Suppose I have a multivariate Gaussian distribution x and a constant matrix A. I know how to calculate the mean and covariance of Ax but how can I prove that Ax will also be multivariate gaussian??
Asked
Active
Viewed 1,153 times
2
-
1@Gue : I know how to find both mean and co variance of the new distribution, but how can I show that new distribution will also be gaussian – Sahil Chadha Jun 28 '19 at 09:17
-
1An easy way to prove this is by using characteristic functions, see e.g. https://math.stackexchange.com/questions/605816/proof-of-the-affine-property-of-normal-distribution-for-a-landscape-matrix – sp59b2 Jun 28 '19 at 09:57
-
There are myriad ways to establish this: see [our list of characterizations of Gaussian distributions](https://stats.stackexchange.com/questions/4364) for some ideas. – whuber Jun 28 '19 at 14:10
1 Answers
2
In addition to characteristic functions, one can use Jacobians to find an expression for RV transformations, i.e. when we set $\mathbf{Y}=\mathbf{AX}+\mathbf{b}$, it's further simplified to: $$f_\mathbf{Y}(\mathbf{y})=\frac{1}{\vert\mathbf{A}\vert}f_\mathbf{X}(\mathbf{A}^{-1}\vert\mathbf{y}-\mathbf{b}\vert)$$
Since $\vert\mathbf{A}\vert$ is constant, $f_{\mathbf{Y}}(y)\propto f_\mathbf{X}(\mathbf{A}^{-1}\vert\mathbf{y}-\mathbf{b}\vert)$, and since $f_\mathbf{X}(\mathbf{.})$ is in MV normal form, so is $f_\mathbf{Y}(\mathbf{y})$.

gunes
- 49,700
- 3
- 39
- 75