Using result from calculus:(1) $ \log(x\log x)>\log\vartheta \cdot \log(2x\log 2x)$ for $2\leq \vartheta<e$ ($e$-Euler number) and $ x>x(\vartheta )>2$, I am able to prove $ p_{2n}<\frac{2n}{\log(\vartheta )}\cdot \log(p_{n})$ for $n>n(\vartheta)$, where $p_{n}$ is the $n$-th prime number. The proof does not work for $\vartheta=e$, since inequality (1) is obviously not valid in this case. But data suggests also stronger result:
$ p_{2n}<2n\cdot \log(p_{n})$ for $n>52$. How can it be proved?
-
3$\begingroup$ Since $p_n \sim n\log(n)$, your inequality is certainly true for large enough $n$. $\endgroup$– Dave BensonCommented Apr 25, 2023 at 10:23
-
$\begingroup$ Sorry, could you be more explicit?,because i dont see the line of the proof betwen the asymptotic estimate and explicit inequality. Thank you. $\endgroup$– Andrej LeškoCommented Apr 25, 2023 at 12:26
-
1$\begingroup$ I think @DaveBenson means we already know $p_{2n} \sim 2n\log(n) \sim 2n\log(p_n) < c 2n\log(p_n)$ for any $c>1$ and large enough $n$ depending on $c$. Something like what you stated as an introduction to your question. I don't see either how it could answer your question ($c=1$). $\endgroup$– Claude ChaunierCommented Apr 25, 2023 at 13:30
-
$\begingroup$ please use $\log n$ etc, your equations look terrible $\endgroup$– kodluCommented Apr 25, 2023 at 16:57
1 Answer
Let's use some standard bounds on the $n$th prime, as found in this paper by Pierre Dusart.
We have $$ p_{2n} \leq 2n[\ln(2n)+\ln(\ln(2n))-0.9484] = 2n\left[\ln(n)+\ln\left(\frac{2}{e^{0.9484}}\ln(2n)\right)\right] $$ using an inequality stated on page 414 of the cited paper, valid for $n\geq 19509$. On the other hand, we have $$2n\ln(p_n)\geq 2n\ln[n(\ln(n)+\ln(\ln(n))-1)]=2n[\ln(n)+\ln(\ln(n\ln(n)/e))] $$ by the main result of that paper, for $n\geq 2$.
Thus, your desired inequality will be true once $n\geq 19509$ and when $$\frac{2}{e^{0.9484}}\ln(2n)<\ln\left(\frac{n\ln(n)}{e}\right). $$ That last inequality becomes true when $n\geq 13$. Then a simple check will show that the inequality you want is valid in the region $52\leq n\leq 19509$.
-
$\begingroup$ As i understand, the proof rest on estimate $ n\cdot log(nlogn)-c_{1}n\leqslant p_{n}\leqslant n\cdot log(nlogn)-c_{2}n $ and work for any $ c_{1}\geqslant 1$ and $,c_{2}\leqslant 0.9484$ $\endgroup$ Commented Apr 26, 2023 at 8:49
-
$\begingroup$ @AndrejLeško Yes, as long as $n$ is big enough. Even more terms can be appended, using the asymptotic expansion for the order of $p_n$, whose first few terms were given by Cipolla in 1902 (as mentioned in the linked paper above); but making that effective takes more work. $\endgroup$ Commented Apr 26, 2023 at 16:21