The fraction of ionizing radiation from massive stars that escapes to the intergalactic medium
N. R. Tanvir, J. P. U. Fynbo, A. de Ugarte Postigo, J. Japelj, K. Wiersema, D. Malesani, D. A. Perley, A. J. Levan, J. Selsing, S. B. Cenko, D. A. Kann, B. Milvang-Jensen, E. Berger, Z. Cano, R. Chornock, S. Covino, A. Cucchiara, V. D'Elia, P. Goldoni, A. Gomboc, K. E. Heintz, J. Hjorth, L. Izzo, P. Jakobsson, L. Kaper, T. Kruehler, T. Laskar, M. Myers, S. Piranomonte, G. Pugliese, R. Sanchez-Ramirez, S. Schulze, M. Sparre, E. R. Stanway, G. Tagliaferri, C. C. Thoene, S. Vergani, P. M. Vreeswijk, R. A. M. J. Wijers, D. Watson, D. Xu
aa r X i v : . [ a s t r o - ph . GA ] M a y MNRAS , 1–31 (2018) Preprint 21 May 2018 Compiled using MNRAS L A TEX style file v3.0
The fraction of ionizing radiation from massive stars thatescapes to the intergalactic medium
N. R. Tanvir, ⋆ J. P. U. Fynbo, A. de Ugarte Postigo, J. Japelj, K. Wiersema, D. Malesani, D. A. Perley, A. J. Levan, J. Selsing, S. B. Cenko, , D. A. Kann, B. Milvang-Jensen, E. Berger, Z. Cano, R. Chornock, S. Covino, A. Cucchiara, V. D’Elia, , P. Goldoni, A. Gomboc, K. E. Heintz, , J. Hjorth, L. Izzo, P. Jakobsson, L. Kaper, T. Kr¨uhler, T. Laskar, , M. Myers, S. Piranomonte, G. Pugliese, R. S´anchez-Ram´ırez, S. Schulze, M. Sparre, , E. R. Stanway, G. Tagliaferri, C. C. Th¨one, S. Vergani, P. M. Vreeswijk, R. A. M. J. Wijers, D. Watson, and D. Xu, Affiliations are listed at the end of the paper
Accepted XXX. Received YYY; in original form ZZZ
ABSTRACT
The part played by stars in the ionization of the intergalactic medium (IGM) remainsan open question. A key issue is the proportion of the stellar ionizing radiation thatescapes the galaxies in which it is produced. Spectroscopy of gamma-ray burst (GRB)afterglows can be used to determine the neutral hydrogen column-density, N H i , in theirhost galaxies and hence the opacity to extreme ultra-violet (EUV) radiation alongthe lines-of-sight to the bursts. Thus, making the reasonable assumption that long-duration GRB locations are representative of the sites of massive stars that dominateEUV production, one can calculate an average escape fraction of ionizing radiation ina way that is independent of galaxy size, luminosity or underlying spectrum. Here wepresent a sample of N H i measures for 138 GRBs in the range . < z < . and useit to establish an average escape fraction at the Lyman limit of h f esc i ≈ . , with a98% confidence upper limit of h f esc i ≈ . . This analysis suggests that stars providea small contribution to the ionizing radiation budget of the IGM at z < , where thebulk of the bursts lie. At higher redshifts, z > , firm conclusions are limited by thesmall size of the GRB sample (7/138), but any decline in average H i column-densityseems to be modest. We also find no indication of a significant correlation of N H i withgalaxy UV luminosity or host stellar mass, for the subset of events for which these areavailable. We discuss in some detail a number of selection effects and potential biases.Drawing on a range of evidence we argue that such effects, while not negligible, areunlikely to produce systematic errors (in either direction) of more than a factor ∼ in f esc , and so would not affect the primary conclusions. Given that many GRB hosts arelow metallicity, high specific star-formation rate, dwarf galaxies, these results presenta particular problem for the hypothesis that such galaxies dominated the reionizationof the universe. Key words: dark ages, reionization, first stars – gamma-ray burst: general – galaxies:ISM – intergalactic medium ⋆ E-mail: [email protected] (NRT)
A key question for our understanding of the reionizationof hydrogen in the intergalactic medium (IGM) is the ex- © N. R. Tanvir et al. tent to which ionizing extreme ultraviolet (EUV) radia-tion from massive stars escapes from the galaxies in whichit is produced. This can be parameterised by the escapefraction, f esc , the proportion of photons produced by starsat the Lyman limit wavelength ( λ = ˚A) that leavethe virial radius of their host galaxy. Only if the aver-age escape fraction, h f esc i , is sufficiently high in the era ofreionization (EoR; . z . ; Planck Collaboration et al.2016), i.e. h f esc i at least 0.1–0.2, is it likely that this phasechange was predominantly driven by EUV star-light (e.g.Ouchi et al. 2009; Bouwens et al. 2012; Finkelstein et al.2012; Robertson et al. 2015; Faisst 2016). Otherwise someother significant source of ionizing radiation is required, suchas a large population of faint quasars (Madau & Haardt2015; Khaire et al. 2016, but see Hassan et al. (2018) forcounter arguments), X-ray binaries (Mirabel et al. 2011;Fragos et al. 2013; Knevitt et al. 2014; Madau & Fragos2017) or decaying/annihilating particles (Sciama 1982;Hansen & Haiman 2004).Direct searches for Lyman continuum emission below912 ˚A in the rest frame are compromised by absorption dueto neutral gas in the intergalactic medium (the Ly α for-est), and essentially impossible above z ∼ as the IGM ab-sorption becomes near total – the so-called Gunn-Petersontrough (Gunn & Peterson 1965). Observations at lower red-shifts are still difficult, and there have been extensive effortssearching for such continuum emission from star-forminggalaxies at z = –4 in recent years (e.g Steidel et al. 2001;Shapley et al. 2006; Vanzella et al. 2010, 2012; Nestor et al.2013; Mostardi et al. 2013; Vanzella et al. 2015; Japelj et al.2017; Marchi et al. 2017). Results have been conflicting, par-ticularly due to rare cases of low redshift galaxies aligningby chance with higher redshift targets (e.g. Vanzella et al.2012; Siana et al. 2015), but given the scarcity of high es-cape fraction systems it appears that f esc is not high onaverage (Grazian et al. 2017; Rutkowski et al. 2017). Thisis consistent with quasars being the primary source of EUVradiation maintaining a reionized IGM at z < . However,at least some individual cases at z & appear to havevery high escape fractions f esc & . (de Barros et al. 2016;Vanzella et al. 2016; Shapley et al. 2016), and so might beanalogues of galaxies in the EoR.To account for the discrepancy between the expectedand observed level of the escape fraction, it has been of-ten suggested that h f esc i may actually increase with de-creasing galaxy luminosity and/or with increasing redshift(e.g. Razoumov & Sommer-Larsen 2010; Ciardi et al. 2012;Kuhlen & Faucher-Gigu`ere 2012; Fontanot et al. 2014;Xu et al. 2016a; Anderson et al. 2017). Observationally, itis hard to reach sufficiently stringent constraints on the es-cape fraction for faint galaxy populations to investigate itsdependence on luminosity (e.g. Japelj et al. 2017), while dueto IGM absorption the claim of a changing escape frac-tion with redshift can only be investigated through sec-ondary means and simulations (e.g. Zackrisson et al. 2013;Sharma et al. 2016). The challenge facing simulators is tomodel in sufficient detail the complex baryonic physics andradiative transfer given limited resolution, while also sam-pling a range of galaxies and environments. Typically, mod-els of an instantaneous burst of star formation incorporatingonly single-star stellar evolution produce the large bulk oftheir EUV within a few Myr, limiting the time available for feedback from winds, radiation and supernovae to open win-dows in the surrounding high density gas. Recently, modelswhich include binary stellar evolution have been shown toprolong the period of high EUV production to ∼ Myr, andhence hold more promise for clearing of local gas, and for atleast a relatively high fraction of stars leaving the immedi-ate environment in which they formed (Stanway et al. 2016;Ma et al. 2016).An alternative route, first proposed by Chen et al.(2007), to constraining h f esc i empirically over a broad rangein redshift is via spectroscopy of long-duration gamma-rayburst (GRB) afterglows. These very bright, but short-lived,continuum sources allow detailed abundance studies of hostgalaxy gas along the line of sight. Crucially, this includescalculation of the neutral hydrogen column-density, N H i , inthe host from fitting the Ly α absorption feature when itis seen (e.g. Prochaska et al. 2007; Fynbo et al. 2009). Thiscolumn-density can be directly converted into an opacitymeasure for EUV ionizing radiation and hence f esc . For anyindividual host galaxy, a single sight-line does not providea robust measure of its average escape fraction, but sincelong-duration GRBs are associated with the core-collapseof massive stars (e.g. Hjorth et al. 2003a; Xu et al. 2013), asample of GRB afterglows should be representative of thedistribution of all sight-lines specifically to the locations ofyoung stars largely responsible for EUV production.To date, neutral hydrogen columns reported for GRBhosts have generally been high, mostly classified as dampedLy α absorbers (DLAs; log ( N H i / cm − ) > . ), which is usu-ally taken as being consistent with their massive-star pro-genitors remaining in or close to the dense molecular cloudsin which they formed, and/or more generally residing at thehearts of gas-rich star-forming galaxies (e.g. Jakobsson et al.2006). In fact, observations of the time-variability of fine-structure transitions in GRB afterglow spectra, in the (dozenor so) systems where it has been measured, has allowedthe distance of the dominant absorbing clouds to be estab-lished, ranging from ∼ pc to & kpc (e.g. Vreeswijk et al.2013). These scales are comparable to the sizes of large ion-ized superbubbles around star-forming regions in the lowredshift universe (Oey & Clarke 1997; Camps-Fari˜na et al.2017), and so might indicate that absorption takes placedue to neutral gas piled up at the boundaries of such bub-bles for some GRBs. In any case, this tendency towards highcolumn-densities is potentially a problem since if the EUVescape fraction is to fulfil the requirements for reionization,a significant proportion of GRBs should have very low H i columns ( log ( N H i / cm − ) ≪ ), particularly at z > .This method has the considerable advantage that GRBafterglows readily probe gas in even very faint galaxies,for which either Lyman-continuum observations would beweakly constraining, or which may be missed completelyin traditional galaxy surveys. In principle, then, with asufficiently large sample one could trace the escape frac-tion both as a function of galaxy luminosity and of red-shift. The tendency of GRBs to occur preferentially in lowermetallicity galaxies, Z / Z ⊙ . . –1 (e.g. Perley et al. 2016b;Japelj et al. 2016; Graham & Fruchter 2017; Vergani et al.2017), often dwarfs with high specific star formation rates(e.g. Svensson et al. 2010; Hunt et al. 2014), also suggeststhey should be more representative of the populations dom-inant during the EoR. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Chen et al. (2007) and Fynbo et al. (2009) have previ-ously performed such analyses, each obtaining 95% confi-dence upper limits on h f esc i of only ∼ . based on sam-ples of ∼ GRBs (with some overlap of their samples) withredshifts . z . . Here we reinvestigate this issue, usinga considerably larger sample of 138 GRBs with N H i deter-minations, spanning a redshift range . . z . . . Thestructure of the paper is as follows: in Section 2 we presentthe sample of GRBs and describe its basic properties; inSection 3 we outline the implications for the average escapefraction of ionizing EUV radiation and consider evidence forevolution over cosmic history; in Section 4 we consider arange of potential systematic uncertainties which could biasour conclusions in either direction, and address more fullythe question of how representative GRB sight-lines are likelyto be for the stellar populations of interest for reionization;finally in Section 5 we draw our conclusions. Further de-tails of some individual GRBs, including results not alreadyreported elsewhere, are given in an Appendix. I COLUMN-DENSITY SAMPLE
Spectra of GRB afterglows frequently exhibit strong Ly α absorption lines which can be modelled to constrain thecolumn-density of neutral hydrogen responsible. Particularlyat higher redshifts, this relies largely on fitting the red wingof the line, since absorption by the IGM Ly α forest sig-nificantly affects the blue wing. In the large majority ofcases, the systemic redshifts are known quite precisely frommetal-line detections, which improves the precision of theLy α fits. We have gathered together H i column-densitiestowards GRBs from the literature, and combined them witha large number of new measurements we have made us-ing afterglow spectra from various sources. Many of thesecome from the long-running Very Large Telescope (VLT) X-shooter legacy programme (Selsing et al. 2018), but we alsoinclude data from the Nordic Optical Telescope (NOT), theWilliam Herschel Telescope (WHT), the Gran TelescopioCanarias (GTC), the Telescopio Nazionale Galileo (TNG),the Gemini Telescopes (both North and South), the AsiagoCopernico Telescope (CT), and other VLT spectrographs.The bulk of the GRBs were originally discoveries of the NeilGehrels Swift
Observatory, but there was little consistencyin terms of which bursts were followed-up or the kinds ofobservations obtained (e.g. in terms of spectral resolution,wavelength coverage, sensitivity etc.). The net result is aninhomogeneous sample, and potential effects of selection bi-ases are discussed in Section 4.The HI column densities measured from the afterglowspectra in our sample are plotted in Figure 1 and sum-marised in Table 1; corresponding primary sources for theadopted values of log ( N H i ) are given in the 4th column ofthe table, and readers are refered also to Appendix A forfurther details regarding previously unreported fits and ad-ditional comments on some particular cases. The total num-ber of sight-lines is 138, which represents a more than four-fold increase over similar previous studies (Chen et al. 2007;Fynbo et al. 2009).The median redshift of our sample is ˜ z = . . The lowerredshift cut-off, at z ∼ . , occurs because the observedwavelength of Ly α falls in the near-UV, and begins to be strongly affected by declining atmospheric transmission. Atthe high redshift end, the sample is curtailed due to declin-ing spectral quality; although bursts at z & . have beenfound, either the signal-to-noise has been too poor to reliablymeasure the red damping wing of Ly α (Tanvir et al. 2009;Salvaterra et al. 2009; Tanvir et al. 2017) or the redshift hasbeen inferred photometrically (Cucchiara et al. 2011a). Following Chen et al. (2007), we note that the optical depthfor radiation at the Lyman limit (912 ˚A) along a given sightline due to absorption by neutral hydrogen is given by τ = σ LL N H i , where σ LL = . × − cm is the photoionizationcross-section of hydrogen. Hence the average escape fractionfor n sight lines is given by h f esc i = n n Õ i = exp (− τ i ) . (1)Considering our whole sample, we find a mean value of only h f esc i = . , well below that thought to be required in theEoR.In Figure 2 we plot the cumulative distribution of H i column-density measures for the whole sample. The medianvalue of column-density is log ( N H i / cm − ) = . , consistentwith previous studies (e.g. the equivalent figure is 21.5 forthe sample of Fynbo et al. 2009) and also similar to themedian values of log ( N H i ) towards H ii regions in the Mag-ellanic Clouds (Pellegrini et al. 2012, see Figure 1). We findthat up to the median point the sample is well described bya simple power law distribution P ( < x ) ∝ x . , where x is thevalue of N H i , shown as an orange dashed line in the figure.While this model is not motivated by any particular physi-cal considerations, it does provide a smooth representationof the data, and using it we obtain an average escape frac-tion h f esc i = . , in good agreement with the value foundabove.Most of the rest of this paper is concerned with therobustness of this result, and the statistical and potentialsystematic uncertainties that may affect it. Only two sight-lines have log ( N H i / cm − ) < (correspond-ing to f esc > . ), and as it happens both of theselow column-density systems were already included in theChen et al. (2007) and Fynbo et al. (2009) analyses. Sincethe numerical result for h f esc i depends entirely on these twosight-lines, we review here what is known of their propertiesand in particular consider whether there could be attenu-ation of EUV radiation by dust as well as H i absorption.We also address the effect of direct recombinations to theground-state producing ionizing photons. GRB 050908, at z = . , had a moderately bright opti-cal afterglow, being R ≈ at 15 min post-burst (Torii MNRAS000
Observatory, but there was little consistencyin terms of which bursts were followed-up or the kinds ofobservations obtained (e.g. in terms of spectral resolution,wavelength coverage, sensitivity etc.). The net result is aninhomogeneous sample, and potential effects of selection bi-ases are discussed in Section 4.The HI column densities measured from the afterglowspectra in our sample are plotted in Figure 1 and sum-marised in Table 1; corresponding primary sources for theadopted values of log ( N H i ) are given in the 4th column ofthe table, and readers are refered also to Appendix A forfurther details regarding previously unreported fits and ad-ditional comments on some particular cases. The total num-ber of sight-lines is 138, which represents a more than four-fold increase over similar previous studies (Chen et al. 2007;Fynbo et al. 2009).The median redshift of our sample is ˜ z = . . The lowerredshift cut-off, at z ∼ . , occurs because the observedwavelength of Ly α falls in the near-UV, and begins to be strongly affected by declining atmospheric transmission. Atthe high redshift end, the sample is curtailed due to declin-ing spectral quality; although bursts at z & . have beenfound, either the signal-to-noise has been too poor to reliablymeasure the red damping wing of Ly α (Tanvir et al. 2009;Salvaterra et al. 2009; Tanvir et al. 2017) or the redshift hasbeen inferred photometrically (Cucchiara et al. 2011a). Following Chen et al. (2007), we note that the optical depthfor radiation at the Lyman limit (912 ˚A) along a given sightline due to absorption by neutral hydrogen is given by τ = σ LL N H i , where σ LL = . × − cm is the photoionizationcross-section of hydrogen. Hence the average escape fractionfor n sight lines is given by h f esc i = n n Õ i = exp (− τ i ) . (1)Considering our whole sample, we find a mean value of only h f esc i = . , well below that thought to be required in theEoR.In Figure 2 we plot the cumulative distribution of H i column-density measures for the whole sample. The medianvalue of column-density is log ( N H i / cm − ) = . , consistentwith previous studies (e.g. the equivalent figure is 21.5 forthe sample of Fynbo et al. 2009) and also similar to themedian values of log ( N H i ) towards H ii regions in the Mag-ellanic Clouds (Pellegrini et al. 2012, see Figure 1). We findthat up to the median point the sample is well described bya simple power law distribution P ( < x ) ∝ x . , where x is thevalue of N H i , shown as an orange dashed line in the figure.While this model is not motivated by any particular physi-cal considerations, it does provide a smooth representationof the data, and using it we obtain an average escape frac-tion h f esc i = . , in good agreement with the value foundabove.Most of the rest of this paper is concerned with therobustness of this result, and the statistical and potentialsystematic uncertainties that may affect it. Only two sight-lines have log ( N H i / cm − ) < (correspond-ing to f esc > . ), and as it happens both of theselow column-density systems were already included in theChen et al. (2007) and Fynbo et al. (2009) analyses. Sincethe numerical result for h f esc i depends entirely on these twosight-lines, we review here what is known of their propertiesand in particular consider whether there could be attenu-ation of EUV radiation by dust as well as H i absorption.We also address the effect of direct recombinations to theground-state producing ionizing photons. GRB 050908, at z = . , had a moderately bright opti-cal afterglow, being R ≈ at 15 min post-burst (Torii MNRAS000 , 1–31 (2018)
N. R. Tanvir et al.
Table 1.
The sample of GRBs. References:- (1) Jensen et al. (2001), (2) Fynbo et al. (2002), (3) Vreeswijk et al. (2006), (4) Hjorth et al.(2003b), (5) Fynbo et al. (2005), (6) Møller et al. (2002), (7) Shin et al. (2006), (8) Vreeswijk et al. (2004), (9) Jakobsson et al. (2004),(10) Fynbo et al. (2009), (11) Berger et al. (2006), (12) Totani et al. (2006), (13) Chen et al. (2007), (14) Chary et al. (2007), (15)Ferrero et al. (2009), (16) This work, (17) Wiseman et al. (2017), (18) Patel et al. (2010), (19) de Ugarte Postigo et al. (2012), (20)Kuin et al. (2009), (21) D’Avanzo et al. (2010), (22) Savaglio et al. (2012), (23) Levesque et al. (2010a), (24) Selsing et al. (2018), (25)Cucchiara et al. (2011b), (26) Zafar et al. in prep., (27) Cucchiara et al. (2015), (28) Jeong et al. (2014), (29) Chornock et al. (2014),(30) Melandri et al. (2015), (31) Pugliese et al. in prep., (32) Chen et al. (2009), (33) Perley (2011), (34) Schulze et al. (2015), (35)Greiner et al. (2015b), (36) McGuire et al. (2016), (37) Tanvir et al. (2012a), (38) McGuire et al. in prep., (39) Th¨one et al. (2011), (40)Friis et al. (2015), (41) Perley et al. (2013), (42) Laskar et al. (2011), (43) Perley et al. (2016b), (44) Myers et al. in prep. (in this case,the derived
Spitzer photometry was transformed to stellar mass estimates following Perley et al. 2016b).GRB z log (cid:16) N H i cm − (cid:17) Refs. M UV , AB Refs. log(M ∗ /M ⊙ ) Refs.000301C 2.03 . ± . (1) − . ± . (32)000926 2.04 . ± . (2) − . ± . (32) . (41)011211 2.14 . ± . (3) − . ± . (32) . (41)020124 3.20 . ± . (4) > − . (35) < . (42)021004 2.33 . ± . (5),(6),A1 − . ± . (5) . (41)030226 1.99 . ± . (7)030323 3.37 . ± . (8) − . ± . (32) < . (42)030429 2.65 . ± . (9)050319 3.24 . ± . (10) > − . (33) . (43)050401 2.90 . ± . (10) − . ± . (34) . (43)050505 4.27 . ± . (11) < . (44)050730 3.97 . ± . (10) > − . (35) < . (43)050820A 2.61 . ± . (10) − . ± . (34) . (43)050904 6.29 . ± . (12) − . ± . (36) < . (43)050908 3.34 . ± . (10) − . ± . (34) < . (42)050922C 2.20 . ± . (10) > − . (33) < . (43)060115 3.53 . ± . (10) − . ± . (34) . (43)060124 2.30 . ± . (10)060206 4.05 . ± . (10) − . ± . (35) < . . ± . (10) − . ± . (33) . (43)060223A 4.41 . ± . (13) − . ± . (35) < . (42)060510B 4.94 . ± . (13) − . ± . (35) . (44)060522 5.11 . ± . (14),(16) > − . (37) < . (43)060526 3.21 . ± . (10) > − . (35) . (43)060605 3.77 . ± . (15) − . ± . (35) < . (42)060607A 3.08 . ± . (10) > − . (35) < . (43)060707 3.43 . ± . (10) − . ± . (34) . (43)060714 2.71 . ± . (10) − . ± . (34) . (43)060906 3.69 . ± . (10) > − . (35) < . (42)060926 3.21 . ± . (10) − . ± . (35) . (42)060927 5.47 . ± . (10) > − . (37) < . (43)061110B 3.44 . ± . (10) − . ± . (35) < . (43)070110 2.35 . ± . (10) − . ± . (34) < . (43)070411 2.95 . ± . (10)070506 2.31 . ± . (10) − . ± . (34)070611 2.04 . ± . (10) > − . (34)070721B 3.63 . ± . (10) − . ± . (35) < . (43)070802 2.45 . ± . (10) − . ± . (34) . (41)070810A 2.17 . ± . (16)071031 2.69 . ± . (10)080129 4.35 . ± . (16) < . (44)080210 2.64 . ± . (10) > − . (33) < . (43)080310 2.43 . ± . (10) > − . (33) . (43)080413A 2.43 . ± . (10) < . (43)080603B 2.69 . ± . (10) . (43)080607 3.04 . ± . (10) − . ± . (35) . (43)080721 2.59 . ± . (10) < . (43)080804 2.20 . ± . (10) . (43)080810 3.36 . ± . (17),A6 − . ± . (35) . (44)080905B 2.37 . ± . (16)080913 6.73 . ± . (18),A8 > − . (38)081008 1.97 . ± . (19) . (43)081029 3.85 . ± . (16) > − . (35) < . (43)081118 2.58 . ± . (16) . (43)081203A 2.05 . ± . (20) MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Table 1 – continued GRB z log (cid:16) N H i cm − (cid:17) Refs. M UV , AB Refs. log(M ∗ /M ⊙ ) Refs.081222 2.77 . ± . (16) . (43)090205 4.65 . ± . (21) − . ± . (35) < . (21)090313 3.38 . ± . (16) > − . (35) < . (44)090323 3.58 . ± . (22),A13 − . ± . (35) . (44)090426 2.61 . ± . (23),A14 − . ± . (39)090516A 4.11 . ± . (19) − . ± . (35) . (44)090519 3.85 . ± . (16) > . (35) < . (43)090529 2.62 . ± . (16)090715B 3.01 . ± . (16) . (44)090726 2.71 . ± . (16)090809 2.74 . ± . (24)090812 2.45 . ± . (19) < . (43)090926A 2.11 . ± . (24)091029 2.75 . ± . (16) < . (43)100219A 4.67 . ± . (24) − . ± . (35) < . (44)100302A 4.81 . ± . (16)100316A 3.16 . ± . (16)100425A 1.76 . ± . (24)100513A 4.77 . ± . (16) − . ± . (35) < . (44)100728B 2.11 . ± . (24) < . (43)110128A 2.34 . ± . (24)110205A 2.21 . ± . (25) . (43)110731A 2.83 . ± . (16)110818A 3.36 . ± . (24) − . ± . (35)111008A 4.99 . ± . (24) > − . (35)111107A 2.89 . ± . (24)120119A 1.73 . ± . (24) . (43)120327A 2.81 . ± . (24)120404A 2.88 . ± . (24)120712A 4.17 . ± . (24) < . (44)120716A 2.49 . ± . (24)120811C 2.67 . ± . (16)120815A 2.36 . ± . (24)120909A 3.93 . ± . (24) − . ± . (35)121024A 2.30 . ± . (24) − . ± . (40) . (40)121027A 1.77 . ± . (24),A25121128A 2.20 . ± . (16)121201A 3.39 . ± . (24) − . ± . (35)121229A 2.71 . ± . (24)130408A 3.76 . ± . (24) > − . (35)130427B 2.78 . ± . (24)130505A 2.27 . ± . (27)130518A 2.49 . ± . (16)130606A 5.91 . ± . (24) − . ± . (36)130610A 2.09 . ± . (16)130612A 2.01 . ± . (24)131011A 1.87 . ± . (24)131108A 2.40 . ± . (16)131117A 4.04 . ± . (24)140206A 2.73 . ± . (16)140226A 1.97 . ± . (27)140304A 5.28 . (28)140311A 4.95 . ± . (24) < . (44)140419A 3.96 . ± . (27)140423A 3.26 . ± . (27)140430A 1.60 . ± . (24)140515A 6.32 . ± . (29),(30),A31 − . ± . (36)140518A 4.71 . ± . (27)140614A 4.23 . ± . (24)140629A 2.28 . ± . (16)140703A 3.14 . ± . (16)140808A 3.29 . ± . (16)141028A 2.33 . ± . (24)141109A 2.99 . ± . (24)MNRAS000
Spitzer photometry was transformed to stellar mass estimates following Perley et al. 2016b).GRB z log (cid:16) N H i cm − (cid:17) Refs. M UV , AB Refs. log(M ∗ /M ⊙ ) Refs.000301C 2.03 . ± . (1) − . ± . (32)000926 2.04 . ± . (2) − . ± . (32) . (41)011211 2.14 . ± . (3) − . ± . (32) . (41)020124 3.20 . ± . (4) > − . (35) < . (42)021004 2.33 . ± . (5),(6),A1 − . ± . (5) . (41)030226 1.99 . ± . (7)030323 3.37 . ± . (8) − . ± . (32) < . (42)030429 2.65 . ± . (9)050319 3.24 . ± . (10) > − . (33) . (43)050401 2.90 . ± . (10) − . ± . (34) . (43)050505 4.27 . ± . (11) < . (44)050730 3.97 . ± . (10) > − . (35) < . (43)050820A 2.61 . ± . (10) − . ± . (34) . (43)050904 6.29 . ± . (12) − . ± . (36) < . (43)050908 3.34 . ± . (10) − . ± . (34) < . (42)050922C 2.20 . ± . (10) > − . (33) < . (43)060115 3.53 . ± . (10) − . ± . (34) . (43)060124 2.30 . ± . (10)060206 4.05 . ± . (10) − . ± . (35) < . . ± . (10) − . ± . (33) . (43)060223A 4.41 . ± . (13) − . ± . (35) < . (42)060510B 4.94 . ± . (13) − . ± . (35) . (44)060522 5.11 . ± . (14),(16) > − . (37) < . (43)060526 3.21 . ± . (10) > − . (35) . (43)060605 3.77 . ± . (15) − . ± . (35) < . (42)060607A 3.08 . ± . (10) > − . (35) < . (43)060707 3.43 . ± . (10) − . ± . (34) . (43)060714 2.71 . ± . (10) − . ± . (34) . (43)060906 3.69 . ± . (10) > − . (35) < . (42)060926 3.21 . ± . (10) − . ± . (35) . (42)060927 5.47 . ± . (10) > − . (37) < . (43)061110B 3.44 . ± . (10) − . ± . (35) < . (43)070110 2.35 . ± . (10) − . ± . (34) < . (43)070411 2.95 . ± . (10)070506 2.31 . ± . (10) − . ± . (34)070611 2.04 . ± . (10) > − . (34)070721B 3.63 . ± . (10) − . ± . (35) < . (43)070802 2.45 . ± . (10) − . ± . (34) . (41)070810A 2.17 . ± . (16)071031 2.69 . ± . (10)080129 4.35 . ± . (16) < . (44)080210 2.64 . ± . (10) > − . (33) < . (43)080310 2.43 . ± . (10) > − . (33) . (43)080413A 2.43 . ± . (10) < . (43)080603B 2.69 . ± . (10) . (43)080607 3.04 . ± . (10) − . ± . (35) . (43)080721 2.59 . ± . (10) < . (43)080804 2.20 . ± . (10) . (43)080810 3.36 . ± . (17),A6 − . ± . (35) . (44)080905B 2.37 . ± . (16)080913 6.73 . ± . (18),A8 > − . (38)081008 1.97 . ± . (19) . (43)081029 3.85 . ± . (16) > − . (35) < . (43)081118 2.58 . ± . (16) . (43)081203A 2.05 . ± . (20) MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Table 1 – continued GRB z log (cid:16) N H i cm − (cid:17) Refs. M UV , AB Refs. log(M ∗ /M ⊙ ) Refs.081222 2.77 . ± . (16) . (43)090205 4.65 . ± . (21) − . ± . (35) < . (21)090313 3.38 . ± . (16) > − . (35) < . (44)090323 3.58 . ± . (22),A13 − . ± . (35) . (44)090426 2.61 . ± . (23),A14 − . ± . (39)090516A 4.11 . ± . (19) − . ± . (35) . (44)090519 3.85 . ± . (16) > . (35) < . (43)090529 2.62 . ± . (16)090715B 3.01 . ± . (16) . (44)090726 2.71 . ± . (16)090809 2.74 . ± . (24)090812 2.45 . ± . (19) < . (43)090926A 2.11 . ± . (24)091029 2.75 . ± . (16) < . (43)100219A 4.67 . ± . (24) − . ± . (35) < . (44)100302A 4.81 . ± . (16)100316A 3.16 . ± . (16)100425A 1.76 . ± . (24)100513A 4.77 . ± . (16) − . ± . (35) < . (44)100728B 2.11 . ± . (24) < . (43)110128A 2.34 . ± . (24)110205A 2.21 . ± . (25) . (43)110731A 2.83 . ± . (16)110818A 3.36 . ± . (24) − . ± . (35)111008A 4.99 . ± . (24) > − . (35)111107A 2.89 . ± . (24)120119A 1.73 . ± . (24) . (43)120327A 2.81 . ± . (24)120404A 2.88 . ± . (24)120712A 4.17 . ± . (24) < . (44)120716A 2.49 . ± . (24)120811C 2.67 . ± . (16)120815A 2.36 . ± . (24)120909A 3.93 . ± . (24) − . ± . (35)121024A 2.30 . ± . (24) − . ± . (40) . (40)121027A 1.77 . ± . (24),A25121128A 2.20 . ± . (16)121201A 3.39 . ± . (24) − . ± . (35)121229A 2.71 . ± . (24)130408A 3.76 . ± . (24) > − . (35)130427B 2.78 . ± . (24)130505A 2.27 . ± . (27)130518A 2.49 . ± . (16)130606A 5.91 . ± . (24) − . ± . (36)130610A 2.09 . ± . (16)130612A 2.01 . ± . (24)131011A 1.87 . ± . (24)131108A 2.40 . ± . (16)131117A 4.04 . ± . (24)140206A 2.73 . ± . (16)140226A 1.97 . ± . (27)140304A 5.28 . (28)140311A 4.95 . ± . (24) < . (44)140419A 3.96 . ± . (27)140423A 3.26 . ± . (27)140430A 1.60 . ± . (24)140515A 6.32 . ± . (29),(30),A31 − . ± . (36)140518A 4.71 . ± . (27)140614A 4.23 . ± . (24)140629A 2.28 . ± . (16)140703A 3.14 . ± . (16)140808A 3.29 . ± . (16)141028A 2.33 . ± . (24)141109A 2.99 . ± . (24)MNRAS000 , 1–31 (2018) N. R. Tanvir et al.
Figure 1.
The values of neutral hydrogen column-density in the host plotted against redshift for the sample of GRBs. The correspondingoptical depth at the Lyman limit is shown on the right-hand axis. The large majority are DLAs, with a smaller proportion (17/138)being classified as sub-DLAs, Lyman Limit Systems (LLS) and below. All sight-lines apart from the two with the lowest column-densitiesare essentially opaque ( τ & ) to EUV ionizing radiation. The running median (red line) and interquartile range (pink shading) of20 points shows no evidence for significant variation with redshift. For comparison, on the left axis we also mark the locations of themedian log( N H i /2) values in the directions of LMC and SMC H ii regions from Pellegrini et al. (2012) (halving the measured columnsis appropriate since the they include contributions from both the foreground and background of the H ii region), and the lowest columndensity out of the Milky Way from the position of the Sun, the “Lockman hole” (Lockman et al. 1986). Table 1 – continued GRB z log (cid:16) N H i cm − (cid:17) Refs. M UV , AB Refs. log(M ∗ /M ⊙ ) Refs.150206A 2.09 . ± . (24)150403A 2.06 . ± . (24)150413A 3.14 . ± . (16)150915A 1.97 . ± . (24)151021A 2.33 . ± . (24)151027B 4.06 . ± . (24)151215A 2.59 . ± . (16)160203A 3.52 . ± . (24),(31)160227A 2.38 . ± . (16)160629A 3.33 . ± . (16)161014A 2.82 . ± . (24)161017A 2.01 . ± . (16) > − . (16)161023A 2.71 . ± . (24)170202A 3.65 . ± . (24)170405A 3.51 . ± . (16)170531B 2.37 . ± . (16)180115A 2.49 . ± . (16)180325A 2.04 . ± . (26)180329B 2.00 . ± . (16) log ( N H i / cm − ) = . . The lower value of log ( N H i / cm − ) = . , corresponding to f esc = . , used here, was that de-rived from a VLT/FORS1 spectrum (Fynbo et al. 2009), and it is preferred since it is based on direct evidence of non-zero afterglow continuum emission below the Lyman limit.There is no indication of excess absorption in the Swift
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation X-ray observations (Evans et al. 2009), which is also consis-tent with a low column density and low extinction sight-line. GRB 060607A, at z = . , had a very bright andwell-studied early optical afterglow, which reached r = . at 3 min post-burst (Nysewander et al. 2009). Thissight-line has the lowest column-density of our sampleat log ( N H i / cm − ) = . , from a VLT/UVES spectrum(Fynbo et al. 2009), corresponding f esc = . . The spec-trum also showed evidence for emission below the Lymanlimit, although only for a small stretch of wavelength be-fore it was cut-off by an intervening absorber, but this isconsistent with a very low opacity.The light curve and spectral energy distribution werestudied in detail by Nysewander et al. (2009), who mod-elled their Bgri optical data together with H -band photome-try from Molinari et al. (2007). They concluded that a rest-frame dust extinction of zero was ruled out at the . σ level.The shape of the extinction law is only weakly constrainedby these data, so extrapolating to the Lyman limit intro-duces a large systematic uncertainty, but with reasonabledust laws their favoured extinction would correspond to avalue of A & . If this inference is correct, then it wouldsuggest the actual escape fraction at the Lyman limit forGRB 060607A could be significantly diminished by dust ex-tinction, by a factor ∼ . or more.We note that there is also marginal evidence of X-rayabsorption in the source-frame at a level of N H , X ≈ ( . ± . )× cm − (Evans et al. 2009) over the Milky Way fore-ground (Willingale et al. 2013). This would be broadly con-sistent with a SMC dust-to-gas ratio (Bouchet et al. 1985)providing the hydrogen associated with this gas had largelybeen ionized (so it was not seen in the optical spectrum)but the dust had mostly not been destroyed. On the otherhand, Prochaska et al. (2008) argue that the absence of N v absorption argues for both low density and low metallicitysurrounding the burst location. Gas that has been ionized by massive star radiation withina host galaxy will generally recombine quickly. A fractionof recombining H ions will go directly to the ground-state,and so emit a photon just above the Lyman limit energy(some higher energy photons will also be emitted by recom-bining He ions). In low column-density systems a fraction ofthese will escape the host without further absorption, andtherefore re-boost the escaping ionizing flux, albeit with ra-diation that will soon be redshifted to energies below 1 Ryd.In other words, simply translating H i column-density intoline-of-sight opacity is likely to lead to a small underestimateof the escape fraction in low column-density systems. Thenet effect of this re-boost depends on various factors, butcould provide an increase of up to 10–20% in the effectiveescape fraction (Faucher-Gigu`ere et al. 2009), thus at leastpartially offsetting any dust extinction. For the remainder of Section 3 we will continue to con-sider only the opacity due to H i absorption, but will returnto the potential systematic effects of dust in Section 4.1.1. Even with our considerably larger sample of sight-lines, thefact that only two have any appreciable escape fractionmeans that to some extent we are still dealing with rathersmall number statistics. We also lack a robust theoreticalmodel which could be fit to the data, and so must explorethe statistical uncertainties non-parametrically.Again we first follow Chen et al. (2007) by performinga bootstrap exercise, employing random resamples ofthe data with replacement. From this we estimate a 98%confidence upper limit of h f esc i < . ; the result is thesame whether or not we allow the resampled N H i valuesto have additional scatter based on the error bars for eachpoint.In an alternative approach, we simulated several largepopulations of sight-lines with higher values of average es-cape fraction than found in our data (by replicating theGRB 050908/060607A values), and drew random 138-member samples from each of these. For the case of thepopulation with h f esc i pop = . we found 98% of randomsamples produced h f esc i samp > . . Thus, these two meth-ods agree on an upper limit for h f esc i of . – . . A similaranalysis gives a 98% lower limit of h f esc i > − .These are significantly tighter constraints than found bythe previous studies of Chen et al. (2007) and Fynbo et al.(2009) of h f esc i < . at 95%, due to our larger sample sizeand the fact that no further very low column-density sight-lines have been identified in any of the additional GRBs.We note that our result is also consistent with the h f esc i = . ± . obtained by constraining the flux below theLyman-limit in a stacked spectrum of eleven GRB afterglowswith ¯ z = . by Christensen et al. (2011). It is worth noting that our N H i distribution is inconsis-tent with the predictions of Cen & Kimm (2014) who usedhigh-resolution cosmological radiation-hydro simulations ofgalaxies within the EoR ( z ∼ ) to explore column-densitiesalong GRB sight-lines. They found a bimodal distributionwith a peak at high column-density ( log ( N H i / cm − ) ∼ –22), similar to the observed distribution, but then anothersubstantial peak with column-densities log ( N H i / cm − ) < which is not seen in practice. Part of the explanation couldbe that Cen & Kimm (2014) assumed that the GRB ratetraces the SNII rate and found a large fraction of their lowcolumn-density GRBs occurred in super-solar metallicity en-vironments, whereas in reality GRB progenitors seem to beyounger at explosion (e.g. Larsson et al. 2007) and crucially,unlike SNII, are rarely found in high metallicity galaxies (e.g.Perley et al. 2015). On the other hand, their simulations donot account for the effect of GRBs in ionizing gas local to theburst (see Section 4.1.2), and it seems when a high column-density is found in their models it is often due to such localgas, whereas the low column-density cases occur for pro-genitors that have escaped their birth clouds. This suggests MNRAS000
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation X-ray observations (Evans et al. 2009), which is also consis-tent with a low column density and low extinction sight-line. GRB 060607A, at z = . , had a very bright andwell-studied early optical afterglow, which reached r = . at 3 min post-burst (Nysewander et al. 2009). Thissight-line has the lowest column-density of our sampleat log ( N H i / cm − ) = . , from a VLT/UVES spectrum(Fynbo et al. 2009), corresponding f esc = . . The spec-trum also showed evidence for emission below the Lymanlimit, although only for a small stretch of wavelength be-fore it was cut-off by an intervening absorber, but this isconsistent with a very low opacity.The light curve and spectral energy distribution werestudied in detail by Nysewander et al. (2009), who mod-elled their Bgri optical data together with H -band photome-try from Molinari et al. (2007). They concluded that a rest-frame dust extinction of zero was ruled out at the . σ level.The shape of the extinction law is only weakly constrainedby these data, so extrapolating to the Lyman limit intro-duces a large systematic uncertainty, but with reasonabledust laws their favoured extinction would correspond to avalue of A & . If this inference is correct, then it wouldsuggest the actual escape fraction at the Lyman limit forGRB 060607A could be significantly diminished by dust ex-tinction, by a factor ∼ . or more.We note that there is also marginal evidence of X-rayabsorption in the source-frame at a level of N H , X ≈ ( . ± . )× cm − (Evans et al. 2009) over the Milky Way fore-ground (Willingale et al. 2013). This would be broadly con-sistent with a SMC dust-to-gas ratio (Bouchet et al. 1985)providing the hydrogen associated with this gas had largelybeen ionized (so it was not seen in the optical spectrum)but the dust had mostly not been destroyed. On the otherhand, Prochaska et al. (2008) argue that the absence of N v absorption argues for both low density and low metallicitysurrounding the burst location. Gas that has been ionized by massive star radiation withina host galaxy will generally recombine quickly. A fractionof recombining H ions will go directly to the ground-state,and so emit a photon just above the Lyman limit energy(some higher energy photons will also be emitted by recom-bining He ions). In low column-density systems a fraction ofthese will escape the host without further absorption, andtherefore re-boost the escaping ionizing flux, albeit with ra-diation that will soon be redshifted to energies below 1 Ryd.In other words, simply translating H i column-density intoline-of-sight opacity is likely to lead to a small underestimateof the escape fraction in low column-density systems. Thenet effect of this re-boost depends on various factors, butcould provide an increase of up to 10–20% in the effectiveescape fraction (Faucher-Gigu`ere et al. 2009), thus at leastpartially offsetting any dust extinction. For the remainder of Section 3 we will continue to con-sider only the opacity due to H i absorption, but will returnto the potential systematic effects of dust in Section 4.1.1. Even with our considerably larger sample of sight-lines, thefact that only two have any appreciable escape fractionmeans that to some extent we are still dealing with rathersmall number statistics. We also lack a robust theoreticalmodel which could be fit to the data, and so must explorethe statistical uncertainties non-parametrically.Again we first follow Chen et al. (2007) by performinga bootstrap exercise, employing random resamples ofthe data with replacement. From this we estimate a 98%confidence upper limit of h f esc i < . ; the result is thesame whether or not we allow the resampled N H i valuesto have additional scatter based on the error bars for eachpoint.In an alternative approach, we simulated several largepopulations of sight-lines with higher values of average es-cape fraction than found in our data (by replicating theGRB 050908/060607A values), and drew random 138-member samples from each of these. For the case of thepopulation with h f esc i pop = . we found 98% of randomsamples produced h f esc i samp > . . Thus, these two meth-ods agree on an upper limit for h f esc i of . – . . A similaranalysis gives a 98% lower limit of h f esc i > − .These are significantly tighter constraints than found bythe previous studies of Chen et al. (2007) and Fynbo et al.(2009) of h f esc i < . at 95%, due to our larger sample sizeand the fact that no further very low column-density sight-lines have been identified in any of the additional GRBs.We note that our result is also consistent with the h f esc i = . ± . obtained by constraining the flux below theLyman-limit in a stacked spectrum of eleven GRB afterglowswith ¯ z = . by Christensen et al. (2011). It is worth noting that our N H i distribution is inconsis-tent with the predictions of Cen & Kimm (2014) who usedhigh-resolution cosmological radiation-hydro simulations ofgalaxies within the EoR ( z ∼ ) to explore column-densitiesalong GRB sight-lines. They found a bimodal distributionwith a peak at high column-density ( log ( N H i / cm − ) ∼ –22), similar to the observed distribution, but then anothersubstantial peak with column-densities log ( N H i / cm − ) < which is not seen in practice. Part of the explanation couldbe that Cen & Kimm (2014) assumed that the GRB ratetraces the SNII rate and found a large fraction of their lowcolumn-density GRBs occurred in super-solar metallicity en-vironments, whereas in reality GRB progenitors seem to beyounger at explosion (e.g. Larsson et al. 2007) and crucially,unlike SNII, are rarely found in high metallicity galaxies (e.g.Perley et al. 2015). On the other hand, their simulations donot account for the effect of GRBs in ionizing gas local to theburst (see Section 4.1.2), and it seems when a high column-density is found in their models it is often due to such localgas, whereas the low column-density cases occur for pro-genitors that have escaped their birth clouds. This suggests MNRAS000 , 1–31 (2018)
N. R. Tanvir et al. their simulations do not capture the distributed nature ofneutral hydrogen in these star-forming galaxies, at least asit is found in z ∼ –5 GRB hosts.Our results do correspond much more closely with theearlier simulations of Pontzen et al. (2010), who similarly in-vestigated sight-lines to young star forming regions in modelgalaxies, but in this case did specifically consider the z ∼ –5 range. They found a median log ( N H i / cm − ) of 21.5 andcalculated an escape fraction of h f esc i ≈ . , both close toour findings. Indeed, their default prescription assumes thatGRBs trace star formation up to an age of 50 Myr, whereasrestricting to a perhaps more realistic 10 Myr age reduces h f esc i to 0.007. These simulations included the effect of localionizing sources on the gas proximate to the burst, but againnot the potential additional effect of ionization due to theGRB itself. We return to these issues in Section 4. As shown by the red line in Figure 1, there is no evidence forsignificant variation in the median value of log ( N H i ) betweenredshifts z ∼ to z ∼ . To investigate this further, in Fig-ure 2 we plot cumulative N H i distributions for three subsetsof the whole sample cut in redshift. It is apparent that thereis little difference between the low ( z < ) and intermediateredshift ( < z < ) sub-samples – the median values arethe same, and a two-sample Kolmogorov-Smirnov (KS) testfinds them to be consistent with the null hypothesis thatthey are drawn from the same parent distribution (p-valueof 0.88). The intermediate redshift sub-sample does have asomewhat longer tail to low column-density than the lowredshift sub-sample, and we note that low column-densitysystems are arguably rather more likely to go unrecognisedat lower redshifts (discussed further in Section 4.2.2). How-ever, another potential selection effect is that the proportionof dusty sight-lines appears to decline with increasing red-shift above z ∼ (Kann et al. 2010; Perley et al. 2016a,b),and if dusty bursts are systematically lost from the low red-shift sub-sample, due to the difficulty of locating and obtain-ing spectra for the afterglows, then it could mask a more sig-nificant evolutionary trend. The issue of biases due to dustis one we return to in subsequent sections.There is a suggestion of a more significant decline inthe typical values of log ( N H i ) at z & (as also pointed outby Chornock et al. 2014; Melandri et al. 2015), which influ-ences the final z > bin (red line). However, the conclusionis still limited by small number statistics, and a KS testagain finds this z > sub-sample to be consistent with be-ing drawn from the same distribution as the lower redshift( z < ) sub-sample (p-value of 0.31). In Figure 3 we plot log ( N H i ) versus host UV absolute mag-nitude, M UV , which is a gauge of the current (unobscured)star formation rate, for those bursts in our sample wheregood constraints on host luminosity are available in the lit-erature (37 measurements and 18 upper limits, detailed inTable 1). The restricted number of cases for which deep hostsearches have been conducted, and the fact that many areupper limits, means we cannot draw firm conclusions, but Figure 2.
Cumulative distribution of H i for the whole sampleand split into subsets by redshift, as indicated. The correspondingoptical depth at Ly α is shown on the top axis. A power-law fitto the lower column-density half of the sample is shown by theorange dashed line (see text). there is little indication of a dependence of the average H i column-density on the current star formation rate. Note, wehave converted these values to a common cosmology (a flatFriedmann model with Ω M = . , H = km s − ), but nototherwise attempted to correct for differences in the proce-dures used by different authors (approaches to k-corrections,for example), or small deviations from the reference 1600 ˚Awavelength generally adopted. These distinctions should notbe at a level that would affect the conclusion.Similarly, in Figure 4 we plot log ( N H i ) against host stel-lar mass, M ∗ , for those galaxies in our sample for whichestimates (or limits) are available in the literature (Ta-ble 1). The bulk are based on Spitzer infrared photometry,particularly from Perley et al. (2013), Perley et al. (2016b),Laskar et al. (2011) and Myers et al. in prep. Here we arerestricted to 28 galaxies with M ∗ estimates and 31 galax-ies with upper limits. Again there is little indication ofany trend, contrary to suggestions that f esc may correlatestrongly with galaxy size (e.g Anderson et al. 2017).One trend that is conspicuous in Figure 3, and in par-ticular Figure 4, is that the UV brighter and more massivehosts are predominantly in the < z < bin. It is true thatsmall and faint hosts are harder to detect at higher redshifts,and that may explain some part of this trend. However, an-other factor seems to be that higher mass ( M ∗ & M ⊙ )hosts at z < are more often associated with heavily ex-tinguished bursts, and hence more likely to be highly dusty,than at higher redshifts (Perley et al. 2016b); these are pre-sumably systematically under-represented in our compila-tion. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Figure 3.
The values of neutral hydrogen column-density plottedagainst host rest-frame UV absolute AB magnitude and corre-sponding star formation rate (top axis, following the calibrationfor a Z = Z ⊙ / , constant star-formation rate and age > Myrpopulation from Madau & Dickinson 2014), for 37 detections and19 upper limits collected from various sources (Table 1). Colourcoding is as with the lines in Figure 2 and upper limits are σ . A number of systematic uncertainties may affect our anal-ysis, potentially biasing the conclusions. These include ob-servational selection effects: in order to be included in thesample afterglows must be localised and spectra obtainedwhich cover the Ly α region with reasonable signal-to-noise.Other concerns relate to the still-uncertain nature ofthe GRB progenitor itself, both in terms of the assump-tion that their sight-lines are representative of dominantEUV-producing stars, and specifically whether there couldbe special circumstances required to produce GRBs that ne-cessitate an atypical environment. It is also pertinent to askwhether the results can be extrapolated to the EoR.In this section we begin by investigating systematic ef-fects which may influence our calculation of h f esc i in termsof its application to the GRB progenitors. The majority ( ∼ %; Gehrels & Razzaque 2013) of Swift -discovered GRBsdo not have redshifts, which, it may be thought, could leadto large biases. With a more uniformly selected sample, someof these effects might be quantified via simulated data-sets,but that approach would be of limited utility here. However,by considering the nature of these selection effects, and sur-veying the available data, we shall show that these biases canbe understood well enough to confirm that they are unlikelyto affect the main conclusions.We then consider more carefully how representative the
Figure 4.
The values of neutral hydrogen column-density plot-ted against host stellar mass estimates from a range of sources(Table 1) for 60 hosts, 32 of which are upper limits (shown at2 σ ). Colour coding is as with the lines in Figure 2. The approxi-mate stellar masses of the Milky Way and Magellanic Clouds areindicated on the top axis for comparison. GRB progenitors are of the EUV-producing massive stars,and the potential systematic uncertainties that may intro-duce. Finally we discuss in more detail the application ofour results to the EoR. h f esc i Two factors are very likely to produce an over-estimate ofthe escape fraction, as discussed here.
GRB afterglow surveys are biased against high column-density sight-lines since they tend to be dustier andhence harder to locate and obtain redshifts for inthe optical band (Fynbo et al. 2009; Greiner et al. 2011;Watson & Jakobsson 2012). Dusty systems at higher red-shift would be particularly susceptible to being lost sinceeven near-IR observations will be looking in the source-frameoptical or near-UV. However we expect dusty systems tobe rarer at z & (Perley et al. 2016b) and particularly by z ∼ (cf. Zafar et al. 2011; Schaerer et al. 2015). RecentlyBolmer et al. (2018) have shown that the decline of obscu-ration in GRB afterglows with increased redshift is likelyprimarily due to their hosts being less dusty rather than anobservational selection effect.Over all redshifts, Perley et al. (2016a) estimate thatapproximately 20% of GRBs detected by Swift are heav-ily dust obscured, and often reside in globally dusty, mas-sive galaxies (Rossi et al. 2012; Perley et al. 2013); this sub-population is significantly biased against in afterglow sam-
MNRAS , 1–31 (2018) N. R. Tanvir et al. ples. A similar proportion show signs of more moderate dustobscuration (see also Covino et al. 2013). Thus this effect,while important for understanding the overall GRB hostpopulation, would likely only necessitate a comparativelyminor correction to the escape fraction estimate (i.e. a ∼ –40% reduction, assuming that all very dusty sight-lines areopaque to EUV), and the correction is likely to be greatestat z < , as discussed in Section 3.4.Even in some cases where afterglows have been detectedand redshifts measured, there may be significant attenuationin the UV by dust. However, from the point of view of escapefraction, this is only relevant for the two sight-lines with non-negligible f esc , and as already discussed in Section 3.1 maylead to corrections (downward) of a factor ∼ or more forour sample. The GRB prompt flash and early afterglow produces anintense radiation field that is expected to quickly de-stroy dust (Waxman & Draine 2000; Morgan et al. 2014)and ionize gas to distances of up to several tens of pcwhen the ambient medium has a low to moderate density, n . cm − (Perna & Lazzati 2002; Vreeswijk et al. 2007;Krongold & Prochaska 2013). Therefore the opacity mea-sured to the afterglow could in principle be less than theopacity that would have been seen to the progenitor starsystem. Various lines of evidence suggest the existence of anionized gas component, likely reasonably local to the GRB.In particular, it has been argued that high column-densitiesmeasured in X-ray absorption, which in some afterglows sig-nificantly exceed the optical measures, are at least partiallydue to denser gas close to the progenitor in which the hy-drogen has been ionized (Watson et al. 2007; Schady et al.2011). Furthermore, the observed correlation of X-ray ab-sorption with local galaxy surface brightness (in opticallybright GRBs) may support a local origin for a significantproportion of the absorbing gas (Lyman et al. 2017). Highlyionized species have also been seen in some afterglow spec-tra, which are likely to be of circumburst origin (Fox et al.2008; De Cia et al. 2011, Heintz et al. submitted.).On the other hand, significant ionization may have beenbrought about by the stellar radiation field prior to the burst(Watson et al. 2013; Krongold & Prochaska 2013), render-ing the ionizing effect of the GRB largely irrelevent. Thisis supported by models of feedback from star formation inmassive molecular clouds which show that hot ionized bub-bles can grow to tens or hundreds of pc in a few Myr, al-though in some circumstances, in particular if star formationis relatively inefficient, the outflow can stall and the cloudrecollapse, leading to further star formation episodes (e.g.Rahner et al. 2017).In principle, deep time-resolved spectroscopy may al-low the GRB-driven ionization of H i to be observed directly(Perna & Lazzati 2002), although in practice, sufficientlygood data-sets have rarely been acquired. Only in one case,namely GRB 090426, has time-variability of Ly α been seen,between spectra obtained at 1.1 hr and 12 hr post-burst,suggesting the influence of the GRB. Here photoionizationmodelling placed the absorbing gas at ∼ pc (Th¨one et al.2011). As discussed in Appendix A14, there is uncertainty regarding this particular burst as to whether it is of the longor short duration class, but nonetheless it confirms the localionizing effect of GRB emission can occur even at quite largedistances.In one other case, GRB 080310, Vreeswijk et al. (2013)found their time-dependent photoionization model to beimproved with the addition of a cloud at ∼ –50 pcfrom the burst, which became fully ionized by the earlyafterglow emission. The required column-density of thiscloud, of log ( N H i / cm − ) ∼ –20, was greater than the log ( N H i / cm − ) = . inferred for the observed neutral ab-sorber.In conclusion, it is likely that some fraction of GRBs ex-hibit a reduced H i column-density due to the ionizing effectof the burst itself. If windows to the IGM often occur whensuperbubbles puncture low-density channels out of galacticneutral gas (e.g. Dove et al. 2000; Roy et al. 2015), then thismay be largely irrelevant as far as high escape fraction sight-lines are concerned. On the other hand, regarding our sam-ple, if GRB 060607A was dust extinguished (Section 3.1.2)then it would be surprising if the dust was not associatedwith some neutral gas which was ionized by the burst. h f esc i A reasonable question is whether some very low N H i sys-tems may not have been recognised in GRB afterglow ob-servations. Omitting from our sample, or over-estimating thecolumn-density, of even a fairly small number of such burstscould lead to significant under-estimation of h f esc i . Thereare several circumstances in which this could plausibly arisethat are discussed below. The nature of such biases dependson the quality of the afterglow spectroscopy, and so we splitour discussion into three broad categories: good S/N spectra(Sections 4.2.2 and 4.2.3), poor S/N spectra that were stillsufficient to provide a redshift (Section 4.2.5), and instanceswhere no redshift was obtained due to the faintness of theoptical afterglow (Section 4.2.6). These divisions are some-what qualitative, but this is appropriate given that we do nothave access to many individual spectra, and also noting thatcontinuum S/N can vary significantly within a spectrum.However, first we consider what can be said about thepotential level of such selection effects, by reference to anearly redshift-complete GRB sample. Only a fraction of X-ray localised GRBs have optical/nIRafterglow identifications, and only a fraction of those haveredshifts from afterglow spectra. Thus, if the chances of aredshift being obtained depend on the H i column-densitythen it could bias our results. On the other hand, many By way of illustration, to the end of 2017,the database maintained by Jochen Greiner, , lists 1200GRBs with X-ray counterparts and 720 with opticalidentifications. Of these, 301 have afterglow redshifts ac-cording to the database maintained by Daniel Perley, .MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRBs receive little ground-based follow-up for other rea-sons, for example, because of poor weather at major obser-vatories, because they are badly placed for observation dueto proximity to the Sun, Moon or Galactic plane, or simplybecause of the limited availability of large telescopes to makethe necessary rapid target-of-opportunity observations.Since our sample is not selected or observed in a uni-form way, and many values are taken from the literature, itis hard to assess the maximum scale of these effects directly.However, we can get a handle on them by considering thewell-defined
Swift
Gamma-Ray Burst Host Galaxy LegacySurvey (SHOALS) sample (Perley et al. 2016a), which con-sists of 119 long-duration GRBs discovered by
Swift up untilOctober 2012, and excludes bursts that were poorly placedfor ground observation (whether or not they have redshifts).SHOALS imposes a threshold on the prompt γ -ray fluenceof S −
150 keV > − erg cm − , which does mean it excludessome intrinsically very weak events, but prompt emissionis thought to arise from internal processes within the GRBjet and so not to be dependent on the nature of the ambi-ent environment. Furthermore, despite many searches, thereis little indication that prompt γ -ray behaviour dependson other properties of the host, such as metallicity (e.g.Levesque et al. 2010b; Japelj et al. 2016). SHOALS also re-quires that bursts have identified X-ray counterparts, butsince all long-bursts are detectable in X-rays if observed suf-ficiently early by Swift , the selection criterion employed wassimply that a rapid autonomous slew was performed. Thisis consistent with theoretical expectations that the X-rayafterglow flux should be independent of ambient density, n ,unless it is very low ( n < − cm − , e.g. Hasco¨et et al. 2011).The SHOALS sample has a high degree of redshift com-pleteness thanks in large part to major efforts to obtain red-shifts from host galaxy observations (identified within X-rayor optical afterglow error boxes) where they had not alreadybeen obtained from afterglow spectroscopy. Specifically, 92%have spectroscopic or (in a few cases) good photometric red-shifts, and all but one have some photometric constraint onthe redshift.Here we restrict our attention to the 80 bursts for whichthe redshift is z > . or for which the constraints allow thepossibility of the burst being in that range. Of these wecan immediately say that 52 were very likely high column-density sight-lines, either because N H i was measured di-rectly (39) or because they were found to have faint after-glows with indications of high levels of extinction (14) ac-cording to Perley et al. (2016a, see also Section 4.1.1). One,namely GRB 060607A, is the same low-column system in-cluded in our sample.Of the remainder, 19 appear to have had at least moder-ately bright optical afterglows, but either spectroscopy wasobtained which did not cover the wavelength of Ly α (8)and the redshifts rely on metal lines (although in no casewere these lines reported as being unusually weak) or sim-ply no spectroscopy was attempted to our knowledge (11).A further four had little afterglow follow-up of any kindreported; these were GRB 050128, early in the Swift mis-sion, GRB 050726, for which real-time alerts were not sentto the ground, GRBs 050922B which occurred on the sameday as several other high-priority bursts, and GRB 070328.We find no reason to think any of these bursts lacked Ly α measurements due to observational selection effects, since they all seem to be cases where only limited follow-up wasattempted.This leaves only three sources, which merit more thor-ough scrutiny. One of these, GRB 071025, was observed withKeck/HIRES, but the spectrum was low-S/N with flux onlybeing detected at λ > ˚A. Fynbo et al. (2009) arguedthat this may be due to a Ly α break at z ≈ . (the lowS/N precluding measurement of the line strength), or alter-natively that it could indicate a highly dust reddened af-terglow at lower redshift. A photometric redshift constraintfrom multi-band afterglow imaging supports a high redshift( z ∼ ) interpretation, whilst also favouring fairly substan-tial dust extinction (Perley et al. 2010). Thus it seems thisis an intrinsically bright event but, again, most likely witha high column-density sight-line.GRB 100305A was the target of several deep imagingobservations within the first hour post-burst, but the onlycandidate afterglow (Gemini/GMOS observations in riz ;Cucchiara 2010) was subsequently found to be outside therevised X-ray error circle and to be present as a steadysource in later imaging (Perley et al. 2016a). We haveanalysed previously unpublished early UKIRT data, andalso find no afterglow down to a 2 σ limit of K AB = . at40 min after the trigger. In fact the Swift /XRT spectrum(see ;Evans et al. 2009) does show significant X-ray absorp-tion above the Galactic value, suggesting a high columnsightline, possibly combined with moderately high redshiftmaking the optical/nIR afterglow faint.Finally we have GRB 070223 is known to be at redshift z = . from the host galaxy (Perley et al. 2016a). Here theafterglow was faint in both the optical and near-infrared,despite early follow-up, meaning that no spectroscopywas attempted. The host galaxy was detected in Spitzer µ m imaging, but the implied stellar mass is a relativelymodest log ( M ∗ / M ⊙ ) ≈ . (Perley et al. 2016b). We havereanalysed the early imaging obtained at the LiverpoolTelescope and the WHT at ≈ hr post-burst (detailsare given in Appendix A3), finding AB magnitudes of r = . ± . and K = . ± . for a faint source at theX-ray afterglow position. However, we have also analysedthe SDSS and PanSTARRS imaging of the same region,and in both cases find a persistent source, presumably thehost, at the same location, with a magnitude r = . ± . .Thus it seems clear that the optical source seen by theLT (and also the MDM 1.3m; Mirabal et al. 2007) wasactually host dominated, and hence the optical afterglowmust have been substantially fainter. By contrast, the K -band source faded by 0.7 mag by the following week,confirming an afterglow detection in the near-IR (Rol et al.2007). Thus it seems likely that, despite not being ina massive dusty host, this event too was heavily extin-guished, which is consistent with the high column-densityinferred from the X-ray spectrum of log ( N H i , X / cm − ) ≈ . (see ;Evans et al. 2009).In summary, from our analysis of the SHOALS sam-ple, of 80 bursts that may be at z > . , 55 have evidenceof high- N H i and/or high extinction and 1 has low column( log ( N H i ) < ). In all other cases, limited follow-up seems tobe the primary reason for a lack of a constraint on N H i . Thisis worth emphasising: even amongst bursts that were chosen MNRAS , 1–31 (2018) N. R. Tanvir et al. as being well-placed for follow-up and which had at leastmoderately bright afterglows, a significant number of events( ∼ %) lack spectroscopic constraints on Ly α absorptionfor reasons that seem to be unrelated to the afterglow prop-erties. Thus, it seems that the large majority of opticallyfaint bursts are dust extinguished, with a smaller numberat high redshift and hence optical “drop-outs”. The predom-inant selection effect, then, leads to high- N H i bursts beinglost from the sample (already discussed in Section 4.1.1).This suggests that any bursts that are lost from our sampledue to selection against low column-density systems mustbe few in number. In rare cases, like GRB 071025 discussed above, afterglowspectra are acquired in which no absorption features can beseen at a reasonable confidence level, or that exhibit onlymarginal features that cannot be unambiguously identified.This could be due to foreground gas in the host having verylow column-density such that it produces neither a clear Ly α feature nor detectable metal lines, with the net result thatno redshift is obtained. However, in our experience, suchapparently featureless spectra are nearly always cases whereeither the continuum level has very low signal-to-noise (S/N)ratio (as was the case for GRB 071025), thus not necessitat-ing an especially low column-density, and/or the spectrumonly covers a relatively short wavelength range and so mayeasily miss prominent absorption features.Problems associated with low-S/N spectra are discussedin Section 4.2.5. The possibility that intrinsically fainter af-terglows, which typically result in no afterglow redshift be-ing determined, may on average have low column-densityabsorbers, we return to in Section 4.2.6. Here we restrict at-tention to whether weak absorption features could have ledto no redshift being found despite the spectra being of mod-erate to good S/N and spanning a wide wavelength range.Again, our experience suggests such circumstances arevery rare: we are not aware of any compelling examples,published or unpublished. A much discussed near-miss wasGRB 070125, for which the absorption lines were very weak,but ultimately the redshift was found to be z = . from Mg ii absorption seen in a Gemini/GMOS spectrum(Cenko et al. 2008). A later Keck LRIS spectrum of theafterglow, which extended to shorter wavelengths, showedmarginal evidence for Ly α absorption, but this was onlysufficient to conclude log ( N H i / cm − ) < . in the host(Updike et al. 2008). Based on the weakness of the metallines, De Cia et al. (2011) argued that the neutral hydrogencolumn-density was probably low, likely in the LLS range,but that this could have been substantially diminished bythe particularly intense afterglow radiation ionizing gas toa considerable distance. Given that Ly α was so far into thenear-UV in this case, around 3100 ˚A, which is hard to cal-ibrate in ground-based data, we did not include it in oursample. The unusual nature of this system is illustratedby the fact that GRB 070125 had the lowest “line-strength-parameter” (an index based on the strength of absorptionlines compared to the average over the sample) out of 69spectra studied by de Ugarte Postigo et al. (2012).Another instructive case is GRB 140928A, for which spectroscopy was obtained with Gemini/GMOS-S(Cucchiara et al. 2014). Here the afterglow continuumwas clearly detected, but no unambiguous lines were seen,despite the S/N being moderately good (S/N ≈ α to fall within the spectrum wouldhave required z > . , which would mean we would haveexpected a clear break due to the onset of the Ly α -forest,irrespective of the host column-density. This is not seen, sowe can conclude that Ly α very likely was not within thespectral window in this case.Thus this example highlights an important point re-garding weak-lined spectra, namely that at least above red-shift z ∼ strong attenuation due to the Ly α forest wouldnormally be expected to be clearly seen in reasonable S/Noptical spectra covering the relevant wavelength range, giv-ing good indications of the redshift, even in the absence ofany host absorption.Finally we note that, while there have been occasionalinstances when host galaxy follow-up has revealed an earlierclaimed afterglow redshift (based on a low-S/N spectrum)to be mistaken (e.g. Jakobsson et al. 2012), to our knowl-edge none of these have indicated a case where the after-glow spectrum should have revealed Ly α absorption whichwas not seen. All these considerations suggest that any biasintroduced by the effect of low host column-density goingunrecognised despite good afterglow spectroscopy should beminor compared to the other effects we consider. A more subtle question is whether N H i values may be over-estimated simply due to the measurement process, partic-ularly for low-S/N spectra. This should not be a majorconcern in the majority of cases, where damping wings areclearly seen and fitted, confirming the high column-density.For cases with rest-frame equivalent width of Ly α less than W ∼ ˚A (roughly log ( N H i / cm − ) ∼ . ), especially whenobserved at low spectral resolution (typically R . ), un-certainty in the velocity structure of the absorbing gas leadsto relatively high uncertainty in the inferred H i column-density. If the range in velocity of the absorbing gas is under-estimated, for example if due to several clouds with differ-ent velocities, then it would lead to an over-estimate of thecolumn-density.Of our sample, three bursts both fall into this categoryand lack direct evidence of emission or otherwise below theLyman limit, namely GRBs 060124, 060605 and 090426. Thelast of these was unusual in exhibiting apparent variabilityof Ly α absorption (Appendix A14), suggesting absorptiondominated by a single absorber. The other two are moredifficult cases, although the spectral resolution is sufficientto rule out a high spread in velocity (cf. GRB 021004, Ap-pendix A1), and the inferred H i columns (and error bars)appear to have considered a fairly conservative range ofDoppler parameters, making a significant over-estimate un-likely. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation A similar possible scenario involving very low column-density would be where the host absorption lines were notidentified in the spectrum at all, but instead chance align-ment with a stronger intervening absorption system led tothe incorrect assignation of its redshift as the redshift ofthe burst, along with an erroneous column-density. Again,this is likely to be a rare circumstance since the incidenceof strong intervening absorbers is not high and one wouldnormally expect to see the Ly α forest from the IGM, par-ticularly above z ∼ , which would allow identification of anunassociated Ly α absorber as being due to an interveningsystem. We also note that in some spectra we detect metalfine-structure lines, which are thought to be the result ofexcitation by the burst itself of gas within its host galaxy,confirming the association (e.g. Vreeswijk et al. 2006).A particular example that highlighted this concern wasGRB 071003, in which it was found that the highest red-shift system, a detection of Mg ii , presumed to be fromthe host, was notably weaker than some intervening ab-sorbers (Perley et al. 2008). Similarly, GRB 060605 exhib-ited weak Ly α from the host, but stronger ( log ( N H i / cm − ) = . ) from an intervening system at slightly lower redshift(Ferrero et al. 2009).Another pertinent case is GRB 141026A, the afterglowof which was observed by GTC, with a spectrum cover-ing wavelength range 5100–9800 ˚A. The S/N was ratherpoor, but an absorption line was seen close to the blueend of the spectrum that if interpreted as Ly α would im-ply z = . and a low column-density of log ( N H i / cm − ) . (de Ugarte Postigo et al. 2014b). In this instance there wereno other features seen to confirm the line identification, andno evidence of a decrement that could be ascribed to the Ly α forest, for which reasons we chose not to include this burstin our sample. Thus, this example illustrates that misidenti-fication of redshift might in some circumstances result in abias in the opposite direction, namely toward lower column-density.Once again, we conclude that whilst it is hard to rule outcompletely, the rate of strong intervening absorbers beingfalsely identified as host systems, providing good spectraare obtained, must be very low. Some afterglow spectra are sufficient to provide redshifts,but the signal-to-noise, at least around the Ly α region, ispoor. This may lead to N H i being undetermined, particu-larly if it is low, thus creating a bias in favour of includinghigher column-density systems. Amongst our sample, onlyeleven bursts lack clear metal line detections, and of thesefive have tentative metal line detections (GRBs 020124,060927, 080129, 080913, 121229A), and five have no metalline detections but do show an unambiguous continuumbreak at Ly α that is sufficiently well defined to constrainthe wing profile (GRBs 060522, 081203, 090519, 100316A,130427B, 140515A). The latter subset all have low S/N, andthe search for metal lines was complicated by low spectralresolution and/or being in a difficult region of the spectrum,but reassuringly they span a wide range of N H i values, whichis not suggestive of any particular bias. This gives confidence that our sample derives predominantly from high-S/N spec-tra, and contains few bursts which are only included becausethey had a particularly high value of H i column-density.Several other bursts have a redshift determined from theLy α break, but the S/N proved insufficient to estimate the N H i value. These cases are few in number: apart from severalat z & , from a search of spectra we have ourselves andthe literature we have only identified GRBs 071025 ( z ≈ ;Section 4.2.1), 140428A ( z ≈ . ; Perley 2014) and 160327A( z ≈ ; de Ugarte Postigo et al. 2016). Thus we believe thatthese cases, while they may be below the median N H i forall bursts, are not likely to be unusually low- N H i . A smallbias could partially offset the bias against dusty sight-linesdiscussed previously.It is notable that redshifts can be obtained from low-S/N spectra when the redshift is comparatively high, whichcan be understood because the strength of the Ly α -breakincreases with redshift. At redshifts below z ∼ such spec-tra likely will not yield secure redshifts, a category that isdiscussed in the next section. We have argued in the preceding sections that bursts areunlikely to have been lost from our sample due to weak ab-sorption lines providing that good spectra were obtained.However, bursts with very faint optical afterglows will beunder-represented due to the increased difficulty of arcsec-ond localisation and redshift determination (either becausespectroscopy was not attempted, or because spectra had toolow S/N to give a conclusive redshift or N H i measure). Thus,if bursts occurring in low density environments had weakerlines and also on average fainter afterglows, then that po-tentially may lead us to systematically lose bursts with high f esc . One way a GRB progenitor could find itself in a lowerdensity environment, would be if it was formed by a so-called “runaway” star. We consider this particular issue inSection 4.4, but here focus on the potential effect of lowdensity on the brightness of afterglows and the likelihoodthat such systems have been missed.As discussed above, the majority of optically faint after-glows are dust extinguished, and have high EUV opacities,while a smaller number are high redshift optical drop-outs.We should also remember that some afterglows were faintwhen observed simply due to the delay in acquiring spec-troscopy. This suggests that the fraction of systems thatare faint due to low density circumburst media is low. Onthe other hand, basic synchrotron afterglow theory providessome motivation for thinking such a trend might occur. Inparticular, for a relativistic jet shocking a medium of uni-form density, n , in typical circumstances the optical after-glow flux should scale with n / (Granot & Sari 2002). Infact, with sufficiently good wide-band monitoring of the af-terglow, the ambient density of the medium in which the jetis travelling (i.e. sub-pc scales) can be calculated. The rangeof circumburst medium densities inferred from such mod-elling is quite wide, from − to cm − (e.g. Laskar et al.2014), but equally is subject to model assumptions and largeuncertainties in many cases.However, the crucial point is that the density struc- MNRAS000
150 keV > − erg cm − , which does mean it excludessome intrinsically very weak events, but prompt emissionis thought to arise from internal processes within the GRBjet and so not to be dependent on the nature of the ambi-ent environment. Furthermore, despite many searches, thereis little indication that prompt γ -ray behaviour dependson other properties of the host, such as metallicity (e.g.Levesque et al. 2010b; Japelj et al. 2016). SHOALS also re-quires that bursts have identified X-ray counterparts, butsince all long-bursts are detectable in X-rays if observed suf-ficiently early by Swift , the selection criterion employed wassimply that a rapid autonomous slew was performed. Thisis consistent with theoretical expectations that the X-rayafterglow flux should be independent of ambient density, n ,unless it is very low ( n < − cm − , e.g. Hasco¨et et al. 2011).The SHOALS sample has a high degree of redshift com-pleteness thanks in large part to major efforts to obtain red-shifts from host galaxy observations (identified within X-rayor optical afterglow error boxes) where they had not alreadybeen obtained from afterglow spectroscopy. Specifically, 92%have spectroscopic or (in a few cases) good photometric red-shifts, and all but one have some photometric constraint onthe redshift.Here we restrict our attention to the 80 bursts for whichthe redshift is z > . or for which the constraints allow thepossibility of the burst being in that range. Of these wecan immediately say that 52 were very likely high column-density sight-lines, either because N H i was measured di-rectly (39) or because they were found to have faint after-glows with indications of high levels of extinction (14) ac-cording to Perley et al. (2016a, see also Section 4.1.1). One,namely GRB 060607A, is the same low-column system in-cluded in our sample.Of the remainder, 19 appear to have had at least moder-ately bright optical afterglows, but either spectroscopy wasobtained which did not cover the wavelength of Ly α (8)and the redshifts rely on metal lines (although in no casewere these lines reported as being unusually weak) or sim-ply no spectroscopy was attempted to our knowledge (11).A further four had little afterglow follow-up of any kindreported; these were GRB 050128, early in the Swift mis-sion, GRB 050726, for which real-time alerts were not sentto the ground, GRBs 050922B which occurred on the sameday as several other high-priority bursts, and GRB 070328.We find no reason to think any of these bursts lacked Ly α measurements due to observational selection effects, since they all seem to be cases where only limited follow-up wasattempted.This leaves only three sources, which merit more thor-ough scrutiny. One of these, GRB 071025, was observed withKeck/HIRES, but the spectrum was low-S/N with flux onlybeing detected at λ > ˚A. Fynbo et al. (2009) arguedthat this may be due to a Ly α break at z ≈ . (the lowS/N precluding measurement of the line strength), or alter-natively that it could indicate a highly dust reddened af-terglow at lower redshift. A photometric redshift constraintfrom multi-band afterglow imaging supports a high redshift( z ∼ ) interpretation, whilst also favouring fairly substan-tial dust extinction (Perley et al. 2010). Thus it seems thisis an intrinsically bright event but, again, most likely witha high column-density sight-line.GRB 100305A was the target of several deep imagingobservations within the first hour post-burst, but the onlycandidate afterglow (Gemini/GMOS observations in riz ;Cucchiara 2010) was subsequently found to be outside therevised X-ray error circle and to be present as a steadysource in later imaging (Perley et al. 2016a). We haveanalysed previously unpublished early UKIRT data, andalso find no afterglow down to a 2 σ limit of K AB = . at40 min after the trigger. In fact the Swift /XRT spectrum(see ;Evans et al. 2009) does show significant X-ray absorp-tion above the Galactic value, suggesting a high columnsightline, possibly combined with moderately high redshiftmaking the optical/nIR afterglow faint.Finally we have GRB 070223 is known to be at redshift z = . from the host galaxy (Perley et al. 2016a). Here theafterglow was faint in both the optical and near-infrared,despite early follow-up, meaning that no spectroscopywas attempted. The host galaxy was detected in Spitzer µ m imaging, but the implied stellar mass is a relativelymodest log ( M ∗ / M ⊙ ) ≈ . (Perley et al. 2016b). We havereanalysed the early imaging obtained at the LiverpoolTelescope and the WHT at ≈ hr post-burst (detailsare given in Appendix A3), finding AB magnitudes of r = . ± . and K = . ± . for a faint source at theX-ray afterglow position. However, we have also analysedthe SDSS and PanSTARRS imaging of the same region,and in both cases find a persistent source, presumably thehost, at the same location, with a magnitude r = . ± . .Thus it seems clear that the optical source seen by theLT (and also the MDM 1.3m; Mirabal et al. 2007) wasactually host dominated, and hence the optical afterglowmust have been substantially fainter. By contrast, the K -band source faded by 0.7 mag by the following week,confirming an afterglow detection in the near-IR (Rol et al.2007). Thus it seems likely that, despite not being ina massive dusty host, this event too was heavily extin-guished, which is consistent with the high column-densityinferred from the X-ray spectrum of log ( N H i , X / cm − ) ≈ . (see ;Evans et al. 2009).In summary, from our analysis of the SHOALS sam-ple, of 80 bursts that may be at z > . , 55 have evidenceof high- N H i and/or high extinction and 1 has low column( log ( N H i ) < ). In all other cases, limited follow-up seems tobe the primary reason for a lack of a constraint on N H i . Thisis worth emphasising: even amongst bursts that were chosen MNRAS , 1–31 (2018) N. R. Tanvir et al. as being well-placed for follow-up and which had at leastmoderately bright afterglows, a significant number of events( ∼ %) lack spectroscopic constraints on Ly α absorptionfor reasons that seem to be unrelated to the afterglow prop-erties. Thus, it seems that the large majority of opticallyfaint bursts are dust extinguished, with a smaller numberat high redshift and hence optical “drop-outs”. The predom-inant selection effect, then, leads to high- N H i bursts beinglost from the sample (already discussed in Section 4.1.1).This suggests that any bursts that are lost from our sampledue to selection against low column-density systems mustbe few in number. In rare cases, like GRB 071025 discussed above, afterglowspectra are acquired in which no absorption features can beseen at a reasonable confidence level, or that exhibit onlymarginal features that cannot be unambiguously identified.This could be due to foreground gas in the host having verylow column-density such that it produces neither a clear Ly α feature nor detectable metal lines, with the net result thatno redshift is obtained. However, in our experience, suchapparently featureless spectra are nearly always cases whereeither the continuum level has very low signal-to-noise (S/N)ratio (as was the case for GRB 071025), thus not necessitat-ing an especially low column-density, and/or the spectrumonly covers a relatively short wavelength range and so mayeasily miss prominent absorption features.Problems associated with low-S/N spectra are discussedin Section 4.2.5. The possibility that intrinsically fainter af-terglows, which typically result in no afterglow redshift be-ing determined, may on average have low column-densityabsorbers, we return to in Section 4.2.6. Here we restrict at-tention to whether weak absorption features could have ledto no redshift being found despite the spectra being of mod-erate to good S/N and spanning a wide wavelength range.Again, our experience suggests such circumstances arevery rare: we are not aware of any compelling examples,published or unpublished. A much discussed near-miss wasGRB 070125, for which the absorption lines were very weak,but ultimately the redshift was found to be z = . from Mg ii absorption seen in a Gemini/GMOS spectrum(Cenko et al. 2008). A later Keck LRIS spectrum of theafterglow, which extended to shorter wavelengths, showedmarginal evidence for Ly α absorption, but this was onlysufficient to conclude log ( N H i / cm − ) < . in the host(Updike et al. 2008). Based on the weakness of the metallines, De Cia et al. (2011) argued that the neutral hydrogencolumn-density was probably low, likely in the LLS range,but that this could have been substantially diminished bythe particularly intense afterglow radiation ionizing gas toa considerable distance. Given that Ly α was so far into thenear-UV in this case, around 3100 ˚A, which is hard to cal-ibrate in ground-based data, we did not include it in oursample. The unusual nature of this system is illustratedby the fact that GRB 070125 had the lowest “line-strength-parameter” (an index based on the strength of absorptionlines compared to the average over the sample) out of 69spectra studied by de Ugarte Postigo et al. (2012).Another instructive case is GRB 140928A, for which spectroscopy was obtained with Gemini/GMOS-S(Cucchiara et al. 2014). Here the afterglow continuumwas clearly detected, but no unambiguous lines were seen,despite the S/N being moderately good (S/N ≈ α to fall within the spectrum wouldhave required z > . , which would mean we would haveexpected a clear break due to the onset of the Ly α -forest,irrespective of the host column-density. This is not seen, sowe can conclude that Ly α very likely was not within thespectral window in this case.Thus this example highlights an important point re-garding weak-lined spectra, namely that at least above red-shift z ∼ strong attenuation due to the Ly α forest wouldnormally be expected to be clearly seen in reasonable S/Noptical spectra covering the relevant wavelength range, giv-ing good indications of the redshift, even in the absence ofany host absorption.Finally we note that, while there have been occasionalinstances when host galaxy follow-up has revealed an earlierclaimed afterglow redshift (based on a low-S/N spectrum)to be mistaken (e.g. Jakobsson et al. 2012), to our knowl-edge none of these have indicated a case where the after-glow spectrum should have revealed Ly α absorption whichwas not seen. All these considerations suggest that any biasintroduced by the effect of low host column-density goingunrecognised despite good afterglow spectroscopy should beminor compared to the other effects we consider. A more subtle question is whether N H i values may be over-estimated simply due to the measurement process, partic-ularly for low-S/N spectra. This should not be a majorconcern in the majority of cases, where damping wings areclearly seen and fitted, confirming the high column-density.For cases with rest-frame equivalent width of Ly α less than W ∼ ˚A (roughly log ( N H i / cm − ) ∼ . ), especially whenobserved at low spectral resolution (typically R . ), un-certainty in the velocity structure of the absorbing gas leadsto relatively high uncertainty in the inferred H i column-density. If the range in velocity of the absorbing gas is under-estimated, for example if due to several clouds with differ-ent velocities, then it would lead to an over-estimate of thecolumn-density.Of our sample, three bursts both fall into this categoryand lack direct evidence of emission or otherwise below theLyman limit, namely GRBs 060124, 060605 and 090426. Thelast of these was unusual in exhibiting apparent variabilityof Ly α absorption (Appendix A14), suggesting absorptiondominated by a single absorber. The other two are moredifficult cases, although the spectral resolution is sufficientto rule out a high spread in velocity (cf. GRB 021004, Ap-pendix A1), and the inferred H i columns (and error bars)appear to have considered a fairly conservative range ofDoppler parameters, making a significant over-estimate un-likely. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation A similar possible scenario involving very low column-density would be where the host absorption lines were notidentified in the spectrum at all, but instead chance align-ment with a stronger intervening absorption system led tothe incorrect assignation of its redshift as the redshift ofthe burst, along with an erroneous column-density. Again,this is likely to be a rare circumstance since the incidenceof strong intervening absorbers is not high and one wouldnormally expect to see the Ly α forest from the IGM, par-ticularly above z ∼ , which would allow identification of anunassociated Ly α absorber as being due to an interveningsystem. We also note that in some spectra we detect metalfine-structure lines, which are thought to be the result ofexcitation by the burst itself of gas within its host galaxy,confirming the association (e.g. Vreeswijk et al. 2006).A particular example that highlighted this concern wasGRB 071003, in which it was found that the highest red-shift system, a detection of Mg ii , presumed to be fromthe host, was notably weaker than some intervening ab-sorbers (Perley et al. 2008). Similarly, GRB 060605 exhib-ited weak Ly α from the host, but stronger ( log ( N H i / cm − ) = . ) from an intervening system at slightly lower redshift(Ferrero et al. 2009).Another pertinent case is GRB 141026A, the afterglowof which was observed by GTC, with a spectrum cover-ing wavelength range 5100–9800 ˚A. The S/N was ratherpoor, but an absorption line was seen close to the blueend of the spectrum that if interpreted as Ly α would im-ply z = . and a low column-density of log ( N H i / cm − ) . (de Ugarte Postigo et al. 2014b). In this instance there wereno other features seen to confirm the line identification, andno evidence of a decrement that could be ascribed to the Ly α forest, for which reasons we chose not to include this burstin our sample. Thus, this example illustrates that misidenti-fication of redshift might in some circumstances result in abias in the opposite direction, namely toward lower column-density.Once again, we conclude that whilst it is hard to rule outcompletely, the rate of strong intervening absorbers beingfalsely identified as host systems, providing good spectraare obtained, must be very low. Some afterglow spectra are sufficient to provide redshifts,but the signal-to-noise, at least around the Ly α region, ispoor. This may lead to N H i being undetermined, particu-larly if it is low, thus creating a bias in favour of includinghigher column-density systems. Amongst our sample, onlyeleven bursts lack clear metal line detections, and of thesefive have tentative metal line detections (GRBs 020124,060927, 080129, 080913, 121229A), and five have no metalline detections but do show an unambiguous continuumbreak at Ly α that is sufficiently well defined to constrainthe wing profile (GRBs 060522, 081203, 090519, 100316A,130427B, 140515A). The latter subset all have low S/N, andthe search for metal lines was complicated by low spectralresolution and/or being in a difficult region of the spectrum,but reassuringly they span a wide range of N H i values, whichis not suggestive of any particular bias. This gives confidence that our sample derives predominantly from high-S/N spec-tra, and contains few bursts which are only included becausethey had a particularly high value of H i column-density.Several other bursts have a redshift determined from theLy α break, but the S/N proved insufficient to estimate the N H i value. These cases are few in number: apart from severalat z & , from a search of spectra we have ourselves andthe literature we have only identified GRBs 071025 ( z ≈ ;Section 4.2.1), 140428A ( z ≈ . ; Perley 2014) and 160327A( z ≈ ; de Ugarte Postigo et al. 2016). Thus we believe thatthese cases, while they may be below the median N H i forall bursts, are not likely to be unusually low- N H i . A smallbias could partially offset the bias against dusty sight-linesdiscussed previously.It is notable that redshifts can be obtained from low-S/N spectra when the redshift is comparatively high, whichcan be understood because the strength of the Ly α -breakincreases with redshift. At redshifts below z ∼ such spec-tra likely will not yield secure redshifts, a category that isdiscussed in the next section. We have argued in the preceding sections that bursts areunlikely to have been lost from our sample due to weak ab-sorption lines providing that good spectra were obtained.However, bursts with very faint optical afterglows will beunder-represented due to the increased difficulty of arcsec-ond localisation and redshift determination (either becausespectroscopy was not attempted, or because spectra had toolow S/N to give a conclusive redshift or N H i measure). Thus,if bursts occurring in low density environments had weakerlines and also on average fainter afterglows, then that po-tentially may lead us to systematically lose bursts with high f esc . One way a GRB progenitor could find itself in a lowerdensity environment, would be if it was formed by a so-called “runaway” star. We consider this particular issue inSection 4.4, but here focus on the potential effect of lowdensity on the brightness of afterglows and the likelihoodthat such systems have been missed.As discussed above, the majority of optically faint after-glows are dust extinguished, and have high EUV opacities,while a smaller number are high redshift optical drop-outs.We should also remember that some afterglows were faintwhen observed simply due to the delay in acquiring spec-troscopy. This suggests that the fraction of systems thatare faint due to low density circumburst media is low. Onthe other hand, basic synchrotron afterglow theory providessome motivation for thinking such a trend might occur. Inparticular, for a relativistic jet shocking a medium of uni-form density, n , in typical circumstances the optical after-glow flux should scale with n / (Granot & Sari 2002). Infact, with sufficiently good wide-band monitoring of the af-terglow, the ambient density of the medium in which the jetis travelling (i.e. sub-pc scales) can be calculated. The rangeof circumburst medium densities inferred from such mod-elling is quite wide, from − to cm − (e.g. Laskar et al.2014), but equally is subject to model assumptions and largeuncertainties in many cases.However, the crucial point is that the density struc- MNRAS000 , 1–31 (2018) N. R. Tanvir et al. ture of the immediate circumburst environment is likelydetermined by the recent mass loss history of the pro-genitor system, and potentially that of any companions(van Marle et al. 2006, 2008). This is very unlikely to becorrelated with the density of neutral gas producing theLy α absorption, which is generally situated at significantdistances of at least tens and often hundreds of pc from theburst site (e.g. Prochaska et al. 2006; Vreeswijk et al. 2013).From an empirical point of view, there are few indi-cations of any correlations of afterglow intrinsic luminositywith other host properties, including column-density. For ex-ample, de Ugarte Postigo et al. (2012) found no evidence ofa correlation of afterglow luminosity at the time of observa-tion with the observed line strength in a sample of 69 bursts.In fact, some low column-density systems actually have no-tably bright optical afterglows, including, as mentioned inSection 3.1.2, GRB 060607A, which has the lowest value of N H i in our sample. Other low column-density GRBs withbright afterglows (given the time post-burst they were firstobserved) were GRBs 070125 ( V = . at 13 hr post-burstUpdike et al. 2008), 071003 ( R = . at 42 s post-burstPerley et al. 2008) and 140928A ( r = . at 22 hr post-burstVarela et al. 2014), all discussed above. To some extent thiscould be regarded as a selection effect, since weak lines canonly be detected, or even searched for, in high-S/N spectra.It is also the case that one expects GRBs with highly lumi-nous optical flashes to be more effective at ionizing gas tolarger distances (Section 4.1.2). However, this does at leastindicate a large scatter in any putative correlation of intrin-sic afterglow luminosity and Ly α strength, and combinedwith the comparative dearth of featureless afterglow spectra(Section 4.2.2), leads us to conclude that any such bias mustbe small. An essential assumption in our analysis is that GRBs aregood tracers of the locations of populations of massive starslikely to be responsible for the bulk of EUV radiation pro-duction. In this section we discuss the extent to which this istrue, and consider the potential implications for our results.
GRBs preferentially occur in low (sub-solar) metallic-ity environments (Kr¨uhler et al. 2015; Japelj et al. 2016;Perley et al. 2016b; Graham & Fruchter 2017; Vergani et al.2017), which are typically (but not solely) in less dusty andsmaller galaxies (e.g. Schulze et al. 2015; Blanchard et al.2016), and therefore might be expected to have lower neutralgas column-densities (although see e.g. Gnedin et al. 2008;Sharma et al. 2016, for counter arguments). Lower metal-licity populations also produce more EUV for a given starformation rate (Stanway et al. 2016). These factors may re-sult in an over-estimate of the escape fraction averaged overall galaxies at lower redshifts, but at higher redshifts (above z ∼ ) we would expect low metallicity to be the dominantmode of star formation, and for it to be increasingly occur-ring in small galaxies (we return to this issue in Section 4.5). GRB positions are well correlated spatially with regions ofhigh mid- and near-UV emission in their hosts, and specif-ically more highly correlated than are most (type II) core-collapse supernovae (Fruchter et al. 2006; Svensson et al.2010; Blanchard et al. 2016). This has previously been usedto argue that if the GRB progenitor is a single star it islikely to have initial mass M > –25 M ⊙ (Larsson et al.2007; Anderson et al. 2012), but in any case, whether sin-gle or binary, it suggests average lifetimes less than those ofmore common core-collapse supernovae.We can investigate this question more quantitativelyusing stellar population synthesis models. Specifically, weconsider the BPASS models of Eldridge & Stanway (2009).These include prescriptions for the contribution of binarystars, which is essential given the importance of binary inter-actions to the evolution of massive stars (Sana et al. 2012).In fact binaries both enhance the total EUV output for agiven stellar mass and extend the emission in time. Thiswill increase the total number of ionizing photons producedand also the effectiveness of feedback, tipping the balanceof ionization versus recombination in the environs of newlyformed stars and allowing a greater period for the gas to becleared. The net result is an increase of up to factors of sev-eral in the predicted escape fraction (Stanway et al. 2016;Ma et al. 2016).In Figure 5 we show the cumulative EUV production fora single burst population as a function of time for a range ofmetallicities. A single burst of star formation represents oneextreme: if, as is quite plausible, star formation is more con-tinuous in a region of a galaxy (c.f. Ochsendorf et al. 2017),then GRBs will also be spread over time and their locationswill naturally be more representative of the EUV productionsites. The figure shows that the large bulk ( & %) of suchradiation is still produced in the first 10 Myr, in other wordsduring the lifetimes of all but the most massive stars. Com-paratively little EUV is produced after an age of ∼ Myr(cf. Ma et al. 2016), so even if older stars are less deeplyembedded in their nascent gas clouds (e.g. through havingmoved from their birth sites and/or there being more timefor stellar feedback, and in particular the accumulated ac-tion of supernovae, to carve low density channels throughthe neutral gas in their vicinity) they can contribute ratherlittle to the total ionizing radiation output. A caveat is thatbinary interaction, particularly at low metallicity, may resultin envelope stripping of rather less massive stars ( ∼ M ⊙ ),which may then emit ionizing radiation over a longer periodof time ( ∼ Myr), so could make a significant contributionto the total EUV output that has generally been neglectedto-date (G¨otberg et al. 2017).Although the nature of GRB progenitors remains un-certain, most scenarios suggest they are indeed likely tohave lifetimes of order 5–10 Myr. For example, this is in-dicated by their correlation with regions of high UV emis-sion (Larsson et al. 2007), and is roughly the range spannedby the viable single star chemically homogeneous evolutionmodels studied by Yoon et al. (2006). To place this in con-text, the relative numbers of type Ic supernovae in logarith-mic bins are also shown on Figure 5, and are predominantlyin the range of 3 Myr to 15 Myr. Since the supernovae ac-
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Figure 5.
Output of EUV radiation for a single burst ofstar formation as a function of age from the BPASS mod-els, which include prescriptions for binary stellar evolution(Eldridge & Stanway 2009). Shown are three models, . Z ⊙ (red), . Z ⊙ (black), Z ⊙ (blue), for a stellar initial mass function ofpower-law slope α = − . between 0.1 and 0.5 M ⊙ and slope α = − . from 0.5 to 100 M ⊙ . Dashed lines are the equivalentmodels with an IMF going up to 300 M ⊙ . Also shown for the . Z ⊙ model are the relative numbers of type Ic supernovae (shadedhistogram), and type II supernovae (dotted histogram) in eachlogarithmic age bin. companying long-duration GRBs are also stripped-envelopeevents (Hjorth & Bloom 2012), and correlate with the UVlight of their hosts in a similar way (Kelly et al. 2008), it isreasonable to suppose that they span a comparable range ofages at explosion. Even if GRB locations are good tracers of the dominantsources of stellar EUV radiation, it might be that theyunder-represent the stars from which the bulk of the es-caping radiation is produced. Here we consider several suchscenarios.
In Section 4.3.2 we argued that GRBs likely do explodeon time-scales relevant for a significant proportion of EUVemission from a single age stellar population. However, ifthe peak episodes of EUV escape generally occur before thefirst GRBs explode following a burst of star formation, it isplausible we may not sample the relevant periods (again, forstar formation of a more continuous nature, this will not bea concern).Modelling in detail the escape of ionizing radiation fromyoung massive stars in a range of realistic scenarios is highlychallenging. On the scales of individual clouds outward pres-sure created by winds, radiation and supernovae competeswith gravitational infall, while ionization competes with re-combination. These clouds also vary in size and shape, andhave complex turbulent internal structure and magneticfields. It is also essential to incorporate the processes of star formation and evolution, which introduces further uncer-tainties. Finally the star forming regions exist in larger scalegalactic environments. In order to escape a galaxy, radiationmust first escape its local environment, likely the molecu-lar cloud in which the stars formed, and subsequently leakout through the larger scale neutral gas distribution. Herewe consider the lessons from recent state-of-the-art modelswhich focus on different aspects of the problem.Howard et al. (2018) examined the escape of EUV radi-ation from a range of massive ( – M ⊙ ) giant molecularclouds (GMCs) containing multiple massive star clusters,based on 3D simulations with ongoing star formation. Theyfound the EUV escape from the clouds themselves to be vari-able in time, with occasional peaks above 10% from 2 Myrafter the onset of star formation. The lower mass clouds tendto achieve high escape fractions of 20–100% by ∼ Myr, dueto near complete ionization of the clouds. This is within thetime-frame that some GRBs likely occur, even if the bulkof progenitors have longer lifetimes (Section 4.3.2). The in-termittency here is partly the result of small-scale densitystructure due to the turbulent nature of the cloud, whichproduces time-changing local density field around the clus-ters. This results in a very anisotropic directional distribu-tion of escaping radiation (something also seen in 3D modelsof small molecular clouds dominated by a single O star, byWalch et al. 2012). Placing these clouds into a galactic con-text produces galaxies with rapid (10–20 Myr) fluctuationsin SFR and f esc , particularly in dwarf galaxies, with a gen-eral trend of stronger episodes of star formation being as-sociated with lower f esc . However, these simulations did notinclude the effects of winds or supernovae, and also use sin-gle star stellar population prescriptions. As noted previously,inclusion of binaries is likely to increase the effectiveness offeedback, and extend the time-scales.Rahner et al. (2017) performed 1D spherically symmet-ric calculations, including winds, radiative transfer and theeffects of supernovae, covering a range of cloud masses, den-sities, star-formation efficiencies and metallicities. Here ab-sorption is dominated by neutral gas in the swept up shellof material surrounding the central low density ionized bub-ble. In those models that show any appreciable EUV escapefrom the birth cloud at all, they also generally find a highproportion occurs during the first 2–6 Myr. Of course, thesecalculations are not able to include anisotropies in the shellstructure, which may be key to understanding the escapefraction and, again, presumably star formation and/or EUVproduction more extended in time would modify these re-sults.Simulations that place star formation in a more cosmo-logical context, while reliant on less sophisticated prescrip-tions for the feedback physics, have tended to find episodesof high escape fraction, at least in early galaxies, to havescale times of 10-20 Myr (e.g. Wise et al. 2014; Kimm & Cen2014; Ma et al. 2016; Trebitsch et al. 2017). This reflectstimescales of star formation activity and consequent super-nova feedback which has the dominant effect on the galacticscale gas distribution. Interestingly, Toy et al. (2016) sug-gested that GRB hosts likely have had episodic star for-mation, based on comparison of enrichment timescales withobserved metallicities.For comparison, star formation in the 30 Doradus H ii region (the Tarantula Nebula) in the Large Magellanic MNRAS , 1–31 (2018) N. R. Tanvir et al.
Cloud, often regarded as a local prototype of low-metallicitystar-forming regions that may have been highly abundantin the early universe, has been occurring in different clumpsand clusters for at least ∼ Myr (Sabbi et al. 2016). In-ferring the EUV escape fraction from the Tarantula Nebularegion is subject to large uncertainties, but a recent detailedstudy of its massive star population constrained it to be inthe range 0–0.6, with a preferred value of 0.06 (Doran et al.2013).Clearly this is a field where much work remains to bedone. It is plausible that in some circumstances, a burst ofstar formation in a molecular cloud towards the edge of itsgalaxy may lead to a brief period of high EUV escape tothe IGM before the first supernovae explode. However, itdoes not seem likely this could be a common occurrence,and that to produce high average escape fractions wouldrequire the combined feedback effects of radiation, windsand supernovae to disperse and ionize both local and globalgas, on timescales comparable to GRB progenitor lifetimes.
It has been suggested that GRBs may favour not onlylow metallicity, but possibly also high density sites (e.g.Kelly et al. 2014; Perley et al. 2015), for example dueto dynamical processes in young dense stellar clustersbeing important in the formation of their progenitors(van den Heuvel & Portegies Zwart 2013).However, even if this is true, it is not obvious thatit would significantly affect our conclusions. Very massiveand dense clouds are likely to recollapse without dispersal(Rahner et al. 2017), and would also be much less affectedby the GRB event itself, so GRBs forming preferentially insuch environments seems to contradict the observation thatonly a minority of bursts are heavily dust obscured, and thatabsorption often is predominantly at large distances. Casesof massive GMCs where feedback does drive a strong outflowmight in fact provide the best chances of creating windowsof low density ionized gas to the IGM, thus favouring lowcolumn-density systems. In any event, it remains the casethat GRBs occur in a range of environments, based on theirgalactic locations and the evidence we have of the local den-sity, which all suggests little bias compared to the stars weexpect to dominate the escaping EUV radiation.
A sizeable fraction ( ∼ %) of OB stars in the Milky Wayare found to have sufficiently high space velocities (several10s of km s − ), presumably as a result of dynamical interac-tions, that they will end their lives well outside their nascentbirth clouds (e.g. Tetzlaff et al. 2011). Such runaway starsmay sometimes spend much of their lives in relatively lowdensity regions, and so could have a higher h f esc i than starsthat remain close to their birth sites. If for some reason,such as a requirement to be a binary system, GRB progeni-tors were less likely to be runaways, then, on the face of it,sight-lines to GRBs would not sample that population. Onthe other hand, since GRBs themselves ionize gas in their locality to significant distance (see Section 4.1.2), and giventhat including runaways in hydrodynamic simulations onlyresults in a modest increase of h f esc i of ∼ % (Kimm & Cen2014), missing runaways are unlikely to have a large effecton our conclusions. We can ask whether the special nature of the GRB pro-genitor may influence the column-densities we measure. Inparticular, to allow a GRB jet to reach highly relativisticvelocities it is thought that their progenitors must haveno extended envelope. Indeed the lack of hydrogen andhelium in the spectra of supernovae accompanying GRBsconfirms this picture. Expelling the envelope without alsolosing significant angular momentum is a potential prob-lem for the collapsar scenario for GRB production (e.g.Detmers et al. 2008). One possibility is that high rotationcould lead to chemically homogeneous evolution, essentiallyconsuming the envelope (Yoon & Langer 2005). Alterna-tively, the hydrogen and helium layers might be lost, forexample through explosive common-envelope ejection in atight binary (Podsiadlowski et al. 2010). In such cases it isplausible that the expelled material, which could amount to ∼ M ⊙ , might provide enhanced absorption if it remainsclose enough to provide a significant column-density, but farenough not to be ionized by the ambient UV radiation fieldprior to the burst. Thus, even though large column-densitiesgenerally seem to be produced by gas at relatively large dis-tance from the GRB site, it could be that a modest contribu-tion from gas at 10–20 pc expelled by the progenitor sets aneffective floor to the distribution of log ( N H i / cm − ) ∼ . –18in the GRB sample. Other massive stars that did not pro-duce such high mass loss would therefore have higher escapefractions.However, this ignores the ionizing flux of the opticalflash and early afterglow of the GRBs themselves. In casessuch as the two lowest column-density systems in our sam-ple, GRBs 050908 and 060607A (Section 3.1), the bright af-terglows exceeded the flux required to ionize this mass oflocal gas by orders of magnitude. Only if the peak UV lu-minosity of the burst was at least as faint as M AB ≈ − could a proportion of the expelled gas remain neutral, andeven at z ≈ this translates to a peak apparent magni-tude of m V ≈ , which is effectively unfeasible for follow-upspectroscopy with reasonable S/N given current technology.Thus, excess absorption from gas expelled by GRB progeni-tors can have no effect on our sample of N H i measurements. As already discussed, GRBs favour low metallicity en-vironments, and in particular appear to have a roughlyconstant efficiency of occurrence up to around 0.3–1 times Solar metallicity, with a rapid drop abovethat threshold (Cucchiara et al. 2015; Perley et al. 2016b;Graham & Fruchter 2017; Vergani et al. 2017). This cer-tainly suggests that GRBs should form during the EoRwhen few galaxies were highly metal enriched. Further-more, known GRB hosts span a wide range in stellar mass
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation (Perley et al. 2016b), including small, low metallicity galax-ies likely representative of the galaxy populations predomi-nant in the EoR. In particular, while three GRB hosts havenow been detected by the Hubble Space Telescope ( HST ) at z ∼ with properties similar to those of Lyman-break galax-ies at the same redshift (McGuire et al. 2016), it remainsthe case that the majority of GRB hosts at z > appearto be undetected to HST limits, consistent with them beingdrawn from a galaxy luminosity function in which the faintend dominates (Tanvir et al. 2012a; Trenti et al. 2015).Further evidence that GRBs effectively sample stel-lar populations predominant in the EoR is that therate of GRBs relative to total star formation appears torise with redshift even faster than expected due to themetallicity sensitivity already discussed (Kistler et al. 2009;Robertson & Ellis 2012; Perley et al. 2016b), although num-ber statistics are small at the highest redshifts.Thus to the extent to which analogues of EoR galax-ies exist at lower redshifts, we would expect them to ap-pear in the GRB host samples, and to be included in ourcolumn-density sample. As discussed in Section 3.3, thehigh-resolution cosmological simulations of galaxies at z ∼ of Cen & Kimm (2014) find that roughly half of GRBsshould have log ( N H i / cm − ) < , which we certainly do notsee at lower redshifts. If the prediction is correct – and lim-ited statistics rule out a strong test at this stage – then itwould require that there is a large sub-population of high- f esc hosts at z > that barely exist in the z < host sample. We have compiled a large sample of H i column-density mea-surements obtained from GRB afterglow spectroscopy. Be-cause GRBs select hosts over a very broad range of luminos-ity, we consider that they provide a good representation ofthe dominant populations of (not highly dusty) star form-ing galaxies at z & . Despite uncertainties about the exactnature of the GRB progenitor, we also argue that the life-times and locations of GRB progenitors likely make themwell suited to sampling the periods of high EUV produc-tion, and in particular the peak episodes for its escape.Out of 138, only two sight-lines have sufficiently lowcolumn-density to allow any significant EUV to emerge. As-suming this sample is representative of the sight-lines tothe massive stars dominating ionizing radiation production,we conclude an average escape fraction at the Lyman limitof h f esc i ≈ . , with a 98% confidence upper limit of h f esc i ≈ . . This value is in reasonable agreement withthe h f esc i ∼ . predicted for GRB sight-lines based on thehydrodynamic simulations of Pontzen et al. (2010). It sug-gests that only in rare cases does stellar feedback punctureholes out of the dense ISM, providing clear windows to theIGM. This is a more stringent limit than was obtained fromthe recent direct search for escaping EUV at z ≈ . byGrazian et al. (2017), who found h f esc i < . (67% confi-dence) for galaxies brighter than L ∗ and h f esc i . . for asample brighter than . L ∗ .If we account for the additional opacity due to dust forthe two sight-lines for which there is any appreciable escapefraction, and the loss of highly dusty bursts from the samplealtogether, then these numbers could reduce by factors of – . If radiation from the GRBs themselves ionized some gas inthese two high- f esc cases, then the estimate of h f esc i shouldbe reduced further.On the other side of the balance sheet, the difficultyof finding low column-density systems in spectra with poorsignal-to-noise does provide a modest selection effect in theopposite direction. However, inspection of more completesub-samples suggests that even missing one or two caseswould be surprising, so the net effect is unlikely to be morethan a factor ∼ increase in our estimate of h f esc i .The bulk of the events are at redshifts z = –5, which,even accounting for these systematic uncertainties, indicatesthat stellar EUV falls short of providing the radiation fieldneeded to explain the properties of the Ly α forest in thisredshift range (Becker & Bolton 2013; Stanway et al. 2016).More crucially, if their properties apply to galaxies duringthe EoR, these limits on h f esc i are at least an order of mag-nitude below what is required to maintain an ionized IGMin that era (Robertson et al. 2015; Stanway et al. 2016).There is a weak suggestion of a decline in average H i column-density at the highest redshifts ( z & ), but statis-tics remain too poor for definitive conclusions, and none ofthe z > sight-lines have an appreciable escape fraction.Furthermore, there is no indication of a strong correlationof neutral hydrogen column-density with either galaxy UVluminosity or stellar mass, for the subsets of our sample forwhich these measures are available. Thus we find no evidenceto support the suggestion that ionising escape fraction maybe much higher for the small galaxies that likely dominatedstar formation in the EoR.Avoiding this conclusion would seem to require eitherthat GRBs are not good tracers of the primary sources ofescaping EUV, for instance because the large majority oc-curs before GRBs explode or because there are classes ofolder and less-massive stars that produce substantially moreEUV than has hitherto been appreciated (e.g. as a result ofbinary interactions; G¨otberg et al. 2017), or that there isvery marked evolution in the properties of GRB sight-linesbetween z ∼ and z ∼ –5.Overall this work shows the power of GRBs to addressthe difficult question of the Lyman continuum escape frac-tion averaged over the dominant populations of high-redshiftstar forming galaxies, and demonstrates the benefits of long-term campaigns to obtain GRB afterglow spectroscopy. Theevidence that GRBs occur preferentially in low metallicitysystems, and that their rate relative to the star formationrate increases with redshift, all suggest that they are likelygood tracers of star formation in the EoR. Future samplesof larger numbers of z > GRBs with good H i column-density determinations may benefit soon from the availabil-ity of JWST , and in the 2020s through the
SVOM satelliteworking in conjunction with follow-up spectroscopy on 30-mclass ground-based telescopes. Hence we may hope to obtainmuch tighter and more direct constraints on the contributionof stars to reionization (Yuan et al. 2016).
ACKNOWLEDGEMENTS
This work has benefitted significantly from the leading contri-butions to our field over many years of three sadly departedcolleagues: Neil Gehrels, Javier Gorosabel and Peter Curran.
MNRAS000
MNRAS000 , 1–31 (2018) N. R. Tanvir et al.
The authors acknowledge useful discussions with AviLoeb, Andrew Pontzen, Martin Haehnelt and Alex de Koter.Partly based on observations made with ESO Tele-scopes at the La Silla Paranal Observatory under pro-gramme IDs 280.D-5059, 081.A-0856, 082.A-0301, 083.A-0644, 091.A-0442, 100.D-0648.Partly based on observations made with the Nordic Op-tical Telescope, operated by the Nordic Optical TelescopeScientific Association at the Observatorio del Roque de losMuchachos, La Palma, Spain, of the Instituto de Astrof´ısicade Canarias, under programs 31-014, 32-010, 39-023, 48-005and 51-504.Partly based on observations made with the Gran Tele-scopio Canarias (GTC), installed in the Spanish Observa-torio del Roque de los Muchachos of the Instituto de As-trofˇSsica de Canarias, in the island of La Palma, Spain.Partly based on observations made with the ItalianTelescopio Nazionale Galileo (TNG) operated on the is-land of La Palma by the Fundaci´on Galileo Galilei of theINAF (Istituto Nazionale di Astrofisica) at the Spanish Ob-servatorio del Roque de los Muchachos of the Instituto deAstrof´ısica de Canarias, under programs A26TAC 63 andA32TAC 5.The WHT and its override programme (for 090715B:programme W09AN001, P.I. Curran; for 161017A: pro-gramme W/2017A/23, P.I. Levan) are operated on the is-land of La Palma by the Isaac Newton Group in the Span-ish Observatorio del Roque de los Muchachos of the Insti-tuto de Astrof´ısica de Canarias. We thank A. Kamble, R.Starling and P. Curran for their help with the 090715B ob-servations, and Marie Hrudkova for executing the 161017Aobservations.Partly based on observations obtained at the GeminiObservatory, which is operated by the Association of Univer-sities for Research in Astronomy, Inc., under a cooperativeagreement with the NSF on behalf of the Gemini partner-ship: the National Science Foundation (United States), theNational Research Council (Canada), CONICYT (Chile),Ministerio de Ciencia, Tecnolog´ıa e Innovaci´on Productiva(Argentina), and Minist´erio da Ciˆencia, Tecnologia e Ino-va¸c˜ao (Brazil).Partially based on data from the GTC Public Archiveat CAB (INTA-CSIC).This work made use of the GRBspec database http://grbspec.iaa.es o University of Leicester, Department of Physics & Astron-omy and Leicester Institute of Space & Earth Observation,University Road, Leicester, LE1 7RH, UK Dark Cosmology Centre, Niels Bohr Institute, Universityof Copenhagen, Juliane Maries Vej 30, 2100 CopenhagenØ, Denmark Instituto de Astrof´ısica de Andaluc´ıa (IAA-CSIC), Glori-eta de la Astronom´ıa s/n, E-18008, Granada, Spain Astronomical Institute Anton Pannekoek, University ofAmsterdam, PO Box 94249, 1090 GE Amsterdam, the
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Netherlands Astrophysics Research Institute, Liverpool John MooresUniversity, IC2, Liverpool Science Park, 146 Brownlow Hill,Liverpool L3 5RF, UK Department of Physics, University of Warwick, Coventry,CV4 7AL, UK Harvard-Smithsonian Center for Astrophysics, 60 GardenStreet, Cambridge, MA 02138, USA NASA’s Goddard Space Flight Center, Greenbelt, MD20771, USA Joint Space-Science Institute, University of Maryland,College Park, MD 20742, USA Astrophysical Institute, Department of Physics andAstronomy, Ohio University, Athens, OH 45701, USA INAF-Osservatorio Astronomico di Brera, Via Bianchi46, 23807 Merate, Italy University of the Virgin Islands, College of Science andMathematics, INAF - Osservatorio Astronomico di Roma, Via Frascati33, I-00040 Monte Porzio Catone (RM), 00078, Italy ASI-Science Data Centre, Via del Politecnico snc, I-00133Rome, Italy APC, Astroparticule et Cosmologie, Universit´e ParisDiderot, CNRS/IN2P3, CEA/Irfu, Observatoire de Paris,Sorbonne Paris Cit´e, 10, Rue Alice Domon et L´eonieDuquet, 75205, Paris Cedex 13, France Centre for Astrophysics and Cosmology, University ofNova Gorica, Vipavska 11c, 5270 Ajdovˇsˇcina, Slovenia Centre for Astrophysics and Cosmology, Science Institute,University of Iceland, Dunhagi 5, 107 Reykjav´ık, Iceland Max-Planck-Institut f¨ur Extraterrestrische Physik,Giessenbachstrasse, 85748 Garching, Germany National Radio Astronomy Observatory, 520 EdgemontRoad, Charlottesville, VA 22903, USA Department ofAstronomy, University of California, 501 Campbell Hall,Berkeley, CA 94720-3411, USA Department of Particle Physics and Astrophysics, Weiz-mann Institute of Science, Rehovot 7610001, Israel Heidelberger Institut f¨ur Theoretische Studien, Schloss-Wolfsbrunnenweg 35, 69118 Heidelberg, Germany GEPI, Observatoire de Paris, PSL Research University,CNRS, Univ. Paris Diderot, Sorbonne Paris Cit´e, PlaceJules Janssen, 92195, Meudon, France CAS Key Laboratory of Space Astronomy and Tech-nology, National Astronomical Observatories, ChineseAcademy of Sciences, Beijing 100012, China
REFERENCES
Anderson J. P., Habergham S. M., James P. A., Hamuy M., 2012,MNRAS, 424, 1372Anderson L., Governato F., Karcher M., Quinn T., Wadsley J.,2017, MNRAS, 468, 4077Becker G. D., Bolton J. S., 2013, MNRAS, 436, 1023Berger E., Penprase B. E., Cenko S. B., Kulkarni S. R., Fox D. B.,Steidel C. C., Reddy N. A., 2006, ApJ, 642, 979Blanchard P. K., Berger E., Fong W.-f., 2016, ApJ, 817, 144Bolmer J., Greiner J., Kr¨uhler T., Schady P., Ledoux C., TanvirN. R., Levan A. J., 2018, A&A, 609, A62Bouchet P., Lequeux J., Maurice E., Prevot L., Prevot-BurnichonM. L., 1985, A&A, 149, 330 Bouwens R. J., et al., 2012, ApJ, 752, L5Camps-Fari˜na A., Zaragoza-Cardiel J., Beckman J. E., Font J.,Vel´azquez P. F., Rodr´ıguez-Gonz´alez A., Rosado M., 2017,MNRAS, 468, 4134Castro-Tirado A. J., et al., 2010, A&A, 517, A61Castro-Tirado A. J., et al., 2014, GRB Coordinates Network,16505Castro-Tirado A. J., et al., 2016, GRB Coordinates Network,19632Cen R., Kimm T., 2014, ApJ, 794, 50Cenko S. B., Berger E., Djorgovski S. G., Mahabal A. A., FoxD. B., 2006, GRB Coordinates Network, 5155Cenko S. B., et al., 2008, ApJ, 677, 441Cenko S. B., Perley D. A., Morgan A. N., Klein C. R., Bloom J. S.,Butler N. R., Cobb B. E., 2010, GRB Coordinates Network,10752Chary R., Berger E., Cowie L., 2007, ApJ, 671, 272Chen H.-W., Prochaska J. X., Gnedin N. Y., 2007, ApJ, 667, L125Chen H.-W., et al., 2009, ApJ, 691, 152Chornock R., Perley D. A., Cenko S. B., Bloom J. S., Cobb B.,Prochaska J. X., 2009a, GRB Coordinates Network, 8994Chornock R., Perley D. A., Cobb B. E., 2009b, GRB CoordinatesNetwork, 10100Chornock R., Cucchiara A., Fox D., Berger E., 2010, GRB Coor-dinates Network, 10466Chornock R., Berger E., Fox D. B., Fong W., Laskar T., RothK. C., 2014, preprint, ( arXiv:1405.7400 )Christensen L., Fynbo J. P. U., Prochaska J. X., Th¨one C. C., deUgarte Postigo A., Jakobsson P., 2011, ApJ, 727, 73Ciardi B., Bolton J. S., Maselli A., Graziani L., 2012, MNRAS,423, 558Covino S., et al., 2013, MNRAS, 432, 1231Cucchiara A., 2010, GRB Coordinates Network, Circular Service,No. 10478, 000
Anderson J. P., Habergham S. M., James P. A., Hamuy M., 2012,MNRAS, 424, 1372Anderson L., Governato F., Karcher M., Quinn T., Wadsley J.,2017, MNRAS, 468, 4077Becker G. D., Bolton J. S., 2013, MNRAS, 436, 1023Berger E., Penprase B. E., Cenko S. B., Kulkarni S. R., Fox D. B.,Steidel C. C., Reddy N. A., 2006, ApJ, 642, 979Blanchard P. K., Berger E., Fong W.-f., 2016, ApJ, 817, 144Bolmer J., Greiner J., Kr¨uhler T., Schady P., Ledoux C., TanvirN. R., Levan A. J., 2018, A&A, 609, A62Bouchet P., Lequeux J., Maurice E., Prevot L., Prevot-BurnichonM. L., 1985, A&A, 149, 330 Bouwens R. J., et al., 2012, ApJ, 752, L5Camps-Fari˜na A., Zaragoza-Cardiel J., Beckman J. E., Font J.,Vel´azquez P. F., Rodr´ıguez-Gonz´alez A., Rosado M., 2017,MNRAS, 468, 4134Castro-Tirado A. J., et al., 2010, A&A, 517, A61Castro-Tirado A. J., et al., 2014, GRB Coordinates Network,16505Castro-Tirado A. J., et al., 2016, GRB Coordinates Network,19632Cen R., Kimm T., 2014, ApJ, 794, 50Cenko S. B., Berger E., Djorgovski S. G., Mahabal A. A., FoxD. B., 2006, GRB Coordinates Network, 5155Cenko S. B., et al., 2008, ApJ, 677, 441Cenko S. B., Perley D. A., Morgan A. N., Klein C. R., Bloom J. S.,Butler N. R., Cobb B. E., 2010, GRB Coordinates Network,10752Chary R., Berger E., Cowie L., 2007, ApJ, 671, 272Chen H.-W., Prochaska J. X., Gnedin N. Y., 2007, ApJ, 667, L125Chen H.-W., et al., 2009, ApJ, 691, 152Chornock R., Perley D. A., Cenko S. B., Bloom J. S., Cobb B.,Prochaska J. X., 2009a, GRB Coordinates Network, 8994Chornock R., Perley D. A., Cobb B. E., 2009b, GRB CoordinatesNetwork, 10100Chornock R., Cucchiara A., Fox D., Berger E., 2010, GRB Coor-dinates Network, 10466Chornock R., Berger E., Fox D. B., Fong W., Laskar T., RothK. C., 2014, preprint, ( arXiv:1405.7400 )Christensen L., Fynbo J. P. U., Prochaska J. X., Th¨one C. C., deUgarte Postigo A., Jakobsson P., 2011, ApJ, 727, 73Ciardi B., Bolton J. S., Maselli A., Graziani L., 2012, MNRAS,423, 558Covino S., et al., 2013, MNRAS, 432, 1231Cucchiara A., 2010, GRB Coordinates Network, Circular Service,No. 10478, 000 , 1–31 (2018) N. R. Tanvir et al.
Finkelstein S. L., et al., 2012, ApJ, 758, 93Fiore F., et al., 2005, ApJ, 624, 853Fontanot F., Cristiani S., Pfrommer C., Cupani G., Vanzella E.,2014, MNRAS, 438, 2097Fox A. J., Ledoux C., Vreeswijk P. M., Smette A., Jaunsen A. O.,2008, A&A, 491, 189Fragos T., Lehmer B. D., Naoz S., Zezas A., Basu-Zych A., 2013,ApJ, 776, L31Friis M., et al., 2015, MNRAS, 451, 167Fruchter A. S., et al., 2006, Nature, 441, 463Fynbo J. P. U., et al., 2002, in Gilfanov M., Sunyeav R., Chu-razov E., eds, Lighthouses of the Universe: The Most Lumi-nous Celestial Objects and Their Use for Cosmology. p. 187( arXiv:astro-ph/0110603 ), doi:10.1007/10856495 24Fynbo J. P. U., et al., 2005, ApJ, 633, 317Fynbo J. P. U., et al., 2009, ApJS, 185, 526Gehrels N., Razzaque S., 2013, Frontiers of Physics, 8, 661Gnedin N. Y., Kravtsov A. V., Chen H.-W., 2008, ApJ, 672, 765Gorosabel J., de Ugarte Postigo A., Th¨one C., Perley D., GarciaRodriguez A., 2014, GRB Coordinates Network, 16671G¨otberg Y., de Mink S. E., Groh J. H., 2017, A&A, 608, A11Graham J. F., Fruchter A. S., 2017, ApJ, 834, 170Granot J., Sari R., 2002, ApJ, 568, 820Grazian A., et al., 2017, A&A, 602, A18Greiner J., et al., 2009a, ApJ, 693, 1610Greiner J., et al., 2009b, ApJ, 693, 1912Greiner J., et al., 2011, A&A, 526, A30Greiner J., et al., 2015a, Nature, 523, 189Greiner J., et al., 2015b, ApJ, 809, 76Gunn J. E., Peterson B. A., 1965, ApJ, 142, 1633Hansen S. H., Haiman Z., 2004, ApJ, 600, 26Hasco¨et R., Uhm Z. L., Mochkovitch R., Daigne F., 2011, A&A,534, A104Hassan S., Dav´e R., Mitra S., Finlator K., Ciardi B., Santos M. G.,2018, MNRAS, 473, 227Hjorth J., Bloom J. S., 2012, in Chapter 9 in “Gamma-RayBursts”, Cambridge Astrophysics Series 51, eds. C. Kouve-liotou, R. A. M. J. Wijers and S. Woosley, Cambridge Uni-versity Press (Cambridge). pp 169–190Hjorth J., et al., 2003a, Nature, 423, 847Hjorth J., et al., 2003b, ApJ, 597, 699Howard C. S., Pudritz R. E., Harris W. E., Klessen R. S., 2018,MNRAS, 475, 3121Hunt L. K., et al., 2014, A&A, 565, A112Izzo L., et al., 2018, GRB Coordinates Network, Circular Service,No. 22567, arXiv:1706.00601 )Kelly P. L., Kirshner R. P., Pahre M., 2008, ApJ, 687, 1201Kelly P. L., Filippenko A. V., Modjaz M., Kocevski D., 2014,ApJ, 789, 23Khaire V., Srianand R., Choudhury T. R., Gaikwad P., 2016,MNRAS, 457, 4051Kimm T., Cen R., 2014, ApJ, 788, 121Kistler M. D., Y¨uksel H., Beacom J. F., Hopkins A. M., WyitheJ. S. B., 2009, ApJ, 705, L104Knevitt G., Wynn G. A., Power C., Bolton J. S., 2014, MNRAS,445, 2034Krongold Y., Prochaska J. X., 2013, ApJ, 774, 115Kr¨uhler T., et al., 2015, A&A, 581, A125 Kuhlen M., Faucher-Gigu`ere C.-A., 2012, MNRAS, 423, 862Kuin N. P. M., et al., 2009, MNRAS, 395, L21Larsson J., Levan A. J., Davies M. B., Fruchter A. S., 2007,MNRAS, 376, 1285Laskar T., Berger E., Chary R.-R., 2011, ApJ, 739, 1Laskar T., et al., 2014, ApJ, 781, 1Levan A. J., et al., 2014, ApJ, 781, 13Levesque E. M., et al., 2010a, MNRAS, 401, 963Levesque E. M., Soderberg A. M., Kewley L. J., Berger E., 2010b,ApJ, 725, 1337Lockman F. J., Jahoda K., McCammon D., 1986, ApJ, 302, 432Lyman J. D., et al., 2017, MNRAS, 467, 1795Ma X., Hopkins P. F., Kasen D., Quataert E., Faucher-Gigu`ereC.-A., Kereˇs D., Murray N., Strom A., 2016, MNRAS,459, 3614Madau P., Dickinson M., 2014, ARA&A, 52, 415Madau P., Fragos T., 2017, ApJ, 840, 39Madau P., Haardt F., 2015, ApJ, 813, L8Malesani D., Fynbo J. P. U., D’Elia V., de Ugarte Postigo A.,Jakobsson P., Thoene C. C., 2009, GRB Coordinates Network,9457Malesani D., et al., 2014, GRB Coordinates Network, 15800Marchi F., et al., 2017, A&A, 601, A73McGuire J. T. W., et al., 2016, ApJ, 825, 135Melandri A., et al., 2008, ApJ, 686, 1209Melandri A., et al., 2015, A&A, 581, A86Mirabal N., Melandri A., Halpern J. P., 2007, GRB CoordinatesNetwork, 6162Mirabel I. F., Dijkstra M., Laurent P., Loeb A., Pritchard J. R.,2011, A&A, 528, A149Molinari E., et al., 2007, A&A, 469, L13Møller P., et al., 2002, A&A, 396, L21Morgan A. N., et al., 2014, MNRAS, 440, 1810Mostardi R. E., Shapley A. E., Nestor D. B., Steidel C. C., ReddyN. A., Trainor R. F., 2013, ApJ, 779, 65Nestor D. B., Shapley A. E., Kornei K. A., Steidel C. C., SianaB., 2013, ApJ, 765, 47Nicuesa Guelbenzu A., et al., 2011, A&A, 531, L6Nysewander M., Reichart D. E., Crain J. A., Foster A., HaislipJ., Ivarsen K., Lacluyze A., Trotter A., 2009, ApJ, 693, 1417Ochsendorf B. B., Meixner M., Roman-Duval J., Rahman M.,Evans II N. J., 2017, ApJ, 841, 109Oey M. S., Clarke C. J., 1997, MNRAS, 289Ouchi M., et al., 2009, ApJ, 706, 1136Page K. L., et al., 2009, MNRAS, 400, 134Patel M., Warren S. J., Mortlock D. J., Fynbo J. P. U., 2010,A&A, 512, L3Pellegrini E. W., Oey M. S., Winkler P. F., Points S. D., SmithR. C., Jaskot A. E., Zastrow J., 2012, ApJ, 755, 40Perley D. A., 2011, PhD thesis, University of California, BerkeleyPerley D. A., 2014, GRB Coordinates Network, 16181Perley D. A., et al., 2008, ApJ, 688, 470Perley D. A., et al., 2010, MNRAS, 406, 2473Perley D. A., et al., 2013, ApJ, 778, 128Perley D. A., et al., 2015, ApJ, 801, 102Perley D. A., et al., 2016a, ApJ, 817, 7Perley D. A., et al., 2016b, ApJ, 817, 8Perna R., Lazzati D., 2002, ApJ, 580, 261Planck Collaboration et al., 2016, A&A, 596, A108Podsiadlowski P., Ivanova N., Justham S., Rappaport S., 2010,MNRAS, 406, 840Pontzen A., et al., 2010, MNRAS, 402, 1523Prochaska J. X., Chen H.-W., Bloom J. S., 2006, ApJ, 648, 95Prochaska J. X., Chen H.-W., Dessauges-Zavadsky M., BloomJ. S., 2007, ApJ, 666, 267Prochaska J. X., Perley D., Howard A., Chen H.-W., Marcy G.,Fischer D., Wilburn C., 2008, GRB Coordinates Network,8083 MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Rahner D., Pellegrini E. W., Glover S. C. O., Klessen R. S., 2017,MNRAS, 470, 4453Razoumov A. O., Sommer-Larsen J., 2010, ApJ, 710, 1239Robertson B. E., Ellis R. S., 2012, ApJ, 744, 95Robertson B. E., Ellis R. S., Furlanetto S. R., Dunlop J. S., 2015,ApJ, 802, L19Rol E., et al., 2007, GRB Coordinates Network, 6221Rossi A., et al., 2012, A&A, 545, A77Roy A., Nath B. B., Sharma P., 2015, MNRAS, 451, 1939Rutkowski M. J., et al., 2017, ApJ, 841, L27Sabbi E., et al., 2016, ApJS, 222, 11Salvaterra R., et al., 2009, Nature, 461, 1258Sana H., et al., 2012, Science, 337, 444S´anchez-Ram´ırez R., et al., 2013a, in Revista Mexicana de As-tronomia y Astrofisica Conference Series. pp 113–113Sanchez-Ramirez R., Gorosabel J., Castro-Tirado A. J., Cepa J.,Gomez-Velarde G., 2013b, GRB Coordinates Network, 14685Savaglio S., et al., 2012, MNRAS, 420, 627Schady P., Savaglio S., Kr¨uhler T., Greiner J., Rau A., 2011,A&A, 525, A113Schaerer D., Boone F., Zamojski M., Staguhn J., Dessauges-Zavadsky M., Finkelstein S., Combes F., 2015, A&A, 574, A19Schlafly E. F., Finkbeiner D. P., 2011, ApJ, 737, 103Schulze S., et al., 2015, ApJ, 808, 73Sciama D. W., 1982, MNRAS, 198, 1PSelsing J., et al., 2018, preprint, ( arXiv:1802.07727 )Shapley A. E., Steidel C. C., Pettini M., Adelberger K. L., ErbD. K., 2006, ApJ, 651, 688Shapley A. E., Steidel C. C., Strom A. L., Bogosavljevi´c M.,Reddy N. A., Siana B., Mostardi R. E., Rudie G. C., 2016,ApJ, 826, L24Sharma M., Theuns T., Frenk C., Bower R., Crain R., SchallerM., Schaye J., 2016, MNRAS, 458, L94Shin M.-S., et al., 2006, preprint, ( arXiv:astro-ph/0608327 )Siana B., et al., 2015, ApJ, 804, 17Smette A., Ledoux C., Vreeswijk P., De Cia A., Petitjean P.,Fynbo J., Malesani D., Fox A., 2013, GRB Coordinates Net-work, 14848Stanway E. R., Eldridge J. J., Becker G. D., 2016, MNRAS,456, 485Starling R. L. C., Wijers R. A. M. J., Hughes M. A., TanvirN. R., Vreeswijk P. M., Rol E., Salamanca I., 2005, MNRAS,360, 305Steidel C. C., Pettini M., Adelberger K. L., 2001, ApJ, 546, 665Svensson K. M., Levan A. J., Tanvir N. R., Fruchter A. S., StrolgerL.-G., 2010, MNRAS, 405, 57Tanvir N. R., et al., 2009, Nature, 461, 1254Tanvir N. R., Wiersema K., Levan A. J., Cenko S. B., Geballe T.,2011, GRB Coordinates Network, 12225Tanvir N. R., et al., 2012a, ApJ, 754, 46Tanvir N. R., Levan A. J., Matulonis T., 2012b, GRB CoordinatesNetwork, 14009Tanvir N. R., et al., 2017, preprint, ( arXiv:1703.09052 )Tetzlaff N., Neuh¨auser R., Hohle M. M., 2011, MNRAS, 410, 190Thoene C. C., et al., 2009, GRB Coordinates Network, 9409Th¨one C. C., Perley D. A., Cooke J., Bloom J. S., Chen H.-W.,Barton E., 2007, GRB Coordinates Network, 6741Th¨one C. C., et al., 2011, MNRAS, 414, 479Th¨one C. C., de Ugarte Postigo A., Gorosabel J., Sanchez-Ramirez R., Fynbo J. P. U., Gomez Velarde G., 2012, GRBCoordinates Network, 13628Torii K., 2005, GRB Coordinates Network, 3943Totani T., Kawai N., Kosugi G., Aoki K., Yamada T., Iye M.,Ohta K., Hattori T., 2006, PASJ, 58, 485Toy V. L., et al., 2016, ApJ, 832, 175Trebitsch M., Blaizot J., Rosdahl J., Devriendt J., Slyz A., 2017,MNRAS, 470, 224Trenti M., Perna R., Jimenez R., 2015, ApJ, 802, 103 Updike A. C., et al., 2008, ApJ, 685, 361Vanzella E., et al., 2010, ApJ, 725, 1011Vanzella E., et al., 2012, ApJ, 751, 70Vanzella E., et al., 2015, A&A, 576, A116Vanzella E., et al., 2016, ApJ, 825, 41Varela K., Kann D. A., Klose S., Greiner J., 2014, GRB Coordi-nates Network, 16849Vergani S. D., et al., 2017, A&A, 599, A120Vreeswijk P. M., et al., 2004, A&A, 419, 927Vreeswijk P. M., et al., 2006, A&A, 447, 145Vreeswijk P. M., et al., 2007, A&A, 468, 83Vreeswijk P. M., Fynbo J. P. U., Malesani D., Hjorth J., de UgartePostigo A., 2008, GRB Coordinates Network, 8191Vreeswijk P. M., et al., 2013, A&A, 549, A22Walch S. K., Whitworth A. P., Bisbas T., W¨unsch R., HubberD., 2012, MNRAS, 427, 625Watson D., Jakobsson P., 2012, ApJ, 754, 89Watson D., Hjorth J., Fynbo J. P. U., Jakobsson P., Foley S.,Sollerman J., Wijers R. A. M. J., 2007, ApJ, 660, L101Watson D., et al., 2013, ApJ, 768, 23Waxman E., Draine B. T., 2000, ApJ, 537, 796Wiersema K., Levan A., Kamble A., Tanvir N., Malesani D., 2009,GRB Coordinates Network, 9673Willingale R., Starling R. L. C., Beardmore A. P., Tanvir N. R.,O’Brien P. T., 2013, MNRAS, 431, 394Wise J. H., Demchenko V. G., Halicek M. T., Norman M. L.,Turk M. J., Abel T., Smith B. D., 2014, MNRAS, 442, 2560Wiseman P., Perley D. A., Schady P., Prochaska J. X., de UgartePostigo A., Kr¨uhler T., Yates R. M., Greiner J., 2017, A&A,607, A107Xu D., et al., 2013, ApJ, 776, 98Xu D., Tanvir N. R., Malesani D., Fynbo J. P. U., Jakobsson P.,Saario J., 2015, GRB Coordinates Network, 18696Xu H., Wise J. H., Norman M. L., Ahn K., O’Shea B. W., 2016a,ApJ, 833, 84Xu D., Fynbo J. P. U., Malesani D., de Ugarte Postigo A., Petru-shevska T., Saario J., Telting J., 2016b, GRB CoordinatesNetwork, 19109Yoon S.-C., Langer N., 2005, A&A, 443, 643Yoon S.-C., Langer N., Norman C., 2006, A&A, 460, 199Yuan W., et al., 2016, Space Sci. Rev., 202, 235Zackrisson E., Inoue A. K., Jensen H., 2013, ApJ, 777, 39Zafar T., Watson D. J., Tanvir N. R., Fynbo J. P. U., StarlingR. L. C., Levan A. J., 2011, ApJ, 735, 2de Barros S., et al., 2016, A&A, 585, A51de Ugarte Postigo A., Tomasella L., 2015, GRB Coordinates Net-work, 17710de Ugarte Postigo A., et al., 2012, A&A, 548, A11de Ugarte Postigo A., Thoene C. C., Gorosabel J., Sanchez-Ramirez R., Fynbo J. P. U., Tanvir N. R., Cabrera-LaversA., Garcia A., 2013, GRB Coordinates Network, 15470de Ugarte Postigo A., Blazek M., Janout P., Sprimont P., Th¨oneC. C., Gorosabel J., S´anchez-Ram´ırez R., 2014a, in Soft-ware and Cyberinfrastructure for Astronomy III. p. 91520B,doi:10.1117/12.2055774de Ugarte Postigo A., Th¨one C. C., Tanvir N. R., Gorosabel J.,Fynbo J., Lombardi G., Reverte-Paya D., Perez D., 2014b,GRB Coordinates Network, 16968de Ugarte Postigo A., Tanvir N. R., Cano Z., Izzo L., FynboJ. P. U., Sanchez-Ramirez R., Thoene C. C., Pesev P., 2016,GRB Coordinates Network, 19245de Ugarte Postigo A., Kann D. A., Thoene C., Izzo L., TanvirN. R., Lombardi G., Marante A., 2017a, GRB CoordinatesNetwork, 20990de Ugarte Postigo A., Izzo L., Kann D. A., Thoene C., Cano Z.,Fynbo J. P. U., Garcia Alvarez D., 2017b, GRB CoordinatesNetwork, 21177MNRAS000
Finkelstein S. L., et al., 2012, ApJ, 758, 93Fiore F., et al., 2005, ApJ, 624, 853Fontanot F., Cristiani S., Pfrommer C., Cupani G., Vanzella E.,2014, MNRAS, 438, 2097Fox A. J., Ledoux C., Vreeswijk P. M., Smette A., Jaunsen A. O.,2008, A&A, 491, 189Fragos T., Lehmer B. D., Naoz S., Zezas A., Basu-Zych A., 2013,ApJ, 776, L31Friis M., et al., 2015, MNRAS, 451, 167Fruchter A. S., et al., 2006, Nature, 441, 463Fynbo J. P. U., et al., 2002, in Gilfanov M., Sunyeav R., Chu-razov E., eds, Lighthouses of the Universe: The Most Lumi-nous Celestial Objects and Their Use for Cosmology. p. 187( arXiv:astro-ph/0110603 ), doi:10.1007/10856495 24Fynbo J. P. U., et al., 2005, ApJ, 633, 317Fynbo J. P. U., et al., 2009, ApJS, 185, 526Gehrels N., Razzaque S., 2013, Frontiers of Physics, 8, 661Gnedin N. Y., Kravtsov A. V., Chen H.-W., 2008, ApJ, 672, 765Gorosabel J., de Ugarte Postigo A., Th¨one C., Perley D., GarciaRodriguez A., 2014, GRB Coordinates Network, 16671G¨otberg Y., de Mink S. E., Groh J. H., 2017, A&A, 608, A11Graham J. F., Fruchter A. S., 2017, ApJ, 834, 170Granot J., Sari R., 2002, ApJ, 568, 820Grazian A., et al., 2017, A&A, 602, A18Greiner J., et al., 2009a, ApJ, 693, 1610Greiner J., et al., 2009b, ApJ, 693, 1912Greiner J., et al., 2011, A&A, 526, A30Greiner J., et al., 2015a, Nature, 523, 189Greiner J., et al., 2015b, ApJ, 809, 76Gunn J. E., Peterson B. A., 1965, ApJ, 142, 1633Hansen S. H., Haiman Z., 2004, ApJ, 600, 26Hasco¨et R., Uhm Z. L., Mochkovitch R., Daigne F., 2011, A&A,534, A104Hassan S., Dav´e R., Mitra S., Finlator K., Ciardi B., Santos M. G.,2018, MNRAS, 473, 227Hjorth J., Bloom J. S., 2012, in Chapter 9 in “Gamma-RayBursts”, Cambridge Astrophysics Series 51, eds. C. Kouve-liotou, R. A. M. J. Wijers and S. Woosley, Cambridge Uni-versity Press (Cambridge). pp 169–190Hjorth J., et al., 2003a, Nature, 423, 847Hjorth J., et al., 2003b, ApJ, 597, 699Howard C. S., Pudritz R. E., Harris W. E., Klessen R. S., 2018,MNRAS, 475, 3121Hunt L. K., et al., 2014, A&A, 565, A112Izzo L., et al., 2018, GRB Coordinates Network, Circular Service,No. 22567, arXiv:1706.00601 )Kelly P. L., Kirshner R. P., Pahre M., 2008, ApJ, 687, 1201Kelly P. L., Filippenko A. V., Modjaz M., Kocevski D., 2014,ApJ, 789, 23Khaire V., Srianand R., Choudhury T. R., Gaikwad P., 2016,MNRAS, 457, 4051Kimm T., Cen R., 2014, ApJ, 788, 121Kistler M. D., Y¨uksel H., Beacom J. F., Hopkins A. M., WyitheJ. S. B., 2009, ApJ, 705, L104Knevitt G., Wynn G. A., Power C., Bolton J. S., 2014, MNRAS,445, 2034Krongold Y., Prochaska J. X., 2013, ApJ, 774, 115Kr¨uhler T., et al., 2015, A&A, 581, A125 Kuhlen M., Faucher-Gigu`ere C.-A., 2012, MNRAS, 423, 862Kuin N. P. M., et al., 2009, MNRAS, 395, L21Larsson J., Levan A. J., Davies M. B., Fruchter A. S., 2007,MNRAS, 376, 1285Laskar T., Berger E., Chary R.-R., 2011, ApJ, 739, 1Laskar T., et al., 2014, ApJ, 781, 1Levan A. J., et al., 2014, ApJ, 781, 13Levesque E. M., et al., 2010a, MNRAS, 401, 963Levesque E. M., Soderberg A. M., Kewley L. J., Berger E., 2010b,ApJ, 725, 1337Lockman F. J., Jahoda K., McCammon D., 1986, ApJ, 302, 432Lyman J. D., et al., 2017, MNRAS, 467, 1795Ma X., Hopkins P. F., Kasen D., Quataert E., Faucher-Gigu`ereC.-A., Kereˇs D., Murray N., Strom A., 2016, MNRAS,459, 3614Madau P., Dickinson M., 2014, ARA&A, 52, 415Madau P., Fragos T., 2017, ApJ, 840, 39Madau P., Haardt F., 2015, ApJ, 813, L8Malesani D., Fynbo J. P. U., D’Elia V., de Ugarte Postigo A.,Jakobsson P., Thoene C. C., 2009, GRB Coordinates Network,9457Malesani D., et al., 2014, GRB Coordinates Network, 15800Marchi F., et al., 2017, A&A, 601, A73McGuire J. T. W., et al., 2016, ApJ, 825, 135Melandri A., et al., 2008, ApJ, 686, 1209Melandri A., et al., 2015, A&A, 581, A86Mirabal N., Melandri A., Halpern J. P., 2007, GRB CoordinatesNetwork, 6162Mirabel I. F., Dijkstra M., Laurent P., Loeb A., Pritchard J. R.,2011, A&A, 528, A149Molinari E., et al., 2007, A&A, 469, L13Møller P., et al., 2002, A&A, 396, L21Morgan A. N., et al., 2014, MNRAS, 440, 1810Mostardi R. E., Shapley A. E., Nestor D. B., Steidel C. C., ReddyN. A., Trainor R. F., 2013, ApJ, 779, 65Nestor D. B., Shapley A. E., Kornei K. A., Steidel C. C., SianaB., 2013, ApJ, 765, 47Nicuesa Guelbenzu A., et al., 2011, A&A, 531, L6Nysewander M., Reichart D. E., Crain J. A., Foster A., HaislipJ., Ivarsen K., Lacluyze A., Trotter A., 2009, ApJ, 693, 1417Ochsendorf B. B., Meixner M., Roman-Duval J., Rahman M.,Evans II N. J., 2017, ApJ, 841, 109Oey M. S., Clarke C. J., 1997, MNRAS, 289Ouchi M., et al., 2009, ApJ, 706, 1136Page K. L., et al., 2009, MNRAS, 400, 134Patel M., Warren S. J., Mortlock D. J., Fynbo J. P. U., 2010,A&A, 512, L3Pellegrini E. W., Oey M. S., Winkler P. F., Points S. D., SmithR. C., Jaskot A. E., Zastrow J., 2012, ApJ, 755, 40Perley D. A., 2011, PhD thesis, University of California, BerkeleyPerley D. A., 2014, GRB Coordinates Network, 16181Perley D. A., et al., 2008, ApJ, 688, 470Perley D. A., et al., 2010, MNRAS, 406, 2473Perley D. A., et al., 2013, ApJ, 778, 128Perley D. A., et al., 2015, ApJ, 801, 102Perley D. A., et al., 2016a, ApJ, 817, 7Perley D. A., et al., 2016b, ApJ, 817, 8Perna R., Lazzati D., 2002, ApJ, 580, 261Planck Collaboration et al., 2016, A&A, 596, A108Podsiadlowski P., Ivanova N., Justham S., Rappaport S., 2010,MNRAS, 406, 840Pontzen A., et al., 2010, MNRAS, 402, 1523Prochaska J. X., Chen H.-W., Bloom J. S., 2006, ApJ, 648, 95Prochaska J. X., Chen H.-W., Dessauges-Zavadsky M., BloomJ. S., 2007, ApJ, 666, 267Prochaska J. X., Perley D., Howard A., Chen H.-W., Marcy G.,Fischer D., Wilburn C., 2008, GRB Coordinates Network,8083 MNRAS , 1–31 (2018) he escape fraction of ionizing radiation Rahner D., Pellegrini E. W., Glover S. C. O., Klessen R. S., 2017,MNRAS, 470, 4453Razoumov A. O., Sommer-Larsen J., 2010, ApJ, 710, 1239Robertson B. E., Ellis R. S., 2012, ApJ, 744, 95Robertson B. E., Ellis R. S., Furlanetto S. R., Dunlop J. S., 2015,ApJ, 802, L19Rol E., et al., 2007, GRB Coordinates Network, 6221Rossi A., et al., 2012, A&A, 545, A77Roy A., Nath B. B., Sharma P., 2015, MNRAS, 451, 1939Rutkowski M. J., et al., 2017, ApJ, 841, L27Sabbi E., et al., 2016, ApJS, 222, 11Salvaterra R., et al., 2009, Nature, 461, 1258Sana H., et al., 2012, Science, 337, 444S´anchez-Ram´ırez R., et al., 2013a, in Revista Mexicana de As-tronomia y Astrofisica Conference Series. pp 113–113Sanchez-Ramirez R., Gorosabel J., Castro-Tirado A. J., Cepa J.,Gomez-Velarde G., 2013b, GRB Coordinates Network, 14685Savaglio S., et al., 2012, MNRAS, 420, 627Schady P., Savaglio S., Kr¨uhler T., Greiner J., Rau A., 2011,A&A, 525, A113Schaerer D., Boone F., Zamojski M., Staguhn J., Dessauges-Zavadsky M., Finkelstein S., Combes F., 2015, A&A, 574, A19Schlafly E. F., Finkbeiner D. P., 2011, ApJ, 737, 103Schulze S., et al., 2015, ApJ, 808, 73Sciama D. W., 1982, MNRAS, 198, 1PSelsing J., et al., 2018, preprint, ( arXiv:1802.07727 )Shapley A. E., Steidel C. C., Pettini M., Adelberger K. L., ErbD. K., 2006, ApJ, 651, 688Shapley A. E., Steidel C. C., Strom A. L., Bogosavljevi´c M.,Reddy N. A., Siana B., Mostardi R. E., Rudie G. C., 2016,ApJ, 826, L24Sharma M., Theuns T., Frenk C., Bower R., Crain R., SchallerM., Schaye J., 2016, MNRAS, 458, L94Shin M.-S., et al., 2006, preprint, ( arXiv:astro-ph/0608327 )Siana B., et al., 2015, ApJ, 804, 17Smette A., Ledoux C., Vreeswijk P., De Cia A., Petitjean P.,Fynbo J., Malesani D., Fox A., 2013, GRB Coordinates Net-work, 14848Stanway E. R., Eldridge J. J., Becker G. D., 2016, MNRAS,456, 485Starling R. L. C., Wijers R. A. M. J., Hughes M. A., TanvirN. R., Vreeswijk P. M., Rol E., Salamanca I., 2005, MNRAS,360, 305Steidel C. C., Pettini M., Adelberger K. L., 2001, ApJ, 546, 665Svensson K. M., Levan A. J., Tanvir N. R., Fruchter A. S., StrolgerL.-G., 2010, MNRAS, 405, 57Tanvir N. R., et al., 2009, Nature, 461, 1254Tanvir N. R., Wiersema K., Levan A. J., Cenko S. B., Geballe T.,2011, GRB Coordinates Network, 12225Tanvir N. R., et al., 2012a, ApJ, 754, 46Tanvir N. R., Levan A. J., Matulonis T., 2012b, GRB CoordinatesNetwork, 14009Tanvir N. R., et al., 2017, preprint, ( arXiv:1703.09052 )Tetzlaff N., Neuh¨auser R., Hohle M. M., 2011, MNRAS, 410, 190Thoene C. C., et al., 2009, GRB Coordinates Network, 9409Th¨one C. C., Perley D. A., Cooke J., Bloom J. S., Chen H.-W.,Barton E., 2007, GRB Coordinates Network, 6741Th¨one C. C., et al., 2011, MNRAS, 414, 479Th¨one C. C., de Ugarte Postigo A., Gorosabel J., Sanchez-Ramirez R., Fynbo J. P. U., Gomez Velarde G., 2012, GRBCoordinates Network, 13628Torii K., 2005, GRB Coordinates Network, 3943Totani T., Kawai N., Kosugi G., Aoki K., Yamada T., Iye M.,Ohta K., Hattori T., 2006, PASJ, 58, 485Toy V. L., et al., 2016, ApJ, 832, 175Trebitsch M., Blaizot J., Rosdahl J., Devriendt J., Slyz A., 2017,MNRAS, 470, 224Trenti M., Perna R., Jimenez R., 2015, ApJ, 802, 103 Updike A. C., et al., 2008, ApJ, 685, 361Vanzella E., et al., 2010, ApJ, 725, 1011Vanzella E., et al., 2012, ApJ, 751, 70Vanzella E., et al., 2015, A&A, 576, A116Vanzella E., et al., 2016, ApJ, 825, 41Varela K., Kann D. A., Klose S., Greiner J., 2014, GRB Coordi-nates Network, 16849Vergani S. D., et al., 2017, A&A, 599, A120Vreeswijk P. M., et al., 2004, A&A, 419, 927Vreeswijk P. M., et al., 2006, A&A, 447, 145Vreeswijk P. M., et al., 2007, A&A, 468, 83Vreeswijk P. M., Fynbo J. P. U., Malesani D., Hjorth J., de UgartePostigo A., 2008, GRB Coordinates Network, 8191Vreeswijk P. M., et al., 2013, A&A, 549, A22Walch S. K., Whitworth A. P., Bisbas T., W¨unsch R., HubberD., 2012, MNRAS, 427, 625Watson D., Jakobsson P., 2012, ApJ, 754, 89Watson D., Hjorth J., Fynbo J. P. U., Jakobsson P., Foley S.,Sollerman J., Wijers R. A. M. J., 2007, ApJ, 660, L101Watson D., et al., 2013, ApJ, 768, 23Waxman E., Draine B. T., 2000, ApJ, 537, 796Wiersema K., Levan A., Kamble A., Tanvir N., Malesani D., 2009,GRB Coordinates Network, 9673Willingale R., Starling R. L. C., Beardmore A. P., Tanvir N. R.,O’Brien P. T., 2013, MNRAS, 431, 394Wise J. H., Demchenko V. G., Halicek M. T., Norman M. L.,Turk M. J., Abel T., Smith B. D., 2014, MNRAS, 442, 2560Wiseman P., Perley D. A., Schady P., Prochaska J. X., de UgartePostigo A., Kr¨uhler T., Yates R. M., Greiner J., 2017, A&A,607, A107Xu D., et al., 2013, ApJ, 776, 98Xu D., Tanvir N. R., Malesani D., Fynbo J. P. U., Jakobsson P.,Saario J., 2015, GRB Coordinates Network, 18696Xu H., Wise J. H., Norman M. L., Ahn K., O’Shea B. W., 2016a,ApJ, 833, 84Xu D., Fynbo J. P. U., Malesani D., de Ugarte Postigo A., Petru-shevska T., Saario J., Telting J., 2016b, GRB CoordinatesNetwork, 19109Yoon S.-C., Langer N., 2005, A&A, 443, 643Yoon S.-C., Langer N., Norman C., 2006, A&A, 460, 199Yuan W., et al., 2016, Space Sci. Rev., 202, 235Zackrisson E., Inoue A. K., Jensen H., 2013, ApJ, 777, 39Zafar T., Watson D. J., Tanvir N. R., Fynbo J. P. U., StarlingR. L. C., Levan A. J., 2011, ApJ, 735, 2de Barros S., et al., 2016, A&A, 585, A51de Ugarte Postigo A., Tomasella L., 2015, GRB Coordinates Net-work, 17710de Ugarte Postigo A., et al., 2012, A&A, 548, A11de Ugarte Postigo A., Thoene C. C., Gorosabel J., Sanchez-Ramirez R., Fynbo J. P. U., Tanvir N. R., Cabrera-LaversA., Garcia A., 2013, GRB Coordinates Network, 15470de Ugarte Postigo A., Blazek M., Janout P., Sprimont P., Th¨oneC. C., Gorosabel J., S´anchez-Ram´ırez R., 2014a, in Soft-ware and Cyberinfrastructure for Astronomy III. p. 91520B,doi:10.1117/12.2055774de Ugarte Postigo A., Th¨one C. C., Tanvir N. R., Gorosabel J.,Fynbo J., Lombardi G., Reverte-Paya D., Perez D., 2014b,GRB Coordinates Network, 16968de Ugarte Postigo A., Tanvir N. R., Cano Z., Izzo L., FynboJ. P. U., Sanchez-Ramirez R., Thoene C. C., Pesev P., 2016,GRB Coordinates Network, 19245de Ugarte Postigo A., Kann D. A., Thoene C., Izzo L., TanvirN. R., Lombardi G., Marante A., 2017a, GRB CoordinatesNetwork, 20990de Ugarte Postigo A., Izzo L., Kann D. A., Thoene C., Cano Z.,Fynbo J. P. U., Garcia Alvarez D., 2017b, GRB CoordinatesNetwork, 21177MNRAS000 , 1–31 (2018) N. R. Tanvir et al. de Ugarte Postigo A., Cano Z., Izzo L., Thoene C. C., KannD. A., Castro-Rodriguez N., Valladares D. P., 2018, GRB Co-ordinates Network, Circular Service, No. 22346,
APPENDIX A: INDIVIDUAL BURSTS
In this appendix we provide information about selectedGRBs. In particular we present our Ly α fits for those caseswhere the column-densities have not previously been re-ported and are not included in Selsing et al. (2018). Unlessstated otherwise, the precise redshift is taken from metalabsorption lines seen in the spectra, which reduces the freeparameters, and improves accuracy of the fits. Details of thefitting procedure are given in Selsing et al. (2018).We caution that the nature of target-of-opportunity ob-servations of variable sources means that the source mag-nitude is frequently poorly known prior to observation,in many cases there was limited time available, and non-optimal conditions or sky location. Thus, some of the spectraare unusually low signal-to-noise and/or suffer from imper-fect flux calibration or other anomalies. Furthermore, thereare cases where host galaxy emission, including Ly α emissionlines, contaminates the afterglow signal, and where interven-ing absorbers introduce metal lines. However, fortunately forthe purposes of this analysis, the red-wing of the Ly α ab-sorption features can still be measured to an adequate levelof precision, with little overall systematic bias in determina-tion of N H i .Data are taken from various sources: some from observa-tions we obtained ourselves and in other cases from archives.Spectrographs used include the Gemini Multi-Object Spec-trograph North and South (GMOS-N, GMOS-S), the VLTFOcal Reducer and low dispersion Spectrograph (FORS1and FORS2) and UltraViolet Echelle Spectrograph (UVES),the William Herschel Telescope (WHT) Intermediate dis-persion Spectrograph and Imaging System (ISIS) and Aux-iliary port CAMera (ACAM), the Gran Telescopio Canarias(GTC) Optical System for Imaging and low-Intermediate-Resolution Integrated Spectroscopy (OSIRIS), the NordicOptical Telescope (NOT) Andalucia Faint Object Spectro-graph and Camera (ALFOSC), the Telescopio NazionaleGalileo (TNG) Device Optimized for the Low Resolution(DOLoRes), the Asiago Copernico Telescope (CT) AsiagoFaint Object Spectrograph and Camera (AFOSC), the KeckLow Resolution Imaging Spectrograph (LRIS).Spectra presented in this appendix will be made avail-able in the GRBspec database http://grbspec.iaa.es (de Ugarte Postigo et al. 2014a). A1 GRB 021004
The bright afterglow of GRB 021004 was well studied, andspectroscopy revealed an unusually complex velocity struc-ture for the absorbing gas (Fiore et al. 2005; Starling et al.2005; Castro-Tirado et al. 2010). Combined with saturationof the lines, establishing the H i column-density was not GRB 060522, z=5.11 N o r m a li s ed F l u x Figure A1.
Fit of the red wing of Ly α for the GRB 060522 Keck-I/LRIS spectrum. straight-forward, and here we adopt a reasonable value anderror based on the log ( N H i / cm − ) > from the the absenceof continuum below the Lyman limit (Fynbo et al. 2005) and log ( N H i / cm − ) < from analysis of the multiple Ly α linesthemselves (Møller et al. 2002). A2 GRB 060522
GRB 060522 was observed with Keck-I/LRIS, starting 14:24UT on 22-May-2006. A total exposure of s was ob-tained. The fit to the Ly α line is shown in Figure A1; theinferred column log ( N H i / cm − ) = . ± . is consistent withprevious estimates (Cenko et al. 2006; Chary et al. 2007).Note, the region redward of Ly α is badly affected by fring-ing, impeding searches for metal absorption lines. A3 GRB 070223
Near-IR imaging of GRB 070223 was obtained with theWHT Long-slit Intermediate Resolution Infrared Spectro-graph (LIRIS) instrument in the
JHK filters between 2.7and 3.7 hr post-burst. A faint source was detected at the lo-cation of the X-ray afterglow (Rol et al. 2007), for which wefind a magnitude K AB = . ± . calibrated against 2MASSstars in the field (and corrected for small foreground extinc-tion via Schlafly & Finkbeiner 2011). Subsequent imagingobtained 8 day post-burst showed this source to have de-clined to K AB = . ± . , confirming the identification ofthe afterglow, but also indicating the presence of an under-lying host galaxy (also detected in 3.6 µ m Spitzer imagingby Perley et al. 2016b).GRB 070223 was observed rapidly with the 2 m Liv-erpool Telescope (LT) in various optical filters, beginningonly 18 min post-burst (Melandri et al. 2008). We created a30 min stacked integration from r -band imaging taken be-tween 3 and 3.7 hr post-burst. At the location of the K -bandtransient, there is a faint detection of a source with fore-ground corrected magnitude r = . ± . , calibrated againstSDSS stars in the field. This object was also seen in R -bandearly imaging by the MDM 1.3 m telescope (Rol et al. 2007),but, as detailed in the text, appears to be a constant source,presumably the host galaxy. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRB 070810A, z=2.17 N o r m a li s ed F l u x Figure A2.
Fit of the red wing of Ly α for the GRB 070810AKeck-I/LRIS spectrum. GRB080129, z=4.349 N o r m a li s ed F l u x Figure A3.
Fit of the red wing of Ly α for the GRB 080129VLT/FORS1 spectrum. A4 GRB 070810A
GRB 070810A was observed with Keck-I/LRIS, starting05:47 UT on 10-Aug-2007. A total exposure of × swas obtained (Th¨one et al. 2007). The fit to the Ly α lineis shown in Figure A2. A5 GRB 080129
GRB 080129 was observed with VLT/FORS1, starting 05:24UT on 30-Jan-2008. A total exposure of × s wasobtained with the OG590 blocking filter and 300I grism,and reduced with the standard ESO pipeline (Greiner et al.2009b). The redshift is fixed to that of the metal absorptionlines, and the fit to the red wing of Ly α (Figure A3) is ratherpoor in this case, plausibly due to velocity structure in lowmetallicity and low column-density gas close to the host. A6 GRB 080810
GRB 080810 was observed at high resolution byKeck/HIRES (Prochaska et al. 2008), and the spec-trum showed a somewhat complex H i absorption system,with two main components separated by ≈ km s − . Thelower redshift system showed Si ii* fine-structure lines,suggesting this gas was closer to the GRB location, and GRB 080905B, z=2.374 N o r m a li s ed F l u x Figure A4.
Fit of the red wing of Ly α for the GRB 080905BVLT/FORS2 spectrum. likely that the other system is infalling on the near-side.Page et al. (2009) found an upper limit for the combinedabsorber of log ( N H i / cm − ) < . and a lower limit for thehigher redshift component of log ( N H i / cm − ) > . fromthe absence of emission below the Lyman limit. Recently,Wiseman et al. (2017) reanalysed the HIRES spectrum,finding log ( N H i / cm − ) = . ± . for this dominantcomponent, which we use here. A7 GRB 080905B
GRB 080905B was observed by VLT/FORS2, starting 01:16UT on 6-Sep-2008. A total exposure of × s was obtainedwith the GRIS 300V grism, and reduced with the standardESO pipeline (Vreeswijk et al. 2008). The fit to the Ly α lineis shown in Figure A4. A8 GRB 080913
GRB 080913 was observed with VLT/FORS2, being foundto be at redshift z ≈ . from the location of the Ly α breakby Greiner et al. (2009a). Lacking a precise metal line red-shift, Greiner et al. (2009a) were only able to place weakconstraints on the H i column-density concluding . < log ( N H i / cm − ) < . . A subsequent reanalysis of the spec-trum by Patel et al. (2010) located a weak S ii +Si ii blend,establishing a firmer redshift of z = . . This alloweda more precise determination of the H i column-density, log ( N H i / cm − ) = . assuming no neutral component ofthe IGM, and log ( N H i / cm − ) = . in a fit that allowed theneutral fraction of the IGM to be a free parameter (specif-ically they found x H i = . ). Given this uncertainty, andthe poor S/N of the spectrum, we adopt log ( N H i / cm − ) = . ± . here. A9 GRB 081029
GRB 081029 was observed by Gemini/GMOS-S, starting07:04 UT on 29-Oct-2008 (Cucchiara et al. 2008a). A to-tal exposure of × s was obtained with the R400 gratingset at 6000 ˚A central wavelength and reduced using the stan-dard Gemini reduction tools within Iraf . The fit to the Ly α line is shown in Figure A5. MNRAS000
GRB 081029 was observed by Gemini/GMOS-S, starting07:04 UT on 29-Oct-2008 (Cucchiara et al. 2008a). A to-tal exposure of × s was obtained with the R400 gratingset at 6000 ˚A central wavelength and reduced using the stan-dard Gemini reduction tools within Iraf . The fit to the Ly α line is shown in Figure A5. MNRAS000 , 1–31 (2018) N. R. Tanvir et al.
GRB 081029, z=3.847 N o r m a li s ed F l u x Figure A5.
Fit of the red wing of Ly α for the GRB 081029Gemini/GMOS-S spectrum. GRB 081118, z=2.581 N o r m a li s ed F l u x Figure A6.
Fit of the red wing of Ly α for the GRB 081118VLT/FORS2 spectrum. A10 GRB 081118
GRB 081118 was observed by VLT/FORS2, starting 02:48UT on 19-Nov-2008. A total exposure of × s was ob-tained with the GRIS 300V grism, and reduced with thestandard ESO pipeline (D’Elia et al. 2008). The fit to theLy α line is shown in Figure A6. A11 GRB 081222
GRB 081222 was observed by Gemini/GMOS-S, starting01:02 UT on 23-Dec-2008 (Cucchiara et al. 2008b). A totalexposure of × s was obtained with the R400 grating setat 6000 ˚A central wavelength, and reduced using the stan-dard Gemini reduction tools within Iraf . The fit to the Ly α line is shown in Figure A7. A12 GRB 090313
GRB 090313 was observed by Gemini/GMOS-S, starting04:20 UT on 14-Mar-2009 (Chornock et al. 2009a). A totalexposure of × s was obtained with the R400 grating setat 6000 ˚A central wavelength, and reduced using the stan-dard Gemini reduction tools within Iraf . The fit to the Ly α line is shown in Figure A8. GRB 081222, z=2.771 N o r m a li s ed F l u x Figure A7.
Fit of the red wing of Ly α for the GRB 081222Gemini/GMOS-S spectrum. GRB 090313, z=3.373 N o r m a li s ed F l u x Figure A8.
Fit of the red wing of Ly α for the GRB 090313Gemini/GMOS-S spectrum. Note that there is a gap betweenchips of the camera around 5280 ˚A. This produces an anomaly inthe trace which was ignored in the fit. A13 GRB 090323
GRB 090323 was unusual exhibiting two absorption sys-tems separated by ≈ km s − , and a relatively high metalabundance (Savaglio et al. 2012). In this case both systemsshowed Si ii* fine structure lines, likely indicating fairly closeproximity to the GRB. The value for N H i used here was ob-tained by summing the column-densities of the two systems. A14 GRB 090426
The prompt duration of GRB 090426 was T ≈ . s, sug-gesting it could be a short-duration burst, particularlygiven that cosmological time-dilation makes this less than0.4 s in the source-frame. However since it was intrinsicallybright and took place in an interacting star-forming system(Th¨one et al. 2011), we include it in our sample as a possi-ble long-duration GRB (see also Nicuesa Guelbenzu et al.2011). The H i column-density in this case was seen tovary, and we take here that measured at 1.1 hr post-burstby Levesque et al. (2010a) using Keck/LRIS. RemovingGRB 090426 from the sample would not have a significantaffect on any of the conclusions. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRB090519, z=3.845 N o r m a li s ed F l u x Figure A9.
Fit of the red wing of Ly α for the GRB 090519VLT/FORS2 spectrum. GRB090529, z=2.624 N o r m a li s ed F l u x Figure A10.
Fit of the red wing of Ly α for the GRB 090529VLT/FORS2 spectrum. A15 GRB 090519
GRB 090519 was observed by VLT/FORS2 starting at 01:03on 20-May-2009 UT (Thoene et al. 2009). A total exposureof × s was obtained, covering the wavelength range3500–9200˚A. The fit to the Ly α line is shown in Figure A9.The afterglow was faint, and the redshift is estimated fromthe Ly α and Ly β breaks since no clear metal lines were seen. A16 GRB 090529
GRB 090529 was observed by VLT/FORS2 starting at01:52 on 31-May-2009 UT, roughly 1.5 days post-burst(Malesani et al. 2009). A total exposure of × s wasobtained, covering the wavelength range 3500–9200˚A. Thefit to the Ly α line is shown in Figure A10. Although theS/N is unusually poor (largely due to the lateness of the ob-servation), the fit benefits from the redshift being fixed bymetal absorption lines. A17 GRB 090715B
GRB 090715B was observed with the WHT/ISIS starting23:46 UT on 15-Jul-2009 (Wiersema et al. 2009). This spec-trograph has a blue and a red arm, separated by a dichroic;
GRB 090715B, z=3.013 N o r m a li s ed F l u x Figure A11.
Fit of the red wing of Ly α for the GRB 090715BWHT/ISIS spectrum. GRB 090726, z=2.713 N o r m a li s ed F l u x Figure A12.
Fit of the red wing of Ly α for the GRB 090726 SAORAS 6-m / SCORPIO spectrum. we used the 300B and 316R gratings. Spectroscopic obser-vations consisted of × s exposure time. The data werereduced using standard techniques in IRAF . Several metalabsorption lines give a redshift z = . . The fit to the Ly α line is shown in Figure A11. A18 GRB 090726
GRB 090726 was observed by the SAO RAS 6-m telescopeusing the SCORPIO spectrograph starting at 00:15 on 27-Jul-2009 UT (Fatkhullin et al. 2009). A total exposure of s was obtained, covering the wavelength range 3700–7800˚A. The fit to the Ly α line is shown in Figure A12. A19 GRB 091029
GRB 091029 was observed by Gemini/GMOS-S, starting06:05 UT on 29-Oct-2009 (Chornock et al. 2009b). A totalexposure of × s was obtained with the R400 grating setat 6000 ˚A central wavelength and reduced using the standardGemini reduction tools within Iraf . The fit to the Ly α lineis shown in Figure A13. MNRAS000
GRB 091029 was observed by Gemini/GMOS-S, starting06:05 UT on 29-Oct-2009 (Chornock et al. 2009b). A totalexposure of × s was obtained with the R400 grating setat 6000 ˚A central wavelength and reduced using the standardGemini reduction tools within Iraf . The fit to the Ly α lineis shown in Figure A13. MNRAS000 , 1–31 (2018) N. R. Tanvir et al.
GRB 091029, z=2.751 N o r m a li s ed F l u x Figure A13.
Fit of the red wing of Ly α for the GRB 091029Gemini/GMOS-S spectrum. GRB 100302A, z=4.813
Observed Wavelength (
Å)0.00.20.40.60.81.01.21.4 N o r m a li s ed F l u x Figure A14.
Fit of the red wing of Ly α for the GRB 100302AGemini/GMOS spectrum. A20 GRB 100302A
GRB 100302A was observed by Gemini/GMOS-N, starting09:42 UT on 03-Mar-2010 (Chornock et al. 2010). A totalexposure of s was obtained with the R400 grating. Thecontinuum S/N is rather poor, but weak metal lines indicatea redshift of z = . . The fit to the Ly α red wing is shownin Figure A14. A21 GRB 100316A
GRB 100316A was observed by GTC/OSIRIS, starting 06:13UT on 16-Mar-2010 (S´anchez-Ram´ırez et al. 2013a). A totalexposure of × s was obtained with the R300B grating.The fit to the Ly α line is shown in Figure A15. The preciseredshift is known from the Ly α emission line in a late-timespectrum of the host. A22 GRB 100513A
GRB 100513A was observed by Gemini/GMOS-N, starting06:13 UT on 13-May-2010 (Cenko et al. 2010). A total expo-sure of × s was obtained with the R400 grating set at8000 ˚A central wavelength and reduced using the standardGemini reduction tools within Iraf . The fit to the Ly α lineis shown in Figure A16. GRB 100316A, z=3.155 N o r m a li s ed F l u x Figure A15.
Fit of the red wing of Ly α for the GRB 100316AGTC/OSIRIS spectrum. Ly α line emission from the host galaxyis evident in the absorption trough, but does not affect the fit. GRB 100513A, z=4.773 N o r m a li s ed F l u x Figure A16.
Fit of the red wing of Ly α for the GRB 100513AGemini/GMOS-N spectrum. The large dips redward of the Ly α line are residuals due to gaps between the detectors in the spec-trograph. A23 GRB 110731A
GRB 110731A was observed by Gemini/GMOS-N, starting09:08 UT on 01-Aug-2011 (Tanvir et al. 2011). A total ex-posure of × s was obtained with the B600 grating set at5250 ˚A central wavelength and reduced using the standardGemini reduction tools within Iraf . The fit to the Ly α lineis shown in Figure A17. A24 GRB 120811C
GRB 120811C was observed by the GTC/OSIRIS, starting15:35 UT on 11-Aug-2012 (Th¨one et al. 2012). A total ex-posure time of 2400 s was obtained, spanning a wavelengthrange 3640–7875 ˚A. The fit to the Ly α line is shown in Fig-ure A18. A25 GRB 121027A
GRB 121027A has been suggested as a member of the ‘ultra-long’ class of GRBs, whose exact nature remains uncer-tain, but since they also appear to be associated with mas-sive stars in low metallicity galaxies (Levan et al. 2014;
MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRB 110731A, z=2.83 N o r m a li s ed F l u x Figure A17.
Fit of the red wing of Ly α for the GRB 110731AGemini/GMOS-N spectrum. The apparent feature at 4800 ˚A isdue to the gap between the detectors in the spectrograph. GRB 120811C, z=2.671 N o r m a li s ed F l u x Figure A18.
Fit of the red wing of Ly α for the GRB 120811CGTC/OSIRIS spectrum. Greiner et al. 2015a; Kann et al. 2017), we include it in oursample.
A26 GRB 121128A
GRB 121128A was observed by Gemini/GMOS-N, starting06:28 UT on 28-Nov-2012 (Tanvir et al. 2012b). A total ex-posure of × s was obtained with the B600 grating set at5250 ˚A central wavelength and reduced using the standardGemini reduction tools within Iraf . The fit to the Ly α lineis shown in Figure A19. A27 GRB 130518A
GRB 130518A was observed at high resolution withGTC/OSIRIS, starting 04:47 UT on 20-May-2013. A to-tal exposure time of 840 s was obtained, spanning a wave-length range 3700–7800 ˚A and was originally reported inSanchez-Ramirez et al. (2013b). The fit to the Ly α line isshown in Figure A20. GRB 121128A, z=2.199 N o r m a li s ed F l u x Figure A19.
Fit of the red wing of Ly α for the GRB 121128AGemini/GMOS-N spectrum. GRB 130518A, z=2.488 N o r m a li s ed F l u x Figure A20.
Fit of the red wing of Ly α for the GRB 130518AGTC/OSIRIS spectrum. A second, intervening DLA, at z = . is apparent. GRB 130610A, z=2.092 N o r m a li s ed F l u x Figure A21.
Fit of the red wing of Ly α for the GRB 130610AVLT/UVES spectrum. A28 GRB 130610A
GRB 130610A was observed at high resolution withVLT/UVES, starting 03:25 UT on 10-Jun-2013(Smette et al. 2013). A series of exposures were ob-tained totalling 6500 s. The fit to the Ly α line is shown inFigure A21. MNRAS , 1–31 (2018) N. R. Tanvir et al.
GRB 131108A, z=2.400 N o r m a li s ed F l u x Figure A22.
Fit of the red wing of Ly α for the GRB 131108AGTC/OSIRIS spectrum. GRB 140206A, z=2.730 N o r m a li s ed F l u x Figure A23.
Fit of the red wing of Ly α for the GRB 140206ANOT/ALFOSC spectrum. A29 GRB 131108A
GRB 131108A was observed with the GTC/OSIRIS, starting20:42 UT on 08-Nov-2013 (de Ugarte Postigo et al. 2013).A total exposure of 1800 s was obtained covering a spectralrange 3700–7870 ˚A. The fit to the Ly α line is shown in Fig-ure A22. A30 GRB 140206A
GRB 140206A was observed with the NOT/ALFOSC start-ing 19:56 UT on 6-Feb-2014 (Malesani et al. 2014, see alsoD’Elia et al. (2014) for TNG/DOLoRes spectroscopy). A to-tal exposure of 3600 s was obtained covering a spectral range3750–9000 ˚A. The fit to the Ly α line is shown in Figure A23. A31 GRB 140515A
GRB 140515A was a high redshift burst observed at sev-eral facilities. No metal lines were confidently detected sothe redshift could only be estimated from the Ly α break it-self. This limits the conclusions that can be drawn, since thedamping wing must be decomposed into ISM and IGM con-tributions, which is less certain in the absence of a precise GRB 140629A, z=2.275 N o r m a li s ed F l u x Figure A24.
Fit of the red wing of Ly α for the GRB 140629ATNG/DOLoRes spectrum. redshift. Nonetheless, the sharpness of the break clearly in-dicates a relatively low H i column-density. Chornock et al.(2014) obtained a value of log ( N H i / cm − ) = . ± . assuming an ionized IGM and log ( N H i / cm − ) = . ina joint fit including a neutral IGM component from anearly Gemini-N/GMOS spectrum. Melandri et al. (2015)analysed later GTC and VLT spectroscopy and concluded log ( N H i / cm − ) < . whereas Selsing et al. (2018) estimate log ( N H i / cm − ) = . ± . . In this paper we therefore adopta compromise value of log ( N H i / cm − ) = . ± . . A32 GRB 140629A
GRB 140629A was observed with the TNG/DOLoRes, start-ing 02:07 UT on 30-Jun-2014 (D’Avanzo et al. 2014). A totalexposure of 1200 s was obtained, covering a spectral range3000–8000 ˚A. The fit to the Ly α line is shown in Figure A24. A33 GRB 140703A
GRB 140703A was observed with the GTC/OSIRIS, start-ing 03:16 UT on 3-Jul-2014. A total exposure of 450 s wasobtained covering a spectral range 3700–10000 ˚A and wasoriginally reported in Castro-Tirado et al. (2014). The fit tothe Ly α line is shown in Figure A25. A34 GRB 140808A
GRB 140808A was observed with the GTC/OSIRIS, starting00:54 UT on 08-Aug-2014 (Gorosabel et al. 2014). A totalexposure of 3600 s was obtained covering a spectral range3700–7800 ˚A. The fit to the Ly α line is shown in Figure A26. A35 GRB 150413A
GRB 150413A was observed with the Asi-ago(CT)/AFOSC, starting 20:53 UT on 13-Apr-2015(de Ugarte Postigo & Tomasella 2015). A total exposureof 1800 s was obtained covering a wavelength range 3400–8200 ˚A. The fit to the Ly α line is shown in Figure A27. MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRB 140703A, z=3.144 N o r m a li s ed F l u x Figure A25.
Fit of the red wing of Ly α for the GRB 140703AGTC/OSIRIS spectrum. GRB 140808A, z=3.29 N o r m a li s ed F l u x Figure A26.
Fit of the red wing of Ly α for the GRB 140808AGTC/OSIRIS spectrum. GRB 150413A, z=3.139 N o r m a li s ed F l u x Figure A27.
Fit of the red wing of Ly α for the GRB 150413ACT/AFOSC spectrum. A36 GRB 151215A
GRB 151215A was observed with the NOT/ALFOSC, start-ing 04:13 UT on 15-Dec-2015 (Xu et al. 2015). A total ex-posure of × s was obtained covering a spectral range3200–9000 ˚A. The fit to the Ly α line is shown in Figure A28. GRB 151215A, z = 2.586 N o r m a li s ed F l u x Figure A28.
Fit of the red wing of Ly α for the GRB 151215ANOT/ALFOSC spectrum. GRB 160227A, z=2.373 N o r m a li s ed F l u x Figure A29.
Fit of the red wing of Ly α for the GRB 160227ANOT/ALFOSC spectrum. A37 GRB 160227A
GRB 160227A was observed with the NOT/ALFOSC, start-ing at 20:19 UT on 27-Feb-2016 (Xu et al. 2016b). A totalexposure of 4800 s was obtained covering a wavelength range3200–9000 ˚A. The fit to the Ly α line is shown in Figure A29. A38 GRB 160629A
GRB 160629A was observed with the GTC/OSIRIS, start-ing at 04:40 UT on 30-Jun-2016 (Castro-Tirado et al. 2016).A total exposure of 600 s was obtained covering a wavelengthrange 3700–7880 ˚A. The fit to the Ly α line is shown in Fig-ure A30. A39 GRB 161017A
GRB 161017A was observed with the TNG/DOLoRes, start-ing at 04:27 UT on 18-Oct-2016 (D’Avanzo et al. 2016). Atotal exposure of 1200 s was obtained covering a wavelengthrange 3500–8000 ˚A. The fit to the Ly α line is shown in Fig-ure A31.For this GRB we also conducted a host search usingthe WHT/ACAM on 7 April 2017. In seeing of 1.1 arcsecwe obtained a 45 min integration in the g -band. No source MNRAS000
GRB 161017A was observed with the TNG/DOLoRes, start-ing at 04:27 UT on 18-Oct-2016 (D’Avanzo et al. 2016). Atotal exposure of 1200 s was obtained covering a wavelengthrange 3500–8000 ˚A. The fit to the Ly α line is shown in Fig-ure A31.For this GRB we also conducted a host search usingthe WHT/ACAM on 7 April 2017. In seeing of 1.1 arcsecwe obtained a 45 min integration in the g -band. No source MNRAS000 , 1–31 (2018) N. R. Tanvir et al.
GRB 160629A, z=3.332 N o r m a li s ed F l u x Figure A30.
Fit of the red wing of Ly α for the GRB 160629AGTC/OSIRIS spectrum. GRB 161017A, z=2.0127 N o r m a li s ed F l u x Figure A31.
Fit of the red wing of Ly α for the GRB 161017ATNG/DOLoRes spectrum. was detected at the GRB position down to a 2 σ limitingmagnitude of g = . , which is corrected for foregroundMilky Way extinction (Schlafly & Finkbeiner 2011). A40 GRB 170405A
GRB 170405A was observed with the GTC/OSIRIS, startingat 02:14 UT on 6-Apr-2017 (de Ugarte Postigo et al. 2017a).A total exposure of × s was obtained covering a wave-length range 3700–7800 ˚A. The fit to the Ly α line is shownin Figure A32. A41 GRB 170531B
GRB 170531B was observed with the GTC/OSIRIS, startingat 02:47 UT on 1-Jun-2017 (de Ugarte Postigo et al. 2017b).A total exposure of × s was obtained covering a wave-length range 3700–7880 ˚A. The fit to the Ly α line is shownin Figure A33. A42 GRB 180115A
GRB 180115A was observed with the GTC/OSIRIS, startingat 20:32 UT on 15-Jan-2018 (de Ugarte Postigo et al. 2018).
GRB 170405A, z=3.510 N o r m a li s ed F l u x Figure A32.
Fit of the red wing of Ly α for the GRB 170405AGTC/OSIRIS spectrum. GRB 170531B, z=2.366 N o r m a li s ed F l u x Figure A33.
Fit of the red wing of Ly α for the GRB 170531BGTC/OSIRIS spectrum. GRB 180115A, z=2.487 N o r m a li s ed F l u x Figure A34.
Fit of the red wing of Ly α for the GRB 180115AGTC/OSIRIS spectrum. A total exposure of × s was obtained covering a wave-length range 3700–7880 ˚A. The fit to the Ly α line is shownin Figure A34. In this case, no metal lines were detected, sothe redshift is based solely on Ly α . MNRAS , 1–31 (2018) he escape fraction of ionizing radiation GRB 180329B, z=1.998 N o r m a li s ed F l u x Figure A35.
Fit of the red wing of Ly α for the GRB 180329BVLT/X-shooter spectrum. A43 GRB 180329B
GRB 180329B was observed with the VLT/X-shooter, start-ing at 00:10 UT on 30-Mar-2018 (Izzo et al. 2018). A totalexposure of × s was obtained covering a wavelengthrange 3000–21000 ˚A. The fit to the Ly α line is shown in Fig-ure A35. This paper has been typeset from a TEX/L A TEX file prepared bythe author.MNRAS000