[gmx-users] Differences between 4.5.5 and 4.6.2-dev?

Tue Apr 23 14:54:30 CEST 2013

Dear Berk, dear mailing list,

On Apr 23, 2013, at 2:18 PM, Berk Hess <gmx3 at hotmail.com> wrote:

> The PME settings you mention won't make any difference.

Thanks for clarification. I was expecting that but they somehow were the best candidates in my view.

> I don't see anything that can explain the differnce.
> But are you sure that the difference is statistically relevant?
> How did you determine sigma?
> There could be long time correlations in your system.

I would be very surprised to find long time correlations as the equilibrium distribution of ions is (almost) flat. In the stationary state the motion of the water will be very slow (I did check that) and "friction time" thus velocity autocorrelation decay time caused by water/ion friction is very short. And that should be identical for both systems.

About statistics: The 128 ions of both species move virtually independently and thus the standard error of the travelled distances is a good measure for the statistical accuracy. But I also made independent runs, also with different ion numbers, and really did a careful statistical analysis of the data. (E.g.: I simulated with different numbers of ions and fitted a of polynomial orders to the conductance. Then performed a chi^2 test to decide which is necessary, and that seemed OK. I do trust my error bars. My profs in my undergrads were exceptionally picky about them.)

> Have you check the temperature? You could be putting a lot of energy into the system.
> To simplify things, you might want to set rvdw=0.9 which removes integration errors
> due to the twin-range cut-off, makes your simulations faster and will have little effect
> on your results.

The temperature is 300.6, target temperature was 300. That should be fine. I did check weaker fields and weaker thermostat coupling. Everything stayed optimally consistent, within 4.5.5, however incompatible with 4.6.2.
I'm rerunning with cutoffs=1.4 now. I should get results by tomorrow. 

Thanks a lot for your help. I'll tell my news tomorrow and meanwhile I have a thesis to write :-).
Cheers
Stefan
> 
> Cheers,
> 
> Berk
> 
> ----------------------------------------
>> From: kessel at icp.uni-stuttgart.de
>> Date: Tue, 23 Apr 2013 10:20:50 +0200
>> To: gmx-users at gromacs.org
>> Subject: [gmx-users] Differences between 4.5.5 and 4.6.2-dev?
>> 
>> Dear Gromacs users,
>> a short disclaimer first: I'm new to using GROMACS and new to doing atomistic resolution modelling. If I'm doing anything very wrong, I'd be very happy to hear.
>> 
>> I'm trying to simulate ion current in a nanopore. My nanopore consists of LJ particles positioned on the surface of a cylinder, that is closed with itself over PBC, thus my system is quasi-infinite. The pore is filled with SPC/E water and (in this particular simulation) 128 NA+ and CL- ions where I'm using the gromos53a6 ion parameters. An electric field is applied in the periodic direction. The current is then the sum of the distances traveled by all ions in a production run divided by the length of the box times +- 1 (depending on the ion species), divided by the simulation time.
>> 
>> I noticed now the following:
>> With GROMACS 4.5.5 and 4.6.2 I obtained different values for the currents; the NA ions travel faster in 4.6.2 while the Cl ions travel faster in 4.5.5. The difference is about 20% in both cases and it is statistically significant (5 or more sigma).
>> 
>> I'm using PME for electrostatics as later a DNA molecule will be added, and the long range nature of electrostatics will most likely be quite important. I am using a twin range cutoff scheme with
>> rlist = 0.9
>> rcoulomb = 0.9
>> rvdw = 1.4
>> and
>> ewald_rtol = 1e-05
>> and the default fourier_spacing (which should be 1.2 nm).
>> According to g_pme_error this choice is not particularly smart (I will do better, I promise) but however should not explain any differences between the two versions.
>> 
>> Comparing the gmxdump output of both tprs i noticed the following differences:
>> 4.6.2 | 4.5.5
>> verlet-buffer-drift = 0.005 | verlet-buffer-drift = 0
>> fourierspacing = 0.12 | fourierspacing = 0
>> dihre-fc = 0 | dihre-fc = 1000
>> 
>> These parameters are my top candidates to explain differences, but I have attached the rest of the production run mdp below.
>> 
>> My 4.5.5 version was the official one compiled on our local supercomputer (by the admins) and my 4.6.2 version is from the git repository, branch release-4.6, last commit 873b98540a47a5727e69342117ab71f8c8b75072. No GPU usage involved. 4.5.5 with "usual" mpi, 4.6.2 with thread-mpi.
>> 
>> Can anybody think of an explanation? My hope would be that some default behaviour has changed between the versions.
>> My short-term strategy is rerunning with a single cutoff of 1.4, an optimal choice of ewald_rtol (tuned with g_pme_error) and hope that the differences disappear. This however will take a while.
>> Cheers and thanks in advance
>> Stefan Kesselheim
>> 
>> 
>> Here is the rest of my mdp file.
>> 
>> define = -DPOSRES_P
>> integrator = md
>> tinit = 0
>> dt = 0.002
>> nsteps = 4000000
>> init_step = 0
>> comm_mode = None
>> nstxout = 0
>> nstvout = 0
>> nstfout = 0
>> nstxtcout = 100
>> nstcheckpoint = 10000
>> nstlog = 10000
>> nstenergy = 1000
>> energygrps = POR SOL NA CL
>> energygrp_excl = POR POR
>> nstlist = 5
>> ns_type = grid
>> pbc = xyz
>> periodic_molecules = yes
>> rlist = 0.9
>> domain-decomposition = yes
>> coulombtype = PME
>> rcoulomb-switch = 0
>> rcoulomb = 0.9
>> epsilon_r = 1
>> epsilon_rf = 1
>> vdwtype = Cut-Off
>> rvdw-switch = 0.
>> rvdw = 1.4
>> DispCorr = EnerPres
>> table-extension = 1
>> energygrp_table =
>> fourier_nx = 0
>> fourier_ny = 0
>> fourier_nz = 0
>> ; EWALD/PME/PPPM parameters
>> pme_order = 4
>> ewald_rtol = 1e-05
>> ewald_geometry = 3d
>> epsilon_surface = 0
>> optimize_fft = no
>> implicit_solvent = No
>> tcoupl = v-rescale
>> tc-grps = Water_and_ions POR
>> tau-t = 5.0 5.0
>> ref-t = 300 300
>> nsttcouple = 1
>> pcoupl = no
>> Pcoupltype = Isotropic
>> tau-p = 1.0
>> compressibility = 4.5e-5
>> ref-p = 1.0
>> gen_vel = yes
>> gen_temp = 300
>> gen_seed = 32293
>> ld_seed = 32293
>> E-x =
>> E-xt =
>> E-y =
>> E-yt =
>> E-z = 1 0.2 0
>> E-zt =
>> constraints = hbonds
>> 
>> 
>> -----------------------------------------------
>> Stefan Kesselheim
>> Institute for Computational Physics
>> Allmandring 3
>> +49 711 685 63630
>> 70184 Stuttgart
>> kessel at icp.uni-stuttgart.de
>> 
>> 
>> 
>> 
>> 
>> --
>> gmx-users mailing list gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/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/Support/Mailing_Lists 		 	   		  --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/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/Support/Mailing_Lists