Sparsity pattern of Jacobian for fmincon

Hello

I am busy with a multiple shooting trajectory optimization problem. I would like to attempt this problem in fmincon using interior point as I've heard it is large scale (I have ~4000 variables).

I just wanted to find out if its possible if one can give fmincon the sparsity pattern of the constraint jacobian? I know some of the elements, but others I cannot differentiate analytically.

Thanks in advance!

Amir

Answer 1

[Alan Weiss]

The fmincon 'trust-region-reflective' algorithm accepts a HessPattern option, but requires you to give an analytic gradient function, so won't be suitable. The 'interior-point' algorithm has no such option, but you can set the Hessian option to 'lbfgs', which saves memory in the Hessian approximation.

And allow me to suggest that you might want to consult Optimizing a Simulation or ODE for suggestions on using optimization routines with an ODE solver. In particular, you might need to take larger-than-default finite difference steps.

Alan Weiss

MATLAB mathematical toolbox documentation

[Amir Patel]

Hi Alan

Thanks for the reply.

I will try the 'lbfgs' option and see what happens.

Just as a suggestion, you guys should think of adding the ability for the use to supply sparsity information into interior-point or sqp. That would make it a lot faster I think.

Regards Amir