Hi,
What Pavel describes is how to get a “difference map”. But it is not an “omit map” per the original definition.
perhaps we are getting into terminology here.. I'm sticking to the original paper by T. Bhat. Difference map is F1-F2, F1=w1*Fobs, F2=w2*Fmodel. I believe you can prefix it with "omit" if you omit something (remove part of model) from F2.
The omitted part should never have been part of the model during any refinement step before the omit map is calculated. The purpose is to prevent bias.
This is where it gets into nuances not discussed in the original paper, if memory serves.
If the model has already been refined with a ligand, then the technique of simulated annealing refinement can be used to remove the bias after removing the ligand. This approximates an “omit map” (still one may argue not a real “omit map”, but close enough).
Yes, this seems to be the common mindset. But I'm yet to see a convincing example with clear illustration of such bias. I believe this 1) remove ligand; 2) do some refinement (optionally, don't believe really necessary!); 3) compute mFo-DFc map; should be good enough in most cases. Pavel