Repository navigation
The Scalar–Singlet Saddle (S³): Minimal Repro
Deterministic “theoretical experiment” scripts that produce the key
certificates/constants referenced in the paper.
- No network access required
- CPU only, fast, fully reproducible
- Outputs written to
outputs/
What this is
This repo contains a single-file, reproducible pipeline that ties together:
- TT sector normalization (\kappa_\infty) computed ab initio (Bessel-factorized baseline) with an independent continuum “Route A” cross-check (TT-complete via symmetry ×6, polar patch around the diagonal, and (r_0!\to!0) extrapolation).
- The Fejér–York lock that produces the effective length (L_{\mathrm{eff}}), the small parameter (\varepsilon), reduced mass (m_r=1/\varepsilon), and a prediction for (\alpha^{-1}) from the same rigidity pipeline.
- Rigorous error control: absolutely convergent expressions, analytic tail/remainder bounds, and guarded quadrature near the (t_1!\approx!0) singular manifold.
- Ablations & diagnostics: York toggle, parity “what-if”, sensitivity (d(\alpha^{-1})/dL), cached vs. computed (\kappa_\infty), and convergence/extrapolation scans.
- Commutator experiments (z_\chi): projector-based and true low-(q) commutator probes that do not alter the kinematic York coefficient (used only for diagnostics; CODATA is shown for context rather than as an input).
The goal is to blunt “numerology” critiques by (i) deriving every ingredient from the same (L_{\mathrm{eff}})/rigidity scheme with explicit bounds, and (ii) providing an out-of-sector cross-check (Route A) that shares no discretization artifacts with the baseline.
Highlights
- One-file run: No figures, just scalars suitable for the paper’s “Lock Universality and Ablations” section.
- Two independent (\kappa_\infty) routes:
- Baseline: Bessel-factorized TT integrals.
- Route A: Continuum integral with TT completion (×6), polar patch around (t_2=t_3=0), and linear (r_0)-extrapolation to remove the regulated core.
- Fejér–York lock (series-exact): (L_0=\log(15/8)), York increment (11/1560), odd/even tails evaluated with absolute convergence and rigorous bounds.
- Precision on rails: Configurable decimal precision for the lock and integrators; all tolerances are finite and guarded (no infinite loops).
- Remainder bounds everywhere: Poisson-summed/series tails and polar-patch contributions carry explicit, printed bounds.
- Ablations you can trust: York toggle, optional parity knob, and first-order (\Delta L) needed to match an external target (diagnostic only).
- Sensitivity reporting: Exact (d(\alpha^{-1})/dL) at the evaluated (L_{\mathrm{eff}}) to translate tiny theory shifts into ppb predictions.
- Convergence dashboards: GL node sweeps, Richardson-style extrapolation for Route A, and slope diagnostics.
- Safe near singularities: Open-interval and polar decompositions plus small-(r) analytic limits avoid division-by-zero traps.
- Cache or compute: Use the cached (\kappa_\infty) for quick runs or recompute both routes when you want a clean ab initio pass.
- Transparent (z_\chi) probes: Implements projector and true commutator measurements to quantify (not force) any mismatch; by design, York’s kinematic coefficient is unchanged by these probes.
Requirements
- Python 3.11 (recommended)
- Packages pinned in
requirements.txt(e.g.mpmath)
Install once:
python -m pip install -r requirements.txtQuick start (Windows PowerShell)
python -m venv .venv
. .\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
.\run_all.cmdQuick start (macOS / Linux)
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
# Manual sequence (same as run_all.cmd)
python run.py xq.py --force
python run.py m.py --force # run.py passes --xq from xq.py into m.py
python run.py kappa.py --force
python run.py a1g.py --force
python run.py alpha.py --forceTip: Omit
--forceto re-use cached results inoutputs/.
Expected checks (sanity)
After a successful run you should see:
outputs/m.json"m_first": 19- Row for
m=19shows"ord2": 18,"dyadic_pass": true,"isolation_pass": true
outputs/xq.jsonx_q ≈ 1.3297024653297962…L0 = log(15/8)
outputs/alpha.jsonalpha^{-1} ≈ 137.035999176279489…- Internal uncertainty split consistent with the paper
How things plug together
xq.pycomputes the EL-lock abscissax_q(Fejér lock).m.pyreads--xqvia utils and certifies the dyadic gate & isolation at the EL lock.kappa.pyproduces the Schlömilch/Bessel integral certificateκ∞.a1g.pyand companions derive group-theoretic factors used in the master relation.alpha.pyassembles everything to reproduce the predictedα^{-1}.
All scripts follow the same CLI contract and shared helpers in utils.py.
Re-running & cache
Results are cached under outputs/. Use --force on any run.py <script> --force
invocation to recompute that step from scratch.
Troubleshooting
- Module not found: activate the virtual environment (
. .\.venv\Scripts\Activate.ps1on Windows orsource .venv/bin/activateon Unix) and reinstall requirements. - Different Python version: prefer Python 3.11 to match expected numerics and JSON formatting.
License
See LICENSE for code licensing. If a separate data license applies to files in outputs/, see the repository notes.
Cite this software
BibTeX
@software{stieger_york_fejer_dL2_2025_v110,
author = {Gustav Stieger},
title = {quantumagi/york: York--Fej{\'e}r $\Delta L^{2}$ (v1.0)},
version = {v1.0.1},
year = {2025},
publisher = {Zenodo},
doi = {10.5281/zenodo.17179996},
url = {https://doi.org/10.5281/zenodo.17179996}
}Related Paper
BibTeX
@misc{Stieger2025S3,
author = {Stieger, Gustav},
title = {The Scalar--Singlet Saddle (S$^3$): Predicting the Fine-Structure Constant $\alpha$ from a Strict Variational Principle},
howpublished = {Zenodo},
year = {2025},
month = {September},
doi = {10.5281/zenodo.17180680},
url = {https://doi.org/10.5281/zenodo.17180680},
note = {Preprint, version v1.1-preprint. License: Other (Not open/All rights reserved).}
}Changelog
v1.0 (initial):
- Minimal script set (
src/), deterministicoutputs/, PowerShell runner, and JSON quick checks. - Matches paper constants and downstream ((\alpha^{-1}, m_r)) within the stated tolerances.
V1.0.1:
- MIT LICENSE added.
v1.1:
- Scripts added/updated