Abstract vs Concrete Models

[sserita]

We have been toying around with splitting models into "abstract" and "concrete" models. Our current models in pyGSTi are "concrete" models, whereas an abstract model would be one that has some sort of lazy evaluation that allows for an unspecified state space. This description is a little vague because we're not sure yet what that API would look like, but one of the clear use cases is when we want to have an context-independent crosstalk-free model for a large device, but we only want to perform simulations on small subsets of the device. Currently, you would have to either initialize a model for the entire device (which may be infeasible for certain evotypes on large numbers of qubits), or build a series of processorspecs for each qubit subset and regenerate the model as needed. We could instead having an ImplicitOpModel-like object that lazily allocate per-qubit representations with layer rules that only built to a subset of the full device state space. Basically, we should able to decouple the representation of the device from the layers we build up for forward simulation.