Abstract
This paper discusses the additive prime divisor function \(A(n) := \sum \limits _{p^\alpha || n} \alpha \, p\) which was introduced by Alladi and Erdős in 1977. It is shown that A(n) is uniformly distributed (mod q) for any fixed integer \(q > 1\) with an explicit bound for the error.
This paper is dedicated to Krishna Alladi on the occasion of his 60th birthday
This research was partially supported by NSA grant H98230-15-1-0035.
Access provided by CONRICYT-eBooks. Download conference paper PDF
Similar content being viewed by others
Keywords
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
1 Introduction
Let \(n = \prod \limits _{i=1}^r p_i^{a_i}\) be the unique prime decomposition of a positive integer n. In 1977, Alladi and Erdős [1] introduced the additive function
Among several other things they proved that A(n) is uniformly distributed modulo 2. This was obtained from the identity
together with the known zero-free region for the Riemann zeta function. As a consequence they proved that there exists a constant \(c > 0\) such that
for \(x \rightarrow \infty .\)
In 1969 Delange [3] gave a necessary and sufficient condition for uniform distribution in progressions for integral valued additive functions which easily implies that A(n) is uniformly distributed (mod q) for all \(q \ge 2\) (although without a bound for the error in the asymptotic formula). The main goal of this paper is to show that A(n) is uniformly distributed modulo q for any integer \(q \ge 2\) with an explicit bound for the error.
Unfortunately, it is not possible to obtain such a simple identity as in (1) for the Dirichlet series
when \(q > 2\) and h, q are coprime. Instead we require a representation involving a product of rational powers of Dirichlet L-functions which will have branch points at the zeros of the L-functions.
The uniform distribution of A(n) is a consequence of the following theorem (1.1) which is proved in §3. To state the theorem we require some standard notation. Let \(\mu \) denote the Mobius function and let \(\phi \) denote Euler’s function. For any Dirichlet character \(\chi \pmod {q}\) (with \(q > 1\)) let \(\tau (\chi ) = \sum \limits _{\ell \pmod {q}} \chi (\ell ) e^{\frac{2\pi i\ell }{q}}\) denote the associated Gauss sum and let \(L(s, \chi )\) denote the Dirichlet L-function associated to \(\chi .\)
Theorem 1.1.
Let h, q be fixed coprime integers with \(q > 2.\) Then for \(x \rightarrow \infty \) we have the asymptotic formula
where \(c_0>0\) is a constant depending at most on h, q,
and
Theorem 1.1 has the following easily proved corollary.
Corollary 1.2.
Let \(q >1\) and let h be an arbitrary integer. Then
The above corollary can then be used to obtain the desired uniform distribution theorem.
Theorem 1.3.
Let h, q be fixed integers with \(q > 2.\) Then for \(x\rightarrow \infty \), we have
We remark that the error term in theorem 1.3 can be replaced by a second order asymptotic term which is not uniformly distributed (mod q).
The proof of theorem (1.1) relies on explicitly constructing an L-function with coefficients of the form \(e^{2\pi i \frac{h A(n)}{q}}\). It will turn out that this L-function will be a product of Dirichlet L-functions raised to complex powers. The techniques for obtaining asymptotic formulae and dealing with branch singularities arising from complex powers of ordinary L-series were first introduced by Selberg [7], and see also Tenenbaum [8] for a very nice exposition with different applications. In [4,5,6], one finds a larger class of additive functions where these methods can also be applied yielding similar results but with different constants.
2 On the function \(L(s, \psi _{h/q})\)
Let h, q be coprime integers with \(q > 1\). In this paper we shall investigate the completely multiplicative function
Then the L-function associated to \(\psi _{h/q}\) is defined by the absolutely convergent series
in the region \(\mathfrak {R}(s) > 1,\) and has an Euler product representation (product over rational primes) of the form
The Euler product (3) converges absolutely to a non-vanishing function for \(\mathfrak {R}(s) > 1.\) We would like to show it has analytic continuation to a larger region.
Lemma 2.1.
Let \(\mathfrak {R}(s) > 1.\) Then
where, for any \(\varepsilon > 0\), the function
is holomorphic for \(\mathfrak {R}(s) > \frac{1}{2} + \varepsilon \) and satisfies \(|T_{h,q}(s)| = \mathscr {O}_\varepsilon \left( 1\right) \) where the \(\mathscr {O}_\varepsilon \)-constant is independent of q and depends at most on \(\varepsilon \).
Proof.
Taking log’s, we obtain
Hence, we may take
which is easily seen to converge absolutely for \(\mathfrak {R}(s) > \frac{1}{2}.\) \(\square \)
For \(q > 2,\) let \(\chi \) denote a Dirichlet character \(\pmod {q}\) with associated Gauss sum \(\tau (\chi ).\) We also let \(\chi _0\) be the trivial character \(\pmod {q}.\)
We require the following lemma.
Lemma 2.2.
Let \(h, q\in \mathbf Z\) with \(q > 2\) and \((h,q) = 1.\) Then
Proof.
Since \((h,q) = 1,\) it follows that for \(\chi \pmod {q}\) with \(\chi \ne \chi _0,\)
This implies that
The proof is completed upon noting that the Ramanujan sum on the right side above can be evaluated as
\(\square \)
Theorem 2.3.
Let \(s\in \mathbf C\) with \(\mathfrak {R}(s)>1.\) Then we have the representation
where
Proof.
If we combine lemmas (2.1) and (2.2) it follows that for \(\mathfrak {R}(s) > 1\),
Hence
The theorem immediately follows after taking exponentials. \(\square \)
The representation of \(L(s, \psi _{h/q})\) given in theorem 2.3 allows one to analytically continue the function \(L(s, \psi _{h/q})\) to a larger region which lies to the left of the line \(\mathfrak {R}(s) = 1 +\varepsilon \) (\(\varepsilon > 0\)). This is a region which does not include the branch points of \(L(s, \psi _{h/q})\) at the zeros and poles of \(L(s,\chi ), \zeta (s)\).
Assume that \(q > 1\) and \(\chi \pmod {q}\). It is well known (see [2]) that the Dirichlet L-functions \(L(\sigma +{ it},\chi )\)) do not vanish in the region
unless \(\chi \) is the exceptional real character which has a simple real zero (Siegel zero) near \(s = 1.\)
Similarly, \(\zeta (\sigma +{ it})\) does not vanish for
Assume \(q > 1\) and that there is no exceptional real character (mod q). It follows from (4) and (5) that \(L(s, \psi _{h/q})\) is holomorphic in the region to the right of the contour \(\mathscr {C}_q\) displayed in Figure 1.
To construct the contour \(\mathscr {C}_q\) first take a slit along the real axis from \(1 - \frac{c_2}{\log q}\) to 1 and construct a line just above and just below the slit. Then take two asymptotes to the line \(\mathfrak {R}(s) = 1\) with the property that if \(\sigma + { it}\) is on the asymptote and \(|t| \ge 1\), then \(\sigma \) satisfies (4). If \(q =1\), we do a similar construction using (5).
3 Proof of theorem 1.1
The proof of theorem 1.1 is based on the following theorem.
Theorem 3.1.
Let h, q be fixed coprime integers with \(q > 2\) and \(\mu (q) \ne 0.\) Then for \(x\rightarrow \infty \) there exist absolute constants \(c,c' > 0\) such that
On the other hand if \(\mu (q) = 0\), then \( \sum \limits _{n\le x} e^{2\pi i \frac{h A(n)}{q}} = \mathscr {O}\left( x e^{-c'\,\sqrt{\log x}} \right) .\)
Proof.
The proof of theorem 3.1 relies on the following lemma taken from [2].
Lemma 3.2.
Let
then for \(x, T > 0\), we have
It follows from lemma 3.2, for \(x, T \gg 1\) and \(c = 1 + \frac{1}{\log x}\), that
Fix large constants \(c_1, c_2 > 0.\) Next, shift the integral in (6) to the left and deform the line of integration to a contour
as in figure 2 below which contains two short horizontal lines:
together with the contour \(C_{T,x}\) which is similar to \(C_q\) except that the two curves asymptotic to the line \(\mathfrak {R}(s) = 1\) go from \(1 -\frac{c_1}{\sqrt{\log qT}} + iT\) to \(1 -\frac{c_2}{\sqrt{\log x}} +i\varepsilon \) and \(1 -\frac{c_2}{\sqrt{\log x}} - i\varepsilon \) to \(1 -\frac{c_1}{\sqrt{\log qT}} - iT\), respectively, for \(0 <\varepsilon \rightarrow 0. \)
Now, by the zero-free regions (4), (5), the region to the right of the contour \(L^+ + \mathscr {C}_{T,x} + L^-\) does not contain any branch points or poles of the L-functions \(L(s, \chi )\) for any \(\chi \pmod {q}\). It follows that
The main contribution for the integral along \(L^+ + \mathscr {C}_{T,x} + L^-\) in (7) comes from the integrals along the straight lines above and below the slit on the real axis\(\Big [1-\frac{c_2}{\sqrt{\log x}}, \;1\Big ].\) These integrals cancel if the function \(L\big (s, \psi _{h/q}\big )\) has no branch points or poles on the slit. It follows from theorem 2.3 that this will be the case if \(\mu (q) = 0\). The remaining integrals in (7) can then be estimated as in the proof of the prime number theorem for arithmetic progressions (see [2]), yielding an error term of the form \( \mathscr {O}\left( x e^{-c'\,\sqrt{\log x}} \right) \). This proves the second part of theorem 3.1.
Next, assume \(\mu (q) \ne 0.\) In this case \(L(s,\psi _{h/q})\) has a branch point at \(s=1\) coming from the Riemann zeta function, it is necessary to keep track of the change in argument. Let \(0^+i\) denote the upper part of the slit and let \(0^- i\) denote the lower part of the slit. Then we have \(\log [\zeta (\sigma +0^+ i) = \log |\zeta (\sigma )| - i\pi \) and \(\log [\zeta (\sigma +0^- i) = \log |\zeta (\sigma )| + i\pi \).
By the standard proof of the prime number theorem for arithmetic progressions it follows that (with an error \(\mathscr {O}\big (e^{-c'\sqrt{\log x}} \big )\)) the right hand side of (7) is asymptotic to
We may evaluate \(\mathscr {I}_{\text {slit}}\) using theorem 2.3. This gives
As in the previous case when \(\mu (q) = 0,\) the remaining integrals in (7) can then be estimated as in the proof of the prime number theorem for arithmetic progressions, yielding an error term of the form \( \mathscr {O}\left( x e^{-c'\,\sqrt{\log x}} \right) \). This completes the proof of theorem 3.1. \(\square \)
The proof of theorem 1.1 follows from theorem 3.1 if we can obtain an asymptotic formula for the integral
Since we have assumed q is fixed, it immediately follows that for arbitrarily large \(c \gg 1\) and \(x \rightarrow \infty ,\) we have
Now, in the region \(1-\frac{c \log \log x}{\log x} \le \sigma \le 1\),
Consequently,
It remains to compute the integral of \(|\zeta (\sigma )|^{\frac{\mu (q)}{\phi (q)}}\) occurring in (10). For \(\sigma \) very close to 1, we have
It follows that
Combining equations (10) and (11) we obtain
Remark: As pointed out to me by Gérald Tenenbaum, it is also possible to deduce Corollary 1.2 directly from theorem 2.3 by using theorem II.5.2 of [8]. In this manner one can obtain an explicit asymptotic expansion which, furthermore, is valid for values of q tending to infinity with x.
4 Examples of equidistribution (mod 3) and (mod 9)
Equidistribution (mod 3): Theorem (1.1) says that for \(h = 1, \; q = 3:\)
We computed the above sum for \(x = 10^7\) and obtained
Our theorem predicts that
Since \(\log \left( 10^7\right) \approx 16.1\) is small, this explains the discrepancy between the actual and predicted results.
As \(x \rightarrow \infty ,\) we have
where
Equidistribution (mod 9):
Our theorem says that for \(h \ne 3,6\) (\(1\le h<9)\) and \(q = 9\):
Surprisingly!! there is a huge amount of cancellation when \(x = 10^7:\)
References
K. Alladi, P. Erdős, On an additive arithmetic function. Pac. J. Math. 71(2), 275–294 (1977)
H. Davenport, Multiplicative Number Theory, vol. 74, 2nd edn., Graduate Texts in Mathematics (Springer, Berlin, 1967). (revised by Hugh Montgomery)
H. Delange, On integral valued additive functions. J. Number Theory 1, 419–430 (1969)
A. Ivić, On Certain Large Additive Functions, (English summary) Paul Erdős and His Mathematics I, vol. 11, Bolyai Society Mathematical Studies (János Bolyai Mathematical Society, Budapest, 1999,2002), pp. 319–331
A. Ivić, P. Erdős, Estimates for sums involving the largest prime factor of an integer and certain related additive functions. Studia Sci. Math. Hungar. 15(1–3), 183–199 (1980)
A, Ivić, P. Erdős, The distribution of quotients of small and large additive functions, II, in Proceedings of the Amalfi Conference on Analytic Number Theory (Maiori, 1989), (University of Salerno, Salerno, 1992), pp. 83–93
A. Selberg, Note on a paper by L.G. Sathe. J. Indian Math. Soc. (N.S.) 18, 83–87 (1954)
G. Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, vol. 163, 3rd edn., Graduate Studies in Mathematics (American Mathematical Society, Providence, 2015). Translated from the 2008 French edition by Patrick D.F. Ion
Acknowledgements
The author would like to thank Ada Goldfeld for creating the figures in this paper and would also like to thank Wladyslaw Narkiewiz for pointing out the reference [3].
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this paper
Cite this paper
Goldfeld, D. (2017). On an Additive Prime Divisor Function of Alladi and Erdős. In: Andrews, G., Garvan, F. (eds) Analytic Number Theory, Modular Forms and q-Hypergeometric Series. ALLADI60 2016. Springer Proceedings in Mathematics & Statistics, vol 221. Springer, Cham. https://doi.org/10.1007/978-3-319-68376-8_17
Download citation
DOI: https://doi.org/10.1007/978-3-319-68376-8_17
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-68375-1
Online ISBN: 978-3-319-68376-8
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)