[gmx-users] Long-term drift, crash

[BON Michael]

Hi,

I have been struggling for some time with a nasty problem with my simulations of RNA in water
using Gromacs 3.3.1 and the AMBER ports. Nobody ever reported it so I am afraid I did some
obvious mistake that I just cannot see.
After some steps (between 50000 and 150000), my simulations crash, and I hardly found a good
reason for it : the only clue I found is given by a MSD analysis of the RNA (with g_msd) : the diffusion, 
theoretically linear over time, is disturbed by sinusoidal-like noise whose magnitude increase
over time. Instead of <x²> = 6Dt, I see something that looks like 6Dt + a*t*sin(2*Pi*t/T), where
T is generally around 5 ps. After some time, the magnitude of this "noise" becomes too big and
the simulation crashes.

All the energies look normal, even just before crash. 

I tried to change a lot of things but I still face the same problem on Mac OS X, Ithanium
cluster, simple precision, double precision, AMBER99, AMBER96, TIP3P, SPC, TIP4P, TIP4PEW

Here is the standard .mdp file I am using (I have the impression I use the same parameters as in
most published papers): 

constraints = all-bonds
constraint_algorithm = lincs
lincs_order = 4 ( or 6)
integrator = md
dt = 0.002
nsteps = 200000
comm_mode = Angular (or none)
nstcomm = 25
coulomb_type = PME (or cut-off)
fourierspacing = 0.12
optimize_fft = yes
vdwtype = cut-off (or switch)
rcoulomb = 1.2 (or anything else)
rvdw = 1.2 (or anything else)
rlist = 1.2 (or anything else)
nstlist = 1
nstype = grid
pbc = xyz

tcoupl = berendsen
tcgrps = RNA Other
tau_t = 0.5 0.5 (or lesser values)
ref_t = 300 300
pcoupl = no;

gen_vel = yes
gen_temp = 300

Please ask for any additional information if you have an idea of where this problem comes from.
Many thanks,

Michaël Bon
-------------- next part --------------
A non-text attachment was scrubbed...
Name: winmail.dat
Type: application/ms-tnef
Size: 3531 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20070814/33ff6a8e/attachment.bin>