meeting 2025 03 19 n57 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki
Context
This bullet provides some additional thoughts and diagnostics on reconciling the old and new pPXF fitting routines. While the old and new codes are not yet in perfect agreement, I've subtantially improved the new code's fitting of the N57 spectra. The wonky results were almost entirely due to the initial guess for V and sigma that was in the old code vs. the new code.
In the old code, the default guess was set to [0, 200] km/s for V and sigma, whereas the new code had a default of [0, 1] km/s if there were no other inputs. I hadn't been specifying an initial guess in the old code, so I hadn't checked the new code for this same initial guess.
Updating the new code's initial guess to the same values has substantially improved both the fits and the agreement between the old and new code's kinematics. I no longer have any spectra with strange "wiggles" or artificially deep or shallow absorportion features. I'm still trying to track down what might be the remaining differences between the fits, but the situation is now currently much better than yesterday and think we can say that largely things are consistent with one another between the old and new code now.
Diagnostics
Spectral Fits
First, here's a dropwdown containing the pPXF fits using the new code with the old initial guess of [0,1] km/s compared to the same bins for the initial guess of [0, 200] km/s. As I mentioned above, there are no longer any of the strange, very-wiggly spectra due to this new initial guess, and no spectra resulting in "unphysical" sigmas values.
Start = [0,1] km/s
Start = [0,200] km/s
One-to-One Comparison
Left column: comparison of the old code vs. the new code for the same input settings (correct Barth stars + bias = 0.2 + nMC=100 + adeg=0). Note that the agreement is subtantially improved, with only very minor offsets/roughly scattering on or near the one-to-one line. I haven't done any point-symmetrization here so there's are straight from pPXF.
Right column: comparison of the most recent set of fits with the new code (correct barth stars + bias = 0.2 + adeg = 0, etc) vs. the current kinematics we've been using (which has bias = 0.0, incorrect Barth stars, adeg=0, etc). Considering the vastly different settings and templates used in these realizations, I'm happy with the level of agreement we currently have
Comparison (old code vs. new code)
Comparison (Newest code kinematics vs. current kinematics used in the minimization)
with errors
without errors
Following Email Exchanges
Uploading a few plots and additional information here following some emails back and forth.
In the two plots below, I'm showing the the results with the new code vs old code in the left column (preliminary fits only). In the right column, I'm showing the new code's preliminary fits vs. the current kinematics plotted without errorbars. I don't have the preliminary pPXF fits for the original fit anymore from what I can tell (I think they were lost when my computer died), so this is I think the closest we can get to a one-to-one comparison. Note however that the old kinematics did used substantially different settings, so it's unsurprising there is not a perfect level of agreement.
There are similar plots showing cases for N315 here. I need to look a bit more closely still but it looks like the level of agreement we have in the N57 cases is roughly similar to what we had for N315, with there being slightly larger of a difference between the old and new code for N315 especailly at high sigma.
I'll try to get a few more representative cases up here before the meeting to for us to discuss.
Prelim Fit Comparison (old code vs. new code)
Comparison (Newest code kinematics vs. current kinematics used in the minimization)
