I did not use autodiff. I would have needed to many dual types in my code.
All my values are real. The values are exactly the same for the ode_functions and the jacobian in Matlab and Julia.
From here I don’t know if we can help any more without seeing code that reproduces the issue.
Understandable. Thanks for your time anyway :).
A small update.
I think I found the problem:
In the function of the nonlinear force, a certain variable is set to a very small value if its drops below 0.
In Matlab this never occurs. I think the solver steps are small enought. So this jumps never occur.
In Julia this value drops below 0, so the solver has to handle the jumps. I think this leads to so many function calls.