What is your evidence that Solver gets confused by functions that can't
go negative (since that would impact all kinds of minimizations)? I
have always assumed that the issue was that the defaults are set way too
loosly.

I have not looked hard for alternate settings that would work in one
pass, but if delta is the quantity that I am trying to minimize, I can
usually improve the initial solution with a second pass to minimize
c*delta, where c is suitably large (say 10^5).

Jerry

Dana DeLouis wrote:

Hi. If your data on the x-axes is evenly spaced out, Excel has a Fourier
Transform function under Data | Analysis. However, it's a Radix-2 algorithm
only.
My opinion is that Solver can not do a LSQ very well, especially with more
than just a few data points. I've never had much success with a LSQ
fitting. The main problem is that by definition, one is squaring the error,
so the "error" never goes negative. This confuses Solver. With multiple
values, Solver gets confused, and will quickly give up.