Документ взят из кэша поисковой машины. Адрес оригинального документа : http://hea-www.harvard.edu/~pgreen/Papers/lbqslines00.pdf
Дата изменения: Wed Sep 14 22:52:44 2005
Дата индексирования: Tue Oct 2 01:42:58 2012
Кодировка: koi8-r
Поисковые слова: guide 8.0

THE ASTROPHYSICAL JOURNAL SUPPLEMENT SERIES, 134 : 35 х 51, 2001 May
( 2001. The American Astronomical Society. All rights reserved. Printed in U.S.A.

V

EMISSION LINE PROPERTIES OF THE LARGE BRIGHT QUASAR SURVEY KARL FORSTER
kforster=cfa.harvard.edu

PAUL J. GREEN AND THOMAS L. ALDCROFT
Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138 ; pgreen=cfa.harvard.edu, aldcroft=cfa.harvard.edu

M. VESTERGAARD
The Ohio State University, 140 West 18th Avenue, Columbus, OH 43210-1173 ; vester=astronomy.ohio- state.edu

CRAIG B. FOLTZ
Multiple Mirror Telescope Observatory, University of Arizona, Tucson, AZ 85721 ; cfoltz=as.arizona.edu

AND PAUL C. HEWETT
Institute of Astronomy, Madingley Road, Cambridge, CB3 0HA, UK ; phewett=ast.cam.ac.uk Received 2000 August 9 ; accepted 2001 January 2

ABSTRACT We present measurements of the optical/UV emission lines for a large homogeneous sample of 993 quasars from the Large Bright Quasar Survey. Our largely automated technique accounts for continuum breaks and galactic reddening, and we perform multicomponent пts to emission line proпles, including the eects of blended iron emission, and of absorption lines both galactic and intrinsic. Here we describe the пtting algorithm and present the results of line пts to the LBQS sample, including upper limits to line equivalent widths when warranted. The distribution of measured line parameters, principally W and j FWHM, are detailed for a variety of lines, including upper limits. We thus initiate a large-scale investigation of correlations between the high-energy continuum and emission lines in quasars, to be extended to complementary samples using similar techniques. High-quality, reproducible measurements of emission lines for uniformly selected samples will advance our understanding of active galaxies, especially in a new era of large surveys selected by a variety of complementary methods. Subject headings : galaxies : active х quasars : emission lines х quasars : general х ultraviolet : galaxies On-line material : machine-readable tables
1

. INTRODUCTION

Quasars, some of the most luminous objects in the universe, allow us to observe toward the start of cosmic time, probing extremes of physics, and illuminating the intervening matter. Though most of the QSOs now known were discovered by their unique spectral energy distributions (SEDs) or emission lines, several interrelated and fundamental questions remain : 1. What signiпcant correlations exist between the observed SED and emission lines in large uniform quasar samples ? 2. How are the observed SED and emission lines aected by intrinsic absorption and orientation ? 3. Can the shape of the intrinsic SED be determined ? 4. What eect does the intrinsic SED have on emission lines and what does that tell us about physical conditions in the QSO ? The production of emission lines in QSO spectra is widely attributed to photoionization and heating of the emitting gas by the UV to X-ray continuum (e.g., Ferland & Shields 1985 ; Krolik & Kallman 1988). Emission lines from a given ion are particularly sensitive to photons of energy above the corresponding ionization threshold, but ionization from excited states and heating via free-free and H~ absorption also determine the lineоs principal ionizing/ heating continuum. Many important lines respond to the extreme ultraviolet (EUV) or soft X-ray continuum. Unfor35

tunately, the EUV band is severely obscured by Galactic absorption, although some constraints on the EUV ionizing continuum are available through analysis of the adjacent UV and soft X-ray windows. The overall similarity of QSO emission line spectra has on occasion been taken as evidence of fairly uniform physical conditions in the broad emission line region (BELR). At пrst, this encouraged the assumption that clouds in the BELR inhabit a narrow swath of parameter space in density, size, and ionization parameter (U). Observed correlations of line equivalent widths (hereafter W ) with SEDs j present challenges to geometric and photoionization models that need to be answered. Photoionization pioneers such as Mushotzky & Ferland (1984) ran models on a single cloud, while photoionization models assuming a distribution of clouds (i.e., not a single U) predict a strong dependence of W on the (assumed power law) slope of the SED (Binette et j 1989 ; Baldwin et al. 1995 ; Korista, Baldwin, al. & Ferland 1998). Large uniform samples of QSO emission line measurements are key to test photoionization models but are also necessary for other important tests of quasar triggering and evolution. Tests of the binary versus lens hypothesis for quasar pairs at the same redshift require quantitative measures of the probability of пnding the measured spectral similarity derived from complete samples (Mortlock, Webster, & Francis 1999). Since optically selected QSOs at similar redshift may tend to be more similar than two QSOs chosen entirely at random, studies of QSO spectral properties should be based on samples large enough to provide

36

FORSTER ET AL.

Vol. 134

statistically signiпcant subsamples across a wide range of luminosity and redshift (Peng et al. 1999). In a small, complete sample of optically selected QSOs studied by Laor et al. (1997), the strongest correlation found between X-ray continuum and optical emission line parameters was of the soft X-ray spectral slope (a ), and the x FWHM of Hb. Strong correlations between a and the relax tive strengths of [O III] j5007 and iron emission were seen both there and in previous studies, particularly Boroson & Green (1992, hereafter BG92). A principal component analysis (PCA) was employed by BG92 to examine an unwieldy set of emission line and continuum correlations within a sample of 87 QSOs taken from the Palomar-Green Bright Quasar Survey (BQS ; Schmidt & Green 1983). This indicated that the strongest sources of variance in the optical spectra, the "" primary eigenvector,оо involved an anticorrelation between the strength of [O III] j5007 and iron emission that was also linked to the FWHM of Hb. This work was extended by Marziani et al. (1996, hereafter M96) and Sulentic et al. (2000) to include a study of the spectral region around C IV j1549. These papers identiпed Narrow-line type 1 Seyfert galaxies (NLS1) and steep spectrum radio galaxies as occupying opposite ends of the primary eigenvector. Other work utilizing the high spectral resolution of the Hubble Space T elescope Faint Object Spectrograph (HST FOS) showed a link between Al III j1859, Si III j1892, and C III j1909 emission line strengths and the primary eigenvector but suered from the small sample size and a focus on NLS1 or the low-redshift radioquiet objects in the BQS (Aoki & Yoshida 1999 ; Kuraszkiewicz et al. 2000). The samples of active galactic nuclei (AGNs) used by most authors are almost entirely taken from BG92 and so do not represent a statistically complete sample and may not be truly representative of the QSO population (Wisotzki et al. 2000). They also deal with the strongest emission features in the OUV spectrum and, while several hypotheses have been proposed, the physical origin of the observed correlations remains a puzzle. There is also a concern that biases within these small samples may have exaggerated the signiпcance of the primary eigenvector. Even in the clear presence of extrinsic eects such as absorption along the line of sight, emission lines have been used to infer the strength and shape of the high-energy SED. As an example, the similarity of UV emission-line properties in broad absorption line (BAL) and non-BAL QSOs (Weymann et al. 1991) has been cited as evidence that orientation is the cause of the BAL phenomenon (i.e., that all radio-quiet QSOs have BAL clouds). However, BALQSOs are now known to exhibit markedly weak X-ray emission as a class (Green & Mathur 1996 ; Gallagher et al. 1999 ; Brandt, Laor, & Wills 2000). But if similar emission lines indeed vouch for similar intrinsic high-energy SEDs, then the large X-ray to optical continuum яux ratio (a ) values observed for BALQSOs are likely to be caused ox by strong absorption along the line of sight rather than by dierences in their intrinsic SEDs. In fact, the soft X-ray deпcit requires absorption at least an order of magnitude larger than those estimated from optical/UV (OUV) spectra alone and the UV and X-ray absorbers have yet to be deпnitively identiпed as one (e.g., see discussion in Mathur, Wilkes, & Elvis 1998). Since BALQSOs may be heavily absorbed, the question of whether similar emission lines are testimony for similar intrinsic SEDs must be answered

through the systematic study of line and continuum correlations in unabsorbed QSOs. Even in supposedly unabsorbed QSOs, a strong inverse correlation exists between optical iron emission, and O III j 5007 in QSOs (BG92). Iron emission is underpredicted by a large factor in standard photoionization models relative to observations (Collin & Joly 2000), and the continuum source responsible for creating rest-frame optical iron emission lines comes almost entirely from X-rays above about 1 keV. QSOs with large a of which BALQSOs represent an ox extreme, have not only weak O III, but feeble narrow-line emission in general (Green 1996). We suggest that the observed correlations may be an eect of absorption of the intrinsic high-energy continuum by thick clouds interior to the NELR. The X-ray heating of these clouds makes them emit copious Fe II lines but prevents the ionizing continuum from reaching the NELR. This is reinforced by another observed correlationхthat the most highly absorbed BALQSOs, the low-ionization BALQSOs, albeit X-ray quiet, are generally very strong optical Fe II, emitters (Stocke et al. 1992 ; Lipari, Terlevich, & Macchetto 1993). Our interpretation needs to be tested across a range of absorption, in large representative QSO samples. We have therefore undertaken a major study of quasar line emission, with careful accounting for absorption lines and blended iron emission, using largely automated procedures on carefully selected samples. The analysis of samples of QSOs with more than a few hundred spectra requires some amount of automation to give consistent results. The largest sample of QSO spectra currently available is that of the Large Bright Quasar Survey, and here we describe the initial results from the measurement of this sample and the analysis methods that will also be applied to other large samples of QSO spectra that will become available in the near future.
2

. THE LBQS SAMPLE

The Large Bright Quasar Survey (LBQS ; see Hewett, Foltz, & Chaee 1995) is a sample of 1058 QSOs selected using the Automatic Plate Measuring Machine from UK Schmidt photographic direct and objective prism plates. The combination of quantiпable selection techniques, including overall spectral shape, strong emission lines, and redshifted absorption features has been shown to be highly efficient at пnding QSOs with 0.2 \ z \ 3.3, a signiпcantly broader range than previous work. The LBQS thus avoids selection eects common in other optical quasar samples that tend to exclude weak-lined quasars or to undersample certain redshift ranges or colors. There appears to be a deпciency in the numbers of QSOs with z D 0.8 in the LBQS sample, although the eect is not statistically signiпcant (Hewett et al. 1995). The LBQS has been for the last decade the principal source of intermediate brightness optically selected quasars available ; 1/9 of the Veron-Cetty & Veron (1998) catalog of QSOs and D50% of all known quasars within a similar magnitude range. Radio data are currently available for about 1/3 of the sample (Hooper et al. 1995) and soft X-ray яuxes and upper limits for at least 85% of the sample can be obtained from the ROSAT All-Sky Survey. Follow-up optical spectra with S/N B 10 in the continuum at 4500 A and 6х10 A resolution were obtained between 1986 and 1990 at the MMT on Mount Hopkins, Arizona, and the 2.5 m duPont in Las Campanas, Chile, for all LBQS QSOs. Error spectra (1 p) are available

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY

37

for 1009 of the objects. The selection criteria and observing procedures for each of the survey пelds can be found in Hewett et al. (1995) and references therein (LBQS Papers IхVI). Three QSOs appear here that were identiпed subsequently to the 1055 QSOs in the Hewett et al. (1995) LBQS catalog. They are 0052]0148 (z \ 0.595), 1027[0149 (z \ 0.754), and 2132[4227 (z \ 0.569), see Hewett, Foltz, & Chaee (2001, in preparation) for further details.
3.

ANALYSIS OF THE SPECTRA

The LBQS spectra were analyzed using "" Sherpa,оо1 ageneralized пtting engine designed primarily for spatially resolved spectroscopy of observations with NASAоs Chandra X-ray Observatory. Sherpa enables data to be modeled using a variety of optimization methods and with a number of built-in statistical tests. The ease with which user-models can be created and the ability to simultaneously model a number of input data sets that may be in dierent formats (e.g., modeling an ASCII and a PHA style FITS пle together) makes Sherpa suitable for our goals. In brief, the process begins with modeling the continuum emission and then accounting for any emission by iron complexes in the spectra. We then model the emission lines using Gaussian proпles and, after a search for signiпcant absorption features, the results are inspected and the emission line пtting procedure repeated to improve the model of each spectrum. The model parameters were determined from a minimization of the s2 statistic with modiпed calculations of uncertainties in each bin (Gehrels 1986). We found that a Powell optimization method gave a balance between an efficient processing time and consistency of results. Because
1 Sherpa is available from the Chandra science center (http :// asc.harvard.edu).

of the large database of observations the modeling of the spectra proceeds in a largely noninteractive manner. We do not present measurements of emission lines for objects with strong BALs. The diversity in the strength and shape of BALs precludes even semiautomated measurements, and their emission line properties have already been well characterized for a strongly overlapping sample in Weymann et al. (1991). We also exclude objects that have multiple narrow absorption features that signiпcantly mask the true shape and strength of the emission features of Lya and C IV j1549. A list of the 65 objects excluded from this study can be found in Table 1. This list includes two objects not previously classiпed as BAL in LBQS Papers IхVI, 0059[2545 and 1242]1737. The measurements for the remaining 993 LBQS spectra are presented here. 3.1. T he Continuum Our пrst step is to model the continuum emission by пtting one or two power laws to ranges in the spectrum that are free from any strong emission lines. We found the second power-law component to be necessary due to the steepening of the continuum toward the UV in low-redshift objects, i.e., for spectra that extended redward of rest frame 4200 A. A list of the continuum modeling "" windows,оо along with the nearest strong emission lines, can be found in Table 2 (see Figs. 1 and 2). We apply a Galactic reddening correction to the power-law continua that follows the prescription given by Cardelli, Clayton, & Mathis (1989), which uses A(j) \ E(B [ V )[aR ] b] , V where a and b are polynomial functions of wavelength derived for 3.3 km є j є 1000 A. We use a ratio of total to selective extinction of R \ 3.1, the standard value for the V diuse ISM, and take the relationship between the color

FIG. 1.хContinuum and iron template modeling windows. The composite QSO spectrum from Francis et al. (1991) is shown above the iron emission templates for the UV (upper panel) and optical (lower panel) regions. The vertical dashed lines mark the positions of the narrow windows used to constrain the continuum shape for each spectrum. The spectral windows used in the modeling of iron emission are shown. See °° 3.1, 3.2, and Table 2 for further details.

38

FORSTER ET AL.
TABLE 1 LBQS QUASARS EXCLUDED FROM EMISSION LINE MEASUREMENTS Designation 0004 0010 0013 0018 0018 0019 0021 0022 0025 0029 0045 0051 0054 0059 0059 0059 0100 0103 0109 1009 1016 1029 1133 1138 1203 1205 1208 1212 1214 1216 1219 1224 1228 ] [ [ [ ] ] [ ] [ ] [ [ ] [ [ [ [ [ [ ] [ [ ] [ ] ] ] ] ] ] ] ] ] 0147 0012 0029 0220 0047 0107 0213 0150 0151 0017 2606 0019 0200 2545 0206 2735 2809 2753 0128 0222 0248 0125 0214 0126 1530 1436 1535 1445 1753 1103 1244 1349 1216 ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... ...... Redshift 1.710 2.154 2.083 2.596 1.835 2.130 2.293 2.826 2.076 2.253 1.242 1.713 1.872 1.955 1.321 1.593 1.768 0.848 1.758 1.349 0.717 2.029 1.468 1.266 1.628 1.643 1.961 1.627 0.679 1.620 1.309 1.838 1.408 Ref.a II II 1, 2 1 II II II II II II V IV IV 3, 4 IV V V V IV III III III III III III I I I I I III I I Notes Designation 1230] 1230] 1231] 1235] 1235] 1235] 1239] 1240] 1240] 1240] 1242] 1243] 1314] 1331[ 1332[ 1333] 1442[ 1443] 2111[ 2113[ 2116[ 2140[ 2201[ 2208[ 2211[ 2212[ 2241] 2244[ 2350[ 2351] 2355[ 2358] 1627B ...... 1705........ 1320........ 1453........ 0857........ 1807B ...... 0955........ 1516........ 1551........ 1607........ 1737........ 0121........ 0116........ 0108........ 0045........ 0133........ 0011........ 0141........ 4335........ 4345........ 4439........ 4552........ 1834........ 1720........ 1915........ 1759........ 0016........ 0105........ 0045A ...... 0120........ 0209........ 0216........ Redshift 2.735 1.420 2.380 2.699 2.898 0.449 2.013 2.297 0.573 2.360 1.863 2.796 2.686 1.881 0.672 1.577 2.226 2.451 1.708 2.053 1.480 1.688 1.814 1.210 1.952 2.217 1.394 2.030 1.617 2.068 1.802 1.872 Ref.a i III I I I I I i iii III III III III iii III III V v V V V V V V II ii II II IV 6 2 Notes 1, 2

Vol. 134

1, 2 5 1, 2, 3, 4

1, 2

1, 2, 7 1, 2

NOTES.х(1) NAL on Lya ; (2) NAL on C IV ; (3) BAL near Lya ; (4) BAL near C IV ; (5) NAL on Mg II ; (6) BAL near Mg II ; (7) NAL on N V. a Original LBQS paper that identiпed BAL (upper case) or associated narrow absorption (lower case). REFERENCES.х(I, i) Foltz et al. 1987 ; (II, ii) Foltz et al. 1989 ; (III, iii) Hewett et al. 1991 ; (IV, iv) Chaee et al. 1991 ; (V, v) Morris et al. 1991.

excess and the line of sight column of neutral hydrogen for each object to be N (Galactic)(1020 cm~2) E(B [ V ) \ H 58.0 (Bohlin, Savage, & Drake 1978). Many of the spectra in the LBQS sample depart from a simple reddened power-law continuum for wavelengths

blueward of D4000 A in the observed frame as noted in the original LBQS papers. This depression in the observed ultraviolet is due to the use of 2A circular apertures in the .5 MMT observations coupled with the eects of atmospheric dispersion at large zenith distances and a guider system that was red-sensitive. To model this part of the continuum, we use a polynomial of up to second order with the яux tied to that of the power-law continuum model in the nearest continuum window with j (observed frame) greater than 4000 c

FIG. 2.хContinuum model for LBQS 1228]1116 (z \ 0.237). The bond between the polynomial continuum and the пrst power law is at 4000 A in the observed frame and the inяection point between the two power-law continua is at rest frame 4220 A (see ° 3.1).

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY
TABLE 2 CONTINUUM AND IRON FITTING WINDOWS REST FRAME WAVELENGTH RANGE (A) Continuum 1140х1150a 1275х1280 1320х1330 1455х1470 1690х1700 2160х2180 2225х2250 3010х3040b 3240х3270 3790х3810 4210х 4230 5080х5100 5600х5630 5970х 6000 Iron EMISSION LINES NEARBY Blueward O VI j1035 N V j1243 O I j1305 Si IV ] O IV j1400 He II j1640 2020х2120 2250х2650 2900х3000 [0pt]C III j1909 [0pt]Mg [O II] Hd j [O III] II j2800 j3728 4102 j4363 Redward Lya j1215 O I j1305 Si IV ] O IV j1400 C IV j1549 Al III j1859 Mg II j2800 [Ne V] j3426 [Ne III] j3869 Hc j4340 Hb j4861 He I j5876 [N II] j6549

39

4400х 4750c 5150х5500

[O III] j5007 He I j5876

a This window lies on the blue side of the Lya emission line and is only used where no other continuum window is available. b May have some iron emission contamination. c He II j4686 lies in this window.

A. We did not apply a polynomial continuum to objects with z є 2.42 because the Lya emission line appears redward of 4000 A and the continuum on the blue side of Lya can be signiпcantly reduced by Lya forest lines. The polynomial continuum model is not physically meaningful and may be inaccurate, particularly where strong iron emission complexes appear around the Mg II j2800 emission line, i.e., for [ 3100A . rest We investigated the eect of the polynomial continuum on W measurements and on the error estimates for the emisj sion line parameters by comparing lines of the same species that occurred in the polynomial continuum region to those that did not. A Kolmogorov-Smirnov test (e.g., Press et al. 1992) shows that the distributions of W measurements for j the UV iron emission, Si IV ] O IV j1400, C IV, and Mg II emission lines are signiпcantly dierent when measured above the polynomial continuum compared to when measured using a power-law continuum. The median increase in W for C IV and Mg II is D20% when these lines fall above j the polynomial continuum but genuine observational trends like the Baldwin Eect (Baldwin 1977 ; Osmer, Porter, & Green 1994) may account for this measurement trend. The intrinsic continuum luminosity of objects with C IV redshifted to j (observed frame) greater than 4000 A is on average twice that of the lower redshift objects where C IV falls above the polynomial continuum. A 20% increase in W for C IV and Mg II is predicted using the relationships j between W and continuum luminosity determined by j Zamorani et al. (1992) but may not fully account for the dierences seen in the measurements of W for these and j other emission lines. Rather than correcting for this eect, we added a systematic uncertainty of 20% to the estimation of emission line strengths and upper limits for all emission lines measured above the polynomial continuum. Finally, we visually inspected the continuum model for each observation and found that minor adjustments were 2200A [ j

required in D20% of the sample ; a large proportion of these were caused by strong intrinsic absorption lines. 3.2. Iron Emission Our second step accounts for iron emission line complexes in the spectra. This emission occurs most strongly in the regions around the Mg II j2800 and Hb emission lines. We follow the prescription used by BG92 of subtracting a template of iron emission lines created from the spectrum of I Zw 1, an NLS1 which exhibits the typically strong iron emission of this class of AGNs. We use two templates : the optical emission line template is identical to that used by BG92 and corrects for iron emission between 4400 A \ j \ 7000 A, while the UV template was developed by rest Vestergaard & Wilkes (2000) from HST FOS observations of I Zw 1 and covers the region 1250 A \j \ 3100 A. rest The FWHM of the iron emission features seen in I Zw 1 is similar to that of the Hb emission line (900 km s~1) and so sets the minimum template FWHM. We convolved this template with a series of Gaussian functions of increasing width to produce a grid of 38 template spectra with 900 [ FWHM [ 10,000 km s~1, conserving the total яux in each template. This provides us with a nominal resolution of 250 km s~1, more than adequate for the quality of the spectra in the LBQS sample. The пrst part in modeling the iron emission is to compare the 2000 km s~1 templates to windows in each spectrum on both sides of the Mg II j2800 emission line and redward of C III j1909 for the UV emission, and to either side of the Hb ] [O III] jj4959,5007 complex for the optical emission (see Table 2). After the relative amplitude of the iron emission has been modeled, the full grid of templates is applied to the spectra to determine the approximate FWHM of the iron emission. This is followed by a пnal modeling of the template amplitude simultaneously with the FWHM of the template spectra. Adjustments are made from a visual inspection of the results to give the best representation of the iron emission complexes present in each spectrum.

40

FORSTER ET AL.
TABLE 3 INVENTORY OF EMISSION LINES j c (A) 1030.0 1215.7 1241.5 1305.0 1400.0 1549.0 1640.0 1859.0 1909.0 2800.0 3426.0 3728.0 3869.0 4101.7 4352.0 4686.5 4861.3 4959.0 5007.0 5875.6 FWHMa (km s~1) 5000 7000s 2500n, 8000b 6000 2500 5800 6500s 3000n, 11000b 10000 3500 5500 4000s 3000n, 8000b 1000 600 900 450 3500 1200 4000s 1000n, 5500b 700 600 2000 Windowb (A) 1010х1060 1170х1350 ... ... ... 1350х1450 1350х1720 ... ... 1820х1970 ... 2700х2900 ... 3390х3460 3700х3760 3810х3930 4000х 4200 4240х 4440 4580х 4790 4750х5100 ... ... ... 5825х5900 See Appendix A

Vol. 134

Emission Line Lyb j1025.7 ] O VI j1035 ....... Lya j1215.7 ....................... N O Si C V j1241.5 ....................... I j1305 .......................... IV ] O IV j1400................ IV j1549 .........................

Notes

He II j1640 ........................ Al III j1859 ........................ C III j1909 ......................... Mg II j2800 ....................... [Ne V] j3426 ...................... [O II] j3728 ....................... [Ne III] j3869 ..................... Hd j4101.7 ........................ Hc j4340.5 ] [O III] j4363 ...... He II j4686.5 ...................... Hb j4861.3 ........................ [O III] j4959 ...................... [O III] j5007 ...................... He I j5875.6 .......................

See Appendix A

Iron emission may be strong in this window See Appendix A See Appendix A See Appendix A

See Appendix A Iron emission may be strong in this window Relative strength and separation of [O III] lines is пxed in пrst iteration ...

a This is the FWHM used for estimating the upper limits of W in weak emission lines. More than one width is given for emission lines that can have two j Gaussian components ; s \ single component, n \ narrow component, b \ broad component. b The wavelength range over which the emission lines are modeled.

Note that we measure the strengths of the iron emission in the UV and optical independently. This is important as the relative strength of the UV and optical iron emission may allow an approximate determination of the conditions in the emitting region, e.g., density and U (Verner et al. 1999). UV and optical iron emission may also correlate dierently with X-ray emission. Green et al. (1995) found that QSOs in the LBQS with strong UV Fe II emission (based on the iron feature under [Ne IV] j2423) are anomalously X-ray bright in the ROSAT passband. By contrast, several studies (e.g., Corbin 1993, Lawrence et al. 1997) пnd an anticorrelation between soft X-ray luminosity and R(Fe II) (the equivalent width ratio of optical iron emission to Hb). Several such intriguing correlations have been noted, but among samples of varying size and using heterogeneous measurement techniques. Our project attempts to remedy this situation using large samples and uniform analysis. When пrst scaling the UV iron template in the LBQS spectra to the iron emission complexes straddling Mg II (Djj2200х3300), we found that the template tended to overpredict the iron emission line strength blueward of C IV j1549. This occurred in 27 of the 238 spectra that cover both C IV and Mg II and that have measurable iron emission. The UV iron template is based on an HST FOS spectrum of I Zw 1 for which the region blueward of C IV was observed six months earlier than the remaining spectrum. I Zw 1 clearly brightened between the two observations, for which reason the bluest subspectrum was scaled to match the redder spectrum. The reasons for the overprediction of iron яux blueward of C IV in our sample when using the I Zw 1 template are not clear, but may include a change in

iron emission strength and multiplet ratios with luminosity indicating variations in the physical conditions to which the iron emission is sensitive (e.g., Netzer 1980). See Vestergaard & Wilkes (2000) for further details and discussion. To compensate for this eect, the UV iron template яux in the region j \ 1530 A was reduced by 50%, resulting in an rest improved agreement with all the sample spectra. 3.3. Emission L ines The emission lines in the spectra are modeled by the addition of Gaussian features at the expected wavelengths to the continuum ] iron template model. The initial value of the peak яux near the position of the line is estimated from the raw data assuming a continuum level under each emission line calculated by пtting a straight line between the continuum windows nearest the line to be modeled. The initial estimate of the FWHM of the emission line component is assumed to be 3000х5000 km s~1 for broad components and 350 km s~1 for narrow components. These reяect the typical widths of emission lines produced in the BELR and NELR (e.g., Peterson 1997). Where the S/N in the spectra are high enough, the Lya, C IV, Mg II, and Hb emission lines may be modeled using two Gaussian components. The initial value of the peak amplitude of each component is set to be 40% of the peak яux estimated from the data. Lower S/N spectra are modeled with a single Gaussian but a visual inspection of the results is made and a second Gaussian component is added where signiпcant residuals appear. The FWHM and peak amplitude of the Gaussian model components are then optimized within a spectral region covering the emission line. The position of the emission line

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY
4

41

is пxed at the expected wavelength in the initial modeling until after the presence of narrow absorption features has been determined (° 3.4). The region of the spectrum used in modeling an emission feature is chosen to best optimize the component of interest, e.g., the Lya emission line is simultaneously пtted with N V j1240 and O I j1305 because the broad component to Lya may have a signiпcant eect on the spectrum near O I. A list of the emission lines modeled in the LBQS spectra is presented in Table 3. There are a number of important emission lines (e.g., C II j1336) that are not modeled here due to the quality of the spectra but will be included in the application of this procedure to samples of higher spectral resolution (e.g., HST FOS spectra). 3.4. Absorption L ines One of the goals of this study is to determine the frequency of occurrence of absorption in the spectra of QSOs. We used the program FINDSL (Aldcroft 1993) to detect signiпcant narrow absorption features and model them with multicomponent Gaussian пtting. Our input to FINDSL is the sum of the reddened continuum, iron template(s), and all the emission line proпles from the пrst Sherpa modeling. This constitutes a "" continuum оо from which FINDSL detects signiпcant deviations in the observed spectrum. We rescale the error array of each spectrum for use with FINDSL so that s2 becomes unity, which yields a better estimate of the true lerror array and aids in the detection of signiпcant absorption features. We exclude the Lya forest region blueward of rest frame 1065 A from the absorption line detection and also exclude the Balmer continuum region (between 3360 and 3960 A) where the global contin uum model may lie below the spectrum causing many spurious absorption lines to be detected. The absorption line measurements from FINDSL were then included with the emission line and continuum results for further modeling using Sherpa. In this second iteration the emission line positions are also пtted, and the absorption lines modeled simultaneously with the emission line components. The results from this automated process were inspected and adjustments made to those spectra that the algorithms could not successfully model. This was not unexpected as the wide variety of spectra in the sample and the chance superposition of absorption features on the emission lines makes it unlikely that any fully automatic procedure would be 100% successful. 3.5. Error Analysis An estimate of the 2 p error range for each parameter of the continuum and emission line components was determined from the s2 conпdence interval bounds (*s2\ 4.0). Where the amplitude of an emission line could not be constrained to within 2 p, the line position was reset to the expected wavelength and the FWHM to a median value that was determined from the distribution of wellconstrained emission line measurements within the LBQS sample. These пxed emission line parameters are given in Table 3. The emission line amplitude was then reпtted and, if remaining unconstrained at the 2 p level, then the 2 p upper limit on the amplitude was estimated. The determination of upper limits for lines that may be present at low яux levels in the spectra of QSOs is one of the primary objectives of this work. This allows the use of survival analysis techniques in determining a more realistic distribution of emission line strengths.

. EMISSION LINE RESULTS

The total number of emission lines measured from the spectra in the LBQS sample are presented in Table 4. The numbers of upper limits measured, and the number of twocomponent emission lines are also tabulated. The full table of all emission line measurements and uncertainties of the 993 spectra modeled here is available in the electronic edition of the Journal, but due to the large size it cannot be reproduced in the print edition. To show the form and content of the large electronic table we present in Table 5 a digested version containing measurements of the emission lines present in three LBQS spectra. The emission line measurements are quoted to a signiпcance level determined by the 2 p uncertainties in the parameters (e.g., Bevington & Robinson 1992). The format is as follows : Each object is represented by a row containing the designation and redshift followed by a row for each measured emission line. The name of the emission line is given in column (1), column (2) gives the FWHM in km s~1, and column (3) gives the oset of the peak of the Gaussian emission line model, in km s~1, from the expected position based on the tabulated redshift. Note that no peak oset measurements appear for the iron emission line measurements. Column (4) gives the rest frame equivalent width of the emission line in A. Each emission
TABLE 4 TOTAL NUMBER OF EMISSION LINES MODELED Fixed FWHM and Positiona ... ... 10 2 0 0 24 20 12 3 0 0 66 63 20 3 0 0 10 27 12 6 4 5 4 0 0 6 8 0 305 W j Upper Limits 294 109 27 0 0 0 8 121 19 2 0 0 69 181 26 42 0 0 365 272 259 222 87 155 18 0 0 73 54 30 2437 Total Out of 993 953 247 130 164 96 96 260 260 414 408 80 80 488 667 667 559 118 118 488 393 363 309 251 187 141 7 7 148 148 33 8288

Emission Line UV iron ........... Optical iron ....... Lyb ] O VI ........ Lya : Single ........... Narrow ......... Broad ........... N V ................ O I ................. Si IV ] O IV ....... C IV : Single ........... Narrow ......... Broad ........... He II j1640 ........ Al III ............... C III ................ Mg II : Single ........... Narrow ......... Broad ........... [Ne V] ............. [O II] .............. [Ne III] ............ H d ................. H c ] [O III] ...... He II j4686 ........ H b: Single ........... Narrow ......... Broad ........... [O III] j4959 ...... [O III] j5007 ...... He I ................ TOTALS ..........

a Does not include W upper limits. j

42
TABLE 5 REPRESENTATIVE EMISSION LINE MEASUREMENTS Designation (z) Emission Line (1) FWHM (km s~1) (2) *j p (km s~1) (3) W j (A) (4)

FORSTER ET AL.

Vol. 134

Absorption Lines (5)

0025]0009 z\ 0.205 UV iron ........... Optical iron ....... Mg II single ....... [Ne V] ............. [O II] .............. [Ne III] ............ Hd .................. Hc ] [O III] ....... He II j4686 ........ Hb single .......... [O III] j4959 ...... [O III] j5007 ...... He I ................ 900`9000 ... 70`65 ~250 ~65 1200`8750 ... 74`15 ~250 ~15 2800`1000 [500`440 40`35 ~850 ~440 ~30 1200`1100 [100`380 5.5`10.0 ~750 ~380 ~4.3 210`440 100`140 3.6`6.1 ~200 ~140 ~3.1 1000`3000 0`2200 5.0`30.0 ~700 ~2200 ~4.4 5500`5800 0`1100 35`55 ~1100 ~1000 ~14 2800`850 [200`380 34`20 ~700 ~380 ~14 1200 ... №15 3000`550 300`240 105`35 ~500 ~240 ~30 600`1100 0`550 15.0`40.0 ~280 ~550 ~8.5 450`120 100`50 33`13 ~120 ~50 ~11 600`650 [300`240 8.0`25.0 ~460 ~240 ~7.9 2244]0020 z \ 0.973 ... 900`280 ~280 [400`90 ~90 [250`120 ~140 200`900 ~900 ... 50.6`1.7 ~1.7 12.8`4.7 ~4.4 27.9`7.3 ~7.2 35.6`4.7 ~4.0 27.4`8.5 ~6.3 №2.8 ... ... ... ... ... ... ... 1 ... ... ... ... ...

UV iron ........... 4250`5750 ~3250 Al III ............... 6400`600 ~500 C III ................ 5550`180 ~160 Mg II narrow ...... 4000`280 ~260 Mg II broad ....... 12500`2100 ~1500 [Ne V] ............. 1000

... ... ... ... ... ...

2354[0134 z \ 2.211 UV iron ........... 2000`8000 ... 35`25 ... ~1000 ~25 Lyb ] O VI ........ 5400`850 1300`420 24.9`7.8 ... ~800 ~420 ~6.5 Lya narrow ....... 3500`200 150`90 33`11 ... ~200 ~140 ~10 Lya broad ......... 8100`440 500`180 57`17 1 ~340 ~240 ~17 N V ................. 5400`400 0`160 34`11 ... ~380 ~220 ~11 O I ................. 600`420 100`160 1.3`1.5 1 ~440 ~160 ~0.8 Si IV ] O IV ....... 5000`700 0`380 13.4`3.8 ... ~650 ~380 ~3.2 C IV single ......... 6100`360 50`180 52.7`6.0 ... ~360 ~180 ~5.7 He II j1640 ........ 17000`3800 [1200`650 26.0`12.0 ... ~3100 ~650 ~8.6 Al III ............... 3500 ... №7.0 ... C III ................ 6500`1900 [400`900 32`17 ... ~1500 ~600 ~12 NOTES.хTable 5 is available in its entirety in machine-readable form in the electronic edition of the Astrophysical Journal. Only emission lines measured in three example LBQS spectra are presented here ; see ° 4 for more details.

where only an upper limit on W could be measured, there j are no values quoted for the peak oset because we пxed the position of the line at its expected wavelength. The FWHM value also has no associated errors in such a case because it was пxed at an approximate median value for the sample. (See ° 3.5.) Note that some detected but poorly constrained lines may have an error estimate for the W measurement j even while the FWHM is пxed (see ° 3.5 and Table 4). Finally, column (5) gives the number of narrow absorption features used in modeling the emission lines. They are tabulated in the row of the closest emission line to the position of the absorption feature, e.g., see Lya (broad) and O I for LBQS 2354[0134 in Table 5. The format of the machinereadable version of Table 5 is explained in Appendix C. To generate a measure of the strength of iron emission in each spectrum, the integrated яux of the model template in the regions 2240х2655 A (blueward of Mg II) or 4434х 4684 A (blueward of Hb) is used and combined with the continuum speciпc яux at 2448 and 4559 A, respectively, to calculate W . For objects where these regions were not j present in the observed spectrum, we extrapolate the continuum to these positions. Note that we integrate the optical iron template яux over a wavelength range that matches the window used by BG92. We present in Table 6 the continuum parameters for the same LBQS spectra that appear in Table 5. The full table is available in the electronic edition of the Journal with one row for each QSO in the sample. The format of Table 6 is as follows. Columns (1), (2), and (3) give the QSO designation, redshift, and the Galactic neutral hydrogen column density (in units of 1020 cm~2), respectively. The latter measurements are taken from the Bell Laboratory H I survey (Stark et al. 1992). Columns (4)х(6) contain the measured values of the power-law continuum ; continuum slope, the rest frame wavelength at which the continuum is normalized, and the continuum яux at that wavelength (ergs cm~2 s~1 A~1), respectively, along with their 2 p uncertainties. The observed continuum яux f in ergs cm~2 s~1 A~1 at a j wavelength j is parameterized such that j ~! f \A , j jj c where A is the normalization of the power-law continua in units of j10~14 ergs cm~2 s~1 A~1 at a rest frame wave length j and ! is the continuum slope. Where two powerc law continua are used in the spectral model, the normalization of both continua are matched at j and the second continuum slope is tabulated in column 7 ocf Table 6. Note that in Table 6 the normalization units are 10~16,

AB

line parameter is quoted with positive and negative 2 p error estimates. The errors quoted for W are based on the j uncertainties in the amplitude and FWHM of the Gaussian model and do not include an error from an uncertainty in the underlying continuum яux level. For emission lines

TABLE 6 REPRESENTATIVE CONTINUUM MEASUREMENTS POWER-LAW CONTINUA DESIGNATION (1) 0025]0009 ...... 2244]0020 ...... 2354[0134 ...... REDSHIFT (2) 0.205 0.973 2.211 NGal H (3) 3.010 5.318 3.368 ! 1 (4) j c (5) A j (6) ! 2 (7) POLYNOMIAL ORDER (8)

3.99`1.37 4220 1.27`0.16 1.08`0.48 2 ~1.51 ~0.15 ~0.48 2.50`0.01 2170 15.4`0.9 ... 2 ~0.01 ~0.1 2.45`0.47 1463 1.19`0.04 ... 1 ~0.24 ~0.08 NOTES.хTable 6 is available in its entirety in machine-readable form in the electronic edition of the Astrophysical Journal. Only the continuum parameters from three example LBQS spectra are presented here ; see ° 4 for more details. The units for the normalization of the power-law continuum (A ) presented here are 10~16 ergs cm~2 s~1 A~1. In the j machine-readable version of this table the normalization factor is 10~14.

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY

43

while the units are 10~14 in the machine-readable table. Column (7) lists the slope of the second power-law continuum if used, the normalization of which is identical to the пrst power-law continuum. Column (8) presents the order of the polynomial continuum if required to model the blue region of the spectra. The format of the machine-readable version of Table 6 is explained in Appendix C.

We present in Figures 3a, 3b, and 3c the proпles and residuals from the models of the three spectra presented in Tables 5 and 6. The measurements presented here were chosen to show a range of spectral qualities present in the LBQS sample. Figure 3 includes a number of panels for each spectrum in each case all wavelengths are observed frame and яuxes are 10~14 ergs cm~2 s~1 A~1. The top

FIG.3a FIG. 3.хSpectral models of three QSOs from the LBQS sample. For each QSO, the top panel shows three sections. The continuum model plotted over the observed spectrum including error bars on each bin, the continuum ] iron emission template plotted over the spectrum, and the iron template proпle alone. Note that for clarity the яux scales are dierent in these sections. Smaller panels for each emission line component show the total best-пt model plotted over the relevant region of each spectrum, the residuals, and the individual Gaussian components (and the proпle of the iron template emission). Flux units are 10~14 ergs cm~2 s~1 A~1, wavelength units are in A and are observed frame values. Broad and narrow components of the same emission line species are marked "" b оо or "" n оо, respectively. See ° 4 for more details. (a) The spectral model of LBQS 0025 ] 0009. The spectral model of LBQS 2244 ] 0020. (c) The spectral model of LBQS 2354 [ 0134.

44

FORSTER ET AL.

Vol. 134

FIG.3b

panel has three sections, the upper section shows the continuum model overlying the spectrum, including error bars on each bin, the middle section displays the iron emission template plotted over the spectrum (no error bars) and the lower section shows template proпle alone. Note that the яux scales are dierent in these sections to display the template emission more clearly. This panel is followed by a number of separate panels, one for each emission line region. These emission line panels have three sections ; the top section shows the total continuum ] emission line model over the spectrum. (The model will include яux from the iron emission line template(s) if used. For clarity, яux errors on each bin are not shown in this section.) The middle section shows the residuals from the emission line model and include the error bars for each spectral bin. The lower section shows the proпles of the emission line models, with multiple components separated into single Gaussian proпles. The shape of the iron emission template used in the modeling of the spectrum is also shown in the lower panel. Dashed vertical lines indicate the expected position of emission lines based on tabulated redshifts. Absorption line proпles are shown along the top edge of the lower panels, e.g., the Lya region panel for LBQS 2354[0134 in Figure 3c. For the panels that show the emission line models of [Ne V], [O II], and [Ne III] the top sections also plot the local continua used for

these lines (see Appendix A). Note that to help visualize the quality of the emission line models the wavelength scale is identical for each of the emission line panels and within each panel the яux scale is identical in the sections displaying the total model, residuals, and individual model components.
5.

OVERALL STATISTICS

The analysis of the LBQS sample of QSO spectra has provided us with over 8000 emission line measurements, of which approximately 30% are upper limits to the equivalent width of low intensity lines. Without the correct inclusion of these "" censored оо data in the analysis of this complete sample, a biased estimation of the properties may occur and a number of important relationships may be masked or, even worse, appear more signiпcant than in reality. We use a nonparametric survival analysis technique to estimate the means and medians that characterize the emission line parameter distributions. The Kaplan-Meier (KM) estimator of a sample distribution is a maximum likelihood method that provides a reconstruction of information lost by censoring of data. This well-established statistical technique has been thoroughly examined for use in astronomical studies by Feigelson & Nelson (1985), Isobe, Feigelson, & Nelson (1986) and references therein. We recommend these papers to the reader.

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY

45

FIG.3c

The astronomical survival statistics package ASURV Rev. 1.1 (Isobe & Feigelson 1990 ; LaValley, Isobe, & Feigelson 1992) was used to provide the means, errors on the mean, and the medians of the W distributions for the LBQS j sample given in Table 7 and the distributions of W shown j in Figure 4. The format of Table 7 is as follows. Column (1) lists the emission line or line blend measured. The distribution for single Gaussian component models are tabulated separately from narrow and broad components. The distribution of sum of the W of the broad and narrow comj ponent models included with the single-component measurements is also tabulated. Columns (2)х(4) give the number of detected emission lines, their mean W and the standard deviation (SD) of the distribution of W . jColumns j (5) and (6) list the total number of emission lines measured for each species and the number of upper limits estimated for W . Columns (7) and (8) give the KM means, error on j the means, and medians of W for each line. Columns (9)х j (11) give the number, mean, and median of the FWHM of the Gaussian components used to model each emission feature. All equivalent widths are rest frame and the error on the W means is estimated from the 1 p of the KM j reconstructed distribution. Note that there are no upper limit measurements for two-component emission line пts, in

cases where any of the parameters of the two components could not be constrained to 2 p, a single Gaussian was used to model the line. We should add a note of caution here about the reconstructed distribution of emission line properties for samples that contain a large proportion of censored data (e.g., [Ne V], He II j4686). Formally, an emission line is detected if the amplitude of a feature can be constrained to within 2 p, based on the s2 conпdence interval bounds. The performance of the KM estimator degrades if the censored fraction is high or if the pattern of censored measurements is not random. We do not believe the latter is a concern here because the observing procedure used in the generation of the sample spectra produced spectra of similar continuum S/N, independent of the strength of emission features and the intrinsic luminosity of the QSO. In support of this conclusion consider the statistics of the O I emission line, where approximately half of the spectra with coverage in the region of this line possess detected emission features. Detected emission lines have 0.6 A [ W [ 14.0 A but the j range of upper limits for the undetected lines is 0.2 A [ W [ 15.0 A. The large proportion of censored data for j some emission lines lowers our conпdence in the KM estimate of the means and medians of a reconstructed W disj

46

FORSTER ET AL.
TABLE 7 EMISSION LINE PARAMETER DISTRIBUTIONS FOR THE LBQS SAMPLE W Detected EMISSION LINE (1) UV iron ........... Optical iron ....... Lyb ] O VI ....... Lyaa : Single ........... Narrow ......... Broad ........... Sumb ............... N V ................ O I ................. Si IV ] O IV ....... C IV a : Single ........... Narrow ......... Broad ........... Sumb ............... He II j1640 ....... Al III ............... C III ................ Mg II a : Single ........... Narrow ......... Broad ........... Sumb ............... [Ne V] ............. [O II] .............. [Ne III] ............ Hd ................. Hc ] [O III] ...... He II j4686 ....... Hb a : Single ........... Narrow ......... Broad ........... Sumb ............... [O III] j4959 ...... [O III] j5007 ...... He I ................ Num (2) 659 138 103 164 96 96 260 252 139 395 406 80 80 486 419 486 641 517 118 118 635 123 121 104 87 164 32 123 7 7 130 75 94 3 Mean (3) 40.2 39.0 11.7 56.4 28.1 68.3 71.2 18.8 3.1 13.0 38.1 17.8 44.8 42.1 18.7 8.7 28.4 39.1 28.1 36.6 43.9 5.8 7.8 7.5 12.3 18.3 21.0 62.4 34.3 101.6 66.4 13.8 30.4 10.7 SD (4) 23.0 21.5 10.1 28.8 21.3 31.6 40.4 10.7 2.5 8.8 19.3 9.3 23.1 23.1 13.5 6.0 14.9 21.4 12.9 21.5 25.1 5.8 14.6 5.0 8.6 11.1 20.1 36.0 13.6 58.9 41.0 14.0 39.0 3.8 Num (5) 953 247 130 164 96 96 260 260 260 414 408 80 80 488 488 667 667 559 118 118 677 488 393 363 309 251 187 141 7 7 148 146 146 33 j Kaplan-Meier Limits (6) 294 109 27 0 ... ... 0 8 121 19 2 ... ... 2 69 181 26 42 ... ... 42 365 272 259 222 87 155 18 ... ... 18 71 52 30 Mean (7) 29.9 ^ 0.8 23.8 ^ 1.6 9.5 ^ 0.9 56.4 28.1 68.3 71.2 18.3 2.0 12.6 38.0 17.8 44.8 42.0 16.4 6.8 27.6 37.1 28.1 36.6 42.0 2.1 3.2 3.4 3.8 14.6 4.2 57.4 34.3 101.6 61.2 8.5 21.6 4.7 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ 2.2 2.2 3.2 2.5 0.7 0.1 0.4 1.0 1.0 2.6 1.0 0.6 0.2 0.6 0.9 1.2 2.0 1.0 0.2 0.4 0.2 0.4 0.7 0.9 3.1 4.8 0.6 3.4 1.0 2.8 1.1 Median (8) 27.3 21.3 7.3 53.0 22.0 64.0 61.0 16.5 1.4 10.9 34.5 15.5 37.4 37.0 12.9 5.2 24.8 33.8 26.0 31.6 36.8 0.4 1.0 1.8 0.1 11.8 0.1 49.8 31.4 77.5 53.8 5.5 13.7 1.7 Num (9) 659 138 93 162 96 96 ... 228 119 383 403 80 80 ... 353 423 621 514 118 118 ... 113 94 92 81 160 27 119 7 7 ... 69 86 3 FWHM Detected Mean (10) 4610 ^ 130 4630 ^ 310 4870 ^ 280 7820 2650 10250 . 5580 2990 6780 7720 2860 10960 . 14670 5740 7820 5160 3510 8660 . 1570 900 1770 2860 2920 3690 4370 1160 6560 . 940 820 820 ^ ^ ^ .. ^ ^ ^ ^ ^ ^ .. ^ ^ ^ ^ ^ ^ .. ^ ^ ^ ^ ^ ^ ^ ^ ^ .. ^ ^ ^ 230 120 420 170 190 160 150 110 360 390 130 170 120 110 300 90 70 130 220 140 550 250 270 850 80 70 120 Median (11) 3470 2600 4410 7650 2470 9330 ... 5400 2460 5940 7300 2750 10530 ... 13590 5620 6670 4440 3370 8880 ... 1420 610 1380 2300 2500 2350 3760 900 5300 ... 770 590 750

NOTES.хCol. (1), emission line or line blend ; col. (2), number of detected emission lines ; col. (3), mean W of detected emission lines ; col. j (4), standard deviation (SD) of W measurements for detected emission lines ; col. (5), total number of measured emission lines ; (6) the number of upper limits estimates of W ;jcols. (7)х(8), the KM reconstructed mean and median of the W distributions (see ° 5) ; cols. (9)х(11), the j j number, mean, and median of the distribution of FWHM of the Gaussian components used to model each emission feature. All W are rest j frame measurements in A and the FWHM are in km s~1. a The distribution for single Gaussian component models are tabulated separately from narrow and broad components. b The distribution of sum of the broad and narrow component W included with the single-component measurements (see ° 5). j

tribution. Any conclusions based on these emission lines should rather use the results from the detected lines alone. The histograms shown in Figure 4 are taken from the KM estimation of the number of data points in each bin. The W (A) and FWHM (km s~1) distribution for each j emission line are shown in the upper and lower panels for each emission line, respectively. For emission lines that have been modeled using two Gaussian components, we include the distribution for the narrow and the broad components separately from the single-component lines. For completeness we also show in Figure 4 the distribution of the full sample of measured lines where the W of j

broad ] narrow components of an emission line have been summed and included in the single-component distribution (dashed line histograms).
6.

DISCUSSION

To examine the robustness of the analysis procedure in OUV spectra with improved S/N and spectral resolution we have modeled two composite QSO spectra using the techniques described above. The пrst was constructed by Francis et al. (1991, hereafter F91) from 718 spectra in the LBQS sample itself, and the second from HST FOS observations of 101 QSOs presented in Zheng et al. (1997, here-

FIG. 4.хKM estimated distribution of the emission line properties of the LBQS sample. The upper panels are rest frame W (A) and the lower panels are j the FWHM (km s~1) of the Gaussian proпle used to model the emission lines (see °° 3.2 and 4 for an explanation of the meaning of the iron emission W and j FWHM). For emission lines that may be modeled with two Gaussian proпles, the narrow component distributions are marked with "" n оо and broad components with "" b оо. The distribution of the sum of the narrow ] broad component W is also shown as the dashed line in the panels for the j single-component models.

48

FORSTER ET AL.

Vol. 134

FIG. 4.хContinued

after Z97). We пnd that even where the limit of two Gaussian components for an emission line does not reproduce the line proпle very well, the measured W is within j 10% of values tabulated in the above papers. The exception is for Mg II, where the F91 measurement is higher and the Z97 measurement is lower than that measured using the techniques presented above. Both discrepancies can be attributed to the presence of iron emission features blending into the wings of the Mg II emission line. Although F91 пtted a global continuum and corrected for iron emission using local spline пts to emission complexes, the W was j measured as the total яux above the continuum and iron emission across more than 50 A in the spectrum and so may include some hidden iron emission. This is also likely to be the explanation of a similar discrepancy in the measurements of Hc ] [O III] j4363 and [O III] j4959 in the F91 composite. The measurements in Z97 use a local continuum which is difficult to determine accurately in regions where iron emission complexes are strong, and may result in their lower W measurement. This conпrms that correct accountj ing for iron emission is crucial in the measurement of emission lines in QSO spectra. We can also compare the global properties of the LBQS emission line measurements with the measurements of these composite QSO spectra. The Z97 composite shows larger W in Lyb ] O VI, Lya, and C IV. The discrepancy remains j even if the means of only our detected emission lines are compared. (The inclusion of the censored data will naturally lower the means of a distribution, but it has little eect on the distribution of strong lines like Lya and C IV which are well constrained in nearly all cases). We suspect that the dierence is attributable to the large (60%) fraction of radio-loud objects in the heterogeneous sample used to con-

struct the Z97 composite. The LBQS sample is predominantly radio quiet (RQ) with D10% radio loud (RL) (Hooper et al. 1995), and Z97 showed that there are signiпcant dierences in the strength of these lines between RL and RQ populations, independent of luminosity. A marginally signiпcant RL/RQ dierence was also seen in a sample of 255 of the optically brightest QSOs in the LBQS sample (Francis, Hooper, & Impey 1993). A more detailed examination of this eect will be made in a subsequent paper that will contrast the continuum and emission line properties within the full LBQS sample. In contrast, the W of the emission features in the F91 composite are lowerj than the mean and medians of the measured emission lines in the LBQS sample. Although the F91 composite was created using a large proportion of the same spectra measured here, it is not surprising that the measurements should dier. Lya, which shows the largest dispersion of relative яux within the sample (see Fig. 5 in F91), is most aected by narrow absorption features. Allowance for the eects of narrow absorption increases the W j modeled with Gaussian components relative to an estimate based on the integration of яux above a continuum. For [Ne III] and Hd, the F91 composite indicates a stronger feature than suggested by the KM mean presented in Table 7. However, the measurements are similar when only the detected lines are studied. There has been no investigation of inclusion of censored data on the emission line and continuum correlations, particularly the Baldwin eect (Baldwin 1977 ; Zamorani et al. 1992 ; Osmer, Porter, & Green 1994) and the correlations associated with the primary eigenvector of spectral variance in AGNs (BG92 ; M96 ; Sulentic et al. 2000). Previous studies have focussed on a small number of bright emission

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY
TABLE 8 EMISSION LINE PARAMETER DISTRIBUTIONS FOR PUBLISHED COMPARISON SAMPLES W EMISSION LINE (1) UV iron ........... Optical iron ....... REF. (2) 2 1 2 3Q 3L 4Q 4L 3Q 3L 3Q 3L 2 2 2 3Q 3L 3Q 3L 1 2 1 3Q 3L 4Q 4L 2 2 3Q 3L 1 2 3Q 3L 4Q 4L Num. (3) 38 87 45 14 17 80 45 11 6 13 13 39 52 44 10 13 7 6 70 32 87 14 17 80 45 42 52 ... ... 83 49 15 17 80 45 Limits (4) 8 7 5 . . . . . . . . 0 0 3 . . . . 0 5 0 . . . . 0 0 . . 0 0 . . . . j Mean (5) 28.1 47.5 178.5 25.6 34.4 52 21.5 72.2 99.2 22.6 20.0 11.2 107.5 25.3 17.9 17.9 23.1 29.1 11.2 16.8 95.5 64.8 73.2 99 84 4.9 105.3 . . 24.8 58.8 20.6 10.2 23.5 23.0 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ .. .. ^ ^ ^ ^ ^ ^ 4.8 2.8 6.2 4.9 4.2 2.8 2.5 9.0 9.0 3.4 2.5 1.4 9.4 2.7 1.6 0.7 2.3 4.1 1.1 2.4 4.0 6.9 5.1 4.2 5.2 0.6 8.4 a Median (6) 21.5 43.8 110.9 ... ... ... ... ... ... ... ... 8.7 96.0 23.4 ... ... ... ... 7.5 13.5 92.5 ... ... ... ... 4.0 107.0 ... ... 16.8 23.4 ... ... ... ... Num. (7) ... ... ... ... ... ... ... ... ... 15 15 39 51 24 12 14 ... ... ... 14 87 14 17 ... ... 38 51 14 17 ... 49 14 15 ... ... FWHM Mean (8) ... ... ... ... ... ... ... ... ... 4880 ^ 6020 ^ 1560 ^ 6990 ^ 12940 ^ 6430 ^ 8420 ^ ... ... ... 7480 ^ 3790 ^ 4430 ^ 5100 ^ ... ... 630 ^ 5950 ^ 9870 ^ 11890 ^ ... 700 ^ 1160 ^ 1150 ^ ... ... Median (9) ... ... ... ... ... ... ... ... ... ... ... 1560 6610 12400 ... ... ... ... ... 7900 3160 ... ... ... ... 590 4780 ... ... ... 610 ... ... ... ...

49

Lya ................ C IV ................

. . . . . . . .

. . . . . . . .

b b

He II j1640 ....... C III j1909 ........ Mg II j2800 ....... He II j4686 ....... Hb .................

. . . .

. . . .

460 670 80 310 600 730 620

. . . .

. . . .

770 220 590 470

b b

.. ..

40 550 970 600 40 90 130

[O III] j5007 ......

. . . .

. . . .

2.7 6.8 3.5 1.9 2.9b 2.2b

NOTES.хCol. (1), emission line or line blend. Col. (2) reference. Col. (3), the total number of measured emission lines. Col. (4), number of upper limits on W . Cols. (5)х(6), the mean and median of the W distributions. Where upper limits are j j present the KM estimator was used (see ° 5). Cols. (7)х(9), the number, mean and median of the distribution of published FWHM. All W are rest frame measurements in A and the FWHM are in km s~1. j a This measurement is not suitable for direct comparison with the mean from the LBQS sample (Table 8) because it is measured in a dierent region of the spectrum (see ° 6). b Errors on mean calculated from p/JN. REFERENCESх(1) Boroson & Green 1992 ; (2) Marziani et al. 1996 ; (3) McIntosh et al. 1999 ; (4) Sulentic et al. (2000) ; Q \ radio quiet, L \ radio loud.

features ; particularly C IV j1549, He II j4686, Hb, O III j5007, and iron emission. We brieяy summarize the published measurements for QSO samples of BG92 (87 low-z QSOs, optical spectra), McIntosh et al. (1999) (32 high-z QSOs with IR spectra), M96 (52 QSOs with optical and UV spectra) and Sulentic et al. (2000) (125 QSOs) in Table 8. Where upper limits are quoted, we have used survival analysis to enable a more direct comparison to our results in Table 7. Some of the published work has also been presented separately as RL or RQ subsamples. While all these studies suer from diering selection eects, the comparison samples also dier from the LBQS sample in redshift range and magnitude limits. The most notable dierence between the properties of the LBQS and the measurements shown in Table 8 can be seen

in the relative strength of optical iron emission, the mean value of which appears signiпcantly higher in BG92 and MS96. The region used to measure the W of optical iron emission is identical to that used in BG92 j (and subsequent authors) and a test using the methods described in ° 3 to measure the BG92 sample spectra (kindly supplied by T. Boroson) shows reasonable (D20%) agreement in Fe II W . j The measurement of UV iron emission in M96 appears similar to the mean of the distribution of iron emission strength measured in the LBQS sample. However, the window used by M96 to characterize the iron emission strength was 1550х1750 A and the iron template яux in this region is approximately 6.5% of the яux in the region used for the study of the LBQS sample presented here (i.e., the mean UV iron emission in M96 is 15 times higher than the

50

FORSTER ET AL.

Vol. 134

mean in the LBQS sample !). The unusually strong iron emission was noted by M96 as a possible bias in their sample. The heterogeneous sample (52 AGNs for which HST FOS spectra covering C IV were available in 1994 and for which the authors had matching optical spectra) has an overabundance of strong iron emitters relative to the BG92 sample. The dierence in iron emission distributions contrasts to the similarity seen in the distributions of W for j [O III] j5007. The strongest correlations reported for optical spectra of AGNs involve iron emission, Hb, and [O III]. We expect that our new, more homogeneous emission line measurements, of iron in particular, may signiпcantly reshape the discussion of correlations that produce the largest variance in PCA analysis of quasar emission lines.
7

. SUMMARY

es, narrow absorption lines, and continuum breaks, and we include upper limits on the strength of low-intensity emission lines. The measurements are presented with uncertainties generated in an objective manner and with a rigorous inclusion of censored data in tabulated emission line parameter distributions. A more exhaustive analysis of the relationships between the emission line and continuum properties of the LBQS sample will be presented in a later paper. High-quality, reproducible measurements of emission lines for uniformly selected samples will advance our understanding of active galaxies, especially in a new era of large surveys selected by a variety of complementary methods, such as the Sloan Digital Sky Survey (SDSS ; York et al. 2000), the FIRST Bright Quasar Survey (White et al. 2000), or the Chandra Multiwavelength Project (Green et al. 1999). The authors gratefully acknowledge support provided NASA through grant NAG5-6410 (LTSA). C. B. acknowledges the support of NSF grant AST 98-03072. F. thanks Matt Malkan and the UCLA Division Astronomy and Astrophysics for their hospitality. by F. K. of

This paper makes available measurements from the largest database of OUV spectra of QSOs to date. Our analysis of 993 spectra of similar quality from the LBQS yields over 8000 measurements of the 20 most prominent emission lines between rest frame 1025х5900 A. We have accounted for the eects of blended iron emission complex-

APPENDIX A NOTES ON SPECIFIC EMISSION LINES Lyb ] O VI j1035.хA яat "" pseudo оо continuum and a single Gaussian emission feature were used to model the spectrum near this emission line blend. The resulting equivalent widths should be viewed as approximate due to the nature of the QSO spectra in this region. Only in one spectrum of the sample was the Lyb and O VI emission lines clearly distinct, but the spectrum was modeled with a single Gaussian for consistency. He II j1640.хThe single Gaussian component used to model the region on the red side of C IV j1549 will account for emission from He II j1640, [Ne V] jj1575,1593, [Ne IV] jj1602,1609, Si II j1650, [O III] jj1661,1663,1668, and Al II j1670. The use of multiple Gaussian components in all but a few spectra did not improve the model пt to this region due to the quality of the LBQS sample. The tabulated values for this emission line component, particularly the FWHM, will be much larger than expected for just the He II line. [Ne V] j3426, [O II] j3728, [Ne III] j3869.хThe global continuum model was not successful for the region near these lines due to the blend of Balmer emission lines from transitions with m є 7 in the low-resolution spectra and so a local power-law continuum was created. This was measured from windows 30 A wide at least 30 A from the expected line position. The emission lines are measured above this local continuum. He II j4686.хFor spectra where two Gaussian were used to model the Hb emission line, the Hb broad component was included in the model пt of the He II j4686 emission line. APPENDIX B INDIVIDUAL OBJECTS LBQS 1206]1052.хThis QSO was assigned a redshift of 0.402 ^ 0.005 (Hewett et al. 1995) based on the cross correlation with the Francis et al. (1991) LBQS composite spectrum. Although all the features in the spectrum will contribute to the redshift estimate the stronger features will contribute more than the weak so it is unclear as to the cause of the discrepancy as the strongest lines of [O III] jj4959,5007 are clearly not at the tabulated redshift. Only the measurement of the Mg II emission line agrees with this redshift, this is likely due to the presence of an absorption feature on the blue side of Mg II that is not modeled in the automated пtting procedure used here. Other lines in the spectrum all show a lower redshift. The value adopted here is z \ 0.396 ^ 0.003 based on the median of 10 emission measurements and the 1 p value of the distribution of z. LBQS 0023]0228.хThe spectrum of this QSO shows very strong and narrow (FWHM ^ instrumental resolution) forbidden emission lines, no evidence of broad Hb emission and only a weak яat continuum. The strength of [O II] (W \ 148`8 A) and [O III] (W \ 258`12 A) in the spectrum of LBQS 0023]0228 is much higher than all the other QSOs ~11 j ~14 in j LBQS except for LBQS 0004]0224, which shows a rising blue continuum and strong broad emission from Mg II and the Hb as well as having forbidden line FWHM Z 1200 km s~1. The spectrum of LBQS 0023]0228 resembles that of a starburst galaxy (e.g., M82) rather than a QSO. The presence of these two strong narrow-line objects in the sample does not aect the parameter distributions due to the large numbers of objects in each sample.

No. 1, 2001

LARGE BRIGHT QUASAR SURVEY APPENDIX C ELECTRONIC TABLES

51

The machine-readable version of Table 5 has an identical format to that published here but includes one row for each emission line that is listed in Table 3 even if the line lies outside the wavelength range of the observed spectrum. The machine-readable table contains zeros for unmeasured emission line parameters, rather than the dots (. . .) present in the sample. This will aid in the use of this large machine-readable ASCII table. The пrst line of the machine-readable table contains shortened column headings. This is followed by 31 lines for each QSO (a total of 30753 lines). The пrst line for each object tabulates the QSO designation and redshift and has a format (a10, f6.3). This is followed by 30 rows with a format (a16, 6i7, 3f9.2, i3), note that the positive and negative error estimates appear in separate columns. Where only an upper limit for W j of an expected emission line is measured, the value is given in the positive error column of W . The FWHM for these lines is j given with zeros in the error columns (the position oset is by deпnition 0 km s~1). Note that there are emission lines where the FWHM and position oset could not be constrained but the W is constrained to 2 p. j The format of the machine-readable version of Table 6 is identical to that presented here and as with Table 5, the positive and negative errors have separate columns. The пrst line of the electronic table contains the column headings and is followed by a single row for each object with the ASCII format (a10, 2f6.3, 3f7.2, i6, 3f11.5, 3f7.2, i3). The normalization units for the power-law continuum (cols. [8], [9], and [10] in the machine-readable table) are 10~14 ergs cm~2 s~1 A~1. We stress that in the machine-readable versions of Tables 5 and 6 zeros appear in all blank spaces that would normally appear blank.
REFERENCES Aldcroft, T. A. 1993, Ph.D. thesis, Stanford Univ. LaValley, M., Isobe, T., & Feigelson, E. D. 1992, in ASP Conf. Ser. 25, Aoki, K., & Yoshida, M. 1999, in ASP Conf. Ser. 162, Quasars and CosAstronomical Data Analysis Software and Systems I, ed. D. M. Worrall, mology ed. G. Ferland, & J. Baldwin (San Francisco : ASP), 385 C. Biemesderfer, & J. Barnes (San Francisco : ASP), 245 Baldwin, J. A. 1977, ApJ, 214, 679 Lawrence, A., Elvis, M., Wilkes, B. J., McHardy, I., & Brandt, N. 1997, Baldwin, J. A., Ferland, G., Korista, K., & Verner, D. 1995, ApJ, 455, 119 MNRAS, 285, 879 Bevington, P. R., & Robinson, D. K. 1992, Data Reduction and Error Lipari, S., Terlevich, R., & Macchetto, F. 1993, ApJ, 406, 451 Analysis for the Physical Sciences (2d ed. ; New York : McGraw-Hill) Marziani, P., Sulentic, J. W., Dultzin-Hacyan, D., Calvani, M., & Moles, Binette, L., Prieto, A., Szusziewicz, E., & Zheng, W. 1989, ApJ, 343, 135 M. 1996, ApJS, 104, 37 (M96) Bohlin, R. C., Savage, B. D., & Drake, J. F. 1978, ApJ, 244, 132 Mathur, S., Wilkes, B., & Elvis, M. 1998, ApJ, 503, L23 Boroson, T. A., & Green, R. F. 1992, ApJS, 80, 109 (BG92) McIntosh, D. H., Rieke, M. J., Rix, H.-W., Foltz, C. B., & Weymann, R. J. Brandt, W. N., Laor, A., & Wills, B. J. 2000, ApJ, 528, 637 1999, ApJ, 514, 40 Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245 Morris, S. L., Weymann, R. J., Anderson, S. F., Hewett, P. C., Francis, P. J., Chaee, F. H., Foltz, C. B., Hewett, P. C., Francis, P. A., Weymann, R. J., Foltz, C. B., Chaee, F. H., & MacAlpine, G. M. 1991, AJ, 102, 1627 Morris, S. L., Anderson, S. F., & MacAlpine, G. M. 1991, AJ, 102, 461 (LBQS Paper V) (LBQS Paper IV) Mortlock, D. J., Webster, R. L., & Francis, P. J. 1999, MNRAS, 309, 836 Collin, S., & Joly, M. 2000, New Astron. Rev., 44, 531 Mushotzky, R. F., & Ferland, G. J. 1984, ApJ, 278, 558 Corbin, M. R. 1993, ApJ, 403, L9 Netzer, H. 1980, ApJ, 236, 406 Feigelson, E. D., & Nelson, P. J. 1985, ApJ, 293, 192 Osmer, P. S., Porter, A. C., & Green, R. F. 1994, ApJ, 436, 678 Ferland, G. F., & Shields, G. A. 1985, in Astrophysics of Active Galaxies Peng, C. Y., Impey, C. D., Falco, E. E., Kochanek, C. S., Lehar, J., McLeod, and Quasi-Stellar Objects, ed. J. Miller (Mill Valley : University Science B. A., Rix, H.-W., & Keeton, C. R. 1999, ApJ, 524, 572 Books), 57 Peterson, B. M. 1997, An Introduction to Active Galactic Nuclei Foltz, C. B., Chaee Jr., F. H., Hewett, P. C., MacAlpine, G. M., Turnshek, (Cambridge : Cambridge Univ. Press) D. A., Weymann, R. J., & Anderson, S. F. 1987, AJ, 94, 1423 (LBQS Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. 1992, Paper I) Numerical Recipes in FORTRAN : The Art of Scientiпc Computing (2d Foltz, C. B., Chaee, F. H., Hewett, P. C., Weymann, R. J., Anderson, Scott ed. ; Cambridge : Cambridge Univ. Press) F., & MacAlpine, G. M. 1989, AJ, 98, 1959 (LBQS Paper II) Schmidt, M., & Green, R. F. 1983, ApJ, 269, 325 Francis, P. J., Hewett, P. C., Foltz, C. B., Chaee, F. H., Weymann, R. J., Stark, A. A., Gammie, C. F., Wilson, R. W., Bally, J., Linke, R. A., Heiles, Morris, S. L. 1991, ApJ, 373, 465 (F91) C., & Hurwitz, M. 1992, ApJS, 79, 77 Francis, P. J., Hooper, E. J., & Impey, C. D. 1993, AJ, 106, 417 Stocke, J. T., Morris, S. L., Weymann, R. J., & Foltz, C. B. 1992, ApJ, 396, Gallagher, S. C., Brandt, W. N., Sambruna, R. M., Mathur, S., & Yama487 saki, N. 1999, ApJ, 519, 549 Sulentic, J. W., Zwitter, T., Marziani, P., & Dultzin-Hacyan, D. 2000, ApJ, Gehrels, N. 1986, ApJ, 303, 336 536, L5 Green, P. J. 1996, ApJ, 467, 61 Verner, E. M., Verner, D. A., Korista, K. T., Ferguson, J. W., Hamann, F., Green, P. J., & Mathur, S. 1996, ApJ, 462, 637 & Ferland, G. J. 1999, ApJS, 120, 101 Green, P. J., et al. 1995, ApJ, 450, 51 Veron-Cetty, M.-P., & Veron, P. 1998, A Catalogue of Quasars and Active Green, P. J., et al. 1999, BAAS, 195, 8008 Nuclei (8th ed. ; Garching : ESO) Hewett, P. C., Foltz, C. B., & Chaee, F. H. 1995, AJ, 109, 1498 (LBQS Vestergaard, M., & Wilkes, B. J. 2000, ApJ, submitted Paper VI) Weymann, R. J., Morris, S. L., Foltz, C. B., & Hewett, P. C. 1991, ApJ, 373, Hewett, P. C., Foltz, C. B., Chaee, F. H., Francis, P. J., Weymann, R. J., 23 Morris, S. L., Anderson, S. F., MacAlpine, G. M. 1991, AJ, 101, 1121 White R. L., et al. 2000, ApJS, 126, 133 (LBQS Paper III) Wisotzki, L., Christlieb, N., Bade, N., Beckmann, V., Kohler, T., Vanelle, Hooper, E., Impey, C. D., Foltz, C., & Hewett, P. 1995, ApJ, 445, 62 C., & Reimers, D. 2000, A&A, 358, 77 Isobe, T., & Feigelson, E. D. 1990, BAAS, 22, 917 York, D. G., et al. 2000, AJ, 120, 1579 Isobe, T., Feigelson, E. D., & Nelson, P. J. 1986, ApJ, 306, 490 Zamorani, G., Marano, B., Mignoli, M., Zitelli, V., & Boyle, B. J. 1992, Korista, K., Baldwin, J., & Ferland, G. 1998, ApJ, 507, 24 MNRAS, 256, 238 Krolik, J. H., & Kallman, T. R. 1988, ApJ, 324, 714 Zheng, W., Kriss, G. A., Telfer, R. C., Grimes, J. P., & Davidsen, A. F. 1997, Kuraszkiewicz, J., Wilkes, B. J., Czerny, B., & Mathur, S. 2000, ApJ, in ApJ, 475, 469 (Z97) press Laor, A., Fiore, F., Elvis, M., Wilkes, B. J., McDowell, J. C. 1997, ApJ, 477, L93