Browsing by Author "Järvenpää, Anni"

During the nearly a century the approximate structure of the Local Group has been known, many methods for estimating the masses of both the individual galaxies within it and the group as a whole have been constructed. Still, to this day, estimates from different credible sources vary by a factor of 2–3. In this master's thesis I construct a new model for estimating the combined mass of the Milky Way and Andromeda galaxies based on the kinematics of the galaxy pair and properties of the Hubble flow surrounding it, aiming to improve on the accuracy of the timing argument. The model is based on a sample of subhalo catalogues from cosmological dark matter only simulations containing a system resembling the Local Group. From these catalogues, I identified the Local Group analogues based on the presence of a pair of dark matter haloes with mutual kinematics and masses resembling those of the Milky Way and Andromeda galaxies in the Local Group, with no other massive objects in the vicinity. For the Hubble flow fitting I used an algorithm for automatically choosing the best range of objects to include in the fit. The surrounding Hubble flows showed clear signs of anisotropy and existence of substructures within the flow. In order to capture the properties of these structures, I clustered the subhaloes within 1.5 to 5.0 Mpc from the Milky Way analogue in each simulation using the DBSCAN clustering algorithm. I then fitted separate Hubble flows for subhaloes outside clusters, within each cluster and within each cluster with all members less massive than 8*10^11 solar masses in each simulation. I used twelve variables to construct the model for predicting the Local Group mass. Nine of these were the Hubble constants, Hubble flow zero point distances and velocity dispersions around the fit measured separately for all haloes, clustered haloes and haloes outside clusters. The remaining three consisted of the radial and tangential velocity components and the distance of the Andromeda Galaxy analogue as seen from the Milky Way analogue. I split the data set consisting of 119 subhalo catalogues into a training set with approximately 60% of the whole data set (71 subhalo catalogues) and a test set containing the remaining subhalo catalogues. I then extracted the principal components of the training set and selected the two first to be used in predicting the mass. A linear regression model was fitted to these components using 10-fold cross-validation. The error of the resulting model was estimated by applying the model on the test set and comparing its predictions to the known masses of the subhalo pair. The obtained root-mean-square error was 1.06*10^12 solar masses. This is a clear improvement over the timing argument, which had a root-mean-square error of 1.42*10^12 solar masses in the same data set.