The Simple Explainer

What Did We Discover?

The short version: We found two ways to test whether the Big Bang was actually a "Big Bounce" — a rebound from a universe that existed before ours. One test involves a specific pattern in how galaxies cluster together. The other involves a subtle rotation in the oldest light in the universe. Both tests will be performed by upcoming space missions within the next few years.

We also built AI pipelines that discovered 195,829 previously uncharacterized astronomical objects in publicly available survey data — objects that nobody had cataloged before.

The honest part: We started by trying to connect the bounce to dark energy (the mysterious force accelerating the universe's expansion). That failed — we found 14 independent reasons why it can't work. But the failure was productive: it told us exactly where to look instead, and the two surviving predictions are sharp, parameter-free, and testable.

The Two Big Cosmic Puzzles

Right now, our best theory of the universe is called Lambda-CDM LAM-duh see-dee-em. It stands for "Lambda Cold Dark Matter" and it's been the gold standard for over 20 years. It explains the Big Bang, how galaxies form, dark matter, dark energy — basically everything we see.

But there's a problem. Actually, two problems. And they're getting worse.

Puzzle #1: The Hubble Tension H-naught TEN-shun

H₀ H-naught is the Hubble constant. It's a single number that tells us how fast the universe is expanding right now.

The "tension" is that we get two different answers depending on how we measure it:

Method 1 — Looking at nearby galaxies: When astronomers measure the distances and speeds of galaxies near us (using supernovae as "standard candles"), they get 73.0 km/s/Mpc.

Method 2 — Looking at the oldest light in the universe: When we analyze the Cosmic Microwave Background (the afterglow of the Big Bang), the math says it should be 67.4 km/s/Mpc.

Those two numbers disagree by ~4.9 sigma SIG-muh — that's a statistical way of saying there's roughly a 1-in-1,000,000 chance this is just a fluke. In physics, 5 sigma is the threshold for calling something a discovery. At ~4.9, we're essentially there.

It's like two thermometers giving different temperatures for the same room. Either one of them is broken, or something weird is going on with the room.

Puzzle #2: The Sigma-Eight Tension SIG-muh eight TEN-shun

σ₈ sigma-eight measures how clumpy the universe is — specifically, how much matter clumps together on scales of about 26 million light-years.

Again, two methods give two different answers. The primary tension is between the CMB (Planck) and direct weak lensing measurements (KiDS-1000, DES Y3):

CMB (Planck) predicts: S₈ = 0.832 ± 0.013 (more lumpy)

KiDS-1000 measures: S₈ = 0.759 ± 0.024 (less lumpy)

DES Y3 measures: S₈ = 0.776 ± 0.017 (less lumpy)

This disagreement sits at 2–3 sigma — not as dramatic as the Hubble tension, but persistent across multiple independent weak lensing experiments. (Note: S₈ ≡ σ₈√(Ω_m/0.3) is the preferred quantity for comparing CMB and lensing results.)

Why These Puzzles Matter

These aren't minor technical glitches. They may be cracks in the foundation of our understanding of the universe. In the history of physics, small mismatches like these have led to revolutions:

• Mercury's orbit was slightly off from Newton's predictions → led to Einstein's general relativity
• Blackbody radiation didn't match classical predictions → led to quantum mechanics
• Neutrinos had a "solar neutrino problem" → led to the discovery that neutrinos have mass

The H₀ and σ₈ tensions might be the next crack that opens the door to entirely new physics.

The Big Idea: Can We Test a Bouncing Universe?

Here's the central proposal of this research, in plain language:

This isn't pure speculation. It's built on three well-established pillars of theoretical physics:

Pillar 1: Einstein-Cartan Theory KAHR-tahn

In standard general relativity, spacetime can curve but it can't twist. Einstein-Cartan theory adds torsion TOR-shun — the idea that spacetime can also twist and warp in a spiraling way, especially when matter with intrinsic spin (like electrons) is packed extremely tightly.

The key insight: at extreme densities (like inside a black hole), this torsion creates an enormous outward pressure that prevents a singularity from forming. Instead of crushing into an infinitely small point, the matter bounces back.

Pillar 2: Loop Quantum Gravity LQG

Loop Quantum Gravity is one of the leading approaches to unifying quantum mechanics with gravity. It says that space itself has a minimum size — you can't zoom in forever. At the tiniest scales (the Planck scale PLAHNK, about 10⁻³⁵ meters), space is made of discrete quantum "atoms."

This quantum structure of space gives rise to a special parameter called the Barbero-Immirzi parameter bar-BEAR-oh im-MEER-zee, written as γ gamma. This parameter connects to parity violation — the idea that certain processes treat left and right differently at the quantum level.

Pillar 3: The Holst Action HOHLST

When you write down the mathematics of Loop Quantum Gravity with fermions (matter particles like electrons and quarks), a special term appears called the Holst term. In empty space, this term doesn't do anything observable. But when matter with spin is present, it activates and generates parity-violating effects — processes that distinguish between clockwise and counterclockwise.

This is what connects quantum gravity to things we can actually observe in the sky.

How It Works: The Bounce and What Survives

Here's the chain of logic, step by step:

Step 1: A massive star in a parent universe collapses into a rotating black hole.

Step 2: As matter falls inward and density approaches the Planck scale, quantum gravity effects create enormous outward pressure. Instead of forming a singularity, the interior bounces — a "Big Bounce" — at a critical density of about 0.27 to 0.41 times the Planck density (depending on the entropy-counting scheme; roughly 10⁹³ kg/m³, incomprehensibly dense).

Step 3: This bounce creates a new expanding region — a baby universe. Because the parent black hole was rotating, the baby universe inherits angular momentum. It has a cosmic rotation axis.

Step 4 (what was hoped): The original idea was that quantum gravity effects from the Holst term would generate a parity-violating correction to the cosmological constant, naturally explaining dark energy. After systematic investigation, all four derivation routes are closed: the mass-coupling lock, topological-shift duality, environmental-mass genericity, and disformal-Planck suppression each independently block the mechanism.

Step 4 (what actually works): The same quantum gravity sector that fails to produce dark energy does naturally produce a spectator axion-like particle (ALP) with a Planck-scale decay constant. This ALP rotates CMB polarization by a predicted β = 0.27° — matching the 3.6σ cosmic birefringence signal observed by Planck + ACT.

Dark energy in this framework is modeled phenomenologically (assumed, not derived). The honest contribution is the bounce mechanism, the 14-barrier closure argument, and the ALP birefringence prediction.

The Cosmological Constant Problem (Honest Status)

The standard model has an embarrassing problem: when physicists calculate what dark energy "should" be based on quantum field theory, the answer is wrong by a factor of 10¹²⁰. That's a 1 followed by 120 zeros. It's been called the worst prediction in the history of physics.

Earlier versions of these papers claimed the spin-torsion framework reduced this mismatch to 10⁵ via "inflationary suppression." After rigorous review, this claim has been retracted. The mechanism reparameterizes the fine-tuning rather than solving it — the suppression factor must itself be tuned. The cosmological constant problem remains open. This is one of the 14 structural barriers documented in the papers.

The Two Surviving Predictions

A theory is only as good as what it can be tested against. After extensive MCMC verification and systematic auditing, here is the honest status of each signature:

Prediction 1: Matter-Bounce Non-Gaussianity (the decisive test)

The strongest prediction from bounce cosmology is a specific pattern in the statistics of the early universe called non-Gaussianity. In simple terms: if the universe bounced, the tiny fluctuations that seeded galaxies would have a particular "skewness" to their distribution.

The prediction: f_NL = −35/8 = −4.375. This number comes directly from the physics of matter-dominated contraction—no free parameters, no tuning. Standard inflation predicts f_NL ≈ +0.015, which is 300 times smaller and the wrong sign.

The SPHEREx satellite (launched 2025, data expected ~2028) will measure this at ~5.0–5.5σ significance (template-corrected; range reflects ε uncertainty from the slow-roll correction) through the galaxy bispectrum. If detected, Bayesian model comparison shows the bounce would be favored over tuned multifield competitors at odds of ~8-17:1 (prior-dependent). If not detected, the quasi-dust matter bounce is ruled out at >4σ. Either way, we get a definitive answer.

Why the Detection Is Slightly Harder Than It Looks

When scientists search for non-Gaussianity in galaxy surveys, they use a mathematical template — like a search pattern — to find the signal. The standard template is called the "local" template, and it's what SPHEREx and other surveys will use.

We discovered that the matter-bounce signal doesn't perfectly match this template. The bounce prediction is −4.375 in the squeezed configuration (when one triangle side is very short), but only −2.25 in the folded configuration (when one side equals the other two). The local template treats all configurations equally at −4.375.

This means a standard search will only capture about 85–90% of the actual bounce signal. Using the physics-derived full-commutator polynomial (6,2,−18,10,−66,18)—derived algebraically from Cai's own intermediate equations—the template overlap is r ≈ 0.85–0.90 (CMB Fisher near 0.90, LSS/SDB nearer 0.85), validated by ℓ-space Fisher (0.878) and injection recovery (0.90 ± 0.01). Instead of a 6.2σ detection with SPHEREx, the realistic significance is about 5.0–5.5σ — still very strong, but the correction matters for planning.

We also resolved a factor-of-two disagreement between two research groups (Cai et al. and Li et al.) about the prediction's exact value. By checking their calculations line-by-line, we confirmed they computed the same physics but used different mathematical conventions. The correct value in the standard convention is −35/8 = −4.375.

Prediction 2: Cosmic Birefringence

The Cosmic Microwave Background CMB is the oldest light in the universe, released about 380,000 years after the Big Bang. This light is slightly polarized (its electromagnetic waves oscillate in preferred directions).

The framework's spectator ALP (axion-like particle), with a Planck-scale decay constant, predicts a specific rotation of CMB polarization called cosmic birefringence: β = 0.27°. This is a genuine prediction, not a fit.

In 2020, Minami and Komatsu found a hint of exactly this kind of twist in Planck satellite data, at 2.4 sigma significance, later refined to 2.7σ. ACT DR6 independently confirmed the signal at 2.9σ (Diego-Palazuelos & Komatsu 2025). The combined Planck + ACT measurement gives β = 0.34° ± 0.09° at 3.6σ significance — consistent with the framework's prediction. Our independent NaMaster EB analysis at NSIDE=1024 gives β = 0.19° ± 0.03° (lead result, without miscalibration marginalization). At higher resolution (NSIDE=2048), β drops to 0.07° ± 0.02°, suggesting high-ℓ contamination/noise—NSIDE=1024 remains the headline result. NaMaster injection: ALL levels pass including β = 0.5° and β = 0.27° with zero bias. The upcoming LiteBIRD satellite (JAXA JFY2032, early 2030s) will test the birefringence definitively.

Historical: Galaxy Spin Patterns (not a prediction of this framework)

If our universe was born from a rotating black hole, galaxies might show a slight preference for spinning one way versus the other. However, this signature has serious problems:

• The predicted coupling between cosmic rotation and galaxy angular momentum has a 9–12 order-of-magnitude gap that cannot be bridged without fine-tuning
• The asymmetry amplitude (A₀ ≈ 0.3%) is an empirical fit parameter, not derived from first principles
• The observational signal is contested — multiple independent reanalyses have reported null results
• CMB isotropy bounds constrain cosmic rotation to (ω/H)₀ < 5 × 10⁻¹¹, far too small to produce observable galaxy spin asymmetry through any known coupling

This signature should be considered speculative and unsubstantiated at present.

Historical: Cosmological Tensions (not resolved by this framework)

Earlier versions of these papers claimed the framework could reduce the Hubble tension from ~4.9σ to 2.9σ. After careful MCMC verification with 424,181 samples across 3 dataset combinations (2 frozen + 1 exploratory), the honest result is:

The spin-torsion extension alone does not resolve cosmological tensions. The MCMC data prefer standard ΛCDM values. The earlier H₀ = 69.2 claim was based on a prior that assumed ΔN_eff > 0; when the data are allowed to speak, they pull it back to zero. This is an important null result.

When Will We Know If It's Right?

The framework's testable prediction is now focused on cosmic birefringence from the spectator ALP. Here's when we'll get decisive data:

In physics, 5 sigma five SIG-muh is the gold standard for discovery — it means there's less than a 1-in-3.5-million chance the result is a statistical fluke. LiteBIRD should reach that threshold for birefringence by the early 2030s.

How This Compares to Other Theories

There are many proposals to explain dark energy. Here's how the spin-torsion framework honestly stacks up:

The unique surviving feature of this framework is cosmic birefringence from a Planck-scale ALP. The 14-barrier closure argument is itself a scientific contribution — it maps out what doesn't work and why, saving future researchers from repeating the same dead ends.

AI Discovery Pipeline

Beyond the theoretical predictions, this research program now includes a large-scale observational discovery effort. We built an AI system (a "spectral autoencoder") that learns what normal astronomical spectra look like, then flags anything unusual. We pointed it at the entire DESI DR1 survey — nearly 18 million spectra, the largest dataset of its kind.

What We Found

The AI identified 195,829 spectral anomalies — objects whose light signatures don't match anything the model learned as "normal." This is the largest autoencoder anomaly search ever run on DESI data, about 90 times bigger than previous efforts that only used early data releases.

The most striking result: galaxies are 20 times more anomalous than quasars. About 0.75% of galaxies trigger the anomaly detector, versus only 0.037% of quasars (the extremely bright galactic nuclei). This was unexpected — quasars are often considered the "weird" objects in astronomy, but it turns out galaxies harbor far more uncharacterized spectral behavior at scale.

Are These Real?

We cross-matched every anomaly against 6 major astronomical databases: SIMBAD, NED, AllWISE, Milliquas, Gaia, and SDSS. The result: 99.8% of our anomalies are not cataloged in SIMBAD, the largest astronomical reference database. NED (the NASA/IPAC Extragalactic Database) recognizes 12.7% of them, AllWISE has 1.5%, and Gaia has 0.6%. The vast majority are genuinely uncataloged objects.

This doesn't mean they're all astrophysically exotic — some may be instrument artifacts or unusual but understood phenomena. That's why the catalog is still in draft stage: we need to visually inspect the top candidates, run injection/recovery tests, and systematically investigate artifacts before making strong claims.

The Enhanced 18M Catalog

Alongside the anomaly search, we're building an enhanced catalog of all 18 million DESI DR1 spectra with 45 computed columns per object — anomaly scores, spectral feature measurements, cross-match results, and quality flags. This is the first catalog of its kind: a uniform machine-learning characterization of every spectrum in DESI DR1. It's 36% complete and running on an H200 GPU.

The w₀-w_a Dark Energy Result

On the cosmology side, we ran a new MCMC analysis testing whether dark energy's equation of state changes over time. The result: w₀ = −0.871 ± 0.061 and w_a = −0.542 ± 0.247, with 98% probability of "quintom-B" behavior — meaning dark energy's equation of state crosses the w = −1 boundary from below. This is consistent with DESI DR2 findings and supports the quintom bounce scenario as a viable cosmological model. The chains fully converged (R−1 = 0.009).

The Key Equations, Explained Simply

The Modified Friedmann Equation

The Friedmann equation is the master equation of cosmology — it describes how the universe expands. In our framework, it gets modified near the bounce:

When density approaches the critical value (ρ_crit ≈ 0.27–0.41 times the Planck density), the term in brackets goes to zero. The expansion rate hits zero — that's the bounce. Then it flips sign and expansion begins. No singularity ever forms. This is the best-established part of the framework.

The ALP Birefringence Angle

The spectator ALP, with a Planck-scale decay constant f_a ≈ M_Pl, rolls a small amount between recombination and today. This produces a uniform rotation of CMB polarization — the birefringence angle β. For the natural parameter values, this gives β ≈ 0.27°, matching observations.

How Could This Be Proven Wrong?

Good science defines how it could be falsified. Here's what would rule out the surviving predictions:

• If LiteBIRD measures β = 0 (no cosmic birefringence) at high significance — this would kill the ALP prediction
• If the birefringence is confirmed but at a very different angle (e.g., β > 1° or β < 0.05°) — would disfavor the Planck-scale decay constant
• If SPHEREx measures f_NL consistent with zero (±1) — would strongly disfavor the matter bounce (which predicts −4.375)
• The 14-barrier closure for geometric dark energy is already a form of self-falsification — the framework tested and rejected its own original claim

The framework is designed to be testable, not unfalsifiable. The birefringence prediction will be decisively tested within the next decade.

Quick Pronunciation Guide

The Bottom Line

This research program asked whether bounce cosmology can be tested. The answer is yes—through two specific, falsifiable predictions:

1. f_NL = −35/8 — a parameter-free non-Gaussianity prediction from matter-dominated contraction, testable by SPHEREx at ~5.0–5.5σ (template-corrected, ~2028)
2. β = 0.27° — a cosmic birefringence prediction from a natural Planck-scale axion, already matching a 3.6σ observed signal, testable by LiteBIRD at 9σ (early 2030s)

The journey to these predictions involved building and systematically dismantling an ambitious theoretical framework (Einstein-Cartan-Holst gravity), cataloging 14 structural barriers, and proving a perturbation-transparency theorem. Most of what we tried didn't work—and we documented every failure honestly. But the predictions that survived are mechanism-independent: they don't depend on the specific bounce model, only on the contracting-phase dynamics.

We'll know within the next few years.