Let $X \sim\mathcal{N}(0, I_n)$ be a random Gaussian vector. Suppose we have two fixed vectors $a,b\in\mathbb{R}^{n}$. We know
\begin{equation}
\langle X, a \rangle \sim \mathcal{N}(0, \|a\|_2^2), \quad \langle X, b \rangle \sim \mathcal{N}(0, \|b\|_2^2),
\end{equation}
and
\begin{equation}
\mathbb{E}\left[\langle X, a \rangle\, \langle X, b \rangle \right] = \langle a, b\rangle.
\end{equation}
I am wondering what else we could say about the distribution of the random variable $\langle X, a \rangle\, \langle X, b \rangle$. Under what conditions are $\langle X, a \rangle$ and $\langle X, b \rangle$ independent? Assuming this is only when $a$ and $b$ are unit orthogonal?
Distribution of product of inner products with random normal vector and two fixed vectors
probability
Best Answer
Since these two random variables are both linear combinations of the components of a multivariate Gaussian random vector, they are indepedent if they are uncorrelated. And \begin{align} & \operatorname{cov}(\langle X,a\rangle, \langle X,b\rangle) \\[8pt] = {} & \operatorname{cov}(a^\top X, b^\top X) \\[8pt] = {} & a^\top \operatorname{cov}(X,X) b \\[8pt] = {} & a^\top I_n b \\[8pt] = {} & 0 \text{ if, but only if, } a,b \text{ are orthogonal.} \end{align}