From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Mon Sep 09 2013 - 01:26:14 CDT
Hi Hailey,
1st thing to mention is that the parallelization itself can change results
slightly. This happens due varying orders of arriving part results that get
computed together. If you changed the number of processors between the 5
trajectories you mentioned, this could be a 1st likely reason. Furthermore,
but only an assumption, the load balancer can have additional impact as
individual core performance can have a random nature and not every core is
as fast as the others, even if same model. Moreover I'm not sure if the
langevin thermostat comes with random forces.
To clarify all these things, try using only a single cpu core for some
tests. This will eliminate the load balancer and the parallelization and
will show if other thing in your simulation come with any randomness.
Norman Geist.
> -----Ursprüngliche Nachricht-----
> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
> Auftrag von Hailey Bureau
> Gesendet: Montag, 9. September 2013 00:54
> An: NAMD list
> Betreff: namd-l: explicit NVT simulation
>
> Hello,
>
> I am running an explicit NVT simulation and I am having trouble
> reproducing data, using the same starting coordinates and random seed
> value. In a batch of 5 trajectories, that I run two times with
> identical seed values, I see different results. However, sometimes the
> trajectories do turn out the same. I have been trying to find any
> previous interest in this sort of problem and how to solve it, but I
> haven't had much luck. Has anyone encountered a problem like this
> before? I am running on one CPU; below is my starting configuration
> file:
>
>
> #############################################################
> ## ADJUSTABLE PARAMETERS ##
> #############################################################
> structure ../../../../00.struc/03.exp/00.psf
> coordinates ../../../../00.struc/03.exp/00.pdb
> outputName daOut
> #############################################################
> ## SIMULATION PARAMETERS ##
> #############################################################
> # Input
> seed xxxxx
> paraTypeCharmm on
> parameters ../../../../toppar/par_all27_prot_lipid.prm
> temperature 300
>
> # Force-Field Parameters
> exclude scaled1-4
> 1-4scaling 1.0
> cutoff 12.0
> switching on
> switchdist 10.0
> pairlistdist 13.5
>
> # Integrator Parameters
> timestep 2.0 ;# 2fs/step
> rigidBonds all ;# needed for 2fs steps
> nonbondedFreq 1
> fullElectFrequency 2
> stepspercycle 10
>
> # Constant Temperature Control
> langevin on ;# do langevin dynamics
> langevinDamping 5 ;# damping coefficient (gamma) of 5/ps
> langevinTemp 300
> langevinHydrogen no ;# don't couple langevin bath to hydrogens
>
> # Periodic Boundary conditions
> # NOTE: Do not set the periodic cell basis if you have also
> # specified an .xsc restart file!
> if {1} {
> cellBasisVector1 27.0 0.0 0.0
> cellBasisVector2 0.0 24.0 0.0
> cellBasisVector3 0.0 0.0 54.0
> cellOrigin 2.28 -0.35 16.49
> }
> wrapWater on
> wrapAll on
>
> # PME (for full-system periodic electrostatics)
> if {1} {
> PME yes
> #PMEGridSpacing 1.0
> #manual grid definition
> PMEGridSizeX 27
> PMEGridSizeY 24
> PMEGridSizeZ 54
> }
>
> # Constant Pressure Control (variable volume)
> useGroupPressure yes ;# needed for rigidBonds
> useFlexibleCell no
> useConstantArea no
>
> langevinPiston off
> #langevinPistonTarget 1.01325 ;# in bar -> 1 atm
> #langevinPistonPeriod 100.0
> #langevinPistonDecay 50.0
> #langevinPistonTemp 300
>
> # Output
> binaryoutput no
> dcdfreq 100 ;# 500steps = every 1ps
> outputEnergies 500
>
> #############################################################
> ## EXTRA PARAMETERS ##
> #############################################################
> # Tcl interface
> tclForces on
> tclForcesScript smdforce.tcl
>
> run 10000 ;# 20 ps
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:38 CST