Data preparation initializer.py is calling a method on StatePreparation that no longer exists

[woodsp-ibm]

Environment

• Qiskit version: 1.2
• Python version: [3.8, 3.12]
• Operating system: Linux, Windows, Mac

What is happening?

Given 1.2 just released I quickly looked around the application modules just as a quick check all was good (algorithms runs tests against qiskit main so as to get some heads up, but the apps just run/test off stable release) - Qiskit Finance nightly CI tests failed in several of them due to this:

      File "/home/runner/work/qiskit-finance/qiskit-finance/qiskit_finance/circuit/library/probability_distributions/normal.py", line 218, in __init__
    circuit = initialize.gates_to_uncompute().inverse()

      File "/opt/hostedtoolcache/Python/3.10.14/x64/lib/python3.10/site-packages/qiskit/circuit/library/data_preparation/initializer.py", line 91, in gates_to_uncompute
    return self._stateprep._gates_to_uncompute()

    AttributeError: 'StatePreparation' object has no attribute '_gates_to_uncompute'

qiskit/qiskit/circuit/library/data_preparation/initializer.py

Line 91 in 713ab38

return self._stateprep._gates_to_uncompute()

How can we reproduce the issue?

Create an initializer Initialize instance and call its method gates_to_uncompute

What should happen?

The gates_to_uncompute public method should work, as it used to, even if internal implementation is altered.

The change, where the method that initializer was calling got removed, seems to have come in #12178

Any suggestions?

No response

[ShellyGarion]

Thank you @woodsp-ibm for detecting this bug. Could you please check if #12976 fixes it?

[alexanderivrii]

Thanks @ShellyGarion. Looking at the code, I believe that the fix should work (though let's wait for the confirmation from @woodsp-ibm).

I do have one one possibly naive question: the qiskit-finance code calls

# use default the isometry (or initialize w/o resets) algorithm to construct the circuit
if upto_diag:
    inner.append(Isometry(np.sqrt(normalized_probabilities), 0, 0), inner.qubits)
    self.append(inner.to_instruction(), inner.qubits)  # Isometry is not a Gate
else:
    initialize = Initialize(np.sqrt(normalized_probabilities))
    circuit = initialize.gates_to_uncompute().inverse()
    inner.compose(circuit, inplace=True)
    self.append(inner.to_gate(), inner.qubits)

If I understand correctly, initialize.gates_to_uncompute().inverse() amounts to calling Isometry(...) and applying inverse() twice in a row. Would it make sense to just call Isometry(...) and then maybe we don't even need the if -- else? I actually also don't understand how exactly up_to_diag is used when calling the Isometry code.

Of course, this does not invalidate the PR, as we still need to ensure that the publicly documented API function Initialize.gates_to_uncompute() works correctly.

[ShellyGarion]

The original synthesis code (before it was replaced by Isometry) was:

       # call to generate the circuit that takes the desired vector to zero
        disentangling_circuit = self._gates_to_uncompute()
              if self._inverse is False:
            initialize_instr = disentangling_circuit.to_instruction().inverse()
        else:
            initialize_instr = disentangling_circuit.to_instruction()

        q = QuantumRegister(self.num_qubits, "q")
        initialize_circuit = QuantumCircuit(q, name="init_def")
        initialize_circuit.append(initialize_instr, q[:])
        return initialize_circuit

So, it basically only does _gates_to_uncompute()

[Cryoris]

If I understand correctly, initialize.gates_to_uncompute().inverse() amounts to calling Isometry(...) and applying inverse() twice in a row. Would it make sense to just call Isometry(...) and then maybe we don't even need the if -- else

The problem was that Isometry is a Instruction but we needed a Gate, since we later on used the circuit as building block and required e.g. controlling or computing inverses. But at that point StatePreparation did not exist if I recall correctly, and we should probably change to using that class in Qiskit Finance.

[woodsp-ibm]

Thank you @woodsp-ibm for detecting this bug. Could you please check if #12976 fixes it?

@ShellyGarion I copied the initializer.py from your branch into my local install off Qiskit main and with that tests that failed before now pass. So yes it fixes things.