[ODE] solving penetrations "statically"

[Anton Shekhovtsov]

Erin Catto wrote:
> Otherwise, you'll need to compute two sets of Lagrange multipliers
> (lambdas). The first set includes the position correction while the second
> set does not. Apply the first set to generate non-persistant velocities
that
> are simply used to update the positions. Use the second set of lambdas to
> generate the final velocities for the step. This way position correction
> won't affect the momentum of the system.

Looks promising, however I do not understand the subject of lambdas.
What part of ODE source is involved with this? How much I am about to change
to do that? :)
could you please point me to some more info