Systematics in Weak Lensing and Galaxy Clustering

With a new generation of surveys on the horizon (e.g. Euclid, LSST, DESI) we are entering a new era of high-precision cosmology. The increase in precision and the scale of data collections compared to previous surveys gives us a new insight into cosmological information. The large amount of the data makes fast and computationally cheap analysis difficult, but if this can be done very precise and robust constraints can be achieved. Through matter probes such as cosmic shear analysis and galaxy clustering we can trace the matter distribution in our universe and use this information to constrain cosmology. These methods can be combined in multi-probe methods such as thein 3x2-point correlation function analysis, which uses both shear-shear correlation, clustering, and their cross-term, to allow more precise parameter constraints. However, are our statistical uncertainties are reduced, we must look closer at the systematic uncertainties induced due to assumptions, degeneracies, and generalisations in our analysis. It's possible that some systematic biases previously concealed by statistical uncertainty are large enough to severely limit our constraining power. On top of this, interactions between systematic effects within and between probes may cause further limitations, such as that between intrinsic alignments and photometric redshift uncertainty. Therefore, an important step in preparation for these surveys is to understand and mitigate the limitations systematics may impose upon them.

My PhD project revolves around optimising parameter constraints from weak lensing and galaxy clustering analysis, with a focus on the systematic effects and combinations of systematics that may affect the constraining power of upcoming surveys. I am pursuing my PhD as a part of the CDT, and a component of this is taking part in modules teaching data science methods. I will use the techniques I learn in these modules to understand problematic systematic effects and improve the errors they induce. I aim to use machine learning and deep learning techniques to remove systematic effects and decouple systematics. More broadly, I hope to apply these methods to the parameter constraints from the 3x2 correlation function with the aim to improve the speed and constraining power.