Suppose that $X\sim N(0,1)$ is a standard Gaussian and $Y\sim N(0,\sigma^2)$ is Gaussian with variance $\sigma^2>1$. Informally, since $Y$ has a larger variance than $X$, it is more spread out than $X$, and thus we expect that for a typical realization, $|Y|$ should be larger than $|X|$.

If we want a more rigorous statement of this fact, we can for example prove that $|Y|$ stochastically dominates $|X|$, that is, $$P[|Y|\leq t]=P[|X|\leq t/\sigma]\leq P[|X|\leq t]$$ for all $t$.

Question. Suppose now that $X\sim N(0,\Sigma_1)$ and $Y\sim N(0,\Sigma_2)$ are Gaussian vectors of the same dimension with $0<\det(\Sigma_1)<\det(\Sigma_2)$. Is there an easy way to see (if even true) that $\|Y\|_2$ stochastically dominates $\|X\|_2$?


It's not true at all. Try $\Sigma_1 = \pmatrix{r&0\cr0 & 0}$ while $\Sigma_2 = I$, where $r$ is very large. $\det(\Sigma_1) = 0 < \det(\Sigma_2)$, but $\|X\|_2$ is very large with high probability.

  • $\begingroup$ Good catch. I've edited the question so as to avoid such examples when the two Gaussians don't really have the same dimension. $\endgroup$ – user78270 Dec 8 '17 at 2:50
  • 2
    $\begingroup$ So let $\Sigma_1=\pmatrix{r&0\\0&1/(2r)}$. $\endgroup$ – kimchi lover Dec 8 '17 at 3:03

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.