The Dark Energy Spectrometer (despec): a multi-Fiber Spectroscopic Upgrade of the Dark Energy Camera and Survey for the Blanco Telescope



Download 4.5 Mb.
Page5/10
Date conversion08.07.2018
Size4.5 Mb.
1   2   3   4   5   6   7   8   9   10

3.D Emission-line Galaxy Selection at High Redshift

Here we describe simulations to investigate selection criteria for a survey targeting emission-line galaxies at redshifts higher than accessible with luminous red galaxies, for BAO and RSD studies. Using the CMC (Jouvel et al. 2009) to generate the flux and S/N for each emission line, the survey is expected to reach the line sensitivities described earlier, namely a signal-to-noise ratio of around 5 for fluxes larger than 3×10-17 erg/s/cm2 from wavelengths 600 to 1000 nm (Fig. 3.2). This is reached for a 30 min exposure, which we take as the baseline.



We can use photometric redshifts from DES + VHS to select high-redshift galaxies. Fig. 3.6 (left panel) shows the redshift distribution of emission-line galaxies that would be obtained with DESpec with a photometric redshift (using ANNz) cut between 1< zphot < 2, i-magnitude<23, and a S/N=5 detection of one of various lines: OII line (blue), H (red), OIIIa (magenta), OIIIb (brown), or H (gold). (Color-color instead of photo-z selection, similar to that of the DEEP2 redshift survey of ELGs over the redshift range 0.7 to 1.4, gives a similar result.) Exercises such as this give a first indication of the spectroscopic success rates, redshift distribution, and numbers of galaxies that would result from such a survey. A wavelength range 600–1000 nm allows H to be detected up to z=0.52 and [OII] to be detected at z>0.6; the upper limit in redshift for [OII] is z=1.7. We reach a total number of 2500 galaxies/deg2 between redshift 1 and 1.7 for galaxies with i<23. Redshift measurements coming from lines other than [OII] provide an additional 300 gal/deg2 in this redshift range. We can further increase the high redshift range of this selection by selecting slightly deeper galaxies and increasing the photometric redshift cut to higher redshifts. In Fig. 3.7 we have done such selection where some of the galaxies with photo-z closer to 1 have been removed from the selection. This selection has around 50 galaxies in a bin of 0.1 in redshift at redshift 1.7, which would be roughly enough to control the shot noise in a power spectrum measurement at these redshifts. In Fig. 3.7 we compare the redshift distribution of such a selection at i<23 and i<23.5 after further cuts in photo-z in order to increase the high-redshift tail with a simple color-cut selection based on three-optical-band data. This shows the extra redshift leverage obtained by deeper optical data in the selection process at the expense of having a more complicated target selection. These selections represent only a few percent of the galaxies available at those magnitude limits: we have tens of galaxies per square arcminute available for selection, yet we are restricting selection to under one galaxy per square arcminute.






Figure 3.6: Example of the yield of a particular ELG target-selection scheme based on the COSMOS Mock Catalog (CMC) simulations, assuming 5-sigma line detections as expected for a 30-minute DESpec exposure.  Galaxies brighter than i=23 are selected that have photometric redshifts between 1 and 2 (left); the distribution of those satisfying a much broader photometric redshift cut between 0 and 3 are shown in the right panel. This shows that selection of high-redshift sources via photometric redshift is effective, and it shows that, as expected, [OII] is the principal spectroscopic feature in this range of redshifts. Furthermore other redshifts can be accessible with other lines, including H, H, and OIII.



Figure 3.7: Predicted redshift distribution for three methods of selecting emission-line galaxies based on the Cosmos Mock Catalog.

In Figure 3.8 we show photometric vs. true redshift for the same flux-limited sample of galaxies. This allows us to estimate the galaxies we miss due to the photo-z selection. For galaxies selected with photometric redshifts between 1 and 2, 4.5% of the galaxies will have true redshifts outside the desired redshift range of 1-2. If we choose instead to select a low-redshift emission-line sample (photo-z’s between 0.5 and 1), 12% of the target galaxies will have true redshifts outside the selection range. This study indicates that DES+VHS imaging can yield high-efficiency target selection of high-redshift galaxies based on photometric redshift cuts.





Figure 3.8: Example of the yield of a particular target-selection scheme based on the COSMOS Mock Catalog (CMC) simulations. If photo-z selection is applied to the DES + VHS catalog, we can select galaxies at high redshift (z=1-2) with 95.5% success and galaxies at intermediate redshift (between 0.5 and 1) with 88% success rate. DES+VHS photometry assumed. Upper panel: ANNz photo-z estimator; lower panel: Le Phare estimator.

Neural networks were used in the above analysis to estimate the photometric redshift of a galaxy with a training set that includes the known redshift of the galaxy. We have also used neural networks to estimate which galaxies would be successful in yielding a redshift if targeted by DESpec. It has been shown that an efficient targeting strategy can indeed be realized using ANNz (Collister & Lahav 2004), and Abdalla et al. (2008) showed that there is some predictability of emission line strength from galaxy colors. This is done in the following way: training-set galaxies with no emission lines were assigned the value zero and training-set galaxies with sufficiently strong emission lines were assigned the value one. Because of the Bayesian nature of ANNz, we can interpret the value generated by the trained network as a probability of each galaxy having yielded a successful redshift. (Strictly speaking, this is not a real probability because it does not include the regularization function used to train the network.) Here we define the output of the network as the ANNz target selection value for that galaxy, ranging generally between zero and one. Using the CMC catalog in the DES+VHS bands, we show in Fig. 3.9 that we can select strong-emission-line galaxies via their target selection values.



Figure 3.9: The number of galaxies per square degree as a function of the ANNz target selection value. The blue line corresponds to galaxies for which we will be able to measure a redshift, the “target galaxies:” they have at least one emission line detected at 5 sigma using DESpec and we assign them a value of 1. The black line corresponds to galaxies that do not have a strong emission line detectable; we assign them a value of 0.

Using 10,000 galaxies as a training sample, we thus derive ANNz target selection values. In a real survey we would have to spend a few nights obtaining a training set by targetting galaxies in a wider color range to make sure that our selection is not slightly biased from possible miscalibrations of our mock catalog. Obtaining this training set with DESpec would be feasible in only a few nights. To define a selection criterion, we plot the cumulative number of galaxies as a function of the target selection value as shown in Fig. 3.10. A cut at a value of 0.8 selects most of the emission-line galaxies and has small contamination from weak-emission-line galaxies. We then derive the Spectroscopic Success Rate (SSR) as the percentage of galaxies for which we will be able to measure a redshift (as before, based on the significance of detection of an emission line of given flux), as shown in Figs 3.10 and 3.11. The dotted lines in Fig. 3.11 represent the SSR for different magnitude cuts, while the solid lines represent the percentage of galaxies compared to the total number of galaxies at the magnitude cut.





Figure 3.10: Cumulative number of galaxies/deg2 as a function of the ANNz target selection value for the target galaxies in blue, the target galaxies with a photometric redshift between [0.5-1] in green and [1-2] in magenta, and the galaxies not targeted in black. Using an ANNz target criterion of 0.8, we are able to reach a SSR of 95% at i<23 AB mag, targeting close to 75% of all the galaxies at i<23. We will have a yield of 24000 galaxies/deg2.










Figure 3.11: Spectroscopic Success Rate (dashed lines) and cumulative number of galaxies (solid lines) as a function of ANNz target selection values for magnitude i<23 in blue, i<22 in green, and i<21 in red. The left figure represents all the galaxies meeting the different magnitude cuts, the middle figure represents the galaxies for which 0.5 phot< 1, and the right figure 1< zphot < 2.
1   2   3   4   5   6   7   8   9   10


The database is protected by copyright ©dentisty.org 2016
send message

    Main page