Hi, everyone
I am using fix/deform to run shear simulation about coarse-grain polymer systems. There is one doubt here:
Is the neighbor length needed to be enlarged than the equilibrium one?
Since I update the neighbor list every step, then I believe the small “bin” size won’t be a problem.
Following is from my script, the global cutoff is 1.5.
communicate single cutoff 4.0 vel yes
neighbor 0.2 bin
neigh_modify every 1 delay 0
Thanks for any reply, Cheers
Xinmeng
Hi, everyone
I am using fix/deform to run shear simulation about coarse-grain polymer
systems. There is one doubt here:Is the neighbor length needed to be enlarged than the
equilibrium one?
do you have any problems?
Since I update the neighbor list every step, then I believe the
small "bin" size won't be a problem.
please re-read the documentation. the value of 0.2 is the neighbor
skin and not the bin size. most parameters to the neighbor list
settings are chosen to achieve optimal performance. some details and
explanation are given in the manual.
axel.
Hi, Axel
Really thanks for your reply 
About question 1 : do you have any problems?
When I do the shear simulations, I find that after changing random number in thermostat or change the MPI cores influence the final evolution morphology slightly, so the whole simulation process cannot be totally repeated. I doubt maybe the neighbor list is not stable in different simulation runs.
In fact what I want to be sure is that is there any special care of the neighbor list in fix/deform simulation (the tilted box).
please re-read the documentation. the value of 0.2 is the neighbor skin
Sorry for my misstatement, I clear know the 0.2 is the neighbor list skin to guarantee safe delayed update of neighbor list.
Xinmeng
Hi, Axel
Really thanks for your reply
>>About question 1 : do you have any problems?
When I do the shear simulations, I find that after changing random
number in thermostat or change the MPI cores influence the final evolution
morphology slightly, so the whole simulation process cannot be totally
repeated. I doubt maybe the neighbor list is not stable in different
simulation runs.
this is complete nonsense. MD is based on solving coupled (linear)
partial differential equations and such as system is in its very
nature subject to lyapunov divergence and chaotic behavior. this is
exposed by changing random numbers or changing the number of
processors or changing the frequency of sorting/frequency the
neighborlists builds and so on. so some differences should *always*
appear.
if the divergence _in the averaged results_ is significant, you have
to review your methodology. you may have to run a larger system or
rerun the simulation multiple times (with randomized starting
conditions) and average over the results or document the distribution
(who says that there is only one final state? there may be multiple
equivalent ones). this is just standard MD stuff.
In fact what I want to be sure is that is there any special care of
the neighbor list in fix/deform simulation (the tilted box).
just check your output for messages about dangerous neighborlist builds.
Hi, Axel
Thanks for your advise and solving my puzzle.
Xinmeng