THE ASTROPHYSICAL JOURNAL, 560 :4971, 2001 October 10 V( 2001. The American Astronomical Society. All rights reserved. Printed in U.S.A.

THE FARTHEST KNOWN SUPERNOVA: SUPPORT FOR AN ACCELERATING UNIVERSE AND AGLIMPSE OF THE EPOCH OF DECELERATION1

ADAM G. RIESS,2 PETER E. NUGENT,3 RONALD L. GILLILAND,2 BRIAN P. SCHMIDT,4 JOHN TONRY,5 MARK DICKINSON,2RODGER I. THOMPSON,6 TAMA S BUDAVA RI,7 STEFANO CASERTANO,2 AARON S. EVANS,8 ALEXEI V. FILIPPENKO,9

MARIO LIVIO,2 DAVID B. SANDERS,5 ALICE E. SHAPLEY,10 HYRON SPINRAD,9 CHARLES C. STEIDEL,10DANIEL STERN,11 JASON SURACE,12 AND SYLVAIN VEILLEUX13

Received 2001 March 12; accepted 2001 May 18

ABSTRACTWe present photometric observations of an apparent Type Ia supernova (SN Ia) at a redshift of D1.7,

the farthest SN observed to date. The supernova, SN 1997, was discovered in a repeat observation bythe Hubble Space Telescope (HST ) of the Hubble Deep Field-North (HDF-N) and serendipitously moni-tored with NICMOS on HST throughout the Thompson et al. Guaranteed-Time Observer (GTO) cam-paign. The SN type can be determined from the host galaxy type : an evolved, red elliptical lackingenough recent star formation to provide a signicant population of core-collapse supernovae. The classi-cation is further supported by diagnostics available from the observed colors and temporal behavior ofthe SN, both of which match a typical SN Ia. The photometric record of the SN includes a dozen uxmeasurements in the I, J, and H bands spanning 35 days in the observed frame. The redshift derivedfrom the SN photometry, z\ 1.7^ 0.1, is in excellent agreement with the redshift estimate ofz\ 1.65^ 0.15 derived from the photometry of the galaxy.U300B450V606 I814 J110J125H160H165 KsOptical and near-infrared spectra of the host provide a very tentative spectroscopic redshift of 1.755. Fitsto observations of the SN provide constraints for the redshift-distance relation of SNe Ia and a powerfultest of the current accelerating universe hypothesis. The apparent SN brightness is consistent with thatexpected in the decelerating phase of the preferred cosmological model, It is inconsis-)

MB 1/3, )"B 23.tent with gray dust or simple luminosity evolution, candidate astrophysical eects that could mimic pre-

vious evidence for an accelerating universe from SNe Ia at zB 0.5. We consider several sources ofpotential systematic error, including gravitational lensing, supernova misclassication, sample selectionbias, and luminosity calibration errors. Currently, none of these eects alone appears likely to challengeour conclusions. Additional SNe Ia at z[ 1 will be required to test more exotic alternatives to the accel-erating universe hypothesis and to probe the nature of dark energy.Subject headings : cosmology : observations supernovae : generalOn-line material : color gure

1 Based on observations with the NASA/ESA Hubble Space Telescope,obtained at the Space Telescope Science Institute, which is operated byAURA, Inc., under NASA contract NAS 5-26555.

2 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore,MD 21218 ; ariess=stsci.edu.

3 Lawrence Berkeley National Laboratory, Berkeley, CA 94720.4 Mount Stromlo and Siding Spring Observatories, Private Bag,

Weston Creek P.O. 2611, Australia.5 Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive,

Honolulu, HI 96822.6 Steward Observatory, University of Arizona, Tucson, AZ 85721.7 Department of Physics and Astronomy, The Johns Hopkins Uni-

versity, Baltimore, MD 21218, and Department of Physics, Uni-Eo tvo sversity, Budapest, Pf. 32, Hungary, H-1518.

8 Department of Physics and Astronomy, State University of New York(SUNY) at Stony Brook, NY 11794-3800.

9 Department of Astronomy, University of California, Berkeley, CA94720-3411.

10 Palomar Observatory, California Institute of Technology, Mail Code105-24, Pasadena, CA 91125.

11 Jet Propulsion Laboratory, California Institute of Technology, MailCode 169-327, Pasadena, CA 91109.

12 SIRTF Science Center, California Institute of Technology, MailCode 314-6, Pasadena, CA 91125.

13 Department of Astronomy, University of Maryland, College Park,MD 20742.

1. INTRODUCTION

The unexpected faintness of Type Ia supernovae (SNe Ia)at zB 0.5 provides the most direct evidence that the expan-sion of the universe is accelerating, propelled by darkenergy (Riess et al. 1998 ; Perlmutter et al. 1999). Thisconclusion is supported by measurements of the character-istic angular scale of uctuations in the cosmic microwavebackground (CMB) that reveal a total energy density well inexcess of the fraction attributed to gravitating mass (de Ber-nardis et al. 2000 ; Balbi et al. 2000 ; Jae et al. 2001).

However, contaminating astrophysical eects can imitatethe evidence for an accelerating universe. A pervasive screenof gray dust could dim SNe Ia with little telltale reddeningapparent from their observed colors (Aguirre 1999a, 1999b ;Rana 1979, 1980). Although the rst exploration of a distantSN Ia at near-infrared (NIR) wavelengths provided no evi-dence of nearly gray dust, more data are needed to performa denitive test (Riess et al. 2000).

A more familiar challenge to the measurement of theglobal acceleration or deceleration rate is luminosity evolu-tion (Sandage & Hardy 1973). The lack of a complete theo-

49

50 RIESS ET AL. Vol. 560

retical understanding of SNe Ia and an inability to identifytheir specic progenitor systems undermines our ability topredict with condence the direction or degree of lumi-nosity evolution Wheeler, & Thielemann 1998 ;(Ho ich,Umeda et al. 1999a, 1999b ; Livio 2000 ; Drell, Loredo, &Wasserman 2000 ; Pinto & Eastman 2000 ; Yungelson &Livio 2000). The weight of empirical evidence appears todisfavor evolution as an alternative to dark energy as thecause of the apparent faintness of SNe Ia at zB 0.5 (seeRiess 2000 for a review). However, the case against evolu-tion remains short of compelling.

The extraordinary claim of the existence of dark energyrequires a high level of evidence for its acceptance. Fortu-nately, a direct and denitive test is available. It should bepossible to discriminate between cosmological models and impostors by tracing the redshift-distance relation toredshifts greater than one.

1.1. T he Next Redshift Octave and the Epochof Deceleration

If the cosmological acceleration inferred from SNe Ia isreal, it commenced rather recently, at 0.5\ z\ 1. Beyondthese redshifts, the universe was more compact and theattraction of matter dominated the repulsion of darkenergy. At z[ 1, the expansion of the universe should havebeen decelerating (see Filippenko & Riess 2000). Theobservable result at z 1 would be an apparent increasedbrightness of SNe Ia relative to what is expected for a non-decelerating universe. However, if the apparent faintness ofSNe Ia at zB 0.5 is caused by dust or simple evolution, SNeIa at z[ 1 should appear fainter than expected from decel-erating cosmological models. More complex param-eterizations of evolution or extinction that can match boththe accelerating and decelerating epochs of expansionwould require a higher order of ne-tuning and are there-fore less plausible.

Measuring global deceleration at z[ 1 provides addi-tional cosmological benets. To constrain the equation ofstate of dark energy (and distinguish a cosmological con-stant from the decaying scalar elds described by the quintessence hypothesis ; Peebles & Ratra 1988 ; Cald-well, & Steinhardt 1998), it is necessary to breakDave ,degeneracies that exist between the global densities of massand dark energy. Observations of SNe Ia in this next red-shift octave are well suited to deciphering the nature of darkenergy and have motivated recent proposals to develop awide-eld optical and NIR space mission (Curtis et al.200014 ; Nugent 2000). To better determine the merits andtechnical requirements of such a mission, it will be impor-tant to study closely the rst few SNe detected at theseredshifts. In addition, the study of SNe Ia at z[ 1 canprovide meaningful constraints on progenitor models (Livio2000 ; Nomoto et al. 2000 ; Ruiz-Lapuente & Canal 1998)after surveys of such SNe over a range of high redshifts arecompleted.

Both the Supernova Cosmology Project (SCP; Perlmut-ter et al. 1995) and the High-z Supernova Search Team(HZT; Schmidt et al. 1998) have pursued the discovery ofSNe Ia in this next redshift interval. In the fall of 1998, theSCP reported the discovery of SN 1998ef at z\ 1.2(Aldering et al. 1998). The following year, the HZT dis-

14 See D. Curtis et al. 2000, Supernova/Acceleration Probe (SNAP), pro-posal to DOE and NSF at http ://snap.lbl.gov/.

covered an SN Ia at z\ 1.2 (SN 1999fv) as well as at leastone more at zB 1.05 (Tonry et al. 1999 ; Coil et al. 2000).These data sets, while currently lacking the statistical powerto discriminate between cosmological and astrophysicaleects, are growing and may provide the means to breakdegeneracies in the future.

In early 1998, Gilliland & Phillips (1998) reported thedetection of two SNe, SN 1997 and SN 1997fg, in a reob-servation of the Hubble Deep Field-North (HDF-N) withWFPC2 through the F814W lter. The elliptical host of SN1997 indicated that this supernova was most probably aSN Ia. . .[at] the greatest distance reported previously forSNe, but the observations at a single epoch and in a singleband were insufficient to provide useful constraints on theSN and, hence, to perform cosmological tests (Gilliland,Nugent, & Phillips 1999, hereafter GNP99).

Here we report additional, serendipitous observations ofSN 1997 obtained in the Guaranteed-Time Observer(GTO) NICMOS campaign (Thompson et al. 1999) and inthe General Observer (GO) program 7817 (M. Dickinson etal. 2001, in preparation), as well as spectroscopy of the host.The combined data set provides the ability to put strongconstraints on the redshift and distance of this supernovaand shows it to be the highest redshift SN Ia observed (todate). These measurements further provide an opportunityto perform a new and powerful test of the accelerating uni-verse by probing its preceding epoch of deceleration.

In 2 of this paper, we describe the observations of theSN and its host in the HDF-N and report photometry ofthe SN from the NICMOS campaign. In 3, we analyze theobservations to constrain the SN parameters : redshift,luminosity, age of discovery, and distance. The constraintsare used to extend the distance-redshift relation of SNe Ia toz[ 1 and to discriminate between cosmological models andcontaminating astrophysical eects. Section 4 contains adiscussion of the systematic uncertainties in our measure-ments and their implications. We summarize our ndingsin 5.

2. OBSERVATIONS

2.1. T he Discovery of SN 1997Between 1997 December 23 and December 26, Gilliland

& Phillips (GO 6473) reobserved the HDF-N with theHubble Space Telescope (HST ) to detect high-redshift SNe.These observations were obtained with WFPC2 (F814W)during 18 HST orbits in the continuous viewing zone(CVZ) and at a spacecraft orientation as closely matched tothe original HDF-N as possible. To critically sample theWF point-spread function (PSF) and robustly reject all hotpixels, CCD defects, and noise uctuations, the exposureswere well dithered using 18 dierent subpixel and multipixelosets. Additional F300W frames were obtained during thebright portion of the CVZ orbits to support improved rejec-tion of transient hot pixels. The total F814W exposure timein the second HDF-N epoch (6300 s) was 51% of thatobtained in the original epoch 2.0 yr prior.

After careful processing, the second epoch was registeredwith the rst and dierence frames in both temporal direc-tions were produced. Robust SN detection thresholds weredetermined by a Monte Carlo exercise of adding PSFs ofvarying brightness onto host galaxies of varying redshifts.From this exercise, it was determined that a brightnessthreshold of (Johnson-Cousins) would ensure them

I\ 27.7

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 51

rejection of all spurious transients. Simulations of complete-ness indicated that 95% of SNe at coincident withm

I\ 27

host galaxies at 1.5\ z\ 1.9 would be discovered. Onlytransients that were brighter than the rejection threshold,visible in each of three subsets and near a host galaxy, wereidentied as SNe. Candidates coincident with hosts centerswere discarded as possible active galactic nuclei. Theharvest from the HDF-N SN search was two robust SNdetections : SN 1997fg and SN 1997 at a signal-to-noiseratio (S/N) of 20 and 9, respectively. The former was hostedby a late-type galaxy with a spectroscopic redshift of 0.95.

SN 1997 was discovered at mag, R.A.\mI\ 27.0

(equinox J2000),12h36m44s.11, decl.\ ]6212@44A.8 0A.16southwest of the center of the host galaxy, 4-403.0 (Williamset al. 1996). The host has been classied as an ellipticalgalaxy based on measurements of its surface brightnessprole, concentration, asymmetry, and colors as well as byvisual inspection (Williams et al. 1996 ; Fernandez-Soto,Lanzetta, & Yahil 1999 ; Dickinson 1999 ; Thompson et al.1999 ; et al. 2000). A section of the HDF-N nearBudava rithe SN host is shown in Figure 1 as observed with WFPC2and NICMOS. GNP99 favored the classication of SN1997 as Type Ia because of the red, elliptical host.

Photometric redshift determinations of the host had beenpublished using only photometry of theU300 B450 V606 I814HDF-N (z\ 0.95 ; Sawicki, Lin, & Yee 1997), as well asfrom the later addition of data from theJ125 H165Ksground (z\ 1.32 ; Fernandez-Soto et al. 1999). To span therest-frame optical breaks in the spectral energy distribution(SED) of galaxies with z[ 1 and reliably estimate theirphotometric redshift, it is necessary to employ both opticaland NIR data. GNP99 assumed the Fernandez-Soto et al.(1999) redshift which employed the best available coverageof the host SED to date. However, even with the monochro-matic detection of a probable SN Ia and an estimate of itsredshift, the extraction of useful cosmological informationfrom SN 1997 was not feasible and was not attempted.

2.2. Serendipity : T he NICMOS CampaignsTwo NIR assaults on the HDF-N with NICMOS on

HST provided a wealth of data and understanding on the

natural history of galaxies (see Ferguson, Dickinson, & Wil-liams 2000 for a review). The GTO program of Thompsonet al. (1999 ; GTO 7235) consisted of D100 orbits of F110Wand F160W exposures of a single 55@@] 55@@ Camera-3 eld,reaching a limiting AB magnitude (Oke & Gunn 1983) of 29in the latter. The observations were gathered during 14 con-secutive days and the eld was contained within the WF4portion of HDF-N, serendipitously imaging the host of SN1997. (It is interesting to note that the placement of theGTO eld within the HDF-N had less than a 20% chanceof containing SN 1997.) Although the program did notbegin until 1998 January 19, about 25 days after the dis-covery of the SN, a series of single-dither exposures (GO7807) was taken between the discovery of the SN and thestart of the GTO program for the purpose of verifying thesuitability of the chosen guide stars. Each of these exposureswas for a duration of 960 s. A single F110W and F240Mexposure on 1998 January 6 included the host, as did aF160W exposure from 1998 January 2 and another on 1997December 26. T he December 26 NICMOS exposure wascoincident within hours of the W FPC2 discovery exposures.(It is of further interest to note the low likelihood of thechance temporal coincidence of the HDF-N SN Search andthe GTO program, each initially scheduled in dierent HSTcycles.)

A second program was undertaken 6 months after theGTO program, between 1998 June 14 and June 22, by M.Dickinson et al. (2001, in preparation ; GO 7817). Thisprogram observed the entire HDF-N in F110W andF160W to a limiting AB magnitude of D26.5 by mosaicingCamera 3 of NICMOS to study a wider eld of galaxies.This program also contained the host galaxy and thegreatly faded light of the supernova. The space-based NIRphotometry of the SN host oered greater precision andcoverage of the SED than the ground-based data alone andallowed an improved estimate of the photometric redshift.Using the space-based pho-U300 B450 V606 I814J110 H160tometry and the ground-based photometryJ125H165 Kscontained in Table 1, et al. (2000) determined theBudava riredshift of the host to be z\ 1.65^ 0.15 from ts to eithergalaxy SED eigenspectra or these same eigenspectra mildly

FIG. 1.Color-composite images of the region of the HDF-N near the host of SN 1997. The WFPC2 images were taken during the HDF-N campaign(Williams et al. 1996) and the NICMOS images were taken during the GTO campaign (Thompson et al. 1999). The arrow indicates the SN host galaxy.

0 10 20 30 40time (days)

102

103

104

105

SN 1

997f

f exp

tim

e (s

econ

ds)

WFPC2 F814W (I)NICMOS F110W (J) NICMOS F160W (H)

HDF SN Search

Guide Star Tests

GTO Campaign

52 RIESS ET AL. Vol. 560

TABLE 1

AB MAGNITUDES OF THE HOST GALAXYOF SN 1997FF

Bandpass log(j/km) AB Magnitude

FUV160b . . . . . . [0.80 \30.0NUV250b . . . . . . [0.60 \29.2U300 . . . . . . . . . . [0.53 27.84~0.70`2.48B450 . . . . . . . . . . . [0.34 26.67^ 0.16V606 . . . . . . . . . . . [0.22 25.64^ 0.04I814 . . . . . . . . . . . [0.10 24.42^ 0.02J110 . . . . . . . . . . . 0.04 22.60^ 0.02J125a . . . . . . . . . . 0.10 21.96^ 0.03H160 . . . . . . . . . . 0.20 21.59^ 0.01H165a . . . . . . . . . 0.22 21.55^ 0.03K

sa . . . . . . . . . . . . 0.33 21.03^ 0.02

a Ground-based observation ; Fernandez-Sotoet al. (1999).

b Ninety-ve percent limits from HST STIS ; H.C. Ferguson (2001, private communication).

corrected to improve the agreement between spectroscopicand photometric redshifts. The redshift constraints derivedfrom the host photometry are analyzed in 3.1.

2.3. Supernovae PhotometryThe two HDF-N NICMOS campaigns taken together

oer a rare opportunity to measure the behavior of a super-nova at a redshift not accessible from the ground andperhaps to discriminate between the inuence of darkenergy and contaminating astrophysical eects at z[ 1.For favored cosmological models an SN()

MB 13, )"B 23),Ia at z\ 1.65 is expected to peak in F160W at D24 mag

and in F110W at D24.5 mag. The 130 ks of exposure timein each bandpass of the GTO program would be expectedto reach S/NB 100 for an SN Ia at peak, though actualmeasurements impacted by the shot noise of the bright hostmay be more uncertain. Because the SN would not be atpeak for some or all of the observations, we expect furtherreductions in the measurement precision. If the apparentbrightnesses of high-redshift SNe are dominated by evolu-tion and/or dust, and not by cosmology, the S/N of the SNmight be further reduced. For the single dithers used to testguide stars, we expect an S/N no better than D10 andpossibly worse because of the above mitigating factors.

Another valuable and fortuitous feature of the NICMOScampaigns is that they likely sample the rest-frame light ofthe supernova in the B and V bands, the most studied andbest understood wavelength region of nearby SNe. Indeed,the great difficulty in observing these wavelengths from theground has generally limited the detection and monitoringof SNe Ia to In the D2 rest-frame months expected toz[ 1.have elapsed between the two separate NICMOS pro-grams, an SN Ia is expected to fade D3 mag, resulting in asignicant surplus of ux in the dierence image of the twoepochs.

Our goal is to measure the photometry of the SNthroughout the 35 days of the GTO program. However, ourtask is complicated by the proximity of the SN to its brighthost. GNP99 found that the SN was located at the half-lightradius of the galaxy in F814W. As we will nd, the hostcontains 2 to 6 times as much ux at the position of the SNas the peak of the SN PSF, depending on the band and thedate of the exposure. The strategy of digitally subtracting an

image of the host obtained when the SN has faded (templateimage) from one taken when the SN is relatively bright hasbeen successfully employed by the SCP (Perlmutter et al.1995) and the High-z Team (Schmidt et al. 1998), as well asby GNP99 using the original HDF-N images as a templateimage. This is the method we employed.

The task of geometrically mapping (i.e., registering) theimages from the Thompson et al. campaign to align withthe template image from the Dickinson et al. campaign wascomplicated by the timing and eld location of the former.As seen in Figure 2, the exposure time for the SN in theThompson et al. campaign was dispersed irregularly over a35 day time interval, requiring careful consideration of theoptimal way to measure the temporal behavior of the SNwhile still yielding robust photometry (see below). In addi-tion, the location of the SN was always extremely close tothe corner of the Camera-3 eld of 256] 256 pixels, missingthe chip during one-third of the dithers, and landing 1 to 15pixels from the corner in the rest. Although the Camera-3eld is remarkably distortion-free in its interior, a milddegree of pincushion distortion exists in the extremecorners. Even distortions of a few tenths of a pixel are intol-erable for the accurate subtraction of the host ux from theSN (Cox et al. 1997). Although application of the NIC-3geometric distortion map removes much of the distortion inthe eld corners, we applied an empirically derived linearmapping between the SN image and template to furtherreduce host contamination.

Our rst step was to use the SExtractor algorithm (Bertin& Arnouts 1996) to detect sources and measure their cen-troids in the template and SN images. Custom software wasused to match identical sources in the two lists (Schmidt etal. 1998). Next, the geomap routine in IRAF was used toderive a ux-conserving mapping of the SN images to thetemplate coordinate system. In practice we found that manyof the individual 900 s dithers in the GTO campaign did notprovide the desired S/N in the centroid measurements to

FIG. 2.HST exposures obtained for SN 1997 as a function of time indierent bandpasses. Time in days (the abscissa) is given relative to 1997December 23, the start of the HDF-N SN search with WFPC2 (GO 6473).The subsequent GTO campaign with NICMOS (Thompson et al. 1999 ;GTO 7235) and its preceding tests for guide star suitability (7807) providedvaluable coverage of the light curve of the SN found during the search. Asubsequent NICMOS campaign (M. Dickinson et al. 2001, in preparation ;GO 7817) provided templates to remove the contaminating light of thehost after the SN had faded.

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 53

derive a robust mapping to the template coordinate system.A tenable alternative is to rst drizzle together a subset ofthe dithers (Fruchter & Hook 1997 ; Thompson et al. 1999 ;GNP99) before deriving the nonlinear coordinate trans-form. This practice has the advantage of increasing the pre-cision of the centroid measurements of the sources, allowingfor the critical sampling of the PSF, reducing the eects ofNICMOS interpixel sensitivity (Storrs et al. 1999), and pro-viding the ability to further remove cosmic rays (Thompsonet al. 1999). However, the obvious disadvantage of combin-ing the dithers before further processing (i.e., binning) is areduction in our ability to resolve the temporal behavior ofthe SN. After much experimentation, we chose an interme-diate strategy of combining the dithers from the main GTOcampaign into three temporal bins of observed-frame widthD2 days for each of the F110W and F160W data sets. Asimilar strategy was used by GNP99 to provide time-re-solved magnitudes of the SN in F814W. The necessaryexception to the practice of binning was for the measure-ment of the SN ux in the individual 960 s dithers used totest guide stars before the GTO campaign.

We employed the Alard (2000) algorithm to match thePSF, mean intensity, and background in the template andSN images. Using the nearest visible sources to the SNsposition, we rst derived and then applied a convolutionkernel with a linear variation across the sources. (Again,experimentation showed a constant convolution kernel wasinadequate for matching the two image PSFs, and second-order convolutions were unstable because of a lack ofenough sources with sufficient S/N to measure the kernelvariation.) Next, we subtracted the template image from theSN images. As seen in Figure 3, the resulting image containsthe SN without the contaminating light of the host. Animportant test of these image-processing routines is toverify a lack of signicant ux residuals in the vicinity ofother galaxies (which did not host SNe) in the eld. Conr-mation of this test can be seen in Figure 3 (see also GNP99,Fig. 1 for the comparable F814W discovery images of SN1997).

Next, we measured the ux of the SN. Again, the formatof the NICMOS campaigns (optimized for galaxy studies,not for the monitoring of SN 1997) presented challengesrarely encountered by past high-redshift supernova pro-grams. Typically, the leading source of noise in the measure-ments of the ux in high-redshift SNe is the shot noise in thesky (or the host galaxy in HST observations). For theNICMOS observations of SN 1997, the dominant sourceof uncertainty is the host galaxy residuals in the dierenceimage. The location of the SN near the core of a muchbrighter host results in the appearance of signicant galaxyresiduals from typically tolerable errors of 0.1 pixels in theimage pair registration. Eradicating registration errors atthe position of the SN is made more difficult by the locationof the host galaxy near the corner of the Camera-3 eld inthe Thompson et al. (1999) campaign. For some images, wereduced this error by processing only a subset of the imagearound the SN. Galaxy residuals in the dierence image canalso result from the undersampling of Camera 3 and fromvariations of the intrapixel sensitivity. To derive a robustmeasure of the SN ux, we used a relatively large apertureto contain the net ux near the SN impacted by the galaxyresiduals. In practice, apertures with a radius of 5 to 10oversampled-by-two pixels to were used. The ux(0A.5 1A.0)of the SN was measured relative to that contained in an

equal-sized aperture of a bright comparison star at thecenter of the GTO eld. Flux uncertainties were determinedby a Monte Carlo exercise of adding and measuring arti-cial SNe in the eld with the same brightness and back-ground as SN 1997 (Schmidt et al. 1998). For the singleF110W dither from 1998 January 6, the SN was notdetected, but a ux upper limit was established by theMonte Carlo exercise.

We transformed the relative SN ux onto the F110W ABand F160W AB magnitude systems by applying the zeropoints of the transformation equation from M. Dickinson etal. (2001, in preparation). A count rate for the comparisonstar of 3.22 and 1.85 ADU per second was measured inF110W and F160W, respectively. Using the zero points of22.89 for AB F110W and 22.85 for AB F160W gave a mea-sured magnitude of 21.62 and 22.18 for the star in these twobandpasses, respectively. (An expected uncertainty of D5%in these zero points is insignicant in comparison with theuncertainties in the relative photometry.) Addition of themeasured magnitude dierences between the star and theSN yielded the measurements for the SN on the AB system.To transform the AB system magnitudes onto the Vegasystem, we calculated the bandpass-weighted magnitudes ofspectrophotometry of Vega (relative to a at spectrum) andderived the zero-point osets of [1.34 and [0.75 mag forF160W and F110W, respectively. The Vega system F160Wand F110W magnitudes are given in Table 2, as are theF814W Vega system magnitudes from GNP99. It is impor-tant to note that the magnitudes of the SN in F110W andF160W as listed in Table 2 are underestimates of the uxcaused by the presence of SN ux in the template imagesfrom Dickinson et al. The templates were obtained 177 daysafter the discovery epoch of 1997 December 23. To accu-rately correct the SN magnitudes in Table 2 for the over-subtraction, it is necessary to t the light curve to determinethis correction. This step is performed in 3.2.

2.4. Host SpectroscopyOn the nights of 2000 November 21, 26, and 27 UT,

about 2 hr of optical spectroscopy were obtained of the SNhost using the echelle spectrograph and imager (ESI) onKeck II under poor conditions (seeing and an airD1A.3mass of 1.72). During the nights of 2001 February 23 and24 UT, 3.5 hr of optical spectroscopy were gathered usingthe low-resolution imaging spectrograph (LRIS ; Oke et al.1995) on Keck I in good conditions (seeing and an airD0A.8mass of 1.4). An additional 2.5 hr of optical spectroscopywere obtained with ESI on Keck II on the nights of 2001February 26 and 27 UT. The total data set of optical spec-troscopy consisted of 8 hr with a wavelength coverage of4000 to 10000 A small amount of continuum ux wasA .detected in the composite spectrum with evidence of someminor breaks in the galaxy SED, but there was no evidenceof either a strong break in the galaxy SED or any emissionlines.

Observations were made with the NIR spectrographNIRSPEC (McLean et al. 1998) on Keck II on the nights of2001 March 16 and 17 UT and on 2001 April 14 UT. On alldates, the slit was oriented at a position angle (P.A.) of0A.76

to include a galaxy D3A to the north.184.6In the March campaign, the slit was rotated to a P.A.

such that light from both the host galaxy and the nearbybright galaxy fell onto the slit. The host galaxy was thenpositioned on the slit by rst obtaining short-integration

12 a

rc s

eco

nd

s

F814W

z=2.80z=0.56

SN 1997ffz~1.7

z~2.0

z=0.88z=1.23z~1.8

F110W

z=2.80z=0.56

SN 1997ffz~1.7

z~2.0

z=0.88z=1.23z~1.8

F160W

z=2.80z=0.56

SN 1997ffz~1.7

z~2.0

z=0.88z=1.23z~1.8

Template Subtraction

F814W

F110W

F160W

N

E

N

E

N

E

54 RIESS ET AL. Vol. 560

FIG. 3. SN 1997 in F814W (I), F110W (J), and F160W (H). Images on the left show the region of the HDF-N near the SN host without the SN(template images). Images on the right show the dierence in intensity between an SN image and the template image. Superimposed on this image areintensity contours. Spectroscopic redshifts (Cohen et al. 2000) are listed as exact while photometric redshifts et al. 2000) are listed as approximate.(Budava ri

images of the eld, then moving the telescope to positionthe nearby galaxy onto the slit. Spectroscopic observationswere obtained by taking four 600 s integrations. Afterthe rst integration, the slit was moved 15A in the spatialdirection on the array for the second integration, thenmoved back to the original position, where the process wasrepeated.

In the April campaign, the target was acquired by placinga nearby bright star onto the slit position desired and off-

setting the telescope D85A to place the target at the sameposition on the slit. The conditions were excellent, withseeing estimated at FWHM for the duration of the0A.45observation. Eight exposures of 900 s each were obtained ;after each exposure, the osets were reversed and the align-ment of the star on the slit was checked using imagesobtained with the IR slit-viewing camera. Judging by thesechecks, the telescope osetting and guiding was accurate tobetter than 1 pixel For each exposure, the target(D0A.14).

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 55

TABLE 2

VEGA MAGNITUDES OF SN 1997FFa

F814W F110Wa F160Wa KBb meff(B)

Days since 1997 Dec 23 (mag) (mag) (mag) (mag) (mag)

[770. . . . . . . . . . . . . . . . . . . . . [27.3c . . . . . . . . . . . .[229. . . . . . . . . . . . . . . . . . . . . [26.0d . . . . . . . . . . . .0.83 . . . . . . . . . . . . . . . . . . . . . . . 26.94 (0.15)e . . . . . . 0.96 25.98 (0.15)2.78 . . . . . . . . . . . . . . . . . . . . . . . 27.09 (0.15)e . . . . . . 1.01 26.08 (0.15)3.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.49 (0.20) [2.57 26.00 (0.20)3.74 . . . . . . . . . . . . . . . . . . . . . . . 26.92 (0.13)e . . . . . . 1.03 25.89 (0.13)10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.56 (0.15) [2.69 26.17 (0.15)14. . . . . . . . . . . . . . . . . . . . . . . . . . . . [23.3 . . . [1.64 [24.8628.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.84 (0.20) [3.14 26.83 (0.20)30.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.26 (0.25) [3.20 27.30 (0.25)32.0 . . . . . . . . . . . . . . . . . . . . . . . . . . 25.67 (0.30) . . . [1.65 27.16 (0.30)32.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.27 (0.20) [3.26 27.36 (0.20)34.5 . . . . . . . . . . . . . . . . . . . . . . . . . . 25.67 (0.20) . . . [1.66 27.14 (0.20)37.0 . . . . . . . . . . . . . . . . . . . . . . . . . . 25.92 (0.20) . . . [1.66 27.38 (0.20)

a Underestimates caused by ux in template observed at ]177 days.b K-corrections to the B band for best t : z\ 1.7, SN discovered one week past maximum bright-

ness.c Estimated from initial HDF-N.d Estimated from HST archive images (GO 7588).e From GNP99.

was placed on a dierent position along the 42A slit toreduce detector systematics and to allow for more accuratesky subtraction.

All the data were reduced following a procedure verysimilar to that described in Pettini et al. (2001). The datafrom 2001 March and 2001 April were reduced indepen-dently and then combined with appropriate weighting intothe nal co-added two-dimensional spectrum. The one-dimensional spectrum was extracted using an aperture of1A.2.

The feature which may be very tentatively identied asthe [O II] 3727 line at z\ 1.755 falls in a relatively rareAregion that is unaected by bright OH lines in the night sky,although there is a strong sky emission feature at 1.029 km,i.e., just to the red of the putative [O II] feature. The excessemission is present in individual subsets of the data andbecomes more prominent when the 40 minutes of integra-tion time from 2001 March are combined with the 120minutes from 2001 April. The line is well resolved at a spec-tral resolution of D7.5 consistent with expectations forA ,the [O II] doublet, which has a rest-frame separation of D2

(Single emission lines in the spectrum of the nearbyA .z\ 0.556 galaxy, and most but not all sky-subtractionresiduals, are signicantly narrower.) This feature is note-worthy and may function as a useful hypothesis to test withfuture observations, but at this time its validity is highlyuncertain.

3. ANALYSIS3.1. T he Redshift of the Supernova and Its Host

As discussed in 2, tting the U300 B450 V606 I814 J110space-based photometry and theH160 J125 H165Ksground-based photometry of 4-403.0 to galaxy SEDs yieldsa photometric redshift of z\ 1.55 to 1.70 with variationsdepending on whether the tted model is based on templategalaxy SEDs (Coleman, Wu, & Weedman 1980, hereafterCWW) or galaxy eigenspectra et al. 2000). These(Budava rits can be seen in Figure 4. The well-constrained tsbetween model and data indicate a far greater degree of

precision in the redshift (1 pB 0.02) than is empiricallyfound by comparing photometric and spectroscopic red-shifts. We therefore consider the empirical dispersion from

et al. (2000) as the measure of the individualBudava riuncertainty.

Using a set of orthogonal eigenspectra derived from theCWW galaxy template SEDs yields the lowest redshift,z\ 1.55^ 0.15, and empirically the least precise and robust

et al. 2000). Indeed, signicant outliers(Budava rioccasionally result from the application of this method. Amore robust and precise redshift estimate comes from thetting of improved eigenspectra. These are derived from theCWW SED eigenspectra, which are rst repaired (see

et al. 2000) as required to improve the agreementBudava ribetween the photometric and spectroscopic redshifts. Mildrepairing yields z\ 1.65^ 0.13 (model KL2) and furtherrepairing (model KL5) provides z\ 1.70^ 0.10 (Budava riet al. 2000). The biggest advantage of rst repairing theeigenspectra is the suppression of outliers, yielding a morerobust estimate. In addition, the relatively simple SEDs ofearly-type galaxies such as 4-403 generally provide morerobust photometric redshifts et al. 2000). For(Budava ri4-403.0, we will adopt z\ 1.65^ 0.15 as a measurementthat is representative of the photometric redshift.

An additional and independent pathway to determine theredshift is from the SN colors. For the following analysis,we will provisionally adopt the classication from GNP99of SN 1997 as a Type Ia supernova based on its red, ellip-tical host galaxy. However, in 4.1 we will analyze thedegree to which this classication is merited.

As can be seen in Table 2, coincident or near-coincidentmeasurements of SN 1997 in dierent bands provide anobserved I[H color of 3.5^ 0.2 mag and a J[H color of1.6^ 0.2 mag, 30/(1] z) days later in the rest frame. InFigure 5, we plot these measurements as a function ofexpected colors of SNe Ia over their temporal evolution atdierent redshifts (note that the size of the point scales withthe temporal proximity to B maximum). SNe Ia are bluestshortly after explosion and become redder with age. They

30

28

26

24

22

20

AB

(mag

)

U300 B450 V606 I814 J110 J125 H160 H165 KS

SED zphotKL2 1.65KL5 1.70CWW 1.55

Photometric z for host of SN 1997ff

Best Fit to UBVIJJHHK

28

26

24

22

AB

(mag

)

SED zphotKL2 1.65KL2 1.30KL2 1.90

Fixed zphot

-0.6 -0.4 -0.2 -0.0 0.2 0.4log (/m)

30

28

26

24

22

20

AB

(mag

)

SED zphotKL2 0.75KL5 0.76CWW 0.71

Best Fit to UBVI

56 RIESS ET AL. Vol. 560

FIG. 4.Photometric redshift estimate for the host of SN 1997. This estimate employs six photometric magnitudes from HST observations and threefrom ground-based observations et al. 2000). Galaxy magnitudes are given in Table 1. The top panel shows the best ts for the CWW (galaxy(Budava riSEDs), KL2 (eigenspectra), and KL5 (eigenspectra) models. The middle plot shows the sensitivity of the t to the redshift. The bottom plot shows the biasesthat result from using only the WFPC2 data.

reach their reddest color D25 days after maximum in thesebandpasses and return to a modestly bluer color during thesubsequent nebular phase. As seen in Figure 5, either of theobserved colors of SN 1997 is redder than an SN Ia at anyphase for z\ 1. The I[H color sets a limit of z[ 1.2 whilethe J[H color is more stringent with z[ 1.4, both at thegreater than 95% condence level.

However, the constraint obtained from each observedcolor treated independently is less restrictive than if we con-

sider their separation in time. In this case, our lower limiton the redshift comes from assuming that the later colormeasurement occurs at the reddest phase of the SN andfrom requiring the earlier color measurement to be consis-tent with an earlier, bluer phase of the SN. In this way wend z[ 1.45 at the 95% condence level. An upper limit onthe redshift comes from the colors and the observation thatthe SN is declining during the ve F160W measurements byD1 mag in the D30 observed-frame days. As seen in Figure

01

2

3

4

56

I 814-H

160 (

mag

) SN 1997ff at Discovery

maximum+25 days

1.0 1.2 1.4 1.6 1.8 2.0z

0.0

0.5

1.0

1.5

2.0

J 110-H

160 (

mag

)

30/(1+z) days later

7000 7500 8000 8500

0.0

0.5

1.0

1.5f (

Jy)

HDF 4-403.0

LBDS 53W091

B2640? B2900?

10000 10500 11000Observed Wavelength (A)

-1

0

1

2

3

f (J

y)

[OII]3727?, z=1.755

HDF 4-403.0

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 57

FIG. 5.Comparison between the observed and expected NIR colors ofan SN Ia as a function of assumed redshift. For a given redshift, the colorevolution is plotted beginning 15 days before maximum (bluest point) andagain in 5 day intervals. SNe Ia redden with time, reach their reddest colorat 2530 days after maximum, and then become bluer by a few tenths of amagnitude. The relative size of the point scales with temporal proximity toB maximum. The observed colors of SN 1997 at discovery and 30/(1] z)days later puts strong constraints on the redshift and age of the SN.

5, the redshift of the SN must be less than 2.0 to both matchthe I[H color and be discovered on the decline. A redshiftin the range of 1.5\ z\ 1.8 is coarsely consistent with theobserved colors and temporal behavior.

This simple analysis assumes negligible reddening fromthe host. While this assumption is appropriate for mostelliptical hosts, we will consider the eect of reddeningexplicitly in 4.2. Note that Galactic reddening is very lowtoward the HDF-N.

The above exercise cannot be performed readily if weassume the SED of a common SN II instead of an SN Ia.Not surprisingly, a comparison of the observed colors of SN1997 to those expected for a blue SN II (similar to thewell-observed SN 1979C; e.g., Schmidt et al. 1994) yieldspoor ts to both color measurements and their time separa-tion. For any value of the redshift, the observed J[H coloris far bluer than the color of a common SN II-P or SN II-L(for denitions, see Barbon, Ciatti, & Rosino 1979) at aphase dictated by the requirement of matching the earlierI[H color. A common SN II would not match both colormeasurements of SN 1997 unless it were observed shortlyafter explosion and at zB 2. However, the observed declineof SN 1997 appears inconsistent with the expected rising

luminosity or plateau phase of an SN II shortly after explo-sion. Reddening of the SN would result in an even greaterdierence between the data and a common SN II. Based onthe observed colors and declining luminosity alone, theidentication of SN 1997 as a normal SN II is stronglydisfavored, a conclusion which is discussed further in 4.2.

The blue color, bright magnitude, and observed decline ofSN 1997 might be consistent with some SNe IIn(Filippenko 1997), which show considerable heterogeneityin their light curves (Schlegel 1990 ; A. V. Filippenko 2001,private communication). However, these objects are rare,and they are not found in old stellar populations ( 4.2).Similarly, SN 1997 was unlikely to be an SN Ib or SN Ic,which occur in very young stellar populations, are generallyredder than SN 1997, and are rare (see 4.2).

The spectroscopy of the host presented in 2.4 providessome evidence that is consistent with the preceding redshiftdeterminations and potentially more precise but currentlyunreliable. As can be seen in Figure 6 (top panel), the opticalspectrum of the host suggests two minor breaks that couldbe identied with the rest-frame breaks at 2640 and 2900 Abecause of blends of metals (Spinrad et al. 1997). A simple s2

FIG. 6.Optical and NIR spectroscopy of the host, HDF 4-403.0, fromthe Keck telescope. The top panel shows the optical spectra of the SN hostcompared to the spectrum of an old, red elliptical galaxy, LBDS 53w091(z\ 1.55 ; Spinrad et al. 1997), transformed to z\ 1.755. A simple s2 mini-mization provides a possible match at a redshift for the SN host of z\ 1.67to 1.79, but this match is not robust. The bottom panel shows the NIRspectroscopy of the host. An apparent weak emission line, if identied as[O II] j3727, would yield z\ 1.755, but this redshift determination istentative. A gradient in the detected continuum is apparent, with anincrease to the red.

58 RIESS ET AL. Vol. 560

minimization between the optical spectrum of the host andthe same region of the SED of the z\ 1.55 elliptical LBDS53w091 (Spinrad et al. 1997) yields a signicant minimumboundedby z\ 1.67and1.79 (3p condence level).However,this minimum does not appear robust and we cannot ruleout the possibility that other redshift matches are possiblegiven other models for the host SED.

The shape of the extracted continuum (Fig. 6b) is consis-tent with the broadband photometry from HST WFPC-2and NICMOS observations and is not inconsistent with thepresence of a break at 4000 in the galaxy rest frame (forAzB 1.7), although for a possible redshift of zB 1.8, theregion longward of the 4000 break is beyond the wave-Alength range covered by the NIRSPEC spectrum.

Although the spectroscopic redshift indicators are sug-gestive of a match with the photometric indicators, thequality of the spectra is too low and the identicationof spectral features too uncertain to reach a robust deter-mination of the redshift from the spectroscopy alone.Therefore in the following section we derive constraintsfrom the SN without employing the spectroscopic redshiftindications.

3.2. Probability Density Functions for SN 1997The simple method for constraining the redshift

described in the previous section can be rened to make useof all of the SN photometric data simultaneously. Byvarying the parameters needed to empirically t an SN Ia,such as the light-curve shape, distance, redshift, and age, wecan use the quality of the t to determine the probabilitydensity function (PDF) of these parameters. An additionalcomponent of this tting process is to include the knowncorrelation between SN Ia light-curve shapes and their peakluminosities (Phillips 1993 ; Phillips et al. 1999 ; Riess, Press,& Kirshner 1996 ; Perlmutter et al. 1997). Examples of thistting process can be seen in Figure 7 as applied to SN1997.

In Appendix A, we develop a simple formalism for usingthe observations of SN 1997 and any prior informationthat is appropriate to determine the PDF of the parametersof luminosity, distance, redshift, and age commonly used toempirically model SNe Ia. This method is quite general andits application to SN Ia photometry is equivalent to the useof a common light-curve tting method, such as *m15(B)

FIG. 7.Comparison between the B-band light curve of a normal SN Ia and the observed data transformed to rest-frame B for dierent assumed redshiftsand discovery ages. The observed SN colors, or, for the transformation to a xed bandpass shown here, the K-corrections are a strong function of redshiftand SN age. The distance modulus may be constrained by osetting the model light curve in magnitudes. A good t between model and data occurs only in anarrow range of redshifts and ages as shown in the middle panel.

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 59

(Phillips et al. 1999 ; Hamuy et al. 1996), the MulticolorLightcurve Shape method (MLCS; Riess et al. 1996, 1998),or the stretch method (Perlmutter et al. 1997), in cases forwhich the light-curve information is more constraining thanprior information. The advantage of this method is itsability to incorporate prior information (e.g., a photometricredshift) in a statistically sound way that properly assignsweights to the relative constraints provided by data andpriors.

Given the presence of SN light in the M. Dickinson et al.(2001, in preparation) template images taken 177 days afterthe SN discovery, it is necessary to restore the ux that isnecessarily oversubtracted from the F110W and F160Wmagnitudes listed in Table 2. The size of the correctiondepends on the redshift, age of discovery, and shape of theSN light curve. Therefore, we implemented this correctionduring the process of tting the data to parameterizedmodels. In the case of the best-tting light curve, the under-estimate of the SN magnitudes from the GTO campaign is0.1 to 0.2 mag (depending on the phase), and for the guidestar test exposures (taken when the SN is D1 mag brighter)the correction is 0.05 mag.

We determined the PDF for SN 1997 using the methodsoutlined in Appendices A and B, the previously describeddata, and specic priors we discuss here.

Riess et al. (1998) found that the observed peak B-bandluminosities of SNe Ia at low redshift (0.01\ z\ 0.1) andhigh redshift (0.3\ z\ 1.0) are characterized by distribu-tion functions with mag. An even narrower lumi-p

M 0.25

nosity function of mag for SNe Ia was found bypM

\ 0.17Perlmutter et al. (1999) for a similar set of low-redshift SNeIa and an independent set of high-redshift SNe Ia. Althoughthe peak luminosity function of very nearby SNe Ia(z\ 0.01) includes a low-luminosity tail populated byso-called SN 1991bg-like SNe Ia (Filippenko et al. 1992b ;Leibundgut et al. 1993 ; Modjaz et al. 2001), such SNe whichare dimmer at peak by D2 mag are undetected in high-redshift, magnitude-limited surveys (Li, Filippenko, & Riess2001). We dened a normal function prior for the observedpeak luminosity of SNe Ia with a standard deviation of 0.25mag. Assuming that SN 1997 was drawn from the samepopulation of SNe Ia that lower redshift SNe Ia havesampled, we expect this prior to be valid. If, however, theluminosities of SNe Ia have signicantly evolved by zB 1.7,then this should be apparent in the divergence of theredshift-magnitude relation of SNe Ia from cosmologicalmodels. We also considered a much less constraining priorof mag. Though this prior underutilizes ourp

M\ 0.50

empirical knowledge of SN Ia luminosities, it does providea wider latitude to allow the photometry of SN 1997 toconstrain the t to its model light-curve shape.(Quantitatively, this prior yields similar results as a per-fectly at luminosity prior.) It also provides for a possiblylarger dispersion in peak luminosities of SNe Ia at higherredshifts.

As provided in 3.1, the photometric redshift of the hostgalaxy from space-based U300B450 V606 I814 J110H160photometry and the ground-based photo-J125 H165Ksmetry results in the prior constraint z\ 1.65^ 0.15

et al. 2000). (This constraint is also consistent(Budava riwith the host spectroscopy as presented in 3.1.) We deter-mined the PDF of SN 1997 both with and without thisprior photometric constraint. The latter approach, whilenot optimal for determining the best constraints, does allow

us to determine the SN redshift solely from the SN data andtest its compatibility with the photometric and spectro-scopic redshift determinations.

Finally, because no additional information is available toconstrain the remaining two SN Ia parameters, distanceand age at discovery (see Appendix A), no further know-ledge of these variables was included in the priors.

By marginalizing the four-dimensional PDF for SN1997 over any three parameters, we determined the PDFfor the fourth parameter of interest. The marginal probabil-ity for the redshift, age at discovery, and luminosity areshown in Figure 8. The marginalized PDF of the redshift isnot a simple function, though it is strongly peaked nearzB 1.7 and is insignicant outside the range 1.4\ z\ 1.95.A much lower local maximum is seen at z\ 1.55. The red-shift measurement of SN 1997 can be crudely approx-imated by z\ 1.7~0.15`0.10.The consistency of the three redshift indicators, deter-mined independently from the galaxy colors, the supernovacolors, and the host spectroscopy, provides a powerful andsuccessful cross-check of our redshift determination.Excluding any galaxy redshift information has little impacton the marginalized redshift PDF of the SN because the SNdata are signicantly more constraining for the redshiftdetermination. The cause of the dierence in measuredphotometric redshift precision lies in the dierence in therelative homogeneity of galaxy and SN Ia colors. For thegalaxy photometric determination, the precision of thismethod is limited by the variations of galaxy SEDs beyondthose which can even be accounted for from the super-position of eigenspectra. In contrast, SNe Ia colors are farmore homogeneous and their mild inhomogeneities are wellcharacterized, leading to more precise constraints on thephotometric redshift.

From the redshift determination, we conclude that SN1997 is the highest redshift SN observed to date (assuspected by GNP99), easily surpassing the two SNe Ia atz\ 1.2 from the HZT (Tonry et al. 1999 ; Coil et al. 2000)and the SCP (Aldering et al. 1998). The statistical con-dence in this statement is very high.

Marginalizing the PDF for SN 1997 over the age-of-discovery parameter yields the function shown in Figure 8.We conclude that the SN was discovered by Gilliland &Phillips (1998) at an age of a week past B-band maximumwith an uncertainty of D5 days ; however, this estimate ishighly non-Gaussian, as can be seen in Figure 8. An addi-tional, local maximum in the marginalized probability isevident at an age of D15 days after maximum. The possi-bility that the SN was discovered at this later age corre-sponds to the same model for which the weaker maximumin the redshift PDF indicated that z\ 1.55. This correlationbetween the redshift and age parameters is a natural conse-quence of the reddening of an SN Ia as it ages and is shownin Figure 9.

Little additional constraint on the peak-luminosity/light-curve-shape parameter is gained from the t, beyond whatis provided by the luminosity function prior as seen inFigure 8. The B-band light curve of a typical SN Ia (e.g.,Leibundgut 1988 ; mag from Hamuy et al.*m15(B)\ 1.11996 ; *\ 0 mag from Riess et al. 1998) provides an excel-lent t to the SN 1997 data when the other three parame-ters are set to their most likely values, as in the middle panelof Figure 7. By relaxing the prior to mag, we canp

M\ 0.50

better determine the degree to which the SN t constrains

1.0 1.2 1.4 1.6 1.8 2.0 2.2z

0.0

0.2

0.4

0.6

0.8

1.0

rela

tive

prob

abili

ty

spectroscopy

CombinedGalaxySN 1997ff

-20 -10 0 10 20 30 40age at discovery (days)

0.0

0.2

0.4

0.6

0.8

1.0

rela

tive

prob

abili

ty

-20.5 -20.0 -19.5 -19.0 -18.5 -18.0luminosity at B max (mag)

0.0

0.2

0.4

0.6

0.8

1.0

rela

tive

prob

abili

ty

Fit0.25Fit0.50m=0.25 magm=0.50 mag

60 RIESS ET AL. Vol. 560

FIG. 8.Marginalized probability density functions for SN Ia model parameters used to t SN 1997. The top panel shows the PDF for the redshiftconstraints from the galaxy photometry, from the SN, and from both sources. The middle panel shows the constraint on the age of the discovery relative to Bmaximum. The bottom panel shows two forms of the observed luminosity function for SNe Ia and the constraints on the luminosity of SN 1997 using thepriors and the light-curve ts.

its possible light-curve shape and, hence, its correlated peakluminosity. As shown in Figure 8, the t to the SN 1997data disfavors a fast declining, subluminous SN Ia light-curve shape. Quantitatively, this result excludes the hypoth-esis that SN 1997 is as much as 0.5 mag subluminous atpeak.

On the bright limb of the peak-luminosity/light-curve-shape relationship, the constraints are nearly parallel to theluminosity function prior. The bright limb is dened byspectroscopically peculiar SNe Ia known as SN 1991T-likeevents (Filippenko et al. 1992a ; Phillips et al. 1992 ; Li et al.

1999), which may be overluminous by 0.30.6 mag. It isimportant to note that SNe Ia are neither observed norexpected to reach peak magnitudes of more than D0.6 magbrighter than the typical (Hamuy et al. 1996 ; &Ho ichKhokhlov 1996). Therefore, we do not consider light-curveshape models that are extrapolated beyond the most lumi-nous SNe Ia observed locally.

Of critical importance to cosmological hypothesis testingis the determination of the luminosity distance to such high-redshift SNe Ia. Because of the signicant correlationbetween the distance and photometric redshift parameters

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 61

FIG. 9. Condence intervals for the discovery age (days relative toB-band maximum) and redshift of SN 1997. Because the observed SNcolors are a strong function of both of these parameters, a high degree ofcorrelation exists in their simultaneous determination.

and the need to use both of these parameters for cosmo-logical applications, we determined the two-dimensionalPDF for distance and redshift simultaneously. This functionis shown in Figure 10 for the dierent priors describedabove. Although it is preferable to include prior informa-tion from the galaxy photometric redshift estimate and theobserved luminosity function of SNe Ia, the constraintsshown in the distance-redshift plane are only minimallyimproved with this information, and our subsequent con-clusions are insensitive to these priors.

We also determined the likelihood function for the dis-tance assuming the tentative spectroscopic redshift. We ndm[M\ 45.15^ 0.34 mag. This likelihood function isquite Gaussian within the 2 p boundaries but attensbeyond for shorter distances (corresponding to an olderdiscovery age) and steepens beyond for longer distances

FIG. 10.Condence intervals for the distance modulus (m[M) andredshift of SN 1997. The intervals were calculated using the galaxy photo-metric redshift and the observed luminosity function of SNe Ia (optimal),neglecting the galaxy photometric redshift and using a weak prior on SNIa luminosities described in the text.

(corresponding to a discovery near maximum). If we assumethe tentative spectroscopic redshift, we nd the age of dis-covery to be 6^ 2 days after B-band maximum and tighterconstraints on the peak-luminosity/light-curve-shapeparameter. For the latter, we nd the SN to be 0.05^ 0.20mag fainter at peak than average, which makes it a highlytypical supernova.

3.3. Cosmological ConstraintsIn Figure 11 we show the redshift and distance data (i.e.,

the Hubble diagram) for SNe Ia as presented by the Super-nova Cosmology Project (Perlmutter et al. 1999) and theHigh-z Supernova Search Team (Riess et al. 1998). Thesedata have been binned in redshift to depict the statisticalleverage of the SN Ia sample. Overplotted are the cosmo-

FIG. 11.Hubble diagram of SNe Ia minus an empty (i.e., empty )\ 0) universe compared to cosmological and astrophysical models. The points arethe redshift-binned data from the HZT (Riess et al. 1998) and the SCP (Perlmutter et al. 1999). Condence intervals (68%, 95%, and 99%) for SN 1997 areindicated. The modeling of the astrophysical contaminants to cosmological inference, intergalactic gray dust or simple evolution, is discussed in 4.2. Themeasurements of SN 1997 are inconsistent with astrophysical eects that could mimic previous evidence for an accelerating universe from SNe Ia at zB 0.5.

62 RIESS ET AL. Vol. 560

logical models (favored),)M

\ 0.35, )" \ 0.65 )M \(open), (Einsteinde0.35, )" \ 0.0 )M \ 1.00, )" \ 0.0Sitter), and an astrophysical model representing a progres-sive dimming in proportion to redshift caused by gray dustor simple evolution within an open cosmology. This modelis further described in 4.2. All data and models are plottedas their dierence from an empty universe ()

M\ 0.0, )" \0.0).

All models are equivalent in the limit of z\ 0. Dierencesin the models are considerable and detectable at z[ 0.1.Evidence for a signicant dark-energy density and currentacceleration is provided by the excessive faintness of thebinned data, with 0.3\ z\ 0.8 compared with that of theopen model, yielding a net dierence of D0.25 mag.

The lack of SNe Ia at an independent redshift interval,beyond z\ 1, provides only the slimmest of margins forinferring the need for dark energy. An alternative explana-tion for the faintness of SNe Ia at zB 0.5 is a contaminatingastrophysical eect. Two often-cited candidates for theseeects are SN evolution and gray intergalactic dust.Although direct tests for these eects have thus far yieldedlittle evidence to support either (Riess 2000 ; Riess et al.2000), the standard of proof for accepting vacuum energy(or quintessence) is high. A more powerful test for any astro-physical eect that continues to dim SNe at ever greaterredshifts is to observe SNe Ia at z[ 1 (Filippenko & Riess2000). At these redshifts, the universe was more compactand familiar gravity would have dominated cosmologicalrepulsion. The resulting deceleration at these redshiftswould be apparent as a brightening of SNe Ia relative to acoasting cosmology or to the aforementioned astrophysicaleects. The redshift of SN 1997 is high enough to probethis earlier epoch and, together with the distance measure-ment, provides the means to discriminate between thesehypotheses.

In Figure 11 we show the constraints derived from SN1997. In the redshift-distance plane, the principal axes ofthe error matrix from the photometric analysis are not quiteperpendicular and the condence contours are complex.Because there is only one object available in this highest

redshift interval, we prefer to interpret Figure 11 with broadbrush strokes.

SN 1997 is brighter by D1.1 mag (and therefore closer)than expected for the persistence of a purported source ofastrophysical dimming at zB 0.5 and beyond. The sta-tistical condence of this statement is high ([99.99%). Thisconclusion supports the reality of the measured accelerationof the universe from SNe Ia at zB 0.5 by excluding themost likely, simple alternatives. To avoid this conclusionrequires the addition of an added layer of astrophysicalcomplexity (e.g., intergalactic dust that dissipates in theinterval 0.5\ z\ 1.7 or luminosity evolution that is sup-pressed or changes sign in this redshift interval). Otherastrophysical eects, such as a change in the SN Ia lumi-nosity distance caused by a change in metallicity with red-shift, are also disfavored (Shanks et al. 2001). Systematicchallenges to these conclusions are addressed in 4.

Other cosmological models that predict a relativedimming of SNe Ia at z[ 1, such as the quasi-steady statehypothesis, appear to be in disagreement with this obser-vation (Banerjee et al. 2000 ; Behnke et al. 2001). Similarly,models with relatively high vacuum energy and relativelylow mass density are excluded (e.g., If we)"B 1, )MB 0).assume an approximately at cosmology, as required byobservations of the CMB, and a cosmological constant-likenature for dark energy, the observations of SN 1997 dis-favor or, alternately,)"[ 0.85 )M \ 0.15.SN 1997 also provides an indication that the universewas decelerating at the time of the supernovas explosion.To better understand this likelihood, in Figure 12 we showthe redshift-distance relation of SNe Ia compared to that ofa family of at, cosmologies. For such cosmologies, the)"transition redshift between the accelerating and deceler-ating epochs occurs at a redshift of (M.[2)"/)M]1@3[ 1Turner, 2001, private communication). For increasingvalues of the transition point (i.e., the coasting point))",occurs at increasing redshifts. The highest value of that)"is marginally consistent with SN 1997 is (at the)" \ 0.85D3 p condence level), for which the transition redshiftoccurs at z\ 1.25, which is signicantly below the redshift

FIG. 12.Same as Fig. 11 with the inclusion of a family of plausible, at cosmologies. The transition redshift (i.e., the coasting point) between the)"accelerating and decelerating phases is indicated and is given as SN 1997 is seen to lie within the epoch of deceleration. This conclusion is[2)"/)M]1@3 [ 1.drawn from the result that the apparent brightness of SN 1997 is inconsistent with values of and, hence, a transition redshift greater than that of)" 0.9SN 1997. [See the electronic edition of the Journal for a color version of this gure.]

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 63

of SN 1997. For the universe to have commenced acceler-ating before the explosion of SN 1997 requires a value of

a result that is highly in conict with the SN)"[ 0.9,brightness. We conclude that, within the framework of thesesimple but plausible cosmological models, SN 1997exploded when the universe was still decelerating. Indeed,the increase in the measured luminosity distance of SNe Iabetween zB 0.5 and zB 1.7, a factor of 4.0, is signicantlysmaller than in most eternally coasting cosmologies (e.g.,

and appears to favor the empirical reality)M

\ 0, )" \ 0)of a net deceleration over this range in redshift. However, arigorous and quantitative test of past deceleration requiresa more complete consideration of the possible nature ofdark energy and is beyond the scope of this paper.

The above conclusions are unchanged if we adopt thetentative spectroscopic redshift of the SN host in place ofthe photometric redshift indicators [in Figs. 11 and 12, thecontours would be replaced by a point at z\ 1.755,*(m[M)\ [0.74^ 0.34]. In this case, the redshift uncer-tainty is greatly diminished and the distance uncertainty ismildly reduced. However, the dominant source of statisticaluncertainty in the testing of this cosmological hypo-thesis remains the distance uncertainty, not the redshiftuncertainty.

A great deal of theoretical eort has been expendedrecently to understand the nature of dark energy. Some ofthe possibilities include Einsteins cosmological constant, adecaying scalar eld ( quintessence ; Peebles & Ratra1988 ; Caldwell et al. 1998), and so on. A large sample ofSNe Ia distributed over the redshift interval of 0.5\ z\ 2.0could empirically break degeneracies between these models(by distinguishing among dierent average equations ofstate, w\P/oc2, where P is the pressure and o is thedensity) if theory alone is insufficient to explain dark energy.SN 1997 is in the right redshift range to discriminatebetween dierent dark-energy models, and if one assumesthat high-redshift SNe are tracing the cosmological modeland not an astrophysical eect, then SN 1997 may beuseful for this task. However, the large uncertainty presentin the measurement of only one SN Ia provides very littleleverage to discriminate between dark-energy models at thistime.

4. DISCUSSION

The results of 3 indicate that SN 1997 is the mostdistant SN Ia observed to date with a redshift of z\

Moreover, an estimate of its luminosity distance is1.7~0.15`0.1 .consistent with an earlier epoch of deceleration and isinconsistent with astrophysical challenges (e.g., simple evol-ution or gray dust) to the inference of a currently acceler-ating universe from SNe Ia at zB 0.5. In this section, weexplore systematic uncertainties in these conclusions.

4.1. Supernovae ClassicationSNe are generally classied by the presence or absence of

characteristic features in their spectra. For example, SNe Iaare distinguished by the absence of hydrogen lines and thepresence of Si II j6150 absorption (see Filippenko 1997 for areview). Unfortunately, our inability to observe the deningregions of an SN SED at high redshift necessitates the use ofadditional indicators of SN type.

However, an alternate way to discriminate some SNe Iais from the morphology of their host galaxies and theirassociated star formation histories. While all types of SNe

have been observed in late-type galaxies, SNe Ia are theonly type to have been observed in early-type galaxies.Although this lore is well known by experienced observersof SNe, this correlation is empirically apparent from anupdate of the Asiago Supernova Catalog (Cappellaro et al.1997 ; Asiago Web site15) and can be seen in Figure 13. Ofthe more than 1000 SNe for which type and host galaxymorphologies are all well-dened and have been classiedin the modern scheme, there have been no core-collapseSNe observed in early-type galaxies (that is, only SNe Iahave been found in such galaxies). All D40 SNe in ellipticalhosts, and classied since the identication of the SN Iasubtype, have been SNe Ia. The same homogeneity of typeis true for the D40 SNe classied in S0 hosts. Core-collapseSNe (types II, Ib, and Ic) rst appear along the Hubblesequence in Sa galaxies, and even within these hosts, theyform a minority and their relative frequency to SNe Ia issuppressed by a factor of D6 compared to their presence inlate-type spirals (Cappellaro et al. 1997, 1999). Evolvedsystems lose their ability to produce core-collapse SNe.

The explanation for this well-known observation isdeeply rooted in the nature of supernova progenitors andtheir ages. Unlike all other types of SNe that result fromcore collapse in massive stars, SNe Ia are believed to arisefrom the thermonuclear disruption of a white dwarf nearthe Chandrasekhar limit and thus occur in evolved stellarpopulations (see Livio 2000 for a review). The loss ofmassive stars in elliptical and S0 galaxies, without compara-ble replacement, quenches the production of core-collapseSNe, while SNe Ia, arising from relatively old progenitors,persist. From the host type of SN 1997, we might readilyconclude, as did GNP99, that it is of Type Ia.

However, more careful consideration is needed to classifySN 1997. Because of its high redshift, we need to determine

15 The Asiago Supernova Catalog is located athttp ://merlino.pd.astro.it/Dsupern/.

FIG. 13.SN type vs. host morphology as compiled by the Asiagocatalog (Cappellaro et al. 1997). This set includes all SNe through SN2001X for which a modern SN classication and galaxy classication areavailable.

30

28

26

24

22

AB

mag

nitu

de

Burst (tau=0.01 Gyr)

0.5 Gyr1.0 Gyr2.0 Gyr

0.5 1.0 1.5 2.0wavelength (m)

30

28

26

24

22

AB

mag

nitu

de

Extended (tau=0.3 Gyr)

1.0 Gyr2.0 Gyr2.5 Gyr

64 RIESS ET AL. Vol. 560

the degree of ongoing star formation and hence the likeli-hood of the appearance of a core-collapse SN from a young,massive star. While the question of how and when ellipticalgalaxies form and evolve is beyond the scope of this paper,we are concerned only with the nature of the star formationhistory and stellar populations in the host of SN 1997.

One study of the host galaxy was made previously byDickinson (1999 ; see also M. Dickinson et al. 2001, inpreparation), who used IR NICMOS photometry tocompare the rest-frame B[V colors of high-redshiftHDF-N ellipticals to those expected for dierent formationand evolution scenarios. Dickinson found the host andanother nearby red elliptical to have rest-frame B[V colorsconsistent with a burst of star formation at zB 4 or 5 fol-lowed by passive evolution.

Here we extend this analysis by comparing the completeultraviolet-optical-infrared SED of the host galaxy toBruzual & Charlot (1993) population synthesis models. Theupper panel of Figure 14 superimposes the galaxy photo-metry with models that assume a single, short burst of star

FIG. 14.Possible star formation histories of the host of SN 1997compared to its observed SED. The upper panel shows the expected SEDfor a single burst of star formation occurring at 0.5 Gyr, 1 Gyr, and 2 Gyrbefore the explosion SN 1997. The favored age of 1 Gyr is long after theloss of the progenitors of core-collapse SNe. For a star formation historyconsisting of a burst (lower panel) followed by continuous and exponentialdecay of extended star formation (q\ 0.3 Gyr), the expected ratio of SNeIa to core-collapse SNe is between 20 and 70.

formation with a Salpeter initial mass function. Such amodel has negligible ongoing star formation after the initialburst, and thus its rest-frame ultraviolet (UV) to opticalcolors redden as quickly as possible. After 1 Gyr haselapsed, this model approximately matches the observed-frame colors of the host galaxy ; a burst occurring 0.5 or 2.0Gyr before the SN appears too short and too long, respec-tively. We expect very few remaining massive stars 1 Gyrafter the cessation of star formation and, therefore, a negli-gible chance that SN 1997 could be a core-collapse SN (theprogenitors of which live for less than 40 Myr). An alterna-tive history would extend the star formation timescale, pro-viding a small residual of ongoing star formation to boostthe UV ux while allowing the rest-frame optical colors toredden to match the IJHK photometry. The bottom panelof Figure 14 shows such a model, with an exponential starformation timescale of 0.3 Gyr. This model matches theobserved SED at an age between 2.0 and 2.5 Gyr (with thefar-UV limit favoring the older age). Normalized to theH-band magnitude of the galaxy, and assuming )

M\ 0.3,

and km s~1 Mpc~1 to compute the)" \ 0.7, H0 \ 70luminosity distance, this model provides an ongoing starformation rate of 0.7 to 0.2 yr~1 at the time SN 1997M

_exploded [(2.410)] 10~4 times the initial rate]. Theremaining population of massive stars should produce 0.004to 0.001 core-collapse SNe per year (in the rest frame). Weemploy a more empirical route to determine the expectedrate of SNe Ia caused by our inability to identify conclu-sively their progenitor systems. Estimates for the rate ofSNe Ia at high redshift from Pain et al. (1996) yield 0.48 SNeIa per century per 1010 solar blue luminosities km(H0 \ 70s~1 Mpc~1). Sullivan et al. (2000) and Kobayashi, Tsuji-moto, & Nomoto (2000) predict a rise in this rate by a factorof D2 at the redshift of SN 1997. The host galaxy lumi-nosity is hence, we expect a rate of D0.07M

B\ [21.9 ;

SNe Ia per year. We thus expect the host to produce 20 to70 times as many SNe Ia as core-collapse SNe at the timeSN 1997 exploded (with the far-UV limit favoring thelarger ratio), favoring its classication as an SN Ia indepen-dent of the cosmological model.

A longer timescale of star formation pushes the time ofthe initial burst uncomfortably close to the formation ofglobular clusters without signicantly altering the expectedproduction ratio of core-collapse SNe to SNe Ia.

The very tentative identication of a noisy spectralfeature with [O II] emission would provide an [O II] uxf ([O II])B 8.7] 10~18 ergs s~1, with a very large uncer-tainty (at least 50%) because of its low S/N. As discussedby Kennicutt (1998), the conversion from [O II] line ux tostar formation rate is imprecise, and large variations in[O II]/Ha (as much as 0.51 dex) are seen among localgalaxies. Nevertheless, adopting Kennicutts conversion fora Salpeter initial mass function (IMF), and assuming (as in 4.1) a cosmology with and km)

M\0.3, )"\0.7, H0\70s~1 Mpc~1, we estimate a host galaxy star formation rate of

2.6 yr~1 from the tentative [O II] line identication.M_This is 4 to 13 times larger than the rates we estimated from

the broadband photometric modeling and would result in arate of core-collapse SNe of D0.01 yr~1, still a factor ofD10 smaller than our estimate of the Type Ia SN rate.However, we consider the identication of this spectralfeature as [O II] emission very tentative and the putativeux very uncertain ; therefore, the calculation of its impliedstar formation rate is highly speculative.

No. 1, 2001 FARTHEST KNOWN SUPERNOVA 65

Another route to estimating the fraction of core-collapseprogenitors comes from a direct conversion of rest-frameUV ux (e.g., from the magnitude, which correspondsB450to roughly 1650 in the host-galaxy rest frame at z\ 1.7)Ato star formation rates, following the conversion for a Sal-peter IMF from Kennicutt (1998). This yields an estimatedstar formation rate of 0.8 yr~1, which is consistent withM

_the results derived previously using an extended star forma-tion scenario. However, this value is likely to be an overesti-mate since the standard conversion factor is estimated frommodels with constant star formation rates while the galaxycolors resemble those of an early-type galaxy for which thecurrent star formation rate is almost certainly far lowerthan its past average. Some signicant fraction of the UVlight may therefore come from longer lived stars and lessneed be attributed to ongoing star formation.

A potentially powerful tool to discriminate between SNtypes comes from enlisting the observed SN data set. Boththe High-z Supernova Search Team and the SupernovaCosmology Project have relied on the photometric behav-ior of an SN when a useful spectrum was not available(Riess et al. 1998 ; Perlmutter et al. 1999). The distance-independent observables of color and light-curve shapehave the potential to discriminate Type Ia SNe from otherSN types. As discussed in 3.1, from the observed colorsand decline of SN 1997, we conclude that its photometricbehavior is inconsistent with a Type II supernova at anyredshift. Also, the scarcity of SNe IIn having similar photo-metric properties argues against SN 1997 being an SN IInon photometric grounds. In contrast, the goodness-of-tbetween SN 1997 and an empirical model of an SN Ia atz\ 1.7 discovered a week after maximum and with atypical light-curve shape (see the middle panel of Fig. 7) ishighly consistent with the SN Ia identication (reduceds2 \ 0.5 for D10 degrees of freedom). Coupled with theapparent consistency of the two photometric redshifts andthe tentative spectroscopic redshift, we might consider theSN data to be the best arbiter of type.

Unfortunately, when using only photometric informa-tion, it may be possible to confuse an SN Ia with a luminousSN Ic or SN Ib (Clocchiatti et al. 2000 ; Riess et al. 1998).However, SNe Ib and Ic are far rarer than SNe Ia(Cappellaro et al. 1997, 1999), and they are expected to arisefrom even more massive progenitors than SNe II (e.g., Wolf-Rayet stars) ; if so, these progenitors should be even lesspopulous than those of SNe II in the comparatively redhost of SN 1997. (However, the masses of progenitors ofSNe II and SNe Ib/Ic can overlap, if the hydrogen envelopeof the progenitor can be lost through mass transfer in abinary system; e.g., Filippenko 1997) Empirically, SNe Iband Ic are common only in very late-type galaxies (i.e., Sc ;Cappellaro et al. 1997, 1999), environments of marked con-trast to the host of SN 1997. The rare, peculiar, highlyluminous supernovae ( hypernovae ) that may be associ-ated with gamma-ray bursts, such as SN 1998bw (e.g.,Galama et al. 1998 ; Iwamoto et al. 1998 ; Woosley,Eastman, & Schmidt 1999), SN 1997cy (Germany et al.2000 ; Turatto et al. 2000), and SN 1999E (Filippenko 2000),also seem to be produced by core collapse in very massivestars.

Based on the nature of the host galaxy (an evolved, redelliptical) and diagnostics available from the observedcolors and temporal behavior of the SN, we nd the mostlikely interpretation is that SN 1997 was of Type Ia.

4.2. Caveats

Because we have employed the SN colors to seek con-straints on the SN redshift and age at discovery, both ofwhich are strong functions of SN color, we cannot employthe colors to determine if there is any reddening by inter-stellar dust. The HDF-N was chosen (at high Galacticlatitude) in part to minimize foreground extinction, so weassume that Milky Way reddening of SN 1997 is negligi-ble. Similarly, given the evolved nature of the red, ellipticalhost, we assume negligible reddening by the host of the SN.However, the apparent consistency with past cosmologicaldeceleration and the apparent inconsistency with contami-nating astrophysical eects reported here would not bechallenged by unexpected, interstellar reddening to SN1997. To demonstrate this conclusion, we reddened the SNby mag in the rest frame and recalculated theA

B\ 0.25

PDF in the distance-redshift plane. As shown in Figure 11,the t is shifted along a reddening vector farther awayfrom the model of astrophysical eects or cosmologicalnondeceleration. While we consider solutions along thereddening vector less likely, they are important to bear inmind when assessing quantitative estimates of cosmologicalparameters based on the previous analysis.

Our empirical model of evolution or intergalactic grayextinction (in magnitudes) is highly simplistic (i.e., linear)and consists only of the product of a constant and the red-shift. Relative to the empty cosmology ()

M\ 0, )" \ 0),this constant is chosen to be 0.3 mag per unit redshift to

match the observed distances of SNe Ia at zB 0.5. Thefunctional form of this model is the same as that derived byT. York et al. (2001, in preparation) from the considerationof a dust-lled universe and is shown to be valid for red-shifts near unity. It also approximates the calculations ofAguirre (1999a, 1999b). However, depending on the epochwhen the hypothetical dust is expected to form, the opticaldepth might be expected to drop at a redshift higher thantwo.

Evolution is far more difficult to model and predictet al. 1998 ; Umeda et al. 1999a, 1999b ; Livio 2000).(Ho ich

For this hypothetical astrophysical eect, our model is thatthe amount of luminosity evolution would scale with themean age available for the growth of the progenitor system(for zB 1). While more complex parameterizations are pos-sible, the salient feature of our simple luminosity evolutionmodel is its monotonic increase with redshift.

Drell et al. (2000) considered somewhat more complexphenomenological models of evolution (in magnitudes) con-sisting of a variable oset and a variable coefficient multi-plied by the logarithm of (1] z). While the functional formof this model may be less motivated by considerations of thenatural scaling of the physical parameters involved (e.g.,time or metallicity), the additional free parameters make itpossible to empirically t both the observed dimming(zB 0.5) and the observed brightening (zB 1.7) withina wider range of underlying cosmological models(e.g., Einsteinde Sitter). To test higher order parameteriza-tions of evolution than the one we considered here willrequire measurements of more distant SNe Ia in new red-shift intervals.

As discussed in Appendices A and B, constraints derivedfrom the photometry of SN 1997 can be recovered fromany of the methods used to characterize the relationshipbetween SN Ia light curves, color curves, and luminosity.

66 RIESS ET AL. Vol. 560

To test the sensitivity of our analysis to the light-curveshape method employed, we rederived the SN constraints in 3 using the stretch method described by Perlmutter etal. (1997). The results were in excellent agreement with thosepresented using the MLCS method (Riess et al. 1998) in 3.The only noteworthy exception is that the constraints onthe expected luminosity and distance of SN 1997 weresomewhat narrower for the stretch-method analysis. Theexplanation for this dierence can be found in the cali-bration of the relationship between the peak luminosity andlight-curve shape used by each method. The stretch methodexpects a somewhat smaller change in peak luminosity for agiven variation in light-curve shape than the MLCSmethod. For the range of possible light-curve shapesallowed by the quality of the t to the SN 1997 data, thestretch method therefore predicts a smaller variation inluminosity (and hence distance) for SN 1997 than theMLCS method. The cosmological conclusions reportedhere are supported by either method.

Sample selection biases can be important factors to con-sider when employing sources detected in a magnitude-limited survey (e.g., Li et al. 2001). However, becausecosmological measurements from SNe Ia are generallybased on the dierence in the apparent luminosities of SNefrom such surveys, a propagated bias in the cosmologicalmeasurements is greatly diminished (Schmidt et al. 1998 ;Riess et al. 1998 ; Perlmutter et al. 1999). In addition, therelatively low intrinsic scatter of SN Ia distance measure-ments (p\ 0.15 mag) further reduces such biases. MonteCarlo simulations of these biases indicate that the appropri-ate corrections to the measured distance moduli are lessthan 0.05 mag, which is negligible compared with the sta-tistical uncertainties for SN 1997 presented here.

It is important to devote special consideration to thelikelihood that SN 1997 resembles the overluminous andslow declining SN 1991T (Filippenko et al. 1992a). An unex-pected overluminosity of SN 1997 would result in an over-estimation of the apparent disagreement with anastrophysical source of dimming. The best estimate of theoverluminosity of SN 1991T is 0.3^ 0.3 mag based on arecent determination of the Cepheid distance to the hostusing HST (Saha et al. 2001). If SN 1997 were unex-pectedly overluminous by this amount, it would still remainbrighter than expected for the dust or simple evolutionmodel (by about 0.8 mag), but the signicance of the dier-ence would be reduced. However, the possibility that SN1997 matches SN 1991T-like SNe is explicitly included inthe analysis in 3 by comparing the t between the photo-metry of the former and the latter (and by employing thewidest luminosity prior). The good t between SN 1997and a typical SN Ia disfavors its identication with theslower declining SN 1991T. Empirically, SN 1991T-likeevents appear to favor hosts with younger stellar popu-lations (Hamuy et al. 2000 ; Howell 2000), but this diagnos-tic is not as useful as the observed light curve shape indetermining the likelihood that SN 1997 resembles SN1991T.

Clustering of mass in the universe can cause the line ofsight to most SNe to be underdense relative to the mean,while an occasional supernova may be seen through anoverdense region. In Riess et al. (1998) and Perlmutter et al.(1999), stochastic lensing that decreased typical uxescaused by underdense lines of sight to the SNe was con-sidered and found to have little eect on the cosmological

measurements. The typical deamplication would be largerat zB 1.7 and may approach a 10%15% decrease in theobserved brightness of a typical SN such as SN 1997 (Holz1998). For a large sample of SNe, the mean would providean unbiased estimate of the unlensed value, but for a singleSN, median statistics are more robust and deamplicationof SN 1997 is applicable.

Lensing by the foreground large-scale structure can alsoalter the apparent brightness of a distant supernova(Metcalf & Silk 1999) in the opposite sense of the precedingconsideration. There is a pair of galaxies in the foregroundof SN 1997 at z\ 0.56, with a separation of D3A and 5A.5from the SN. If we assume km s~1 Mpc~1,H0 \ 70 )M \0.3, and these galaxies are found at projected dis-)" \ 0.7,tances of D20 and 35 kpc in the lens plane and haveapproximately L* luminosity, with implying aM

VB[21,

velocity dispersion of D200 km s~1. Assuming that the twogalaxies have an approximately isothermal mass distribu-tion, the resulting magnication of SN 1997 would beD0.3 mag, in good agreement with the results of Lewis &Ibata (2001). However, without detailed knowledge of theform of the mass potential, it is not possible to recover theprecise amplication (and to accurately correct the mea-sured luminosity). Our estimate is only an approximationsince the mass prole could fall o more steeply thanassumed, or there could be an excess dark matter concen-tration associated with this pair of galaxies.

The Ha line width measured from the NIRSPEC spec-trum for the nearby z\ 0.56 foreground galaxy is not sig-nicantly resolved at an instrumental resolution of about180 km s~1 (FWHM), indicating p 90 km s~1 with noindication of any rotational sheer. This may not provide astrong constraint on the foreground galaxy mass, givenuncertainties in the galaxy orientation (it appears to besomewhat face-on, although this is difficult to assess givenits irregular morphology) and the fact that the line emissionmay trace only one star-forming region within the galaxyrather than the full potential well depth. We can only saythat there is no immediate kinematic evidence from existingspectroscopy for a large mass for the closest foregroundgalaxy.

It is possible to derive a useful constraint on the maximumlikely amplication of the SN by the closest foregroundlenses by examining the shape of the host galaxy that wouldbe stretched in the tangential direction by an amount thatdepends on the SN amplication. This calculation is per-formed in Appendix C and the results can be seen in Figure15. From this calculation, we conclude that the lack ofapparent tangential ellipticity of the host galaxy (for thedegree of specic SN amplication estimated above) is notvery surprising (D20% of randomly selected hosts wouldexhibit as little tangential stretching as seen). However, sig-nicantly greater amplications are not very likely ; 0.6 or0.8 mag amplications would produce galaxies with no evi-dence of tangential stretching only in 6% and 3% of identi-cal ensembles, respectively. The observed roundness of theimage of the host galaxy, with an axis ratio of 0.85, is furthercircumstantial evidence against the presence of substantiallensing. While unremarkable in the absence of lensing, suchroundness is unusual in im