I have a question, if we do a non linear transformation on Gaussian random vector, will it give us Gaussian as a result? If No which techniques can we use to make sure the result is finally gaussian.
Thanks for your time
I have a question, if we do a non linear transformation on Gaussian random vector, will it give us Gaussian as a result? If No which techniques can we use to make sure the result is finally gaussian.
Thanks for your time
Copyright © 2021 JogjaFile Inc.
As mentioned, if the transformation $g$ is nonlinear, then $g(X)$ is in general not Gaussian. (Practically any nonlinear $g$ will give you an immediate counterexample.)
But if $g$ is smooth enough (so that locally it is approximately linear), and the variance of $X$ is small (so that its values are concentrated in a small neighborhood), then the distribution of $g(X)$ will be approximately Gaussian. This is a special case of the delta method, which you can read for a precise statement. The Jacobian of $g$ will help compute the variance of $g(X)$.