The Soft X#Ray Properties of a Complete Sample of Optically Selected Quasars. II. Final Results

We present the Ðnal results of a ROSAT PSPC program to study the soft X-ray emission properties of a complete sample of low-z quasars. This sample includes all 23 quasars from the Bright Quasar Survey with z ¹ 0.400 and cm ~2 . Pointed ROSAT PSPC observations were made for all N H IGal \ 1.9 ] 10 20 quasars, yielding high signal-to-noise (S/N) spectra for most objects, which allowed an accurate determination of the spectral shape. The following main results were obtained:

1.The spectra of 22 of the 23 quasars are consistent, to within D30%, with a single power-law model at rest-frame 0.2È2 keV.There is no evidence for signiÐcant soft excess emission with respect to the best-Ðt power law.We place a limit (95% conÐdence) of D5 ] 1019 cm~2 on the amount of excess foreground absorption by cold gas for most of our quasars.The limits are D1 ] 1019 cm~2 in the two highest S/N spectra.
2. The mean 0.2È2 keV continuum of quasars agrees remarkably well with an extrapolation of the mean 1050È350 continuum recently determined by et al.
for z [ 0.33 quasars.This sug-A Zheng (1996) gests that there is no steep soft component below 0.2 keV.
3. SigniÐcant X-ray absorption (q [ 0.3) by partially ionized gas ("" warm absorber ÏÏ) in quasars is rather rare, occurring for of the population, which is in sharp contrast to lower luminosity active [5% galactic nuclei (AGNs), where signiÐcant absorption probably occurs for D50% of the population.
4. Extensive correlation analysis of the X-ray continuum emission parameters with optical emissionline parameters indicates that the strongest correlation is between the spectral slope and the Hb a x FWHM.A possible explanation for this remarkably strong correlation is a dependence of on a x L /L Edd , as seen in Galactic black hole candidates.
5. The strong correlations between and Fe II/Hb, and the peak [O III] to Hb Ñux ratio are a x L *O III+ , veriÐed.The physical origin of these correlations is still not understood.
6.There appears to be a distinct class of "" X-rayÈweak ÏÏ quasars, which form D10% of the population (three out of 23), where the X-ray emission is smaller, by a factor of 10È30, than expected based on their luminosity at other bands and on their Hb luminosity.These may be quasars in which the direct X-ray source is obscured and only scattered X-rays are observed.
7. Thin accretion disk models cannot reproduce the observed 0.2È2 keV spectral shape, and they also cannot reproduce the tight correlation between the optical and soft X-ray emission.An as yet unknown physical mechanism must be maintaining a strong correlation between the optical and soft X-ray emission.
8. The H I/He I ratio in the high Galactic latitude ISM must be within 20%, and possibly within 5%, of the total H/He ratio of 10, which indicates that He in the di †use H II gas component of the interstellar medium is mostly ionized to He II or He III.
We Ðnally note the intriguing possibility that although in radio-loud quasars ([1.15 ^0.14) is Sa x T signiÐcantly Ñatter than in radio-quiet quasars ([1.72 ^0.09) the X-ray emission may not be related to the presence of radio emission.The di †erence in may result from the strong versus Hb FWHM Sa x T a x correlation and the tendency of radio-loud quasars to have broader Hb.Subject headings : galaxies : active È galaxies : nuclei È quasars : general È X-rays : galaxies

INTRODUCTION
Quasars emit most of their power in the ultraviolet to soft X-ray regime.The position-sensitive proportional counter (PSPC) detector aboard ROSAT allowed a signiÐcantly improved study of the soft X-ray emission of quasars com-pared with earlier missions (some of which were not sensitive below 2 keV), such as HEAO-  Mushotzky, 1993).indicated that the X-ray emission above 1-2 keV is well described by a power law with a spectral slope a x \ of about [0.5 for radio-loud quasars and about d ln f l /d ln l [1.0 for radio-quiet quasars.Large heterogeneous samples of AGNs were recently studied using the ROSAT PSPC by & Fink and by Brinkmann, & Walter (1993) Wang, Bergeron However, the objects studied in the papers (1996).mentioned above do not form a complete sample, and the available results may be biased by various selection e †ects that were not well deÐned a priori.In particular, most of the studied objects are nearby, intrinsically X-rayÈbright AGNs.
To overcome the potential biases in existing studies, we initiated a ROSAT PSPC program to make an accurate determination of the soft X-ray properties of a well-deÐned and complete sample of quasars, selected independently of their X-ray properties.This program was designed to address the following questions : 1. What are the soft X-ray spectral properties of the lowredshift quasar population ?
2. Are simple thin accretion disk models (e.g., Laor 1990) able to Ðt the observed optical/UV/soft X-ray continuum ?Are other modifying mechanisms, such as a hot corona (e.g., & Elvis required ?Are models invoking opti-Czerny 1987), cally thin free-free emission possible (e.g., Korista, Ferland, & Peterson 1990 ;Barvainis 1993) ? 3. Do the observed soft X-ray properties display any sig-niÐcant correlations with other properties of these quasars ?Are these correlations compatible with various models for the continuum and line emission mechanisms ?
Our sample includes all 23 quasars from the Bright Quasar Survey (BQS) sample & Green with (Schmidt 1983) z ¹ 0.400 and cm~2, where is the N H I Gal \ 1.9 ] 1020 N H I Gal H I Galactic column density as measured at 21 cm.These selection criteria allow optimal study of soft X-ray emission at the lowest possible energy.The additional advantages of the BQS sample are that it has been extensively explored at other wavelengths (see et al. hereafter Laor 1994, Paper I, for further details) and that it includes only bright quasars, thus allowing high signal-to-noise (S/N) X-ray spectra for most objects.The sample selection criteria are independent of the quasarÏs X-ray properties, and we thus expect our sample to be representative of the low-redshift, optically selected quasar population.
Preliminary results from the analysis of the Ðrst 10 quasars available to us were described in Here we Paper I. report the analysis of the complete sample, which allows us to address the three questions posed above.The outline of the paper is as follows.In we describe the observations °2 and the analysis of the spectra.
describes the Section 3 analysis of correlations between the soft X-ray properties and other continuum and emission-line properties.In we °4 compare our results with other soft X-ray observations and discuss some of the implications.We conclude in with °5 answers to the questions raised above, and with some new questions to be addressed in future studies.

OBSERVATIONS AND ANALYSIS OF THE SPECTRA
The complete sample of 23 quasars is listed in  (1989).four of the 23 quasars in our sample are radio loud (deÐned here as R º 10).N H I Gal AGNs, including most AGNs observed by ROSAT , which would be very useful for eliminating the signiÐcant systematic uncertainty in the PSPC spectral slope that must be present when a low-accuracy is used.N H I Gal lists the PSPC observations of all the quasars.Table 2 For the sake of completeness we include also the 10 quasars already reported in All sources were detected, and Paper I. their net source counts range from 93 to 38,015, with a median value of about 1900 counts.The PROS software package was used to extract the source counts.Table 2 includes the exposure times, the dates of the observations, the net number of counts and their statistical error, the count rate, the radius of the circular aperture used to extract the source counts, the o †set of the X-ray position from the center of the PSPC Ðeld of view, the ROSAT sequence identiÐcation number, and the SASS version used for the calibration of the data.All objects, except one, are typically within 15@@ of the center of the PSPC Ðeld of view, so all the identiÐcations are secure.The one exception is PG 1440]356, where the pointing was o †set by 40@ from the position of the quasar et al.
Note that (Gondhalekar 1994).the exposure times are uncertain by about 4% as the result of a number of possible systematic errors, as described by et al.
The typically large number of counts for Fiore (1994).each object allows an accurate determination of the spectral slope for most objects, as described below.
Model Ðts to the extracted number of source counts per pulse-invariant (PI) channel, were carried out using the N ch ob, XSPEC software package.PI channels 1È12 of the original 256 channel spectra (E \ 0.11 keV), were ignored, since they are not well calibrated and are inherently uncertain because of the large Galactic optical depth.Paper I power-law model could not provide an acceptable Ðt to three of the 10 quasars, although in two of them the apparent features could not be Ðtted with a simple physical model, and in one of them this may be due to calibration errors (see °4.1).
As mentioned above, a comparison of the free Ðt (Ðt 1) N H with the Ðt (Ðt 2) allows us to look for evidence for an N H I Gal absorption or an emission excess.We measure the statistical FIG.1a FIG.1.ÈObserved vs. single power-law Ðt PSPC spectra of the 13 quasars not reported in The upper panel for each object shows the observed Paper I. count rate (points with error bars), and the expected count rate using the best-Ðt single power-law model with a free (solid-line histogram).The middle panel N H shows the observed minus the expected Ñux in units of standard deviations, and the lower panel for each object shows the fractional deviations from the expected Ñux.The best-Ðt the Ðt s2, and the number of degrees of freedom (dof ) are indicated for each object.The spectra of all quasars are consistent with a x , a simple power law.Note in particular the high S/N spectrum of PG 1116]215, where intrinsic features below 1 keV must have an amplitude of less than ^10%.

FIG. 1b
signiÐcance of the reduction in with the addition of s fit 2 N H as a free parameter using the F-test In PG (Bevington 1969).1440]356 we Ðnd a signiÐcant reduction with Pr \ 7.5 ] 10~4 (F \ 12.98 for 26 dof ), where Pr is the probability that the reduction in s2 is not statistically signiÐcant (calculated using the FTEST (1996) 1440]35 at 0.1È0.15keV (80È120 using the Extreme A ) Ultraviolet Explorer.There is no indication for such a component in the PSPC spectrum below 0.15 keV (Fig. 1c).However, given the very low sensitivity of the PSPC below 0.15 keV, such a steep soft component may still be consistent with the PSPC spectrum.In the other 12 objects the free Ðt does not provide a signiÐcant improvement (i.e., N H Pr [ 0.01), and thus there is no clear evidence for either intrinsic absorption or low-energy excess emission above a simple power law.
compares the Galactic deduced from the Figure 2 N H accurate 21 cm measurements with the best-Ðt X-ray column deduced using the free Ðt.The straight line N H represents equal columns.The s2 of the cm) \ N H (21 model is 31.9 for 22 dof (PG 1001]054 was not N H (X-ray) included because of the low S/N), which is acceptable at the 8% level.This result demonstrates that there is no sig-niÐcant excess absorption over the Galactic value in any of our objects.It is interesting to note that in our highest S/N spectra, those of PG 1116]215 and PG 1226]023, is determined to a level of (0.8È1) ] 1019 cm~2 N H (X-ray) and is still consistent with cm), indicating that the N H (21 two methods agree to better than 10%. The average hard ROSAT band (0.5È2 keV) slope for the complete sample is [1.59 ^0.08 (excluding PG 1114]445, which is a †ected by a warm absorber, and PG 1001]054 and PG 1425]267, for which the S/N is very low).This slope is not signiÐcantly di †erent from the average slope for the full ROSAT band, [1.63 ^0.07.(1996), Wang The results of the single power-law Ðt with a free (1996).
N H in these papers are all consistent with our results for the overlapping objects.The only discrepancy is with PG 1444]407 for which, although both the present authors and Wang et al. (1996) Ðnd a similar slope, Wang et al.Ðnd evidence for absorption while we Ðnd no such evidence.For a simple power-law Ðt to PG 1444]407 et al.
As discussed in (°5.1.3),the di †erence in spectral Paper I slopes at hard (2È10 keV) and soft X-rays raises the possibility that may be changing within the PSPC band itself.a x The individual spectra are well Ðtted by a simple power law, and thus any spectral curvature must be consistent with zero.Stronger constraints on the spectral curvature may be obtained by measuring the average curvature parameter (b, deÐned in for the complete sample, since the Paper I) random error in the mean is smaller by N1@2 D 5 than the random error for individual objects.Unfortunately, the PSPC calibration uncertainty at low energy, discussed in introduces a systematic error in b (which obviously Paper I, does not cancel out as N1@2) and, as shown in does Paper I, not allow a reliable determination of the curvature parameter.We therefore did not try to constrain the spectral curvature parameter in this paper.

CORRELATION ANALYSIS
presents eight of the 12 rest-frame continuum  (1992).
H 0 \ 50 km s~1 Mpc~1 and q 0 \ 0.5.The four additional continuum parameters that are omitted from for the sake of brevity are  2 of & Green Boroson (1992).The signiÐcance of the correlations was tested using the Spearman rank-order correlation coefficient which is (r S ), sensitive to any monotonic relation between the two variables.A summary of the main correlation coefficients and their two-sided signiÐcance is given in Table 5.

SigniÐcance L evel
In the correlation analysis was carried out using Paper I 10 objects, and only relatively strong correlations (r S º 0.76) could be detected at the required signiÐcance level (Pr ¹ 0.01).Here, with 23 objects, Pr ¹ 0.01 corresponds to and we can thus test for the presence of weaker r S º 0.52, correlations and check whether the correlations suggested in remain signiÐcant.We have searched for corre-Paper I lations among the 12 continuum emission parameters, and between these 12 parameters and the 18 emission-line parameters listed above, which gives a total of 294 di †erent correlations.One thus expects about 1 spurious correlation with Pr ¹ 3.4 ] 10~3 in our analysis, and for a signiÐcance level of 1% one would now have to go to Pr ¹ 3.4 ] 10~5 rather than Pr ¹ 1 ] 10~2.However, we Ðnd that there are actually 42 correlations with Pr ¹ 3.4 ] 10~3, rather than just one, in our sample.Thus, the probability that any one of them is the spurious one is only 2.4%, and the signiÐcance level of these correlations is reduced by a factor of 7 (\ 0.024/0.0034)rather than a factor of 300.Below we assume that correlations with Pr ¹ 1 ] 10~3 are signiÐcant at the 1% level (there are 30 correlations with Pr ¹ 1 ] 10~3, versus an expected number of 0.3).Thus, given the large number of correlations we looked at, we can only test reliably for correlations with (which corr S º 0.64 responds to Pr ¹ 1 ] 10~3 for 23 data points).is associated with a low 2 keV Ñux.Note also the two lowest a x spectra (PG 1001]054 and PG 1411]442), which appear to form a distinct group of "" X-rayÈweak ÏÏ quasars.L ower panel : As in middle panel, for the four RLQs.PG 1425]267 may also be an "" X-rayÈweak ÏÏ quasar.also suggested by the presence of a marginally signiÐcant correlation between and Pr \ 0.0089 ; see a x a ox (r S \ 0.533, below), and the absence of a signiÐcant correlation between and Pr \ 0.36).a x a os (r S \ [0.201,The X-ray luminosity distribution appears to be bimodal, with two quasars, PG 1001]054 and PG 1411]442, being a factor of 30 weaker than the mean radio-quiet quasar.These two quasars appear to form a distinct group of "" X-rayÈweak quasars.ÏÏ The statistics for the radio-loud quasars (RLQs) are much poorer, and there is no well-deÐned mean, but PG 1425]267 may be a similar X-rayÈweak RLQ.

Correlations with Emission-L ine Properties
presents the correlations between the hard X-ray Figure 4 luminosity, (log l \ 17.685) or the soft X-ray lumi-L 2 keV nosity, and the luminosity of Hb, [O III], He II, or L 0.3 keV , Fe II.The value of and the two-sided signiÐcance level r S (Pr) of are indicated above each panel.Upper limits were r S not included in the correlations.Thus, the actual correlations for He II, where there are Ðve upper limits, are likely to be smaller than found here (there is only one upper limit for [O III] and Fe II, and none for Hb).Excluding He II, the X-ray luminosity is most strongly correlated with L Hb (r S \ 0.734, Pr \ 6.6 ] 10~5).We note in passing that has an L Hb even stronger correlation with the luminosity at 3000 A Pr \ 9 ] 10~8) and with the near where *v \ Hb FWHM.This Figure 5f 1.5L 14.25 1@2 *v~2, parameter is related, under some assumptions, to the luminosity in Eddington units, as further discussed in °4.7. ) ( P r ) better with the optical and near IR luminosity).Note the position of the X-rayÈweak quasars.These appear to be fainter by factors of 10È30 in the X-ray than expected based on their This factor is similar to the deÐciency seen in L Hb .Fig. 3.

Urry, & Canizares
In particular, earlier missions sug-1990).gested that the hard X-ray slope et al. (Lawson 1992 ;et al. extends down to D0.5 keV, with a Williams 1992) steep rise at lower energy.Here we Ðnd that the 0.2-2 keV spectrum is Ðtted well by a single power law with Galactic absorption.This indicates that (1) the break between the soft and hard X-ray slope must occur well above 0.5 keV, (2) the break must be gradual, and ( 3 (1996), keV.However, the exact break energy cannot be accurately determined from the ASCA spectra because of likely calibration uncertainties below 1 keV.
As mentioned in the di †erent Einstein IPC and Paper I, EXOSAT LE ] ME results may be traced back to the combined e †ect of the lower sensitivity of these instruments below D0.5 keV, and possibly some calibration errors.Small systematic errors in the PSPC response function appear to be present below 0.2 keV et al. and (Fiore 1994), this instrument is thought to be signiÐcantly better calibrated at low energy than earlier instruments.
No signiÐcant spectral features are present in the PSPC spectra of any of 13 additional quasars reported here, indicating that intrinsic features must have an amplitude of less than 10%È20%.Note in particular the high S/N spectrum of PG 1116]215, where the number of counts is about 12 times the median sample counts, yet this spectrum is still consistent with a simple power law.For the complete sample we Ðnd that only one quasar, PG 1114]445, has a signiÐcant physical feature that is well described by a warm absorber model.In two other quasars, PG 1226]023 and PG 1512]370, there are signiÐcant features below 0.5 keV In the case of PG 1512]370 the features are at a (Paper I).level of D30%, and in PG 1226]023 they are at a level of and may well be due to small calibration errors.[10% This result is consistent with the result of et al.Fiore (1994), who found that a simple power law provides an acceptable Ðt to the individual spectra of six high S/N PSPC quasar spectra.
A composite optical to hard X-ray spectral energy dis-FIG.5.ÈVarious correlations with (aÈd) Emission-line parameters that correlate strongly with Open symbols are for radio-loud quasars.The a x .a x .a x versus Hb FWHM correlation is the strongest in our sample.Note that the X-rayÈweak quasars do not stand out in these correlations ; they thus have the "" right ÏÏ spectral slope but the "" wrong ÏÏ luminosity.(e) Note the trend of increasing with increasing which indicates that a steep tends to be a ox a x , a x associated with a weak 2 keV Ñux rather than a strong 0.3 keV Ñux.The position of the X-rayÈweak quasars are marked.( f ) The Y -axis is a measure of under some simpliÐed assumptions, as described in L /L Edd °4.7.
tribution for RLQs and RQQs is displayed in To Figure 6.construct it, we used the mean the mean L 14.25 (Table 5), a o , the mean and the mean in our sample.We excluded a x , a ox from the mean the three X-rayÈweak quasars, and PG 1114]445, where is highly uncertain because of the pres-a x ence of a warm absorber.The mean spectra were extended above 2 keV assuming a slope of [1 for RQQs and [0.7 for RLQs.The & Ferland hereafter Mathews (1987, MF) quasar energy distribution is also displayed for the purpose of comparison.The shape assumes a steep soft com-MF ponent with a break to the hard X-ray slope above 0.3 keV, and it therefore signiÐcantly underestimates the soft X-ray Ñux at D0.2È1 keV.
RLQs tend to be somewhat stronger hard X-ray sources than RQQs.This trend, together with the Ñatter X-ray slope of RLQs, was interpreted by & Elvis as pos-Wilkes (1987) sible evidence for a two-component model.In this interpre-tation RLQs have the same hard X-ray component with as in RQQs, with an additional contribution from a x D [1 a Ñatter component, making their overall X-ray a x D [0.5 emission Ñatter and brighter.The additional X-ray component in RLQs could be related to the radio jet, e.g., through inverse Compton scattering.The composite spectrum suggests that although RLQs are brighter at 2 keV, they may actually be fainter at lower energy because of their Ñatter The RLQ composite is based only on four objects a x .and is therefore rather uncertain.In addition, the results of et al. °IIIc) suggest that RLQs in the PG Sanders (1989, sample are about twice as bright at 2 keV compared with RQQs of similar optical luminosity, rather than the D30% found for the composite ; thus the di †erence in PSPC a x would imply a smaller di †erence at 0.2 keV than is shown in the composite.If RLQs are indeed weaker than RQQs at 0.2 keV, then the two-component model suggested above would not be valid, and RLQs need to have a di †erent X-ray emission process, rather than just an additional component.
The di †erence between RLQs and RQQs may actually be unrelated to the radio emission properties, as discussed in °4.4.also displays a simple cuto † power-law model of Figure 6 the form with and L l P laoe~hl@kTcut a o \ [0.3 T cut \ 5.4 ] 105 K.This is an alternative way to interpolate between the UV and soft X-ray emission, and it is also a reasonable approximation for an optically thick thermal component.The lack of a very steep low-energy component down to 0.2 keV allows us to set an upper limit on The upper limit T cut .is set using and the slope from 3000 to rest-frame a o a os @ , A 0.15(1 ] z) keV (the lowest energy where the Galactic absorption correction error is ¹30%), given by where (1 ] z).The upper limit on the log l s @ \ 16.560 ] log cuto † temperature is related to the spectral slopes by T cut ul [ a os @ ) log (l s @ /1015) K .
We Ðnd a rather small dispersion in with T cut ul ST cut ul T \ (5.5 ^2.6) ] 105 K, averaged over the complete sample which corresponds to a cuto † energy of 47 eV, or (Table 4), about 3.5 rydbergs.This value of corresponds very T cut ul closely to the far-UV continuum shape assumed by (see MF Fig. 6 6, with the PSPC mean spectra, suggest that the FUV powerlaw continuum extends to the soft X-ray band.In the case of RQQs there is remarkable agreement in both slope and normalization of the soft X-ray and the FUV power-law continua.RLQs are predicted to be weaker than RQQs at D100 eV by both the FUV and the PSPC composites.It thus appears that there is no extreme UV sharp cuto † in quasars, and that the fraction of bolometric luminosity in the FUV regime is signiÐcantly smaller than assumed. The UV to X-ray continuum suggested in is very Figure 6 di †erent from the one predicted by thin accretion disk models and suggested by photoionization models.In (°4.5) particular, it implies about a 4 times weaker FUV ionizing continuum compared with the continuum that was MF deduced based on the He II j1640 recombination line.
One should note, however, that the Zheng et al. sample is practically disjoint from our low-z sample, so it may still be possible that low-z quasars have a di †erent FUV continuum.

Intrinsic Absorption
As shown in the H I column deduced from our Figure 2, accurate 21 cm measurements is consistent for all objects with the best-Ðt X-ray column.It is quite remarkable that even in our highest S/N spectra the 21 cm and X-ray columns agree to a level of about 1 ] 1019 cm~2, or 5%È7%.This agreement is remarkable, since the 21 cm line and the PSPC are actually measuring the columns of di †erent elements.Most of the soft X-ray absorption is due to He I or He II rather than H I, and the H I column is indirectly inferred assuming the column ratio H I/He I \ 10.The fact that the 21 cm line and the PSPC give the same H I column implies that the H I/He column ratio at high Galactic latitudes must indeed be close to 10.The dispersion in the H I/He column ratio is lower than 20% (based on typical quasars in our sample), and may even be lower than 5% (based on our highest S/N spectra).There is therefore no appreciable Galactic column at high Galactic latitudes where the ionized fraction of H di †ers signiÐcantly from the ionized fraction of He, as found for example in H II regions (e.g., Osterbrock 1989).
The consistency of the 21 cm and X-ray columns also indicates that the typical column of cold gas intrinsic to the quasars in our sample must be smaller than the X-ray N H uncertainty, or about 3 ] 1019(1 ] z)3 cm~2.An additional indication for a lack of an intrinsic cold column in quasars comes from the fact that the strong correlations of with a x the emission-line parameters described above become (°3.Nandra using the Ginga LAC for a largely overlapping (1994) sample, show low-energy absorption, or spectral features inconsistent with the simple power law typically observed above 2 keV.Quasars are very di †erent.Excess absorption produces signiÐcant spectral features only in one object (PG 1114]445 ; see i.e., D5% (1/23) of the objects, and Paper I), the absorbing gas is partially ionized ("" warm ÏÏ) rather than neutral.Given the typical S/N in our sample, we estimate that a partially ionized absorber that produces q [ 0.3 can be ruled out in most of our objects.We cannot rule out partial absorption, or complete absorption and scattering, by a very high column density cm~2) gas, since (N H [ 1024 such e †ects may only suppress the Ñux level without a †ecting the spectral shape, and without inducing signiÐcant spectral features.As described in we suspect that such °4.8,strong absorption may indeed be present in D10% (3/23) of the quasars in our sample (the X-rayÈweak quasars).

Implications of the Continuum-Continuum Correlations
The continuum-continuum luminosity correlations found here are all weaker than those found in This is Paper I. mostly due to the three X-rayÈweak quasars that were not present in For example, in Paper I we found that Paper I. can be predicted to within a factor of 2, once f 0.3 keV f 1.69 km is given.This statement is still valid if the four extreme objects labeled in (middle panel) are excluded.The Figure 3 implications of the near-IR versus X-ray luminosity corre-lation on the X-ray variability power spectrum were discussed in Paper I.
In  ^0.28, where the ^here and above refers to the dispersion about the mean rather than the error in the mean.
The versus correlation found here is weaker than in a x a ox owing to the presence of the X-rayÈweak quasars.Paper I, However, the other 20 quasars appear to follow a trend of increasing with increasing indicating, as a x a ox (Fig. 5e), discussed in that a steep is generally associated Paper I, a x with a weak hard X-ray component (at 2 keV) rather than a strong soft excess.The only object that clearly violates this trend is PG 1626]554 which has both a steep (Fig. 3), a x and a strong soft excess.Mason, & Co rdova Puchnarewicz, and Puchnarewicz et al. present (1994) (1995a, 1995b ) PSPC spectra of three AGNs with extremely strong soft excess, where Our sample suggests that L 0.2 keV [ L 3000 z .such objects are most likely rare, as can also be inferred from the selection criteria of Puchnarewicz et al. (1995a, who selected their three objects from the ROSAT 1995b), WFC all-sky survey, in which only Ðve AGNs were detected.This selection criterion implies that these AGNs must have a very high far-UV Ñux.
The soft X-rayÈselected quasars in the et Puchnarewicz al.
sample have and none of (1996) Sa ox T \ [1.14 ^0.02, their quasars is "" X-ray weak,ÏÏ i.e., with The a ox \ [1.6.absence of X-rayÈweak quasars in their sample is clearly a selection e †ect.The small survey area implies that most quasars in their sample are optically rather faint (m B D 18È19)."" Normal ÏÏ quasars in their sample produce a few a ox hundred PSPC counts, but "" X-rayÈweak ÏÏ quasars are below their detection threshold.The abundance of "" X-rayÈ loud ÏÏ quasars (i.e., in the Puchnarewicz et al.

Implications of the Continuum-L ine Correlations
The presence of the strong correlations of with the Hb a x FWHM, with and with the Fe II/Hb ratio described L *O III+ , in is veriÐed here.The correlation coefficients for Paper I the complete sample are comparable to or somewhat smaller than those found in but since the sample is Paper I, larger, the signiÐcance level is now much higher We (Fig. 5).also report here an additional strong correlation of with a x the ratio of [O III] peak Ñux to the Hb peak Ñux, which is one of the emission-line parameters measured by Boroson & Green The versus Hb FWHM correlation is the (1992).
a x strongest correlation we Ðnd between any of the X-ray continuum emission parameters and any of the emission-line parameters reported by Boroson & Green (with the addition of line luminosities reported in Shastri 1993 the fact that the average in RLQs is signiÐcantly Ñatter a x than in RQQs may be completely unrelated to the (°4.1) presence of radio emission, it may just reÑect the fact that RLQs tend to have broader lines than RQQs (e.g.,Tables 3  and 5 in & Green This appears to be the Boroson 1992).case in our sample, where the RLQs follow the same a x versus Hb FWHM distribution deÐned by the RQQs (Fig. This intriguing suggestion can be clearly tested by com-5a).paring for RLQs and RQQs of similar Hb FWHM. a x Brandt, & Fink studied in detail narrow-Boller, (1996) line Seyfert 1 galaxies (NLS1), and they also Ðnd an apparently signiÐcant trend of increasing with increasing Hb a x FWHM.However, the scatter in their sample is signiÐcantly larger than that found here.In particular, they Ðnd a large range of for Hb FWHM \ 2000 km s~1, where a x only a few objects are available in our sample.The overall larger scatter in the et al. data is probably due Boller (1996) in part to the generally larger statistical errors in their a x determinations.Large systematic errors may also be induced by the use of Hb FWHM from a variety of sources.The measured Hb FWHM can be sensitive to the exact measuring procedure, such as continuum placement, subtraction of Fe II blends, and subtraction of the narrow component of the line (produced in the narrow-line region), which may not be well resolved in low-resolution spectra.

Inconsistency with T hin Accretion Disk Models
presents the continuum emission from two thin Figure 7 accretion disk models.The models are for a disk around a rotating black hole, and viscous stress that scales as where is the radiation pressure and is (P gas P rad )1@2, P rad P gas the gas pressure & Netzer SigniÐcant soft (Laor 1989).X-ray emission is obtained for disks with a high accretion rate and a small inclination.However, as discussed by Fiore FIG.7.ÈObserved energy distribution of some of the quasars in our sample vs. two accretion disk spectra.The disk models are for a rotating black hole, a \ 0.1, and viscous (the other parameters stress P (P gas P rad )1@2 are indicated).Note that although signiÐcant soft X-ray Ñux can be produced by the thin-disk models, the spectral slope is always much steeper than observed.et al.
the observed soft X-ray spectral slope is always (1995), much Ñatter than the one produced by a thin "" bare ÏÏ accretion disk model.As noted above, there is no indication in the 0.2È2 keV band for a very steep and soft "" accretion disk ÏÏ component.
Although thin disks cannot reproduce the 0.2È2 keV spectral shape, they may still be able to contribute a signiÐcant fraction of the Ñux at the lowest observed energy, i.e., 0.2È0.3keV, above which a nonthermal power-law component sets in.As noted by et al. and in Walter (1994) Paper accretion disk models predict a large dispersion in the I, optical/soft X-ray Ñux ratio, and the strong correlation between these Ñuxes argues against the idea that a thin disk produces both the optical and soft X-ray emission.The arguments put by et al. and in were Walter (1994) Paper I only qualitative, and were not based on actual disk models.Furthermore, the objects in the small sample of Walter et al. were selected from known optically bright AGNs, and they also had to be bright soft X-ray sources, since most spectra were obtained from the ROSAT all-sky survey.Thus, these objects were a priori selected to be bright at both optical-UV and at soft X-rays, and the absence of a large scatter in the UV/soft X-ray Ñux ratio may just reÑect the sample selection criteria.Such selection e †ects are not present in our sample, since the sample was deÐned independently of the X-ray properties, and X-ray spectra were obtained for all objects.
Below we describe a detailed calculation of the expected distribution of the optical/soft X-ray Ñux ratio for a complete optically selected sample based on the thin disk models of & Netzer and show that such models Laor (1989), cannot be reconciled with the observed distribution of the optical/soft X-ray Ñux ratio in our complete sample.
The optical/soft X-ray Ñux ratio, of a given disk a os , model depends on the black hole mass, accretion rate m 5 , and inclination angle h \ cos~1 k.We now need to determine what distributions of these parameters will be consistent with the observed luminosity function in a complete optically selected sample.The intrinsic distribution of disk inclinations must be random.However, the observed distribution depends on the shape of the luminosity function of quasars, and possible obscuration e †ects, as described below.
The luminosity function of quasars is parameterized using the number density of quasars per unit volume per unit magnitude, ' 4 d2N/dM dV , and it is well Ðtted by a power law over a restricted range of magnitude M. Using Figure 2  The observed dn/dm 9 P m 9 ~2.5.number of objects in a Ñux-limited sample is dN ob /dL P (dn/dL )V (L ), where V (L ) P L 3@2 is the observable volume for a Ñux-limited sample, such as the BQS sample.We therefore select a mass distribution of dN ob /dm 9 P m 9 ~1.compares the observed distribution of as a Figure 8 a os , function of at 3000 with the one expected from thin lL l A , accretion disk models with the parameter distribution described above.Thin-disk models cannot account for the very small scatter in a os .The range of observed disk inclinations may actually be smaller than assumed here.For example, for a certain range of inclinations the disk may be completely obscured by an optically thick torus, as suggested in uniÐcation schemes for RQQs (e.g., However, even if k is Ðxed at a Antonucci 1993).given value for all AGNs (say k \ 1, which corresponds to the points extending from on the left axis to a os \ [1.5 on the bottom axis of the range in log lL l \ 46.5 Fig. 8), m 9 and will still produce a range in which is much larger m 5 a os than observed.
The X-ray power-law emission is most likely produced by Comptonization of the thermal disk emission in a hot corona above the disk (e.The small range in Titarchuk 1995).a os implies that some physical mechanism that couples the optical and soft X-ray emission processes must be operating, e.g., through a feedback that regulates both the temperature (see & Maraschi and the optical Haardt 1993) depth of the corona.

FIG. 8.ÈObserved
vs. luminosity distribution, and the one expected a os from a distribution of accretion disks that conforms to the observed luminosity function.The observed range in is much smaller than the a os expected range, indicating that the UV and soft X-ray emission mechanisms are much better correlated than expected from thin-disk models.
As pointed out by various authors Fabian, & Min-(Ross, eshige & Takahara et al. 1993 ;Shimura 1995 ;Dorrer 1996), and shown in simple thin accretion disks with no Figure 7, corona can produce a signiÐcant Ñux below 1 keV.For various disk model parameters can in fact be signiÐ-a os cantly Ñatter than observed yet such extreme Ñat (Fig. 8), opticalÈsoft X-ray spectra are only rarely observed (e.g., The Ñattest spectra are expected for Puchnarewicz 1995b).disks that are close to edge-on (e.g., et al. and Laor 1990), one therefore needs to assume that such disks are not observable.This is indeed expected in AGN uniÐcation schemes that invoke obscuring material close to the disk plane.Alternatively, the accreted material may form a geometrically thick, rather than a thin, conÐguration close to the center, which would display a smaller inclination dependence.1984 ;Fiore 1995 ;Pounds, 1995), which display a steep slope in the soft and the hard X-ray bands when their brightness increases.The physical interpretation for this e †ect described by et al. is Pounds (1995) that the hard X-ray power law is produced by Comptonization in a hot corona and that, as the object becomes brighter in the opticalÈUV, Compton cooling of the corona increases, and the corona becomes colder, thus producing a steeper X-ray power law.This is obviously far from being a predictive model, since the coronal heating mechanism is not speciÐed, and it is implicitly assumed that the coronal heating does not increase much as the quasar becomes brighter.However, the narrow Hb line proÐles provide independent evidence that steep quasars may indeed a x have a higher as further described below.L /L Edd , The of quasars can be estimated under two L /L Edd assumptions : (1) The bulk motion of the gas in the broadline region is virialized, i.e., *v ^(GM/r)1@2, where *v \ Hb FWHM.(1995).lower black hole mass.The broad-lineÈemitting gas does not extend much closer to the center in narrow-line AGNs, as it does not extend much closer to the center in other AGNs, simply because of the e †ects of a higher ionization parameter, and a higher gas density, each of which quenches line emission.4.8.T he X-RayÈweak Quasars Two of the quasars in our sample, the RQQs PG 1001]054 and PG 1411]442, and possibly also the RLQ PG 1425]267, appear to form a distinct group, which we term here "" X-rayÈweak ÏÏ quasars, where the normalized X-ray luminosity is a factor of 10È30 smaller than the sample median.The position of these quasars as outliers can be noticed in the near-IR normalized Ñux distribution in the versus correlation and in the (Fig. 3), a x a ox (Fig. 5e), Hb versus 2 keV and 0.3 keV luminosity correlations (Fig.
The Ðrst two indicators are based on the spectral shape, 4).but the last one is independent of the spectral shape, and it also suggests a deÐciency of the X-ray luminosity by a factor of 10È30 relative to the one expected based on the Hb luminosity.An apparently bimodal distribution in can a ox also be seen in Figure 5b  Although the three X-rayÈweak quasars in our sample stand out in luminosity correlations, they conform well to the correlations (Figs. They thus have the "" right ÏÏ a x 5aÈ5d).slope but the "" wrong ÏÏ Ñux level.Why are these quasars di †erent ?A simple answer is that for some unknown reason the X-ray emission mechanism, most likely Comptonization by T º 108 K electrons, tends to be bimodal, and in about 10% of quasars (or in all quasars for D10% of the time) the X-ray Ñux level is strongly suppressed, while the spectral slope is not a †ected.Another option is that these are just normal quasars where the direct X-ray Ñux happens to be obscured.In this case what we see is only the scattered X-ray Ñux.Photoionization calculations indicate that a few percent of the direct Ñux will be scattered, depending on the covering factor of the absorber and the ionization parameter.If the ionization parameter is large enough, then the scattering will be mostly by free electrons, which preserves the spectral shape (see and & Kriss Netzer 1993 Krolik Such scattering will explain why the Ñux level is 1995).strongly reduced, while the spectral shape is not a †ected.Note that the obscuring matter should be transparent in the visible range, as is the case with the absorbing matter in broad absorption line QSOs.
Additional hints toward this interpretation come from the fact that PG 1411]442 is a broad absorption line quasar (BALQSO ; Green, & Hutchings and Malkan, 1987), the UV absorbing gas may also produce soft X-ray absorption, as may also be the case in PHL 5200 observed here in the X-rayÈweak RQQs.Another hint is provided by the fact that PG 1114]445 is also somewhat underluminous at 0.3 keV and this quasar is most (Fig. 3b), likely seen through a warm absorber.The X-rayÈweak quasars could therefore be more extreme cases of PG 1114]445 and have an absorbing column that is large enough to completely absorb the direct soft X-ray emission.
Note Forthcoming HST spectra of all 23 quasars in our sample will allow us to test whether there is a one-to-one correspondence between X-ray weakness and broad absorption lines, i.e., whether all X-rayÈweak quasars are BALQSOs, and not just that all BALQSOs are X-ray weak, as strongly suggested by the & Mathur and Green (1996) et al.
results.Green (1996) A simple test of whether these are truly "" X-rayÈweak quasars ÏÏ or just normal highly absorbed quasars can be done by looking at their hard X-ray emission.If the X-ray column is below 1024 cm~2, then the obscuring material would become transparent at E \ 10 keV, and the observed hard X-ray emission will rise steeply above the cuto † energy, as seen in various highly absorbed AGNs, such as Another prediction is that the X-rayÈweak quasars should show lower variability compared with other quasars of similar X-ray luminosity.This is true for the following reasons : (1) They are intrinsically more X-ray luminous, and variability amplitude tends to drop with increasing luminosity & Mushotzky also Fig. 9 in et (Barr 1986 ; Boller al. (2) The scattering medium must be signiÐcantly 1996).larger than the X-ray source, and short-timescale variability will be averaged out.If the X-rayÈweak quasars are just due to large-amplitude intrinsic variability of the soft X-ray emission, as seen in some steep-spectrum narrow-line Seyfert 1 galaxies then one may expect the exact (°4.7), opposite behavior, i.e., these quasars may become signiÐcantly brighter at soft X-rays at some stage in the future.

SUMMARY
We deÐned a complete sample of 23 optically selected quasars that includes all the PG quasars at z ¹ 0.400, and cm~2.Pointed ROSAT PSPC obser-N H I Gal \ 1.9 ] 1020 vations were made for all quasars, yielding high S/N spectra for most objects.The high quality of the ROSAT spectra allows one to determine the best-Ðtting with about an a x order of magnitude higher precision compared with previously available X-ray spectra.In this paper we report the observations of 13 quasars not described in analyze Paper I, the correlation of the X-ray properties of the complete sample with other emission properties, determine the mean X-ray spectra of low-z quasars, and discuss the possible origin of the versus Hb FWHM correlation, the nature of a x X-rayÈweak quasars, and the physical origin of the soft X-ray emission.Our major results are the following : 1.The spectra of 22 of the 23 quasars are consistent, to within D10%È30%, with a single power-law model over the rest-frame range 0.2È2 keV.There is no evidence for signiÐcant soft excess emission with respect to the best-Ðt power law.We place a limit of D5 ] 1019 cm~2 on the amount of excess foreground absorption by cold gas in most of our quasars.The limits are D1 ] 1019 cm~2 in the two highest S/N spectra.
2. SigniÐcant X-ray absorption by partially ionized gas ("" warm absorber ÏÏ) in quasars is rather rare, occurring for of the population, which is in sharp contrast to lower [5% luminosity AGNs, where signiÐcant absorption probably occurs for D50% of the population.
3. 4. Extensive correlation analysis of the X-ray continuum emission parameters with optical emission-line parameters indicates that the strongest correlation is between and a x , the Hb FWHM.A possible explanation for this remarkably strong correlation is a dependence of on as a x L /L Edd , observed in Galactic black hole candidates.
5. There appears to be a distinct class of "" X-rayÈweak ÏÏ quasars, which form D10% of the population, where the X-ray emission is smaller by a factor of 10È30 than expected based on their luminosity at other bands, and on their Hb luminosity.
6. Thin accretion disk models cannot reproduce the observed 0.2È2 keV spectral shape, and they also cannot reproduce the tight correlation between the optical and soft X-ray emission.
7. The H I/He I ratio in the ISM at high Galactic latitudes must be within 20%, and possibly within 5%, of the total H/He ratio.
The main questions raised by this study are the following : 1. What is the true nature of X-rayÈquiet quasars ?Are these quasars indeed intrinsically X-ray weak, or are they just highly absorbed but otherwise normal quasars ?
2. What physical mechanism is maintaining the strong correlation between the opticalÈUV and the soft X-ray continuum emission, or, equivalently, maintaining a very small dispersion in the maximum possible far-UV cuto † temperature ?
3. What is the physical origin for the strong correlations between and Fe II/Hb, and the peak [O III]-to-Hb a x , L *O III+ , Ñux ratio ? 4. Is the soft X-ray emission indeed related to the presence of radio emission, or is it just a spurious relation, and the primary e †ect is related to the Hb line width ?Or, put di †erently, do RLQs and RQQs of similar Hb FWHM have similar a x ?
Extensions of the ROSAT PSPC survey described in this paper to the hard X-ray regime with ASCA and SAX, and to the UV with HST , and soft X-ray variability monitoring with the ROSAT HRI, which are currently being carried out, may provide answers to some of the questions raised above.These studies will also allow us to ( 1 FIG. 1c the power-law component are determined by the temperature and electron scattering optical depth in the corona (e.g. L /L Edd test whether soft X-ray variability is indeed strongly tied to the Hb FWHM, as expected if the Hb FWHM is an indicator of and (3) explore the relation of the UV line L /L Edd ; emission properties to the ionizing spectral shape.

TABLE 1 THE
(5) F. J. Lockman 1995, private communication.(4) Savage The 1993 January PSPC calibration matrix was used for observations made after 1991 October 14, and earlier observations were Ðtted with the 1992 March calibration matrix.The best-Ðt model parameters are obtained by s2 minimization.

TABLE 2 JOURNAL
OF ROSAT PSPC OBSERVATIONS

Table 5
III] EW, Fe II EW, Hb EW, He II EW, Table5[O III]/Hb and He II/Hb Ñux ratios, [O III] peak Ñux to Hb peak Ñux ratio, radio to optical Ñux ratio, and the Hb asymmetry, shape, and shift parameters.All these 11 parameters are listed in Table the absolute luminosities, and the two lower panels the luminosity normalized to unity at log l \ 14.25 for radio-quiet quasars and for radio-loud quasars.Note the relatively small dispersion in the normalized 0.3 keV (log l \ 16.861) luminosity.The outlying objects are labeled.PG 1626]554 is the only object where a steep is clearly associated with a strong soft excess a x (relative to the near-IR Ñux).In other objects a steep a x tends to be associated with a low 2 keV Ñux.This trend is

TABLE 4 CONTINUUM
AND EMISSION-LINE PARAMETERS Name

TABLE 5 SPEARMAN
ÈComparison of the spectral energy distribution of all 23 quasars.Upper panel : Absolute luminosity.Middle panel : Luminosity of the 19 RQQs normalized to unity at log l \ 14.25 (the longest wavelength available for all objects).Note that PG 1626]554 is the only quasar for which a steep is associated with a strong soft excess ; in all other objects a x a steep III] peak Ñux to Hb peak Ñux (as deÐned by Boroson & Green) are the emission-line parameters that correlate RLQs and RQQs are plotted in a thin solid line.They suggest that the FUV power law extends into the soft X-ray regime, with no extreme UV spectral break and no steep soft component below 0.2 keV.
with more stringent upper limits set by the lack of a Lyman limit edge, as well as the He I and the He II bound-free edges in a few very high z quasars.
Ñux-limited sample, such as the Puchnarewicz et al. sample.

Table 5
III], and the [O III] to Hb peak Ñux ratio may thus partly reÑect the FWHM ratio of these lines.Thus, this correlation may represent a correlation of the [O III] FWHM with High spectral resolution mea-a x .surements of the [O III] line proÐle are required to test this possibility.
\ 0.55M ] c for z \ 0.2 and log ' \ 0.66M ] c for 0.4 \ z \ 0.7, where c is a constant.Since our sample is restricted to z \ 0.4, we assume log ' \ 0.6M ] c.Using the relation M \ [2.5 log L ] c, we get dn/dL P L ~2.5, where n 4 dN/dV . in the expression for dn/dL , we get dn/ dk P k0.5.Thus although the disks are assumed to have a Edd both the luminosity and the velocity Ðeld in the broad-line region may not be isotropic and therefore the presence of cannot be securely deduced.
that PG 1425]267 is a RLQ, while all BALQSOs are known to be RQQs et al.
The average soft X-ray spectral slope for RQQs is and it agrees remarkably well with that RLQs are weaker than RQQs below 0.2 keV, as suggested also by the Zheng et al. mean RLQ continuum.These results suggest that there is no steep soft component below 0.2 keV.