Lower bound for the remainder in the prime-pair conjecture
Jacob Korevaar (University of Amsterdam)

TL;DR
This paper investigates the lower bounds of the error term in the prime-pair conjecture, linking it to the zeros of the Riemann zeta function and proposing an approximation analogous to Riemann's for prime counting.
Contribution
It establishes a heuristic connection between the remainder in prime-pair counts and the zeros of the zeta function, providing insights into the conjecture's limitations.
Findings
Heuristic argument suggests the remainder cannot be smaller than x^beta.
Proposes an approximation for pi_{2r}(x) similar to Riemann's for pi(x).
Links the size of the error term to the supremum of zeta zeros' real parts.
Abstract
For any positive integer r, let pi_{2r}(x) denote the number of prime pairs (p, p+2r) with p not exceeding (large) x. According to the prime-pair conjecture of Hardy and Littlewood, pi_{2r}(x) should be asymptotic to 2C_{2r}li_2(x) with an explicit positive constant C_{2r}. A heuristic argument indicates that the remainder e_{2r}(x) in this approximation cannot be of lower order than x^beta, where beta is the supremum of the real parts of zeta's zeros. The argument also suggests an approximation for pi_{2r}(x) similar to one of Riemann for pi(x).
Click any figure to enlarge with its caption.
Figure 1
Figure 2| 4 | ||||
| 13 | ||||
| 52 | ||||
| 259 | 274 | 0.945 | ||
| 1595 | 1599 | 0.997 | ||
| 10548 | 10560 | 0.999 | ||
| 74914 | 75223 | 0.996 | ||
| 563533 | 563804 | 0.9995 |
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAnalytic Number Theory Research · Limits and Structures in Graph Theory · Algebraic Geometry and Number Theory
Lower bound for the
remainder
in the Prime-Pair Conjecture
Jacob Korevaar
(Date: Spring 2008)
Abstract.
Taking let denote the number of prime pairs with . The prime-pair conjecture of Hardy and Littlewood (1923) asserts that with an explicit constant . A heuristic argument indicates that the remainder in this approximation cannot be of lower order than , where is the supremum of the real parts of zeta’s zeros. The argument also suggests an approximation for similar to one of Riemann for .
2000 Mathematics Subject Classification:
11P32
1. Introduction
For let denote the number of prime pairs with . The famous prime-pair conjecture (PPC) of Hardy and Littlewood [11] asserts that for ,
[TABLE]
Here is the ‘twin-prime constant’,
[TABLE]
and the general ‘prime-pair constant’ is given by
[TABLE]
No proof of (1.1) is in sight, but our arguments make it plausible that the best asymptotic estimate for the remainder
[TABLE]
cannot be as small as ; see Section 10.
For the following we set
[TABLE]
where runs over the complex zeros of . Recall that Riemann’s Hypothesis (RH) asserts that . For the case of the prime number theorem it is known that the remainder
[TABLE]
is for every , but cannot be for any . Indeed, a formula from Riemann’s work suggests the approximation
[TABLE]
for any . Here the sum over is a limit of ‘symmetric’ partial sums; it becomes significant for very large . In 1895 von Mangoldt obtained the following formula, from which he derived a proof of (1.6); cf. Davenport [5], Edwards [7]:
[TABLE]
The formula is exact for all where is continuous.
For prime pairs one would expect that
[TABLE]
Here the symbol is shorthand for the -notation. Some time ago, Dan Goldston [9] suggested that the author’s complex method (now in [14]) might provide a good lower bound for . In this note we use such an approach to obtain a conditional proof for
Metatheorem 1.1**.**
Statement is correct.
For our analysis we introduce an analog to :
[TABLE]
It is not difficult to see that the PPC (1.1) is equivalent to the asymptotic relation
[TABLE]
For our subsequent analysis it is convenient to work with the following series of Dirichlet-type, where :
[TABLE]
Note that for the boundary behavior of as , the denominators may be replaced by . Hence by a two-way Wiener–Ikehara theorem for Dirichlet series with positive coefficients, the PPC in the form (1.11) is true if and only if the difference
[TABLE]
has ‘good’ boundary behavior as . That is, should tend to a distribution which is locally equal to a pseudofunction. By a pseudofunction we mean the distributional Fourier transform of a bounded function which tends to zero at infinity; see [13]. It cannot have poles and is locally given by Fourier series whose coefficients tend to zero. In particular itself would have to show pole-type behavior, with residue , for angular approach of to from the right; there should be no other poles on the line .
Heuristic arguments make it plausible that has a meromorphic extension to some half-plane where :
Metatheorem 1.2**.**
For every there is a number such that
[TABLE]
where is holomorphic in .
Our approach would take care of Metatheorem 1.1 in the case . Metatheorem 1.2 suggests the following approximation for :
Metatheorem 1.3**.**
For each there is a number such that
[TABLE]
The case of Metatheorem 1.1 is more subtle. It requires consideration of the function
[TABLE]
and the associated Dirichlet series
[TABLE]
Here our arguments suggest
Metatheorem 1.4**.**
If there is a representation for similar to the one for . However, if one has
[TABLE]
with constants and a function that is holomorphic for and has ‘good’ boundary behavior as .
Metatheorems 1.2 and 1.4 lead to plausible approximations for and finally, :
Metatheorem 1.5**.**
There are constants such that
[TABLE]
The constants come from the special case of the Bateman–Horn conjecture [1], [2] that involves the prime pairs : the number of such pairs with should satisfy an asymptotic relation
[TABLE]
with certain specific constants . The analysis in Sections 8–10, which includes computations by Fokko van de Bult [3], supports and utilizes
Metatheorem 1.6**.**
The Bateman–Horn constants in have mean value one (just like the Hardy–Littlewood constants ).
2. Auxiliary functions
Integration by parts shows that the estimate with small would be equivalent to the inequality
[TABLE]
Note that (1.17) and (2.1) would imply holomorphy of the difference
[TABLE]
Comparison of the series for and will show that the difference is holomorphic for ; cf. Lemma 7.1 below. Hence an estimate would imply holomorphy of the difference in (1.13) for , provided .
We need precise information on the function derived from (1.12).
Lemma 2.1**.**
For one has
[TABLE]
where has an analytic continuation to the half-plane . This gives a meromorphic continuation of :
[TABLE]
where is holomorphic for .
Proof.
Taking one has
[TABLE]
It follows that
[TABLE]
where is holomorphic for . Hence by differentiation,
[TABLE]
[TABLE]
where is also holomorphic for . Finally use a standard formula for :
[TABLE]
cf. Titchmarsh [16], and set . ∎
We need the representation in Theorem 3.1 below. It involves sufficiently smooth even sieving functions depending on a parameter . The basic functions have and support ; we require that , and are absolutely continuous with of bounded variation. An example involving the Jackson kernel for is given by
[TABLE]
An important role is played by a Mellin transform associated with the Fourier transform . For
[TABLE]
In the special case of one finds
[TABLE]
The function extends to a meromorphic function for with simple poles at the points . The residue of the pole at is with , and . Furthermore, the standard order estimates
[TABLE]
for and imply the useful majorization
[TABLE]
3. A basic representation
The following result is related to Theorem in [14], but more precise. It will be verified in Section 6.
Theorem 3.1**.**
For any and with there is a meromorphic representation
[TABLE]
Here is given by , also for ; the functions are holomorphic for . The function has a purely quadratic pole at ; see . On the basis of the PPC one expects that for , the function has a first-order pole at with residue . The functions and are described in – below. The error term is holomorphic for .
The function is given by the sum
[TABLE]
Here runs over the complex zeros of . The combination is meromorphic for , with poles at and the points ; the apparent poles at the points cancel each other. The simple poles at and have residues
[TABLE]
The function is given by the sum
[TABLE]
Here and independently run over the complex zeros of . It is convenient to denote the sum of the first two terms by ; for it has poles at and at the points . The double series defines a function which we call . Under RH the series is absolutely convergent for . Indeed, setting , and , the inequalities (2.12), (2.13) show that the terms in the double series are majorized by
[TABLE]
Observing that the number of zeros with is , the convergence now follows from a discrete analog of Lemma 5.1 below.
If there is absolute convergence for . For the double sum may be interpreted as a limit of sums over the zeros , whose imaginary part has absolute value less than , as through suitable values; see [14]. By the apparent poles of at the points with must cancel each other. Formally, there is cancellation also at the other points .
4. Metatheorem 1.1 for and
Metatheorem 1.2
. Taking , formulas (3.1)–(3.5) show that
[TABLE]
with a ‘symmetric’ sum over and a remainder that is holomorphic for . Recall from Section 2 that an inequality with would imply holomorphy of the difference
[TABLE]
for . Hence if such holomorphy leads to a contradiction, so does (1.9). This would prove Metatheorem 1.1 for the case .
Suppose now that for all and some , the differences are holomorphic in the strip given by . Then by (4), the function has a meromorphic continuation [also called ] to , with poles at and some points . The pole at will have residue
[TABLE]
At this point we use the fact that the prime-pair constants have mean value one. Good estimates were obtained by Bombieri–Davenport and Montgomery; these were later improved by Friedlander and Goldston [8] to
[TABLE]
It follows that is as , and even . Hence by (3) the residue at of (the meromorphic continuation of) the double sum also is . [By (2.4) the pole of at is purely quadratic.] The estimate is not surprising if one observes that occurs in the terms of only as a factor ; cf. (2.5). For the exponents have real part , which is less than if .
If the latter kind of heuristic has general validity, the residues of the poles of or at the points in must also be (at least when and is small). In view of (4) this would imply that many of the functions must become singular at points in , which would contradict our assumption on the differences .
What would be a reasonable hypothesis on the form of the singularities? Let us start with and suppose that is holomorphic in . The residue will equal . Thus it changes character as passes through the value : it will be linear in , of the form , for , and this linear term is augmented by the nonlinear term as enters the interval . It is plausible that the poles of at the points in will be affected in a corresponding manner. More precisely, the residues should change from the linear form to as enters the interval . If that is correct, the function must have first-order poles at the points in with residue . The combination
[TABLE]
would be holomorphic in .
Next taking (and if desired, using a modified function which vanishes on , say), one may pass to the case , etc. Thus one is led to the postulate that each function has poles at the points in some strip with residue . If this is correct, the residue of or at the poles in will be
[TABLE]
Since the constants have average this would be consistent with the earlier argument that should be .
It follows that Metatheorem 1.2 is altogether plausible, and this suggests Metatheorem 1.3.
5. Integral representations
Setting (and later ), we write for the ‘vertical line’ ; the factor in complex integrals will be omitted. Thus
[TABLE]
Since it is important for us to have absolutely convergent integrals, we often have to replace a line by a path with suitable and :
[TABLE]
cf. Figure 1.
Thus, for example,
[TABLE]
with absolute convergence if and . Similarly for . For the combination
[TABLE]
with , one can now write down an absolutely convergent repeated integral. In [14] it was combined with (2) to obtain a repeated complex integral for the sieving function in which and occur separately. Taking , , it was found that
[TABLE]
We then considered the following integral:
[TABLE]
with suitable paths of integration and for appropriate ; cf. Section 6. Next, substituting the Dirichlet series for , formula (5) led to the expansion
[TABLE]
where is holomorphic for . Indeed, for odd numbers , the product can be only if either or is of the form for some . Thus was extended to a holomorphic function on the half-plane .
To verify the absolute convergence of the repeated integral in (5) we substituted , , and used the inequalities (2.12), (2.13) together with a simple lemma:
Lemma 5.1**.**
For real constants , the function
[TABLE]
is integrable over if and only if , , and .
For the convergence of the repeated integral in (5) we also used the fact that the quotient grows at most logarithmically in for , and for under RH; cf. (2.6) and Titchmarsh [16]. The holomorphy of the integral for then followed from locally uniform convergence in .
The following sections serve as preparations for the case of Theorem 1.1, so that RH is satisfied.
6. Derivation of Theorem 3.1 under RH
Changing variables in (5) one obtains
[TABLE]
with new paths and the point to the left of them. Using Cauchy’s theorem and assuming RH, one may take , with small and , . [Without RH one could take , and .] The absolute convergence of the repeated integral follows from Lemma 5.1.
We now move the paths of integration across the poles of the integrand, the points where or is equal to , or . For the transition one may use quasi-rectangular contours , see Figure 2, where runs through a sequence such that the horizontal segments at level are as far from zeros of the zeta function as possible.
Moving the -path to a line with , one gets
[TABLE]
say, where by the residue theorem
[TABLE]
with
[TABLE]
Observe that for given with , and small , the function is holomorphic in on and between the paths and . Defining for by continuity at the points and , it becomes holomorphic in for . Indeed, the poles at the point cancel each other, as do the poles at the points .
What conditions do and have to satisfy? The double integral for must be absolutely convergent, which requires ; cf. Lemma 5.1. Also, one should not cross a pole of during the shifting operation. Thus should remain less than . Taking small, this allows values of close to . Since we ultimately want to consider values of around , we take . Varying and , the double integral will define as a holomorphic function for and .
We next consider the single integral for . Moving the path across the points , and to the line , we obtain the decomposition
[TABLE]
Working out the residue with the aid of (6) one obtains nine terms. Five of these combine into the function of (3). Using the pole-type behavior of at the point (Section 2), the first term in provides an important pole at the point :
[TABLE]
where is holomorphic for . The other terms in only present simple poles at the points . A short computation shows that the residues at those poles are all equal to . The four remaining terms coming from the big residue provide the function of (3).
It remains to consider the single integral along in (6), let us call it , which we want to define a holomorphic function in a relatively wide strip. For that we need absolute convergence of the ‘double sum’, formed by the -integral along and the sum over in (6). With and , the standard estimates give the following majorant for the integrand:
[TABLE]
Taking , the analog of Lemma 5.1 for the integral of a sum proves the absolute convergence and holomorphy of the integral when .
Combination of the above results with (5) will verify Theorem 3.1 under RH.
7. The differences and
To treat the case of Theorem 1.1 one has to work with the function of (1.17) instead of . In the following and are primes that run over all pairs of the type indicated with the sums. By the definition of ,
[TABLE]
where is holomorphic for . The final sum comes from the cases , and , . There are only finitely many of the form such that . The function includes these and the cases where or is a prime power with exponent . Continuing one obtains a sum over the prime pairs and a sum over prime pairs :
Lemma 7.1**.**
One has
[TABLE]
where and are holomorphic for , and , respectively.
We also consider corresponding partial sums, from (1.16) and
[TABLE]
A sieving argument would show that ; cf. [2], [10], [12].
Lemma 7.2**.**
By and ,
[TABLE]
We can now formulate a refinement of Theorem 3.1. In view of (7.1) the discussion in Section 6 shows the following.
Theorem 7.3**.**
For and , one has
[TABLE]
where the error terms are holomorphic for .
We wish to use (7.3) for the study of the prime-pair functions when , and for that we need information on the functions near the line . This requires the consideration of prime pairs .
8. Prime pairs
Let with , and define
[TABLE]
Does tend to infinity as ? Not if can be factored, nor if , for then is divisible by when . However, if is irreducible and for every prime , there is a positive integer such that does not divide , one would expect that as . This is a very special case of what is usually called Schinzel’s conjecture [15]. More generally, let be any polynomial of degree with integer coefficients. For irreducible we set
[TABLE]
and define
[TABLE]
The product will converge, but may be zero; if can be factored, we define . Then a special case of the general conjecture of Bateman and Horn [1], [2] asserts the following:
Conjecture 8.1**.**
As , one has
[TABLE]
Cf. Davenport and Schinzel [6], and Hindry and Rivoal [12]. In the special case of the polynomial
[TABLE]
one finds that for , using the Legendre symbol,
[TABLE]
Here generates a real character (different from the principal character) belonging to a modulus . The convergence of the product for thus follows from the known convergence of series .
Fokko van de Bult [3] has computed
[TABLE]
and counted
[TABLE]
for . His results are in excellent agreement with Conjecture 8.1. In the table the number is compared to rounded values
[TABLE]
The table also gives some ratios
[TABLE]
These seem to converge to rather quickly!
We can now discuss the functions
[TABLE]
of (7.1). Assuming that the Bateman–Horn conjecture is true for the polynomials , one obtains the following asymptotic relation for the functions of (7.3):
[TABLE]
For us it will be convenient to write this relation in the form
[TABLE]
By the two-way Wiener–Ikehara theorem of [13] and integration by parts, relation (8.10) is equivalent to the statement that the difference
[TABLE]
has good (that is, pseudofunction) boundary behavior as . In particular must have a first-order pole at with residue , and no other poles on the line .
Before returning to the proof of Theorem 1.1 we give a supporting argument for Metatheorem 1.6, which asserts that the constants have mean value one.
9. A function . Metatheorem
Using paths specified below we will study the function
[TABLE]
Here analogs to (5), (5) provide the following expansion for , cf. (7.1):
[TABLE]
where and is holomorphic for . Comparison with in (2.3) shows that
[TABLE]
where is holomorphic for . Formula (9) may be used to define as a holomorphic function for .
In (9), assuming RH, one may take , and , with small . Varying , the integral thus represents as a holomorphic function for and . We now move the -path across the poles at the points , and to the path , where and . Then the residue theorem gives
[TABLE]
say, where
[TABLE]
with as in (6). Recall that the apparent poles of at the points and cancel out.
We next move the -path in the integral for to . Picking up residues at , and the zeros of , the result is
[TABLE]
say, where
[TABLE]
The integrals for and in (9.4) and (9) will define holomorphic functions for .
Let denote the strip . We have to know the boundary behavior of as . What sort of poles on the line will result from the three products in the formula for ? The first product involves , which by (6) is holomorphic on , and , which has poles at the points . The resulting poles have principal parts
[TABLE]
Turning to the second product, the function is holomorphic on , except for a simple pole at due to the pole of for . The other factor is , and by a short calculation, cf. (2), the principal part of the pole at works out to
[TABLE]
In the third product the function is holomorphic on . However, the factors introduce poles at the points . The poles in the product have principal part
[TABLE]
hence they cancel the poles at the points in (9.8). The third product also generates a double series :
[TABLE]
The series is absolutely convergent for . Its sum will have an analytic continuation to , also denoted , but we do not know much about its behavior near the line ; see below.
In support of the hypothesis that the poles of at the points cancel out one may analyze an integral related to . It is obtained from (9) by interchanging the roles of and . The new integral is of course equal to . In the analysis the role of is now taken by
[TABLE]
Here the apparent poles at the points and cancel out.
Summary 9.1**.**
Assume RH. Combination of (9) and the subsequent results shows that for ,
[TABLE]
where and are holomorphic for .
Observe that the (analytic continuation of the) sum must have a second-order pole at the point . Indeed, has a quadratic pole at , see (9.3), and by sieving, the functions cannot have a worse singularity at than a first-order pole. In Section 8 it was made plausible that the functions indeed have a first-order pole at . What can we say about the mean value of the residues , or of the numbers ? By (9.1) and (8.10) the residue of at is equal to
[TABLE]
Now it is plausible that this residue is as . Indeed, occurs in the terms of only as a factor ; cf. the considerations in Section 4. Assuming , and letting approach the constant function on , it follows from (9.14) that
[TABLE]
Thus the numbers should have mean value , as asserted in Metatheorem 1.6. The metatheorem is supported by numerical evidence: a computation of the first fifteen constants by Fokko van de Bult [3] gave their average as .
Remark 9.2**.**
Simple adaptation of our heuristics and accompanying numerical results indicate that relation (9.15) and Metatheorem 1.6 can be extended to the case of prime pairs with ; see [4].
10. Metatheorem 1.1 for and
Metatheorem 1.4
Taking , Theorem 7.3 shows that
[TABLE]
where . The error term is holomorphic for . To complete the proof of Theorem 1.1 we have to deal with the case , so that RH holds. Suppose now that for and ,
[TABLE]
Then the corresponding functions of (2.2) have continuous boundary values for ; cf. (1.17).
On the basis of Section 8 we may plausibly assume that the functions show ‘good’ (pseudofunction) boundary behavior for . Hence by (10), the function would have a ‘good’ extension to the strip , apart from first-order poles at , and the points . ‘Good’ meaning: holomorphy for and good boundary behavior after subtraction of the poles. As in Section 4, the pole at of , or of the double sum in (3), will have residue as in (4.3). By the mean-value property of the constants this residue is as . We recall that this was not surprising because occurs in the terms of only as a factor .
Since by our assumption (10.2) the functions would have no pole at , the pole of or at that point would have residue
[TABLE]
with as in (9.14). In Section 9 it was made plausible that as . We used both numerical evidence and the argument that the terms of the double sum contain only as a factor . However, the latter argument would also suggest that . Indeed, the terms in the double series of (3) contain only as a factor !
The contradiction indicates that assumption (10.2) is false, and that formula (10.3) for is incorrect. It is most likely that the functions have poles at the point , and that these poles more or less cancel those of the functions . Thus the true residue of at the point may still be as . Note also that by (10), the (true) residue is equal to [math] for . Combining our observations, the simplest hypothesis would be that does not have a pole at for any value of ! Letting increase from on, it would follow that has a pole at with residue for every . This contradiction to (10.2) would establish Metatheorem 1.1!
One could also argue on the basis of the points . Since would have no poles at those points, assumption (10.2) would require poles of or at with residue . But this would contradict the assumption that the residues are which was reasonable because the terms of contain only as a factor . Thus (10.2) must be incorrect for many values of . The simplest explanation of a residue for would be that the functions have poles at with residue . Indeed, we know that
[TABLE]
We now turn to Metatheorem 1.4. Using Lemma 7.1, the preceding arguments make it plausible that, indeed,
[TABLE]
where is holomorphic for and has good boundary behavior for .
In the case Metatheorem 1.4 suggests the approximation
[TABLE]
Finally, to arrive at Metatheorem 1.5 one would use the formula
[TABLE]
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] P. T. Bateman and R. A. Horn, A heuristic asymptotic formula concerning the distribution of prime numbers . Math. Comp. 16 (1962), 363–367. [ sec 1 , 8 ]
- 2[2] P. T. Bateman and R. A. Horn, Primes represented by irreducible polynomials in one variable . Proc. Sympos. Pure Math., vol. VIII pp 119–132. Amer. Math. Soc., Providence, R.I., 1965. [ sec 1 , 7 , 8 ]
- 3[3] F. J. van de Bult, Counting prime pairs ( p , p 2 ± 2 r ) 𝑝 plus-or-minus superscript 𝑝 2 2 𝑟 (p,\,p^{2}\pm 2r) . In e-mails of December 2007 and January 2008. [ sec 1 , 8 , 9 ]
- 4[4] Fokko van de Bult and Jaap Korevaar, Mean value one of prime-pair constants . Manuscript, Amsterdam, June 2008. See ar Xiv:0806.1667 v 1 [math.NT]. [ sec 9 ]
- 5[5] H. Davenport, Multiplicative number theory . (Third edition, revised by H. L. Montgomery.) Graduate Texts in Math., 74. Springer, New York, 2000. [ sec 1 ]
- 6[6] H. Davenport and A. Schinzel, A note on certain arithmetical constants . Illinois J. Math. 10 (1966), 181–185. [ sec 8 ]
- 7[7] H. M. Edwards, Riemann’s zeta function . Academic Press, New York, 1974. Reprinted by Dover Publications, Mineola, N.Y., 2001. [ sec 1 ]
- 8[8] J. B. Friedlander and D. A. Goldston, Some singular series averages and the distribution of Goldbach numbers in short intervals . Illinois J. Math. 39 (1995), 158–180. [ sec 4 ]
