[gmx-users] restart

[Mark Abraham]

Yujie Wu wrote:
> Hi,
> 
> I am wondering why in gromacs a restarted MD simulation will not reproduce the
> trajectory by a continuous one.
> 
> The test I did was to run a 4-step MD and got the final .gro file. Then ran a
> 2-step MD, generating the tpr using "tpbconv -s -f -e -extend", then ran it and
> got the final .gro file. A diff of the two .gro files tells me that they are
> different even in the third decimal place. The `nstxout', `nstvout', `nstfout',
> and `nstenergy' variables were set to 1. The Gromacs version is 3.3 (single
> precision).
> 
> Any idea where and what the problem is? Thanks in advance.

It isn't clear to me that you have a problem yet - are you attempting to 
reach the same end point with two 2-step simulations as you reach with 
one 4-step simulation?

If so, it sounds like you may have reproduced a problem that I posted 
last year which was never satisfactorily resolved. Another user may also 
have had the problem. See 
http://www.gromacs.org/pipermail/gmx-users/2005-July/016105.html and 
http://www.gromacs.org/pipermail/gmx-users/2005-July/016181.html and 
http://www.gromacs.org/pipermail/gmx-developers/2005-July/001240.html 
and links within them. The short version was that the initial pressure 
on restarts of Berendsen thermostatted simulations was sometimes not the 
same as the pressure at that time in a continuous simulation. There was 
further off-list discussion where I reported that I could do the same 
restart 100+ times on different hardware and gromacs versions and get an 
incorrect initial pressure a few percent of the time.

Try using "gmxcheck -s your.tpr > file" on the various .tpr files and 
diff them to see what differences exist. Also, comparing .gro files 
isn't a definitive test for equivalence, because of their low precision. 
Comparing relevant parts of .edr files using g_energy on each .edr file 
is probably better.

Mark