Inconsistent semantics of `leakage_scale` between `general.rs` and `error_h1.py`

[PabloAndresCQ]

In Rust PECOS' general error model, leakage_scale is not a scale parameter in the same sense as the rest, but actually a probability that when the noise model requests a leak event, the leak is modelled as escaping to a leaked state |2> or a depolarising channel is applied to it (see here). In Python PECOS for H1/H2, there was only a binary choice (provided by leak2depolar).

There is a leakage_scale in Python PECOS H1/H2, but it does not have the same semantics as Rust PECOS. It is actually a standard scaling parameter there (see here) that does not have a corresponding parameter in Rust PECOS.

IMO, the current Rust leakage_scale should be renamed to something else (maybe p_leak2depolar) and the H1/H2 leakage_scale be added.

[ciaranra]

p_leakage2depolar or leakage2depolar_ratio (since maybe it isn't really a probability either) sounds good to me. Eventually I would like to switch H1/H2 over to the same thing. I felt that leakage_scale was a bit confusing and being able to tune between full leakage to fill depolarizing was more useful.

[ciaranra]

Maybe... Get rid of leakage_scale, keep the name leakage2depolar have True map to 1.0 and False map to 0.0. And allow users to also use floats. Maybe this is what you are already doing for Helios.

[PabloAndresCQ]

For the Selene noise model, atm we only have the leak2depolar boolean and do not support leakage_scale. What's the use case of being able to tune a leakage2depolar_ratio? I had expected that a binary option was all that was needed.

I felt that leakage_scale was a bit confusing

Ah, I just noticed that in H1/H2 it was not just a scaling parameter as I claimed in the description of the issue. It seems to have two purposes:

1. It is used as a scaling parameter (see here).
2. But it is also passed as the p_leak argument to leak, which is then its own thing.

To me, it looks like these two purposes of leakage_scale are disjoint and there should be one parameter for each purpose. What's the reason it's a single parameter? Should we add a parameter for purpose 1 in the Rust PECOS implementation?

[ciaranra]

leakage_scale is just an awkward thing that affects many different noise channels and can reduce the error rate in somewhat unpredictable ways. It is better to scale noise channel error rates. People also want to understand how leakage affects their experiments, so converting it to completely depolarizing noise is useful. Therefore, I think it is better to just get rid of the confusing leakage_scale and have one knob for making a circuit more or less leaked (unless you go in a modify the individual noise channels yourself).

[ciaranra]

Here I mean the original leakage_scale in Python is awkward. In Rust I got rid of leakage2depolar and moved that functionality to leakage_scale. But I think that was the wrong call. Rust's leakage_scale should be called leakage2depolar, I think. And the Python H1/H2 should be adjusted to match

[PabloAndresCQ]

Those changes sound reasonable to me, and I am happy for leakage2depolar in Selene to become a float point rather than a boolean.

[ciaranra]

Alright, sounds good.