Nevermind, I realize that thought was silly because of how linear stability is defined. Carry on.
Though I would still give a fixed timestep method a try to narrow down the bug to adaptive stepping. There may be some issue somewhere, something weird like how it chooses min/max timesteps? On Monday, June 27, 2016 at 6:15:45 AM UTC+1, Chris Rackauckas wrote: > > Won't some of the Runge-Kutta solvers be really unstable with a small > negative Delta t? What happens when you use a fixed time step method: does > it run and diverge? >