Here is another interesting observation:
In the past I've coded routines to calculate torsional energies and
forces and found that unexpected results could occur for calculating
acos of a dot product due to precision. For example, sometimes vectors
which are very nearly parallel can have a unit dot product of something
like -1.0001, which of course throws an error when acos is taken unless
the value is clipped to the range [-1,1].
Given the fact that our minimization was erratic and not entirely
reproducible and knowing that many of our initial pdb configurations had
"perfect" dihedral angles, I decided to perturb all initial pdb phi/psi
angles by a random amount by up to +- 1 degree. As it turns out, this
fixed our problem and we no longer saw amber hanging during minimization.
Might this be due to the minimization procedure itself when symmetries
are perfect along the backbone, or does this sound like an internal
coding problem perhaps akin to the acos example mentioned above? We
haven't seen this problem in amber7 or amber8.
-Scott
M. Scott Shell
Postdoctoral scholar, Dill group
Department of Pharmaceutical Chemistry
University of California San Francisco
M. Scott Shell wrote:
> John Mongan wrote:
>> Can you be more specific about "hangs"? Does the program exit or stay
>> running? Do you continue to see output to the mdout file? If you run
>> top, is it using the maximum processor that it can, or is it
>> stopped/blocked?
>>
> The program doesn't exit. The minout file is one byte in length and
> the minimization program just continues to run. Right now I can't see
> the process load because we're running it in a shared cluster without
> individual node access.
>> Also, what compiler and platform are you using?
>>
> We use "./configure g95" for Xeons running Linux version
> 2.6.16-1.2122_FC5smp (bhcompile.hs20-bc1-3.build.redhat.com) (gcc
> version 4.1.0 20060304 (Red Hat 4.1.0-3)) #1 SMP Sun May 21 15:18:32
> EDT 2006.
>> This seems very curious because igb=7 uses all of the code used by
>> igb=5 (plus a little more for the neck correction).
>>
>>
> After running many additional tests, I've found that the problem is
> not specific to igb=5. It seems to occur regardless of the igb
> setting, although it does not reproducibly occur every time sander is
> run. I've also tried varying the PB radii set, and this doesn't seem
> to have an effect either. This seems very unusual behavior to me; I'm
> wondering if this could be some problem during compilation. Or could
> it be specific to the pdb file which is used as an initial
> configuration, as I've only seen this error when we use a pdb template
> in tleap? The pdb file we're using seems quite reasonable.
>
> Thanks,
> Scott
>
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun Jun 11 2006 - 06:07:15 PDT
