meeting 2024 09 03 n57 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki

Lots of updates that are a bit all over the place, but I think there is some good progress being made here.

Emily and I have updated the TriOS Tools/TriOS to output the h1/h2 information along with our standard outputs. This allows us to compute the NNLS chi2 for each moment and each bin, and compare those to the kinematic chi2s that we have been working with lately.

Summary/Takeaway Points:

• I've managed to rerun GPR/dynesty on a variety of relevant cases below using the outputs directly from TriOS. These are GPR/dynesty computed with the kinematic chi2, with the NNLS chi2, and with including/excluding dummy moments and the outermost bins. In the past, we hadn't had access to the dummy moments to either exclude/include, and we hadn't been able to compute the NNLS chi2 associated with specific bins/moments.

• With these changes, I think I have managed to converge on a final set of posteriors that agree quite well with one another, and I think one of the lingering issues was how we were treating the dummy moments in the NNLS case. We had been essentially forced to include the dummies in the NNLS chi2, and removing them seems to make the various posteriors generally agree.

• It looks like most of the discrepency between the NNLS and Kinem chi2 is in sigma and in the GMOS bins, as we can see from the comparison plots below.

• Specifically, if we choose the fiducial case of NNLS or kinem chi2, but exclude the dummies and include all bins, the results are roughly in agreement, with the NNLS landscape favoring a slightly large black hole and slightly larger halo (though within uncertainties). Additionally, the kinem chi2 landscape doesn't have the bimodal feature we see in the NNLS landscape.

  • One interesting note is that, as you see below, the largest discrepency between NNLS and kinem chi2 is in the Mitchell data, and excluding the outer 4 Mitchell data from the kinem chi2 GPR/dynesty run brings the NNLS and kinem closer into agreement.
  • It seems like the outer bins are the deciding factor between high/low Tmaj, and the presence of dummies vs. no dummies is driving the change in BH Mass.
  • Running NNLS + excluding the outer bins seems to push the halo and black hole to quite large values, and I don't think there is any reason for us to be including the dummies in our GPR/dynesty.

• There are a few things I think that would still be valuable: I want to go back and look at the LOSVDs again now that I have the full set of information; additionally, I think looking at the discrepency between kinem and NNLS per bin and per moment would be helpful -- we currently are looking at the two separately. We also can still run the slightly shifted scales we discussed after we figure out the Expanse account issues. I don't expect this to significantly change our results, however, given how nicely our GPR and dynesty is already doing. Anything else worth checking out?

I've also begun to look at how the chi2 between the two cases (kinem vs. NNLS) compare by breaking these out by moment. Note that the h1/h2 panels compare h1/h2 chi2 to the V/sigma chi2.

• This case uses all bins and all moments.

I did the same thing broken down by bin, and I think this is a bit more illuminating. This seems to suggest to me that there really is quite a discrepency between the Mitchell data NNLS and kinem values, rather than the GMOS data which seems to correlate quite tightly:

• Note that in the previous version of TriOS/scraping and such, we only had access to the total NNLS chi2 (which includes all bins and all dummies). This is in stark contrast to the kinematic chi2 we have been using, which I've been excluding the dummy moments.
• Also note that in the cornerplots below, I've chosen to show only the K = 50, nu = 1.5 cases. I also ran the results for various K and nu = 0.5, but the wiki here has been running low on space.