Time step and numerical accuracy

[willstring]

Hello again,

I'm trying to run a basic collision study to show convergence of the results with the time step size.

The simulation is pretty simple, about 100 cubes being shaken around in a container.

If i'm not mistaken, the time step should be much smaller than the time scale of impact between contacting discrete elements (dt<sqrt(m/k)) for both numerical accuracy and stability of the integration schemes?  (where m is the mass of the discrete elements and k is the contact stiffness)

I've tried running time steps where dt is about 5000 times smaller than this ratio but still can't get convergence: it gives a different configuration of blocks every time.

Eventually shouldn't I be able to get convergence with fine enough time steps?  I guess I should keep going smaller?  Or am I missing something...?

[Damien André]

Hello,
what is your initial configuration ?
Do the blocks are of equal sizes and aligned on a grid like this ?


Do you apply Gravity ? How do you shake your box ?
Thx, Damien.

[willstring]

Hold up... might be an issue on my end... let me fix some things and i'll get back to you...

Thank you again, for your help!