Re: non-repeatability NAMD simulations

From: Jérôme Hénin (jerome.henin_at_uhp-nancy.fr)
Date: Wed May 11 2005 - 03:00:55 CDT

This has already been discussed several times on this list.

See Jim's answer:
http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l/1317.html

Or mine, previously (says essentially the same thing):
http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l/0274.html

If you want reproducible MD runs, the solution is: run on 1 processor.

Jerome

On Wednesday 11 May 2005 00:08, roberte_at_organ.su.se wrote:
> --
> I've build a cluster of 4 computers with different intel cpu's ( and one
> amd 1.8GHz) using Clustermatic 5 ...
> However when running the same NAMD simulation on the machine (2 processors
> or on all 4 ) the results from the simulations differ, even when using the
> same SEED number and starting conditions.. Is this a common problem with
> NAMD, as I have not encountered it with Charmm ( when running on a itanium
> cluster)..
>
> The following input configuration file was used : This is an octasaccharide
> in a 50Å box already minimized and heated but not fully equilibrated.
>
> Has any one else encountered this problem. It would be nice to be able to
> get reproducable results.
> (this configuration file was run on two cpu's twice, resulting in two
> different energy out-puts. They start to drift apart at about step 1000)
> -------------------------------------
>
> #Input
> structure p22_wb.psf
> coordinates p22_wb_min_heat_eq_long.coor
>
> #temperature 300
> velocities p22_wb_min_heat_eq_long.vel
> extendedSystem p22_wb_min_heat_eq_long.xsc
>
> #firsttimestep 100000
>
> #molly on
>
> #parameters
> paraTypeCharmm on
> parameters /home/rob/NAMD_2.5_Linux-i686/namd-tutorial-files/parm/parmsu_01
>.prm exclude scaled1-4
> 1-4scaling 1.0
> dielectric 1.0
>
> #setting parameters for long-range
> switching on
> switchdist 10.0
> cutoff 12.0
> pairlistDist 15.0
> stepspercycle 4
> rigidBonds none
> #rigidTolerance 0.0001
> #rigidIterations 100
>
>
> # Constant Temperature Control
> #langevin on ;# do langevin dynamics
> #langevinDamping 5 ;# damping coefficient (gamma) of 5/ps
> #langevinTemp 300
> #langevinHydrogen off ;# don't couple langevin bath to hydrogens
>
> #PME
> PME on
> #PMETolerance 0.00001
> PMEGridSizeX 50
> PMEGridSizeY 50
> PMEGridSizeZ 50
>
> timestep 1.0
> fullElectFrequency 2
>
> #
> #minimization on
>
> outputenergies 250
> outputtiming 250
> binaryoutput no
> outputname p22_wb_test5_2proc_a
> restartname p22_wb_test5_2proc_a
> restartfreq 100
> binaryrestart yes
> DCDfile p22_wb_test5_2proc_a.dcd
> DCDfreq 250
> #numsteps 2000 # this is the steps for minimization
> outputPressure 250
>
> #MD protocol Block
> seed 2010
> numsteps 2000
> #temperature 300
> rescaleFreq 40
> rescaleTemp 300
>
> useGroupPressure yes ;# needed for rigidBonds
> useFlexibleCell no
> useConstantArea no
>
> BerendsenPressure on
> BerendsenPressureTarget 1.01325
> BerendsenPressureCompressibility 0.0000457
> BerendsenPressureRelaxationTime 100
> BerendsenPressureFreq 10
>
>
> #langevinPiston on
> #langevinPistonTarget 1.01325 ;# in bar -> 1 atm
> #langevinPistonPeriod 100.
> #langevinPistonDecay 50.
> #langevinPistonTemp 300
>
>
> #Boundrary Conditions
> #cellBasisVector1 49.0 0. 0.
> #cellBasisVector2 0. 49.0 0.
> #cellBasisVector3 0. 0. 49.0
> #cellOrigin 0. 0. 0.
> #wrapWater on
> wrapAll on

-- 
Jérôme Hénin
Equipe de Dynamique des Assemblages Membranaires
Université Henri Poincaré / CNRS
Tel : (33) 3 83 68 43 95        Fax : (33) 3 83 68 43 71
http://www.edam.uhp-nancy.fr/

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:39:26 CST