2

I found an asymptotic for $\sum\limits_{p\leq x}\dfrac{p-1}{p}$ where the sum is extended over primes not exceeding $x$ in an article but I cannot find it now. I can find the following using Abel partial summation (assuming my calculation is correct) $$\dfrac{x}{2\log{x}}-x+O\!\left(\dfrac{x}{\log^{2}{x}}\right)$$ but I think the bound I saw was sharper. Does anybody know a sharper bound for this? This is probably a duplicate but the search function does not really work well for mathematical terms and the expression is not easy to express verbally.

Gary
  • 31,845
Masum
  • 270

2 Answers2

2

Your sum is $$ \sum\limits_{p \le x} {\frac{p-1}{p}}= \pi (x) - \sum\limits_{p \le x} {\frac{1}{p}} . $$ The sum on the right-hand side is $\sim \log\log x$ and any error term of any reasonable asymptotics for $\pi(x)$ is larger in size than that. Thus, your original sum will have the same asymptotic approximations as $\pi(x)$. Hence, for example, $$ \sum\limits_{p \le x} {\frac{p-1}{p}}= \int_2^x {\frac{{dt}}{{\log t}}} + \mathcal{O}(xe^{ - c\sqrt {\log x} } ). $$ Under the Riemann hypothesis, the error term can be improved to $\mathcal{O}(\sqrt{x}\log x)$.

Gary
  • 31,845
  • I am starting to think the bound I saw was actually the one I got because it doesn't look like it's possible to improve this further without RH. So for now I will close this although unsure which one to accept as the answer. – Masum Sep 05 '21 at 03:15
1

It is equivalent to $$\pi(x)-\sum_{p\le x}\frac{1}{p}$$ It's well known that $\sum_{p\le x}\frac{1}{p}=\log(\log(x))+O(1)+O(1/\log^2(x))$, see Does the sum of reciprocals of primes converge?.

It's also well known that $\pi(x)=\frac{x}{\log(x)}\left(1+O(1/\log x)\right)$.

Hence the entire sum is $$\frac{x}{\log(x)}\left(1+O(1/\log x)\right)-\log\log x-O(1)-O(1/\log^2(x))$$

  • I have a similar result by Abel's summation formula. I ignored the lesser dominating terms. Your result is actually the same as mine. I am kinda looking for a sharper result. Edited the post to reflect that – Masum Sep 03 '21 at 18:57
  • 1
    @Masum Well to get a sharper result you need to sharpen $\pi(x)$ and $\sum 1/p$. These are well known sums and I think this is the best that you can do. – TheBestMagician Sep 03 '21 at 18:59