Transpile and optimize your PennyLane circuits into IonQ's native trapped-ion gate set (GPI, GPI2, MS) with just a single extra line of code!
from ionizer.transforms import ionize
@qml.qnode(dev)
@ionize
def circuit(x):
qml.Hadamard(wires=0)
qml.CNOT(wires=[0, 1])
qml.RX(x, wires=1)
return qml.expval(qml.PauliZ(0))>>> qml.draw(circuit)(0.3)
0: ──GPI2(0.00)─╭MS──GPI2(-1.57)─────────────────────────┤ <Z>
1: ──GPI2(3.14)─╰MS──GPI2(1.57)───GPI(-1.42)──GPI2(1.57)─┤The Ionizer is available via PyPI:
pip install ionizer
The core requirement is PennyLane 0.37.
Python versions 3.10-3.12 are supported and tested against.
To install from source, clone this repository and use
Poetry to install the dependencies listed in the
pyproject.toml file.
For more detailed explanations and usage examples, please check the full online documentation.
The Ionizer is implemented using quantum function
transforms, similar to PennyLane's existing
compilation
tools. To
compile and execute the circuit using trapped ion gates, the
@ionize transform will
- Decompose all operations into Paulis/Pauli rotations, Hadamard, and CNOT
- Cancel inverses and merge single-qubit rotations
- Convert everything except RZ to GPI, GPI2, and MS gates
- Virtually apply RZ gates
- Repeatedly apply single-qubit gate fusion and commutation through MS gates, and perform simplification based on a database of circuit identities.
from ionizer.transforms import ionize
@qml.qnode(dev)
@ionize
def circuit_ionized(params):
for idx in range(5):
qml.Hadamard(wires=idx)
for idx in range(4):
qml.RY(params[idx], wires=idx)
qml.CNOT(wires=[idx + 1, idx])
for wire in dev.wires:
qml.PauliX(wires=wire)
return qml.expval(qml.PauliX(0))>>> circuit_ionized(params)
tensor(0.99500417, requires_grad=True)
>>> qml.draw(circuit_ionized)(params)
0: ──GPI2(-1.57)──GPI(-1.52)──GPI2(-3.04)─╭MS───────────────────────────────────────────────────
1: ──GPI2(-1.92)──GPI(3.14)───GPI2(-1.22)─╰MS──GPI2(2.36)──GPI(1.67)──GPI2(0.99)─╭MS────────────
2: ──GPI2(-1.92)──GPI(3.14)───GPI2(-1.22)────────────────────────────────────────╰MS──GPI2(2.36)
3: ──GPI2(-1.92)──GPI(3.14)───GPI2(-1.22)───────────────────────────────────────────────────────
4: ──GPI2(-1.92)──GPI(3.14)───GPI2(-1.22)───────────────────────────────────────────────────────
────────────────────────────────────────────────────────────────────────────────────────────┤ <X>
────────────────────────────────────────────────────────────────────────────────────────────┤
───GPI(1.72)──GPI2(1.09)─╭MS────────────────────────────────────────────────────────────────┤
─────────────────────────╰MS──GPI2(2.36)──GPI(1.77)──GPI2(1.19)─╭MS─────────────────────────┤
────────────────────────────────────────────────────────────────╰MS──GPI2(0.00)──GPI2(1.57)─┤
The consistuent transforms can also be accessed and used independently.
There is currently not direct support for other frameworks. However, if you
would like to apply the transforms to Qiskit circuits, this can be accomplished
using the
pennylane-qiskit package as
shown belown.
qiskit_circuit = QuantumCircuit(...)
# Turns a Qiskit circuit into a PennyLane quantum function
qfunc = qml.from_qiskit(qiskit_circuit)
@qml.qnode(dev)
@ionize
def pennylane_circuit():
qfunc()
return qml.expval(qml.PauliX(0))This package is a work in progress. While it has been verified to work on some fairly large circuits, we still need to work on:
- finding circuit identities involving the 2-qubit gate
- ensuring differentiability of variational parameters
- writing more tests (compile at your own risk!)
The Ionizer is available open source under the MIT License. Contributions are welcome. Please open an issue if you are interested in contributing, or if you encounter a bug.
If you use the Ionizer as part of your workflow, we would appreciate if you cite it using the BibTeX below.
@software{di_matteo_2024_13370396,
author = {Di Matteo, Olivia},
title = {The Ionizer},
month = aug,
year = 2024,
publisher = {Zenodo},
version = {0.3},
doi = {10.5281/zenodo.13370396},
url = {https://doi.org/10.5281/zenodo.13370396}
}