[phenixbb] zero B-factors
Pavel Afonine
pafonine at lbl.gov
Mon Nov 18 13:38:24 PST 2013
If you see this problem using the latest Phenix version then please send
me files (off-list) necessary to reproduce it and I will investigate.
If interested what was the "problem" causing zero B-factors please see
below.
Thanks,
Pavel
In a nutshell the problem is this. Assuming total model structure factor
is (for simplicity):
F ~ exp(-Boverall_scale * s**2) * exp(-Batoms * s**2)
it's clear that any combination of Boverall_scale and Batoms that
maintains Boverall_scale+Batoms=const will not change total model
structure factors (and therefore R-factors, maps, etc).
The new bulk-solvent and overall scaling algorithm that we implemented
sometime in April 2012 and that went into 1.8_1069 originally did not
care about returning isotropic component of Boverall_scale back to atoms
(Batoms):
http://journals.iucr.org/d/issues/2013/04/00/dz5273/dz5273.pdf
In fact, it assumed that overall contribution (overall B that is equal
for all atoms) should stay in Boverall_scale (indeed, why keep common B
in all individual atomic B-factors!?), and the rest goes into Batoms.
This behavior typically generates user concerns about "too small"
B-factors.
An alternative agreement is to postulate that the matrix Boverall_scale
is traceless meaning that overall B-factor goes into individual atomic
B-factors.
This behavior occasionally generates user concerns about "too large"
B-factors.
A few months later (sometime in end summer/fall 2012) we changed scaling
behavior such that overall B-factor stays in individual atomic B-factors.
All in all, either way is correct and a matter of agreement and
preferences, but definitely not a bug.
More information about the phenixbb
mailing list