T Reduction by Optimization

This software implements a technique developed by Marc Davis to use numerical optimization to reduce the number of T gates in a quantum circuit. This is implemented as a BQSKit pass.

The technique used here is described in the paper "T Count as a Numerically Solvable Optimization Problem".

Installation

This is available for Python 3.10+ on Linux, macOS, and Windows.

git clone https://github.com/WolfLink/trbo
pip install ./trbo

Optionally, if you want to use trbo.gridsynth.GridsynthPass:

pip install gridsynth

Basic Usage

TRbO provides tools to be used in a BQSKit workflow to convert circuits to the Clifford+T+Rz gate set with a minimized Rz and T count.

from bqskit import Circuit
from bqskit.compiler import Compiler
import trbo

before_circuit = Circuit.from_file("input.qasm")
with Compiler() as compiler:
    after_circuit = compiler.compile(before_circuit, trbo.workflows.default())
print(after_circuit.gate_counts)
after_circuit.save("output.qasm")

If the resulting circuit after TRbO still has Rz gates, they can be converted using Gridsynth. You can also do this in one step just by adding it to the workflow:

from trbo.gridsynth import GridsynthPass
with Compiler() as compiler:
    after_circuit = compiler.compile(before_circuit, trbo.workflows.default() + [GridsynthPass()])

We also provide trbo.workflows.fast() which provides lower-quality results at a faster runtime, and trbo.workflows.slow() which will more often find optimal results but at a slower runtime. You can further tune TRbO by controlling the following parameters passed to trbo.workflows.default():

• mutistarts (Default: 64) The number of random starting points to for numerical optimization. Larger values may offer better results at the cost of increased runtime.
• success_threshold (Default: 1e-6) The maximum allowed synthesis error using BQSKit's matrix distance (circuit.get_unitary().get_distance_from(original_circuit.get_unitary()))
• strict_opt (Default: False) Small quality increase at a significant performance cost. When the minimum Rz count that TRbO can find is not zero, the number of T gates added by gridsynth is expected to be large compared to the number of remaining T gates that could be removed, so TRbO normally skips the second step of removing T gates. Setting this value to True will always perform T gate reduction even when the final Rz count is not zero.
• partition_size (Default: 4) The maximum number of qubits to include in a block when partitioning a large circuit. Larger values may offer better results at the cost of increased runtime.

If you want to try tweaking settings we recommend starting with changin multistarts. You can also use trbo.utils.MultiStartPass to run the entire trbo routine multiple times and get the best result, which is sometimes more effective than just increasing the multistarts value.

If you are experienced with BQSKit, the main Pass is trbo.trbo.TRbOPass, and we recommend looking at trbo/workflows.py for suggestions on how to use it.

For more details on input, output, and working with quantum circuits in BQSKit, see the BQSKit documentation.