beginning of biased pauli noise for PF sim

[amicciche]

This is the code that was mentioned in our Zulip messages. One note, it seemed to be that line 41 in the noise.jl, n = nqubits(s) wasn't doing anything, so I went ahead and deleted that in addition to the functions I added.

As mentioned over Zulip, because of the error, this code is untested, aside from being able to generate the object from the Julia REPL.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 2.77778% with 35 lines in your changes missing coverage. Please review.

Project coverage is 82.24%. Comparing base (1fb1783) to head (e6548f9).

Files	Patch %	Lines
src/noise.jl	0.00%	18 Missing :warning:
src/pauli_frames.jl	0.00%	17 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #295 +/- ## ========================================== - Coverage 83.01% 82.24% -0.77% ========================================== Files 61 61 Lines 4033 4067 +34 ========================================== - Hits 3348 3345 -3 - Misses 685 722 +37 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[amicciche]

I'm now having this error when I try to use QuantumClifford:

[Krastanov]

yeah, tests fail, so I have definitely done something silly. Should have waited for tests to pass before pinging you

[Krastanov]

it is fixed now

[amicciche]

The applynoise!() that takes a PauliFrame seems to work correctly, however with the way the constructors are written for PauliError, one can't do something like the following (which will probably a common enough case to warrant fixing): I tried to fix this by changing Integer to Real on line 37 of noise.jl, but this caused an error. Here:

[Krastanov]

What error did the change to Real cause?

[Krastanov]

marking this as a draft just to make my review queue a bit more manageable. Mark it as finished when ready for review.

[amicciche]

What error did the change to Real cause?

When I change it to Real:

I get this error:

[Krastanov]

Float is a Real so when you do PauliNoise(x::Real, y::Real, z::Real) = PauliNoise(float(x), float(y), float(z)) you end up with infinite recursion.

A cleaner solution than both what I have written and what you have written is to specify the type for the constructor:

julia> struct A{T} a::T end

julia> A(a::Real) = A{Float64}(a)
A

julia> A(1)
A{Float64}(1.0)

Or even better, something that does not hardcode Float64

julia> struct A{T} a::T end

julia> function A(a::Real) 
           a = float(a)
           return A{typeof(a)}(a)
       end
A

julia> A(1)
A{Float64}(1.0)

julia> A(BigInt(1))
A{BigFloat}(1.0)

So the cleanest solution should be something along the lines of

function PauliNoise(px::Real, py::Real, pz::Real)
    px, py, pz = float.((px, py, pz))
    px, py, pz = promote(px, py, pz)
    T = typeof(px)
    return PauliNoise{T}(px, py, pz)
end

[amicciche]

Ah, I see! I'll try to fix this later today, using what you suggested, thanks!

[amicciche]

Your code worked as provided! Thanks! With that, I think everything works. I checked for a few codes that doing PauliError(qubit, p) after decoding results in the same logical vs physical error rate plots as PauliError(qubit, p/3, p/3, p/3). For some CSS codes, I also checked that the X and Z logical error corresponding to those simulations also matches doing PauliError(qubit, p*2/3, 0, 0) and PauliError(qubit, 0, 0, p*2/3) correspondingly (and the other X/Z logical error is 0). Similarly, PauliError(qubit, 0, p*2/3, 0) gives the same performance as PauliError(qubit, p).