### Install and Build MadNLPPardiso

Source: https://madsuite.org/MadNLP.jl/stable/installation

Install the MadNLPPardiso package and then build it. This process requires a C compiler.

```julia
pkg> add MadNLPPardiso
pkg> build MadNLPPardiso
```

--------------------------------

### Re-run MadNLP with Different Initial Point

Source: https://madsuite.org/MadNLP.jl/stable/quickstart

This example demonstrates how to re-initialize the optimization problem with a new starting point (`x0`) and run MadNLP again. This is useful for exploring different local optima in non-convex problems.

```julia
x0 = ones(2)
nlp = HS15Model(x0)
results = madnlp(nlp)
nothing
```

```text
This is MadNLP version v0.9.1, running with MUMPS v5.8.2

Number of nonzeros in constraint Jacobian............:        4
Number of nonzeros in Lagrangian Hessian.............:        3

Total number of variables............................:        2
                     variables with only lower bounds:        0
                variables with lower and upper bounds:        0
                     variables with only upper bounds:        1
Total number of equality constraints.................:        0
Total number of inequality constraints...............:        2
        inequality constraints with only lower bounds:        2
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
   0  5.8004900e+01 5.20e-01 8.04e+01 1.49e+00  -1.0     -   0.00e+00  1  0
   1  5.5900651e+01 5.15e-01 7.60e+01 1.43e+00  -1.0     -   9.17e-03  1  1h
   2  5.3726392e+01 5.10e-01 8.07e+01 2.01e+00  -1.0     -   1.04e-02  1  1h
```

--------------------------------

### Install MadNLPHSL

Source: https://madsuite.org/MadNLP.jl/stable/installation

Add the MadNLPHSL package after successfully installing HSL_jll.

```julia
pkg> add MadNLPHSL
```

--------------------------------

### Import MadNLPGPU and CUDA

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

Import the necessary packages for using MadNLP on the GPU. This is the initial setup step.

```julia
using MadNLPGPU
using CUDA
```

--------------------------------

### Run MadNLP with Initial Solution

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/warmstart

Solve the optimization problem using the `madnlp` function, starting from the previously obtained initial solution `x0`. This performs a warm start.

```julia
results = madnlp(nlp)
nothing
```

--------------------------------

### Install MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/installation

Use this command to add the MadNLP package to your Julia environment.

```julia
pkg> add MadNLP
```

--------------------------------

### Solve KKT System with SparseKKTSystem

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/kktsystem

This snippet demonstrates how to solve the KKT system using MadNLP's SparseKKTSystem and LapackCPUSolver. Ensure MadNLP is installed and the necessary models are defined.

```julia
nlp = HS15Model()
results = madnlp(nlp; kkt_system=MadNLP.SparseKKTSystem, linear_solver=LapackCPUSolver)
nothing
```

--------------------------------

### Develop HSL_jll.jl Package

Source: https://madsuite.org/MadNLP.jl/stable/installation

Install HSL_jll.jl by developing the package from a local path. Ensure you have obtained the necessary license.

```julia
import Pkg
Pkg.develop(path = "/full/path/to/HSL_jll.jl")
```

--------------------------------

### Install MadNLPGPU for GPU Support

Source: https://madsuite.org/MadNLP.jl/stable/installation

Add the MadNLPGPU package to enable GPU acceleration. It automatically detects NVIDIA or AMD GPUs.

```julia
pkg> add MadNLPGPU
```

--------------------------------

### Access Initial Solution with get_x0

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/warmstart

Retrieve the initial primal variable stored in `nlp.meta.x0` using the `get_x0` function. This is useful for understanding the starting point of the optimization.

```julia
x0 = NLPModels.get_x0(nlp)
```

--------------------------------

### MadNLP Optimization Iteration Log (Early Stage)

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/quasi_newton

This log shows the initial iterations of the MadNLP optimization process, highlighting rapid changes in objective and infeasibility values. It is useful for observing the algorithm's behavior at the start of a run.

```text
iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
   0  4.8810842e+01 2.22e-16 1.71e+01 0.00e+00  -1.0     -   0.00e+00  1  0
   1  2.2956342e+00 1.50e+03 2.04e+02 0.00e+00  -1.0     -   1.25e-01  1  4f
   2  4.4967927e+00 3.76e+02 5.48e+01 0.00e+00  -1.0     -   1.00e+00  1  1h
```

--------------------------------

### Get KKT Matrix from AbstractKKTSystem

Source: https://madsuite.org/MadNLP.jl/stable/man/linear_solvers

Retrieve the assembled KKT matrix from an AbstractKKTSystem instance. Ensure the KKT system is properly assembled before calling this function.

```julia
K = MadNLP.get_kkt(kkt)
```

--------------------------------

### Get KKT System Components

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/kktsystem

Provides basic functions to retrieve components of the KKT system, such as the number of variables, the KKT matrix in CSC format, and the Jacobian callback.

```julia
MadNLP.num_variables(kkt::DiagonalHessianKKTSystem) = length(kkt.diag_hess)
MadNLP.get_kkt(kkt::DiagonalHessianKKTSystem) = kkt.aug_com
MadNLP.get_jacobian(kkt::DiagonalHessianKKTSystem) = kkt.jac_callback
```

--------------------------------

### MadNLP Solver Output

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

Example output from the MadNLP solver, detailing iteration progress, infeasibilities, and final solution status. This output is useful for diagnosing convergence issues.

```text
This is MadNLP version v0.9.1, running with cuDSS v0.7.1

Number of nonzeros in constraint Jacobian............:       84
Number of nonzeros in Lagrangian Hessian.............:     5250

Total number of variables............................:       84
                     variables with only lower bounds:        0
                variables with lower and upper bounds:       84
                     variables with only upper bounds:        0
Total number of equality constraints.................:        0
Total number of inequality constraints...............:       42
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:       42

iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
   0  0.0000000e+00 1.04e+02 1.00e+00 1.00e+01  -1.0     -   0.00e+00  0  0
   1  1.4808378e+01 1.00e+02 2.64e+01 6.04e+00  -1.0     -   3.37e-02  1  1h
   2  1.3660555e+04 2.50e+01 1.79e+02 1.87e+00  -1.0     -   1.00e+00  1  1h
   3  1.5757708e+04 2.11e+01 1.55e+02 8.19e-01  -1.0     -   1.66e-01  1  1h
   4  2.2700976e+04 1.19e+01 1.31e+02 3.46e-01  -1.0     -   4.94e-01  1  1h
   5  3.6245598e+04 2.96e+00 2.13e+02 2.84e-01  -1.0     -   1.00e+00  1  1h
   6  4.0274013e+04 1.53e+00 1.64e+02 1.19e-01  -1.0     -   5.59e-01  1  1h
   7  4.5453607e+04 3.61e-01 1.98e+02 1.16e-01  -1.0     -   1.00e+00  1  1h
   8  4.7405169e+04 7.22e-02 1.00e+02 1.00e-01  -1.0     -   1.00e+00  1  1h
   9  4.7892154e+04 8.03e-03 3.84e+01 9.05e-02  -1.0     -   1.00e+00  1  1h
iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
  10  4.7949829e+04 3.07e-04 6.16e+00 7.82e-02  -1.0     -   1.00e+00  1  1h
  11  4.7951497e+04 1.23e-06 5.90e-02 7.81e-02  -1.0     -   1.00e+00  1  1h
  12  4.7947424e+04 3.39e-06 1.21e-03 2.85e-03  -2.5     -   1.00e+00  1  1f
  13  4.7947303e+04 4.19e-09 1.24e-06 7.80e-06  -5.0     -   1.00e+00  1  1h

Number of Iterations....: 13

                                   (scaled)                 (unscaled)
Objective...............:   4.7947302735661782e+04    4.7947302735661782e+04
Dual infeasibility......:   1.2358625376867822e-06    1.2358625376867822e-06
Constraint violation....:   4.1921440895476697e-09    4.1921440895476697e-09
Complementarity.........:   7.8004282261790513e-06    7.8004282261790513e-06
Overall NLP error.......:   7.8004282261790513e-06    7.8004282261790513e-06

Number of objective function evaluations              = 14
Number of objective gradient evaluations              = 14
Number of constraint evaluations                      = 14
Number of constraint Jacobian evaluations             = 14
Number of Lagrangian Hessian evaluations              = 13
Number of KKT factorizations                          = 13
Number of KKT backsolves                              = 13

Total wall secs in initialization                     =  1.257 s
Total wall secs in linear solver                      =  unavailable
Total wall secs in NLP function evaluations           =  0.004 s
Total wall secs in solver (w/o init./fun./lin. alg.)  =  unavailable
Total wall secs                                       =  8.150 s

EXIT: Optimal Solution Found (tol = 1.0e-04).

```

--------------------------------

### KKT System Creation and Initialization

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Functions for creating and initializing KKT systems.

```APIDOC
## Create KKT System

`MadNLP.create_kkt_system` — Function

```
create_kkt_system(
    ::Type{KKT},
    cb::AbstractCallback,
    ind_cons::NamedTuple,
    linear_solver::Type{LinSol};
    opt_linear_solver=default_options(linear_solver),
    hessian_approximation=ExactHessian,
) where {KKT<:AbstractKKTSystem, LinSol<:AbstractLinearSolver}
```

Instantiate a new KKT system with type `KKT`, associated to the the nonlinear program encoded inside the callback `cb`. The `NamedTuple` `ind_cons` stores the indexes of all the variables and constraints in the callback `cb`. In addition, the user should pass the linear solver `linear_solver` that will be used to solve the KKT system after it has been assembled.

## Initialize KKT System

`MadNLP.initialize!` — Function

```
initialize!(kkt::AbstractKKTSystem)
```

Initialize KKT system with default values. Called when we initialize the `MadNLPSolver` storing the current KKT system `kkt`.
```

--------------------------------

### Initialize KKT System

Source: https://madsuite.org/MadNLP.jl/stable/man/kkt

Initializes the KKT system after it has been built. This step is necessary before using the KKT system within an interior-point algorithm.

```julia
MadNLP.initialize!(kkt);
```

```julia
3-element Vector{Float64}:
 0.0
 0.0
 0.0
```

--------------------------------

### Create KKT System

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Instantiate a new KKT system with a specified type, callback, constraint indices, and linear solver. Optional arguments for linear solver options and Hessian approximation can be provided.

```julia
create_kkt_system(
    ::Type{KKT},
    cb::AbstractCallback,
    ind_cons::NamedTuple,
    linear_solver::Type{LinSol};
    opt_linear_solver=default_options(linear_solver),
    hessian_approximation=ExactHessian,
) where {KKT<:AbstractKKTSystem, LinSol<:AbstractLinearSolver}
```

--------------------------------

### Configure and Run MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/kktsystem

Instantiates an optimization model and solves it using MadNLP with the DiagonalHessianKKTSystem and LapackCPUSolver. Ensure the necessary model and solver are imported.

```julia
nlp = HS15Model()
results = madnlp(nlp; kkt_system=DiagonalHessianKKTSystem, linear_solver=LapackCPUSolver)
nothing
```

--------------------------------

### Get Dual Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns the dual values (y, z) stored in the KKT vector.

```julia
dual(X::AbstractKKTVector)
```

--------------------------------

### Get Primal Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns the primal values (x, s) stored in the KKT vector.

```julia
primal(X::AbstractKKTVector)
```

--------------------------------

### Get Nonzero Count of Jacobian

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns the number of non-zero elements in the Jacobian matrix.

```julia
nnz_jacobian
```

--------------------------------

### Load Quadmath Library

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/multiprecision

Import the Quadmath library to enable working with quadruple-precision floating-point numbers.

```julia
using Quadmath
```

--------------------------------

### Get Full KKT Vector Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns all values stored within the KKT vector.

```julia
full(X::AbstractKKTVector)
```

--------------------------------

### Create Sparse KKT System

Source: https://madsuite.org/MadNLP.jl/stable/man/kkt

Instantiates a sparse KKT system using the `create_kkt_system` function. Requires the KKT system type, a callback function `cb`, and the chosen linear solver.

```julia
kkt = MadNLP.create_kkt_system(
    MadNLP.SparseKKTSystem,
    cb,
    linear_solver,
)
```

```julia
MadNLP.SparseKKTSystem{Float64, Vector{Float64}, SparseArrays.SparseMatrixCSC{Float64, Int32}, MadNLP.ExactHessian{Float64, Vector{Float64}}, LapackCPUSolver{Float64, SparseArrays.SparseMatrixCSC{Float64, Int32}}, Vector{Int64}, Vector{Int32}}([0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0], MadNLP.ExactHessian{Float64, Vector{Float64}}(), [2.0e-323, 6.34436211249205e-310, 6.3443652045335e-310, 6.3443652045999e-310], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0], [6.34437148191976e-310, 6.34437148191976e-310], [6.3442752171701e-310], [6.34427528331247e-310, 6.34430136495017e-310], [6.34427522052977e-310], [0.0 0.0 … 0.0 0.0; 0.0 0.0 … 0.0 0.0; … ; 0.0 0.0 … 0.0 0.0; 0.0 0.0 … 0.0 0.0], sparse(Int32[1, 2, 5, 6, 2, 5, 6, 3, 5, 4, 6, 5, 6], Int32[1, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 5, 6], [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0], 6, 6), [1, 5, 8, 10, 1, 2, 5, 3, 6, 4, 7, 9, 11, 12, 13], [0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0], sparse(Int32[1, 2, 1, 2, 1, 2], Int32[1, 1, 2, 2, 3, 4], [1.0, 2.0, 3.0, 4.0, 5.0, 6.0], 2, 4), [1, 3, 2, 4, 5, 6], LapackCPUSolver{Float64, SparseArrays.SparseMatrixCSC{Float64, Int32}}(sparse(Int32[1, 2, 5, 6, 2, 5, 6, 3, 5, 4, 6, 5, 6], Int32[1, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 5, 6], [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0], 6, 6), [6.3443751916674e-310 6.34430040933314e-310 … 6.3443651934443e-310 NaN; 6.3443713372534e-310 6.34437195755003e-310 … 2.5e-323 6.34437325290557e-310; … ; 6.3443004186113e-310 6.34436900948325e-310 … 6.3443651934443e-310 6.34436900948325e-310; 5.0e-324 6.3443732524518e-310 … 2.5e-323 6.3443721236952e-310], 6, Float64[], Float64[], Float64[], [384.0, 0.0, 3.466909703e-315, 6.3443794058355e-310, 0.0, 1.51e-321, 5.0e-324, 0.0, 0.0, 3.466909703e-315  …  0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 8.0e-323, 3.88e-321], 384, [128411514206256], -1, Base.RefValue{Int64}(0), [1, 1, 0, 0, 128411514810656, 128411589880704], MadNLP.LapackOptions(MadNLP.BUNCHKAUFMAN), MadNLP.MadNLPLogger(MadNLP.INFO, MadNLP.INFO, nothing)), [1, 2], [3, 4], [1])
```

--------------------------------

### Initialize LapackCPUSolver with KKT Matrix

Source: https://madsuite.org/MadNLP.jl/stable/man/linear_solvers

Create a LapackCPUSolver instance using the assembled KKT matrix. This solver does not copy the matrix, maintaining a reference for direct updates.

```julia
linear_solver = LapackCPUSolver(K)
```

--------------------------------

### Get Upper-Bound Dual Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns the dual values (ω) associated with the upper-bound stored in the KKT vector.

```julia
dual_ub(X::AbstractKKTVector)
```

--------------------------------

### Instantiate Model on CPU

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

Instantiate your optimization model on the CPU using the specified data type and parameters. This is a prerequisite for wrapping it with SparseWrapperModel.

```julia
nlp_cpu = airport_model(Float64, nothing)
```

--------------------------------

### Import ExaModels for GPU Optimization

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

Import the ExaModels package, which is recommended for implementing optimization models that run entirely on the GPU.

```julia
using ExaModels
```

--------------------------------

### Get Lower-Bound Dual Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns the dual values (ν) associated with the lower-bound stored in the KKT vector.

```julia
dual_lb(X::AbstractKKTVector)
```

--------------------------------

### Get Primal and Dual Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Returns both the primal (x, s) and dual (y, z) values stored in the KKT vector.

```julia
primal_dual(X::AbstractKKTVector)
```

--------------------------------

### Display Model Summary

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

This output provides a summary of the instantiated ExaModel, detailing the number of variables, constraints, and their types (free, lower/upper bounded, etc.), along with sparsity information for the Hessian and Jacobian.

```julia
An ExaModel{Float64, CUDA.CuArray{Float64, 1, CUDA.DeviceMemory}, ...}

  Problem name: Generic
   All variables: ████████████████████ 84     All constraints: ████████████████████ 42    
            free: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0                 free: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
           lower: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0                lower: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
           upper: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0                upper: ████████████████████ 42    
         low/upp: ████████████████████ 84             low/upp: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
           fixed: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0                fixed: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
          infeas: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0               infeas: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
            nnzh: (-47.06% sparsity)   5250            linear: ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅ 0     
                                                      nonlinear: ████████████████████ 42    
                                                         nnzj: ( 97.62% sparsity)   84    
                                                     lin_nnzj: (------% sparsity)         
                                                     nln_nnzj: ( 97.62% sparsity)   84    


```

--------------------------------

### Get Number of Dual Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Retrieves the total count of dual values (y, z) within a given KKT vector.

```julia
number_dual(X::AbstractKKTVector)
```

--------------------------------

### Get Number of Primal Values

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Retrieves the total count of primal values (x, s) within a given KKT vector.

```julia
number_primal(X::AbstractKKTVector)
```

--------------------------------

### Initialize KKT System

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Initialize the KKT system with default values. This function is called during the initialization of the `MadNLPSolver` that stores the KKT system.

```julia
initialize!(kkt::AbstractKKTSystem)
```

--------------------------------

### Create SparseCallback for KKT System

Source: https://madsuite.org/MadNLP.jl/stable/man/kkt

Instantiate a SparseCallback to wrap an NLPModel for KKT system assembly. This is typically done before each iteration of an interior-point algorithm.

```julia
cb = MadNLP.create_callback(
    MadNLP.SparseCallback,
    nlp,
)
```

--------------------------------

### Get KKT Matrix Pointer

Source: https://madsuite.org/MadNLP.jl/stable/lib/kkt

Return a pointer to the KKT matrix implemented in `kkt`. This pointer is intended for use with a linear solver.

```julia
get_kkt(kkt::AbstractKKTSystem)::AbstractMatrix
```

--------------------------------

### Initial Solve with MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/warmstart

Perform the first solve of the optimization problem using the instantiated solver.

```julia
results = MadNLP.solve!(solver)
```

--------------------------------

### Initialize Linear Solver

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/kktsystem

Initializes the linear solver with the sparse matrix in CSC format. This prepares the system for solving linear equations.

```julia
_linear_solver = linear_solver(aug_com; opt = opt_linear_solver)
```

--------------------------------

### Instantiate and Solve with MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/quickstart

Creates an instance of the custom HS15Model, initializes a MadNLP solver, and solves the optimization problem.

```julia
x0 = zeros(2) # initial position
nlp = HS15Model(x0)
solver = MadNLP.MadNLPSolver(nlp; print_level=MadNLP.INFO)
results = MadNLP.solve!(solver)
nothing
```

--------------------------------

### Solve JuMP Model on GPU with MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

This example demonstrates how to solve a JuMP model on the GPU by specifying `CuArray` as the `array_type` and `CUDSSSolver` as the `linear_solver` when optimizing.

```julia
using CUDA
using JuMP
using MadNLP, MadNLPGPU

model = Model(MadNLP.Optimizer)
@variable(model, x1 <= 0.5)
@variable(model, x2)

@objective(model, Min, 100.0 * (x2 - x1^2)^2 + (1.0 - x1)^2)
@constraint(model, x1 * x2 >= 1.0)
@constraint(model, x1 + x2^2 >= 0.0)

JuMP.set_optimizer_attribute(model, "array_type", CuArray)
JuMP.set_optimizer_attribute(model, "linear_solver", CUDSSSolver)
JuMP.optimize!(model) # Solve the model on the GPU!
```

--------------------------------

### Set Initial Guess from Solution

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/warmstart

Copy the previously found solution to the initial guess vector of the NLP model. This ensures the algorithm starts from a known point.

```julia
copyto!(NLPModels.get_x0(nlp), results.solution)
```

--------------------------------

### Get KKT Matrix

Source: https://madsuite.org/MadNLP.jl/stable/man/kkt

Retrieves a reference to the internal KKT matrix stored within the `kkt` object. This matrix is automatically updated whenever `kkt` is assembled.

```julia
kkt_matrix = MadNLP.get_kkt(kkt)
```

```julia
6×6 SparseArrays.SparseMatrixCSC{Float64, Int32} with 13 stored entries:
 1.0   ⋅    ⋅     ⋅     ⋅     ⋅ 
 2.0  5.0   ⋅     ⋅     ⋅     ⋅ 
  ⋅    ⋅   8.0    ⋅     ⋅     ⋅ 
  ⋅    ⋅    ⋅   10.0    ⋅     ⋅ 
 3.0  6.0  9.0    ⋅   12.0    ⋅ 
 4.0  7.0   ⋅   11.0    ⋅   13.0
```

--------------------------------

### Precompiling MadNLPGPU

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

This output indicates the precompilation process for the MadNLPGPU package, showing the time taken and successful completion.

```julia
Precompiling packages...
  14040.0 ms  ✓ MadNLPGPU
  1 dependency successfully precompiled in 16 seconds. 123 already precompiled.
```

--------------------------------

### Tune LBFGS max_history in MadNLP

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/quasi_newton

Increase the LBFGS `max_history` to 20 to use more past iterates for the low-rank approximation. This example also specifies the `LapackCPUSolver` and `MadNLP.CompactLBFGS` for the optimization.

```julia
qn_options = MadNLP.QuasiNewtonOptions(; max_history=20)
results_qn = madnlp(
    nlp;
    linear_solver=LapackCPUSolver,
    hessian_approximation=MadNLP.CompactLBFGS,
    quasi_newton_options=qn_options,
)
nothing
```

```text
This is MadNLP version v0.9.1, running with Lapack-CPU (BUNCHKAUFMAN)

Number of nonzeros in constraint Jacobian............:       30
Number of nonzeros in Lagrangian Hessian.............:      975

Total number of variables............................:       30
                     variables with only lower bounds:        0
                variables with lower and upper bounds:        0
                     variables with only upper bounds:        0
Total number of equality constraints.................:       10
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
   0  4.8810842e+01 2.22e-16 1.71e+01 0.00e+00  -1.0     -   0.00e+00  1  0
   1  2.2956342e+00 1.50e+03 2.04e+02 0.00e+00  -1.0     -   1.25e-01  1  4f
   2  4.4967927e+00 3.76e+02 5.48e+01 0.00e+00  -1.0     -   1.00e+00  1  1h
   3  8.7093834e+00 9.38e+01 1.37e+01 0.00e+00  -1.0     -   1.00e+00  1  1h
   4  1.4770500e+01 2.32e+01 3.20e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
   5  1.8176499e+01 9.94e+00 1.50e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
   6  2.5109575e+01 2.32e+00 1.92e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
   7  2.8512602e+01 2.52e+00 2.52e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
   8  1.2759872e+01 3.54e+01 1.13e+01 0.00e+00  -1.0     -   1.00e+00  1  1h
   9  1.5624022e+01 1.58e+01 2.83e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
  10  2.1982310e+01 5.68e+00 1.86e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  11  2.2542843e+01 4.22e+00 1.14e+00 0.00e+00  -1.0     -   5.00e-01  1  2h
  12  2.7255011e+01 1.47e+00 2.96e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  13  2.9232894e+01 1.25e+00 2.66e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  14  2.9261067e+01 9.81e-01 3.03e+00 0.00e+00  -1.0     -   5.00e-01  1  2h
  15  2.7142323e+01 2.39e+00 3.27e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  16  2.7776481e+01 1.35e+00 1.98e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  17  3.0166747e+01 7.86e-01 1.80e+00 0.00e+00  -1.0     -   1.00e+00  1  1h
  18  3.0609464e+01 3.09e-01 8.42e-01 0.00e+00  -1.0     -   1.00e+00  1  1h
  19  3.2450081e+01 6.00e-02 5.04e-01 0.00e+00  -1.7     -   1.00e+00  1  1h
iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
  20  3.2636130e+01 3.88e-02 3.94e-01 0.00e+00  -1.7     -   1.00e+00  1  1h
  21  3.2706987e+01 9.35e-03 1.66e-01 0.00e+00  -1.7     -   1.00e+00  1  1h
  22  3.2699721e+01 6.51e-03 9.41e-02 0.00e+00  -2.5     -   1.00e+00  1  1h
  23  3.2682575e+01 5.11e-03 1.14e-01 0.00e+00  -2.5     -   1.00e+00  1  1h
  24  3.2701207e+01 4.15e-03 1.48e-01 0.00e+00  -2.5     -   1.00e+00  1  1h
  25  3.2712954e+01 1.75e-03 4.88e-02 0.00e+00  -2.5     -   1.00e+00  1  1h
  26  3.2719381e+01 2.29e-04 2.32e-02 0.00e+00  -2.5     -   1.00e+00  1  1h
  27  3.2720051e+01 3.26e-05 6.65e-03 0.00e+00  -3.8     -   1.00e+00  1  1h
  28  3.2720061e+01 2.64e-05 7.96e-03 0.00e+00  -3.8     -   1.00e+00  1  1h
  29  3.2715805e+01 6.76e-04 2.85e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
iter    objective    inf_pr   inf_du inf_compl lg(mu) lg(rg) alpha_pr ir ls
  30  3.2702239e+01 2.73e-03 5.82e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  31  3.2697960e+01 2.28e-03 7.47e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  32  3.2707411e+01 2.46e-03 1.24e-01 0.00e+00  -3.8     -   1.00e+00  1  1h
  33  3.2709995e+01 9.33e-04 1.60e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  34  3.2712746e+01 6.45e-04 2.11e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  35  3.2704283e+01 3.24e-03 5.49e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  36  3.2709923e+01 1.48e-03 2.22e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  37  3.2715749e+01 4.96e-04 5.36e-02 0.00e+00  -3.8     -   1.00e+00  1  1h
  38  3.2714941e+01 4.33e-04 2.54e-03 0.00e+00  -3.8     -   1.00e+00  1  1h
```

--------------------------------

### Configure MadNLP with HybridKKT for GPU

Source: https://madsuite.org/MadNLP.jl/stable/tutorials/gpu

Use HybridKKT.jl for accurate solutions by employing the Golub & Greif augmented Lagrangian formulation. This strategy is more accurate but slightly slower than Lifted-KKT.

```julia
using HybridKKT

results = madnlp(
    nlp;
    linear_solver=MadNLPGPU.CUDSSSolver,
    kkt_system=HybridKKT.HybridCondensedKKTSystem,
    equality_treatment=MadNLP.EnforceEquality,
    fixed_variable_treatment=MadNLP.MakeParameter,
)
```

--------------------------------

### Instantiate MadNLP Solver

Source: https://madsuite.org/MadNLP.jl/stable/lib/ipm

Instantiate a new `MadNLPSolver` associated with the nonlinear program `nlp::AbstractNLPModel`. Options are passed as optional arguments. The constructor allocates memory for the interior-point algorithm.

```julia
MadNLPSolver(nlp::AbstractNLPModel{T, VT}; options...) where {T, VT}
```