Re: AMBER: Binary file curiosities

From: Thomas Cheatham III <tec3.utah.edu>
Date: Fri, 3 Aug 2007 17:32:55 -0600 (Mountain Daylight Time)

> I have been playing with reading the various binary AMBER trajectory data
> types lately, and I noticed a few things that strike me as puzzling
> "features" of AMBER. For instance, from what I can tell from mdwrit.f, if
> you choose the option NTRX=1 in sander you get an unformatted file that
> looks exactly like the formatted mdcrd file. However, I do not see the

NTRX is for specifying the format of the "refc" coordinates that are read
in so will not alter output. For output, you would play with NTXO.

The binary formats in AMBER are old; they are not transferable among
different hardware (for example SGI vs. Linux) meaning that in general you
need to re-read them on the platform that produced them to be safe using
the same compiler.

It would not be too hard to adapt ptraj to read some of these (although I
haven't seen the need except perhaps to provide greater precision plus
there is the added difficulty that these are fortran implementation
dependent). Instead we went the NetCDF route...

> option to write these types of files from ptraj, only binpos binary files
> are available (from my inspection of trajectory.c). Also, possibly due to
> my lack of c expertise, I do not see the option to read binary mdcrd files
> within ptraj itself. So my questions are the following: 1) Why such
> inconsistencies with the binary file types across the various sander
> modules? I understand that you must try to be as backward compatible as
> possible, but consistency between file reads and writes would make things
> much easier for the users (including myself). 2) Is the magic number in the

The older binary format is not "transferable" and Fortran-implementation
specific. To my knowledge, very few people use it (except perhaps with
NMODE restrt files).

> binpos file type only used only to determine the difference between binpos
> and CHARMM trajectories? 3) Has anyone worked with reading binpos files in
> fortran, particularly using direct access reads? If so, can you please

DAC may have more to say on this regarding the binpos format, but the
ptraj parser for file types if fairly simple in that it looks at the top
of the files and makes some assumptions based on what is seen.

> provide any examples? 4) Is there any documentation anywhere on the binpos
> file type? I have been looking all over, and only believe I know what is in
> the file because of the "od" command and code/comments inside ptraj.
> Please, if I am missing an obvious reference to the binpos file structure,
> do not hesitate to send it my way.

If I were to play with binary formats, I would opt for the NetCDF format
which is supported in current releases of ptraj/AMBER. The binpos is a
simple format but may depend on endian order (and hence not be
transferable among machines?). It will be the most compact. NetCDF is
less compact but more transferable.

-- tom
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun Aug 05 2007 - 06:08:00 PDT
Custom Search