Skip to main content

 

I have large a planning problem in AIMMS, which is linear optimization. I am using GUROBI 9.0 for solving it and the solver is set to use the Barrier method based on our experience that Barrier converges faster than Simplex in our large-scale problem. 

After around 56 hours, the program stopped and I got the “Warning: Model status 13: Error no solution, solver status 1: Normal Completion”.

Here are screenshots from the warning and progress panels of AIMMS. 
 

Error/Warnings Panel
Progress Panel
​​​​​

None of the existing questions and discussions on “Warning: Model status 13” are related to the case that it says “Solver status 1: Normal Completion”. I could not find any more information on software or the AIMMS manual regarding this issue. I need to find what is the source of this problem. I would be grateful if you can help me with this issue.

Thanks,

Hi @rezakhalili. It seems that you are solving a MIP model and that Gurobi terminated while solving the root node. The combination of ‘Error no solution’ and ‘Normal completion’ only seems to occur if the optimization was terminated due to unrecoverable numerical difficulties (but then AIMMS should also print a warning in the Messages Window, and in the file ‘messages.log’ in the log folder). The very large number of iterations during the crossover phase of the barrier algorithm also is a strong indication that the model has numerical issues.

You could print the Gurobi log file by setting the following options:

  • Solvers General option 'Solver window messages' to 'All';
  • Gurobi option 'Output file' to 'Yes'.

The Gurobi log file will be printed in the log folder. It often contains useful information.

It could be that the model has scaling issues. If you can share the project then I can take a look. You can also send an email to our support, with a reference to this topic, if your project contains sensitive information.


Thank you very much @Marcel Hunting 

I should clarify that it's a MIP model but for this run, all of the binary variables are relaxed. Hence, I am solving a linear optimization. I will follow all of your recommendations regarding my issue. Also, I would be grateful if you can provide me with any tutorials on finding and solving numerical issues in my model using AIMMS if there exist. I believe AIMMS has some built-in features regarding these issues, which I am not very familiar with. Unfortunately, I cannot share my project with anyone unless an NDA is in place, which seems to be impossible.


You can use the Math Program Inspector to inspect the matrix for large coefficients or coefficients very close to 0 by using the Matrix View. You can also use the Scaling Tool in the Math Program Inspector to see whether that helps to improve the scaling.

To open the Math Program Inspector AIMMS needs to generate the math program first. If is sufficient to solve the model using a time limit of, e.g., 10 seconds. You can also use the function GMP::Instance::Generate to generate the math program without solving it.

AIMMS can setup an NDA if needed but maybe the bottleneck is on your side.


Reply


Didn't find what you were looking for? Try searching on our documentation pages:

AIMMS Developer & PRO | AIMMS How-To | AIMMS SC Navigator