2.1 Environmental chamber experiments and instrumentation

Experiments were conducted inside a 10 m³ Teflon^® chamber at 298 K, using UV lights to generate radicals. The NO₂ photolysis rate was measured to characterize the UV intensity (Carter et al., 2005) and was found to be similar to ambient levels (0.53 min⁻¹ at 0^∘ zenith angle) at approximately 0.5 min⁻¹. The chamber relative humidity (RH) ranged from 0 to 67 %. For dry experiments, the chamber was filled with dried clean air supplied by a clean-air generator (model 737R, Aadco). Under typical atmospheric conditions, elevated RH can be expected, especially within the marine boundary layer near coastal regions, where Cl-alkane chemistry may be important (Riedel et al., 2012). Therefore, SOA formation under humid conditions was also investigated. For humid experiments (i.e., RH > 20 %), the chamber was flushed overnight with humidified clean air. To reduce wall loss, dried ammonium sulfate seed particles were injected into the chamber. The seed particles were generated from a 0.01 M aqueous ammonium sulfate solution using an aerosol generation system (AGS 2002, Brechtel). Chlorine gas (101 ppm in N₂, Airgas) was injected as the Cl radical precursor. Each n-alkane (n-octane at 99 %, n-decane at 99 %, dodecane > 99%; Sigma-Aldrich) was first injected into a glass gas sampling tube (Kimble-Chase, 250 mL), which was flushed with gently heated clean air into the chamber at 2 L min⁻¹ for at least 30 min. Precursor NO (9.98 ppm in N₂, Airgas) and NO₂ (9.86 ppm in N₂, Airgas) were injected into the chamber using a mass flow controller (GFC17, Aalborg). Initial concentrations of VOC and oxidant precursors are summarized in Table 1. Gas-phase NO and NO₂ concentrations were monitored by a chemiluminescence monitor (200E, Teledyne). The O₃ concentration was monitored using a photometric ozone analyzer (400E, Teledyne). The start of photooxidation, initiated by turning on all the UV lights after the precursors were well mixed in the chamber, was designated as the reference point (i.e., time 0 min) for each experiment. UV lights were turned off after 60 min. The day before and after each SOA formation experiment, ammonium sulfate seed particles and 50 ppb of chlorine gas were injected into the chamber. UV lights were turned on to generate Cl radicals and remove residual reactive organic compounds in the chamber. Minimal SOA formation was observed during these cleaning experiments.

Particle size distributions were characterized using a scanning electrical mobility system (SEMS, Brechtel model 2002). The particle-phase bulk chemical composition was measured using an aerosol chemical speciation monitor (ACSM, Aerodyne). The ACSM was calibrated with 300 nm size-selected ammonium nitrate and ammonium sulfate aerosols generated from nebulized 0.005 M solutions using the AGS to determine the nitrate response factor (RF) and relative ionization efficiencies (RIE) for ammonium and sulfate, which are required for ion-to-mass signal conversions. Using electron impact ionization, the ACSM can measure the submicron, nonrefractory aerosol bulk composition at 1 min intervals (Budisulistiorini et al., 2013; Ng et al., 2011a). Using a standard fragmentation table (Allan et al., 2004), the ACSM can speciate the aerosol content into organics, nitrate, sulfate, ammonium, and chloride (Ng et al., 2011a). The ability of the ACSM to detect organic chloride using HCl⁺ (m∕z 36) has been demonstrated previously for isoprene–Cl SOA (Wang and Hildebrandt Ruiz, 2017). The default RIE_Chl value of 1.3 was used for chloride mass conversion. The Cl⁺ (m∕z 35) ion was excluded from chlorine quantification as it showed inconsistent response to nonrefractory chlorides (e.g., ammonium chloride). ACSM data were analyzed in Igor Pro V6.37 (Wavemetrics, Inc.) using ACSM local v1603 (Aerodyne) and other custom routines. Organic aerosol concentrations were calculated using ACSM measurements assuming a collection efficiency of 0.5 and RIE of 1.4, corrected for depositional particle wall loss (Pathak et al., 2007) but not for organic vapor loss (Huang et al., 2018b; Krechmer et al., 2017; Nah et al., 2017).

2.2 FIGAERO–CIMS

A high-resolution time-of-flight chemical ionization mass spectrometer (CIMS, Aerodyne) was used to measure the gas-phase chemical composition using I⁻ as the chemical ionization reagent. Humidified UHP N₂ was flushed over a methyl iodide permeation tube and then through ²¹⁰Po ionizer into the ion-molecule reaction chamber of the CIMS. Theory and operation of the CIMS are described in detail elsewhere (Aljawhary et al., 2013; Bertram et al., 2011; Lee et al., 2014; Wang and Hildebrandt Ruiz, 2017). The CIMS inlet was coupled to a Filter Inlet for Gases and AEROsols (FIGAERO), which has been used in a number of studies to investigate the chemical composition and volatility of particle-phase compounds (D'Ambro et al., 2017; Gaston et al., 2016; Huang et al., 2018a; Lee et al., 2016; Lopez-Hilfiker et al., 2014, 2015; Stark et al., 2017; Thompson et al., 2017). The FIGAERO system alternated between two operational modes. In the aerosol collection mode, a gas-aerosol mixture was drawn through a PTFE filter (Zefluor^® 2.0 µm 24 mm, Pall Corp.) at 3 SLPM for 15 to 45 min while gas species were sampled and analyzed. In the desorption mode, clean air or UHP N₂ was passed through the filter and heated to 200 ^∘C (measured just above the filter) at a rate of approximately 5 to 10 ^∘C min⁻¹ for 40 to 20 min, respectively. The filter was then soaked at 200 ^∘C for 20 min. The volatilized vapor was sampled by the CIMS, and the desorption signal as a function of temperature can be used to construct a one-dimensional (1-D) thermogram (e.g., Fig. 6a). For a monomeric compound, the desorption signal as a function of temperature is expected to be monomodal (Lopez-Hilfiker et al., 2014). The temperature at the peak desorption signal, T_max correlates with the enthalpy of sublimation (dH_sub) and the saturation vapor pressure (C^∗) of the compound (Lopez-Hilfiker et al., 2014, 2015). However, some compounds may exhibit bimodal or multimodal behavior, with a second desorption mode occurring at much higher temperatures than expected, which is often interpreted as the result of thermal decomposition of larger oligomeric compounds (Huang et al., 2018a; Lopez-Hilfiker et al., 2014; Stark et al., 2017; Wang et al., 2016).

Figure 1Representative trends of SOA and trace-gas species during the photooxidation period. Data from dodecane oxidation (Exp. 11) are shown. Particulate chlorine concentrations were multiplied by 70 for ease of comparison. Ions consistent with HONO and HO₂NO₂ were observed, indicative of secondary HO_x chemistry. Formation of ClONO and ClNO₂ due to oxidation of NO₂ by Cl^• was also observed.

Download

A FIGAERO–CIMS data analysis was conducted in Igor Pro with Tofware v2.5.10 (Tofwerk) and custom routines. Hundreds of ions belonging to diverse chemical families can be retrieved from the mass spectra. To simultaneously represent the chemical composition, relative aerosol volatility (i.e., T_max distribution), and multimodal thermal desorption behaviors, a two-dimensional (2-D) thermogram framework was developed. The 2-D thermogram is comprised of normalized unit-mass resolution 1-D thermograms, each expressed as a percentage color scale of the maximum desorption signal. Two-dimensional thermogram applications are discussed in Sect. 3.2. The advantage of using UMR over HR data is the ability to investigate the SOA thermal desorption behavior over the entire m∕z and volatility (i.e., T_max) range without having to assign chemical formulae to all ion. This HR analysis is time consuming, especially for high-molecular-weight compounds, the exact molecular composition of which can be difficult to ascertain. The disadvantage of using UMR over HR data is the overlapping of ions and potential interference by isotopic signals or nonadduct ions. Therefore, the 2-D thermogram should be used as a complement rather than a replacement of the HR analysis. Based on the SOA molecular composition as observed by the FIGAERO–CIMS, the average oxidation state of carbon (OS_C) may be estimated:

where NO₃ : C, Cl : C, O : C, and H : C are the molecular ratios of the number of -NO₃ functional groups, -Cl functional groups, non-NO₃ oxygen atoms, and H atoms to the number of carbon atoms for any given compound. The average SOA OS_C is calculated based on iodide adducts only. For simplicity, all organic ions were assumed to have equal sensitivity, which is known to vary with ion cluster binding energy and sample RH (Hyttinen et al., 2018; Iyer et al., 2016; Lopez-Hilfiker et al., 2016).

3.1 SOA and organic chloride formation

Experimental conditions and results are summarized in Table 1. Chlorine-alkane SOA yields increased with VOC precursor length, consistent with the trend observed for OH-alkane SOA (Jordan et al., 2008; Lim and Ziemann, 2009b; Presto et al., 2010; Schilling Fahnestock et al., 2015; Yee et al., 2012, 2013). For similarly functionalized alkane oxidation products, the vapor pressure decreases roughly loglinearly as the precursor carbon number increases from 8 to 15 (Jordan et al., 2008; Presto et al., 2010). Comparison of the unit-mass ACSM spectra for octane (Exp. 3), decane (Exp. 7), and dodecane (Exp. 11) SOA in Fig. S1 in the Supplement shows a consistent increase in the fractional contributions to the bulk OA mass (i.e., f_m∕z) by organic ions at m∕z 27 (${\mathrm{C}}_{\mathrm{2}}{\mathrm{H}}_{\mathrm{3}}^{+}$), 41 (${\mathrm{C}}_{\mathrm{3}}{\mathrm{H}}_{\mathrm{5}}^{+}$), 55 (${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{7}}^{+}$) and 57 (${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{9}}^{+}$) with increasing alkane length. Select odd m∕z ions, noticeably m∕z 55 and 57, have been used as tracers for hydrocarbon-like organic aerosol (HOA), and sometimes for primary aerosol emissions as well (Ng et al., 2011b; Ulbrich et al., 2009). Given the same oxidation conditions, SOA products derived from longer alkane precursors appeared less oxidized, as seen by the decrease in f₄₄ (presumably mostly ${\mathrm{CO}}_{\mathrm{2}}^{+}$), but are more hydrocarbon-like, as seen by the increase in f₅₇, consistent with previous observations (Lambe et al., 2012; Loza et al., 2014). As the alkane chain length increases, the increased SOA production also favors the partitioning of semivolatile compounds into the particle phase (Donahue et al., 2006; Pankow, 1994), further lowering f₄₄.

Time series of gas-phase NO, NO₂, O₃, Cl₂, and suspended particle-phase organics and chlorine concentrations during photooxidation are shown in Fig. 1 using dodecane–Cl oxidation under low RH (Exp. 11) as the representative example. The Cl₂ concentration was estimated using I⁻ CIMS by tracking the Cl₂I⁻ ion. In addition, ions consistent with HO₂NO₂, HONO, and ClNO₂ were observed in the gas phase, as shown in Fig. 1. The ClNO₂I⁻ ions are expected to correspond to both ClNO₂ (< 20 %) and chlorine nitrite (ClONO, > 80 %) from the reaction between Cl^• and NO₂ (Golden, 2007; Niki et al., 1978). Formation of HO₂NO₂ from the reaction of HO₂ and NO₂ has been observed using I⁻ CIMS previously (Veres et al., 2015). Formation of HONO was likely due to the interactions between HO_x and NO (see Reactions S2 and S3 in the Supplement). Under UV, HONO decays to background levels. Production of HO₂NO₂ and HONO is indicative of secondary HO_x chemistry enabled by primary Cl-alkane oxidation chemistry, consistent with previous studies (Wang and Hildebrandt Ruiz, 2017; Young et al., 2014).

Figure 2Formation pathway of chlorinated organics via chlorine addition to the DHF. H abstraction from DHF is also possible. Ozone may also react with the double-bond on the DHF, competing with chlorine radicals. Reaction pathways leading to DHF are adapted from OH-initiated oxidation of alkanes (Lim and Ziemann, 2009b).

Download

The NO concentration decreased sharply at the beginning of the photooxidation accompanied by increases in the concentrations of NO₂ and ozone. SOA production was most rapid during this initial period (0 to 10 min). Afterwards, at around 15 min in Fig. 1, the concentrations of NO₂ and ozone stabilized. Oxidation continued under UV driven by Cl radicals, and the SOA concentration began to decay due to oxidative fragmentation (Kroll et al., 2011; Lambe et al., 2012; Wang and Hildebrandt Ruiz, 2017) and wall loss. Ozone production continued slowly under UV at ∼0.24 ppb min⁻¹. Slight NO₂ production (< 0.1 ppb min⁻¹) was also observed, which may be due to production of nitrous acid (HONO) on a Teflon^® surface, a common background contaminant in environmental chambers (Carter et al., 2005; Rohrer et al., 2005).

SOA production was dependent on the initial NO and NO₂ concentrations, as shown in Fig. S2. Higher initial NO concentrations led to higher-SOA yields and lower ozone production for all precursors, as shown in Table 1 and Fig. S2. This is similar to alkane–OH SOA formation, where higher NO concentrations lead to more abundant organic nitrate formation, which increases the SOA volume (Schilling Fahnestock et al., 2015) and density (Loza et al., 2014). This is consistent with FIGAERO–CIMS results where the particulate organonitrate molar ratio (calculated as the average number of -NO₃ per molecule) and mass fraction were the highest in SOA produced under NO-only conditions (see Table S1 in Supplement). The organonitrate molar ratio also increased with the alkane precursor length, from 0.57–0.64 for octane to 0.75–0.93 for dodecane. A similar trend was observed for the organonitrate mass fraction, which increased from 0.53–0.58 for octane to 0.66–0.72 for dodecane. The alkane–Cl SOA yields, ranging from 0.16 (octane) to 1.65 (dodecane), are 4 times higher than the alkane–OH SOA yields (0.04 for octane and 0.35 for dodecane) that were obtained using 1 ppm alkane, 10 ppm methyl nitrite, and 10 ppm NO (Lim and Ziemann, 2009a), which are much higher than the precursor concentrations used here (13 to 15 ppb alkanes, < 40 ppb NO_x, 40 ppb Cl₂). Formation of primary alkyl nitrates due to terminal H abstraction by Cl could contribute to the higher SOA yields observed for alkane–Cl than for alkane–OH oxidation, as primary alkyl nitrates have lower saturation vapor pressures than secondary alkyl nitrates (Lim and Ziemann, 2009b; Yeh and Ziemann, 2014, 2015). The apparent SOA density, calculated using the ACSM SOA mass measurement and the SEMS SOA volume measurement, was 2.1±0.3 g cm⁻³, which is substantially higher than that reported for OH-alkane SOA formed under high-NO_x conditions, 1.06 to 1.28 g cm⁻³ (Lim and Ziemann, 2009a; Loza et al., 2014). These differences could be due to uncertainties associated with the ACSM and similar instruments, specifically the RIE (Li et al., 2018; Xu et al., 2018) and CE (Docherty et al., 2013; Middlebrook et al., 2012; Robinson et al., 2017), which may vary with SOA composition, oxidation state, or phase state. Assuming an SOA density of 1.06 g cm⁻³, the SOA yield (0.10 to 0.99) is still higher for Cl-alkane oxidation compared to OH-alkane oxidation. Molar SOA yields, calculated using the average molecular weight of species identified with the FIGAERO–CIMS, are summarized in Table S1.

Although the initial Cl-alkane reactions proceed via a H-abstraction pathway, particulate chlorine was observed using the ACSM as shown in Fig. 1. Direct halogenation of the C₈₋₁₂ alkyl radicals is expected to be minimal, given the low amounts of Cl₂ present (maximum of 40 ppb). A carbon–carbon double bond is required to enable Cl-addition reactions and organic chlorine formation. The heterogeneous production of dihydrofuran (DHF) via 1,4-hydroxycarbonyl uptake, acid-catalyzed isomerization, and dehydration reactions (Atkinson et al., 2008; Jordan et al., 2008; Lim and Ziemann, 2009c, b), followed by Cl addition to the DHF double bond, could be the source of observed organic chlorine. A condensed reaction pathway is shown in Fig. 2. The delay in particulate chlorine formation relative to that of bulk organics, as shown in Fig. 1, is consistent with the rate-limiting heterogeneous DHF production (Holt et al., 2005; Zhang et al., 2014). Under high-NO_x conditions, the peroxy radical product from the chlorine-addition pathway could react with NO to form a cyclic hemiacetal chloronitrate or an alkoxy radical that would undergo ring-opening reactions. Compounds resembling chloronitrates (e.g., ONO₂−C₁₂H₁₈ClO₃ ⋅ I⁻ for dodecane) were tentatively identified in the particle phase using the FIGAERO, but they were not well separated from the shoulder of nearby organonitrate peaks (e.g., ONO₂−C₁₂H₂₁O₅ ⋅ I⁻), as shown in Fig. S9. Select C₂₋₆ organochlorides were well separated from nearby peaks and confirmed by the distinct Cl isotopic signal, the thermograms of which are shown in Fig. S10. The most abundant gas-phase organic chlorine compounds observed were products of ring-opening reaction pathways such as C₂H₃ClO₂ and C₄₋₆ compounds. Small amounts of C₂H₂Cl₂ were also observed.

Figure 3Trends of early-generation gas-phase oxidation products (hydroxynitrates) and multigeneration-oxidation products (hydroxycarbonyl nitrates) observed during chlorine-initiated oxidation of octane (Exp. 3), decane (Exp. 7), and dodecane (Exp. 11). Species shown were first normalized against the I⁻ reagent ion signal and then normalized against their respective maxima.

Download

Organochloride formation is expected to be lower under humid conditions, where DHF formation is inhibited (Holt et al., 2005; Zhang et al., 2014; Ziemann, 2011). Evidence consistent with organochloride suppression under humid conditions was observed for dodecane SOA only, where the organochloride (including chloronitrates) mass fraction decreased from 0.15 (Exp. 11, < 5 % RH) to 0.13 (Exp. 12, 67 % RH) as measured by the FIGAERO–CIMS. The mass fraction of the -Cl functional group decreased from 1.8 E⁻² to 1.6 E⁻² as measured by the FIGAERO–CIMS (Table S1) or from 1.4 E⁻² to 1.1 E⁻² as measured by the ACSM (Table 1) as the RH increased. No clear differences were observed for octane or decane SOA, which may be due to the less extreme RH conditions investigated, uncertainties with organochloride ion identification in the CIMS, or the lower organochloride concentrations observed in Exps. 1–8, which is especially challenging for chloride quantification using the ACSM.

DHF ozonolysis can compete with chlorination. For instance, under typical marine boundary layer (MBL) conditions, the chlorine-initiated oxidation was estimated to be a significant sink of 2,5-DHF, accounting for 29 % of the reaction in the presence of OH and O₃ (Alwe et al., 2014). In contrast, only 1.8 % of 2,3-DHF consumption was attributed to Cl radical chemistry in the MBL, owing to the increased reactivity of 2,3-DHF (relative to 2,5-DHF) towards ozone and OH radicals (Alwe et al., 2014). The reported alkane–OH reaction mechanisms expect the formation of substituted 2,3-DHF (Ziemann, 2011). The alkane-derived organochloride yield under ambient conditions is therefore expected to be smaller in the presence of elevated RH and O₃ levels. Although ozone can impose an upper limit on alkane SOA yields and oxidation state, suppressing the multigenerational OH-alkane oxidation chemistry (Zhang et al., 2014), the continued gas-phase processing of OH-alkane and DHF-ozonolysis products via H abstraction by chlorine radicals could counteract these limitations.

The gas-phase alkane–Cl oxidation products formed under high-NO_x conditions were dominated by organonitrates, which was also reflected in the dominance of organonitrates in the particles phase (see Table 1 and Sect. 3.2). Figure 3 shows the formation of early-generation (e.g., hydroxynitrates, ${\mathrm{ONO}}_{\mathrm{2}}-{\mathrm{C}}_{\mathrm{n}}{\mathrm{H}}_{\mathrm{2}\mathrm{n}+\mathrm{1}}\mathrm{O}$) and multigenerational (e.g., oxidized organonitrates, ${\mathrm{ONO}}_{\mathrm{2}}-{\mathrm{C}}_{\mathrm{n}}{\mathrm{H}}_{\mathrm{2}\mathrm{n}-\mathrm{5}}{\mathrm{O}}_{\mathrm{4}}$) oxidation products from different alkane precursors. Like the bulk particle-phase composition, gas-phase compounds derived from smaller alkane precursor were more oxidized given similar oxidation conditions (e.g., NO, NO₂, and Cl₂): the signal of oxidation products with similar oxygen numbers (and higher oxidation state) peaked earlier in the photooxidation period for smaller precursors (e.g., ONO₂−C₈H₁₁O₄ vs. ONO₂−C₁₂H₁₉O₄ in Fig. 3). As oxidation continued, driven primarily by gas-phase chemistry (Aimanant and Ziemann, 2013), the importance of fragmentation reactions increased relative to that of functionalization reactions (Lambe et al., 2012). The heterogeneous oxidation of SOA (Bertram et al., 2001; George and Abbatt, 2010), which is expected to drive the oxidation of very large (n > 30) alkanes (Lim and Ziemann, 2009b), may also contribute to oxidation and fragmentation observed here, but its impacts are beyond the scope of this work. Assuming a uniform CIMS sensitivity, Figure 4a shows that the gas phase was dominated by C₃ to C₅ organic nitrates. Figure 4b shows that the OS_C increased, while n_C decreased as photooxidation continued, indicative of fragmentation reactions. In addition to hydroxynitrates and hydroxycarbonyl nitrates, dinitrates and trinitrates were also observed. The CIMS was not sensitive towards simple, alkane-derived alkyl nitrates (${\mathrm{ONO}}_{\mathrm{2}}-{\mathrm{C}}_{\mathrm{n}}{\mathrm{H}}_{\mathrm{2}\mathrm{n}+\mathrm{1}}$). The lack of sensitivity of the I⁻ reagent ion towards simple alkyl and keto nitrates has been reported previously for isoprene- and monoterpene-derived organic nitrates (Lee et al., 2016).

Figure 4Comparison of the (a) average composition and (b) the increases in average OS_C of gas-phase reaction products over the photooxidation period (0–60 min), as well as the (c) average particle-phase composition and (d) the evolution of average SOA OS_C during FIGAERO desorption for octane (Exp. 3), decane (Exp. 7), and dodecane (Exp. 11). Color-coded ions in (a) and (c) indicate molecules that were much more abundant in reaction products derived from a particular alkane precursor. In (c), O_x, dO_x, and tO_x were used as shorthand notations for oxidized organic nitrate, dinitrate, and trinitrates with x number of oxygens (excluding those from ONO₂). Sticks in (a) and (c) were not stacked, and no x-axis offset was applied during plotting. In (b), OS_C increased and n_C decreased as photooxidation continued, consistent with photooxidative fragmentation. In (d), SOA derived from larger alkane precursors appeared less oxidized, consistent with results shown in Table 1 and Fig. S1. Methods used to calculate the average OS_C and n_C are detailed in the Supplement.

Download

3.2 Two-dimensional thermogram

The FIGAERO filter spectra are shown in Fig. 4c for octane (Exp. 3), decane (Exp. 7), and dodecane (Exp. 11). A logscale version of Fig. 4c is shown in Fig. S3. The spectra are calculated from the average desorption ion signals observed when the filter temperature was between 40 and 140 ^∘C. As shown below and in Fig. 5, SOA components desorbed most effectively in this temperature range, with most organic ions having T_max within this temperature range. At desorption temperature above 140 ^∘C, inorganic ions began to dominate the spectra. The C₈₋₁₂ alkane–Cl SOA share many similarities towards the lower m∕z (< 320) range, which consisted of C_≤7 oxidized organic compounds. For larger alkane precursors, the particle-phase composition was dominated by C_n organic nitrates, which are grouped by their degree of oxygenation (i.e., number of nonnitrate O atoms), O_x, as shown in Fig. 4c. O₁ to O₅ mononitrates, O₂ to O₄ dinitrates, and O₁ to O₃ trinitrates dominated each oxygenation group. O₆ to O₈ mononitrates were present in the particle phase but were less abundant than the nearby O₂ to O₄ dinitrates (e.g., (ONO₂)₂−C₁₂H₂₀O₂ at m∕z 447 > (ONO₂)₂−C₁₂H₂₂O₂ at m∕z 449 > (ONO₂)−C₁₂H₁₇O₆ m∕z 446 > (ONO₂)−C₁₂H₁₉O₆ at m∕z 448). An example of a C₁₂ mononitrate distribution is shown in Fig. S4. Assuming the same sensitivity towards the different organic nitrates observed, the organic nitrate abundance follows a bell-shaped distribution, similarly to field observations for C₅ and C₁₀ organic nitrates derived from isoprene and monoterpenes, respectively (Lee et al., 2016). For dodecane mononitrates, the abundance peaked around O₃ and O₄ and decreased towards O₂ and O₅, as shown in Fig. S4. Dinitrate abundance decreased from O₂ to O₄. Trinitrate decreased from O₁ to O₃. Similar trends can be observed for octane and decane SOA. As the precursor length increased, the SOA appeared less oxidized, as shown in Fig. 4d, consistent with ACSM observations for f₄₄ (see Table 1 and Fig. S1), which is correlated with OS_C (Canagaratna et al., 2015). Simple hydroxynitrates (e.g., ONO₂−C_nH_2n+1O) were not observed in the particle phase as they were completely oxidized in the gas phase, as shown in Fig. 3.

Figure 5Comparison of two-dimensional thermograms for (a) octane–Cl SOA under low RH from Exp. 3, (b) dodecane–Cl SOA under low RH from Exp. 11, and (c) dodecane–Cl SOA under high RH (67 %) from Exp. 12. The color represents the ion intensity at a m∕z as a percentage of the maximum desorption signal observed at that m∕z. Different thermal desorption regions in (a) are dominated by (1) low-temperature thermal decomposition products and nonadducts (2) monomers (3) oligomers and their decomposition products and (4) thermal decomposition of ammonium sulfate or extremely low-volatility compounds. Region (5) is unresolved but appears to be thermally unstable oligomers. Region (2) and (3) can overlap, as shown in (d) for (b) and as shown in (e) for (c). At each nominal m∕z, only the dominant ion is labeled in (d) and (e). If multiple ions of similar intensity were present, only general descriptions are given in the annotations, where ON stands for organonitrate and CN stands for chloronitrate. Nonnitrated organic ions were observed but not labeled because they were positioned between more intense organonitrates at neighboring m∕z coordinates. T_max values are included in parenthesis. Increasing the RH suppressed oligomer formation, enhancing the monomer features in (c) and (e) when compared with (b) and (d). The m∕z values shown includes the chemical ionization reagent I⁻ (m∕z 127).

Download

In addition to the identification of aerosol chemical composition, the FIGAERO–CIMS can be used to estimate the aerosol volatility using empirical correlations between C^∗ and T_max (Lopez-Hilfiker et al., 2014, 2015). The particle-phase chemical composition and volatility distribution can be represented in several ways. For instance, desorption signals of select compounds can be normalized against the maximum and plotted on the same 1-D thermogram (ion signal vs. filter temperature) to compare their relative volatilities (Lopez-Hilfiker et al., 2014). Scatterplots of oxygen number or O : C ratio vs. carbon number colored by T_max or particle-phase mass fraction have been used to show changes in aerosol composition (Huang et al., 2018a; Lopez-Hilfiker et al., 2014, 2015). To show the differences between various types of thermal desorption products (i.e., monomers, thermal decomposition products, and oligomers), averages of the 1-D thermograms can be shown (Huang et al., 2018a; Lopez-Hilfiker et al., 2015). Scatterplots of T_max or the OS_c vs. molecular weight have been used to distinguish monomers and oligomers for select compounds (Wang et al., 2016). Here we propose a new framework to represent aerosol composition, relative volatility, and oligomer decomposition simultaneously using 2-D thermograms shown in Fig. 5a–e.

As illustrated in Fig. 5a for octane–chlorine SOA, the thermal desorption products can be separated into five different groups. Region 1 (m∕z < 350, 40 < T_max < 90 ^∘C) was composed of a group of semivolatile compounds with similar T_max values. Iodide adducts in this region correspond to species that are too volatile to be present as molecular compounds in the particle phase and are likely low-temperature decomposition products. A prominence of ions smaller than m∕z 127 including Cl⁻ and a range of organic ions could be due to acid exchange or charge transfer products (as opposed to I⁻ adducts) of low-temperature decomposition products. Although I⁻ is a relatively soft ionization method compared to electron impact ionization, collision-induced ion fragmentation and cluster dissociation cannot be ruled out as a potential source for nonadducts. Region 2 (350 < m∕z < ∼700, 40 ^∘C < T_max < 120 ^∘C) consisted of monomers, dimers, and some oligomers. Oligomerization may proceed via condensed-phase reactions between cyclic hemiacetal compounds, forming acetal dimers, or reactions between cyclic hemiacetal and 1,4-hydroxycarbonyl compounds, forming hemiacetal dimers (Aimanant and Ziemann, 2013). Condensed-phase, acid-catalyzed isomerization of hemiacetal and acetal dimers is also possible (Aimanant and Ziemann, 2013). For OH-initiated oxidation of dodecane under high-NO_x conditions, hemiacetals and nitrate hemiacetals are reported to dominate the oligomer composition (Schilling Fahnestock et al., 2015), where nitrate hemiacetals could form via the reactive uptake of hydroxy dihydrofuran and hydroxycarbonyl nitrate compounds (Schilling Fahnestock et al., 2015). The three oligomerization pathways are illustrated in Fig. S11. Ions consistent with acetal, hemiacetal, and nitrate hemiacetal compounds were observed in the particle phase using the FIGAERO–CIMS in regions 2 and 3. Oligomer formation via aldol condensation by highly oxidized compounds is also possible (Schilling Fahnestock et al., 2015). Oligomerization may also proceed through a NO₃ elimination and a reverse esterification process involving two alkyl nitrate compounds to form a single nitrate oligomer, which was recently proposed for α-Pinene SOA, though the exact mechanism remains elusive (Faxon et al., 2018).

An overall positive correlation between T_max and molecular weight was observed in region 2. A similar correlation was also reported for SOA derived from oleic acid (Wang et al., 2016) and from α-Pinene (Faxon et al., 2018). In other words, overall, volatility decreased with molecular weight for compounds in region 2. However, within small m∕z segments in region 2, the correlation was not strictly linear and the T_max varied cyclically. When the carbon and oxygen numbers were fixed (i.e., fixed O : C), the T_max decreased (i.e., volatility increased) with increasing H : C (i.e., decreasing OS_C), as shown in Fig. 5e for C₁₂ organic nitrates (e.g., ${\mathrm{ONO}}_{\mathrm{2}}-{\mathrm{C}}_{\mathrm{12}}{\mathrm{H}}_{\mathrm{19}-\mathrm{25}}$O₃ ⋅ I⁻ from m∕z 400 to 406) observed in dodecane–Cl SOA desorption. Although decreases in OS_C are often associated with increases in the aerosol volatility (Donahue et al., 2011; Kroll et al., 2011), fragmentation notwithstanding, the increase in H : C in each fixed O : C ion group was expected to decrease in aerosol volatility. The increase in H : C was likely due to increased hydroxy functionalization over ketone functionalization, which should result in lowering of the saturation vapor pressure according to group contribution theory (Pankow and Asher, 2008). In addition, the T_max decreased from O₃ : C₁₂ organonitrates (e.g., ONO₂−C₁₂H₂₁O₃, T_max 95 ^∘C) to O₄ : C₁₂ organonitrates (e.g., ONO₂−C₁₂H₂₁O₄, T_max 89 ^∘C). The inconsistencies observed within small m∕z segments may be due to loading-dependent T_max shift (see Sect. 3.3), where T_max increases with mass loading. As shown in Fig. 4c, the organonitrate ion intensities decreased from O₃ onwards. Overlapping of monomer and oligomer desorption peaks could also increase the apparent T_max.

Region 3 (T_max > 90 ^∘C) consisted of thermal decomposition products across a wide m∕z range. Select ions were present in both region 2 and region 3, resulting in bimodal thermal desorption, which can complicate T_max estimation using the 2-D thermogram, especially when ions from the two desorption modes overlap. The thermal decomposition products may be the result of reversible dissociation of oligomers (Faxon et al., 2018; Schobesberger et al., 2018) or the cleavage of C-O or C-C bonds, which has been observed in the thermal decomposition of monomeric organic acids (Stark et al., 2017). Additionally, SOA may exist in a “glassy” state (Koop et al., 2011), trapping volatile components (Perraud et al., 2012) that are released as the SOA mass is depleted. Region 4 (150 ^∘C < T_max < 170 ^∘C) consisted of mostly thermal decomposition products of the (NH₄)₂SO₄ seed particles. There was a clear gap between region 3 and 4. It is possible that the edge of regions 2 and 3, towards a high T_max range, is the FIGAERO–CIMS's volatility detection limit for alkane–Cl SOA, where the energy required for vaporization exceeds that for thermal decomposition (Lopez-Hilfiker et al., 2014). Plateauing of T_max towards the high m∕z range of region 2 appeared to agree with this hypothesis, for which further increases in molecular weight, carbon number, and the degree of oxidation translated to negligible T_max increase. Region 5 (m∕z > 700) likely consisted of high molecular weight, low volatility oligomers. Region 5 therefore likely consisted of C${}_{\mathrm{24}-\mathrm{36}+}$ (dimers and above) compounds. Thermal desorption of some high-molecular-weight compounds at temperatures much lower than expected (T_max∼80 ^∘C) was observed for decane SOA and dodecane SOA. No definitive molecular composition was determined for these ions due to low signal intensity and wide range of potential chemical formulae.

Figure 6(a) The 1-D thermograms for levoglucosan (C₆H₁₀O₅) at different loading conditions and (b) the correlation between filter loading and T_max. Levoglucosan aerosol was generated by nebulizing a 1.2 E⁻² M aqueous solution. The aerosol was injected into the clean Teflon chamber, collected onto the FIGAERO filter, and analyzed. The T_max-loading correlation for pure levoglucosan could be described by a sigmoid function, leveling off at 41 and 64 ^∘C under very low and very high loading conditions, respectively.

Download

The region boundaries can vary depending on the SOA composition as well as mass loading, as will be discussed in Sect. 3.3. The distribution and general features of the regions are consistent for alkane–Cl SOA formed in different experiments (Figs. 5 and S8). Comparison of 2-D thermograms for octane–Cl SOA (Fig. 5a) and dodecane–Cl SOA (Fig. 5b) shows that the dodecane SOA contains unique, high-molecular-weight oligomers (m∕z > 750), which are accompanied by significantly stronger thermal decomposition features in region 3. Bimodal thermal desorption behavior was observed, as shown in Fig. 5d, which masked the volatility m∕z dependence expected for region 2. Compared to SOA formed under dry conditions (Fig. 5b, Exp. 11), SOA formed under humid conditions (Exp. 12, 67 % RH) contained less high-molecular-weight oligomers in region 5 and exhibited much less thermal decomposition behavior in region 3, as shown in Fig. 5e, where a more distinct T_max-m∕z correlation could be established compared to Fig. 5d. RH-induced oligomer suppression has been reported for toluene and α-Pinene SOA formation (Hinks et al., 2018; Huang et al., 2018a). The oligomer decrease under high RH could also be due to RH-volatility dependent organic vapor wall loss (Huang et al., 2018b). A clear m∕z dependence of signal reduction under humid conditions compared to the dry conditions is shown in Fig. S5. Additionally, SOA yield decreased (37 % for decane and 22 % for dodecane) under humid conditions while ozone production increased, which could be related to humidity-induced inhibition of 1,4-hydroxycarbonyl uptake and heterogeneous DHF formation (Holt et al., 2005). DHF is a key intermediate product for multigenerational oxidation chemistry and is reactive towards ozone (Zhang et al., 2014; Ziemann, 2011). Reduction in SOA yields under humid conditions has also been reported for dodecane–OH SOA formation (Schilling Fahnestock et al., 2015). The combined effects of DHF reduction and ozone enhancement would suppress organic chloride formation, which was observed for dodecane SOA, as discussed in Sect. 3.1.

3.3 T_max shift

As the desorption temperature increased above 140 ^∘C, all dominant ions observed shared the same T_max, except for some background ions (e.g., ${\mathrm{IHNO}}_{\mathrm{3}}^{-}$), as shown in Fig. 5a–c, which were identified to be related to the thermal decomposition of (NH₄)₂SO₄ seed particles. From this process, H₂SO₄ vapor molecules produced were either detected as H₂SO₄ ⋅ I⁻ or were deprotonated to produce ${\mathrm{HSO}}_{\mathrm{4}}^{-}$, acting as a secondary chemical ionization reagent. Examples of first-order (e.g., ${\mathrm{HSO}}_{\mathrm{4}}^{-}$), second-order (e.g., H₂SO₄ ⋅ ${\mathrm{HSO}}_{\mathrm{4}}^{-}$) and third-order sulfate (e.g., (H₂SO₄)₂ ⋅ ${\mathrm{HSO}}_{\mathrm{4}}^{-}$) clusters are shown in Fig. S6. It is worth noting that, while the T_max of sulfate ions was uniform within each filter desorption run, the T_max varied between filter runs, which can be seen in Fig. 5a–c to be between 146 and 153 ^∘C. Characterization experiments suggested that T_max may increase with filter loading, as shown in Fig. 6a for pure levoglucosan aerosols and in Fig. S7a for pure (NH₄)₂SO₄ aerosols. Whereas the T_max-mass loading correlation appeared to be linear for pure (NH₄)₂SO₄, as shown in Fig. S7b. for T_max between 136 and 149 ^∘C, a roughly sigmoidal correlation was observed for pure levoglucosan, as shown in Fig. 6b, where T_max increased quickly from 43 to 62 ^∘C as filter loading increased from 0.2 to 0.8 µg.

T_max shift phenomena have been observed in several FIGAERO–CIMS studies (D'Ambro et al., 2017; Gaston et al., 2016; Huang et al., 2018a; Lopez-Hilfiker et al., 2015; Thompson et al., 2017). Explanations vary for the observed T_max shifts. For example, pinonic acid (C₁₀H₁₆O₃) identified in α-Pinene ozonolysis SOA had a much higher T_max (40 ^∘C) than expected from calibration (T_max < 32 ^∘C), which was attributed to interactions between pinonic acid and other SOA components, causing a decrease in the measured apparent vapor pressure (Lopez-Hilfiker et al., 2015). However, no matrix effects were reported for the pinonic acid T_max value (now around 65 ^∘C instead of < 32 ^∘C) obtained from a synthetic mixture of organic acid calibrants (Thompson et al., 2017), where some interactions between different organic molecules might have been expected. Whereas the pure compounds are shown to have sharp monomodal desorption peaks (Lopez-Hilfiker et al., 2014), the field campaign-average thermograms for individual compounds have broader desorption peaks (Thompson et al., 2017), which could be due to the presence of isomers (Thompson et al., 2017), variations in filter loading, differences in aerosol viscosity (Huang et al., 2018a), or interferences from oligomer decomposition products (Lopez-Hilfiker et al., 2014). Another T_max shift example was reported for ambient biomass burning measurements, where the levoglucosan T_max varied between 58–70 ^∘C, in the range of the 61.5 ^∘C literature value (Lopez-Hilfiker et al., 2014) and the values observed here (41–64 ^∘C), between 17:00 and 10:00 over the course of the campaign (Gaston et al., 2016). Between 10:00 and 19:00, the levoglucosan T_max was significantly higher at 100 ^∘C, which was attributed to thermal decomposition of oligomers produced from acid-catalyzed heterogenous reactions, as indicated by the increase in sulfate ion intensities during the same period (Gaston et al., 2016). It was not clear whether the sulfate ions measured were derived from sulfuric acid or from the decomposition of (NH₄)₂SO₄, which could change the assumption made for aerosol acidity. A T_max shift was also observed for isoprene SOA, where the mass and T_max of nonnitrated OA decreased with increasing NO_x concentration (D'Ambro et al., 2017). A recent investigation reported increases in T_max for α-Pinene ozonolysis SOA produced under dry conditions compared to SOA produced under humid conditions (Huang et al., 2018a). A positive correlation between aerosol loading and T_max was reported for the first time, and the uniform thermal desorption peak shape assumption was questioned (Huang et al., 2018a). The loading dependence was reported to plateau at around 2 to 4 µg (converted from CIMS ion intensity assuming maximum sensitivity, i.e., that of formic acid), similarly to observations for levoglucosan standards as shown in Fig. 6b, possibly due to saturation effects. Because increased oligomer content was observed under dry conditions, the authors suggested that viscosity effects were responsible for the observed T_max shift (Huang et al., 2018a). It should be noted that plateauing of the T_max-loading dependence was demonstrated using averaged values, where ions were lumped based on carbon number. Those that had less than or equal to 10 carbons were designated as monomers, which likely included desorption ions across multiple desorption regions (see Fig. 5). This categorization was perhaps too broad, and the loading dependence of individual ions could be lost after averaging. In contrast, the T_max of oligomers, which were defined as compounds that contained more than 10 carbons – thus were a more precisely defined group – only began to plateau at higher concentrations.

The T_max mass dependence was also observed for alkane–Cl SOA. For Exp. 6, three filters were collected and analyzed. The first filter was collected at 3 SLPM for 45 min, followed by the second filter at 3 SLPM for 30 min, and afterwards the third filter at 3 SLPM for 15 min. The ratios of the unit-mass integrated ion signals during the desorption period for the first two filters relative to that of the third filter are shown in Fig. S8a. The associated 2-D thermograms are shown in Fig. S8b–d. Compounds in the T_max-m∕z dependence region (Region 2) showed a 10 to 20 ^∘C T_max increase from the lowest filter-loading to the highest filter-loading conditions. T_max of sulfate-related ions increased from 149 to 172 ^∘C with filter loading. The increase of enhancement ratio with m∕z, as shown in Fig. S8a, reflects the changes in the composition of suspended aerosol in the chamber over time. Between the first (high-loading) filter collection and the third (low-loading) filter collection, volatility-dependent vapor wall loss may lead to a disproportionate decrease in high-molecular-weight, low-volatility compounds (Huang et al., 2018b; Krechmer et al., 2016), resulting in increasingly greater enhancement towards the high m∕z region shown in Fig. S8a. The correlation between T_max shift and integrated desorption signal for organics was not linear, unlike for ammonium sulfate, which may be due to matrix or saturation effects as previously suggested (Huang et al., 2018a). T_max-loading dependence may differ for individual FIGAEROs due to design variations or artifacts such as sample distribution within the filter matrix or nonuniform heating, which would result in different aerosol evaporation kinetics (Schobesberger et al., 2018). Overall, our results show that the T_max for organic and inorganic compounds varies with loading, which needs to be accounted for when estimating aerosol component volatility from the empirical correlations between T_max and C^∗ using the FIGAERO–CIMS.