[Ipopt] 2 function evaluations for 1 iteration

[Hans Pirnay]

Hi Simon,

>
> for a rather time critical application I limited the maximal iterations to
> 1. The IPOPT algorithm uses two function evaluations (objective, gradient,
> inequality constraint, inequality constraint Jacobian), see below.
>
> How does IPOPT need 2 function evaluations for 1 iteration?
>
> Is there an option in IPOPT to reduce the function evaluation to 1?

In this one iteration that you allow, Ipopt does two things:

1. compute the values at the starting value
2. compute a search direction and an acceptable trial point using a line search

The 'acceptable trial point' calculation is where your second function
evaluations take place. It could even be more than one additional
function call if the line search fails.
Actually, there are probably even more function calls that are done
before the main iteration of IPOPT when setting up the problem, for
example for computing the scaling parameters.

Hope this helps
Hans