Date: Mon, 29 Jul 2002 22:13:16 -0700
From: David Case
Subject: Re: a water jumps

On Mon, Jul 29, 2002, Masaki Tomimoto wrote:
>
> I still have a trouble on MD simulation with an error message, "The
> systen has extended byhond the extent of the virtual box". I prepared a
> protein and a ligand. Solvent waters were added with solvatecap command
> with WATBOX216 in tleap. Radius of the sphere of solvent waters was 28 A,
> whose center was an atom of the ligand. Then, all residues with in 10 A
> from any atom of the ligand and all solvent waters were specified as movable
> residues, all others were fixed with BELLY option. In MD, a water molecule
> close to boundary moved about 13 A to vacuum area in 5 ps. This water
> molecule caused the error, I think. Then, the question is why the water
> molcule moved such a lot. Please give me any clue!

I sounds like one of the waters "evaporated": moved from the condensed
solvent into vacuum; at this point, it will just keep moving in a straight
line in whatever direction it got started, and will indeed move outside
of the virtual box in a short period of time.

The movement of the water is "physical": waters do evaporate from small
droplets. The problem with the virtual box is a limitation in Amber 7:
basically, we did not anticipate how often people would be carrying out
non-periodic simulations where the configuration of the system would change
dramatically over the course of the simulation (in this, one water molecule
is getting very far away from the rest of the atoms in the system.) The
code was written with the assumption that the system would stay inside a
"virtual box" this is (say) 30 Ang. bigger than the intial configuration in
all directions.

[When the "virtual box" idea is valid, there is a substantial improvement in
efficiency in calculating non-bonded interactions. When it is not valid,
the code simply halts, which is admittedly not very helpful....]

In this particular case, you should be able to keep the waters from
evaporating by setting a small "cap force constant" to penalize large
motions of waters (see section 5.6.13 of the Users' Manual).

However, in a way this is just a band-aid. Amber is a code that reflects
the types of research carried out by its developers. Most of the current
developers don't use "cap" or "droplet" simulations, in which a sphere
(droplet) of solvent waters is used to represent solvation. If we did want
to carry out simulations of this sort, more work would go into implementing
and testing stochastic boundary potentials and reaction field models, which
are required to make these models reasonably realistic. [There is some work
along these lines going on, which may show up in future releases.]

Currently, the things Amber is "good at" are periodic solvated simulations
(using PME for long-range electrostatics) and non-periodic generalized Born
simulations (where solvation effects are treated in a continuum fashion.)
Other non-periodic simulations are possible, but have more limitations (such
as the lack of a non-periodic reaction-field potential, the virtual box
limitation, and the absence of a smooth truncation scheme for non-periodic
simulations.) These limitations don't necessarily invalidate the types of
simulation you are describing, but they should be kept in mind. And we are
receptive to proposals from the community to help add some needed
functionality in this area.

...hope this helps, and good luck...dac