Suppose that vector $x \in \mathcal{R}^n$, and $x$ conforms to the standard gaussian distribution $\mathcal{N}(0, I_n)$. Can we calculate $E(\|x\|^2_2)$? Can we further extend to the case of $E(\|x\|^p_2)$, when $p\in \mathcal{N}, p\geq 1$?

Note that there is already a link for the $p=1$ case, i.e., Average norm of a N-dimensional vector given by a normal distribution

  • $\begingroup$ Is this supposed to be the $p$-th power of the $2$-norm or just the $p$-norm? Of maybe the $p$-th power of the $p$-norm? Your formula and the title are conflicting on this. $\endgroup$ – Dominik Dec 20 '16 at 9:18
  • $\begingroup$ Actually p-th power euclidean norm. $\endgroup$ – Zhuanghua Liu Dec 20 '16 at 9:39

If $X$ is a standard $n$-dimensional Gaussian random variable, then $\|X\|_2^2 \sim \chi_n^2$. The chi-squared distribution has density $$2^{-n/2} \Gamma(n/2)^{-1} x^{n/2 - 1} e^{-x/2}I\{x > 0\} \, dx.$$

With this we can find an explicit formula for the expectation of $\|X\|_2^p$ $$E[\|X\|_2^p] = E[(\|X\|_2^2)^{p/2}] = \int_0^\infty 2^{-n/2} \Gamma(n/2)^{-1} x^{(p + n)/2 - 1} e^{-x/2} \, dx = 2^{-n/2} \Gamma(n/2)^{-1} 2^{(p + n)/2} \Gamma((p + n)/2) = 2^{p/2} \frac{\Gamma\Big(\frac{p + n}{2}\Big)}{\Gamma\Big(\frac{n}{2}\Big)}.$$

To get approximations and inequalities, you just need to apply some approximations for the gamma-function. A standard reference for this would be Abramowitz and Stegun.


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.