Let $X_1,\ldots, X_n$ be $n$ i.i.d. copies of Gaussian random variable $X \sim N(0, \sigma^2)$. It is known that \begin{align} \mathbb{P}\Bigl( \Bigl|\frac{1}{n}\sum_{j=1}^n X_j \Bigl| >t\Bigr) &\leq 2 \exp( cnt^2)~~\text{and}\\ \mathbb{P}\Bigl( \Bigl|\frac{1}{n}\sum_{j=1}^n (X_j^2 - \mathbb{E}X_j^2) \Bigl| >t\Bigr) &\leq 2 \exp( cn\min \{t^2, t\}). \end{align} These two results follows from the concentration inequality of sub-Gaussian and sub-exponential random variables and the second one is a Bernstein-type inequality. I wonder if there are similar results for the higher moments of Gaussian random variables. Can one have \begin{align} \mathbb{P}\Bigl( \Bigl|\frac{1}{n}\sum_{j=1}^n (X_j^4 - \mathbb{E}X_j^4) \Bigl| >t\Bigr) \leq 2 \exp( cnt)? \end{align} And more generally, for a centered random variable $Y$ satisfying $\mathbb{P}(|Y| > t) \leq 2 \exp(c t^{\alpha})$ for $\alpha > 0$, can we get an exponential inequality for the concentration of $\sum_{i=1}^n Y_i$? That is, can we have \begin{align} \mathbb{P}\Bigl( \Bigl|\frac{1}{n}\sum_{j=1}^n (Y_j - \mathbb{E}Y_j) \Bigl| >t\Bigr) \leq 2 \exp( cnt^\beta) \end{align} for some $\beta > 0$?


1 Answer 1


See Theorem 23 in Section 9.3 of Ryan O'Donnell's book Analysis of Boolean functions. Even though the theorem there is stated for $\pm 1$ variables, it holds for Gaussians as well (see Chapter 10 of the book for details on that).

For the more general statement, see Exercise 7 in these lecture notes of Terry Tao.

  • $\begingroup$ Sorry the previous statement of the question is not correct. Now I've made some corrections. And what can we say for more general random variables with some decaying tail behavior? $\endgroup$
    – Steve
    Nov 22, 2015 at 10:55
  • $\begingroup$ Your more general question is addressed in lecture notes of Terry Tao. $\endgroup$ Nov 22, 2015 at 13:38

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.