A comparative and experimental study of the reactivity with nitrate radical of two terpenes: α-terpinene and γ-terpinene

. Biogenic volatile organic compounds (BVOCs) are intensely emitted by forests and crops into the atmosphere. During the night, they react very rapidly with the nitrate radical (NO 3 ), leading to the formation of a variety of functionalized products including organic nitrates and to large amounts of secondary organic aerosols (SOA). Organic nitrates (ONs) have been shown to play a key role in the transport of reactive nitrogen and consequently in the ozone budget, but also to be important components of the total organic aerosol while SOA 15 are known to play a direct and indirect role on the climate. However, the reactivity of BVOCs with NO 3 remains poorly studied. The aim of this work is to provide new kinetic and mechanistic data for two monoterpenes (C 10 H 16 ), α- and γ-terpinene, through experiments in simulation chambers. These two compounds, which have very similar chemical structures, have been chosen in order to fill the lack of experimental data but also to highlight the influence of the chemical structure on the reactivity. 20 Rate constants have been measured using both relative and absolute methods. They were found to be (1.2 ± 0.5)  10 -10 and (2.9 ± 1.1)  10 -11 cm 3 molecule -1 s -1 for α- and γ-terpinene respectively. Mechanistic studies have also been conducted in order to identify and quantify the main reaction products. Total organic nitrate and SOA yields have been determined. While organic nitrate formation yields appear to be similar, SOA yields exhibit large differences with γ-terpinene being a much more efficient precursor of aerosols. In order to provide 25 explanations for this difference, chemical analysis of the gas phase products were performed at the molecular scale. Detected products allowed proposing chemical mechanisms and providing explanations through peroxy and alkoxy reaction pathways. coming from RO 2 + RO 2 pathway and (ii) a primary trifunctional nitrate formed by the decomposition of alkoxy radical and (iii) detection of secondary products, all presenting very low vapor pressures. On the other hand, for α-terpinene (i) no hydroxynitrate and (ii) no secondary products with low vapor pressure were detected in the oxidation products. The absence of hydroxynitrate may be explained by the fact that, due to relocation of the free electron by 585 mesomeric effect for alkyl radicals, the most stable radicals are tertiary and cannot undergo H-shift to produce the hydroxynitrate. Hence, two aspects of the mechanism appear to be critical for the SOA formation: i) the peroxy radical reaction pathways: when the carbon that bears the radical group, has a hydrogen available, the reaction RO 2 + RO 2  ROH + R(O) can occur, leading here to the formation of an hydroxynitrate. Due to hydrogen bonds, this product has very low vapor pressure. This has already been reported by Ng et al. 2008 who 590 compared isoprene SOA formation for reactions RO 2 + RO 2 et RO 2 + NO 3 . The study reported higher formation of SOA for the pathway RO 2 + RO 2 due to the formation of the specific hydroxynitrate and its secondary reaction. In our study, this product was detected for γ-terpinene, but not for α-terpinene for which we expect that most stable radicals are tertiary ones and cannot undergo this reaction pathway. ii) the alkoxy radical reaction pathways: several decomposition pathways can occur, leading to products which have very different volatilities.

at very low concentration (ppt level), these two features being mandatory for kinetic study of fast reactions. For γ-terpinene, a relative rate determination is also performed. In addition, mechanistic studies have been performed for both compounds by providing total organic nitrate and SOA yields and identification of individual gas phase products. It allows proposing reaction mechanisms for the two compounds. For γ-terpinene, these results are 80 compared to those obtained by the previous study. Finally, differences observed for -and γ-terpinene in kinetic and mechanistic data are discussed in regards to the chemical structures of the two compounds.

85
-and γ-terpinene were purchased from Sigma-Aldrich Co. at respectively 85% and 97% of purity and were used without any additional purification. NO 3 radicals were generated in situ, by thermal dissociation of N 2 O 5 (Eq. (1)), previously synthesized in a vacuum line by the reaction between O 3 and NO 2 (Eq. (2) and Eq. (3)) adapted from (Atkinson et al., 1984a;Schott and Davidson, 1958). The detailed protocol is presented in Picquet-Varrault et al., 2009.

95
Experiments were conducted in two different simulation chambers: CSA and CESAM chambers. The CSA chamber was used for kinetic experiments. It is made of a 6 meters long Pyrex reactor having a volume of 977 L (Doussin et al., 1997) and being equipped with a homogenization system that allows a mixing time below one minute. This chamber is dedicated to gas phase studies and is hence equipped with several analytical devices for gas phase monitoring. An in situ multiple reflection optical system coupled to a FTIR (Bruker Vertex 80) 100 spectrometer allows monitoring organic species in the chamber. Infrared spectra were recorded with a resolution of 0.5 cm -1 , an optical path length of 204 m and a spectral range of 700-4000 cm -1 . Integrated band intensities (in cm molecule -1 , logarithm base e) used to quantify the species are: IBI -terpinene (790-850 cm -1 ) = (1.9 ± 0.2)  10 -18 , IBI γ-terpinene (920-990 cm -1 ) = (1.01 ± 0.02)  10 -18 , IBI NO2 (1530-1680 cm -1 ) = (5.6 ± 0.2)  10 -17 , IBI HNO3 (840-930 cm -1 ) = (2.1 ± 0.2)  10 -17 , IBI N2O5 (1205-1275 cm -1 ) = (1.7 ± 0.1)  10 -17 .This technique also allowed 105 measuring the total organic nitrate concentration by considering that all these species absorb at 1250 cm -1 and 850 cm -1 which correspond to absorptions of -ONO 2 function and by assuming that their absorption cross sections are similar whatever the organic nitrate considered. This hypothesis was verified in our research group with the analysis of standards. In this study, we used IBI ON (900-820 cm -1 ) = (9.5 ± 2.9)  10 -18 cm molecule -1 .
Uncertainties on concentrations of species measured by FTIR include the uncertainty on the IBIs and the 110 uncertainty on the spectra analysis.
For absolute rate determination, NO 3 was monitored from its visible absorption at 662 nm with an in situ IBBCEAS technique which has recently been coupled to the CSA. It is described in details in Fouqueau et al., 2020a. This technique also allows NO 2 monitoring. Having a precise knowledge of the wavelength dependent https://doi.org/10.5194/acp-2020-504 Preprint. Discussion started: 8 June 2020 c Author(s) 2020. CC BY 4.0 License. mirror reflectivity, R(λ), is one of the most critical point of the IBBCEAS technique. It was therefore determined 115 prior to each experiment by introducing a known amount of NO 2 (several hundred ppb) into the chamber. The cross sections used to quantify NO 2 are from (Vandaele et al., 1997) and those used for NO 3 quantification are from Orphal et al., 2003. At 662.1 nm, which corresponds to the NO 3 maximum absorption, the cross section is (2.13 ± 0.06)  10 -17 cm 2 molecule -1 . Thanks to the very high reflectivity of mirrors (99.974 ± 0.002%), the optimum optical path length was found to be 2.5 km leading to a detection limit of 6 ppt for 10 s of integration 120 time. The uncertainty on NO 3 concentrations by IBBCEAS was estimated to be 9%, with a minimum absolute value of 3 ppt (Fouqueau et al., 2020). This uncertainty includes the uncertainties on the reflectivity of the mirrors, the NO 3 absorption cross sections, and the data treatment.
Finally, in order to monitor organic reactants and products, a high resolution PTR-ToF-MS (Kore Series 2e, mass resolution of 4000) was used in two ionization modes, H 3 O + and NO + . When used in standard operational 125 conditions, i.e. with H 3 O + ionization, it has been shown that organic nitrates are subject to important fragmentation (Müller et al. 2012 ;Aoki, Inomata, and Tanimoto 2007). In order to reduce this fragmentation, Duncianu et al. 2017 has adapted the instrument operating procedure for organic nitrate detection by reducing the electric field in the reactor. The same study has also developed a NO + ionization mode, by replacing the ionization gas (water vapor) with dry air and also by applying a reduced electric field. These two modes allow 130 cross checking for the identification of the products. This method was characterized and validated thanks to experiments with various standards of organic nitrates (alkyl nitrates, carbonyl nitrates and hydroxynitrates) allowing to propose ionization patterns for each type of organic nitrates and for both ionization modes. Hence, in NO + ionization mode, organic nitrates were shown to be ionized by charge transfer or by an NO + adduct formation and therefore to be detected at their own mass (M) or at M+30. Hydroxynitrates have been detected at 135 M-1 suggesting that the ionization proceeds mainly by a hydrogen loss.
The second simulation chamber is CESAM chamber (Experimental Chamber of Multiphase Atmospheric Simulation) (Wang et al., 2011). It has been specifically designed to study multiphase processes. In this work, it was used to investigate the mechanisms and the SOA formation. CESAM is a 4177 liters stainless-steel evacuable reactor and is equipped with a fan that allows mixing time of approximatively one minute. Aerosol 140 lifetimes in CESAM being very long (up to 4 days), it is particularly suited for SOA studies. This chamber is equipped with dedicated analytical instruments for gas and aerosol phases. It is coupled with an in situ long path FTIR spectrometer (Bruker Tensor 37) allowing acquiring spectra in the 700-4000 cm -1 spectral range with a resolution of 0.5 cm -1 and an optical path of 174.5 m. It is also equipped with a PTR- To convert size distribution into mass distribution, a particle density of 1.4 g cm -3 was used, estimated to be the density of SOA formed by BVOC+NO 3 reactions (Fry et al. 2014;Draper et al. 2015;Boyd et al. 2015). Some experiments lead to the formation of large particles (diameters > 800 150 nm). In this case, a Palas Welas was used in addition to the SMPS, to measure the particle size distribution. This instrument is based on an optical measurement and allows covering a wider size range (0.2-17 µm).
During experiments, filter sampling was proceeded, allowing the measurement of total ONs yield in the aerosol phase. The filter analysis was performed by FTIR after extraction of particles in liquid phase, following a protocol described by Rindelaub et al., 2015: SOA are extracted in 5 mL of CCl 4 . Organic nitrates were 155 quantified from standard solutions of 2 types of organic nitrates (nitrooxypropanol and tert-butyl nitrate). IBIs for the two standards were 510 and 580 L mol -1 cm -2 respectively. The difference between the two of them is small, so the integrated absorption cross section of organic nitrates in liquid phase was considered to be: IBI ONs (1264-1310 cm -1 ) = 557 ± 110 L mol -1 cm -2 .
The pressure into the chamber was maintained constant by introducing synthetic air in order to compensate the 160 decrease of pressure due to instrument sampling. This leads to a weak dilution of the mixture, here less than 20% for an experiment length of 3 hours. All data presented in the following sections were corrected by dilution and, for the SOA measurements, by the particles wall losses, which have been characterized in CESAM: wall loss rates were determined as a function of the diameter of the particles and interpolated using the (Lai and Nazaroff, 2000) model (friction velocity u*=3.7 cm s -1see Lamkaddam et al., 2017). Due to the material used for the 165 walls (stainless steel), this correction has been found to be small in the present case.

Kinetic study
Kinetic experiments were performed in the CSA chamber at room temperature and atmospheric pressure, in a Then, the second-order kinetic equation is:  (Atkinson, 1986): where In this work, two different reference compounds with well-known rate constants were used: 2,3-dimethyl-2butene and 2-methyl-2-butene. Due to the lack of recommendation by IUPAC for the reaction between these compounds and NO 3 radical, rate constants were calculated as the mean values of the determinations available in 205 the literature (Atkinson, 1988;Atkinson et al., 1984;Atkinson et al., 1984;Benter et al., 1992;Berndt et al., 1998). The uncertainties on reference rate constants were calculated as twice the standard deviation of all the values. The obtained rate constants are: k 2,3-dimethyl-2-butene = (5.5 ± 1.7) × 10 -11 cm 3 molecule -1 s -1 et k 2-methyl-2-butene = (9.6 ± 1.6) × 10 -12 cm 3 molecule -1 s -1 . Finally, the uncertainty on k BVOC was calculated by considering the relative uncertainty corresponding to the statistical error on the linear regression (2) and the error on the Ref.

Mechanistic study
Mechanistic experiments were performed in CESAM chamber at room temperature and atmospheric pressure, in were performed for high concentration experiments (>150 ppb) in order to determine the total organic nitrate 225 concentration in the particle phase (see section 2.2). The sampling on filters started at the end of the oxidation (when the precursor has completely reacted) and lasted between 3 and 6 hours. A charcoal denuder was used to remove organic compounds from gas phase.
Total organic nitrate yields in gas phase were determined by plotting the molecular concentration of organic nitrates as a function of the molecular concentration of the BVOC reacted and by calculating the slope of the 230 straight line. Organic nitrate yields in SOA phase were calculated by measuring the final concentration of organic nitrates and by dividing it by the total reacted BVOC concentration for each experiment. Uncertainty on the yield was calculated as the sum of the relative uncertainties on organic nitrates and BVOC concentrations.
The SOA yield is defined as the ratio of the mass concentration of SOA produced, M 0 , divided by the mass concentration of the BVOC reacted, ∆BVOC. For all experiments, the SOA yield was calculated for each data 235 point, but also once the BVOC has been totally consumed, hence providing both time-dependent and overall SOA yields. These yields were plotted as a function of the organic aerosol mass and fitted by a two-product model described by Odum et al., 1996: Where 1 , 2 and ,1 , ,2 are stoichiometric factors and partitioning coefficients (in m 3 µg -1 ) of the two 240 hypothetical products respectively. Due to the slow injections of N 2 O 5 , SOA equilibrium was expected to be reached at small time steps and time-dependent yields have been used. This also allowed obtaining yields for small aerosol content in the chamber.
As described in section 2.2, oxidation products were detected thanks to PTR-ToF-MS measurements in two ionization modes (H 3 O + and NO + ). However, quantification of these products was not performed due to the lack 245 of standards. Finally, vapor pressures have been estimated using SIMPOL-1 method of Pankow and Asher, 2008 in order to evaluate their contribution to SOA formation via the GECKO-A website (http://geckoa.lisa.upec.fr/generateur_form.php, last access May 12 th 2020). Raoult law (Valorso et al., 2011) has been used to estimate also the fraction of a product i in the condensed phase : where N i,gas and N i,aer are the concentrations (in molecule cm −3 ) of the product i in the gas and particle phases respectively, M aer ̅̅̅̅̅̅ is the mean molecular weight of SOA species (g mol -1 ), C aer is the total SOA mass concentration (µg m −3 ), R is the gas constant (atm m 3 K -1 mol -1 ), T the temperature (K), P i vap is the vapor pressure and γ i is the activity coefficient of product i (in this study, γ i = 1). The mean molecular weight has been estimated to be the mean value for low volatility products which were detected.

255
The calculation of is highly dependent of the estimated vapor pressure. Pankow and Asher, 2008 showed that SIMPOL-1 technique allows predicting with an error between 50 % and 60 % for < 10 -6 atm.
is therefore associated with a high uncertainty and can only be used as an indicator. can also be compared to partitioning coefficients in Eq. (8), using the following equation:

Kinetic results
The list of kinetic experiments and their corresponding experimental conditions are presented in Table 1.
Absolute rate determinations were conducted for and -terpinene while relative rate one was performed only 265 for γ-terpinene. For each method, between three and five experiments were conducted.
Kinetic results obtained for γ-terpinene by relative rate method are presented in Fig regression was applied to all the values (by mixing data sets), leading to k -terpinene = (3.0 ± 1.1) × 10 -11 cm 3 molecule -1 s -1 with 2,3-dimethyl-2-butene and (2.7 ± 0.6) × 10 -11 cm 3 molecule -1 s -1 with 2-methyl-2-butene. It can be concluded that rate constants obtained with the two references are in very good agreement. observed for -terpinene and between FTIR and PTR-ToF-MS data. Good agreement is also observed for NO 2 between FTIR and IBBCEAS data, with an exception for the first experimental point for NO 2 following the injection of N 2 O 5 for which a good mixing is probably not fully achieved yet. These agreements are particularly satisfying considering the fact that the two instruments do not sample in the same volume of the chamber: FTIR provides an integrated measurement of the absorbing species over the whole length of the chamber, IBBCEAS

280
provides an integrated measurement in the width of the chamber and PTR-ToF-MS samples the mixture in one point. This comparison demonstrates that the mixing of the chamber is efficient enough to allow combining data from different instruments for absolute rate determination.
Kinetic plots for absolute kinetic determinations gathering results from all experiments are presented in Fig. 3 for both BVOC. As explained above, the first experimental point following the injection of the reactants was not 285 taken into account. Due to the low integration time used for both measurement techniques, a relatively high noise has been observed for BVOC and NO 3 concentrations. Kinetic results are thus subject to relatively high uncertainties. Rate constants measured by the absolute rate method are (3.0 ± 0.9) × 10 -11 cm 3 molecule -1 s -1 for -terpinene and (1.2 ± 0.3) × 10 -10 cm 3 molecule -1 s -1 for -terpinene.
The absolute values are compared to those obtained by the relative method (for γ-terpinene) and to those already 290 published in the literature in Table 2. It should be noticed that relative rate determinations from Atkinson et al. 1985 andBerndt et al. 1996 have been updated by using the same reference rate constants as the one used for this study (see section 2.3). Uncertainties on the reference rate constants have also been added to the statistical errors provided by the authors.
For γ-terpinene, the three rate constants obtained by this study, i.e. absolute and relative determinations with two 295 reference compounds, are in very good agreement. They are also in good agreement with the relative kinetic study of Atkinson et al. 1985 and with the absolute study of Martinez et al. 1999, even if the second one appears to be 20 % lower. For -terpinene, the absolute rate determination provided by this study has been compared to the previous relative determinations provided by Atkinson et al., 1985 andby Berndt et al., 1996. These two relative rate studies were performed with the same reference compound but using two different experimental 300 setups: a flow reactor (Berndt et al., 1996) and a simulation chamber (Atkinson et al., 1985). When considering the overall uncertainties on these rate constants (approx. 40 %) which include the uncertainty on the reference rate constant, the data seem to be in agreement but when comparing the ratio k BVOC /k ref , it appears that they are in fact, not congruent. No explanation was provided by the authors to explain this disagreement. However, for comparison with absolute rate determination, the overall uncertainty had to be considered. Within uncertainties,

305
the value provided here is in agreement with these two previous. In conclusion, this study allows providing new kinetic data for and -terpinene and confirming the values obtained by the few previous studies. It also provides the first absolute rate determination for α-terpinene.
When comparing the reactivity of the two terpenes, it can be seen that -terpinene is much more reactive than γterpinene (by a factor of approximatively four) and this can easily be explained by the conjugation of the double 310 bonds for -terpinene. Indeed, after the addition of NO 3 radical on one of the double bonds, the alkyl radical formed is stabilized by the delocalization of the single electron. Experimental data have also been compared to rate constants estimated by the structure-activity relationship (SAR) developed by Kerdouci et al., 2014 for the reaction between BVOCs and NO 3 (see Table 2). This SAR has been shown to estimate rate constants within a factor of 2. By taking into account these uncertainties, it can be considered that experimental and estimated rate 315 constants are in good agreement. In particular, the significant increase of the rate constant due to the conjugation of the double bonds is well reproduced by the SAR.

Mechanistic results
Eleven mechanistic experiments were conducted in CESAM chamber for γ-terpinene and eight for α-terpinene, during which the formation of gas phase products and SOA was monitored. Experimental conditions as well as 320 organic nitrate and SOA yields obtained for all experiments are presented in Table 3.

SOA yields 335
Time-dependent and overall SOA yields (Y SOA ) for both compounds have been plotted as a function of the aerosol mass (M 0 ) in Fig. 5. A two products model, defined by Odum (Odum et al., 1996), see Section 2. 4, has been applied for the two curves. For each compound, experimental points obtained from different experiments are in fairly good agreement and show similar tendencies. Yields obtained for γ-terpinene can reach 40 % whereas they are below 2% for α-terpinene. These results demonstrate that γ-terpinene is a very efficient SOA 340 precursor which is not the case for α-terpinene.

350
One can estimate the uncertainties of these parameters by looking at the fit sensitivity. It appears to be very sensitive to (with an associated error estimated to 5 %) and less to (with an error estimated to 50 %).
For -terpinene, our study provides the first determination of SOA yields. For -terpinene, SOA yields have been compared with those provided by Slade et al., 2017. This study used seeds for half of the experiments, and did not observe significantly different yields for experiments conducted with and without seeds. In the study of In conclusion, both compounds appear to have very different behavior towards SOA production: for an ambient aerosol mass loading of 10 µg m -3 , which is typical of biogenic SOA impacted environments (Slade et al. 2017), yields of 10 % and 1 % have been found respectively for γ-terpinene and α-terpinene. For higher aerosol mass https://doi.org/10.5194/acp-2020-504 Preprint. Discussion started: 8 June 2020 c Author(s) 2020. CC BY 4.0 License.

Organic nitrates yields
The formation yields of total organic nitrates in the gas phase (Y ONg ), have been investigated by plotting their concentration as a function of the BVOC consumption for both and -terpinene (see Fig. 6). The linearity of these plots and the fact that the slopes at the origin are different from zero indicate that i) organic nitrates are 375 primary products and ii) if primary organic nitrates are subject to loss processes, e.g. through reaction with NO 3 radicals, they may produce secondary organic nitrates, so that the ON yield is constant. Molar Y ONg obtained for both BVOCs are very similar: 47 ± 10 % for γ-terpinene and 43 ± 10 % for -terpinene. These results confirm that organic nitrates are major products of BVOC+NO 3 reactions. For γ-terpinene, the yield obtained here has been compared to the only value previously reported in the literature by Slade et al. 2017: 11 ± 1%. Despite the 380 fact that experimental conditions are very similar, Y ONg differ by a factor of four. Loss reactions of organic nitrates in particle phase that shift the partitioning equilibrium are invoked by the authors to explain the surprisingly low obtained yield. No influence of RH on organic nitrate yields has been noticed. Another suggested hypothesis advanced by the authors is an epoxidation of hydroxynitrates in particle phase, followed by a loss of the NO 2 group. However, it is expected that these reactions also occur in our experiments and this 385 hypothesis cannot explain the differences observed between the two studies. Concerning -terpinene, our study provides the first organic nitrate yield.
Considering the fact that organic nitrates may partition into/onto aerosols, yields of total organic nitrates in the particle phase (Y ONp ) have also been measured by collecting particles on filters and analyzing them by FTIR.

395
Values are presented in Table 3. For γ-terpinene, molar yields range between 1 and 8 %, presenting a high dispersion and for α-terpinene range between 1 and 3 %. This dispersion can be explained by the fact that i) Y ONp directly depends on the total SOA mass concentration, and consequently on the concentration of reacted BVOC and ii) the concentrations of ON are low (minimum is 5 × 10 -5 mol L -1 ) and thus subject to high variability. The first correlation appears to be clear in Table 3: for high concentration experiments (~400 ppb), yields reach 10 400 %, and for low concentration ones, yields appear to be between 2 and 3 %. The molar Y ONp for γ-terpinene are in  Table 3. For γ-terpinene, the mass yields have been found to range between 3 and 13 %, and for α-terpinene, between 3 and 6 %. By comparison with SOA yields, it is estimated that organic nitrates represent 50 % of the SOA for γ-terpinene and 100 % for α-terpinene. Organic nitrates are therefore major components of the SOA produced by the NO 3 oxidation of these two BVOCs. This conclusion can be compared with some 2007) and in forest region affected by urban air mass (Hao et al., 2014). This result thus confirms the major 420 contribution of organic nitrates in SOA formation and also the importance of NO 3 chemistry in this process.

Products at molecular scale and mechanisms
In order to provide mechanisms and to propose explanations for the different SOA yields between and terpinene, an identification of gas phase products at the molecular scale has been performed by PTR-ToF-MS.
The combination of two different ionization modes, with H 3 O + and NO + , for the detection of products allowed an 425 accurate identification of the molecules. Detected signals in both ionization modes and corresponding raw formula are summarized in Table 4. Products with molecular weights of 114, 152 and 168 g mol -1 for γ-terpinene and 58, 152, 168 and 171 g mol -1 for α-terpinene have been detected with high intensities. For γ-terpinene the molecular weights of 184, 215 and 229 have also been detected with lower intensities. Many of the detected products are nitrogenous species which is coherent with high production yields of organic nitrates. To explain 430 these observations, mechanisms have been proposed in Fig. 7 for γ-terpinene and in Fig. 8 for α-terpinene. All detected products are framed. Time profiles of PTR-ToF-MS signals were also used to determine whether the products are primary or secondary ones. Typical PTR-ToF-MS profiles are shown in the Fig. S2. First generation products are framed in blue and second generation ones in pink.

435
NO 3 radical can react by addition on one of the two double bonds (H-atom abstraction is considered to be negligible), each addition leading to the formation of two possible nitrooxy alkyl radicals. Kerdouci et al., 2014, who developed a structure-activity relationship for VOC+NO 3 reactions, suggest that the additions on the two double bonds of -terpinene have the same branching ratios. Thus, the four possible nitrooxy alkyl radicals were considered here. However, to facilitate the reading of the mechanism, only two radicals are shown in Fig. 7,

440
considering that in most cases, products obtained are isomers and cannot be distinguished by the analytical techniques used in this work. Nitrooxy alkyl radicals react with O 2 to form a peroxy radical (RO 2 ) (reaction 2).
However, the formation of an epoxide (reaction 3) with a molecular weight of 152 g mol -1 has also been detected finally, they can decompose by ring opening on the other side of the alkoxy group (reaction 8) leading to the formation of an alkyl radical which reacts to form a trifunctional compound (two carbonyl and one nitrate groups) of molecular weight MW = 229 g mol -1 (framed in green in Fig. 7). This product has been detected with a weak signal at m/z 230 in H 3 O + ionization mode but was not observed in NO + mode.
Because quantification of these products was not possible, branching ratios between these different pathways 460 could not be determined. However, if one hypothesizes that the products have similar responses, one can consider that signal intensities provide information on major products and that a product which has a strong signal is largely formed. In both modes, the dicarbonyl compound coming from the decomposition of the RO radical by reaction 7 has an intense signal, suggesting a high formation yield. According to the SAR developed by Vereecken and Peeters, 2009 from DFT calculation, energy barriers for reactions 7 and 8 appear to be similar 465 (E b,7 = 6.0 kcal mol -1 et E b,8 = 6.5 kcal mol -1 , with an error estimated by the authors of 0.5 kcal mol -1 ), leading to similar branching ratios. The signal of the trifunctional compound formed by reaction 8 (m/z 230) is weak and can be explained by its low volatility which suggests a partition in favor of the aerosol phase. It should also be noticed that no peroxynitrate (RO 2 NO 2 ) was detected in our experiments suggesting that RO 2 + NO 2 reactions are minor pathways. Due to high concentrations of NO 2 in the simulation chamber generated by the injection 470 mode of N 2 O 5 the detection of products coming from the reaction of peroxy and/or alkoxy radicals with NO 2 (respectively nitrooxy-peroxynitrate and dinitrate) were expected to be detected. Two hypothesis can be made regarding their absence (i) they can be formed but not detected by PTR-MS, or (ii) the structures of monoterpenes do not allow the formation of these compounds. Nevertheless, the detection of these compounds is usually not considered as a representative pathway in atmospheric conditions.

475
Primary products that still have a double bound can also react with nitrate radical to form second generation products, framed in pink in the mechanism. This is confirmed by time profiles of primary products MW = 152 and 168 g mol -1 which are decreasing with time (see Fig. S2) and by the detection of secondary products with MW = 114, 86, 184, 247 and 245 g mol -1 . The product with MW = 184 g mol -1 correspond to a second generation epoxide. Other products can be explained by the reaction of di-carbonyl compounds with NO 3 radical.
the RO + O 2 reaction (reaction 6-2) can lead to the formation of highly functionalized products with MW = 247 g mol -1 and MW = 245 g mol -1 .
Vapor pressures and percentage of partitioning into the aerosol phase were calculated as described in section 2.4

485
for aerosol mass loading of 800 µg.m -3 , typical of an experiment end, and are also presented next to the molecules in the Fig. 7. Among the first generation products, two types of products have low vapor pressures and can thus participate to SOA formation: first, the hydroxynitrate (e.g. MW = 215 g mol -1 ) which is characteristic of the RO 2 + RO 2 pathway (Fig. 7 -pathway 4). It was estimated to partition at 40% in particle phase. Secondly, trifunctional molecules ( Fig. 7 -pathway 8) are expected to have very low volatility and to be 490 present mainly in the aerosol phase (close to 100%). For these two products, the associate partitioning coefficient, following Eq. (10) has been calculated. Considering the uncertainty on due to the vapor pressure estimation, they can vary from 5.7 × 10 -4 to 2.0 × 10 -3 m 3 µg -1 for the hydroxynitrate (MW = 215 g mol -1 ) and from 5.3 × 10 -3 to 2.1 × 10 -2 m 3 µg -1 for the trifunctional compound (MW = 229 g mol -1 ). They appear to be consistent with the partitioning coefficients found with the two product model from Eq. (8) ( ,1 ,1 = 3.4 495 10 -3 m 3 µg -1 and ,2 = 4.5 x 10 -2 m 3 µg -1 ), especially by considering the associated uncertainty estimated on .
One can then consider that the two groups of products used by the two products model can be constituted by the hydroxynitrate and the trifunctional compounds, or by similar products.
Other first generation products are estimated to play a minor role in SOA formation. For secondary products, multifunctional products (with 4 chemical groups) are estimated to be between 80 and 100 % in the particle 500 phase. Other secondary products which are formed by fragmentation processes (dicarbonyl compounds) are expected to be volatile.
To conclude, the oxidation of γ-terpinene by NO 3 radical leads to the formation of several functionalized products and particularly to multifunctional organic nitrates which were detected in both phases. Most detected products can be formed by different pathways thus no preferential pathway could clearly be identified.

505
Nevertheless, products with up to 4 chemical groups (nitrate, carbonyl and alcohol) have been identified and explain the high SOA formation. In particular, the reaction RO 2 + RO 2  ROH + R(O) seems to play a significant role in the SOA formation as hydroxynitrates formed have low vapor pressures (both first and second generation products) with large probability of partitioning into the particle phase.

α-terpinene oxidation scheme
As described for γ-terpinene, addition of NO 3 radical on the double bonds of α-terpinene can lead to the formation of four nitrooxy alkyl radicals. Nevertheless, in this case, the conjugation of the two double bonds allows a delocalization of the single electron and hence, a stabilization of the two corresponding nitrooxy alkyl 525 radicals (see Fig. S3 in SI). Thus, nitrooxy alkyl radicals expected to be the most favorable are those which can undergo an electron delocalization. Note that this delocalization leads to the formation of two tertiary nitrooxy alkyl radicals. Mechanism has been established by considering all possible radicals but for clarity in Fig. 8 propose the following decomposition pathways for RO radicals: i) alkoxy radicals can lose the isopropyl group leading to the formation of a cyclic ketonitrate and an isopropyl radical which then evolve to form acetone. The cyclic ketonitrate and the acetone have been detected by PTR-ToF-MS with high intensity signals in both ionization modes. Acetone was also detected by FTIR but its concentration was close to the detection limit of 10 ppb and formation yield could not be precisely measured. Nevertheless, by considering the 10 ppb detection 540 limit, the acetone formation yield is expected to be below 3 %; ii) they can also decompose by a scission of the C(ONO 2 )-CH(O•) bond (reaction 7, framed in orange) leading to a ring opening and to the formation of a dicarbonyl product with MW = 168 g mol -1 . This compound was detected with high signals by PTR-ToF-MS in H 3 O + and NO + modes, respectively at m/z 169 and m/z 168; iii) the formation of trifunctional species (one nitrate group and two carbonyl group) with MW = 229 g mol -1 , coming from the ring opening on the other side 545 of alkoxy group (reaction 8) has also been observed but it needs confirmation. Indeed, mass m/z 230 has been detected in H 3 O + mode, but neither m/z 229 nor 259 were detected in NO + mode. As for γ-terpinene, they can be formed with low concentrations and/or be mostly in particle phase leading to weak signals with PTR-ToF-MS.
The SAR proposed by Vereecken and Peeters, 2009 allowed estimating the branching ratios between the different evolution pathways of alkoxy radicals. It suggests that ring openings (reaction 7 and 8) are the two most 550 likely pathways with similar branching ratios (differing of 0.6 kcal mol -1 ). Reaction 7' corresponding to the loss of the isopropyl group appears to be less likely. This is in agreement with the low acetone formation yield. This is also in agreement with the detection of the trifunctional species for which the weak signal observed can be explained by its low volatility. This product is the only primary product expected to contribute to SOA formation (with = 50 %). As a reminder, the associated partitioning coefficient for this trifunctional compound 555 (MW = 229 g mol -1 ) was estimated to range between 5.3 × 10 -3 and 2.1 × 10 -2 m 3 µg -1 . It can correspond to the ,2 estimated by the two-product model in Eq. (8) ( ,2 = 3.5 × 10 −2 3 µ −1 ). The second partitioning coefficient in the two-product model ( ,1 = 4.5 × 10 −1 3 µ −1 ) cannot be attributed, and it can be explained https://doi.org/10.5194/acp-2020-504 Preprint. Discussion started: 8 June 2020 c Author(s) 2020. CC BY 4.0 License.
by i) the fact that this product can be totally in the particle phase and thus not detectable in the gas phase and ii) the very low production of SOA, leading to a low precision on the fit.

560
The only secondary products identified are epoxides that can come from the reaction of the primary epoxide with NO 3 radical, but also from the oxidation of the unsaturated dicarbonyl compounds. In the last case, they are epoxides coming from the loss of -NO 2 from alkyl radicals (reaction 3-2). Other secondary products were expected, but not detected. Signal m/z 243 has been detected in NO + mode but this signal could not be attributed to a product.

565
In conclusion, products detected with highest signals are cyclic ketonitrates, dicarbonyl compounds and epoxides. These three families of products have high vapor pressures and are thus not expected to significantly contribute to SOA formation. Trifunctional products with low vapor pressures have also been detected with low signals, suggesting low formation yields or strong partition in aerosol phase. Low SOA yields measured for αterpinene suggest that the first explanation is more favorable.

Comparative discussion
Even though α-and γ-terpinene have similar chemical structure only differing by the position of the double bonds, their reactions with nitrate radical have different consequences especially regarding SOA formation.
Mean SOA and organic nitrate yields obtained for both compounds are presented in Table 5. Several previous studies on BVOC + NO 3 reactions suggest a correlation between organic nitrate yield and SOA formation (Fry et 575 al., 2014;Hallquist et al., 1999). α-pinene presents indeed a low organic nitrate yield, in good agreement with a SOA yield close to zero, when limonene and ∆-carene both present high SOA and organic nitrate yields. In our study, Y ONg and Y ONp for α-and γ-terpinene are similar regarding the uncertainty. Thus α-terpinene does not follow this correlation, as it produces a high amount of organic nitrates, but almost no SOA.
To interpret the difference in SOA formation, the mechanisms have to be compared: for γ-terpinene, the major 580 SOA production can be explained by the formation of (i) a primary hydroxynitrate coming from RO 2 + RO 2 pathway and (ii) a primary trifunctional nitrate formed by the decomposition of alkoxy radical and (iii) detection of secondary products, all presenting very low vapor pressures. On the other hand, for α-terpinene (i) no hydroxynitrate and (ii) no secondary products with low vapor pressure were detected in the oxidation products.
The absence of hydroxynitrate may be explained by the fact that, due to relocation of the free electron by 585 mesomeric effect for alkyl radicals, the most stable radicals are tertiary and cannot undergo H-shift to produce the hydroxynitrate. Hence, two aspects of the mechanism appear to be critical for the SOA formation: i) the peroxy radical reaction pathways: when the carbon that bears the radical group, has a hydrogen available, the reaction RO 2 + RO 2  ROH + R(O) can occur, leading here to the formation of an hydroxynitrate. Due to hydrogen bonds, this product has very low vapor pressure. This has already been reported by Ng et al. 2008 who 590 compared isoprene SOA formation for reactions RO 2 + RO 2 et RO 2 + NO 3 . The study reported higher formation of SOA for the pathway RO 2 + RO 2 due to the formation of the specific hydroxynitrate and its secondary reaction. In our study, this product was detected for γ-terpinene, but not for α-terpinene for which we expect that most stable radicals are tertiary ones and cannot undergo this reaction pathway. ii) the alkoxy radical reaction pathways: several decomposition pathways can occur, leading to products which have very different volatilities. that for ∆-carene, the decomposition of alkoxy radicals can lead to the formation of keto-nitrooxy-alkyl radicals, whereas for α-pinene, the alkoxy radicals decompose almost exclusively to form the dicarbonyl compound.
One interesting fact is that for both compounds, epoxides have been detected, whereas it is usually admitted that their formation is favored only at low oxygen concentration (Berndt and Böge, 1995). Many studies did not detect these compounds (Jaoui et al., 2013;Slade et al., 2017;Spittler et al., 2006), but their formation has already been observed by Skov et al., 1994, which studied NO 3 radical oxidation of several alkenes and isoprene. isoprene. In our study, epoxides were not quantified, but based on previous studies, their formation yields are expected to be low.

Conclusions & atmospheric impacts
In summary, this work provides kinetic and mechanistic data on the oxidation by NO 3 radical of αand γ-625 terpinene, using simulation chambers. The two compounds present very similar chemical structures (the same carbon skeleton and two double bonds which are conjugated in the case of α-terpinene, and not for γ-terpinene) and this work aimed at highlighting the influence of the structure on the reactivity, in particular on the SOA formation.
Absolute and relative kinetic determinations have been performed. This study provides the first absolute determination for α-terpinene. Both α-and γ-terpinene appear to be very reactive towards nitrate radical due to https://doi.org/10.5194/acp-2020-504 Preprint. Discussion started: 8 June 2020 c Author(s) 2020. CC BY 4.0 License.
the presence of two double bonds. αterpinene is particularly reactive with NO 3 radicals due to the conjugation of the double bonds while γterpinene is four times less reactive.
As far as we know, this study is the first mechanistic study for the oxidation of α-terpinene by NO 3 radicals. Our study has confirmed that the NO 3 oxidation of α-and γ-terpinene produces large amounts of organic nitrates

635
(with overall yields ~ 50%) which have been shown to be present in both gas and aerosol phases. Nevertheless, major differences in the SOA formation for the two compounds have been pointed out despite their similar structure. γ-terpinene has been shown to be an efficient SOA precursor, when α-terpinene is a poorly efficient SOA precursor. To explain these differences, a molecular scale study has been conducted and two reaction pathways have been shown to play a key role in the SOA formation: i) the peroxy radical reaction pathways 640 leading to the formation of low volatility hydroxynitrates, which were detected for γ-terpinene but not for αterpinene ii) the alkoxy radical scission pathways which can either form high volatility dicarbonyl compounds or low volatility trifunctional products, depending on where the scission occurs.
The atmospheric lifetimes of the two compounds have been estimated by using typical nighttime NO 3 concentration (10 ppt) and low insolation diurnal concentration (0.1 ppt, Corchnoy and Atkinson, 1990). These 645 lifetimes are presented and compared to those estimated for oxidation by OH and ozone in Table 6. Prior to the discussion, it is important to remind that monoterpenes are intensively emitted during both day and night (Lindwall et al., 2015). From Table 6, it can be observed that the two monoterpenes exhibit very short lifetimes towards NO 3 radical for nighttime conditions (40 s for α-terpinene and 2 min for γ-terpinene) confirming that NO 3 oxidation is a major sink for these compounds. As expected, lifetimes estimated for low sunlight diurnal 650 conditions are longer (few hours), but are still fairly short. By comparison with lifetimes estimated for other oxidants, it is concluded that all three oxidants are very efficient sinks. For low insolation diurnal conditions, even though diurnal chemistry is clearly led by OH and O 3 , NO 3 oxidation is not negligible. This result can be compared to the modeling study of Forkel et al., 2006, which has shown that NO 3 oxidation is an important sink of BVOCs even during the day under specific conditions (under canopy, with low luminosity and considering a 655 calculated mixing ratio for NO 3 of 3 ppt).
The short lifetimes indicate that oxidation products are formed close to the BVOCs emission areas. If all three oxidants are major sinks, the products formed by the different processes are very different. Organic nitrates can also be formed by OH oxidation through RO 2 + NO reactions but yields are much lower. Lee et al. 2006 have investigated the products formed by the OH oxidation of 16 terpenoids, including α-and γ-terpinene, in presence 660 of NOx. Organic nitrate yields were shown to be less than 1%. So a major impact of BVOC oxidation by NO 3 radicals is the formation of organic nitrates which are known to act as NOx reservoirs. In our study, dicarbonyl compounds have also been shown to be formed but the same compounds have been detected as major products of the OH chemistry of α-and γ-terpinene by Lee et al. 2006. So, this is not a special feature of the nighttime chemistry.

665
SOA yields produced by NO 3 oxidation can be compared to those formed by ozonolysis and OH oxidation. Friedman and Farmer, 2018, Griffin et al., 1999, Lee et al., 2006 have measured the SOA yields for the OH oxidation of several terpenes, including α-and γ-terpinene. In the experiments performed by Griffin et al., 1999, mixing of terpenes were introduced leading to overall SOA yields which were found to be 4 % at 10 µg m -3 and https://doi.org/10.5194/acp-2020-504 Preprint. Discussion started: 8 June 2020 c Author(s) 2020. CC BY 4.0 License.
are below 20 %. Finally, the study of Friedman and Farmer, 2018, was performed with low NOx conditions and SOA yields were found to be very low (≤ 1 %) at M 0 = 10 µg m -3 . Regarding these results, the oxidation by NO 3 appears to be a much more efficient SOA source than the OH oxidation. This observation has already been made by several previous studies (Hallquist, Wangberg, and Ljungstrom 1997;Griffin et al. 1999;Spittler et al. 2006;Ng et al. 2008;Fry et al. 2014;Boyd et al. 2015, Slade et al. 2017. Regarding the ozonolysis of and -675 terpinene, there is, to our knowledge, no data on SOA yields in the literature. However, more generally, the ozonolysis of BVOCs is known to be an important source of SOA. In conclusion, the most important impacts of this chemistry rely on the formation of large amounts of organic nitrates (present in both gas and aerosol phases) and SOA. Organic nitrates play a key role in tropospheric chemistry because they behave as NOx reservoirs, carrying reactive nitrogen in remote areas. Their chemistry in 680 gas and aerosol phases is nevertheless still not well documented. Considering that our study shows a large production of multifunctional organic nitrates, it is necessary to better understand their reactivity in order to better evaluate their impacts. Formation of SOA seems on the other hand, strongly dependent on the structure of the BVOC. Studies at molecular scale are thus mandatory to better evaluate the impact of this chemistry on the SOA formation.