QEncode Benchmark

Documentation

Technical references for benchmark methodology, rules, and reproducibility. The full suite is open source — all scripts, specs, and data are in the GitHub repository.

Already have a VQE energy?

Score it against the exact active-space ground state without running the pipeline. No clone, and no chemistry stack — the reference energies ship inside the package.

pip install qencode-benchmark

python -c "
import qencode
s = qencode.score(-7.9835, molecule='LiH', active_space=(4, 4),
                  optimizer='COBYLA', ansatz='hea')
print(s.report())"

Reports the gap to exact diagonalisation, which of the two thresholds it clears, the certification margin, whether your optimiser and ansatz make that margin fragile across machines, and the rank among published entries. Full guide →

Generate a benchmark entry

The full pipeline: computes the CASCI reference, runs the VQE, and writes a hashed entry. Needs the chemistry stack, which the package pulls in.

pip install qencode-benchmark

qencode run --molecule H2 --mapping jordan_wigner \
  --ansatz-type uccsd --out-dir out

Output: a JSON entry with PySCF reference energies, the VQE result, circuit metrics and a SHA-256 provenance hash. About ten seconds for H₂. Clone the repository instead if you want the entry database and the producing commit recorded inside each entry. GPU backend available with --backend lightning.gpu.

On-site documentation

GitHub repository docs

Scoring notebook

Executable walkthrough of qencode.score: edit one cell with your energy and run it top to bottom. Outputs are committed, so it reads on GitHub without being run.

Quick Start Guide

Run your first entry in under 10 minutes. Covers environment setup, entry generation, and verification.

Trust Policy

The single definition of certified: gap below 10 mHa against the active-space CASCI reference. What certification attests, what it does not, and the markers that are reported but are not certification.

Leaderboard Rules

Eligibility, accuracy ranking, hardware cost ranking, balanced score formula, research tier policy, and the dated amendments on reproducibility and certification margin.

Verification sweep

The first end-to-end re-run of all 54 published entries, the three verifier bugs it found, and what reproducibility means across machines for gradient-free optimisers.

Noisy tier

What every certified circuit loses on a device: the measured hardware penalty under named gate-noise models, the gap under noise, zero-noise extrapolation, the corrected depolarizing convention, and two findings about the published entries the measurement exposed.

requirements-v4.txt

Pinned environment: PySCF 2.6.2, PennyLane 0.45.0, openfermion 1.6.1, NumPy 2.2.6, SciPy 1.13.1. Exact pins, not lower bounds — a VQE result is only reproducible if the stack is.

CITATION.cff

How to cite QEncode in papers and grant applications.

License

Open-source licensing terms for using, modifying, and distributing QEncode.

Need managed certification?

If you need signed artifacts for a paper, grant, or hardware evaluation, apply for managed certification. The self-run path is always free.