Strict inequalities in fmincon

Hello,

I'd like to use fmincon with strict inequalities (A.x<b and not A.x<=b). I've looked on different websites but I didn't find any information about this. Is it possible? How can I do that ?

Thanks for helping

Tom

Answer 1

[the cyclist]

I don't have the optimization toolbox, so I cannot test this idea, but I wonder if you can effectively handle this by including "eps" (the floating point spacing) as part of the input specification of "b".

Answer 2

[Steve Grikschat]

Effectively, you're defining a new strict inequality A*x <= b-e where "e" can be arbitrarily small. As the cyclist pointed out, there are floating point limitations to how small "e" can be.

However, also note that solvers (including fmincon) have tolerances for constraint violation as well. In this case, it means that at a solution, your worst case now is A*x <= b-e + TolCon (possibly with a relative factor on TolCon). You should factor that in if you need to satisfy the strict inequality.