meeting 2025 03 07 n315 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki

Context

  • I've added the Grid D Base models + equivalent scalings to our set of models and run GPR + dynesty on the combined "best scaled" set.

  • Some quick reminders:

    • The kinematics we're using are the (bias=0.2, Barth Stars) except for the three strange bins which use (bias=0.2, subet of Barth stars). Instead of removing the inner 5 GMOS points, we have no included 4 of the 5 as "normal" bins, with the central most GMOS point now being properly masked from the fit. We saw that the results from these kinematics track the original masking procedure quite well.
    • The models we're using are Grid D + E with scalings at s=(0.985, 0.99, 0.995, 1.0, 1.005, 1.01, 1.015). There are ~1000 models in each of Grid D and E, so the total number of models with scalings is ~14,000. This is very close to what we used in the final cornerplot in the paper, with the caveat that we had some additional wider scalings for Grid D in the fiducial results.

  • Results/Takeaways:

    • The addition of Grid D has greatly improved the constraint on T, and the posteriors are looking much better.
    • The black hole is ~3.1 billion with a M/L of ~2.57, so really not too different from the current results the paper.
    • Tmaj and Tmin are near the (0,0) boundary as before, and their 3 sigma contours are not fully resolved just yet whereas they are in our current results, but I'm unsure if we want to put much stock in this since they're already quite near the boundary.
    • There are definitely a few specific models which have not "bottomed out" yet as a function of their scaling, but I don't think adding more scalings to either grid will improve on our current results.

  • Moving forward, I think it just depends on how much more we want to explore constraing Tmaj/Tmin to full resolve the 3 sigma contours, or if we just want to move on with the results below.

Diagnostic Plots

  • First, here's the 1d panels for Grid D, Grid E, and the combined Grids D + E for all the scalings we have.
All Scalings
images/250307/1d_panels_grids_d_and_e.png
  • We can then select the best set of scales, which gives these 1d chi2 panels. I've colored the points according to which grid they're coming from.
Best Scaled Models
images/250307/1d_panels_grids_d_and_e_best_scales.png
  • And finally I took the best scaled Grids D + E set of models and ran them through GPR and dynesty using nIter=8, nu = 1.5, and K=60 and K=100 (which are quite similar to one another, though we typicaly use something like K=60 in our papers. This currenly has ~1000 models within K=60 of the minimum).
K=60 K=100
[images/250307/grids_D_and_E_BEST_SCALES_nIter8_grid_alpha_K60_nu1.5-1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_5/wiki/[[images/250307/grids_D_and_E_BEST_SCALES_nIter8_grid_alpha_K100_nu1.5-1.png)