In a force-based graph-layout simulation using Barnes-Hut, what are the conditions for collapse? With collapse I mean multiple (or even all) nodes "collapsing" into a single point. Is there some research I could look into? Any hint is appreciated.
Background: We've written a force simulation to compute a graph-layout. But when running the simulation a bit too long all nodes of the graph collapse into a (or a few) single point(s). Here You can see a grid of 16 successful simulation runs, and afterwards a collapse, where we chose the parameters a little bit differently.