search

unitaryfund / mitiq

Mitiq is an open source toolkit for implementing error mitigation techniques on most current intermediate-scale quantum computers.
https://mitiq.readthedocs.io
GNU General Public License v3.0
358 stars 157 forks source link

Behavior of `zne.scaling.fold_all` for scale factors that are not odd integers #755

Closed andreamari closed 3 years ago

andreamari commented 3 years ago

Users may be tempted to use this function: https://github.com/unitaryfund/mitiq/blob/2cb68d64a0f2ebfc98f6c3fbb03cb9c67f516cdf/mitiq/zne/scaling/folding.py#L240

with arbitrary scale factors. However this function, by construction, makes sense only if the scale factor is an odd integer. We may sill keep the type of scale_factor as float for consistency with other methods, but we should check if scale_factor is approximately equal to an odd integer. If not, we can raise a ValueError.

asuzukosi commented 3 years ago

Hey, i'll like to work on this issue

asuzukosi commented 3 years ago

@andreamari I've made the suggested changes and created a pull request, you can review it now.

rmlarose commented 3 years ago

Should we round it to the nearest odd integer instead of raising an error @andreamari?

andreamari commented 3 years ago

Should we round it to the nearest odd integer instead of raising an error @andreamari?

Actually, the rounding that you suggest is what the function is already doing.

Maybe we can only raise a warning (instead of an error) if the scale_factor is not np.isclose to its odd integer approximation.

My only concern is that the user may not be aware of the such a crude rounding approximation. E.g. scale_factor = 2 is rounded to 1.

rmlarose commented 3 years ago

I'm ok with a warning but I don't think its necessary as long as the function is well-documented and clearly states it will round the scale factor.

I think raising an error is bad behavior though.