[Ipopt] finalize solution

[Joel Andersson]

Oops, you beat me to it. Just adding one thing, since I know Brad is the
author of an excellent AD tool...

Keep in mind that you can always calculate the simple bounds multipliers
yourself after IPOPT finishes. It follows easily from the KKT conditions.
All you need is a reverse mode sweep on the Lagrangian function.

Joel
On Jul 20, 2015 12:18 AM, "Tony Kelman" <kelman at berkeley.edu> wrote:

> This may change depending on the setting of the option
> fixed_variable_treatment
> http://www.coin-or.org/Ipopt/documentation/node43.html#SECTION000114060000000000000
>
> "For both "make_constraints" and "relax_bounds", bound multipliers are
> computed for the fixed variables."
>
>
> -----Original Message----- From: Brad Bell
> Sent: Sunday, July 19, 2015 3:01 PM
> To: ipopt at list.coin-or.org
> Subject: [Ipopt] finalize solution
>
> Consider the case where the lower limit and upper limit is equal for one
> of the primal variable components; e.g.,
>     x_l[j] = x_u[j]
> in the get_bounds_info function.
>
> I appears that in this case, finalize_solution will have
>     z_L[j] = z_U[j] = 0.
> Is this how Ipopt works ?
>
> If so, consider the optimization problem
>     minimize x[0]^2 subject to  1 <= x[0] <= 1
> In this case, it appears to me, that the final multiplier z_L[0], in
> finalize_solution, should be non-zero.
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20150720/9f24b028/attachment.html>