Problem with Ackland Analysis

Dear Lammps users,

We are still working with Lammps 12Apr13. When we perform an ackland analysis with this version of LAMMPS, we obtain the following image for some atomic configuration. BCC atoms are represented in blue and FCC atoms are green.

box-ack-12Apr13.jpg

If we perform the same calculation with a version of LAMMPS 7Dec15 or 11Aug17, the result is drastically different. Many BCC atoms of the structure are erroneously tagged as FCC.

box-ack-7Dec15.jpg

I also computed the atomic environment with the CNA adaptative algorithm available in OVITO. The result obtain with OVITO is coherent with the oldest version of LAMMPS (12 Apr 13). BCC atoms are represented in blue and FCC atoms are green and white atoms are unknown.

box-cna-7Dec15.jpg

With the cross-checking of the results with the CNA adaptative procedure of OVITO, I suspect the Ackland analysis to be erroneous in the newer version of LAMMPS.

I went through the file « compute_ackland_atom.cpp » in the USER-MISC directory and I found some modifications between the versions 12Apr13 and 7Dec15 (or later). However, I did not find mentions of these changes in the webpage "Latest Features and Bug Fixes in LAMMPS » or in the source file.

So, I’m wondering if someone also noticed this drastic difference of result for the Ackland analysis and if there is a bug or a problem with the newer implementation of Ackland.

Regards,

Olivier

Dear Lammps users,

We are still working with Lammps 12Apr13. When we perform an ackland
analysis with this version of LAMMPS, we obtain the following image for
some atomic configuration. BCC atoms are represented in blue and FCC atoms
are green.

​[...]

I went through the file « compute_ackland_atom.cpp » in the USER-MISC
directory and I found some modifications between the versions 12Apr13 and
7Dec15 (or later). However, I did not find mentions of these changes in the
webpage "Latest Features and Bug Fixes in LAMMPS » or in the source file.

​after a little digging in the git repository and the mailing list
archives, i think i have found the origin of the change you noticed in this
contribution posted to the mailing list:

https://sourceforge.net/p/lammps/mailman/message/34258128/​

it comes with some detailed explanations, so please have a look.

So, I’m wondering if someone also noticed this drastic difference of
result for the Ackland analysis and if there is a bug or a problem with the
newer implementation of Ackland.

​that e-mail claims to have removed some issues with the "old"
implementation and the changes were approved by the original contributor of
compute ackland/atom.

i am copying brian, who contributed these changes, in the hope, that he may
be able to comment on his changes and your observations.

and i am also copying alex, the author of ovito in the hope, that he can
provide some insight into what algorithm/implementation his code uses and
how that stacks up with what the code in LAMMPS does.

it would be fairly easy to roll back the changes, but unless there is some
tangible information that can prove what is correct for all cases and where
the problem is at the implementation level, it is difficult to make a
choice here.

axel.

The implementation of the Ackland analysis found in OVITO is, according to my first assessment, is still based on the “old” algorithm version from LAMMPS. The algorithm in OVITO hasn’t changed since the very first program release as far as I can recall. The "n0 < 11” check, for example, mentioned by Brian Barnes in his contribution, is missing in OVITO’s implementation of the Ackland analysis.

-Alex