2
$\begingroup$

I am looking for a derivation of the following sum: $$\sum_{n=1}^{\infty}\bigg(\left(n+\frac{1}{2}\right)\ln\left(1+\frac{1}{n}\right)-1\bigg)=1-\ln(\sqrt{2\pi})$$ My current derivation(s) uses the zeta function at negative integers (and or Stirling approximation/ the derivative of $\zeta'(0)$). I want to avoid those.

How I got an answer was via regularization of $$-\sum_{i=1}^{\infty}\frac{\zeta(-i)}{i}=\sum_{n=1}^{\infty}\bigg(\left(n+\frac{1}{2}\right)\ln\left(1+ \frac{1}{n}\right)-1\bigg)$$ My own other try was rewriting it via: $$\sum_{n=1}^{\infty}\bigg(\left(n+\frac{1}{2}\right)\ln\left(1+\frac{1}{n}\right)-1\bigg)=\sum_{k=2}^{\infty} \zeta(k)(-1)^k \bigg(\frac{1}{k+1}-\frac{1}{2k}\bigg)$$ If this works I am already happy. If there's another simple way I'd love to hear it as well.

$\endgroup$
3
  • 5
    $\begingroup$ This doesn't make any sense because $n$ is an index variable $\endgroup$ May 10 '20 at 12:14
  • $\begingroup$ Should it be a summation on RHS? $\endgroup$
    – Henry Lee
    May 10 '20 at 12:19
  • $\begingroup$ I made a typo the rhs should be not n but pi. I am sorry. $\endgroup$
    – Gerben
    May 10 '20 at 13:08
4
$\begingroup$

One has \begin{align*} & \sum\limits_{n = 1}^\infty {\left[ {\left( {n + \frac{1}{2}} \right)\log \left( {1 + \frac{1}{n}} \right) - 1} \right]} = \sum\limits_{n = 1}^\infty {\int_0^1 {\frac{{\frac{1}{2} - t}}{{n + t}}dt} } = \sum\limits_{n = 1}^\infty {\int_0^1 {\frac{{\frac{1}{2} - (t - \left\lfloor t \right\rfloor )}}{{n + t}}dt} } \\ & = \sum\limits_{n = 1}^\infty {\int_{n - 1}^n {\frac{{\frac{1}{2} - (t - \left\lfloor t \right\rfloor )}}{{t + 1}}dt} } = \int_0^{ + \infty } {\frac{{\frac{1}{2} - (t - \left\lfloor t \right\rfloor )}}{{t + 1}}dt} . \end{align*} Now, by the Euler--Maclaurin formula, $$ \log k! = \left( {k + \frac{1}{2}} \right)\log k- k + C + \int_0^{ + \infty } {\frac{{\frac{1}{2} - (t - \left\lfloor t \right\rfloor )}}{{t + k}}dt} $$ with some constant $C$. It can be shown that the integral is $\mathcal{O}(k^{-1})$ and so by Stirling's formula (or the Wallis product), $C=\frac{1}{2}\log (2\pi )$. Thus \begin{align*}\sum\limits_{n = 1}^\infty {\left[ {\left( {n + \frac{1}{2}} \right)\log \left( {1 + \frac{1}{n}} \right) - 1} \right]} & = \log 1! - \left( {\left( {1 + \frac{1}{2}} \right)\log 1 - 1 + \frac{1}{2}\log (2\pi )} \right) \\ &= 1 - \frac{1}{2}\log (2\pi ). \end{align*}

$\endgroup$
3
  • $\begingroup$ I got here, trying to derive Stirlings formula. That's why I wanted to avoid it. Hoping there was an easier way to solve this sum. Why wallis product? $\endgroup$
    – Gerben
    May 10 '20 at 13:44
  • $\begingroup$ The use of Wallis product is a common way to determine the constant $\sqrt{2\pi}$ in Stirling's formula. The series you are looking at is a special case of the Gudermann series. See Chapter 2, §4 in "Classical Topics in Complex Function Theory" by Reinhold Remmert. $\endgroup$
    – Gary
    May 10 '20 at 15:05
  • $\begingroup$ Thanks. It's really not that exciting indeed. Think i actually seen it before. I really like your integral technique though $\endgroup$
    – Gerben
    May 10 '20 at 21:45
3
$\begingroup$

New Answer. Let $S_N$ denote the partial sum for the first $N$ terms. Then $S_N$ is related to the Stirling's Formula by the following computation:

\begin{align*} S_N &= \sum_{n=1}^{N} \left(n+\frac{1}{2}\right)\log(n+1) - \sum_{n=1}^{N} \left(n+\frac{1}{2}\right)\log n - N \\ &= \left(N+\frac{1}{2}\right)\log (N+1) - \log (N!) - N. \end{align*}

Now we consider $e^{-S_N}$ instead. Using the formula $\int_{0}^{\infty}x^{n}e^{-sx}\,\mathrm{d}x=\frac{n!}{s^{n+1}}$,

\begin{align*} \exp(-S_N) &= \frac{N!e^{N}}{(N+1)^{N+\frac{1}{2}}} \\ &= \frac{N^{N+1}}{(N+1)^{N+\frac{1}{2}}} \int_{0}^{\infty} x^N e^{-N(x-1)} \, \mathrm{d}x \\ &= \frac{1}{(1+\frac{1}{N})^{N+\frac{1}{2}}} \int_{-\infty}^{\infty} \left(1 + \frac{u}{\sqrt{N}}\right)_{+}^N e^{-\sqrt{N}u} \, \mathrm{d}u, \end{align*}

where we utilized the substitution $x=1+\frac{u}{\sqrt{N}}$ in the last step and $x_{+}:=\max\{0,x\}$ denotes the positive part of $x$. Then, taking limit as $N\to\infty$ and assuming for a moment that the order of limit and integral can be swapped, we get

\begin{align*} \lim_{N\to\infty} \exp(-S_N) &= \biggl( \lim_{N\to\infty} \frac{1}{(1+\frac{1}{N})^{N+\frac{1}{2}}} \biggr) \int_{-\infty}^{\infty} \lim_{N\to\infty} \left(1 + \frac{u}{\sqrt{N}}\right)_{+}^N e^{-\sqrt{N}u} \, \mathrm{d}u \\ &= \frac{1}{e} \int_{-\infty}^{\infty} e^{-u^2/2} \, \mathrm{d}u = \frac{\sqrt{2\pi}}{e}. \end{align*}

Here, the last step follows from the gaussian integral. Therefore

$$ \sum_{n=1}^{\infty} \left[ \left(n+\frac{1}{2}\right)\log\left(1+\frac{1}{n}\right)-1 \right] = \lim_{N\to\infty} S_N = 1 - \log\sqrt{2\pi} $$

provided the interchange of limit and integral is justified. For this, we note the following inequality:

$$ \log(1+x) \leq x - \frac{x^2}{2(1+x_+)}, \qquad x > -1 $$

From this, we deduce that

$$ \left(1 + \frac{u}{\sqrt{N}}\right)_{+}^N e^{-\sqrt{N}u} \leq \exp\left(-\frac{u^2}{2(1+u_+)}\right) $$

holds for all $N\geq 1$ and for all $u \in \mathbb{R}$. Therefore the dominated convergence theorem is applicable and the desired step is justified, completing the proof.


Old Answer. The sum converges absolutely by the Limit Comparison Test with $\zeta(2)$. Now for each given $n \geq 1$,

\begin{align*} \left(n+\frac{1}{2}\right)\log\left(1+\frac{1}{n}\right)-1 &= \left(n+\frac{1}{2}\right)\left(\sum_{j=1}^{\infty}\frac{(-1)^{j-1}}{jn^j} \right)-1\\ &= - \frac{1}{4n^2} + \left(n+\frac{1}{2}\right)\sum_{j=3}^{\infty}\frac{(-1)^{j-1}}{jn^j}\\ &= - \frac{1}{4n^2} + \sum_{j=3}^{\infty}\frac{(-1)^{j-1}}{j}\left(\frac{1}{n^{j-1}}+\frac{1}{2n^j}\right). \end{align*}

Using the formula $\int_{0}^{\infty}x^{s-1}e^{-nx}\,\mathrm{d}x=\frac{\Gamma(s)}{n^s}$, this may be recast as

\begin{align*} &= \int_{0}^{\infty}\left[ - \frac{x}{4} + \sum_{j=3}^{\infty}\frac{(-1)^{j-1}}{j}\left( \frac{x^{j-2}}{(j-2)!} + \frac{x^{j-1}}{2(j-1)!} \right)\right] e^{-nx}\, \mathrm{d}x \\ &= \int_{0}^{\infty} \left( \frac{1}{x} - \left(\frac{1}{2x}+\frac{1}{x^2}\right)(1-e^{-x}) \right) e^{-nx} \, \mathrm{d}x. \end{align*}

Summing this for $n = 1, 2, \dots$, we get

\begin{align*} S &:= \sum_{n=1}^{\infty} \left[ \left(n+\frac{1}{2}\right)\log\left(1+\frac{1}{n}\right)-1 \right] \\ &= \int_{0}^{\infty} \left( \frac{1}{x} - \left(\frac{1}{2x}+\frac{1}{x^2}\right)(1-e^{-x}) \right) \frac{1}{e^x - 1} \, \mathrm{d}x \\ &= \int_{0}^{\infty} \left( \frac{1}{x(e^x - 1)} - \left(\frac{1}{2x}+\frac{1}{x^2}\right)e^{-x} \right) \, \mathrm{d}x. \end{align*}

To compute the right-hand side, we consider the following regularization:

\begin{align*} S(s) &:= \int_{0}^{\infty} \left( \frac{1}{x(e^x - 1)} - \left(\frac{1}{2x}+\frac{1}{x^2}\right)e^{-x} \right) x^s \, \mathrm{d}x \\ &= \int_{0}^{\infty} \left( \frac{x^{s-1}}{e^x - 1} - \frac{1}{2}x^{s-1}e^{-x} - x^{s-2}e^{-x} \right) \, \mathrm{d}x. \end{align*}

This function is analytic for $\operatorname{Re}(s) > -1$, and $S = S(0)$. Moreover, for $s > 2$, we easily find that

\begin{align*} S(s) &= \Gamma(s)\zeta(s)-\frac{1}{2}\Gamma(s)-\Gamma(s-1) \\ &= \Gamma(s+1)\biggl( \frac{\zeta(s)-\frac{1}{2}-\frac{1}{s-1}}{s} \biggr). \end{align*}

By the principle of analytic continuation, this identity must hold on all of $\operatorname{Re}(s)>-1$. So, letting $s \to 0$ to the above formula yields

$$ S = \lim_{s\to 0}S(s) = 1 + \zeta'(0). $$

Now the desired formula follows from $\zeta'(0) = -\log\sqrt{2\pi}$.

$\endgroup$
6
  • $\begingroup$ Maybe a question on its own but how do I find S(s) in terms of gamma and zeta? Somehow I am much more familiar with series expension and summations then integerals. $\endgroup$
    – Gerben
    May 10 '20 at 14:53
  • $\begingroup$ @Gerben, This follows from the formula $$\int_{0}^{\infty}\frac{x^{s-1}}{e^x-1}\,\mathrm{d}x=\sum_{n=1}^{\infty}\int_{0}^{\infty}x^{s-1}e^{-nx}\,\mathrm{d}x=\sum_{n=1}^{\infty}\frac{\Gamma(s)}{n^s}=\Gamma(s)\zeta(s)$$ that holds for $\operatorname{Re}(s)>1$. $\endgroup$ May 10 '20 at 17:51
  • $\begingroup$ @Gerben, But if your goal is to derive the Stirling's formula, then this computation might not be best suited for the purpose, since you will need to derive $\zeta'(0) = -\log\sqrt{2\pi}$ while avoiding a circular argument. $\endgroup$ May 10 '20 at 18:12
  • $\begingroup$ Not the most solid proof like way, but i think i done that ages ago some what intinuative, although i would rephrase the beginning a bit now. math.stackexchange.com/questions/1325169/… $\endgroup$
    – Gerben
    May 10 '20 at 21:30
  • $\begingroup$ And I think i can craft atleast an argument for the integral, I can atleast look into it. $\endgroup$
    – Gerben
    May 10 '20 at 21:33
0
$\begingroup$

The series is not convergent, so the formula is wrong. $ (n+\frac 1 2 ) \ln (1+\frac 1 n) \to 1$ as $ n \to \infty$ and this proves that LHS is $\infty$. Also RHS depends on $n$.

$\endgroup$
2
  • $\begingroup$ the series does converge i think? You just missed the "- 1" part. It's $(n+\frac{1}[2})\ln(1+\frac{1}{n}-1$ $\endgroup$
    – Gerben
    May 10 '20 at 13:07
  • $\begingroup$ Maybe i should clarify that with big bracelets around the sum. When i compute the imo convergent sum asked, it clearly converge to the value given. Similair for the divergent sum over the negative zeta, but that one diverge once above a certain value, yet for the first few value's it's ofcourse very close to the regularized value given. $\endgroup$
    – Gerben
    May 10 '20 at 13:13
0
$\begingroup$

Consider the integral $$\displaystyle \int\limits_{\displaystyle j}^{\displaystyle j+1}\frac{\displaystyle \left \{ x \right \}-\frac{\displaystyle 1}{\displaystyle 2}}{\displaystyle x}dx $$

$\endgroup$

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.