[gmx-users] Constraints not working in pull code (sometimes, sometimes not)

[chris.neale at utoronto.ca]

Are you 100% sure that you have the correct index group for your large  
box? Did you use a .ndx file that you prepared as input to mdrun or  
did you rely on the creation of a standard group? If you relied on the  
creation of a standard group, can you please try again with the larger  
box and a .ndx file that you created (preferably the exact one that  
you used with g_dist) and supplied to mdrun?

If that doesn't fix the problem, then I'd consider filing a redmine issue.

Chris.

-- original message --


Dear all,

I have reported a week ago update about my troubles with octanol/water slab
and pulling simulations, where COM code starts to report a wrong distances
between centre-of-masses between two groups when bigger box (see bellow) is
used.

So far, smaller box cured our actual calculation problems, but I am still
curious why g_dist and pullx.xvg are giving so different values, when I
enlarge the simulation box?

Best wishes
Karel Berka

Original mail down there
>>

Dear all,

We have tested more strange behavior of constraints from pull code for
calculation of free energy differences small molecule on DOPC membrane and
octanol slab.
As I have reported previously, while the errors in free energy differences
on DOPC were rather small, the errors on octanol were strangely high.

We have prepared smaller slabs (as Chris Neale suggested) of octanol with
comparable size of box and we have also tried to analyze the position of
centres either via pullx.xvg provided by pull code, but also by g_dist
utility and to our surprise, size matters. A lot.
(Same groups were used in mdp for pull and for g_dist analysis, only
distances in z-axis is shown)

DOPC, 128 molecules, box size 6.2 6.4 8.3 , position of small molecule on
the edge of membrane
Time(ps)   pullf.xvg    dist.xvg
      0.00    -0.12416    -0.12436
      1.00    -0.12416    -0.12433
      2.00    -0.12416    -0.12428
      3.00    -0.12416    -0.12430
      4.00    -0.12416    -0.12430
      5.00    -0.12416    -0.12432
      6.00    -0.12416    -0.12434
      7.00    -0.12416    -0.12449
As can be seen, the variation of distances reported by these two programs is
not perfect, but it is rather ok with precision of about 0.001 nm.

Octanol, 957 molecules, box size 5.7 5.9 13.7, position is similar on the
edge of slab, errors in free energy profiles are mainly here.
Time(ps)    pullf.xvg    dist.xvg
      0.00    -2.81855    -3.41133
      1.00    -2.81076    -3.44213
      2.00    -2.82005    -3.27949
      3.00    -2.81856    -3.35097
      4.00    -2.82016    -3.50378
      5.00    -2.81849    -3.63261
      6.00    -2.81855    -3.60058
      7.00    -2.81870    -3.80251
      8.00    -2.81859    -3.32790
      9.00    -2.81849    -3.24329
     10.00    -2.81855    -3.28394
     21.00    -2.81855    -4.02524
Here, the distances reported by pull code and by g_dist are completely
different with differences in about 1 nm! Visual inspection in VMD says that
small molecule is highly mobile a actual distances are reported rightly by
g_dist program.
Strangely enough, when we put distance measured by g_dist as a start into
the pull code without guessing the distances, but this also did not worked,
since the molecule tried to be at some completely different position and the
simulation crashed as the molecule speeded up.

Octanol, 475 molecules, box size 5.5 5.5 9.5, position also on the edge of
the slab
Time(ps)      pullf.xvg    dist.xvg
      0.00    -0.49996    -0.49989
      1.00    -0.49996    -0.49988
      2.00    -0.49996    -0.49981
      3.00    -0.49996    -0.49981
      4.00    -0.49996    -0.50001
      5.00    -0.49996    -0.49987
      6.00    -0.49996    -0.49986
      7.00    -0.49996    -0.49994
      8.00    -0.49997    -0.49988
      9.00    -0.49996    -0.49980
     10.00    -0.49996    -0.49988
Smaller box cured miraculously the problem, since here the difference is
similarly small as in DOPC case.

However just to remind another ache we had with octanol slab, in case of
octanol, the size of box in xy plane had to be held constant as the box
narrowed thorough the whole simulation when in anisotropic NPT ensemble.

Why g_dist and pullx show so different results in large box, I still do not
understand. Any suggestions?
And possibly any suggestions whether our box narrowing in case of octanol is
common or do we have another devil hidden there?

Best wishes
--
Zdraví skoro zdravý
Karel "Krápník" Berka