A Specific SLP/QP Preset

[jeffreydeankelly2]

Dear Charlie;

I tried the undocumented preset = filterslp as you suggested and found it was not effective on the types of SLP problems we solve.

I am wondering if it would be possible for Uno to implement a true SLP approach which would require the following:

1. Allow the modeler to identify variables and constraints as linear and nonlinear. Only nonlinear variables have convergence control (i.e., trust region step-bounding) and only nonlinear constraints have 1-norm penalty-error variables augmented. And technically, only nonlinear variables require initial-values.

2. Converge on the objective function and the constraints only after the 1-norm penalty-errors have converged. If the variables converge first with the objective function and constraints unconverged, then declare the problem as unconverged. If the 1-norm penalty-errors do not converge after a user specified number of iterations, then report the worst 1-norm penalty-error activities as the most-likely nonlinear constraints that are infeasible (usually a small number). This has proven to be a very fast and reliable diagnostic to identify root-cause inconsistencies.

Just thought I would pass this along for your reflection to see if perhaps it fits with your unifying nonlinear optimization thesis :-)

All the best - Jeff