Originally posted by: SystemAdmin

Hi, all,

I have a MIP instance. With CPLEX default parameters, I can solve this instance within 0.28 seconds. But some binary varaibles have values such as 0.000001 and 0.999999.

Then I tried to set MIP tolerance integrity to 0 and SIMPLEX tolerance feasibility to the minimum value, 1e-9. Now the solution course is significantly slowed down and the tree size is very large. After 1800 seconds, the GAP is still 26.32% and the tree size is 120 M. Finally CPLEX failed to solve this instance and terminated out of memory.

Is there any to figure out such a numerical trouble?

Thanks.

Shaon
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: EdKlotz

> shaonli wrote:
> Hi, all,
>
> I have a MIP instance. With CPLEX default parameters, I can solve this instance within 0.28 seconds. But some binary varaibles have values such as 0.000001 and 0.999999.
>
> Then I tried to set MIP tolerance integrity to 0 and SIMPLEX tolerance feasibility to the minimum value, 1e-9. Now the solution course is significantly slowed down and the tree size is very large. After 1800 seconds, the GAP is still 26.32% and the tree size is 120 M. Finally CPLEX failed to solve this instance and terminated out of memory.
>
> Is there any to figure out such a numerical trouble?
>
> Thanks.
>
> Shaon

Take a look at http://www-01.ibm.com/support/docview.wss?uid=swg21399984
for a detailed discussion of this. In a nutshell, examine
the constraints containing integer variables that took values such
as .000001 in the original run. They probably have have large coefficients.
Try either reducing those coefficients if that doesn't compromise the
meaning of your model, or replacing those constraints with indicator
constraints if it does.
#CPLEXOptimizers
#DecisionOptimization