Suppression of new particle formation from monoterpene oxidation by NO x

The impact of nitrogen oxides (NO x = NO + NO 2 ) on new particle formation (NPF) and on photochemical ozone production from real plant volatile organic compound (BVOC) emissions was studied in a laboratory set up. At high NO x conditions (BVOC/NO x < 7, NO x > 23 ppb) no new particles were formed. Instead photochemical ozone formation 5 was observed resulting in higher hydroxyl radical (OH) and lower nitrogen monoxide (NO) concentrations. As soon as [NO] was reduced to below 1 ppb by OH reactions, NPF was observed. Adding high amounts of NO x caused NPF orders of magnitude slower than in analogous experiments at low NO x conditions (NO x ∼ 300 ppt), although OH concentrations were higher. Varying NO 2 photolysis enabled showing that NO was 10 responsible for suppression of NPF suggesting that peroxy radicals are involved in NPF. The rates of NPF and photochemical ozone production were related by power law dependence with an exponent of approximately − 2. This exponent indicated that the overall peroxy radical concentration must have been the same whenever NPF ap-peared. Thus permutation reactions of ﬁrst generation peroxy radicals cannot be the 15 rate limiting step in NPF from monoterpene oxidation. It was concluded that permutation reactions of higher generation peroxy radical like molecules limit the rate of new particle formation. In contrast particle mass sensitive 20 about an order of magnitude.

was observed resulting in higher hydroxyl radical (OH) and lower nitrogen monoxide (NO) concentrations. As soon as [NO] was reduced to below 1 ppb by OH reactions, NPF was observed. Adding high amounts of NO x caused NPF orders of magnitude slower than in analogous experiments at low NO x conditions (NO x ∼ 300 ppt), although OH concentrations were higher. Varying NO 2 photolysis enabled showing that NO was 10 responsible for suppression of NPF suggesting that peroxy radicals are involved in NPF. The rates of NPF and photochemical ozone production were related by power law dependence with an exponent of approximately −2. This exponent indicated that the overall peroxy radical concentration must have been the same whenever NPF appeared. Thus permutation reactions of first generation peroxy radicals cannot be the 15 rate limiting step in NPF from monoterpene oxidation. It was concluded that permutation reactions of higher generation peroxy radical like molecules limit the rate of new particle formation.
In contrast to the strong effects on the particle numbers, the formation of particle mass was less sensitive to NO x concentrations, if at all. Only at very high NO x concen-

Introduction
Secondary organic aerosols (SOA) are an important component of tropospheric aerosols hence affecting the radiation balance of the Earth's atmosphere by direct effects (scattering or absorbing sun light) and by indirect effects (as cloud condensation tion of SOA has natural sources. Plant emitted biogenic volatile organic compounds (BVOC) are oxidized in the atmosphere and the oxidation products condense on preexisting particulate matter increasing their size and modifying their properties. In clean background air, where only small amounts of particulate matter exist, new particle formation (NPF) has been observed (e.g. Dal Maso et al., 2007). Despite extensive stud- 5 ies, the mechanisms of NPF are not yet clarified. It is generally accepted that sulfuric acid (H 2 SO 4 ) plays a key role for the formation of critical seed clusters (Kirkby et al., 2011). However, formation of particulate matter with diameters above a few nanometres requires additional precursors (Kulmala et al., 2013).
Oxidized organic matter is postulated to be the complementary precursor to H 2 SO 4 10 (Kulmala et al., 2004Metzger et al., 2010;Riipinen et al., 2011Riipinen et al., , 2012Zhang et al., 2011). However, many oxidation steps are required for the formation of highly oxidized organic products with low enough vapour pressures that enable condensation onto critical clusters (Kiendler-Scharr et al., 2009;Ehn et al., 2010Ehn et al., , 2012. The complexity of these oxidation steps hampers elucidating the basic processes of NPF. production rates may be used as a tool to gain more insight into the role of peroxy radicals in particle formation: one peroxy radical is consumed per NO 2 formed in the NO + RO 2 reaction and furthermore, one O 3 molecule is formed from the photolysis of one NO 2 molecule. Hence the rate of photochemical ozone production, P(O 3 ), is a quantity that is linearly related to the consumption of peroxy radicals in reactions with 15 NO. Consequently P(O 3 ) can be used as quantity giving hints to the suppression of permutation reaction rates.
To gain more insights into the role of peroxy radicals in NPF, we used P(O 3 ) as a tool. We found that the photochemical system behaved more complicated than expectable from a simplistic approach: assuming that NPF would be controlled by permutation floors mounted in separate climate chambers. Both chambers were operated as continuously stirred tank reactors (CSTR) with Teflon fans ensuring homogeneous mixing with mixing times of about two minutes. The smaller chamber contained the plants. A fraction of the air exiting this plant chamber was fed into the larger chamber that served as reaction chamber. Residence time of the air in the plant chamber was about 10 22 min; residence time of the air in the reaction chamber was about 63 min.
Air was purified by an adsorption dryer (KEA 70; Zander Aufbereitungstechnik GmbH & Co. KG, Essen, Germany) and by a palladium catalyst operating at 450 • C. Ozone, NO, NO 2 , and volatile organic compounds (> C 3 ) were removed by the purification system. Concentrations of CO 2 and water vapour were further reduced by the adsorption 15 dryer. By adding CO 2 from cylinders the CO 2 concentration in the chamber was kept at 350 ppm. The dew point in the plant chamber was restricted to a maximum of 13 • C to avoid condensation in the transfer line to the reaction chamber which was at a temperature of 17 ± 0.5 • C.
Discharge lamps (HQI 400 W/D; Osram, Munich, Germany) were used to simulate 20 the solar light spectrum in the plant chamber. At full illumination and at typical midcanopy height photosynthetic photon flux density (PPFD) was 480 µmol m −2 s −1 . Infrared radiation (between 750 and 1050 nm) was reflected by filters (type IR3; Prinz Optics GmbH, Stromberg, Germany) placed between the lamps and the plant chamber in order to minimize radiative heating of the plants. 25 A fraction of the air leaving the plant chamber (∼ 12 L min −1 ) was fed into the reaction chamber. In addition, another air stream was introduced to the reaction chamber (∼ 11 L min −1 ). This second air stream was used to add O 3 to the reaction chamber 25831 The reaction chamber was equipped with the same HQI 400 W/D lamps as the plant chamber. The HQI lamps had a spectrum similar to sunlight but with low intensity in the near UV. To obtain sufficient NO 2 photolysis frequencies (J(NO 2 )) another 12 discharge lamps (Phillips, TL 60 W/10-R, 60 W, λ max = 365 nm, from here on termed UVA lamps) were used. During the measurements described here, the reaction chamber was illuminated with 2 of the HQI 400 W/D lamps and all 12 UVA Lamps resulting in an NO 2 photolysis frequency of J(NO 2 ) = 4.3 × 10 −3 s −1 . These lamps were installed outside of the reaction chamber. Due to the thick glass walls, light with wavelengths shorter than 350 nm was absent in the chamber.
OH radicals were generated by O 3 photolysis and reaction of the O 1 D-atoms with 15 water vapor. As efficient ozone photolysis in the Hartley band requires wavelengths shorter than 350 nm, a UVC lamp was installed inside the reaction chamber (Philips, TUV 40 W, λ max = 254 nm, from here on termed as TUV lamp). Whenever the TUV lamp was switched on, OH radicals were generated at concentrations > 10 7 cm −3 .
These concentrations are much higher than the OH concentrations expected as a by-20 product of BVOC ozonolysis with the TUV lamp switched off. In the following we will always refer to the photolytical generated OH radicals, although there were always some OH radicals in the reaction system. The TUV lamp was shielded by glass tubes imposed on the lamp with a gap between the glass tubes. The photolytical OH production was adjusted by varying this gap. During the experiments described here J(O 1 D) 25 was constant at about 9 × 10 −4 s −1 . NO 2 photolysis by the TUV lamp was negligible but HNO 3 was photolysed leading to a background of ∼ 300 ppt NO x (see below). Nitrogen monoxide (Linde, 99.5 ± 5 ppm NO in Nitrogen) was added to the air introduced from the plant chamber into the reaction chamber. The NO x concentrations in 25832 The UVA lamps were switched on long before the TUV lamp was switched on. Without the TUV light i.e. in the absence of high OH concentrations [NO], [NO 2 ] and [O 3 ] were near the photostationary steady state (PSS). At high NO x conditions, ozone concentrations increased after switching on the TUV lamp. As NO 2 photolysis by the TUV lamp was inefficient, the increases of [O 3 ] in the chamber were due to photochemical 10 ozone production and not due to variation of J(NO 2 ). Based on differences of ozone concentrations between the chamber inlet and outlet, respectively, the rates of photochemical ozone production (P(O 3 )) were determined (for more details see Supplement Sect. S2).
Trace gases were measured using commercial equipment. Ozone concentrations 15 were determined by UV absorption (Thermo Environmental instruments, model 49), NO was measured by chemiluminescence (Eco Physics, CLD 770 AL ppt), and NO 2 by chemiluminescence after photolysis (Eco Physics, PLC 760). Mixing ratios of biogenic volatile organic compounds (BVOC) in plant chamber and reaction chamber were determined by measurements at the outlets of the respective chambers. These The GC-MS system operated at plant chamber outlet was used to quantify the BVOC introduced into reaction chamber. Note that the concentration of BVOC introduced into the reaction chamber was smaller than measured at the outlet of the plant chamber because of the second air flow into reaction chamber. For all BVOC values given here this dilution was always considered. 5 The GC-MS operated at the outlet of the reaction chamber was used to quantify OH concentrations by measuring the decrease of certain BVOC. During the experiments with plants we used α-pinene and β-pinene to determine OH concentrations. With the onset of O 3 photolysis the concentrations of both BVOC decreased substantially but were still measurable. Using the rate constants of their reactions with OH 10 (α-pinene = 5.37 × 10 −11 cm 3 s −1 , β-pinene = 7.89 × 10 −11 cm 3 s −1 , Atkinson, 1997)  nected to the reaction chamber by a straight stainless steel tube (diameter: 6 mm, lengths: 0.5 m). This UCPC had a nominal activation diameter of 7 nm and was used to count the total number of particles formed in the chamber. A Scanning Mobility Particle Sizer (SMPS, TSI3081 + TSI3786) also directly coupled to the reaction chamber measured the number size distribution between 10 and 500 nm. The obtained size dis-20 tributions were converted into volume distributions to determine particle volume-and mass yields. The yield of particle formation was related to the total BVOC consumption, given by the difference between inlet and outlet of the reaction chamber. Thus, the yield considers also the consumption of BVOC by ozone reactions for particle mass formation. 25 Nucleation rates (J 7 ) were determined from the first derivative of particle number concentrations as a function of time. The particle counter was sensitive only to particles which had already reached diameters of about 7 nm. Assuming these small particles to be spherical and having a density of ∼ 1.2 g cm −3 , the masses of such 7 nm particles Introduction are in the range of ∼ 1000 monoterpene masses. In the context used here, J 7 does not mean the formation rate of critical clusters but shows the appearance of small particles that already comprise of many molecules that have participated in early particle formation. Direct plant emissions were used as SOA precursor because these are realistic 5 BVOC mixes making the results of such experiments independent of specific effects of individual BVOC. Plants were delivered from Israel (details see Lang-Yona et al., 2010) and stored in a growth room before the measurements. The plant chamber contained 5 three to four years old tree seedlings: two Aleppo pines (Pinus halepensis L.), one Holm oak (Quercus ilex L.), one Palestine oak (Quercus calliprinos L.), and one Pistachio (Pistacia palestina L.). A diurnal light cycle was simulated in the plant chamber by switching on and off the HQI lamps (06:00 to 18:00 LT full illumination, 18:00 to 19:00 LT simulation of twilight by switching off individual lamps, 19:00 to 05:00 LT darkness, and 05:00 to 06:00 LT simulation of twilight by switching on individual lamps). Not later than three hours after twilight in the morning the VOC emissions from the plants 15 were quite constant and another three hours later conditions in the reaction chamber were near steady state. Then the TUV lamp was switched on to induce OH production and particle formation. One measurement was conducted per day with NO x addition every second day.
In some experiments α-pinene was used as sole SOA precursor by using air from 20 a permeation/diffusion source instead of the air from the plant chamber. Details of this source are described in Mentel et al. (2009) Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | reaction chamber, HNO 3 produced in reactions of NO 2 with OH diffused into the plate. The next day, when NO x was removed from the air at chamber inlet, HNO 3 diffused out of the plate. As HNO 3 was photolysed at 254 nm, NO x was produced in the reaction chamber. Together with some NO emitted from the plants, the NO x production by HNO 3 photolysis determined lower limit of around 300 ppt NO x in the chamber. Hence we 5 performed measurements at low NO x conditions but no measurements at zero NO x . When OH was produced and BVOC concentrations decreased, NO x concentrations also decreased due to reactions with OH. In accordance with Pandis et al. (1991) and related publications we will parameterize our experiments by [NO x ] 0 , the initial NO x concentrations in the reaction chamber before OH production. In case of the 10 experiments without NO x addition, [NO x ] 0 was set to 300 ppt, i.e. the NO x concentration measured shortly after the TUV lamp was switched on and HNO 3 was photolysed. 3 Methods: peroxy radicals and their reaction system 15 Details of the photochemical system, the chemical reactions and the derivation of the equations are given in the Supplement (Sect. S1). For better comparability, numbering of reactions and rate constants is adapted to that in the Supplement. Peroxy radicals (RO 2 + HO 2 ) are key species in the photochemical reaction system of interest. Figure 1 illustrates the known reactions of peroxy radicals as well as their 20 trend with changing NO x in the reaction system. Added is a pathway from permutation reactions to particle formation. Such pathway was suggested before for the growth of particulate matter (e.g. Kroll et al., 2006) Reaction (R2) (Reaction R2 = Reaction R2a + Reaction R2b with rate constants k 2a and k 2b ) is the major loss path of RO 2 at high NO x conditions: and leads to the formation of alkoxy radicals (RO) and NO 2 in Reaction (R2a) and 5 to the formation of organic nitrates (RONO 2 ) in Reaction (R2b). Upon NO 2 photolysis ozone is formed. The sum of the rate constants k 2a and k 2b (k 2 ), is the average rate constant for the reaction of NO with RO 2 . The branching ratios for ozone formation and organic nitrate formation in (R2) are termed Y(O 3 ) and Y(RONO 2 ), respectively.
Reaction (R3) (R3 = R3a + R3b) is the main loss for RO 2 radicals at low NO x conditions: Reaction (R3a) forms hydroperoxides. Reaction (R3b) forms different products includ- 15 ing alkoxy radicals, alcohols, carbonyl compounds etc. (e.g. master chemical mechanism, MCM for α-pinene), and probably also alkyl peroxides ROOR (e.g. Hallquist et al., 2009). The formation of ROOH can be considered as a special case of ROOR' formation with R = H. In Reaction (R3) different peroxy radicals (RO 2 and HO 2 ) react with each other with permutation of all pairs being a priori possible. Reaction (R3) is 20 therefore termed "permutation reaction". An average rate constant k 3 can be specified for a given reaction system of RO 2 and HO 2 . Products of Reaction (R3) will be termed as permutation reaction products, PRP, and their production rates as P(PRP). If PRP are involved in NPF, NO will switch a photochemical system containing BVOC into either O 3 formation or new particle formation. 25 The photochemical system switches between P(PRP) and P(O 3 ). Considering that the chemical systems were quite similar this switch was approximated by Eq. (1) [NO] 2 (1) In Eq. (1), const is a constant. Varying [NO] leads to complex relationships between P(O 3 ) and P(PRP), and eventually to particle formation. If [RO 2 ] is constant, P(PRP) and P(O 3 ) are related by an inverse power law dependence with an exponent close to 5 −2 (for details see Supplement, Sect. 1). During our experiments we found power law dependence in this range.

BVOC emissions
Main emissions from the Mediterranean plants were those of monoterpenes (MT). Iso-10 prene emissions were substantially lower and emissions of other biogenic VOC were negligible. Figure 2 shows the emission pattern expressed as mixing ratios of BVOC introduced into the reaction chamber. During the twenty six experiments conducted within about a month, day to day fluctuations in the plants' emissions were unavoidable. The BVOC concentrations introduced into the reaction chamber varied between 15 ∼ 77 ppb C to ∼ 130 ppb C (∼ 44 to ∼ 73 µg m −3 ). Although contributions of isoprene were always less than 10 % they were always considered in summing up BVOC concentrations. further. For example, during low NO x experiments α-pinene concentrations in the reaction chamber were at maximum 200 ppt indicating that more than 95 % of the α-pinene was oxidized. At high NO x conditions ([NO x ] > 30 ppb) α-pinene concentrations at the chamber outlet were in the range of several hundreds of ppt for the first hours. In the course of such experiments the α-pinene-and β-pinene concentrations decreased to 5 ∼ 20 ppt because OH concentrations increased with time. Nevertheless, in nearly all measurements, α-pinene and β-pinene concentrations remained high enough to allow reasonable determinations of OH concentrations. Because the isoprene contribution was less than 10 % and fairly constant from experiment to experiment, the suppressing impact of isoprene on J 7 (see Kiendler-Scharr 10 et al., 2009) was low and constant from experiment to experiment. In the following analyses the impact of isoprene was neglected for the interpretation of NO x impacts on NPF. Figure 3 shows the temporal development of particle number density for three experi- 15 ments with different [NO x ] 0 . Two striking observations were made. The first was a decrease of J 7 with increasing NO x . The second was a delay between the start of photolytic OH production and the onset of NPF. During long lag times [NO x Fig. 4). 20 For 1.1 < BNR < 10, rates of NPF and P(O 3 ) were inversely related. With increasing BNR, J 7 increased by two orders of magnitude whereas P(O 3 ) decreased from ∼ 60 ppb h −1 to non-measurable values (Fig. 5). In low NO x regimes with 10 < BNR < 30 P(O 3 ) was negligible while J 7 increased further by one order of magnitude. For BNR > 30 both, J 7 and P(O 3 ), were insensitive to BNR. 25 Assuming power law dependences for J 7 = f (BNR) in the range 1.1 < BNR < 30 and for P(O 3 ) = f (BNR) in the range 1.1 < BNR < 10 we derived an exponent of 2.3 ± 0.1, Introduction to 65 µg m −3 . Normalization was conducted using the results from the linear regression analysis of data obtained without NO x addition. Considering that more than 95 % of the 10 BVOC introduced into the reaction chamber were oxidized PM MAX ,norm was calculated as:

NO x dependence of ozone-and new particle formation
PM MAX , norm is shown in Fig. 7 as a function of BNR. The normalized data show still some scatter. However, as long as BNR > 10, PM MAX ,norm was independent of NO x 15 within the error margins of the data. J 7 was affected at BNR < 30 (see Fig. 5 and compare ratio J 7 /PM max in Table 1). For better overview, data obtained during the experiments using BVOC emissions from plants as SOA precursors are listed in Table 1. As shown in Table 1, the ratio J 7 /PM max increased by about two orders of magnitude when BNR increased from 3 to 20 30 ppb C ppb −1 . At BNR > 30 J 7 /PM max leveled out. and PM max dropped from 6 to 2 µg m −3 . To remove the effect of the suppressions induced by UVA light, we normalized the data for the two different light conditions sepa-10 rately. As reference points we choose J 7 and PM MAX as measured without NO x addition with UVA light off (J(NO 2 ) = 0 s −1 ) and UVA light on (J(NO 2 ) = 4.3×10 −3 s −1 ). Since the data were normalized separately for the two light conditions, the impact of UVA light on NPF and SOA mass cancelled out. Figure 8 shows that the suppression of J 7 with NO x addition was stronger at J(NO 2 ) = 4.3 × 10 −3 s −1 than at J(NO 2 ) = 0 s −1 . This suggests 15 that NO is the NO x component which is mainly responsible for suppression of new particle formation. The OH concentrations were not systematically affected by the addition of NO x . After taking out the suppression by UVA light, the residual relative suppression of PM MAX with increasing [NO x ] was small (max. 50 %) and showed no systematic varia-20 tion with [NO x ] 0 . The suppression of PM MAX was within the uncertainty of the data and small compared to the suppression of J 7 . The lack of significant suppression of mass formation was consistent with the results obtained with the plant emitted BVOC mix. As all measurements with α-pinene were conducted at BNR > 15; suppression of SOA mass formation should thus not occur. It is therefore not possible to decide whether NO 25 or NO 2 caused the suppression of PM MAX in the BVOC mix at BNR < 10.

Comparison with literature data
We found a strong suppression of NPF with increasing NO x for the BVOC mix emitted from Mediterranean species and dominated by monoterpenes. With BNR decreasing 5 from 30 to 1.1 J 7 decreased by more than three orders of magnitude.
Limited data exist in the literature with respect to impacts of NO x on NPF. Pandis et al. (1991) found nucleation thresholds for β-pinene in the range of 300 to 500 ppb C which is quite high. According to Kroll et al. (2006) the high threshold found by Pandis et al. can be explained by high background NO concentrations. These high background 10 NO concentrations might have suppressed RO 2 formation. Hence formation of PRP is also suppressed. From this Kroll et al. (2006) concluded that hydroperoxides formed in Reaction (R3a) are important for NPF. If so, NPF should be totally suppressed at high NO x conditions as found in this study.
In contrast, Lim and Ziemann (2005) measured particle formation from OH initiated 15 oxidation of long chain alkanes even at [VOC] 0 /[NO x ] 0 < 0.05. This would not easily be explainable if PRP would be the only compounds involved in early particle formation. But Lim and Ziemann also observed that the particles formed in their experiments comprised mostly oxidation products of organic nitrates. The observation of Lim and Ziemann are in line with observations on the sesquiterpene -NO x system by 20 Ng et al. (2007b). Mechanisms of SOA formation from molecules much larger than monoterpenes may favor a RONO 2 route and may differ from those discussed here for monoterpenes.

25843
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Working hypothesis on the mechanism of NO x impacts on NPF
Only when OH was produced from ozone photolysis we observed formation of new particles with diameters above 7 nm, otherwise NPF was insignificant. We conclude that OH is a necessary oxidant for NPF in the range of [BVOC] 0 applied here (see also Mentel et al., 2009;Kiendler-Scharr et al., 2009, 2012Lang-Yona et al., 2010).
At high [NO x ] 0 conditions, NPF was not observed during the first hours of the experiments (as long as [NO x ] was high), although OH concentrations reached about 2.6 × 10 6 cm −3 (see Table 1). Under atmospheric conditions, NPF is observed at com- steadily increased in this time period (see Fig. 4). OH concentrations reached high levels of 2.2-2.6 × 10 7 cm −3 when NPF was observed (compare red square in Fig. 6). The threshold OH concentration was clearly higher than measured at the onset of NPF at low to medium [NO x ] 0 (Table 1) but J 7 was orders of magnitude lower. The enhanced OH threshold for NPF in presence of NO x implies that the mechanism responsible for 20 the suppression of NPF by NO x is superimposed to impacts of [OH].
Results of the experiment with α-pinene as SOA precursor were also consistent with this conclusion. In this experiment [OH] was not systematically affected by the addition of NO x because the reactivity of the total mix increased by less than 20 %. But J 7 decreased by an order of magnitude confirming that there must have been another 25 process suppressing J 7 by the addition of NO x .
The experiment with α-pinene furthermore confirmed that it is NO that is mainly responsible for suppression of J 7 . This is in accordance with the hypothesis that PRP Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | are precursors in NPF: reactions of NO with peroxy radicals (Reaction R2) suppress the formation of PRP and therewith NPF. The same arguments apply for suppression of NPF as assumed for the suppression of SOA mass formation by NO x (e.g. Pandis et al., 1991;Presto et al., 2005b;Kroll et al., 2006;Ng et al., 2007b). 5 As suggested above, peroxy radicals play a crucial role for NPF. Their concentrations are difficult to measure, however, they can be estimated using the deviation of [

New particle formation in relation to photochemical ozone production
Photochemical ozone formation was insignificant for BNR > 10. However, P(O 3 ) increased when BNR decreased from 10 to 1.1. In BVOC rich systems with BNR > 10 the main RO 2 losses proceeded via Reaction (R3), because RO 2 consumption by NO via (R2) could not compete. At low BNR P(O 3 ) became significant, indicating that Reactions (R2) proceeded with considerable rates. This behavior is known from the classic empirical kinetic modeling approach (EKMA) ozone isopleth diagram (Seinfeld and Pandis, 2006). NPF exhibited a behavior opposite to P(O 3 ), namely, J 7 decreased with decreasing BNR. J 7 dropped by three orders of magnitude when BNR decreased from 30 to 10 1.1. Obviously, increasing [NO x ] 0 had caused a switch of the chemical system from Reaction (R3) to Reaction (R2) (from the red to the blue arrows in Fig. 1 Fig. 5 might therefore be fortuitous at least for BNR < 10 with lag times exceeding several minutes. We therefore made use of Eq. (1) which relates [RO 2 ] and P(O 3 ). 20 From the overall increase of P(O 3 ) for 10 > BNR > 1.1 in Fig. 5, we conclude that P(O 3 ) is NO x -limited in that BNR range. Moreover, P(O 3 ) increases approximately linearly with decreasing BNR, indicating that the rate of Reaction (R2a) must also increase approximately linearly. P(O 3 ) is thus a linear measure of Reaction (R2) and therefore also a measure of the [RO 2 ] which is withdrawn from the PRP channels. 25 According to our working hypothesis P(O 3 ) is therefore also a measure of the RO 2 withdrawn from NPF and SOA mass formation. P(O 3 ) as measured at the onset of NPF should therefore provide hints on the impacts of RO 2 on J 7 . Introduction  Figure 10 shows a plot of ln(J 7 ) vs. ln(P(O 3 )). Linear regression analysis resulted in a slope of −1.9 ± 0.37 (R 2 = 0.87) i.e. in a slope near to −2.
The observed relationship indicated a mechanistic inverse coupling between J 7 and P(O 3 ). Considering the competition of reaction pathways (R2) and (R3), the simplest way of explaining the coupling would be a direct involvement of PRP in NPF. Since To confirm this consequence of strictly applying Eq. (1) we calculated the peroxy radical concentrations at the onset of NPF for all experiments relative to the experiment 15 with the highest NO x addition (BNR = 1.1, Fig. 9). Details of these calculations are given in the Supplement and we here summarize the basic result: at the onset of NPF, [RO 2 ] was indeed quite constant in all experiments (see Supplement Table S1 and Sect. S4 and S5). Within the uncertainties of such calculations [RO 2 ] did not vary by more than a factor of 2 and there were no systematic variations of [RO 2 ] with variations 20 of BNR.
Invariable [RO 2 ] at high variability of J 7 suggested that J 7 was not determined by P(PRP). We therefore conclude that Reaction (R3) as depicted in Fig. 1 is not the rate limiting step for new particle formation (see also Supplement Sect. S6). PRP formed from first generation peroxy radicals are not the vapors that directly aid NPF, probably 25 because they do not reach supersaturations required to grow nanometer particles. This may be because their vapor pressures are too high.
Evidence that the abundance of first generation PRP does not limit the rates of new particle formation is supported by the suppression of J 7 in the range 10 < BNR < 30 in 25847 absence of significant P(O 3 ). Here NPF was still suppressed by an order of magnitude suggesting that reaction pathways leading to NPF were substantially affected by NO. The same amount of NO was not efficient enough to generate measurable P(O 3 ) (see Fig. 5 and compare to Fig. 10). Inefficient P(O 3 ) indicates that rates of Reaction (R2) are low and PRP production remain high. Suppression of J 7 by 90 % without signifi-5 cant increases in rates of Reaction (R2) can only be explained by compounds that are involved in NPF but do not significantly contribute to photochemical ozone formation.
Higher generation molecules produced by PRP oxidation may be a comprehensible explanation for the observed suppression of J 7 without photochemical ozone formation. These RO 2 -like molecules must be abundant at much lower concentrations than first 10 generation RO 2 but capable of reacting with NO even faster than first generation RO 2 .
The assumption of higher generation RO 2 -like molecules being produced from further oxidation is supported by another finding: as shown by Kiendler-Scharr et al. (2009), NPF in dependence on [OH] exhibits power law dependence with an exponent of ∼ 4. This power law dependence indicates that at least four OH radicals 15 are consumed until new particles are formed from the oxidation of monoterpenes. The chemical system studied here is similar to that shown by Kiendler-Scharr et al. (2009). Hence, similar numbers of oxidation steps with OH are likely also needed here to produce the vapors that are able to grow nanometer particles. Formation of two first generation peroxy radicals requires less than 4 OH radicals, at maximum one for each 20 RO 2 . The requirement of more oxidation steps by OH radicals is in accordance with the requirement of further PRP oxidation to produce the vapors directly involved in NPF.
Further oxidation of first generation PRP may form the intermediate vapors that directly grow smaller particles to 7 nm particles. However, these higher generation vapors 25 are not yet identified. May be that some of the high molecular weight oxidation products shown to be produced during α-pinene oxidation (Ehn et al., 2012;Zhao et al., 2013;Kulmala et al., 2013) are such vapors or precursors of them. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Despite of the unknown identity of these compounds, our data hint to some of their chemical properties and to the character of the rate limiting step of NPF: -Essential gas phase precursors must be produced in a reaction chain induced by OH oxidation of BVOC: NPF was not observable without photolytic OH generation and from previous experiments we know that NPF increases with increasing 5 [BVOC] 0 (Mentel et al., 2009). The vapors therefore must have been produced in an OH initiated reaction from the BVOC.
-The vapors capable of condensing on nanometer particles are produced by precursors with chemical properties similar to those of peroxy radicals: suppression of NPF by NO shows that the concentrations of the RO 2 -like molecules are 10 efficiently suppressed with increasing NO. Affinity to NO and chemical properties similar to those of peroxy radicals is the only explanation for the observed behavior.
-Rate limiting step in NPF from monoterpene oxidation is a permutation reaction: Reaction (R2) consumes one NO molecule and the inverse squared dependence of NPF on P(O 3 ) shows that two molecules are involved in the rate limiting step. Either two of the RO 2 -like molecules react among each other or one RO 2 -like molecule reacts with a first generation peroxy radical. However, it is a permutation reaction that limits the formation of a 7 nm particle and not a simple oxidation by OH or O 3 . 20

Comparison with literature data
We found SOA mass yields in the range of 12 % which is substantially higher than the yields observed in Mentel et al. (2009)  pacts of NO x on SOA mass formation were reliably determinable because the dynamic range of [NO x ] 0 spanned more than 2 orders of magnitude.
There are several studies on the impact of NO x on particle mass formation (e.g. Pandis et al., 1991;Martin-Reviejo and Wirtz, 2005;Song et al., 2005;Presto et al., 2005b;Ng et al., 2007a, b;Kroll et al., 2006;Zhang et al., 2006). We restrict the comparison of our data with studies on BVOC. Table 2 gives examples for results obtained with respect to mass yields from BVOC oxidation in dependence of NO x .
The BVOC mix emitted from the set of plants investigated here predominantly comprised of monoterpenes and we studied photooxidation without seed aerosol. Appropriate comparisons of our results are those to results of Pandis et al. (1991) at NO x ∼ 800 ppb, Kim et al.: BNR between ∼ 33 and ∼ 6). All these studies found that SOA yields were lower at low BNR than at high BNR. This is consistent with our results. Pandis et al. report increasing yields with BNR values from 1 to 15. For BNR > 15 they report decreasing yields with increasing BNR. Such a decrease at high BNR was not found here. But for BNR < 15 our results are in qualitative agreement to those of Pandis 25 et al. (1991). Our data regarding impacts of NO x on SOA mass formation are nearly identical to those of Presto et al. (2005b). For the BVOC mix investigated here, NO x addition had no significant impact on SOA yield as long as BNR was above 10. Only for the two data points obtained at BNR < 3 suppression of SOA yield was obvious.
Decrease in SOA mass formation with decreasing BNR is attributed to the loss of peroxy radicals and the formation of organic nitrates (Reaction R2b). In case of monoterpene-or isoprene oxidation, organic nitrates produced in Reaction (R2b) are expected to have higher volatility than products formed in the absence of NO x (e.g. Presto et al., 2005b;Kroll et al., 2006). Thus, when more organic nitrates are formed, 5 SOA formation should be lower than at low NO x conditions.
In principle the routes via organic nitrates are possible. However, the degree of suppression in our experiments was substantially higher than can be explained by formation of organic nitrates. Of course organic nitrate formation is increasingly favored over PRP formation with increasing [NO]. But this holds only as long as Adding more NO can then not produce more organic nitrates and cannot produce more ozone. For a given BVOC mixture the branching ratio of organic nitrate formation from reactions of alkylperoxy radicals with NO is fixed. Assuming that the data for α-pinene, β-pinene and limonene in the master chemical mechanism (MCM3.2) are applicable 15 to our BVOC mix, we expect that about 25 % of the peroxy radicals can form organic nitrates via (R2b), the other 75 % form O 3 via (R2a). Scavenging of about 25 % of the intermediates involved in SOA mass formation cannot lead to a drop of SOA mass formation by more than an order of magnitude even if organic nitrates would not at all contribute to particle mass formation. However, an order of magnitude reduction in 20 PM MAX was observed here for the highest NO x conditions. The strong drop in PM MAX must have other reasons. Our measurements were made without seed aerosols and thus particle mass could not be accumulated without NPF before. Inhibition of NPF causes limitations in particle number density, particle surface, and particle volume precluding efficient condensation 25 or solution of non-nucleating vapors (e.g. Kerminen et al., 2000). The observed suppression of PM MAX at low BNR (see Fig. 7 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | seed aerosol (e.g. Kroll et al., 2006;Ng et al., 2007b). We therefore suppose that NO x also affects SOA mass formation.

Impact of first generation RO 2 on SOA yield
So far we cannot distinguish between NO or NO 2 as the NO x component causing the reduction of SOA mass yields. In accordance to most of the other published data we 5 assume that NO is responsible for suppression of PM MAX . If so, changes in SOA mass formation by NO x addition might be caused by impacts of NO on first generation [RO 2 ]. As shown above, [RO 2 ] was quite insensitive to [NO x ] 0 and we concluded that formation rates of first generation PRP are not rate limiting for NPF. Similarly we can conclude that formation rates of first generation PRP are also not rate limiting for SOA mass for-10 mation as SOA mass was formed timely after NPF appeared, i.e. when [NO x ] [NO x ] 0 in the high NO x cases. Therefore, [RO 2 ] and P(PRP) were similar in all cases when particle mass was formed. Again, presuming that particle mass formation is affected by NO x , further PRP oxidation seems to be required to produce the condensing vapors aiding SOA mass formation. As in case of NPF, the highly oxidized compounds 15 detected by Ehn et al. (2012) and Kulmala et al. (2013) are candidates of vapors aiding SOA mass formation.
As obvious from the systematic change of J 7 /PM MAX (Table 1), chemical properties of vapors aiding mass formation is somewhat different from that of intermediates aiding NPF. If NO is the compound causing suppression of PM MAX , suppression of J 7 at 20 constant PM MAX indicates that the vapors aiding NPF have higher affinity to NO than vapors aiding mass formation. Higher oxidative stage of the former might be a reasonable explanation for this.

Summary and conclusion
We found that NO acts as a switch between NPF and photochemical ozone formation, suggesting a mechanistic connection between both processes. However, the behavior of NPF was too complex for a simplistic chemical mechanism that includes first generation oxidation products only. Our data indicate that higher generation intermediates are 5 involved in NPF. These intermediates have similar chemical properties as peroxy radicals. They react with NO and they react among each other in permutation reactions. Permutation reactions are supposed here to be rate limiting for NPF. This means that the formation rates of intermediates with vapor pressures low enough to allow condensation on nanoparticles are controlled by rates of permutation reactions. At least one 10 of the reactants must be a RO 2 -like molecule produced in a reaction chain as a higher generation oxidation product.
From previous studies, we know that traces of H 2 SO 4 in a range of 10 5 -10 6 cm −3 are present in our system and H 2 SO 4 might have acted as a nucleating agent also in the experiments described here. Nevertheless, the abundance of volatile organic 15 compounds was necessary for particles to reach diameters of 7 nm and basic principles of RO 2 chemistry were suitable to explain the dynamic behavior of the chemical system including NPF. cleation and growth on high-molecular-weight gas-phase products during ozonolysis of αpinene, Atmos. Chem. Phys., 13, 7631-7644, doi:10.5194/acp-13-7631-2013Phys., 13, 7631-7644, doi:10.5194/acp-13-7631- , 2013