[gmx-users] Differences between gmx bar and pymbar free energy

[Gmx QA]

Hi gmx-users

I am running a series of free-energy of solvation calculations for small
molecules in water and octanol, to calculate partition coefficients.

For this, I have set up a series of lambda-points in water and in octanol,
and then haves used primarily gmx bar to evaluate the results.

Using gmx bar (v.5.0.4) with the following command line
$ gmx bar -f md*xvg -o -oi -oh -b 100 &>FEP.dat

I get for eg. ethanol in water a Delta-G of 14.33 kJ/mol (which is very
close to the value (14.42 kj/mol) reported by eg. Lundborg and Lindahl in
JPCB, 119, 2015, for the same combination of water model and forcefield, so
that is reassuring).

Then I tried the same analysis using pymbar (cloned today from git) like
this:
$ python alchemical_analysis.py -t 300 -p 1 -q xvg -p md -m BAR -v -s 100
&> FEP_pymbar.dat

Now, I get a Delta-G of 64.235 kJ/mol, so clearly not the same as with gmx
bar (and not simply a kJ vs kcal issue either I think)

I feel like I am missing something fundamental since I get so different
values, and would appreciate any feedback. The simulations have been run
using gmx 5.0.4 with a total of 47 lambda points, and each run is 10 ns in
length.

I can provide mdp-files and such if needed.

Cheers
/PK