[gmx-users] box size changing during isotropic pressure coupling
chris.neale at utoronto.ca
Thu Jul 24 20:51:13 CEST 2008
Thanks for the assistance Berk,
>Hmm... this is a (seemingly( really systematic decrease in x and y and increase in z.
>But is x at the start really EXACTLY identical to y?
No. Here are the starting dimensions:
19.58000 19.39000 21.04200
That is prior to 200ps run with position restraints on the DPC. The box dimensions during that position-restrained
section are not included in the plot that I made available online. When the unrestrained section began, the box was:
19.370279 19.182320 20.816620
which may be slightly different ratios, but does not show any systematic difference between X/Y and Z
cneale at cneale-desktop:~$ echo 19.58/19.370279|bc -l
cneale at cneale-desktop:~$ echo 19.39/19.18232|bc -l
cneale at cneale-desktop:~$ echo 21.042/20.816620|bc -l
>Because in that case I would think they should always stay identical.
>But looking at the code I see that it is always uses triclinic scaling.
>Thus x and y are treated differently (although adding only zeros),
>so depending on the compiler, the results could differ by a bit.
>What system and compiler are you using?
I am using a variety of systems and compilers. As you can imagine, it takes some maneuvering to get 150 ns for a system of 750,000 atoms.
Most of this simulation was carried out using gromacs 3.3 on 2 x Opterons running HP Linux XC 3.1. It was compiled with:
PathScale EKOPath(TM) Compiler Suite: Version 2.2.1
Built on: 2005-08-05 13:34:17 -0700
Thread model: posix
GNU gcc version 3.3.1 (PathScale 2.2.1 driver)
Other sections of this same run were simulated using gromacs-3.3.1 compiled using a gcc 3.x compiler on a 2 x opteron cluster
running SUSE Linux
I have also some results from a system that is a triclinic box that I have posted to the same webpage:
below the previously mentioned images.
More information about the gromacs.org_gmx-users