JuMP register polynomial with its gradient

Specific Domains Optimization (Mathematical)
1

Dear all,

I am using jump and I want to register a polynomial function with its gradient, with

register(model2, :poly, num_var, poly, gradient)

I build the polynomial with AbstractAlgebra.jl and the gradient with the function

function gradient(x...)
    point = Float64[]
    for i in 1:length(x)
        append!(point,x[i])
    end
    gradient = Float64[]
    for i in 1:length(x)
        deriv = evaluate(derivative(f,i),point)
        append!(gradient,convert(Float64,deriv))
    end
    return gradient
end

If I call the gradient on a point with

gradient(zeros(100))

there is no error. However, when I solve my problem with Ipopt as follows

num_var = 100
model = Model(Ipopt.Optimizer)
register(model, :poly, num_var, poly, gradient)
@variable(model, 0.0 <= z[1:num_var] <= 1.0)
@NLobjective(model, Min, poly(z...))
optimize!(model)

JuMP return me the following error message referring to the evaluation step in the gradient of the function.

ERROR: LoadError: Incorrect number of values in evaluation

I would ask you if you have any insight or suggestion to solve this error. Thanks very much

2

Your gradient function need to accept the gradient as the first argument and modify it in place.

https://jump.dev/JuMP.jl/stable/manual/nlp/#Multivariate-functions