Dear users and developers,
After a few months of trial and error (and no success), I finally gave up using "conventional" CVs with my system and decided to move on to path CVs. I read the original path CV paper [Branduardi, Gervasio and Parrinello (2007) J. Chem. Phys. 126, 054103] along with several papers of its application to cyclin dependent kinases (CDKs) and small drug candidates by Gervasio & Co. (because my system also contains a CDK, despite being different other aspects that I do not wish to dwell upon), and everything seemed to be more or less clear to me.
In short, I ran a short plain MetaD simulation with a contact map of interest and used the first unbinding event to define states A and B, along with 3 intermediates, for a total N = 5. I used the driver to calculate the optimal RMSD of the unbinding process and chose equally distant frames within RMSD space. I dumped the frames containing only the atoms of interest and, from the Belfast tutorial on adaptive variables, I figured out how to estimate my lambda value.
As for sigma, there's where I started to get in trouble. I failed to understand where I can get a good estimate of both sigmas, one for S and another for Z. Thus, I decided to give them both the same value, which was taken as the standard deviation of the RMSD (of this subset of atoms of interest) from a long unbiased MD simulation of this system. I should have probably used adaptive sigmas...
Anyway, I started the simulation with the following plumed code:
"""
p: PATH REFERENCE=frames.pdb LAMBDA=15 TYPE=OPTIMAL
# every 1 ps
m: METAD ARG=p.spath,p.zpath HEIGHT=0.5 SIGMA=0.015,0.015 PACE=500 FILE=HILLS
# every 10 ps
PRINT ARG=p.spath,p.zpath,m.bias STRIDE=5000 FILE=COLVAR
"""
The simulation ran for about 10 ns before it crashed with the classic:
"""
X particles communicated to PME rank Y are more than 2/3 times the cut-off
"""
However, this is not what worried me the most. What really worried me was that after a quick plot of S vs. Z, I noticed that Z takes on negative values. From Branduardi's paper mentioned above, Z(R), as defined in Eq. 9, should not take negative values assuming lambda is positive. Right? If so, does anyone have any idea of what is going on? Any input would be really appreciated.
Best regards,
J