Im implements gpu-based simulation using spring mass model with reverse Euler integration. The linear system is solved by solving the conjugate gradient with a filter. Everything is done in the GPU. I think that, as an implicit integration, it should be stable, as indicated in many papers, but it is unstable, as is the implicit method. In most cases, when the time step size reaches a certain value (depending on stiffness), CG refuses to converge and goes into an endless loop.
Although I checked the code again and again and read a lot of documents, I still could not find out the reason. The fabric moves correctly, and the animation is much more convincing than the one that uses explicit integration, so I think the forces are calculated correctly. Is there something that I missed to cause instability?
I ran into this strange problem for several days ......
Anyone can help? any suggestion that I would really appreciate
Thanks in advance!
source share