AMBER Archive (2006)

Subject: Re: AMBER: Ewald and ibelly

From: David A. Case (
Date: Mon Mar 20 2006 - 00:26:17 CST

On Sun, Mar 19, 2006, Austin B. Yongye wrote:

> I ran some periodic boundary, constant pressure simulations with both
> Ewald (pme) and ibelly options turned on.
> The output file has this warning section:
> Warning: Although EWALD will work with belly
> (for equilibration), it is not strictly correct!
> Besides this, my Ewald error estimates are of the order of a tenth of a kcal.
> However, when I use the restraint option (ntr=1) to restrain certain
> residues in my molecule and with use_pme=1, I get reasonable error
> estimates (~10^-4).
> I would like to know if there is some literature as to why this happens.

If you really want to understand this, you will need to spend time with
the references cited in the Ewald section of the Users' Manual; or start
with the discussion of PME in the J Comput. Chem. paper of last December
that gives an overview of the Amber programs.

Briefly: the "belly" option artificially prevents certain atoms from moving,
basically by have a "Maxwell daemon" come in and zero out their forces once
the regular force calculation is completed. Unlike SHAKE, which applies a
dynamic theory of constraints to certain bonds, this daemon violates physical
principles and Newton's laws of motion. Since it is so artificial, there is
no good way to correctly compute the system pressure (which has a large
kinetic energy component, some of which is not correct for frozen atoms.) If
the belly is fairly small, the errors introduced are often also small, but the
"not strictly correct" warning is valid.

The "error estimate" (described in the papers alluded to above) involves
calculations of the virial (another part of pressure estimation). These are
also artificially altered by the belly assumptions. This doesn't mean that
the PME is inaccurate, but it does mean (as you noted) that the estimates used
to monitor its accuracy won't work.


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to