I sent e-mail to dac earlier today, however I am spreading this wider
among the developers. On the reflector there have been some observations
of strange LEaP behavior with solvateBox or solvateOct in AMBER8 with
particular compilers. No one had been able to reproduce. I am now able
to reproduce and the bug is wierd.
Two problems result:
(1) solvate commands place much more water than normal and place it in
overlapping boxes that overlap the solute (if the solute was centered
otherwise the box is at the origin). The boxes are translated in one
direction correctly (say X) but then not translated in Y or Z. If you
look at such a system it will not solvate the system and will definately
blow up in simulation due to waters overlapping the solute (if centered).
The resulting box + water is too big / too many and the density will come
out too low (~0.3 g/cc).
(2) My student Elijah also found that addions adds ions on a line on the
first grid direction; each ~1 A apart. On a DNA, this results in a line
of ions at the first base pair center (or at the origin of the DNA is not
at the origin). (that is addIons both to an unsolvated OR solvated DNA).
My amber8 LEaP was compiled with pgcc 5.1-5. I tried removing all the
options (-O2 -tp px) and still retain the problems. I cannot figure out a
way to avoid this with pgcc (except to use amber7).
The problems miraculously go away with gcc 2.96 (in my experience) and
later versions (as posted on the AMBER reflector).
Any suggestions from the audience on how to debug/proceed?
I hope everyone is peachy.
--tom
Received on Wed Apr 05 2006 - 23:50:00 PDT
