bamfai/bigbounce-mcmc

BigBounce MCMC chains

Canonical Markov-Chain Monte Carlo posteriors for the BigBounce spin-torsion cosmology program (Golden 2026, Papers 1A/1B/2/3/4).

Latest converged chain — iter2_converged_2026-05-18/

DESI DR2 BAO + Planck PR4 + DES-Y5 + Pantheon+ joint w₀wₐ posterior.

Sampler: Cobaya v3.6.1 Metropolis-Hastings, 16-chain MPI, OMP=6, RTX A5000.

Headline result

• —w₀ = −0.812 ± 0.044 — departing from ΛCDM (w=−1) at +4.3σ
• —wₐ = −0.667 ± 0.186 — departing from ΛCDM (wₐ=0) at −3.6σ
• —w₀ + wₐ = −1.48 ± 0.15 — phantom crossing required (quintom signature)
• —H₀ = 67.185 ± 0.455 km/s/Mpc (no-SH0ES posterior — chain is BAO + CMB + SN only)

This is the empirical posterior used as the quintom-B empirical anchor in Paper 1A §VII.H Structural Tension and Paper 1B Table 1B.

Files

• —iter2_converged_2026-05-18/chains_w0wa_dr2_iter2_chain_bundle.tar.gz (24.9 MB compressed; 89 MB uncompressed) — 16 chain .txt files + Cobaya configs (input.yaml, updated.yaml, progress, checkpoint, covmat)
• —iter2_converged_2026-05-18/mcmc_iter2_omp6.log (10.2 MB) — full sampler stdout/stderr with per-flush diagnostics
• —iter2_converged_2026-05-18/posterior_summary.txt — GetDist mean ± σ summary for all 17 sampled + 11 derived params + χ² decomposition
• —iter2_converged_2026-05-18/shoes_yaml_audit.md — verification audit of the full_tension chain's SH0ES likelihood configuration (resolves multi-round reviewer BLOCKER R5+R7+R10)
• —iter2_converged_2026-05-18/BACKUP_MANIFEST.md — full backup provenance with sha256 checksums for the 3-location backup (HF + local disk + GitHub manifest)

Reproducibility

huggingface-cli download bamfai/bigbounce-mcmc \
    iter2_converged_2026-05-18/chains_w0wa_dr2_iter2_chain_bundle.tar.gz \
    --repo-type dataset --local-dir ./

tar -xzf iter2_converged_2026-05-18/chains_w0wa_dr2_iter2_chain_bundle.tar.gz

# Expected sha256 of the bundle: 74cf4e609ab6b2f02fc04b3f930f805607f53181868c50eaa9bcd536443d8a4e

python3 -c "
from getdist import loadMCSamples
s = loadMCSamples('chains_w0wa_dr2_iter2/spin_torsion_dr2', settings={'ignore_rows': 0.3})
for p in ['H0', 'omegam', 'sigma8', 'S8', 'w', 'wa']:
    print(f'{p:<8} = {s.mean(p):.4f} ± {s.std(p):.4f}')
"

License & citation

CC BY 4.0. If you use these chains, please cite the relevant paper in the BigBounce program (Golden 2026a/b/c/d, see github.com/Hubify-Projects/bigbounce).