Properties of this Fourier Transform

82 Views Asked by At

Let $\phi$ be the following function $$ \phi (\omega, n) \equiv \left( 4 e^{\gamma - 2} \right)^{\sqrt{n} \, i \omega / \Sigma} \left[ \frac{2}{\sqrt{\pi}} \Gamma \left( \frac{3}{2} + \frac{i \omega}{\sqrt{n} \, \Sigma} \right) \right]^n $$ where $\omega$ is a continuous variable and $n$ is a parameter. $e, \pi, \gamma$ are the Euler's number, Ludolph's constant and the Euler-Mascheroni constant, respectively. The capital $\Sigma$ is defined via $\Sigma^2 = \pi^2 / 2 - 4$, thus, also just a numerical parameter.

Now we define the following Fourier transform $$ f (x, n) \equiv \frac{1}{2 \pi} \int \limits_{- \infty}^\infty \mathrm{d} \omega \, e^{- i \omega x} \phi (\omega,n) $$

It's not super hard to prove, that as $n \to \infty$, $f (x, n) \to \mathcal{N} (x; \mu = 0, \sigma = 1)$ (normal distribution with zero mean and unit variance).

  1. As a first order business, I would like to prove, that the integral defining $f$ results in a valid distribution only when $n$ is an integer. What I know: the integral of $f$ over the whole real line is always 1, regardless of $n$. What I suspect: $f$ is strictly positive only when $n$ is a positive integer. I don't know how to prove it.

  2. Since computing the integral numerically (there's no other way for general $n$, as far as I'm aware) is expensive (especially in the tails, where greater numerical precision is needed), I'd like to find the asymptotic/limiting behavior of $f$ for $x \to \pm \infty$, for any integer $n$ (assuming we already proved the previous point). That way, the numerical integration gives the main shape, while the tails can be extrapolated with the asymptotic forms.

  • For example, we know immediately, that for $n \to \infty$, the function is the normal distribution $ \sim e^{-x^2/2}$ (for both $x \to \pm \infty$), which also gives its asymptotic behavior.

  • The integral is also computable for $n = 1$ and it's asymptotic behavior is much more complicated:

$$ f(x \to - \infty, 1) \sim \frac{\Sigma (e^{2 - \gamma} / 2)^{3/2}}{\sqrt{2 \pi}} \exp \left( \frac{3 \Sigma}{2} x \right) $$ $$ f(x \to + \infty, 1) \sim C \exp \left( - \frac{e^{2 - \gamma}}{4} e^{\Sigma \, x} \right) $$ with $C$ being a constant I have had trouble to determine.

Can an analogous analysis be done for any integer $n \geq 2$?