Distribution of a quadratic function involving a random gaussian matrix

117 Views Asked by At

Assume $A$ is a gaussian matrix of size $n \times N$ where $n\ll N$. All the entries of $A$ are independent and identically distributed normal $\mathcal N(0, 1/n)$.

Moreover, assume that $x$ is a vector of size $N$.

Denote $H=A^{*}A-I$, where $I$ is the identity matrix. Can someone help me understand the distribution of $Hx$?

In a research paper I read (pg 3, section D), they state that $Hx$ behaves as a kind of noisy random vector and is accurately modeled as a vector with i.i.d. Gaussian entries with variance $ \frac{1}{n} \|x\|^{2}$

Can someone help me prove it?

I also have a follow-up question:

What can I say about distribution of $Hx$, if all entries of $A$ were independent and identically distributed $\operatorname{Bern}(p)$ variables. .