Sometimes I got an optimal solution using CPLEX but it is not the minimal one. I got a minimal solution using some other solvers. When I force the other decision variables (using other solvers) solution in CPLEX I got a solution that is less than the optimal. Some times the difference is so big. Optimality tolerance is the default one 1.0E-6. Please advice on this issue.


I am not able to attach any file for a technical issue.

------------------------------
Tareq Oshan
------------------------------

#DecisionOptimization

Do you set the absolute and relative MIP gap to 0 and all resource limits to infinity? By default absolute and relative MIP gap are not 0. This allows CPLEX to stop with (slightly) non-optimal solutions. If you set a time limit then CPLEX may of course stop with non-optimal solutions.

What is the solution status when you observe suboptimal solutions being claimed optimal? Can you share one of the offending models, the solution CPLEX produces and the solution you got from another solver? If you cannot share these, can you share the log for the offending CPLEX solve?

------------------------------
Daniel Junglas
------------------------------

Thanks Daniel,

attached are the CPLEX files. The solution is 70159 \times 10^6. But using other solvers I got different decision variables with a solution of 3.26 \times 10^6. Forcing the decision variables of the other solver into CPLEX will get us a lower solution than the one we got using CPLEX.

------------------------------
Tareq Oshan
------------------------------

Original Message:
Sent: Wed July 22, 2020 06:41 AM
From: Daniel Junglas
Subject: Optimal

Do you set the absolute and relative MIP gap to 0 and all resource limits to infinity? By default absolute and relative MIP gap are not 0. This allows CPLEX to stop with (slightly) non-optimal solutions. If you set a time limit then CPLEX may of course stop with non-optimal solutions.

What is the solution status when you observe suboptimal solutions being claimed optimal? Can you share one of the offending models, the solution CPLEX produces and the solution you got from another solver? If you cannot share these, can you share the log for the offending CPLEX solve?

------------------------------
Daniel Junglas
------------------------------

In your OPS file I can see that you set the time limit to 86400 and the dettime limit to 86400 as well. This may result in premature terminate. Again, what is the solution status returned by CPLEX? Are you sure CPLEX claims the solution optimal.

Also, given that your model is written in OPL and that does only support CPLEX, how do you feed your problem to other solvers. Are you absolutely sure you feed the exact same model to the other solvers?

------------------------------
Daniel Junglas
------------------------------

Original Message:
Sent: Wed July 22, 2020 07:27 AM
From: Tareq Oshan
Subject: Optimal

Thanks Daniel,

attached are the CPLEX files. The solution is 70159 \times 10^6. But using other solvers I got different decision variables with a solution of 3.26 \times 10^6. Forcing the decision variables of the other solver into CPLEX will get us a lower solution than the one we got using CPLEX.

------------------------------
Tareq Oshan
------------------------------


Original Message:
Sent: Wed July 22, 2020 06:41 AM
From: Daniel Junglas
Subject: Optimal

Do you set the absolute and relative MIP gap to 0 and all resource limits to infinity? By default absolute and relative MIP gap are not 0. This allows CPLEX to stop with (slightly) non-optimal solutions. If you set a time limit then CPLEX may of course stop with non-optimal solutions.

What is the solution status when you observe suboptimal solutions being claimed optimal? Can you share one of the offending models, the solution CPLEX produces and the solution you got from another solver? If you cannot share these, can you share the log for the offending CPLEX solve?

------------------------------
Daniel Junglas

I changed the time limit and it does not work. 

I used LINGO and it took long time to get a way lower solution that the one CPLEX got me. I was sure that the codes are equivalent. When I forced the decision variables that I got from LINGO into CPLEX I got the same solution in both LINGO and CPLEX which means that the codes are representation the same problem.

------------------------------
Tareq Oshan
------------------------------

Original Message:
Sent: Wed July 22, 2020 09:10 AM
From: Daniel Junglas
Subject: Optimal

In your OPS file I can see that you set the time limit to 86400 and the dettime limit to 86400 as well. This may result in premature terminate. Again, what is the solution status returned by CPLEX? Are you sure CPLEX claims the solution optimal.

Also, given that your model is written in OPL and that does only support CPLEX, how do you feed your problem to other solvers. Are you absolutely sure you feed the exact same model to the other solvers?

------------------------------
Daniel Junglas
------------------------------


Original Message:
Sent: Wed July 22, 2020 07:27 AM
From: Tareq Oshan
Subject: Optimal

Thanks Daniel,

attached are the CPLEX files. The solution is 70159 \times 10^6. But using other solvers I got different decision variables with a solution of 3.26 \times 10^6. Forcing the decision variables of the other solver into CPLEX will get us a lower solution than the one we got using CPLEX.

------------------------------
Tareq Oshan


Original Message:
Sent: Wed July 22, 2020 06:41 AM
From: Daniel Junglas
Subject: Optimal

Do you set the absolute and relative MIP gap to 0 and all resource limits to infinity? By default absolute and relative MIP gap are not 0. This allows CPLEX to stop with (slightly) non-optimal solutions. If you set a time limit then CPLEX may of course stop with non-optimal solutions.

What is the solution status when you observe suboptimal solutions being claimed optimal? Can you share one of the offending models, the solution CPLEX produces and the solution you got from another solver? If you cannot share these, can you share the log for the offending CPLEX solve?

------------------------------
Daniel Junglas

What exactly do you mean by "it does not work"? And what version of CPLEX do you use? When I solve your model with 12.10 here then after half a day the process is not finished but I have  feasible solution about 3.3e6 and a gap of 0.85%, which means that the solution is much better than what you report and an objective value of 3.26e6 is not ruled out yet.

Can you show the full engine log of the offending CPLEX run?

------------------------------
Daniel Junglas
------------------------------

Original Message:
Sent: Wed July 22, 2020 10:50 AM
From: Tareq Oshan
Subject: Optimal

I changed the time limit and it does not work. 

I used LINGO and it took long time to get a way lower solution that the one CPLEX got me. I was sure that the codes are equivalent. When I forced the decision variables that I got from LINGO into CPLEX I got the same solution in both LINGO and CPLEX which means that the codes are representation the same problem.

------------------------------
Tareq Oshan
------------------------------


Original Message:
Sent: Wed July 22, 2020 09:10 AM
From: Daniel Junglas
Subject: Optimal

In your OPS file I can see that you set the time limit to 86400 and the dettime limit to 86400 as well. This may result in premature terminate. Again, what is the solution status returned by CPLEX? Are you sure CPLEX claims the solution optimal.

Also, given that your model is written in OPL and that does only support CPLEX, how do you feed your problem to other solvers. Are you absolutely sure you feed the exact same model to the other solvers?

------------------------------
Daniel Junglas


Original Message:
Sent: Wed July 22, 2020 07:27 AM
From: Tareq Oshan
Subject: Optimal

Thanks Daniel,

attached are the CPLEX files. The solution is 70159 \times 10^6. But using other solvers I got different decision variables with a solution of 3.26 \times 10^6. Forcing the decision variables of the other solver into CPLEX will get us a lower solution than the one we got using CPLEX.

------------------------------
Tareq Oshan


Original Message:
Sent: Wed July 22, 2020 06:41 AM
From: Daniel Junglas
Subject: Optimal

Do you set the absolute and relative MIP gap to 0 and all resource limits to infinity? By default absolute and relative MIP gap are not 0. This allows CPLEX to stop with (slightly) non-optimal solutions. If you set a time limit then CPLEX may of course stop with non-optimal solutions.

What is the solution status when you observe suboptimal solutions being claimed optimal? Can you share one of the offending models, the solution CPLEX produces and the solution you got from another solver? If you cannot share these, can you share the log for the offending CPLEX solve?

------------------------------
Daniel Junglas