Re: Problem with 'lsqnonlin' function


In article <14515320.1201524250454.JavaMail.jakarta@xxxxxxxxxxxxxxxxxxxxxx>,
miki lo <myartsev@xxxxxxxxx> writes:
Hello,
I am trying to do a non-linear fit using the 'lsqnonlin' Matlab function.
Some times the termination is successful and I get a message that:
Optimization terminated: relative function value changing by less than OPTIONS.TolFun.
However, there are times where I get the following termination message:
Optimization terminated: first-order optimality less than OPTIONS.TolFun,
and no negative/zero curvature detected in trust region model.

What does it mean?
I tried looking at the Matlab help of the function but couldn't understand what this message is supposed to tell me?

Thank you

the second case is the better one!
if you obtain the first message, then this simply means that the sum of
squares function is very flat such that functions values change from move
to move (a solution of a trust region model) by less than the
parameter set by Options.Tolfun (you obtain the value calling
optimset with no parameters)
so, with say Tolfun=1.0e-8 your final precision may be much worse
1.0e-4.
in the second case, the gradinet norm has become small and the
Hessian seems to be positive definite (large scale version), i.e.
the PCG method didn't detect a direction of negative curvature.

hth
peter

.