pyoptsparse | oriented framework

despite both having the SLSQP name, Scipy and pyoptsparse use different implementations of the algorithm. The Scipy one was updated with stricter bounds enforcement a few years ago, but the one in pyoptsparse was not.

To fix this, someone would need to spend some time updating the SLSQP fortran code pyoptsparse itself. Jacob Williams has done a decent job of creating an updated SLSQP codebase. Perhaps you could collaborate with him to add his version of the algorithm to pyoptsparse.

SNOPT installation is purely a related to pyOptSparse. OpenMDAO uses pyOptSparse as is. So if SNOPT is not working in OpenMDAO, then it is because something has gone wrong with your SNOPT install in pyOptSparse itself.

One way you can test if things are working is to run the pyOptSparse test for SNOPT.

You want to order a source distribution of SNOPT from here

I've pushed up a couple of fixes to the repository so that you can run it without SNOPT. The basic Branin problem in the examples works and gets to the expected answer now. I can't promise that SLSQP is the best choice for more complicated problems as we usually favor SNOPT over SLSQP in our work. This is still very experimental code, so the documentation is weak and there are still a lot of control knobs and flags that are buried as subcomponent attributes (including ideas that we tried that didn't pan out). But we appreciate users who are willing to try AMIEGO and help us improve it.

The first step in making any optimizer available to OpenMDAO is to make it accessible in Python itself. It looks like APOPT is available in Python through the GEKKO library. So Step 1 is for you to get that library, and figure out how to use it.

Then you can write a driver that integrates GEKKO in OpenMDAO. To do that you will write your own plugin that wraps GEKKO. You can search for the openmdao github topic to see a list of existing plugins for some examples (you can make plugins for anything, not just drivers)

One example driver plugin for is the OpenMDAO wrapper for the NLopt library, written by the National Renewable Energy Laboratory. https://github.com/johnjasa/nrel_openmdao_extensions/blob/master/nrel_openmdao_extensions/nlopt_driver.py

The actual driver code is here, but the full repo is necessary to actually register it as a plugin.

You can also look into the MINLP algorithm AMIEGO, which is also available as a plugin. You can read about it in detail in this paper.

All of the expressions in Gekko must use Gekko variables so that the model can be compiled to byte-code. If you open the run folder and gk_model0.apm, you will see the model that you've created with this script: