This gradient function should not be needed. generate_ode_function creates the gradient function.

This may not work:

• the signature of solve_ivp is not the same as generate_ode_function (I believe generate_ode_function has the signature of odeit.)
• If in solve_ivp a method other than RK45 is used, the y is not C contiguous, so y = np.ascontiguousarray(y) is needed.

Doing solve_ivp(lambda t, x: rhs(x, t) should be sufficient. If there is some issue with RK4 and y not being contiguous, this sounds like a new bug. That has never been required in the past.

I must admit, I am not goot with these lambda functions.
Yes, there is an issue if e.g. method='Radau' in solve_ivp: I raised issue # 23657 with scipy, so
``y = np.ascontiguousarray(y) is needed, best I can tell.

This is also needed, where I calculate the reaction forces.

There are several instances of variables that need to be in double backticks throughout the page. Fixed width font in ReST is via double backticks.

RK45 is likely not appropriate choice for a model with collision (i.e. it is a stiff system).

If you use odeint() then you don't need the gradient function and it uses LSODA as the integrator by default which switches to a stiff integrator when needed.

scipy discourages the use of odeint for new code: https://docs.scipy.org/doc/scipy/reference/generated/scipy.integrate.odeint.html
I could try 'Radau', which is for stiff systems.

PyDy uses odeint by default. So, until PyDy changes that I don't think we need to abide by SciPy's statement.

You can run LSODA with odeint or solve_ivp, btw.

That's what I assumed, given the signature of generate_ode_function.
Of course I can switch to odeint, just thought that given scipy's recommendation it may be good to show how to use it with generate_ode_function. Mostly this np.ascontiguousarray(..) was not trivial for me.

Radau works fine.

However, if rhs_gen was generated with 'cython' 'and is used to, say, calculate the accelerations needed for the reaction forces still np.ascontiguousarray(..) is needed.

You'll need a minimal example to show this.

The example is pushed shows this:

.. jupyter-execute::

    RHS = np.empty((resultat.shape))
    #pl_vals = np.ascontiguousarray(pL_vals)
    for i in range(resultat.shape[0]):
        #res = np.ascontiguousarray(resultat[i])
        res = resultat[i]
        RHS[i] = rhs_gen(res, 0.0, np.array(pL_vals))

If the np.ascontiguousarray(..) are blanked out like above, it will raise an error.

However, if rhs_gen was generated with 'cython' 'and is used to, say, calculate the accelerations needed for the reaction forces still np.ascontiguousarray(..) is needed.

You'll need a minimal example to show this.

I made a minimum simulation and could NOT repeat this problem!

• I had failed to convert my constants from list into an np.array
• Strangely, when I use resultat[i] as given from solve_ivp, it does not work with generator='cython'. (it works with generator='symjit'). If I put resultat[i] = np.array(resultat[i]) it works with generator='cython'

Le me know if I should send you this minimum example.

I played around some more and found that somehow applying np.array(x) to an np.array x seems to 'heal' to non C contiguous issue.

Interesting!

You can always report a new issue. The best is to include a minimal reproducing example that I can copy/paste from the issue.

Interesting!

You can always report a new issue. The best is to include a minimal reproducing example that I can copy/paste from the issue.

I raised issue #547 with code showing it.

Many of these non-sentences are likely meant to be section headers.

Will change accordingly.