Postsolve indicates that solution to FO is not reliable

[Carlos Bustos]

Hi all, 

when I run a MIP, the model provides a solution but it shows me the following message:

Postsolve: there is a big difference between the optimum value of the final LP

and the optimum value that CPLEX returned first.

Warning: Postsolve indicates that solution to FO is not reliable.

Why is this? How can you fix this?

Thanks!!

[Marcel Hunting]

Hi,

CPLEX will find a solution according to its tolerance settings. For example, the default setting of the CPLEX option Integrality is 1e-5 and therefore CPLEX could return 0.99999999 for a binary variable. In the postsolve step AIMMS rounds all integer variables to their nearest integer value, and the LP model derived in this way is solved by CPLEX. As a result you obtain a solution in which all integer variables have a true integer value.

That warning indicates that CPLEX might not have found a true optimal solution for the MIP. Often this warning indicates that the model is poorly scaled in which case you have to improve the scaling of the model. Instead, you could also try setting the CPLEX option 'Integrality' to 0.

Best regards,

Marcel Hunting

AIMMS

[MPS]

hi Marcel Hunting,
In similar situation,when Aimms says this message, is the solution correct and feasible? because when i do what you say And run again,

 After a long time

 i see that solver says Best LP Bound is 69 while Best solution is 70 (gap=1.40%) and Continues to run With the previous results.(Best LP Bound AND Best solution that Obtained by default integrality in Little time are 69 and my model is minimizing). i can't check my model because of the complexity...