[ODE] How to do iterative

david@csworkbench.com david at csworkbench.com
Wed Mar 19 13:45:02 2003


Ok, I sat down and read over this strategy in Numerical Recipes (it
described this method as a refinement of a good guess... exactly what I'm
looking for), and it makes a LOT more sense now.  My question is, if I use
dSolveLCP to solve for delta_lambda, will the LCP constraints (i.e. lo and
hi) still be right, or will they need to be transformed somehow?  I don't
think solving the linear algebra system directly will give me the right
solution (wouldn't contact and joint limits become sticky?).

Thanks again.... I think I'm beginning to see the light here,
David Whittaker

>> A*lamda = b (plus some stuff for the LCP solver to simulate friction
>> and limits..)
>
> Once you've got a lambda, why not do this:
>
> next_lambda = prev_lambda + delta_lambda
>
> A * next_lambda = A * (prev_lambda + delta_lambda) = b
>
> therefore,
>
> A * delta_lambda = b - A * prev_lambda
>
> So you can solve for delta_lambda directly if you make your rhs   b - A
> * prev_lambda
>
> This is why I asked whether you'd gotten all the physics out of the loop
> and were left only with linear algebra...
>
> - Martin
> _______________________________________________
> ODE mailing list
> ODE@q12.org
> http://q12.org/mailman/listinfo/ode