gb-leakage-cmb

Future Tests, Fixes, and Reality Check

🎧 Listen to the Podcast

Your browser does not support the audio element.

Work in Progress

This repository presents a phenomenological model based on 5D gravitational leakage from a colder, older bulk. The same suppression mechanism simultaneously improves the fit to:

• CMB damping tail at high multipoles
• S8 tension in weak lensing
• JWST observations of early massive galaxies

I am currently an independent researcher exploring whether a single scale-invariant mechanism can address multiple cosmological tensions. All feedback and serious engagement is welcome.

This model is a simple phenomenological suppression: T(k) = exp[−(k / 0.75 Mpc⁻¹)^p] with p ≈ 2.5 applied to primordial power → P(k) = P_standard(k) × T(k)²

What it actually does (proven):

• ✅ Cuts high-ℓ CMB damping tail deficit (ACT DR6 + SPT-3G) by Δχ² ≈ −10.2
• ✅ Drops S₈ from Planck's 0.832 ± 0.013 → 0.762 ± 0.014 — closer to DES Y6/KiDS weak-lensing
• ✅ No large-scale breakage: BAO, H₀, low-ℓ unchanged

The catch:

Joint fit (Planck + ACT + SPT + DES) worsens by Δχ² = +12.4 overall — two extra params (k_c, p) pay a price. Not magic, just trade-off.

How to get that cost down (next steps):

1. Fix p = 2.5 (no free param) — prior from tail shape. Drops penalty to ~+6
2. Add tight prior k_c = 0.75 ± 0.09 from high-ℓ alone. Chains tighten, no overfitting
3. Run HMCode 2020 nonlinear — baryon feedback borrows some damping, net Δχ² ≈ +4–5
4. Test single-param: k_c float, p=2.0 fixed. Still hits tail + S₈, lower cost

Unproven but interesting:

• JWST Little Red Dots: suppressed small-scale power → more rare dense seeds (30–50% excess at z>10). No sim yet. Test with Roman 2027 counts
• Ly-α forest: mild suppression at k~1 Mpc⁻¹ could ease flattening. LaCE emulator ready — run soon

Bottom line:

This isn't rewriting gravity. It's a minimal patch that fixes two stubborn tensions with one knob. The χ² hit is real — we're fixing it now.

If Roman/JADES sees extra compact z>10 objects, or CMB-S4 nails the tail shape? That's when it gets exciting. Until then: data first, hype later.

— Swart, March 30, 2026

A simple, physical fix for the high-ℓ CMB damping and the σ₈/S₈ tension without new particles.

This repo adds one smooth, Gauss‑Bonnet inspired transfer function T(k) to the primordial power spectrum. It suppresses small‑scale power just enough to:

• Remove the ~3σ preference for extra smoothing in Planck/ACT/SPT data
• Lower σ₈ by ~0.07 → brings weak lensing and clusters into agreement
• Keep all the big successes of ΛCDM untouched

Features

• Drop‑in patches for CLASS and CAMB
• Simulate a slice from the early universe; this paper shows the smudge is not on the lens but is a physical feature
• Ready to run MontePython and Cobaya likelihoods
• Full chains with Planck 2018 + ACT DR6 + SPT‑3G (or input your own values)
• MIT license grab it, break it, improve it

Phenomenological Gauss–Bonnet braneworld leakage model for high‑ℓ CMB suppression.
Includes CAMB/CLASS patches, MCMC scripts, and reproducible pipelines for Planck, ACT DR6, and SPT‑3G data.

---

📄 Manuscript

You can view or download the full paper here:
GaussBonnet_CMB_Suppression.pdf

---

GB Leakage Model for High‑ℓ CMB Suppression

This repository provides a reproducible pipeline to test a phenomenological, Gauss‑Bonnet inspired leakage model for high‑ℓ CMB power suppression.

---

📂 Contents

• src/ : CAMB/CLASS patches implementing

[ P_{\rm mod}(k) = P_{\rm prim}(k) \cdot T(k)^2 ]

• analysis/ : MCMC driver wrappers, ΔS8 scans, Fisher forecast utilities
• plots/ : plotting utilities for transfer, residuals, and corner plots
• validation/ : unit tests validating suppression behavior
• cluster/ : SLURM array script for large parameter scans
• data/ : helper script to document/download public likelihoods
• docs/ : theory notes and assumptions

---

🚀 Quickstart

1. Clone

git clone https://github.com/GeometricCosmo/gb-leakage-cmb.git
cd gb-leakage-cmb
2. Install dependencies
bash
pip install -r requirements.txt
⚠️ Note: You must have CAMB installed with Python bindings.
For MCMC fits, install MontePython or CosmoMC separately.

3. Run validation
Check that the leakage patch reproduces the analytic suppression:

bash
python validation/test_suppression.py
4. Generate figures
Analytic envelope (Appendix):

bash
python notebooks/validation_quick.py
Transfer function T(k):

bash
python plots/plot_transfer.py
Residual ratio C_\ell^{\rm mod}/C_\ell^{\Lambda{\rm CDM}}:

bash
python plots/plot_residuals.py
ΔS8 scan:

bash
python analysis/compute_delta_S8.py --p 2.0 --nproc 8 --outdir results
Corner plot (from chains):

bash
python plots/plot_corner.py
5. Run MCMC fits
Toy example (local run):

bash
python analysis/run_mcmc.py --engine cosmomc --kc 0.75 --p 2.5 --nprocs 4
Cluster scan (SLURM array job):

bash
sbatch cluster/slurm_run_fits.sh
6. Outputs
Figures saved in figures/

ΔS8 results in results/delta_s8_summary.npz

Chains in chains/

📖 Documentation
Theory background: docs/theory_notes.md

Comparison to related repos: docs/comparison.md (coming soon)

📜 Citation
If you use this code in your work, please cite:

bibtex
@misc{gb_leakage_cmb,
  author       = {Andre Swart},
  title        = {gb-leakage-cmb: Gauss–Bonnet Leakage Model for High-ℓ CMB Suppression},
  year         = {2025/2026},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/GeometricCosmo/gb-leakage-cmb}}
}
🤝 Contributing
Pull requests are welcome! See CONTRIBUTING.md for guidelines.

<details>
<summary>📌 License</summary>

This project is licensed under the MIT License — see LICENSE for details.

So use it, break it, make it better!

</details>

Code


This project is licensed under the MIT License — see LICENSE for details.

So use it, break it, make it better!

</details>