A PAIR CORRELATION PROBLEM, AND COUNTING LATTICE

POINTS WITH THE ZETA FUNCTION

CHRISTOPH AISTLEITNER, DANIEL EL-BAZ, MARC MUNSCH

Dedicated to the memory of Professor Aleksandar Ivić (1949-2020)

Abstract. The pair correlation is a localized statistic for sequences in the unit interval.
arXiv:2009.08184v3 [math.NT] 15 Feb 2021

Pseudo-random behavior with respect to this statistic is called Poissonian behavior. The
metric theory of pair correlations of sequences of the form (an α)n≥1 has been pioneered
by Rudnick, Sarnak and Zaharescu. Here α is a real parameter, and (an )n≥1 is an integer
sequence, often of arithmetic origin. Recently, a general framework was developed which
gives criteria for Poissonian pair correlation of such sequences for almost every real number
α, in terms of the additive energy of the integer sequence (an )n≥1 . In the present paper we
develop a similar framework for the case when (an )n≥1 is a sequence of reals rather than
integers, thereby pursuing a line of research which was recently initiated by Rudnick and
Technau. As an application of our method, we prove that for every real number θ > 1, the
sequence (nθ α)n≥1 has Poissonian pair correlation for almost all α ∈ R.

1. Introduction and statement of results

A sequence (yn )n≥1 of real numbers is called uniformly distributed (or equidistributed) modulo
one if for all intervals A ⊂ [0, 1) the asymptotic equality
N
1 X
(1) lim 1A (yn ) = λ(A)
N →∞ N
n=1

holds. Here 1A is the indicator function of A, extended periodically with period 1, and λ
denotes Lebesgue measure. Uniform distribution theory has a long history, going back to
the seminal paper of Hermann Weyl [44]. For general background, see [15, 24]. Uniform
distribution of a sequence can be seen as a pseudo-randomness property, in the sense that
a sequence (Yn )n≥1 of independent, identically distributed random variables having uniform
distribution on [0, 1) satisfies (1) almost surely as a consequence of the Glivenko–Cantelli
theorem; thus a deterministic sequence (yn )n≥1 which is uniformly distributed mod 1 exhibits
the same behavior as a typical realization of a random sequence.

A sequence (yn )n≥1 is said to have Poissonian pair correlation if for all real numbers s ≥ 0,
1 X
lim 1[−s/N,s/N ] (yn − ym ) = 2s.
N →∞ N
1≤m,n≤N,
m6=n

2010 Mathematics Subject Classification. Primary 11K06, 11J83, 11M06; Secondary 11B05, 11J25, 11J71 .
Key words and phrases. Pair correlation, Riemann zeta function, lattice points, Diophantine inequality.
1
2 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

This notion is motivated by questions from theoretical physics, and plays a key role in the
Berry–Tabor conjecture; see [28] for more information. Just like equidistribution, Poisso-
nian pair correlation can be seen as a pseudo-randomness property, since a random sequence
(Yn )n≥1 as above almost surely has Poissonian pair correlation. However, clearly the two
properties are of a rather different nature. While equidistribution is a “large-scale” statistic
(where the test interval always remains the same), the pair correlation is a highly localized
statistic (where the size of the test interval shrinks in proportion with N ). Note that the two
properties are not independent: it is known that a sequence having Poissonian pair correlation
necessarily must be equidistributed [3, 17, 29], whereas the opposite implication is generally
false. An illustrative example is the sequence (nα)n≥1 , which is equidistributed if and only
if α 6∈ Q, but which fails to have Poissonian pair correlation for any α (see [25] for a more
general result along those lines).

The theory of uniform distribution modulo one can be said to be relatively well understood
(at least in the one-dimensional case). Many specific sequences are known which are uni-
formly distributed mod one. In contrast, only very few specific
√ results are known in the pair
correlation setting. A notable exception is the sequence ( n)n∈Z≥1 \ , which is known to have
Poissonian pair correlation [16]. The sequence (n2 α)n≥1 is conjectured to have Poissonian pair
correlation under mild Diophantine assumptions on α, but only partial results are known in
this direction [20, 30, 34, 43]. Lacking specific examples, it is natural to turn to a metric the-
ory instead. Let (an )n≥1 be a sequence of distinct integers, let α ∈ R, and consider sequences
of the form (an α)n≥1 . The metric theory of such sequences with respect to equidistribution
is very simple: for every such (an )n , the sequence (an α)n is uniformly distributed mod 1
for almost all α [44]. The situation with respect to pair correlation is much more delicate.
Pioneering work in this area was carried out by Rudnick, Sarnak and Zaharescu [33, 36]. As
noted above, (nα)n≥1 does not have Poissonian pair correlation for any α. However, for any
polynomial p ∈ Z[X] of degree at least 2, the pair correlation of (p(n)α)n is Poissonian for
almost all α. For related results, see for example [6, 12, 37].

Recently, a simple criterion was established in [5] which allows to decide whether the sequence
(an α)n has Poissonian pair correlation for almost all α for many naturally arising integer
sequences (an )n . Let EN denote the number of solutions (n1 , n2 , n3 , n4 ) of the equation
(2) an1 − an2 + an3 − an4 = 0,
subject to 1 ≤ n1 , n2 , n3 , n4 ≤ N . This quantity is called the additive energy in the additive
combinatorics literature (see [18, 40]). Note that N 2 ≤ EN ≤ N 3 for every (an )n and every
N . The criterion is as follows. If a sequence (an )n satisfies EN ≪ N 3−ε for some ε > 0, then
(an α)n has Poissonian pair correlation for almost all α. If in contrast EN ≫ N 3 , then the
conclusion fails to be true. For further refinements of this criterion, and for remaining open
problems, see [4, 7, 8, 26]. We emphasize that all that was written in this paragraph requires
(an )n to be a sequence of integers.

Very little is known in the metric theory of pair correlation of sequences (xn α)n when (xn )n is a
sequence of reals rather than integers. One step in this general direction is [12], where (xn )n is
allowed to take rational values and the results obtained depend on the size of the denominators
of these rationals. A general result was obtained recently in [35], where the authors gave a
criterion formulated in terms of the number of solutions of a certain Diophantine inequality.
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 3

The criterion is as follows: for a sequence (xn )n , assume that there exist some ε > 0 and
δ > 0 such that the number of integer solutions (n1 , n2 , n3 , n4 , j1 , j2 ) of the equation
(3) |j1 (xn1 − xn2 ) − j2 (xn3 − xn4 )| < N ε ,
subject to 1 ≤ |j1 |, |j2 | ≤ N 1+ε , 1 ≤ n1 , n2 , n3 , n4 ≤ N, n1 6= n2 , n3 6= n4 , is of order
≪ N 4−δ , then (xn α)n has Poissonian pair correlation for almost all α. It is verified in [35]
that this condition is satisfied for lacunary sequences. A condition in the spirit of (3) arises
very naturally when studying this sort of problem (cf. also [36]); we will encounter a variant
of this condition in Equation (16) below. In particular, it is very natural that in the integer
case one has to count solutions of Diophantine equations, while in the real-number setting
one has to count solutions of Diophantine inequalities. The problem with (3) is that it is in
general rather difficult to verify whether this condition is satisfied for a given sequence or
not, with issues being caused in particular by the presence of the coefficients j1 and j2 . The
purpose of the present paper is to give a simplified criterion, in the spirit of the criterion of
[5] which was specified in terms of the number of solutions of the equation (2).

Theorem 1. Let (xn )n≥1 be a sequence of positive real numbers for which there exists a
constant c > 0 such that xn+1 − xn ≥ c, n ≥ 1. Let EN ∗ denote the number of solutions

(n1 , n2 , n3 , n4 ) of the inequality

(4) |xn1 − xn2 + xn3 − xn4 | < 1,
subject to ni ≤ N, i = 1, 2, 3, 4. Assume that there exists some δ > 0 such that EN∗ ≪

N 183/76−δ as N → ∞. Then the sequence (xn α)n≥1 has Poissonian pair correlation for al-
most all α ∈ R.

The exponent 183/76 ≈ 2.408 in the conclusion of the theorem comes from a bound for the
178/13-th moment of the Riemann zeta function on the critical line due to Ivic [23], building
on earlier work of Heath-Brown [19]. Conditionally under the Lindelöf hypothesis, our bound
for EN∗ can be relaxed to E ∗ ≪ N 3−ε for any ε > 0, which would be in accordance with the
N
results known for the integer case.

Theorem 1 applies, for example, to all sequences of the form xn = p(n), n ≥ 1, where p is a
quadratic polynomial with real coefficients. For such a sequence (xn )n we have EN ∗ ≪ N 2+ε

for any ε > 0 by Lemma 5.2 of [11]. Theorem 1 also applies to xn = p(n) for every polyno-
mial p ∈ R[X] of degree d ≥ 3, under the additional assumption that the coefficient of xd−1
is rational1; the required bound for EN ∗ then follows, after eliminating this coefficient, from

Lemma 7 below (with the choice of θ = d and γ = N d−2 ). The extra assumption on the sec-
ond coefficient is most likely redundant, but we have not been able to establish the necessary
bound for EN ∗ without it. A famous open conjecture in additive combinatorics asserts that

EN ≪ N 2+ε for all convex sequences (xn )n , which would provide many further applications
of our theorem; however, unfortunately the best current bound in this direction (Shkredov’s
32/13 ≈ 2.46 from [39]) is just beyond the range of applicability of our theorem.

Bounding the number of solutions of (4) is necessary to control the variance of the pair
correlation function. When carefully reading the proof of Theorem 1 it becomes visible
1This could be relaxed to assuming some Diophantine condition on this coefficient.
4 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

that not all solutions of (4) contribute equally to the variance, but that rather a 4-tuple
(n1 , n2 , n3 , n4 ) with xn1 − xn2 + xn3 − xn4 = γ for some γ ∈ (−1, 1) has a stronger effect on
the variance the smaller the absolute value of γ is. This suggests to consider the quantity
∗ , which is defined as the number of solutions (n , n , n , n ) of the inequality
EN,γ 1 2 3 4

|xn1 − xn2 + xn3 − xn4 | < γ,

for γ ∈ (0, 1] and subject to ni ≤ N, i = 1, 2, 3, 4. Very informally speaking, one might expect
∗
that EN,γ ∗
scales as EN,γ ≈ γEN ∗ for a “randomly behaved” real sequence (x ) , except for the
n n
contribution of the trivial solutions n1 = n2 and n3 = n4 which always is of order N 2 . The
following theorem states that being able to control EN,γ ∗ as a function of γ indeed allows us
to deduce metric pair correlations in some cases where the condition on the additive energy
in Theorem 1 fails to hold.

Theorem 2. Let (xn )n≥1 be a sequence of positive real numbers for which there exists c > 0
such that xn+1 − xn ≥ c, n ≥ 1. Assume that there exists some δ > 0 such that for all η > 0
we have

∗
(5) EN,γ ≪η,δ N 2+η + γN 3−δ

as N → ∞, uniformly for γ ∈ (0, 1]. Then the sequence (xn α)n≥1 has Poissonian pair corre-
lation for almost all α ∈ R.

Theorem 2 is tailor-made for an application to the sequence xn = nθ . For that sequence,

equation (5) holds provided θ > 1 as a consequence of Lemma 7 below. Since that particular
problem was a key driving force for writing the present paper, we formulate this conclusion
as a theorem rather than just as a corollary.

Theorem 3. For every real number θ > 1 and almost every α ∈ R, the sequence (nθ α)n≥1
has Poissonian pair correlation.

As noted above, the conclusion of Theorem 3 is not true when θ = 1. It seems plausible that
the conclusion of the theorem is valid again for 0 < θ < 1. However, this cannot be proved
with the methods used in the present paper, which break down in the case of a sequence (xn )n
whose order of growth is only linear or even slower. We will address this aspect at the very
end of the paper, where we also formulate some further open problems.

In conclusion we note that Technau and Yesha recently obtained a result which is somewhat
similar to our Theorem 3, but which is “metric” in the exponent rather than in a multiplicative
parameter. More precisely, they showed that (nθ )n has Poissonian pair correlation for almost
all θ > 7. Their paper also contains similar results on higher correlations, which require a
larger value of θ. From a technical perspective, their problem is rather different from ours.
For details see their paper [42].
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 5

2. Preliminaries
As in the introduction, let 1[−s/N,s/N ] (x) denote the indicator function of the interval [−s/N, s/N ],
extended with period 1. That is,

1 if x − hxi ∈ [−s/N, s/N ],
1[−s/N,s/N ] (x) =
0 otherwise,
where hxi denotes the nearest integer to x. We wish to show that under the assumptions of
Theorem 1 we have for almost all α ∈ R
1 X
(6) 1[−s/N,s/N ] (xm α − xn α) → 2s
N
1≤m,n≤N,
m6=n

as N → ∞ for all s ≥ 0. It is well-known that for any s and N , and for any positive integer
+ −
K there exist trigonometric polynomials fK,s,N (x) and fK,s,N (x) of degree at most K such
that
− +
(7) fK,s,N (x) ≤ 1[−s/N,s/N ] (x) ≤ fK,s,N (x)
for all x, and such that
Z 1
± 1
(8) fK,s,N (x) dx = 2s/N ± .
0 K +1
−
Furthermore, the j-th Fourier coefficient cj of fK,s,N satisfies
 
2s 1 1
(9) |cj | ≤ min , +
N π|j| K +1
+
for all j, and an analogous bound holds for the Fourier coefficients of fK,s,N . These trigono-
metric polynomials are called Selberg polynomials, and their construction is described in
detail in Chapter 1 of [31].

Instead of establishing the required convergence relation (6) for indicator functions, we will
+ −
rather work with the trigonometric polynomials fK,s,N and fK,s,N instead, which is technically
more convenient. More precisely, in order to obtain (6) it suffices to prove the following. For
every fixed positive integer r, and for every fixed real number s ≥ 0, we have
X Z 1
1 + +
(10) frN,s,N (xm α − xn α) ∼ N frN,s,N (x)dx
N 0
1≤m,n≤N,
m6=n

as N → ∞, for almost all α ∈ R, and the same is true when f + is replaced by f − . The
desired result for indicator functions then follows from (7) and (8) and letting r → ∞ (see
[33, 35] for more details).
To establish (10) we prove that the “expected value” (with respect to α) of the left-hand side
is asymptotic to the right-hand side, and that the “variance” of the left-hand side of (10)
is not too large. An application of Chebyshev’s inequality together with the Borel–Cantelli
lemma then gives the desired result. As usual in such problems, controlling the expectation
is easier than controlling the variance.
6 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

We will obtain the required bound for the expectation in Section 3, and the bound for the
variance in Sections 4 and 5. In Section 6 we conclude the proof of Theorem 1. Section 7
contains all of the necessary modifications for the proof of Theorem 2, and in Section 8 we
show that the sequence (nθ )n indeed allows an application of Theorem 2. Finally, in Section
9 we discuss limitations of our method, and outline open problems and directions for future
research.

3. Proof of Theorem 1: Controlling expectations

Throughout the argument we assume that a positive integer r and a positive real s are fixed.
+
We write fN for the function frN,s,N , as defined in the previous section (or for the function
−
frN,s,N — both cases work in exactly the same way). We want to control the “expected value”
with respect to α of the left-hand side of (10) as N → ∞. In the case when (xn )n≥1 is an
integer sequence everything is periodic with period 1, and it is appropriate to integrate over
α ∈ [0, 1] with respect to the Lebesgue measure. In our case, when (xn )n≥1 is a sequence of
reals, we do not have such periodicity. We thus have to integrate over all α ∈ R with respect
to an appropriate measure µ, which is absolutely continuous with respect to the Lebesgue
measure, so that a µ-almost everywhere conclusion implies a Lebesgue-almost everywhere
conclusion. A good choice for the measure µ is the measure whose density with respect to
the Lebesgue measure is given by
2(sin(x/2))2
(11) dµ(x) = dx.
πx2
2
The Fourier transform of x 7→ 2(sin(x/2))
πx2
is a non-negative real function which is supported
√
on the interval (−1, 1), and which is uniformly bounded by 1/ 2π. Note that the measure µ
is normalized such that µ(R) = 1.

Expanding the function fN (x) into a Fourier series

X
cj e2πijx ,
j∈Z

by construction we have cj = 0 when |j| > rN , and |cj | ≤ 2s/N + 1/(rN ) ≪ N −1 for all
R1
j (recall that r and s are assumed to be fixed). Moreover we have c0 = 0 fN (x)dx. Using
the fact that the Fourier transform of the measure µ is supported on (−1, 1) and uniformly
bounded, we obtain
 
 
Z Z 
 1 X 1 
 fN (xm α − xn α) dµ(α) − N fN (x)dx
 N
 R 1≤m,n≤N, 0 
 m6=n

 Z 1
N (N − 1)
≪ N− fN (x)dx
N 0
| {z }| {z }
=1 ≪N −1
 
 
Z 
1 X  X 
+ |cj |  e2πij(x m α−x n α)
dµ(α)
N  R 1≤m,n≤N, 
1≤|j|≤rN  
m6=n
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 7
X X

≪ N −1 + N −2 1 |j(xm − xn )| < 1
1≤|j|≤rN 1≤m,n≤N,
m6=n
| {z }
≪N due to the growth assumption on (xn )n≥1

(12) ≪ N −1 ,
where we estimated |cj | using (9). Thus we have
Z X Z 1
1
(13) fN (xm α − xn α) dµ(α) = N fN (x)dx + O(1/N ),
R N 1≤m,n≤N, 0
m6=n

as desired.
Controlling the variances is more difficult, and will be done in the next two sections.

4. Proof of Theorem 1: Controlling variances

We keep the setup as in Section 3 above, that is, we assume that r and s are fixed, and we
+ −
write fN for either frN,s,N or frN,s,N . Furthermore, we write hN for the centered version of
fN , that is, for the function
Z 1 X
(14) hN (x) = fN (x) − f (x) dx = cj e2πijx .
0 j∈Z,
j6=0

We wish to estimate the “variance” of our localized counting function, or more precisely the
quantity
 2
Z X
1 
(15) Var(hN , µ) :=  hN (xm α − xn α)
N  dµ(α).
R 1≤m,n≤N,
m6=n

The following bound on Var(hN , µ) is the crucial ingredient in our proof of Theorem 1.
Lemma 1. For every ε > 0 we have, as N → ∞,
 
Var(hN , µ) ≪ max N −ε/8 + N −183/89+3ε (EN
∗ 76/89
) ∗
, EN N −2.49+4ε .

For the convenience of the reader, we note at this point that our assumption that there is
some δ > 0 such that EN ∗ ≪ N 183/76−δ ensures together with Lemma 1 that there is some
′
δ′ > 0 such that Var(hN , µ) ≪ N −δ , which is sufficient to deduce Theorem 1 (see Section 6
for details). We also note that conditionally under the Lindelöf hypothesis the bound which
follows from our method is Var(hN , µ) ≪ N −3+ε EN ∗ .

5. Proof of Lemma 1: Lattice point counting via the Riemann zeta function
5.1. A first reduction. Squaring out in (15) and using again the properties of the Fourier
transform of the measure µ, we can bound Var(hN , µ) by
Z X X
1
2
|cj1 cj2 | e2πiα(j1 (xn1 −xn2 )−j2 (xn3 −xn4 )) dµ(α)
R N | {z }
1≤n1 ,n2 ,n3 ,n4 ≤N, j1 ,j2 ∈Z, j1 ,j2 6=0
n1 6=n2 , n3 6=n4 |j1 |,|j2 |≤rN ≪N −2
8 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

1 X X

≪ 1 |j1 (xn1 − xn2 ) − j2 (xn3 − xn4 )| < 1 ,
N4
1≤n1 ,n2 ,n3 ,n4 ≤N, 1≤j1 ,j2 ≤rN
n1 >n2 , n3 >n4

thereby essentially arriving at (3). For technical reasons, in this paper we prefer to localize
the variables j1 , j2 into dyadic regions and thus apply the Cauchy–Schwarz inequality to (15).
To simplify later formulas we also replace the differences xn1 − xn2 and xn3 − xn4 by their
respective absolute values using the parity of hN . Then, writing U for the smallest integer
for which 2U ≥ rN , we can bound Var(hN , µ) by
 2
Z U
1 X X X 
 cj e2πij|xm −xn |α 
N  dµ(α)
R u=1 1≤m,n≤N, 2
u−1 u
≤|j|<2
m6=n
 2
!  
Z XU XU  
1  X X 
≪ 1  cj e2πij|xm −xn |α 
N2   dµ(α)
R k=1 u=1 1≤m,n≤N, 2u−1 ≤|j|<2u 

m6=n
U
log N X X X   
 
(16) ≪ 1 j1 |xn1 − xn2 | − j2 |xn3 − xn4 | < 1 .
N4
u=1 1≤n1 ,n2 ,n3 ,n4 ≤N, 2u−1 ≤j1 ,j2 <2u
n1 6=n2 , n3 6=n4

Thus we have reduced the problem of estimating the variance to a problem of bounding the
number of solutions of a Diophantine inequality.

5.2. Counting solutions by using the Riemann zeta function. We will relate the count-
ing problem in Equation (16) to the problem of bounding a twisted moment of the Riemann
zeta function. Before we return to the proof, we point out the difference between the real-
number case (in this paper) and the corresponding results for the case of (xn )n≥1 being an
integer sequence. In the integer case, the problem of estimating the variance of the pair cor-
relation function can be reduced to counting solutions of j1 (xn1 − xn2 ) = j2 (xn3 − xn4 ). Note
that this is in accordance with the situation in the present paper, where we count solutions to
|j1 (xn1 − xn2 ) − j2 (xn3 − xn4 )| < 1, with the difference that in the integer case “< 1” implies
“= 0”. The number of solutions of the counting problem in the integer case is essentially
governed by what is called a GCD sum. It is known that such sums have a connection with
the Riemann zeta function (see [1, 21]), and strong estimates for such sums were obtained in
[2, 9, 14]. Our approach below is motivated by a beautiful argument of Lewko and Radziwill
[27], who showed how the relevant GCD sum can be estimated in terms of a twisted moment
of a random model of the Riemann zeta function on the critical line.2 The randomization
was crucial in their argument for different reasons, one being that the required distributional
estimates for extreme values of the actual Riemann zeta function are not known uncondition-
ally. Their argument relied crucially on the fundamental theorem of arithmetic, and thus on
the fact that they were dealing with integer sequences. In the real-number case the situation
is much more delicate. We will relate our counting problem to a convolution formula for
2See also [13] for links between twisted moments of character sums and GCD sums. Furthermore, see [38]
for a very recent paper of Shkredov, where he applies GCD sums and methods from [27] to give upper bounds
for the maximal length of arithmetic progressions contained in sets with small product set.
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 9

the Riemann zeta function. The kernel will be chosen for its good properties with respect
to the Fourier transform (positivity and localized support) which allow to overcount without
substantial loss. To summarize, in our argument below we will use of a combination of ideas
from [1, 9, 10, 14] and [27].

Let (xn )n≥1 be the sequence from the statement of Theorem 1. Let M = N 2 − N , and let
{z1 , . . . , zM } be the multi-set of all absolute differences {|xm − xn | : 1 ≤ m, n ≤ N, m 6= n},
meaning that we allow repetitions in the definition. For a given positive integer u with
2u ≤ 2rN , we wish to estimate
X X

(17) 1 |j1 zm − j2 zn | < 1 .
1≤m,n≤M 2u−1 ≤j1 ,j2 <2u
2
−t /2 , and note that
We write ζ(σ + it) for the Riemann zeta function. We also write√ Φ(t) = e
this function has a positive Fourier transform given by Φ b = 2πΦ. Throughout the proof
ε > 0 is a small constant, and we take it for granted that N is “large”.

Our argument proceeds by splitting into two cases depending on the size of min{zm , zn }. We
first treat the case when zm , zn are both at least of size N 1.01 . We then treat the case when
one of zm or zn is “small”, which because of our dyadic splitting essentially amounts to saying
that both variables are small.

• Case 1: Counting solutions for zm , zn ≥ N 1.01 .

Let u be given such that 2u−1 ≤ j1 , j2 ≤ 2u . Set T = 2u N 1+ε/2 . For any integer k ≥ N 1.01 ,
we set
XM


(18) bk = 1 zm ∈ [k, k + 1) ,
m=1

while for k < N 1.01 we set bk = 0. Clearly we have

X∞
bk ≤ M = N 2 − N.
k=1
We split the interval [1, ∞) into a disjoint union of intervals Ih for h ≥ 0, where
"&  ' &  '!
1 h 1 h+1
Ih = 1+ , 1+ ,
T T
and we set
 1/2
X
(19) ah =  b2k  , h ≥ 0.
k∈Ih

Finally, we define a function

∞  
X 1 iht
(20) P (t) = ah 1 + .
T
h=0
The following four lemmas correspond to the key steps in our Case 1 analysis.
10 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

Lemma 2 (Controlling the square-integral of P in terms of the additive energy). We have

Z
(21) |P (t)|2 Φ(t/T )dt ≪ T EN
∗
.
R
Lemma 3 (Counting solutions of Diophantine inequalities in terms of ah ). We have
X X
X X
1 |j1 zm − j2 zn | < 1 ≪ ah1 ah2 .
2u−1 ≤j1 ,j2 <2u 1≤m,n≤M 2u−1 ≤j1 ,j2 ≤2u  h1 ,h2 ≥0, 
h −h j 
(1+ T1 ) 1 2 − j2 ≤ T4

1

The next step, in Lemma 4 below, is to relate thePsum on the right-hand side of the equa-
tion above to a complex integral, where the term j1 ,j2≥1 (j1 j2 )−1/2 (j1 /j2 )it can be roughly
interpreted as |ζ(1/2 + it)|2 . However, instead of simply using a truncated expression or an
approximate functional equation for ζ(1/2 + it), we will rather use of a convolution formula
to improve the analysis near t = 0. To do so, we introduce the function K defined by
sin2 ((1 + ε/4)u log N )
K(u) := ,
πu2 (1 + ε/4)(log N )
whose Fourier transform is given by
 
b |ξ|
K(ξ) = max 1 − ,0 .
2(1 + ε/4) log N
A similar idea was also fruitfully used in a paper of Bondarenko and Seip [10]. The function K
b 2(log rN )
is chosen in such a way that we have K(log j1 j2 ) ≫ 1 − 2(1+ε/4) log N ≫ 1 (where we suppress
the dependence on the constants ε and r).

Lemma 4 (Counting solutions of the Diophantine inequality by complex integration). We

have
Z X b  
X X 2u K(log j1 j2 ) j1 it
ah1 ah2 ≪ 1/2
|P (t)|2 Φ(t/T )dt.
T R (j j
1 2 ) j2
u−12 u ≤j1 ,j2 <2  h1 ≥0,h2 ≥0,  j1 ,j2 ≥1
h −h j 
(1+ T1 ) 1 2 − j2 ≤ T4

1

Lemma 5 (Estimating the complex integral). We have

Z X b  
2u K(log j1 j2 ) j1 it
(22) |P (t)|2 Φ(t/T )dt ≪ N 4−ε/4 + N 173/89+2ε (EN
∗ 76/89
) .
T R (j1 j2 )1/2
j1 ,j2 ≥1
j2

We now prove these four lemmas.

Proof of Lemma 2. We have
Z Z X  
1 (h1 −h2 )it
|P (t)|2 Φ(t/T )dt = ah1 ah2 1 + Φ(t/T )dt
R R h ,h ≥0 T
1 2
X Z    
1
=T ah1 ah2 exp log 1 + T (h1 − h2 )iy Φ(y)dy
R T
h1 ,h2 ≥0
X    
b 1
=T ah1 ah2 Φ log 1 + T (h1 − h2 )
T
h1 ,h2 ≥0
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 11

X  
b h1 − h2
(23) ≪T ah1 ah2 Φ
2
h1 ,h2 ≥0
X∞
(24) ≪T a2h
h=0
X∞ M
X
≪T b2k ≪ T ∗
1 = T EN .
k=1 m,n=1
|zm −zn |<1

Here we used that T log(1 + 1/T ) ≥ 1/2 for sufficiently large N (note that large N implies
b to pass from
large T ), and the Cauchy–Schwarz inequality together with the rapid decay of Φ
(23) to (24). We will further comment on the construction of P (t) at the very end of our
Case 1 analysis. 

Proof of Lemma 3. Let j1 and j2 be fixed, and assume without loss of generality that j1 ≥ j2 .
Let k ≥ N 1.01 be an integer in Ih1 , and assume that zm ∈ [k, k + 1). Then the inequality
|j1 zm − j2 zn | < 1 is only possible when
  
 j1 k 
 
(25)  j2 − zn  < 4

(recall that j1 /j2 ≤ 2 because j1 , j2 are located in the same dyadic interval). We write
ℓ(k) = ⌈j1 k/j2 ⌉. Recall that j1 /j2 ≥ 1 by assumption, so the mapping k 7→ ℓ(k) is injective.
Thus we have
X

1 |j1 zm − j2 zn | < 1
zm ∈Ih1 ,zn ∈Ih2
X X X
≪ 1
k∈Ih1 zm ∈[k,k+1) zn ∈Ih2 ,
|ℓ(k)−zn |<4
X X X X
≪ 1
k∈Ih1 zm ∈[k,k+1) −4≤v≤3 zn ∈Ih2 ,
zn ∈[ℓ(k)+v,ℓ(k)+v+1)
X X
≪ bk bℓ(k)+v
−4≤v≤3 k∈Ih1 such that
ℓ(k)+v∈Ih2
 1/2  1/2
X X
≪ b2k   b2ℓ 
k∈Ih1 ℓ∈Ih2

(26) ≪ ah 1 ah 2
by Cauchy–Schwarz.

When j1 and j2 are fixed, there can only be solutions of |j1 zm − j2 zn | < 1 with zm ∈ Ih1
and zn ∈ Ih2 for particular pairs (h1 , h2 ). Assume that zm ∈ Ih1 and
 zn ∈ Ih2 such that
 
|j1 zm − j2 zn | < 1. Recall that j1 ≥ j2 by assumption, so we have  zzmn
− jj21  < j11zn and
12 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

consequently zzm
n
≤ jj21 + j11zn ≤ 2. Since zm ∈ Ih1 and zn ∈ Ih2 , the quotient zm /zn is
somewhere between (1 + 1/T )h1 −h2 −1 and (1 + 1/T )h1 −h2 +1 , so that
zm 3
(27) ≤ (1 + 1/T )h1 −h2 (1 + 1/T ) ≤ (1 + 1/T )h1 −h2 + .
zn | {z } T
≤2(1+1/T )≤3, since zm /zn ≤ 2

Similarly
zm 3
(28) ≥ (1 + 1/T )h1 −h2 − .
zn T
 
 zm j 2 
Since j1 ≥ 2 u−1 and zn ≥ N 1.01 by assumption, we have  zn − j1  ≤ 2u−11N 1.01 ≤ T1 , where
the last inequality follows from our choice of T . Overall, together with (27) and (28) this
shows that the inequality |j1 zm − j2 zn | < 1 for zm ∈ Ih1 and zn ∈ Ih2 is only possible when
  
 1 h1 −h2 j2  4

 1+ − ≤ .
 T j1  T
Note that, for fixed j1 , j2 , this is an inequality which only depends on h1 , h2 and not on zm , zn
anymore. Thus in combination with (26) we obtain
X
X
1 |j1 zm − j2 zn | < 1 ≪ ah1 ah2
1≤m,n≤M h1 ,h2 ≥0, 
h −h

j 
(1+ T1 ) 1 2 − j2 ≤ T4

1

for all fixed j1 and j2 . When summing over j1 and j2 , we obtain the conclusion of Lemma
3. 
b and Φ we have
Proof of Lemma 4. By the properties of K
X 1 X
1/2
ah1 ah2
u−1 u
(j1 j2 ) h ≥0,h ≥0,
2 ≤j1 ,j2 <2 1 2
h −h
 
j 
(1+ T1 ) 1 2 − j2 ≤ T4

1

X K(log   
b j1 j2 ) X b j1 h1 −h2
≪ ah ah Φ T log (1 + 1/T )
j1 ,j2 ≥1
(j1 j2 )1/2 h ,h ≥0 1 2 j2
1 2
Z X b  
1 K(log j1 j2 ) j1 it
(29) ≪ |P (t)|2 Φ(t/T )dt.
T R
j1 ,j2 ≥1
(j1 j2 )1/2 j2

Note that we crucially used the fact that in all three lines of the displayed equation above, all
terms in the summations are non-negative, because K, b Φ and Φ b are all non-negative. Thus,
u 1/2 u
using that 2 ≪ (j1 j2 ) ≪ 2 , we have
Z X b  
X X 2u K(log j1 j2 ) j1 it
ah1 ah2 ≪ 1/2
|P (t)|2 Φ(t/T )dt,
T R (j1 j2 ) j2
u−1
2 u
≤j1 ,j2 ≤2  h1 ,h2 ≥0,  j1 ,j2 ≥1
h −h j 
(1+ T1 ) 1 2 − j2 ≤ T4

1

as claimed. 
In order to prove Lemma 5, we need the following technical tool, which is Lemma 5.3 of [14].
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 13

Lemma 6. Let σ ∈ (−∞, 1) and let F be a holomorphic function in the strip y = ℑz ∈

[σ − 2, 0], such that
1
(30) sup |F (x + iy)| ≪ .
σ−2≤y≤0 x2 +1
Then for all s = σ + it ∈ C, t 6= 0, we have
X Fb (log kℓ)
k s ℓs
k,ℓ>1
Z
= ζ(s + iu)ζ(s − iu)F (u)du + 2πζ(1 − 2it)F (is − i) + 2πζ(1 + 2it)F (is − i).
R

The proof of this lemma (Lemma 6) is only briefly sketched in [14]. However, a detailed proof
of a similar lemma is given in [10, Lemma 1]; for the proof of our lemma one can exactly
follow the argument given there, just using the function f : z 7→ ζ(z + it)ζ(z − it)K(iσ − iz)
instead of the one considered there.

Proof of Lemma 5. For simplicity of writing we define

 
X K(log
b j1 j2 ) j1 it
G(t) = 1/2
.
(j1 j2 ) j2
j ,j ≥1
1 2

We note that K can be extended analytically and satisfies assumption (30). Furthermore
(31) |K(t − i/2)| ≪ N 1+ε/4 /(t + 1)2 , |K(−t − i/2)| ≪ N 1+ε/4 /(t + 1)2 .
We first note that we have the pointwise bound
 2 !2
X ∞
X
(32) |P (t)|2 ≤ |P (0)|2 =  ah  ≪ bn ≪ N 4, t ∈ R.
h≥0 n=1

This allows us to see that

Z 1
G(t)|P (t)|2 Φ(t/T )dt ≪ N 5
−1
and thus we can restrict our integration domain to |t| ≥ 1 below. By Lemma 6 we have
Z
G(t)|P (t)|2 Φ(t/T )dt = Int1 + Int2 + Int3 ,
|t|≥1

where
Z Z
2
Int1 = |P (t)| Φ(t/T ) ζ(1/2 + it + iu)ζ(1/2 − it + iu)K(u) du dt,
|t|≥1 R
Z
Int2 = 2π ζ(1 − 2it)K(−t − i/2)|P (t)|2 Φ(t/T ) dt,
|t|≥1
Z
Int3 = 2π ζ(1 + 2it)K(t − i/2)|P (t)|2 Φ(t/T ) dt.
|t|≥1
14 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

Using (31) together with the easy estimate |ζ(1 + it)| ≪ log t, we obtain
Z
1+ε/4 4 log t
Int2 ≪ N N 2
dt
t≥1 t

≪ N 5+ε/4 .
Exactly the same estimate holds for Int3 .

The classical convexity bound |ζ(1/2 + it)| ≪ |t|1/4 gives |ζ(1/2 + it + iu)ζ(1/2 − it + iu)| ≪
(|t| + |u|)1/2 ≪ |t|1/2 + |u|1/2 . Hence we can bound the contribution of the domain |u| ≥ T to
Int1 by
Z Z Z Z
|t|1/2 |P (t)|2 Φ(t/T ) K(u) du dt + |P (t)|2 Φ(t/T ) |u|1/2 K(u) du dt
R |u|≥T R |u|≥T
| {z } | {z }
≪T −1 ≪T −1/2
Z !
|t|1/2 1
≪ + 1/2 |P (t)|2 Φ(t/T )dt ≪ T 1/2 EN
∗
,
R T T

where we used K(u) ≪ u−2 and Lemma 2 together with the quick decay of Φ.
Let A = 178
13 . Then by Ivić’s theorem [23, Theorem 8.3] we have
Z T
3(A−12)
(33) |ζ(1/2 + it)|A dt ≪ T 2+ 22 +ε = T 29/13+ε .
0
The contribution to Int1 of small u is
Z Z 
2
K(u) |ζ(1/2 + it + iu)||ζ(1/2 − it + iu)||P (t)| Φ(t/T ) dt du.
|u|≤T R

To estimate the term inside the brackets, we use Hölder’s inequality with parameters 1/A +
A 89
1/A + 1/B = 1, so that B = A−2 = 76 , and write |P (t)|2 = |P (t)|2−2/B |P (t)|2/B . By Lemma
2 together with (32) and (33) we deduce
Z
(34) |ζ(1/2 + it + iu)||ζ(1/2 − it + iu)||P (t)|2 Φ(t/T ) dt
R
Z 1/A Z 1/A
A A
≪ |ζ(1/2 + it + iu)| Φ(t/T ) dt |ζ(1/2 − it + iu)| Φ(t/T ) dt ×
R R
Z 1/B
2(1−1/B) 2
× |P (0)| |P (t)| Φ(t/T ) dt
R
 2/A
29/13+ε
≪ T N 4(1−1/B) T 1/B (EN∗ 1/B
)
 2/A
∗ (A−2)/A
= T 29/13+ε N 8/A T (A−2)/A (EN ) .

Integrating over u we deduce that

 2/A
Int1 ≪ T 29/13+ε N 8/A T (A−2)/A (EN
∗ (A−2)/A
)

= T 105/89+ε N 52/89 (EN

∗ 76/89
) .
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 15

Using Lemma 3 and (29) and inserting our bounds for Int1 , Int2 and Int3 we obtain
X X

1 |j1 zm − j2 zn | < 1
1≤m,n≤M 2u−1 ≤j1 ,j2 <2u
Z
2u 2
≪ G(t)|P (t)| Φ(t/T )dt
T R
2u  5+ε/4 
≪ N + T 105/89+ε N 52/89 (EN
∗ 76/89
) + T 1/2 EN
∗
T
2u  5+ε/4 
≪ N + T 105/89+ε N 52/89 (EN
∗ 76/89
) ,
T
where we used that the term T 1/2 EN
∗ is dominated by the other two summands. Substituting

T = 2u N 1+ε/2 and using that 2u ≪ N , we finally get the upper bound

X X
1(|j1 zm − j2 zn | < 1)
1≤m,n≤M 2u−1 ≤j1 ,j2 <2u
1
≪ (N 5+ε/4 + N 52/89 N 2×105/89+2ε (EN
∗ 76/89
) )
N 1+ε/2
= N 4−ε/4 + N 173/89+2ε (EN
∗ 76/89
) .
This establishes Lemma 5. 
We note here that conditionally under the Lindelöf hypothesis we could estimate the integral
in line (34) much more efficiently, by using a pointwise bound for the zeta function and esti-
mating the remaining integral with Lemma 2.

Combining Lemmas 3, 4 and 5, we have shown in our Case 1 analysis that the contribution
of pairs zm , zn with zm , zn ≥ N 1.01 to the counting problem (17) is
X X

(35) 1 |j1 zm − j2 zn | < 1 ≪ N 4−ε/4 + N 173/89+2ε (EN
∗ 76/89
) .
1≤m,n≤M, 2u−1 ≤j1 ,j2 <2u
zm ,zn ≥N 1.01

Before we move on to Case 2, we make some further comments P on our argument above. Intu-
itively, it would seem more natural to work with Q(t) := m zm it rather than with the more

complicated function P (t). However, from a technical point of view the key problem in the
whole argument is to be able to choose an appropriate value of T which balances the contri-
bution to our integrals of those values of t for which |t| is “large” (this gets worse when T is
larger, since the bound for the zeta function grows polynomially in T ) against the contribu-
tion coming from those t for which |t| is “small” (this contribution can only be compensated
in the final estimate when T is sufficiently large). When working directly with RQ, the size
of T would need to depend on the size of the zm in order to be able to control |Q|2 . The
“orthogonalization” procedure leading to our definition of P (t) gives us more freedom in our
choice of T . The whole problem discussed in this paragraph occurs only in the real-number
setting, in contrast to the integer setting.

• Case 2: Counting solutions for zm , zn with min{zm , zn } < N 1.01 .

First we consider the contribution to (17) of those zm and zn for which max{zm , zn } < 4N 1/4 .
We assumed that xn+1 − xn ≥ c > 0, so zn ≥ c for all n. Furthermore, we deduce that among
16 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

z1 , . . . , zM there are at most ≪ N 5/4 many elements which are smaller than 4N 1/4 (we
suppress the dependence of the implied constant on c). Note that whenever j1 and zm , zn are
fixed, there are at most ≪ 1 many possible choices for j2 such that |j1 zm − j2 zn | < 1, again
since zn ≥ c. Thus the total contribution of pairs zm , zn with max{zm , zn } < 4N 1/4 to our
counting problem is at most
X X
 2
1 |j1 zm − j2 zn | < 1 ≪ N 5/4 2u ≪ N 7/2 .
1≤m,n≤M 2u−1 ≤j1 ,j2 <2u
max{zm ,zn }<4N 1/4

Now consider the case when max{zm , zn } ≥ 4N 1/4 . Recall that we have localized j1 , j2 into
a dyadic interval in the counting problem. This implies a similar localization for zm and zn ,
since j1 /j2 ∈ [1/2, 2] and |j1 zm − j2 zn | < 1 are only possible if we have zm /zn ∈ [1/4, 4], given
the fact that max{zm , zn } ≥ 4N 1/4 .

Thus we can restrict ourselves in the counting problem (17) to the case when zm ∈ [4N β , 8N β )
for some β ≥ 1/4, and when consequently zn needs to be in [N β , 32N β ). Note that there are
≪ log N many intervals of this form necessary to cover the whole relevant range [N 1/4 , N 1.01 ],
and clearly we only need to consider 1/4 ≤ β ≤ 1.01. We count more solutions if we relax
the condition to zm , zn ∈ [N β , 32N β ). Thus, let us consider
X X

(36) 1 |j1 zm − j2 zn | < 1
1≤m,n≤M, 2u−1 ≤j1 ,j2 <2u
zm ,zn ∈[N β ,32N β )

for some β ∈ [1/4, 1.01]. We set up everything as in Case 1, but now we define T = 2u N β .
We define the bk ’s as before but restricting ourselves to those zm contained in [N β , 32N β ).
That is, we set
X
bk = 1(zm ∈ [k, k + 1)).
1≤m≤M,
zm ∈[N β ,32N β )
P
Note that previously we had k bk = M ≤ N 2 , whereas now we have a stronger bound.
Applying the Cauchy–Schwarz inequality we obtain
X p ∗ β/2
(37) bk ≪ EN N .
k

We define (ah )h≥0 and P (t) as in Case1; see (19)  and (20). In the present case the inequality
 zm j 2 
|j1 zm −j2 zn | < 1 is only possible when  zn − j1  ≤ 2u−11 N β . By construction, 2u−1 N β becomes
large in comparison with T , and we can continue to argue as in Case 1. Note that now, as
a consequence of (37), we have |P (0)|2 ≪ EN ∗ N β instead of |P (0)|2 ≪ N 4 as in Case 1.

Proceeding as in Case 1 we obtain

Int2 , Int3 ≪ N 1+β+ε/4 EN
∗
+ T 1/2 EN
∗
.
178
As for Int1 , we now obtain, again writing A = 13 ,

Int1 ≪ (T 29/13+ε )2/A (EN

∗
N β )2/A (EN
∗ 1−2/A 1−2/A
) T
∗
= EN (T 29/13+ε )2/A (N β )2/A T 1−2/A .
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 17

Substituting T = 2u N β ≪ N 1+β , we conclude that (36) is bounded by

2u  ∗ 1+β+ε/4 ∗ 105/89+ε 13β/89

≪ EN N + EN T N
T
∗
≪ EN N 1+ε/4 + EN ∗
N 105/89+29β/89+(1+β)ε .
Recall that we only need to consider β ≤ 1.01, and that there are at most ≪ log N many
different values of β to consider for Case 2. Hence it follows that, for every fixed value of u,
(36) is bounded by
∗
(38) ≪ EN N 1.51+3ε .
We could have used a different value of A here (for instance A = 12, namely Heath-Brown’s
bound on the twelfth moment from [19]) to arrive at (38). However, we kept the same pa-
rameters as in Case 1 to simplify the writing.

Finally, inserting in (16) resp. (17) the bound (35) from Case 1 together with the bound (38)
yields  
Var(hN , µ) ≪ max N −ε/8 + N −183/89+3ε (EN ∗ 76/89
) ∗
, EN N −2.49+4ε ,
which concludes the proof of Lemma 1.

6. Proof of Theorem 1: conclusion of the proof

The crucial ingredient in the proof of Theorem 1 is the variance bound from Lemma 1. We
record that we have, for every sufficiently small ε > 0,
 
Var(hN , µ) ≪ max N −ε/8 + N −183/89+3ε (EN ∗ 76/89
) ∗
, EN N −2.49+4ε .
∗ ≪ N 183/76−δ shows that for every sufficiently small δ ′ > 0 we have
Inserting EN
′
(39) Var(hN , µ) ≪ N −δ .
Everything else now follows from a standard procedure. To be a bit more specific, conver-
gence in (10) can be established using the estimate for the expectations in Section 3, and
using the variance bound (39) together with Chebyshev’s inequality and the Borel–Cantelli
lemma. From that we get a convergence result for almost all α ∈ R, for fixed values of r and
s. One notes that there are only countably many possible values of r, and that by continu-
ity/monotonicity it is sufficient to consider countably many values of s. Since a countable
union of sets of measure zero has measure zero as well, almost all α ∈ R have the property
that (10) holds for all r and all s, as desired. We refer the reader to [5] or [35], where this
argument is carried out in full detail. It applies without any modifications to the situation in
the present paper.

7. Proof of Theorem 2.
We set up the same machinery as in the proof of Theorem 1. Controlling the expectations,
as in Section 3 above, is unproblematic. The crucial part is again the variance estimate. As
in Section 4, we are led to the counting problem
X X

1 |j1 zm − j2 zn | < 1
1≤m,n≤M 2u−1 ≤j1 ,j2 <2u
18 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

where {z1 , . . . , zM } is the multi-set of all the absolute differences {|xm − xn | : 1 ≤ m, n ≤

N, m 6= n}. As above, u is a positive integer with 2u ≤ 2rN , and M = N 2 − N .

As in the general argument before, we can easily dispose of the contribution of those zm , zn
for which max{zm , zn } < 4N 1/4 . Thus again we can localize zm and zn , and restrict ourselves
to counting
X X

(40) 1 |j1 zm − j2 zn | < 1
1≤m,n≤M, 2u−1 ≤j1 ,j2 <2u
zm ,zn ∈[N β ,32N β )

for some 1/4 ≤ β ≤ 1.01, where we understand that the final range for β = 1.01 extends
over all of [N β , ∞) rather than only [N β , 32N β ). Note that in total at most ≪ log N many
different values of β need to be considered. Let β ∈ [1/4, 1.01] be fixed. Let u in (40) be
fixed. For integers k ≥ 0 we define
X   
k k+1
(41) bk = 1 zm ∈ u , u ,
2 2
1≤m≤M,
zm ∈[N β ,32N β )

where again in case β = 1.01 the range [N β , 32N β ) is understood to be replaced by [N β , ∞).
Note the difference in comparison with (18). There we collected all zm in a range of the
form [k, k + 1), since we could only control the number of solutions of the specific inequality
(4), which has “< 1” on the right-hand side. In contrast we can now control the number of
solutions on a finer scale, and can accordingly set shorter ranges for the grouping of the zm
(where γ = 2−u ).

Let ε > 0 be a small constant (chosen depending on the size of η and δ in the statement of
the theorem). Set T = 2u N min{β−ε,1+ε}. Unlike the argument in the general case in Section
4, we do not explicitly distinguish between Case 1 and Case 2, but have implicitly included
S way that T is defined. As in Section 4, we split the interval [1, ∞)
this distinction into the
into a disjoint union ∞ h=0 Ih , where
"&  ' &  '!
1 h 1 h+1
Ih = 1+ , 1+ ,
T T

and set
 1/2
X
ah =  b2k  , h ≥ 0,
k: k/2u ∈Ih

as well as
∞  
X 1 iht
P (t) = ah 1 + .
T
h=0
Then by construction we again have
Z ∞
X
2
|P (t)| Φ(t/T )dt ≪ T a2h
R h=0
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 19

as during the proof of Lemma 2 in Section 4, but now we can continue to estimate this by
(5) and obtain
X X  
(42) T a2h ≪ T b2k ≪ T EN,2
∗
−u ≪ T N 2+η + 2−u N 3−δ .
h≥0 k≥0

Now we establish the necessary upper bound on |P (0)|. Trivially we always have |P (0)| ≪ N 2 .
For small values of β we obtain a better estimate. Note that
X X n o
|P (0)| = ah ≤ bk = # m : zm ∈ [N β , 32N β )
h k
⌊32N β ⌋
X
≤ # {m : zm ∈ [a, a + 1)}
a=⌊N β ⌋
 1/2
⌊32N β ⌋
X  2
(43) ≪ N β/2  # {m : zm ∈ [a, a + 1)} 
a=⌊N β ⌋
q
≪ N β/2 ∗
EN,1

(44) ≪ N β/2 N (3−δ)/2 = N 3/2+β/2−δ/2 ,

where for (43) we used Cauchy–Schwarz and for (44) our assumption (5) in the statement of
Theorem 2. So overall we have |P (0)| ≪ N min{2,3/2+β/2−δ/2} .

As in the proof of Lemma 3 in Section 4, we assume without loss of generality that j1 ≥ j2 .

Recall that j1 , j2 ≥ 2u−1 by assumption. Assume that zm ∈ [k/2u , (k + 1)/2u ) for some k.
Then the inequality |j1 zm − j2 zn | ≤ 1 is only possible when
l m 
 j1 k 
 j2  5
 − z 
n ≤ u ,
 2u 2
 

which is a version of (25) that is adapted to the construction in (41). Arguing as in the lines
leading to (26), this again gives
X

1 |j1 zm − j2 zn | < 1 ≪ ah1 ah2 ,
zm ∈Ih1 ,zn ∈Ih2

which perfectly resembles (26) but where now the bk and ah are defined in a different way
according to (41). Note that T is chosen in such a way that j1 zm and j2 zn exceed T ; indeed,
by assumption j1 , j2 ≥ 2u−1 and zm , zn ≥ N β , while T ≤ 2u N β−ε by definition. Thus we can
continue the argument as in Section 5. It turns out that in this setting in order to bound Int1
it is sufficient to use |ζ(1/2 + it)| ≪ |t|1/6 (which is essentially the Weyl–Hardy–Littlewood
bound), rather than the more elaborate argument relying on estimates for moments of the
Riemann zeta function. We obtain
X X

1 |j1 zm − j2 zn | < 1
1≤m,n≤M, 2u−1 ≤j1 ,j2 <2u
zm ,zn ∈[N β ,32N β )
20 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

2u  
≪ |P (0)|2 N 1+ε/4 + T 1+1/3 EN,2
∗
−u
T
2u N min{4,3+β−δ} N 1+ε/4
≪ + (2u )4/3 N (1+ε)/3 N 2+η + (2u )4/3 N (1+ε)/3 2−u N 3−δ
T
≪ N 1+ε/4+min{4,3+β−δ}−min{β−ε,1+ε} + N 11/3+ε+η + N 11/3+ε−δ
≪ N 4−ε/2
if ε was chosen sufficiently small (with respect to δ). Here we used (5) as well as T ≪
2u N 1+ε , 2u ≪ N . Noting that we need to consider at most ≪ log N different values of β,
this gives the necessary variance estimate. The remaining part of the proof of Theorem 3 can
be carried out exactly as in the proof of Theorem 1. We remark that any subconvex bound
for the Riemann zeta function would be sufficient to derive the same conclusion.

8. Proof of Theorem 3.
We assume that θ > 1 is fixed, and consider the sequence (xn ) defined by xn = nθ , n ≥ 1. Note
that with this definition we have xn+1 − xn ≥ 1 for all n ≥ 1, so the assumption xn+1 − xn ≥ c
of Theorem 2 is satisfied in this case with c = 1. The following lemma of Robert and Sargos
shows that the necessary bound on EN,γ∗ also is satisfied for this sequence.3
Lemma 7 ([32, Theorem 2]). Let θ 6= 0, 1 be a fixed real number. For any γ > 0 and B ≥ 2,
let N (B, γ) denote the number of 4-tuples (n1 , n2 , n3 , n4 ) ∈ {B + 1, B + 2, . . . , 2B}4 for which
 
 θ θ θ θ
(45) n1 − n2 + n3 − n4  ≤ γ.
Then for every ε > 0,
N (B, γ) ≪ε B 2+ε + γB 4−θ+ε .
The restriction to a dyadic range for (n1 , n2 , n3 , n4 ) in the statement of the lemma does
not actually play a role. This is easily seen by interpreting the number of solutions of the
inequality as an L4 -norm. Indeed, generalizing the definition in (11) and setting
(sin(γx))2
dµ2γ (x) = dx,
πγx2
we have a measure whose Fourier transform is a (normalized) tent function on [−2γ, 2γ]. Let
∗
EN,γ denote the number of solutions of (45), subject to (n1 , n2 , n3 , n4 ) ∈ {1, . . . , N }4 . Assume
for simplicity of writing that N is a power of 2, i.e. N = 2L for some L ≥ 1. Then applying
Hölder’s inequality we obtain
 4
Z X L X
∗ θ
EN,γ ≪  e2πin x  dµ2γ (x)
R ℓ=0 2ℓ−1 <n≤2ℓ
!3  4
Z L
X XL  
 X θ 
≪ 1  e2πin x 
  dµ2γ (x)
R ℓ=0 
ℓ=0 2ℓ−1 <n≤2ℓ 

3We thank Niclas Technau for pointing out to us that the estimate in Lemma 7 is also contained as a special
case in a general result in a very recent paper of Huang [22]. Huang’s result gives improved error terms, but
for our application this does not play a role. However, the generality of Huang’s results could allow further
applications of our method in the spirit of our Theorem 3.
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 21

 4
XL Z  X 
 
2πinθ x 
(46) ≪ (log N )3  e
  dµ2γ (x)
ℓ=0 R 2ℓ−1 <n≤2ℓ 
L
X
≪ (log N )3 E2ℓ−1 ,2γ
ℓ=0
2+ε
≪ε N + γN 4−θ+ε ,
which is obtained by interpreting the integrals in line (46) in terms of solutions of the Dio-
phantine inequality (45), and applying Lemma 7 with parameters 2γ and B = 2ℓ−1 . Thus we
have
∗
(47) EN,γ ≪ε N 2+ε + γN 4−θ+ε
for any ε > 0. Consequently all assumptions of Theorem 2 are satisfied, and we can conclude
that (nθ α)n has Poissonian pair correlation for almost all α.

9. Closing remarks
As remarked in the introduction, our method breaks down completely when the growth order
of the sequence (xn )n≥1 is only linear or even slower. Not only does the “lattice point counting
with the zeta function” argument from Section 5 fail to work in this situation, but there is a
much more fundamental reason why the whole approach based on calculating first and second
moments (expectations and variances, as in Sections 3 and 4) fails to work in this setup. To
give a brief sketch of what causes the problem, assume that (xn )n is a sequence of reals such
that xn ≤ n, n ≥ 1. Assume that we want to bound the variance in analogy with (15), so
say we want to show that
 2
Z X
1 
(48)  1[−1/N,1/N ] (xm α − xn α)
N  dµ(α)
R 1≤m,n≤N,
m6=n

tends to zero as N → ∞ (where we write the original indicator function instead of its approx-
imation by a trigonometric polynomial, and where for simplicity of writing we set s = 1). By
our assumption on the growth of (xn )n , all differences xm − xn appearing in the sum above
are uniformly bounded by N . Thus we have 1[−1/N,1/N ] (xm α − xn α) = 1 throughout the
range α ∈ [−1/N 2 , 1/N 2 ], for all m, n ≤ N . Consequently
 2  2
Z X Z 2
1/N  X
1  
 (x α − x α) dµ(α) ≥ 1 1
N 1[−1/N,1/N ] m n   N  dµ(α)
R 1≤m,n≤N, −1/N 2 1≤m,n≤N,
m6=n m6=n
≫ 1.
Thus the variance fails to tend to zero for a slowly growing (xn )n , due to the fact that the
contribution of small values of α to the variance integral is too large.4 The argument used
in Section 5 fails to work in a similar way for slowly growing (xn )n , since the error terms
4A similar argument appears at the end of [33], where it is used to show that the L2 approach fails to work
in the case of the triple correlation of (n2 α)n ; cf. also [41].
22 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

coming from the contribution to the integrals of values of t near zero become too large.5
Consequently, it seems that for establishing Poissonian pair correlation of (xn α)n for almost
all α for slowly growing (xn )n some genuine new ideas are necessary. Note that we cannot
simply remove all values of α near zero from the variance integral (48) by replacing µ with
some other measure which vanishes for small α, since such a measure would fail to have non-
negative Fourier transform (thereby causing major problems in other places). Note also that
all these problems with slowly growing sequences (xn )n are a novel aspect which only shows
up in the real-number setup – in contrast, when (an )n is an integer sequence which grows
at most linearly, then (an α)n is known to fail to have Poissonian pair correlation for any α,
because the additive energy of (a1 , . . . , aN ) necessarily is of maximal possible order (cf. [26]).

We emphasize that the fact that our method fails to work in the case of slowly growing
sequences (xn )n should not be understood as indicating that in such a case (xn α)n should
necessarily fail to have Poissonian pair correlation for almost all α. Quite on the contrary,
there are good reasons to expect that also for slowly growing (xn α)n one should in “generic”
situations obtain Poissonian pair correlation for almost all α. It seems that the property
of having Poissonian pair correlation for (xn α)n for almost all α can only be prevented by
a certain (“small-scale”) combinatorial obstruction, in such a way that the case of integer
sequences (xn )n with slowly growing (xn )n can be seen as a degenerate situation exhibiting
exactly this type of combinatorial obstruction (coming from the fact that in the integer setup
everything which is smaller than one in absolute value necessarily equals zero). We believe
that these are very interesting phenomena, and we propose the following open problems.

Open Problem 1: Let θ ∈ (0, 1). Show that (nθ α)n≥1 has Poissonian pair correlation for
almost all α. Note that Lemma 7 is still valid for this range of θ.

Open Problem 2: Let xn = n + log n. Show that (xn α)n has Poissonian pair correlation
for almost all α. We note that it is possible to establish a variant of Lemma 7 for this setting
(with exponent 3 in place of 4 − θ).

Open Problem 3: Let xn = n log n, n ≥ 1. Show that (xn α)n has Poissonian pair correla-
tion for almost all α.

Clearly the exponent 183/76 − δ in the statement of Theorem 1 is not optimal, and most
likely it can be improved to 3 − δ (which is the case conditionally under the Lindelöf hypoth-
esis). It seems to us that the method of Bloom and Walker [8], which led to a quantitative
improvement of the results of [5], cannot be used here. Their method relied on sum-product
5It might be difficult to spot at a quick glance, so we briefly comment on where the speed of growth of (x )
n n
was used in our argument in Sections 5 and 7. There is a term |P (0)|2 N 1+ε/4 coming from the contribution
of values of t near the origin to the integral. This term is divided by T at the end of the calculation, so we
cannot take T too small since we need N |P (0)|2 N 1+ε/4 /N 4 T → 0. On the other hand, we cannot take T too
large, since we need T ≪ 2u zn to be able to detect the solutions of our Diophantine inequality. To balance
everything out, we need to be able to assure that there are not too many small values of zn (i.e., not too
many differences xm − xn which are “small”). In our proof of Theorem 1 our assumption on the order of the
∗
additive energy takes care of this: it is easy to see that an upper bound on EN implies an upper bound on the
number of “small” differences xm − xn , which is what we used in Case 2 of Section 5. In the setting of Theorem
2 a similar argument based on the energy assumption allowed us to control the number of small differences
xm − xn ; the relevant equations there are (37) and (44).
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 23

estimates, which, roughly speaking, leads to an integrand |P (t)|2 being replaced by |P (t)|4 .
In the case of integer sequences (when Rworking with the random model of the zeta function)
one has perfect orthogonality, so that |P |4 can be efficiently bounded. In our setting the
situation is quite different – we have constructed our
R function P (t) in such a way that the
diagonal contribution dominates when calculating |P |2 , but we do not have orthogonality
R
for |P |4 and cannot efficiently bound this integral.

Open Problem 4: Show that Theorem 1 remains valid under the weaker assumption EN ∗ ≪

N 3−δ for some δ > 0. Show that this can be further relaxed to assuming EN,γ ∗ ≪ γN 4−δ , for
all γ in a range from roughly 1/N to 1. It might even be the case that only values of γ near a
∗
critical size of roughly 1/N are relevant. Note that if the condition EN,γ ≪ γN 4−δ uniformly
for γ ∈ [1/N, 1] truly is the “right” condition, then this would give a unified picture for the
real-sequence case as well as for the integer-sequence case. Indeed, in the latter case clearly
∗
EN,γ = EN for all γ < 1 and thus the condition would reduce to EN ≪ N 3−δ , in accordance
with the criterion stated after (2).

As noted, in the case of an integer sequence (xn )n it is known that (xn α)n cannot have Pois-
sonian pair correlation for almost all α when EN ≫ N 3 . It would be interesting to obtain an
analogous result in the case of real sequences.

Open Problem 5: Show that unlike in the integer case, it is possible for an increasing
∗ ≫ N 3 and that (x α) has Poissonian pair correlation for
sequence (xn )n≥1 of reals that EN n n
almost all α (compare Open Problems 1 and 2 above, where EN ∗ ≫ N 3 ). Establish a criterion
∗
(stated for example in terms of EN,γ ) which ensures that (xn α)n does not have Poissonian
∗
pair correlation for almost all α. A candidate for such a criterion is that EN,γ ≫ γN 4 for
some γ = γ(N ) for infinitely many N , where maybe one also has to assume that these values
of γ are of size γ ≈ 1/N .

Acknowledgements
CA is supported by the Austrian Science Fund (FWF), projects F-5512, I-3466, I-4945 and
Y-901. DE is supported by FWF projects F-5512 and Y-901. MM is supported by FWF
project P-33043. We thank Winston Heap, Olivier Robert, Zeev Rudnick, Ilya Shkredov,
Igor Shparlinski, Athanasios Sourmelidis and Niclas Technau for discussions and comments.

References
[1] C. Aistleitner. Lower bounds for the maximum of the Riemann zeta function along vertical lines. Math.
Ann., 365(1-2):473–496, 2016.
[2] C. Aistleitner, I. Berkes, and K. Seip. GCD sums from Poisson integrals and systems of dilated functions.
J. Eur. Math. Soc. (JEMS), 17(6):1517–1546, 2015.
[3] C. Aistleitner, T. Lachmann, and F. Pausinger. Pair correlations and equidistribution. J. Number Theory,
182:206–220, 2018.
[4] C. Aistleitner, T. Lachmann, and N. Technau. There is no Khintchine threshold for metric pair correla-
tions. Mathematika, 65(4):929–949, 2019.
[5] C. Aistleitner, G. Larcher, and M. Lewko. Additive energy and the Hausdorff dimension of the exceptional
set in metric pair correlation problems. Israel J. Math., 222(1):463–485, 2017.
[6] I. Berkes, W. Philipp, and R. Tichy. Pair correlations and U -statistics for independent and weakly depen-
dent random variables. Illinois J. Math., 45(2):559–580, 2001.
24 C. AISTLEITNER, D. EL-BAZ, M. MUNSCH

[7] T. F. Bloom, S. Chow, A. Gafni, and A. Walker. Additive energy and the metric Poissonian property.
Mathematika, 64(3):679–700, 2018.
[8] T. F. Bloom and A. Walker. GCD sums and sum-product estimates. Israel J. Math., 235(1):1–11, 2020.
[9] A. Bondarenko and K. Seip. Large greatest common divisor sums and extreme values of the Riemann zeta
function. Duke Math. J., 166(9):1685–1701, 2017.
[10] A. Bondarenko and K. Seip. Extreme values of the Riemann zeta function and its argument. Math. Ann.,
372(3-4):999–1015, 2018.
[11] J. Brüdern, K. Kawada, and T. D. Wooley. Additive representation in thin sequences, VIII: Diophantine
inequalities in review. In Number theory, volume 6 of Ser. Number Theory Appl., pages 20–79. World Sci.
Publ., Hackensack, NJ, 2010.
[12] S. Chaubey, M. Lanius, and A. Zaharescu. Pair correlation of fractional parts derived from rational valued
sequences. J. Number Theory, 151:147–158, 2015.
[13] R. de la Bretèche, M. Munsch, and G. Tenenbaum. Small Gál sums and applications. Journal of the
London Mathematical Society, 103(1):336–352, 2021.
[14] R. de la Bretèche and G. Tenenbaum. Sommes de Gál et applications. Proc. Lond. Math. Soc. (3),
119(1):104–134, 2019.
[15] M. Drmota and R. F. Tichy. Sequences, discrepancies and applications, volume 1651 of Lecture Notes in
Mathematics. Springer-Verlag, Berlin, 1997.
[16] √
D. El-Baz, J. Marklof, and I. Vinogradov. The two-point correlation function of the fractional parts of
n is Poisson. Proc. Amer. Math. Soc., 143(7):2815–2828, 2015.
[17] S. Grepstad and G. Larcher. On pair correlation and discrepancy. Arch. Math. (Basel), 109(2):143–149,
2017.
[18] D. J. Grynkiewicz. Structural additive theory, volume 30 of Developments in Mathematics. Springer, Cham,
2013.
[19] D. R. Heath-Brown. The twelfth power moment of the Riemann-function. Quart. J. Math. Oxford Ser.
(2), 29(116):443–462, 1978.
[20] D. R. Heath-Brown. Pair correlation for fractional parts of αn2 . Math. Proc. Cambridge Philos. Soc.,
148(3):385–407, 2010.
[21] T. Hilberdink. An arithmetical mapping and applications to Ω-results for the Riemann zeta function. Acta
Arith., 139(4):341–367, 2009.
[22] J.-J. Huang. The density of rational points near hypersurfaces. Duke Math. J., 169(11):2045–2077, 2020.
[23] A. Ivić. The Riemann zeta-function. Dover Publications, Inc., Mineola, NY, 2003. Theory and applications,
Reprint of the 1985 original [Wiley, New York; MR0792089 (87d:11062)].
[24] L. Kuipers and H. Niederreiter. Uniform distribution of sequences. Wiley-Interscience [John Wiley &
Sons], New York-London-Sydney, 1974.
[25] G. Larcher and W. Stockinger. Pair correlation of sequences ({an α})n∈N with maximal additive energy.
Math. Proc. Cambridge Philos. Soc., 168(2):287–293, 2020.
[26] G. Larcher and W. Stockinger. Pair correlation of sequences with maximal additive energy. Math. Proc.
Cambridge Philos. Soc., 168(2):287–293, 2020.
[27] M. Lewko and M. Radziwill. Refinements of Gál’s theorem and applications. Adv. Math., 305:280–297,
2017.
[28] J. Marklof. The Berry-Tabor conjecture. In European Congress of Mathematics, Vol. II (Barcelona, 2000),
volume 202 of Progr. Math., pages 421–427. Birkhäuser, Basel, 2001.
[29] J. Marklof. Pair correlation and equidistribution on manifolds. Monatsh. Math., 191(2):279–294, 2020.
[30] J. Marklof and N. Yesha. Pair correlation for quadratic polynomials mod 1. Compos. Math., 154(5):960–
983, 2018.
[31] H. L. Montgomery. Ten lectures on the interface between analytic number theory and harmonic analysis,
volume 84 of CBMS Regional Conference Series in Mathematics. Published for the Conference Board of
the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI,
1994.
[32] O. Robert and P. Sargos. Three-dimensional exponential sums with monomials. J. Reine Angew. Math.,
591:1–20, 2006.
[33] Z. Rudnick and P. Sarnak. The pair correlation function of fractional parts of polynomials. Comm. Math.
Phys., 194(1):61–70, 1998.
 A PAIR CORRELATION PROBLEM AND COUNTING LATTICE POINTS 25

[34] Z. Rudnick, P. Sarnak, and A. Zaharescu. The distribution of spacings between the fractional parts of
n2 α. Invent. Math., 145(1):37–57, 2001.
[35] Z. Rudnick and N. Technau. The metric theory of the pair correlation function of real-valued lacunary
sequences. Illinois J. Math., 2020. Advance publication.
[36] Z. Rudnick and A. Zaharescu. A metric result on the pair correlation of fractional parts of sequences.
Acta Arith., 89(3):283–293, 1999.
[37] Z. Rudnick and A. Zaharescu. The distribution of spacings between fractional parts of lacunary sequences.
Forum Math., 14(5):691–712, 2002.
[38] I. Shkredov. On some applications of GCD sums to arithmetic combinatorics, 2020. Preprint,
arXiv:2010.03020.
[39] I. D. Shkredov. Some new results on higher energies. Trans. Moscow Math. Soc., pages 31–63, 2013.
[40] T. Tao and V. Vu. Additive combinatorics, volume 105 of Cambridge Studies in Advanced Mathematics.
Cambridge University Press, Cambridge, 2006.
[41] N. Technau and A. Walker. On the triple correlations of fractional parts of n2 α, 2020. Preprint,
arXiv:2005.01490.
[42] N. Technau and N. Yesha. On the correlations of nα mod 1, 2020. Preprint, arXiv:2006.16629.
[43] J. L. Truelsen. Divisor Problems and the Pair Correlation for the Fractional Parts of n2 α. International
Mathematics Research Notices, 2010(16):3144–3183, 01 2010.
[44] H. Weyl. Über die Gleichverteilung von Zahlen mod. Eins. Math. Ann., 77(3):313–352, 1916.