Originally posted by: SystemAdmin

Hi, there.
I found a very interesting problem when I was trying to solve a MIP problem. I try to improve the computation by solving it from a feasible solution which is read from MST file. What is interesting is that cplex gave the message to me that so solution is refined from the MST while the MST file has the solution obtained from the exactly the same problem by only relaxing the relative gap to a larger number...
So may anyone explain this to me? Does it matter with the inequality/equality constraints tolerance? I really appreciate for any help.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

You should make sure that your starting solution satisfies the starting formulation. If your provided solution is infeasible, I think that CPLEX will keep it and try to repairt it.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Yes, I think the solution is feasible because it is obtained from exactly the same formulation.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

This could be a numerical issue. For example, if the solution provided by CPLEX satisfies integrality only within the tolerance (default is 1e-5), and such integer variables are member of constraints with large coefficients, then it can happen that this solution is rejected as a MIP start.

But this is pure speculation. For a better answer, I would need to get your model and your MST file.
#CPLEXOptimizers
#DecisionOptimization