Example for RungeKutta4 is unstable

[John-MF]

Hi,
I tried running the example examples/Component/ODESolver/Forward/RungeKutta4Solver.scn, but the simulation always explodes around the 5th second. Even setting a dt of 0.001 I get this problem.

Can someone check if they have the same behaviour ? I should have recompiled a clean version of SOFA without fast math, but I may have missed something still.

Here is how it behaves on my machine (instability appears at 00:14)

SOFA.v24.06.00.-sofa_src_examples_Component_ODESolver_Forward_RungeKutta4Solver.scn.2024-09-27.10-13-23.mp4

[hugtalbot]

Hi @John-MF

Thanks a lot for reporting this.
Unfortunately I do not use the RK4 often (nor other explicit schemes). Since the regression tests never complained I guess it has this behavior for some time. I will discuss this with other devs but I checked and this behavior occurs also:

• with lower dt, meaning that it does not seem to be a stability issue
• with a direct solver, this does not seem to be a matrix-conditioning issue
• with a some of the point partially fixed instead of fully fixed in order to decrease the over-constraint effect, this did not change anything
• with a lower Poisson ratio, it helped a bit

I only reached a stable simulation using a higher stiffness ($E=5000$) and a smaller time step : $dt=1e^{-4}$
Possibly the RK4 is just not the most appropriate scheme in such a case.

I hope this helps

[John-MF]

Thank you, at least I can stop looking for a problem in my setup for now.

[hugtalbot]

Hi @John-MF

FYI the PR #5036 fixed the scene.
Again, the RK4 is just not the most appropriate scheme in such a case. To use RK4, damping should be added to stabilize the integration scheme.

Hope your work with SOFA is now running smoothly. Cheers