Suppose I have a random $m \times m$ matrix $R \sim \mu$ that is possibly singular. Is it true that $ E[R] \propto I$ implies that there exists a scalar $r_{\mu, m}$ such that $E[R^+] = r_{\mu, m}I$, where ${}^+$ denotes the Moore-Penrose pseudo-inverse?

If this is false but there are certain conditions that are needed to make this true, then I'd be interested to know those conditions.

For instance it is true when $R = X^\top X$ and $X$ is $n \times m$ normally distributed (at least I have found references for this when $n\not\in [m-3, m+1]$). Not sure if this is the only non-trivial case.

  • $\begingroup$ what does $R\sim \mu$ mean? $\endgroup$ Jan 28 '21 at 12:12
  • $\begingroup$ The random matrix $R$ has distribution $\mu$. $\endgroup$
    – user27182
    Jan 28 '21 at 16:00
  • $\begingroup$ you write $R^+ = r_{\mu, m}I$ --- don't you want to take the expectation value of $R^+$ ? $\endgroup$ Jan 28 '21 at 17:58
  • $\begingroup$ yes, that's a typo! thanks $\endgroup$
    – user27182
    Jan 28 '21 at 20:27

In general no.

If $X$ and $Y$ are independent scalar random variables satisfying $\mathbb E X = \mathbb E Y = 1$ but $\mathbb E X^{-1} \neq \mathbb E Y^{-1}$, then the matrix $R = \begin{pmatrix} X & 0 \\ 0 & Y \end{pmatrix}$ is a counterexample.


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.