[gmx-users] micelle disaggregated in serial, but not parallel, runs using sd integrator

Berk Hess gmx3 at hotmail.com
Thu Feb 5 10:25:09 CET 2009


I don't know why I did not add checks for ld-seed before.
Now grompp gives a note when continutation=yes and ld-seed!=-1.

tpbconv will now generate a new ld-seed when reading a trajectory
(but you should not use tpbconv, use a checkpoint file instead).

But yesterday I forgot to tell that there is a bug in the checkpointing
of mdrun in 4.0 - 4.0.3. Without domain decomposition the initial box
size would always be stored in the checkpoint file, which causes problems
with NPT simulations. NPT simulations with domain decomposition and
all NVT simulations were fine.

Gromacs 4.0.4 will all bugs fixed and extra checks should be released today.


> Date: Wed, 4 Feb 2009 19:30:47 -0500
> From: chris.neale at utoronto.ca
> To: gmx-users at gromacs.org
> Subject: [gmx-users] micelle disaggregated in serial, but not parallel, runs using sd integrator
> It appears as if you were correct Berk. I will report on the results of my 24h test tomorrow, but I also set up another system 
> that used ld_seed=1993 and ran in 20 ps segments instead of the 200 ps segments that I was previously using. This system shows 
> signs of disaggregation on the 200 ps time-scale as opposed to the 2 ns time-scale that I observed for 200 ps segments.
> I don't know how you figured that one out, but I am very grateful.
> Now that I see the trajectories, it does make sense that any net movement applied to an individual molecule by the noise will 
> lead to directed movement over many separate segments.
> I think this is probably worth a note in the grompp output for sd runs when a user sets ld_seed to something other than -1 and 
> utilizes the -t option (or some other indication that this is intended as a continuation).
> Chris.
> -- original message --
> Thank you Berk,
> I will repeat my runs using the checkpoint file and report my findings back to this list. Thank you for this advice.
> Chris.
> -- original message --
> Hi,
> In this manner you use the same random seed and thus noise for all parts.
> In most cases this will not lead to serious artifacts with SD,
> but you can never be sure.
> When checkpoints are used, you do not repeat random numbers.
> This also gives a difference between serial and parallel in 4.0.
> With serial you get exactly the same noise per atom, in parallel not,
> since atoms migrate from one node to another (with domain decompostion).
> If you do not use checkpoints, use ld_seed=-1 and do not use tpbconv.
> Berk
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php

Express yourself instantly with MSN Messenger! Download today it's FREE!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20090205/c6369c16/attachment.html>

More information about the gromacs.org_gmx-users mailing list