[gmx-users] R: Md continuation with cpt

Anna Marabotti anna.marabotti at isa.cnr.it
Fri Apr 29 10:45:45 CEST 2011

 Dear gmx-users, Mark and Justin,
thank you for your previous and precious support, but now I have another
doubt. As I told you previously, looking at my trajectory I can see that the
protein is still not fully equilibrated, then I would like to continue the
simulation. However, now if I use tpbconv with -extend and mdrun with -cpi,
I'm forced to use the file .tpr that still contains the erroneous
information about the init_step. On the other hand, if I make a new .tpr
file using an .mdp file with init_step = 0, this will no longer be an exact
continuation (at least I think so). Is there a way to manage this problem? I
see that in tpbconv options there is the flag -nsteps, but I don't know if
(and how) it would be a good idea to use it for this problem.

BTW, I tried to recover my .edr file using the command 
eneconv -e 1.edr 2.edr 3.edr 4.edr -o full.edr -settime
as Mark suggested, and using option 0 for the first file, and "l" to set the
time as the continuation from the end of the previous one, but it didn't
work: I obtained a new .edr file still with the error message "timestep
don't match", and when I transformed it into a .xvg file, there was the same
error as previously: only few points were saved, at 0, 60, 160, 260, 5000,
10000 and 15000 ps (instead of 140, 200, 300 400, 5140, 10140, 15140 ps as
previously). I don't know if I used it incorrectly. However, using the .trr
files and the option mdrun -rerun, I obtained a new .edr file in which I
could obtain a "coarse-grained" (the sampling was every 50000 steps) .edr
file in which I could see that the energy values are stable. Any suggestions
on this point?

Many thanks and best regards

-----Messaggio originale-----
Da: Anna Marabotti [mailto:anna.marabotti at isa.cnr.it] 
Inviato: giovedì 28 aprile 2011 16.55
A: 'gmx-users at gromacs.org'
Oggetto: R: Md continuation with cpt

Dear Mark,
thank you very much for your hint. It's a relief to know that I haven't to
redo the simulation!
>From now on I will set init_step = 0 for my simulations.


Message: 3
Date: Thu, 28 Apr 2011 20:56:00 +1000
From: Mark Abraham <Mark.Abraham at anu.edu.au>
Subject: Re: [gmx-users] R: Md continuation with cpt
To: Discussion list for GROMACS users <gmx-users at gromacs.org>
Message-ID: <4DB947C0.90702 at anu.edu.au>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

On 4/28/2011 5:47 PM, Anna Marabotti wrote:
> Dear Justin,
> thank you for your precious help. Yes, the problem is what you say and
> I was suspecting too (but I was not sure, that's why I was searching for a
> confirmation of my suspects from more expert people). The problem however
> present in the NVT->NPT junction. My .mdp settings were
> NVT:
> tinit = 0
> init_step = 0
> nsteps = 10000
> dt = 0.002
> tinit = 0
> init_step = 20000 (instead of 10000)
> nsteps = 50000
> dt = 0.002
> That's why I had a last NPT step of 70000 instead of 60000, and that's why
> set init_step = 70000 in the production MD.
> What you are saying also explains the anomalous behaviour of the xvg file
> created from the .edr file, in which only few points were saved, therefore
> the graphs created with xmgrace contain only 7 values (at 140, 200, 300
> 5140, 10140, 15140 ps).
> I always set nstep identical to the last step of the previous
> MD because I understood that it was necessary to do it to obtain an exact
> continuation, there is no other special reason.

The label on the step number has nothing to do with the physics. 
Instructions for restarts preserving simulation ensembles are on the 
website. IIRC init_step refers to the label given to the step whose time 
is tinit (see manual 7.3.3), and haphazardly changing only one of them 
might lead to such problems.

>   Since you are saying that
> this is no (longer?) necessary (I'm currently using Gromacs 4.0.7 on this
> machine), I'm very happy to set always nsteps = 0, to avoid future

Do that.

> In the meantime, what I have to do? I had a look at the .xtc files with
> gmxcheck and apparently all is fine, with no error messages, also the
> trajectory seems to be coherent with my expectations. I run g_rms and
> g_gyrate and the behaviour of the systems appears "normal" (apart from the
> fact that the system is not stabilized yet, so I have to continue the
> trajectory). I can provide you (offlist) the .xvg graphs, if you want.
> Based on these data, what do you suggest to do? Do you think that the
> simulation could be considered formally correct (i.e. this is only a
> of erroneous writing of .edr file) or do I have to repeat
> (AAAAAARRRRRRGGGGHHHHH) the whole simulation? I saw in the manual that the
> option -rerun of mdrun command allows to recalculate forces and energies
> from a trajectory file, do you think that I can use it to obtain a new
> file?

If it's just a problem of labelling the frames, then trjcat and eneconv 
have -settime to deal with these kinds of scenarios.


More information about the gromacs.org_gmx-users mailing list