Skip to content

A transpiler pass for Qiskit which uses ZX-Calculus for circuit optimization, implemented using PyZX.

License

Notifications You must be signed in to change notification settings

dlyongemallo/qiskit-zx-transpiler

Repository files navigation

Qiskit ZX Transpiler

A transpiler pass for Qiskit which uses ZX-Calculus for circuit optimization, implemented using PyZX.

License PyPI - Python Version PyPI - Package Version CI Status

Example usage

from qiskit.circuit import QuantumCircuit
from qiskit.transpiler import PassManager
from zxpass import ZXPass

# Create a qiskit `QuantumCircuit` here...
qc = QuantumCircuit(...)

pass_manager = PassManager(ZXPass())
zx_qc = pass_manager.run(qc)

It is also possible to initialise ZXPass with a custom optimization function. (The default, if none is supplied, is to call pyzx.simplify.full_reduce on the graph of the circuit.)

import pyzx

def my_optimize(c: pyzx.Circuit) -> pyzx.Circuit:
    g = c.to_graph()
    # do stuff to simplify `g`...
    return pyzx.extract.extract_circuit(g)

pass_manager = PassManager(ZXPass(my_optimize))
my_qc = pass_manager.run(qc)

The transpiler is also exposed as a pass manager stage plugin at the optimization stage.

from qiskit import transpile

zx_qc = transpile(qc, optimization_method="zxpass")

Running benchmarks

To perform some benchmarks based on the QASMBench suite, run the following:

cd benchmarking
python run_benchmarks.py

This will output some statistics and produce 2 PNG files showing the depth compression ratio between both Qiskit- and ZX-optimized circuits and the original circuits, and ratio of non-local gates beween the Qiskit- and ZX-optimized circuits.

Previous work

There have been two previous attempts to create a transpiler pass for Qiskit using PyZX which I'm aware of.

The first attempt was made in 2019 by @lia-approves, @edasgupta, and @ewinston when they were interns at IBM Quantum, as documented in this Qiskit issue. That code used qasm as an intermediate format when converting between a Qiskit DAGCircuit and a PyZX Circuit, which is undesirable for reasons noted in that issue. Furthermore, the code is out of date with subsequent changes made to both Qiskit and PyZX.

The second attempt was made by @gprs1809 et al. as a part of the Qiskit Advocate Mentorship Program (QAMP) in the fall of 2022. The code is found in this repository. This implementation converts a PyZX Circuit directly to a Qiskit QuantumCircuit, without going through qasm. However, the code is incomplete and produces the wrong output for some circuits (which may have been due to a difference between how PyZX and Qiskit implements certain gates, fixed in this PR), and (as far as I can tell) the code to convert in the other direction is not available.