ 1,
6c6 k 4 (ln k)2 ln ln(q + 2) ≥ 6c6 (ln ln 3)k 4 ≥ 6c6 (ln ln 3)C 0
e0 is chosen large enough. The first inequality in (5.73) is equivalent
if C
to the inequality
6c6 k 4 (ln k)2 ln ln(q + 2) ≤ (ln x)9/10 .
66 ARTYOM RADOMSKII

Since q ≤ ln x and k ≤ (ln x)1/5 , we have

1
6c6 k 4 (ln k)2 ln ln(q + 2) ≤ 6c6 (ln x)4/5 (ln ln x)2 ln ln(ln x + 2) ≤
25
≤ (ln x)9/10 ,
if c(ε) is chosen large enough. Thus, (5.73) holds. We have x ≥ c, where
c is the absolute constant in Lemma 5.2, if c(ε) is chosen large enough.
Applying Lemma 5.2 and taking into account that ln(k + 1) ≤ 2 ln k,
we have
X ∆L X ∆L
≤ ≤ 4C ln ln(q + 2)(ln k)η ln x =
1≤b≤2ηy
ϕ(∆ L)
1≤b≤2η ln x
ϕ(∆L)
L=qt+b∈L
/ L=qt+b∈L
/

= 36C ln ln(q + 2)(ln k)η ln R.

Substituting this estimate into (5.70), we obtain (see also (5.71), (5.72)
and (5.67))
Bk
S4 ≤ 36Cc5 k 4 (ln k)2 SB (L)x(ln R)k Ik η ln ln(q + 2) = c6 k 4 (ln k)2 ·
ϕ(B)k
Bk k 1
· S B (L)x(ln R) I k ln ln(q + 2) =
ϕ(B)k 12c6 k 4 (ln k)2 ln ln(q + 2)
1 Bk k m Bk 1
= k
S B (L)x(ln R) I k ≤ k
SB (L)x(ln R)k Ik ≤ S2 .
12 ϕ(B) 12 ϕ(B) 6
(5.74)
From (5.69) and (5.74) we obtain
e 4
S(L) = S1 − S2 − S3 − S4 ≥ S1 − S2 .
3
We have (see (5.2))
X Bk
S2 = m wn = m(1 + o(1)) SB (L)x(ln R)k Ik ≤
ϕ(B)k
n∈A(x)

3 Bk
≤ m SB (L)x(ln R)k Ik ,
2 ϕ(B)k
if c(ε) is chosen large enough. Applying (5.66), where we replace c by
3c1 , we obtain
e Bk
S(L) ≥ SB (L)x(ln R)k Ik (3c1 ln k − 2m),
ϕ(B)k
where c1 > 0 is an absolute constant. We put

c=C
e e0 + 1 , (5.75)
c1
k = ⌈exp(ecm)⌉. (5.76)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 67

It is not hard to see that

k≥C e0 and 3c1 ln k − 2m ≥ m.
Since m is a positive integer, we see that
3c1 ln k − 2m ≥ 1.
Hence,
e Bk
S(L) ≥ SB (L)x(ln R)k Ik .
ϕ(B)k
Since B k /ϕ(B)k ≥ 1, ln R = (1/9) ln x, SB (L) ≥ exp(−c2 k) and Ik ≥
c3 (2k ln k)−k , where c2 and c3 are positive absolute constants (see (5.6)
and (5.7)), we obtain
e 1
S(L) ≥ k c3 (2k ln k)−k exp(−c2 k)x(ln x)k ≥ exp(−k 2 )x(ln x)k ,
9
e
if C0 is chosen large enough. We obtain
X
S= e
S(L) ≥
1≤b1 <...<bk
qbk ≤ηy
L=L(b) admissible
X
≥ exp(−k 2 )x(ln x)k 1=
1≤b1 <...<bk
qbk ≤ηy
L=L(b) admissible

= exp(−k 2 )x(ln x)k S ′ . (5.77)

Now we obtain a lower bound for S ′ . First let us show that
h ηy i
2≤k≤ /2. (5.78)
q
The first inequality obviously holds, since we may assume that C e0 ≥ 2.
To prove the second inequality it suffices to show that
ηy
2k ≤ . (5.79)
q
We have (see (5.62), (5.72))
ηy c4 y ε
≥ ηy ε = 4 ,
q k (ln k)2 ln ln(q + 2)
where c4 > 0 is an absolute constant. Thus, to prove (5.79) it suffices
to show that
2k 5 (ln k)2 ln ln(q + 2) ≤ c4 y ε .
In particular, from (5.62) it follows that q ≤ y. Applying (5.63), we
have
ε2
2k 5 (ln k)2 ln ln(q + 2) ≤ 2y 5ε/14 (ln y)2 ln ln(y + 2) ≤ c4 y ε ,
196
if c(ε) is chosen large enough. Thus, (5.78) is proved.
68 ARTYOM RADOMSKII

We put
n h ηy i o
Ω= 1≤n≤ : (n, p) = 1 ∀p ≤ k .
q
Applying Lemma 2.14, we have
h ηy i  [ηy/q]
#Ω = Φ , k ≥ c0 ,
q ln k
where c0 > 0 is an absolute constant. In particular, from (5.78) it
follows that ηy/q ≥ 4, and hence
h ηy i ηy ηy
≥ −1≥ .
q q 2q
We obtain
ηy
#Ω ≥ c5 , (5.80)
q ln k
where c5 > 0 is an absolute constant. Let us show that
ηy
c5 ≥ 2k. (5.81)
q ln k
Applying (5.62) and (5.72), we have
ηy c6 y ε
c5 ≥ 4 ,
q ln k k (ln k)3 ln ln(q + 2)
where c6 > 0 is an absolute constant. Therefore it suffices to show that
2k 5 (ln k)3 ln ln(q + 2) ≤ c6 y ε .
Applying (5.63) and taking into account that q ≤ y, we have
 3
5 3 5ε/14 ε
2k (ln k) ln ln(q + 2) ≤ 2y (ln y)3 ln ln(y + 2) ≤ c6 y ε ,
14
if c(ε) is chosen large enough. Thus, (5.81) is proved.
Let b1 < . . . < bk be positive integers from the set Ω. Let us show
that for any prime p with p ∤ q there is an integer mp such that mp 6≡
bi (mod p) for all 1 ≤ i ≤ k. Let p be a prime with p ∤ q. If p > k, then
the statement is obvious. If p ≤ k, then we may put mp = 0; from the
definition of the set Ω it follows that bi 6≡ 0 (mod p) for all 1 ≤ i ≤ k.
Thus, the statement is proved. By Lemma 5.1, L(b) is an admissible
set. Hence (see also Lemma 2.13, (5.80), (5.81) and (5.72)),
   ηy k
′ #Ω
S ≥ ≥ k −k (#Ω − k)k ≥ k −k c5 −k ≥
k q ln k
 c ηy k  y k
5
≥ k −k = k −k c6 =
2 q ln k q ln ln(q + 2)k 4 (ln k)3
 y k  c6 k
= ,
q ln ln(q + 2) k 5 (ln k)3
 CONSECUTIVE PRIMES IN SHORT INTERVALS 69

where c6 > 0 is an absolute constant. We have

 c6 k
≥ exp(−k 2 ),
k 5 (ln k)3
e0 is chosen large enough. Hence,
if C
 y k
S′ ≥ exp(−k 2 ).
q ln ln(q + 2)
Substituting this estimate into (5.77), we obtain
 y k
S ≥ exp(−2k 2 )x(ln x)k ≥
q ln ln(q + 2)
 y k
≥ exp(−2k 5 )x(ln x)k . (5.82)
q ln ln(q + 2)
Now we obtain an upper bound for S. Applying (5.64) and (5.65),
we have
X X
S≤ An (L)wn (L) ≤
1≤b1 <...<bk n∈A(x):
qbk ≤ηy An (L)>0
L=L(b) admissible
X X
≤ Ck(ln R)2k e(4k)/ρ 1.
1≤b1 <...<bk n∈A(x):
qbk ≤ηy An (L)>0
L=L(b) admissible

We have (see parts 1) – 3), i) and ii) above)

X X
1≤
1≤b1 <...<bk n∈A(x):
qbk ≤ηy An (L)>0
L=L(b) admissible

≤ #{x ≤ n < 2x : there are ≥ (m + 1) consecutive primes

all congruent to a (mod q) in the interval [qn + 1, qn + 2ηy]} ≤
≤ #{x ≤ n < 2x : there are ≥ (m + 1) consecutive primes
all congruent to a (mod q) in the interval [qn + 1, qn + y]} := N1 .
Hence,
S ≤ Ck(ln R)2k e(4k)/ρ N1 .
Since ρ lies in the interval (5.68), we have
4k 8k 4 ln k
≤ = c4 k 4 ln k,
ρ c3
where c4 > 0 is an absolute constant. Since ln R = (1/9) ln x, we have
k
Ck(ln R)2k e(4k)/ρ ≤ C 2k exp(c4 k 4 ln k)(ln x)2k ≤ exp(k 5 )(ln x)2k ,
9
e0 is chosen large enough. Hence,
if C
S ≤ exp(k 5 )(ln x)2k N1 . (5.83)
70 ARTYOM RADOMSKII

From (5.82) and (5.83) we obtain

 y k  1 k
N1 ≥ x exp(−3k 5 ). (5.84)
ln x q ln ln(q + 2)
We define
Ω1 = {x ≤ n ≤ 2x − 1 : there are ≥ (m + 1) consecutive primes
all congruent to a (mod q) in the interval [qn + 1, qn + y]};
Ω2 = {qx + 1 ≤ pn ≤ q(2x − 1) + y : pn ≡ . . . ≡ pn+m ≡ a (mod q),
pn+m − pn ≤ y}.
We put N2 = #Ω2 . Since x is a positive integer, we have N1 = #Ω1 .
Let us show that
N1 ≤ (⌈y⌉ + 1)N2 . (5.85)
Let n ∈ Ω1 . Then there are ≥ (m + 1) consecutive primes all congruent
to a (mod q) in the interval [qn + 1, qn + y]. Let p be the first of them.
Then p ∈ Ω2 . We put
Λ = {j ∈ Z : qj + 1 ≤ p ≤ qj + y}.
We claim that
#Λ ≤ ⌈y⌉ + 1. (5.86)
We put Ij = [qj + 1, qj + y], j ∈ Z. Since p ∈ In , we have Λ 6= ∅. Let
l be the minimal element in Λ. We put t = ⌈y⌉ + 1. Then t > y. We
have
q(l + t) + 1 > q(l + t) = ql + qt ≥ ql + t > ql + y ≥ p.
Hence, p ∈ / Ij for j ≥ l + t and j ≤ l − 1. We obtain #Λ ≤ t. Thus,
(5.86) is proved; (5.85) follows from (5.86). We have
⌈y⌉ + 1 ≤ y + 2 ≤ 2y,
if c(ε) is chosen large enough. Since
N2 ≤ #{qx + 1 ≤ pn ≤ 2qx + y : pn ≡ . . . ≡ pn+m ≡ a (mod q),
pn+m − pn ≤ y} =: N3 ,
we obtain (see (5.84))
1 x  y k  1 k
N3 ≥ exp(−3k 5 ). (5.87)
2 y ln x q ln ln(q + 2)
We put
N4 = #{qx < pn ≤ 2qx − 5q : pn ≡ . . . ≡ pn+m ≡ a (mod q),
pn+m − pn ≤ y}; (5.88)
N5 = #{2qx − 5q < pn ≤ 2qx + y : pn ≡ . . . ≡ pn+m ≡ a (mod q),
pn+m − pn ≤ y}.
Then
N3 = N4 + N5 . (5.89)
 CONSECUTIVE PRIMES IN SHORT INTERVALS 71

Since q ≤ y, we have
N5 ≤ 5q + [y] ≤ 5q + y ≤ 6y. (5.90)
Let us show that
1 x  y k  1 k
y≤ exp(−3k 5 ) := T1 . (5.91)
24 y ln x q ln ln(q + 2)
Since q ≤ y 1−ε ≤ y and k ≤ y ε/14 , we have
1 x yε k
T1 ≥ exp(−3y 5ε/14 ).
24 y ln x ln ln(y + 2)
Therefore to prove (5.91) it suffices to show that
1 x yε k
y≤ exp(−3y 5ε/14 ).
24 y ln x ln ln(y + 2)
Taking logarithms, we obtain
ln y ≤ − ln 24 + ln x − ln y + k(ε ln y − ln ln x − ln ln ln(y + 2)) − 3y 5ε/14
or, that is equivalent,
T2 := 2 ln y + ln 24 − εk ln y + k ln ln x + k ln ln ln(y + 2) + 3y 5ε/14 ≤ ln x.
Since y ≤ ln x, 0 < ε < 1, we have k ≤ (ln x)ε/14 ≤ (ln x)1/14 . We have
T2 ≤ 2 ln ln x + ln 24 + (ln x)1/14 ln ln x + (ln x)1/14 ln ln ln(ln x + 2)+
+ 3(ln x)5/14 ≤ ln x,
if c(ε) is chosen large enough. Thus, (5.91) is proved. From (5.90) and
(5.91) it follows that
1 x  y k  1 k
N5 ≤ exp(−3k 5 ). (5.92)
4 y ln x q ln ln(q + 2)
Applying (5.87), (5.89) and (5.92), we obtain
1 x  y k  1 k
N4 ≥ exp(−3k 5 ) =: T3 . (5.93)
4 y ln x q ln ln(q + 2)
We have (see (2.2))
2qx 2qx qx
π(2qx) ≤ c1 ≤ c1 = c2 ,
ln(2qx) ln x ln x
where c1 > 0 and c2 = 2c1 > 0 are absolute constants. We obtain
qx ln x 1 x  y k  1 k
T3 = exp(−3k 5 ) ≥
ln x qx 4 y ln x q ln ln(q + 2)
1  y  k−1 1
≥ π(2qx) k+1 k
exp(−3k 5 ).
4c2 ln x q (ln ln(q + 2))
Using the inequality ln(1 + x) ≤ x, x > 0, we obtain
ln ln(q + 2) ≤ ln(1 + q) ≤ q.
72 ARTYOM RADOMSKII

Hence,
1 1 1
≥ ≥ .
q k+1 (ln ln(q + 2))k q 2k+1 q 3k5
We have
5
4c2 ≤ 23k ,
e0 is chosen large enough. We have
if C
 y k−1 1
T3 ≥ π(2qx) .
ln x (2eq)3k5
We have
3k 5 ≤ k 6 ,
e0 is chosen large enough. Hence,
if C
1 1
5 ≥ .
(2eq) 3k (2eq)k6
We have
6 7
(2e)k ≤ 2k ,
e0 is chosen large enough. It is clear that
if C
6 7
qk ≤ qk .
We obtain
1 1
6 ≥ .
(2eq) k (2q)k7
Since (see (5.61))
y
0< ≤ 1,
ln x
we have  y k−1  y k7
≥ .
ln x ln x
We obtain  y k7
T3 ≥ π(2qx) .
2q ln x
We have (see (5.75), (5.76))
k = ⌈exp(e
cm)⌉ ≤ exp(e
cm) + 1 ≤ exp(2e
cm), (5.94)
e0 is chosen large enough. Therefore
if C
k 7 ≤ exp(14e
cm).
Since Ce0 is a positive absolute constant, we see from (5.75) that e
c is a
positive absolute constant. We have
 y exp(14ecm)  y exp(Cm)
T3 ≥ π(2qx) = π(2qx) , (5.95)
2q ln x 2q ln x
 CONSECUTIVE PRIMES IN SHORT INTERVALS 73

where C = 14e c > 0 is an absolute constant. From (5.88), (5.93) and

(5.95) we obtain
#{qx < pn ≤ 2qx − 5q : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm)
pn+m − pn ≤ y} ≥ π(2qx) .
2q ln x
Applying (5.94), we see that the inequality
k ≤ y ε/14
holds if
exp(2e cm) ≤ y ε/14 .
This inequality is equivalent to
ε
m≤ ln y = c · ε ln y,
28e c
where c = 1/(28e c) > 0 is an absolute constant. Let us denote c(ε) by
c0 (ε). Lemma 5.4 is proved.
Lemma 5.5. There are positive absolute constants c and C such that
the following holds. Let ε be a real number with 0 < ε < 1. Then there
is a number c0 (ε) > 0, depending only on ε, such that if x ∈ R, y ∈ R,
m ∈ Z, q ∈ Z, a ∈ Z are such that c0 (ε) ≤ y ≤ ln x, 1 ≤ m ≤ c · ε ln y,
1 ≤ q ≤ y 1−ε and (a, q) = 1, then
#{qx < pn ≤ 2qx : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y}
 exp(Cm)
y
≥ π(2qx) .
2q ln x
Proof of Lemma 5.5. Let c, C, c0 (ε) be the quantities in Lemma
5.4. We choose a quantity e c0 (ε) and an absolute constant C e later; they
will be large enough. Let e e
c0 (ε) ≥ c0 (ε) and C ≥ C. Let x ∈ R, y ∈ R,
m ∈ Z, q ∈ Z, a ∈ Z are such that e c0 (ε) ≤ y ≤ ln x, 1 ≤ m ≤ c · ε ln y,
1 ≤ q ≤ y 1−ε , (a, q) = 1. We put l = ⌈x⌉. Then, by Lemma 5.4, we
have
N1 = #{ql < pn ≤ 2ql − 5q : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 exp(Cm)
y
pn+m − pn ≤ y} ≥ π(2ql) =: T1 . (5.96)
2q ln l
Since x ≤ l < x + 1, we have
ql ≥ qx,
2ql − 5q ≤ 2q(x + 1) − 5q = 2qx − 3q < 2qx.
Therefore
N1 ≤ #{qx < pn ≤ 2qx : pn ≡ . . . ≡ pn+m ≡ a (mod q),
pn+m − pn ≤ y} =: N2 . (5.97)
74 ARTYOM RADOMSKII

We have
x + 1 ≤ x2 ,
if e
c0 (ε) is chosen large enough. Hence,
ln l ≤ ln(x + 1) ≤ 2 ln x.
Since
π(2ql) ≥ π(2qx),
we have
 exp(Cm)  exp(Cm)  exp(Cm)
y y 1
T1 ≥ π(2qx) = π(2qx) .
4q ln x q ln x 4
We have
e
2exp(Cm) ≤ exp(2Cm),e
e is chosen large enough. Since C
if C e ≥ C, we have
 exp(Cm)  exp(Cm) e  2 exp(Cm)
e  exp(2Cm)
e
1 1 1 1
≥ = ≥ .
4 4 2 2
Since
y
0< ≤ 1,
q ln x
we have
 exp(Cm)  exp(2Cm)
e
y y
≥ .
q ln x q ln x
Hence,
 exp(2Cm)
e
y
T1 ≥ π(2qx) . (5.98)
2q ln x
From (5.96), (5.97) and (5.98) we obtain
#{qx < pn ≤ 2qx : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 exp(2Cm) e
y
pn+m − pn ≤ y} ≥ π(2qx) .
2q ln x
Let us denote ec0 (ε) by c0 (ε) and 2C e by C. Lemma 5.5 is proved.
Let us complete the proof of Theorem 1.1. Let c0 (ε), c, C be the
quantities in Lemma 5.5. We choose a quantity e c0 (ε) and an absolute
e
constant C later; they will be large enough. Let e c0 (ε) ≥ c0 (ε) and
e
C ≥ C.
I) Let us prove the following statement.
Let ε be a real number with 0 < ε < 1. Let t ∈ R, y ∈ R, m ∈ Z,
q ∈ Z, a ∈ Z be such that
 t 
t ≥ 100, e c0 (ε) ≤ y ≤ ln ,
2 ln t
1 ≤ m ≤ c · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 75

Then
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y}
 y exp(Cm) e
≥ π(t) . (5.99)
2q ln t
Indeed, since t ≥ 100, we have 2 ln t ≥ 1. Hence,
 t 
y ≤ ln ≤ ln t.
2 ln t
We have
q ≤ y 1−ε ≤ y ≤ ln t.
Therefore  t   t
y ≤ ln ≤ ln .
2 ln t 2q
We put
t
x= .
2q
We have x ∈ R, y ∈ R, m ∈ Z, q ∈ Z, a ∈ Z,
c0 (ε) ≤ y ≤ ln x,
1 ≤ m ≤ c · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
By Lemma 5.5, we have
#{qx < pn ≤ 2qx : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y}
 exp(Cm)  exp(Cm)
e
y y
≥ π(2qx) ≥ π(2qx) ≥
2q ln x 2q ln x
 exp(Cm)
e
y
≥ π(2qx) .
2q ln(2qx)
Returning to the variable t, we obtain (5.99).
II) Let us prove the following statement.
Let ε be a real number with 0 < ε < 1. Let t ∈ R, m ∈ Z, q ∈ Z,
a ∈ Z be such that
ε  t 
t ≥ 100, ec0 ≤ ln ,
2 2 ln t
c
1 ≤ m ≤ · ε ln ln t, 1 ≤ q ≤ (ln t)1−ε , (a, q) = 1.
4
Then
n
# t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 t o  1 exp(Cm)
e
pn+m − pn ≤ ln ≥ π(t) .
2 ln t 4q
Proof. We need the following
76 ARTYOM RADOMSKII

Lemma 5.6. Let t be a real number with t ≥ 100. Then

√
1) 2 ln t ≤ t;
 t  1
2) ln ≥ ln t;
2 ln t 2
 t  1
3) ln ln ≥ ln ln t;
2 ln t 2
ln(2 ln t) 1
4) 1 − ≥ .
ln t 2
Proof of Lemma 5.6.
1) Let us consider the function
√
f (x) = x − 2 ln x.
Then
1 2 1 √
f ′ (x) = √ − = ( x − 4) > 0,
2 x x 2x
if x > 16. Hence, the function f (x) is increasing on the interval
(16, +∞). Since
f (100) = 0.78 . . . > 0,
we see that f (x) > 0 on [100, +∞). Hence, f (t) > 0. The inequality
1) is proved.
2) Taking into account that t ≥ 100 and applying the inequality 1),
we have
t √
≥ t.
2 ln t
Therefore  t  √ 1
ln ≥ ln t = ln t.
2 ln t 2
The inequality 2) is proved.
3) If z ≥ 4, then
z √
≥ z.
2
Since
ln t ≥ ln 100 = 4.605 . . . > 4,
we have
1 √
ln t ≥ ln t.
2
Applying the inequality 2), we have
 t  √
ln ≥ ln t.
2 ln t
Taking logarithms, we obtain
 t  √ 1
ln ln ≥ ln ln t = ln ln t.
2 ln t 2
The inequality 3) is proved.
 CONSECUTIVE PRIMES IN SHORT INTERVALS 77

4) Let us consider the function

x
f (x) = − ln(2x).
2
We have
1 1
f ′ (x) = − > 0,
2 x
if x > 2. Hence, the function f (x) is increasing on the interval (2, +∞).
Since
f (4.5) = 0.05 . . . > 0,
we see that f (x) > 0 on [4.5, +∞). Since
ln t ≥ ln 100 = 4.605 . . . > 4.5,
we have
ln t
− ln(2 ln t) ≥ 0.
2
We obtain
ln(2 ln t) 1
≤ .
ln t 2
Hence,
ln(2 ln t) 1
1− ≥ .
ln t 2
The inequality 4) is proved. Lemma 5.6 is proved.
We put  t 
y = ln .
2 ln t
Since t ≥ 100, we have (see Lemma 5.6, the inequality 3))
 t  1
ln y = ln ln ≥ ln ln t.
2 ln t 2
Therefore
ε
1 ≤ m ≤ c · ln y.
2
We may assume that
c0 (ε) ≥ 21/ε .
e
Since t ≥ 100, we have
t
≤ t.
2 ln t
We have ε  t 
c0
e ≤ ln ≤ ln t.
2 2 ln t
Hence,   ε 
t ≥ exp e c0 ≥ exp(22/ε ).
2
Therefore
1
(ln t)1−ε/2 ≥ (ln t)1−ε . (5.100)
2
78 ARTYOM RADOMSKII

From (5.100) and the inequality 4) of Lemma 5.6 we have

  1−ε/2  1−ε/2
1−ε/2 t  1−ε/2 ln(2 ln t)
y = ln = (ln t) 1− ≥
2 ln t ln t
 1−ε/2
1−ε/2 1 1
≥ (ln t) ≥ (ln t)1−ε/2 ≥ (ln t)1−ε .
2 2
Since
1 ≤ q ≤ (ln t)1−ε ,
we have
1 ≤ q ≤ y 1−ε/2 .
Applying the statement of part I) with ε/2 and the inequality 2) of
Lemma 5.6, we have
n
# t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 t o  ln(t/(2 ln t)) exp(Cm)
e
pn+m − pn ≤ ln ≥ π(t) ≥
2 ln t 2q ln t
 1 exp(Cm)e
≥ π(t) .
4q
The statement is proved.
III) Let ε be a real number with 0 < ε < 1. Let t ∈ R, y ∈ R, m ∈ Z,
q ∈ Z, a ∈ Z be such that
ε  t 
t ≥ 100, e c0 ≤ ln ≤ y ≤ ln t,
2 2 ln t
c
1 ≤ m ≤ · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
4
Then
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm)
e
pn+m − pn ≤ y} ≥ π(t) .
4q ln t
Proof. Since
y ≤ ln t,
we have
c
1 ≤ m ≤ · ε ln ln t, 1 ≤ q ≤ (ln t)1−ε .
4
 CONSECUTIVE PRIMES IN SHORT INTERVALS 79

Applying the statement of part II), we have

#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q), pn+m − pn ≤ y}
n
≥ # t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 t o  1 exp(Cm)
e
pn+m − pn ≤ ln ≥ π(t) ≥
2 ln t 4q
 y exp(Cm)
e
≥ π(t) .
4q ln t
The statement is proved.
IV) For 0 < ε < 1 we define the quantity t0 (ε) as follows:
1) t0 (ε) ≥ 100;
 t   ε 
2) ln ≥ max e
c0 ,e
c0 (ε) for all t ≥ t0 (ε).
2 ln t 2
Let us prove the following statement.
Let ε be a real number with 0 < ε < 1. Let t ∈ R, y ∈ R, m ∈ Z,
q ∈ Z, a ∈ Z be such that
 ε 
t ≥ t0 (ε), max e c0 ,e
c0 (ε) ≤ y ≤ ln t,
2
c
1 ≤ m ≤ · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
4
Then
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm) e
pn+m − pn ≤ y} ≥ π(t) .
4q ln t
Proof. Let us consider two cases.
 t 
1) ln < y ≤ ln t.
2 ln t
Then
ε  t 
t ≥ 100, ec0 ≤ ln ≤ y ≤ ln t,
2 2 ln t
c
1 ≤ m ≤ · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
4
Applying the statement of part III), we have
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm)
e
pn+m − pn ≤ y} ≥ π(t) .
4q ln t
Now we consider the second case:
 t 
2) y ≤ ln .
2 ln t
80 ARTYOM RADOMSKII

Then
 t 
t ≥ 100, e c0 (ε) ≤ y ≤ ln ,
2 ln t
1 ≤ m ≤ c · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
Applying the statement of part I), we have
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm) e  y exp(Cm)
e
pn+m − pn ≤ y} ≥ π(t) ≥ π(t) .
2q ln t 4q ln t
The statement is proved.
V) For 0 < ε < 1 we put
 ε 
ρ(ε) = max e c0 ,e
c0 (ε) + t0 (ε).
2
Let us prove the following statement.
Let ε be a real number with 0 < ε < 1. Let t ∈ R, y ∈ R, m ∈ Z,
q ∈ Z, a ∈ Z be such that
ρ(ε) ≤ y ≤ ln t,
c
1≤m≤ · ε ln y, 1 ≤ q ≤ y 1−ε , (a, q) = 1.
4
Then
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(2Cm)
e
pn+m − pn ≤ y} ≥ π(t) .
2q ln t
Proof. We have
 ε 
max e c0 ,e
c0 (ε) ≤ y ≤ ln t
2
and
t ≥ exp(ρ(ε)) ≥ ρ(ε) ≥ t0 (ε).
Applying the statement of part IV), we have
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm)e
pn+m − pn ≤ y} ≥ π(t) . (5.101)
4q ln t
We may assume that
Ce ≥ 2.
Therefore
e
exp(Cm) e ≥C
≥ Cm e ≥ 2.
Hence,
e
2 exp(Cm) ≤ exp(2Cm).e
 CONSECUTIVE PRIMES IN SHORT INTERVALS 81

We have
 exp(Cm)
e  2 exp(Cm)
e  exp(2Cm)
e
1 1 1
= ≥ .
4 2 2
Since
y
0< ≤ 1,
q ln t
we have
 y exp(Cm)
e  y exp(2Cm)
e
≥ .
q ln t q ln t
We obtain
 y exp(Cm)
e  y exp(2Cm)
e
≥ . (5.102)
4q ln t 2q ln t
From (5.101) and (5.102) we obtain
#{t/2 < pn ≤ t : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(2Cm)
e
pn+m − pn ≤ y} ≥ π(t) .
2q ln t
e
The statement is proved. Let us denote ρ(ε) by c0 (ε), c/4 by c and 2C
by C. Theorem 1.1 is proved.
Proof of Corollary 1.1. Let c0 (ε), c and C be the quantities
in Theorem 1.1. We put
 
2 1
C1 = max , c0 ,C .
c 2
Let m be a positive integer. Let x ∈ R, y ∈ R be such that
exp(C1 m) ≤ y ≤ ln x.
Then
1
y ≥ exp(C1 m) ≥ C1 m ≥ C1 ≥ c0 ,
2
2 
y ≥ exp(C1 m) ≥ exp m .
c
From the last inequality we obtain
1
m ≤ c · ln y.
2
We put q = 1, a = 1. We have
1
c0 ≤ y ≤ ln x,
2
1
1 ≤ m ≤ c · ln y, 1 ≤ q ≤ y 1/2 , (a, q) = 1.
2
82 ARTYOM RADOMSKII

Applying Theorem 1.1 with ε = 1/2, we have

#{x/2 < pn ≤ x : pn+m − pn ≤ y} =
= #{x/2 < pn ≤ x : pn ≡ . . . ≡ pn+m ≡ a (mod q),
 y exp(Cm)  y exp(Cm)
pn+m − pn ≤ y} ≥ π(x) = π(x) ≥
2q ln x 2 ln x
 y exp(C1 m)
≥ π(x) .
2 ln x
Let us denote C1 by C. Corollary 1.1 is proved.
6. Acknowledgements
The author is deeply grateful to Sergei Konyagin and Maxim Korolev
for their attention to this work and useful comments.
The author also expresses gratitude to Mikhail Gabdullin and Pavel
Grigor’ev for useful comments and suggestions.
References
[1] H. Davenport, Multiplicative number theory, volume 74 of Graduate Texts in
Mathematics. Springer-Verlag, New York, third edition, 2000. Revised and with
a preface by Hugh L. Montgomery.
[2] H. Halberstam and H. E. Richert, Sieve methods, L.M.S. monographs. Aca-
demic Press, 1974.
[3] R. R. Hall and G. Tenenbaum, Divisors, Cambridge Tracts in Mathematics,
90. Cambridge University Press, Cambridge, 1988.
[4] A. E. Ingham, The distribution of prime numbers, Cambridge Tracts in Math-
ematics, no. 30, Cambridge University Press, 1932.
[5] J. Maynard, Dense clusters of primes in subsets, Compositio Mathematica, 152
(2016), 1517–1554.
[6] K. Prachar, Primzahlverteilung, Berlin, Springer, 1957.
[7] I. M. Vinogradov, Elements of number theory, Translated from the Fifth Re-
vised Edition by Saul Kravetz, New York, Dover, 1954.

Steklov Mathematical Institute of Russian Academy of Sciences, 8

Gubkina St., Moscow 119991, Russia
Email address: [email protected]