How to evaluate uncertainty of solution of over-determined system using least square method?

17 Views Asked by At

I'm trying to solve the over-determined system of linear form:

$$ Ax = b $$

I know we can consider the problem as a least-square regression[1]:

$$ min_x \|Ax - b\| $$

and the solution is

$$ x = (A^\top A)^{-1}A^\top b $$

But I don't know how accurate this approximation is. How to properly evaluate the uncertainty of x of this solution?

Also appreciate if you share some article about this theory.


Note:

I found related answer[2]. According to the article, the covariance matrix of x is calculated as

$$ C_x = (A^\top A)^{-1}A^\top C_b A (A^\top A)^{-1} $$

If it is true, then, how can we set covariance matrix of b ($C_b$)? Just calculating standard deviation of $\|Ax - b\|$ (we denote this as $\sigma$) and set $C_b = eye(\sigma^2)$ is O.K.?