Large Gas-Phase Source of Esters and Other Accretion Products in the Atmosphere

Dimeric accretion products have been observed both in atmospheric aerosol particles and in the gas phase. With their low volatilities, they are key contributors to the formation of new aerosol particles, acting as seeds for more volatile organic vapors to partition onto. Many particle-phase accretion products have been identified as esters. Various gas- and particle-phase formation pathways have been suggested for them, yet evidence remains inconclusive. In contrast, peroxide accretion products have been shown to form via gas-phase peroxy radical (RO2) cross reactions. Here, we show that these reactions can also be a major source of esters and other types of accretion products. We studied α-pinene ozonolysis using state-of-the-art chemical ionization mass spectrometry together with different isotopic labeling approaches and quantum chemical calculations, finding strong evidence for fast radical isomerization before accretion. Specifically, this isomerization seems to happen within the intermediate complex of two alkoxy (RO) radicals, which generally determines the branching of all RO2-RO2 reactions. Accretion products are formed when the radicals in the complex recombine. We found that RO with suitable structures can undergo extremely rapid C–C β scissions before recombination, often resulting in ester products. We also found evidence of this previously overlooked RO2–RO2 reaction pathway forming alkyl accretion products and speculate that some earlier peroxide identifications may in fact be hemiacetals or ethers. Our findings help answer several outstanding questions on the sources of accretion products in organic aerosol and bridge our knowledge of the gas phase formation and particle phase detection of accretion products. As esters are inherently more stable than peroxides, this also impacts their further reactivity in the aerosol.

1 Initial RO 2 radical formation in OH oxidation RO 2 radicals are formed in the OH oxidation of α-pinene as well: as compared to ozonolysis, the reaction proceeds more straightforwardly 1 . A simplified schematic on the formation of the three initial RO 2 radicals is shown in Fig. S1.  Figure S1: Formation of the three initial RO 2 radicals in OH oxidation of α-pinene, adapted from Xu et al. and Rolletter et al.. 2,3 Again, the mechanism is not exhaustive, but only to show the formation of the known RO 2 radicals with the formula C 10 H 17 O 3 .

Accretion product formation patterns
The first organic peroxy radicals formed in α-pinene ozonolysis have the formula C 10 H 15 O 4 .
Some of these can undergo autoxidation, in which O 2 molecules are sequentially added to the peroxy radical to yield a more highly oxygenated peroxy radical with the formula C 10 H 15 O 4+2n . Here n denotes the number of autoxidation steps. Correspondingly, in the OH oxidation of α-pinene, the first peroxy radicals have the formula C 10 H 17 O 3 , and C 10 H 17 O 3+2m upon autoxidation. Peroxy radicals can react together to form ROOR accretion products (and an O 2 molecule). The different possible combinations of the peroxy radicals generated in α-pinene ozonolysis and OH reactions to yield a ROOR accretion product are listed in Table S1. Table S1: Compositions of C 20 accretion products formed from RO 2 radicals generated in the O 3 or OH initiated autoxidation of α-pinene (C 10 H 16 ). The n,m, p and q denote the number of autoxidation steps each radical has gone through: with no autoxidation, these are equal to zero. All of the precursor RO 2 and the resulting accretion products, have been observed 4 . The highest oxygenated compound for each type of RO 2 and accretion product observed were C 10  In addition to C 20 accretion products, Berndt et al. also observed C 19 H 28 O odd and C 19 H 30 O even accretion products. The molecular formulae correspond to the C 20 accretion products on the first row of Table S1, with the elimination of an additional CH 2 O. As compared to the C 20 accretion products, the C 19 accretion products had lower oxygen numbers. The majority of them, and all of the most abundant ones, can be explained with the reactions of C 10

Laboratory experiments
The experimental investigations at TROPOS have been conducted in a free-jet flow system at T = 295 ± 2 K, a pressure of 1 bar with a total flow of 100 l min −1 (STP) of purified air and a reaction time of 7.9 s. More detailed information on the experimental setup is given elsewhere. 5,6 Ozone was produced by passing 2 l min −1 (STP) air through an ozone generator (UVP OG-2). α-pinene was taken from a gas mixture in helium prepared in a gas-metering unit. All gas flows were set by means of calibrated gas flow controllers (MKS 1259/1179).
Ozone concentrations were measured at the outflow of the reactor by a gas monitor (Thermo Environmental Instruments 49C) and α-pinene by means of a PTR mass spectrometer (HS PTR-QMS 500, Ionicon). 7 Detection of reaction products was conducted using a CI-APi-TOF mass spectrometer (Airmodus, Tofwerk) sampling from the centre flow of the free-jet flow system with a rate of 10 l min −1 (STP). The ion-molecule reaction (IMR) proceeded at atmospheric pressure using a Boulder-type inlet system. 8 Product ionisation was carried out by means of C 2 H 5 NH + 3 (ethylaminium). 6 Stated lower limit concentrations were obtained using a calibration factor f = 1.85 × 10 9 molecules cm −3 . The "lower limit" approach assumes collision limit in the ion-molecule reaction and no ion losses within the mass spectrometer. 6 The experiments at the University of Helsinki were carried out in the COALA chamber, a two cubic meter (2 m 3 ) continuous stirred-tank reactor (CSTR) made out of teflon (FEP) foil. The setup is described in more detail elsewhere, 9,10 but a brief overview is given here.
Purified air, along with reactants, is continuously injected into the chamber at a total rate of 51.5 l min −1 . Nearly the same amount of air is continuously sampled, with the excess amount flushed out of the chamber to maintain a slight overpressure in the chamber. We injected the isotope-labelled α-pinene in liquid phase into a small (0.5 l min −1 ) flow of high purity nitrogen using a syringe pump with variable injection speeds. The α-pinene evaporated into the flow, and entered the chamber in the gas phase. We produced ozone using an ozone generator, and humidified the flow of clean air into the chamber in select experiments by bubbling it through either purified (MilliQ) water or heavy water (D 2 O).

Accretion products detected at the TROPOS flow tube
If the RO radical originating from RO 2 -Kb undergoes a scission in the complex of two alkyl radicals to yield the C 19 ester accretion products, we would expect this scission to occur also when RO 2 -Kb is in complex with various other RO 2 radicals. We found this to be true at least for RO 2 generated from the oxidation of isoprene and ethylene as well: in these cases, the non-scissioned accretion products contain 15 and 13 carbon atoms, respectively.

Examples of accretion product spectra
We have included tables and spectra of accretion products detected both in in the TROPOS flow tube and the COALA chamber (Tables S3 -S7 and Figs. S4 -S7). Tentative formation mechanisms, including the reactant RO 2 radicals and the assumed linkage type, for the most abundant accretion products are summarized in Table S8. Table S3: Obtained accretion products (C 2 H 5 NH + 3 adducts) and their signal intensities from the ozonolyis of α-pinene measured in the free-jet flow system at TROPOS applying product dilution by a factor of 7, see spectrum in Fig. 4b   3 ) + signal((C 2 H 5 NH 2 )C 2 H 5 NH + 3 )) background corrected; duty-cycle corrected § likely influenced by processes within the IMR Figure S4: Spectrum of accretion products (C 3 H 7 NH + 3 adducts) in the C 18 -C 20 range from the ozonolyis of α-pinene measured in the free-jet flow system at TROPOS, re-analysis of data published in ref. 4 ) background corrected; duty-cycle corrected § likely influenced by processes within the IMR Figure S5: Spectrum of accretion products (C 3 H 7 NH + 3 adducts) in the C 18 -C 20 range from the ozonolyis of α-pinene in the presence of ethylene measured in the free-jet flow system at TROPOS, re-analysis of data published in ref. 4. Corresponding peak composition and intensity of main signals are given in Table S5. Reactant concentrations were: [α-pinene] = 3.9 ×10 11 , [ethylene] = 8.03 ×10 12 and [O 3 ] = 2.0 ×10 12 molecules cm −3 . Table S6: Main signals of accretion products (C 3 H 7 NH + 3 adducts) and their intensities from the ozonolyis of α-pinene in the presence of isoprene measured in the free-jet flow system at TROPOS, re-analysis of published data in ref. 4 ) background corrected; duty-cycle corrected § likely influenced by processes within the IMR Figure S6: Spectrum of accretion products (C 3 H 7 NH + 3 adducts) in the C 18 -C 20 range from the ozonolyis of α-pinene in the presence of isoprene measured in the free-jet flow system at TROPOS, re-analysis of data published in ref. 4. Corresponding peak composition and intensity of main signals are given in Table S6 Table S7: Main signals of accretion products (NO -3 adducts) and their intensities from the ozonolyis of α-pinene measured in COALA chamber, spectra in Figs. 3a and 3e in the main text and spectrum in Fig. S7. Along with the detected accretion product, we also list tentative RO 2 participating in its formation. For accretion products with higher oxygen numbers, multiple combinations of reactant RO 2 are possible. Also listed is whether we assume a beta-scission has occurred in the formation, and the tentative accretion product type. Based on reanalysis of data given in Berndt et al., no ROOR 4 Quantum chemical calculations

Conformational Sampling of Isolated alkoxy and alkyl radicals
Systematic conformer searches for both C 10 O 3 alkoxy ("RO") and C 9 O 2 alkyl radical ("R") "monomer" systems were performed using the Spartan 18 program. 11 In these searches, every non-terminal bond was rotated a certain number of degrees (120 for sp 3 hybridized atoms and 180 for sp 2 hybridized atoms). A molecular mechanics force field (MMFF) optimization was performed to find distinct minimum energy structures on the PES. We found 11 conformers for the alkoxy (RO) and 8 conformers for the alkyl (R) radical monomers. The number of unpaired electrons was set to one, and the total charge were set to zero. Note that unlike earlier versions of Spartan, where peroxy radicals were sometimes treated as anions when default force-field settings were used, Spartan 18 does not require manually setting the type of the radical oxygen atom (e.g. to generic divalent oxygen).

Systematic sampling of the 3 (RO · · · OR) and 3 (R · · · OR) clusters
Systematic conformational sampling of 3 (RO · · · OR) and 3 (R · · · OR) clusters was done using the building up approach described by Kubečka et al.. 12 The purpose of this sampling was to find representative local minima structure on the potential energy surface (PES). The details of the sampling are similar to our previous work on smaller 3 (RO · · · OR) clusters. 12,13 In brief, the monomer conformers from the previous steps were taken and optimized at the ωb97X-D/6-311++G** level of theory. Afterward, the Artificial Bee Colony (ABC) algorithm 14,15 was used to generate thousands of cluster structures for the both 3 (RO · · · OR) and 3 (R · · · OR) clusters. The ABC algorithm performs rigid-body molecular dynamics, which requires partial charges and Lennard Jones parameters for each of the atoms in the cluster.
Partial charges were obtained from natural bonding orbital (NBO) population calculations at the ωb97X-D/6-311++G** level, while Lennard-Jones 16 parameters were taken from the CHARMM force field database. 17 Three thousand cluster conformers were initially generated by the ABC algorithm for each combination of monomer conformers. For 11 alkoxy ("RO") conformers, there are in total 66 possible distinct pairs, leading to 198 000 initial conformers for the 3 (RO · · · OR) cluster. Similarly, the 11 RO and 8 R conformers (88 distinct pairs) led to a total of 264 000 initial conformers for the 3 (R · · · OR) cluster.
Next, GFN-xTB semiempirical optimization was performed for all the conformers generated from ABC using the XTB program. 18 Especially in the 3 (R · · · OR) cluster, various unwanted reactions (e.g. recombination) happened during the XTB optimization due a combination of strained input geometries, and possibly low reaction barriers (at least at this level of theory). While some of these reactions may be real, they are undesirable artifacts from the point of view of the conformational sampling of 3 (R · · · OR) clusters. We first tested different optimization criteria available in the XTB program to maximize the number of intact conformers for our higher level (DFT) calculations. However, this was unsuccessful. We therefore used a systematic filtering approach to remove the unwanted conformers

Sampling of the ROR ester accretion product
About 2703 conformers of the ROR ester accretion product, generated by MMFF conformational sampling in Spartan 18 as described above, were optimized using two different semiempirical methods. First, the GFN-xTB method was applied on both the triplet and singlet surfaces using the XTB program. 18 Redundant and high-energy conformers were then filtered out using the approach described above. (Various unwanted conformers, corresponding to artifact reactions, were found on the triplet surface, and manually discarded). Second, the same set of conformations was sampled using the ODM3 method, 23 as implemented in the MNDO2020 semiempirical code, 24 on both singlet and triplet surfaces. On the triplet surface, approximately one third of the optimizations failed to converge due to failure of the SCF procedure. Despite this, we were able to obtain a large number (on the order of a thousand) of optimized low-energy configurations on both surfaces. The 20 best (lowest-energy) conformers each were selected from the singlet and triplet surfaces from both semiempirical methods for DFT optimization at the ωb97X-D/6-31++G** level. Recombination energetics were then calculated for both singlet and triplet conformers.

ISC rate Calculation
The global minimum-energy conformer, and one representative local minimum conformer with a clearly different bonding pattern, were selected from the ωb97X-D/6-31++G** optimized 3 (RO · · · OR) and 3 (R · · · OR) cluster structures for the inter-system crossing (ISC) rate constant calculation. We used the XMC-QDPT2 multireference method since we are working with open-shell systems. 25,26 The details of the calculations are described in our previous work. 13,27 Briefly, the ISC rate constant (k ISC ) was calculated using the following formula: 28 where ⟨ϕ(T i )|Ĥ SO |ϕ(S j )⟩ 2 are the spin orbit coupling matrix elements (SOCME) in cm −1 , and F ij is the Franck-Condon factor for the transition. Both singlet and triplet energies were The matrix element of spin-orbital coupling interaction (SOCME) between triplet states T1-T4 and singlet states S1-S4 were computed at the CASSCF level of theory, using the XMC-QDPT2/6-311++G(d,p) energies. 30 We used GAMESS-US for this calculation. 31

Alkoxy Bond Scission Rate Calculation
In addition to these calculations, the rate of the RO 2 -Kb derived alkoxy bond scission reaction was determined computationally to further explain our experimental results of the C 19 ester being the main product. Conformer sampling was performed according to the scheme described in Section 4.1.1. Conformers of the transition state were sampled with the scissioning C-C bond frozen after an initial saddle point structure was found, after which the same optimization and filtering workflow was applied as for the alkoxy radicals. All alkoxy radical conformers of suitably low energy were optimized at the ωB97X-D/aug-cc-pVTZ level of theory, after which a transition state structure was searched for by taking the lowest-energy conformer and streching the scissioning OC-C=O bond length to 2.0 Å, perfoming a low-level constrained optimization (B3LYP/6-31+G*) and performing another round of conformational sampling for the formed structure. The energetically lowest structures were again optimized at the ωB97X-D/aug-cc-pVTZ level of theory, after which the reaction rate was calculated using the Eyring equation (2), in this case neglecting quantum tunneling as it is highly unlikely to impact a reaction dependent of heavy atom motion at 300 K, and as calculation of a tunneling coefficient would have required optimization of the formed C 9 H 13 O 2 radical as well.
where RO and TS refer to the alkoxy radical and bond scission transition state, respectively.

Results and discussion
The stability of both 3 (RO · · · OR) and 3 (R · · · OR) clusters are shown in the Table S9.
The binding energies of the two clusters are roughly comparable to each other, and also to the larger of the 3 (RO · · · OR) clusters investigated in our previous study. 13 The weaker electronic binding energy of the 3 (R · · · OR) cluster is probably related to the smaller number of C-H. . . O interactions, as evident from Fig. S10. However, this is partially compensated by the more favorable entropy of the looser 3 (R · · · OR) structure, resulting in almost identical Gibbs free energies of dissociation. Using the detailed-balance approach in our previous study, these free energies would imply cluster dissociation rates on the order of 10 11 s −1 .
However, as discussed extensively in the same study, this is almost certainly an overestimate of the true dissociation rate.
The model does not account for the impact of the dissociation energy of the radical pair, which likely slows the dissociation down by several orders of magnitude. The dissociation energy of the triplet state (RO · · · OR') is 14 kcal/mol (Table S9). A dissociation energy this high means the dissociation rate is going to be substantially lower than 10 11 s −1 . Using a novel first-principles model 32 on calculating the dissociation rate, we can estimate it to be in the range of 10 4 s −1 range, several orders of magnitude below 10 11 s −1 .
To verify that the relatively low binding energies (and thus high detailed-balance evaporation rates) are not caused by conformer issues (i.e. the neglect of low-lying conformers), or by errors in the commonly employed rigid rotor-harmonic oscillator approximation, we have carried out additional test calculations. First, we recomputed all thermal contributions to the free energies of all species using the quasi-harmonic approximation. 33 This was especially important for the 3 (RO · · · OR) system, as the lowest free-energy conformer had multiple vibrational modes with wavenumbers on the order of 10 cm −1 . Next, we recomputed the free energies for the 3 (RO · · · OR) => RO + RO and 3 (R · · · OR) => RO + R reactions using all cluster conformers within 2 kcal/mol in free energy, and the approach of Partanen et al. 34 to account for multiple conformers. The resulting free energies were -2.2 kcal/mol for the 3 (RO · · · OR) => RO + RO reaction, and -4.7 kcal/mol for the 3 (R · · · OR) => RO + R reaction. As these free energies would still imply detailed-balance evaporation rates in excess of 10 11 s −1 for both systems, we can conclude that neither conformer effects nor errors in the harmonic oscillator entropies are behind the qualitative failure of the detailed-balance approach. The distance between the radical centers is 6.7 Å in the 3 (RO · · · OR) cluster and 8.9 Å in the 3 (R · · · OR) cluster.
For both of 3 (RO · · · OR) and 3 (R · · · OR) clusters, we computed the ISC rate both for the global minima conformers, as well as another representative low-energy conformer with different bonding pattern, in order to see how the rates changes between conformers.
Our results confirmed that the ISC rate in both clusters are reasonably fast, similar to our previous study on smaller 3 (RO · · · OR) clusters, though with a large variation between conformers. 13 Notably, the rates predicted for the 3 (R · · · OR) cluster are even faster than for the 3 (RO · · · OR) cluster, and the rate for the second 3 (R · · · OR) conformer exceeds even the upper limit dissociation rate estimated by the detailed balance approach. Thus, provided that the 3 (R · · · OR) cluster forms in the first place (e.g. due to rapid alkoxy scission in the 3 (RO · · · OR) cluster), ISC to the singlet state is very likely a competitive pathway.

Stability of the ROR ester accretion product
The recombination energetics of ROR ester accretion product (with respect to the free RO and R radicals) were calculated on both singlet and triplet surfaces (Table S11). Unsurprisingly, recombination on the singlet surface is extremely exergonic. However, also recombination on the triplet surface was found to be exergonic, due to the presence of carbonyl groups and thus low-lying triplet states in the ester product. Thus, direct recombination of 3 (R · · · OR) to 3 (ROR) is thermodynamically possible -albeit likely associated with an energy barrier making it uncompetitive with either ISC or dissociation (Fig. S11).

Alkoxy Bond Scission Rate Calculation
Using Eq. (2), we calculated the rate for the beta scission of the RO derived from RO 2 -Kb to be 2 × 10 9 s −1 . The optimized structure for the scissioning RO, and the transition state towards the scissioned products, are shown in Fig. S13.

Radical complex
Transition state Figure S11: Optimized structures (at the ωB97X-D/aug-cc-pVTZ level) of a representative transition state connecting a CH 3 C · (O). . . · O CH 2 CH(O) alkyl-alkoxy radical complex to the corresponding ester, on the triplet surface. The electronic energy (free energy at 298 K and 1 atm) difference between the two is 13.6 (11.5) kcal/mol. Note that a full conformer sampling could not be performed for this reaction, as e.g. transition state conformer guesses with the aldehyde oxygen in a "cis" position led to other transition states interconverting different covalently bound diradical species. As these transition states did not connect (through IRC paths) to the correct reactant complex, their existence should not change the qualitative conclusion that direct recombination of alkyl and alkoxy radicals on the triplet surface is likely associated with a substantial barrier.

Exchange of labile deuteriums to hydrogen with H 2 O
Even though we did not humidify the COALA chamber for the majority of the experiments, enough water vapour was present in the system to exchange all the labile deuteriums to hydrogens. We verified this by actively humidifying the chamber during the oxidation of deuterated α-pinene. The deuteration shifted certain HOM compounds by a number of mass units, often fewer than three, as compared to the non-deuterated case. This indicates that the HOM had lost deuterium during oxidation, most probably through intramolecular hydrogen abstraction in autoxidation. Through this, the deuterium atoms became bound to hydroperoxide groups, and thus labile. No change in the spectrum was observed during humidification, meaning that all of the labile deuteriums were already exchanged in the non-humidified chamber (Fig. S14). This means that all of the labile deuteriums (-OOD) have already been exchanged to hydrogen in the dry case. All peaks are observed clustered with the nitrate ion.

α-pinene
Any deuterium atoms abstracted during autoxidation are exchanged with H 2 O for ordinary hydrogen. This means that the loss of a deuterium atom from an observed compound means that that specific deuterium has been abstracted during autoxidation. The most highly oxygenated RO 2 radicals produced in autoxidation have the formula C 10 H 15 O 10 : these form C 20 H 30 O 18 accretion products in their self reactions. 4 As compared to ordinary α-pinene, C 20 H 30 O 18 accretion products formed from the oxidation of d3-α-pinene shift by four or five mass units (Fig. S15). This means that they have lost either one or two deuteriums as compared to the total of six originally present in the two precursor α-pinene molecules. The loss of one deuterium is explained by one of the reacting RO 2 radicals losing one deuterium, while the loss of two could be a result of both losing one, or one RO 2 losing two, while the other has lost none. However, if the latter was the case, we would also expect RO 2 s that have lost two deuterium atoms to react with others that have lost one or two as well, resulting in C 20 H 30 O 18 accretion products that have lost more than two deuterium atoms. As these are not observed, we can deduce that up to one of the three deuterium atoms in the d3-α-pinene can be lost in autoxidation. Thus, for the deuterated version of C 19 H 28 O 11 that has lost a total of four, it is very likely that one has been lost in autoxidation, and the remaining three have been lost together with the carbon atom they were bound to.

Non-ideal behaviour of certain accretion products in dilution experiments
The experiments at TROPOS were conducted using a dilution system for the inlet flow of the CI-APi-TOF. A dilution with nitrogen by up to a factor of seven was performed before the inlet of the instrument. If no unwanted processes occur inside the inlet, and the mea-   Figure S16: Product spectrum of the α-pinene ozonolysis in the range 351.5 -354.0 Th obtained using ethylaminium ionization. The product signal appears as an adduct with C 2 H 5 NH + 3 and is shifted by 46.09 Th. The spectrum represents the measured raw signal recorded from 10 min data accumulation. The red trace is the signal of (C 18 H 26 O 4 )C 2 H 5 NH + 3 from peak fitting and the blue trace the signal of (C 13 H 22 O 8 )C 2 H 5 NH + 3 .

Molecular Dynamics Simulations of Interactions Between α-
Pinene Derived Peroxy Radicals and the Ethyl-ammonium Cation We have previously modelled the association lifetimes of the pre-reactive complex formed by two peroxy radicals by simulating collisions between them using empirical force fields (for details of the force field models, simulation procedures, and results, we refer the reader to our previous publications 35,36   To study the influence of the ethyl-ammonium cation, we first optimized the structure of the complex between a single RO 2 -Kb and CH 3 CH 2 NH + 3 by running molecular dynamics simulations at low temperature (5 K). The structure with the lowest binding energy is shown in Figure S17. This structure shows the formation of very strong hydrogen bonds between the ammonium group and the oxygens in the peroxy radical, leading to a binding energy of −28 kcal/mol. Then, we simulated collisions between the RO 2 -Kb + ethylammonium ionic complex and another RO 2 -Kb. In each of the 100 collisions simulated, we observed the formation of an associated complex of all three molecules which remained held together for the entire 2 ns of the simulation. Based on this, and assuming that any energetic barriers for tetroxide formation are not significantly altered by the presence of the cation, we can predict that the overall reactivity of the cation-bound complex should be essentially collision-limited. This is supported by the structure of the lowest energy tri-molecular complex, also shown in Figure S17. When the second α-pinene peroxy is included, the interaction energy gets lowered by an additional 20 kcal/mol, for a total association energy of −48 kcal/mol. Figure S17: Left) Minimum energy configuration of RO 2 -Kb + CH 3 CH 2 NH + 3 . Right) Minimum energy configuration of 2 RO 2 -Kb + CH 3 CH 2 NH + 3 . Configurations are from LAMMPS simulations run at 5 K.
In summary, the presence of the ethylammonium ion stabilizes the clusters, likely increasing reaction rates. Further, it seems that without the ion, the clusters of two RO 2 -Kb are more weakly bound as compared to clusters of two RO 2 -Ka. Clusters between RO 2 -Ka and RO 2 -Kb lie between the two. As a result, the reaction rate of two RO 2 -Kb is farther away from collision limited, and there is more room for its ion enhancement. This fits qualitatively with our results, where C 18 H 26 O 4 were the most affected by ion mediated processes, while C 19 H 28 O 5 was only slightly affected, and C 20 H 30 O 6 not at all. The first is expected to form from the reaction between two RO 2 -Kb, second in e.g. reaction of RO 2 -Ka with RO 2 -Kb, and the last e.g. in the reaction of two RO 2 -Ka.

Scissions of other RO
We have extensively studied the potential beta scission of the RO derived from RO 2 -Kb (main text Fig. 1). In addition to RO 2 -Kb, α-pinene ozonolysis is also know to product at least four other RO 2 radicals with the formula C 10 H 15 O 4 , some of which can undergo autoxidation to lead to an even larger variety of RO 2 . 37,38 In addition, OH-oxidation of αpinene produces RO 2 -radicals with the formula C 10 H 17 O 3 (Fig. S1), again some of which can undergo autoxidation. 3,39 When these types of RO 2 react together, forming the RO · · · OR' complex, they too may undergo scission reactions while still in the complex. We studied the scissions available to the RO derived from the four other C 10 H 15 O 4 RO 2 radicals from α-  Based on the SAR, 40 we found fast scission reactions for all of the studied RO radicals.
In some cases, such as for RO 2 -I, the scission would lead to fragmentation of the molecule: in this case it would leave a C 7 H 9 O 2 alkyl radical product. Sometimes the radical product would be small, such as for RO 2 -Ka where a CHO alkyl radical is formed. In other cases, the scission would break a ring structure, without fragmenting the molecule in parts. An example of this would be the extremely fast scission available to RO 2 -Kc and RO 2 -Kd, leaving a C 10 H 15 O 3 alkyl radical. If such a radical would recombine with a non-scissioned ten-carbon RO from α-pinene ozonolysis, the resulting product would be an ether (ROR) bonded C 20 accretion product. Further, if two RO 2 -Kc react together, it is conceivable that they would, after scission on both sides, form a C 20 accretion product with a carbon bound backbone.
APINBO2 APINCO2 Figure S18: The scission products for the RO radicals derived from the first-generation RO 2 radicals formed in the ozone (top row) or OH radical (bottom row) oxidation of α-pinene. RO 2 -Kc and RO 2 Kd are stereoisomers, and have the same potential scission points. These isomers have two competing alkoxy bond scission reactions, of which the much faster one breaks the 4-carbon ring, thus having no closed-shell products: both reactions are shown.