AbstractThis thesis explores a forward modelling approach for simulating galaxy catalogues solely based on the galaxy luminosity function. The approach developed by Herbel et al. (2017) enables accurate simulations of observed galaxy redshift distributions, which is very useful for constraining the cosmological model with shear measurements without relying on individual redshift measurements. The parameterisation of the luminosity function includes a linear and an exponential model with redshift for the characteristic magnitude and the amplitude, respectively. Apparent magnitudes are simulated by defining the flux density as a linear combination of the K-correct templates with Dirichlet distributed coefficients.
The sensitivity of the simulated redshift distribution to the luminosity function parameters is investigated in the first study. Almost all parameters showed strong impact on the redshift distribution. The slope of the characteristic magnitude affects the redshift distribution the most and needs tight constraints to use the forward model for simulating the redshift distribution precisely. The study also indicated high sensitivity of the simulated colour distributions to the model parameters. These distributions can therefore be used for constraining the model parameters.
A second study tested Sequential Neural Posterior Estimation for measuring the luminosity function against default parameter values and applied the method to COSMOS2020 data. The tests showed excellent performance and the fitted model parameters matched the default values with high accuracy. We also recovered the luminosity function of the data with the simulation-based inference approach and obtained more precise constraints compared to our default parameter set. However, the fits still need improvement and the results suggest to modify the model in upcoming studies.
The last part of the thesis focused on my contributions to the open-source Python package SkyPy. First, I derive and explain the implementation of the luminosity and redshift sampling according to the model of the luminosity function. Afterwards, I expand this approach to stellar masses of star-forming and quenched galaxies. It is shown that stellar masses can be modelled in a similar way to luminosities. Furthermore, by exploiting and improving the continuity equations of galaxy evolution, the parameter space was optimised. Comparisons with observations are showing good agreement and the correctness of the model and implementation.
In the last section, the forward model is compared to COSMOS2020 data. Even though, the simulations recover the data well, some details of the data could not be accurately simulated. The comparison suggests including instrumental selection effects such as signal-to-noise cuts and pixel saturation in the model. Additionally, higher-order redshift terms for the characteristic magnitude should be considered. The comparison further points to an inaccurate model of the template coefficients in the flux density model. Dirichlet distributed coefficients cannot be assumed. The primary conclusion of the comparisons is that a new model for the coefficients has to be found in upcoming work.
|Date of Award||3 Oct 2023|
|Supervisor||Adam Amara (Supervisor), David Bacon (Supervisor) & Andrew Lundgren (Supervisor)|