Using 12-bit tables for long-range coulomb

Hi,

I have been trying to run the 14May16 version (latest stable version downloaded from the LAMMPS website). I am using atom style full with non-bonded interactions are modeled using the Buckingham potential and Coulomb interactions with fixed partial charges and pppm kspace style. The relevant commands are below:

boundary p p p
units real
pair_style buck/coul/long 11.0 11.0
neigh_modify exclude molecule all

kspace_style pppm 1e-6

The compilation was done with Intel cluster tools, MKL, mpic++, FFTW3, MVAPICH2 MPI.

I get a WARNING: Using 12-bit tables for long-range coulomb (…/kspace.cpp:316). I am not able to find an explanation about this WARNING in the online manual or the PDF file shipped with the tarball. Previous compilation (30Oct2014) on the same machine using the same Makefile didn’t throw this WARNING.

Can someone please explain what this WARNING means and the possible implications? Thank you in advance.

Hi,

I have been trying to run the 14May16 version (latest stable version
downloaded from the LAMMPS website). I am using atom style full with
non-bonded interactions are modeled using the Buckingham potential and
Coulomb interactions with fixed partial charges and pppm kspace style. The
relevant commands are below:

boundary p p p
units real
pair_style buck/coul/long 11.0 11.0
neigh_modify exclude molecule all
kspace_style pppm 1e-6

The compilation was done with Intel cluster tools, MKL, mpic++, FFTW3,
MVAPICH2 MPI.

I get a WARNING: Using 12-bit tables for long-range coulomb
(../kspace.cpp:316). I am not able to find an explanation about this WARNING
in the online manual or the PDF file shipped with the tarball. Previous
compilation (30Oct2014) on the same machine using the same Makefile didn't
throw this WARNING.

Can someone please explain what this WARNING means and the possible
implications? Thank you in advance.

this warning is an additional information for users. there is no
change in the long-range coulomb implementation of LAMMPS.

the warning indicates, that you are using interpolation tables to
approximate the short-range contributions of the coulomb potential.
those tables are set up with 12-bit accuracy, which will result in an
overall accuracy on the coulomb force which is slightly worse than
single precision floating point math. this is (and has been for a very
long time) the default setting in LAMMPS for all pair styles that
support this feature. it has been deemed a good compromise between
performance and accuracy.

you can change the table spacing using the pair_modify command. there
are changes forthcoming to upgrade the analytical long-range coulomb
to be computed in full double precision. in general, forces are not
very susceptible to errors made in this part of the calculation (same
as for the kspace part), however, stresses are much more sensitive
(and that includes the kspace accuracy parameter).

HTH,
    axel.