The importance of spectroscopic redshifts in EMU Minnie Yuan Mao UTas/CASS/AAO

Outline ATLAS EMU Synergies The importance of redshift information – Photometric – Spectroscopic The need to train photometric redshifts and statistical redshifts Specific case study - WATs

ATLAS Australia Telescope Large Area Survey Uses the Australia Telescope Compact Array in Narrabri, NSW ATLAS is wide and deep –Wide = 7 square degrees –Deep = ~10 µJy rms at 1.4 GHz ATLAS comprises two fields so as to mitigate cosmic variance –CDFS – Chandra Deep Field South –ELAIS-S1 – European Large Area ISO Survey-South 1

CDFS & ELAIS These fields were chosen as they have deep optical, infrared and X-ray data. ~2000 radio sources so far (~30 µJy rms), expect ~16000 when we get down to a rms of 10 µJy

ATLAS  EMU My role in ATLAS has been to obtain spectroscopic redshifts for the radio sources so as to study their cosmic evolution

Cosmic Evolution Luminosity functions EMU: 10μJy rms Flux limit ~0.05mJy (5 x rms) Cf Mauch & Sadler (2007): ~2.8mJy EMU can probe the luminosity function for low-luminosity (> W/Hz) out to z ~ 0.6! 9/23/2015Minnie Mao - University of Tasmania Flux limit of EMU (50μJy)

The importance of redshift information One of the key science goals of EMU, and indeed one of the most pressing questions in astronomy, is to understand how galaxies have formed and evolved with cosmic time In order to do this we need redshift information!

Surveys that EMU with cross-match with (Norris et al. 2011)

Photometric redshifts Photometric redshifts (photo-z) can be used to estimate the distance of an objects The standard method is to fit the photometry to a set of template spectra, drawn from population synthesis models, or using the observed SEDs of low-redshift galaxies These photometric points can be thought about as really really low- resolution spectra EMU will have photometric redshifts for ~30% of its sources by 2013 and 70% by 2020 (Norris et al. 2011) Padmanabhan (2007), this figure shows the way the spectrum of an elliptical galaxy moves through various photometric bands as it increases in redshift.

Problems with photo-z Choice of template? – Galaxies evolve so templates constructed from low-z galaxies do not necessarily match very well at high-z – Using models makes it difficult to determine whether the models are good representations of real galaxies and the application of corrections (such as for dust) is difficult Best results are obtained if large training sets derived from real galaxy spectra are used – E.g. Salvato et al. (2008) for the COSMOS survey – The training set must be analogous to the full dataset! Bell et al. (2009) note that using “state-of-the-art techniques”, typical errors on photometric redshifts using spectroscopic training sets quite small for red galaxies with strong 4000A breaks, but are larger for blue galaxies, which have featureless spectra dominated by emission lines that vary strongly from object to object Furthermore, EMU is a radio survey comprising a larger fraction of “odd” objects

AAT Anglo-Australian Telescope

Spectroscopic redshifts Spectrographs can measure the spectrum of a galaxy and hence the redshift can be derived from doppler shifting of known emission/absorption lines Olden days: a few spectra per night Nowadays: hundreds of spectra can be measured at once (e.g. using multi-object spectrographs like AAOmega on the AAT) note: still marked slower than photometric plates.

Star-forming galaxies Strong Balmer lines Forbidden lines including [OII], [OIII] and [NII]

Elliptical galaxies Emission dominated by G & K stars The strong break at 4000Å is used to determine photometric redshifts

A wealth of data from galaxy spectra Accurate redshift determination High-resolution galaxy spectra allow unambiguous classification of star-forming galaxies and AGN Chemical composition of galaxy BPT diagram (Baldwin et al. 1981) from Peterson book on AGN.

ATLAS: Spec z vs. Phot z Photo-z from Rowan-Robinson et al. (2008) 5 optical bands and 2 IR bands Broad agreement but basically a scatter plot at z>1

Surveys that EMU with cross-match with (Norris et al. 2011)

ECDFS: Spec z vs. Phot z Phot-z from MUSYC survey (Cardomone et al. 2010) and COMBO-17 (Wolf et al. 2004) COMBO-17 uses 17 bands to derive phot-zs MUSYC uses 32 bands to derive photo-zs These are among the best photometric surveys conducted to date! Good agreement but still basically a scatter plot at z>1 (The mean z of EMU sources will be z ~ 1 for SF sources and z~1.9 for AGNs)

Spectral Classifications

SF or AGN We compared our spectral classifications with mid-infrared colour- colour diagrams similar to using the SWIRE data from Spitzer (e.g. Lacy et al. 2004, Richards et al. 2006).

SF or AGN We compared our spectral classifications with mid-infrared colour- colour diagrams similar to using the SWIRE data from Spitzer (e.g. Lacy et al. 2004, Richards et al. 2006).

A quick look at statistical redshifts “Lacy Wedge” Mean z ~ 0.6 Mean z ~ 0.2 Mean z ~ 0.4

Far-infrared Radio Correlation Mean z ~ 0.37 Mean z ~ 0.7 Mean z ~ 0.37 Mean z ~ 0.7

Mid-way summary While spectroscopy provides a wealth of data about an individual source – it is nigh impossible to obtain spectroscopic redshifts for all 70 million EMU sources – but a good training set of spectroscopic redshifts is necessary to train both photometric redshifts and statistical redshifts. EMU will have photometric redshifts for ~30% of its sources by 2013, increasing to 70% by 2020

A Case Study: One of the science goals of EMU is to identify clusters using diffuse radio sources We did this in ATLAS by searching for overdensities around bent-radio sources

WATs in ATLAS z = 0.15z = 0.37 z = 0.38z = 0.22 z = 0.32

Spatial Distribution 9/23/2015Minnie Mao - University of Tasmania

Redshift Distribution Peak at z~0.22! 42 galaxies in 3 ‘peak’ bins Significant to 7σ Velocity dispersion 870 km/s Similar to rich, local Universe clusters undergoing mergers (e.g. A3667, A3376) Inset shows 3 ‘peak’ bins The distribution in the inset is not Gaussian and shows substructure, consistent with merging systems Bin size = 1500 km/s z There are a total of 309 sources with spectroscopic redshifts within a degree of the WAT Bin size (inset) = 375 km/s 9/23/2015 Minnie Mao - University of Tasmania

Spectroscopic vs. Photometric Binsize z=0.05 Photometric redshifts from Rowan-Robinson (2008) Spectroscopic redshifts from our AAOmega observations

Spectroscopic vs. Photometric Binsize z=0.05 Photometric redshifts from Rowan-Robinson (2008) Spectroscopic redshifts from our AAOmega observations

Spectroscopic vs. Photometric Binsize z=0.025 Photometric redshifts from Rowan-Robinson (2008) Spectroscopic redshifts from our AAOmega observations

Spectroscopic vs. Photometric Binsize z=0.025 Photometric redshifts from Rowan-Robinson (2008) Spectroscopic redshifts from our AAOmega observations

Spectroscopic vs. Photometric Photometric redshifts from Rowan-Robinson (2008) Spectroscopic redshifts from our AAOmega observations

This overdensity can only be seen when we have spectroscopic redshifts!

Take home message While we won’t (ever?) get spectroscopic redshifts for all 70 million EMU sources, it’s imperative that we have a good sample of spectroscopic redshifts so that we can train photometric/statistical redshift algorithms. We need to get a complete spectroscopic catalogue for a subset of EMU – We’ve already started doing this in ATLAS!