A POPULATION OF X-RAY WEAK QUASARS: PHL 1811 ANALOGS AT HIGH REDSHIFT

We first compiled a sample of the 1621 objects classified as QSO or HIZ_QSO in the SDSS Data Release 7 (DR7; Abazajian et al. 2009) Catalog Archive Server (CAS) with m_r ⩽ 18.8 and 2.125 ⩽ z ⩽ 2.385 (see Figure 1). The magnitude limit selects objects that are sufficiently bright for short Chandra observations to provide tight constraints on their optical-to-X-ray SED properties. The selected redshift range provides the best possible coverage between Lyα and Fe ii (and usually Mg ii) in the SDSS spectra, which allows reliable identification of PHL 1811 analogs. These analogs were systematically selected as quasars having weak and blueshifted high-ionization lines, such as C iv λ1549 and Si iv λ1400,⁹ as well as strong UV Fe ii (2200–2600 Å) and Fe iii UV48 (2080 Å) emission. The criteria upon rest-frame equivalent widths (W_r) of high-ionization lines extracted from the SDSS CAS SpecObj table were W_r (Si iv) ⩽15 Å, −15 Å ⩽ W_r (C iv) ⩽20 Å, and χ² ⩽ 20 for the C iv line fit; the χ² fit-quality criterion ensures reliable measurements of C iv line properties.¹⁰ All 54 spectra meeting these criteria were then inspected visually by P.B.H. We first excluded sources that are misclassified stars or are otherwise spurious, as well as sources that have bad redshift measurements or detectable UV broad absorption lines (BALs). BAL quasars were excluded because they typically show substantial X-ray absorption (e.g., Gallagher et al. 2002, 2006; Gibson et al. 2009b). After this examination we had 32 sources that are non-BAL weak-line quasars (WLQs; see Section 4.6 for further details). Among these 32 sources, objects with strong UV Fe ii and Fe iii UV48 emission and whose high-ionization lines showed a strong blueshifted component (with the magnitude of the blueshift exceeding 800 km s⁻¹) were kept in the sample. We rejected 13 objects primarily due to weak Fe emission, and 8 objects primarily due to a low or unknown blueshift. After doing this, 11 quasars remained in our sample.¹¹ Figure 2 shows the SDSS spectra for these potential PHL 1811 analogs. The SDSS quasar composite spectrum from Vanden Berk et al. (2001) and the spectrum for PHL 1811 from the Hubble Space Telescope (HST; Leighly et al. 2007a) are also included in Figure 2. Nine of the 11 objects are radio-quiet (R < 10; see Section 3 for the definition of R). J1454+0324 and J1618+0704 are radio-intermediate with R = 12.8 and R = 35.0, respectively. Three of the 11 objects already had sensitive archival X-ray coverage (J1219+1244 and J1521+5202 by Chandra, J1230+2049 by XMM-Newton). We proposed short (5–13 ks) Chandra observations for the other eight objects, and were awarded observing time in Cycle 11 (see Table 1 for an X-ray observation log).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The redshifts and their uncertainties (see Table 2) for our PHL 1811 analogs are the spectro1d redshifts from the SDSS CAS in most cases. These redshifts were obtained by fitting the peaks in the SDSS spectra with a set of strong quasar emission lines (see Section 4.2.10.1 of Stoughton et al. 2002 for details). The SDSS CAS redshift measurements may not be precise for our sample because of the weak, blueshifted, and asymmetric emission lines. We therefore examined the redshift values for our PHL 1811 analogs carefully. For four quasars whose CAS redshifts were suspect based on this examination, we adopt redshifts and uncertainties from other approaches: for J1125+4028, the redshift of the longer-wavelength peak of the double-peaked Mg ii emission is adopted as systemic; for both J1219+1244 and J1230+2049, we adopt the redshift of a strong (but not broad) absorption system; for J1521+5202, we follow Just et al. (2007) and adopt z = 2.19 based on Mg ii emission. The differences between these updated redshift values and the CAS values have a magnitude of Δz = 0.02–0.05. The final redshift values adopted are listed as z_q in Table 2. Redshifts of narrow absorption-line systems (NALs) are also listed in Table 2. Three PHL 1811 analogs (J0848+5408, J1219+1244, and J1230+2049) have associated NALs (defined as |v| < 5000 km s⁻¹), which are labeled in Figure 2 and in Table 2. The other NALs with v < −5000 km s⁻¹ are most likely intervening absorption.

Hewett & Wild (2010) reported improved redshift measurements with lower systematic uncertainties for SDSS quasars. Their measurements for our candidate PHL 1811 analogs are also included in Table 2, denoted as z_HW. These redshifts are either consistent with or slightly lower than z_q. However, the z_HW values were determined via cross-correlation with quasar spectral templates. Given the unusual spectral properties of our candidate PHL 1811 analogs, we have not adopted the z_HW values. Nevertheless, we will test whether adopting these different redshift values would affect our main conclusions in the following sections, particularly regarding C iv blueshifts.

We measured emission-line properties interactively for C iv, Si iv, the λ1900 complex,¹² and Fe iii UV48 (Table 3), since the automatic measurements stored in the CAS may be unreliable because of the weak, highly blueshifted lines of PHL 1811 analogs. We first fitted the continuum for each source following the method of Vanden Berk et al. (2001). The SDSS spectra were smoothed with a five-pixel sliding-box filter. Regions of strong narrow absorption were interpolated across manually. The fitting region for each emission line was then defined between lower and upper wavelength limits λ_lo and λ_hi (see Table 2 of Vanden Berk et al. 2001). A power-law local continuum was fit to the lower and upper 10% of the wavelength region between λ_lo and λ_hi. After subtracting the continuum, we measured W_r, FWHM, the line dispersion (σ_line, the second moment relative to the flux-weighted mean wavelength of the line), and the associated uncertainties for each line. Blueshifts were calculated between the lab wavelength of a line in the quasar rest frame (see Table 2 of Vanden Berk et al. 2001) and the observed mode of all pixels with heights greater than 50% of the peak height of this line, where mode = 3 × median − 2 × mean. In addition to the C iv blueshift errors quoted in Table 3, the uncertainties in the redshift measurements also produce further uncertainties for the C iv blueshift. We also include corresponding measurements of the spectrum of PHL 1811 (Leighly et al. 2007a), the composite quasar spectrum of Vanden Berk et al. (2001), the "B4" composite spectrum of high-redshift (2.06 < z < 3.33), high-luminosity (−28 < M_i < −26) quasars of Yip et al. (2004), and the spectrum of PHL 1092 (Leighly et al. 2007a) in Table 3 for comparison.

Using our refined W_r values, all our targets and PHL 1811 have W_r(C iv) < 11 Å, W_r(Si iv) < 8 Å, 3.5 Å < W_r(λ1900 Å) <12.5 Å, and 2.5 Å < W_r(Fe iii) <8 Å. Our targets have significantly weaker C iv and Si iv lines than the required selection criteria. For PHL 1811 analogs, the first three of these lines have average W_r values that are only ≈40% of the average for normal quasars of similar redshift and luminosity. Therefore, the sources in our sample have weak high-ionization lines and semi-forbidden lines like PHL 1811. In contrast, Fe iii UV48 has a 50% larger average W_r in PHL 1811 analogs. As compared to PHL 1811, our PHL 1811 analogs have lines that are just as weak but are typically broader (perhaps partly because our sources are more luminous and thus likely have SMBHs of higher mass; e.g., Laor 2000) and more blueshifted.

Some key emission lines in the rest-frame optical band, e.g., Hα, Hβ, and [O iii] λ5007, are not covered by the SDSS spectra of our PHL 1811 analogs due to their high redshifts. PHL 1811 has a narrow-line type 1 optical spectrum as well as very weak [O iii] emission compared to typical type 1 quasars (Leighly et al. 2007a). Near-infrared spectroscopy of our PHL 1811 analogs is needed to measure the properties of these emission lines. We report near-infrared spectroscopy of our most luminous PHL 1811 analog, J1521+5202, in the Appendix of this work.

One radio-quiet source in our sample, J0903+0708, shows notably different X-ray properties from the other radio-quiet sources; it is much brighter in X-rays (see Section 3). After the Chandra observations, we observed this source with the Hobby–Eberly Telescope (HET; Ramsey et al. 1998) because it was a potential outlier. We obtained spectroscopy for J0903+0708 using the G2 grating and the 1'' slit of the Low-Resolution Spectrograph (Hill et al. 1998) on the HET on 2010 November 5. Two HET spectra with spectral resolutions R ≈ 1000 were extracted, each of which has an exposure time of 600 s. The spectra were calibrated to a standard star (G191 B2B). Figure 3 shows the co-added HET spectrum and a comparison to the SDSS spectrum. The rest-frame wavelength range of the HET spectra is 1350–2250 Å. The rest-frame UV emission-line measurements of the co-added HET spectrum are also listed in Table 3. The strengths of the emission lines (C iv, Si iv, C iii], and Fe iii) in the HET spectrum agree with those of the SDSS spectrum, while the blueshift of the C iv line in the HET spectrum is −610 ± 300 km s⁻¹, different at a 2.5σ level from the SDSS measurement (−2300 ± 600 km s⁻¹). In order to test for any possible variation of the C iv line, we performed a χ² test on the C iv region (1486–1574 Å) of the two spectra. These two spectra are consistent with each other within the noise (χ² = 132.3 for 146 spectral bins). The HET spectrum has a much higher signal-to-noise ratio than the SDSS spectrum (see the inset of Figure 3 for the C iv line region), which indicates that the SDSS blueshift measurement was probably erroneous due to the SDSS spectrum's limited signal-to-noise ratio. With the improved HET data, J0903+0708 would not have passed our original selection criterion of C iv blueshift for PHL 1811 analogs (see Section 2.1), and thus it should not be included in our sample. Note that J0903+0708 was one of the lowest C iv blueshift sources among our candidate PHL 1811 analogs even with only the SDSS measurements. J0903+0708 is probably a regular WLQ (see Section 4.6 for further discussions on the relation between PHL 1811 analogs and WLQs). We will therefore exclude this source from discussion hereafter unless noted. Nevertheless, we will report the X-ray properties of J0903+0708 in Section 3. Our final sample of PHL 1811 analogs includes 10 quasars; 8 are radio quiet, while 2 are radio intermediate.

Zoom In Zoom Out Reset image size

We also estimated the spectral index of the presumed power-law continuum in the rest-frame 1200–2800 Å range (α_ν, where $f_\nu \propto \nu ^{\alpha _\nu }$) for each source by fitting the SDSS spectra in four line-free regions using the method described in Section 2.1 of Gibson et al. (2008b). The results are listed in Table 3. The range of spectral indices for our candidate PHL 1811 analogs is −0.97 to −0.28 with a mean value of 〈α_ν〉 = −0.61. This value is within the range seen for typical quasars, although it is somewhat steeper (i.e., redder) than the "canonical" value of α_ν = −0.5 the optical and UV region because of the presence of the "small blue bump" due to the Balmer continuum and Fe emission (see discussion in Section 2.1 of Strateva et al. 2005; also see Natali et al. 1998; Schneider et al. 2001; Vanden Berk et al. 2001).

The somewhat redder than average nature of our quasars is evidenced by their Δ(g − i) values (the g − i color minus the average g − i color for quasars at the same redshift) taken from Schneider et al. (2010). On average, our sample is redder than ≈75% of other quasars at the same redshift, with 〈Δ(g − i)〉 = 0.28. About a quarter of this excess can be attributed to weak C iv emission in the g band and about a quarter to strong Fe ii emission in the i band (see Figure 2). The remaining half corresponds to the redder continuum expected for quasars with spectral index α_ν = −0.61 instead of α_ν = −0.5.

In spite of their somewhat redder colors, we do not expect substantial incompleteness in the SDSS selection of PHL 1811 analogs. Our PHL 1811 analogs lie within the loci of u − g, g − r, r − i, and i − z colors for SDSS color-selected quasars at similar redshifts and luminosities, but they are somewhat offset from the regions with the highest density of quasars (see Figure 4). They are within the inclusion region of (u − g) < 0.6 for SDSS color selection of quasar candidates (see Section 3.5.2 of Richards et al. 2002). Our objects are also in a redshift range where the SDSS color selection algorithm has high completeness (≈75%; see Figure 6 of Richards et al. 2006).

Zoom In Zoom Out Reset image size

The Δα_ox parameter (see Column 12 of Table 5) quantifies the relative X-ray brightness of a quasar with respect to a typical radio-quiet quasar with the same UV luminosity. Figure 2 shows the SDSS spectra of our PHL 1811 analogs (and also J0903+0708; see Section 2.3) ordered by Δα_ox. No strong trends are apparent among the PHL 1811 analogs between Δα_ox and the UV emission-line characteristics (see Section 4.5 for quantitative correlation analyses). Two radio-intermediate sources in our sample¹⁵ are X-ray bright (see further discussion in Section 4.7). The distribution of Δα_ox values for the eight radio-quiet PHL 1811 analogs is shown in Figure 5(a), compared to that for the 132 radio-quiet, non-BAL quasars in Sample B of Gibson et al. (2008a).¹⁶ The Sample B quasars were all serendipitously detected by Chandra (with off-axis angles greater than 1'). They are representative of typical SDSS quasars in the redshift range of 1.7 ⩽ z ⩽ 2.7. We use a Peto–Prentice test (e.g., Latta 1981), implemented in the Astronomy Survival Analysis (ASURV) package (e.g., Lavalley et al. 1992), to assess whether these two samples follow the same distribution. The Peto–Prentice test is our preferred approach because it is the least affected by the factors of unequal sample sizes or different censoring patterns that exist in our case (e.g., Latta 1981). The distribution of Δα_ox values for our sample is found to be significantly different from that for typical SDSS quasars; the probability of the null hypothesis (the two samples following the same distribution) is reported to be 1 × 10⁻²⁸ by the Peto–Prentice test. Other two-sample tests for censored data (e.g., Gehan, logrank, and Peto–Peto) give similar results (see Table 6). The two-sample test results do not change materially if J0903+0708 (see Section 2.3) is included (also see Table 6).

Zoom In Zoom Out Reset image size

The mean value of Δα_ox for radio-quiet PHL 1811 analogs in our sample is −0.423 ± 0.049, which is calculated using the Kaplan–Meier estimator¹⁷ (also implemented in the ASURV package), while the mean Δα_ox value for radio-quiet, non-BAL quasars in Sample B of Gibson et al. (2008a) is −0.001 ± 0.011. Therefore, radio-quiet PHL 1811 analogs in our sample are anomalously X-ray weak by a mean factor of ≈13. If J0903+0708 is included, the mean value of Δα_ox changes to −0.384  ±  0.057, corresponding to a mean X-ray weakness factor of ≈10. All of our radio-quiet PHL 1811 analogs are X-ray weak by factors >4.8; we only have lower limits on the range of X-ray weakness factors because four of the PHL 1811 analogs are undetected in X-rays. Our X-ray detected sources show that the X-ray weakness factors span at least the range from 8.7 to 34.5.

For the three radio-quiet PHL 1811 analogs not detected by Chandra (J0848+5408, J1219+1244, and J1230+3320), we performed a stacking analysis to obtain an average constraint on their relative X-ray weakness. We added the total counts and background counts of the three sources in their apertures for the full band, soft band, and hard band. For each band, we then calculated the Poisson probability of the total counts arising from a fluctuation of the corresponding background counts. If this probability is less than 0.1%, we take the stacked source to be detected in that band. The stacked source is detected in the full band, with 4.42 total counts and 0.35 background counts; the Poisson probability is 1 × 10⁻⁴. The stacked source is not detected in the soft or hard band, so a band-ratio analysis is not possible. The combined effective exposure time for the stacked source is 23.3 ks. After an aperture correction was applied, the average full-band count rate of the stacked source is 1.84 × 10⁻⁴ counts s⁻¹. We adopted the average values of the parameters for the three sources, such as Galactic N_H, z, and f_2500 Å, to calculate the X-ray properties of the stacked source. Under the assumption of a Galactic-absorbed power-law model with Γ = 2, the average X-ray flux density at rest-frame 2 keV is 2.90 × 10⁻³³ erg cm⁻² s⁻¹ Hz⁻¹, while the average rest-frame X-ray luminosity from 2–10 keV is 2.75 × 10⁴³ erg s⁻¹. The average α_ox value for the stacked source is −2.26. The Δα_ox value is −0.57, corresponding to an X-ray weakness factor of 30.5 (cf. Figure 5). At least on average, these undetected PHL 1811 analogs appear to be only a few times X-ray weaker than those we have detected individually. PHL 1811 itself has similar X-ray weakness (≈30–100) to these undetected PHL 1811 analogs on average (as does J1521+5202).

We also performed a stacking analysis of the four radio-quiet PHL 1811 analogs that are detected by Chandra (J0000+1356, J0946+2744, J1125+5028, and J1521+5202) to constrain their average X-ray spectral properties. We added the counts for each source in soft band and hard band. The total background-subtracted counts are 12.9^+4.7 _{− 3.5} in the soft band and 7.7^+3.9 _{− 2.7} in the hard band. The band ratio is 0.60^+0.38 _{− 0.26}, corresponding to an effective photon index of Γ = 1.22^+0.52 _{− 0.45} using the average Galactic neutral hydrogen column density of these four sources. This photon-index value suggests a harder average X-ray spectrum for the X-ray detected, radio-quiet PHL 1811 analogs than that for typical radio-quiet quasars. This average X-ray spectrum is also harder than the X-ray spectrum of PHL 1811 (Γ ≈ 2.2; Leighly et al. 2007b). However, even the quoted 68% confidence error bars are large due to limited counts. If quoting 90% confidence error bars, the photon index is Γ = 1.22^+0.85 _{− 0.66}. If we furthermore add the counts from the three PHL 1811 analogs undetected by Chandra (see the previous paragraph), the average effective photon index becomes Γ = 1.10^+0.45 _{− 0.40} (68% confidence level), or 1.10^+0.77 _{− 0.59} (90% confidence level).

Our sample selection and observational results allow us to investigate the demographics of PHL 1811 analogs in the total quasar population. Our PHL 1811 analogs were selected from high-redshift (2.125 ⩽ z ⩽ 2.385), optically bright (m_r ⩽ 18.8) objects classified as QSO or HIZ_QSO in the SDSS DR7 CAS (see Section 2.1). There are 1358 radio-quiet, non-BAL SDSS DR7 quasars within the above redshift and magnitude ranges. Our sample of PHL 1811 analogs contains eight radio-quiet and two radio-intermediate sources. The radio-quiet PHL 1811 analogs thus appear to be only 8/1358 = (0.6^+0.5 _{− 0.3})% (at the 90% confidence level) of the total radio-quiet, non-BAL quasar population. Considering the potential incompleteness of our sample selection, we take 0.3% as a lower limit on the fraction of PHL 1811 analogs in the radio-quiet, non-BAL quasar population. On the other hand, Figure 5(b) shows upper limits on the fraction of optically selected, radio-quiet, non-BAL quasars compiled from samples using SDSS only the SDSS+BQS (Bright Quasar Survey; Schmidt & Green 1983) that are X-ray weak by a given factor (adapted from Figure 5 of Gibson et al. 2008a). Only ⩽2.8% (at the 90% confidence level) of those quasars are as X-ray weak (Δα_ox=−0.427; see Section 4.1) as the radio-quiet sources in our sample. In summary, the fraction of radio-quiet PHL 1811 analogs should be within the range of 0.3%–2.8%. The upper limit of 2.8% is obtained solely based on the factor of X-ray weakness. Since our selection of PHL 1811 analogs should not have strong incompleteness (a factor of ≲ 2), we estimate the fraction of PHL 1811 analogs in the total quasar population to be ≲ 1.2%.

Radio-quiet PHL 1811 analogs, which are all X-ray weak in our sample, only compose a small fraction of the total quasar population. They do not appear to present fractionally significant difficulties to the utility of X-ray surveys for finding AGNs throughout the universe (see Section 1). The majority (≳ 80%) of BAL quasars are also X-ray weak owing to intrinsic X-ray absorption (e.g., Gibson et al. 2009b). Considering the fraction of BAL quasars (≈15%–20%; e.g., Gibson et al. 2009b) found in SDSS quasar samples, X-ray weak BAL quasars outnumber X-ray weak PHL 1811 analogs by a factor of ≳ 10.

Our radio-quiet PHL 1811 analogs could be X-ray weak either because they are intrinsically X-ray weak, perhaps due to quenching of the ADC (see Section 1), or because they are X-ray absorbed. In this section, we will discuss the strengths and weaknesses of each of these hypotheses.

There are three indirect arguments supporting the intrinsic X-ray weakness scenario. First, as discussed in Section 2.1, we have selected against objects with detectable broad C iv absorption (or any other broad absorption lines, including Si iv and Mg ii), since most heavily X-ray absorbed type 1 quasars show broad C iv absorption (e.g., Brandt et al. 2000; Gibson et al. 2008a; Page et al. 2010). If we assume the PHL 1811 analogs have underlying X-ray spectra with Γ = 2, the level of intrinsic neutral absorption would need to be N_H ≈ 5 × 10²³ cm⁻² (≳ 2 × 10²³ cm⁻² to ≳ 8 × 10²³ cm⁻²) at z ≈ 2.2 to generate the observed X-ray weakness factor of 13 (and the range of >4.76 to ⩾34.5). Three of our eight radio-quiet PHL 1811 analogs (J0848+5408, J1219+1244, and J1230+2049) have narrow associated C iv absorption lines (within ±5000 km s⁻¹ of the quasar redshift). Quasars with associated NALs generally have similar X-ray properties to unabsorbed quasars and are not heavily X-ray absorbed (e.g., Misawa et al. 2008; Chartas et al. 2009). The intrinsic X-ray absorption for quasars with associated NAL systems is typically N_H ≲ 10²² cm⁻², much lower than the level required for our radio-quiet PHL 1811 analogs. The frequency of associated NAL systems in our sample (3/8 = 38%) is consistent with that of typical quasars (see Section 3.2.1 of Ganguly & Brotherton 2008). For the total number of NAL systems (both associated and intervening), the work of Nestor et al. (2008) predicts that between Lyα and C iv in the spectra of our eight radio-quiet PHL 1811 analogs we should find 16 C iv absorbers with FWHM < 600 km s⁻¹ and W_r(λ1548) ⩽ −0.3 Å. In fact, we find 10. Within the limitations of small-number statistics, there is nothing unusual about the total number of NAL systems in our PHL 1811 analogs. One source, J1219+1244, has a C iv absorber with a total velocity width of 1100 km s⁻¹, just above the 1000 km s⁻¹ threshold to be termed a mini-BAL quasar. Mini-BAL quasars are also generally not strongly X-ray absorbed (e.g., Gibson et al. 2009a; Wu et al. 2010a). The intrinsic X-ray absorption of mini-BAL quasars is also typically at the level of N_H ≲ 10²² cm⁻² (Wu et al. 2010a). While it is perhaps possible that a type 1 quasar could be strongly X-ray absorbed without showing notable UV absorption lines (see below), the available empirical evidence indicates that our removal of quasars with BALs should eliminate the vast majority of strongly X-ray absorbed quasars.

Second, the quasars in our sample show the blue UV/optical continua typical of type 1 quasar spectra. Our sample was chosen to avoid objects with detectable spectral curvature indicative of strong dust reddening. Nonetheless, the somewhat redder than average colors of our sample (see Section 2.4) suggest that we cannot rule out an average dust reddening of up to E(B − V) ≈ 0.05 magnitudes. For an SMC dust-to-gas ratio (Bouchet et al. 1985), such reddening would imply a column density of N_H ≈ 2.2 × 10²¹ cm⁻². Again, this column density is ≳ 100 times too small to explain the X-ray weakness of our radio-quiet PHL 1811 analogs. Note, however, that this is not a strong argument against X-ray absorption, since many BAL quasars (which usually have heavy X-ray absorption) also have blue UV/optical continua (e.g., Trump et al. 2006).

Third, the unusual UV emission-line properties of PHL 1811 can be explained naturally by an intrinsically X-ray weak SED (Leighly et al. 2007a). Such a continuum lacks adequate photons to create highly ionized ions, which results in weak high-ionization lines (e.g., C iv, Si iv). The temperature of the gas is too low to excite semi-forbidden lines (e.g., C iii]). The strong Fe ii and Fe iii emission can also be qualitatively explained by the PHL 1811-like SED (Leighly et al. 2007a). Since our PHL 1811 analogs have the same unusual UV emission-line properties, Occam's razor would suggest their SEDs are also probably intrinsically X-ray weak.

In spite of the indirect arguments above, it is notable that the average X-ray spectrum for our radio-quiet PHL 1811 analogs appears to be harder than for typical radio-quiet quasars and for PHL 1811 itself (see Section 4.2), which hints at the presence of heavy X-ray absorption.¹⁸ It is possible that X-ray absorption could also lead to the X-ray weak SEDs required to create the unusual UV emission lines of our PHL 1811 analogs. However, if heavy X-ray absorption is responsible, it must lie closer to the SMBH than the BELR does, so that this emission-line region "sees" a soft ionizing SED and produces a PHL 1811-like UV emission-line spectrum. It must also have a remarkable lack of accompanying C iv, Si iv, or Mg ii BALs (or even mini-BALs in all but one case). These requirements might be achieved by a line of sight that passes through "shielding gas" but not a typical UV absorbing wind (e.g., Gallagher et al. 2005). The shielding gas, which resides interior to the UV absorbing wind, is sufficiently highly ionized that it is transparent to UV photons but absorbs soft X-ray photons, preventing overionization of the wind (e.g., Murray et al. 1995). To explain PHL 1811 analogs, the shielding gas could have an atypically larger covering factor of the X-ray continuum source that prevents most of the X-rays from reaching the BELR; we return to this idea in Section 4.6. The shielding gas might betray its presence via very highly ionized absorption, either in the UV or in X-rays. For example, Telfer et al. (1998) report that the X-ray weak quasar SBS 1542+541 has BAL troughs much stronger in O vi and Ne viii than in N v or C iv (see their Figure 5), and O vi and Ne viii have been seen in narrower absorption systems as well (e.g., Scott et al. 2004).

In fact, PHL 1811 has a narrow O vi absorber at z = 0.192186 (consistent with the Balmer-line redshift of PHL 1811 in Leighly et al. 2007a) in which the O vi column density is ∼100 times the H i column density (Tripp et al. 2008). Our inspection of the HST Space Telescope Imaging Spectrograph (STIS) spectrum of PHL 1811 also reveals N v absorption at the same redshift, with W_r an order of magnitude smaller than that of O vi; C iv absorption is not seen in a lower-resolution STIS spectrum. Associated O vi absorption is not rare, being seen in about 23% (Frank et al. 2010) to 63% (Tripp et al. 2008) of quasars. However, the z = 0.192186 absorber in PHL 1811 is unusual, having the highest O vi column density and the highest O vi/H i ratio among the 14 associated O vi absorbers in Tripp et al. (2008). If this unusual O vi absorber is connected with the unusual X-ray and UV properties of PHL 1811, future spectroscopy should reveal unusual associated O vi absorption in our PHL 1811 analogs as well.

In summary, the currently available data are unable to discriminate rigorously between the intrinsic X-ray weakness and X-ray absorption scenarios. Occam's razor might initially suggest favoring the hypothesis that our PHL 1811 analogs are intrinsically X-ray weak like PHL 1811 itself appears to be. However, the suggestion of heavy X-ray absorption from the likely hard X-ray spectra definitely needs further investigation. If our objects are indeed established to be heavily X-ray absorbed, then Occam's razor might alternatively be used to motivate reconsideration of the evidence that PHL 1811 is intrinsically X-ray weak. The evidence for intrinsic X-ray weakness presented by Leighly et al. (2007b) appears strong (e.g., steep X-ray spectrum with no apparent photoelectric absorption cutoff, X-ray variability). Nevertheless, absorption effects in local AGNs are sometimes observed to be exceedingly complex, and perhaps some unusual absorbers or special absorption geometry for this unusual quasar might yet be consistent with the observed X-ray properties.

We have quantitatively searched for relations between relative X-ray weakness and the unusual UV emission-line properties. Correlation analyses were performed between Δα_ox and the UV emission-line parameters listed in Table 3, using both Kendall's τ test and Spearman's rank-order analysis in the ASURV package. Kendall's τ test is preferred over Spearman's rank-order analysis in small-sample cases, where the sample size is N ⩽ 30.¹⁹ The results are presented in Table 7. We found only a marginal correlation (94.9% correlation probability) between Δα_ox and the blueshift of the C iv line when including both radio-quiet and radio-intermediate PHL 1811 analogs in our sample. The effective significance of this correlation is further weakened by the number of trials, since we have tested for multiple correlations in Table 7. No other correlations were found; this is likely due to the small sample size and the limited range of parameter space spanned by the PHL 1811 analogs alone. If we use the redshift values of Hewett & Wild (2010) for our PHL 1811 analogs, the correlation between Δα_ox and the C iv blueshift remains insignificant (91.2% correlation probability).

Figure 6 shows Δα_ox plotted against the UV emission-line parameters of W_r(C iv), W_r(λ1900 Å), and C iv blueshift for radio-quiet PHL 1811 analogs, PHL 1811 itself, and the radio-quiet, non-BAL quasars in Sample B of Gibson et al. (2008a). The addition of the Sample B quasars and PHL 1811 increases the sample size to N = 148 and broadens the parameter space of line measurements. The values of W_r(C iv) and C iv blueshift for the quasars in Sample B are taken from Shen et al. (2011); we modified the C iv blueshift values for the Sample B quasars to account for the improved redshift measurements of Hewett & Wild (2010). The values of W_r(λ1900 Å) are taken from Gibson et al. (2008a). PHL 1092 (see Section 1) also has PHL 1811-like UV emission-line properties and shows dramatic X-ray variability.²⁰ We include this source in Figures 6 and 7, but not in the correlation analyses.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

A significant correlation between W_r(C iv) and Δα_ox was reported by Gibson et al. (2008a) for their sample. After adding our radio-quiet PHL 1811 analogs and PHL 1811, the correlation becomes more significant (see Table 7 and the red solid line in Figure 6(a)). However, all of the radio-quiet PHL 1811 analogs and PHL 1811 itself lie well below the best-fit linear correlation (in log space) in Gibson et al. (2008a) between W_r(C iv) and Δα_ox. This remains true even if we consider the 95% confidence uncertainty range shown in Figure 6(a), obtained using a nonparametric bootstrap method (Efron 1979). This deviation from the correlation found for typical radio-quiet quasars may indicate that the radio-quiet PHL 1811 analogs arise from a distinct population of typical radio-quiet quasars. There is no significant correlation between W_r(λ1900 Å) and Δα_ox (see Table 7 and Figure 6(b)).

The C iv emission lines of quasars are generally known to have systematic blueshifts (e.g., Gaskell 1982; Richards et al. 2011). We performed Spearman rank-order analysis and found a correlation between C iv blueshift and Δα_ox for radio-quiet PHL 1811 analogs, PHL 1811, and the Sample B quasars. However, this correlation shows substantial scatter (see Figure 6(c)). Although the correlation probability (>99.99%) is formally high, the correlation coefficient (r_S = 0.39) shows that this correlation is weaker than that between W_r(C iv) and Δα_ox (r_S = 0.49). Furthermore, it is difficult to determine if there is truly a single correlation or instead two populations artificially producing an apparent correlation.

The red triangles in Figure 6(c) show the four radio-quiet, non-BAL quasars from Gallagher et al. (2005) which have "large" C iv blueshifts compared to most typical radio-quiet, non-BAL quasars. Their X-ray brightnesses appear to be normal. However, the magnitudes of their C iv blueshifts are ⩽2000 km s⁻¹, while all but one of the radio-quiet PHL 1811 analogs in our sample have C iv blueshift magnitudes >2000 km s⁻¹. The four sources from Gallagher et al. (2005) are more similar to the typical radio-quiet, non-BAL quasars. Gallagher et al. (2005) also suggested evidence for intrinsic absorption (at the level of N_H ≈ 10²² cm⁻²) for their "large" C iv blueshift objects via joint X-ray spectral analyses. In order to search for any possible trend between intrinsic absorption and the C iv blueshift, we performed a band-ratio analysis for all the Sample B quasars from Gibson et al. (2008a). A harder X-ray spectrum (with smaller effective photon index) could indicate the presence of intrinsic absorption. However, no correlation was found between the effective photon indices and C iv blueshifts.

In Figure 7, the C iv blueshift is plotted against W_r(C iv) for our radio-quiet PHL 1811 analogs, PHL 1811, and the Sample B quasars in Gibson et al. (2008a). The color for each source shows its Δα_ox value, indicating its relative X-ray brightness. The gray dots represent the 13,582 radio-quiet Sample A quasars from Richards et al. (2011) which do not generally have constraining information on their X-ray properties. Motivated by much past work, Richards et al. (2011) described a BELR model with both "disk" and "wind" components. The C iv blueshift and W_r(C iv) show the tradeoff between these two components, which ultimately depends upon the shape of the ionizing continuum. The claimed interpretation is that quasar BELRs change from disk-dominated (with a more ionizing SED) to wind-dominated (with a less ionizing SED) as a quasar's location moves from the upper right corner to the lower left corner of Figure 7. The relative X-ray brightness becomes moderately weaker for the Sample B quasars in Gibson et al. (2008a) following this trend. Our radio-quiet PHL 1811 analogs show extreme behavior in the Δα_ox−C iv blueshift− W_r(C iv) parameter space; note they are essentially disjoint from the main quasar population. This suggests they may have extreme wind-dominated BELRs.

The SDSS has discovered ∼80 high-redshift (z > 2.2) quasars with extremely weak or undetectable UV emission lines (WLQs; e.g., Diamond-Stanic et al. 2009); similar objects also likely exist at lower redshifts (e.g., Plotkin et al. 2010a). In this section, we discuss the observational relation between our PHL 1811 analogs and WLQs more generally. WLQs are defined as quasars with W_r (Lyα + N v) < 15.4 Å, measured between 1160 Å and 1290 Å in the rest frame (Diamond-Stanic et al. 2009). Measurements of Lyα W_r are difficult with our wavelength coverage, but many of our PHL 1811 analogs likely meet the WLQ definition; for example, J0946+2744 was identified as a WLQ by Shemmer et al. (2009).

There has been some past speculation about a possible connection between PHL 1811 itself and WLQs (e.g., Leighly et al. 2007a; Shemmer et al. 2009), but we here have the advantage of having a carefully selected sample of PHL 1811 analogs with established basic X-ray properties. Figure 8 compares the median composite spectrum of radio-quiet PHL 1811 analogs with the mean spectra of WLQs from Shemmer et al. (2009) and Diamond-Stanic et al. (2009).²¹ PHL 1811 analogs and WLQs both have weak Lyα, Si iv, and C iv emission lines. PHL 1811 analogs have prominent UV Fe ii and Fe iii emission, while most WLQ samples have been selected at high redshift and generally lacked spectral coverage of UV Fe emission. However, inspection of the lower-redshift WLQ samples of Plotkin et al. (2010a, Plotkin et al. 2010b) shows that their UV Fe emission can be just as prominent relative to the broad lines as it is in PHL 1811 analogs. Furthermore, the two WLQs studied by Shemmer et al. (2010) both have prominent optical Fe ii, similar to PHL 1811 itself (Leighly et al. 2007a).

Zoom In Zoom Out Reset image size

When defining our sample of PHL 1811 analogs (see Section 2.1), we first selected 54 objects with weak high-ionization lines, 32 of which can be considered non-BAL WLQs. Twenty-nine of the 32 non-BAL WLQs are radio quiet. Eight of these 29 sources having strong UV Fe emission and C iv blueshifts were retained as our radio-quiet PHL 1811 analogs. At least from an observational point of view, PHL 1811 analogs thus appear to be a subset (≃ 30%) of WLQs. Our radio-quiet PHL 1811 analogs are X-ray weak compared to typical quasars, while WLQs are generally not X-ray weak as a population (Shemmer et al. 2009). However, there are a few known X-ray weak WLQs (e.g., J1302+0030, J1421+3433, and perhaps J1335+3533 and J1532−0039 in Shemmer et al. 2006, 2009) that may be PHL 1811 analogs. We do not have reliable measurements of the C iv blueshifts for these sources because of either the wavelength coverage or the quality of their SDSS spectra. Near-infrared spectroscopy for many WLQs (e.g., Shemmer et al. 2010) is required to assess if the X-ray weak WLQs have notably strong UV Fe ii and Fe iii emission.

While the observational relation between PHL 1811 analogs and WLQs appears fairly simple, as described above, any physical relation between these two classes may be much more complex. The X-ray weakness and distinctive emission lines of PHL 1811 analogs may ultimately be due to some extreme physical parameter (or parameters), such as accretion rate (see Section 1). In contrast, at least some WLQs appear simply to have anemic broad line regions (Shemmer et al. 2010), and some may be in a specific evolutionary stage where the quasar activity has only recently begun (Hryniewicz et al. 2010).

However, there is another scenario that might unify PHL 1811 analogs and most WLQs, which we illustrate in Figure 9. As mentioned in Section 4.4, suppose there is a subset of quasars in which the shielding gas covers all or most of the BELR (which likely consists of "clouds"), but little more than the BELR. A sufficient column of such gas could absorb X-ray and other ionizing photons before they reach the BELR, resulting in weak broad-line emission and wind acceleration without overionization. (In normal quasars with strong broad-line emission, a range of shielding gas covering factors and columns are likely to exist, but the majority of the BELR gas cannot be shielded by such high columns and must be exposed to the ionizing continuum.) When such quasars are observed through the shielding gas, a PHL 1811 analog can be seen; when they are observed from other directions (where shielding gas is unlikely to be seen), an X-ray normal WLQ can be seen. A PHL 1811 analog fraction of 30% among WLQs is consistent with the estimated BELR covering fraction (e.g., Maiolino et al. 2001). The low C iv W_r values in PHL 1811 analogs require that ≳ 80% ± 20% of the BELR be covered by shielding gas.²² As coverage of randomly distributed BELR clouds by shielding gas with the same covering factor would be physically implausible, this scenario requires the BELR in PHL 1811 analogs and WLQs to have a non-random geometrical distribution (see, e.g., Section 4.3 of Risaliti et al. 2011 and references therein). The large C iv blueshifts of PHL 1811 analogs could be produced in lines of sight into an accelerating wind, for example, from a rotating, disk-like BELR (Murray et al. 1995).

Zoom In Zoom Out Reset image size

This simple unification scenario makes several predictions. The O vi absorption in PHL 1811 (see Section 4.4) is evidence for the existence of highly ionized UV-absorbing gas along our sightline to its central engine; PHL 1811 analogs might have an excess of such absorption, but WLQs should not. Broad emission line W_r values in WLQs and PHL 1811 analogs should be comparable, consistent with the limited data available to date (e.g., the Appendix and Shemmer et al. 2010). Narrow-line emission could be seen in WLQs and PHL 1811 analogs in this model, but is unlikely given the observed anti-correlation (Boroson & Green 1992; Netzer et al. 2004) of [O iii] strength with the Fe ii/Hβ ratio,²³ which is large in PHL 1811 (Leighly et al. 2007a), J1521+5202 (see the Appendix), and at least two WLQs (Shemmer et al. 2010). Additional observations to test these predictions are needed.

The two radio-intermediate sources in our sample, J1454+0324 and J1618+0704, are X-ray bright with Δα_ox=0.42 and Δα_ox=0.28, respectively. Radio-intermediate sources probably have significant relativistic jets, which often have associated X-ray emission (e.g., Worrall et al. 1987; Miller et al. 2011). Figure 10 compares the X-ray "excesses" in these two PHL 1811 analogs (assessed with Δα_ox) with those of typical radio-intermediate quasars from the primary sample of Miller et al. (2011). The Δα_ox values of our two radio-intermediate PHL 1811 analogs are relatively large compared to those of typical radio-intermediate quasars with similar radio loudness values (R = 10–50), and also compared to the whole radio-intermediate quasar sample (R = 10–100; also see Figure 7 of Miller et al. 2011). Note that the radio-loudness values in Miller et al. (2011) have been converted to our definition (they used f_2500 Å for the optical flux density).

Zoom In Zoom Out Reset image size

We performed X-ray spectral analyses for the two radio-intermediate PHL 1811 analogs, since both of them have sufficient counts for basic spectral fitting. The X-ray spectra were extracted with the psextract routine in CIAO v4.2, using a 3'' radius aperture centered on the X-ray position for each source. Background spectra were extracted using annular regions with inner radii of 6'' and outer radii of 9''. Both background regions are free of X-ray sources. Spectral fitting was executed using XSPEC v12.5.1 (Arnaud 1996). The X-ray spectra were grouped to have at least 20 (10) counts per bin for J1454+0324 (J1618+0704). We used a power-law model with a Galactic absorption component, in which the Galactic column density is fixed to the values from Table 5. We also tried another model similar to the first, but adding an intrinsic neutral absorption component. Table 8 shows the X-ray spectral fitting results. The quoted errors or the upper limits for the best-fit parameters are at 90% confidence for one parameter of interest (Δχ² = 2.71; Avni 1976). For both sources, we found no evidence for strong intrinsic neutral absorption; adding an intrinsic neutral absorption component did not improve the fit quality. The best-fit photon indices for both sources are consistent with those from band-ratio analysis (see the last column of Table 4).

J1618+0704, which has moderate radio emission (R = 35), has a relatively flat X-ray power-law continuum (Γ = 1.48^+0.29 _{− 0.28}) indicating that its X-ray flux probably has a substantial contribution from relativistic jets (e.g., Wilkes & Elvis 1987; Page et al. 2005). The X-ray brightness and X-ray spectral shape of J1618+0704 suggest that the X-ray emission associated with jets has not been significantly diminished, even if the X-ray emission from the ADC has been absorbed or quenched as for the radio-quiet PHL 1811 analogs (i.e., the observed spectrum may be dominated by jet-linked emission).

The other radio-intermediate PHL 1811 analog, J1454+0324, has a steeper X-ray continuum (Γ = 2.13^+0.18 _{− 0.17}). The radio-loudness value (R = 12.8) of this source is just above the upper limit for radio-quiet sources. However, J1454+0324 is brighter in X-rays than most typical radio-intermediate quasars with significantly higher R values (see Figure 10). The notable X-ray brightness and the steep X-ray continuum indicate that this source may have a different physical nature from that of J1618+0704. One possibility is that J1454+0324 may be similar to BL Lac objects. Although BL Lac objects are generally radio-loud, it is possible that the BL Lac population has a small radio-faint tail (e.g., Plotkin et al. 2010a). In this scenario, J1454+0324 would have a relatively featureless UV spectrum because its emission lines have been somewhat diluted by a relativistically boosted UV continuum (though not enough to make it a bona fide BL Lac object), which allows it to pass our criteria for selecting PHL 1811 analogs. The α_ox value for J1454+0324 is consistent with those for the majority of BL Lac objects (e.g., Shemmer et al. 2009; Plotkin et al. 2010b) for which the X-ray emission is relativistically boosted. The X-ray spectral slope of J1454+0324 is also consistent with those of BL Lac objects (e.g., Donato et al. 2005). However, there are several problems with a BL Lac-like interpretation. First, J1454+0324 has relatively strong observed Fe ii and Fe iii emission, which should also be diluted by a relativistically boosted continuum. Second, J1454+0324 has a blue UV/optical continuum (α_ν = −0.28), while typical BL Lac objects have red continua (e.g., Stein et al. 1976). Third, there is no significant variability (only ≈3% level) between the two SDSS spectroscopic epochs of J1454+0324. Future observations (e.g., polarization measurements) of J1454+0324 in the UV/optical band may further constrain its nature which presently remains uncertain.