Repeatability of Forward Dynamic Simulations

[James Buffi]

Hi all,

Several colleagues and I are attempting to perform forward dynamic simulations on different computers (same versions of Opensim) using the exact same input information. When we do this, our simulations match well for the first part of the trial but afterwards all diverge in very different ways. I understand that numerical error builds up, which is why the simulations diverge, but why are the results from different computers so different? Does it have to do with differences in the computer hardware or processing speeds?

Thanks!

[Tim Dorn]

I think its reasonable to expect differences using different machines precisely because the systems are highly non linear and unstable. Even a one bit difference in the state at any time will result in divergance for the rest of the simulation. This could be hardware dependant or operating system dependant. Its impossible to evaluate how 'normal' the change is - it depends on your dynamic system. One important thing however to evaluate is that the same simulation on the same machine is repeatable.

[James Buffi]

Thanks Tim,

We are even getting different results on the same machine using the exact same inputs. The simulations match for much of the durations of the trials but when they diverge, they diverge in different ways. Based on your previous answer, I am guessing that this is because of the complexity and instability of the upper extremity system we are simulating in addition to the processing speeds and availability of memory on our computers at the times we are running the simulations.

I really appreciate the feedback and any more insight you can give about this would be great.