coul/wolf bug

Sorry if you get this twice, I made a small error in describing the test
job -- it is r=1.0 angstrom to r=12.0 angstrom with a step size of 0.01
angstrom/step. Best Regards, John

Dear Lammps-users,
I think I found a bug in coul/wolf. It doesn't look like the energy goes
to zero for r > rcut.

Here is a test job which scans the coulombic energy for an arbitrary dimer
with +1.0 charge on each atom for r = 1.0 angstroms to r = 10.0 angstroms
in increments of 0.1 angstroms per step.

Thanks,
John

coul.lammpstrj (172 KB)

data.dimer (231 Bytes)

dimer.in (471 Bytes)

dimer.out (26.1 KB)

I'll see if Ray can look at this.

Steve

Hi John,

Thanks for pointing this out. After examining the potential and the
original paper, I don't think it is a bug that energy is not zero when
using the Wolf Summation method when rij > Rc. Rather it is the nature of
the summation method.

Wolf sum is a method based on Ewald sum with the following
modifications/approximations: spherical truncation with a finite cutoff,
damping via the complementary error function, charge neutralized within
the cutoff sphere, and dropping off the slow converging error function
term by adding and subtracting a self energy term. The total energy from
Wolf after all that becomes a sum of a shifted Ewald (charge-neutralized)
potential and a self term (Etot = Eew - Eself).

Ewald energy (Eew) does go to zero when rij > Rc, but Eself is a constant
that depends on damping, Rc and charge only. This constant does not
vanish when rij > Rc, and it is why you see different energies when rij >
Rc with different damping and Rc values. Two atoms in a vacuum is an
extreme case for Wolf summation method, and the PotE does not go to zero
if you use Ewald sum with commands "pair_style coul/long" with
"kspace_style ewald". I don't think you can compare Wolf/Ewald with a
pure cutoff method like coul/cut.

By the way, the damping (0.3 you used) is a factor determining how fast
the complementary error function (erfc) falls off from one to zero with
increasing rij. Erfc falls off slower with a smaller value of damping
(0.1-0.2), and the energy/forces converges better.

For more info, Sections III and V of "D. Wolf, J. Chem. Phys. 110 17 1999"
have detailed descriptions.

Regards,
Ray