[Ipopt] Hessians of objective and constraints have different sparse structures

Stefan Vigerske stefan at math.hu-berlin.de
Mon Dec 3 06:05:11 EST 2018


no, Ipopt works with the Hessian of the Lagrangian, which includes 
objective and all (nonlinear) constraints.


On 12/1/18 10:07 PM, R zu wrote:
> Hi.
> I can divide the hessian to 2 x 2 blocks:
> D, A
> B, C
> The Hessian of the objective has non-zero elements only in the D and C
> blocks. The D block is diagonal.
> The Hessian of each constraint is all zero except for one element in the A
> or B blocks.
> The sum of the Hessian of all constraints has non-zero elements over many
> places in blocks A and B.
> Ipopt seems to require me to provide the sum of all Hessians, which is
> quite dense. Is it possible to provide the Hessian of the objective and the
> weighted sum of the Hessian of the constraints separately?
> doc: https://www.coin-or.org/Ipopt/documentation/node22.html#eq:IpoptLAG
> Thanks again.
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/ipopt

More information about the Ipopt mailing list