Bug in CCFD stress profile calculation of GROMACS-LS?

In Martini simulations (15 us, last 5 us analyzed) of an asymmetric GM1+POPC lipid bilayer (5+95 upper, 0+100 lower leaflet) strong stress fluctuations arising from the dihedral contributions of CCF decomposition are found. It is possible that these are due to a slight bug in the GROMACS-LS (http://mdstress.org) code, because:

1) The fluctuations do not show converge, when more data is added, see the attached plot ‘effect_of_sampling_rate.pdf’.

2) The fluctuations (but no other features of the stress profiles) visibly depend on the chosen frame of reference, see ‘effect_of_centering_schemes.pdf’.

3) The stress profiles calculated along the coordinates that lay in the membrane plane (x and y), are not constant in CCFD, see ‘profiles_along_x_and_y.pdf’.

To assist debugging the code, this repository contains:

5to95_0to100_0.trr --- 1st 33.3% of a 5 us trajectory with snapshots every 100 ps
5to95_0to100_0.trr --- 2nd 33.3% of a 5 us trajectory with snapshots every 100 ps
5to95_0to100_0.trr --- 3rd 33.3% of a 5 us trajectory with snapshots every 100 ps

5to95_0to100.tpr --- tpr used for production (Gromacs 5.1.1)
5to95_0to100_rerun.tpr --- tpr used for stress analysis and centering

5to95_0to100.gro --- gro file after 15 us (10 us relaxation, 5 us production)

index.ndx --- index file used in production
5to95_0to100.ndx --- index file used in analysis

system.top --- topology file
martini_v2.0_ions.itp --- martini FF file (system.top expects this to be in folder toppar/)
martini_v2.0_lipids_all_201506.itp --- martini FF file (system.top expects this to be in folder toppar/)
martini_v2.2.itp --- martini FF file (system.top expects this to be in folder toppar/)

martini_straight_GM1_saveVels.mdp --- run input file used for production (Gromacs 5.1.1)
martini_straight_GM1_rerunForP.mdp --- run input file used for analysis and centering

effect_of_sampling_rate.pdf --- plot of results showing that fluctuations in CCFD do not appear to converge when data is added
profiles_along_x_and_y.pdf --- plot of results showing that in CCFD pressure profiles along the membrane directions are not constant
effect_of_centering_schemes.pdf --- plot of results showing that choice of center of mass (here three possibilities are shown: CoM of POPC GL1 beads, CoM of POPCs, and CoM of all lipids) visibly affects CCFD fluctuations, but not other features of the stress profile
effect_of_centering_schemes_2.pdf --- plot of results (similar to 'effect_of_centering_schemes.pdf', but the center of the bilayer is at 3.7 nm instead at 0.0 / 7.4 nm)
individual_components.pdf --- plot of results showing that the fluctuations in CCFD arise from dihedrals
CCFD_gridsize.pdf --- plot of results showing that decreasing gridsize has no clear effect on the CCFD fluctuations