How can I make the simulation more precise in this case where a pool of particles collapses into itself?

[jt0dd]

In a simulation where I have a pool of particles freely moving around (with upward and downward forces involved based on a heat gradient), when a certain number of particles get involved, suddenly they all collapse in on eachother, as if the simulation decides "I no longer care whether these particles overlap".

I've tinkered with the broadphase and positionIterations settings and it didnt seem to fix it, any suggestions?

Here's a video of it happening:

https://i.gyazo.com/fa2052ca1c7243b3353d3f37d1e0e28e.mp4

[liabru]

Can you check out my response in #951?

It's worth trying each of those but in particular here it looks like you should switch to a fixed timestep, though the simulation may slow down due to the amount of collisions in your example.