15
$\begingroup$

I would like to prove that

$$\int_{0}^{\infty} \cos(\omega x) \exp(-x^{\alpha}) \, {\rm d} x \ge {\alpha^2 \sqrt{\pi} \over 8} \exp \left( -\frac{\omega^2}{4} \right)$$

for any $\omega > 0$ and $1 < \alpha < 2$.

Here is some research effort. By using the representation of the generalized Gaussian density ($c_\alpha\exp(-|x|^{\alpha})$, $c_{\alpha}>0$ normalizing constant) as Gaussian mixture, I can prove that $$\int_{0}^{\infty} \cos(\omega x) \exp(-x^{\alpha}) \, {\rm d} x \ge {\Gamma(1/\alpha) \over \alpha} \exp \left( -\frac{\omega^2}{2} {\Gamma(3/\alpha)\over \Gamma(1/\alpha)} \right).$$

When $\alpha=2$, ${\Gamma(3/\alpha)\over 2 \Gamma(1/\alpha)}={1\over 4}$ and $ {\Gamma(1/\alpha) \over \alpha}= {\alpha^2 \sqrt{\pi} \over 8} $ and $\mbox{Var}(\xi)= {\Gamma(3/\alpha)\over\Gamma(1/\alpha)}$ when $ \xi \sim c_\alpha\exp(-|x|^{\alpha})$.

Both ${\Gamma(3/\alpha)\over 2 \Gamma(1/\alpha)}$ and $ {\Gamma(1/\alpha) \over \alpha}$ are decreasing in $\alpha \in (1,2)$. I need the result as originally stated as the coefficient in front of $\omega$ in the exponent has to be free from $\alpha$. All of my attempts are tied to probablisitic arguments and hence I fail to get rid of ${\Gamma(3/\alpha)\over 2 \Gamma(1/\alpha)}$ coefficient.


Related question

$\endgroup$
11
  • $\begingroup$ It is false for $\omega=0$, $\alpha=2$. $\endgroup$ Commented Jul 22, 2020 at 17:22
  • $\begingroup$ I presume your $dc$ should be $dx$; the inequality is badly violated; say for $\omega=1$ the l.h.s. and r.h.s. cross at $\alpha=1.4$. $\endgroup$ Commented Jul 22, 2020 at 17:22
  • 1
    $\begingroup$ this inequality does hold ($\omega>0$, $1<\alpha<2$): $$\int_{0}^{\infty}\cos(\omega x)\exp(-x^{\alpha})dx \ge {\alpha^2 \sqrt{\pi} \over 8} \exp(-\omega^2/4) .$$ Equality is reached at $\alpha=2$. $\endgroup$ Commented Jul 22, 2020 at 17:41
  • $\begingroup$ @CarloBeenakker, yes, thank you, it should be $\sqrt{\pi}$- I have edited the question $\endgroup$ Commented Jul 22, 2020 at 17:44
  • $\begingroup$ More precise bounds are known; see e.g. Chen, Z-Q., Kim, P. and Song, R., Heat kernel estimates for the Dirichlet fractional Laplacian, 2010. $\endgroup$
    – user90189
    Commented Jul 23, 2020 at 6:52

0

You must log in to answer this question.