Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Introduce $$\varrho(n) = \sum\limits_{\substack{1\le k\le n \\ (k,n)=1}} \frac{1}{k}.$$ The following thread at proposes to analyse the average order of $\varrho(n)$, i.e. $$\frac{1}{n} \sum_{k=1}^n \varrho(k).$$

I have tried to duplicate this calculation but I don't arrive at the same result. My question is, which one is right, the original post or my findings. My calculation follows.

First we need an identity for $\varrho$ that will prove very useful later on. Observe that $$ \varrho(n) + \sum_{\substack{d\mid n \\ d>1}} \sum^n_{\substack{k=1 \\ (k, n)=d}} \frac{1}{k} = H_n.$$ Now the LHS is $$ \varrho(n) + \sum_{\substack{d\mid n \\ d>1}} \sum^{n/d}_{\substack{m=1 \\ (m, n/d)=1}} \frac{1}{md} = \varrho(n) + \sum_{\substack{d\mid n \\ d>1}} \frac{1}{d} \varrho\left(\frac{n}{d}\right) = \sum_{d\mid n} \frac{1}{d} \varrho\left(\frac{n}{d}\right).$$ Switching to Dirichlet convolutions, we have $$\varrho \star \frac{1}{n} = H_n \sim \log n + \gamma + \frac{1}{2n}.$$ With $$ A(s) = \sum_{n\ge 1}\frac{\varrho(n)}{n^s}$$ this gives $$ A(s) \zeta(s+1) \sim -\zeta'(s) + \gamma \zeta(s) + \frac{1}{2} \zeta(s+1)$$ or $$ A(s) \sim \frac{1}{\zeta(s+1)} \left( -\zeta'(s) + \gamma \zeta(s) \right) + \frac{1}{2}.$$ To find the average order use the Mellin-Perron type integral $$\int_{3/2-i\infty}^{3/2+i\infty} A(s) n^s \frac{ds}{s} = -\frac{1}{2} \varrho(n) + \sum_{k=1}^n \varrho(k)$$ and shift to the left to pick up the residue at $s=1$, getting $$ \frac{6}{\pi^2} n \log n + \left(\frac{6(\gamma-1)}{\pi^2} - \frac{36}{\pi^4} \zeta'(2)\right) n + O(\log n)$$ so that the average order is $$ \frac{1}{n} \sum_{k=1}^n \varrho(k) \sim \frac{6}{\pi^2} \log n + \left(\frac{6(\gamma-1)}{\pi^2} - \frac{36}{\pi^4} \zeta'(2)\right) + O\left(\frac{\log n}{n}\right).$$

Which one is right?

Addendum. In view of Eric Naslunds excellent comment below maybe we can ask whether anyone is able to supply those missing bounds on the rest of the contour for the Mellin-Perron integral, thereby turning this question into a useful reference. Here is a MSE challenge of the same type.

share|cite|improve this question
The numerical evidence is unambiguously in your favour; $$\frac1n\sum_{k=1}^n\varrho(k)-\frac6{\pi^2}\log n$$ fluctuates near $0.0895$, which coincides up to three significant digits with your constant. – joriki Oct 30 '12 at 23:12
I accidentally let the calculation continue in the background :-) It came up to $n=733184$ and seemed to be hovering around $0.089478\pm0.000005$, in agreement with your constant $0.894727\dots$ – joriki Oct 31 '12 at 6:12
I tried to duplicate Eric's calculation and am very certain that you are correct. I guess that Eric accidentally wrote $\sum \frac{\mu(n)}{n^s} = \zeta(s)$ when he tried to evaluate $\sum \frac{\mu(n) \log(n)}{n^s}$, which causes the error. – user27126 Nov 6 '12 at 1:11

Your calculation is correct. When I reached the sum $$\sum_{n=1}^\infty \frac{\mu(n)\log n}{n^2}$$ I incorrectly wrote that it equals $-\zeta^{'}(2)$, rather than $\frac{\zeta^{'}(2)}{\zeta(2)^2}$. To see why it is $\frac{\zeta^{'}(2)}{\zeta(2)^2}$,simply take the derivative of $$\frac{1}{\zeta(s)}=\sum_{n=1}^\infty \mu(n)n^{-s}.$$

Remark: I note again that your proof is missing most of the critical details. When using the residue theorem to evaluate the sum of a multiplicative function, the majority of the work appears when bounding the other three parts of the contour, something which you have taken for granted. Evaluating the residue is the easiest part, and only a minor detail. You must prove that this is actually the asymptotic. This is why I shied away from the residue theorem in my other answers. It provides a nice and fast heuristic, allowing us to see what the answer should be, but to actually prove the results and find the correct error term requires bounds on zeta and lemmas I didn't want to use. It is not always trivial to guess the correct error term when using residue methods.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.