I have two different versions of the AMBER 11 code that are giving me
different results with the same input files.
Here is what I have:
My "older AMBER 11" is a CVS checkout from February 22nd, 2009. It was
compiled with the Intel compiler version 9.1, version 9.0 of the Intel
MKL libraries and version 1.2.2 of Open MPI.
My "newer AMBER 11" is a CVS checkout from the middle of November,
2009. It was compiled with the Intel compiler version version 11.0,
version 10.2.1.017 of the Intel MKL libraries and version 1.3.2 of Open MPI.
I am doing some TI's on nucleobases in solution to determine their pKa's
(so that I can later determine pKa shifts between the solution phase and
the environment of my system).
I discovered the problem when I tried to do some additional windows
(additional values of lambda) using the newer code for a TI calculation
that I had already done using the older code. The simulations using the
newer code were not stable: I was getting the error message that vlimit
was being exceeded. When I looked at the trajectory I saw that an amino
group on my nucleobase was starting to flail around spontaneously after
about 20 ps (by spontaneously I mean that I could not see anything
colliding with the amino group, the hydrogens just started jumping around).
Here are the endpoints of the TI that was unstable:
lambda = 0: 9-methyladenine
lambda = 1: 9-methyladenine with a hydrogen van der Waals radius at the
N1 position, also the atom types of N1 and the exocyclic amine nitrogen
were changed to give them new van der Waals radii.
The van der Waals radius at the N1 was added using soft core
potentials. The instabilities get worse (vlimit is exceeded earlier in
the simulation) as lambda gets closer to 1.
It seemed strange that only one of my nucleobases would be unstable, so
I tried redoing some of my other older nucleobase calculations with the
newer code. I do see differences in energy between the newer code and
older code. In one case, the simulations produced the same output for
the first 1500 fs or so, then I started to see differences in energy in
the fourth decimal place. By 70 ps, the newer code energy is -8211 and
the older code energy is -8260.
I am using Langevin dynamics (using the same random number seed to
compare the older and newer codes), so I wonder if that isn't masking
small instabilities in my other nucleobase simulations using the newer code.
So, have there been any changes to the TI code that might be causing
this problem? I won't clutter up the listserv with parameter and input
files, but I would be happy to send them to anyone who would like to try
to reproduce this or examine them.
-Hugh Heldenbrand
Graduate Student,
Chemistry Dept.
University of Minnesota
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Jan 28 2010 - 10:30:02 PST
