MultiLevel SABRE

MultiLevel SABRE is a Qiskit transpiler pass that applies a multi-level approach to the SABRE qubit routing algorithm. The algorithm coarsens the circuit to a simplified representation, solves the routing problem at the coarsest level and then refines the solution back to the original circuit. This often yields fewer SWAP gates compared to running SABRE directly on the full circuit.

Installation

The package is not published on PyPI yet, so install it from the repository root using pip:

pip install -e .

Usage

Example notebooks and scripts are available in the examples directory. They show how to optimise circuits and compare the results with the original SABRE pass.

Simple Example

from qiskit import QuantumCircuit
from qiskit.transpiler import CouplingMap, PassManager
from multilevel_sabre import MultiLevelSabre

# Create a simple circuit
circuit = QuantumCircuit(5)
circuit.h(0)
circuit.cx(0, 1)
circuit.cx(1, 2)
circuit.cx(2, 3)
circuit.cx(3, 4)

# Linear chain hardware topology
coupling_map = CouplingMap.from_line(5)

# Run the MultiLevel SABRE optimisation
pass_manager = PassManager([
    MultiLevelSabre(
        coupling_graph=coupling_map,
        cycles=10,
        random_seed=1,
        coarsest_solving_trials=50,
        num_interpolation=10,
        use_initial_embedding=True,
        verbose=0,
    )
])
optimised = pass_manager.run(circuit)

Comparison with SABRE

python examples/comparison_example.py

The script loads a benchmark circuit, runs both SABRE and MultiLevel SABRE and prints the number of SWAP gates and compilation time for each.

Parameters

The MultiLevelSabre class exposes the following arguments:

• coupling_graph (CouplingMap): hardware coupling map.
• cycles (int, default 10): number of optimisation cycles. The best result across cycles is returned.
• random_seed (int, default 1): seed for randomness.
• coarsest_solving_trials (int, default 50): number of attempts at the coarsest level.
• num_interpolation (int, default 10): number of interpolation steps during refinement.
• use_initial_embedding (bool, default True): whether to start from an initial heuristic mapping.
• verbose (int, default 0): verbosity level (0, 1, or 2).

Citation

If you use this code in your research, please cite the MultiLevel SABRE paper and reference this repository.

License

This project is licensed under the MIT License. See LICENSE for details.