2

I have the series for the Taylor expansion of $\arctan(x)$ multiplied by a fraction of sinh functions: $$\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}\frac{\sinh(\frac{(2n+1)\pi}{2})}{2\sinh((2n+1)\pi)}.$$ Using $\int_0^1 (-x^2)^n dx = \frac{(-1)^n}{2n+1}$, I get:

$$\int_0^1 \sum_{n=0}^\infty \frac{\sinh(\frac{(2n+1)\pi}{2})}{2\sinh((2n+1)\pi)}(-x^2)^ndx$$

I don't know if this is the case, but I'm assuming $\sum_{n=1}^\infty (-ax^2)^n$ is a desirable form that can readily be evaluated. With that in mind, after rewriting $\sinh((2n+1)\pi/2) = 2(e^{n\pi}e^{\pi/2}-e^{-n\pi}e^{-\pi/2})$, I get:

$$2\int_0^1 \bigg(e^{\pi/2}\sum_{n=0}^\infty \frac{1}{2\sinh((2n+1)\pi)}(-e^\pi x^2)^n - e^{-\pi/2}\sum_{n=0}^\infty \frac{1}{2\sinh((2n+1)\pi)}(-e^\pi x^2)^n\bigg)dx \\ =4\sinh(\pi/2)\int_0^1 \sum_{n=0}^\infty \frac{1}{2\sinh((2n+1)\pi)}(-e^\pi x^2)^ndx$$

My problem for getting to this now is being stuck with $\sinh((2n+1)\pi)$ in the denominator.

Any suggestions?


Note: we can also write: $\sinh(n\pi/2)/\sinh(n\pi) = \frac{1}{2\sinh(\frac{(2n+1)\pi}{2})}$, but I don't know if that helps us

EDIT: I thought about trying:

$$\frac{1}{2\sinh((2n+1)\pi)} = \int_0^1 x^{2\sinh((2n+1)\pi)-1} dx$$

And $$x^{2\sinh((2n+1)\pi)-1}=\frac{x^{e^{2n\pi+\pi}-e^{-2n\pi-\pi}}}{x} \\ = \frac{(x^{e^{2n\pi+\pi}})(x^{-e^{-2n\pi-\pi}})}{x} \\ = \frac{(x^{e^{2n\pi}})^{e^\pi}(x^{-e^{-2n\pi}})^{-e^{-\pi}}}{x}$$

And while it is a way to rewrite the sinh term, I don't think this helps.

Gary
  • 31,845

2 Answers2

5

Let $S$ be the sum to compute. Using $$\frac{\sinh\left(\frac{(2n+1)\pi}{2}\right)}{\sinh((2n+1)\pi)} = \frac 1 {2\cosh\left(\frac{(2n+1)\pi}{2}\right)}$$ and the known pole expansion for $\frac 1 {\cosh}$: $$\frac 1 {\cosh z} =\sum_{k=0}^{+\infty} (-1)^k \frac{(2k+1)\pi}{\left(\frac{2k+1}2 \right)^2\pi^2 + z^2}$$ $$\begin{split} \pi S & = \frac \pi 4 \sum_{n=0}^{+\infty} \frac{(-1)^n}{2n+1}\frac{1}{\cosh\left(\frac{(2n+1)\pi}2\right)}\\ & = \frac \pi 4 \sum_{n=0}^{+\infty} \frac{(-1)^n}{2n+1}\sum_{k=0}^{+\infty} (-1)^k\frac{(2k+1)\pi}{\left(\frac{2k+1}2 \right)^2\pi^2 + \left(\frac{2n+1}2 \right)^2\pi^2}\\ &= \sum_{n=0}^{+\infty} \frac{(-1)^n}{2n+1}\sum_{k=0}^{+\infty} (-1)^k\frac{(2k+1)}{\left(2k+1 \right)^2 + \left(2n+1\right)^2} \end{split}$$ In other words $$\pi S = \sum_{n=0}^{+\infty} \sum_{k=0}^{+\infty} \frac{2k+1}{2n+1}\cdot \frac {(-1)^{k+n}}{\left(2k+1 \right)^2 + \left(2n+1\right)^2} \tag{1}$$ Now, using partial fractions, $$ \frac{2k+1}{2n+1}\cdot \frac {1}{\left(2k+1 \right)^2 + \left(2n+1\right)^2} = \frac{1}{2k+1}\left ( \frac 1 {2n+1} - \frac{2n+1}{\left(2k+1 \right)^2 + \left(2n+1\right)^2}\right) $$ We have $$\begin{split} \pi S & = \sum_{n=0}^{+\infty} \sum_{k=0}^{+\infty} \frac {(-1)^{k+n}}{(2k+1)(2n+1)} - \sum_{n=0}^{+\infty} \sum_{k=0}^{+\infty}\frac{2n+1}{2k+1}\cdot\frac {(-1)^{k+n}}{\left(2k+1 \right)^2 + \left(2n+1\right)^2}\\ &= \left(\sum_{n=0}^{+\infty}\frac {(-1)^{n}}{2n+1}\right)^2-\pi S \end{split}$$ where we identified $\pi S$ using $(1)$. In other words, $$\pi S = \frac{\pi^2}{16} - \pi S$$ and thus we conclude with the result: $$\boxed{S=\frac \pi {32}}$$

Edit: If you're curious, the classical proof for the "known pole expansion" of $\frac 1 {\cosh}$ is using Complex Analysis (Mittag-Leffler theorem). I personally prefer to use Fourier analysis:

  • Define $t\mapsto e^{iat}$ for $|t|<\pi$ and $a\notin \mathbb Z$.

  • Compute its Fourier series on $(-\pi, \pi)$: $$e^{iat} = \sum_{k\in\mathbb Z} \frac {(-1)^k} \pi \frac{\sinh(\pi a)}{ik+a}e^{int}$$

  • Evaluate at $t=0$ and rearrange a bit: $$\frac{\pi}{\sinh(\pi a)}=\sum_{k\in\mathbb Z}\frac{(-1)^k}{ik+a}$$

  • Finally, set $a=\frac i 2-\frac z \pi$, and multiply both sides by $\frac i \pi$: $$\begin{split} \frac{1}{\cosh(z)}&= \sum_{k\in\mathbb Z}\frac{(-1)^k}{iz+\left( k+\frac 1 2\right)\pi}\\ &=\sum_{k=0}^{+\infty}\frac{(-1)^k}{iz+\left( k+\frac 1 2\right)\pi } +\sum_{k=-\infty}^{-1}\frac{(-1)^k}{iz+\left( k+\frac 1 2\right)\pi}\\ &=\sum_{k=0}^{+\infty}\frac{(-1)^k}{iz+\left( k+\frac 1 2\right)\pi } +\sum_{p=0}^{+\infty}\frac{(-1)^{p+1}}{iz-\left( p+\frac 1 2\right)\pi} & \text{ with } p=-(k+1)\\ &= \sum_{k=0}^{+\infty}(-1)^k \left (\frac{1}{iz+\left( k+\frac 1 2\right)\pi}-\frac{1}{iz-\left( k+\frac 1 2\right)\pi}\right)\\ &= \sum_{k=0}^{+\infty}\frac{(-1)^k (2k+1)\pi}{z^2+\left( k+\frac 1 2\right)^2\pi^2} \end{split}$$

Stefan Lafon
  • 12,256
  • 11
  • 29
  • I'll have to look into this theorem further and try to apply it myself. So the source doesn't explicitly show the expansion for cosh, but I'll have to trust that what you obtained is correct. It's a method from complex analysis which I have very little experience in, so I'll likely be taking my time looking into this. Thank you! – Researcher R Dec 19 '23 at 05:57
  • 1
    Yes I verified it numerically. For the expansion, go back to the Wikipedia page and scroll down to the examples. I prefer to use the Fourier series method over the Complex Analysis one. – Stefan Lafon Dec 19 '23 at 05:59
  • nevermind, I got it to work when I used cosh. The infinite series evaluates to $$\pi S = \frac{\pi}{4}\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}\frac{1}{\cosh(\frac{(2n+1)\pi}{2})} = \frac{\pi^2}{32}$$ Here is the verification Thanks again! – Researcher R Dec 19 '23 at 06:15
  • 2
    (+1) Great answer! The expansion of $1/\cosh z$ follows from that of $\sec z$ on the Wikipedia through the substitution $z \to \mathrm{i} z$. – Gary Dec 19 '23 at 06:15
  • @StefanLafon I actually don't know how to do create either series representations of the function, but if you say the Fourier Series expansion is more preferential, I can try that instead. I'll look into it more and see if I can't replicate your results. Thanks again! – Researcher R Dec 19 '23 at 06:26
  • It's usually a matter of personal preference. I like Fourier. I haven't worked out the exact way for this expansion, it's possible I got the exact function to expand wrong, but I suspect it's either that one or $\sinh (at)$. Sorry if I'm not more helpful. Maybe @Gary knows. – Stefan Lafon Dec 19 '23 at 06:29
  • 1
    @Gary I just saw the latter half of your comment. Yes, sec(iz) would expand to $$\sum_{n=0}^\infty \frac{(-1)^n(2n+1)\pi}{(n+1/2)^2 \pi^2 -i^2z^2}$$ which would give us what Stefan arrived to. I'll have to look through additional literature for deriving the summations for the examples as I didn't find the process explained in the wiki to be didactic. Hopefully I can get it as this prove useful in the future – Researcher R Dec 19 '23 at 07:39
  • 1
    @ResearcherR I have edited my post to include a proof of the expansion, using Fourier series. I hope that helps. – Stefan Lafon Dec 20 '23 at 02:20
  • Where is $t$ in the Fourier series of $\mathrm{e}^{\mathrm{i} a t}$? – Gary Dec 20 '23 at 09:00
  • Good catch. Fixed. – Stefan Lafon Dec 20 '23 at 15:47
  • @StefanLafon Providing the Fourier series absolutely helped! I don't think I'd ever have gotten that on my own. Currently I'm digging into the other method of expansion and I'm stuck on trying to find the principal part of a Laurent Series as I'm terribly inexperienced with line integration. To clarify, $$p_a(z) = \sum_{k=-\infty}^{-1} a_k(z-c)^k$$ and the coefficients, $$a_k = \frac{1}{2\pi i} \oint_\gamma \frac{f(z)}{(z-c)^{k+1}}dz$$. If I figure this out, I'll add an additional answer using the pole expansion technique. – Researcher R Dec 21 '23 at 23:59
  • 1
    Hi! I'm finally back. I finally got around to familiarizing myself with Mittag-Leffler expansions and I found it to be simple for some cases, somewhat aggravating for others like $\csc(z)$. I'll be writing up a solution using that and writing out the derivation process for anyone who wants to see that worked out as well. I just wanted to comment that the trick with the partial fraction expansion which held onto the original sum and substituting it with $\pi S$ was very clever. I don't know if I would have ever realized that that double sum coincidentally was the same as what we started with. – Researcher R Jan 12 '24 at 02:01
  • It was a bit of luck. I tinkered with the sum and stumbled upon it. – Stefan Lafon Jan 12 '24 at 04:07
0

I just wanted to add this answer only to show the Pole expansion derivation for $\csc(z)$. So from what I've learned about it, the Mittag-Leffler expansion (which is how we get the pole expansion for all poles $(n+\frac{1}{2})\pi$) is the sum of all the principal parts of all possible Laurent Series of $f(z)$.

Using the fact that $\tag{1}\sec(z) = -\csc(z-\frac{\pi}{2})$ we can actually instead start with writing out all the principal parts of $\frac{1}{\sin(z)}$. This function has all of it's poles centered at $n\pi$ and all possible Laurent Series can be expressed by simply plugging in $n\pi$ into a normal Laurent Series. Doing so would give:

$$\frac{1}{z-n\pi}\frac{z-n\pi}{\sin(z)} =\frac{1}{z-n\pi}\frac{z-n\pi}{T_{n\pi}(\sin(z))} =\frac{1}{z-n\pi} \frac{z-n\pi}{\Big\{0+(-1)^n(z-n\pi)--0+-\frac{(-1)^n}{6}(z-n\pi)^3-0+\frac{(-1)^n}{120}(z-n\pi)^5--0+....\Big\}}$$

And performing long division and then multiplying the series yields the Laurent Series about all poles $n\pi$: $$\frac{1}{\sin(z)} = (-1)^n\{\boxed{...+0(z-n\pi)^{-3}+0(z-n\pi)^{-2}+1(z-n\pi)^{-1}}+0(z-n\pi)^{0}+\frac{1}{6}(z-n\pi)^{1}+0(z-n\pi)^{2}+\frac{7}{360}(z-n\pi)^{3}+...\}$$

And lastly we want all the principal parts which I have boxed and these are going to be the negative power pole terms only ($(z-n\pi)^{n<0}$), and thankfully we only have one nonzero term $\frac{(-1)^n}{z-n\pi}$, so our Mittag-Leffler expansion representation of $f(z)$ (which again, is the sum of all the principal parts which make up all possible Laurent series), is:

$$\tag{2}\frac{1}{\sin(z)} = \sum_{n=-\infty}^\infty \frac{(-1)^n}{z-n\pi}$$

And now we just have to use (1) from earlier and rewrite (2):

$$\tag{3}\frac{1}{\cos(z)} = -\sum_{n=-\infty}^\infty \frac{(-1)^n}{(z-\frac{\pi}{2})-n\pi}$$

And split up the sum such that $\sum_{-N}^N a_n = \sum_{n=0}^N a_n + \sum_{n=0}^{N-1}a_{-n-1} \approx \sum_{n=0}^N a_n + \sum_{n=0}^{N}a_{-n-1} \text{ for } N\rightarrow \infty$ to get:

$$\frac{1}{\cos(z)} = \sum_{0}^\infty \frac{(-1)^{n+1}}{(z-\frac{\pi}{2})-n\pi} + \frac{(-1)^{-n-1+1}}{(z-\frac{\pi}{2})-(-n-1)\pi} \\ =\sum_{0}^\infty (-1)^{n+1}\Big(\frac{1}{z-\frac{\pi}{2}-n\pi} + \frac{-1}{z+\frac{\pi}{2}+n\pi}\Big) \\ = \sum_{n=0}^\infty (-1)^n\frac{(2n+1)\pi}{(n+\frac{1}{2})^2\pi^2-z^2}$$

And finally $\cos(iz) = \cosh(z)$ and therefore:

$$\tag{4} \frac{1}{\cosh(z)} = \sum_{n=0}^\infty (-1)^n\frac{(2n+1)\pi}{(n+\frac{1}{2})^2\pi^2+z^2}$$


Now that we have an alternate way of representing our pesky hyperbolic trig function inside of our sum:

$$\tag{5} u(1/2,1/2) = \frac{1}{4}\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}\frac{1}{\sinh((n+\frac{1}{2})\pi)}$$

by plugging in (4) into (5) after changing our iteration variable to k and substitution of z:

$$\tag{6} u(1/2,1/2) = \frac{1}{4}\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}\sum_{k=0}^\infty (-1)^k\frac{(2k+1)\pi}{(k+\frac{1}{2})^2\pi^2+(n+\frac{1}{2})^2\pi^2}$$

And from here it should be fairly simple what to do next, being combine the sums $\sum_n a_n \sum_k a_k = \sum_n\sum_k a_n a_k$, factor out $\pi$, and the tricky part would be to keep a mental note of this sum and compare it to what it looks like after performing a partial fraction expansion. After that it become clear what happens next.

Starting with combining everything and noting that $a^n(b+c)^n = (ab+ac)^n$:

$$\tag{7} \boxed{S = \frac{1}{\pi}\sum_{n=0}^\infty \sum_{k=0}^\infty \frac{2k+1}{2n+1}\frac{(-1)^{n+k}}{(2k+1)^2+(2n+1)^2}}$$

Partial fraction expansion of $\frac{A}{B(A^2+B^2)} = \frac{1}{AB}-\frac{B}{A(A^2+B^2)}$ and splitting up $\sum_n\sum_k \frac{1}{AB}-\frac{B}{A(A^2+B^2)}$ into two sums yields:

$$S = \frac{1}{\pi}\tag{8}\sum_{n=0}^\infty\sum_{k=0}^\infty \frac{(-1)^{n+k}}{(2n+1)(2k+1)} - \boxed{\frac{1}{\pi}\sum_{n=0}^\infty \sum_{k=0}^\infty \frac{2k+1}{2n+1}\frac{(-1)^{n+k}}{(2k+1)^2+(2n+1)^2}}$$

Notice how the boxed term is exactly the same as (7), so if we were to substitute this sum with S as Stefan Lafon suggested in their solution, we would get:

$$ S = \frac{1}{\pi}\sum_{n=0}^\infty\frac{(-1)^n}{2n+1}\sum_{k=0}^\infty\frac{(-1)^k}{2k+1} - S \\ 2S = \frac{1}{\pi} \arctan(1)^2 \\ S = \frac{1}{2\pi}\Big(\frac{\pi}{4}\Big)^2 = \frac{\pi}{32}$$

And therefore we can say that:

$$u(1/2,1/2) = \frac{1}{4}\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}\frac{1}{\sinh\big((n+1/2)\pi\big)} = \frac{\pi}{32}$$

So it looks like the trick with dealing with hyperbolic trig functions (at least in this example) is by... not evaluating them, we basically cancel out and get rid of it.

If by chance you came to this question trying to evaluate the solution to a PDE (partial differential equation, specifically Poisson's Equation) like I did, I recommend heading to THIS QUESTION, it might prove useful.