Kronos Quantum

Honest, runnable quantum computing for fusion. Nine real quantum programs that execute on a free local simulator today, and on real quantum hardware the moment you plug in an account — the same code, one environment variable. Built and benchmarked by Kronos Fusion Energy, a family-owned American fusion company.

The honest headline. A quantum advantage for fusion is not here yet. Fault-tolerant hardware at the scale a fusion kernel needs is a post-~2036 prospect — our own resource estimates put a useful electronic-structure crossover at tens of thousands of physical qubits, against the roughly 1e2–1e3 physical qubits (and zero error-corrected logical qubits) hardware has today. What is real and runnable now is the tooling and the testing. This repository is where we do that testing in the open: real circuits, real fusion data, real numbers — including the null results.

What's here

• kronos_quantum/ — nine runnable quantum codes (see CODES.md): VQE, a fault-tolerant resource estimator, quantum metrology, a tensor-network ROM, a quantum kernel classifier, QAOA, Trotterized dynamics, error mitigation, and a variational quantum linear solver.
• data/ — the runnable data subset (a completed CGYRO turbulence flux map and real MAST disruption features) plus pointers to the full published datasets.
• WHITEPAPER.md — the technical white paper: the honest quantum-for-fusion thesis, the method behind each code, the results, and where each ties to the machine.
• papers/PAPERS.md — the peer-reviewable Kronos quantum paper series (2.90–2.95) that these codes accompany.
• research/ — the deeper research tracks the polished codes were distilled from.
• benchmarks/, tests/ — reproducible, honest benchmarks and a green test suite.

Where the published work sits relative to our actual research

Everything shared here is a public snapshot. Our internal research runs roughly three to four years ahead of what we have published — not because the later work is secret, but because writing it all up rigorously takes time we have not yet spent. We intend to keep papering and releasing it as we can. Read this repository as a faithful, conservative floor on what Kronos can do with quantum computing for fusion, not the ceiling.

Quickstart

pip install -e .            # numpy, scipy, scikit-learn, pandas, pennylane (free local simulator)
PYTHONPATH=. python -c "
import kronos_quantum as Q
print(Q.list_codes())                       # the nine codes
p = Q.run('KQUBIT')                          # VQE on the local simulator
print(p.note)                                # recovers H2 ground state to ~0.0002 mHa
"

Regenerate the honest benchmark table and run the tests:

PYTHONPATH=. python benchmarks/run_all.py    # writes BENCHMARKS.md (9 codes)
PYTHONPATH=. python -m pytest tests/ -q       # 8 passed

Registry-by-design: you always go through Q.get(name) / Q.run(name, x); from kronos_quantum import KQUBIT fails on purpose. Every code returns a Prediction(y, uncertainty, in_domain) and exposes an honest card() and benchmark().

Run it on real quantum hardware

The default backend is a free, exact local simulator. The same circuits run on real hardware or a shot-based simulator by setting one environment variable — no code change:

export KODEX_QC_BACKEND=ibm            # or 'aer' for local shot noise
export KODEX_QC_TOKEN=<your IBM Quantum API token>
PYTHONPATH=. python -c "import kronos_quantum as Q; print(Q.run('KQOPT').note)"

See docs/RUN_ON_HARDWARE.md. Install the extra with pip install -e '.[hardware]'.

The nine codes at a glance

Full detail in CODES.md; numbers in BENCHMARKS.md.

Data

Two small real datasets run everything here (the circuit codes build their own problems):

• data/cgyro_a1e_flux_map.csv — the completed 16/16 CGYRO A1e nonlinear kinetic-electron turbulence flux map (representative fidelity, reduced electron mass μ=400). Full campaign: Zenodo DOI 10.5281/zenodo.22136279.
• data/mast_disruption_features.csv — physics features for 591 real MAST shots (FAIR-MAST), a fixed 50 ms pre-event window; labels derived from the Ip current-quench (heuristic).

See data/README.md.

Honesty discipline

• Real benchmarks on real held-out data; null and negative results are kept, not hidden (KQERN ties classical, KSENSE erases its own advantage, KTENSOR corrects the low-rank expectation).
• No quantum-advantage claim this decade, on every card — a meaningful fusion role is post-~2036.
• CPU-only, fixed seeds, no network in the benchmarks.

License & citation

Apache-2.0 (see LICENSE). Please cite via CITATION.cff.

Kronos Fusion Energy. Quantum computing will matter for fusion; we are honest that its hardware role is still roughly a decade out, and we are doing the real testing today so we are ready when it arrives.