Description of the Basic Ideas

The combinatorial optimization problem is rewritten as a single (quadratic unconstrained binary) objective function. The usual way to build it up is to use the following structure: In the Instance we describe the problem defining parameters. From the instance, we construct the ObjectiveTerms, containing the different contributions to the objective function, in particular the ones derived from problem constraints. The objective terms can be implemented directly or derived from a ConstrainedObjective, which contains the objective function and multiple constraints, implemented as ConstraintBinary. The objective terms can now be used to create the Objective by summing up the single terms weighted with a certain so-called penalty weight.

All objective objects contain Polynomials representing the functions. There are special polynomials, PolyBinary and PolyIsing, which take advantage of the restriction to either binary (0 or 1) or spin (-1 or 1) variables.

The ScipModel is an interface to the classical MILP solver SCIP, which can solve a ConstrainedObjective or a (small enough) Objective for comparison. In Solution, we store not only the optimal variable assignment but also further information, like runtime etc., which are obtained during the solving process.

Furthermore, we have the HardwareAdjacency and the Embedding, which are useful when dealing with actual hardware.

All mentioned objects also provide methods to store and load their information in and from hdf5 files.