Understanding how edge misfit dislocations (MDs) form in a GeSi/Si(001) film has been a long standing issue. The challenge is to find a mechanism accounting for the presence of these dislocations at the interface since they are not mobile and cannot nucleate at the surface and glide towards the interface. Furthermore, experiments can hardly detect the nucleation and early stages of growth because of the short time scale involved. Here we present the first semi-quantitative atomistic calculation of the formation of edge dislocations in such films. We use a global optimization method and density functional theory calculations, combined with computations using potential energy functions to identify the best mechanisms. We show that those previously suggested are relevant only for a low film strain and we propose a new mechanism which accounts for the formation of edge dislocations at high film strain. In this one, a 60° MD nucleates as a "split" half-loop with two branches gliding on different planes. One branch belongs to the glide plane of a complementary 60° MD and therefore strongly favors the formation of the complementary MD which is immediately combined with the first MD to form an edge MD.