Pulse class: consider changing description or code refactoring

[nathanshammah]

It has been brought up that the Pulse class can be confusing for the user, as it brings together control pulses, the relative Hamiltonian, and noise added to the control pulses. This issue is opened to discuss this point and see what action to take.

There is also some conceptual overlap between qutip-qip.Pulse and qutip.QobjEvo (and qutip.Qobj for the time-constant part).

Right now the Pulse class is also discussed in https://qutip-qip.readthedocs.io/en/latest/qip-processor.html#modelling-quantum-hardware-with-processor.

The actions to be taken can be discussed here, for example:

• A. Code refactoring.
• B. Remove pulse from the public APIs
• C. Reframing the presentation of Pulse in the documentation (and white paper)

[BoxiLi]

My preferred solution would be:

• Add a Model class that represents the Hamiltonian model in Processor (see below).
• Bring the introduction of Pulse a bit down as an advanced use, both in the paper and in the tutorial. Phrase it as a data class rather than focusing too much on the concepts. It is different from QobjEvo because it separates noise from the ideal dynamics. And it needs to handle concatenation of coefficients, which is not a problem of QobjEvo.

The biggest problem I see is that the list of Pulse is partially initialized at the initialization of a Processor (including only the Hamiltonians, no coeff), which is indeed a less-optimal design. So I suggest to add a Model class that contains only Hamiltonian information. The class takes the hardware parameters to initialize

class Model():
    def __init__(params):
        ...

model = SpinChainModel({"h":0.01, ...})

and has a `call' that returns the corresponding Hamiltonian. The Hamiltonian Qobj can be obtained by

model("sx0")

or whatever identifier that matches the implementation of the class, i.e.

model("sx", 0)

Advantages are:

• Hamiltonian models are now generated separately, there is not need for a partially initialized Pulse.
• There is no need to list all possible Hamiltonians ahead, which may take lots of memory for all-connected systems like ion trap or atoms.

@nathanshammah @quantshah Any opinions?

[BoxiLi]

I also mentioned in https://github.com/qutip/qutip-qip/discussions/101 a possible upgrade for the Hamiltonian coefficients.