Stéphane MOTTELET

Thanks for this confirmation. I thought that DASSL had the same limitation, is it the case ?

Thanks a lot, using `IDASuppressAlg` solved it for the sliding problem. For the above pendulum I compared with and without indication of algebric variable and suppressing it in the error...

In the next days I will write a pure C implementation of the failing example to allow further investigations.

In your pendulum.c example (and in my Scilab code), lambda is initially 0 which is correct considering the initial position (0,1), but when I test with (sqrt(2)/2,sqrt(2)/2)), initializing lambda with...

Here is the example (a sliding pendulum) that fails with IDA: [sliding_pendulum.c.zip](https://github.com/LLNL/sundials/files/9654530/sliding_pendulum.c.zip) unless `IDASuppressAlg` is used (the corresponding part has been commented out in the source). In that case the...

Here is the same example with the code using dassl : [sliding_pendulum_dassl.c.zip](https://github.com/LLNL/sundials/files/9654124/sliding_pendulum_dassl.c.zip) and default options. I got the following output (at t=0.05): ```` y[0]=0.994788 y[1]=0.659589 y[2]=0.00521073 y[3]=-0.209267 y[4]=-0.44875 y[5]=0.209173 y[6]=-2.0401...

The dassl example has been linked against netlib's Slatec and both examples against macOS Lapack.

OK thanks. More generally speaking, would you recommend using IDA instead of DASLL suite in every situation ?

OK. I was just asking because some software (e.g. DiffEq in Julia) propose DASSL. But the logic of DiffEq is to rewrite all solvers natively, hence simpler to do for...

> If your function is `f(x) = x^2-2` then `df(x)/dx = 2.0*x`. Your Jacobian returns `df(x)/dx = 2.0`. I know how to compute a Jacobian, please see the Kinsol documentation...