meeting 2025 07 02 n57 - JacobPilawa/TriaxSchwarzschild_wiki_6 GitHub Wiki
- Following up on some open question/suggestions from the meeting.
- Additive Polynomial Follow Ups/Distributions of Adeg:
- I've added plots below showing the additive polynomial as a function of sky position, the radial kinematics (straight from pPXF) for the cases below, and comparisons of the distributions of adeg for the cases as well.
- It seems like in general, limiting the spectra to the Barth/near-Barth subset results in additive polynomials that are consistently positive, whereas opening the library up to the full set of templates gives a much more "centered" set of polynomials with means much closer to 0.
- While I'm not sure this is a bad thing per se, it's definitely noticeable looking at either the raw distributions, the 1-to-1 plots, or the sky maps of the adeg.
- Another interesting note is that there does not appear to be systematic trends as a function of aperture for the additive polynomial degree (e.g., it doesn't appear to be highest at the center and decrease outwards). Given that, I think we can rule out any spatial correlation with the wonky kinematic behavior.
- Refitting Spectra with Top 15/384
- Another sanity check we wanted to look at was the result of taking the highest weight stars when using the full set of 384 templates, refit the spectra, and compare the results.
- This test seems to show that the top 15 templates by weight are ~all that is needed to reproduce the kinematics. If we compare the 384 + 3 to the results from just the top 15 alone, the median differences are very small, centered roughly around 0, and definitely within the size of the errors.
- Stars Selection by Spectral Type:
- This is perhaps the most interesting set of new results. Near the bottom of the page, I've included some plots where I break down our various fits into the weights by spectral type, and there are a few notable trends.
- First, if we look at the full weight distribution when we have the full set of 384 templates + 3 missing Barth templates, it appears as though the Barth templates are not actually given a substantial amount of weight (in fact, only 6 Barth templates are given any weight in the first place, and scan603.fits is notably absent).
- Moreover, it looks like pPXF is favoring many different spectral types when we open the fit to the full library, whereas the Barth results are placing essentially all the weight on K stars (since the majority of Barth are K stars anyway). F stars in particular are reciving a lot of weight when we open the full template library.
- Limiting the library to K/G+K Stars Alone
- Given the results showing that K stars are far and away dominating the weight of the fits, I tried two additional tests where I limit the library to only K stars (~185 templates) and G+K stars (409 templates), and refit the spectra.
- The results here are a bit intriguing -- when I use only the K stars, I get results quite similar to that of the "12 overlapping" case, which I think makes sense since those are also predominately K type stars.
- What's interesting is that, when I include all G + K stars, the fits actually seem...to move in the right direction? At least to me it seems like a good number of bins are in nice agreement, and the systematic offset in h4/h6/h8 seems to be slightly mitigated. There is a new "wing" of high sigma/h4/h6 points which appear to be inflated relative to Barth, but to my eye this seems like a smaller effect than we have seen before.
-
Note that these are taken from the meeting two days ago, but put here for quick access.
-
One to one plots:
| Barth vs. Barth Reprocessed | Barth vs. 209 Trimmed | Barth vs. 384 Trimmed | Barth vs. 12 Overlapping | Barth vs. 384 Trimmed + 3 Missing | Barth vs. 3 Missing |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
- Histograms of Differences:
| Barth vs. Barth Reprocessed | Barth vs. 209 Trimmed | Barth vs. 384 Trimmed | Barth vs. 12 Overlapping | Barth vs. 384 Trimmed + 3 Missing | Barth vs. 3 Missing |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
- Fit qualities:
| Barth | Barth Reprocessed | Trimmed 209 | Trimmed 384 | Overlapping 12 | Trimmed 384 + 3 Missing | 3 Missing |
|---|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
- Maps of the additive polynomial spatially:
| Barth | Barth Reprocessed | Trimmed 209 | Trimmed 384 | Overlapping 12 | Trimmed 384 + 3 Missing | 3 Missing | |
|---|---|---|---|---|---|---|---|
| Map |  |
 |
 |
 |
 |
 |
 |
- Radial profiles (straight from pPXF):
| Barth | Barth Reprocessed | Trimmed 209 | Trimmed 384 | Overlapping 12 | Trimmed 384 + 3 Missing | 3 Missing | |
|---|---|---|---|---|---|---|---|
| Profile |  |
 |
 |
 |
 |
 |
 |
| Barth Reprocessed | Trimmed 209 | Trimmed 384 | Overlapping 12 | Trimmed 384 + 3 Missing | 3 Missing | 12 Overlapping + 603.fits |
|---|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
-
We wanted to run one test where we take the results from the "trimmed 384" set of stars, selected only the top N set of templates by weight, and refit to see how the kinematics change. It appears as though using the top 15 of the set of 384 trimmed templates results in very nice agreement, suggesting that the 369 lowest weight templates are not contributing apprecaibly to the fit.
-
Here's the 1-to-1 plot for this test, as well as the histograms of differences:
| One to One | Histogram of Differences |
|---|---|
 |
 |
- And here are the RMSs for the two cases:
| Top 15 RMS | Full 384 RMS |
|---|---|
 |
 |
- One thing I was curious about -- when we open up our fit to the full set of 384 templates + the 3 "missing" Barth templates, are the Barth templates given any weight in the first place? Or is pPXF finding an entirely different set of stars that are consistent with the data?
- It seems like it's more of the latter -- in this plot, I've included the weight associated with each template (for all templates with non-zero weight). The Barth subset if makred in red, with the other stars marked in blue. The total additive polynomial is on the left in gray.
- Seems like only 6 of the 15 Barth stars receive any weight, with scan603.fits notably getting ZERO weight.
| 384 + 3 Full Set of Weights |
|---|
 |
- First, here are the results summed over all bins:
| Barth vs. Barth Reprocessed | Barth vs. 209 Trimmed | Barth vs. 384 Trimmed | Barth vs. 12 Overlapping | Barth vs. 384 Trimmed + 3 Missing | Barth vs. 3 Missing |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
And here are the results on a per-bin basis
| Barth vs. Barth Reprocessed | Barth vs. 209 Trimmed | Barth vs. 384 Trimmed | Barth vs. 12 Overlapping | Barth vs. 384 Trimmed + 3 Missing | Barth vs. 3 Missing |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
- Given that the Barth results seem to prefer only K stars (sort of by virtue of Barth basically ony having K stars), I tried to run two extra tests where I limit the template library to only contain K stars, and a second test where the library contains G + K stars.
- I haven't figured out exactly why this is happening yet, but it appears like limiting the library to K stars only gives something similar to ~12 Overlapping case (where Barth appears to be inflated relative to these results). When I add the G stars in the mix, the agreement for a lot of the bins actually appears to be decent, but there is a new "inflated" section relative to Barth which starts to appear. I feel like there is something here that we can work with, but I haven't disentangled it just yet...
| Barth vs. K Stars Alone | Barth vs. G+K Stars |
|---|---|
 |
 |
 |
 |
- I also put together a quick test where I fit the superspectrum with various sets of templates and have some diagnostics below.
- For the main test, I've fit the superspectrum with the set of 384 templates + 3 missing Barth templates from our last few recent meetings. Here's what the fit to the superspectrum looks like using those 387 templates:
| Superspec Fit |
|---|
 |
-
Only 11/387 templates receive any weight from the superspectrum:
- (filename, weights) --> Spectral Type
- (scan423.fits, 0.185) --> Spectral Type = K
- (scan554.fits, 0.146) --> Spectral Type = K
- (scan670.fits, 0.118) --> Spectral Type = F
- (scan638.fits, 0.103) --> Spectral Type = K
- (scan488.fits, 0.081) --> Spectral Type = K
- (scan262.fits, 0.079) --> Spectral Type = F
- (scan671.fits, 0.066) --> Spectral Type = G
- (scan470.fits, 0.050) --> Spectral Type = M
- (scan387.fits, 0.034) --> Spectral Type = M
- (scan628.fits, 0.028) --> Spectral Type = B
- (scan685.fits, 0.000) --> Spectral Type = M
-
And here's what those spectra look like:
| Superspectra Templates |
|---|
 |
- I then took these 11 files and refit our set of 215 bins using ONLY these for the template library:
| Barth vs. 11 Superspectra | Barth vs. 11 Superspectra Differences |
|---|---|
 |
 |
- I also tried adding in scan603.fits into the mix (bringing the total to 12 templates). Notably, scan603.fits being added to this set of templated does not appreciably change the quality of agreement between Barth and this new set of templates, contrary to what we had seen in the "12 overlapping" case.
| Barth vs. 11 Superspectra + Scan603.fits | Barth vs. 11 Superspectra Differences + Scan603.fits |
|---|---|
 |
 |
- It's also worth checking to see if the superspectra alone results agree with other libraries we have created on the way. It appears like superspectrum subset of templates agrees between with the set of (384 + 3 missing) and the (209 trimmed) than we see with Barth. I think this makes sense given how much of a struggle it has been to make 209 trimmed or 384+3 trimmed libraries agree with the Barth set alone, whereas this new superspectrum set of templates is more "data driven" in a sense, rather than just using what Barth says.
| Superspectrum vs. Trimmed 384 + 3 | Superspectrum vs. Trimmed 209 | Superspectrum vs. 12 Overlapping | Superspectrum vs. 12 Overlapping + 603.fits |
|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
- One thing we conjectued in the last meeting was that maybe the absorption lines in the scan603.fits spectrum are in some way different than the other stars in the Barth sample. So to look at this a bit, I've plotted ratios of the Barth spectra below. The lowest line is scan603.fits straight from the FITS file, and the others are each Barth template divided by the scan603.fits spectrum.
- It seems quite clear the scan603.fits has the "shallowest" absorpotion features near the calcium triplet, which is likely the feature that is causing there to be so much weight given to this template/why this template in particular is so necessary for robust Barth results.
| Barth Ratios |
|---|
 |