METAL: The Metal Evolution, Transport, and Abundance in the Large Magellanic Cloud Hubble program. II. Variations of interstellar depletions and dust-to-gas ratio within the LMC
Julia Roman-Duval, Edward B. Jenkins, Kirill Tchernyshyov, Benjamin Williams, Christopher J.R. Clark, Karl D. Gordon, Margaret Meixner, Lea Hagen, Joshua Peek, Karin Sandstrom, Jessica Werk, Petia Yanchulova Merica-Jones
DDraft version January 26, 2021
Preprint typeset using L A TEX style emulateapj v. 08/22/09
METAL: THE METAL EVOLUTION, TRANSPORT, AND ABUNDANCE IN THE LARGE MAGELLANICCLOUD HUBBLE PROGRAM. II. VARIATIONS OF INTERSTELLAR DEPLETIONS AND DUST-TO-GASRATIO WITHIN THE LMC
Julia Roman-Duval , Edward B. Jenkins , Kirill Tchernyshyov , Benjamin Williams , Christopher J.R. Clark ,Karl D. Gordon , Margaret Meixner , Lea Hagen , Joshua Peek , Karin Sandstrom , Jessica Werk , PetiaYanchulova Merica-Jones Draft version January 26, 2021
ABSTRACTA key component of the baryon cycle in galaxies is the depletion of metals from the gas to the dustphase in the neutral ISM. The METAL (Metal Evolution, Transport and Abundance in the LargeMagellanic Cloud) program on the
Hubble Space Telescope acquired UV spectra toward 32 sightlinesin the half-solar metallicity LMC, from which we derive interstellar depletions (gas-phase fractions)of Mg, Si, Fe, Ni, S, Zn, Cr, and Cu. The depletions of different elements are tightly correlated,indicating a common origin. Hydrogen column density is the main driver for depletion variations.Correlations are weaker with volume density, probed by C I fine structure lines, and distance to theLMC center. The latter correlation results from an East-West variation of the gas-phase metallicity.Gas in the East, compressed side of the LMC encompassing 30 Doradus and the Southeast H i over-density is enriched by up to +0.3 dex, while gas in the West side is metal-deficient by up to − ∼
20 cm − ) and molecular (log N(H) ∼
22 cm − ) ISM observed fromfar-infrared, 21 cm, and CO observations. The variations of dust-to-metal and dust-to-gas ratioswith column density have important implications for the sub-grid physics of chemical evolution, gasand dust mass estimates throughout cosmic times, and for the chemical enrichment of the Universemeasured via spectroscopy of damped Lyman- α systems. Subject headings:
ISM: atoms - ISM: Dust INTRODUCTION
The transfer of metals between interstellar gas anddust constitutes an important component of the baryoncycle in galaxies, the incessant recycling of gas, dust,and metals between stars, the interstellar medium, andgalaxy halos. Recent observations and modeling haveshown that interstellar dust can grow and evolve in theinterstellar medium (ISM). The evolution of the dust con-tent of galaxies over cosmic times (Morgan & Edmunds2003; Boyer et al. 2012; Rowlands et al. 2012, 2014;Zhukovska & Henning 2013) cannot be explained by bal-ance of the dust production rates in evolved stars (Bladh& H¨ofner 2012; Riebel et al. 2012; Srinivasan et al. 2016)and supernova remnants (Matsuura et al. 2011) and thedust destruction rates in interstellar shocks (Jones et al.1994, 1996). This so-called dust budget crisis can beresolved by dust growth in the ISM, via accretion of gas-phase metals onto pre-existing dust grains (Zhukovska Space Telescope Science Institute, 3700 San Martin Drive, Bal-timore, MD 21218; [email protected] Princeton University Observatory, Peyton Hall, Princeton Uni-versity, Princeton, NJ 08544-1001 USA Department of Astronomy, Box 351580, University of Wash-ington, Seattle, WA 98195, USA SOFIA Science Mission Operations, NASA Ames ResearchCenter, Building N232, M/S 232-12, P.O. Box 1, Moffett Field,CA 94035-0001 Center for Astrophysics and Space Sciences, Department ofPhysics, University of California, 9500 Gilman Drive, La Jolla, SanDiego, CA 92093, USA NASA Goddard Space Flight Center, Greenbelt, MD 20771,USA et al. 2008; Draine 2009; McKinnon et al. 2016), effec-tively modifying the relation between dust and gas mass.The dust-to-metal ratio (D/M) and the dust-to-gas ra-tio (D/H = D/M × Z, where Z is the metallicity) arefundamental parameters resulting from the interstellargas-dust cycle, and are expected to substantially varywith environment, in particular metallicity (Asano et al.2013; Feldmann 2015).Owing to the key role of dust in the radiative transfer,chemistry, and thermodynamics, galaxy evolution can-not be understood without accounting for dust, and thusfor the interstellar gas-dust cycle. Because 30%-50% ofstellar light is absorbed by dust in the optical-ultraviolet(UV) and re-emitted in the far-infrared (FIR), our under-standing of dust is critical to correct for reddening effects,interpret observations of galaxies, and trace their stellar,metal, dust, and gas content over cosmic times. Addi-tionally, a comprehensive understanding of how metalsdeplete from the gas to the dust phase via dust forma-tion in the ISM, and from the dust to the gas phase viadust destruction by sputtering in Supernova (SN) shockwaves, is critical to understand the chemical enrichmentof the universe over cosmic times. Indeed, the chemicalenrichment of the universe is traced by quasar absorptionspectroscopy through damped Lyman- α systems (DLAs,e.g., Rafelski et al. 2012), and the resulting DLA neu-tral gas abundances have to be corrected for depletioneffects, particularly above 1% solar metallicity.D/M and D/H can either be estimated in two ways.The first method consists in using emission-based trac- a r X i v : . [ a s t r o - ph . GA ] J a n ers of gas (H i
21 cm and CO rotational transitions)and dust (FIR emission) as in, e.g., R´emy-Ruyer et al.(2014); De Vis et al. (2019). However, the degeneraciesbetween the dust opacity, dust mass, and CO dark-gas(e.g., Roman-Duval et al. 2014; Galliano et al. 2018) pre-clude an unambiguous characterization of the variationsof D/M and D/H with metallicity using these emission-based ISM tracers. The second approach compares chem-ical abundances in neutral interstellar gas based on UVspectroscopy to stellar abundances of young stars re-cently formed out of ISM (e.g., Luck et al. 1998), whichcan be used as a proxy for the total (gas + dust) abun-dances of the ISM (i.e., interstellar depletions, as in Sav-age & Sembach 1996; Jenkins 2009; Tchernyshyov et al.2015). In order to understand how metals deplete fromthe gas into the dust phase, and thus how D/M varieswith environment, a detailed census of metals in neutralgas and dust from UV spectroscopy is therefore required.Interstellar depletions are the logarithm of the fractionof metals in the gas-phase. Thus, the depletion for ele-ment X is δ ( X ) = log ( X/H ) gas − log ( X/H ) total (1)where ( X/H ) gas is the abundance of X in the gas-phaseand ( X/H ) total is the total abundance of X (gas+dust),assumed to be equal to the abundance of X in the pho-tospheres of young stars that have formed out of the ISMrecently. Metallicities estimated from the emission linesof H ii regions suffer from relatively large systematics dueto 1) the need to estimate a temperature from line ra-tios and temperature fluctuations inside these regions, 2)the poorly understood discrepancy between recombina-tion and collisionnally excited lines, and 3) the presenceof dust in H ii regions, removing some of the metals fromthe gas.In the Milky Way (Jenkins 2009) and SMC (Jenkins &Wallerstein 2017), the fraction of metals in the gas-phasedecreases with increasing hydrogen volume density andcolumn density, albeit at different rates for different el-ements. In this paper, we derive interstellar depletionsin the Large Magellanic Cloud (LMC) using recent ob-servations with the Hubble Space Telescope obtained aspart of the METAL large Cycle 24 program (GO-14675,see Roman-Duval et al. 2019). The LMC metallicity (1/2solar, see Russell & Dopita 1992) lies approximately mid-way between that of the MW (solar) and the SMC (1/5solar, see Russell & Dopita 1992), and provides the linkbetween the large differences in dust properties seen be-tween the MW and SMC, which is below the ”criticalmetallicity” (Feldmann 2015) where the dust-to-gas ra-tio departs from a linear scaling with metallicity (R´emy-Ruyer et al. 2014; Roman-Duval et al. 2014, 2017; Chianget al. 2018), the PAH fraction is an order of magnitudelower than in the MW (Sandstrom et al. 2010), and theUV extinction curves distinctively lacks a 2175 ˚A bumpand are steeper in the FUV than in the MW (Gordonet al. 2003). In addition, the LMC’s gas disk is thinner(120 pc, Elmegreen et al. 2001) and less inclined thanthe SMC, alleviating confusion in velocity and distancestructure along the line-of-sight.In this paper, we focus on the variations of interstellardepletions with environment within the LMC. In an up-coming paper, we will perform a detailed comparison of depletions between the Milky Way, LMC, and SMC. Thepaper is organized as follows. We describe the observa-tions and abundance measurements in Sections 2 and 3.In Section 4, we describe the derivation of volume den-sity and radiation fields from the C I fine structure line.The correlations between depletions of different elementsand with environment are examined in Sections 5 and 6.Combining depletions of different dust constituents, thevariations of the dust-to-gas ratio with environment arederived in Section 7. Section 8 provides a summary ofthis paper. OBSERVATIONS
The details of the observing strategy, sample proper-ties, and survey parameters were covered in the METALSurvey paper (Roman-Duval et al. 2019, ; hereafter Pa-per I). For convenience we provide a brief summary andexplanation of the sample here. In short, this study ofinterstellar depletions in the LMC is based on STIS andCOS medium-resolution spectra of 32 massive stars, ob-tained predominantly as part of the METAL large HSTprogram (GO-14675, see Roman-Duval et al. 2019), butalso from archival HST spectra of the same target sample(DOI 10.17909/t9-g6d9-rj76). The target sample used inthis analysis is listed in Table 1, along with the LMC H i and H column densities toward each star, and the he-liocentric velocity intervals with detectable absorption inthe Fe II lines (1608 ˚A, 2249 ˚A, 2260 ˚A). Information inTable 1 is directly taken from the survey paper (Roman-Duval et al. 2019, , hereafter Paper I).The COS spectra were obtained with the G130M andG160M gratings, while the STIS spectra used the E140Mand E230M. Two of the targets (SK-70 115 and SK-6873) had archival E230H spectra as well, which were pre-ferred since they minimized the effects of unresolved sat-uration. The complete list of medium-resolution spec-tra used in this analysis, including the program IDs ofarchival data, is included in Paper I (Tables A1 and A2). GAS-PHASE COLUMN DENSITY AND ABUNDANCEMEASUREMENTS
The quality of the METAL spectra allows us to deriveneutral gas abundances for Fe, Mg, Si, Ni, Cu, Cr, Zn, S,in all METAL targets listed in Table 1 (i.e., all METALtargets except SK-69 220, which is an LBV with a verycomplex stellar spectrum precluding accurate measure-ments of interstellar features). Gas-phase column den-sities for transitions observed in the STIS/E140M ( R =45,000) and STIS/E230M ( R = 30,000) spectra were de-rived using the apparent optical depth method (AOD,Savage & Sembach 1991; Jenkins 1996), as outlined inPaper I for the Si abundances. The methodology forthe STIS-based AOD measurements is described in Sec-tion 3.1. For transitions in the COS G130M and G160Mspectra ( R = 15,000-20,000), we used profile fitting toderive gas-phase column densities, following the methoddescribed in Section 3.3. In some instances of targets andions covered by both instruments, we chose the STIS-based abundances for this analysis given the higher res-olution. We compare the column densities derived bythese two methods at different spectral resolutions inSection 9.3. Lastly, ionization corrections for singly ion-ized column densities are negligible in the column densityrange of our sight-lines (log N(H) = 20-22 Tchernyshyov TABLE 1Spectroscopic targets and their interstellar parameters
Target Ra Dec E ( B − V ) log N (H i ) LMCa log N (H ) b V helio h deg mag cm − cm − km s − SK-67 2 04h47m04.451s -67d06m53.12s 0.26 21.04 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± a The H i column densities are from Roman-Duval et al. (2019) b The H column densities are from Welty et al. (2012) et al. 2015; Jenkins & Wallerstein 2017), so the abun-dance of an element is taken to be that of the measuredlow ion.The resulting equivalent widths and column densitiesfor each individual spectral line in the STIS spectra arelisted in Table 3, while the combined column density andabundance measurements for each element (combiningdifferent spectral lines) in the METAL survey are in-cluded in Table 5. Column densities with the AOD method in theSTIS spectra
We first determined the continuum levels by best-fitting Legendre polynomials (Sembach & Savage 1992)to fluxes on either side of the absorption profiles (seethe top panels of Figures 1 and 2 for some examples).We then applied the AOD method to each feature anddetermined the apparent column density N a as a func-tion of velocity. The velocity ranges used to integratethe AOD-based apparent column densities in the STISspectra are listed in Table 1. The upper portions of theintegration velocity ranges are well defined by the sharp,long wavelength edges of strong transitions, such as theFe II i
21 cm emission in the GASS III sur-vey (McClure-Griffiths et al. 2009; Kalberla et al. 2010;Kalberla & Haud 2015) in the directions of our targetstars, which reveal a well defined feature arising fromgas in the LMC. Specifically, the lower velocity limit ofAOD integration corresponds to the edge of the H i
21 cmemission, where the brightness temperature rises above0.2 K ( ∼ σ ). Our basic premise is therefore that muchof the foreground gas at lower velocities contains fullyionized hydrogen and thus should not be included.To convert the integrated apparent optical depths tocolumn densities, we assume the oscillator strengthslisted in Table 2 for each element/transition. For ele-ments with several transitions from which column densi-ties can be measured, we perform a weighted average ofthe different column densities, where the weight is givenby the inverse square of the error.In some cases the lines are saturated, or nearly so, andwhen this saturation is not properly resolved, a straightevaluation of the AOD will underestimate the true col-umn density. If the weakest (or only) line has a centralintensity relative to the continuum level I /I c < .
05, wesimply declare a lower limit to the column density. Formilder cases of saturation with I /I c (cid:46) .
3, we either sig-nificantly increased the upper error bound for the derivedcolumn density or, when two lines of differing strengthswere available, we applied a correction for saturation tak-ing the following approach. In cases where the saturation F l ux ( - e r g c m - s - Å - ) C on ti nuu m N o r m a li ze d F l ux -60 -40 -20 0 20 40 60Velocity offset (km s - )-101234 N a ( c m - [ k m s - ] - ) Sk-69 104S II 1250.578 Å F l ux ( - e r g c m - s - Å - ) C on ti nuu m N o r m a li ze d F l ux -60 -40 -20 0 20 40 60Velocity offset (km s - )-10123 N a ( c m - [ k m s - ] - ) Sk-69 104
S II 1253.805 Å
Fig. 1.—
Demonstrations of various aspects of deriving AOD outcomes and their uncertainties for Sk-69 104 and S II ( λ λ N a ( v ), withthe black trace showing the preferred values, the blue traces on either side showing deviations that arise from the upper and lower limitsfor the adopted continuum, and the green error bars illustrating the random uncertainties arising from photon counting noise. The verticaldashed lines indicate the velocity limits (see Table 1). in the lines of S II λλ II λλ II , where AOD determinations at spe-cific velocities are compared with each other and treatedin the same manner as equivalent widths in a standardcurve-of-growth analysis. The corrected outcomes for logN(S II ) and N(Zn II ) are can be identified in Table 3 be-cause they show no specifications for wavelengths, oscil-lator strengths, or equivalent widths (as opposed to themeasurement of the strong and weak lines). The magni-tudes of the corrections are typically +0.15 dex for S II and +0.2 for Zn II .The lines of Zn II presented special challenges aris- ing from the interference of other lines at similar wave-lengths, which can be troublesome because the velocityranges of absorption features from the LMC are large.The left-hand side of the Zn II line at 2062.6604 ˚A mayhave an overlap from the Cr II line at 2062.2361 ˚A, whichis situated a relative velocity of − − , andthe right-hand side of the Zn II line at 2026.1370 ˚A cansuffer from interference by the Mg I line at 2026.4768˚A (at +50.3 km s − ). To compensate for such inter-fering features, we use apparent optical depths of otherCr II and Mg I lines, multiply them by the appropriateratios of f λ λ , and then subtract them with their respec-tive velocity offsets from the Zn II features. This proce-dure works well when using the Cr II line at 2056.2569˚A because its value of f λ λ is not much different fromthat of the 2062.2361˚A line. The Mg I comparison line at1827.9351˚A is considerably weaker than that of the inter-fering 2026.4768 ˚A transition, and if the latter has unre-solved saturation, the compensation will be larger thanwarranted. When it was clearly evident that this was TABLE 2Spectral lines and oscillator strengths used forabundance and depletion measurements
Element/ion 12 + log(X/H)
LMC , tot Wavelength log λf λ ˚A ˚AMg II ± II ± I ± II ± II ± II ± II ± II ± II ± II ± Note . — Mg, Si, O, P, Cr, Fe, Ni, Zn LMC total (gas + dust)abundances are from Tchernyshyov et al. (2015); S, Cu abun-dances are from Asplund et al. (2009) scaled by a factor 0.5. Oscil-lator strengths are from Morton (2003), except for Zn II (Kisieliuset al. 2015), S II (Kisielius et al. 2014), and Ni II (Jenkins & Tripp2006) happening, we disregarded the (usually extreme) right-hand portion of the Zn II II feature when we cal-culated the effects of saturation by comparing the AODoutcomes for the two lines. For the individual resultsthat we obtained for the 2062 ˚A line of Zn II in Table 3,we list the AOD column density outcomes after subtract-ing off the contribution from Cr II . For the 2026 ˚A line,the corrections were often larger than they should havebeen, for the reason discussed above. Hence, for this linewe listed the uncorrected column densities.At the opposite extreme, we considered a measurementto be marginal if the equivalent width outcome was lessthan the 2 σ level of uncertainty from noise and contin-uum placement. For weak lines below this uncertaintythreshold, we specified an upper limit for the columndensity based on a completely unsaturated line having astrength at the measurement value plus a 1 σ positive ex-cursion, but with an allowance for the fact that negativereal line strengths are not allowed even though we occa-sionally obtained negative measurement outcomes causedby downward noise fluctuations (or a continuum place-ment that was too low). Details of how we calculatedthese 1 σ upper limits are given in Appendix D of Bowenet al. (2008). Such a calculation avoids the unphysical F l ux ( - e r g c m - s - Å - ) C on ti nuu m N o r m a li ze d F l ux -100 -50 0 50 100Velocity offset (km s - )-101234 N a ( c m - [ k m s - ] - ) Sk-69 175
Fe II 2260.780 Å
Fig. 2.—
Same as Figure 1, but for Sk-69 175 and Fe II λ conclusion that an upper limit for a column density canbe nearly zero or negative when the measurement yieldsan outcome that is < − σ . It also yields a smooth tran-sition to a conventional expression of an upper limit asa value plus 1 σ when the value is larger than twice thenoise level.Errors on the column densities stem from the effects ofthree different sources: (1) noise in the absorption pro-file, (2) errors in defining the continuum level, and (3)uncertainties in the transition oscillator strengths, f λ , allof which were combined in quadrature in our error esti-mation. We evaluate the expected deviations producedby continuum definition by remeasuring the AODs at thelower and upper bounds for the continua, which are de-rived from the expected formal uncertainties in the poly-nomial coefficients of the fits as described by Sembach &Savage (1992). We multiply these coefficient uncertain-ties by 2 in order to make approximate allowances for ad-ditional deviations that might arise from some freedomin assigning the most appropriate order for the polyno-mial.When two or more, non-saturated lines - as shown bythe weak line strengths and the consistency of the de-rived column densities between lines of different oscillatorstrengths - were available for a given element (e.g., Fe II,
180 200 220 240 260 280v [km s - ]-0.50.00.51.01.52.02.5 N a ( v ) [ c m - / ( k m s - )] Corrected1253.8051250.578
Fig. 3.—
Apparent column density of S II toward BI 173 as mea-sured from the λ λ Ni II, Mg II), we derived the final column density valueas average of the column densities derived from each line,weighted by the inverse of their squared errors.
The problematic case of the single Si II λ For Si II , the only line that is not always badly sat-urated is the λ II column density determination usingdifferent transitions of varying oscillator strengths, as istypically done with the AOD method (see previous Sec-tion). Nonetheless, the λ II being a key component of dust, itis important to measure its abundance. Therefore, wehave devised a hybrid method to constrain the columndensity and abundance of Si II . The method uses boththe AOD and curve-of-growth (COG) analysis performedon multiple non- or very mildly saturated transitions forother elements (Fe II , Mg II , S II ) with similar equiva-lent widths to that of the single Si II λ b . Theapproach then applies this b value to the Si II COG toadjust the AOD-determined Si II column density and as-sociated errors, which can be impacted by saturation.The methodology for evaluating and correcting the ef-fects of saturation on the Si II column density determi-nation from the AOD method is illustrated in Figures 4(for Sk-67 5) and 5 (for BI 173). These two figures showprobability contours based on a χ distribution with twodegrees of freedom for various combinations of log N (x-axis) and b (y-axis) for a standard COG, given the mea-sured values of equivalent widths and their uncertaintiesfor each of the two transitions used (e.g., λλ II , λλ II , λλ II ). The column density determined from the AODmethod is also overlaid on the contours. The COG andAOD results are shown for Mg II , S II , Fe II , and Si II . ForMg II , S II , Fe II , which have two measured transitions,the contours are closed or half-closed and log N and b are constrained, while for Si II , we have only one mea-surement but two parameters, and hence the contours are open ended and only indicate unacceptable combi-nations of log N and b . Our objective is to determine aplausible range of b values for Si II relying on the com-bined COG and AOD analyses for Mg II , Fe II , S II , sothat log N (Si II ) can be constrained given the Si II COGcontours.For transitions that do not suffer from saturation(Mg II , Fe II ), the column densities determined fromthe AOD method applied to lines with different oscil-lator strengths are within errors, and the column densitydetermination is consistent with the COG contours. ForS II , saturation effects can be evaluated and correctedfor using the Jenkins (1996) method. In this case, theintersection of the corrected column density is generallyin agreement with the COG contours. Thus, the inter-section of the AOD-determined N and the COG contoursprovides a well-constrained range of b values, shown asgray lines in Figures 4 and 5.The range of allowed b values depends on the strengthof the transition, on the measurement errors on the AOD-derived column density, and on the tightness of the COGcontours, which in turn depends on the error bars on themeasured equivalent widths. Thus, different elementsprovide constraints on b that are generally consistent,but can differ between different elements. One impor-tant criterion in prioritizing the b -values derived fromdifferent elements is that the two features have transitionprobabilities that differ enough to give good indicationsof the COG behavior. The lines in the Mg II doublethave strengths that differ by 1.77, and the depletion be-havior of Mg II is similar to that of Si II in the MW.However, both of the lines in the doublet are consider-ably weaker than the Si II feature, and effective b valuescan change with increasing line strength as weaker non-Gaussian wings start to become more influential. S II has lines that differ in strength by a factor 2.01, and theequivalent widths are more similar to those of Si II (withthe benefit that saturation effects can be corrected forusing the Jenkins (1996) method). One possible disad-vantage with S II is that its depletion behavior may differfrom that of Si II , which may drive a difference in the ve-locity behaviors. The two Fe II lines that we investigatedalso have equivalent widths similar to that of Si II , butthe strengths of the two lines differ by only a factor of1.35, which weakens the COG test. Additionally, Fe II issubstantially more depleted than Si II . In our analysis,we therefore determine the most plausible range of b val-ues by prioritizing the constraints on b from S II owing toits benefits outlined above, and the fact that is generallyoffers the tightest constraints (narrowest b range). WhenS II is not available, we prefer Fe II over Mg II .In the example of Sk-675 (Figure 4), Mg II and Fe II in-dicate b = 10 ± − , while the strongest constraintscomes from S II with b = 10.75 ± − . For Si II ,the intersection of the innermost COG contour and theAOD column density measurement are consistent with b = 10.75 km s − , and the error bars on the AOD mea-surement are also consistent with the range of b valuesdetermined from S II , Mg II , and Fe II . From the loca-tion of this intersection below where the contours becomecurved, it is clear that there is some saturation of the line,but the AOD analysis seems to have handled it well.In the second example of BI 173 (Figure 5), the TABLE 3Equivalent widths and individual AOD and column density measurements for the STIS observations
Sight-line grating Element Wavelength log λf λ EW log( NX )˚A ˚A m˚A cm − BI 173 STIS-M OI 1355.598 -2.805 10.4 ± < ± +0 . − . BI 173 STIS-M MgII 1240.395 -0.355 44.7 ± +0 . − . BI 173 STIS-M SiII 1808.013 0.575 236.6 ± +0 . − . BI 173 STIS-M PII 1152.818 2.451 166.4 ± > ± +0 . − . BI 173 STIS-M SII 1250.578 0.809 183.2 ± +0 . − . BI 173 STIS-M SII · · · · · · · · · +0 . − . BI 173 STIS-M CrII 2056.254 2.326 118.2 ± +0 . − . BI 173 STIS-M CrII 2066.161 2.024 77.1 ± +0 . − . BI 173 STIS-M FeII 2260.780 0.742 157.2 ± +0 . − . BI 173 STIS-M FeII 2249.877 0.612 114.2 ± +0 . − . BI 173 STIS-M NiII 1370.132 1.906 86.0 ± +0 . − . BI 173 STIS-M NiII 1317.217 1.876 53.9 ± +0 . − . BI 173 STIS-M NiII 1741.549 1.871 79.4 ± +0 . − . BI 173 STIS-M NiII 1709.600 1.743 66.1 ± +0 . − . BI 173 STIS-M NiII 1751.910 1.686 79.1 ± < ± +0 . − . BI 173 STIS-M CuII 1358.773 2.569 17.9 ± +0 . − . BI 173 STIS-M ZnII 2026.136 3.106 153.8 ± +0 . − . BI 173 STIS-M ZnII 2062.664 2.804 119.8 ± +0 . − . BI 173 STIS-M ZnII · · · · · · · · · +0 . − . BI 173 STIS-M GeII 1237.059 3.033 8.1 ± < ± > ± +0 . − . SK-66 19 STIS-M CrII 2066.161 2.024 94.7 ± +0 . − . SK-66 19 STIS-M FeII 2260.780 0.742 161.0 ± +0 . − . SK-66 19 STIS-M FeII 2249.877 0.612 127.0 ± +0 . − . SK-66 19 STIS-M NiII 1741.549 1.871 37.5 ± < ± +0 . − . SK-66 19 STIS-M ZnII 2062.664 2.804 270.6 ± +0 . − . SK-66 19 STIS-M ZnII · · · · · · · · · +0 . − . SK-68 73 STIS-H OI 1355.598 -2.805 16.9 ± +0 . − . SK-68 73 STIS-H MgII 1240.395 -0.355 118.2 ± > ± > ± +0 . − . SK-68 73 STIS-H NiII 1317.217 1.876 77.8 ± +0 . − . SK-68 73 STIS-H CuII 1358.773 2.569 22.4 ± +0 . − . SK-68 73 STIS-M SiII 1808.013 0.575 234.5 ± > ± +0 . − . SK-68 73 STIS-M CrII 2066.161 2.024 69.4 ± +0 . − . SK-68 73 STIS-M FeII 2260.780 0.742 163.4 ± +0 . − . SK-68 73 STIS-M FeII 2249.877 0.612 132.8 ± +0 . − . SK-68 73 STIS-M NiII 1741.549 1.871 95.9 ± +0 . − . SK-68 73 STIS-M NiII 1709.600 1.743 45.8 ± +0 . − . SK-68 73 STIS-M NiII 1751.910 1.686 58.6 ± +0 . − . SK-68 73 STIS-M ZnII 2026.136 3.106 199.0 ± +0 . − . SK-68 73 STIS-M ZnII 2062.664 2.804 264.2 ± +0 . − . SK-68 73 STIS-M ZnII · · · · · · · · · +0 . − . Note . — The entirety of this table is available online in machine-readable format
TABLE 4Column density adjustments for Si II based on thehybrid AOD and COG approach Target N (Si II ) AOD N (Si II ) AOD+COG cm − cm − SK-67 2 15.63 +0 . − . > +0 . − . > +0 . − . +0 . − . SK-66 35 15.75 +0 . − . +0 . − . SK-65 22 15.58 +0 . − . +0 . − . SK-68 26 15.73 +0 . − . > +0 . − . > +0 . − . +0 . − . BI 184 15.80 +0 . − . +0 . − . SK-67 191 15.61 +0 . − . +0 . − . SK-67 211 15.74 +0 . − . +0 . − . SK-66 172 15.71 +0 . − . > +0 . − . +0 . − . SK-70 115 15.96 +0 . − . +0 . − . weighted average of the strong and weak Fe II lines givesa most plausible b value of 22 km s − , with a possiblerange b >
16 km s − . Mg II yields b > − . Theweak Mg II lines show smaller b values than the strongerFe II lines for the reasons outlined above. This can beunderstood in terms of a velocity profile that is narrow inthe central portion but then has broad wings in the lowerportions. Such a behavior will cause a shift to higher b values for stronger lines. The best constraints on b againarise from S II for this target, with a most plausible valueof 16 ± − . In this case, b = 16 ± − in theSi II COG corresponds to N = 15.98 ± − , whilethe column determined from the AOD is N = 15.78 ± − . In this case, saturation did therefore impact theSi II column density determination. In Table 3, we re-port N (Si II ) determined from the hybrid AOD/COGmethod, with upper and lower error bars of ± N is insensi-tive to changes in b .We proceed with this analysis for all sight-lines andfind that 14 out of 32 targets in the sample need adjust-ments to their AOD-derived Si II column densities. Thisincludes 5 targets for which I /I c > .
05 and therefore N (Si II ) was derived from the AOD, but the examina-tion of the COG contours revealed that only a lower limitcould be estimated. The results of these corrections arelisted in Table 4. Column densities with profile fitting in the COSspectra
The lower resolution of COS dictated the need forprofile fitting, in order to overcome the effects ofadditional smoothing of the absorption profiles. Weperformed such fitting to the COS G130M spectraand derive column densities of Mg II ( λ λ II ( λ λ − interval over the range in whichabsorption can be detected. Each component has acolumn density, central velocity (within its 10 km s − interval), and width. One might anticipate that reduc-ing the component spacing could lead to larger columndensities since smaller b values would be associatedwith more tightly spaced components. In AppendixA, we show that the column density measurementsobtained from profile fitting do not change beyond theiruncertainties when a shorter component spacing of 5 kms − is used, although a handful of sight-lines do havehigher (by +0.5-0.7 dex) column densities of Mg II , withcorrespondingly larger uncertainties, with the tightercomponent spacing.The modeled absorption features are multiplied by thecontinuum and convolved with the COS instrumentalline spread functions (LSF) in order to forward-modelthe observations, and allow for the correct propagationof uncertainties. We marginalize over the individualcomponent parameters using a custom implementationof the simplified Manifold Metropolis-adjusted Langevinalgorithm (Girolami & Calderhead 2011) and sum thecolumn densities of all of the components. The infer-ences of the Mg II , Ni II , and S II column densities areperformed independently. For each element, differentspectral lines observed in different exposures (e.g.,different FP-POS dithers) are fit simultaneously.We use samples drawn from the posterior probabilitiesusing MCMC to build posterior probability distributionsfor each species column density and the model fluxat each wavelength. The reported column densitiescorrespond to the 50 th percentile of the posteriordistribution. The resulting uncertainties, taken asthe difference between the 16 th and 50 th percentile(lower uncertainties) and between the 50 th and 84 th percentile (upper uncertainties) of the posterior dis-tribution, include uncertainties on the measurement(noise), continuum estimation, and the possibility ofobservationally similar but physically different velocitycomponent structures. Figure 6 provides an exampleof profile fitting of the Mg II ( λλ II ( λλ Gas–phase abundances and depletions
Gas-phase abundances are derived by taking the ratioof the measured column densities to the total hydrogencolumn density, N(H) = N(H i ) + 2N(H ), where N(H)is listed in Table 1. The H i column densities aredetermined from the METAL spectra (see Paper I),while the H column densities are from Welty et al.(2012). The depletion (logarithm of the fraction of Log N (cm -2 ) b ( k m / s ) Log N (cm -2 ) b ( k m / s ) b ( k m / s ) Log N (cm -2 ) Log N (cm -2 ) b ( k m / s ) Sk-67 5Mg IISk-67 5Fe II Sk-67 5S IISk-67 5Si II
Fig. 4.—
Outcome of the AOD and COG approaches to determine column densities of Mg II (top left), S II (top right), Fe II (bottomleft) and Si II (bottom right) toward Sk-67 5. Each panel shows the probability contours based on a χ distribution with two degrees offreedom for various combinations of log N (x-axis) and b (y-axis) for a standard COG, given the measured values of equivalent widths andtheir uncertainties for each of the two transitions used (e.g., λλ II , λλ II , λλ II ).The column density determined from the AOD method is also overlaid on the contours in blue (strong line) and cyan (weak line). For Fe II ,Mg II , and S II , two transitions are available to evaluate and correct for the effects of saturation, and the column densities determined fromthe AOD and COG are in agreement, and the intersection of the AOD-determined N and the COG contours provides a well-constrainedrange of b values, shown as gray lines. This value of b can then be applied to the COG contours obtained from the single Si II transition inorder to evaluate and correct for the effects of saturation. For this Sk-67 5 sight-line, the Si II COG contours and AOD-determined columndensity are in agreement for the b value derived from the other element. Therefore, saturation is not an issue for this sight-line and noadjustment to the AOD outcome is necessary. element X in the gas-phase) is then calculated assumingthat the total (gas and dust) neutral ISM abundanceof X is equal to the photospheric abundance of X inyoung stars. Because young stars recently formed outof the ISM and have not yet undergone self-enrichment,they are good proxies for the present-day ISM com-position. A number of studies have spectroscopicallyinvestigated the composition of luminous young starsin the LMC. However, no single study includes all theelements for which we wish to compute interstellardepletions. Tchernyshyov et al. (2015) carefully pooledmeasurements of young star abundances across studiesusing a multilevel linear model to account for differencesbetween studies and missing uncertainty information.In this work, we assume the LMC stellar abundancescompiled in Tchernyshyov et al. (2015) for the totalISM abundances. These reference abundances are listedin Table 2 for each element. The measurement errorson the depletions are obtained by summing the errors on the logarithms of the column densities of X andH in quadrature. Systematic errors on the depletionsdue to uncertainties on the photospheric abundancesare not included in Table 5, because they do not affectthe relative trends examined here (e.g., environmentalparameters). An estimate of these systematic errors canbe found in Table 4 of Tchernyshyov et al. (2015). HYDROGEN DENSITIES AND RADIATION FIELDSFROM THE C I FINE STRUCTURE LINES
The C I lines (C I at λ I ∗ at λ I ∗∗ at λ I ∗ )/N(C I ) tot and N(C I ∗∗ )/N(C I ) tot provide an es-0 Log N (cm -2 ) b ( k m / s ) b ( k m / s ) Log N (cm -2 ) Log N (cm -2 ) b ( k m / s ) BI 173S IILog N (cm -2 ) b ( k m / s ) BI 173Mg II
BI 173Fe II
BI 173Si II
Fig. 5.—
Same as Figure 4, but for the sight-line toward BI 173. In this case, the b value derived from the intersection of the AOD andCOG contours for Mg II , Fe II , and S II suggests that the column density of Si II derived from the AOD should be increased by +0.2 dex(solid gray line) and the error bars increased to − timate of the number density of hydrogen atoms, which,combined with estimate of the N(C II )/N(C I ) tot ratioand an iterative approach, yields an estimate of the in-tensity of the UV radiation field I normalized to the to avalue I specified by Mathis et al. (1983) for the averageintensity of ultraviolet starlight in the solar neighbor-hood. We applied a line profile fitting method to the C I lines in the METAL spectra and followed this approachto estimate n ( H ) and I/I in 26 out of 32 METAL sight-lines.We first derive column densities of N(C I ), N(C I ∗ ),and N(C I ∗∗ ) (summing up to N(C I ) tot ) using the sameprofile fitting method as described in Section 3.3 (seeexample in Figure 7). We also assume a 10 km s − com-ponent spacing for C I . As for Mg II and Ni II , we in-vestigate possible systematic differences in the N(C I ),N(C I ∗ ), and N(C I ∗∗ ) outcomes for a tighter componentspacing of 5 km s in Appendix A and find results withinuncertainties. The resulting column densities are listedin Table 6 and the ratios f = N(C I ∗ )/N(C I ) tot and f = N(C I ∗∗ )/N(C I ) tot are shown in Figure 8.As explained in Jenkins & Tripp (2001, 2011) and ref-erences therein, for a given kinetic temperature and radi-ation field intensity, the location of the ( f , f ) measure-ments follows a track that is dependent on the volume density of hydrogen, n ( H ) (or equivalently on pressurefor a given temperature), and the fraction of low-pressuregas, g low . Geometrically, the composite measurementof ( f , f ) toward a sight-line, which includes contribu-tions from high pressure gas with C I column density(1 − g low )N(C I ) tot and low pressure gas with C I col-umn density g low N(C I ) tot , corresponds to the center ofmass of the points ( f hp , lp1 , f hp , lp2 ) associated with thehigh and low pressure contributors, with the weights foreach component corresponding to the corresponding frac-tion of the C I column density they represent, (1- g low )and g low , respectively. The high pressure component isassumed to have ( f hp1 , f hp2 ) = (0.38, 0.49), as in Jenkins& Tripp (2011). We will discuss the effects of differentassumptions on the location of the high-pressure compo-nent further in this section. The low pressure componentfollows tracks that depend on density, temperature, andradiation field intensity (responsible for optical pumpingof the excited C I states). The model is identical to theone described in Jenkins & Tripp (2001), in particulartheir equations 10–12. This geometrical estimation ofthe low and high pressure components is illustrated inFigure 9, showing the ( f , f ) measurement for Sk-68135.By geometrically matching the ( f , f ) measurements1 LMC Mg II l l l l N o r m a li z e d F l u x ( A r b i t r a r y U n i t s ) LMC Ni II l l N o r m a li z e d F l u x ( A r b i t r a r y U n i t s ) - N(Mg II)N(Mg II) = 16.05 N(Ni II) = 14.18 - N(Ni II)
Fig. 6.—
Example of profile fitting of the Mg II ( λλ II ( λλ to the model tracks for the low pressure componentshown in Figures 8 and 9, and assuming that the kinetictemperature of the gas is equal to the rotational temper-ature of H , T (H ) reported in Welty et al. (2012), g low and n ( H ) can be derived, if the radiation field intensityfor the low-pressure is known. Fortunately, as describedin Jenkins & Tripp (2011, their Equation 1), an estimateof the ionized to neutral carbon ratio, N(C II )/N(C I ) tot ,can provide the necessary constraints on the radiationfield intensity. Because the calculation of the radiationfield intensity from Equation 1 of Jenkins & Tripp (2011)depends on the local hydrogen volume density n (H) (inaddition to N(C II )/N(C I ) tot ), and conversely, the deter-mination of n (H) from f and f requires the knowledgeof the radiation field intensity, an iterative approach is necessary to solve for both. We start the iterative cal-culation by assuming the standard radiation field inten-sity specified by Mathis et al. (1983) ( I/I = 1) in thecomputation of n (H) using the model shown in Figure8 and the ( f , f ) measurements. We can then applythis density to calculate the radiation field intensity fromN(C II )/N(C I ) tot , following the approach described laterin this section. From the density and radiation field, wederive the electron density n e (the procedure for derivingthe electron density is described in the next paragraph).These are the initial values for the radiation field inten-sity, hydrogen density, and electron density. Next, weproceed with the iterative approach below: • Compute an updated radiation field intensity
I/I using the density n (H) and electron density2 TABLE 5Column densities and depletions
Target log N(H) Element Grating log N(X) 12 + log(
X/H ) LMC , gas Depletion δ ( X ) a cm − cm − BI 173 21.25 ± ± ± ± ± ± ± ± ± ± ± ± ± < < < -0.42BI 173 21.25 ± ± ± ± ± ± ± ± ± < < < ± > > > -0.75BI 173 21.25 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± > > > -1.31SK-66 19 21.87 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± > > > -0.89SK-68 73 21.68 ± > > > -1.11SK-68 73 21.68 ± · · · . 12.52 ± ± ± ± ± ± ± Note . — This entirety of this table is available online in machine-readable format a The statistical uncertainty is listed here. Systematic errors on the depletions due to uncertainties on the photospheric abundances are not included,because they do not affect the relative trends examined here (e.g., environmental parameters). An estimate of these systematic errors can be foundin Table 4 of Tchernyshyov et al. (2015). n e from the previous iteration, as well as fixedmeasurements (N(C I ) tot , N(C II ), T), as inputs toEquation 1 of Jenkins & Tripp (2011). • Using this updated radiation field intensity asinput for the model tracks, derive an updateddensity for the low pressure component, n (H),from the observed ( f , f ) line ratios using thegeometrical approach described above. • Using the fixed temperature T (H ) and the up-dated I/I and n (H) values as inputs in Equation24 of Weingartner & Draine (2001), compute anupdated electron density n e (see next paragraphfor more details on this step). • Repeat this process using the radiation field inten-sity, density, and electron density output by an it-eration as input for the next iteration. Continueiterating on the radiation field intensity, density,and electron density until convergence is reached,i.e. when the difference between successive itera-tions is less than 5% on all parameters.During the iteration on radiation field and density, acomputation of the electron density is required. For this,we solve for the ionization equilibrium electron densityfollowing Equation 24 in Weingartner & Draine (2001),given the density and radiation field at each step of the iteration. The calculation also includes physical coef-ficients such as the ionization from metals of x e, =8.7 × − in the LMC (scaled by a factor of 1/2 to ac-count for the metallicity difference between the LMC andMilky Way), and the cosmic ray ionization rate in theLMC, found to be 30% (Abdo et al. 2010) of the MilkyWay value 2 × − s − (Indriolo et al. 2007; Neufeldet al. 2010) (although the results are largely insensitiveto the assumed cosmic ray ionization rate within a factorof a few of this value). The radiative plus di-electronicrecombination coefficients, α r (H) and the recombinationrates for different elements due to collisions with dustgrains, α g (X), are taken from Shull & van Steenberg(1982) and Weingartner & Draine (2001), respectively.We scale α g (X) down by a factor 1/3 to account forthe lower abundance of dust grains in the LMC. Whilethe LMC has half-solar metallicity, Roman-Duval et al.(2019) showed that the fraction of Si in the dust-phaseis a factor 1.5 in the LMC than in the MW, leading toa dust-to-gas ratio 3 times lower in the LMC than inthe Milky Way. The results described further in this pa-per confirm this result for other elements than Si. Sincethe Equation 24 of Weingartner & Draine (2001) used tocompute the electron density cannot be solved analyti-cally, we take an iterative approach.As explained above, we use the N(C II )/N(C I ) tot ratioto estimate the radiation field intensity, using the den-sity and electron density at each step of the iterationas inputs to Jenkins & Tripp (Equation 1 of 2011). In3 TABLE 6Measurements of the C I , C I *, and C I ** column densities and ratios, derived radiation fields, volume densities, andelectron densities Target Instr. T ( H ) a log N(C I ) tot log N(C II ) f f I/I n (H) g low n e K cm − cm − cm − cm − SK-67 2 STIS 46.0 14.73 ± ± ± ± ± ±
37 0.86 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± · · · · · · · · · · · · · · · · · · · · · · · · · · · SK-66 19 COS 77.0 14.37 ± ± ± ± ± ±
74 0.43 ± ± ± ± ± ± ± ±
31 0.84 ± ± ± ± ± ± ± ±
19 0.59 ± ± ± ± ± ± ± ±
23 0.63 ± ± ± ± ± ± ± ±
48 0.37 ± ± ± ± ± ± ± ±
95 1.00 ± ± ± ± ± ± ± ±
34 0.67 ± ± ± ± ± ± ± ±
180 0.75 ± ± · · · · · · · · · · · · · · · · · · · · · · · · · · · SK-68 73 STIS 57.0 14.49 ± ± ± ± ± ±
44 0.70 ± ± · · · · · · · · · · · · · · · · · · · · · · · · · · · SK-67 105 STIS 62.0 14.09 ± ± ± ± ± ±
129 1.00 ± ± ± ± ± ± ± ±
25 0.50 ± ± ± ± ± ± ± ±
180 0.86 ± ± ± ± ± ± ± ±
24 0.62 ± ± · · · · · · · · · · · · · · · · · · · · · · · · · · · SK-67 191 · · · · · · · · · · · · · · · · · · · · · · · · · · ·
SK-67 211 · · · · · · · · · · · · · · · · · · · · · · · · · · ·
BI 237 COS 61.0 14.04 ± ± ± ± ± ±
109 0.99 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
72 1.00 ± ± ± ± ± ± ± ±
138 1.00 ± ± ± ± ± ± ± ±
14 0.87 ± ± ± ± ± ± ± ±
18 0.92 ± ± ± ± ± ± ± ±
29 1.00 ± ± ± ± ± ± ± ±
40 1.00 ± ± ± ± ± ± ± ±
26 0.60 ± ± ± ± ± ± ± ±
25 0.95 ± ± a Rotational temperatures T , computed as in Equation 5 of Tumlinson et al. (2002), are taken from Welty et al. (2012) and references therein.For Sk-66 172, the iterative computation of density and radiation field from C I and C II line ratios diverges with the H T rotational temperaturegiven in Welty et al. (2012) (41 +110 − K) input to the model as the kinetic temperature. The closest temperatures for which the models converge is110K, which are well within the error bars of the temperature estimation. the METAL sample of sight-lines, the C II λ λ II column density with the following procedure. First,we scale the hydrogen column density N ( H ) for eachsight-line by the carbon abundance in the LMC (12 +log(O/H) = 7.94, see Table 2), and obtain an estimate ofthe total carbon column density in the ISM (gas + dust).Second, we compute the carbon depletion correspondingto the measured iron depletion for each sight-line, usingthe coefficients presented in Jenkins (2009). We appliedthis depletion value to the total carbon column density,thus yielding an estimate of the gas-phase carbon columndensity for each sight-line. Since the C I column densityfor the range of hydrogen column densities probed by theMETAL sight-lines is negligible compared to the C II col-umn density, we can then estimate N(C I ) tot /N(C II ) (cid:39) N(C I ) tot /N(C). Other ways to estimate N(C II ) fromthe measured S II and Mg II column densities are ex-amined in Appendix A, but yield very similar results tothe method used here. Furthermore, N(H) is measuredfor all targets, unlike N(Mg II ) and N(S II ), giving thisapproach a fundamental advantage.In most cases, the H rotational temperature T wasreported in Welty et al. (2012). When this was not the case, we assumed the median value of the Welty et al.(2012) sample, or 80 K. For Sk-66 172, the iterative com-putation of density and radiation field from C I and C II line ratios oscillates between two very different but notvery physical models (very low radiation field, very highdensity and vice-versa), given the H T rotational tem-perature given in Welty et al. (2012) (41 +110 − K) input tothe model as the kinetic temperature. The closest tem-peratures for which the models converge is 110 K, whichare well within the error bars of the temperature estima-tion. We assumed these temperatures in the modelingand report these values in Table 6.To propagate the errors on f and f on the density andradiation field intensity determination, we use a Monte-Carlo approach. We draw sample of ( f , f ) within Gaus-sian distributions of standard deviation equal to the 1 σ error on f and f , and repeat the procedure to compute n ( H ) and I/I for each draw. The error on these pa-rameters is then taken as the standard deviation of theresulting distributions. The errors on n ( H ) and I/I arereported in Table 6.We have explored the systematic effect of assuming adifferent ( f hp1 , f hp2 ) location for the high-pressure com-ponent on the resulting density derivations. Given thegeometrical set-up of the approach, the density of the4 N(C I) tot = 14.37±0.72N(C I) tot
LMC C I l l N o r m a li z e d F l u x ( A r b i t r a r y U n i t s ) LMC C I l Fig. 7.—
Example of profile fitting of the C I band system in the COS spectrum of BI 184. The left panel shows the spectrum nearthe lines of interest (scaled by a constant), and different realizations of the continuum (gray) and absorption (orange) models sampling theposterior probability density functions shown in the right panel. N ( CI * )/ N ( CI ) tot N ( C I ** ) / N ( C I ) t o t T = 55.0 KT = 75.0 KT = 95.0 KT = 115.0 KT = 250.0 KHigh pressure gas
Fig. 8.—
Distribution of line ratios f = N(C I ∗∗ )/N(C I ) tot versus f = N(C I ∗ )/N(C I ) tot for the sight-lines in the METALprogram. The model tracks for the low pressure gas are computedfor several temperatures indicated in the legend, and the logarithmof the densities along the tracks are indicated on the figure. low-pressure component is very weakly dependent on f hp2 (only g low would change in this case). Given a dis-placement ∆ f hp1 of the assumed location of the high-pressure component, and using the Thales theorem, ∆ f lp1 = g high /g low ∆ f hp1 = (1- g low ) /g low ∆ f hp1 . For all our tar-gets, g low < f lp1 < ∆ f hp1 . Furthermore,the location of the high-pressure component cannot slidetoo far off the high-density end of the model tracks. Thefiducial value we assume, ( f hp1 , f hp2 ) = (0.38, 0.49) (as f1 = N(CI*)/N(CI) tot f = N ( C I ** ) / N ( C I ) t o t g low High pressure gasMeasurementLow pressure gas
Fig. 9.—
Example of modeling and geometrical derivation of thelow ( g low ) and high (1 − g low ) pressure gas components fractionsfrom the C I fine structure line ratios, f and f , for the Sk-68 135sight-line. The black tracks represent a model of the variationsof the ( f , f ) distribution for a fixed temperature (equal to the H rotational temperature for this target, or 91 K), as the volumedensity changes (log n(H) values are indicated next to each blackpoint). The red point with error bars is the ( f , f ) measurement,which corresponds to the geometric average of the high (pink starat f , f = 0.38, 0.49) and the low pressure component, located atthe intersection of the line joining the high pressure component andthe measurement, and the model track, thus yielding the densityof the low pressure component. in Jenkins & Tripp 2011), corresponds to n ( H ) ∼ cm − and T ∼
250 K as seen in the gray model tracks inFigure 8. The model for densities > cm − cm − f , f ) = (0.36, 0.51). For lower temperaturemodels, the extremity of the model tracks would lie ontop of the 10 cm − point as well. For T ∼
100 K, thismeans the high-pressure point is at (0.39, 0.46). Thus,for a reasonable temperature, the high-pressure compo-nent is constrained between f = 0.36 and 0.39, meaningonly a maximum displacement of ∆ f hp1 = 0.03 is possi-ble. Given the high ratio of g low / g high , this implies thatthe effects of moving the location of the high-pressurepoint within reasonable model bounds has a negligibleeffect on the outcome. We have verified this numericallyby recomputing densities using ( f hp1 , f hp2 ) = (0.36, 0.51)and (0.39, 0.46) and found that the resulting differencesin density determinations were typically lower than the1 σ statistical error on n(H).Finally, we note that C I toward 3 of the sight-lines(SK-70 115, SK-68 73, and SK-67 5) was previously ana-lyzed by Welty et al. (2016, (W16)). Our measurementsof N (C I ) tot , f , and f are in reasonable agreementwith those in Welty et al. (2016). For SK-67 5, the val-ues compare as follows: N (C I ) tot = 13.75 ± ± f = 0.28 ± ± f = 0.09 ± ± N (C I ) tot = 14.29 ± ± f = 0.45 ± ± f =0.27 ± ± N (C I ) tot = 13.7 ± ± f = 0.36 ± ± f = 0.16 ± ± I differ be-yond the uncertainties, but not by a large amount, whilethe f and f values, which are the driving parameters inthe density derivation, are within 1 σ . There are differ-ences in the assumptions for the estimation of N (C II ),and possibly in the modeling of the C I line ratios, leadingto difference in the output parameters. We derive n (H)= 101 ±
6, 268 ±
44, and 310 ±
25 cm − for the 3 sight-lineswhile W16 obtain n (H) = 119, 1919, and 212 cm − (nouncertainties reported). Densities for SK-67 5 and SK-70 116 are roughly consistent within our uncertaintiesbetween the two studies, but densities toward SK-68 73differ by a large factor (beyond our uncertainties, butnot necessarily beyond the W16 uncertainties, which arenot reported). This is due to the location of SK-68 73in the ( f , f ) diagram, near the curve where change of5% in f can lead to a large change in density. Moregenerally, this difference indicates densities derived fromC I integrated along the entire line-of-sight may not bephysically meaningful when the physical conditions showdrastic changes from one component to the next. Finally,for the radiation fields, we find I/I = 2.3 ± ± ± THE CORRELATION OF DEPLETIONS BETWEENDIFFERENT ELEMENTS
As pointed out by Jenkins (2009) in the Milky Way,depletions for different elements correlate well with eachother, indicating that the depletion process is a collectiveone. We investigate these correlations in the LMC usingthe METAL data in Figure 10. As expected, we alsofind tight correlations between the depletions of differentelements and that of iron. Following Jenkins (2009) andJenkins & Wallerstein (2017), we fit those correlations with linear functions of the form δ ( X ) = a ( X ) ( δ ( F e ) − z ( X )) + b ( X ) (2)where a ( x ) and b ( X ) (slope and intercept of the relationbetween depletions of element X and Fe depletions) arefitted for, and where z ( X ) = (cid:80) los δ ( F e ) σ ( δ ( X )) (cid:80) los σ ( δ ( X )) (3)Here σ ( δ )(X) is the error on δ (X). Introducing the zero-point reference z ( X ) in δ ( F e ) has the benefit to re-duce the covariance between the formal fitting errorsin a ( X ) and b ( X ) to near zero, as explained in Jenkins(2009). The resulting parameters and their uncertaintiesare listed in Table 7, where we also list the p-value andcorrelation coefficients for this relation for each element.The correlation coefficients can be artificially enhancedby the covariant errors on δ (Fe) and δ (X), through com-mon errors on N(H). We account for this following themethod described in Jenkins et al. (Appendix B of 1986).The ”corrected” correlation coefficient is also listed inTable 7. Accounting for covariant errors only marginallyreduces the correlation coefficient, which is > δ ( X ) - δ ( F e ) relation as a func-tion of condensation temperature of element X, and thezero-point of this relation (at δ ( F e ) = − . − − α systems (DLAs), in which S and Znare often used as metallicity tracers.Given the tight correlations between depletions of dif-ferent elements, Jenkins (2009) introduced the parameter F ∗ to describe the collective advancement of the deple-tion process in the Milky Way. The depletion levels ofelement X is then modeled by a linear relation with F ∗ ,of the form δ ( X ) = A X × ( F ∗ − z X ) + B X . F ∗ = 0corresponds to the most lightly depleted sight-lines inthe Milky Way with log N(H) > , where ion-ization corrections are negligible, and F ∗ =1 correspondsto the well studied heavily depleted velocity componenttoward ζ Oph. Since the F ∗ scale is tied to the particularsight-lines used to anchor the F ∗ = 0 and 1 extremes, acomparison of depletion patterns using the F ∗ parameterin other galaxies requires one to use the same normaliza-tions for F ∗ . Therefore, similar to the computation of F ∗ in SMC by Jenkins & Wallerstein (2017), the F ∗ param-eter in the LMC is given by: F ∗ ( LM C ) = δ ( F e ) − B F e A F e + z F e (4)where A F e = − .
285 , B F e = − .
513 , and z F e = 0.437are the coefficients of the linear relation between δ ( F e )and F ∗ in the Milky Way given in Table 4 of Jenk-ins (2009). Thus, F ∗ in galaxies other than the MilkyWay corresponds to a scaling of iron depletions, whichmakes it easier to compare depletion patterns to thosein the Milky Way. The top axis of Figure 10 showsthe F ∗ scale in the LMC. Iron was chosen as a proxyfor F ∗ due to its abundance of spectral lines of differentoscillator strengths, allowing straight-forward measure-ments of column densities, abundances, and depletions(all METAL sight-lines have an iron depletion determi-nation). VARIATIONS OF DEPLETIONS WITH LOCALENVIRONMENT
In this Section, we explore the environmental pa-rameters driving variations of the interstellar depletionsand subsequently dust-to-metal ratio within the LMC.Thanks to the tight correlation between depletions ofiron and other elements, we can explore the variationsof depletions as a function of environment using iron asa representative element. We examine the correlationsbetween iron depletions and hydrogen column density(N(H)), H fraction (f(H )), hydrogen volume density(n(H)), radiation field intensity ( I/I ), and distance fromvarious landmarks in the LMC, in particular to its cen-ter, located about 1 kpc West of the 30 Dor massivestar-forming region. Multi-linear regression of iron depletions versuslocal environment parameters
In order to determine which environmental parametersamongst N(H), n(H),
I/I , f ( H ) drive the variations ofdepletions and dust-to-metal ratio, we first perform amulti-linear regression of the iron depletions ( δ ( F e )) asa function of combinations of 3 of these parameters. Wecannot perform a robust multi-linear regression analysison more parameters in one instance with only 32 mea-surements. The covariance of errors in the depletions (recall δ ( X ) ∝ N ( X ) /N ( H )) with N ( H ) and f ( H ) canstrengthen or weaken the inferred correlation betweenparameters. To account for this, we use the mlinmix err IDL package developed by Kelly (2007), which uses aBayesian approach to multi-linear regression, accountingfor errors and covariances between dependent and inde-pendent variables.We explore the multi-linear correlations between δ ( F e )and log N(H), f ( H ) and n ( H ) in Figure 12, which showsthe distributions of the slopes of the iron depletions ver-sus each parameter, α log N ( H ) (Fe), α log f ( H (Fe), and α log n ( H ) (Fe), such that: δ (Fe) = α log N(H) (Fe) log N(H)+ α log f (H ) (Fe) log f (H )+ α log n(H) (Fe) log n(H) + β (Fe) (5)We find α log N ( H ) (Fe) = − . ± . α log f ( H (Fe) =0.01 ± α log n ( H ) (Fe) = − ± f ( H ) in the regime probed by the METALsight-lines (3 × − — 0.62 with a median value of 0.03).There is a secondary marginal anti-correlation with hy-drogen volume density, as traced by the C I gas. In the-ory, one would expect a strong correlation between deple-tions and volume density, because dust growth timescalesare inversely proportional to density. However, since C II is the dominant form of carbon in the neutral translucentISM probed by this spectroscopic program, the C I gasmay represent a small mass and volume fraction of thegas traced by our sight-lines, and thus the density inthe C I gas may not be representative of the mean den-sity along the line of sight, traced by other metals andH i . Rather, since we are viewing the LMC nearly faceon, variations in the path length are effectively drivenby any changes in the scale height of the gas perpendic-ular to the plane of the LMC. The magnitudes of suchvariations are probably small compared to the variabilityof N(H) in our sample. Hence, N(H) should be a goodproxy for the average n(H) over the entire line of sight toa star embedded near the plane of the LMC, explainingthe resulting strong correlation with depletions.In Figure 13, we perform a multi-linear regres-sion with the combination of log N ( H ), log f ( H ),and I/I . In this case, we find α log N ( H ) (Fe)= − ± α log f ( H (Fe) = 0.00 ± α log I/I (Fe) = − ± fraction, at least in the parameterspace probed by the METAL survey. We note however,that the radiation field probed by the C I gas, as for thedensity, may not be representative of the radiation fieldilluminating the gas along the entire line of sight, sincethe C I gas is associated with a small fraction of the H i . Depletions vs hydrogen column density
Because timescales for accretion of gas-phase metalsonto dust grains become shorter as density increases(Asano et al. 2013; Zhukovska et al. 2016), it is expectedthat the fraction of metals in the gas (i.e., depletion)7 ( X ) Si Mg Ni Cu -3 -2 -1 0 (Fe) ( X ) S -3 -2 -1 0 (Fe) Zn -3 -2 -1 0 (Fe) Ti -3 -2 -1 0 (Fe) Cr -1.00.01.0 F * -1.00.01.0 F * -1.00.01.0 F * -1.00.01.0 F * Fig. 10.—
Depletions (log fraction in the gas-phase) for Si, Mg, Ni, Cu, S, Zn, Ti, Cr, as a function of iron depletions ( δ (Fe)). Measurementsmade with COS at slightly lower resolution are outlined in magenta. The fits of the depletions are shown by the gray lines, the transparencyof which is equal to the square-root of the probability of a given fit. TABLE 7Fits of depletions vs iron depletion δ (Fe) Elements a X σ ( a X ) b X σ ( b X ) a z X r b r c c p − value STD d Si 0.871 0.093 -0.68 0.030 -1.27 0.85 0.80 2.56e-07 0.11Ni 1.006 0.059 -1.26 0.017 -1.38 0.98 0.98 3.84e-21 0.04Mg 0.472 0.087 -0.50 0.024 -1.47 0.74 0.72 1.07e-05 0.08Cu 0.900 0.330 -0.44 0.094 -1.37 0.88 0.88 1.21e-01 0.34Cr 0.925 0.061 -1.13 0.016 -1.42 0.87 0.86 3.46e-09 0.09Zn 0.567 0.056 -0.36 0.015 -1.41 0.66 0.56 5.09e-05 0.10S 0.801 0.075 -0.31 0.025 -1.13 0.93 0.91 3.00e-08 0.08Ti 1.156 0.120 -1.63 0.023 -1.46 0.99 0.99 2.08e-07 0.07 a Systematic errors on b X due to uncertainties on the photospheric abundances are not included, because they do not affect the relative trendsexamined here (e.g., environmental parameters). An estimate of these systematic errors can be found in Table 4 of Tchernyshyov et al. (2015). b Correlation coefficient c Correlation coefficient corrected for covariant errors (through the log N(H) dependence of δ (Fe) and δ (X)) following Jenkins et al. (Appendix Bof 1986) d Standard deviation of the measurements about the fit decreases with increasing hydrogen column density orvolume density (they are related). Section 6.1 demon-strates that the strongest correlation between depletionsand environment is with hydrogen column density, evenaccounting for the covariance between depletions and logN(H). Such a trend has also been observed in the MilkyWay (Wakker & Mathis 2000; Jenkins 2009) for the fullsuite of the major components of the ISM, and Magel-lanic Clouds (Tchernyshyov et al. 2015; Roman-Duvalet al. 2019), albeit for a limited range of elements.Here, we quantify the correlation between depletionsand log N(H) for all elements probed by the METAL spectra. In Figure 14, the depletions toward the 32METAL sight-lines in the LMC decrease with increas-ing hydrogen column density for all elements probed bythe survey: Si, Fe, Mg, O, Ni, Cu, S, Zn, and Cr. TheTi depletions are taken from Welty & Crowther (2010),an optical spectroscopic study of a large number of LMCsight-lines. We fit the relation between elemental de-pletions and log N ( H ) taking into account the errors inlog N ( H ) and depletions, and using a linear function ofthe form:8 a ( X ) Si NiMgCu CrZnS Ti600 800 1000 1200 1400 160050% Condensation Temperature (K)1.51.00.5 ( X )(( F e ) = . ) Si NiMgCu CrZnS Ti
Fig. 11.—
Slope a (top) and zero-point (bottom) of the rela-tion between depletion of an element X, δ ( X ), and iron depletion, δ ( F e ), as a function of the 50% condensation temperature. Thezero-point of the δ ( X ) - δ ( F e ) relation is computed at δ ( F e ) = − . TABLE 850% CondensationTemperatures
Elements T c O 182Mg 1336Si 1310S 664Ti 1582Cr 1296Fe 1334Ni 1353Cu 1037Zn 726
Note . — 50% conden-sation temperatures arefrom Lodders (2003) δ ( X ) = a H ( X ) (log N ( H ) − z H ( X )) + b H ( X ) (6)where a H ( x ) and b H ( X ) (slope and intercept of the rela-tion between depletions of element X and log N(H)) arefitted for, and where z H ( X ) = (cid:80) los log N ( H ) σ ( δ ( X )) (cid:80) los σ ( δ ( X )) (7)Again, introducing the zero-point reference z H ( X ) in logN(H) has the benefit to reduce the covariance betweenthe formal fitting errors in a H ( X ) and b H ( X ) to nearzero. As pointed out by Jenkins (2009), introducing the z H ( X ) parameter practically removes the covariance inthe errors on the fitted solutions a H ( X ) and b H ( X ). Thebest-fit parameters for each element, as well as the p and r values, are listed in Table 9. The tight correlations seen in Figure 14 are reflected in the r values, whichrange from − − p < × − ). What other parameter(s) drives depletionvariations?
While hydrogen column density appears to be a maindriver of the depletion levels, and depletions do not ap-pear to correlate with volume density, radiation fieldintensity, or H fraction, the residuals of the δ ( X )—log N ( H ) for different elements X correlate strongly witheach other, as seen in Figure 15. This indicates a phys-ical origin for these residuals and subsequently a sec-ondary correlation with another parameter other thanvolume density, H fraction, or radiation field intensity.We note that the residuals of the δ ( X )—log N ( H ) corre-lation could also be caused by variations of similar mag-nitude of the true total metallicity of the ISM (gas +dust).We investigated possible correlations of these residu-als with tracers of interstellar shocks, star-formation andfeedback, such as the distance to the closet supernova(SN) remnant (Temim et al. 2015), the H α surface bright-ness from the SHASSA survey at 1 (cid:48) (15 pc) resolution(Gaustad et al. 2001), 24 µ m surface brightness in theSAGE Spitzer survey of the LMC (Meixner et al. 2006)at 6 (cid:48)(cid:48) resolution (1.5 pc)), and the dust temperature at36 (cid:48)(cid:48) (10 pc) resolution (Gordon et al. 2014). We found nocorrelation between the residuals of the δ ( X )—log N ( H )correlation and any of these parameters.Ultimately, we uncovered a significant correlation be-tween the residuals of the δ ( X )—log N ( H ) correlationand distance to the center of the LMC, located at RA =82.25 ◦ and DEC = − ◦ , about 1 kpc to the West of30 Doradus. The relation is shown in Figure 16 for iron.The correlation coefficient is − α log N ( H ) (Fe) = − ± α log n ( H ) (Fe) = − ± α d center (Fe) = − ± α log N ( H ) (X), α log n ( H ) (X), α d center (X) for all elements X, as well asthe intercept of the multilinear correlation, β . The re-sults are listed in Table 10. All elements exhibit similaranti-correlations between their depletions, the hydrogenvolume density, and the distance to the LMC center.The anti-correlations between depletions, hydrogencolumn density and volume density are expected if metalsaccrete onto dust grains in the ISM, since the timescalefor accretion is inversely proportional to density (andsubsequently column density). The anti-correlation withdistance to the LMC center is a relatively surprising find-ing, which could result from two effects. The first is apossible metallicity gradient. We assume constant total9 logN ( H ) = 0.69 +0.170.18 . . . . . l o g f ( H ) logf ( H ) = 0.03 +0.040.04 . . . . . logN ( H ) . . . . . l o g n ( H ) . . . . . logf ( H ) . . . . . logn ( H ) logn ( H ) = 0.23 +0.150.14 Fig. 12.—
Distributions of the slopes of the iron depletions versus the logarithm of the hydrogen column density, α logN ( H ) , the logarithmof the H fraction, α f ( H , and the hydrogen volume density, α logn ( H ) , such that δ ( F e ) = α logN ( H ) log N ( H ) + α f ( H f ( H ) + α n ( H ) log n ( H ) + β (Fe) abundances for the ISM based on the lack of observedgradient in > ii regions (Toribio San Ciprianoet al. 2017, and references therein). If a metallicity gra-dient did exist in young (a few tens of million years old)stars, it would give the appearance of a gradient in thedepletions, as observed here. The second effect possiblycausing the observed negative radial gradient in deple-tions is dust processing (formation and destruction). Inthis case, metals are less depleted from the gas-phasenear the LMC center and 30 Dor, and more depleted into dust grains away from 30 Dor.To get more insight into which effect might be at play,we map out the residuals of the main trend between irondepletions and hydrogen column density relative to theLMC gas disk in Figure 18. There is a clear gradientin gas-phase metallicity (and/or depletions) from Eastto West. Sight-lines on the East (left) side of the LMCcenter, along the H i filament associated with the South-East H i overdensity (Nidever et al. 2008; Mastropietroet al. 2009), have positive residuals up to +0.3 dex, i.e.,higher gas-phase metallicities for their H i column den-0 logN ( H ) = 0.64 +0.200.20 . . . . . l o g f ( H ) logf ( H ) = 0.00 +0.040.04 . . . . logN ( H ) . . . . . I / I . . . . . logf ( H ) . . . . . I / I I / I = 0.01 +0.010.01 Fig. 13.—
Distributions of the slopes of the iron depletions versus the logarithm of the hydrogen column density, α logN ( H ) (Fe), thelogarithm of the H fraction, α f ( H (Fe), and the radiation field intensity, α I/I (Fe), such that δ ( F e ) = α logN ( H ) (Fe) log N ( H ) + α f ( H (Fe) f ( H ) + α I/I (Fe) I/I + β ( F e ) sity than the fiducial trend. Conversely, sight-lines on theWest (right) side of the LMC center has negative residu-als (down to − i column density.Stars and H ii regions in the LMC do no exhibit a sim-ilar pattern in metallicity variations. While a shallow( − ± −
1) radial metallicity gradient,similar to the one we derive, is observed in 1-2 Gyr oldAGB stars (Cioni 2009, and references therein). Themetallicity gradient for red giants in clusters in the diskis also negligible (Grocholski et al. 2006a,b). Further- more, abundances in H ii regions show either a mild, notstatistically significant gradient ( − − Pagelet al. 1978) or no gradient at all (Toribio San Ciprianoet al. 2017), albeit with sparse samples (11 and 4 H ii re-gions respectively). Similarly, measurements in OB starsin N11 and NGC 2004 (Trundle et al. 2007), as well as 30Doradus (Markova et al. 2020) show very similar abun-dances, within errors.On the other hand, one could argue that the increasedturbulence and feedback from active star-formation onthe East side of the LMC, compressed as a result of the1 ( X ) a = 0.61 ± 0.069 b = 0.67 ± 0.028 z =21.04r=-0.77 ;p=3.1E-05 Si a = 0.71 ± 0.029 b = 1.39 ± 0.011 z =21.29r=-0.85 ;p=1.7E-09 Fe a = 0.32 ± 0.085 b = 0.50 ± 0.023 z =21.37r=-0.40 ;p=3.6E-02 Mg ( X ) a = 0.71 ± 0.040 b = 1.26 ± 0.014 z =21.32r=-0.82 ;p=2.8E-08 Ni a = 0.88 ± 0.296 b = 0.44 ± 0.093 z =21.35r=-0.99 ;p=1.4E-02 Cu a = 0.64 ± 0.056 b = 0.32 ± 0.021 z =20.84r=-0.88 ;p=3.6E-06 S
19 20 21 22 log N(H) ( X ) a = 0.43 ± 0.043 b = 0.36 ± 0.014 z =21.30r=-0.55 ;p=1.6E-03 Zn
19 20 21 22 log N(H) a = 0.74 ± 0.065 b = 1.63 ± 0.017 z =21.43r=-0.74 ;p=1.4E-02 Ti
19 20 21 22 log N(H) a = 0.65 ± 0.042 b = 1.14 ± 0.014 z =21.38r=-0.77 ;p=3.1E-06 Cr Fig. 14.—
Depletions (log fraction in the gas-phase) for Si, Fe, Mg, O, Ni, Cu, S, Zn, Ti, Cr, as a function of log N(H), where N(H)is from Roman-Duval et al. (2019) and includes both H i and H . Measurements made with COS at slightly lower resolution than STISare outlined in magenta. The a H , b H , z H parameters of the best-fit correlations, as well as the correlation coefficient r and p -value areindicated in each panel. collision with the SMC (Tsuge et al. 2019), might resultin increased dust destruction rates returning metals tothe gas-phase faster than in the quiescent, trailing Westside of the LMC. This would result in a gas-phase metal-licity gradient, with the East side being more metal richthan the West side, while the total metallicity of theLMC remains uniform. This would be consistent withthe finding reported in Tsuge et al. (2019) that the gas-to-dust ratio is 30% higher on the East side than the Westside of the LMC. While Tsuge et al. (2019) interpret thisresult as an indication that the East side is more metalpoor as a result of the mixing with the lower metallic-ity SMC gas, the observed reduced dust abundance onthe East side might actually result from dust process-ing. The gas-phase metal enhancement observed in thisstudy on the East side compared to the West side of theLMC would be a direct consequence from metals beingreturned from the dust to the gas-phase as a result of this increased large-scale shock-induced dust processing.Metallicity gradient or variations in dust processing?The jury is still out. Given the sparse spatial coverage ofH ii regions and OB stars on the face of the LMC, it ispossible that metallicity variations such as those shownin Figure 18 might have been missed. Stellar abundancesin the 100+ LMC young massive stars from the ULLY-SES HST Director’s discretionary program will providean essential dataset to test whether the metallicity ofstars formed out of the ISM in the last few millions yearsexhibits spatial variations similar to the ones observed inthe neutral gas.If we are indeed looking at East-West variations inthe metallicity of the ISM (dust and gas), models ac-counting for the dynamical interaction between the LMC,SMC, and the MW halo in their predictions for the star-formation history and chemical enrichment of the LMCwill be required. A possible (albeit admittedly highly2 TABLE 9Fits of depletions vs log
N(H)
Elements a H b H a z H r b p − value STD c Fe -0.711 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± a The first and second uncertainties reported represent the statistical error and systematic error on the photospheric abundances of young stars(see Table 4 in Tchernyshyov et al. (2015)) b Correlation coefficient c Standard deviation of the measurements about the fit ( Ni ) - ( Ni ) fit ( F e ) - ( F e ) f i t Fig. 15.—
Residuals of the fitted correlation between iron deple-tion and log N(H) ( δ (Fe) - δ (Fe) fit ) versus residuals of the fit-ted correlation between nickel depletions and logN(H) ( δ (Ni) - δ (Ni) fit ). The residuals are computed as δ (X) - δ (X) fit = δ (X) − a H (X)(log N(H) − z H (X)) − b H (X). The residuals are not ran-dom noise: residuals for all elements correlate with each other,indicating a physical origin to this secondary depletion variation.The red dashed line indicates a 1:1 correlation. speculative) explanation is that intense star formationoccurred in the last few tens of millions years ago nearthe leading edge of the LMC, which is compressed byram pressure from the MW halo. This star-formationcould have induced fountains of chemically enriched gas(Bustard et al. 2018), which would then be ejected highenough above the clockwise rotating disk to be swept bythe ∼
250 km s − motion of the LMC through the MWhalo, flowing south along the H i filament, before con-densing back into the disk. This process can occur on50-100 Myr timescales (Kim & Ostriker 2018), recentlyenough to affect the metallicity of young stars and theISM. GAS-TO-DUST RATIO FROM DEPLETIONS ANDVARIATIONS WITH HYDROGEN COLUMN DENSITY
With depletion measurements in hand, we can estimatehow the dust-to-gas ratio (D/H) varies with environmentin the LMC. Note that we do not include the contributionof helium here, which would increase the gas mass by 36%
Distance to LMC center [kpc] ( F e ) - ( F e ) f i t r=-0.3, p=1.09e-01(-0.087±0.000)×d(center) + 0.168±0.001 Fig. 16.—
Residuals of the correlation between iron depletion andlog N(H) ( δ (Fe) - δ (Fe) fit ) versus distance to the LMC center. Theresiduals are computed as δ (Fe) - δ (Fe) fit = δ (Fe) − a H (Fe)(logN(H) − z H (Fe)) − b H (Fe). from the hydrogen mass (G/D = 1.36 H/D). Since thedepletions correlate in the strongest way with hydrogencolumn density, we focus here on the variations of D/Hwith N(H). The dust-to-gas ratio, given by Equation 8,can be obtained from the fractions of each element in thegas (10 δ ( X ) ) and dust (1 − δ ( X ) ) phases. D/H = (cid:88) X (cid:16) − δ ( X ) (cid:17) A ( X ) W ( X ) (8)where A(X) is the abundance of element X given in Table2 and W(X) is the atomic weight of element X.The METAL survey provided interstellar depletionsfor Mg, Si, S, Fe, Ni, Cr, Cu, Zn, with Ti from Welty& Crowther (2010). METAL only yielded 4 measure-ments of oxygen depletions, which are not sufficient toconstrain variations with column density. The carbon λ λ logN ( H ) (Fe) = 0.82 +0.150.15 . . . . l o g n ( H ) ( F e ) logn ( H ) (Fe) = 0.13 +0.110.11 . . . . logN ( H ) (Fe) . . . . . d c e n t e r ( F e ) . . . . logn ( H ) (Fe) . . . . . d center (Fe) d center (Fe) = 0.10 +0.040.04 Fig. 17.—
Distributions of the slopes of the iron depletions versus the logarithm of the hydrogen column density, α logN ( H ) , the hydrogenvolume density, α logn ( H ) , and the distance to the LMC center, α d ( center ) , such that δ ( F e ) = α logN ( H ) log N ( H ) + α n ( H ) log n ( H ) + α d ( center d ( center ) + β in our estimation of D/H as a function of hydrogen col-umn density, we assume that the depletions of C andO as a function of iron depletions behave similarly tothose in the Milky Way, where the weak C and O linesare detected and interstellar depletions can be measured.Figure 3 of Jenkins & Wallerstein (2017) shows that therelation between iron depletions and those of other ele-ments probed by METAL only very weakly vary betweenthe MW and SMC, so this is the best assumption we canmake given the data we have for C and O at the inter-mediate metallicity of the LMC. Nonetheless, we make a note of caution that a deficiency of carbon relative toother elements in the LMC (log C/O = − − − Fig. 18.—
Map of the H i column density in the LMC derived from 21 cm emission (Kim et al. 2003), with the METAL sight-lines overlaid.Cyan and magenta circles correspond to sight-lines with negative and positive residuals from the main relation between iron depletionsand hydrogen column density, respectively. The size of the circles is proportional to the absolute value of the residuals. The blue and red(very small) circles similarly scale with the depletion measurement error, which is negligible compared to the magnitude of the residualsfor all sight-lines except Sk-69 104. There is a clear East-West split in the gas-phase abundances and depletions, with sight-lines along theSouth-East over-dense H i filament (Mastropietro et al. 2009) having enhanced gas-phase metallicities (up to +0.3 dex), and sight-lines onthe West side of the LMC center having depressed metallicities (up to − TABLE 10Multilinear regressions results for depletions as a function of hydrogen column density, hydrogen volume density, anddistance to the LMC center
Elements α log N ( H ) α log n ( H ) α d ( center ) β Fe 32 -0.82 +0 . − . -0.13 +0 . − . -0.10 +0 . − . +3 . − . Si 23 -0.73 +0 . − . -0.45 +0 . − . -0.08 +0 . − . +5 . − . Mg 27 -0.34 +0 . − . -0.09 +0 . − . -0.07 +0 . − . +2 . − . Ni 30 -0.82 +0 . − . +0 . − . -0.10 +0 . − . +3 . − . S 18 -1.38 +0 . − . -0.10 +0 . − . -0.13 +0 . − . +10 . − . Zn 31 -0.42 +0 . − . -0.26 +0 . − . -0.05 +0 . − . +2 . − . Ti 10 -1.32 +0 . − . +0 . − . -0.22 +0 . − . +10 . − . Cr 27 -0.62 +0 . − . -0.10 +0 . − . -0.06 +0 . − . +4 . − . D u s t - t o - g a s R a t i o ( D / H ) D/H from fit to depletionsD/H from depletion measurementsD/H from FIR emission )10 F r a c t i o n o f D / H c o n t r i b u t e d b y X COMgSiFeNiZnTi P r o b a b ili t y Fig. 19.— (Top) Dust-to-gas ratio (D/H), obtained from the collection of depletions measured by the METAL program, as a function ofthe logarithm of the hydrogen column density, N(H) (orange points and red lines). The orange points are measurements for each sight-line,while the red lines, the transparency of which scales with the posterior probability of the realization, were obtained from the fits of theindividual depletions with log N(H). For comparison, the D/H measured from FIR, 21 cm, and CO 1-0 in Roman-Duval et al. (2017) isshown in black. (Bottom) Contributions of different elements X to D/H. δ (C) or δ (O) and δ (Fe) (Jenkins 2009) and obtain an estimate of δ (C) and δ (O) for each sight-line. We then compute the D/H ac-cording to Equation 8 for each sight-line. The resultingdust-to-gas ratio values are plotted in the top panel ofFigure 19 as orange points.In addition, we perform the same type of estimate ofD/H for for each N(H) in a grid of column densities cov-ering the observed range (log N(H) = 20—22 cm − ). Wedetermine the depletion of Fe, δ ( F e ) LMC , correspond-ing to this value of N(H) from Equation 6 and its best-fit coefficients in Table 9 ( a H ( F e ) LMC , b H ( F e ) LMC , and z H ( F e ) LMC ). We then use the fitted relations betweendepletions of different elements, δ ( X ) LMC and that ofiron, described by Equation 4 and its best-fit coefficientslisted in Table 7 ( a ( X ) LMC , b ( X ) LMC , z ( X ) LMC ), to esti-mate the depletions of each element covered by METALfor each N(H) in the grid. We take this approach insteadof using the fits between log N(H) and δ (X) directly be-cause the fitted correlations between log N(H) and δ (Fe)on the one hand, and δ (Fe) and δ (X) on the other hand,are less noisy. For C and O, for which we do not havesufficient measurements in the LMC, we proceed as de-scribed above, by using the seemingly ”universal” rela-tionship between the depletion of iron and that of otherelements. The resulting trend between D/H and N(H) isplotted as a red line in Figure 19.We propagate the uncertainties at each step ofthe computation of the D/H using a Monte-Carloapproach, by drawing realizations of each param-eter ( a H ( F e ) LMC , b H ( F e ) LMC , a ( X ) LMC , b ( X ) LMC , a ( C, O ) MW , b ( C, O ) MW ) within their uncertainties. InFigure 19, the uncertainties on the D/H measurementsfor each sight-line are shown as orange error bars, whilethe fitted trend of D/H vs N(H) for each realization hasa color proportional to its posterior probability.D/H in the LMC increases by a factor 4 between logN(H) = 20 and 22 cm − , from D/H (cid:39) (cid:39) i
21 cm emission that G/D decreases from1500 (H/D = 1100) at Σ g = 9 M (cid:12) pc − , correspondingto log N(H) = 8 × cm − , to 500 (H/D = 370) at Σ g = 100 M (cid:12) pc − (log N(H) = 10 cm − (black pointsin Figure 19), albeit with large systematic uncertaintiesinherent to the lack of constraints on the FIR dust emis-sivity and the CO-to-H conversion factor. Figure 19shows that the slope of the relation between log N(H) andD/H is consistent between the FIR and UV-spectroscopybased depletions, but the zero-point of the FIR-basedtrend lies a factor of 2 lower than the depletion-basedtrend. Two possible effects could explain this difference.First, as pointed out earlier in this paper, and alsoby author studies (e.g., Roman-Duval et al. 2014; Clarket al. 2019), the dust FIR opacity is degenerate with dustsurface density: even if the dust temperature is knownfrom fitting the FIR SED, the FIR surface brightness ata given wavelength is still proportional to the product ofopacity and dust surface density (see, e.g., Equation 7 ofGordon et al. 2014). Furthermore, the dust opacity can vary significantly with environment (e.g. Stepnik et al.2003; K¨ohler et al. 2012; Demyk et al. 2017). A factor of2 systematic uncertainty on FIR opacity, and thereforeon dust surface densities derived from FIR maps, wouldnot be surprising at all. It is therefore possible that theoffset seen in Figure 19 between the D/H derived fromdepletions and FIR maps is simply due to an uncertainassumption on the FIR opacity.Second, the geometrical set-up is fundamentally differ-ent between pencil-beam UV spectroscopy toward mas-sive stars randomly distributed in the dust disk and FIRobservations in a 10-15 pc beam probing the interstellardust through the entire depth of the disk. Already, acomparison of H i column densities derived from fittingthe damping wings of the Ly- α line at 1216 ˚A towardmassive stars and those derived from 1’ resolution (15pc) 21 cm observation yield vastly different column den-sities. To evaluate the magnitude of these geometric ef-fects on the D/H vs N(H) trends, we used a prototypicalhydrodynamic simulation from Federrath et al. (2010) tocreate a simple toy model. We scale the 256 pixel boxto a mean density of 20 cm − , and assume that a pixelin gas density cube is 0.2 pc wide. Following the chem-ical evolution model by Asano et al. (2013), we further-more assume that the G/D is given by G/D = (G/D) f /f dust , where f is the fraction of metals locked in dustat (G/D) (we assume f = 0.15 and (G/D) = 1200 inthe LMC), and f dust is given by Equations 16 of Roman-Duval et al. (2017), repeated here for clarity: f dust = f exp (cid:16) tτ acc (cid:17) − f + f exp (cid:16) tτ acc (cid:17) (9)and (cid:18) τ acc yr (cid:19) = 2 × × (cid:18) < a > . µm (cid:19) (cid:16) n H − (cid:17) − (cid:18) T d K (cid:19) − (cid:18) Z . (cid:19) − (10)with Z = 0.01 for the LMC (half solar), and we assume t =1Gyr. This simple model basically assumes that dustgrowth with a timescale inversely proportional to densityand metallicity, and this allows us to compute the H/Dand n dust = n gas × (D/H) in each model cell based onits gas density.We place 32 random sight-lines in the simulation boxsuch that their hydrogen column density is between 10 and 10 cm − as in our sample, and compute the surfacedensity of gas Σ( H ), the surface density of dust Σ dust and the D/H = Σ dust /Σ gas toward each sight-line. Toemulate the nature and lower spatial resolution of FIRobservations, we re-bin the simulated cube to a 5 cubeof 10 pc wide pixels, and compute surface densities ofgas and dust and the dust-to-gas ratio through the en-tire depth of this simulated box. The resulting trendsof D/H vs hydrogen column density in this simple toymodel are shown in Figure 20. Owing to the differentregions probed by these different types of measurements(absorption spectroscopy toward point sources in the diskvs FIR emission photometry through the entire disk)7and the hierarchical density structure of the ISM, therelation between D/H and log N(H) recovered from therandomly placed pencil-bean sight-lines and the coarserresolution integrated measurements are offset by a factor > I fine structure lines is only representative of theC I gas, which represents a small fraction of the gas alongthe line of sight. In the bottom panel of Figure 19, weshow the fractional contribution of each element to theD/H, as a function of hydrogen column density. This fig-ure provides some constraints about the composition ofdust expected from depletion measurements. In partic-ular, the mass contributions of refractory elements, suchas Fe, Ti, Ni, and that of carbon, decrease with increas-ing column density. Conversely, volatile elements suchas O, S, Si, Mg, Zn increasingly contribute to the dustmass budget as the column density of ISM increases. Inparticular, O, S, and Zn only contribute above log N(H) >
21 cm − . SUMMARY
In this paper, we presented the analysis of the 32 UVmedium-resolution COS and STIS spectra toward mas-sive stars in the half-solar metallicity LMC taken as partof the METAL large program. From the spectra, chemi-cal abundances and interstellar depletions (fraction of anelement in the gas-phase) for key constituents of dust inneutral gas (Mg, Si, Fe, Ni, S, Zn, Cr, and Cu) are de-rived using the apparent optical depth method. For eachsight-line, we also derive the volume density, radiationfield, and electron density from the C I fine structurelines and an estimate of the C I /C II line ratio. Com-bined with previously determined atomic and molecularhydrogen column densities, this allows us to probe theenvironmental dependence of depletions.We find that the depletions of different elements aretightly correlated with each other (even after accountingfor covariant errors), implying that 1) depletions respondto the same environmental parameters causing their vari-ations, and 2) depletions for an element (e.g., Fe forwhich many lines of different oscillator strengths are ac-cessible throughout the UV) can be used as a proxy fordepletions for another element. For all elements, deple-tions decrease with increasing hydrogen column density,at rates that do not systematically depend on condensa-tion temperature.We find that hydrogen column density is the maindriver for depletion variations, even after accounting forthe covariant errors between depletions and hydrogen col-umn density. Within the parameter space probed byMETAL, no correlations with molecular fraction or ra-diation field intensity are found. There are, however,secondary, weaker correlations with volume density, de- rived from the C I fine structure lines, and distance to theLMC center ( − − ), located about 1 kpc Westof the massive star-forming region 30 Doradus. Theoreti-cally, one would expect the dust-to-gas ratio to correlatemainly with volume density. However, the density de-rived by the C I fine structure lines only traces a smallfraction of the gas along the line of sight (dominatedby C II , not C I ), while the mean density is not ob-servable. Hence, it is not too surprising that depletionsare not well correlated with the volume density derivedfrom the C I lines, but better correlated with hydrogencolumn density, which is a good proxy for the averagevolume density along the line of sight for a galaxy seenface-on. A simple simulation furthermore shows that apurely volume density-dependent dust-to-gas ratio willtranspire as correlated with column density as well, onceintegrated along the line of sight. The two-dimensionaldistribution of the gas metallicity reveals that the gra-dient in depletions with distance to the LMC center re-sults from an East-West pattern in metallicity, with gasin the H i -rich East side being more enriched than onWest side. The split in metallicity occurs roughly alongthe line perpendicular to the leading edge of the LMC.On the compressed side of the LMC traced by the South-east H i overdensity (SEHO Nidever et al. 2008), whereintense star-formation is occurring (in 30 Dor for exam-ple), and has occurred in the last few ten millions years(as evidences by the super-shells), the gas is more en-riched in metals (by up to +0.3 dex), while gas on thewestern side of the LMC reaches gas-phase metallicitieslower by up to − (cid:39) (cid:39) − . We therefore confirm the factor 3-4increase in dust-to-metal and dust-to-gas ratios betweenthe diffuse and translucent/molecular ISM observed fromfar-infrared, 21 cm, and CO 1-0 observations in the LMC.While the slope of the relation between hydrogen col-umn density and D/H is consistent between depletionsand FIR measurements, the zero point of the relationbased on UV depletions lies a factor of 2 higher com-pared to the FIR measurements. This discrepancy canbe explained by systematic uncertainties in the FIR opac-ity of dust assumed to convert the surface brightness tosurface density measurements, and/or by geometrical ef-fects combining the hierarchical nature of the ISM andthe different volumes probed by the FIR measurements(corresponding to ∼
10 pc pixels on the sky probing theentire depth of the gas + dust disk) and the D/H mea-8
Fig. 20.—
Dust-to-gas ratio (D/H) as a function of the logarithm of hydrogen column density (left) and mean density (right), in a toymodel based on a hydrodynamic simulation. In the toy model, the local dust-to-gas ratio is assumed to be dependent on the local volumedensity, via density-dependent timescales for the formation of dust (see Equations 10). The red points, emulating pencil beam absorptionspectroscopy measurements, were obtained by computing the integrated D/H (ratio of surface densities of dust to gas integrated along thesight-lines all the way to the depth of the source) toward 32 point sources (1 pixel wide) randomly placed in the 256 , which is assumed tobe 50 pc wide (0.2 pc pixels). The black points, emulating D/H measurements at the lower resolution of FIR observations, were computedby integrating the dust and gas surface densities over the entire depth of the box, as well as 50 pixels on each side (10 pc). Owing to thedifferent regions probed by these different types of measurements and the hierarchical density structure of the ISM, the resulting integrateddust-to-gas ratio follows trends with hydrogen column density that are offset from one another. The trends agree between the two types ofmeasurements as a function of mean volume density. surements based on UV spectroscopy (pencil-beam sight-lines scattered with the depth of the disk).Future work based on the parallel imaging obtained aspart of METAL to produce extinction maps in the LMCand compare them to FIR emission based obtain withthe Spitzer and
Herschel space telescopes will allow us tocharacterize the FIR opacity of dust and its variations inthis galaxy. This work will in turn substantially reducethe systematic uncertainties on D/H measured from FIR,and disentangle opacity and geometric effects responsi-ble for the factor 2 discrepancy between D/H measuredfrom FIR and from UV-spectroscopy-based depletions.Additionally, an upcoming paper will examine how therelation between dust-to-metal ratio and environmentchanges at different metallicities, based on similar de-pletion samples in the Milky Way (Jenkins 2009) andSmall Magellanic Cloud (Jenkins & Wallerstein 2017).The characterization of depletion patterns at a range ofmetallicities will provide key information for interpret-ing metallicity measurements in DLAs, reduce system-atic uncertainties on dust-based gas mass estimates ingalaxies in the nearby and high-redshift universe, andconstrain chemical evolution models.We thank the referee for providing insightful commentsand suggestions that contributed to validating the ro- bustness of the results. We thank Chris Evans andDanny Lennon for discussing and providing literature ref-erences for abundances of OB stars in the LMC. EdwardB. Jenkins was supported by grant HST-GO-14675.004-A to Princeton University. Benjamin Williams acknowl-edges support from grant HST-GO-14675.009-A to theUniversity of Washington. Lea Hagen and Karl Gor-don were supported by grant HST-GO-14675.002-A tothe Space Telescope Science Institute. Karin Sandstromand Petia Yanchulova Merica-Jones acknowledge supportfrom grant HST-GO-14675.008-A to the University ofCalifornia, San Diego. This work is based on observa-tions with the NASA/ESA Hubble Space Telescope ob-tained at the Space Telescope Science Institute, whichis operated by the Associations of Universities for Re-search in Astronomy, Incorporated, under NASA con-tract NAS5-26555. These observations are associatedwith program 14675. Support for Program number 14675was provided by NASA through a grant from the SpaceTelescope Science Institute, which is operated by theAssociation of Universities for Research in Astronomy,Incorporated, under NASA contract NAS5-26555. Thispaper makes uses of the corner python code (Foreman-Mackey 2016).
REFERENCESAbdo, A. A., Ackermann, M., Ajello, M., et al. 2010, A&A, 512,A7Asano, R. S., Takeuchi, T. T., Hirashita, H., & Inoue, A. K. 2013,Earth, Planets, and Space, 65, 213 Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009,ARA&A, 47, 481Bladh, S., & H¨ofner, S. 2012, A&A, 546, A76 Bowen, D. V., Jenkins, E. B., Tripp, T. M., et al. 2008, ApJS,176, 59Boyer, M. L., Srinivasan, S., Riebel, D., et al. 2012, ApJ, 748, 40Bustard, C., Pardy, S. A., D’Onghia, E., Zweibel, E. G., &Gallagher, J. S., I. 2018, ApJ, 863, 49Carswell, R. F., & Webb, J. K. 2014, VPFIT: Voigt profile fittingprogram, Astrophysics Source Code Library, ascl:1408.015Chiang, I. D., Sandstrom, K. M., Chastenet, J., et al. 2018, ApJ,865, 117Cioni, M. R. L. 2009, A&A, 506, 1137Clark, C. J. R., De Vis, P., Baes, M., et al. 2019, MNRAS, 489,5256De Vis, P., Jones, A., Viaene, S., et al. 2019, A&A, 623, A5Demyk, K., Meny, C., Leroux, H., et al. 2017, A&A, 606, A50Draine, B. T. 2009, in Astronomical Society of the PacificConference Series, Vol. 414, Cosmic Dust - Near and Far, ed.T. Henning, E. Gr¨un, & J. Steinacker, 453Elmegreen, B. G., Kim, S., & Staveley-Smith, L. 2001, ApJ, 548,749Federrath, C., Roman-Duval, J., Klessen, R. S., Schmidt, W., &Mac Low, M. M. 2010, A&A, 512, A81Feldmann, R. 2015, MNRAS, 449, 3274Field, G. B. 1974, ApJ, 187, 453Foreman-Mackey, D. 2016, The Journal of Open Source Software,1, 24Galliano, F., Galametz, M., & Jones, A. P. 2018, ARA&A, 56, 673Gaustad, J. E., McCullough, P. R., Rosing, W., & Van Buren, D.2001, PASP, 113, 1326Girolami, M., & Calderhead, B. 2011, Journal of the RoyalStatistical Society: Series B (Statistical Methodology), 73, 123Gordon, K., Clayton, G., Misselt, K., & et al. 2003, ApJ, 594, 279Gordon, K. D., Roman-Duval, J., Bot, C., et al. 2014, ApJ, 797,85Grocholski, A. J., Cole, A. A., Sarajedini, A., Geisler, D., &Smith, V. V. 2006a, AJ, 132, 1630—. 2006b, AJ, 132, 1630Indriolo, N., Geballe, T. R., Oka, T., & McCall, B. J. 2007, ApJ,671, 1736Jenkins, E. B. 1996, ApJ, 471, 292—. 2009, ApJ, 700, 1299Jenkins, E. B., Savage, B. D., & Spitzer, L., J. 1986, ApJ, 301, 355Jenkins, E. B., & Tripp, T. M. 2001, ApJS, 137, 297—. 2006, ApJ, 637, 548—. 2011, ApJ, 734, 65Jenkins, E. B., & Wallerstein, G. 2017, ApJ, 838, 85Jones, A. P., Tielens, A. G. G. M., Hollenbach, D. J., & McKee,C. F. 1994, ApJ, 433, 797Jones, A. P., Tielens, A. G. G. M., Hollenbach, D. J., & McKee,C. F. 1996, in The Role of Dust in the Formation of Stars, ed.H. U. K¨aufl & R. Siebenmorgen, 419–+Kalberla, P. M. W., & Haud, U. 2015, A&A, 578, A78Kalberla, P. M. W., McClure-Griffiths, N. M., Pisano, D. J., et al.2010, A&A, 521, A17Kelly, B. C. 2007, ApJ, 665, 1489Kim, C.-G., & Ostriker, E. C. 2018, ApJ, 853, 173Kim, S., Staveley-Smith, L., Dopita, M. A., et al. 2003, ApJS,148, 473Kisielius, R., Kulkarni, V. P., Ferland, G. J., Bogdanovich, P., &Lykins, M. L. 2014, ApJ, 780, 76Kisielius, R., Kulkarni, V. P., Ferland, G. J., et al. 2015, ApJ,804, 76K¨ohler, M., Stepnik, B., Jones, A. P., et al. 2012, A&A, 548, A61Lodders, K. 2003, Meteoritics and Planetary Science Supplement,38, 5272Luck, R. E., Moffett, T. J., Barnes, III, T. G., & Gieren, W. P.1998, AJ, 115, 605Markova, N., Puls, J., Dufton, P. L., et al. 2020, A&A, 634, A16Mastropietro, C., Burkert, A., & Moore, B. 2009, MNRAS, 399,2004 Mathis, J. S., Mezger, P. G., & Panagia, N. 1983, A&A, 500, 259Matsuura, M., Dwek, E., Meixner, M., et al. 2011, Science, 333,1258McClure-Griffiths, N. M., Pisano, D. J., Calabretta, M. R., et al.2009, ApJS, 181, 398McKinnon, R., Torrey, P., & Vogelsberger, M. 2016, MNRAS,457, 3775Meixner, M., Gordon, K. D., Indebetouw, R., et al. 2006, AJ, 132,2268Morgan, H. L., & Edmunds, M. G. 2003, MNRAS, 343, 427Morton, D. C. 2003, ApJS, 149, 205Neufeld, D. A., Goicoechea, J. R., Sonnentrucker, P., et al. 2010,A&A, 521, L10Nidever, D. L., Majewski, S. R., & Butler Burton, W. 2008, ApJ,679, 432Pagel, B. E. J., Edmunds, M. G., Fosbury, R. A. E., & Webster,B. L. 1978, MNRAS, 184, 569Rafelski, M., Wolfe, A. M., Prochaska, J. X., Neeleman, M., &Mendez, A. J. 2012, ApJ, 755, 89Rasmussen, C. E., & Williams, C. K. I. 2006, Gaussian Processesfor Machine Learning, ed. Cambridge, MA: MIT PressR´emy-Ruyer, A., Madden, S. C., Galliano, F., et al. 2014, A&A,563, A31Riebel, D., Srinivasan, S., Sargent, B., & Meixner, M. 2012, ApJ,753, 71Ritchey, A. M., Federman, S. R., & Lambert, D. L. 2018, ApJS,236, 36Roman-Duval, J., Bot, C., Chastenet, J., & Gordon, K. 2017,ApJ, 841, 72Roman-Duval, J., Gordon, K. D., Meixner, M., et al. 2014, ApJ,797, 86Roman-Duval, J., Jenkins, E. B., Williams, B., et al. 2019, ApJ,871, 151Rowlands, K., Gomez, H. L., Dunne, L., et al. 2014, MNRAS,441, 1040Rowlands, K., Dunne, L., Maddox, S., et al. 2012, MNRAS, 419,2545Russell, S. C., & Dopita, M. A. 1992, ApJ, 384, 508Sandstrom, K. M., Bolatto, A. D., Draine, B. T., Bot, C., &Stanimirovi´c, S. 2010, ApJ, 715, 701Savage, B. D., & Sembach, K. R. 1991, ApJ, 379, 245—. 1996, ARA&A, 34, 279Sembach, K. R., & Savage, B. D. 1992, ApJS, 83, 147Shull, J. M., & van Steenberg, M. 1982, ApJS, 48, 95Srinivasan, S., Boyer, M. L., Kemper, F., et al. 2016, MNRAS,457, 2814Stepnik, B., Abergel, A., Bernard, J.-P., et al. 2003, A&A, 398,551Tchernyshyov, K., Meixner, M., Seale, J., et al. 2015, ApJ, 811, 78Temim, T., Dwek, E., Tchernyshyov, K., et al. 2015, ApJ, 799,158Toribio San Cipriano, L., Dom´ınguez-Guzm´an, G., Esteban, C.,et al. 2017, MNRAS, 467, 3759Trundle, C., Dufton, P. L., Hunter, I., et al. 2007, A&A, 471, 625Tsuge, K., Sano, H., Tachihara, K., et al. 2019, ApJ, 871, 44Tumlinson, J., Shull, J. M., Rachford, B. L., et al. 2002, ApJ,566, 857Wakker, B. P., & Mathis, J. S. 2000, ApJ, 544, L107Weingartner, J. C., & Draine, B. T. 2001, ApJ, 563, 842Welty, D. E., & Crowther, P. A. 2010, MNRAS, 404, 1321Welty, D. E., Lauroesch, J. T., Wong, T., & York, D. G. 2016,ApJ, 821, 118Welty, D. E., Xue, R., & Wong, T. 2012, ApJ, 745, 173Zhukovska, S., Dobbs, C., Jenkins, E. B., & Klessen, R. S. 2016,ApJ, 831, 147Zhukovska, S., Gail, H.-P., & Trieloff, M. 2008, A&A, 479, 453Zhukovska, S., & Henning, T. 2013, A&A, 555, A99 APPENDIX A: VALIDATION OF THE MEASUREMENTS
Impact of the component spacing in the profile fitting
In fitting the COS line profiles (Mg II , Ni II , C I ), we nominally use a component spacing of 10 km s − . One mightanticipate that reducing the component spacing could lead to larger column densities since smaller b values would beassociated with more tightly spaced components. To evaluate the impact of the choice of component spacing on the0column density outcomes, we have performed the profile fitting on the COS spectra for the Mg II , Ni II , and C I lineswith 5 km − component spacing. The comparison of the results for 10 km − and 5 km − component spacing isshown in Figure 21.For Ni II , the column density measurements with 5 and 10 km s − component spacing are in excellent agreementfor all but 2 sight-lines (BI 184 and BI 253), for which the column density derived with a 5 km s − component spacingexceeds the column density derived from 10 km s − spacing by 1-1.5 dex. The uncertainties are nonetheless muchlarger in the 5 km s − case, and thus both results are within uncertainties. Furthermore, the AOD method applied tothe Ni II lines in the STIS NUV spectra (1709, 1741, and 1751 ˚A) for both sight-lines yields column density outcomesthat are in near perfect agreement (within 0.1 dex) of the column densities derived from the profile fitting applied tothe COS FUV spectra (1317, 1370 ˚A) with the more realistic 10 km s − component spacing.For Mg II , 5 out of 10 sight-lines show excellent agreement between the results obtained with 5 and 10 km s − component spacings, while for the other half (Sk-69 279, Sk-66 19, Sk-68 73, BI 253, Sk-68 140), the tighter componentspacing yields column densities 0.1 dex in excess of the ones derived from the 10 km s − spacing (albeit still withinuncertainties given the larger errors associated with the 5 km s − spacing). Thus, it is possible that our Mg II valuesderived from profile fitting with 10 km s − applied to COS spectra underestimate the true column densities. However,we deem the values derived from a 5 km s − spacing implausible given the excellent agreement between the valuesderived from the AOD applied to STIS spectra for Ni II and their counterpart derived from COS spectra and theprofile fitting method with 10 km s − component spacing.Finally, for C I , the values of N (C I ) tot , f and f are all within uncertainties whether a component spacing of 5 kms − or 10 km s − is assumed. Comparison of column densities derived from the AOD and profile fitting applied to STIS spectra
Owing to the different spectral resolutions of COS and STIS, different approaches were selected to determine columndensities (AOD for the higher resolution STIS spectra, profile fitting for the lower resolution COS spectra). In orderto evaluate systematic differences between the two methods, we applied the profile fitting method to the Fe II andSi II lines for a few STIS spectra. These lines were selected as benchmarks because 1) Fe II column densities are veryrobustly determined from the AOD thanks to the many lines available and the lack of saturation deduced from theconsistency between weak and strong lines, and 2) Si II column density measurements with the AOD are difficult dueto the availability of only one transition ( λ Comparison between COS-, STIS-, and FUSE-based column densities
A few sight-lines in the METAL sample were observed with COS in the FUV (G130M and G160M), andSTIS/E230M/1978 in the NUV (see Table 7 in Roman-Duval et al. 2019). For those sight-lines, the Ni II columndensities estimated from the λ λ ∼ λ λ λ ∼ II , Si II , Zn II , and Cr II in FUSE and COS NUV spectra, a few ofwhich were also observed with STIS as part of the METAL survey. We can therefore also compare the column densitymeasurements of these ions obtained at two different resolutions. The central optical depth of a line, τ , is determinedby the column density, oscillator strength f λ , and velocity dispersion b through τ = 1 . × − N f λ λ/b . Thevelocity dispersions for our sight-lines are not known, but assuming that they do not exhibit significant variations fromone case to the next, we can state that W λ /λ = 2 bF ( τ ) /c , where the function F increases monotonically (see Eq. 4of Jenkins 1996). We thus compare the column densities derived from higher resolution (STIS) observations to thosefrom lower resolution (COS, FUSE) as a function of equivalent width. The results of these comparisons are shown inFigure 23.In Figure 23, it is clear that, as the strength of a line increases, measurements made in lower resolution spectra(FUSE, COS) increasingly underestimate the column density obtained from the higher resolution (STIS) spectra.While the vast majority of column density measurements presented here were performed using STIS spectra, theeffects of unresolved saturation constitute a limitation of our measurements for a few sight-lines for which the Mg II ,S II and/or Ni II column densities were measured with COS only (see Table 3).For Mg II , column densities derived from COS range between log N(Mg II ) = 15.77 and 16.06 cm − . With log f λ λ = − λ II measurements occupy the range log W λ /λ < −
4, which empirically appears to beless susceptible to unresolved saturation according to the differences in column densities derived at different resolutionsin Figure 23. All the S II column density measurements (except one) derived from COS only yielded lower limits andthus account for the significant effects of saturation. Lastly, the Ni II measurements derived from COS spectra only ( λ f λ λ = 2.02 ˚A) have a range of COS-based W λ /λ = − − log N(MgII), 10 km/s component spacing l o g N ( M g II ) , k m / s c o m p o n e n t s p a c i n g log N(NiII), 10 km/s component spacing l o g N ( N i II ) , k m / s c o m p o n e n t s p a c i n g log N(CI), 10 km/s component spacing l o g N ( C I ) , k m / s c o m p o n e n t s p a c i n g f , 10 km/s component spacing f , k m / s c o m p o n e n t s p a c i n g f , 10 km/s component spacing f , k m / s c o m p o n e n t s p a c i n g Fig. 21.—
Comparison of column densities derived from the profile fitting method applied to COS spectra with 10 km s − componentspacing (x-axis) and 5 km s − spacing (y-axis) for Mg II (top left), Ni II (top right), C I (middle left), f = N (C I ∗ )/ N (C I ) tot (middleright) and f = N (C I ∗∗ / N (C I ) tot (bottom). The red dashed line indicates a 1:1 comparison. error bars). Thus, the effects of saturation in our COS-based Ni II column density estimates are probably relativelylarge, but our large error bars should capture this uncertainty. Comparison of different estimations of the column density of C II in calculating the radiation field intensity and volume density in Section 4, we estimate the column density of C II by scaling the hydrogen column density with the total (stellar) carbon abundance in the LMC and the depletion ofcarbon derived from the relation between iron depletion and carbon depletion established in the Milky Way (Jenkins2009). To validate this estimation, we have compared this approach with scaling the Mg II and S II column densities2 N(FeII) - AOD N ( F e II ) - P r o f il e F i tt i n g N(SiII) - AOD N ( S i II ) - P r o f il e F i tt i n g Fig. 22.—
Comparison of Fe II (left) and Si II (right) column densities derived from the AOD (x-axis) and profile fitting (y-axis) methodsapplied to STIS spectra (with 10 km s − component spacing for the profile fitting method). The red dashed line indicates a 1:1 comparison. log W / F U S E o r C O S l o g N ( X ) - S T I S l o g N ( X ) ( c m ) FeII (FUSE)SiII (COSNUV)ZnII (COSNUV)CrII (COSNUV)NiII (COSFUV)
Fig. 23.—
Difference between column densities of various low ions X measured in COS or FUSE spectra ( R ∼ R > W λ /λ measured with the higher resolution STIS instrument. The FUSE and COS/NUV-based column densitiesare from Tchernyshyov et al. (2015), while the COS/FUV-based column densities are measured in METAL spectra (Ni II lines only). AllSTIS-based measurements are from METAL (Table 3). The empty gray circles represent binned averages, with the error in x being thewidth of the bin and the error in y the error on the mean in each bin. As the central optical depth (and hence equivalent width) increases,the lower resolution data (FUSE or COS) increasingly underestimates the true column density (as estimated from the higher resolutionSTIS data), due to unresolved saturation. measured in the LMC (Table 5) with the ratio of the LMC carbon/magnesium or carbon/sulfur stellar abundancesand the ratio of the carbon and magnesium or sulfur depletions. In this case also, the carbon depletions are estimatedfrom the iron depletions measured in the LMC and the relation between iron and carbon depletions established in theMilky Way (Jenkins 2009). The magnesium and sulfur depletions are estimated from the relation between iron andmagnesium (or sulfur) depletions established in the LMC from this study (Table 7).Owing to the weak carbon lines not being observable in the LMC and the strong carbon lines always being badlysaturated, we have no other option than to use the relation between iron and carbon depletions established in theMilky Way applied to the iron depletions measured in the LMC in order to estimate the carbon depletions in theLMC. However, it is possible that the relation between carbon and iron depletion in the LMC could differ from theone in the Milky Way. While we cannot compare the relations between carbon and iron depletions in the Milky Wayand LMC, we can perform this exercise for magnesium, which was measured in the Milky Way by Jenkins (2009)and in the LMC by this study. Therefore, in order to gauge the impact of possible differences in depletion patternsbetween the LMC and Milky Way on the estimate of C II column densities, we also scaled the measured LMC Mg II column densities (Table 5) by the ratio of carbon/magnesium stellar abundances and the ratio of carbon/magnesiumdepletions as described above, but this time using the iron depletion measured in the LMC applied to the relation3 N(CII) = A C N H ( C ) MW N ( C II ) e s t i m a t e N(CII) = N(SII)10 ( C ) MW /10 ( S ) A C /A S N(CII) = N(MgII)10 ( C ) MW /10 ( Mg ) A C /A Mg N(CII) = N(MgII)10 ( C ) MW /10 ( Mg ) MW A C /A Mg Fig. 24.—