[gmx-users] Possible bug in the temperature calculation from rerun

[Justin Lemkul]

On 9/21/12 8:29 AM, Bastien Loubet wrote:
> Dear gmx users,
>
> We recently got a problem with the rerun feature of mdrun, and we request
> your help in order to help to solve it.
>
> We have run a simulation of a large POPC membrane using the coarse grained
> Martini force fields. From these simulations we obtained a trr trajectory
> file which contains both the position and the velocity of each martini
> particle every 20 ps. From this trajectory we rerun the simulation using the
> -rerun option of the mdrun program but with a different topology for which
> we have set all the bonded and non bonded interaction to zero. Specifically
> we have set all the force constant to zero in the martini_v2.P.itp and the
> martini_v2.0_lipids.itp file, the aim is to calculate the virial due to the
> electrostatic forces alone.
> The problem is the following: from the edr file we get from the rerun we
> extract the mean value of the energies using g_energy. For the edr file
> obtained during the simulation (and not the rerun) we obtained for the
> temperature and the kinetic energy, using also a 20ps time interval:
>
> Energy                      Average   Err.Est.       RMSD  Tot-Drift
> -------------------------------------------------------------------------------
> Temperature                 324.998     0.0049   0.413669 -0.00743801  (K)
> Kinetic En.              1.9525e+06         29    2485.21   -44.6815
> (kJ/mol)
>
> This is to be expected because we have a Nose-Hover temperature coupling of
> 325 K. The kinetic energy is also equals to the number of degree of freedom
> times half the Boltzmann constant times the temperature.
> However for the rerun, with no bonded and non-bonded interactions, we get:
>
> Energy                      Average   Err.Est.       RMSD  Tot-Drift
> -------------------------------------------------------------------------------
> Temperature                  327.35     0.0053   0.415978 -0.00514402  (K)
> Kinetic En.              1.96663e+06         32    2499.08   -30.8973
> (kJ/mol)
>
> The kinetic energy is again equals to the number of degree of freedom times
> half the Boltzmann constant times the temperature, however the temperature
> is 2 K off ! This is surprising as only the velocities and the masses of the
> particle should enter the kinetic energy. The velocities are taken from the
> trr trajectory file and we have checked using the Linux diff utility that
> our topology have the same masses than the original one.
> Another cause can be that we are doing the rerun on a local machine while
> the original simulation was run on a cluster, however we believe that this
> kind of numerical error cannot be responsible for a 2 degree Kelvin mistake.
> Furthermore we have rerun the simulation using the original topology and we
> obtained:
>
> Energy                      Average   Err.Est.       RMSD  Tot-Drift
> -------------------------------------------------------------------------------
> Temperature                 324.998      0.005   0.411827 -0.00761929  (K)
> Kinetic En.              1.9525e+06         30    2474.15   -45.7788
> (kJ/mol)
>
> The slightly different value are certainly due to the previously mentioned
> fact that the reruns were not run on the same machine as the simulation and
> numerical rounding errors. This point out that it is really the topology
> that changes the value of the temperature. However we really have no masses
> differences between the two topologies and the velocities are draw from the
> same trr trajectory file.
>
> Any thought on this problem would be welcome.
>

Another thing to consider is the fact that an .edr file is accurate over all 
simulation steps, while reruns only do calculations present in the frames 
supplied.  If you are using frames every 20 ps, that may only represent less 
than 1% of the total data collected in the simulation, depending on the value of dt.

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================