6

In the fifth method of deriving the functional equation for Riemann's zeta function in Titchmarsh's book the following 'expansion' is assumed without proof : $$\dfrac{1}{e^x-1} = \dfrac{1}{x} - \dfrac{1}{2} + 2x \sum_{n=1}^{\infty} \dfrac{1}{4n^2 \pi^2 +x^2}.$$ I have never encountered this type of expansion and also WolframAlpha does not give a clue for the identity. How this equality is derived?

I have a second question regarding the integral just below this identity in the book

enter image description here

How to compute the integral $\int_0^{\infty} \dfrac{x^s}{4n^2 \pi^2 +x^2} dx$? Again using WolframAlpha I couldn't get any hint.

Ali
  • 143
  • https://www.wolframalpha.com/input?i=%5Csum_%7Bn%3D1%7D%5E%7B%5Cinfty%7D+%282x%2F%284n%5E2+pi%5E2+%2Bx%5E2%29%29 The series simplifies into an expression involving coth, which isn't suprising since it does look pretty similar to the coth series representation. I'd look into similar series like these ig – Max0815 Jan 08 '24 at 06:49
  • 1
    For your second question, would this help: https://math.stackexchange.com/questions/2398165/choosing-branch-cuts-for-complex-integration?noredirect=1 ? Found using Approach0. – Alexander Burstein Jan 08 '24 at 06:51
  • @Max0815, you are right I edited my OP – Ali Jan 08 '24 at 07:08
  • @AlexanderBurstein, in https://math.stackexchange.com/questions/2319721/partial-fraction-expansion-of-frac1ez-1?noredirect=1 renus' answer uses Liouville's theorem but it states that if the entire function is bounded in whole plane and not just on imaginary axis. – Ali Jan 08 '24 at 07:37
  • @AlexanderBurstein, I can't comment there because it requires 50 points to ask him/her – Ali Jan 08 '24 at 08:00
  • @Max0815, in https://math.stackexchange.com/questions/2319721/partial-fraction-expansion-of-frac1ez-1?noredirect=1 the series is calculated but I can't comment there to ask him/her as it requires 50 points in MSE, do you know if Liouville's theorem that states that if the entire function is bounded in whole plane and not just on imaginary axis, also holds only if it is bounded on imaginary axis? – Ali Jan 08 '24 at 08:07
  • Consider the function $\sin$. It's holomorphic on the whole plane and bounded on the real axis, yet it's non-constant, hence unbounded by Liouville's. Thus a counterexample for your claim is simply $z \mapsto \sin(-iz)$, since for $ix$ you'll get $\sin(x)$. – Bruno B Jan 08 '24 at 08:35
  • @BrunoB, yes! Thank you very much. Then renus' answer in that post is incorrect. If I could justify that last step I could prove the series. Other answers on MSE or Internet are not clear/simple. Would you please help me with my first question in OP if you know a rigorous approach? – Ali Jan 08 '24 at 08:40
  • reuns' answer is not necessarily incorrect since they're also using some periodicity condition, so they have stronger assumptions that yours. – Bruno B Jan 08 '24 at 08:51
  • @BrunoB it's only periodic on imaginary axis so how it implies that it is bounded on whole plane? Liouville's theorem that I have studied from Churchil's book supposes boundedness of f on whole plane. – Ali Jan 08 '24 at 10:18
  • @BrunoB I could comment there. Nonetheless, what is the new theorem you are mentioning? I mean that you said periodicity on only imaginary axis does not imply boundedness so how the function is constant? – Ali Jan 08 '24 at 11:25
  • I was not mentioning a new theorem at all, however I will admit that I did misinterpret what reuns was saying and that I didn't check if my interpretation was correct. I thought reuns was saying that there was $1$-periodicity overall, aka along all axes parallel or equal to the real axis, but that's not the case at all... It wasn't the case in my earlier example either so it wouldn't have been a counterexample to my (now faulty) previous interpretation. – Bruno B Jan 08 '24 at 11:55

2 Answers2

6

In this answer, it is shown that $$ \sum_{n=1}^\infty\frac1{n^2+u^2}=\frac\pi{2u}\coth(\pi u)-\frac1{2u^2}\tag1 $$ Therefore, $$ \begin{align} \sum_{n=1}^\infty\frac1{4n^2\pi^2+x^2} &=\frac1{4\pi^2}\sum_{n=1}^\infty\frac1{n^2+u^2}\tag{2a}\\ &=\frac1{4\pi^2}\left(\frac\pi{2u}\coth(\pi u)-\frac1{2u^2}\right)\tag{2b}\\ &=\frac1{4x}\coth\left(\frac{x}2\right)-\frac1{2x^2}\tag{2c}\\ &=\frac1{4x}\frac{e^x+1}{e^x-1}-\frac1{2x^2}\tag{2d}\\ &=\frac1{2x}\frac1{e^x-1}+\frac1{4x}-\frac1{2x^2}\tag{2d}\\ \end{align} $$ Explanation:
$\text{(2a):}$ $x=2\pi u$
$\text{(2b):}$ apply $(1)$
$\text{(2c):}$ undo the substitution from $\text{(2a)}$
$\text{(2d):}$ $\coth\left(\frac{x}2\right)=\frac{e^x+1}{e^x-1}$
$\text{(2e):}$ $\frac{e^x+1}{e^x-1}=\frac2{e^x-1}+1$

Solving $(2)$ for $\frac1{e^x-1}$ gives $$ \frac1{e^x-1}=\frac1x-\frac12+2x\sum_{n=1}^\infty\frac1{4n^2\pi^2+x^2}\tag3 $$ In this answer, it is shown that $$ \int_0^\infty\frac{x^n}{1+x^m}\,\mathrm{d}x=\frac{\pi}{m}\csc\left(\pi\frac{n+1}{m}\right)\tag4 $$ Therefore, $$ \begin{align} \int_0^\infty\frac{x^s}{4n^2\pi^2+x^2}\,\mathrm{d}x &=(2n\pi)^{s-1}\int_0^\infty\frac{x^s}{1+x^2}\,\mathrm{d}x\tag{5a}\\ &=(2n\pi)^{s-1}\frac\pi2\csc\left(\pi\frac{s+1}2\right)\tag{5b}\\ &=(2n\pi)^{s-1}\frac\pi2\sec\left(\frac{s\pi}2\right)\tag{5c} \end{align} $$ Explanation:
$\text{(5a):}$ substitute $x\mapsto2n\pi x$
$\text{(5b):}$ apply $(4)$
$\text{(5c):}$ $\csc(x+\pi/2)=\sec(x)$

robjohn
  • 345,667
3

This is an alternative answer to your first question. Let $n\ge 2$ be an integer and $x \ne 1$ be a real number. By partial fraction decomposition, we have \begin{align*} \frac{1}{{x^{2n} - 1}} & = \frac{1}{n}\frac{1}{{x^2 - 1}} + \frac{1}{n}\sum\limits_{k = 1}^{n - 1} {\frac{{x\cos \left( {\frac{{\pi k}}{n}} \right) - 1}}{{x^2 - 2x\cos \left( {\frac{{\pi k}}{n}} \right) + 1}}} \\ & = \frac{1}{n}\frac{1}{{x^2 - 1}} - \frac{{n - 1}}{{2n}} + \frac{1}{n}\sum\limits_{k = 1}^{n - 1} {\frac{{(x - 1)^2 + x - 1}}{{(x - 1)^2 + 4x\sin ^2 \left( {\frac{{\pi k}}{{2n}}} \right)}}} . \end{align*} Substituting $x = 1 + \frac{t}{2n}$ with $t\ne 0$ a real number, we find $$ \frac{1}{{\left( {1 + \frac{t}{{2n}}} \right)^{2n} - 1}} = \frac{1}{{t\left( {1 + \frac{t}{{4n}}} \right)}} - \frac{{n - 1}}{{2n}} + \sum\limits_{k = 1}^{n - 1} {\frac{{\frac{{t^2 }}{n} + 2t}}{{t^2 + 4\pi ^2 k^2 \left( {1 + \frac{t}{{2n}}} \right)\left( {\left( {\frac{{2n}}{{\pi k}}} \right)^2 \sin ^2 \left( {\frac{{\pi k}}{{2n}}} \right)} \right)}}} . $$ Passing to the limit $n\to+\infty$ yields the desired result: $$ \frac{1}{{{\rm e}^t - 1}} = \frac{1}{t} - \frac{1}{2} + \sum\limits_{k = 1}^{\infty} {\frac{{2t}}{{t^2 + 4\pi ^2 k^2 }}} . $$

Gary
  • 31,845