The Astrophysical Journal, 751:107 (15pp), 2012 June 1 doi:10.1088/0004-637X/751/2/107
C
2012. The American Astronomical Society. All rights reserved. Printed in the U.S.A.
A 10 kpc SCALE SEYFERT GALAXY OUTFLOW: HST/COS
OBSERVATIONS OF IRAS F22456−5125
Benoit C. J. Borguet
1
, Doug Edmonds
1
, Nahum Arav
1
, J ay Dunn
2
, and Gerard A. Kriss
3,4
1
Department of Physics, Virginia Tech, Blacksburg, VA 24061, USA; benbo@vt.edu
2
Augusta Perimeter College, Atlanta, GA, USA
3
Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
4
Center for Astrophysical Sciences, Department of Physics and Astronomy, Johns Hopkins University, Baltimore, MD 21218, USA
Received 2012 January 6; accepted 2012 March 25; published 2012 May 11
ABSTRACT
We present analysis of the UV spectrum of the low-z AGN IRAS F22456−5125 obtained with the Cosmic Origins
Spectrograph on board the Hubble Space Telescope. The spectrum reveals six main kinematic components, spanning
a range of velocities of up to 800 km s
−1
, which for the first time are observed in troughs associated with C ii,Civ,
N v,Siii,Siiii,Siiv, and S iv. We also obtain data on the O vi troughs, which we compare to those available from an
earlier Far Ultraviolet Spectroscopic Explorer epoch. Column densities measured from these ions allow us to derive
a well-constrained photoionization solution for each outflow component. Two of these kinematic components show
troughs associated with transitions from excited states of Si ii and C ii. The number density inferred from these
troughs, in combination with the deduced ionization parameter, allows us to determine the distance to these outflow
components from the central source. We find these components to be at a distance of ∼10 kpc. The distances and
the number densities derived are consistent with the outflow being part of a galactic wind.
Key words: galaxies: active – line: formation – quasars: absorption lines – quasars: individual (IRAS
F22456−5125)
Online-only material: color figures, figure set
1. INTRODUCTION
Mass outflows are detected in the UV spectra of more than
50% of low-redshift active galactic nuclei (AGNs), mainly
Seyfert galaxies, e.g., Crenshaw et al. (1999), Kriss et al.
(2002), Dunn et al. (2007), and Ganguly & Brotherton (2008).
These outflows are observed as narrow absorption lines (a few
hundred km s
−1
in width) blueshifted with respect to the AGN
systemic redshift.
In this paper, we determine the ionization equilibrium, dis-
tance, mass flow rate, and kinetic luminosity of the UV outflow
observed in the luminous Seyfert 1 galaxy IRAS F22456−5125
(z = 0.1016; Dunn et al. 2010). The bolometric luminosity
of this object, L
bol
= 10
45.6
erg s
−1
(see Section 4), places it
at the Seyfert/quasar border defined to be 10
12
L
, where L
is the luminosity of the Sun (Soifer et al. 1987). Several ab-
sorption systems are resolved in the UV spectrum in five main
kinematic components ranging in velocity from −20 km s
−1
to
−820 km s
−1
. A detailed analysis of the physical properties of
the UV absorber determined from Far Ultraviolet Spectroscopic
Explorer (FUSE) archival spectra has been published by Dunn
et al. (2010). These authors report a lower limit on the distance
R of the absorbing material from the central source of 20 kpc
using photoionization timescale arguments.
In 2010 June we observed IRAS F22456−5125 with the
Cosmic Origins Spectrograph (COS) on board the Hubble Space
Telescope (HST) as part of our program aiming at determining
the cosmological impact of AGN outflows (PI: Arav, PID:
11686). The high signal-to-noise spectrum obtained reveals the
presence of absorption troughs associated with high-ionization
species (C iv,Nv,Ovi,Siiv, and S iv), as well as lower ones
(Si ii,Siiii,Cii), thus increasing the number of constraints
on the photoionization analysis of the absorber compared
to Dunn et al. (2010). We also identify absorption troughs
corresponding to excited states of Si ii and C ii associated with
two kinematic components of the UV outflow. The population of
the excited state relative to the resonance counterpart provides
an indirect measurement of the number density of the gas
producing the lines (Osterbrock & Ferland 2006). These number
densities allow us to determine reliable distances to these two
components and hence derive their mass flow rates and kinetic
luminosities.
The plan of the paper is as follows: in Section 2, we present
the COS observations of IRAS F22456−5125, as well as the
reduction of the data and identification of the spectral features
within the COS range. In Section 3, we detail the computation
of the column densities associated with every species. We
present the photoionization analysis of the outflow components
in Section 4 and report the derived distance, mass flow rate, and
kinetic luminosity in Section 5. We conclude the paper by a
discussion of our results in Section 6. This paper is the second
of a series and the reader will be referred to Edmonds et al.
(2011, hereafter Paper I) for further details on the techniques
used throughout the paper.
2. HST/COS OBSERVATIONS AND DATA REDUCTION
We observed IRAS F22456−5125 using the COS instrument
(Osterman et al. 2011) on board the HST in 2010 June using both
medium-resolution (Δλ/λ ∼ 18,
000) far ultraviolet gratings
G130M and G160M. Sub-exposures of the target were obtained
for each grating through the Primary Science Aperture using
different central wavelength settings in order to minimize the
impact of the instrumental features and to fill the gap between
detector segments, providing a continuous coverage over the
spectral range between roughly 1135–1795 Å. We obtained a
total integration time of 15,056 s and 11,889 s for the G130M
and G160M gratings, respectively.
1
Copyright by the IOP PUBLISHING LTD. Benoit C. J. Borguet et al. 2012. " a 10 kpc scale seyfert galaxy outflow: hst/cos
observations of iras f22456-5125," ApJ 751 107 doi:10.1088/0004-637X/751/2/107
The Astrophysical Journal, 751:107 (15pp), 2012 June 1 Borguetetal.
Figure 1. Full FUV spectrum of IRAS F22456−5125 obtained by COS. The major absorption troughs related to the intrinsic absorber are labeled. The green line
represents our fit to the non-absorbed emission model (see Section 2.3).
(A color version and the complete figure set (23 images) of this figure are available in the online journal.)
The data sets processed through the standard CALCOS
5
pipeline were retrieved from the MAST archive. They were then
flat-fielded and combined together using the COADD_X1D
6
IDL pipeline developed by the COS GTO team (see Danforth
et al. 2010 for details). The reduced spectrum with its original
5
Details on CALCOS can be found in the COS Data Handbook.
6
The routine can be found at http://casa.colorado.edu/∼danforth/science/
cos/costools.html.
∼2kms
−1
oversampling has an overall signal-to-noise ratio
(S/N) 15 pixel
−1
in most of the continuum region. Typical
errors in the wavelength calibration are less than 15 km s
−1
.In
Figure 1, we show the majority of the spectrum on which we
identified major intrinsic absorption features associated with
the outflow. The COS FUV spectrum of IRAS F22456−5125
is presented in greater detail along with the identification of
most absorption features (interstellar, intergalactic, and intrinsic
lines) in the online version of Figure 1.
2
The Astrophysical Journal, 751:107 (15pp), 2012 June 1 Borguetetal.
Figure 2. Illustration of the necessity of using a deconvolution algorithm when dealing with COS data (see the text for details). Troughs associated with the intrinsic
absorber significantly deepen while the saturated interstellar C ii line exhibits the expected squared black bottom p rofile. The main difference between the deconvolved
spectrum using the RL method and the modified RL algorithm respecting the sampling theorem is the significant reduction of oscillations due to the total deconvolution
process performed in RL, even more so when considering a high number of iterations.
(A color version of this figure is available in the online journal.)
2.1. Identification of Spectral Features
Using archived FUSE spectra, Dunn et al. (2007, 2010)
reported the first detection of five distinct kinematic components
with centroid velocities v
1
=−800 km s
−1
, v
2
=−610 km s
−1
,
v
3
=−440 km s
−1
, v
4
=−320 km s
−1
, v
5
=−130 km s
−1
and FWHM ∈ [50, 200] km s
−1
associated with an intrinsic UV
outflow in IRAS F22456−5125. These components, spanning a
total velocity range of 800 km s
−1
, were detected in O vi,Ciii,
and several lines of the Lyman series (Lyβ to Lyη). Using the
kinematic pattern reported by Dunn et al. (2010) as a template,
we identify absorption features in our COS spectrum related to
both low-ionization (C ii,Siii,Siiii) and high-ionization species
(Si iv,Siv,Civ,Nv,Ovi), as well as in the Lyα transition.
Absorption troughs from the metastable level C ii* λ1335.704
are detected in components 2 and 3, and troughs from metastable
Si ii* λλ1264.738 and 1194.500 are detected in component 2.
While the absorption troughs associated with the higher-
ionization lines generally exhibit broader profiles, we observe
a 1:1 kinematic correspondence between the core of these
components and the narrower features associated with the lower-
ionization species of the outflow. Given the significantly broad
range of velocities covered by the components and their net
kinematic separations, such a match is not likely to occur by
chance. This argues in favor of a scenario where the troughs
of the different ionic species detected in a given kinematic
component are generated in the same region. This observation
is strengthened by the fact that most of the troughs have a line
profile similar to that of the non-blended N v λ1238.820 line
when properly scaled.
The high S/N of our COS observations (S/N 40 per
resolution element on most of the spectral coverage) reveals
the presence of kinematic substructures in several components
of the outflow compared to the lower-S/N FUSE observations
(S/N ∼ 7; Dunn et al. 2010). Nevertheless, given the self-
blending of these features in the strongest lines (e.g., O vi) and
the absence of apparent change between the FUSE and COS
observations, we will use the labeling of components as defined
in Dunn et al. (2010). We will, however, separate their trough
5 into low- and high-velocity components given the apparent
difference in ionization suggested by the presence of a stronger
Si iii in subcomponent 5A
(v
5A
=−40 km s
−1
) than in 5B
(v
5B
=−130 km s
−1
) relative to the higher-ionization lines
(C iv,Nv,Ovi; see Figure 3). Most of our analysis in this
paper concentrates on components 2 and 3 of the outflow, for
which absorption features associated with an excited state have
been detected.
2.2. Deconvolution of the COS Spectrum
Detailed analysis of the on-orbit COS line-spread function
(LSF) revealed the presence of broadened wings that scatter
a significant part of the continuum flux inside the absorption
troughs (see Kriss et al. 2011 for details). This continuum
leaking is particularly strong for narrow absorption troughs
(FWHM ∼50 km s
−1
) in which this effect may significantly
affect the estimation of the true column density by artificially
increasing the residual intensity observed inside the troughs.
Given the overall good S/N of our data, we can correct the
effect of the poor LSF by deconvolving the COS spectrum.
Adopting the procedure described in Kriss et al. (2011), we
deconvolve the spectrum obtained for each grating in 50 Å
intervals using the wavelength-dependent LSFs and an IDL
implementation of the stsdas Richardson–Lucy (RL) “lucy”
algorithm (G. Schneider & B. Stobie 2011, private commu-
nication). The main effect of the deconvolution is illustrated in
Figure 2, in which we clearly see that the deconvolved spectrum
shows significantly deepened intrinsic Lyα absorption troughs
and produces a square, black bottom for the saturated interstellar
line C ii λ1334.532.
However, the main drawback of the deconvolution algorithms
commonly used, such as the RL algorithm, is a significant
increase of the noise in the deconvolved spectrum due to the
fact that these techniques try to perform a total deconvolution of
the data, i.e., in which the LSF of the deconvolved spectrum is
a Dirac delta function, violating the Shannon sampling theorem
3
The Astrophysical Journal, 751:107 (15pp), 2012 June 1 Borguetetal.
Figure 3. Normalized absorption line profile of the metal lines associated with the outflow in IRAS F22456−5125. The line profiles have been deconvolved using
the modified RL algorithm described in Section 2.2 and rebinned to a common ∼5kms
−1
dispersion velocity scale. For doublets, we overplot the expected residual
intensity in the strongest component based on the residual flux observed in the weakest component assuming an AOD absorber model. For C iv we only plot that
quantity in regions free of self-blending (mainly T 3; see the text).
(A color version of this figure is available in the online journal.)
(see Magain et al. 1998 for a thorough discussion of these
issues). In order to decrease these effects, we modified the RL
algorithm by forcing the deconvolved spectrum to have an LSF
satisfying the sampling theorem. We choose the deconvolved
LSFtobeaGaussianwitha2pixelFWHM(∼5kms
−1
given the COS detector sampling). This operation prevents the
appearance of strong unwanted oscillations since we force the
maximum resolution that can be achieved in the deconvolved
data to agree with the sampling theorem. The deconvolved
spectrum produced by this modified RL algorithm is similar to
the one obtained by the traditional RL algorithm (see Figure 2),
the main difference being the significant decrease of the high
frequencies and high-amplitude features artificially generated
by RL deconvolution with a high number of iterations. In our
analysis, we will derive the column density for each ionic
species using the spectrum deconvolved with the modified RL
algorithm, allowing us to derive more accurate column densities
associated with the narrow absorption components observed in
IRAS F22456−5125.
2.3. Unabsorbed Emission Model
The unabsorbed emission model F
0
(λ) of IRAS F22456−
5125 is constructed in a similar manner to the one described
in detail for IRAS F04250−5718 in Paper I, in which we
consider three main sources of emission: a continuum, a broad
emission line (BEL) component, and a narrow emission line
(NEL) component. Adopting a single power law (PL) F (λ) =
F
1150
(λ/1150)
α
to describe the deredenned (E(B − V ) =
0.01035; Schlegel et al. 1998) continuum emission, we obtain
4
The Astrophysical Journal, 751:107 (15pp), 2012 June 1 Borguetetal.
a reduced χ
2
red
= 1.413 over emission/absorption line free
regions of the rest wavelength spectrum ([1115, 1130] Å,
[1340,1360] Å, and [1455,1475] Å) with α =−1.473 ± 0.068
and F
1150
= 2.13010
−14
± 0.003310
−14
erg cm
−2
Å
−1
s
−1
.
Prominent BEL features observed in the spectrum (Lyα,Civ,
O vi) are fit using two broad Gaussians of FWHM ∼ 9000 and
2000 km s
−1
. The NEL component of each line of a doublet is fit
by a single narrower Gaussian (FWHM ∼ 600 km s
−1
) centered
around the rest wavelength of each line, with the separation of
the two Gaussians fixed to the velocity difference between the
doublet lines. The NEL of the strong Lyα lineisbestfitbytwo
Gaussians of FWHM ∼ 1200 and 400 km s
−1
. The remaining
weaker emission features in the spectrum (Si iv+O iv,Cii,Nv,
O i, etc.) are modeled by a smooth cubic spline fit. A normalized
spectrum is then obtained by dividing the data with the emission
model. We present our best fit to the unabsorbed spectrum of
IRAS F22456−5125 in Figure 1.
3. COLUMN DENSITY DETERMINATION
3.1. Methodology
The column density associated with a given ionic species
detected in the outflow is determined by modeling the residual
intensity in the normalized data of the absorption troughs.
Assuming a single homogeneous emission source F
0
(v) and
a one-dimensional spatial distribution of optical depth across
the emission source τ
i
(x,v), we can express the intensity F
i
(v)
observed for a line i as (Arav et al. 2005)
F
i
(v) = F
0
(v)
1
0
e
−τ
i
(x,v)
dx, (1)
where v is the radial velocity of the outflow and the spatial
extension of the emission source is normalized to 1. Once the
optical depth solution τ
i
(x,v) is derived at a given radial veloc-
ity, we link the observed residual intensity I
i
(v) = F
i
(v)/F
0
(v)
to the ionic column density using the relation
N
ion
(v) =
3.8 × 10
14
f
i
λ
i
τ
i
(v)(cm
−2
km
−1
s), (2)
where f
i
, λ
i
, and τ
i
(v) are the oscillator strength, the rest
wavelength, and the average optical depth across the emission
source of line i (see Paper I), respectively.
We consider here the three absorber models (i.e., optical depth
distributions) discussed in Paper I: the apparent optical depth
(AOD), partial covering (PC), and PL models. We use these three
models in order to account for possible inhomogeneities in the
absorber (see Section 6), which cause t he apparent strength ratio
R
a
= τ
i
/τ
j
of two lines i, j from a given ion to deviate from the
expected laboratory ratio R
l
= λ
i
f
i
/λ
j
f
j
(e.g., Wampler et al.
1995; Hamann 1997;Aravetal.1999). Wherever possible we
derive these three optical depth solutions for ions with multiple
transitions. However, as mentioned in Paper I, we consider the
results obtained with the PL model performed on doublets with
caution given its increased sensitivity to the S/N, which can
lead to severe overestimation of the underlying column density
(Arav et al. 2005). For singlet lines we will generally only derive
a lower limit on the column density using the AOD method. This
lower limit will be considered a measurement in cases where the
singlet line is associated with a kinematic component for which
other multiplets do not show signs of saturation. In the following
subsections, we use the term (non-black) saturation to qualify
Tab le 1
Atomic Data for the Observed Transitions
Ion E
low
a
λ
i
b
g
low
c
f
i
d
(cm
−1
)(Å)
H i 0.00 1215.670 2 0.4164
C ii 0.00 1334.532 2 0.1290
C ii* 63.42 1335.704
e
4 0.1277
C iv 0.00 1548.202 2 0.1900
C iv 0.00 1550.774 2 0.0952
N v 0.00 1238.821 2 0.1560
N v 0.00 1242.804 2 0.0780
O vi 0.00 1031.912 2 0.1330
O vi 0.00 1037.613 2 0.0660
Si ii 0.00 1190.416 2 0.2770
Si ii 0.00 1193.280 2 0.5750
Si ii* 287.24 1194.500 4 0.7370
Si ii 0.00 1260.422 2 1.2200
Si ii* 287.24 1264.730 4 1.0900
Si ii 0.00 1304.370 2 0.0928
Si ii 0.00 1526.720 2 0.1330
Si iii 0.00 1206.500 1 1.6700
Si iv 0.00 1393.760 2 0.5130
Si iv 0.00 1402.770 2 0.2550
S iv 0.00 1062.656 2 0.0500
Notes.
a
Lower-level energy.
b
Wavelength of the transition.
c
Statistical weight.
d
Oscillator strength. We use the oscillator strengths from the National Institute
of Standards and Technology (NIST) database, except for S iv, for which we
use the value reported in Hibbert et al. (2002).
e
Blend of two transitions; we report the sum of the oscillator strength and the
weighted average of λ
i
.
absorption troughs of doublets in which R
a
= τ
i
/τ
j
< 0.75 R
l
,
where τ
i
and τ
j
are the AOD of the strongest and the weakest
component of the doublet, respectively.
3.2. Column Density Measurements
Computed ionic column densities are determined using the
deconvolved line profiles presented in Figure 3 and the ionic
transition properties reported in Table 1. The computed column
densities are reported in Table 2 for the three absorber models
when possible. The adopted values shown in the last column
of Table 2 are the ones used in the photoionization analysis.
When available, we choose to use the value reported in the PC
column as the measurement and use the PL measurement and
error as the upper error in order to account for the possible
inhomogeneities i n the absorbing material distribution. If only
the AOD determination is available, we will consider the
reported value minus the error as a lower limit unless we have
evidence suggesting a high covering.
3.2.1. H
i
The spectral coverage of the COS G130M and G160M
gratings only allows us to cover the Lyα line that shows a deep
and smooth profile in which the different kinematic components
blend. The absence of higher-order Lyman series lines restricts
us to put a lower limit on the H i column density by applying
the AOD method to the Lyα profile.
A better constraint on the H i column density is determined
by using higher-order Ly-series lines from earlier FUSE data
(Dunn et al. 2010). In Figure 4, we compare the 2010 June
5
