5
$\begingroup$

What does rotational invariance mean in statistics? The property that the normal distribution satisfies for independent normal distributed $X_i$, $\Sigma_i X_i$ is also normal with variance $\Sigma_i Var(X_i)$ is referred to as rotational invariance and I want to know why.

$\endgroup$
8
$\begingroup$

Let $X_i$ have mean $\mu_i$ and standard deviation $\sigma_i$, and write $X_i = \mu_i + \sigma_i Z_i$ where $Z_i$ are independent standard normal random variables. The joint density of $Z_1, \ldots, Z_n$ is $$f(z_1, \ldots, z_n) = (2 \pi)^{-n/2} e^{-(z_1^2 + \ldots + z_n^2)/2} = (2 \pi)^{-n/2} e^{-\|{\bf z}\|^2/2}$$ which is rotationally invariant, i.e. invariant under rotations of $n$-dimensional space, because it only depends on the length of the vector ${\bf z} = (z_1, \ldots, z_n)$. Now $Z_1 = (1,0,\ldots, 0) \cdot \mathbb (Z_1, \ldots, Z_n)$ has a standard normal distribution. But because of the rotational invariance, so does ${\bf u} \cdot (Z_1, \ldots, Z_n)$ for any unit vector $\bf u$.
In particular, take ${\bf u} = (\sigma_1, \ldots, \sigma_n)/\sigma$ where $\sigma = \sqrt{\sigma_1^2 + \ldots + \sigma_n^2}$, and we get that $(\sigma_1 Z_1 + \ldots + \sigma_n Z_n)/\sigma$ has a standard normal distribution, which means $$X_1 + \ldots + X_n = (\mu_1 + \ldots + \mu_n) + \sigma \dfrac{\sigma_1 Z_1 + \ldots + \sigma_n Z_n}{\sigma}$$ has a normal distribution with mean $\mu_1 + \ldots + \mu_n$ and standard deviation $\sigma$.

$\endgroup$

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.