2

As mentioned in another post, as a consequence of Mittag-Leffler's theorem combined with the Weierstrass factorization theorem, after reducing to the common denominator, any meromorphic function can be expressed as

$$f(z) = e^{g(z)}\prod_k(z-z_k)^{n_k} \tag{1}\label{1}$$

where $n_k > 0$ ($n_k<0$) denotes the multiplicity of a zero (pole) at $z_k$ and $g(z)$ is an entire function.

This can be reformulated and generalized into

$$\begin{align} f(z) &= \exp\left[g(z) + \sum_k n_k\ln(z-z_k)\right] \\ &\to \exp\Big[g(z) + \underbrace{\int_\mathbb C n(y)\ln(z-y)\,dy}_{=(n\ast_\mathbb C\ln)(z)}\Big] \tag{2}\label{2} \end{align}$$

where $n(z)$ is a strictly integer-valued distribution for meromorphic functions, but could in general also be something else, e.g. a meromorphic function itself. Due to its origin I'd like to refer to $\eqref{2}$ as the Weierstrass-Mittag-Leffler-Transformation (or WeMiLe-Transformation, if you allow for that acronym). So now my question is

How can the transformation $\eqref{2}$ be inverted?

For a meromorphic $f(z)$ with finitely many zeros of finite order that inversion is obviously $n(y) = \sum_k n_k \delta(y-z_k)$ and $g(z)=\ln\big[f(\zeta)/\prod_k(\zeta-z_k)\big]$ for any $\zeta\in\mathbb C\backslash\{z_k\}$, i.e. one has to determine all zeros and poles, but what about non-meromorphic $f(z)$, e.g. what about $f(z)=\delta(z)$?

1 Answers1

2

first attempt at a partial answer, to be improved...

I've noticed the exponent in $\eqref{2}$ is proportional to the antiderivative of the Hilbert-Transform of $n(z)$ if all zeroes and poles lie on the real axis, so formally this means

$$n(z) = -\frac1\pi\mathcal H \frac{d}{dz}\ln\frac{f(z)}{c} = -\frac1\pi\mathcal H \frac{f'(z)}{f(z)} \tag{n}\label{n}$$

where $\mathcal H$ denotes the Hilbert transform $\mathcal H f(z) = \Big(\frac1{\pi z}\ast f\Big)(z)$. But how valid is this really?

Ok, let's have some examples:

$f(z) = c$

Clearly $f'(z)=0=n(z)$

$f(z) = z - z_0$

So $f'(z) = 1$ and therefore \begin{align} n(z) &= -\frac1\pi\mathcal H \frac1{z-z_0} \\ &= \delta(z-z_0) \end{align}

So far, so good, though I didn't require $z_0\in\mathbb R$ here at first glance...

to do: do some more examples

  • I'm commenting here as to not clutter the other question, and because annoyingly we have no message system in mse... anyway, I put a bounty on the Laurent ring question, we'll see if anything comes of it. I am not well-versed in Mittag-Leffler yet, it wasn't in my coursework as a student and I haven't taught it properly yet... maybe later this semester more will shake loose in my dense head. – James S. Cook Oct 28 '14 at 00:12
  • @JamesS.Cook hey thanks :) yeah, SE persistently refuses to implement any more social features, though that might be justified in order to protect regular users from "vampires" anyway, at least in my case external means of communication can be found via my gaming.se name on Facebook... – Tobias Kienzler Oct 28 '14 at 08:40
  • @JamesS.Cook So, how'd the semester go? ;) – Tobias Kienzler Sep 13 '15 at 18:46