Numerical instabilities

[lukovdm]

Dear PAYNT team,

While using PAYNT to model check some POMDPs I have come across some numerical instability problems. I have now found an example where cegis and ar disagree on whether there is a counterexample, and the counterexample found seems to be wrong. Changing the linear_equation_solver_type and the minmax_solver_environment also of course have impact on the numerical instabilities found. I have attached the example below.
Sketch.zip

How could I make the numerical instabilities less impactful? Would an exact method be possible?

Let me know if you have any questions!

• Luko @ Radboud