Spotting high-z molecular absorbers using neutral carbon: Results from a complete spectroscopic survey with the VLT

While molecular quasar absorption systems provide unique probes of the physical and chemical properties of the gas as well as original constraints on fundamental physics and cosmology, their detection remains challenging. Here we present the results from a complete survey for molecular gas in thirty-nine absorption systems selected solely upon the detection of neutral carbon lines in SDSS spectra, without any prior knowledge of the atomic or molecular gas content. H2 is found in all twelve systems (including seven new detections) where the corresponding lines are covered by the instrument setups and measured to have log N(H2)>=18, indicating a self-shielded regime. We also report seven CO detections (7/39) down to log N(CO)~13.5, including a new one, and put stringent constraints on N(CO) for the remaining 32 systems. N(CO) and N(CI) are found to be strongly correlated with N(CO)/N(CI)~1/10. This suggests that the CI-selected absorber population is probing gas deeper than the HI-H2 transition in which a substantial fraction of the total hydrogen in the cloud is in the form of H2. We conclude that targeting CI-bearing absorbers is a very efficient way to find high-metallicity molecular absorbers. However, probing the molecular content in lower metallicity regimes as well as high column density neutral gas remains to be undertaken to unravel the processes of gas conversion in normal high-z galaxies


Introduction
The detection and analysis of molecular absorption lines along the lines of sight to background light sources has proven to be an extremely useful tool to investigate the physical and chemical state of the interstellar medium (ISM) thanks to the sensitive formation, destruction and excitation processes of molecules. Such technique applies from the Solar neighbourhood towards nearby stars (e.g. Savage et al. 1977;Boissé et al. 2013) to the gas in and around high redshift galaxies revealed by Damped Lymanα systems (DLAs) (e.g. Levshakov et al. 1989;Ge et al. 1997;Petitjean et al. 2000;Cui et al. 2005;Srianand et al. 2005;Noterdaeme et al. 2008;Jorgenson et al. 2010;Carswell et al. 2011;Balashev et al. 2017). In addition, the detection of molecular species at high-redshift provides original and sensitive probes of fundamental physics and cosmology. Tiny shifts in the relative wavelengths of H 2 Lyman and Werner lines can be used to constrain the pos-⋆ Based on observations and archival data from the European Southern Observatory (ESO) prog. IDs 060.A-9024, 072.A-0346, 278.A-5062, 080.A-0482, 080.A-0795, 081.A-0242, 081.A-0334, 082.A-0544, 082.A-0569, 083.A-0454, 084.A-0699, 086.A-0074 and 086.A-0643. using the Ultraviolet and Visual Echelle Spectrograph (UVES) and X-shooter at the Very Large Telescope (VLT), on Cerro Paranal, Chile. ⋆⋆ email: noterdaeme@iap.fr sible space-time variation of the proton-to-electron mass ratio down to a few parts-per-million over a timescale of Gyrs (see Ubachs et al. 2016, and references therein). The excitation of CO rotational levels provides in turn one of the best thermometers for measuring the temperature of the cosmic microwave background (CMB) radiation at high redshift Noterdaeme et al. 2011). Last but not least, the molecular phase of the ISM makes the link between the gas accreted onto galaxy and its gravitational collapse that gives birth to stars. However, the small number of known molecular absorbers contrasts with the huge number of DLAs detected so far (e.g. Prochaska et al. 2005;Noterdaeme et al. 2012): only about 25 confirmed high-redshift H 2 -bearing DLAs have been reported to date (see Balashev et al. 2017 and references therein), highlighting the small covering factor of the molecular gas and the need of efficient selection techniques.
In the local ISM, early works using Copernicus showed that H 2 and neutral carbon (C i) were frequently observed in the same absorption systems (e.g. Liszt 1981). Despite the high abundance of carbon, it is usually found in ionised forms in high-redshift DLAs and the neutral carbon is see only in a small fraction of DLAs that also show H 2 absorption (e.g. Ge et al. 2001;Srianand et al. 2005). This is likely due to the first ionisation potential of carbon (11.26 eV) being close to the energy of Lyman-Werner photons that lead to H 2 dissociation (through Solomon process, see e.g. Stecher & Williams 1967). C i also conveniently produces absorption lines out of the Lyman-α forest that can be identified even at low spectral resolution. We have therefore performed the first blind survey for neutral carbon lines in quasar spectra from the Sloan Digital Sky Survey (Ledoux et al. 2015), without any prior knowledge of the associated atomic and molecular content. The 66 C i candidates constitute our parent sample. We report here on the complete follow up of this sample with the Ultraviolet and Visual Echelle Spectrograph (UVES) at a resolving power R ∼ 50 000 and the X-Shooter spectrograph (R ∼ 5000) at the Very Large Telescope.

Observations and Results
We obtained spectra for almost all systems that are observable from Paranal Observatory, i.e. a sample of thirty nine confirmed C i absorbers. Details about the observing procedures, data reduction, and metal line measurements are presented in Ledoux et al. (in prep). A near-infrared study of the Na i and Ca ii lines as well as the dust extinction properties are presented in Zou et al. (submitted). Here, we focus on the detection of H 2 and CO. Wavelengths and oscillator strengths for H 2 and CO lines are from the compilations of Malec et al. (2010) and Daprà et al. (2016), respectively.

Molecular hydrogen
We detect H 2 absorption lines whenever covered by our spectra (twelve systems). Five of these are already reported in the literature (Noterdaeme et al. 2007;Srianand et al. 2008;Jorgenson et al. 2010;Noterdaeme et al. 2010;Klimenko et al. 2016), from which we have taken the H 2 column densities, and seven are new detections. We estimated the total H 2 column densities for the new detections through Voigt-profile fitting, focusing on the low rotational levels that contain most of the H 2 . We note that while the velocity profile of singly ionised metals is wide with a large number of components, we detect H 2 only at velocities where C i is also detected. Below we comment on each system, in order of increasing right ascension of the background quasar.
J091721+015448, z abs = 2.107 This system was observed with X-shooter at a spectral resolution of ∼60 km s −1 . We obtain an accurate measurement of the total H 2 column density thanks to the damping wings that are seen for the low rotational levels in the four bands covered by our spectrum (see Fig. 1) and obtain log N (H 2 ) = 20.11 ± 0.06.

J111756+143716, z abs = 2.001
This system is characterised by two narrow H 2 components seen in the UVES spectrum ( Fig. 2) in different rotational levels. These components also correspond to those seen in the neutral carbon lines. While our best-fit value is found to be around log N (H 2 ) ∼ 18, we note that the data quality is poor and that only one band is covered, making it impossible to assess the presence of blends. In addition, at such column density, the absorption is in the logarithmic part of the curve of growth. We are therefore unable to associate an uncertainty to this measurement that we display with a large arbitrary (albeit quite conservative) 1 dex error bar in Fig. 9.
J131129+222552, z abs = 3.092 Thanks to the high absorption redshift, no less than twenty Lyman and Werner H 2 bands are covered by our UVES spectrum, shown on Fig. 3. Four components can be distinguished in the high rotational levels but lines from the J = 0 and J = 1 rotational levels are strongly damped and therefore modelled using a single component. The damping wings together with the large number of detected transitions and the high achieved S/N allows a very accurate measurement of the total H 2 column density which we found to be log N (H 2 ) = 19.69 ± 0.01.
J164610+232922, z abs = 1.998 While the S/N of our UVES spectrum in the region of H 2 lines (see Fig. 4) is quite low 1 , two narrow H 2 components are clearly seen in rotational levels J = 0-3 and our spec-  trum covers four Lyman bands, that span more than an order of magnitude in oscillator strengths. We find a total column density N (H 2 ) ≈ 10 18 cm −2 with a ∼30% uncertainty.
J225719−100104, z abs = 1.836 This system is more complex with no less than eight H 2 components, strongly blended with each other. Unfortunately, only three Lyman H 2 bands are covered by the UVES data (Fig. 5), the bluest of which in a region with low S/N ratio. To remove strong degeneracy between parameters, we had to fix the excitation temperature T 01 to 100 K. While this is a strong assumption, we note that varying T 01 within a factor of two has little effect on the total column H 2 density (changes ∼0.1 dex). Still, we caution that this error may be underestimated and covering bluer transitions is required to confirm our column density measurement (log N (H 2 ) = 19.5 ± 0.1).  J233156-090802, z abs = 2.142 In spite of the low S/N achieved this system, shown on Fig. 6, the data is clearly consistent with strongly damped H 2 lines at the same redshift as that of CO lines (see next section). We fitted the J = 0, 1, 2 lines, from which we obtain realistic excitation temperatures, T 01 ∼ 140 K and T 02 ∼ 180 K. The total H 2 column density is found to be log N (H 2 ) = 20.57 ± 0.05.
J233633-105841, z abs = 1.829 The H 2 profile in this system is well modelled by two components, that are partially blended at the X-Shooter spectral resolution. The bluest component dominates however the total column density, and the measurement is facilitated by the presence of damping wings and the high S/N achieved. We note that the L0-0 band is partially blended with unrelated absorption lines, which we modelled when fitting H 2 (see Fig. 7). We obtain log N (H 2 ) = 19.0 ± 0.12.
Article number, page 3 of 6  Fig. 6. Same as Fig. 1 for the UVES spectrum of J2331−0908. The data has been rebinned by 7 pixels for visual purpose only. The contribution from H2 alone is shown in red and the total absorption-line profile is depicted in orange.

Carbon monoxide
CO is detected in seven systems in our sample, six of them already reported by our group and one being a new detection presented here for the first time. This brings the number of known high-z CO-bearing quasar absorbers to nine 2 . We measured upper limits on N (CO) for all systems assuming the Doppler parameter to be 1 km s −1 , similar to what has been measured in all high-z CO absorbers to date. We also assume the CMB radiation to be the main excitation source in diffuse gas at high-z (as observed by Srianand et al. 2008;Noterdaeme et al. 2011). We calculated the local (i.e. for each band individually) and global χ 2 values for a range of total column densities. CO is detected when the χ 2 curves are consistent with each other and present a clear inflexion point, defining the best-fit value. For non-detections, χ 2 (N (CO)) is generally monotonic with a minimum consistent with that 2 The detections towards J1211+0833 (Ma et al. 2015) and J0000+0015  are not formally part of the statistical sample although selected upon their C i content.  of N (CO)=0 within uncertainty. Our 3 σ upper limit corresponds to the column density where the χ 2 is 9 above this minimum. With this method, not only we recover all the known CO absorbers but we also identify the new CO system, at z = 2.143 towards SDSS J2331−0908 (Fig. 8), observed by Nestor and collaborators (Prog. ID 080.A-0795). This is only the fourth high-z system with direct and simultaneous measurements of N (CO) and N (H 2 ).
Before discussing our findings, it is worth mentioning that, in the local ISM, the excitation temperature of CO is found to be a few degrees above the CMB temperature (e.g. Burgh et al. 2007), owing to additional excitation processes such as collisions, far-infrared dust emission and possibly cosmic rays. Relaxing our assumptions we find that the derived CO column density limits (as well as the CO column density for the new detection at z abs = 2.143 towards SDSS J2331−0908) are not changed significantly as the total band equivalent width is almost conserved. For example, allowing an excitation temperature 5 K above the CMB temperature only increases the derived values by less than 0.04 dex. Table 1 summarises the H 2 and CO detections and column density measurements. Figure 9 presents the H 2 and CO column densities as well as the overall molecular fraction f H2 = 2N (H 2 )/(2N (H 2 ) + N (H i)) as a function of N (C i) for our complete sample. Known systems from the literature are also added for comparison but not considered for statistical analysis.

Discussion
We find that H 2 is detected with N (H 2 ) > ∼ 10 18 cm −2 in all systems with log N (C i) > 13.5. In this regime, H 2 is likely to be self-shielded and the molecular fraction substantial in the H 2 -bearing gas. We also observe a possible trend for increasing N (H 2 ) with increasing N (C i) (Spearman rank correlation coefficient r = 0.4, 1.2 σ significance)  in our statistical sample, albeit with a large dispersion. We note that systems that were not C i-selected (from literature) seem not to follow this trend. Four of them indeed have N (H 2 ) > 5 × 10 19 cm −2 in spite of relatively low C i column density (log N (C i) < ∼ 14). This difference is likely due to the different chemical enrichments: C i-selected systems probe mostly high-metallicity gas (Zou et al. sub., Ledoux et al. in prep.) while the four mentioned literature systems all have low metallicities. . Squares correspond to high-z H2 DLA systems from the literature (Balashev et al. 2010(Balashev et al. , 2011Carswell et al. 2011;Guimarães et al. 2012;Noterdaeme et al. 2015Noterdaeme et al. , 2017Petitjean et al. 2002).
Since the column density at which H i is converted into H 2 strongly depends on the chemical properties of the gas, in particular the abundance of dust grains (e.g. Bialy & Sternberg 2016), we can expect less H i in the molecular cloud envelope for high-metallicity systems compared to low-metallicity ones. In addition, contrary to DLAs, C isystems were selected without any a priori knowledge of the H i content (Ledoux et al. 2015) and should have less contribution from unrelated atomic gas that does not belong to the envelope of the H 2 cloud. This is seen in the bottom panel of Fig. 9, where the correlation between f and N (C i) is seen with r = 0.6 at 2.1 σ: the average overall H 2 molecular fraction is about 15% in our sample (and about 30% when CO is detected) but < 3% at log N (C i) < 14.
The correlation between N (CO) and N (C i) for CO detections is strong with r = 0.88 (2.6 σ). From the column density distributions, we can see that the probability to detect CO becomes much larger above N (C i) ∼ 5 × 10 14 cm −2 (6/12) than below this value (1/27). In addition, there is no CO detection among the 18 systems with log N (C i) < 14.4. Since the CO detection limits are signif-