How to evaluate this limit with harmonic numbers

$$\lim_{n\rightarrow \infty }n\left [ \widetilde{H_{n}}-H_{2n}+H_n \right ]$$


$$\displaystyle \widetilde{H_{n}}=\sum_{j=1}^{n}\frac{\left ( -1 \right )^{j-1}}{j}$$

is the alternating harmonic number?

  • $\begingroup$ Hint: Write the alternating harmonic number as a linear combination of two other harmonic numbers, and approximate all of them with natural logarithms, then use their logarithmic properties. $\endgroup$ – Lucian Jan 11 '17 at 21:21

One may recall the asymptotics of harmonic numbers, as $ n \to \infty$, $$ H_n=\log n+\gamma+\frac1{2n}+\mathcal{O}\left(\frac1{n^2} \right) $$ and one may observe that $$ \widetilde{H_{n}}-\log 2=\sum_{j=1}^{n}\frac{\left ( -1 \right )^{j-1}}{j}-\log 2=(-1)^{n-1}\int_0^1\frac{t^n}{1+t}dt $$ Then, we may write $$ \begin{align} &n\left [ \widetilde{H_{n}}-H_{2n}+H_n \right ] \\\\&=n\left [ \left(\widetilde{H_{n}}-\log 2\right)-\left(H_{2n}-\log (2n) \right)+\left(H_{n}-\log n \right)\right ] \\\\&=n\left [ (-1)^{n-1}\int_0^1\frac{t^n}{1+t}dt-\left(\gamma+\frac1{4n}+\mathcal{O}\left(\frac1{n^2} \right) \right)+\left(\gamma+\frac1{2n}+\mathcal{O}\left(\frac1{n^2} \right) \right)\right ] \\\\&=n (-1)^{n-1}\int_0^1\frac{t^n}{1+t}dt+\frac14+\mathcal{O}\left(\frac1{n} \right) . \end{align} $$ Now, integrating by parts, $$ \begin{align} n (-1)^{n-1}\int_0^1\frac{t^n}{1+t}dt&=\left. \frac{n (-1)^{n-1}t^{n+1}}{(n+1)}\frac{1}{1+t}\right|_0^1+\frac{n (-1)^{n-1}}{(n+1)}\int_0^1\frac{t^{n+1}}{(1+t)^2}\:dx\\\\ &=\frac12\frac{n}{n+1}(-1)^n+\mathcal{O}\left(\frac1{n} \right)\tag1 \end{align} $$ and, as $n \to \infty$,

$$ n\left [ \widetilde{H_{n}}-H_{2n}+H_n \right ]=\frac12\frac{n}{n+1}(-1)^n+\frac14+\mathcal{O}\left(\frac1{n} \right) $$

which does not admit a limit.

  • $\begingroup$ There's an alternating $(-1)^{n-1}$ in front of the $\int_0^1 \frac{t^{n+1}}{(1+t)^2}\,dt$ too right? That should give some $\frac{(-1)^{n-1}c}{n}$ contribution too I think :o (+1) for the nice approach though :-) $\endgroup$ – r9m Apr 6 '16 at 13:11
  • 1
    $\begingroup$ One may notice that $(-1)^n \times \mathcal{O}\left(\frac1{n} \right) =\mathcal{O}\left(\frac1{n} \right)$. Thanks! $\endgroup$ – Olivier Oloa Apr 6 '16 at 13:15

Ok, seems to me that there are two different problems here.

On one end : $n(-H_{2n}+H_n)$. You can rewrite this one as $-\sum_{i=1}^n\frac{1}{1+\frac{i}{n}}$ which is, when you think of it, a way to approximate $-(\int_0^1\frac{1}{1+x} - 1)=1-\ln(2)$ (See https://en.wikipedia.org/wiki/Riemann_sum).

On the other hand, it is commonly known that, being an alternating series, $\tilde H_n$ converges towards a finite (non-zero) number, see http://pirate.shu.edu/~wachsmut/ira/numser/proofs/altharm.html for instance.

so $n\tilde H_n$ has an infinite limit.

So overall $\lim_\infty n(\tilde H_n - H_{2n} + H_n)=\infty$

EDIT : OK so seeing the other answers I just hav a slight doubt about what you mean by $H_{2n}$. Using the standard definition ($H_n=\sum_{k=1}^n\frac 1 k$) it should be $H_{2n}=\sum_{k=1}^{2n}\frac 1 k$ but it seems to me that you imply $\sum_{k=1}^{n}\frac 1 {2k}$

  • $\begingroup$ The limit of $-\frac{1}{n}\sum_{i=1}^n\frac{1}{1+\frac{i}{n}}$ is $-\int_0^1 \frac{1}{1+x}\,dx = -\ln 2$ .. $\endgroup$ – r9m Apr 6 '16 at 13:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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