Checking the pion scattering lengths at threshold
Extrapolate the archived S-wave median meshes toward threshold and compare the two isospin channels with the scattering lengths reported in Table IV.
Approach threshold from the mesh
The authors report S-wave scattering lengths for the isoscalar (S0) and isotensor (S2) channels. Their archived energy grid begins at 280 MeV, just above the two-pion threshold of 279.14 MeV. A convenient near-threshold variable is the squared center-of-mass momentum, q² = E²/4 − mπ², with mπ = 139.57 MeV.
Take the archived pointwise median of the real partial wave at 280 and 280.5 MeV. Fit a straight line in q² through these two values and evaluate it at zero. In the authors’ convention this intercept estimates the dimensionless product of the pion mass and the scattering length. Repeat with 281 MeV as the second energy to see how the choice of grid points affects the estimate.
Compare the two channels
| Wave | Table IV median | 280–280.5 MeV extrapolation | 280–281 MeV extrapolation |
|---|---|---|---|
| S0 | 0.222 | 0.221886 | 0.221885 |
| S2 | -0.041 | -0.040534 | -0.040478 |
Both narrow-grid estimates round to the reported medians at the precision of Table IV. Extrapolating the lower and upper pointwise curves separately gives two further diagnostics in the output file. The individual replicas can change rank between energies. A threshold interval would require threshold estimates for the individual replicas before taking quantiles.
Check the phase convention
For selected individual-replica mesh values, reconstruct the real amplitude using Eq. (6), Re t = η sin(2δ)/(2ρ), where ρ = √(1 − (2mπ/E)²). The archived phase shifts are in degrees, so the script converts them to radians. It compares 18 values across 3 replicas to their stored real amplitudes at absolute tolerance 1e-12. The largest difference is 1.8e-15. Omitting the degree conversion makes 18 comparisons fail.
These checks use selected cells of the authors’ archived fit. They do not retrain the network, recompute all replica quantiles, or test the Roy dispersion relations across the fitted energy range.
Repeat the calculation
Save the Python script, selected mesh inputs, and recorded output together. The inputs are selected and restructured values from Wyatt A. Smith’s production ensemble, released under CC BY 4.0. Run the script with Python 3 and its standard library.
python3 -I check.py > rerun.json
cmp result.json rerun.jsonFor a fresh extraction, the Zenodo release contains Git LFS pointers with file sizes and digests. Download the two ensemble arrays and three replica arrays listed in the script as numerical NPZ files from the matching repository revision. Use the raw file download to resolve each pointer and save the five NPZ files together. Supply that directory with python3 -I check.py --from-mesh DIR. The script checks the five file digests. The JSON result retains the values before rounding for display.
Sources and interpretation
S-matrix informed neural networks for amplitude analysis. arXiv:2608.23750v1. Eqs. (5)–(6), partial-wave convention. Eqs. (10)–(11), Roy equations. Eqs. (19)–(20), threshold definition and pionium input. §§IV–V, representation, training and selection. §VI C, Table IV and Fig. 9, scattering lengths. Appendix F, initialization and numerical settings.
Read the research entry for the role of AI in the original work and the range of its conclusions.