Fixes for dependency issues caused by 0.14 release

[doichanj]

Summary

This PR fixed dependency issues caused by release 0.14.
Also this fix includes fix for #2084

Details and comments

• fixed sampling measure > 63 qubits
• ccz was accepted when input backend does not have ccz
• fixed smaller coupling map was generated in Primitives V1
• Remove skipping Python 3.8
• Hide primitives V2 for Qiskit < 1.0 setup.py requirements for qiskit are outdated #2096

[chriseclectic]

Are there any python unit tests for large Pauli's on the stabilizer simulator? If not you should add some (or if so add some more).

A basic one which is similar to the test I saw fail on 0.14 for my repo is something like preparing +1 Pauli eigenstates, measuring them, and checking the expectation value computed from the final counts is close to 1

Eg

    @ddt.data(65, 127, 433)
    def test_expectation_value_large_paulis(self, num_qubits):
        """Test expectation value circuit with large Pauli objects"""
        paulis = qi.PauliList([num_qubits * i for i in ["X", "Y", "Z"]])
        circuits = []
        for pauli in paulis:
            qc = QuantumCircuit(num_qubits, num_qubits)
            for i, p in enumerate(pauli):
                if p == Pauli("Y"):
                    qc.sdg(i)
                    qc.h(i)
                elif p == Pauli("X"):
                     qc.h(i)
                if p != Pauli("I"):
                    qc.measure(i, i)
        
        # Run circuits
        ...
        
        # Check all expvals are close to 1 where each expval is computed
        #  expval = sum_i freq_i * (1 - 2 * parity(bitsting_i)) / shots

[doichanj]

@chriseclectic I tested with the circuit below but both non-fixed version and this PR returns 1.0

        paulis = qi.PauliList([num_qubits * i for i in ["X", "Y", "Z"]])
        qc = QuantumCircuit(num_qubits, num_qubits)
        for pauli in paulis:
            for i, p in enumerate(pauli):
                if p == qi.Pauli("Y"):
                    qc.sdg(i)
                    qc.h(i)
                elif p == qi.Pauli("X"):
                     qc.h(i)
                if p != qi.Pauli("I"):
                    qc.measure(i, i)
        qubits = list(range(num_qubits))
        ostr = "I"*num_qubits
        oper = qi.Pauli(ostr)
        backend = self.backend(method=method)
        label = "expval"
        qc.save_expectation_value(oper, qubits, label=label)
        result = backend.run(transpile(qc, backend, optimization_level=0), shots=1).result()
        self.assertTrue(result.success)
        simdata = result.data(0)
        self.assertIn(label, simdata)
        value = simdata[label]
        self.assertAlmostEqual(value, 1.0)

something wrong with my code?
And I noticed that only randomize_measurement=True cases failed in pec_runtime test case

[chriseclectic]

For this test you need to specifically test the measure instruction, not the save_expectation_value.

I'm not sure why the failure only occurred when randomizing the output and correcting in post-processing. I did double check the with result metadata used in post-processing to undo the randomization with a fixed generation seed and it was the same both in 0.13 and 0.14, so that leads me to think the issue is the count data itself when randomizing.

One thing when randomizing is the stabilizer state being sampled from (which is a simple tensor product state) won't be symmetric over different qubits anymore, when without randomization for this test it was. You should be able to capture this without randomization by Paulis that mix X, Y, Z.

[hhorii]

Aer C++ library supports only the following gates natively.

"ccx", "ccz", "cp", "cswap", "csx", "cx", "cy", "cz", "delay", "ecr", "h",
"id", "mcp", "mcphase", "mcr", "mcrx", "mcry", "mcrz", "mcswap", "mcsx",
"mcu", "mcu1", "mcu2", "mcu3", "mcx", "mcx_gray", "mcy", "mcz", "p", "r"

If basis_gates is not subset of the above set, simulation will fail. I think it is better to check name is in the above set.

Also, I guess backend.configuration().basis_gate is list and not set. basis_gates should be set because this is one of hot loops.

I think that at the beginning of assemble_circuit(), basis_gates should be re-created as a set (while reducing measure, reset and etc). If None is passed, we can use a default set.

[doichanj]

When AerSimulator is made by from_backend() given basis_gates does not contain all the gates Aer supports, Aer should return error

[hhorii]

As I mentioned the above, list should be converted to set here.

[hhorii]

Just to test whether stabilizer can work for large-qubit circuits, we can use GHZ and then validate result with more simpler counts.