meeting 2025 03 06 n315 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki

Context

  • I started by taking our newest set of kinematics (bias = 0.2, Barth Stars with the three bins replaced), and reminimized Grid E base models with a proper mask over the central bin. We wanted to verify here that this agrees with the set where we simply are not counting the chi2 in the minimization since this is a more "legit" way to mask this central bin (which looks better now with the new kinematics).

    • RESULT: The two sets of results are in very nice agreement at this stage. The chi2's track each other very well, with a slight slope away from 1-to-1 but not obviously "wrong."
    • The cornerplots are also in excellent agreement at this stage, considering that we're only using Grid E's models alone. I think that the addition of scaling from Grid E + potentially Grid D added to the mix will help.

  • Since the results were not appreciably changed, we'll move forward with this newest set of kinematics by adding scalings of Grid E.

    • For completeness/as a reminder, these new kinematics are produced with (Barth stars + bias = 0.2) except for the three strange bins, which we replaced with the (bias = 0.2, subset of Barth stars) results. We then masked the central Mitchell + Outer 5 Mitchell, as well as the central most GMOS. We do this mask by setting the errors on all moments to 10000.

Diagnostics

  • First, here are the 1d panels for the two minimizations. The left panels show the old approach which simply leave out the 5 central GMOS bins from the chi2 calculation. The right panels show the results in which I keep 4 out of the 5 central bins but mask the very center by increasing its error and reminimizing. The 1d panels in both cases appear to be in great agreement with roughly a ~40 chi2 offset between the two.
1d Panels
images/250306/1d_profiles.png
  • Here's the NNLS chi2 from each case plotted against one another. They track each other very well!
One to One One to One Zoomed In
[images/250306/one_to_one_with_fit.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250306/one_to_one_with_fit_zoomed.png)
  • And here's the result from feeding in the points in the panels above straight to GPR and dynesty. Again, these are considering only models from Base Grid E, so the sampling is not ideal. Even then, the two sets of results agree VERY well at this stage.
Old Masking Procedure New Masking Procedure
[images/250306/new_grid_E_only_original_mask_grid_alpha_K100_nu1.5-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250306/new_grid_E_only_new_mask_grid_alpha_K100_nu1.5-1.png)