Given a 2-D parameter vector, $\mathbf{X} = [x_1, x_2]$, let the corresponding $2\times2$ Fisher Information Matrix be $\mathbf{F}$. The Cramer-Rao Lower Bound (CRLB) is the inverse of the FIM. I obtain the following matrix by inverting $\mathbf{F}$ as: $$\mathbf{C} = \begin{bmatrix} \mathbf{C}_{1,1} & \mathbf{C}_{1,2}\\\mathbf{C}_{2,1} & \mathbf{C}_{2,2}\end{bmatrix}$$

I wanted to use a single error metric for both $x_1$ and $x_2$. I came across the following expression in one of the papers: $$ f(\mathbf{X}) = \frac{1}{\sqrt{2}}\left(\mathbf{C}_{1,1}^2 + \mathbf{C}_{2,2}^2\right)^{1/2}$$

What does this quantity imply? Is this a joint CRLB for both parameters?

  • $\begingroup$ I understand that $f(\mathbf{X})$ is actually the rms value of the lower bounds of both parameters. But I have not seen usage of this quantity before. $\endgroup$
    – r2d2
    Commented Feb 1, 2018 at 3:13
  • $\begingroup$ Hi: If it was $\frac{1}{\sqrt n}$ rather $\frac{1}{\sqrt2}$, where $n$ was the sample size, then it would be the CRLB for the standard deviation of the sum of the 2 parameters. But, given that it's $\frac{1}{ \sqrt n}$, I can't help. maybe it's a typo ? $\endgroup$
    – mark leeds
    Commented Feb 1, 2018 at 7:07


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.