Estimating the mass of the Local Group using machine learning applied to numerical simulations

摘要

We present a new approach to calculating the combined mass of the Milky Way (MW) and Andromeda (M31), which together account for the bulk of the mass of the Local Group (LG). We base our work on an ensemble of 30,190 halo pairs from the Small MultiDark simulation, assuming a ΛCDM (Cosmological Constant and Cold Dark Matter) cosmology. This is used in conjunction with machine learning methods (artificial neural networks, ANN) to investigate the relationship between the mass and selected parameters characterising the orbit and local environment of the binary. ANN are employed to take account of additional physics arising from interactions with larger structures or dynamical effects which are not analytically well understood. Results from the ANN are most successful when the velocity shear is provided, which demonstrates the flexibility of machine learning to model physical phenomena and readily incorporate new information. The resulting estimate for the Local Group mass, when shear information is included, is 4.9 × 10~(12)M_⊙, with an error of ±0.8 × 10~(12)M_⊙ from the 68% uncertainty in observables, and a r.m.s. scatter interval of _(-1.3)~(+1.7) × 10~(12)M_⊙ estimated scatter from the differences between the model estimates and simulation masses for a testing sample of halo pairs. We also consider a recently reported large relative transverse velocity of M31 and the Milky Way, and produce an alternative mass estimate of 3.6 ± 0.3 _(-1.3)~(+2.1) × 10 ~(12)M_⊙. Although the methods used predict similar values for the most likely mass of the LG, application of ANN compared to the traditional Timing Argument reduces the scatter in the log mass by approximately half when tested on samples from the simulation.