meeting 2025 03 27 n57 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki
-
Some additional diagnostics on the spectral fitting, including a closer look at the RMS/residuals and the MC realizations/specific bins which seem to be a bit weird. This bullet has a bunch of different tests so things will likely bounce a bit over the place.
-
Takeaways:
- The largest discrepency at this point is between the old code and new code for the incorrect set of settings. The new code with the correct settings is in decent agreement with the old code + incorrect settings. My best guess at this point for the high h4 using the new code + incorrect settings is probably how the spectral mask is being treated? It seems like, looking at the spectral fits below, the mask on BOX is missing some "spiky" features which may be driving the kinematics in some way. However it's not entirely clear to me that we need to worry much, but I'll try to mask these additional regions as a quick sanity check.
-
To orient ourselves, here's a quick comparison of three sets of data:
- "Published" = Produced with old code + incorrect settings, currently what is used in the minimization. I only have the MC realizations of this since the original fits were on my old laptop/run ~4 years ago.
- New Kinematics (New Code) = kinematics produced with the correct settings + new code. Uses the proper Barth stars/starting guess/bias values.
- Reproduction (Prelim) = using the new code with the same settings as the "Published" set of kinematics. These are the prelim set of fits (but I also have the MC versions, too).
-
The key takeaway here seems to be that the high h4 points only seem to appear when we're fitting the data with the new code + incorrect settings. This doesn't seem to be the case when we use the correct settings and the new code. It's not entirely clear why the new code prefers different h4 than the old code for what I think are the same input settings.
| New Code, Correct Settings vs. "Published" | New Code, Incorrect Settings vs. "Published" | New Code, Incorrect Settings vs. New Code, Correct Settings | |
|---|---|---|---|
| MC |  |
 |
 |
| Prelim |  |
- We have seen these a bit before, and a few things stood out.
- First, in the new code with the incorrect settings, there are two bins which had suspicious h4 values (both prelim and MC). The full kinematics for these two bins are:
| Bin | v | v_err | sigma | sigma_err | h3 | h3_err | h4 | h4_err | h5 | h5_err | h6 | h6_err | h7 | h7_err | h8 | h8_err |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 49.79863 | 8.85107 | 327.60508 | 8.42134 | -0.00201 | 0.02020 | 0.10338 | 0.02153 | 0.01728 | 0.02326 | -0.01379 | 0.02523 | 0.08345 | 0.02331 | 0.11009 | 0.02719 |
| 19 | 56.81734 | 7.18033 | 278.63727 | 8.27980 | 0.06473 | 0.01635 | 0.09912 | 0.01800 | 0.05684 | 0.01953 | -0.01938 | 0.02214 | 0.04483 | 0.01763 | 0.01100 | 0.01915 |
- And the same kinematics from the new code + correct settings are:
| Bin | v | v_err | sigma | sigma_err | h3 | h3_err | h4 | h4_err | h5 | h5_err | h6 | h6_err | h7 | h7_err | h8 | h8_err |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 72.74025 | 9.88060 | 288.56591 | 9.56852 | 0.00356 | 0.02192 | 0.03647 | 0.02471 | -0.03388 | 0.02823 | -0.02733 | 0.03058 | 0.05272 | 0.02604 | 0.05129 | 0.02784 |
| 19 | 64.13413 | 6.32078 | 278.15299 | 6.71504 | 0.07163 | 0.01580 | 0.07986 | 0.01680 | 0.04964 | 0.01902 | -0.02751 | 0.01937 | 0.04338 | 0.01571 | 0.01249 | 0.01738 |
- And here are the spectra and the accompanying fits for these two bins. It seems like me (looking at these and now looking back at the full spectal fits on this page), there appears to be a few features that should probably be masked out from this fit/might be causing a bit of weirdness in our spectra and driving a difference between the two sets.
- There appears to be some very large "absorption"-like or "emission"-like features on the red side of some of the spectra which were not included in the initial spectral mask. It's likely that there is some variation in the kinematics coming from these weird features as you can almost start to see in the spectral fits below. I'll look more closely below at how the MC's for these spectra look.
- You can compare the top and bottom rows here to see how this manifests -- I imagine that the "dip" toward the right side (present in the top row, absent in the bottom row) could be contributing to this h4 weirdness. I'll try to mask this and re-fit.
| Case | Bin 1 | Bin 19 |
|---|---|---|
| New Code + Incorrect Settings |  |
 |
| New Code + Correct Settings |  |
 |
- We also noticed that there appeared to be one bin in particular which had a pretty large difference between the "published" sigma value and the value with the new code. This is actually the same Bin 1 posted above. The two lines here are the kinematics between the new code incorrect settings and the "published" data for comparison:
| Bin | v | v_err | sigma | sigma_err | h3 | h3_err | h4 | h4_err | h5 | h5_err | h6 | h6_err | h7 | h7_err | h8 | h8_err |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 (New Code Incorrect settings) | 49.79863 | 8.85107 | 327.60508 | 8.42134 | -0.00201 | 0.02020 | 0.10338 | 0.02153 | 0.01728 | 0.02326 | -0.01379 | 0.02523 | 0.08345 | 0.02331 | 0.11009 | 0.02719 |
| 1 (Published) | 3.538 | 7.394 | 250.082 | 16.660 | -0.020 | 0.018 | 0.039 | 0.026 | -0.054 | 0.028 | -0.031 | 0.034 | 0.028 | 0.026 | -0.069 | 0.074 |
| 1 (New Code Correct Settings) | 72.74025 | 9.88060 | 288.56591 | 9.56852 | 0.00356 | 0.02192 | 0.03647 | 0.02471 | -0.03388 | 0.02823 | -0.02733 | 0.03058 | 0.05272 | 0.02604 | 0.05129 | 0.02784 |
-
I've recreated versions of the spectra fitting plots to hopefully make a bit more sense of why there is this discrepant sigma bin, for example.
-
In the plots below, I've got the spectra + fits using the new code for the new correct kinematics and my attempt at reproducing the original kinematics using the new code (basically comparing the fits immediately in the table above for all spectra):
-
First, here's basically the takeaway of what we wanted to see -- this plot compares the RMS of the fit residual to the pPXF S/N which in fact follows y=1/x exactly.
| Plot |
|---|
 |
Huge plot here -- likely will take a minute to load both sets of plots. The left panels here are the spectra + residuals + RMS value + distribution of the MC moments for the new kinematics (correct settings) using the new code. The right set of plots are showing the new code results with the settings of the old code/"published" kinematics.
| New Kinematics + New Code | Reproduced Kinematics + New Code |
|---|---|
 |
 |
- I've created a set of diagnostic input plots for the New Code + Correct Settings case from above so we can get a sense of how the current iteration looks. Below are some of those diagnositcs:
| Symmetrized | Raw Data | Equivalent "Old" Plot | |
|---|---|---|---|
| bin centers |  |
||
| radial moments |  |
 |
 |
| maps |  |
 |
- And here's one other quick sanity check -- this is showing the GMOS + Mitchell combined maps:
| No Zoom | Zoomed |
|---|---|
 |
 |
- Haven't had the chance to explore this in full depth, but I added two quick additional masks to the cases above and have a quick comparison of the before and after. It seems like my hunch was in the right direction that the mask was affecting the high h4 points in at least one of the bins. Adding the same mask to the correct settings + new code case doesn't seem to change our answers:
| Correct Settings | Incorrect Settings |
|---|---|
 |
 |