[ODE] Other integrators (was: ODE implementors guide)

[Russ Smith]

> "The time stepping methods presented here are partially implicit
> Euler methods ... So, ti is probably best to combine DAE methods and
> time stepping methods; when the DAE method fails, use a timestepping
> method with small step-size h>0 to get past the impulsive part of the
> solution, and then re-start the DAE method."

i don't recall what this paper is about, but i think the point is that
more "traditional" DAE solution methods can be taylored to give a
certain desired level of accuracy, but these methods become
significantly less accurate (and also potentially a lot less stable)
when the system encounters impulses (e.g. collisions). implicit
time stepping methods handle the impulses better.

> As I understand it ODE does not currently use a hybrid method, but
> exclusively uses a partially implicit Euler time stepping method,
> right?

yes.

> How difficult would it be to use a hybrid scheme, where the DAE part
> allows a user-defined integrator?

not too difficult. currently the time stepping scheme is mixed in
with all the other system calculation, not separated out as you 
would do if you thought of your integrator as a separate module.
but this separation is not hard, and after all an integrator is
really a small piece of code compared to the complexities of
calculating the system matrix.

russ.

--
Russell Smith
http://www.q12.org