Integer Optimization: performance help

[odow]

I don’t really understand what your nonanticipativity constraints are doing, or why they are quadratics.

You could consider doing this binary * binary to linear reformulation:

Specifying datatype in JuMP expressions and constraints to save memory

You may be able to reformulate as a linear program, or at least a linearly-constrained one. This may not help the memory by added even more objects, but it may help the solver. Since r_i_j and y_i_k are binary variables, their product can be replaced with an additional variable and four sparse inequalities. This is the McCormick envelope, but enforcing binary variables means it is an exact reformulation (e.g. cf. McCormick envelopes - optimization). Setting x = ry, with r and y binary variabl…