calling KNITRO with no derivatives

14 views

Skip to first unread message

Ziena

unread,

Feb 23, 2007, 9:30:05 AM2/23/07

to KNITRO Nonlinear Optimization Solver

I have a problem coded in C in which I cannot specify exact gradients
or
hessians. I am using gradopt=3 (central) and hessopt=2 (bfgs).

I'm getting the following message in the debug log file:
WARNING: Nonzero constraint derivative found with no Jacobian
element.
Sparsity pattern of Jacobian may be wrong.

Can you advise me as to what to do about this?

Todd Plantenga

unread,

Feb 23, 2007, 9:38:47 AM2/23/07

to KNITRO Nonlinear Optimization Solver

KNITRO wants to exploit partial derivative sparsity
even when using finite difference estimates. For that reason,
you supply the Jacobian sparsity pattern when calling
KTR_init_problem(); specifically, the arguments nnzJ,
jacIndexVars, and jacIndexCons.
The warning message means that the finite difference
computation in KNITRO discovered nonzero derivatives that
are not listed in your sparsity pattern of nonzeroes.

If you truly have no idea what the sparsity pattern is,
then tell KTR_init_problem() that all nonzeroes are present.
The down side is this means the Jacobian will be a dense
matrix, and KNITRO will run slowly if the number of variables
is large.