Re: AMBER: antechamber, how does it work

From: Gustavo Seabra <>
Date: Wed, 15 Aug 2007 10:15:48 -0400

OK, I'm not an expert in antechamber, but I'll try some of the points here:

> (1) Can divcon be launched selectively with amber (and an important option for
> RESTARTING should a crash occur), or AM1 is what can be launched? It seems to
> me that it is not the same thing.

I believe there's some confusion here. First, note that 'antechamber'
is the name of both a collection of programs, and a script that
executes them in a definite order. DIVCON is one of those programs,
which is called when the antechamber script is required to generate
quantum information itself. This is not a required step, and others
have already suggest to use some external program for this step, in
place of DIVCON.

DIVCON has a number of QM Hamiltonians. AM1 is one of those. So, AM1
is *not* something "to be lauched", but rather one of the QM methods
available in DIVCON, and in many other QM programs. DIVCON just
happens to be the one that is plugged into the antechamber package,
but nothing stops you of using any other, if you wish.

About "Launching with Amber", we need to understand what you mean by
that. SANDER has 2 ways of doing QM/MM: One is using DIVCON, which
which has an interface with SANDER, and the other is using the QM
methods provided natively within SANDER. SANDER provides, natively,
support for MNDO, AM1, PM3, SCC-DFTB and variations of those.

> (2) When launching antechamber as in tutorial B4 (the - more or less - way I am
> using it), is "divide and conquer" at work? I see, during present antechamber
> job, listed in divcon.out statements like "Resetting search direction". What
> does that mean? That guess Hessian is recomputed from the last good
> coordinates? Or does that mean that minimization of my molecule is done by
> formally cutting it into pieces (divide and conquer)? The way I was used at
> when computing charges "ab initio" via HF-6-31G* by formally cutting the
> molecule into residues.

This question is probably better answered by one of the DIVCON's
people, but my guess is that, if it's using DIVCON, it is "divide and
conquer". After all, that's what DIVCON means... But notice that this
is not necessarily related to the calculation of charges. DIVCON is a
method to solve the SCF equations by dividing the molecules into
pieces. The nature of "divide and conquer" probably has nothing to do
with the "resetting search direction" message. This message only means
that whatever vector the minimizer was following is not minimizing the
energy, and it is going to try something else. (But I'd wait for the
DIVCON people to come in here...)

> (3) Amber9 manual, after stating that divcon is called by antechamber (p 74),
> refers to AM1 in the following pages, none of the many options of divcon
> (chapter 7) being mentioned.

Again, that just means they are just using DIVCON to do an AM1 calculation.

> (4) When launching a QMMM procedure from sander.MPI, is the QM part running
> parallel? If so, why not applying parallelization for the AM1 minimization
> within antechamber?

That goes back to the different implementations of QM inside SANDER
(see above). Notice that antechamber calls DIVCON and *not* SANDER
when it needs a minimization. SANDER is not exactly known for it's
great minimization abilities...

Also, I cannot say much about DIVCON, but the native SANDER QM
implementation is not fully parallel yet. The greatest difficulty is
with the parallelization of the eigenvalue solver, which is far from
trivial. This is a problem not only in SANDER, but in most QM programs
out there, and is a field of active research.

> This is a message from a user who has to rely heavily on antechamber, and is
> carrying out a job for a non-polymeric, flexible molecule made of 500 atoms
> with a huge number of chiral centers and many charges. Just the job of creating
> a pdb file was real "work". I omit discussing why I am not using an external
> program parallelized in AM1. Who had ever to prepare files for MD for large,
> flexible non-polymeric molecule will understand why.

I just wanted to add here that, for a molecule this size, DIVCON is
probably one of the best possibilities, since it has been built
exactly with that in mind. I can't tell you much about parallelization
there, though.

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Sun Aug 19 2007 - 06:07:07 PDT
Custom Search