Infeasible model after consecutive invocation

As per your instructions, I created the 3 exports (model1.lp, model2.lp, model3.lp). Model1.lp and model3.lp are indeed identical. Model2.lp indeed has a subset of variables fixed. I can solve each of these LPs through the interactive optimizer, and the results are as expected: Model1.lp and Model3.lp produce a feasible solution, Model2.lp is infeasible.

I am not sure I understood correctly: are you saying that model B is found infeasible only if it is solved right after model A in your application but when solved standalone then it is found feasible? In particular, if you solve the original model before even creating model B, is that found feasible?

That is correct. In fact, if I disable steps 2, 3, 4 and 5 in the above workflow, and hence solve the original model with no variables fixed then I obtain a correct, feasible solution.

I played around with the PRESOLVEOPS parameters but was unable to find a combination of parameters that allowed me to solve model B correctly. Only setting PRESOLVE=0 allows me to solve model B (I did not attempt an exhaustive search over all 2^19 PRESOLVE parameter combinations..).

I also tried inserting the following lines right before invoking xprs.mipOptimize("p"):

xprsProb.writeProb(config.OUTPUT_DIRECTORY+"/"+solveInstanceName, "l");
 
xprsProb.readProb(config.OUTPUT_DIRECTORY+"/"+solveInstanceName, "v");

This simply writes the problem to a file, and immediately reads the same file again. This solves the issue, i.e. model A is again infeasible, and model B produces a feasible solution. Also adding the following 2 lines right before invoking mipOptimize solves the problem:

xprsProb.saveAs(config.OUTPUT_DIRECTORY+"/"+instanceName);
xprsProb.restore(config.OUTPUT_DIRECTORY+"/"+instanceName);

I've email the requested svf to you ([email protected])