Re: LuaJIT2 performance for number crunching

[T T <t34www@...>]

On 16 February 2011 08:47, Francesco Abbate <francesco.bbt@gmail.com> wrote:
>
> I'm working now to implement the system to control the error and
> adjust the step of automatically. This is something required in real
> cases where a fixed step size would be a too much coarse approach.

If you are serious about writing RK (explicit) ODE integrator with
auto time step control, then the Dormand–Prince method with embedded
RK 4/5 pair is the one to go for:

http://en.wikipedia.org/wiki/Dormand–Prince_method

See also:

"A survey of the MATLAB ODE suite"
http://www.osaka-kyoiku.ac.jp/~ashino/pdf/2651.pdf

Double time step error control is not exactly the state-of-the-art any more.

Cheers,

Tomek