How to choose a fulcrum for local refinement?



Also - during local refinement I often see a population of particles that retain large shifts in both orientation and offset through the final iteration (see below) - they don’t seem to converge, whereas the rest of the particles do. It might be nice to have a way to select and test the removal of such particles, as I suspect they are garbage.



@olibclarke yes, that is how the dynamic masking should work.

We’ve also noticed the population of particles that seem to drift far from their original alignments in a few datasets. We will work on something to filter them out!



Doesn’t seem to be quite working as expected though - see PM


@ali.h most of what you said makes sense (I like the dynamic-static mask approach as well), but I am not so sure about the fulcrum position.

I think you should draw pictures of a stack of several particles with a rigid body displacement, centered on the CoM or the fulcrum, and think about how the whole cost function will evolve over a few iterations in either case. It’s ultimately an empirical question, but there is a reason why CoM-centered refinements have been preferred throughout the history of single-particle analysis. Perhaps branch-and-bound changes the calculus since fine global search can be made reasonably fast.