Initial guess in MINLP

#1

Hello,
Something strange happens with my program.

using JuMP, Ipopt, CPLEX, Juniper
m = Model(solver=juniper)
@variable(m, x, start = 0.0)
@variable(m, y, start = 0.0)
@NLobjective(m, Min, sin((1-5x))^2 + 100(cos(y-x^2))^2)
solve(m)
println("x = ", getvalue(x), " y = ", getvalue(y))

where, I set the initial guess to x = 0 and y = 0. The solution provided is

Obj: 7.434409052608867e-20
x = 0.828318530664551 y = -0.8846847385581247

I want to resolve this MINLP and change the starting point to optimize. However, when I change the starting point, the optimal solution obtained is the same. This not happen.

Can anybody help me or give a advise how can I set the initial guess in MINLP?

#2

It doesn’t look like Juniper supports warm-starts:

#3

This should be a fairly straightforward feature to add. I would recomend posting a feature request as an Issue in the Juniper repo.

#4

Der collegue,
How can I do this? Is there a Juniper repo?
Thank you!

#5

Go to https://github.com/lanl-ansi/Juniper.jl/issues and click “New issue”