the Creative Commons Attribution 3.0 License. Atmospheric Chemistry

Secondary organic aerosol (SOA) formation from the reaction of isoprene with nitrate radicals (NO 3 ) is investigated in the Caltech indoor chambers. Experiments are performed in the dark and under dry conditions (RH&lt10%) using N 2 O 5 as a source of NO 3 radicals. For an initial isoprene concentration of 18.4 to 101.6 ppb, the SOA yield (defined as the ratio of the mass of organic aerosol formed to the mass of parent hydrocarbon reacted) ranges from 4.3% to 23.8%. By examining the time evolutions of gas-phase intermediate products and aerosol volume in real time, we are able to constrain the chemistry that leads to the formation of low-volatility products. Although the formation of ROOR from the reaction of two peroxy radicals (RO 2 ) has generally been considered as a minor channel, based on the gas-phase and aerosol-phase data it appears that RO 2 +RO 2 reaction (self reaction or cross-reaction) in the gas phase yielding ROOR products is a dominant SOA formation pathway. A wide array of organic nitrates and peroxides are identified in the aerosol formed and mechanisms for SOA formation are proposed. Using a uniform SOA yield of 10% (corresponding to M o ≅10 μg m −3 ), it is estimated that ~2 to 3 Tg yr −1 of SOA results from isoprene+NO 3 . The extent to which the results from this study can be applied to conditions in the atmosphere depends on the fate of peroxy radicals in the nighttime troposphere.

and it has been suggested that the reaction with nitrate radicals, NO 3 , is a major contributor to isoprene decay at night (Curren et al., 1998;Starn et al., 1998;Stroud et al., 2002;Steinbacher et al., 2005). Typical NO 3 radical mixing ratios in boundary layer continental air masses range between ∼10 to ∼100 ppt (Platt and Janssen, 1995;Smith et al., 1995;Heintz et al., 1996;Carslaw et al., 1997). However, concen- 15 trations as high as several hundred ppt have been observed over northeastern USA and Europe (Platt et al., 1981;von Friedeburg et al., 2002;Brown et al., 2006;Penkett et al., 2007). Given the rapid reaction rate between isoprene and NO 3 radicals (k NO3 =7×10 −13 cm 3 molecule −1 s −1 at T =298 K, IUPAC), it is likely that NO 3 radicals play a major role in the nighttime chemistry of isoprene. 20 The kinetics and gas-phase products of the isoprene-NO 3 reaction have been the subject of several laboratory and theoretical studies (Jay and Stieglitz, 1989;Barnes et al., 1990;Skov et al., 1992;Kwok et al., 1996;Berndt and Böge, 1997;Suh et al., 2001;Zhang et al., 2002;Fan et al., 2004). In many studies, C 5 -nitroxycarbonyl is identified as the major first-generation gas-phase reaction product (Jay and Stieglitz, Little is known beyond the formation of the first-generation products of the reaction of NO 3 with isoprene. The isoprene nitrates and other first-generation products still contain a double bond, and it is likely that the further oxidation of these species will lead to low volatility products that can contribute to SOA formation at nighttime. In this work, SOA formation from the reaction of isoprene with NO 3 radicals is in-10 vestigated. Laboratory chamber experiments are performed in the dark using N 2 O 5 as a source of NO 3 radicals. Aerosol yields are obtained over a range of initial isoprene concentrations (mixing ratios). By examining the time evolutions of aerosol volume and different intermediate gas-phase products, we are able to constrain the chemistry that leads to the formation of low-volatility products. Mechanisms for SOA formation are 15 proposed and chemical composition data of the SOA formed are also presented.

Experimental section
Experiments are carried out in the Caltech dual 28 m 3 Teflon chambers. A detailed particles cm −3 , with a geometric mean diameter of ∼50 nm. The initial seed volume is 10-12 µm 3 cm −3 . In some experiments, no seed particles are added and aerosols are formed via nucleation. After introduction of the seed aerosols (in seeded experiments), a known volume of isoprene (Aldrich, 99%) is injected into a glass bulb and introduced into the chamber by an air stream. The mixing ratio of isoprene is monitored with a 10 gas chromatograph equipped with a flame ionization detector (GC-FID, Agilent model 6890N). The column used is a bonded polystyrene-divinylbenzene based column (HP-PLOT Q, 15 m×0.53 mm, 40 µm thickness, J&W Scientific). The oven temperature is held at 60 • C for 0.5 min, ramped at 35 • C min −1 to 200 • C, and held constant for 3.5 min. The thermal decomposition of N 2 O 5 serves as a source of NO 3 radicals in these 15 experiments. N 2 O 5 is prepared and collected offline by mixing a stream of nitric oxide (≥99.5%, Matheson Tri Gas) with a stream of ozone in a glass bulb (Davidson et al., 1978): Ozone is generated by flowing oxygen through an ozonizer (OREC model V10-0, Phoenix, AZ) at ∼1 L min −1 . The mixing ratio of ozone is measured by a UV/VIS spectrometer (Hewlett Packard model 8453) to be ∼2%. The flow rate of nitric oxide into the glass bulb is adjusted until the brown color in the bulb disappears. The N 2 O 5 is 25 trapped for 2 h in an acetone-dry ice bath ( Interactive Discussion dangerous) as a white solid, and stored between experiments under liquid nitrogen temperature. Once the seed and isoprene concentrations in the chamber stabilize, reaction is initiated by vaporizing N 2 O 5 into an evacuated 500 mL glass bulb and introduced into the chamber with an air stream of 5 L min −1 . The amount of N 2 O 5 injected is estimated based on the vapor pressure in the glass bulb, which is measured using 5 a capacitance manometer (MKS); this amount corresponds to an initial mixing ratio of ∼1 ppm in the chamber. The thermal decomposition of N 2 O 5 forms NO 2 and NO 3 radicals. Impurities in the N 2 O 5 starting material are quantified by FTIR spectroscopy (Nicolet model Magna 550). N 2 O 5 is vaporized into an evacuated pyrex cell (18 cm in length and 300 cm 3 ) with CaF 2 windows. Spectra are collected immediately upon 10 addition over the 1000 cm −1 to 4000 cm −1 window allowing for quantification of NO 2 (1616 cm −1 band) and HNO 3 (3550 cm −1 band) impurities. A custom-modified Varian 1200 Chemical Ionization Mass Spectrometer (CIMS) is used to continuously monitor the concentrations of various gas-phase intermediates and products over the course of the experiments. The CIMS instrument is operated 15 mainly in negative mode using CF 3 O − as a reagent ion, which selectively clusters with compounds having high fluorine affinity (e.g., acidic compounds and many hydroxyand nitroxy-carbonyls), forming ions at m/z MW+85. In some experiments, the CIMS instrument is also operated in the positive mode using H 2 O·H + as a reagent ion forming ions at m/z MW+1. The ionization schemes are as follows: 20 Negative chemical ionization: CF 3 O − +HB ->CF 3 O − ·HB Positive chemical ionization: H 2 O·H + +D->D·H + +H 2 O (where D has a proton affinity >H 2 O) The term "product ion" is used throughout this manuscript to describe the ionized products formed through the above chemical reaction schemes. Typically, we scan Introduction Aerosol physical and chemical properties are monitored by many instruments. Realtime particle mass spectra are obtained with an Aerodyne quadrupole Aerosol Mass Spectrometer (Q-AMS) (Jayne et al., 2000). A Particle-Into-Liquid Sampler (PILS, Brechtel Manufacturing, Inc.) coupled with ion chromatography (IC) is employed for quantitative measurements of water-soluble ions in the aerosol phase (Sorooshian et 5 al., 2006). Duplicate Teflon filters (PALL Life Sciences, 47-mm diameter, 1.0-µm pore size, teflo membrane) are collected from a selected number of experiments for offline chemical analysis. Filter sampling is initiated when the aerosol volume reaches its maximum value. Depending on the total volume concentration of aerosol in the chamber, the filter sampling time is 2-4 h, which results in ∼2-5 m 3 of total chamber air sampled.

10
Teflon filters used for high-resolution electrospray ionization-time-of-flight mass spectrometry (ESI-TOFMS) analysis are extracted in 5 mL of high-purity methanol (LC-MS CHROMASOLV-Grade, Sigma-Aldrich) by 45 min of sonication. Methanol sample extracts are then blown dry under a gentle N 2 stream (without added heat) once the filters are removed and archived at −20 • C. Dried residues are then reconstituted with 500 mL 15 of a 1:1 (v/v) solvent mixture of 0.1% acetic acid in water (LC-MS CHROMASOLV-Grade, Sigma-Aldrich) and 0.1% acetic acid in methanol (LC-MS CHROMASOLV-Grade, Sigma Aldrich). All resultant filter extracts are analyzed by a Waters ACQUITY ultra performance liquid chromatography (UPLC) system, coupled to a Waters LCT Premier XT time-of-flight mass spectrometer (TOFMS) equipped with an ESI source 20 that is operated in the negative (-) ionization mode. Detailed operating conditions for the UPLC/(-)ESI-TOFMS instrument have been described previously (Ng et al., 2007a). A Waters ACQUITY UPLC HSS column is selected to separate the SOA components because of its increased retention of water-soluble polar organics; separation is achieved as a result of trifunctionally-bonded (T3) C 18 alkyl residues on this col- 25 umn, which prevent stationary phase collapse when a 100% aqueous mobile phase is used and result in better retention of water-soluble polar organic compounds. In addition to the UPLC/(-)ESI-TOFMS analysis, all remaining Teflon filters are extracted and analyzed for total peroxide content (sum of ROOR and ROOH) by using an iodometric- Interactive Discussion spectroscopic method previously described by Surratt et al. (2006). To study the mechanism of SOA formation, in several experiments the experimental protocols are slightly modified: (1) An excess amount of isoprene (relative to N 2 O 5 concentration) is injected into the chamber to prevent the further reaction of first-generation gas-phase products, allowing these products to be detected more readily; (2) After the addition of isoprene, pulses of N 2 O 5 are introduced into the chamber to study the evolution of different intermediate gas-phase products; (3) With isoprene well mixed in the chamber, N 2 O 5 is introduced slowly to maximize the self-reaction of peroxy radicals (see Sect. 4.2). This is achieved by first injecting N 2 O 5 into a 65 L Teflon bag; then an air stream of 1 L min −1 is passed through the Teflon bag to introduce N 2 O 5 into the 10 chamber over a 7-h period. We refer to this as the "slow N 2 O 5 injection experiment"; and (4) With N 2 O 5 well mixed in the chamber, isoprene is introduced slowly to maximize the reaction between peroxy radicals and nitrate radicals (see Sect. 4.2). This is achieved by first injecting isoprene into a 65 L Teflon bag, and then introduced into the chamber with an air stream of 0.1 L min −1 for 7 h. We refer to this as the "slow isoprene Introduction within an hour after the introduction of N 2 O 5 . About 2.5 µg m −3 of inorganic nitrate is measured by PILS/IC, which agrees well with the amount of nitrates detected by Q-AMS. FTIR analysis indicates the presence of ∼10% HNO 3 and 4% NO 2 impurity in the N 2 O 5 prepared, thus the nitrates measured by PILS/IC and Q-AMS likely arise 5 from the partitioning or reactive uptake of gas-phase HNO 3 into the aerosol phase, or HNO 3 produced from heterogeneous hydrolysis of N 2 O 5 . As in the Q-AMS analysis, no organic species are detected in the filter samples collected from these blank experiments.
3.2 Aerosol yields 10 A series of experiments with different initial isoprene concentrations are carried out (these are referred to as "typical yield experiments" hereafter). The initial isoprene concentration ranged from 18.4 to 203.4 ppb. Figure 2 shows the reaction profile of the oxidation of an initial mixture containing 203.4 ppb isoprene. Since the chamber is NO x -free at the beginning of the experiment, once N 2 O 5 is introduced into the chamber 15 the equilibrium in Reaction (3) favors the formation of NO 3 . This generates a relatively high concentration of NO 3 radicals and results in rapid isoprene decay. Aerosol growth is observed and aerosol volume continues to increase even after all the isoprene is consumed. Owing to the rapid isoprene decay and the relatively long time between each GC measurement (12 min), the isoprene decay over time is captured only in experi-20 ments in which the initial isoprene concentration is >100 ppb. Based on the observed isoprene decay in these experiments and the isoprene-NO 3 rate constant k NO3 , the NO 3 concentration in the chamber is estimated to be ∼140 ppt.
The SOA yield of each experiment (Table 1) is shown in Fig. 3 Interactive Discussion aerosol mass is corrected for the amount of inorganic nitrate measured in each experiment. For convenience, SOA yields can be parameterized by a semi-empirical model based on absorptive gas-particle partitioning of two semivolatile products (Odum et al., 1996: in which Y is the aerosol yield, ∆M o is the organic aerosol mass produced, M o is the organic aerosol mass present (equal to ∆M o in chamber experiments with no absorbing organic mass present initially), α i is the mass-based gas-phase stoichiometric fraction for semivolatile species i , and K om,i is the gas-particle partitioning coefficient for species i . With this two-product model, Eq. (4) is fit to the experimental yield data (data 10 with ∆M o <100µg m −3 ) and the yield parameters obtained are: α 1 =0.089, α 2 =0.203, K om,1 =0.182 m 3 µg −1 , and K om,2 =0.046 m 3 µg −1 . For an organic aerosol mass of ∼10 µg m −3 , the aerosol yield is ∼10%. Also shown in Fig. 3 are aerosol yields from the slow isoprene/N 2 O 5 injection experiments. Since the PILS/IC is not employed in these experiments, in calculating SOA 15 yields it is assumed that the amount of inorganic nitrate formed in these slow injection experiments is roughly the same as that in other experiments. For the slow isoprene injection experiment, no isoprene is observed by GC-FID, indicating that once the isoprene enters the chamber, it is quickly consumed by reaction with NO 3 . The time profile of isoprene injection is obtained in a separate experiment, in which the same amount 20 of isoprene is added into the chamber without N 2 O 5 present. Assuming the amount of isoprene injected into the chamber is the same as the isoprene reacted, the amount of isoprene reacted over the course of the slow isoprene experiment can be deduced. As seen in Fig. 3, the SOA yield from the slow N 2 O 5 injection experiment is roughly the same as those in the other yield experiments; the yield from the slow isoprene injection Introduction Interactive Discussion carbon reacted, ∆HC) over the course of the slow N 2 O 5 injection experiment and the slow isoprene injection experiment are shown in Fig. 4. As hydrocarbon measurements are made with a lower frequency than particle volume, the isoprene concentrations shown are obtained by interpolating GC-FID measurements. In both experiments about 50 ppb of isoprene is consumed, the only difference being the order of isoprene/N 2 O 5 injection. From Fig. 4 it is clear that as the reaction proceeds, more aerosol is formed in the slow isoprene injection experiment for the same amount of isoprene reacted. However, the final SOA yield under the slow N 2 O 5 injection conditions is higher due to continued aerosol formation even after the complete consumption of isoprene. The presence of a "hook" at the end of the growth curve for the slow N 2 O 5 10 injection experiment indicates that further reactions are contributing to aerosol growth after isoprene is consumed (Ng et al., 2006). This is further discussed in Sect. 4.3.

Gas-phase measurements
The CIMS technique measures the concentrations of different gas-phase products over the course of the experiments. A series of experiments is carried out to study the 15 mechanisms of SOA formation by varying the relative amount of isoprene and N 2 O 5 injected and monitoring the time evolution of the intermediate products. Shown in Fig. 5 are the time profiles of three major gas-phase products and the corresponding aerosol growth from the excess isoprene experiment. In this experiment, ∼120 ppb of N 2 O 5 is first injected into the chamber, followed by the introduction of ∼800 ppb isoprene. 20 The initial concentration of isoprene is estimated based on the volume of the isoprene injected and the chamber volume. Once isoprene is injected, a number of product ions are formed immediately, with m/z 230, 232, and 248 being the most dominant ones. Several minor product ions at m/z 185, 377, and 393 are also observed (not shown). With the presence of excess isoprene, it is expected that the three major 25 products detected are first-generation products. Their further reaction is suppressed, as indicated by the relatively constant concentrations of the product ions once they are formed. At the end of the experiment, 725 ppb of isoprene is measured by GC-FID. A small amount of aerosol is formed instantaneously, likely from the condensation of relatively nonvolatile first-generation products, or from further generation products that are formed at a very rapid rate.
To study further the evolution of the gas-phase products, an experiment is performed in which pulses of N 2 O 5 are introduced into the chamber (with isoprene present) 5 (Fig. 6). The top panel shows the isoprene decay and aerosol formation; the middle panel shows the time profiles of the three major first-generation products (m/z 230, 232, and 248); the bottom panel shows the time profiles of three minor products (m/z 185, 377, and 393). In this experiment, 179 ppb of isoprene is first injected into the chamber, followed by the addition of 3 pulses of N 2 O 5 (∼120, 50, 210 ppb). The obser-10 vations after the addition of the first pulse of N 2 O 5 are similar to the excess isoprene experiment described above. With the addition of ∼120 ppb N 2 O 5 , 97 ppb of isoprene is reacted away, m/z 230, 232, and 248 are formed with concentrations of ∼53 ppb, 25 ppb, and 21 ppb, respectively. Because of the lack of authentic standards, the concentrations are uncertain. Because the sum of the ion concentrations derived from 15 our estimated sensitivities is equal to the reacted isoprene, our estimated sensitivity must represent a lower limit for the actual sensitivity of the CIMS technique to these compounds. Similar to the data in Fig. 5, the concentrations of these product ions stay relatively constant owing to the presence of excess isoprene. The minor products at m/z 185, 377, and 393, are formed with the concentrations of ∼1.5 ppb, 1 ppb, and 20 1 ppb, respectively. It is noted that the m/z 393 ion is formed with a relatively slower rate than all other product ions. A small amount of aerosol is observed. At t=15:40, a second pulse of N 2 O 5 (∼50 ppb) is introduced into the chamber and the remaining 82 ppb isoprene is completely consumed. As seen from Fig. 6, the concentrations of all intermediate products increase accordingly and more aerosol is produced. The relative 25 increase in the concentration of m/z 232 ion is not as high as would be expected if we assume the ratio of m/z 230 to 232 formed to be the same as in the first pulse of N 2 O 5 addition (i.e. ∼2:1). This indicates that some of m/z232 ion has reacted with NO 3 radicals. The last pulse of N 2 O 5 (∼210 ppb) is added at t=19:00. Since all isoprene Interactive Discussion has been consumed, the additional NO 3 radicals react mainly with the first-generation products, as indicated by the decay of m/z 230, 232, and 248, 185, 377, and 393 ions. Of all of the observed products, it appears that m/z 232 and 377 ions are the most reactive with NO 3 radicals, and their decays in excess NO 3 are strongly correlated with aerosol growth. The rest of the product ions display relatively slower decay 5 kinetics. The decay of the major product ion at m/z 230 does not appear to correlate with aerosol growth, as the concentration of the m/z 230 ion continues to decrease throughout the experiment but there is no further aerosol growth. Since the CIMS instrument has only 0.5 AMU resolution and it cannot distinguish products of similar or identical molecular weight, it is likely that many of observed ions comprise isomers 10 formed from the NO 3 attack at different positions. The fact that many of the observed product ions show two distinct decay time scales indicates that these isomers have substantially different reactivity towards NO 3 radicals.  Figure 7 shows the AMS spectrum of SOA formed in the typical yield experiments. Each mass fragment is normalized by the total signal. The SOA exhibits relatively high signals at m/z 30, 43, and 46. The signals at m/z 30 and 46 likely correspond to NO + (30) and NO + 2 (46) fragments from the nitrates in the aerosol. The spectrum shown in Fig. 7 is obtained when aerosol volume reaches its maximum value; the spectrum 20 obtained several hours after aerosol volume peaks shows minimal changes in the mass fractions of different fragments, indicating that the aerosol composition is not changing significantly over time. Figure 8 shows the mass spectrum of the slow N 2 O 5 injection experiment versus a typical yield experiment; Fig. 9 shows the mass spectrum of the slow isoprene injection  Figure 10 shows the representative UPLC/(-)ESI-TOFMS base peak ion chromatograms (BPCs) for different types of experiments conducted. The numbers denoted above the selected chromatographic peaks correspond to the most abundant 5 negative ions observed in their respective mass spectra. Comparison of the BPCs shown in Fig. 10 indicates that the compositions of the SOA are quite similar for the typical yield experiment, slow isoprene injection experiment, and the acid seed experiment, suggesting a common SOA formation pathway. The SOA composition from the excess isoprene experiment, however, is different from these experiments. This will be discussed further in Sect. 4.4. Accurate mass measurements for all ions observed by the UPLC/(-)ESI-TOFMS technique for a typical yield experiment are listed in Table 2. The error between the measured mass and theoretical mass is reported in two different ways, ppm and mDa. Overall, the error between the measured and theoretical masses is found to be less 15 than ±2 mDa and ±5 ppm, allowing for generally unambiguous identification of molecular formulae. None of the listed ions is observed in solvent blanks and control filters. By combining the elemental SOA composition (i.e. TOFMS suggested ion formula) data and the gas-phase data from CIMS, structures for each of the SOA components are also proposed. As shown in Table 2, the types of compounds formed 20 included nitroxy-organic acids, hydroxynitrates, nitroxy-organic peroxides (e.g. nitroxyhydroxyperoxides), and nitroxy-organosulfates. It should be noted that the data presented in Table 2 are also applicable to all other types of experiments conducted in this study; however, none of the organosulfates are observed in the nucleation experiments, consistent with previous work (Liggio et al., 2005;Liggio et al., 2006;Surratt et 25 al., 2007ab;Iinuma et al., 2007ab). Surprisingly, previously characterized organosulfates of the 2-methyltetrols and the 2-methyltetrol mono-nitrates detected at m/z 215 and m/z 260 (not listed in Table 2 Interactive Discussion tooxidation of isoprene in the presence of acidified sulfate seed aerosol (Surratt et al., 2007ab;Gómez-González et al., 2007), are also observed in the acid seed experiment shown in Fig. 10, suggesting that nighttime oxidation of isoprene in the presence of acidic seed may also be a viable pathway for these known ambient tracer compounds.

Offline chemical analysis
Owing to the implementation of reverse-phase chromatography, the SOA components that are more hydrophilic elute from the column the earliest, while the more hydrophobic components elute the latest. It is clear from Table 2 that compounds with the same carbon number and general functionality (i.e. carboxylic acid, alcohol, or organosulfate), but differing number of nitroxy groups, exhibit distinctly different chromatographic behaviors. The presence of more nitroxy groups appears to increase the 10 retention time of the SOA compound. For example, it is found that m/z 194 organic acid compound (C 5 H 8 NO − 7 ) containing one nitroxy group elutes earlier than that of the m/z 239 organic acid compounds (C 5 H 7 N 2 O − 9 ) containing two nitroxy groups. Similarly, the m/z 305 organosulfate (C 5 H 9 N 2 O 11 S − ) elutes earlier than that of the m/z 349 organosulfate (C 5

15
SOA components that are either nitroxy-organic acids or nitroxy-organosulfates are detected strongly as the [M-H] − ion, consistent with previous work (Surratt et al., 2006Gao et al., 2004ab, 2006, whereas the hydroxynitrates and nitroxyhydroxyperoxides are detected as both the [M-H] − and [M-H+C 2 H 4 O 2 ] − ions, with the latter acetic acid adduct ion, in most cases, being the base peak ion (i.e. dominant ion). 20 The acetic acid adduct ions for the hydroxynitrates and the nitroxy-hydroxyperoxides are formed owing to the presence of acetic acid in the UPLC mobile phase. Previous studies have shown that non-acidic hydroxylated species (such as the 2-methyltetrols) and organic peroxides formed from the photooxidation of isoprene (Claeys et al., 2004;Edney et al., 2005;Surratt et al., 2006) are either undetectable or yield weak nega-25 tive ions when using (-)ESI-MS techniques. However, it appears that the co-presence of nitroxy groups in the hydroxylated SOA components allow for these compounds to become acidic enough to be detected by the UPLC/(-)ESI-TOFMS technique, or allow for adduction with acetic acid. Further confirmation for the presence of organic perox- Interactive Discussion ides in the isoprene SOA produced from NO 3 oxidation is provided by the iodometricspectroscopic measurements shown in Table 3. Based upon the UPLC/(-)ESI-TOFMS measurements shown in Table 2, an average molecular weight of 433 for the organic peroxides is assumed for the calculations shown in Table 3. The contribution of organic peroxides to the SOA mass concentration is found to be fairly reproducible for dupli-5 cate typical experiments (i.e. 8/22/07 and 10/24/07). The amount of organic peroxides in the excess isoprene experiment is below detection limits.

Formation of various gas-phase products
As seen from Fig. 5 and Fig. 6, the three major first-generation products formed from 10 isoprene-NO 3 reaction are the m/z 230, 232, and 248 ions. Since the CIMS technique uses CF 3 O − (anionic mass 85 Da) as the reagent ion, compounds are detected at a m/z value of their molecular weight (MW) plus 85. The product ions at m/z 230, 232, and 248 likely correspond to C 5 -nitroxycarbonyl (MW 145), C 5 -hydroxynitrate (MW 147), and C 5 -nitroxyhydroperoxide (MW 163). These products have been observed 15 in previous studies (Jay and Stieglitz, 1989;Skov et al., 1992;Kwok et al., 1996;Berndt and Böge, 1997) and their formation from the isoprene-NO 3 reaction is relatively straightforward (Fig. 11). The reaction proceeds by NO 3 addition to the C=C double bond, forming four possible nitroxyalkyl radicals depending the position of the NO 3 attack. Previous studies suggest that NO 3 radicals predominantly attack isoprene in 20 the 1-position, with a branching ratio (C1-position/C4-position) varying between 3.5 and 7.4 (Skov et al., 1992;Berndt and Böge, 1997;Suh et al., 2001). In Fig. 11, only the nitroxyalkyl radical formed from the C1 attack is shown. The nitroxyalkyl radicals then react with O 2 to form RO 2 radicals, which react further with HO 2 , RO 2 , or NO 3 radicals under the experimental conditions in this study. The reaction of RO 2 radicals 25 and HO 2 radicals leads to the formation of C 5 -nitroxyhydroperoxide (m/z 248). The Interactive Discussion reaction of two RO 2 radicals (self reaction or cross reaction) has three different possible channels: The second channel results in the formation of C 5 -nitroxycarbonyl (m/z 230) and C 5hydroxynitrate (m/z 232). According to channel (5b), these two products should be formed with a 1:1 ratio; however, C 5 -nitroxycarbonyl can also be formed from alkoxy radicals (either from RO 2 +RO 2 reaction or RO 2 +NO 3 reaction). In Fig. 6, about 53 ppb of C 5 -nitroxycarbonyl and 25 ppb of C 5 -hydroxynitrate are formed after the addition of 10 the first pulse of N 2 O 5 , indicating ∼28 ppb of C 5 -nitroxycarbonyl is formed from the fragmentation of alkoxy radicals. The branching ratios for the reaction of small peroxy radicals have been investigated in previous studies. It is found that the branching ratio for channel (5a) for methylperoxy and ethylperoxy radicals is ∼0.3-0.4 and ∼0.6, respectively (Lightfoot et al., 1992;Wallington et al., 1992;Tyndall et al., 1998). It is likely 15 that the isoprene peroxy radicals react via this pathway to form alkoxy radicals and contribute to the "extra" 28 ppb of C 5 -nitroxycarbonyl. Although the concentrations of the product ions measured by CIMS are only rough estimates, the above observation is indicative that most RO 2 radicals react with other RO 2 radicals instead with NO 3 or HO 2 radicals. 20 Other than C 5 -nitroxycarbonyl, C 5 -hydroxynitrate, and C 5 -nitroxyhydroperoxide, three other minor products (m/z 185, 377 and 393 ions) are also observed as intermediate products. The proposed mechanisms for the formation of these gas-phase products are also shown in Fig. 11. Although channel (5c) in the RO 2 +RO 2 reaction is found to be minor for small peroxy radicals such as methylperoxy and ethylperoxy Introduction  , 1998, 2001), the product ion at m/z 377 could be the corresponding ROOR product formed from the self reaction of isoprene peroxy radicals. The product ion at m/z 185 likely corresponds to the C 5 -hydroxycarbonyl. It has been observed in previous studies and it likely arises from the isomerization of nitroxyalkoxy radicals through a 6member transition state to form a hydroxynitroxy alkyl radical, which then decomposes 5 to form NO 2 and C 5 -hydroxycarbonyl (Kwok et al., 1996). Such isomerization has also been proposed to occur in the photooxidation of isoprene (Paulson and Seinfeld, 1992;Carter and Atkinson, 1996;Dibble, 2002). It is possible that the hydroxynitroxy alkyl radical formed proceeds to react with O 2 to form a peroxy radical, which then reacts with the isoprene peroxy radical to form the product ion m/z at 393. The product ion at m/z 393 shows a slower rate of formation ( Fig. 6) compared to other product ions suggesting that it might also be formed from the further oxidation of a first-generation product. 2-methyl-2-vinyl-oxirane has been observed from isoprene-NO 3 reaction in previous studies at 20 mbar in helium (Berndt and Böge, 1997) and 20 Torr in argon (Skov et al., 1994), respectively. When operated in positive mode with H 2 O·H + as 15 the reagent ion (products are observed at m/z=MW+1), CIMS shows a protonated molecule at m/z 85. Although the epoxide yield is found to be <1% of the total reacted isoprene at atmospheric pressure (Skov et al., 1994), the signal at m/z 85 can arise in part from the epoxide. The further oxidation of the epoxide results in the formation of an epoxide peroxy radical, which can react with the isoprene peroxy radical to form 20 the peroxide at m/z 393. It is noted that a product ion at m/z 246 is detected in CIMS, which could arise from the corresponding carbonyl product formed from the reactions of two epoxide peroxy radicals, or from the fragmentation of the epoxide alkoxy radicals. Unlike m/z 393, which decays after the addition of the last pulse of N 2 O 5 , m/z 246 stays relatively constant suggesting that it is not being further oxidized by NO 3 radicals. 25 To examine further the possibility of peroxide formation (m/z 377 and 393) in the gas phase, an experiment is conducted using 1,3-butadiene as the parent hydrocarbon.
The analogous product ions for the 1,3-butadiene system, i.e. m/z 349 and 365, are observed in CIMS, providing further indication that the formation of ROOR products Interactive Discussion from two RO 2 radicals is occurring in the gas phase. Further details of the gas-phase chemistry of isoprene and 1,3-butadiene will be forthcoming in a future manuscript.

Effect of peroxy radical chemistry on SOA yield
The SOA yield ranges from 4.3% to 23.8% for an initial isoprene concentration of 18.4 to 101.6 ppb in the typical yield experiments. While the SOA yield from the slow N 2 O 5 5 injection experiment is roughly the same as that in the typical yield experiments, the SOA yield from the slow isoprene injection experiment is lower (Fig. 3). In both cases, ∼40 ppb of isoprene is consumed, the main difference being the relative importance of RO 2 +RO 2 reaction versus RO 2 +NO 3 reaction in each system. In the slow N 2 O 5 injection experiment, a relatively small amount of NO 3 is available in the chamber.

10
Once RO 2 radicals are formed, it is expected that they would react primarily with other RO 2 radicals instead of NO 3 radicals owing to the presence of a relatively higher isoprene concentration in the chamber. On the other hand, the slow isoprene injection experiment favors RO 2 +NO 3 reaction owing to the presence of excess N 2 O 5 in the chamber. Thus the higher SOA yield observed in the slow N 2 O 5 injection experiment 15 suggests the products formed via RO 2 +RO 2 reaction partition more readily into the aerosol phase, or the RO 2 +RO 2 reaction forms products that further react and contribute significantly to aerosol growth. The fact that the SOA yield from the slow N 2 O 5 injection experiment is roughly the same as in the typical yield experiments implies that RO 2 +RO 2 reaction dominates in typical yield experiments. 20 The time profile for the three major first-generation gas phase products and SOA growth from the slow N 2 O 5 injection experiment and slow isoprene injection experiment are shown in Fig. 12 and Fig. 13, respectively. In both cases, once the first-generation products are formed they can react further with NO 3 radicals, making it difficult to estimate the formation yields of these products based on the measured concentrations. 25 The extent to which these products react further is expected to be higher in the slow isoprene injection experiment owing to the presence of excess NO 3 in chamber; this is consistent with the relatively lower concentrations of first-generation products ob-3181 Introduction served. As mentioned before, it is possible that the CIMS signal at the observed m/z comprises isomers formed from the NO 3 attack at positions other than the C1 carbon. Such isomers have slightly different structures but they could exhibit a very different reaction rate towards NO 3 radicals. For instance, studies have shown that the reaction rates of NO 3 radicals with unsaturated alcohols and unsaturated carbonyl compounds 5 can vary by several orders of magnitude depending on the position of the substituted methyl group (Noda et al., 2002;Canosa-Mas et al., 2005). It is possible that the minor products formed from NO 3 attack at other positions react much slower with NO 3 radicals, hence the concentrations of the observed product ions do not decay to zero towards the end of the experiment. At the end of the experiment, about 8 ppb and 10 3 ppb of C 5 -hydroxynitrate is left in the slow N 2 O 5 injection experiment and slow isoprene injection experiment, respectively. Assuming the amount of reactive isomers and unreactive (or relatively slow reacting) isomers are formed in the same ratio in the slow N 2 O 5 injection experiment and the slow isoprene injection experiment, we can deduce that a relatively higher concentration of reactive C 5 -hydroxynitrate (as well as the two 15 other first-generation products) is formed in the slow N 2 O 5 injection experiment. This is consistent with the larger extent of RO 2 +RO 2 reaction (which forms C 5 -hydroxynitrate) and the higher SOA yield observed in the slow N 2 O 5 injection experiment, as it appears that C 5 -hydroxynitrate is an effective SOA precursor (Fig. 6).

20
By examining the time-dependent growth curves (organic aerosol, ∆M o as a function of hydrocarbon reacted, ∆HC) we can gain insights into the general mechanisms of SOA formation (Ng et al., 2006. Figure 4 shows the time-dependent growth curves for the slow N 2 O 5 injection experiment and the slow isoprene injection experiment, respectively. For the slow N 2 O 5 injection experiment, the initial aerosol growth 25 likely arises from the condensation of first-generation products as the presence of excess isoprene in the chamber suppresses their further oxidation. If higher generation products do contribute to SOA formation, they would have to be formed at very rapid 3182 Introduction rates. After isoprene is consumed, aerosol mass continues to increases and results in a "hook" in the growth curve. This indicates that secondary products (or higher generation products) also contribute significantly to SOA formation. The same observation can be made if we examine the reaction profile of a typical yield experiment (Fig. 2): there is further SOA growth after all isoprene is reacted away, indicating that the further 5 oxidation of first generation products are contributing to SOA formed. These observations are consistent with the fact that the decay of first-generation products observed in CIMS (especially the m/z 232 and m/z 377 ions) is strongly anticorrelated with further SOA growth (Fig. 6). On the other hand, the slow isoprene injection experiment does not allow us to differentiate the contribution of first-and second-generation products to 10 SOA formation. With the presence of excess NO 3 radicals in the chamber, the firstgeneration products formed in the slow isoprene injection experiment would be further oxidized once they are formed. The SOA growth observed throughout this experiment is from the partitioning of these highly oxidized and nonvolatile products. Hence, at the beginning of the experiment, for the same amount of ∆HC, the amount of SOA 15 formed in this experiment is higher than that in the slow N 2 O 5 injection experiment, in which the aerosol growth is probably from the condensation of relatively more volatile first-generation products. Both the AMS data and filter sample data (Figs. 8,9,and 10) show a very similar composition for the final SOA formed in slow N 2 O 5 injection experiment and the slow isoprene injection experiment, suggesting a common SOA forming 20 channel. Based on the previous discussion on the effect of peroxy radical chemistry on SOA yields, it is likely that the RO 2 +RO 2 reaction is the SOA-forming channel in both cases; such a reaction occurs to a large extent in the slow N 2 O 5 injection experiments and results in the formation of more SOA.

Proposed mechanisms of SOA formation 25
The combination of CIMS gas-phase data and elemental SOA composition data provides substantial insights into the mechanisms of SOA formation. Shown in Figs. 14-17 are the proposed SOA formation mechanisms from the further oxidation of the various 3183 Introduction Interactive Discussion gas-phase products measured by CIMS. The compounds in the boxes are the SOA products detected by UPLC/(-)ESI-TOFMS. Owing to multiple chromatographic peaks observed in the UPLC/(-)ESI-TOFMS extracted ion chromatograms (EICs) for the negative ions of the proposed SOA products, structural isomers are likely; however, for simplicity we show only one possible isomer for each product formed from a particular 5 reaction pathway. Many of the SOA products detected are formed from the further oxidation of first-or higher-generation products, which is consistent with the observation of continual SOA growth after the complete consumption of isoprene (hence a "hook" in the growth curve). With the large number of nitrate-substituted compounds detected by UPLC/(-)ESI-TOFMS technique, it is also not surprising that AMS shows strong 10 signals at m/z 30 (NO + ) and m/z 46 (NO + 2 ). Shown in Figs. 14 and 15 are the proposed SOA formation pathways from the further oxidation of the m/z 232 (i.e. C 5 -hydroxynitrate) and 377 gas-phase product ions (as detected by CIMS). The decay of these two products has been found to be strongly correlated with aerosol growth (Fig. 6), which is consistent with the large number of 15 SOA products formed from their further oxidation. The further oxidation of these two gas-phase products also yields SOA compounds of the same molecular weight (compounds of MW 371 and 450). Although m/z 393 is a minor gas-phase product, the further oxidation of this compound leads to formation of several SOA products (Fig. 16). As mentioned before, there are two possible formation routes for m/z 393, and the 20 further oxidation of both products is shown in Fig. 16. The further oxidation of the m/z 393 ion appears to yield SOA products that are specific only to this gas-phase product: these include the SOA products of MW 387 and 467. Figure 17 shows the proposed SOA formation mechanisms from three other gasphase products (m/z 185, m/z 230, and m/z 277); the further oxidation of these 25 product ions leads to relatively minor SOA products. Although C 5 -nitroxycarbonyl (m/z 230) is the most abundant gas-phase product detected by CIMS, its further oxidation is not well correlated with aerosol growth (Fig. 6). The further oxidation of m/z 230 yields an SOA product at MW 240. This organic acid product is found to be quite mi- Interactive Discussion nor when examining the peak area in its corresponding extracted ion chromatogram (EIC). It is noted that no SOA products are detected from the further oxidation of the C 5 -nitroxyhydroperoxide (m/z 248) (also a major gas-phase product); it is possible that these hydroperoxide products are not acidic enough to be detected by the UPLC/(-)ESI-TOFMS technique, or degrade during sample workup and/or analysis 5 procedures. It has been shown that hydroxycarbonyl plays a key role in SOA formation from the reaction of linear alkenes with NO 3 radicals (Gong et al., 2005), however, in the isoprene-NO 3 system, the further oxidation of the minor gas-phase product C 5hydroxycarbonyl (m/z 185) leads to the formation of only one minor aerosol product at MW 195. Some evidence for the formation of a C 5 -dinitrate first-generation gas-10 phase product is indicated from the CIMS and UPLC/(-)ESI-TOFMS data. This firstgeneration gas-phase product has been observed previously by Werner et al. (1997). The CIMS shows a weak signal at m/z 277, which could be associated to the dinitrate product; we do not know, however, whether the negative ion efficiently clusters with such compounds. Further evidence for the dinitrate gas-phase product is provided by 15 the UPLC/(-)ESI-TOFMS detection of an SOA product at MW 495, which could result from the further oxidation of a C 5 -dinitrate precursor. The precursor compound before the last oxidation step shown in this mechanism in Fig. 17 may exist in the particle phase; however, this compound is not likely to be detected by the UPLC/(-)ESI-TOFMS technique owing to the lack of acidic hydrogens from neighboring hydroxyl and/or car-20 boxyl groups. The SOA products highlighted in Figs. 14-17 are observed in all major experiments conducted; however, not all of these products are strongly detected in the excess isoprene experiment (Fig. 10c). With the presence of excess isoprene, further oxidations of first-generation products should be minimal and no significant SOA formation 25 is expected. However, SOA growth is observed and it appears from the UPLC/(-)ESI-TOFMS data that enough RO 2 +RO 2 chemistry is occurring to yield many of the products shown in Figs. 14-17. When comparing the UPLC/(-)ESI-TOFMS BPCs (Fig. 10 Interactive Discussion the excess isoprene experiment, while m/z 333 is the dominant chromatographic peak in other experiments. The chromatographic peak at m/z 430 corresponds to the acetic acid cluster ion for the compound at MW 371, which can be formed from the further oxidation of CIMS m/z 232 and 377 ions (Figs. 14 and 15). The chromatographic peak at m/z 446 corresponds to the acetic acid cluster ion for the compound at MW 387, which 5 is formed from the further oxidation of CIMS m/z 393 (Fig. 16). The detection of these two SOA products (MW 371 and MW 387) suggests that further oxidation of m/z 232, 377, and 393 is occurring in the excess isoprene experiment and contributing to SOA growth. It is also possible that CIMS m/z 393 (a first-generation product according to one of the formation routes) is nonvolatile enough that it partitions into the aerosol 10 phase and its further oxidation proceeds heterogeneously. Chromatographic peaks such as m/z 333 (associated with MW 271 compound), 449 (MW 450 compound) and 554 (MW 495 compound) are not as strong in the excess isoprene experiment owing to the fact there is not enough NO 3 in the system to allow for the formation of these highly oxidized compounds. 15 As discussed earlier, the formation yields of ROOR from the reaction of two peroxy radicals is very low for small peroxy radicals (Kan et al., 1980;Niki et al., 1981Niki et al., , 1982Wallington et al., 1989;Tyndall et al., 1998Tyndall et al., , 2001. However, according to both gasphase and aerosol-phase data in this study, it appears that RO 2 +RO 2 reaction (self reaction or cross-reaction) in the gas phase yielding ROOR products is a dominant SOA 20 formation pathway. Such reaction has been proposed to form low-volatility diacyl peroxides in the SOA formed from cyclohexene ozonolysis (Ziemann, 2002). In the case of self-reaction of peroxy radicals, the molecular weight of the product is essentially doubled, providing an efficient way to form products of low volatility. Owing to the lack of authentic standards, we cannot accurately quantify how much each of the peroxide 25 products contributes to the SOA mass. Nevertheless, the total peroxide measurement (Table 3) indicates that they contribute significantly to the total SOA formed.
From the UPLC/(-)ESI-TOFMS (Table 2) and PILS/IC measurements, it appears that organic acids are not a major contributor to SOA formation from the oxidation of iso- high-and low-NO x conditions), in which a large number of organic acids, such as 2methylglyceric, formic, and acetic acid, are observed (Surratt et al., 2006;Szmigielski et al., 2007). In the photooxidation experiments, the level of organic acids detected under low-NO x conditions is lower than under high-NO x conditions. The low-NO x isoprene SOA was previously found to also have a significant amount of organic peroxides, as detected in the current study (Table 3); however, organic peroxides detected previously in low-NO x isoprene SOA were not structurally elucidated through MS techniques performed in the present study (Table 2,, possibly owing to the lack of nitroxy groups which seem to induce acidity and/or increase the adductive abilities of organic peroxides with acetic acid during the ESI-MS analysis. Overall, it appears that the 15 isoprene-NO 3 SOA is much more similar to the previously studied low-NO x isoprene SOA. More specifically, it appears that both contain a large amount of organic peroxides, organosulfates (if conducted in the presence of sulfate seed aerosol), and neutral hydroxylated compounds, such as the hydroxynitrates observed in Fig. 14   The global chemical transport model GEOS-Chem (v. 7-04-11) (http://www.as.harvard. edu:16080/chemistry/trop/geos/) is used to estimate, roughly, global SOA formation from the isoprene + NO 3 reaction. The current version of GEOS-Chem treats mechanistically SOA formation from isoprene +OH, monoterpenes and sesquiterpenes, and 25 aromatics; here we will estimate SOA formation from isoprene +NO 3 by using an approximate, uniform SOA yield of 10% (corresponding to M o ∼ =10 µg m −3 in Fig. 3 Interactive Discussion noted that this yield is quite uncertain and the importance of peroxy radical self reactions in this study suggest that the SOA yield in the atmosphere will be highly sensitive to the nature of the nighttime peroxy radical chemistry. Here, we seek to obtain only a "back-of-the-envelope" estimate. Two global isoprene emissions are available in GEOS-Chem: GEIA (Global Emission Inventory Activity) (Guenther et al., 1995) and MEGAN (Model of Emissions and Gases from Nature) (Guenther et al., 2006). Both models require, as input, meteorological data such as temperature to calculate the amount isoprene emitted. For the present estimate, the meteorological fields employed by Wu et al. (2007), generated by the Goddard Institute for Space Studies (GISS) General Circulation Model III, are used.

10
Meteorological conditions correspond approximately to those of year 2000. Table 4 presents the annual emissions of isoprene as predicted by each of the emission models, together with the amount of isoprene predicted to react via OH, O 3 , and NO 3 , the global burden, and lifetime. We note that there is a significant difference between the annual isoprene emissions predicted by the earlier and newer emission 15 models. Isoprene + OH accounts for 300 to 400 Tg yr −1 of isoprene consumption. Henze et al. (2007) predict that annual SOA production from isoprene + OH is about 13 Tg yr −1 (based on the MEGAN inventory and GEOS-4 meteorological fields, which are assimilated fields from actual year 2004). Note that SOA production from isoprene + OH, or any other pathway for that matter, is sensitive to the production of SOA from 20 other hydrocarbon precursors since gas-aerosol partitioning depends on the total organic aerosol mass.
If we take as a rough estimate a 10% SOA yield from the isoprene + NO 3 pathway from the results in Table 4, 2 to 3 Tg yr −1 of SOA results from isoprene + NO 3 . This rate of production would make SOA from isoprene + NO 3 as significant as that from 25 sesquiterpenes, biogenic alcohols, and aromatics, each of which produces about 2 to 4 Tg yr −1 of SOA (Henze et al., 2007). Owing to efficient photodissociation, NO 3 achieves its highest concentrations at night. By contrast, isoprene emissions are assumed to be zero at night in both emission models. Consequently, the isoprene + NO 3 Introduction Interactive Discussion reaction occurs only at night, involving isoprene that remains unreacted after each daytime period. We caution that the estimates above are obtained at the crudest level of approximation, in which a globally uniform SOA yield of 10% from isoprene + NO 3 is applied. As we note from Table 4, there is also as substantial difference between predictions of the 5 two available isoprene emission models; the more recent MEGAN model represents an improved level of understanding over the earlier GEIA model. Predictions of SOA formation from the isoprene + NO 3 pathway are, of course, highly dependent on ambient NO 3 radical concentrations. Nitrate radical concentrations predicted in the current simulations vary from about 0.1 ppt in remote regions of South America to 20 ppt or more 10 in the southeastern USA (in August). Future work will address the simulation of SOA formation from isoprene + NO 3 following the microphysical treatment in GEOS-Chem.

Implications
We report a series of chamber experiments investigating the formation of secondary organic aerosols from the reaction of isoprene with nitrate radicals. For an initial isoprene 15 concentration of 18.4 to 101.6 ppb, the SOA yield ranges from 4.3% to 23.8% (typical yield experiments). The SOA yield from the slow N 2 O 5 injection experiment (RO 2 +RO 2 reaction dominates) is much higher than that from the slow isoprene injection experiment (RO 2 +NO 3 dominates), implying that RO 2 +RO 2 is a more effective channel of forming SOA. The SOA yield from the slow N 2 O 5 experiment is roughly the same as 20 that in the typical yield experiments, suggesting that SOA yields obtained in this study likely represent conditions in which peroxy-peroxy radical reactions are favored. Using a uniform SOA yield of 10% (corresponding to M o ∼ =10 µg m −3 ), ∼2 to 3 Tg yr −1 of SOA results from isoprene + NO 3 , which is about 1 / 4 of the amount of SOA estimated to be formed from isoprene + OH (∼13 Tg yr −1 ) (Henze et al., 2007). 25 The extent to which the results from this study can be applied to conditions in the atmosphere depends on the relative importance of RO 2 +RO 2 versus RO 2 +NO 3 re- Interactive Discussion actions in the nighttime troposphere. However, the fate of peroxy radicals in the atmosphere is uncertain owing to the large uncertainties in the reaction rate constants and ambient concentration of both RO 2 and NO 3 radicals (Skov et al., 1992;Kirchner and Stockwell, 1996;Bey et al., 2001ab;Vaughan et al., 2006). A modeling study by Kirchner and Stockwell et al. (1996) suggests that RO 2 +NO 3 reaction is important at 5 night; for a moderately polluted site, ∼77-90% of the total RO 2 at night is predicted to react with NO 3 . These results are at odds with the study by Bey et al. (2001ab), which suggests that NO 3 radicals are not involved significantly in the propagation of RO 2 radicals. Currently, only the reaction rate constants for small, relatively simple RO 2 radicals with NO 3 radicals have been reported (e.g. Biggs et al., 1994;Daele et al., 1995;10 Canosa -Mas et al., 1996;Vaughan et al., 2006) and they are roughly in the range of (1-3)×10 −12 cm 3 molecule −1 s −1 . With the oxidation of various volatile organic compounds by O 3 and NO 3 under nighttime conditions, it is expected that multi-functional peroxy radicals would be prevalent; the reaction rates of these complex peroxy radicals warrants future study. Furthermore, more field measurements on the concentrations of 15 peroxy radicals and nitrate radicals would also help to constrain the relative importance of RO 2 +RO 2 versus RO 2 +NO 3 reaction.
In this study, we have shown that the formation of ROOR from the reaction of two peroxy radicals is an effective SOA-forming channel based on gas-phase data and elemental SOA composition data. This reaction has generally been considered as a 20 minor channel and has not been widely studied. If the results from this study can be applied to other systems (i.e. the reaction of NO 3 radicals with other volatile organic compounds), the organic peroxides could possibly be very important SOA components in all systems; they may not have been identified previously owing to the lack of suitable analytical techniques (such as accurate mass measurements from high resolution MS) 25 and clearly more study is needed. Introduction   , Atmos. Environ., 28, 1583-1592, 1994 Products and mechanisms of the reactions of the nitrate radical (NO 3 ) with isoprene, 1,3-butadiene and 2,3-dimethyl-1,3butadiene in air, Atmos. Environ., 26A, 15, 2771-2783  M., Sorooshian, A., Seinfeld, J. H., and Claeys, M.: Characterization of 2-methylglyceric acid oligomers in secondary organic aerosol formed from the photooxidation of isoprene using trimethylsilylation and gas chromatography/ion trap mass spectrometry, J. Mass Spectrom., 42, 101-116, 2007. Tyndall, G. S., Cox, R. A., Granier, C., Lesclaux, R., Moortgat, G. K., Pilling, M. J., Ravis-5 hankara, A. R., and Wallington, T. J.: Atmospheric chemistry of small peroxy radicals, J. Geophys. Res., 106, D11, 12 157-12 182, 2001. Tyndall, G. S., Wallington, T. J., and Ball, J. C.: FTIR product study of the reactions of CH 3 O 2 +CH 3 O 2 and CH 3 O 2 +O 3 , J. Phys. Chem., 102, 2547-2554, 1998 Kinetic studies of reactions of the nitrate radical (NO 3 ) with peroxy radicals (RO 2 ): an indirect source of OH at night?, Phys. Chem. Chem. Phys., 8, 3749-3760, 2006. von Friedeburg, C., Wagner, T., Geyer, A., Kaiser, N., Platt, U., Vogel, B., and Vogel, H.: Deriva-tion of tropospheric NO 3 profiles using off-axis differential optical absorption spectroscopy measurements during sunrise and comparison with simulations, J. Geophys. Res., 107, D13, Introduction  Table 2. SOA products identified using UPLC/(-)ESI-TOFMS.     These compounds appear to be very minor SOA products due to very small chromatographic peak areas, confirming that the further oxidation of the nitroxycarbonyl and hydroxycarbonyl first-generation gas-phase products do not yield significant quantities of SOA. c A blank cell indicates that the detected SOA product had no obervable acetic acid adduct ion (i.e. [M H + C 2 H 4 O 2 ] ). d These organosulfate SOA products were observed only in experiments employing either (NH 4 ) 2 SO 4 (i.e. neutral) or MgSO 4 + H 2 SO 4 (i.e. acidic) seed aerosol. These organosulfate SOA products were also observed in the excess isoprene experiments. e In addition to the acetic acid adduct ion, these compounds also had a significant adduct ion at [M H + HNO 3 ] (m/z 333), indicating that these compounds are likely not very stable due to the fragmentation of one of the NO 3 groups during the MS analysis. f These compounds were only weakly detected in the excess isoprene experiments.           Fig. 15. Proposed mechanism for SOA formation from the formation and decay of the CIMS m/z 377 gas-phase product formed from the isoprene + NO 3 reaction. Boxes indicate UPLC/(-)ESI-TOFMS detected SOA products; molecular formulas were confirmed by the accurate mass data provided by the UPLC/(-)ESI-TOFMS. Multiple structural isomers are possible, consistent with the multiple chromatographic peaks observed in the extracted ion chromatograms; however, only one structural isomer is shown for simplicity. a This first-generation gas-phase product was detected as the [M+CF 3 O] − ion by the CIMS instrument. b These particle-phase compounds were detected as both their [M-H] Fig. 17. Proposed mechanism for SOA formation from the formation and decay of the C 5nitrooxycarbonyl, C 5 -hydroxycarbonyl, and C 5 -dinitrate first-generation products formed from the isoprene + NO 3 reaction. Boxes indicate UPLC/(-)ESI-TOFMS detected SOA products; molecular formulas were confirmed by the accurate mass data provided by the UPLC/(-)ESI-TOFMS. Multiple structural isomers are possible, consistent with the multiple chromatographic peaks observed in the extracted ion chromatograms; however, only one structural isomer is shown for simplicity. a These first-generation gas-phase products were previously observed by Skov et al. (1994) and Kwok et al. (1996); these gas-phase products were detected as the [M+CF 3 O] − ion by the CIMS instrument. b These are minor SOA products, confirming that the further oxidation of the C 5 -nitrooxycarbonyl and C 5 -hydroxycarbonyl first-generation products do not yield significant amounts of SOA. c This first-generation gas-phase product was previously observed by Werner et al. (1999); this gas-phase product was also detected as the [M + CF 3 O − ] − ion by the CIMS instrument.