There might be some issues in the NoiseModelTQ class. And provided a potential fix to it.

[caitaozhan]

Hi Hanrui,

In this open issue: #104 (comment), I mentioned that the profiling of sampling Pauli error gates is not what I expected.

In the old code, I mostly see Pauli Z errors sampled for sx gate, and a few Pauli Z errors sampled for cx gate. What should be like is that 1) the number of errors for double qubit errors (cx) are orders of magnitude higher than the number of errors for single qubit gates (sx); and 2) the number of Pauli X, Y, Z errors are kind of even.

I figured out that there might be some issues in the NoiseModelTQ class, and fixed them (mostly, may still have issues). Now the profiling of Pauli error gates is close to what I expect. An example of the current Pauli errors sampled is shown below (factor=1, no need to increase factor to a larger number):

single qubit error: pauli x = 0, pauli y = 0, pauli z = 6
double qubit error: pauli x = 73, pauli y = 54, pauli z = 65

References:

1. 1 and 2 qubit gate errors consisting of a depolarizing_error() followed by a thermal_relaxation_error()
2. depolarization.

Thanks,
Caitao

[Hanrui-Wang]

Hi Caitao,

Thank you for the pull request! I have reviewed it and the quantumnat.py looks good to me. For the noise model, I have a question, when will the 'pauli' directly appear in the instructions?

[caitaozhan]

Hi Hanhui,

According to what I observed, instruction["name"] == "pauli" appears when the op_name is a 'cx' gate.

[Hanrui-Wang]

Hi Caitao,

yes, I see the 'pauli' in the 'cx' noise dict. for my understanding, the code now also record the prob for the two qubit pauli and applied them one by one if the error is sampled (the 'kraus' and 'id' are ignored for now). I will merge it.