François Pacaud

An idea is to implement Hessian-vector product as a linear operator, such as ``` struct HessVecProd

See: http://stanford.edu/class/cme338/notes/notes09-PDCO.pdf

Required for the MOI to NLPModels conversion in https://github.com/MadNLP/MadNCL.jl/pull/2

Distributed OPF leads to the formulation of degenerate nonlinear programs. Preliminary experiments on rosetta_distribution_opf show that MadNLP has different behavior than Ipopt. [This script](https://github.com/hei06j/rosetta_distribution_opf.jl/blob/main/script_4wire/IVR_EN_vectorized.jl) returns the following output with Ipopt:...

Pass option `prod` to ExaModel when defining a new model on the GPU from JuMP. Fix ambiguity for `hprod!`.

I am currently working on the MacMPEC benchmark, and using AmplNLReader to import the instances implemented by [Sven Leyffer](https://wiki.mcs.anl.gov/leyffer/index.php/MacMPEC). I noticed I got an error if there is multiple dots...

- add explicit return for the callbacks to fix `hess_structure` and `jac_structure` - fix type ambiguity in KernelAbstractions wrapper xref #111

Revisit fixed variables treatment

2

* Change the behavior of MakeParameter, depending on the callback: - `SparseCallback`: remove the fixed variables and reduce problem's dimension in MadNLP - `DenseCallback`: keep the fixed variables frozen (as...

- relax initial bounds with `bound_relax_factor` instead of `tol`. - significantly improve the performance of MadNLP on CUTEst [pprof-cpu.pdf](https://github.com/user-attachments/files/16383748/pprof-cpu.pdf)