would be even better to be able to avoid the divergence withtout stoping the simulation: for example unactivate collision or adding damping / more iteration to the solver