I fit a spectrum using binning likelihood method, and want to draw a n-sigma contour about two parameters.
The fitting status is converged. But it breaks down when drawing a n-sigma contour and says that it can’t find 4 points. Here, I wonder whether a contour is sensitive to the parameters’ initial values. Shall I adjust the initial values again and again so as to draw a good n-sigma contour (in the codes I require 20 points for a contour)? Anyone help me? Thanks!
The problem is probably due to fit failure when you evaluate the profile likelihood for drawing the cantor.
You should understand why the minimizations are failing, maybe your function is having some discontinuities in the derivatives. Look at the result using an higher print level in the minimisation.
You could also try to perform a grid scan of the profile-likelihood and get the contour from the grid.
In this way you can understand better the locations where the minimizations are failing.