[gmx-users] Add additional LJ to certain atom pair- experience and a funny result

[jia jia]

Dear all:
   I'm trying to add additional LJ potential to certain atom pair, e.g. for
atom 1 and 2, except normal atom LJ and Coulomb depend on atom type, I want
add an additional LJ to them (so, only for atom 1 and 2, they will have
Vorin+Vadd). I'm using gromcas 4.0.5 (double precision).
   My way to do this is add a line for " 1 2 1 C6value C12value" to the
paires as the gromacs does no scale 1-4 potential.
   To test whether it work, I've set a system contain only two non-bonded
atom, then perform two 1-cpu, 40 step simulation (output v, f every step),
one simulation is with origin LJ potential, one simulation is add the line
" 1 2 1 C6value C12value" to paires, and change the C6 C12 value of atom 1
2 in XXnb.itp (or other nonbond parameter file) to "0 0".
   With this change, I'm expecting the two simulation give identical result.

   I've checked the result with gmxdump, when using G53a6, yes, after 40
step two siimulations give identical results.

   However, when use Martini, the 0 - 10 step is identical, after that
there is growing difference between them. At the end of 40steps, the
difference in coordinations is about 0.0002 nm, and for the force is also
about 0.0002.
   If I run a simulation from a second time, the result from second run is
indentical with  previous run. This suggest the differences are not from
machine  error.

   Has any one test this before or have any idea where this error from?

   I doubt the error may from define variable, e.g. the progrom treat LJ
and 1-4 interaction in different precision.
   I'll contine test and check source, If I find anything I'll report here.

Cheers
Guang