A common, but minor reason why negative displacement values occur is due to
ambiguity in the placement of the background at high Q (angle).
Another common reason why Uiso is low is because of the neglect of absorption.
(We have been through this many times - see the Rietveld archive eg
With regard to constrained refinement: Should one first refine lattice
parameters and then fix them before proceeding with structure refinement? As
I see it, lattice parameters are directly observable in the xrd pattern; if one
allows lattice parameters to change during structure refinement,
