- 113

- 0

y(x) = a/(x+b) + c

My problem is that least-squares fitting using this equation is extremely unstable and heavily dependent on initial guess. No matter how accurate parameters I start with, the algorithm often goes berserk (trying 3.5e+58 instead of 1.2 etc.). It also doesn't matter which algorithm I use.

I guess there must be some mathematical preconditioning voodoo or something that I could use, but I can't find anything that works for me.

Any ideas?

Thanks!