Dear Steve and lammps users,
I ran an NEMD simulation to compute thermal conductivity of a CNT using the AIREBO potential. I plotted the instantaneous temperature gradient slope with respect to time for the duration of the simulation. I also ran two restarts from different points in time, and plotted those slope evolutions in the same plot. It turned out that the slope and trajectory would change significantly after the simulation was restarted. The two plots that I am attaching followed the same procedure, and they are just two different replicate runs.
In the initial run, I used the command:
restart 500000 cnt.*.restart
I copied the desired restart file to a new folder and renamed it “cnt.restart”. In the restart input deck, I used the command:
read_restart cnt.restart
I used the same number of processors in all simulations.
Do you have any idea what could cause the trajectories to change so much? Thanks.
Jonathan
slope_evo1.pdf (138 KB)
slope_evo2.pdf (122 KB)
Hello Jonny,
At first you need to specify how do you get the “instantaneous temperature gradient slope with respect to time”. Because you are using NEMD, the slope significantly depends on how you define it. If you count the hot or cold regions or adjacent regions into the slope calculation, it is not a good idea due to the non-equilibrium status.
Second, you also should give more details about your input for restart. When you restart from a restart file, you must be clear about what information is or not stored in it. For example, if you redefine a region and assign with a group name which has been previously used, it will be messed up.
Xiaopeng
Hi Xiaopeng,
I am using the fix heat command to inject and remove heat from predefined source/sink regions. The source/sink regions are selected at the very beginning via the region command to capture the first and last several rings of the CNT. After the region command, I use a group command to pass to “fix heat” as well as the “set” command to redefine the atom types. The instantaneous temperature gradient is a linear least squares line fit to the middle region, but sufficiently far away from the source/sink regions. I have plotted the average temperature with the fit line and it is indeed linear in the region I have selected.
In my restart, instead of using the region command, I define my groups based on the types that were set in the original simulation. And then I proceed in the same manner (fix nve with two fix heat commands).
Thank you.
Jonathan
In general, if you run a long time from a restart, you will not
get identical answers to the previous run. Due to round-off
or other small factors. All that is guaranteed is statistically
identical results, assuming you have setup your 2nd simulation
the same as the first. See secton 9.1 of the manual for
more details.
Steve
Hi Steve,
I can understand if the two are subtly different, but my simulations seemingly show a statistically different result. The time-averaged slope essentially doubles during the restarted runs. Moreover, the slope changes almost immediately, reaching a steady state in just a couple hundred picoseconds – about half a million timesteps, whereas the original run which had been going for several million timesteps could not reach the “new” steady state. Which time-averaged slope should I trust?
Jonathan
Hi Jonny,
I think you don’t need to redefine your group in your restart input. I suggest you read carefully the manual again about what information is stored in the restart file. It is very possible the problem is your restart input. Did you visualize your system? If not, it may also tell you something. You can also attach your two input files and then others can more understand what problem you are in.
Xiaopeng
If you ran simulation A to equilibrium and saved a restart,
then started simulation B from the restart file and run it
to equilibrium, then the 2 simulations should agree statistically.
If they don't then either you didn't run one or both of them
to equilibrium or you are not running the same simulation (i.e.
you did not re-specify all the needed quantities in simulation
B after reading the restart file). There's no other magic involved
with restart files.
Steve
Can Steve or anyone else affirm this issue?