The Goldbach Conjecture

Ricardo Guillermo Barca

To cite this version:

Ricardo Guillermo Barca. The Goldbach Conjecture. 2018. �hal-01662191v3�

HAL Id: hal-01662191

https://hal.science/hal-01662191v3
Preprint submitted on 17 Aug 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 The Goldbach Conjecture
Ricardo G. Barca

Abstract
The binary Goldbach conjecture asserts that every even integer greater than 4 is the sum of two primes. In order
to prove this statement, we begin by introducing a kind of double sieve of Eratosthenes as follows. Given a positive
even integer x > 4, we sift out from [1, x] all those elements that are congruents to 0 modulo p, or congruents to x
√ √
modulo p, where p is a prime less than x. So, any integer in the interval [ x, x] that remains unsifted is a prime
q for which either x − q = 1 or x − q is also a prime. Then, we introduce a new way of formulating a sieve, which we
call the sequence of k-tuples of remainders. Using this tool, we prove that there exists an integer Kα > 5 such that
pk /2 is a lower bound for the sifting function of this sieve, for every even number x that satisfies p2k < x < p2k+1 ,
where k > Kα . This result implies that every even integer x > p2k (k > Kα ) can be expressed as the sum of two
primes. Furthermore we provide the upper estimation Kα < 89.

1 Introduction
1.1 The sieve method and the Goldbach’s problem
In the year 1742 Goldbach wrote a letter to his friend Euler telling him about a conjecture involving prime numbers.
Goldbach’s conjecture: Every even number greater than 4 is the sum of two primes. The Goldbach Conjecture is
one of the oldest unsolved problems in number theory [6]. This conjecture was verified many times with powerful
computers, but could not be proven. In May 26, 2013, T. Oliveira e Silva verified the conjecture for n ≤ 4 × 1017 [10].
Mathematicians had achieved some partial results in their efforts to prove this conjecture. Vinogradov proved, in
1937, that every sufficiently large odd number is the sum of three primes [12]. Later, in 1973, J.R. Chen showed that
every sufficiently large even number can be written as the sum of either two primes or a prime and the product of two
primes [5]. In 1975, H. Montgomery and R.C. Vaughan showed that ‘most’ even numbers were expressible as the sum
of two primes [9]. Recently, a proof of the related ternary Goldbach conjecture, that every odd integer greater than 5
is the sum of 3 primes, has been given by Harald Helfgott [13].
In this paper we prove (Main Theorem, Section 8) the following: There exists an integer Kα > 5 such that every
even integer greater than p2k (k > Kα ) is the sum of two primes. We shall also derive an upper bound for the number
Kα .
It is well known that one of the principal ways of attacking the problem of the Goldbach’s conjecture has been
through the use of sieve methods. Viggo Brun [4] was the first to obtain a result, as an approximation to Goldbach’s
conjecture: Every sufficiently large even integer is a sum of two integers, each having at most nine prime factors.
Later, other mathematicians in the area of sieve theory have improved this initial result.
In the context of sieve theory, the sieve method consist in removing elements of a list of integers, according to a set
of rules; for instance, given a finite sequence A of integers, we could remove from A those members which lie in a given
collection of arithmetic progressions. In the original sieve of Eratosthenes, we start with the integers√in the interval
[1, x], where x is a positive real number, and sift out all those√ which are divisible by the primes p < x. Therefore,
any integer that remains unsifted is a prime in the interval [ x, x].
We begin by describing briefly the sieve problem; we use, as far as possible, the concepts and notation of the book
by Cojocaru and Ram Murty [2], chapters 2 and 5. Let A be a finite set of integers and let P be the sequence of all
primes; let z ≥ 2 be a positive real number. Furthermore, to each p ∈ P, p < z we have associated a subset Ap of
A . The sieve problem is to estimate, from above and below, the size of the set

[
A \ Ap ,
p∈P
p<z

which consists of the elements of the set A after removing the elements of all the subsets Ap . We call the procedure
of removing the elements of the subsets Ap from the set A the sifting process. The sifting function S(A , P, z) is
defined by the equation

1
 [
S(A , P, z) = A \ Ap ,
p∈P
p<z

and counts the elements of A that have survived the sifting process. Now, let Pz be the set of primes p ∈ P, p < z;
and for each subset I of Pz , denote by

\
AI = Ap .
p∈I

Then, the inclusion–exclusion principle gives us

|I|
X
S(A , P, z) = (−1) |AI | ,
I⊆Pz

where for the empty set ∅ we have A∅ = A . We often take A to be a finite set of positive integers, and Ap to be
the subset of A consisting of elements lying in some congruence classes modulo p.
Using this notation, we can now define formally the sieve of√Eratosthenes. Let A = {n ∈ Z+ : n ≤ x}, where
x ∈ R, x > 1, and let P be the sequence of all primes. Let z = x, and

Y
P (z) = p.
p∈P
p<z

Now, to each p ∈ P, p < z, we associate the subset Ap of A , defined as follows: Ap = {n ∈ A : n ≡ 0 (mod √ p)}.
Then, when we sift out from A all those elements of every set Ap , the unsifted members of A in the interval √ [ x, x]
are the integers that are not divisible by primes of P less than z; that is to say, any integer remaining in [ x, x] is a
prime. Furthermore, if d is a squarefree integer such that d|P (z), we define the set

\
Ad = Ap .
p|d

So, from the inclusion–exclusion principle we obtain

X
S(A , P, z) = µ (d) |Ad | , (1)
d|P (z)

where µ(d) is the Möbius function; and from (1) it can be derived the well-known formula of Legendre

X X jxk
S(A , P, z) = µ (d) |Ad | = µ (d) .
d
d|P (z) d|P (z)
√
In a first instance, the sieve of Eratosthenes is very useful for finding the prime numbers between x and x.
However, from a theoretical point of view, the experts in sieve theory are interested in estimating for every x the
number of integers remaining after the sifting process has been performed.
The use of the Möbius function is a simple way to approach a sieve problem; however, satisfactory results are rather
hard to achieve, unless z is very small. We shall illustrate this with the special case given in the book by Halberstam
and Richert [1], Chapter 1, Section 5.
Let A = {n ∈ Z+ : n ≤ x}, and let 2 ≤ z ≤ x. As usual in sieve theory, instead of |A | we can use a close
approximation X to |A |. Furthermore, for each prime p we choose a multiplicative function w(p) so that (w(p)/p)X
approximates to |Ap |. Then, for each squarefree integer d we have that (w(d)/d)X approximates to |Ad |, and we can
write

w(d)
|Ad | = X + Rd ,
d
where Rd is the remainder term. Then, substituting this into (1),

2
  
X w(d) X w(d) X
S(A , P, z) = µ (d) X + Rd = µ (d) X+ µ (d) Rd . (2)
d d
d|P (z) d|P (z) d|P (z)

On the other hand, since w is a multiplicative function,

X w(d) Y  w (p)

µ (d) = 1− = W (z) ,
d p
d|P (z) p∈P
p<z

and, substituting this into (2),

X
S(A , P, z) = XW (z) + µ (d) Rd .
d|P (z)

Hence, we can write

X
S(A , P, z) = XW (z) + θ |Rd | (|θ| ≤ 1).
d|P (z)

Furthermore, if we impose the conditions |Rd | ≤ w(d) and w(p) ≤ A0 , for some constant A0 ≥ 1, we get

z
S(A , P, z) = XW (z) + θ (1 + A0 ) .

See [1, Theorem 1.1] for details. Now, taking X = x, w(p) = 1, A0 = 1, these conditions are satisfied, and we
obtain

Y  1

S(A , P, z) ≤ x 1− + 2z . (3)
p
p∈P
p<z

From this illustrative case we can see that the error term will be very large provided that z is not sufficiently small
compared with x. In spite of this, taking z = log x, the formula in (3) can be used to obtain an elementary upper
bound for π(x). See [1, Ch. 1, (5.8)].
Now, suppose that we express the Goldbach’s problem as a sieve problem; it is clear that in order to prove this
conjecture what we require is a lower bound for the sifting function. However, there is a well-known phenomenon
in sieve theory, called the ‘parity barrier’ or the ‘parity problem’, which was explained first by Selberg (see [15]). It
appears that sieve methods cannot distinguish between numbers with an even number of prime factors and an odd
number of prime factors. The parity problem was described briefly by Terence Tao [14] as follows: ‘If A is a set whose
elements are all products of an odd number of primes (or are all products of an even number of primes), then (without
injecting additional ingredients), sieve theory is unable to provide non-trivial lower bounds on the size of A.’ This
means that in order to solve the Goldbach’s problem we need first to find a suitable sieve, and then to introduce new
procedures for estimating the sifting function, very far from the usual methods in the current sieve theory.

1.2 A sieve for the Goldbach’s problem

Let P be the sequence of all primes; and given pk ∈ P, let mk = p1 p2 p3 · · · pk . From now on, and throughout this
paper,√for convenience, we take x to be an even integer greater than p24 = 49. Note that if pk is the greatest prime less
than x, every even number x > 49 satisfies p2k < x < p2k+1 < mk ; this fact is very important for our purposes, as we
shall see later.
Now, how can we construct a sieve to tackle the Goldbach’s problem? Given a positive even √ integer x, as we have
seen in the previous subsection,
√ using the sieve of Eratosthenes we can get the primes between x and x. Assume that
among the primes between x and x there is at least a prime q such that x − q is also a prime. Then, to attack the
Goldbach’s
√ problem we need a sieve that sift out all the integers in the interval [1, x] which are divisible by the
√ primes
p < x, as the sieve of Eratosthenes does, and that additionally sift out, from the primes q remaining in [ x, x], all
those such that x − q is not a prime.
Then, in order to construct such a sieve, we propose to modify the sieve of Eratosthenes
√ as follows: First, we sift
out all those integers n in the interval
√ [1, x] such that n ≡ 0 (mod p), where p < x; thus, any integer
√ that remains
unsifted is a prime in the interval [ x, x]. Next, we sift out all those integers n that remains in [ x, x] such that

3
 √
n ≡ x (mod p). It is easy to√see that any number that remains unsifted in [ x, x] is a prime q such that x − q is not
divisible by the primes p < x; so, either x − q = 1 or x − q is a prime.
Let us define formally this sieve, which we call the Sieve associated √with x, or alternatively the Sieve I. Let
A = {n ∈ Z+ : n ≤ x}. Let P be the sequence of all primes; and let z = x. Let

Y
P (z) = p = mk .
p∈P
p<z

Now, to each p ∈ P, p < z, we associate the subset Ap of A , defined as follows: Ap = {n ∈ A : n ≡ 0

(mod p) or n ≡ x (mod p)}. Furthermore, if d is a squarefree integer such that d|P (z), we define the set

\
Ad = Ap .
p|d

In this case, the sifting function

[
S(A , P, z) = A \ Ap
p∈P
p<z
√ √
counts the primes q in the interval [ x, x], such that x − q is also a prime in [ x, x]; and furthermore counts 1
and x − 1 whenever x − 1 is a prime. As in the case of the sieve of Eratosthenes-Legendre, the inclusion–exclusion
principle gives us

X
S(A , P, z) = µ (d) |Ad | .
d|P (z)

Now, S(A , P, z) > 2 implies that x is the sum of two primes; and if this is proved for all x, the Goldbach’s
conjecture would be proved. Then, the solution of the Goldbach’s problem depends on establishing a positive lower
bound for the sifting function. However, we can not hope to find a suitable lower bound by using the usual sieve
methods, due to the parity problem, which was already mentioned in this Introduction. So far, all attempts to solve the
Goldbach’s problem by the usual sieve techniques did not have the expected success. For these reasons, the strategy
used in this paper differs quite a lot from the usual approach in sieve theory. In the next subsection we shall begin by
introducing another way of formulating a sieve problem.

1.3 The sequence of k-tuples of remainders

In this paper we propose to use another formulation for this kind of sieves, which is able to show all the details of
the sifting process, and will allow us to obtain a lower bound for the number of elements that remain unsifted. For
this purpose, we begin by introducing the notion of sequence of k-tuples of remainders. Let {p1 , p2 , p3 , . . . , pk } be the
ordered set of the first k prime numbers. Suppose that for every natural number n we form a k-tuple, the elements
of which are the remainders of dividing n by p1 , p2 , p3 , . . . , pk ; so, we have a sequence of k-tuples of remainders.
If we arrange these k-tuples from top to bottom, the sequence of k-tuples of remainders can be seen as a matrix
formed by k columns and infinitely many rows, where each column is a periodic sequence of remainders modulo
ph ∈ {p1 , p2 , p3 , . . . , pk }. It is easy to prove that the sequence of k-tuples of remainders is periodic, and the period is
mk = p1 p2 p3 · · · pk .
Suppose that within the periods of every sequence of remainders modulo ph (a given column of the matrix), we
define some (not all) of the remainders as selected remainders, no matter the criterion for selecting the remainders.
Consequently, some k-tuples have one or more selected remainders, and other k-tuples do not have any selected
remainder. If a given k-tuple has one or more selected remainders, we say that it is a prohibited k-tuple; otherwise
we say that it is a permitted k-tuple. We shall define more formally the sequence of k-tuples and related concepts in
Section 2.
Now, in a general context, a sieve is a tool or device that separates, for instance, coarser from finer particles.
Then, given a sieve device we can define a ‘sieve problem’, for instance, to count the number of finer particles that
pass through the sieve device. We can think of a sequence of k-tuples as a ‘sieve device’, in the sense that when a set
of integers is ‘fed’ into the sieve device (the sequence of k-tuples), it separates the integers associated to permitted
k-tuples from integers associated to prohibited k-tuples. The sieve problem, in this case, is to estimate the number of
integers that ‘pass through’ the sieve device; that is, to estimate the number of permitted k-tuples attached to some
of the integers in the input set.

4
 Given an even integer x > 49, we formulate the Sieve I (the Sieve √ associated with x) by means of a sequence of
k-tuples as follows. Let P be the sequence of all primes; let z = x, and let pk be the greatest prime less than
z. With the index k corresponding to the prime pk , we construct the sequence of k-tuples of remainders, where the
rules for selecting remainders are the following: If a given k-tuple of the sequence has 0 as an element, or has its
hth element equal to the remainder of dividing x by ph ∈ {p1 , p2 , p3 , . . . , pk }, these elements are defined as selected
remainders. So, within the periods of every sequence of remainders modulo ph (a given column of the matrix), the
remainder 0 is always a selected remainder, and besides, if ph does not divide x, the resulting remainder is a second
selected remainder. Let A be the set consisting of the indices n of the sequence of k-tuples that lie in the interval
[1, x]. For each p ∈ P, p < z, the set Ap ⊂ A consists of the indices n for which the corresponding element in the
sequence of remainders modulo p is a selected remainder. Then, the indices of the prohibited k-tuples lying in A are
sifted out; and the indices of the permitted k-tuples lying in A remain unsifted. The sifting function is given by the
the number of permitted k-tuples whose indices lie in the interval A . In Section 8 we shall define more formally the
formulation of the Sieve I based on a sequence of k-tuples.
Remark
√ 1.1. Note that given a k-tuple whose index is n < x, if n ≡ 0 (mod p) or n ≡ x (mod p)√for at least one
p < x, then it is a prohibited k-tuple; and if n 6≡ 0 (mod p) and n 6≡ x (mod p) for every p < x, then it is a
permitted k-tuple.

n 2 3 5 7 n 2 3 5 7
1 1 1 1 1 31 1 1 1 3
2 0 2 2 2 32 0 2 2 4
3 1 0 3 3 33 1 0 3 5
4 0 1 4 4 34 0 1 4 6
5 1 2 0 5 35 1 2 0 0 11 + 41 = 52
6 0 0 1 6 36 0 0 1 1
7 1 1 2 0 37 1 1 2 2
8 0 2 3 1 38 0 2 3 3
9 1 0 4 2 39 1 0 4 4
10 0 1 0 3 40 0 1 0 5
11 1 2 1 4 41 1 2 1 6
12 0 0 2 5 42 0 0 2 0
13 1 1 3 6 43 1 1 3 1
14 0 2 4 0 44 0 2 4 2
15 1 0 0 1 45 1 0 0 3 23 + 29 = 52
16 0 1 1 2 46 0 1 1 4
17 1 2 2 3 47 1 2 2 5
18 0 0 3 4 48 0 0 3 6
19 1 1 4 5 49 1 1 4 0
20 0 2 0 6 50 0 2 0 1
21 1 0 1 0 51 1 0 1 2
22 0 1 2 1 52 0 1 2 3
23 1 2 3 2
24 0 0 4 3
25 1 1 0 4
26 0 2 1 5
27 1 0 2 6
28 0 1 3 0
29 1 2 4 1
30 0 0 0 2

Figure 1

Therefore, given an even integer x ≥ 49 (p2k < x < p2k+1 ), it is easy to see that the ordered set of k-tuples whose
indices lie in the interval [1, x] of the sequence is only an alternative formulation of the Sieve associated with x (the

5
Sieve I), which was described before by using the usual sieve theory notation. We shall prove (Theorem 8.1) that the
indices (greater than 1) of the permitted k-tuples lying within [1, x] are primes q such that either x − q is a prime or
x − q = 1. Note that this form of the sieve gives us a detailed picture of the sifting process; other reasons for using
this formulation for sieves based on a sequence of k-tuples will be explained later.
Example 1.1. Figure 1 illustrates how the Sieve I can be used to find some Goldbach partitions for the even number
x = 52. We proceed as follows:
√
1. We make a list of the primes less than 52. We obtain {2, 3, 5, 7}.
2. We compute the remainders of dividing x = 52 by the prime moduli of the list. We obtain {0, 1, 2, 3}.
3. In every k-tuple we select each 0, and also the elements {1, 2, 3}, corresponding to the moduli {3, 5, 7}, respec-
tively. (The selected remainders are circled.)

4. Now, we colour gray the permitted k-tuples. The arrows show the corresponding Goldbach partitions. Note that
there is no permitted k-tuple for the partition 47 + 5.

1.4 The auxiliary Sieve II

To prove the Main Theorem we need to find a lower bound for the sifting function of the Sieve associated to x, for
every even number x > p2k , where k is sufficiently large.
However, we can see that, no matter the formulation, the Sieve I is a ‘static’ sieve; that is, given an even number
x, we can formulate a specific Sieve I for this even number x. For our purposes, we need a ‘dinamic’ sieve, which is
able of working as x → ∞. Suppose that given x > 49 and using the Sieve I we have a way to compute the number
of permitted k-tuples whose indices lie in [1, x]; then, we could prove the Main theorem by constructing a sequence
of sieves associated with every even number x > 49. That is, we could construct a sequence where the elements are
sequences of k-tuples, each one for every even number x > 49, and compute the number of permitted k-tuples whose
indices lie in the interval [1, x] of each sequence of k-tuples.
Now, using the Sieve I, the implementation of this idea finds some √ difficulties. For instance, if x = 50 the Sieve
I can be described as follows: Since the greatest prime less than 50 is p4 = 7, we have k = 4; so, we construct
the sequence of 4-tuples of remainders. In every 4-tuple of the sequence, if the hth element is 0, or is equal to the
remainder of dividing x by ph ∈ {p1 , p2 , p3 , p4 }, this element is a selected remainder. Let A be the set consisting of
the indices of the sequence of 4-tuples that lie in the interval [1, 50].
Suppose that we go to the next even integer x = 52. In this case, we have again k = 4, and the sequence of 4-tuples
of remainders is the same as before, but now the set A consists of the indices that lie in [1, 52], and the selected
remainders take specific values for x = 52. In √ addition, as x runs through the even numbers, when x > 121 we have
k > 4, because the greatest prime less than x will be pk > p4 = 7. The difficulty resides in the handling of all these
variables as x runs through all the even numbers. On the other hand, given k ≥ 4, when x is divisible by a prime
ph ∈ {p1 , p2 , p3 , . . . , pk }, the remainder is 0; so, in each sequence of remainders modulo ph (1 ≤ h ≤ k) that form the
sequence of k-tuples, there could exist one or two selected remainders within the period of the sequence (if there is
only one selected remainder, it is always 0). This is an additional serious difficulty in order to derive a formula for
computing the sifting function.
For all these reasons it is preferable to work with a more general kind of sieve, for which the sequence of k-tuples is
more ‘homogeneous’ than that corresponding to the Sieve I, in the sense that in each sequence of remainders modulo
ph (1 < h ≤ k) that form the sequence of k-tuples of this new sieve there exist always two selected remainders in every
period of the sequence. So, we introduce another sieve, which we call simply the Sieve II. We describe the Sieve II in
the form proposed before, by means of a sequence of k-tuples, as follows. Let P be the sequence of all primes; and
let pk (k ≥ 4) be a prime of the sequence. With the index k corresponding to the prime pk , we construct the sequence
of k-tuples of remainders, where the rules for selecting remainders are the following: In every sequence of remainders
modulo ph (1 < h ≤ k) that form the sequence of k-tuples there are always two selected remainders r and r0 modulo
ph ; in the sequence of remainders modulo p1 = 2 there is only one selected remainder r modulo p1 . Let B be the set
consisting of the indices of the sequence of k-tuples that lie in the interval [1, y], where y is an integer that satisfies
y > pk . For each p ∈ P, p ≤ pk , the set Bp ⊂ B consists of the indices n for which the corresponding element in
the sequence of remainders modulo p is a selected remainder. The indices of the prohibited k-tuples lying in B are
sifted out; and the indices of the permitted k-tuples lying in B remain unsifted. The sifting function is defined by the
equation

[
T (B, P, pk ) = B \ Bp ,
p∈P
p≤pk

6
 and counts the number of permitted k-tuples whose indices lie in B. We shall define more formally the Sieve II in
Section 2.
Remark 1.2. In this case, given a k-tuple whose index is n, if n ≡ r (mod p) or n ≡ r0 (mod p) for at least one p ≤ pk ,
where r, r0 are the selected remainders modulo p, then it is a prohibited k-tuple; and if n 6≡ r (mod p) and n 6≡ r0
(mod p) for every p ≤ pk , then it is a permitted k-tuple.
Note that the unsifted elements in B may be or may be not prime numbers; indeed, the Sieve II is a collection of
sieves, one for each particular choice of the selected remainders.

n 2 3 5 7 n 2 3 5 7
1 1 1 1 1 · · · · ·
2 0 2 2 2 176 0 2 1 1
3 1 0 3 3 177 1 0 2 2
4 0 1 4 4 178 0 1 3 3
5 1 2 0 5 179 1 2 4 4
6 0 0 1 6 180 0 0 0 5
7 1 1 2 0 181 1 1 1 6
8 0 2 3 1 182 0 2 2 0
9 1 0 4 2 183 1 0 3 1
10 0 1 0 3 184 0 1 4 2
11 1 2 1 4 185 1 2 0 3
12 0 0 2 5 186 0 0 1 4
13 1 1 3 6 187 1 1 2 5
14 0 2 4 0 188 0 2 3 6
15 1 0 0 1 189 1 0 4 0
16 0 1 1 2 190 0 1 0 1
17 1 2 2 3 191 1 2 1 2
18 0 0 3 4 192 0 0 2 3
19 1 1 4 5 193 1 1 3 4
20 0 2 0 6 194 0 2 4 5
21 1 0 1 0 195 1 0 0 6
22 0 1 2 1 196 0 1 1 0
23 1 2 3 2 197 1 2 2 1
24 0 0 4 3 198 0 0 3 2
25 1 1 0 4 199 1 1 4 3
26 0 2 1 5 200 0 2 0 4
27 1 0 2 6 201 1 0 1 5
28 0 1 3 0 202 0 1 2 6
29 1 2 4 1 203 1 2 3 0
30 0 0 0 2 204 0 0 4 1
31 1 1 1 3 205 1 1 0 2
32 0 2 2 4 206 0 2 1 3
33 1 0 3 5 207 1 0 2 4
34 0 1 4 6 208 0 1 3 5
35 1 2 0 0 209 1 2 4 6
· · · · · 210 0 0 0 0

Figure 2

Now, suppose that in the Sieve II we take B = {n : 1 ≤ n ≤ p2k }. Given an even number x > 49 that satisfies
p2k< x < p2k+1 , we can construct the sequence of k-tuples associated to the Sieve I; and using the same k, we can
construct the sequence of k-tuples associated to the Sieve II. So, we can compare for every even number x > 49 the

7
sifting function of the Sieve I with the sifting function of the attached Sieve II. That is, we can compare the number of
permitted k-tuples whose indices lie in the interval [1, x] of the sequence of k-tuples corresponding to the Sieve I, with
the number of permitted k-tuples whose indices lie in the interval [1, p2k ] of the sequence of k-tuples corresponding to
the Sieve II. We shall prove later (Lemma 8.2) for every even number x > 49 that, under the given conditions, the
value of the sifting function corresponding to the Sieve I is greater than or equal to the minimum value of the sifting
function corresponding to the Sieve II.
Example 1.2. For k = 4 (pk = 7), the period of the sequence of k-tuples is equal to 210. The first 35, and the last
35 of the 4-tuples in the interval [1, 210] (the first period of the sequence), are pictured in Figure 2, for a given choice
of selected remainders. The Sieve II is given by the k-tuples whose indices lie in [1, 72 ].
We can now construct a sequence indexed by k where every element of this sequence is a sequence of k-tuples; that
is, we have a sequence of sequences of k-tuples. In each of these sequences of k-tuples we have a Sieve II, which is given
by the ordered set of k-tuples whose indices lie in the interval [1, p2k ] of the sequence of k-tuples. So, our problem now
is, given the Sieve II, how to compute the number of permitted k-tuples whose indices lie within [1, p2k ] (the sifting
function). We shall see how the study of the sequences of k-tuples reveals the way to derive a lower bound for this
number.

1.5 Using the inclusion-exclusion principle for computing the number of permitted
k-tuples in a period of the sequence of k-tuples of the Sieve II
Usually, the sieve method consist in operate on the formula given by the inclusion-exclusion principle to obtain bounds
for the sifting function, as we have illustrated in the first subsection. In our approach, the starting point is also the
inclusion-exclusion principle, but only as a first step towards obtaining a lower bound for the sifting function of the
Sieve II. That is, from the formula given by this principle we shall compute the number of permitted k-tuples within
a period of the corresponding sequence of k-tuples, as follows.
Let us consider again the Sieve II, but now taking B = {n : 1 ≤ n ≤ mk }; that is, B is now the set of the indices
corresponding to the first period of the sequence of k-tuples. Given p ∈ P, 2 < p ≤ pk , we have |Bp | = 2mk /p, since
p|mk and there are two selected remainders r, r0 for each modulus p > 2, by definition. Furthermore, given a squarefree
integer d such that d|mk , 2 - d, the set Bd is the intersection of the subsets Bp such that p|d (p 6= 2). Hence,

2ν(d)
|Bd | = mk (d|mk , 2 - d),
d
where ν(d) is the number of distinct prime divisors of d. Furthermore, we have the identity

2ν(d) Y  
X 2
µ (d) = 1− . (4)
d p
d|mk 2<p≤pk
2-d p∈P

On the other hand, the subset Bp1 consist of the integers n ∈ B such that n ≡ r (mod p1 ), where r is the selected
remainder for the modulus p1 in the sequence of k-tuples of the Sieve II. Then |Bp1 | = mk /p1 , since p1 |mk and there
is only one selected remainder for the modulus p1 , by definition. Furthermore, given a squarefree integer d such that
d|mk , 2 | d, the set Bd is now the intersection of the subsets Bp such that p|d, and one subset Bp is Bp1 . Hence,

2ν(d)−1
|Bd | = mk (d|mk , 2 | d).
d
Now, by the inclusion-exclusion principle,

X X 2ν(d) X 2ν(d)−1
T ({n : 1 ≤ n ≤ mk }, P, pk ) = µ (d) |Bd | = µ (d) mk + µ (d) mk =
d d
d|mk d|mk d|mk
2-d 2|d
X 2ν(d) 1 X 2ν(d) 1 X 2ν(d)
= µ (d) mk − µ (d) mk = µ (d) mk .
d 2 d 2 d
d|mk d|mk d|mk
2-d 2-d 2-d

So, using (4) we can see that the number of permitted k-tuples whose indices lie in the interval [1, mk ] (the first
period of the sequence of k-tuples associated to the Sieve II) is given by

8
 1 Y  2

T ({n : 1 ≤ n ≤ mk }, P, pk ) = mk 1− , (5)
2 p
2<p≤pk
p∈P

whatever the selected remainders r, r0 (mod p), for every p ∈ P, p ≤ pk .

Note that since mk is a multiple of every p ∈ {p1 , p2 , p3 , . . . , pk }, the size of the sets Bp , and so the size of
the sets Bd can be computed exactly. This is the reason why in this case we can obtain a precise result from the
inclusion-exclusion principle.

1.6 The structure of the first period of the sequence of k-tuples of remainders
Until now, we have arranged the elements of each k-tuple horizontally, from left to right; and we have arranged the
k-tuples of the sequence vertically, from top to bottom. Hence, the first period of the sequence of k-tuples can be
seen as a matrix, with columns from h = 1 to h = k, and mk = p1 p2 p3 · · · pk rows. Note that for each h (1 ≤ h ≤ k),
we also have a sequence of h-tuples with period mh = p1 p2 p3 · · · ph , which fits into the period mk a whole number of
times.

7 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 ...

5 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 ...

3 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 ...

k
2 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 ...

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 ...

... 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0

... 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0

... 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0

... 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0

... 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210

Figure 3

Suppose that we rotate (for convenience) the entire sequence 90 degrees counterclockwise. Then, the index n of
the sequence of k-tuples increases from left to right, and the index k of the elements of each k-tuple increases from the
bottom up. Consequently, we can think of the first period of the sequence of k-tuples as a matrix formed by k rows
and mk columns. Each row of this matrix, from h = 1 to h = k, is formed by the remainders of dividing the integers
from n = 1 to n = mk by the modulus ph . For every n (1 ≤ n ≤ mk ), the corresponding column matrix is the k-tuple
of the remainders of dividing n by the moduli p1 , p2 , . . . , pk .
Note that if we let k → ∞, the period of the sequence and the size of the involved k-tuples grow simultaneously.
Example 1.3. Figure 3 illustrates the first period of the sequence of 4-tuples pictured in Figure 2, but now arranged
horizontally from left to right.
The sequences of k-tuples in general shall be defined more formally in Section 2, but now we need the following
definition:
Definition 1.1. Given a sequence of k-tuples, and using the order relation given by the index n, we define an
interval of k-tuples, denoted by I[m, n]k , to be the set of consecutive k-tuples associated with an integer interval
[m, n] ∩ Z+ , where m is the index of the first k-tuple, and n is the index of the last k-tuple. We also use the notation
I[m, n] = I[m, n]k for this interval. We define the size of I[m, n] by the equation |I[m, n]| = n − m + 1; and we use
the notation I[]k , or alternatively I[], to denote the empty interval.

9
 In particular, let us consider the sequence of k-tuples associated to the Sieve II. Since this sequence is periodic, it
suffices to consider its first period, between n = 1 and n = mk (the interval I[1, mk ]). Note that for pk ≥ 7 (k ≥ 4), the
interval I[1, p2k ] is completely included within the first period of the sequence of k-tuples. Although this is the interval
that interests us, in order to understand the properties of the sequence of k-tuples, and its behaviour as k → ∞, it is
necessary to study the whole fundamental period of the sequence, not just the interval I[1, p2k ].
The following step in our approach consists of dividing into two parts the first period of the sequence of k-tuples,
as follows: the left interval I[1, p2k ], and the right interval I[p2k + 1, mk ]. So, since for every h (1 ≤ h ≤ k) there is a
sequence of h-tuples of remainders, the interval I[1, mk ]h of each sequence turns out subdivided into two intervals: the
left interval I[1, p2k ]h , and the right interval I[p2k + 1, mk ]h . If we think of the first period of the sequence of k-tuples as
a matrix, we can see that this matrix has been now partitioned into two blocks: the left block, formed by the columns
from n = 1 to n = p2k ; and the right block, formed by the columns from n = p2k + 1 to n = mk . Each row of the left
block is formed by the remainders of dividing the integers from n = 1 to n = p2k by the modulus ph (1 ≤ h ≤ k); and
each row of the right block is formed by the remainders of dividing the integers from n = p2k + 1 to n = mk by the
modulus ph .
Recall that within the first period of the sequence of k-tuples (the interval I[1, mk ]), the exact number of permitted
k-tuples is given by (5). Furthermore, for every h such that 1 ≤ h < k, since the number of permitted h-tuples in a
period of the sequence of h-tuples is given also by (5), and the period mh divides mk , we can compute precisely the
number of permitted h-tuples in each interval I[1, mk ]h as well. We can see that for every h (1 ≤ h ≤ k), the number
of permitted h-tuples in I[1, mk ]h is the same, whatever the choice of the selected remainders in the sequence of
h-tuples. However, within both the left interval I[1, p2k ]h and the right interval I[p2k + 1, mk ]h , the number of permitted
h-tuples could change when the selected remainders in the sequence of h-tuples are changed, because the positions of
the permitted h-tuples along the interval I[1, mk ]h are modified.
Note that for every sequence of h-tuples of remainders (1 ≤ h ≤ k), the intervals I[1, mk ]h , I[1, p2k ]h and I[p2k +
1, mk ]h are itself sieve devices, that separate prohibited h-tuples from permitted h-tuples.
On the other hand, attached to the first period of the sequence of k-tuples there is a k × 2 matrix, where for every
h (1 ≤ h ≤ k), the entry in the row h and first column is the number of permitted h-tuples in I[1, p2k ]h , and the entry
in the row h and second column is the number of permitted h-tuples in I[p2k + 1, mk ]h . Of course, the entries in the
matrix depends on the choice of the selected remainders in the sequence of k-tuples. Note that if we take y = p2k in
the Sieve II, the sifting function is the entry in the first row and first column of this matrix; that is, to compute the
sifting function for the Sieve II we ought to be able to compute this entry in the matrix. Note that this quantity is
related to the entry in the first row and second column of the matrix, since the sum of both entries is given by (5).
A question may have already occurred to the reader at this point: What is the advantage of the formulation of the
sieves based in a sequence of k-tuples of remainders? We shall explain the principal reason in what follows.
Let us consider the sequence of k-tuples of the Sieve II, in horizontal position, where k ≥ 4. For a given choice of
selected remainders, the interval I[1, mk ] of this sequence is a sieve device, that sift out the prohibited k-tuples that
lie in I[1, mk ], and allows to survive the permitted k-tuples in this interval. Furthermore, for every h (1 ≤ h < k)
there is a sequence of h-tuples of remainders as well. And the interval I[1, mk ]h of every sequence of h-tuples is also
a sieve device, that sift out the prohibited h-tuples and allows to survive the permitted h-tuples in I[1, mk ]h . So,
we have decomposed the sifting process into several stages, from h = 1 to h = k, where each ‘partial’ sieve device
contributes to the whole sifting process. Hence, we can study the behaviour of this partial sieve devices to determine
the behaviour of the whole sieve; the advantage of this perspective will become apparent in the rest of this section.
Note that as h goes from 1 to k, the number of permitted h-tuples decreases, as a result of the sifting process in each
stage of the whole sifting process. Of course, there is also a similar structure in the left block and the right block of
the first period of the sequence of k-tuples.

1.7 The density of permitted k-tuples

In the Sieve II we have taken first the set B = {n : 1 ≤ n ≤ p2k }, and so, the sifting function T ({n : 1 ≤ n ≤ p2k }, P, pk )
is equal to the number of permitted k-tuples in the interval I[1, p2k ] of the sequence of k-tuples associated to the Sieve
II. Note that the evaluation of this sifting function is what we need to solve the Goldbach problem. However, this
sifting function depends on the choice of the selected remainders in the sequence of k-tuples associated to the Sieve
II, and it can not be computed exactly. The obtaining of a lower bound for this sifting function is the main task that
we must perform in this paper.
On the other hand, in the Sieve II we have next taken the set B = {n : 1 ≤ n ≤ mk }; here, the sifting function
T ({n : 1 ≤ n ≤ mk }, P, pk ) is equal to the number of permitted k-tuples in the interval I[1, mk ] of the sequence of
k-tuples associated to the Sieve II. In this case, the sifting function does not depend on the choice of the selected
remainders in the sequence of k-tuples, and it can be computed precisely using (5).
Then, a natural question arises: How can we take advantage of the exact computation of T ({n : 1 ≤ n ≤
mk }, P, pk ) for obtaining an estimate of T ({n : 1 ≤ n ≤ p2k }, P, pk )?
Let us consider the interval I[1, mk ] (first period of the sequence of k-tuples of the Sieve II); furthermore, consider
the intervals I[1, p2k ] and I[p2k + 1, mk ]. We can see that, for a given choice of selected remainders in the sequence of

10
k-tuples, if the proportion of permitted k-tuples in I[1, p2k ] is less than the proportion in I[1, mk ], the proportion of
permitted k-tuples in I[p2k + 1, mk ] must be greater than the proportion in I[1, mk ]; and vice versa.
Suppose that the proportion of permitted k-tuples in the interval I[1, p2k ] were equal to the proportion of permitted
k-tuples in the interval I[1, mk ]. In this case, we could compute at once the exact number of permitted k-tuples in
the interval I[1, p2k ], since we know this quantity for the interval I[1, mk ], by (5). Certainly, it is unlikely that our
assumption on the proportion of permitted k-tuples in these intervals could be true; however, we could say that in
some sense this assumption is ‘aproximately’ true. This suggests the possibility of working with this amount (the
proportion of permitted k-tuples in a given interval) to obtain the expected results.
Now, assume that for every k the proportion of permitted k-tuples in I[1, p2k ] were greater than some constant
C > 0; in this case, the number of permitted k-tuples within this interval would be greater than Cp2k ; this would imply
that the number of permitted k-tuples within I[1, p2k ] tends to infinity with k; and, indeed, this is the result that we
wish to prove. However, it seems unlikely that this constant exists, since from (5) it follows that the proportion of
permitted k-tuples in the interval I[1, mk ] is given by

k  
1Y 2
1− , (6)
2 ph
h=2

which tends slowly to 0, as k → ∞. This fact makes working with this amount (the proportion of permitted
k-tuples) not very useful.
For this reason it is more convenient to work with a new quantity, that we call the density of permitted k-tuples,
or simply the k-density, which is defined formally in Section 3. It is defined for a given interval as the quotient of
the number of permitted k-tuples within this interval and the number of subintervals of size pk . That is, for a given
interval, is the average number of permitted k-tuples within the subintervals of size pk . We denote by ck and δk the
number of permitted k-tuples, and the density of permitted k-tuples within the period I[1, mk ] of the sequence of
k-tuples, respectively (see Definition 2.9 and Definition 3.3). Since ck does not depend on the choice of the selected
remainders in the sequence of k-tuples, neither does δk depend on that choice. We shall prove later (Lemma 3.2 and
Theorem 3.4) that δk increases and tends to infinity as k → ∞. For some values of k, Table 1 gives ck , the ratio
ck /mk , and δk . Note that given the proportion of permitted k-tuples within a given interval, multiplying by pk we
obtain the density of permitted k-tuples within this interval.
Suppose that the minimum value of the density of permitted k-tuples within the left interval I[1, p2k ] were greater
than some constant C > 0; this would imply that the number of permitted k-tuples within this interval is greater than
Cpk , since the number of subintervals of size pk in I[1, p2k ] is equal to pk . And this, in turn, would imply that the
number of permitted k-tuples in I[1, p2k ] tends to infinity, as k → ∞. In other words, the sifting function of the Sieve
II would tend to infinity, as k → ∞; and this is the result that we need in order to prove the Main theorem.
Attached to the first period of the sequence of k-tuples there is now another k × 2 matrix, where for every
h (1 ≤ h ≤ k), the entry in the row h and first column is the density of permitted h-tuples in I[1, p2k ]h , and the entry
in the row h and second column is the density of permitted h-tuples in I[p2k + 1, mk ]h . The entries in this matrix
also depends on the choice of the selected remainders in the sequence of k-tuples. Note that the entry in the first row
and first column of this new k × 2 matrix, multiplied by pk , is equal to the entry in the first row and first column of
the k × 2 matrix described in the preceding subsection. The relationships in the matrix of h-densities, between the
elements of the rows, and between the elements of the columns are very important for our purposes, as we shall see
later.

1.8 Simple explanation of the main ideas

Let us consider again the interval I[1, mk ] (the first period) of the sequence of k-tuples of the Sieve II, in horizontal
position, for k sufficiently large. As we have seen before, we can consider the first period of the sequence of k-tuples as
a matrix of k rows and mk columns. Recall that for every h (1 ≤ h ≤ k), the rows from 1 to h are part of a sequence
of h-tuples; and we shall say that h is the level of this sequence. Note that for k large enough δh increases between
h = 4 and h = k (see Lemma 3.2 and Corollary 3.3); in this case δk > δ4 = 1/2 (see Table 1).
Suppose that for every level h (1 ≤ h ≤ k), the permitted h-tuples were placed in positions that follow an
approximately regular pattern along the interval I[1, mk ] of the corresponding sequence of h-tuples, whatever the
choice of the selected remainders in the sequence of k-tuples. In this case, for each level h, the density of permitted
h-tuples in both intervals I[1, p2k ] and I[p2k + 1, mk ] of the sequence of h-tuples should be close to the density within
I[1, mk ]; that is, it should be close to δh , since mk is a multiple of mh . Therefore, since δh increases from level h = 4 to
level h = k, the density of permitted h-tuples in the interval I[1, p2k ] of every sequence of h-tuples should also increase
from h = 4 to h = k, whatever the choice of the selected remainders. So, the density of permitted k-tuples in the
interval I[1, p2k ] of the sequence of k-tuples (no matter the choice of the selected remainders) should be greater than
δ4 .
Now, in order to explain in a way as simple as possible how through these ideas can be achieved a favorable outcome
in the resolution of the Goldbach’s problem, we shall make the following assumption. We assume that there exists an

11
 Table 1: Quotient ck /mk and density δk .

k pk mk ck ck /mk δk
4 7 210 15 0.071 0.500
5 11 2310 135 0.058 0.643
6 13 30030 1485 0.049 0.643
7 17 510510 22275 0.044 0.742
8 19 9699690 378675 0.039 0.742
9 23 223092870 7952175 0.036 0.820
10 29 - - 0.033 0.962
11 31 - - 0.031 0.962
12 37 - - 0.029 1.087
13 41 - - 0.028 1.145
14 43 - - 0.027 1.145
15 47 - - 0.025 1.199
16 53 - - 0.024 1.301
17 59 - - 0.024 1.399
18 61 - - 0.023 1.399
19 67 - - 0.022 1.490
20 71 - - 0.022 1.535
21 73 - - 0.021 1.535
22 79 - - 0.020 1.619
23 83 - - 0.020 1.660
24 89 - - 0.019 1.740
. . . . . .

integer Kα > 5 such that the density of permitted k-tuples in the interval I[1, p2k ] of the sequence of k-tuples (for all
the choices of the selected remainders) is greater than δ4 = 1/2 for every k > Kα . So, since the interval I[1, p2k ] is
the union of pk subintervals of size pk , we can see that the quantity of permitted k-tuples in I[1, p2k ] (the value of the
Sifting function of the Sieve II) must be greater than pk δ4 = pk /2, for k > Kα . Now, we have seen before that for
all even number x that satisfies p2k < x < p2k+1 , the minimum value of the Sifting function of the Sieve II is a lower
bound for the Sifting function of the Sieve I. From this it follows that the Sifting function of the Sieve I is also greater
than pk /2 for all even number x such that p2k < x < p2k+1 (k > Kα ). And it is easy to see that this fact proves the
Goldbach’s conjecture for all even number x > p2k , where k > Kα . We shall prove the existence of Kα and we shall
give an upper bound for this number.

2 Periodic sequences of k-tuples

General Notation. We write (a, b) for the greatest common divisor of a and b, if no confusion will arise. In addition,
lcm is used as an abbreviation for the least common multiple. Given a set A, we denote by |A| the cardinality of A.
For each a ∈ R, the symbol bac denotes the floor function, and the symbol dae denotes the ceiling function.
In the Introduction we began by describing a first kind of sieve to attack the Goldbach’s problem, which we call
the Sieve associated with x (or Sieve I); then, we have introduced the notion of sequence of k tuples of remainders as a
new formulation for sieves in general, and for this sieve in particular. The Sieve associated with x (Sieve I) is directly
related to the Goldbach’s problem; we defer to Section 8 the formal definition of this sieve. On the other hand, we
have also described in the Introduction a second sieve more general, which we call the Sieve II. As we have seen in the
Introduction, the sequence of k-tuples corresponding to the Sieve II is more homogeneous than that corresponding to
the Sieve I, in the sense that in every sequence of remainders modulo ph (1 < h ≤ k) there are always two selected
remainders. This fact is very important in order to compute the minimum value of the sifting function of the Sieve II.
The Sieve II is not directly related to the Goldbach’s problem, but, as we have seen in the Introduction, the
minimum number of permitted k-tuples in the interval I[1, p2k ] of the sequence of k-tuples corresponding to the Sieve
II (the minimum value of the sifting function of the Sieve II), is a lower bound for the number of permitted k-
tuples in the interval I[1, x] of the sequence corresponding to the Sieve I (the sifting function of the Sieve I), where
p2k < x < p2k+1 ; we shall prove this fact in Section 8. In this section we define formally the Sieve II; and we shall deal
with the properties of this sieve until Section 7.
We begin by defining the concepts of sequence of remainders and sequence of k-tuples of remainders, and other
associated concepts.

12
 Definition 2.1. Let P be the sequence of all primes, and consider the subset {p1 , p2 , p3 , . . . , pk } of the first k
primes.

(1) Given ph (1 ≤ h ≤ k), we define the periodic sequence {rn }, where rn denotes the remainder of dividing n by
the modulus ph . We denote the sequence {rn } by the symbol sh . The period of the sequence is equal to ph . See
Example 2.1.
(2) We define the sequence {(r1 , r2 , r3 , . . . , rk )n }, the elements of which are k-tuples of remainders obtained by dividing
n by the moduli p1 , p2 , p3 , . . . , pk . We arrange the sequence of k-tuples of remainders vertically; we usually omit
the comma separators in the k-tuples. Then, the sequence of k-tuples can be seen as a matrix formed by k columns
and infinitely many rows, where each column of the matrix is a periodic sequence sh (1 ≤ h ≤ k). We call the
index k the level of the sequence of k-tuples of remainders. See Example 2.2.

Example 2.1. For the modulus p3 = 5 we have s3 = {1, 2, 3, 4, 0, 1, 2, 3, . . .}.

Example 2.2. Table 2 shows the first elements of the sequence of 5-tuples of the remainders of dividing n by
{2, 3, 5, 7, 11}.

Table 2: Sequence of 5-tuples of remainders.

n 2 3 5 7 11
1 1 1 1 1 1
2 0 2 2 2 2
3 1 0 3 3 3
4 0 1 4 4 4
5 1 2 0 5 5
6 0 0 1 6 6
7 1 1 2 0 7
8 0 2 3 1 8
9 1 0 4 2 9
10 0 1 0 3 10
11 1 2 1 4 0
12 0 0 2 5 1
13 1 1 3 6 2
14 0 2 4 0 3
15 1 0 0 1 4
16 0 1 1 2 5
17 1 2 2 3 6
18 0 0 3 4 7
. . . . . .

Definition 2.2. Given a sequence {rn } with prime modulus pk we assign to the remainders rn one of the two
following states: selected state or not selected state.
Definition 2.3. Given a sequence of k-tuples of remainders, we define a k-tuple to be prohibited if it has one or
more selected remainders, and we define it to be permitted if it contains no selected remainders.
Definition 2.4. We denote by mk the product p1 p2 p3 · · · pk .
Proposition 2.1. The sequence of k-tuples of remainders is periodic, and its fundamental period is equal to mk =
p1 p2 p3 · · · pk .

Proof. Let sh (1 ≤ h ≤ k) be the sequences of remainders that form a given sequence of k-tuples. Let m0 be a multiple
of all the primes p1 , p2 , p3 , . . . , pk . The period of every sequence sh is equal to ph ∈ {p1 , p2 , p3 , . . . , pk }. Therefore, for
every sequence sh , the remainders are repeated for all the integer intervals of size m0 , starting from the index n = 1
onward. Since p1 , p2 , p3 , . . . , pk are primes, the product mk is the lcm. Consequently, the fundamental period of the
sequence of k-tuples is equal to mk .

13
 So far, we have defined the sequence of k-tuples of remainders without defining any rules for selecting remainders;
note that without selected remainders, the sequence of k-tuples does not work as a sieve. Before defining these rules,
we shall consider another important question concerning the sequence of k-tuples of remainders. As we have seen in
the Introduction, in the case of the Sieve II it will be necessary to deal with the behaviour of the sequence of k-tuples as
k increases indefinitely. Consequently, we need two more definitions before defining the rules for selecting remainders.
Definition 2.5. Sum of sequences.
Let {p1 , p2 , p3 , . . . , pk } be the set of the first k primes. Let {(r1 r2 r3 . . . rk )n } be the sequence of k-tuples of
the remainders of dividing n by the k prime moduli {p1 , p2 , p3 , . . . , pk }, and let {(rk+1 )n } be the sequence of the
remainders of dividing n by the prime modulus pk+1 . We define the sum {(r1 r2 r3 . . . rk )n } + {(rk+1 )n }, of the
sequence {(r1 r2 r3 . . . rk )n } and the sequence {(rk+1 )n }, to be the sequence of (k + 1)-tuples given by the equation

{(r1 r2 r3 . . . rk )n } + {(rk+1 )n } = {(r1 r2 r3 . . . rk rk+1 )n } ,

and formed by the ordered juxtaposition of each k-tuple of the first sequence with each element (index n modulo
pk+1 ) of the second sequence.

Definition 2.6. Let P be the sequence of all primes, and let pk ∈ P. Let sk be the sequence of the remainders
P of
dividing n by the modulus pk . Let {sk } be the sequence of sequences sk . We define the series denoted by sk to be
the sequence {Sk }, where Sk denotes the partial sum:

S1 = s1 ,
S2 = s1 + s2 ,
S3 = s1 + s2 + s3 ,
. . . . . . . . . . . .
Sk = s1 + s2 + s3 + s4 + ··· + sk ,
P
and the symbol refers to the formal addition of sequences. In each partial sum Sk , the greatest prime modulus
pk will be called the characteristic prime modulus of the partial sum Sk . The index k will be called the level, and we
shall say that Sk is the partial sum of level k.
Example 2.3. Table 3 shows the partial sum S4 and the formal addition of the sequence of remainders s5 to obtain
the partial sum S5 .
On the one hand, we can look at a given partial sum Sk as a sequence indexed by n, of the k-tuples of remainders
obtained by dividing n by the moduli p1 , p2 , p3 , . . . , pk . On the other hand, the partial sum Sk can be seen as a finite
sequence indexed by the set {1, . . . , k} (k ∈ Z+ ), of sequences P of remainders modulo ph ∈ {p1 , p2 , p3 , . . . , pk }, where
the indices {1, . . . , k} increase from left to right. And the series sk is the sequence indexed by k, of the partial sums
Sk .
Now we are ready to define P the rules for selecting remainders in the sequences sh (1 ≤ h ≤ k) that make up every
partial sum Sk of the series sk .
Definition 2.7. Let sh (1 ≤ h ≤ k) be one of the sequences of remainders that form the partial sum Sk .

Rule 1. If h = 1, in the sequence of remainders s1 there will be selected one remainder, the same one in every period
of the sequence.
Rule 2. If 1 < h ≤ k, in every sequence of remainders sh there will be selected two remainders, the same two in every
period of the sequence.

Example 2.4. Table 4 shows the partial sum of level k = 4, where the selected remainders can be seen marked
between the square brackets [ ]. Note that the 4-tuples 1 and 7 are permitted k-tuples.
Properly speaking, a given partial sum Sk is a sequence of k-tuples of remainders. However, from now on, when
we refer to a given partial sum Sk , we mean Sk together with the selected remainders, unless we specifically state
otherwise. Now we are ready to define formally the Sieve II.
Definition 2.8. Let P be the sequence of all primes; and let pk (k ≥ 4) be a prime of the sequence. Let B be
the set consisting of the indices of the partial sum Sk that lie in the interval [1, y], where y is an integer that satisfies
y > pk . For each p = ph ∈ P (1 ≤ h ≤ k), the subset Bp of B consists of the indices whose remainder modulo p = ph
is one of the selected remainders r or r0 . The indices of the prohibited k-tuples lying in B are sifted out; and the
indices of the permitted k-tuples lying in B remain unsifted. See Remark 1.2. The sifting function

14
 Table 3: Partial sums S4 and S5 .

S4 s5 S5
n 2 3 5 7 11 2 3 5 7 11
1 1 1 1 1 1 1 1 1 1 1
2 0 2 2 2 2 0 2 2 2 2
3 1 0 3 3 3 1 0 3 3 3
4 0 1 4 4 4 0 1 4 4 4
5 1 2 0 5 5 1 2 0 5 5
6 0 0 1 6 6 0 0 1 6 6
7 1 1 2 0 7 1 1 2 0 7
8 0 2 3 1 8 0 2 3 1 8
9 1 0 4 2 + 9 = 1 0 4 2 9
10 0 1 0 3 10 0 1 0 3 10
11 1 2 1 4 0 1 2 1 4 0
12 0 0 2 5 1 0 0 2 5 1
13 1 1 3 6 2 1 1 3 6 2
14 0 2 4 0 3 0 2 4 0 3
15 1 0 0 1 4 1 0 0 1 4
16 0 1 1 2 5 0 1 1 2 5
17 1 2 2 3 6 1 2 2 3 6
18 0 0 3 4 7 0 0 3 4 7
. . . . . . . . . . .

[
T (B, P, pk ) = B \ Bp ,
p∈P
p≤pk

is given by the the number of permitted k-tuples whose indices lie in the interval B.
Hereafter until the end of the paper, we take B = {n : 1 ≤ n ≤ p2k }. P
In the following theorems we prove some other properties of the partial sums of the series sk , which will be used
throughout this paper.
Proposition 2.2. Let Sk be a given partial sum. Let sk+1 be the sequence of remainders modulo pk+1 . Let
r (0 ≤ r < pk+1 ) be one of the remainders modulo pk+1 of the sequence sk+1 . Let n ∈ Z+ be the index of a given
k-tuple of Sk . Then, when we juxtapose the elements of the sequence sk+1 to the right of each k-tuple of Sk , we have
the following.

(1) If the k-tuple at position n is prohibited, then the (k + 1)-tuple of Sk+1 at position n will be prohibited as well.
(2) If the k-tuple at position n is permitted and n ≡ r (mod pk+1 ), then:

(a) The (k + 1)-tuple of Sk+1 at position n is prohibited if and only if r is a selected remainder;
(b) The (k + 1)-tuple of Sk+1 at position n is permitted if and only if r is not a selected remainder.

Proof. By definition, a given k-tuple is prohibited if it has one or more selected remainders; if it has no selected
remainder, the k-tuple is permitted. The proof is immediate.

Definition 2.9. For a given partial sum Sk , we denote by ck the number of permitted k-tuples within a period of
Sk .
Proposition 2.3. Let Sk be a given partial sum. We have: ck = (p1 − 1)(p2 − 2)(p3 − 2) · · · (pk − 2).
Proof. It follows from (5), by simplifying the expression.

15
 Table 4: Partial sum S4 with selected remainders.

n 2 3 5 7
1 1 1 1 1
2 [0] [2] 2 2
3 1 [0] [3] [3]
4 [0] 1 4 4
5 1 [2] [0] [5]
6 [0] [0] 1 6
7 1 1 2 0
8 [0] [2] [3] 1
9 1 [0] 4 2
10 [0] 1 [0] [3]
11 1 [2] 1 4
12 [0] [0] 2 [5]
13 1 1 [3] 6
14 [0] [2] 4 0
15 1 [0] [0] 1
16 [0] 1 1 2
. . . . .

Lemma 2.4. Let Sk be a given partial sum. Let pk be the characteristic prime modulus of the partial sum Sk . Let
ck be the number of permitted k-tuples within the period of Sk . We have p2k = o(ck ).
Proof. Using Proposition 2.3, we have

p2k p2k
= = (7)
ck (p1 − 1) (p2 − 2) (p3 − 2) · · · (pk − 2)
   
1 pk pk
= .
(p1 − 1) (p2 − 2) (p3 − 2) · · · (pk−2 − 2) pk−1 − 2 pk − 2

Let gk−1 denote the gap pk − pk−1 ; so, pk /(pk−1 − 2) = (pk−1 + gk−1 )/(pk−1 − 2). By the Bertrand-Chebyshev
theorem, we have gk−1 < pk−1 =⇒ (pk−1 +gk−1 )/(pk−1 −2) < 2pk−1 /(pk−1 −2). It follows that limk→∞ pk /(pk−1 −2) <
limk→∞ 2pk−1 /(pk−1 − 2) = 2. Since limk→∞ pk /(pk − 2) = 1, returning to (7), it is easy to see that limk→∞ p2k /ck = 0.

Lemma 2.5. Let Sk be a given partial sum. Let pk be the characteristic prime modulus of the partial sum Sk . Let
mk be the period of Sk . We have p2k = o(mk ).
Proof. Since mk = p1 p2 p3 · · · pk by definition, the proof follows at once from Proposition 2.3 and Lemma 2.4.

Proposition 2.6. The Construction Procedure P

Let Sk and Sk+1 be consecutive partial sums of the series sk . Let mk and mk+1 be the periods of Sk and Sk+1 ,
respectively. Consider the following procedure. First we take pk+1 periods of the partial sum Sk . Next we juxtapose the
remainders of the sequence sk+1 to the right of each k-tuple of Sk (that is to say, we perform the operation Sk + sk+1 ).
This produces a whole period of the partial sum Sk+1 .
Proof. By Proposition 2.1, the period mk of the partial sum Sk is equal to mk = p1 p2 p3 · · · pk . If we repeat pk+1 times
the period of the partial sum Sk , the total number of k-tuples will be mk pk+1 = p1 p2 p3 · · · pk pk+1 = mk+1 . Thus,
when we add the sequence sk+1 , the number of (k + 1)-tuples of Sk+1 that we obtain is equal to mk+1 , that is to say,
a period of Sk+1 .

By the Construction Procedure, to get a period of the partial sum Sk+1 , we first take pk+1 periods of the partial
sum Sk . The following proposition shows that the distribution of the permitted k-tuples that are within the pk+1
periods of the partial sum Sk over the residue classes modulo pk+1 is uniform.

16
 Proposition 2.7. The permitted k-tuples within the first pk+1 periods of the partial sum Sk are uniformly distributed
over the residue classes modulo pk+1 .
Proof. Let ck be the number of permitted k-tuples within a period of Sk . Let [y] = [0], [1], [2], . . . , [pk+1 − 1] be
the residue classes modulo pk+1 . Let n ∈ Z+ be the index of a given permitted k-tuple within the first period of
the partial sum Sk . Thus, within pk+1 periods of the partial sum Sk there are pk+1 permitted k-tuples with indices
n0 = mk x + n, where x = 0, 1, 2, 3, . . . , pk+1 − 1 represents each period. Because (mk , pk+1 ) = 1, for each residue class
[y] the congruence mk x + n ≡ y (mod pk+1 ) has a unique solution x. Therefore, since there are ck permitted k-tuples
within the period of Sk , it follows that there are ck permitted k-tuples within each residue class modulo pk+1 , and the
resulting distribution is uniform.

Corollary 2.8. If there are m0 consecutive periods of the partial sum Sk (including the first), where m0 is a multiple
of pk+1 , the permitted k-tuples within these m0 periods are also uniformly distributed over the residue classes modulo
pk+1 .

3 Definition and properties of the density of permitted k-tuples

In this section, we define more formally the concept of the density of permitted k-tuples, and we prove that the density
of permitted k-tuples within a period of the partial sum Sk is increasing and tends to ∞ as k → ∞.
P
Definition 3.1. Let Sk be a given partial sum of the series sk ; let I[m, n] be a given interval of k-tuples. We
I[m,n]
denote by ck the number of permitted k-tuples within I[m, n]. By abuse of notation, we normally omit specific
I[m,n]
mention of the integer interval [m, n] ∩ Z+ and write cIk instead of ck if no confusion will arise.
P
Definition 3.2. Let Sk be a partial sum of the series sk ; let I[m, n] be a given interval of k-tuples. The number
of subintervals of size pk in this interval is equal to |I[m, n]|/pk . We define the density of permitted k-tuples in the
interval I[m, n] (or simply the k-density) by

I[m,n] cIk
δk = .
|I [m, n]| /pk
I[]
For the empty interval we define δk = 0. By abuse of notation, we often omit specific mention of the integer
I[m,n]
interval [m, n] ∩ Z+ and write δkI instead of δk .
Remark 3.1. The density of permitted k-tuples is the average number of permitted k-tuples inside subintervals of
size pk .
P
Definition 3.3. Let Sk be a given partial sum of the series sk ; let mk be the period of Sk . Recall that we have
I[1,mk ]
used the notation ck = ck for the number of permitted k-tuples within the interval I[1, mk ] (the first period of
I[1,mk ]
Sk ). We normally use the notation δk = δk for the density of permitted k-tuples within the interval I[1, mk ].
Since mk /pk is the number of subintervals of size pk within a period of Sk , by definition, we have

ck
δk = .
mk /pk
By Proposition 2.3, the number of permitted k-tuples within the interval I[1, mk ] (the first period of Sk ), does
not depend on which are the selected remainders in the sequences of remainders that form Sk . Therefore, we think of
I[1, mk ] as being a special interval, and this explains why we use the special notation ck for the number of permitted
k-tuples within I[1, mk ], and δk for the density of permitted k-tuples within I[1, mk ].
Example 3.1. The period of the partial sum S4 is equal to m4 = 2 × 3 × 5 × 7 = 30 × 7 = 210, and the number of
permitted 4-tuples within the period is equal to c4 = (2 − 1)(3 − 2)(5 − 2)(7 − 2) = 15. Then

c4 15 1
δ4 = = = .
m4 /p4 30 2
The following lemma gives a formula for computing δk .
Lemma 3.1. We have

     
p1 − 1 p2 − 2 p3 − 2 pk−1 − 2
δk = ··· (pk − 2) .
p1 p2 p3 pk−1

17
Proof. The size of the period of the partial sum Sk is equal to mk = p1 p2 p3 p4 · · · pk−1 pk (see Proposition 2.1).
Therefore, the number of subintervals of size pk is equal to (p1 p2 p3 p4 · · · pk−1 pk )/pk = p1 p2 p3 p4 · · · pk−1 . On the other
hand, the number of permitted k-tuples within a period of Sk is equal to ck = (p1 −1)(p2 −2)(p3 −2) · · · (pk−1 −2)(pk −2),
by Proposition 2.3. Consequently, by definition, we obtain

(p1 − 1) (p2 − 2) (p3 − 2) · · · (pk−1 − 2) (pk − 2)

δk = =
p1 p2 p3 p4 · · · pk−1
     
p1 − 1 p2 − 2 p3 − 2 pk−1 − 2
= ··· (pk − 2) .
p1 p2 p3 pk−1

The next lemma shows that δk is increasing if k > 1.

P
Lemma 3.2. Let Sk and Sk+1 be consecutive partial sums of the series sk . If δk denotes the density of permitted
k-tuples within a period of Sk , and δk+1 denotes the density of permitted (k + 1)-tuples within a period of Sk+1 , then

 
pk+1 − 2
δk+1 = δk .
pk

The proof is given in Section 6.

Corollary 3.3. By Lemma 3.2,

1. pk+1 − pk < 2 =⇒ δk+1 < δk .

2. pk+1 − pk = 2 =⇒ δk+1 = δk .
3. pk+1 − pk > 2 =⇒ δk+1 > δk .

Example 3.2. The characteristic prime moduli of the partial sums S4 and S5 are p4 = 7 and p5 = 11. The
period of the partial sum S4 is m4 = 2 × 3 × 5 × 7 = 30 × 7 = 210, and the number of permitted 4-tuples is
c4 = (2 − 1) (3 − 2) (5 − 2) (7 − 2) = 15. Then δ4 = 15/30 = 0.500. On the other hand, the period of the par-
tial sum S5 is m5 = 2 × 3 × 5 × 7 × 11 = 210 × 11 = 2310, and the number of permitted 5-tuples is c5 =
(2 − 1) (3 − 2) (5 − 2) (7 − 2) (11 − 2) = 135. Then δ5 = 135/210 ≈ 0.643. Note that since 7 and 11 are not twin
primes, δ5 > δ4 (see Corollary 3.3).
Now we prove that δk → ∞ as k → ∞. First, we make a definition.
Definition 3.4. Let pk > 2 and pk+1 be consecutive primes. We denote by θk the difference pk+1 − pk − 2.
Theorem 3.4. Let Sk be a given partial sum. Let δk be the density of permitted k-tuples within a period of Sk . As
k → ∞, we have δk → ∞.
Proof. Lemma 3.1 implies

       
p1 − 1 p2 − 2 p3 − 2 p4 − 2 p5 − 2 pk−1 − 2
δk = ··· (pk − 2) .
p1 p2 p3 p4 p5 pk−1

If we shift denominators to the right, we obtain

       
p2 − 2 p3 − 2 p4 − 2 p5 − 2 pk−1 − 2 pk − 2
δk = (p1 − 1) ··· .
p1 p2 p3 p4 pk−2 pk−1

By definition, θk = pk+1 − pk − 2 =⇒ pk+1 − 2 = pk + θk . Consequently, we can write the expression of δk as

      
1 p 2 + θ2 p3 + θ3 p4 + θ4 pk−2 + θk−2 pk−1 + θk−1
δk = ··· =
2 p2 p3 p4 pk−2 pk−1
      
1 θ2 θ3 θ4 θk−2 θk−1
= 1+ 1+ 1+ ··· 1 + 1+ =
2 p2 p3 p4 pk−2 pk−1
       
1 1 θ2 θ3 θ4 θk−1 θk pk
= 1+ 1+ 1+ 1+ ··· 1 + 1+ .
3 p1 p2 p3 p4 pk−1 pk pk + θk

18
 Then

∞ 
" Y #
1 1 θk pk
lim δk = 1+ 1+ lim . (8)
k→∞ 3 p1 pk k→∞ pk + θk
k=2

The infinite product between square brackets diverges if the series

∞
1 X θk
+ (9)
p1 pk
k=2

diverges. In the series

P∞ (9), if pk is the first of a pair of twin primes, by definition we have θk = 0; otherwise
we have θk ≥ 2. Let j=1 1/qj denote the series where every prime qj is the first of a pair of twin primes. Since
P∞
the series of reciprocals of the twin primes converges [3], the series j=1 1/qj also converges. Therefore, the series
P∞ P∞ P∞ P∞ P∞
k=1 1/pk − j=1 1/qj diverges, because k=1 1/pk diverges. By comparison with the series k=1 1/pk − j=1 1/qj ,
it follows that the series (9) diverges, because θk /pk > 1/pk for the terms where θk > 0. Thus, the infinite product in
(8) tends to ∞ as well. On the other hand, by the Bertrand–Chebyshev theorem, pk < pk+1 < 2pk =⇒ θk < pk =⇒
pk /(pk + θk ) > 1/2. Consequently, δk → ∞ as k → ∞.

4 The average density of permitted k-tuples within a given interval

I[m, n]
P
Let Sk be a given partial sum of the series sk . In Section 3 we showed that, for the interval I[1, mk ] of the partial
sum Sk (the first period), the density of permitted k-tuples does not depend on the choice of the selected remainders
in the sequences sh (1 ≤ h ≤ k) that form the partial sum Sk (see Lemma 3.1). However, it is easy to see that this
assertion does not hold for all the intervals I[m, n] of the partial sum Sk . In this section we prove that, within a given
interval I[m, n] of the partial sum Sk , the average of the values of the k-density for all the possible choices of the
selected remainders is equal to δk . First, we make some definitions.
Definition 4.1. Let sh (1 ≤ h ≤ k) be the sequences of remainders that form the partial sum Sk . A given choice
of the selected remainders within the period of one of the sequences sh , or within the periods of all the sequences
sh (1 ≤ h ≤ k), will be called a combination of selected remainders. We denote by νh the number of combinations of
selected remainders within the period of a given sequence sh . Since, by definition, for the sequences sh (1 < h ≤ k)
there are two selected remainders within the period ph ,

 
ph
νh = . (10)
2

In the sequence s1 there is only one selected remainder within the period; then, p1 = 2 =⇒ ν1 = 2. We denote by
Nk the number of combinations of selected remainders within the periods of all the sequences sh (1 ≤ h ≤ k). Then

     
p1 p2 p3 pk
Nk = ··· . (11)
1 2 2 2

Convention. From now on, when we refer to the average density of permitted k-tuples within a given interval I[m, n]
of the partial sum Sk , we mean that this average is computed taking into account all the combinations of selected
remainders in the sequences sh that form the partial sum Sk . We use the same convention when we refer to the average
number of permitted k-tuples.
Definition 4.2. The operation of Type A.
Let sh (1 ≤ h ≤ k) be the sequences of remainders that form the partial sum Sk . For h > 1, let r, r0 (mod ph ) be
the selected remainders within the period ph of the sequence sh . We define the operation that changes the selected
remainders r, r0 (mod ph ) to r + 1, r0 + 1 (mod ph ) to be the Type A operation.
For the sequence s1 , we also define the operation of changing the selected remainder r (mod p1 ) to r + 1 (mod p1 )
to be the operation of Type A.
Example 4.1. Table 5 shows the first period of the sequence of remainders s4 (p4 = 7), where initially we select the
remainders [1] and [3] and then we apply successively the Type A operation.

19
 Table 5: First period of the sequence of remainders s4 .

n
1 [1] 1 1 1 1 [1] 1
2 2 [2] 2 2 2 2 [2]
3 [3] 3 [3] 3 3 3 3
4 4 [4] 4 [4] 4 4 4
5 5 5 [5] 5 [5] 5 5
6 6 6 6 [6] 6 [6] 6
7 0 0 0 0 [0] 0 [0]

Definition 4.3. The operation of Type B.

Let sh (1 < h ≤ k) be the sequences of remainders that form the partial sum Sk . Let r, r0 (mod ph ) be the selected
remainders (in that order), within the period ph of the sequence sh . We define the Type B operation as follows:
1) The remainder r holds selected.
2) We change the other selected remainder r0 (mod ph ) to r0 + 1 (mod ph ), r 6= r0 + 1.

Example 4.2. Table 6 shows the first period of the sequence of remainders s4 (p4 = 7), where initially we selected
the remainders [1] and [2], and then we applied successively the Type B operation.

Table 6: First period of the sequence of remainders s4 .

n
1 [1] [1] [1] [1] [1] [1]
2 [2] 2 2 2 2 2
3 3 [3] 3 3 3 3
4 4 4 [4] 4 4 4
5 5 5 5 [5] 5 5
6 6 6 6 6 [6] 6
7 0 0 0 0 0 [0]

Definition 4.4. Let sh (1 ≤ h ≤ k) be a given sequence of remainders modulo ph . We define νhA by νhA = ph , and
we define νhB (h > 1) by νhB = (ph − 1)/2.
Remark 4.1. Suppose that we choose two consecutive selected remainders r, r0 within the period of the sequence
sh (1 < h ≤ k). So, we have one out of νh combinations of selected remainders. Repeating the Type A operation νhA −1
times, we obtain νhA = ph different combinations of selected remainders. Now, if for each one of these combinations
we leave unchanged the selected remainder r, and then we repeat νhB − 1 times the Type B operation, we obtain all
the νh combinations of selected remainders within the period of the sequence sh . This is expressed by the equation

 
ph ph ! ph − 1
νh = = = ph = νhA νhB .
2 2! (ph − 2)! 2

Definition 4.5. Let Sk and Sk+1 be the partial sums of level k and k + 1. Let sk+1 be the sequence of remainders
of level k + 1. Let I[m, n]k be an interval of k-tuples of Sk , and let I[m, n]k+1 be an interval of (k + 1)-tuples of Sk+1 ,
where the indices m, n are the same for both intervals. When we juxtapose the remainders of the sequence sk+1 to
the right of each k-tuple of Sk , then, by Proposition 2.2, the permitted k-tuples of Sk , whose indices are congruent
to a given selected remainder of sk+1 modulo pk+1 , are converted to prohibited (k + 1)-tuples of Sk+1 . We denote by
fk+1 the fraction of the permitted k-tuples within the interval I[m, n]k that are converted to prohibited (k + 1)-tuples
within the interval I[m, n]k+1 . For the partial sum S1 , let f1 denote the fraction of the prohibited 1-tuples within the
interval I[m, n]k=1 .
We denote by f k+1 the average of fk+1 for all the combinations of selected remainders in the sequence sk+1 (k ≥ 1).
For the partial sum S1 , let f 1 denote the average of f1 for the 2 combinations of selected remainders in the sequence
s1 .
The following lemma gives a formula for computing the average fraction f k+1 .

Lemma 4.1. For k ≥ 1 we have f k+1 = 2/pk+1 . For S1 we have f 1 = 1/p1 .

20
Proof. Let [0], [1], [2], . . . , [pk+1 − 1] be the residue classes modulo pk+1 . Let cIk be the number of permitted k-tuples
within I[m, n]k . We denote by η0 , η1 , η2 , . . . , ηpk+1 −1 the number of permitted k-tuples within I[m, n]k whose indices
belong to the residue classes [0], [1], [2], . . . , [pk+1 − 1] respectively. Therefore, cIk = η0 + η1 + η2 + · · · + ηpk+1 −1 .
We wish to compute the average fraction of the permitted k-tuples within the interval I[m, n]k that are converted
to prohibited (k + 1)-tuples within the interval I[m, n]k+1 , for all the νk+1 combinations of selected remainders in the
sequence sk+1 (k ≥ 1). Now,

A B (pk+1 − 1)
νk+1 = νk+1 νk+1 = pk+1 ,
2
A
by Remark 4.1. Consequently, we begin by taking the average over the νk+1 combinations obtained by Type A
B
operations, and then we take the average of the previous averages over the νk+1 combinations obtained by Type B
operations.

Step 1. Suppose that we choose two selected remainders r, r0 within the period of the sequence sk+1 . By Proposi-
tion 2.2, the indices of the permitted k-tuples within the interval I[m, n]k of Sk that are converted to prohibited
(k + 1)-tuples within the interval I[m, n]k+1 of Sk+1 belong to one of the residue classes [r] or [r0 ]. It follows
that the fraction of the cIk permitted k-tuples within the interval I[m, n]k of Sk that are converted to prohibited
(k + 1)-tuples within the interval I[m, n]k+1 of Sk+1 is equal to (ηr + ηr0 )/cIk . Taking the average over the
A
νk+1 combinations of selected remainders obtained by repeated Type A operations, we obtain

A
νk+1 pk+1 pk+1 −1 pk+1 −1
!
X ηr + ηr0 X ηr + ηr 0
X X
1/cIk ηr + ηr0
cIk cIk



i=1 i=1 r=0 r 0 =0 1/cIk cIk + cIk 2
A
= = = = .
νk+1 pk+1 pk+1 pk+1 pk+1

B
Step 2. Now, if we take the average over the νk+1 = (pk+1 − 1)/2 combinations of selected remainders obtained by
repeated Type B operations from each one of the combinations obtained before, we obtain

B
νk+1
X 2
j=1
pk+1 2
f k+1 = B
= ,
νk+1 pk+1

B
because pk+1 does not depend on the index j (1 ≤ j ≤ νk+1 ).
For the partial sum S1 , there are two residue classes modulo p1 = 2 and one selected remainder. Therefore, it
is easy to see that f 1 = 1/p1 .

Definition 4.6. It follows from Proposition 2.2 that when we juxtapose the remainders of the sequence sk+1 to the
right of each k-tuple of Sk , the permitted k-tuples of Sk whose indices are not congruent to any of the two selected
0
remainders of sk+1 modulo pk+1 are, as (k + 1)-tuples of Sk+1 , still permitted. We denote by fk+1 the fraction
of permitted k-tuples within the interval I[m, n]k of Sk that are transferred to the interval I[m, n]k+1 of Sk+1 as
permitted (k + 1)-tuples. For the partial sum S1 , let f10 denote the fraction of the permitted 1-tuples within the
interval I[m, n]k=1 .
0 0
We denote by f k+1 the average of fk+1 for all the combinations of selected remainders in the sequence sk+1 . For
0
the partial sum S1 , let f 1 denote the average of f10 for the 2 combinations of selected remainders in the sequence s1 .
0
Now, using the preceding lemma, we can calculate the average fraction f k+1 .
0 0
Lemma 4.2. We have f k+1 = (pk+1 − 2)/pk+1 . For S1 , we have f 1 = (p1 − 1)/p1 .

Proof. By Proposition 2.2, a given permitted k-tuple within the interval I[m, n]k of Sk can be transferred to the interval
0
I[m, n]k+1 of Sk+1 either as a permitted (k + 1)-tuple or as a prohibited (k + 1)-tuple. Consequently, fk+1 + fk+1 = 1,
0 0
and so f k+1 + f k+1 = 1. Therefore, using Lemma 4.1, we obtain f k+1 = 1 − f k+1 = 1 − 2/pk+1 = (pk+1 − 2)/pk+1 .
0
For the partial sum S1 , we have f 1 = 1/p1 =⇒ f 1 = (p1 − 1)/p1 .

21
 Definition 4.7. Let Sk be the partial sum of level k. Let I[m, n] be an interval of k-tuples of Sk . We denote
I
by cIk the average number of permitted k-tuples within the interval I[m, n]. We denote by δ k the average density of
permitted k-tuples within the interval I[m, n].
Finally, using the preceding lemmas, we calculate the average k-density within a given interval I[m, n], and show
that it is equal to the k-density within the period of Sk .
I
Theorem 4.3. Let δk be the density of permitted k-tuples within a period of the partial sum Sk . Then δ k = δk .
Proof. Let sh (1 ≤ h ≤ k) be the sequences of remainders that form Sk . If there were no selected remainders within
the sequences sh , all the k-tuples within the interval I[m, n] would be permitted k-tuples, and then cIk = |I[m, n]|,
where |I[m, n]| is the size of the interval I[m, n]. However, since we have selected remainders in every sequence
sh (1 ≤ h ≤ k), using Lemma 4.2 at each level transition from h = 1 to h = k, we can write

     
p1 − 1 p2 − 2 p3 − 2 pk − 2
cIk = |I [m, n]| ··· . (12)
p1 p2 p3 pk

Now, the number of intervals of size pk within the interval I[m, n] is equal to |I[m, n]|/pk . Consequently, by
definition,

I cIk pk
δk = = cI . (13)
|I[m,n]| |I [m, n]| k
pk

Therefore, substituting (12) for cIk in (13), and using Lemma 3.1, we obtain

      
I pk p1 − 1 p2 − 2 p3 − 2 pk − 2
δk = |I [m, n]| ··· =
|I [m, n]| p1 p2 p3 pk
   
p1 − 1 p2 − 2 p3 − 2
= · · · (pk − 2) = δk .
p1 p2 p3

5 The density of permitted k-tuples within the interval I[1, n] as n → ∞

P
Let Sk (k > 2) be a partial sum of the series sk . Let pk be its characteristic prime modulus, and let mk be its
period. Let δk be the density of permitted k-tuples within the period of Sk . Let I[1, n] (n ≥ mk ) be a given interval of
k-tuples of the partial sum Sk . We denote by cIk the number of permitted k-tuples, and by δkI the k-density in I[1, n].
In this section we shall show that δkI converges to δk as n → ∞. First, we make a definition.
Definition 5.1. Let bn/mk c denote the integer part of n/mk (n ≥ mk ). We denote by cη the number of permitted
k-tuples within the interval I[1, bn/mk cmk ] ⊆ I[1, n]. If n is not a multiple of mk , we denote by c the number
of permitted k-tuples within the interval I[bn/mk cmk + 1, n] ⊂ I[1, n]; otherwise c = 0. We call the interval
I[bn/mk cmk + 1, n] the incomplete period of the interval I[1, n].
The following lemma gives us a formula for the k-density in the interval I[1, n].
Lemma 5.1. We have

j k
n
mk mk pk c
δkI = δk + .
n n
Proof. By definition,

cIk
δkI = n .
pk

Since bn/mk c is the number of times that the period of Sk fits in the interval I[1, n], the interval I[1, bn/mk cmk ]
is that part of I[1, n] whose size is a multiple of mk , and so bn/mk cmk /pk is the number of subintervals of size pk
within this part of the interval I[1, n]. Consequently, multiplying by the k-density in the period of Sk , we obtain

22
 j k
n
mk mk
cη = δk .
pk
Since cIk is the number of permitted k-tuples within I[1, n], we have cIk = cη + c . Then

j k
n j k
mk mk n
cIk cη + c δk + c mk mk pk c
pk
δkI = n = n = n = δk + .
pk pk pk n n

Remark 5.1. Let ck be the number of permitted k-tuples within the period of Sk . By definition,

ck
δk = =⇒ ck = δk mk /pk .
mk /pk
Now, using the formula from the preceding lemma, we find lower and upper bounds for the k-density within the
interval I[1, n].
Lemma 5.2. Let I[1, n] (n ≥ mk ) be an interval of k-tuples of a given partial sum Sk . For k > 2,

j k j k 
n n
mk mk mk + 1 mk
j k δk < δkI < j k δk . (14)
n n
mk mk + (mk − 1) mk mk + 1

Proof. Step 1. We first consider the case where n is not a multiple of mk . By Lemma 5.1,

j k
n
mk mk p k c
δkI = δk + . (15)
n n

To obtain bounds for δkI , we proceed as follows. We begin by obtaining bounds for c . By Remark 5.1, δk mk /pk
is the number of permitted k-tuples within the period of Sk . Since by assumption n is not a multiple of mk ,
it is easy to see that

0 ≤ c ≤ δk mk /pk . (16)

Next, we obtain bounds for the denominator in (15); since n is not a multiple of mk ,

bn/mk cmk + 1 ≤ n ≤ bn/mk cmk + (mk − 1) . (17)

Step 2. We obtain a lower bound for δkI . If we replace the denominator in (15) by the upper bound in (17),

j k
n
mk mk pk c
j k δk + j k ≤ δkI . (18)
n n
mk mk + (mk − 1) mk mk + (mk − 1)

Note that if n is equal to the upper bound in (17), the size of the incomplete period differs from the period
mk by one. On the other hand, it is easy to check, using Proposition 2.3, that within the period of the partial
sum Sk (k > 2) there is more than one permitted k-tuple. It follows that if n is equal to the upper bound
in (17), then there is at least one permitted k-tuple within the incomplete period of I[1, n], and so c > 0.
Therefore, if we replace c in (18) by the lower bound in (16),

j k
n
mk mk
j k δk < δkI . (19)
n
mk mk + (mk − 1)

23
Step 3. We now obtain an upper bound for δkI . If we replace the denominator in (15) by the lower bound in (17),

j k
n
mk mk p c
δkI ≤ j k δk + j kk  . (20)
n n
mk mk + 1 mk mk + 1

Note that if n is equal to the lower bound in (17), the size of the incomplete period is equal to 1, and so there
can not be more than one permitted k-tuple within the incomplete period of I[1, n]. On the other hand, we
saw in Step 2 that for a level k > 2, there is more than one permitted k-tuple within the period of the partial
sum Sk . It follows that if n is equal to the lower bound in (17), then c ≤ 1 < δk mk /pk . Therefore, if we
replace c in (20) by the upper bound in (16),

j k j k 
n n
mk mk pk δkpm k
mk + 1 mk
δkI < j k δk + j k k
= j k δk . (21)
n n n
mk mk + 1 mk mk + 1 mk mk + 1

Step 4. Now we complete the proof. Suppose that n is a multiple of mk . Then, the density of permitted k-tuples
within the interval I[1, n] will be equal to the density within the period of Sk ; that is, δkI = δk . Since the
lower bound in (19) is less than δk , and the upper bound in (21) is greater than δk , we conclude that for
every interval I[1, n] of the partial sum Sk (k > 2), the inequalities (19) and (21) are always satisfied, and the
lemma is proved.

Remark 5.2. It is easy to check that in (14) the upper bound is decreasing, and the lower bound is increasing, as
n → ∞.
Finally, we show that the k-density in the interval I[1, n] of a given partial sum Sk tends to δk as the size n of the
interval increases.

sk . As n → ∞, the density δkI converges to

P
Proposition 5.3. Let Sk (k > 2) be a given partial sum of the series
δk , whatever the combination of selected remainders in the sequences sh that form the partial sum Sk .
Proof. Using the inequalities of Lemma 5.2, if we take limits as n → ∞,

j k j k 
n n
mk mk mk + 1 mk
lim j k δk < lim δkI < lim j k δk .
n→∞ n n→∞ n→∞ n
mk mk + (mk − 1) mk mk + 1

Now, dividing the numerator and denominator by bn/mk c, we obtain

 
1+ j 1 k mk
n
mk mk
lim δ < lim δkI < lim
−1) k
δk .
n→∞ mk + (m
jk k n→∞ n→∞ mk + j 1 k
n n
mk mk

Since for a given level k, the values mk and δk are constants, as n → ∞ we have bn/mk c → ∞, and the lower and
upper bounds tend to δk . This implies that δkI converges to δk as n → ∞.

6 The k-density within the intervals I[1, p2k ] and I[p2k + 1, mk ]

P
Let Sk (k ≥ 4) be a given partial sum of the series sk . In this section, we shall subdivide the interval I[1, mk ] of Sk
into two parts, and shall establish the relationship between the density of permitted k-tuples within one part and the
density of permitted k-tuples within the other part. We begin by introducing some terminology and notation.
P
Definition 6.1. Let Sk and Sk+1 be consecutive partial sums of the series sk . We use the notation pk → pk+1
or alternatively k → k + 1 to denote the transition from level k to level k + 1. For the level transition pk → pk+1 , we
call the difference pk+1 − pk the order of the transition.

24
 Definition 6.2. When we juxtapose the remainders of the sequence sk+1 to the right of each k-tuple of Sk , by
Proposition 2.2, a given permitted k-tuple of Sk , whose index is congruent to a selected remainder of sk+1 modulo
pk+1 , is converted to a prohibited (k + 1)-tuple of Sk+1 . In that case, we say that at the level transition k → k + 1
one permitted k-tuple is removed.
Let sh (1 ≤ h ≤ k) be the periodic sequences of remainders that form the partial sum Sk . Let mh be the period of
every partial sum Sh from level h = 1 to level h = k. Let ch be the number of permitted h-tuples, and let δh be the
h-density within the period of every partial sum Sh (1 ≤ h ≤ k).
Definition 6.3. If we write the index n of the sequences sh from top to bottom, and the level h from left to right
(see Table 2) we say that the partial sum Sk is in vertical position. Now, suppose that the partial sum Sk is in vertical
position, and we rotate it 90 degrees counterclockwise. Then, the index n of the sequences sh increases from left to
right, and the level h increases from the bottom up. In this case, we say that the partial sum Sk is in horizontal
position.
For every partial sum Sh from level h = 1 to level h = k in horizontal position, let us consider the interval I[1, mk ]h ,
whose size is the period mk of Sk .
Remark 6.1. Using Proposition 2.1, it is easy to check that the period of the partial sum S1 is equal to m1 = p1 = 2.
On the other hand, by Proposition 2.3, within every period of S1 we have only one permitted 1-tuple. Therefore, the
interval I[1, mk ]1 of the partial sum S1 is divided into subintervals of size m1 = 2, each one containing one permitted
1-tuple. The position of the permitted 1-tuple is the same within every subinterval, and is determined by the selected
remainder in the sequence s1 . Note that the positions of consecutive permitted 1-tuples in the partial sum S1 differ
by two.
Remark 6.2. By the preceding remark, the positions of the permitted 1-tuples show a regular pattern along the
interval I[1, mk ]1 of the partial sum S1 . However, when we add the sequences sh from level h = 2 to level h = k, the
selected remainders in each sequence sh remove permitted (h − 1)-tuples from the partial sum Sh−1 . Consequently,
we obtain an interval I[1, mk ]k where the permitted k-tuples are spread along the interval, in positions that show an
irregular pattern. Note that if we change the combination of selected remainders in the sequences sh (1 ≤ h ≤ k),
within the interval I[1, mk ]k some permitted k-tuples ‘disappear’, and other permitted k-tuples ‘appear’, although the
number of permitted k-tuples within the interval I[1, mk ]k of the partial sum Sk does not change (see Proposition 2.3).
The following lemma gives us the number of permitted h-tuples within the interval I[1, mk ]h of every partial sum
Sh where h < k.
P
Lemma 6.1. Let Sk (k ≥ 4) be a given partial sum of the series sk . Let us consider the interval I[1, mk ]h in
every partial sum Sh , from level h = 1 to level h = k.
For any given partial sum Sh (h < k), the number of permitted h-tuples within the interval I[1, mk ]h is equal to
ch ph+1 ph+2 · · · pk .
Proof. Choose a level h < k. By definition, we have mk = p1 p2 p3 · · · ph ph+1 ph+2 · · · pk = mh ph+1 ph+2 · · · pk . That is,
the size of the interval I[1, mk ]h of the partial sum Sh is equal to ph+1 ph+2 · · · pk times the period mh of the partial
sum Sh . Consequently, it is easy to see that the number of permitted h-tuples within the interval I[1, mk ]h is equal
to ch ph+1 ph+2 · · · pk .

Remark 6.3. By Proposition 2.3 and Lemma 6.1, the number of permitted h-tuples within the interval I[1, mk ]h (1 ≤
h ≤ k) does not depend on the combination of selected remainders in the sequences sh that form the partial sum Sh ;
therefore, neither does the density of permitted h-tuples within this interval. Furthermore, since the size of I[1, mk ]h is
multiple of mh , the density of permitted h-tuples within the interval I[1, mk ]h is equal to δh (the density of permitted
h-tuples within the period mh of the partial sum Sh ).
Now, let us denote by c0h the number of permitted h-tuples within the interval I[1, mk ]h of every partial sum
Sh (1 ≤ h ≤ k), which is computed using Proposition 2.3 and Lemma 6.1. We have a question at this point: What is
the behaviour of c0h as h goes from level 1 to level k? This behaviour can be described as follows.
Remark 6.4. For every partial sum Sh (h < k), suppose that we juxtapose the remainders of the sequence sh+1 to
each h-tuple of Sh . By Proposition 2.2, the permitted h-tuples within the interval I[1, mk ]h whose indices are included
in two residue classes modulo ph+1 are removed by the selected remainders within the sequence sh+1 ; and the permitted
h-tuples whose indices are not included in these residue classes are transferred to level h + 1 as permitted (h + 1)-
tuples within the interval I[1, mk ]h+1 of the partial sum Sh+1 , whatever the combination of selected remainders in the
sequence sh+1 . Since for every level h < k the size of the interval I[1, mk ]h is a multiple of ph+1 , by Proposition 2.7
and Corollary 2.8, the permitted h-tuples within the interval I[1, mk ]h of Sh are distributed uniformly over the residue
classes modulo ph+1 . Therefore, for each level h < k, a fraction 2/ph+1 of the permitted h-tuples within the interval
I[1, mk ]h of Sh have been removed, and a fraction (ph+1 − 2)/ph+1 have been transferred to level h + 1 as permitted
(h + 1)-tuples within the interval I[1, mk ]h+1 of Sh+1 , whatever the combination of selected remainders in the sequence
sh+1 .

25
 The next lemma follows at once from the preceding remark.
Lemma 6.2. Let h and h + 1 be consecutive levels, where 1 ≤ h < k. We have

 
ph+1 − 2
c0h+1 = c0h .
ph+1
Let us examine now the behaviour of δh as h goes from level 1 to level k. Since the selected remainders of the
sequences sh+1 remove permitted h-tuples within the interval I[1, mk ]h of the partial sum Sh , at each level transition
h → h + 1, the number of permitted h-tuples decreases as the level increases from h = 1 to h = k. However,
by Lemma 3.2 and Corollary 3.3, the h-density within the interval I[1, mk ]h of the partial sum Sh grows at each
transition ph → ph+1 of order greater than 2, because to compute the h-density we count the permitted h-tuples
within subintervals of size ph , which grow by more than 2, overcompensating for the permitted h-tuples removed.
If ph → ph+1 is a level transition of order 2, the h-density does not change, because the increase in the size ph
is compensated for by the permitted h-tuples removed. (Note that p1 → p2 is the only level transition where δh
decreases.) Therefore, the h-density increases between h = 1 and h = k if we choose k so large that there are a
sufficient number of level transitions of order greater than 2 between h = 1 and h = k.
We have the following lemma:
Lemma 6.3. Let h and h + 1 be consecutive levels, where 1 ≤ h < k. Then

 
ph+1 − 2
δh+1 = δh . (22)
ph

Proof. By Lemma 6.2, the number of permitted (h+1)-tuples in the interval I[1, mk ]h+1 is given by c0h (ph+1 −2)/ph+1 .
On the other hand, in the interval I[1, mk ]h+1 there is mk /ph+1 subintervals of size ph+1 . Then, by definition

c0h (ph+1 − 2) /ph+1

δh+1 = .
mk /ph+1
Multiplying numerator and denominator by ph , using Lemma 6.1, and simplifying, we obtain

ph (c0h (ph+1 − 2) /ph+1 ) c0h

   
ph+1 − 2 ch ph+1 − 2
δh+1 = = = .
ph (mk /ph+1 ) mk /ph ph mh /ph ph
So, by definition,

 
ph+1 − 2
δh+1 = δh .
ph

Now, the proof of Lemma 3.2 is immediate.

Proof of Lemma 3.2. It follows at once from Remark 6.3 and the preceding lemma.

Now, if we ‘cut’ the first period of Sk into two parts, between the indices p2k and p2k + 1, we obtain a left-hand
subinterval and a right-hand subinterval.
Definition 6.4. Let Sk (k ≥ 4) be a given partial sum, in horizontal position. We subdivide the interval I[1, mk ]
(its first period) into two intervals: I[1, p2k ], which we call the Left interval, and I[p2k + 1, mk ], which we call the Right
interval. We often denote the Left interval I[1, p2k ] by the symbol Lk , and the Right interval I[p2k + 1, mk ] by the
symbol Rk . For every partial sum Sh from level h = 1 to level h = k − 1 there is also a Left interval I[1, p2k ]h , and a
Right interval I[p2k + 1, mk ]h . See Figure 4.
As we have seen in the Introduction, the first period of the sequence of k-tuples can be seen as a matrix, with mk
columns and k rows. Each row of this matrix (from h = 1 to h = k), is formed by the remainders of dividing the
integers from n = 1 to n = mk by the modulus ph . In addition, this matrix has been partitioned into two blocks: the
Left block formed by the columns from n = 1 to n = p2k ; and the Right block formed by the columns from n = p2k + 1
to n = mk . Each row of the Left block is formed by the remainders of dividing the integers from n = 1 to n = p2k by
the modulus ph ; and each row of the Right block is formed by the remainders of dividing the integers from n = p2k + 1
to n = mk by the modulus ph .

26
    
Level h = k I 1, p2k h=k ∪ I p2k + 1, mk h=k = I [1, mk ]h=k

Level h = h0
   
I 1, p2k h=h0 ∪ I p2k + 1, mk h=h0 = I [1, mk ]h=h0

   
Level h = 1 I 1, p2k h=1 ∪ I p2k + 1, mk h=1 = I [1, mk ]h=1

Figure 4: Left and Right intervals

Definition 6.5. For a given partial sum Sh (1 ≤ h ≤ k), we use the notation cL h to denote the number of permitted
k

2 Rk
h-tuples within the Left interval I[1, pk ]h , and we use the notation ch to denote the number of permitted h-tuples
within the Right interval I[p2k + 1, mk ]h .
Although the number of permitted h-tuples within the interval I[1, mk ]h of every partial sum Sh (1 ≤ h ≤ k) does
not change if we choose another set of selected remainders, the positions of the permitted h-tuples along the period
of Sh will be changed. Then, it seems reasonable to expect that some permitted h-tuples will be transferred from the
Left interval I[1, p2k ]h to the Right interval I[p2k + 1, mk ]h , or vice versa, because the size of the Left interval and the
size of the Right interval are not a multiple of mh . Hence, the number of permitted h-tuples within the Left interval
I[1, p2k ]h and within the Right interval I[p2k + 1, mk ]h is determined by the combination of selected remainders in the
sequences sh that form the partial sum Sh .
Definition 6.6. For a given partial sum Sh (that is to say, a partial sum where we have a given combination of
selected remainders in the sequences that form the partial sum Sh ), we use the notation δhLk to denote the density of
permitted h-tuples within the Left interval I[1, p2k ]h , and we use the notation δhRk to denote the density of permitted
h-tuples within the Right interval I[p2k + 1, mk ]h .
By Remark 6.3, the h-density within the interval I[1, mk ]h does not depend on the combination of selected remain-
ders in the sequences sh that form the partial sum Sh . However, the transfer of some permitted h-tuples from the Left
interval I[1, p2k ]h to the Right interval I[p2k + 1, mk ]h , or in the opposite direction, when we change the combination
of selected remainders, brings about changes in the h-density within both intervals. The crossing of some permitted
h-tuples from I[1, p2k ]h to I[p2k + 1, mk ]h decreases δhLk and increases the δhRk , and vice versa. By Theorem 4.3, the
average of δhLk within I[1, p2k ]h is equal to δh , and the average of δhRk within I[p2k + 1, mk ]h is also equal to δh . Hence

δhLk > δh ⇐⇒ δhRk < δh , (23)

δhLk < δh ⇐⇒ δhRk > δh .

Definition 6.7. We often call δhLk (δhRk ) the true h-density to distinguish it from the average δh within the Left
interval I[1, p2k ]h (the Right interval I[p2k + 1, mk ]h ).
The following lemma shows that for every partial sum Sh (1 ≤ h ≤ k), the h-density within the Left interval
I[1, p2k ]h can not be equal to the h-density within the Right interval I[p2k + 1, mk ]h .
P
Lemma 6.4. Let Sk (k ≥ 4) be a given partial sum of the series sk . Let us consider the interval I[1, mk ]h (whose
size is the period mk of Sk ), the Left interval I[1, p2k ]h and the Right interval I[p2k + 1, mk ]h , in every partial sum
Sh from level h = 1 to level h = k. Let us denote by mh the period of the partial sum Sh , and by ch the number of
permitted h-tuples within a period of the partial sum Sh (1 ≤ h ≤ k).
For every partial sum Sh we have δhLk 6= δhRk .

Proof. Step 1. By Remark 6.1, the positions of consecutive permitted 1-tuples in the partial sum S1 differ by 2. It
follows that the number of permitted h-tuples in every Left interval I[1, p2k ]h (1 ≤ h ≤ k) is less than the size
of the interval. In symbols, cL 2
h < pk (1 ≤ h ≤ k).
k

Step 2. Let us consider a given partial sum Sh , where 1 ≤ h < k. Consider the number of permitted h-tuples in the Left
interval I[1, p2k ]h , denoted by cL 2
h , and the number of permitted h-tuples in the Right interval I[pk + 1, mk ]h ,
k

denoted by ch . By Lemma 6.1, we have ch + ch = ch ph+1 ph+2 · · · pk ; so, if ch is a multiple of pk , then cR

Rk Lk Rk Lk
h
k

Lk Lk
is a multiple of pk as well. In this case, (ch /pk ) is a whole number; and so (ch /pk )/pk , is a reduced fraction,

27
 since cL 2 Rk
h < pk , by Step 1. Hence, it is easy to check that this fraction is not equal to (ch /pk )/(mk−1 − pk ),
k

since (cR
h /pk ) is also a whole number, and mk−1 − pk is not a multiple of pk .
k

On the other hand, if cL Lk 2

h is not a multiple of pk , then ch /pk is a reduced fraction; and it is easy to check that
k

Rk
this fraction is not equal to ch /(mk − p2k ), since mk − p2k is not a multiple of p2k . In either case, we proved
that cL 2 Rk 2
h /pk is not equal to ch /(mk − pk ).
k

Step 3. Now, let us consider the partial sum Sk . Consider the number of permitted k-tuples in the Left interval
I[1, p2k ], denoted by cL
k , and the number of permitted k-tuples in the interval I[1, mk ] (complete period of
k

Sk ), denoted by ck . By Proposition 2.3 we have ck = (p1 − 1)(p2 − 2)(p3 − 2) · · · (pk − 2). Now, if cL
k is not a
k

Lk
multiple of pk , we can see that ck /p2k is a reduced fraction; and it is easy to check that this fraction can not
be equal to ck /mk , since mk is a squarefree integer.
On the other hand, if cL Lk Lk
k is a multiple of pk , then (ck /pk ) is a whole number; and (ck /pk )/pk , is a reduced
k

fraction, since cL 2
k < pk , by Step 1. From Proposition 2.1 and Proposition 2.3 follows
k

ck (p1 − 1) (p2 − 2) (p3 − 2) · · · (pk−1 − 2) (pk − 2)

= =
mk p1 p2 p3 · · · pk−1 pk
      
p1 − 1 p2 − 2 p3 − 2 pk−1 − 2 pk − 2
= ··· ,
p1 p2 p3 pk−1 pk

and shifting the denominators to the right, we obtain

     
ck p2 − 2 p3 − 2 pk−1 − 2 pk − 2 1
= ··· .
mk p1 p2 pk−2 pk−1 pk

Note that p1 can not be canceled with any numerator of the fractions in parentheses, since all these are odd
integers. Thus, it is easy to check that this fraction is not equal to the reduced fraction (cL
k /pk )/pk .
k

In either case, the proportion of permitted k-tuples in the interval I[1, p2k ], given by cL 2
k /pk , is not equal to
k

Lk
the proportion of permitted k-tuples in the interval I[1, mk ], given by ck /mk . Thus, if ck /p2k > ck /mk , it
must be cR 2 Lk 2 Rk 2
k /(mk − pk ) < ck /mk ; and vice versa; this implies ck /pk 6= ck /(mk − pk ).
k

Step 4. We prove the lemma. From Steps 2 and 3, for every partial sum Sh (1 ≤ h ≤ k) it follows that cL 2
h /pk 6=
k

Rk L R L R
ch /(mk − p2k ); multiplying by ph we obtain ph ch k /p2k 6= ph ch k /(mk − p2k ); and so δh k 6= δh k .

Even though the increase of the number of permitted h-tuples within one interval is equal to the decrease of the
number of permitted h-tuples within the other interval, the increase of the h-density within one interval is not equal
to the decrease of the h-density within the other interval. This is due to their being more subintervals of size ph within
I[p2k + 1, mk ]h than within I[1, p2k ]h , for k > 3. The following lemma gives the relationship between the h-density
within I[1, p2k ]h and the h-density within I[p2k + 1, mk ]h . First, a definition:
Definition 6.8. Let Sh be the partial sums from level h = 1 to level h = k (k ≥ 4). Let I[1, p2k ]h be the Left
interval, and let I[p2k + 1, mk ]h be the Right interval, in every partial sum Sh (1 ≤ h ≤ k). For a given partial sum
Sh (1 ≤ h ≤ k), let δhLk be the density of permitted h-tuples within the Left interval I[1, p2k ]h , and let δhRk be the
density of permitted h-tuples within the Right interval I[p2k + 1, mk ]h . We use the notation {δhLk } to denote the set
of values of δhLk , and we use the notation {δhRk } to denote the set of values of δhRk , for all the combinations of selected
remainders in the sequences that form the partial sum Sh .
Lemma 6.5. There is a bijective function fh : {δhLk } → {δhRk } such that

p2k
fh (x) = δh − (x − δh ) ,
mk − p2k

and

mk − p2k
fh−1 (x) = δh + (δh − x) .
p2k

28
Proof. For a given level h (1 ≤ h ≤ k), if we change the combination of selected remainders in the partial sum Sh ,
some permitted h-tuples will be transferred from the Left interval I[1, p2k ]h to the Right interval I[p2k + 1, mk ]h , or vice
versa, as we have seen before. So, it is easy to see that there is a set of values of the number of permitted h-tuples
within the Left interval, and a set of values of the number of permitted h-tuples within the Right interval. However,
there exists a one-to-one correspondence between both sets, since the number of permitted h-tuples within the interval
I[1, mk ]h is the same, whatever the combination of selected remainders in the sequences sh that form the partial sum
Sh , by Proposition 2.3 and Lemma 6.1. It follows that there is also a one-to-one correspondence between the set of
values of δhLk , and the set of values of δhRk . So, for a given level h (1 ≤ h ≤ k), we can define a bijective function
fh : {δhLk } → {δhRk }.
Now, for the partial sum Sh , assume that the density of permitted h-tuples within I[1, p2k ]h , and within I[p2k +
1, mk ]h , is equal to the average δh . Then, suppose that some permitted h-tuples are transferred from the Right interval
to the Left interval. We have an increase (δhLk −δh ) of the h-density within the Left interval, and a decrease (δh −δhRk ) of
the h-density within the Right interval. See (23). Because within the Left interval I[1, p2k ]h we have p2k /ph subintervals
of size ph , by definition, the number of permitted h-tuples entering the Left interval is equal to (δhLk − δh )p2k /ph . In
the same way, within the Right interval I[p2k + 1, mk ]h we have (mk − p2k )/ph subintervals of size ph , and then, by
definition, the number of permitted h-tuples coming out of the Right interval is equal to (δh − δhRk )(mk − p2k )/ph .
Since the number of permitted h-tuples entering the Left interval must be equal to the number of permitted h-tuples
coming out of the Right interval,

  p2   m − p2   p2k
k
δhLk − δh k
= δh − δhRk k
=⇒ δhRk = δh − δhLk − δh .
ph ph mk − p2k

Therefore, we have a bijective function fh : {δhLk } → {δhRk }, such that fh (x) = δh − (x − δh )(p2k /(mk − p2k )), and it
is easy to check that fh−1 (x) = δh + (δh − x)(mk − p2k )/p2k .

7 The sifting function of the Sieve II

7.1 The behaviour of the h-density within the Right interval as k → ∞
In this section we shall establish a lower bound for the minimum value of the sifting function of the Sieve II, for k
sufficiently large. However, before achieving this result, we need to establish a lower bound for the k-density within
the interval I[1, p2k ] of the partial sum Sk , where k is large enough. Now, for reasons that will be clear later, we begin
by studying the behaviour of the h-density (1 ≤ h ≤ k) within the Right block of the partition of the first period of
Sk ; the following example illustrates this behaviour for a small level k.
P
Example 7.1. Let Sk be a partial sum of the series sk . Suppose that we take first k = 4, and then we let
k → ∞. We can see that, as the level k increases, for every partial sum Sh from h = 1 to h = k, the size of the
Right interval I[p2k + 1, mk ]h grows very fast, since p2k = o(mk ), by Lemma 2.5. Note that for h = 1 there is one
permitted 1-tuple within every period of size p1 = 2 of the partial sum S1 , by Remark 6.1. So, as the size of the
Right interval I[p2k + 1, mk ]1 increases, the number of permitted 1-tuples grows very fast as well, and the density of
permitted 1-tuples becomes more and more closely to the average δ1 . Therefore, we can reach a level k not too large
(for example k = 8 (pk = 19)) such that the distribution of the permitted 1-tuples that are within I[p2k + 1, mk ]1 over
the residue classes modulo p2 = 3 is not far from uniform. So, the fraction of permitted 1-tuples within the Right
interval of S1 that are transferred to the Right interval of S2 as permitted 2-tuples is approximately (p2 − 2)/p2 (see
Remark 6.4).
However, as h goes from level 1 to level k (where k is this level not too large), the number of permitted h-tuples that
are within the Right interval decreases (see Remark 6.2), and ph increases; in addition, the number of combinations
of selected remainders within the sequences that form every partial sum Sh increases as well. See (10) and (11).
Therefore, as h goes from level 1 to level k − 1, for some combinations of selected remainders, the distribution of the
permitted h-tuples that are within I[p2k + 1, mk ]h over the residue classes modulo ph+1 becomes far from uniform.
So, for these combinations of selected remainders, the fraction of permitted h-tuples within the Right interval of Sh
that are transferred to the Right interval of Sh+1 as permitted (h + 1)-tuples moves away from (ph+1 − 2)/ph+1 (see
Remark 6.4), and consequently, the values of δhRk moves away from δh , as h goes from level 1 to level k (see Lemma 6.3).
Now, we consider in the following two remarks the case where k is larger than in the previous case.
Remark 7.1. For every partial sum Sh from level h = 1 to level h = k (k ≥ 4), let us consider the interval I[1, mk ]h ,
the Left interval I[1, p2k ]h and the Right interval I[p2k + 1, mk ]h . We denote by ck the number of permitted k-tuples
within the interval I[1, mk ]k , and we denote by c0h the number of permitted h-tuples within every interval I[1, mk ]h
where 1 ≤ h < k. Since p2k = o(mk ), by Lemma 2.5, and p2k = o(ck ), by Lemma 2.4, we can take k large enough that
the size of the Left interval I[1, p2k ] is negligible compared to the size of the interval I[1, mk ], and furthermore the size

29
of I[1, p2k ] is negligible compared to ck . Therefore, since c0h > ck (see Proposition 2.3 and Lemma 6.1), almost all the
permitted h-tuples in every interval I[1, mk ]h (1 ≤ h ≤ k) belong to the Right interval I[p2k + 1, mk ]h .
Remark 7.2. Assume that k satisfies the conditions required in the preceding remark. Then, as we have remarked,
almost all the permitted h-tuples in every interval I[1, mk ]h (1 ≤ h ≤ k) belong to the Right interval I[p2k + 1, mk ]h ;
furthermore, from the second condition it follows that ck  p2k > pk . Note that in this case, c0h  ph for every h < k,
since ph < pk and c0h > ck . Hence, it is easy to see that for each level from h = 1 to h = k − 1, the distribution of
the permitted h-tuples that are within the Right interval over the residue classes modulo ph+1 will be almost uniform,
whatever the combination of selected remainders in the sequences sh that form the partial sum Sk ; so, the fraction of
the permitted h-tuples within the Right interval of Sh that are transferred to the Right interval of Sh+1 as permitted
(h + 1)-tuples will be very close to the average fraction (ph+1 − 2)/ph+1 (see Remark 6.4 and Lemma 6.2). Therefore,
for every level transition h → h + 1 (1 ≤ h < k), the relationship between δhRk and δh+1 Rk
will be very close to the
relationship between δh and δh+1 given by Lemma 6.3 (see the proof of this lemma).
By the preceding remark and what we have seen in this remark we conclude that, if k satisfies the conditions
required in Remark 7.1, for each level from h = 1 to h = k the values of δhRk will be very close to δh , whatever the
combination of selected remainders in the sequences sh that form the partial sum Sk .
The following lemma shows that as k → ∞, the true h-density within the Right interval I[p2k + 1, mk ]h of every
partial sum Sh (1 ≤ h ≤ k) converges uniformly to the average δh .
sk . Let us consider the Right interval I[p2k + 1, mk ]h
P
Lemma 7.1. Let Sk (k ≥ 4) be a partial sum of the series
in every partial sum Sh , from level h = 1 to level h = k. For every  > 0, there exists N (depending only on ) such
that k > N implies |δhRk − δh | < , for every partial sum Sh from level h = 1 to level h = k, whatever the combination
of selected remainders in the sequences sh that form every partial sum Sk .

Proof. Step 1. The size of the Right interval I[p2k + 1, mk ]h of the partial sum Sh , by definition, is equal to mk − p2k ,
and so the number of subintervals of size ph within the Right interval is equal to (mk − p2k )/ph (1 ≤ h ≤ k).
Denoting by cR 2
h the number of permitted h-tuples within I[pk + 1, mk ]h , by definition, we have
k

cR k

δhRk = h
(1 ≤ h ≤ k). (24)
(mk − p2k ) /ph

Step 2. Let us denote by mh the period of the partial sum Sh , and by ch the number of permitted h-tuples within a
period of the partial sum Sh . For every level from h = 1 to h = k, let c0h be the number of permitted h-tuples
within the interval I[1, mk ]h of the partial sum Sh . Using Lemma 6.1, we obtain

c01 = c1 p2 p3 · · · pk , (25)
c02 = c2 p3 p4 · · · pk ,
...
c0h = ch ph+1 ph+2 · · · pk ,
...
c0k = ck .

Note that c0h increases as the level decreases from h = k to h = 1 (see Proposition 2.3). For every level from
h = 1 to h = k, since I[1, mk ]h = I[1, p2k ]h ∪ I[p2k + 1, mk ]h , the number of permitted h-tuples within the Right
interval I[p2k + 1, mk ]h can not be greater than c0h , and so we have cR 0
h ≤ ch . On the other hand, the number
k

2
of permitted h-tuples within the Left interval I[1, pk ]h of the partial sum Sh can not be greater than the size
p2k of the Left interval. Therefore, c0h − p2k ≤ cR 0
h . Consequently, replacing the numerator in (24) by ch − pk
k 2
0
and by ch , we obtain

c0h − p2k Rk c0h

≤ δ h ≤ (1 ≤ h ≤ k).
(mk − p2k ) /ph (mk − p2k ) /ph

Extracting the common factors c0h and mk , we obtain

c0h 1 − p2k /c0h c0h

   
1
≤ δhRk ≤ .
mk /ph 1 − p2k /mk mk /ph 1 − p2k /mk

30
 Now, by definition, mk = p1 p2 p3 · · · ph ph+1 ph+2 · · · pk = mh ph+1 ph+2 · · · pk . Then, using (25), we can simplify
both sides:

1 − p2k /c0h
   
ch ch 1
≤ δhRk ≤ .
mh /ph 1 − p2k /mk mh /ph 1 − p2k /mk

By definition,

ch
δh = .
mh /ph

Therefore, for every partial sum Sh from level h = 1 to level h = k, whatever the combination of selected
remainders in the sequences sh that form the partial sum Sk , we have the bounds

1 − p2k /c0h
   
1
δh ≤ δhRk ≤ δh . (26)
1 − p2k /mk 1 − p2k /mk

Step 3. Now, let  > 0 be a given small number, and let N ≥ 12. For level h = k, from (26), we obtain

1 − p2k /c0k
   
1
δk ≤ δkRk ≤ δk .
1 − p2k /mk 1 − p2k /mk

On the one hand, p2k = o(mk ), by Lemma 2.5. On the other hand, ck = c0k by (25), and so, by Lemma 2.4,
p2k = o(c0k ). Besides, it follows from Proposition 2.3 that c0k < mk . Therefore, we can take N large enough
that for level k > N ,

1 − p2k /c0k
   
 1 
δk − < δk ≤ δkRk ≤ δk < δk + , (27)
2 1 − p2k /mk 1 − p2k /mk 2

at level h = k.
Step 4. Now, the rightmost inequality in (27) implies

 
1 
δk 2 −1 < .
1 − pk /mk 2

For a given level h < k, since k > N ≥ 12 by assumption, it is easy to verify using Lemma 3.2 and Corollary 3.3
(see Table 1), that δh < δk . Hence,

   
1  1 
δh −1 < =⇒ δh < δh + . (28)
1 − p2k /mk 2 1 − p2k /mk 2

Step 5. The leftmost inequality in (27) implies

1 − p2k /c0k p2k /c0k − p2k /mk

   

δk 1− = δk < .
1 − p2k /mk 1 − p2k /mk 2

For a given level h < k, since k > N ≥ 12, we have δh < δk (see Step 4). On the other hand, c0k = ck < c0h < mk ,
where 1 ≤ h < k (see (25) and Proposition 2.3). Hence, replacing δk by δh , and c0k by c0h , we obtain

1 − p2k /c0h 1 − p2k /c0h

   
 
δh 1 − < =⇒ δh − < δh . (29)
1 − p2k /mk 2 2 1 − p2k /mk

31
Step 6. We now prove the lemma. By (26), (27), (28), and (29), for k > N we can write

1 − p2k /c0h
   
 1 
δh − < δh ≤ δhRk ≤ δh < δh + ,
2 1 − p2k /mk 1 − p2k /mk 2

for every level from h = 1 to h = k. This implies |δhRk − δh | <  for every level from h = 1 to h = k (k > N ),
whatever the combination of selected remainders in the sequences sh that form every partial sum Sk .

7.2 A lower bound for the sifting function of the Sieve II

Now we are ready to establish a lower bound for the true density of permitted k-tuples within the Left interval I[1, p2k ]
of the partial sum Sk , denoted by δkLk , for k sufficiently large. By Theorem 4.3, the average density of permitted
k-tuples within I[1, p2k ] is equal to δk , that is to say, is equal to the k-density within the period of Sk . By Lemma 3.2
and Corollary 3.3, the density δk increases at each level transition pk → pk+1 of order greater than 2, and this implies
that for level k > 4 we have δk > δ4 . However, what happens to the true density δkLk ? The next lemma shows that
there exists Kα ∈ Z+ , Kα > 5 such that δkLk > δ4 for every level k > Kα , whatever the combination of selected
remainders in the sequences sh that form every partial sum Sk .
Recall the notation {δhLk } to denote the set of values of δhLk , and the notation {δhRk } to denote the set of values
of δhRk , for all the combinations of selected remainders in the sequences that form the partial sum Sh (1 ≤ h ≤ k).
Furthermore, recall that δh denotes the density of permitted h-tuples within the period of the partial sum Sh .
Definition 7.1. We use the notation δ̂hLk and δ̌hLk to denote, respectively, min{δhLk } and max{δhLk }, and the notation
δ̂hRk and δ̌hRk to denote, respectively, min{δhRk } and max{δhRk }.

Remark 7.3. By Lemma 6.4, for every partial sum Sh (1 ≤ h ≤ k, k ≥ 4), we have δhLk 6= δhRk . On the other hand,
for any given level h, the average density of permitted h-tuples within the Left interval I[1, p2k ]h (the Right interval
I[p2k + 1, mk ]h ) is equal to δh , by Theorem 4.3. Then, for each level between h = 1 and h = k, we have δ̂hLk < δh < δ̌hLk
(δ̂hRk < δh < δ̌hRk ). See (23).
Remark 7.4. Note that for a given level h (1 ≤ h ≤ k), the image of δ̂hLk under the function fh of Lemma 6.5 is
δ̌hRk , and the image of δ̌hLk under fh is δ̂hRk . See (23).

sk . There exists an integer Kα > 5 such that δ̂kLk > δ4

P
Lemma 7.2. Let Sk (k ≥ 5) be a partial sum of the series
for every k > Kα .
Proof. Step 1. Let L be a number such that 0 < L < (δ5 − δ4 ); let us consider the Left interval I[1, p2k ]5 of the partial
sum S5 . Since the size of the interval I[1, p2k ]5 tends to infinity as k → ∞, by Proposition 5.3 it must exist an
integer N ≥ 5 such that

δ5 − δ̂5Lk < L (30)

for every k > N .

Step 2. Let us consider the Right interval I[p2k + 1, mk ]5 of S5 ; and let

p2k
R = L . (31)
mk − p2k

By Lemma 6.5 and Remark 7.4, we can write

  p2k   p2k
δ̌5Rk = δ5 − δ̂5Lk − δ5 = δ 5 + δ 5 − δ̂ Lk
5 ,
mk − p2k mk − p2k

and in view of (30), for k > N we obtain

p2k
δ̌5Rk < δ5 + L .
mk − p2k

32
 So, by (31) we have

δ̌5Rk < δ5 + R , (32)

for k > N .
Step 3. Note that the inequality (32) is only valid for h = 5, whenever k > N . However, by Lemma 7.1 there exists an
integer N 0 > N such that δ̌hRk < δh + R if k > N 0 , for every partial sum Sh from level h = 1 to level h = k;
in particular, for h = k (k > N 0 ) we have δ̌kRk < δk + R , and from this we obtain

R > δ̌kRk − δk . (33)

On the other hand, again by Lemma 6.5 and Remark 7.4, we can write

  m − p2   m − p2
k k
δ̂kLk = δk + δk − δ̌kRk k
= δ k − δ̌ Rk
k − δ k
k
. (34)
p2k p2k

In view of (33), if we replace (δ̌kRk − δk ) by R in the last member of (34), we obtain

mk − p2k
δ̂kLk > δk − R (k > N 0 ),
p2k

and using again the relationship (31), for k > N 0 we have

δ̂kLk > δk − L .

Step 4. We prove the lemma. Since N 0 > N ≥ 5, it is easy to check using Lemma 3.2 that δk > δ5 if k > N 0 . So,
taken Kα = N 0 , it follows that δ̂kLk > δk − L > δ5 − L for every k > Kα , by Step 3. Since L < (δ5 − δ4 ) by
hypothesis, for every k > Kα we obtain δ̂kLk > δ5 − (δ5 − δ4 ) = δ4 .

Definition 7.2. Let Sk be the partial sum associated to the Sieve II; recall that in Section 2 we have taken
B = {n : 1 ≤ n ≤ p2k }; let T (B, P, pk ) be the sifting function of the Sieve II. We denote by {T (B, P, pk )} the set of
the values of T (B, P, pk ) for all the combinations of selected remainders in the sequences that form the partial sum
Sk .

Now, we can obtain a lower bound for the sifting function of the Sieve II (that is, a lower bound for the number
of permitted k-tuples within the Left interval I[1, p2k ] of Sk ), for k sufficiently large.
Lemma 7.3. Let Kα be the number whose existence is established in Lemma 7.2. For level k > Kα , we have
min{T (B, P, pk )} > pk /2.
sk . Recall the notation {δkLk } to denote the set of
P
Proof. Step 1. Consider a given partial sum Sk of the series
values of δkLk , for all the combinations of selected remainders in the sequences that form the partial sum Sk ;
and recall the notation δ̂kLk to denote min{δkLk }. Note that within the Left interval I[1, p2k ] of Sk we have pk
subintervals of size pk . So, the minimum number of permitted k-tuples within the Left interval I[1, p2k ] of Sk
is pk δ̂kLk . Then, by definition, min{T (B, P, pk )} = pk δ̂kLk .

Step 2. Now, by Lemma 7.2, if k > Kα then δ̂kLk > δ4 . From this and Step 1 it follows that min{T (B, P, pk )} >
pk δ4 , whenever k > Kα . Using Lemma 3.1, it is easy to check that δ4 = 1/2 (see Table 1), and so
min{T (B, P, pk )} > pk /2 if k > Kα .

33
8 Proof of the Main Theorem
In this section we prove the Main Theorem. We begin by defining the sequence of k-tuples of the Sieve associated
with x (the Sieve I), where x > 49 is an even number.
√
Definition 8.1. Let x > 49 be an even number, and let k be the index of the greatest prime less than x. Let
{b1 , b2 , b3 , . . . , bk } be the ordered set of the remainders of dividing x by p1 , p2 , p3 , . . . , pk . We define the sequence of
k-tuples of remainders of level k, where in the sequences of remainders modulo ph (1 ≤ h ≤ k) that form this sequence
of k-tuples are applied the following rules for selecting remainders:

Rule 1. Within every period of size ph of the sequence sh (1 ≤ h ≤ k), the remainder 0 is selected.
Rule 2. Within every period of size ph of the sequence sh (1 ≤ h ≤ k), the remainder bh is selected.

Now we can define formally the Sieve I, as follows.

√
Definition 8.2. Let P be the sequence of all primes; let z = x, and let pk be the greatest prime less than z. Let A
be the set consisting of the indices of the sequence of k-tuples of the preceding definition that lie in the interval [1, x].
For each p = ph ∈ P (1 ≤ h ≤ k), the subset Ap of A consists of the indices n of the sequence of k-tuples such that
the remainder of dividing n by the modulus ph is a selected remainder. Then, the indices of the prohibited k-tuples
lying in A are sifted out; and the indices of the permitted k-tuples lying in A remain unsifted. See Remark 1.2. The
sifting function

[
S(A , P, z) = A \ Ap ,
p∈P
p<z

is given by the number of permitted k-tuples whose indices lie in the interval A .
Remark 8.1. Every sequence sh (1 ≤ h ≤ k) that form the sequence of k-tuples associated to the Sieve I consists
of the remainders of dividing n by ph . If a remainder is equal to 0, it is always a selected remainder. If a remainder
is equal to bh , it is also a selected remainder. If x is divisible by ph , then bh = 0 and therefore, in every period ph of
sh there is only one selected remainder.
The following theorem shows that if n is the index of a permitted k-tuple belonging to the set A and 1 < n < x,
then n is a prime such that either x − n = 1 or x − n is also a prime.
Theorem 8.1. Let us consider the Sieve I, and its associated sequence of k-tuples. If n (1 < n < x) is an unsifted
element of the set A , then n is a prime such that either x − n = 1 or x − n is also a prime.
Proof. Step 1. By definition, the set A consists of the indices of the sequence of k-tuples associated to the Sieve I,
which lie in the interval [1, x]. Since n is an unsifted element of the set A , by definition, n is the index of a
permitted k-tuple. In the sequences of remainders modulo ph (1 ≤ h ≤ k) that form the sequence of k-tuples
associated to the Sieve I, if a remainder is equal to 0 then it is a selected remainder. Then, by definition,
a permitted k-tuple in this sequence has no element equal to√0 (see Remark 1.1). This means that n is not
divisible by any of the primes p1 , p2 , p3 , . . . , pk less than z = x. Since 1 < n < x, it follows at once that n is
a prime.
Step 2. Let {b1 , b2 , b3 , . . . , bk } be the ordered set of the remainders of dividing x by p1 , p2 , p3 , . . . , pk . Let rh (1 ≤
h ≤ k) be the elements of the permitted k-tuple whose index is n. In the sequences of remainders modulo
ph (1 ≤ h ≤ k) that form the sequence of k-tuples associated to the Sieve I, by definition, if a given remainder
of the sequence is equal to bh ∈ {b1 , b2 , b3 , . . . , bk }, then it is a selected remainder. Consequently, by definition,
for the permitted √ k-tuple whose index is n we have rh 6= bh (1 ≤ h ≤ k); this implies n 6≡ x (mod ph ), for
every prime ph < x (see Remark 1.1).
√
Step 3. By Step 1, n is a prime; furthermore √ n 6≡ x (mod √ ph < x, by Step 2. This last implies that x − n
√ ph ), where
is not divisible by any prime ph < x. Since x − n < x, it follows that either x − n = 1 or x − n is also a
prime.

Note that, given the level k, and given an even integer x (p2k < x < p2k+1 ), there is a sequence of k-tuples associated
to the Sieve I, which has specific selected remainders for this particular x. On the other hand, given k, there is a
partial sum Sk associated to the Sieve II, where there are multiple choices for selecting remainders, allowed by the rules
defined in Section 2. Both are sequences of k-tuples of remainders, but they differ in the rules for selecting remainders
in each one of them. The following lemma gives the relationship between the number of permitted k-tuples within

34
the interval I[1, p2k ] of the partial sum Sk (the sifting function of the Sieve II), and the number of permitted k-tuples
within the interval I[1, x] of the sequence of k-tuples associated to the Sieve I (the sifting function of the Sieve I).
Recall that we denote by {T (B, P, pk )} the set of the values of T (B, P, pk ) for all the combinations of selected
remainders in the sequences that form the partial sum Sk associated to the Sieve II.

Lemma 8.2. Let P be the√sequence of all primes. Let x > 49 be an even number, and let k be the index of the
greatest prime less than z = x; that is, p2k < x < p2k+1 . Consider the Sieve I, the Sieve II, and their associated
sequences of k-tuples. We have S(A , P, z) ≥ min{T (B, P, pk )}.
Proof. By definition, the sequences of remainders modulo ph (1 < h ≤ k) that form the sequence of k-tuples associated
to the Sieve I can have one or two selected remainders in every period (see Remark 8.1). However, the sequences
sh (1 < h ≤ k) that form the partial sum Sk associated to the Sieve II, by definition, have always two selected
remainders in every period. Suppose that we perform on the sequence of k-tuples associated to the Sieve I the
following operation: in each sequence of remainders modulo ph (1 < h ≤ k) that have only one selected remainder in
every period, we choose an arbitrary second selected remainder (the same element in every period of the sequence). We
obtain a partial sum Sk with a particular combination of selected remainders, where the number of permitted k-tuples
within the interval I[1, p2k ] is greater than or equal to min{T (B, P, pk )}. It is obvious that in the interval I[1, p2k ] of the
sequence of k-tuples associated to the Sieve I before performing the operation, the number of permitted k-tuples is also
greater than or equal to min{T (B, P, pk )}. Since I[1, p2k ] ⊂ I[1, x], it follows that S(A , P, z) ≥ min{T (B, P, pk )}.

We need one more lemma before proving the Main theorem.

Lemma 8.3. In the sequence of k-tuples associated to the Sieve I, if n (1 < n < x) is the index of a permitted
k-tuple, then n0 = x − n is the index of another permitted k-tuple.
√
Proof. Step 1. Let {p1 , p2 , p3 , . . . , pk } be the ordered set of the primes less than z = x; and let {b1 , b2 , b3 , . . . , bk } be
the ordered set of the remainders of dividing x by p1 , p2 , p3 , . . . , pk . Recall that in the sequences of remainders
modulo ph (1 ≤ h ≤ k) that form the sequence of k-tuples associated to the Sieve I, we have that 0 is a
selected remainder, and bh is also a selected remainder. Therefore, a given k-tuple whose elements are neither
0 nor bh (1 ≤ h ≤ k), by definition, is a permitted k-tuple.
Step 2. Let rh (1 ≤ h ≤ k) be the elements of the permitted k-tuple whose index is n. By definition, for the permitted
k-tuple whose index is n we have rh 6= √ bh (1 ≤ h ≤ k), since every bh is a selected remainder; this implies
√
n 6≡ x (mod ph ), for every prime ph < x. Hence n0 = x − n 6≡ 0 (mod ph ), for every prime ph < x. It
follows that the k-tuple whose index is n0 has no element equal to 0.

Step 3. Let rh0 (1 ≤ h ≤ k) be the elements of the k-tuple whose index is n0 = x − n. By definition, the permitted
k-tuple whose index is n has no element equal to 0, since √ it is a selected remainder. This means that n is not
divisible by any of the primes p1 , p2 , p3 , . . . , pk less than x. It follows that n 6≡ 0 (mod ph ) =⇒ n + x 6≡ x
(mod ph ) =⇒ n0 = x − n 6≡ x (mod ph ), for every prime ph ∈ {p1 , p2 , p3 , . . . , pk }. So, for the k-tuple whose
index is n0 = x − n we have rh0 6= bh (1 ≤ h ≤ k). From Step 1, Step 2 and this step, the k-tuple whose index
is n0 is a permitted k-tuple.

Finally, we prove the Main Theorem.

Remark 8.2. Let x > 49 be an even number. Assume that in the sequence of k-tuples associated to the Sieve I
there is a permitted k-tuple at position n = x − 1. Then, by Lemma 8.3, there is another permitted k-tuple at position
1; and furthermore, n = x − 1 is a prime, by Lemma 8.1, Step 1. So, 1 and x − 1 will appear among the unsifted
members of the set A . Note that in this case x is an even number of the form p + 1, where p is a prime.
Theorem 8.4. The Main Theorem √
Let x be an even number, and let k be the index of the greatest prime less than z = x. Furthermore, let Kα be the
number whose existence is established in Lemma 7.2. Every even integer x > p2k (k > Kα ) is the sum of two primes.

Proof. Step 1. Recall that S(A , P, z) denotes the sifting function of the Sieve I; assume that S(A , P, z) ≥ 3. By
Remark 8.2, among the unsifted members of the set A might appear 1 and x − 1. So, we can see that there
are at least S(A , P, z) − 2 integers n in A such that n is a prime and x − n is also a prime, by Theorem 8.1.
Step 2. By Lemma 7.3, for every level k > Kα we have min{T (B, P, pk )} > pk /2. On the other hand, S(A , P, z) ≥
min{T (B, P, pk )} for every even number x such that p2k < x < p2k+1 , by Lemma 8.2. It follows that
S(A , P, z) > pk /2 for every even number x > p2k , where k > Kα > 5, by definition (see Lemma 7.2). Then,
by Step 1, if x > p2k (k > Kα ) there must be at least one unsifted member n < x of A , which is a prime such
that x − n is also a prime. The theorem is proved.

35
9 Estimation of the number Kα
In this section we shall give an upper bound for Kα , whose existence is guaranteed by Lemma 7.2. In order to obtain
this upper bound we need to advance our understanding of the behaviour of the maximum density of permitted
k-tuples within the Right interval, and the minimum density of permitted k-tuples within the Left interval.

9.1 A formula for the maximum density of permitted k-tuples within the Right interval
P
Let Sk (k > 4) be a given partial sum of the series sk , where the first period is partitioned as we have seen before, in
a Left block and a Right block. In particular, consider for level h = k the Left interval I[1, p2k ]k and the Right interval
I[p2k + 1, mk ]k ; and for level h = 4 the Left interval I[1, p2k ]4 and the Right interval I[p2k + 1, mk ]4 . By Lemma 6.5 there
is a bijection between the values of the k-density in I[1, p2k ]k , and the values of the k-density in I[p2k + 1, mk ]k , and
there is also a bijection between the values of the 4-density in I[1, p2k ]4 , and the values of the 4-density in I[p2k +1, mk ]4 .
However, what is the relationship between the k-density within the Left interval I[1, p2k ]k , and the 4-density within the
Left interval I[1, p2k ]4 ? And what is the relationship between the k-density within the Right interval I[p2k + 1, mk ]k ,
and the 4-density within the Right interval I[p2k + 1, mk ]4 ? These are the questions in which we are interested.
By now, we know that the h-density within each interval I[1, mk ]h is equal to δh (1 ≤ h ≤ k). Consequently,
for a level k sufficiently large, when we subdivide every interval I[1, mk ]h into a Left interval and Right interval, it
seems reasonable to expect that the behaviour of δhLk and δhRk , between level h = 1 and level h = k, is analogous to
the behaviour of δh . In particular, with regard to the Right block, the values of δhRk (1 ≤ h ≤ k) approximate the
respective average δh more and more closely as the level k becomes large, by Lemma 7.1. Thus, it is easy to see that
in the Right block of the partition, between h = 1 and h = k, the maximum values of δhRk within the Right interval
I[p2k + 1, mk ]h tend to be proportional to the values of δh as k → ∞. This will allow us to derive a formula for δ̌hRk ,
in the Right block of the partition.
Now, given to levels h = i and h = j (1 ≤ i < j ≤ k), the following lemma shows that δ̌jRk is greater than the value
Rk
δ̌i δj /δi , computed by assuming δ̌hRk ∝ δh , between level h = 1 and level h = k.
Lemma 9.1. Let Sh be the partial sums from level h = 1 to level h = k (k ≥ 4). Let us consider the Right interval
in every partial sum Sh (1 ≤ h ≤ k). Let i, j be two levels such that 1 ≤ i < j ≤ k. Then

δj
δj < δ̌iRk < δ̌jRk .
δi
In other words, the value of δ̌jRk exceeds the value calculated by assuming δ̌hRk ∝ δh , between level h = 1 and level
h = k.
Proof. By Remark 7.3, for each level between h = 1 and h = k we have δ̂hRk < δh < δ̌hRk . Now, for level h = i, there
exists one combination of selected remainders in the sequences sh that form the partial sum Si , such that the density
of permitted i-tuples within the Right interval I[p2k + 1, mk ]i is equal to the maximum value δ̌iRk . Since by definition
the size of the Right interval I[p2k + 1, mk ]i of the partial sum Si is equal to mk − p2k , the number of subintervals of
size pi within this interval is (mk − p2k )/pi . Consequently, for this particular combination of selected remainders, the
number of permitted i-tuples within the Right interval I[p2k + 1, mk ]i is equal to δ̌iRk (mk − p2k )/pi . Therefore, using
Lemma 4.2 at each level transition h → h + 1, up to level h = j, we obtain

mk − p2k
       
pi+1 − 2 pi+2 − 2 pj − 2
δ̌iRk ··· =
pi pi+1 pi+2 pj
mk − p2k
     
pi+1 − 2 pi+2 − 2 pj − 2
= δ̌iRk ··· ,
pi pi+1 pj−1 pj
which is the average number of permitted j-tuples within the Right interval I[p2k + 1, mk ]j , for all the combinations
of selected remainders in the sequences sh from level i + 1 to level j, starting with the combination corresponding
to the maximum value δ̌iRk . Now, dividing by (mk − p2k )/pj (the number of subintervals of size pj within the Right
interval I[p2k + 1, mk ]j , for level h = j), we get

    
pi+1 − 2 pi+2 − 2 pj − 2
δ̌iRk ··· ,
pi pi+1 pj−1
which is the corresponding average j-density. We can write this expression as

    
pi+1 −2 pi+2 −2 pj −2

pi+1 − 2

pi+2 − 2
 
pj − 2
 δi pi pi+1 ··· pj−1
δ̌iRk ··· = δ̌iRk ,
pi pi+1 pj−1 δi

36
 and, using Lemma 3.2, it is easy to see that

    
pi+1 − 2 pi+2 − 2 pj − 2 δj
δ̌iRk ··· = δ̌iRk .
pi pi+1 pj−1 δi

Therefore, we can see that δ̌iRk δj /δi is the average density of permitted j-tuples within the Right interval I[p2k +
1, mk ]j , for all the combinations of selected remainders in the sequences sh from level h = i + 1 to level h = j such that
the combination of selected remainders in the sequences sh from level h = 1 to level h = i is the one corresponding to
the maximum value δ̌iRk . That is, δ̌iRk δj /δi is an average, not a maximum value. Consequently, it is easy to see that
δ̌jRk must be greater than δ̌iRk δj /δi , and then we can write

δj
δj < δ̌iRk < δ̌jRk .
δi

In particular for h = 4 and h = k (k > 4) it follows from Lemma 9.1 that

δk
δk < δ̌4Rk < δ̌kRk . (35)
δ4

We can see that δ̌kRk tends asymptotically to (δk /δ4 )δ̌4Rk as k → ∞, since the difference (δ̌kRk − δk ) tends to 0 as
k → ∞, by Lemma 7.1. In other words, as k → ∞ the ratio δ̌kRk /δ̌4Rk tends to be proportional to the ratio δk /δ4 .
On the other hand, we have the elementary identity

δ k Rk δ k  Rk 
δ̌4 = δk + δ̌4 − δ4 , (36)
δ4 δ4

which can be easily verified. (Note that this identity is not valid in the limit as k → ∞, since in this limit δ̌4Rk = δ4 ,
by Lemma 7.1.) So, by (35) and the preceding identity, we also have

δ k  Rk 
δk < δk + δ̌4 − δ4 < δ̌kRk . (37)
δ4
Therefore, by the rightmost inequality in (37) we can write

δ k  Rk 
δ̌kRk = δk + β4k δ̌4 − δ4 (k > 4), (38)
δ4

where β4k is some real number greater than 1. The number β4k measures how far the ratio (δ̌kRk − δk )/(δ̌4Rk − δ4 )
‘deviates’ from being exactly proportional to the ratio δk /δ4 .
Now, an obvious question occurs at this point: What is the behaviour of β4k as k → ∞? The answer to this question
is given in the next subsection.

9.2 The behaviour of β4k as k → ∞

We need some more lemmas.
P
Lemma 9.2. Let Sk (k ≥ 4) be a given partial sum of the series sk ; consider the interval I[1, mk ]h in every
partial sum Sh , from level h = 1 to level h = k. We have the identity

k−1
Y 
ph+1 − 2 δk
= . (39)
ph δ4
h=4

Proof. Using the formula (22) at each level transition from h = 4 to h = k − 1, we can write

     k−1
Y 
p5 − 2 p6 − 2 pk − 2 ph+1 − 2
δk = δ4 ··· = δ4 ,
p4 p5 pk−1 ph
h=4

37
 and then

k−1
Y 
ph+1 − 2 δk
= .
ph δ4
h=4

sk . Let us consider the Right interval I[p2k + 1, mk ]h

P
Lemma 9.3. Let Sk (k > 4) be a partial sum of the series
for a fixed level h < k. We have

δhRk
→1 as k → ∞.
δh
Remark 9.1. Note that since h is a fixed level, δh does not change as k → ∞.
Proof. By Lemma 7.1 there exist a function (k) such that (k) → 0 and

δh − (k) < δhRk < δh + (k)

as k → ∞. Dividing by δh we obtain

(k) δ Rk (k)
1− < h <1+ .
δh δh δh
Since (k)/δh → 0 as k → ∞, the lemma follows.

sk . Let us consider the Right interval I[p2k + 1, mk ]k

P
Lemma 9.4. Let Sk (k > 4) be a partial sum of the series
for the level k. We have

δkRk
→1 as k → ∞.
δk
Remark 9.2. Note that in this case, δk → ∞ as k → ∞, by Theorem 3.4.
Proof. The proof uses exactly the same argument as given in the preceding lemma, replacing h by k throughout.

Definition 9.1. Suppose given the partial sum Sk , and a particular combination of selected remainders in the
sequences sh (1 ≤ h ≤ k) that form Sk . Let I[p2k + 1, mk ]h be the Right interval for every partial sum Sh from h = 1
to h = k, and let δhRk be the true density of permitted h-tuples within every interval I[p2k + 1, mk ]h . We denote by
φh the ‘true’ factor by which we must multiply the h-density within the interval I[p2k + 1, mk ]h (denoted by δhRk ) to
Rk
obtain the (h + 1)-density within the interval I[p2k + 1, mk ]h+1 (denoted by δh+1 ), for every level transition h → h + 1
from h = 1 to h = k − 1. In symbols

Rk
δh+1 = δhRk φh
(1 ≤ h < k).
P
Lemma 9.5. Let Sk (k > 4) be a partial sum of the series sk . The partial product

k−1
Y
φh (40)
h=4

tends asymptotically to the average partial product

k−1
Y 
ph+1 − 2
(41)
ph
h=4

as k → ∞, whatever the combination of selected remainders in the sequences sh (1 ≤ h ≤ k) that form Sk .

38
Proof. Let us consider the levels h = 4 and h = k. Using the factors φh at each level transition from h = 4 to h = k −1,
we can write

k−1
Y
δkRk = δ4Rk φ4 φ5 · · · φk−1 = δ4Rk φh . (42)
h=4

(Note that each set of factors φ4 , φ5 , . . . , φk−1 is attached to every particular combination of selected remainders
in the sequences sh (1 ≤ h ≤ k) that form Sk .) On the other hand, by (39) we have

k−1
Y 
ph+1 − 2
δk = δ4 . (43)
ph
h=4

Dividing side by side (42) and (43), we obtain

k−1
Y
δ4Rk φh
δkRk h=4
= k−1 ,
δk Y ph+1 − 2
δ4
ph
h=4

and so

k−1
Y
φh R
δk k
h=4 δk
k−1
Y  ph+1 = R
δ4 k
.
−2 δ4
ph
h=4

Since δkRk /δk → 1 as k → ∞, by Lemma 9.4, and δ4Rk /δ4 → 1 as k → ∞, by Lemma 9.3, the lemma follows.

Now we are ready to give an answer to the question about the behaviour of β4k as k → ∞.
Lemma 9.6. The number β4k in (38) tends to 1 by the right as k → ∞.

sk ; consider the Right interval I[p2k + 1, mk ]h in every

P
Proof. Let Sk (k > 4) be a given partial sum of the series
partial sum Sh , from level h = 1 to level h = k. Recall that β4k > 1 by definition.

Step 1. For level h = 4, there exists one combination of selected remainders in the sequences sh that form the partial
sum S4 , such that the density of permitted 4-tuples within the Right interval I[p2k + 1, mk ]4 is equal to the
maximum value δ̌4Rk . Suppose that the factor by which we must multiply the h-density within the interval
I[p2k + 1, mk ]h (denoted by δhRk ) to obtain the (h + 1)-density within the interval I[p2k + 1, mk ]h+1 (denoted by
Rk
δh+1 ) is equal to the average factor (ph+1 − 2)/ph , for every level transition h → h + 1 from h = 4 to h = k − 1,
whatever the combination of selected remainders in the sequences sh from h = 5 to h = k; that is, we are
assuming that φh = (ph+1 − 2)/ph (4 ≤ h < k). In this case, δ̌kRk would be equal to

k−1
Y 
ph+1 − 2 δ k Rk
δ̌4Rk = δ̌ (44)
ph δ4 4
h=4

(see identity (39)); that is, δ̌kRk would be equal to

δk  Rk 
δk + δ̌4 − δ4 (45)
δ4

(see identity (36)).

39
Step 2. Clearly, the assumption made in the preceding step is not true (see (35) and (37)). We know certainly that
for every combination of selected remainders in the sequences sh (5 ≤ h ≤ k) that form Sk the factors
φh (4 ≤ h < k) are different from the corresponding average factors (ph+1 − 2)/ph . And there exists one of
these combinations of selected remainders (and its attached factors φh (4 ≤ h < k)) such that the density of
permitted k-tuples within the Right interval I[p2k + 1, mk ]k is the maximum value δ̌kRk . Now, for a level k not
too large, the partial product

k−1
Y
φh (46)
h=4

corresponding to this particular combination of selected remainders could be quite larger than the average
product (41), and so, δ̌kRk could be quite far from the proportional value given by (45). So, since the factor β4k
in the formula (38) stands for taking into account the excess over the proportional value, we conjecture that,
in this case, β4k will be far above 1. However, as k → ∞, the partial product (46) approaches the average
product (41), whatever the combination of selected remainders in Sk , by Lemma 9.5; and furthermore, we
know that δ̌kRk tends asymptotically to (45) as k → ∞, by (37) and Lemma 7.1. Therefore, it is reasonable to
think that β4k approaches 1 as k → ∞.
Step 3. On the contrary, suppose that β4k > C holds as k → ∞, for some constant C > 1. This means that as k → ∞,
for every partial sum Sk there exists a combination of selected remainders in the sequences sh that form Sk
for which the partial product (46) holds too far from the average partial product (41). That is, there exists a
combination of selected remainders in every Sk such that the inequality

k−1 k−1
Y 
Y ph+1 − 2
φh > B
ph
h=4 h=4

holds as k → ∞, for some constant B > 1; and this contradicts Lemma 9.5. Therefore, we conclude that
β4k → 1 by the right as k → ∞.

9.3 A formula for the maximum density of permitted k-tuples within the Left interval
We recall that the first period of the partial sum Sk (k ≥ 4), in horizontal position, can be seen as a matrix, with mk
columns and k rows; and we recall also that this matrix was partitioned into two blocks: the Left block formed by the
columns from n = 1 to n = p2k ; and the Right block formed by the columns from n = p2k + 1 to n = mk . From (38)
we have a formula for δ̌kRk within the Right block of the partition. However, we need a similar formula for δ̂kLk within
the Left block of the partition; in order to derive this formula, we proceed as follows. Suppose that for the level 4 we
know the minimum value of the density of permitted 4-tuples δ̂4Lk , within the Left interval I[1, p2k ]4 (Left block of the
partition). Using the function f4 of Lemma 6.5 we can compute, for the level 4, the maximum value of the density
of permitted 4-tuples δ̌4Rk , within the Right interval I[p2k + 1, mk ]4 (Right block of the partition). Then, using the
formula in (38) we can compute, for the level k, the maximum value of the density of permitted k-tuples δ̌kRk , within
the Right interval I[p2k + 1, mk ]k (Right block of the partition). Finally, using the function fk−1 of Lemma 6.5 we can
compute, for the level k, the minimum value of the density of permitted k-tuples δ̂kLk , within the Left interval I[1, p2k ]k
(Left block of the partition). This formula is given in the following lemma.
P
Lemma 9.7. Let Sk (k > 4) be a partial sum of the series sk . Within the first period of Sk , consider the Left
interval I[1, p2k ]h and the Right interval I[p2k + 1, mk ]h of every partial sum Sh from h = 1 to h = k. We have

δk  
δ̂kLk = δk − β4k δ4 − δ̂4Lk , (47)
δ4
where β4k is the number which appears in (38).

Proof. Step 1. We compute the value of f4 (see Lemma 6.5) at x = δ̂4Lk (the minimum density of permitted 4-tuples
within the Left interval I[1, p2k ]4 ). We obtain

    p2k
δ̌4Rk = f4 δ̂4Lk = δ4 − δ̂4Lk − δ4 .
mk − p2k

See Remark 7.4.

40
Step 2. Next, we take the maximum density of permitted 4-tuples within the Right interval I[p2k + 1, mk ]4 obtained
in the previous step, and using the formula in (38), we get

p2k
  
δk   δk  
δ̌kRk = δk + β4k δ̌4Rk − δ4 = δk + β4k δ4 − δ̂4Lk − δ4 − δ4 .
δ4 δ4 mk − p2k

Step 3. Finally, we compute the value of fk−1 (see Lemma 6.5) at x = δ̌kRk (the maximum density of permitted k-tuples
within the Right interval I[p2k + 1, mk ]k , obtained in the preceding step). See Remark 7.4. We obtain

    m − p2
k
δ̂kLk = fk−1 δ̌kRk = δk + δk − δ̌kRk k
=
p2k
p2k mk − p2k
    
δk  
= δk + δk − δk + β4k δ4 − δ̂4Lk − δ4 2 − δ 4 =
δ4 mk − pk p2k
p2k mk − p2k
    
δk 
= δk + δk − δk + β4k − δ̂4Lk − δ4 2 =
δ4 mk − pk p2k
p2k mk − p2k
   
δk 
= δk + −β4k − δ̂4Lk − δ4 2 =
δ4 mk − pk p2k
δk  
= δk − β4k δ4 − δ̂4Lk .
δ4

9.4 An upper bound for Kα

Before computing an upper bound for Kα , we need to obtain an upper bound for β4k . Since β4k > 1 by definition,
it is convenient to compute the upper bound for β4k in a range of values of k where this number is close to 1. By
Lemma 9.6, we know that β4k → 1 as k → ∞. Now, how large must k be for β4k to be close to 1?
In the case examined in Remark 7.1 and Remark 7.2, we have seen that, if k satisfies the conditions required in
Remark 7.1, for each level from h = 1 to h = k the values of δhRk will be very close to δh , whatever the combination
of selected remainders in the sequences sh that form the partial sum Sk , and furthermore, for every level from h = 4
to h = k − 1, the factor φh will be very close to the average factor (ph+1 − 2)/ph (see Lemma 6.3). It follows that
for every combination of selected remainders in the sequences sh that form the partial sum Sk , the partial product in
(40) will be very close to the average partial product in (41). In particular, this is true for the combination of selected
remainders corresponding to the maximum value δ̌kRk . Therefore, δ̌kRk will be very close to (45), and β4k must be close
to 1 (see the proof of Lemma 9.6). So, we formulate the following criterion.
Criterion . Let k be large enough that the size of the Left interval I[1, p2k ] is negligible compared to the size of the
interval I[1, mk ], and the size of I[1, p2k ] is negligible compared to ck . Under these conditions k is sufficiently large to
guarantee that the number β4k which appears in (38) is close to 1.

For a partial sum Sk where k ≥ 35, the ratio of the size of I[1, p2k ] to the size of the interval I[1, mk ] is less than
1.5 × 10−53 , and the ratio of the size of I[1, p2k ] to ck is less than 0.9 × 10−51 . Clearly, the size of the Left interval
I[1, p2k ] is negligible compared to the size of I[1, mk ], and compared to ck . Consequently, by the preceding criterion,
the number β4k must be close to 1 for a level k ≥ 35.
We proceed to compute an upper bound for β4k , valid for k ≥ 35. From the formula (47) we have

 
Lk
δ
δ4 k − δ̂ k
β4k =  ,
δk δ − δ̂ Lk
4 4

for any given level k > 4; and since δ̂kLk can not be less than zero, we can write

δ4 δk δ4
1 < β4k <  = . (48)
δk δ − δ̂ Lk δ − δ̂ Lk
4 4 4 4

Now, since β4k is close to 1 for a level k ≥ 35, and β4k → 1 as k → ∞ by Lemma 9.6, we can see that (48) holds for
every k ≥ 35.

41
 In order to compute the upper bound in (48), we need the value of δ̂4Lk . Note that by (11) there are 1260
combinations of selected remainders in the sequences sh (1 ≤ h ≤ 4) that form the partial sum S4 . The minimum
number of permitted 4-tuples within the Left interval I[1, p2k ]4 of the partial sum S4 (for k = 35) can be obtained by
explicit computation of the number of permitted 4-tuples for every combination of selected remainders in S4 , and then
taking the minimum among these values1 . Then, with the minimum number of permitted 4-tuples within I[1, p2k ]4
(where k = 35) we can compute δ̂4Lk ; in this way we obtain δ̂4Lk = 0.49945 (rounded up).
With this value of δ̂4Lk and taking δ4 = 1/2, we can return to (48), obtaining

1 < β4k < 909.1. (49)

Now we are ready to compute an upper bound for Kα , using the formula (47); we proceed in the following way.
For every k starting from level k = 35 we perform the following procedure:
Procedure . Step 1. Compute a lower bound for the density of permitted 4-tuples within the Left interval I[1, p2k ]4 .

Step 2. Compute  as the difference between δ4 and the lower bound of the preceding step.
Step 3. Compute δk .
Step 4. Compute a lower bound for the density of permitted k-tuples within the Left interval I[1, p2k ]k .

Step 5. Increase k by 1.
We use Lemma 5.2 and Lemma 3.1 in the steps 1 and 3 respectively; and we take δ4 = 1/2 in Step 2. We use the
formula (47) in the Step 4, replacing (δ4 − δ̂4Lk ) by , β4k by its upper bound in (49), and δk by the value obtained in
the Step 3. The procedure is carried out until the lower bound for the density of permitted k-tuples within I[1, p2k ]k ,
computed in Step 4, overcome the value δ4 = 1/2; at this point, the value of k is the upper bound for Kα . We obtain

Kα < 89. (50)

For some values of k, Table 7 gives pk , , δk and the lower bound for δkLk computed by using the preceding
procedure.

Table 7: Computed lower bounds for δ4Lk .

k pk  δk lb δkLk
35 149 46.947 × 10−4 2.4099 -18.161
36 151 45.655 × 10−4 2.4099 -17.595
37 157 42.173 × 10−4 2.4738 -16.495
38 163 39.184 × 10−4 2.5368 -15.536
39 167 37.416 × 10−4 2.5679 -14.902
40 173 34.800 × 10−4 2.6294 -14.008
. . . . .
81 419 5.9453 × 10−4 4.7246 -0.3826
82 421 5.8890 × 10−4 4.7246 -0.3342
83 431 5.6228 × 10−4 4.8144 -0.1076
84 433 5.5725 × 10−4 4.8144 -0.0635
85 439 5.4207 × 10−4 4.8589 0.0670
86 443 5.3222 × 10−4 4.8810 0.1578
87 449 5.1782 × 10−4 4.9251 0.2881
88 457 5.0012 × 10−4 4.9909 0.4526
89 461 4.9124 × 10−4 5.0127 0.5355
90 463 4.8739 × 10−4 5.0127 0.5706
91 467 4.7894 × 10−4 5.0344 0.6504
92 479 4.5528 × 10−4 5.1422 0.8855
93 487 4.4037 × 10−4 5.2066 1.0378
. . . . .

1 The numerical computation were carried out on a desktop computer, using a program written in language C.

42
9.5 Conclusion
By the Main theorem and the upper bound in (50) we can say that every even integer x > p2k , where k ≥ 89 (p2k ≥
212521), is the sum of two primes. Now, it is a known fact that the strong Goldbach conjecture has already been
verified for all n ≤ 4 × 1017 [10]. Therefore, we conclude that every even number x > 4 can be expressed as the sum
of two odd primes; and then, the binary Goldbach conjecture is proved.

Acknowledgements

The author wants to thank Dra. Patricia Quattrini (Departamento de Matematicas, FCEyN, Universidad de
Buenos Aires) for helpful conversations and Dr. Hendrik W. Lenstra (Universiteit Leiden, The Netherlands) for his
extremely useful suggestions.

References
[1] H. Halberstam and H.-E. Richert, Sieve Methods, Academic Press, London, 1974.
[2] A. C. Cojocaru and M. Ram Murty, An Introduction to Sieve Methods an Their Applications, Cambridge University Press,
London, 2005.
[3] V. Brun, La série 1/5 + 1/7 + 1/11 + 1/13 + 1/17 + 1/19 + 1/29 + 1/31 + 1/41 + 1/43 + 1/59 + 1/61 + . . ., où les dénominateurs
sont nombres premieres jumeaux est convergente où finie, Bull. Sci. Math., 43, (1919), 124–128.
[4] V. Brun, Le crible d’Eratosthène et le théorème de Goldbach, C. R. Acad. Sci. Paris, 168, (1919), 544–546.
[5] J. R. Chen, On the representation of a large even integer as the sum of a prime and the product of at most two primes. II,
Sci. Sinica 21 (1978), no. 4, 421–430. MR0511293 (80e:10037)
[6] R. K. Guy, Unsolved Problems in Number Theory, third edition, Springer-Verlag, Berlin, 2004. MR2076335 (2005h:11003).
[7] G. H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers, sixth edition, Oxford Univ. Press, Oxford,
2008. MR2445243 (2009i:11001).
[8] E. Landau, Elementary Number Theory, translated by J. E. Goodman, Chelsea, New York, 1958. MR0092794 (19,1159d).
[9] H. L. Montgomery and R. C. Vaughan, The exceptional set in Goldbach’s problem, Acta Arith. 27 (1975), 353–370.
MR0374063 (51 #10263).
[10] T. Oliveira e Silva, Goldbach conjecture verification,
http://www.ieeta.pt/~tos/goldbach.html.
[11] J. Rey Pastor, P. Pi Calleja and C. A. Trejo, Mathematical Analysis. Vol. I (in Spanish), Fifth edition], Editorial Kapelusz,
Buenos Aires, 1960. MR0145013 (26 #2552a).
[12] I.M. Vinogradov, The Representation of an Odd Number as the Sum of Three Primes, Dokl. Akad. Nauk SSSR 16 (1937),
139–142.
[13] H. Helfgott, The ternary Goldbach conjecture is true, http://www.arXiv.org/abs/1312.7748.
[14] Terence Tao, Open question: The parity problem in sieve theory,
https://terrytao.wordpress.com/2007/06/05/open-question-the-parity-problem-in-sieve-theory/.
[15] A.Selberg, The general sieve method and its place in prime number theory, Proc. Int. Congress of Mathematicians Cam-
bridge, MA (Providence, RI. American Mathematical Society), 1 (1950), 262–289.
Ricardo G. Barca
Universidad Tecnologica Nacional
Buenos Aires (Argentina)
E-mail address: [email protected]

43