Atmospheric Chemistry and Physics Isoprene Photooxidation: New Insights into the Production of Acids and Organic Nitrates

We describe a nearly explicit chemical mechanism for isoprene photooxidation guided by chamber studies that include time-resolved observation of an extensive suite of volatile compounds. We provide new constraints on the chemistry of the poorly-understood isoprene δ-hydroxy channels, which account for more than one third of the total isoprene carbon flux and a larger fraction of the nitrate yields. We show that the cis branch dominates the chemistry of the δ-hydroxy channel with less than 5% of the carbon following the trans branch. The modelled yield of iso-prene nitrates is 12±3% with a large difference between the δ and β branches. The oxidation of these nitrates releases about 50% of the NO x. Methacrolein nitrates (modelled yield 15±3% from methacrolein) and methylvinylketone nitrates (modelled yield 11±3% yield from methylvinylke-tone) are also observed. Propanone nitrate, produced with a yield of 1% from isoprene, appears to be the longest-lived nitrate formed in the total oxidation of isoprene. We find a large molar yield of formic acid and suggest a novel mechanism leading to its formation from the organic nitrates. Finally, the most important features of this mechanism are summarized in a condensed scheme appropriate for use in global chemical transport models.


Introduction
Isoprene (2-methyl-1,3-butadiene, C 5 H 8 ) is a short-lived compound (τ 1/2 =1−2 hours) emitted by many deciduous trees during daylight hours.Between 0.5% and 2% of the carbon fixed by isoprene emitting plants is released to the atmosphere as isoprene (Harley et al., 1999), a flux accounting for about one third of the total anthropogenic and natural volatile organic compounds (VOC) emissions (Guenther et al., 2006).Isoprene plays a crucial role in determining the oxidative chemistry of the troposphere.Ozone levels in urban as well as in rural sites are impacted by the sequestration and transport of NO x via formation of isoprene nitrates (Horowitz et al., 1998) and various isoprene-derived peroxyacylnitrates.Moreover, field (Claeys et al., 2004) and chamber studies (Kroll et al., 2006;Surratt et al., 2006) have recently shown that compounds formed in isoprene photooxidation, such as methylglyceric acid or methylerythritol are ubiquitous in aerosol particles and may contribute significantly to the aerosol global burden (Henze and Seinfeld, 2006;van Donkelaar et al., 2007).
In the light of the potential for significant change in isoprene emissions due to climate and land use changes (Shallcross and Monks, 2000), studies have been made to predict the impact of altered isoprene emissions on tropospheric ozone (Sanderson et al., 2003;Wiedinmyer et al., 2006).von Kuhlmann et al. (2004) and Fiore et al. (2005) note, however, that quantifying this impact is difficult due to uncertainties regarding: 1) the dependence of isoprene emissions on temperature (Harley et al., 2004) and CO 2 concentration (Rosenstiel et al., 2003); and 2) the isoprene photooxidation Published by Copernicus Publications on behalf of the European Geosciences Union.
F. Paulot et al.: Isoprene photooxidation: new insights into the production of acids and organic nitrates scheme, especially the yields and fates of isoprene nitrates.
In this study, we use anion chemical ionization mass spectrometry (CIMS) to monitor the photooxidation of isoprene.This technique greatly expands the range of compounds that can be observed during the photooxidation of isoprene and other hydrocarbons (Ng et al., 2008).Quantitative interpretation is challenging, however, because 1) calibration standards are not available for many of the compounds identified and 2) mass analogs (compounds having the same mass) are not differentiated.Therefore the iterative development of a detailed mechanism is used to analyze the different signals and derive branching ratios and yields for the compounds identified.
First, we briefly describe the experiment emphasizing the calibration of CIMS measurements.Next, we report and identify the largest signals monitored by CIMS.We then discuss how these signals help constrain the development of the model emphasizing the δ-hydroxy channels, the organic nitrate yield and fate, as well as some routes to organic acids.Finally we discuss the potential implications of our findings for tropospheric chemistry and present a reduced mechanism suitable for inclusion in chemical transport model.

Experimental setting
The data of the present study were collected in the 28 m 3 Caltech atmospheric chamber, in an experiment similar to those described by Kroll et al. (2006).Initial concentrations of isoprene, NO and H 2 O 2 were 94 ppbv, 500 ppbv and 2.1 ppmv.The photolysis of H 2 O 2 constitutes the primary OH source in the experiment.NO was added prior to isoprene so that the chamber was initially ozone free.The initial relative humidity was less than 6% and is assumed to be constant in this study.The temperature increased by about 5 degrees in the first one hundred minutes and remained constant thereafter at 296.5 K .To simplify modelling, we consider this temperature to hold during the whole experiment.
Isoprene decay was monitored using GC-FID.Ozone was measured by UV absorption (Horiba) and NO and NO 2 (after conversion to NO) by chemiluminescence.The size distribution and the volume concentration of secondary organic aerosol (SOA) were measured using a differential mobility analyzer (DMA,TSI 3760).Further details are available in Kroll et al. (2006).

CIMS
Gas-phase products were monitored using a novel CIMS technique (Crounse et al., 2006) with CF 3 O − as the reagent anion.Non reactive with ozone, carbon dioxide and dioxygen (Huey et al., 1996), CF 3 O − is a versatile reagent ion suitable for the study of many oxygenated compounds.In general, two primary ionization pathways are observed: A minor ionization pathway is observed for certain compounds: In Reaction (R3), CF 2 O is incorporated into the original neutral molecule.Reaction (R3) has been observed for peroxynitric acid (PNA) and for peroxy acetic acid (PAA).While Reaction (R3) is not the major ionization pathway, in several cases it is useful for distinguishing certain mass analogs.The dominant ionization pathway for an analyte depends mostly on the acidity (or fluoride affinity) of the neutral species (Amelynck et al., 2000;Crounse et al., 2006).Highly acidic compounds, such as nitric acid, only form the transfer product ion through Reaction (R1) while hydrogen peroxide and methylhydrogen peroxide (MHP) form only the cluster product ions through Reaction (R2).Species with intermediate acidity (e.g.formic and acetic acids) form both the transfer and cluster products.Most of the VOC measured in this study follow Reaction (R2).
In this study, air was drawn from the chamber through a perfluoroalkoxy Teflon line of 1.2 m length and 0.635 cm outer diameter (OD), at a rate of 10 standard liters per minute (slm), and then sub-sampled into the CIMS flow tube using a critical orifice made of glass.The orifice constrained the flow from the chamber into the CIMS to be 145 standard cubic centimeters per minute (sccm).Upon introduction to the CIMS flow tube, the chamber gas was diluted with 1760 sccm of UHP N 2 (99.999%) to a total pressure of 35 hPa, primarily to reduce the concentration of H 2 O 2 to manageable levels.The gas is expanded in a flow tube ( 17.8 cm, 2.54 cm OD Pyrex glass coated with a thin layer of Teflon (Fluoropel 801A, from Cytonix Corp.)) before reacted with a transverse ion beam of the reagent anion (Crounse et al., 2006, Fig. 1).
Mass scans were conducted using a quadrupole mass spectrometer from m/z=18 to m/z=275 dwelling on each mass for 1 s (giving a scan cycle of about 4 1 2 min).The mass scans were repeated throughout the duration of the experiment (17 h).Zero scans were conducted periodically throughout the experiment by overfilling the critical orifice on the high pressure (chamber) side with UHP N 2 .In addition to providing instrumental backgrounds, the temporal response of the zero scans give insight into the strength of the interaction of the measured compounds with the equipment walls.
The instrumental background signals for most of the large molecular weight products produced in isoprene oxidation are very small (after the instrument has been sampling clean zero air for an extended period of time), which suggests that variations in instrumental background over the course   of the experiment are not important for these signals.Compounds with a smaller molecular weight (e.g.formic and acetic acids) do have instrumental backgrounds, but the level of the instrumental background is small relative to the signal generated in the isoprene oxidation experiment (more than 10 times smaller), so ignoring instrumental background changes over the course of the experiment for these species does not introduce a substantial error.

Calibration
The concentration of a compound X, whose product ion is detected at m/z=p, is calculated through: where Signal(m/z=p) is the normalized signal associated with X (cf.Appendix B1) and c X is the calibration constant for the compound X in ppbv −1 .In many cases, no standard is readily available and no experimental determination of c X can be made.In such cases, we assume that c X is related to the thermal capture rate (k X ) and the binding energy of the cluster.k X is estimated from the Langevin-Gioumousis-Stevenson-based collision rate.We use the empirical approach developed by Su and Chesnavich (1982) to calculate k X from the dipole moment (µ X ) and the polarizability (α X ) of X.
In the absence of experimental determinations of µ X and α X , we use quantum chemical calculations to estimate them.The lowest energy conformers of the molecules are found with the conformer searches method within the Spartan06 quantum package (Wavefunction Inc., 2006) at the B3LYP/6-31G(d) level of theory.The dipole moments and static polarizabilities are then calculated for the optimized geometries at the same level of theory.When a molecule has several low energy conformers, a thermally-weighted average of their reaction rate coefficients is used for k X (cf.Appendix B2 for details).
The sensitivity of the instrument to X also depends on the binding energy between X and the reagent anion as well as the nature of the reagent anion.In the presence of abundant ligands (L) such as water or hydrogen peroxide, the sensitivity of the CIMS to some VOC is modified through two different processes: 1) CF 3 O − •L may react faster with X than the bare CF 3 O − anion because ligand exchange reactions can stabilize the product ion to a greater extent; 2) the cluster CF 3 O − •L may be sufficiently stable leading to a lower sensitivity at higher L mixing ratio due to ligand exchange: For example, Crounse et al. (2006) reported that the sensitivity to methylhydroperoxide (MHP) decreases with the water vapor mixing ratio due to Reaction (R4).
In general large molecules featuring several functional groups (peroxide, nitrooxy, alcohol, carbonyl) exhibit only a weak dependence on the amount of water.Therefore, we neglect the binding energy effect in this study and take: where k HNO 3 =1.93×10−9 cm 3 molecule −1 s −1 is calculated using the experimental dipole and polarizability of nitric acid and c HNO 3 is the sensitivity to nitric acid for typical conditions where the flow tube was operated (water vapor mixing ratio=150 ppmv).HNO 3 is used as the calibration reference because of the weak dependence of the sensitivity with water and its thorough laboratory study (Huey et al., 1996;Amelynck et al., 2000;Crounse et al., 2006).When several compounds are observed at the same m/z, we report the signal calibrated with a reference calibration c ref and the modeled concentrations of each compound X i multiplied by s X i =c X i /c ref , so that one can compare measured signal with the prediction of the mechanism.The predicted concentration of a specific compound is therefore Finally, molecules such as isoprene, methacrolein (MACR), methylvinylketone (MVK) or peroxyacylnitrate (PAN) are not observed with our measurement technique despite their relatively large dipole moment.More generally, the method is not sensitive to simple aldehydes, alcohols, and ketones, presumably due to the low binding energy of these compounds with CF 3 O − .

Results
After the photolysis lights are turned on, isoprene decays with a half life of ∼20 min (Fig. A3).Several inorganic markers of the chemical evolution of the system can be monitored by CIMS (Table 1 and Fig. A4).Nitrous acid (transfer at m/z=66, cluster at m/z=132, Fig. A4), peaks after 50 min and has mostly disappeared after 150 min.HONO is associated with a very large concentration of NO which defines our first chemical regime (0≤t≤150 min).Given the high concentration of NO, little ozone is formed and isoprene photooxidation proceeds almost entirely through OH addition.Pernitric acid (PNA, transfer at m/z=98, cluster at m/z=164, Fig. A4) grows steadily peaking at ∼600 min.Given the sensitivity of PNA to the ratio HO 2 :NO 2 , the time when PNA reaches its maximum indicates the transition from a NO x (Regime 2: 150≤t≤600 min) to a HO x -dominated chemistry (Regime 3: t≥600 min).In this study, we focus on the NO xdominated chemistry, limiting our discussion to the first and second regimes.Studies of low NO x chemistry will follow in a separate manuscript.Nitric acid (transfer at m/z=82) grows steadily during the experiment to reach ∼430 ppb at the end of the experiment.We estimate dinitrogen pentoxide profile by removing the nitric acid contribution to the NO − 3 (m/z=62) temporal signal (Huey et al., 1996).The corrected signal exhibits a shape similar to PNA, peaking after ∼500 min at ∼ 3ppb.
In Table 1 and Figs. 1 to 15, we report the main signals measured by CIMS, the chemical formula of the associated compounds as well as their most likely identification using mechanistic considerations.To our knowledge, this is the first time that the temporal evolutions of isoprene nitrates (cluster at m/z=232, Fig. 9) and methacrolein/methylvinylketone nitrates (cluster at m/z=234, Fig. 13) have been monitored.We also observe the formation of small carboxilic acids such as formic and acetic acid, which can be clearly identified given that they undergo both Reactions (R1) and (R2).This specificity helps identify larger acidic compounds such as (Z)-2/3methyl-4-oxobut-2-enoic acid (MOBA): the signal recorded at m/z=199 (cluster) correlates with the associated transfer at m/z=133 (ρ=0.93 for the first 400 min).This also allows differentiation of certain mass analogs, e.g. the contribution of acetic acid cluster to m/z=145 can be removed using its experimental ratio between transfer and cluster.The residual is the cluster of glycolaldehyde (m/z=145).
Unfortunately most mass analogs, such as isoprene nitrates (m/z=232) are positional isomers, and thus cannot be specifically identified using this approach, thus precluding the derivation of their concentrations.To overcome this difficulty, a detailed mechanism has been developed iteratively using the constraints of organic and inorganic signals in association with previously identified mechanisms (Appendix A).

δ-hydroxy channels
Under chamber experimental conditions, isoprene photooxidation proceeds primarily through the addition of OH to the two double bonds (position 1, 2, 3 and 4, in Fig. 1).In the following we will denote the different channels by the couple (i, j), where i and j refer, respectively, to the carbon on which the HO and O 2 addition occurs.Besides β-hydroxy peroxy radicals (1, 2) and (4, 3), additions to positions 1 and 4 can lead, after addition of O 2 , to four δ-hydroxy peroxy radicals (Sprengnether et al., 2002), referred to as Z 1, 4 , E 1, 4 , Z 4, 1 , E 4, 1 .The branching ratio between these different channels remain uncertain (cf.Sect.C1).Here we use a combination of theoretical (Lei et al., 2000) and experi-mental results (Sprengnether et al. (2002) and this study) as constraints: As most studies of isoprene photooxidation have focused on the main decomposition channels (1, 2 and 4, 3) yielding MACR and MVK (Paulson et al., 1992;Sprengnether et al., 2002;Karl et al., 2006), the δ-hydroxy channels, which account for about 30% of the carbon and a large fraction of the organic nitrates, remain poorly constrained.A large number of products originating from the δ-hydroxy channels can be monitored by CIMS which motivates our emphasis on their chemistry.

Chemistry of the δ-hydroxy channels
Z 1,4 .The reaction of the peroxy radical with NO yields an alkoxy radical which undergoes a δ 5 1 isomerization (Atkinson, 1997;Park et al., 2004).The resulting β-hydroxy allyl radical can then react with O 2 and form a 1,4-hydroxycarbonyl, (2Z)-4-hydroxy-2methylbut-2-enal (HC5 Z(1,4)) detected as a cluster at m/z=185 (Fig. 2).The detection of its 13 C isotope at m/z=186 supports the attribution of the signal to HC5.Formation of glycolaldehyde (GLYC cluster at m/z=145, Fig. 3) and methylglyoxal (MGLYX) at this stage of the photooxidation have also been described by Dibble (2004a,b).This reaction is based on the stabilization of the alkoxy radical, reproduced below, through a double hydrogen bond which prevents its decomposition while enhancing a double hydrogen shift involving the hydrogen of the alcohol groups.
OH can add to the HC5 Z(1, 4) double bond (channels 2 and 3 in Fig. 1), abstract the aldehydic hydrogen (channel 4 ) or the hydrogen α to the alcohol (channel 1 ).Addition on position 2 is expected to yield GLYC and MGLYX.The signal detected at m/z=217 (Fig. 4) suggests the existence of channel 3 .In this pathway the alkoxy radical formed after addition of OH on position 3 is stabilized enough by a double hydrogen bond, so that reaction of O 2 becomes competitive with unimolecular decomposition.This mechanism yields 2,4-dihydroxy-2-methyl-3-oxobutanal (DHMOB (1, 4)).
The reaction of OH with HOPL is expected to yield an acylperoxy radical which can further react with NO 2 yielding a PAN-like compound, 3-hydroxy-oxo-peroxyacylnitrate (PAN 250).This compound may be associated with the signal monitored at m/z=250, although the identification is not certain.Signals originating from the E branch are much smaller than those originating from its Z counterpart.This suggests a large asymmetry between the E and Z δ-hydroxy channels.
Due to the similarity between the (4, 1) and (1, 4) branch, we only address the major differences.

Consequences
The observations of numerous compounds formed at different stages of δ-hydroxy pathways lead to several inferences about the general mechanism: Channel asymmetry.If an equal partitioning of the carbon is assumed between E 1,4 and Z 1,4 as suggested by the theoretical work of Dibble (2002), the concentrations of both HOPL and DHPN are greatly overestimated while the concentrations of HC5=HC5 Z(1, 4)+HC5 E(4, 1)+HC5 Z(4, 1) and its products (DHMOB, MOBA) are underestimated.Good agreement with the observations is obtained when, An additional piece of evidence suggesting that little flux of carbon occurs through E 1,4 is the low signal recorded at m/z=201, which should include 3-methylhydroxy-4-hydroxybutenal (MHBL) based on its structural similarity with HC5.
The asymmetry between Z/E isomers contradicts the conclusions drawn from quantum mechanical calculations (Dibble, 2002) as well as the assumption made by most kinetic models of isoprene photooxidation (Paulson and Seinfeld, 1992;Fan and Zhang, 2004).The discrepancy with quantum mechanical calculations may be related to a difference in the reaction of the cis/trans radical with O 2 .The radicals are formed with approximately 40 kcal/mol excess energy.The minimum isomerization barrier is estimated to be about 15 kcal/mol (Dibble, 2002).Therefore, assuming a collision stabilization of 100 cm −1 collision −1 , the radicals undergo nearly 100 collisions (20 with O 2 ) before they are cooled below the isomerization barrier.If, based on reported rate coefficients for R • +O 2 (Atkinson et al., 2006), one reaction among ten is assumed to be reactive, peroxyradicals are likely to be formed before the isomers are cooled below the isomerization barrier.Therefore, the equilibrium may be shifted if the reaction of the trans radical with O 2 is faster than the reaction of the cis radical.Measurements made with reduced partial pressure of O 2 could test this hypothesis.The cis and trans forms can also be interconverted latter in the photooxidation by the δ 5 1 isomerization.Therefore the observed discrepancy may also be attributed to the additional stability of the cis β-hydroxy allyl radical provided by the interaction between the alcohol groups.
Evidence for Dibble's mechanism.Both HACET (Fig. 7) and GLYC (Fig. 3) exhibit a very prompt source.To our knowledge, the mechanism proposed by Dibble (2004a,b) and reproduced in Sect.4.1.1 is the only mechanism able to yield both compounds after a single OH reaction.For GLYC, we set the branching ratio quenching:thermalization to 7:3 in good agreement with the theoretical estimate (Dibble, 2004b).The same branching ratio was applied to capture the prompt formation of HACET from the Z 4,1 branch.Theoretical considerations do not support such a large hydroxyacetone formation (Dibble, 2004b).The hydroxyacetone rate constant with OH is set to k OH =5.98×10 −12 cm 3 molecule −1 s −1 (Dillon et al., 2006) and the rate constant of glycolaldehyde with OH is set to k OH =8×10 −12 cm 3 molecule −1 s −1 (Karunanandan et al., 2007).HC5 chemistry.HC5 (Fig. 2) exhibits a very fast decay consistent with a reaction rate coefficient with OH similar to isoprene (1.0−1.2×10−10 cm 3 molecule −1 s −1 ).This estimate is consistent with the fastest rate recently derived by Baker et al. (2005) and ∼80% greater than the SAR estimate (k SAR OH =6.82×10 −11 cm 3 molecule −1 s −1 or 7.9×10 −11 cm 3 molecule −1 s −1 with the correction from Bethel et al., 2001;Papagni et al., 2001) .This discrepancy can be partly explained by the effect of the alcohol group α of the double bond which enhances the addition of OH (Papagni et al., 2001).The large measured yield of MOBA=MOBA Z(1, 4)+MOBA Z(4,1) (Fig. 5) also suggests that the abstraction of the aldehydic hydrogen (channel 4 ) is faster than predicted, possibly related to a long distance interaction between the alcohol group and the carbonyl group (Neeb, 2000).
In our model, 3-MF yield is set to 4.5% based on experimental results (Atkinson et al., 1989;Paulson et al., 1992) and formed from the parent alkoxy of HC5.As a result, 37% of the alkoxyradical formed in the Z 1, 4 and Z 4, 1 branches must decompose to 3-MF in order to match the experimental yield.We can not rule out 3-MF heterogeneous formation.We note, however, that if heterogeneous processes yield 3-MF, the calculated HC5 yield would be ∼20% higher.Moreover, the decay rate required to match HC5 profile would likely be faster than observed.3-MF formation mechanism has little impact on the conclusions of this paper but has significant consequences for atmospheric chemistry.Indeed if formed through heterogeneous processes, 3-MF yield is likely to be smaller than determined in atmospheric chambers.Further work is clearly required to quantify this issue and determine the products of 3-MF photooxidation.
We note, finally, that the observation of large yields for HC5 and 3-MF are consistent with an asymmetry between the E and Z branches.If the branching ratio E:Z were close to 1:1, the fraction of peroxy radical Z 1, 4 and Z 4, 1 required to decompose to 3-MF would be 62% and the yield of HC5 only 6.1% This is inconsistent with the yields previously reported (Baker et al., 2005).

Organic nitrates
The observation of the organic nitrates of isoprene and MVK/MACR as well as some of the products of their photooxidation (Figs. 8 to 13), provides constraints on the isoprene nitrate yields, their lifetimes and the amount of NO x recycled through the first stage of their photooxidation as well as their lifetimes.
PROPNN features a very prompt source, which requires a fast reaction rate coefficient of ISOPN(4, 1) with OH: k ISOPN(4,1) OH =9.5×10 −11 cm 3 molecule −1 s −1 .This is ∼45% faster than SAR and suggests an inadequate parameterization of the effects of nitroxy groups on the reactivity of the double bond (Neeb, 2000).No significant signal is observed at m/z=230, confirming that the abstraction of the hydrogen in α of the alcohol of ISOPN is negligible compared to addition on the double bond.SAR suggests that ISOPN(1, 4) and ISOPN(4, 1) are similarly short-lived with respect to OH.This is consistent with the prompt source of ETHLN (Fig. 10), a product of the oxidation of ISOPN (1, 4) (Fig. 8).The use of the primary nitrate photolysis rate (cf.A3) and SAR rate estimate for the reaction ETHLN + OH underpredicts its decay.To match the measured profile (Fig. 10), we take k OH =1×10 −11 cm 3 molecule −1 s −1 , three times faster than the SAR estimate.If the ETHNL photolysis rate is larger than estimated (J∼4×10 −7 s −1 using 1−C 4 H 9 ONO 2 ), ETHLN reaction rate coefficient with OH would be commensurately slower.
The yield of ETHLN is substantially overestimated if the yield of the reaction RC  unimolecular decomposition of the α-hydroxyperoxy radical from ISOPN(1, 4) yields formic acid, resolving this discrepancy.
The reaction of the isoprene nitrate with ozone is included for ISOPN (1, 2) and ISOPN (4, 3), because their long lifetimes allow them to encounter high concentrations of ozone in the chamber (Fig. A2).We do not observe the formation of 3 hydroxy-2-nitrooxy-2-methyl propanoic acid (no correlation between m/z=184 and m/z=250).Therefore, we use a simplified version of the ozonolysis products proposed by Giacopelli et al. (2005) assuming that this reaction yields only MACRN and MVKN.These reactions contribute significantly to the total yield of MVKN and MACRN in this experiment.
MACRN features an aldehyde group which suggests a much faster decomposition than MVKN.This is confirmed by the profile of hydroxyacetone which does not exhibit any significant late source.As a result, the third regime of the experiment is dominated by MVKN and its reaction rate coefficient with OH can be estimated: k MVKN OH =2.8×10 −12 cm 3 molecule −1 s −1 .The yield of MVKN can be constrained using GLYC profile since MVK is its major source: α MVK =(11±3)%.Applying the same approach to MACR/MACRN/HACET is more complicated since hydroxyacetone has many more sources than glycolaldehyde (Fig. 7).We find that a yield of MACRN of (15±3)% and a reaction rate coefficient with OH of 5×10 −11 cm 3 molecule −1 s −1 best match the peak time of m/z=234.These values are consistent with the study of Chuong and Stevens (2004).
The abstraction of the hydrogen α to the alcohol in MACRN, MVKN and MVKN (m) is expected to yield dicarbonyl nitrates.We expect CIMS to be sensitive to this class of compounds since we are able to measure compounds featuring a carbonyl in β of the nitrate group, such as PROPNN or ETHLN.Since m/z=232 only features early stages compounds, isoprene nitrates, there is no evidence for the formation of dicarbonyl nitrates.Recent theoretical studies (Peeters et al., 2001;Hermans et al., 2005) show that primary α-hydroxy-alkylperoxy radicals can be sufficiently stabilized to undergo reactions with NO and yield formic acid.Nevertheless the photooxidation of MVKN occurs in a mostly low-NO x environment which suggests that α-hydroxy-alkylperoxy radicals may undergo an intramolecular decomposition to yield a carboxilic acid and the nitrate radical.Such a reaction may involve a four or six-e − mechanism.
Finally, m/z=234 signal features a prompt source which can not be accounted for by MVK or MACR nor by the βhydroxy isoprene nitrates, which have a similar lifetime with respect to OH. Conversely ISOPN (1, 4) is very short-lived and a MVKN (m) yield of 10-15% enables to capture this feature (Fig. 13).(∼600 min), a molar yield of about (10±3)% is obtained (Fig. 14).
The measured profile of formic acid (Fig. 14) features the three characteristic chemical regimes of this experiment as described (cf.Sect.3).
First regime.A very early source of formic acid is noticeable (Fig. 14).Several observations suggest that formic acid may be formed from an intramolecular decomposition of ISOPN (1, 4) similar to the one described for MVKN earlier: 1) This source is absent from the experiments performed in the absence of NO x 2) the early formic acid profile correlates very well with the propanone nitrate which originates from ISOPN (4, 1) decomposition 3) ETHNL would be largely overevaluated in the absence of other decomposition channel for ISOPN (1, 4).Matching the ETHLN profile (Fig. 10) results in a branching ratio for formic acid to ETHLN of 3:1.ISOPN(2, 1) may yield acetic acid, but is not included since its contribution would be negligible.Bierbach et al. (1995) report 4-oxo-pentenal as the major product of the photooxidation of 2-methylfuran in the absence of NO x , while formic acid accounts for about 6%.We are unaware of any study of the photooxidation of 3-MF in the presence of NO x .Since the reaction rate coefficient of methylfuran with OH is similar to that derived for ISOPN (1, 4) and (4, 1), its photooxidation may contribute to the early sources of formic acid: Second regime.Butkovskaya et al. (2006a,b) report a formic acid yield from the photooxidation of GLYC (HAC) of 18% (respectively 7%).The formation of formic acid from the decomposition of MVKN described in Sect.4.2.3 also plays a minor role in this regime.

Third regime (not shown).
As NO x becomes limiting, hydroxymethyl hydroperoxide (HMHP) formation from the reaction of CH 2 OO with water is enhanced.HMHP can account for most of the late formation of formic acid through its reaction with OH and its photolysis.A large additional source is missing, however, in the mechanism.Heterogeneous decomposition of HMHP (Neeb et al., 1997) and aerosol processes (Walser et al., 2007) are likely to account for this missing source.An upper limit for the strength of the aerosol source can be estimated from the decrease of the aerosol volume −2.5 µm 3 /cm 3 which would represent a release of 2.6 ppbv C=O in the chamber assuming a density of 1.25 g/cm 3 (Kroll et al., 2006).Since the same phenomenon is observed for acetic acid (Fig. 15), most likely both HO x -dominated VOC oxidation as well as organic aerosol oxidation are needed to explain the observed increase in formic and acetic acid in the third regime.

Acetic acid
The production of acetic acid (Fig. 15) occurs primarily through the oxidation of hydroxyacetone as described by Butkovskaya et al. (2006b).Additional routes include direct formation from CH 3 CHO + OH (Cameron et al., 2002) as well as CH 3 C(O)OO + HO 2 following Reaction (R11).
Two additional minor routes are hypothesized: 1) decomposition of MACRN(m), 2) decomposition of DHMOB (1, 4) (m/z=217), inspired by the mechanism proposed by Butkovskaya et al. (2006a).Following their analysis, we assume a 37% acetic molar yield, the remaining falling apart as CO 2 and hydroxybutane-2,3-dione (m/z=187  and 199 in Sect 4.1 can react with OH and ozone under the chamber conditions.The slow decay of the signal suggests that its reactivity is dominated by ozone in the chamber and that the acid group significantly hinders the addition of OH onto the double bond.A good match is obtained by setting k OH equal to 3×10 −12 cm 3 molecule −1 s −1 (F COOH =0.1 in terms of SAR) and k O 3 =2×10 −17 cm 3 molecule −1 s −1 .Little signal is observed at m/z=93 which suggests no or minor formation of oxoacetic acid.This suggests that the reaction with OH does not conserve the acid group, which is likely lost as CO 2 .Pyruvic acid can be expected to be a major product of MOBA ozonolysis.

Pyruvic acid
Pyruvic acid is a precursor for glyoxylic and oxalic acids, two carboxylic acids detected in the aerosol phase (Carlton et al., 2006).It is observed as a cluster at m/z=173 in association with HOPL, a product of DHPL photooxidation.Due to its stickiness to the walls of the flow tube, the theoretical calibration is expected to largely overestimate our sensitivity to this compound.The yield of pyruvic acid after 600 min is 2±1%.
Qualitatively, the major sources of pyruvic acid are expected to include (heterogeneous) hydrolysis of the Criegee intermediate produced in the ozonolysis of MVK and MOBA as well as the decomposition of MVKN(m) following the scheme presented in Sect.4.2.3.

Fate of organic nitrogen
The formation of organic nitrates, and more specifically isoprene nitrates, play an important role in determining the amount of NO x and thus ozone production in many environments.
The observation of isoprene nitrate clusters with CF 3 O − as well as some of the products of their photooxidation, provides constraints on the isoprene nitrate yields, the amount of NO x recycled through the first stage of their photooxidation, and their lifetimes (Table 2).These three parameters are necessary to accurately assess the influence of isoprene photooxidation on atmospheric chemistry.2007), 7% at 130 hPa.Using experimental yields collected for compounds similar to isoprene, Giacopelli et al. (2005) estimated the nitrate yield of the β and δ-hydroxy isoprene nitrates to be respectively 5.5% for the former and 15% for the latter, for an overall yield of 8.6%.Since organic nitrates sequester NO x , such a large variation in the estimated yields has profound implications to the assessments of ozone production caused by isoprene photooxidation (von Kuhlmann et al., 2004;Fiore et al., 2005;Horowitz et al., 2007).
We report a yield of (11.7±3)% with a large discrepancy between the yields of the nitrates originating from δ ( 24%) and β-hydroxy channels ( 6.7%).We emphasize, however, that we derived a total yield rather than specific branching ratios so that the specific organic nitrate yields are affected by the choice of the initial β:δ-hydroxy channel ratio (cf.Sect.C1).Nevertheless, the discrepancy between δ-hydroxy channels and β-hydroxy channels is a reliable feature, with the δ-hydroxy isoprene nitrates accounting for about 60% of the total isoprene nitrate yield.Giacopelli et al. (2005) suggested this behaviour previously using measurement collected for similar compounds.
Lifetime.The efficiency of both NO x transport and removal through organic nitrates is related to their lifetimes.The transport of isoprene-nitrates and their alkylnitrate degradation products is of special importance since it is thought to be a major source of NO x in rural areas (Horowitz et al., 1998).In this NO xlimited environment, the release of NO x through their decomposition would greatly influence O 3 production.In the absence of experimental data, Giacopelli et al. (2005) estimated using Kwok's SAR that the δ-hydroxy isoprene nitrates should be significantly shorter-lived than the β-hydroxy isoprene nitrates.With the help of the propanone nitrate profile, we can experimentally confirm this discrepancy (Table 2).With [OH] =10 6 molecule −1 cm −3 , the photochemical lifetime with OH of the δ-hydroxy isoprene nitrates (respectively the β-hydroxy isoprene nitrate) is Horowitz et al. (2007) show that the deposition of isoprene nitrates is likely to be dominated by dry processes and that τ As a result, the fate of the δ-hydroxy isoprene nitrates is likely to be dominated by their reactivity with OH and possibly O 3 similar to isoprene while other processes such as dry deposition and reaction with NO 3 must be taken into account for proper modeling of the β-hydroxy isoprene nitrates.Therefore, the latter are likely to have greater influence on tropospheric chemistry.
The large difference in the lifetime of the organic nitrates formed in the δ-hydroxy and β-hydroxy channels may explain some of the spread in the reported yields and NO x recyclings.Studies focusing on the very first step of isoprene photooxidation (e.g.Sprengnether et al., 2002) tend to report the highest nitrate yield, suggesting that the short-lived δ-hydroxy isoprene nitrates may have been underestimated in some previous experiments (see inset of Fig. 9).The same argument may also explain the observations of a greater variety of isoprene nitrates in laboratory experiments than in the field (Giacopelli et al., 2005).
Recycling.The efficiency of the NO x sequestration depends on the fate of the isoprene nitrates and more specifically on how much NO x is released in their subsequent photooxidation.NO x -recycling is defined as the difference between the NO x released by the reaction and the NO consumed.As a result, since ISOPN(2, 1) oxidation does not yield any NO 2 , its recycling is negative due to the formation of dinitrates (Fig. 8 and Table 2), which we have observed at m/z=311 in another experiment.Horowitz et al. (2007) obtain the best agreement with the boundary layer data when 40% of the NO x is recycled with a low nitrate yield (4%).We find a NO x recycling of (55%±10)% by the isoprene peroxy radicals consistent with Horowitz et al. (2007) conclusion despite our very different yields.As highlighted in the inset of (Fig. 9), this may be related to the wrong estimation of isoprene nitrate yield due to the short lifetime of the δ-hydroxy channels.
The photooxidation of isoprene leads to the formation of other significant organic nitrates MVKN, MACRN, PROPNN ( 1%) and ETHLN ( 1%).All these compounds are substantially longer-lived than isoprene nitrates and therefore are more likely to influence the NO x -balance on a larger scale (assuming a similar deposition velocity).The formation of PROPNN and MVKN appear especially important as their photochemical sinks are very slow: τ MVKN OH 100 h and τ PROPNN OH >200 h.Therefore, they may constitute important pathways for NO x transport as well as significant NO x -sinks through deposition.They also can contribute to the growing source of atmospheric nitrogen to the open ocean (Duce et al., 2008).
In contrast to ISOPN, MACRN and MVKN release most of their NO x in the course of their decomposition, possibly through the formation of formic and pyruvic acids.These organic nitrate channels may contribute significantly to the missing source of small carboxylic acid in the free troposphere.

Acids
Small carboxilic acids are ubiquitous in the atmosphere both in the gas-phase and in the aqueous phase, playing an important role in rain acidity and cloud reactions (Chebbi and Carlier, 1996).
The photooxidation of isoprene under high NO x produces substantial amounts of formic (yield (10±3)% after 600 min) and acetic acids (yield (3±1)% after 600 min).Acetic and formic acids are highly correlated after the first 150 minutes (ρ=0.988),since their main source, hydroxyacetone for acetic acid and glycolaldehyde for formic acid, share a similar origin and lifetimes (Figs. 14 and 15).We find A strong correlation between formic and acetic acids has been observed previously over Amazonia (Andreae et al., 1988) and Virginia (Talbot et al., 1995).In most large scale chemical models, these compounds originate primarily from biomass burning and to a lesser extent from ozonolysis of alkenes.Since the main source of both acids in the chamber is unlikely to result from the ozonolysis of the alkenes, our study shows that additional channels for their formation should be included.The main identified sources (hydroxyacetone, glycolaldehyde, organic nitrates) are much longerlived than the ones currently included in global model which may help resolve part of the discrepancy between models (Jacob and Wofsy, 1988) and atmospheric observations (Andreae et al., 1988;Talbot et al., 1990).
Finally the identification of MOBA, a five carbon acid, could be important for aerosols as its vapor pressure and the vapor pressures of the products of its photooxidation are expected to be low.

Development of a reduced mechanism
The new constraints derived in this study primarly originate from our observation of the chemistry of the δ-hydroxy channels.In particular, we have shown that these channels account for a large fraction of the isoprene nitrates and small carboxylic acids, whose role is important on a global scale.Most of the reduced isoprene photooxidation mechanism implemented in chemical transport models, e.g.MOZART (Pfister et al., 2008), do not adequatly account for the chemistry of the δ-hydroxy channels, thus impeding proper modeling of the consequences of isoprene photooxidation on troposheric chemistry.
The detailed chemical mechanism described in this study is too complex to be included in large scale atmospheric chemistry simulations.To aid in such investigations, we propose that a few modifications of these simplified mechanisms be implemented (Table 3).While adding little complexity and maintaining carbon balance, these few changes describe more accurately the formation and fate of nitrates as well as the yield of carboxylic acid.In particular, the long life time of PROPNN and MVKN enables transport of organic nitrates over long distances.
In our reduced mechanism, we neglect the E δ-hydroxy channel branch ( 5% of the carbon) as well as the non-Dibble branch yielding MPDL and OBL (Fig. 1).We have only included the formation of organic nitrates which were directly constrained in this study (ISOPN, PROPNN, ETHNL).The yield of minor organic nitrates can be derived using the number of carbons of the parent peroxy radical, a common approach in most chemical transport models.We introduce a generic four carbon hydrocarbon, HC4, to account for the decomposition of MOBA.This study does not constrain the fate of 3-methylfuran and the reaction of isoprene nitrates with ozone; more theoretical and experimental work is required.

Conclusions
A substantial fraction of the terrestrial Northern Hemisphere is characterized by conditions in which the fate of isoprene peroxy radicals is dominated by reactions with NO.Chameides et al. (1988) demonstrated that they play a major role in the formation of ozone in urban areas.This study complements previous investigations of isoprene photooxidation mechanism by focusing on the δ-hydroxy channels, whose chemistry is not adequately represented in chemical transport models.We focus on the large yields of small carboxylic acids and propose new constraints for the yield and the fate of organic nitrates.Both constitute outstanding uncertainties in the photooxidation of isoprene, impeding proper model-ing of tropospheric chemistry on a global scale.To aid in the development of improved simulations of this chemistry, we propose simple modifications of current condensed mechanism which maintains carbon balance and accounts for the new constraints and mechanisms identified in this study (Table 3).

A1 VOC chemistry
Except as noted below, we use the known rate coefficients of bimolecular and termolecular reactions as tabulated in IU-PAC (Atkinson et al., 2004(Atkinson et al., , 2006) ) and JPL (Sander, 2006) reports.

A1.1 HO
Reactions of HO with VOC are limited to its addition on a double bond and the abstraction of the aldehydic hydrogen and the hydrogen in α to an alcohol, i.e. the abstraction of hydrogens from alcohols is neglected.For the addition of HO onto double bonds, in the absence of data or previous Atmos.Chem.Phys., 9, 1479-1501, 2009 www.atmos-chem-phys.net/9/1479/2009/information enabling differentiation between the two carbons, we assume that the reaction occurs only on the most favorable location based on steric considerations.A structureactivity relationship (SAR) method is used to determine unknown reaction rate coefficients (Kwok and Atkinson, 1995).
Following the studies of Orlando and Tyndall (2001) and Méreau et al. (2001), acylradicals are assumed to decompose promptly when the alkyl group features a carbonyl or an alcohol (with R secondary or tertiary) in β to the carbonyl: We also assume that acylradicals featuring a nitrooxy group in β to the carbonyl undergo unimolecular decomposition.In all other cases, the acyl radical is assumed to add O 2 to yield the associated peroxy radical.
In case of resonance, the branching between the addition of O 2 on the carbon in α of alcohol (denoted a in Fig. 1) or in γ ( b ) is unknown.Addition on a features a more stable double bond as well as a kinetically favored radical.In the mechanism, we set the branching a : b to 65%:35%.Furthermore we note that HOPL does not exhibit any early source, suggesting that the yield of HMPL is negligible.Indeed, its formation appears unfavorable on both a thermodynamic (the double bond is less substituted) and a kinetic (formation of a secondary radical) basis.Thus, the branching ratio is set to

A1.2 Ozone
Ozone reacts with alkenes via the formation of a molozonide, quickly followed by its decomposition into a carbonyl and a Criegee intermediate.Assuming a generic rate constant for the reaction of alkenes with ozone, 10 −17 cm 3 molecule −1 s −1 , the reaction of an alkene with ozone is included if τ HO > τ O 3 10 , where τ HO and τ O 3 are the lifetimes of the alkenes with respect to HO and O 3 respectively.
Ozone reactions are included for isoprene, MACR and MVK, following IUPAC recommendations.

A1.3 NO 3
Reactions of NO 3 with alkenes and aldehydes have been neglected, since 1 throughout the experiment.
Carter's parameterization is used to compute the alkyl nitrate yield (Carter and Atkinson, 1989;Arey et al., 2001): .43, Y 298 0 =γ e βn , where n is the number of carbons in the molecule.The parameter m is set to 0.4, 1.0 and 0.3 for primary, secondary and tertiary nitrates, respectively (Arey et al., 2001).
For β-hydroxy peroxy radicals, α is divided by two to account for the effect of the hydroxy group as highlighted by O'Brien et al. (1998).
For acylnitrates, the yield is set to the alkyl tertiary nitrate yield, providing it does not exceed 4%.

A2.2 NO 2
NO 2 reacts with peroxy acyl radicals to yield peroxyacylnitrate-like compounds, which decompose thermally or photolytically: The rates of formation and decomposition of methyl peroxyacylnitrates (MPAN) are used for all PAN-like compounds except PAN itself.Most PAN-like compounds except PAN itself have other reactive groups (aldehyde, primary or a secondary alcohol, double bond) causing their major sink to be reaction with HO.A2.3 NO 3 NO 3 reacts with peroxy radicals through The rate coefficient is set to 2.3×10 −12 cm 3 molecule −1 s −1 independent of both the temperature and the peroxyradical.A2.4 HO 2 and peroxy radicals HO 2 reacts with peroxy radicals through four different channels: Reaction (R13) has only been observed for compounds such as RCH 2 OCH 2 OO and is not considered in this study.Acyl peroxides are assumed to react through Reaction (R10), Reaction (R11) and Reaction (R12) with a branching ratio 0.4:0.2:0.4 (Hasson et al., 2004;Jenkin et al., 2007).Acetonylperoxy radicals have also been shown to react through channels Reaction (R10) and Reaction (R12) with a branching ratio 1:2 (Hasson et al., 2004).The other alkylperoxy are assumed to react through Reaction (R10) only.
RO 2 +RO 2 reactions are neglected in this study.In the early stages of isoprene photooxidation the chemistry of peroxyradicals is entirely dominated by NO.At the end of the experiment, peroxy radical chemistry is dominated by HO 2 , which concentration is high enough so that RO 2 + RO 2 reactions can be safely neglected.

A3 Photolysis
The photolysis rate of a compound i is computed via: The effective light flux I e is computed using an experimental determination of J HONO and a spectrum of the lamp output made every nanometer (LI-COR LI1800 λ 1 =300 nm, λ 2 =600 nm).σ HONO is scaled using the oscillator strength recently reported by Wall et al. (2006).This gives J HOOH = 3.1×10 −6 s −1 .The photolysis of compounds with unknown absorption cross sections is estimated from the known photolysis rate constants of similar compounds.The photolysis of organic nitrates is assumed to yield only RO + NO 2 .For primary organic nitrate, the photolysis rate is taken from 1−C 4 H 9 ONO 2 , for secondary organic nitrates from 2−C 4 H 9 ONO 2 and for tertiary nitrates from tertbutylnitrate (Roberts and Fajer, 1989;Atkinson et al., 2006).

A4 Fate of the alkoxy radicals
Alkoxy radicals can react following three different pathways: Since the isomerization reaction, Reaction (R16), requires at least four carbons (Atkinson, 1997), it occurs only in the first stages of isoprene photooxidation, when major products retain five carbons.In the case of isoprene, isomerization (Reaction (R16)) is faster than decomposition (Reaction (R14)) and reaction with O 2 (Reaction (R15)).Alkoxy radicals which cannot undergo Reaction (R16) are assumed to decompose through Reaction (R14), i.e. their reaction with O 2 (Reaction (R15)) is generally neglected except for a few cases detailed in the discussion section.
Generally the decomposition of an alkoxy radical can occur through different channels, whose branching ratios (Y i ) are estimated using their respective activation energies, E b is calculated using the generalized structure-activity relationship developed by Peeters et al. (2004).

A5 Skill of the model
The evolution of the modelled chemical system can be examined through its instantaneous speciation (Figs.A1).Furthermore, given that the oxidation of CO by HO is negligible in the time scale of experiment, we define the chemical speed of the system, V, as

First regime (0<t<150 min
).This regime is characterized by a large supply of NO, as well as very reactive compounds featuring a double bond.V reaches a maximum after a few minutes at 0.7 ppv(C)/min.O 3 and PNA are very low in this regime, underlying a chemistry dominated by NO.The organic nitrate concentration reaches its maximum at the end of this regime.The reduction in [HO] corresponds to an increase of [NO 2 ] leading to the formation of nitric acid.
Second regime 150<t<550 min.This regime is characterized by a very stable V (0.5 ppbv(C)/min) with a chemistry dominated by aldehydes.HO recycling though HO 2 + NO is less efficient than in the first regime due to the abundance of O 3 which favors the formation of PAN.Nevertheless, the reduction in the chemical speed due to the transition from "double bond dominated" to "aldehyde dominated" reduces HO sinks which ultimately leads to a slow increase in HO, leveling off when PNA peaks, i.e. when the NO x is titrated.
Third regime 550 <t<1000 min.After the PNA peak, the chemistry is dominated by HO 2 , as evidenced by the formation of peracetic acid (PAA cluster at m/z=161) and methylhydroperoxide (MHP cluster m/z=133).Low-reactivity compounds such as ketones or long-lived nitrates dominate the chamber composition.Despite the almost constant HO, the chemical speed drops significantly to 0.1 ppbv(C)/min.
These three distinct chemical regimes are consistent with the ones derived using the experimental profiles of HONO and PNA, i.e. the mechanism accurately represents the average evolution of the chemical system (Figs.A1 and A2).
The skill of the mechanism can be evaluated in greater details by comparing the times when various species peak as well as their maxima between mechanism and experiment  (Table A1).The mechanism generally captures correctly the peak times indicating that the chemical speed is properly modeled in the first and second regime.The error regarding the maximum intensity falls within the uncertainty of this study (±20−30%).The sensitivity of the CIMS to PNA is probably overevaluated due to ligand exchange with H 2 O 2 .Satisfactory representation for the background chemistry species is also reached (Figs.A3 and A4).In particular, HO 2 NO 2 , a very sensitive marker for the ratio of NO x and HO x , is well captured during the first and second regimes.
For the chamber experimental conditions, the reagent ion was found in several forms: CF Due to the high count rates for the primary isotopes of the reagent ions (sum ∼ 14 MHz), the 13 C isotopes were monitored instead: As stated in Sect.2.3, in order to get the concentration for an analyte, X, detected as a product ion with m/z=p, we divide the normalized signal for m/z by the sensitivity (c X ) for that analyte under chamber conditions Eq. (1).
The above method fails when mass analog ions, i.e. different ions with the same mass-to-charge ratio, exist at the m/z of interest.The mass analog ions correspond to different analytes in the chamber, which have different reaction rate coefficients with the reagent ion.While the CIMS instrument can not separate mass analogs, the explicit model can.To compare the model results with a measured signal composed of mass analogs, we use the following: where a is a m/z featuring mass analog ions, A the subset of compounds yielding product ions with m/z=a and c ref =3.85×10 −4 pptv −1 is taken as an approximate general calibration.Nominally, Signal(CF 3 O − )=120 kcounts/s, this gives c ref =46 counts.s−1 .pptv−1 , in the CIMS flow tube.Including the dilution factor (13.2), the sensitivity is 3.5 counts.s−1 .pptv−1 in the chamber air.
B2 Dipoles and polarizabilities computed by quantum mechanics The dipole moment and polarizability of a molecule depend on its charge distribution.Thus, different conformers of a molecule can have very different dipole moments.The polarizability is essentially determined by the number of electrons and so is not significantly altered by conformers.
We have calculated the dipole moment and polarizability using density functional theory.Many of the molecules of interest have a large number of structural conformers and we have calculated a conformer distribution for all molecules.To generate the initial set of conformers, we have allowed 3 fold rotation about all CC, CO, CN single bonds.This leads to, for example 34 guess structures in the case of ISOPN Z(1, 4).For each guess conformer, geometry optimization is conducted at the B3LYP/6-31G(d) level.The optimized conformers are ranked by energy and relative population for a temperature of 298 K is determined.We have only calculated the polarizability of the lowest energy structure for each of the molecules as we found this to be relatively insensitive to structure.All calculations were performed with Spartan'06, with the default convergence criteria (Wavefunction Inc., 2006).Test calculations on a few small molecules for which the dipole moment has been measured show that the B3LYP/6-31G(d) calculated dipole moments are in good agreement with experiment (Table B1).
MACR and MVK are only produced through the reaction of the β-hydroxy alkoxy radicals with NO and by ozonolysis of isoprene (Fig. 1).Since the latter accounts for less than 0.5% of the total isoprene consumption in the chamber, we can use the direct determination of the yield of these products (Sprengnether et al., 2002) as a constraint.
We consider that all β-hydroxy peroxy radical have the same nitrate branching ratio, α β , as suggested by Giacopelli et al. (2005).In this study, we assume that channels (2, 1) (respectively (3, 4)) yield MVK and MACR.Park et al. (2003) proposed that the radicals formed in these channels undergo a cyclization, thus reducing the yield of MVK and MACR.The yield of the nitrates which should originate from the hydrocarbons proposed by Park et al. (2003) is too small to provide conclusive experimental evidence in favor or against this mechanism.Given the small combined yield of these channels, this uncertainty remains small compared to the ones affecting the major channels.Theoretical determinations of the branching ratio have also been made (Lei et al., 2000;Greenwald et al., 2007): The product of the decomposition of the isoprene nitrates formed in the (4, 1) branch, dihydroxybutanone (DHB) and propanone nitrate (PROPNN), provide an additional constraint (Fig. 8): where α δ is the nitrate yield from the δ-hydroxy peroxy radicals, γ is the branching ratio of the pathway yielding propanone nitrate from ISOPN(2, 1), computed using (A3), α dn 2,1 and α dn 4,1 , the respective organic dinitrate branching ratios from ISOPN(2, 1) and ISOPN(4, 1).
The upper and lower bounds reflect the uncertainty on the identification of the m/z=189 signal.The upper bound is derived assuming all m/=189 signal originates from the DHB yielded by ISOPN(4, 1).The lower limit assumes that no DHB is formed from ISOPN but rather that the signal measured at m/z=189 results from the photooxidation of HC5 Z(1, 4) (negligible based on Peeters' SAR) and ISOPN (1, 4) (formation of dihydroxymethylpropanal (DHMPL) from addition of HO on the less-sterically favored carbon).
In the mechanism, we use the constraints implied by the study of Lei et al. (2000), Y MACR =0.26 and Y MVK =0.40 and the upper bound of Eq.C7.Since the use of the non linear system formed by Eqs.(C1) to (C7) in order to solve for α β , α δ and Y i,j does not yield a single solution, we use the branching ratios derived by Lei et al. (2001) to initialize the numerical solution of this non-linear system and obtain: Y 1,2 41%, Y 1,4 15%, Y 2,1 2%, Y 4,3 23%, Y 4,1 14%, Y 3,4 5%, α δ 24%, α β 6.7%.This set of parameters yields a self-consistent mechanism which captures correctly most of our observations.The constraints implied by the theoretical study of Greenwald et al. (2007) and the experimental work of Sprengnether Table B1.Weighted average dipoles (µ) and polarizabilities (α).Experimental determinations are indicated in parenthesis when available.k X , is the weighted average of the collision rates calculated for conformers with an abundance greater than 5%.k HNO 3 =1.92×10−9 cm 3 molecule −1 s −1 .σ is the weighted standard deviation of the distribution of thermal collision rate constants, i.e. it indicates the sensitivity of the calibration to the calculated distribution of conformers.A direct consequence of the small branching ratio for (4, 1) branch is to preclude DHBN from being an important product of ISOPN(4, 1).The signal measured at m/z=189 would therefore mostly correspond to DHMPL formed from the photooxidation of ISOPN(1, 4).There are two major inconsistencies with this hypothesis.First, due to the small carbon flux through the (4, 1) branch, we are not able to capture the prompt source of hydroxyacetone attributed to Dibble mechanism.Second, DHMPL features an aldehydic group so that its lifetime with respect of HO is expected to be much shorter than DHB inconsistent with the signal recorded at m/z=189 (Fig. 12).
F. Paulot et al.: Isoprene photooxidation: new insights into the production of acids and organic nitrates While the derivation of the specific branching ratio is affected by this major uncertainty, the determination of the overall yield of the different products identified is, on the other hand, strongly constrained by observations and thus relatively insensitive to our choice.The asymmetry of the nitrate yields between the β and δ-hydroxy channels is therefore a reliable feature.Indeed this conclusion bears a striking similarity with the estimate derived by Giacopelli et al. (2005)

C2 Quantum mechanics
In this study, we have assumed that ligand exchange has a negligible impact on CIMS sensitivity.Therefore to assess the accuracy of our calibration, we compare the calculated collision rate with the fastest experimental collision rate: where k r HNO 3 =2.2×10−9 cm 3 molecule −1 s −1 (Huey et al., 1996;Amelynck et al., 2000) and c e X is the maximum sensitivity of this technique determined experimentally by changing the water vapor mixing ratio (often found at zero water vapor mixing ratio).
Conversely, the sensitivity to smaller molecules such as formic (r=1.5) or acetic acid (r=2) is largely overpredicted.If the experimental rates of Amelynck et al. (2000) are used, the agreement is much better with r=1.0 for formic acid and r=1.1 for acetic acid.The discrepancy may be explained by the smaller collisional energy used in the Amelynck et al. (2000)'s experiment which would result in fewer A − .HF complexes being broken during the expansion into the high vacuum.

C3 Inorganic chemistry uncertainties
In addition to uncertainties associated with the VOC chemistry, proper modeling of the background chemistry must be achieved to derive conclusions regarding the VOC chemistry.
The model is especially sensitive to the following parameters: Nitric acid.The rate of HO + NO 2 + M→HNO 3 is an important uncertainty regarding the background chemistry.We use the recently reported rate coefficient of 9.16×10 −12 cm 3 molecule −1 s −1 (Okumura and Sander, 2005) which tends to reduce the rate of formation of nitric acid and conversely increases the formation rate of ozone in comparison with the previous estimates (Atkinson et al., 2006).
Dinitrogen Pentoxide.N 2 O 5 is known to react with water on surfaces (aerosol, walls) to yield nitric acid: The DMA measurements can be used to obtain the aerosol surface area S and the collision rate, k coll = 1 4 8RT π M S=2×10 −3 cm 3 molecule −1 s −1 .The accommodation coefficient is set to 0.05.Initial concentration of H 2 O 2 .Due to the technique used in this experiment to introduce H 2 O 2 into the chamber, its concentration is not known accurately.No calibration is available at such a high hydrogen peroxide level, so that its estimate based on CIMS measurement is uncertain: 1.9-2.3ppm.A new injection method has been developed to this uncertainty in future studies.

Fig. 7 .
Fig. 7. Signal recorded at m/z=159 (black circles) compared with modeled HACET profile.Colored bars indicate the instantaneous modeled contributions of the different sources of HACET.MPAN source has been described byOrlando et al. (2002).

Fig. 8 .
Fig. 8. Decomposition pathway of the different isoprene nitrates after their reaction with OH.Reaction of ISOPN with OH, O 2 , and NO, also yields dihydroxy-dinitrates through Reaction (R7).Color code is identical to Fig. 1.

Fig. 11 .Fig. 12 .
Fig. 11.Signal recorded at m/z=204 (black circles) compared to modeled PROPNN (measured as a cluster).Colored bars indicate the instantaneous modeled contributions of the different sources of PROPNN.

4. 3 Fig. 14 .
Fig. 14.Signal recorded at m/z=65 (black circles) compared to modeled formic acid.Colored bars indicate the instantaneous modeled contributions of the different sources of formic acid.

Fig. 15 .
Fig. 15.Signal recorded at m/z=79 (black circles) compared to modeled acetic acid (observed as a transfer at this mass).Colored bars indicate the instantaneous modeled contributions of the different sources of acetic acid.

F.
Fig. A4.Comparison of CIMS measurements of inorganic species with their modeled concentrations using experimental calibrations.Calibrations for HO 2 NO 2 and N 2 O 5 are uncertain.
using a corrected Carter's parameterization.It is also consistent with the suggestion of O'Brien et al. (1998) that hydrogen bonding in β-hydroxy substituted ROONO intermediate weakens the RO−ONO bond, enhancing RO + NO 2 production.

Table 1 .
Signals monitored by Chemical Ionization Mass Spectrometry.C denotes a cluster (Reaction (R2)) and T a transfer (Reaction (R1)).Uncertain identifications (cf.text) are highlighted by a .

Table 2 .
Isoprene nitrate kinetic data.The uncertainty is smaller than 30%.
MOBA, the class of five-carbon acids identified at m/z=133

Table A1 .
Skill of the model.