Gateset Tomography does not correctly estimate readout errors

[ChriPiv]

Informations

• Qiskit Ignis version: 0.4.0.dev0+e3b227e
• Python version: 3.7.3
• Operating system: Debian 10 (WSL)

What is the current behavior?

GST should give us an estimate of the readout error in the circuit, by giving us a single POVM measurement operator. In the current implementation in qiskit, this POVM operator seems to be a perfect projection.

Steps to reproduce the problem

Following code runs a GST on a simple basis set, using a noise_model that includes readout noise. GST should be able to give us an estimate of that readout error.

from qiskit import *
from qiskit.extensions import HGate
from qiskit.compiler import transpile
from qiskit.providers.aer.noise import *
from qiskit.ignis.verification.tomography.basis.gatesetbasis import default_gateset_basis
from qiskit.ignis.verification.tomography import *

# create some noise model
noise_model = NoiseModel()
depol_error = depolarizing_error(0.1, 1)
readout_error = ReadoutError([[0.9, 0.1], [0.15, 0.85]])
noise_model.add_all_qubit_quantum_error(depol_error, ['u2', 'u3'])
noise_model.add_all_qubit_readout_error(readout_error)

# create GST circuits
gate = HGate()
basis = default_gateset_basis()
basis.add_gate(gate)
circuits = gateset_tomography_circuits(gateset_basis=basis)
for i in range(len(circuits)):
    circuits[i] = transpile(circuits[i], basis_gates=['id', 'u2', 'u3'])

# Run GST circuits
job = qiskit.execute(circuits, Aer.get_backend('qasm_simulator'), noise_model=noise_model, shots=10000)
result = job.result()

# Run GST fitter
fitter = GatesetTomographyFitter(result, circuits, basis)
result_gates = fitter.fit()
print(result_gates.keys())
# 'E', 'rho', 'Id', 'X_Rot_90', 'Y_Rot_90', 'h'
print(result_gates['rho'])
# [[1., 0.], [0., 0.]]
print(result_gates['E'])
# [[1., 0.], [0., 0.]]

What is the expected behavior?

The above code should not return a perfect projection [[1., 0.], [0., 0.]] for result_gates['E'].

Suggested solutions

[gadial]

This was a whole missing piece (thanks for catching it!), fixed in #408 .