yade-users team mailing list archive

[Bruno Chareyre <question695226@xxxxxxxxxxxxxxxxxxxxx>]

Question #695226 on Yade changed:
https://answers.launchpad.net/yade/+question/695226

Bruno Chareyre posted a new comment:
You are past-the-end of my checklist. :)
RungeKuttaCashKarp54Integrator makes sense then.

Last one maybe: in a working case (small damping), did you find
consistency _at the structural scale_ in terms of damping the
oscillations? It could be that there is a mismatch between what you
input at the interactions scale and what you expect in terms of damping
at the structural scale?

> is similar to forward Euler

Yes and no. It is a 2nd order time centered scheme from Newton's point
of view. Problem is velocities are not known on step, they are know at
mid-steps. So this wonderful scheme for mass-spring systems becomes a
poor 1st order Euler when there are viscous terms. If you take mid-steps
as your frame of reference and you think to what happens to velocities
wrt. visous forces it is plain explicit Euler, you are right.

There are ways to evaluate on-step velocites (something similar to
"velocity Verlet" [1]) which we don't use. It should improve your case,
it needs some changes in the integrator and it needs two calculations of
forces per time iteration.

> the RungeKuttaCashKarp54Integrator is extremely slow.

Because there are even more than  two calculations of forces per time
iteration, I guess?

Do you mean that despite larger timesteps it is still slower?

Bruno

[1] https://en.wikipedia.org/wiki/Verlet_integration

-- 
You received this question notification because your team yade-users is
an answer contact for Yade.