fix mapping of meta data for variable bounds

[svigerske]

Mapping from full x to x without fixed vars was not taken into account.

@danieloliveira56 I think this should fix the issue with the variable names for variable bounds that you reported to GAMS support.
It looks to me like when going from full x to the reduced x (fixed variables removed), things were ok and that's why we got

initial x unscaled[    1]{y}= 0.0000000000000000e+00
initial x unscaled[    2]{z}= 0.0000000000000000e+00
initial x unscaled[    3]{w}= 0.0000000000000000e+00

But for the lower variable bounds (x_L), both the mapping from full x to the reduced x and then from the reduced x to the variables with lower bound needs to be taken into account. But only the latter was used in the code. (same for x_U)

So here I change it to use pos_idx (full x -> reduced x) and pos_idxL (reduced x -> x_L).
I now get

DenseVector "original x_L unscaled" with 3 elements:
original x_L unscaled[    1]{y}=-2.0000000000000000e+00
original x_L unscaled[    2]{z}=-3.0000000000000000e+00
original x_L unscaled[    3]{w}=-4.0000000000000000e+00
DenseVector "original x_U unscaled" with 3 elements:
original x_U unscaled[    1]{y}= 2.0000000000000000e+00
original x_U unscaled[    2]{z}= 3.0000000000000000e+00
original x_U unscaled[    3]{w}= 4.0000000000000000e+00

As you are an expert on this kind of Ipopt code (#570), I thought I may ask you to have another look. :)

[danieloliveira56]

Hi Stefan,

Yes, the variable naming for my model makes sense now.
I see the exact same output for x_L and x_U with holdfixed=0 and 1.

Looks good to me, thank you!

[svigerske]

Thank you. I'll put this into Ipopt 3.14.9 and GAMS 40.1.0.