2
$\begingroup$

I am reading the High-Dimensional Probability by Dr.Roman Vershynin , where I stuck on some statement at page 28. where state as below:

Any bounded random variable $X$ is sub-gaussian with: $$\newcommand\norm[1]{\left\lVert#1\right\rVert} \norm{X}_{\psi_2}\leq \frac{\norm{X}_{\infty}}{\sqrt{log2}} $$

where $\newcommand\norm[1]{\left\lVert#1\right\rVert} \norm{X}_{\psi_2}$ is the sub-gaussian norm define as:

$$\newcommand\norm[1]{\left\lVert#1\right\rVert} \norm{X}_{\psi_2} =inf \left\{ t>0 : \mathbb{E} \exp{(\frac{X^2}{t^2}) \leq 2} \right\} $$

where $\newcommand\norm[1]{\left\lVert#1\right\rVert} \norm{X}_{\infty} :=( \mathbb{E} |X|^p)^{1/p}$ as $p \to \infty$

I can see how why the bounded random variable is sub-gaussian (hoeffing lemma ),but How could I see this upper bound of sub-gaussian norm?

$\endgroup$
  • $\begingroup$ What is $p$? I think there is something wrong with the definition of $\|X\|_{\infty}$ $\endgroup$ – Kavi Rama Murthy Sep 16 '18 at 0:32
  • $\begingroup$ @KaviRamaMurthy sorry I fix that $\endgroup$ – ShaoyuPei Sep 16 '18 at 0:35
2
$\begingroup$

If $t=(\sqrt {\log 2})^{-1} \|X\|_{\infty}$ then $Ee^{\frac {X^{2}} {t^{2}}} \leq e^{\log 2}= 2$ and hence $\|X\|_{\psi_2} \leq t$. I have used the fact that $\|X\|_{\infty}$ is nothing but the essential supremum of $X$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.