<<

Proc. Natl. Acad. Sci. USA Vol. 95, pp. 47–52, January 1998 Colloquium Paper

This paper was presented at a colloquium entitled ‘‘The Age of the , , and ,’’ organized by David N. Schramm, held March 21–23, 1997, sponsored by the National Academy of Sciences at the Beckman Center in Irvine, CA.

The universe at z > 5: When and how did the ‘‘dark age” end?

MARTIN J. REES

Institute of Astronomy, Madingley Road, Cambridge, CB3 OHA, United Kingdom

ABSTRACT This paper considers how the first subgalac- We now know that at least some and quasars had tic structures produced the UV radiation that ionized the already formed by a billion . But how much earlier did ?and the ‘‘feedback’’ effects structures form, and what were they like 5 ؍ intergalactic medium before z of the UV radiation on structure formation. The first ‘‘pre- In most cosmological theories, especially those that postu- galaxies’’ may eventually be detectable by their direct UV late adiabatic Gaussian irregularities in the early universe, emission, with characteristic spectral features at Lyman ␣; quasars and large galaxies should thin out beyond z ϭ 5, but high-z supernovae may also be detectable. Other probes of the subgalactic structures may exist even at exceeding 10. —and of the epochs of I shall discuss the effects of the earliest and supernovae ,5 ؍ intergalactic medium beyond z reheating and , are discussed, along with possible production of UV radiation, nonuniform reheating of the links between the diffusion of pregalactic metals and the intergalactic medium (IGM), and the production of the first origin of magnetic fields. heavy elements—and the implications for observations at ultra-high redshifts. I shall discuss some potential observations The structures in our present universe are the outcome of more that can probe pregalactic era, telling us when the first energy than 10 of evolution. Slight irregularities im- input occurred, and what scale of objects already existed at z Ͼ printed at very early eras led to increasing contrasts in the 5. But it may first be useful to summarize our theoretical density from place to place, until overdense regions evolved picture of how the first cosmic structure emerged. into bound structures. Quantitative and detailed (albeit spec- ulative) theories for the origin of the fluctuations (in an Clustering in Hierarchical Models ultra-early inflationary phase, or else as the outcome of topological defects, etc.) can now be tested against increasingly Characteristic Scales in Cold Dark Matter (CDM) Models. detailed and precise data. I will focus on the CDM model. But this is just a ‘‘template’’ The data are of two kinds. (i) CMB fluctuations probe for some more general deductions, which essentially apply to irregularities on scales relevant to observed cosmic structure any ‘‘bottom up’’ model for structure formation. There is no and on larger scales that are still in the linear regime. The minimum scale for the aggregation, under gravity, of cold detected have (according to most assumptions) prop- nonbaryonic matter. However, the baryons constitute a gas agated almost uninterruptedly since the recombination era, whose pressure opposes condensation on very small scales. when the universe had been expanding for less than a million Therefore, the gas does not ‘‘feel’’ the very smallest conden- years. (ii) The other line of evidence on cosmic structure comes sations. In the context relevant here, where the dark matter’s from traditional astronomy. Any acceptable theory must ac- gravity dominates, the baryonic Jeans mass is

count for the present clustering properties of galaxies and dark 3 matter; it must also match the actual universe at all past eras ⍀ T 2 Ӎ ϫ 4ͩ b ͪͩ g ͪ M 5 10 MJ [1] that can be probed observationally. We have for some years J 0.025h2 T known about quasars with redshifts up to 5. Quasars them- rad selves may be associated with atypical (even exceptional) Only on scales larger than this can baryons promptly condense galaxies, so their intrinsic properties are hard to relate to the into bound systems, along with the dark matter. [The proviso general trend of formation. What has been especially ‘‘promptly’’ is nontrivial because, as emphasized by Loeb exciting about recent developments is that the morphology and (private communication), the baryons may be able to condense clustering of ordinary galaxies can now be probed out to similar later into smaller systems.] During the ‘‘dark age’’ the gas redshifts: the powerful combination of the Hubble Space became even cooler than the background (i.e., Tg Telescope (HST) and the Keck Telescope has revealed many Ͻ Trad): if it had cooled adiabatically, with no heat input since galaxies at z Ͼ 3. Also, the absorption features in quasar ϭ recombination, Tg would, at z 10, have been below 5 K. The spectra (the Lyman forest, etc.) probe the history of the ϳ smallest bound structures, with mass MJ, would have virial- clumping and temperature of a typical sample of the universe ized at a temperature a few times larger than Tg(z). Larger on galactic (and smaller) scales. masses would virialize at temperatures higher by a further 1͞ The mystery lies at still higher redshifts, between (in round ͞ 2 ϭ ϭ factor (M MJ) . This virial temperature would be reached not numbers) a million years (z 1,000) and a billion years (z solely by adiabatic compression, but also because of a shock, 5). When the primordial radiation cooled below a few thou- which would typically occur before the radius had decreased by sand degrees, it shifted into the infrared. The universe then a factor of 2. entered a dark age, which continued until the first bound These virialized systems would, however, have a dull exis- structures formed, releasing gravitational or nuclear energy tence as stable clouds unless they could lose energy and deflate that lit up the universe again. How long did the ‘‘dark age’’ last?

Abbreviations: IGM, intergalactic medium; CDM, cold dark matter; © 1998 by The National Academy of Sciences 0027-8424͞98͞9547-6$2.00͞0 Mpc, million persec; NGST, New Generation Space Telescope; PNAS is available online at http:͞͞www.pnas.org. AGN(s), active galactic nucleus(i); yr, .

47 Downloaded by guest on September 28, 2021 48 Colloquium Paper: Rees Proc. Natl. Acad. Sci. USA 95 (1998)

due to atomic or molecular radiative processes—clouds that background intensity gets a boost, because the contributions couldn’t cool would simply remain in equilibrium, being later from remote regions (which dominate in Olbers-type ) incorporated in a larger scale of structure as the hierarchy built are less severely attenuated. The UV is then enough to up. On the other hand, clouds that can cool radiatively will maintain the very high mean ionization implied by transpar- deflate. Most cooling mechanisms are more efficient at higher ency of the IGM beyond the Lyman limit. This means that it temperatures, as well as at higher densities. Once collapse can maintain high ionization of a cloud until it has either starts, it proceeds almost isothermally, so that the internal collapsed to an overdensity exceeding the IGM ratio of ions to Jeans mass falls as the density rises. A virialized, self- neutrals or until it becomes self-shielding (which happens at gravitating cloud that can cool radiatively would eventually go more modest overdensities for large clouds). Until that hap- into free-fall collapse and (perhaps after a disc phase) frag- pens the cooling rate will be reduced by the lack of bound ment into smaller pieces. and consequent elimination of the (otherwise dom- Three ‘‘cooling regimes’’ are relevant during successive inant) ‘‘line’’ contribution to the cooling. phases of the cosmogonic process, each being associated with When this third phase is reached, the thermal properties of a characteristic temperature. the uncollapsed gas will resemble those of the structures (i) For a H–He plasma the only effective cooling at low Ͻ 4 responsible for the observed Lyman–forest lines in high-z temperatures ( 10 K) comes from molecular hydrogen. Even quasars spectra—these are mainly filaments, draining into this process cuts off below a few hundred degrees; but above virialized systems. Such systems have velocity dispersions of 50 that temperature it allows contraction within the cosmic km͞sec and are destined to turn into galaxies of the kind whose expansion timescale. The H fraction is never high, and it is, in 2 descendants are still recognizable. I shall return (section 4) to any case, not a very efficient coolant—indeed, systems that discuss the detectability of these early galaxies. collapse at z Ͻ 10 fail to form enough molecules for effective cooling [e.g., see figure 1 of Tegmark, et al. (1)]—but molec- Evidence for Diffuse Gas at High z ular cooling almost certainly played a role in forming the very first objects that lit up the universe CMB Fluctuations as a Probe of the Ionization Epoch. If the (ii)IfH is prevented from forming, so that molecular 2 intergalactic medium were suddenly reionized at a z, cooling is ineffective, then a H–He mixture behaves adiabat- ϾϾ ically unless T is as high as 8,000–10,000°, when excitation of then the optical depth to Thomson scattering back to zi( 1) Lyman ␣ by the Maxwellian tail of the electrons provides would be efficient cooling whose rate rises steeply with temperature. 3 1 ϩ z 2 ⍀ Because of this steep temperature dependence, gas in this ␶ ϭ Ϫ1ͩ iͪ ͩ b ͪ 0.05h Ϫ [2] regime contracts almost isothermally, so that its Jeans mass es 10 0.025h 2 decreases as the density rises. (iii) The UV from early stars will photoionize some (and (the generalization to a more realistic scenario of gradual eventually almost all) of the diffuse gas. When this happens, reionization is straightforward). Even when this optical depth the HI fraction is suppressed to a very low level, so there is no is far below unity, the ionized gas constitutes a ‘‘fog’’ that cooling by collisional excitation of Lyman lines; moreover, the attenuates the fluctuations imprinted at the recombination Ͻ energy radiated whenever a recombination occurs is quickly era; the photons that are scattered at zi then manifest a canceled by the energy input from a photoionization, so the different pattern of fluctuations, characteristically on larger only net cooling is via bremsstrahlung. The cooling is, in effect, angular scales. This optical depth is consequently one of the then reduced by a factor of 100 (see, for instance, ref. 2). The parameters that can, in principle, be determined from CMB minimum temperature (below which there is a net heating anisotropy measurements. It is feasible to detect a value as from the UV) depends on the UV spectrum and on whether small as 0.1— measurements may allow even the He is doubly ionized: it is in the range of 20,000–40,000°. greater precision, because the scattered component would (These three regimes refer to a H–He plasma. When heavy imprint polarization on angular scales of a few degrees, which elements are present they can dominate the low-T cooling; would be absent from the Sachs–Wolfe fluctuations on that ionization is still important in suppressing the most efficient angular scale originating at trec. channels for cooling.) Twenty-one-centimeter Emission, Absorption, and Tomog- The Role of Molecular Hydrogen and the UV Feedback. The raphy. The 21-cm line of HI at redshift z would contribute to role of molecular cooling at early cosmic epochs has been the background spectrum at a wavelength of 21(1 ϩ z) cm. This considered by many authors, dating back to the 1960s; recent contribution amounts to a brightness temperature of order discussions are given, for instance, by Tegmark et al. (1) and 1͞ 0.05 (1 ϩ z) 2. This is very small compared with the 2.7 K of Haiman, Rees, and Loeb (3). The exact efficiency depends on the CMB and smaller still compared with the nonthermal the density and, therefore, on the redshift when the first collapse occurs. background, which swamps the CMB, even at high galactic latitudes, at the long wavelengths where high-z HI should show However, even at high redshifts, H2 cooling would be quenched if there were a UV background able to dissociate the up. Nonetheless, inhomogeneities in the HI may be detectable, molecules as fast as they formed. Photons of h␯ Ͼ 11.18 eV can because they would give rise not only to angular fluctuations but also to spectral structure (5, 6). If the same strip of sky were photodissociate H2, as first calculated by Stecher and Williams (4). These photons can penetrate a high column density of HI scanned at two radio frequencies differing by (for example) 1 and destroy molecules in virialized and collapsing clouds. (If MHz, the temperature fluctuations due to the CMB itself, to the incident spectrum has a nonthermal component extending galactic thermal and synchrotron backgrounds, and to discrete up to KeV energies, there is a counterbalancing positive sources would track each other closely. Contrariwise, there feedback because the number of photoelectrons is increased, would be no correlation between the 21-cm contributions, and this enhances molecule formation.) because the two frequencies would be probing ‘‘shells’’ in Only a small fraction of the UV that ionized the IGM could redshift space whose radial separation would exceed the have been produced in systems where formation was correlation length. Consequently, it is not necessarily unfea- triggered by molecular cooling. Most must have formed in sible to distinguish the 21-cm background, utilizing a radio systems large enough to have been able to cool by atomic line telescope with a large collecting area. That line radiation effects. There is then a further transition when the medium allows three-dimensional tomography of the high-z HI renders breaks through and becomes completely ionized: the UV this a specially interesting technique. Downloaded by guest on September 28, 2021 Colloquium Paper: Rees Proc. Natl. Acad. Sci. USA 95 (1998) 49

For the 21-cm contribution to be observable, the spin observations is [C͞H] ϭϪ2.5. However, the metal abundances temperature Ts must differ from that of the black-body cosmic are similar to those of Population II stars, where oxygen is the background. The gas would be detected in absorption or in most abundant element and is overabundant by a factor ϳ2 ϭ emission depending on whether Ts is lower or higher than Trad. relative to carbon. With ZJ 0.02, the metallicity of the The hyperfine levels of HI are affected by the microwave Lyman ␣ forest is then Z Ӎ 10Ϫ4. This metallicity should be background itself, by collisional processes, and by Lyman ␣ approximately the same as the mean metallicity of the universe (whose profile is itself controlled by the kinetic temperature). at z ϭ 3 if the Lyman ␣ forest contains most of the baryons, Ts will therefore be a weighted mean of the CMB and gas as is found to be the case in models of structure formation for temperatures. the Lyman ␣ forest similar to those analyzed in refs. 10 and 11. Before there had been any heat input due to the develop- The mean metallicity of the universe could be significantly ment of nonlinear structures, the kinetic temperature would be lower only if most of the baryons were metal-free and in NHI lower than that of the radiation, and the 21 cm would be an Ͻ 1014 cmϪ2 absorption systems, or in a very diffuse, unob- absorption feature. In principle, one might be able to detect served intergalactic medium. incipient large-scale structure, even when still in the linear The ratio of the mass of heavy elements ejected by a star to regime, because it leads to variations in the column density of the energy in ionizing photons emitted over the lifetime of the HI, per unit redshift interval, along different lines of sight (5). star, as derived from models of and supernova When reheating occurs, the situation becomes more com- explosions, turns out to be about constant over the relevant plicated (6). The kinetic temperature can rise due to the weak mass range 10 MJ Շ M Շ 50 MJ, so given a mean metallicity ៮ shocking and adiabatic compression that accompanies the Z we can predict the energy in ionizing photons that was emergence of the first (very small scale) nonlinear structure emitted for each baryon in the universe. According to Madau ៮ 2 ៮ (compare section 2). When photoionization starts, there will and Shull (12), this energy is 0.002 Zmpc per baryon. For Z also, around each HII domain, be a zone of predominantly ϭ 10Ϫ4, 10 ionizing photons were emitted per baryon by the neutral hydrogen that has been heated by hard UV or x-ray stars that produced the Lyman ␣ forest heavy elements photons. This latter effect would be more important if the first (assuming a mean energy of 20 eV per ionizing ). UV sources emitted radiation with a power-law (rather than Most of these photons must also have been absorbed by just exponential) component. neutral hydrogen. A fraction fi will be absorbed internally, in Because the signal is so weak, there is little prospect of the galaxies where the emitting stars were formed; the rest detecting high-z, 21-cm emission unless the signal displays should be absorbed after having escaped their original galax- structure on (comoving) scales of several million parsec (Mpc) ies, in other dense absorbing systems where protons can (corresponding to angular scales of several arc minutes). recombine many times during the reionization epoch (similar According to CDM-type models, the gas is likely to have been to the observed Lyman limit systems). Of the photons that already ionized, predominantly by numerous ionizing sources, were absorbed internally, about 70% resulted in the produc- each of subgalactic scale, before such large structures become tion of Lyman ␣ photons. These will then be scattered many conspicuous. On the other hand, if the primordial gas were times in the host galaxy, until they move to the wings of the ␣ heated by widely spaced, quasar-level sources, each of these Lyman line and escape. We define fd as the fraction of Lyman would be surrounded by a shell that could feasibly be revealed ␣ photons that can eventually escape without being absorbed by 21-cm tomography using, for instance, the new Giant Meter by dust. The mean comoving number density of these Lyman ␣ ៮ Wave Telescope (7). photons is then 7 fi fd(ZϪ4)nb, where nb is the comoving ៮ ϭ ៮ ͞ Ϫ4 baryon density and ZϪ4 Z 10 . The mean surface bright- Detection of First Galaxies at z տ 6 via Lyman ␣ Features ness of the Lyman ␣ emission line from the galaxies is simply obtained by multiplying the photon comoving number density ϳ ͞ ␲ Ϫ7 ៮ ⍀ 2͞ Arguments that 10 UV Photons Were Generated for Each by c 4 , which yields 10 fi fd ZϪ4 ( bh 0.023) photons Ϫ2⅐ Ϫ1⅐ Ϫ2 ⍀ 2 Baryon. According to most ‘‘hierarchical’’ cosmological mod- cm sec arcsec . The fiducial value of bh used here is els, the UV that reheated and photoionized the IGM before close to that implied by a primordial deuterium abundance z ϭ 5 came from stars that formed in dark matter clumps with D͞H Ӎ 3ϫ10Ϫ5, as reported by Tytler, Fan, and Burles (13). total mass greater than 108 solar masses and internal velocity The Likely Luminosity of Individual ‘‘Galaxies.’’ The above dispersion greater than about 15 km͞sec. What is the chance arguments determine the average surface brightness of the of detecting these ancient stellar systems? high-redshift sky in the Lyman ␣ emission line or, in other Estimating the cumulative amount of activity at high red- words, the product of the number of galaxies per solid angle shifts is straightforward. The total integrated UV production and their individual, mean Lyman ␣ flux. But the ease of beyond z ϭ 5 must have been enough to ionize the IGM and detection obviously depends crucially on whether the UV build up the UV background, whose strength can be inferred background comes from huge numbers of individually ultra- directly out to z ϭ 5 from models of the Lyman forest, etc. faint systems or (more optimistically) at least in part from Therefore, we have a firm lower limit to the ionizing UV systems of higher luminosity. The uncertainty here is much generated. The total amount could be substantially above this greater—it is essentially tied to the poorly understood effi- limit, because much could ‘‘go to waste’’ through reprocessing ciency of forming massive stars in low-mass dark matter halos in dense clouds, local absorption in the sources, etc. with shallow potential wells. A separate estimate of the number of O and B stars at these If star formation were highly efficient in all collapsed halos early eras comes from the heavy element abundance in high-z with velocity dispersion ␴ տ 15 km⅐sϪ1, corresponding to the absorption clouds. The diffuse gas in the Lyman ␣ forest was virial temperature T ϭ ␮␴2͞k Ӎ 104 K, then reionization would Ϫ already enriched to a heavy element abundance z ϳ 10 2ZJ by take place as soon as the highest-density peaks on the scales of a redshift z ϭ 3. The evidence comes from absorption lines of these low-velocity dispersions collapse. For a top-hat spherical CIV and other species associated with Lyman ␣ forest lines region of mass M that turns around when the age of the տ 14 Ϫ2 ͞ with NHI 10 cm (ref. 8 and references therein). Detailed universe is tf 2, the relation between mass and velocity dis- 3͞ ϭ 2͞ ␲ ␴3 ϳ 9 ␴͞ ⅐ Ϫ1 3 ͞ 9 calculations of the expected column densities of the observed persion is M 2 (2 G) tf 10 MJ ( 20 km s ) (tf 10 absorption lines, using hydrodynamic simulations of the Ly- yr). If all the baryons turned into stars over a timescale of order man ␣ forest and realistic models for the spectrum of the of the free-fall time of the halo, the star formation rate would Ϫ Ϫ ionizing background (ref. 9; U. Hellsten, R. Dave, L. Hern- then be 0.3MJ yr 1(␴͞20 km⅐s 1)3. For a normal IMF, this can quist, D. H. Weinberg, and N. Katz, unpublished work), have yield a luminosity close to the peak luminosity of a supernova. shown that the carbon abundance needed to reproduce the The Lyman ␣ luminosity from this efficient, rapid starburst Downloaded by guest on September 28, 2021 50 Colloquium Paper: Rees Proc. Natl. Acad. Sci. USA 95 (1998)

52 ␴͞ ⅐ Ϫ1 3 ͞ Ϫ1 ϳ would be 10 fi fd ( 20 km s ) (fb 0.1), photons per sec . be 400 times rarer, which still implies a number density of a ␣ ⍀ϭ ϳ The corresponding Lyman flux at present (for 1 and H0 few galaxies per square arc minute with AB magnitudes of 28 ϭ ⅐ Ϫ1⅐ Ϫ1 Ϫ6 ␴͞ ⅐ Ϫ1 3 ͞ Ϫ 70 km s Mpc )is10 fi fd ( 20 km s ) (fb 0.1) [1 (with the optimistic assumption of a rapid and efficient ϩ Ϫ1/2 Ϫ2 ϩ Ϫ1 Ϫ2⅐ Ϫ1 (1 zf) ] (1 zf) photons per cm sec . If the sources starburst in these more massive galaxies). that reionized the universe have these characteristics, then we Detection with ground-based telescopes should be possible at can use the mean Lyman ␣ surface brightness obtained earlier z Ͻ 6. The Keck telescope can detect point sources to AB to conclude that the number density of these sources in the sky magnitudes R Ͻ 28, and I Ͻ 27, in a night of observing time [M. should be ϳ1 per arcsec2 for the fiducial numbers we have Rauch, personal communication; Cohen (23, 24)]. Galaxies up to ϳ ϭ chosen, and a formation redshift zf 10. The redshift zf should z 6 can be detected in the I band; for higher redshift, the of course depend on the detailed model for the amplitude of magnitude limit from a ground-based telescope degrades rapidly the primordial fluctuations on small scales. due to the high atmospheric background at ␭ Ͼ 8,500 Å. The The main problem with this scenario is that supernova magnitude limit could be improved significantly with adaptive explosions in galaxies of such low velocity dispersions may well optics. And the HST detects galaxies in the expel the gas before more than a small fraction has turned to down to I Ӎ 28.5 (ref. 21). The New Generation Space Telescope stars. This leaves us with two other options regarding the (NGST) could image galaxies to AB magnitudes ϳ31 in the sources of the reionization photons. (i) If just a small fraction near-infrared and should be able to detect galaxies to a much of the gas in each object turns to stars before the remainder is higher redshift (24) with a much higher number density than has expelled in a wind, then each galaxy emits fewer photons. been seen so far. The prospect for detecting the high-redshift Therefore, more have to form from lower amplitude peaks. galaxies responsible for the enrichment of the Lyman ␣ forest has The luminosity of each galaxy would be reduced. (ii)Star also been analyzed by Haiman and colleagues (25) and by Cen (R. formation may be so inefficient in the low-escape-velocity Cen, unpublished work). halos that reionization has to await the formation of more In general, the lensing magnification in rich lensing clusters massive galaxies with ␴ տ 100 km⅐sϪ1 (photoionization has may be used here to stretch the magnitude limit (see also R. Cen, little effect on the cooling rate in halos with velocity dispersion unpublished work). As an example, a lensing cluster with an ␴ տ 100 km⅐sϪ1; see refs. 14 and 15). Einstein ring radius b ϭ 30Љ should magnify to A Ͼ 10 an area of Some moderately massive galaxies are in any case expected ϳ30 arcsec2 in the source plane. In the example used above, about to form before reionization, because the power spectrum at 30 galaxies with AB ϭ 32 could be in this area, which would be small scales in cold dark matter models flattens to a slope of magnified to AB ϭ 29.5. Magnified images of high-redshift about n ϭϪ2.7 (so the amplitude of fluctuations decreases galaxies should characteristically appear in pairs around the very slowly with scale), implying that at the same epoch, halos critical lines, in a region that can be predicted from lensing models with ␴ ϭ 70 km⅐sϪ1 (with a mass ϳ40 times larger than halos [see Miralda-Escude´and Fort (26), Kneib et al. (27), and refer- with ␴ ϭ 20 km⅐sϪ1) would be collapsing from 3 Ϫ ␴ peaks on ences therein], so this should help in their identification. These this larger scale. In a Gaussian theory, the 3 Ϫ ␴ peaks should numbers indicate that a new deep field (similar to the Hubble contain ϳ10% as much mass as the 2 Ϫ ␴ peaks at a fixed Deep Field) imaged with HST in a rich cluster, adding also the epoch; so we see that the mass distribution of these first- H and J filters in the near-infrared, might well identify several generation galaxies should probably extend well above the galaxies at z Ͼ 5. In fact, the largest redshift object known at minimum mass for efficient atomic cooling. present (at z ϭ 4.92) is already a gravitationally lensed galaxy (M. Detectability of Galaxies Beyond the Reionization Redshift Franx, G. D. Illingworth, D. D. Kelson, P. G. van Dokkum, and zi. What is the chance of detecting these galaxies? Given the K.-V. Tran, unpublished work). mean surface brightness of Lyman ␣ photons derived previ- The foregoing discussion has assumed that these high- ously, the surface brightness in the UV continuum follows with redshift galaxies would be sufficiently small to remain unre- little additional uncertainty, because the stars emitting most of solved. Resolved objects would need to have higher fluxes to the light in the range 1,216 Å Ͻ ␭ Շ 2,500 Å are also mostly be detected, because the detection is limited by the sky very young stars. For every Lyman ␣ photon produced from background. The likely dimensions of the star-forming regions ionizing photons there should be ϳ10 UV continuum photons. are ϳ100 parsec (pc), corresponding to angular sizes ϳ0.01 The mean surface brightness from all galaxies in the rest-frame arcsec—not resolvable even with NGST. UV continuum is 10Ϫ6 photons cmϪ2⅐secϪ1⅐arcsecϪ2. The main The Lyman ␣ Emission Line. Another possible way to detect spectral feature that should identify any such galaxies at z տ 5 the faint galaxies is to search directly for the Lyman ␣ emission is a sharp break of the UV continuum at the Lyman ␣ line. As we discussed before, for a normal starburst spectrum wavelength, due to the Gunn–Peterson trough (16). In addi- we expect ϳ10% of the UV photons to be in the Lyman ␣ tion, the Lyman ␣ emission line may be present, depending on emission line if dust absorption is not important. Therefore, if dust absorption and scattering of the Lyman ␣ photons. Notice the sensitivity for detecting galaxies is still limited by the sky that the redshift at which the IGM was reionized is only background for emission-line searches, the width of the line marginally relevant to the presence of the Gunn–Peterson should be ⌬␭͞␭ Ͻ 0.01 to allow detection on a shorter time edge, because even if the medium was reionized at z ϾϾ 5, we than for the UV continuum. The width of the Lyman ␣ line in ␣ know that the flux decrement caused by the Lyman forest emission from a region of neutral gas with column density NHI Ӎ ϭ 22 Ϫ2 ␴ ϭ ␴ ⅐ Ϫ1 reaches a factor of 2 at z 4 and grows rapidly with redshift. 10 N22 cm and velocity dispersion 10 6 km s is 1͞ Ӎ ⌬␭͞␭ Ӎ ϫ Ϫ3 ␴ 3 Thus, the technique of identifying galaxies at z 3 from the 2 10 (N22 6) (28). Thus, the sensitivity of Lyman continuum break (17, 18) should be replaced by the emission-line searches might be, at best, similar to searches for Gunn–Peterson trough at z տ 5 (see refs. 19 and 20 for a careful the UV continuum, and of course a much smaller range of analysis of the effects of the Lyman ␣ forest on galaxy colors). redshift is searched because a very narrow band must be used. The mean surface brightness in the rest-frame UV contin- The Lyman ␣ emission line may be suppressed because of uum from these galaxies can also be expressed as S␯ ϭ 6 ϫ internal dust extinction. In addition, any photon emitted in the 10Ϫ33 erg cmϪ2⅐secϪ1⅐HzϪ1⅐arcsecϪ2 ϭ 32AB arcsecϪ2 (where blue side of the Lyman ␣ line will be scattered in the AB denotes AB magnitudes in the band where the UV surrounding IGM, reducing the strength of the line by a factor continuum is observed). If there is one galaxy per arcsec2 at of 2. For a galaxy observed when the surrounding IGM is still ϳ redshifts zf 10, each galaxy would have an AB magnitude of mostly neutral, even photons on the red side can be scattered 32, i.e., about 3 magnitudes fainter than the detection limit in because of the damping wing of the Gunn–Peterson trough the Hubble Deep Field for the I-band (21). However, as (34). This width of the damped absorption from the IGM is mentioned above, galaxies with a mass 40 times larger would broader than the emission lines from starburst galaxies, so any Downloaded by guest on September 28, 2021 Colloquium Paper: Rees Proc. Natl. Acad. Sci. USA 95 (1998) 51

Lyman ␣ emission line should be suppressed by the IGM in any nature of the simulation changes as soon as the first stars (or galaxy observed before the reionization. The edge of the other compact objects) form. The first stars exert crucial Gunn–Peterson trough should be ‘‘rounded off’’ into a char- feedback—the remaining gas is heated by ionizing radiation acteristic shape (34),except that the inhomogeneity of the IGM and perhaps also by an injection of kinetic energy via winds and may alter the profile (in particular, the presence of a halo of even supernova explosions. The three uncertainties here are: gas accreting on the galaxy may substantially increase the (i) What is the IMF of the first ? The column density contributing to the damped profile from gas at high-mass stars are the ones that provide efficient (and a redshift close to that of the galaxy). relatively prompt) feedback. It plainly makes a big difference Thus, even though the Lyman ␣ emission line might not be whether these are the dominant type of stars or whether the as good a spectral signature as the Gunn–Peterson trough as initial IMF rises steeply toward low masses, so that very many a technique to search for high redshift galaxies, its presence or faint stars form before there is a significant feedback. absence should be an important diagnostic of the ionization (ii) The influence of the early stars depends on whether their state of the IGM in the neighborhood of any detected galaxy. energy is deposited locally or penetrates into the all-pervasive A Note on Active Galactic Nuclei (AGNs) and Nonthermal medium that is not yet in contracting systems. The UV UV Background. AGN formation requires virialized systems radiation could, for instance, be mainly absorbed in the gas with larger masses and deeper potential wells (cf. ref. 29). immediately surrounding the first stars, so that it exerts no AGNs may ‘‘take over’’ as the dominant UV source at redshifts feedback on the condensation of further clumps—the total below 5 (and the second ionization of He may be delayed until number of massive stars needed to build up the UV back- AGNs can provide a power-law contribution to the spectrum), ground, and the concomitant contamination by heavy ele- but at higher z, when the H itself was reheated and ionized, OB ments, would then be greater. stars would dominate. (iii) Quite apart from the uncertainty in the IMF, it is also It is worth noting also that the diffuse IGM at redshifts of unclear what fraction of the baryons that falls into a clump would 3–5 has such a long thermal timescale that it retains a actually be incorporated into stars before being reejected. The ‘‘memory’’ of its thermal history at even higher redshifts. The ϭ retained fraction would almost certainly depend on the virial adiabat this gas is on at z 5 depends on the value of zi and velocity; gas more readily escapes from the shallow potential wells also on the slope of the UV spectrum. This adiabat can be relevant to dwarf galaxies. Ejection is even easier in potential constrained by observations of weak and narrow lines in the wells so shallow that they cannot confine gas at the photoioniza- Lyman forest (30, 31). tion temperature. This effect may lead to a reduced efficiency during the second, as compared with the third, of the three Very High Redshift Supernovae (and Gamma-Ray Bursts?) cosmogonic stages discussed in section 2. All of these three uncertainties would, for a given fluctua- The intergalactic gas was already highly photoionized by z ϭ tion spectrum, affect the redshift at which molecules were 5; also, the mean abundance of heavy elements had attained a destroyed and at which full ionization occurred. level about 0.01 solar by that time, this degree of contamina- tion being about what would be expected if the reheating and Heavy Elements and Magnetic Fields ionization were due to OB stars (25, 32). It is straightforward to calculate how many supernovae would have gone off, in each If the main UV source is stars, there is inevitably an associated comoving volume, as a direct consequence of this output of build-up of heavy elements. (In more radical pictures where black UV and heavy elements; there would be one, or maybe several, holes are involved in the early energy input, this inference doesn’t per year in each square arc minute of sky (33). These would be hold, because the energy supply could be gravitational rather than primarily of type 2. The typical observed has a flat nuclear.) The question then arises of how this processed gas would maximum lasting 80 days. One would therefore (taking the be distributed. Would it be confined in the virialized systems, or time dilation into account) expect each supernova to be near its maximum for nearly a year. It is possible that the explosions could it spread through the entire IGM? The ubiquity of carbon features in intermediate- and high- proceed differently when the stellar envelope is essentially տ 14 Ϫ2 metal-free, yielding different light curves, so any estimates of (N 3.10 cm ) column-density systems implies that heavy detectability are tentative. However, taking a standard type 2 elements are dispersed broadly enough to have a large cover- light curve (which may, of course, be pessimistic), one calcu- ing factor. These absorption systems may be associated with lates (33) that these objects should be approximately 27th the subgalactic sites of star formation that produce the first magnitude in J and K bands even out beyond z ϭ 5. The heavy elements. The nucleosynthesis sites therefore cannot be detection of such objects would be an easy task with the NGST. too sparse if these elements are, within the time available, to With existing facilities it is marginal. The best hope would be diffuse enough so that they are encountered somewhere along that observations of clusters of galaxies might serendipitously every line of sight through a typical high-column-density cloud. detect a magnified, gravitationally lensed image from far (Moreover, efficient mixing would then certainly imply bulk behind the cluster. motions so high that their neglect would be a serious inade- ␣ As a speculative addendum, note that a few percent of quacy in simulations of Lyman forest formation, etc.) The observed gamma-ray bursts may come from redshifts as large absorption line data tell us the mean abundance through the as 5; this would be expected if the burst rate, as a function of relevant cloud. They are compatible with 99% of the material cosmic epoch, tracks the star-formation rate. At the time of being entirely unprocessed, the heavy elements being re- writing, data on optical and x-ray afterglows are still very stricted to 1% of the material—the early heavy elements need sparse, but it is at least an exciting possibility that there may be not be thoroughly mixed, but they must have spread sufficiently occasional flashes, far brighter than supernovae, from very to have a large ‘‘covering factor’’ in the intermediate- and large redshifts. high-N clouds. The first stars are important for another reason: they may The First Stars: Some Uncertainties generate the first cosmic magnetic fields. Moreover, mass loss (via winds or supernovae permeated by magnetic flux) would The gravitational aspects of clustering can all be modeled disperse magnetic flux along with the heavy elements. This convincingly by computer simulations. Now the dynamics of flux, stretched and sheared by bulk motions, can be the ‘‘seed’’ the baryonic (gaseous) component—including shocks and for the later amplification processes that generate the larger- radiative cooling—can be modeled this way as well. But the scale fields pervading disc galaxies. Downloaded by guest on September 28, 2021 52 Colloquium Paper: Rees Proc. Natl. Acad. Sci. USA 95 (1998)

Where Are the Oldest Stars? photoionized gas comes from bremsstrahlung, which is less effective than the collisionally excited line emission from gas The efficiency of early mixing is important for the interpretation that is only partly ionized. The completion of photoionization of stars in our own galaxy that have ultra-low metallicity (33)— may therefore signal another pause in the cosmogonic process, lower than the mean metallicity that would have been generated associated with a further increase in the minimum scale that in association with the UV background at z Ͼ 5. If the heavy can collapse, and in the efficiency of cooling. elements were efficiently mixed, then these stars would them- By the epoch z ϭ 5, some structures (albeit perhaps only selves need to have formed before galaxies were assembled. To exceptional ones) must have attained galactic scales. But huge a first approximation they would cluster nondissipatively; they numbers of lower-mass systems should already exist at higher would therefore be distributed in halos (including the halo of our redshifts, and we can make quite firm estimates of their integrated own galaxy) like the dark matter itself. More careful estimates UV output. This paper has addressed how we can probe their slightly weaken this inference. This is because the subgalaxies indirect effects and perhaps even detect them directly. would tend, during the subsequent mergers, to sink via dynamical friction toward the centers of the merged systems. There would I am especially grateful to Jordi-Miralda Escude and would also like nevertheless be a tendency for the most extreme metal-poor stars to thank Zoltan Haiman, Avi Loeb, and Max Tegmark for discussion to have a more extended distribution in our galactic halo and to and collaboration on some of the topics described here. have a bigger spread of motions. The number of such stars depends on the IMF. If this were 1. Tegmark, M., Silk, J., Rees, M. J., Blanchard, A., Abel, T. & Palla, F. (1997) Astrophys. J. 474, 1. flatter, there would be fewer low-mass stars formed concur- 2. Efstathiou, G. (1992) Mon. Not. R. Astron. Soc. 256, 43. rently with those that produced the UV background. If, on the 3. Haiman, Z., Rees, M. J. & Loeb, A. (1997a) Astrophys. J. 476, 458. other hand, the IMF were initially steeper, there could, in 4. Stecher, T. P. & Williams, D. A. (1967) Astrophys. J. Lett. 149, L1. principle, be a lot of very low mass (macho) objects produced 5. Scott, D. & Rees, M. J. (1990) Mon. Not. R. Astron. Soc. 247, 510. at high redshift. These could provide a few percent of the halo 6. Madau, P., Meiksin, A. & Rees, M. J. (1997) Astrophys. J. 475, if omega were 1 (and a higher proportion of the Շ30-kpc inner 429. halo probed by lensing searches); a larger proportion could be 7. Swarup, G. (1996) in Cold Gas at High Redshifts, ed. Bremer, M. provided in a low-density universe. (Kluwer, Dordrecht, The Netherlands), 457. Most of the small, first-generation galaxies by now should 8. Songaila, A. & Cowie, L. L. (1996) Astrophys. J. 112, 335. 9. Rauch, M., Haehnelt, M. G. & Steinmetz, M. (1997) Astrophys. have merged into more massive systems (forming part of the J., in press. halo population of stars in normal galaxies like the Milky 10. Hernquist, L., Katz, N., Weinberg, D. H. & Miralda-Escude´,J. Way), but some could survive until today in galactic halos or (1996) Astrophys. J. 457, L51. even as isolated objects. This may be the explanation of 11. Miralda-Escude´,J., Cen, R., Ostriker, J. P. & Rauch, M. (1996) present-day dwarf spheroidal galaxies (34). Astrophys. J. 471, 582. 12. Madau, P. & Shull, J. M. (1996) Astrophys. J. 457, 551. Summary 13. Tytler, D., Fan, X.-M., Burles, S., Cottrell, L., Davis, C., Kirkman, D. & Zuo, L. (1995) in QSO Absorption Lines, ed. Meylan, G. Our general conclusions are relevant to any model in which the (Springer, Heidelberg), p. 289. 465, initial fluctuations have amplitudes decreasing with scale, so 14. Thoul, A. & Weinberg, D. H. (1996) Astrophys. J. 608. 15. Navarro, J. F. & Steinmetz, M. (1997) Astrophys. J. 478, 13. that cosmic structures form ‘‘bottom-up.’’ Such models differ, 16. Gunn, J. E. & Peterson, B. A. Astrophys. J. 142, 1633. of course, in the epoch at which ‘‘first light’’ would have 17. Guhathakurta, P., Tyson, J. A. & Majewski, S. R. (1990) Astro- occurred. In models with primordial baryon fluctuations phys. J. 357, L9. (PIB), this may be at z Ͼ 100; for CDM (primarily discussed 18. Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M. & here), it is in the range 10–20; and for ‘‘mixed dark matter’’ Adelberger, K. L. (1996) Astrophys. J. 462, L17. models, the first structures may form still more recently. 19. Madau, P. (1995) Astrophys. J. 441, 18. Molecular cooling tends to be more efficient at high densities 20. Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., and, therefore, at large redshifts; but in all cases it determines Steidel, C. C. & Fruchter, A. (1996) Mon. Not. R. Astron. Soc. 283, 1388. the scale of the first objects that condense out and contribute 21. Williams, R. E., Blacker, B., Dickinson, M., Van Dyke Dixon, W., the first injection of heat into the universe. Ferguson, H., et al. (1996) Astron. J. 112, 1335. The amount of background UV generated per solar-mass of 22. Cohen, J. G. (1995) Keck LRIS Quick Reference Guide (California material in these first objects is very uncertain—it depends on the Polytechnic State University, San Luis Obispo, CA). efficiency of star formation, on whether the IMF favors massive 23. Cohen, J. G. (1995) The Efficiency of the LRIS in the Spectroscopic stars (or even supermassive objects or black holes), and on how Mode (California Polytechnic State University, San Luis Obispo, much of the UV is ‘‘soaked up’’ by dense gas within the bound CA). objects themselves. But irrespective of all these uncertainties, the 24. Mather, J. & Stockman, H. (1996) NASA Report (National UV background exerts an important feedback on the cosmogonic Aeronautics and Space Administration, Washington, D.C.). 25. Haiman, Z. & Loeb, A. (1997) Science with the Next Generation process, by quenching H2 cooling, long before reaching the level Space Telescope (National Aeronautics and Space Administra- needed to photoionize the entire IGM. tion, Washington, D.C.), in press. The IGM remained predominantly neutral until a sufficient Ϫ3͞ 26. Miralda-Escude´,J. & Fort, B. (1993) Astrophys. J. 417, L5. number of objects above 109((1 ϩ z)͞10) 2MJ had gone 27. Kneib, J. P., Ellis, R. S., Smail, I. R., Couch, W. J. & Sharples, nonlinear. Such systems are massive enough to have virial R. (1996) Astrophys. J. 471, 643. temperatures above 10,000 K—hot enough for HI line emis- 28. Harrington, J. P. (1973) Mon. Not. R. Astron. Soc. 162, 43. sion to permit very efficient cooling. Most of the O-B stars (or 29. Haehnelt, M. & Rees, M. J. (1993) Mon. Not. R. Astron. Soc. 263, accreting black holes) that photoionized the IGM had to form 168. in systems at least as large as this. 30. Haehnelt, M. & Steinmetz, M. (1997) Mon. Not. R. Astron. Soc., in press. Formation of such systems would have continued unim- 31. Miralda-Escude´,J. & Rees, M. J. (1994) Mon. Not. R. Astron. Soc. peded until ionization was complete; the UV background rises 266, 343. Ϫ21 Ϫ2⅐ Ϫ1⅐ Ϫ1 sharply, to a value of order 10 ergs per cm Hz ster , 32. Ostriker, J. P. & Gnedin, N. (1996) Astrophys. J. 472, 630. when the universe becomes, in effect, an HII region. This must 33. Miralda-Escude´,J. & Rees, M. J. (1997) Astrophys. J. 478, L57. have happened before z ϭ 5. The only net cooling of a fully 34. Miralda-Escude´,J. & Rees, M. J. (1998) Astrophys. J., in press. Downloaded by guest on September 28, 2021