6
$\begingroup$

$\newcommand{\mean}{\mathop{\mathrm{mean}}}$

Define $$ S(d) = \prod_{p|d\atop p>2}{p-1\over p-2}. $$ Bombieri and Davenport (1966) proved that $$ \mean\limits_{d\in{\mathbb N}} S(d) = \mean\limits_{d\in{\mathbb N}} \prod_{p|d\atop p>2}{p-1\over p-2} ~=~ \Pi_2^{-1} = 1 / 0.66016\ldots = 1.51478\ldots\,, \tag{1} $$ where $\Pi_2=0.66016\ldots$ is the twin prime constant.

It is not difficult to check that the average value of $S(d)$ remains unchanged ($=\Pi_2^{-1}$) if $d$ runs through an arithmetic progression $d=qn$ with $q=2^k$; in particular the average of $S(d)$ is $\Pi_2^{-1}$ when $d$ runs through all positive even integers.

Some experimentation with PARI/GP leads me to the following

Generalization of $(1)$ for an arithmetic progression $d=nq$, $n\in{\mathbb N}$: $$ \mean\limits_{d=nq\atop n\in{\mathbb N}} S(d) = \mean\limits_{d=nq\atop n\in{\mathbb N}} \prod_{p|d\atop p>2}{p-1\over p-2} ~=~ \Pi_2^{-1} \prod_{p|q\atop p>2}{p\over p-1}. \tag{2} $$

For example, if $q=5$, then average of $S(d)$ over the progression $d=5n$ is $$\mean\limits_{d=5n\atop n\in{\mathbb N}} S(d) = {5\over4}\Pi_2^{-1}.$$

Can anyone please point me to existing proofs of $(2)$? An idea/sketch of a proof would also be greatly appreciated!

Note: Let $q'$ be the largest odd divisor of $q$, then the "extra factor" in $(2)$ is

$$\prod_{p|q\atop p>2}{p\over p-1} = \prod_{p|q'}{p\over p-1} = {q'\over\varphi(q')}. $$ In particular, if $q$ itself is odd, then the "extra factor" is $q/\varphi(q)$, as was noted in comments.

$\endgroup$
5
  • 4
    $\begingroup$ Note that the extra factor equals $q/\varphi(q)$. $\endgroup$ Commented Jan 8, 2020 at 8:21
  • $\begingroup$ You may also find mathoverflow.net/questions/61842/about-goldbachs-conjecture insightful. $\endgroup$ Commented Jan 8, 2020 at 11:58
  • 3
    $\begingroup$ $(1)$ follows from the same Tauberian theorem as in the proof of the PNT. Then for $(2)$ use that $S(nq)= S(q/\gcd(n,q))S(n)$ $\endgroup$
    – reuns
    Commented Jan 8, 2020 at 14:58
  • $\begingroup$ Many thanks! Yes, for odd $q$ the extra factor is $q/\varphi(q)$. For even $q$, it's slightly less pretty. $\endgroup$
    – Alex
    Commented Jan 8, 2020 at 15:37
  • $\begingroup$ Thanks a lot @reuns for your insight! Could you possibly expand your comment into an answer, especially the $(1)\Rightarrow(2)$ part, so I can upvote... :) It looks less and less likely that someone comes up with a reference for proof of $(2)$... $\endgroup$
    – Alex
    Commented Jan 10, 2020 at 18:13

0

You must log in to answer this question.