Next_step! resulting in slightly different results for the same inputs

Specific Domains Modelling & Simulations
9

I used the example Tethers.jl/src/Tether_07.jl at main · ufechner7/Tethers.jl · GitHub and added using Sundials.

Getting the following results for tol = 1e-6:

  • CVODE_BDF()
    Elapsed time: 0.001612388 s, speed: 6202.0 times real-time
    Number of evaluations per step: 1.7
  • QNDF(autodiff=true)
    Elapsed time: 0.00189734 s, speed: 5271.0 times real-time
    Number of evaluations per step: 2.2

@Bart_van_de_Lint So you are right, until now CVODE_BDF() is the fastest solver, about 18% faster than the fastest Julia solver I found so far.

But less accurate… I mean, there is always a trade-off, fast solution or robust and accurate solution. Your choice.

By the way, try:
?QNDF() and
?CVODE_BDF()

Both of them have a lot of options.

@ChrisRackauckas Why don’t we have a Julia solver yet that is faster than CVODE_BDF() ?

Related: CVODE_BDF outperforms Julia solvers for stiff system biology model?

1 Like