IX-Fusion

Evaluation-only computational research framework for falsifiable stellarator hypothesis testing.

IX-Fusion asks whether a C6-seeded quasi-isodynamic / omnigenous stellarator starting condition is worth pursuing once magnetic, particle-confinement, active-control, structural, and reactor-facing engineering penalties are evaluated against matched controls.

The project is not built around preserving a symbol. The C6 / inverted-Flower geometry is only the historical source of the initial six-field-period hypothesis. Physics is allowed to modify or reject it.

Current release verdict: FAIL_OR_INCONCLUSIVE at the reduced-model gate.
Scientific stage: geometry_hypothesis.
IX-Fusion does not claim demonstrated plasma confinement, ignition, net energy, net electricity, or reactor feasibility.

Current proof-of-concept result

Release 0.1 compares the C6 seed with a matched five-field-period helical control using the same deterministic optimization budget.

Reduced-order metric	C6 candidate	Matched control	C6 outcome
Composite screening objective — lower is better	0.0318820	0.0304006	4.87% worse
Bounce-action variation proxy — lower is better	0.000258110	0.000257673	0.17% worse
Mean radial-excursion proxy — lower is better	0.00930135	0.0130368	28.65% better
Field-line escape fraction in tested reduced set	0	0	tie
Engineering burden proxy — lower is better	0.870566	0.788074	10.47% worse

The C6 candidate therefore does not clear the predeclared overall advantage gate, even though one field-line metric is favorable. That distinction is intentional: IX-Fusion is designed so one attractive number cannot override contrary evidence.

See PROOF_OF_CONCEPT.md and results/poc/POC_RESULT.md.

Research question

Does a C6-seeded quasi-isodynamic/omnigenous stellarator, co-optimized with distributed phase-coherent RF actuation and realistic engineering constraints, provide measurable improvements in confinement or controllability relative to matched conventional stellarator baselines without worsening islands, transport, stability, structural burden, actuator complexity, or whole-system penalties?

Release 0.1 does not have enough model authority to answer that full question. It provides a reproducible lower-fidelity screen and the architecture required to escalate the question to specialist solvers without changing the claim rules after seeing results.

What is implemented now

C6 geometry and magnetic screening

• versioned six-field-period C6 seed;
• matched five-field-period helical control;
• axisymmetric negative control;
• dimensionless 3-D toroidal boundary/axis representation;
• reduced field-strength spectrum;
• deterministic field-line integration and Poincaré output;
• rotational-transform screening output;
• resonant-overlap proxy.

Omnigenity-oriented screening

• bounce-action variation proxy inspired by trapped-particle orbit physics;
• explicit distinction between the proxy and the second adiabatic invariant of a solved equilibrium;
• equal-budget deterministic seed optimization;
• multi-gate verdict logic rather than a single-objective success claim.

Active phase-control research layer

• six-source phase-coherent actuator model;
• spatial-mode spectrum and target-mode purity;
• amplitude, phase, and geometry error injection;
• deterministic Monte Carlo;
• abstract feedback correction.

Under the committed error assumptions, median target spatial-mode purity increases from approximately 99.21% open loop to 99.93% with abstract feedback. This is a signal-processing result only; it is not evidence of RF/plasma coupling, heating, current drive, or instability suppression.

Magneto-structural and engineering screens

• magnetic-axis curvature and torsion;
• normalized support-burden proxy;
• blanket-space proxy;
• shielding-penetration burden proxy;
• plasma-facing heat-distribution proxy;
• cryogenic/control burden proxy;
• geometry/parameter-error Monte Carlo.

These outputs intentionally have screening authority only. They do not replace FEA, materials qualification, neutronics, thermal-hydraulics, or cryogenic engineering.

Evidence and falsification infrastructure

• A/B candidate/control protocol;
• ablations;
• negative controls;
• deterministic seeds;
• provenance hashes;
• machine-readable evidence bundle;
• explicit loss ledger;
• kill criteria;
• staged claim authority;
• release SHA-256 manifest;
• one authoritative GREEN/RED command.

A useful negative result already exposed

The no_axis_helical_shaping ablation can improve the internal composite objective.

That is not evidence that the 3-D stellarator shape should be removed. It exposes an important limitation in the reduced model: rotational transform is currently parameterized rather than generated self-consistently from a solved coil/equilibrium field.

IX-Fusion retains that result instead of hiding it. It is one of the reasons the next gate must use an actual equilibrium and coil solver.

Model-authority ladder

geometry hypothesis
        ↓
solved 3-D equilibrium + nested surfaces + transform
        ↓
magnetic candidate
        ↓
coil realization + error fields + particle orbits
        ↓
confinement candidate
        ↓
finite-pressure stability + turbulence + edge/divertor + RF full-wave
        ↓
plasma candidate
        ↓
blanket/neutronics + fuel cycle + complete plant energy ledger
        ↓
fusion-performance candidate

Release 0.1 remains at the first stage.

Intended external solver path

The repository detects but does not bundle specialist solvers. Its evidence contracts are designed for later integration of:

• DESC or equivalent for 3-D MHD equilibrium/optimization;
• SIMSOPT or equivalent for stellarator coil and field optimization;
• VMEC-class equilibrium as an independent cross-check;
• guiding-center / fast-particle tools;
• MHD stability analysis;
• gyrokinetic / validated transport analysis;
• RF full-wave / deposition tools;
• neutronics / blanket / fuel-cycle tools.

An unavailable or failed external solver must remain UNKNOWN; the internal model is not allowed to impersonate a higher-authority result.

Run it

Python 3.11 or later is required.

python -m venv .venv

Windows PowerShell:

.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -e .
python check_green.py

Linux/macOS:

. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e .
python check_green.py

Expected final line for an intact release:

IX-FUSION: GREEN

GREEN means repository integrity and reduced-order evidence reproduce. It does not mean fusion has been achieved.

Full walkthrough: docs/08_BUILD_AND_VALIDATE.md.

Reproduce the research artifacts

python scripts/run_poc.py
python scripts/run_secondary_studies.py
python scripts/reproduce_release.py --verify

To intentionally refresh committed release artifacts and their hash manifest:

python scripts/generate_release.py

Repository map

IX-Fusion/
├── src/ix_fusion/              # reduced scientific/control/evidence models
├── configs/                    # candidate, control, optimization and system inputs
├── tests/                      # deterministic unit and scientific-regression tests
├── results/                    # generated POC, baselines, ablations, Monte Carlo, evidence
├── schemas/                    # candidate/evidence/external-solver contracts
├── provenance/                 # hashes of user-supplied conceptual source repositories
├── BOM/                        # software, conceptual-system and research evidence BOMs
├── docs/                       # claim boundary, validation path, safety, assumptions, roadmap
├── scripts/                    # reproduction, manifest and release commands
├── check_green.py              # authoritative local/CI quality gate
├── PROOF_OF_CONCEPT.md
├── VALIDATION_REPORT.md
├── LICENSE                     # IX-Fusion Research & Evaluation License
└── MANIFEST.sha256

Recommended reviewer path

For a skeptical technical review:

1. docs/02_CLAIM_BOUNDARY.md
2. PROOF_OF_CONCEPT.md
3. results/poc/verdict.json
4. src/ix_fusion/field.py
5. src/ix_fusion/omnigenity.py
6. src/ix_fusion/optimizer.py
7. results/ablations/c6_ablations.json
8. docs/12_MODEL_LIMITATIONS.md
9. results/evidence/loss_ledger.json
10. run python check_green.py

Safety and physical-build boundary

IX-Fusion is not a physical reactor construction package. The conceptual system BOM keeps major engineering categories visible but intentionally does not contain operational reactor specifications, tritium-handling procedures, radiation-producing experiment instructions, high-voltage/RF build procedures, superconducting-magnet fabrication instructions, or cryogenic plant construction procedures.

See docs/10_SAFETY_AND_SCOPE.md and docs/21_PHYSICAL_BUILD_BOUNDARY.md.

License

IX-Fusion is source-available for research/evaluation only under the custom IX-Fusion Research & Evaluation License 1.0. It is not Apache-2.0 and is not presented as OSI open-source software. Production, operational, manufacturing, distribution, and commercial use require separate written permission.

See LICENSE.

Technical basis and provenance

The validation path is informed by current stellarator work on direct omnigenity optimization, equilibrium optimization, coil optimization, coil/support co-optimization, HTS field shaping, and real-time plasma-control methods. References are separated from IX-Fusion findings in docs/17_REFERENCES.md.

Concepts translated from earlier project repositories are documented in docs/16_SOURCE_REUSE_MAP.md and source-archive hashes are recorded in provenance/ORIGIN_INPUTS.json. Prior-repository claims are not inherited as IX-Fusion evidence.

---

The standard for this repository is not whether C6 wins. It is whether the repository can tell us, reproducibly and without moving the goalposts, when C6 loses.