I have $N$ data points that I am trying to fit using a function of the form

$y_i = \prod_j {X_{i,j}}^{b_j}, \quad j=1..N$

where $\mathbf X$ and $\mathbf y$ are measured values. The form of this function is motivated from physics. I also have a variance-covariance matrix containing the estimated uncertainties of $\mathbf X$ and $\mathbf y$, which I assume to be distributed according to a multivariate normal distribution. Finally, $\mathbf b$ are the exponents I would like to determine.

I am not sure how to handle the fact there are uncertainties in the $\mathbf X$ variables and furthermore that they are correlated with the errors in $\mathbf y$. If the errors in $\mathbf X$ did not exist then I would simply find the $\mathbf b$ that minimize

$\chi^2 = \sum_i \frac{(y_i - \prod_j X_{i,j}^{b_j})^2}{\sigma_y^2}$

What should I do in this case? Is there a similarly simple functional form that I can optimize?

It is not lost on me that I can take the log of both sides and it looks like a standard regression problem, except without an intercept. Is this a useful line to pursue?


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.