Mass-morphology relation of TNG50 galaxies

We used the cosmological hydrodynamical simulation TNG50 to study the galaxy mass-morphology relation, as measured by the rotational support of the stellar component of simulated galaxies. For isolated galaxies with a stellar mass in the range of 8 < log( M * / M ⊙ ) < 11, rotational support increases with M * , from dispersion-supported spheroidal dwarfs to massive galaxies with prominent, rotationally supported discs. Our results indicate that this correlation arises from the spatial distribution of star formation in TNG50 galaxies, which occurs primarily in two distinct regions: an unresolved, non-rotating central baryonic clump ( r ≲ 1 kpc) and a rotationally supported outer disc, separated by a quiescent region. The importance of the inner clump increases with decreasing M * ; it makes up less than 20% of all stars in the most massive galaxies, but more than 80% in dwarfs. This explains why dwarfs have less rotational support than massive galaxies and why all dwarfs have similar stellar half-mass radii, regardless of M * . It also explains why massive galaxies in TNG50 appear to form inside-out (as the outer disc grows), whereas dwarfs form outside-in, as star formation in the dominant inner clump moves progressively inward. The clump-disc segregation of star formation in TNG50 galaxies is probably numerical in origin. Inner clumps are formed by the accumulation of low-angular-momentum gas supported by the equation of state introduced to prevent artificial fragmentation. The decoupled-wind feedback implementation in TNG50 helps to preserve the clumps, but disrupts disc formation in its immediate surroundings. This hinders the formation of discs in (dwarf) galaxies whose sizes are not substantially larger than the clump, but it has little effect on the larger discs of more massive systems. Our results argue in favour of taking caution when interpreting the dependence on stellar mass of TNG50 galaxy morphologies, or the evolution of galaxy sizes, especially at the dwarf end.