[gmx-developers] Thoughts on pressure control / constraints / leapfrog

[Berk Hess]

>The difference occurs is that because with velocity verlet, to
>correctly implement constraints, at time t, it is possible (necessary,
>actually) to correct both the velocities and the positions such that
>they obey constraints (i.e., SHAKE and RATTLE). In GROMACS currently,
>this constrained velocity at t+0.5*dt is determined by taking the
>difference between the constrained positions r(t+dt) - r(t) / dt.
>This is not equivalent, and it doesn't appear that v(t+0.5*dt) is
>actually obeying any constraints with respect to the bonds, but I
>haven't finished with the math yet.  Does anybody have any references
>that leapfrog and verlet are equivalent with constraints?

I just derived that v(t+0.5*dt) in the constraint direction at t+0.5*dt is 
zero.
I don't know of any references, but I would say that also with constraints
the algorithms should be identical (unless you constrain the velocities
incorrectly).

Berk.

_________________________________________________________________
MSN Search, for accurate results! http://search.msn.nl