AMBER Archive (2008)
Subject: Re: AMBER: steered MD initial state
From: M. L. Dodson (activesitedynamics_at_comcast.net)
Date: Tue May 06 2008 - 12:27:03 CDT
Gustavo Seabra wrote:
> Hi Bud,
> If I understood correctly, what you are getting are maybe different
> trajectories, but all started from the same replica, that only differ
> on the random number used as seed. After the temperature
> equilibration, You end up with a number of different restart files
> which you then use for the SMD runs, is that correct? If this is the
> case, I'm afraid you may be right and the different starting points
> are not necessarily of the same ensemble.
Well, I would not phrase it that way, but I think I understand you.
All replicas start from only one restart file, however each SMD
replica is preceded by a 100000 step temperature equilibration (as I
described). To be specific, I have 12 replicas at this time, each
corresponding to a different random number seed during the temperature
equilibration. That means 12 temperature equilibrations, followed by
12 SMD runs, each SMD run initiated with the last frame of the
corresponding temperature equilibration run.
> The idea in Jarzynski calculations is that each replica will be a
> different *starting point* for the process, all belonging to the same
> initial ensemble. The way we do it here is extracting frames from an
> equilibrated trajectory, for example, one frame every 100ps or so, so
> that each replica is a different starting point, all belonging to this
> same initial ensemble. Still, it is a good idea to give each replica a
> different random number seed, especially if you are using the Langevin
> thermostat (which you SHOULD, especially for QM/MM calculations).
OK, that is the way I had planned to do it until I became (maybe too)
concerned that such replicas would not adequately sample the starting
ensemble. Remember that each such starting point is deterministically
reproducible from earlier starting point instances (modulo truncation
to the precision of formatted restart files). That was my thinking as
of a couple of weeks ago. Using different seeds for Langevin
temperature control during the SMD runs handles such concerns
adequately, though, I agree.
So I understand your protocol description. But think about an
experimentalist doing an AFM pulling experiment. They do the pull,
relax back to the starting distance, then pull again (at least that is
my understanding from seminars I have attended). If they did not do
that, they would have to set up each replica independently, which is
even more different (as far as the initial states) than just relaxing
back to the initial distance. So it seems that the same logic that
concludes we must rigorously start from the same ensemble (in the
manner you describe), if that invalidates my replica set, also
invalidates the experimentalists' protocol. Or am I being too obtuse
I have no qualms about doing it exactly as you describe in the future,
but I have about two weeks computation already in hand I would like to
avoid throwing away unless I can convince myself it is worthless. And
as I get further into the analysis, I may need to do more replicas.
I'll bite the bullet if I need to, though. I just thought it was
worthwhile to tap the expertise at hand first.
> On Tue, May 6, 2008 at 12:04 PM, M. L. Dodson
> <activesitedynamics_at_comcast.net> wrote:
>> Hello Ambers,
>> I have question about setup of SMD runs having to do with selecting
>> the initial coordinates and velocities. I understand from the
>> literature that all replicates should be drawn from the same well
>> equilibrated ensemble. To ensure that I was getting sufficiently
>> independent trajectories for the SMD replicates, I took my (well
>> equilibrated) restart file, truncated the velocities with NTX=1, and
>> set each replicate up with different random number generator seeds
>> (chosen from output from random.org). This is a QMMM run
>> investigating a proton transfer reaction, so I carried out a 100000
>> step (0.05fs step size, QM shake turned off) equilibration of the
>> temperature. Analysis showed the temperature across the trajectory to
>> have stabilized. I followed this with the SMD run. These SMD
>> trajectories, differing by random number generator seed, constitute
>> the replicates to be averaged with the Jarzynski procedure.
>> Due to a question on the list a couple of days ago, I reread the
>> Theor. Chem. Acc. paper by Dr. Roitberg and his coworkers. They point
>> out that the SMD trajectory replicas going into the Jarzynski
>> averaging algorithm should be drawn from the same starting ensemble.
>> After rereading the paper, I got nervous. Does my protocol render the
>> replicate trajectories invalid for use because my protocol means they
>> are no longer drawn from the same initial ensemble?
>> My thought process was perhaps too focused on the independence of the
>> trajectories, but it seems to me this protocol is no more invalid than
>> repeated experimental atomic force microscope pulling runs.
>> Does anyone have any comments on this? I would appreciate hearing
>> Bud Dodson
M. L. Dodson
Business Email: activesitedynamics-at-comcast-dot-net
Personal Email: mldodson-at-comcast-dot-net
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu