srp with triclinic box

Hellow LAMMPS users,

Using the srp potential I noticed something unusual. When using a triclinic box I started encountering the following error message “Communication cutoff too small for fix srp. Need 2.155384, current 0.15625”. This was happening despite the fact that I specified “comm_modify mode single cutoff 5 vel yes” (i.e. cutoff of 5 not 0.037161). To figure out why that was happening I made a small test script which I could run on a single processor and I debugged it. For some reason when you switch to a triclinic box in your data file (“0 0 0 xy xz yz”) the ghost cutoff you originally specified gets rescaled as comm->cutghost[0]/Lx, comm->cutghost[0]/Ly, comm->cutghost[0]/Lz. I was wondering why is this happening? It seems like a bug to me but I was hoping someone can confirm this.

I am running the latest stable LAMMPS version, lammps-16Mar18. It is compiled with MISC and USER-MISC packages. I have attached the input script and data file to this email. The data file is called “Chains.txt” and the input script is “input.txt”.

Chains.txt (4.7 KB)

input.txt (2.8 KB)

Made some typos in previous email: ghost cutoff you originally specified gets rescaled as comm->cutghost[0]/Lx, comm->cutghost[1]/Ly, comm->cutghost[2]/Lz

Hellow LAMMPS users,

Using the srp potential I noticed something unusual. When using a triclinic
box I started encountering the following error message "Communication cutoff
too small for fix srp. Need 2.155384, current 0.15625". This was happening
despite the fact that I specified "comm_modify mode single cutoff 5 vel yes"
(i.e. cutoff of 5 not 0.037161). To figure out why that was happening I made
a small test script which I could run on a single processor and I debugged
it. For some reason when you switch to a triclinic box in your data file ("0
0 0 xy xz yz") the ghost cutoff you originally specified gets rescaled as
comm->cutghost[0]/Lx, comm->cutghost[0]/Ly, comm->cutghost[0]/Lz. I was
wondering why is this happening? It seems like a bug to me but I was hoping
someone can confirm this.

no, this is not a bug, but exchange of ghost atoms (and reneighboring)
for triclinic cells is done by switching temporarily to fractional
coordinates so it can utilize the same code as for orthogonal cells.

so the issue would rather be with fix srp. i'll take a closer look.

axel.

John,

a bugfix for the issue is now available on github as pull request #856
https://github.com/lammps/lammps/pull/856
this should make it into the next LAMMPS patch release, which is
supposed to be released "soon".

Axel.