meeting 2024 10 30 n315 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki
Context
Following up on email discussions from yesterday where Chung-Pei and I were discussing the "bottomed out" and "non-bottomed-out" subsets of models for Grids C and D.
I've run additional scalings on Grid C (1.02, 1.025, 1.03) and Grid D (0.97, 0.975, 0.98, 1.02, 1.025, 1.03), targeting only well-fitting models (within 50 of the minimum of each grid).
The idea was to increase the fraction of well-fitting models which have their chi2 turn around as a function of scale factor, and these new scalings seem to be doing well at that.
As a reminder, Grid C had 132 models not-bottomed out, and Grid D had 89 models. With the additional scalings, Grid C is reduced to 57 models not bottoming out, and Grid D is reduced to 17 models.
Plots
First, here are some summary plots for each of the cubes.
Grid A
Grid B
Grid B Prime
Grid C
Grid D
All Base Models
And here are the current set of best scaled models from all the grids, plotted in the relevant spaces:
T Space
Angle Space
UPQ Space
Here's a quick CDF-style plot of the best-scaled set of models:
Plot
Focusing in on Grids C and D exclusively:
More Zoomed
Less Zoomed
Cornerplots
I've run our GPR + dynesty routines on the newest set of best-scaled models, and here are those posteriors:
These are still quite rough just as I was adding the new scalings in this morning. I want to test things a bit more here because the linking lenghts associated with this fit are currently at the boundaries (at least it's giving me flags when running the code), and I also want to run things with higher numbers of iterations (still currently nIter=1 to get some fast results).
K=60
K=80
K=100
nu=0.5
nu=1.5
K=60
K=80
K=100
nu=0.5
nu=1.5
(STILL RUNNING) One concern was that our posteriors were being dominated by the single low lying point near Mbh~2.5e9. I've run a second set of dynesty + GPR runs which remove this single model from the fit. Here are those posteriors
K=60
K=80
K=100
nu=0.5
nu=1.5
Investigating Scalings for Grids C and D
Here are some follow ups from yesterday showing the improving coverage of Grid C and D with the additional scalings.
As mentioned above, yesterday we highlighted the 132 models in Grid C and 89 models in Grid D which had yet to bottom out in their chi2 vs. scale factor. With the additional Grid C and Grid D scalings, these numbers are reduced to 57 models and 17 models, respectively.
Grid C
Grid D
Bottomed Out
No Bottom Out
And now here are the 2d plots showing where the remaining bottomed-out/non-bottomed-out models are:
Grid C
Grid D
Grid B Prime Completed
I've filled out the remaining ~100 models from Grid Bprime. There's actually hardly a change in any of the results, but here are two plots showing a quick comparison of the d1 and d3 grids:
