For a random variable $X$, I have a large deviation inequality of the form \begin{equation} P(|X-\mathbb EX|\geq r)\leq ce^{-\alpha r}\,. \end{equation}

Consider a sample mean $S_n=\frac{1}{n}(X_1+X_2+\dots+X_n)$. I would to obtain a central limit theorem of the form $\sqrt n(S_n-\mu)\equiv N(0,\sigma^2)$. Is this possible?

It seems that I need to know the variance of $X$ to apply the Lindeberg–Lévy CLT. I don't see how I can get the variance of $X$ from its tail bound.

  • $\begingroup$ Typed a more relevant title. $\endgroup$ – Did Aug 28 '18 at 19:09

You have, for a non-negative random variable $Z$, $$ \mathbb{E}[Z] = \int_{0}^\infty \mathbb{P}\{Z \geq z\}dz \tag{1} $$ from which you can bound the variance of $X$: $$ \operatorname{Var} X = \mathbb{E}[(X-\mathbb{E}[X])^2] = \int_{0}^\infty \mathbb{P}\{(X-\mathbb{E}[X])^2 \geq z\}dz = \int_{0}^\infty \mathbb{P}\{\lvert X-\mathbb{E}[X]\rvert\geq \sqrt z\} dz\tag{2} $$ and using your concentration bound you thus obtain $$ \operatorname{Var} X \leq \int_{0}^\infty c e^{-\alpha \sqrt{z}}dz = \frac{c}{\alpha^2} \tag{3} $$ Does that suffice for your purposes?

  • $\begingroup$ yes, that helps. thank you! $\endgroup$ – Alice Schwarze Aug 28 '18 at 19:03
  • $\begingroup$ @AliceSchwarze You're welcome! $\endgroup$ – Clement C. Aug 28 '18 at 19:05

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.