I am trying to measure instantaneous temperature profile along the z axis of an elongated box (x=y=45 and z=180 A) of 4000 water molecules divided into 100 equal slabs.
Assuming each slab to be a separate NVT box, temperature of each slab should be almost equal (linear profile with zero slope) at certain time step. However, I see quite large fluctuations in temperature from slab to slab along the z axis.
Increasing number of water molecules and temperature reassignment frequency of the thermostat did not solve the problem.
Should changing the thermostating algorithm (in this case , velocities were take from Maxwell-Boltzmann distribution) or strength of coupling effect the profile?
How can I reduce fluctuations?
Hi, J.,
If you mean, that the temperature T of a given slub is defined as kB T = 2/3 Ek / N, where Ek is the instantaneous kinetic energy of the particles in the slub, why shouldn't it fluctuate? That's quite OK, if it does, since N is finite. What you could do is to take the time average of the instantaneous temperature as the measure of temperature in the slub.
R.
Roman,
You are right, averaging over time steps gives a linear profile and temperature of each slab converge to the temperature of the overall system.
But i want to measure the profile in a certain time (say t ) during an NVT run.
But because the density is uniform, the instantaneous number of atoms in each slab N at the time t , is almost equal and the only parameter which effect the formula you mentioned is the instantaneous kinetic energy or average velocity of atoms in the slab.
The point is why the instantaneous kinetic energy of slabs differ from slab to slab at a given time step t?
Even if one increase the number of atoms or reduce the number of slabs to remove finite size effects.
J.,
I don't know, what is your idea behind looking at the "instantaneous" profiles, but you could perhaps look at the time average of instantaneous temperature, and define your profile as T+err, where err is the finite size fluctuation. If it makes any physical significance, you could also smooth the profile using linear least squares fit for instance, or "moving frame" of the slub.
Hi Roman,
In effect, i have suddenly imposed both ends of the box (initially at T0) to a hot bath and want to measure the system reaction to the perturbation in the boundaries. Initially the profile should be linear with zero slope(-------------) and then it become quadratic.
The question i posted is concerned with the initial instance of the system.
linear least squares fit is a good idea, but first i wanted to have a smoother profile then use data regression.
Any way , your comments was very helpful.
- Badges
- Science topic
- Similar topics
- Chemistry
- Theoretical Chemistry
- Computational Chemistry
- Molecular Dynamics