4

I just come across this question by trying to analyze the pseudoinverse of some infinite matrix (the matrix T as interpreted in my answer to this MSE-question), where this series occurs from some dotproducts. I think it can also be expressed in terms of the Moebius-function: $$\sum_{k=1}^\infty \left| {\operatorname{moebius}(k) \over k} \right| $$ Remembering, that the sum of the reciprocals of the primes is divergent I think I should assume divergence here as well, but I'm not sure.

  • Q1: Is this sum convergent (and if, what is its value)?

  • Q2: if it is divergent, is there possibly some finite value related, like for instance the Euler/Mascheroni-$\gamma$ for the harmonic series?

  • It's divergent because it's at least $\sum_p\frac{1}{p}$. – Unit May 08 '15 at 07:56
  • 2
    Well... prime numbers are certainly square-free, thus $$\sum_{p ; \text{prime}} \frac{1}{p} \leq \sum_{k = 1}^{\infty} \frac{|\mu(k)|}{k}$$... – A.P. May 08 '15 at 07:56
  • 4
    $\displaystyle\sum_{n\text{ squarefree}} \dfrac{1}{n} = \displaystyle\prod_{p\text{ prime}} \left(1+\dfrac{1}{p}\right) $ – Darth Geek May 08 '15 at 08:00
  • @A.P. : reading your argument makes me feeling nearly stupid (but well, it's ok). Of course! (I asked question Q2 there more out of couriosity.... ) – Gottfried Helms May 08 '15 at 08:04
  • 1
    @Darth: thanks too - that formal re-expression of the sum is a nice reminder to look at - perhaps it helps to understand things better if needed later. – Gottfried Helms May 08 '15 at 08:06
  • @GottfriedHelms Don't worry, we all overlook something once in a while. Paraphrasing one of my professors: the only way not to make mistakes is to do nothing. I have no clue about Q2, though... – A.P. May 08 '15 at 08:28

2 Answers2

5

The asymtotic growth of the sum of the reciprocals of squarefree numbers $n\le x$ is $\frac{6 e^\gamma}{\pi^2} \log x$, because we have

$$ \prod\limits_{p≤x}\left(1 + \frac{1}{p} \right) = \frac{\prod\limits_{p≤x}\left(1 - \frac{1}{p^2} \right)}{\prod\limits_{p≤x}\left(1 - \frac{1}{p} \right)} \sim \frac{\frac{1}{\zeta(2)}}{\frac{1}{e^\gamma \log x}} = \frac{6 e^\gamma}{\pi^2} \log x. $$ In particular we have $$ \sum_{n {\rm \; squarefree}}\frac{1}{n}=\prod_{p}\left(1 + \frac{1}{p} \right) =\infty. $$

Dietrich Burde
  • 130,978
1

Let $Q(x)$ denote the number of squarefree positive integers $n$ with $1\le n\le x$. Then we may write $Q(x)=\frac{6x}{\pi^2}+R(x)$, where $R(x)=O(\sqrt{x})$ as $x\to \infty$ ($R(x)$ is in fact smaller than this. See: wikipedia: square-free integer). Now we apply Abel's summation formula, to find \begin{align*}\sum_{\substack{n\le x \\ n \text{ squarefree}}}\frac{1}{n}=&\;\frac{Q(x)}{x}+\int_{1}^{x}\frac{Q(t)}{t^2}\text{d}t \\[-1mm] =&\; \frac{6}{\pi^2}+\frac{R(x)}{x}+\frac{6}{\pi^2}\int_{1}^{x}\frac{\text{d}t}{t}+\int_{1}^{x}\frac{R(t)}{t^2}\text{d}t \\[2mm] =&\; \frac{6\log x}{\pi^2}+\Big(\frac{6}{\pi^2}+\int_{1}^{\infty}\frac{R(t)}{t^2}\text{d}t\Big)+\Big(\frac{R(x)}{x}-\int_{x}^{\infty}\frac{R(t)}{t^2}\text{d}t\Big).\end{align*} The first bracketed term converges as $x\to \infty$ (to $C_{\text{sf}}$, say), while the second bracketed term vanishes as $x\to \infty$ (the strength of the bound may depend on if the Riemann hypothesis is assumed or not).

The constant $C_{\text{sf}}$ can be expressed more transparently as follows: First note the identity
$$(\ast) \hspace{1cm} \sum_{n\le x}\sum_{d|n}f(d)=\sum_{d\le x}f(d)\Big\lfloor \frac{x}{d}\Big\rfloor=x\sum_{d\le x}\frac{f(d)}{d}-\sum_{d\le x}f(d)\Big\{\frac{x}{d}\Big\}.$$ Second, note that we can write $$Q(x)=\sum_{n\le x}\sum_{d|n}f(d),$$ where $$f(d)=\begin{cases}\mu(\sqrt{d}) & \text{if } d \text{ is a perfect square,} \\ 0 & \text{otherwise.} \end{cases}$$ Using this in the identity $(\ast)$, we obtain \begin{align*}Q(x)=&\; x\sum_{d^2\le x}\frac{\mu(d)}{d^2}-\sum_{d^2\le x}\mu(d)\Big\{\frac{x}{d^2}\Big\} \\ =&\; \frac{6x}{\pi^2} \hspace{2mm}\underbrace{-x\sum_{d^2>x}\frac{\mu(d)}{d^2}-\sum_{d^2\le x}\mu(d)\Big\{\frac{x}{d^2}\Big\}}_{R(x)}. \end{align*}

In summary, the squarefree harmonic series diverges to $+\infty$, is asymptotic to $ \frac{6\log x}{\pi^2}$, and has an associated constant $C_{\text{sf}}$, where \begin{align*}C_{\text{sf}}=&\;\lim_{x\to \infty}\Big(-\frac{6\log x}{\pi^2}+\sum_{\substack{n\le x \\ n \text{ squarefree}}}\frac{1}{n}\Big)=\frac{6}{\pi^2}+\int_{1}^{\infty}\frac{R(t)}{t^2}\text{d}t \\[1.5mm] =&\; \frac{6}{\pi^2}-\int_{1}^{\infty}\frac{1}{t^2}\bigg(\sum_{d^2>t}\frac{\mu(d)}{d^2}+\sum_{d^2\le t}\mu(d)\Big\{\frac{t}{d^2}\Big\}\bigg) \text{d}t.\end{align*}

Some parts of the calculation above may be found with a more detailed explanation in Montgomery & Vaughan's Multiplicative Number Theory I: Classical Theory (Chapter 2.1).

AfterMath
  • 852
  • 1
    Thanks @Aftermath! Unfortunately it is -at the moment- too much for me and I have to come back to this later (not this week, for instance). But I hope anyway it is useful for some lurkers here as well! – Gottfried Helms Jun 19 '23 at 11:51