From: Marco on
Hi,
I'm using the interior-point algorithm in FMINCON with the analytical
expression of the Hessian, setting the following option
....'Hessian','user-supplied','HessFcn',@funHess...

To reduce the amount of memory, I define the Hessian matrix
as a sparse matrix in the function @funHess

I also notice that it is possible to provide the function with the sparsity
pattern of the Hessian, using the option "HessPattern".

Is that always convenient, even if you provide the analytic expression of the
(sparse) Hessian?

Actually I have tried it with some problems, and I haven't noticed any improvement
in terms of computing speed.

By the way, the Matlab help mentions the KNITRO toolbox which seems to be
more powerful than FMINCON in case of large and sparse Hessian.
Is there anyone who tested the two functions on large systems?

Best
Marco