New color painting/matching criterion

[mbaumer]

• Do matching based on redshift alone, subject to further cuts:
  • ZSRC < 4.5
  • ZLENS < 3
  • 200 < VELDISP < 300 (target matches to most massive galaxies)

[mbaumer]

Hey @drphilmarshall , I'm using my own naive redshift matching, which just finds the lens/qso in the sloan reference with the closest redshift to each simulated object and assigns the sim that color. Quasar painting looks good (there had been a bug before--on my end, not Adri's), but for the lenses there are a good number of color outliers at low redshift, especially in g-r for z < 0.2 -- any thoughts?

[drphilmarshall]

Looks like some blue galaxies, perhaps with spatially unresolved high
rotation velocities? I suggest cutting them out with a g-r plus i-r color
box (they seem to form a sequence at r-i ~ 0.0 and g-r >~ 0.2 or so) - what
do you think?

On Tue, Jul 14, 2015 at 6:33 AM, Mike Baumer notifications@github.com
wrote:

Hey @drphilmarshall https://github.com/drphilmarshall , I'm using my
own naive redshift matching, which just finds the lens/qso in the sloan
reference with the closest redshift to each simulated object and assigns
the sim that color. Quasar painting looks good (there had been a bug
before--on my end, not Adri's), but for the lenses there are a good number
of color outliers at low redshift, especially in g-r for z < 0.2 -- any
thoughts?

[image: image]
https://cloud.githubusercontent.com/assets/7607631/8674316/0ace2732-2a35-11e5-99ea-1baeb5366de0.png

—
Reply to this email directly or view it on GitHub
#25 (comment).

[mbaumer]

OK, I put in cuts demanding g-r < 3, r-i > 0, and i-z > 0. The plots look much better, but I think now we should discuss the issue of high-redshift colors (column 1 of attached plot).

Right now, my redshift matching is grabbing the colors of the object with the closest z, but for all objects with z > ~1, this is simply the highest redshift object in the reference catalog, meaning all lenses past z=1 have exactly the same color. This isn't ideal, but do you think it's ok for the first pass? I could put in some (true) variation by randomly choosing colors from the N closest reference redshifts or something...

[aagnello]

In fact, for my needs I bypassed it because I was drawing from a distribution with interpolated mean and interpolated covariances. I think this alleviates the problem because it blurs things. If you want to improve on it, besides adding higher-z galaxies, you can try to draw from a sort of de-trended distribution:
-- use spectral templates (Mannucci; Kinney and Calzetti; ...) to set the average colour with redshift
-- use them to determine the average colour as fnct.of z
-- in the detrended colours (swarm-minus-template) at 0<z<1, check if the std.dev. in colour varies with z, extrapolate it to higher z
-- given a galaxy at z>1, draw its magnitudes/colours from the template plus a scatter given by the extrapolated std.dev.
Same goes for the i-band magnitude as fnct. of z and sigma (or whatever magnitude you're using as the reference one).

[drphilmarshall]

Let's go with Adri's first suggestion and investigate templates. I think we'll end up with a more flexible system.