Galaxy mergers on a moving mesh: a comparison with smoothed-particle hydrodynamics

Galaxy mergers have been investigated for decades using smoothed particle hydrodynamics (SPH), but recent work highlighting inaccuracies inherent in the traditional SPH technique calls into question the reliability of previous studies. We explore this issue by comparing a suite of Gadget-3 SPH simulations of idealised (i.e., non-cosmological) isolated discs and galaxy mergers with otherwise identical calculations performed using the moving-mesh code Arepo. When black hole (BH) accretion and active galactic nucleus (AGN) feedback are not included, the star formation histories (SFHs) obtained from the two codes agree well. When BHs are included, the code- and resolution-dependent variations in the SFHs are more significant, but the agreement is still good, and the stellar mass formed over the course of a simulation is robust to variations in the numerical method. During a merger, the gas morphology and phase structure are initially similar prior to the starburst phase. However, once a hot gaseous halo has formed from shock heating and AGN feedback (when included), the agreement is less good. In particular, during the post-starburst phase, the SPH simulations feature more prominent hot gaseous haloes and spurious clumps, whereas with Arepo, gas clumps and filaments are less apparent and the hot halo gas can cool more efficiently. We discuss the origin of these differences and explain why the SPH technique yields trustworthy results for some applications (such as the idealised isolated disc and galaxy merger simulations presented here) but not others (e.g., gas flows onto galaxies in cosmological hydrodynamical simulations).