LAXFORGE is a gauge-aware discovery engine for zero-curvature representations and Lax pairs. Its purpose is not to generate decorative symbolic coincidences, but to produce candidates that survive algebraic validation, gauge/fake-pair reduction, invariant comparison, hierarchy extraction, and prior-art collision checks.
This repository scaffold is designed for incremental implementation by Codex-style coding agents. It includes:
- A formal implementation specification.
- Module boundaries and function contracts.
- Agent prompts for staged implementation.
- A first calibration example: the second-jet nilpotent mKdV lift.
- A minimal symbolic validation script and pytest target.
The philosophical rule is simple:
A pair is not new because we have not seen it before. It is new only after it survives zero-curvature validation, gauge reduction, invariant comparison, and serious prior-art collision.
python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'
pytest -q
python scripts/run_mkdv_validation.pyTo emit a LAXCERT-ingestable calibration artifact:
python scripts/export_laxcert_calibration.py runs/laxcert_calibration --overwriteThe output directory contains laxforge_manifest.json, candidate.json, and
source_report.json. The default candidate is the LAXCERT section-10
LaxforgeAKNSD2TransportZero calibration with second-order diagonal operators
and paired-field off-diagonal entries. LAXCERT can certify that directory
directly.
The first milestone is not to claim novelty. It is to calibrate the symbolic engine on a known-valid construction:
[ Q = u + \epsilon v + \epsilon^2 w, \qquad \epsilon^3=0, ]
with the AKNS/mKdV pair lifted into the truncated algebra (\mathbb{R}[\epsilon]/(\epsilon^3)). This validates the algebra, curvature expansion, coefficient extraction, and proof-emission pipeline.
src/laxforge/
algebra/ # coefficient algebras, truncated and finite-table arithmetic
core/ # curvature, ansatz, gauge, cyclic basis, invariants
examples/ # verified examples and calibration targets
tests/ # pytest suite
scripts/ # runnable demos and validation reports
docs/ # formal specs, architecture, metrics, run matrix
prompts/ # Codex agent prompts by module
LAXFORGE should act like a mathematical clean room. Every candidate receives a structured dossier:
- Candidate pair ((U,V)).
- Generated PDE or target PDE.
- Zero-curvature proof.
- Gauge/fake-pair tests.
- Cyclic-basis fingerprint.
- Essential spectral-parameter test.
- Conservation-law extraction.
- Hamiltonian/bi-Hamiltonian attempts.
- Known-hierarchy collision search.
- Publishability classification.