plaquette-unionfind 0.0.1a2

No project description provided

Table of Contents

• About

• Installation

• Usage

  • Python Frontend

  • C++ Backend

  • Interface to Plaquette

• Benchmarks

• Documentation

About

The Plaquette-UnionFind plugin extends the Plaquette error correction software, providing a fast unionfind decoder written in C++. See the Benchmarks section for a comparison to other known decoding packages.

Installation

The basic dependencies for installation are cmake and ninja.

The C++ tests/examples and python bindings can be built independently by

cmake -Bbuild -G Ninja -DPLAQUETTE_UNIONFIND_BUILD_TESTS=On -DPLAQUETTE_UNIONFIND_BUILD_BINDINGS=On
cmake --build ./build

You can run the C++ backend tests with

make test-cpp

You can install just the python interface with (this quietly builds the C++ backend):

pip install -r requirements.txt
pip install .

or get the latest stable release from PyPI

pip install plaquette_unionfind

You can run the python frontend tests with

make test-python

Usage

Python Frontend

import plaquette_unionfind as puf
import plaquette_graph as pg

vertex_boundary = [False] * 12 + [True] * 8
edges = [(0, 12), (0, 1), (1, 2), (2, 13), (0, 3), (1, 4), (2, 5), (3, 14), (3, 4),
         (4, 5), (5, 15), (3, 6), (4, 7), (5, 8), (6, 16), (6, 7), (7, 8), (8, 17),
         (6, 9), (7, 10), (8, 11), (9, 18), (9, 10), (10, 11), (11, 19)]
syndrome = [False, False, True, True, False, True, True, False, True, False, True,
            False, False, False, False, False, False, False, False, False]
dg = pg.DecodingGraph(num_vertices, edges, vertex_boundary)
uf = puf.UnionFindDecoder(dg)
correction = uf.decode(syndrome)

C++ Backend

#include "DecodingGraph.hpp"
#include "UnionFindDecoder.hpp"

int main(int argc, char *argv[]) {

    using namespace Plaquette;
    //a vector storing a flag that is 1 if the vertex is on the boundary
    std::vector<bool> vertex_boundary = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                         0, 0, 1, 1, 1, 1, 1, 1, 1, 1};
    std::vector<std::pair<size_t, size_t>> edges = {
        {0, 12}, {0, 1},  {1, 2},   {2, 13}, {0, 3}, {1, 4},  {2, 5},
        {3, 14}, {3, 4},  {4, 5},   {5, 15}, {3, 6}, {4, 7},  {5, 8},
        {6, 16}, {6, 7},  {7, 8},   {8, 17}, {6, 9}, {7, 10}, {8, 11},
        {9, 18}, {9, 10}, {10, 11}, {11, 19}};

    std::vector<bool> syndrome = {false, false, true,  true,  false, true,  true,
                                  false, true,  false, true,  false, false, false,
                                  false, false, false, false, false, false};

    auto decoding_graph = DecodingGraph(vertex_boundary.size(),
                                      edges, vertex_boundary);
    UnionFindDecoder decoder(decoding_graph);
    auto correction = decoder.Decode(syndrome);
}

Interface to Plaquette

Plaquette is undergoing heavy development, so this interface is likely to change. If you are benchmarking our decoder, please do not use the plaquette interface unless you know what you are doing. You will be timing other computations unrelated to the decoding.

from plaquette.codes import LatticeCode
import plaquette_unionfind

    code = LatticeCode.make_planar(n_rounds=1, size=size)
    qed = {
        "pauli": {q.equbit_idx: {"x": error_rate} for q in code.lattice.dataqubits}
    }
    decoder = plaquette_unionfind.UnionFindDecoderInterface.from_code(code, qed, weighted=False)

Benchmarks

We benchmark plaquette-unionfind along with two of the leading open source decorders PyMatching-v2 and FusionBlosson.

For our benchmarks, we have been careful to only time the intrinsic PyMatching and FusionBlossom decoding functions. Only the intrinsic speed of the decoder was timed, so e.g., we do not yet benchmark along with the decode_batch function for Pymatching-v2. We also set FusionBlossom to single-threaded mode for a fair comparison with PyMatching and plaquette-unionfind. All benchmarks are performed on a m6id.4xlarge AWS node. All benchmarks are reproducible (see below) using our scripts on a m6id.4xlarge.

For our first benchmark, we use the 2-D (perfect measurement) Planar Code with p = 0.05 depolarization error. We observe up to 8-10x speedup over PyMatching and FusionBlossom.

For our second benchmark we use the 3-D Rotated Planar Code to account for measurement errors, with p = 0.01 depolarization error and p = 0.01 measurement error. We observe that Pymatching is more sensitive to the lattice size than the other decoders.

Finally we benchmark a 30x30 Rotated Planar Code (29 rounds of measurement) with varying probability. We observe that FusionBlossom is sensitive to the error probability.

To run all three benchmarks, use the following bash commands:

source benchmarks/run_perfect_planar_benchmark.sh 0.05
source benchmarks/run_imperfect_rotated_planar_benchmark.sh 0.01
source benchmarks/run_imperfect_rotated_planar_fixed_size_benchmark.sh

python plot_benchmark_1.py perfect_planar_0.05.dat perfect_planar_0.05.png "Surface Code Benchmark #1" 5
python plot_benchmark_2.py imperfect_rotated_planar_0.01.dat imperfect_rotated_planar_0.01.png "Surface Code Benchmark #2" 5
python plot_benchmark_3.py imperfect_rotated_planar_fixed_size.dat imperfect_rotated_planar_fixed_size.png "Surface Code Benchmark #3" 1

Documentation

To generate the documentation you will need to install graphviz and doxygen. Then run

pip install -r doc/requirements.txt
make docs
firefox ./doc/_build/html/index.html

Here is a live link to the documentation: https://docs.plaquette.design/projects/unionfind