Infinite minimization with mmxnlp module

Question

I'm studying nonlinear optimization on Xpress Mosel with mmxnlp module.

How to do infinite minization to find better local minimum on every step?

Only so?:

while (true)
do
minimize(Obj)
writeln(X.sol)
copysoltoinit
end-do

Answer 1

The problem with a local minimum is that the solver has no idea how to continue, that is why it stops there. So there is no point in going on.

There is another way to find more solutions, though: multistarts. With multistarts you can ask the solver to start the search from multiple different points. With that it is likely to find different local minima.

In order to use multistart with mosel, you use the addmultistart function. There also is an example for this in the public example repository. That example does

addmultistart("random points", XNLP_MSSET_INITIALVALUES, 50)

to make the solver start from 50 random points.