Improving the reliability of Fe-and S-XANES measurements in silicate glasses : 4 correcting beam damage and identifying Fe-oxide nanolites in hydrous and 5 anhydrous melt inclusions 6 7

7 Allan H. Lerner, Michelle J. Muth, Paul J. Wallace, Antonio Lanzirotti, Matthew Newville, 8 Glenn A. Gaetani, Proteek Chowdhury, Rajdeep Dasgupta 9 10 1 Department of Earth Sciences, University of Oregon, Oregon 97403, USA 11 2 Center for Advanced Radiation Sources, The University of Chicago, Illinois 60637, USA 12 3 Department of Geology and Geophysics, Woods Hole Oceanographic Institution, Woods Hole, 13 Massachusetts 02543, USA 14 4 Department of Earth and Planetary Sciences, University of California, Riverside, CA 92521, USA 15 5 Department of Earth, Environmental, and Planetary Sciences, Rice University, Texas 77005, USA 16 * Corresponding author at: 1272 University of Oregon, Eugene, OR 97403, USA. E-mail address: 17 lerner.allan@gmail.com (A. H. Lerner). 18 19

Of course, inferring magma redox state from iron and sulfur valence in quenched glasses 122 requires accurate XANES measurements. The large penetrative depths of high-energy X-rays 123 and oblique incident beam trajectories at many analytical facilities require careful sample 124 preparation and analytical strategies to avoid signal contamination during Fe-and S-XANES 125 measurements of MI and matrix glasses (Figure 1). It has also been recognized that many glass 126 compositions are susceptible to X-ray induced changes in iron and sulfur speciation during 127 analysis (i.e., beam damage) ( beamlines are available at a number of synchrotron light sources worldwide with incident X-ray 145 intensities ranging from 10 8 -10 12 photons/s (Sutton et al., 2020), where higher X-ray fluxes 146 shorten analysis time by providing lower detection limits, but amplify beam damage concerns. 147 Many of these approaches to lessen beam damage are challenging to apply to MI owing 148 to their small sizes. Smaller MI require more focused beam diameters to avoid contamination by 149 the host-phase, but are thereby subject to higher photon densities and thus possible beam damage 150 (e.g., Gaborieau et al., 2020;Tassara et al., 2020). Melt inclusions are often targeted for 151 petrological investigation specifically because they can retain magmatic volatiles that are otherwise lost from the external magma during ascent and degassing (Kent, 2008; Métrich and 153 Wallace, 2008). However, silicate glasses with high H2O content have been observed to undergo 154 larger changes in iron (and potentially sulfur) speciation during irradiation than what is observed 155 in anhydrous silicate glasses Moussallam et al., 2019). Hydrous MI may 156 also be susceptible to the formation of nanolite crystals during quenching (Danyushevsky et al.,157 2002; Di , which may lead to spurious interpretation of XANES spectra. 158 These combined properties make it particularly challenging to apply XANES oxybarometry 159 methods to the analysis of MI from volcanic arc environments, which tend to be both small and 160 H2O-rich. 161 For these more challenging MI, it is beneficial to develop XANES approaches that both 162 minimize changes in elemental speciation during irradiation and correct for changes that do 163 occur. Applying generalized corrections to datasets is not ideal because differences in glass 164 compositions and H2O contents (e.g., caused by variable diffusive H + loss from MI before 165 quenching) can lead to different MI susceptibilities to beam damage within the same deposit or 166 even within the same host mineral. 167 In this study, we present techniques that aid in recognizing X-ray--induced changes in 168 iron and sulfur valence in volcanic glasses and MI that result from XANES analysis. We then 169 propose new time-dependent corrections for beam damage that does occur. For S-XANES, we 170 also introduce a new spectral fitting approach that may better account for reduction of S 6+ to S 4+ 171 during analysis. Finally, we present a method to identify the presence of Fe-oxide nanolites in 172 MI during Fe-XANES analysis. Collectively, these methods enable reliable quantification of iron 173 and sulfur valence, and thereby melt redox state, from small and/or beam damage-susceptible 174 glasses and MI. 175 176

Geometric considerations 178
At the Fe K-edge, the characteristic 1/e X-ray absorption depth in basaltic glass is ≈ 40 179 μm (Elam et al., 2002), and 120 μm (1/e 3 ) thick glass is therefore required for 95% absorption of 180 X-rays during Fe-XANES measurements. X-ray absorption by Fe-bearing inclusions or 181 crystalline host phases that may be present within the analytical path will be mixed with the signal of the targeted glass. This is a particular problem for analyzing MI, as MI are often less 183 than 100 μm thick. Consequently, most MI must be doubly intersected for Fe-XANES analysis 184 to avoid signal contamination from the host mineral. A further complication in XANES 185 measurements of MI, particularly for highly penetrative Fe-XANES analyses, is that many 186 μXANES beamline configurations utilize a ~45° slant geometry of incoming X-ray beam in 187 fluorescence operating modes. The inclined incidence angle means that as wafer thickness 188 increases in the beam direction, progressively wider doubly-intersected MI areas are needed to 189 keep the analytical path free of mineral contamination (Figure 1). Throughout the X-ray 190 penetration volume, the minimum required doubly-intersected MI dimensions for a host-free 191 glass measurement are roughly equal to MI thickness plus the beam diameter size (assuming a 192 cylindrical doubly-intersected MI area). This requires MI to be either sufficiently wide or ground 193 very thin for clean glass analyses using high energy X-rays (e.g., Fe-XANES, V-XANES, . Thus, small MI in olivine and other Fe-bearing phases can be challenging to measure 195 for Fe-XANES. Even for analyses of MI in phases that have low, but non-zero,  concentrations (e.g., feldspars), the high penetrative depth of Fe-XANES can excite a large 197 volume of the host phase, so that the host contribution to the Fe-XANES signal may be 198 significant. This issue is of particular concern for small MI and for glass compositions with 199 relatively low Fe-contents, such as dacites and rhyolites. At the lower energy S K-edge (~2500 200 eV), X-rays are more strongly attenuated, with the 1/e X-ray absorption depth in basaltic glass 201 only ~5 μm. Consequently, 95% of the S-XANES X-ray absorption occurs within the upper 15 202 μm and most of these geometric concerns are accordingly lessened (Figure 1A (brown) analyzed with a 20×20 μm X-ray beam. For many XANES fluorescence measurements, 208 the X-ray beam (black arrow) is ~45° incident to the sample surface and the sample fluorescent 209 energy (gray arrows) is measured at 45° in the opposite direction. Depending on the penetration 210 depth of the X-ray energy being used, the beam may interact with substantial host mineral both 211 laterally and at depth, leading to mineral-contaminated spectra. For Fe-XANES (red arrows), 212 37% and 86% of the X-ray signal are absorbed in 40 and 80 μm hypotenuse paths through 213 basaltic glass (28 and 56 μm vertical thicknesses), requiring the MI be both doubly intersected 214 and sufficiently wide to avoid host mineral contamination. S-XANES X-ray energies are much 215 less penetrating (blue arrows), so MI geometry and thickness concerns are lessened. (B) A series 216 of Fe-XANES measurements of a doubly-intersected olivine-hosted MI from the southern 217 Cascades (BORG-1_37, Table 1) showing a traverse from within the MI into the olivine-host,  218 demonstrating the difference in absorption edge shape between analyses of glass and of olivine. 219 Measurement locations are shown atop a Ca Kα X-ray map (inset), with symbol colors matching 220 the shown spectra. 221 222

Analytical details and sample descriptions 223
Fe-and S-XANES measurements were conducted on a variety of volcanic and 224 experimental silicate glasses at GSECARS beamline 13-ID-E at Argonne National Laboratory's 225 (Illinois, USA) Advanced Photon Source (APS), a third generation synchrotron light source 226 . Details of the 13-ID-E beamline configuration are described in Head et al. (2018) and are consistent with measurements conducted here, except for differences in photon 228 flux and analytical times described below. 229 To account for differences in monochromator calibrations between synchrotron facilities, 230 a set of standards (minerals, metal foils, synthetic glasses) were measured at the onset of each 231 analytical session to determine the appropriate energy offset to apply to Fe-and S-XANES 232 oxybarometer calibration curves relative to reference energy fitting ranges (details below). At 233 beamline 13-ID-E, the lattice constants for the monochromator Si(111) and Si(311) crystals are 234 calculated from reference foils measured throughout the analyzable energy range of the crystals, 235 and provide excellent consistency with absorption edge energies determined by Kraft et al. 236 (1996). The 13-ID-E beamline has excellent reproducibility in measured reference materials over 237 the course of standard two to three-day measurement periods and therefore no within-session 238 drift corrections were applied during either Fe-or S-XANES measurements. Prior to each 239 XANES analysis, an X-ray map was made by rapidly rastering across the sample to identify 240 areas in MI and other glass targets that were free of host mineral and microlite crystals in the 241 beam path. The X-ray beam was then turned off to prevent any further unnecessary beam 242 interaction with the glasses until XANES measurements began. 243 Analyzed samples include doubly-intersected MI and matrix glasses mounted on Fe-free 244 glass rounds and thin sections. Samples were embedded in CrystalBond®, EpoThin® epoxy, or 245 thin section resin. All bonding material and glass substrates were analyzed to confirm that they 246 contained only trace iron and had negligible contribution to Fe-XANES signals. The bonding 247 materials did contain substantial S, but the low energy X-rays for S-XANES measurements are 248 fully absorbed within a ~20 μm path within basaltic glasses (15 μm vertical path with 45º 249 incident beam angle, Figure 1A). All analyzed MI and matrix glass areas are thicker than 20 μm, 250 so that the bonding materials contributed no appreciable signal to S-XANES measurements. We 251 also analyzed singly intersected experimental glass charges, where glass thicknesses of multiple 252 mm fully absorbed X-rays at both Fe-and S-Kα energies so that contamination from the capsule 253 material was insignificant. In experimental glass charges, care was taken to analyze only crystal-254 poor glass areas and to avoid measurements near capsule edges. 255 256 257  2) Flux density on the sample was further decreased by defocusing the incident X-ray beam 274 so that photon densities were generally 1 -1.5×10 8 , 2 -4×10 7 , and 6 -9×10 6 275 photons/s/μm 2 for 5×5, 10×10, and 20×20 μm beam footprints, respectively. 276 3) Analysis times were minimized as much as possible while still providing interpretable 277 spectra, which allowed us to reduce beam exposure. 278 The 13-ID-E monochromator calibration provides a first derivative of the Fe K-edge peak 279 of iron foil at ~7110.7 eV, consistent with values determined by Kraft et al. (1996). We followed 280 the Fe-XANES measurement methodology outlined in Head et al. (2018), but with modified scan 281 times and energy ranges used to further identify and correct for beam damage. Two different 282 analytical setups were used: rapid pre-edge scans and slower full energy scans. For rapid scans, 283 the incident beam was scanned from 7092 -7107 eV in 2.5 eV steps, from 7107 -7119 eV in 0.1 eV steps, and from 7119 -7144 eV in 0.05 Å -1 (0.5 -1.0 eV) steps (continuous steps rather 285 than discrete). Each scan step was 0.5 seconds (s) and the total scan time was 82 s, with ~10 s 286 delay prior to the next analysis for beamline adjustment and computational processing. The rapid 287 pre-edge scans quickly measure over a reduced energy range to minimize beam exposure to the 288 extent possible while still collecting spectra with high enough resolution for peak fitting in the 289 pre-edge region. The 82 s scan is much faster than typical Fe-XANES scan durations reported in 290 the literature, which usually range from 270 s to >700 s (4.5 to >10 minutes) (e.g., Cottrell  Beam-induced oxidation causes a shift in Fe-Kα pre-edge peak intensities but does not produce 339 any uniquely identifiable spectral features. Consequently, it is impossible to know from a single 340 Fe-XANES analysis whether a sample had suffered from beam-induced photo-oxidation 341 (compare with S-XANES beam damage, which produced diagnostic spectral features, as 342 discussed in section 2.3). Therefore, samples must either be analyzed under carefully tested 343 analytical conditions to ensure that no significant beam damage occurs for the particular glass 344 composition and analysis duration, or a method must be employed that can identify and correct 345 for beam damage in each individual analysis spot. We emphasize the latter approach in this study, presenting a method that allows us to reliably analyze small, hydrous glass inclusions with 347 a relatively high-flux beam. 348 To identify and correct for beam damage within each analysis spot, we conducted 349 multiple rapid scans of the Fe K pre-edge region to create a time series of progressive oxidation 350  of the host mineral is less of a concern than for Fe-XANES analyses, where iron signal 549 contribution from the host phase can be significant. Consequently, larger X-ray analysis 550 footprints can generally be used for S-XANES, which reduces X-ray dose and thereby 551 ameliorates some of the beam damage potential. However, when analyzing sulfur-poor MI 552 (<~400 ppm S), even slight contributions from the host phase might be significant relative to the 553 low-sulfur glass signal, and beam overlap of the host phase should be avoided. Additionally, 554 cracks and surface contaminants (e.g., oils) may be present on prepared surfaces, both of which can contain undesired sulfur-bearing material ( not been thoroughly examined. It is consequently difficult to currently predict whether any 588 particular sample will be susceptible to S-XANES beam damage. Therefore, as with iron beam 589 damage, it is important to be able to account for beam damage within each individual 590 measurement rather than applying generalized corrections to an entire sample suite. 591 Our approach in managing and correcting S-XANES beam damage is similar to that for 592 reducing Fe-XANES beam damage, namely minimizing pre-analysis X-ray irradiation, 593 decreasing photon dose as much as possible while maintaining sufficient signal, and using repeat 594 rapid scans to observe beam-induced changes in sulfur speciation. Where S-XANES photo-595 reduction is observed, we correct affected spectra by calculating the peak area of the beam 596 damage-induced S 4+ signal and restoring this to original S 6+ intensity via a calibrated conversion 597 factor (details below). 598 599 As with our Fe-XANES beam damage correction approach, we conducted repeat rapid 612 scans to identify S-XANES beam damage and, if necessary, applied sample-specific corrections. 613

S-XANES analytical conditions
Sulfur K-edge spectra were collected by scanning the incident beam from 2437 -2467 eV in 2.5 614 eV steps, from 2467 -2487 eV in 0.1 eV steps, and from 2487 -2622 eV in 1.5 eV steps. Short 615 analysis times of either 0.5 or 1.0 s per step bin were used (continuous steps rather than discrete) 616 for rapid scans with total durations of 154 or 308 s, respectively. Three repeat scans were typically conducted for each analysis spot, with cumulative measurement times of ~8 -15 618 minutes per location. If S 4+ peak growth was identified during successive scans, only the first 619 scan was used to quantify sulfur speciation, as this scan would have undergone the least S 6+ to 620 S 4+ photo-reductive beam damage. If no S 4+ peak ingrowth was observed, the repeat scans were 621 merged to improve signal quality. 622 In beam-damaged samples, S 6+ to S 4+ photo-reduction can be corrected by restoring the 623 S 4+ 2477.5 eV peak intensity back to a S 6+ signal. This correction requires knowing an 624 appropriate signal intensity scaling factor to restore a S 4+ signal to the original S 6+ intensity. uncertain. To determine how the loss of S 6+ intensity relates to the growth of S 4+ , and therefore 629 how to calculate an appropriate signal intensity scaling factor between these peak intensities, we 630 conducted a series of measurements on a hydrous, sulfate-dominated, sulfur-rich experimental 631 basaltic glasses from Chowdury and Dasgupta (2019) ( Table 1). The large area of this 632 experimental glass allowed a series of measurements with multiple spot sizes (2×2, 10×10, 633 20×20, and 50×50 μm) to observe varying degrees of beam damage under photon densities 634 ranging from 6.9×10 6 -1.1×10 10 photons/s/μm 2 . The sulfate-only initial composition of this 635 oxidized glass made the identification of S 4+ peak ingrowth obvious. With repeat measurements, 636 we are able to track the ingrowth of the S 4+ 2477.5 eV peak (hereafter the "S 4+ peak") at the 637 expense of the S 6+ 2481.3 -2482 eV peak. We can thereby quantify how the S 4+ peak intensity 638 relates to the loss of S 6+ intensity, and how consistent the S 4+ to S 6+ intensity scaling relationship 639 is with increasing degrees of beam damage. reduction induced S 4+ signal intensity as S 2intensity. Our peak fitting approach differentiates S 2-658 , S 4+ , and S 6+ absorption intensities, enabling us to quantify beam damage by isolating S 4+ from 659 the S 2peak. We can then restore the S 4+ photo-reduction signal to an original S 6+ intensity to 660 calculate the undamaged sulfur speciation of the glass. 661 Our S-XANES peak fitting method again uses the spectral fitting program XAS viewer 662 (Newville, 2013) to correct for instrument deadtime and to fit the S-XANES data. Measured 663 spectra were first scaled by the Si-Kα signal intensity, to avoid aberrations in incident beam 664 intensity over the analysis energy range due to possible contaminants within the beamline optics. approach for all spectra (Table 2). 677 Assessing S-XANES spectra of >100 reduced and oxidized glass analyses across a 678 compositional range from basaltic to rhyolitic (Table 1; Data supplement), we identify the energy ranges of five peaks within the S-Kα absorption region. We distinguish four absorption 680 peak ranges that have been recognized as corresponding to sulfide complexes, and S 2-, S 4+ (Table 2). We additionally identify an absorption peak between 2483.5 -2486 eV, which is 683 slightly higher energy than the main S 6+ peak. This 2483.5 -2486 eV energy peak was similarly 684 identified by Konecke et al. (2017), who refer to it as the sulfur "ionization peak", a term we 685 adopt here. The sulfur-ionization peak intensity seems partially correlated to S 6+ intensity, but is 686 also present in S 2--dominated spectra. After normalizing the spectra, we simultaneously fit the 687 background with an error function and Gaussian and fit five separate Gaussian functions for each 688 of the sulfur absorption features (Figure 7, Figure A.6). Table 2  and S 6+ peak intensities to sulfur speciation, however we find that the following empirical 711 polynomial relationship is more appropriate for our peak fitting method (Figure A The average precision of our S-XANES peak fitting method, based on multiple analyses 729 in single MI and within regions of mid-ocean ridge basalt (MORB) glasses, is ±7% relative (2 RSE, 19 analyses in glasses ranging from 0.07 to 0.85 S 6+ /ΣS; see Data supplement). When 731 considering further uncertainties in the peak fitting calibration and from the non-uniqueness of 732 spectra normalization (particularly in signal-limited samples), we assume the total accuracy of 733 this method to be better than ±10% relative. 734 735 736 Figure 7. (A) Example S-XANES peak fitting to oxidized experimental glass G466. This 737 spectrum is the 2 nd of 3 repeat scans with a 20×20 μm beam (photon flux density of 1.1×10 8 738 photons/s/μm 2 ) and shows a dominant S 6+ peak (2480 -2482.3 eV) and a substantial beam 739 damage-induced S 4+ peak (2476.8 -2477.7 eV). No S 2intensity is observed. The fit residual 740 shows slight remaining unfit structure. (B) Example S-XANES peak fitting of reduced VG-2 741 MORB glass analyzed using a 50×50 μm beam (photon flux density of 6.2×10 6 photons/s/μm 2 ). 742 The noisier spectrum is due to lower sulfur content in VG-2 than G466, as well as a difference in 743 vertical scale. A main glassy S 2peak (2475.3 -2477 eV) is present, as well as a lesser S 6+ peak 744 and a minor sulfide peak (2465 -2470 eV). Minimal S 4+ beam damage ingrowth is observed 745 with this diffuse beam analysis (compare to Figure 9). A sulfur-ionization peak (2483.5 -2486 746 eV) is present in S-XANES spectra of the both oxidized and reduced glasses. See Table 2 for  747 identification of peaks and fit parameters. Reference peak position lines may vary slightly 748 between samples depending on bond coordination environments. 749 750

Correcting S-XANES beam damage 751
Since we include the S 4+ peak in our fitting methodology, we can quantitatively separate 752 the beam damage-induced S 4+ signal from the overlapping broad S 2peak in S-XANES spectra. 753 This was not possible with the Jugo et al. (2010) method because all signal intensity over this 754 region was considered as S 2-, which would lead to spurious results in beam-damaged spectra 755 (Figures 8, 9). During repeat measurements of hydrous, sulfur-rich, oxidized, anhydrite-saturated 756 experimental basaltic glasses G466 and G479 (50 -51 wt% SiO2, 9000 -15000 ppm S, 6.5 -8.9 757 wt% H2O, 1300 -1325 °C, 1.5 -2.0 GPa; Table 1 these glasses are highly oxidized, they contain no S 2signal to overlap with the S 4+ peak, which 761 makes observation of the S 4+ signal straightforward. As expected, increased photon doses with 762 more focused beams cause more rapid S 6+ to S 4+ photo-reduction. Comparing the intensity ratio 763 of S 4+ peak ingrowth and S 6+ peak loss during progressive beam damage from repeat 764 measurements with photon flux densities ranging from 10 6 to 10 10 photons/s/μm 2 , we find that 765 S 4+ peak ingrowth relates to S 6+ intensity decrease by a factor of 1.2 ± 0.1 (1 SE; n = 7) (see 766 Data supplement). We apply this scaling factor to observed S 4+ peak intensities in beam 767 damaged samples to restore original S 6+ peak intensities via: 768

I[S 6+ ] / ΣI[S T ] = ΣI[S 6+ ]/ (I[S 2-] + ΣI[S 6+ ]), [Eq. 3]
Inputting this value into our peak fitting calibration based on the Jugo et al. (2010) glass suite 775 (Eq. 1) calculates the beam damage-restored sulfur speciation. 776 In addition to the obvious S 4+ peak growth during beam damage of G466 and G479 777 glasses, we observe the ingrowth of a very small peak between 2471.6 -2472.0 eV (Figure 8  discussed]), where the S 4+ photo-reduction peak would be counted as part of the S 2signal. 804 Reference peak position lines may vary slightly between samples. 805 806

Observations of natural glasses and melt inclusions 807
We observe the same rapid S 6+ to S 4+ photo-reduction in numerous natural glasses. 808 Hydrous basaltic MI from the southern Cascades (up to 3.7 wt% H2O) undergo rapid photo-809 reduction (Muth and Wallace, 2021), which is consistent with hydrous basalts being highly 810 susceptible to speciation changes during X-ray irradiation (Cottrell et   of repeated analyses (20 repeated scans for JDF-46N; 2 -6 repeated scans for ALV892-1). 866 Cumulative irradiation durations are listed on the right, and S 6+ /ΣS calculations using the peak 867 fitting approach with and without correcting for S 4+ photo-reduction are compared. As in Figure  868 9, the ingrowth of S 4+ (2476 -2477.7 eV) and loss of S 6+ (2480.5 -2483.3 eV) is increasingly 869 apparent during longer analyses and those with more focused beams. Note that S 4+ corrections do 870 not reproduce the S 6+ /ΣS observed with low photon density measurements, indicating that 871 challenge of applying beam damage corrections in reduced glasses with overlapping S 2and S 4+ 872 peak areas. Reference peak position lines may vary slightly between samples. 873 874 We also observe S 6+ to S 4+  atmospheric interaction prior to quenching, the Kīlauea olivine-hosted MI range from reduced to 878 highly oxidized (FMQ -0.7 to +2.4; Lerner, 2020). S-XANES beam damage occurs in Kīlauea 879 MI throughout this wide range of oxidation states (Figure 11). The S 6+ to S 4+ photo-reduction 880 during X-ray irradiation in Kīlauea MI and in MORB glasses is particularly interesting because 881 these low-H2O ocean island basalt (OIB) and MORB glasses are stable during Fe-XANES 882 measurements (Figure 11), having Φ values ≤0.1 (Table 1). These observations highlight that 883 major (e.g., iron) and minor (e.g., sulfur) elements may have different susceptibilities to X-ray  peak is challenging in more reduced samples due to the overlap of the dominant S 2peak with the 919 relatively minor S 4+ peak, and we might be under-fitting the S 4+ peak in the MORB spectra. 920 Additionally, in samples with mixed sulfur speciation, the slight beam damage-induced energy 921 increase in the 2470 -2475 eV range (Figure 8 inset) would be completely masked by, and 922 included within, the broad S 2peak area. Further characterizing the complete range of sulfur 923 complexing and valence changes during beam damage will be important for further improving S-924 XANES correction methods. The uncertainties in the S 4+ to S 6+ intensity corrections underscore 925 that the foremost approach during S-XANES measurements should be to minimize beam damage 926 as much as possible, so that the overall uncertainties stemming from any S 4+ corrections are 927 small. 928 In summary, S-XANES beam damage can occur in both hydrous and anhydrous silicate 929 glasses, but can be identified through repeat rapid scans by the presence and growth of a S 4+ 930 peak. If beam damage is found to occur, we suggest focusing on the least damaged spectra for 931 each measurement, and then applying a S 4+ to S 6+ scaling factor to restore S 4+ signal to the 932 original S 6+ intensity. In high-sulfur samples, where signal intensity is sufficient even with rapid 933 scans, this is the ideal approach as beam damage is first limited and then restored to a good approximation of original S 6+ intensity. Low-sulfur samples may require merging multiple rapid 935 scans to obtain quantifiable spectra, despite the longer cumulative analysis time inducing more 936 photo-reduction. In long duration or merged scans, irradiation-induced S 4+ signal can still be 937 restored to S 6+ intensity, and although this introduces greater uncertainty (due to imprecisely 938 known S 4+ to S 6+ scaling factors), it is still a better approach than not applying any beam damage 939 correction. In highly oxidized samples lacking S 2-, accounting for S 4+ is less important as it can 940 simply be assumed that all sulfur was originally present as S 6+ . However, in samples with mixed 941 sulfur speciation, separating any S 4+ photo-reduction signal from the overlapping S 2peak, and 942 restoring the S 4+ to original S 6+ is important in accurately determining the initial sulfur speciation 943 of the glass. 944 945

Identifying Fe-oxide nanolite crystals in Fe-XANES spectra 946
In addition to beam damage concerns during XANES analyses of glasses, the possible 947 cryptic occurrence of nanolite crystals in glasses must also be considered to avoid spurious 948 interpretations of XANES spectra. Nanolites are minerals in the sub-micron range that are 949 typically undecipherable with optical microscopes or even with electron microscopes, but can 950 form in MI during quenching under certain conditions. In some settings, dispersed nanolite 951 crystals become large enough to appear as a fine "dust" within MI ( Melt inclusions in Augustine feldspar and pyroxene grains that contain Fe-oxide 990 (presumably magnetite or maghemite) nanolites are consistently a brown color, although no 991 distinct fine-scale crystals are observable with either optical or electron microscopes (Figures 12,  992 13). Optically colorless MI are also present in the same samples from Augustine, and these 993 colorless MI have smooth Fe-XANES absorption edge spectra that are indicative of glass with no magnetite-like structure (Figures 12, 13). The occurrence of colorless and brown MI, even within