There are two main models of star formation. Gravitational collapse is a top-down process in which molecular clumps that are hundreds of times heavier than the Sun fragment into gaseous cores, which then collapse to make individual stars. Competitive accretion, on the other hand, is bottom-up process: stars are born as small seeds that grow by accreting material from nearby clouds of gas and, sometimes, colliding with one another.

Based on computer simulations Mark Krumholz from Princeton University, Christopher McKee from the University of California at Berkeley, and Richard Klein from Berkeley and Lawrence Livermore National Laboratory now claim that the bottom-up theory is incorrect because the seeds cannot grow fast enough during the lifetimes of the clouds to reach typical star sizes. Krumholz and co-workers simulated the accretion process for different types of molecular clumps and identified those in which the accretion rate is high enough for stars to form. However, the types of clumps in which this happens do not correspond to any that have been seen in observations.

"Our result is that the bottom-up idea doesn't work," Krumholz told PhysicsWeb, "because seeds can't accrete quickly enough to grow to stellar masses within the lifetimes of the clouds out of which they are born. Instead, stars form by fragmentation, and the fragmentation process determines their masses."

The results also explain, the team says, why observations suggest that objects as different as small brown dwarfs and massive stars have a common formation mechanism. In contrast, the accretion model involves different mechanisms for making objects with different masses. A universal formation process might also explain why the mass distribution of newly formed stars - the initial mass function - seems to be constant throughout our galaxy and other galaxies.

"Many earlier simulations of star formation processes made a significant error because they modelled environments with properties that are very different from those observed," says Krumholz. "A lot of these simulations are now going to have to be reconsidered and probably re-done."