Showing that $E[\|X_k\|^2]$ approaches $O(k)$ from below for random $k\times d$ matrix $X$?

30 Views Asked by At

Suppose $X_k$ is a matrix made by stacking $k$ random vectors of unit norm sampled from some sufficiently nice $d$-dimensional distribution.

How would I show that the following quantity is monotonically increasing in $k$ for operator norm $\|\cdot\|$?

$$E[(\|X_k\|-1)^2]/k$$

Expecting it to be true because when $k$ is small relative to $d$, a random vector has high probability to be orthogonal to existing ones and may not affect the norm much.

Simple simulation

enter image description here