I'm trying to build a new instruction for my circuit. This instruction needs both a controller qubit qctl and an arbitrary register qreg. When qctl is set then the Qiskit's initialize function is applied to qreg.

The original Initialize gate (version 10.5) can be found in the official documentation or locally at path: /anaconda3/envs/<environment name>/lib/python3.7/site-packages/qiskit/extensions/initializer.py. It follows a particular procedure which consists of applying a sequence of RY and RZ gates in order to match the desired state.

The idea is:

  • copy the Initialize instruction, naming it ControlledInitialize;
  • pass an additional single qubit register qctl to ControlledInitialize;
  • change all RZ, RY gates with CRZ, CRY gates (the first one is already available, the second one have to be made from scratch).

The problem

It seems that I have passed qctl register in the wrong way, in fact the below minimum example throws the error:

DAGCircuitError: '(qu)bit qctl[0] not found'

Minimum example

import numpy as np
from qiskit import QuantumCircuit, ClassicalRegister, QuantumRegister
from qiskit import BasicAer, execute

# copy here CRY and ControlledInitialize implementation

desired_vector = [ 1 / math.sqrt(2), 0, 0, 1 / math.sqrt(2) ]
qctl = QuantumRegister(1, "qctl")
qreg = QuantumRegister(2, "qreg")
creg = ClassicalRegister(2, "creg")
circuit = QuantumCircuit(qctl, qreg, creg)

circuit.controlled_initialize(qctl, desired_vector, qreg)
circuit.measure(qreg, creg)

job = execute(circuit, BasicAer.get_backend('qasm_simulator'), shots=10000)
print('Counts: ', job.result().get_counts(circuit))

The implementation

The whole code can be seen here as .py and here as Jupyter notebook.

CRZ, CRY gates

CRZ is a standard gate, you can use it with

from qiskit.extensions.standard.crz import CrzGate

CRY is present in Aqua module as a function but not as a subclass of Gate class. You can easily derive the gate implementation:

from qiskit.circuit import CompositeGate
from qiskit.circuit import Gate
from qiskit.circuit import QuantumCircuit
from qiskit.circuit import QuantumRegister
from qiskit.circuit.decorators import _op_expand, _to_bits
from qiskit.extensions.standard.u3 import U3Gate
from qiskit.extensions.standard.cx import CnotGate

class CryGate(Gate):
    """controlled-rz gate."""

    def __init__(self, theta):
        """Create new cry gate."""
        super().__init__("cry", 2, [theta]) # 2 = number of qubits

    def _define(self):
        self.u3(theta / 2, 0, 0, q_target)
        self.cx(q_control, q_target)
        self.u3(-theta / 2, 0, 0, q_target)
        self.cx(q_control, q_target)
        definition = []
        q = QuantumRegister(2, "q")
        rule = [
            (U3Gate(self.params[0] / 2, 0, 0), [q[1]], []),
            (CnotGate(), [q[0], q[1]], []),
            (U3Gate(-self.params[0] / 2, 0, 0), [q[1]], []),
            (CnotGate(), [q[0], q[1]], [])
        for inst in rule:
        self.definition = definition

    def inverse(self):
        """Invert this gate."""
        return CrzGate(-self.params[0])

def cry(self, theta, ctl, tgt):
    """Apply crz from ctl to tgt with angle theta."""
    return self.append(CryGate(theta), [ctl, tgt], [])

QuantumCircuit.cry = cry
CompositeGate.cry = cry

ControlledInitialize instruction

Any modification of original Initialize instruction is denoted with WATCH ME comment. Here an overview:

  • in __init__ I just save the single qubit control register;
  • in _define, gates_to_uncompute, multiplexer the temporary circuit built will have also qctl register;
  • in _define, gates_to_uncompute, multiplexer any append function call is enriched with qctl register in the list of qubits taken as second parameter;
  • in gates_to_uncompute just substitute RYGate/RZGate with CryGate/CrzGate.

    class ControlledInitialize(Instruction):
    """Complex amplitude initialization.
    Class that implements the (complex amplitude) initialization of some
    flexible collection of qubit registers (assuming the qubits are in the
    zero state).
    def __init__(self, controlled_qubit, params):
        """Create new initialize composite.
        params (list): vector of complex amplitudes to initialize to
        # WATCH ME: save controlled qubit register
        self.controlled_qubit = controlled_qubit
        num_qubits = math.log2(len(params))
        # Check if param is a power of 2
        if num_qubits == 0 or not num_qubits.is_integer():
            raise QiskitError("Desired statevector length not a positive power of 2.")
        # Check if probabilities (amplitudes squared) sum to 1
        if not math.isclose(sum(np.absolute(params) ** 2), 1.0,
            raise QiskitError("Sum of amplitudes-squared does not equal one.")
        num_qubits = int(num_qubits)
        super().__init__("controlledinitialize", num_qubits, 0, params) # +1 per il controllo
    def _define(self):
        """Calculate a subcircuit that implements this initialization
        Implements a recursive initialization algorithm, including optimizations,
        from "Synthesis of Quantum Logic Circuits" Shende, Bullock, Markov
        Additionally implements some extra optimizations: remove zero rotations and
        double cnots.
        # call to generate the circuit that takes the desired vector to zero
        disentangling_circuit = self.gates_to_uncompute()
        # invert the circuit to create the desired vector from zero (assuming
        # the qubits are in the zero state)
        initialize_instr = disentangling_circuit.to_instruction().inverse()
        q = QuantumRegister(self.num_qubits, 'q')
        initialize_circuit = QuantumCircuit(self.controlled_qubit, q, name='init_def')
        for qubit in q:
            initialize_circuit.append(Reset(), [qubit])
        # WATCH ME: cambiati registri
        temp_qubitsreg = [ self.controlled_qubit[0] ] + q[:]
        # initialize_circuit.append(initialize_instr, q[:])
        initialize_circuit.append(initialize_instr, temp_qubitsreg)
        self.definition = initialize_circuit.data
    def gates_to_uncompute(self):
        Call to create a circuit with gates that take the
        desired vector to zero.
            QuantumCircuit: circuit to take self.params vector to |00..0>
        q = QuantumRegister(self.num_qubits)
        # WATCH ME: aggiunto registro controlled_qubit
        circuit = QuantumCircuit(self.controlled_qubit, q, name='disentangler')
        # kick start the peeling loop, and disentangle one-by-one from LSB to MSB
        remaining_param = self.params
        for i in range(self.num_qubits):
            # work out which rotations must be done to disentangle the LSB
            # qubit (we peel away one qubit at a time)
             phis) = ControlledInitialize._rotations_to_disentangle(remaining_param)
            # WATCH ME: Initialize._rotations_to_disentangle diventa ControlledInitialize._rotations_to_disentangle
            # perform the required rotations to decouple the LSB qubit (so that
            # it can be "factored" out, leaving a shorter amplitude vector to peel away)
            # WATCH ME: substitute RZ with CRZ
            # rz_mult = self._multiplex(RZGate, phis)
            rz_mult = self._multiplex(CrzGate, phis)
            # WATCH ME: substitute RY with CRY
            # ry_mult = self._multiplex(RYGate, thetas)
            ry_mult = self._multiplex(CryGate, thetas)
            # WATCH ME: cambiati registri
            temp_qubitsreg = [ self.controlled_qubit[0] ] + q[i:self.num_qubits]
            # circuit.append(rz_mult.to_instruction(), q[i:self.num_qubits])
            # circuit.append(ry_mult.to_instruction(), q[i:self.num_qubits])
            circuit.append(rz_mult.to_instruction(), temp_qubitsreg)
            circuit.append(ry_mult.to_instruction(), temp_qubitsreg)
            print("Z: ", phis, " | Y: ", thetas)
        return circuit
    def _rotations_to_disentangle(local_param):
        Static internal method to work out Ry and Rz rotation angles used
        to disentangle the LSB qubit.
        These rotations make up the block diagonal matrix U (i.e. multiplexor)
        that disentangles the LSB.
        [[Ry(theta_1).Rz(phi_1)  0   .   .   0],
         [0         Ry(theta_2).Rz(phi_2) .  0],
          0         0           Ry(theta_2^n).Rz(phi_2^n)]]
        remaining_vector = []
        thetas = []
        phis = []
        param_len = len(local_param)
        for i in range(param_len // 2):
            # Ry and Rz rotations to move bloch vector from 0 to "imaginary"
            # qubit
            # (imagine a qubit state signified by the amplitudes at index 2*i
            # and 2*(i+1), corresponding to the select qubits of the
            # multiplexor being in state |i>)
             add_phi) = ControlledInitialize._bloch_angles(local_param[2 * i: 2 * (i + 1)])
            # WATCH ME: Initialize._bloch_angles diventa ControlledInitialize._bloch_angles
            # rotations for all imaginary qubits of the full vector
            # to move from where it is to zero, hence the negative sign
        return remaining_vector, thetas, phis
    def _bloch_angles(pair_of_complex):
        Static internal method to work out rotation to create the passed in
        qubit from the zero vector.
        [a_complex, b_complex] = pair_of_complex
        # Force a and b to be complex, as otherwise numpy.angle might fail.
        a_complex = complex(a_complex)
        b_complex = complex(b_complex)
        mag_a = np.absolute(a_complex)
        final_r = float(np.sqrt(mag_a ** 2 + np.absolute(b_complex) ** 2))
        if final_r < _EPS:
            theta = 0
            phi = 0
            final_r = 0
            final_t = 0
            theta = float(2 * np.arccos(mag_a / final_r))
            a_arg = np.angle(a_complex)
            b_arg = np.angle(b_complex)
            final_t = a_arg + b_arg
            phi = b_arg - a_arg
        return final_r * np.exp(1.J * final_t / 2), theta, phi
    def _multiplex(self, target_gate, list_of_angles):
        Return a recursive implementation of a multiplexor circuit,
        where each instruction itself has a decomposition based on
        smaller multiplexors.
        The LSB is the multiplexor "data" and the other bits are multiplexor "select".
            target_gate (Gate): Ry or Rz gate to apply to target qubit, multiplexed
                over all other "select" qubits
            list_of_angles (list[float]): list of rotation angles to apply Ry and Rz
            DAGCircuit: the circuit implementing the multiplexor's action
        list_len = len(list_of_angles)
        local_num_qubits = int(math.log2(list_len)) + 1
        q = QuantumRegister(local_num_qubits)
        # WATCH ME: aggiunto registro controlled_qubit
        circuit = QuantumCircuit(self.controlled_qubit, q, name="multiplex" + local_num_qubits.__str__())
        lsb = q[0]
        msb = q[local_num_qubits - 1]
        # case of no multiplexing: base case for recursion
        if local_num_qubits == 1:
            temp_qubitsreg = [ self.controlled_qubit[0], q[0] ]
            circuit.append(target_gate(list_of_angles[0]), temp_qubitsreg)
            return circuit
        # calc angle weights, assuming recursion (that is the lower-level
        # requested angles have been correctly implemented by recursion
        angle_weight = scipy.kron([[0.5, 0.5], [0.5, -0.5]],
                                  np.identity(2 ** (local_num_qubits - 2)))
        # calc the combo angles
        list_of_angles = angle_weight.dot(np.array(list_of_angles)).tolist()
        # recursive step on half the angles fulfilling the above assumption
        multiplex_1 = self._multiplex(target_gate, list_of_angles[0:(list_len // 2)])
        temp_qubitsreg =  [ self.controlled_qubit[0] ] + q[0:-1]
        circuit.append(multiplex_1.to_instruction(), temp_qubitsreg)
        # attach CNOT as follows, thereby flipping the LSB qubit
        circuit.append(CnotGate(), [msb, lsb])
        # implement extra efficiency from the paper of cancelling adjacent
        # CNOTs (by leaving out last CNOT and reversing (NOT inverting) the
        # second lower-level multiplex)
        multiplex_2 = self._multiplex(target_gate, list_of_angles[(list_len // 2):])
        temp_qubitsreg = [ self.controlled_qubit[0] ] + q[0:-1]
        if list_len > 1:
            circuit.append(multiplex_2.to_instruction().mirror(), temp_qubitsreg)
            circuit.append(multiplex_2.to_instruction(), temp_qubitsreg)
        # attach a final CNOT
        circuit.append(CnotGate(), [msb, lsb])
        return circuit

Qiskit version

Latest version is used:

import qiskit

{'qiskit': '0.10.5',
 'qiskit-terra': '0.8.2',
 'qiskit-ignis': '0.1.1',
 'qiskit-aer': '0.2.1',
 'qiskit-ibmq-provider': '0.2.2',
 'qiskit-aqua': '0.5.2'}
  • $\begingroup$ Just a comment to congratulate you for this question! You included everything and the whole question is well structured! I hope you will have a good answer. $\endgroup$ Commented Jul 17, 2019 at 7:47
  • $\begingroup$ If anybody stumbles upon this: The CRYGate is a circuit method by now and you can do circuit.cry. $\endgroup$
    – Cryoris
    Commented May 20, 2020 at 12:29

1 Answer 1


This is occurring because you declare your definition rule on two registers, but the way nodes are added to the DAG, only one register will be added. It is defined over both the QuantumRegister "q" you define in the method, and also the register passed in to self.params.

To fix this therefore you need to update your definition to work on only one register. For example, the first qubit could be the control and the other qubits can be initialised.

  • 1
    $\begingroup$ Thank you for your comment. I need some time to test it. $\endgroup$
    – incud
    Commented Jul 17, 2019 at 14:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.