[gmx-developers] DIfference in Coulomb terms 5.0/4.6

[David van der Spoel]

We're trying to debug the difference between Coulomb SR/Recip division 
in 4.6 and 5.0. To simplify things we made a system of 2 SPC molecules. 
They use the same cut-off, even after mdrun changes if, and the distance 
between the molecules is such that they are completely within the short 
range cutoff. Results are below. Any clue what causes the difference?

gmx-master:
Statistics over 1 steps [ 0.0000 through 0.0000 ps ], 5 data sets
All statistics are over 1 points

Energy                      Average   Err.Est.       RMSD  Tot-Drift
-------------------------------------------------------------------------------
LJ (SR)                     7.92898         --          0          0 
(kJ/mol)
Disper. corr.            -0.00105424         --          0          0 
(kJ/mol)
Coulomb (SR)               -41.7249         --          0          0 
(kJ/mol)
Coul. recip.                6.32359         --          0          0 
(kJ/mol)
Potential                  -27.4734         --          0          0 
(kJ/mol)


gmx-4.6.7:
Statistics over 1 steps [ 0.0000 through 0.0000 ps ], 5 data sets
All statistics are over 1 points

Energy                      Average   Err.Est.       RMSD  Tot-Drift
-------------------------------------------------------------------------------
LJ (SR)                     7.92898         --          0          0 
(kJ/mol)
Disper. corr.            -0.00105424         --          0          0 
(kJ/mol)
Coulomb (SR)               -36.0859         --          0          0 
(kJ/mol)
Coul. recip.               0.684564         --          0          0 
(kJ/mol)
Potential                  -27.4734         --          0          0 
(kJ/mol)


David.