Decomposition Algorithms: - Dantzig-Wolfe Decomposition: If the problem has a block structure, decompose it into smaller subproblems that can be solved independently and then combined. - Benders Decomposition: For problems with complicating variables, separate the problem into a master problem and subproblems, solving them iteratively.

- Deactivate constraints - Relax binary / integer variables - Test on tiny instances - Play with the solver parameters Not so much different than any classical programming troubleshooting challenge.

If there is a time dimension, break the problem into smaller time ranges. Note that deterministic approximations do not necessarily make a problem easier. Optimizing a sequential problem over multiple time periods because we assume we know the future can make the problem much harder.

- Develop unit tests (small examples that should work with your complex model) - Deactivate teh constraints one by one and run the model to spot the problem making consraints. - Visualize the output to find the problem - Print the values of inoput values to the functions (sometimes the value that the function gets is not what we expect) - Relax the integer variables and check the feasibility

In my experience, there are several ways of identifying constraints that make a model infeasible. Some of them are: - Isolate them. Start by reducing the set of constraints of your problem and adding them one by one. That way you know which one is breaking the model. - Move the constraint as a penalty of the objective function. That way you will spot constraints that breaks the model without actually breaking it. - If you're using Google OR-Tools, it's good to have binary variables that apply only when a constraint is activated. Each solver may have different ways of spotting them.