Let $U$ be a random matrix, supported on the positive-definite cone of matrices. We denote $\sqrt{U}$ to be the principal square root of $U$. That is, the unique positive-definite matrix such that $\sqrt{U}^2 = U$.

I'm interested in bounding the quantitiy $\text{Tr}({\mathbb{E}[\sqrt{U}]^2})$.

In the given answer here a solution is given for random variables which shows that $$\mathbb{E}[X]\left(1 - \frac{\mathbb{E}(X - \mathbb{E}[X])^2}{2\mathbb{E}[X]}\right)^2 \leq E[\sqrt{X}]^2$$.

However, the method used above is inapplicable for non-commutative matrices.

I'm wondering if there is any equivalent results known for matrices.

  • $\begingroup$ isn't it true that the entities involved in the estimate you mention are $X$ and the identity matrix $I$? They do commute, and thus formulae would still hold... $\endgroup$ Feb 19 '17 at 22:37
  • $\begingroup$ The problem is that we really need $X$ to commute with $\mathbb{E}[X]$, which is generally not true. $\endgroup$
    – Cain
    Feb 21 '17 at 11:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.