Volatile organic compounds (VOCs) are ubiquitous in the atmosphere and major precursors for secondary organic aerosol (SOA). SOA represents a dominant fraction of the tropospheric aerosol (Hallquist et al., 2009; Shrivastava et al., 2017; Gentner et al., 2017) and affects climate (Intergovernmental Panel on Climate, 2014) and health (WHO, 2016). Consequently research interest in SOA formation and properties ranges from remote atmospheres (Ehn et al., 2012, 2014; Kristensen et al., 2016) to densely populated and polluted environments (Chan and Yao, 2008; Hu et al., 2015; Guo et al., 2014; Hallquist et al., 2016). Following the study by Ehn et al. (2012), highly oxygenated organic molecules (HOMs) with low volatilities, formed from the oxidation of biogenic volatile organic compounds (BVOCs), have attracted much research interest (Crounse et al., 2013; Ehn et al., 2014; Jokinen et al., 2014, 2015; Mentel et al., 2015; Yan et al., 2016; Berndt et al., 2016; Bianchi et al., 2019). These compounds have been shown to contribute to new particle formation (NPF) and to SOA growth (Ehn et al., 2014; Bianchi et al., 2016; Kirkby et al., 2016; Trostl et al., 2016; McFiggans et al., 2019), making them an important factor in the formation of atmospheric SOA. These oxidation products can be either described as HOM based on their high oxygen number (O > 6) (Bianchi et al., 2019) or as extremely low volatility organic compounds (ELVOCs) based on their volatility (Donahue et al., 2012; Trostl et al., 2016). In this study, we will refer to oxidation products as HOMs, because not all of the measured compounds may fulfill the criteria for ELVOCs (Trostl et al., 2016; Kurtén et al., 2016). Gas-phase autoxidation of alkylperoxy radicals (RO2) has been proposed as the formation mechanism for HOMs (Crounse et al., 2013; Ehn et al., 2014; Jokinen et al., 2014). After the initial reaction of an oxidant with the VOC and subsequent addition of O2 to the alkyl radical (R), the produced RO2 isomerises via intra-molecular H abstraction (H shift). During this process a hydroperoxide group and a new R are formed. Additional O2 addition and H-shift sequences can introduce large amounts of oxygen to the molecule and subsequently lower the vapour pressure. The chemistry of aromatic compounds is somewhat different compared to other VOCs as they can lose their aromaticity during the initial OH attack while retaining their ring structure. Moreover, reaction products are more reactive than the parent compound. The produced RO2 species form an oxygen bridge, a bicyclic and potentially tricyclic alkyl radical (Molteni et al., 2018; Wang et al., 2017), before further oxidation processes open the ring structure. For 1,3,5-trimethylbenzene (TMB), emitted from combustion sources in the urban environment, Molteni et al. (2018) proposed a generalised reaction scheme for HOM formation after OH addition. According to their scheme, first-generation alkylperoxy radicals with the general formula of C9H13O5-11 were formed from the initial OH attack and subsequent H-shift and O2 addition sequences. A postulated second OH attack would result in propagating peroxy radical chains, yielding radicals with the general formula C9H15O7-11.

Generally, the termination reaction of RO2 (with a general m/z=x) with HO2 leads to the formation of hydroperoxides (m/z=x+1), while termination reactions with other RO2 species can lead to the formation of a carbonyl (m/z=x-17), a hydroxy group (m/z=x-15) or dimers (m/z=2x-32) (Mentel et al., 2015; Jokinen et al., 2014; Rissanen et al., 2014). The propagation reaction of RO2 with another RO2 or NO also results in the formation of RO (m/z=x-16). The RO can undergo internal H shift leading to the formation of a hydroxy group and subsequently a new peroxy radical, which can continue the autoxidation sequences. The alkoxy step shift of the observed m/z by 16 leads to an overlap of different termination product sequences. During extensive oxidation the first-generation products are subject to secondary chemistry, increasing the numbers of products. Molteni et al. (2018) found several closed-shell monomer products with the general formula C9H12-16O5-11 from the oxidation of TMB with OH. The formation of dimers with different numbers of H atoms in their study was explained by reactions of two first-generation RO2 radicals (C18H26O5-10): one first- and one second-generation RO2 (C18H28O9-12) or two second-generation RO2 species resulting in the dimer C18H30O11. Although dimers have in general lower O:C ratios than monomers, they are expected to be less volatile due to higher molecular weight and more functional groups, making them candidates to participate in nucleation processes (Kirkby et al., 2016).

Organonitrates (ONs) are formed as soon as sufficient NOx is present in the atmosphere. ONs are highly important for the reactive nitrogen budget wherein the formation of highly functionalised organic nitrates can contribute significantly to secondary organic aerosol (Lee et al., 2016; Bianchi et al., 2017). In this study we refer to compounds that only consist of H, C and O as HOM monomers or HOM dimers and to N-containing compounds as ONs.

NOx influences the oxidation of organics directly by changing oxidant levels (reducing or increasing OH, depending on the NOx regime) and indirectly by influencing RO2 chemistry. In high NOx environments such as urban areas, the reaction of NO with RO2 radicals can compete with the autoxidation mechanism (Reactions R1 and R2) and thus potentially inhibit HOM while favouring some ON formation (Reaction R1 with a yield of up to 0.3). R1RO2+NO→RONO2,R2→RO+NO2. Although ONs may have high O:C ratios, they differ from HOMs as they contain at least one nitrogen atom. While HOM formation from BVOCs has been intensively studied, only few studies have been conducted focusing on the formation of HOMs from aromatic anthropogenic volatile organic compounds (AVOCs) (Molteni et al., 2018; Wang et al., 2017). However, these studies indicate that AVOCs have a strong potential to form HOMs under NOx-free conditions and proposed that they may play a crucial role in NPF and particle growth of SOA in urban areas. Regarding SOA yields, a number of smog chamber studies have been conducted in order to investigate the oxidation of TMB with OH radicals under different NOx and aerosol seed conditions (Paulsen et al., 2005; Rodigast et al., 2017; Wyche et al., 2009; Huang et al., 2015; Sato et al., 2012). They reported that under higher NOx:VOC conditions, SOA yield is reduced compared to medium or lower NOx:VOC conditions. The NOx:VOC ratio, in almost all studies, was <1, apart from one experiment (Wyche et al., 2009), with NOx:VOC=1.9 in which the SOA yield was 0.29, compared to yields of up to 7.47 under lower NOx conditions.

In this study we investigate the oxidation of TMB in a laminar-flow reactor, while different NOx and OH conditions were applied. A nitrate chemical ionisation atmospheric pressure interface time-of-flight mass spectrometer (CI-APi-TOF-MS) (Junninen et al., 2010; Jokinen et al., 2012) was used to monitor the oxidation product distribution. We show the formation of HOMs and nitrate-containing compounds with and without NOx added to the reaction system. Possible mechanisms leading to the formation of ONs and suppression of particle formation are discussed.

The measured HOMs were generated using the laminar-flow Gothenburg potential aerosol mass reactor (Go:PAM), initially described by Watne et al. (2018). The Go:PAM is a 100 cm long and 9.6 cm wide quartz glass cylinder which is irradiated over a length of 84 cm by two 30 W Phillips TUV lamps (254 nm); a schematic is shown in Fig. 1. The OH radicals are produced inside Go:PAM by photolysing O3 in the presence of water vapour. The O3 is generated outside Go:PAM by photolysing pure O2 (UVP Pen-Ray^® mercury lamps, 185 nm) and distributed in 3 L min-1 particle-free and humidified air (Milli-Q) over the reactor cross section. The VOC was introduced through a gravimetrically characterised diffusion source (see Fig. S2) centrally at the top of the reactor with a flow of 8 L min-1 while NO was introduced via a NO gas cylinder. Flows were adjusted for a median residence time of 34 s in Go:PAM. A funnel-shaped device is subsampling the central part of the laminar flow to minimise wall effects on the sample. A condensation particle counter (CPC, 3775 TSI) was used to measure the number particle concentration in the sample flow. O3 was monitored by a model 202 monitor (2B Technologies), relative humidity by a Vaisala HMP60 probe and NOx by a model 42i monitor (Thermo Scientific) over the course of the experiments. The OH exposure, over the residence time in the reactor, for NOx-free conditions without added TMB, was measured using SO2 titration (Teledyne T100) as described by Kang et al. (2007). Gas-phase oxidation products were measured with an atmospheric pressure interface high-resolution time-of-flight mass spectrometer (APi-TOF-MS, Aerodyne Research Inc. and Tofwerk AG) (Junninen et al., 2010; Jokinen et al., 2012) in connection with a A70 CI-inlet (Airmodus Ltd) (Eisele and Tanner, 1993). The CI inlet is a laminar-flow inlet operated with a sheath flow of 20 L min-1 containing NO3- ions that are generated by ionising HNO3 using an 241Am foil upstream in the inlet design. The sample stream from Go:PAM is introduced in the centre of the sheath flow at a rate of 8 L min-1. The NO3- ions are electrostatically pushed into the sample flow and form stable adducts with sample molecules as described by Ehn et al. (2012). The reaction time of oxidation products and NO3- is a few hundred milliseconds before being subsampled into the TOF-MS at 0.8 L min-1 by a critical orifice. Differential pumping decreased the pressure from 103 mbar in the CI source to 10-6 mbar in the TOF extraction region, where HOMs are detected as negatively charged clusters with NO3-.

A kinetic box model was used to simulate the chemistry in the Go:PAM reactor. The core of the model was first described by Watne et al. (2018). The model consists of 32 species and 68 reactions now including TMB chemistry partly from the MCM v3.3.1 (Jenkin et al., 2003) as well as proposed mechanisms and rate coefficients for NO2 chemistry (Atkinson et al., 1992; Finlayson-Pitts and Pitts, 2000) and highly oxygenated compounds (Ehn et al., 2014; Berndt et al., 2018; Zhao et al., 2018) (see Table S2). The photon flux used in the simulations was tuned to match measured decay of O3, and an OH sink was added to match the observed OH exposure in the background experiment, i.e. without the addition of TMB. The model was run for all experiments with and without NOx. Primarily, the modelled output on OH exposure for each experiments was used to interpret the results and for calculating the consumed TMB. However, generally the modelling output was also used to understand the effects of changing experimental conditions on monomer, dimer and organonitrate (ON) production. The experiments without NOx were named 1–4, denoting the increase in OH exposure, and experiments with NOx were denoted according to their NOx/ΔTMB ratio and high (H) or low (L) OH exposure as seen in Table 1.

Table 1 summarises eight experiments where TMB has been oxidised by various amounts of OH using the Go:PAM unit. Generally, a high OH production, induced either by increased light exposure (two lamps) or elevated ozone concentration, resulted in new particles (e.g. experiments 3 and 4), while addition of NOx reduced or suppressed the particle formation. The results from the kinetic model show that the amount of reacted TMB ranges from 5 to 30 ppb, depending on OH exposure (Fig. S1 in the Supplement). An overview of the oxidation product distribution measured with the CI-APi-TOF-MS for different conditions is shown in Figs. 3 and 4. The compounds were detected as nitrate clusters at m/z = masscompound+62. The spectra in Figs. 3 and 4 show significant ion signals from oxygenated hydrocarbons retaining the nine carbons from the original TMB with either even H numbers (closed shell) or odd H numbers (open shell) with a limited number of products from fragmentation, i.e. ions with less than nine carbon atoms. C9 compounds with an O/C ratio of 6/9 or higher were classified as HOM monomers with the general formula C9H12-16O6-11 in the mass m/z range 280–360. Oxygenated hydrocarbons found in the m/z range 460–560 containing 18C were classified as dimers with chemical formulas C18H24-30O10-16. The monomer with the highest intensity detected was C9H14O7, m/z 296. The highest intensities among the dimers were C18H26O10, C18H28O11 and C18H28O12 at m/z 464, m/z 482 and m/z 498, respectively. In addition to HOM monomers and dimers, nitrogen-containing compounds were found as C9 compounds with one or two N or C18 compounds with one N. The nitrogen-containing compounds were of the general formulas C9H12-18NO6-13, C9H12-18N2O8-15 and C18H18-24NO6-10. The dominating ONs were C9H13NO8 at m/z 325, C9H15NO10 at m/z 359 and C9H14N2O10 at m/z 372. In the experiments where NOx was added, the formation of ON compounds was increasing with the NOx concentration. In parallel, the levels of HOM monomers and dimers were reduced with NOx concentration, where dimers were affected stronger than monomers. Even if the fragmentation products were limited some fragmentation leading to less than nine carbons could be observed. The most prominent fragments were assigned molecular formulas C4H7NO7 at m/z 243, C4H6O12 at m/z 246, C5H6O12 at m/z 258 and C6H9NO7 at m/z 269. Some compounds with C numbers of 15 and 17 were detected in the m/z range 270–560, but their contribution to the total signal was negligible (Fig. 3).

The relative contributions of the different compound classes to the total assigned signal are shown in Fig. 2. It is apparent that HOM monomers and dimers dominate the experiments with low NOx. The contribution of the monomers to the total oxidation product signal ranges from 20.7 %–42.1 % and dimers make 6.8 %–43.3 % of the total signal, depending on experimental conditions. Dimer contributions are highest at high OH exposure (experiments 3 and 4 with estimated OH exposures of 3.8×1010 and 2.1×1011 molec s cm-3, respectively) and decrease with increasing NOx. Nitrated compounds dominated the spectra with contributions of up to ∼75 % in the experiments with the highest amount of NOx. Surprisingly, some nitrated compounds were also found in the experiments without added NOx, which may stem from background NO contamination (∼3–5 ppb).

Recently, Molteni et al. (2018) assigned 17 compounds making up 80 % of the total detected signal for HOM oxidation products from the reaction of TMB with OH. Their compound with the highest fraction of the signal (24.2 %) was the dimer C18H26O8. This compound was not detected in our study. We did not detected deprotonated compounds in the mass m/z range 270–560, HOM monomers with 17 H atoms, nor compounds with O/C<0.55 which were found by Molteni et al. (2018). However, we did find 10 of the previously reported 17 monomers and dimers in our spectra. The oxidation product distributions in our experiments are in general term more diverse, i.e. we found more compounds with smaller yields compared to Molteni et al. (2018). In our experiments the highest 20 compounds together explain 46 %–63 % of the total signal with the individual highest oxidation products contributing only between 4.4 % and 16 %. These differences can be the result of different experimental conditions and setup. In our study the residence time is almost double compared to Molteni et al. (2018), leading to the formation of more oxidised compounds, especially more oxidised dimers, which have been reported in this study. In addition we produce OH radicals through irradiation at 254 nm in the full length of the flow reactor, enhancing the effects of secondary chemistry. Despite these differences there is a general agreement on the conclusions for the NOx-free conditions with the Molteni et al. (2018) study where one rapidly forms HOMs of very low volatility, which can initiate NPF.

The amount of TMB reacted after 34 s in the experiments ranges from 30 ppb (almost all) in experiment 4 to 5 ppb in experiments 1 and NOx9 (compare Table 1 and Fig. S1). The signal intensities of the HOM monomers are increasing with increasing OH exposure. Dimer compounds also increase with increasing OH and reach their highest levels in experiment 3, resulting in the highest ratio of dimer to monomer (see Fig. 2). Faster conversion of TMB will result in higher initial RO2 levels enabling faster RO2 + RO2 (self-)reaction. The concentration profile of RO2 in experiments 1 and 2 is lower and more evenly spread out over the length of the Go:PAM (Fig. S1), and the influence of the RO2 + RO2 reaction will be less in these experiments compared to experiments 3 and 4. Enhanced OH exposure does not only affect the monomer/dimer ratio but also the total amount of compounds measured. Observed compounds were 10 times higher in the high OH exposure experiments 3 and 4 compared to experiments 1 and 2. This was also valid for the experiments with added NOx (NOx1 and NOx3H vs. NOx3L and NOx9). Under high OH exposure (experiment 4) the major HOM monomer is C9H14O7 with a contribution of 5.5 % to the total explained signal, followed by C9H16O9 and C9H14O8, contributing 4.6 % and 4.5 %, respectively.

At the lowest OH exposure (experiment 1) the major signals comprise the monomer C9H12O10 and the dimers C18H26O10 and C18H28O11, contributing 4.4 %, 4.0 % and 3.4 % to the total, respectively. These dimer compounds, C18H26O10 and C18H28O11, are the highest signals in experiments 2 and 3. Two open-shell species were found among the larger signals in experiments 1 and 2 (see Fig. 3): C9H15O8 and C9H15O7.

The number of H atoms is a characteristic for HOM monomers and dimers. An overview of different oxidation product generations is given in Table 2. We observe HOM monomers with 12–16 H atoms and all compounds with an even number of H are closed-shell products. Compounds with an uneven number of H atom changes. are open-shell molecules (radicals). Following the proposed termination scheme by Mentel et al. (2015), compounds with 12 H atoms can be identified as first-generation monomers (terminated from C9H13Ox radicals, Reaction R3) and compounds with 16 H atoms as second-generation monomers (terminated from C9H15Ox radicals, Reaction R4). C9H14Ox can be either first- or second-generation products and originate from either C9H13Ox or C9H15Ox (Reactions R5 and R6). R3C9H13Ox+RO2→C9H12Ox-1+ROH+O2,R4C9H15Ox+RO2→C9H16Ox-1+R′CHO+O2,R5C9H13Ox+1+RO2→C9H14Ox+R′CHO+O2,R6C9H15Ox+1+RO2→C9H14Ox+ROH+O2. The RO2 radicals, C9H13,15Ox, can form dimers likely via the reaction R7HOM-RO2+HOM-RO2→HOM-DIMER+O2. Such a reaction is possible between any HOM-RO2 first- or second-generation (and other C9 peroxy radicals with sufficient high abundance). Two first-generation HOM-RO2 species will result in C18H26Ox dimers (Reaction R8), while one first- and one second-generation HOM-RO2 produce a C18H28Ox (Reaction R9) and dimerisation of two second-generation HOM-RO2 species will result in dimers of the formula C18H30Ox (Reaction R10). R8C9H13Ox+C9H13Ox→C18H26Ox-2+O2,R9C9H13Ox+C9H15Ox→C18H28Ox-2+O2,R10C9H15Ox+C9H15Ox→C18H30Ox-2+O2. A closer examination of the contribution of different HOM generations in Table 2 shows that first-generation monomers with 12 H atoms are showing the highest contribution in experiments with low OH exposure (experiments 1, 2), while monomers with 14 H atoms gain importance with higher OH exposure. The second-generation monomers with 16 H atoms dominate in experiments with the highest OH exposure (experiments 3, 4). The dimer population with 28 H atoms has a larger fraction of the total signal at higher OH exposures compared to dimers with 26 H atoms. Dimer population with 30 H atoms is generally lower than other dimers but has the highest fraction in experiment 4. The overall dimer fraction of up to 43 % in this study (Fig. 2) is similar to the dimer fraction of 40 % reported by Molteni et al. (2018). However, the relative contributions of monomer and dimer generations differ. We find higher contributions of H12 monomers (up to 11 %) and a higher contribution of H28 dimer (up to 12 %) under our experimental conditions. The contribution of H14 monomers and H26 dimers is significantly less compared to that reported in Molteni et al. (2018).

Increasing OH exposure promotes second OH attacks on oxidation products leading to the observed reduction of first-generation products (C9H12Ox) as well as an increase in second-generation products (C9H14Ox and C9H16Ox). The C9H14Ox products have mainly characteristics of second-generation products, as their contribution is enhanced in the experiments with higher OH exposures (Table S1 in the Supplement), in which there is an enhanced possibility for secondary chemistry initiated by reaction of OH with the first-generation products. The increased oxidation degree can also explain the formation of dimers (C18H28Ox) from first- and second-generation RO2 (C9H13Ox and C9H15Ox) at higher OH exposure and the increase in second-generation dimers. Open-shell species are observed as first-generation RO2 (C9H13Ox) and have a higher contribution at lower OH exposures. Second-generation RO2 species (C9H15Ox) have the highest contribution in experiment 3. At the highest OH exposures in experiment 4, the contribution of C9H15Ox radicals, one of the top-10 contributors to the signal (Table S1), is reduced, while the contribution of the second-generation products (C9H14Ox and C9H16Ox) and dimers increased.

In this study the contribution of C18H28Ox shows that both first- and second-generation HOM-RO2 species were dimerising. The kinetics of dimer formation, if produced from RO2 self-reaction, depends on the square of [HOM-RO2] and their relative importance will increase with the RO2 concentration. Increased local RO2 concentrations (in the first part of the Go:PAM) would explain the increase of dimers with increasing OH exposure.

Substantial particle formation was observed in experiment 4, under the highest OH exposure. Although the amount of reacted TMB in experiments 3 and 4 is similar (26 and 30 ppb, respectively), significant particle formation was not observed under the conditions of experiment 3. The rate at which new particle formation (nucleation) occurs is related to the chemical composition and concentration of the nucleating species (McGraw and Zhang, 2008). After reaching the critical nucleus, particle growth becomes spontaneous in the presence of condensable vapour. Apparently, the local concentration of nucleating species or condensable vapour was not high enough in experiment 3 to yield large numbers of particles, compared to experiment 4. According to recent studies (Ehn et al., 2014; Trostl et al., 2016; Mohr et al., 2017; McFiggans et al., 2019) dimers play an important role in new particle formation. Mohr et al. (2017) found decreased levels of gas-phase dimers in ambient air during NPF events, which is in line with our observations of lower dimer levels in the presence of particles in experiment 4, compared to experiment 3. A large enough concentration of low-volatility dimers obviously helps form critical nuclei that then grow by condensation. Note that the newly formed particles will provide an additional sink for dimers and thus reduce their presence as observable gas-phase products at the end of the flow reactor.

We have measured the formation of HOM monomers and dimers from OH-initiated oxidation of TMB. The experiments with highest OH exposures led to particle formation when NOx was not added. With increasing OH exposure and increased likelihood of a second OH attack, we observe a higher contribution from second-generation oxidation products and dimers in general. The latter is attributed to the increased RO2 concentrations from the increased and fast TMB consumption by OH. The observed products in this study match what would be expected as termination products from previously proposed reaction mechanisms for HOM formation.

The addition of NOx to simulate urban conditions leads to the formation of ON in addition to HOM and a reduction in particle formation potential. We observe that the formation of ON is increasing with increasing NOx/ΔTMB ratio, mostly at the expense of dimers. The presence of ONs, formed at the expense of dimers, can explain the decreased tendency for particle formation. We therefore suggest that the reaction of HOM-RO2 + NO competes with the HOM-RO2 self-reaction, yielding primarily a reduction in dimer formation, which is responsible for the reduction in particle formation. The experimental design using the Go:PAM with concentrations of HOx (and ROx) higher than ambient would attenuate the influence of added NOx. This further emphasises the implication of our findings that most likely the NOx effect would be even more important in the urban atmosphere.

According to studies by Molteni et al. (2018) and Wang et al. (2017), HOM formation from AVOCs was observed and consequently AVOC-HOM were suggested as potential contributors to observed NPF in urban atmospheres. In our study, under NOx-free conditions, we found several previously identified HOMs even if we did not fully agree with the identity and relative importance of all HOMs. However, the oxidation in polluted environments will happen under elevated NOx levels and, as has been shown here, this can lead to the formation of ONs instead of HOMs and subsequently a reduction in NPF potential. We conclude that for interpretation of NPF from aromatics in urban areas, care should be taken and the OH exposure, NOx levels and RO2 concentrations need to be considered in detail since they will largely determine if the HOM-RO2 + NO can compete with reactions yielding HOM, and especially HOM dimers.