meeting 2023 06 13 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki
First, here's an example plot showing the results for P1 in a bit more detail -- this shows the recoveries for r1 through r5 plotted as the posterior in BHxML space. The recovered best fits and 1-sigma values are plotted above the posteriors. The red points show all models within sqrt(2*Nkin) of the minimum chi2 associated with this P. The red point and errorbar represent the "error" if we went the sqrt(2*Nkin) route for our uncertainties.
| Plot |
|---|
| images/230613/test_P1.png |
Result for each P
| P1 | P2 | P3 | P4 | P5 | |
|---|---|---|---|---|---|
| images/230613/test_P1.png | images/230613/test_P2.png | images/230613/test_P3.png | images/230613/test_P4.png | images/230613/test_P5.png |
Here is the plot Emily made last time but also split into each parameter/split into the different P's -- just as a reminder, these plots are showing essentially a "CDF"-like quantity starting at the 50th percentile symmetrically outward insted of the 0th
| Plot from Last time |
|---|
| images/230613/full.png |
- Split by parameter
| BH | ML | MDM | T | Tmaj | Tmin | |
|---|---|---|---|---|---|---|
| images/230613/full_bh.png | images/230613/full_ml.png | images/230613/full_mdm.png | images/230613/full_t.png | images/230613/full_tmaj.png | images/230613/full_tmin.png |
- Split by P and Parameter
I slightly modified her script for these plots, which I think are more or less what we discussed last time -- these are essentially plotting the CDF of the posteriors vs. the "empirical" CDF from our recovered values. Basically I start at the 0th percentile of each parameter/realization, and incrememnts toward the 100th percentile, keeping track of the fraction of models less than a given percentile. In the ideal case this is 1-to-1.
| Plot from Last time |
|---|
| images/230613/test_cdf.png |
- Split by parameters
- Split by P and Parameter
- Last time, I had a few plots showing the results of our models in nearby chi2; we wanted instead to see some models in nearby parameter spaces (which admittedly are a bit difficult to find). Here are some of those diagnostics for P1 for example:
P1 Example
| Case 1 | Case 2 | Case 3 | |
|---|---|---|---|
| GMOS Moments 1 | |||
| GMOS Moments 2 | |||
| GMOS Moments 3 | |||
| Mitchell Moments 1 | |||
| Mitchell Moments 2 | |||
| Mitchell Moments 3 | |||
| Beta Profiles | |||
| Weights (colored by I2, final type) | |||
| Weights (colored by I3, final type) | |||
| Weights (colored by I2, initial type) | |||
| Weights (colored by I3, initial type) | |||
| Long Weights (colored by I2) | |||
| Intermediate Weights (colored by I2) | |||
| Short Weights (colored by I2) | |||
| Box Weights (colored by I2) | |||
| Long Weights (colored by I3) | |||
| Intermediate Weights (colored by I3) | |||
| Short Weights (colored by I3) | |||
| Box Weights (colored by I3) | |||
| Orbit Classifications | |||
| Regular Weights | |||
| Box Weights | |||
| Retrograde Weights | |||
| Regular Weights Summed over Energies | |||
| Box Weights Summed over Energies | |||
| Retrograde Weights Summed over Energies |
We wanted to explore the axisymmetric recovery with meff a bit more to see if we can find a specific w to mulitply our meff surface which would recover our true inclination. I went ahead and tried our GPR + dynesty routine with a host of weights w.
- r1
- r2
- r3
- r4
- r5: Running
For reference, here are the corresponding 1d panels for the cases above to see how successive w's change the landscape:
- r1
- r2
- r3
- r4
- r5
