6
$\begingroup$

I am a beginner in complex analysis. I have come across a problem that, in essence, asks me to find the residue of $e^{\frac{1}{\sin z}}$ at the isolated essential singularity $z=0$.

Until now, I have only seen problems where the sine is multiplied or divided to an analytic function, in which case it sufficed to expand the Taylor series of $\sin z$, or even consider only the linear term of it.

However, in this case, $\sin z$ is not only inverted but also exponentiated, which made it too difficult for me to directly write the Laurent series of the function. I briefly tried writing $$ \csc z = \sum_{n=-1}^{\infty} a_n z^n$$ because it has a simple pole at $z = 0$, but exponentiating it lead to a representation of the residue as an infinite sum including $a_n$, which I had no idea how to calculate using the inductive formlua for $a_n$.

I would like to ask what kind of techniques can be used in such problems, when complicated formulae are composed and inverted and such, making a direct derivation of Laurent series inapplicable.

Thank you in advance.

$\endgroup$
2
  • 4
    $\begingroup$ This is actually pretty difficult. I tried this for a while before giving up. Though, I did find math.stackexchange.com/questions/120282/… which basically is a very similar problem 👀 hopefully you can salvage a series answer from the answer there $\endgroup$
    – Max0815
    Commented Sep 27, 2022 at 4:08
  • 1
    $\begingroup$ @Max0815 Thanks a lot!! I just solved it (with a series answer; no surprises there...). I'm now writing my answer to post it here!! $\endgroup$
    – 이희원
    Commented Sep 27, 2022 at 7:54

1 Answer 1

5
$\begingroup$

Okay, based on Max0815's suggested link, along with a handful of Wikipedia pages, I came up with a somewhat self-contained solution to this problem for anyone unfamiliar with ring and/or field theory.

The main idea is to use the Lagrange Inversion formula, which I frankly still don't fully understand. But let me show only its application for this problem.

Consider any function $f$ that is analytic on some deleted neighborhood of $z=0$. It can be expanded as the Laurent series $$ f(z) = \sum_{n = -\infty} ^ {\infty} a_n z^n.$$ Let us denote the coefficient of $x^n$ as $\left[ z^n \right] f(z)$, which is determined uniquely since $f$ is analytic. We clearly see that for any integers $n$ and $k$, $$\left[ z^n \right] f(z) = \left[ z^k \right] \left( z^{k - n} f(z) \right). $$


The first goal of the proof is to show that for any integers $k$ and $n$, $$ k \cdot \left[ z^k \right] \left( \sin^{-1} z \right)^n = n \cdot [z^{-n}] \left( \sin z \right)^{-k}. $$ For the record, since $\sin^{-1}$ is multivalued, we choose the branch of $Re\sin^{-1} z \in \left( -\frac{\pi}{2}, \frac{\pi}{2} \right]$. (Not that it matters for our purposes...)

From the above discussion, we have $$ k \cdot \left[ z^k \right] \left( \sin^{-1} z \right)^n = k\cdot \left[ z^{-1} \right] \left( z^{-k-1} \left( \sin^{-1} z \right)^n \right). $$ Let $g(z):=z^{-k-1} \left( \sin^{-1} z \right)^n$. It clearly has a convergent Laurent series near $z=0$, which may or may not include a $z^{-1}$ term. Let $g_{-1}:=\left[ z^{-1} \right]g(z)$. Then, the Laurent series for $g(z)-\frac{g_{-1}}{z}$ can be termwise integrated to yield a new function $G(z)$, analytic on a deleted neighborhood of $0$.

Rewriting $g(z)=\frac{g_{-1}}{z} + G'(z)$, we see that $$ g(\sin z) \cdot \cos z = g_{-1} \cot z + (G(\sin z))'. $$ It is straightforward to calculate that $\left[ z^{-1} \right] \cot z = 1$. As for the latter, notice that the derivative of no integer power of $z$ yields a $z^{-1}$ term. Hence, we have $$ \left[ z^{-1} \right] (G(\sin z))' = 0 ~\Rightarrow~ \left[ z^{-1} \right] \left( g(\sin z) \cos z \right) = g_{-1} = \left[ z^{-1} \right] g(z)$$ $$ \begin{equation} \begin{split} \therefore k\cdot \left[ z^{-1} \right] \left( z^{-k-1} \left( \sin^{-1} z \right)^n \right) & = k\cdot \left[ z^{-1} \right] \left( (\sin z)^{-k-1} z^n \right) \\ & = -\left[ z^{-1} \right] \left( z^n \left( \left( \sin z \right)^{-k} \right)' \right) \end{split} \end{equation} $$

Recall from before that the derivative of any analytic function cannot have any $z^{-1}$ terms. Then, it follows that $$ \left[ z^{-1} \right] \left( z^n \left( \sin z \right)^{-k} \right)' = 0. $$ Therefore, by chain rule, we have $$ \begin{equation} \begin{split} -\left[ z^{-1} \right] \left( z^n \left( \left( \sin z \right)^{-k} \right)' \right) & = \left[ z^{-1} \right] \left( \left( z^n \right)' \left( \sin z \right)^{-k} \right) \\ & = \left[ z^{-1} \right] \left( nz^{n-1} \cdot \left( \sin z \right)^{-k} \right) \\ & = n \cdot [z^{-n}] \left( \sin z \right)^{-k} \end{split} \end{equation} $$

Therefore, connecting the results, we have the desired inversion formula $$ k \cdot \left[ z^k \right] \left( \sin^{-1} z \right)^n = n \cdot [z^{-n}] \left( \sin z \right)^{-k}. $$


Notice that, by definition, $ Res_{z = 0} f(z) = \left[ z^{-1} \right] f(z) $. Thus, the residue that we want is $$ \begin{equation} \begin{split} Res_{z=0} e^{\frac{1}{\sin z}} & = \left[ z^{-1} \right] e^{\frac{1}{\sin z}} \\ & = \left[ z^{-1} \right] \sum_{k=0}^{\infty} \frac{1}{k!} \cdot \left( \frac{1}{\sin z} \right)^k \\ & = \sum_{k=1}^{\infty} \frac{1}{k!} \cdot 1 \cdot \left[ z^{-1} \right] \left( \sin z \right)^{-k} \\ & = \sum_{k=1}^{\infty} \frac{1}{k!} \cdot k \cdot \left[ z^k \right] \left( \sin^{-1} z \right)^{1} \\ & = \sum_{k=1}^{\infty} \frac{1}{(k-1)!}\left[ z^k \right] \sin^{-1} z \end{split} \end{equation} $$ Noticing $\left( \sin^{-1} z \right)' = \left( 1 - z^2 \right)^{-1/2}$ and termwise integrating the Taylor series of the right hand side, we get $$ \begin{equation} \begin{split} \sin^{-1} z & = \sum_{k=0}^\infty \frac{(-1)^k}{2k+1} \begin{pmatrix} -1/2 \\ k \end{pmatrix} z^{2k+1} \\ & = \sum_{k=0}^\infty \frac{(2k)!}{\left( 2^k \cdot k! \right)^2 \cdot (2k+1)} z^{2k+1} \end{split} \end{equation} $$

Finally, substituting it into our residue result, we get the final answer $$ \begin{equation} \begin{split} Res_{z=0} e^{\frac{1}{\sin z}} & = \sum_{k=0}^{\infty} \frac{1}{((2k+1)-1)!} \frac{(2k)!}{\left( 2^k \cdot k! \right)^2 \cdot (2k+1)} \\ & = \sum_{k=0}^{\infty} \frac{1}{\left( 2^k \cdot k! \right)^2 \cdot (2k+1)} \end{split} \end{equation} $$


Any details I glossed over or criticisms are welcomed. Thank you again to Max.

$\endgroup$
2
  • 1
    $\begingroup$ Just as a side note, your residue value can be expressed in closed form with a hypergeometric function or bessel/struve functions wolframalpha.com/… $\endgroup$ Commented Jan 17 at 5:18
  • 1
    $\begingroup$ @CaptainChicky Oh wow, I didn't even consider putting it to WolframAlpha. It seems hypergeometric functions appear just about everywhere :O $\endgroup$
    – 이희원
    Commented Jan 17 at 10:06

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .