Plotting the bimodal distribution of galaxies

Markus Pössel (Haus der Astronomie), May 2013

This page offers various versions of a histogram that is originally from Blanton et al. 2003 (where it's part of Fig. 7). It must be one of the first comprehensive histograms showing the bimodal distribution of galaxies by color: bluish-starforming on the one hand, "red and dead" (that is, non-starforming) on the other. If you're not familiar with what this means, here's a blog article that explains a bit more.

In its original version, the diagram requires some familiarity with the ways astronomers measure color: with differences of brightnesses measured through different filters, expressed in the arcane astronomical system of magnitudes. Since this is a very cool diagram, I wanted to make it more accessible. That's why I introduced galaxy colors into the histogram as a color gradient, in a semi-realistic way that corresponds to the colors those particular galaxies have on images served by the SDSS Skyserver Object Explorer (more about the color scheme below).

The result looks like this:

Diagrams for download

As this particular form of the diagram might be useful to others, I'm releasing it into public domain. Use it any way you see fit. Here are some different versions for download.

Basic histogram with title:

Basic histogram without title:

In order to explain what a histogram is to people who aren't familiar with the concept, I've also made a version with auxiliary lines delineating the bins:

Histogram with lines and title:

Histogram with lines, without title:

Finally, if you want to make your own version, or use your own data, here is the python file I used for plotting. It uses numpy and matplotlib and reads this data file.

About the colors

To make the diagram more accessible, I wanted to use natural colors, for a given definition of natural. One way would have been to take some of the transformation formulae from the ugriz filter system used by SDSS to Johnson UBV and attempt to reconstruct an as-natural as possible RGB. However, the issue of what are true colors in astronomical images is quite thorny.

Instead, I opted to choose my colors similar to the way these objects are shown in the SDSS images served by the Skyserver Object Explorer, as probably the most frequent version of an SDSS image one is likely to encounter. To that end, I drew a sample of 45 galaxies around z=0.1, in groups of 5 around 9 g-r values from DR8 using the SQL Search. The SQL queries are here. I assumed the top 5 returned would be a representative sample.

Here are some of the galaxies I used (scaled up slightly):

For each galaxy, I extracted color in Gimp in an admittedly less than rigorous way, using a color picker with a 12 px radius, which I centered on the galaxy if the galaxy image appeared small enough for this to result in a fair sampling; in cases where this would have given only the color of the central region, I moved the color picker so that the central region would lie near the circumference.

To characterize color, as opposed to brightness, I pulled the value (or brightness), V, in the HSV system all the way up, leaving hue and saturation unchanged.

I then averaged R, G and B values separately within each of the 9 groups and used python to fit custom-guessed curves to each color. The curves, as well as the data points, can be seen here:

The curve's equations can be found in the file I used for plotting (see above) as Rfunc, Gfunc and Bfunc, respectively.

Many thanks to Kate H.R. Rubin for pointing me to this diagram and for helpful discussions.