Signatures of sulfur mass-independent fractionation (S-MIF) are observed for sulfur minerals in Archean rocks, and for modern stratospheric sulfate aerosols (SSA) deposited in polar ice. Ultraviolet light photolysis of SO2 is thought to be the most likely source for these S-MIF signatures, although several hypotheses have been proposed for the underlying mechanism(s) of S-MIF production. Laboratory SO2 photolysis experiments are carried out with a flow-through photochemical reactor with a broadband (Xe arc lamp) light source at 0.1 to 5 mbar SO2 in 0.25 to 1 bar N2 bath gas, in order to test the effect of SO2 pressure on the production of S-MIF. Elemental sulfur products yield high δ34S values up to 140 ‰, with δ33S/δ34S of 0.59 ± 0.04 and Δ36S/Δ33S ratios of −4.6 ± 1.3 with respect to initial SO2. The magnitude of the isotope effect strongly depends on SO2 partial pressure, with larger fractionations at higher SO2 pressures, but saturates at an SO2 column density of 1018 molecules cm−2. The observed pressure dependence and δ33S/δ34S and Δ36S/Δ33S ratios are consistent with model calculations based on synthesized SO2 isotopologue cross sections, suggesting a significant contribution of isotopologue self-shielding to S-MIF for high SO2 pressure (>0.1 mbar) experiments. Results of dual-cell experiments further support this conclusion. The measured isotopic patterns, in particular the Δ36S/Δ33S relationships, closely match those measured for modern SSA from explosive volcanic eruptions. These isotope systematics could be used to trace the chemistry of SSA after large Plinian volcanic eruptions.
 Sulfate and sulfide minerals in sedimentary rocks older than 2.4 Giga years ago (Ga) commonly show signatures of sulfur isotope mass-independent fractionation (S-MIF), anomalous isotope compositions that deviate from the mass-dependent scaling law of isotope fractionation [e.g., Farquhar et al., 2000a; Pavlov and Kasting, 2002; Ono et al., 2003]. Equilibrium, kinetic, and biological processes nominally fractionate four stable isotopes of sulfur (32S, 33S, 34S, and 36S) in proportion to their difference in mass as predicted by the quantum mechanical theory of isotope fractionation [e.g., Urey, 1947]. Therefore, the preservation of MIF signatures indicate that a fundamental change in the sulfur cycle occurred at about 2.4 Ga. Assuming that the MIF signature is sourced from ultraviolet (UV) photolysis of SO2, production and preservation of S-MIF signatures are thought to only be possible under an anoxic early atmosphere that is UV transparent and allows rich sulfur atmospheric chemistry [Farquhar et al., 2000a, 2001; Pavlov and Kasting, 2002]. In addition, several works suggest potential secular structures in the Archean S-MIF record, such as changes in the magnitude of 33S anomaly or in the relationship between the 33S and 36S anomalies [Ono et al., 2006a; Ohmoto et al., 2006; Farquhar et al., 2007; Zerkle et al., 2012]. A better understanding of the mechanisms responsible for the production of S-MIF during SO2 photolysis would provide additional constraints on the early evolution of the atmosphere beyond atmospheric oxygen levels.
 S-MIF signatures are also measured in stratospheric sulfate aerosols (SSA) deposited in polar ice [Savarino et al., 2003, Baroni et al., 2007, 2008]. These S-MIF signatures are only associated with large stratospheric eruptions (e.g., Pinatubo, 1991, Agung, 1963, Tambora, 1815) but not with predominantly tropospheric eruptions (e.g., Cerro Hudson, 1991), suggesting S-MIF can be used as a proxy for stratospheric volcanic events in the past [Savarino et al., 2003; Baroni et al., 2007, 2008]. The source reaction for S-MIF in the modern stratosphere, however, has been debated, because the MIF source mechanism is still not well constrained, and because of the difficulty in preserving S-MIF signatures in a present-day atmosphere [Savarino et al., 2003; Pavlov et al., 2005].
 Laboratory photochemical studies have shown that photolysis of SO2 and photopolymerization of CS2 both produce S-MIF [Farquhar et al., 2001; Masterson et al., 2011; Whitehill and Ono, 2012; Zmolek et al., 1999], but photolysis of H2S and OCS does not [Farquhar et al., 2000b; Lin et al., 2011]. Both CS2 and SO2 exhibit fine structures in their UV absorption spectrum, suggesting that S-MIF is associated with the process of predissociation [Farquhar et al., 2001; Zmolek et al., 1999; Danielache et al., 2008; Lyons, 2007, 2008]. Two absorption bands of SO2, from 190 to 220 nm and 240 to 340 nm, are thought to be important for S-MIF production [Farquhar et al., 2001; Danielache et al., 2008]. Laboratory experiments by Farquhar et al.  showed that the pattern of S-MIF depends on the wavelength of excitation. This was used to link the 190 to 220 nm band to Archean S-MIF [Farquhar et al., 2001; Ueno et al., 2008; Lyons, 2007], and the 240 to 340 nm band to the S-MIF in SSA [Savarino et al., 2003; Baroni et al., 2007, 2008]. However, recent work using broadband radiation sources shows that S-MIF associated with the 240 to 340 nm is characterized by a positive correlation between Δ36S and Δ33S values, whereas negative correlations are observed in SSA [Whitehill and Ono, 2012]. Thus, in this study, we focus on the excitation of SO2 under the 190 to 220 nm absorption region.
 Several hypotheses have been proposed for the mechanism underlying the production of S-MIF. A symmetry-dependent isotope effect, which has been suggested as the mechanism for oxygen isotope MIF during the ozone recombination reaction [Gao and Marcus, 2001], does not apply to SO2, as sulfur isotope substitution does not change the molecular symmetry of SO2. The SO2 absorption band in the 190 to 220 nm region exhibits vibrational fine structure due to bending mode progressions of the 1B2 ← X1A1 system (Figure 1-A) [Freeman et al., 1984]. Isotope substitutions shift the band positions and can produce isotope self-shielding due to differential optical column density (Figure 1) [Lyons, 2007, 2008]. In addition to the shift of band positions, sulfur isotope substitution can also affect the cross-section amplitude due to isotope differences in the Franck-Condon envelope and vibrational structures [Danielache et al., 2008; Tokue and Nanbu, 2010]. Molecular dynamics during predissociation, such as nonadiabatic resonances among various bound excited states of SO2, may also contribute to S-MIF production [e.g., Masterson et al., 2011 and Zmolek et al., 1999 for CS2].
 The goal of this study, therefore, is to test which of these three proposed mechanisms contributes predominantly to the observed S-MIF in Archean rocks, SSA, and laboratory photochemical experiments. Previous laboratory photochemical experiments employed static photochemical cells with SO2 partial pressures (pSO2) ranging from 6 mbar to over one bar [Farquhar et al., 2001; Masterson et al., 2011; Whitehill and Ono, 2012]. Isotopologue self-shielding is expected to take effect under these high SO2 column densities (>3.3 × 1019 molecules cm−2) [Lyons, 2007, 2008]. We used a flow-through photochemical cell to measure S-MIF at pSO2 as low as 0.1 mbar (corresponding column density at 3.8 × 1016 molecules cm−2), in order to test the contribution of isotopologue self-shielding. In addition, two photochemical cells are placed in series for some experiments to isolate the spectrum effects by only varying pSO2 of the front cell, while holding pSO2 of the rear cell constant. The experimental results are compared with isotope fractionation expected from synthetic isotopologue-dependent cross sections by Lyons  that account only for the effect of band-position shifts. By experimentally studying the effect of SO2 pressure, we also aimed to isolate the production of S-MIF by mechanisms other than self-shielding.
2.1 Photochemical Experiments
 Photolysis experiments were carried out using a dual-cell flow-through photochemical reactor illustrated in Figure 2. Each cylindrical glass photochemical cell (15 cm length, 5 cm inside diameter) is equipped with two UV grade windows (Corning 7980 grade, > 90% UV transmittance above 190 nm) sealed by o-rings, and inlet and outlet ports (made of 3/8" o.d. glass tubes). For dual-cell experiments, two identical flow-through cells were placed in series. Premixed SO2–N2 gas (100 ppm or 5% SO2) was diluted with pure N2 (UHP grade) using digital mass flow controllers and flowed continuously through the photochemical cell. Pressure inside the cell was monitored with capacitance manometers (MKS, 1000 torr full scale). The reaction cell was pumped continuously with a diaphragm pump through a needle valve and a vacuum regulator for experiments under pN2 less than 1 bar. Errors in pSO2 are estimated to be between 5 and 10%, based upon the precision and accuracy of the mass flow controllers and the manometer. A 150 W Xenon arc lamp (Model 6254, Newport) in a lamp housing (Model 67005, Newport) was used as a light source for all experiments. The irradiance spectrum between 190 and 400 nm was monitored with a UV spectrometer (SPM-002-BT, Photon Control, Burnaby, Canada), which was calibrated against Hg lamp lines for wavelength. The Xe arc lamp emits continuum radiation down to at least 190 nm (and possibly lower).
 While the excitation of SO2 by radiation in the 240 to 340 nm region (1A2, 1B1 ← X1A1) occurs in our experiments, the resulting low-energy excited states are rapidly quenched (within 4 to 300 ns for the singlet and triplet states, respectively) by collisions with N2 before participating in any chemical reactions such as self-reaction with SO2 [Sidebottom et al., 1972; Calvert et al., 1978; Whitehill and Ono, 2012].
 UV irradiation of the SO2 produces elemental sulfur (S0), sulfur trioxide (SO3), and residual SO2 with the overall reaction stoichiometry [Ustinov et al., 1988]:
 Upon photolysis, S0 and SO3 condensed inside the photochemical cell and on a glass wool filter placed at the outlet port, and were collected by rinsing with dichloromethane and deionized water, respectively. The S0 was crystallized from dichloromethane by evaporation and reduced to H2S by Cr reduction and precipitated as Ag2S by a procedure modified from Gröger et al. . SO3 rapidly hydrolyzes to sulfate in water and was precipitated as barium sulfate by the addition of a barium chloride solution. For two experiments (S-1020 and S-1021), residual SO2 was collected by bubbling effluent in a series of two bubblers, the first containing 60 mL of 80% isopropyl alcohol, and the second 60 mL of 3% hydrogen peroxide, to capture SO3 and residual SO2, respectively. After photolysis, these trapping solutions were acidified to pH < 4 with 6 N HCl, and sulfate was precipitated as barium sulfate. The first trap did not yield quantifiable (less than 1 mg) BaSO4, suggesting that the majority of the SO3 was trapped at the glass wool trap or on the walls of the photochemical cell. The barium sulfate was reduced to H2S and precipitated as Ag2S by the method described in Forrest and Newman .
2.2 Isotope Ratio Analysis
 A detailed description of the isotope ratio analysis method can be found in Ono et al. [2006b]. Approximately 2 mg of Ag2S was reacted with elemental fluorine gas in externally heated nickel tubes (at 300 °C) to form SF6. The product SF6 is purified by gas chromatography, introduced to a gas source isotope ratio mass spectrometer (MAT 253, Thermo-Fisher), and measured for four ion beams of SF5+. A microvolume (0.4 mL) procedure was used for small samples (<1 mg Ag2S). Replicate analyses (N = 28) of the reference Ag2S IAEA-S-1 yielded 2σ standard deviations of 0.26‰, 0.014‰, and 0.19‰ for δ36S, Δ33S, and Δ36S, respectively [Ono et al., 2012]. Typical reproducibility for microvolume analyses of IAEA-S-1 is 0.9‰, 0.08‰, and 0.8‰ for δ36S, Δ33S, and Δ36S, respectively (2σ for 14 replicate analyses).
 Sulfur isotope compositions are reported in conventional δ notation, defined as:
where xR is the ratio xS/32S (x = 33, 34, or 36) of sample (Rsa) and initial SO2 (Ri), respectively. The common multiplication factor of 1000 is omitted because it technically belongs to ‰ symbol [Coplen, 2011]. The magnitude of S-MIF is reported using capital delta notations calculated according to:
 We use the definitions above among various definitions because they are mathematically consistent with the δ notation (i.e., deviation of the ratio from the expected ratio) [see Kaiser et al., 2004].
2.3 Modeling the Effect of Isotopologue Spectrum Overlap
 The isotope fractionation due to spectrum overlap was modeled using the SO2 cross section reported by Lyons [2007, 2008] (Figure 1). These cross sections are based on the high-resolution (0.002 nm) 32SO2 cross-section measurements of Freeman et al. . Cross sections for minor isotopologues are estimated by shifting band positions of the 32SO2 cross sections. Lyons [2007, 2008] used the energy shift parameters adapted and extrapolated from ab-initio model study of Ran et al. . Figure 3 compares the isotopologue shift parameters from Lyons  and Ran et al.  as well as those from ab-initio wave-packet model calculation [Tokue and Nanbu, 2010], and experimental measurement [Danielache et al., 2008]. The isotope shift parameters used in this and previous studies agree well with the latest theoretical model by Tokue and Nanbu , but experimental data by Danielache et al.  show smaller magnitude energy shifts compared to theoretical values. A smaller isotope shift would produce a smaller effect of isotopologue self-shielding due to a higher degree of overlap in the fine vibrational structures.
 Following previous studies [e.g., Lyons, 2008; Ueno et al., 2009], photolysis quantum yields for 32SO2 are assumed to be wavelength (but not isotopologue) dependent above 205 nm and assumed to be unity below 205 nm [Okazaki et al. 1997]. The lamp output power (FXe) was approximated from the manufacturer's data sheet as a function of wavelength (λ in nm):
 The first and second factors are to correct the efficiency of the condenser (F/1) and rear reflector, respectively. Transmission of quartz windows (τw) are measured and fitted as:
 Total photolysis rates for each isotopologue are derived by integrating over wavelength between 190 and 220 nm and over the pathlength (15 cm) of the flow-through reactor. The major sources of error are the spectrum shape of the Xe lamp, and absorption by the optics and Schumann–Runge band of O2 in the optical path (which is applicable for wavelengths shorter than 195 nm) [Whitehill and Ono, 2012]. The estimated fractionation factor, however, is only a weak function of spectrum shape of the incident light [Lyons, 2007]. The overall magnitude of the photon flux affects photolysis yield but not isotope fractionation. We did not use the cross section from Danielache et al. . As discussed in Whitehill and Ono , the cross section from Danielache et al.  predicts negative Δ33S values under the light spectrum of Xe lamp, which is not reproduced by laboratory Xe lamp SO2 photolysis experiments.
 Self-shielding of 32SO2, in an ideal case, is expected to fractionate all the minor isotopes at the same magnitude. The isotope composition of the resulting product would follow δ33S ≃ δ34S ≃ δ36S. This is only true, however, when there is no spectrum overlap between major and minor isotope absorption lines, and when the underlying continuum is very weak, such as for the 12C16O absorption band at 105.2 nm. In the case of SO2 in the 190 to 220 nm range, the magnitude of isotope shift is relatively small compared to the peak width of individual vibrational bands (Figure 1-A and 1-B), and a substantial continuum is present at laboratory and atmospheric temperatures. These significant spectrum overlaps among different SO2 isotopologue lines cause mutual shielding (e.g., 32SO2 shielding 34SO2), which contributes mass-dependent fractionation. Due to a combination of self- and mutual-shielding effects, spectrum overlaps produce an isotope fractionation pattern that falls between ideal self-shielding (δ33S ≃ δ34S ≃ δ36S) and mass-dependent fractionation (i.e., δ33S/0.515 ≃ δ34S ≃ δ36S/1.9) (Lyons 2009).
 Elemental sulfur products (S0) are characterized by large δ34S values, as high as 140 ‰, and significant Δ33S and Δ36S anomalies up to 11.7 ‰ and −40.5 ‰, respectively (Table 1; Figure 4). The magnitude of the δ34S fractionation is a strong function of pSO2, increasing with increasing SO2 column density (Figure 4-A). The results for high pSO2 experiments (pSO2 > 2.5 mbar) closely match those of static cell experiments by Masterson et al. . Different pressures of nitrogen bath gas (0.25, 0.5, and 1 bar pN2) produced similar δ34S values (Figure 4-A). The pN2, however, affects the relationship between the two isotope fractionation factors, as lower pN2 produces higher δ33S/δ34S ratios (0.57 to 0.64 for 1 to 0.25 bar N2, respectively) and lower δ36S/δ34S values (1.41 to 1.68 for 1 to 0.25 bar N2, respectively) (Figures 4-B and 4-C). When photochemical cells are placed in series, the rear cell (held at 0.1 mbar pSO2) consistently yields larger δ34S values compared to the front cell. The rear cell did not produce a sufficient quantity of S0 for isotope analysis when the front cell pSO2 was higher than 0.3 mbar, since the majority of photons were absorbed in the front cell.
Table 1. Results of Flow-through SO2 Photolysis Experiments. All Isotope Ratios Are Normalized Against That of Starting SO2
Flow Rate (sccm)
These are pSO2 of the front cell. Rear cell pSO2 is held constant at 0.1 mbar
S0, Front Cell
S0, Rear Cell
SO3 Front Cell
SO3 Rear Cell
 Figure 4 also shows the results of the model calculation to be compared with experimental results. The contribution of spectrum overlap is expected for pSO2 ranging from 0.01 mbar to 10 mbar. Above 10 mbar pSO2 (column density > 1018 molecules cm−2), the value of δ34S reaches the high pSO2 limit (Figure 4-A). The model calculation and experimental results [including those by Masterson et al., 2011] show this saturation effect at high pSO2, which is expected purely from spectrum overlap. The agreement, however, is not quantitative since the model significantly overestimates the δ34S effect by a factor 1.9. Dashed lines in Figure 4-A represent the modeled δ34S values multiplied by the factor 1/1.9 to show this difference. Possible causes of this discrepancy will be discussed.
 The model and experimental results agree well for δ33S/δ34S and δ36S/δ34S ratios; both show isotope fractionation is rather closer to mass-dependent fractionation (δ33S/0.515 ≃ δ34S ≃ δ36S/1.9) than ideal self-shielding (δ33S ≃ δ34S ≃ δ36S) (Figures 4-B and 4-C). This is an expected result due to the significant spectrum overlap among various isotopologue of SO2 (i.e., mutual shielding) (Figure 1-B) [Lyons, 2009]. The experimental results yield δ33S/δ34S ratios from 0.57 to 0.64 and δ36S/δ34S ratios from 1.41 to 1.68. These experimental values agree well with the model results, which show δ33S/δ34S from 0.55 to 0.67 and δ36S/δ34S from 1.61 to 1.85. We cannot model the effect of pN2, as pressure-broadening parameters have not been reported for the 190–220 nm regions.
4.1 Source of S-MIF During Laboratory SO2 Photolysis in the 190 to 220 nm Region
 The following reactions represent the photolysis of SO2 under our experimental conditions:
where M is the third body molecule (N2 in our experiments) [Farquhar et al., 2001; Okabe, 1978]. Reactions ((R2)), ((R3)), and ((R4)) are favored under high pSO2 (> ca. 1 mbar), while ((R5)) and ((R6)) become important in low pSO2 experiments. Atomic sulfur from reactions ((R4)) or ((R6)) polymerizes and forms S0 aerosols. The above reactions do not include the excitation of SO2 by radiation in the 240 to 340 nm region (1A2, 1B1 ← X 1A1), since the resulting low-energy excited states are rapidly quenched by N2 [Sidebottom et al., 1972; Calvert et al., 1978; Lyons, 2009; Whitehill and Ono, 2012]. Previous static cell experiments used a 200 ± 35 nm band-pass filter to isolate the photochemistry of SO2 excited in the 190 to 220 nm versus the 240 to 340 nm absorption region [Whitehill and Ono, 2012]. We did not use the band-pass filter in this study because it would significantly limit the number of available photons (maximum transmission of the band-pass filter is 30%). The isotope pattern observed in this study (large δ34S values and small δ33S/δ34S, negative Δ36S/Δ33S values) is consistent with the 190 to 220 nm excitation band origin of the S-MIF. Isotope signatures for the 240–340 nm band (δ33S/δ34S >1 and Δ36S/Δ33S = 0.64) [Whitehill and Ono, 2012] are distinctly different from the isotope pattern observed in this study, suggesting only a minor (if not entirely absent) contribution from the excitation in the 240 to 340 nm regions.
 The qualitative agreement between measured S-MIF in the elemental S products and estimated S-MIF from spectrum overlap suggests a critical and significant contribution of isotopologue self-shielding for the observed S-MIF during SO2 photolysis under 190 to 220 nm excitation. Similarity between cross-section calculations and experimental results include (1) small S-MIF (and small δ34S values) at low pSO2, increasing with increasing pSO2 and reaching saturation at an SO2 column density of 1018 molecules cm−2, (2) δ33S/δ34S and δ36S/δ34S relationships, and (3) dual-cell experiments showing the change of S-MIF in the rear cell by changing only the pSO2 of the front cell. The model calculation, however, overestimates the δ34S value by a factor of 1.9. This could be, in part, due to different temperatures for the cross-section measurement [213 K, Freeman et al., 1984] compared to our photochemical experiment (ca. 300 K), as well as a possible effect of chemistry following photolysis (i.e., R2 to R6) as discussed below. Higher temperature may affect the population of rotational energy levels and affect both the width and the amplitude of the absorption bands.
 Our experimental results indicate that Δ33S/δ34S and Δ36S/δ34S relationships also depend on pN2 (low pN2 favors larger Δ33S and Δ36S anomalies), consistent with the results of static cell experiments for pSO2 with added helium [Masterson et al., 2011]. If observed S-MIF is indeed predominantly caused by self-shielding, the effect of bath gas pressure can be explained by pressure broadening of the absorption features at the low-energy side of the spectrum (>210 nm), where the fluorescence lifetime of the excited state SO2 is comparable to the collision frequency [Katagiri et al., 1997]. Measurements of the pressure-broadening factor may provide key insights into the effect of bath gas on the S-MIF pattern. Masterson et al.  observed decreasing magnitude of MIF at high pSO2 (>30 mbar) by static cell experiments. While this may appear contradictory to our results showing smaller magnitude MIF at lower pSO2 (<1 mbar), it is likely that the excess SO2 at very high pSO2 (>30 mbar) would act as a bath gas, resulting in the observed lower S-MIF anomaly.
 The observed pSO2 dependence provides critical insights into the potential contribution to MIF from mechanisms other than self-shielding. The S-MIF may originate—independent of a self-shielding mechanism—from quantum efficiencies for photolysis that are isotopologue dependent, such as nonadiabatic surface crossings [Zmolek et al., 1999; Muskatel et al., 2011] or from isotopologue-dependent absorption line strengths [Danielache et al., 2008]. Our experimental results suggest that these two effects make only a minor contribution to S-MIF (Δ33S < 2.5 ‰) under our experimental conditions, because both mechanisms should be effective even at the lowest optical density. This does not, however, exclude the possibility of S-MIF in nature at much lower pSO2. One also needs to be cautious in applying experimental results to natural conditions because sulfur isotope exchange reactions, such as,
may be important and effectively minimize the S-MIF signatures in SO (x = 33, 34, or 36). Isotope exchange kinetics, however, are expected to be first order with respect to pSO2. The saturation of 34S fractionation at 10 mbar pSO2 (Figure 4-A) suggests that the isotope exchange reaction ((R7)) is relatively slow: if R7 were fast, one would expect to see decreasing δ34S with increasing pSO2 above 10 mbar.
 Other than SO2 photolysis itself ((R1)), any one of the reactions above (R2 to R6) could contribute to mass-dependent or mass-independent fractionation. Our experimental results can be used to eliminate some reactions from all the potential S-MIF source reactions. Small isotope effects at low pSO2 suggest that ((R5)) and ((R6)) are not the main source of MIF, as these two reactions are favored under low pSO2. The SO2 oxidation reaction ((R2)) is mass dependent because the product SO3 is fractionated only mass dependently (Table 1). The reactions SO + SO3 ((R3)) or SO + SO ((R4)) are potential sources for S-MIF. Results of our experiments cannot fully exclude the possibility of S-MIF during ((R3)) and ((R4)). It is difficult to test the possibility, however, without knowing the mechanism and predicted isotope pattern (e.g., Δ33S/δ34S or Δ36S/Δ33S ratios). The SO + SO reaction ((R4)) may produce SO-dimer intermediates, of which some of the low electronic states are symmetric, which could potentially produce symmetry-dependent MIF [Lyons, 2008]. Symmetry-dependent MIF, in an ideal case, would produce nearly equal enrichment in all minor isotopes (i.e., δ33S ≃ 1.04 δ34S ≃ δ36S, where the factor 1.04 accounts for the symmetric species with two 34S) and result in the Δ36S/Δ33S ratios of −1.6 [Ono et al., 2009]. This is much higher than what we have measured.
4.2 Implication for the MIF in SSA
 Significant SO2 isotope self-shielding is expected when the SO2 column density exceeds ca. 1017 molecules cm−2 (Figure 4-A). During large explosive volcanic events, such as the Mt. Pinatubo (1991) eruption, the SO2 column density can reach and exceed these levels. The SO2 column density following the Pinatubo (1991) eruption initially reached as high as 1019 molecules cm−2. Within two weeks, the SO2 cloud covered an area in excess of 1017 km2 with an SO2 column density above 1017 molecules cm−2 [e.g., Guo et al., 2004]. Therefore, isotopologue self-shielding is expected to occur during direct photolysis of SO2 following stratospheric volcanic events. In support of this hypothesis, S-MIF signatures (i.e., Δ33S/δ34S and Δ36S/Δ33S ratios) produced by our experiments agree well with those for SSA trapped in polar ice [Savarino et al., 2003; Baroni et al., 2007, 2008] (Figures 5-C and 5-F).
 The isotope effect in the 190 to 220 nm region is exclusive to the stratosphere, since radiation in the 190 to 220 nm range is only available above 20 km due to absorption by O2 and O3 in this wavelength region [e.g., Farquhar et al., 2001]. This is consistent with the observation of S-MIF signatures for SSA exclusively from large stratospheric eruptions. Preservation of the S-MIF signature, however, remains problematic, since reaction with O2 rapidly oxidizes SO back to SO2.
Instead of SO2 photolysis ((R1)), Savarino et al.  suggest that the S-MIF in SSA sourced from photo-oxidation of SO2:
where *SO2 is excited state SO2 produced by excitation under 240 to 330 nm absorption bands [Calvert et al., 1978; Chung et al., 1975; Savarino et al., 2003]. Transfer of sulfur isotope signatures from *SO2 to SO3 would allow preservation of the S-MIF signature in *SO2. A recent photochemical study, however, showed that the sulfur in SO3 in reaction ((R9)) is largely (if not exclusively) derived from ground state SO2 [Whitehill and Ono, 2012]. The study also showed that the 240 to 330 nm excitation band produces positive Δ36S/Δ33S values (~0.63) rather than the negative values observed in SSA, suggesting photo-oxidation ((R9)) is not a likely candidate for S-MIF in modern SSA.
 The preservation of SO isotope signatures (rather than through R9), thus, implies the presence of an unknown reaction that preserves the S-MIF signatures in SO. A reaction such as
is only postulated by some researchers [Myerson et al., 1957; Wood and Heicklen, 1971; Black et al., 1982], but may provide a mechanism to preserve S-MIF signatures. The extent that reaction ((R10)) may contribute to sulfate formation can be estimated by comparing the kinetics of SO2 oxidation by OH:
At steady state with respect to SO, the fraction of SO3 produced from SO (fSO) is
where k8, k10, and k11 are the rate constants for R8, R10, and R11, respectively, and J1 is the photolysis rate of SO2. As an example of SO2 photolysis in a stratospheric volcanic plume, we consider a total SO2 column depth of 10 DU (2.7 × 1017 cm−2), which is distributed at height 27 ± 7 (FWHM) km [e.g., Aquila et al., 2012]. The model result shows that a rate constant k10 of 10−36 cm6/s (which is the maximum rate suggested by Black et al. ) can allow R10 to contribute up to 10% of SO3 formation at above 30 km (Figure 6). A smaller value of k10 will decrease this contribution as shown in Figure 6. The maximum Δ33S value observed for sulfate aerosol in polar ice [1.4 ‰, Baroni et al., 2007] is only 12% of the maximum Δ33S value observed in this study (11.8 ‰), suggesting that the contribution of SO2 photolysis of 12% or less can explain the polar ice S-MIF signatures (i.e., SO2 + OH reaction is still a dominant source of SO2 oxidation).
 The presence of ash, ice particles, and other molecular constituents in the volcanic plume (e.g., HCl, H2S) could modify sulfur chemistry in several ways. First, the fraction of O2 in the plume will be somewhat lower than in clean stratospheric air. Second, O3 and OH will be lower, possibly substantially lower, than in the background stratosphere due to surface reactions on plume particles. Lower OH will decrease the sulfate formation rate by reaction ((R11)) and increase the fraction of sulfate produced from SO (i.e., the second term in equation ((6)) increases). Faster conversion of SO to sulfate ((R10)) versus SO2 to sulfate ((R11)) during initial plume chemistry, would be consistent with the observation by Baroni et al.  that SSA with Δ33S > 0 was deposited first in Antarctic snow, followed by SSA with Δ33S < 0. Finally, reaction ((R10)) is likely to have a large activation energy and could be entirely negligible at stratospheric temperatures. Heterogeneous addition of SO and O2 on ash or ice particles,
could be rapid compared to gas-phase oxidation of SO2.
 Detailed modeling of stratospheric sulfur chemistry is beyond the scope of this study. Understanding the production and preservation of S-MIF signatures will provide a unique constraint on the origin and fate of stratospheric sulfur aerosols, which have a significant impact on global climate by compensating the greenhouse warming effects [Turco et al., 1982; Robock et al., 2008].
4.3 Implication to Archean S-MIF
 Some well-preserved Archean rocks show characteristic Δ33S/δ34S and Δ36S/Δ33S ratios of 0.9 and −0.9, respectively [e.g., Ono et al., 2003; Kaufman et al., 2007; Zerkle et al., 2012]. The ranges of Δ33S/δ34S and Δ36S/Δ33S ratios produced by our experiments (0.086 ± 0.035 and −4.6 ± 1.3, respectively) do not match these main Archean arrays, suggesting that the main source reaction for Archean S-MIF may not be linked to our experimental results (Figures 4-A and 4-D). It is, however, conceivable that the isotopologue self-shielding could have made a secondary contribution to the structure of Δ36S/Δ33S record when high SO2 column density was achieved after large volcanic eruptions. Studies of rocks from Neoarchean age (2.5 to 2.65 Ga) have shown several stratigraphic intervals with lower Δ36S/Δ33S ratios of −1.5 compared to a more common value of −0.9 [Kaufman et al., 2007; Zerkle et al., 2012]. In addition, data for rocks between 3.2 and 2.8 Ga are characterized by relatively small Δ33S values and with more negative Δ36S values (Figure 4-E). The cause of these secular changes in the structure of Archean S-MIF has been attributed to a variety of changes in atmospheric chemistry [Farquhar et al., 2007], including ephemeral oxidation [Ono et al., 2006a; Ohmoto et al., 2006; Kaufman et al., 2007], the development of organic haze [Domagal-Goldman et al., 2008; Zerkle et al., 2012], or changes in volcanic gas SO2/H2S ratios [Halevy et al., 2010]. The results of this study suggest that the observed Δ36S/Δ33S trend can be explained by a combination of the main Archean S-MIF reaction (with Δ36S/Δ33S ≈ −0.9) and the isotope self-shielding due to increased volcanic SO2 loading (with Δ36S/Δ33S ≈ −4.6). In particular, the Mesoarchean increase of SO2 loading is consistent with abundant detrital and diagenetic pyrite in the Witwatersrand Supergroup, despite presumably low sulfate levels of the Archean oceans [Habicht et al., 2002]. Increased SO2 loading would have had significant consequences on the early Earth's climate, because a temporal increase of sulfate in surface environments would suppress microbial methanogenesis by competing for H2 with sulfate reducers [Lovley and Klug, 1983]. The collapse of methanogens would have triggered the oldest known glaciations in the Witwatersrand-Pongola basins at ca. 2.9 Ga [Young et al., 1998; Ono et al., 2006a].
 Sulfur isotope effects during the UV photolysis of SO2 under a broadband light source were investigated with a flow-through photochemical reactor. The S-MIF signature and the δ34S values of S0 products increase with increasing SO2 pressure, but saturate at a column density of 1018 molecules cm−2. The SO2 column density dependence, large δ34S values, and Δ33S/δ34S and Δ36S/δ34S relationships suggest an important contribution of isotopologue self-shielding to the observed mass-independent isotope effect. Results from dual-cell experiments further support this conclusion. The measured isotope pattern, in particular the Δ36S/Δ33S ratios, show good agreement with data for SSA, suggesting that photolysis of SO2 in the 190 to 220 nm region following large volcanic events could be the dominant source of the modern S-MIF in SSA. This implies there is a currently unknown mechanism for preserving the isotope signature of SO formed from SO2 photolysis in the stratosphere. Although the results do not agree with the main Archean S-MIF array, SO2 self-shielding could have contributed to the Δ36S/Δ33S variations during parts of the Archean.
 Authors thank William J. Olszewski and Kat Thomas for assistance in sulfur isotope analysis, and Eliza Harris for proofreading. This work was supported by the NASA Exobiology Program, Grant No. NNX10AR85G to Ono and NNX10AR80G to Lyons. The authors also thank Sebastian Danielache and two anonymous reviewers for constructive comments to the earlier version of the manuscript.