Obtaining expectation values

Given a quasi- or standard probability distribution, it is possible to compute the expectation value of diagonal operators directly from the distributions (or collections) of distributions. This can be done using string representation for standard diagonal operators such as I, Z, 0 or 1, or via dictionaries for custom operators.

Let us first generate some quasi-distributions by mitigating 2- and 3-qubit GHZ circuits on a noisy-simulator.

import numpy as np
from qiskit import *
from qiskit_ibm_runtime.fake_provider import FakeAthensV2
import mthree

backend = FakeAthensV2()

ghz2 = QuantumCircuit(2)
ghz2.h(0)
ghz2.cx(0,1)
ghz2.measure_all()

trans_ghz2 = transpile(ghz2, backend)

ghz3 = QuantumCircuit(3)
ghz3.h(0)
ghz3.cx(0,1)
ghz3.cx(1,2)
ghz3.measure_all()

trans_ghz3 = transpile(ghz3, backend)

raw2 = backend.run(trans_ghz2, shots=4000).result().get_counts()
raw3 = backend.run(trans_ghz3, shots=4000).result().get_counts()

mit = mthree.M3Mitigation(backend)
mit.cals_from_system()

quasi2 = mit.apply_correction(raw2, [0,1], return_mitigation_overhead=True)
quasi3 = mit.apply_correction(raw3, [0,1,2], return_mitigation_overhead=True)

Now let us compute the expectaion values of these distributions for the default case of Z operators on each qubit:

print('GHZ2:', quasi2.expval())
print('GHZ3:', quasi3.expval())

GHZ2: 0.9812714457511902
GHZ3: 0.020822674036026

The values are close to one and zero, respectively. We can use strings to repeat the above via:

print('GHZ2:', quasi2.expval('ZZ'))
print('GHZ3:', quasi3.expval('ZZZ'))

GHZ2: 0.9812714457511902
GHZ3: 0.020822674036026

Replacing a Z measurement with an I on one of the qubits has the affect of changing the sign for the \(|1>^{\otimes N}\) component:

print('GHZ2:', quasi2.expval('IZ'))
print('GHZ3:', quasi3.expval('ZIZ'))

GHZ2: -0.005565941333770752
GHZ3: 0.969853401184082

We can also pass lists of strings:

quasi3.expval_and_stddev(['ZZZ','ZIZ'])

(array([0.02082267, 0.9698534 ], dtype=float32), 0.018225846468499913)

Alternatively, users can specify their own custom diagonal operators using dictionaries. Here we form the projectors on the all ones and zeros states:

all_zeros_proj = {'000': 1}
all_ones_proj = {'111': 1}
quasi3.expval(all_zeros_proj)

0.49364039301872253

Like strings, one can pass an array of dicts:

quasi3.expval([all_zeros_proj, all_ones_proj])

array([0.4936404, 0.4861155], dtype=float32)

We can verify that the projectors return the correct values:

p0s, p1s = quasi3.expval([all_zeros_proj, all_ones_proj])
np.allclose([p0s, p1s], [quasi3['000'], quasi3['111']])

True