On Aug 9, 2013, at 7:20 AM, Brian Radak <radak004.umn.edu> wrote:
> Page 13 of the manual is probably supposed to cover this, but I agree that
> there are a few sneaky things and "gotchas" in mdout files. I'm not at all
> an AMBER code expert, so this may not be 100% accurate:
>
> Etot = EKtot + EPtot - total energy
> EKtot = 1/2 sum_i m_i v_i^2 - total kinetic energy
> EPtot - total potential energy
>
> NB: Because of Leapfrog Verlet, the velocities and positions are NEVER
> propagated to the same point in time. Therefore the kinetic energy never
> corresponds to an actually realized velocity vector of the system. I
> believe the solution in AMBER is to either average the velocities from a
> half step back and a half step forward or else somehow do a half step
> integration to get EKtot. This also results in EPtot not corresponding to
> the configurations written to mdcrd unless each and every integration step
> is written to disk.
The velocities are averaged between t+1/2 dt and t-1/2 dt. This is accurate to the order of dt^2 I think. Perhaps marginally worse than velocity verlet, but still quite good. (And a lot better than straight verlet for velocities).
HTH,
Jason
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Aug 09 2013 - 14:30:02 PDT
