Nature © Macmillan Publishers Ltd 1997NATURE | VOL 390 | 27 NOVEMBER 1997 377

letters to nature

Formationofelliptical galaxiesatmoderate redshiftsStephen E. Zepf*†

* Department of Astronomy, University of California, Berkeley,California 94720-3411, USA† Department of Astronomy, Yale University, New Haven,Connecticut 06520-8101, USA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Different cosmological models make specific predictions aboutthe number of elliptical galaxies as a function of redshift1, soobservations can in principle be used to discriminate betweenthose models. Traditionally, elliptical galaxies have been thoughtto have formed in a single, rapid burst of star formation at highredshifts (z . 5), and then evolved quietly—with no significantfurther star formation—since that time2,3. Yet evidence suggeststhat at least some ellipticals formed for the merger of two spiralgalaxies4,5. It remains unclear which process dominates the for-mation of elliptical galaxies. Here I use the results of deep optical6

and near-infrared7–10 images to show that there are fewer galaxieswith very red colours than predicted by models in which typicalellipticals have completed their star formation by z < 5, whichmeans that elliptical galaxies must have significant star formationat z , 5. This requirement, combined with constraints on burstsof star formation in lower-redshift galaxies11,12, and the observedproperties of galaxies in the redshift range 0 , z , 1 (refs 13–17),suggests either that ellipticals form at moderate redshifts, where alarge initial burst of star formation is shrouded by dust, or thatthey form through the merging of smaller galaxies.

Very deep images at optical and near-infrared wavelengthsprovide a sensitive test of models of elliptical galaxy evolution.Passive models in which ellipticals form in a single, rapid burst athigh redshift predict the existence of a population of high-redshiftgalaxies with extremely red colours. In particular, galaxies that haveformed the bulk of the their stars at high redshifts will haveextremely red optical and near-infrared colours when observed at(1 ( z , zf ), where zf is the formation redshift. The red colours arethe result of the absence of hot, young massive stars combined withthe effect of redshift, which causes the observed near-infrared andoptical colours to become redder as they probe farther into the rest-frame blue and ultraviolet. This trend towards redder colours athigher redshifts changes only as the redshift of formation isapproached, when the hot and blue massive stars have not yetdied out.

The predictions of these models are shown in Fig. 1, which tracesthe colours of galaxies formed in bursts of star formation at variousredshifts for representative cosmologies. This diagram demon-strates that galaxies that have completed a large majority of theirstar formation by z < 5 reach extremely red colours ofV 606 2 K . 7. Galaxies with significant star formation at lowerredshifts do not have these colours unless reddened greatly bydust. Therefore the number of objects with V 606 2 K . 7 gives anupper limit to the co-moving number density of passively evolvingelliptical galaxies with zf ) 5.

Only recently have surveys become deep enough in both theoptical and near-infrared bands to test for the presence of theseextremely red galaxies at z . 1. This advance is due in large part tothe optical images of the Hubble Deep Field (HDF)6 that allowdetections or useful upper limits on objects with the very faint Vmagnitudes expected of red objects at these redshifts. These opticalimages have been followed up by several near-infrared surveys thatcan be used to determine the number of galaxies in the HDF withvery red V 606 2 K colours. One of these surveys provides K magni-tudes to very faint limits for ,20% of the HDF7. Near-infrared

images of the entire HDF to a somewhat brighter limit have alsobeen obtained and made publicly available8, and these have beenanalysed as part of this work. Two ground-based surveys in differentfields provide additional constraints on the number of extremelyred galaxies9,10. Although these ground-based surveys are not asdeep as the HDF surveys, particularly at optical wavelengths, theyare valuable because they are in completely independent fields,widely separated on the sky.

The results of these surveys are compared with the predictions ofpassively evolving models in Fig. 2. The surface density of extremelyred galaxies objects is far below the expectations of a model in whichall early-type galaxies form in bursts at high redshift. The predic-tions for the surface density of extremely red galaxies are calculatedby using a constant co-moving number density of early-typegalaxies set by studies of the K-band luminosity function at lowredshift18, the stellar populations models discussed earlier, and theappropriate cosmology for each plot. This calculation might

Figure 1 A plot of the observed V606 2 K colour as a function of redshift for a stellar

population formed in a single short burst (t ¼ 0:01Gyr): a, Ω ¼ 1:0, h ¼ 0:50; b,

Ω ¼ 0:2, h ¼ 0:65. Galaxieswith higher formation redshifts (zf ) 5) haveextremely

red colours that are not reached by galaxies with lower formation redshifts

(zf ( 3). The arrow marks the colour cut used in Fig. 2. Both the closed model

plotted in the top panel and the open model in the bottom panel exhibit the same

effect. An approximate translation for the HST V606 bandpass to the standard V

bandpass is given on the right side of the plots. These figures are based on the

Bruzual and Charlot models (unpublished data) with solar metalllicity and a

Miller–Scalo stellar initial mass function (IMF). The results are not sensitive to

the IMF within the usual range of parameters (forexample, Salpeter versusMiller–

Scalo), and a short burst duration is used only to isolate the effect of the formation

redshift. For simplicity, absorption by interveningH I has not been accounted for26,

which will only make the colours redder at z ) 3.

Nature © Macmillan Publishers Ltd 1997

underestimate the number density of extremely red galaxiesexpected in passively evolving models because the adopted localluminosity function has a density normalization about a factor oftwo lower than more global measures at modest redshifts19. Alter-natively, the calculation might be an overestimate in that theluminosity function used includes both elliptical and S0 galaxies,whereas passive models could be said to only apply to ellipticals,which are ,50% of the total.

The strong deficit of galaxies with extremely red colours, seen inwidely separated fields and at different flux limits, rules out modelsin which typical elliptical galaxies are fully assembled and haveformed all of their stars at z ) 5. This constraint cannot be escapedsimply by breaking ellipticals into many pieces that formed theirstars at high redshift. Near-infrared surveys are now deep enough todetect the individual pieces, and thus such models dramaticallyoverproduce red galaxies within the observed range of K magni-tudes. An illustrative example with zf ¼ 5, a constant total stellarmass in the early-type population and a co-moving number densityof this population that increases as ð1 þ zÞ3 is plotted in Fig. 2.Models with more modest density evolution follow curves betweenthis model and the zf ¼ 5 passive model with constant numberdensity. Therefore typical elliptical galaxies must have had signifi-cant star formation at z , 5.

Single-burst models at lower redshifts are strongly constrained byother observations. In particular, the modest ultraviolet luminositiesobserved for galaxies in the HDF11,12, as well as the failure of searchesfor strong-emission-line objects20 rule out these models unless theobjects are obscured greatly by dust21,22. Dusty models produce highluminosities in the far infrared and submillimetre range, and areclose to current observational limits at these wavelengths22.

Alternatively, the simple passive model at high redshift can be

altered to incorporate a much longer starburst duration thatextends to lower redshift. To make the colours sufficiently blue,this component must form ,5% or more of the stellar mass ofellipticals from zf to z ¼ 1. For example, if 5% of the stellar massforms in an additional star-formation component that is constantfrom zf to z ¼ 1 and then shuts off, the resulting V 606 2 K colour isslightly redder than 5 over most of this redshift range. If the mass inthe extended component is only a few per cent, then the V 606 2 Kcolour is greater than 6 and far more red galaxies are expected thanobserved. Thus, the extended star formation must represent at least5% of the stellar mass to accommodate the absence of very redgalaxies. This modification of the passive model runs into severalproblems with other observational constraints. First, it increases theevolution to brighter luminosities at higher redshifts expected inpassive models, contrary to the results of deep, red-selected galaxyredshift surveys, which find little or no luminosity evolution13–15.Moreover, an extended star formation history is unable to accountfor the elemental abundance ratios in ellipticals16.

A more natural answer is that most ellipticals form at z , 5through merging and associated starbursts, which are likely to be atleast moderately dusty. This result is consistent with a large body ofother evidence that indicates that merging has a major role in theformation of ellipticals, ranging from detailed studies of nearbygalaxy mergers to the discovery of bimodality in the globular clustersystems of elliptical galaxies17. The formation of elliptical galaxies inthis way is consistent with the predictions of hierarchical clusteringmodels of galaxy formation23–25. M

Received 9 July; accepted 7 October 1997.

1. Fukugita, M., Hogan, C. J. & Peebles, P. J. E. The history of galaxies. Nature 381, 489–495 (1996).2. Eggen, O. J., Lynden-Bell, D. & Sandage, A. R. Evidence from the motions of old stars that the galaxy

collapsed. Astrophys. J. 136, 748–766 (1962).

letters to nature

378 NATURE | VOL 390 | 27 NOVEMBER 1997

Figure 2 The surface density of extremely red galaxies (V606 2 K . 7) at different

limiting K magnitudes compared with the predictions of passive models of

elliptical galaxy formation: a, Ω ¼ 1:0, h ¼ 0:50; b, Ω ¼ 0:2, h ¼ 0:65. The data

points are inconsistent with zf ) 5 models, including the light dotted line repre-

senting a pure density evolution model with zf ¼ 5. The two circles represent

surveys in HDF. The deeper of these is the Hogg et al. survey with the Keck

telescope of ,1arcmin2 to a 50% completeness limit of K < 23:5, in which no

objects with colours as red as V606 2 K . 7 are found, and only one is even within

one magnitude of this colour7. The second HDF data point is based on the publicly

available Kitt Peak imaging survey8 that covers the full 5 arcmin2 of the HDF, to a

50% completeness limit of K < 22:25. Using an updated version of the SExtractor

image analysis program27, we find that all sources with K , 22 are detected in the

HDF optical images and have significantly bluer colours than the passive model

predictions. At the fainter limit of K , 22:5, there is one good candidate and two

marginal candidates for objects with V606 2 K . 7:0. The point plotted accounts for

all three of these candidates and is therefore shown as an upper limit. The square

and the triangle are from ground-based surveys in other fields by Moustakas

et al.9 and Cowie et al.10, respectively. The former covers ,2arcmin2 to a 50%

completeness limit of K < 22:5. Within these limits there are three galaxies that are

undetected in V, with approximate lower limits to their (V 2 K) colour of 7.5, 6.1 and

5.3 (ref. 9). These three represent an upper limit to the number of extremely red

objects in this field. The latter survey covers ,6 arcmin2 to a limiting K magnitude

of 20.9. Because the bluer bandpass is not very deep in these data, I adopt

ðI 2 KÞ . 4:5 as the best estimate of the equivalent V606 2 K . 7 cut used above.

There are approximately nine galaxies with colours this red. This number is more

properly considered an upper limit, as there are several objects with red (I 2 K) but

blue optical colours9,10.

Nature © Macmillan Publishers Ltd 1997

3. Partridge, R. B. & Peebles, P. J. E. Are young galaxies visible? Astrophys. J. 147, 868–886 (1967).4. Toomre, A. in The Evolution of Galaxies and Stellar Populations (eds Tinsley, B. M. & Larson, R. B.)

401–416 (Yale Univ. Observatory, New Haven, 1977).5. Schwezier, F. Colliding and merging galaxies III. The dynamically young merger remnant NGC 3921.

Astron. J. 111, 109–129 (1996).6. Williams, R. E. et al. The Hubble Deep Field: observations, data reduction, and galaxy photometry.

Astron. J. 112, 1335–1389 (1996).7. Hogg, D. W., Neugebauer, G., Armus, L., Matthews, K., Pahre, M. A., Soifer, B. T. & Weinberger, A. J.

Near infrared imaging of the Hubble Deep Field with the Keck Telescope. Astron. J. 113, 474–482(1997).

8. Dickinson, M. et al. (http://www.stsci.edu/ftp/science/hdf/hdf.html).9. Moustakas, L. A., Davis, M., Graham, J. R., Silk, J., Peterson, B. & Yoshii, Y. Colors and K-band counts

of extremely faint field galaxies. Astrophys. J. 475, 445–456 (1997).10. Cowie, L. L., Gardner, J. P., IIu, E. M., Songaila, A., Hodapp, K.-W. & Wainscoat, R. J. The Hawaii K-

band galaxy survey. I. Deep K-band imaging. Astrophys. J. 434, 114–127 (1994).11. Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., Steidel, C. C. & Fruchter, A. High-

redshift galaxies in the Hubble Deep Field: colour selection and star formation history to z,4. Mon.Not. R. Astron. Soc. 283, 1388–1404 (1996).

12. Steidel, C. C., Giavalisco, M., Dickinson, M. & Adelberger, K. I. Spectroscopy of Lyman break galaxiesin the Hubble Deep Field. Astron. J. 112, 352–358 (1996).

13. Cowie, L. L., Songaila, A., Hu, E. M. & Cohen, J. G. New insight on galaxy formation and evolutionfrom Keck spectroscopy of the Hawaii deep fields. Astron. J. 112, 839–864 (1996).

14. Kauffmann, G., Charlot, S. & White, S. D. M. Detection of strong evolution in the poulation of early-type galaxies. Mon. Not. R. Astron. Soc. 283, L117–L122 (1996).

15. Lilly, S. J., Tresse, L., Hammer, F., Crampton, D. & LeFevre, O. The Canada–France Redshift Survey.VI. Evolution of the galaxy luminosity function to z,1. Astrophys. J. 455, 108–124 (1995).

16. Worthey, G., Faber, S. M. & Gonzalez, J. J. Astrophys. J. 398, 69–73 (1992).17. Ashman, K. M. & Zepf, S. E. Globular Cluster Systems (Cambridge University Press, in the press).18. Gardner, J. P., Sharples, R. M., Frenk, C. S. & Carrasco, B. E. A wide field K-band survey. The

luminosity function of galaxies. Astrophys. J. 480, L99–L103 (1997).19. Ellis, R. S. Faint blue galaxies. Ann. Rev. Astron. Astrophys. (in the press).20. Djorgovski, S. G. in New Light on Galaxy Evolution (eds Bender, R. & Davies, R. L.) 277–286 (Kluwer,

Dordrecht, 1996).21. Meurer, G., Heckman, T. M., Lehnert, M. D., Leitherer, C. & Lowenthal, J. The panchromatic starburst

intensity limit at low and high redshift. Astron. J. 114, 54–68 (1997).22. Zepf, S. E. & Silk, J. On the effects of bursts of massive star formation during the evolution of ellipitcal

galaxies. Astrophys. J. 466, 114–121 (1996).23. Baugh, C. M., Cole, S. & Frenk, C. S. Evolution of the Hubble sequence in hierarchical models for

galaxy formation. Mon. Not. R. Astron. Soc. 283, 1361–1378 (1996).24. Cole, S., Aragon-Salamanca, A., Frenk, C. S., Navarro, J. F. & Zepf, S. E. A recipe for galaxy formation.

Mon. Not. R. Astron. Soc. 271, 781–806 (1994).25. Kauffmann, G., Guiderdoni, B. & White, S. D. M. Faint galaxy counts in a hierarchical universe. Mon.

Not. R. Astron. Soc. 267, 981–999 (1994).26. Madau, P. Radiative transfer in a clumpy universe: the colors of high-redshift galaxies. Astrophys. J.

441, 18–27 (1995).27. Bertin, E. & Arnouts, S. SExtractor: software for source extraction. Astron. Astrophys. Suppl. Ser. 117,

393–404 (1996).

Acknowledgements. I thank colleagues at Berkeley, Cambridge and Durham for stimulating conversa-tions; L. Moustakas for assisting with the use of image analysis software; R. Bouwens for providing code tomanipulate stellar populations models; and D. Hogg for providing data in tabular form. This work hasbeen supported by NASA grants.

Correspondence should be addressed to the author at Yale University (e-mail: zepf@astro.yale.edu).

letters to nature

NATURE | VOL 390 | 27 NOVEMBER 1997 379

Layeringof a liquidmetal incontactwithahardwallWillem Jan Huisman*, Joost F. Peters§,Michel J. Zwanenburg§, Steven A. de Vries*,Trevor E. Derry†, Douglas Abernathy‡& J. Friso van der Veen§

* FOM-Institute for Atomic and Molecular Physics, Kruislaan 407,1098 SJ Amsterdam, The Netherlands† Schonland Research Centre for Nuclear Sciences, University of Witwatersrand,Private Bag 3, P.O. Wits, 2050 Johannesburg, South Africa‡ European Synchrotron Radiation Facility, BP 220, 38043 Grenoble Cedex,France§ University of Amsterdam, Van der Waals-Zeeman Institute, Valckenierstraat 65,1018 XE Amsterdam, The Netherlands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

When a liquid makes contact with a solid wall, theoreticalstudies1–4 indicate that the atoms or molecules will becomelayered adjacent to the wall, giving rise to an oscillatory densityprofile. This expectation has not, however, been directly verified,although an oscillatory force curve is seen for liquids compressedbetween solid surfaces5. Here we present the results of an X-rayscattering study of liquid gallium metal in contact with a (111)diamond surface. We see pronounced layering in the liquid

density profile which decays exponentially with increasing dis-tance from the wall. The layer spacing is about 3.8 A, which isequal to the repeat distance of (001) planes of upright galliumdimers in solid a-gallium. Thus it appears that the liquid nearthe wall assumes a solid-like structure similar to the a-phase,which is nucleated on freezing at lower temperatures. This kind ofordering should significantly influence flow, capillary osmosis,lubrication and wetting properties5,6, and is likely to triggerheterogeneous nucleation of the solid.

Wall-induced oscillations in the atomic density are difficult todetect because they occur at a deeply buried interface over a depthinterval of typically less than a nanometre. An earlier study of anelectrolyte–silver (111) interface showed layering of the watermolecules close to the electrode7. This ordering phenomenon,however, relates to the presence of a strong electric field at theinterface, causing the dipolar water molecules to be preferentiallyoriented and attracting them to the silver surface. For the freesurfaces of liquid Hg and Ga, the in-plane pair distributionfunction8 and the density profile along the surface normal9,10 havebeen determined. In the latter case the strong gradient in the densitydependent one-body potential across the liquid–vapour interface,which acts as an effective force on the surface ion cores, isresponsible for a layering effect25. The layering reported here ispresent in a nonpolar elemental liquid lying against a hard wall, inthe absence of an external field. This reflects the commonlyoccurring situation of having a liquid metal contained in a crucible.

Our choice of materials is convenient for several reasons: Ga isliquid at 300 K (supercooled by 8 K) and diamond is essentially

diamond (111)

θ θ ki

liquid Ga

Qkf

10–10

10–8

10–6

10–4

10–2

100

0 1 2

Ref

lect

ivity

Q (Å–1)

Figure 1 Reflectivity of the Ga/diamond (111) interface as a function of

perpendicular momentum transfer Q'. Measured reflectivities obtained from

transverse momentum scans and v–2v scans are represented by filled and open

circles, respectively. The solid curve is the best-fit reflectivity curve, obtained by

Fourier transform of the electron density profile of Fig. 2; see also equation (1).

The dashed curve is a model calculation for a non-oscillatory density profile, see

text. The inset shows a schematic of the scattering geometry. The incoming

beam is represented by the wavevector ki, the outgoing beam by kf. The interface

is illuminated from the substrate side with the monochromatic X-ray beam. The

top surface of the liquid and the bottom surface of the substrate are shielded from

the incoming beam. After scattering, the photons travel through the crystal again

on their way to the detector. The momentum transfer Q, the difference between

the wavevectors of the incident and scattered photons, is normal to the interface.

Note that the indexof refractionof Ga is smaller than that of diamond. As the cross

section of carbon for Compton scattering is relatively large and a significant

volume of bulk liquid is illuminated, there is a high background of diffuse

scattering. For the highest momentum transfers, the signal-to-background ratio is

a few per cent. At a photon energy of 17 keV, the beam is attenuated by a factor of

four along the total pathlength travelled in the diamond. The data were corrected

for a 12% variation in the attenuation over the Q' range covered. These

experiments have only become feasible because third-generation synchrotron

sources deliver suchbrilliant beamsof energetic X-rays that sufficient count-rates

can be obtained from the small ordering effects in a few atomic layers.