Is it necessary to apply normalization to unknown parameters with different scales?

Hello everybody!

I have a model built in AMPL. The scales of my unknown values are different. Two of them are coordinates, for the another; one has a scale between 2 and 9, and the other is between -150 and 50.

I want to apply some solvers like “Ipopt” and “Gurobi.” However, I’m not sure if it’s necessary to apply normalization to these values before using the solvers?

Thank you!

You don’t need to worry about scaling when the differences are this small. Also, many solvers, including Ipopt and Gurobi, scale the problem automatically, so they usually work on a well-normalized problem. You might want to think about improving your problem’s scaling if it has ranges than differ by a factor of more than 10^6, however.

Gurobi has a good discussion of scaling in https://www.gurobi.com/documentation/current/refman/tolerances_and_user_scalin.html and the subsections listed at the bottom of that page.

1 Like

thank you for your response, the coordinates means latitude and longitude in a small space.
According to these, is this ok again?

I expect that would be OK.

1 Like