dear hui,
2009/9/4 惠Hui 欧阳Ouyang <[email protected]...>:
Hi Ritwik,
I solved my problem before I got responses from Axel and Steve, so I didn't
try nph yet. In my problem, the problem occurred because I take lj/cut
instead of lj/gromacs. And I solved it by short the cut off distance. In my
i beg to disagree. you didn't _solve_ the problem, you at best side stepped it.
due to changing cutoffs and replacing a hard cutoff with a smooth cutoff, you
reduce the total system momentum generated by your initial conditions.
if you have other initial conditions, you may see the same thing all over again.
also, the total momentum will still build, so you may get back where you were,
just with a longer delay. as steve wrote, the langevin thermostat is
by construction
less sensitive to this effect as it is supposed to dissipate the total
system momentum
through the balance of friction and randomization. if you stick with
nvt integration,
it is _very_ advisable to monitor the total system momentum, since it
_will_ build
up over time in long trajectories and usually does not dissipate with a single
nose-hoover thermostat, thus (infrequently invoking) fix momentum linear is
advisable. in any case, those changes are _not_ transferrable to an all atom
water potential with long range coulomb treatment.
opinion, first check the density of water from your calculations. If the
density is higher than 1000kg/m^3, and it's not reasonable, you can try give
a smaller cut off distance, as slightly smaller than 8.0 in your
lj/cut/coul/long for lj potential, and keep tracking your density. And also
ouch. this is no good advice for a multi-phase system. you can at best
monitor the density in a bulk area of the water phase, and then it has to
match the "natural" density of the water potential which may _not_ be 1g/cc
and then using a cutoff change to adjust the density is in my personal opinion
a very bad idea. the cutoff has to be chosen, that the interactions beyond
the cutoff are negligible, if you reduce the cutoff, the way that you propose,
you would modify the potential and thus the model and the simulation would
be worthless in my book.
I think you need use fix npt first, to relax your system, then switch to nvt
when the density is reasonable and so are the temperature. The last thing
please note, that we no next to nothing about the real starting structure
and the system setup, since this is all hidden in the data file.
cheers,
axel.