Re: Maximum likelihood estimator and multiple maxima

I have met this problem of multiple global maxima of the likelihood (minima of chisquare) trying a six parameter fit in the context of a cosmological model.

Using global optimizers I have found already about ten chisquare minima whose relative diference of chisquare is less than 10^(-6), and many more with slighty larger relative difference. They are located in a "wide area" of the the allowed region of the parameter space.

Because of numerical errors that might be of the same order of these relative differences, I found doubtful that any of these chisquare minima locations could be identified with the point in parameter space refered to in the maximum likelihood method.

So, I ask you some advice/pointers on alternative methods to get estimates of the values of the parameters and their confidence regions for this kind of problems.

Regards, Alejandro Jakubi
.