I am using lammps-6Jun14.
There seems to be a discrepancy between the kspace_styles ewald, ewald/disp, and pppm.
When a triclinic box is used with positive tilts, the total energy reported by all three solvers is the same ( -424.09 eV, for the test system i’ve attached, with xy = xz = yz = 2 lattice units). However, if any of the tilts are negative, the solvers do not agree. There are lots of possible combinations, but here are the results for two of them (energies in eV, tilts in lattice units):
xy xz yz -2 0 0
ewald -427
ewald/disp -424
pppm -424
xy xz yz -2 -2 -2
ewald -789
ewald/disp -424
pppm -795
The input script I used to generate these numbers is attached. It loops through the three kspace styles automatically. I manually tried various tilt combinations by changing the ‘region’ command.
I used a kspace accuracy of 10^-6. Changing the accuracy changes the values of the “wrong” energies (those that aren’t -424 eV) dramatically. Also, it seems that the problem only appears for the pppm solver when box is “very” tilted, but I haven’t looked in to this much. By “very” tilted, I mean that more than one tilt is negative, and they are close or equal to the maximum tilt.
Any help would be appreciated.
Thanks, Steven
in.test (630 Bytes)