lidmas 1.2.1

LiDMaS+ (Logical Injection & Decoding Modeling System) quantum error-correction simulator

LiDMaS+

Logical Injection & Decoding Modeling System

LiDMaS+ is a C++ research platform for quantum error-correction studies, covering surface-code threshold analysis, hybrid CV-DV (including digitized GKP) noise modeling, and hardware-to-decoder replay benchmarking under a common decoder IO contract. It also includes CSS and LDPC engine paths for comparative studies and supports cross-stack decoder comparability workflows.

Install from PyPI:

python -m pip install --upgrade lidmas

Then run:

lidmas --help

Documentation:

• Lidmas+ Documentation
• Read the Docs mirror is currently unavailable.

Statement of Need

Benchmarking decoder behavior and threshold trends requires reproducible, scriptable, and inspectable simulation pipelines. LiDMaS+ provides:

• deterministic Monte Carlo runs with explicit seed control,
• multiple decoders under a common interface,
• confidence-interval-aware threshold outputs,
• numerous reproducible workflows and scripts across examples/.

This makes it suitable for method development, reproducibility appendices, and comparative decoder studies.

Core Capabilities

• Surface-code simulation with configurable code distance and trial counts.
• CSS and LDPC demo/threshold workflows via engine switching.
• Decoder plugins: mwpm, uf, neural_mwpm.
• Noise modes:
  • pauli: sweep logical error rate versus physical Pauli error rate p.
  • hybrid: sweep logical error rate versus CV displacement scale sigma using GKP digitization.
• Optional threshold analysis tools (crossing estimates and scaling fits).
• Reproducible example suite under examples/.

Requirements

• C++20 compiler
• CMake >= 3.16
• Optional: OpenMP for parallel threshold runs
• Optional: CUDA toolkit (for GPU-accelerated Pauli surface_threshold sampling)
• Optional (for plots): Python 3 with matplotlib and pandas

Build

cmake -S . -B build
cmake --build build -j

Primary executable:

• build/lidmas

Packaging Notes

• Brand name remains LiDMaS+.
• PyPI package name is lidmas.
• CLI command is lidmas.
• Published wheels are CPU-oriented; CUDA builds are supported from source builds.

Optional CUDA build (Pauli surface_threshold sampling)

cmake -S . -B build -DLIDMAS_ENABLE_CUDA=ON
cmake --build build -j

At runtime, enable with:

./build/lidmas --engine=surface --surface_threshold --mode=pauli --gpu ...

Quick benchmark:

./build/lidmas --gpu_bench
./build/lidmas --gpu_bench_quick
./build/lidmas --gpu_bench_full

Quick Start

Show CLI help:

./build/lidmas --help

Run deterministic smoke test:

./build/lidmas --smoke

Run a Pauli threshold sweep (surface engine):

./build/lidmas --engine=surface --surface_threshold \
  --mode=pauli \
  --decoder=mwpm \
  --d=3,5,7 \
  --p_start=0.01 --p_end=0.15 --p_step=0.01 \
  --trials=2000 \
  --seed=1337 \
  --out=surface_threshold.csv

Run a hybrid CV sweep (surface engine):

./build/lidmas --engine=surface --surface_threshold \
  --mode=hybrid \
  --decoder=mwpm \
  --d=3,5,7 \
  --sigma_start=0.05 --sigma_end=0.60 --sigma_step=0.05 \
  --trials=2000 \
  --seed=1337 \
  --out=surface_threshold.csv

Run a native GKP sweep (surface engine):

./build/lidmas --engine=surface --surface_threshold \
  --mode=gkp \
  --decoder=mwpm \
  --d=3,5,7 \
  --sigma_start=0.05 --sigma_end=0.60 --sigma_step=0.05 \
  --gkp_gate=0.0005 --gkp_meas=0.0005 --gkp_idle=0.0002 \
  --gkp_loss=0.001 \
  --trials=2000 \
  --seed=1337 \
  --out=gkp_surface_threshold.csv

Neural decoder note:

• --decoder=neural_mwpm requires --neural_model=<path>.
• A trained reference model is provided at examples/decoder_comparison/trained_model.json.
• To retrain it, run python3 examples/decoder_comparison/train_neural_model.py.

CSS engine demo / threshold (experimental):

./build/lidmas --engine=css \
  --css_spec=examples/css_codes/steane/spec.yaml

./build/lidmas --engine=css \
  --css_repetition=7
./build/lidmas --engine=css \
  --css_shor

./build/lidmas --engine=css --css_threshold --mode=pauli --trials=2000 \
  --css_spec=examples/css_codes/steane/spec.yaml \
  --out=css_threshold.csv

CSS matrix files are dense 0/1 text (space or comma separated). Logical files can include multiple rows.

--css_repetition=<n> builds a bit-flip repetition code automatically (Hx empty, Hz chain). --css_shor builds the Shor [[9,1,3]] code automatically.

LDPC engine (default):

./build/lidmas --engine=ldpc

Reproducible Examples

The examples/ directory contains ready-to-run scripts for smoke checks, Pauli/hybrid/GKP thresholds, hardware-integration replay, stack-comparison benchmarking, scaling workflows, and plotting.

Setup once:

./examples/setup_env.sh

Run a minimal end-to-end check:

bash examples/quick_smoke/run.sh

Generated artifacts are written to:

• examples/results/<example_name>/

Output Schema

Threshold CSV output uses:

• mode,distance,sigma,pauli_p,trials,ler,ci_low,ci_high,defect_mean,weight_mean,decoder_fail_rate,mwpm_weight_scale,mwpm_graph,timestamp

Validation

For quick validation in local or CI environments:

./build/lidmas --smoke

Hardware Integration

See hardware-integration for the decoder IO schema, recommended data transport, and adapter API.

Xanadu-style job conversion example:

bash hardware_integration/xanadu/run.sh

Aurora/QCA/GKP fixture conversions:

bash hardware_integration/xanadu/run_public_datasets.sh

One-command real-data slice (download + convert + replay):

bash hardware_integration/xanadu/xandau_hardware_data.sh --install-deps

For large real datasets, use converter streaming/chunk flags: --stream --shot-start <N> --max-shots <K> [--append-out].

Replay converted NDJSON through the C++ adapter:

./build/lidmas --decoder_io_replay \
  --decoder_io_in=examples/results/hardware_integration/decoder_requests.ndjson \
  --decoder_io_out=examples/results/hardware_integration/decoder_responses.ndjson \
  --decoder_io_continue_on_error

Project Layout

include/   # public headers and interfaces
src/       # simulator and decoder implementations
examples/  # reproducible runs and plotting scripts
hardware_integration/ # provider-oriented hardware data ingestion

Release Notes

Detailed release notes and version-specific changes are tracked in Git tags and GitHub Releases.

Citation

If you use LiDMaS+ in academic work, cite the software release used for your experiments (tag + commit hash).

Paper reference (paper_03):

@misc{wayo2026unifiedhardwaretodecoderarchitecturehybrid,
  title={A Unified Hardware-to-Decoder Architecture for Hybrid Continuous-Variable and Discrete-Variable Quantum Error Correction in LiDMaS+},
  author={Dennis Delali Kwesi Wayo and Chinonso Onah and Leonardo Goliatt and Sven Groppe},
  year={2026},
  eprint={2604.15389},
  archivePrefix={arXiv},
  primaryClass={quant-ph},
  url={https://arxiv.org/abs/2604.15389}
}

License

This project is released under the MIT License (see LICENSE).

Contributing

Issues and pull requests are welcome. Please include:

• a clear problem statement,
• reproduction steps,
• expected versus observed behavior,
• and, where possible, a minimal test or script.