2
$\begingroup$

If $X_1, X_2, \dots, X_M$ are independent identically distributed (i.i.d.) normal random variables, how to calculate the PDF and mean of the Euclidean norm of $X_1,X_2,\ldots,X_M$?

$$X_i\sim N(0,\sigma^2)$$

$$Y=\sqrt{\sum_{i=1}^M X_i^2}$$

There is a similarity between $Y$ and the square root of the $\chi$-squared distribution, i.e., the sum of the squares of $k$ independent standard normal random variables; however, in the case of $Y$, $\sigma$ may not necessarily be $1$.

$$f_Y(y) = \text{?}$$

$$E[Y] = \text{?}$$

$\endgroup$
2
$\begingroup$

The fact that the variance of the normals is not one is immaterial. You could just compute everything for $\sigma=1$ and then rescale $Y\to \sigma Y$ at the end. Recall that if $X$ is $N(0,\sigma^2)$ then $X/\sigma$ is $N(0,1).$ So $Y/\sigma = \sqrt{\sum _i(X_i/\sigma)^2}$ which is $\chi^2(M).$

So you just need to compute square root of a $\chi^2(M).$ This is a usual change of variables. So let $Z$ be a $\chi^2(M)$ and let $Y = \sigma\sqrt{Z}$ where I included the scaling factor. Then you can write the cumulative distribution for $Y$ as $$ F_Y(y) = P(Y\le y) = P(\sigma\sqrt Z < y) = P(Z<y^2/\sigma^2) = F_Z(y^2/\sigma^2)$$ where $F_Z$ is the CDF for a $\chi^2(M).$ To get the PDF, take the derivative: $$ f_Y(y) = F'_Y(y) = \frac{2y}{\sigma^2}F'_Z(y^2/\sigma^2) = \frac{2y}{\sigma^2}f_Z(y^2/\sigma^2)$$ where $f_Z$ is the PDF for a $\chi^2(M).$

The expectation can be computed by either integrating the above PDF or computing the expectation of the square root of a chi squared. Remember, you don't need to worry about the $\sigma.$ The answer's going to be proportional to it... just do the computation with $\sigma = 1$ and then scale by $\sigma$ at the end.

$\endgroup$

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.