From: Peter Freddolino (petefred_at_umich.edu)
Date: Tue Nov 08 2016 - 15:10:04 CST
I don’t think this is off topic — definitely useful methods for people to have in mind, and some of these ideas can be added pretty painlessly to existing workflows. For cases requiring implementation in NAMD itself, there has been some discussion over the last couple days of coming up with better ways to have distributed development of new features for NAMD moving forward, so hopefully we’ll have good mechanisms for adding some of these features in the near future.
> On Nov 8, 2016, at 10:18 AM, Brian Radak <bradak_at_anl.gov> wrote:
> I'm probably fairly off-topic at this point in the thread, but there have been quite a few recent papers regarding timesteps and integrators that I think might be of interest and also deserve wider attention:
> Long-Time-Step Molecular Dynamics through Hydrogen Mass Repartitioning
> Using Nonequilibrium Fluctuation Theorems to Understand and Correct Errors in Equilibrium and Nonequilibrium Simulations of Discrete Langevin Dynamics
> Efficient molecular dynamics using geodesic integration and solvent–solute splitting
> These are fairly well-defined advancements that would greatly benefit NAMD.
> On 11/07/2016 09:21 PM, Peter Freddolino wrote:
>> Following up on this, I would recommend looking at the standard output and error messages produced by your simulation, which likely include ‘atoms moving too fast errors’ and a crashing simulation due to the timestep being used.
>>> On Nov 7, 2016, at 8:12 PM, Richard Overstreet <roverst_at_g.clemson.edu>
>>> You need to do some research on what the parameters you are changing do to the simulation. The integrator time step affects how accurately system trajectories/forces are calculated. I have never seen someone use a 10fs time step before usually this is at most 2fs when the rigidbonds parameter is set to "all". I would recommend reading. "understanding molecular dynamics simulations" by Daan Frenkel before running any more simulations.
>>> On 11/07/2016 05:10 PM, Oscar Bastidas wrote:
>>>> I am trying to run a simulation in NAMD for 100 picoseconds but the output files (*.coor and *.dcd) never show when I execute the simulation for this simulation time. Is there a range of acceptable values for the simulation time parameters (timestep, dcdfreq and total number of runs) that if I go beyond, NAMD will not work? I have successfully run 5 and 10 picosecond simulations by altering those above three parameters in the *.conf file:
>>>> for 5 picoseconds:
>>>> Timestep=2 femtosec/step
>>>> dcdfreq=250 steps/trajectory snapshot
>>>> Total run steps=2500 steps
>>>> for 10 picoseconds:
>>>> Timestep=2 femtosec/step
>>>> dcdfreq=500 steps/trajectory snapshot
>>>> Total run steps=5000 steps
>>>> Both of these scenarios yield results (*.coor and *.dcd files), but when I try to do a 100 picosecond simulation according to the following parameters, I never get my *.coor and *.dcd output files even after waiting for several days:
>>>> for 100 picoseconds:
>>>> Timestep=10 femtosec/step
>>>> dcdfreq=1000 steps/trajectory snapshot
>>>> Total run steps=10000 steps
>>>> Is there some formula or heuristic to keep in mind when one wishes to alter these variables for longer timescales? Is there another variable that I should alter in the *.conf file in order to run at 100 picoseconds that I may have missed? Thank you.
>>>> Oscar Bastidas, Graduate Student
>>>> Department of Chemical and Life Science Engineering
>>>> Virginia Commonwealth University
> Brian Radak
> Postdoctoral Appointee
> Leadership Computing Facility
> Argonne National Laboratory
> 9700 South Cass Avenue, Bldg. 240
> Argonne, IL 60439-4854
> (630) 252-8643
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:35 CST