search

michellab / plumed2

Development version of plumed 2
GNU Lesser General Public License v3.0
1 stars 0 forks source link

Grid update and reproducibility #5

Open carajillu opened 5 years ago

carajillu commented 5 years ago

It bugged me a lot that post-processing with driver gave different JEDI score than monitoring on the fly. There has to be a problem with the way the grid is updated, and I think it is that the position of the grid ata given time depends on its position the last time JEDI was calculated. If you are biasing JEDI, the grid is being updated at every step and therefore it makes a lot of sense that the JEDI score is different upon post-processing. The way the grid is updated should be changed so that the final position doesn't depend on the initial (although I have no idea how to do that).

carajillu commented 5 years ago

This is something we need to look at. Reference values are being updated for binding site atom positions, binding site COM and center of the grid, but not for the positions of the grid points. This means that every time JEDI is called the grid is translated from the place it was in when it was generated. Sounds like a massive translation to me and is probably related to the fluctuations in JEDI score.

Problem: I remember putting the reference grid positions inside this loop a while ago, and then I would get JEDI equal to 0 for all snapshots except the first

image

jmichel80 commented 5 years ago

If I recall well the grid is centred on the origin (0,0,0) after initialisation by center_grid. This is needed for calculate_rotation_rmsd. Why do you think this is a problem to apply a large translation to the grid coordinates ?

carajillu commented 5 years ago

Picking up on this issue. To be fair I don't even remember why I thought bringing the grid from far away would be a problem.

However, I have been looking at the code over the last few days and I have noticed that although the grid is centered on the origin upon initialisation, the binding site is not. Now I am not saying that is certainly a problem, but centering both grid and binding site on the origin prior to simulation shouldn't do any harm that I can think of, and having them centered at the origin before the simulation starts would prevent us from potentially dealing with something that is out of the periodic box.

I have no idea how plumed deals with things that are out of the box when using PBC, but better to not have to find out. just something worth a thought.