meeting 2025 03 17 n57 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki

Context

We're revisiting N57's spectral fits following the Barth stars error from a few weeks ago, and here are some diangotics and thoughts on the newest rounds of fits:
- So far, I'm having a hard time getting nice looking/stable fits with the newest pPXF code, and also having hard time reproducing results from the old code with the new code (my current guess is that the new code and old code are doing something subtly different, since I can't seem to get the same output for [what I think are] identical inputs to the code).
- I've got some diagnostics and thoughts below, but things are quite all over the place right now with many knobs turning and not a good sense of what is causing things to go awry. In the meantime, while I try to get a handle on things and why they are giving strange fits, I'm focusing on running only the new pPXF code to get a sensibile set of outputs.
Here are some key takeaways though:
- With the new pPXF code:
  - The primary problem with the new pPXF code is that I can't seem to get stable results across all spectra. There seem to consistently be at least a handful of spectra which are horribly misfit depending on what exactly I change in the fit.
  - The best set of results I've obtained so far are using the incorrect set of Barth stars, bias = 0.0, adeg = -1, and a redshift = 0.017880 (from NED). This should be similar to the current kinematics with the caveat that I had been using adeg=0 rather than adeg=-1 (for my own reference/keeping track, I've been calling this Case A).
  - My initial hunch is that the adegree is causing some weirdness in the new code, but I need to test this a bit further. There doesn't seem to be the same sensitivity to the adgree whe using the old code.
- With the old pPXF code:
  - I seem to be able to get much more stable results with the old pPXF code. If I use the same inputs mentioned above (incorrect Barth stars, bias = 0.0, adeg=-1, redshift=0.017880), I can again get a decent set of fits using these exact settings or with adeg=0 (for my own reference, I've been calling this Case F Comp and Case G Comp, where the difference between them is a slightly different mask, but the results are in decent agreement).
  - What's most perplexing is that Case F Comp and Case G Comp (old pPXF code with very similar settings) give very nice fits, but the new pPXF code with the same settings is very poor. I'm trying to figure out why this is the case now but am sort of grapsing at straws.
- With all this said:
  - Despite the fits currently being all over the place (and me still sorting things out), there is already a pretty decent agreement between my Best Case in the new code and the old set of kinematics (both of which are using teh incorrect stars). At least we're not totally discrepent with the old data.
Currently I'm focusing on using the new code alone to troubleshoot the fits. It's quite challenging bouncing between the old and new code, trying to see exactly where things are breaking down between the two especially when the new set of results can be all over the place. So I think more a more effective use of time is trying to get a decent set of fits with the new code (since I can already do that with the old code) to see how to make sense of both sets of results.

Diagnostics

First, here's a random sample of spectra fits with the new pPXF code.
- The first column is my current "best set" of fits. The main settings here are (incorrect Barth stars) + bias = 0.0 + adeg = -1. This set of fits is special since it's the only case with the new pPXF code in which all spectral bins seem to be well behaved.
- The second column should be the same settings as the first column, with the only change being that I'm using the correct Barth stars instead of the off-by-one stars. Note that the spectra are not exactly the same between columns, but I've chosen spectra trying to represent the range in quality. Note that some of the fits look great, whereas others are likely being dominated by specific stellar templates much like we saw with N315.
- The third column is the same settings except I use the full CAT library of stars. Again a similar trend where some of the spectra look great, whereas others are sigma >= 600 km/s.
- The fourth column should be the same settings but using adeg=-1 rather than adeg=0. These are better behaved, but they still have some sigmas that are >=500 or 600 km/s.

	Current "Best" Fits	Same Settings but Correct Barth Stars
Bin 1	[images/250317/A/pPXF_plot_00000-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/C/pPXF_plot_00000-1.png)	[images/250317/B/pPXF_plot_00000-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/D/pPXF_plot_00009-1.png)
Bin 2	[images/250317/A/pPXF_plot_00010-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/C/pPXF_plot_00003-1.png)	[images/250317/B/pPXF_plot_00010-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/D/pPXF_plot_00012-1.png)
Bin 3	[images/250317/A/pPXF_plot_00050-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/C/pPXF_plot_00006-1.png)	[images/250317/B/pPXF_plot_00050-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/D/pPXF_plot_00016-1.png)
Bin 4	[images/250317/A/pPXF_plot_00100-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/C/pPXF_plot_00007-1.png)	[images/250317/B/pPXF_plot_00100-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/D/pPXF_plot_00021-1.png)
Bin 5	[images/250317/A/pPXF_plot_00200-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/C/pPXF_plot_00008-1.png)	[images/250317/B/pPXF_plot_00200-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/D/pPXF_plot_00042-1.png)

Here's a similar set of plots, but this time comparing three sets of fits with the old code which seem to be a bit better behaved overall despite some drastic parameter changes between the cases. There aren't any spectra in these sets of fits which "explode" in sigma like we get with the new pPXF code.
- The first column should be the old code run with the correct Barth stars and a mask I had created when initially fitting the spectra. It's encouraging that this case (which has bascially all the corrected parameters/templates/etc) is quite robust:
- The second and third columns are showing essentially reproductions of the last column in the table above but run with the new code. I was trying to test here if I'm getting the same outputs for the same inputs between the old and new code. The old code results with different adgree values (below) are much more similar to another than either result is to the columns above, suggesting that something is different between the old and the new code that I am currently missing. This I think makes sense given than the old code results are still quite similar to one another and the new code is what keeps shifting around.

Corrected Barth Stars	"Best" Reproduction with Old Code adeg=0	"Best" Reproduction with deg=0 and slightly different mask
[images/250317/E/pPXF_plot_0-1.png]] ](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/F/pPXF_plot_0-1.png--)	images/250317/G/pPXF_plot_0-1.png
[images/250317/E/pPXF_plot_10-1.png]] ](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/F/pPXF_plot_10-1.png-)	images/250317/G/pPXF_plot_10-1.png
[images/250317/E/pPXF_plot_50-1.png]] ](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/F/pPXF_plot_50-1.png-)	images/250317/G/pPXF_plot_50-1.png
[images/250317/E/pPXF_plot_100-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/F/pPXF_plot_100-1.png)	images/250317/G/pPXF_plot_100-1.png
[images/250317/E/pPXF_plot_200-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250317/F/pPXF_plot_200-1.png)	images/250317/G/pPXF_plot_200-1.png

I realize things are quite a mess currently, but I did make one quick comparison of the "Current "Best" Fits" with the new code versus the old kinematics from the previous reduction. These currently have very different settings as you might imagine, but the kinematics seem to be relatively close to one another despite the challenges. Again, take these with a grain of salt since I still need to improve the fits, but I hope the new minimum is not completely different from where it currently is given how things are shaking out so far:

Plot
images/250317/quick_check.png

More to come....