follower as maximization problem

[leomink]

I found this very weird behavior on one instance that I wanted to share with you.

When I solve the problem with the auxiliary file presenting the follower as maximization (problem.mps paired with followerasMax.aux), I get an output which I am pretty sure is not bilevel feasible.

Now if I solve the same problem but with the auxiliary file presenting the follower as minimization (testProblem.mps paired with followerasMin.aux), then I get an optimal solution to the problem. Now the only difference between the auxiliary files is OS=1 vs OS=-1 and the same exact coefficients but multiplied by (-1).

Thanks in advance,

Leo

problem.txt
followerasMin.txt
followerasMax.txt

[yuxies]

Hello Leo,

Could you provide the version of MibS you used to solve the problem?

I tried to solve the problem using the latest version of stable/1.2 and the two different aux files. But I received the same solution below. Could you also help confirm whether these look like what you saw ealier?

Optimal solution:
Cost = 494
First stage (upper level) variable values:
Second stage (lower level) variable values:
Y2 = 1
Y7 = 1
Y8 = 1
Y9 = 1
Y15 = 1
Y17 = 1
Y22 = 1
Y23 = 1
Y26 = 1

Best,
Yu