forked from lijiext/lammps
f6096c899a | ||
---|---|---|
.. | ||
README | ||
in.heat | ||
in.heatflux | ||
in.langevin | ||
in.mp | ||
log.heat.12Sep13 | ||
log.heatflux.12Sep13 | ||
log.langevin.12Sep13 | ||
log.mp.12Sep13 | ||
profile.heat.12Sep13 | ||
profile.heatflux.12Sep13 | ||
profile.langevin.12Sep13 | ||
profile.mp.12Sep13 |
README
This directory has 4 scripts that compute the thermal conductivity (kappa) of a Lennard-Jones fluid using 4 different methods. See the discussion in Section 6.20 of the manual for an overview of the methods and pointers to doc pages for the commands which implement them. Citations for the various methods can also be found in the manaul. These scripts are provided for illustration purposes. No guarantee is made that the systems are fully equilibrated or that the runs are long enough to generate good statistics and highly accurate results. ------------- These are the 4 methods for computing thermal conductivity. The first 3 are non-equilibrium methods; the last is an equilibrium method. in.langevin = thermostat 2 regions at different temperatures via fix langevin in.heat = add/subtract energy to 2 regions via fix heat in.mp = use fix thermal/conductivity and the Muller-Plathe method in.heatflux = use compute heat/flux and the Green-Kubo method The NEMD systems have 8000 atoms with a box length 2x larger in z, the non-equilibrium direction. The G-K system has 4000 atoms and a cubic box; it also needs to be run longer to generate good statistics. The scripts were all run on 8 processors. They all run in a minute or so and produce the accompanying log files and profile files (for temperature or heat-flux). The state point of the LJ fluid is rho* = 0.6, T* = 1.35, and Rcut = 2.5 sigma. This was chosen to agree with a 1986 paper by D Evans in Phys Rev A, 34, p 1449, where he computed the thermal conductivity of a small 108-atom system using a thermostatting method. Fig 1 in the paper shows his simulations produced a kappa of around 3.4 for this system, in agreement with an experimental data point as well. ------------- Here is how to extract Kappa from the log file output for each method. The NEMD methods use the formula kappa = dQ * dZ/dTemp where dQ = energy flux, and dTemp/dZ = temperature gradient. (1) in.langevin dQ = 8000 * 0.5*(0.905+0.947) / 100 / 18.82^2 / 2 8000 atoms 0.5*(0.905+0.947) = from log file = ave of total in/out energy for 2 regions normalized by # of atoms 100 = 20,000 steps at 0.005 tau timestep = run time in tau xy box area = 18.82^2 divide by 2 since energy flux goes in 2 directions due to periodic z dTemp = 0.578 from log file for average Temp difference between 2 regions dZ = 18.82 Kappa = 3.41 (2) in.heat dQ = (100*100) / 100 / 18.82^2 / 2 100*100 = 100 (time in tau) * 100 (energy delta specified in fix heat) 100 = 20,000 steps at 0.005 tau timestep = run time in tau xy box area = 18.82^2 divide by 2 since energy flux goes in 2 directions due to periodic z dTemp = 0.783 from log file for average Temp difference between 2 regions dZ = 18.82 Kappa = 3.39 (3) in.mp dQ = 15087 / 100 / 18.82^2 / 2 15087 = cummulative delta energy, tallied by fix thermal/conductivity 100 = 20,000 steps at 0.005 tau timestep = run time in tau xy box area = 18.82^2 divide by 2 since energy flux goes in 2 directions due to periodic z dTemp = 1.16 from log file for average Temp difference between 2 regions dZ = 18.82 Kappa = 3.45 (4) in.heatflux kappa is computed directly within the script, by performing a time integration of the formulas discussed on the compute heat/flux doc page - the resulting value prints at the end of the run and is in the log file Kappa = 3.78