Define for $n \in \mathbb{N}$ the function $$\tau_1(n):=\sum_{\substack{d|n, \\ d+1|n}}1,$$ i.e. the number of consecutive divisors of an integer. The average of $\tau_1(n)$ is $1$ since $$\sum_{n\leq x}\tau_1(n)=\sum_{d<\sqrt{x}}\Big[\frac{x}{d(d+1)}\Big]=x+O(\sqrt{x}).$$ I was wondering whether more is known about $\tau_1(n)$, for example does it have a distribution function, meaning does the following limit exist for all $z \in \mathbb{R}$, $$F(z):=\lim_{x \to +\infty}\frac{1}{x}\#\{n\leq x \ \text{such that} \ \tau_1(n)\geq z\}? $$

  • $\begingroup$ Of course a suitably modulated question can be posed for $\tau_f(n):=\sum_{[f(d)]|n}1$, where $f:\mathbb{N}\to \mathbb{R}$ is any function, the previous example having $f(d)=d(d+1)$. $\endgroup$ – Captain Darling May 24 '15 at 23:59

Yes, $F(z)$ exists for all $z$. Let $y\ge 2$, and define $\tau_{1,y}(n)$ as the number of $d\le y$ for which both $d,d+1$ divide n. Let $F_y$ denote the analogue of $F$ with $\tau_{1,y}$ replacing $\tau_1$. It's clear that $F_y(z)$ exists for all $z$, since $\tau_y(n)$ is actually a periodic arithmetic function (with period the lcm of $1, 2, \dots, \lfloor y\rfloor +1$). Moreover, $F_y(z)$ is increasing as a function of $y$, and $F_y(z) \le 1$. So for each $z$, the limit $\lim_{y\to\infty} F_y(z)$ exists.

Claim: The limit definining $F(z)$ coincides with $\lim_{y\to\infty} F_y(z)$.

If $n \le x$ and $\tau_1(n) \ge z$, then either $\tau_{1,y}(n) \ge z$, or $n$ is divisible by some $d(d+1)$ with $d > y$. Hence, $$ \#\{n \le x: \tau_1(n) \ge z\} \le \#\{n \le x: \tau_{1,y}(n) \ge z\} + O(x/y). $$ Divide by $x$, and take the limsup as $x\to\infty$, and then let $y\to\infty$. This shows that the $\limsup$ of the expression in the definition of $F$ is at most $\lim_{y\to\infty} F_y(z)$. On the other hand, since $\tau_{1,y} \le \tau_{1}$, a similar but simpler argument shows that the $\liminf$ in the definition of $F$ is at least $\lim_{y\to\infty} F_y(z)$.

It seems this was known to Erdős and Hall; in their paper, they remark "It is easy to see that the density of integers $n$ for which $\tau_k(n)=r$ exists." Probably they had the above argument in mind.

  • $\begingroup$ That is rather great, could you please link their paper ? could not find it... $\endgroup$ – Captain Darling May 25 '15 at 21:45
  • 1
    $\begingroup$ Here's a PDF: renyi.hu/~p_erdos/1978-26.pdf $\endgroup$ – so-called friend Don May 25 '15 at 22:01
  • 1
    $\begingroup$ Ok so the probability distribution result is valid with the same proof even taking any polynomial $f(d)$ with integer coefficients in place of $d(d+1)$ and such that the leading coefficient is positive and the degree of $f$ is greater than $1$. It might be interesting to try and figure explicit formulas for $F(z)$... $\endgroup$ – Captain Darling May 25 '15 at 22:13

See Katalin Gyarmati, On the density of integers with consecutive divisors, Publ. Math. Debrecen 74 (2009), no. 1-2, 1–17, MR2490419 (2010e:11087).

The first paragraph of the review:

P. Erdős and R. R. Hall [J. Austral. Math. Soc. Ser. A 25 (1978), no. 4, 479–485; MR0506088 (58 #21975)] first studied the number of consecutive divisors. Let $k\ge2$ be a fixed integer; the function $\tau_k(n)$ is defined as the number of positive divisors of the form $m(m+1)\cdots(m+k−1)$ with $m\ge1$. Erdős and Hall proved that the estimate $\tau_k(n)>(\log n)^{e^{1/k}−\epsilon}$ holds for infinitely many $n$. They also proved that $$\sum_{n\le x}\tau_k(n)={x\over(k-1)(k-1)!}+O(x^{1/k})$$

The review goes on to discuss the work of Tenenbaum and other authors, and then gets to Gyamarti's paper, which is concerned with $$A_P(K)=\lim_{N\to\infty}\#\{\,n:1\le n\le N,P(m)\mid n{\rm\ for\ some\ }m\ge K\,\}$$ and finds asymptotics for $P$ a quadratic polynomial with integer coefficients.

  • $\begingroup$ The reviews indicate that there is work by Tenenbaum and dL Breteche for bounding $\tau_k$ by a power of $\tau$ which is smaller than $1$. The asymptotic estimate is also rather straightforward. But it gives the interesting result that the average number of more than one consecutive divisors is $\int_0^1 \frac{e^x-1}{x}dx$.I am wondering whether it is any easy to prove asymptotics for the higher moments of $\tau_2$. $\endgroup$ – Captain Darling May 25 '15 at 1:42

This supplements the two previous answers, giving further references that provide information on $F(z)$. Theorem 2 of Konyagin and Soundararajan shows that there exist arbitrarily (square-free) large integers $N$ with $$ \# \{ d: d(d+1)| N \} \ge \exp ((\log N)^{\frac 1{16}}). $$ Therefore for large $z$, there exists an integer $N$ below $\exp( (\log z)^{16})$ with $\# \{ d: d(d+1)|N \} \ge z$. Since all multiples of $N$ have at least $z$ consecutive divisors, it follows that $$ F(z) \ge \exp(- (\log z)^{16}) $$ for large $z$. This improves substantially upon the results of Erdos and Hall, who had a bound like $F(z) \ge \exp(-z^{1/\sqrt{e}+o(1)})$. Harper's paper could probably be adapted slightly to improve the $16$ in the result of Konyagin and Soundararajan to $6$, and further improvements could also be made. One might guess that $F(z)$ (see the heuristic in the paper of Konyagin and Soundararajan) should be something like $\exp(-(\log z)^{2+o(1)})$, but this appears difficult.

  • $\begingroup$ That is interesting, doesn't the last guess suggest that the distribution of $\tau_1(n)$ should have a heuristic behavior conforming to the gaussian law ? $\endgroup$ – Captain Darling May 26 '15 at 16:35
  • $\begingroup$ @CaptainDarling: Note it's $\exp(-(\log z)^{2+o(1)})$, not like $\exp(-z^2/2)$ (Gaussian). I don't expect it to be anything nice. $\endgroup$ – Lucia May 26 '15 at 16:46
  • $\begingroup$ Indeed you are right, however my implication was that considering $F\left(\exp\left(\frac{t}{\sqrt{2}}\right)\right)$ would lead to a probabilistic interpretation, after all the events $d|n$ and $d+1|n$ are morally independent. $\endgroup$ – Captain Darling May 26 '15 at 17:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.