5
$\begingroup$

In this previous question it is shown that the convolution of the prime counting function $\pi$ with itself, is related to the Goldbach conjecture:

$$\pi^*(n):=\sum_{k=0}^n \pi(k) \pi(n-k)$$

The Goldbach conjecture might be written as:

$$\forall n \ge 2: \frac{\pi^*(2n)+\pi^*(2n-2)}{2} > \pi^*(2n-1)$$

I have plotted the function $\pi^*$ and it looks like a parabola:

Goldbach_conjecture_prime_counting_function_convolution

Question: Is there any explanation for this observation or is this superficial observation maybe wrong?

Here is some SageMath Code to compute parameters $a,b,c,d,e$ for regression:

$$an^4+bn^3+cn^2+dn+e \approx \pi^*(n)$$

Thanks for your help!

$\endgroup$
6
  • $\begingroup$ Given the prime number theorem asymptotic, it looks heuristically like one could expect this function to grow polynomially (sort of). Ignoring the logarithm in the denominator of the asymptotic, $\sum\limits_{k=0}^n k(n-k) \approx \frac{n^3}{2} - \frac{n^3}{3}$. I know little about analytic number theory so take this with a grain of salt. $\endgroup$
    – Bma
    Commented Jul 10, 2022 at 12:22
  • $\begingroup$ It seems your reformulation of Goldbach's conjecture boils down to showing the function $\pi^{*}$ is convex, which may be achieved through mathematicals means outside mere analytical number theory. $\endgroup$ Commented Jul 11, 2022 at 2:42
  • $\begingroup$ Ok, I just found out I made a similar comment to your previous question. Have you tried replacing $\pi$ by its expression worked out by Riemann in his famous memoir? The conjectured convexity may be tightly related to RH and maybe to the vertical distribution of critical zeros of Zeta as well. $\endgroup$ Commented Jul 11, 2022 at 2:58
  • $\begingroup$ @SylvainJULIEN no, not yet. have started it but not gone through the computation. $\endgroup$ Commented Jul 11, 2022 at 4:40
  • $\begingroup$ Do you mean the Goldbach conjecture that every even counting number greater than 2 is equal to the sum of two prime numbers? If not, could you please give a reference to the conjecture you're referring to as the Goldbach conjecture? $\endgroup$ Commented Jul 11, 2022 at 21:17

1 Answer 1

11
$\begingroup$

It is straightforward to show that $$\pi^*(n)\sim\frac{n^3}{6\log^2 n}.\tag{$\ast$}\label{ast}$$ So the graph of $\pi^*(n)$ does not look like a parabola. Instead, it looks like the graph of $\frac{n^3}{\log^2 n}$.

Indeed, we have $$\sum_{\min(k,n-k)<\frac{n}{\log n}} \pi(k) \pi(n-k)\ll\frac{n}{\log n}\pi(n)^2\ll\frac{n^3}{\log^3 n},$$ and also \begin{align*}\sum_{\min(k,n-k)\geq\frac{n}{\log n}} \pi(k) \pi(n-k)&\sim\frac{1}{\log n^2}\sum_{\min(k,n-k)\geq\frac{n}{\log n}}k(n-k)\\&\sim\frac{1}{\log n^2}\sum_{k=0}^n k(n-k)\\&\sim\frac{n^3}{6\log^2 n}.\end{align*} Here we used that for $\min(k,n-k)\geq\frac{n}{\log n}$, both $\log k$ and $\log(n-k)$ are asymptotically $\log n$.

Added. The asymptotic formula \eqref{ast} converges rather slowly. Here are some numeric data: \begin{align*} \pi^*(10^2)&=16329\\ \pi^*(10^3)&=6311273\\ \pi^*(10^4)&=3119183737\\ \pi^*(10^5)&=1817310193749\\ \pi^*(10^6)&=1181102034701650\\ \pi^*(10^7)&=827525141442938787\\ \pi^*(10^8)&=611768346585852887680 \end{align*} The corresponding ratios of the two sides of \eqref{ast} are: \begin{align*} n=10^2\quad \rightsquigarrow\quad 2.0778\\ n=10^3\quad \rightsquigarrow\quad 1.8069\\ n=10^4\quad \rightsquigarrow\quad 1.5876\\ n=10^5\quad \rightsquigarrow\quad 1.4453\\ n=10^6\quad \rightsquigarrow\quad 1.3526\\ n=10^7\quad \rightsquigarrow\quad 1.2899\\ n=10^8\quad \rightsquigarrow\quad 1.2455 \end{align*}

$\endgroup$
8
  • $\begingroup$ Thanks, for this proof. :-) How difficult is it to prove asymptotically the Goldbach conjecture? :-) If it is possible at all, I do not know. (Sorry for my naive question concerning Goldbach.) $\endgroup$ Commented Jul 10, 2022 at 18:58
  • $\begingroup$ @mathoverflowUser We don't know if the Goldbach conjecture is true. If it is false, then it is impossible to prove it. If it is true, then it is probably very hard to prove it as it has been an open problem for 280 years. $\endgroup$
    – GH from MO
    Commented Jul 10, 2022 at 19:01
  • $\begingroup$ Yes ok, I understand. I just checked that your approximate formula for $\pi^*$ satisfies the inequality of Goldbach given above ( up to n=3000).... So this is funny. :-) I have updated the quesiton with your formula. $\endgroup$ Commented Jul 10, 2022 at 19:05
  • $\begingroup$ @mathoverflowUser Yes, but the asymptotic formula is useless for the inequality. $\endgroup$
    – GH from MO
    Commented Jul 10, 2022 at 19:08
  • $\begingroup$ Right, I was too entusiasthic. :-) Sorry for that! $\endgroup$ Commented Jul 10, 2022 at 19:09

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .