[gmx-users] Wrong potential energies in stochastic dynamics

[reich at mpikg.mpg.de]

Dear Gromacs-users,

I have been simulating a harmonic oscillator (two covalently bound dummy
atoms) using stochastic dynamics. The average potential energy seems to be
fine when I use a time step of 1 fs but when using a time step of 0.1 fs I
observe extremly large potential energies -much to large to get the
correct average potential energy of 1/2kT.
I wonder whether there is a problem with the implemented algorithm when
the timesteps are chosen too small. At least the paper on this algorithm
(Mol. Sim. 1:3, 173-185) states that there is a problem when the product
of time step and friction coefficient gets too small.
Strangely enough the average potential energy gets significantly smaller
when I use a time step of 0.1 fs but masses of about 0.1 amu for each
atom. This I cannot explain since the friction coefficient is not a
function of the mass. I have set t_tau to 400 ps anyway, so I suppose the
friction coefficient would be pretty large anyway (does anyone know how
t_tau relates to the friction coefficient exactly?).

What´s the problem with using small time steps and why can small masses
alleviate this problem?

Best

Luther