Joint Astronomy Centre
Google
Show document only
JAC Home
JCMT
UKIRT
Contact info
JAC Divisions
OMP
Outreach
Seminars
Staff-only Wiki
Weather
Web Cameras
____________________

Observing at UKIRT
Service Observing
UKIDSS Survey Operations
Target of Opportunity
Calibration & Utilities
UKIRT Archive
Public wiki
Accessing Flexed Data
Accessing UKIDSS Data
Reduction Cookbooks
Telescope
Site Quality
Instruments
Newsletter/Publications
UKIRT Faults
JAC Safety Manual

Detection of H-alpha in z=1 Galaxies

Chris Blake1, Karl Glazebrook2, Frossie Economou3

1 : Nuclear & Astrophysics Laboratory, Oxford University, Keble Road, Oxford OX1 3RH, UK

2 : Anglo-Australian Observatory, PO Box 296, Epping, NSW 2121, Australia

3 : Joint Astronomy Centre, 660 N. A'ohoku Place, Hilo, Hawaii 96720, USA

 

Introduction

This report concerns spectroscopic observations of 13 galaxies performed at UKIRT using CGS4. The aim was to try and detect and measure H-alpha emission from NORMAL galaxies at redshifts 0.7 < z < 1.1. At these redshifts, the H-alpha line is shifted into the J-band, which is heavily contaminated with bright OH emission lines from the sky. To cope with this, the galaxies were selected so the H-alpha line would lie at a redshifted wavelength in between these strong sky lines. The CGS4 is a particularly suitable instrument for these observations, as its 150 lines/mm grating is sufficient to resolve (at R=2200) the narrow OH lines (which make up 95% of the sky background). So by choosing the
redshifts of the galaxies to lie between them, we can avoid almost all the OH contamination.

Measurement of the H-alpha luminosity of a galaxy is important, because from it the star-formation rate (SFR) in the galaxy can be inferred. The SFR is related to the H-alpha luminosity because the radiation from young forming stars ionizes the surrounding hydrogen gas, giving rise to H-alpha emission by recombination. Measurement of the SFR is very important in the ultimate scientific goal of understanding galaxy evolution. By measuring the SFR as a function of redshift for various types of galaxies, we may determine at what stage of their evolution different types of galaxies form most of their stars. For example Madau et al (1997) has produced the first 'star-formation history of the universe', which shows a peak at z=1-2, declining both to higher and lower redshifts. A significant result, and in agreement with theoretical models.

However these results were based upon taking the UV luminosity density of galaxies, and converting this in to a star-formation rate via population synthesis models. This would break down for example, if the there was an unexpectedly large amount of dust in high-z galaxies or the model calibration was wrong. Thus an independent check is vital and for this Halpha studies are ideal: the line strength is a direct measure of the number of young OB stars and lies in the red, making it less subject to dust.
 
We selected galaxies from the Canada France Redshift Survey - this is an I<22 selected sample of field galaxies with a high-end redshift tail out to z=1.1. About 50% of the galaxies in the range 0.7<z<1.1 survived our OH-avoidance selection criteria. This sample is particularly suitable as a large fraction of the objects have HST images and have had extensive morphological analysis. A total of 13 galaxies were observed on two separate runs (9-11 May 1996, 3-5 October 1996).

Observations and Data reduction

The project depends on subtracting the bright OH lines to reveal the weaker galaxy spectrum beneath, which typically falls onto 2 or 3 rows of the array. There are two levels of sky-subtraction involved.  Firstly we adopted the standard near IR method of taking 'object' and 'sky' paired observations, where for the second 'sky' exposure, the telescope
is slightly moved so the position of galaxy shifts along the length of the slit, and its spectrum appears in different rows of the array. The first stage of the sky subtraction is to subtract the 'sky' exposure from the 'object' exposure for each pair of images. This will remove much of the sky line intensity - and as the galaxy spectra are in different sets of rows in the pair of images, we end up with a 'positive' and 'negative' galaxy spectrum in the resulting subtracted image. In practice, due to difficulties in keeping the faint galaxy in the slit, the 'positive' spectrum (obtained from the initial 'object' exposure) was usually somewhat stronger than the 'negative' spectrum. However this subtraction is imperfect, because of the strength of the OH lines and their time variability. Because of the high dispersion, and the relative weakness of the sky continuum BETWEEN the lines we used rather longer exposures then usual with CGS4 (15
minutes) to ensure we were background limited. Thus there is still a systematic sky residual.

To deal with this remaining sky residual, we performed secondary sky subtraction by interpolation along the slit, a standard optical technique. There is a standard piece of Figaro software to do this called 'polysky', which fits and subtracts a polynomial from the data points in each array column (in the slit direction). Using it somewhat more complex in the CGS4 case as there is a positive and a negative object row to ignore. CGS4DR handles this by calling 'polysky' from a script for multiple regions, however this gives unsatisfactory results as there are more degrees of freedom and the sky is not constrained to be smoothly varying across the object rows. We adopted the solution of rewriting polysky and using a new script so that it was possible to define three sky regions, and do a global polynomial fit. This gave excellent results.
 
** Figure 1 **
Figure 1 : Top - Object exposure of C14.0600, observed on 11 May. Middle - result of subtracting the corresponding 'sky exposure'. Bottom - result of 2nd order sky subtraction along the slit with improved POLYSKY.
 These stages of data reduction are illustrated by Figure 1. The first image is an object exposure of C14.0600, observed on 11 May. The second image is the result of subtracting the corresponding 'sky exposure'. The OH lines have been considerably reduced in intensity as we can now just about see the object spectrum around row 30. The third image shows the result of the polynomial sky subtraction along the slit. There is now only a faint OH residual visible, (there is always extra Poisson noise) and we can see the positive object continuum clearly, as well as positive and negative H-alpha emission features around (180,30) and (180,42).  A close-up of this region is displayed in Figure 2.
 
** Figure 2 **
Figure 2 : Close up of the H-alpha region at the bottom of Figure 1.
 
For the October data, there was another problem to overcome - the slit on the spectrograph had jammed out of position, slanted relative to the array, which meant that the sky lines were slanted across different columns of the array at about 15 degrees. To deal with this, we wrote some software to straighten up the image by cross-correlating between rows and working out their relative shifts, before performing the usual polynomial subtraction. Figure 3 shows the procedure with the tilted data,  the raw object-sky image, after straightening and after polynomial sky-subtraction.
 
** Figure 3 **
Figure 3 : Correction of tilted data. Top - raw object-sky image. Middle - after straightening. Bottom - after polynomial sky subtraction.
To extract a galaxy spectra with maximum H-alpha emission signal, the array rows were weighted optimally in the extraction. For each galaxy, positive and negative spectra were extracted, added together where appropriate, and corrected to unit exposure time. The resulting galaxy spectra were then flux-calibrated using flux standard stars observed on the nights.

H-alpha fluxes and limits

In 8 of the 13 different galaxies, H-alpha emission was detected. Figure 4 shows the  IR spectra of our 8 detections, and one non-detection for good measure, the vertical dashed line shows the predicted position from the optical redshift (error dz= +/- 0.001-0.002).  In these plots, OH-line regions have been set to zero, as these regions still have higher noise than elsewhere, despite the sky subtraction process. The units on the x-axis are microns, and the flux units are milli-Janskys. The continuum emission has also been subtracted from the spectrum, by determining the mean of the data points in a region 0.02 microns either side of the line (but not including the line, or the masked-out OH regions), and subtracting this mean value from the spectrum. The fluxes under the lines were measured simply by adding up the values of the data points across the range of N pixels encompassing the emission feature. The error in the fluxes due to Poisson noise were obtained from the standard deviation of the continuum data points. In the 5 galaxies where H-alpha emission was not detected the continuum noise was used to place a limit on the flux, assuming a typical line width.
 
** Figure 4 **
Figure 4 : Spectra of our 8 H-alpha detections, and one non-detection for good measure. The vertical dashed line shows the predicted position from the optical redshift (error dz = +/- 0.001 - 0.002). Regions of extra OH residual (more Poisson noise) have been masked out, and the continuum subtracted.
The H-alpha fluxes and limits (in W/m2) were converted to luminosities (in W) using the cosmological distance of the object (taking omega = 1, h = 0.5). The H-alpha luminosity can be converted into a SFR using the scaling given by Kennicutt (1983, ApJ, 272, 54) which is

                SFR(Msun yr-1) = 8.93 * 10-35 * LH-alpha        (W)
 

The SFR by an independent method

We also computed the SFR for each galaxy by a totally independent method, following Madau (1996, MNRAS, 283, 1388). He gives a conversion between the continuum luminosity density at 2800 A and the SFR, calibrated against
population synthesis models. This is quite convenient for us, as at z~1 this corresponds to the observed V band, and we have V band magnitudes for our CFRS galaxies. Thus we eliminate K-correction uncertainty.
 
** Figure 5 **
Figure 5 : Star Formation rate inferred from direct UKIRT/CGS4 observations of H-alpha at z=1, against SFR inferred for the same galaxies from the UV continuum flux, via stellar population modelling, at 2800A (V-band observed frame).
In Figure 5 we plot the SFRs inferred by the two different, totally independent methods against each other.
This is still preliminary work, but of note are several important points. We also plot the one:one line.
 
    (i) We see there is reasonable agreement. The fact that two completely independent and apparently indirect methods of inferring the SFR of a galaxy should produce such comparable results is very significant. It provides strong evidence for the validity of the methods we used to infer the SFR from H-alpha luminosity, which is so important to our understanding of galactic evolution.
    (ii) There is a lack of points with low UV but strong Halpha, (i.e. scattering above the line), as might be the case if for example there was strong dust extinction in many of these objects. Rather we conclude that the UV method of Madau IS giving an approximately fair estimate of the SFR at z=1.
    (iii) There is an interesting set of points with zero Halpha but appreciable UV (i.e. scattering below the line). There may be a number of explanantions for this (some of this light may not be coming from stars for one). But we feel this probably lies in the calibration between UV and star-formation via the population synthesis codes. The 2800A light is coming from a population of young stars, F stars, but of a slightly older age (1-2 Gyr instead of 0.01-0.1 Gyr). Thus even if the models had perfect IMFs you would see variations in the amount of H-alpha/UV produced depending on the episodic vs continuous nature of the star-formation. e.g. if the star-formation was in a burst, just extinguished, there would be UV but no H-alpha.
 In conclusion we are encouraged by the broad agreement with our independent measures of star-formation, the small differences we see are interesting in themselves and tell us about the history and physics of these objects. We should add that because of the high-resolution of our data, the lines are also resolved, so we have velocity information (the velocities are typical) to add to our database. Finally we will be tying together this physical information with the morphological information from the team's HST images.
 

Acknowledgements

We would like to thank the staff of UKIRT for their excellent support in this project. CAB thanks the AAO for receipt of a Summer Vacation Studentship.
 

Back to the Contents
Contact: Chris Davis. Updated: Thu Dec 23 14:49:39 HST 2004

Return to top ^