We investigate the star formation-feedback cycle in cosmological galaxy formation simulations , focusing on progenitors of Milky Way ( MW ) -sized galaxies .
We find that in order to reproduce key properties of the MW progenitors , such as semi-empirically derived star formation histories and the shape of rotation curves , our implementation of star formation and stellar feedback requires 1 ) a combination of local early momentum feedback via radiation pressure and stellar winds and subsequent efficient supernovae feedback , and 2 ) efficacy of feedback that results in self-regulation of the global star formation rate on kiloparsec scales .
We show that such feedback-driven self-regulation is achieved globally for a local star formation efficiency per free fall time of \epsilon _ { ff } \approx 10 \% .
Although this value is larger that the \epsilon _ { ff } \sim 1 \% value usually inferred from the Kennicutt-Schmidt ( KS ) relation , we show that it is consistent with direct observational estimates of \epsilon _ { ff } in molecular clouds .
Moreover , we show that simulations with local efficiency of \epsilon _ { ff } \approx 10 \% reproduce the global observed KS relation .
Such simulations also reproduce the cosmic star formation history of the Milky Way sized galaxies and satisfy a number of other observational constraints .
Conversely , we find that simulations that a priori assume an inefficient mode of star formation , instead of achieving it via stellar feedback regulation , fail to produce sufficiently vigorous outflows and do not reproduce observations .
This illustrates the importance of understanding the complex interplay between star formation and feedback and the detailed processes that contribute to the feedback-regulated formation of galaxies .