[MMTK] Triple bonds?

Christopher Drost chris.drostie at gmail.com
Wed Jul 13 13:51:42 UTC 2011


On Thu, Jul 7, 2011 at 10:27 AM, Ramon Crehuet <rcrehuet at gmail.com> wrote:
> Dear Christopher,
> Considering the amount of human time vs. computer time, and if the molecules
> are smallish as you said, why don't you go for a QM method? Semi-empirical
> methods can treat prettly large molecules and you will not need to
> parameterize anything.

I might be able to get a friend who has access to the more-expensive
software  to do this for me, since MMTK is getting to be absurdly
difficult to debug.  I have also contemplated spending a week to learn
some of the languages of the non-Python programs. I also have a copy
of GAMESS that I might install and peek around. Still, I'm surprised
that what seems like the most basic molecular-mechanics tasks are
impossible for open-source projects. PyQuante's energy minimization
system is buggy to the point of nonfunctional and would take several
days to debug; MMTK does not believe in error handling and instead
throws Segmentation Faults which I must trace with gdb. The one thing
that has kind-of worked for some purposes is GPAW, but unfortunately
it seems to take ages with 3-atom systems; my 54-atom system would
probably tax it beyond all recognition.

2011/7/7 Konrad Hinsen <research at khinsen.fastmail.net>
> Amber doesn't have the notion of a double or triple bond. The molecule
> topology specifies just a bond, without any parameters. The interactions are
> then specified in terms of atom types. To get a triple bond, you thus have
> to introduce new atom types for the atoms on either side of the bond, and
> then define an appropriate force constant for a bond between the atoms of
> the corresponding types. This would typically be done in the form of a
> "modification file" (in Amber terminology), i.e. an extension to the list of
> standard Amber parameters. MMTK uses standard Amber modification files, so
> you can follow the Amber documentation for the details:
>
>        http://ambermd.org/#ff

When I tried this advice, I obtained another outside-of-Python
segmentation fault. This one is beyond my abilities to debug. The
segmentation fault is issued by line 202 of nonbonded.c, which reads:

    nblist->boxes[box].n++;

The problem is that 'box' is at this time set to -2147483648 (= 0x8000 0000).

For the sake of context, this is the second time that nblist_update is
called from MMTK. The first time it appears to run through the entire
loop perfectly successfully. Both times when it enters the loop n_sub
is 0 and n is 54.

One potential cause of the error is that loop variable 'ai' gets
similarly set to -2147483648 for reasons which I cannot properly
comprehend. (C isn't really my thing, and whenever someone has to type
'gdb python' an angel loses its wings.)

I can send the parameter files and code, etc. that I have constructed
to anybody who is perhaps interested in debugging the problem, but I
don't really have the time, resources, or energy to join MMTK as a
proper contributor. This was supposed to be a quick calculation, not a
code audit.

-- Chris



More information about the mmtk mailing list