I still do not see an image...
On Tue, Oct 27, 2015 at 12:10 PM, Fabian gmail <fabian.glaser.gmail.com>
wrote:
> Yeah I know, but still there is a qualitatively jump that stays stable for
> each different 5ns simulation, I attach a figure hope you can see it.
> Carlos Simmerling explained me what you say that there is jump in time, and
> restarting I understand also produces some "instantaneous jump in the
> restraint energy, which changes the forces and the simulation adjusts to a
> new average structure with different energy” in Carlos words.
>
> But still it puzzled me a little, here is the jump please notice it gets
> smaller and smaller with simulation time…. but there is a jump specially
> after 5ns and 10 ns, there it gets much better.
>
Restraints with constant pressure are a little weird. To implement
constant pressure, what sander/pmemd does is scale the molecule positions
slightly (either bringing all of them closer to the center of the box or
farther away, depending on whether the pressure needs to be decreased or
increased to match the external barostat, respectively).
If the restraint positions stay the same, you now have an external force
pulling the atoms back to where they *used* to be, thereby opposing the
scaling move. I believe that to avoid this, what sander/pmemd does is to
scale the restraint coordinates *as well*. So what I think might be
happening here is that when you are restarting *with the same restraint
coordinate file*, you are also introducing a (very slightly) different set
of restraint coordinates as sander/pmemd was using at the end of the
previous simulation.
This will introduce an impulse upon restart (since the restraint force is
now a bit different than it was at the end of the previous simulation).
Since this is what you say you are seeing (send us pictures! ;)), it would
seem to me to be a reasonable explanation.
You can test this idea in one of two ways:
1. Remove the restraints
2. Remove the barostat
If either eliminates the artifact you are seeing, chances are increased
that what I described is the culprit (which is less of a problem than using
restraints for your whole simulation).
I plan to use your MMPBSA.py script only starting at 20 or 30 ns for 20 ns
> (2000 frames) is that good enough?
>
You will probably get the same results using 200 frames over 20 ns than
2000. MMPBSA.py results converge within a few hundred frames (properly
sampled and independent from each other, of course).
HTH,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 27 2015 - 10:00:03 PDT
