[ODE] ODE Logo

[GARY VANSICKLE]

> What kind of spring constant are you thinking?

Well, let's take the ever-present automobile suspension.  Please ;-).

>  If it's the sort of
> thing where compression of the spring won't change much from one
> timestep to the next (say by < 25%), you're probably better off doing it
> by simply adding a force. 

 In the auto suspension you're going to have a stiff spring and large
displacements.  Mix in Euler, and I smell an explosion brewing.

> The solver's computationally expensive, don't
> make more work for it than you need to.
> 

No I know, that's why I have yet another wishlist item:

- Global System Simplification
 The name here probably makes no sense, but I don't quite know what to call
it.  Right now, everything goes in one big matrix and gets LCP'ed.  Great,
but is it somehow possible to break the system in question into subproblems
which don't need the full treatment and hence have simpler (perhaps even
closed-form and hence unconditionally stable) solutions?  Russ touches on
this concept himself in the manual when he mentions that a future direction
might be to integrate Featherstone for islands that have no cycles, and
hence reap the improved speed of that method.  Well, if'n ya can do that,
seems to me that it should be even simpler to plug some spring equations in
there somewhere.  And likewise with any number of special cases which I
can't think of right now.

Maybe that's just not possible, I don't know and probably never will unless
somebody with more quarters of calc than I explains it to me.  But the thing
is, if my system needs a spring, it needs a spring.  Computational cost is a
secondary issue.  And it seems to me that springs are as basic a component
as joints, hence the wishlist item.