Tibbs Morgan, Harry David (2026) Classifying DE vs DM birefringence regimes on CMB maps using CNNs. Bachelor's Thesis, Physics.
|
Text
BachelorThesis-FINAL.pdf Download (12MB) | Preview |
Abstract
Cosmic birefringence—the rotation of the linear polarization of cosmic microwave background photons—could reveal whether an axion-like scalar field behaves like dark energy or dark matter. This thesis investigates whether convolutional neural networks (CNNs) can distinguish these regimes directly from simulated polarization maps, and if this outperforms simpler power-spectrum-based methods. A complete pipeline was developed, generating polarization spectra using CAMB-based prescriptions and birefCLASS scalar-field evolution. These were converted into flat-sky map representations (e.g., Stokes, E/B-modes, Fourier inputs) and evaluated against a spectral baseline under rigorous testing, including simplified foreground contamination. In the weak CAMB benchmark, the spectral baseline outperformed the CNNs. In the birefCLASS benchmark, all classifiers perfectly separated the selected scalar-field masses, even when controlling for overall birefringence amplitude and foregrounds. However, map-level networks never exceeded the spectral baseline's performance. While these regimes are highly distinguishable in idealized simulations, map-level learning does not yet demonstrate a clear advantage over spectral classification. This validated pipeline provides a strong foundation for future research incorporating more closely spaced masses, realistic observational effects, and complex foregrounds.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Meerburg, P.D. |
| Degree programme: | Physics |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 16 Jul 2026 14:24 |
| Last Modified: | 16 Jul 2026 14:24 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/37438 |
Actions (login required)
 |
View Item |