How do can I improve my results using lsqnonlin

Hi,

I'm trying to solve a trim equation of 6 parameters using lsqnonlin. My problem is that I keep getting VERRYY different results depending on starting conditions and optimset. I set lb and ub so that keeps the values somewhat in place, but still the answers give basically random results.

I tried using ms to try and better search for a trim value but again I don't find any consistent results.

How do can I make it converge to a smaller resnorm? I reached 6, but I think I need a lower value to be sure my results are accurate.

Thanks

[Matt J]

We need to see the code and how you're running it.

[Walter Roberson]

"My problem is that I keep getting VERRYY different results depending on starting conditions and optimset."

That is very common for lsqnonlin. lsqnonlin is not a global optimizer.

[Noa Regev]

I tried also fsolve but it keeps converging to a solution which isn't good for me.

[Noa Regev]

I can't give out the code, that's thousands of lines!!

I use:

    options=optimset('Display','final','DiffMinChange',1e-5,'DiffMaxChange',0.5,'TolX',1e-10,'TolFun',1.e-5,'MaxFunEvals',1000,'Algorithm','levenberg-marquardt')

[Walter Roberson]

You should try some of the Global Optimizations on the residue (sum of squares of differences between the desired values and the computed values.)

[Matt J]

I can't give out the code, that's thousands of lines!!

Well, what about a mathematical description of the problem? What are the equations? Is the objective function smooth? Definitely no quantization operations like round(), ceil(), fix(), etc...? What do you do currently to obtain an initial guess?