Hi Urzula,
I am going to answer in respect to running explicit solvent PME
calculations here since for GB simulations the considerations are
different and there one should probably not use a cutoff (or use a large
one).
In PME explicit solvent simulations the cutoff value controls two things.
Firstly it controls the distance to which VDW interactions are calculated.
These are short range and so do not need a long cutoff. Second it
determines the division of electrostatic calculations between direct space
and reciprocal space. This division is somewhat arbitrary. The PME
approach effectively gives you infinite electrostatics independent of the
cutoff value - so 'cutoff' is somewhat of a misnomer name wise. The more
work you do in the direct space sum the coarser you can make the
reciprocal space FFT grid for a given level of accuracy. Thus the value of
cut in the electrostatics can effectively be tuned for optimum performance
depending on how well the hardware does inverse square roots in compared
with FFTs. This can vary wildly especially when running in parallel. In
reality, however, one always has to do the VDW interactions and these
typically run to about 8 angstroms before becoming negligent. Thus you
have to do at least 8 angstroms of direct space so things tend to be
optimized for this balance. Thus the typical value used for the cutoff is
around 8 or 9 angstroms and mostly affects the accuracy of the VDW
interactions to a small extent rather than radically affecting the
electrostatic calculation. Increasing the direct space cutoff will
generally decrease performance since the balance for most hardware lies on
the side of FFTs being very fast compared with direct space calculations
thus using a value of say 10 or 12 will typically just slow the
calculation down for very little increase in accuracy, since the
electrostatics are not appreciably affected and the VDW terms have mostly
decayed to zero by that distance. Thus 8 or 9 angstroms is typically
recommended. Reducing the cutoff below 8 angstroms is NOT recommended (and
may actually be trapped in sander and pmemd to stop one doing it without
modifying the code) since then you start truncating the VDW interactions.
In terms of the water box size, this is what is known as the minimum image
convention and is a restriction of the PME algorithm used for the long
range electrostatics. Essentially it is that an atom should NOT see it's
own image (in an adjacent cell in the periodic boundaries) in the direct
space sum. If this is not enforced then you would likely double count
electrostatic interactions which would be bad. The simplest way to avoid
this is to enforce the restriction that the smallest periodic box
dimension must be at least twice the size of the cutoff. Thus this would
typically be 16 angstroms for an 8 angstrom cutoff - although you'd
typically want to build a box bigger than this to allow for reduction in
box size that can occur if you run constant pressure and the density
increases.
That of course is the minimum water box size, in reality you want to use a
bigger box than this especially if your solute is non spherical in shape
and can rotate or unravel during a simulation - e.g. if it rotates it
could stick out of both ends of the box causing other problems. Typically
people use a minimum box buffer (in leap) of about 12 angstroms. This
means that when leap builds your solvent box it places solvent sufficient
to make the minimum distance between the edge of the solvent and any
solute atom 12 angstroms. Larger values may be necessary for non spherical
solutes or cases where you have multiple molecules in your solute etc.
I hope that makes sense and helps. The restrictions in sander and pmemd
should be the same. The 'sanity-checks' for values below 8 angstroms may
be different, however.
All the best
Ross
On 1/18/13 6:02 AM, "Urszula Uciechowska"
<urszula.uciechowska.chem.umu.se> wrote:
>Dear Amber users,
>
>I would like to ask about the cut off value and the water box size in
>PMEMD. I have read that the radius of the water box should be two times
>larger than the value of cutoff?
>What does it mean? Could someone explain that to me?
>What is the lowest cut off value that I can use in PMEMD?
> How it is in case of sander?
>
>best regards
>Urszula
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jan 18 2013 - 07:00:02 PST
