Diagnostics

Mass Correction

• I first started with the "correction" to the mass we were discussing in the last meeting, hit a wall, and then turned my attention toward the GSMF fitting via the poisson process approach, and also hit a wall there. So this page attempts to just show that work for our discussion.
  • The gaussian process regression becomes impossible with the number of data points in our set, so I had to take an approach where I randomly sample groups of 5k points, and fit a GPR to those data.
  • The result is shown in the plot below -- I show one GP with its confidence interval as the shaded region, but I also include the 10 other mean GPR fits to the data. The nice thing is that they are all consistent with one another AND consistent with the linear fits, so maybe we should just default to the global linear fit for the data?
• The "correction" becomes a little more complicated and where I started to get a little stuck. Technically we're not fitting vs. the K-band stellar mass but rather the Cluver stellar mass, so the second plot shows that relationship in the same redshift bins.
  • It seems like the overall linear trend with Cluver+14 is much weaker than versus the K-band masses. To my eye, this looks much more consistent with a redshift dependence rather than

GPR Fit vs. K Band Stellar Mass	vs. Cluver+14 Stellar Mass
[images/260303/repro_fig_gp_v2.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/260303/repro_fig_cluver14_xaxis.png)

• I've also split this into a plot versus redshift, and also split into mass-binned panels:

vs. Redshift	Mass Binned Panels
[images/260303/repro_fig_z_xaxis.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/260303/repro_fig_z_xaxis_4panels.png)

GSMF Construction

• I've been trying to get reasonable results from Emily's poisson process fitting, and managed to get the code up and running this morning. I have a few results here I'm still working through and trying to understand (as well as making sure my use of the code is correct), but the idea is that:

  • I'm effectively swapping out the MASSIVE mass and redshift for the masses and redshifts inside of my catalog. I've been "building up" the number of points I feed to the fitter just so I can make sure everything runs in a reasonable time, and to ensure that the outputs are reasonable.

• I unfortunately don't have that great of diagnostic plots just yet. I'm having a hard time getting the fits to converge and think I need to be playing around with the priors a bit more. I hope to have a few more robust results in the next day or two, but I just haven't had the chance to get this running properly.

• For now, here are a few attempts:

Getting the code to run	Sampling 100 galaxies from my catalog	Expanding priors a bit
[images/260303/gsmf_fit_gsmf.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/260303/test_gsmf_fit_gsmf.png)	images/260303/test_gsmf_fit_1000_samples_fiducial_prior_WITH_leja_gsmf.png

• And the posteriors from my most recent run (I have tried expanding beyond these but it's getting caught in some local minimum I think). Note that the parameteer order is logphi1, logphi2, logMstar, alpha1, alpha2:

Sample Posteriors
images/260303/sample_posterior.png