meeting 2025 05 22 n410 - JacobPilawa/TriaxSchwarzschild_wiki_6 GitHub Wiki

Trying to clean up some of the N57 template issues by taking an iterative approach to the template selection. At least for N410, I was finding that the trimmed library still had quite a few odd looking spectra that were getting picked in the pPXF fit. I've got a host of equivalent diagnostics for N57 here which seem to broadly show the same issue of strange templates getting picked, and thus resulting in weird spectra.

Results:

• Still working through things, but I've finding that I can make the full library and Barth library agree very nicely by trimming out many of the strange template spectra. There were still quite a few templates that were obviously unsuitable being picked by pPXF, and this seemed to be driving the differences in the kinematics between cases.
• After trimming away ~25% or so of the templates, I'm able to get comparable agreement with what Irina quotes in MASSIVE XIII. Still pushing this a bit further since there are definitely some remaining wonky spectra, but the progress has been nice so far.

Just to set the stage here, I'll start with essentially our fiducial kinematic settings. This is basically adeg=0, bias=0.2, startguess=[0, 200], mdeg=3, etc. The only difference in these set of plots is whether or not I'm using only the Barth library vs. the trimmed library from Emily.

Here's a comparison of the kinematics from these cases:

• Notably, none of the Barth templates are selected as being in the Top 15 most important templates in the trimmed library. Instead, the fit seems to prefer stars with very strange looking stellar continuua, like due to miscalibration when creating the library.

Given how distinct the preferred templates are, and following the advice from Emily, I've removed these 15 preferred spectra (since none of them overlapped with the Barth stars), and refit the spectra. For completeness, I've removed the following from the trimmed library:

• ['scan638.fits', 'scan262.fits', 'scan670.fits', 'scan405.fits', 'scan488.fits', 'scan475.fits', 'scan353.fits', 'scan547.fits', 'scan171.fits', 'scan685.fits', 'scan168.fits', 'scan336.fits', 'scan423.fits', 'scan692.fits', 'scan307.fits']
• I'm temporarily calling the trimmed libray minus these 15 templates "Trimmed B."

Here are the same set of results from above but without these 15 specific templates.

• In short, there are still a few funky looking spectra here, but the results have started to improve with the RMS differences coming donw. There is still a bit of variability in the higher order moments, so I'm going to try a few more trims to the library.

Here's a comparison of the kinematics from these cases:

• Notably, none of the Barth templates are selected as being in the Top 15 most important templates in the trimmed library. Instead, the fit seems to prefer stars with very strange looking stellar continuua, like due to miscalibration when creating the library.

Doing another round of cuts since there are still no overlaps between the Barth subset and the full trimmed library, and there are still some templates that are quite funky looking that are preferred in these spectra. For library C, I've removed 30 templates now, adding these 15 to be removed:

• ['scan398.fits','scan694.fits','scan516.fits','scan408.fits','scan704.fits','scan572.fits','scan268.fits','scan479.fits','scan406.fits','scan270.fits','scan023.fits','scan472.fits','scan563.fits','scan352.fits','scan335.fits']
• This library is Trimmed C

This is the first case where we start to get an overlap with the Barth stars! Specifically scan304.fits is present in both sets of top 15 templates, which is a good sign. I'm going to continue with another trimming of the library to see where things fall.

Here's a comparison of the kinematics from these cases:

• Notably, none of the Barth templates are selected as being in the Top 15 most important templates in the trimmed library. Instead, the fit seems to prefer stars with very strange looking stellar continuua, like due to miscalibration when creating the library.

Doing another round of trimming here, but only cutting 14 templates this time to retain the one Barth star that was found to be overlapping with Barth vs. Trimmed C.

This new round continues to improve, and things are starting to look really nice! This set of fits picks up a second Barth star among the top 15, with the new entry being 'scan286.fits.' Going to try one more trim to see if we eventually converge on the Barth set of stars.

Here's a comparison of the kinematics from these cases:

• We now have two templates that overlap with the Barth set as mentioned above.

There's still one obviously weird template that keeps getting picked, so trying another round of cuts by removing the 13 stars from Trimmed D that are not in Barth (keeping scan304 and scan286).

• Strangely, this Trimmed E library, though it still has scan286, seems to prefer to no longer have that template in its top 15, instead favoring a few more spectra that are extremely weird.
• Given this strange result of one of the Barth stars now dropping out, I'm going to try a slightly different approach for the next mask. For the remaining ~150 templates, I'm going to go through them all and try to remove strange looking ones by hand rather than this iterative approach, since it seems like this iterative approach will still eventually converge to the Barth results

Here's a comparison of the kinematics from these cases:

• We now have two templates that overlap with the Barth set as mentioned above.

Going through the above plots by hand a bit quickly, I identify the following templates as looking problematic:

• ['scan032.fits','scan034.fits','scan233.fits','scan290.fits','scan315.fits','scan331.fits','scan344.fits','scan404.fits','scan441.fits','scan447.fits','scan512.fits','scan539.fits','scan643.fits']
• I've removed these from the trimmed E library (leaving Trimmed F with 139 templates still able to be used), and again reprocessed the spectra:
  • Note that there is still only a single template that overlaps with the Barth set (scan304.fits)

The results are still very similar to what I find for Trimmed Grid E, and I think we are at the point of diminishing returns now (unless there is a better approach here). I still notice that some of the spectra that are being selected are quite wonky, with obvious slopes across the data or variabilities in data quality that I personally don't see among the Barth templates.

• All of this is to say I think that we can definitely converge on the Barth results if we continue down this path by removing these strange looking templates.

Here's a comparison of the kinematics from these cases:

• We now have two templates that overlap with the Barth set as mentioned above.