We compare the results of various cosmological gas-dynamical codes used to simulate the formation of a galaxy in the Λ cold dark matter structure formation paradigm. The various runs (13 in total) differ in their numerical hydrodynamical treatment [smoothed particle hydrodynamics (SPH), moving mesh and adaptive mesh refinement] but share the same initial conditions and adopt in each case their latest published model of gas cooling, star formation and feedback. Despite the common halo assembly history, we find large code-to-code variations in the stellar mass, size, morphology and gas content of the galaxy at z= 0, due mainly to the different implementations of star formation and feedback. Compared with observation, most codes tend to produce an overly massive galaxy, smaller and less gas rich than typical spirals, with a massive bulge and a declining rotation curve. A stellar disc is discernible in most simulations, although its prominence varies widely from code to code. There is a well-defined trend between the effects of feedback and the severity of the disagreement with observed spirals. In general, models that are more effective at limiting the baryonic mass of the galaxy come closer to matching observed galaxy scaling laws, but often to the detriment of the disc component. Although numerical convergence is not particularly good for any of the codes, our conclusions hold at two different numerical resolutions. Some differences can also be traced to the different numerical techniques; for example, more gas seems able to cool and become available for star formation in grid-based codes than in SPH. However, this effect is small compared to the variations induced by different feedback prescriptions. We conclude that state-of-the-art simulations cannot yet uniquely predict the properties of the baryonic component of a galaxy, even when the assembly history of its host halo is fully specified. Developing feedback algorithms that can effectively regulate the mass of a galaxy without hindering the formation of high angular momentum stellar discs remains a challenge

Collisionless simulations of the cold dark matter (CDM) cosmology predict a plethora of dark matter substructures in the haloes of Milky Way sized galaxies, yet the number of known luminous satellites galaxies is very much smaller, a discrepancy that has become known as the 'missing satellite problem'. The most massive substructures have been shown to be plausibly the hosts of the brightest satellites, but it remains unclear which processes prevent star formation in the many other, purely dark substructures. We use high-resolution hydrodynamic simulations of the formation of Milky Way sized galaxies in order to test how well such self-consistent models of structure formation match the observed properties of the Galaxy's satellite population. For the first time, we include in such calculations feedback from cosmic rays injected into the star-forming gas by supernovae as well as the energy input from supermassive black holes growing at the Milky Way's centre and its progenitor systems. We find that non-thermal particle populations quite strongly suppress the star formation efficiency of the smallest galaxies. In fact, our cosmic ray model is able to reproduce the observed faintend of the satellite luminosity function, while models that include only the effects of cosmic reionization, or galactic winds, do significantly worse. Our simulated satellite population approximately matches available kinematic data on the satellites and their observed spatial distribution. We conclude that a proper resolution of the missing satellite problem likely requires the inclusion of non-standard physics for regulating star formation in the smallest haloes, and that cosmic reionization alone may not be sufficient.