We perform radiation hydrodynamical simulations of protostellar collapse in spherical symmetry, with a special focus on very low-mass objects, i.e. brown dwarfs and sub-brown dwarfs. The inclusion of a realistic equation of state that includes the effect of hydrogen dissociation allows for a modeling of the complete process from the beginning of the collapse until the formation of the protostar. We solve the frequency-dependent radiative transfer equation without any diffusion approximation, using realistic dust and gas opacities. Our results show that the properties of the protostar are essentially independent of the initial conditions, which had previously only been confirmed for higher mass ranges. For very low mass initial conditions, however, we find that the first core phase of the collapse shows some significant differences in the time evolution, with the first core lifetime increasing dramatically because of the reduced accretion rate from the surrounding envelope. We consider the observational implications of this. We also investigate the opposite case of a collapse without any first core phase, which may occur for very unstable initial conditions. In the appendix, we describe a severe numerical problem that causes an unphysical expansion after the formation of the protostar, which may affect other attempts at similar calculations of self-gravitational collapse. We explain the origin of the unphysical behavior and present a solution that can be used in similar investigations.

We develop a new numerical scheme for solving the radiative transfer equation in a spherically symmetric system. This scheme does not rely on any kind of diffusion approximation and it is accurate for optically thin, thick, and intermediate systems. In the limit of a homogeneously distributed extinction coefficient, our method is very accurate and exceptionally fast. We combine this fast method with a slower, but more generally applicable method to describe realistic problems. We perform various test calculations including a simplified protostellar collapse simulation. We also discuss possible future improvements.