3 ms·
It's a pity there are no JavaScript options listed.
by jxcole 9y ago
It's a pity there are no JavaScript options listed.
- tlarkworthy 9y agoWouldn't it be a single entry, numeric.js?
- ChrisRackauckas 9y agoYeah, I'm not sure of any others. But looking at numeric.js, it has a single ODE solver which is Dopri, and thus matches the tableau of ode45 and dopri5. However, it's easy to see that their implementation is very different just by this line: h = min(0.8*h*pow(tol/erinf,0.25),4*h); That means that the timesteps are changes via Cechino's optimality theorem, the "standard" timestepping method which isn't actually used in production quality solvers. The reason is because this form of adaptivity causes solvers to change their h too rapidly, which actually decreases the stability region and increases the number of step rejections. So all of the other explicit RK solvers mentioned (except when I single it out) all use a PI-controller. And this makes a pretty big difference: I've found something like a 10x performance change by using a PI-controller on many equations, and then there's many other equations where at a "simple" tolerance like reltol=10^-3 the dopri-type method will diverge without a PI-controller on the timesteps. But it does have event handling and dense output. However, since it's lacking any solvers for stiff ODEs, I am not sure it really is a full suite.