Flat primes and thin primes

Flat primes and thin primes Kevin A. Broughan and Zhou Qizhi University of Waikato, Hamilton, New Zealand Version: 0th October 2008 E-mail: kab@waikato.ac.nz, qz49@waikato.ac.nz Flat primes and thin primes are primes where the shift by ± has a restricted form, namely a power of 2 or that times a square free number or odd prime respectively. They arise in the study of multi-perfect numbers. Here we show that the flat primes have asymptotic density relative to that of the full set of primes given by twice Artin s constant, that more than 50% of the primes are both lower and upper flat, and that the series of reciprocals of thin primes converges. Key Words: flat prime, thin prime, twin prime, sieve. MSC2000: A4, B05, B83.. INTRODUCTION Some interesting subclasses of primes have been identified and actively considered. These include Mersenne primes, Sophie Germain primes, Fermat primes, Cullen s primes, Wieferich primes, primes of the form n 2 +, of the form n! ±, etc. See for eample [4, Chapter 5] and the references in that tet. For any one of these classes, determining whether or not it is infinite has proved to be a very difficult problem. In this article we eplore two classes of primes, the so-called flat primes and the thin primes. They have simple representations, and we are able to get an idea of their densities relative to the full set of primes. These primes are similar to primes of the form k 2 e + considered by Erdös and Odlyzko, Chen and Sierpiński among others [5, 6, 6]. There the focus is mainly on the admissible values of odd integers k with k, rather on the density of primes themselves having that structure. Erdös showed [5, Theorem ] that the number N( of odd numbers less than or equal to of the form (p + /2 e satisfies c N( c 2,

2 BROUGHAN AND ZHOU where c and c 2 are positive absolute constants. In the opposite direction, a simple modification of the derivation of Sierpinski [6] gives an infinite number of integers n (including an infinite set of primes such that n 2 e is composite for every e =, 2, 3,. Definition.. We say a natural number n is a flat number if n + = 2 e or n + = 2 e q q m where e and the q i are distinct odd primes. If a prime p is a flat number we say p is a (upper flat prime. Let F ( := #{p : p is a flat prime}. Then it is straight forward to show that the density of flat numbers is the same as that of the odd square free numbers, i.e. the number of flat numbers up to is given by 4/π 2 + O( [7]. Definition.2. We say a natural number n is a thin number if n + = 2 e q or n + = 2 e where e and q is an odd prime. If a prime p is a thin number we say p is a thin prime. Let T ( := #{p : n is a thin prime}. For eample, among the first 00 primes, 75 primes are flat and among the first 000 primes, 742 are flat. For thin primes the corresponding numbers are 38 and 23 respectively. The first 0 flat primes are 3, 5, 7,, 3, 9, 23, 29, 3, and 37. The first 0 thin primes are 3, 5, 7,, 3, 9, 23, 3, 37 and 43. If M( is the number of Mersenne primes up to then clearly, for all : M( T ( F ( π(. Figure shows the ratio of F (/π( over a small range. This gives some indication of the strength of Theorem 3. below - in the given range over 70% of all primes are flat. Figure 2 shows the ratio of the number of thin primes up to to the number of twin primes up to. The relationship between thin and twin primes comes from the method of proof of Theorem 4. below and there is no known direct relationship. These types of number arise frequently in the contet of multiperfect numbers, i.e. numbers which satisfy k n = σ(n where σ(n is the sum of the positive divisors of n. For eample, when k = 3 all of the known

FLAT AND THIN PRIMES 3 0.76 0.755 0.75 0.745 2000 4000 6000 8000 0000 0.735 FIG.. The ratio F (/π( for 0 4..8.7.6.5.4.3 2000 4000 6000 8000 0000 FIG. 2. The ratio of thin primes to twin primes up to for 0 4. eamples of so-called 3-perfect numbers are c = 2 3 3 5, c 2 = 2 5 3 7, c 3 = 2 9 3 3, c 4 = 2 8 5 7 9 37 73, c 5 = 2 3 3 43 27, c 6 = 2 4 5 7 9 3 5.

4 BROUGHAN AND ZHOU Each c i is a flat number and each odd prime appearing on the right hand side is thin. The paper is organized as follows: In Section 2 we first show that the asymptotic density of thin numbers up to is the same as that of the primes up to. In Section 3 we show that the density of flat primes up to, relative to the density of all primes, is given by 2A where A is Artin s constant. A corollary to this is that there is a flat prime in every interval [, ( + ɛ]. This is followed by a demonstration that primes which are both lower and upper flat have density and constitute more than half of all primes. In Section 4 we then show that the thin primes are sufficiently sparse that the sum of their reciprocals converges. The final section is a numerical validation of what might be epected for the density of thin primes under the Bateman-Horn conjectures. We use Landau s O,o,and notation. The symbols p, q are restricted to be rational primes. 2. THIN AND FLAT NUMBERS Theorem 2.. The asymptotic density of thin numbers up to is the same as that of the primes up to. Proof. The number of thin numbers up to is given by N( = log 2 n= ( π 2 n + O(. We will show that lim N(/π( =. To this end first consider a single term in the sum. By [5], there is a positive real absolute constant α such that for sufficiently large, + α < π( < α. Therefore, for n N such that n 4 log 2, lb := α + α n log 2 < 2n π( 2 n < π( + α α n log 2 =: ub.

FLAT AND THIN PRIMES 5 Clearly lb and ub tend to as for fied n. The difference between the upper and lower bounds is ub lb = d d = ub lb ( 4α 2αn log 2 log 2 ( α n log 2 = α2 log 2 + n2 log 2 2 log 2 6α. 4α d, where ( + α n log 2 2n log 2 for n in the given range and sufficiently large. Hence log 2 /4 n= π( 2 n π( n log n log n log π( π( 2 n n π( 2 n n log π( 2 2 n π( + o( 6α 2 n + o( + o( = o( log 4 so π( 2 n + n> log as. For the remaining part of the summation range for N(, note that this corresponds to values of and n which satisfy /2 n 3/4, so, using π( /2 and S( := log 2 π n= log 2 /4 ( 2 n 3 4 = O( 3 4 it follows that S(/π( 0 as. Hence N(/π(. 2 n 3. FLAT PRIMES

6 BROUGHAN AND ZHOU Theorem 3.. For all H > 0 F ( = 2 p ( ( Li( + O p(p log H i.e. the relative density of flat primes is 2A = 0.7480 where A is Artin s constant., Proof. We begin following the method of Mirsky [0]. Fi e and let and y satisfy < y < and be sufficiently large. Let H > 0 be the given positive constant. Define F e ( := #{p : p is prime and m square free such that 2 e m = p + }. Then, if µ 2 (m is the characteristic function of the square free numbers, F e ( = p p+=2 e m µ 2 (m = p { = Σ + Σ 2, where Σ := p { µ(a}, and Σ 2 := p { a: a y a 2 b2 e =p+ a>y a 2 b2 e =p+ µ(a}. a:a a 2 b2 e =p+ µ(a} Now using the Bombieri-Vinogradov theorem [4, Section 28] for the number of primes in an arithmetic progression, which is valid with a uniform error bound for the values of e which will be needed: Σ = a y µ(a p:p p mod 2 e a 2 = ( Li( µ(a φ(2 e a 2 + O( log 2H+ a y ( µ(a ( = φ(2 e a 2 Li( + O a a>y ( φ(2 e a 2 + O y log 2H+.

FLAT AND THIN PRIMES 7 Note that the function g(n := 2 e φ(2 e n 2 is multiplicative, so the coefficient of Li( may be rewritten 2 e a 2 e µ(a φ(2 e a 2 = ( 2 e 2e φ(2 e p 2 p = 3 ( 2 e 4 p 2 p p odd = 3A 2 e. Now consider the sum in the first error term: Therefore a>y For the second sum: and therefore φ(2 e a 2 a>y 2 e φ(a 2 2 e ( O φ(2 e a 2 = O a>y Σ 2 p<{ a>y p+=2 e a 2 b a>y ( log log y 2 e y log log a a 2.. } ( = O 2 e, y a>y 2 e a 2 b F e ( = 3A ( log log y ( ( 2 e Li( + O 2 e + O y 2 e + O y If we choose y = log H, then Now let F e ( = 3A 2 e Li( + O(. log H+ y log 2H+ D e ( := #{p : p is prime, p + = 2 e m, with m square free and odd}. By, [0, Theorem 2], D ( = A Li(+O (. Considering the even H+ and odd cases, for all e, we have F e ( = D e ( + D e+ ( so and therefore F ( + F 2 ( + = D ( + 2(D 2 ( + D 3 ( +.

8 BROUGHAN AND ZHOU Then F ( = log 2 e= and this completes the proof. D e ( + O( = 2 (D ( + F ( + F 2 ( + + O( = A 2 ( + 3 2 + 3 2 2 + Li( + O( log H+ ( = 2ALi( + O log H Note that if we call primes with the shape p = 2 e p p m + lower flat, their density is the same as that of the flat primes, so more than 20% of all primes are both flat and lower flat. However, this figure very significantly underestimates the proportion of such primes - see Theorem 3.2 and its corollary below. By analogy, flat primes are also called upper flat primes. Corollary 3.. For all ɛ > 0 and ɛ there eists a flat prime in the interval [, ( + ɛ]. Note also that it would be possible to adapt the method of Adleman, Pomerance and Rumley [, Proposition 9] to count flat primes in arithmetic progressions. Theorem 3.2. Let the constant H > 0 and the real variable be sufficiently large. Let the set of primes which are both lower and upper flat which are less than be given by B( = {p : e, f and odd square free u, v so p = 2 e v, Then p + = 2 f u}. where the constant B( = A 2 Li( + O( log H A 2 = p odd ( 2 = 0.53538. p 2 p

FLAT AND THIN PRIMES 9 Proof. Let e, f and define the sets: L e := {p : odd square free v so p = 2 e v} U f := {p : odd square free u so p + = 2 f u}. Then L U = and L e U f = for all e 2, f 2 so we can write B( = { f 2 L U f } { e 2 U L e } where all of the unions are disjoint. Now fi e 2. We will first estimate the size of U L e,where U L e = {p : odd square free u, v so p + = 2u, p = 2 e v}. Then #U L e = p { = p { = p { = p τ (d{ p+=2u, p =2 e v, u, v odd and square free a,b odd,(a,b=, p mod a 2, p mod b 2, p +2 e mod 2 e+ a,b odd,(a,b=, p u mod 2 e+ a 2 b 2 d odd, p u mod 2 e+ d 2 } µ(aµ(b} µ(aµ(b} µ(d} where u, the residue obtained through an application of the Chinese Remainder Algorithm, is dependent on d and e, and τ (d is the number of unitary divisors of d, a multiplicative function with τ (p = 2. We then split and reverse the sum in a similar manner as in the proof of Theorem 3. to arrive at #U L e = ( d,d odd = 2 e p odd τ (d ( Li( + O φ(2 e+ d 2 log H+ ( 2 ( Li( + O p 2 p log H+

0 BROUGHAN AND ZHOU Summing over e 2 and, noticing that the sizes for each corresponding L U e are the same, we obtain the stated value of B(. Figure 3 compares the number of primes up to 80, 000 with the number of primes up to 80, 000 which are both lower and upper flat. 0.542 0.538 20000 40000 60000 80000 0.536 0.534 0.532 FIG. 3. The ratio B(/π( for 8 0 4. Corollary 3.2. It follows from Theorems 3. and 3.2 that the set of rational primes may be divided into 4 disjoint classes: those both lower and upper flat - about 54%, those either lower or upper flat but not both - each about 2%, and those neither upper nor lower flat - 4%. 4. THIN PRIMES In the paper [9, Theorem 3] a proof is set out for a result given below on the number of primes up to giving a lower bound for the number primes with fied consecutive values of the number of distinct prime divisors of shifts of the primes by a, with the parameter a having the eplicit value 2. It is remarked that a similar proof will work for all integer (non-zero a. Here is the statement taken from Mathematical Reviews (although the lower bound for m is not given: Let a be a non-zero integer and (for m define P(m,, ω := #{p : p, ω(p + a = m}.

FLAT AND THIN PRIMES Then there eist positive absolute constants b and c such that as (log m P(m,, ω + P(m +,, ω c (m! log 2 holds for m b log. If we use the result in case a =, we are able to show the number of thin primes is infinite. To see this let a =, m = and be sufficiently large. Then T ( + M( = P(,, ω + P(2,, ω where = #{p : p + = 2 e or p + = 2 e q f, e, f, or p = 2} Then M( := #{p : p + = 2 e q f, e, f 2}. M( log log e= f=2 log e= π(( 2 e f + O( π( 2 e log 2 π( Therefore, by the quoted result above, the number of thin primes less than or equal to is bounded below by a constant times / log 2, so must be infinite. However there are parts of the proof of [9, Theorem 3] that do not appear to work, even for the given case a = 2, and, in addition, the implied lower bound should be m 2. Apparently the best available safe result, using the method of Chen, appears to be that of Heath-Brown [8, Lemma ] from which we can easily show that the number of primes H( such that p and either p + = 2p or p + = 2p p 2, with the p i being odd primes, is infinite, indeed H( / log 2. Based on this evidence, the Bateman-Horn conjecture set out in Section 5 below, and numerical evidence, we are led to the conjecture: Conjecture: The number of thin primes up to satisfies T ( log 2. The order of difficulty of this conjecture appears to be similar to that there are an infinite number of twin primes or Sophie Germain primes. As usual upper bounds are much easier to obtain.

2 BROUGHAN AND ZHOU Theorem 4.. As T ( log 2. Proof. First let e be fied and apply the sieve of Brun in the same manner as for the classical twin primes problem (for eample [8, Theorem 4] or [3, Theorem 3.] to count J e ( := #{p : 2 e p is prime}, noting that if A = {m(2 e m : m } and ρ(d is the number of solutions modulo d which satisfy m(2 e m 0 mod d, then ρ is a multiplicative function, ρ(2 = and ρ(p = 2 for odd primes p leading to the same bound as in the twin primes problem, namely J e ( log 2. Now we use the fact, proved using induction for m 4, that, for all m, m n= 2 n n 2 < 5 2m m 2 (2. (For sufficiently large m, the 5 can be replaced by 2 + ɛ but we don t need this.

FLAT AND THIN PRIMES 3 Finally, let be large and choose m N so 2 m < 2 m+. Then T ( = log 2 e= log 2 e= log 2 e=0 ( Je ( 2 e + O( 2 e log 2 + O( 2 e 2 e log 2 + O( 2 e = = log 2m+ log 2 e=0 log 2 2 log 2 2 m e=0 m+ 2 m+ 2 e log 2 2m+ 2 e + O( 2 m e+ + O( (m + e 2 n= 2 m+ 2 n + O( n2 < 5 log 2 + O( by (2 2 (m + 2 2 log 2 2 < 0 log 2 2 log 2 + O( log 2, completing the proof of the theorem. So the asymptotic bound is the same as that for twin primes. In the same manner as originally derived by Brun for the sum of reciprocals of the twin primes (for eample [2, Theorem 6.2] we obtain: Corollary 4.. The sum of the reciprocals of the thin primes is finite. Proof. If p n is the n th thin prime then, by Theorem 4., p n n = T (p n log 2 p n p n (log n 2 so n log 2 n. p n

4 BROUGHAN AND ZHOU 5. HARDY-LITTLEWOOD-BATEMAN-HORN CONJECTURES The well known Hardy-Littlewood-Bateman-Horn conjectures [7, 3] give an asymptotic formula for the number of simultaneous prime values of sets of polynomials in Z[], with some restrictions on the polynomials. In the case of twin primes the polynomials are f o ( =, f ( = + 2 and if then the formula predicted is π 2 ( := #{p : p + 2 is prime} π 2 ( 2C 2 2 du log 2 u where C 2 is the so-called twin prime constant [3] defined by C 2 := p>2 ( (p 2. In the case of thin primes the conjectures only apply to forms with fied e with polynomials f 0 ( =, f e ( = 2 e. If Then the formulas predict T e ( := #{p : p + = 2 e q} T e ( 2C 2 2 e 2 du log 2 u. The factor /2 e occurs simply because p + if and only if q /2 e. Hence / T ( π 2 ( e= T e (. π 2 ( To test this numerically we evaluated the ratio of the number of thin primes up to to the number of twin primes up to for up to 4 0 6 in steps of 0 5 and obtained the following values: {.,.20343,.6852,.734,.6036,.5882,.489,.447,.4499,.428,.355,.2896,.2543,.234,.75,

FLAT AND THIN PRIMES 5.84,.729,.438,.68,.099,.69,.06,.25,.095,.22,.37,.34,.25,.8,.306,.79,.05,.0986,.096,.0876,.0924,.092,.0676,.0623,.0536}. demonstrating some convergence towards the predicted value. If the relationship between the thin and twin primes could be made eplicit this would assist in a proof of the twin primes conjecture. 6. ACKNOWLEDGEMENTS We are grateful to Igor Shparlinski for making as aware of the theorem of Heath-Brown. The computations and graphics were produced using Mathematica. REFERENCES. Adleman, L.M., Pomerance, C. and Rumley, R.S. On distinguishing prime numbers from composite numbers, Ann. of Math. (2 7 (983, p73-206. 2. Baker, R.C. and Pintz, J. The distribution of squarefree numbers, Acta Arithmetica 46 (985, p7-79. 3. Bateman, P. T. and Diamond, H. Analytic Number Theory: An introductory course, World Scientific, 2004. 4. Davenport, H. Multiplicative Number Theory, 3rd Edition, Springer, 2000. 5. Erdös, P. and Odlyzko, A.M. On the density of odd integers of the form (p 2 n and related questions, Journal of Number Theory, (979 p257-263. 6. Chen, Y.G. On integers of the form k 2 n +, Proceedings of the American Mathematical Society, 29 (2000, p355-36. 7. Hardy, G.H. and Littlewood, J. E., Some problems of Partitio numerorum ; III: on the epression of a number as a sum of primes, Acta Math. 44 (923, p-70. 8. Heath-Brown, D.R., Artin s conjecture for primitive roots, Quart. J. Math. Oford Ser. (2 37 (986, p27-38. 9. Hooley, C. Application of sieve methods to the theory of numbers, Cambridge, 976. 0. Mirsky, L. The number of representations of an integer as the sum of a prime and a k-free integer, American Math. Monthly 56 (949, p7-9.. Montgomery, H.L. and Vaughan, R.C. On the large sieve, Mathematica, 20 (973, p9-34. 2. Nathanson, M.B. Additive Number Theory: The classical bases, Springer, 996. 3. Nicely, T.R. http://www.trnicely.net/twins/twins.html. 4. Ribenboim, P. The new book of prime number records, Springer, 996. 5. Rosser, J.B. and Schoenfeld, L. Approimate formulas for some functions of prime numbers, Illinois J. Math. 6 (962, p64-94. 6. Sierpiński, W. Sur un problém concernant les nombres k 2 n +, Elem. Math. 5 (960, p73-74, Corrigendum 7 (962, p85.

6 BROUGHAN AND ZHOU 7. Suryanarayana, The number of k-free integers which are coprime to n, Indian J. Math. (969, p3-39. 8. Tenenbaum, G. Introduction to analytic and probabilistic number theory, Cambridge, 995. 9. Timofeev, N. M. The Hardy-Ramanujan and Halasz inequalities for shifted primes, Mathematical Notes, 57 (995, p522-535.