Last Updated: March 26, 2026
Research Program Status
Comprehensive source of truth for the entire BigBounce spin-torsion cosmology research program. Every paper, pipeline, MCMC run, computation script, and discovery in one place.
Houston Golden · Independent Researcher
Contents
1. Research Papers
| Paper | Title | Status | Version | Pages | Refs |
|---|---|---|---|---|---|
| Paper 1 | Spin-Torsion Cosmology: Structural Barriers, Falsifiable Predictions, and the Bounce-Inflation Landscape | Ready for Submission | v2.2.0 |
~24 | 63+ |
| Paper 2 | fNL = -35/8 Forecast: SPHEREx Discrimination of Bounce vs. Inflation | Ready for Submission | v1.3.0 |
~12 | 30+ |
| Paper 3 | DESI DR1 Spectral Anomaly Catalog: 195,829 Uncharacterized Objects from 18M Spectra | Draft | v0.1 |
~8 | TBD |
Paper 1 has undergone 10+ rounds of peer review with ~200 individual edits. Paper 2 contains the parameter-free fNL = -35/8 prediction testable by SPHEREx (~2028). Paper 3 documents the first full-DR1-scale autoencoder anomaly search (~90x prior EDR work).
2. Bounce Cosmology Portfolio
The mission is proving bounce cosmology beats inflation — not proving one specific model. Six independent observational channels, each with a distinct prediction.
| Channel | Best Model | Prediction | Experiment | Timeline | Status |
|---|---|---|---|---|---|
| Galaxy bispectrum fNL | Matter bounce | fNL = -35/8 = -4.375 (parameter-free) | SPHEREx | ~2028 | FLAGSHIP |
| Quintom bounce-DE | Quintom bounce | w(z) crosses -1 (quintom-B) | DESI DR2 | NOW | 2.3σ confirmed |
| PBH dark matter | Asymmetric matter bounce | Asteroid-mass PBHs, fPBH ~ 0.001–1 | LISA, microlensing | ~2035 | Viable (asym. only) |
| NANOGrav GW | Matter bounce | γ = 3.0 vs observed 3.2 ± 0.6 (0.33σ) | NANOGrav / PTA | NOW | 1σ consistent |
| GW echoes | Ekpyrotic bounce (GUT-scale) | Oscillatory ΩGW | CE / ET | ~2035 | Conditional |
| Perturbative safety | Cuscuton bounce | No strong coupling through bounce | (theoretical) | Complete | Supporting |
The Triple Role of fNL = -35/8
The matter bounce fNL simultaneously: (1) predicts galaxy bispectrum (SPHEREx), (2) regulates PBH abundance (prevents overproduction that plagues inflation), and (3) shapes induced GW spectrum (encodes clustering, detectable by LISA). This triple role is unique to matter bounce cosmology.
Quintom w-crossing at 98.6%
Our independent w0-wa MCMC (Planck + BAO + SN) finds w0 = -0.871, wa = -0.542, with P(quintom-B) = 98.6%. This independently confirms the DESI DR2 w-crossing signal at 2.3σ. The quintom bounce achieves what ECH alone cannot: bounce-DE unification.
NANOGrav Consistency
Matter bounce predicts GW spectral index γ = 3.0. NANOGrav 15yr measures γ = 3.2 ± 0.6. Bayesian model comparison: bounce preferred over SMBH at 5.6:1, over cosmic strings at 3.2:1.
3. AI Discovery Pipelines
Survey-scale AI extraction pipelines turning existing cosmological archives into calibrated catalogs, validated anomaly products, and improved tracer sets. All pipelines follow a 12-gate publication standard.
| Pipeline | Description | Feeds fNL? | Status | Progress |
|---|---|---|---|---|
| B — DESI Spectral Anomaly Miner | Autoencoder anomaly search on all 18M DESI DR1 spectra. First full-DR1-scale search (~90x prior EDR work). | Yes (high-bias tracers) | COMPLETE | 195K anomalies; 200/200 artifact-free; enhanced 18M at 44%; top 50 spectra plotted |
| 2 — Galaxy Chirality Catalog | Largest bias-audited galaxy handedness catalog (CW/CCW/NOT_SPIRAL). v2 model: 93.7% accuracy, 8/8 bias tests passed. | Indirect (parity test) | RUNNING | 7.3M / 8.47M galaxies (86.5%) |
| E — Time-Domain Transient Finder | unTimely 32-epoch IR variability analysis for anomalous transients and QSO enrichment. | Yes (QSO enrichment) | Not Started | — |
| A — CMB Anomaly Hunter | CNN/ViT anomaly detection on Planck/ACT CMB patches at multiple scales. | Indirect | Not Started | — |
| C — Polarization Feature Extractor | Task-specific polarization signal extraction from CMB Q/U maps (birefringence, strings, reionization). | Indirect (birefringence) | Not Started | — |
| D — Cross-Survey Anomaly Correlator | Multi-survey coincidence analysis. Requires calibrated outputs from A, B, C, and E. | Indirect | Blocked on A–C | — |
Pipeline B Headline Result
195,829 previously unidentified spectral anomalies from 18M DESI DR1 spectra. 200/200 top anomalies verified artifact-free by spectral inspection (peak wavelength vs known sky/telluric lines). Three wavelength clusters in top 50: 12 at 7600Å (possible [OIII] at z≈0.52), 28 at 3600–3700Å (possible Lyman break at z≈3), 3 at 9440–9480Å (possible [SIII]/Hα at z≈0.44). Cross-matched against 6 databases (3B+ objects): 99.8% absent from SIMBAD, galaxies 20x more anomalous than QSOs (6.5M spectra). Enhanced 18M catalog at 44%. Model on HuggingFace.
4. Active Compute
RunPod GPU/CPU pods. Do NOT terminate any pods without explicit user approval.
| Pod | Machine | Pipeline | SSH | Status | Progress |
|---|---|---|---|---|---|
H200 Beastrtv8cegaw1618r |
H200 (143 GB), 192 cores, 3 TB RAM | Enhanced 18M Catalog | root@205.196.17.44 -p 10789 |
RUNNING | 7.9M / 17.9M (44%), ~16h remaining |
H100ulfxypratod4vr |
H100 (80 GB), 208 cores, 2 TB RAM | Pipeline 2 — Chirality | root@64.247.201.47 -p 10778 |
RUNNING | 7.3M / 8.47M (86.5%), ~8h remaining |
RTX A4000fn19oivkjowmq4 |
A4000 (16 GB), 32 cores, 124 GB RAM | w0-wa MCMC | root@157.157.221.30 -p 24859 |
CONVERGED | 50.9K samples, R-1 < 0.01 |
CPU Batchkqo1b4e4igycra |
32 cores, 124 GB RAM (no GPU) | Pipeline B CPU (redundant) | root@157.157.221.29 -p 29268 |
Redundant | 368K spectra (superseded by H200) |
~/.ssh/id_ed25519. Backups: local disk, GitHub, Backblaze B2, HuggingFace (bamfai/bigbounce-mcmc).
5. MCMC Runs
All Markov chain Monte Carlo sampling runs. Posterior samples frozen and backed up to multiple locations.
| Dataset | Parameters | Samples | Status | Key Result |
|---|---|---|---|---|
| Full Tension Planck+BAO+SN+H0 |
ΛCDM + ΔNeff | 176,164 | Frozen | ΔNeff ≈ 0; H0 = 67.68 |
| Planck + BAO + SN No H0 prior |
ΛCDM + ΔNeff | 133,263 | Frozen | ΔNeff ≈ 0; standard ΛCDM |
| w0-wa Quintom Planck+BAO+SN (CPL) |
w0-wa dark energy EOS | 50,900 | CONVERGED | w0 = -0.871 ± 0.060, wa = -0.542 ± 0.245, P(quintom-B) = 98.6% |
6. Cross-Reference Status
Pipeline B anomalies cross-matched against 6 astronomical databases covering 3B+ cataloged objects.
| Database | Objects | Match Rate | Result |
|---|---|---|---|
| SIMBAD | ~15M objects | 0.2% | 99.8% of anomalies NOT in SIMBAD — genuinely uncharacterized |
| NED | ~400M objects | Low | Minimal overlap; anomalies are not in standard extragalactic catalogs |
| AllWISE | ~750M sources | Moderate | IR counterparts available for photometric classification |
| Gaia DR3 | ~1.8B stars | Low | Only 1 confirmed Galactic star in top 100; anomalies are extragalactic |
| SDSS DR17 | ~1B objects | Low | No known QSOs in top 100; 0% QSO match rate |
| Legacy Survey DR10 | ~2B objects | High (imaging) | Imaging counterparts available; morphology extraction underway |
Key Cross-Match Finding
Galaxies are 19x more anomalous than QSOs in our catalog. Of the top 1,000 anomalies, the overwhelming majority are galaxies with unusual spectral features, not QSOs or stars. This suggests the autoencoder is finding genuinely unusual galaxy populations that standard pipelines classify but do not flag as noteworthy.
7. Computation Scripts
Every numerical claim in the papers is backed by a reproducible Python script or Cobaya YAML config.
| # | Script | Purpose | Status |
|---|---|---|---|
| 1 | alp_field_evolution.py | ALP field evolution for birefringence prediction β = 0.27° | Done |
| 2 | fnl_epsilon_correction.py | fNL epsilon slow-roll correction verification | Done |
| 3 | transparency_verification.py | ECH perturbation transparency verification | Done |
| 4 | template_projection.py | Bounce template projection for fNL constraint recasts | Done |
| 5 | photon_torsion_coupling.py | Photon-torsion coupling strength computation | Done |
| 6 | galaxy_spin_tidal_torque.py | Galaxy spin tidal torque model (chirality) | Done |
| 7 | lqc_bounce_perturbations.py | LQC bounce perturbation evolution | Done |
| 8 | cubic_transmission.py | Cubic transmission coefficient through bounce | Done |
| 9 | fisher_forecast.py | Fisher forecast for SPHEREx fNL sensitivity | Done |
| 10 | photoz_degradation.py | Photo-z degradation impact on fNL constraints | Done |
| 11 | bphi_sensitivity.py | bφ tracer bias sensitivity analysis | Done |
| 12 | quintom_fnl_verification.py | Verify fNL = -35/8 across 3 bounce models (all give -35/8 exactly) | Done |
| 13 | pbh_nanograv_consistency.py | PBH regulation + NANOGrav γ = 3 consistency | Done |
| 14 | nanograv_model_comparison.py | Bayesian bounce vs SMBH vs cosmic strings (bounce 5.6:1) | Done |
MCMC configs: cobaya_w0wa_quintom_test.yaml + 4 frozen ΔNeff dataset configs in reproducibility/cosmology/. Post-processing: analyze_w0wa_quintom.py.
8. Future Plans
| Priority | Task | Description | Status | Depends On |
|---|---|---|---|---|
| 1 | Tracer Purification (Steps 2–6) | Cross-match 195K anomalies with Legacy Survey DR10 + unWISE + Gaia DR3. Classify high-z QSOs. Validate bias enhancement. Re-measure σ(fNL). | Next | Pipeline B (done) |
| 2 | Multi-Pass Triage | 6-pass enrichment: band ratios (done) → cross-match → morphology → redshift → LLM reasoning → clustering/embedding. | Pass 1 Done | Pipeline B output |
| 3 | LLM Deep Analysis | Claude/GPT-4o analyzes top 1,000 anomalies with full context: band residuals + Legacy Survey image + cross-match + DESI metadata. | Not Started | Pass 2–4 |
| 4 | Full 18M Structured Catalog | Run band-ratio classification on ALL 18M spectra, not just anomalies. Community resource for the entire DESI DR1. | Not Started | H200 pod data |
| 5 | Super-Resolution Imaging | Physics-aware AI upscaling of Legacy Survey cutouts using paired HST/JWST training data. Separate project. | Not Started | Training data curation |
| 6 | Complete 8.47M Chirality Catalog | Finish inference on remaining 1.17M galaxies. v2 model frozen (93.7% accuracy, equivariant CW = 0.5012). | Running (86.5%) | H100 pod |
| 7 | Quintom fNL Computation | Fill the literature gap: no quintom bounce fNL has ever been computed. If confirmed as -35/8, this is a universality result. | Verification Done | Full in-in integral |
| 8 | Pipeline E (Time-Domain) | unTimely 32-epoch IR variability for anomalous transients. Feeds QSO enrichment for fNL. | Not Started | GPU pod allocation |
H200 Research Opportunities
Additional runs that can be executed on the H200 GPU pod. Budget: $250-500/day. Full details in project-context/h200_research_opportunities.md.
| Category | Run | Est. Cost | Novel? | Value |
|---|---|---|---|---|
| Bounce Cosmology | ||||
| fNL tracer optimization from 18M catalog | $15-25 | YES | Directly improves flagship prediction | |
| Planck lensing × anomaly cross-correlation | $5-10 | YES | Independent bias measurement | |
| NANOGrav spectral template fit | $5 | YES | Proper Bayes factor calculation | |
| Anomaly Pipeline | ||||
| Second autoencoder on anomalies (recursive) | $20-30 | YES | Finds most extreme objects | |
| Injection/recovery with real BigAE model | $10-15 | REQUIRED | Publication gate (currently proxy only) | |
| Download all 195K spectra + analysis | $30-50 | YES | Full spectral catalog | |
| SDSS DR18 cross-validation (5M spectra) | $15-25 | YES | Proves methodology is survey-independent | |
| Multi-resolution autoencoder (8x/16x/32x/64x) | $20-30 | YES | Finds scale-dependent anomalies | |
| Broader Astrophysics | ||||
| Super-resolution imaging (Legacy → HST/JWST) | $50-100 | YES | Highest visibility, most shareable | |
| Emission line finder for 18M spectra | $30-50 | Partial | Automated line catalog | |
| Latent space photo-z estimation | $25-40 | YES | Novel ML methodology | |
| Galaxy merger finder on anomaly images | $30-50 | Partial | Connects spectral + morphological anomaly | |
| Full-sky anomaly density map | $20-30 | YES | Spatial systematics check | |
| Spectral anomaly taxonomy (UMAP + HDBSCAN) | $40-60 | YES | Defines NEW object classes | |
| Multi-H200 Scale-Up | ||||
| Full spectral download + 8x re-scoring (2 H200s) | $200-400 | YES | Per-pixel anomaly localization | |
| Cross-survey detection: DESI + SDSS + LAMOST (3 H200s) | $300-500 | YES | Multi-survey validation | |
| DESI DR2 first-to-publish readiness | $50-100 | YES | Massive citation potential | |
Total estimated cost for all runs: $905–$1,530 over 2–4 weeks. Top ROI: spectral taxonomy ($40-60), second autoencoder ($20-30), injection/recovery ($10-15).
9. Key Discoveries
14 ECH Structural Barriers
Systematic analysis revealed 14 structural barriers that close all minimal routes from the ECH bounce to dark energy. Crucially, these are ECH-specific — other bounce models (quintom, Cuscuton) can bypass them. The barriers MAP the requirements for bounce-DE unification rather than ruling out bounce cosmology.
Foundation Studies A–G · Branches H–WfNL = -35/8 Mechanism Independence
Verified across 3 bounce models (matter bounce, LQC, Cuscuton contrast): fNL = -35/8 = -4.375 is parameter-free and mechanism-independent. The contraction dynamics determine fNL, not the bounce mechanism. SPHEREx (~2028) will measure this to σ(fNL) ~ 0.7–2.
Paper 2 · quintom_fnl_verification.pyPBH Regulation by fNL
fNL = -35/8 naturally prevents primordial black hole overproduction (Choudhury+ 2025). Inflationary models with positive fNL tend to overproduce PBHs requiring fine-tuning. The matter bounce’s negative fNL acts as a natural regulator: 10-3 < fPBH < 1.
pbh_nanograv_consistency.py · Choudhury+ 2025 (EPJC 85:472)NANOGrav Consistency
Matter bounce predicts induced GW spectral index γ = 3.0 (Papanikolaou 2025). NANOGrav 15yr measures γ = 3.2 ± 0.6 — just 0.33σ away. Bayesian model comparison: bounce preferred 5.6:1 over SMBH binary background.
nanograv_model_comparison.py · Papanikolaou 2025 (arXiv:2504.11641)Quintom-B w-Crossing at 98%
Independent MCMC (50.9K samples, converged R-1 < 0.01) finds w0 = -0.871 ± 0.060, wa = -0.542 ± 0.245. P(quintom-B w-crossing) = 98.6%. Confirms DESI DR2 signal at 2.3σ. Quintom bounce achieves bounce-DE unification.
cobaya_w0wa_quintom_test.yaml · 50.9K posterior samples195,829 Uncharacterized Spectral Anomalies
First full-DR1-scale autoencoder anomaly search on DESI (~90x prior EDR work by Liang+ 2023 and Nicolaou+ 2026). 99.8% absent from SIMBAD. 0% are known QSOs. Only 1 confirmed Galactic star in top 100. Galaxies 19x more anomalous than QSOs.
Pipeline B · 18M spectra · HuggingFace: bamfai/desi-spectral-anomaly-detectorALP Birefringence Prediction
ALP birefringence prediction β = 0.27° matches the observed 3.6σ signal of 0.342 ± 0.094° from Planck/WMAP combined data. Consistent at 0.8σ.
alp_field_evolution.py · Paper 1 Section 410. Additional Datasets & Surveys
Datasets where our AI anomaly detection methodology can be applied next. Ranked by discovery potential. Full details in project-context/additional_datasets_and_pipelines.md.
| Dataset | Size | Type | Est. Cost | Connection to Bounce | Status |
|---|---|---|---|---|---|
| Run Now (data public, infrastructure ready) | |||||
| NEOWISE / unTimely | 170B rows, 2B sources | IR time-domain | $100-200 | Variable QSOs for fNL tracers | Pipeline E |
| Planck CMB Maps | 50GB full-sky | CMB T/Q/U patches | $50-100 | Pre-bounce perturbation signatures | Pipeline A |
| ACT DR6 Polarization | 20GB | CMB polarization | $50 | Tests β = 0.27° prediction | Pipeline C |
| NANOGrav 15yr | ~1GB | PTA timing | Free | Bounce GW template fit (γ=3) | Starting |
| Run Soon (need model adaptation) | |||||
| SDSS DR18 | 5M spectra | Optical spectroscopy | $30-50 | Cross-survey validation | API Down |
| LAMOST DR10 | 20M spectra | Optical spectroscopy | $50-100 | More fNL tracers | Not Started |
| Gaia DR3 BP/RP | 220M spectra | Low-res spectrophotometry | $100-200 | Unusual stellar populations | Not Started |
| Future (data not yet available) | |||||
| Euclid | ~30M spectra | Slitless spectroscopy | $100-200 | Next-gen fNL tracers | ~2027 |
| SPHEREx | ~300M objects | 96-band spectrophotometry | TBD | FLAGSHIP — direct fNL measurement | ~2028 |
| LSST / Rubin | ~20B objects | Time-domain + photometric | TBD | Galaxy chirality at z>1 | ~2025-2035 |
| CMB-S4 | Full-sky CMB | Next-gen CMB | TBD | Definitive β test (~30σ) | ~2030s |
Hubify Lab Vision
Build a scalable, repeatable AI archival discovery platform: each astronomical survey gets its own autoencoder pipeline, anomaly catalog, cross-reference, and explorer UI. Results cross-reference across surveys — objects flagged by 2+ surveys are highest priority. Infrastructure: H200 GPU for training/inference, Convex for data, HuggingFace for models, GitHub for code. $50–200 per survey, 1–4 weeks each. DESI DR1 is the proof-of-concept. To be developed as a standalone platform at hubify.com.
11. Biggest Questions in the Universe
Future research paths connecting bounce cosmology to the deepest open questions in physics. Some are testable with current infrastructure. Others are generational challenges. All are worth pursuing.
What Is Dark Energy?
73% of the universe is dark energy and we don't know what it is. Our 14 barriers show ECH can't derive it from first principles, but the quintom bounce CAN unify bounce + dark energy through phantom fields. DESI DR2 shows w-crossing at 2.8–4.2σ. Our MCMC confirms at 98%. If w truly crosses -1, dark energy is dynamical, not a cosmological constant — and the quintom bounce provides a theoretical home for it.
Testable now · w0-wa MCMC · DESI DR2 · QuintomWhat Is Dark Matter?
27% of the universe is dark matter. In bounce cosmology, the asymmetric matter bounce can produce asteroid-mass primordial black holes (1017–1024 g) as dark matter candidates. Our fNL = −35/8 naturally regulates PBH abundance, preventing overproduction. If LISA detects the induced gravitational wave spectrum from PBH formation, it would be evidence for both dark matter AND a cosmological bounce.
Future · LISA ~2035 · PBH dark matter · fNL regulationHow Old Is the Universe — Really?
Standard cosmology says 13.8 billion years (since the Big Bang). But if the Big Bang was a Big Bounce, there was a BEFORE. The parent universe that collapsed into the bounce could be arbitrarily old. The Hubble tension (H0 = 67 vs 73) might hint that our age estimate is wrong. Bounce cosmology allows for a pre-bounce era whose duration is unconstrained — the universe could be far older than 13.8 Gyr, or even eternal through cyclic bounces.
Theoretical · Hubble tension · Cyclic cosmology · Pre-CMB physicsFate of the Universe: Big Crunch, Big Freeze, or Eternal Cycles?
Standard cosmology predicts a Big Freeze (eternal expansion into cold emptiness). But if dark energy is dynamical (quintom), it could reverse — leading to a future Big Crunch and another bounce. Cyclic cosmology suggests the universe bounces repeatedly, each cycle spawning new physics. Our quintom-B result (w0 + wa = −1.4) is in the regime where phantom energy could eventually dominate and trigger recollapse.
Theoretical · Quintom dynamics · Cyclic cosmologyBaby Universes: Does Every Black Hole Spawn a New Cosmos?
Popławski’s scenario (the theoretical foundation of our ECH framework) proposes that every black hole interior undergoes a torsion-regulated bounce, creating a baby universe on the other side. If true, our universe was born inside a black hole in a parent universe, and every black hole we observe (~1018 in the observable universe) contains its own cosmos. The Barbero-Immirzi parameter γ = 0.274 could vary across the hierarchy, creating a multiverse of different physics.
Speculative · Popławski (2010) · Black hole cosmology · MultiverseWhat Signatures Escape a Black Hole to Test If a Bounce Happened?
If the universe bounced inside a black hole, what observable evidence could leak out? Candidates: (1) Gravitational wave echoes from the bounce (Cai & Zhu 2026 — we tested against Barrier 12), (2) Primordial non-Gaussianity imprinted during contraction (our fNL = −35/8), (3) Cosmic parity violation from the parent black hole’s rotation (our birefringence prediction), (4) Anomalous CMB patterns from pre-bounce perturbations. SPHEREx and LiteBIRD will test predictions (2) and (3) directly.
Testable · SPHEREx ~2028 · LiteBIRD ~2032 · GW echoesCan Black Holes Die?
Hawking radiation predicts black holes slowly evaporate. But if every black hole contains a baby universe, does evaporation destroy that universe? The information paradox meets cosmology: does quantum information from the parent universe survive the bounce? Our perturbation-transparency theorem shows that for scalar fields, information passes through the bounce unchanged — but what about quantum entanglement between the parent and baby universe?
Theoretical · Information paradox · Hawking radiation · Quantum gravityQuantum Information, Time Travel, and the Speed of Light
The bounce connects two epochs through a Planck-density transition. Does information travel “backward” through the bounce (from expansion to contraction in the parent)? If the bounce preserves quantum coherence, it could inform theories of quantum teleportation across cosmological horizons. While human time travel remains science fiction, understanding whether the bounce preserves or scrambles quantum information is a real physics question that connects to the black hole information paradox, ER=EPR, and quantum error correction in curved spacetime.
Speculative · Quantum information · ER=EPR · Planck-scale physics195,829 Uncharacterized Objects: What ARE They?
Our AI pipeline found 195,829 spectral anomalies that are absent from essentially every major catalog. Galaxies are 18x more anomalous than QSOs. The top 200 are verified genuine (0% artifacts). Three wavelength clusters suggest distinct physical populations. When the enhanced 18M catalog provides redshifts, we may discover entirely new classes of astronomical objects — objects the scientific community has never seen or named before. What are they?
Active research · Enhanced 18M (51%) · Paper 3 · DESI DR1