AMBER Archive (2007)

Subject: Re: Fwd: Re: AMBER: Minimization amber9 segmentation fault

From: David A. Case (case_at_scripps.edu)
Date: Mon Dec 03 2007 - 11:08:48 CST


On Mon, Dec 03, 2007, Francesco Pietra wrote:

> I removed all WAT residues at a distance <=1.5A from POPC residues from the
> membrane already bearing a cavity. Then inserted the protein-complex and
> built the TIP3P box with leap. Then run the minimization again. Starting
> energy was 9 orders of magnitude less that previous run (which had crashed
> because WAT 534 had clashes with POP 24) and gradient and energy decreased
> by only 2 orders of magnitude. Again segmentation fault.

Segmentation faults can be very difficult to diagnose, since there can be many
causes. Ordinarily, sander would not segfault on minimization just because
of bad overlaps.

What I would do:

1. Try a serial minimzation to see if there is some weird parallelization
problem. You are only trying to do a short minimization, so you should reduce
your exposure to other problems.

2. Only do steepest descent minimization with a bad structure. Conjugate
gradient minimization (which by defaults starts after 10 steps,) can take very
large trial steps, which might somehow be violating implicit assumptions about
periodicity that leads to memory corruption (just guessing here, but avoiding
conjugate gradients until you have a better structure is definitely a good
idea.)

3. If that fails, you will probably have to remove the initial bad contacts
and see if that helps.

...dac

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu