Noise model does not support custom op

[Squirtle007]

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

Currently, noisy simulations do not support user-defined custom operations. Specifically, adding noise channels (e.g., cudaq.BitFlipChannel, cudaq.PhaseFlipChannel, cudaq.KrausChannel) to cudaq.NoiseModel() with a custom op will result in an error:

RuntimeError: Invalid quantum op for noise_model::add_channel

Steps to reproduce the bug

import cudaq
import numpy as np

# Identity matrix as a custom unitary operator
matrix = np.array([[1, 0, 0, 1]])
cudaq.register_operation("id", matrix)


# Let's define a simple kernel that we will add noise to.
qubit_count = 1

@cudaq.kernel
def kernel(qubit_count: int):
    qvector = cudaq.qvector(qubit_count)
    id(qvector)

print(cudaq.draw(kernel, qubit_count))


# Define the Kraus Error Operator as a complex ndarray.
error_probability = 0.2
kraus_0 = np.array([[1.0, 0.0], [0.0, np.sqrt(1 - error_probability)]], dtype=np.complex128)
kraus_1 =  np.array([[0.0, np.sqrt(error_probability)], [0.0, 0.0]], dtype=np.complex128)

# Add the two channels to our Noise Model.
noise_model = cudaq.NoiseModel()

# Add the Kraus Operator to create a quantum noise channel.
damping_channel = cudaq.KrausChannel([kraus_0, kraus_1])

# Apply the damping channel to the custom op specified as "id" on the 1st qubit.
noise_model.add_channel("id", [0], damping_channel)

Expected behavior

cudaq.NoiseModel() should accept custom operator: [str], like "id" defined above by users, as a valid unitary quantum operator, and a subsequent cudaq.sample call can return SampleResult accordingly.

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA Quantum version: 0.8.0
• Python version: 3.10

Suggestions

No response

[1tnguyen]

Hi @Squirtle007

The noise for custom op feature will be available in the next release (0.9). It is available now in the nightly docker image.

An example to test:

import cudaq
import numpy as np

cudaq.set_target("density-matrix-cpu")

# Identity matrix as a custom unitary operator
matrix = np.array([[1, 0, 0, 1]])
cudaq.register_operation("id", matrix)


# Let's define a simple kernel that we will add noise to.
qubit_count = 1

@cudaq.kernel
def kernel(qubit_count: int):
    qvector = cudaq.qvector(qubit_count)
    x(qvector)
    id(qvector)


# Define the Kraus Error Operator as a complex ndarray.
error_probability = 0.2
kraus_0 = np.array([[1.0, 0.0], [0.0, np.sqrt(1 - error_probability)]], dtype=np.complex128)
kraus_1 =  np.array([[0.0, np.sqrt(error_probability)], [0.0, 0.0]], dtype=np.complex128)

# Add the two channels to our Noise Model.
noise_model = cudaq.NoiseModel()

# Add the Kraus Operator to create a quantum noise channel.
damping_channel = cudaq.KrausChannel([kraus_0, kraus_1])

# Apply the damping channel to the custom op specified as "id" on the 1st qubit.
noise_model.add_channel("id", [0], damping_channel)

sample = cudaq.sample(kernel, qubit_count, noise_model=noise_model)
print(sample)

This will print a noisy result like { 0:211 1:789 }.

Note: in the kernel, I added an X operation to set the qubit to 1 so that the noise effect can be observed as we are using an amplitude damping channel which only affects excited states.

[Squirtle007]

Thanks, @1tnguyen , this works on my end using the CUDA-Q nightly image. Thanks for your confirmation.