The local AHS device¶
The local analog Hamiltonian simulation (AHS) device of the PennyLane-Braket plugin runs simulation on the local Braket SDK. This could be either utilizing the processors of your own PC, or those of a Braket notebook instance hosted on AWS.
This device is useful for small-scale simulations in which the time of sending a job to a remote service would add an unnecessary overhead. It can also be used for rapid prototyping before running a computation on a paid-for remote service.
After the Braket SDK and the plugin are installed you immediately have access to the local Braket AHS simulator in PennyLane.
Note that pulse programming in PennyLane requires the module
jax, which can be installed
following the instructions [here](https://github.com/google/jax#installation).
To instantiate the local Braket simulator, simply use:
import pennylane as qml device_local = qml.device("braket.local.ahs", wires=2)
This device can be used with a QNode within PennyLane. It accepts circuits with a single ParametrizedEvolution operator based on a ParametrizedHamiltonian compatible with the simulated hardware. More information about creating PennyLane operators for AHS can be found in the PennyLane docs.
It is important to keep track of units when specifying electromagnetic pulses for hardware control. The frequency and amplitude provided in PennyLane for Rydberg atom systems are expected to be in units of MHz, time in microseconds, phase in radians, and distance in micrometers. All of these will be converted to SI units internally as needed for upload to the hardware, and frequency will be converted to angular frequency (multiplied by \(2 \pi\)).
When reading hardware specifications from the Braket backend, bear in mind that all units are SI and frequencies are in rad/s. This conversion is done when creating a pulse program for upload, and units in the PennyLane functions should follow the conventions specified in the PennyLane docs to ensure correct unit conversion. See rydberg_interaction and rydberg_drive in Pennylane for specification of expected input units, and examples for creating hardware-compatible ParametrizedEvolution operators in PennyLane.
Creating a register¶
The atom register defines where the atoms will be located, which determines the strength of the interaction between the atoms. Here we define coordinates for the atoms to be placed at (in micrometers), and create a constant interaction term for the Hamiltonian:
# number of coordinate pairs must match number of device wires coordinates = [[0, 0], [0, 5]] H_interaction = qml.pulse.rydberg_interaction(coordinates)
Creating a drive¶
We can create a drive with a global component and (positive) local detunings. If the local detunings are time-dependent, they must all have the same time-dependent envelope, but can have different, positive scaling factors.
from jax import numpy as jnp # gaussian amplitude function (qml.pulse.rect enforces 0 at start and end for hardware) def amp_fn(p, t): f = p * jnp.exp(-(t - p)**2 / (2 * p**2)) return qml.pulse.rect(f, windows=[0.1, 1.7])(p, t) # defining a linear detuning def det_fn_global(p, t): return p * t def det_fn_local(p, t): return p * t**2 # creating a global drive on all wires H_global = qml.pulse.rydberg_drive(amplitude=amp_fn, phase=0, detuning=det_fn_global, wires=[0, 1]) # creating local drives # note only local detuning is currently supported, so amplitude and phase are set to 0 H_local0 = qml.pulse.rydberg_drive(amplitude=0, phase=0, detuning = det_fn_local, wires=) H_local1 = qml.pulse.rydberg_drive(amplitude=0, phase=0, detuning = det_fn_local, wires=) # full hamiltonian H = H_interaction + H_global + H_local0 + H_local1
Executing an AHS program¶
@qml.qnode(device_local) def circuit(params): qml.evolve(H)(params, t=1.5) return qml.sample() # amp_fn expects p to contain 3 parameters amp_params = [2.5, 1, 0.3] # global_det_fn expects p to be a single parameter det_global_params = 0.2 # each of the local drives take a single parameter for p # the detunings have the same shape, but vary by scaling factor p local_params1 = 0.5 local_params2 = 1
When executed, the circuit will perform the computation on the local machine.
>>> circuit([amp_params, det_global_params, local_params1, local_params2]) array([[0, 0], [0, 0], [0, 0], ..., [1, 0], [1, 0], [1, 0]])