Chemical characterisation of benzene oxidation products under high and low NOx conditions using chemical ionisation mass spectrometry

Aromatic hydrocarbons are a class of volatile organic compounds associated with anthropogenic activity and 35 make up a significant fraction of urban VOC emissions that contribute to the formation of secondary organic aerosol (SOA). Benzene is one of the most abundant species emitted from vehicles, biomass burning and industry. An iodide time of flight chemical ionisation mass spectrometer (ToF-CIMS) and nitrate ToF-CIMS were deployed https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Benzene is an aromatic volatile organic compound (VOC) commonly used as a vehicular fuel additive (Verma 65 and Des Tombe, 2002) and as a chemical intermediate in the manufacture of a range of products e.g. detergents (Oyoroko and Ogamba, 2017), lubricants (Rodriguez et al., 2018), dyes (Guo et al., 2018) and pesticides (Wang et al., 2014). Whilst current global estimates of benzenoid (molecules containing a benzene ring) emission to the atmosphere by vegetation is of a comparable order to that of anthropogenic activity (~ 5 times lower, Misztal et al. 2015), and background concentrations of benzene are enhanced by increased biomass burning (Archibald et 70 al., 2015), emission of benzene to the urban atmosphere is dominated by vehicle exhausts (Gentner et al., 2012) and solvent evaporation (Hoyt and Raun, 2015). As well as a toxin (Snyder, Kocsis and Drew, 1975) and carcinogen (Ruchirawat et al., 2005(Ruchirawat et al., , 2007Bird et al., 2010), benzene is photochemically active and contributes to the formation of ozone (O3) and secondary organic aerosol (SOA), both of which act to modify the climate and contribute to poor air quality (Henze et al., 2007;Ng et al., 2007). SOA formation from benzene has previously 75 been quantified with a focus on the contribution from smaller mass, ring breaking reaction products such as https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License. epoxides (Glowacki, Wang and Pilling, 2009) and dicarbonyl aldehydes (Johnson et al., 2005). For example the 37% of the glyoxal formed in the LA basin is estimated to derive from aromatic sources (Knote et al., 2014).
These smaller ring breaking products typically make up a larger fraction of SOA mass than the ring retaining products (Borrás and Tortajada-Genaro, 2012) and so have traditionally been the main focus for SOA 80 quantification. Other major, toxic benzene oxidation products such as catechol, nitrophenol and maleic anhydride are also known components of SOA (Borrás and Tortajada-Genaro, 2012).
More recently the autoxidation mechanism has been demonstrated to occur in aromatic systems (Wang et al., 2017, Wang et al., 2020Molteni et al., 2018;Tsiligiannis et al., 2019;Garmash et al., 2020;Mehra et al., 2020) producing highly oxygenated organic molecules (HOM, defined as containing 6 or more oxygen atoms which is 85 a product of the autoxidation mechanism) incorporating up to 11 oxygen atoms (Bianchi et al., 2019). The autoxidation mechanism of intra molecular H shifts from the carbon backbone to the peroxy radical centre forming peroxide groups is consistent with other VOC systems (Rissanen et al., 2014) initiated by a variety of different oxidants (e.g. Mentel et al. 2015;Berndt et al. 2016). The inclusion of an alkyl group to a benzene ring facilitates HOM formation (Wang et al., 2017) as a consequence of the greater degrees of freedom afforded to the molecule.

90
Whilst benzene is not the most reactive of aromatic VOCs, τOH ≈ 9.5 days for benzene vs. τOH ≈ 6-10 hours for xylenes ([OH] = 2.0 × 10 6 molecules cm -3 , Atkinson and Arey, (2007)), it is ubiquitous in the urban environment and is the simplest C6 aromatic ring system to study.
Oxidation of benzene occurs nearly exclusively via hydroxyl radical (OH) addition to form the cyclohexadienyl radical/benzene-OH adduct, which subsequently adds O2 to form the hydroxycyclohexadiene peroxy radical 95 (C6H6-OH-O2) (Volkamer et al., 2002) (Fig. 1). Two subsequent reaction pathways are postulated for this peroxy radical: either elimination of HO2 yields phenol and secondary OH attack must occur again; or an endocyclic dioxygen bridged carbon centred radical intermediate is formed by the addition of the peroxy group to (typically a β-carbon (Glowacki, Wang and Pilling, 2009)). This di-oxygen bridge carbon centred radical may add another O2 and form a peroxy radical (named as BZBIPERO2 in the master chemical mechanism, MCM, Saunders et al. 100 (2003)). However it is not known if autoxidation may continue to form a second oxygen bridged radical, described as type II autoxidation (Molteni et al., 2018, Fig 1, BZBIPERO2-diB), or form a hydroperoxide carbon centred radical, formed through intra molecular hydrogen abstraction (type I autoxidation). At each step, termination of the peroxy radical to hydroxyl, hydroperoxyl, nitrate, peroxy nitrate or reduction to an alkoxy radical is possible, from which further termination via formation of a nitrite, nitro-or acyl group can occur.

105
Typically the presence of NOx alters the product distribution of VOC oxidation and reduces SOA formation (e.g. Stirnweis et al. 2017). With other atmospherically relevant VOC precursors of SOA e.g. isoprene or monoterpenes, high NO conditions can supress SOA formation (Wildt et al., 2014;Sarrafzadeh et al., 2016) as reduction of RO2 to RO ultimately leads to fragmentation of the RO species (Surratt et al., 2006;Nguyen et al., 2015), but it can also form epoxides, aldehydes and hydroperoxides which readily partition to the aerosol phase 110 and contribute to SOA growth (Surratt et al., 2010).
The further reaction of NOx with oxygenated VOCs produces species including nitro organics, nitrates, peroxy nitrates and peroxyacyl nitrates (Atkinson, 2000) that can also condense and contribute to SOA formation. Under conditions where NOx is present in moderate amounts, the HO2:OH ratio is low as HO2 is reduced by NO to form OH. This allows more OH oxidation to occur and less termination of RO2 by HO2 to occur; whereas under low 115 https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License.
NOx conditions VOC consumption is lower, as OH recycling from HO2 relies on the slower HO2 + O3 reaction (Atkinson, 2000). Here low and high NOx are relative terms and are defined by the available VOC. This reduction of the HO2:OH ratio as a function of NO has been observed at Mace Head, Ireland where clean low NOx marine influenced air can be contrasted with polluted NOx containing continental air (Creasey, Heard and Lee, 2002). In that instance, an increase in NO concentration from 0.01 ppb to 1 ppb reduces the HO2:OH ratio from 200:1 to 120 10:1.
The autoxidation mechanism is known to form biogenic HOM, and as a consequence SOA, in ambient rural environments (e.g. Ehn et al. 2014). The mechanism is increasingly a competitive process in urban and suburban environments where NOx concentrations have seen significant reductions in recent years (Praske et al., 2018).
ToF-CIMS is a measurement technique frequently used to probe VOC oxidation due to the ability to detect many low concentration compounds simultaneously in real time (e.g. Chhabra et al. 2015). As a result of the sensitivity 130 of the nitrate ionisation scheme towards HOM, this reagent ion is typically used to study the autoxidation mechanism and HOM formation. However, to achieve carbon mass closure of the system, multiple ionisation schemes are required due to their complimentary yet differing sensitivities towards OVOCs with different oxidation states and functional groups (e.g. Isaacman-VanWertz et al., 2017;Riva et al., 2019).
In this study, the oxidation of benzene by OH under atmospherically relevant high and low urban NOx conditions 135 is investigated in the Jülich plant atmosphere chamber (JPAC) with two time of flight chemical ionisation mass spectrometers (ToF-CIMS) using the iodide and nitrate ionisation schemes. The properties of the oxidation products are compared between the two ionisation schemes, as well as the high and low NOx conditions.

Methods and Experimental
The experiments were performed in a 1450 L borosilicate continuous flow reactor of the Jülich Plant Atmospheric 140 Chamber (JPAC) described elsewhere (Mentel et al., 2009 It is expected that as the NOx concentration decreases, peroxy radical lifetime is much longer than the required time scale for autoxidation hydrogen shifts (Praske et al., 2018) and so autoxidation products can be expected in this low NOx case here (≤ 300 pptv). Where the UVA lamps are active, the NO/NOx ratio is increased, thus increasing the ratio of RO/ROx (RO + RO2). Formic acid calibrations and chamber and instrumental backgrounds were taken before every experiment to assess 185 instrument sensitivity changes. Backgrounds were performed by overflowing the CIMS inlet with N2 and calibrations were performed by flowing 50 sccm N2 over a formic acid permeation tube held at 40 o C. The formic acid sensitivity was 3.15 ± 0.26 Hz ppt -1 (2σ) measured as a 5 minute average normalised to 10 6 Hz iodide. The background was 2.72 ± 0.66 ppt (2σ) measured as a 5 minute average normalised to 10 6 Hz iodide. The limit of detection (3σ) was 100 ppt. Calibration of all detected masses was not possible, so data are presented either as 190 signal (Hz) from the instrument or as a number of species.
It is not possible to identify a compound from its formula alone. Oxidation is initiated with OH addition to the benzene ring, but fragmentation occurs and so the OH moiety may be lost; additionally multiple OH attacks make the counting of oxygen more difficult (Garmash et al., 2020). It is for these reasons that RNOx species are described more generally and any discussion of compound classes is speculative. Data analysis was performed 195 using the Tofware (V 3.1.0, (Stark et al., 2015) ).

Calculation of average carbon oxidation state ( ̅̅̅̅̅ )
Average carbon oxidation state ̅̅̅̅̅ (Kroll et al., 2011) is commonly used to describe the degree of oxidation within a complex oxidation reaction. It is calculated using the oxygen to carbon ratio (O:C), hydrogen to carbon It is more difficult to justify this assumption for gas phase data and as this work explores different N-containing functional groups, this nitrate assumption cannot be used. Instead of assuming all N is in the form of nitrate, a 215 range of OSN (5+, 3+ and 0) are considered and ̅̅̅̅̅ is calculated as a range from a theoretical minimum, where all N is considered nitrate (5+), to a theoretical maximum, where all N is considered cyanate (0). Nitro and nitroso compounds are also considered (3+), but amines and amides are not as they are not expected to be present in the system. As neither the maximum or minimum OSN is a likely or an accurate description of the system, they merely https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License.
represent theoretical limits to calculate ̅̅̅̅̅ ; the true value will lie somewhere in the middle, in all likelihood 220 between the mean and lower limit as most species are expected not to be cyanates. The ̅̅̅̅̅ reported here is the mean average of three scenarios where all N is considered to exist in 0, 3+ or 5+ oxidation states. This methodology reduces the accuracy of the reported ̅̅̅̅̅ but it clearly and accurately presents the ̅̅̅̅̅ range.
In this document ̅̅̅̅̅ are reported as a function of carbon number. The variation in ̅̅̅̅̅ between all CHON of the same carbon number is much greater than the variability in ̅̅̅̅̅ of a single CHON compound as a consequence 225 of varying OSN for that compound.
Where a species has the form RNO1,2 the species is considered to be either a cyanate or hydroxy cyanate and the OSN is set to 0. For RNO3 species the nitrogen is considered to be either a dihydroxy cyanate or a hydroxy RONO or RNO2 compound and so OSN is set to either 0 or 3+. RNO>3 have OSN set to 0, 3+ or 5+. This leads to the modification of the ̅̅̅̅̅ parameterisation:

235
Hierarchical cluster analysis (HCA) is an analytical technique used to simplify datasets containing large numbers of individual observations into groups or clusters defined by their similarity with the aim of increasing interpretability of the dataset. HCA is an agglomerative technique which allows for the successive merging of individuals or clusters to form larger clusters. Due to the complex nature of atmospheric oxidation and the detection of many oxidation products by mass spectrometry, it is common practice to reduce the dimensionality 240 of mass spectrometry datasets in order to better describe bulk properties of the process being studied (Sekimoto et al., 2018;Isokääntä et al., 2019;Koss et al., 2019) with HCA being one such method. HCA is independent of calibration as it relies on relative differences between time series' rather than exact concentrations, making it a suitable technique to apply to mass spectrometry data where authentic standards do not exist and individual calibrations are next to impossible. The final number of clusters is selected based on the distance between them 245 and is determined by the user. Here we use HCA to group the time series' of benzene oxidation products to investigate its utility as a tool for investigating oxidation processes or product properties where the reaction occurs in a continually stirred flow reactor.
In order to use HCA the linkage criterion and distance metric must be selected. The root of the sum of the squares of a pair of observations (time series), A and B, (i.e. the Euclidean distance) is chosen as the distance metric to 250 define the similarity of the time series' at each time step, t.
The Ward linkage criterion was chosen to determine the distance between sets of observations (U, V, which can be single or multimember clusters) as it gave similar or more interpretable results compared with other linkage https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License.
criteria. Here, the sum of the squares of the cluster members (xi where i iterates over the members U, V or their 255 agglomeration) from their cluster means (mU, mV or mU∪V) are calculated. This is performed for the two candidate clusters for agglomerating (U, V) and for the new cluster they would form through agglomeration (U∪V). The difference in the sum of the squares between this potential merged cluster and the initial two candidate clusters is the distance used to assess whether the agglomeration of the candidate clusters is acceptable: When the increase in the sum of the squares (i.e. distances) is minimal after agglomerating, the two candidate clusters, the new cluster is formed. Here, the time series for all CHO and CHON compounds from the iodide ToF-CIMS measurements were selected for clustering. Time series were scaled between 0-1 to remove the effect of their differing magnitudes and focus on the changes in trends. The scaled time series were then smoothed to a 10 minute average to remove noise that may affect the results of the HCA. This treatment is similar to other studies 265 using HCA with mass spectrometric data (e.g. Koss et al., 2019). The analysis was implemented using the cluster.hierarchy module from SciPy (1.2.0) scientific python library using Python 3.6.

Results and Discussion
We detect a range of benzene oxidation products, including many with the same formula as those listed in the MCM, as well as highly oxidised species with both ToF-CIMS instruments. Well resolved peaks at a given unit 270 mass were frequently observed in the Ispectrum. The separation between the Iion cluster with its large negative mass defect and the positive defect (excess) of the high H containing deprotonated species allows good peak separation at the resolution of the ToF-CIMS (Fig 2). Beyond Iand H2O, the largest adduct signals we identified with the I -CIMS were low mass species: formic acid (CH2O2), nitric acid (HNO3), CH2O3, C3H6O3, HONO, C4H4O3 and C3H8O3. The prevalence of these ions is likely do to instrument sensitivity rather than chemistry.

CHON product distributions: comparison between nitrate and iodide ionisation
Forty-two formulae are found to be common in the low NOx experiment between the two instruments representing the overlap of oxidation product signals of which 20 are C6 compounds and the rest have lower carbon numbers.
The time series behaviour of the signals at the start of oxidation is highly variable and is specific to each instrument 280 and ion measured by that instrument as oxidant levels and wall partitioning equilibrates. Typically, the level of highly oxidised species increases significantly from the outset but relaxes to lower levels rapidly before they grow again to reach steady state. This early spiking behaviour has been observed in other studies investigating VOC oxidation using the nitrate ToF-CIMS (e.g. Ehn et al. 2014) and is thought to be a consequence of a latent aerosol sink shifting the equilibrium species from the gas to the aerosol phase. The early spiking also describes the 285 dynamic nature of oxidation concentrations (OH and O3) before the steady state is reached. The iodide CIMS measures a delayed response from wall equilibration processes (Krechmer et al., 2016;Pagonis et al., 2017) making the detection of these rapid changes more challenging. For the range of species measured here, the variability between the two measurements for the same species vary (Fig S1). In general the nitrate observes https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License.
square waveforms whereas the iodide observes saw tooth waveforms. Where the agreement is particularly bad it 290 is most likely that different isomers contribute to the different signals measured on the two instruments.
The nitrate ionisation scheme detects a greater number of higher mass species with a higher oxygen content (Fig   3c, 3e) and higher OS c (Fig 3h, 0.84 vs 0.42), whereas the iodide ionisation scheme detects lower mass compounds with a lower oxygen number and lower OS c , as has been reported elsewhere in different VOC oxidation systems (e.g. Isaacman-VanWertz et al., 2017). The average mass of a nitrate adduct (not including the reagent ion) is

295
~240 g mol -1 with ~30% of that attributable to oxygen, whereas the average mass of an iodide adduct is 140 g mol -1 with ~20% attributable to oxygen (Fig 3g).
The nitrate CIMS is able to detect a greater number of high mass species greater than C6 including a large number of dimers (C12) whereas the iodide scheme did not detect many species greater than C6 in this instance but did measure a greater number of low C molecules compared to the nitrate (Fig 3a). Few species greater than C12 are 300 detected with either reagent ion, suggesting that in this system, their formation was not likely, as both the nitrate and iodide ionisation schemes are capable of detecting >C20 compounds (e.g. Mohr et al., 2019). The iodide scheme detects a maximum number of molecules containing O3 and O4 and very few with more than O7 (Fig 3c).
The nitrate scheme observes a broad range of oxygen numbers peaking between O9 and O11 but up to O18. Iodide detects an average H:C/O:C of ~1.5 whereas nitrate detects H:C/O:C of approximately 1.1 due to higher O:C 305 ratios (Fig 3f).
With both nitrate and iodide ionisation, the maximum variation in oxygen number per molecule is detected for C6 compounds (Fig 4a) although for nitrate this is closely followed by C12 i.e. C6 dimers as well as C5, C9 and C10 compounds. The nitrate CIMS detects approximately half the number of low oxygen content (O2-4) (and N containing) oxidation products compared to the iodide CIMS. Subsequently, OS as a function of carbon number 310 differs between the ionisation schemes at higher carbon numbers (Fig 4b). Below C8 the OS broadly agree however above C8 the iodide observes negative OS whereas nitrate observes positive values. This is likely due to the detection of different isomers, which become more prevalent at these at these higher masses.
The saturation concentrations (C*) of the detected species were calculated using the relation of Mohr et al. (2019).
The C* indicates that there is significant overlap of SVOC and IVOC detection between the two instruments (Fig   315   4c). Whilst the iodide ionisation scheme detected HOM, it did not detect any ELVOC (extremely low volatile compounds) and very little LVOC (low volatile compounds), detecting instead mainly IVOC (intermediate volatile compounds), SVOC (semi volatile compounds) and some VOC (Fig 4c). This is in contrast to the nitrate scheme which detected mainly ELVOC, SVOC and some IVOC, although less than the iodide. This highlights that sampling with both nitrate and iodide schemes captures a large range of gas phase oxidation products, as has 320 been shown in other systems (e.g. Riva et al., 2019).

Iodide CIMS detected CHO and CHON under different NOx conditions
A total of 132 and 195 adduct and deprotonated peaks identified as CHO or CHON in the low and high NOx experiments respectively using the iodide CIMS of which 126 are common between the experiments (Fig 5). The number of detected compounds is much higher in the high NOx case compared with the low NOx case. This is 325 expected as the complexity of the system increases and additional reaction pathways are viable. This is especially true regarding the detection of N containing compounds. As the chamber had a NOx background of ~ 300 ppt, https://doi.org/10.5194/acp-2020-819 Preprint. Discussion started: 17 August 2020 c Author(s) 2020. CC BY 4.0 License.
formation of N-containing compounds still occurred during the low NOx experiment. Twenty four N-containing products were detected during the low NOx experiment compared with 70 detected during the high NOx experiment (20 ppb NOx) of which 23 were common.

330
The majority of species detected are C6 and C4 compounds (Fig 6a) in both cases with O3-4 also most abundant.
The atom distributions (Fig 6 a-d) appear very similar, with the largest difference being the number of detected N containing compounds (Fig 6 d), where more than twice as many are detected in the high NOx case. The average carbon and hydrogen atom number for the detected species is marginally higher for the low NOx case than for the high NOx case (Fig 6e). This observation is further enhanced when weighting the average composition by signal  Fig 6h). The small differences in oxidation product characteristics between the two NOx conditions indicates bulk analysis of all oxidation products is not sensitive enough. In order to better understand the differences between oxidation products from the different NOx cases, the ions are grouped by preference to formation under high or low NOx conditions (defined as the regime in which signal enhancement is greatest) and re-analysed ( Fig S2). Despite In order to better describe these oxidation products, five different groupings of compounds are used to categorise detected oxidation products. Background species refer to those present in the chamber before oxidation and comprise < 0.05 % of total signal. Compounds of formulae that match species described in the benzene oxidation highly oxidised and ring retaining products equilibrate and remain flat for the duration of the experiment (Fig 7).
This is also true of the high NOx experiment, although when NO2 photolysis stops halfway through the oxidation, the growth of signal is diminished and the saw tooth profile less apparent. The MCM and ring breaking groups contain small, oxidised molecules that are near the end of the oxidative chain representing large sink of the carbon in the system and so continually grow. The ring retaining and highly oxidised species reach an equilibrium quickly 365 as they tend to be earlier generation products and so their profiles remain flat.

MCM products
The MCM lists 85 multigenerational oxidation products of benzene including important intermediate and radical species (Jenkin et al., 2003;Bloss et al., 2005). The iodide ToF-CIMS detects 19 and 26 of these oxidation products under low and high NOx conditions. These signals are a mixture of adducts and non-adduct peaks that

Mechanistic investigation
To test the extent to which the iodide CIMS is able to detect highly oxidised molecules that may originate from autoxidative processes, a number of theoretically suggested formulae based on the autoxidation mechanism with propagation and termination steps were devised and then searched for within the spectra. The applied mechanism includes the benzene oxidation scheme from the MCM with two additional autoxidation steps added in. It consists 385 of autoxidative intra molecular hydrogen shifts from the carbon backbone to a peroxy radical group forming a hydroperoxide group, followed by O2 addition to form a new peroxy radical (Fig 1). The peroxy radical groups were terminated to hydroxyl, peroxyl, nitrate or peroxy nitrate groups, or reduced to form the alkoxy equivalent which then terminates to a carbonyl group or a nitro group (isomeric with nitrite). This was repeated for phenol and catechol precursors providing a total of 53 individual oxidation products with 21 unique formula. We observe 390 14 and 21 individual deprotonated and adduct signals in the low and high NOx spectra that correspond to 9 and 11 unique formula respectively. These 9 and 11 unique formula correspond to a maximum of 27 and 33 individual oxidation products (Table S1).
At least one signal (adduct or deprotonated) for all theoretic CHO products are observed in the low NOx case apart from C6H8O9 and C6H8O10. No derived CHON are observed in the low NOx case indicating the incorporation of 395 nitrogen into compounds observed under these conditions occurs at a later point. C6H8O6, C6H8O7, C6H8O8 can only be formed through the autoxidation mechanism from either phenol or catechol precursors. All three are found in the low NOx case and C6H8O7 and C6H8O8 are found in the high NOx case. C6H8O8 is an exclusively second generation autoxidation product that has been observed previously in benzene oxidation studies (Molteni et al., 2018). C6H8O9 and C6H8O10 are not observed in the high NOx case. C6H7NO4, C6H7NO6, C6H7NO7, are the only

400
NOx containing species observed in the high NOx case. The formation of C6H7NO4 has two mechanistic routes, the nitration of the initial hydroxy benzene peroxy radical or the addition of a peroxy nitrate group to the phenol, whereas C6H7NO6, C6H7NO7 have numerous routes to formation with the only non-autoxidation route being the nitration of phenol and catechol respectively. Exclusively second generation high oxygen content CHON species C6H7NO8, C6H7NO9, C6H7NO10, C6H7NO11 are not observed. Figure 9 summarises these ions in the mass spectra.

Ring retention and ring breaking products
In terms of signal, most of the ring retaining products are O1,2,3 of which O1,2 is dominated by the formation of phenol and catechol. Conversely, O3 isn't dominated by any one product, but instead comprises of both CHO and CHON products in both NOx conditions (Fig 10). Although O1-7 species are observed in both cases, higher O 410 numbers of 10, 11 and 12 are only visible in the high NOx case due to the presence of nitrogen groups that can incorporate more oxygen.
For the ring breaking products, the majority of signal can be attributed to the presence of species containing 3 and 4 oxygen atoms, although at lower carbon numbers of 1 and 2 lower oxygen numbers of 1 and 2 become important ( Fig 11). The majority of signal for both cases is due to C1,2,3 compounds. C3 compounds show the greatest signal 415 contribution for the largest range of oxygen numbers. This is especially true in the high NOx case where signal C3 compounds have a greater contribution from O>4. Similarly at higher carbon numbers e.g. 5 and 6, higher oxygen numbers of between 6 and 12 are more readily observed, again where N is incorporated in the high NOx case a greater inclusion of oxygen is also observed. The form of this organic nitrogen is perturbed throughout the experiment by altering the NO/NOx fraction by photolysing NO2.

Effect of NO2 photolysis during oxidation under high NOx conditions
A four cluster solution was chosen from analysis of the dendrogram (Fig S3) and is sufficient to provide enough detail that can be explained, but not so much that interpretation becomes unclear. Further refinement of clusters greater than 4 offers no more insight in time series behaviour due to oxidation behaviour but rather describes signal delay likely due to gas wall partitioning (Krechmer et al., 2016;Pagonis et al., 2017). This behaviour is 425 still observed in the four cluster solution, however other affects pertaining to chemistry rather than partitioning are more visible at this low cluster number. Therefore the four cluster solution is chosen (Fig 12). See Table S2.
for a list of MCM defined products and their assigned clusters.
Cluster 3 and cluster 4 do not display much effect of NO2 photolysis. Cluster 3 is a slow formation cluster that grows as photo-oxidation begins. This is likely to represent semi-volatile material that partitions to walls and 430 equilibrates with the instrument lines and IMR. Cluster 4 is a background cluster that decays when photo-oxidation begins (TUV on) and recovers when photo-oxidation stops. These species make up 2.75 % of the total signal during oxidation and so are thought not to contribute majorly to the reaction. The background cluster 4 has the lowest number of RNOx containing members, and the slow growth, semi-volatile cluster 3 has the greatest number (Fig 13c). Cluster 4 contains the greatest number of species with high carbon numbers with a negative OS . These 435 characteristics are reflected in cluster 3 but to a lesser extent as carbon numbers are smaller and OS less negative.
It is also notable that cluster 1 products remain at elevated levels in the chamber after all photochemistry has stopped. This again may be indicative of more semi volatile material formed. This is anecdotally corroborated by the higher oxygen and carbon numbers of cluster 1 compared to cluster 2 (Fig 13a). Cluster 2 shares time series features that are similar to clusters 1 and 3. It has the same long range response as cluster 3 but the same short 440 range increase from photo-oxidation as cluster 1. This is reflected in Fig. 13b where cluster 2 overlaps with clusters 1 and 3, which themselves are well separated. Both these clusters have lower average carbon numbers and higher, positive OS compared with clusters 3 and 4 (Fig 13b).
Clusters 1 and 2 represent the formation of oxidation products. They increase similarly when photo-oxidation is initiated, however when the NO2/NOx ratio is increased, cluster 1 continues to increase but cluster 2 decreases.

445
This suggests cluster 1 products are independent of NO2/NOx ratio at these high NO2 fractions; the growth curve is independent of the decrease in NO concentration, suggesting this cluster is characterised by more reactions of NO2 and peroxy radicals. Conversely to cluster 1, cluster 2 decreases when the NO2/NOx increases suggesting NO is an important route to formation, either from NO addition products or the increased alkoxy radical fraction RO/RO2, which is supported by cluster 2 products having higher OS and lower carbon numbers than cluster 1 450 products.
Both clusters 1 and 2 contain a similar number of RNO and RNO2 species, but cluster 1 has a greater, odd number of oxygen as RNO3 and RNO5, whereas cluster 2 has a greater even number of oxygen in RNO4 and RNO6. The inclusion of these groupings in one cluster leads to a lack of them in the other; no RNO4 or RNO6 are found in cluster 1 nor is there a large amount of RNO5 or any RNO3 in cluster 2.

455
Ideally, NO and NO2 addition to peroxy precursors (cluster 1) would produce nitrates and peroxy acyl nitrates (PANs) that have an odd number of oxygens (RNO3 and RNO5), and addition to alkoxy precursors (cluster 2) would produce nitro-and nitrite compounds that have an even number of oxygens (RNO2). This is indeed observed, however in reality this distinction is not so clear cut as OH addition vs abstraction would reverse this.
So although these observations are broadly followed, they are not exact resulting in a blurring of RNOx groupings, 460 especially at higher O numbers (Fig 13c). Both clusters 1 and 2 have the same number of RNO7 species despite their time series profiles showing a different dependency on the NO2/NOx ratio. It is likely that these RNO7 are structurally different and the oxygen is incorporated into their structures in different ways. For example cluster 1 RNO7, which are all C5 and C6 compounds, may contain more PAN like compounds as these are formed from peroxy acyl radicals and NO2 addition e.g. to BZEMUCCO3 (MCM), which is the peroxy radical formed from 465 BZEPOXMUC, a first generation ring opening product of benzene, to form BZEMUCPAN (C6H5NO7). In contrast, cluster 2 RNO7 are smaller C3 and C4 compounds, which may require the fragmentation of larger organic precursors enhanced by the presence of NO e.g. the unimolecular decomposition of the alkoxy radical MALANHYO to form HCOCOHCO3 and then HCOCOPAN (C3H3NO7). Both ionisation schemes detect benzene oxidation products including highly oxidised organic molecules. Nitrate CIMS detects many C12 dimers and a greater number of species with high oxygen (O9-O11). This translates to 475 higher OS , especially at higher carbon numbers (C≥8), and lower C* indicating the detection of ELVOC, SVOC and IVOC. In contrast, the iodide CIMS detects no dimers, but many more monomers and ring breaking products (C≤6) than the nitrate scheme, with most common oxygen numbers of (O3-O4). The OS of high carbon species is lower, although at lower carbon numbers (C<6) OS between the two ionisation schemes broadly agree due to the increased likelihood of measuring the same species, rather than isomers. The corresponding C* measured by the and 12 are observed. Within the context of the theoretical mechanistic investigation (section 3.3), iodide ionisation 500 is able to detect 27 and 33 species belonging to potential autoxidation reaction pathways. These detected species include one previously observed, exclusively second generation autoxidation product C6H8O8 and all derived 1st generation autoxidation products. However it is noted that only two of these first generation products steps (C6H8O6, C6H8O7) are formed exclusively through autoxidation whilst the rest have other routes to formation.

Conclusion
Higher oxygen content products C6H8O9 and C6H8O10 are not observed in either case.

505
Clustering the time series in the high NOx experiment into four clusters distinguishes two clusters that contain products formed from photo-oxidation. One of these clusters (cluster 1) is independent of the NO2/NOx ratio whereas a second cluster (cluster 2) decreases, as a result of NO dependent formation, either through addition and/or an increased RO/RO2 fraction. Cluster 2 has higher OS and lower carbon numbers than cluster 1 suggesting it consists of more oxidised and fragmented compounds consistent with an increased RO/RO2. Cluster 510 2 contains RNO4 and RNO6 but no RNO3 and little RNO5 whereas the opposite is true for cluster 1. This somewhat agrees with theoretical RNOx product distributions as NOx addition to alkoxy radicals (cluster 2) is more likely to produce even oxygen content RNOx (through the formation of nitrites and nitro compounds), and odd oxygen RNOx through addition to peroxy radicals (cluster 1, such as nitrates and PANs). For species with larger oxygen content e.g. RNO7 which is detected in both clusters, the carbon number is lower for cluster 2 (C3,4) compared to 515 cluster 1 (C5,6) indicating more fragmentation has occurred, again implying a greater contribution from the alkoxy channel. It is noted that the effect of OH addition rather than H abstraction as the initiation step to the reaction will reverse this pattern, and along with other unimolecular rearrangements, may explain some of the RNOx cluster variability observed.