Simplified arc line centroiding.

[cmccully]

I've simplified the peak refinement algorithm. Rather than trying to fit all lines simultaneously, we only fit one line at a time and only the data near the original peak. This has fixed some catastrophic failures that cause later stages to crash because the wavelength solution is no longer monotonic.

[jchate6]

I guess my question is if this is the right way to do things.
Isn't fitting them all the more robust method? Should this be a fallback option to find a solution in case a single solution doesn't work?

[cmccully]

This is only breaking apart the line centroiding part to work line by line. Our line atlases are not complete (and may not be possible to complete) so we don't have good enough atomic data to centroid all blends. There are cases where the fitters can get into a part of parameter space where they centroid the wrong feature entirely which was happening here.

For the final wavelength solution we still fit the wavelength polynomial to all the lines simultaneously.

This is a case of when the "proper" thing to do makes the actual reduction process more brittle.

[sfoale]

lgtm