Suppose the random variables $X_i$ are independent and satisfy $E[X_i] = 0$. Then the following inequality holds: $$E\left[\left(\sum \limits_{i = 1}^n X_i\right)^4\right] = \sum \limits_{i = 1}^n E[X_i^4] + 3\sum \limits_{i \ne j} E[X_i^2X_j^2] \le 3\left(\sum \limits_{i = 1}^n E[X_i^4] + \sum \limits_{i \ne j} \sqrt{E[X_i^4]}\sqrt{E[X_j^4]}\right) = 3\left(\sum \limits_{i = 1}^n \sqrt{E[X_i^4]}\right)^2$$

Now I would like to generalize this inequality to moments of order 6 [or higher]. My best guess is that the inequality

$$E\left[\left(\sum \limits_{i = 1}^n X_i\right)^6\right] \le C \left(\sum \limits_{i = 1}^n E[X_i^6]^{1/3}\right)^3$$

holds, where $C$ is an absolute constant.

Can someone provide a proof/ source for the last inequality? Or perhaps a source for a similar inequality?

The "similar inequality" should provide at least $E\left[\left(\sum \limits_{i = 1}^n X_i\right)^6\right] = O(n^3)$ in the i.i.d. case.


The inequality is actually correct. It follows from Theorems 3 in a Rosenthal paper from 1970 (Israel J. Math. 8, 273-303).

Rosenthal proves the following inequality for $p > 2$ and independent, centered Random variables $X_i \in L^p$: $$E\left[\left|\sum \limits_{i = 1}^n X_i\right|^p\right] \le C(p) \max\left\{\sum \limits_{i = 1}^n E[ \left|X_i\right|^p], \left(\sum \limits_{i = 1}^n E[X_i^2]\right)^{p/2}\right\}.$$

The inequality in my original post follows from two applications of Jensen's inequality: $$\left(\sum \limits_{i = 1}^n E[\left|X_i\right|^p]\right)^{2/p} = n^{2/p} \left(\sum \limits_{i = 1}^n \frac{1}{n}E[\left|X_i\right|^p]\right)^{2/p} \le n^{2/p - 1} \sum \limits_{i = 1}^n E[\left|X_i\right|^p]^{2/p} \le \sum \limits_{i = 1}^n E[\left|X_i\right|^p]^{2/p}$$ $$E[X_i^2] \le E[\left|X_i\right|^p]^{2/p}$$


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.