- Compare with MadNCL-CPU+ExaModels, MadNCL-CPU+JuMP, MadNCL-CPU+JuMP+SymbolicAD
- Explain why product w_1w_2 <= 0 is equivalent to the MPCC (hard to interpret currently)
- Improve technical description for Section IV
- Definition of the index sets
- Sign of r and t
- Update rule for rho
- Definition of variable x
- Explain better the signed Cholesky factorization and MadNLP (how does it differ from other solver?)
- Include information about constraint violation for equality and inequality
- Motivate why MINLP reformulation is intractable
- Condition number of the Newton system as function of \rho for ACOPF
- Comments better why Knitro needs more iterations, and why MadNLP-CPU needs more iterations than MadNCL-GPU
- Solve instance from the GO-competition (allow for infeasible contingencies)
- Remove III.B and Lagrangian in III.A? Also easier description of strong stationarity
- Clarify contingency screening procedure (low or high infeasibility)
- Provide more details about the base-case AC OPF model
- Motivate better why problem violates MFCQ
- Issue with Eq (11) (w1, w2) \geq 0
- Clarify inner and outer iterations in IV.A
- Clarify that solver can converge to a spurious locally infeasible solution