Jacobian from lsqnonlin with multistart

Dear all,

For my research, I'm running multiple lsqnonlin-fits using Matlab's MultiStart function. My objective function does not supply a jacobian.

If I run a single lsqnonlin-fit (from one start point; no MultiStart), I'm able to retrieve the (approximated) jacobian using: [x,resnorm,residual,exitflag,output,lambda,jacobian] = lsqnonlin(...) I need the jacobian to calculate a confidence interval for my estimated parameters. The run command (to run MultiStart), however, does not return a jacobian.

Is there a way to get the jacobian of the best fit of a MultiStart-run (again using lsqnonlin as the local solver)?

Thanks in advance!

Bart

Answer 1

[Bart]

Is there anybody who has some thoughts on solving this problem? Thanks!

Bart.

Answer 2

[Cody]

yes I'd also like to know how this can be accomplished

Answer 3

[Kapil Dave]

I am also interested in the same

Answer 4

[Carlos Henriques]

I'm doing it in this way:

you can extract the best X0 from multistart and then use it to run the lsqnonlin again.

For instance:

[params,fval,eflag,output,manymins] = run(ms,problem,20); % manymins have all the X0 used by multistart, you can take the better

[params,resnorm,residual,exitflag,fit_out,lambda,jacobian] = lsqnonlin(fun,manymins(1).X0{1},lb,ub,options);

cheers