JuMP new nonlinear interface: adding constraint programatically

Specific Domains Optimization (Mathematical)
1

ModelPredictiveControl.jl is still using the legacy syntax for NLP, but I’m starting the migration right now. The nonlinear inequality constraints are currently added programmatically with (simplified for the purpose of the explanation):

using JuMP, Ipopt
model = Model(Ipopt.Optimizer)
nx = 2
@variable(model, x[1:nx])
gfunc_i(i, x) = 10x[i]^2 - 10
gfunc = [(x...) -> gfunc_i(i, x) for i in 1:nx]
ymin, ymax = [-1, -1], [1, 1]
for i in eachindex(ymin)
    sym = Symbol("g_ymin_$i")
    register(model, sym, nx, gfunc[i], autodiff=true)
    add_nonlinear_constraint(model, :(-$(sym)($(x...)) <= -$(ymin[i])))
end
for i in eachindex(ymax)
    sym = Symbol("g_ymax_$i")
    register(model, sym, nx, gfunc[i], autodiff=true)
    add_nonlinear_constraint(model, :(+$(sym)($(x...)) <= +$(ymax[i])))
end

is there an equivalent for the add_nonlinear_constraint method in the new NLP syntax? I’ve tried add_constraint but I’m not sure how to use it since the docstring is not very detailed:

help?> add_constraint
search: add_constraint add_nonlinear_constraint

  add_constraint(model::GenericModel, con::AbstractConstraint, name::String="")

  Add a constraint con to Model model and sets its name.

Thanks for the help :slight_smile:

1 Like
2 
using JuMP
gfunc_i(i, x) = 10x[i]^2 - 10
ymin, ymax = [-1, -1], [1, 1]
nx = length(ymin)
model = Model()
@variable(model, x[1:nx])
# Option 1: function tracing
@constraint(model, [i in 1:nx], ymin[i] <= gfunc_i(i, x) <= ymax[i])

# Option 2: operator
for i in 1:nx
    g = (x...) -> g_func_i(i, x)
    op = add_nonlinear_operator(model, nx, g; name = Symbol("gfunc_$i"))
    @constraint(model, ymin[i] <= op(x...) <= ymax[i])
end

For option 1, see: Nonlinear Modeling · JuMP

For options 2, see: Nonlinear Modeling · JuMP

1 Like
3

Wow, it is way cleaner like that haha. The operator option works well, thanks!

1 Like
4

Wow, it is way cleaner like that haha

Yip! I put a bit of thought into making sure things like this were now well supported :smile:

1 Like
5

It shows, I really like the new syntax.

1 Like