8

More specifically, how would you evaluate the below formula? $$\lim_{n\to\infty}\sum_{k=n/2}^{n}\frac{1}{k}$$ I know that the harmonic series starting at any point diverges, but when we limit it in this way, does the series diverge or converge?

If it diverges:

  • How might you determine that?

  • Is there some $d$ that we can replace with $2$ to make the sequence converge?

If it converges:

  • What does it converge to, and how might you determine that?

  • The sequence must converge for any $d>2$. Is there a formula for the series generalized for any $d$?

4 Answers4

13

For simplicity let's assume that $n$ is even (if it is odd your sequence has to be modified using integer part of $\frac{n}{2}$). Let's compute $$ \int\limits_{\frac{n}{2}}^{n} \frac{1}{ x} dx = \ln n-\ln(\frac{n}{2}) = \ln 2 $$ Now we have estimations $$ \int\limits_{\frac{n}{2} - 1}^{n} \frac{1}{ x} dx \ge \sum_{k=n/2}^{n}\frac{1}{k} \ge \int\limits_{\frac{n}{2}}^{n + 1} \frac{1}{ x} dx $$ and both estimates obviously tend to $\ln 2$ as $n \rightarrow \infty$.

With arbitrary $d$ replacing $2$ we have $\ln d$ as limit.

Matsmir
  • 4,159
12

Let $H_n = \sum_{k=1}^n 1/k$ denote the $n$-th Harmonic number. It is well-known that $(H_n - \log n)_n$ converges to some constant $\gamma$, called the Euler-Mascheroni constant.

You ask for $(H_n - H_{n/2})_n$. We write it as:

$$(H_n - \log n) - (H_{n/2} - \log (n/2)) + (\log (n) - \log (n/2))$$

The two former terms tend to $\gamma$ thus it remains $\log (n) - \log (n/2)$, which is just $-\log (1/2) = \log 2$.

This works for any $d$.

quid
  • 42,835
8

I thought it might be instructive to present a way forward that does not rely on Riemann sums or properties of Harmonic numbers.

Rather, we rely on only elementary analysis and the Taylor series for $\displaystyle \log(1+x)$. To that end, we now proceed.


The limit $\lim_{n\to\infty}\sum_{k=n/2}^n \frac1k$ converges. To see this we write

$$\begin{align} \sum_{k=1}^{2m} \frac1k&=\sum_{k=1}^m \frac1{2k}+\sum_{k=1}^m\frac1{2k-1}\\\\\ &=\sum_{k=1}^{m} \frac1{k}+\sum_{k=1}^m\left(\frac1{2k-1}-\frac1{2k}\right)\\\\ \sum_{k=m+1}^{2m}\frac1k&=\sum_{k=1}^m\left(\frac1{2k-1}-\frac1{2k}\right)\\\\ &=\sum_{k=1}^{2m}\frac{(-1)^{k+1}}{k}\tag1 \end{align}$$

Letting $m=n/2$ in $(1)$ yields

$$\lim_{n\to\infty}\left(\sum_{k=n/2}^n\frac1k-\frac1{n/2}\right)=\sum_{k=1}^\infty\frac{(-1)^{k+1}}{k}\tag2$$

The right-hand side of $(2)$ is $\log(2)$ as recognized from the Taylor Series of $\log(1+x)=\sum_{k=1}^\infty \frac{(-1)^{k+1}}{k}x^k$ evaluated at $x=1$. And the right-hand side reduces to the limit of interest. Hence, we see that

$$\lim_{n\to\infty}\sum_{k=n/2}^n\frac1k=\log(2)$$

And we are done!

Mark Viola
  • 184,670
  • Thanks for your answer! While I think Matsmir's answer is cleaner, this one definitely helped understand this conceptually. – Erez Abrams Dec 31 '19 at 17:36
  • 1
    You're welcome. My pleasure. – Mark Viola Dec 31 '19 at 18:28
  • Sorry, my wording was bad. I can only choose one answer (from what I understand), and I just subjectively like theirs more. That's all :) – Erez Abrams Dec 31 '19 at 18:30
  • That's OK. But that other answer requires one to understand Riemann integrals and their relationship to summation. – Mark Viola Dec 31 '19 at 18:32
  • 1
    I think that knowing how to integrate is more common than knowing the Taylor series for log(1+x), but I'm no expert... – Erez Abrams Jan 01 '20 at 01:17
  • You could briefly highlight, at the end of eq 2, that the series converges (before using the Taylor series to show what it converges to) by mentioning that the RHS of 2 is an alternating harmonic series. – Luca Citi Jan 01 '20 at 09:44
  • @LucaCiti If one knows how to develop that Taylor series for $\log(1+x)$, then one likely knows that the alternating harmonic series converges. – Mark Viola Jan 02 '20 at 03:59
  • @MarkViola I was thinking of the opposite: someone who may know that the alternating harmonic series converges but not know about that Taylor series. Also, at that stage you have already answered part of the question (whether it converges) and you may briefly draw the attention of the reader to this fact. I was merely suggesting adding something like "We recognise that this is the alternating harmonic series and thus we have proven that the OP's series converges. To determine what value it converges to, we observe that the right-hand side of (2) is..." – Luca Citi Jan 02 '20 at 11:22
3

Concerning the limit itself, you already received the good answers.

What I found interesting is the problem of the partial sums for large values of $n$. $$S_{n,d}=\sum_{k=\frac n d}^{n}\frac{1}{k}=H_n-H_{\frac{n}{d}-1}$$ Using, for large $p$, the expansion $$H_p=\gamma +\log \left({p}\right)+\frac{1}{2 p}-\frac{1}{12 p^2}+O\left(\frac{1}{p^4}\right)$$ apply it twice an continue with Taylor series to get $$S_{n,d}=\log (d)+\frac{d+1}{2 n}+\frac{d^2-1}{12 n^2}+O\left(\frac{1}{n^4}\right)$$ which shows the respective roles of $n$ and $d$.