Hi all,
i know the difference in the way these two commands compute the potentials. But i cannot understand if for example we use the same cuttof distance in both cases. I mean ok the first is for short range coulombic and the 2nd for long range but with a same cutoff distance?
Regards
Nelson
Hi all,
i know the difference in the way these two commands compute the potentials.
But i cannot understand if for example we use the same cuttof distance in
both cases. I mean ok the first is for short range coulombic and the 2nd for
long range but with a same cutoff distance?
nelson,
that depends on what kind of error you are willing to accept.
if you use coul/long you also have to add a kspace style to
compensate for the coulomb contribution you dampen out
in the real space part with coul/long. with coul/cut you don't
do that and thus it will run much faster, but your coulobm potential
energy will still be finite at the cutoff radius. to get the same
kind of accuracy you would have to use a much, _much_
larger cutoff for coulomb than for lennard jones.
sometimes people don't care and then coul/cut is faster.
also if you have a small cluster, then using coul/cut is the
best choice, provided the cutoff is long enough to include
all charges in all interactions. this way you have no artificial
periodicity imposed on you.
HTH,
   axel.
Thank you for your answer. And how someone can decide if he is willing to accept this error? I mean without the kspace and coul/long it is much much faster.i was thinking that if you gain in accurancy with coul/long is it worth that it is much much slower?
that is a very good question and the answer to that is: it depends
if you cut off a potential, if there is still significant interaction, you
will generate artefacts in your structure. taking the example of
sodium chloride in water, you would see that in the g(r) around
sodium and chloride there will be an anomaly around the cutoff
distance. charged sites that are inside the cutoff will feel a
significant force toward the sodium as soon as they are beyond
the cutoff, it is gone as a consequence, there will be a dip in
the probability density.
now this can be to some degree alleviated by using a smoothing
function (e.g. coul/charmm or coul/gromacs) and not only shifts
the potential energy but also the force. and for homogeneous systems
that may be good enough for the most part. however, if your system
is inhomogeneous (e.g. a water film on a metal surface, a lipid bilayer,
a surfactant vesicle), then you will incur serious artefacts.
our group has tried and tested many options when studying ionic
surfactants with coarse grain models, bit so far none of the options
without kspace worked to our satisfaction.
i am currently experimenting with different options to reduce the
impact of the computational cost of kspace. one way is to tune
the coulomb cutoff in the coul/long part. if the kspace takes too
much time, it may be faster to reduce its impact and then use
alternate parallelization options to make up for the addtional
computation in real space. please see the post on that subject
that i'll be sending to the list shortly.
the only option that you have is to make careful tests and
decide on what is acceptable to you (and to reviewers of
your publications, and _those_ often demand proper handling
of long-range electrostatics unless you can prove beyond
any doubt that it is not needed, which is very difficult).
cheers,
   axel.
thank you again for your answer. You mentioned something about charmm and gromacs and smoothing Meaning?
For charmm fields there are different coefficients which we use that with the usual lj? Are these accessible for various molecules or only for water?Because i had tried to do it for CO2 and CO but nothing…
Nelson
thank you again for your answer. You mentioned something about charmm and
gromacs and smoothing Meaning?
they both implement functionality, that makes the forces go smoothly
to zero at the cutoff. both options are available for lammps.
this is documented in the lammps manual and the corresponding
publications for the respective force fields.
For charmm fields there are different coefficients which we use that with
the usual lj? Are these accessible for various molecules or only for
this question makes no sense to me. the cutoff for a forcefield is
typically a global value and has little to do with the coefficients.
for coulomb there are no coefficients. you assign (partial) charges
to each site through the force field.
water?Because i had tried to do it for CO2 and CO but nothing..
i don't understand what you want to say here.
axel.
In order to use charmm force field with coul/long or coul/charm we need some coefficients. For water someone can find them easily, but for others like co2 or co it’s difficult that’s what i meant…
you have to search the literature or parametrize the interactions yourself.
that is the nature of the game. for adding new interaction parameter
sets to CHARMM there is a specific "recipe" the the corresponding
publications by alexander mackerell.
there have been studies in the past. it all depends on what you
want to learn from your simulations and what thermodynamic
state your molecules have to be in.
you have to realize that molecular dynamics simulations have
been around for quite a while now. if something is easy to do,
it has already been done. sorry.
axel.
Dear Dr Axel ,
i run some simulations the first one with lj/cut/coul/long 10 and the other with lj/cut/coul/cut 10 60 . I compute several things lke msd and conductivity and i found them in a very good agreement (below 2%). This means that from practical view i found something equivalent with the “physical” system that contains the long range forces or it is the same?
this is impossible to tell from remote without knowing many details
of what you did exactly. also, the effects of long-range electrostatics
can be very subtle and not show in "simple" parameters like g(r) or MSD.
you'd have to look at some dielectric properties like the size depended
kirkwood g factor G_k(R) to get a feeling for this.
here is a collection of papers that i used when writing my phd thesis:
P. MADDEN, D. KIVELSON. A consistent molecular treatment of
dielectric phenomena. Adv. Chem. Phys., 56, 467–567, 1984.
[55] M. NEUMANN. Dipole moment fluctuation formulas in computer
simulations of polar systems. Mol. Phys., 50(4), 841–858, 1983.
[56] M. NEUMANN, O. STEINHAUSER. The influence of boundary conditions
used in machine simulations on the structure of polar systems.
Mol. Phys., 39(2), 437–454, 1980.
[57] M. NEUMANN, O. STEINHAUSER. On the calculations of the dielectric
constant using the Ewald-Kornfeld tensor. Chem. Phys. Lett.,
95(4,5), 417–422, 1983.
i am certain that there are additional, newer publications on these issues.
i also suggest (also with reference to your other discussions) you find
somebody nearby that has experience in MD and meet and discuss
with this person. a mailing list is no substitute for good counseling.
cheers,
    axel.