The flow reactor (4.6 cm i.d. and 85 cm long) used in this study has been described in detail previously (Zhao et al., 2015). α-Cedrene was added to the flow reactor by injecting the pure liquid (Sigma-Aldrich and Extrasynthese, > 98 %) into a flow of 1.76–2.96 L min-1 of clean, dry air (Praxair, ultra zero air) using an automated syringe pump (Pump Systems Inc., model NE-1000). Ozone was generated by passing a flow of O2 (Praxair, Ultra High Purity, 99.993 %) at 0.24 L min-1 through a pen-ray mercury lamp (model 11SC-2) and subsequently added to the flow reactor downstream of the α-cedrene inlet. The O3 concentration, determined using a UV-VIS spectrometer (Ocean Optics, HR4000), was adjusted by changing the UV exposure of the O2 via a movable metal cover surrounding the lamp. The total flow rate in the reactor was 2.0, 2.9, or 3.2 L min-1, corresponding to a residence time of 44, 30, or 27 s, respectively. Some of the experiments were carried out in the presence of water vapor or formic acid, both of which can act as SCI scavengers. Water vapor was added by bubbling a flow of clean air through nanopure water (18.2 MΩ cm) into the flow reactor. The relative humidity of the airflow (∼ 75 % relative humidity (RH)) in the reactor was measured using a humidity probe (Vaisala, HMT234). Formic acid (Sigma-Aldrich, ≥ 95 %) was added to the reactor using the same method as for α-cedrene.

The particle size distributions were measured at the outlet of the reactor using a scanning mobility particle sizer (SMPS, TSI), which is equipped with an electrostatic classifier (model 3080), a long differential mobility analyzer (model 3081), and a condensation particle counter (model 3776). When the size distribution of SOA was stable, the polydisperse particles were collected onto a 47 mm PTFE filter (Millipore Fluoropore, 0.45 µm pore size) at a flow rate of 1.9, 2.8, or 3.1 L min-1, with venting of the remaining 0.1 L min-1 aerosol flow to the hood. In order to obtain sufficient particle mass for the analysis, the collection lasted 3–15 h, depending on the particle mass concentrations. Excellent collection efficiency (> 99 %) was obtained for particles of all diameters as established by SMPS measurements downstream of the filter. The filter samples were extracted immediately with a 2 mL mixture (5 : 3 in v / v) of methanol (OmniSolv; liquid chromatography–mass spectrometry (LC-MS) grade) and hexanes (a mixture of hexane isomers; Fisher Scientific; 99.9 %) under ultrasonication in an ice bath for 30 min. The extracts were then purged gently using a flow of dry N2 at room temperature to evaporate down to 1 mL. As hexanes are more volatile than methanol, the remaining solvent in the extracts would be primarily methanol. The resulting extracts were either analyzed directly using an LCT Premier electrospray ionization time-of-flight mass spectrometer (ESI-ToF-MS, Waters) or diluted with water (OmniSolv; LC-MS grade) to yield a final 50 : 50 water : methanol mixed solution, followed by analysis with a Xevo TQS electrospray ionization triple quadrupole mass spectrometer (ESI-TQ-MS, Waters). In order to verify that the sonication has no significant impact on SOA measurements, filter extraction was performed instead, in some experiments, in the mixture of methanol and hexanes by shaking for 30 min. No significant difference in product distribution was observed in ESI mass spectra of the SOA extracted with and without sonication. In separate experiments, the chemical composition of α-cedrene SOA was also measured in real time using direct analysis in real-time mass spectrometry (DART-MS).

Chamber experiments were carried out in 450 L Teflon chambers. Two types of α-cedrene ozonolysis experiments were conducted: (i) under dry or humid (72 % RH) conditions and (ii) in the absence or presence of gas phase 2-EHN, which was incorporated into SOA during ozonolysis and used as a tracer molecule to probe the viscosity of SOA. The α-cedrene was added using an automated syringe pump to inject a defined volume of pure liquid into a flow of clean, dry air, which was directed into the chamber. Water vapor was added by bubbling air through nanopure water into the chamber. 2-EHN (Sigma-Aldrich, 97 %) was added by injecting a known volume of liquid into the chamber. After 10 min to allow for mixing of the gases (or 2 h for the evaporation of the 2-EHN), ozonolysis was initiated by adding O3 generated by a commercial ozone generator (Polymetrics, Model T-816) to the chamber.

For experiments CH1–CH3 (Table 1), SOA composition was examined online by a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS; Aerodyne Research Inc.) or by collection onto a PTFE filter at a flow rate of 12 L min-1. Filter sampling started after 30 min reaction time and lasted about 30 min. Because of the small volume of the chamber (450 L) and relatively low SOA mass loading in most experiments (see Table 1), in each experiment the SOA was sampled onto one filter to ensure enough mass for ESI-MS analysis. The filter was extracted using the same method as for SOA samples obtained in flow reactor experiments, followed by ESI-MS analysis. For experiments CH4 and CH5, SOA formed without or with added 2-EHN at a 60 min reaction time was collected onto a germanium (Ge) attenuated total reflectance (ATR) crystal using a custom-designed impactor (Kidd et al., 2014b) at a flow rate of 30 L min-1. The SOA impacted on the ATR crystal was subsequently probed using attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy.

Most experiments were performed with excess O3. In order to evaluate the influence of excess O3, which could lead to secondary oxidation of first-generation products, some experiments were performed with excess α-cedrene (Table 1, experiment CH3). The ESI mass spectra of SOA formed in the presence of excess O3 or excess α-cedrene are very similar, suggesting that the excess O3 has no observable influence on SOA formation. In addition, no significant difference was observed in ESI mass spectra of SOA formed with or without added 2-EHN, suggesting that the incorporation of 2-EHN into the SOA does not change its overall composition.

In both flow reactor and chamber experiments, aerosol samples were directed through a 10 cm monolith extruded carbon denuder (Novacarb^™; Mast Carbon, Ltd.) to remove the gas phase species prior to particle collection on the filter. No significant difference in particle size distribution was observed with or without the denuder. Blank experiments were also carried out under the same experimental conditions as the ozonolysis experiments but without adding O3 to the flow reactor or the chamber.

Figure 2a is a typical ATR-FTIR spectrum recorded immediately following impaction of α-cedrene SOA formed in the chamber in the absence of 2-EHN. The strong C=O band at 1706 cm-1 suggests that aldehydes and ketones are important SOA components. A shoulder at 1762 cm-1 may indicate the presence of carboxylic acids, esters, or other species containing C=O groups, with a more electronegative atom such as oxygen being attached to the carbonyl carbon (Socrates, 2001; Kidd et al., 2014a).

Figure 2b is a difference spectrum showing the changes in SOA after exposure to a flow of clean dry air for 20 h. The positive and negative peaks in the spectrum represent an increase or decrease, respectively, in the functional groups in SOA due to air exposure. There is a decrease in peaks at 3416, 1762, 1371, and 1076 cm-1 and an increase in peaks at 1735, 1706, 1410, and in the 1100–1350 cm-1 region. Note that there is no obvious change in the C-H peaks around 2957 cm-1, suggesting that evaporation of organic species from SOA during air exposure is not important. Therefore, the changes in SOA as indicated by the difference spectrum are due to chemistry occurring in SOA during air exposure. One of the possible processes is the decomposition of oligomers that comprise a significant fraction of α-cedrene SOA as discussed later. For example, decomposition of peroxyhemiacetals and aldol condensation products to their precursors can lead to the loss of O-H groups and concomitant formation of ketone and aldehyde C=O groups. The decomposition of oligomers in SOA was supported by ESI-MS measurements of SOA collected on Teflon filters and then exposed to a flow of clean dry air; the relative ion intensity of oligomers to LMW products in the mass spectrum of SOA extracted after 20 h of air exposure is ∼ 15 % lower compared to that of SOA extracted immediately following collection. The difference spectrum of the SOA film exposed to clean humid air at 89 % RH for 40 min (Fig. S3) shows a broad band centered at 3426 cm-1 and a narrow band at 1640 cm-1 resulting from the stretching and bending vibration of adsorbed water, respectively, with relative strength of stretching vs. bending of ∼ 2. The absence of a negative peak at 1640 cm-1 in the difference spectrum after dry air exposure as shown in Fig. 2b, therefore, suggests that the contribution of water to the loss in the O-H band at 3416 cm-1 is minor, and the decrease in this region must be due to a change in SOA components.

Figure 2c shows the difference spectrum of SOA upon exposure to 8 ppm O3 for 20 h, which is very similar to the difference spectrum of SOA after 20 h of exposure to clean air (Fig. 2b). This shows that the α-cedrene SOA oxidation products are not reactive toward O3, at least as detectable by changes in the infrared spectrum.

In order to probe the phase state of α-cedrene SOA, a relatively volatile organic species, 2-EHN (Psat=1.4 × 10-4 atm at 295 K; Pankow and Asher, 2008), was included in one type of experiment (CH5, Table 1) so that it could be incorporated into the SOA as it is formed in the chamber. The evaporation of 2-EHN from the SOA impacted onto the ATR crystal upon exposure to a flow of clean dry air was followed with time to probe the phase state of SOA.

Figure 3a shows a digital photograph of a typical impaction pattern on the ATR crystal of α-cedrene SOA formed in the chamber in the presence of gas phase 2-EHN. The SOA impacts and adheres to the crystal, forming a narrow film ∼ 1.0 mm in width at the centerline. Similar impaction patterns were also observed for the SOA formed without added 2-EHN. Figure 3b shows the SOA number and mass size distributions after 60 min reaction time in the chamber, as well as the collection efficiency of the impactor as a function of particle diameter (Kidd et al., 2014b). By dividing the particles into a number of 10 nm size bins and applying the average particle collection efficiency at each bin, the SOA mass collected on the crystal is estimated to be 42 µg using ∑mifi, where mi and fi are the particle mass and average particle collection efficiency at each size bin, respectively. Since the particle wall loss in the chamber during impaction was not considered, this mass should be an upper limit. Assuming a particle density of 1.1 g cm-3 (Yao et al., 2014), the maximum average thickness of the SOA film on the ATR crystal is estimated to be L=0.54 µm. This estimate of L relies on the assumption that SOA impacted on the crystal forms a uniform narrow film. This may underestimate the film thickness by as much as a factor of 2 if, rather than forming a thin film, the SOA is collected in separate columns immediately below the impactor holes. An additional, smaller uncertainty in L results from variation in the collection efficiency of the impactor over the particle size range of interest (see Fig. 3b). We estimate the combination of these two to give an uncertainty in L of about a factor of 2.

The refractive index of α-cedrene SOA is not known, but a value of ∼ 1.5 is reasonable based on literature values for SOA from other biogenic organic oxidations (Lambe et al., 2013; Kim et al., 2014). The depth of penetration of the IR beam at 1280 and 1635 cm-1, peaks which correspond to the absorption bands of organic nitrate as discussed below, is then calculated to be 0.52 and 0.41 µm (Harrick, 1967), respectively, for SOA on a Ge crystal. This suggests that the entire depth of the SOA film on the ATR crystal can be probed reasonably well by the IR beam in the region of interest.

Figure 4a is a typical ATR-FTIR spectrum of α-cedrene SOA formed in the presence of 2-EHN in the chamber under dry conditions. The spectrum is essentially the same as that of SOA formed without 2-EHN (Fig. 2a) except that there are two new bands at 1635 and 1280 cm-1 resulting from the vibrations of the nitrate functional group (Socrates, 2001; Bruns et al., 2010; Perraud et al., 2012) when 2-EHN is present. Figure 4b shows the difference spectrum of SOA after exposure to a flow of clean dry air for 20 h. In addition to the spectral features similar to the difference spectrum of 2-EHN-free SOA (Fig. 2b), there is a small loss of nitrate peaks at 1635 and 1280 cm-1, indicating some evaporation of 2-EHN from SOA during air exposure. Figure 5 shows the temporal evolution of the integrated area of the nitrate peak at 1280 cm-1 over 20 h of air exposure. It can be seen that after 20 h of air exposure, only ∼ 27 % of 2-EHN evaporated from the SOA. This slow evaporation indicates that 2-EHN is incorporated in the bulk of the SOA and that the SOA must be a high-viscosity semisolid rather than a liquid where diffusion would be much more rapid (Shiraiwa et al., 2011; Koop et al., 2011).

The incorporation of high-volatility 2-EHN into the bulk of high-viscosity SOA is consistent with a condensation particle growth mechanism (Finlayson-Pitts and Pitts, 2000; Seinfeld and Pandis, 2006; Riipinen et al., 2011; Perraud et al., 2012), for which organic species condense kinetically onto the surface of pre-existing particles and become buried and incorporated into the bulk by incoming low-volatility organics. It is important to note that in the ATR-FTIR spectrum (Fig. 4c) of α-cedrene SOA formed without added 2-EHN in the chamber and then exposed to 20 ppm 2-EHN in the ATR cell for 1 h, the absorption bands associated with organic nitrate at 1635 and 1280 cm-1 are negligible. This observation suggests that the uptake of 2-EHN into or onto previously formed SOA is not important.

Assuming that the evaporation of 2-EHN is determined only by its diffusion in the film of SOA impacted on the ATR crystal and that the diffusion follows Fick's Law, for a film with uniform 2-EHN concentration (C0) at t=0 and zero concentration at the surface for t>0, the total fraction (F) of 2-EHN remaining in the film at time t can be expressed as (Crank, 1975; Mehrer, 2007) F=8π2∑j=0∞1(2j+1)2exp⁡-(2j+1)2π2Dt4L2j=0,1,2,…  . In Eq. (1), D is the diffusion coefficient of 2-EHN in the SOA and L is the thickness of the SOA film on the ATR crystal. A best fit of the evaporation data of 2-EHN in Fig. 5 to Eq. (1) gives D/L2=5.6×10-5 min-1. With the SOA film thickness L of 0.54 µm estimated above, the diffusion coefficient (D) of 2-EHN in SOA is calculated to be 3 × 10-15 cm2 s-1, consistent with the D values of the order of ∼ 10-15 cm2 s-1 predicted from the timescale for evaporation combined with the estimated thickness of the SOA layer (Shiraiwa et al., 2011; Koop et al., 2011). The uncertainty in L of about a factor of 2 described above translates into an uncertainty in D of about a factor of 4. However, this value for D is within the range of diffusion coefficients that would be expected in a highly viscous semisolid matrix, i.e., D=10-10–10-20 cm2 s-1 (Shiraiwa et al., 2011; Koop et al., 2011), confirming that α-cedrene SOA is a high-viscosity semisolid. However, because of their small sizes, these particles do not have enough momentum to bounce as they impact on the ATR crystal, as seen for larger SOA from α-pinene oxidation in earlier studies (Kidd et al., 2014a, b). Recently, Saukko et al. (2012) and Pajunoja et al. (2015) reported the formation of semisolid SOA from the oxidation of the sesquiterpene longifolene based on particle bounce measurements. However, to our knowledge, the present work shows the first measurement of the diffusion coefficient of an organic species in sesquiterpene SOA.

Figure 6 shows typical ESI+ and ESI- mass spectra of SOA formed from α-cedrene ozonolysis in the flow reactor at 30 s reaction time. Ions in the mass range m/z 220–350 correspond to LMW products derived directly from oxidation of α-cedrene and retaining much of the structure of the parent compound (hereafter termed P1 products). Those with m/z > 420 correspond to HMW products formed using two, three, or four P1 products (hereafter termed P2, P3, and P4 products, respectively) as building blocks. We try to avoid using the terms “dimer”, “trimer”, etc. which may imply simple combinations of smaller species whereas P2, P3, etc. are clearly complex combinations of different LMW products. As discussed above, there is no evidence that any of these peaks arise from multiple charging of higher molecular weight species. The data presented in Fig. 6 suggest that P1 and P2 products comprise a dominant fraction of α-cedrene SOA. Compared to the ESI+ spectrum, the corresponding ions in the ESI- spectrum are generally 24 Da lower in mass. This is consistent with the fact that the ions observed in the ESI+ mode are primarily sodium adducts, while those in the ESI- mode are deprotonated ions with a predominant contribution from carboxylic acids.

Figure 7 shows the ESI mass spectra of polydisperse α-cedrene SOA (size distributions are given in Fig. S1a) with geometric mean diameters of 15 and 23 nm formed in the flow reactor at the same concentrations of α-cedrene and O3 but different reaction times (27 and 44 s) under dry conditions (Table 1, experiments FR1–FR2). Among P2 products, those at m/z 481–543 are the most abundant ones in the flow reactor SOA and they have a greater contribution to the smaller particles formed at shorter reaction times (Fig. 7a). Conversely, P1 products contribute more to larger particles formed at longer reaction times (Fig. 7b). Figure S4 shows additional ESI mass spectra of polydisperse α-cedrene SOA particles of different geometric mean diameters (13, 18, and 26 nm) formed in the flow reactor at the same reaction time (30 s) but different concentrations of α-cedrene under dry conditions (Table 1, experiments FR3–FR5). A similar size-dependent distribution of P1 and P2 products is observed. These results suggest that P2 products at m/z 481–543 may play an important role in initial particle formation, while P1 products contribute significantly to particle growth.

Figure 8 shows typical ESI+ and ESI- mass spectra of α-cedrene SOA formed in the chamber. The chamber SOA has a geometric mean diameter of 66 nm, larger than those formed in the flow reactor (13–26 nm, depending on the experimental conditions) due to the longer reaction times and hence greater extents of reaction. Figure 8 shows that the chamber particles also have relatively more P1 products, in agreement with the important role of P1 products in particle growth. In addition, the subset of P2 products with m/z > 543 in the positive ion mode accounts for a greater fraction of total P2 products in the chamber SOA than in the flow reactor SOA (Figs. 6a and 7). This suggests that these larger P2 products (i.e., m/z > 543) are mainly formed at longer reaction times in the chamber and contribute mainly to particle growth.

It is known that P1 products may undergo ion–molecule reactions to form noncovalently bound clusters in the ESI source (Müller et al., 2009; Gao et al., 2010), and the formation of such clusters, if it occurs, is expected to be positively correlated with the abundance of P1 products. However, as shown in Figs. 7 and S4, the signal intensity ratio of P2 to P1 products is inversely dependent on the size of SOA. This suggests that the P2 products observed in ESI mass spectra are unlikely to be artifacts of in-source clustering. In addition, HMW products in α-cedrene SOA are also detected in real time by DART-MS, which employs a different ionization method. Figure S5 shows typical DART+ (positive ion mode) and DART- (negative ion mode) mass spectra of flow reactor α-cedrene SOA, which was heated to 160 ∘C before introduction into the ionization region. The masses of P1 and P2 products in the DART+ mass spectrum are generally 22–24 Da lower than those in the ESI+ mass spectrum (Fig. 6), consistent with the ions detected by DART-MS in the positive mode being primarily [M + H]+, M+, and [M - H]+ (Cody et al., 2005; Nah et al., 2013). In contrast, essentially the same masses for P1 and P2 products were observed in DART- and ESI- mass spectra, in both of which the ions are predominantly [M - H]-. There are some differences between ESI and DART mass spectra that are likely due to the heating of the incoming aerosol stream for DART-MS and the higher DART source temperature. Overall, DART-MS measurements further confirm that the P2 products observed by ESI-MS are not artifacts from the ion source.

Previous studies have reported an OH yield of 62–67 % from ozonolysis of α-cedrene (Shu and Atkinson, 1994; Yao et al., 2014). Therefore, reaction with OH could play a role in α-cedrene oxidation in the absence of an OH scavenger. We reported in a previous study (Zhao et al., 2015) ESI-MS spectra in the absence and presence of cyclohexane, which showed that the relative intensity of P2–P4 peaks in the presence of cyclohexane is smaller. This suggests that OH oxidation may contribute to the formation of HMW species.

Possible structures and formation mechanisms of typical P1 and P2 products (as labeled in Fig. 6) were explored based on their fragmentation mass spectra (MS2), accurate mass data, previously identified P1 products (Jaoui et al., 2004, 2013; Reinnig et al., 2009; Yao et al., 2014), and likely reaction mechanisms of terpene ozonolysis. To the best of our knowledge, this is the first detailed examination of HMW products initially reported (Zhao et al., 2015) in this system.

The reaction of α-cedrene with ozone has a very high SCI yield (> 88 %) (Yao et al., 2014). If SCI reactions are key to SOA formation in this system, the addition of SCI scavengers such as water vapor and formic acid should have significant influence on the formation of SOA.

Figure 11 shows the size distributions of SOA formed from ozonolysis of α-cedrene in the flow reactor under dry conditions, at 75 % RH, or in the presence of 15 ppm formic acid. SOA with very similar number concentrations is formed with or without added water vapor, while significantly fewer, but larger, particles are formed in the presence of formic acid. Recent kinetics measurements of the CH2OO CI reaction with water vapor (monomer and dimer) give rate constants of kdimer = (4.0–6.5) × 10-12 cm3 molecule-1 s-1 (Chao et al., 2015; Lewis et al., 2015) and kmonomer = 3.2 × 10-16 cm3 molecule-1 s-1 (Berndt et al., 2015). The rate constant for the reaction of CH2OO with formic acid is much larger, kHCOOH = 1 × 10-10 cm3 molecule-1 s-1 (Welz et al., 2014). The reactivity of Criegee intermediates has been shown to depend on structure for a number of reasons, including different extents of pressure stabilization of excited CI, different reactivities for the syn- and anti-forms, and different natures of the groups attached to the central CI carbon (Ryzhkov and Ariya, 2004; Anglada et al., 2011; Donahue et al., 2011; Vereecken et al., 2012; Taatjes et al., 2013; Welz et al., 2014). While the α-cedrene CI is clearly very different than CH2OO, if the rate constants for the latter are representative in a relative sense, the ratio of the rates for reaction of the α-cedrene CI with formic acid and water dimer relative to water monomer are ∼ 240 : 18 : 1 under our experimental conditions (3.7 × 1014 HCOOH cm-3, 4.8 × 1017 H2O cm-3 and 5.1 × 1014 (H2O)2 cm-3, based on a water equilibrium constant of 0.0536 atm-1; Ruscic, 2013). Thus, it is not surprising that formic acid has a much greater impact than water. Furthermore, the large effects of formic acid further support the importance of the α-cedrene CI in SOA formation. These impacts of water and formic acid are in agreement with those reported by Yao et al. (2014), who reported a similar quenching effect of added acetic acid on particle formation in the α-cedrene ozonolysis but relatively little impact of water vapor.

Figure 12a and b show the ESI mass spectra of α-cedrene SOA formed in the flow reactor without and with added water vapor, respectively. There are no significant differences in product distribution with and without added water vapor. In addition, ESI-MS2 measurements show that typical P1 and P2 products as labeled in both mass spectra have very similar MS2 spectra, suggesting that the product composition is not impacted by the presence of water. Measurements of the chamber SOA by AMS (see Fig. S11) also did not show significant differences in the particle mass spectra upon water addition. On the other hand, Fig. 12c shows the ESI mass spectra of SOA formed in the presence of 15 ppm formic acid. The formation of HMW products is significantly reduced as are the number of peaks in the P1 product region. However, P1 products at m/z 259, 275, and 291, as well as P2 products, for example, at m/z 527, 543, and 559, are still evident. All of these P1 products have very similar MS2 spectra (and thus likely similar structures) to those formed in the absence of formic acid. In contrast, P2 products at m/z 527 and 543 show significantly different MS2 spectra compared to those obtained without added formic acid.

Yao et al. (2014) reported an increase in the formation of α-cedronaldehyde in the presence of acetic acid, consistent with an increase in the present study in the relative abundance of the P1 product at m/z 259 assigned to α-cedronaldehyde when formic acid was added (Fig. 12c). The reaction of α-cedrene SCI with formic acid is expected to form α-formyloxy hydroperoxide (MW 298 Da), which is believed to contribute to SOA growth. While the corresponding sodium adduct at m/z 321 is not observed in the mass spectrum, this is not surprising as hydroperoxides are likely to undergo decomposition during SOA sampling and analysis (Witkowski and Gierczak, 2013).

Comparison of Fig. 12c to a shows that HMW products (P2–P4) formed with added formic acid are less important than those produced in the absence of formic acid. Combined with the smaller number concentration (Fig. 11), the data suggest that HMW products must be important in new particle formation. As fewer particles are formed with the addition of formic acid, more P1 products are available for each particle to grow to larger sizes.

The size-dependent composition of α-cedrene SOA and the effect of SCI scavengers on particle formation suggest that HMW products (P2–P4) play an important role in the initial stages of particle formation. This is consistent with earlier suggestions in simpler systems that higher molecular weight products of alkene ozonolysis are primarily responsible for nucleation, while many different products from low to high molecular weight can contribute to growth (Lee and Kamens, 2005; Sadezky et al., 2008; Winkler et al., 2012; Zhao et al., 2013; Ehn et al., 2014; Kidd et al., 2014a; Zhao et al., 2015; X. Zhang et al., 2015). Bonn and Moortgat (2003) estimated an upper limit for the saturated vapor pressure (Psat) of nucleating species produced from sesquiterpene ozonolysis to be 1.2 × 10-13 atm. Donahue et al. (2013) also suggested that organic vapors with Psat < 10-13 atm may contribute to particle nucleation. The Psat of typical P1 products, as well as the P2 products (e.g., ester and peroxyhemiacetal) are estimated by averaging the predictions (see Table S1 in the Supplement) from two group contribution methods – SIMPOL.1 (Pankow and Asher, 2008) and EVAPORATION (Compernolle et al., 2011) – and are shown in Fig. 13. The Psat for the P2 products are 3–6 orders of magnitude lower than the suggested nucleation threshold, consistent with their important contributions to initial particle formation. The newly observed products (P1-321 and P1-323) also have vapor pressures lower than the nucleation threshold and thus may in principle also contribute to nucleation. This is also expected to be the case for the HMW P3 and P4 products. However, without information on their chemical composition and structures, data on P3 and P4 could not be included in Fig. 13. In contrast, the smaller, more volatile P1 products will mainly contribute to particle growth, as suggested by their relatively larger abundance in ESI mass spectra of larger particles (Figs. 7 and S4).

Compared to P2 products at m/z 481–543, those with m/z > 543 are significantly less important in the flow reactor SOA (geometric mean diameter 13–26 nm) but contribute a greater fraction to total P2 products in the chamber SOA (geometric mean diameter 66 nm). This suggests that these larger P2 products (i.e., m/z > 543) play an important role in particle growth and may be formed mainly via condensed phase reactions of P1 products in the SOA.

Elemental analysis of SOA from chamber experiments using AMS results in an average O : C ratio of 0.34 ± 0.03 (1σ) and an H : C ratio of 1.51 ± 0.02. These values are within the range given for P1 products detected by ESI-MS (Table 2) and are consistent with our observation that P1 products predominantly contribute to particle growth. Along with the potential mechanisms discussed above to explain the formation of observed P1 and P2 products and their contribution to particle formation and growth, another mechanism considered for particle formation and growth is the production of extremely low-volatility organic compounds (ELVOCs), which was proposed to occur via sequential intramolecular hydrogen abstraction and O2 addition of RO2 radicals (Vereecken et al., 2007; Crounse et al., 2013; Ehn et al., 2014; Rissanen et al., 2015) and shown to play an important role in particle formation and growth from monoterpene oxidation (Zhao et al., 2013; Ehn et al., 2014; Jokinen et al., 2015). Because of the high oxygen content in ELVOCs, particles formed from these compounds are expected to have high O : C ratios. The ELVOC mechanism cannot be ruled out as being involved in initial particle formation during α-cedrene ozonolysis. However, based on the O : C ratios measured for chamber SOA, it is unlikely to be a dominant contributor to total particle mass in this system. Measurements by AMS are not possible for the flow reactor experiments as the particles are too small to be efficiently transmitted into the AMS.

It is clear from the impact of formic acid on particle formation that the SCI plays a key role in forming the HMW products and new particles. In the case of the trans-3-hexene ozonolysis (Sadezky et al., 2008; Zhao et al., 2015), the composition of the SOA clearly showed evidence for oligomer formation from the sequential reaction of RO2 radicals with SCI, leading to ESI mass spectra that showed the repeat unit of oligomers corresponding to SCI. A search for similar products in the α-cedrene reaction was not successful, indicating that while this may contribute, other mechanisms are more important (Zhao et al., 2015). This is not surprising, given the number of potential reaction paths for the SCI from α-cedrene (e.g., Fig. S9) and the large number of low-volatility products that can quickly contribute to growth and the SOA composition once nucleation has occurred.

The incorporation of 2-EHN into SOA and its very slow evaporation back out (Figs. 4 and 5) are consistent with a condensation type of growth mechanism (Finlayson-Pitts and Pitts, 2000; Seinfeld and Pandis, 2006) in which organic species are irreversibly taken up by the particle surface and thus their incorporation into particles depends on the collision frequency of the gas with the particle surface, and the magnitude of the uptake coefficient (Perraud et al., 2012). Such a growth mechanism is characteristic of highly viscous SOA, in which incorporated organic species undergo very slow evaporation because of the diffusion limitation, in contrast to an equilibrium mechanism that applies for low-viscosity liquid particles. This result also suggests that growth will occur not only via low-volatility products but also via uptake of higher-volatility species such as the smaller P1 products identified here.