Thanks Axel,
Well I tried one of the jobs on a different version of lammps on a different supercomputer and also got the segmentation fault.
I started getting this error when I moved to a new data file which was the same as the old one but just had maybe 3000 more atoms.
And changing the box dimension in the data file can also have an impact. So the error does seem related to reading the data file.
The one thing common to both lammps installations is that they use a custom pair_airebo.cpp
Maybe that is causing difficulty for certain atom numbers/box dimensions
I am surprised that some input/data files work and others give this error even though the changes I make seem trivial
I should also say that I have been running quite a number of jobs without any difficulty up to now (that was with the older data file with somewhat less atoms)
Daniel
Thanks Axel,
Well I tried one of the jobs on a different version of lammps on a different
supercomputer and also got the segmentation fault.
I started getting this error when I moved to a new data file which was the
same as the old one but just had maybe 3000 more atoms.
And changing the box dimension in the data file can also have an impact. So
the error does seem related to reading the data file.
The one thing common to both lammps installations is that they use a custom
pair_airebo.cpp
Maybe that is causing difficulty for certain atom numbers/box dimensions
i don't like to speculate over things that can be easily validated by
some proper tests.
I am surprised that some input/data files work and others give this error
even though the changes I make seem trivial
trivial changes can have lots of bad results. you may not get
segmentation faults, but rather incorrect results from slight data
corruption. i prefer a solid immediate crash over a slight or creeping
data corruption every day.
I should also say that I have been running quite a number of jobs without
any difficulty up to now (that was with the older data file with somewhat
less atoms)
this is all insubstantial. run a test with the current LAMMPS version
and if it still segfaults provide a suitable input deck that reliably
reproduces the crash. if you want this issue to be properly resolved,
that is what you need to do.
axel.
Daniel,
as a first test - have you run your job with non-modified pair_airebo code? Is it the custom pair style that causes the error?
Oleg.
Oleg and Axel,
Thank you both.
I will try it with the non-modified pair_airebo and see what happens
It could be related to the modified pair_airebo but the problem seems to occur while reading the data file
Daniel