Unprecedented Ambient Sulfur Trioxide (SO3) Detection: Possible Formation Mechanism and Atmospheric Implications

Sulfur trioxide (SO3) is a crucial compound for atmospheric sulfuric acid (H2SO4) formation, acid rain formation, and other atmospheric physicochemical processes. During the daytime, SO3 is mainly produced from the photo-oxidation of SO2 by OH radicals. However, the sources of SO3 during the early morning and night, when OH radicals are scarce, are not fully understood. We report results from two field measurements in urban Beijing during winter and summer 2019, using a nitrate-CI-APi-LTOF (chemical ionization-atmospheric pressure interface-long-time-of-flight) mass spectrometer to detect atmospheric SO3 and H2SO4. Our results show the level of SO3 was higher during the winter than during the summer, with high SO3 levels observed especially during the early morning (∼05:00 to ∼08:30) and night (∼18:00 to ∼05:00 the next day). On the basis of analysis of SO2, NOx, black carbon, traffic flow, and atmospheric ions, we suggest SO3 could be formed from the catalytic oxidation of SO2 on the surface of traffic-related black carbon. This previously unidentified SO3 source results in significant H2SO4 formation in the early morning and thus promotes sub-2.5 nm particle formation. These findings will help in understanding urban SO3 and formulating policies to mitigate secondary particle formation in Chinese megacities.


39
Text S1. Sampling site 40 Our field measurements were conducted at BUCT (Beijing University of Chemical Technology) sampling 41 site (39.94° N, 116.30° E), which was located on the west campus of BUCT 1 . This site was located at the 42 roof of a teaching building, which is approximately 15 m above the ground level. Around 130 m to the north 43 and 550 m to the west are Zizhuyuan Road and West Third Ring Road, respectively. The "West Third Ring

44
Road" is one of the main "Ring" roads in Beijing. Besides the influence of traffic, this site is also affected by 45 local commercial and residential activities. Therefore, the BUCT monitoring site is representative of an urban

49
Aerodyne Research Inc, USA and Tofwerk AG, Switzerland) mass spectrometer was deployed to detect SO 3 50 and gas-phase sulfuric acid. The CI-APi-LTOF consists of an optimized inlet for chemical ionization (CI-51 inlet) 2, 3 and an APi-LTOF mass spectrometer with the mass resolving power of ~10000 Th/Th. Nitrate ions 52 (NO 3 -·(HNO 3 ) n , n=0,1 and 2) were used as reagent ions. The working principle of nitrate-CI-APi-LTOF has 53 been described in many previous studies 2, 4 .

54
In the charging part of CI-inlet, the nitrate ions are electrostatically pushed into ambient sample flow to react 55 with SO 3 and H 2 SO 4 . In the CI-inlet, the ion-molecule reaction time was ∼200 ms 4 . Pure air originated from 56 a pure air generator (Aadco 737) was used as the sheath air. Ambient air was sampled into the CI-inlet through 57 a ¾ inch stainless steel tube. A 0.8 L min -1 flow from the mixed flow entered the APi-LTOF. Data of CI-

58
APi-LTOF were acquired at 5 s time resolution and analyzed with a MATLAB tofTools package 5 .

59
Text S3. Detection of sulfuric acid with nitrate reagent ions

63
where n = 0, 1 or 2 and j = 0 or 1. Due to H 2 SO 4 being a stronger acid, de-protonation occurs during its 64 collision with nitrate ions. Thus, H 2 SO 4 molecules can be detected as de-protonated monomer ions and cluster 65 ions with HNO 3 in CI-APi-LTOF.

66
To take the variation in the total reagent ions into account, neutral sulfuric acid was quantified according to 67 the following equation: where C (in units of cm -3 ) is a calibration coefficient from in-situ calibration.

70
Text S4. Calibration experiment for SO 3

71
The calibration of SO 3 was implemented by introducing a known amount of gaseous SO 3 produced by the 72 reaction of SO 2 and OH radicals formed by UV photolysis of water vapour, which is similar to the method 73 for sulfuric acid calibration in the previous literature (i.e. Kürten et al., 2012) 6 . During the calibration 74 experiment, a 10 L•min -1 N 2 flow, a 100 mL•min -1 pure air flow, a 300 mL•min -1 SO 2 flow and a set of 20 -3 75 400 mL•min -1 saturated water vapour flow were mixed together as the calibration sampling flow. Then, the 76 mixed flow was exposed to 184.9 nm UV light to produce OH radicals which reacted with SO 2 to produce 77 SO 3 . The schematic of the experimental setup was shown in Figure S2. The UV lamp was turned on in an N 2 78 environment at least one hour before the actual calibration measurement in order to achieve a stable light 79 intensity. During the calibration, the box was flushed with a 1 -2 L•min -1 dry N 2 flow to avoid the absorption  Figure S3. The correlation between 84 normalized SO 3 signals measured by CI-APi-LTOF and SO 3 concentrations formed by photo-oxidation of 85 SO 2 by OH radicals was depicted in Figure S4. After taking the diffusion loss of the sampling line into 86 account, a calibration coefficient of 1.7 × 10 10 molecule cm -3 was obtained. The diffusion loss was assumed 87 as same as that of sulfuric acid.

88
Similar to the quantification of sulfuric acid 2 , to regard the variation in the total reagent ions, SO 3 was 89 quantified according to: The main charging ions are HNO 3 ·NO 3and NO 3 -. The corresponding ratios (collision rate of H 2 SO 4 / 115 collision rate of SO 3 ) are 3.07, 3.12, 1.39 and 1.42, respectively. The newer ion-molecule collision rate 116 parametrization (Su & Chesnavich, 1982) thus predicts a difference of a factor of 3, whereas the older one 117 only predicts a difference of about a factor of 1.5. A non-polar molecule collides much slower with an ion 118 than a strongly polar molecule. The product ions H 2 SO 4 ·NO 3 -(binding Gibbs free energy of -32.6 kcal/mol 119 at the wB97xd/aug-cc-pVTZ level) and SO 3 ·NO 3 -(binding Gibbs free energy -28.4 kcal/mol at the same 120 level) are very strongly bound and stable. The sensitivity of SO 3 could be less than that of H 2 SO 4 by a factor 121 of 3. Thus, a factor of 3 difference would lead to an underestimation of SO 3 if the calibration factor for 122 H 2 SO 4 was used to quantify SO 3 .

123
Text S6. Quantum chemical calculations 124 Quantum chemical calculations demonstrate that the SO 3 ·(NO 3 -) cluster is very strongly bound compared to 125 the HNO 3 ·(NO 3 -) cluster (Table S1). The difference in binding is over 10 kcal/mol both in electronic and free 126 energies, with the more rigorous coupled-cluster methods predicting a larger difference than the density 127 functional theory method used here. SO 3 molecules will thus be very efficiently charged by nitrate ions in a 128 nitrate-CI-APi-LTOF instrument, as the charge transfer reaction HNO 3 ·(NO 3 -) + SO 3 → SO 3 ·(NO 3 -) + HNO 3 129 is highly favourable. Furthermore, the thermal evaporation rate of SO 3 ·(NO 3 -) clusters in the CI-inlet will be 130 negligible, and also the (non-thermal) fragmentation of the cluster in the ion optics of the instrument will be 131 considerably smaller than for example that of the (H 2 SO 4 ) 2 ·HSO 4cluster, which has binding energy 132 comparable to HNO 3 ·(NO 3 -) 9 . All of this supports the hypothesis that the instrument sensitivity toward SO 3 133 will be very high. The optimized structure of the SO 3 ·(NO 3 -) cluster is shown in Figure S5. The strength of 134 the O 3 S…ONO 2interaction is reflected in the relatively short S…O distance.

135
The thermodynamics of the SO 3 · H 2 O + NO 3 -· (HNO 3 ) → SO 3 · (NO 3 -) · H 2 O + HNO 3 reaction (R5) was 136 assessed at the wB97X-D/aug-cc-pVTZ level. Three different hydrogen bonding patterns (conformers) for 137 SO 3 ·(NO 3 -)·H 2 O were assessed, with the H 2 O molecule placed either close to the SO 3 moiety, the NO 3moiety, 138 or in a bridging position between the two. The latter structure, where H 2 O H-bonds to both O-S and O-N 139 oxygen atoms, was found to be the lowest in free energy (at 298 K) (see Figure S6), though the differences 140 between conformers were fairly small (below 2 kcal/mol). By comparison to the results in Table S1, it is 141 likely that higher-level energy corrections (omitted here for computational reasons) would lead to an even 142 more negative (favourable) reaction free energy.

160
Mass concentrations of PM 2.5 were recorded by a TEOM (tapered element oscillating microbalances) monitor.

188
Combining Eqs. 5, 6 and 7, we obtain Eq.8 to calculate the concentration of BC: where k is a loading effect parameter.

191
The mass concentrations of non-refractory PM 2.5 including sulfate concentration were measured by an online

210
Text S10. Calculation of condensation sink

211
Condensation sink (CS) describes the condensing vapour sink caused by the particle population 24 :

213
where D is the diffusion coefficient of the condensing vapour (usually assumed to be sulfuric acid), and β m , 214 dp is the transitional regime correction factor.

215
Text S11. Source identification of SO 2 in winter

216
In winter, the median concentrations of SO 2 exhibited similar diurnal trends as SO 3 (Figure 2A). A similar 217 diurnal variation of SO 2 with an early morning peak has already been reported from another site in urban

218
Beijing 25 . We also studied the evaluation of median mixing layer height (MLH) together with the diurnal 219 trend of the median concentration of SO 2 , SO 3 and UVB ( Figure S8). The median MLH was merely 200-300 220 m, and stable in the morning (~05:00 to ~08:30). Many studies have reported regional SO 2 is tightly linked