Convergence

[finnBsch]

Been working on something similar and noticed that circles tend to stay in collision if not interesting sufficiently many times. I saw that in your implementation, you only iterate two times, not until convergence, correct?
However, the results seem to show pretty much perfect collision resolution, which doesn't make any sense to me. Is there any intuitive explanation?

[aevinj]

I think its a mixture of substepping and decreased levels of forces acting on the particles (gravity). I myself am working on my own implementation, and realised with more substeps, there's less overlap in collisions - thus less collision resolution applied removing the need for multiple iterations for collision resolution. Furthermore, I started with a high value for gravity which caused densely packed regions of particles - inevitably requiring more iterations of collision resolution - therefore reducing this value, helped.