I made some numerical simulations about the number of primes which are the sum of 3 consecutive primes (OEIS A034962), that is for instance: $$5+7+11=23$$ $$7+11+13=31$$ $$11+13+17=41$$ $$17+19+23=59$$ $$19+23+29=71$$ $$23+29+31=83$$ $$29+31+37=97$$ $$...$$ The number of such triplets, till a certain integer n, seems to be well approximated by the following function: $$\frac{e\cdot\pi(n)}{\log(n)}$$

If the previous was true, the density of such primes in the whole set of prime numbers would be comparable to that of primes inside the set of naturals.

I ask if this estimate is correct and some references about this matter.


  • $\begingroup$ is your use of $\pi$ supposed to be a capital $\prod$ to indicate a product? $\endgroup$ – Carl Witthoft Mar 4 at 14:58
  • 2
    $\begingroup$ Not at all. It indicates the primes counting function. See also Lucia's answer. $\endgroup$ – Augusto Santi Mar 4 at 15:08
  • $\begingroup$ Thanks - new notation for me! $\endgroup$ – Carl Witthoft Mar 4 at 19:24

The question asks how many primes $p_n \le x$ are there such that $p_n + p_{n+1}+p_{n+2}$ is also prime. This is beyond our reach to answer, but one can use Hardy-Littlewood type heuristics to attack this. Since $p_n + p_{n+1}+ p_{n+2}$ is roughly of size $x$, it has about $1/\log x$ chance of being prime, and so one should expect the answer to be on the scale $\pi(x)/\log x \approx x/(\log x)^2$. But the asymptotic will be a little different, since $p_n + p_{n+1} +p_{n+2}$ is not quite random -- for example it will always be odd (omitting $2+3+5$).

Let's flesh out the usual heuristic. To be prime, a number must be coprime to each prime number $\ell$. A random integer has a chance $(1-1/\ell)$ of being coprime to $\ell$. What is the chance that $p_n+ p_{n+1} + p_{n+2}$ is coprime to $\ell$? Each of the primes $p_n$, $p_{n+1}$, $p_{n+2}$ itself lies in some reduced residue class $\mod \ell$. Assuming these possibilities are equally distributed, one should get the chance $$ \frac{1}{(\ell-1)^3} \sum_{\substack{a, b, c=1 \\ (a+b+c,\ell)=1}}^{\ell-1} 1. $$ If $a$ and $b$ are given with $a\not\equiv -b \mod \ell$, then $c$ has $\ell-2$ possible residue classes, and if $a\equiv -b \mod \ell$ then $c$ has $\ell-1$ possible residue classes. So the answer is $$ \frac{1}{(\ell-1)^3} \Big( (\ell-1)(\ell-2)(\ell-2) + (\ell-1)(\ell-1)\Big)= \frac{\ell^2-3\ell+3}{(\ell-1)^2}. $$ So we adjust the random probability at $\ell$ by the factor $$ \Big(1-\frac {1}{\ell}\Big)^{-1} \frac{(\ell^2-3\ell+3)}{(\ell-1)^2}= 1 +\frac{1}{(\ell-1)^3}. $$ For example, when $\ell=2$, this adjustment factor is $2$ reflecting the fact that $p_n+p_{n+1}+p_{n+2}$ is guaranteed to be odd.

Then the analog of the Hardy--Littlewood conjecture will be $$ \sim \prod_{\ell} \Big(1+\frac{1}{(\ell-1)^3} \Big) \frac{x}{(\log x)^2}. $$ The constant in the product is approximately $2.3$; definitely not $e$.

There are interesting biases concerning why this conjecture would be an underestimate. Work of Lemke Oliver and Soundararajan makes precise conjectures on the distribution of consecutive primes in arithmetic progressions. For example, if we look at $\ell=3$, the sum $p_n + p_{n+1} +p_{n+2}$ can be a multiple of $3$ only if all three primes are congruent to each other $\mod 3$. This configuration is not preferred, and there are significant biases against it for small numbers! Similarly for other $\ell$ also, there are biases that indicate that the actual probability would be a tiny bit larger (with the effect wearing off slowly for large $x$). In other words, one might be able to identify lower order terms on the scale of $x(\log \log x)/(\log x)^3$, which might explain why the numerics suggest a larger value for the constant.

| cite | improve this answer | |

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.