Load balancing and fixed boundaries conditions

Dear Lammps users,

I'm using f boundaries during collision cascade to remove automatically the sputtered atoms while they are leaving the simulation box.
But I saw that the atoms are not removed when i'm using "balance rcb" (they keep "moving away") whereas they are removed properly while using other balance options.

Since I didn't see any comment about it in the documentation, is this behavior expected ?

Thanks !

Here is a "quick&dirty " MWE I can propose:

Dear all,

I update my previous post to point the fact that the "bug" described below seems to be partially corrected in the current version (15May15). Now, the atoms leaving the simulation box are properly removed in two directions, but not in the third one.
Since I didn't see a clear mention of that in the lasts patch note, I preferred to mention here.

If anyone have some news about-it or found something really wrong in my MWE, feel free to comment/correct me!


You’re saying the script you posted shows different
behavior for lost atoms thru the fixed boundaries
depending on the “balance” command?

How many procs are you running on?


Actually, what would help ever more is

if you post an input script (similar to the one

you posted) for a problem where if you run

it you get a dump file with an atom clearly

outside the simulation box which has not

been “lost” as you expect it to be. Also

the dump file. Also info on how you ran it,

e.g. on how many procs.

Note that atoms can be slightly outside the

simulation box until a reneighboring occurs,

at which time they should become “lost”

if you are using fixed boundaries.


Good advices!
As attachment, are the lammps script, a snapshot and the screen output.
To ensure that's not a reneighboring issue, I forced the calculation of the neighbor list each step.
The two behaviors can be tested by commenting/uncommenting the "balance" command.

I run this simulation on 8 cpu and 12 cpu with similar results (I didn't tested again, but I first had these issue on a cluster with hundreds of cpu).

Since it might also be a compilation issue on my side, here are what I'm using:
mpirun: openmpi-1.8
executable: lmp_mpi_15May15_MANYBODY (no other package)

Do you need more information?


mwe.lmp (916 Bytes)

mwe.lmp.screen (8.87 KB)

snap.2300.atom (133 KB)

I’m looking at the dump file for step 2300. What is

the problem with it? Is there one or more atoms

far outside the box? All the scaled coords appear

to be between 0.0 and 1.0.


My bad, sorry, wrong file. It was the one generated with the other balance option, not the rcb.
Here is the good one.


snap.2300.atom (134 KB)