NLPSaUT

📚Read the docs here!📚

The NLPSaUT module constructs a JuMP model for a generic nonlinear program (NLP). The expected use case is solving a differentiable (either analytically or numerically) nonconvex NLP with gradient-based algorithms such as Ipopt or SNOPT.

The user is expected to provide a "fitness function" (pygmo-style), which evaluates the objective, equality, and inequality constraints. Below is an example:

function f_fitness(x::T...) where {T<:Real}
    # compute objective
    f = x[1]^2 - x[2]
    
    # equality constraints
    h = zeros(T, 1)
    h = x[1]^3 + x[2] - 2.4

    # inequality constraints
    g = zeros(T, 2)
    g[2] = -0.3x[1] + x[2] - 2   # y <= 0.3x + 2
    g[1] = x[1] + x[2] - 5      # y <= -x + 5
    return [f; h; g]
end

Derivatives of f_fitness is taken using ForwardDiff.jl (which is the default JuMP behavior according to its docs); as such, f_fitness should be written in a way that is compatiable to ForwardDiff.jl (read here as to why it is ForwardDiff, not ReverseDiff). For reference, here's the JuMP docs page on common mistakes when using ForwardDiff.jl.

The model constructed by NLPSaUT utilizes memoization to economize on the fitness evaluation (see JuMP Tips and tricks on NLP).

Quick start

1. git clone this repository
2. start julia-repl
3. activate & instantiate package (first time)

pkg> activate .
julia> using Pkg                # first time only
julia> Pkg.instantiate()        # first time only

4. run tests

(NLPSaUT) pkg> test

Note

• To use with SNOPT, it's probably better to go through GAMS.jl rather than to use SNOPT7.jl directly (installing SNOPT7.jl currently errors on julia v1.10)

Examples

For examples, see the examples directory.

TODO notes

• Development

  • Finite difference gradient option
  • Analytical gradient option

• Examples/documentation

  • Simple example with Ipopt
  • Simple example with SNOPT via GAMS
  • Example with ODEProblem
  • Demonstration of memoization benefits
  • Parallelized fitness function