This topic is now archived and is closed to further replies.


Adaptive stepsizing

Recommended Posts

Hi, I''ve been implementing a little physics simulator in my spare time. Obviously fast adaptive stepsizing is of great importance in interactive simulations such as games. One alternative is step-doubling, which is quite costy, though. It seems to me, that generally the fastest method for adaptive stepsizing is to build a criterion function which determines if a new computed state is accepted or not, and if it isn''t, the new state is rejected, old state is restored and time step is decreased. Many papers have proposed this technique. For example, Joukhadar''s paper (Adaptive time step for fast converging dynamic simulation system) and Baraff''s paper (Large Steps in Cloth Simulation). I''m not, I repeat NOT interested how to build a criterion function. I''m just interested about the algorithm that controls the time step. The former paper (Joukhadar) gives a nice pseudo-code of the algorithm for controlling the time step after a suitable criterion is built. The paper by Baraff, on the other hand, gives no pseudo code, but their method sounds quite elegant and efficient in the text, making intelligent choices about, for example, when not increase time step for avoiding useless tests for larger time steps. Has anyone got experience on this kind of adaptive stepsizing algorithms? What are the more efficient algorithms available these days? Any literature tips? - Mikko Kauppila

Share this post

Link to post
Share on other sites