Near Infrared Spectroscopy of High Redshift Active Galactic Nuclei. II. Disappearing Narrow Line Regions and the Role of Accretion

Abstract

We present new near infrared spectroscopic measurements for 29 luminous
high-z quasars and use the data to discuss the size and other properties of the
NLRs in those sources. The high resolution spectra have been used to carefully
model the Fe II blends and to provide reliable [O III], Fe II and Hb
measurements. We find that about 2/3 of all high luminosity sources show strong
[O III] lines while the remaining objects show no or very weak such line. While
weak [O III] emitters are also found among lower luminosity AGN, we argue that
the implications for very high luminosity objects are different. In particular,
we suggest that the averaging of these two populations in other works gave rise
to claims of a Baldwin relationship in [O III] which is not confirmed by our
data. We also argue that earlier proposed relations of the type R_NLR \propto
L_[O III]^{1/2}, where R_NLR is the NLR radius, are theoretically sound yet
they must break down for R_NLR exceeding a few kpc. This suggests that the NLR
properties in luminous sources are different from those observed in nearby AGN.
In particular, we suggest that some sources lost their very large, dynamically
unbound NLR while others are in a phase of violent star-forming events that
produce a large quantity of high density gas in the central kpc. This gas is
ionized and excited by the central radiation source and its spectroscopic
properties may be different from those observed in nearby, lower luminosity
NLRs. We also discuss the dependence of EW(Hb) and Fe II/Hb on L, M_BH, and
accretion rate for a large sample of AGNs. The strongest dependence of the two
quantities is on the accretion rate and the Fe II/Hb correlation is probably
due to the EW(Hb) dependence on accretion rate. We show the most extreme values
measured so far of Fe II/Hb and address its correlation with EW([O III]).

Full text

Page 1

arXiv:astro-ph/0406560v1 24 Jun 2004
THE ASTROPHYSICAL JOURNAL, 614:???–???, 2004 OCTOBER 10, ASTRO-PH/0406560
Preprint typeset using LATEX style emulateapj v. 11/26/03
NEAR INFRARED SPECTROSCOPY OF HIGH REDSHIFT ACTIVE GALACTIC NUCLEI.
II. DISAPPEARING NARROW LINE REGIONS AND THE ROLE OF ACCRETION
H. NETZER,1O. SHEMMER,1R. MAIOLINO,2E. OLIVA,3S. CROOM,4E. CORBETT,4AND L. DI FABRIZIO3
Received 2004 March 28; accepted 2004 June 22
ABSTRACT
We present new near infrared spectroscopic measurements for 29 luminous high-redshift active galactic nuclei
(AGNs) and use the data to discuss the size and other properties of the narrow line regions (NLRs) in those
sources. The high resolution spectra have been used to carefully model the Fe II blends and to provide reli-
able [O III]λ5007, Fe II and Hβ measurements. We find that about 2/3 of all very high luminosity sources
show strong [O III]λ5007 lines while the remaining objects show no or very weak such line. While weak
[O III]λ5007 emitters are also found among lower luminosity AGNs, we argue that the implications for very
high luminosity objects are different. In particular, we suggest that the averaging of these two populations in
other works gave rise to claims of a Baldwin relationship in [O III]λ5007 which is not confirmed by our data.
We also argue that earlier proposed relations of the type RNLR∝ L1/2
are theoretically sound yet they must break down for RNLRexceeding a few kpc. This suggests that the NLR
properties in high luminosity sources are very different from those observed in nearby AGNs. In particular,
we suggest that some sources lost their very large, dynamically unbound NLR while others are in a phase of
violent star-forming events that produce a large quantity of high density gas in the central kpc. This gas is
ionized and excited by the central radiation source and its spectroscopic properties may be different from those
observed in nearby, lower luminosity NLRs. We also discuss the dependence of EW(Hβ) and Fe II/Hβ on
luminosity, black hole mass, and accretion rate for a large sample of AGNs. The strongest dependence of the
two quantities is on the accretion rate and the Fe II/Hβ correlation is probably due to the EW(Hβ) dependence
on accretion rate. We show the most extreme values measured so far of Fe II/Hβ and address its correlation
with EW([O III]λ5007).
Subject headings: galaxies: active – galaxies: nuclei – galaxies: Seyfert – quasars: emission lines – galaxies:
starburst
[O III], where RNLRis the radius of the NLR,
1. INTRODUCTION
The narrow line regions (NLRs) of active galactic nuclei
(AGNs) have been studied, extensively, from the ground and
from space. This component of the nucleus is spatially re-
solved from the ground in nearby sources and HST observa-
tions extend the range to a redshift of about 0.5. Thus, de-
tailed NLR mappings are now available for a large number of
sourcescoveringa largerangeofluminosityandredshift(Fal-
cke, Wilson, & Simpson 1998; Bennert et al. 2002, hereafter
B02, Schmitt et al. 2003).
The spectroscopic characteristics of the NLR gas have been
studied, extensively, over several decades and high quality
data are now available (e.g., Veilleux & Osterbrock 1987).
The main source of excitation of the NLR gas is photoion-
ization by the the central continuum (see review and refer-
ences in Netzer 1990) but shock excitation must be important
in some parts of this region, most notably is NLRs that are as-
sociated with jet-like radio structures (Schiano 1986; Dopita
et al. 2002 and references therein). The gas dynamics has
been studied too with detailed results concerning the profiles
of various emission lines and their dependenceon the level of
1School of Physics and Astronomy and the Wise Observatory, The Ray-
mond and Beverly Sackler Faculty of Exact Sciences, Tel-Aviv University,
Tel-Aviv 69978, Israel; netzer@wise.tau.ac.il
2INAF - Osservatorio Astrofisico di Arcetri, Largo E. Fermi 5, I-50125
Firenze, Italy; maiolino@arcetri.astro.it
3Istituto Nazionale di Astrofisica, Centro Galileo Galilei, and Telescopio
Nazionale Galileo, P.O. Box 565, E-38700 Santa Cruz de la Palma, Spain;
oliva@tng.iac.es
4Anglo-Australian Observatory, PO Box 296, Epping, NSW 1710, Aus-
tralia; scroom@aaoepp.aao.gov.au
ionization, the density and the dust content of the gas (e.g.,
Veilleux 1991; Nelson & Whittle 1996; Barth et al., 2001).
A major emphasis in recent years has been the extension of
such works to higher luminosity sources. Some such studies
suggest that the [O III]λ5007 line width is correlated with the
stellar velocity distribution in the bulge and thus also with the
mass of the central black hole (Nelson, 2000; Shields et al.,
2003).
Several recent attempts to study NLRs in large samples
of AGNs lead to apparently conflicting results.
tained narrow band HST images of seven luminous radio-
quiet Palomar-Green(PG) quasars with z <0.5. They argued,
onthebasis ofcomparisonwith nearbyless luminoussources,
that the NLR size (radius) scales with the [O III]λ5007 and
the Hβ line luminosities roughly as RNLR∝ L0.5. The mea-
sured NLR sizes in their most luminous sources approached
10 kpc (throughout this work we assume H0= 70 km s−1
Mpc−1, Ωm= 0.3, and ΩΛ= 0.7). This dependence has been
questioned by Schmitt et al. (2003) who studied a much
larger sample (22 Seyfert 1’s and 38 Seyfert 2’s), albeit with
much lower luminosity, and found RNLR∝ L0.33. Croom et
al. (2002) analyzed the spectra of ∼ 22,000 AGNs from the
2dF quasar redshift survey (2QZ) and claimed to see a de-
crease in the equivalent width (EW) of several narrow lines
([O II]λ3727, [Ne V]λ3426, and [Ne III]λ3870) with source
luminosity. They suggested that at least part of this “Bald-
win effect” (Baldwin 1977) is due to the increase in NLR size
with source luminosity which leads to galactic-scale dimen-
sions in the most luminous objects. Such NLRs are likely
to escape the system leading to AGNs with weak or no NLR
B02 ob-

Page 2

2 NETZER ET AL.
emission. Testing this idea for the most intense narrow line,
[O III]λ5007, was limited by the low redshift, and hence rela-
tivelylow luminosityin theCroomet al. (2002)ground-based
sample.
This paper addresses the issues of “the disappearingNLRs”
and the Fe II/Hβ ratio in high luminosity AGNs. The work
complements the Shemmer et al. (2004; hereafter Paper I)
study and is based on the same data set. The paper is arranged
as follows: § 2 presents the new observations, § 3 shows var-
ious correlations involving the [O III]λ5007, Hβ, and Fe II
lines and § 4 discusses the implications regardingthe size and
the physics of the NLR as well as the Fe II/Hβ ratio in the
most luminous AGNs.
2. OBSERVATIONS
We obtained new near infrared (IR) spectroscopic observa-
tions for a sample of 29 high-redshift, high-luminosity AGNs
at the Anglo-Australian Telescope (AAT) in Australia and at
Telescopio Nazionale Galileo (TNG) in Spain. The observa-
tions and data reduction are described in Paper I where all IR
spectra are also shown (their Figures 1–3).
In this work we focus on the [O III]λ5007 emission line
that, when observed, is the strongest narrow line in the spec-
trum (all numbers given in this paper refer to the 5007Å
component only). We also show and discuss various corre-
lations involving the broad Fe II lines. The measurement of
the [O III]λ5007 line is complicated due to the presence of
strong, broad Fe II blends in this part of the spectrum. The ac-
curate modeling of these blends is crucial to our study of the
Fe IIspectrumas well as themeasurementofthe[O III]λ5007
line. Below we give a detailed description of this process and
illustrate the results for the case of [HB89] 1346−036.
The near-IR spectrum of the quasar [HB89] 1346−036
(z=2.370) was observed by McIntosh et al. (1999). The same
observation was later used by Yuan & Wills (2003) who re-
measured the McIntosh et al. (1999) spectrum and used it
in their study of the Eddington ratio in z ∼ 2 quasars. Both
McIntosh et al. (1999) and Yuan & Wills (2003) found an
[O III]λ5007 line in this source, which was later used in sev-
eral of their correlations. Our superior, high S/N, better res-
olution IR spectroscopy of the source shows broad lines due
to Hβ and Fe II and a weak emission feature very close to
the putative [O III]λ5007 position (Fig. 1 top curve). We
have used the empirical Fe II emission template of Boroson
& Green (1992; hereafter BG92), scaled to the intensity of
the strongest iron features in order to removethose Fe II lines.
The broadenedtemplate is shown in the diagram and the Fe II
subtracted spectrum is given below the original spectrum. As
seen in the diagram, the process completely removed all trace
of [O III]λ5007 emission. We applied a similar procedure to
the spectra of all other sources and the results listed below
are all corrected for the Fe II blends. We suspect that other
sources in the McIntosh et al. (1999) and the Yuan and Wills
(2003) samples suffer from a similar problem and hence de-
cided not to include these objects in our analysis. We note
that Sulentic et al. (2004) also question some of the McIntosh
et al. (1999) measurements.
Table 1 gives a summary of the data used in this paper. It
includes the object name (column 1), systemic redshift (col-
umn 2), continuum luminosity defined as λLλat rest wave-
length 5100Å (column 3) and the basic [O III]λ5007 line
measurements (rest-frame EW in columns 4 & 5, luminos-
ity in columns 6 & 7, and FWHM in columns 8 & 9). Table 1
also gives the best-fit Hβ luminosity in column 10, and the Fe
400045005000550060006500
Rest-Frame Wavelength [Å]
0
0
2
2
4
4
6
6
8
8
10
10
12
12
14
14
16
16
18
18
20
20
22
22
Fλ [10-17 erg s-1 cm-2 Å-1]
1.5 1.6
Observed-Frame Wavelength [µm]
1.71.81.92.02.12.22.31.4
Fλ [10-17 erg s-1 cm-2 Å-1]
broadened Fe II template
Fe II subtracted spectrum
spectrum
Hα broad
Hβ broad
Fe II λλ 4924,5018
no [O III]
[HB89] 1346-036
FIG. 1.— An example of the [O III]λ5007 line measurement process for the
z=2.370 quasar [HB89] 1346−036 we observed at TNGin 2002. The diagram
shows the reduced calibrated spectrum with emission lines resembling the
[O III]λ5007 doublet (top curve) and the Fe II subtracted spectrum (middle
curve) where no sign of [O III]λ5007 is seen. The Fe II template is also
shown (bottom curve).
II/Hβ fluxratio5in column11. Regardingtheuncertaintieson
those numbers, some of those are discussed in Paper I and the
others, related to the [O III]λ5007 line, were obtained using
the procedure explained in Paper I as applied to this line.
The main result, which is apparent in Table 1, is that the
population of high-redshift, high-luminosity quasars is di-
vided into two distinct groups. One group (22 sources) con-
tains objects with strong [O III]λ5007 lines (EW∼ 10−80Å)
and with I([O III]λ5007)/I(broad Hβ) similar to the ratio ob-
served in many low luminosity type-I AGNs. The second
group (seven sources) shows no [O III]λ5007 line within
the observational uncertainty. To obtain the upper limits on
EW([O III]λ5007) in those sources, we assumed a “typical”
[O III]λ5007 line with FWHM of 1000 km s−1(aboutthe me-
dian in our sample, see below) and looked for the weakest
such feature that would have been detected in our spectra af-
ter the removal of the Fe II blends. For the best S/N spectra
(three sources), this translates to an EW which is approxi-
mately 0.05×EW(Hβ). For the other four cases the upper
limits correspond to 0.1−0.2×EW(Hβ). The luminosities
correspondingto these upper limits are listed in Table 1.
The division into two distinct groups of very high luminos-
ity AGNs is further confirmed by the Dietrich et al. (2002a)
observations. These authors found that two out of the six
luminous z ≃ 3.5 quasars in their sample have prominent
[O III]λ5007 lines while the remaining four had no trace of
this line. Combining with our new data we find that out of 35
high luminosity quasars, 24 show strong [O III]λ5007 and 11
others are consistent with no such line in their spectrum.
We have checked this findingin various ways. In particular,
we have examined the distribution of I([O III]λ5007)/I(Hβ)
whichis shownin Fig.2. The histogramis madeupof a broad
distribution centered at about 0.3 and a group of sources with
I([O III]λ5007)/I(Hβ)< 0.1. Unfortunately, the uncertainty
on the upper limit of I([O III]λ5007) in two of the sources
is very large and those objects bridge the gap between the
strong and the very weak [O III]λ5007 emitters. We must
5This ratio matches the BG92 definition of R Fe II, i.e., the ratio between
the EW of the Fe II blends in the λ4434–λ4684 band and EW(Hβ)

Page 3

DISAPPEARING NARROW LINE REGIONS AND THE ROLE OF ACCRETION3
TABLE 1. CONTINUUM AND EMISSION LINE MEASUREMENTS.
Quasar Nameza
Log λLλ(5100)
EW([O III])
Best Fit
[Å]
(4)
L[O III]
FWHM([O III)
Best Fit
[km s−1]
(8)
LHβ
Best Fit
[erg s−1]
(10)
Fe II/Hβ
Direct
[Å]
(5)
Best Fit
[erg s−1]
(6)
Direct
[erg s−1]
(7)
Direct
[km s−1]
(9)
[erg s−1]
(3)(1)(2)(11)
2QZ J001221.1−283630
2QZ J002830.4−281706
UM 667
LBQS 0109+0213
[HB89] 0123+257
2QZ J023805.8−274337
SDSS J024933.42−083454.4
[HB89] 0329−385
[HB89] 0504+030
SDSS J100428.43+001825.6
TON 618
[HB89] 1246−057
[HB89] 1318−113
[HB89] 1346−036
SDSS J135445.66+002050.2
UM 629
UM 632
UM 642
UM 645
SBS 1425+606
SDSS J170102.18+612301.0
SDSS J173352.22+540030.5
[HB89] 2126−158
[HB89] 2132+014
2QZ J221814.4−300306
2QZ J222006.7−280324
[HB89] 2254+024
2QZ J231456.8−280102
2QZ J234510.3−293155
2.339
2.401
3.132
2.349
2.369
2.471
2.491
2.435
2.473
3.046
2.226
2.240
2.306
2.370
2.531
2.460
2.517
2.361
2.257
3.202
2.301
3.428
3.282
3.199
2.389
2.414
2.083
2.400
2.382
46.26
46.58
46.28
46.80
46.57
46.57
46.38
46.71
46.32
46.44
47.31
47.16
46.89
46.88
46.49
46.56
46.54
46.29
46.31
47.38
46.34
47.00
47.25
45.77
46.54
47.22
46.45
46.31
46.32
< 10
32
16
25
27
< 7
27
20
73
54
< 3
< 5
14
< 3
< 3
35
15
11
23
23
< 7
11
13
59
12
13
15
15
20
···
31
16
31
27
···
27
24
74
60
···
···
13
···
···
23
16
14
30
19
···
8
9
52
13
16
13
15
28
< 43.59
44.40
43.78
44.50
44.31
< 43.70
44.11
44.30
44.49
44.46
< 44.11
< 44.14
44.31
< 43.72
< 43.22
44.40
44.03
43.63
43.98
45.03
< 43.50
44.34
44.64
43.83
43.94
44.63
43.94
43.79
43.96
······
1511
759
1398
587
···
815
491
1065
527
···
···
1903
···
···
1413
196
1696
525
1382
···
1335
1327
1315
1791
1019
612
1267
887
···
1823
1468
1341
532
···
394
450
836
996
···
···
2303
···
···
929
492
1859
633
659
···
833
1040
1328
2343
908
1390
1654
758
44.29
44.63
44.44
44.93
44.75
44.70
44.54
44.82
44.38
44.76
45.42
45.14
44.84
45.02
44.52
44.77
44.85
44.53
44.67
45.37
44.50
44.99
45.47
44.27
44.54
45.14
44.78
44.47
44.79
1.27
0.37
2.08
0.14
0.26
1.57
< 0.1
0.30
0.49
0.59
0.65
1.20
< 0.1
0.87
0.63
1.29
0.35
0.41
0.10
0.34
1.06
0.39
0.49
0.54
0.57
0.42
1.27
< 0.1
0.82
44.38
43.79
44.59
44.32
···
44.10
44.37
44.51
44.51
···
···
44.28
···
···
44.23
44.05
43.74
44.09
44.96
···
44.22
44.51
43.77
43.96
44.73
43.86
43.81
44.11
NOTE. — The methods used to obtain ’Best Fit’ and ’Direct’ measurements are outlined in Paper I.
aSystemic redshift (see Paper I).
00.10.2 0.30.40.50.6
F[O III]/FHβ
0
2
4
6
8
10
Number
FIG. 2.— [O III]λ5007/Hβ line ratio histogram for our sources. The dark
region indicates sources with upper limits (arrows) on the [O III]λ5007 flux.
therefore consider two hypotheses: one is a real dichotomy
in the [O III]λ5007 line intensity and the other, a continuous
distributionin EW([O III]λ5007)which has a longtail at very
small EWs.
Very high luminosity sources that are also weak
[O III]λ5007 emitters have been found by Yuan and Wills
(2003) in their IR study of high redshift quasars.
ever, most of those objects are broad absorption line quasars
(BALQSOs) that are known to have a weak [O III]λ5007
line. Regarding intermediate luminosity AGNs, BG92 find
that 26 out of the 87 sources in their PG quasar sample show
EW([O III]λ5007)< 10Å and 12 show EW([O III]λ5007)<
6Å. In addition, about half of the quasars in the new Sulen-
tic et al. (2004) high-z sample have EW([O III]λ5007)< 6Å.
Assuming the upper limits we obtained can be translated to
actual EW measurements, we conclude that the EW distribu-
tion in our sample is not very different from that of BG92.
Combining with the information about nearby Seyfert 1s we
conclude that there are very few weak [O III]λ5007 emitters
among low luminosity AGNs, but their fraction increases to-
wards intermediate and high luminosity. However, L[O III]of
the veryluminoussources in oursample is two ordersof mag-
nitude larger than observed in the most luminous BG92 ob-
jects. This has important implications for the physics and the
structure of the NLR in these extreme cases as discussed in
the following sections.
How-
3. LUMINOSITY, EQUIVALENT-WIDTH, AND SIZE
CORRELATIONS
3.1. Measured and predicted NLR sizes
B02 measured NLR sizes in seven PG quasars and com-
pared them with sizes obtained by Falcke et al. (1998) in
nearby Seyfert 2 galaxies. Their main finding is a strong cor-
relationbetween the NLR radius (RNLR) and the [O III]λ5007

Page 4

4NETZER ET AL.
434445
Log LHβ[erg s-1]
22
33
55
1010
2020
3030
5050
100100
RNLR [kpc]
PG QSOs (Bennert et al. 2002)
z>2 QSOs w/[O III]
z>2 QSOs w/o [O III]
FIG. 3.— The RNLR vs. LHβdiagram including our luminous high-z
quasars. Filled squares are the original B02 quasar data, and the straight
line is the B02 RNLR−LHβbest-fit line. Empty symbols represent high-z
quasars with measured [O III] lines, for which RNLRwas inferred from the
B02 RNLR−L[O III]relation. One-sided error bars on the empty symbols in-
dicate the difference between best-fit and ’direct’ measurements (see text).
Arrows represent high-z quasars, for which we have an upper limit on the
[O III]λ5007 line flux and hence a limit on RNLRfrom the B02 relation. Note
the two distinct groups and the enormous predicted RNLRat high Hβ lumi-
nosity.
line luminosity. Their relationship (B02 Eq. 1) scaled to the
somewhat different cosmology adopted here, can be written
as
RNLR= 2.1L0.52±0.06
[O III],42kpc,
(1)
where L[O III],42= L[O III]/1042erg s−1.
agreementwith RNLR∝L1/2
2.1 kpc is of order 15%). For reasons that shall become ap-
parent later, we prefer to use the equivalent relation involving
the Hβ luminosity (B02 Eq. 3 converted to our assumed cos-
mology)
RNLR= 1.15L0.49±0.06
This is in perfect
[O III](theuncertaintyontheconstant
Hβ,42
kpc,
(2)
where LHβ,42= LHβ/1042erg s−1.
The recent, more detailed work of Schmitt et al. (2003) use
a sample of 60 Seyfert 1 and Seyfert 2 galaxies and discuss
in detail the differences between the two sub-groups,the con-
centration of the [O III]λ5007 emission, etc. The main find-
ing which is relevant to our work is the following relationship
(adjusted to the cosmology used here)
Rmaj≃ 1.2L0.33±0.04
[O III],42kpc,
(3)
where Rmaj is the size of the semi-major axis of the
[O III]λ5007 nebulosity. This is significantly different from
the B02 results in both the R−L dependence and the normal-
ization. Schmitt et al. (2003)have also investigatedthe corre-
lation when the seven B02 sources are added to their sample.
The result is a steeper dependence of the form Rmaj∝ L0.42
A new work by Bennert et al. (2004) argues that much of the
difference is due to orientation since most of the sources in
Schmitt et al. (2003) are Seyfert 2s while the more luminous
sources in B02 are all type-I AGNs. We shall return to this
issue in § 4.
Theoreticalsuggestionsthat the NLR size shouldscale with
L1/2
discussed in many papers (see Netzer 1990 for references
[O III].
ion, where Lionis the ionizing source luminosity, have been
560 565570 575 580 585 590595
590
585
580
575
570
[Ο ΙΙΙ]λ5007
FIG. 4.— Two dimensional spectrum around the [O III]λ5007 region of
2QZ J222006.7−280324. The coordinates of the vertical and horizontal axes
are given in pixels, where each pixel in the spatial (vertical axis) corresponds
to 0.446′′(∼ 3.5 kpc). Note that there is no trace of extended [O III]λ5007
emission beyond a ∼ 1′′(∼ 7.2 kpc) radius from the center.
prior to 1990 and Dopita et al 2002 for more recent publi-
cations). This is based on the assumption that both the broad
line region (BLR) gas and the NLR gas are photoionized by
a central source whose spectral energy distribution changes
only slightly with source luminosity. Spectroscopic studies
show a remarkable similarity between the emission line spec-
trumofhighandlow luminosityAGNs. This suggests thatthe
ionization parameter, U (defined here as the ratio of the Ly-
man continuum photon density to the hydrogen number den-
sity NH), and the typical gas density, are basically the same in
all sources. Since U ∝ Lion/NHR2, we find R ∝ L1/2
ditional assumption is that most of the [O III]λ5007 emission
originates in radiation-bounded clouds, because of the fairly
uniform value of I([O III]λ5007)/I(narrow Hβ). This allows
to replace Lionby the luminosity of any hydrogen recombina-
tion line, e.g., Hβ. We note that there is a well established
R−L correlation for the BLR gas (e.g., Kaspi et al. 2000)
where reverberation mappings show that RBLR∝ L0.6±0.1.
A real physical explanation of any R−L dependence is still
lacking since the mechanism controlling the gas density and
location is unknown. Some papers assume a stratified, radia-
tionboundedNLR with a pre-chosenrunof densityandhence
level of ionization (e.g., Netzer 1990; Komossa & Schulz
1997). Such models are naturally normalized in incident flux
units (Lion/R2) and can be tuned to produce the same mean
U for all sources. However, some narrow emission lines are
probablyproducedin density boundedgas (e.g., Binette, Wil-
son, & Storchi-Bergmann 1996) which considerably compli-
cate the models. It is also clear (e.g., Alexander et al. 1999)
that well studied NLRs contain a large range of conditions
with a spread in ionization parameter and gas density. An
almost orthogonal approach is provided by the “locally opti-
mally emitting cloud” (LOC) model (Ferguson et al. 1997).
The assumption in this case is of a large range of conditions
(density and covering factor) at each location, where the in-
tensities of the various lines reflect the line production effi-
ciency at each location. This efficiency is the highest for
densities that are close to the critical density of the line in
question. The model provides a natural scaling of Lionwith
RNLRprovided there is a large reservoir of gas with simi-
lar properties in all AGNs and on all scales. Finally, there
are equally complex NLR models where shock excited gas
contributes significantly to the observed NLR emission (e.g.,
Contini, Prieto, & Viegas 1998; Schiano 1986; Dopita and
ion. An ad-

Page 5

DISAPPEARING NARROW LINE REGIONS AND THE ROLE OF ACCRETION5
Sutherland 1995) but no natural R−L scaling.
The recent papers by Dopita et al. (2002) and Groves et
al. (2004) provide a more solid foundation to NLR modeling.
Theseauthorsassumeddusty,stratified NLR cloudswherethe
external radiation pressure acts mostly on the dust particles
and forces the local ionization parameter to certain specific
values such that in the [O III]λ5007 producinggas,U ≃10−2.
The model suggests a natural RNLR∝ L1/2dependence. It also
implies that the composition and temperature of the gas are
rather different from those assumed in other models because
of metal depletion.
3.2. New [O III] and RNLRmeasurements
The combination of the B02 observations and the recent
theoretical developments point to a “natural” RNLR∝ L1/2
pendence yet raises a severe problem regarding the NLR size
in high luminosity AGNs. Any such scaling will lead, at large
enough Lion, to sizes that are larger than galactic sizes. The
B02 results addressed here are used for normalizing this rela-
tionship, but the problem exists at high luminosity whether or
not their scaling is correct. Given this, we would not expect
to see any strong [O III]λ5007 emitters in high luminosity
sources, yet our new observations clearly show such objects.
To define the problem in a more quantitative way we plot
in Fig. 3 three quantities vs. LHβ(luminosity of the entire
emission line). The first is the measured RNLRfrom the B02
sample (seven sources) where we also show the best (modi-
fied) B02 fit (Eq. 2). The second is from our newly observed
Hβ and [O III]λ5007 lines with two additional sources from
Dietrich et al. (2002a). For these we use Eq. 1 to guess
RNLR, given the observed L[O III]. The predicted RNLRfor
most sources in this group lie close to the value predicted
from LHβ(the straight line) confirming the small scatter in
I([O III]λ5007)/I(Hβ). The third group includes the seven
sources from our sample and the four sources from Dietrich
et al. (2002a) where no [O III]λ5007 has been detected. For
these we use RNLRderived from the upper limits on L[O III].
The upper limits on RNLRobtained in this way are a factor of
2−3 smaller than those derived from LHβ.
The implications of Fig. 3 are clear. For those sources with
measured Hβ and [O III]λ5007 lines, the derived RNLR is
enormous, exceeding 70 kpc in the most luminous sources.
We consider those sizes completely unreasonable for reasons
thatarediscussedinthenextsection. LikeFig.2, this diagram
suggest a dichotomy in the properties of the high luminosity
quasars, where some sources show strong [O III]λ5007 lines
and others show no or very weak such emission.
There are two other ways to verify, experimentally, the
B02 claim. Most of our sample is at z ≃ 2.5. At this red-
shift, and the chosen cosmology, the angular diameter dis-
tance is a weak function of redshift and corresponds to about
7 kpc per arc-sec. The predicted RNLR, using the B02 rela-
tionships and our measured [O III]λ5007 luminosities, cor-
responds to a total extent of 2′′-10′′. This can be tested by
spatially resolved space and ground-based observations. The
2D spectra of two of our sources ([HB89] 0329−385, and
2QZJ222006.7−280324)showprominent[O III]λ5007emis-
sion which allow such measurements (see Fig. 4). In these
cases, most of the line flux (>99%) is emitted within the cen-
tral four pixels corresponding to a radius of 7.2 kpc at each
source. This size is a strong upper limit since much of the
flux is likely to be due to the PSF (corresponding to ∼ 1′′at
the time of observations). The two upper limits on RNLRob-
ionde-
4344 45 464748
Log λLλ(5100) [erg s-1]
0
0.5
1
1.5
2
Log EW([O III]λ5007) [Å]
2QZ
This work
BG92
Sulentic et al. (2004)
FIG. 5.— Baldwin relationship for [O III]λ5007. Mean (over luminosity
bins) values for the Croom et al. (2002) 2QZ sources are marked with as-
terisks. The data presented in this paper are marked with filled squares. The
BG92, and Sulentic et al. (2004) samples are marked with (circles), and (di-
amonds), respectively. Upper limits on EW([O III]λ5007) are marked with
arrows). The mean EW([O III]λ5007) of the 2QZ sample is marked by a
dotted line.
tained in this way are a factor of ∼ 5 smaller than the radii
derived from the B02 relationships.
The term “NLR radius” used by B02 is ambiguous since
those authors used very low surface brightness features to de-
fine the dimension of the [O III]λ5007 nebulosity. We used
the HST archive to extract and re-analyze the B02 images.
In particular, we examined the source showing the largest
[O III]λ5007 nebulosity, PG 1049−005, and remeasured its
observed [O III]λ5007 image. We found that 95% of the line
emission is encircled within a radius of 1.1′′which corre-
sponds to a radius of 5.5 kpc at the source. This is half of the
radius deduced by B02 and suggests that the bulk of the NLR
emission is emitted within a volume which is much smaller
than inferred by their relationships. The different way of
measuring the [O III]λ5007 nebulosity is probably the main
source of discrepancy in normalization (i.e., the NLR radius
atL[O III],42=1)betweentheB02andtheSchmittetal. (2003)
works.
In summary, our new observations contradict the B02 re-
sults in two ways. First, about one third of the high luminos-
ity sources show no trace of an NLR. Second, there are direct
indications in three cases, and sound theoretical reasons (see
below), to suggest that most of the NLR emission is restricted
to a volumewhich is muchsmaller thaninferredfromthe B02
relationships.
3.3. The Baldwin relationship for [O III]
The new data show the presence of two groups of lumi-
nous AGNs, those with strong [O III]λ5007 and those with
no (or very weak) such line. Out of the 35 sources investi-
gated by us (6 from Dietrich et al. 2002a and 29 from our
IR sample), 24 belong to the first group and 11 to the second.
As mentioned above, earlier studies, like the Yuan & Wills
(2003) work, already found very weak [O III]λ5007 in sev-
eral AGNs. Most or all of these are known BALQSOs while
only two of the 10 sources discussed here (one from Dietrich
et al. 2002a, [HB89] 0105−265 and one, [HB89] 1246−057,
from our sample) show BALQSO properties. Thus, weak or
no [O III]λ5007 seems to be a common property of many lu-

Page 6

6NETZER ET AL.
FIG. 6.— FWHM([O III]λ5007) vs. (a) λLλ(5100), (b) MBH, and (c) L/LEddfor the Shields et al. (2003) sample (empty symbols) and the high-z quasars
presented in this paper (filled symbols). The strongest correlation is between FWHM([O III]λ5007) and luminosity. The correlation weakens as the dependence
on luminosity drops from L through MBH(∝ L0.6) to L/LEdd(∝ L0.4), as indicated by the Spearman rank correlation coefficients at the top left of each panel.
minous AGNs.
To further illustrate this point we plot in Fig. 5
EW([O III]λ5007) versus λLλ(5100Å) for the high-z quasars
from our sample and that of Sulentic et al. (2004), for which
B magnitudes were transformed to λLλ(5100) assuming Lν∝
ν−αwith α = 0.5. We also included the BG92 sample in the
diagram, and the 2QZ data of Croom et al. (2002), for which
bJmagnitudeswere transformedto λLλ(5100)using the same
methods as above. The diagram shows that for the popula-
tion of strong [O III]λ5007 emitters, there is no reduction of
EW([O III]λ5007)with sourceluminosity. Onthe otherhand,
there are many weak, or no [O III]λ5007 emitters at high lu-
minosity that could give the impression that the line EW de-
creases with increasingsourceluminosity. (Infact, theCroom
et al. 2002 data represent mean EW([O III]λ5007) in several
luminosity bins. Assuming a fraction of weak [O III]λ5007
emitters in those bins similar to the one found by BG92, we
find that the plotted average may underestimate the typical
EW([O III]λ5007) in those sources by about 25%). We sus-
pect that the Dietrich et al. (2002b) claim of a Baldwin rela-
tionship for this line is the result of their averaging together
four very weak [O III]λ5007 emitters with two sources show-
ing “typical” EW([O III]λ5007).
For completeness, we tested also the correlation of
EW([O III]λ5007) with accretion rate (in terms of the
LBol/LEddratio,hereafterL/LEdd; seePaperIformoredetails)
Correlations with accretionrate are foundto be extremelyim-
portant in Paper I and in § 3.5 below. For EW([O III]λ5007),
this correlation is not significant (see Table 2).
3.4. FWHM correlations
We have tested our sample for possible correlations of
luminosity, black hole (BH) mass, and accretion-rate with
FWHM([O III]λ5007). Such correlations have been in-
vestigated in the past (e.g., Shields et al.
FWHM([O III]λ5007) has been suggested as a potentially
useful surrogate for σ∗.
Fig. 6 shows FWHM([O III]λ5007) vs. source luminosity,
MBH, and L/LEddfor our sample and the Shields et al. (2003)
sources. To obtain the intrinsic line width we assumed
∆λ2
where ∆λinstis the instrumental resolution. Since the slit-
2003), and
true= ∆λ2
obs−∆λ2
inst,
(4)
width and the seeing disk sizes were comparable during the
time of observations, this is a reasonable assumption for
line profiles that are indistinguishable from a Gaussian. The
uncertainty on FWHM([O III]λ5007) is quite large for the
poorer S/N spectra and is of the order of the instrumental (or
rebinnedinstrumental)resolution(∼600km s−1). BH masses
and L/LEddfor all sources on these diagrams were calculated
as prescribed in Paper I.
On their own, the measured FWHM([O III]λ5007) for our
high-zsources show no correlationwith luminosity,BH mass,
or accretion-rate, due to the narrow luminosity range of our
sample. The additional Shields et al. (2003) data increase
this range considerablyand show significant correlationswith
all three parameters. The strongest correlation is between
FWHM([O III]λ5007) and luminosity. The strength of the
correlation increases with the dependence on luminosity, i.e.
less significant correlations as one goes from L to MBH(∝
L0.6) to L/LEdd(∝ L0.4).
3.5. Correlations involving Hβ and Fe II
In Paper I and in Table 1 we give LHβ, which we trans-
formed to EW(Hβ), and Fe II/Hβ. Here, again, we have
tested the correlations of these quantities against L, MBH, and
L/LEddusing our sample, and the samples of Sulentic et al.
(2004) and BG92 (accretion rates for the BG92 sample were
calculated as prescribed in Paper I). For EW(Hβ) we find sig-
nificant correlations with both L and L/LEdd. Details of the
correlations are given in Table 2 and in Fig. 7. In the case of
Fe II/Hβ the only significant correlation is with L/LEdd. De-
tails of those correlations are also given in Table 2. Fig. 8
shows the strongest correlationof Fe II/Hβ (against the accre-
tion rate) for our new sources combined with those of BG92.
The Baldwin relationship found here for EW(Hβ), and
shown in Fig. 7 is interesting since it is in contradiction with
the Croom et al.(2002) finding for this line.
our combined sample is far from being complete, in partic-
ular at the high luminosity range where we specifically chose
to observe high luminosity sources. Thus it is likely that
lower luminosity sources at high redshift would have a larger
EW(Hβ) that will spoil the correlation. Regarding the de-
pendence on accretion rate, similar selection effects may be
operating but the observed correlation is so strong that we
suggest that this may be a real effect. We also note that in
However,

Page 7

DISAPPEARING NARROW LINE REGIONS AND THE ROLE OF ACCRETION7
TABLE 2. SPEARMAN RANK CORRELATION COEFFICIENTS MATRIX
Property vs.
(1)
MBH
(2)
L/LEdd
(3)
Fe II/Hβ
(4)
EW([O III])
(5)
EW(Hβ)
(6)
λLλ(5100)
MBH
L/LEdd
Fe II/Hβ
EW([O III])
EW(Hβ)
0.85a
···
···
···
···
···
0.48a
0.03a
···
···
···
···
-0.04b
-0.22c
0.48c
···
···
···
-0.09d
0.14e
-0.13e
-0.39f
···
···
-0.37g
-0.15h
-0.47h
-0.33b
0.28d
···
NOTE. — Significant correlations with chance probabilities smaller than
1% are given in bold face. Upper limits were not included in the correlations.
a118 sources from BG92 and this work
b124 sources from BG92, Sulentic et al. (2004), and this work
c107 sources from BG92 and this work
d121 sources from BG92, Sulentic et al. (2004), and this work
e104 sources from BG92 and this work
f110 sources from BG92, Sulentic et al. (2004), and this work
g135 sources from BG92, Sulentic et al. (2004), and this work
h118 sources from BG92 and this work
44454647
Log λLλ(5100) [erg s-1]
1.5
2
2.5
Log EW(Hβ) [Å]
-2-1.5-1 -0.50
Log L/LEdd
a) b)
FIG. 7.— EW(Hβ) vs. (a) luminosity and (b) L/LEddfor the BG92 sam-
ple (including broad-line AGNs, empty circles, and NLS1s filled circles), the
Sulentic et al. (2004) sample (empty diamonds in panel a) and the high-z
quasars presented in this paper (filled squares).
Fig. 7, the narrow-lineSeyfert1 galaxies(NLS1s) are situated
very close to the high luminosity, high accretion rate sources.
Thesituationresemblesthe Baskin & Laor(2004)findingthat
EW(C IVλ1549) depends strongly on L/LEddeven in low lu-
minosity AGNs that do not show the Baldwin relationship for
this line.
Fig. 8 is hard to interpret. Considering only sources with
measurable Fe II/Hβ (i.e., a ratio larger than ∼ 0.1) we find
a clear trend of increased Fe II/Hβ for higher accretion rates
(see Table 2). In particular,we notethe similar locationon the
diagramof our high-z quasars and the NLS1s. However,there
is a significant number of sources with very small Fe II/Hβ
and very large L/LEdd. We note, in this respect, that our new
observations extend the measurement of Fe II/Hβ to values of
L/LEddnever investigated before. The only exceptions, per-
haps, are a few BALQSOs in the Yuan & Wills (2003)sample
with spectral properties very different from those of our (non-
BAL) sources (note that these authors measured the entire op-
tical Fe II blends and their values must be scaled down by a
factor 3.58 for comparison with the measurements presented
in our work; B. Wills private communication).
The above two findings suggest that the main reason for the
increase of Fe II/Hβ with the accretion rate is the decrease of
EW(Hβ) with L/LEdd. In fact, looking at EW(Fe II) against
L/LEdd(not shown here) we find no correlation at all. The
fractional increase in Fe II/Hβ with accretion rate is also con-
sistent with the decrease in EW(Hβ). However, we cannot
0.01 0.1
L/LEdd
1
0.1
1
Fe II / Hβ
BG92 NLS1
BG92
This work
FIG. 8.— Fe II/Hβ vs. L/LEddfor the BG92 sample, and the high-z quasars
presented in this paper (symbols are as in Fig. 7). Arrows represent no de-
tection of iron emission (in the BG92 sample the upper limit on the Fe II/Hβ
ratio was set to 0.09, which is the lowest value they reported, and in our
sample the upper limits are given in Table 1).
rule out the possibility that changes in the Fe/H abundancera-
tio are involved too, as shown, in Paper I, to be the case for
N/C.
In Fig. 9 we plot the well known anti-correlation between
EW([O III]λ5007) and Fe II/Hβ (e.g., BG92) for our sources,
for the low-z BG92 sources, and for the new intermediate-z
sample of Sulentic et al. (2004). Our sources are consistent
with this trend. Fe II/Hβ for our sources is generally higher
than in BG92 and Sulentic et al. (2004) indicating, perhaps,
the strong dependenceof this propertyon the total luminosity.
However, there are clear exceptions, i.e., sources with very
highluminosity,yet relativelysmallFe II/Hβ. Here,again,we
note similar large values of Fe II/Hβ in several of the BALQ-
SOs of Yuan & Wills (2003) .
4. DISCUSSION
4.1. Do enormous NLRs really exist?
A major goal of the present investigation is to test the
NLR properties and the NLR spectrum in very high luminos-
ity sources where the theoretically predicted RNLR, as well
as the empirical B02 relationship, results in unreasonably
large dimensions. The B02 sample already included claims
for sources with RNLR∼ 10 kpc and our bright [O III]λ5007
emitters would continue this relationship to enormously large
NLRs (more than 100 kpc in diameter, see Fig. 3). There
are, however, fundamental problems in this suggested size-
luminosity interpretation on both observational and theoreti-
cal grounds.
B02 suggested two estimates of the NLR size (Eq. 1 &
2). We separate the discussion of those claims into two, ac-
cording to the two sub-groups discovered here (the weak and
the strong [O III]λ5007 emitters). For the weak [O III]λ5007
emitters we find a clear contradictionbetween the RNLRbased
on the observed Hβ luminosity and the one based on the
observed (or the upper limit on) [O III]λ5007 luminosity
(Fig. 3). As for the strong [O III]λ5007 emitters, 2D spec-
tra of two of our sources rule out the large predicted dimen-
sions (§ 3.2). Moreover, our new measurements of the largest
[O III]λ5007 nebula in the B02 sample are also in conflict
with the B02 relationships. Based on the evidence in hand

Page 8

8NETZER ET AL.
we suspect that the RNLR∝ L1/2
some intermediate luminosity scale and that the “true” NLR
radius, defined here as the radius encompassing 95% of the
line emission, does not exceed a few kpc even in the most
luminous quasars.
There are other predictions that make us question the ex-
istence of such enormous NLRs. The suggested RNLR∝ L1/2
relationship would predict NLR sizes, in the most luminous
AGNs, that exceed the size of the largest known bulges and,
in fact, the size of the largestknowngalaxies(exceptfor some
cD galaxies). Such sizes are unacceptable for several reasons.
First, the escape velocity from a spherical galaxy is roughly
290M1/2
M⊙and R10the radial distance in units of 10 kpc. The new
FWHMs listed in Table 1 and shown in Fig. 6, compared
with the predicted RNLR, suggest therefore, dynamically un-
bounded NLRs. Given those velocities and dimensions, the
dynamical time for the most luminous [O III]λ5007 emitters,
is a few ×107R10years suggesting a short lived phenomenon.
The inferred amount of ionized gas in such NLRs, given radi-
ation bounded gas and a “typical” ionization parameter, is
iondependence breaks down at
ion
11R−1/2
10kpckm s−1, where M11is the mass in units of 1011
MNLR≃ 109
?Cf
0.1
?
R2
10N21M⊙,
(5)
where Cf is the covering fraction and N21the column den-
sity in units of 1021cm−2.
ments show that for strong [O III]λ5007 emitters N21> 1.
This would give MNLR> 109M⊙for all of our sources and
MNLR> 1010M⊙for the most luminous [O III]λ5007 emit-
ters. Judging by the observed FWHM, most of this material
is probably unbound and thus flows from the center at out-
flow rates approaching 106R2
of ∼ 103cm−3. All those numbers seem incompatible with
long term mass ejection in AGN-hosting galaxies.
The Schmitt et al. (2003) results alleviate some of these
difficulties because of the smaller size normalization and the
flatter RNLR-L[O III]dependence. However, it is not at all clear
that extrapolatingtheir results to much higherluminositywith
the suggested L1/3dependence (Eq. 3) is justified in view of
the B02 measurements.
We are facing a situation where sound physical arguments
(the narrow line spectrum, the Groves et al. 2004 dusty NLR
model, etc.) support the expected RNLR∝ L1/2relationship,
yet its application to the most luminous quasars give unrea-
sonably large sizes, masses and mass outflow rates. As ex-
plained below, the most likely explanation in our opinion is
that “typical” NLRs (i.e., those similar in their properties to
the ones observed in Seyfert 1 galaxies) cannot last very long
in high and perhaps also in intermediate luminosity AGNs.
This means that the [O III]λ5007emitting regionsin our sam-
ple may be of a different origin and physical properties.
Simple photoionization argu-
10M⊙yr−1assuming a density
4.2. Luminous NLRs as star forming regions
If compact NLRs are indeed typical of many high luminos-
ityAGNs, thentheirpropertiesmustbeverydifferentfromthe
properties of those NLRs observed in nearby sources. In par-
ticular, the gas density in the highest luminosity NLRs must
be several orders of magnitudelarger. Consider for example a
maximum NLR size of ∼ 3 kpc and assume a similar ioniza-
tionparameterinall NLRs. This wouldmeanthat thegasden-
sity in the most luminous [O III]λ5007 emitters is 102−103
times larger than the density in nearby Seyfert 1 galaxies.
0.11
Fe II / Hβ
1
10
100
EW([O III]λ5007) [Å]
BG92
BG92 NLS1
This work
Sulentic et al. (2004)
FIG. 9.— EW([O III]λ5007) vs. Fe II/Hβ for the BG92 sample, the Su-
lentic et al. (2004) sample, and the high-z quasars presented in this paper
(symbols are as in Fig. 7). Upper limits on Fe II/Hβ (arrows pointing left)
are similar to those shown in Fig. 8. Upper limits on EW([O III]λ5007), i.e.,
no [O III]λ5007 detection, for the BG92 (Sulentic et al. 2004) sample were
set to 1 (1.3) Å, respectively, which is the lowest value each of them reported.
The upper limits on EW([O III]λ5007) for our sample are given in Table 1.
The spectroscopicpropertiesmust be verydifferenttoo which
can, in principle, be tested by accurate observations. Unfor-
tunately, present day IR spectroscopy is very limited in this
respect because of the restricted wavelength bands available
to ground-based observations.
A possible origin of a high density gas in kpc-scale nuclear
regions is violent star-forming activity. Such events can pro-
duce high density, large column density, non-solar composi-
tion dusty gas. The overall spectrum of such regions is likely
to differ from the spectrum of nearby, lower density NLRs.
At present we can only observe [O III]λ5007 which is the
strongest emission line under a variety of conditions. Future,
space-bornspectroscopy,will be able to test this idea by look-
ing for other emission lines.
The scenario we propose to explain the observations of
our high luminosity AGNs, and the apparent break down of
the RNLR∝ L1/2
One where such scaling continues to high luminosity due
to radiation pressure force or other effects. This results in
short-lived enormous NLRs that will show basically no nu-
clear narrow emission lines during most of their life. The
other group is those sources where starburst or another un-
known process ejects high density gas into their nuclear re-
gion. This gas is ionized and excited by the central radia-
tion source and produces the observed strong [O III]λ5007
lines. Such “star-forming NLRs” would have spectral proper-
ties that are rather different from those observed in nearby
less luminous sources. Given the similar fraction of weak
[O III]λ5007 emitters in BG92 and in our new sample, we
suggest that the phenomenon is of a continuous nature and
starts at some intermediate luminosity. Its clearest manifes-
tation is in the highest luminosity sources such as the ones
observed here.
ionrelationship, is of two distinct populations.
4.3. Correlations involving Fe II, Hβ, and [O III]
The sample used here, which combines data from various
different sources, is not complete. However, it allows the
best test, so far, of the BG92 Fe II/Hβ –[O III]λ5007 rela-
tionship and several other suggested correlations at the high

Page 9

DISAPPEARING NARROW LINE REGIONS AND THE ROLE OF ACCRETION 9
end of the AGN luminosity function. The results presented
here suggest that the most extreme values of Fe II/Hβ require
very high luminosity. They also suggest that higher accretion
rate results in larger Fe II/Hβ (Fig. 8) and that L/LEddis the
most important factor for determining several other correla-
tions. There are, however, some exceptions, i.e., sources with
large L/LEddyet small Fe II/Hβ. Our work shows that the
decrease of EW(Hβ) with increasing L/LEddin the present
sample is probably the cause for the increase of Fe II/Hβ with
the accretion rate. EW(Fe II) by itself does not depend on
the accretion rate and it is therefore possible that the iron-to-
hydrogenabundanceratio playsno rolein this correlation. On
the other hand, we do not have the data or a good enough
theory (e.g., Verner et al. 2004 and references therein) to
completelyrule out the possibility that the iron abundancede-
pends on luminosity or on the accretion rate. In a similar way,
EW([O III]λ5007) is not correlated with L/LEdd, but we sug-
gest that the Fe II/Hβ – EW([O III]λ5007) anti-correlation is
driven by the accretion rate.
Sulentic et al. (2004) and others argued that the general
AGN population should be divided into two sub-groups ac-
cording to the so-called Eigenvector 1 (E1, see BG92) which
contains four main observables. According to this scheme,
group A sources show the strongest E1 properties.
radio-loud AGNs belong to group B, i.e., those with weaker
E1 properties. Here we are not interested in E1 as such since
our sample is not complete and so is the combination of our
sample with the BG92 and the Sulentic et al. (2004) samples.
We are however in a position to look at the extreme end of
the distribution in L and in L/LEddand test for those quanti-
ties that depend on them. We have already discussed all the
relevant correlations. Here we note that all those involving
L/LEddshow that NLS1s occupy the same part of parame-
ter space as the high luminosity, high accretion rate AGNs
observed by us. These sources are thus the high luminos-
ity analogs of NLS1s and should, perhaps, be referred to as
narrow-line type 1 quasars (NLQ1s).
Most
5. CONCLUSIONS
We have discussed the near-IR spectra of 29 newly ob-
served high-luminosity high-redshift AGNs and used the data
to argue that previous claims for expected and observed
RNLR∝ L1/2
tions. About 2/3 of all very high luminosity sources show
strong [O III]λ5007 lines while the remaining objects show
no or very weak such line. We suggest that the NLR proper-
ties in high luminosity, strong [O III]λ5007 emitters are very
different from those observed in nearby AGNs, and possibly
imply denser than “usual” NLRs. The origin of the high den-
[O III]dependence are in conflict with the observa-
sity gas is likely to be a violent star-forming event in the nu-
cleus. We also investigated Fe II/Hβ and EW([O III]λ5007)
at the high end of the luminosity and L/LEddin AGNs and
showed that the first of those is probably driven by the over-
all accretion rate while the second is independent of source
luminosity (i.e., no Baldwin relationship) or accretion rate.
The Fe II/Hβ ratio may be an iron abundance indicator but
this cannot be provenobservationallybecause of the EW(Hβ)
dependence on accretion rate.
We are grateful to the technical staff at the AAT and
TNG observatories for invaluable help during the observa-
tions. We acknowledge constructive remarks made by an
anonymous referee, which helped to improve this work. This
work is based on observations made with the Italian Tele-
scopio Nazionale Galileo (TNG) operated on the island of
La Palma by the Centro Galileo Galilei of the INAF (Isti-
tuto Nazionale di Astrofisica) at the Spanish Observatorio del
Roquede los Muchachosof the Institutode Astrofisica de Ca-
narias. We would like to thank Angela Cotera for allowing us
to use half a night of her time at AAT, and Dirk Grupe for
providing us an electronic version of the Boroson & Green
(1992) Fe II template. We gratefully acknowledge construc-
tive remarks from an anonymous referee, who helped to im-
prove this work considerably. The 2dF QSO Redshift Survey
(2QZ) was compiled by the 2QZ survey team from observa-
tions made with the 2-degree Field on the Anglo-Australian
Telescope. Funding for the creation and distribution of the
SDSS ArchivehasbeenprovidedbytheAlfredP. SloanFoun-
dation, the Participating Institutions, the National Aeronau-
tics and Space Administration, the National Science Founda-
tion, the U.S. Department of Energy, the Japanese Monbuk-
agakusho, and the Max Planck Society. The SDSS Web site
is http://www.sdss.org/. The SDSS is managed by the As-
trophysical Research Consortium (ARC) for the Participating
Institutions. The Participating Institutions are The Univer-
sity of Chicago, Fermilab, the Institute for Advanced Study,
the Japan Participation Group,The Johns HopkinsUniversity,
Los Alamos National Laboratory, the Max-Planck-Institute
for Astronomy (MPIA), the Max-Planck-Institute for Astro-
physics (MPA), New Mexico State University, University of
Pittsburgh, Princeton University, the United States Naval Ob-
servatory, and the University of Washington. This research
has made use of the NED database which is operated by the
Jet Propulsion Laboratory,California Institute of Technology,
under contract with the National Aeronautics and Space Ad-
ministration. This work is supported by the Israel Science
Foundation grant 232/03. RM acknowledges partial support
by the Italian Ministry of Research (MIUR).
REFERENCES
Alexander, T., Sturm, E., Lutz, D., Sternberg, A., Netzer, H., & Genzel, R.
1999, ApJ, 512, 204
Baldwin, J. A. 1977, ApJ, 214, 679
Barth, A. J., Ho, L. C., Filippenko, A. V., Rix, H., & Sargent, W. L. W. 2001,
ApJ, 546, 205
Baskin, A. & Laor, A. 2004, MNRAS, 350, L31
Bennert, N., Falcke, H., Schulz, H., Wilson, A. S., & Wills, B. J. 2002, ApJ,
574, L105
Bennert, N., Falcke, H., Shchekinov, Y., & Wilson, A. S. 2004, in The
Interplay among Black Holes, Stars and ISM in Galactic Nuclei, IAU 222,
eds. T. Storchi-Bergmann, L.C. Ho, and H.R. Schmitt
Binette, L., Wilson, A. S., & Storchi-Bergmann, T. 1996, A&A, 312, 365
Boroson, T. A. & Green, R. F. 1992, ApJS, 80, 109
Contini, M., Prieto, M. A., & Viegas, S. M. 1998, ApJ, 492, 511
Croom, S. M. et al. 2002, MNRAS, 337, 275
Dietrich, M., Appenzeller, I., Vestergaard, M., & Wagner, S. J. 2002a, ApJ,
564, 581
Dietrich, M., Hamann, F., Shields, J. C., Constantin, A., Vestergaard, M.,
Chaffee, F., Foltz, C. B., & Junkkarinen, V. T. 2002b, ApJ, 581, 912
Dopita, M. A. & Sutherland, R. S. 1995, ApJ, 455, 468
Dopita, M. A., Groves, B. A., Sutherland, R. S., Binette, L., & Cecil, G. 2002,
ApJ, 572, 753
Falcke, H., Wilson, A. S., & Simpson, C. 1998, ApJ, 502, 199
Ferguson, J. W., Korista, K. T., Baldwin, J. A., & Ferland, G. J. 1997, ApJ,
487, 122
Groves, B. A., Dopita, M. A., & Sutherland, R. S. 2004 (submitted to ApJ)
Kaspi, S., Smith, P. S., Netzer, H., Maoz, D., Jannuzi, B. T., & Giveon, U.
2000, ApJ, 533, 631
Komossa, S. & Schulz, H. 1997, A&A, 323, 31
McIntosh, D. H., Rieke, M. J., Rix, H.-W., Foltz, C. B., & Weymann, R. J.
1999, ApJ, 514, 40

Page 10

10NETZER ET AL.
Nelson, C. H. 2000, ApJ, 544, L91
Nelson, C. H. & Whittle, M. 1996, ApJ, 465, 96
Netzer, H. 1990, in Active Galactic Nuclei, ed., T. J.-L. Courvoisier & M.
Mayor (Berlin: Springer), 57
Netzer, H. 2003, ApJ, 583, L5
Schiano, A. V. R. 1986, ApJ, 302, 81
Schmitt, H. R., Donley, J. L., Antonucci, R. R. J., Hutchings, J. B., Kinney,
A. L., & Pringle, J. E. 2003, ApJ, 597, 768
Shemmer, O., Netzer, H., Maiolino, R., Croom, S., Oliva, T., & Di Fabrizio,
L. 2004, ApJ, in press (Paper I)
Shields, G. A., Gebhardt, K., Salviander, S., Wills, B. J., Xie, B., Brotherton,
M. S., Yuan, J., & Dietrich, M. 2003, ApJ, 583, 124
Sulentic, J. W., Stirpe, G. M., Marziani, P., Zamanov, R., Calvani, M., &
Braito, V. 2004, A&A, in press (astro-ph/0405279)
Veilleux, S. 1991, ApJS, 75, 383
Veilleux, S. & Osterbrock, D. E. 1987, ApJS, 63, 295
Verner, E., Bruhweiler, F., Verner, D., Johansson, S., Kallman, T., & Gull, T.
2004, ApJ, in press
Yuan, M. J. & Wills, B. J. 2003, ApJ, 593, L11