Bruno Chareyre a écrit : >> Hmm... As me seems, prevVelocities is specific for leapfrog method. >> For another integrator may be another implementation (Gear >> predictor-corrector as example...) >> > > Given that the kinematics (pos,vel) at time t depends on previous > kinematics. I don't see how an integrator could compute the motion > without previous velocity (the result would be independant on e.g. > initial velocities!). > Yes, different schemes could need vel(t-2dt) instead of vel(t-dt) (or > both of them), but you always need a "previous" velocity. > > Bruno >
Mmmmh... Perhaps I missed your point. It may be true that some schemes don't need previousVel, and in fact it rings a bell for me : Indeed, even in leap-frog, we could write v(t+dt) = v(t) + acceleration*dt (in the code, it would read : v += acceleration*dt). where v(t) is the previous velocity, but it's erased as soon as current velocity is computed. This is how I write the second order finite difference scheme usually, and I'm still not sure that the equations in Yade are equivalent to this. I will double-check that point as it could explain why we need unusually low timesteps... Bruno > _______________________________________________ > Yade-users mailing list > [email protected] > https://lists.berlios.de/mailman/listinfo/yade-users > _______________________________________________ Yade-users mailing list [email protected] https://lists.berlios.de/mailman/listinfo/yade-users
