Context:

• I've fianlly managed to catch all of our scalings up to speed and have the results posted here.
• Just as a quick reminder, we currently have ~2000 models for N57, and 8 scalings of those 2000 models. The 2000 models were chosen to be identical to 2000/4000 of the final models in the old N57 posteriors, which were computed with the incorrect binning scheme.
• The scale factors are (0.97, 0.99, 0.995, 1.0, 1.005, 1.01, 1.02, and 1.03). In the plots below, I'm including only the (0.99, 0.995, and 1.00) scale factors in the case where I selected the "best scaled" models, because scales outside of this range lead to very lopsided posteriors due to how the points are distributed.

Summary/Takeaways:

• The individual scales themselves look well (given the caveat that some of the posteriors are against the boundary due to where the points lie/get scaled), but the parameters change continutously in a way that I would expected given the 1d chi2 vs. paramters.
• In general:
  • It seems that going from NNLS --> kinem reduces the black hole mass by ~1e9
  • Going from all bins --> no outer 4 increases the black hole mass by ~1e9 and slightly, and changes the Tmaj constraint slightly in the kinem case.
  • Including vs. excluding dummies --> leads to very small changes in the parameters.
• As a quick reminder, I believe we decided that the NNLS, no dummies, all bins case was our "fiducial" case moving forward. I think that this posterior in particular looks alright, especially in the context of the individual scaling results being consistent with one another around scale = 1.0.

Moving Forward:

• I think we have a few options with what we'd like to do. The simplest thing we can do is say that we're done and quote the fiducial case we discussed above. If we wanted to push a bit furhter, we could try to add one more round of models to bring the "active" models from ~2000 to ~3000 or so to ensure that we have sufficient point sampling in the space. If we want to do more on this galaxy, I think that this would be the thing to do since our model points weren't chosen optimally for our data/results (given the issues we've had).
• On the other hand, I don't anticipate the results to appreciably change by adding this round -- if anything, I just expect the posteriors to look a little "cleaner" (that is to say, I think the posteriors are a bit ratty due to there only being ~2000 models here). I will also try to run dynesty+GPR with a very large number of iterations now to see if that cleans anything up.

One other note to keep in mind for the time being:

• I am still running into one strange result that you might get a sense of below. Despite the individual scalings all looking quite well constrained (aside from being on the boundary in some cases), the results from the "best scales" cornerplots below are a bit noisier than I would have expected, and I am a bit surprised to see some of the "best scales" getting suspiciously close to the boundary.
• All of this is to say I am unsure why the individual results can look so great, but the combined "best scales" looked like they're simply noisier. I'm trying a few things our today and will continue to upload any new results on this front/will keep you posted!