All experiments were performed in batch mode in the Caltech 17.9 m3 fluorinated ethylene propylene (FEP) Teflon-walled environmental chamber, which hangs in a temperature-controlled enclosure. The chamber volume was characterized according to the procedure outlined in . While the chamber pressure remains constant throughout the duration of an experiment, the volume decreases as air is sampled by various instruments; the fraction of the volume at the end of the experiment compared to the beginning of it is given in Table . Before each experiment, the chamber was flushed for >24 h with clean air (compressed air with ozone, nitrogen oxides, water vapor, and organic carbon removed). The radical source H2O2 was injected at 42 ∘C and 5 L min-1 into the chamber, followed by the injection of benzyl alcohol (Sigma Aldrich ReagentPlus, ≥ 99 %) with gentle heating (60 ∘C) at 2 L min-1 (5 L min-1 for experiments S1–S3 and E1) for > 50 min. The purity of the benzyl alcohol was verified with nuclear magnetic resonance (NMR) spectroscopy. Meanwhile, a 0.06 M (NH4)2SO4 solution (0.15 M for experiments S2 and E1) was atomized and the resulting particles dried, charge-conditioned with a TSI Model 3088 soft X-ray neutralizer, and then injected into the chamber for varying lengths of time (depending on the desired initial seed concentration; note that no particles were injected for experiment S1). The solution was sonicated before each injection. Then, NO (506.9 ppm ± 2 %, Airgas Specialty Gases, Certified Standard) or, for experiment E1, NO2 (488 ppm, Air Liquide) was injected into the chamber at 5 L min-1 to achieve the desired initial NO or NO2 concentration. Ultraviolet broadband lights centered around 350 nm were used to photolyze H2O2 with a rate of jH2O2≈4.7×10-6 s-1, calculated using the measured variation in irradiance with wavelength and the NO2 photolysis rate (jNO2=6.2(±0.1)×10-3 s-1) found using a 0.29 L quartz tube and the procedure outlined in . Experiment L1 was performed at 8 % of the light strength of the other experiments (jH2O2≈3.7×10-7 s-1).

A Vaisala HMM211 probe was used to measure the temperature and humidity of the chamber. Humidity was calibrated for RH from 11 % to 95 % (using LiCl, KNO3, Mg(NO3)2, and MgCl2 salts). A Teledyne nitrogen oxide analyzer (Model T200) was used to measure the NO and NO2 concentrations throughout the experiments; note that this instrument measures the contribution of NOy compounds (e.g., organic nitrates) as NO2. Due to some drift between experiments, linear fits were performed on the slope and offset calibrations, except for experiments S2–S3 and U5, due to a calibration problem. Ozone was measured with a Horiba ambient monitor. NO, NO2, and O3 concentrations were recorded every 30 s. Humidity and temperature uncertainties were calculated as standard deviations from the mean value, where measurements were taken every 30 s throughout the experiment. Initial NO and NO2 mixing ratios were determined (as well as their standard deviations) prior to irradiation during the background collection period (usually ≳ 60 min). For experiments N1–N6 and U6, NO was continuously injected during oxidation to maintain a stable NO mixing ratio.

A CF3O- chemical ionization mass spectrometer (CIMS) measured oxidation products and the benzyl alcohol concentration by scanning m/z ratios between 50 and 330. The CIMS is equipped with a Varian 1200 triple quadrupole mass analyzer. A custom-built inlet was used to ensure that the sample was taken at a constant temperature (the top of the inlet was 25 ∘C). To reduce loss of vapor to the tubing prior to analysis, the CIMS sampled off of a bypass flow that was accelerated using a mechanical pump.

The 193 m/z signal (the mass of benzyl alcohol and CF3O-), which was measured every 162 to 172 s, was normalized to the 86 m/z signal (the M+1 peak for CF3O-) and used to measure the benzyl alcohol concentration. This signal was calibrated using dilutions of an 800 L Teflon bag of ∼ 44 ppb benzyl alcohol. The concentration in this bag was verified using Fourier transform infrared absorption (FT-IR) spectroscopy with a 19 cm path length and absorption cross sections from the Pacific Northwest National Laboratory (PNNL) database. In this way, any wall or sampling loss was accounted for since the CIMS sampled from the same volume as the FT-IR. Multiple FT-IR samples were taken in succession until there were consistent spectra; this was to ensure a minimal effect from any compound deposited on the FT-IR instrument walls or sampling lines.

For the experiments labeled U3 and U7–U8, there were errors with the CIMS measurements. Correspondingly, Table  does not report an initial benzyl alcohol concentration, a first-order exponential fit to the benzyl alcohol decay, or any SOA yields. The experiments are still included in Table  because their results are used to understand differences in chemical composition.

During the background collection period of ∼ 1 h for each experiment, the standard deviation of the benzyl alcohol mixing ratio, along with the uncertainty in the calibration, was used to estimate the uncertainty of the initial benzyl alcohol mixing ratio (see Table ). This combined standard deviation was also considered as the uncertainty in the measurement of the time-resolved gas-phase mixing ratio throughout the experiment. The SOA yield is determined from the reacted benzyl alcohol, which is the difference between the measured benzyl alcohol concentration at any given time and the initial benzyl alcohol concentration. The variance of the reacted benzyl alcohol is the sum of the variances of the initial and measured benzyl alcohol mixing ratios. The uncertainty reported in Table  is, then, the square root of the reacted benzyl alcohol mixing ratio variance.

The conversion from mixing ratio to mass concentration of reacted benzyl alcohol was performed assuming a constant pressure of 1 atm. Note that the chamber is located three floors from a weather station, which reported an average atmospheric pressure of 0.97 atm in the year 2019 ; thus, 1 atm is a reasonable estimate of the pressure in the experiments.

To measure the particle size distribution, a custom-built SMPS with a 308100 TSI differential mobility analyzer (DMA) and a TSI 3010 butanol condensation particle counter (CPC) was used with a sheath flow rate of 2.64 L min-1, an aerosol flow rate from the chamber of 0.515 L min-1, and a dilution flow of 0.485 L min-1. A full size scan was collected every 5.5 min (for experiments S1–S3 and E1 scans were performed every 6 min), and the voltage was scanned over 4 min from 15 to 9875 V. Data inversion was performed using the method described in . Total number, volume, and surface area concentrations were determined assuming 431 size bins between 22 and 847 nm. When the sample flow was < 0.515 L min-1, an adjustment to the total number concentration was performed to account for the sampled flow. Particles were charged with a 500-microcurie Po-210 source, except for experiments S1–S3 and E1, which used an X-ray source.

When the aerosol size distribution was close to the edges of the measurable range, a logarithmic fit of the distribution tail was performed on the edges of the distribution: diameters of 382 to 600 nm were used to fit particles above 600 nm, and those with diameters 35 to 200 nm were used to fit particles with diameters smaller than 35 nm. Fits of the tail distribution were performed on the upper end of the size distribution for experiment N5, which produced an average of a 3.4 % decrease from the raw measurement in the volume concentration; the lower end of the size distribution for experiment S2, which led to a volume concentration adjustment of < 0.1 %; and on both the upper and lower ends of the size distribution for the nucleation experiment S1, which (for those points after at least 100 min of oxidation) led to a volume concentration difference of < 1 % from that measured in the absence of any adjustment. Particle volume was converted to particle mass with a SOA density of 1.4 g cm-3, consistent with past work on isoprene and on benzyl alcohol .

Uncertainty in the particle size was assumed not to exceed 2 nm, as is typical. For the CPC-associated margin of error, according to approximate Poisson statistics, the uncertainty of the number in each particle size bin was taken as the square root of the number concentration in that bin and that value of uncertainty was propagated into surface area and volume measurements both by bin and, eventually, for the total number concentration. Additionally, an uncertainty in the measured volume concentration due to sample noise was added from the uncertainty of the wall-loss-corrected volume concentrations in the background collection period prior to lights on (see Sect. ).

For experiments U1–U8, there were issues with the particle-volume measurements or with the particle-wall-deposition correction (see Sect. ). While these experiments were used for the analysis of chemical composition, no SOA yields or wall-loss slopes are reported. Additionally, experiments U1, U3–U4, and U7 report approximate initial seed surface area concentrations. There is no initial measured seed surface area concentration for experiment U8.

Aerosol-phase bulk composition was determined using an in situ high-resolution time-of-flight aerosol mass spectrometer (AMS, Aerodyne Research) in the high-sensitivity V mode. Data were analyzed with Igor Pro (version 6.37) and the Squirrel (1.57l) and Pika (1.16l) toolkits. Elemental composition was determined following the improved-ambient method from and . Absolute uncertainties of O:C and H:C ratios are ±28% and ±13%, respectively .

Measurements from the AMS can be utilized to determine the mass fraction of organonitrates (RONO2) in the aerosol phase following the method described by . Both inorganic and organic nitrates fragment to an m/z of 30 (NO+) and an m/z of 46 (NO2+), but the ratio of these two fragments for organonitrates (including those derived from aromatic hydrocarbons) and for ammonium nitrate is quite different, and this difference can be utilized to determine the contribution of organonitrates to the nitrate signal in the AMS . The measured mass ratio of NO/NO2 (called the NOx+ ratio) is used to show the contribution of organonitrates to aerosol mass (see Appendix ). Note that fragments of the form CxHyNz+ are sufficiently scarce that they are neglected: the N:C ratio was never more than 0.026 for the experiments considered here.

For experiments N1–N3 and U1–U6, the chemical composition of particle-phase compounds was further analyzed using offline ultra-high-performance liquid chromatography electrospray ionization quadruple time-of-flight mass spectrometry (UPLC/ESI-QToF-MS) . This method is described in Appendix .

The secondary organic aerosol yield (SOA Y) is given by 1Y=ΔSOAmeasΔBnOHmeas, where ΔBnOHmeas is the reacted mass of benzyl alcohol, that is, the difference between the initial concentration and the measured concentration at a given time. ΔSOAmeas is the difference between the measured and wall-deposition-corrected aerosol mass concentration at a given time and the aerosol concentration prior to the beginning of oxidation. The wall-deposition correction assumes that once a particle deposits on the wall, suspended gas-phase molecules no longer condense onto it; its growth ceases. This corresponds to the technical assumption that ω=0, where ω is a proportionality factor that describes the degree to which vapor condenses onto particles already deposited on the chamber walls compared to those suspended in the bulk of the chamber: if ω=0, once a particle deposits on the chamber wall it is lost to the system and no longer acts as a condensation sink; if ω=1, a particle deposited on the chamber wall acts as a condensation sink identically to that of a suspended particle .

The SOA yield is bounded by the assumptions that ω=0 and ω=1. The extent of difference between these cases is dependent on characteristics of the chamber (e.g., the rate of particle-wall deposition) and of the chemical system (e.g., the amount of kinetic vs. equilibrium particle growth that occurs) . Appendix  describes the calculation of ΔSOAmeas,ω=1 and the corresponding assumptions.

Table  shows the SOA yields calculated with uncertainties for the ω=0 and the ω=1 assumption. The SOA yield calculation with both ω=0 and ω=1 is shown for experiment R1 in Fig. . Since the difference between the SOA yield calculated with ω=1 and with ω=0 is dependent on the amount of organic aerosol that deposits onto the chamber walls, experiments with a higher initial aerosol concentration or that simply last for a longer period tend to have a greater disparity between SOA yields calculated with the ω=0 assumption and those calculated with the ω=1 assumption. Even so, for all the experiments considered here, the ω=1 calculated SOA yield is within the uncertainty of the SOA yield found assuming that ω=0. Furthermore, optimization of chamber parameters indicates that the ω=0 case is closer to reality than the ω=1 case (see Table  and the discussion in Appendix ).

The chamber walls have, primarily, two effects on the SOA yield results: (1) particles with organic mass on them may deposit on the chamber walls and not be detected (this is called particle-wall deposition) or (2) low-volatility compounds that, in the atmosphere, would condense onto suspended particles and form secondary organic aerosol mass instead deposit directly onto the chamber walls (this is called vapor-wall deposition).

Since vapor-wall deposition can involve loss to the wall of the oxidation products and not just the precursor compound, it is difficult to directly correct for the effect of vapor-wall deposition on the observed SOA yield. This is because, often, as is the case here, not all the oxidation products are fully measured and characterized. Instead, one can minimize its effect by increasing the presence of the suspended aerosol surface area concentration so that the suspended aerosol outcompetes the chamber wall as a condensation sink. To do so, however, increases the effect of particle-wall deposition because, as there are more particles in the chamber, a greater fraction will generally deposit onto the chamber walls (due to a nonlinear decay) .

Noting that one must always account for particle-wall deposition, since even a nucleation experiment will produce particles that may deposit on the chamber walls while one is attempting to measure them, we take this approach of correcting for particle-wall deposition and operating our experiments in a regime that minimizes the effect of vapor-wall deposition.

The SOA yield is defined as the ratio of the mass of aerosol formed to the mass of precursor reacted (see Eq. ). One may overestimate the yield by underestimating the amount of benzyl alcohol reacted or by overestimating the amount of aerosol formed. If the particle-wall-deposition adjustment overcorrects the aerosol formed, it would seem as if a higher yield exists than that in actuality. Table  shows the SOA yield that would be calculated assuming that no particles were lost to the chamber walls during the experiment: this is simply the difference between the measured aerosol mass at the end of the experiment and that at the beginning, divided by the total reacted benzyl alcohol mass. Except for experiment R3 and L1, which ran for 12 and 17 h, respectively, the raw particle volumes at the end of the experiments were >80 % of the wall-deposition-corrected volumes. So, even if there are errors in the particle-wall-deposition correction, the SOA yields will still be quite large. For experiment R1, the assumed uncertainty that comes from particle-wall deposition is ∼8%. This dominates, for ΔSOAmeas, the random and counting error. The total uncertainty in ΔSOAmeas for experiment R1 is, including the uncertainty in the aerosol density, the wall deposition, and the random error, ∼9%.

Most of the reported uncertainty in the SOA yield comes not from the wall-deposition correction but from the uncertainty in the benzyl alcohol concentration. For experiment R1, the random error in the benzyl alcohol signal, measured during the background collection period, was 15 %. Combined with the uncertainty of the calibration (6 %), this was a 16 % uncertainty. This same error was applied to the concentration of benzyl alcohol measured at the end of the experiment. Since Δ[BnOH]meas=[BnOH]0-[BnOH]t=end, the uncertainty of Δ[BnOH]meas is 19.5 %.

With the 9 % and 19.5 % uncertainties in ΔSOAmeas and Δ[BnOH]meas, respectively, we get a 21 % uncertainty in the final calculated SOA yield. Most of this comes from the precursor concentration.

Uncertainty from vapor-wall deposition is not included in the calculated error, but any vapor-wall deposition would only decrease the fraction of organic aerosol observed. That is, the true ΔSOAmeas would be larger than the calculated ΔSOAmeas. If experiments were not run at a sufficiently large aerosol surface area concentration to neglect the loss of gas-phase products to the chamber walls, the true SOA yield will only be larger than what is reported here.

The uptake and growth of aerosol can occur either through adsorption or absorption of oxidation products. Generally, we think of secondary organic aerosol growth as governed by absorption, though adsorption is also possible, especially at the large surface area concentrations used in this study to reduce the effect of vapor-wall deposition. To estimate the relative effects of these two processes, we use the gas–particle partitioning coefficient given by 2Kp=1pL0NsAtspRTeΔQ/RT+fomRTMWomγ, where the first term comes from adsorption and the second from absorption. The absorbent vapor pressure, pL0 is in units of atm. If we assume that the molecular weight of the organic material MWom=188 g mol-1=1.88×108 µg mol-1, which is the molecular weight of the major low-volatility oxidation product of benzyl alcohol calculated by ; the activity coefficient of a compound in the organic phase is γ=1; and the temperature is T=291 K (matching that in experiment R1), the absorptive term is ∼(1.3×10-10)fom m3 atm µg-1, where fom is the mass fraction of absorbing organic in the aerosol phase.

The specific surface area of the particulate matter, Atsp, changes little throughout experiment R1. At the beginning of the experiment, when particles are the smallest, Atsp≈0.14 cm2 µg-1. Using Eq. (60) from , the surface concentration of sorption sites on an adsorbing surface is Ns,om≈4.5×10-10 mol cm-2 for the organic phase and Ns,ammsulf≈6.7×10-10 mol cm-2 for ammonium sulfate. Note that the calculation for the organic phase uses ρom=1.4 g cm-3. To get an upper-bound estimate of adsorption, if we take Ns=Ns,ammsulf, the adsorptive term is ∼(2.2×10-12)e1.7ΔQ m3 atm µg-1, where ΔQ is the enthalpy difference between desorption from the particle surface and vaporization of the pure liquid and has units of kcal mol-1.

To determine the relative importance of adsorption and absorption, we need ΔQ and fom. For liquid-like adsorption, ΔQ≈0, but for SOA from polycyclic aromatic hydrocarbons rings and organochlorines, ΔQ≈2–4 kcal mol-1 and ΔQ≈1–2 kcal mol-1, respectively . For experiment R1, fom is 0.1 by 10 min and 0.5 by 1 h. At the end of the experiment, fom=0.8.

Depending on the value of ΔQ, the length into the experiment at which adsorption is insignificant changes. If ΔQ≈0, adsorption will be responsible for <15 % partitioning 10 min into the experiment. If ΔQ≲0.9 kcal mol-1, adsorption will be responsible for <15 % partitioning 1 h into the experiment, and if ΔQ≲1.2 kcal mol-1, adsorption will be responsible for <15 % partitioning at the end of the experiment. Note that, since prior to the commencement of oxidation, no aerosol growth is observed, the seed aerosol neither adsorbs nor absorbs benzyl alcohol.

If absorption dominates gas–particle partitioning, the SOA yield would depend on the amount of organic material in the aerosol phase (ΔSOAmeas, which varies with fom) if equilibrium growth occurs, as is shown in Fig.  . Traditionally, this partitioning is given by 3Y=ΔSOAmeas∑i=1nαiKom,i1+Kom,iΔSOAmeas, where a one-product model has n=1 and a two-product model has n=2 . The stoichiometric fraction of product i in mass units is αi. Kom,i is the absorptive partitioning coefficient for the organic phase for species i, which is Kp,ifom from Eq. () .

The two-product model does not improve from the one-product model (dotted curve in Fig. a) but only creates a very large non-volatile compound (Kom≫1) that is formed in very small quantities (α≪1) and the other compound nearly matches the compound found in the one-product optimization. The one-product optimization gives α=0.97 and Kom=0.009 if all points are equally weighted. If we only include the end points, this gives α=1.05 and Kom=0.005.

At ≳500 µg m-3, the SOA yield flattens out. This indicates that, above this ΔSOAmeas, the partitioning coefficients for the oxidative products are sufficiently large (that is, the products are sufficiently non-volatile), that Y approaches α, the gas-phase stoichiometric fraction in mass units for the oxidation products .

While, usually, the SOA yield is reported as a single number at the end of an experiment, it can also be understood as a function of time since multiple generations of oxidation products usually exist . For example, in the α-pinene system, the SOA yield has been shown to depend on the total hydroxyl radical exposure . Figure  shows, for each experiment, the terminal SOA yield and the bands indicating at which times each of the experiments lie within 10 %, 5 %, and 1 % of the final reported yield. The most atmospherically representative value of α is that to which the experiments converge. For almost all the experiments, the yields appear to have converged sufficiently to justify the reporting of the final yield, though the benzyl alcohol concentration may not yet have all reacted (see Table ); as more reacts, more aerosol is formed but the SOA yield levels out. Experiments R3 and R5, which were run for considerably longer than other experiments, show that the final SOA yield changed little from earlier in oxidation, when the other experiments were terminated.

Note that for experiment L1, also run for considerably longer than the other experiments, the light strength was ∼8% of that in all the other experiments. At this lower oxidation rate, the SOA yield takes much longer to converge but does appear to be a function of ΔSOAmeas (Fig. ). This shows that the convergence time depends on the rate of oxidation. Table  includes the amount of benzyl alcohol that reacted by the end of the experiment and a first-order exponential fit to the decay (kBnOH+OH[OH]): for all except experiment L1, the kBnOH+OH[OH] is similar, indicating a similar decay. Note that since the exact [OH] is not calculated, kBnOH+OH[OH] is reported assuming that [OH] is constant throughout the experiment.

Figure  shows the SOA yield of benzyl alcohol over a range of temperatures, all corresponding to approximately the same initial surface area range (1500–2800 µm2 cm-3) and the same initial NO mixing ratio of ∼80 ppb (see R1–R5 and T1–T4 in Table ). In general, a lower yield of benzyl alcohol exists at higher temperatures; this is expected due to the decreased volatility of oxidation products at lower temperatures and to the increased rapidity of second-generation reactions, which may potentially form high-volatility fragments before the lower volatility first-generation products have time to partition into the particle phase.

At the lowest temperature measured, where one would expect the greatest seed surface area effect (that is, the most competition between the wall and suspended aerosol condensation sinks), we have already determined that we are outside the range of the seed surface area effect (Fig. ). So, one would not expect that the difference in SOA yield is related to competition with the chamber wall.

A higher SOA yield at lower temperatures is also supported in Fig. , which shows how the chemical makeup of the aerosol is different for aerosol formed at different temperatures: the O:C ratio is higher and the H:C ratio is lower on aerosol formed at higher temperatures, meaning that more volatile compounds that might condense at lower temperatures (and have a smaller O:C ratio and a lower H:C ratio) do not condense at the higher temperature (Fig. a and c). Though the difference is slight, there is a trend for a larger NOx+ ratio (Fig. b) and, correspondingly, a larger mass fraction of organonitrates at higher temperatures. The former indicates that the organonitrates may be less volatile than other nitrogen-containing compounds that may condense into the aerosol phase (including, potentially, inorganic ammonium nitrate). The latter suggests that the gas-phase branching may be different. It may be that fewer organonitrates are formed at lower temperatures.

If Eq. () governs the SOA yield curve, then one would expect the curve to asymptotically approach α as ΔSOAmeas increases. The partitioning coefficient Kom is explicitly temperature dependent, but α is temperature dependent only insofar as the oxidation products preferred for formation change with temperature. In Fig. , the flattening out of the SOA yield curves indicates that it is α (and not just Kom) that changes with temperature. This points to a change in chemistry accounting for the difference in SOA yields at different temperatures.

This difference in chemistry accounting for the difference observed in the SOA yield is also supported by observations in the gas phase. The gas-phase concentration of hydroxybenzyl alcohol (HOBnOH) has a molar mass of 124 g mol-1 and is detected at M+19, corresponding to the addition of F- . This signal normalized to the reactant ion signal by the initial benzyl alcohol concentration (expressed in signal normalized to reactant ion signal) for each of the experiments described here is given in Fig. . Note that this is, essentially, the HOBnOH concentration divided by the initial benzyl alcohol concentration. The temporal evolution of HOBnOH for nearly identical experiments is fairly reproducible, as shown in Fig. a. The formation of HOBnOH or the rate at which it reacts away seems to increase slightly at higher temperatures (Fig. d), which also indicates that it is a change in chemical composition that accounts for changes in the SOA yield.

To probe the different chemical pathways that form, the SOA yield dependence on variable NO concentrations was investigated (Fig. ). NO mixing ratios were maintained throughout experiments N1–N6 and U6, leading to an increase in the total NOx in the system. NOx increased by ∼ 60 ppb for experiment N1 and ∼ 100–200 ppb for experiments N2–N6 and U6. Generally, the SOA yield seems to decrease with increased NO concentration.

As shown in Fig. c, there are also larger O:C ratios after ∼ 2 h of oxidation for the lower NO mixing ratios (N1, N2, and N4). Note that experiment N4 appears to behave more similarly to N1–N2 than to N5–N6 and U6; the control on the NO mixing ratio for N4 was much less successful than for the other constant NO experiments (see the error bars in Fig. ). While the [NO] throughout experiment N4 was, on average, 74 ppb, it was only 62 ppb on average during the first 3 h of oxidation (experiment N3 had an average [NO] of 62 ppb during the first 3 h of oxidation).

Figure e shows a difference in the gas-phase chemistry at different NO mixing ratios for the lower constant [NO] experiments (N1–N3), but there appears to be little difference in the HOBnOH concentration for the higher constant [NO] experiments (Fig. f). This also matches the observed change in O:C ratios (indicating a change in chemical composition) between experiments N1, N2, and N3 (but not between N5, U6, and N6).

This change in chemical composition could be in the prevalence of organonitrates. It appears that at the beginning of each experiment, the first secondary organic aerosol formed comprised a significant portion of organonitrates (as much >20 % by mass), as shown in Fig. . While the mass fraction of organonitrates is not reported for the experiments shown in Figs.  and (due to calibration issues), the NOx+ ratio trend is the same as that for the experiments shown in Fig. , where the mass fraction can be reported. Note that one pathway to form organonitrates is by reaction with the nitrate radical; since all our analysis from the AMS is of experiments with the ultraviolet lights on, one does not expect a significant concentration of nitrate radicals . Instead, we expect the organonitrates to have been formed by a RO2⚫+NO reaction; this reaction has a high gas-phase yield for organonitrates for large compounds . Nitroaromatics could also form from the addition of NO2 to a radical intermediate, as has been suggested as the formation mechanism for nitrocatechols from laboratory studies of m-cresol .

Indeed, UPLC analysis found a high prevalence of RNO2 compounds (see Table ), which likely will not lead to the same NOx+ ratios as organonitrates and might contribute NO2+ fragments that could lower the NOx+ ratio. For all experiments with filters collected (N1–N3 and U1–U6), nearly all compounds detected with UPLC analysis were nitroaromatics. This indicates that the low-volatility products that condense into the aerosol phase retain their aromatic rings. Some of the ring-retaining compounds have C7 structures, as does benzyl alcohol. However, several of the compounds detected are C6 structures, indicating the possible loss of the methanol group. In particular, UPLC analysis showed a particularly high concentration of nitrocatechol in the aerosol. The atomic ratios of oxygen-to-carbon atoms (O:C) are quite large: between 0.6 and 1.0, which matches that of very oxygenated rings (Fig. ) but could also match nitrocatechol (O:C of 0.67).

As oxidation continued, more non-nitrogenated organic compounds condensed into the particle phase, decreasing the mass concentration of organonitrates. Simultaneously, the NOx+ ratio decreased, which could have been caused by nitric acid, formed from OH+NO2, partitioning into the aerosol phase and forming nitrate ions. Partitioning of HNO3 into secondary organic aerosol has been observed by . Another possibility is that other compounds, such as organonitrites, might produce NO2+ fragments that lower the NOx+ ratio throughout the experiment.

It is possible, however, that there are non-ring-retaining compounds which condense onto SOA that are simply not detectable by the UPLC. Additionally, the prevalence of nitroaromatics may be a result of the UPLC analysis method that is particularly sensitive to nitroaromatics: the detection of aerosol-phase compounds via the UPLC/MS method is limited to detecting compounds that are water soluble and lie within the detection limits of the instrument. Though filters were stored at low temperatures, on-filter chemistry or hydrolysis in the aqueous phase could occur. This could alter the molecular weight of the original compounds collected in the particle phase .

Nevertheless, the presence of many nitrogen-containing compounds in the particle phase is clear. This is supported by other studies: nitroaromatic hydrocarbons have been observed from daytime oxidation, sourced from anthropogenic sources, and attributed to the particle phase . These compounds have been observed in toluene chamber oxidation studies run at constant 15–20 ppb NOx mixing ratios, for which benzyl alcohol was one of the measured oxidation products .

Perhaps at higher NO concentrations there are more nitroaromatics, and these compounds are more volatile than the nitrogen-free oxidation products (such as the very oxygenated rings). Though the differences in H:C and O:C ratios are slight, the larger O:C ratios – corresponding to the very oxygenated rings – that are seen at lower NO concentrations support the theory that the compounds formed differ (see Fig. ).

Experiment E1, which is similar to experiments R1–R5 except that, prior to the beginning of oxidation, it begins with 71.0 ± 0.8 ppb of NO2 and no NO, shows a much lower SOA yield than that from experiments R1–R5. This suggests that it is the NO that is the relevant reactant that causes initially high SOA formation. This is supported by the significant mass fraction of organonitrates at the beginning of the experiments; organonitrates are formed by RO2⚫ reaction with NO.

Throughout all the experiments, the O:C ratio also first decreases and then increases. If particle growth is kinetically controlled (supported by a modeled αp∼10-2; see Appendix ), the change in O:C ratio throughout the experiment might simply be a result of the greater abundance of higher volatility oxidation products at the beginning of the experiment. Only the lowest volatility products (which are, presumably, compounds with the highest O:C ratios) condense initially, but as higher volatility compounds build up they may eventually partition into the aerosol phase, decreasing the O:C ratio. As lower volatility second- and third-generation compounds are formed, these might then increase the O:C ratio observed. There may also be particle-phase chemical reactions, such as oligomerization , that lead to the change in O:C ratio throughout the experiment, or the observed change could result from a change in the nitrogen-containing compounds in the aerosol phase. Note that, when there is a large contribution of organonitrates to the aerosol, the O:C ratio will be an underestimate .