Absorption signatures of warm-hot gas at low redshift: OVI

Abstract

We investigate the origin and physical properties of OVI absorbers at low
redshift (z = 0.25) using a subset of cosmological, hydrodynamical simulations
from the OverWhelmingly Large Simulations (OWLS) project. Intervening OVI
absorbers are believed to trace shock-heated gas in the Warm-Hot Intergalactic
Medium (WHIM) and may thus play a key role in the search for the missing
baryons in the present-day Universe. When compared to observations, the
predicted distributions of the different OVI line parameters (column density,
Doppler parameter, rest equivalent width) from our simulations exhibit a lack
of strong OVI absorbers. This suggests that physical processes on sub-grid
scales (e.g. turbulence) may strongly influence the observed properties of OVI
systems. We find that the intervening OVI absorption arises mainly in highly
metal-enriched (0.1 << Z/Z_sun < 1) gas at typical overdensities of 1 <<
rho/ < 100. One third of the OVI absorbers in our simulation are found to
trace gas at temperatures T < 10^5 K, while the rest arises in gas at higher
temperatures around T =10^5.3 K. The OVI resides in a similar region of
(rho,T)-space as much of the shock-heated baryonic matter, but the vast
majority of this gas has a lower metal content and does not give rise to
detectable OVI absorption As a consequence of the patchy metal distribution,
OVI absorbers in our simulations trace only a very small fraction of the cosmic
baryons (<2 percent) and the cosmic metals. Instead, these systems presumably
trace previously shock-heated, metal-rich material from galactic winds that is
now cooling. The common approach of comparing OVI and HI column densities to
estimate the physical conditions in intervening absorbers from QSO observations
may be misleading, as most of the HI (and most of the gas mass) is not
physically connected with the high-metallicity patches that give rise to the
OVI absorption.

Full text

Page 1

Mon. Not. R. Astron. Soc. 000, 1–18 (—-)Printed 19th July 2010(MN LATEX style file v2.2)
Absorption signatures of warm-hot gas at low redshift: Ovi
Thorsten Tepper-Garc´ ıa,1?Philipp Richter,1Joop Schaye,2C. M. Booth,2
Claudio Dalla Vecchia,2,3Tom Theuns4,5and Robert P.C. Wiersma2,6
1Universit¨ at Potsdam, Karl-Liebknecht-Str. 24/25, 14476 Potsdam, Germany
2Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, The Netherlands
3Max Planck Institut f¨ ur Extraterrestrische Physik, Giessenbachstraße 1, 85748 Garching, Germany
4Institute for Computational Cosmology, Department of Physics, University of Durham, South Road, Durham, DH1 3LE, UK
5Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium
6Max Planck Institut f¨ ur Astrophysik, Karl-Schwarzschild-Str. 1, 8574, Garching, Germany
Accepted —-. Received —-; in original form —-
ABSTRACT
We investigate the origin and physical properties of Ovi absorbers at low redshift
(z = 0.25) using a subset of cosmological, hydrodynamical simulations from the OverWhelm-
ingly Large Simulations (OWLS) project. Intervening Ovi absorbers are believed to trace
shock-heated gas in the Warm-Hot Intergalactic Medium (WHIM) and may thus play a key
role in the search for the missing baryons in the present-day Universe. When compared to
observations, the predicted distributions of the different Ovi line parameters (column dens-
ity NO VI, Doppler parameter bO VI, rest equivalent width Wr) from our simulations exhibit a
lack of strong Ovi absorbers, a discrepancy that has also been found by Oppenheimer &
Dav´ e (2009). This suggests that physical processes on sub-grid scales (e.g. turbulence) may
strongly influence the observed properties of Ovi systems. We find that the intervening Ovi
absorption arises in highly metal-enriched (10−1? Z/Z? ? 1) gas at typical overdensities
of 1 ? ρ/?ρ? ? 102and temperatures T = 105.3±0.5K. These temperatures are much higher
than inferred by Oppenheimer & Dav´ e (2009), probably because that work did not take the
suppression of metal-line cooling by the photo-ionising background radiation into account.
While the Ovi resides in a similar region of (ρ,T)-space as much of the shock-heated ba-
ryonic matter, the vast majority of this gas has a lower metal content and does not give rise to
detectable Ovi absorption. As a consequence of the patchy metal distribution, Ovi absorbers
in our simulations trace only a very small fraction of the cosmic baryons (< 2 percent) and the
cosmic metals. Instead, these systems presumably trace previously shock-heated, metal-rich
material from galactic winds that is now cooling. The common approach of comparing Ovi
and Hi column densities to estimate the physical conditions in intervening absorbers from
QSO observations may be misleading, as most of the Hi (and most of the gas mass) is not
physically connected with the high-metallicity patches that give rise to the Ovi absorption.
Key words: cosmology: theory — methods: numerical — intergalactic medium — quasars:
absorption lines — galaxies: formation
1 INTRODUCTION
Diffuse ionised gas in the intergalactic medium (IGM) represents
the major baryon reservoir in the Universe at any redshift. From
observations of the Lyα forest line density at high redshift in quasar
(QSO) and active galactic nuclei (AGN) spectra it can be deduced
that the photo-ionised IGM contains more than ninety per cent of
the baryons at redshift z = 3 (Rauch et al. 1997; Weinberg et al.
1997; Schaye 2001). However, at z = 0 the fraction of baryons
?E-mail: tepper@astro.physik.uni-potsdam.de
residing in the Lyα forest is strongly reduced to ∼ 30 − 40 per
cent (Penton et al. 2004), while at the same time the (observable)
amountofbaryonsincondensedstructures(i.e.,baryonsingalaxies
and galaxy clusters) has increased to only a few per cent (Fukugita
2004). These observations thus indicate that a significant fraction
of the baryons formerly residing in the photo-ionised IGM have
“disappeared”. Cosmological simulations have predicted that, as a
result of the large-scale structure formation in the Universe, most of
these “missing baryons” have moved into a hot, shock-heated inter-
galactic gas phase (Cen & Ostriker 1999; Dav´ e et al. 2001; Bertone
et al. 2008). This shock-heated intergalactic gas phase is referred to
c ? —- RAS
arXiv:1007.2840v1 [astro-ph.CO] 16 Jul 2010

Page 2

2Tepper-Garc´ ıa et al.
as the Warm-Hot Intergalactic Medium (WHIM) and is expected to
have characteristic temperatures in the range T ∼ 105− 107K and
densities of nH∼ 10−4− 10−6cm−3. Constraining the distribution
and physical properties of the WHIM is important to understand
how the gas (and the metals) are transported from galaxies into the
IGM and recycled into galaxies, and what role the shock-heated
IGM has for the evolution of galaxies in the local Universe.
Since emission from such a thin plasma is extremely dim (Fur-
lanetto et al. 2004; Bertone et al. 2009, 2010) and the hydrogen
is almost fully ionised, the analysis of highly ionised heavy ele-
ments (in particular the high ions of oxygen, Ovi Ovii , and Oviii)
in Ultra-Violet (UV) and X-ray absorption against distant extra-
galactic background sources has become the leading method to
study the properties and baryon content of the WHIM at low red-
shift (for a recent review see Richter et al. 2008). Most promising is
the search for five-times ionised oxygen (Ovi) in the UV spectra of
low-redshift AGN, as obtained with space-based UV spectrographs
such as Hubble Space Telescope (HST) Space Telescope Imaging
Spectrograph (STIS) and the Far Ultraviolet Spectroscopic Ex-
plorer (FUSE) (e.g. Tripp et al. 2000; Richter et al. 2004). Oxygen
is a relatively abundant element with two strong Ovi transitions at
λ1031.9Å and λ1037.6Å, and so far, more than 50 intervening Ovi
absorbers at low redshift have been identified and analysed (e.g.
Tripp et al. 2008; Danforth & Shull 2008).
In spite of the large Ovi samples obtained to date, there is still
no general consensus about the physical conditions of the gas giv-
ing rise to Ovi absorption. While Danforth & Shull (2008) find that
Ovi (and associated Nv) are reliable tracers of collisionally ionised
gas at temperatures 105K < T < 106K (i.e., the low-temperature
WHIM), Thom & Chen (2008a) argue that Ovi arises mainly in
photo-ionised gas at temperatures T < 105K. Similarly, Tripp et al.
(2008) find that well-aligned Ovi-Hi absorbers have line widths
that are consistent with photo-ionised gas. Nevertheless, these au-
thors show that more than half of their Ovi absorbers are complex,
i.e., multi-phase, and could thus trace both cold photo-ionised gas
and lower metallicity collisionally ionised gas at T > 105K.
The interpretation of the abundance and nature of interven-
ing Ovi absorbers arising in collisionally ionised gas in terms of
the distribution and baryon content of the WHIM is not straight-
forward. In collisional ionisation equilibrium (CIE; as usually as-
sumed for a shock-heated plasma like the WHIM), Ovi predomin-
antly traces gas at T ∼ 3 × 105K, while in the higher temperature
regime of the WHIM oxygen is further ionised to Ovii and Oviii,
observable only in the X-ray band for which high-quality spectral
data are sparse. Thus, only a (small) fraction of the WHIM can
in principle be traced by intervening Ovi absorbers. In addition,
several authors (e.g. Gnat & Sternberg 2007) have argued that the
assumption of CIE is not valid in case of the WHIM (i.e., the gas
may be out of an ionisation equilibrium) and more recently, Wi-
ersma et al. (2009a) have shown that photo-ionisation of the WHIM
strongly influences the cooling efficiency of the gas in the temper-
ature range of interest. These physical arguments, together with the
not well established oxygen abundance of the WHIM at low red-
shift, indicate that a reliable estimate of the baryon budget of the
WHIM (i.e., the amount of ionised hydrogen) from Ovi observa-
tions requires a deep understanding of the physical conditions of
the gas.
In this paper, we make use of a set of numerical simula-
tions from the OverWhelmingly Large Simulations (OWLS) pro-
ject (Schaye et al. 2010) to study the physical properties and baryon
content of intervening Ovi absorbers at low redshift. The main ad-
vantage of these simulations compared to previous ones is the im-
plementationofimportantphysicalprocessesthathavebeenlargely
ignored in earlier studies. As will be shown here, the influence of
the photo-ionisation on the cooling function of the WHIM (Wi-
ersma et al. 2009a) represents a particularly important aspect for
a correct interpretation of the observed properties of the Ovi ab-
sorbers and their role for our understanding of the WHIM.
Even though we do not attempt to tune our simulations to re-
produce any Ovi observables in this paper, we will compare our
results to observations where possible. Moreover, a thorough ana-
lysis of variations with respect to our fiducial model and the effect
of post-run physics variations on the resulting physical properties
of the simulated Ovi absorbers are left for a future paper. Note
that we will often refer to Oppenheimer & Dav´ e (2009) and make
comparisons between our and their results, since Oppenheimer &
Dav´ e (2009) is the most comprehensive study on Ovi absorbers in
simulations to date. As we will show, the fact that Oppenheimer
& Dav´ e (2009) assumed CIE when computing the contribution of
heavy elements to the radiative cooling rates will likely have af-
fected their conclusions.
This paper is organised as follows: in Sec. 2 we briefly intro-
duce the simulation we use, i.e., our fiducial model. We describe
how we create our synthetic spectra and present their analysis in
Sec. 3. In Sec. 4 we discuss a comparison of our results to ob-
servations, while in Sec. 5 we analyse the physical properties of
the Ovi gas in our simulation in detail. We present and discuss our
conclusions in Sec. 6. A detailed analysis of the effect of the signal-
to-noise ratio, simulation box size, and resolution on the Ovi linea
parameter distributions are left for the Appendix.
2SIMULATIONS
The simulations used in this work are part of a large set of
cosmological simulations that together comprise the OWLS
project, described in detail in Schaye et al. (2010, and references
therein). Briefly, the simulations were performed with a signi-
ficantly extended version of the N-Body, Tree-PM, Smoothed
Particle Hydrodynamics (SPH) code gadget iii – which is a
modified version of gadget ii (last described in Springel 2005)
–, a Lagrangian code used to calculate gravitational and hydro-
dynamical forces on a system of particles. The initial conditions
were generated from an initial glass-like state (White 1996) with
cmbfast (version 4.1; Seljak & Zaldarriaga 1996) and evolved
to redshift z = 127 using the Zel’Dovich (1970) approximation.
The reference model of OWLS adopts a flat ΛCDM cosmology
characterised by the set of parameters {Ωm, Ωb, ΩΛ, σ8, ns, h} =
{0.238, 0.0418, 0.762,0.74, 0.95, 0.73}
Wilkinson Microwave Anisotropy Probe (WMAP) 3-year data1
(Spergel et al. 2007). The reference model includes prescriptions
for star formation (SF) and for feedback from core collapse
supernovae (SNe) described in Schaye & Dalla Vecchia (2008) and
Dalla Vecchia & Schaye (2008), respectively. The implementations
of radiative cooling and heating are described in Wiersma et al.
(2009a) and summarised below (cf. Sec. 5.3). For a thorough de-
scription of the implementation of stellar evolution in the reference
model we kindly refer the reader to Wiersma et al. (2009b).
asderivedfromthe
1These parameter values are largely consistent with the WMAP 7-year
results (Jarosik et al. 2010), the largest difference being the value of σ8,
which is 2σ lower in the WMAP 3-year data than allowed by the WMAP
7-year data.
c ? —- RAS, MNRAS 000, 1–18

Page 3

Ovi absorbers at low redshift3
Table 1. List of simulations –which assume the reference model of the OWLS project– used in this study. Our fiducial run REF L050N512 (shown in bold) is
used for the comparison to observations and our general analysis, whilst all other simulations are used in the Appendix when addressing the convergence of our
results with respect to box size and resolution. The columns showa: simulation identifierb; comoving box size (L); number of dark matter particles (N; there
are equally many baryonic particles); baryonic particle mass (mb); dark matter particle mass (mdm); comoving (Plummer-equivalent) gravitational softening
(?com); maximum physical softening (?prop); final redshift (zend).
SimulationLNmb
mdm
(h−1M?)
4.1 × 108
3.3 × 109
4.1 × 108
5.1 × 107
4.1 × 108
?com
?prop
zend
(h−1Mpc)(h−1M?)
8.7 × 107
6.9 × 108
8.7 × 107
1.1 × 107
8.7 × 107
(h−1kpc)(h−1kpc)
REF L025N128
REF L050N128
REF L050N256
REF L050N512
REF L100N512
25.00
50.00
50.00
50.00
100.00
1283
1283
2563
5123
5123
7.81
15.62
7.81
3.91
7.81
2.00
4.00
2.00
1.00
2.00
0
0
0
0
0
aFor a complete list of the OWLS runs as well as the detailed description of the corresponding model see Schaye et al. (2010).
bThe name convention is [model] LxxxNyyy, where ’xxx’ and ’yyy’ are, respectively, the box size in comoving h−1Mpc and cube root of the number of
particles per species (dark matter or baryonic).
We analyse the physical properties of Ovi absorbers identi-
fied in simulated spectra, focusing on a high resolution simula-
tion run which assumes the reference model of the OWLS dubbed
REF L050N512. This simulation was run down to z = 0 in a peri-
odic box of size L = 50 comoving h−1Mpc, using 5123dark matter
particles and equally many baryonic particles. This run has been
shown to have a large enough box size and high enough resolution
to obtain a converged prediction for the cosmic star formation his-
tory for z ? 4, which, however, drops off less rapidly with time than
is observed for z < 1 (Schaye et al. 2010). Also, the predicted metal
mass distribution for different components (stars, star-forming gas,
non-star-forming gas) and gas phases from this particular run is
shown to be converged for z < 2 and agrees broadly with observa-
tions (Wiersma et al. 2009b, see discussion in their Appendix C).
We refer the reader to Schaye et al. (2010, and references therein)
for a more detailed description of our fiducial run (and variations
thereof). In Table 1 we summarise its basic numerical properties,
together with the corresponding properties of other runs assum-
ing the reference model. These runs are discussed in Appendix B,
where we address the convergence of our fiducial run with respect
to box size and resolution, using various quantities derived from
synthetic spectra as explained in the following section.
3 SPECTRAL ANALYSIS
In this section we briefly describe our method to compute synthetic
spectra using the spectra generating package specwizard written by
Schaye, Booth, & Theuns, which follows the approach described
in Theuns et al. (1998, their Appendix A4). A detailed analysis of
the effect of the signal-to-noise ratio (S/N) on the distributions of
line parameters (column density, Doppler parameter, rest equival-
ent width) derived from the spectra is presented in Appendix A, and
the main results are only summarised here for brevity (see below).
3.1 Synthetic spectra
Quite generally, the first step to compute a synthetic spectrum is to
draw a random physical sightline from a simulation box at a given
redshift z. A physical sightline is simply defined as the line between
a given point on opposite faces of the simulation box, and the col-
lection of SPH particles with projected distances to this line smaller
than their corresponding smoothing length. The next step is to cal-
culate the ionisation balance for each SPH particle as a function of
redshift, density, and temperature, which we do using precomputed
tables obtained with the photoionisation package cloudy (version
07.02 of the code last described by Ferland et al. 1998), assuming
the gas is exposed to the Haardt & Madau (2001) model for the
X-Ray/UV background radiation from galaxies and quasars.
A synthetic spectrum is then computed as follows. For a
chosen pixel size ∆x (in proper length units), the sightline is di-
vided into Npix = [a(z)L/h]/∆x bins in distance, where h and
a(z) are the normalized Hubble constant and the expansion factor
at the box’s redshift z, respectively. Given the positions, peculiar
velocities, and temperatures of all relevant SPH particles as well as
the densities for each ion species (e.g., Ovi), we compute the ion
density-weighted temperature and peculiar velocity for each spe-
cies at each bin according to the SPH interpolation scheme. Proper
distance bins ∆x along the sightline are transformed into velocity
bins ∆v via ∆v = H(z) ∆x, ion number densities nioninto column
densities Nion= nion∆x, and temperatures into Doppler widths. The
optical depth τ(v) at each pixel for a particular transition is com-
puted assuming a thermal (i.e., Gaussian) profile, taking peculiar
velocities into account. Finally, the optical depth spectrum is trans-
formed into a continuum-normalised flux via F(v) = exp[−τ(v)].
We generate 1000 random sightlines through the simulation
box of our fiducial run at 5 different redshifts spanning the range2
z = 0.0−0.5, from which we synthesise a total of 5000 continuum-
normalised spectra containing absorption by the strongest trans-
ition λ1031.93 Å of the Ovi doublet (λ1031.93 Å, λ1037.62 Å).
In order to mimic (and thus compare our results to) observations
performed by HST/STIS as closely as possible, we convolve our
spectra with a instrumental Gaussian Line-Spread Function (LSF)
with a Full-Width-At-Half-Maximum FWHM = 7 kms−1, and res-
ample our spectra onto 3.5 kms−1pixels. We add Gaussian noise to
each spectrum3assuming a flux-dependent root-mean-square (rms)
amplitude given by4(S/N)−1F(v) for S/N = 10, 30, and 50. Note
2The redshift step is dz = 0.125.
3Unless stated otherwise, the S/N values quoted throughout this work de-
note the S/N per pixel in the continuum.
4We assume a minimum, i.e. flux-independent noise level (S/N)min= 102,
which roughly corresponds to the typical read-out noise σronof the STIS
detector of 4e−/ADU. Note, however, that our results are not sensitive to
the assumed (S/N)minvalue.
c ? —- RAS, MNRAS 000, 1–18

Page 4

4Tepper-Garc´ ıa et al.
that the typical S/N of STIS observations average S/N ∼ 10 (e.g.
Tripp et al. 2008), and hence we use our synthetic spectra with
S/N = 10 when testing our results against observations in the next
section. Spectra with S/N = 30 and S/N = 50 are used to assess
the effect of the signal-to-noise ratio on the line statistics (see Ap-
pendix A). Also, and even though a S/N of 50 is obviously much
higher than the typical S/N of STIS observations, we choose it as
a standard value for the spectra used in the analysis of the phys-
ical conditions of Ovi bearing gas in our simulation – presented
in Sec. 5 –, since this results in a larger sample of absorption lines.
Besides, the Cosmic Origins Spectrograph (COS) recently installed
on HST is expected to provide data at such high S/N for reasonable
integration times5, although at somewhat lower spectral resolution.
We fit Ovi λ1031.93 Å absorption features identified in our
spectra with individual Voigt-profile components using a modified
version of the package autovp (Dav´ e et al. 1997), which includes
the Voigt-profile approximation of Tepper-Garc´ ıa (2006). The fit
results include the column density NO VI and Doppler parameter
bO VIwith corresponding uncertainties, as well as the rest equivalent
width Wrfor each component. An example of a synthetic spectrum
and its corresponding fit is shown in top panel of Fig. 1.
The analysis of our 5000 spectra in the range 0 ? z ? 0.5
yields a total surveyed redshift path ∆z = 98.5, along which we
identify a total of 1093, 2246, and 3115 Ovi absorbers for
S/N=10, 30, and 50, respectively, thus obtaining a statistically sig-
nificant sample of Ovi absorption lines. This allows a detailed
analysis of the line parameter distributions for each absorption
line sample obtained from spectra with different S/N, discussed
in detail in Appendix A. Briefly, we show that we can reliably
identify lines with Ovi column densities above log(NO VI/cm−2) ≈
12.3 at S/N = 50, log(NO VI/cm−2) ≈ 12.6 at S/N = 30, and
logNO VI/cm−2≈ 13.2 at S/N = 10, even though we systematic-
ally underestimate the total column density along each sightline,
indicating that we miss a fraction of the absorption lines present in
the spectra. For all identified lines, however, we are able to meas-
ure rest equivalent widths quite accurately. Moreover, we find that
the Doppler parameter distribution is sensitive to signal-to-noise
ratios S/N ? 10, but fitted Doppler parameters bO VI ? 40kms−1
are reliable if S/N ? 30. Furthermore, we show in Appendix B
that the line parameter distributions are converged with respect to
box size and resolution. These results are of great importance when
testing the predictions from simulations against observations us-
ing quantities derived from synthetic spectra, since they allow us to
bracket the sources of a potential disagreement between predicted
and observed quantities more easily. Also, they are important for
the analysis of the connection between physical conditions of the
absorbing gas and direct observables, i.e., Ovi column density and
Doppler parameter (see Sec.5.2).
For ease of discussion, in what follows we shall refer to our
Ovi absorber sample described above as sample 1. The particular
line sample obtained from spectra with S/N = 10 is discussed in
the next section where we compare the results from our simula-
tion to observations. We generate and analyse another 5000 spectra
with S/N = 50 for our fiducial run at z = 0.25. In these spectra
we identify a total of 3034 Ovi absorbers over a total redshift path
∆z = 97.7, and we shall refer to this Ovi line sample as sample
2. The latter is used and analysed in detail in Sec. 5. Note that the
5So far, QSO spectra with a mean S/N = 50 have been obtained with COS
for a total observing time of ∼ 9ks, compared to S/N = 10 for ∼ 28ks using
STIS.
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
−500 −400 −300 −200 −100 0 100 200 300
Flux (normalised)
|| | |||||
100
107
101
102
103
104
105

∆OVI
OVI λ1032 Å
log TOVI/K = 5.74
log ∆OVI = 3.14
log NOVI/cm−2 = 14.36
bOVI = 23.91 km s−1
104
105
106
−500 −400 −300 −200 −100
Restframe velocity [km s−1]
0 100 200 300
TOVI [K]
Figure 1. The top panel shows a synthetic, normalised spectrum with
S/N = 50 (black) and the corresponding fit (red) along a random sight-
line through a simulation box of size L = 50h−1Mpc at z = 0.25, show-
ing a series of λ1031.93 Å absorption features. The vertical black marks
flag the positions of the centroids of individual identified components. Note
that only the relevant velocity range is shown. The middle and bottom pan-
els show, respectively, the optical-depth weighted overdensity (magenta)
and temperature (orange) along the sightline in redshift-space (see Sec. 5).
The values included in the middle panel correspond to the optical-depth
weighted gas temperature, optical-depth weighted overdensity, Ovi column
density, and Ovi Doppler parameter for the line flagged by the blue dashed
line. That the optical-depth weighted quantities correspond in each case to
an optical-depth weighted average over the full line profile. Note in partic-
ular the correspondence between density peaks and aborption features
number of identified components and the total surveyed redshift
path in sample 2 are very similar to the corresponding quantities of
sample 1. Also, the distribution of column density, Doppler para-
meter, and rest equivalent width resulting from sample 2 are very
similar to those of sample 1, but not identical, since the latter aver-
ages the evolution of the Ovi absorbers from z = 0.5 → 0, while
the former is restricted to z = 0.25.
4COMPARISON TO OBSERVATIONS
The Ovi line parameter distributions (presented in Appendix A)
from our sample 1 for an adopted S/N = 10 can readily be
compared to three well measured Ovi observables, namely, the
column density distribution function (CDDF for short), the equival-
c ? —- RAS, MNRAS 000, 1–18

Page 5

Ovi absorbers at low redshift5
ent width distribution, and the correlation between Doppler para-
meter and column density.
The top-leftpanelof
f(NO VI) ≡ ∂2N/∂NO VI∂χ, i.e., the number of lines per logar-
ithmic column density interval per unit absorption path length
χ (see Sec. 5.1.3). Comparison of the result from our fiducial
run to observations by Thom & Chen (2008b, black symbols)
shows that the amplitude of the predicted CDDF is slightly lower
at column densities [1013, 1014]cm−2, but it agrees well with
the high-column density data point, which has, however, a large
associated uncertainty. Assuming the CDDF can be described
in terms of a single power-law, Danforth & Shull (2008) find
that a fit in the column density range [1013, 6.3 × 1014]cm−2to
the CDDF obtained from their Ovi absorber sample results in
a power-law index β = −1.98 ± 0.11. For the column density
range [1.6 × 1013, 6.3 × 1014]cm−2, we find β = −2.11 ± 0.15
(solid line), which is consistent with the observed value. Note
that this slope (with the appropriate normalisation) is also con-
sistent with the data from Thom & Chen (2008b). If instead
we consider only lines with column densities in the narrower
range [1.6 × 1013, 2 × 1014]cm−2, we find β = −1.78 ± 0.08
(dashed line), which fits the distribution better as can be judged
by the smaller uncertainty in β. We conclude that the slope
of the Ovi CDDF resulting from our fiducial run is consistent
with current observational constraints up to column densities
NO VI = 5 × 1014cm−2, but the amplitude is lower (by a factor 2)
than observed. It is worth noting at this point that a perfect match
between predicted and observed CDDF would be surprising, since
the stellar yields used in our simulation are uncertain at the factor
of two level (see Wiersma et al. 2009b, their Appendix A3).
Given that most of our identified lines lie on the linear part
of the curve-of-growth (see Fig. A2), there is a (nearly) one-to-
one correspondence between the predicted CDDF and the predicted
equivalent width distribution, and we thus do not expect the latter to
agree well with the observed one. The comparison between the cu-
mulative equivalent width distribution both from observations and
our sample 1 is shown in the top-right panel of Fig. 2. The ob-
servations are taken from Danforth & Shull (2008, their Table 4,
containing 75 confirmed equivalent width measurements) and from
Tripp et al. (2008, their Table 2 with 77 single components). While
Tripp et al. (2008) explicitly distinguish between components and
systems, Danforth & Shull (2008) do not make a clear distinction
between both types of absorbers. We follow the approach by Tripp
et al. (2008) for a better comparison with their results. We define an
absorption system as a simply connected region along the normal-
ised, fitted spectrum with an Ovi optical depth at each pixel above
a given threshold τth. We adopt6τth = 0.097, which corresponds
to the central optical depth of a Ovi λ1031.93 Å absorption line
with a column density log(NO VI/cm−2) = 12.3 – corresponding to
the smallest detectable column density in our spectra with S/N =
10; see. Fig. A1 –, and a Doppler parameter at the STIS resolution
limit bmin= 4.2 kms−1. Note that the resulting rest equivalent dis-
tribution for systems is not completely insensitive to the adopted
value of τth. As can be seen in the top-right panel of Fig. 2, our
fiducial run is not able to reproduce the amplitude of the observed
distribution, neither for components nor for systems, as anticipated
given the disagreement between the amplitude of the observed and
Fig.2 displaysthe CDDF
6For comparison, Tripp et al. (2008) estimate a detection threshold given
by τth= 0.1 for their data with an average S/N ≈ 10.
predicted CDDF. Also, our predicted turn-over at Wr∼ 100mÅ in
the distribution of components is stronger than observed.
To gain a better insight into the reason behind the discrep-
ancy between the predicted and observed equivalent width distribu-
tions for single components, we replot the predicted and observed
distributions shown in the top-right panel of Fig. 2 in differential
form, as shown in the bottom-left panel of the same figure. Note
that the data and the results from our spectra have been binned
using ∆logWr= 0.2 dex, and that the resulting distributions have
been normalised to unit area. It is apparent that the observed dis-
tributions are broadly consistent with each other, even though the
Tripp et al. (2008) distribution peaks at higher Wr. In contrast,
the distribution resulting from our fiducial run shows an excess
of absorption lines at Wr≈ 30mÅ, as well as a lack of absorp-
tion lines with Wr> 100mÅ, which explains why our predicted
turn-over at Wr∼ 100mÅ in the cumulative equivalent width dis-
tribution is stronger than observed. As is shown in Appendix A
these mismatches are not due to fitting inaccuracies but are rather
intrinsic to our simulation. Note further that, while Tripp et al.
(2008)’s measurements are restricted to Wr? 30mÅ, Danforth &
Shull (2008) estimate their Ovi system sample to be complete
down to Wr≈ 10mÅ.
In principle, there are two plausible reasons for the lack of
strong (i.e., higher equivalent width) lines. On the one hand, it
could be due to the apparent deficit of absorption components with
NO VI> 1014.5cm−2in the predicted CDDF (top-left panel of Fig. 2),
when compared to the extrapolated power-law, assuming of course
that the latter gives a correct description of the data. On the other
hand, this lack could also arise due to the absence of lines with
NO VI∼ 1014.5cm−2and Doppler parameters much larger than the
predictedmedianb-value(bO VI= 12.9 kms−1atS/N=50).Indeed,
since an Ovi line saturates at these high column densities, the equi-
valent width is particular sensitive to the Doppler parameter. An
Ovi line with NO VI∼ 1014.5cm−2can, for example, have an equi-
valent width as high as Wr∼ 400mÅ if bO VI ∼ 50 kms−1. Lines
with such high b-values are, however, almost absent from our spec-
tra (see Fig. A2). Since there is no reason to assume that the ob-
served power-law can be extrapolated to high column density val-
ues, and given the good agreement between our predicted and the
observed CDDF slope, we consider the absence of lines with large
b-values at a given column density as the most plausible explana-
tion for the lack of high equivalent width lines in our simulation as
compared to the data.
To further investigate this possibility, let us now consider the
correlation between Doppler parameters and Ovi column dens-
ities. In the framework of a unifying model, Heckman et al.
(2002) have shown that a bO VI− NO VIcorrelation, as is observed
in Ovi absorption systems in a variety of environments (Milky
Way disk and halo, High-Velocity Clouds, the Magellanic Clouds,
IGM), naturally arises in diffuse, radiative cooling gas at temperat-
ures T ∼ 105− 106K. Nevertheless, there is still some controversy
about the validity of such a model for IGM Ovi absorbers. Indeed,
while Danforth et al. (2006) find no such correlation in their data,
Lehner et al. (2006) show that their (own and compiled) data fol-
low the predicted relation rather well. Similarly, Tripp et al. (2008)
find a correlation between the line widths and column densities in
their Ovi sample, but these authors note that the significance is not
enough to support the model by Heckman et al. (2002).
Bearing this controversy in mind, we compare the bO VI−NO VI
correlation resulting from our simulation to observations. For this
purpose, we use the results by Danforth et al. (2006, their 40 Ovi
single component sample) and Tripp et al. (2008, their high quality
c ? —- RAS, MNRAS 000, 1–18

Page 6

6 Tepper-Garc´ ıa et al.
10−17
10−16
10−15
10−14
10−13
10−12
10−11
10−10
1012
1013
1014
1015
f(NOVI)
NOVI [cm−2]
REF_L050N512; components
β = −2.106
β = −1.777
Thom and Chen 08a
10−1
100
101
102
103
101
102
dN / dz (>Wr)
Wr [m Å]
REF_L050N512; components
systems
Danforth and Shull 2008
Tripp et al. 2008a; components
systems
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
101
102
103
dN / d(log Wr)
Wr [m Å]
REF_L050N512; components
Danforth and Shull 2008
Tripp et al. 2008a; components
0
1013
10
20
30
40
50
60
70
80
1014
1015
bOVI [km s−1]
NOVI [cm−2]
REF_L050N512; components
Danforth and Shull 2008a
Tripp et al. 2008a; components
Figure 2. Comparison between the observed Ovi line parameter distributions and results from our fiducial run using spectra with S/N = 10 (our sample 1),
which roughly corresponds to the average signal-to-noise of the data. Top left: Ovi column density distribution function. The x-bars indicate the bin size, while
the y-error bars show Poisson single-sided 1σ confidence limits based on the tables by Gehrels (1986). The dashed and solid lines show, respectively, a fit in
form of a power-law to the CDDF in the column density ranges [1.6 × 1013, 6.3 × 1014]cm−2and [1.6 × 1013, 2 × 1014]cm−2. Top right: Cumulative line
number density from our simulation for single components (red) and systems (blue), and observations (symbols). In each case, the black solid line shows the
actual value at each given limiting equivalent width, while the blue and red lines display Poisson noise. Bottom left: Differential column density distribution
for single components. The result for our simulation is highlighted by the shaded area. Bottom right: Line Doppler width vs. Ovi column density from our
synthetic spectra (red) and observations by Tripp et al. (2008, black) and Danforth & Shull (2008, green). For ease of comparison, our predictions and the
data have been binned in column density; in each bin, the dots indicate the median value, and the x- and y-bars indicate the bin size, and the 25th and 75th
percentiles, respectively (see text for details).
sample of 77 intervening Ovi absorbing components) only, since
both these samples are large enough, since each of them provides
evidence either against or in favour of the predicted correlation, and
because we use one of these data sets for the comparison between
our predicted and the observed cumulative equivalent width distri-
bution. To facilitate the comparison between our results and obser-
vations, we bin the Ovi Doppler parameters from our simulation
and the data in column density bins of size ∆logNO VI= 0.3 dex,
and compute the median, and the 25th and 75th percentiles in each
bin. The result is shown in the bottom-right panel of Fig. 2. Appar-
ently, the predicted correlation between Doppler parameters and
column densities matches the observations at the lowest column
densities log(NO VI/cm−2) ? 13.5, but does not so at higher column
densities. While the observations show an increase in the Doppler
width with increasing column density, the result from our simula-
tion shows no clear trend. We warn, however, that the trend sug-
gested by the binned data is to be taken with caution, given the
enormous scatter and some discrepancy between the observations
(compare, e.g., the point at NO VI= 2 × 1014cm−2).
The bottom-right panel of Fig. 2 clearly shows that
the observed b-values are much larger at column densities
NO VI> 5 × 1013than the corresponding median b-values of our
identified Ovi absorbers. This is consistent with our suspicion that
the mismatch between the predicted and observed equivalent width
distribution is due to a lack of Ovi lines with high column densities
and large Doppler parameters. Furthermore, this result also indic-
ates that the observed Ovi absorption systems are subject to a sub-
stantial broadening mechanism, either in the form of (small-scale)
turbulence (which is not captured/properly modeled by the SPH
scheme in our simulation) or Hubble broadening. The alternative
that the observed Ovi absorbers arise in gas at higher temperat-
ures, thus resulting in lines with larger Doppler parameters, can be
ruled out, since a Doppler parameter of, e.g., bO VI= 40 kms−1cor-
responds to a gas temperature T ∼ 1.5 × 106K for oxygen, which
would result in a too low Ovi ion fraction (nO VI/nO) < 10−2, either
in CIE (Sutherland & Dopita 1993) or non-equilibrium conditions
(Gnat & Sternberg 2007).
It is noteworthy that a similar disagreement between the pre-
c ? —- RAS, MNRAS 000, 1–18

Page 7

Ovi absorbers at low redshift7
dicted and the observed Ovi cumulative line-number density and
bO VI−NO VIcorrelation is reported by Oppenheimer & Dav´ e (2009).
These authors argue that turbulence is the crucial mechanism lead-
ing to the large b-values measured in IGM Ovi systems, and solve
the discrepancy between their simulations and observations by
adding sub-resolution turbulent broadening as a function of hy-
drogen density – partly constrained by observations– to the Dop-
pler parameter of their simulated absorption lines. Quite remark-
ably, they find that this approach, besides reproducing the b−NO VI
correlation by construction, simultaneously brings their predicted
equivalent width distribution into better agreement with observa-
tions. Since the CDDF slope resulting from our simulation is con-
sistent with the data, while the b-value distribution is not, it is quite
plausible that broadening our Ovi lines following the approach by
Oppenheimer & Dav´ e (2009) would help to reconcile our results
with observations. Although this is tempting, we will not consider
it here. Instead, we will present and discuss a series of modifica-
tions to our reference run (i.e. simulation runs with different para-
meters) and post-run variations thereof – including the addition
of sub-resolution turbulence–, and their implications for Ovi ab-
sorbers statistics in a future paper. For now, we will proceed with
the analysis of the physical properties of the Ovi bearing gas in our
reference run as is.
5PHYSICAL PROPERTIES OF Ovi ABSORBERS
In this section we analyse the physical properties (density, temper-
ature, metallicity, ionisation state, baryon content) of the Ovi bear-
ing gas traced by the absorption features identified in our synthetic
spectra. For the sake of simplicity, we restrict ourselves to the ana-
lysis of our fiducial run at z = 0.25, which roughly corresponds
to the median redshift of the Ovi observations found in the liter-
ature. Hence, in what follows, we will use our Ovi line sample 2
introduced at the end of Sec. 3.1.
5.1Optical depth-weighted physical quantities
Defining physical properties such as the density or temperature of
the gas responsible for the absorption features identified in simu-
lated spectra is not a trivial task, and different methods have been
described in the literature. Oppenheimer & Dav´ e (2009, their equa-
tion 7), for example, ascribe physical properties to simulated Hi
and Ovi absorption systems by weighting the desired quantity by
the product of the SPH particle mass, the mass fraction of the cor-
responding element (e.g. oxygen), and the ionisation fraction of
the corresponding species (e.g. Ovi ) at the line centre, taking both
peculiar and thermal velocities into account. Schaye et al. (1999),
on the other hand, define the density (temperature) of the absorb-
ing gas at the line centre as the sum of the density (temperature)
of all gas elements that contribute to the absorption in that pixel,
weighted by their contribution to that pixel’s optical depth. A sim-
ilar approach is followed by Richter et al. (2006) to ascribe a dens-
ity and a temperature to the absorption features identified as Broad
Lyα Absorbers (BLAs) in their simulations.
By definition, optical-depth weighted physical quantities dir-
ectly relate an absorption feature identified in a synthetic spectrum
to the absorbing gas in the simulation, and this gives insight about
the physical state of the gas in which absorption actually takes
place. This in turn allows for a meaningful comparison between
quantities obtained from a spectral analysis (e.g. , column density,
Doppler parameter) and the ‘true’ physical properties as given in
the simulations (e.g., volume density, temperature). For example,
for a uniform density, uniform temperature gas cloud, the optical-
depth weighted properties of the cloud would be simply its given
density and temperature. In case there is, e.g., a density (temper-
ature) gradient towards the centre, then naturally the denser gas
will be more important in determining the absorption line prop-
erties. The optical-depth weighted quantity properly reflects this,
and provides the properly weighted average density or temperat-
ure across the absorbing cloud. We therefore choose to use optical-
depth weighted physical quantities for our analysis of simulated
spectra. In contrast to the approaches mentioned above, however,
we do not only consider the line centre, but compute a weighted
average over the line profile. So, to compute a physical quantity,
say the density, associated with a certain line, we first compute the
optical-depth weighted density in redshift space along the sightline
as in Schaye et al. (1999). Next we compute the average of the
optical-depth weighted density over the line profile, weighted by
the optical depth of each pixel and assign this last weighted aver-
age to the line. We have compared the physical quantities resulting
from this approach with using just the value of the optical-depth
weighted quantity at the line centre and found that the difference is
small, with averaging over the full profile typically giving slightly
smaller values.
In what follows, quantities weighted by Ovi optical depth
will be denoted by adding a corresponding subscript; thus, for ex-
ample, the optical-depth weighted gas overdensity (temperature) is
denoted by ∆O VI(TO VI). Note that we define ∆ ≡ ρb/?ρb?, where
?ρb? = 4.18 × 10−31(h/0.73)2g cm−3is the mean cosmic baryon
density. An example of the optical-depth weighted temperature and
overdensity along a random sightline through a simulation box at
z = 0.25, as well as the corresponding quantities for a given Ovi
absorption line are shown in middle and bottom panels of Fig.1.
5.1.1 Temperatures and overdensities
In a recent study, Wiersma et al. (2009b) investigated the distri-
bution of metals using a simulation run which assumed the same
reference model as our fiducial run but at 8x lower mass resolu-
tion and twice as large a box size. Their results, which were shown
to be insensitive to box size and nearly converged for both their
and our fiducial runs, show that the diffuse, photo-ionised IGM,
i.e. , gas at overdensities ∆ ? 101.5and temperatures 103K ?
T ? 104.5K, harbours a large fraction of the mass in the simu-
lation, but contains a negligible fraction of the metals. They find
that the metals are mostly spread over low-density structures at
temperatures T ? 105K, despite the inclusion of metal-line cool-
ing, indicating that the Warm-Hot Intergalactic Medium (WHIM)
at 105K ? T ? 107K and ∆ ? 103contains a significant amount
of metals (Wiersma et al. 2009b, their Figure 10). As discussed by
these authors, the reason that a very high fraction of the metals are
in low-density gas at T ? 105K is that high-velocity winds trans-
port metals from galaxies out to large distances, and these winds
shock-heat the gas to such high temperatures.
Motivated by their findings, here we want to investigate if and
how the baryon and metal distributions, as well as different gas
phases in our simulation are traced by Ovi detected in absorp-
tion. To this end, we compute an optical-depth weighted temper-
ature, TO VI, and overdensity, ∆O VI, for each of our Ovi absorbers
in sample 2, and compare the resulting (TO VI, ∆O VI)-distribution to
the intrinsic mass and metal distributions in our simulation.
The top panel of Fig. 3 shows the gas mass distribution
∂2Mgas/(Mgas,tot∂log∆∂logT) (coloured areas) at z
=
0.25
c ? —- RAS, MNRAS 000, 1–18

Page 8

8Tepper-Garc´ ıa et al.
and the distribution of Ovi absorbers (white contours) in the
temperature-density plane. For the case of the absorbers (contours)
the axes correspond to the Ovi optical-depth weighted temperature
and density. The colour coding shows the amplitude of the gas mass
distribution. The red/green/blue/black areas enclose 50/75/90/100
per cent of the total gas mass, with the gas mass fraction in each
bin given by the value indicated in the colour bar. The white con-
tourscontain,startingfromtheinnermost,25,50,75,90,and99per
cent of the total number (3034) of identified Ovi absorbers. For ref-
erence and for the subsequent analysis, we include the sub-panels
below and to the right which show, respectively, the distributions
(normalised to their peak value) marginalised over ∆ (lower sub-
panel)andT (rightsub-panel),ofgasmass(red),andOviabsorbers
(black). Note that we have excluded from this and all subsequent
phasediagramsthestar-forminggas(i.e.,theinterstellarmedium)–
defined as gas with densities exceeding our adopted star-formation
threshold n∗
z = 0.25 –, since the temperature of this gas simply reflects the
pressure, imposed via an equation of state P ∝ ρ4/3, of the un-
resolved multiphase ISM (see Schaye & Dalla Vecchia 2008, for
details).
Focusing first on the gas mass distribution, we can see that
approximately half of the baryonic mass in our simulation is
contained in a phase corresponding to the diffuse IGM (red re-
gion), which shows a tight correlation between temperature and
(over)density governed by the balance between adiabatic cooling
duetothecosmicexpansionandphoto-heatingbythemeta-galactic
UV background (Hui & Gnedin 1997). We also find a signific-
ant amount of gas at low overdensities (1 ? ∆ ? 102) and much
higher temperatures (105K ? T ? 107K), corresponding to shock-
heated material. The plume of the distribution at ∆ ? 103and
T ? 107K , represents hot gas in galaxy clusters, i.e. the so-called
intracluster medium (ICM). Note that there is no well-defined de-
marcation between gas phases at different temperatures and over-
densities (see red histograms); rather, the transition between these
phases is smooth.
In contrast to the extended temperature and overdensity distri-
butions of the gas mass in our simulation, the distribution of TO VI
and ∆O VIshown by the white contours is constrained to a relat-
ively narrow temperature range 104.5K ? T ? 106K and overdens-
ities 1 ? ∆ < 103. Our simulation thus indicate that Ovi traces
shock-heated material with temperatures around T ∼ 105.3±0.5K,
and at slightly higher overdensities than the diffuse IGM, around
∆ ∼ 10 − 102typically. The comparison to the overall gas mass
distribution also suggests that Ovi traces a significant amount of
the baryons in the simulation. We will discuss the baryon content
of Ovi bearing gas in more detail in Sec. 5.1.3, and will show that,
contrary to our expectation, the Ovi bearing gas contains only a
small fraction of the warm-hot baryons.
The middle panel of Fig. 3 shows the gas metal mass distri-
bution ∂2MZ/(MZ,tot∂log∆∂logT). Each particle’s metal mass is
defined by mZ≡ Zsm× mg, where Zsmis the smoothed metallicity
(see Sec. 5.1.2 and Wiersma et al. 2009b, for a detailed discussion
on smoothed metallicities; we note that this plot would look nearly
identical if particle metallicities had been used). Again, the col-
H= 0.1cm−3which corresponds to7∆ ≈ 3 × 105at
7The relation between hydrogen number density nHand (baryonic) over-
density ∆ is given by
nH=?ρb?
mH
XH(1 + z)3∆ ≈ 1.9 × 10−7cm−3?XH
0.752
?
(1 + z)3∆
10−1
100
101
102
103
104
105
∆
−6−5.5−5−4.5−4−3.5−3−2.5
log (fraction of gas mass)
103
104
105
106
107
108
Temperature [K]
10−1
100
101
102
103
104
105
∆
−6−5.5−5−4.5−4−3.5−3
log (fraction of gas metal mass)
103
104
105
106
107
108
Temperature [K]
10−1
100
101
102
103
104
105
∆
−6 −5.5−5−4.5−4−3.5−3−2.5
log (fraction of OVI mass)
103
104
105
106
107
108
Temperature [K]
Figure 3. Distribution (coloured areas) of gas mass (top), gas metal mass
(middle), and Ovi mass (bottom) in the temperature-overdensity plane at
z = 0.25, together with the distribution of Ovi absorbers (white contours) of
sample 2. The histograms at the bottom and to the right of each panel show
the corresponding gas (red) and Ovi absorbers (black) distributions mar-
ginalised over T and ∆, respectively, normalised to their peak values. The
colour coding has been chosen in such a way that the red/green/blue/black
areas contain 50/75/90/100 per cent of the total gas mass (top), total gas
metal mass (middle), total Ovi mass (bottom), respectively. The white con-
tours enclose, starting from the innermost, 25, 50, 75, 90, and 99 per cent
of the total number of Ovi absorbers. Note that the white contours and the
black histograms are the same in all three panels.
c ? —- RAS, MNRAS 000, 1–18

Page 9

Ovi absorbers at low redshift9
our scheme shows the amplitude of the distribution with the colour
cuts chosen in such a way that the red/green/blue/black areas en-
close 50/75/90/100 per cent of the total gas metal mass, with the
gas metal mass fraction in each bin given by the value indicated in
the colour bar. The white contours are the same as in the top panel.
Clearly, we can distinguish two phases that harbour a substantial
fraction of the metals in the simulation at z = 0.25: shock heated
material at temperatures T ? 105K and overdensities ∆ ? 103,
and cold-warm gas at T ? 104.5K and ∆ ? 102. In agreement with
Wiersma et al. (2009b, their Fig. 7) we estimate that these phases
together contain ∼ 30 per cent of the total metals in the simula-
tion, while the rest is contained in star-forming gas (∼ 10 per cent)
or locked up in stars (∼ 60 per cent). We find that nearly 20 per
cent of the metals in our simulations are contained in WHIM gas
at 105K ? T ? 107K, but the Ovi absorbers together contain ony
< 0.1 per cent of the metal budget. Hence, Ovi absorbers do not
trace the bulk of metals, even though the majority arise in gas with
a high metal fraction, as can be judged from the (TO VI, ∆O VI) dis-
tribution.
Presumably, there is a continuous exchange of material
between the cold-warm, diffuse IGM and these metal-rich gas
phases. Intergalactic gas is accreted onto the potential wells of
galaxies and thereby shock-heated. At the same time, galactic
winds shock-heat and transport metal-rich gas from the high-
density, star-forming regions into the low-density IGM, some of
which may then cool down and may eventually be re-accreted onto
galaxies to fuel star formation (“intergalactic fountain”). In this
scenario, and given the location of the Ovi distribution with respect
to the gas metal mass distribution (middle panel), Ovi absorbers
would mainly arise in enriched, shock-heated galactic-wind ma-
terial that has probably started to cool down, wandering from the
high-temperature, low-density region to the low-temperature, high-
density regime on phase space. As such, many of the Ovi absorbers
tracing warm-hot gas may be short-lived. We will provide evidence
for this later on when we discuss the metallicity of Ovi bearing gas
below and in Sec. 5.1.2.
To better understand the nature of these Ovi absorbers, we
investigate if (and how much of) the Ovi bearing gas in our sim-
ulation is traced by the Ovi we see in absorption. To this end,
we compare in the bottom panel of Fig. 3 the Ovi mass distribu-
tion ∂2MO VI/(MO VI,tot∂log∆∂logT) to the distribution of Ovi ab-
sorbers in the (TO VI, ∆O VI) plane. A particle’s oxygen mass is given
by mO VI≡ (nO VI/nO) × XO× mg, where XOis the smoothed oxy-
gen mass fraction. Colors and contours have the same meaning as
in the middle and top panels. We find that the bulk of the Ovi is
found in gas at moderate overdensities ∆ ∼ 100.5− 102.5, and tem-
peratures between T = 104K and 106K, which is consistent with
the overall distribution of our Ovi absorbers. Also, the marginal-
ised temperature distribution (lower sub-panel, red histogram) is
clearlybi-modal,showingthatOviisdistributedamongtwophases
with different temperatures: T ∼ 104.5K, corresponding to photo-
ionised gas, and T ∼ 105.5K, corresponding to collisionally ionised
gas. In contrast, and quite remarkable, is the fact that the Ovi seen
in absorption traces mainly the hotter phase, although at slightly
lower temperatures, T ∼ 105.3K, while the cooler Ovi bearing gas
phase is only marginally traced. Note that this is not an effect of the
limited S/N of our spectra, nor of a limited resolution, since qual-
itatively the same result is found in noise-free spectra with very
high resolution. As can be judged by the amplitude of the (two-
dimensional) distribution around T ∼ 104.5K and 10 < ∆ < 102,
the Ovi content in this gas phase is low, which could mean that also
theOvicolumndensityistoolowforthisgastobedetectableinab-
sorption.Ourassumptionisreinforcedbythefact,goingbacktothe
middle panel, that the overall metal content of gas at temperatures
and overdensities comparable to the low temperature Ovi bearing
gas is quite low. The fact that the peak temperature of the hotter
phase is slightly shifted to lower values as compared to the high-
temperature peak of the intrinsic temperature distribution suggests
that Ovi absorption arising in cooler gas phases overlaps in redshift
space with Ovi absorption from hotter gas, thus leading to slightly
lower optical-depth weighted gas temperatures. To a lesser extent,
it is a consequence of averaging optical-depth weighted quantities
over the full line profile, as discussed in Sec. 5.1.
An interesting fact is that the low temperature Ovi bearing gas
closes the gap between the high-temperature, low-density and the
cold, high-density metal-rich gas phases shown in the middle panel,
which suggests that some of the gas traced by Ovi has already
cooled down to an equilibrium temperature T ∼ 104.5K, which
is typical for enriched gas at z = 0.25 (see Sec. 5.3, Fig. 9).
Summarising, we find that, while the mass in our simulation is
more or less equally distributed between two phases with temperat-
ures T ? 104.5K and T ? 104.5K, and overdensities 0.1 ? ∆ ? 103,
Ovi traces gas at high temperatures T > 105K and moderate over-
densities 10 < ∆ < 102, which are typical for the warm-hot inter-
galactic medium (WHIM) in our (Wiersma et al. 2009b) and other
simulations (Cen & Ostriker 1999; Dav´ e et al. 2001; Bertone et al.
2008; Tornatore et al. 2009). This gas phase typically contains ∼ 40
of the baryons in our simulation, but the Ovi bearing gas turns out
to contain only a small fraction of the cosmic baryons at low red-
shift (see Sec. 5.1.3). Furthermore, we find that roughly 40 per cent
of the metals in our simulation are distributed between warm-hot
gas at moderate to high overdensities 1 < ∆ < 103(∼ 30 per
cent), and cold star-forming gas at very high densities (∼ 10 per
cent). While Ovi certainly arises in gas containing metals, it is not
a tracer of the main metal reservoirs in our simulation.
Our result that absorption by Ovi in our simulations is pref-
erentially related to gas at temperatures 104.5K < T < 106K, with
a high fraction (more than 65 per cent) having temperatures fall-
ing in the range 105K < T < 106K, i.e. , in the low temperature
regime of the WHIM, is quite interesting, since Oppenheimer &
Dav´ e (e.g. 2009) find exactly the opposite trend, namely that the
overwhelmingly majority of the Ovi in their simulations is found
in gas a temperatures T ∼ 104.2±0.2K, typical of photo-ionised gas.
It is important to note that our simulation does show a gas phase
containing Ovi at photoionisation temperatures (see Fig. 3, bottom
panel), which is however not detected in absorption, most probably
due to its low overall metal content. In Sec.5.3 we will present a
deeper analysis of the disagreement between the findings by Op-
penheimer & Dav´ e (2009) and our results, which we will attribute
to their neglect of the effect of photo-ionisation on metal-line cool-
ing.
5.1.2Metallicity
In this section, we analyse the local metallicity of the gas traced by
the Ovi absorbers (in sample 2). We proceed in the same manner
as with gas temperatures and overdensities, and estimate optical-
depth weighted metallicities ZO VIfor each of our identified Ovi ab-
sorbers.Notethatanaccuratemeasurementoftheelementalabund-
ances in the absorbing gas is very important, since this quantity is
crucial for an indirect estimate of the baryon fraction in the gas
traced by a given absorption species (see Sec. 5.1.3).
In our simulation, as is done in all OWLS runs, we trace how
stellar evolution alters the abundance of 11 elements (H, He, C, N,
c ? —- RAS, MNRAS 000, 1–18

Page 10

10Tepper-Garc´ ıa et al.
10−6
10−5
10−4
10−3
10−2
10−1
100
101
Metallicity [Z⊙]
−6.5 −6−5.5 −5−4.5 −4−3.5
log (fraction of gas mass)
10−1
100
101
102
103
104
105
∆
10−6
10−5
10−4
10−3
10−2
10−1
100
101
Metallicity [Z⊙]
−6−5.5−5−4.5 −4−3.5−3
log (fraction of gas mass)
103
104
105
106
107
108
Temperature [K]
10−2
10−1
100
101
Metallicity [Z⊙]
−6−5.5−5 −4.5−4−3.5−3
log (fraction of gas metal mass)
10−1
100
101
102
∆
103
104
105
10−2
10−1
100
101
Metallicity [Z⊙]
−6−5.5 −5−4.5−4 −3.5−3
log (fraction of gas metal mass)
103
104
105
106
107
108
Temperature [K]
Figure 4. Top row: Gas mass distribution (coloured areas) on the overdensity-metallicity (left) and temperature-metallicity (right) planes, together with the
distribution of the Ovi absorbers (white contours) for our fiducial run REF L050N512 at z = 0.25. Note that the marginalised overdensity distributions (lower
sub-panels) are identical to those in the top panel of Fig. 3 (right sub-panel). Bottom row: Gas metal mass distribution on the overdensity-metallicity (left)
and temperature-metallicity (right) planes for our fiducial run REF L050N512 at z = 0.25. White contours are as in the top row. Note that the marginalised
temperature distributions (lower sub-panels) are identical to those in the middle panel of Fig. 3 (lower sub-panel). Colour coding and contour levels are as in
Fig. 3. We use a solar metallicity of Z?= 0.0127 to express the absolute metallicities predicted by the simulation in solar units.
O, Ne, Mg, Si, Fe, Ca, S) explicitly, while simultaneously follow-
ing the metallicity of each particle (see Wiersma et al. 2009b, for
further details). We choose to use smoothed metallicities – as op-
posed to particle metallicities – for the various processes that are
metallicity-dependent, in particular for the calculation of radiative
metal cooling. The smoothed metallicity of a particle is defined
as the ratio of the SPH estimates of the metal mass density and
the total gas mass density at the location of the particle, while the
particle metallicity is given by the ratio of the particle’s metal mass
and its total gas mass. Wiersma et al. (2009b) have shown that
in low-metallicity gas, smoothed metallicities are generally higher
than particle metallicities. The use of smoothed metallicities in part
counters (but does not solve) the lack of metal mixing inherent
to SPH, which leads to differences when compared to the use of
particle metallicities (although these differences decrease with in-
creasing resolution). First, the use of smoothed metallicities results
in higher fractions of metals residing in gas both at lower temperat-
ures (T < 105K as compared to T ∼ 106K) and lower metallicities.
Second, the use of smoothed abundances enhances radiative cool-
ing and increases the predicted SFR. These differences may be im-
portant for the comparison of our results to the results from studies
employing particle metallicities (e.g. Oppenheimer & Dav´ e 2009).
In the top row of Fig. 4, we show the gas mass distribu-
tions ∂2Mgas/(Mgas,tot∂logZ ∂log∆) (left panel; coloured areas)
and ∂2Mgas/(Mgas,tot∂logZ ∂logT) (right panel, coloured areas),
together with the distribution of the Ovi absorbers in the (∆, Z)
and (T, Z) planes (white contours), respectively. Again, for the
case of the absorbers the axes correspond to the Ovi optical-depth
weighted temperature TO VI, density ∆O VI, and metallicity ZO VI. The
meaning of colours and contours is the same as in Fig. 3. The his-
tograms in the lower and right sub-panels show the correspond-
ing distributions marginalised over metallicity and overdensity (left
panel) or temperature (right panel) normalised to their peak values.
The panels in the bottom row show the gas metal mass distribu-
tions ∂2MZ/(MZ,tot∂logZ ∂log∆) (left panel; coloured areas), and
∂2MZ/(MZ,tot∂logZ ∂logT) (right panel; coloured areas), as well
as their corresponding marginalised distributions (histograms in the
low and right sub-panels), together with the distribution of the Ovi
absorbers (white contours). Note that these panels follow the same
colour scheme and include identical contours as the corresponding
top-row panels, but that the y-axis range is smaller.
Several interesting conclusions can be drawn from this figure.
First, the top-row panels reveal that a high fraction of the baryonic
mass in our simulation is contained in gas with local metallicit-
ies −2 < log(Z/Z?) < −1. In contrast, Ovi absorbers clearly trace
over-enriched material with metallicities log(ZO VI/Z?) > −1. The
c ? —- RAS, MNRAS 000, 1–18

Page 11

Ovi absorbers at low redshift 11
median metallicity of the gas traced by Ovi is ZO VI≈ 0.6 Z?, which
is almost an order of magnitude higher than the median gas metal-
licity at similar overdensities and temperatures. Note that the use
of optical-depth weighted metallicities ensures that we track (and
average over) the metallicity of all individual particles that contrib-
ute to the Ovi absorption. Hence, ZO VIis an estimate of the local,
i.e. around the spatial resolution limit, metallicity. Note that the
overall mean metallicity of the structures giving rise to the Ovi ab-
sorption – as is derived from observations by using the ratio of the
Ovi column density to the total Hi column density in the absorbing
gas – can be much smaller than ZO VI. In our case, this is supported
by the fact that gas at overdensities and temperatures comparable
to the centres of mass of the (∆O VI, ZO VI) and (TO VI, ZO VI) distri-
butions display a large range of metallicities. The large scatter in
metallicities, in particular at the lowest overdensities, indicates that
the metal distribution is quite inhomogeneous. This will in part be
due to the metal-mixing problem inherent to SPH which is alle-
viated (but by no means solved) by the use of smoothed abund-
ances, as mentioned above. While the implementation of metal dif-
fusion (Greif et al. 2009; Shen et al. 2009) could, depending on the
choice of parameter values, reduce the scatter in the metallicities,
this may in fact be in conflict with observations which indicate that
intergalactic metals are poorly mixed on small scales (Schaye et al.
2007).
Intermsofthemetalmassdistribution,wecanseethatthevast
majority of the metals in our simulation reside in gas with metalli-
cities −1 ? log(Z/Z?) ? 0, temperatures T ? 105K, and overdens-
ities 10−1< ∆ ? 103. From the overlap of the Ovi distribution with
the overall metal mass distribution, we may conclude that the Ovi
we see in absorption traces the moderate-density, warm, enriched
gas relatively well. Nevertheless, despite the coexistence of the Ovi
absorbers with this gaseous phase on the (T, Z)- and (∆, Z)-planes,
the gas traced by Ovi contains actually only a vanishingly small
(less than 0.1 per cent) fraction of the metals (see Sec 5.1.1).
To sum up, the results of this section indicate that Ovi traces
over-enriched gas, which is neither representative for the bulk of
the baryons nor for the bulk of metals. The metallicity of the gas
traced by Ovi is high, with a narrow distribution extending over
the range (0.1, 1) Z?, and a peak at ZO VI≈ 0.6 Z?, which is con-
sistent with results from other simulations (e.g. Cen & Ostriker
2006; Oppenheimer & Dav´ e 2009), and metallicity measurements
in metal-line absorption systems at low redshift (e.g. Savage et al.
2002; Prochaska et al. 2004; Tripp et al. 2005; Jenkins et al. 2005;
Cooksey et al. 2008). In spite of the agreement with observations,
we want to caution again that our quoted values refer to local es-
timates of the metallicity, while observed values are generally de-
rived from metal-line systems aligned (i.e. within a given range
in velocity space) with Hi absorption, which can lead to signific-
antly lower mean metallicities if the metal distribution is patchy
(see Schaye et al. 2007, for a discussion).
5.1.3Baryon content
The baryon content of gas traced by Ovi is a quantity of great in-
terest since it has long been anticipated by cosmological simula-
tions that nearly half of the baryons in the Universe (absent in ob-
servational inventories, see e.g. Fukugita 2004) might be hidden in
the WHIM (Cen & Ostriker 1999; Dav´ e et al. 2001; Bertone et al.
2008), and that this gas phase might be detectable through its dif-
ferent emission and absorption signatures, in particular absorption
by Ovi (Cen & Ostriker 1999).
A standard approach used in the literature to indirectly estim-
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
10−3
10−2
(nOVI/nO)OVI
10−1
100
PDF
−1−0.5 0 0.5 1
[O/H]OVI
0
0.68
0.2
0.4
0.6
0.8
1
0.7 0.72
(XH)OVI
0.74 0.76
PDF
10−3
10−2
10−1
100
Ωb(OVI)/Ωb
Figure 5. Distribution of optical-depth weighted Ovi fraction (top-left),
local oxygen abundance (top-right), and local hydrogen mass fraction
(bottom-left) for our sample of 3034 Ovi absorbers at z = 0.25 in spec-
tra with S/N = 50 (sample 2). The bottom-right panel shows the distri-
bution of the baryon density, relative to the cosmic mean, traced by Ovi
along individual sightlines. Note that we use the default cloudy solar oxy-
gen abundance (nO/nH)?= 4.9 × 10−4to compute [O/H]O VI.
ate the baryon fraction of Ovi bearing gas is to write
Ωb(Ovi) =mH
ρc

c
H0
NLOS
?
i=1
∆χi

−1NLOS
?
i=1
Nabs
?
j=1
NO VI,ij
XH(nO VI/nO) (nO/nH),
(1)
where mHis the hydrogen mass, ρcthe critical density, and XHthe
(assumed) hydrogen mass fraction. The factor in brackets is the
total surveyed physical length with ∆χi the so-called absorption
path length (see e.g. Schaye 2001) along an individual sightline.
One major caveat of this approach is that the estimated value
for Ωb(Ovi) is highly sensitive to both the ion fraction and the
metallicity of the gas and that, owing to the difficulties in measur-
ing these quantities, plausible values need to be assumed. Usually,
the Ovi ion fraction is set to (nO VI/nO) = 0.2 – corresponding to
the peak ion fraction of Ovi at T ∼ 2 − 3 × 105K either in CIE
(Sutherland & Dopita 1993) or non-equilibrium (Gnat & Sternberg
2007) –, and the metallicity to [O/H] = 0.1 dex (see e.g. Tripp
et al. 2000).
Here we want to estimate the baryon content of the gas traced
by our sample of Ovi absorbers, and we want to do this in such a
way that we can compare our result to the observed range of val-
ues. Therefore, we follow the above approach, i.e., using equation
(1), but exploit the advantage of our simulation, which allows us to
measuretheionfraction(nO VI/nO),oxygenabundance(nO/nH),and
hydrogen mass fraction XHof the gas giving rise to each Ovi ab-
sorber. To this end, and to be consistent, we again use optical-depth
weighted quantities, estimated for the Ovi absorbers in sample 2.
The resulting distributions of the values for each of these quantities
are shown in Fig. 5 (top row; bottom-left).
Inserting the values for each of the latter quantities and
for each identified Ovi absorber in equation (1), we obtain
Ωb(Ovi)/Ωb = 0.014 over 5000 sightlines spanning a total ab-
sorption path ∆χ = 139.3. The distribution of the baryon content
along individual sightlines is shown in the bottom-right panel of
c ? —- RAS, MNRAS 000, 1–18

Page 12

12Tepper-Garc´ ıa et al.
Fig. 5. Using the same 5000 spectra8, this time with S/N = 10
rather than S/N = 50, we find Ωb(Ovi)/Ωb= 0.0096. Either value
is lower than values estimated from observations which range from
Ωb(Ovi)/Ωb? 0.019 (Tripp et al. 2000), through ∼ 0.045 (Thom
& Chen 2008b), ? 0.054 (Sembach et al. 2004), and 0.086 ± 0.008
(Danforth & Shull 2008).
Two things are noteworthy. First, as can be judged from the
bottom-right panel of Fig. 5, the distribution of Ωb(Ovi) along in-
dividual sightlines shows a large scatter, reflecting the effect of cos-
mic variance. This indicates that measurements of the baryon con-
tent using data taken along different sightlines might vary drastic-
ally, even in case that the ionisation fraction and the metallicity
are known to high accuracy, which they are usually not (although
note that our sightlines span ∆z = 0.02 at z = 0.25). Second, and
related to the previous point, is the fact that the median optical-
depth weighted ionisation fraction resulting from our simulation
is (nO VI/nO)O VI ≈ 0.052, i.e., is a factor four smaller than usu-
ally assumed, and the median optical-depth weighted metallicity
is [O/H]O VI≈ −0.19 dex or about six times larger than the ‘canon-
ical’ 10 per cent solar-value. Also, the distribution of both these
quantities shows a large dispersion (see top panels in Fig. 5). This
in turn implies that baryon density measurements from observa-
tions which assumed fixed values for the Ovi ionisation fraction
and the metallicity may be off (most probably overestimated) by
non-negligible factors, independent of the effect of cosmic vari-
ance. This in turn might explain why we find an average baryon
density which is lower than the various measured values.
5.2Physical conditions of the gas and Ovi observables
Now that we have analysed the physical conditions (density, tem-
perature, etc.) of the gas giving rise to the Ovi absorption in our
simulation, we want to make a connection between these and the
actual Ovi observables, i.e. the Ovi column density, NO VI, and the
Ovi line width as measured by the Doppler parameter, bO VI.
For this purpose we bin the set of (bO VI, NO VI) values for our
Ovi sample 2 using ∆bO VI= 2.0 kms−1and ∆logNO VI= 0.2 dex,
and compute in each cell the median of the physical quantity un-
der consideration, e.g. density. The result of this exercise is shown
in Fig. 6. The panels show the median overdensity (top-left), tem-
perature (top-right), hydrogen mass fraction (middle-left), metalli-
city (middle-right), oxygen abundance (bottom-left), and Ovi ion-
isation fraction (bottom-right) as a function of bO VIand NO VI. For
reference, we have included contours showing the distribution (by
number) of Ovi absorbers in the bO VI−NO VIplane. These contours
contain, starting from the innermost, 25, 50, 75, 90, and 99 per cent
of the identified Ovi absorbers.
Looking at the top-left panel, we can see that there is a
mild correlation between column density and overdensity, with
no apparent correlation between the line width and overdensity,
which is consistent with our missing bO VI− NO VIcorrelation (see
Fig. 2, bottom-right panel). In contrast, there is a strong correl-
ation between Doppler parameter and the optical-depth weighted
gas temperature, as shown by the top-right panel. Furthermore, the
middle-right panel shows a good correlation between column dens-
ity and gas metallicity, i.e. a metallicity-density relationship. This is
consistent with the results shown in the middle-left and bottom-left
panels, where we can see that the hydrogen mass fraction of the gas
8In this case we identify a total of 1044 Ovi components rather than 3034.
giving rise to Ovi absorption decreases with NO VI, and correspond-
ingly, the oxygen abundance increases. Finally, the bottom-right
panel shows that there is no clear trend of the ionisation state of the
Ovi bearing gas with bO VInor with NO VI.
All these results together imply that some information about
the physical conditions can be gained from the measured Ovi
column densities and line widths alone, but only in a statistical
sense, given the large scatter in the correlations between directly
observable and derived physical quantities. In particular, the bot-
tom panels of Fig. 6 reinforce our previous statement that assuming
a fixed value for the oxygen abundance and Ovi ionisation fraction
to estimate the baryon content of Ovi bearing gas is dangerous.
5.2.1Gas temperature, Doppler widths, and line strengths
One of the main purposes of recent observational studies about in-
tergalactic Ovi absorption at low redshift has been to determine the
temperature, and thus the ionisation state, of the gas giving rise to
the observed absorption. Some of these (e.g. Thom & Chen 2008b;
Tripp et al. 2008) have estimated the gas temperature using the line
width of well-aligned, i.e., within some velocity uncertainty σ(∆v),
Ovi - Hi absorbers, and have found that the implied gas temperat-
ure are typically T < 105K, indicating that the associated Ovi is
mainly photo-ionised. This approach, however, implicitly assumes
that Ovi and Hi absorbers arise in the same gas, which might not
be true in general, even for small velocity displacements. As also
noted by Thom & Chen (2008b) and Tripp et al. (2008), the gas is
most probable multi-phase, and a broader Hi component related to
the Ovi absorption might be too weak to be detected, particularly
if the metallicity of the component is high. Such a broad compon-
ent would relax the upper limit on the temperature imposed by the
Hi line width, leading to higher gas temperatures. Indeed, Tripp
et al. (2008) have estimated an upper limit on the temperature of
the Ovi bearing gas using the Ovi line widths alone and assuming
pure thermal broadening, and they find significantly higher temper-
atures than those allowed by well-aligned OVI - Hi absorbers.
Danforth & Shull (2008) find a strong correlation between the
column densities of Ovi and Nv absorbers, and use this to model
the ionisation state of the absorbing gas, concluding that the ob-
served column density ratios are consistent with collisionally ion-
ised gas at T = 105.3±0.1K, assuming a solar (N/O) abundance.
Danforth& Shull(2008)notethat althoughtheobservedNO VI/NN V
ratios are also consistent with pure photo-ionisation models, the
implied ionisation parameters, the required metallicities, and/or the
spectral hardness of the photo-ionising radiation are not compatible
with other constraints.
Given the current disagreement about the temperature, and
hencethe ionisationstate,of thegastraced bytheobserved Ovi,we
inspect the correlation between Ovi line width and gas temperature
suggested by the top-right panel of Fig. 6 in more detail. For this
purpose, we bin the TO VIvalues using bins of size ∆bO VI= 2kms−1
for all identified Ovi absorbers with bO VI< 40 kms−1in sample 2,
and compute the median, and 25th and 75th percentiles in each bin.
The result is shown in Fig. 7. The relation between gas temperature
and Doppler parameter assuming pure thermal broadening,
TO VI=mOb2
O VI
2k
≈ 9.7 × 104K
?
bO VI
10 kms−1
?2
,
(2)
has been included in this plot for reference (blue dashed curve).
Clearly, there is a tendency for the the optical-depth weighted
gas temperature to increase with bO VI. For bO VI> 15kms−1(cor-
responding to log(T/K) > 5.3), non-thermal broadening becomes
c ? —- RAS, MNRAS 000, 1–18

Page 13

Ovi absorbers at low redshift13
bOVI [km s−1]
NOVI [cm−2]
1012
1013
1014
1015
1.0
1.2
1.4
1.6
1.8
2.0
log ∆OVI
bOVI [km s−1]
NOVI [cm−2]
4.6
4.8
5.0
5.2
5.4
5.6
log TOVI / K
bOVI [km s−1]
NOVI [cm−2]
1012
1013
1014
1015
0.71
0.71
0.72
0.72
0.73
0.73
0.74
log (XH)OVI
bOVI [km s−1]
NOVI [cm−2]
−0.6
−0.5
−0.4
−0.3
−0.2
−0.1
0.0
log ZOVI / Z⊙
bOVI [km s−1]
NOVI [cm−2]
5 10 15 20 25 30 35 40
1012
1013
1014
1015
−3.9
−3.8
−3.7
−3.6
−3.5
−3.4
−3.3
−3.2
(nO /nH)OVI
bOVI [km s−1]
NOVI [cm−2]
5 10 15 20 25 30 35 40
−2.0
−1.8
−1.6
−1.4
−1.2
−1.0
log (nOVI /nO)OVI
Figure 6. Physical conditions of the Ovi bearing gas as a function of Ovi observables, i.e. column density and Doppler parameter. The colour scale shows,
for each (bO VI,NO VI)-cell of size ∆bO VI= 2.0 kms−1and ∆logNO VI= 0.2 dex, the median overdensity (top-left), temperature (top-right), hydrogen mass
fraction (middle-left), metallicity (middle-right), oxygen abundance (bottom-left), and Ovi ionisation fraction (bottom-right). For reference, we have included
contours showing the distribution (by number) of Ovi absorbers on the bO VI− NO VIplane. These contours contain, starting from the innermost, 25, 50, 75,
90, and 99 per cent of the identified Ovi absorbers and they are the same in all panels.
progressively more important, leading to larger b-values for a given
temperature than expected for pure thermal broadening. Note that
for the bin just above the resolution limit in our synthetic spectra,
bmin= 4.2 kms−1, the median temperature is larger than expected
for pure thermal broadening, as is the case for the 75th percentiles
at higher b-values, which is unphysical. As discussed in Sec. A2,
this is an artifact of our fitting procedure in which weak, narrow
components are added to broad, shallow absorption features to im-
prove the fit. Some of these broader components arise in gas with
relatively high temperatures and, correspondingly, high optical-
depth weighted temperatures are assigned to the related (narrow)
components.
From the previous discussion we can conclude that line widths
do reflect the temperature of the absorbing gas in our simulation, al-
though in a statistical sense, but that no definite conclusion can be
reached for individual absorbers, unless further information, e.g.
other related absorbing ions, are available. Also, it is plausible that
the correlation between gas temperature and line widths might be
washed out by line-broadening mechanisms such as turbulence,
which are not resolved by our simulation, or by a too low instru-
mental resolution.
Nevertheless, assuming we would estimate an upper limit on
the gas temperature from the Ovi line widths alone, e.g. by invert-
ing the relation shown in Fig. 7, we would find a good agreement
with the upper limits estimated by Tripp et al. (2008, their Table 7)
from their Ovi line widths, which are, however, much higher than
the temperatures implied by their sample of well-aligned Ovi -
Hi absorbers. This, in turn, suggests that gas temperatures derived
from bH I/bO VIratios alone might be too low. Furthermore, this sup-
ports the idea that there are broad Hi components which truly are
associated with the Ovi components, but which are too weak to be
detected, as anticipated by Tripp et al. (2008).
In addition, Ovi detections might be biased towards lower
temperatures. This is plausible since absorption strength, as meas-
ured by the optical depth at the line centre9, τ0, is proportional to
9The central optical depth of an absorption line with column density N
c ? —- RAS, MNRAS 000, 1–18

Page 14

14Tepper-Garc´ ıa et al.
104
105
106

TOVI [K]

0 5 10 15 20 25 30 35 40
bOVI [km s−1]
Figure 7. Comparison between fitted Doppler parameters and Ovi gas tem-
peratures for all components (3034) in sample 2. The filled dots show the
median value in each bin of size∆bO VI= 2kms−1(indicated by the x-bars),
while the y-bars show the 25th and 75th percentiles, respectively. The blue
dashed curve shows the relation between temperature and Doppler width
assuming pure thermal broadening (see text for details). The (normalised)
histogram in the sub-panel shows the distribution of bO VIvalues.
104
105
106

1011
1012
1013
1014
TOVI [K]
NOVI /bOVI [cm−2 km−1 s]

10−1
100
101

Line Depth (τ0)
Figure 8. Optical depth-weighted gas temperature of the Ovi absorbing
gas as a function of central optical depth τ0(or line strength NO VI/bO VI)
for all components identified in 5000 synthetic spectra with S/N = 50 at
z = 0.25. The blue dashed lines indicate the median gas temperature at the
smallest and largest line depths. The (normalised) histogram in the lower
panel shows the distribution of Ovi components contributing to each bin.
the ratio of the column density to the Doppler width, N/b. Hence,
at a fixed column density, line strength is inversely proportional to
the Doppler parameter and thus, assuming pure thermal broaden-
ing, inversely proportional to the square-root of the gas temperat-
ure. Also, at a fixed b-value, line strength is directly proportional to
column density, and given the correlation between column density
and metallicity presented in Sec. 5.1.2, also directly proportional
and Doppler parameter b is given by
τ0=
√πe2
mecλfN
b≈ 2.048
?N/1014cm−2
b/10kms−1
?
,
where e and meare the electron’s charge and mass, λ and f are the trans-
ition’s rest wavelength and oscillator strength, respectively, and the last nu-
merical factor is valid for the Ovi λ1031.93 Å transition.
Z = Z⊙
Z = 0.1 Z⊙
106
104
105
107
Temperature [K]
10−1
100
101
102
103
104
∆
Figure 9. Distribution of Ovi mass (coloured regions) and distribution of
Ovi absorbers (white contours) in the temperature-overdensity plane. Note
that these distributions are identical to those shown in the bottom panel of
Fig. 3. Cooling time contours (magenta lines) for two different metallicities
(solar, solid; 10 per cent solar, dashed) split the plane into regions corres-
ponding to temperatures and overdensities for which the cooling time is
shorter (upper part) and longer (lower part) than a Hubble time.
to the gas metallicity. Since metallicity enhances cooling, it is thus
natural to expect that stronger Ovi absorption traces gas at lower
temperatures.
In order to test this, we bin the distribution of gas temperat-
ures TO VIas as function of the line depth (central optical depth) for
our sample 2, and compute the median temperature and the 25th
and 75th percentiles in each bin. The result is shown in Fig. 8.
As we anticipated, we see a clear anti-correlation between the me-
dian gas temperature and absorption strength. We note that Tripp
et al. (2008) estimated their detection threshold to be τ0 ≈ 0.1.
We can see from the histogram in the sub-panel – which shows the
distribution of lines as a function of line strength – that there is
a non-negligible number of weak Ovi absorption features arising
in warm-hot gas, which would not have been detected by Tripp
et al. (2008). Qualitatively, this conclusion is neither affected by
the assumed S/N (the same behaviour is found for spectra with
S/N = 10), nor by the fact that in our synthetic spectra there is
no apparent correlation between b and NO VI. Incidentally, the lat-
ter facilitates the interpretation of the anti-correlation between gas
temperature and line strength shown in Fig. 8 as a signature of en-
hanced cooling in enriched gas, which is discussed in the next sec-
tion.Thebottomlineofthisexerciseisthatobservations,evenifthe
gas temperature is measured accurately, may be biased towards ab-
sorption systems tracing gas at somewhat lower temperatures than
expected for collisionally ionised gas, due to the limited S/N of the
data.
5.3Cooling times
We presented in Sec. 5.1.1 our key result that simulated Ovi
absorbers preferentially trace gas at temperatures in the range
T ∼ 105− 106K, and thus trace gas in the low temperature regime
of the WHIM. This is in marked contrast with the results of Oppen-
heimer & Dav´ e (2009, their Fig. 14), who find that the majority of
Oviabsorbersarise ingas attemperatures103.8? T ? 104.8K with
a peak at T ≈ 104.2±0.2K. The purpose of this section is to address
the source of this discrepancy.
c ? —- RAS, MNRAS 000, 1–18

Page 15

Ovi absorbers at low redshift15
10−25
10−24
10−23
10−22
10−21
104
105
106
107
108
|Λnet| / nH
2 [erg cm3 s−1]
Temperature [K]
PI (Z=0.1Z⊙)
CIE (Z=0.1Z⊙)
Hybrid (Z=0.1Z⊙)
10−25
10−24
10−23
10−22
10−21
104
105
106
107
108
|Λnet| / nH
2 [erg cm3 s−1]
Temperature [K]
PI (Z=Z⊙)
CIE (Z=Z⊙)
Hybrid (Z=Z⊙)
Figure 10. Normalised, absolute net cooling rates at z = 0.25 for gas with nH= 10−5cm−3, for gas with metallicity of 10 per cent solar (left panel) and solar
(right panel), computed assuming collisional ionisation equilibrium (CIE; red), including photoionisation (Wiersma et al. 2009a, PI; black), and including
photoionisation only for gas of primordial composition while assuming CIE for metal-line cooling (Hybrid; blue dashed). See text for details.
Oppenheimer & Dav´ e (2009) find that in their simulations
Ovi traces over-enriched (by factors of four to six) regions with
a clumpy metallicity distribution, and they argue that these re-
gions are thus subject to enhanced metal-line cooling, such that
Ovi bearing gas which was initially shock-heated to temperatures
T > 106K is able to cool to photo-ionised temperatures well
within a Hubble time. Our results are consistent with Oppenheimer
& Dav´ e (2009) inasmuch as we also find that Ovi traces inhomo-
geneously enriched gas with relatively high metallicities, but at sig-
nificantly higher temperatures. In other words, our results indic-
ate that, in our simulation, the vast majority of the gas traced by
Ovi that was shock-heated to temperatures T > 105K has not yet
cool down to photo-ionisation temperatures, in spite of its enhanced
metallicity.
This fact is demonstrated in Fig. 9, where we reproduce the
bottom panel of Fig. 3 showing the Ovi mass distribution (colour
shading) and the distribution of Ovi absorbers (white contours) in
the T − ∆ phase diagram, together with the locus of (T, ∆)-values
for which the cooling time equals the Hubble time for gas of solar
(solid) and ten per cent solar (dashed) metallicity. Note that the
metallicity values used roughly span the range of gas metallicities
traced by our Ovi absorbers (see Fig. 4). Here, the cooling time is
defined as
T
dT/dt=3
tcool≡
2
nkT
|Λnet|,
(3)
where Λnetis the net (i.e., cooling minus heating) radiative cooling
rate at a fixed density. Note that the contour segments to the left
from T ≈ 104.5K correspond actually to heating times. The point
where heating and cooling time contours merge defines the equi-
librium temperature, i.e. the temperature at which the net cooling
rate vanishes and the cooling time becomes effectively infinite (see
also Fig. 10). From Fig. 9, it becomes apparent that, while virtually
all Ovi absorbers in our simulation at z = 0.25 will cool down to
the equilibrium temperature in less than a Hubble time10(in the ab-
sence of further shock-heating), most will still be hot by z = 0. Note
that, as anticipated in Sec. 5.1.1, the low-temperature Ovi bearing
10For reference, the Hubble time τHubble≡ tcosmic(z = 0) = 13.8 Gyr and
tcosmic(z = 0.25) = 10.9 Gyr for our adopted cosmology.
gas has nearly reached the equilibrium temperature T ∼ 104.5K,
typical for photo-ionised Ovi.
The above result strongly suggests that an important factor
leading to the notable difference between the results of Oppen-
heimer & Dav´ e (2009) and our findings regarding the mean tem-
perature of gas traced by Ovi is the way the radiative cooling of gas
is treated in the simulations. As described in Oppenheimer & Dav´ e
(2009), metal-line cooling is included in their simulations using the
models by Sutherland & Dopita (1993), which assume collisional
ionisation equilibrium and fixed relative abundances. They only in-
cluded photo-ionisation for hydrogen and helium. In contrast, as
described in detail in Schaye et al. (2010), the OWLS runs include
radiative cooling implemented according to the method by Wi-
ersma et al. (2009a), who compute cooling rates on an element-by-
element basis in the presence of the cosmic microwave background
(CMB) and an ionising (i.e., UV/X-Ray) radiation field as modeled
by Haardt & Madau (1996). Wiersma et al. (2009a) showed that
including photo-ionisation by the meta-galactic UV/X-Ray back-
groundnotonlydrasticallydecreasesthecoolingratesforhydrogen
and helium, but also for heavy elements. Moreover, they showed
that photo-ionisation of heavy elements shifts the equilibrium tem-
peratures to higher values. In other words, photo-ionisation can
drastically affect the radiative cooling rates of enriched gas and cal-
culations neglecting photo-ionisation will thus underestimate the
temperature of the gas.
A comparison of the cooling rates computed according to each
method is shown in Fig. 10. In each case, the gas is exposed to the
CMB and the UV/X-Ray background at z = 0.25, and has a hy-
drogen density nH = 10−5cm−3(∆ ≈ 50), a hydrogen mass frac-
tion XH= 0.752 – which roughly correspond to the median values
of (nH)O VIand (XH)O VIin sample 2 – and a metallicity of 10 per
cent solar (left panel) and solar (right panel). The black curve in
either panel shows the net cooling rate including photo-ionisation
of heavy elements (PI); the red curve, the net cooling rate assum-
ing pure collisional ionisation equilibrium (CIE) and fixed (solar)
abundances, while the blue curve shows the net cooling rate as-
suming CIE, but including photo-ionisation for hydrogen and he-
lium (Hybrid). The dip apparent in each cooling curve shows the
equilibrium temperature. Note, in particular, the difference in equi-
librium temperatures and overall amplitude between the black and
blue curves, which correspond to the cooling rates implemented in
c ? —- RAS, MNRAS 000, 1–18