search

michellab / plumed2

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

grid_update branch #8

Open carajillu opened 5 years ago

carajillu commented 5 years ago

ref_xlist and mov_xlist have length = n_apolarpolar and they should have length = new_bsite.size(). when new_bsite.size() is very small (small bsite_cutoff), ref_xlist and mov_xlist must be having a lot of zeros and that is probably why the calculation of eigenvalues and eigenvectors doesn't converge and JEDI profile is even worse.

carajillu commented 5 years ago

This bug has been fixed but the results using a grid_cutoff (the distance with the grid at which an atom is NOT taken into account for the grid update) of 99999 and 0.2 are extremely similar. The actual scores at each snapshot are NOT the same, varying up to 0.3-0.4 units in some cases, but the overall trend and the fluctuations and the spikiness are the same.

@jmichel80 you are the one who coded the grid update in the first place (I understand it might not be fresh in mind), right? Can you have a look at the code when you have a moment and let me know if you see something clearly wrong?

In case you want to run any tests, the branch is grid_update and you will need to add the keyword bsite_cutoff in jedi.params.

Example files are in ghidorah:/home/joan/PhD/year_5/grid_update_cutoff