meeting 2025 06 26 n57 - JacobPilawa/TriaxSchwarzschild_wiki_6 GitHub Wiki
Context
- Note this is still a bit of a work in progress as of 6/26 afternoon. Adding some more tests/thoughts to what is below.
- I've refit the N57 data with the new trimmed template library Emily had created in the meeting. This library contains 384 templates with known/obvious spectral types.
- Note I'm also including the quick test in which I re-ran a subset of the line of BH models from N315, this time using the "fixed seed" verison of the code.
N57 Diagnostics
- First, here's some comparisons of the fiducial Barth set of templates (x-axes) for a few different library choices:
- It's not obvious to me that the trimmed libraries are performing differently from one another; I realize that the RMS are different for each moment between each case, but it's not obvious that the systematics are improving or that we are finding a "better" minimum location with the different template libraries.
| Fiducial vs. 384 Trimmed Templates | Fiducial vs. 209 Trimmed Templates (Previous "Trimmed" Set) | Fiducial vs. GKM | Fiducial vs. GKM Main Sequence |
|---|---|---|---|
| [images/250626/fiducial_vs_trimmed_library.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/250512/n57_250624_plot.png) | [images/250528/250528_barth_vs_gkm_stars.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/250528/250528_barth_vs_gkm_main_sequence_stars.png) |
Quick Polynomial Reminders:
- I wanted to include these plots which show the kinematics as a function of adeg/mdeg for our reference. I've excluded the high adeg values from the plot so we can better see variations around our fiducial adeg=0.
| Adeg Varying | Mdeg Varying |
|---|---|
| [images/250626/N57_adeg_testing_zoom_to_small_adeg.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/250626/N57_mdeg_testing_250627.png) |
N315: Fixed Seed Model Results
- I ran a subset of 20 of the models from the line of BH models we ran for the referee report so we can assess how the code performs. It seems to me like the model-to-model variation has decreased (in particular there is pretty solid "line" of continous models in the plot below), but I'm still a bit surprised at the variation between some of the adjacent models. With that said, it appears to my eye like there is less model-to-model variation in the new models, even though we only have ~20 such cases:
- It might be interesting to try the test I had done before, say with only 5 models, to get a sense of how much of the reamining variation is model-to-model vs. potentially something else we are missing in the code? I should check how the NNLS solution gets initialized to ensure that it is deterministic...
| Line of BH w/ Fixed Seed Results |
|---|
| images/250626/all_fit.png |
