[gmx-users] simulation problems

[baloilgiullare]

Hello gmx users

I am simulating a system composed of of an organic molecule
(a 10 -CH=CH- thread with polyciclyc ends) in CH3CN. The system is made
up of 6000 atoms in a cubic box. I am using the gromos the 96 force field.
The solute molecule is elongated like : A---------B where A and B are 
the ends
but (via h-bond) can fold like:
--A-B--
|     |
-------
(sorry for poor ASCII art but I hope it helps).
Since I want to study
the kinetics of this process I have performed several simulations of the
system starting from different staring conditions. The problem I have
is that some of simulations (4 out of 15) crash due to problems with
lincs. The strange thing is that this happens after the simulation has
go on for quite a long time (never less than 5 ns) so I do not think
that it is related to the equilibration at the start of the simulation.
The initial lincs warning always concerns different bonds and the
energies some 1 ps intervals before the crash seem not to show
"patological" behavior.
I would discard machine and/or compiler problems too, since i have used
two different gromacs versions (3.3.2 and 3.3.3) and both gave the same
problem. Both versions were tested with the gmxtest suite using
gcc-3.4.6. The simulations were run of 1 to 4 cpus on a single AMD
Opteron node.

Here is my mdp file:

cpp                      = /lib/cpp
include                  =
define                   =

integrator               = md
tinit                    = 0000.0
dt                       = 0.002
nsteps                   = 100000
nstcomm                  = 1
comm-grps                = System
comm_mode                = Linear

emtol                    = 100
emstep                   = 0.01
niter                    = 0
nstcgsteep               = 1000

nstxout                  = 2500
nstvout                  = 2500
nstfout                  = 0
nstlog                   = 2500
nstenergy                = 250
nstxtcout                = 250
xtc_precision            = 1000
xtc_grps                 = System
; Selection of energy groups =
energygrps               = System

nstlist                  = 5
ns_type                  = grid
pbc                      = xyz
rlist                    = 1.1
domain-decomposition     = no

coulombtype              = PME
rcoulomb_switch          = 0
rcoulomb                 = 1.00
epsilon_r                = 1
vdw_type                 = Cut-off
rvdw_switch              = 0
rvdw                     = 1.00
DispCorr                 = no
fourierspacing           = 0.12
fourier_nx               =
fourier_ny               =
fourier_nz               =
pme_order                = 4
ewald_rtol               = 1e-05
epsilon_surface          = 0
optimize_fft             = no

tcoupl                   = Berendsen
tc-grps                  = System
tau_t                    = 0.1
ref_t                    = 300.
Pcoupl                   = No
Pcoupltype               = Isotropic
tau_p                    = 5.0
compressibility          = 11.2e-5
ref_p                    = 1.0


annealing                = no
annealing_npoints        =
annealing_time           =
annealing_temp           =

gen_vel                  = no
gen_temp                 = 300
gen_seed                 = 135907

constraints              = all-bonds

constraint-algorithm     = Lincs
unconstrained-start      = yes
Shake-SOR                = no
shake-tol                = 0.0001
lincs_order              = 4
lincs-iter               = 1
lincs_warnangle          = 30
morse                    = no

Any ideas? Should I keep separated thermostats for solute and solvent?
Even increasing lincs-order to 2 did not help.

Greetings,
giordano