[WIP] MVP for lower dimensional reformulation

[dfridovi]

Reworking the KKT system to be lower dimensional by handling each nested problem as follows:

• use an interior point relaxation on all inequality constraints
• keep the primal feasibility constraints, dual feasibility, and complementarity (actually, relax via the interior point scheme and ignore dual feasibility and interior point slack feasibility constraints)
• every time we move up a level in the hierarchy, do the following:
  1. add the constraints above to a list
  2. create a constraint on the current level that enforces Lagrangian stationarity of the lower level, but only taking the gradient wrt the upper level primal and not wrt duals or slacks at lower levels (i.e., no "induced primals")

Main features:

• new KKT system structure
• general construction of the new KKT system
• recursive symbolic derivation of the KKT system for equality-constrained problems
• custom Newton solver with fraction-to-the-boundary linesearches that handles non-square kkt system
• add preference slacks/inequality constraints
• test problem set up with CI
• profile vs original GOOP
• try dropping \nabla^k \pi for k \geq 2

[dfridovi]

@leedh0124 why is there a dependency on IntelOpenMP? Can we rm that? It's causing the tests to fail: https://github.com/CLeARoboticsLab/Quasi-GOOP/actions/runs/16454832091/job/46509114257?pr=1

[leedh0124]

I never noticed that. I think we can remove it Dong Ho Lee, PhD Student The University of Texas at Austin | CLeAR Lab, Department of Aerospace Engineering and Engineering Mechanics | 737-230-7783 | https://clearoboticslab.github.io/ <https://www.utexas.edu/>

…

On Tue, Jul 22, 2025 at 4:10 PM David Fridovich-Keil < ***@***.***> wrote: *dfridovi* left a comment (CLeARoboticsLab/Reduced-GOOP#1) <#1 (comment)> @leedh0124 <https://github.com/leedh0124> why is there a dependency on IntelOpenMP? Can we rm that? It's causing the tests to fail: https://github.com/CLeARoboticsLab/Quasi-GOOP/actions/runs/16454832091/job/46509114257?pr=1 — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIPBX4GB6CYGC2HUEVEE4E33J2SDTAVCNFSM6AAAAACB25G4XKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCMBUHA2DCNRQGU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[dfridovi]

@leedh0124 I think it should be fixed now. I just removed a bunch of unnecessary dependencies. Btw, when you put back in the examples with intersections and stuff that you want to visualize, put those into a separate examples module that has its own separate dependencies. You can see how I did it in MCPs.jl so it doesn't pollute the main package deps.