Per particle scale factors in local refinement?


In local refinement, there is an option to “reset the per particle scale factors to 1.0” before refinement.

This implies that local refinement inherits per-particle scale factors from the preceding global refinement - is this the case?

It seems that while these scale factors might be reasonable if the local region of interest is in the center of the particle, or comprises a large fraction of the mass, they are likely to be quite wrong if the region of interest is on the periphery of the particle, or comprises a very small fraction of the total mass. Is that a reasonable inference?


Hi @olibclarke,

Yes this is true – the input scales are held constant at the input values unless that parameter is set (or of course if “Minimize over per-particle scale” is activated).

That may be a reasonable inference – we have not done extensive investigation into how the optimal scale would vary across the structure (aside from observing that often scale estimation can worsen results when the desired mask covers a small fraction of the total structure). Due to how varied the masks for local refinement can be, we chose to keep the default to preserving the input scales, since it was not clear to us if it is categorically better to reset them to 1 than to keep them.