GROMACS

g_chi -h reports:

* phi and psi dihedrals are calculated in a non-standard way, using H-N-CA-C
  for phi instead of C(-)-N-CA-C, and N-CA-C-O for psi instead of
  N-CA-C-N(+). This causes (usually small) discrepancies with the output of
  other tools like g_rama.

The file chi.log then lists the atoms used to evaluate the dihedral angles:

Residue VAL2
 Angle [   AI,   AJ,   AK,   AL]  #tr/ns  S^2D
--------------------------------------------
   Phi [   19,   18,   20,   32]    0.00    1.00
   Psi [   18,   20,   32,   33]    0.00    1.00
   Chi1[   18,   20,   22,   24]    0.00    1.00

And, consistently, the phi dihedral (19,18,20,32 from above) is H-N-CA-C :

-bash-4.1$ head -40 restr.gro 
Green Red Orange Magenta Azure Cyan Skyblue in water
54303
    1GLU      N    1   3.250   1.491   2.817
    1GLU     H1    2   3.327   1.535   2.864
    1GLU     H2    3   3.273   1.395   2.792
    1GLU     H3    4   3.231   1.539   2.731
    1GLU     CA    5   3.128   1.494   2.900
    1GLU     HA    6   3.099   1.598   2.916
    1GLU     CB    7   3.012   1.432   2.819
    1GLU    HB1    8   2.999   1.328   2.848
    1GLU    HB2    9   3.039   1.434   2.712
    1GLU     CG   10   2.884   1.512   2.837
    1GLU    HG1   11   2.907   1.618   2.820
    1GLU    HG2   12   2.848   1.500   2.938
    1GLU     CD   13   2.777   1.466   2.737
    1GLU    OE1   14   2.709   1.556   2.684
    1GLU    OE2   15   2.773   1.345   2.711
    1GLU      C   16   3.141   1.430   3.039
    1GLU      O   17   3.053   1.450   3.121
    2VAL      N   18   3.253   1.362   3.073
    2VAL      H   19   3.318   1.335   3.000
    2VAL     CA   20   3.276   1.304   3.207
    2VAL     HA   21   3.181   1.292   3.258
    2VAL     CB   22   3.338   1.164   3.193
    2VAL     HB   23   3.416   1.167   3.117
    2VAL    CG1   24   3.402   1.111   3.322
    2VAL   HG11   25   3.435   1.008   3.308
    2VAL   HG12   26   3.489   1.170   3.349
    2VAL   HG13   27   3.329   1.114   3.404
    2VAL    CG2   28   3.229   1.065   3.149
    2VAL   HG21   29   3.272   0.966   3.133
    2VAL   HG22   30   3.151   1.058   3.225
    2VAL   HG23   31   3.184   1.098   3.056
    2VAL      C   32   3.359   1.396   3.297
    2VAL      O   33   3.318   1.422   3.410
    3TRP      N   34   3.471   1.452   3.248
    3TRP      H   35   3.508   1.420   3.160
    3TRP     CA   36   3.549   1.552   3.324
    3TRP     HA   37   3.579   1.510   3.421
    3TRP     CB   38   3.676   1.589   3.246

However, looking at the value returned from g_chi in the .xvg file, it is clear that g_chi is either (a) actually using C(i-1)-N-CA-C for phi, or (b) somehow using the computed H-N-CA-C dihedral angle to estimate the C(i-1)-N-CA-C dihedral angle.

This lead me to scripting problems when I tried to extract dihedral angle definitions and reference angles for use in dihedral restraints because the reported angle does not match the reported atomic definition.

Complicating things even more, bug 53 suggests that g_chi and g_rama are now exactly the same for phi and psi (except the terminal residues). See also https://www.mail-archive.com/gmx-users@gromacs.org/msg19991.html and my new calculations indicate that g_rama and g_chi give the same backbone dihedrals.

Therefore, I suggest that:
(a) the warning about non-standard torsions be changed to indicate that this only relates to terminal residues
(b) the definition of atoms printed in chi.log be corrected.