Slow behaviour of functools reduce() in VibrationalOp

[paulineollitrault]

What is the expected enhancement?

This line: https://github.com/Qiskit/qiskit-nature/blob/76c9f976c08ba49c6397a40cc138e62a9f827b7a/qiskit_nature/operators/second_quantization/vibrational_op.py#L162 is too slow to treat large system sizes.

[mrossinek]

This is literally just summing the list of operators, ops. Since that basically amounts to simply concatenating the underlying list of labels and coefficients, this can definitely improved significantly by simply performing said list concatenation in place.

@paulineollitrault Would you be interested in working on a fix for this yourself?

[paulineollitrault]

Can be fixed by modifying these lines https://github.com/Qiskit/qiskit-nature/blob/f0a4be91e82133b35560dcb34149d1956ad1b431/qiskit_nature/operators/second_quantization/vibrational_op.py#L149-L164 as follows:

        else:
            # Sparse label
            dense_labels = self._convert_to_dense_labels(data, num_modals)

            new_dense_labels = []
            new_dense_coeffs = []

            labels = []
            for dense_label, coeff in dense_labels:
                new_op = reduce(
                    lambda a, b: a @ b,
                    (VibrationalOp((label, 1), num_modes, num_modals) for label in dense_label),
                )
                # We ignore the type here because mypy only sees the complex coefficient
                labels.extend(new_op._labels.copy())
            self._labels = labels.copy()