From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Thu Jan 05 2012 - 16:48:47 CST
On Thu, Jan 5, 2012 at 4:42 PM, <gaborekt_at_duq.edu> wrote:
> I have been trying to perform NVE simulations of a peptide with a fixed
> conformation in aqueous solutions and have been having problems... The
> basic problem is that the energy is not remaining constant during my
> simulations. Here are some more details on what I've been doing:
> What I'm doing is simulating a small peptide in aqueous solution and
> holding the backbone atoms of the peptide fixed during the simulation so
> that it remains in the same conformation. I initially ran these systems
> using the NPT ensemble for 140 ns... during this time, I examined the
> potential and kinetic energies of the systems, and noticed that they would
> change rapidly during the beginning of the simulation and then quickly
> plateau to a given value and remain there for the majority of the
> simulation... I saved the atomic coordinates and velocities from these NPT
> simulations and used them to begin my NVE simulations, where I turned off
> the pressure and temperature controls, and ran the simulation for ~70 ns.
> I have also kept the backbone atoms of the peptide fixed for these NVE
> simulations. What I've noticed is that the total energy of the system
> seems to gradually increase (along with the temperature) throughout the
> entire simulation... For example, the total energy in one of my systems
> increased from approximately -17,000 to -14,000 kcal/mol over the course
> of the 70 ns simulation. These simulations all used periodic boundary
> conditions and PME, and I haven't noticed anything strange happening when
> I visually inspect the trajectory from these simulations.
> If anybody has any idea why this is happening, any help would be appreciated.
this is mostly due to the numerical integration of the equations
of motions using finite time steps. this is a basic property of
doing classical MD. since errors due to finite time steps will
increase energy. you can rationalize this easily
when you consider a harmonic oscillator.
the second contributions are approximations in your forces,
e.g. due to multi-timestepping, due to using floating point
numbers with limited accuracy (and thus truncation errors),
due to not fully converged SHAKE/RATTLE and so on.
to compensate for this intrinsic error is the primary reason
why people started using thermostat algorithms in MD.
a small amount of energy gain (or energy conservation),
is one of the "quality" criteria that people apply to MD,
and in general one should test settings in an input file
(e.g. time step) to get a sufficiently well energy conservation.
due to using thermostats, many people are not aware
of *how* aggressive their time step or other run time
parameter choices are.
it is a bit surprising that you are surprised by this.
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 College of Science and Technology Temple University, Philadelphia PA, USA.
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:06 CST