Qual
Almost 4 decades ago, Partridge and Peebles predicted that galaxies undergoing their first throes of star formation should be strong emitters in the Lyman-alpha emission line. Coupled with this prediction was the notion that the detection of such young galaxies would:
Motivated by these considerations, observers spent the next 30 years unsuccessfully searching for the expected widespread population of galaxies undergoing their first starbursts. Today I'm going to present my work with Hy Spinrad, which constitutes a modern, highly successful version of the search for these primeval galaxies. Specifically, I'll be presenting results from our various attempts to detect and characterize Lyman-alpha emitting galaxies at redshifts as high as 6.5. At that epoch, the Universe is just 800 million years old, or something like 6% its current age.
The collaborators in this work include some of Hy's former students: namely Daniel Stern (now at JPL), and Arjun Dey (now at NOAO), as well as James Rhoads and Sangeeta Malhotra (both at Space Telescope in Baltimore).
Naturally, this talk will start with the recent history of searches for Lyman-alpha-emitting primeval galaxies, which, as I said, were uniformly unsuccessful at discovering such a population until about 1996. The main issue hampering the unsuccessful searches was a general failure to achieve sufficient limiting flux sensitivities -- a situation which has since been remedied by the commissioning of 10-meter class telescope. That said, the search techniques used in past decades are fundamentally the same as the techniques used today to great success. I'll describe search techniques typically employed to search for galaxies at high redshift, emphasizing the two main techniques I have used while working with Spinrad and collaborators: namely, serendipitous slit spectroscopy searches, and narrow band imaging. I'll highlight results from the serendipitous work, which have been published for the most part, and then I'll present our results from the on-going Large Area Lyman Alpha (or LALA) survey, a narrow band imaging survey which constitutes one of the largest systematic searches for Lyman-alpha emitting galaxies in the young Universe. And of course I'll conclude by outlining the remaining work which I'd like to complete for my thesis, and I'll give the timeline according to which I expect to complete that work, and to write it all up.
Okay. So the salient feature of primeval galaxies relevant to this work can be summarized: "Nascent galaxies undergoing their first major burst of star formation are expected to contain many hot, young, massive stars which ionize interstellar gas." You can think of a primeval galaxy as a giant HII region that has been photoionized in its first burst of star formation. As a consequence, strong emission lines are expected to dominate the galaxy's SED, and for the record, this inference is relatively insensitive to the stellar initial mass function, since it's only the massive stars which produce the UV ionizing radiation which is ultimately responsible for the emission lines.
If you're thinking of it as a giant HII region, what's the size scale?
Lya Luminosity: That said, to detect these galaxies, you might think to exploit the expected emission line spectrum. At redshifts beyond 3, the Lya emission line is observable in the optical passband, and under standard conditions, about two Lya photons are produced for every three ionizing photons from the stars. Thus, Lya emission is a natural choice to serve as the observational signpost of primeval galaxies in formation. And consequently, one of the most intensively applied methods for searching for primeval galaxies historically and today is to search for the redshifted Lyman-alpha line.
To this end, Partridge and Peebles made an optimistic prediction based on converting 2% of gas into metals in 3 x 10^7 years in Milky Way-sized galaxies. This translates into expected line luminosities of 10^44 erg/s. At a redshift of 5 in the concordance cosmology, this luminosity corresponds to a flux of about 5 x 10^-16 erg/cm2/s, which, as you'll see shortly, was a totally accessible flux to spectroscopy and imaging surveys of the last couple of decades.
Continuum Luminosity: By the same token, the continuum luminosity of primeval galaxies is expected to be somewhat faint. The continuum for a constant star formation rate is predicted to be roughly flat between the Lyman and Balmer breaks, and can be expressed like this, where the flux of the galaxy is something like 10^28 times the star formation rate in solar masses per year.
Initial, naive predictions of the star formation rate argued that the formation of 10^11 solar masses in stars in about 10^9 years would result in an average SFR on the order of 100 solar masses per year. Plugging that into this result and putting the galaxy at redshift 5 gives you an expected continuum brightness of about 23.5 mags. Let's fast forward to the present, for a moment: We now know that typical SFRs (uncorrected for dust) are actually more like 10 solar masses per year, which drops the continuum brightness to about 26th mag. Either way, you're left with the sense that you need to pin your hopes on the emission line as your redshift indicator, as direct spectroscopic observation of absorption line signatures in the continuum is a tall order. For the record, if you adopt this emission line luminosity and this continuum flux, you get a line with a rest frame equivalent width of about 100 angstroms.
And this ignores dust which can drop the UV flux density by more than an order of magnitude. 1 mag of absorption in the V band is equivalent to 4.4 mags at 1000 angstroms.
Surface density: Finally, primordial galaxies are also expected to be common: if all L* galaxies have undergone such a phase of rapid star-formation, one should see a surface density of about 10^4 - 10^5 per square degree.
This, of course, assumes no evolution; in a hierarchical formation scenario... Mergers would give you more galaxies at higher redshifts, but I assume their fluxes would be smaller... So for the same flux limit, maybe you catch fewer of those parent galaxies... (Shore this up!)
Maybe add an emergency slide showing the Pritchet Sec. 2.1 estimate: This is based on adopting a B-band luminosity density of the Universe (2.8 x 10^8 L_B,solar/Mpc^3), and then dividing that by the B-band luminosity of an L* galaxy (L_B* = 1.6 x 10^10 L_B,solar) for a local estimate of the number of L* galaxies of n=0.015 Mpc^3. Then just do the solid angle stuff...
So you want for redshifted Lyman-alpha emission. Slit spectroscopy and/or narrow band imaging are natural choices for search techniques, as they are well-suited to take advantage of emission lines.
Deep spectroscopy alone is an efficient means to detect Lya emission from very high-redshift systems. Long spectroscopic observations are sensitive to line-emitting sources which serendipitously fall within the slit, potentially out to a redshift of z = 6.5. This redshift limit is set by the plummeting response of CCDs in the near-IR, as well as by the strong night sky emission lines which dominate ground-based spectroscopy at wavelengths longer than 9300 angstroms.
Spectroscopy is an efficient way to achieve excellent limiting fluxes for detecting unresolved emission lines, because the spectral resolution helps lower contamination by sky background in any given resolution element. In a 1.5 hr spectrum at moderate resolution with Keck/LRIS, the limiting flux probed for a spectroscopically unresolved emission line in a 1" aperture is better than 10^-17 erg/cm2/s, though this limit is strongly wavelength-dependent.
Obviously, the major limitation to this method is that the survey volume is fairly limited. Graphically, you can see that whereas narrowband searches, to be discussed momentarily, probe a thin shell of redshift space, deep spectroscopy probes a pencil-beam of lookback time. So even though you're sensitive in redshift space to Lyman-alpha covering the entire optical window, you're very limited by the small solid angle of your spectroscopic slit. For example, a single longslit on Keck/LRIS is typically 175 arcsec long by 1.0 arcsec wide. In the contemporary cosmology, this corresponds to a comoving volume of about 600 cubic Mpc. This is something like 2 orders of magnitude smaller than what's available in contemporary narrowband imaging surveys.
One way to mitigate this limitation is to adopt the attitude that every time you're doing any kind of reasonably deep slit spectroscopy for whatever reason, you're simultaneously performing a serendipitous spectroscopy survey within the region of the spectroscopic slit not occupied by the primary target. The main advantage of this sort of serendipitous survey is that you get it for free, at no cost to the primary observing campaign.
Before I talk about narrow band selection, let's take a quick look at how such a serendipitous survey works in practice. This figure shows a portion of the Hubble Deep Field flanking field with a projection of a spectroscopic slit. This observation was made as part of Spinrad's campaign of spectroscopic follow-up to Lyman break galaxy candidates in the Hubble Deep Field, and this is actually a scan out the notebook I kept during the project. You can see that the spectroscopic slit hits the target, but you get free spectra of these neighboring galaxies as well.
Lest you think that relying on serendipity is just totally haphazard, it's important to stress the significant role serendipity has historically played in astronomy. Foremost, serendipitous surveys are extremely efficient, since they require no direct allocation of telescope time, and I'll stress shortly they they are directly competitive with and complementary to narrowband surveys. In extra-galactic astronomy alone, serendipitous detections were responsible for the first known object at a redshift beyond 5, and in the mid-to-late nineties, serendipity was the dominant technique for locating the most distant known objects, as the high redshift pundits moved from radio-selection to systematic photometric selection.
In any case, the process of cataloging serendipitous detections proceeds backward from the usual steps involved in compiling redshifts. In the usual scenario, one begins with photometry for a galaxy whose location is known and subsequently obtains a spectrum. In a serendipitous survey, you begin with a spectrum and work backward to the progenitor's location and photometry. To accomplish this, one combines what was known about the observation -- the location of the target, the dimensions and orientation of the target slit, and the position of the target within the slit -- and then reconstructs the position of the slit on the sky. You then map the reconstructed slit image to the target field and thereby identify a posteriori the objects which you in fact observed.
Keep this process in mind, because four years of serendipitous results from Spinrad's campaign of deep spectroscopy in the Hubble Deep Field are going to occupy the middle third of this talk.
So that's simple slit spectroscopy. Narrowband imaging surveys, on the other hand, impose a restriction on the redshift range covered, but they offer a much larger solid angle. With a modern 8k by 8k mosaic CCD, and a narrow band filter that's, say, 80 angstroms wide, you survey a volume of about 10^5 cubic megaparsecs. But by the same token, because you're not dispersing the light, you're a bit more challenged to achieve the faintest flux limits. Whereas slit spectroscopy gets you to better than 10^-17 ergs per second per square centimeter, typical narrow band surveys hover around 2 to 5 10^-17. It's also worth noting that narrowband imaging generally requires follow-up spectroscopy to discriminate Lya lines, for instance, from metal and or AGN emission lines.
What does narrow band imaging look like in practice? Simply, you make an image through a narrow band filter at conveniently located windows in the night sky emission spectrum. If you see emission in the narrow band filter which is not matched by significant emission in some broad band comparison filter, then you know you've got an emission line. I'll come back to the details concerning how you make a guess as to the identity of the emission line. This schematic gives a rough indication of what we're actually doing with a narrow band survey right now (the survey is called LALA: the Large Area Lyman Alpha survey); a full account of that survey, including the details of candidate selection, will occupy the last third of this talk.
In any case, finding Lyman-alpha in these narrow band filters corresponds to finding galaxies at the following redshifts. And in turn, those redshifts correspond these fractions of the total age of the Universe. For comparison, reionization is thought to have finished up by about redshift 6, so these handful of galaxies at z=6.5 are actually sending us Lyman-alpha emission from beyond the epoch of reionization.
So those are the principle observational techniques we'll be dealing with today. As an aside, I should briefly mention one other search technique for distant galaxies, though it doesn't hinge on the galaxy having emission lines in its SED. Steidel et al. in 1992 began systematically exploiting the so-called Lyman break method to identify galaxies at redshift 3. The expectation for a star-forming galaxies is that spectrum in the UV restframe will be dominated by hot O and B stars, resulting in a pronounced continuum discontinuity at the Lyman break (912 A) seen hear in this model SED for a star-forming galaxy at z=3.151; photons more energetic than 912 angstroms get absorbed in the stellar photospheres of the stars whose light dominates the spectrum.
This break can easily be picked off in broad band photometry. The broken lines show the transmission filter of three broad band filters chosen specifically to select the redshifted Lyman break of the galaxy, such that the U-G color will be red, as well as the relatively flat rest UV continuum in G-R longward of the break.
This spectrum, by the way, shows a QSO at redshift z=3.295. In this case, the pronounced spectral break is due to an optically thick Lyman limit system at a slightly lower redshift.
LBGs show Lya in emission about half the time. Continuum mags on this plot correspond to about I_AB = 23 to 25 for the LBGs.
I bring this up because narrow band searches for Lyman-alpha emission can be thought of as complementary to photometric searches for the Lyman break. The Lyman break method at typical sensitives requires moderately bright galaxies, while Lyman-alpha searches can identify intrinsically fainter galaxies, though they will only select the fraction of objects that have strong line emission. These histograms depict the distribution of the continuum I band fluxes for Lyman-alpha galaxies at redshift 4.5 in the LALA survey, and for Lyman break galaxies at a redshift of 3.6 from the NOAO Deep Wide Field survey, both of which I'm involved with. You can see how Lyman-alpha selection allows us to probe galaxies further down the continuum luminosity function than LBG samples alone can achieve. The implication is that a large population of Lya emitters exist which have fainter continua than the limits of photometric, Lyman-break surveys, and it's that population that we probe with narrowband selection.
As a final note, I'll mention that at higher redshifts, intergalactic absorption due to the Lyman-alpha and Lyman-beta forests play an equal or greater role in attenuating the spectrum of star-forming galaxies as does the photospheric discontinuity at the Lyman break. These hydrogen absorptions cause high-redshift galaxies to effectively disappear from bluer passbands; hence, this technique is sometimes dubbed the "drop-out" technique. This image from a review by Stern and Spinrad shows this effect quite dramatically. These three sources are B-drops, because IGM absorption has killed all emission in passbands at B or bluer, corresponding to being located at redshifts around 4. And this object here is an R-drop for precisely the same reason, except that it's redshift of 5.34 means that all light in R-band or bluer has been absorbed by the IGM.
What's F160W? It's NICMOS H-band, AKA 1.6 microns.
Okay, so that's a sampling of search techniques used in the past and today to detect this expected population of primeval galaxies. Recall the first thing I said, which is that the existence of a significant population of Lyman-alpha emitting galaxies residing in the early Universe was predicted nearly 40 years ago, and that astronomers subsequently spent 30 years trying and failing to detect this population.
This plot from a review in 1994 summarizes that situation as it stood just under a decade ago. Plotted are the upper limit to the volume density and apparent Lya luminosity for 16 surveys aimed at detecting redshifted Lya emission. The points can be interpreted as excluding the region to the upper left of each point; that is, for a given survey, the non-detection of primeval Lyman-alpha emitters requires that the population be either less luminous, and/or less numerous.
The solid line is a theoretical Lya luminosity function for primeval galaxies based on the sort of naive picture I showed you earlier. Clearly, primeval galaxies as depicted in that simple model should have been easily detected. But as was finally discovered with the advent of the 10m-class telescopes, the Lya emission from primeval galaxies is indeed out there, it's just on the order of 100 times fainter than was predicted.
So why is this the case? This is worth spending a few moments on, because the same physical effects which prevented the early detection of primeval galaxies are determining the particulars of the population as it emerges in current studies.
PGs are not as massive as expected? The most obvious thing is simply that, fundamentally, young galaxies just aren't as massive as was once thought. The initial models for the formation of primeval galaxies involved the creation of a basically a Milky Way-sized object over just a gigayear of intense star formation. By contrast, the dominant contemporary paradigm for galaxy formation suggests that galaxies start as many smaller subclumps which later merged to form the large galaxies we see today. In this case, the Lya line luminosity should be smaller than predicted, while its equivalent width remains large.
Of course, this also means that the local density of L* galaxies (0.015 Mpc^-3) may substantially underestimate the comoving space density of small objects at high z. To wit, models employed by Baugh et al. and Giavalisco et al. to reproduce the properties of the Lyman break galaxy population suggest that the average L* galaxy today was about 4 subunits at a redshift of 1, and about 14 subunits at a redshift of 5. This kind of thinking definitely increases the expected space density of primeval galaxies (or proto-galaxies, rather) at high redshift, but by the same token, it would lower the flux threshold necessary to detect them.
PGs contain dust? Perhaps more problematically, the assumption the production of Lya by pure recombination in a gaseous medium is almost certainly too simple. Consider the effect of adding dust to the medium: Lya photons experience a large number of resonant scatterings in neutral atomic hydrogen, thereby increasing the path length and the likelihood of dust scattering and absorption. To make matters worse, the known extinction curve for dust peaks in the UV range, such that extinction actually has its maximum around Lya. For example, a modest reddening of E(B-V) = 0.2 will yield a 38% attenuation of Ha luminosity, and a 96% attenuation of Lya! The implication is that even a small amount of dust can quench Lyman-alpha emission.
LBGs are modeled with E(B-V) = 0.1.
And it's a near certainty that primeval galaxies contain dust. On the observational side, local star-formers show less Lya than expected from theoretical Lya to Hb ratios. Lya/Hb is expected to be 33; we see < 10. Small amounts of dust mixed in with the neutral gas is the assumed culprit. On the theoretical side, it appears that substantial metals and dust can be produced in just a few generations of massive stars, requiring less than 10^8 years. This means that even galaxies at the highest accessible redshifts, though young by all other standards, are still old enough to contain dust.
It's important to note, however, that to first approximation, the surrounding continuum should experience the same extinction by dust as does the Lyman-alpha line, such that the equivalent width is not affected. Therefore, in regions where continuum is detected, dust absorption alone cannot explain the absence of Lya emission. Recall that only about half of the continuum-selected galaxies at high redshift show Lya in emission; hence, dust can't be the end of the story.
Lya emission is a strong function of kinematics? On that note, recently it has become apparent that more than the dust content alone, the velocity structure of the neutral gas in a galaxy is the driving factor that determines the detectability of Lya in emission. And the evidence is coming both from star-formers locally, and at high redshift:
Locally, in a sample of 8 HST UV spectra of nearby star-forming galaxies, Kunth et al. found that the primary indicator of Lya is kinematics, not metallicity. In 4 cases, they observed ISM metal lines which were static with respect to the the ionized gas; in those cases, they saw Lya in absorption. In the remaining 4 cases, the metal lines were blueshifted with respect to the the ionized gas by about 200 km/s; in those cases, they saw asymmetric Lya in emission. Significantly, the galaxies in the sample span a metal abundance of more than a factor of 10, and no correlation was found between metallicity and Lya strength (which had been postulated by to go hand-in-hand with the appearance of dust). So what's going on? Doppler shifting the neutral hydrogen away from the Lya line center has the effect of turning off resonant scattering, and thereby significantly diminishing a Lyman-alpha photon's pathlength to dust absorption. Or stated in reverse, the implication is that even in nearly primordial systems undergoing star formation, there is enough dust in the column to suppress Lya emission provided the kinematics of the neutral gas allow resonant scattering of the Lya photons.
What lines? Kunth got Lya, OI 1302, and SiII 1304; the former gets a handle on the HI column; the latter 2 give you a crude estimate of the chemical composition. Importantly, they also give you the velocity of the gas with respect to the systemic velocity of the system (take from the optical emission lines).
The situation is similar at high redshift: About half the Lyman break galaxies show Lya in emission. These lines are asymmetric, and when continuum is observable, is typically shifted by 10^2 km/s from metallic ISM absorption lines.
What's the story on the ubiquitous asymmetry of the Lya emission? Well the emerging picture is that with or without the presence of dust, outflowing gas is an important condition on Lya radiation. Even if the star formation is dust free, the resonant scattering of Lya photons, though it doesn't destroy them, causes them to diffuse out of the neutral hydrogen envelope, such that Lya emission would is extended and likely falls below surface brightness detection limits.
That said, the idea which ties asymmetry to the emergence of Lya photons is that the star-forming galaxy is driving an thick expanding shell of neutral gas. More specifically: for a sufficiently massive starburst, the hot ionized gas created in the vicinity of the stars vents into the halo of the galaxy, where it sweeps up neutral hydrogen into an optically thick shell. Recombination in the ionized gas converts Lyman continuum photons escaping from the surface of the hot stars into line photons. Then, from the vantage of an observer, the near side of the expanding shell absorbs photons on the blue side of the resonant Lya emission line, causing a flux decrement on what would otherwise be the blue wing of the Lya emission. The far side of the shell backscatters Lya photons into the observer's line of sight; as these photons are offset redward by hundreds of km/s from both the rest frame of the galaxy and the approaching side of the neutral shell, they escape the galaxy and impose a pronounced red wing on the emission-line profile. The net effect is to create the P Cygni-like profile ubiquitous in observations of expanding shells, and evidently, ubiquitous in the spectra of high redshift galaxies. We'll fit a model like this in all its gory detail to one the Spinrad group's high-redshift galaxies shortly.
So all this has been to set the stage for the work I've done with Hy over the past four or five years. I don't think there's really any controversy surrounding this first point here. However, we are just now beginning to disentangle the effects of the quantity and kinematics of gas and dust on the emergent Lya profiles in young galaxies, and consequently, on what can be gleaned about the early history of structure and star formation from their detection. And all such work is predicated on the detection and characterization of the population of young galaxies in the first place. Which brings me to my first bit of observational work: a serendipitous survey in the Hubble Deep Field.
So, between February 1997 and February 2001, Spinrad and collaborators obtained deep Keck spectra of photometrically selected high redshift galaxy candidates in the Hubble Deep Field and its environs. This campaign involved slit spectroscopy of roughly 65 high-z candidates, and those observations resulted in at least 125 serendipitous detections. Of those serendipitous detections, 74 resulted in reliable redshifts.
These histograms plot the distribution of serendipitously determined at that time redshifts against all known redshifts determined in targeted surveys. (These histograms show the same sample, by the way; this bottom plot just zooms in on the 0 to 1 redshift region, with finer bins.) The efficiency of a serendipitous is apparent: accounting for those redshifts that were already known, the serendipitous survey contributed an additional 10% to the list of known redshifts within the central HDF, and an additional 30% to the known redshifts in the surrounding fields.
The lion's share of these detections were at redshifts which are too low to be germane to today's talk. You can see in the top histogram, however, that a handful of galaxies were discovered to be at redshifts beyond 4 and 5. These redshifts were uniformly determined thanks to the serendipitous detection of high-redshift Lyman-alpha emission. When these data were published, there were only 3 published Lyman-alpha emitting galaxies at z > 5, so this small spectroscopic samples represented a significant contribution to what was then known about Lyman-alpha emitters at the earliest observable epoch.
Let me forestall for one moment a discussion of the high-redshift detections, as even in the low sample we made some remarkable finds. Foremost, you can see that the serendipitous sample reproduces the redshift clustering seen in targeted surveys. We find that just about 1/3 of our detections fall into redshift peaks at 98% significance. These same structures had been noted by at least one prior author, and they remain in the much larger sample recently produced as follow-up to the expanded GOODS imaging survey in the HDF-N.
I'll just mention in passing two other items of particular interest:
This galaxy at a redshift of about 2 turned out to be interesting for two reasons. Foremost, its redshift is unexpectedly large for an object showing identifiable spiral structure, making it the most distant known spiral galaxy. Furthermore, because of the spiral structure in the optical, and due to its steep radio spectrum and its significant detection by ISOCAM at 15 microns, observers performing multi-wavelength follow-up to the HDF initially classified this source as a low-redshift disk, with star formation as its primary emission mechanism. Our spectroscopy, however, which showed narrow, high-ionization state emission lines, and coupled with a Chandra detection of this source as a hard X-ray emitter, this indicates that this galaxy in fact harbors a comparatively rare, high-redshift Type II AGN.
How do I know its the highest z spiral? Basically, there are at least 2 samples of spirals complete to I and K mags fainter than HDFX28 which show know galaxies beyond z = 1.5 or so.
This image here shows a small galaxy cluster at a redshift of z=0.85. That redshift peak at 0.85 in the histogram I just showed you was largely due to this structure here, for which we have 11 spectroscopic redshifts in a radius of about 45 arcsecs. After the initial redshifts were determined serendipitously, we noticed an over-density of objects with like colors in this particular region. This is an I-band image, and even without taking color into account, you can see that there's a concentration of objects here. But if you select objects which are redder than V-I > 1.5 -- which is what you expect when you redshift the Balmer break to z=0.85 -- you find that the these sources here are over-dense by a factor of 3 compared to the field. Subsequent targeted spectroscopy confirmed the spatially clustered red objects were indeed of like redshift, with a line-of-sight velocity dispersion of something like 600 km/s.
Okay, back to the matter at hand. What did we learn about star-forming galaxies at redshifts beyond 5 in our serendipitous sample? The first thing that must be mentioned is a caveat concerning derived quantities from serendipitous spectroscopy. Because you have no guarantee that the object was optimally aligned with the spectroscopic slit, it is likely that fluxes are underestimates. Nonetheless, at least coarsely, these objects appear to have line fluxes on the order of 10^42 erg/s, which corresponds to star formation rates of about 1 solar mass per year. We generally detect little or no continuum, so the equivalent widths we estimate are made with upper limits to the continuum based on Poisson noise, which means that they are lower limits. The equivalent widths are large -- and this is going to be a theme throughout this talk -- which is suggestive of primordial star formation. More on that shortly; I'll be able to be more quantitative about the properties of individual sources when I turn attention to our narrowband imaging survey results.
Based on the surface area covered by our spectroscopic slits, we estimate a surface density of Lya-emitters at redshift 5 of roughly 2 to 3 sources per arcminute squared per unit-z. At this risk of over-interpreting these small numbers, this abundance is in good agreement with theoretical predictions made by Haimann and Spaans for the evolution of Lya-emitters that follow a Press Schechter formalism, and with the only competing observations available at the time, which was a targeted survey by a Hawaii group covering redshifts around 3 and 4.
One last detail concerning the sources: where the Lya lines are spectrally resolved, we uniformly see asymmetric emission lines, in these sense of a sharp blue cut-off and a broad red wing. This has been interpreted as evidence for the expanding shell model of high redshift galaxies I alluded to above.
This source here at a redshift of 5.631 is typical of these detections. (And I now realize that it's a rather poor example of an asymmetric line, as it's barely resolved. But you'll see plenty of higher resolution spectra shortly...) In any case, what you see in the top spectrum is actually two sources: the continuum object here is actually a foreground galaxy at z=0.64. This isolated emission line is what we interpreted as Lya at z=5.63; it's spatially offset from the continuum source by about 2 arcseconds.. To check for the presence of red side continuum (and therefore, for a spectral break), I fit a Gaussian along the spatial direction of the continuum source on the blue side, and then subtracted that off column-wise along the wavelength direction.
The bottom plot is just a one-dimensional extraction of the subtracted spectrogram. The continuum detection, as you can see, is maybe of slight significance.
And this brings me to the question: why do we believe that this is Lyman-alpha at all? Or, when you have only a solo emission line, how do you rule out other interpretations?
Other lines: Foremost, most alternative line interpretations suggest secondary emission features, which are not observed in this case. This table shows the other common, strong nebular emission lines in the spectra of star-forming galaxies as they would be interpreted for this line at 8000 angstroms or so. First, Ha is out, because you don't see sulfur or nitrogen, and anyway at the resulting redshift, we would have seen the oxygen complex at 5007 as well as 3727. That said, [OIII] 5007 is also out, because you don't see neighboring 4959 or Hb, and anyway at this redshift, we would have picked up [OII] 3727. This -- as is almost always the case -- leaves [OII] as the real possibility for a low redshift interloper, because the [OII]-interpretation does not imply any neighboring lines against which to check our interpretation.
Continuum: That said, we now look at the details of the line itself. In this case, there's no continuum detection to provide any clues. We've already seen that a nice big continuum break would be evidence for Lya; in contrast, a solo line (at even moderate redshift) with detectable continuum at equal strengths on either side of the line must be [OII].
Line profile: The next thing to check is the structure of the line profile. [OII] is a doublet, with a line separation of 3 angstroms. At the implied redshift, this separation becomes just over 6 angstroms (or 200 km/s). Unfortunately, our spectral resolution in the serendipitous survey was generally just about 200 km/s at best; just as often, we were working at lower resolution, sometimes as low as 1000 km/s. This means that we really never had any hope of resolving the [OII] doublet.
On the other hand, the Lya lines themselves are sometimes as broad 3 to 5 hundred kilometers per second, in which case we have a chance to resolve its characteristically asymmetric shape. There's a hint of that asymmetry in this spectrum here, though you'll see much better examples in a minute. And from an observational point of view, here there's a bit of a trade-off. In the lowest resolution spectroscopy, it's unlikely that you'll resolve the line asymmetry, but you're more sensitive to the continuum, so hopefully you pick up the decrement. In higher resolution spectroscopy, you're more likely to see the asymmetry, but less likely to get a continuum detection.
The [OII] doublet is 3726 and 3729.
Equivalent width: This brings us to equivalent width. Large continuum-selected and Ha-selected samples indicate that [OII] very rarely exceeds a rest frame equivalent width of 100 angstroms. In this particular case, we estimate an equivalent width of just about 300 angstroms, which put the rest frame equivalent width of about 150 angstroms on the [OII] interpretation. Such large values for [OII] are very rarely observed.
What do we do about the continuum not being so well detected? The equivalent width is the ratio of the flux in the line to the flux in the continuum. Let's say our continuum detection is consistent with zero; we can still use the noise properties to put a 2-sigma (say) upper limit on the continuum. In that sense, that's as large as the denominator can get, which means the EW is a lower limit. An observation deep enough to detect the continuum would find some number smaller than our limit, in which case the denominator shrinks and the final EW grows. So in general, continuum non-detections result in lower limits to the equivalent width, which means the true values are higher, which is almost always the interesting direction.
In some sense, the crown jewel of the solo-line, serendipitous detections was this galaxy here at a redshift of 5.2, dubbed ES1. It was actually detected in an echelle spectrum obtained in high-resolution follow-up to our work in the HDF. The target in this observation was this source "D16", a Lyman break galaxy at a redshift of 3.125. While ES1 did not constitute the most distant known galaxy at the time of its discovery, the observation was almost certainly the best resolved, highest signal-to-noise detection of Lyman-alpha for a very young star-forming galaxy at such an early epoch.
In this HST I-band image, ES1 appears faint and just barely resolved. The isophotal magnitude comes to about 25.4 AB, which is challengingly faint, but quite typical for galaxies at this redshift. Moreover, because the flux in the I-band is dominated by the emission line, the continuum magnitude is much fainter. That said, the continuum appears to be just barely detected here in the spectrum on the red side, but not at all on the blue side. The integrated flux in the emission translates to a luminosity of almost 10^43 erg/s, which in turn translates into the rather vigorous star formation rate of about 10 solar masses per year.
The most dramatic feature of the spectrum, however, is the very pronounced line morphology, with its sharp blue cut-off and a broad red wing. This detection therefore fit right in with the emerging picture of optically thick expanding regions surrounding star-forming galaxies, most naturally driven by the starbursts that render them visible in the first place.
That said, we decided to capitalize on the excellent resolution of this detection by attempting to probe the emission line structure with a simple model based on the expanding shell picture.
Central emission:
Absorption:
Notice that the centroid of the Lya emission appears to be redshifted by something on the order of 100 km/s. This redshift is clearly not of cosmological origin! That is, it is generated artificially by the absorption of the blue part of the profile, and should not be considered a tracer of the flow of ionized gas. Bear this caveat in mind any time anyone quotes a redshift based on the peak of a Lya emission line; it's likely to be an over-estimate by a couple of hundred km/s.
Additional emission components:
That's this component here, and it should be weaker and broader than the central component: weaker, because only a fraction of the photons get scattered back into the line of site, and broader because resonant scattering in the neutral layer is expected to affect a rather wide range of energies.
Additionally, an ionized region can develop at the external shock front of the shell, producing an additional Lya recombination emission component. And since the foreground component is in front of the neutral shell, it remains unaffected by scattering. The background component, on the other hand, will be fully absorbed by the large column density of neutral matter at the rear-ward edge of the shell.
My model for ES1 did not contain this foreground emission component. However, you'll see evidence for it in later spectra.
Now it remains to put these components together to form the net emergent spectrum. Finally, because this object is at a very high redshift, we must consider the intergalactic medium, which has the effect of attenuating the spectrum by absorption in the Lya forest.
IGM absorption:
So that's the model, and here's how it looks when fit to the emission line. At the end of the day, we find that the outflow velocity necessary to produce this shape is just in excess of 300 km/s.
This conclusion bears on a host of cosmological issues surrounding the evolution of galaxies and the IGM at high redshift. Foremost, it suggests that processed material from ES1 will become available to the IGM, potentially providing the enrichment necessary to account for the amount of metals observed there. Indeed, recent observations of C IV absorption systems along the lines of sight to lensed quasi-stellar objects (QSOs) call for enrichment at increasingly high redshift, beyond even z > 5 (e.g., Aguirre et al. 2001; Rauch, Sargent, & Barlow 2001). In addition, both detailed observations and careful theoretical studies demand a mechanism for preheating the material out of which galaxy clusters ultimately collapse and become bound (e.g., Kaiser 1991; Mushotzky & Scharf 1997 and references therein). Here again, galaxy-scale outflows at high redshift are the likely culprit (e.g., Renzini et al. 1993). Finally, galactic winds have proved important in reproducing the faint-end slope of the observed field galaxy luminosity function in semianalytic models of galaxy formation. Outflows are invoked to suppress star formation in low-mass dark matter halos, either via the direct escape of gas-phase baryons in the outflow itself (e.g., Somerville & Primack 1999) or by ram-pressure stripping of the gas-phase baryons by energetic winds from neighboring galaxies (Scannapieco & Broadhurst 2001).
Dude, you better actually *read* some of these references. Or possibly swap the least-understood out for:
...the expected correlation between strong galactic outflows and the escape of Lyman continuum radiation from star-forming galaxies. This correlation bears directly on the much-debated physical nature and relative contributions of the sources that comprise the UV background, as a significant contribution by sources other than QSOs is required at high redshift, owing to the rapid decline in the space density of optical and radio-loud quasars at z > 3 (Bianchi, Cristiani, & Kim 2001; Madau, Haardt, & Rees 1999). It is likely that star-forming galaxies fill this niche...
I chose absorption and emission components merely in an effort to reproduce the velocity structure in the line, without reference to how the line strengths reflect actual emissivity or absorbing column. We just qualitatively show that both back-scattered photons and photons originating at the internal, ionized layer of the expanding shell contribute significantly to the emergent line profile.
You could be more sophisticated about this. And indeed, a more careful line transfer model of ES1's spectrum in press as of a couple of months ago provides a richer account of the interplay between gas and dust in the system, and constrains the over-all energetics.
Ahn's model fits for, among other things, the HI column density and the dust opacity in the supershell, and the expansion velocity of the supershell. With these quantities, he derives the total kinetic energy in the shell, which he finds to be equivalent to the mechanical output of about 4000 supernovae explosions. From there, you can adopt an initial mass function, and then normalize it such that the integral beyond 8 solar masses gives you 4000 stars -- the progenitors of the supernovae. Now that you have a normalized IMF, you can say how many stars were once in the system, and how many stars are still in it... This turns out to be about 10^5 solar masses, or similar to a globular cluster.
How does the dust opacity come in? It has the affect of an over-all attenuation to the emission, but especially to those photons that are multiply scattered. Without dust, Ahn finds multiple emission peaks corresponding to an additonal redshift by Vexp with each back-scattering event. These extra peaks get nailed by dust, which brings the model back into agreement with observations. Sketchy? Sounds like it, but Mas-Hesse buys it...
For the energy: E = 1/2 M_shell V_exp^2 = 2 pi R_shell^2 N_HI m_HI V_exp^2. That is, he only needs the derived quantities: radius (which he takes from the HST imaging), and the column density and the expansion velocity (which he gets out of his model).
Having modeled the transfer of the Lya radiation through the dust, Ahn also has a handle on the survival rate, or the escape fraction, of Lya photons. (That is, he knows how many Lya photons he puts in, and he knows how many he gets out.) This can be used to determined the un-extincted value of the Lya luminosity, which in turn translates into a dust-corrected star formation rate, which comes out to about 14 solar masses per year.
Finally, if you approximate the dust radius as roughly equal to the wavelength of Lya radiation, and you can use the modeled optical depth to dust absorption to estimate the dust density, and ultimately the dust-to-gas ratio.
That said -- and I realize that this is just way to much text for a slide -- but the following quote is so apt that I wanted to read it to you verbatim. "This galaxy exemplifies the paradigm defining the detectability of the Lya emission line: while most of the line emission has been destroyed by dust absorption, the line is still detectable, and indeed quite prominently, due to the kinematical configuration of the neutral gas surrounding the HII region. If there were no dust at all, the line would remain undetectable unless the neutral gas were expanding. On the other hand, if there were no neutral gas at all, both the Lya emission line and the surrounding continuum would experience roughly the same amount of absorption, so that the equivalent width of the line would remain unaffected, independent of the amount of dust. Only the line intensity would be strongly dependent on the amount of dust." This then, is the legacy of the serendipitous survey -- from the initial first confirmations of the existence of a widespread population of primeval galaxies, to the first ever in situ observations thereof with sufficient signal and resolution to bear fundamentally on the details of the central emitting mechanisms.
One last serendipitous detection: spectroscopic confirmations from our recent Keck/LRIS campaigns are tantalizingly suggest of the detection of a massive structure at z=5.2. The two galaxies pictured here were serendipitously detected within a projected separation of 300 kpc and 400 km/s (and you've already met ES1). Since then, by way of follow-up to the initial serendipitous discoveries, two more galaxies within roughly the same spatial and velocity separation have been spectroscopically confirmed, and 10 more have been photometrically selected to be at z=5.2 within a radius of 1 Mpc.
At z=5.2, 1" corresponds to 6.3 kpc.
This sort of detection fits in well with the contemporary theory. The dominant paradigm for understanding galaxy formation is the dark halo model, in which galaxies form stars quiescently at the bottom of potential wells of massive dark matter halos. The most massive galaxies should form first in regions where the density is the highest. Since these regions are expected to be strongly clustered spatially, it is reasonable to expect high-redshift, large scale structures like we seem to be observing here.
If substantiated, these detections would represent an over-density of Lya emitters rivaling the well-known proto-cluster detected by Venemans et al. at z=4.1, or the quasar pair detected by Djorgovski at z=5.2. At this redshift, the Universe is too young for galaxies with this velocity dispersion to have virialized, implying that we are witnessing infall into a massive, early-time over-density, the highest redshift structure currently known.
Brother, you better bone up time to virialize... Read Venemans, and Djorgovski, and maybe Bahall and Fan 1998.
So that's it for the serendipitous detections. To review, Lya-emitting galaxies at high redshift were only first detected in the mid-nineties, and even then, the number of detected objects remained small. In order to get a statistically useful sample, Sangeeta Malhotra and James Rhoads, along with Spinrad and collaborators, initiated the Large Area Lyman Alpha survey in 1998. I am now going to highlight results from my work on spectroscopic follow-up to that survey.
The LALA survey supplements the excellent broad band photometry of the NOAO Deep Wide Field Survey with deep narrowband imaging aimed at identifying Lya emission at z=4.5, z=5.7, and z=6.5. I'm going to restrict this discussion to the z=4.5 survey, which uses 5 over-lapping filters, each with FWHM of 80 angstroms, as pictured here. The comparison broadband filter is Kron-Cousins R, also pictured. Again, the idea is that a strong Lya line will easily stand out against the continuum background when comparing the filters.
The redshifts surveyed z=4.5, 5.7, and 6.5 correspond to the Universe at 1.2, 0.9, and 0.8 Gyrs old.
LALA concentrates on two fields: Cetus and Bootes, each is 36 x 36 square arcminutes in size. These are the particulars of the narrowband filters employed in the z of 4.5 survey; the actually redshift ranged spanned is 4.37 to 4.57, corresponding to a comoving volume just about 10^6 comoving cubic megaparsecs in each field.
The sensitivity of the survey is competitive with the deepest narrow band surveys to date (down to about 10^-17 ergs/cm2/s), while its solid angle coverage was comparable to all previous surveys combined at the time of its inception. (Since then, a Japanese group has emerged with a narrowband imaging survey of similar size.)
Briefly, candidates are selected based on a 5-sigma detection in a narrow band filter, the flux density of which must exceed the R-band flux density at the 4-sigma confidence level. To guard against foreground interlopers, we set a minimum observed equivalent width of EW_obs > 80 A, and the candidate must not be detected in the Bw-band. That way we guarantee a flux decrement indicative of the intervening Lya forest.
So far, nearly 400 candidates across all the redshifts covered by the survey; about 350 of those constitute the sample at z=4.5. I'm about to talk in considerable detail about the 4.5 sample, but before I do, I'll just mention that we have about 10 sources spectroscopically confirmed at z=5.7, published and unpublished, and one at z=6.5, soon to appear in the literature.
How big a field must you achieve to overcome LSS effects? Of course, having two separate fields helps.
Between 2000 April and 2003 May, we obtained spectra of a cross-section of emission line candidates, again using slitmasks on Keck/LRIS. Out of 25 spectroscopic candidates, we obtained confirming spectra for 18. 17 of the confirmed galaxies are strong Lya-emitters, generally asymmetric and of large equivalent width. One object shows no Lya emission, but there's a large spectral discontinuity that we identify with the onset of absorption by the Lyman alpha forest. Of the 7 targets that were not confirmed, 6 were non-detections, and 1 was a low-redshift interloper, showing [OII] 3727 emission at a redshift of 0.8.
The observations themselves were a mixture of the higher resolution 400l grating -- pictured here -- and of lower resolution 150l grating, which I'll show you in a moment. Each plot in this case highlights the Lya emission line, and because these are the higher resolution observations, you can generally see some hint of asymmetry. Over-plotted are the narrow band filter transmission curves, which show you how this particular object was picked up. These are all very typical of Lya at high redshift. One interesting detail: in two cases we see this hint of blueward emission in an otherwise classically asymmetric line -- which maybe this is evidence for an ionization front, as suggested by the expanding shell model?
In any case, here are the 150l spectra. Thanks to the lower resolution, you can't see much structure in the line, but as I'll show in a moment, you're more sensitive to the continuum on the red side.
So we've talked at length about the ubiquitous asymmetry of Lya emission; this data set offered the first chance to characterize the asymmetry in a systematic way, which of course is the first step in turning asymmetry into a really useful diagnostic of the line identification. That said, we defined two extremely simple measure of asymmetry, shown here. The wavelength-based asymmetry measure says, okay, go to the peak of the emission line, then measure the distance in wavelength out until the line crosses 10% of the peak value, and then take the ratio. (We toyed with other flux cut-offs, and it doesn't seem to make that much of a difference.) The flux-based asymmetry measure goes to the peak, and then just integrates the flux out to the 10% mark on either side, and then takes the ratio.
These points here show a scatter plot of the asymmetry measures applied to the LALA sources. The circles are the redshift 4.5 galaxies I'm talking about today; these stars are 3 galaxies at redshift 5.7, and this asterisk is our one spectroscopically confirmed galaxy at 6.5.
For comparison, I borrowed a sample of [OII] lines at redshift 1 from the excellent dataset of the of the DEEP II team, smoothed them to the appropriate resolution, and then applied the asymmetry measures. You can see, first of all, that the measures track each other pretty well, but more importantly, that the Lya sample segregates fairly well from the [OII] sample. The [OII] emitters are clustered around 0.8 +/- 0.1 in both measures, while every Lya-emitter (save one) satisfies either a_f > 1, a_lambda > 1, or both. The implication is that these empirical measures are a pretty clean diagnostic of the line ID.
Notice that one source I've identified as a Lya-emitter falls down here among the [OII] camp. The lack of asymmetry does not favor the Lya interpretation, on the one hand; the lack of blue continuum does not favor the [OII] interpretation, on the other. The rest frame equivalent width, if Lya, is a reasonable 111 angstroms. Alternatively, the rest frame equivalent width if [OII] is a whopping 340 angstroms. So, either this is an exceptional Lya source, for its lack of asymmetry, or its an exceptional [OII] source, for its incredible equivalent width. I'm only aware of one other [OII] line with a similarly large equivalent width.
Look at Osterbrock to see what is allowed for [OII].
Okay. I now want to turn to the ensemble results. We find a large fraction of Lyman-alpha emitters with very high rest frame equivalent widths. The distribution shown here shows the EWs measured in spectroscopy for a sample of 17 Lya-emitting galaxies confirmed in with Keck/LRIS. Taken at face value, the histogram implies that four galaxies in the sample exceed the rest frame equivalent width of 240 angstroms. That's a significant limit, because 240 angstroms is the maximum possible equivalent width predicted by the canonical models of normal star formation. In the seminal paper on the topic (Charlot & Fall 1993), the equivalent widths of Lya alpha in young star-forming systems of solar metallicity is expected to hover in the 100 to 200 angstrom range, and even then, these large equivalent widths are maintained only in the first million to 10 million years of the system's life.
The worry, of course, is that those equivalent widths that are indicated here as being highest generally have the faintest, and hence least certain, continuum estimates. So to get a handle on what fraction of our galaxies reliably exceed the 240 angstrom mark, given the differences in each object's uncertainty, I made a careful monte carlo simulation, defining probability density functions for each measurement, and then asking the final ensemble how many objects are permitted by the observed distribution to exceed 240 angstroms, and at what confidence. The result is that at least 2 out of the 17 sources exceed the 240 mark at 96% confidence. And since normal star formation isn't powering these sources, we're left with the question: what is?
And this is actually likely to be a systematic underestimate of the emission line equivalent widths. This is the expression we use, which is simply the emission in the line divided by the continuum as measured to the red side of the line, corrected for cosmological expansion. Notably, I have not accounted for IGM absorption, which is expected to cut the emission line flux roughly in half, while not attenuating the red side continuum. So the suggestion is that things are actually a factor of 2 worse than they appear here.
One possibility is that we have uncovered a large population of AGN, which would actually have interesting implications for, for instance, the pace of black hole formation in the universe, and for cosmological backgrounds. The evidence, however, has been amassing against this scenario.
The first (and weakest) piece of evidence against the AGN scenario is that a situation in which all the LALAs are AGN would violate X-ray background. CDF observations indicate that there are 2-3 x 10^3 X-ray sources per square degree with X-ray fluxes corresponding to the Lya flux limit of our sample (using the X-ray to Lya flux of the Normal object). Thus, if all the Lya sources were Type II Q's, we'd violate the X-ray background constraint unless there were no X-ray sources at lower or higher redshifts! That doesn't sound right, and anyway it violates the observation that the CDF redshift distribution peaks at z=0.7-0.8.
Far more problematic is the recent non-detection in deep Chandra/ACIS imaging of the sample. Out of about roughly 150 individual Lya candidates, none was detected with Chandra/ACIS to a 3-sigma limiting X-ray luminosity of L(2-8 keV) = 3.3 x 10^43 erg/s (where the cut-off for a typical quasar is something like 10^44 erg/s).
HDFX28 has an uncorrected fullband luminosity of 9.3 x 10^43 erg/s. This is typical of that population.
The constraint is even stronger in the stacked X-ray image shown here, which suggests a 3-sigma limit to the average X-ray luminosity of L(2-8 keV) < 2.8 x 10^42 erg/s. This figure shows two versions of the stacked Chandra images of the Lya emitters. The left panel shows the stack of all Lya emitters which fall within the off-axis angle of 8 arcminutes; this amounts to about 70 sources, and no cumulative x-ray emission is seen. The right panel shows the stack of just Chandra images of Lya sources whose rest frame equivalent widths exceed 240 angstroms; again, there's no cumulative x-ray flux. The panels are 20" x 20" in size, and the circle is a 2.5" radius circle center on the stacking position, The final limit to the x-ray flux based on this stacking analysis is roughly two orders of magnitude fainter than what is typically observed for even the heavily obscured, Type II AGN.
On top of the X-ray non-detection, the case against AGN is further borne by this optical spectrum, which is the composite of 11 individual galaxy spectra at z=4.5. The narrow physical widths of the Lya emission lines (v < 500 km/s) definitively rule out conventional broad-lined AGN, and is indeed strongly suggestive against narrow-lined AGN, with their typical v ~ 1000 km/s.
And yet more problematically, no individual spectrum shows evidence of the high-ionization state UV emission lines symptomatic of AGN activity (e.g. NV, CIV, HeII, CIII), nor is there evidence of such lines in this composite spectrum. This panel here highlights the spectroscopic non-detection of CIV; this panel here highlights the non-detection of HeII, both typically observed in AGN spectra.
So AGN are out, leaving star formation as the likely power source for our Lyman-alpha emitters. But as I mentioned before, the distribution of EWs cannot be reasonably explained by ordinary stellar populations.
To this end, Malhotra et al. (2002) modeled the the equivalent distribution of the LALA survey Lyman-alpha galaxy candidates with the following three models: a Salpeter IMF with low metallicity gas, an IMF which is heavily skewed toward massive stars in similarly low metallicity gas, and a Salpeter IMF in zero metallicity gas. They also have to assume a luminosity function, and they make Lyman-alpha emission lines by assuming that all the ionizing flux is absorbed by neutral hydrogen and produces two Ly photons per three ionizing photons.
Interpreting the model is a little bit convoluted, so in the interest of time, take my word for it that model A is pretty heavily disfavored; it fails to reproduce the median equivalent width that is observed. Models B and C do a good job of reproducing the median equivalent width, though both struggle somewhat to reproduce the low EW end of the distribution. One idea is that the low-EW end constitutes the slightly more evolved end of the population.
Between that result and the AGN non-detection, it appears that we're closing in on the conclusion that "he identification of a population of large equivalent width Lya-emitters evidently powered by star formation originating in low metallicity gas suggests that we are closing the gap between the first, little-enriched primordial galaxies and the higher metallicities of massive galaxies in the local universe." Indeed, recent numerical studies of the rest frame UV and optical properties of very low metallicity stellar populations indicate that Lya emission increases strongly with decreasing metallicity, far exceeding EW > 500 angstroms for Z < 10^-5 Z_solar. The tantalizing limit of such studies is star formation in zero metallicity gas, the so-called Population III, which constitutes the first bout of star formation in the pre-galactic Universe.
The striking features of massive Population III stars are their high effective temperatures (T_eff = 10^5 K for M > 100 M_solar) and consequent hard ionizing spectra, resulting in the production of 60% more hydrogen-ionizing photons than their Population II counterparts, and up to 10^5 times more HeII-ionizing photons. As a consequence, in addition to high equivalent width Lya emission, a unique observational signature of this primeval population are the HeII recombination lines, e.g. n = 3 to 2 at 1640 angstroms, and n = 5 to 3 at 4686 angstroms, which derive for large HeIII regions. These lines are particularly attractive for a detection experiment since the suffer minimal effects of scattering by gas and decreasing attenuation by intervening dust.
This figure from Tumlinson et al shows one model for the flux of HeII 1640 expected from a Population III cluster as a function of redshift. Two cases are shown: this parameter f_evol comes out of the way Tumlinson model the production of HeII ionizing photons; it accounts for the time evolution of the HeII ionizing continuum relative to the evolution of the HI ionizing continuum (i.e. f_evol is 1 is the evolution of the HeII ionizing photons is the same as the evolution of HI ionizing photons; high f_evol means HeII stays strong -- even brightens -- over the lifetime of the Pop III star; low f_evol means HeII fades). The two values here bracket the extreme cases, based on model atmospheres of massive stars. The top line in each case is an SFR of 20 M_sun / yr; the bottom line is an SFR of 5 M_sun / yr.
The main issue governing f_evol is mass loss in Pop III stars. HeII photons are produced as a direct result of the high temperature of metal free stars. However, evolutionary tracks show that these stars may evolve to cooler temperatures over their lifetimes if they do not experience mass loss. (If they do experience mass loss, hotter layers underneath are exposed (?).) Anyway, if mass loss is unimportant, then the general trend will be for T_eff to lower, and Pop III stars will favor low f_evol. (That is, HeII fades, while HI emission [Balmer lines] brightens.) If mass loss is substantial, then the trend will be toward higher T_eff, and larger f_evol. (That is, HeII emission increases as star of successively lower mass make excursions to higher T_eff.) The two values here bracket the extreme cases, based on model atmospheres of massive stars.
I've marked the flux levels expected at redshift 4.5 here with stars. And this line here shows our detection limits in spectroscopy. In the most optimistic scenario, we'd detect HeII at nearly the same flux levels that we're detecting Lyman-alpha. In the most pessimistic scenario, HeII falls below our detection limit.How to tell HeII from Wolf-Rayet stars vs. HeII from Pop III? Wolf-Rayet stars have really broad lines, much broader than what's expected from nebular emission (according to Alice, bless her).
So *if* there's HeII emission by a Pop III cluster at z=4.5, there's a chance we could detect it. But how feasible is the formation of a Pop. III cluster at the comparatively advanced age of a presumably polluted Universe at at our redshift? Current models suggest that though substantial metals can be produced in just a few generations of massive stars, requiring less than 10^8 years, Population III objects tend to form in the 10^(6.5 - 7) M_solar mass range, just large enough to cool within a Hubble time, but small enough that they are not clustered near areas of previous star formation. In other words, cosmic metal enrichment has proceeded very inhomogeneously, with regions close to star formation rapidly becoming metal polluted and overshooting the critical metallicity necessary for the onset of normal star formation, while other regions remain essentially metal free. Thus, Pop III and normal star formation must have been active at the same time, possibly down to relatively low redshift. Scannapieco predicts Pop III contributing to as much as 10% of the SFR at z = 6, for their most favorable choice of parameters. Thus, having established that observation of Population III objects is likely within the capabilities of current instruments, theorists have even suggested that, in fact, HeII may already been seen in our growing catalogs of solo-line emitters at high redshift, mistaken for Lya.
To that end, this plots the probability that a given Lya detection at a given redshift is due to a cluster of Pop III stars. That is, it shows the fraction of the total numbers of Lya-emitting galaxies at a given redshift predicted to contain metal-free star formation.
The isocontours give the probability that a detected galaxy hosts Pop III stars. The contours are > 10^-2, > 10^-1.5, > 10^-1, and > 10^-0.5. The various panels show the effect of varying feedback efficiencies, parameterized by the value E_g^III. Basically, you can think of E_g^III as the relative ability of forming stars to pollute the Universe with their metals. (For the record, E_g^III is the energy input per unit gas mass in the form of spherical galaxy outflows powered by SN explosions; it has units 10^51 erg/M_sun). The dashed line shows a flux of 1.5 x 10^-17 erg/cm2/s, and the data points are various galaxies from various surveys.
So what do you see? Pop II objects occupy a well-defined region of the luminosity - redshift plane, the extent of which is governed by the strength of the feedback. At low feedback values, the non-zero probability region widens considerably; this is because low feedbacks mean a smaller volume of the universe is polluted, such that Pop III formation can continue to lower redshifts in the high mass, high luminosity objects that form later in hierarchical formation scenarios.
Why does the upper envelope change with feedback and not the lower? Pop III objects occupy a limited mass range, such that they are large enough to cool efficiently, and small enough so that they are not clustered near areas of previous star formation. Turning down the feedback means that you can be yet larger, and so more tightly clustered near regions of existing star formation -- but because the feedback is low, you can get that close without picking up polluted gas.
Our LALA sample has a median luminosity (for a WMAP cosmology) of 6 x 10^42 erg/s and a redshift of 4.5. In the model with the strongest feedback, this implies that we'd have to detect more than 100 sources before the sample would reliably contain 1 Pop III object. In the model with the least feedback (and hence the most metal-free universe), the fraction of Pop III objects climbs to something like 10 percent, in which case (barring Poisson statistics), we've already caught 5 of them.
You can anticipate what I'm going to show next. I've already mentioned that we don't see HeII emission in any spectra by way of ruling out the scenario in which our Lya-emitters harbor AGN. And in this composite spectrum, the HeII flux is formally consistent with zero, with a 2-sigma (3-sigma) upper limit of 13% (20%) of the flux in Lya. The corresponding 2-sigma (3-sigma) upper limit to the HeII rest frame equivalent width is 17 A (25 A).
Now I showed you one model prediction for HeII flux; there's actually been a number of increasingly revised predictions for the surface density and detectability of Pop III clusters in the last few years. This equivalent width limit is sufficient to rule out the youngest zero-metallicity instantaneous burst and continuous SFR models, though metallicities of Z < 10^-7 Z_solar are still permissible. At best, we therefore conclude that this data set cannot corroborate the proposition Tumlinson (2003) that the high equivalent widths of the z = 4.5 narrow band-selected Lya-emitters suggest that the first metal-free stars have already been found.
So what's next? Clearly, this line of inquiry could benefit from a larger sample size, and from spectra with better signal-to-noise. To that end, in the 2003 observing campaign, we obtained spectra of roughly 40 additional candidates at z=4.5 with the multi-object spectrograph Keck/DEIMOS. DEIMOS benefits from enhanced red throughput over LRIS, and by that observing season the LALA sample had benefited from an additional round of R-band imaging, both of which combined to enhance our hit rate from about 70% to about 80%. The result of this campaign was roughly 35 detections of Lya at z=4.5, which more than doubles our sample from LRIS. The data have been largely processed (this required writing an additional reduction package to supplement the excellent software provided by the DEEP team), and they are currently being analyzed.
This is truly just a random sampling of spectra here. You can see right away the enhanced signal-to-noise, and the higher resolution does a magnificent job of bringing out the morphology of the lines. Notice also the we resolve [OII], so our main low redshift interloper is a thing of the past. No matter how you slice it, these new data will afford a powerful lever arm on all of the issues I've raised over the last hour.
Which brings me to my last slide.