Add a default max_step to the processors

[BoxiLi]

Similar to qutip/qutip#2040, with a separate handling of options for qutip-v5.

[BoxiLi]

Although this does solve the skipping pulse issue. On my computer, it increases the total time for the tests from 40 seconds to 3 min...

[BoxiLi]

The main slow down is actually at the computation of the full tlist, not the time evolution.

[hodgestar]

That is quite a big slowdown. Can we make get_full_tlist more efficient? Or avoid calculating the full tlist and just get the smallest difference?

[BoxiLi]

Having a closer look again, the main problem is actually with the CircuitQED, where a much smaller sampling step is needed for coeff to represent a continuous pulse shape. What we get from get_full_tlist is only the smallest sampling step size in tlist, rather than the gate time. The gate time information is not kept, unfortunately. What you suggest in the other repo in qutip, saving the max_step in QobjEvo would be helpful.

This problem does not exist in qutip.qip inside qutip--v4, because there is no CircuitQED model.

[BoxiLi]

An alternative is to use something like one-tenth of the total circuit pulse time. This does not solve the problem completely. But it will avoid those mistakes that keep popping up when people try and compile one gate and send it to the solver. When there are more gates and a non-trivial circuit, the probability of this happening will be very low.

[hodgestar]

I'm okay with a hackier solution like 0.1 * total_time for now, and a comment explaining the problem and perhaps referencing a GitHub issue that explains the difficulties.