I have been working with optimization based bound tightening code in PowerModels for a while. I have a question about setting solver tolerance in which the OBBT code can work for different test cases without any warning (e.g. adjust tolerances in solver to avoid the issue). Can you please let me know what is the best value of solver tolerance that can work for most test systems? Thanks in advance!
Given the numerical accuracy of most modern solvers I don’t think their is a simple or ideal answer to this question. In the release version of PowerModels we choose values that seemed to be robust across a multiple solvers and test cases.
In this report, we considered for the following slightly different tolerances, which seemed to be suitable for the specific case of using Guorbi v8.0,