[gmx-users] free energy and soft core

[Stefano Piana]

Hi,
I am very much interested in this discussion about free energy 
calculations as I am doing some myself. As this can be a related issue I 
attach this mail to this topic.
I am "vaporizing" a molecule from a solution by turning the LJ 
parameters and the charges to 0 during a perturbation run.
The funny thing is that if I vaporize the molecule alone (without water) 
in a non-periodic system I still get a free energy difference although 
there is no Coulomb or LJ interaction at all in the molecule (they are 
exactly 0 in the .edr file) and no contribution to the kinetic energy as 
I am not changing the masses.
Energy                      Average       RMSD     Fluct.      Drift 
Tot-Drift
-------------------------------------------------------------------------------
Angle                       10.4334    5.38315    4.38697   0.108069 
10.8071
Proper Dih.                 4.84332     3.0113    3.00803 -0.00485981 
-0.485991
Improper Dih.             0.0462622  0.0583175  0.0582569 9.20078e-05 
0.00920096
LJ (SR)                           0          0          0          0 
       0
Coulomb (SR)                      0          0          0          0 
       0
Potential                    15.323    5.86594    5.05137   0.103301 
10.3303
Kinetic En.                 14.2877    5.47298    5.46365  0.0110618 
  1.1062
Total Energy                29.6107    4.23443    2.65159   0.114363 
11.4365
Temperature                 312.437    119.681    119.477   0.241895 
   24.19
Pressure (bar)                    0          0          0          0 
       0
dVpot/dlambda              0.141056    4.96762    4.94791  0.0153115 
1.53118
dEkin/dlambda                     0          0          0          0 
       0
Heat Capacity Cv:      15.9914 J/mol K (factor = 0.146731)
----------------------------------------------------------------------

The free energy difference that I obtain is small (0.141 kj mol-1) but 
reproducible to the third decimal digit in runs of different length and 
with different values for sc-alpha.
Any idea of where this contribution is coming from?

Stefano