Basic under/over approximation operations.

[daemontus]

An over/under approximation of a BDD can be easily obtained by erasing the smallest cube. Other heuristics can be considered as well, for example the cube can be pseudo-randomly selected (probably with some weight given to the size of the cube). This reduces the size of a BDD while minimising loss of information, which is useful for approximative techniques.

[daemontus]

This does not seem to be as simple as expected. We definitely need some further research into such algorithms and hence push it to 0.4.0.