Average Orders of Certain Arithmetical Functions

Transcription

1 Average Orders of Certain Arithmetical Functions Kaneenika Sinha July 26, 2006 Department of Mathematics and Statistics, Queen s University, Kingston, Ontario, Canada K7L 3N6, skaneen@mast.queensu.ca ABSTRACT. We consider the functions Mn), the maximum exponent of any prime power dividing n and mn), the minimum exponent of any prime power dividing n. The sums Mn) and mn) have been well investigated in the literature. In this note, we will improve known estimates of both the above sums under the assumption of the Riemann hypothesis. We will also obtain Ω-type estimates for these sums unconditionally. Mathematics Subject Classification 2000): Primary A5, Secondary A25. Introduction A natural number n 2 can be written uniquely as a product of prime numbers, We define and n = p α pα 2 2 pα k k, α i. Mn) = max{α, α 2,, α k } mn) = min{α, α 2,, α k }. We also define M) = m) =. The sums Mn) and mn) have been well investigated in the literature. In 969, Ivan Niven [0] proved that Mn) Bx, as x ) where B = + ). ζk) B can be evaluated to be.7052 approximately. He also showed that mn) = x + ζ ) 3 2 ζ3) x 2 + ox 2 ). 2) The aim of this note is to prove the following theorems: Research partially supported by a Queen s graduate fellowship.

2 Theorem Assuming the Riemann hypothesis, Mn) = Bx + Ox ɛ ). 3) Moreover, unconditionally, we have Mn) = Bx + Ωx 4 ). 4) Theorem 2 Assuming the Riemann hypothesis, we have where and γ i,k = 0 j k j i Also, unconditionally, we have mn) = x + p 7 i=2 ) k + j ζ + k + i Ai)x i + Ox ɛ ), 5) 2k m=k+ p m k+i 3k m=2k+2 A2) = γ 0,2, A3) = γ,2 + γ 0,3, A4) = γ,3 + γ 0,4, A5) = γ 2,3 + γ,4 + γ 0,5, A6) = γ 2,4 + γ,5 + γ 0,6, A7) = γ 3,4 + γ 2,5 + γ,6 + γ 0,7. mn) = x + 7 i=2 p m k+i Ai)x /i + Ωx 0 ). 6) Before proving these theorems, we would like to briefly review what is known about these sums. Perhaps unaware of Niven s work, in 975, C. W. Anderson proposed the following problem in the American Mathematical Monthly: Problem: Prove that lim x x where µm) is the Möbius function. Mn) = + m=2 µm) mm ), In response, T. Salat pointed out that the above sum had already been investigated by Niven in his 969 paper. A solution to Anderson s problem was also submitted by Ram Murty and Kumar Murty. In fact, they reported a mistake in the question and showed that the correct term on the right hand side should be µm) mm ), m=2 but this was not mentioned by the editor when responses to Anderson s problem were referenced in a later issue of the Monthly see [].) It is not difficult to see that this correct term is indeed equal to Niven s constant B. Since ) ζk) < for k 2, 2k 2 ),

3 the sum defining B in ) is absolutely convergent. Thus, by interchanging summations, we see that ) µm) µm) = ζk) m k = mm ). m=2 Niven s result was made effective by D. Suryanarayana and R. Sitaramachandrarao in 977. In [2], they proved that Mn) = Bx + x 2 exp c log 3/5 xlog log x) 5 ), 7) where c is an absolute positive constant. The main ingredient in their proof is a result of Walfisz proved in [4]), which we will state in the next section. They also prove that mn) = x + γ 0,2 x 2 + γ,2 + γ 0,3 )x 3 + γ,3 + γ 0,4 )x 4 m=2 + γ 2,3 + γ,4 + γ 0,5 )x 5 + Ox 6 ), 8) where γ i,k s are as defined in Theorem 2. Unaware of the results of [2], in 99, Hui Zhong Cao obtained an error estimate of Ox 2 log x) in [4]. This estimate is weaker than that obtained in [2]. Cao also provides an estimate for mn), but this is also not as sharp as equation 8). In the next two sections, we will estimate Mn). We will also refine some known estimates about mn) in the last section. The two main theorems proved in this note are Theorems and 2. 2 Preliminary results In this section, we will mention some results which are needed in this paper. For a fixed positive integer k 2, we define a k-free integer as a positive integer which is not divisible by the k-th power of any of its prime factors. Let S k x) denote the number of k-free integers less than or equal to x. It turns out that Mn) is directly related to the distribution of k-free numbers. By elementary number theory we know that for an integer k 2, S k x) = The direct connection between the error terms x ζk) + Ox k ). E k x) = S k x) x ζk) for k 2 and the error term in the sum Mn) is indicated in the following proposition, which is also mentioned in [0] and [2]: Proposition 3 where Mn) = Bx j = j E k x) + O), log x. log 2 3

4 Proof. We begin by observing that the maximum of {M), M2), Mn)} is log x log 2. Also, for k 2, the number of integers less than or equal to x such that Mn) = k is S k x) S k x). Thus, if we get that j = log x, log 2 Mn) = k )S k x) S k x)). j+ Since S j+ x) = x, the above sum is equal to jx + j k )S k x) which, on further simplification, is equal to Writing the above expression becomes This is equal to Since we get that Thus, k>j x + x jx S k x) = j Bx x k>j j ks k x), k= j S k x). x ζk) + Ox k ), ) ζk) ) ζk) ) < ζk) k>j j E k x). j E k x). ζk) < 2 k, Mn) = Bx 2 k = 2 j = O j E k x) + O). ). x While the trivial estimate for E k x) is Ox k ), substantially better estimates can be found, especially under the assumption of the Riemann hypothesis. We refer the reader to an excellent survey of Pappalardi [] on various improvements of this error term. In particular, the following results are relevant to this note: 4

5 In 963, Walfisz [4] proved that E k x) = x k exp ck 8/5 log 3/5 xlog log x) /5 ). 9) We advise the reader that Walfisz s theorem has not been recorded correctly in []. The exponent /5 in the third displayed formula on page 73 should be replaced by /5. The authors, in [2], prove 7) by applying Walfisz s result 9) for k = 2 and the trivial estimate Ox k ) for k 3 to the right hand side of Proposition 3. Clearly, any improvement in the estimates for E k x) will give a better error term for Mn). We now state two better estimates for E kx) which have been obtained by assuming the Riemann hypothesis. In 993, Jia [8] proved the following result: Proposition 4 Assuming the Riemann hypothesis, E 2 x) = Ox 7/54+ɛ ). The next result was proved in 98 by Montgomery and Vaughan [9]: Proposition 5 Assuming the Riemann hypothesis, for k 2, E k x) = Ox k+ +ɛ ) for any ɛ > 0. In 989, Graham and Pintz [5] obtained the following better estimate for E 3 x): Proposition 6 Assuming the Riemann hypothesis, 3 Proof of Theorem We recall, from Proposition 3, we get that By Proposition 4, E 3 x) = Ox 7/30 ). Mn) = Bx Moreover, by Proposition 5, for 3 k j, Thus, we get j E k x) + O). E 2 x) = Ox 7/54+ɛ ). E k x) = Ox 4 +ɛ ). Mn) = Bx + Ox 7/54+ɛ ). This proves equation 3), the first part of Theorem. In order to obtain equation 4), it suffices to show that E 2 x) + E 3 x) = Ωx 4 ). 0) This is because equation 9) gives us an unconditional estimate for E 4 x), which is sharper than Ox /4 ). Substituting the trivial estimate Ox /k ) for k 5, equation 9) for k = 4 and equation 0) in Proposition 3, we obtain equation 4). Equation 0) follows as a corollary to the following general result, which seems not to have been noticed before and is interesting in its own right: 5

6 Proposition 7 Let Sx) be a complex-valued function. Suppose that Sx) fs) = s dx xs+ has a pole at a complex number with real part θ. Then Sx) = Ωx θ ). Proof. The main idea of the proof of Proposition 7 is inspired by a technique followed by Vaidya in [3] to find out an Ω-type estimate for E k x) conditional upon either the Riemann hypothesis or the existence of a simple zero of ζks) on the line Res) = k/2 which is not a zero of ζs). Let us assume that Sx) = ox θ ). Then, for every ɛ > 0, we can find x 0 = x 0 ɛ) such that for x x 0, we have Sx) < ɛx θ. Let Sx) A for 0 < x < x 0. Thus, for s = σ + it with σ > θ, we have fs) < s A x θ dx + s ɛ = s A θ + s ɛ σ θ. x σ +θ dx Now, let us first consider the case when fs) has a simple pole at ρ = θ + it 0 with non-zero residue c. Then, as σ θ, t = t 0 ), we get that lim σ θ)fσ + it 0) = c. σ θ Thus, by the above inequality, we get that for every ɛ > 0, c = lim σ θ)fσ + it 0 ) σ θ lim σ θ = ɛ σ + it 0. { σ + it 0 σ θ) A θ + σ + it 0 ɛ This is not possible as c is non-zero. Thus, our assumption that Sx) = ox θ ) is false. Therefore, Sx) = Ωx θ ). The case when fs) has a pole of order 2 can also be dealt with by a similar analysis. This proves Proposition 7. In particular, if we let then it is not difficult to see that s Sx) = E 2 x) + E 3 x), Sx) ζs) dx = xs+ ζ2s) + ζs) ζ3s) ζs) ζ2) ζs) ζ3). ) Let ϱ = 4 + iγ be a pole of the right hand side of equation ). We prove the existence of such a pole by applying some zero-density arguments. A classical result of Selberg tells us that for T } 6

7 sufficiently large, the number of zeros of ζ2s) on the line segment joining 4 to 4 + it is at least AT log T for some A > 0. Such zeroes are the singularities of ζs) ζ2s) + ζs) ζ3s) ζs) ζ2) ζs) ζ3) unless possibly they are also zeros of ζs) or ζ3s). Since the real part of ϱ is /4, by the functional equation, we get a zero of ζs) where the real part of s is equal to 3/4. However, by a theorem of Carlson, see [6], Theorem 0.) we know that the number of zeros of ζs) up to height T with real part 3/4 is OT 3/4 log T ). Therefore, not every zero of ζ2s) is a zero of ζs) or ζ3s). Thus, the function s E 2 x) + E 3 x) x s+ dx will have at least one pole with real part /4. Thus, by applying Proposition 7 to the function Sx) = E 2 x) + E 3 x), we deduce equation 0). This concludes the proof of Theorem. 4 Proof of Theorem 2 In this section, we refine known estimates of the sum mn). We recall that equation 8) has been proved unconditionally in [2]. Once again, this estimate can be improved by assuming the Riemann hypothesis. Analogous to a k-free integer, a k-full integer is defined as a positive integer which is divisible by the k-th power of all its prime factors. We observe that analogous to Proposition 3), if L k x) denotes the number of k-full integers less than or equal to x, then mn) = x + L 2 x) + L 7 x) + Ox 8 ), 2) since j = Olog x) and k 8 L kx) = Ox /8 ) + Ox /9 log x). We now state some known results about the distribution of k-full integers. The following result has been proved in [7]: Proposition 8 For every k, L k x) = γ 0,k x /k + + γ k,k x /2k ) + k x), where the γ i,k s are as given after equation 8). Moreover, 3 x) = Ox ), 4 x) = Ox ), 5 x) = Ox 0 ) and 6 x) = Ox 2 ). The following conditional result proved in [5] in 200 gives an estimate for 2 x) : Proposition 9 Assuming the Riemann hypothesis, 2 x) = Ox ɛ ) for any ɛ > 0. We also recall the following unconditional Ω-type result of Balasubramanian, Ramachandra and Subbarao proved in [3]: Proposition 0 2 x) = Ωx /0 ). 7

8 We are now ready to prove Theorem 2: We had observed above in equation 2) that mn) = x + L 2 x) + + L 7 x) + Ox 8 ). Thus, by using Proposition 9 for k = 2, Proposition 8 for 3 k 6, and the elementary estimate for k = 7, we get that mn) = x + k x) = Ox k+ ) 7 i=2 Ai)x i + Ox ɛ ), where the Ai) s are as given above. We also observe that by applying equation 8), Proposition 8 and Proposition 0 to the right hand side of equation 2), we get that This proves Theorem 2. mn) = x + 5 Concluding remarks 7 Ai)x /i + Ωx 0 ). i=2 Thus, in this paper, we saw how closely the Riemann hypothesis is linked to the averages of exponents in factoring integers. The key ingredients in the proofs of Theorems and 2, as we saw in the last two sections, are the refined estimates of E k x) for k 2 obtained by assuming this ubiquitous conjecture of Riemann. We should note that these estimates were not available to Suryanarayana and Sitaramachandrarao when they wrote their paper [2] in 977. One would also like to know if the estimates for k x) for k 3 can be further improved under the assumption of the Riemann hypothesis or even some quasi-riemann hypothesis. Could one also obtain optimal error terms for the sums Mn) and mn)? We relegate these questions to future research. Acknowledgments. I would like to thank Professors E. Bertram and R. Murty for bringing this problem to my attention. I would also like to thank the referee for many helpful suggestions on a previous version of this paper. References [] C. W. Anderson, K. Murty, R. Murty and T. Salat, Solutions of advanced problems: Problem 605, The American Mathematical Monthly, Vol ), no. 0, pp [2] R. Balasubramanian and K. Ramachandra, On square-free numbers, Proceedings of the Ramanujan Centennial International Conference Annamalainagar, 987), RMS Publications, Ramanujan Math. Soc., Annamalainagar, 988, pp

9 [3] R. Balasubramanian, K. Ramachandra and M. V. Subbarao, On the error function in the asymptotic formula for the counting function of k-full numbers, Acta Arithmetica, Vol ), no. 2, pp [4] Hui Zhong Cao, The asymptotic formulas related to exponents in factoring integers, Math. Balkanica N.S.), Vol. 5 99), no. 2, pp [5] S. W. Graham and J. Pintz, The distribution of r-free numbers, Acta Math. Hungar., Vol ), no. -2, pp [6] H. Iwaniec and E. Kowalski, Analytic Number Theory, Vol ), AMS Colloquium Publications, American Math. Soc., Providence, Rhode Island. [7] A. Ivić and P. Shiu, The distribution of powerful integers, Illinois J. Math., Vol ), no. 4, pp [8] C. H. Jia, The distribution of square-free numbers, Sci. China Ser. A, 36, 993, No. 2, pp [9] H. L. Montgomery and R. C. Vaughan, The distribution of square-free numbers, Recent progress in analytic number theory, Vol., Durham 979), Academic Press, London-New York, 98, pp [0] I. Niven, Averages of exponents in factoring integers, Proceedings of the American Mathematical Society, Vol. 22, No. 2 August 969), pp [] F. Pappalardi, A survey on k-freeness, Ramanujan Mathematical Society Lecture Notes Series, No., edited by S. D. Adhikari, R. Balasubramaniam, K. Srinivas, 2004, pp [2] D. Suryanarayana and R. Sitaramachandrarao, On the maximum and minimum exponents in factoring integers, Arch. Math. Basel), Vol ), no. 3, pp [3] A. M. Vaidya, On the changes of sign of a certain error function connected with k-free integers, J. Indian Math. Soc, Vol. 32, 968, pp [4] A. Walfisz, Weylsche Exponentialsummen in der neueren Zahlentheorie, Mathematische Forschungsberichte, XV. VEB Deutscher Verlag der Wissenschaften, Berlin 963. [5] J. Wu, On the distribution of square-full integers, Arch. Math. Basel), Vol ), no. 3, pp