[gmx-users] Bug (?) with FEP while using particle decomposition in charge transformation

[Alexey Zeifman]

Dear all,

I have faced a very strange results while using the FEP together with particle decomposition in charge transformation. Situation is as follows:

gromacs-4.5.5, single precision, mpi-run on 32 nodes;
only charges are modified in the topology file (VdW and massess remains the same for A and B state); 
particle decomposition (-pd) option is used.

The combination of this conditions lead to a very strange *.xvg file like this (lambda=0):

@TYPE xy
@ subtitle "T = 300 (K), \xl\f{} = 0"
@ view 0.15, 0.15, 0.75, 0.85
@ legend on
@ legend box on
@ legend loctype view
@ legend 0.78, 0.8
@ legend length 2
@ s0 legend "dH/d\xl\f{} \xl\f{} 0"
@ s1 legend "\xD\f{}H \xl\f{} 0.1"
0.0000 7.74067 -2.30013
0.0200 8.83184 -2.08346
0.0400 6.61338 -2.43848
0.0600 3.85357 -2.58843
0.0800 9.49265 -2.03448
0.1000 2.4197 -2.7838
0.1200 4.81406 -2.64074
0.1400 3.79622 -2.83082
0.1600 4.9235 -2.53202
0.1800 7.24284 -2.41357
0.2000 3.79017 -2.68861
0.2200 7.84849 -2.26953
0.2400 10.1044 -2.11132
0.2600 1.89623 -3.10296
0.2800 11.6953 -2.08701
0.3000 11.103 -1.96233
0.3200 4.68676 -2.62905
0.3400 9.83457 -2.11386
0.3600 6.12451 -2.27113
0.3800 8.56351 -2.17592
0.4000 11.0368 -1.93516
0.4200 4.80366 -2.49621
0.4400 9.58472 -2.08431
0.4600 5.04113 -2.60035

The usual file (obtained without -pd) looks like this:

@    title "dH/d\xl\f{}, \xD\f{}H"
@    xaxis  label "Time (ps)"
@    yaxis  label "(kJ/mol)"
@TYPE xy
@ subtitle "T = 300 (K), \xl\f{} = 0"
@ view 0.15, 0.15, 0.75, 0.85
@ legend on
@ legend box on
@ legend loctype view
@ legend 0.78, 0.8
@ legend length 2
@ s0 legend "dH/d\xl\f{} \xl\f{} 0"
@ s1 legend "\xD\f{}H \xl\f{} 0.1"
0.0000 7.77708 0.777688
0.0200 10.322 1.03217
0.0400 8.82036 0.882043
0.0600 9.46574 0.946536
0.0800 10.2726 1.02731
0.1000 12.0395 1.20386
0.1200 6.85707 0.685689
0.1400 8.99015 0.899035
0.1600 8.68244 0.868208
0.1800 8.98454 0.898397
0.2000 9.21256 0.921202
0.2200 7.16041 0.716125
0.2400 5.1431 0.514307
0.2600 3.34832 0.334797
0.2800 3.14324 0.314328
0.3000 1.83755 0.183755
0.3200 3.91737 0.391783
0.3400 5.79081 0.579126
0.3600 4.92766 0.492766
0.3800 8.06557 0.806605
0.4000 7.00964 0.700994
0.4200 12.1488 1.21481

So the signs of the dH/dlambda are opposite while using particle decomposition. 
There is no such problem while running VdW transformation.

Is it a Gromacs bug or I'm doing something wrong?

With best regards,
Alexey Zeifman