09-15-2017 11:47 AM
I am trying to solve an MINLP problem by using CEIL function in the formulation for integer variables. So technically, the NLP solver is able to solve the equations as they underlying variables are all real. However, at some cases the solver doesnt not converge and I was wondering if changing step size could help in solving the problem. I tried with CDIGITS and FDIGITS, both = 0.01 however I did not find any improvement or evidence of algorithm taking steps of 0.01 from log of iterations.
Please help if anybody has previous experience on working with step size in a PROC OPTMODEL NLP solver.