Let $X\in\mathbb{R}^n$ be a zero-mean Gaussian random vector with covariance matrix $\Sigma\in\mathbb{R}^{n\times n}$. And let $Y = \text{sign}(X)\in\{-1,1\}^n$.

Has anyone studied this object $Y$? Does it have a name, or does it show up in some branch of probability? I'm wondering if there's a characterization of the joint density of $Y$ (i.e., suppose $v\in\{-1,1\}^n$, then what is $\mathbb{P}(Y=v)$) in terms of the covariance matrix $\Sigma$. Thanks!


You could call $Y$ a generalized multi-variate indicator function that gives values $-1$ or $1$ on the different parts of the support. Note you have to choose which sign to assign to $0$ (however, a single point that is not a point mass doesn't change the values of any integrals).

In general there is no closed form for $P(X = v)$. However you can express it as an integral of the Gaussian density over an orthant. You can estimate such an integral efficiently with sampling.

  • $\begingroup$ Yes, I see. So, for example for $n=2$, $\mathbb{P}(Y_1=1,Y_2=1) = \mathbf{1}(X_1>0,X_2>0) = \mathbb{P}(X_1>0,X_2>0)$, which is the integral of the joint Gaussian density over the top right orthant. Thanks! $\endgroup$ – paulcon Oct 5 '15 at 0:25

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.