Add a tool for measuring expectation values of Pauli strings with readout error mitigation #7067
Conversation
…of qubits as input
… readout error mitigation
…CalibrationResult. Besides, fixed some lints
|
Hey @eliottrosenberg and @NoureldinYosri if you could take a look at this. Thanks! |
There was a problem hiding this comment.
Very nice work Danni!! This looks really great!
A few initial comments:
Regarding line 55 in shuffle_circuits_with_readout_benchmarking.py, it should be possible to specify 0 for num_random_bitstrings in order to do the measurement without readout mitigation (or have a different way of running without readout mitigation).
I think you went with lists of tuples for circuits_to_pauli because you can't put a cirq.Circuit into a dictionary, but you can put a cirq.FrozenCircuit into a dictionary, so maybe that would be more convenient (not sure). Also, the object that measure_pauli_strings outputs is pretty complicated, with lots of nested tuples, so maybe we want to create some data class for the output?
…_random_bitstrings, and add a test to cover the situation. The design is the run_shuffled_with_readout_benchmarking method will return a empty SingleQubitReadoutCalibrationResult if no calibration is actually done. 2. Allow measure_pauli_strings to take num_random_bitstrings = 0. In this case, no mitigation is actually done, and the mitigated result and unmitigated result are the same. Also add a test to handle this situation. 3. Make the return type of measure_pauli_strings a data class.
Done. |
There was a problem hiding this comment.
Thanks, Danni!! I tried testing it in this colab and got the error shown there.
Can you try again? The problem should be fixed in the latest commit ;) |
There was a problem hiding this comment.
Thanks, Danni! I tested it again, and there seems to be a bug in the way it is computing the expectation value. I tried it on a simulator without noise in a case where the expectation value should be 1 and got something close to 0. https://colab.sandbox.google.com/drive/1_on4xIHQNMH_2km3RjlfoZxBcYim9A3y
Let me know if you have any trouble identifying the cause of this and I can step in and help.
|
@ddddddanni Here is a simpler example that illustrates the bug: https://colab.sandbox.google.com/drive/1DUOEbrJSHIvlIN1CI7T6lug4zGDB5nyG |
Thanks for catching this! This is due to the code doesn't cover the situation that some pauli strings could be pauli I. I will fix it. |
There was a problem hiding this comment.
@ddddddanni LGTM ... I will take another look after you fix the failing CIs
| from cirq.study import ResultDict | ||
|
|
||
|
|
||
| @dataclasses.dataclass |
There was a problem hiding this comment.
prefer to use attrs over dataclasses
| elif pauli_op == ops.Y: | ||
| # Rotate to Y basis: Rx(pi/2) | ||
| operations.append(ops.rx(np.pi / 2)(qid_list[qubit_index])) | ||
| # No operation needed for Pauli Z or I (identity) |
There was a problem hiding this comment.
there shouldn't be identity in this case
| # Extract unique qubit tuples from all circuits | ||
| unique_qubit_tuples = set() | ||
| for circuit in circuits_to_pauli.keys(): | ||
| unique_qubit_tuples.add(tuple(sorted(set(circuit.all_qubits())))) |
There was a problem hiding this comment.
circuit.all_qubits() returns a frozenset, so you can dop the set here and elsewhere
Done! Can you try test again? The issue was Pauli I was incorrectly treated as Z in expectation calculation, and now the issue is fixed, and I also modified tests to test this. |
There was a problem hiding this comment.
Thanks, @ddddddanni! This is really great! One more request: can we multiply the result of measuring the PauliString by its coefficient? For example, right now your code does not distinguish between the following PauliStrings:
cirq.PauliString(cirq.Z.on_each(cirq.LineQubit.range(2)))and
-1 * cirq.PauliString(cirq.Z.on_each(cirq.LineQubit.range(2)))
|
(And when you implement my previous suggestion, make sure to also scale the statistical uncertainty by |
This PR does the following two things:
Improved Readout Benchmarking: Modified run_shuffled_with_readout_benchmarking to accept a list of qubit tuples (List[Tuple[Qubit]]). This allows for calculating the readout error rate on specific qubit subsets, ensuring alignment with the qubit sets used in non-identity Pauli operators within Pauli strings.
Added a tool to compute expectation values for Pauli operators. This tool calculates both unmitigated and readout-error-mitigated expectation values for each circuit, and integrates the results from the readout benchmarking.