Originally posted by: MachexC

Hi Alex,

My model runs very fast when I scale the data for by 1e-11 to get smaller than original data. But I'm worried if I scale the data for by 1e-11 because as you know objective coefficients are below CPLEX's default optimality tolerance of 1e-6. I am worry about the narrow model is essentially the same as solving a model reach to zero objective. Thus, given CPLEX's tolerance, the model is over optimality tolerance of 1e-6? and could you show how I check it?

Please attached engine log and statistics.

Thank you for help.

Bests and regards,

Mac

If you scale your data by 1e-11 then it probably solves fast because many coefficients are essentially ignored for being too small. So may end up solving a different model. I am not sure your approach is reasonable. What are the minimum and maximum absolute values of your coefficients with and without scaling?

Originally posted by: MachexC

Hi Daniel Junglas,

The minimum and maximum absolute values of coefficients without scaling are 400 and 433600. Is it ok?

Thank you,

Mac

This range of coefficients looks fine. But with these numbers scaling by 1e-11 seems really questionable. Did you compare the sizes of the presolved models with and without your scaling? Just to make that your scaling does not make many numbers so small that they are considered 0. And even if that is not the case, you will have a lot of numbers below or very close to CPLEX feasibility/optimality tolerances. That is usually not a good idea.

Also note that scaling the objective affects the absolute gap termination criterion!