Even the smallest of correlations between galaxies and their environment can impact measurements on the largest scales.
Claire Lamman, Harvard University
September 9, 2022
Understanding the largest structures in the Universe and how they evolve is more convoluted than just mapping out the positions of galaxies. This is because galaxy characteristics are not random; they are tangled both with how we choose targets and the underlying cosmic structure that we’re measuring. Interpreting DESI data involves carefully separating these threads, making sure to leave no loose ends.
An important idea to keep in mind is that DESI is not directly measuring the large-scale structure, because it’s mostly dark matter. Instead, we measure “tracers”: galaxies that trace the underlying web. It’s difficult to choose a sample of tracers which perfectly represent the true density. Certain properties of galaxies make them more likely to be observed, and these properties are often correlated with the structure itself. For instance, we’re more likely to observe big, luminous galaxies, which are more likely to be found in dense clusters.
These ideas are not new. Any survey must be careful when their measurement is correlated with the way they select their sample. Here’s a simple example:
The purpose of DESI is to measure out structure very accurately. This means it’s more important than ever to account for even tiny sources of bias, and it can get complicated. Another example is that denser regions have more active galaxies, which can have more distinct spectral lines that make them easier to get redshifts for. Or we can undersample galaxies in dense regions because they are so close together that DESI’s fibers may collide if we observe some at the same time. Yet another example is that we correct for the dust in our own galaxy, but the dust maps we use include a little bit of dust from beyond the Milky Way! This extragalactic dust is more likely to be found in dense regions, so we can end up over-correcting for dust in denser regions and over-estimate their mass.
Each bias must be carefully accounted for. This can be done by comparing our sample to similar surveys or simulations to check for systematic incompleteness. However, many effects must be carefully studied on their own, and there’s enough examples to fill many papers!
I recently finished one such study, about a bias created by the alignment of galaxy shapes. Elliptical galaxies tend to align themselves with density filaments. That is, their shapes stretch along the direction of strands in the cosmic web. This effect is very small, but becomes noticeable when sampling millions of galaxies. This is tied to target selection because DESI is slightly more likely to choose a galaxy that is pointed at Earth; this is the orientation where a galaxy’s light is the most concentrated on the sky. Therefore, we are more likely to observe galaxies which lie in strands that are parallel to our line of sight, and undersample galaxies in density strands stretching across our line of sight.
The contrast in galaxy clustering we measure in one direction versus another tells us how fast the cosmic web forms (Redshift Space Distortions). Galaxy alignments can bias our measurements of this by about 0.5%. Although this seems tiny, it could make the difference between competing models of dark energy. It’s called “precision cosmology” for a reason! Here is a version of my paper written for the general public.
Even small knots like galaxy alignment can tangle up our measurements, which is one reason you need a large team like DESI to unravel the cosmic web.
In honor of the opening of our online shop, I made a special DESI-gn that highlights both the science and instrumentation parts of our collaboration. “Dark Energy” is represented by a map of large-scale structure, and “Spectroscopic Instrument” is represented by our focal plane and its 5,000 robotic positioners.
LRG selection is done by the cut (red line) in the (r – z) – (z – W1) space. This selection uses the W1 infrared band to separate galaxies (color points) from stars (grey points). The different colors show the redshift of the galaxies in the color–color space. This selection will observe ~615 targets per square degree. (See Zou et al., in prep.)
To avoid stellar contamination (black smooth line) in the target selection, ELG selection uses a cut in (r – z) – (g – r) space. The color histogram is the redshift distribution of ELGs in the color space. This selection will observe ~2,387 targets per square degree. (See Raichoor et al., in prep.)
Only point sources are considered during the selection and the target selection aims to separate quasars from stars. However, the separation between these two classes is less obvious than in the previous cases. A more sophisticated selection has to be applied. Quasars are selected via a Random Forest classification instead of the classical color cuts selection done in previous spectroscopic surveys. The idea of the selection is to separate quasars from stars based on the “infrared excess” of QSOs. The stellar locus is illustrated by the red line and the quasars by the blue/green/yellow points. This selection will observe ~308 targets per square degree. (See Chaussidon et al., in prep.)
