[ODE] Speed of ODE's constraint method

[Russ Smith]

hi richard,

> I should say though that LCP methods are very stable,
> and once Russ sorts out box-box, ODE is going to be
> rock solid.

thanks for the vote of confidence!

> The reason that there are direct LCP games
> in the top 20 at the moment despite the above hurdles
> is because the stability is so good.

i am curious about one thing: do your customers percieve that the
stability itself is the good thing, or that the shortened development
time that results is the good thing? (both? neither?)

> <Marketing hat on>
> <Marketing hat off>
> Sorry about that...

as a penance for shamelessly marketing your stuff here :), perhaps you
could tell us exactly what an 'iterative LCP' solver is? i'm genuinely
curious. are you:
  * running a fixed number of LCP iterations per timestep?
  * factorizing the non-LCP constraints fully?
  * re-using constraint clamping info from step to step?
    - if so, how do you handle contact coherency problems?
  * in general, how is the LCP restarted?

also, there is one thing to add to your explanation of where the time
goes: when a large impact occurs in the system (e.g. a car crashes into
a pile of boxes) there is usually a single time step at the moment of
impact that goes *really* slowly. the problem is that the LCP solver
needs a large number of iterations to effectively spread the impact
force throughout the rest of the system (this problem can be understood
more precisely by considering how the algorithm steps through the
elongated LCP solution space). it sounds like this situation would
benefit the most from your constant iteration technique?

russ.

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