What about using a Runge-Kutta integration scheme? I haven't studied
cloth-simulation in any extent, bu I suppose one has a (poosibly) huge set
of ODEs to solve. Euler is intrisicly unstable, but 4th order Runge-Kutta is
very stable, even with high force gradients in newtonian systems.

regards
Simen Kvaal.


> I've been implementing the Backward Euler method, which is supposed to be
> much more stable in stiff equations (like those which arise in cloth
> simulations). It has not been very easy, mainly because, in contrast with
> forward Euler, one has to find the derivatives of the force functions, and
> solve a linear system.
>
> Even though it takes much more time per iteration, in theory, one can take
> much larger time steps, and it's faster in the long run.
>
> However, I'm still having to take rather small steps, because if not, I
have
> stability problems.
>

Post a reply to this message