Closed anhi closed 1 year ago
Thanks for reporting, I'm tracking down the problem.
MWE:
l = SVector{2, Float64}[[0.0, 0.0], [-1, 0.0]]
unitcell = [14.01, 14.02]
neighborlist(l, 5.0; unitcell=unitcell)
This is a regression, the bug is not present in v0.8.0.
The issue is a combination of the exact particle coordinates being [0.0, 0.0]
and the way the computing cells are filled up, a particle is getting to be copied into two computing cells because of falling at the exact boundary.
The fix will be released asap.
I will release a new version 0.8.15 which will be identical to 0.8.10, which was fine.
The bug was introduced in 0.8.11 and the fix is not that easy, I need some time.
I recommend using version 0.8.10 for now if this affects you, until version 0.8.15 and greater are released.
The issue will be solved in version 0.8.15 that will be released at any moment.
However, the fix includes performance regression, which I will have to deal with later, requiring some careful thought about the general improvements introduced in version 0.8.11, in particular.
Thus, I am not closing the issue now, but changing its label from "bug" to "enhancement".
fixed in 0.8.16 to be released at any moment. I thank you again for reporting, and please let me know of any other issue you face using the package.
Hi,
I've noticed that sometimes, the results of neighborlist are non-unique: