- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- E-B decomposition of CMB polarization using deep learning
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
E-B decomposition of CMB polarization using deep learning Gulati, Puranjay Rohan
Abstract
The polarization of the cosmic microwave background (CMB) can be split into two coordinate independent components: the gradient-like E-mode, and the curl-like B-mode. Primordial B-modes are of particular interest as they do not arise from scalar density perturbations and serve as a direct probe of primordial gravitational waves theorized to be generated during inflation. A direct detection of these elusive B-modes would help determine the energy scale of inflation and constrain possible inflationary models. However, the E-B decomposition of polarization on a partial or cut sky is non-unique and introduces modes that cannot conclusively be classified as pure E or B. This is a source of ambiguity for all CMB experiments probing polarization as a Galactic cut is required even for purported full-sky missions, such as the European Space Agency’s (ESA) Planck mission. We present a map-based technique using machine learning (ML) methods to perform this partial sky decomposition. In particular, we use a deep residual convolutional neural network (CNN) based on the U-Net architecture to perform this decomposition on small patches of the sky 400 square degrees in area, allowing the targeting of regions with low Galactic foreground. Our deep residual U-Net performs exceptionally well at angular scales of a few degrees, corresponding to spherical harmonic Fourier modes l (ell) ≃ 50 to 110, a range of scales ideal for probing the recombination bump of the primordial B-mode power spectrum. Additionally, the map-based nature of our technique allows visual inspection of the E-B separation. This may be especially useful for identifying residual Galactic contamination in polarization data.
Item Metadata
Title |
E-B decomposition of CMB polarization using deep learning
|
Creator | |
Supervisor | |
Publisher |
University of British Columbia
|
Date Issued |
2022
|
Description |
The polarization of the cosmic microwave background (CMB) can be split into two coordinate independent components: the gradient-like E-mode, and the curl-like B-mode. Primordial B-modes are of particular interest as they do not arise from scalar density perturbations and serve as a direct probe of primordial gravitational waves theorized to be generated during inflation. A direct detection of these elusive B-modes would help determine the energy scale of inflation and constrain possible inflationary models.
However, the E-B decomposition of polarization on a partial or cut sky is non-unique and introduces modes that cannot conclusively be classified as pure E or B. This is a source of ambiguity for all CMB experiments probing polarization as a Galactic cut is required even for purported full-sky missions, such as the European Space Agency’s (ESA) Planck mission. We present a map-based technique using machine learning (ML) methods to perform this partial sky decomposition.
In particular, we use a deep residual convolutional neural network (CNN) based on the U-Net architecture to perform this decomposition on small patches of the sky 400 square degrees in area, allowing the targeting of regions with low Galactic foreground. Our deep residual U-Net performs exceptionally well at angular scales of a few degrees, corresponding to spherical harmonic Fourier modes l (ell) ≃ 50 to 110, a range of scales ideal for probing the recombination bump of the primordial B-mode power spectrum. Additionally, the map-based nature of our technique allows visual inspection of the E-B separation. This may be especially useful for identifying residual Galactic contamination in polarization data.
|
Genre | |
Type | |
Language |
eng
|
Date Available |
2022-07-20
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
Attribution-NonCommercial-NoDerivatives 4.0 International
|
DOI |
10.14288/1.0416316
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
2022-11
|
Campus | |
Scholarly Level |
Graduate
|
Rights URI | |
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International