Weak lensing cosmology and its astrophysical systematics through machine learning - Doctor of Philosophy dissertation

Aug 5, 2026

Upload by Finn

This dissertation studies weak lensing cosmology and astrophysical systematics using machine learning. It targets the discrepancy between two prior weak-lensing analyses on CFHTLenS data, quantifies how baryons alter weak-lensing statistics, and employs convolutional neural networks (CNNs) to jointly constrain cosmological and baryonic parameters. The work compares two-point correlation functions with power spectrum analyses, examines small-scale excess power, and models baryonic corrections. It shows that marginalizing baryonic parameters can weaken Ωm–σ8 constraints, while combining lensing power spectrum and peak counts mitigates the degradation. CNNs improve constraint tightness in Ωm–σ8 on simulations and reduce statistical uncertainties in the HSC first-year shear catalog, extracting additional information beyond traditional summary statistics even in real experiments.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages176

Reading Time7 h 24 min

AccessFree to download

Chapter 1: Introduction

Weak lensing basics

Weak lensing sky surveys

Systematic effects in weak lensing signals

Constraining cosmology from weak lensing

The application of deep learning

Chapter 2: The matter fluctuation amplitude inferred from the weak lensing power spectrum and correlation function in CFHTLenS data

Lensing Data

Methods

Results

Discussion

Conclusion

Chapter 3: The Impact of Baryons on Cosmological Inference from Weak Lensing Statistics

Methods

Results

Discussion

Conclusions

Chapter 4: Simultaneously constraining cosmology and baryonic physics via deep learning from weak lensing

Methods

Weak lensing cosmology and its astrophysical systematics through machine learning Tianhuan Lu Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy under the Executive Committee of the Graduate School of Arts and Sciences COLUMBIA UNIVERSITY © 2023 Tianhuan Lu All Rights Reserved Abstract Weak lensing cosmology and its astrophysical systematics through machine learning Tianhuan Lu In this dissertation, we investigate weak lensing cosmology and its astrophysical systematics by employing machine learning techniques. We focus on addressing the discrepancy between two previous weak lensing analyses on CFHTLenS data, understanding the impact of baryons on weak lensing statistics, and leveraging convolutional neural networks (CNNs) for constraining cosmological and baryonic parameters. First, we perform a side-by-side comparison of the two-point correlation function and power spectrum analyses on CFHTLenS data, identifying excess power in the data on small scales and discussing potential origins of this excess power. Next, we study the effect of baryons on weak lensing statistics using the baryonic correction model, demonstrating that marginalizing over baryonic parameters will degrade constraints in the Ωm–􀁦8 parameter space, but the degradation can be mitigated by combining the lensing power spectrum and peak counts. Second, we explore the use of CNNs to constrain cosmological and baryonic parameters. We find that CNNs can achieve tighter constraints in Ωm–􀁦8 space than traditional methods on simulation data. We then apply our pipeline to the HSC first-year weak lensing shear catalog. We find that statistical uncertainties of the parameters by the CNNs are smaller than those from the power spectrum and peak counts, showing that CNNs can extract additional cosmological information from weak lensing data even in a real experiment. Table of Contents Acknowledgments ........................................ xv Dedication ............................................ xvi Chapter 1: Introduction .................................... 1 1.1 Weak lensing basics .................................. 1 1.2 Weak lensing sky surveys ............................... 3 1.3 Systematic effects in weak lensing signals ...................... 3 1.3.1 Point spread function ............................. 4 1.3.2 Photometric redshift ............................. 5 1.3.3 Intrinsic alignments .............................. 6 1.3.4 Baryonic effects ................................ 7 1.4 Constraining cosmology from weak lensing ..................... 8 1.4.1 Commonly used summary statistics ..................... 8 1.4.2 Cosmological inference ............................ 10 1.5 The application of deep learning ........................... 11 1.5.1 Deep learning basics ............................. 11 1.5.2 Deep learning as a summary statistic ..................... 14 1.6 Structure of dissertation ................................ 14 Chapter 2: The matter fluctuation amplitude inferred from the weak lensing power spectrum and correlation function in CFHTLenS data ................. 17 2.1 Introduction ...................................... 17 2.2 Lensing Data ..................................... 19 2.2.1 CFHTLenS shear catalogue .......................... 19 2.2.2 Convergence maps .............................. 19 2.2.3 Mock catalogues and maps .......................... 23 2.3 Methods ........................................ 25 2.3.1 Statistics .................................... 25 2.3.2 Parameter estimation ............................. 27 2.4 Results ......................................... 31 2.4.1 Inferences from the 2PCF vs. the power spectrum .............. 31 2.4.2 Systematic biases ............................... 32 2.4.3 Statistical fluke ................................ 33 2.4.4 Shapes of 2PCFs and power spectra ............

FAQs

How do CNNs improve cosmological constraints compared with traditional methods?

The study finds that CNNs yield tighter constraints in Ωm–σ8 on simulation data than power-spectrum-based and peak-count-based approaches, and that the CNN pipeline produces smaller statistical uncertainties when applied to the HSC first-year shear catalog.

Aug 5, 2026

Upload by Finn